diff --git a/analysis/current/analysis-report-hybrid.json b/analysis/current/analysis-report-hybrid.json index a24520474b268c77deb9b72d0106ea2e4764c386..f3ca9db7dfe531e32f2973dc803cd1d8a8efd030 100644 --- a/analysis/current/analysis-report-hybrid.json +++ b/analysis/current/analysis-report-hybrid.json @@ -1,14 +1,14 @@ { "schema_version": "1.0", "repo": "huggingface/diffusers", - "snapshot_id": "20260427T112541Z", - "generated_at": "2026-04-27T11:44:46Z", + "snapshot_id": "20260427T113512Z", + "generated_at": "2026-04-27T12:02:03Z", "evidence_quality": "full", "llm_enrichment": true, "meta_bugs": [ { - "cluster_id": "cluster-1246-22", - "summary": "Cluster of 17 issues and 5 PRs centered on issue #13124.", + "cluster_id": "cluster-1246-28", + "summary": "Cluster of 23 issues and 5 PRs centered on issue #13124.", "status": "open", "confidence": 0.99, "canonical_issue_number": 13124, @@ -16,15 +16,21 @@ "issue_numbers": [ 1246, 2780, + 4566, + 5412, + 5757, + 5840, 6086, 6098, 6124, 6277, 6363, + 6366, 6442, 6454, 6482, 7330, + 8659, 8871, 9628, 9973, @@ -250,7 +256,7 @@ "cluster_id": "cluster-6585-11", "summary": "Cluster of 1 issues and 10 PRs centered on issue #6585.", "status": "merged", - "confidence": 0.75, + "confidence": 0.8, "canonical_issue_number": 6585, "canonical_pr_number": 6737, "issue_numbers": [ @@ -270,7 +276,8 @@ ], "evidence_types": [ "closing_reference", - "shared_issue_target" + "shared_issue_target", + "soft_similarity" ], "pr_comparisons": [ { @@ -797,4076 +804,460 @@ ] }, { - "cluster_id": "cluster-11390-10", - "summary": "Cluster of 1 issues and 9 PRs centered on issue #11390.", + "cluster_id": "cluster-4478-9", + "summary": "Cluster of 6 issues and 3 PRs centered on issue #4619.", "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 11390, - "canonical_pr_number": 11463, + "confidence": 0.6, + "canonical_issue_number": 4619, + "canonical_pr_number": 4796, "issue_numbers": [ - 11390 + 4478, + 4619, + 4752, + 5510, + 7572, + 7854 ], "pr_numbers": [ - 11395, - 11421, - 11426, - 11427, - 11463, - 11471, - 11472, - 11491, - 11492 + 4632, + 4796, + 7858 ], "evidence_types": [ "closing_reference", - "shared_issue_target", "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 11395, - "right_pr_number": 11421, - "code_similarity": 0.338, - "size_similarity": 0.96, + "left_pr_number": 4632, + "right_pr_number": 4796, + "code_similarity": 0.14, + "size_similarity": 0.682, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.972, + "patch_similarity": 0.026, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11395, - "right_pr_number": 11426, - "code_similarity": 0.271, - "size_similarity": 0.8, + "left_pr_number": 4632, + "right_pr_number": 7858, + "code_similarity": 0.034, + "size_similarity": 0.16, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.739, + "patch_similarity": 0.015, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11395, - "right_pr_number": 11427, - "code_similarity": 0.271, - "size_similarity": 0.8, - "file_overlap": 0.0, + "left_pr_number": 4796, + "right_pr_number": 7858, + "code_similarity": 0.093, + "size_similarity": 0.109, + "file_overlap": 0.111, "area_overlap": 0.0, - "patch_similarity": 0.739, - "shared_filenames": [], - "shared_file_areas": [] - }, + "patch_similarity": 0.25, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py", + "left_ranges": [ + [ + 908, + 926 + ] + ], + "right_ranges": [ + [ + 525, + 532 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9567-8", + "summary": "Cluster of 1 issues and 7 PRs centered on issue #9567.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 9567, + "canonical_pr_number": 9608, + "issue_numbers": [ + 9567 + ], + "pr_numbers": [ + 9579, + 9582, + 9583, + 9606, + 9608, + 9724, + 9815 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ { - "left_pr_number": 11395, - "right_pr_number": 11463, - "code_similarity": 0.179, - "size_similarity": 0.444, + "left_pr_number": 9579, + "right_pr_number": 9582, + "code_similarity": 0.043, + "size_similarity": 0.175, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.604, + "patch_similarity": 0.053, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11395, - "right_pr_number": 11471, - "code_similarity": 0.335, - "size_similarity": 0.923, + "left_pr_number": 9579, + "right_pr_number": 9583, + "code_similarity": 0.061, + "size_similarity": 0.292, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 1.0, + "patch_similarity": 0.015, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11395, - "right_pr_number": 11472, - "code_similarity": 0.335, - "size_similarity": 0.923, + "left_pr_number": 9579, + "right_pr_number": 9606, + "code_similarity": 0.088, + "size_similarity": 0.417, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 1.0, + "patch_similarity": 0.032, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11395, - "right_pr_number": 11491, - "code_similarity": 0.328, - "size_similarity": 0.889, + "left_pr_number": 9579, + "right_pr_number": 9608, + "code_similarity": 0.104, + "size_similarity": 0.376, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 1.0, + "patch_similarity": 0.194, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11395, - "right_pr_number": 11492, - "code_similarity": 0.338, - "size_similarity": 0.96, + "left_pr_number": 9579, + "right_pr_number": 9724, + "code_similarity": 0.054, + "size_similarity": 0.225, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.972, + "patch_similarity": 0.062, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11421, - "right_pr_number": 11426, - "code_similarity": 0.28, - "size_similarity": 0.833, + "left_pr_number": 9579, + "right_pr_number": 9815, + "code_similarity": 0.05, + "size_similarity": 0.2, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.756, + "patch_similarity": 0.068, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11421, - "right_pr_number": 11427, - "code_similarity": 0.28, - "size_similarity": 0.833, + "left_pr_number": 9582, + "right_pr_number": 9583, + "code_similarity": 0.128, + "size_similarity": 0.6, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.756, + "patch_similarity": 0.053, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11421, - "right_pr_number": 11463, - "code_similarity": 0.185, - "size_similarity": 0.463, + "left_pr_number": 9582, + "right_pr_number": 9606, + "code_similarity": 0.088, + "size_similarity": 0.42, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.615, + "patch_similarity": 0.027, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11421, - "right_pr_number": 11471, - "code_similarity": 0.338, - "size_similarity": 0.962, + "left_pr_number": 9582, + "right_pr_number": 9608, + "code_similarity": 0.022, + "size_similarity": 0.066, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.972, + "patch_similarity": 0.06, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11421, - "right_pr_number": 11472, - "code_similarity": 0.338, - "size_similarity": 0.962, + "left_pr_number": 9582, + "right_pr_number": 9724, + "code_similarity": 0.158, + "size_similarity": 0.778, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.972, + "patch_similarity": 0.019, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11421, - "right_pr_number": 11491, - "code_similarity": 0.331, - "size_similarity": 0.926, + "left_pr_number": 9582, + "right_pr_number": 9815, + "code_similarity": 0.179, + "size_similarity": 0.875, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.972, + "patch_similarity": 0.027, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11421, - "right_pr_number": 11492, - "code_similarity": 0.35, - "size_similarity": 1.0, + "left_pr_number": 9583, + "right_pr_number": 9606, + "code_similarity": 0.142, + "size_similarity": 0.7, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 1.0, + "patch_similarity": 0.014, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11426, - "right_pr_number": 11427, - "code_similarity": 0.35, - "size_similarity": 1.0, + "left_pr_number": 9583, + "right_pr_number": 9608, + "code_similarity": 0.024, + "size_similarity": 0.11, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 1.0, + "patch_similarity": 0.013, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11426, - "right_pr_number": 11463, - "code_similarity": 0.194, - "size_similarity": 0.556, + "left_pr_number": 9583, + "right_pr_number": 9724, + "code_similarity": 0.157, + "size_similarity": 0.771, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.55, + "patch_similarity": 0.02, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11426, - "right_pr_number": 11471, - "code_similarity": 0.284, - "size_similarity": 0.867, + "left_pr_number": 9583, + "right_pr_number": 9815, + "code_similarity": 0.141, + "size_similarity": 0.686, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.739, + "patch_similarity": 0.029, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11426, - "right_pr_number": 11472, - "code_similarity": 0.284, - "size_similarity": 0.867, + "left_pr_number": 9606, + "right_pr_number": 9608, + "code_similarity": 0.038, + "size_similarity": 0.157, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.739, + "patch_similarity": 0.044, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 11426, - "right_pr_number": 11491, - "code_similarity": 0.291, - "size_similarity": 0.9, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.739, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11426, - "right_pr_number": 11492, - "code_similarity": 0.28, - "size_similarity": 0.833, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.756, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11427, - "right_pr_number": 11463, - "code_similarity": 0.194, - "size_similarity": 0.556, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.55, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11427, - "right_pr_number": 11471, - "code_similarity": 0.284, - "size_similarity": 0.867, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.739, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11427, - "right_pr_number": 11472, - "code_similarity": 0.284, - "size_similarity": 0.867, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.739, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11427, - "right_pr_number": 11491, - "code_similarity": 0.291, - "size_similarity": 0.9, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.739, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11427, - "right_pr_number": 11492, - "code_similarity": 0.28, - "size_similarity": 0.833, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.756, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11463, - "right_pr_number": 11471, - "code_similarity": 0.187, - "size_similarity": 0.481, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.604, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11463, - "right_pr_number": 11472, - "code_similarity": 0.187, - "size_similarity": 0.481, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.604, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11463, - "right_pr_number": 11491, - "code_similarity": 0.191, - "size_similarity": 0.5, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.604, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11463, - "right_pr_number": 11492, - "code_similarity": 0.185, - "size_similarity": 0.463, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.615, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11471, - "right_pr_number": 11472, - "code_similarity": 0.35, - "size_similarity": 1.0, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11471, - "right_pr_number": 11491, - "code_similarity": 0.343, - "size_similarity": 0.963, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11471, - "right_pr_number": 11492, - "code_similarity": 0.338, - "size_similarity": 0.962, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.972, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11472, - "right_pr_number": 11491, - "code_similarity": 0.343, - "size_similarity": 0.963, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11472, - "right_pr_number": 11492, - "code_similarity": 0.338, - "size_similarity": 0.962, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.972, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11491, - "right_pr_number": 11492, - "code_similarity": 0.331, - "size_similarity": 0.926, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.972, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-4478-9", - "summary": "Cluster of 6 issues and 3 PRs centered on issue #4619.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 4619, - "canonical_pr_number": 4796, - "issue_numbers": [ - 4478, - 4619, - 4752, - 5510, - 7572, - 7854 - ], - "pr_numbers": [ - 4632, - 4796, - 7858 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 4632, - "right_pr_number": 4796, - "code_similarity": 0.14, - "size_similarity": 0.682, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.026, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 4632, - "right_pr_number": 7858, - "code_similarity": 0.034, - "size_similarity": 0.16, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.015, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 4796, - "right_pr_number": 7858, - "code_similarity": 0.093, - "size_similarity": 0.109, - "file_overlap": 0.111, - "area_overlap": 0.0, - "patch_similarity": 0.25, - "shared_filenames": [ - "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py", - "left_ranges": [ - [ - 908, - 926 - ] - ], - "right_ranges": [ - [ - 525, - 532 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-9567-8", - "summary": "Cluster of 1 issues and 7 PRs centered on issue #9567.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 9567, - "canonical_pr_number": 9608, - "issue_numbers": [ - 9567 - ], - "pr_numbers": [ - 9579, - 9582, - 9583, - 9606, - 9608, - 9724, - 9815 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 9579, - "right_pr_number": 9582, - "code_similarity": 0.043, - "size_similarity": 0.175, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.053, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9579, - "right_pr_number": 9583, - "code_similarity": 0.061, - "size_similarity": 0.292, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.015, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9579, - "right_pr_number": 9606, - "code_similarity": 0.088, - "size_similarity": 0.417, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.032, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9579, - "right_pr_number": 9608, - "code_similarity": 0.104, - "size_similarity": 0.376, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.194, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9579, - "right_pr_number": 9724, - "code_similarity": 0.054, - "size_similarity": 0.225, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.062, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9579, - "right_pr_number": 9815, - "code_similarity": 0.05, - "size_similarity": 0.2, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.068, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9582, - "right_pr_number": 9583, - "code_similarity": 0.128, - "size_similarity": 0.6, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.053, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9582, - "right_pr_number": 9606, - "code_similarity": 0.088, - "size_similarity": 0.42, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.027, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9582, - "right_pr_number": 9608, - "code_similarity": 0.022, - "size_similarity": 0.066, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.06, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9582, - "right_pr_number": 9724, - "code_similarity": 0.158, - "size_similarity": 0.778, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.019, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9582, - "right_pr_number": 9815, - "code_similarity": 0.179, - "size_similarity": 0.875, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.027, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9583, - "right_pr_number": 9606, - "code_similarity": 0.142, - "size_similarity": 0.7, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.014, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9583, - "right_pr_number": 9608, - "code_similarity": 0.024, - "size_similarity": 0.11, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.013, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9583, - "right_pr_number": 9724, - "code_similarity": 0.157, - "size_similarity": 0.771, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.02, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9583, - "right_pr_number": 9815, - "code_similarity": 0.141, - "size_similarity": 0.686, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.029, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9606, - "right_pr_number": 9608, - "code_similarity": 0.038, - "size_similarity": 0.157, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.044, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9606, - "right_pr_number": 9724, - "code_similarity": 0.426, - "size_similarity": 0.54, - "file_overlap": 1.0, - "area_overlap": 0.017, - "patch_similarity": 0.081, - "shared_filenames": [ - "src/diffusers/training_utils.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/training_utils.py", - "left_ranges": [ - [ - 36, - 44 - ], - [ - 195, - 207 - ], - [ - 233, - 240 - ], - [ - 253, - 260 - ], - [ - 271, - 279 - ], - [ - 506, - 513 - ], - [ - 541, - 565 - ], - [ - 575, - 584 - ] - ], - "right_ranges": [ - [ - 43, - 51 - ], - [ - 59, - 76 - ] - ] - } - ] - }, - { - "left_pr_number": 9606, - "right_pr_number": 9815, - "code_similarity": 0.413, - "size_similarity": 0.48, - "file_overlap": 1.0, - "area_overlap": 0.018, - "patch_similarity": 0.071, - "shared_filenames": [ - "src/diffusers/training_utils.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/training_utils.py", - "left_ranges": [ - [ - 36, - 44 - ], - [ - 195, - 207 - ], - [ - 233, - 240 - ], - [ - 253, - 260 - ], - [ - 271, - 279 - ], - [ - 506, - 513 - ], - [ - 541, - 565 - ], - [ - 575, - 584 - ] - ], - "right_ranges": [ - [ - 43, - 51 - ], - [ - 61, - 77 - ] - ] - } - ] - }, - { - "left_pr_number": 9608, - "right_pr_number": 9724, - "code_similarity": 0.026, - "size_similarity": 0.085, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.061, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9608, - "right_pr_number": 9815, - "code_similarity": 0.025, - "size_similarity": 0.075, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.067, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9724, - "right_pr_number": 9815, - "code_similarity": 0.884, - "size_similarity": 0.889, - "file_overlap": 1.0, - "area_overlap": 0.893, - "patch_similarity": 0.628, - "shared_filenames": [ - "src/diffusers/training_utils.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/training_utils.py", - "left_ranges": [ - [ - 43, - 51 - ], - [ - 59, - 76 - ] - ], - "right_ranges": [ - [ - 43, - 51 - ], - [ - 61, - 77 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-6984-7", - "summary": "Cluster of 1 issues and 6 PRs centered on issue #6984.", - "status": "open", - "confidence": 0.8, - "canonical_issue_number": 6984, - "canonical_pr_number": 7071, - "issue_numbers": [ - 6984 - ], - "pr_numbers": [ - 7052, - 7071, - 7211, - 7307, - 8376, - 8387 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 7052, - "right_pr_number": 7071, - "code_similarity": 0.063, - "size_similarity": 0.139, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.233, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7052, - "right_pr_number": 7211, - "code_similarity": 0.069, - "size_similarity": 0.185, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.216, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7052, - "right_pr_number": 7307, - "code_similarity": 0.403, - "size_similarity": 0.269, - "file_overlap": 0.5, - "area_overlap": 0.453, - "patch_similarity": 0.274, - "shared_filenames": [ - "examples/community/stable_diffusion_tensorrt_txt2img.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/stable_diffusion_tensorrt_txt2img.py", - "left_ranges": [ - [ - 19, - 28 - ], - [ - 41, - 58 - ], - [ - 592, - 602 - ], - [ - 678, - 685 - ], - [ - 697, - 771 - ], - [ - 1008, - 1013 - ] - ], - "right_ranges": [ - [ - 19, - 36 - ], - [ - 45, - 62 - ], - [ - 594, - 606 - ], - [ - 683, - 780 - ], - [ - 791, - 865 - ], - [ - 1102, - 1107 - ] - ] - } - ] - }, - { - "left_pr_number": 7052, - "right_pr_number": 8376, - "code_similarity": 0.088, - "size_similarity": 0.299, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.187, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7052, - "right_pr_number": 8387, - "code_similarity": 0.088, - "size_similarity": 0.299, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.187, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7071, - "right_pr_number": 7211, - "code_similarity": 0.274, - "size_similarity": 0.754, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.818, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7071, - "right_pr_number": 7307, - "code_similarity": 0.156, - "size_similarity": 0.517, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.352, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7071, - "right_pr_number": 8376, - "code_similarity": 0.186, - "size_similarity": 0.466, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.619, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7071, - "right_pr_number": 8387, - "code_similarity": 0.186, - "size_similarity": 0.466, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.619, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7211, - "right_pr_number": 7307, - "code_similarity": 0.186, - "size_similarity": 0.685, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.326, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7211, - "right_pr_number": 8376, - "code_similarity": 0.223, - "size_similarity": 0.618, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.665, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7211, - "right_pr_number": 8387, - "code_similarity": 0.223, - "size_similarity": 0.618, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.665, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7307, - "right_pr_number": 8376, - "code_similarity": 0.231, - "size_similarity": 0.901, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.341, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 7307, - "right_pr_number": 8387, - "code_similarity": 0.231, - "size_similarity": 0.901, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.341, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8376, - "right_pr_number": 8387, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, - "shared_filenames": [ - "examples/community/regional_prompting_stable_diffusion.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/regional_prompting_stable_diffusion.py", - "left_ranges": [ - [ - 1, - 28 - ], - [ - 33, - 41 - ], - [ - 83, - 182 - ], - [ - 186, - 376 - ], - [ - 399, - 405 - ] - ], - "right_ranges": [ - [ - 1, - 28 - ], - [ - 33, - 41 - ], - [ - 83, - 182 - ], - [ - 186, - 376 - ], - [ - 399, - 405 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-6503-8", - "summary": "Cluster of 1 issues and 7 PRs centered on issue #6503.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 6503, - "canonical_pr_number": 6532, - "issue_numbers": [ - 6503 - ], - "pr_numbers": [ - 6525, - 6526, - 6532, - 6534, - 6550, - 6555, - 6556 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 6525, - "right_pr_number": 6526, - "code_similarity": 0.269, - "size_similarity": 0.893, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.6, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6525, - "right_pr_number": 6532, - "code_similarity": 0.172, - "size_similarity": 0.549, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.417, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6525, - "right_pr_number": 6534, - "code_similarity": 0.18, - "size_similarity": 0.651, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.333, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6525, - "right_pr_number": 6550, - "code_similarity": 0.195, - "size_similarity": 0.757, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.293, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6525, - "right_pr_number": 6555, - "code_similarity": 0.252, - "size_similarity": 0.929, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.444, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6525, - "right_pr_number": 6556, - "code_similarity": 0.149, - "size_similarity": 0.528, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.288, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6526, - "right_pr_number": 6532, - "code_similarity": 0.173, - "size_similarity": 0.49, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.5, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6526, - "right_pr_number": 6534, - "code_similarity": 0.176, - "size_similarity": 0.581, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.4, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6526, - "right_pr_number": 6550, - "code_similarity": 0.196, - "size_similarity": 0.676, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.405, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6526, - "right_pr_number": 6555, - "code_similarity": 0.265, - "size_similarity": 0.962, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.485, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6526, - "right_pr_number": 6556, - "code_similarity": 0.149, - "size_similarity": 0.472, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.364, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6532, - "right_pr_number": 6534, - "code_similarity": 0.264, - "size_similarity": 0.843, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.638, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6532, - "right_pr_number": 6550, - "code_similarity": 0.198, - "size_similarity": 0.725, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.356, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6532, - "right_pr_number": 6555, - "code_similarity": 0.162, - "size_similarity": 0.51, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.4, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6532, - "right_pr_number": 6556, - "code_similarity": 0.252, - "size_similarity": 0.962, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.397, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6534, - "right_pr_number": 6550, - "code_similarity": 0.236, - "size_similarity": 0.86, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.426, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6534, - "right_pr_number": 6555, - "code_similarity": 0.202, - "size_similarity": 0.605, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.541, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6534, - "right_pr_number": 6556, - "code_similarity": 0.244, - "size_similarity": 0.811, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.545, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6550, - "right_pr_number": 6555, - "code_similarity": 0.205, - "size_similarity": 0.703, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.429, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6550, - "right_pr_number": 6556, - "code_similarity": 0.221, - "size_similarity": 0.698, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.543, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6555, - "right_pr_number": 6556, - "code_similarity": 0.166, - "size_similarity": 0.491, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.452, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-9514-5", - "summary": "Cluster of 3 issues and 2 PRs centered on issue #10989.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 10989, - "canonical_pr_number": 10990, - "issue_numbers": [ - 9514, - 10337, - 10989 - ], - "pr_numbers": [ - 10338, - 10990 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 10338, - "right_pr_number": 10990, - "code_similarity": 0.016, - "size_similarity": 0.076, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.005, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-9591-5", - "summary": "Cluster of 5 related pull requests linked by soft_similarity.", - "status": "open", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 9591, - "issue_numbers": [], - "pr_numbers": [ - 9591, - 9706, - 9707, - 9714, - 9715 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 9591, - "right_pr_number": 9706, - "code_similarity": 0.099, - "size_similarity": 0.422, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.097, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9591, - "right_pr_number": 9707, - "code_similarity": 0.099, - "size_similarity": 0.422, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.097, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9591, - "right_pr_number": 9714, - "code_similarity": 0.099, - "size_similarity": 0.422, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.097, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9591, - "right_pr_number": 9715, - "code_similarity": 0.099, - "size_similarity": 0.422, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.097, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9706, - "right_pr_number": 9707, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, - "shared_filenames": [ - "src/diffusers/models/transformers/auraflow_transformer_2d.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", - "left_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ], - "right_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ] - } - ] - }, - { - "left_pr_number": 9706, - "right_pr_number": 9714, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, - "shared_filenames": [ - "src/diffusers/models/transformers/auraflow_transformer_2d.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", - "left_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ], - "right_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ] - } - ] - }, - { - "left_pr_number": 9706, - "right_pr_number": 9715, - "code_similarity": 0.996, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 0.975, - "shared_filenames": [ - "src/diffusers/models/transformers/auraflow_transformer_2d.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", - "left_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ], - "right_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ] - } - ] - }, - { - "left_pr_number": 9707, - "right_pr_number": 9714, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, - "shared_filenames": [ - "src/diffusers/models/transformers/auraflow_transformer_2d.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", - "left_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ], - "right_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ] - } - ] - }, - { - "left_pr_number": 9707, - "right_pr_number": 9715, - "code_similarity": 0.996, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 0.975, - "shared_filenames": [ - "src/diffusers/models/transformers/auraflow_transformer_2d.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", - "left_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ], - "right_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ] - } - ] - }, - { - "left_pr_number": 9714, - "right_pr_number": 9715, - "code_similarity": 0.996, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 0.975, - "shared_filenames": [ - "src/diffusers/models/transformers/auraflow_transformer_2d.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", - "left_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ], - "right_ranges": [ - [ - 187, - 193 - ], - [ - 257, - 277 - ], - [ - 347, - 355 - ], - [ - 373, - 385 - ], - [ - 406, - 415 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-8446-5", - "summary": "Cluster of 5 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 8461, - "issue_numbers": [], - "pr_numbers": [ - 8446, - 8461, - 8476, - 9308, - 9316 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 8446, - "right_pr_number": 8461, - "code_similarity": 0.35, - "size_similarity": 1.0, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8446, - "right_pr_number": 8476, - "code_similarity": 0.35, - "size_similarity": 1.0, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8446, - "right_pr_number": 9308, - "code_similarity": 0.321, - "size_similarity": 0.921, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.914, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8446, - "right_pr_number": 9316, - "code_similarity": 0.35, - "size_similarity": 1.0, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8461, - "right_pr_number": 8476, - "code_similarity": 0.35, - "size_similarity": 1.0, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8461, - "right_pr_number": 9308, - "code_similarity": 0.321, - "size_similarity": 0.921, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.914, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8461, - "right_pr_number": 9316, - "code_similarity": 0.35, - "size_similarity": 1.0, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8476, - "right_pr_number": 9308, - "code_similarity": 0.321, - "size_similarity": 0.921, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.914, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 8476, - "right_pr_number": 9316, - "code_similarity": 0.35, - "size_similarity": 1.0, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 9308, - "right_pr_number": 9316, - "code_similarity": 0.321, - "size_similarity": 0.921, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.914, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-4498-5", - "summary": "Cluster of 5 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 4502, - "issue_numbers": [], - "pr_numbers": [ - 4498, - 4499, - 4500, - 4501, - 4502 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 4498, - "right_pr_number": 4499, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1850, - 1887 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 788 - ] - ], - "right_ranges": [ - [ - 1, - 788 - ] - ] - } - ] - }, - { - "left_pr_number": 4498, - "right_pr_number": 4500, - "code_similarity": 0.999, - "size_similarity": 0.999, - "file_overlap": 1.0, - "area_overlap": 0.999, - "patch_similarity": 1.0, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1850, - 1887 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 788 - ] - ], - "right_ranges": [ - [ - 1, - 789 - ] - ] - } - ] - }, - { - "left_pr_number": 4498, - "right_pr_number": 4501, - "code_similarity": 0.997, - "size_similarity": 0.994, - "file_overlap": 1.0, - "area_overlap": 0.994, - "patch_similarity": 1.0, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1850, - 1887 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 788 - ] - ], - "right_ranges": [ - [ - 1, - 793 - ] - ] - } - ] - }, - { - "left_pr_number": 4498, - "right_pr_number": 4502, - "code_similarity": 0.937, - "size_similarity": 0.953, - "file_overlap": 1.0, - "area_overlap": 0.87, - "patch_similarity": 0.948, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1889, - 1957 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 788 - ] - ], - "right_ranges": [ - [ - 1, - 799 - ] - ] - } - ] - }, - { - "left_pr_number": 4499, - "right_pr_number": 4500, - "code_similarity": 0.999, - "size_similarity": 0.999, - "file_overlap": 1.0, - "area_overlap": 0.999, - "patch_similarity": 1.0, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1850, - 1887 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 788 - ] - ], - "right_ranges": [ - [ - 1, - 789 - ] - ] - } - ] - }, - { - "left_pr_number": 4499, - "right_pr_number": 4501, - "code_similarity": 0.997, - "size_similarity": 0.994, - "file_overlap": 1.0, - "area_overlap": 0.994, - "patch_similarity": 1.0, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1850, - 1887 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 788 - ] - ], - "right_ranges": [ - [ - 1, - 793 - ] - ] - } - ] - }, - { - "left_pr_number": 4499, - "right_pr_number": 4502, - "code_similarity": 0.937, - "size_similarity": 0.953, - "file_overlap": 1.0, - "area_overlap": 0.87, - "patch_similarity": 0.948, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1889, - 1957 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 788 - ] - ], - "right_ranges": [ - [ - 1, - 799 - ] - ] - } - ] - }, - { - "left_pr_number": 4500, - "right_pr_number": 4501, - "code_similarity": 0.997, - "size_similarity": 0.995, - "file_overlap": 1.0, - "area_overlap": 0.995, - "patch_similarity": 1.0, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1850, - 1887 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 789 - ] - ], - "right_ranges": [ - [ - 1, - 793 - ] - ] - } - ] - }, - { - "left_pr_number": 4500, - "right_pr_number": 4502, - "code_similarity": 0.938, - "size_similarity": 0.954, - "file_overlap": 1.0, - "area_overlap": 0.871, - "patch_similarity": 0.948, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1889, - 1957 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 789 - ] - ], - "right_ranges": [ - [ - 1, - 799 - ] - ] - } - ] - }, - { - "left_pr_number": 4501, - "right_pr_number": 4502, - "code_similarity": 0.94, - "size_similarity": 0.958, - "file_overlap": 1.0, - "area_overlap": 0.875, - "patch_similarity": 0.948, - "shared_filenames": [ - "examples/community/README.md", - "examples/community/stable_diffusion_xl_reference.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/README.md", - "left_ranges": [ - [ - 1850, - 1887 - ] - ], - "right_ranges": [ - [ - 1889, - 1957 - ] - ] - }, - { - "filename": "examples/community/stable_diffusion_xl_reference.py", - "left_ranges": [ - [ - 1, - 793 - ] - ], - "right_ranges": [ - [ - 1, - 799 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-9403-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #9403.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 9403, - "canonical_pr_number": 10261, - "issue_numbers": [ - 9403, - 9825 - ], - "pr_numbers": [ - 9787, - 10261 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 9787, - "right_pr_number": 10261, - "code_similarity": 0.589, - "size_similarity": 0.715, - "file_overlap": 0.583, - "area_overlap": 0.468, - "patch_similarity": 0.715, - "shared_filenames": [ - "scripts/convert_flux_xlabs_ipadapter_to_diffusers.py", - "src/diffusers/loaders/__init__.py", - "src/diffusers/loaders/ip_adapter.py", - "src/diffusers/loaders/transformer_flux.py", - "src/diffusers/models/attention_processor.py", - "src/diffusers/models/transformers/transformer_flux.py", - "src/diffusers/pipelines/flux/pipeline_flux.py" - ], - "shared_file_areas": [ - { - "filename": "scripts/convert_flux_xlabs_ipadapter_to_diffusers.py", - "left_ranges": [ - [ - 1, - 97 - ] - ], - "right_ranges": [ - [ - 1, - 97 - ] - ] - }, - { - "filename": "src/diffusers/loaders/__init__.py", - "left_ranges": [ - [ - 55, - 61 - ], - [ - 70, - 89 - ] - ], - "right_ranges": [ - [ - 55, - 61 - ], - [ - 77, - 83 - ], - [ - 87, - 100 - ] - ] - }, - { - "filename": "src/diffusers/loaders/ip_adapter.py", - "left_ranges": [ - [ - 41, - 48 - ], - [ - 348, - 611 - ] - ], - "right_ranges": [ - [ - 38, - 45 - ], - [ - 355, - 644 - ] - ] - }, - { - "filename": "src/diffusers/loaders/transformer_flux.py", - "left_ranges": [ - [ - 1, - 174 - ] - ], - "right_ranges": [ - [ - 1, - 179 - ] - ] - }, - { - "filename": "src/diffusers/models/attention_processor.py", - "left_ranges": [ - [ - 482, - 488 - ], - [ - 1987, - 2215 - ], - [ - 4614, - 4620 - ] - ], - "right_ranges": [ - [ - 575, - 581 - ], - [ - 2653, - 2801 - ], - [ - 6039, - 6045 - ] - ] - }, - { - "filename": "src/diffusers/models/transformers/transformer_flux.py", - "left_ranges": [ - [ - 21, - 27 - ], - [ - 207, - 215 - ] - ], - "right_ranges": [ - [ - 21, - 27 - ], - [ - 177, - 194 - ], - [ - 200, - 207 - ], - [ - 219, - 227 - ], - [ - 491, - 501 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/flux/pipeline_flux.py", - "left_ranges": [ - [ - 17, - 33 - ], - [ - 149, - 155 - ], - [ - 190, - 197 - ], - [ - 203, - 210 - ], - [ - 387, - 436 - ], - [ - 611, - 618 - ], - [ - 662, - 672 - ], - [ - 783, - 799 - ] - ], - "right_ranges": [ - [ - 17, - 33 - ], - [ - 149, - 155 - ], - [ - 177, - 184 - ], - [ - 190, - 197 - ], - [ - 203, - 210 - ], - [ - 389, - 448 - ], - [ - 477, - 509 - ], - [ - 632, - 640 - ], - [ - 645, - 656 - ], - [ - 700, - 716 - ], - [ - 748, - 759 - ], - [ - 775, - 781 - ], - [ - 790, - 810 - ], - [ - 846, - 888 - ], - [ - 898, - 919 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-8810-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #9886.", - "status": "open", - "confidence": 0.6, - "canonical_issue_number": 9886, - "canonical_pr_number": 8864, - "issue_numbers": [ - 9886 - ], - "pr_numbers": [ - 8810, - 8864, - 9917 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 8810, - "right_pr_number": 8864, - "code_similarity": 0.849, - "size_similarity": 0.755, - "file_overlap": 0.875, - "area_overlap": 0.886, - "patch_similarity": 0.835, - "shared_filenames": [ - "docs/source/en/api/pipelines/pag.md", - "src/diffusers/__init__.py", - "src/diffusers/pipelines/__init__.py", - "src/diffusers/pipelines/auto_pipeline.py", - "src/diffusers/pipelines/pag/__init__.py", - "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", - "src/diffusers/utils/dummy_torch_and_transformers_objects.py" - ], - "shared_file_areas": [ - { - "filename": "docs/source/en/api/pipelines/pag.md", - "left_ranges": [ - [ - 44, - 51 - ] - ], - "right_ranges": [ - [ - 71, - 81 - ] - ] - }, - { - "filename": "src/diffusers/__init__.py", - "left_ranges": [ - [ - 290, - 296 - ], - [ - 690, - 696 - ] - ], - "right_ranges": [ - [ - 332, - 338 - ], - [ - 790, - 796 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/__init__.py", - "left_ranges": [ - [ - 145, - 151 - ], - [ - 493, - 499 - ] - ], - "right_ranges": [ - [ - 168, - 174 - ], - [ - 572, - 578 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/auto_pipeline.py", - "left_ranges": [ - [ - 47, - 53 - ], - [ - 93, - 99 - ] - ], - "right_ranges": [ - [ - 59, - 65 - ], - [ - 135, - 141 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/pag/__init__.py", - "left_ranges": [ - [ - 22, - 28 - ], - [ - 37, - 43 - ] - ], - "right_ranges": [ - [ - 23, - 29 - ], - [ - 47, - 53 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", - "left_ranges": [ - [ - 1, - 1297 - ] - ], - "right_ranges": [ - [ - 1, - 1291 - ] - ] - }, - { - "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", - "left_ranges": [ - [ - 1247, - 1267 - ] - ], - "right_ranges": [ - [ - 1832, - 1852 - ] - ] - } - ] - }, - { - "left_pr_number": 8810, - "right_pr_number": 9917, - "code_similarity": 0.808, - "size_similarity": 0.773, - "file_overlap": 0.75, - "area_overlap": 0.877, - "patch_similarity": 0.81, - "shared_filenames": [ - "src/diffusers/__init__.py", - "src/diffusers/pipelines/__init__.py", - "src/diffusers/pipelines/auto_pipeline.py", - "src/diffusers/pipelines/pag/__init__.py", - "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", - "src/diffusers/utils/dummy_torch_and_transformers_objects.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/__init__.py", - "left_ranges": [ - [ - 290, - 296 - ], - [ - 690, - 696 - ] - ], - "right_ranges": [ - [ - 340, - 346 - ], - [ - 806, - 812 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/__init__.py", - "left_ranges": [ - [ - 145, - 151 - ], - [ - 493, - 499 - ] - ], - "right_ranges": [ - [ - 169, - 175 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/auto_pipeline.py", - "left_ranges": [ - [ - 47, - 53 - ], - [ - 93, - 99 - ] - ], - "right_ranges": [ - [ - 64, - 70 - ], - [ - 136, - 142 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/pag/__init__.py", - "left_ranges": [ - [ - 22, - 28 - ], - [ - 37, - 43 - ] - ], - "right_ranges": [ - [ - 25, - 31 - ], - [ - 48, - 54 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", - "left_ranges": [ - [ - 1, - 1297 - ] - ], - "right_ranges": [ - [ - 1, - 1342 - ] - ] - }, - { - "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", - "left_ranges": [ - [ - 1247, - 1267 - ] - ], - "right_ranges": [ - [ - 1786, - 1804 - ] - ] - } - ] - }, - { - "left_pr_number": 8864, - "right_pr_number": 9917, - "code_similarity": 0.912, - "size_similarity": 0.977, - "file_overlap": 0.857, - "area_overlap": 0.9, - "patch_similarity": 0.961, - "shared_filenames": [ - "src/diffusers/__init__.py", - "src/diffusers/pipelines/__init__.py", - "src/diffusers/pipelines/auto_pipeline.py", - "src/diffusers/pipelines/pag/__init__.py", - "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", - "src/diffusers/utils/dummy_torch_and_transformers_objects.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/__init__.py", - "left_ranges": [ - [ - 332, - 338 - ], - [ - 790, - 796 - ] - ], - "right_ranges": [ - [ - 340, - 346 - ], - [ - 806, - 812 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/__init__.py", - "left_ranges": [ - [ - 168, - 174 - ], - [ - 572, - 578 - ] - ], - "right_ranges": [ - [ - 169, - 175 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/auto_pipeline.py", - "left_ranges": [ - [ - 59, - 65 - ], - [ - 135, - 141 - ] - ], - "right_ranges": [ - [ - 64, - 70 - ], - [ - 136, - 142 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/pag/__init__.py", - "left_ranges": [ - [ - 23, - 29 - ], - [ - 47, - 53 - ] - ], - "right_ranges": [ - [ - 25, - 31 - ], - [ - 48, - 54 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", - "left_ranges": [ - [ - 1, - 1291 - ] - ], - "right_ranges": [ - [ - 1, - 1342 - ] - ] - }, - { - "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", - "left_ranges": [ - [ - 1832, - 1852 - ] - ], - "right_ranges": [ - [ - 1786, - 1804 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-3880-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #3880.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 3880, - "canonical_pr_number": 4536, - "issue_numbers": [ - 3880, - 4782 - ], - "pr_numbers": [ - 3921, - 4536 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 3921, - "right_pr_number": 4536, - "code_similarity": 0.061, - "size_similarity": 0.229, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.1, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-6137-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #8236.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 8236, - "canonical_pr_number": 8312, - "issue_numbers": [ - 6137, - 8236 - ], - "pr_numbers": [ - 6143, - 8312 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 6143, - "right_pr_number": 8312, - "code_similarity": 0.239, - "size_similarity": 0.493, - "file_overlap": 0.333, - "area_overlap": 0.0, - "patch_similarity": 0.269, - "shared_filenames": [ - "examples/text_to_image/train_text_to_image_lora.py" - ], - "shared_file_areas": [ - { - "filename": "examples/text_to_image/train_text_to_image_lora.py", - "left_ranges": [ - [ - 630, - 638 - ], - [ - 646, - 651 - ] - ], - "right_ranges": [ - [ - 697, - 718 - ], - [ - 722, - 735 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-5249-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #5249.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 5249, - "canonical_pr_number": 5293, - "issue_numbers": [ - 5249 - ], - "pr_numbers": [ - 5293, - 5299, - 5312 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 5293, - "right_pr_number": 5299, - "code_similarity": 0.044, - "size_similarity": 0.098, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.164, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 5293, - "right_pr_number": 5312, - "code_similarity": 0.028, - "size_similarity": 0.078, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.08, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 5299, - "right_pr_number": 5312, - "code_similarity": 0.188, - "size_similarity": 0.795, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.192, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-9237-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #9237.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 9237, - "canonical_pr_number": 9419, - "issue_numbers": [ - 9237, - 9548 - ], - "pr_numbers": [ - 9419, - 9549 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 9419, - "right_pr_number": 9549, - "code_similarity": 0.143, - "size_similarity": 0.144, - "file_overlap": 0.167, - "area_overlap": 0.029, - "patch_similarity": 0.364, - "shared_filenames": [ - "examples/dreambooth/train_dreambooth_lora_flux.py" - ], - "shared_file_areas": [ - { - "filename": "examples/dreambooth/train_dreambooth_lora_flux.py", - "left_ranges": [ - [ - 170, - 183 - ], - [ - 1786, - 1792 - ], - [ - 1834, - 1840 - ] - ], - "right_ranges": [ - [ - 183, - 190 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-11730-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #11730.", - "status": "open", - "confidence": 0.8, - "canonical_issue_number": 11730, - "canonical_pr_number": 11744, - "issue_numbers": [ - 11730 - ], - "pr_numbers": [ - 11744, - 11782, - 11783 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 11744, - "right_pr_number": 11782, - "code_similarity": 0.339, - "size_similarity": 0.963, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.975, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11744, - "right_pr_number": 11783, - "code_similarity": 0.339, - "size_similarity": 0.963, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.975, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11782, - "right_pr_number": 11783, - "code_similarity": 0.964, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 0.896, - "patch_similarity": 1.0, - "shared_filenames": [ - "examples/dreambooth/test_dreambooth_lora_sd3.py", - "examples/dreambooth/train_dreambooth_lora_sd3.py" - ], - "shared_file_areas": [ - { - "filename": "examples/dreambooth/test_dreambooth_lora_sd3.py", - "left_ranges": [ - [ - 13, - 27 - ], - [ - 209, - 254 - ] - ], - "right_ranges": [ - [ - 13, - 27 - ], - [ - 209, - 254 - ] - ] - }, - { - "filename": "examples/dreambooth/train_dreambooth_lora_sd3.py", - "left_ranges": [ - [ - 53, - 59 - ], - [ - 322, - 333 - ], - [ - 1273, - 1279 - ], - [ - 1302, - 1316 - ], - [ - 1333, - 1339 - ], - [ - 1935, - 1946 - ], - [ - 1957, - 1963 - ] - ], - "right_ranges": [ - [ - 54, - 60 - ], - [ - 323, - 334 - ], - [ - 1274, - 1280 - ], - [ - 1303, - 1317 - ], - [ - 1334, - 1340 - ], - [ - 1936, - 1947 - ], - [ - 1958, - 1964 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-6226-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #11285.", - "status": "open", - "confidence": 0.6, - "canonical_issue_number": 11285, - "canonical_pr_number": 12964, - "issue_numbers": [ - 6226, - 11285 - ], - "pr_numbers": [ - 11306, - 12964 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 11306, - "right_pr_number": 12964, - "code_similarity": 0.057, - "size_similarity": 0.214, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.096, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-8384-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #8384.", - "status": "open", - "confidence": 0.8, - "canonical_issue_number": 8384, - "canonical_pr_number": 11239, - "issue_numbers": [ - 8384 - ], - "pr_numbers": [ - 10973, - 11239, - 11240 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 10973, - "right_pr_number": 11239, - "code_similarity": 0.344, - "size_similarity": 0.972, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 10973, - "right_pr_number": 11240, - "code_similarity": 0.35, - "size_similarity": 1.0, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11239, - "right_pr_number": 11240, - "code_similarity": 0.344, - "size_similarity": 0.972, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 1.0, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-13074-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #13074.", - "status": "open", - "confidence": 0.8, - "canonical_issue_number": 13074, - "canonical_pr_number": 13075, - "issue_numbers": [ - 13074 - ], - "pr_numbers": [ - 13075, - 13093, - 13514 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 13075, - "right_pr_number": 13093, - "code_similarity": 0.695, - "size_similarity": 0.85, - "file_overlap": 1.0, - "area_overlap": 0.5, - "patch_similarity": 0.333, - "shared_filenames": [ - "src/diffusers/models/_modeling_parallel.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/_modeling_parallel.py", - "left_ranges": [ - [ - 22, - 28 - ], - [ - 291, - 304 - ] - ], - "right_ranges": [ - [ - 290, - 301 - ] - ] - } - ] - }, - { - "left_pr_number": 13075, - "right_pr_number": 13514, - "code_similarity": 0.593, - "size_similarity": 0.95, - "file_overlap": 1.0, - "area_overlap": 0.207, - "patch_similarity": 0.205, - "shared_filenames": [ - "src/diffusers/models/_modeling_parallel.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/_modeling_parallel.py", - "left_ranges": [ - [ - 22, - 28 - ], - [ - 291, - 304 - ] - ], - "right_ranges": [ - [ - 299, - 312 - ] - ] - } - ] - }, - { - "left_pr_number": 13093, - "right_pr_number": 13514, - "code_similarity": 0.576, - "size_similarity": 0.895, - "file_overlap": 1.0, - "area_overlap": 0.13, - "patch_similarity": 0.345, - "shared_filenames": [ - "src/diffusers/models/_modeling_parallel.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/_modeling_parallel.py", - "left_ranges": [ - [ - 290, - 301 - ] - ], - "right_ranges": [ - [ - 299, - 312 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-6514-4", - "summary": "Cluster of 4 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 6514, - "issue_numbers": [], - "pr_numbers": [ - 6514, - 6554, - 6840, - 7242 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 6514, - "right_pr_number": 6554, - "code_similarity": 0.182, - "size_similarity": 0.444, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.623, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6514, - "right_pr_number": 6840, - "code_similarity": 0.161, - "size_similarity": 0.378, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.569, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6514, - "right_pr_number": 7242, - "code_similarity": 0.118, - "size_similarity": 0.089, - "file_overlap": 0.333, - "area_overlap": 0.0, - "patch_similarity": 0.0, + "left_pr_number": 9606, + "right_pr_number": 9724, + "code_similarity": 0.426, + "size_similarity": 0.54, + "file_overlap": 1.0, + "area_overlap": 0.017, + "patch_similarity": 0.081, "shared_filenames": [ - "examples/dreambooth/train_dreambooth_lora_sdxl.py" + "src/diffusers/training_utils.py" ], "shared_file_areas": [ { - "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "filename": "src/diffusers/training_utils.py", "left_ranges": [ [ - 34, - 40 + 36, + 44 ], [ - 53, - 65 + 195, + 207 ], [ - 1002, - 1007 + 233, + 240 ], [ - 1058, - 1096 + 253, + 260 ], [ - 1105, - 1121 + 271, + 279 ], [ - 1533, - 1539 + 506, + 513 + ], + [ + 541, + 565 + ], + [ + 575, + 584 ] ], "right_ranges": [ [ - 877, - 884 + 43, + 51 + ], + [ + 59, + 76 ] ] } ] }, { - "left_pr_number": 6554, - "right_pr_number": 6840, - "code_similarity": 0.297, - "size_similarity": 0.85, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.847, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6554, - "right_pr_number": 7242, - "code_similarity": 0.04, - "size_similarity": 0.2, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.0, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 6840, - "right_pr_number": 7242, - "code_similarity": 0.047, - "size_similarity": 0.235, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.0, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-12771-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #12771.", - "status": "open", - "confidence": 0.8, - "canonical_issue_number": 12771, - "canonical_pr_number": 12788, - "issue_numbers": [ - 12771 - ], - "pr_numbers": [ - 12788, - 13239, - 13240 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 12788, - "right_pr_number": 13239, - "code_similarity": 0.238, - "size_similarity": 0.24, - "file_overlap": 0.5, - "area_overlap": 0.038, - "patch_similarity": 0.178, + "left_pr_number": 9606, + "right_pr_number": 9815, + "code_similarity": 0.413, + "size_similarity": 0.48, + "file_overlap": 1.0, + "area_overlap": 0.018, + "patch_similarity": 0.071, "shared_filenames": [ - "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" + "src/diffusers/training_utils.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", + "filename": "src/diffusers/training_utils.py", "left_ranges": [ [ - 24, - 31 + 36, + 44 ], [ - 412, - 445 + 195, + 207 ], [ - 550, - 587 - ] - ], - "right_ranges": [ + 233, + 240 + ], [ - 442, - 452 + 253, + 260 ], [ - 472, - 488 - ] - ] - } - ] - }, - { - "left_pr_number": 12788, - "right_pr_number": 13240, - "code_similarity": 0.236, - "size_similarity": 0.25, - "file_overlap": 0.5, - "area_overlap": 0.038, - "patch_similarity": 0.15, - "shared_filenames": [ - "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", - "left_ranges": [ + 271, + 279 + ], [ - 24, - 31 + 506, + 513 ], [ - 412, - 445 + 541, + 565 ], [ - 550, - 587 + 575, + 584 ] ], "right_ranges": [ [ - 442, - 454 + 43, + 51 ], [ - 465, - 480 + 61, + 77 ] ] } ] }, { - "left_pr_number": 13239, - "right_pr_number": 13240, - "code_similarity": 0.756, - "size_similarity": 0.962, + "left_pr_number": 9608, + "right_pr_number": 9724, + "code_similarity": 0.026, + "size_similarity": 0.085, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.061, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9608, + "right_pr_number": 9815, + "code_similarity": 0.025, + "size_similarity": 0.075, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.067, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9724, + "right_pr_number": 9815, + "code_similarity": 0.884, + "size_similarity": 0.889, "file_overlap": 1.0, - "area_overlap": 0.541, - "patch_similarity": 0.5, + "area_overlap": 0.893, + "patch_similarity": 0.628, "shared_filenames": [ - "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" + "src/diffusers/training_utils.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", + "filename": "src/diffusers/training_utils.py", "left_ranges": [ [ - 442, - 452 + 43, + 51 ], [ - 472, - 488 + 59, + 76 ] ], "right_ranges": [ [ - 442, - 454 + 43, + 51 ], [ - 465, - 480 + 61, + 77 ] ] } @@ -4875,115 +1266,297 @@ ] }, { - "cluster_id": "cluster-13077-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #13077.", - "status": "open", - "confidence": 0.75, - "canonical_issue_number": 13077, - "canonical_pr_number": 13136, + "cluster_id": "cluster-6984-7", + "summary": "Cluster of 1 issues and 6 PRs centered on issue #6984.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 6984, + "canonical_pr_number": 7071, "issue_numbers": [ - 13077 + 6984 ], "pr_numbers": [ - 13102, - 13128, - 13136 + 7052, + 7071, + 7211, + 7307, + 8376, + 8387 ], "evidence_types": [ "closing_reference", - "shared_issue_target" + "shared_issue_target", + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 13102, - "right_pr_number": 13128, - "code_similarity": 0.27, - "size_similarity": 0.102, + "left_pr_number": 7052, + "right_pr_number": 7071, + "code_similarity": 0.063, + "size_similarity": 0.139, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.233, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7052, + "right_pr_number": 7211, + "code_similarity": 0.069, + "size_similarity": 0.185, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.216, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7052, + "right_pr_number": 7307, + "code_similarity": 0.403, + "size_similarity": 0.269, "file_overlap": 0.5, - "area_overlap": 0.253, - "patch_similarity": 0.074, + "area_overlap": 0.453, + "patch_similarity": 0.274, "shared_filenames": [ - "src/diffusers/models/controlnets/controlnet_z_image.py" + "examples/community/stable_diffusion_tensorrt_txt2img.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", + "filename": "examples/community/stable_diffusion_tensorrt_txt2img.py", "left_ranges": [ [ - 517, - 535 - ] - ], - "right_ranges": [ + 19, + 28 + ], [ - 517, - 591 - ] - ] - } - ] - }, - { - "left_pr_number": 13102, - "right_pr_number": 13136, - "code_similarity": 0.693, - "size_similarity": 0.9, - "file_overlap": 1.0, - "area_overlap": 0.531, - "patch_similarity": 0.182, - "shared_filenames": [ - "src/diffusers/models/controlnets/controlnet_z_image.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", - "left_ranges": [ + 41, + 58 + ], [ - 517, - 535 + 592, + 602 + ], + [ + 678, + 685 + ], + [ + 697, + 771 + ], + [ + 1008, + 1013 ] ], "right_ranges": [ [ - 517, - 528 + 19, + 36 ], [ - 531, - 548 + 45, + 62 + ], + [ + 594, + 606 + ], + [ + 683, + 780 + ], + [ + 791, + 865 + ], + [ + 1102, + 1107 ] ] } ] }, { - "left_pr_number": 13128, - "right_pr_number": 13136, - "code_similarity": 0.321, - "size_similarity": 0.092, - "file_overlap": 0.5, - "area_overlap": 0.4, - "patch_similarity": 0.083, + "left_pr_number": 7052, + "right_pr_number": 8376, + "code_similarity": 0.088, + "size_similarity": 0.299, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.187, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7052, + "right_pr_number": 8387, + "code_similarity": 0.088, + "size_similarity": 0.299, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.187, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7071, + "right_pr_number": 7211, + "code_similarity": 0.274, + "size_similarity": 0.754, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.818, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7071, + "right_pr_number": 7307, + "code_similarity": 0.156, + "size_similarity": 0.517, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.352, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7071, + "right_pr_number": 8376, + "code_similarity": 0.186, + "size_similarity": 0.466, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.619, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7071, + "right_pr_number": 8387, + "code_similarity": 0.186, + "size_similarity": 0.466, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.619, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7211, + "right_pr_number": 7307, + "code_similarity": 0.186, + "size_similarity": 0.685, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.326, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7211, + "right_pr_number": 8376, + "code_similarity": 0.223, + "size_similarity": 0.618, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.665, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7211, + "right_pr_number": 8387, + "code_similarity": 0.223, + "size_similarity": 0.618, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.665, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7307, + "right_pr_number": 8376, + "code_similarity": 0.231, + "size_similarity": 0.901, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.341, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7307, + "right_pr_number": 8387, + "code_similarity": 0.231, + "size_similarity": 0.901, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.341, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8376, + "right_pr_number": 8387, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, "shared_filenames": [ - "src/diffusers/models/controlnets/controlnet_z_image.py" + "examples/community/regional_prompting_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", + "filename": "examples/community/regional_prompting_stable_diffusion.py", "left_ranges": [ [ - 517, - 591 + 1, + 28 + ], + [ + 33, + 41 + ], + [ + 83, + 182 + ], + [ + 186, + 376 + ], + [ + 399, + 405 ] ], "right_ranges": [ [ - 517, - 528 + 1, + 28 ], [ - 531, - 548 + 33, + 41 + ], + [ + 83, + 182 + ], + [ + 186, + 376 + ], + [ + 399, + 405 ] ] } @@ -4992,118 +1565,347 @@ ] }, { - "cluster_id": "cluster-1257-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #1257.", + "cluster_id": "cluster-11390-7", + "summary": "Cluster of 1 issues and 6 PRs centered on issue #11390.", "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 1257, - "canonical_pr_number": 1361, + "confidence": 0.8, + "canonical_issue_number": 11390, + "canonical_pr_number": 11463, "issue_numbers": [ - 1257, - 2950 + 11390 ], "pr_numbers": [ - 1361, - 3000 + 11395, + 11426, + 11427, + 11463, + 11491, + 11492 ], "evidence_types": [ "closing_reference", + "shared_issue_target", "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 1361, - "right_pr_number": 3000, - "code_similarity": 0.018, - "size_similarity": 0.041, + "left_pr_number": 11395, + "right_pr_number": 11426, + "code_similarity": 0.271, + "size_similarity": 0.8, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.063, + "patch_similarity": 0.739, "shared_filenames": [], "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-1407-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #1792.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 1792, - "canonical_pr_number": 1416, - "issue_numbers": [ - 1407, - 1792 - ], - "pr_numbers": [ - 1416, - 1796 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 1416, - "right_pr_number": 1796, - "code_similarity": 0.149, - "size_similarity": 0.691, + "left_pr_number": 11395, + "right_pr_number": 11427, + "code_similarity": 0.271, + "size_similarity": 0.8, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.739, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11395, + "right_pr_number": 11463, + "code_similarity": 0.179, + "size_similarity": 0.444, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.604, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11395, + "right_pr_number": 11491, + "code_similarity": 0.328, + "size_similarity": 0.889, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11395, + "right_pr_number": 11492, + "code_similarity": 0.338, + "size_similarity": 0.96, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.972, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11426, + "right_pr_number": 11427, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11426, + "right_pr_number": 11463, + "code_similarity": 0.194, + "size_similarity": 0.556, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.55, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11426, + "right_pr_number": 11491, + "code_similarity": 0.291, + "size_similarity": 0.9, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.739, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11426, + "right_pr_number": 11492, + "code_similarity": 0.28, + "size_similarity": 0.833, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.756, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11427, + "right_pr_number": 11463, + "code_similarity": 0.194, + "size_similarity": 0.556, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.55, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11427, + "right_pr_number": 11491, + "code_similarity": 0.291, + "size_similarity": 0.9, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.739, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11427, + "right_pr_number": 11492, + "code_similarity": 0.28, + "size_similarity": 0.833, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.756, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11463, + "right_pr_number": 11491, + "code_similarity": 0.191, + "size_similarity": 0.5, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.604, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11463, + "right_pr_number": 11492, + "code_similarity": 0.185, + "size_similarity": 0.463, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.615, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11491, + "right_pr_number": 11492, + "code_similarity": 0.331, + "size_similarity": 0.926, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.071, + "patch_similarity": 0.972, "shared_filenames": [], "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-11315-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #11315.", + "cluster_id": "cluster-6503-6", + "summary": "Cluster of 1 issues and 5 PRs centered on issue #6503.", "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 11315, - "canonical_pr_number": 12665, + "confidence": 0.8, + "canonical_issue_number": 6503, + "canonical_pr_number": 6532, "issue_numbers": [ - 11315, - 12653 + 6503 ], "pr_numbers": [ - 11316, - 12665 + 6525, + 6532, + 6534, + 6555, + 6556 ], "evidence_types": [ "closing_reference", + "shared_issue_target", "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 11316, - "right_pr_number": 12665, - "code_similarity": 0.35, - "size_similarity": 1.0, + "left_pr_number": 6525, + "right_pr_number": 6532, + "code_similarity": 0.172, + "size_similarity": 0.549, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 1.0, + "patch_similarity": 0.417, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6525, + "right_pr_number": 6534, + "code_similarity": 0.18, + "size_similarity": 0.651, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.333, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6525, + "right_pr_number": 6555, + "code_similarity": 0.252, + "size_similarity": 0.929, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.444, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6525, + "right_pr_number": 6556, + "code_similarity": 0.149, + "size_similarity": 0.528, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.288, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6532, + "right_pr_number": 6534, + "code_similarity": 0.264, + "size_similarity": 0.843, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.638, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6532, + "right_pr_number": 6555, + "code_similarity": 0.162, + "size_similarity": 0.51, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.4, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6532, + "right_pr_number": 6556, + "code_similarity": 0.252, + "size_similarity": 0.962, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.397, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6534, + "right_pr_number": 6555, + "code_similarity": 0.202, + "size_similarity": 0.605, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.541, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6534, + "right_pr_number": 6556, + "code_similarity": 0.244, + "size_similarity": 0.811, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.545, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6555, + "right_pr_number": 6556, + "code_similarity": 0.166, + "size_similarity": 0.491, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.452, "shared_filenames": [], "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-1466-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #1466.", + "cluster_id": "cluster-2098-6", + "summary": "Cluster of 3 issues and 3 PRs centered on issue #3008.", "status": "merged", "confidence": 0.6, - "canonical_issue_number": 1466, - "canonical_pr_number": 2070, + "canonical_issue_number": 3008, + "canonical_pr_number": 3197, "issue_numbers": [ - 1466, - 2041 + 2098, + 3008, + 3184 ], "pr_numbers": [ - 1547, - 2070 + 3197, + 3201, + 3265 ], "evidence_types": [ "closing_reference", @@ -5111,1085 +1913,1332 @@ ], "pr_comparisons": [ { - "left_pr_number": 1547, - "right_pr_number": 2070, - "code_similarity": 0.211, - "size_similarity": 0.636, - "file_overlap": 0.2, - "area_overlap": 0.0, - "patch_similarity": 0.155, + "left_pr_number": 3197, + "right_pr_number": 3201, + "code_similarity": 0.79, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 0.401, + "patch_similarity": 1.0, "shared_filenames": [ - "examples/dreambooth/train_dreambooth.py" + "examples/community/lpw_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "examples/dreambooth/train_dreambooth.py", + "filename": "examples/community/lpw_stable_diffusion.py", "left_ranges": [ [ - 300, - 309 + 1, + 30 + ], + [ + 135, + 141 + ], + [ + 196, + 202 + ], + [ + 236, + 242 + ], + [ + 252, + 258 + ], + [ + 340, + 354 + ], + [ + 368, + 417 + ], + [ + 440, + 635 + ], + [ + 646, + 655 + ], + [ + 669, + 739 + ], + [ + 742, + 783 + ], + [ + 791, + 797 + ], + [ + 815, + 865 + ], + [ + 875, + 893 + ], + [ + 927, + 952 + ], + [ + 964, + 973 + ], + [ + 982, + 1028 + ], + [ + 1035, + 1043 + ], + [ + 1050, + 1119 + ], + [ + 1130, + 1146 + ], + [ + 1168, + 1187 + ], + [ + 1199, + 1211 + ], + [ + 1223, + 1237 + ], + [ + 1244, + 1259 + ], + [ + 1286, + 1301 + ], + [ + 1313, + 1325 + ], + [ + 1335, + 1349 + ], + [ + 1356, + 1373 + ], + [ + 1401, + 1422 + ], + [ + 1434, + 1446 + ], + [ + 1455, + 1470 + ] + ], + "right_ranges": [ + [ + 1, + 39 + ], + [ + 144, + 150 + ], + [ + 205, + 211 + ], + [ + 245, + 251 + ], + [ + 261, + 267 ], [ - 313, - 349 + 349, + 363 ], [ - 534, - 545 - ] - ], - "right_ranges": [ + 377, + 406 + ], [ - 240, - 253 + 429, + 494 ], [ - 660, - 666 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-11702-4", - "summary": "Cluster of 2 issues and 2 PRs centered on issue #11709.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 11709, - "canonical_pr_number": 11704, - "issue_numbers": [ - 11702, - 11709 - ], - "pr_numbers": [ - 11704, - 11726 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 11704, - "right_pr_number": 11726, - "code_similarity": 0.545, - "size_similarity": 0.711, - "file_overlap": 1.0, - "area_overlap": 0.215, - "patch_similarity": 0.183, - "shared_filenames": [ - "src/diffusers/loaders/lora_conversion_utils.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/loaders/lora_conversion_utils.py", - "left_ranges": [ + 505, + 512 + ], [ - 1605, - 1622 + 526, + 572 ], [ - 1664, - 1679 + 575, + 591 ], [ - 1682, - 1697 + 599, + 605 ], [ - 1726, - 1745 - ] - ], - "right_ranges": [ + 623, + 665 + ], [ - 1596, - 1605 + 675, + 689 ], [ - 1625, - 1681 + 723, + 738 ], [ - 1689, - 1698 + 750, + 755 + ], + [ + 764, + 800 + ], + [ + 807, + 813 + ], + [ + 820, + 862 + ], + [ + 873, + 886 + ], + [ + 908, + 920 + ], + [ + 932, + 938 + ], + [ + 950, + 961 + ], + [ + 968, + 980 + ], + [ + 1007, + 1015 + ], + [ + 1027, + 1034 + ], + [ + 1044, + 1055 + ], + [ + 1062, + 1075 + ], + [ + 1103, + 1114 + ], + [ + 1126, + 1133 + ], + [ + 1142, + 1153 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-6866-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #6866.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 6866, - "canonical_pr_number": 6941, - "issue_numbers": [ - 6866 - ], - "pr_numbers": [ - 6941, - 7030, - 7031 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 6941, - "right_pr_number": 7030, - "code_similarity": 0.345, - "size_similarity": 0.139, + "left_pr_number": 3197, + "right_pr_number": 3265, + "code_similarity": 1.0, + "size_similarity": 1.0, "file_overlap": 1.0, - "area_overlap": 0.039, - "patch_similarity": 0.027, + "area_overlap": 1.0, + "patch_similarity": 1.0, "shared_filenames": [ - "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" + "examples/community/lpw_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", + "filename": "examples/community/lpw_stable_diffusion.py", "left_ranges": [ [ - 19, - 35 + 1, + 30 + ], + [ + 135, + 141 + ], + [ + 196, + 202 + ], + [ + 236, + 242 + ], + [ + 252, + 258 + ], + [ + 340, + 354 + ], + [ + 368, + 417 + ], + [ + 440, + 635 + ], + [ + 646, + 655 + ], + [ + 669, + 739 + ], + [ + 742, + 783 + ], + [ + 791, + 797 + ], + [ + 815, + 865 + ], + [ + 875, + 893 + ], + [ + 927, + 952 + ], + [ + 964, + 973 + ], + [ + 982, + 1028 + ], + [ + 1035, + 1043 + ], + [ + 1050, + 1119 + ], + [ + 1130, + 1146 + ], + [ + 1168, + 1187 + ], + [ + 1199, + 1211 + ], + [ + 1223, + 1237 + ], + [ + 1244, + 1259 + ], + [ + 1286, + 1301 + ], + [ + 1313, + 1325 + ], + [ + 1335, + 1349 + ], + [ + 1356, + 1373 + ], + [ + 1401, + 1422 + ], + [ + 1434, + 1446 + ], + [ + 1455, + 1470 + ] + ], + "right_ranges": [ + [ + 1, + 30 ], [ - 201, - 208 + 135, + 141 ], [ - 218, - 225 + 196, + 202 ], [ - 507, - 572 + 236, + 242 ], [ - 636, - 643 + 252, + 258 ], [ - 824, - 834 + 340, + 354 ], [ - 1177, - 1184 + 368, + 417 ], [ - 1273, - 1282 + 440, + 635 ], [ - 1409, - 1416 + 646, + 655 ], [ - 1463, - 1474 + 669, + 739 ], [ - 1740, - 1748 - ] - ], - "right_ranges": [ + 742, + 783 + ], [ - 28, - 34 + 791, + 797 ], [ - 146, - 152 + 815, + 865 ], [ - 158, - 164 - ] - ] - } - ] - }, - { - "left_pr_number": 6941, - "right_pr_number": 7031, - "code_similarity": 0.356, - "size_similarity": 0.185, - "file_overlap": 1.0, - "area_overlap": 0.044, - "patch_similarity": 0.027, - "shared_filenames": [ - "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", - "left_ranges": [ + 875, + 893 + ], [ - 19, - 35 + 927, + 952 ], [ - 201, - 208 + 964, + 973 ], [ - 218, - 225 + 982, + 1028 ], [ - 507, - 572 + 1035, + 1043 ], [ - 636, - 643 + 1050, + 1119 ], [ - 824, - 834 + 1130, + 1146 ], [ - 1177, - 1184 + 1168, + 1187 ], [ - 1273, - 1282 + 1199, + 1211 ], [ - 1409, - 1416 + 1223, + 1237 ], [ - 1463, - 1474 + 1244, + 1259 ], [ - 1740, - 1748 - ] - ], - "right_ranges": [ + 1286, + 1301 + ], [ - 28, - 39 + 1313, + 1325 ], [ - 151, - 157 + 1335, + 1349 ], [ - 163, - 169 + 1356, + 1373 + ], + [ + 1401, + 1422 + ], + [ + 1434, + 1446 + ], + [ + 1455, + 1470 ] ] } ] }, { - "left_pr_number": 7030, - "right_pr_number": 7031, - "code_similarity": 0.707, - "size_similarity": 0.75, + "left_pr_number": 3201, + "right_pr_number": 3265, + "code_similarity": 0.79, + "size_similarity": 1.0, "file_overlap": 1.0, - "area_overlap": 0.306, + "area_overlap": 0.401, "patch_similarity": 1.0, "shared_filenames": [ - "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" + "examples/community/lpw_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", + "filename": "examples/community/lpw_stable_diffusion.py", "left_ranges": [ [ - 28, - 34 + 1, + 39 ], [ - 146, - 152 + 144, + 150 ], [ - 158, - 164 - ] - ], - "right_ranges": [ + 205, + 211 + ], [ - 28, - 39 + 245, + 251 ], [ - 151, - 157 + 261, + 267 ], [ - 163, - 169 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-5262-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #5262.", - "status": "merged", - "confidence": 0.75, - "canonical_issue_number": 5262, - "canonical_pr_number": 5370, - "issue_numbers": [ - 5262 - ], - "pr_numbers": [ - 5288, - 5366, - 5370 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target" - ], - "pr_comparisons": [ - { - "left_pr_number": 5288, - "right_pr_number": 5366, - "code_similarity": 0.344, - "size_similarity": 0.222, - "file_overlap": 1.0, - "area_overlap": 0.0, - "patch_similarity": 0.0, - "shared_filenames": [ - "tests/pipelines/stable_diffusion/test_stable_diffusion.py" - ], - "shared_file_areas": [ - { - "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", - "left_ranges": [ + 349, + 363 + ], [ - 90, - 97 - ] - ], - "right_ranges": [ + 377, + 406 + ], [ - 103, - 129 + 429, + 494 ], [ - 134, - 170 - ] - ] - } - ] - }, - { - "left_pr_number": 5288, - "right_pr_number": 5370, - "code_similarity": 0.356, - "size_similarity": 0.279, - "file_overlap": 1.0, - "area_overlap": 0.0, - "patch_similarity": 0.0, - "shared_filenames": [ - "tests/pipelines/stable_diffusion/test_stable_diffusion.py" - ], - "shared_file_areas": [ - { - "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", - "left_ranges": [ + 505, + 512 + ], + [ + 526, + 572 + ], + [ + 575, + 591 + ], + [ + 599, + 605 + ], + [ + 623, + 665 + ], + [ + 675, + 689 + ], + [ + 723, + 738 + ], + [ + 750, + 755 + ], [ - 90, - 97 - ] - ], - "right_ranges": [ + 764, + 800 + ], [ - 82, - 88 + 807, + 813 ], [ - 104, - 118 + 820, + 862 ], [ - 123, - 145 + 873, + 886 ], [ - 186, - 192 + 908, + 920 ], [ - 320, - 326 + 932, + 938 ], [ - 338, - 344 + 950, + 961 ], [ - 378, - 384 + 968, + 980 ], [ - 397, - 403 + 1007, + 1015 ], [ - 416, - 422 + 1027, + 1034 ], [ - 488, - 494 + 1044, + 1055 ], [ - 641, - 647 + 1062, + 1075 ], [ - 668, - 674 - ] - ] - } - ] - }, - { - "left_pr_number": 5366, - "right_pr_number": 5370, - "code_similarity": 0.559, - "size_similarity": 0.796, - "file_overlap": 1.0, - "area_overlap": 0.246, - "patch_similarity": 0.088, - "shared_filenames": [ - "tests/pipelines/stable_diffusion/test_stable_diffusion.py" - ], - "shared_file_areas": [ - { - "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", - "left_ranges": [ + 1103, + 1114 + ], [ - 103, - 129 + 1126, + 1133 ], [ - 134, - 170 + 1142, + 1153 ] ], "right_ranges": [ [ - 82, - 88 + 1, + 30 ], [ - 104, - 118 + 135, + 141 ], [ - 123, - 145 + 196, + 202 ], [ - 186, - 192 + 236, + 242 ], [ - 320, - 326 + 252, + 258 ], [ - 338, - 344 + 340, + 354 ], [ - 378, - 384 + 368, + 417 ], [ - 397, - 403 + 440, + 635 ], [ - 416, - 422 + 646, + 655 ], [ - 488, - 494 + 669, + 739 ], [ - 641, - 647 + 742, + 783 ], [ - 668, - 674 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-10357-4", - "summary": "Cluster of 1 issues and 3 PRs centered on issue #10357.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 10357, - "canonical_pr_number": 10498, - "issue_numbers": [ - 10357 - ], - "pr_numbers": [ - 10364, - 10365, - 10498 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 10364, - "right_pr_number": 10365, - "code_similarity": 0.309, - "size_similarity": 0.955, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.786, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 10364, - "right_pr_number": 10498, - "code_similarity": 0.08, - "size_similarity": 0.008, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.524, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 10365, - "right_pr_number": 10498, - "code_similarity": 0.102, - "size_similarity": 0.009, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.667, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-916-4", - "summary": "Cluster of 4 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 1110, - "issue_numbers": [], - "pr_numbers": [ - 916, - 1007, - 1110, - 1225 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 916, - "right_pr_number": 1007, - "code_similarity": 0.334, - "size_similarity": 0.238, - "file_overlap": 0.833, - "area_overlap": 0.065, - "patch_similarity": 0.091, - "shared_filenames": [ - "examples/unconditional_image_generation/train_unconditional.py", - "src/diffusers/models/unet_2d.py", - "src/diffusers/pipeline_utils.py", - "src/diffusers/pipelines/ddpm/pipeline_ddpm.py", - "tests/test_models_unet.py" - ], - "shared_file_areas": [ - { - "filename": "examples/unconditional_image_generation/train_unconditional.py", - "left_ranges": [ - [ - 138, - 144 - ] - ], - "right_ranges": [ - [ - 59, - 69 + 791, + 797 ], [ - 144, - 150 + 815, + 865 ], [ - 231, - 237 - ] - ] - }, - { - "filename": "src/diffusers/models/unet_2d.py", - "left_ranges": [ - [ - 258, - 261 - ] - ], - "right_ranges": [ - [ - 245, - 248 - ] - ] - }, - { - "filename": "src/diffusers/pipeline_utils.py", - "left_ranges": [ + 875, + 893 + ], [ - 72, - 80 - ] - ], - "right_ranges": [ + 927, + 952 + ], [ - 53, - 61 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/ddpm/pipeline_ddpm.py", - "left_ranges": [ + 964, + 973 + ], [ - 77, - 83 - ] - ], - "right_ranges": [ + 982, + 1028 + ], [ - 90, - 96 - ] - ] - }, - { - "filename": "tests/test_models_unet.py", - "left_ranges": [ + 1035, + 1043 + ], [ - 131, - 137 + 1050, + 1119 ], [ - 141, - 147 + 1130, + 1146 ], [ - 219, - 225 + 1168, + 1187 ], [ - 294, - 300 + 1199, + 1211 ], [ - 412, - 418 + 1223, + 1237 ], [ - 437, - 443 - ] - ], - "right_ranges": [ + 1244, + 1259 + ], [ - 77, - 83 + 1286, + 1301 ], [ - 129, - 135 + 1313, + 1325 ], [ - 149, - 155 + 1335, + 1349 ], [ - 202, - 208 + 1356, + 1373 ], [ - 216, - 222 + 1401, + 1422 ], [ - 338, - 344 + 1434, + 1446 ], [ - 363, - 369 + 1455, + 1470 ] ] } ] - }, + } + ] + }, + { + "cluster_id": "cluster-9514-5", + "summary": "Cluster of 3 issues and 2 PRs centered on issue #10989.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 10989, + "canonical_pr_number": 10990, + "issue_numbers": [ + 9514, + 10337, + 10989 + ], + "pr_numbers": [ + 10338, + 10990 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ { - "left_pr_number": 916, - "right_pr_number": 1110, - "code_similarity": 0.734, - "size_similarity": 0.919, - "file_overlap": 0.286, - "area_overlap": 0.955, - "patch_similarity": 0.872, + "left_pr_number": 10338, + "right_pr_number": 10990, + "code_similarity": 0.016, + "size_similarity": 0.076, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.005, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-9237-5", + "summary": "Cluster of 3 issues and 2 PRs centered on issue #9237.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 9237, + "canonical_pr_number": 9419, + "issue_numbers": [ + 9237, + 9497, + 9548 + ], + "pr_numbers": [ + 9419, + 9549 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9419, + "right_pr_number": 9549, + "code_similarity": 0.143, + "size_similarity": 0.144, + "file_overlap": 0.167, + "area_overlap": 0.029, + "patch_similarity": 0.364, "shared_filenames": [ - "examples/unconditional_image_generation/train_unconditional_ort.py", - "src/diffusers/pipeline_utils.py" + "examples/dreambooth/train_dreambooth_lora_flux.py" ], "shared_file_areas": [ { - "filename": "examples/unconditional_image_generation/train_unconditional_ort.py", + "filename": "examples/dreambooth/train_dreambooth_lora_flux.py", "left_ranges": [ [ - 1, - 251 + 170, + 183 + ], + [ + 1786, + 1792 + ], + [ + 1834, + 1840 ] ], "right_ranges": [ [ - 1, - 251 + 183, + 190 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-9591-5", + "summary": "Cluster of 5 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 9591, + "issue_numbers": [], + "pr_numbers": [ + 9591, + 9706, + 9707, + 9714, + 9715 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9591, + "right_pr_number": 9706, + "code_similarity": 0.099, + "size_similarity": 0.422, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.097, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9591, + "right_pr_number": 9707, + "code_similarity": 0.099, + "size_similarity": 0.422, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.097, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9591, + "right_pr_number": 9714, + "code_similarity": 0.099, + "size_similarity": 0.422, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.097, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9591, + "right_pr_number": 9715, + "code_similarity": 0.099, + "size_similarity": 0.422, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.097, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9706, + "right_pr_number": 9707, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipeline_utils.py", + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", "left_ranges": [ [ - 72, - 80 + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 ] ], "right_ranges": [ [ - 78, - 86 + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 ] ] } ] }, { - "left_pr_number": 916, - "right_pr_number": 1225, - "code_similarity": 0.137, - "size_similarity": 0.102, - "file_overlap": 0.333, - "area_overlap": 0.019, - "patch_similarity": 0.063, + "left_pr_number": 9706, + "right_pr_number": 9714, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, "shared_filenames": [ - "examples/unconditional_image_generation/train_unconditional.py", - "src/diffusers/pipeline_utils.py" + "src/diffusers/models/transformers/auraflow_transformer_2d.py" ], "shared_file_areas": [ { - "filename": "examples/unconditional_image_generation/train_unconditional.py", + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", "left_ranges": [ [ - 138, - 144 - ] - ], - "right_ranges": [ + 187, + 193 + ], [ - 59, - 69 + 257, + 277 ], [ - 144, - 153 + 347, + 355 ], [ - 245, - 251 - ] - ] - }, - { - "filename": "src/diffusers/pipeline_utils.py", - "left_ranges": [ + 373, + 385 + ], [ - 72, - 80 + 406, + 415 ] ], "right_ranges": [ [ - 53, - 61 + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 ] ] } ] }, { - "left_pr_number": 1007, - "right_pr_number": 1110, - "code_similarity": 0.103, - "size_similarity": 0.259, - "file_overlap": 0.143, - "area_overlap": 0.0, - "patch_similarity": 0.057, + "left_pr_number": 9706, + "right_pr_number": 9715, + "code_similarity": 0.996, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 0.975, "shared_filenames": [ - "src/diffusers/pipeline_utils.py" + "src/diffusers/models/transformers/auraflow_transformer_2d.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipeline_utils.py", + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", "left_ranges": [ [ - 53, - 61 + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 ] ], "right_ranges": [ [ - 78, - 86 + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 ] ] } ] }, { - "left_pr_number": 1007, - "right_pr_number": 1225, - "code_similarity": 0.494, - "size_similarity": 0.43, - "file_overlap": 0.4, - "area_overlap": 0.614, - "patch_similarity": 0.486, + "left_pr_number": 9707, + "right_pr_number": 9714, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, "shared_filenames": [ - "examples/unconditional_image_generation/train_unconditional.py", - "src/diffusers/pipeline_utils.py" + "src/diffusers/models/transformers/auraflow_transformer_2d.py" ], "shared_file_areas": [ { - "filename": "examples/unconditional_image_generation/train_unconditional.py", + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", "left_ranges": [ [ - 59, - 69 + 187, + 193 ], [ - 144, - 150 + 257, + 277 ], [ - 231, - 237 + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 ] ], "right_ranges": [ [ - 59, - 69 + 187, + 193 ], [ - 144, - 153 + 257, + 277 ], [ - 245, - 251 - ] - ] - }, - { - "filename": "src/diffusers/pipeline_utils.py", - "left_ranges": [ + 347, + 355 + ], [ - 53, - 61 - ] - ], - "right_ranges": [ + 373, + 385 + ], [ - 53, - 61 + 406, + 415 ] ] } ] }, { - "left_pr_number": 1110, - "right_pr_number": 1225, - "code_similarity": 0.106, - "size_similarity": 0.111, - "file_overlap": 0.25, - "area_overlap": 0.0, - "patch_similarity": 0.059, + "left_pr_number": 9707, + "right_pr_number": 9715, + "code_similarity": 0.996, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 0.975, "shared_filenames": [ - "src/diffusers/pipeline_utils.py" + "src/diffusers/models/transformers/auraflow_transformer_2d.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipeline_utils.py", + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", "left_ranges": [ [ - 78, - 86 + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 ] ], "right_ranges": [ [ - 53, - 61 + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-6123-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #6123.", - "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 6123, - "canonical_pr_number": 6328, - "issue_numbers": [ - 6123 - ], - "pr_numbers": [ - 6328, - 6509 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ - { - "left_pr_number": 6328, - "right_pr_number": 6509, - "code_similarity": 0.241, - "size_similarity": 0.686, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.693, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-7736-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #7736.", - "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 7736, - "canonical_pr_number": 7761, - "issue_numbers": [ - 7736 - ], - "pr_numbers": [ - 7739, - 7761 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 7739, - "right_pr_number": 7761, - "code_similarity": 0.155, - "size_similarity": 0.462, - "file_overlap": 0.05, - "area_overlap": 0.071, - "patch_similarity": 0.15, + "left_pr_number": 9714, + "right_pr_number": 9715, + "code_similarity": 0.996, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 0.975, "shared_filenames": [ - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py" + "src/diffusers/models/transformers/auraflow_transformer_2d.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", "left_ranges": [ [ - 19, - 25 + 187, + 193 ], [ - 122, - 128 + 257, + 277 ], [ - 254, - 261 + 347, + 355 ], [ - 969, - 984 + 373, + 385 ], [ - 1023, - 1032 + 406, + 415 ] ], "right_ranges": [ [ - 11, - 24 + 187, + 193 ], [ - 775, - 783 + 257, + 277 ], [ - 847, - 857 + 347, + 355 ], [ - 883, - 891 + 373, + 385 + ], + [ + 406, + 415 ] ] } @@ -6198,862 +3247,689 @@ ] }, { - "cluster_id": "cluster-2642-4", - "summary": "Cluster of 4 related pull requests linked by soft_similarity.", + "cluster_id": "cluster-10109-5", + "summary": "Cluster of 1 issues and 4 PRs centered on issue #10357.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 10357, + "canonical_pr_number": 10498, + "issue_numbers": [ + 10357 + ], + "pr_numbers": [ + 10109, + 10364, + 10365, + 10498 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10109, + "right_pr_number": 10364, + "code_similarity": 0.34, + "size_similarity": 0.952, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10109, + "right_pr_number": 10365, + "code_similarity": 0.3, + "size_similarity": 0.909, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.786, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10109, + "right_pr_number": 10498, + "code_similarity": 0.08, + "size_similarity": 0.008, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.524, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10364, + "right_pr_number": 10365, + "code_similarity": 0.309, + "size_similarity": 0.955, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.786, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10364, + "right_pr_number": 10498, + "code_similarity": 0.08, + "size_similarity": 0.008, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.524, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10365, + "right_pr_number": 10498, + "code_similarity": 0.102, + "size_similarity": 0.009, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.667, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-8446-5", + "summary": "Cluster of 5 related pull requests linked by soft_similarity.", "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 2684, + "canonical_pr_number": 8461, "issue_numbers": [], "pr_numbers": [ - 2642, - 2647, - 2684, - 2748 + 8446, + 8461, + 8476, + 9308, + 9316 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 2642, - "right_pr_number": 2647, - "code_similarity": 0.143, - "size_similarity": 0.7, + "left_pr_number": 8446, + "right_pr_number": 8461, + "code_similarity": 0.35, + "size_similarity": 1.0, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.019, + "patch_similarity": 1.0, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 2642, - "right_pr_number": 2684, - "code_similarity": 0.135, - "size_similarity": 0.6, + "left_pr_number": 8446, + "right_pr_number": 8476, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8446, + "right_pr_number": 9308, + "code_similarity": 0.321, + "size_similarity": 0.921, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.914, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8446, + "right_pr_number": 9316, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8461, + "right_pr_number": 8476, + "code_similarity": 0.35, + "size_similarity": 1.0, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.1, + "patch_similarity": 1.0, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 2642, - "right_pr_number": 2748, - "code_similarity": 0.195, - "size_similarity": 0.9, + "left_pr_number": 8461, + "right_pr_number": 9308, + "code_similarity": 0.321, + "size_similarity": 0.921, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.098, + "patch_similarity": 0.914, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 2647, - "right_pr_number": 2684, - "code_similarity": 0.176, - "size_similarity": 0.857, + "left_pr_number": 8461, + "right_pr_number": 9316, + "code_similarity": 0.35, + "size_similarity": 1.0, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.033, + "patch_similarity": 1.0, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 2647, - "right_pr_number": 2748, - "code_similarity": 0.158, - "size_similarity": 0.778, + "left_pr_number": 8476, + "right_pr_number": 9308, + "code_similarity": 0.321, + "size_similarity": 0.921, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.013, + "patch_similarity": 0.914, "shared_filenames": [], "shared_file_areas": [] }, { - "left_pr_number": 2684, - "right_pr_number": 2748, - "code_similarity": 0.145, - "size_similarity": 0.667, + "left_pr_number": 8476, + "right_pr_number": 9316, + "code_similarity": 0.35, + "size_similarity": 1.0, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.076, + "patch_similarity": 1.0, "shared_filenames": [], "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-6158-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #6158.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 6158, - "canonical_pr_number": 6721, - "issue_numbers": [ - 6158 - ], - "pr_numbers": [ - 6195, - 6721 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 6195, - "right_pr_number": 6721, - "code_similarity": 0.215, - "size_similarity": 0.615, + "left_pr_number": 9308, + "right_pr_number": 9316, + "code_similarity": 0.321, + "size_similarity": 0.921, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.61, + "patch_similarity": 0.914, "shared_filenames": [], "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-4944-2", - "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "cluster_id": "cluster-4498-5", + "summary": "Cluster of 5 related pull requests linked by soft_similarity.", "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 5713, + "canonical_pr_number": 4502, "issue_numbers": [], "pr_numbers": [ - 4944, - 5713 + 4498, + 4499, + 4500, + 4501, + 4502 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 4944, - "right_pr_number": 5713, - "code_similarity": 0.332, - "size_similarity": 0.309, - "file_overlap": 0.071, - "area_overlap": 0.555, - "patch_similarity": 0.366, + "left_pr_number": 4498, + "right_pr_number": 4499, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, "shared_filenames": [ - "src/diffusers/models/attention_processor.py", - "src/diffusers/models/unet_2d_condition.py", - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py" + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/models/attention_processor.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 1973, - 2384 - ], - [ - 2402, - 2409 + 1850, + 1887 ] ], "right_ranges": [ [ - 1975, - 2224 - ], - [ - 2242, - 2249 + 1850, + 1887 ] ] }, { - "filename": "src/diffusers/models/unet_2d_condition.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ - 1022, - 1028 + 1, + 788 ] ], "right_ranges": [ [ - 1022, - 1036 + 1, + 788 ] ] - }, + } + ] + }, + { + "left_pr_number": 4498, + "right_pr_number": 4500, + "code_similarity": 0.999, + "size_similarity": 0.999, + "file_overlap": 1.0, + "area_overlap": 0.999, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 17, - 27 - ], - [ - 70, - 78 - ], - [ - 84, - 90 - ], - [ - 106, - 112 - ], - [ - 119, - 125 - ], - [ - 196, - 202 - ], - [ - 444, - 473 - ], - [ - 648, - 654 - ], - [ - 695, - 701 - ], - [ - 795, - 808 + 1850, + 1887 ] ], "right_ranges": [ [ - 17, - 27 - ], - [ - 70, - 78 - ], - [ - 84, - 90 - ], - [ - 107, - 113 - ], - [ - 120, - 126 - ], - [ - 197, - 203 - ], - [ - 445, - 463 - ], - [ - 667, - 673 - ], - [ - 714, - 720 - ], - [ - 814, - 831 - ], - [ - 846, - 855 - ], - [ - 873, - 879 + 1850, + 1887 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-3896-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 4097, - "issue_numbers": [], - "pr_numbers": [ - 3896, - 4016, - 4097 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 3896, - "right_pr_number": 4016, - "code_similarity": 0.196, - "size_similarity": 0.42, - "file_overlap": 0.095, - "area_overlap": 0.008, - "patch_similarity": 0.535, - "shared_filenames": [ - "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", - "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py" - ], - "shared_file_areas": [ + }, { - "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ 1, - 805 + 788 ] ], "right_ranges": [ [ - 73, - 79 + 1, + 789 ] ] - }, + } + ] + }, + { + "left_pr_number": 4498, + "right_pr_number": 4501, + "code_similarity": 0.997, + "size_similarity": 0.994, + "file_overlap": 1.0, + "area_overlap": 0.994, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 1, - 869 + 1850, + 1887 ] ], "right_ranges": [ [ - 78, - 84 + 1850, + 1887 ] ] - } - ] - }, - { - "left_pr_number": 3896, - "right_pr_number": 4097, - "code_similarity": 0.096, - "size_similarity": 0.214, - "file_overlap": 0.05, - "area_overlap": 0.008, - "patch_similarity": 0.237, - "shared_filenames": [ - "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py" - ], - "shared_file_areas": [ + }, { - "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ 1, - 805 + 788 ] ], "right_ranges": [ [ - 13, - 19 - ], - [ - 810, - 875 + 1, + 793 ] ] } ] }, { - "left_pr_number": 4016, - "right_pr_number": 4097, - "code_similarity": 0.366, - "size_similarity": 0.509, - "file_overlap": 0.4, - "area_overlap": 0.299, - "patch_similarity": 0.267, + "left_pr_number": 4498, + "right_pr_number": 4502, + "code_similarity": 0.937, + "size_similarity": 0.953, + "file_overlap": 1.0, + "area_overlap": 0.87, + "patch_similarity": 0.948, "shared_filenames": [ - "examples/dreambooth/README_sdxl.md", - "examples/dreambooth/train_dreambooth_lora_sdxl.py", - "examples/test_examples.py", - "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py" + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" ], "shared_file_areas": [ { - "filename": "examples/dreambooth/README_sdxl.md", + "filename": "examples/community/README.md", "left_ranges": [ [ - 1, - 178 + 1850, + 1887 ] ], "right_ranges": [ [ - 164, - 180 + 1889, + 1957 ] ] }, { - "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ 1, - 1266 + 788 ] ], "right_ranges": [ [ - 16, - 22 - ], - [ - 46, - 56 - ], - [ - 64, - 70 - ], - [ - 92, - 99 - ], - [ - 128, - 139 - ], - [ - 424, - 455 - ], - [ - 461, - 466 - ], - [ - 485, - 513 - ], - [ - 528, - 556 - ], - [ - 602, - 607 - ], - [ - 635, - 641 - ], - [ - 700, - 713 - ], - [ - 729, - 738 - ], - [ - 772, - 837 - ], - [ - 861, - 871 - ], - [ - 874, - 954 - ], - [ - 969, - 989 - ], - [ - 1042, - 1050 - ], - [ - 1053, - 1068 - ], - [ - 1078, - 1107 - ], - [ - 1129, - 1139 - ], - [ - 1184, - 1205 - ], - [ - 1227, - 1237 - ], - [ - 1257, - 1288 - ], - [ - 1334, - 1340 + 1, + 799 ] ] - }, + } + ] + }, + { + "left_pr_number": 4499, + "right_pr_number": 4500, + "code_similarity": 0.999, + "size_similarity": 0.999, + "file_overlap": 1.0, + "area_overlap": 0.999, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ { - "filename": "examples/test_examples.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 353, - 390 + 1850, + 1887 ] ], "right_ranges": [ [ - 385, - 426 + 1850, + 1887 ] ] }, { - "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ - 73, - 79 + 1, + 788 ] ], "right_ranges": [ [ - 13, - 19 - ], - [ - 810, - 875 + 1, + 789 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-9772-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #9772.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 9772, - "canonical_pr_number": 10131, - "issue_numbers": [ - 9772 - ], - "pr_numbers": [ - 9776, - 10131 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 9776, - "right_pr_number": 10131, - "code_similarity": 0.789, - "size_similarity": 0.969, - "file_overlap": 0.444, - "area_overlap": 0.947, - "patch_similarity": 0.87, + "left_pr_number": 4499, + "right_pr_number": 4501, + "code_similarity": 0.997, + "size_similarity": 0.994, + "file_overlap": 1.0, + "area_overlap": 0.994, + "patch_similarity": 1.0, "shared_filenames": [ - "docs/source/en/_toctree.yml", - "src/diffusers/__init__.py", - "src/diffusers/models/__init__.py", - "src/diffusers/pipelines/__init__.py", - "src/diffusers/pipelines/controlnet/__init__.py", - "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py", - "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py", - "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py" + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" ], "shared_file_areas": [ { - "filename": "docs/source/en/_toctree.yml", + "filename": "examples/community/README.md", "left_ranges": [ [ - 217, - 224 + 1850, + 1887 ] ], "right_ranges": [ [ - 252, - 259 - ], - [ - 368, - 375 + 1850, + 1887 ] ] }, { - "filename": "src/diffusers/__init__.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ - 88, - 94 - ], - [ - 364, - 372 - ], - [ - 571, - 577 - ], - [ - 825, - 833 + 1, + 788 ] ], "right_ranges": [ [ - 91, - 97 - ], - [ - 377, - 385 - ], - [ - 587, - 593 - ], - [ - 851, - 859 + 1, + 793 ] ] - }, + } + ] + }, + { + "left_pr_number": 4499, + "right_pr_number": 4502, + "code_similarity": 0.937, + "size_similarity": 0.953, + "file_overlap": 1.0, + "area_overlap": 0.87, + "patch_similarity": 0.948, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/models/__init__.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 39, - 45 - ], - [ - 94, - 100 + 1850, + 1887 ] ], "right_ranges": [ [ - 44, - 50 - ], - [ - 101, - 107 + 1889, + 1957 ] ] }, { - "filename": "src/diffusers/pipelines/__init__.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ - 156, - 164 - ], - [ - 489, - 497 + 1, + 788 ] ], "right_ranges": [ [ - 162, - 170 - ], - [ - 498, - 506 + 1, + 799 ] ] - }, + } + ] + }, + { + "left_pr_number": 4500, + "right_pr_number": 4501, + "code_similarity": 0.997, + "size_similarity": 0.995, + "file_overlap": 1.0, + "area_overlap": 0.995, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/controlnet/__init__.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 30, - 38 - ], - [ - 60, - 68 + 1850, + 1887 ] ], "right_ranges": [ [ - 1, - 86 + 1850, + 1887 ] ] }, { - "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ 1, - 1824 + 789 ] ], "right_ranges": [ [ 1, - 1801 + 793 ] ] - }, + } + ] + }, + { + "left_pr_number": 4500, + "right_pr_number": 4502, + "code_similarity": 0.938, + "size_similarity": 0.954, + "file_overlap": 1.0, + "area_overlap": 0.871, + "patch_similarity": 0.948, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 1, - 1500 + 1850, + 1887 ] ], "right_ranges": [ [ - 1, - 1531 + 1889, + 1957 ] ] }, { - "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ 1, - 1624 + 789 ] ], "right_ranges": [ [ 1, - 1646 + 799 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-6688-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #6688.", - "status": "closed", - "confidence": 0.55, - "canonical_issue_number": 6688, - "canonical_pr_number": 7005, - "issue_numbers": [ - 6688 - ], - "pr_numbers": [ - 6844, - 7005 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 6844, - "right_pr_number": 7005, - "code_similarity": 0.471, - "size_similarity": 0.968, - "file_overlap": 0.182, - "area_overlap": 0.304, - "patch_similarity": 0.778, + "left_pr_number": 4501, + "right_pr_number": 4502, + "code_similarity": 0.94, + "size_similarity": 0.958, + "file_overlap": 1.0, + "area_overlap": 0.875, + "patch_similarity": 0.948, "shared_filenames": [ - "src/diffusers/models/attention.py", - "src/diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py" + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/models/attention.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 431, - 437 - ], - [ - 475, - 488 - ], - [ - 498, - 510 - ], - [ - 531, - 544 + 1850, + 1887 ] ], "right_ranges": [ [ - 425, - 431 + 1889, + 1957 ] ] }, { - "filename": "src/diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py", + "filename": "examples/community/stable_diffusion_xl_reference.py", "left_ranges": [ [ - 205, - 211 - ], - [ - 227, - 233 - ], - [ - 331, - 341 + 1, + 793 ] ], "right_ranges": [ [ - 205, - 211 - ], - [ - 227, - 233 - ], - [ - 331, - 339 + 1, + 799 ] ] } @@ -7062,536 +3938,846 @@ ] }, { - "cluster_id": "cluster-8710-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #8710.", + "cluster_id": "cluster-4780-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 8710, - "canonical_pr_number": 9386, - "issue_numbers": [ - 8710 - ], + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 5151, + "issue_numbers": [], "pr_numbers": [ - 8725, - 9386 + 4780, + 5151 ], "evidence_types": [ - "closing_reference" + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 8725, - "right_pr_number": 9386, - "code_similarity": 0.463, - "size_similarity": 0.85, - "file_overlap": 0.545, - "area_overlap": 0.029, - "patch_similarity": 0.797, + "left_pr_number": 4780, + "right_pr_number": 5151, + "code_similarity": 0.148, + "size_similarity": 0.208, + "file_overlap": 0.148, + "area_overlap": 0.102, + "patch_similarity": 0.175, "shared_filenames": [ - "docs/source/en/api/pipelines/pag.md", - "src/diffusers/__init__.py", - "src/diffusers/pipelines/__init__.py", - "src/diffusers/pipelines/auto_pipeline.py", - "src/diffusers/pipelines/pag/__init__.py", - "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + "src/diffusers/loaders.py", + "src/diffusers/models/attention.py", + "src/diffusers/models/attention_processor.py", + "src/diffusers/models/resnet.py", + "src/diffusers/models/transformer_2d.py", + "src/diffusers/models/unet_2d_condition.py", + "src/diffusers/utils/__init__.py", + "src/diffusers/utils/peft_utils.py" ], "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/pag.md", + "filename": "src/diffusers/loaders.py", "left_ranges": [ [ - 20, - 30 + 31, + 44 + ], + [ + 52, + 60 + ], + [ + 1392, + 1421 + ], + [ + 1444, + 1449 + ], + [ + 1462, + 1494 + ], + [ + 1497, + 1512 + ], + [ + 1533, + 1538 ] ], "right_ranges": [ [ - 48, - 58 + 11, + 16 + ], + [ + 31, + 46 + ], + [ + 72, + 77 + ], + [ + 400, + 406 + ], + [ + 514, + 523 + ], + [ + 528, + 563 + ], + [ + 680, + 756 + ], + [ + 1169, + 1174 + ], + [ + 1210, + 1216 + ], + [ + 1520, + 1559 + ], + [ + 1571, + 1579 + ], + [ + 1600, + 1655 + ], + [ + 1712, + 1718 + ], + [ + 1725, + 1731 + ], + [ + 1749, + 1779 + ], + [ + 1845, + 1865 + ], + [ + 2235, + 2254 + ], + [ + 2289, + 2295 + ], + [ + 2342, + 2357 + ], + [ + 2384, + 2390 + ], + [ + 2402, + 2408 + ], + [ + 2435, + 2441 + ], + [ + 2452, + 2597 + ], + [ + 3176, + 3187 + ], + [ + 3199, + 3207 + ], + [ + 3219, + 3227 + ], + [ + 3230, + 3236 + ], + [ + 3242, + 3248 + ], + [ + 3309, + 3325 ] ] }, { - "filename": "src/diffusers/__init__.py", + "filename": "src/diffusers/models/attention.py", "left_ranges": [ [ - 304, - 310 + 21, + 26 ], [ - 703, - 709 + 295, + 308 + ], + [ + 339, + 354 ] ], "right_ranges": [ [ - 363, - 369 + 17, + 23 ], [ - 835, - 841 + 301, + 307 + ], + [ + 318, + 332 + ], + [ + 371, + 379 + ], + [ + 382, + 389 ] ] }, { - "filename": "src/diffusers/pipelines/__init__.py", + "filename": "src/diffusers/models/attention_processor.py", "left_ranges": [ [ - 141, - 147 + 21, + 27 ], [ - 492, - 498 + 137, + 158 + ], + [ + 413, + 456 + ], + [ + 585, + 599 + ], + [ + 604, + 610 + ], + [ + 737, + 753 + ], + [ + 761, + 767 + ], + [ + 797, + 803 + ], + [ + 806, + 813 + ], + [ + 824, + 830 + ], + [ + 952, + 966 + ], + [ + 973, + 979 + ], + [ + 1030, + 1044 + ], + [ + 1058, + 1064 ] ], "right_ranges": [ [ - 174, - 180 + 18, + 24 ], [ - 596, - 602 + 137, + 163 + ], + [ + 550, + 557 + ], + [ + 569, + 583 + ], + [ + 588, + 594 + ], + [ + 1014, + 1033 + ], + [ + 1047, + 1055 ] ] }, { - "filename": "src/diffusers/pipelines/auto_pipeline.py", + "filename": "src/diffusers/models/resnet.py", "left_ranges": [ [ - 47, - 53 + 23, + 28 + ], + [ + 126, + 132 + ], + [ + 165, + 173 + ], + [ + 196, + 202 + ], + [ + 212, + 224 + ], + [ + 534, + 546 + ], + [ + 557, + 563 + ], + [ + 583, + 589 + ], + [ + 624, + 635 ], [ - 89, - 95 + 646, + 655 ] ], "right_ranges": [ [ - 66, - 72 + 20, + 26 ], [ - 161, - 167 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/pag/__init__.py", - "left_ranges": [ + 150, + 162 + ], [ - 23, - 29 + 195, + 206 ], [ - 37, - 43 - ] - ], - "right_ranges": [ + 239, + 248 + ], [ - 34, - 41 + 258, + 275 ], [ - 60, - 66 - ] - ] - }, - { - "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", - "left_ranges": [ + 616, + 624 + ], [ - 1232, - 1252 - ] - ], - "right_ranges": [ + 629, + 641 + ], [ - 1757, - 1777 + 652, + 658 + ], + [ + 678, + 684 + ], + [ + 719, + 734 + ], + [ + 745, + 756 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-6891-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #6891.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 6891, - "canonical_pr_number": 7615, - "issue_numbers": [ - 6891 - ], - "pr_numbers": [ - 7613, - 7615 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 7613, - "right_pr_number": 7615, - "code_similarity": 0.319, - "size_similarity": 0.882, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.948, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-7677-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #7677.", - "status": "open", - "confidence": 0.55, - "canonical_issue_number": 7677, - "canonical_pr_number": 7865, - "issue_numbers": [ - 7677 - ], - "pr_numbers": [ - 7865, - 7898 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ - { - "left_pr_number": 7865, - "right_pr_number": 7898, - "code_similarity": 0.159, - "size_similarity": 0.682, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.15, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-12441-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #12441.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 12441, - "canonical_pr_number": 12526, - "issue_numbers": [ - 12441 - ], - "pr_numbers": [ - 12442, - 12526 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 12442, - "right_pr_number": 12526, - "code_similarity": 0.687, - "size_similarity": 0.728, - "file_overlap": 0.684, - "area_overlap": 0.669, - "patch_similarity": 0.68, - "shared_filenames": [ - "docs/source/en/api/pipelines/wan.md", - "scripts/convert_wan_to_diffusers.py", - "src/diffusers/__init__.py", - "src/diffusers/models/__init__.py", - "src/diffusers/models/transformers/__init__.py", - "src/diffusers/models/transformers/transformer_wan_animate.py", - "src/diffusers/pipelines/__init__.py", - "src/diffusers/pipelines/wan/__init__.py", - "src/diffusers/pipelines/wan/pipeline_wan_animate.py", - "src/diffusers/utils/dummy_pt_objects.py", - "src/diffusers/utils/dummy_torch_and_transformers_objects.py", - "tests/pipelines/wan/test_wan_animate.py", - "tests/quantization/gguf/test_gguf.py" - ], - "shared_file_areas": [ + }, { - "filename": "docs/source/en/api/pipelines/wan.md", + "filename": "src/diffusers/models/transformer_2d.py", "left_ranges": [ [ - 40, - 46 + 23, + 28 ], [ - 96, - 110 + 138, + 146 ], [ - 151, - 165 + 196, + 204 ], [ - 250, - 469 + 274, + 280 ], [ - 496, - 505 + 283, + 295 ], [ - 574, - 585 + 325, + 333 ] ], "right_ranges": [ [ - 40, - 46 + 20, + 26 ], [ - 96, - 110 + 100, + 108 ], [ - 151, - 165 + 142, + 150 ], [ - 250, - 469 + 200, + 208 ], [ - 496, - 505 + 295, + 315 ], [ - 574, - 585 + 345, + 361 ] ] }, { - "filename": "scripts/convert_wan_to_diffusers.py", + "filename": "src/diffusers/models/unet_2d_condition.py", "left_ranges": [ [ - 1, - 27 - ], + 731, + 737 + ] + ], + "right_ranges": [ [ - 116, - 308 + 20, + 26 ], [ - 560, - 590 + 995, + 1003 ], [ - 601, - 612 - ], + 1097, + 1106 + ] + ] + }, + { + "filename": "src/diffusers/utils/__init__.py", + "left_ranges": [ [ - 620, - 643 + 67, + 73 ], [ - 1148, - 1311 - ], + 83, + 94 + ] + ], + "right_ranges": [ [ - 1324, - 1330 + 26, + 36 ], [ - 1340, - 1346 + 88, + 94 ], [ - 1352, - 1359 - ], + 98, + 108 + ] + ] + }, + { + "filename": "src/diffusers/utils/peft_utils.py", + "left_ranges": [ [ - 1416, - 1449 + 1, + 153 ] ], "right_ranges": [ [ - 6, + 15, 25 ], [ - 114, - 316 + 56, + 61 ], [ - 568, - 604 + 108, + 118 ], [ - 615, - 626 + 121, + 142 ], [ - 634, - 645 + 167, + 175 ], [ - 1168, - 1174 + 194, + 216 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-10025-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #10025.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 10025, + "canonical_pr_number": 10044, + "issue_numbers": [ + 10025 + ], + "pr_numbers": [ + 10044, + 10051, + 10053 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10044, + "right_pr_number": 10051, + "code_similarity": 0.133, + "size_similarity": 0.25, + "file_overlap": 0.2, + "area_overlap": 0.0, + "patch_similarity": 0.152, + "shared_filenames": [ + "src/diffusers/models/attention_processor.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 731, + 773 ], [ - 1184, - 1190 + 1495, + 1508 ], [ - 1196, - 1203 + 2356, + 2366 + ] + ], + "right_ranges": [ + [ + 1762, + 1770 ], [ - 1260, - 1280 + 1858, + 1866 ] ] - }, + } + ] + }, + { + "left_pr_number": 10044, + "right_pr_number": 10053, + "code_similarity": 0.27, + "size_similarity": 0.508, + "file_overlap": 0.5, + "area_overlap": 0.0, + "patch_similarity": 0.121, + "shared_filenames": [ + "src/diffusers/models/attention_processor.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/__init__.py", + "filename": "src/diffusers/models/attention_processor.py", "left_ranges": [ [ - 259, - 265 - ], - [ - 621, - 627 + 731, + 773 ], [ - 954, - 960 + 1495, + 1508 ], [ - 1286, - 1292 + 2356, + 2366 ] ], "right_ranges": [ [ - 268, - 274 + 1907, + 1915 ], [ - 637, - 643 - ], + 2014, + 2025 + ] + ] + } + ] + }, + { + "left_pr_number": 10051, + "right_pr_number": 10053, + "code_similarity": 0.213, + "size_similarity": 0.127, + "file_overlap": 0.4, + "area_overlap": 0.152, + "patch_similarity": 0.094, + "shared_filenames": [ + "scripts/convert_flux_to_diffusers.py", + "src/diffusers/models/attention_processor.py" + ], + "shared_file_areas": [ + { + "filename": "scripts/convert_flux_to_diffusers.py", + "left_ranges": [ [ - 979, - 985 - ], + 31, + 37 + ] + ], + "right_ranges": [ [ - 1318, - 1324 + 31, + 37 ] ] }, { - "filename": "src/diffusers/models/__init__.py", + "filename": "src/diffusers/models/attention_processor.py", "left_ranges": [ [ - 101, - 107 + 1762, + 1770 ], [ - 201, - 207 + 1858, + 1866 ] ], "right_ranges": [ [ - 108, - 114 + 1907, + 1915 ], [ - 215, - 221 + 2014, + 2025 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-8810-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #9886.", + "status": "open", + "confidence": 0.6, + "canonical_issue_number": 9886, + "canonical_pr_number": 8864, + "issue_numbers": [ + 9886 + ], + "pr_numbers": [ + 8810, + 8864, + 9917 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 8810, + "right_pr_number": 8864, + "code_similarity": 0.849, + "size_similarity": 0.755, + "file_overlap": 0.875, + "area_overlap": 0.886, + "patch_similarity": 0.835, + "shared_filenames": [ + "docs/source/en/api/pipelines/pag.md", + "src/diffusers/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/pipelines/pag/__init__.py", + "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/pag.md", + "left_ranges": [ + [ + 44, + 51 + ] + ], + "right_ranges": [ + [ + 71, + 81 ] ] }, { - "filename": "src/diffusers/models/transformers/__init__.py", + "filename": "src/diffusers/__init__.py", "left_ranges": [ [ - 37, - 41 + 290, + 296 + ], + [ + 690, + 696 ] ], "right_ranges": [ [ - 42, - 46 + 332, + 338 + ], + [ + 790, + 796 ] ] }, { - "filename": "src/diffusers/models/transformers/transformer_wan_animate.py", + "filename": "src/diffusers/pipelines/__init__.py", "left_ranges": [ [ - 1, - 611 + 145, + 151 + ], + [ + 493, + 499 ] ], "right_ranges": [ [ - 1, - 1298 + 168, + 174 + ], + [ + 572, + 578 ] ] }, { - "filename": "src/diffusers/pipelines/__init__.py", + "filename": "src/diffusers/pipelines/auto_pipeline.py", "left_ranges": [ [ - 381, - 393 + 47, + 53 ], [ - 794, - 806 + 93, + 99 ] ], "right_ranges": [ [ - 385, - 397 + 59, + 65 ], [ - 809, - 821 + 135, + 141 ] ] }, { - "filename": "src/diffusers/pipelines/wan/__init__.py", + "filename": "src/diffusers/pipelines/pag/__init__.py", "left_ranges": [ [ - 23, - 29 + 22, + 28 ], [ - 36, - 45 + 37, + 43 ] ], "right_ranges": [ @@ -7600,424 +4786,404 @@ 29 ], [ - 36, - 45 + 47, + 53 ] ] }, { - "filename": "src/diffusers/pipelines/wan/pipeline_wan_animate.py", + "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", "left_ranges": [ [ 1, - 1072 + 1297 ] ], "right_ranges": [ [ 1, - 1204 + 1291 ] ] }, { - "filename": "src/diffusers/utils/dummy_pt_objects.py", + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", "left_ranges": [ [ - 1488, - 1508 + 1247, + 1267 ] ], "right_ranges": [ [ - 1623, - 1643 + 1832, + 1852 ] ] - }, + } + ] + }, + { + "left_pr_number": 8810, + "right_pr_number": 9917, + "code_similarity": 0.808, + "size_similarity": 0.773, + "file_overlap": 0.75, + "area_overlap": 0.877, + "patch_similarity": 0.81, + "shared_filenames": [ + "src/diffusers/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/pipelines/pag/__init__.py", + "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "filename": "src/diffusers/__init__.py", "left_ranges": [ [ - 3407, - 3427 + 290, + 296 + ], + [ + 690, + 696 ] ], "right_ranges": [ [ - 3512, - 3532 + 340, + 346 + ], + [ + 806, + 812 ] ] }, { - "filename": "tests/pipelines/wan/test_wan_animate.py", + "filename": "src/diffusers/pipelines/__init__.py", "left_ranges": [ [ - 1, - 245 + 145, + 151 + ], + [ + 493, + 499 ] ], "right_ranges": [ [ - 1, - 239 + 169, + 175 ] ] }, { - "filename": "tests/quantization/gguf/test_gguf.py", + "filename": "src/diffusers/pipelines/auto_pipeline.py", "left_ranges": [ [ - 16, - 22 + 47, + 53 ], [ - 722, - 754 + 93, + 99 ] ], "right_ranges": [ [ - 16, - 22 + 64, + 70 ], [ - 722, - 754 + 136, + 142 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-5886-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5886.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 5886, - "canonical_pr_number": 5911, - "issue_numbers": [ - 5886 - ], - "pr_numbers": [ - 5910, - 5911 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 5910, - "right_pr_number": 5911, - "code_similarity": 0.294, - "size_similarity": 0.411, - "file_overlap": 0.333, - "area_overlap": 0.26, - "patch_similarity": 0.134, - "shared_filenames": [ - "src/diffusers/loaders/unet.py", - "src/diffusers/models/embeddings.py" - ], - "shared_file_areas": [ + }, { - "filename": "src/diffusers/loaders/unet.py", + "filename": "src/diffusers/pipelines/pag/__init__.py", "left_ranges": [ [ - 21, - 27 - ], - [ - 664, - 670 + 22, + 28 ], [ - 711, - 726 + 37, + 43 ] ], "right_ranges": [ [ - 21, - 27 + 25, + 31 ], [ - 675, - 683 - ], + 48, + 54 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "left_ranges": [ [ - 747, - 778 + 1, + 1297 + ] + ], + "right_ranges": [ + [ + 1, + 1342 ] ] }, { - "filename": "src/diffusers/models/embeddings.py", + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", "left_ranges": [ [ - 458, - 482 + 1247, + 1267 ] ], "right_ranges": [ [ - 461, - 478 + 1786, + 1804 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-4899-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5800.", - "status": "closed", - "confidence": 0.6, - "canonical_issue_number": 5800, - "canonical_pr_number": 5938, - "issue_numbers": [ - 5800 - ], - "pr_numbers": [ - 4899, - 5938 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 4899, - "right_pr_number": 5938, - "code_similarity": 0.472, - "size_similarity": 0.741, - "file_overlap": 0.375, - "area_overlap": 0.354, - "patch_similarity": 0.586, + "left_pr_number": 8864, + "right_pr_number": 9917, + "code_similarity": 0.912, + "size_similarity": 0.977, + "file_overlap": 0.857, + "area_overlap": 0.9, + "patch_similarity": 0.961, "shared_filenames": [ - "src/diffusers/models/controlnet.py", - "src/diffusers/models/embeddings.py", - "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py" + "src/diffusers/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/pipelines/pag/__init__.py", + "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/models/controlnet.py", + "filename": "src/diffusers/__init__.py", "left_ranges": [ [ - 19, - 26 - ], - [ - 81, - 87 - ], - [ - 97, - 103 - ], - [ - 109, - 117 - ], - [ - 251, - 257 - ], - [ - 723, - 738 + 332, + 338 ], [ - 771, - 777 + 790, + 796 ] ], "right_ranges": [ [ - 20, - 27 - ], - [ - 82, - 88 - ], - [ - 108, - 116 + 340, + 346 ], [ - 253, - 259 + 806, + 812 ] ] }, { - "filename": "src/diffusers/models/embeddings.py", + "filename": "src/diffusers/pipelines/__init__.py", "left_ranges": [ [ - 18, - 24 - ], - [ - 167, - 176 + 168, + 174 ], [ - 180, - 186 + 572, + 578 ] ], "right_ranges": [ [ - 18, - 24 - ], - [ - 201, - 210 - ], - [ - 214, - 220 + 169, + 175 ] ] }, { - "filename": "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", + "filename": "src/diffusers/pipelines/auto_pipeline.py", "left_ranges": [ [ - 377, - 397 - ], - [ - 429, - 450 - ], - [ - 458, - 483 - ], - [ - 623, - 631 - ], - [ - 635, - 647 + 59, + 65 ], [ - 652, - 672 + 135, + 141 ] ], "right_ranges": [ [ - 304, - 311 - ], - [ - 323, - 328 - ], - [ - 377, - 397 - ], - [ - 429, - 450 - ], - [ - 458, - 483 - ], - [ - 531, - 536 - ], - [ - 623, - 631 - ], - [ - 635, - 647 - ], - [ - 652, - 672 - ], - [ - 830, - 836 - ], - [ - 965, - 971 - ], - [ - 1178, - 1183 + 64, + 70 ], [ - 1262, - 1275 - ], + 136, + 142 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/__init__.py", + "left_ranges": [ [ - 1308, - 1332 + 23, + 29 ], [ - 1346, - 1353 - ], + 47, + 53 + ] + ], + "right_ranges": [ [ - 1431, - 1439 + 25, + 31 ], [ - 1482, - 1492 - ], + 48, + 54 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "left_ranges": [ [ - 1498, - 1505 - ], + 1, + 1291 + ] + ], + "right_ranges": [ [ - 1509, - 1516 - ], + 1, + 1342 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "left_ranges": [ [ - 1541, - 1549 - ], + 1832, + 1852 + ] + ], + "right_ranges": [ [ - 1576, - 1585 - ], + 1786, + 1804 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-281-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #281.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 281, + "canonical_pr_number": 361, + "issue_numbers": [ + 281, + 725 + ], + "pr_numbers": [ + 361, + 727 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 361, + "right_pr_number": 727, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-6137-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #8236.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 8236, + "canonical_pr_number": 8312, + "issue_numbers": [ + 6137, + 8236 + ], + "pr_numbers": [ + 6143, + 8312 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6143, + "right_pr_number": 8312, + "code_similarity": 0.239, + "size_similarity": 0.493, + "file_overlap": 0.333, + "area_overlap": 0.0, + "patch_similarity": 0.269, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image_lora.py" + ], + "shared_file_areas": [ + { + "filename": "examples/text_to_image/train_text_to_image_lora.py", + "left_ranges": [ [ - 1592, - 1602 + 630, + 638 ], [ - 1632, - 1649 - ], + 646, + 651 + ] + ], + "right_ranges": [ [ - 1663, - 1668 + 697, + 718 ], [ - 1677, - 1690 + 722, + 735 ] ] } @@ -8026,18 +5192,19 @@ ] }, { - "cluster_id": "cluster-10025-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #10025.", - "status": "open", + "cluster_id": "cluster-5249-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #5249.", + "status": "merged", "confidence": 0.8, - "canonical_issue_number": 10025, - "canonical_pr_number": 10053, + "canonical_issue_number": 5249, + "canonical_pr_number": 5293, "issue_numbers": [ - 10025 + 5249 ], "pr_numbers": [ - 10051, - 10053 + 5293, + 5299, + 5312 ], "evidence_types": [ "closing_reference", @@ -8046,53 +5213,178 @@ ], "pr_comparisons": [ { - "left_pr_number": 10051, - "right_pr_number": 10053, - "code_similarity": 0.213, - "size_similarity": 0.127, - "file_overlap": 0.4, - "area_overlap": 0.152, - "patch_similarity": 0.094, + "left_pr_number": 5293, + "right_pr_number": 5299, + "code_similarity": 0.044, + "size_similarity": 0.098, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.164, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5293, + "right_pr_number": 5312, + "code_similarity": 0.028, + "size_similarity": 0.078, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.08, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5299, + "right_pr_number": 5312, + "code_similarity": 0.188, + "size_similarity": 0.795, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.192, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-11730-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #11730.", + "status": "open", + "confidence": 0.6, + "canonical_issue_number": 11730, + "canonical_pr_number": 11744, + "issue_numbers": [ + 11730 + ], + "pr_numbers": [ + 11744, + 11782, + 11783 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 11744, + "right_pr_number": 11782, + "code_similarity": 0.339, + "size_similarity": 0.963, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.975, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11744, + "right_pr_number": 11783, + "code_similarity": 0.339, + "size_similarity": 0.963, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.975, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11782, + "right_pr_number": 11783, + "code_similarity": 0.964, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 0.896, + "patch_similarity": 1.0, "shared_filenames": [ - "scripts/convert_flux_to_diffusers.py", - "src/diffusers/models/attention_processor.py" + "examples/dreambooth/test_dreambooth_lora_sd3.py", + "examples/dreambooth/train_dreambooth_lora_sd3.py" ], "shared_file_areas": [ { - "filename": "scripts/convert_flux_to_diffusers.py", + "filename": "examples/dreambooth/test_dreambooth_lora_sd3.py", "left_ranges": [ [ - 31, - 37 + 13, + 27 + ], + [ + 209, + 254 ] ], "right_ranges": [ [ - 31, - 37 + 13, + 27 + ], + [ + 209, + 254 ] ] }, { - "filename": "src/diffusers/models/attention_processor.py", + "filename": "examples/dreambooth/train_dreambooth_lora_sd3.py", "left_ranges": [ [ - 1762, - 1770 + 53, + 59 ], [ - 1858, - 1866 + 322, + 333 + ], + [ + 1273, + 1279 + ], + [ + 1302, + 1316 + ], + [ + 1333, + 1339 + ], + [ + 1935, + 1946 + ], + [ + 1957, + 1963 ] ], "right_ranges": [ [ - 1907, - 1915 + 54, + 60 ], [ - 2014, - 2025 + 323, + 334 + ], + [ + 1274, + 1280 + ], + [ + 1303, + 1317 + ], + [ + 1334, + 1340 + ], + [ + 1936, + 1947 + ], + [ + 1958, + 1964 ] ] } @@ -8101,314 +5393,555 @@ ] }, { - "cluster_id": "cluster-7038-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #7038.", - "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 7038, - "canonical_pr_number": 7550, + "cluster_id": "cluster-8384-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #8384.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 8384, + "canonical_pr_number": 11239, "issue_numbers": [ - 7038 + 8384 ], "pr_numbers": [ - 7442, - 7550 + 10973, + 11239, + 11240 ], "evidence_types": [ - "closing_reference" + "closing_reference", + "shared_issue_target", + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 7442, - "right_pr_number": 7550, - "code_similarity": 0.083, - "size_similarity": 0.271, + "left_pr_number": 10973, + "right_pr_number": 11239, + "code_similarity": 0.344, + "size_similarity": 0.972, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.192, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10973, + "right_pr_number": 11240, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11239, + "right_pr_number": 11240, + "code_similarity": 0.344, + "size_similarity": 0.972, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, "shared_filenames": [], "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-13377-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #13377.", - "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 13377, - "canonical_pr_number": 13379, + "cluster_id": "cluster-13074-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #13074.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 13074, + "canonical_pr_number": 13075, "issue_numbers": [ - 13377 + 13074 ], "pr_numbers": [ - 13379, - 13449 + 13075, + 13093, + 13514 ], "evidence_types": [ - "closing_reference" + "closing_reference", + "shared_issue_target", + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 13379, - "right_pr_number": 13449, - "code_similarity": 0.212, - "size_similarity": 0.179, - "file_overlap": 0.562, - "area_overlap": 0.002, - "patch_similarity": 0.046, + "left_pr_number": 13075, + "right_pr_number": 13093, + "code_similarity": 0.695, + "size_similarity": 0.85, + "file_overlap": 1.0, + "area_overlap": 0.5, + "patch_similarity": 0.333, "shared_filenames": [ - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage.py", - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet.py", - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet_inpaint.py", - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit.py", - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_inpaint.py", - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py", - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py", - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_layered.py" + "src/diffusers/models/_modeling_parallel.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage.py", + "filename": "src/diffusers/models/_modeling_parallel.py", "left_ranges": [ [ - 311, - 332 + 22, + 28 ], [ - 600, - 606 + 291, + 304 ] ], "right_ranges": [ [ - 585, - 591 + 290, + 301 ] ] - }, + } + ] + }, + { + "left_pr_number": 13075, + "right_pr_number": 13514, + "code_similarity": 0.593, + "size_similarity": 0.95, + "file_overlap": 1.0, + "area_overlap": 0.207, + "patch_similarity": 0.205, + "shared_filenames": [ + "src/diffusers/models/_modeling_parallel.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet.py", + "filename": "src/diffusers/models/_modeling_parallel.py", "left_ranges": [ [ - 101, - 107 - ], - [ - 239, - 245 - ], - [ - 248, - 254 - ], - [ - 287, - 293 - ], - [ - 318, - 330 - ], - [ - 376, - 397 + 22, + 28 ], [ - 718, - 724 + 291, + 304 ] ], "right_ranges": [ [ - 701, - 707 + 299, + 312 ] ] - }, + } + ] + }, + { + "left_pr_number": 13093, + "right_pr_number": 13514, + "code_similarity": 0.576, + "size_similarity": 0.895, + "file_overlap": 1.0, + "area_overlap": 0.13, + "patch_similarity": 0.345, + "shared_filenames": [ + "src/diffusers/models/_modeling_parallel.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet_inpaint.py", + "filename": "src/diffusers/models/_modeling_parallel.py", "left_ranges": [ [ - 74, - 80 - ], - [ - 221, - 227 - ], + 290, + 301 + ] + ], + "right_ranges": [ [ - 230, - 236 - ], + 299, + 312 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6514-4", + "summary": "Cluster of 4 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 6514, + "issue_numbers": [], + "pr_numbers": [ + 6514, + 6554, + 6840, + 7242 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6514, + "right_pr_number": 6554, + "code_similarity": 0.182, + "size_similarity": 0.444, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.623, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6514, + "right_pr_number": 6840, + "code_similarity": 0.161, + "size_similarity": 0.378, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.569, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6514, + "right_pr_number": 7242, + "code_similarity": 0.118, + "size_similarity": 0.089, + "file_overlap": 0.333, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ [ - 247, - 253 + 34, + 40 ], - [ - 269, - 275 + [ + 53, + 65 ], [ - 280, - 286 + 1002, + 1007 ], [ - 300, - 317 + 1058, + 1096 ], [ - 359, - 377 + 1105, + 1121 ], [ - 751, - 757 + 1533, + 1539 ] ], "right_ranges": [ [ - 740, - 746 + 877, + 884 ] ] - }, + } + ] + }, + { + "left_pr_number": 6554, + "right_pr_number": 6840, + "code_similarity": 0.297, + "size_similarity": 0.85, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.847, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6554, + "right_pr_number": 7242, + "code_similarity": 0.04, + "size_similarity": 0.2, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6840, + "right_pr_number": 7242, + "code_similarity": 0.047, + "size_similarity": 0.235, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-12771-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #12771.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 12771, + "canonical_pr_number": 12788, + "issue_numbers": [ + 12771 + ], + "pr_numbers": [ + 12788, + 13239, + 13240 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 12788, + "right_pr_number": 13239, + "code_similarity": 0.238, + "size_similarity": 0.24, + "file_overlap": 0.5, + "area_overlap": 0.038, + "patch_similarity": 0.178, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit.py", + "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", "left_ranges": [ [ - 247, - 261 - ], - [ - 306, - 318 + 24, + 31 ], [ - 360, - 378 + 412, + 445 ], [ - 714, - 720 + 550, + 587 ] ], "right_ranges": [ [ - 706, - 712 + 442, + 452 + ], + [ + 472, + 488 ] ] - }, + } + ] + }, + { + "left_pr_number": 12788, + "right_pr_number": 13240, + "code_similarity": 0.236, + "size_similarity": 0.25, + "file_overlap": 0.5, + "area_overlap": 0.038, + "patch_similarity": 0.15, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_inpaint.py", + "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", "left_ranges": [ [ - 258, - 272 - ], - [ - 318, - 330 + 24, + 31 ], [ - 392, - 412 + 412, + 445 ], [ - 895, - 901 + 550, + 587 ] ], "right_ranges": [ [ - 879, - 885 + 442, + 454 + ], + [ + 465, + 480 ] ] - }, + } + ] + }, + { + "left_pr_number": 13239, + "right_pr_number": 13240, + "code_similarity": 0.756, + "size_similarity": 0.962, + "file_overlap": 1.0, + "area_overlap": 0.541, + "patch_similarity": 0.5, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", + "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", "left_ranges": [ [ - 260, - 274 - ], - [ - 320, - 332 - ], - [ - 375, - 393 + 442, + 452 ], [ - 702, - 708 + 472, + 488 ] ], "right_ranges": [ [ - 694, - 700 + 442, + 454 + ], + [ + 465, + 480 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-13077-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #13077.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 13077, + "canonical_pr_number": 13136, + "issue_numbers": [ + 13077 + ], + "pr_numbers": [ + 13102, + 13128, + 13136 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 13102, + "right_pr_number": 13128, + "code_similarity": 0.27, + "size_similarity": 0.102, + "file_overlap": 0.5, + "area_overlap": 0.253, + "patch_similarity": 0.074, + "shared_filenames": [ + "src/diffusers/models/controlnets/controlnet_z_image.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py", + "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", "left_ranges": [ [ - 358, - 379 - ], - [ - 693, - 699 + 517, + 535 ] ], "right_ranges": [ - [ - 678, - 684 + [ + 517, + 591 ] ] - }, + } + ] + }, + { + "left_pr_number": 13102, + "right_pr_number": 13136, + "code_similarity": 0.693, + "size_similarity": 0.9, + "file_overlap": 1.0, + "area_overlap": 0.531, + "patch_similarity": 0.182, + "shared_filenames": [ + "src/diffusers/models/controlnets/controlnet_z_image.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py", + "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", "left_ranges": [ [ - 385, - 405 - ], - [ - 837, - 843 + 517, + 535 ] ], "right_ranges": [ [ - 823, - 829 + 517, + 528 + ], + [ + 531, + 548 ] ] - }, + } + ] + }, + { + "left_pr_number": 13128, + "right_pr_number": 13136, + "code_similarity": 0.321, + "size_similarity": 0.092, + "file_overlap": 0.5, + "area_overlap": 0.4, + "patch_similarity": 0.083, + "shared_filenames": [ + "src/diffusers/models/controlnets/controlnet_z_image.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_layered.py", + "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", "left_ranges": [ [ - 385, - 403 - ], - [ - 704, - 710 + 517, + 591 ] ], "right_ranges": [ [ - 698, - 704 + 517, + 528 + ], + [ + 531, + 548 ] ] } @@ -8417,87 +5950,160 @@ ] }, { - "cluster_id": "cluster-12458-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #12458.", - "status": "open", - "confidence": 0.55, - "canonical_issue_number": 12458, - "canonical_pr_number": 12698, + "cluster_id": "cluster-1257-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #1257.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 1257, + "canonical_pr_number": 1361, "issue_numbers": [ - 12458 + 1257, + 2950 ], "pr_numbers": [ - 12467, - 12698 + 1361, + 3000 ], "evidence_types": [ - "closing_reference" + "closing_reference", + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 12467, - "right_pr_number": 12698, - "code_similarity": 0.133, - "size_similarity": 0.125, - "file_overlap": 0.067, - "area_overlap": 0.208, - "patch_similarity": 0.105, + "left_pr_number": 1361, + "right_pr_number": 3000, + "code_similarity": 0.018, + "size_similarity": 0.041, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.063, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-1407-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #1792.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 1792, + "canonical_pr_number": 1416, + "issue_numbers": [ + 1407, + 1792 + ], + "pr_numbers": [ + 1416, + 1796 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 1416, + "right_pr_number": 1796, + "code_similarity": 0.149, + "size_similarity": 0.691, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.071, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-11315-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #11315.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 11315, + "canonical_pr_number": 12665, + "issue_numbers": [ + 11315, + 12653 + ], + "pr_numbers": [ + 11316, + 12665 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 11316, + "right_pr_number": 12665, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-1466-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #1466.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 1466, + "canonical_pr_number": 2070, + "issue_numbers": [ + 1466, + 2041 + ], + "pr_numbers": [ + 1547, + 2070 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 1547, + "right_pr_number": 2070, + "code_similarity": 0.211, + "size_similarity": 0.636, + "file_overlap": 0.2, + "area_overlap": 0.0, + "patch_similarity": 0.155, "shared_filenames": [ - "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py" + "examples/dreambooth/train_dreambooth.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", + "filename": "examples/dreambooth/train_dreambooth.py", "left_ranges": [ [ - 252, - 268 - ], - [ - 635, - 646 - ], - [ - 686, - 692 + 300, + 309 ], [ - 695, - 704 + 313, + 349 ], [ - 736, - 760 + 534, + 545 ] ], "right_ranges": [ [ - 226, - 250 - ], - [ - 255, - 286 - ], - [ - 312, - 317 - ], - [ - 322, - 380 - ], - [ - 680, - 706 - ], - [ - 739, - 810 + 240, + 253 ], [ - 853, - 868 + 660, + 666 ] ] } @@ -8506,55 +6112,69 @@ ] }, { - "cluster_id": "cluster-5524-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5524.", + "cluster_id": "cluster-11702-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #11709.", "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 5524, - "canonical_pr_number": 5704, + "confidence": 0.6, + "canonical_issue_number": 11709, + "canonical_pr_number": 11704, "issue_numbers": [ - 5524 + 11702, + 11709 ], "pr_numbers": [ - 5696, - 5704 + 11704, + 11726 ], "evidence_types": [ - "closing_reference" + "closing_reference", + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 5696, - "right_pr_number": 5704, - "code_similarity": 0.183, - "size_similarity": 0.165, - "file_overlap": 0.25, - "area_overlap": 0.194, - "patch_similarity": 0.048, + "left_pr_number": 11704, + "right_pr_number": 11726, + "code_similarity": 0.545, + "size_similarity": 0.711, + "file_overlap": 1.0, + "area_overlap": 0.215, + "patch_similarity": 0.183, "shared_filenames": [ - "src/diffusers/pipelines/wuerstchen/modeling_wuerstchen_prior.py" + "src/diffusers/loaders/lora_conversion_utils.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/wuerstchen/modeling_wuerstchen_prior.py", + "filename": "src/diffusers/loaders/lora_conversion_utils.py", "left_ranges": [ [ - 59, - 64 + 1605, + 1622 + ], + [ + 1664, + 1679 + ], + [ + 1682, + 1697 + ], + [ + 1726, + 1745 ] ], "right_ranges": [ [ - 28, - 36 + 1596, + 1605 ], [ - 41, - 55 + 1625, + 1681 ], [ - 59, - 65 + 1689, + 1698 ] ] } @@ -8563,153 +6183,220 @@ ] }, { - "cluster_id": "cluster-12754-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #12754.", - "status": "open", - "confidence": 0.75, - "canonical_issue_number": 12754, - "canonical_pr_number": 12762, + "cluster_id": "cluster-6866-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #6866.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 6866, + "canonical_pr_number": 6941, "issue_numbers": [ - 12754 + 6866 ], "pr_numbers": [ - 12762, - 13103 + 6941, + 7030, + 7031 ], "evidence_types": [ "closing_reference", - "shared_issue_target" + "shared_issue_target", + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 12762, - "right_pr_number": 13103, - "code_similarity": 0.538, - "size_similarity": 0.923, + "left_pr_number": 6941, + "right_pr_number": 7030, + "code_similarity": 0.345, + "size_similarity": 0.139, "file_overlap": 1.0, - "area_overlap": 0.0, - "patch_similarity": 0.357, + "area_overlap": 0.039, + "patch_similarity": 0.027, "shared_filenames": [ - "src/diffusers/pipelines/pipeline_utils.py" + "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/pipeline_utils.py", + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", "left_ranges": [ [ - 2115, - 2121 + 19, + 35 + ], + [ + 201, + 208 + ], + [ + 218, + 225 + ], + [ + 507, + 572 + ], + [ + 636, + 643 + ], + [ + 824, + 834 + ], + [ + 1177, + 1184 + ], + [ + 1273, + 1282 + ], + [ + 1409, + 1416 + ], + [ + 1463, + 1474 + ], + [ + 1740, + 1748 ] ], "right_ranges": [ [ - 2082, - 2089 + 28, + 34 + ], + [ + 146, + 152 + ], + [ + 158, + 164 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-12905-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #12905.", - "status": "open", - "confidence": 0.8, - "canonical_issue_number": 12905, - "canonical_pr_number": 13183, - "issue_numbers": [ - 12905 - ], - "pr_numbers": [ - 13094, - 13183 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 13094, - "right_pr_number": 13183, - "code_similarity": 0.597, - "size_similarity": 0.632, + "left_pr_number": 6941, + "right_pr_number": 7031, + "code_similarity": 0.356, + "size_similarity": 0.185, "file_overlap": 1.0, - "area_overlap": 0.438, - "patch_similarity": 0.119, + "area_overlap": 0.044, + "patch_similarity": 0.027, "shared_filenames": [ - "src/diffusers/pipelines/z_image/pipeline_z_image.py" + "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/z_image/pipeline_z_image.py", + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", "left_ranges": [ [ - 276, - 282 + 19, + 35 ], [ - 335, - 343 + 201, + 208 + ], + [ + 218, + 225 + ], + [ + 507, + 572 + ], + [ + 636, + 643 + ], + [ + 824, + 834 + ], + [ + 1177, + 1184 + ], + [ + 1273, + 1282 + ], + [ + 1409, + 1416 + ], + [ + 1463, + 1474 + ], + [ + 1740, + 1748 ] ], "right_ranges": [ [ - 276, - 282 + 28, + 39 + ], + [ + 151, + 157 + ], + [ + 163, + 169 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-5712-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5712.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 5712, - "canonical_pr_number": 6438, - "issue_numbers": [ - 5712 - ], - "pr_numbers": [ - 6342, - 6438 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 6342, - "right_pr_number": 6438, - "code_similarity": 0.54, - "size_similarity": 0.917, + "left_pr_number": 7030, + "right_pr_number": 7031, + "code_similarity": 0.707, + "size_similarity": 0.75, "file_overlap": 1.0, - "area_overlap": 0.0, - "patch_similarity": 0.375, + "area_overlap": 0.306, + "patch_similarity": 1.0, "shared_filenames": [ - "docs/source/en/installation.md" + "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" ], "shared_file_areas": [ { - "filename": "docs/source/en/installation.md", + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", "left_ranges": [ [ - 48, - 54 + 28, + 34 + ], + [ + 146, + 152 + ], + [ + 158, + 164 ] ], "right_ranges": [ [ - 37, - 46 + 28, + 39 + ], + [ + 151, + 157 + ], + [ + 163, + 169 ] ] } @@ -8718,18 +6405,19 @@ ] }, { - "cluster_id": "cluster-9841-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #9841.", - "status": "open", + "cluster_id": "cluster-5262-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #5262.", + "status": "merged", "confidence": 0.75, - "canonical_issue_number": 9841, - "canonical_pr_number": 9879, + "canonical_issue_number": 5262, + "canonical_pr_number": 5370, "issue_numbers": [ - 9841 + 5262 ], "pr_numbers": [ - 9842, - 9879 + 5288, + 5366, + 5370 ], "evidence_types": [ "closing_reference", @@ -8737,207 +6425,183 @@ ], "pr_comparisons": [ { - "left_pr_number": 9842, - "right_pr_number": 9879, - "code_similarity": 0.839, - "size_similarity": 0.891, + "left_pr_number": 5288, + "right_pr_number": 5366, + "code_similarity": 0.344, + "size_similarity": 0.222, "file_overlap": 1.0, - "area_overlap": 0.767, - "patch_similarity": 0.614, + "area_overlap": 0.0, + "patch_similarity": 0.0, "shared_filenames": [ - "examples/dreambooth/train_dreambooth_lora_sdxl.py" + "tests/pipelines/stable_diffusion/test_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", "left_ranges": [ [ - 1183, - 1218 + 90, + 97 ] ], "right_ranges": [ [ - 67, - 73 + 103, + 129 ], [ - 1184, - 1216 + 134, + 170 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-13292-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #13292.", - "status": "open", - "confidence": 0.6, - "canonical_issue_number": 13292, - "canonical_pr_number": 13415, - "issue_numbers": [ - 13292 - ], - "pr_numbers": [ - 13307, - 13415 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 13307, - "right_pr_number": 13415, - "code_similarity": 0.48, - "size_similarity": 0.978, - "file_overlap": 0.667, - "area_overlap": 0.138, - "patch_similarity": 0.238, + "left_pr_number": 5288, + "right_pr_number": 5370, + "code_similarity": 0.356, + "size_similarity": 0.279, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, "shared_filenames": [ - "examples/dreambooth/train_dreambooth_lora_flux2.py", - "examples/dreambooth/train_dreambooth_lora_flux2_klein.py" + "tests/pipelines/stable_diffusion/test_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "examples/dreambooth/train_dreambooth_lora_flux2.py", + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", "left_ranges": [ [ - 1734, - 1743 + 90, + 97 ] ], "right_ranges": [ [ - 1740, - 1751 + 82, + 88 ], [ - 1812, - 1822 - ] - ] - }, - { - "filename": "examples/dreambooth/train_dreambooth_lora_flux2_klein.py", - "left_ranges": [ + 104, + 118 + ], [ - 1674, - 1683 - ] - ], - "right_ranges": [ + 123, + 145 + ], [ - 1680, - 1691 + 186, + 192 ], [ - 1755, - 1765 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-12656-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #12656.", - "status": "open", - "confidence": 0.8, - "canonical_issue_number": 12656, - "canonical_pr_number": 13110, - "issue_numbers": [ - 12656 - ], - "pr_numbers": [ - 13110, - 13516 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 13110, - "right_pr_number": 13516, - "code_similarity": 0.696, - "size_similarity": 0.696, - "file_overlap": 1.0, - "area_overlap": 0.632, - "patch_similarity": 0.237, - "shared_filenames": [ - "src/diffusers/models/transformers/auraflow_transformer_2d.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", - "left_ranges": [ + 320, + 326 + ], [ - 76, - 87 - ] - ], - "right_ranges": [ + 338, + 344 + ], + [ + 378, + 384 + ], + [ + 397, + 403 + ], + [ + 416, + 422 + ], + [ + 488, + 494 + ], + [ + 641, + 647 + ], [ - 76, - 94 + 668, + 674 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-12555-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #13062.", - "status": "open", - "confidence": 0.55, - "canonical_issue_number": 13062, - "canonical_pr_number": 12555, - "issue_numbers": [ - 13062 - ], - "pr_numbers": [ - 12555, - 13518 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 12555, - "right_pr_number": 13518, - "code_similarity": 0.599, - "size_similarity": 0.379, - "file_overlap": 0.5, - "area_overlap": 1.0, - "patch_similarity": 0.154, + "left_pr_number": 5366, + "right_pr_number": 5370, + "code_similarity": 0.559, + "size_similarity": 0.796, + "file_overlap": 1.0, + "area_overlap": 0.246, + "patch_similarity": 0.088, "shared_filenames": [ - "src/diffusers/utils/import_utils.py" + "tests/pipelines/stable_diffusion/test_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/utils/import_utils.py", + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", "left_ranges": [ [ - 140, - 146 + 103, + 129 + ], + [ + 134, + 170 ] ], "right_ranges": [ [ - 140, - 146 + 82, + 88 + ], + [ + 104, + 118 + ], + [ + 123, + 145 + ], + [ + 186, + 192 + ], + [ + 320, + 326 + ], + [ + 338, + 344 + ], + [ + 378, + 384 + ], + [ + 397, + 403 + ], + [ + 416, + 422 + ], + [ + 488, + 494 + ], + [ + 641, + 647 + ], + [ + 668, + 674 ] ] } @@ -8946,505 +6610,380 @@ ] }, { - "cluster_id": "cluster-7730-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #7730.", + "cluster_id": "cluster-916-4", + "summary": "Cluster of 4 related pull requests linked by soft_similarity.", "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 7730, - "canonical_pr_number": 11156, - "issue_numbers": [ - 7730 - ], + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 1110, + "issue_numbers": [], "pr_numbers": [ - 7742, - 11156 + 916, + 1007, + 1110, + 1225 ], "evidence_types": [ - "closing_reference" + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 7742, - "right_pr_number": 11156, - "code_similarity": 0.633, - "size_similarity": 0.833, - "file_overlap": 0.917, - "area_overlap": 0.188, - "patch_similarity": 0.837, + "left_pr_number": 916, + "right_pr_number": 1007, + "code_similarity": 0.334, + "size_similarity": 0.238, + "file_overlap": 0.833, + "area_overlap": 0.065, + "patch_similarity": 0.091, "shared_filenames": [ - "examples/instruct_pix2pix/train_instruct_pix2pix.py", - "examples/research_projects/instructpix2pix_lora/train_instruct_pix2pix_lora.py", - "examples/research_projects/promptdiffusion/convert_original_promptdiffusion_to_diffusers.py", - "scripts/convert_dance_diffusion_to_diffusers.py", - "scripts/convert_vae_pt_to_diffusers.py", - "src/diffusers/loaders/single_file_utils.py", - "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", - "src/diffusers/utils/loading_utils.py", - "src/diffusers/utils/testing_utils.py", - "utils/fetch_latest_release_branch.py", - "utils/notify_slack_about_release.py" + "examples/unconditional_image_generation/train_unconditional.py", + "src/diffusers/models/unet_2d.py", + "src/diffusers/pipeline_utils.py", + "src/diffusers/pipelines/ddpm/pipeline_ddpm.py", + "tests/test_models_unet.py" ], "shared_file_areas": [ { - "filename": "examples/instruct_pix2pix/train_instruct_pix2pix.py", + "filename": "examples/unconditional_image_generation/train_unconditional.py", "left_ranges": [ [ - 418, - 424 + 138, + 144 ] ], "right_ranges": [ [ - 49, - 55 + 59, + 69 ], [ - 419, - 425 + 144, + 150 + ], + [ + 231, + 237 ] ] }, { - "filename": "examples/research_projects/instructpix2pix_lora/train_instruct_pix2pix_lora.py", + "filename": "src/diffusers/models/unet_2d.py", "left_ranges": [ [ - 378, - 384 + 258, + 261 ] ], "right_ranges": [ [ - 54, - 60 - ], - [ - 476, - 482 + 245, + 248 ] ] }, { - "filename": "examples/research_projects/promptdiffusion/convert_original_promptdiffusion_to_diffusers.py", + "filename": "src/diffusers/pipeline_utils.py", "left_ranges": [ [ - 1435, - 1441 + 72, + 80 ] ], "right_ranges": [ [ - 59, - 65 - ], - [ - 1436, - 1442 + 53, + 61 ] ] }, { - "filename": "scripts/convert_dance_diffusion_to_diffusers.py", + "filename": "src/diffusers/pipelines/ddpm/pipeline_ddpm.py", "left_ranges": [ [ - 74, - 80 + 77, + 83 ] ], "right_ranges": [ [ - 11, - 17 - ], - [ - 75, - 81 + 90, + 96 ] ] }, { - "filename": "scripts/convert_vae_pt_to_diffusers.py", + "filename": "tests/test_models_unet.py", "left_ranges": [ [ - 122, - 128 + 131, + 137 + ], + [ + 141, + 147 + ], + [ + 219, + 225 + ], + [ + 294, + 300 + ], + [ + 412, + 418 + ], + [ + 437, + 443 ] ], "right_ranges": [ [ - 13, - 19 + 77, + 83 ], [ - 123, - 130 - ] - ] - }, - { - "filename": "src/diffusers/loaders/single_file_utils.py", - "left_ranges": [ - [ - 381, - 387 + 129, + 135 ], [ - 402, - 408 - ] - ], - "right_ranges": [ - [ - 44, - 50 + 149, + 155 ], [ - 444, - 450 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", - "left_ranges": [ + 202, + 208 + ], [ - 1324, - 1330 - ] - ], - "right_ranges": [ + 216, + 222 + ], [ - 52, - 58 + 338, + 344 ], [ - 1325, - 1331 + 363, + 369 ] ] - }, + } + ] + }, + { + "left_pr_number": 916, + "right_pr_number": 1110, + "code_similarity": 0.734, + "size_similarity": 0.919, + "file_overlap": 0.286, + "area_overlap": 0.955, + "patch_similarity": 0.872, + "shared_filenames": [ + "examples/unconditional_image_generation/train_unconditional_ort.py", + "src/diffusers/pipeline_utils.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/utils/loading_utils.py", + "filename": "examples/unconditional_image_generation/train_unconditional_ort.py", "left_ranges": [ [ - 25, - 31 + 1, + 251 ] ], "right_ranges": [ [ - 7, - 13 - ], - [ - 30, - 36 + 1, + 251 ] ] }, { - "filename": "src/diffusers/utils/testing_utils.py", + "filename": "src/diffusers/pipeline_utils.py", "left_ranges": [ [ - 396, - 402 - ], - [ - 417, - 423 - ], - [ - 436, - 442 + 72, + 80 ] ], "right_ranges": [ [ - 26, - 32 - ], - [ - 595, - 601 - ], - [ - 616, - 622 - ], - [ - 635, - 641 + 78, + 86 ] ] - }, + } + ] + }, + { + "left_pr_number": 916, + "right_pr_number": 1225, + "code_similarity": 0.137, + "size_similarity": 0.102, + "file_overlap": 0.333, + "area_overlap": 0.019, + "patch_similarity": 0.063, + "shared_filenames": [ + "examples/unconditional_image_generation/train_unconditional.py", + "src/diffusers/pipeline_utils.py" + ], + "shared_file_areas": [ { - "filename": "utils/fetch_latest_release_branch.py", + "filename": "examples/unconditional_image_generation/train_unconditional.py", "left_ranges": [ [ - 27, - 33 + 138, + 144 ] ], "right_ranges": [ [ - 13, - 24 + 59, + 69 ], [ - 30, - 40 + 144, + 153 + ], + [ + 245, + 251 ] ] }, { - "filename": "utils/notify_slack_about_release.py", + "filename": "src/diffusers/pipeline_utils.py", "left_ranges": [ [ - 26, - 32 - ], - [ - 38, - 44 + 72, + 80 ] ], "right_ranges": [ [ - 17, - 24 - ], - [ - 28, - 34 - ], - [ - 40, - 46 + 53, + 61 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-12540-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #12540.", - "status": "open", - "confidence": 0.55, - "canonical_issue_number": 12540, - "canonical_pr_number": 13109, - "issue_numbers": [ - 12540 - ], - "pr_numbers": [ - 13109, - 13517 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 13109, - "right_pr_number": 13517, - "code_similarity": 0.441, - "size_similarity": 0.667, - "file_overlap": 1.0, + "left_pr_number": 1007, + "right_pr_number": 1110, + "code_similarity": 0.103, + "size_similarity": 0.259, + "file_overlap": 0.143, "area_overlap": 0.0, - "patch_similarity": 0.052, + "patch_similarity": 0.057, "shared_filenames": [ - "src/diffusers/models/transformers/sana_transformer.py" + "src/diffusers/pipeline_utils.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/models/transformers/sana_transformer.py", + "filename": "src/diffusers/pipeline_utils.py", "left_ranges": [ [ - 108, - 121 + 53, + 61 ] ], "right_ranges": [ [ - 457, - 475 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-10076-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #10076.", - "status": "open", - "confidence": 0.55, - "canonical_issue_number": 10076, - "canonical_pr_number": 10449, - "issue_numbers": [ - 10076 - ], - "pr_numbers": [ - 10449, - 13435 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ - { - "left_pr_number": 10449, - "right_pr_number": 13435, - "code_similarity": 0.247, - "size_similarity": 0.979, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.342, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-13411-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #13411.", - "status": "open", - "confidence": 0.55, - "canonical_issue_number": 13411, - "canonical_pr_number": 13466, - "issue_numbers": [ - 13411 - ], - "pr_numbers": [ - 13412, - 13466 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ + 78, + 86 + ] + ] + } + ] + }, { - "left_pr_number": 13412, - "right_pr_number": 13466, - "code_similarity": 0.138, - "size_similarity": 0.05, - "file_overlap": 0.25, - "area_overlap": 0.12, - "patch_similarity": 0.075, + "left_pr_number": 1007, + "right_pr_number": 1225, + "code_similarity": 0.494, + "size_similarity": 0.43, + "file_overlap": 0.4, + "area_overlap": 0.614, + "patch_similarity": 0.486, "shared_filenames": [ - "src/diffusers/schedulers/scheduling_ltx_euler_ancestral_rf.py" + "examples/unconditional_image_generation/train_unconditional.py", + "src/diffusers/pipeline_utils.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/schedulers/scheduling_ltx_euler_ancestral_rf.py", + "filename": "examples/unconditional_image_generation/train_unconditional.py", "left_ranges": [ [ - 1, - 4 + 59, + 69 ], [ - 65, - 73 + 144, + 150 ], [ - 83, - 97 - ], + 231, + 237 + ] + ], + "right_ranges": [ [ - 237, - 259 + 59, + 69 ], [ - 271, - 278 + 144, + 153 ], [ - 358, - 377 + 245, + 251 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 53, + 61 ] ], "right_ranges": [ [ - 233, - 246 + 53, + 61 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-11208-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #11208.", - "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 11208, - "canonical_pr_number": 11251, - "issue_numbers": [ - 11208 - ], - "pr_numbers": [ - 11251, - 12400 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 11251, - "right_pr_number": 12400, - "code_similarity": 0.336, - "size_similarity": 0.584, - "file_overlap": 0.5, - "area_overlap": 0.087, - "patch_similarity": 0.261, + "left_pr_number": 1110, + "right_pr_number": 1225, + "code_similarity": 0.106, + "size_similarity": 0.111, + "file_overlap": 0.25, + "area_overlap": 0.0, + "patch_similarity": 0.059, "shared_filenames": [ - "src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py" + "src/diffusers/pipeline_utils.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py", + "filename": "src/diffusers/pipeline_utils.py", "left_ranges": [ [ - 15, - 22 - ], - [ - 41, - 47 - ], - [ - 229, - 236 - ], - [ - 576, - 639 - ], - [ - 716, - 791 - ], - [ - 1157, - 1180 - ], - [ - 1242, - 1267 + 78, + 86 ] ], "right_ranges": [ [ - 1119, - 1171 - ], - [ - 1174, - 1183 + 53, + 61 ] ] } @@ -9453,18 +6992,19 @@ ] }, { - "cluster_id": "cluster-12913-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #12913.", - "status": "open", + "cluster_id": "cluster-5062-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #5062.", + "status": "merged", "confidence": 0.8, - "canonical_issue_number": 12913, - "canonical_pr_number": 13513, + "canonical_issue_number": 5062, + "canonical_pr_number": 5066, "issue_numbers": [ - 12913 + 5062 ], "pr_numbers": [ - 13272, - 13513 + 5065, + 5066, + 5332 ], "evidence_types": [ "closing_reference", @@ -9473,117 +7013,199 @@ ], "pr_comparisons": [ { - "left_pr_number": 13272, - "right_pr_number": 13513, - "code_similarity": 0.45, - "size_similarity": 0.2, - "file_overlap": 0.5, - "area_overlap": 0.667, - "patch_similarity": 0.18, + "left_pr_number": 5065, + "right_pr_number": 5066, + "code_similarity": 0.257, + "size_similarity": 0.089, + "file_overlap": 0.15, + "area_overlap": 0.543, + "patch_similarity": 0.028, "shared_filenames": [ - "src/diffusers/pipelines/lumina2/pipeline_lumina2.py" + "examples/community/run_onnx_controlnet.py", + "examples/community/run_tensorrt_controlnet.py", + "src/diffusers/models/attention_processor.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/lumina2/pipeline_lumina2.py", + "filename": "examples/community/run_onnx_controlnet.py", "left_ranges": [ [ - 696, - 703 + 8, + 14 ] ], "right_ranges": [ [ - 696, - 707 + 8, + 13 + ], + [ + 21, + 27 + ] + ] + }, + { + "filename": "examples/community/run_tensorrt_controlnet.py", + "left_ranges": [ + [ + 11, + 17 + ] + ], + "right_ranges": [ + [ + 11, + 16 + ], + [ + 25, + 31 + ] + ] + }, + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 382, + 388 + ] + ], + "right_ranges": [ + [ + 382, + 388 ] ] } ] + }, + { + "left_pr_number": 5065, + "right_pr_number": 5332, + "code_similarity": 0.003, + "size_similarity": 0.015, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.003, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5066, + "right_pr_number": 5332, + "code_similarity": 0.033, + "size_similarity": 0.167, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-4704-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #4704.", - "status": "closed", + "cluster_id": "cluster-6123-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6123.", + "status": "merged", "confidence": 0.55, - "canonical_issue_number": 4704, - "canonical_pr_number": 4926, + "canonical_issue_number": 6123, + "canonical_pr_number": 6328, "issue_numbers": [ - 4704 + 6123 ], "pr_numbers": [ - 4707, - 4926 + 6328, + 6509 ], "evidence_types": [ "closing_reference" ], "pr_comparisons": [ { - "left_pr_number": 4707, - "right_pr_number": 4926, - "code_similarity": 0.197, - "size_similarity": 0.962, + "left_pr_number": 6328, + "right_pr_number": 6509, + "code_similarity": 0.241, + "size_similarity": 0.686, "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.029, + "patch_similarity": 0.693, "shared_filenames": [], "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-9610-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #9610.", + "cluster_id": "cluster-7736-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #7736.", "status": "merged", "confidence": 0.55, - "canonical_issue_number": 9610, - "canonical_pr_number": 9659, + "canonical_issue_number": 7736, + "canonical_pr_number": 7761, "issue_numbers": [ - 9610 + 7736 ], "pr_numbers": [ - 9659, - 9843 + 7739, + 7761 ], "evidence_types": [ "closing_reference" ], "pr_comparisons": [ { - "left_pr_number": 9659, - "right_pr_number": 9843, - "code_similarity": 0.292, - "size_similarity": 0.331, - "file_overlap": 0.5, - "area_overlap": 0.11, - "patch_similarity": 0.246, + "left_pr_number": 7739, + "right_pr_number": 7761, + "code_similarity": 0.155, + "size_similarity": 0.462, + "file_overlap": 0.05, + "area_overlap": 0.071, + "patch_similarity": 0.15, "shared_filenames": [ - "tests/lora/test_lora_layers_sd3.py" + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "tests/lora/test_lora_layers_sd3.py", + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", "left_ranges": [ [ - 12, - 40 + 19, + 25 ], [ - 45, - 54 + 122, + 128 ], [ - 128, - 215 + 254, + 261 + ], + [ + 969, + 984 + ], + [ + 1023, + 1032 ] ], "right_ranges": [ [ - 166, - 179 + 11, + 24 + ], + [ + 775, + 783 + ], + [ + 847, + 857 + ], + [ + 883, + 891 ] ] } @@ -9592,2271 +7214,2359 @@ ] }, { - "cluster_id": "cluster-13243-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #13243.", - "status": "open", - "confidence": 0.75, - "canonical_issue_number": 13243, - "canonical_pr_number": 13512, + "cluster_id": "cluster-6158-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6158.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 6158, + "canonical_pr_number": 6721, "issue_numbers": [ - 13243 + 6158 ], "pr_numbers": [ - 13246, - 13512 + 6195, + 6721 ], "evidence_types": [ "closing_reference", - "shared_issue_target" + "shared_issue_target", + "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 13246, - "right_pr_number": 13512, - "code_similarity": 0.475, - "size_similarity": 0.26, - "file_overlap": 0.5, - "area_overlap": 0.714, - "patch_similarity": 0.152, - "shared_filenames": [ - "src/diffusers/schedulers/scheduling_flow_match_euler_discrete.py" - ], - "shared_file_areas": [ - { - "filename": "src/diffusers/schedulers/scheduling_flow_match_euler_discrete.py", - "left_ranges": [ - [ - 127, - 137 - ], - [ - 144, - 149 - ] - ], - "right_ranges": [ - [ - 127, - 139 - ], - [ - 146, - 151 - ] - ] - } - ] + "left_pr_number": 6195, + "right_pr_number": 6721, + "code_similarity": 0.215, + "size_similarity": 0.615, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.61, + "shared_filenames": [], + "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-13445-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #13445.", - "status": "open", - "confidence": 0.8, - "canonical_issue_number": 13445, - "canonical_pr_number": 13455, - "issue_numbers": [ - 13445 - ], + "cluster_id": "cluster-6299-4", + "summary": "Cluster of 4 related pull requests linked by soft_similarity.", + "status": "closed", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 6322, + "issue_numbers": [], "pr_numbers": [ - 13455, - 13509 + 6299, + 6300, + 6301, + 6322 ], "evidence_types": [ - "closing_reference", - "shared_issue_target", "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 13455, - "right_pr_number": 13509, + "left_pr_number": 6299, + "right_pr_number": 6300, "code_similarity": 1.0, "size_similarity": 1.0, "file_overlap": 1.0, "area_overlap": 1.0, "patch_similarity": 1.0, "shared_filenames": [ - "examples/dreambooth/train_dreambooth_lora_flux2_klein.py" + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" ], "shared_file_areas": [ { - "filename": "examples/dreambooth/train_dreambooth_lora_flux2_klein.py", + "filename": "examples/text_to_image/train_text_to_image.py", + "left_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ], + "right_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", "left_ranges": [ [ - 358, - 364 + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 ] ], "right_ranges": [ [ - 358, - 364 + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-4553-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #4553.", - "status": "closed", - "confidence": 0.6, - "canonical_issue_number": 4553, - "canonical_pr_number": 4558, - "issue_numbers": [ - 4553 - ], - "pr_numbers": [ - 4558, - 4603 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 4558, - "right_pr_number": 4603, - "code_similarity": 0.423, - "size_similarity": 0.12, - "file_overlap": 0.125, + "left_pr_number": 6299, + "right_pr_number": 6301, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, "area_overlap": 1.0, - "patch_similarity": 0.078, + "patch_similarity": 1.0, "shared_filenames": [ - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py" + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", + "filename": "examples/text_to_image/train_text_to_image.py", "left_ranges": [ [ - 540, - 546 - ] - ], - "right_ranges": [ + 41, + 47 + ], [ - 540, - 546 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-5214-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5216.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 5216, - "canonical_pr_number": 5305, - "issue_numbers": [ - 5216 - ], - "pr_numbers": [ - 5214, - 5305 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 5214, - "right_pr_number": 5305, - "code_similarity": 0.064, - "size_similarity": 0.022, - "file_overlap": 0.019, - "area_overlap": 0.144, - "patch_similarity": 0.026, - "shared_filenames": [ - "examples/custom_diffusion/train_custom_diffusion.py", - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen.py", - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen_text_image.py" - ], - "shared_file_areas": [ - { - "filename": "examples/custom_diffusion/train_custom_diffusion.py", - "left_ranges": [ + 152, + 158 + ], [ - 1075, - 1104 - ] - ], - "right_ranges": [ + 166, + 179 + ], [ - 207, - 213 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen.py", - "left_ranges": [ + 485, + 515 + ], [ - 26, - 32 + 575, + 586 ], [ - 274, - 283 + 592, + 608 ], [ - 402, - 411 + 1014, + 1021 ], [ - 803, - 811 + 1036, + 1071 ], [ - 840, - 847 + 1125, + 1136 ] ], "right_ranges": [ [ - 803, - 811 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen_text_image.py", - "left_ranges": [ + 41, + 47 + ], [ - 32, - 38 + 152, + 158 ], [ - 306, - 315 + 166, + 179 ], [ - 434, - 443 + 485, + 515 ], [ - 965, - 973 + 575, + 586 + ], + [ + 592, + 608 ], [ 1014, 1021 - ] - ], - "right_ranges": [ + ], [ - 965, - 973 + 1036, + 1071 + ], + [ + 1125, + 1136 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-5302-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5302.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 5302, - "canonical_pr_number": 5304, - "issue_numbers": [ - 5302 - ], - "pr_numbers": [ - 5303, - 5304 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 5303, - "right_pr_number": 5304, - "code_similarity": 0.093, - "size_similarity": 0.059, - "file_overlap": 0.167, - "area_overlap": 0.089, - "patch_similarity": 0.0, - "shared_filenames": [ - "src/diffusers/pipelines/deepfloyd_if/pipeline_if.py" - ], - "shared_file_areas": [ + }, { - "filename": "src/diffusers/pipelines/deepfloyd_if/pipeline_if.py", + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", "left_ranges": [ [ - 158, - 165 + 34, + 40 ], [ - 540, - 546 + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 ] ], "right_ranges": [ [ - 161, - 171 + 34, + 40 ], [ - 174, - 187 + 54, + 60 ], [ - 193, - 200 + 580, + 593 ], [ - 570, - 582 + 874, + 920 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-4629-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 4661, - "issue_numbers": [], - "pr_numbers": [ - 4629, - 4659, - 4661 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 4629, - "right_pr_number": 4659, - "code_similarity": 1.0, - "size_similarity": 1.0, + "left_pr_number": 6299, + "right_pr_number": 6322, + "code_similarity": 0.716, + "size_similarity": 0.748, "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, + "area_overlap": 0.422, + "patch_similarity": 0.791, "shared_filenames": [ - "docs/source/en/_toctree.yml", - "docs/source/en/api/models/overview.md", - "docs/source/en/using-diffusers/push_to_hub.md", - "examples/community/README.md", - "examples/community/lpw_stable_diffusion_xl.py", - "examples/controlnet/train_controlnet.py", - "examples/controlnet/train_controlnet_flax.py", - "examples/controlnet/train_controlnet_sdxl.py", - "examples/custom_diffusion/train_custom_diffusion.py", - "examples/dreambooth/train_dreambooth.py", - "examples/dreambooth/train_dreambooth_flax.py", - "examples/dreambooth/train_dreambooth_lora.py", - "examples/dreambooth/train_dreambooth_lora_sdxl.py", - "examples/instruct_pix2pix/train_instruct_pix2pix.py", - "examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py", - "examples/research_projects/controlnet/train_controlnet_webdataset.py", "examples/text_to_image/train_text_to_image.py", - "examples/text_to_image/train_text_to_image_flax.py", - "examples/text_to_image/train_text_to_image_lora.py", - "examples/text_to_image/train_text_to_image_lora_sdxl.py", - "examples/text_to_image/train_text_to_image_sdxl.py", - "examples/textual_inversion/textual_inversion.py", - "examples/textual_inversion/textual_inversion_flax.py", - "examples/unconditional_image_generation/train_unconditional.py", - "setup.py", - "src/diffusers/__init__.py", - "src/diffusers/pipelines/controlnet/multicontrolnet.py", - "src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py", - "src/diffusers/utils/hub_utils.py", - "tests/models/test_lora_layers.py", - "tests/pipelines/controlnet/test_controlnet_sdxl.py" + "examples/text_to_image/train_text_to_image_sdxl.py" ], "shared_file_areas": [ { - "filename": "docs/source/en/_toctree.yml", - "left_ranges": [ - [ - 32, - 37 - ] - ], - "right_ranges": [ - [ - 32, - 37 - ] - ] - }, - { - "filename": "docs/source/en/api/models/overview.md", - "left_ranges": [ - [ - 11, - 16 - ] - ], - "right_ranges": [ - [ - 11, - 16 - ] - ] - }, - { - "filename": "docs/source/en/using-diffusers/push_to_hub.md", + "filename": "examples/text_to_image/train_text_to_image.py", "left_ranges": [ [ - 0, - 0 - ] - ], - "right_ranges": [ + 41, + 47 + ], [ - 0, - 0 - ] - ] - }, - { - "filename": "examples/community/README.md", - "left_ranges": [ + 152, + 158 + ], [ - 39, - 46 + 166, + 179 ], [ - 1530, - 1567 - ] - ], - "right_ranges": [ + 485, + 515 + ], [ - 39, - 46 + 575, + 586 ], [ - 1530, - 1567 - ] - ] - }, - { - "filename": "examples/community/lpw_stable_diffusion_xl.py", - "left_ranges": [ + 592, + 608 + ], [ - 1, - 1297 - ] - ], - "right_ranges": [ + 1014, + 1021 + ], [ - 1, - 1297 - ] - ] - }, - { - "filename": "examples/controlnet/train_controlnet.py", - "left_ranges": [ + 1036, + 1071 + ], [ - 56, - 62 + 1125, + 1136 ] ], "right_ranges": [ [ - 56, - 62 - ] - ] - }, - { - "filename": "examples/controlnet/train_controlnet_flax.py", - "left_ranges": [ + 41, + 47 + ], [ - 59, - 65 - ] - ], - "right_ranges": [ + 152, + 158 + ], [ - 59, - 65 - ] - ] - }, - { - "filename": "examples/controlnet/train_controlnet_sdxl.py", - "left_ranges": [ + 166, + 179 + ], [ - 58, - 64 - ] - ], - "right_ranges": [ + 485, + 520 + ], [ - 58, - 64 - ] - ] - }, - { - "filename": "examples/custom_diffusion/train_custom_diffusion.py", - "left_ranges": [ + 580, + 591 + ], [ - 58, - 64 - ] - ], - "right_ranges": [ + 597, + 613 + ], [ - 58, - 64 - ] - ] - }, - { - "filename": "examples/dreambooth/train_dreambooth.py", - "left_ranges": [ + 634, + 658 + ], [ - 60, - 66 - ] - ], - "right_ranges": [ + 735, + 741 + ], [ - 60, - 66 - ] - ] - }, - { - "filename": "examples/dreambooth/train_dreambooth_flax.py", - "left_ranges": [ + 1018, + 1024 + ], [ - 36, - 42 - ] - ], - "right_ranges": [ + 1033, + 1040 + ], [ - 36, - 42 - ] - ] - }, - { - "filename": "examples/dreambooth/train_dreambooth_lora.py", - "left_ranges": [ + 1055, + 1090 + ], [ - 70, - 76 - ] - ], - "right_ranges": [ + 1093, + 1100 + ], [ - 70, - 76 + 1107, + 1120 + ], + [ + 1144, + 1155 ] ] }, { - "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", "left_ranges": [ [ - 58, - 64 - ] - ], - "right_ranges": [ + 34, + 40 + ], [ - 58, - 64 - ] - ] - }, - { - "filename": "examples/instruct_pix2pix/train_instruct_pix2pix.py", - "left_ranges": [ + 54, + 60 + ], [ - 52, - 58 + 580, + 593 + ], + [ + 874, + 920 ] ], "right_ranges": [ [ - 52, - 58 - ] - ] - }, - { - "filename": "examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py", - "left_ranges": [ + 34, + 41 + ], [ - 55, - 61 - ] - ], - "right_ranges": [ + 54, + 60 + ], [ - 55, - 61 - ] - ] - }, - { - "filename": "examples/research_projects/controlnet/train_controlnet_webdataset.py", - "left_ranges": [ + 172, + 178 + ], [ - 0, - 0 - ] - ], - "right_ranges": [ + 181, + 192 + ], [ - 0, - 0 + 478, + 485 + ], + [ + 586, + 599 + ], + [ + 801, + 812 + ], + [ + 882, + 940 ] ] - }, + } + ] + }, + { + "left_pr_number": 6300, + "right_pr_number": 6301, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ { "filename": "examples/text_to_image/train_text_to_image.py", "left_ranges": [ [ - 53, - 59 - ] - ], - "right_ranges": [ + 41, + 47 + ], [ - 53, - 59 - ] - ] - }, - { - "filename": "examples/text_to_image/train_text_to_image_flax.py", - "left_ranges": [ + 152, + 158 + ], [ - 33, - 39 - ] - ], - "right_ranges": [ + 166, + 179 + ], [ - 33, - 39 - ] - ] - }, - { - "filename": "examples/text_to_image/train_text_to_image_lora.py", - "left_ranges": [ + 485, + 515 + ], [ - 48, - 54 - ] - ], - "right_ranges": [ + 575, + 586 + ], [ - 48, - 54 - ] - ] - }, - { - "filename": "examples/text_to_image/train_text_to_image_lora_sdxl.py", - "left_ranges": [ + 592, + 608 + ], [ - 57, - 63 - ] - ], - "right_ranges": [ + 1014, + 1021 + ], [ - 57, - 63 - ] - ] - }, - { - "filename": "examples/text_to_image/train_text_to_image_sdxl.py", - "left_ranges": [ + 1036, + 1071 + ], [ - 57, - 63 + 1125, + 1136 ] ], "right_ranges": [ [ - 57, - 63 - ] - ] - }, - { - "filename": "examples/textual_inversion/textual_inversion.py", - "left_ranges": [ + 41, + 47 + ], [ - 79, - 85 - ] - ], - "right_ranges": [ + 152, + 158 + ], [ - 79, - 85 - ] - ] - }, - { - "filename": "examples/textual_inversion/textual_inversion_flax.py", - "left_ranges": [ + 166, + 179 + ], [ - 56, - 62 - ] - ], - "right_ranges": [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], [ - 56, - 62 + 1125, + 1136 ] ] }, { - "filename": "examples/unconditional_image_generation/train_unconditional.py", + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", "left_ranges": [ [ - 30, - 36 - ] - ], - "right_ranges": [ + 34, + 40 + ], [ - 30, - 36 - ] - ] - }, - { - "filename": "setup.py", - "left_ranges": [ + 54, + 60 + ], [ - 233, - 239 + 580, + 593 + ], + [ + 874, + 920 ] ], "right_ranges": [ [ - 233, - 239 + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 ] ] - }, + } + ] + }, + { + "left_pr_number": 6300, + "right_pr_number": 6322, + "code_similarity": 0.716, + "size_similarity": 0.748, + "file_overlap": 1.0, + "area_overlap": 0.422, + "patch_similarity": 0.791, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/__init__.py", + "filename": "examples/text_to_image/train_text_to_image.py", "left_ranges": [ [ - 1, - 4 + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 ] ], "right_ranges": [ [ - 1, - 4 + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 520 + ], + [ + 580, + 591 + ], + [ + 597, + 613 + ], + [ + 634, + 658 + ], + [ + 735, + 741 + ], + [ + 1018, + 1024 + ], + [ + 1033, + 1040 + ], + [ + 1055, + 1090 + ], + [ + 1093, + 1100 + ], + [ + 1107, + 1120 + ], + [ + 1144, + 1155 ] ] }, { - "filename": "src/diffusers/pipelines/controlnet/multicontrolnet.py", + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", "left_ranges": [ [ - 39, - 44 + 34, + 40 ], [ - 53, - 58 + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 ] ], "right_ranges": [ [ - 39, - 44 + 34, + 41 ], [ - 53, - 58 + 54, + 60 + ], + [ + 172, + 178 + ], + [ + 181, + 192 + ], + [ + 478, + 485 + ], + [ + 586, + 599 + ], + [ + 801, + 812 + ], + [ + 882, + 940 ] ] - }, + } + ] + }, + { + "left_pr_number": 6301, + "right_pr_number": 6322, + "code_similarity": 0.716, + "size_similarity": 0.748, + "file_overlap": 1.0, + "area_overlap": 0.422, + "patch_similarity": 0.791, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py", + "filename": "examples/text_to_image/train_text_to_image.py", "left_ranges": [ [ - 149, - 163 + 41, + 47 ], [ - 530, - 535 + 152, + 158 ], [ - 540, - 545 + 166, + 179 ], [ - 551, - 564 + 485, + 515 ], [ - 569, - 574 + 575, + 586 ], [ - 606, - 611 + 592, + 608 ], [ - 888, - 893 + 1014, + 1021 ], [ - 933, - 938 + 1036, + 1071 ], [ - 963, - 974 + 1125, + 1136 ] ], "right_ranges": [ [ - 149, - 163 + 41, + 47 ], [ - 530, - 535 + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 520 + ], + [ + 580, + 591 + ], + [ + 597, + 613 + ], + [ + 634, + 658 ], [ - 540, - 545 + 735, + 741 ], [ - 551, - 564 + 1018, + 1024 ], [ - 569, - 574 + 1033, + 1040 ], [ - 606, - 611 + 1055, + 1090 ], [ - 888, - 893 + 1093, + 1100 ], [ - 933, - 938 + 1107, + 1120 ], [ - 963, - 974 + 1144, + 1155 ] ] }, { - "filename": "src/diffusers/utils/hub_utils.py", + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", "left_ranges": [ [ - 410, - 432 - ] - ], - "right_ranges": [ + 34, + 40 + ], [ - 410, - 432 - ] - ] - }, - { - "filename": "tests/models/test_lora_layers.py", - "left_ranges": [ + 54, + 60 + ], [ - 664, - 669 + 580, + 593 + ], + [ + 874, + 920 ] ], "right_ranges": [ [ - 664, - 669 - ] - ] - }, - { - "filename": "tests/pipelines/controlnet/test_controlnet_sdxl.py", - "left_ranges": [ + 34, + 41 + ], [ - 26, - 31 + 54, + 60 ], [ - 46, - 52 + 172, + 178 ], [ - 297, - 299 - ] - ], - "right_ranges": [ + 181, + 192 + ], [ - 26, - 31 + 478, + 485 ], [ - 46, - 52 + 586, + 599 ], [ - 297, - 299 + 801, + 812 + ], + [ + 882, + 940 ] ] } ] - }, + } + ] + }, + { + "cluster_id": "cluster-3896-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 4097, + "issue_numbers": [], + "pr_numbers": [ + 3896, + 4016, + 4097 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ { - "left_pr_number": 4629, - "right_pr_number": 4661, - "code_similarity": 0.514, - "size_similarity": 0.357, - "file_overlap": 0.065, - "area_overlap": 0.996, - "patch_similarity": 0.5, + "left_pr_number": 3896, + "right_pr_number": 4016, + "code_similarity": 0.196, + "size_similarity": 0.42, + "file_overlap": 0.095, + "area_overlap": 0.008, + "patch_similarity": 0.535, "shared_filenames": [ - "examples/community/README.md", - "examples/community/lpw_stable_diffusion_xl.py" + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py" ], "shared_file_areas": [ { - "filename": "examples/community/README.md", + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", "left_ranges": [ [ - 39, - 46 - ], - [ - 1530, - 1567 + 1, + 805 ] ], "right_ranges": [ [ - 39, - 46 - ], - [ - 1530, - 1573 + 73, + 79 ] ] }, { - "filename": "examples/community/lpw_stable_diffusion_xl.py", + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py", "left_ranges": [ [ 1, - 1297 + 869 ] ], "right_ranges": [ [ - 1, - 1297 + 78, + 84 ] ] } ] }, { - "left_pr_number": 4659, - "right_pr_number": 4661, - "code_similarity": 0.514, - "size_similarity": 0.357, - "file_overlap": 0.065, - "area_overlap": 0.996, - "patch_similarity": 0.5, + "left_pr_number": 3896, + "right_pr_number": 4097, + "code_similarity": 0.096, + "size_similarity": 0.214, + "file_overlap": 0.05, + "area_overlap": 0.008, + "patch_similarity": 0.237, "shared_filenames": [ - "examples/community/README.md", - "examples/community/lpw_stable_diffusion_xl.py" + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py" ], "shared_file_areas": [ { - "filename": "examples/community/README.md", + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", "left_ranges": [ [ - 39, - 46 - ], - [ - 1530, - 1567 + 1, + 805 ] ], "right_ranges": [ [ - 39, - 46 + 13, + 19 ], [ - 1530, - 1573 + 810, + 875 ] ] - }, + } + ] + }, + { + "left_pr_number": 4016, + "right_pr_number": 4097, + "code_similarity": 0.366, + "size_similarity": 0.509, + "file_overlap": 0.4, + "area_overlap": 0.299, + "patch_similarity": 0.267, + "shared_filenames": [ + "examples/dreambooth/README_sdxl.md", + "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "examples/test_examples.py", + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py" + ], + "shared_file_areas": [ { - "filename": "examples/community/lpw_stable_diffusion_xl.py", + "filename": "examples/dreambooth/README_sdxl.md", "left_ranges": [ [ 1, - 1297 + 178 ] ], "right_ranges": [ [ - 1, - 1297 + 164, + 180 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-5725-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 5748, - "issue_numbers": [], - "pr_numbers": [ - 5725, - 5744, - 5748 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 5725, - "right_pr_number": 5744, - "code_similarity": 0.654, - "size_similarity": 0.479, - "file_overlap": 0.5, - "area_overlap": 0.89, - "patch_similarity": 0.643, - "shared_filenames": [ - "docs/source/en/_toctree.yml", - "docs/source/en/api/pipelines/alt_diffusion.md", - "docs/source/en/api/pipelines/animatediff.md", - "docs/source/en/api/pipelines/attend_and_excite.md", - "docs/source/en/api/pipelines/audio_diffusion.md", - "docs/source/en/api/pipelines/audioldm.md", - "docs/source/en/api/pipelines/audioldm2.md", - "docs/source/en/api/pipelines/auto_pipeline.md", - "docs/source/en/api/pipelines/blip_diffusion.md", - "docs/source/en/api/pipelines/consistency_models.md", - "docs/source/en/api/pipelines/controlnet.md", - "docs/source/en/api/pipelines/controlnet_sdxl.md", - "docs/source/en/api/pipelines/cycle_diffusion.md", - "docs/source/en/api/pipelines/dance_diffusion.md", - "docs/source/en/api/pipelines/ddim.md", - "docs/source/en/api/pipelines/deepfloyd_if.md", - "docs/source/en/api/pipelines/diffedit.md", - "docs/source/en/api/pipelines/dit.md", - "docs/source/en/api/pipelines/kandinsky.md", - "docs/source/en/api/pipelines/kandinsky_v22.md", - "docs/source/en/api/pipelines/latent_consistency_models.md", - "docs/source/en/api/pipelines/latent_diffusion.md", - "docs/source/en/api/pipelines/model_editing.md", - "docs/source/en/api/pipelines/musicldm.md", - "docs/source/en/api/pipelines/overview.md", - "docs/source/en/api/pipelines/paint_by_example.md", - "docs/source/en/api/pipelines/panorama.md" - ], - "shared_file_areas": [ + }, { - "filename": "docs/source/en/_toctree.yml", + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", "left_ranges": [ [ - 227, - 233 + 1, + 1266 + ] + ], + "right_ranges": [ + [ + 16, + 22 + ], + [ + 46, + 56 + ], + [ + 64, + 70 + ], + [ + 92, + 99 + ], + [ + 128, + 139 ], [ - 263, - 275 - ] - ], - "right_ranges": [ + 424, + 455 + ], [ - 231, - 237 + 461, + 466 ], [ - 267, - 279 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/alt_diffusion.md", - "left_ranges": [ + 485, + 513 + ], [ - 16, - 22 + 528, + 556 ], [ - 44, - 47 - ] - ], - "right_ranges": [ + 602, + 607 + ], [ - 16, - 22 + 635, + 641 ], [ - 44, - 47 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/animatediff.md", - "left_ranges": [ + 700, + 713 + ], [ - 14, - 24 + 729, + 738 ], [ - 28, - 34 + 772, + 837 ], [ - 211, - 221 + 861, + 871 ], [ - 232, - 234 - ] - ], - "right_ranges": [ + 874, + 954 + ], [ - 14, - 24 + 969, + 989 ], [ - 28, - 34 + 1042, + 1050 ], [ - 211, - 221 + 1053, + 1068 ], [ - 232, - 234 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/attend_and_excite.md", - "left_ranges": [ + 1078, + 1107 + ], [ - 16, - 22 + 1129, + 1139 ], [ - 34, - 37 - ] - ], - "right_ranges": [ + 1184, + 1205 + ], [ - 16, - 22 + 1227, + 1237 ], [ - 34, - 37 + 1257, + 1288 + ], + [ + 1334, + 1340 ] ] }, { - "filename": "docs/source/en/api/pipelines/audio_diffusion.md", + "filename": "examples/test_examples.py", "left_ranges": [ [ - 14, - 19 + 353, + 390 ] ], "right_ranges": [ [ - 14, - 19 + 385, + 426 ] ] }, { - "filename": "docs/source/en/api/pipelines/audioldm.md", + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", "left_ranges": [ [ - 19, - 27 - ], - [ - 47, - 50 + 73, + 79 ] ], "right_ranges": [ [ - 19, - 27 + 13, + 19 ], [ - 47, - 50 + 810, + 875 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-9403-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #9403.", + "status": "merged", + "confidence": 0.75, + "canonical_issue_number": 9403, + "canonical_pr_number": 10261, + "issue_numbers": [ + 9403 + ], + "pr_numbers": [ + 9787, + 10261 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 9787, + "right_pr_number": 10261, + "code_similarity": 0.589, + "size_similarity": 0.715, + "file_overlap": 0.583, + "area_overlap": 0.468, + "patch_similarity": 0.715, + "shared_filenames": [ + "scripts/convert_flux_xlabs_ipadapter_to_diffusers.py", + "src/diffusers/loaders/__init__.py", + "src/diffusers/loaders/ip_adapter.py", + "src/diffusers/loaders/transformer_flux.py", + "src/diffusers/models/attention_processor.py", + "src/diffusers/models/transformers/transformer_flux.py", + "src/diffusers/pipelines/flux/pipeline_flux.py" + ], + "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/audioldm2.md", + "filename": "scripts/convert_flux_xlabs_ipadapter_to_diffusers.py", "left_ranges": [ [ - 12, - 34 - ], - [ - 41, - 47 - ], - [ - 50, - 56 - ], - [ - 75, - 78 + 1, + 97 ] ], "right_ranges": [ [ - 12, - 34 - ], - [ - 41, - 47 - ], - [ - 50, - 56 - ], - [ - 75, - 78 + 1, + 97 ] ] }, { - "filename": "docs/source/en/api/pipelines/auto_pipeline.md", + "filename": "src/diffusers/loaders/__init__.py", "left_ranges": [ [ - 35, - 52 - ], - [ - 56, + 55, 61 ], [ - 69, - 71 + 70, + 89 ] ], "right_ranges": [ [ - 35, - 52 + 55, + 61 ], [ - 56, - 61 + 77, + 83 ], [ - 69, - 71 + 87, + 100 ] ] }, { - "filename": "docs/source/en/api/pipelines/blip_diffusion.md", + "filename": "src/diffusers/loaders/ip_adapter.py", "left_ranges": [ [ - 1, - 24 + 41, + 48 + ], + [ + 348, + 611 ] ], "right_ranges": [ [ - 1, - 25 + 38, + 45 + ], + [ + 355, + 644 ] ] }, { - "filename": "docs/source/en/api/pipelines/consistency_models.md", + "filename": "src/diffusers/loaders/transformer_flux.py", "left_ranges": [ [ 1, - 22 - ], - [ - 30, - 36 - ], - [ - 39, - 61 + 174 ] ], "right_ranges": [ [ 1, - 22 - ], - [ - 39, - 56 + 179 ] ] }, { - "filename": "docs/source/en/api/pipelines/controlnet.md", + "filename": "src/diffusers/models/attention_processor.py", "left_ranges": [ [ - 12, - 24 + 482, + 488 ], [ - 67, - 72 + 1987, + 2215 ], [ - 75, - 78 + 4614, + 4620 ] ], "right_ranges": [ [ - 12, - 24 + 575, + 581 ], [ - 67, - 72 + 2653, + 2801 ], [ - 75, - 78 + 6039, + 6045 ] ] }, { - "filename": "docs/source/en/api/pipelines/controlnet_sdxl.md", + "filename": "src/diffusers/models/transformers/transformer_flux.py", "left_ranges": [ [ - 12, - 24 - ], - [ - 28, - 34 + 21, + 27 ], [ - 50, - 55 + 207, + 215 ] ], "right_ranges": [ [ - 12, - 24 + 21, + 27 ], [ - 28, - 34 + 177, + 194 ], [ - 50, - 55 + 200, + 207 + ], + [ + 219, + 227 + ], + [ + 491, + 501 ] ] }, { - "filename": "docs/source/en/api/pipelines/cycle_diffusion.md", + "filename": "src/diffusers/pipelines/flux/pipeline_flux.py", "left_ranges": [ [ - 16, - 22 - ], - [ - 30, + 17, 33 - ] - ], - "right_ranges": [ + ], [ - 16, - 22 + 149, + 155 ], [ - 30, - 33 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/dance_diffusion.md", - "left_ranges": [ + 190, + 197 + ], [ - 16, - 21 + 203, + 210 ], [ - 28, - 31 - ] - ], - "right_ranges": [ + 387, + 436 + ], [ - 16, - 21 + 611, + 618 ], [ - 29, - 32 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/ddim.md", - "left_ranges": [ + 662, + 672 + ], [ - 26, - 29 + 783, + 799 ] ], "right_ranges": [ [ - 26, - 29 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/deepfloyd_if.md", - "left_ranges": [ + 17, + 33 + ], [ - 10, - 40 + 149, + 155 ], [ - 47, - 53 + 177, + 184 ], [ - 80, - 90 + 190, + 197 ], [ - 115, - 157 + 203, + 210 ], [ - 182, - 211 + 389, + 448 ], [ - 218, - 233 + 477, + 509 ], [ - 257, - 288 + 632, + 640 ], [ - 319, - 331 + 645, + 656 ], [ - 334, - 340 + 700, + 716 ], [ - 348, - 366 + 748, + 759 ], [ - 397, - 417 + 775, + 781 ], [ - 433, - 446 + 790, + 810 ], [ - 454, - 477 + 846, + 888 + ], + [ + 898, + 919 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9772-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #9772.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 9772, + "canonical_pr_number": 10131, + "issue_numbers": [ + 9772 + ], + "pr_numbers": [ + 9776, + 10131 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9776, + "right_pr_number": 10131, + "code_similarity": 0.789, + "size_similarity": 0.969, + "file_overlap": 0.444, + "area_overlap": 0.947, + "patch_similarity": 0.87, + "shared_filenames": [ + "docs/source/en/_toctree.yml", + "src/diffusers/__init__.py", + "src/diffusers/models/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/controlnet/__init__.py", + "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py", + "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py", + "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/_toctree.yml", + "left_ranges": [ + [ + 217, + 224 ] ], "right_ranges": [ [ - 10, - 40 + 252, + 259 ], [ - 47, - 53 - ], + 368, + 375 + ] + ] + }, + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ [ - 72, - 88 + 88, + 94 ], [ - 113, - 155 + 364, + 372 ], [ - 180, - 209 + 571, + 577 ], [ - 216, - 231 + 825, + 833 + ] + ], + "right_ranges": [ + [ + 91, + 97 ], [ - 255, - 286 + 377, + 385 ], [ - 317, - 329 + 587, + 593 ], [ - 332, - 338 - ], + 851, + 859 + ] + ] + }, + { + "filename": "src/diffusers/models/__init__.py", + "left_ranges": [ [ - 346, - 364 + 39, + 45 ], [ - 395, - 415 - ], + 94, + 100 + ] + ], + "right_ranges": [ [ - 431, - 444 + 44, + 50 ], [ - 452, - 475 + 101, + 107 ] ] }, { - "filename": "docs/source/en/api/pipelines/diffedit.md", + "filename": "src/diffusers/pipelines/__init__.py", "left_ranges": [ [ - 22, - 28 - ], - [ - 42, - 48 + 156, + 164 ], [ - 52, - 55 + 489, + 497 ] ], "right_ranges": [ [ - 22, - 28 - ], - [ - 42, - 48 + 162, + 170 ], [ - 52, - 55 + 498, + 506 ] ] }, { - "filename": "docs/source/en/api/pipelines/dit.md", + "filename": "src/diffusers/pipelines/controlnet/__init__.py", "left_ranges": [ [ - 32, - 35 + 30, + 38 + ], + [ + 60, + 68 ] ], "right_ranges": [ [ - 32, - 35 + 1, + 86 ] ] }, { - "filename": "docs/source/en/api/pipelines/kandinsky.md", + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py", "left_ranges": [ [ - 9, - 15 - ], - [ - 23, - 41 + 1, + 1824 ] ], "right_ranges": [ [ - 9, - 15 - ], - [ - 23, - 41 + 1, + 1801 ] ] }, { - "filename": "docs/source/en/api/pipelines/kandinsky_v22.md", + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py", "left_ranges": [ [ - 9, - 15 - ], - [ - 23, - 34 + 1, + 1500 ] ], "right_ranges": [ [ - 9, - 15 - ], - [ - 23, - 34 + 1, + 1531 ] ] }, { - "filename": "docs/source/en/api/pipelines/latent_consistency_models.md", + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py", "left_ranges": [ [ 1, - 32 + 1624 ] ], "right_ranges": [ [ 1, - 22 + 1646 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6688-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6688.", + "status": "closed", + "confidence": 0.55, + "canonical_issue_number": 6688, + "canonical_pr_number": 7005, + "issue_numbers": [ + 6688 + ], + "pr_numbers": [ + 6844, + 7005 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 6844, + "right_pr_number": 7005, + "code_similarity": 0.471, + "size_similarity": 0.968, + "file_overlap": 0.182, + "area_overlap": 0.304, + "patch_similarity": 0.778, + "shared_filenames": [ + "src/diffusers/models/attention.py", + "src/diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/attention.py", + "left_ranges": [ + [ + 431, + 437 + ], + [ + 475, + 488 + ], + [ + 498, + 510 + ], + [ + 531, + 544 + ] + ], + "right_ranges": [ + [ + 425, + 431 ] ] }, { - "filename": "docs/source/en/api/pipelines/latent_diffusion.md", + "filename": "src/diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py", "left_ranges": [ [ - 18, - 24 + 205, + 211 ], [ - 37, - 40 + 227, + 233 + ], + [ + 331, + 341 ] ], "right_ranges": [ [ - 18, - 24 + 205, + 211 ], [ - 37, - 40 + 227, + 233 + ], + [ + 331, + 339 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-8710-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #8710.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 8710, + "canonical_pr_number": 9386, + "issue_numbers": [ + 8710 + ], + "pr_numbers": [ + 8725, + 9386 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 8725, + "right_pr_number": 9386, + "code_similarity": 0.463, + "size_similarity": 0.85, + "file_overlap": 0.545, + "area_overlap": 0.029, + "patch_similarity": 0.797, + "shared_filenames": [ + "docs/source/en/api/pipelines/pag.md", + "src/diffusers/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/pipelines/pag/__init__.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/pag.md", + "left_ranges": [ + [ + 20, + 30 + ] + ], + "right_ranges": [ + [ + 48, + 58 ] ] }, { - "filename": "docs/source/en/api/pipelines/model_editing.md", + "filename": "src/diffusers/__init__.py", "left_ranges": [ [ - 32, - 35 + 304, + 310 + ], + [ + 703, + 709 ] ], "right_ranges": [ [ - 32, - 35 + 363, + 369 + ], + [ + 835, + 841 ] ] }, { - "filename": "docs/source/en/api/pipelines/musicldm.md", + "filename": "src/diffusers/pipelines/__init__.py", "left_ranges": [ [ - 13, - 29 + 141, + 147 ], [ - 49, - 52 + 492, + 498 ] ], "right_ranges": [ [ - 13, - 29 + 174, + 180 ], [ - 49, - 52 + 596, + 602 ] ] }, { - "filename": "docs/source/en/api/pipelines/overview.md", + "filename": "src/diffusers/pipelines/auto_pipeline.py", "left_ranges": [ [ - 31, - 37 + 47, + 53 ], [ - 47, - 81 + 89, + 95 ] ], "right_ranges": [ [ - 31, - 37 + 66, + 72 ], [ - 47, - 81 + 161, + 167 ] ] }, { - "filename": "docs/source/en/api/pipelines/paint_by_example.md", + "filename": "src/diffusers/pipelines/pag/__init__.py", "left_ranges": [ [ - 10, - 16 - ], - [ - 22, - 28 + 23, + 29 ], [ - 36, - 39 + 37, + 43 ] ], "right_ranges": [ [ - 10, - 16 - ], - [ - 22, - 28 + 34, + 41 ], [ - 36, - 39 + 60, + 66 ] ] }, { - "filename": "docs/source/en/api/pipelines/panorama.md", + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", "left_ranges": [ [ - 22, - 33 - ], - [ - 47, - 50 + 1232, + 1252 ] ], "right_ranges": [ [ - 22, - 33 - ], - [ - 47, - 50 + 1757, + 1777 ] ] } ] - }, + } + ] + }, + { + "cluster_id": "cluster-6891-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6891.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 6891, + "canonical_pr_number": 7615, + "issue_numbers": [ + 6891 + ], + "pr_numbers": [ + 7613, + 7615 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ { - "left_pr_number": 5725, - "right_pr_number": 5748, - "code_similarity": 0.654, - "size_similarity": 0.494, - "file_overlap": 0.5, - "area_overlap": 0.885, - "patch_similarity": 0.635, + "left_pr_number": 7613, + "right_pr_number": 7615, + "code_similarity": 0.319, + "size_similarity": 0.882, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.948, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-7677-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #7677.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 7677, + "canonical_pr_number": 7865, + "issue_numbers": [ + 7677 + ], + "pr_numbers": [ + 7865, + 7898 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 7865, + "right_pr_number": 7898, + "code_similarity": 0.159, + "size_similarity": 0.682, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.15, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-12441-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12441.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 12441, + "canonical_pr_number": 12526, + "issue_numbers": [ + 12441 + ], + "pr_numbers": [ + 12442, + 12526 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 12442, + "right_pr_number": 12526, + "code_similarity": 0.687, + "size_similarity": 0.728, + "file_overlap": 0.684, + "area_overlap": 0.669, + "patch_similarity": 0.68, "shared_filenames": [ - "docs/source/en/api/pipelines/paradigms.md", - "docs/source/en/api/pipelines/pix2pix_zero.md", - "docs/source/en/api/pipelines/pixart.md", - "docs/source/en/api/pipelines/pndm.md", - "docs/source/en/api/pipelines/score_sde_ve.md", - "docs/source/en/api/pipelines/self_attention_guidance.md", - "docs/source/en/api/pipelines/semantic_stable_diffusion.md", - "docs/source/en/api/pipelines/shap_e.md", - "docs/source/en/api/pipelines/spectrogram_diffusion.md", - "docs/source/en/api/pipelines/stable_diffusion/adapter.md", - "docs/source/en/api/pipelines/stable_diffusion/depth2img.md", - "docs/source/en/api/pipelines/stable_diffusion/inpaint.md", - "docs/source/en/api/pipelines/stable_diffusion/latent_upscale.md", - "docs/source/en/api/pipelines/stable_diffusion/overview.md", - "docs/source/en/api/pipelines/stable_diffusion/stable_diffusion_2.md", - "docs/source/en/api/pipelines/stable_diffusion/stable_diffusion_xl.md", - "docs/source/en/api/pipelines/stable_diffusion/text2img.md", - "docs/source/en/api/pipelines/stable_diffusion/upscale.md", - "docs/source/en/api/pipelines/stable_unclip.md", - "docs/source/en/api/pipelines/stochastic_karras_ve.md", - "docs/source/en/api/pipelines/text_to_video.md", - "docs/source/en/api/pipelines/text_to_video_zero.md", - "docs/source/en/api/pipelines/unclip.md", - "docs/source/en/api/pipelines/unidiffuser.md", - "docs/source/en/api/pipelines/value_guided_sampling.md", - "docs/source/en/api/pipelines/versatile_diffusion.md", - "docs/source/en/api/pipelines/wuerstchen.md" + "docs/source/en/api/pipelines/wan.md", + "scripts/convert_wan_to_diffusers.py", + "src/diffusers/__init__.py", + "src/diffusers/models/__init__.py", + "src/diffusers/models/transformers/__init__.py", + "src/diffusers/models/transformers/transformer_wan_animate.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/wan/__init__.py", + "src/diffusers/pipelines/wan/pipeline_wan_animate.py", + "src/diffusers/utils/dummy_pt_objects.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "tests/pipelines/wan/test_wan_animate.py", + "tests/quantization/gguf/test_gguf.py" ], "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/paradigms.md", + "filename": "docs/source/en/api/pipelines/wan.md", "left_ranges": [ [ - 16, - 22 + 40, + 46 ], [ - 26, - 40 + 96, + 110 + ], + [ + 151, + 165 + ], + [ + 250, + 469 + ], + [ + 496, + 505 + ], + [ + 574, + 585 ] ], "right_ranges": [ [ - 16, - 22 + 40, + 46 ], [ - 26, - 39 + 96, + 110 + ], + [ + 151, + 165 + ], + [ + 250, + 469 + ], + [ + 496, + 505 + ], + [ + 574, + 585 ] ] }, { - "filename": "docs/source/en/api/pipelines/pix2pix_zero.md", + "filename": "scripts/convert_wan_to_diffusers.py", "left_ranges": [ [ - 20, - 26 + 1, + 27 ], [ - 29, - 39 + 116, + 308 ], [ - 78, - 99 + 560, + 590 ], [ - 117, - 141 + 601, + 612 ], [ - 157, - 172 + 620, + 643 ], [ - 178, - 184 + 1148, + 1311 ], [ - 191, - 201 + 1324, + 1330 ], [ - 212, - 229 + 1340, + 1346 ], [ - 235, - 242 + 1352, + 1359 ], [ - 260, - 288 + 1416, + 1449 ] ], "right_ranges": [ [ - 20, - 26 - ], - [ - 29, - 39 + 6, + 25 ], [ - 78, - 99 + 114, + 316 ], [ - 117, - 141 + 568, + 604 ], [ - 157, - 172 + 615, + 626 ], [ - 178, - 184 + 634, + 645 ], [ - 191, - 201 + 1168, + 1174 ], [ - 212, - 229 + 1184, + 1190 ], [ - 235, - 242 + 1196, + 1203 ], [ - 260, - 288 + 1260, + 1280 ] ] }, { - "filename": "docs/source/en/api/pipelines/pixart.md", + "filename": "src/diffusers/__init__.py", "left_ranges": [ [ - 24, - 42 - ] - ], - "right_ranges": [ - [ - 10, - 16 + 259, + 265 ], [ - 24, - 43 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/pndm.md", - "left_ranges": [ - [ - 12, - 18 + 621, + 627 ], [ - 32, - 35 - ] - ], - "right_ranges": [ - [ - 12, - 18 + 954, + 960 ], [ - 32, - 35 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/score_sde_ve.md", - "left_ranges": [ - [ - 32, - 35 + 1286, + 1292 ] ], "right_ranges": [ [ - 32, - 35 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/self_attention_guidance.md", - "left_ranges": [ - [ - 32, - 35 - ] - ], - "right_ranges": [ + 268, + 274 + ], [ - 32, - 35 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/semantic_stable_diffusion.md", - "left_ranges": [ + 637, + 643 + ], [ - 12, - 23 - ] - ], - "right_ranges": [ + 979, + 985 + ], [ - 12, - 23 + 1318, + 1324 ] ] }, { - "filename": "docs/source/en/api/pipelines/shap_e.md", + "filename": "src/diffusers/models/__init__.py", "left_ranges": [ [ - 9, - 15 + 101, + 107 ], [ - 34, - 37 + 201, + 207 ] ], "right_ranges": [ [ - 9, - 15 + 108, + 114 ], [ - 34, - 37 + 215, + 221 ] ] }, { - "filename": "docs/source/en/api/pipelines/spectrogram_diffusion.md", + "filename": "src/diffusers/models/transformers/__init__.py", "left_ranges": [ [ - 34, - 37 + 37, + 41 ] ], "right_ranges": [ [ - 34, - 37 + 42, + 46 ] ] }, { - "filename": "docs/source/en/api/pipelines/stable_diffusion/adapter.md", + "filename": "src/diffusers/models/transformers/transformer_wan_animate.py", "left_ranges": [ [ - 20, - 26 - ], - [ - 33, - 39 - ], - [ - 42, - 48 - ], - [ - 83, - 103 - ], - [ - 106, - 112 - ], - [ - 122, - 130 - ], - [ - 141, - 153 - ], - [ - 160, - 166 - ], - [ - 182, - 196 - ], - [ - 201, - 213 - ], - [ - 221, - 240 + 1, + 611 ] ], "right_ranges": [ [ - 20, - 26 - ], - [ - 33, - 39 - ], - [ - 42, - 48 - ], - [ - 83, - 103 - ], - [ - 106, - 112 - ], - [ - 122, - 130 - ], - [ - 141, - 153 - ], - [ - 160, - 166 - ], - [ - 182, - 196 - ], - [ - 201, - 213 - ], - [ - 221, - 240 + 1, + 1298 ] ] }, { - "filename": "docs/source/en/api/pipelines/stable_diffusion/depth2img.md", + "filename": "src/diffusers/pipelines/__init__.py", "left_ranges": [ [ - 12, - 22 + 381, + 393 ], [ - 37, - 40 + 794, + 806 ] ], "right_ranges": [ [ - 12, - 22 + 385, + 397 ], - [ - 37, - 40 + [ + 809, + 821 ] ] }, { - "filename": "docs/source/en/api/pipelines/stable_diffusion/inpaint.md", + "filename": "src/diffusers/pipelines/wan/__init__.py", "left_ranges": [ [ 23, 29 ], [ - 54, - 57 + 36, + 45 ] ], "right_ranges": [ @@ -11865,871 +9575,917 @@ 29 ], [ - 54, - 57 + 36, + 45 ] ] }, { - "filename": "docs/source/en/api/pipelines/stable_diffusion/latent_upscale.md", + "filename": "src/diffusers/pipelines/wan/pipeline_wan_animate.py", "left_ranges": [ [ - 16, - 22 - ], - [ - 35, - 38 + 1, + 1072 ] ], "right_ranges": [ [ - 16, - 22 - ], - [ - 35, - 38 + 1, + 1204 ] ] }, { - "filename": "docs/source/en/api/pipelines/stable_diffusion/overview.md", + "filename": "src/diffusers/utils/dummy_pt_objects.py", "left_ranges": [ [ - 34, - 40 - ], - [ - 165, - 168 + 1488, + 1508 ] ], "right_ranges": [ [ - 34, - 40 - ], - [ - 165, - 168 + 1623, + 1643 ] ] }, { - "filename": "docs/source/en/api/pipelines/stable_diffusion/stable_diffusion_2.md", + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", "left_ranges": [ [ - 14, - 25 - ], - [ - 35, - 41 - ], - [ - 55, - 75 - ], - [ - 79, - 92 - ], - [ - 96, - 124 + 3407, + 3427 ] ], "right_ranges": [ [ - 14, - 25 - ], - [ - 35, - 41 - ], - [ - 55, - 75 - ], - [ - 79, - 92 - ], + 3512, + 3532 + ] + ] + }, + { + "filename": "tests/pipelines/wan/test_wan_animate.py", + "left_ranges": [ [ - 96, - 114 - ], + 1, + 245 + ] + ], + "right_ranges": [ [ - 117, - 125 + 1, + 239 ] ] }, { - "filename": "docs/source/en/api/pipelines/stable_diffusion/stable_diffusion_xl.md", + "filename": "tests/quantization/gguf/test_gguf.py", "left_ranges": [ [ - 23, - 29 + 16, + 22 ], [ - 32, - 38 + 722, + 754 ] ], "right_ranges": [ [ - 23, - 29 + 16, + 22 ], [ - 32, - 38 + 722, + 754 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-5886-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5886.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 5886, + "canonical_pr_number": 5911, + "issue_numbers": [ + 5886 + ], + "pr_numbers": [ + 5910, + 5911 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5910, + "right_pr_number": 5911, + "code_similarity": 0.294, + "size_similarity": 0.411, + "file_overlap": 0.333, + "area_overlap": 0.26, + "patch_similarity": 0.134, + "shared_filenames": [ + "src/diffusers/loaders/unet.py", + "src/diffusers/models/embeddings.py" + ], + "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/stable_diffusion/text2img.md", + "filename": "src/diffusers/loaders/unet.py", "left_ranges": [ [ - 20, - 26 + 21, + 27 ], [ - 56, - 59 + 664, + 670 + ], + [ + 711, + 726 ] ], "right_ranges": [ [ - 20, - 26 + 21, + 27 ], [ - 56, - 59 + 675, + 683 + ], + [ + 747, + 778 ] ] }, { - "filename": "docs/source/en/api/pipelines/stable_diffusion/upscale.md", + "filename": "src/diffusers/models/embeddings.py", "left_ranges": [ [ - 16, - 22 - ], - [ - 34, - 37 + 458, + 482 ] ], "right_ranges": [ [ - 16, - 22 - ], - [ - 34, - 37 + 461, + 478 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-3880-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #3880.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 3880, + "canonical_pr_number": 4536, + "issue_numbers": [ + 3880 + ], + "pr_numbers": [ + 3921, + 4536 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 3921, + "right_pr_number": 4536, + "code_similarity": 0.061, + "size_similarity": 0.229, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.1, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-4899-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5800.", + "status": "closed", + "confidence": 0.6, + "canonical_issue_number": 5800, + "canonical_pr_number": 5938, + "issue_numbers": [ + 5800 + ], + "pr_numbers": [ + 4899, + 5938 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 4899, + "right_pr_number": 5938, + "code_similarity": 0.472, + "size_similarity": 0.741, + "file_overlap": 0.375, + "area_overlap": 0.354, + "patch_similarity": 0.586, + "shared_filenames": [ + "src/diffusers/models/controlnet.py", + "src/diffusers/models/embeddings.py", + "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py" + ], + "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/stable_unclip.md", + "filename": "src/diffusers/models/controlnet.py", "left_ranges": [ [ - 22, - 31 + 19, + 26 ], [ - 58, - 69 + 81, + 87 ], [ - 82, - 106 + 97, + 103 ], [ - 113, - 118 + 109, + 117 ], [ - 124, - 130 + 251, + 257 + ], + [ + 723, + 738 + ], + [ + 771, + 777 ] ], "right_ranges": [ [ - 22, - 31 - ], - [ - 58, - 69 + 20, + 27 ], [ - 88, - 106 + 82, + 88 ], [ - 113, - 118 + 108, + 116 ], [ - 124, - 129 + 253, + 259 ] ] }, { - "filename": "docs/source/en/api/pipelines/stochastic_karras_ve.md", + "filename": "src/diffusers/models/embeddings.py", "left_ranges": [ [ - 16, - 22 + 18, + 24 ], [ - 30, - 33 + 167, + 176 + ], + [ + 180, + 186 ] ], "right_ranges": [ [ - 16, - 22 + 18, + 24 ], [ - 30, - 33 + 201, + 210 + ], + [ + 214, + 220 ] ] }, { - "filename": "docs/source/en/api/pipelines/text_to_video.md", + "filename": "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", "left_ranges": [ [ - 12, - 18 - ], - [ - 26, - 38 + 377, + 397 ], [ - 88, - 94 + 429, + 450 ], [ - 118, - 126 + 458, + 483 ], [ - 153, - 159 + 623, + 631 ], [ - 167, - 178 + 635, + 647 ], [ - 184, - 187 + 652, + 672 ] ], "right_ranges": [ [ - 12, - 18 + 304, + 311 ], [ - 26, - 38 + 323, + 328 ], [ - 88, - 94 + 377, + 397 ], [ - 118, - 126 + 429, + 450 ], [ - 153, - 159 + 458, + 483 ], [ - 167, - 178 + 531, + 536 ], [ - 184, - 187 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/text_to_video_zero.md", - "left_ranges": [ + 623, + 631 + ], [ - 12, - 18 + 635, + 647 ], [ - 30, - 54 + 652, + 672 ], [ - 58, - 68 + 830, + 836 ], [ - 90, - 96 + 965, + 971 ], [ - 117, - 132 + 1178, + 1183 ], [ - 143, - 161 + 1262, + 1275 ], [ - 182, - 207 + 1308, + 1332 ], [ - 228, - 235 + 1346, + 1353 ], [ - 247, - 268 - ] - ], - "right_ranges": [ + 1431, + 1439 + ], [ - 12, - 18 + 1482, + 1492 ], [ - 30, - 44 + 1498, + 1505 ], [ - 57, - 73 + 1509, + 1516 ], [ - 115, - 121 + 1541, + 1549 ], [ - 145, - 156 + 1576, + 1585 + ], + [ + 1592, + 1602 ], [ - 188, - 199 + 1632, + 1649 ], [ - 242, - 257 + 1663, + 1668 + ], + [ + 1677, + 1690 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-12004-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 12006, + "issue_numbers": [], + "pr_numbers": [ + 12004, + 12006, + 12024 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 12004, + "right_pr_number": 12006, + "code_similarity": 0.151, + "size_similarity": 0.056, + "file_overlap": 0.222, + "area_overlap": 0.166, + "patch_similarity": 0.103, + "shared_filenames": [ + "src/diffusers/pipelines/skyreels_v2/pipeline_skyreels_v2_i2v.py", + "src/diffusers/pipelines/wan/pipeline_wan_i2v.py" + ], + "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/unclip.md", + "filename": "src/diffusers/pipelines/skyreels_v2/pipeline_skyreels_v2_i2v.py", "left_ranges": [ [ - 9, - 21 + 316, + 321 ] ], "right_ranges": [ [ - 9, - 21 + 370, + 375 ] ] }, { - "filename": "docs/source/en/api/pipelines/unidiffuser.md", + "filename": "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", "left_ranges": [ [ - 14, - 20 + 149, + 180 + ], + [ + 186, + 194 ], [ - 54, - 60 + 339, + 345 ], [ - 65, - 71 + 383, + 394 ], [ - 100, - 106 + 504, + 510 ], [ - 133, - 139 + 549, + 558 ], [ - 147, - 153 + 615, + 621 ], [ - 164, - 170 + 625, + 635 + ], + [ + 662, + 675 + ], + [ + 700, + 730 ], [ - 190, - 205 + 734, + 748 ] ], "right_ranges": [ [ - 14, - 20 + 175, + 181 ], [ - 54, - 60 + 189, + 198 ], [ - 65, - 71 + 420, + 431 ], [ - 100, - 106 + 458, + 470 ], [ - 133, - 139 + 674, + 680 ], [ - 147, - 153 + 694, + 701 ], [ - 164, - 169 + 708, + 717 ], [ - 190, - 205 + 738, + 754 + ], + [ + 792, + 800 ] ] - }, + } + ] + }, + { + "left_pr_number": 12004, + "right_pr_number": 12024, + "code_similarity": 0.387, + "size_similarity": 0.855, + "file_overlap": 0.455, + "area_overlap": 0.157, + "patch_similarity": 0.164, + "shared_filenames": [ + "src/diffusers/models/transformers/transformer_wan.py", + "src/diffusers/pipelines/wan/pipeline_wan.py", + "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", + "tests/pipelines/wan/test_wan.py", + "tests/pipelines/wan/test_wan_image_to_video.py" + ], + "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/value_guided_sampling.md", + "filename": "src/diffusers/models/transformers/transformer_wan.py", "left_ranges": [ [ - 22, - 38 - ] - ], - "right_ranges": [ + 170, + 180 + ], [ - 22, - 38 - ] - ] - }, - { - "filename": "docs/source/en/api/pipelines/versatile_diffusion.md", - "left_ranges": [ + 312, + 334 + ], [ - 12, - 22 + 486, + 507 + ], + [ + 517, + 530 ] ], "right_ranges": [ [ - 12, - 22 + 324, + 330 ] ] }, { - "filename": "docs/source/en/api/pipelines/wuerstchen.md", + "filename": "src/diffusers/pipelines/wan/pipeline_wan.py", "left_ranges": [ [ - 1, - 27 + 112, + 131 ], [ - 33, - 39 + 134, + 142 ], [ - 57, - 63 + 146, + 157 ], [ - 102, - 109 + 285, + 291 ], [ - 126, - 137 + 318, + 326 ], [ - 158, - 165 - ] - ], - "right_ranges": [ + 388, + 394 + ], [ - 1, - 27 + 427, + 436 ], [ - 33, - 39 + 485, + 491 ], [ - 57, - 63 + 495, + 505 ], [ - 102, - 109 + 549, + 592 ], [ - 126, - 132 + 595, + 609 + ] + ], + "right_ranges": [ + [ + 125, + 139 + ], + [ + 526, + 532 ], [ - 153, - 160 + 536, + 546 ] ] - } - ] - }, - { - "left_pr_number": 5744, - "right_pr_number": 5748, - "code_similarity": 0.238, - "size_similarity": 0.969, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.292, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-13462-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #13462.", - "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 13462, - "canonical_pr_number": 13471, - "issue_numbers": [ - 13462 - ], - "pr_numbers": [ - 13468, - 13471 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ - { - "left_pr_number": 13468, - "right_pr_number": 13471, - "code_similarity": 0.43, - "size_similarity": 0.344, - "file_overlap": 1.0, - "area_overlap": 0.132, - "patch_similarity": 0.102, - "shared_filenames": [ - "examples/profiling/profiling_utils.py" - ], - "shared_file_areas": [ + }, { - "filename": "examples/profiling/profiling_utils.py", + "filename": "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", "left_ranges": [ [ - 1, - 6 + 149, + 180 ], [ - 29, - 46 + 186, + 194 ], [ - 49, - 56 + 339, + 345 ], [ - 59, - 95 + 383, + 394 ], [ - 169, - 177 + 504, + 510 ], [ - 217, - 225 - ] - ], - "right_ranges": [ + 549, + 558 + ], [ - 45, - 60 - ] - ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-5285-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5285.", - "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 5285, - "canonical_pr_number": 5478, - "issue_numbers": [ - 5285 - ], - "pr_numbers": [ - 5286, - 5478 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ - { - "left_pr_number": 5286, - "right_pr_number": 5478, - "code_similarity": 0.144, - "size_similarity": 0.035, - "file_overlap": 0.143, - "area_overlap": 0.224, - "patch_similarity": 0.104, - "shared_filenames": [ - "docs/source/ja/index.md" - ], - "shared_file_areas": [ - { - "filename": "docs/source/ja/index.md", - "left_ranges": [ + 615, + 621 + ], + [ + 625, + 635 + ], [ - 1, - 22 + 662, + 675 + ], + [ + 700, + 730 + ], + [ + 734, + 748 ] ], "right_ranges": [ [ - 1, - 98 + 162, + 178 + ], + [ + 669, + 681 + ], + [ + 710, + 716 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-10817-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #10817.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 10817, - "canonical_pr_number": 10888, - "issue_numbers": [ - 10817 - ], - "pr_numbers": [ - 10830, - 10888 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 10830, - "right_pr_number": 10888, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, - "shared_filenames": [ - "src/diffusers/__init__.py", - "src/diffusers/pipelines/auto_pipeline.py", - "src/diffusers/utils/dummy_torch_and_transformers_objects.py" - ], - "shared_file_areas": [ + }, { - "filename": "src/diffusers/__init__.py", + "filename": "tests/pipelines/wan/test_wan.py", "left_ranges": [ [ - 865, - 871 + 85, + 113 ] ], "right_ranges": [ [ - 865, - 871 + 13, + 22 + ], + [ + 87, + 99 + ], + [ + 141, + 185 ] ] }, { - "filename": "src/diffusers/pipelines/auto_pipeline.py", + "filename": "tests/pipelines/wan/test_wan_image_to_video.py", "left_ranges": [ [ - 34, - 43 + 86, + 108 ], [ - 124, - 130 + 126, + 132 ], [ - 183, - 189 + 182, + 193 + ], + [ + 242, + 265 + ], + [ + 283, + 289 + ], + [ + 340, + 348 ] ], "right_ranges": [ [ - 34, - 43 + 12, + 21 ], [ - 124, - 130 + 27, + 33 ], [ - 183, - 189 - ] - ] - }, - { - "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", - "left_ranges": [ + 88, + 93 + ], [ - 1517, - 1537 - ] - ], - "right_ranges": [ + 111, + 117 + ], [ - 1517, - 1537 + 167, + 210 + ], + [ + 260, + 265 + ], + [ + 283, + 289 + ], + [ + 341, + 381 ] ] } ] - } - ] - }, - { - "cluster_id": "cluster-5504-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5504.", - "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 5504, - "canonical_pr_number": 5697, - "issue_numbers": [ - 5504 - ], - "pr_numbers": [ - 5506, - 5697 - ], - "evidence_types": [ - "closing_reference", - "soft_similarity" - ], - "pr_comparisons": [ + }, { - "left_pr_number": 5506, - "right_pr_number": 5697, - "code_similarity": 0.518, - "size_similarity": 0.4, - "file_overlap": 1.0, - "area_overlap": 0.073, - "patch_similarity": 0.75, + "left_pr_number": 12006, + "right_pr_number": 12024, + "code_similarity": 0.121, + "size_similarity": 0.065, + "file_overlap": 0.125, + "area_overlap": 0.161, + "patch_similarity": 0.093, "shared_filenames": [ - "src/diffusers/models/attention_processor.py" + "src/diffusers/pipelines/wan/pipeline_wan_i2v.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/models/attention_processor.py", + "filename": "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", "left_ranges": [ [ - 909, - 916 + 175, + 181 ], [ - 938, - 952 + 189, + 198 ], [ - 959, - 965 - ] - ], - "right_ranges": [ - [ - 879, - 887 + 420, + 431 ], [ - 894, - 910 + 458, + 470 ], [ - 918, - 924 + 674, + 680 ], [ - 949, - 957 + 694, + 701 ], [ - 964, - 970 + 708, + 717 ], [ - 973, - 980 + 738, + 754 ], [ - 991, - 997 - ], + 792, + 800 + ] + ], + "right_ranges": [ [ - 1183, - 1190 + 162, + 178 ], [ - 1215, - 1222 + 669, + 681 ], [ - 1236, - 1242 + 710, + 716 ] ] } @@ -12738,345 +10494,314 @@ ] }, { - "cluster_id": "cluster-3690-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #3693.", + "cluster_id": "cluster-7038-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #7038.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 7038, + "canonical_pr_number": 7550, + "issue_numbers": [ + 7038 + ], + "pr_numbers": [ + 7442, + 7550 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 7442, + "right_pr_number": 7550, + "code_similarity": 0.083, + "size_similarity": 0.271, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.192, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-13377-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13377.", "status": "merged", - "confidence": 0.6, - "canonical_issue_number": 3693, - "canonical_pr_number": 3695, + "confidence": 0.55, + "canonical_issue_number": 13377, + "canonical_pr_number": 13379, "issue_numbers": [ - 3693 + 13377 ], "pr_numbers": [ - 3690, - 3695 + 13379, + 13449 ], "evidence_types": [ - "closing_reference", - "soft_similarity" + "closing_reference" ], "pr_comparisons": [ { - "left_pr_number": 3690, - "right_pr_number": 3695, - "code_similarity": 0.747, - "size_similarity": 0.529, - "file_overlap": 0.5, - "area_overlap": 1.0, - "patch_similarity": 0.938, + "left_pr_number": 13379, + "right_pr_number": 13449, + "code_similarity": 0.212, + "size_similarity": 0.179, + "file_overlap": 0.562, + "area_overlap": 0.002, + "patch_similarity": 0.046, "shared_filenames": [ - "src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py" + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet_inpaint.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_inpaint.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_layered.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py", + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage.py", "left_ranges": [ [ - 41, - 61 + 311, + 332 + ], + [ + 600, + 606 ] ], "right_ranges": [ [ - 41, - 61 + 585, + 591 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-2067-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 2147, - "issue_numbers": [], - "pr_numbers": [ - 2067, - 2096, - 2147 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 2067, - "right_pr_number": 2096, - "code_similarity": 0.393, - "size_similarity": 0.3, - "file_overlap": 1.0, - "area_overlap": 0.0, - "patch_similarity": 0.221, - "shared_filenames": [ - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" - ], - "shared_file_areas": [ + }, { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet.py", "left_ranges": [ [ - 21, - 27 + 101, + 107 ], [ - 45, - 70 + 239, + 245 ], [ - 429, - 435 + 248, + 254 ], [ - 486, - 495 + 287, + 293 ], [ - 522, - 535 + 318, + 330 ], [ - 583, - 609 + 376, + 397 + ], + [ + 718, + 724 ] ], "right_ranges": [ [ - 509, - 519 - ], - [ - 567, - 575 + 701, + 707 ] ] - } - ] - }, - { - "left_pr_number": 2067, - "right_pr_number": 2147, - "code_similarity": 0.457, - "size_similarity": 0.811, - "file_overlap": 0.5, - "area_overlap": 0.296, - "patch_similarity": 0.273, - "shared_filenames": [ - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" - ], - "shared_file_areas": [ + }, { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet_inpaint.py", "left_ranges": [ [ - 21, - 27 + 74, + 80 ], [ - 45, - 70 + 221, + 227 ], [ - 429, - 435 + 230, + 236 ], [ - 486, - 495 + 247, + 253 ], [ - 522, - 535 + 269, + 275 ], [ - 583, - 609 + 280, + 286 + ], + [ + 300, + 317 + ], + [ + 359, + 377 + ], + [ + 751, + 757 ] ], "right_ranges": [ [ - 45, - 78 + 740, + 746 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit.py", + "left_ranges": [ + [ + 247, + 261 ], [ - 515, - 522 + 306, + 318 ], [ - 603, - 609 + 360, + 378 ], [ - 657, - 665 + 714, + 720 + ] + ], + "right_ranges": [ + [ + 706, + 712 ] ] - } - ] - }, - { - "left_pr_number": 2096, - "right_pr_number": 2147, - "code_similarity": 0.256, - "size_similarity": 0.243, - "file_overlap": 0.5, - "area_overlap": 0.068, - "patch_similarity": 0.226, - "shared_filenames": [ - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" - ], - "shared_file_areas": [ + }, { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_inpaint.py", "left_ranges": [ [ - 509, - 519 + 258, + 272 ], [ - 567, - 575 + 318, + 330 + ], + [ + 392, + 412 + ], + [ + 895, + 901 ] ], "right_ranges": [ [ - 45, - 78 + 879, + 885 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", + "left_ranges": [ + [ + 260, + 274 ], [ - 515, - 522 + 320, + 332 ], [ - 603, - 609 + 375, + 393 ], [ - 657, - 665 + 702, + 708 + ] + ], + "right_ranges": [ + [ + 694, + 700 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-5264-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5264.", - "status": "merged", - "confidence": 0.55, - "canonical_issue_number": 5264, - "canonical_pr_number": 5535, - "issue_numbers": [ - 5264 - ], - "pr_numbers": [ - 5280, - 5535 - ], - "evidence_types": [ - "closing_reference" - ], - "pr_comparisons": [ - { - "left_pr_number": 5280, - "right_pr_number": 5535, - "code_similarity": 0.553, - "size_similarity": 0.581, - "file_overlap": 1.0, - "area_overlap": 0.176, - "patch_similarity": 0.5, - "shared_filenames": [ - "tests/pipelines/controlnet/test_controlnet_img2img.py" - ], - "shared_file_areas": [ + }, { - "filename": "tests/pipelines/controlnet/test_controlnet_img2img.py", + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py", "left_ranges": [ [ - 72, - 79 + 358, + 379 ], [ - 83, - 90 + 693, + 699 ] ], "right_ranges": [ [ - 72, - 96 - ], - [ - 102, - 114 - ], + 678, + 684 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py", + "left_ranges": [ [ - 189, - 203 + 385, + 405 ], [ - 207, - 231 - ], + 837, + 843 + ] + ], + "right_ranges": [ [ - 239, - 251 + 823, + 829 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-8960-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #8960.", - "status": "merged", - "confidence": 0.75, - "canonical_issue_number": 8960, - "canonical_pr_number": 8966, - "issue_numbers": [ - 8960 - ], - "pr_numbers": [ - 8961, - 8966 - ], - "evidence_types": [ - "closing_reference", - "shared_issue_target" - ], - "pr_comparisons": [ - { - "left_pr_number": 8961, - "right_pr_number": 8966, - "code_similarity": 0.838, - "size_similarity": 0.786, - "file_overlap": 1.0, - "area_overlap": 0.875, - "patch_similarity": 0.5, - "shared_filenames": [ - "docs/source/en/tutorials/basic_training.md" - ], - "shared_file_areas": [ + }, { - "filename": "docs/source/en/tutorials/basic_training.md", + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_layered.py", "left_ranges": [ [ - 340, - 346 + 385, + 403 + ], + [ + 704, + 710 ] ], "right_ranges": [ [ - 340, - 347 + 698, + 704 ] ] } @@ -13085,89 +10810,87 @@ ] }, { - "cluster_id": "cluster-5062-3", - "summary": "Cluster of 1 issues and 2 PRs centered on issue #5062.", - "status": "merged", - "confidence": 0.8, - "canonical_issue_number": 5062, - "canonical_pr_number": 5066, + "cluster_id": "cluster-12458-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12458.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 12458, + "canonical_pr_number": 12698, "issue_numbers": [ - 5062 + 12458 ], "pr_numbers": [ - 5065, - 5066 + 12467, + 12698 ], "evidence_types": [ - "closing_reference", - "shared_issue_target", - "soft_similarity" + "closing_reference" ], "pr_comparisons": [ { - "left_pr_number": 5065, - "right_pr_number": 5066, - "code_similarity": 0.257, - "size_similarity": 0.089, - "file_overlap": 0.15, - "area_overlap": 0.543, - "patch_similarity": 0.028, + "left_pr_number": 12467, + "right_pr_number": 12698, + "code_similarity": 0.133, + "size_similarity": 0.125, + "file_overlap": 0.067, + "area_overlap": 0.208, + "patch_similarity": 0.105, "shared_filenames": [ - "examples/community/run_onnx_controlnet.py", - "examples/community/run_tensorrt_controlnet.py", - "src/diffusers/models/attention_processor.py" + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py" ], "shared_file_areas": [ { - "filename": "examples/community/run_onnx_controlnet.py", + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", "left_ranges": [ [ - 8, - 14 - ] - ], - "right_ranges": [ + 252, + 268 + ], [ - 8, - 13 + 635, + 646 ], [ - 21, - 27 - ] - ] - }, - { - "filename": "examples/community/run_tensorrt_controlnet.py", - "left_ranges": [ + 686, + 692 + ], [ - 11, - 17 + 695, + 704 + ], + [ + 736, + 760 ] ], "right_ranges": [ [ - 11, - 16 + 226, + 250 ], [ - 25, - 31 - ] - ] - }, - { - "filename": "src/diffusers/models/attention_processor.py", - "left_ranges": [ + 255, + 286 + ], + [ + 312, + 317 + ], + [ + 322, + 380 + ], + [ + 680, + 706 + ], [ - 382, - 388 - ] - ], - "right_ranges": [ + 739, + 810 + ], [ - 382, - 388 + 853, + 868 ] ] } @@ -13176,61 +10899,55 @@ ] }, { - "cluster_id": "cluster-4076-2", - "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "cluster_id": "cluster-5524-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5524.", "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 4079, - "issue_numbers": [], + "confidence": 0.55, + "canonical_issue_number": 5524, + "canonical_pr_number": 5704, + "issue_numbers": [ + 5524 + ], "pr_numbers": [ - 4076, - 4079 + 5696, + 5704 ], "evidence_types": [ - "soft_similarity" + "closing_reference" ], "pr_comparisons": [ { - "left_pr_number": 4076, - "right_pr_number": 4079, - "code_similarity": 0.657, - "size_similarity": 0.79, - "file_overlap": 0.2, - "area_overlap": 0.909, - "patch_similarity": 0.809, + "left_pr_number": 5696, + "right_pr_number": 5704, + "code_similarity": 0.183, + "size_similarity": 0.165, + "file_overlap": 0.25, + "area_overlap": 0.194, + "patch_similarity": 0.048, "shared_filenames": [ - "examples/instruct_pix2pix/train_instruct_pix2pix_xl.py", - "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py" + "src/diffusers/pipelines/wuerstchen/modeling_wuerstchen_prior.py" ], "shared_file_areas": [ { - "filename": "examples/instruct_pix2pix/train_instruct_pix2pix_xl.py", + "filename": "src/diffusers/pipelines/wuerstchen/modeling_wuerstchen_prior.py", "left_ranges": [ [ - 1, - 1051 + 59, + 64 ] ], "right_ranges": [ [ - 1, - 1205 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py", - "left_ranges": [ + 28, + 36 + ], [ - 1, - 963 - ] - ], - "right_ranges": [ + 41, + 55 + ], [ - 1, - 919 + 59, + 65 ] ] } @@ -13239,760 +10956,1199 @@ ] }, { - "cluster_id": "cluster-1028-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 1142, - "issue_numbers": [], + "cluster_id": "cluster-11285-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #11285.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 11285, + "canonical_pr_number": 12964, + "issue_numbers": [ + 11285 + ], "pr_numbers": [ - 1028, - 1071, - 1142 + 11306, + 12964 ], "evidence_types": [ - "soft_similarity" + "closing_reference" ], "pr_comparisons": [ { - "left_pr_number": 1028, - "right_pr_number": 1071, - "code_similarity": 0.317, - "size_similarity": 0.62, - "file_overlap": 0.2, - "area_overlap": 0.236, - "patch_similarity": 0.34, + "left_pr_number": 11306, + "right_pr_number": 12964, + "code_similarity": 0.057, + "size_similarity": 0.214, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.096, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-12754-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12754.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 12754, + "canonical_pr_number": 12762, + "issue_numbers": [ + 12754 + ], + "pr_numbers": [ + 12762, + 13103 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 12762, + "right_pr_number": 13103, + "code_similarity": 0.538, + "size_similarity": 0.923, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.357, "shared_filenames": [ - "README.md", - "docs/source/api/schedulers.mdx", - "docs/source/optimization/fp16.mdx", - "examples/community/README.md", - "examples/community/multilingual_stable_diffusion.py", - "setup.py", - "src/diffusers/dependency_versions_table.py", - "src/diffusers/pipeline_flax_utils.py", - "src/diffusers/pipeline_utils.py", - "src/diffusers/pipelines/README.md", - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", - "src/diffusers/utils/__init__.py", - "src/diffusers/utils/testing_utils.py", - "tests/models/test_models_unet_2d.py", - "tests/models/test_models_vae.py", - "tests/pipelines/stable_diffusion/test_stable_diffusion.py", - "tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py", - "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint.py", - "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint_legacy.py", - "tests/test_pipelines.py" + "src/diffusers/pipelines/pipeline_utils.py" ], "shared_file_areas": [ { - "filename": "README.md", - "left_ranges": [ - [ - 182, - 190 - ], - [ - 331, - 355 - ] - ], - "right_ranges": [ - [ - 142, - 148 - ] - ] - }, - { - "filename": "docs/source/api/schedulers.mdx", - "left_ranges": [ - [ - 89, - 95 - ], - [ - 99, - 107 - ] - ], - "right_ranges": [ - [ - 112, - 130 - ] - ] - }, - { - "filename": "docs/source/optimization/fp16.mdx", + "filename": "src/diffusers/pipelines/pipeline_utils.py", "left_ranges": [ [ - 18, - 26 + 2115, + 2121 ] ], "right_ranges": [ [ - 22, - 28 - ], - [ - 291, - 331 + 2082, + 2089 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-12905-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12905.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 12905, + "canonical_pr_number": 13183, + "issue_numbers": [ + 12905 + ], + "pr_numbers": [ + 13094, + 13183 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13094, + "right_pr_number": 13183, + "code_similarity": 0.597, + "size_similarity": 0.632, + "file_overlap": 1.0, + "area_overlap": 0.438, + "patch_similarity": 0.119, + "shared_filenames": [ + "src/diffusers/pipelines/z_image/pipeline_z_image.py" + ], + "shared_file_areas": [ { - "filename": "examples/community/README.md", + "filename": "src/diffusers/pipelines/z_image/pipeline_z_image.py", "left_ranges": [ [ - 16, - 24 + 276, + 282 ], [ - 374, - 514 + 335, + 343 ] ], "right_ranges": [ [ - 17, - 25 - ], - [ - 375, - 423 - ], - [ - 501, - 572 + 276, + 282 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-5712-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5712.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 5712, + "canonical_pr_number": 6438, + "issue_numbers": [ + 5712 + ], + "pr_numbers": [ + 6342, + 6438 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6342, + "right_pr_number": 6438, + "code_similarity": 0.54, + "size_similarity": 0.917, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.375, + "shared_filenames": [ + "docs/source/en/installation.md" + ], + "shared_file_areas": [ { - "filename": "examples/community/multilingual_stable_diffusion.py", + "filename": "docs/source/en/installation.md", "left_ranges": [ [ - 1, - 356 + 48, + 54 ] ], "right_ranges": [ [ - 1, - 436 + 37, + 46 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-9841-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #9841.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 9841, + "canonical_pr_number": 9879, + "issue_numbers": [ + 9841 + ], + "pr_numbers": [ + 9842, + 9879 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 9842, + "right_pr_number": 9879, + "code_similarity": 0.839, + "size_similarity": 0.891, + "file_overlap": 1.0, + "area_overlap": 0.767, + "patch_similarity": 0.614, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ { - "filename": "setup.py", + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", "left_ranges": [ [ - 94, - 100 - ], - [ - 182, - 188 + 1183, + 1218 ] ], "right_ranges": [ [ - 89, - 98 + 67, + 73 ], [ - 180, - 185 + 1184, + 1216 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-13292-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13292.", + "status": "open", + "confidence": 0.6, + "canonical_issue_number": 13292, + "canonical_pr_number": 13415, + "issue_numbers": [ + 13292 + ], + "pr_numbers": [ + 13307, + 13415 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13307, + "right_pr_number": 13415, + "code_similarity": 0.48, + "size_similarity": 0.978, + "file_overlap": 0.667, + "area_overlap": 0.138, + "patch_similarity": 0.238, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_flux2.py", + "examples/dreambooth/train_dreambooth_lora_flux2_klein.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/dependency_versions_table.py", + "filename": "examples/dreambooth/train_dreambooth_lora_flux2.py", "left_ranges": [ [ - 18, - 24 + 1734, + 1743 ] ], "right_ranges": [ [ - 13, - 22 + 1740, + 1751 + ], + [ + 1812, + 1822 ] ] }, { - "filename": "src/diffusers/pipeline_flax_utils.py", + "filename": "examples/dreambooth/train_dreambooth_lora_flux2_klein.py", "left_ranges": [ [ - 444, - 454 + 1674, + 1683 ] ], "right_ranges": [ [ - 29, - 35 - ], - [ - 272, - 278 - ], - [ - 302, - 314 + 1680, + 1691 ], [ - 319, - 325 + 1755, + 1765 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-12656-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12656.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 12656, + "canonical_pr_number": 13110, + "issue_numbers": [ + 12656 + ], + "pr_numbers": [ + 13110, + 13516 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13110, + "right_pr_number": 13516, + "code_similarity": 0.696, + "size_similarity": 0.696, + "file_overlap": 1.0, + "area_overlap": 0.632, + "patch_similarity": 0.237, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipeline_utils.py", + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", "left_ranges": [ [ - 625, - 635 + 76, + 87 ] ], - "right_ranges": [ - [ - 30, - 38 - ], - [ - 360, - 366 - ], - [ - 398, - 411 - ], + "right_ranges": [ [ - 602, - 608 + 76, + 94 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-12555-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13062.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 13062, + "canonical_pr_number": 12555, + "issue_numbers": [ + 13062 + ], + "pr_numbers": [ + 12555, + 13518 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 12555, + "right_pr_number": 13518, + "code_similarity": 0.599, + "size_similarity": 0.379, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.154, + "shared_filenames": [ + "src/diffusers/utils/import_utils.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/README.md", + "filename": "src/diffusers/utils/import_utils.py", "left_ranges": [ [ - 30, - 49 + 140, + 146 ] ], "right_ranges": [ [ - 16, - 22 + 140, + 146 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-7730-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #7730.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 7730, + "canonical_pr_number": 11156, + "issue_numbers": [ + 7730 + ], + "pr_numbers": [ + 7742, + 11156 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 7742, + "right_pr_number": 11156, + "code_similarity": 0.633, + "size_similarity": 0.833, + "file_overlap": 0.917, + "area_overlap": 0.188, + "patch_similarity": 0.837, + "shared_filenames": [ + "examples/instruct_pix2pix/train_instruct_pix2pix.py", + "examples/research_projects/instructpix2pix_lora/train_instruct_pix2pix_lora.py", + "examples/research_projects/promptdiffusion/convert_original_promptdiffusion_to_diffusers.py", + "scripts/convert_dance_diffusion_to_diffusers.py", + "scripts/convert_vae_pt_to_diffusers.py", + "src/diffusers/loaders/single_file_utils.py", + "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", + "src/diffusers/utils/loading_utils.py", + "src/diffusers/utils/testing_utils.py", + "utils/fetch_latest_release_branch.py", + "utils/notify_slack_about_release.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", + "filename": "examples/instruct_pix2pix/train_instruct_pix2pix.py", "left_ranges": [ [ - 208, - 213 + 418, + 424 ] ], "right_ranges": [ [ - 9, - 21 - ], - [ - 58, - 66 - ], - [ - 80, - 98 - ], - [ - 113, - 136 - ], - [ - 172, - 179 + 49, + 55 ], [ - 373, - 383 + 419, + 425 ] ] }, { - "filename": "src/diffusers/utils/__init__.py", + "filename": "examples/research_projects/instructpix2pix_lora/train_instruct_pix2pix_lora.py", "left_ranges": [ [ - 40, - 55 + 378, + 384 ] ], "right_ranges": [ [ - 42, - 48 + 54, + 60 + ], + [ + 476, + 482 ] ] }, { - "filename": "src/diffusers/utils/testing_utils.py", + "filename": "examples/research_projects/promptdiffusion/convert_original_promptdiffusion_to_diffusers.py", "left_ranges": [ [ - 4, - 17 - ], - [ - 37, - 50 - ], - [ - 168, - 186 + 1435, + 1441 ] ], "right_ranges": [ [ - 139, - 167 + 59, + 65 ], [ - 191, - 203 + 1436, + 1442 ] ] }, { - "filename": "tests/models/test_models_unet_2d.py", + "filename": "scripts/convert_dance_diffusion_to_diffusers.py", "left_ranges": [ [ - 21, - 33 - ], - [ - 68, - 73 - ], - [ - 150, - 163 - ], - [ - 208, - 214 - ], - [ - 288, - 294 - ], - [ - 340, - 348 - ], - [ - 356, - 362 - ], - [ - 378, - 384 - ], - [ - 403, - 596 + 74, + 80 ] ], "right_ranges": [ [ - 21, - 35 - ], - [ - 431, - 437 + 11, + 17 ], [ - 447, - 453 + 75, + 81 ] ] }, { - "filename": "tests/models/test_models_vae.py", + "filename": "scripts/convert_vae_pt_to_diffusers.py", "left_ranges": [ [ - 1, - 303 + 122, + 128 ] ], "right_ranges": [ [ - 20, - 26 + 13, + 19 ], [ - 147, - 153 + 123, + 130 ] ] }, { - "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", + "filename": "src/diffusers/loaders/single_file_utils.py", "left_ranges": [ [ - 528, - 534 - ], - [ - 548, - 554 - ], - [ - 576, - 582 - ], - [ - 595, - 604 - ], - [ - 634, - 643 - ], - [ - 672, - 678 - ], - [ - 714, - 720 + 381, + 387 ], [ - 740, - 746 + 402, + 408 ] ], "right_ranges": [ [ - 24, - 31 - ], - [ - 363, - 458 + 44, + 50 ], [ - 640, - 652 + 444, + 450 ] ] }, { - "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py", + "filename": "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", "left_ranges": [ [ - 488, - 494 - ], + 1324, + 1330 + ] + ], + "right_ranges": [ [ - 530, - 536 + 52, + 58 ], [ - 582, - 588 + 1325, + 1331 + ] + ] + }, + { + "filename": "src/diffusers/utils/loading_utils.py", + "left_ranges": [ + [ + 25, + 31 ] ], "right_ranges": [ [ - 523, - 530 + 7, + 13 + ], + [ + 30, + 36 ] ] }, { - "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint.py", + "filename": "src/diffusers/utils/testing_utils.py", "left_ranges": [ [ - 288, - 294 + 396, + 402 ], [ - 330, - 336 + 417, + 423 ], [ - 368, - 376 + 436, + 442 ] ], "right_ranges": [ [ - 28, - 34 + 26, + 32 ], [ - 278, - 287 + 595, + 601 ], [ - 306, - 312 + 616, + 622 ], [ - 317, - 326 - ], + 635, + 641 + ] + ] + }, + { + "filename": "utils/fetch_latest_release_branch.py", + "left_ranges": [ [ - 347, - 353 - ], + 27, + 33 + ] + ], + "right_ranges": [ [ - 358, - 370 + 13, + 24 ], [ - 385, - 388 + 30, + 40 ] ] }, { - "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint_legacy.py", + "filename": "utils/notify_slack_about_release.py", "left_ranges": [ [ - 368, - 374 - ], - [ - 414, - 420 + 26, + 32 ], [ - 471, - 477 + 38, + 44 ] ], "right_ranges": [ [ - 31, - 37 + 17, + 24 ], [ - 358, - 367 + 28, + 34 ], [ - 388, - 394 - ], + 40, + 46 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12540-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12540.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 12540, + "canonical_pr_number": 13109, + "issue_numbers": [ + 12540 + ], + "pr_numbers": [ + 13109, + 13517 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 13109, + "right_pr_number": 13517, + "code_similarity": 0.441, + "size_similarity": 0.667, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.052, + "shared_filenames": [ + "src/diffusers/models/transformers/sana_transformer.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/sana_transformer.py", + "left_ranges": [ [ - 400, - 412 - ], + 108, + 121 + ] + ], + "right_ranges": [ [ - 432, - 438 + 457, + 475 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-6545-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6545.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 6545, + "canonical_pr_number": 6569, + "issue_numbers": [ + 6545 + ], + "pr_numbers": [ + 6569, + 8283 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6569, + "right_pr_number": 8283, + "code_similarity": 0.258, + "size_similarity": 0.981, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.411, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-10076-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #10076.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 10076, + "canonical_pr_number": 10449, + "issue_numbers": [ + 10076 + ], + "pr_numbers": [ + 10449, + 13435 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 10449, + "right_pr_number": 13435, + "code_similarity": 0.247, + "size_similarity": 0.979, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.342, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-13411-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13411.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 13411, + "canonical_pr_number": 13466, + "issue_numbers": [ + 13411 + ], + "pr_numbers": [ + 13412, + 13466 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 13412, + "right_pr_number": 13466, + "code_similarity": 0.138, + "size_similarity": 0.05, + "file_overlap": 0.25, + "area_overlap": 0.12, + "patch_similarity": 0.075, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_ltx_euler_ancestral_rf.py" + ], + "shared_file_areas": [ { - "filename": "tests/test_pipelines.py", + "filename": "src/diffusers/schedulers/scheduling_ltx_euler_ancestral_rf.py", "left_ranges": [ [ - 77, - 83 + 1, + 4 ], [ - 86, - 92 + 65, + 73 ], [ - 98, - 104 + 83, + 97 ], [ - 111, - 117 + 237, + 259 ], [ - 121, - 127 + 271, + 278 ], [ - 316, - 324 - ], + 358, + 377 + ] + ], + "right_ranges": [ [ - 341, - 349 + 233, + 246 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-11208-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #11208.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 11208, + "canonical_pr_number": 11251, + "issue_numbers": [ + 11208 + ], + "pr_numbers": [ + 11251, + 12400 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 11251, + "right_pr_number": 12400, + "code_similarity": 0.336, + "size_similarity": 0.584, + "file_overlap": 0.5, + "area_overlap": 0.087, + "patch_similarity": 0.261, + "shared_filenames": [ + "src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py", + "left_ranges": [ + [ + 15, + 22 ], [ - 366, - 372 + 41, + 47 ], [ - 382, - 393 + 229, + 236 ], [ - 404, - 418 + 576, + 639 ], [ - 426, - 432 + 716, + 791 ], [ - 448, - 454 + 1157, + 1180 ], [ - 472, - 478 + 1242, + 1267 ] ], "right_ranges": [ [ - 42, - 48 - ], - [ - 446, - 454 - ], - [ - 462, - 485 + 1119, + 1171 ], [ - 494, - 510 + 1174, + 1183 ] ] } ] - }, + } + ] + }, + { + "cluster_id": "cluster-12913-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12913.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 12913, + "canonical_pr_number": 13513, + "issue_numbers": [ + 12913 + ], + "pr_numbers": [ + 13272, + 13513 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ { - "left_pr_number": 1028, - "right_pr_number": 1142, - "code_similarity": 0.304, - "size_similarity": 0.196, - "file_overlap": 0.043, - "area_overlap": 0.569, - "patch_similarity": 0.354, + "left_pr_number": 13272, + "right_pr_number": 13513, + "code_similarity": 0.45, + "size_similarity": 0.2, + "file_overlap": 0.5, + "area_overlap": 0.667, + "patch_similarity": 0.18, "shared_filenames": [ - "examples/community/README.md", - "examples/community/multilingual_stable_diffusion.py" + "src/diffusers/pipelines/lumina2/pipeline_lumina2.py" ], "shared_file_areas": [ { - "filename": "examples/community/README.md", + "filename": "src/diffusers/pipelines/lumina2/pipeline_lumina2.py", "left_ranges": [ [ - 16, - 24 - ], - [ - 374, - 514 + 696, + 703 ] ], "right_ranges": [ [ - 18, - 28 + 696, + 707 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-4704-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #4704.", + "status": "closed", + "confidence": 0.55, + "canonical_issue_number": 4704, + "canonical_pr_number": 4926, + "issue_numbers": [ + 4704 + ], + "pr_numbers": [ + 4707, + 4926 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 4707, + "right_pr_number": 4926, + "code_similarity": 0.197, + "size_similarity": 0.962, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.029, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-9610-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #9610.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 9610, + "canonical_pr_number": 9659, + "issue_numbers": [ + 9610 + ], + "pr_numbers": [ + 9659, + 9843 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 9659, + "right_pr_number": 9843, + "code_similarity": 0.292, + "size_similarity": 0.331, + "file_overlap": 0.5, + "area_overlap": 0.11, + "patch_similarity": 0.246, + "shared_filenames": [ + "tests/lora/test_lora_layers_sd3.py" + ], + "shared_file_areas": [ + { + "filename": "tests/lora/test_lora_layers_sd3.py", + "left_ranges": [ + [ + 12, + 40 ], [ - 504, - 577 + 45, + 54 ], [ - 604, - 607 - ] - ] - }, - { - "filename": "examples/community/multilingual_stable_diffusion.py", - "left_ranges": [ - [ - 1, - 356 + 128, + 215 ] ], "right_ranges": [ [ - 1, - 436 + 166, + 179 ] ] } ] - }, + } + ] + }, + { + "cluster_id": "cluster-13243-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13243.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 13243, + "canonical_pr_number": 13512, + "issue_numbers": [ + 13243 + ], + "pr_numbers": [ + 13246, + 13512 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ { - "left_pr_number": 1071, - "right_pr_number": 1142, - "code_similarity": 0.401, - "size_similarity": 0.122, - "file_overlap": 0.027, - "area_overlap": 0.888, - "patch_similarity": 0.386, + "left_pr_number": 13246, + "right_pr_number": 13512, + "code_similarity": 0.475, + "size_similarity": 0.26, + "file_overlap": 0.5, + "area_overlap": 0.714, + "patch_similarity": 0.152, "shared_filenames": [ - "examples/community/README.md", - "examples/community/multilingual_stable_diffusion.py" + "src/diffusers/schedulers/scheduling_flow_match_euler_discrete.py" ], "shared_file_areas": [ { - "filename": "examples/community/README.md", + "filename": "src/diffusers/schedulers/scheduling_flow_match_euler_discrete.py", "left_ranges": [ [ - 17, - 25 - ], - [ - 375, - 423 + 127, + 137 ], [ - 501, - 572 + 144, + 149 ] ], "right_ranges": [ [ - 18, - 28 - ], - [ - 504, - 577 + 127, + 139 ], [ - 604, - 607 + 146, + 151 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-13445-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13445.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 13445, + "canonical_pr_number": 13455, + "issue_numbers": [ + 13445 + ], + "pr_numbers": [ + 13455, + 13509 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13455, + "right_pr_number": 13509, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_flux2_klein.py" + ], + "shared_file_areas": [ { - "filename": "examples/community/multilingual_stable_diffusion.py", + "filename": "examples/dreambooth/train_dreambooth_lora_flux2_klein.py", "left_ranges": [ [ - 1, - 436 + 358, + 364 ] ], "right_ranges": [ [ - 1, - 436 + 358, + 364 ] ] } @@ -14001,176 +12157,262 @@ ] }, { - "cluster_id": "cluster-2812-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 2815, - "issue_numbers": [], + "cluster_id": "cluster-4553-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #4553.", + "status": "closed", + "confidence": 0.6, + "canonical_issue_number": 4553, + "canonical_pr_number": 4558, + "issue_numbers": [ + 4553 + ], "pr_numbers": [ - 2812, - 2813, - 2815 + 4558, + 4603 ], "evidence_types": [ + "closing_reference", "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 2812, - "right_pr_number": 2813, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, + "left_pr_number": 4558, + "right_pr_number": 4603, + "code_similarity": 0.423, + "size_similarity": 0.12, + "file_overlap": 0.125, "area_overlap": 1.0, - "patch_similarity": 1.0, + "patch_similarity": 0.078, "shared_filenames": [ - "docs/source/en/api/pipelines/stable_unclip.mdx", - "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py" + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py" ], "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", "left_ranges": [ [ - 16, - 25 - ], + 540, + 546 + ] + ], + "right_ranges": [ [ - 28, - 42 - ], + 540, + 546 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12496-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12499.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 12499, + "canonical_pr_number": 12523, + "issue_numbers": [ + 12499 + ], + "pr_numbers": [ + 12496, + 12523 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 12496, + "right_pr_number": 12523, + "code_similarity": 0.059, + "size_similarity": 0.281, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.02, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-5214-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5216.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 5216, + "canonical_pr_number": 5305, + "issue_numbers": [ + 5216 + ], + "pr_numbers": [ + 5214, + 5305 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5214, + "right_pr_number": 5305, + "code_similarity": 0.064, + "size_similarity": 0.022, + "file_overlap": 0.019, + "area_overlap": 0.144, + "patch_similarity": 0.026, + "shared_filenames": [ + "examples/custom_diffusion/train_custom_diffusion.py", + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen.py", + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen_text_image.py" + ], + "shared_file_areas": [ + { + "filename": "examples/custom_diffusion/train_custom_diffusion.py", + "left_ranges": [ [ - 50, - 57 + 1075, + 1104 ] ], "right_ranges": [ [ - 16, - 25 - ], - [ - 28, - 42 - ], - [ - 50, - 57 + 207, + 213 ] ] }, { - "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py", + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen.py", "left_ranges": [ [ - 537, - 563 + 26, + 32 ], [ - 568, - 580 + 274, + 283 + ], + [ + 402, + 411 + ], + [ + 803, + 811 + ], + [ + 840, + 847 ] ], "right_ranges": [ [ - 537, - 563 - ], - [ - 568, - 580 + 803, + 811 ] ] - } - ] - }, - { - "left_pr_number": 2812, - "right_pr_number": 2815, - "code_similarity": 0.601, - "size_similarity": 0.725, - "file_overlap": 0.5, - "area_overlap": 0.66, - "patch_similarity": 0.503, - "shared_filenames": [ - "docs/source/en/api/pipelines/stable_unclip.mdx" - ], - "shared_file_areas": [ + }, { - "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen_text_image.py", "left_ranges": [ [ - 16, - 25 + 32, + 38 ], [ - 28, - 42 + 306, + 315 ], [ - 50, - 57 - ] - ], - "right_ranges": [ - [ - 16, - 25 + 434, + 443 ], [ - 28, - 42 + 965, + 973 ], [ - 50, - 74 + 1014, + 1021 + ] + ], + "right_ranges": [ + [ + 965, + 973 ] ] } ] - }, + } + ] + }, + { + "cluster_id": "cluster-5302-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5302.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 5302, + "canonical_pr_number": 5304, + "issue_numbers": [ + 5302 + ], + "pr_numbers": [ + 5303, + 5304 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ { - "left_pr_number": 2813, - "right_pr_number": 2815, - "code_similarity": 0.601, - "size_similarity": 0.725, - "file_overlap": 0.5, - "area_overlap": 0.66, - "patch_similarity": 0.503, + "left_pr_number": 5303, + "right_pr_number": 5304, + "code_similarity": 0.093, + "size_similarity": 0.059, + "file_overlap": 0.167, + "area_overlap": 0.089, + "patch_similarity": 0.0, "shared_filenames": [ - "docs/source/en/api/pipelines/stable_unclip.mdx" + "src/diffusers/pipelines/deepfloyd_if/pipeline_if.py" ], "shared_file_areas": [ { - "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", + "filename": "src/diffusers/pipelines/deepfloyd_if/pipeline_if.py", "left_ranges": [ [ - 16, - 25 - ], - [ - 28, - 42 + 158, + 165 ], [ - 50, - 57 + 540, + 546 ] ], "right_ranges": [ [ - 16, - 25 + 161, + 171 ], [ - 28, - 42 + 174, + 187 ], [ - 50, - 74 + 193, + 200 + ], + [ + 570, + 582 ] ] } @@ -14179,1060 +12421,728 @@ ] }, { - "cluster_id": "cluster-643-3", + "cluster_id": "cluster-4629-3", "summary": "Cluster of 3 related pull requests linked by soft_similarity.", "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 645, + "canonical_pr_number": 4661, "issue_numbers": [], "pr_numbers": [ - 643, - 644, - 645 + 4629, + 4659, + 4661 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 643, - "right_pr_number": 644, - "code_similarity": 0.505, - "size_similarity": 0.216, + "left_pr_number": 4629, + "right_pr_number": 4659, + "code_similarity": 1.0, + "size_similarity": 1.0, "file_overlap": 1.0, - "area_overlap": 0.304, - "patch_similarity": 0.373, + "area_overlap": 1.0, + "patch_similarity": 1.0, "shared_filenames": [ + "docs/source/en/_toctree.yml", + "docs/source/en/api/models/overview.md", + "docs/source/en/using-diffusers/push_to_hub.md", + "examples/community/README.md", + "examples/community/lpw_stable_diffusion_xl.py", + "examples/controlnet/train_controlnet.py", + "examples/controlnet/train_controlnet_flax.py", + "examples/controlnet/train_controlnet_sdxl.py", + "examples/custom_diffusion/train_custom_diffusion.py", + "examples/dreambooth/train_dreambooth.py", + "examples/dreambooth/train_dreambooth_flax.py", + "examples/dreambooth/train_dreambooth_lora.py", + "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "examples/instruct_pix2pix/train_instruct_pix2pix.py", + "examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py", + "examples/research_projects/controlnet/train_controlnet_webdataset.py", + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_flax.py", + "examples/text_to_image/train_text_to_image_lora.py", + "examples/text_to_image/train_text_to_image_lora_sdxl.py", + "examples/text_to_image/train_text_to_image_sdxl.py", "examples/textual_inversion/textual_inversion.py", - "src/diffusers/dependency_versions_table.py", - "src/diffusers/models/resnet.py", - "src/diffusers/schedulers/scheduling_sde_ve.py", - "src/diffusers/schedulers/scheduling_sde_ve_flax.py" - ], - "shared_file_areas": [ - { - "filename": "examples/textual_inversion/textual_inversion.py", - "left_ranges": [ - [ - 29, - 49 - ], - [ - 552, - 559 - ], - [ - 579, - 585 - ] - ], - "right_ranges": [ - [ - 29, - 49 - ], - [ - 552, - 559 - ], - [ - 579, - 585 - ] - ] - }, + "examples/textual_inversion/textual_inversion_flax.py", + "examples/unconditional_image_generation/train_unconditional.py", + "setup.py", + "src/diffusers/__init__.py", + "src/diffusers/pipelines/controlnet/multicontrolnet.py", + "src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py", + "src/diffusers/utils/hub_utils.py", + "tests/models/test_lora_layers.py", + "tests/pipelines/controlnet/test_controlnet_sdxl.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/dependency_versions_table.py", + "filename": "docs/source/en/_toctree.yml", "left_ranges": [ [ - 17, - 22 + 32, + 37 ] ], "right_ranges": [ [ - 17, - 22 + 32, + 37 ] ] }, { - "filename": "src/diffusers/models/resnet.py", + "filename": "docs/source/en/api/models/overview.md", "left_ranges": [ [ - 1, - 482 + 11, + 16 ] ], "right_ranges": [ [ - 109, - 132 - ], - [ - 200, - 222 - ], - [ - 378, - 400 - ], - [ - 413, - 434 + 11, + 16 ] ] }, { - "filename": "src/diffusers/schedulers/scheduling_sde_ve.py", + "filename": "docs/source/en/using-diffusers/push_to_hub.md", "left_ranges": [ [ - 94, - 101 - ], - [ - 120, - 129 + 0, + 0 ] ], "right_ranges": [ [ - 94, - 101 - ], - [ - 120, - 129 + 0, + 0 ] ] }, { - "filename": "src/diffusers/schedulers/scheduling_sde_ve_flax.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 104, - 111 + 39, + 46 ], [ - 132, - 141 + 1530, + 1567 ] ], "right_ranges": [ [ - 104, - 111 + 39, + 46 ], [ - 132, - 141 + 1530, + 1567 ] ] - } - ] - }, - { - "left_pr_number": 643, - "right_pr_number": 645, - "code_similarity": 0.433, - "size_similarity": 0.03, - "file_overlap": 0.2, - "area_overlap": 1.0, - "patch_similarity": 0.115, - "shared_filenames": [ - "examples/textual_inversion/textual_inversion.py" - ], - "shared_file_areas": [ + }, { - "filename": "examples/textual_inversion/textual_inversion.py", + "filename": "examples/community/lpw_stable_diffusion_xl.py", "left_ranges": [ [ - 29, - 49 - ], - [ - 552, - 559 - ], - [ - 579, - 585 + 1, + 1297 ] ], "right_ranges": [ [ - 29, - 49 - ], - [ - 552, - 559 - ], - [ - 579, - 585 + 1, + 1297 ] ] - } - ] - }, - { - "left_pr_number": 644, - "right_pr_number": 645, - "code_similarity": 0.472, - "size_similarity": 0.138, - "file_overlap": 0.2, - "area_overlap": 1.0, - "patch_similarity": 0.228, - "shared_filenames": [ - "examples/textual_inversion/textual_inversion.py" - ], - "shared_file_areas": [ + }, { - "filename": "examples/textual_inversion/textual_inversion.py", + "filename": "examples/controlnet/train_controlnet.py", "left_ranges": [ [ - 29, - 49 - ], - [ - 552, - 559 - ], - [ - 579, - 585 + 56, + 62 ] ], "right_ranges": [ [ - 29, - 49 - ], - [ - 552, - 559 - ], - [ - 579, - 585 + 56, + 62 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-11445-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 11503, - "issue_numbers": [], - "pr_numbers": [ - 11445, - 11503, - 11507 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 11445, - "right_pr_number": 11503, - "code_similarity": 0.056, - "size_similarity": 0.236, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.056, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11445, - "right_pr_number": 11507, - "code_similarity": 0.111, - "size_similarity": 0.509, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.06, - "shared_filenames": [], - "shared_file_areas": [] - }, - { - "left_pr_number": 11503, - "right_pr_number": 11507, - "code_similarity": 0.03, - "size_similarity": 0.12, - "file_overlap": 0.0, - "area_overlap": 0.0, - "patch_similarity": 0.042, - "shared_filenames": [], - "shared_file_areas": [] - } - ] - }, - { - "cluster_id": "cluster-5626-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 5626, - "issue_numbers": [], - "pr_numbers": [ - 5626, - 7544, - 7868 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 5626, - "right_pr_number": 7544, - "code_similarity": 0.429, - "size_similarity": 0.407, - "file_overlap": 1.0, - "area_overlap": 0.0, - "patch_similarity": 0.317, - "shared_filenames": [ - "tests/pipelines/test_pipelines_common.py" - ], - "shared_file_areas": [ + }, { - "filename": "tests/pipelines/test_pipelines_common.py", + "filename": "examples/controlnet/train_controlnet_flax.py", "left_ranges": [ [ - 742, - 755 + 59, + 65 ] ], "right_ranges": [ [ - 1144, - 1167 - ], - [ - 1193, - 1215 - ] - ] - } - ] - }, - { - "left_pr_number": 5626, - "right_pr_number": 7868, - "code_similarity": 0.427, - "size_similarity": 0.583, - "file_overlap": 1.0, - "area_overlap": 0.0, - "patch_similarity": 0.069, - "shared_filenames": [ - "tests/pipelines/test_pipelines_common.py" - ], - "shared_file_areas": [ + 59, + 65 + ] + ] + }, { - "filename": "tests/pipelines/test_pipelines_common.py", + "filename": "examples/controlnet/train_controlnet_sdxl.py", "left_ranges": [ [ - 742, - 755 + 58, + 64 ] ], "right_ranges": [ [ - 1375, - 1381 - ], - [ - 1440, - 1446 + 58, + 64 ] ] - } - ] - }, - { - "left_pr_number": 7544, - "right_pr_number": 7868, - "code_similarity": 0.352, - "size_similarity": 0.237, - "file_overlap": 1.0, - "area_overlap": 0.0, - "patch_similarity": 0.029, - "shared_filenames": [ - "tests/pipelines/test_pipelines_common.py" - ], - "shared_file_areas": [ + }, { - "filename": "tests/pipelines/test_pipelines_common.py", + "filename": "examples/custom_diffusion/train_custom_diffusion.py", "left_ranges": [ [ - 1144, - 1167 - ], - [ - 1193, - 1215 + 58, + 64 ] ], "right_ranges": [ [ - 1375, - 1381 - ], - [ - 1440, - 1446 + 58, + 64 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-3168-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "closed", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 3168, - "issue_numbers": [], - "pr_numbers": [ - 3168, - 3676, - 3687 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 3168, - "right_pr_number": 3676, - "code_similarity": 0.9, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 0.333, - "shared_filenames": [ - "docs/source/en/index.mdx" - ], - "shared_file_areas": [ + }, { - "filename": "docs/source/en/index.mdx", + "filename": "examples/dreambooth/train_dreambooth.py", "left_ranges": [ [ - 16, - 22 + 60, + 66 ] ], "right_ranges": [ [ - 16, - 22 + 60, + 66 ] ] - } - ] - }, - { - "left_pr_number": 3168, - "right_pr_number": 3687, - "code_similarity": 0.9, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 0.333, - "shared_filenames": [ - "docs/source/en/index.mdx" - ], - "shared_file_areas": [ + }, { - "filename": "docs/source/en/index.mdx", + "filename": "examples/dreambooth/train_dreambooth_flax.py", "left_ranges": [ [ - 16, - 22 + 36, + 42 ] ], "right_ranges": [ [ - 16, - 22 + 36, + 42 ] ] - } - ] - }, - { - "left_pr_number": 3676, - "right_pr_number": 3687, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, - "shared_filenames": [ - "docs/source/en/index.mdx" - ], - "shared_file_areas": [ + }, { - "filename": "docs/source/en/index.mdx", + "filename": "examples/dreambooth/train_dreambooth_lora.py", "left_ranges": [ [ - 16, - 22 + 70, + 76 ] ], "right_ranges": [ [ - 16, - 22 + 70, + 76 ] ] - } - ] - } - ] - }, - { - "cluster_id": "cluster-9080-2", - "summary": "Cluster of 2 related pull requests linked by soft_similarity.", - "status": "merged", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 9082, - "issue_numbers": [], - "pr_numbers": [ - 9080, - 9082 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 9080, - "right_pr_number": 9082, - "code_similarity": 0.947, - "size_similarity": 0.98, - "file_overlap": 0.962, - "area_overlap": 0.936, - "patch_similarity": 0.9, - "shared_filenames": [ - "docs/source/en/api/loaders/single_file.md", - "docs/source/en/api/models/autoencoderkl_cogvideox.md", - "docs/source/en/api/models/cogvideox_transformer3d.md", - "docs/source/en/api/pipelines/cogvideox.md", - "scripts/convert_cogvideox_to_diffusers.py", - "src/diffusers/__init__.py", - "src/diffusers/models/__init__.py", - "src/diffusers/models/autoencoders/__init__.py", - "src/diffusers/models/autoencoders/autoencoder_kl_cogvideox.py", - "src/diffusers/models/downsampling.py", - "src/diffusers/models/embeddings.py", - "src/diffusers/models/normalization.py", - "src/diffusers/models/transformers/__init__.py", - "src/diffusers/models/transformers/cogvideox_transformer_3d.py", - "src/diffusers/models/upsampling.py", - "src/diffusers/pipelines/__init__.py", - "src/diffusers/pipelines/cogvideo/__init__.py", - "src/diffusers/pipelines/cogvideo/pipeline_cogvideox.py", - "src/diffusers/schedulers/__init__.py", - "src/diffusers/schedulers/scheduling_ddim_cogvideox.py", - "src/diffusers/schedulers/scheduling_dpm_cogvideox.py", - "src/diffusers/utils/dummy_pt_objects.py", - "src/diffusers/utils/dummy_torch_and_transformers_objects.py", - "tests/pipelines/cogvideox/__init__.py", - "tests/pipelines/cogvideox/test_cogvideox.py" - ], - "shared_file_areas": [ + }, { - "filename": "docs/source/en/api/loaders/single_file.md", + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", "left_ranges": [ [ - 22, - 28 - ], - [ - 50, - 56 + 58, + 64 ] ], "right_ranges": [ [ - 22, - 28 - ], - [ - 50, - 56 + 58, + 64 ] ] }, { - "filename": "docs/source/en/api/models/autoencoderkl_cogvideox.md", + "filename": "examples/instruct_pix2pix/train_instruct_pix2pix.py", "left_ranges": [ [ - 1, - 37 + 52, + 58 ] ], "right_ranges": [ [ - 1, - 37 + 52, + 58 ] ] }, { - "filename": "docs/source/en/api/models/cogvideox_transformer3d.md", + "filename": "examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py", "left_ranges": [ [ - 1, - 18 + 55, + 61 ] ], "right_ranges": [ [ - 1, - 30 + 55, + 61 ] ] }, { - "filename": "docs/source/en/api/pipelines/cogvideox.md", + "filename": "examples/research_projects/controlnet/train_controlnet_webdataset.py", "left_ranges": [ [ - 1, - 79 + 0, + 0 ] ], "right_ranges": [ [ - 1, - 91 + 0, + 0 ] ] }, { - "filename": "scripts/convert_cogvideox_to_diffusers.py", + "filename": "examples/text_to_image/train_text_to_image.py", "left_ranges": [ [ - 1, - 222 + 53, + 59 ] ], "right_ranges": [ [ - 1, - 222 + 53, + 59 ] ] }, { - "filename": "src/diffusers/__init__.py", + "filename": "examples/text_to_image/train_text_to_image_flax.py", "left_ranges": [ [ - 78, - 88 - ], - [ - 156, - 163 - ], - [ - 253, - 259 - ], - [ - 528, - 538 - ], - [ - 603, - 610 - ], - [ - 681, - 687 + 33, + 39 ] ], "right_ranges": [ [ - 79, - 89 - ], - [ - 157, - 164 - ], - [ - 252, - 258 - ], - [ - 540, - 550 - ], - [ - 615, - 622 - ], - [ - 691, - 697 + 33, + 39 ] ] }, { - "filename": "src/diffusers/models/__init__.py", - "left_ranges": [ - [ - 28, - 34 - ], - [ - 42, - 48 - ], - [ - 79, - 85 - ], + "filename": "examples/text_to_image/train_text_to_image_lora.py", + "left_ranges": [ [ - 95, - 101 + 48, + 54 ] ], "right_ranges": [ [ - 28, - 34 - ], - [ - 42, - 48 - ], - [ - 79, - 85 - ], - [ - 95, - 101 + 48, + 54 ] ] }, { - "filename": "src/diffusers/models/autoencoders/__init__.py", + "filename": "examples/text_to_image/train_text_to_image_lora_sdxl.py", "left_ranges": [ [ - 1, - 6 + 57, + 63 ] ], "right_ranges": [ [ - 1, - 6 + 57, + 63 ] ] }, { - "filename": "src/diffusers/models/autoencoders/autoencoder_kl_cogvideox.py", + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", "left_ranges": [ [ - 1, - 964 + 57, + 63 ] ], "right_ranges": [ [ - 1, - 1035 + 57, + 63 ] ] }, { - "filename": "src/diffusers/models/downsampling.py", + "filename": "examples/textual_inversion/textual_inversion.py", "left_ranges": [ [ - 285, - 358 + 79, + 85 ] ], "right_ranges": [ [ - 285, - 358 + 79, + 85 ] ] }, { - "filename": "src/diffusers/models/embeddings.py", + "filename": "examples/textual_inversion/textual_inversion_flax.py", "left_ranges": [ [ - 78, - 130 - ], - [ - 334, - 377 + 56, + 62 ] ], "right_ranges": [ [ - 78, - 130 - ], - [ - 334, - 379 + 56, + 62 ] ] }, { - "filename": "src/diffusers/models/normalization.py", + "filename": "examples/unconditional_image_generation/train_unconditional.py", "left_ranges": [ [ - 37, - 82 - ], - [ - 349, - 378 + 30, + 36 ] ], "right_ranges": [ [ - 34, - 86 - ], - [ - 355, - 384 + 30, + 36 ] ] }, { - "filename": "src/diffusers/models/transformers/__init__.py", + "filename": "setup.py", "left_ranges": [ [ - 3, - 9 + 233, + 239 ] ], "right_ranges": [ [ - 3, - 9 + 233, + 239 ] ] }, { - "filename": "src/diffusers/models/transformers/cogvideox_transformer_3d.py", + "filename": "src/diffusers/__init__.py", "left_ranges": [ [ 1, - 352 + 4 ] ], "right_ranges": [ [ 1, - 345 + 4 ] ] }, { - "filename": "src/diffusers/models/upsampling.py", + "filename": "src/diffusers/pipelines/controlnet/multicontrolnet.py", "left_ranges": [ [ - 348, - 418 + 39, + 44 + ], + [ + 53, + 58 ] ], "right_ranges": [ [ - 348, - 417 + 39, + 44 + ], + [ + 53, + 58 ] ] }, { - "filename": "src/diffusers/pipelines/__init__.py", + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py", "left_ranges": [ [ - 131, - 137 + 149, + 163 + ], + [ + 530, + 535 + ], + [ + 540, + 545 + ], + [ + 551, + 564 + ], + [ + 569, + 574 + ], + [ + 606, + 611 ], [ - 438, - 444 + 888, + 893 + ], + [ + 933, + 938 + ], + [ + 963, + 974 ] ], "right_ranges": [ [ - 132, - 138 + 149, + 163 ], [ - 452, - 458 + 530, + 535 + ], + [ + 540, + 545 + ], + [ + 551, + 564 + ], + [ + 569, + 574 + ], + [ + 606, + 611 + ], + [ + 888, + 893 + ], + [ + 933, + 938 + ], + [ + 963, + 974 ] ] }, { - "filename": "src/diffusers/pipelines/cogvideo/__init__.py", + "filename": "src/diffusers/utils/hub_utils.py", "left_ranges": [ [ - 1, - 48 + 410, + 432 ] ], "right_ranges": [ [ - 1, - 48 + 410, + 432 ] ] }, { - "filename": "src/diffusers/pipelines/cogvideo/pipeline_cogvideox.py", + "filename": "tests/models/test_lora_layers.py", "left_ranges": [ [ - 1, - 681 + 664, + 669 ] ], "right_ranges": [ [ - 1, - 687 + 664, + 669 ] ] }, { - "filename": "src/diffusers/schedulers/__init__.py", + "filename": "tests/pipelines/controlnet/test_controlnet_sdxl.py", "left_ranges": [ [ - 43, - 56 + 26, + 31 + ], + [ + 46, + 52 ], [ - 143, - 156 + 297, + 299 ] ], "right_ranges": [ [ - 43, - 56 + 26, + 31 + ], + [ + 46, + 52 ], [ - 143, - 156 + 297, + 299 ] ] - }, + } + ] + }, + { + "left_pr_number": 4629, + "right_pr_number": 4661, + "code_similarity": 0.514, + "size_similarity": 0.357, + "file_overlap": 0.065, + "area_overlap": 0.996, + "patch_similarity": 0.5, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/lpw_stable_diffusion_xl.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/schedulers/scheduling_ddim_cogvideox.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 1, - 479 + 39, + 46 + ], + [ + 1530, + 1567 ] ], "right_ranges": [ [ - 1, - 449 + 39, + 46 + ], + [ + 1530, + 1573 ] ] }, { - "filename": "src/diffusers/schedulers/scheduling_dpm_cogvideox.py", + "filename": "examples/community/lpw_stable_diffusion_xl.py", "left_ranges": [ [ 1, - 486 + 1297 ] ], "right_ranges": [ [ 1, - 489 + 1297 ] ] - }, + } + ] + }, + { + "left_pr_number": 4659, + "right_pr_number": 4661, + "code_similarity": 0.514, + "size_similarity": 0.357, + "file_overlap": 0.065, + "area_overlap": 0.996, + "patch_similarity": 0.5, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/lpw_stable_diffusion_xl.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/utils/dummy_pt_objects.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 47, - 67 - ], - [ - 107, - 127 + 39, + 46 ], [ - 1005, - 1040 + 1530, + 1567 ] ], "right_ranges": [ [ - 47, - 67 - ], - [ - 107, - 127 + 39, + 46 ], [ - 1005, - 1040 - ] - ] - }, - { - "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", - "left_ranges": [ - [ - 287, - 307 - ] - ], - "right_ranges": [ - [ - 257, - 277 + 1530, + 1573 ] ] }, { - "filename": "tests/pipelines/cogvideox/test_cogvideox.py", + "filename": "examples/community/lpw_stable_diffusion_xl.py", "left_ranges": [ [ 1, - 289 + 1297 ] ], "right_ranges": [ [ 1, - 290 + 1297 ] ] } @@ -15241,416 +13151,409 @@ ] }, { - "cluster_id": "cluster-6299-3", - "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "closed", - "confidence": 0.5, - "canonical_issue_number": null, - "canonical_pr_number": 6299, - "issue_numbers": [], + "cluster_id": "cluster-13462-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13462.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 13462, + "canonical_pr_number": 13471, + "issue_numbers": [ + 13462 + ], "pr_numbers": [ - 6299, - 6300, - 6301 + 13468, + 13471 ], "evidence_types": [ - "soft_similarity" + "closing_reference" ], "pr_comparisons": [ { - "left_pr_number": 6299, - "right_pr_number": 6300, - "code_similarity": 1.0, - "size_similarity": 1.0, + "left_pr_number": 13468, + "right_pr_number": 13471, + "code_similarity": 0.43, + "size_similarity": 0.344, "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, + "area_overlap": 0.132, + "patch_similarity": 0.102, "shared_filenames": [ - "examples/text_to_image/train_text_to_image.py", - "examples/text_to_image/train_text_to_image_sdxl.py" + "examples/profiling/profiling_utils.py" ], "shared_file_areas": [ { - "filename": "examples/text_to_image/train_text_to_image.py", + "filename": "examples/profiling/profiling_utils.py", "left_ranges": [ [ - 41, - 47 - ], - [ - 152, - 158 - ], - [ - 166, - 179 - ], - [ - 485, - 515 + 1, + 6 ], [ - 575, - 586 + 29, + 46 ], [ - 592, - 608 + 49, + 56 ], [ - 1014, - 1021 + 59, + 95 ], [ - 1036, - 1071 + 169, + 177 ], [ - 1125, - 1136 + 217, + 225 ] ], "right_ranges": [ [ - 41, - 47 - ], - [ - 152, - 158 - ], - [ - 166, - 179 - ], - [ - 485, - 515 - ], - [ - 575, - 586 - ], - [ - 592, - 608 - ], - [ - 1014, - 1021 - ], - [ - 1036, - 1071 - ], - [ - 1125, - 1136 + 45, + 60 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-5285-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5285.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 5285, + "canonical_pr_number": 5478, + "issue_numbers": [ + 5285 + ], + "pr_numbers": [ + 5286, + 5478 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 5286, + "right_pr_number": 5478, + "code_similarity": 0.144, + "size_similarity": 0.035, + "file_overlap": 0.143, + "area_overlap": 0.224, + "patch_similarity": 0.104, + "shared_filenames": [ + "docs/source/ja/index.md" + ], + "shared_file_areas": [ { - "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "filename": "docs/source/ja/index.md", "left_ranges": [ [ - 34, - 40 - ], - [ - 54, - 60 - ], - [ - 580, - 593 - ], - [ - 874, - 920 + 1, + 22 ] ], "right_ranges": [ [ - 34, - 40 - ], - [ - 54, - 60 - ], - [ - 580, - 593 - ], - [ - 874, - 920 + 1, + 98 ] ] } ] - }, + } + ] + }, + { + "cluster_id": "cluster-10817-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #10817.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 10817, + "canonical_pr_number": 10888, + "issue_numbers": [ + 10817 + ], + "pr_numbers": [ + 10830, + 10888 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ { - "left_pr_number": 6299, - "right_pr_number": 6301, + "left_pr_number": 10830, + "right_pr_number": 10888, "code_similarity": 1.0, "size_similarity": 1.0, "file_overlap": 1.0, "area_overlap": 1.0, "patch_similarity": 1.0, "shared_filenames": [ - "examples/text_to_image/train_text_to_image.py", - "examples/text_to_image/train_text_to_image_sdxl.py" + "src/diffusers/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" ], "shared_file_areas": [ { - "filename": "examples/text_to_image/train_text_to_image.py", + "filename": "src/diffusers/__init__.py", "left_ranges": [ [ - 41, - 47 - ], - [ - 152, - 158 - ], - [ - 166, - 179 - ], - [ - 485, - 515 - ], - [ - 575, - 586 - ], - [ - 592, - 608 - ], - [ - 1014, - 1021 - ], - [ - 1036, - 1071 - ], - [ - 1125, - 1136 + 865, + 871 ] ], "right_ranges": [ [ - 41, - 47 - ], - [ - 152, - 158 - ], - [ - 166, - 179 - ], - [ - 485, - 515 - ], - [ - 575, - 586 - ], - [ - 592, - 608 - ], - [ - 1014, - 1021 - ], - [ - 1036, - 1071 - ], - [ - 1125, - 1136 + 865, + 871 ] ] }, { - "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "filename": "src/diffusers/pipelines/auto_pipeline.py", "left_ranges": [ [ 34, - 40 - ], - [ - 54, - 60 + 43 ], [ - 580, - 593 + 124, + 130 ], [ - 874, - 920 + 183, + 189 ] ], "right_ranges": [ [ 34, - 40 + 43 ], [ - 54, - 60 + 124, + 130 ], [ - 580, - 593 - ], + 183, + 189 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "left_ranges": [ [ - 874, - 920 + 1517, + 1537 + ] + ], + "right_ranges": [ + [ + 1517, + 1537 ] ] } ] - }, + } + ] + }, + { + "cluster_id": "cluster-5504-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5504.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 5504, + "canonical_pr_number": 5697, + "issue_numbers": [ + 5504 + ], + "pr_numbers": [ + 5506, + 5697 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ { - "left_pr_number": 6300, - "right_pr_number": 6301, - "code_similarity": 1.0, - "size_similarity": 1.0, + "left_pr_number": 5506, + "right_pr_number": 5697, + "code_similarity": 0.518, + "size_similarity": 0.4, "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, + "area_overlap": 0.073, + "patch_similarity": 0.75, "shared_filenames": [ - "examples/text_to_image/train_text_to_image.py", - "examples/text_to_image/train_text_to_image_sdxl.py" + "src/diffusers/models/attention_processor.py" ], "shared_file_areas": [ { - "filename": "examples/text_to_image/train_text_to_image.py", + "filename": "src/diffusers/models/attention_processor.py", "left_ranges": [ [ - 41, - 47 - ], - [ - 152, - 158 - ], - [ - 166, - 179 + 909, + 916 ], [ - 485, - 515 + 938, + 952 ], [ - 575, - 586 - ], + 959, + 965 + ] + ], + "right_ranges": [ [ - 592, - 608 + 879, + 887 ], [ - 1014, - 1021 + 894, + 910 ], [ - 1036, - 1071 + 918, + 924 ], [ - 1125, - 1136 - ] - ], - "right_ranges": [ - [ - 41, - 47 + 949, + 957 ], [ - 152, - 158 + 964, + 970 ], [ - 166, - 179 + 973, + 980 ], [ - 485, - 515 + 991, + 997 ], [ - 575, - 586 + 1183, + 1190 ], [ - 592, - 608 + 1215, + 1222 ], [ - 1014, - 1021 - ], + 1236, + 1242 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-3690-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #3693.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 3693, + "canonical_pr_number": 3695, + "issue_numbers": [ + 3693 + ], + "pr_numbers": [ + 3690, + 3695 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 3690, + "right_pr_number": 3695, + "code_similarity": 0.747, + "size_similarity": 0.529, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.938, + "shared_filenames": [ + "src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py", + "left_ranges": [ [ - 1036, - 1071 - ], + 41, + 61 + ] + ], + "right_ranges": [ [ - 1125, - 1136 + 41, + 61 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-10453-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #10453.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 10453, + "canonical_pr_number": 10454, + "issue_numbers": [ + 10453 + ], + "pr_numbers": [ + 10454, + 10482 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10454, + "right_pr_number": 10482, + "code_similarity": 0.657, + "size_similarity": 0.611, + "file_overlap": 1.0, + "area_overlap": 0.467, + "patch_similarity": 0.474, + "shared_filenames": [ + "src/diffusers/models/transformers/transformer_hunyuan_video.py" + ], + "shared_file_areas": [ { - "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "filename": "src/diffusers/models/transformers/transformer_hunyuan_video.py", "left_ranges": [ [ - 34, - 40 - ], - [ - 54, - 60 - ], - [ - 580, - 593 - ], - [ - 874, - 920 + 721, + 727 ] ], "right_ranges": [ [ - 34, - 40 - ], - [ - 54, - 60 - ], - [ - 580, - 593 - ], - [ - 874, - 920 + 713, + 727 ] ] } @@ -15659,704 +13562,601 @@ ] }, { - "cluster_id": "cluster-9564-3", + "cluster_id": "cluster-2067-3", "summary": "Cluster of 3 related pull requests linked by soft_similarity.", - "status": "closed", + "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 9564, + "canonical_pr_number": 2147, "issue_numbers": [], "pr_numbers": [ - 9564, - 9854, - 9855 + 2067, + 2096, + 2147 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 9564, - "right_pr_number": 9854, - "code_similarity": 1.0, - "size_similarity": 1.0, + "left_pr_number": 2067, + "right_pr_number": 2096, + "code_similarity": 0.393, + "size_similarity": 0.3, "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.221, "shared_filenames": [ - "src/diffusers/loaders/lora_pipeline.py", - "src/diffusers/loaders/unet.py" + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/loaders/lora_pipeline.py", + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", "left_ranges": [ [ - 88, - 104 - ], - [ - 116, - 122 - ], - [ - 127, - 133 - ], - [ - 241, - 249 - ], - [ - 261, - 276 - ], - [ - 280, - 290 - ], - [ - 297, - 303 - ], - [ - 320, - 335 - ], - [ - 572, - 589 - ], - [ - 603, - 609 - ], - [ - 742, - 748 - ], - [ - 760, - 775 - ], - [ - 779, - 786 - ], - [ - 794, - 800 - ], - [ - 817, - 831 - ], - [ - 1695, - 1711 - ], - [ - 1725, - 1731 - ], - [ - 1738, - 1750 - ], - [ - 1762, - 1774 - ], - [ - 1817, - 1828 - ] - ], - "right_ranges": [ - [ - 88, - 104 - ], - [ - 116, - 122 - ], - [ - 127, - 133 - ], - [ - 241, - 249 - ], - [ - 261, - 276 + 21, + 27 ], [ - 280, - 290 + 45, + 70 ], [ - 297, - 303 + 429, + 435 ], [ - 320, - 335 + 486, + 495 ], [ - 572, - 589 + 522, + 535 ], [ - 603, + 583, 609 - ], - [ - 742, - 748 - ], - [ - 760, - 775 - ], - [ - 779, - 786 - ], - [ - 794, - 800 - ], - [ - 817, - 831 - ], - [ - 1695, - 1711 - ], - [ - 1725, - 1731 - ], - [ - 1738, - 1750 - ], - [ - 1762, - 1774 - ], - [ - 1817, - 1828 ] - ] - }, - { - "filename": "src/diffusers/loaders/unet.py", - "left_ranges": [ - [ - 115, - 121 - ], - [ - 143, - 156 - ], - [ - 216, - 222 - ], - [ - 276, - 284 - ], - [ - 345, - 356 - ] - ], - "right_ranges": [ - [ - 115, - 121 - ], - [ - 143, - 156 - ], - [ - 216, - 222 - ], + ], + "right_ranges": [ [ - 276, - 284 + 509, + 519 ], [ - 345, - 356 + 567, + 575 ] ] } ] }, { - "left_pr_number": 9564, - "right_pr_number": 9855, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, + "left_pr_number": 2067, + "right_pr_number": 2147, + "code_similarity": 0.457, + "size_similarity": 0.811, + "file_overlap": 0.5, + "area_overlap": 0.296, + "patch_similarity": 0.273, "shared_filenames": [ - "src/diffusers/loaders/lora_pipeline.py", - "src/diffusers/loaders/unet.py" + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/loaders/lora_pipeline.py", + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", "left_ranges": [ [ - 88, - 104 - ], - [ - 116, - 122 + 21, + 27 ], [ - 127, - 133 + 45, + 70 ], [ - 241, - 249 + 429, + 435 ], [ - 261, - 276 + 486, + 495 ], [ - 280, - 290 + 522, + 535 ], [ - 297, - 303 - ], + 583, + 609 + ] + ], + "right_ranges": [ [ - 320, - 335 + 45, + 78 ], [ - 572, - 589 + 515, + 522 ], [ 603, 609 ], [ - 742, - 748 - ], - [ - 760, - 775 - ], - [ - 779, - 786 - ], - [ - 794, - 800 - ], - [ - 817, - 831 - ], - [ - 1695, - 1711 - ], - [ - 1725, - 1731 - ], - [ - 1738, - 1750 - ], + 657, + 665 + ] + ] + } + ] + }, + { + "left_pr_number": 2096, + "right_pr_number": 2147, + "code_similarity": 0.256, + "size_similarity": 0.243, + "file_overlap": 0.5, + "area_overlap": 0.068, + "patch_similarity": 0.226, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", + "left_ranges": [ [ - 1762, - 1774 + 509, + 519 ], [ - 1817, - 1828 + 567, + 575 ] ], "right_ranges": [ [ - 88, - 104 - ], - [ - 116, - 122 - ], - [ - 127, - 133 - ], - [ - 241, - 249 + 45, + 78 ], [ - 261, - 276 + 515, + 522 ], [ - 280, - 290 + 603, + 609 ], [ - 297, - 303 - ], + 657, + 665 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5264-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5264.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 5264, + "canonical_pr_number": 5535, + "issue_numbers": [ + 5264 + ], + "pr_numbers": [ + 5280, + 5535 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 5280, + "right_pr_number": 5535, + "code_similarity": 0.553, + "size_similarity": 0.581, + "file_overlap": 1.0, + "area_overlap": 0.176, + "patch_similarity": 0.5, + "shared_filenames": [ + "tests/pipelines/controlnet/test_controlnet_img2img.py" + ], + "shared_file_areas": [ + { + "filename": "tests/pipelines/controlnet/test_controlnet_img2img.py", + "left_ranges": [ [ - 320, - 335 + 72, + 79 ], [ - 572, - 589 - ], + 83, + 90 + ] + ], + "right_ranges": [ [ - 603, - 609 + 72, + 96 ], [ - 742, - 748 + 102, + 114 ], [ - 760, - 775 + 189, + 203 ], [ - 779, - 786 + 207, + 231 ], [ - 794, - 800 - ], + 239, + 251 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-709-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #709.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 709, + "canonical_pr_number": 710, + "issue_numbers": [ + 709 + ], + "pr_numbers": [ + 710, + 798 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 710, + "right_pr_number": 798, + "code_similarity": 0.25, + "size_similarity": 0.5, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-8960-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #8960.", + "status": "merged", + "confidence": 0.75, + "canonical_issue_number": 8960, + "canonical_pr_number": 8966, + "issue_numbers": [ + 8960 + ], + "pr_numbers": [ + 8961, + 8966 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 8961, + "right_pr_number": 8966, + "code_similarity": 0.838, + "size_similarity": 0.786, + "file_overlap": 1.0, + "area_overlap": 0.875, + "patch_similarity": 0.5, + "shared_filenames": [ + "docs/source/en/tutorials/basic_training.md" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/tutorials/basic_training.md", + "left_ranges": [ [ - 817, - 831 - ], + 340, + 346 + ] + ], + "right_ranges": [ [ - 1695, - 1711 - ], + 340, + 347 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9083-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 9083, + "issue_numbers": [], + "pr_numbers": [ + 9083, + 10183, + 10991 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9083, + "right_pr_number": 10183, + "code_similarity": 0.316, + "size_similarity": 0.937, + "file_overlap": 0.333, + "area_overlap": 0.022, + "patch_similarity": 0.14, + "shared_filenames": [ + "src/diffusers/loaders/single_file_model.py", + "src/diffusers/loaders/single_file_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders/single_file_model.py", + "left_ranges": [ [ - 1725, - 1731 + 24, + 30 ], [ - 1738, - 1750 - ], + 75, + 84 + ] + ], + "right_ranges": [ [ - 1762, - 1774 + 23, + 29 ], [ - 1817, - 1828 + 83, + 89 ] ] }, { - "filename": "src/diffusers/loaders/unet.py", + "filename": "src/diffusers/loaders/single_file_utils.py", "left_ranges": [ [ - 115, - 121 - ], - [ - 143, - 156 + 77, + 83 ], [ - 216, - 222 + 111, + 118 ], [ - 276, - 284 + 506, + 516 ], [ - 345, - 356 + 1867, + 2061 ] ], "right_ranges": [ [ - 115, - 121 - ], - [ - 143, - 156 + 92, + 99 ], [ - 216, - 222 + 140, + 149 ], [ - 276, - 284 + 570, + 592 ], [ - 345, - 356 + 2221, + 2295 ] ] } ] }, { - "left_pr_number": 9854, - "right_pr_number": 9855, - "code_similarity": 1.0, - "size_similarity": 1.0, - "file_overlap": 1.0, - "area_overlap": 1.0, - "patch_similarity": 1.0, + "left_pr_number": 9083, + "right_pr_number": 10991, + "code_similarity": 0.186, + "size_similarity": 0.466, + "file_overlap": 0.2, + "area_overlap": 0.003, + "patch_similarity": 0.209, "shared_filenames": [ - "src/diffusers/loaders/lora_pipeline.py", - "src/diffusers/loaders/unet.py" + "src/diffusers/loaders/single_file_model.py", + "src/diffusers/loaders/single_file_utils.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/loaders/lora_pipeline.py", + "filename": "src/diffusers/loaders/single_file_model.py", "left_ranges": [ [ - 88, - 104 - ], - [ - 116, - 122 - ], - [ - 127, - 133 - ], - [ - 241, - 249 - ], - [ - 261, - 276 - ], - [ - 280, - 290 - ], - [ - 297, - 303 - ], - [ - 320, - 335 - ], - [ - 572, - 589 - ], - [ - 603, - 609 - ], - [ - 742, - 748 - ], - [ - 760, - 775 - ], - [ - 779, - 786 - ], - [ - 794, - 800 - ], - [ - 817, - 831 - ], - [ - 1695, - 1711 - ], - [ - 1725, - 1731 - ], - [ - 1738, - 1750 - ], - [ - 1762, - 1774 + 24, + 30 ], [ - 1817, - 1828 + 75, + 84 ] ], "right_ranges": [ [ - 88, - 104 - ], - [ - 116, - 122 - ], - [ - 127, - 133 - ], - [ - 241, - 249 + 39, + 46 ], [ - 261, - 276 - ], + 119, + 132 + ] + ] + }, + { + "filename": "src/diffusers/loaders/single_file_utils.py", + "left_ranges": [ [ - 280, - 290 + 77, + 83 ], [ - 297, - 303 + 111, + 118 ], [ - 320, - 335 + 506, + 516 ], [ - 572, - 589 - ], + 1867, + 2061 + ] + ], + "right_ranges": [ [ - 603, - 609 + 117, + 124 ], [ - 742, - 748 + 178, + 186 ], [ - 760, - 775 + 669, + 689 ], [ - 779, - 786 + 2490, + 2496 ], [ - 794, - 800 + 2499, + 2520 ], [ - 817, - 831 + 2523, + 2606 ], [ - 1695, - 1711 - ], + 2895, + 3146 + ] + ] + } + ] + }, + { + "left_pr_number": 10183, + "right_pr_number": 10991, + "code_similarity": 0.194, + "size_similarity": 0.437, + "file_overlap": 0.2, + "area_overlap": 0.0, + "patch_similarity": 0.31, + "shared_filenames": [ + "src/diffusers/loaders/single_file_model.py", + "src/diffusers/loaders/single_file_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders/single_file_model.py", + "left_ranges": [ [ - 1725, - 1731 + 23, + 29 ], [ - 1738, - 1750 - ], + 83, + 89 + ] + ], + "right_ranges": [ [ - 1762, - 1774 + 39, + 46 ], [ - 1817, - 1828 + 119, + 132 ] ] }, { - "filename": "src/diffusers/loaders/unet.py", + "filename": "src/diffusers/loaders/single_file_utils.py", "left_ranges": [ [ - 115, - 121 - ], - [ - 143, - 156 + 92, + 99 ], [ - 216, - 222 + 140, + 149 ], [ - 276, - 284 + 570, + 592 ], [ - 345, - 356 + 2221, + 2295 ] ], "right_ranges": [ [ - 115, - 121 + 117, + 124 + ], + [ + 178, + 186 + ], + [ + 669, + 689 ], [ - 143, - 156 + 2490, + 2496 ], [ - 216, - 222 + 2499, + 2520 ], [ - 276, - 284 + 2523, + 2606 ], [ - 345, - 356 + 2895, + 3146 ] ] } @@ -16365,61 +14165,61 @@ ] }, { - "cluster_id": "cluster-4131-2", + "cluster_id": "cluster-4076-2", "summary": "Cluster of 2 related pull requests linked by soft_similarity.", "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 4138, + "canonical_pr_number": 4079, "issue_numbers": [], "pr_numbers": [ - 4131, - 4138 + 4076, + 4079 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 4131, - "right_pr_number": 4138, - "code_similarity": 0.129, - "size_similarity": 0.103, - "file_overlap": 0.071, - "area_overlap": 0.219, - "patch_similarity": 0.067, + "left_pr_number": 4076, + "right_pr_number": 4079, + "code_similarity": 0.657, + "size_similarity": 0.79, + "file_overlap": 0.2, + "area_overlap": 0.909, + "patch_similarity": 0.809, "shared_filenames": [ - "src/diffusers/__init__.py", - "src/diffusers/pipelines/__init__.py" + "examples/instruct_pix2pix/train_instruct_pix2pix_xl.py", + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/__init__.py", + "filename": "examples/instruct_pix2pix/train_instruct_pix2pix_xl.py", "left_ranges": [ [ - 59, - 67 + 1, + 1051 ] ], "right_ranges": [ [ - 61, - 69 + 1, + 1205 ] ] }, { - "filename": "src/diffusers/pipelines/__init__.py", + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py", "left_ranges": [ [ - 24, - 37 + 1, + 963 ] ], "right_ranges": [ [ - 17, - 23 + 1, + 919 ] ] } @@ -16428,549 +14228,760 @@ ] }, { - "cluster_id": "cluster-1334-2", - "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "cluster_id": "cluster-1028-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 1426, + "canonical_pr_number": 1142, "issue_numbers": [], "pr_numbers": [ - 1334, - 1426 + 1028, + 1071, + 1142 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 1334, - "right_pr_number": 1426, - "code_similarity": 0.313, - "size_similarity": 0.071, - "file_overlap": 0.188, - "area_overlap": 0.614, - "patch_similarity": 0.183, + "left_pr_number": 1028, + "right_pr_number": 1071, + "code_similarity": 0.317, + "size_similarity": 0.62, + "file_overlap": 0.2, + "area_overlap": 0.236, + "patch_similarity": 0.34, "shared_filenames": [ - ".github/workflows/pr_tests.yml", - ".gitignore", - "docker/diffusers-flax-cpu/Dockerfile", - "docker/diffusers-flax-tpu/Dockerfile", - "docker/diffusers-onnxruntime-cpu/Dockerfile", - "docker/diffusers-onnxruntime-cuda/Dockerfile", - "docker/diffusers-pytorch-cpu/Dockerfile", - "docker/diffusers-pytorch-cuda/Dockerfile", - "docs/source/_toctree.yml", - "docs/source/api/pipelines/audio_diffusion.mdx", - "docs/source/api/pipelines/overview.mdx", - "docs/source/index.mdx", - "docs/source/using-diffusers/audio.mdx", + "README.md", + "docs/source/api/schedulers.mdx", + "docs/source/optimization/fp16.mdx", + "examples/community/README.md", + "examples/community/multilingual_stable_diffusion.py", "setup.py", - "src/diffusers/__init__.py", "src/diffusers/dependency_versions_table.py", - "src/diffusers/pipelines/__init__.py", - "src/diffusers/pipelines/audio_diffusion/__init__.py", - "src/diffusers/pipelines/audio_diffusion/pipeline_audio_diffusion.py", + "src/diffusers/pipeline_flax_utils.py", + "src/diffusers/pipeline_utils.py", + "src/diffusers/pipelines/README.md", + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", "src/diffusers/utils/__init__.py", - "src/diffusers/utils/dummy_pt_objects.py", - "src/diffusers/utils/dummy_torch_and_librosa_objects.py", - "src/diffusers/utils/import_utils.py", - "tests/pipelines/audio_diffusion/test_audio_diffusion.py" + "src/diffusers/utils/testing_utils.py", + "tests/models/test_models_unet_2d.py", + "tests/models/test_models_vae.py", + "tests/pipelines/stable_diffusion/test_stable_diffusion.py", + "tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py", + "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint.py", + "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint_legacy.py", + "tests/test_pipelines.py" ], "shared_file_areas": [ { - "filename": ".github/workflows/pr_tests.yml", + "filename": "README.md", "left_ranges": [ [ - 58, - 67 + 182, + 190 ], [ - 129, - 135 + 331, + 355 ] ], "right_ranges": [ [ - 57, - 63 + 142, + 148 ] ] }, { - "filename": ".gitignore", + "filename": "docs/source/api/schedulers.mdx", "left_ranges": [ [ - 165, - 170 + 89, + 95 + ], + [ + 99, + 107 ] ], "right_ranges": [ [ - 165, - 168 + 112, + 130 + ] + ] + }, + { + "filename": "docs/source/optimization/fp16.mdx", + "left_ranges": [ + [ + 18, + 26 + ] + ], + "right_ranges": [ + [ + 22, + 28 + ], + [ + 291, + 331 + ] + ] + }, + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 16, + 24 + ], + [ + 374, + 514 + ] + ], + "right_ranges": [ + [ + 17, + 25 + ], + [ + 375, + 423 + ], + [ + 501, + 572 + ] + ] + }, + { + "filename": "examples/community/multilingual_stable_diffusion.py", + "left_ranges": [ + [ + 1, + 356 + ] + ], + "right_ranges": [ + [ + 1, + 436 + ] + ] + }, + { + "filename": "setup.py", + "left_ranges": [ + [ + 94, + 100 + ], + [ + 182, + 188 + ] + ], + "right_ranges": [ + [ + 89, + 98 + ], + [ + 180, + 185 + ] + ] + }, + { + "filename": "src/diffusers/dependency_versions_table.py", + "left_ranges": [ + [ + 18, + 24 + ] + ], + "right_ranges": [ + [ + 13, + 22 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_flax_utils.py", + "left_ranges": [ + [ + 444, + 454 + ] + ], + "right_ranges": [ + [ + 29, + 35 + ], + [ + 272, + 278 + ], + [ + 302, + 314 + ], + [ + 319, + 325 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 625, + 635 + ] + ], + "right_ranges": [ + [ + 30, + 38 + ], + [ + 360, + 366 + ], + [ + 398, + 411 + ], + [ + 602, + 608 ] ] }, { - "filename": "docker/diffusers-flax-cpu/Dockerfile", + "filename": "src/diffusers/pipelines/README.md", "left_ranges": [ [ - 11, - 17 - ], - [ - 38, - 44 + 30, + 49 ] ], "right_ranges": [ [ - 11, - 17 - ], - [ - 34, - 40 + 16, + 22 ] ] }, { - "filename": "docker/diffusers-flax-tpu/Dockerfile", + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", "left_ranges": [ [ - 11, - 17 - ], - [ - 40, - 46 + 208, + 213 ] ], "right_ranges": [ [ - 11, - 17 + 9, + 21 ], [ - 36, - 42 - ] - ] - }, - { - "filename": "docker/diffusers-onnxruntime-cpu/Dockerfile", - "left_ranges": [ + 58, + 66 + ], [ - 11, - 17 + 80, + 98 ], [ - 38, - 44 - ] - ], - "right_ranges": [ + 113, + 136 + ], [ - 11, - 17 + 172, + 179 ], [ - 34, - 40 + 373, + 383 ] ] }, { - "filename": "docker/diffusers-onnxruntime-cuda/Dockerfile", + "filename": "src/diffusers/utils/__init__.py", "left_ranges": [ [ - 11, - 17 - ], - [ - 38, - 44 + 40, + 55 ] ], "right_ranges": [ [ - 11, - 17 - ], - [ - 34, - 40 + 42, + 48 ] ] }, { - "filename": "docker/diffusers-pytorch-cpu/Dockerfile", + "filename": "src/diffusers/utils/testing_utils.py", "left_ranges": [ [ - 11, + 4, 17 ], [ 37, - 43 + 50 + ], + [ + 168, + 186 ] ], "right_ranges": [ [ - 11, - 17 + 139, + 167 ], [ - 33, - 39 + 191, + 203 ] ] }, { - "filename": "docker/diffusers-pytorch-cuda/Dockerfile", + "filename": "tests/models/test_models_unet_2d.py", "left_ranges": [ [ - 11, - 17 + 21, + 33 ], [ - 37, - 43 - ] - ], - "right_ranges": [ + 68, + 73 + ], [ - 11, - 17 + 150, + 163 ], [ - 33, - 39 - ] - ] - }, - { - "filename": "docs/source/_toctree.yml", - "left_ranges": [ + 208, + 214 + ], [ - 106, - 127 + 288, + 294 + ], + [ + 340, + 348 + ], + [ + 356, + 362 + ], + [ + 378, + 384 + ], + [ + 403, + 596 ] ], "right_ranges": [ [ - 122, - 129 + 21, + 35 + ], + [ + 431, + 437 + ], + [ + 447, + 453 ] ] }, { - "filename": "docs/source/api/pipelines/audio_diffusion.mdx", + "filename": "tests/models/test_models_vae.py", "left_ranges": [ [ 1, - 108 + 303 ] ], "right_ranges": [ [ - 1, - 102 + 20, + 26 + ], + [ + 147, + 153 ] ] }, { - "filename": "docs/source/api/pipelines/overview.mdx", + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", "left_ranges": [ [ - 45, - 51 + 528, + 534 ], [ - 59, - 72 + 548, + 554 + ], + [ + 576, + 582 + ], + [ + 595, + 604 + ], + [ + 634, + 643 + ], + [ + 672, + 678 + ], + [ + 714, + 720 + ], + [ + 740, + 746 ] ], "right_ranges": [ [ - 45, - 51 + 24, + 31 + ], + [ + 363, + 458 + ], + [ + 640, + 652 ] ] }, { - "filename": "docs/source/index.mdx", + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py", "left_ranges": [ [ - 35, - 41 + 488, + 494 ], [ - 49, - 62 + 530, + 536 + ], + [ + 582, + 588 ] ], "right_ranges": [ [ - 35, - 41 + 523, + 530 ] ] }, { - "filename": "docs/source/using-diffusers/audio.mdx", + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint.py", "left_ranges": [ [ - 12, - 16 + 288, + 294 + ], + [ + 330, + 336 + ], + [ + 368, + 376 ] ], "right_ranges": [ [ - 12, - 16 + 28, + 34 + ], + [ + 278, + 287 + ], + [ + 306, + 312 + ], + [ + 317, + 326 + ], + [ + 347, + 353 + ], + [ + 358, + 370 + ], + [ + 385, + 388 ] ] }, { - "filename": "setup.py", + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint_legacy.py", "left_ranges": [ [ - 105, - 111 + 368, + 374 ], [ - 188, - 195 + 414, + 420 ], [ - 214, - 220 + 471, + 477 ] ], "right_ranges": [ [ - 91, - 97 + 31, + 37 ], [ - 182, - 188 + 358, + 367 + ], + [ + 388, + 394 + ], + [ + 400, + 412 + ], + [ + 432, + 438 ] ] }, { - "filename": "src/diffusers/__init__.py", + "filename": "tests/test_pipelines.py", "left_ranges": [ [ - 9, - 15 + 77, + 83 ], [ - 18, - 24 + 86, + 92 ], [ - 30, - 41 + 98, + 104 ], [ - 71, - 87 - ] - ], - "right_ranges": [ - [ - 30, - 43 - ] - ] - }, - { - "filename": "src/diffusers/dependency_versions_table.py", - "left_ranges": [ + 111, + 117 + ], [ - 29, - 33 - ] - ], - "right_ranges": [ + 121, + 127 + ], [ - 15, - 21 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/__init__.py", - "left_ranges": [ + 316, + 324 + ], [ - 1, - 10 + 341, + 349 ], [ - 20, - 51 - ] - ], - "right_ranges": [ + 366, + 372 + ], [ - 1, - 10 + 382, + 393 ], [ - 20, - 30 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/audio_diffusion/__init__.py", - "left_ranges": [ + 404, + 418 + ], [ - 1, - 3 - ] - ], - "right_ranges": [ + 426, + 432 + ], [ - 1, - 3 - ] - ] - }, - { - "filename": "src/diffusers/pipelines/audio_diffusion/pipeline_audio_diffusion.py", - "left_ranges": [ + 448, + 454 + ], [ - 1, - 238 + 472, + 478 ] ], "right_ranges": [ [ - 1, - 248 - ] - ] - }, - { - "filename": "src/diffusers/utils/__init__.py", - "left_ranges": [ + 42, + 48 + ], [ - 26, - 40 - ] - ], - "right_ranges": [ + 446, + 454 + ], [ - 26, - 32 + 462, + 485 + ], + [ + 494, + 510 ] ] - }, + } + ] + }, + { + "left_pr_number": 1028, + "right_pr_number": 1142, + "code_similarity": 0.304, + "size_similarity": 0.196, + "file_overlap": 0.043, + "area_overlap": 0.569, + "patch_similarity": 0.354, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/multilingual_stable_diffusion.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/utils/dummy_pt_objects.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 34, - 54 - ], - [ - 167, - 187 + 16, + 24 ], [ - 242, - 262 + 374, + 514 ] ], "right_ranges": [ [ - 152, - 172 + 18, + 28 ], [ - 257, - 277 + 504, + 577 + ], + [ + 604, + 607 ] ] }, { - "filename": "src/diffusers/utils/dummy_torch_and_librosa_objects.py", + "filename": "examples/community/multilingual_stable_diffusion.py", "left_ranges": [ [ 1, - 41 + 356 ] ], "right_ranges": [ [ 1, - 34 + 436 ] ] - }, + } + ] + }, + { + "left_pr_number": 1071, + "right_pr_number": 1142, + "code_similarity": 0.401, + "size_similarity": 0.122, + "file_overlap": 0.027, + "area_overlap": 0.888, + "patch_similarity": 0.386, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/multilingual_stable_diffusion.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/utils/import_utils.py", + "filename": "examples/community/README.md", "left_ranges": [ [ - 162, - 178 - ], - [ - 229, - 238 - ], - [ - 271, - 282 - ], - [ - 306, - 312 + 17, + 25 ], [ - 321, - 337 + 375, + 423 ], [ - 376, - 392 + 501, + 572 ] ], "right_ranges": [ [ - 180, - 196 - ], - [ - 251, - 260 + 18, + 28 ], [ - 293, - 304 + 504, + 577 ], [ - 328, - 334 + 604, + 607 ] ] }, { - "filename": "tests/pipelines/audio_diffusion/test_audio_diffusion.py", + "filename": "examples/community/multilingual_stable_diffusion.py", "left_ranges": [ [ 1, - 158 + 436 ] ], "right_ranges": [ [ 1, - 157 + 436 ] ] } @@ -16979,149 +14990,176 @@ ] }, { - "cluster_id": "cluster-2479-2", - "summary": "Cluster of 2 related pull requests linked by soft_similarity.", - "status": "closed", + "cluster_id": "cluster-2812-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 3912, + "canonical_pr_number": 2815, "issue_numbers": [], "pr_numbers": [ - 2479, - 3912 + 2812, + 2813, + 2815 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 2479, - "right_pr_number": 3912, - "code_similarity": 0.43, - "size_similarity": 0.945, - "file_overlap": 0.333, - "area_overlap": 0.315, - "patch_similarity": 0.203, + "left_pr_number": 2812, + "right_pr_number": 2813, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, "shared_filenames": [ - "examples/text_to_image/train_text_to_image_lora.py" + "docs/source/en/api/pipelines/stable_unclip.mdx", + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py" ], "shared_file_areas": [ { - "filename": "examples/text_to_image/train_text_to_image_lora.py", + "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", "left_ranges": [ [ - 15, - 22 - ], - [ - 197, - 240 - ], - [ - 469, - 474 - ], - [ - 478, - 556 - ], - [ - 564, - 569 + 16, + 25 ], [ - 587, - 614 + 28, + 42 ], [ - 729, - 747 - ], + 50, + 57 + ] + ], + "right_ranges": [ [ - 799, - 806 + 16, + 25 ], [ - 847, - 860 + 28, + 42 ], [ - 889, - 895 - ], + 50, + 57 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py", + "left_ranges": [ [ - 925, - 948 + 537, + 563 ], [ - 959, - 1003 + 568, + 580 ] ], "right_ranges": [ [ - 15, - 28 - ], - [ - 42, - 49 - ], - [ - 197, - 207 + 537, + 563 ], [ - 374, - 395 - ], + 568, + 580 + ] + ] + } + ] + }, + { + "left_pr_number": 2812, + "right_pr_number": 2815, + "code_similarity": 0.601, + "size_similarity": 0.725, + "file_overlap": 0.5, + "area_overlap": 0.66, + "patch_similarity": 0.503, + "shared_filenames": [ + "docs/source/en/api/pipelines/stable_unclip.mdx" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", + "left_ranges": [ [ - 467, - 485 + 16, + 25 ], [ - 494, - 501 + 28, + 42 ], [ - 507, - 523 - ], + 50, + 57 + ] + ], + "right_ranges": [ [ - 543, - 598 + 16, + 25 ], [ - 617, - 629 + 28, + 42 ], [ - 749, - 762 - ], + 50, + 74 + ] + ] + } + ] + }, + { + "left_pr_number": 2813, + "right_pr_number": 2815, + "code_similarity": 0.601, + "size_similarity": 0.725, + "file_overlap": 0.5, + "area_overlap": 0.66, + "patch_similarity": 0.503, + "shared_filenames": [ + "docs/source/en/api/pipelines/stable_unclip.mdx" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", + "left_ranges": [ [ - 814, - 821 + 16, + 25 ], [ - 888, - 898 + 28, + 42 ], [ - 947, - 953 - ], + 50, + 57 + ] + ], + "right_ranges": [ [ - 960, - 968 + 16, + 25 ], [ - 984, - 1005 + 28, + 42 ], [ - 1024, - 1030 + 50, + 74 ] ] } @@ -17130,387 +15168,590 @@ ] }, { - "cluster_id": "cluster-12351-2", - "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "cluster_id": "cluster-386-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 12366, + "canonical_pr_number": 386, "issue_numbers": [], "pr_numbers": [ - 12351, - 12366 + 386, + 388, + 417 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 12351, - "right_pr_number": 12366, - "code_similarity": 0.657, - "size_similarity": 0.543, - "file_overlap": 1.0, - "area_overlap": 0.525, - "patch_similarity": 0.433, - "shared_filenames": [ - "docs/source/en/_toctree.yml", - "docs/source/en/optimization/cache_dit.md" - ], - "shared_file_areas": [ - { - "filename": "docs/source/en/_toctree.yml", - "left_ranges": [ - [ - 82, - 89 - ] - ], - "right_ranges": [ - [ - 80, - 87 - ] - ] - }, - { - "filename": "docs/source/en/optimization/cache_dit.md", - "left_ranges": [ - [ - 1, - 516 - ] - ], - "right_ranges": [ - [ - 1, - 270 - ] - ] - } - ] + "left_pr_number": 386, + "right_pr_number": 388, + "code_similarity": 0.089, + "size_similarity": 0.244, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.268, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 386, + "right_pr_number": 417, + "code_similarity": 0.068, + "size_similarity": 0.262, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.106, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 388, + "right_pr_number": 417, + "code_similarity": 0.204, + "size_similarity": 0.933, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.113, + "shared_filenames": [], + "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-12004-2", - "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "cluster_id": "cluster-643-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 12024, + "canonical_pr_number": 645, "issue_numbers": [], "pr_numbers": [ - 12004, - 12024 + 643, + 644, + 645 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 12004, - "right_pr_number": 12024, - "code_similarity": 0.387, - "size_similarity": 0.855, - "file_overlap": 0.455, - "area_overlap": 0.157, - "patch_similarity": 0.164, + "left_pr_number": 643, + "right_pr_number": 644, + "code_similarity": 0.505, + "size_similarity": 0.216, + "file_overlap": 1.0, + "area_overlap": 0.304, + "patch_similarity": 0.373, "shared_filenames": [ - "src/diffusers/models/transformers/transformer_wan.py", - "src/diffusers/pipelines/wan/pipeline_wan.py", - "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", - "tests/pipelines/wan/test_wan.py", - "tests/pipelines/wan/test_wan_image_to_video.py" + "examples/textual_inversion/textual_inversion.py", + "src/diffusers/dependency_versions_table.py", + "src/diffusers/models/resnet.py", + "src/diffusers/schedulers/scheduling_sde_ve.py", + "src/diffusers/schedulers/scheduling_sde_ve_flax.py" ], "shared_file_areas": [ { - "filename": "src/diffusers/models/transformers/transformer_wan.py", + "filename": "examples/textual_inversion/textual_inversion.py", "left_ranges": [ [ - 170, - 180 + 29, + 49 ], [ - 312, - 334 + 552, + 559 ], [ - 486, - 507 + 579, + 585 + ] + ], + "right_ranges": [ + [ + 29, + 49 ], [ - 517, - 530 + 552, + 559 + ], + [ + 579, + 585 + ] + ] + }, + { + "filename": "src/diffusers/dependency_versions_table.py", + "left_ranges": [ + [ + 17, + 22 ] ], "right_ranges": [ [ - 324, - 330 + 17, + 22 ] ] }, { - "filename": "src/diffusers/pipelines/wan/pipeline_wan.py", + "filename": "src/diffusers/models/resnet.py", "left_ranges": [ [ - 112, - 131 - ], + 1, + 482 + ] + ], + "right_ranges": [ [ - 134, - 142 + 109, + 132 ], [ - 146, - 157 + 200, + 222 ], [ - 285, - 291 + 378, + 400 ], [ - 318, - 326 - ], + 413, + 434 + ] + ] + }, + { + "filename": "src/diffusers/schedulers/scheduling_sde_ve.py", + "left_ranges": [ [ - 388, - 394 + 94, + 101 ], [ - 427, - 436 - ], + 120, + 129 + ] + ], + "right_ranges": [ [ - 485, - 491 + 94, + 101 ], [ - 495, - 505 - ], + 120, + 129 + ] + ] + }, + { + "filename": "src/diffusers/schedulers/scheduling_sde_ve_flax.py", + "left_ranges": [ [ - 549, - 592 + 104, + 111 ], [ - 595, - 609 + 132, + 141 ] ], "right_ranges": [ [ - 125, - 139 - ], - [ - 526, - 532 + 104, + 111 ], [ - 536, - 546 + 132, + 141 ] ] - }, + } + ] + }, + { + "left_pr_number": 643, + "right_pr_number": 645, + "code_similarity": 0.433, + "size_similarity": 0.03, + "file_overlap": 0.2, + "area_overlap": 1.0, + "patch_similarity": 0.115, + "shared_filenames": [ + "examples/textual_inversion/textual_inversion.py" + ], + "shared_file_areas": [ { - "filename": "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", + "filename": "examples/textual_inversion/textual_inversion.py", "left_ranges": [ [ - 149, - 180 + 29, + 49 ], [ - 186, - 194 + 552, + 559 ], [ - 339, - 345 - ], + 579, + 585 + ] + ], + "right_ranges": [ [ - 383, - 394 + 29, + 49 ], [ - 504, - 510 + 552, + 559 ], [ - 549, - 558 + 579, + 585 + ] + ] + } + ] + }, + { + "left_pr_number": 644, + "right_pr_number": 645, + "code_similarity": 0.472, + "size_similarity": 0.138, + "file_overlap": 0.2, + "area_overlap": 1.0, + "patch_similarity": 0.228, + "shared_filenames": [ + "examples/textual_inversion/textual_inversion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/textual_inversion/textual_inversion.py", + "left_ranges": [ + [ + 29, + 49 ], [ - 615, - 621 + 552, + 559 ], [ - 625, - 635 + 579, + 585 + ] + ], + "right_ranges": [ + [ + 29, + 49 ], [ - 662, - 675 + 552, + 559 ], [ - 700, - 730 + 579, + 585 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-7072-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 7072, + "issue_numbers": [], + "pr_numbers": [ + 7072, + 7233, + 7235 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 7072, + "right_pr_number": 7233, + "code_similarity": 0.087, + "size_similarity": 0.253, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.243, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7072, + "right_pr_number": 7235, + "code_similarity": 0.272, + "size_similarity": 0.687, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.899, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7233, + "right_pr_number": 7235, + "code_similarity": 0.614, + "size_similarity": 0.368, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.271, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 647, + 661 + ], + [ + 1156, + 1162 ], [ - 734, - 748 + 1168, + 1174 ] ], "right_ranges": [ [ - 162, - 178 + 647, + 661 ], [ - 669, - 681 + 1156, + 1162 ], [ - 710, - 716 + 1168, + 1174 ] ] - }, + } + ] + } + ] + }, + { + "cluster_id": "cluster-11445-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 11503, + "issue_numbers": [], + "pr_numbers": [ + 11445, + 11503, + 11507 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 11445, + "right_pr_number": 11503, + "code_similarity": 0.056, + "size_similarity": 0.236, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.056, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11445, + "right_pr_number": 11507, + "code_similarity": 0.111, + "size_similarity": 0.509, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.06, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11503, + "right_pr_number": 11507, + "code_similarity": 0.03, + "size_similarity": 0.12, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.042, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-326-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 327, + "issue_numbers": [], + "pr_numbers": [ + 326, + 327, + 349 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 326, + "right_pr_number": 327, + "code_similarity": 0.549, + "size_similarity": 0.144, + "file_overlap": 0.067, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/diffusers/pipelines/pndm/pipeline_pndm.py" + ], + "shared_file_areas": [ { - "filename": "tests/pipelines/wan/test_wan.py", + "filename": "src/diffusers/pipelines/pndm/pipeline_pndm.py", "left_ranges": [ [ - 85, - 113 + 15, + 47 ] ], "right_ranges": [ [ - 13, - 22 - ], - [ - 87, - 99 - ], - [ - 141, - 185 + 15, + 47 ] ] - }, + } + ] + }, + { + "left_pr_number": 326, + "right_pr_number": 349, + "code_similarity": 0.13, + "size_similarity": 0.202, + "file_overlap": 0.067, + "area_overlap": 0.119, + "patch_similarity": 0.189, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_ddpm.py" + ], + "shared_file_areas": [ { - "filename": "tests/pipelines/wan/test_wan_image_to_video.py", + "filename": "src/diffusers/schedulers/scheduling_ddpm.py", "left_ranges": [ [ - 86, - 108 - ], - [ - 126, - 132 - ], - [ - 182, - 193 - ], - [ - 242, - 265 - ], - [ - 283, - 289 - ], - [ - 340, - 348 + 60, + 65 ] ], "right_ranges": [ [ - 12, + 15, 21 ], [ - 27, - 33 + 51, + 64 ], [ - 88, + 87, 93 ], [ - 111, - 117 - ], - [ - 167, - 210 - ], - [ - 260, - 265 - ], - [ - 283, - 289 - ], - [ - 341, - 381 + 179, + 191 ] ] } ] + }, + { + "left_pr_number": 327, + "right_pr_number": 349, + "code_similarity": 0.17, + "size_similarity": 0.711, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.189, + "shared_filenames": [], + "shared_file_areas": [] } ] }, { - "cluster_id": "cluster-8992-2", - "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "cluster_id": "cluster-5833-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", "status": "merged", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 9040, + "canonical_pr_number": 5834, "issue_numbers": [], "pr_numbers": [ - 8992, - 9040 + 5833, + 5834, + 7934 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 8992, - "right_pr_number": 9040, - "code_similarity": 0.604, - "size_similarity": 0.481, - "file_overlap": 0.167, - "area_overlap": 0.921, - "patch_similarity": 0.906, - "shared_filenames": [ - "examples/community/pipeline_hunyuandit_differential_img2img.py" - ], - "shared_file_areas": [ - { - "filename": "examples/community/pipeline_hunyuandit_differential_img2img.py", - "left_ranges": [ - [ - 1, - 1242 - ] - ], - "right_ranges": [ - [ - 1, - 1144 - ] - ] - } - ] + "left_pr_number": 5833, + "right_pr_number": 5834, + "code_similarity": 0.027, + "size_similarity": 0.128, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.006, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5833, + "right_pr_number": 7934, + "code_similarity": 0.047, + "size_similarity": 0.22, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.017, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5834, + "right_pr_number": 7934, + "code_similarity": 0.117, + "size_similarity": 0.583, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] } ] } @@ -17533,27 +15774,33 @@ "reason": "Issues in cluster-11702-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." }, { - "cluster_id": "cluster-1246-22", + "cluster_id": "cluster-1246-28", "canonical_issue_number": 13124, "duplicate_issue_numbers": [ 1246, 2780, + 4566, + 5412, + 5757, + 5840, 6086, 6098, 6124, 6277, 6363, + 6366, 6442, 6454, 6482, 7330, + 8659, 8871, 9628, 9973, 10752, 10910 ], - "reason": "Issues in cluster-1246-22 are treated as duplicates because they share closing_reference, duplicate_reference, shared_issue_target, soft_similarity evidence." + "reason": "Issues in cluster-1246-28 are treated as duplicates because they share closing_reference, duplicate_reference, shared_issue_target, soft_similarity evidence." }, { "cluster_id": "cluster-1257-4", @@ -17580,12 +15827,21 @@ "reason": "Issues in cluster-1466-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." }, { - "cluster_id": "cluster-3880-4", - "canonical_issue_number": 3880, + "cluster_id": "cluster-2098-6", + "canonical_issue_number": 3008, + "duplicate_issue_numbers": [ + 2098, + 3184 + ], + "reason": "Issues in cluster-2098-6 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-281-4", + "canonical_issue_number": 281, "duplicate_issue_numbers": [ - 4782 + 725 ], - "reason": "Issues in cluster-3880-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." + "reason": "Issues in cluster-281-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." }, { "cluster_id": "cluster-4478-9", @@ -17608,28 +15864,13 @@ "reason": "Issues in cluster-6137-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." }, { - "cluster_id": "cluster-6226-4", - "canonical_issue_number": 11285, - "duplicate_issue_numbers": [ - 6226 - ], - "reason": "Issues in cluster-6226-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." - }, - { - "cluster_id": "cluster-9237-4", + "cluster_id": "cluster-9237-5", "canonical_issue_number": 9237, "duplicate_issue_numbers": [ + 9497, 9548 ], - "reason": "Issues in cluster-9237-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." - }, - { - "cluster_id": "cluster-9403-4", - "canonical_issue_number": 9403, - "duplicate_issue_numbers": [ - 9825 - ], - "reason": "Issues in cluster-9403-4 are treated as duplicates because they share closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "Issues in cluster-9237-5 are treated as duplicates because they share closing_reference, soft_similarity evidence." }, { "cluster_id": "cluster-9514-5", @@ -17643,13 +15884,14 @@ ], "duplicate_prs": [ { - "cluster_id": "cluster-10025-3", - "canonical_pr_number": 10053, + "cluster_id": "cluster-10025-4", + "canonical_pr_number": 10044, "duplicate_pr_numbers": [ - 10051 + 10051, + 10053 ], "target_issue_number": 10025, - "reason": "PRs in cluster-10025-3 are treated as duplicates because they converge on issue #10025 with closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "PRs in cluster-10025-4 are treated as duplicates because they converge on issue #10025 with closing_reference, shared_issue_target, soft_similarity evidence." }, { "cluster_id": "cluster-10076-3", @@ -17661,33 +15903,43 @@ "reason": "PRs in cluster-10076-3 are treated as duplicates because they converge on issue #10076 with closing_reference evidence." }, { - "cluster_id": "cluster-1028-3", - "canonical_pr_number": 1142, + "cluster_id": "cluster-10109-5", + "canonical_pr_number": 10498, "duplicate_pr_numbers": [ - 1028, - 1071 + 10109, + 10364, + 10365 + ], + "target_issue_number": 10357, + "reason": "PRs in cluster-10109-5 are treated as duplicates because they converge on issue #10357 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-10132-2", + "canonical_pr_number": 10132, + "duplicate_pr_numbers": [ + 10497 ], "target_issue_number": null, - "reason": "PRs in cluster-1028-3 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-10132-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-10299-2", - "canonical_pr_number": 10316, + "cluster_id": "cluster-10279-2", + "canonical_pr_number": 10279, "duplicate_pr_numbers": [ - 10299 + 10323 ], "target_issue_number": null, - "reason": "PRs in cluster-10299-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-10279-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-10357-4", - "canonical_pr_number": 10498, + "cluster_id": "cluster-1028-3", + "canonical_pr_number": 1142, "duplicate_pr_numbers": [ - 10364, - 10365 + 1028, + 1071 ], - "target_issue_number": 10357, - "reason": "PRs in cluster-10357-4 are treated as duplicates because they converge on issue #10357 with closing_reference, shared_issue_target, soft_similarity evidence." + "target_issue_number": null, + "reason": "PRs in cluster-1028-3 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-10386-2", @@ -17707,6 +15959,15 @@ "target_issue_number": null, "reason": "PRs in cluster-10402-2 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-10453-3", + "canonical_pr_number": 10454, + "duplicate_pr_numbers": [ + 10482 + ], + "target_issue_number": 10453, + "reason": "PRs in cluster-10453-3 are treated as duplicates because they converge on issue #10453 with closing_reference, soft_similarity evidence." + }, { "cluster_id": "cluster-10464-2", "canonical_pr_number": 10465, @@ -17734,6 +15995,15 @@ "target_issue_number": null, "reason": "PRs in cluster-10517-2 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-10678-2", + "canonical_pr_number": 10784, + "duplicate_pr_numbers": [ + 10678 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10678-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-10817-3", "canonical_pr_number": 10888, @@ -17780,13 +16050,13 @@ "reason": "PRs in cluster-11208-3 are treated as duplicates because they converge on issue #11208 with closing_reference evidence." }, { - "cluster_id": "cluster-11257-2", - "canonical_pr_number": 11701, + "cluster_id": "cluster-11285-3", + "canonical_pr_number": 12964, "duplicate_pr_numbers": [ - 11257 + 11306 ], - "target_issue_number": null, - "reason": "PRs in cluster-11257-2 are treated as duplicates because they share soft_similarity evidence." + "target_issue_number": 11285, + "reason": "PRs in cluster-11285-3 are treated as duplicates because they converge on issue #11285 with closing_reference evidence." }, { "cluster_id": "cluster-11315-4", @@ -17807,20 +16077,17 @@ "reason": "PRs in cluster-11358-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-11390-10", + "cluster_id": "cluster-11390-7", "canonical_pr_number": 11463, "duplicate_pr_numbers": [ 11395, - 11421, 11426, 11427, - 11471, - 11472, 11491, 11492 ], "target_issue_number": 11390, - "reason": "PRs in cluster-11390-10 are treated as duplicates because they converge on issue #11390 with closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "PRs in cluster-11390-7 are treated as duplicates because they converge on issue #11390 with closing_reference, shared_issue_target, soft_similarity evidence." }, { "cluster_id": "cluster-11445-3", @@ -17832,6 +16099,24 @@ "target_issue_number": null, "reason": "PRs in cluster-11445-3 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-11455-2", + "canonical_pr_number": 11455, + "duplicate_pr_numbers": [ + 11465 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11455-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-11471-2", + "canonical_pr_number": 11471, + "duplicate_pr_numbers": [ + 11472 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11471-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-11522-2", "canonical_pr_number": 12069, @@ -17841,6 +16126,15 @@ "target_issue_number": null, "reason": "PRs in cluster-11522-2 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-11663-2", + "canonical_pr_number": 11738, + "duplicate_pr_numbers": [ + 11663 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11663-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-11702-4", "canonical_pr_number": 11704, @@ -17858,16 +16152,26 @@ 11783 ], "target_issue_number": 11730, - "reason": "PRs in cluster-11730-4 are treated as duplicates because they converge on issue #11730 with closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "PRs in cluster-11730-4 are treated as duplicates because they converge on issue #11730 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-11995-2", + "canonical_pr_number": 11995, + "duplicate_pr_numbers": [ + 12086 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11995-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-12004-2", - "canonical_pr_number": 12024, + "cluster_id": "cluster-12004-3", + "canonical_pr_number": 12006, "duplicate_pr_numbers": [ - 12004 + 12004, + 12024 ], "target_issue_number": null, - "reason": "PRs in cluster-12004-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-12004-3 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-12049-2", @@ -17951,7 +16255,7 @@ "reason": "PRs in cluster-12458-3 are treated as duplicates because they converge on issue #12458 with closing_reference evidence." }, { - "cluster_id": "cluster-1246-22", + "cluster_id": "cluster-1246-28", "canonical_pr_number": 6566, "duplicate_pr_numbers": [ 10783, @@ -17960,7 +16264,16 @@ 13510 ], "target_issue_number": 13124, - "reason": "PRs in cluster-1246-22 are treated as duplicates because they converge on issue #13124 with closing_reference, duplicate_reference, shared_issue_target, soft_similarity evidence." + "reason": "PRs in cluster-1246-28 are treated as duplicates because they converge on issue #13124 with closing_reference, duplicate_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-12496-3", + "canonical_pr_number": 12523, + "duplicate_pr_numbers": [ + 12496 + ], + "target_issue_number": 12499, + "reason": "PRs in cluster-12496-3 are treated as duplicates because they converge on issue #12499 with closing_reference, soft_similarity evidence." }, { "cluster_id": "cluster-12540-3", @@ -18035,6 +16348,15 @@ "target_issue_number": 12771, "reason": "PRs in cluster-12771-4 are treated as duplicates because they converge on issue #12771 with closing_reference, shared_issue_target, soft_similarity evidence." }, + { + "cluster_id": "cluster-12782-2", + "canonical_pr_number": 13039, + "duplicate_pr_numbers": [ + 12782 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12782-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-12869-2", "canonical_pr_number": 12870, @@ -18062,9 +16384,18 @@ "target_issue_number": 12913, "reason": "PRs in cluster-12913-3 are treated as duplicates because they converge on issue #12913 with closing_reference, shared_issue_target, soft_similarity evidence." }, + { + "cluster_id": "cluster-13042-2", + "canonical_pr_number": 13042, + "duplicate_pr_numbers": [ + 13424 + ], + "target_issue_number": null, + "reason": "PRs in cluster-13042-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-13069-2", - "canonical_pr_number": 13071, + "canonical_pr_number": 13082, "duplicate_pr_numbers": [ 13069 ], @@ -18109,6 +16440,15 @@ "target_issue_number": 13243, "reason": "PRs in cluster-13243-3 are treated as duplicates because they converge on issue #13243 with closing_reference, shared_issue_target evidence." }, + { + "cluster_id": "cluster-13259-2", + "canonical_pr_number": 13259, + "duplicate_pr_numbers": [ + 13567 + ], + "target_issue_number": null, + "reason": "PRs in cluster-13259-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-13292-3", "canonical_pr_number": 13415, @@ -18190,6 +16530,15 @@ "target_issue_number": 1792, "reason": "PRs in cluster-1407-4 are treated as duplicates because they converge on issue #1792 with closing_reference, soft_similarity evidence." }, + { + "cluster_id": "cluster-1415-2", + "canonical_pr_number": 2757, + "duplicate_pr_numbers": [ + 1415 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1415-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-1466-4", "canonical_pr_number": 2070, @@ -18245,6 +16594,16 @@ "target_issue_number": null, "reason": "PRs in cluster-2067-3 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-2098-6", + "canonical_pr_number": 3197, + "duplicate_pr_numbers": [ + 3201, + 3265 + ], + "target_issue_number": 3008, + "reason": "PRs in cluster-2098-6 are treated as duplicates because they converge on issue #3008 with closing_reference, soft_similarity evidence." + }, { "cluster_id": "cluster-2206-2", "canonical_pr_number": 2206, @@ -18272,15 +16631,6 @@ "target_issue_number": null, "reason": "PRs in cluster-2383-2 are treated as duplicates because they share soft_similarity evidence." }, - { - "cluster_id": "cluster-2388-2", - "canonical_pr_number": 2388, - "duplicate_pr_numbers": [ - 2406 - ], - "target_issue_number": null, - "reason": "PRs in cluster-2388-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-2417-2", "canonical_pr_number": 2418, @@ -18299,15 +16649,6 @@ "target_issue_number": null, "reason": "PRs in cluster-2450-2 are treated as duplicates because they share soft_similarity evidence." }, - { - "cluster_id": "cluster-2479-2", - "canonical_pr_number": 3912, - "duplicate_pr_numbers": [ - 2479 - ], - "target_issue_number": null, - "reason": "PRs in cluster-2479-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-2541-2", "canonical_pr_number": 2566, @@ -18318,24 +16659,14 @@ "reason": "PRs in cluster-2541-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-2637-2", - "canonical_pr_number": 2637, - "duplicate_pr_numbers": [ - 3825 - ], - "target_issue_number": null, - "reason": "PRs in cluster-2637-2 are treated as duplicates because they share soft_similarity evidence." - }, - { - "cluster_id": "cluster-2642-4", + "cluster_id": "cluster-2647-3", "canonical_pr_number": 2684, "duplicate_pr_numbers": [ - 2642, 2647, 2748 ], "target_issue_number": null, - "reason": "PRs in cluster-2642-4 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-2647-3 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-2741-2", @@ -18346,6 +16677,15 @@ "target_issue_number": null, "reason": "PRs in cluster-2741-2 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-281-4", + "canonical_pr_number": 361, + "duplicate_pr_numbers": [ + 727 + ], + "target_issue_number": 281, + "reason": "PRs in cluster-281-4 are treated as duplicates because they converge on issue #281 with closing_reference, soft_similarity evidence." + }, { "cluster_id": "cluster-2812-3", "canonical_pr_number": 2815, @@ -18374,15 +16714,6 @@ "target_issue_number": null, "reason": "PRs in cluster-2842-2 are treated as duplicates because they share soft_similarity evidence." }, - { - "cluster_id": "cluster-2915-2", - "canonical_pr_number": 3850, - "duplicate_pr_numbers": [ - 2915 - ], - "target_issue_number": null, - "reason": "PRs in cluster-2915-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-3084-2", "canonical_pr_number": 3459, @@ -18403,22 +16734,23 @@ "reason": "PRs in cluster-3168-3 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-3201-2", - "canonical_pr_number": 3265, + "cluster_id": "cluster-3237-2", + "canonical_pr_number": 3285, "duplicate_pr_numbers": [ - 3201 + 3237 ], "target_issue_number": null, - "reason": "PRs in cluster-3201-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-3237-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-3237-2", - "canonical_pr_number": 3285, + "cluster_id": "cluster-326-3", + "canonical_pr_number": 327, "duplicate_pr_numbers": [ - 3237 + 326, + 349 ], "target_issue_number": null, - "reason": "PRs in cluster-3237-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-326-3 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-3414-2", @@ -18430,22 +16762,22 @@ "reason": "PRs in cluster-3414-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-3458-2", - "canonical_pr_number": 3460, + "cluster_id": "cluster-344-2", + "canonical_pr_number": 344, "duplicate_pr_numbers": [ - 3458 + 365 ], "target_issue_number": null, - "reason": "PRs in cluster-3458-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-344-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-3470-2", - "canonical_pr_number": 3470, + "cluster_id": "cluster-3458-2", + "canonical_pr_number": 3460, "duplicate_pr_numbers": [ - 3472 + 3458 ], "target_issue_number": null, - "reason": "PRs in cluster-3470-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-3458-2 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-3554-2", @@ -18456,15 +16788,6 @@ "target_issue_number": null, "reason": "PRs in cluster-3554-2 are treated as duplicates because they share soft_similarity evidence." }, - { - "cluster_id": "cluster-3586-2", - "canonical_pr_number": 5672, - "duplicate_pr_numbers": [ - 3586 - ], - "target_issue_number": null, - "reason": "PRs in cluster-3586-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-3690-3", "canonical_pr_number": 3695, @@ -18484,13 +16807,32 @@ "reason": "PRs in cluster-3709-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-3880-4", + "cluster_id": "cluster-3797-2", + "canonical_pr_number": 3797, + "duplicate_pr_numbers": [ + 3957 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3797-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-386-3", + "canonical_pr_number": 386, + "duplicate_pr_numbers": [ + 388, + 417 + ], + "target_issue_number": null, + "reason": "PRs in cluster-386-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3880-3", "canonical_pr_number": 4536, "duplicate_pr_numbers": [ 3921 ], "target_issue_number": 3880, - "reason": "PRs in cluster-3880-4 are treated as duplicates because they converge on issue #3880 with closing_reference, soft_similarity evidence." + "reason": "PRs in cluster-3880-3 are treated as duplicates because they converge on issue #3880 with closing_reference evidence." }, { "cluster_id": "cluster-3896-3", @@ -18502,6 +16844,24 @@ "target_issue_number": null, "reason": "PRs in cluster-3896-3 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-395-2", + "canonical_pr_number": 395, + "duplicate_pr_numbers": [ + 679 + ], + "target_issue_number": null, + "reason": "PRs in cluster-395-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3968-2", + "canonical_pr_number": 3968, + "duplicate_pr_numbers": [ + 4346 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3968-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-4076-2", "canonical_pr_number": 4079, @@ -18520,6 +16880,24 @@ "target_issue_number": null, "reason": "PRs in cluster-4131-2 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-4315-2", + "canonical_pr_number": 5298, + "duplicate_pr_numbers": [ + 4315 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4315-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-433-2", + "canonical_pr_number": 433, + "duplicate_pr_numbers": [ + 2007 + ], + "target_issue_number": null, + "reason": "PRs in cluster-433-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-4475-2", "canonical_pr_number": 4493, @@ -18615,6 +16993,15 @@ "target_issue_number": 4704, "reason": "PRs in cluster-4704-3 are treated as duplicates because they converge on issue #4704 with closing_reference evidence." }, + { + "cluster_id": "cluster-4780-2", + "canonical_pr_number": 5151, + "duplicate_pr_numbers": [ + 4780 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4780-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-4810-2", "canonical_pr_number": 4810, @@ -18624,6 +17011,15 @@ "target_issue_number": null, "reason": "PRs in cluster-4810-2 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-483-2", + "canonical_pr_number": 483, + "duplicate_pr_numbers": [ + 978 + ], + "target_issue_number": null, + "reason": "PRs in cluster-483-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-4833-2", "canonical_pr_number": 4833, @@ -18643,31 +17039,14 @@ "reason": "PRs in cluster-4899-3 are treated as duplicates because they converge on issue #5800 with closing_reference, soft_similarity evidence." }, { - "cluster_id": "cluster-4944-2", - "canonical_pr_number": 5713, - "duplicate_pr_numbers": [ - 4944 - ], - "target_issue_number": null, - "reason": "PRs in cluster-4944-2 are treated as duplicates because they share soft_similarity evidence." - }, - { - "cluster_id": "cluster-4947-2", - "canonical_pr_number": 4947, - "duplicate_pr_numbers": [ - 5188 - ], - "target_issue_number": null, - "reason": "PRs in cluster-4947-2 are treated as duplicates because they share soft_similarity evidence." - }, - { - "cluster_id": "cluster-5062-3", + "cluster_id": "cluster-5062-4", "canonical_pr_number": 5066, "duplicate_pr_numbers": [ - 5065 + 5065, + 5332 ], "target_issue_number": 5062, - "reason": "PRs in cluster-5062-3 are treated as duplicates because they converge on issue #5062 with closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "PRs in cluster-5062-4 are treated as duplicates because they converge on issue #5062 with closing_reference, shared_issue_target, soft_similarity evidence." }, { "cluster_id": "cluster-5214-3", @@ -18743,6 +17122,15 @@ "target_issue_number": null, "reason": "PRs in cluster-5347-2 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-5382-2", + "canonical_pr_number": 5382, + "duplicate_pr_numbers": [ + 5614 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5382-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-5504-3", "canonical_pr_number": 5697, @@ -18781,32 +17169,14 @@ "reason": "PRs in cluster-5712-3 are treated as duplicates because they converge on issue #5712 with closing_reference, soft_similarity evidence." }, { - "cluster_id": "cluster-5725-3", - "canonical_pr_number": 5748, - "duplicate_pr_numbers": [ - 5725, - 5744 - ], - "target_issue_number": null, - "reason": "PRs in cluster-5725-3 are treated as duplicates because they share soft_similarity evidence." - }, - { - "cluster_id": "cluster-5828-2", - "canonical_pr_number": 5828, - "duplicate_pr_numbers": [ - 7746 - ], - "target_issue_number": null, - "reason": "PRs in cluster-5828-2 are treated as duplicates because they share soft_similarity evidence." - }, - { - "cluster_id": "cluster-5833-2", + "cluster_id": "cluster-5833-3", "canonical_pr_number": 5834, "duplicate_pr_numbers": [ - 5833 + 5833, + 7934 ], "target_issue_number": null, - "reason": "PRs in cluster-5833-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-5833-3 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-5878-2", @@ -18845,13 +17215,22 @@ "reason": "PRs in cluster-5970-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-5983-2", - "canonical_pr_number": 6466, + "cluster_id": "cluster-5982-2", + "canonical_pr_number": 5982, + "duplicate_pr_numbers": [ + 6065 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5982-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6014-2", + "canonical_pr_number": 7638, "duplicate_pr_numbers": [ - 5983 + 6014 ], "target_issue_number": null, - "reason": "PRs in cluster-5983-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-6014-2 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-6119-2", @@ -18890,41 +17269,24 @@ "reason": "PRs in cluster-6158-3 are treated as duplicates because they converge on issue #6158 with closing_reference, shared_issue_target, soft_similarity evidence." }, { - "cluster_id": "cluster-6207-2", - "canonical_pr_number": 6266, - "duplicate_pr_numbers": [ - 6207 - ], - "target_issue_number": null, - "reason": "PRs in cluster-6207-2 are treated as duplicates because they share soft_similarity evidence." - }, - { - "cluster_id": "cluster-6208-2", - "canonical_pr_number": 6208, + "cluster_id": "cluster-6261-2", + "canonical_pr_number": 6261, "duplicate_pr_numbers": [ - 6241 + 6265 ], "target_issue_number": null, - "reason": "PRs in cluster-6208-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-6261-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-6226-4", - "canonical_pr_number": 12964, - "duplicate_pr_numbers": [ - 11306 - ], - "target_issue_number": 11285, - "reason": "PRs in cluster-6226-4 are treated as duplicates because they converge on issue #11285 with closing_reference, soft_similarity evidence." - }, - { - "cluster_id": "cluster-6299-3", - "canonical_pr_number": 6299, + "cluster_id": "cluster-6299-4", + "canonical_pr_number": 6322, "duplicate_pr_numbers": [ + 6299, 6300, 6301 ], "target_issue_number": null, - "reason": "PRs in cluster-6299-3 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-6299-4 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-643-3", @@ -18946,18 +17308,16 @@ "reason": "PRs in cluster-6453-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-6503-8", + "cluster_id": "cluster-6503-6", "canonical_pr_number": 6532, "duplicate_pr_numbers": [ 6525, - 6526, 6534, - 6550, 6555, 6556 ], "target_issue_number": 6503, - "reason": "PRs in cluster-6503-8 are treated as duplicates because they converge on issue #6503 with closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "PRs in cluster-6503-6 are treated as duplicates because they converge on issue #6503 with closing_reference, shared_issue_target, soft_similarity evidence." }, { "cluster_id": "cluster-6514-4", @@ -18970,6 +17330,15 @@ "target_issue_number": null, "reason": "PRs in cluster-6514-4 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-6545-3", + "canonical_pr_number": 6569, + "duplicate_pr_numbers": [ + 8283 + ], + "target_issue_number": 6545, + "reason": "PRs in cluster-6545-3 are treated as duplicates because they converge on issue #6545 with closing_reference, soft_similarity evidence." + }, { "cluster_id": "cluster-6585-11", "canonical_pr_number": 6737, @@ -18985,7 +17354,7 @@ 6781 ], "target_issue_number": 6585, - "reason": "PRs in cluster-6585-11 are treated as duplicates because they converge on issue #6585 with closing_reference, shared_issue_target evidence." + "reason": "PRs in cluster-6585-11 are treated as duplicates because they converge on issue #6585 with closing_reference, shared_issue_target, soft_similarity evidence." }, { "cluster_id": "cluster-6587-2", @@ -19042,6 +17411,15 @@ "target_issue_number": 6891, "reason": "PRs in cluster-6891-3 are treated as duplicates because they converge on issue #6891 with closing_reference, shared_issue_target, soft_similarity evidence." }, + { + "cluster_id": "cluster-6909-2", + "canonical_pr_number": 6909, + "duplicate_pr_numbers": [ + 6939 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6909-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-6980-2", "canonical_pr_number": 6983, @@ -19064,15 +17442,6 @@ "target_issue_number": 6984, "reason": "PRs in cluster-6984-7 are treated as duplicates because they converge on issue #6984 with closing_reference, shared_issue_target, soft_similarity evidence." }, - { - "cluster_id": "cluster-70-2", - "canonical_pr_number": 78, - "duplicate_pr_numbers": [ - 70 - ], - "target_issue_number": null, - "reason": "PRs in cluster-70-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-7038-3", "canonical_pr_number": 7550, @@ -19083,31 +17452,32 @@ "reason": "PRs in cluster-7038-3 are treated as duplicates because they converge on issue #7038 with closing_reference evidence." }, { - "cluster_id": "cluster-7068-2", - "canonical_pr_number": 8572, + "cluster_id": "cluster-7072-3", + "canonical_pr_number": 7072, "duplicate_pr_numbers": [ - 7068 + 7233, + 7235 ], "target_issue_number": null, - "reason": "PRs in cluster-7068-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-7072-3 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-7092-2", - "canonical_pr_number": 7092, + "cluster_id": "cluster-709-3", + "canonical_pr_number": 710, "duplicate_pr_numbers": [ - 8687 + 798 ], - "target_issue_number": null, - "reason": "PRs in cluster-7092-2 are treated as duplicates because they share soft_similarity evidence." + "target_issue_number": 709, + "reason": "PRs in cluster-709-3 are treated as duplicates because they converge on issue #709 with closing_reference, soft_similarity evidence." }, { - "cluster_id": "cluster-7233-2", - "canonical_pr_number": 7235, + "cluster_id": "cluster-7092-2", + "canonical_pr_number": 7092, "duplicate_pr_numbers": [ - 7233 + 8687 ], "target_issue_number": null, - "reason": "PRs in cluster-7233-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-7092-2 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-7677-3", @@ -19155,22 +17525,22 @@ "reason": "PRs in cluster-7736-3 are treated as duplicates because they converge on issue #7736 with closing_reference evidence." }, { - "cluster_id": "cluster-8283-2", - "canonical_pr_number": 8283, + "cluster_id": "cluster-8308-2", + "canonical_pr_number": 8310, "duplicate_pr_numbers": [ - 8335 + 8308 ], "target_issue_number": null, - "reason": "PRs in cluster-8283-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-8308-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-8308-2", - "canonical_pr_number": 8310, + "cluster_id": "cluster-831-2", + "canonical_pr_number": 831, "duplicate_pr_numbers": [ - 8308 + 1490 ], "target_issue_number": null, - "reason": "PRs in cluster-8308-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-831-2 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-8384-4", @@ -19203,15 +17573,6 @@ "target_issue_number": null, "reason": "PRs in cluster-8478-2 are treated as duplicates because they share soft_similarity evidence." }, - { - "cluster_id": "cluster-8607-2", - "canonical_pr_number": 8607, - "duplicate_pr_numbers": [ - 8618 - ], - "target_issue_number": null, - "reason": "PRs in cluster-8607-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-8710-3", "canonical_pr_number": 9386, @@ -19230,15 +17591,6 @@ "target_issue_number": null, "reason": "PRs in cluster-8773-2 are treated as duplicates because they share soft_similarity evidence." }, - { - "cluster_id": "cluster-8775-2", - "canonical_pr_number": 8775, - "duplicate_pr_numbers": [ - 9760 - ], - "target_issue_number": null, - "reason": "PRs in cluster-8775-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-8798-2", "canonical_pr_number": 8798, @@ -19285,6 +17637,24 @@ "target_issue_number": null, "reason": "PRs in cluster-8899-2 are treated as duplicates because they share soft_similarity evidence." }, + { + "cluster_id": "cluster-8932-2", + "canonical_pr_number": 8987, + "duplicate_pr_numbers": [ + 8932 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8932-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8946-2", + "canonical_pr_number": 9131, + "duplicate_pr_numbers": [ + 8946 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8946-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-8949-2", "canonical_pr_number": 8949, @@ -19313,22 +17683,32 @@ "reason": "PRs in cluster-8973-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-8992-2", - "canonical_pr_number": 9040, + "cluster_id": "cluster-9080-2", + "canonical_pr_number": 9082, "duplicate_pr_numbers": [ - 8992 + 9080 ], "target_issue_number": null, - "reason": "PRs in cluster-8992-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-9080-2 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-9080-2", - "canonical_pr_number": 9082, + "cluster_id": "cluster-9083-3", + "canonical_pr_number": 9083, "duplicate_pr_numbers": [ - 9080 + 10183, + 10991 ], "target_issue_number": null, - "reason": "PRs in cluster-9080-2 are treated as duplicates because they share soft_similarity evidence." + "reason": "PRs in cluster-9083-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-912-2", + "canonical_pr_number": 912, + "duplicate_pr_numbers": [ + 1131 + ], + "target_issue_number": null, + "reason": "PRs in cluster-912-2 are treated as duplicates because they share soft_similarity evidence." }, { "cluster_id": "cluster-916-4", @@ -19342,22 +17722,22 @@ "reason": "PRs in cluster-916-4 are treated as duplicates because they share soft_similarity evidence." }, { - "cluster_id": "cluster-9237-4", + "cluster_id": "cluster-9237-5", "canonical_pr_number": 9419, "duplicate_pr_numbers": [ 9549 ], "target_issue_number": 9237, - "reason": "PRs in cluster-9237-4 are treated as duplicates because they converge on issue #9237 with closing_reference, soft_similarity evidence." + "reason": "PRs in cluster-9237-5 are treated as duplicates because they converge on issue #9237 with closing_reference, soft_similarity evidence." }, { - "cluster_id": "cluster-9403-4", + "cluster_id": "cluster-9403-3", "canonical_pr_number": 10261, "duplicate_pr_numbers": [ 9787 ], "target_issue_number": 9403, - "reason": "PRs in cluster-9403-4 are treated as duplicates because they converge on issue #9403 with closing_reference, shared_issue_target, soft_similarity evidence." + "reason": "PRs in cluster-9403-3 are treated as duplicates because they converge on issue #9403 with closing_reference, shared_issue_target evidence." }, { "cluster_id": "cluster-9514-5", @@ -19442,15 +17822,15 @@ } ], "best_issue": { - "cluster_id": "cluster-1246-22", + "cluster_id": "cluster-1246-28", "issue_number": 13124, - "reason": "Issue #13124 is the strongest global issue candidate because it is open, belongs to a cluster with 22 artifacts, and carries 2 discussion comments plus 6 inbound references.", - "score": 330.0 + "reason": "Issue #13124 is the strongest global issue candidate because it is open, belongs to a cluster with 28 artifacts, and carries 2 discussion comments plus 6 inbound references.", + "score": 390.0 }, "best_pr": { - "cluster_id": "cluster-13074-4", - "pr_number": 13075, - "reason": "PR #13075 is the strongest global PR candidate because it is open, belongs to a cluster with 4 artifacts, links to 1 issues, and carries 12 review/discussion events.", - "score": 188.98 + "cluster_id": "cluster-10025-4", + "pr_number": 10044, + "reason": "PR #10044 is the strongest global PR candidate because it is open, belongs to a cluster with 4 artifacts, links to 2 issues, and carries 32 review/discussion events.", + "score": 233.939 } } diff --git a/analysis/current/analysis-report-hybrid.llm-reviews.json b/analysis/current/analysis-report-hybrid.llm-reviews.json index 57e235bbcb831bf893d71fb449e656a451ce9995..86fac2f1a1889e2ddf4df96763c182ad0211d4fb 100644 --- a/analysis/current/analysis-report-hybrid.llm-reviews.json +++ b/analysis/current/analysis-report-hybrid.llm-reviews.json @@ -1,8 +1,8 @@ { "schema_version": "1.0", "repo": "huggingface/diffusers", - "snapshot_id": "20260427T112541Z", - "generated_at": "2026-04-27T11:44:46Z", + "snapshot_id": "20260427T113512Z", + "generated_at": "2026-04-27T12:02:03Z", "model": "gpt-5.4-mini?service_tier=flex", "reviews": [ { @@ -101,10 +101,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -402,10 +402,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -705,10 +705,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -986,10 +986,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -1268,10 +1268,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -1534,10 +1534,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -1843,10 +1843,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -2161,10 +2161,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -2457,10 +2457,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -2746,10 +2746,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -3052,10 +3052,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -3332,10 +3332,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -3620,10 +3620,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -3923,10 +3923,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -4234,10 +4234,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -4538,10 +4538,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -4820,10 +4820,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -5112,10 +5112,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -5397,10 +5397,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -5678,10 +5678,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -5964,10 +5964,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -6240,10 +6240,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -6521,10 +6521,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -6806,10 +6806,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -7068,10 +7068,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -7347,10 +7347,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -7626,10 +7626,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -7907,10 +7907,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -8195,10 +8195,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -8415,10 +8415,8 @@ "issue:11178", "issue:11209", "issue:11224", - "issue:12499", "issue:1260", "issue:1263", - "issue:12896", "issue:13429", "issue:13430", "issue:1495", @@ -8438,6 +8436,7 @@ "issue:3788", "issue:4094", "issue:4241", + "issue:4728", "issue:4779", "issue:4945", "issue:5455", @@ -8445,6 +8444,7 @@ "issue:5689", "issue:6149", "issue:6636", + "issue:7114", "issue:7228", "issue:7513", "issue:7593", @@ -8463,6 +8463,7 @@ "issue:1263|issue:1495", "issue:2904|issue:7228", "issue:4094|issue:4945", + "issue:4728|issue:7114", "issue:7593|issue:858", "issue:11209|issue:11224", "issue:13429|issue:13430", @@ -8481,193 +8482,198 @@ "issue:9497|issue:9500", "issue:2635|issue:2676", "issue:11178|issue:9208", - "issue:3432|issue:3453", - "issue:12499|issue:12896" + "issue:3432|issue:3453" ], - "prepared_review_unit_hash": "d53e39062b6007f22d0486a8154c726d0bf5aa2b0346e07a7ea9f1098c169cb4", + "prepared_review_unit_hash": "df5c3c4fb92ea3d48294abc11942a661c60b835b7b2bc73b89ba09ab59c520ab", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 34244, - "estimated_input_tokens": 8561, - "estimated_eval_tokens": 17378 + "serialized_chars": 34391, + "estimated_input_tokens": 8598, + "estimated_eval_tokens": 17452 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly several small duplicate candidates, not one single tight cluster. The clearest duplicate families are the generic LoRA-loading failures, T2I-Adapter support/pipeline requests, ControlNet from_single_file loading, and the two DPM++ scheduler-name questions.", - "confidence": 0.81, - "canonical_issue_reason": "issue:11224 is the broadest and cleanest umbrella for the LoRA-loading failure family: it is generic, open, and aligns with the more specific report in issue:11062 as well as the exact duplicate-like issue:11209.", + "summary": "This cluster is mostly a grab-bag of model-loading/support issues, with a small number of clear duplicates inside narrower subtopics (LoRA loading, scheduler-name questions, T2I-Adapter support, and ControlNet from_single_file loading).", + "confidence": 0.86, + "canonical_issue_reason": "issue:11209 is the best canonical issue for the LoRA-loading subgroup: it is generic, open, and has a near-identical twin in issue:11224.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11224 is the best overall issue candidate because its title is the most general and reusable for deduplicating related LoRA load failures without being tied to one specific model.", + "best_issue_reason": "issue:11209 is the strongest global issue representative among these items, but only for the LoRA-loading slice of the cluster; the overall set is too mixed to support one broad canonical issue perfectly.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10436", "right": "issue:1925", "accept": false, - "reason": "Both are inpainting-loading complaints, but they involve different models and different failure modes (runwayml/stable-diffusion-inpainting vs SD2 half-precision)." + "reason": "Both mention inpainting, but one is a model-loading failure and the other is a half-precision runtime issue on a different model/version." }, { "left": "issue:10933", "right": "issue:6149", "accept": false, - "reason": "Both mention IP-Adapter load errors, but they are for different model variants and likely different code paths." + "reason": "Both are IP-Adapter load errors, but they target different adapter/model variants and do not look like the same concrete bug." }, { "left": "issue:1260", "right": "issue:6636", "accept": false, - "reason": "Both are import errors, but they concern unrelated symbols and unrelated breakages." + "reason": "Different import symbols and different API breakages; same general category, not the same bug." }, { "left": "issue:1263", "right": "issue:1495", "accept": false, - "reason": "Different missing imports with no clear shared bug or code path." + "reason": "Both are import errors, but for unrelated symbols and modules." }, { "left": "issue:2904", "right": "issue:7228", "accept": false, - "reason": "AudioLDM scheduler failure and traced-model dtype attribute error are unrelated." + "reason": "Unrelated failures: AudioLDM scheduler initialization vs traced StableDiffusionPipeline missing dtype." }, { "left": "issue:4094", "right": "issue:4945", "accept": false, - "reason": "One is a feature request, the other is a test request; not the same underlying change." + "reason": "One asks for a training script, the other asks for tests for training; related area but not the same change." + }, + { + "left": "issue:4728", + "right": "issue:7114", + "accept": false, + "reason": "Both involve OOM during training, but on different scripts and contexts, so not the same underlying issue." }, { "left": "issue:7593", "right": "issue:858", "accept": false, - "reason": "Different subsystems and different errors; no shared concrete bug." + "reason": "Different attribute errors in different code paths; only superficial 'bug' similarity." }, { "left": "issue:11209", "right": "issue:11224", "accept": true, - "reason": "Near-identical generic LoRA loading failure titles; these look like the same underlying bug." + "reason": "Near-identical titles and very high textual overlap; both are generic 'Error in loading Lora' reports." }, { "left": "issue:13429", "right": "issue:13430", "accept": true, - "reason": "Both are the same JoyAI-Image-Edit support request, phrased as add/support." + "reason": "Same JoyAI-Image-Edit support request, phrased as 'add' vs 'support'." }, { "left": "issue:11062", "right": "issue:11209", - "accept": true, - "reason": "Specific LoRA load failure report vs the generic LoRA loading error; same family and likely same bug." + "accept": false, + "reason": "Both are LoRA loading problems, but 11062 is a specific CivitAI LoRA load error and not clearly the same bug as the generic loading failure." }, { "left": "issue:11062", "right": "issue:11224", - "accept": true, - "reason": "Same LoRA loading failure family; the specific CivitAI model report fits the broader generic issue." + "accept": false, + "reason": "Same reason as 11062 vs 11209: similar area, but not enough evidence of the same concrete failure." }, { "left": "issue:9500", "right": "issue:9501", "accept": false, - "reason": "Both are Dreambooth Flux save-related, but one is a checkpoint failure and the other is a long delay; not clearly the same bug." + "reason": "Both concern Flux DreamBooth saving, but one is a checkpoint failure and the other is a long save delay; likely distinct symptoms." }, { "left": "issue:5455", "right": "issue:7513", "accept": false, - "reason": "Both involve training problems, but one is a notebook subprocess failure and the other is a GPU/memory training issue." + "reason": "Both mention DreamBooth/LoRA training problems, but the titles point to different workflows and failure modes." }, { "left": "issue:3728", "right": "issue:3788", "accept": false, - "reason": "Both are new pipeline feature requests, but for different pipelines/models." + "reason": "Separate new-pipeline requests for different models; same broad theme, not the same change." }, { "left": "issue:9161", "right": "issue:9163", "accept": false, - "reason": "Same script, but the issues describe different bug classes (distributed bugs vs tensor bugs) rather than one concrete defect." + "reason": "Same script area, but one is distributed bugs and the other is tensor bugs; too vague to merge as one issue." }, { "left": "issue:2830", "right": "issue:4779", "accept": false, - "reason": "Both reference ModelScope video ideas, but one is a capability question and the other is a pipeline request; not the same change." + "reason": "Both are about video generation on ModelScope, but one is a question and the other a pipeline request; not clearly the same request." }, { "left": "issue:1962", "right": "issue:2984", "accept": false, - "reason": "Generic video-generation interest vs a specific latent video diffusion paper/model request; not the same issue." + "reason": "Different video-generation topics and different model families." }, { "left": "issue:10612", "right": "issue:10989", "accept": false, - "reason": "Both mention quantized Flux + LoRA loading, but they target different pipelines/components and are not clearly the same bug." + "reason": "Both involve LoRA loading with quantization, but one is Flux Fill and the other is Flux ControlNet; related bug class, not the same concrete code path." }, { "left": "issue:5630", "right": "issue:5689", "accept": false, - "reason": "Similar symptom, but different schedulers and different issue scopes; too broad to treat as one bug." + "reason": "Both report residual noise in SDXL, but they refer to different schedulers and scenarios." }, { "left": "issue:2838", "right": "issue:4241", "accept": false, - "reason": "Unrelated ControlNet pipeline behavior question vs unidiffusers LoRA fine-tuning support request." + "reason": "Both are about control/T2I-adapter style support, but one is a ControlNet pipeline question and the other is fine-tuning unidiffusers with LoRA." }, { "left": "issue:2390", "right": "issue:2437", "accept": true, - "reason": "Same T2I-Adapter feature area; one asks for support, the other for the pipeline addition." + "reason": "Both are the same T2I-Adapter feature request, one phrased as support and the other as adding the pipeline." + }, + { + "left": "issue:9497", + "right": "issue:9500", + "accept": false, + "reason": "Both are Flux DreamBooth training issues, but one is a matrix-shape runtime error and the other is checkpoint saving failure." }, { "left": "issue:2635", "right": "issue:2676", "accept": true, - "reason": "Both ask what diffusers scheduler names correspond to DPM++ Karras variants." + "reason": "Same underlying scheduler-name mapping question for DPM++ Karras variants." }, { "left": "issue:11178", "right": "issue:9208", "accept": true, - "reason": "Both are about ControlNet models failing to load with from_single_file(); one is specific, the other general." - }, - { - "left": "issue:9497", - "right": "issue:9500", - "accept": false, - "reason": "Both are Flux Dreambooth training issues, but one is a matmul shape error and the other is a checkpoint-saving problem." + "reason": "Same from_single_file ControlNet loading problem; 11178 is a concrete instance of the broader 9208 issue." }, { - "left": "issue:12499", - "right": "issue:12896", + "left": "issue:3432", + "right": "issue:3453", "accept": false, - "reason": "Same Wan 2.2 I2V area, but the failures are described differently and are not clearly the same concrete bug." + "reason": "Both are precision/runtime errors in examples, but on different ops, dtypes, and scripts." } ] }, "evaluator_result": { "accept": true, - "feedback": "The judgments are conservative and supported by the packet. The summary correctly frames this as several small duplicate/support families rather than one tight cluster, and the accepted soft-edge pairs all have clear title-level similarity. No pair seems overclaimed beyond the evidence provided." + "feedback": "The summary is well grounded in the packet: it correctly treats the cluster as mixed and only elevates clearly duplicated subtopics where the textual overlap is strong. The accepted soft-edge pairs are conservative enough, and the rejected pairs avoid overclaiming sameness from broad topical similarity alone. No PR-pair mergeability issue applies here." }, "evaluator_used": true, "retried": false, @@ -8699,11 +8705,11 @@ "issue:11657", "issue:12047", "issue:12052", - "issue:1214", "issue:12458", "issue:12459", + "issue:12499", + "issue:12896", "issue:13311", - "issue:1925", "issue:1962", "issue:197", "issue:199", @@ -8737,6 +8743,7 @@ "issue:9692" ], "soft_pairs": [ + "issue:12499|issue:12896", "issue:12458|issue:12459", "issue:3284|issue:3363", "issue:11209|issue:11286", @@ -8760,193 +8767,192 @@ "issue:197|issue:199", "issue:9237|issue:9497", "issue:5633|issue:7418", - "issue:11657|issue:13311", - "issue:1214|issue:1925" + "issue:11657|issue:13311" ], - "prepared_review_unit_hash": "a6f9e9840246a0acc5af535d32210fc9ad5e5cfaa348aaaf05510d27217e9d65", + "prepared_review_unit_hash": "89cc845ffdb129dfe49ad785f6410e104a612028193b9f84bdca2ef65539bb8c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 33404, - "estimated_input_tokens": 8351, - "estimated_eval_tokens": 16958 + "serialized_chars": 33415, + "estimated_input_tokens": 8354, + "estimated_eval_tokens": 16964 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The only strong duplicate cluster here is the trio of generic LoRA-loading reports (11209/11224/11286). Most other pairs are broad topical neighbors but describe different models, pipelines, or symptoms, so they should stay separate.", - "confidence": 0.68, - "canonical_issue_reason": "issue:11224 is the best representative of the LoRA-loading error cluster: it has the same generic failure theme as 11209 and 11286, is open, and has the most discussion among the near-identical reports.", + "summary": "This set is mostly heterogeneous: many feature requests and unrelated model-specific bugs. The clearest duplicate pocket is the LoRA-loading reports (11209/11224/11286), with a likely duplicate pair for Flux Dreambooth training (9237/9497).", + "confidence": 0.74, + "canonical_issue_reason": "issue:11224 is the strongest representative of the LoRA-loading duplicate set: it has the same generic failure title as 11209/11286 and the most discussion activity among them.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11224 is the most suitable global issue to keep as the canonical LoRA-loading problem report because it is the most central and actively discussed instance of that duplicate set.", + "best_issue_reason": "issue:11224 is the best consolidation target for the only clearly duplicated issue family in this batch.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12499", + "right": "issue:12896", + "accept": false, + "reason": "Same Wan 2.2 I2V area, but one is a condition-shape mismatch and the other is an MPS-specific device/backend failure; not the same concrete bug." + }, { "left": "issue:12458", "right": "issue:12459", "accept": false, - "reason": "Both mention batch inference, but one is a Qwen-image-edit feature/issue and the other is a FLUX.1-Kontext-dev throughput problem; different concrete use cases and likely different fixes." + "reason": "Both mention batch inference, but they are for different models/pipelines and one is a throughput request while the other is an issue report." }, { "left": "issue:3284", "right": "issue:3363", - "accept": false, - "reason": "Both involve train_dreambooth_lora.py failures, but the reports point to different runtime contexts ('accelerate' vs 'two machines') and no evidence shows the same underlying bug." + "accept": true, + "reason": "Both report train_dreambooth_lora.py failing, with very similar wording and no sign of different code paths; plausible duplicate." }, { "left": "issue:11209", "right": "issue:11286", "accept": true, - "reason": "Same generic LoRA loading failure report with near-identical titles and very similar problem framing; likely duplicate reports of the same bug." + "reason": "Near-identical LoRA loading failure wording in the same subsystem; likely the same underlying bug." }, { "left": "issue:11224", "right": "issue:11286", "accept": true, - "reason": "Same generic LoRA loading failure report with near-identical titles and very similar problem framing; likely duplicate reports of the same bug." + "reason": "Near-identical LoRA loading failure wording in the same subsystem; likely the same underlying bug." }, { "left": "issue:3094", "right": "issue:3451", "accept": false, - "reason": "Both are training/memory related, but one is a CUDA OOM during a finetuning example and the other is fp16 using more memory than fp32; these are different issues." + "reason": "Both are training-memory related, but one is an OOM during finetuning and the other is an fp16-vs-fp32 memory regression; different bugs." }, { "left": "issue:1962", "right": "issue:5889", "accept": false, - "reason": "One is a general video-generation feature request, the other is the Stable Video Diffusion project/topic; they are related in area but not the same issue." + "reason": "One is a generic request for video generation with Stable Diffusion, the other is about Stable Video Diffusion specifically; not the same issue." }, { "left": "issue:4680", "right": "issue:5594", "accept": false, - "reason": "Both mention SDXL fine-tuning, but one is about inpainting fine-tuning and the other is about memory savings with LoRA; different requests." + "reason": "Both touch SDXL fine-tuning, but one asks about inpainting support and the other about LoRA memory usage; different requests." }, { "left": "issue:6584", "right": "issue:6689", "accept": false, - "reason": "Both are conversion-to-Diffusers problems, but for different source models (DreamShaper v8 Inpaint vs SSD-1B); not the same bug." + "reason": "Both are conversion-to-Diffusers problems, but they involve different source models and no shared concrete failure is evident." }, { "left": "issue:9042", "right": "issue:9148", "accept": false, - "reason": "These are separate feature requests: a Flux pipeline and a Flux inpainting pipeline; related, but not the same concrete change." + "reason": "Related Flux feature requests, but one is the base pipeline and the other is inpainting support; not the same change." }, { "left": "issue:12047", "right": "issue:12052", "accept": false, - "reason": "One is a LoRA fusion failure on Wan2.2 GGUF, the other is an offloading device-mismatch runtime error; different code paths and symptoms." + "reason": "Both involve Wan 2.2 and LightX2V, but one is LoRA fusion on GGUF and the other is an offloading device-mismatch bug; different code paths." }, { "left": "issue:3884", "right": "issue:7330", "accept": false, - "reason": "Both reference train_text_to_image_lora.py, but the reported failures are different and the titles do not establish the same underlying defect." + "reason": "Same training script, but the failure descriptions are too generic vs. a specific FP16 unscale error; not enough evidence of the same bug." }, { "left": "issue:4708", "right": "issue:5603", "accept": false, - "reason": "Both are import-related, but one is a generic diffusers.utils import error and the other is a missing specific symbol; insufficient evidence they are the same bug." + "reason": "Both are import errors from diffusers.utils, but the concrete missing symbol/module failure is different and not clearly the same root cause." }, { "left": "issue:4708", "right": "issue:6636", "accept": false, - "reason": "Same as above: both are import problems, but the exact missing names/problems differ and could stem from separate compatibility issues." + "reason": "Both are import errors from diffusers.utils, but they point to different missing names and could be separate regressions." }, { "left": "issue:2462", "right": "issue:2838", "accept": false, - "reason": "One requests IterableDataset support for finetuning, the other asks for a controlnet_pipeline behavior change; unrelated." + "reason": "Different feature requests in different areas; only loosely related by training/pipeline support." }, { "left": "issue:2941", "right": "issue:3061", "accept": false, - "reason": "Both concern Stable Diffusion behavior, but one is a scheduler quality issue and the other is a reproducibility issue; not the same defect." + "reason": "Both concern Stable Diffusion outputs, but one is scheduler-specific undesired results and the other is general reproducibility; not the same bug." }, { "left": "issue:5118", "right": "issue:9551", "accept": false, - "reason": "Both mention ControlNet usage, but one is about reference_only in diffusers and the other about x-labs Flux controlnet models; different models and integrations." + "reason": "Both mention ControlNet, but one is about reference_only usage and the other about Flux ControlNet models; different tasks." }, { "left": "issue:11374", "right": "issue:11569", "accept": false, - "reason": "Separate feature requests for different model families (SkyReels-V2 vs ltxv-13b-0.9.7-distilled); not duplicates." + "reason": "Both are model-support feature requests, but for different models and different integration work." }, { "left": "issue:10144", "right": "issue:6418", "accept": false, - "reason": "Both involve video diffusion, but one is a quality comparison against official Mochi code and the other is stable video diffusion frames collapsing to the first frame; distinct problems." + "reason": "Both are video-quality complaints, but they involve different models and different symptoms; not the same underlying issue." }, { "left": "issue:11062", "right": "issue:11286", "accept": false, - "reason": "Both are LoRA loading issues, but they target different model families and one is a concrete CivitAI LoRA error while the other is a generic loading failure." + "reason": "Both are LoRA-loading problems, but 11062 is specific to a particular CivitAI LoRA while 11286 is a generic loading error; not enough to treat as the same bug." }, { "left": "issue:10343", "right": "issue:9692", "accept": false, - "reason": "Both mention CogView3-Plus-3B, but one is a black-image runtime warning and the other is a startup error; not enough evidence they are the same bug." + "reason": "Same CogView3-Plus family, but one is a black-image/cast warning and the other is a startup error; different failures." }, { "left": "issue:197", "right": "issue:199", "accept": false, - "reason": "Both are dead-code cleanup requests in different modules (ResNet vs Attention), not duplicates." + "reason": "Both are dead-code cleanup items, but they target different modules and are not the same change." }, { "left": "issue:9237", "right": "issue:9497", - "accept": false, - "reason": "Both are about Flux training, but one is a general dreambooth LoRA issue and the other is a specific mat2 dimensionality runtime error; different defects." + "accept": true, + "reason": "Both are Flux Dreambooth/LoRA training problems, and 9497 looks like a concrete runtime manifestation of the broader training issue in 9237." }, { "left": "issue:5633", "right": "issue:7418", "accept": false, - "reason": "Both involve transformer-related bugs, but they are in different files/components and describe different failures." + "reason": "Different files and different bug reports; no evidence they are the same defect." }, { "left": "issue:11657", "right": "issue:13311", "accept": false, - "reason": "Both are LoRA-key-mismatch reports, but they affect different model architectures (WanTransformer3DModel vs Flux2Transformer2DModel) and are not the same code path." - }, - { - "left": "issue:1214", - "right": "issue:1925", - "accept": false, - "reason": "Both concern half precision, but one is CLIPTextModel-specific and the other is a stable-diffusion-2-inpainting pipeline issue; not the same underlying bug." + "reason": "Both mention missing LoRA keys with prefix='transformer', but they affect different model families and likely different loaders." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet, and the duplicate judgments are conservative overall. The canonical LoRA-loading cluster around 11209/11224/11286 is well supported by the near-identical titles and higher discussion on 11224, and the false pairs are mostly justified by distinct models, pipelines, or symptoms." + "feedback": "The summary is grounded in the packet: the LoRA-loading trio is the strongest duplicate pocket, and the 9237/9497 Flux Dreambooth pair is plausibly related. The non-duplicate calls are generally conservative and supported by the differing titles/symptoms." }, "evaluator_used": true, "retried": false, @@ -8978,9 +8984,11 @@ "issue:11824", "issue:11839", "issue:11865", + "issue:1214", "issue:12216", "issue:12645", "issue:12653", + "issue:1925", "issue:2830", "issue:2966", "issue:3019", @@ -8988,8 +8996,6 @@ "issue:3420", "issue:3774", "issue:3885", - "issue:4185", - "issue:4398", "issue:4520", "issue:4708", "issue:4733", @@ -9016,6 +9022,7 @@ "issue:9857" ], "soft_pairs": [ + "issue:1214|issue:1925", "issue:11315|issue:12653", "issue:4708|issue:6794", "issue:7168|issue:7306", @@ -9038,187 +9045,186 @@ "issue:11865|issue:9141", "issue:2830|issue:3019", "issue:10282|issue:11839", - "issue:9497|issue:9548", - "issue:4185|issue:4398" + "issue:9497|issue:9548" ], - "prepared_review_unit_hash": "7f539a42fa98424d88fdbb802043cf22c45e52da1c8c75d5bc3d077d6f9c0108", + "prepared_review_unit_hash": "c8d315b9364c00ee5b1089325b80a4be00eeeeb25158216210e1892d3cc59ae0", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33890, - "estimated_input_tokens": 8473, - "estimated_eval_tokens": 17202 + "serialized_chars": 33858, + "estimated_input_tokens": 8465, + "estimated_eval_tokens": 17186 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly heterogeneous, but there is one clear duplicate pair around the HiDream RoPE float64/MPS regression. Most other soft pairs look like related symptom families, not the same concrete bug.", - "confidence": 0.82, - "canonical_issue_reason": "Issue 11315 is the cleanest canonical anchor: it states the concrete failure (RoPE float64 on MPS) clearly, and 12653 is effectively the same bug with the same root symptom.", + "summary": "This cluster is mostly a grab bag of unrelated issues across different pipelines/models. The only clear duplicate-looking pair is the rope float64 MPS regression. No PRs are present.", + "confidence": 0.71, + "canonical_issue_reason": "issue:9237 is the best representative anchor because it is the broadest FLUX dreambooth/LoRA training thread and has the strongest follow-up activity among the Flux-training issues.", "canonical_pr_reason": null, - "best_issue_reason": "11315 is the most specific and actionable issue in the only strong duplicate pair, making it the best representative for triage.", + "best_issue_reason": "issue:9237 is the most useful global issue to keep as the canonical reference for the recurring Flux dreambooth/LoRA reports; it is broad, central, and has the most discussion/inbound references in that subtopic.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1214", + "right": "issue:1925", + "accept": false, + "reason": "Both mention half-precision, but one is CLIPTextModel-specific and the other is SD2 inpainting; too broad to treat as the same bug." + }, { "left": "issue:11315", "right": "issue:12653", "accept": true, - "reason": "Same RoPE float64 regression causing MPS breakage; 12653 is a near-direct restatement of the same bug on related hardware." + "reason": "Both describe the same rope implementation float64 regression breaking MPS/NPU; same underlying bug and likely one fix." }, { "left": "issue:4708", "right": "issue:6794", "accept": false, - "reason": "Both are import errors, but they point to different missing symbols/modules and do not look like the same underlying defect." + "reason": "Both are import errors, but they point to different symbols/modules and do not clearly share the same root cause." }, { "left": "issue:7168", "right": "issue:7306", "accept": false, - "reason": "Both involve tensor-shape/type complaints, but they are different pipelines and different failure modes." + "reason": "Different problems: IP-Adapter image-embed shape handling versus float16 support in LEditsPPPipelineStableDiffusionXL." }, { "left": "issue:2966", "right": "issue:3026", "accept": false, - "reason": "Same example script family, but one is multi-GPU training and the other is a single-GPU launch error; not the same bug." + "reason": "Same script family, but one is multi-GPU training and the other is a single-GPU accelerate launch failure; not enough evidence of the same concrete bug." }, { "left": "issue:7621", "right": "issue:8359", "accept": false, - "reason": "Different features: one is an InstructPix2Pix/cosxl_edit incompatibility, the other is from_single_file/custom checkpoint loading." + "reason": "Different features: cosxl_edit pipeline compatibility versus from_single_file with custom checkpoints." }, { "left": "issue:4736", "right": "issue:4872", "accept": false, - "reason": "Both mention SDXL finetuning instability/OOM, but one is a specific training-script crash with invalid casts and the other is generic low-VRAM OOM." + "reason": "Both mention SDXL finetuning OOM, but one also has an invalid-cast error and they do not look like the same specific failure." }, { "left": "issue:10749", "right": "issue:10872", "accept": false, - "reason": "Completely different requests: GGUF support for Lumina2 versus from_single_file support for SanaTransformer2DModel." + "reason": "Both are feature requests for new loading support, but they target different models and formats." }, { "left": "issue:7263", "right": "issue:9448", "accept": false, - "reason": "Both are IP-Adapter-related, but the errors and pipelines differ; this is not the same concrete code-path problem." + "reason": "Both involve IP-Adapter, but the failures are different: attention slicing versus a tuple/shape error in ControlNet inpaint." }, { "left": "issue:5226", "right": "issue:5227", "accept": false, - "reason": "One is about load_lora_weights on SDXL AdapterPipeline, the other about load_lora_weight with torch.compile; related area, different bugs." + "reason": "Both concern LoRA loading, but one is AdapterPipeline compatibility and the other is torch.compile interaction; different code paths." }, { "left": "issue:3420", "right": "issue:4520", "accept": false, - "reason": "Different functionality: ControlNet/inpaint pipeline failure versus LoRA scale with attention slicing." + "reason": "Different subsystems and symptoms: ControlNet/Diffusion pipeline mismatch versus LoRA scale with attention slicing." }, { "left": "issue:7513", "right": "issue:9237", "accept": false, - "reason": "Both involve training, but they are different pipelines (SDXL DreamBooth vs Flux DreamBooth LoRA) and different issue reports." + "reason": "Both are Dreambooth/LoRA training reports, but one is SDXL on a 3060 and the other is FLUX training; not the same bug." }, { "left": "issue:10226", "right": "issue:9857", "accept": false, - "reason": "Both are FLUX training shape mismatches, but one is DreamBooth LoRA and the other is ControlNet embedding mismatch; not clearly the same bug." + "reason": "Both are Flux tensor-shape failures, but one is Dreambooth LoRA training and the other is ControlNet embedding mismatch; too different to merge." }, { "left": "issue:11824", "right": "issue:11839", "accept": false, - "reason": "Same pipeline family, but one is multi-image input handling and the other is GGUF invalid input sizing; different problems." + "reason": "Same pipeline family, but one is multi-image input handling and the other is GGUF loading with invalid input size; not clearly the same defect." }, { "left": "issue:10848", "right": "issue:9026", "accept": false, - "reason": "Both are signature-mismatch errors in loading helpers, but they involve different functions and different kwargs." + "reason": "Both are unexpected-kwarg errors, but they occur in different loader functions with different arguments and likely different fixes." }, { "left": "issue:4733", "right": "issue:5757", "accept": false, - "reason": "Both mention checkpoint resume in DreamBooth LoRA scripts, but they are separate reports with no evidence of the same root cause." + "reason": "Both are checkpoint-resume issues, but they are on different training scripts and the reports do not show the same underlying failure." }, { "left": "issue:10869", "right": "issue:10958", "accept": false, - "reason": "Both involve meta tensors, but the exact errors and affected pipelines differ; too broad to treat as the same bug." + "reason": "Both are meta-tensor errors, but one is Tensor.item() and the other is copy-out; similar class of symptom, not the same concrete bug." }, { "left": "issue:7492", "right": "issue:8939", "accept": false, - "reason": "Unrelated loading failures: one is a missing repository, the other is a tokenizer loading error for a different model." + "reason": "Both are loading failures, but one is a missing repository and the other is a tokenizer load error for a different model." }, { "left": "issue:12216", "right": "issue:12645", "accept": false, - "reason": "Both are Qwen-Image-related, but one is quality regression and the other is performance; not the same issue." + "reason": "Same model family, but one is quality comparison and the other is speed comparison; not the same issue." }, { "left": "issue:3774", "right": "issue:3885", "accept": false, - "reason": "Both concern VRAM/ZeRO memory, but one is a specific zero3 training observation and the other is a general memory-reduction question." + "reason": "Both concern VRAM/memory in train_text_to_image.py, but one is zero3 not saving memory and the other is a general VRAM reduction question." }, { "left": "issue:11865", "right": "issue:9141", "accept": false, - "reason": "Both ask for from_single_file support, but for different model classes and different pipelines." + "reason": "Both mention missing from_single_file support, but they target different model classes and different user requests." }, { "left": "issue:2830", "right": "issue:3019", "accept": false, - "reason": "Completely different feature requests with no shared bug/change." + "reason": "Image-to-video/modelscope question versus segment-anything for text-guided editing; unrelated." }, { "left": "issue:10282", "right": "issue:11839", "accept": false, - "reason": "Shape mismatch while loading x_embedder versus invalid input size for GGUF; different failures." + "reason": "Both involve shape/input-size mismatch, but one is a Flux x_embedder weight shape issue and the other is GGUF load input-size validation; different failures." }, { "left": "issue:9497", "right": "issue:9548", "accept": false, - "reason": "Related FLUX DreamBooth/LoRA area, but 9548 is a follow-up to a different issue chain and does not clearly match the mat2/1-D tensor failure." - }, - { - "left": "issue:4185", - "right": "issue:4398", - "accept": false, - "reason": "Both are black-image reports, but one is SDXL ControlNet inference and the other is SDXL LoRA DreamBooth training; different bugs." + "reason": "Both are Flux Dreambooth/LoRA training reports, but one is a mat2 dimensionality error and the other is a follow-up on the earlier Flux issue; not the same concrete bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The assessment is well grounded in the packet and stays conservative. The 11315/12653 pair is clearly the strongest match from the titles/body, while the other soft pairs are appropriately treated as related-symptom or feature-adjacent rather than definite duplicates." + "feedback": "The summary is grounded in the packet: issue:9237 is a reasonable canonical anchor for the Flux dreambooth/LoRA cluster, the rope float64 MPS pair is the only clearly duplicate-looking one, and the note that no PRs are present is supported. The soft-edge verdicts are generally conservative and do not overstate duplicate evidence." }, "evaluator_used": true, "retried": false, @@ -9252,15 +9258,15 @@ "issue:2244", "issue:2250", "issue:2636", - "issue:2883", "issue:3289", "issue:3391", "issue:3728", "issue:3788", "issue:3826", + "issue:4185", "issue:4213", + "issue:4398", "issue:4733", - "issue:4779", "issue:4872", "issue:5005", "issue:5226", @@ -9288,6 +9294,7 @@ "issue:9914" ], "soft_pairs": [ + "issue:4185|issue:4398", "issue:2636|issue:5005", "issue:8723|issue:9012", "issue:6274|issue:6275", @@ -9311,193 +9318,192 @@ "issue:3788|issue:3826", "issue:5412|issue:5840", "issue:4213|issue:5226", - "issue:12009|issue:12047", - "issue:2883|issue:4779" + "issue:12009|issue:12047" ], - "prepared_review_unit_hash": "a4211212ed4f6adec3b0e07b021d947af007803cf3df869fe092745bd94ffeea", + "prepared_review_unit_hash": "4a9321ff7cf7fd748e76f181fc590f2669c771739f70523e759d5abf66c8e845", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 33766, - "estimated_input_tokens": 8442, - "estimated_eval_tokens": 17140 + "serialized_chars": 33845, + "estimated_input_tokens": 8462, + "estimated_eval_tokens": 17180 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous cluster of training, loading, and feature-request issues. The only clear duplicates are a few checkpoint-resume, SD3ControlNet from_transformer, and Stable Cascade resolution pairs; most other soft pairs are just same-subsystem similarity, not the same bug.", - "confidence": 0.74, - "canonical_issue_reason": "If a single issue must anchor this mixed set, #9343 is the best representative because it is a concrete, active FLUX loading bug rather than a feature request or a vague symptom.", + "summary": "This set is mostly a mix of unrelated issues, with a few clear duplicate-style pairs around the same concrete bug (resume-from-checkpoint progress handling, Stable Cascade resolution failures, and some model-loading errors). I did not find a single global canonical issue/PR for the whole cluster.", + "confidence": 0.79, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "#9343 is the strongest global issue candidate: specific, open, and central to the cluster\u2019s recurring model-loading failures.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4185", + "right": "issue:4398", + "accept": false, + "reason": "Both mention black images, but they involve different pipelines and likely different root causes." + }, { "left": "issue:2636", "right": "issue:5005", "accept": true, - "reason": "Both describe the same checkpoint-resume progress bar miscount/truncation behavior in training scripts." + "reason": "Same resume-from-checkpoint progress-bar/skip-step bug in training scripts." }, { "left": "issue:8723", "right": "issue:9012", "accept": true, - "reason": "Same SD3ControlNetModel.from_transformer initialization/load-weights failure." + "reason": "Both target SD3ControlNetModel.from_transformer loading/initialization failures on the same code path." }, { "left": "issue:6274", "right": "issue:6275", "accept": false, - "reason": "Related adapter area, but one is about multiple adapters breaking and the other about unsupported adapter models; not the same concrete bug." + "reason": "Both are T2I-Adapter issues, but one is about multiple adapters and the other about model coverage." }, { "left": "issue:11464", "right": "issue:12459", "accept": false, - "reason": "Different FLUX problems: one is LoRA loading into FluxControlPipeline, the other is batch inference throughput." + "reason": "Different FLUX problems: model loading vs batch inference throughput." }, { "left": "issue:10937", "right": "issue:11864", "accept": false, - "reason": "Both mention torch.compile, but they involve different encode paths and different failures, so not the same bug." + "reason": "Both are torch.compile encode failures, but they affect different models and error modes." }, { "left": "issue:6277", "right": "issue:6363", - "accept": false, - "reason": "One is a vague gradient issue; the other is a specific FP16 unscale error in a training script." + "accept": true, + "reason": "Same train_text_to_image_lora.py gradient/unscale failure reported at different specificity levels." }, { "left": "issue:3728", "right": "issue:9873", "accept": false, - "reason": "Both are pipeline requests, but for different models/features and not the same change." + "reason": "Both are new pipeline requests, but for completely different models." }, { "left": "issue:9343", "right": "issue:9667", "accept": false, - "reason": "Both are FLUX loading issues, but they hit different code paths and failure modes." + "reason": "Different FLUX load-time issues with different flags and failure contexts." }, { "left": "issue:7142", "right": "issue:7432", "accept": false, - "reason": "Both compare Diffusers to WebUI output, but they are different inference-quality complaints and not a duplicate bug." + "reason": "Both compare diffusers to WebUI, but the model/context differ and the reports are too broad to treat as the same bug." }, { "left": "issue:7355", "right": "issue:7644", "accept": true, - "reason": "Both target Stable Cascade failures on non-standard image sizes / arbitrary resolutions." + "reason": "Same Stable Cascade resolution-handling failure with arbitrary / non-power-of-two sizes." }, { "left": "issue:3788", "right": "issue:9873", "accept": false, - "reason": "Different new-pipeline requests for different models." + "reason": "Different new pipeline proposals with no shared concrete bug or feature." }, { "left": "issue:9115", "right": "issue:9187", "accept": false, - "reason": "Different adapters and different failure modes; not the same underlying bug." + "reason": "Different adapter integrations and different failure modes." }, { "left": "issue:2052", "right": "issue:4733", "accept": false, - "reason": "Both are resume-related training issues, but for different scripts and errors." + "reason": "Both involve checkpoint resume, but they are different training scripts and errors." }, { "left": "issue:3289", "right": "issue:3391", "accept": false, - "reason": "Same DeepFloyd SR area, but the reported symptoms and triggers are not clearly the same bug." + "reason": "Same DeepFloyd area, but the bugs are different symptoms and likely different causes." }, { "left": "issue:11630", "right": "issue:11674", "accept": false, - "reason": "Separate feature requests: single-file loading versus extra ref-image/multi-video support." + "reason": "Same Wan VACE family, but one is single-file support and the other is ref-image/multi-control-video support." }, { "left": "issue:2244", "right": "issue:2250", "accept": false, - "reason": "Both are DreamBooth inference bugs, but one is a CPU/half-precision runtime error and the other is a prompt-generation failure." + "reason": "Both are DreamBooth inference issues, but the failures are unrelated." }, { "left": "issue:9343", "right": "issue:9914", "accept": false, - "reason": "Both are FLUX-related loading problems, but one is a mismatched-size loading bug and the other is a LoRA loading issue." + "reason": "Both concern FLUX loading, but one is a mismatched-size/load flag issue and the other is LoRA loading with Xlabs." }, { "left": "issue:7057", "right": "issue:771", "accept": false, - "reason": "Both concern checkpoint/model compatibility, but they are unrelated model/training failures." + "reason": "Both are checkpoint or precision-related, but the actual models and failure paths differ." }, { "left": "issue:6201", "right": "issue:8645", "accept": false, - "reason": "Prompt-encoding bug versus a missing helper function; unrelated issues." + "reason": "Different code paths entirely: prompt embedding handling vs missing single_file_utils helper." }, { "left": "issue:2089", "right": "issue:4872", "accept": false, - "reason": "Both are OOM training reports, but they involve different models and setups." + "reason": "Both are OOM training reports, but on different models and scenarios." }, { "left": "issue:3788", "right": "issue:3826", "accept": false, - "reason": "Different new pipeline requests." + "reason": "Both are new pipeline requests, but they are unrelated models/features." }, { "left": "issue:5412", "right": "issue:5840", "accept": true, - "reason": "Same train_dreambooth_lora_sdxl.py resume-from-checkpoint failure." + "reason": "Same train_dreambooth_lora_sdxl.py checkpoint-resume failure." }, { "left": "issue:4213", "right": "issue:5226", "accept": false, - "reason": "Different feature/compatibility problems: from_single_file without accelerate versus LoRA loading on AdapterPipeline." + "reason": "Different SDXL loading features and different failures." }, { "left": "issue:12009", "right": "issue:12047", "accept": false, - "reason": "Same Wan2.2 GGUF area, but one is model loading from single file and the other is LoRA fusion; different code paths." - }, - { - "left": "issue:2883", - "right": "issue:4779", - "accept": false, - "reason": "Both are image-to-video related, but one is a generic pipeline request and the other is a specific ModelScope pipeline request." + "reason": "Same model family, but one is single-file loading and the other is LoRA fusion on GGUF; not clearly the same bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet and stays conservative on the ambiguous pairs. The accepted soft edges are defensible duplicates or near-duplicates based on the issue titles and described failure modes, and the rejected pairs are appropriately treated as same-subsystem similarity rather than clear duplicates. The canonical issue choice is also reasonable for this mixed cluster." + "feedback": "The cluster summary is grounded in the packet, and the soft-edge decisions are conservative overall. The accepted pairs are plausible duplicates or near-duplicates based on the titles, while the rejected pairs are appropriately cautious and do not overstate identity across different models, scripts, or failure modes. No canonical issue/PR is asserted without support, and the summary does not overclaim a single global bug." }, "evaluator_used": true, "retried": false, @@ -9535,17 +9541,17 @@ "issue:2560", "issue:2772", "issue:2774", + "issue:2883", "issue:2891", "issue:3008", "issue:3094", - "issue:3112", "issue:3184", "issue:3569", "issue:3631", "issue:3639", "issue:4000", "issue:4580", - "issue:5314", + "issue:4779", "issue:5433", "issue:5465", "issue:5577", @@ -9568,6 +9574,7 @@ "issue:9953" ], "soft_pairs": [ + "issue:2883|issue:4779", "issue:10653|issue:9953", "issue:10350|issue:12082", "issue:3569|issue:3639", @@ -9591,196 +9598,195 @@ "issue:10472|issue:10848", "issue:8494|issue:852", "issue:2294|issue:7147", - "issue:6183|issue:7210", - "issue:3112|issue:5314" + "issue:6183|issue:7210" ], - "prepared_review_unit_hash": "b30b36c82858c05d5fa132dcdb183e92720e72d2e897f606125b7f52a862459a", + "prepared_review_unit_hash": "3869d19c7f88ed381f1cd72f0597f44f79c60aa0528ca70d94baf2dcb24dacca", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34789, - "estimated_input_tokens": 8698, - "estimated_eval_tokens": 17652 + "serialized_chars": 34770, + "estimated_input_tokens": 8693, + "estimated_eval_tokens": 17642 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly noisy, but there are two credible duplicate-like issue pairs: `train_controlnet.py` local data loading (`2772`/`7267`) and `lpw_stable_diffusion` generator handling (`3008`/`3184`). The rest are broad-similarity or clearly distinct bugs/features.", - "confidence": 0.8, - "canonical_issue_reason": "issue:2772 is the best issue anchor because it gives the clearest, most concrete description of the `train_controlnet.py` local dataset loading failure. It reads like the most substantive report in that thread, while issue:7267 is also similar but not clearly a subordinate follow-up from the packet alone.", + "summary": "This cluster is mostly heterogeneous. Two pairs look like true duplicates or near-duplicates around the same code path; the rest are only loosely similar by topic and should stay separate.", + "confidence": 0.95, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:2772 is the strongest representative issue because it targets a specific code path (`train_controlnet.py` local data loading) and has the most actionable bug framing. The cluster also contains another likely duplicate-like pair (`3008`/`3184`), so `2772` is the best anchor rather than the only duplicate-like thread.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2883", + "right": "issue:4779", + "accept": false, + "reason": "Both mention image-to-video, but one is a generic pipeline request and the other is a ModelScope-specific pipeline item; not the same bug/change." + }, { "left": "issue:10653", "right": "issue:9953", "accept": false, - "reason": "Both are quantized/model device-movement complaints, but one is about `.to(cpu)` after 8bit load and the other about `.to(cuda)` on a quantized component; similar symptom, different bug path." + "reason": "Both involve quantized modules and device moves, but one is 8-bit to CPU and the other is moving a quantized component to CUDA; different failure paths." }, { "left": "issue:10350", "right": "issue:12082", "accept": false, - "reason": "Both mention high VAE memory use, but they concern different models and scenarios (`SD2.1` decoder vs `wan2.1` VAE after compile)." + "reason": "Both mention VAE memory, but they target different models and triggers (SD2.1 decoder vs Wan2.1 after compile)." }, { "left": "issue:3569", "right": "issue:3639", "accept": false, - "reason": "These are separate new-pipeline requests (`Uni-ControlNet` vs `GlyphControl`), not the same underlying change." + "reason": "These are separate new-pipeline requests for different features (Uni-ControlNet vs GlyphControl)." }, { "left": "issue:5433", "right": "issue:6295", "accept": false, - "reason": "Both discuss scheduler artifacts, but the concrete settings and failure modes differ enough that they do not look like one fixable bug." + "reason": "Both are about DPM++ artifacts, but the reported conditions differ enough (missing refiner vs use_lu_lambdas) that they do not clearly describe one concrete bug." }, { "left": "issue:2772", "right": "issue:7267", "accept": true, - "reason": "Both report `train_controlnet.py` failing on valid local dataset loading, and the second is a much shorter restatement of the same underlying bug." + "reason": "Both describe the same train_controlnet.py local-dataset/local-loading problem; the titles point to the same underlying script bug." }, { "left": "issue:5577", "right": "issue:6183", "accept": false, - "reason": "Both are conversion/loading issues, but one is a `ControlNetModel.from_single_file` KeyError and the other is an SDXL shape mismatch; different failures." + "reason": "Both involve from_single_file, but they hit different model types and different errors (ControlNet key error vs SDXL shape issue)." }, { "left": "issue:9501", "right": "issue:9727", "accept": false, - "reason": "Both involve FLUX DreamBooth saving, but one is a delayed save and the other is a multigpu save problem; not clearly the same code-path bug." + "reason": "Both are DreamBooth/FLUX save problems, but one is delayed saving and the other is a multigpu save failure; not clearly the same bug." }, { "left": "issue:2089", "right": "issue:3094", "accept": false, - "reason": "Both are OOM training reports, but they are for different scripts and contexts; the shared symptom is too broad to treat as a duplicate." + "reason": "Both are OOM reports during training, but they involve different examples and contexts; too broad to be one bug." }, { "left": "issue:1495", "right": "issue:7147", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols from different modules." + "reason": "Both are import errors, but for different symbols and different releases; not the same issue." }, { "left": "issue:3008", "right": "issue:3184", "accept": true, - "reason": "Both point to `lpw_stable_diffusion` mishandling generator inputs on the same pipeline/code path, so they look like the same bug." + "reason": "Both are about lpw_stable_diffusion handling generator lists/invalid torch.randn arguments, which looks like the same concrete pipeline bug." }, { "left": "issue:12195", "right": "issue:7185", "accept": false, - "reason": "`torch.library.custom_op` missing in PyTorch 2.3.1 is a specific compatibility break, while the other issue is a broad question about PyTorch 2.2.0 support." + "reason": "One is a specific AutoencoderKL failure on PyTorch 2.3.1; the other is a general question about PyTorch 2.2.0 support." }, { "left": "issue:5661", "right": "issue:5917", "accept": false, - "reason": "These are different import/version failures in different parts of the library." + "reason": "Both are upgrade-related errors, but they concern unrelated missing names in different code paths." }, { "left": "issue:2774", "right": "issue:4000", "accept": false, - "reason": "Both are conversion-related, but one is ControlNet safetensors conversion and the other is Stable Diffusion 2.0 ckpt conversion; too broad to merge." + "reason": "Both are conversion problems, but one is ControlNet safetensors conversion and the other is SD 2.0 ckpt conversion; different tools and likely different bugs." }, { "left": "issue:11948", "right": "issue:7209", "accept": false, - "reason": "Both involve offline loading, but one is `from_pretrained` on WanTransformer3DModel and the other is `from_single_file` on a local safetensor; different API paths." + "reason": "Both concern offline loading, but one is from_pretrained for WanTransformer3DModel and the other is from_single_file with a local safetensor; different loading paths." }, { "left": "issue:12458", "right": "issue:12645", "accept": false, - "reason": "Batch inference/feature-request discussion is not the same as a slowdown comparison against ComfyUI." + "reason": "One is a batch inference issue/feature request and the other is a speed comparison complaint; not the same bug." }, { "left": "issue:3631", "right": "issue:4580", "accept": false, - "reason": "Different pipelines and failures: latents in inpainting vs `use_karras_sigmas` in UniPC scheduling." + "reason": "Different pipelines and errors: inpainting latents exception vs UniPC schedule/use_karras_sigmas exception." }, { "left": "issue:5846", "right": "issue:6073", "accept": false, - "reason": "A specific `resume_from_checkpoint` error in SDXL LoRA training is unrelated to a beginner help/setup request." + "reason": "Both are beginner DreamBooth support reports, but they do not identify the same concrete failure." }, { "left": "issue:2392", "right": "issue:2560", "accept": false, - "reason": "One is a prompt_embeds validation bug; the other is differing outputs with generator lists. They do not share the same failure mode." + "reason": "Both touch StableDiffusionImg2ImgPipeline, but one is a prompt_embeds validation error and the other is differing outputs with a generator list." }, { "left": "issue:2459", "right": "issue:771", "accept": false, - "reason": "Different training-time compatibility issues (`xformers` enablement vs BF16 support) with no shared bug path." + "reason": "Unrelated training/setup failures: xformers enablement vs BF16 support." }, { "left": "issue:2891", "right": "issue:5465", "accept": false, - "reason": "Docs/support mismatch for embeds in an upscale pipeline is unrelated to DPM++ artifact issues." + "reason": "Both mention unsupported options/docs mismatches, but they are different features (embeds vs inpaint_strength) and not one bug." }, { "left": "issue:10472", "right": "issue:10848", "accept": false, - "reason": "One is a LoRA `lora_bias` constructor argument error; the other is a `load_model_dict_into_meta` `device` keyword error. Different APIs and failures." + "reason": "Both are unexpected-keyword errors, but for different arguments and loading functions; too generic to merge." }, { "left": "issue:8494", "right": "issue:852", "accept": false, - "reason": "Both are unexpected keyword-argument errors, but for different options (`force_hooks` vs `device_map`) and different versions/platforms." + "reason": "Different unexpected keyword arguments in different versions/code paths; not the same defect." }, { "left": "issue:2294", "right": "issue:7147", "accept": false, - "reason": "Different missing imports (`StableDiffusionVideoInpaintPipeline` vs `StableCascadeDecoderPipeline`) and not the same underlying issue." + "reason": "Both are import errors, but they reference different missing pipeline symbols and different eras." }, { "left": "issue:6183", "right": "issue:7210", "accept": false, - "reason": "`from_single_file` shape mismatch and `named_parameters` missing during accelerate dispatch are separate pipeline/model integration bugs." - }, - { - "left": "issue:3112", - "right": "issue:5314", - "accept": false, - "reason": "Both are ControlNet-related, but they are different pipeline features rather than the same change." + "reason": "One is a from_single_file shape issue and the other is an accelerate dispatch attribute error; unrelated failures." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The core anchor choice is reasonable, but the reasoning is a bit too strong for the packet. In particular, calling issue:2772 the cluster\u2019s \"only clear duplicate-like thread\" overstates the evidence because the packet also supports another accepted duplicate-like pair (issue:3008 / issue:3184). Also, describing issue:7267 as a \"follow-up\" is not directly grounded in the provided data. The rest of the soft-edge calls look appropriately conservative." + "accept": true, + "feedback": "The evaluation is grounded in the packet and stays conservative. The two accepted pairs are reasonably supported as likely duplicates/near-duplicates, and the remaining pairs are appropriately rejected for being only loosely related or clearly different failure modes. The cluster summary also matches the evidence: mostly heterogeneous with only two plausible duplicate pairs." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -9814,6 +9820,7 @@ "issue:2163", "issue:2371", "issue:2480", + "issue:3112", "issue:3221", "issue:3450", "issue:3694", @@ -9826,12 +9833,12 @@ "issue:5025", "issue:5128", "issue:5226", + "issue:5314", "issue:5442", "issue:5633", "issue:579", "issue:5931", "issue:6574", - "issue:6791", "issue:692", "issue:7223", "issue:7239", @@ -9839,7 +9846,6 @@ "issue:7483", "issue:7708", "issue:7871", - "issue:8659", "issue:9343", "issue:9525", "issue:9803", @@ -9848,6 +9854,7 @@ "issue:9971" ], "soft_pairs": [ + "issue:3112|issue:5314", "issue:12195|issue:12666", "issue:1013|issue:1538", "issue:2371|issue:2480", @@ -9870,187 +9877,186 @@ "issue:9343|issue:9525", "issue:1614|issue:993", "issue:13277|issue:13319", - "issue:2163|issue:3694", - "issue:6791|issue:8659" + "issue:2163|issue:3694" ], - "prepared_review_unit_hash": "288343a31a244270d462d2ea8d4d26be16ea93cc94abac4086d205d6b96b2006", + "prepared_review_unit_hash": "964a7151bb94f18a804812e8aaa39c3bb98644817efdda03b912b21b9665ed7b", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34034, - "estimated_input_tokens": 8509, - "estimated_eval_tokens": 17274 + "serialized_chars": 33985, + "estimated_input_tokens": 8497, + "estimated_eval_tokens": 17250 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous and not a true duplicate set. The only likely duplicate pair is the SDXL LoRA conversion-script reports (4616/7483); the rest look like distinct bugs, feature requests, or unrelated failure modes.", - "confidence": 0.58, - "canonical_issue_reason": "Issue 7483 is the clearest, most specific report of the SDXL LoRA conversion-script failure, and it matches the only plausible duplicate pair in this set.", + "summary": "Most items are unrelated diffusers bugs/feature requests; only one soft pair looks like a true duplicate around the SDXL LoRA-to-Diffusers conversion script. The rest are too broad, different code paths, or only share a subsystem name.", + "confidence": 0.84, + "canonical_issue_reason": "issue:7483 is the better canonical issue for the accepted duplicate pair because it is more specific about the failing SDXL 1.0 base-model conversion path, while issue:4616 is broader and less explicit.", "canonical_pr_reason": null, - "best_issue_reason": "7483 is the best duplicate target here because it describes a concrete, actionable SDXL LoRA conversion problem with the most precise scope.", + "best_issue_reason": "issue:7483 is the strongest representative issue in this set: it clearly states the broken script and the SDXL 1.0 base-model condition, making it easier to anchor discussion and deduping.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3112", + "right": "issue:5314", + "accept": false, + "reason": "Both concern ControlNet/inpainting, but one is a multi-controlnet request and the other is reference inpainting; different change requests, not the same bug." + }, { "left": "issue:12195", "right": "issue:12666", "accept": false, - "reason": "Different problems: PyTorch custom_op incompatibility vs a missing pipeline feature." + "reason": "Completely different problems: an AutoencoderKL custom_op compatibility error versus a missing Kandinsky video pipeline release feature." }, { "left": "issue:1013", "right": "issue:1538", "accept": false, - "reason": "Both mention StableDiffusionMegaPipeline, but one is a version incompatibility and the other is a model-specific failure." + "reason": "Same pipeline family, but the failures are different: one is a version-compatibility issue and the other is model-specific compatibility with stable-diffusion-2-base." }, { "left": "issue:2371", "right": "issue:2480", "accept": false, - "reason": "Dreambooth-related, but the symptoms differ: missing outputs vs checkpoint directory structure." + "reason": "Different Dreambooth output/checkpoint problems; one is about generated objects not appearing, the other about checkpoint directory layout." }, { "left": "issue:4708", "right": "issue:5025", "accept": false, - "reason": "Both are import issues, but one is a generic diffusers.utils import error and the other is a specific randn_tensor import failure." + "reason": "Both are import-related, but one is a generic diffusers.utils import error and the other is specifically missing randn_tensor; not enough evidence they share the same root cause." }, { "left": "issue:9803", "right": "issue:9914", "accept": false, - "reason": "Both involve Flux LoRA loading, but they appear to be different code paths and failure modes." + "reason": "Both mention Flux LoRA loading, but the titles point to different failure modes and Xlabs-specific loading behavior; too broad to treat as the same concrete bug." }, { "left": "issue:1181", "right": "issue:692", "accept": false, - "reason": "Training from scratch is not the same as asking for a fine-tuning example." + "reason": "A training-from-scratch issue and a request for a fine-tuning example are not the same underlying bug or change." }, { "left": "issue:4480", "right": "issue:5931", "accept": false, - "reason": "One is an MPS fp16 kernel crash; the other is nondeterministic generation." + "reason": "Different symptoms and likely different root causes: MPS kernel crash with fp16 versus nondeterministic generation." }, { "left": "issue:10994", "right": "issue:11045", "accept": false, - "reason": "Same script family, but batch-size crash and model-loading crash are different bugs." + "reason": "Both involve train_dreambooth_lora_flux, but one crashes with batch size > 1 during training while the other crashes when loading Flux Schnell 1; not the same code-path problem." }, { "left": "issue:4566", "right": "issue:7239", "accept": false, - "reason": "Checkpoint resume failure vs SDXL LoRA shape mismatch are distinct issues." + "reason": "Checkpoint resume failure and SDXL LoRA shape-mismatch error are distinct issues." }, { "left": "issue:11127", "right": "issue:9971", "accept": false, - "reason": "Different components: QwenImage/Ulysses batch-input failure vs Flux scheduler behavior." + "reason": "Both are Flux-related, but one is a QwenImage single-file loading issue and the other is a scheduler issue; different concrete bugs." }, { "left": "issue:5442", "right": "issue:7243", "accept": false, - "reason": "Different pipelines/features; only a broad diffusion-model similarity." + "reason": "These are separate model/pipeline feature requests (ScaleCrafter vs ResAdapter), not duplicates." }, { "left": "issue:4724", "right": "issue:6574", "accept": false, - "reason": "Img2Img checkpoint behavior and SVD tensor warning are unrelated." + "reason": "An SDXL img2img correctness issue and an SVD pipeline warning are unrelated." }, { "left": "issue:3221", "right": "issue:3450", "accept": false, - "reason": "Both are loading-related, but they target different APIs and failures." + "reason": "Different loading APIs and different failure contexts; only the general theme of pipeline loading overlaps." }, { "left": "issue:579", "right": "issue:7708", "accept": false, - "reason": "Both mention config.json, but one is textual inversion and the other is an SDXL model packaging issue." + "reason": "Both mention missing config.json, but they target different loading workflows and model types; not clearly the same bug." }, { "left": "issue:4616", "right": "issue:7483", "accept": true, - "reason": "Same underlying bug family: the LoRA-to-Diffusers conversion script failing for SDXL base models." + "reason": "Both describe the SDXL LoRA-to-Diffusers conversion script failing for SDXL-based models; same concrete bug/change with closely matching scope." }, { "left": "issue:5128", "right": "issue:7871", "accept": false, - "reason": "Single-file loading error vs blockwise-scale LoRA error are different problems." + "reason": "A single-file loading error and a blockwise-scale LoRA issue are different problems." }, { "left": "issue:1214", "right": "issue:5226", "accept": false, - "reason": "Half-precision CLIPTextModel issue is unrelated to LoRA loading in StableDiffusionXLAdapterPipeline." + "reason": "Half-precision CLIPTextModel behavior and SDXL adapter LoRA loading are unrelated." }, { "left": "issue:1962", "right": "issue:7223", "accept": false, - "reason": "One is a generic video-generation request; the other is a specific Latte feature proposal." + "reason": "A generic video generation question and a request to add Latte are not duplicates." }, { "left": "issue:4210", "right": "issue:5633", "accept": false, - "reason": "UNet attention_mask usage question is unrelated to a transformer_temporal.py bug." + "reason": "UNet attention_mask usage and a transformer_temporal bug are different code paths and different bugs." }, { "left": "issue:9343", "right": "issue:9525", "accept": false, - "reason": "Both are Flux-related, but model-loading flags and lora_scale behavior are different code paths." + "reason": "Both are Flux-related, but one is a loading/size mismatch error and the other is lora_scale not taking effect; different defects." }, { "left": "issue:1614", "right": "issue:993", "accept": false, - "reason": "Both involve precision, but one is black outputs and the other is a dtype mismatch error." + "reason": "Both are precision-related, but black-image output on autocast and generic fp16 dtype mismatch are not the same concrete failure." }, { "left": "issue:13277", "right": "issue:13319", "accept": false, - "reason": "Same model family, but forward batch-input failure and backward-pass failure are not the same bug." + "reason": "Same Ulysses SP topic, but one fails with batch inputs in the pipeline while the other fails in backward pass; different bugs." }, { "left": "issue:2163", "right": "issue:3694", "accept": false, - "reason": "Training script dtype mismatch vs missing model files are different setup/runtime problems." - }, - { - "left": "issue:6791", - "right": "issue:8659", - "accept": false, - "reason": "Same example script, but a general run failure is not the same as resume/inference-loading problems." + "reason": "UNet dtype mismatch during training and missing model/config files in the example are unrelated failures." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative overall. The accepted 4616/7483 pair is plausibly the same SDXL LoRA conversion-script bug from the titles alone, and the other soft-edge rejections are reasonable given the differing failure modes or feature requests." + "feedback": "The analyst output is grounded in the packet and stays conservative on the soft-edge pairs. The one accepted duplicate pair (issue:4616 / issue:7483) is plausibly the same SDXL LoRA conversion-script bug, and the rest are reasonably rejected as different failure modes, feature requests, or only loosely related subsystem overlaps. The summary does not overstate evidence." }, "evaluator_used": true, "retried": false, @@ -10099,6 +10105,7 @@ "issue:6415", "issue:6548", "issue:6741", + "issue:6791", "issue:6914", "issue:7057", "issue:7185", @@ -10107,10 +10114,9 @@ "issue:7589", "issue:8236", "issue:8307", - "issue:8534", "issue:861", "issue:8613", - "issue:8708", + "issue:8659", "issue:8839", "issue:8909", "issue:9141", @@ -10121,6 +10127,7 @@ "issue:9857" ], "soft_pairs": [ + "issue:6791|issue:8659", "issue:5963|issue:6028", "issue:4593|issue:4697", "issue:6383|issue:7185", @@ -10145,199 +10152,198 @@ "issue:3325|issue:3884", "issue:11417|issue:4492", "issue:2459|issue:7057", - "issue:8307|issue:9141", - "issue:8534|issue:8708" + "issue:8307|issue:9141" ], - "prepared_review_unit_hash": "966490c542651aeb8a3b5f57448c152f626c680d23b0f2b86f5d3631e1ade51f", + "prepared_review_unit_hash": "935735cde9dff9cc7f164243105317ba0189630ce8f92a638882fd9f74a87d28", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 34795, - "estimated_input_tokens": 8699, - "estimated_eval_tokens": 17654 + "serialized_chars": 34775, + "estimated_input_tokens": 8694, + "estimated_eval_tokens": 17644 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are only broadly similar (same model family or same kind of error) and should not be merged. The only clear duplicate-like match is the multi-GPU LR scheduler step-count bug: #6137 and #8236.", - "confidence": 0.93, - "canonical_issue_reason": "#6137 is the broader, more general report of the underlying scheduler step-count bug in multi-GPU training, so it is the best canonical issue for the duplicate pair.", + "summary": "The cluster is mostly heterogeneous: many unrelated bug reports across different models, training scripts, and loading paths. Only one clear duplicate pair stands out: 6137 and 8236, both about incorrect LR scheduler step-count calculation in multi-GPU training.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 6137 is the best canonical issue because 8236 is a near-duplicate of the same multi-GPU LR scheduler step-count bug.", "canonical_pr_reason": null, - "best_issue_reason": "#6137 best represents the shared issue because it states the core bug directly, while #8236 is a later, more specific manifestation tied to `--num_train_epochs` in examples.", + "best_issue_reason": "6137 is the strongest representative of the only real duplicate thread in this set; the rest of the items are too diverse to share one underlying fix.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6791", + "right": "issue:8659", + "accept": false, + "reason": "Both mention SDXL training, but one is a generic can't-run issue and the other is checkpoint resume/load behavior; different failures and code paths." + }, { "left": "issue:5963", "right": "issue:6028", "accept": false, - "reason": "Both are Kandinsky 3.0, but one is about embeds-vs-prompts behavior and the other is CUDA OOM; different bugs." + "reason": "Same model family, but one is failing when passing embeds instead of prompts and the other is a CUDA OOM; not the same bug." }, { "left": "issue:4593", "right": "issue:4697", "accept": false, - "reason": "Both concern LoRA loading, but one is a rank mismatch with an A1111 LoRA and the other is a specific LoRA failing after an update; not the same concrete failure." + "reason": "Both are LoRA-loading problems, but they affect different scenarios and error conditions; not clearly the same underlying defect." }, { "left": "issue:6383", "right": "issue:7185", "accept": false, - "reason": "Multi-stage learning rates vs PyTorch 2.2 compatibility are unrelated topics." + "reason": "A scheduler feature question and a PyTorch version-compatibility question are not duplicate bug reports." }, { "left": "issue:6176", "right": "issue:8839", "accept": false, - "reason": "Both mention LoRA weights, but one is post-training DreamBooth LoRA loading and the other is an SD3 LoRA weight error; different code paths." + "reason": "Different LoRA issues on different models/scripts; one is loading weights after DreamBooth training, the other is an SD3 LoRA weight error." }, { "left": "issue:7513", "right": "issue:9548", "accept": false, - "reason": "Different models and symptoms: SDXL DreamBooth resource issues vs FLUX DreamBooth LoRA training issues." + "reason": "Both involve FLUX/LoRA training, but the reported problems and scripts differ too much to be the same bug." }, { "left": "issue:3094", "right": "issue:4872", "accept": false, - "reason": "Both are OOM reports, but they affect different training scripts/models and likely different causes." + "reason": "Both are OOM reports, but they concern different examples and model setups; likely separate memory issues." }, { "left": "issue:1214", "right": "issue:7589", "accept": false, - "reason": "Half-precision CLIPTextModel failure is unrelated to GLIGEN not working." + "reason": "Half-precision CLIPTextModel failure is unrelated to GLIGEN layout-to-image behavior." }, { "left": "issue:9799", "right": "issue:9857", "accept": false, - "reason": "Both are tensor-size mismatch errors, but the model, script, and failure context differ; too generic to treat as the same bug." + "reason": "Both are tensor-size mismatch errors, but the titles point to different FLUX training paths and likely different root causes." }, { "left": "issue:10671", "right": "issue:9450", "accept": false, - "reason": "One is a deterministic dtype/device lookup issue in DiffusionPipeline; the other is a multi-GPU device placement problem in FluxPipeline." + "reason": "Both mention device/dtype issues, but one is DiffusionPipeline dtype/device access and the other is a multi-GPU tensor device mismatch; not the same bug." }, { "left": "issue:7057", "right": "issue:861", "accept": false, - "reason": "A checkpoint loading failure is not the same as a ckpt-to-diffusers conversion problem." + "reason": "A broken checkpoint load is not the same as a conversion-from-ckpt problem." }, { "left": "issue:2132", "right": "issue:6741", "accept": false, - "reason": "Different script-specific errors: missing `trackers` on Accelerator vs missing `no_flip` on Namespace." + "reason": "Different missing-attribute errors in different scripts: Accelerator.trackers vs Namespace.no_flip." }, { "left": "issue:3706", "right": "issue:7405", "accept": false, - "reason": "They are both model announcements, not the same bug or change." + "reason": "These are model announcement/new-model issues, not duplicates of the same bug or change." }, { "left": "issue:4389", "right": "issue:9343", "accept": false, - "reason": "ControlNet loading failure vs FLUX loading with `low_cpu_mem_usage=False` and `ignore_mismatched_sizes=True` are different problems." + "reason": "Both are model-loading reports, but ControlNet loading and FLUX low_cpu_mem_usage/ignore_mismatched_sizes failures are distinct." }, { "left": "issue:4697", "right": "issue:6176", "accept": false, - "reason": "Both involve LoRA-related loading, but they describe different workflows and failure modes." + "reason": "Both mention loading LoRA weights, but they are different scenarios with different symptoms and no clear shared code-path bug." }, { "left": "issue:6137", "right": "issue:8236", "accept": true, - "reason": "These describe the same underlying LR scheduler step-count bug under multi-GPU training; #8236 is a later, more specific report of the same issue." + "reason": "These describe the same bug: LR scheduler step-count is computed incorrectly in multi-GPU training, with 8236 giving a specific num_train_epochs case of 6137." }, { "left": "issue:10722", "right": "issue:9857", "accept": false, - "reason": "Both are tensor mismatch errors, but they occur in different FLUX training scenarios with different shapes and contexts." + "reason": "Both are shape-mismatch reports, but they affect different training setups and do not clearly share one root cause." }, { "left": "issue:6415", "right": "issue:6914", "accept": false, - "reason": "One is about unsupported IP-Adapter-FaceID/FaceID plus, the other is an AttributeError from `enable_model_cpu_offload()`." + "reason": "Both involve IP-Adapter-FaceID, but one is about unsupported features and the other is an offload-related AttributeError; not the same defect." }, { "left": "issue:8613", "right": "issue:8909", "accept": false, - "reason": "OOM while using `caption_column` is unrelated to problems saving `text_encoder_ti` in a DreamBooth LoRA script." + "reason": "OOM while using caption_column is unrelated to problems saving text_encoder_ti in a DreamBooth LoRA script." }, { "left": "issue:2941", "right": "issue:6548", "accept": false, - "reason": "Scheduler quality issue vs img2img RGBA latent misclassification are unrelated." + "reason": "Scheduler output quality and RGBA latent handling are different pipeline bugs." }, { "left": "issue:10591", "right": "issue:11314", "accept": false, - "reason": "Both are training-script complaints, but they target different models and different bugs." + "reason": "Different training scripts, different models, and different reported mistakes; not a duplicate bug." }, { "left": "issue:10169", "right": "issue:1976", "accept": false, - "reason": "Missing `config.json` is too generic here; #1976 is specifically about an incorrect `tokenizer/config.json` lookup, not the same reported failure." + "reason": "Both mention config.json, but 1976 is specifically about tokenizer/config.json lookup; 10169 is too generic to confirm the same bug." }, { "left": "issue:3325", "right": "issue:3884", "accept": false, - "reason": "CUDA invalid argument in DreamBooth is not the same as an error using `train_text_to_image_lora.py`." + "reason": "Different scripts and errors: DreamBooth CUDA invalid argument vs text-to-image LoRA failure." }, { "left": "issue:11417", "right": "issue:4492", "accept": false, - "reason": "DistributedDataParallel missing `dtype` and a `false` NameError are unrelated." + "reason": "An AttributeError on DistributedDataParallel.dtype is unrelated to a lowercase boolean NameError." }, { "left": "issue:2459", "right": "issue:7057", "accept": false, - "reason": "xFormers attention not working in a LoRA training script is unrelated to a checkpoint loading issue." + "reason": "xFormers attention not working in a training script is unrelated to a checkpoint load failure." }, { "left": "issue:8307", "right": "issue:9141", "accept": false, - "reason": "Missing `StableCascadeUNet` and missing `FluxPipeline.from_single_file()` are different API gaps." - }, - { - "left": "issue:8534", - "right": "issue:8708", - "accept": false, - "reason": "Both are SD3 training-script issues, but one is logit-normal weighting and the other is a missing `vae.config.shift_factor`; different concrete bugs." + "reason": "Missing StableCascadeUNet API and missing FluxPipeline.from_single_file are different missing-symbol issues." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative on the soft edges. The #6137/#8236 match is well supported by the titles, and the other pairs are reasonably rejected as different bugs rather than over-merged on superficial similarity." + "feedback": "Grounded and suitably conservative. The summary stays within the packet evidence, and the only affirmative soft-edge pair (6137/8236) is well-supported by the closely matching LR-scheduler/multi-GPU titles. The other pairs are rejected cautiously rather than overstated." }, "evaluator_used": true, "retried": false, @@ -10375,9 +10381,7 @@ "issue:2089", "issue:2299", "issue:2857", - "issue:3032", "issue:3325", - "issue:3497", "issue:3706", "issue:3764", "issue:4000", @@ -10390,7 +10394,6 @@ "issue:4736", "issue:4969", "issue:5578", - "issue:5919", "issue:5934", "issue:6444", "issue:6794", @@ -10401,13 +10404,16 @@ "issue:7854", "issue:8414", "issue:8500", + "issue:8534", "issue:8557", "issue:8614", + "issue:8708", "issue:8833", "issue:8939", "issue:9421" ], "soft_pairs": [ + "issue:8534|issue:8708", "issue:6444|issue:7728", "issue:8500|issue:8557", "issue:1398|issue:1552", @@ -10430,198 +10436,187 @@ "issue:1398|issue:1614", "issue:4566|issue:7133", "issue:12657|issue:13249", - "issue:4229|issue:7757", - "issue:3032|issue:3497", - "issue:3497|issue:5919" + "issue:4229|issue:7757" ], - "prepared_review_unit_hash": "00021578414f3cc04c550e40b5eef149bd3d779357e1a0397e9cc8b3e2055362", + "prepared_review_unit_hash": "df4e2576b5d9e3761534940afdb53d42c41937e94e7a389001aa04b7a4f13a9d", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 34988, - "estimated_input_tokens": 8747, - "estimated_eval_tokens": 17750 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 34171, + "estimated_input_tokens": 8543, + "estimated_eval_tokens": 17342 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous set of support/issues around model loading, training, precision, and quantization. Only the SD3 token-limit pair looks like a true duplicate; the rest are related only at a broad subsystem level and should stay separate.", - "confidence": 0.88, - "canonical_issue_reason": "issue:8557 is the clearest, most specific representative of the only true duplicate pair (SD3 token limit in training).", + "summary": "This cluster is a heterogeneous mix of unrelated issues (feature requests, model-support questions, file-loading/import errors, and distinct training/runtime bugs). I found no convincing duplicate pairs, and there are no PRs in the cluster.", + "confidence": 0.95, + "canonical_issue_reason": "No canonical issue: the items span multiple unrelated problem types and code paths, so collapsing them into one issue would be incorrect.", "canonical_pr_reason": null, - "best_issue_reason": "issue:8557 is the best global issue candidate because it is the most precise statement of the shared SD3 training token-limit bug.", + "best_issue_reason": "issue:10381 is the most detailed, diagnosis-rich report, so it is the best representative anchor if one must pick an issue.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:8534", + "right": "issue:8708", + "accept": false, + "reason": "Both mention SD3 training, but one is about wrong logit-normal weighting and the other about a missing `shift_factor`; different bugs." + }, { "left": "issue:6444", "right": "issue:7728", "accept": false, - "reason": "Both are SDXL LoRA/accelerate training failures, but the reported errors differ (peft_config/checkpointing vs accelerate error) and do not look like one concrete bug." + "reason": "Both involve SDXL LoRA training, but one is a `peft_config` checkpointing issue and the other is an Accelerate runtime error; not the same concrete failure." }, { "left": "issue:8500", "right": "issue:8557", - "accept": true, - "reason": "Same underlying SD3 training token-limit problem; titles and scope match closely enough to treat as duplicates." + "accept": false, + "reason": "Both mention token limits, but one is a general question and the other is a training-specific SD3 report; not enough to treat as one bug." }, { "left": "issue:1398", "right": "issue:1552", "accept": false, - "reason": "Both mention Stable Diffusion 2 on Apple/MPS-related settings, but one is a platform compatibility issue and the other is attention slicing behavior; not the same bug." + "reason": "Different problems: Apple Silicon/MPS support for SD2 versus `enable_attention_slicing()` incompatibility." }, { "left": "issue:3706", "right": "issue:8414", "accept": false, - "reason": "Both are model-related announcements/requests, but Wuerstchen and ConsistencyTTA are different models and changes." + "reason": "Both are model-related announcements, but Wuerstchen and ConsistencyTTA are different models/changes." }, { "left": "issue:10381", "right": "issue:10798", "accept": false, - "reason": "Both involve Flux quantization, but one is inference with quantization+LoRA/Quanto and the other is a bitsandbytes device-combination bug; too different to merge." + "reason": "Both are Flux quantization/LoRA reports, but they involve different backends and symptoms (`Quanto`/Zero GPU spaces vs `bitsandbytes` device-combination bug)." }, { "left": "issue:1166", "right": "issue:1475", "accept": false, - "reason": "Different training scenarios and failures; one is TPU textual inversion saving hang, the other is mixed class-image DreamBooth/textual inversion training." + "reason": "Textual inversion appears in both, but one is a TPU save hang and the other is a multi-class DreamBooth/textual inversion training question." }, { "left": "issue:3325", "right": "issue:4701", "accept": false, - "reason": "Different runtime errors in different paths (CUDA invalid argument during DreamBooth vs AutoencoderKL CUDA error when differentiating)." + "reason": "Different failures: DreamBooth CUDA invalid argument versus AutoencoderKL differentiation CUDA error." }, { "left": "issue:10128", "right": "issue:2857", "accept": false, - "reason": "Both are feature-support questions, but fastercache and UniDiffuser are unrelated requests." + "reason": "These are unrelated feature requests for different models/capabilities (`fastercache` vs UniDiffuser)." }, { "left": "issue:11396", "right": "issue:4453", "accept": false, - "reason": "Both ask about LoRA conversion, but they target different ecosystems and formats (ComfyUI vs AUTO1111) and are not the same change." + "reason": "Both ask about LoRA conversion, but they target different source models and output ecosystems (hidream/comfyui vs SDXL/AUTO1111)." }, { "left": "issue:4648", "right": "issue:5578", "accept": false, - "reason": "Both are missing-file loading errors, but they reference different expected files/configs and different model-loading failures." + "reason": "Both are missing-file load errors, but they refer to different missing files and different loading contexts." }, { "left": "issue:4708", "right": "issue:4969", "accept": false, - "reason": "Import errors in the same library area, but the missing symbols/modules differ and do not point to one specific defect." + "reason": "Both are import errors, but they involve different missing symbols/modules and likely different root causes." }, { "left": "issue:3764", "right": "issue:5934", "accept": false, - "reason": "Both involve torch.compile, but one is an interpreter-shutdown backend failure and the other is a general compile error; not the same bug." + "reason": "Both mention `torch.compile`, but the reported failures are different and not clearly the same code-path bug." }, { "left": "issue:10928", "right": "issue:8939", "accept": false, - "reason": "Both are model-loading issues, but one is a missing fp16 weight file and the other is a tokenizer loading failure for SD3; different root causes." + "reason": "Different loading issues: missing `diffusion_pytorch_model.fp16.bin` versus tokenizer loading failure for SD3 medium." }, { "left": "issue:6794", "right": "issue:7681", "accept": false, - "reason": "Both are import-time failures, but they come from unrelated modules and error causes." + "reason": "Both are import-time errors, but the missing symbol and dependency failure are unrelated." }, { "left": "issue:8614", "right": "issue:8833", "accept": false, - "reason": "Both complain about config.json missing, but they involve different models and loading contexts; too broad to unify." + "reason": "Both mention `config.json`, but they concern different artifacts/models and are not the same loading bug." }, { "left": "issue:11117", "right": "issue:7854", "accept": false, - "reason": "Both are dtype mismatch runtime errors, but the exact mismatch and affected pipeline differ." + "reason": "Both are dtype mismatch runtime errors, but they affect different code paths and error conditions." }, { "left": "issue:2299", "right": "issue:4000", "accept": false, - "reason": "One asks about fine-tuned checkpoints in Diffusers; the other asks about converting SD2 ckpt to Diffusers format. Related topic, not same bug/change." + "reason": "One asks about fine-tuned checkpoints in Diffusers; the other is about converting an SD2 checkpoint to Diffusers format. Related topic, not the same issue." }, { "left": "issue:1976", "right": "issue:9421", "accept": false, - "reason": "Both are tokenizer/path loading errors, but they concern different model identifiers and likely different missing artifacts." + "reason": "Both are model access/load errors, but they refer to different models and different missing-path situations." }, { "left": "issue:2089", "right": "issue:4736", "accept": false, - "reason": "Both are memory-related training issues, but they affect different scripts/models and one includes an invalid-value cast error; not the same defect." + "reason": "Both mention training and OOM, but one is DreamBooth on AltDiffusion and the other is SDXL LoRA with an additional invalid-value symptom." }, { "left": "issue:1398", "right": "issue:1614", "accept": false, - "reason": "Different Stable Diffusion 2/MPS vs autocast-black-image failures; no clear shared underlying bug." + "reason": "Different SD2/MPS versus autocast black-image behavior; not the same bug." }, { "left": "issue:4566", "right": "issue:7133", "accept": false, - "reason": "Both involve resuming SDXL LoRA training from checkpoints, but the evidence is insufficient to say they are the same concrete failure." + "reason": "Both are checkpoint-resume problems for SDXL LoRA, but one is a generic resume failure and the other is a text-encoder size mismatch; too different to merge." }, { "left": "issue:12657", "right": "issue:13249", "accept": false, - "reason": "Different model-specific regressions: quantized Qwen Image Edit vs Z-Image LoRA support." + "reason": "Different models and failure reports: Qwen Image Edit quantization versus Z-Image LoRA support." }, { "left": "issue:4229", "right": "issue:7757", "accept": false, - "reason": "Both are multi-GPU training problems, but one is an accelerated DreamBooth error and the other is ControlNet OOM; not one bug." - }, - { - "left": "issue:3032", - "right": "issue:3497", - "accept": false, - "reason": "Both are missing pipeline methods, but they are different methods/classes and separate API issues." - }, - { - "left": "issue:3497", - "right": "issue:5919", - "accept": false, - "reason": "Different API breakages: missing from_ckpt on a pipeline vs missing set_lora_layer on Linear." + "reason": "Both are multi-GPU training issues, but they involve different scripts and different error classes (accelerated training error vs OOM)." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The only accepted duplicate pair, issue:8500 / issue:8557, is well supported by the SD3 token-limit titles, and the rest of the soft-edge verdicts appropriately avoid overclaiming duplication from broad topical similarity alone." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -10643,14 +10638,15 @@ "issue:10760", "issue:11417", "issue:11839", - "issue:12009", "issue:1214", "issue:1925", "issue:2091", + "issue:3032", "issue:3061", "issue:3325", "issue:3382", "issue:3454", + "issue:3497", "issue:3774", "issue:4213", "issue:4325", @@ -10663,6 +10659,7 @@ "issue:5811", "issue:5884", "issue:5897", + "issue:5919", "issue:6096", "issue:6131", "issue:6159", @@ -10678,16 +10675,15 @@ "issue:7681", "issue:7735", "issue:7848", - "issue:8871", - "issue:9208", "issue:9497", - "issue:9548", "issue:964", "issue:9643", "issue:9727", "issue:9799" ], "soft_pairs": [ + "issue:3032|issue:3497", + "issue:3497|issue:5919", "issue:4325|issue:6492", "issue:5811|issue:5884", "issue:3325|issue:7619", @@ -10710,194 +10706,192 @@ "issue:9497|issue:9799", "issue:5455|issue:5897", "issue:10760|issue:7681", - "issue:7418|issue:7848", - "issue:8871|issue:9548", - "issue:12009|issue:9208" + "issue:7418|issue:7848" ], - "prepared_review_unit_hash": "bc55ddf53b56ff5edb778c7a748964f2ec886125a6cee8c286b441907ea0cb0f", + "prepared_review_unit_hash": "766a327c71b41c64442ec4085daf13dcfeb6314509edb5a2fd6149b59050b7b6", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 35013, - "estimated_input_tokens": 8754, - "estimated_eval_tokens": 17764 + "serialized_chars": 34384, + "estimated_input_tokens": 8596, + "estimated_eval_tokens": 17448 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a mixed set of superficially similar issues. The only likely duplicate pair is the two black-image generation reports; the rest are distinct bugs, model-loading failures, or training/config errors that just share broad subsystem keywords.", - "confidence": 0.61, - "canonical_issue_reason": "issue:6492 is the clearest canonical target for the black-image symptom: it is broader and more general than issue:4325, which includes a more specific SDXL warning but the same visible failure.", + "summary": "This cluster is mostly heterogeneous: many unrelated training/inference/import/runtime bug reports with only one clear duplicate-like pair around `caption_column` validation. Most soft matches are superficial (same pipeline, same framework, or same broad error class) rather than the same underlying bug.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 6821 is the clearest, older report of the `caption_column` validation failure; 7735 appears to be the same bug phrased differently.", "canonical_pr_reason": null, - "best_issue_reason": "issue:6492 is the best representative issue overall because it matches the only strong duplicate pair and is phrased as the general user-facing symptom.", + "best_issue_reason": "6821 is the strongest representative issue in the only near-duplicate pair and has the most explicit, concise failure description.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3032", + "right": "issue:3497", + "accept": false, + "reason": "Both are attribute errors on different pipeline APIs (`from_pretrained` vs `from_ckpt`); not the same bug." + }, + { + "left": "issue:3497", + "right": "issue:5919", + "accept": false, + "reason": "Different missing attributes (`from_ckpt` vs `set_lora_layer`) and different code paths." + }, { "left": "issue:4325", "right": "issue:6492", - "accept": true, - "reason": "Both report black images during generation; 4325 adds a warning, but the core symptom and likely fix area are the same." + "accept": false, + "reason": "Both mention black images, but one is an SDXL runtime warning/cast issue and the other is a broad symptom report; too vague to treat as the same bug." }, { "left": "issue:5811", "right": "issue:5884", "accept": false, - "reason": "One is about callback_on_step_end support, the other about IP-Adapter support; different features and different code paths." + "reason": "One is about callback arguments, the other about IP-Adapter support; distinct feature changes." }, { "left": "issue:3325", "right": "issue:7619", "accept": false, - "reason": "Both are Dreambooth-related, but one is a CUDA invalid-argument crash and the other is a channel-shape error; not the same bug." + "reason": "Both involve `train_dreambooth.py`, but the reported failures are different (`CUDA invalid argument` vs channel mismatch)." }, { "left": "issue:1925", "right": "issue:964", "accept": false, - "reason": "Both mention inpainting, but 1925 is specifically half-precision failure while 964 is a general inpainting example failure." + "reason": "Half-precision inpainting vs generic inpainting failure; not clearly the same issue." }, { "left": "issue:6226", "right": "issue:6446", "accept": false, - "reason": "These describe unrelated loading problems in different contexts; shared diffusion-model loading is too broad to merge." + "reason": "Both are loading problems, but the targets and failure modes differ substantially." }, { "left": "issue:11417", "right": "issue:6351", "accept": false, - "reason": "Both involve DDP wrapper attribute access, but one is missing dtype and the other is missing disable_adapters; different concrete failures." + "reason": "Both are DDP attribute errors, but they involve different missing attributes and likely different fixes." }, { "left": "issue:5314", "right": "issue:5474", "accept": false, - "reason": "A new pipeline request and a different reference-only pipeline proposal are not the same bug or change." + "reason": "Related ControlNet/reference themes, but one is a pipeline request and the other is a different reference-only concept." }, { "left": "issue:3454", "right": "issue:5502", "accept": false, - "reason": "Different training scripts and different failures; one is a latest-version runtime issue, the other is fp16 support in custom diffusion training." + "reason": "Different training scripts and different compatibility problems; not the same bug." }, { "left": "issue:5465", "right": "issue:6159", "accept": false, - "reason": "Both are parameter-support gaps, but for different pipelines and different arguments; not one underlying defect." + "reason": "Different pipelines and different unsupported parameters (`inpaint_strength` vs `denoising_end`)." }, { "left": "issue:1214", "right": "issue:4213", "accept": false, - "reason": "Half-precision CLIPTextModel failure and missing-accelerate from_single_file loading are unrelated." + "reason": "Half-precision CLIP bug vs single-file SDXL loading without accelerate; unrelated." }, { "left": "issue:6131", "right": "issue:6561", "accept": false, - "reason": "One is an IP-Adapter feature not working, the other is a missing config file for a model repo; different root causes." + "reason": "One is an IP-Adapter support failure, the other is a missing model file/config error." }, { "left": "issue:2091", "right": "issue:3382", "accept": false, - "reason": "Checkpoint resume/multi-GPU behavior and multi-GPU OOM are different training issues." + "reason": "Both concern multi-GPU training, but one is checkpoint resume and the other is OOM behavior." }, { "left": "issue:10453", "right": "issue:9643", "accept": false, - "reason": "Both are model capability limitations, but batch-size >1 for Hunyuan Video is unrelated to multiple ControlNets in Flux." + "reason": "Different model families and different limitations (`batch size > 1` vs multiple ControlNets)." }, { "left": "issue:4375", "right": "issue:6096", "accept": false, - "reason": "Both involve torch.compile crashes, but the model paths and exceptions differ; too broad to treat as one bug." + "reason": "Both involve `torch.compile`, but the model-specific failure modes differ; not enough to call them the same bug." }, { "left": "issue:10650", "right": "issue:11839", "accept": false, - "reason": "Both are invalid-size errors, but they affect different models/pipelines and likely different shape assumptions." + "reason": "Both are shape/input-size errors, but on different models and different loading paths." }, { "left": "issue:2091", "right": "issue:3774", "accept": false, - "reason": "Checkpoint resumption and ZeRO3 memory savings are separate training concerns, not the same bug." + "reason": "Different multi-GPU training issues: checkpoint resume vs zero3 memory savings." }, { "left": "issue:9497", "right": "issue:9727", "accept": false, - "reason": "Both are Flux dreambooth issues, but one is a matmul shape error and the other is a multigpu save problem." + "reason": "Both are Flux Dreambooth reports, but the errors are different (`mat2` shape vs save problem on multigpu)." }, { "left": "issue:3061", "right": "issue:6548", "accept": false, - "reason": "A reproducibility complaint is not the same as an RGBA latents-handling bug in img2img." + "reason": "Reproducibility issue vs RGBA/image-as-latents bug; not the same underlying defect." }, { "left": "issue:6821", "right": "issue:7735", - "accept": false, - "reason": "Both mention caption_column, but one is about an invalid allowed value and the other about a missing column; different validation failures." + "accept": true, + "reason": "Near-identical `caption_column` validation failure in the training CLI; same underlying bug and likely duplicate report." }, { "left": "issue:9497", "right": "issue:9799", "accept": false, - "reason": "Different runtime shape mismatches in different Flux-related contexts; no clear shared code-path." + "reason": "Both are Flux training errors, but the tensor-shape failures are different." }, { "left": "issue:5455", "right": "issue:5897", "accept": false, - "reason": "A notebook subprocess failure and a device-mismatch LoRA training error are unrelated." + "reason": "Both are LoRA training failures, but they are different scripts and different error types." }, { "left": "issue:10760", "right": "issue:7681", "accept": false, - "reason": "Importing a removed class and an OpenSSL/X509 import failure are unrelated import-time errors." + "reason": "An import/export attribute bug vs an environment/OpenSSL import failure; unrelated." }, { "left": "issue:7418", "right": "issue:7848", "accept": false, - "reason": "These are different community pipelines with different decoding/implementation issues." - }, - { - "left": "issue:8871", - "right": "issue:9548", - "accept": false, - "reason": "An fp16 gradient unscale failure and a Flux Dreambooth multigpu save issue are distinct training bugs." - }, - { - "left": "issue:12009", - "right": "issue:9208", - "accept": false, - "reason": "Both are from_single_file loading failures, but they concern different models and formats; not the same concrete bug." + "reason": "Same community example file family, but one is a code bug and the other is a decoding issue; not clearly the same bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The analysis is grounded in the packet and stays conservative overall. The only accepted soft pair, issue:4325 \u2194 issue:6492, is supported by the shared black-image symptom, and the rest of the rejections avoid overclaiming duplicate status from broad subsystem similarity. The summary matches the evidence without overstating it." + "feedback": "The summary is well grounded in the packet: it correctly identifies the only clear duplicate-like pair (6821/7735) and treats the rest conservatively as superficially similar rather than the same bug. The soft-edge rejections are cautious and supported by the titles/body metadata. No overstatement noted." }, "evaluator_used": true, "retried": false, @@ -10921,6 +10915,7 @@ "issue:1082", "issue:11247", "issue:11410", + "issue:12009", "issue:12587", "issue:1260", "issue:12699", @@ -10934,10 +10929,8 @@ "issue:2437", "issue:2578", "issue:281", - "issue:3061", "issue:3152", "issue:4213", - "issue:4483", "issue:4566", "issue:4911", "issue:4969", @@ -10958,15 +10951,19 @@ "issue:7147", "issue:725", "issue:7757", - "issue:8686", + "issue:8871", "issue:8939", "issue:8956", "issue:9051", + "issue:9208", "issue:9527", + "issue:9548", "issue:9667", "issue:9841" ], "soft_pairs": [ + "issue:8871|issue:9548", + "issue:12009|issue:9208", "issue:1672|issue:2299", "issue:3152|issue:4969", "issue:1260|issue:7147", @@ -10988,189 +10985,190 @@ "issue:11410|issue:9667", "issue:281|issue:725", "issue:8939|issue:8956", - "issue:13073|issue:9051", - "issue:3061|issue:4483", - "issue:6964|issue:8686" + "issue:13073|issue:9051" ], - "prepared_review_unit_hash": "fd99d15f6d3508c1c6e6c5af9e00f5202ed36bd812da6a18df2b0419fb824159", + "prepared_review_unit_hash": "13c5d6ef4dbf97229d641e2f2f6b6d9660dc32031f0ca176be8da301e28ffce2", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33950, - "estimated_input_tokens": 8488, - "estimated_eval_tokens": 17232 + "serialized_chars": 34565, + "estimated_input_tokens": 8642, + "estimated_eval_tokens": 17540 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly thematically related diffusers issues, but the proposed soft pairs do not describe the same concrete bug/change. They span different models, pipelines, training scripts, and failure modes, so none should be merged as duplicates from the available evidence.", - "confidence": 0.81, - "canonical_issue_reason": "Issue #2331 is the broadest, most central ControlNet feature request in the set and has the strongest discussion/inbound-reference signal, making it the best representative issue if one must be chosen.", + "summary": "This cluster is mostly a loose set of unrelated training/loading bugs. Only the Accelerate/model-loading feature requests (281 and 725) look like the same underlying change; the rest are too contextually different to merge as duplicates.", + "confidence": 0.63, + "canonical_issue_reason": "Issue 725 is the clearest concrete version of the shared model-loading optimization request and overlaps directly with issue 281\u2019s ask to natively use Accelerate to reduce memory during model loading.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #2331 is the most suitable global issue representative because it anchors the ControlNet pipeline area that several other issues touch, while still being a distinct, well-scoped canonical discussion.", + "best_issue_reason": "Issue 725 is the strongest canonical candidate because it is specific, actionable, and the only near-duplicate pair in the set clearly points to the same feature area.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:8871", + "right": "issue:9548", + "accept": false, + "reason": "Both mention LoRA/Flux training, but the reported failures and scopes differ; this is not the same concrete bug." + }, + { + "left": "issue:12009", + "right": "issue:9208", + "accept": false, + "reason": "Both are from_single_file loading failures, but they affect different model families and fail for different reasons." + }, { "left": "issue:1672", "right": "issue:2299", "accept": false, - "reason": "Both concern converted Stable Diffusion checkpoints, but one is about output image size after ckpt-to-diffusers conversion and the other is about loading fast-stable-diffusion fine-tuned checkpoints; different failure modes." + "reason": "Both relate to checkpoint conversion/loading, but one is about wrong output image size and the other is about fast-stable-diffusion checkpoint support." }, { "left": "issue:3152", "right": "issue:4969", "accept": false, - "reason": "Both are import errors, but they involve unrelated TensorFlow and diffusers symbols/modules, so they are not the same bug." + "reason": "Generic import errors, but they refer to different missing symbols and unrelated code paths." }, { "left": "issue:1260", "right": "issue:7147", "accept": false, - "reason": "Both are import-name problems, but they target different missing pipeline classes and different releases/features." + "reason": "Both are import-name failures, but they concern different pipeline classes and do not indicate the same bug." }, { "left": "issue:12699", "right": "issue:12720", "accept": false, - "reason": "One is about video quality degrading after repeated prompt calls, the other about quantizing Wan 2.2 VACE after loading LoRA; different behavior and code path." + "reason": "One is a video quality regression over repeated prompts, the other is a quantization-after-LoRA question; different problems." }, { "left": "issue:4911", "right": "issue:5128", "accept": false, - "reason": "Both involve from_single_file loading, but one is a custom text encoder loading problem and the other is a local safetensors model loading issue." + "reason": "Both involve from_single_file, but they concern different model types and different missing/invalid loading data." }, { "left": "issue:2578", "right": "issue:4213", "accept": false, - "reason": "Editable install failure vs. SDXL from_single_file requiring accelerate; related packaging/loading area, but not the same underlying defect." + "reason": "Editable install trouble is unrelated to SDXL from_single_file needing accelerate." }, { "left": "issue:6353", "right": "issue:7146", "accept": false, - "reason": "Both mention SDXL training and errors, but one is a controlnet training example failure while the other is a CUDA OOM during text-to-image fine-tuning." + "reason": "Training-example error vs CUDA OOM; same broad area, but not the same underlying bug." }, { "left": "issue:6154", "right": "issue:6967", "accept": false, - "reason": "Both are SDXL training issues, but one is on M2 Mac LoRA training and the other is a prior-preservation shape mismatch in an advanced DreamBooth LoRA script." + "reason": "Different training scripts and different failures; not mergeable as one defect." }, { "left": "issue:6415", "right": "issue:6561", "accept": false, - "reason": "Both are about IP-Adapter-related support/loading, but one is FaceID feature support and the other is a missing image_encoder config file." + "reason": "One is about feature/model support, the other about a missing repository file during loading." }, { "left": "issue:11247", "right": "issue:9527", "accept": false, - "reason": "Both mention multi-GPU/controlnet/Flux-style distribution, but one is about distributing FluxTransformer2DModel with controlnet and the other is a dtype mismatch between controlnet and main pipeline." + "reason": "Multi-GPU distribution problem vs dtype mismatch; related to ControlNet/FLUX but not the same code-path bug." }, { "left": "issue:2331", "right": "issue:2437", "accept": false, - "reason": "Different feature requests: ControlNet pipeline versus T2I-Adapter pipeline. Same ecosystem, not the same change." + "reason": "Separate feature requests for different pipelines; same subsystem, different changes." }, { "left": "issue:222", "right": "issue:684", "accept": false, - "reason": "Training runtime error on generic setup vs. AMD-specific memory optimization failure; different environments and causes." + "reason": "Broad training runtime failures, but one is general and the other is AMD-specific memory optimization behavior." }, { "left": "issue:1082", "right": "issue:5897", "accept": false, - "reason": "Both are training runtime failures, but one is a CUDA cublas execution error during DreamBooth and the other is a device-mismatch error in LoRA training." + "reason": "Different LoRA/DreamBooth runtime errors with distinct failure modes and code paths." }, { "left": "issue:12587", "right": "issue:6203", "accept": false, - "reason": "One is a question about context_pre_only usage, the other is about SDXL ControlNet image-size conditioning; not the same bug." + "reason": "These are different questions about different ControlNet conditions." }, { "left": "issue:4566", "right": "issue:9841", "accept": false, - "reason": "Both are LoRA training issues, but one is resume-from-checkpoint and the other is a TypeError when use_dora is False; different code paths." + "reason": "Checkpoint resume issue vs use_dora TypeError; both in SDXL LoRA training, but not the same bug." }, { "left": "issue:1976", "right": "issue:579", "accept": false, - "reason": "Both are missing-config/file lookup errors, but they concern different components and assets (tokenizer config vs textual inversion config)." + "reason": "Both are config-file lookup errors, but they are for different missing files and different loading flows." }, { "left": "issue:6146", "right": "issue:7757", "accept": false, - "reason": "Both relate to multi-GPU training, but they are different scripts and problems: DreamBooth LoRA multi-GPU vs ControlNet SDXL OOM." + "reason": "Both concern multi-GPU training, but one is DreamBooth LoRA and the other is ControlNet SDXL OOM; not the same defect." }, { "left": "issue:6218", "right": "issue:6964", "accept": false, - "reason": "Both involve ControlNet pipelines, but one is batched seeds behavior and the other is 4-channel UNet support; not the same defect." + "reason": "Both involve ControlNet pipelines, but the failures are different and not the same underlying bug." }, { "left": "issue:11410", "right": "issue:9667", "accept": false, - "reason": "Both reference attention/transformer code, but one is an unpacking error in attention_processor and the other is loading a flux fp8 model with a local transformer file." + "reason": "Completely different failures: attention tensor unpacking vs loading a flux fp8 model." }, { "left": "issue:281", "right": "issue:725", - "accept": false, - "reason": "Both concern memory/loading optimizations, but one asks to natively use Accelerate and the other is about adding device map and accelerate to DiffusionPipeline; overlapping area, not duplicate." + "accept": true, + "reason": "Same underlying feature request: integrate Accelerate/device_map into DiffusionPipeline loading to reduce memory use; 725 is a more concrete restatement of 281." }, { "left": "issue:8939", "right": "issue:8956", "accept": false, - "reason": "Both are model-loading errors for diffusion models, but they affect different model families and tokenizer/model asset layouts." + "reason": "Both are model-loading issues, but they involve different pipelines/models and different missing tokenizer/model assets." }, { "left": "issue:13073", "right": "issue:9051", "accept": false, - "reason": "Both are pipeline capability limitations, but one is ZImageControlNet guidance_scale behavior and the other is FluxPipeline batch inference support." - }, - { - "left": "issue:3061", - "right": "issue:4483", - "accept": false, - "reason": "Both are reproducibility-related, but one is general Stable Diffusion pipeline reproducibility and the other is SDXL generator determinism; similar class of issue, not the same bug." - }, - { - "left": "issue:6964", - "right": "issue:8686", - "accept": false, - "reason": "Both are ControlNet pipeline issues, but one is 4-channel UNet compatibility and the other is SDXLCFGCutoffCallback support." + "reason": "Different pipeline behavior problems: guidance-scale support in ZImageControlNetPipeline vs batch inference in FluxPipeline." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "Grounded in the packet and appropriately conservative: the 281/725 pair is the only clearly same-bug/feature match, and the other soft-edge rejections do not overclaim equivalence." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -11196,14 +11194,13 @@ "issue:1305", "issue:13079", "issue:1486", - "issue:2185", "issue:2210", "issue:2774", - "issue:2875", "issue:2904", "issue:2941", "issue:2947", "issue:3019", + "issue:3061", "issue:3451", "issue:3565", "issue:4137", @@ -11215,6 +11212,7 @@ "issue:6224", "issue:6280", "issue:6481", + "issue:6964", "issue:7142", "issue:7146", "issue:7178", @@ -11227,6 +11225,7 @@ "issue:7782", "issue:8504", "issue:852", + "issue:8686", "issue:9026", "issue:9141", "issue:9270", @@ -11236,6 +11235,8 @@ "issue:9911" ], "soft_pairs": [ + "issue:3061|issue:4483", + "issue:6964|issue:8686", "issue:2904|issue:2947", "issue:10920|issue:1486", "issue:5510|issue:6082", @@ -11258,181 +11259,186 @@ "issue:6280|issue:9141", "issue:10026|issue:9911", "issue:6481|issue:730", - "issue:3451|issue:3565", - "issue:2185|issue:2875" + "issue:3451|issue:3565" ], - "prepared_review_unit_hash": "304da58e4e207a97290f3551cbc301a3d6f995dab02f2163b60d681c98e6b1e0", + "prepared_review_unit_hash": "634f678bdf46a0b81863211b16965048c72ad2a00f506b35a3584cd8041a6230", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 33599, - "estimated_input_tokens": 8400, - "estimated_eval_tokens": 17056 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34469, + "estimated_input_tokens": 8618, + "estimated_eval_tokens": 17492 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issues with superficial title similarity only; none of the soft pairs look like true duplicates or the same fix path.", + "summary": "The cluster is heterogeneous: the soft pairs share only broad topical similarity (e.g., memory, LoRA, scheduler, Flux) but not the same concrete bug or fix path. No duplicate pair is strong enough to merge, so there is no clear canonical issue in this set.", "confidence": 0.95, - "canonical_issue_reason": "No single canonical issue stands out: the cluster is heterogeneous (training, schedulers, conversion, memory, LoRA, and pipeline bugs) and the soft pairs are only broad topical overlaps.", + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 9497 is the most concrete and well-scoped Flux training failure in the set, but it still should not be treated as the umbrella for the cluster.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3061", + "right": "issue:4483", + "accept": false, + "reason": "Both mention nondeterminism, but one is a generic Stable Diffusion reproducibility complaint and the other is an SDXL generator determinism issue; different code paths and failure modes." + }, + { + "left": "issue:6964", + "right": "issue:8686", + "accept": false, + "reason": "One is about 4-channel UNet support in inpaint pipelines; the other is a callback incompatibility with ControlNet pipelines. Related subsystem, different bug." + }, { "left": "issue:2904", "right": "issue:2947", "accept": false, - "reason": "Different failures in different components: AudioLDM pipeline scheduler NoneType vs UNet attention-proc attribute missing." + "reason": "AudioLDM scheduler AttributeError and missing load_attn_procs on UNet are unrelated API/runtime failures." }, { "left": "issue:10920", "right": "issue:1486", "accept": false, - "reason": "Both involve sample-mode scheduler math, but one is DDIMInverseScheduler first-iteration infs and the other is DDIMScheduler behavior; not the same bug." + "reason": "Both involve DDIM-like schedulers, but one is DDIMInverseScheduler inf on first iteration and the other is DDIMScheduler with prediction_type='sample'; different bug locations." }, { "left": "issue:5510", "right": "issue:6082", "accept": false, - "reason": "Both are dtype mismatch runtime errors, but in different pipelines and with no evidence of the same code path or fix." + "reason": "Both are dtype mismatch errors, but they occur in different pipelines with different concrete errors and likely different fixes." }, { "left": "issue:2941", "right": "issue:4483", "accept": false, - "reason": "One is EulerAncestralDiscreteScheduler output quality, the other SDXL generator nondeterminism; different symptoms and causes." + "reason": "EulerAncestralDiscreteScheduler incorrect results is not the same as generator non-determinism; different symptoms and likely different causes." }, { "left": "issue:6224", "right": "issue:7146", "accept": false, - "reason": "TorchDynamo fine-tuning issue vs SDXL CUDA OOM during training; same broad area, not the same defect." + "reason": "TorchDynamo fine-tuning failure and SDXL training OOM are separate training issues." }, { "left": "issue:10350", "right": "issue:10569", "accept": false, - "reason": "Both mention memory use, but for different models/components: SD2.1 VAE decoder vs HunyuanVideo on CPU." + "reason": "Both are memory complaints, but one is VAE decoder memory in SD2.1 and the other is CPU memory usage in HunyuanVideo; different components." }, { "left": "issue:6481", "right": "issue:761", "accept": false, - "reason": "DreamBooth batch-size failure and 8-bit Adam failure are distinct training problems with different triggers." + "reason": "DreamBooth batch-size failure and 8-bit Adam failure are distinct training/optimizer problems." }, { "left": "issue:7185", "right": "issue:9810", "accept": false, - "reason": "PyTorch version support question vs LoRA module_to_save API gap; unrelated concerns." + "reason": "PyTorch version support question and LoRA module_to_save() support are unrelated issues." }, { "left": "issue:4213", "right": "issue:7178", "accept": false, - "reason": "Missing from_single_file support for an SDXL pipeline vs export_to_video malfunction; unrelated code paths." + "reason": "from_single_file() dependency issue and export_to_video malfunction are unrelated pipeline problems." }, { "left": "issue:9270", "right": "issue:9514", "accept": false, - "reason": "Both concern Flux/LoRA, but one is SimpleTuner training support and the other is loading weights after fp8 quantization." + "reason": "Both are Flux/LoRA-related, but one is incomplete support in SimpleTuner and the other is loading LoRA after fp8 quantization; not the same bug." }, { "left": "issue:1305", "right": "issue:3019", "accept": false, - "reason": "Completely different feature requests/papers: Direct Inversion vs segment-anything text-guided editing." + "reason": "Direct Inversion and SAM-based text-guided editing are different feature requests/proposals." }, { "left": "issue:13079", "right": "issue:7629", "accept": false, - "reason": "Both are OOM reports, but on different models and scenarios: FLUX.2-klein loop vs SVD on old PyTorch." + "reason": "Both mention OOM, but one is FLUX.2-klein in a loop and the other is SVD on PyTorch 1.13.1; different model/runtime contexts." }, { "left": "issue:5846", "right": "issue:7330", "accept": false, - "reason": "Both are training-script errors, but one is resume/checkpoint handling and the other is FP16 gradient unscale during LoRA training." + "reason": "Both are training-script errors, but resume_from_checkpoint and FP16 unscale are different bugs." }, { "left": "issue:2210", "right": "issue:8504", "accept": false, - "reason": "Both are LoRA/text-encoder related, but one requests training the text encoder and the other asks for selecting which encoders to train in SD3-Medium." + "reason": "General request to train text encoder with LoRA and a specific SD3-Medium text-encoder selection feature are overlapping themes, but not the same change." }, { "left": "issue:2774", "right": "issue:7782", "accept": false, - "reason": "Both are conversion/unpickling problems, but they involve different failure modes and file handling paths." + "reason": "Both are conversion errors, but one is ControlNet safetensors-to-diffusers and the other is invalid safetensors loading; different failure modes." }, { "left": "issue:11023", "right": "issue:7221", "accept": false, - "reason": "Blurred output in a custom pipeline and noisy output from a model checkpoint are too vague to be the same underlying bug." + "reason": "Both concern bad image outputs, but one is blur with a custom pipeline and the other is noisy output with a model file; not enough to unify." }, { "left": "issue:4137", "right": "issue:7142", "accept": false, - "reason": "Inference mismatch with WebUI in two different contexts; no concrete shared code path or identical defect." + "reason": "Both compare diffusers to WebUI, but one is LoRA result mismatch and the other is SDXL-lightning inference mismatch; different pipelines and change targets." }, { "left": "issue:852", "right": "issue:9026", "accept": false, - "reason": "Both are API incompatibilities, but against different functions and kwargs; not the same issue." + "reason": "Both are unexpected keyword-argument errors, but they affect different APIs and versions; not the same underlying bug." }, { "left": "issue:10226", "right": "issue:9497", "accept": false, - "reason": "Both are Flux Dreambooth training errors, but one is tensor size mismatch and the other is mat2 dimensionality; distinct failure points." + "reason": "Both are Flux DreamBooth/LoRA training errors, but the reported failures are different shape/type mismatches and may stem from different code paths." }, { "left": "issue:6280", "right": "issue:9141", "accept": false, - "reason": "Same missing-from_single_file theme, but for different pipeline classes; separate support gaps rather than one duplicate bug." + "reason": "Both are about missing from_single_file support, but one is a specific ControlNet img2img pipeline and the other is the FluxPipeline API." }, { "left": "issue:10026", "right": "issue:9911", "accept": false, - "reason": "Both mention Flux controlnet training, but one is memory scaling under DeepSpeed and the other is a multi-controlnet shape mismatch." + "reason": "DeepSpeed memory scaling and multi-ControlNet layer-length mismatch are unrelated Flux issues." }, { "left": "issue:6481", "right": "issue:730", "accept": false, - "reason": "DreamBooth batch-size >1 failure vs local-vs-Colab training behavior; different underlying problems." + "reason": "DreamBooth batch-size >1 failure and local-vs-Colab training differences are not the same defect." }, { "left": "issue:3451", "right": "issue:3565", "accept": false, - "reason": "fp16 higher memory than fp32 in a training script vs generic text-to-image OOM; related theme, but not the same bug." - }, - { - "left": "issue:2185", - "right": "issue:2875", - "accept": false, - "reason": "Missing config files for a naive text-to-image example vs training loss becoming NaN; unrelated issues." + "reason": "Both involve memory during training, but one is fp16 using more memory than fp32 and the other is generic OOM in a text-to-image example; different root causes." } ] }, @@ -11468,12 +11474,14 @@ "issue:1540", "issue:1541", "issue:1672", - "issue:1988", + "issue:2185", + "issue:2875", "issue:3284", "issue:3454", "issue:4086", "issue:4094", "issue:4125", + "issue:4368", "issue:4398", "issue:4405", "issue:4875", @@ -11484,19 +11492,17 @@ "issue:5791", "issue:5798", "issue:5956", - "issue:5966", "issue:6117", "issue:6154", "issue:6232", "issue:6297", + "issue:6362", "issue:6382", "issue:6967", "issue:7589", - "issue:7658", "issue:8589", "issue:9278", "issue:9420", - "issue:9476", "issue:9497", "issue:9546", "issue:9667", @@ -11506,11 +11512,11 @@ "issue:993" ], "soft_pairs": [ + "issue:2185|issue:2875", "issue:4125|issue:5956", "issue:11853|issue:6382", "issue:11117|issue:5510", "issue:1672|issue:993", - "issue:1988|issue:9476", "issue:10226|issue:6967", "issue:5791|issue:8589", "issue:4398|issue:9841", @@ -11528,187 +11534,187 @@ "issue:10392|issue:9420", "issue:5016|issue:6297", "issue:1540|issue:1541", - "issue:6117|issue:6232", - "issue:5966|issue:7658" + "issue:4368|issue:6362", + "issue:6117|issue:6232" ], - "prepared_review_unit_hash": "64c563357038b673ea22b8478f2eee4fd42e769760bb8d2d7264d65dce647605", + "prepared_review_unit_hash": "26b2ca33dc142aeb2bbbf022113fff08b71276a52ee6d194dbb861ad20722b53", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34274, - "estimated_input_tokens": 8569, - "estimated_eval_tokens": 17394 + "serialized_chars": 34163, + "estimated_input_tokens": 8541, + "estimated_eval_tokens": 17338 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous and dominated by title similarity rather than true duplicates. Only the Dreambooth MODEL_TYPE pair looks like a real duplicate; the rest should stay separate.", - "confidence": 0.71, - "canonical_issue_reason": "issue:1541 is the clearest concrete bug report and best matches the only obvious duplicate pair about Dreambooth MODEL_TYPE / CKPT loading.", + "summary": "This cluster is mostly a mixed set of training/loading/performance reports across DreamBooth, LoRA, Flux, SDXL, IP-Adapter, and schedulers. Only the two DreamBooth MODEL_TYPE issues look like a true duplicate pair; the rest are similar in wording but describe different bugs or change requests.", + "confidence": 0.79, + "canonical_issue_reason": "issue:1541 is the clearest representative of the duplicate DreamBooth MODEL_TYPE / CKPT loading failure and states the root symptom more directly than 1540.", "canonical_pr_reason": null, - "best_issue_reason": "issue:1541 is the most specific and actionable issue in the cluster; it names the actual failure mode more clearly than 1540.", + "best_issue_reason": "issue:1541 is the best single issue to anchor this cluster because it cleanly captures the DreamBooth checkpoint-loading / MODEL_TYPE failure mode.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2185", + "right": "issue:2875", + "accept": false, + "reason": "One is missing training config files; the other is a training-stability/nan-loss problem." + }, { "left": "issue:4125", "right": "issue:5956", "accept": false, - "reason": "Both are training-quality complaints, but they concern different models and symptoms, not the same bug." + "reason": "Both concern training behavior, but they are different models and different symptoms (expected loss curve vs degraded image quality)." }, { "left": "issue:11853", "right": "issue:6382", "accept": false, - "reason": "Both mention IP-Adapter, but one is a load-time key error and the other is a controlnet/IP-adapter conflict; different failure paths." + "reason": "Both mention IP-Adapter, but one is a ControlNet/IP-Adapter loading size conflict and the other is an LCM LoRA key error after training." }, { "left": "issue:11117", "right": "issue:5510", "accept": false, - "reason": "Both are dtype mismatch errors, but in different pipelines and code paths, so they are not clearly the same bug." + "reason": "Different pipelines and different dtype mismatch sites; not the same bug." }, { "left": "issue:1672", "right": "issue:993", "accept": false, - "reason": "Image-size conversion failure vs fp16 dtype mismatch; unrelated issues." - }, - { - "left": "issue:1988", - "right": "issue:9476", - "accept": false, - "reason": "Both are type/bias mismatches, but they occur in different scripts and model families; not the same concrete defect." + "reason": "Wrong image size after conversion is unrelated to an fp16 data-type mismatch." }, { "left": "issue:10226", "right": "issue:6967", "accept": false, - "reason": "Both involve shape mismatches in Dreambooth/LoRA, but the model setups and reported failures differ too much to merge." + "reason": "Both are shape-mismatch reports, but they involve different models and different training contexts (Flux LoRA vs SDXL prior preservation)." }, { "left": "issue:5791", "right": "issue:8589", "accept": false, - "reason": "Both mention prior preservation, but one is about interaction with SNR gamma and the other is a separate SD3 training issue." + "reason": "Both mention prior preservation, but one is a parameter-combination question and the other is an SD3-specific prior-preservation training failure." }, { "left": "issue:4398", "right": "issue:9841", "accept": false, - "reason": "Black-image generation and a use_dora TypeError are different bugs." + "reason": "Black outputs and a use_dora TypeError are different failure modes." }, { "left": "issue:3284", "right": "issue:3454", "accept": false, - "reason": "Same script family, but these read as broad compatibility complaints rather than one shared code-path bug." + "reason": "Same script family, but the reported failures are not specific enough to conclude the same underlying bug." }, { "left": "issue:9667", "right": "issue:9914", "accept": false, - "reason": "Both are Flux loading problems, but one is fp8 model loading and the other is LoRA loading; different defects." + "reason": "One is loading a Flux fp8 model through a local transformer file; the other is loading a trained LoRA with Xlabs on Diffusers." }, { "left": "issue:4086", "right": "issue:7589", "accept": false, - "reason": "Different subsystems entirely: SDXL ControlNet training vs GLIGEN layout synthesis." + "reason": "Different features entirely: SDXL ControlNet training vs GLIGEN layout-to-image synthesis." }, { "left": "issue:9278", "right": "issue:9497", "accept": false, - "reason": "One is multigpu training support and the other is a mat2 dimensionality error; not the same issue." + "reason": "Both are Flux DreamBooth issues, but one is multigpu training and the other is a matmul shape error." }, { "left": "issue:9546", "right": "issue:9857", "accept": false, - "reason": "OOM during validation and embedding size mismatch are different failures." + "reason": "Validation OOM and embedding tensor size mismatch are different problems." }, { "left": "issue:4875", "right": "issue:5178", "accept": false, - "reason": "Both are DPM-SDE-related crashes when inputs change shape, but the scheduler variants and reported triggers differ; too weak to treat as one bug." + "reason": "Both involve DPM SDE state across runs, but the triggering condition and scheduler variant differ enough that they do not look like the same bug." }, { "left": "issue:5565", "right": "issue:6154", "accept": false, - "reason": "Both are SDXL LoRA training problems, but one is a script/run issue and the other is an M2 Mac failure." + "reason": "Generic script failure versus an M2 Mac-specific LoRA training problem." }, { "left": "issue:11922", "right": "issue:12459", "accept": false, - "reason": "Different FLUX tasks: fill-dev running problem vs Kontext batch inference throughput." + "reason": "Different model families and symptoms; one is FLUX.1-Fill-dev running problem, the other is a Kontext-dev batch-throughput issue." }, { "left": "issue:4094", "right": "issue:4405", "accept": false, - "reason": "Both are feature requests for training scripts, but they target different techniques and are not duplicates." + "reason": "These are separate feature requests for different training scripts." }, { "left": "issue:10848", "right": "issue:5798", "accept": false, - "reason": "Both are unexpected keyword argument errors, but they involve different functions and unrelated APIs." + "reason": "Both are unexpected-argument errors, but they hit different functions/pipelines and do not share the same code path." }, { "left": "issue:12047", "right": "issue:12535", "accept": false, - "reason": "Both concern Wan2.2 LoRAs, but one is fusion on GGUF and the other is loading a Lightning LoRA; not the same bug." + "reason": "Both are Wan2.2 LoRA-related, but one is fusing a Lightx2v LoRA on GGUF and the other is failing to load a Lightning LoRA." }, { "left": "issue:10392", "right": "issue:9420", "accept": false, - "reason": "Both are Flux/Hyper-SD-related, but one is an fp8 loading setup and the other is a saved LoRA load failure." + "reason": "Different Flux+LoRA loading problems with different model variants and failure points." }, { "left": "issue:5016", "right": "issue:6297", "accept": false, - "reason": "Both are Dreambooth SDXL LoRA training complaints, but one is specifically about text encoder training and the other is a vague inability to train on photos." + "reason": "Text-encoder training is a different issue from a general complaint that the current script cannot train on photos." }, { "left": "issue:1540", "right": "issue:1541", "accept": true, - "reason": "Both report the same Dreambooth MODEL_TYPE/CKPT loading failure; they are clearly the same underlying bug." + "reason": "Both report the same DreamBooth MODEL_TYPE / CKPT loading failure; 1541 is the clearer root-cause description." }, { - "left": "issue:6117", - "right": "issue:6232", + "left": "issue:4368", + "right": "issue:6362", "accept": false, - "reason": "Both are scheduler add_noise issues, but they affect different schedulers and settings." + "reason": "One is training-time OOM for SDXL LoRA; the other is dataset mapping OOM." }, { - "left": "issue:5966", - "right": "issue:7658", + "left": "issue:6117", + "right": "issue:6232", "accept": false, - "reason": "Feature/question about ControlNet support vs asking about multiple conditioning; not duplicates." + "reason": "Both are scheduler add_noise bugs, but they involve different scheduler classes and different failing settings." } ] }, "evaluator_result": { "accept": true, - "feedback": "The canonical choice and summary are grounded in the packet. The Dreambooth MODEL_TYPE pair is the only clearly supported duplicate, and the soft-edge rejections are appropriately conservative given that the evidence is mostly title-level similarity. No PR-mergeability issue applies here." + "feedback": "The summary is grounded in the packet and stays conservative. The only clearly supported duplicate is the DreamBooth MODEL_TYPE / CKPT-loading pair (1540/1541), and the other soft-edge verdicts appropriately avoid overclaiming equivalence." }, "evaluator_used": true, "retried": false, @@ -11734,7 +11740,6 @@ "issue:11117", "issue:11127", "issue:11133", - "issue:11285", "issue:11352", "issue:11432", "issue:12216", @@ -11759,18 +11764,19 @@ "issue:5455", "issue:5769", "issue:5805", + "issue:5966", "issue:6159", "issue:6278", "issue:6695", "issue:6832", "issue:7466", + "issue:7658", "issue:7666", "issue:771", "issue:7799", "issue:8405", "issue:8414", "issue:8667", - "issue:8939", "issue:9112", "issue:9237", "issue:938", @@ -11779,6 +11785,7 @@ "issue:9953" ], "soft_pairs": [ + "issue:5966|issue:7658", "issue:11352|issue:9953", "issue:6832|issue:7799", "issue:3803|issue:6278", @@ -11803,193 +11810,192 @@ "issue:10833|issue:5075", "issue:5455|issue:6695", "issue:11023|issue:7666", - "issue:11133|issue:13043", - "issue:11285|issue:8939" + "issue:11133|issue:13043" ], - "prepared_review_unit_hash": "dfd5587a9e57af423c12d206e8e454b65e5c04abf73b772c087e309ad0addd6b", + "prepared_review_unit_hash": "7264d9dac5b6d610f15a81d5433473f2f4196c1fd2d5ad568e1e98e8ac904f67", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 35161, - "estimated_input_tokens": 8791, - "estimated_eval_tokens": 17838 + "serialized_chars": 34814, + "estimated_input_tokens": 8704, + "estimated_eval_tokens": 17664 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are only superficially similar: they share broad vocabulary like shape mismatch, dtype mismatch, quantization, multi-GPU, or specific model names, but they point to different models, pipelines, or code paths. I would not merge any of the soft edges; the closest recurring theme is quantized-model device transfer, best represented by issue 9953.", - "confidence": 0.87, - "canonical_issue_reason": "Issue 9953 is the broadest and most representative concrete bug in the set: moving a pipeline with quantized components to CUDA. It is more general than the nearby CPU-warning report, but still distinct enough that the pair should not be treated as duplicates.", + "summary": "This cluster is a broad grab-bag of diffusion/pipeline bug reports, mostly about model compatibility, dtype/device mismatches, quantization, and training/inference failures. The soft matches are mostly superficial (same subsystem or similar error wording) rather than true duplicates.", + "confidence": 0.42, + "canonical_issue_reason": "issue:9953 is the best representative of the cluster\u2019s most recurring concrete theme: pipeline failures when moving quantized components across devices. It is specific enough to anchor the cluster, while still broader than the more model-specific reports.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 9953 is the best overall representative issue among these items because it is concrete, widely scoped, and closest to the strongest recurring subsystem theme in the set.", + "best_issue_reason": "issue:9953 is the most generally useful issue here because it describes a common runtime/device-transfer bug pattern that plausibly spans multiple similar reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5966", + "right": "issue:7658", + "accept": false, + "reason": "Related ControlNet topic, but one is SD 2.1 support and the other is SDXL multiple-conditioning; not the same bug." + }, { "left": "issue:11352", "right": "issue:9953", "accept": false, - "reason": "Same broad quantization/device-move area, but one is a missing CPU warning and the other is a CUDA transfer error; different symptom and likely different fix." + "reason": "Both involve quantized pipelines and device moves, but one is missing a CPU warning and the other is a CUDA error; different failures." }, { "left": "issue:6832", "right": "issue:7799", "accept": false, - "reason": "Both mention matmul/shape mismatch, but they involve different pipelines and different adapter combinations, so they are not the same bug." + "reason": "Both are IP-Adapter matmul/shape problems, but they hit different pipelines and likely different code paths." }, { "left": "issue:3803", "right": "issue:6278", "accept": false, - "reason": "Both are multi-GPU training problems, but one is LoRA model sharing and the other is LCM distillation startup; different failure modes." + "reason": "Both mention multi-GPU training, but the scripts and failure modes are different; not a duplicate bug." }, { "left": "issue:4206", "right": "issue:4925", "accept": false, - "reason": "Both touch SDXL controlnet training, but one is a broken main-branch regression while the other is distributed OOM; not the same defect." + "reason": "Both concern SDXL ControlNet training, but one is a general main-branch breakage and the other is a distributed OOM issue." }, { "left": "issue:11127", "right": "issue:9667", "accept": false, - "reason": "Both are Flux single-file/loading complaints, but one is a CivitAI model not working and the other is a local transformer file loading error; too different to merge." + "reason": "Both are Flux loading issues, but one is a single-file compatibility problem and the other is fp8/local transformer loading." }, { "left": "issue:12216", "right": "issue:12458", "accept": false, - "reason": "Both are Qwen-image-edit reports, but one is output quality and the other is batch inference/feature request; not a duplicate bug." + "reason": "One reports inferior Qwen-Image-Edit results, the other is a batch inference request/issue; different concerns." }, { "left": "issue:3382", "right": "issue:4925", "accept": false, - "reason": "Both mention GPU memory, but one is a multi-GPU training OOM and the other is SDXL controlnet distributed OOM; different code paths and contexts." + "reason": "Both are training OOM reports, but they are from different training setups and not the same underlying defect." }, { "left": "issue:8405", "right": "issue:9112", "accept": false, - "reason": "Both have tensor-size issues, but they are for different models and pipelines, so they do not appear to be the same underlying bug." + "reason": "Both mention tensor-size errors, but on different models and different inference paths." }, { "left": "issue:8414", "right": "issue:938", "accept": false, - "reason": "A new text-to-audio model request is unrelated to a human motion diffusion model issue." + "reason": "Completely different models and domains: text-to-audio vs text-to-motion." }, { "left": "issue:8667", "right": "issue:9112", "accept": false, - "reason": "Both are dimension mismatch errors, but one is SD3ControlNetModel and the other is Auraflow inference; different model-specific bugs." + "reason": "Similar size-mismatch wording, but the model families and likely root causes differ." }, { "left": "issue:4872", "right": "issue:4925", "accept": false, - "reason": "Both are OOM reports, but one is SDXL finetuning on T4 and the other is distributed ControlNet training; not the same problem." + "reason": "Both are OOM complaints, but one is SDXL finetuning on T4 and the other is distributed ControlNet training." }, { "left": "issue:175", "right": "issue:5769", "accept": false, - "reason": "One is about example training seeding and the other is about mixed precision correctness; unrelated concerns." + "reason": "Both ask about training correctness, but one is about seeding and the other about mixed precision; not the same issue." }, { "left": "issue:13105", "right": "issue:13258", "accept": false, - "reason": "Both concern Wan models, but one is noisy output from T2V and the other is missing TI2V functionality; different issues." + "reason": "Different Wan model versions and different symptoms: noisy output vs missing image-to-video behavior." }, { "left": "issue:1552", "right": "issue:771", "accept": false, - "reason": "Different feature bugs: attention slicing with SD2 versus BF16 with dreambooth." + "reason": "One is attention slicing on SD2, the other is BF16 in dreambooth; unrelated precision/memory issues." }, { "left": "issue:9237", "right": "issue:9500", "accept": false, - "reason": "Both involve Flux training, but one is LoRA training and the other is checkpoint saving during dreambooth; not the same bug." + "reason": "Both are Flux dreambooth-related, but one is a training failure and the other is a checkpoint-saving failure." }, { "left": "issue:2891", "right": "issue:6159", "accept": false, - "reason": "Unsupported embeddings in one pipeline and unsupported denoising_end in another are separate API-support issues." + "reason": "Different pipelines and different API gaps; one is embeds support, the other is denoising_end support." }, { "left": "issue:2549", "right": "issue:4520", "accept": false, - "reason": "One is SD2 UNet compatibility, the other is LoRA scale with attention slicing; different functionality." + "reason": "Different bugs in different features: SD 2.0 UNet loading vs LoRA scale with attention slicing." }, { "left": "issue:3382", "right": "issue:3451", "accept": false, - "reason": "Both are training memory topics, but one is multi-GPU OOM and the other is fp16 using more memory than fp32; different root causes." + "reason": "Both are training-memory reports, but one is multi-GPU OOM and the other is fp16 memory regression." }, { "left": "issue:11117", "right": "issue:5805", "accept": false, - "reason": "Both are dtype mismatch errors, but they affect different pipelines and backends; too model-specific to treat as duplicates." + "reason": "Both are dtype mismatch errors, but they occur in different pipelines and environments." }, { "left": "issue:5254", "right": "issue:7466", "accept": false, - "reason": "A feature request for multi-ControlNet support is not the same as a torch.compile failure in multi-ControlNet inpaint." + "reason": "One is a feature request for multi-ControlNet support, the other is a torch.compile failure; not the same change or bug." }, { "left": "issue:11432", "right": "issue:4724", "accept": false, - "reason": "torch_dtype='auto' loading behavior and SDXL img2img quality/behavior are unrelated." + "reason": "A from_pretrained dtype-auto issue and an SDXL img2img behavior problem are not the same underlying defect." }, { "left": "issue:10833", "right": "issue:5075", "accept": false, - "reason": "Both are matrix-shape errors, but they happen in different pipelines and on different model families; not the same concrete bug." + "reason": "Both are matmul shape errors, but they happen in different pipelines with different inputs and likely different root causes." }, { "left": "issue:5455", "right": "issue:6695", "accept": false, - "reason": "Both mention subprocess errors, but the underlying notebook/workflow failures are unrelated." + "reason": "Same generic subprocess error string, but the surrounding workflows are too different to call them duplicates." }, { "left": "issue:11023", "right": "issue:7666", "accept": false, - "reason": "Both mention lpw_stable_diffusion_xl, but one is blurry output on Playground v2.5 and the other is from_single_file incompatibility; different problems." + "reason": "Both mention lpw_stable_diffusion_xl, but one reports blurry output and the other single-file loading failure." }, { "left": "issue:11133", "right": "issue:13043", "accept": false, - "reason": "A CogVideoX image-to-video bug is unrelated to a GPU kernel support issue on RTX 5090." - }, - { - "left": "issue:11285", - "right": "issue:8939", - "accept": false, - "reason": "A conversion/value-error report and a tokenizer-loading failure are different failure modes with no clear duplicate evidence." + "reason": "Unrelated models and failures: CogVideoX pipeline bug vs Blackwell kernel-image compatibility issue." } ] }, @@ -12016,12 +12022,11 @@ "issue:10040", "issue:1082", "issue:10937", - "issue:1100", "issue:11133", + "issue:11285", "issue:12108", "issue:12195", "issue:12235", - "issue:1249", "issue:12666", "issue:13351", "issue:1875", @@ -12058,11 +12063,13 @@ "issue:8613", "issue:8667", "issue:8817", + "issue:8939", "issue:9112", "issue:9187", "issue:9202" ], "soft_pairs": [ + "issue:11285|issue:8939", "issue:5455|issue:7133", "issue:1082|issue:7146", "issue:5006|issue:7146", @@ -12086,194 +12093,196 @@ "issue:2234|issue:2296", "issue:4316|issue:8613", "issue:7263|issue:9187", - "issue:7147|issue:758", - "issue:1100|issue:1249" + "issue:7147|issue:758" ], - "prepared_review_unit_hash": "16cf077ca266ec3a8300b23713746c6bf24ae0dfe94deed0fffb0c9e8ba450dd", + "prepared_review_unit_hash": "ec26074868e2c6559578d3129c346308007255b7cf092be68bf3b5a42282c67d", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34716, - "estimated_input_tokens": 8679, - "estimated_eval_tokens": 17614 + "serialized_chars": 34966, + "estimated_input_tokens": 8742, + "estimated_eval_tokens": 17740 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All candidate edges are weak title-level similarities; none look like the same concrete bug/change, so I rejected every soft pair. The issues span unrelated failures, model families, and code paths.", - "confidence": 0.79, - "canonical_issue_reason": "Issue 7146 is the clearest, most concrete bug report in the set: a specific SDXL training OOM with script/version context. It is the best available representative only because there is no strong duplicate cluster here.", + "summary": "This cluster is heterogeneous: most items are unrelated API, training, memory, or model-specific bugs that only share superficial wording. There is no single coherent duplicate target across the whole set.", + "confidence": 0.94, + "canonical_issue_reason": "No clear canonical issue: the items span many unrelated failure modes and subsystems, so no single issue cleanly represents the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 7146 has a well-scoped reproduction target and specific failure mode, making it the most suitable single issue to anchor triage if one must be chosen from this mixed set.", + "best_issue_reason": "No strong best issue candidate for the whole cluster; the only near-duplicate-looking pair is a narrow API-missing report, but it does not unify the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11285", + "right": "issue:8939", + "accept": false, + "reason": "Different problems: conversion/value errors vs tokenizer loading failure." + }, { "left": "issue:5455", "right": "issue:7133", "accept": false, - "reason": "Both involve Dreambooth/training, but one is a notebook subprocess failure and the other is a checkpoint-resume text-encoder size mismatch; different bugs." + "reason": "Dreambooth notebook subprocess failure is not the same as SDXL checkpoint-resume size mismatch." }, { "left": "issue:1082", "right": "issue:7146", "accept": false, - "reason": "Both are CUDA failures during training, but the error types and contexts differ (CUBLAS matmul crash vs SDXL training OOM)." + "reason": "CUDA cublas execution failure differs from a generic SDXL OOM during finetuning." }, { "left": "issue:5006", "right": "issue:7146", "accept": false, - "reason": "Both mention SDXL training memory issues, but one is end-of-training BF16 OOM and the other is a general training OOM; not the same concrete failure." + "reason": "Both mention OOM, but one is a BF16 end-of-training spike and the other is a broader training OOM; not the same concrete bug." }, { "left": "issue:3382", "right": "issue:4872", "accept": false, - "reason": "Both are OOM reports, but one is multi-GPU text-to-image training and the other is T4 Colab SDXL finetuning; too broad to be the same bug." + "reason": "Different training contexts and symptoms: multi-GPU OOM vs T4 Colab SDXL finetuning OOM." }, { "left": "issue:2578", "right": "issue:7178", "accept": false, - "reason": "Editable install failure and export_to_video malfunction are unrelated code paths and different kinds of breakage." + "reason": "Editable install failure is unrelated to export_to_video malfunction." }, { "left": "issue:12235", "right": "issue:13351", "accept": false, - "reason": "Both relate to Wan fun/control features, but one is a support request and the other is a specific model/controlnet combination; not clearly the same issue." + "reason": "Related feature area, but different request scopes and model variants; not one concrete change." }, { "left": "issue:4450", "right": "issue:7735", "accept": false, - "reason": "Multi-text inversion not working is unrelated to a missing caption column in a training dataset." + "reason": "Textual inversion multi-text support is unrelated to a missing caption_column field." }, { "left": "issue:4542", "right": "issue:7378", "accept": false, - "reason": "One is a pipeline output-count limitation; the other is a memory-cost complaint about latent decoding. Different symptoms and likely different fixes." + "reason": "One is a multi-image output limit, the other is memory use in latent decoding." }, { "left": "issue:539", "right": "issue:6761", "accept": false, - "reason": "Both are API/keyword-argument errors, but they hit different functions with different missing parameters; not the same underlying bug." + "reason": "Different unexpected-argument errors on unrelated APIs." }, { "left": "issue:7644", "right": "issue:8667", "accept": false, - "reason": "Both concern non-power-of-two image handling, but they affect different models/pipelines and the exact failure modes are not shown to be the same." + "reason": "Both involve non-power-of-two images, but different model/pipeline code paths and failure modes." }, { "left": "issue:8817", "right": "issue:9202", "accept": false, - "reason": "Both involve IP-Adapter/FaceID, but one is a general FaceID problem and the other is a specific SDXL PAGPipeline incompatibility; too distinct." + "reason": "Different IP-Adapter variants and pipeline behavior; not the same bug." }, { "left": "issue:10040", "right": "issue:3481", "accept": false, - "reason": "Grey-image generation is not the same as DDIM step-count changing image determinism; one is a rendering defect, the other is expected sampling behavior." + "reason": "Grey-image generation is unrelated to DDIM determinism across inference steps." }, { "left": "issue:1875", "right": "issue:720", "accept": false, - "reason": "Both are mixed precision/type mismatch errors, but one is Imagic fp16 training and the other is traced UNet documentation code; different code paths." + "reason": "Both are dtype mismatch errors, but in different pipelines and code paths." }, { "left": "issue:11133", "right": "issue:12108", "accept": false, - "reason": "CogVideoX image-to-video and Qwen/Chroma flow-matching scheduler issues are separate pipeline/scheduler problems." + "reason": "CogVideoX image-to-video bug is unrelated to Qwen/Chroma flow-matching scheduler breakage." }, { "left": "issue:4052", "right": "issue:6280", - "accept": false, - "reason": "Missing from_single_file on different pipeline classes is a shared API gap, but these are separate classes and not the same concrete bug." + "accept": true, + "reason": "Both report the same missing `from_single_file` pipeline API; the titles differ only in how specifically the pipeline is named." }, { "left": "issue:12666", "right": "issue:4356", "accept": false, - "reason": "A missing Kandinsky 5.0 video pipeline and a missing dummy_torch_and_transformers module are unrelated release/package issues." + "reason": "Kandinsky pipeline availability is unrelated to a missing dummy module import." }, { "left": "issue:5607", "right": "issue:9112", "accept": false, - "reason": "Tensor size mismatch here appears in unrelated contexts: one during concatenation with 64 vs 512, the other in Auraflow with 264 vs 4096." + "reason": "Tensor size mismatch here is a generic shape issue, but the pipelines and reported failure contexts differ." }, { "left": "issue:10937", "right": "issue:6239", "accept": false, - "reason": "Both touch torch.compile/FX around VAE encode, but one is an error on vae.encode and the other is a request to better support latent_dist.sample(); not the same bug." + "reason": "Both touch VAE/torch FX, but one is a compile error on vae.encode and the other is a torch FX support request for latent sampling." }, { "left": "issue:2904", "right": "issue:3497", "accept": false, - "reason": "AudioLDM scheduler NoneType and StableDiffusionInpaintPipeline from_ckpt are unrelated pipeline API failures." + "reason": "AudioLDM scheduler NoneType error is unrelated to a missing from_ckpt attribute." }, { "left": "issue:12195", "right": "issue:4356", "accept": false, - "reason": "A PyTorch 2.3.1 custom_op incompatibility is not the same as the missing dummy module import issue." + "reason": "A PyTorch version incompatibility around `torch.library.custom_op` is unrelated to the missing dummy module import." }, { "left": "issue:2234", "right": "issue:2296", "accept": false, - "reason": "Both concern xformers attention, but one is training fine-tuning failure and the other is a T4 inference/runtime issue; different manifestations and likely different root causes." + "reason": "Both mention xformers attention, but one is training failure and the other is T4 inference incompatibility." }, { "left": "issue:4316", "right": "issue:8613", "accept": false, - "reason": "Validation-option failure in dreambooth training is unrelated to an OOM from caption_column configuration." + "reason": "Validation-option failure during training is unrelated to an OOM triggered by caption_column usage." }, { "left": "issue:7263", "right": "issue:9187", "accept": false, - "reason": "IP Adapter + attention slicing and IP Adapter + image_encoder_folder=None after Kolors update are different incompatibilities." + "reason": "Different IP-Adapter failure modes: attention slicing vs SDXL/image_encoder_folder=None after Kolors update." }, { "left": "issue:7147", "right": "issue:758", "accept": false, - "reason": "Missing StableCascadeDecoderPipeline import and urllib3 PROTOCOL_TLS import error are unrelated dependency/API breakages." - }, - { - "left": "issue:1100", - "right": "issue:1249", - "accept": false, - "reason": "Both mention missing .device attributes, but one is CLIPTextTransformer and the other is numpy.ndarray in OnnxPipeline; different objects and code paths." + "reason": "StableCascade import failure is unrelated to the urllib3 PROTOCOL_TLS import error." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall: the cluster is clearly heterogeneous, the no-canonical-issue conclusion is conservative, and the one accepted soft edge (issue:4052 \u2194 issue:6280) is a reasonable same-API-missing match. The false soft-edge rejections are also cautious and do not overstate similarity." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -12294,11 +12303,13 @@ "issue:10428", "issue:10848", "issue:10937", + "issue:1100", "issue:11033", "issue:11149", "issue:11464", "issue:11651", "issue:1214", + "issue:1249", "issue:13137", "issue:13446", "issue:13484", @@ -12309,7 +12320,6 @@ "issue:2856", "issue:3045", "issue:3480", - "issue:3582", "issue:4194", "issue:4368", "issue:4453", @@ -12320,7 +12330,6 @@ "issue:4969", "issue:5006", "issue:5374", - "issue:5619", "issue:5846", "issue:6278", "issue:6524", @@ -12340,6 +12349,7 @@ "issue:9841" ], "soft_pairs": [ + "issue:1100|issue:1249", "issue:10848|issue:11149", "issue:10937|issue:9091", "issue:6879|issue:9497", @@ -12363,187 +12373,186 @@ "issue:2780|issue:5846", "issue:1434|issue:4874", "issue:6278|issue:8236", - "issue:4534|issue:9841", - "issue:3582|issue:5619" + "issue:4534|issue:9841" ], - "prepared_review_unit_hash": "a54afc570cbad43d2248a298b78308bab1855f4edcb222e09ec39977622e8a7b", + "prepared_review_unit_hash": "84b65760e8007d4542dda8ad5c8523265c7b1113e30aa07424cefa405b7b8337", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34204, - "estimated_input_tokens": 8551, - "estimated_eval_tokens": 17358 + "serialized_chars": 34268, + "estimated_input_tokens": 8567, + "estimated_eval_tokens": 17390 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a noisy, weakly related cluster: most pairs only share a broad subsystem/theme (Flux, LoRA, VAE, training, etc.) but not the same concrete bug or fix path. I would not merge any of the soft pairs as duplicates.", - "confidence": 0.23, - "canonical_issue_reason": "issue:10381 is the broadest and earliest Flux/LoRA-related report here, so it is the closest thing to a representative issue, but it still does not cover the rest of the heterogeneous set.", + "summary": "The set is heterogeneous; most soft pairs share only broad subsystem vocabulary and do not look like the same underlying bug. The closest mini-cluster is Flux/LoRA loading, but even there the reports appear to be different variants rather than duplicates.", + "confidence": 0.68, + "canonical_issue_reason": "No single issue cleanly represents the whole set. If a representative is needed, issue 11464 is the best fit for the Flux/LoRA-loading subset because it is the broadest FluxControlPipeline report and is nearest to 11651.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10381 is the most usable umbrella entry if forced to pick one issue, because it is general and central to the Flux + LoRA subset; still, the cluster is too diverse for it to be a strong canonical for all items.", + "best_issue_reason": "Issue 11464 is the most suitable canonical issue among the provided items: it is the clearest, broadest Flux/LoRA loading failure and the only one with an obvious nearby sibling in the set.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1100", + "right": "issue:1249", + "accept": false, + "reason": "Both mention device-related attribute errors, but one is a CLIPTextTransformer attribute lookup and the other is a numpy.ndarray/device issue in OnnxPipeline. Different code paths and root causes." + }, { "left": "issue:10848", "right": "issue:11149", "accept": false, - "reason": "Different failures: IP-Adapter meta-loading keyword mismatch vs bitsandbytes quantizer dtype argument mismatch. Same general area, but not the same bug." + "reason": "Both are unexpected keyword-argument failures, but they occur in different loaders: load_model_dict_into_meta(device) vs BnB4BitDiffusersQuantizer.create_quantized_param(dtype). Not the same bug." }, { "left": "issue:10937", "right": "issue:9091", "accept": false, - "reason": "Both mention torch.compile and VAE, but one is a vae.encode error and the other is VAE decoder compile support. Different code paths and fixes." + "reason": "Both involve torch.compile and VAE components, but one is an encode-time error while the other is decoder support for a specific VAE. Different failure modes." }, { "left": "issue:6879", "right": "issue:9497", "accept": false, - "reason": "Both are dtype/matmul training errors, but one is Custom Diffusion mixed-precision mismatch and the other is Flux Dreambooth mat2 shape error. Not the same issue." + "reason": "Both are dtype/matmul runtime errors, but one is Custom Diffusion training and the other is Flux Dreambooth training. Different models and likely different code paths." }, { "left": "issue:10381", "right": "issue:10428", "accept": false, - "reason": "Flux LoRA/quantization inference error versus a Flux inference failure on Ascend NPU. Similar model family, but no evidence of the same underlying bug." + "reason": "Similar Flux inference wording, but one is quantization + LoRA on zero-GPU spaces and the other is Ascend NPU inference. Different backend/hardware-specific problems." }, { "left": "issue:1614", "right": "issue:4724", "accept": false, - "reason": "Both are image-quality/behavior problems in diffusion pipelines, but one is blank output under autocast and the other is img2img mismatch with an SDXL checkpoint. Different symptoms and causes." + "reason": "Both are Stable Diffusion pipeline failures, but black output with autocast is unrelated to an img2img checkpoint mismatch. Not the same underlying bug." }, { "left": "issue:11033", "right": "issue:1214", "accept": false, - "reason": "From_single_file loading failure for SD1.5 UNet versus CLIPTextModel half-precision failure. Same broad model stack, different broken component." + "reason": "One is SD1.5 from_single_file loading for UNet, the other is CLIPTextModel half-precision behavior. Different components and symptoms." }, { "left": "issue:4368", "right": "issue:5006", "accept": false, - "reason": "Both are OOM reports during LoRA/SDXL training, but one is general LoRA training OOM and the other is BF16-specific OOM at training end. Not the same bug." + "reason": "Both are training memory issues, but one is SDXL LoRA OOM and the other is end-of-training BF16 OOM. Too broad to treat as the same bug." }, { "left": "issue:1214", "right": "issue:7660", "accept": false, - "reason": "CLIP half-precision failure is unrelated to a trace example not working. No shared concrete code path." + "reason": "Half-precision CLIPTextModel failure and a trace example failure are unrelated." }, { "left": "issue:13137", "right": "issue:13484", "accept": false, - "reason": "Both involve Flux 2 and LoRA loading, but one is a Klein/LoKr loading error and the other is FAL LoRA loading failure. Too model-variant-specific to assume one bug." + "reason": "Both are Flux 2 LoRA loading problems, but one is LoKr weight loading and the other is FAL LoRA loading. Close topic, but not clearly the same code-path bug." }, { "left": "issue:13446", "right": "issue:9660", "accept": false, - "reason": "Custom pipeline trust-check issue versus from_pretrained not setting eval mode. Different API behaviors and fixes." + "reason": "Custom pipeline trust checking and from_pretrained eval-mode handling are unrelated issues in different parts of the API." }, { "left": "issue:2210", "right": "issue:3480", "accept": false, - "reason": "Training text encoder with LoRA is a feature request; adding a scale parameter for text-encoder LoRA layers is a separate API enhancement." + "reason": "Both concern text-encoder LoRA, but one is a training-capability request and the other is a scale-parameter feature request. Same area, not the same change." }, { "left": "issue:2856", "right": "issue:6584", "accept": false, - "reason": "Generic conversion failure versus a specific DreamShaper v8 inpaint model conversion problem. Related conversion domain, not the same bug." + "reason": "Generic SD-to-Diffusers conversion versus a specific DreamShaper v8 inpaint conversion failure. Not the same underlying conversion bug." }, { "left": "issue:3045", "right": "issue:4453", "accept": false, - "reason": "Both concern LoRA conversion/export, but one is general safetensors-to-diffusers conversion and the other is SDXL LoRA compatibility with A1111. Different targets." + "reason": "These are opposite directions and different ecosystems: converting LoRA safetensors to Diffusers vs converting Diffusers SDXL LoRA to AUTO1111-compatible safetensors." }, { "left": "issue:4194", "right": "issue:6524", "accept": false, - "reason": "VAE not working in from_single_file versus reloading a LoRA a second time. Different pipelines and failure modes." + "reason": "VAE not working in from_single_file and LoRA failing on second load are different subsystems." }, { "left": "issue:4969", "right": "issue:6794", "accept": false, - "reason": "ImportError for CrossAttention removal versus import error for cast_training_params. Both are import issues, but entirely different symbols." + "reason": "Importing CrossAttention and importing cast_training_params are unrelated import errors." }, { "left": "issue:5374", "right": "issue:6761", "accept": false, - "reason": "Both are unexpected-keyword-argument errors, but against different APIs: checkpoint_path vs VaeImageProcessor.preprocess. Not the same bug." + "reason": "Both are unexpected keyword-argument errors, but they affect different methods with different parameters and components." }, { "left": "issue:7033", "right": "issue:7238", "accept": false, - "reason": "IP-Adapter attention masking warning logs versus IP-Adapter masking behavior. Same subsystem, but one is logging/warning noise and the other is functional masking behavior." + "reason": "Both mention IP-Adapter masking, but one is warning-log noise and the other is a masking behavior issue. Not enough evidence of the same bug." }, { "left": "issue:8708", "right": "issue:8887", "accept": false, - "reason": "Missing shift_factor in SD3 dreambooth examples versus wrong VAE encode in SD3 training script. Both SD3 training issues, but distinct defects." + "reason": "Both are SD3 training/VAE-related, but one is a missing config field in dreambooth examples and the other is a wrong VAE encode path. Related, but not the same defect." }, { "left": "issue:11464", "right": "issue:11651", "accept": false, - "reason": "Both are Flux Control LoRA loading failures, but they target different LoRA variants/models (FLUX.1-Canny-dev-lora vs flux-fill-lora). Not enough evidence of a single shared bug." + "reason": "Both are FluxControl LoRA loading failures, but they refer to different LoRA variants (Canny-dev vs flux-fill) and likely separate loading cases. Too distinct to merge." }, { "left": "issue:4835", "right": "issue:4874", "accept": false, - "reason": "ControlNet MultiControlNetModel bug versus UpscalePipeline low_res_scheduler bug. Same family of pipelines but different concrete errors." + "reason": "Different pipeline bugs: ControlNet img2img with MultiControlNetModel versus UpscalePipeline low_res_scheduler when loading locally." }, { "left": "issue:2780", "right": "issue:5846", "accept": false, - "reason": "Both are LoRA training script errors, but one is FP16 gradient unscale and the other is resume_from_checkpoint. Different training flow issues." + "reason": "Both are LoRA training issues, but one is FP16 gradient unscale handling and the other is resume_from_checkpoint behavior in a different script/version." }, { "left": "issue:1434", "right": "issue:4874", "accept": false, - "reason": "OOM during upscaling large images versus low_res_scheduler bug when loading locally. One is resource usage, the other is scheduler handling." + "reason": "Memory blowout during upscaling is unrelated to the low_res_scheduler local-loading bug." }, { "left": "issue:6278", "right": "issue:8236", "accept": false, - "reason": "Both involve multi-GPU training, but one is inability to run a distillation script and the other is wrong LR scheduler step count. Different bugs." + "reason": "Both involve multi-GPU examples, but one cannot run and the other has an incorrect scheduler step count. Different training bugs." }, { "left": "issue:4534", "right": "issue:9841", "accept": false, - "reason": "MPS validation crash in SDXL dreambooth versus TypeError when use_dora is false. Same training area, but unrelated errors." - }, - { - "left": "issue:3582", - "right": "issue:5619", - "accept": false, - "reason": "ControlNet inpaint strength=1 bug versus AutoencodeTiny/img2img encode issue. Different pipelines and code paths." + "reason": "SDXL dreambooth validation crash on mps and SDXL LoRA type error with use_dora=False are different failures in different execution paths." } ] }, @@ -12575,7 +12584,6 @@ "issue:10952", "issue:11209", "issue:11224", - "issue:11247", "issue:11285", "issue:11702", "issue:1214", @@ -12588,6 +12596,7 @@ "issue:2856", "issue:3382", "issue:3481", + "issue:3582", "issue:3619", "issue:3774", "issue:3910", @@ -12598,6 +12607,7 @@ "issue:4476", "issue:4736", "issue:5259", + "issue:5619", "issue:6130", "issue:6176", "issue:6226", @@ -12612,10 +12622,10 @@ "issue:8514", "issue:9179", "issue:9180", - "issue:9911", "issue:9962" ], "soft_pairs": [ + "issue:3582|issue:5619", "issue:3619|issue:771", "issue:1979|issue:2856", "issue:10055|issue:7310", @@ -12640,203 +12650,199 @@ "issue:11224|issue:6130", "issue:10266|issue:10738", "issue:3382|issue:4736", - "issue:1931|issue:234", - "issue:11247|issue:9911" + "issue:1931|issue:234" ], - "prepared_review_unit_hash": "367e261684105b00f645bfa6923954135066cc7c77a13b16b66edf02e8203b73", + "prepared_review_unit_hash": "eb8fbc1e0e9e88ab8ccfb0df70cb0dd9818349af7b968c3838bb7833831fe5fe", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 26, - "serialized_chars": 33831, - "estimated_input_tokens": 8458, - "estimated_eval_tokens": 17172 + "serialized_chars": 33814, + "estimated_input_tokens": 8454, + "estimated_eval_tokens": 17164 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a mixed cluster of unrelated issues, with a few clear duplicate-like pairs around SD/LoRA conversion, LoRA loading, and SDXL+ControlNet training results. No PRs are present.", - "confidence": 0.77, - "canonical_issue_reason": "issue:6130 is the best canonical target: it is the earliest and most-discussed generic \"Error while loading Lora\" report, and it matches the later exact-title reports more closely than the more specific LoRA failures.", + "summary": "This cluster is mostly a loose collection of Diffusers issues around training, LoRA loading, conversion, and a few scheduler/runtime bugs. The only clear duplicate pair is the repeated generic \"Error in loading Lora\" reports; the rest are thematically similar but not the same underlying defect.", + "confidence": 0.66, + "canonical_issue_reason": "issue:11224 is the clearest representative of the repeated generic LoRA-loading failure reports, with the same broad title as issue:11209 and slightly richer discussion activity.", "canonical_pr_reason": null, - "best_issue_reason": "issue:6130 is the strongest overall representative of the repeated LoRA-loading problem in this set; it is broad enough to anchor the duplicate reports and has the most discussion among the exact-match candidates.", + "best_issue_reason": "issue:11224 is the best cluster anchor because it is a broad, user-facing LoRA loading failure that matches multiple nearby reports, while still being specific enough to represent the duplicate family.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3582", + "right": "issue:5619", + "accept": false, + "reason": "Different problems: ControlNet inpaint strength handling vs LCM img2img AutoencodeTiny encode behavior." + }, { "left": "issue:3619", "right": "issue:771", "accept": false, - "reason": "Different subsystems and failure modes: Text2Video-Zero attention vs DreamBooth bf16 precision." + "reason": "Unrelated runtime areas: Text2Video-Zero CrossFrameAttention vs BF16 dreambooth precision support." }, { "left": "issue:1979", "right": "issue:2856", - "accept": true, - "reason": "Same underlying conversion problem: both are about converting Stable Diffusion models to diffusers and failing during that path." + "accept": false, + "reason": "Both are conversion-related, but one is a general SD-to-diffusers conversion failure and the other is a specific SD conversion script issue; not clearly the same bug." }, { "left": "issue:10055", "right": "issue:7310", "accept": false, - "reason": "Both are training-script problems, but for different models/scripts and different concrete failures." + "reason": "Both mention training scripts, but one is SD3 ControlNet training and the other is SDXL LoRA defaults; different code paths and failure modes." }, { "left": "issue:4340", "right": "issue:6176", "accept": false, - "reason": "LoRA-related, but one is SDXL text-encoder training and the other is loading LoRA weights after dreambooth training; not the same bug." + "reason": "Both concern LoRA training/loading, but one is SDXL LoRA training with text encoder and the other is loading LoRA weights after DreamBooth LoRA training; not the same defect." }, { "left": "issue:11285", "right": "issue:6226", - "accept": true, - "reason": "Both point to conversion-script compatibility issues that break downstream loading; likely the same stable-diffusion<->diffusers conversion bug." + "accept": false, + "reason": "One is about conversion value errors between original Stable Diffusion and diffusers, the other about loading errors caused by a conversion script; related area but not the same underlying issue." }, { "left": "issue:4073", "right": "issue:9180", "accept": false, - "reason": "Unrelated runtime errors in different code paths: safety checker astype vs missing _execution_device." + "reason": "Different failure classes: DeepFloyd IF safety checker attribute error vs pipeline missing _execution_device." }, { "left": "issue:13162", "right": "issue:13484", "accept": false, - "reason": "Both mention Flux LoRA support/loading, but they are different model variants and not clearly the same concrete defect." + "reason": "Both are Flux LoRA support/load issues, but they appear to target different models and loading scenarios; too broad to treat as the same concrete bug." }, { "left": "issue:5259", "right": "issue:7708", "accept": false, - "reason": "Both are missing-file errors, but for different expected files and different model assets." + "reason": "Both are missing-file load errors, but one is generic model_index.json missing and the other is a tokenizer/config.json path issue for SDXL." }, { "left": "issue:1214", "right": "issue:7178", "accept": false, - "reason": "Completely different areas: half-precision CLIP text model vs video export behavior." + "reason": "Half-precision CLIPTextModel failure is unrelated to export_to_video behavior." }, { "left": "issue:10421", "right": "issue:4468", "accept": false, - "reason": "Both involve multi-GPU behavior, but one is a CPU memory leak and the other is a performance difference; not the same bug." + "reason": "Both involve multi-GPU behavior, but one is a CPU memory leak during pipeline migration and the other is differing inference performance on multi-GPU." }, { "left": "issue:10952", "right": "issue:3910", "accept": false, - "reason": "Both are LoRA loading failures, but one is a specific adapter-name mismatch and the other is a generic statement about certain LoRAs not loading." + "reason": "Both are LoRA-related, but one is a specific adapter-name mismatch in SDXL long-prompt weighting and the other is a generic \"certain LoRA will not load\" report." }, { "left": "issue:1742", "right": "issue:3481", "accept": false, - "reason": "Different concerns: prompt-list inference speed vs image output differences across inference steps." + "reason": "Performance variability vs output-dissimilarity across inference steps are different issues despite both mentioning prompt/inference behavior." }, { "left": "issue:4086", "right": "issue:7483", "accept": false, - "reason": "Different operations and models: SDXL ControlNet training vs LoRA-to-diffusers conversion for SDXL." + "reason": "Both are about ControlNet/SDXL training or conversion, but one is SDXL ControlNet training not working and the other is LoRA-to-Diffusers conversion for SDXL 1.0 base models." }, { "left": "issue:4476", "right": "issue:9179", - "accept": true, - "reason": "Both describe SDXL+ControlNet training producing poor/unexpected results on the same general setup, likely the same underlying training issue." + "accept": false, + "reason": "Both mention SDXL+ControlNet, but one is fill50k results not matching expectations and the other is training results generally not making sense; too broad to merge." }, { "left": "issue:7056", "right": "issue:8514", "accept": false, - "reason": "Different errors in different fine-tuning scenarios; no clear shared code-path." + "reason": "Both are training errors, but one is SDXL text-to-image LoRA with train_text_encoder and the other is EMAModel causing InstructP2P parallel fine-tuning failure." }, { "left": "issue:3382", "right": "issue:3774", "accept": false, - "reason": "Both are distributed-training memory complaints, but one is multi-GPU OOM and the other is zero3 not saving memory; not the same concrete issue." + "reason": "Multi-GPU OOM in text-to-image training and ZeRO3 memory not saving are related only at a high level; different training setups and symptoms." }, { "left": "issue:10823", "right": "issue:6536", "accept": false, - "reason": "Different symptoms and workflows: missing inference files vs UNet3DConditionModel OOM during fine-tuning." + "reason": "Different problems: missing inference files from fine-tuning script vs UNet3DConditionModel fine-tuning out-of-memory." }, { "left": "issue:6761", "right": "issue:9962", "accept": false, - "reason": "Both are unexpected keyword-argument errors, but against different APIs and unrelated call sites." + "reason": "Different API breakages: unexpected keyword on VaeImageProcessor.preprocess vs unexpected keyword use_cuda_graph." }, { "left": "issue:11209", "right": "issue:11702", "accept": false, - "reason": "Both are LoRA loading errors, but 11702 is a specific AccVid key error and 11209 is a generic loading failure." + "reason": "Both are LoRA loading issues, but 11702 is an AccVid-specific key error with diffusers; not the same as the generic loading failure." }, { "left": "issue:11209", "right": "issue:6130", - "accept": true, - "reason": "Same generic LoRA-loading failure; the titles are effectively the same and 6130 is the clearest duplicate target." + "accept": false, + "reason": "Same broad theme, but 6130 is an older generic loading issue with no evidence it is the exact same adapter/key failure as 11209." }, { "left": "issue:11224", "right": "issue:11702", "accept": false, - "reason": "Generic LoRA load error vs specific AccVid LoRA key error; not clearly the same bug." + "reason": "Generic LoRA loading failure vs AccVid LoRA key error; different concrete failure modes." }, { "left": "issue:11224", "right": "issue:6130", - "accept": true, - "reason": "Same generic LoRA-loading failure; 11224 reads like a later duplicate of the 6130 report." + "accept": false, + "reason": "Both are generic LoRA-loading complaints, but the evidence is too weak to say they are the same underlying bug." }, { "left": "issue:10266", "right": "issue:10738", "accept": false, - "reason": "Both mention index out-of-bounds, but one is UniPC+FlowMatch and the other is a scheduler sigma issue; too little evidence they are the same defect." + "reason": "Both are out-of-bounds scheduler issues, but one is UniPC with FlowMatch and the other is a scheduler sigma index error; not enough to identify a single bug." }, { "left": "issue:3382", "right": "issue:4736", "accept": false, - "reason": "Both are training-related OOM-style reports, but one is multi-GPU memory use and the other includes invalid-value casting in a different script." + "reason": "Different training failures: multi-GPU OOM in text-to-image vs SDXL LoRA OOM plus invalid cast." }, { "left": "issue:1931", "right": "issue:234", "accept": false, - "reason": "Related half-precision issues, but one is SD2 half/float mismatch and the other is a pipeline half-precision general failure." - }, - { - "left": "issue:11247", - "right": "issue:9911", - "accept": false, - "reason": "Both involve Flux/controlnet multi-GPU behavior, but one is a distribution attribute issue and the other is a multi-controlnet layer-length error." + "reason": "Both concern half precision, but one is a Stable Diffusion 2 dtype mismatch and the other is a generic pipeline half-precision failure." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays mostly conservative. The chosen canonical issue, issue:6130, is well supported by the repeated generic LoRA-loading reports, and the accepted soft pairs are plausible matches rather than overclaims. No PRs are present, and the cluster summary appropriately notes a mixed set with a few likely duplicate-like groups." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -12859,6 +12865,7 @@ "issue:1070", "issue:10860", "issue:11160", + "issue:11247", "issue:11396", "issue:12108", "issue:12141", @@ -12875,7 +12882,6 @@ "issue:3191", "issue:3235", "issue:3812", - "issue:4478", "issue:4551", "issue:4619", "issue:4768", @@ -12897,19 +12903,19 @@ "issue:8833", "issue:8922", "issue:9343", - "issue:9476", "issue:9500", "issue:9548", + "issue:9911", "issue:9970" ], "soft_pairs": [ + "issue:11247|issue:9911", "issue:11160|issue:12896", "issue:4911|issue:7228", "issue:1988|issue:4619", "issue:2578|issue:5226", "issue:10169|issue:3812", "issue:4768|issue:5884", - "issue:4478|issue:9476", "issue:12141|issue:7057", "issue:7057|issue:7666", "issue:13300|issue:8668", @@ -12929,184 +12935,184 @@ "issue:10860|issue:11396", "issue:5886|issue:7974" ], - "prepared_review_unit_hash": "16246af649ec749cd3f91429ff8ae469df78658be913815ac49d76acaaf0e9ba", + "prepared_review_unit_hash": "b7e2aed33fba7a1bfe6a8fadaef0b31b21ee0ed2659b2814cf8e43f56685ac5f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34484, - "estimated_input_tokens": 8621, - "estimated_eval_tokens": 17498 + "serialized_chars": 34435, + "estimated_input_tokens": 8609, + "estimated_eval_tokens": 17474 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly separate bug reports that only overlap at a high level (black/grey images, dtype mismatches, config/load failures, training issues). I do not see any strong duplicate pairs among the soft edges, so the cluster should not be collapsed aggressively.", - "confidence": 0.82, - "canonical_issue_reason": "issue:9500 is the most detailed and discussion-heavy report here, and it best serves as a representative anchor for the repeated DreamBooth/Flux training-failure theme among the training-related issues.", + "summary": "This cluster is very loose: most items are distinct model-loading, training, or GPU/device-specific bugs rather than true duplicates. No soft pair looks safe to merge as the same underlying issue.", + "confidence": 0.37, + "canonical_issue_reason": "issue:9500 is the most substantial and actionable report in the set, with the richest reproduction details and highest discussion activity, so it is the best representative anchor for this heterogeneous cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9500 has the richest repro/context and the highest activity, so it is the most useful single issue to keep as the representative bug report in this set.", + "best_issue_reason": "issue:9500 is the strongest standalone issue for triage: detailed, active, and clearly described, even though the surrounding items are not true duplicates.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11247", + "right": "issue:9911", + "accept": false, + "reason": "Both concern Flux/controlnet multi-device or multi-controlnet behavior, but they describe different failure modes and likely different code paths." + }, { "left": "issue:11160", "right": "issue:12896", "accept": false, - "reason": "Both mention WAN 2.1/2.2 I2V, but one is about inconsistent inference results and the other is a CUDA-vs-MPS failure; different bug modes and likely different causes." + "reason": "Same WAN i2v family, but one is inconsistent inference output and the other is a CUDA-vs-MPS execution failure; too different to treat as one bug." }, { "left": "issue:4911", "right": "issue:7228", "accept": false, - "reason": "Both are load-time errors, but one is about custom text_encoder loading from_single_file and the other is a traced-model dtype attribute error; not the same failure." + "reason": "Both are loading-related, but one is about missing custom text_encoder in from_single_file and the other is a traced-model dtype attribute error." }, { "left": "issue:1988", "right": "issue:4619", "accept": false, - "reason": "Both are mixed-precision type-mismatch errors, but they affect different training scripts and model variants; too broad to treat as one concrete bug." + "reason": "Both mention fp16/bf16 type mismatch during training, but they are different training scripts and not clearly the same concrete bug." }, { "left": "issue:2578", "right": "issue:5226", "accept": false, - "reason": "Editable install problems and LoRA-loading failure in SDXL adapter pipeline are unrelated issues." + "reason": "Editable install failure and LoRA loading failure are unrelated problems." }, { "left": "issue:10169", "right": "issue:3812", "accept": false, - "reason": "Missing config.json and missing {'controlnet'} config entry are different configuration-loading problems." + "reason": "Missing config.json and missing controlnet key in config are different config-loading issues." }, { "left": "issue:4768", "right": "issue:5884", "accept": false, - "reason": "One is a checkpoint-saving feature request for multiple T2I adapters, the other is adding IP-Adapter support across pipelines; not the same change." - }, - { - "left": "issue:4478", - "right": "issue:9476", - "accept": false, - "reason": "Both involve dtype/loading problems, but SDXL inference bias-type mismatch and FLUX loading with ignore_mismatched_sizes are distinct failures." + "reason": "One is about checkpointing multiple T2I adapters; the other is a feature request for broader IP-Adapter support." }, { "left": "issue:12141", "right": "issue:7057", "accept": false, - "reason": "WanVACEPipeline VAE dtype incompatibility and a generic checkpoint loading failure are not the same bug." + "reason": "VAE dtype compatibility and checkpoint/model loading failure are not the same underlying bug." }, { "left": "issue:7057", "right": "issue:7666", "accept": false, - "reason": "A checkpoint load failure and a custom pipeline from_single_file failure are separate issues with different code paths." + "reason": "Different loading paths and different errors; no evidence they share the same root cause." }, { "left": "issue:13300", "right": "issue:8668", "accept": false, - "reason": "CPU generation failures on SD3.5/FLUX and an image-size alignment bug in SD3 img2img are unrelated." + "reason": "CPU generation failure for SD3.5/FLUX is unrelated to the image-size alignment bug in SD3 img2img." }, { "left": "issue:9500", "right": "issue:9970", "accept": false, - "reason": "Both are DreamBooth training reports, but one is about checkpoint saving in FLUX and the other is an SDXL LoRA training failure; different code paths." + "reason": "Both are DreamBooth training failures, but one is Flux checkpoint saving and the other is SDXL LoRA script failure; too broad to merge." }, { "left": "issue:2991", "right": "issue:817", "accept": false, - "reason": "Both are mixed-precision example bugs, but they affect different examples and likely different call sites." + "reason": "Both are mixed-precision example failures, but they affect different examples and are not clearly the same code-path defect." }, { "left": "issue:12771", "right": "issue:9343", "accept": false, - "reason": "A scheduler out-of-bounds error and a FLUX loading/mismatched-sizes error are unrelated." + "reason": "Scheduler out-of-bounds with Karras sigmas is unrelated to FLUX loading with mismatched sizes." }, { "left": "issue:3235", "right": "issue:4885", "accept": false, - "reason": "Custom Diffusion weights with biases failing to load and SDXL ControlNet Img2Img LoRA loading are different loading features." + "reason": "Custom Diffusion weights with biases failing to load is not the same as SDXL ControlNet Img2Img LoRA loading." }, { "left": "issue:6082", "right": "issue:7854", "accept": false, - "reason": "Both are Half/Float runtime mismatches, but the titles indicate different contexts and not a clearly shared concrete bug." + "reason": "Both are half/float mismatch reports, but they are separate reports without enough evidence of the same concrete failure." }, { "left": "issue:12108", "right": "issue:12359", "accept": false, - "reason": "A scheduler/flow-matching pipeline break and a documentation bug about guidance_scale are not duplicates." + "reason": "Pipeline breakage with flow-matching schedulers is not the same as the Chroma documentation issue about guidance_scale." }, { "left": "issue:10040", "right": "issue:8922", "accept": false, - "reason": "Grey-image generation in Stable Diffusion 3 and black-image output from a quantized SD3 single-file setup are similar symptoms but not the same underlying issue." + "reason": "Grey images and black images are similar symptoms, but they are not enough to conclude the same root cause." }, { "left": "issue:3191", "right": "issue:4936", "accept": false, - "reason": "Dreambooth+LoRA broken and Dreambooth LoRA resume broken since 0.19.0 are different bugs in the same area." + "reason": "Dreambooth+LoRA being broken and resume in Dreambooth LoRA training are different training problems." }, { "left": "issue:1565", "right": "issue:4551", "accept": false, - "reason": "GPU selection/OOM behavior and a DreamBooth crash on SDXL are unrelated." + "reason": "GPU selection/OOM on SD2 is unrelated to SDXL Dreambooth crashing." }, { "left": "issue:9500", "right": "issue:9548", "accept": false, - "reason": "These are both Flux DreamBooth reports, but one is checkpoint saving and the other is a broader follow-up issue; not enough to merge as the same bug." + "reason": "Both involve Flux DreamBooth/LoRA, but one is checkpoint saving failure and the other is a broader 'still issue' follow-up; not the same concrete bug." }, { "left": "issue:579", "right": "issue:8833", "accept": false, - "reason": "Textual Inversion config.json missing and SD3 medium model missing config.json are different model-loading cases." + "reason": "Both mention missing config.json, but one is textual inversion and the other is a model repo/config packaging issue." }, { "left": "issue:10436", "right": "issue:1070", "accept": false, - "reason": "An inpainting model failing to load and a Stable Diffusion inpainting bug are too different in scope and symptom." + "reason": "Inpainting model not loading and inpainting faces bug are different issues." }, { "left": "issue:1251", "right": "issue:2495", "accept": false, - "reason": "Both produce black images on MPS, but they are different pipelines (RePaint vs Upscale) and there is no evidence of the same code-path." + "reason": "Black output on MPS for RePaint and black output on MPS for upscale pipeline may look similar, but they affect different pipelines and code paths." }, { "left": "issue:10860", "right": "issue:11396", "accept": false, - "reason": "Both mention LoRA compatibility with ComfyUI, but they concern different model families and different conversion/loading questions." + "reason": "Both involve LoRA interoperability with ComfyUI, but one is about using a trained LoRA and the other is about converting a HiDream LoRA format." }, { "left": "issue:5886", "right": "issue:7974", "accept": false, - "reason": "IP-Adapter face-model support and inpaint-with-IP-Adapter behavior are separate feature/bug reports." + "reason": "Face model support request and an inpaint bug with IP-Adapter face+style are related only broadly; they are not the same bug or change." } ] }, @@ -13210,10 +13216,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -13485,10 +13491,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -13758,10 +13764,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -14047,10 +14053,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -14327,10 +14333,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -14605,10 +14611,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -14888,10 +14894,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -15166,10 +15172,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -15409,10 +15415,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -15685,10 +15691,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -15965,10 +15971,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -16241,10 +16247,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -16510,10 +16516,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -16781,10 +16787,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -17061,10 +17067,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -17338,10 +17344,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -17614,10 +17620,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -17884,10 +17890,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -18154,10 +18160,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -18429,10 +18435,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -18713,10 +18719,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -18985,10 +18991,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -19255,10 +19261,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -19524,10 +19530,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -19798,10 +19804,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -20062,10 +20068,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -20332,10 +20338,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -20625,10 +20631,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -20643,174 +20649,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly heterogeneous: a few pairs are true near-duplicates, but most are only loosely related by subsystem or phrasing. The strongest duplicate clusters are DPM++/SDXL artifacts, img2img/inpainting breakage, and from_single_file offline loading errors.", - "confidence": 0.87, - "canonical_issue_reason": "issue:5689 is the clearest active report in the set and the best anchor for the DPM++/SDXL residual-noise/artifact duplicate pair.", + "summary": "This set is mostly heterogeneous, but a few issue pairs look like real duplicates: from_single_file offline-loading errors, img2img/inpainting pipeline failure, and a DPM++/scheduler add_noise bug. Most other pairs are only broadly similar and should stay separate.", + "confidence": 0.79, + "canonical_issue_reason": "No single issue cleanly represents the whole set because it spans unrelated topics; within the strongest duplicate-like scheduler pair, issue 6232 is the best umbrella since it states the broader add_noise failure while 6069 is a specific t=0 case.", "canonical_pr_reason": null, - "best_issue_reason": "issue:5689 is the most specific, actionable issue here and has the strongest duplicate-like counterpart among the provided items.", + "best_issue_reason": "Issue 6232 is the best global issue candidate among the duplicate-like pairs because it is the broadest, most reusable description of the scheduler bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10049", "right": "issue:8975", "accept": false, - "reason": "Both are import errors, but they fail on different symbols and likely different code paths." + "reason": "Both are import errors, but they involve different missing symbols/modules and no shared code path is evident." }, { "left": "issue:7931", "right": "issue:9202", "accept": false, - "reason": "Both involve IP-Adapter/SDXL, but one is ControlNet inpaint with image_embeds and the other is PAG/FaceID Plus V2; too different to merge." + "reason": "Both mention IP-Adapter, but they target different pipelines and different failure modes." }, { "left": "issue:5433", "right": "issue:5689", - "accept": true, - "reason": "Same underlying SDXL/DPM++ artifact problem: one reports visual artifacts without the refiner, the other residual noise from DPM++ schedulers." + "accept": false, + "reason": "Very similar DPM++/SDXL symptom, but one is artifacting without the refiner and the other is residual noise; not enough to confirm the same bug." }, { "left": "issue:4731", "right": "issue:5227", "accept": false, - "reason": "Both mention torch.compile, but one is multi-ControlNet SDXL and the other is LoRA loading; different failures." + "reason": "Both involve torch.compile, but one is multi-ControlNet and the other is load_lora_weight, so the concrete failures differ." }, { "left": "issue:5314", "right": "issue:9148", "accept": false, - "reason": "Both are inpainting-related feature requests, but they target different pipelines and models." + "reason": "These are both inpainting-related feature requests, but for different models/pipelines and not the same change." }, { "left": "issue:8493", "right": "issue:8723", "accept": false, - "reason": "Both are SD3 model-loading problems, but one is SD3Transformer2DModel state_dict loading and the other is SD3ControlNetModel initialization." + "reason": "Both are SD3 model issues, but one is state_dict loading for SD3Transformer2DModel and the other is SD3ControlNetModel initialization." }, { "left": "issue:4185", "right": "issue:4206", "accept": false, - "reason": "Both are ControlNet/SDXL issues, but one is black-image inference and the other is training breakage; not the same bug." + "reason": "One is SDXL ControlNet inference producing black images; the other is SDXL ControlNet training broken on main, so the code paths differ." }, { "left": "issue:7690", "right": "issue:7716", "accept": false, - "reason": "Both are model-loading errors, but one is missing tokenizer files and the other is missing config.json; different root causes." + "reason": "Both are loading failures, but one is tokenizer lookup and the other is a missing config.json in a local model directory." }, { "left": "issue:1572", "right": "issue:337", "accept": true, - "reason": "Same img2img/inpainting malfunction in Stable Diffusion; the titles describe the same broken workflow." + "reason": "Both report img2img/inpainting failing in Stable Diffusion-style pipelines, with 1572 being the broader pipeline-specific version." + }, + { + "left": "issue:2984", + "right": "issue:7243", + "accept": false, + "reason": "Completely different model/paper requests with no shared bug or implementation target." + }, + { + "left": "issue:3432", + "right": "issue:7330", + "accept": false, + "reason": "Both are training/runtime failures, but one is CPU Half convolution support and the other is FP16 gradient unscaling in LoRA training." }, { "left": "issue:3706", "right": "issue:3728", "accept": false, - "reason": "Different new-model/pipeline requests with no concrete shared bug or implementation." + "reason": "Different new-model/pipeline requests; same general area but not the same feature." }, { "left": "issue:3728", "right": "issue:4779", "accept": false, - "reason": "Both are new pipeline requests, but for unrelated tasks/models." + "reason": "Text-to-image shifted diffusion and image-to-video ModelScope are unrelated pipeline requests." }, { "left": "issue:6914", "right": "issue:8827", "accept": false, - "reason": "Both involve IP-Adapter FaceID, but one is an offload-induced AttributeError and the other is a generic usage error; not enough to treat as the same bug." + "reason": "The FaceID/IP-Adapter context overlaps, but the specific error and triggering condition are not clearly the same." }, { "left": "issue:3094", "right": "issue:3453", "accept": false, - "reason": "Both are training/runtime errors, but one is CUDA OOM and the other is a BFloat16 kernel limitation." + "reason": "Different training problems with different root causes: OOM vs unsupported bfloat16 CUDA op." }, { "left": "issue:10940", "right": "issue:7681", "accept": false, - "reason": "Both are import failures, but they reference unrelated missing dependencies and unrelated traceback roots." + "reason": "Both are import-time failures, but they point to unrelated missing dependencies/environment problems." }, { "left": "issue:2984", - "right": "issue:7243", - "accept": false, - "reason": "Both are diffusion-model research/proposal issues, not the same bug or change." - }, - { - "left": "issue:3432", - "right": "issue:7330", - "accept": false, - "reason": "Both affect text-to-image training/examples, but one is CPU half-precision support and the other is FP16 gradient scaling." - }, - { - "left": "issue:12412", - "right": "issue:13232", - "accept": false, - "reason": "Both are support requests for different models, so they are not duplicates." - }, - { - "left": "issue:4305", - "right": "issue:4356", - "accept": false, - "reason": "One is a torch.compile graph-break regression; the other is a missing dummy module import error." - }, - { - "left": "issue:1677", - "right": "issue:993", + "right": "issue:5889", "accept": false, - "reason": "Both mention type mismatches, but one is wrong call data type and the other is fp16 dtype mismatch; too vague to merge." + "reason": "A research/model request versus a stable video diffusion feature request; not duplicates." }, { - "left": "issue:4828", - "right": "issue:7209", + "left": "issue:6069", + "right": "issue:6232", "accept": true, - "reason": "Same from_single_file loading failure path: both describe connection/HTTPS errors when loading a local model, especially offline." + "reason": "These are the same scheduler.add_noise failure family; 6232 is the broader description and 6069 is a specific t=0 manifestation." }, { "left": "issue:4042", "right": "issue:5314", "accept": false, - "reason": "Related around reference/inpainting, but one asks how to combine features while the other proposes a new pipeline." + "reason": "Both mention reference/inpainting, but one asks how to combine features while the other requests a dedicated pipeline." }, { "left": "issue:202", "right": "issue:338", "accept": false, - "reason": "Both are test-addition tasks, but for different files/features." + "reason": "Both are test-related, but they target different test additions and do not describe the same bug or change." }, { "left": "issue:202", "right": "issue:339", "accept": false, - "reason": "Same as above: different test targets, not the same change." + "reason": "Same as above: different test targets, not a duplicate." }, { "left": "issue:8589", "right": "issue:8828", "accept": false, - "reason": "Both are SD3 training issues, but one is prior-preservation support and the other is a 16GB training error." + "reason": "Both are SD3 training issues, but one is prior-preservation behavior and the other is a general 16GB training error." + }, + { + "left": "issue:4828", + "right": "issue:7209", + "accept": true, + "reason": "Both describe from_single_file offline/local loading hitting an unwanted network/connection path, so they look like the same underlying bug." }, { "left": "issue:2984", - "right": "issue:5889", + "right": "issue:7243", "accept": false, - "reason": "Unrelated model/paper issue versus Stable Video Diffusion request." + "reason": "Unrelated feature requests; no shared concrete problem." }, { - "left": "issue:6069", - "right": "issue:6232", + "left": "issue:4305", + "right": "issue:4356", + "accept": false, + "reason": "A torch.compile graph-break report and a missing dummy module import error are different issues." + }, + { + "left": "issue:1677", + "right": "issue:993", "accept": false, - "reason": "Both concern scheduler.add_noise, but the affected scheduler classes and failure descriptions differ." + "reason": "Both are dtype-related, but one is a wrong-call-data-type problem and the other is a stable diffusion fp16 mismatch." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet, and the accepted soft-edge pairs are reasonably supported by the issue titles: DPM++/SDXL artifact reports, img2img/inpainting breakage, and from_single_file offline loading errors. The false verdicts are also appropriately conservative, mostly distinguishing different symbols, pipelines, or root causes." + "feedback": "The summary is grounded in the packet and stays conservative overall. The accepted soft-edge pairs are well supported by the issue titles, especially 6069/6232, 4828/7209, and 1572/337. The rejected pairs are also reasonably distinguished by different pipelines, failure modes, or feature requests. No overstatement stood out." }, "evaluator_used": true, "retried": false, @@ -20909,10 +20915,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -20927,162 +20933,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous set of unrelated issues. The only clear duplicate-like pair is the Flux inpainting request. The rest are broader topic similarities at best, not the same concrete bug/change.", - "confidence": 0.84, - "canonical_issue_reason": "issue:9148 is the clearest, most general formulation of the only confirmed duplicate pair (Flux inpainting request), so it works best as the canonical issue anchor.", + "summary": "Only one soft edge looks like a true duplicate: the LTX single-file loading/support reports. The rest are title-level similarities across different models, pipelines, or unrelated bugs/features and should stay separate.", + "confidence": 0.62, + "canonical_issue_reason": "issue:11258 is the broader LTX single-file support request and best captures the underlying problem behind the missing-file error in issue:10321.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9148 is the best issue representative here: concise, explicit, and directly matches the duplicate request in issue:9264.", + "best_issue_reason": "issue:11258 is the clearest representative of the LTX single-file issue family; issue:10321 reads as a specific failure symptom of that same gap.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10798", "right": "issue:10833", "accept": false, - "reason": "Both involve Flux + bitsandbytes, but one is a general device-combination bug and the other is a specific FluxFill matmul shape error; not the same concrete bug." + "reason": "Both mention Flux + bitsandbytes quantization, but they describe different bugs: one is a device-combination issue, the other a FluxFillPipeline matmul shape failure." }, { "left": "issue:3638", "right": "issue:3639", "accept": false, - "reason": "Different new pipeline requests with distinct functionality: Prompt-Free Diffusion vs GlyphControl." + "reason": "These are separate new-pipeline proposals (Prompt-Free Diffusion vs GlyphControl), not the same feature or bug." }, { "left": "issue:8503", "right": "issue:8839", "accept": false, - "reason": "Both are SD3-related errors, but one is about gradient checkpointing crashes and the other is a LoRA weight error." + "reason": "Both are SD3-related, but one is a gradient-checkpointing crash and the other is an SD3 LoRA weight error; different code paths and symptoms." }, { "left": "issue:6149", "right": "issue:6415", "accept": false, - "reason": "Both mention IP-Adapter face variants, but one is a load error for a specific model and the other is lack of support for FaceID/FaceID Plus." + "reason": "Different IP-Adapter topics: one is a specific full-face_sd15 load error, the other is support for FaceID/FaceID Plus." }, { "left": "issue:7330", "right": "issue:7646", "accept": false, - "reason": "Same training script, but different problems: FP16 unscale failure vs dataloader worker settings." + "reason": "Same training script, but unrelated problems: FP16 gradient unscale error versus dataloader worker configuration." }, { "left": "issue:3357", "right": "issue:5227", "accept": false, - "reason": "Both touch torch.compile, but the failing code paths and concrete errors are different." + "reason": "Both involve torch.compile, but the underlying failures differ: a Dynamo hasattr limitation versus load_lora_weight incompatibility." }, { "left": "issue:1076", "right": "issue:1962", "accept": false, - "reason": "Unrelated topics: Stable Diffusion model question vs video generation with Stable Diffusion." + "reason": "These are unrelated questions about stable diffusion and video generation, not the same underlying bug or change." }, { "left": "issue:7799", "right": "issue:9202", "accept": false, - "reason": "Both are IP-Adapter shape/mismatch issues, but they affect different pipelines and adapters." + "reason": "Both are IP-Adapter shape mismatches, but they affect different pipelines/models and appear to be different failures." }, { "left": "issue:7219", "right": "issue:9303", "accept": false, - "reason": "Both are upscaler-related feature requests, but for different models and use cases." + "reason": "Both are upscaler requests, but SUPIR and VEnhancer for CogVideoX-5b are distinct feature requests." }, { "left": "issue:3357", "right": "issue:4731", "accept": false, - "reason": "Both are torch.compile issues, but one is a generic dynamo hasattr failure and the other is a multi-ControlNet SDXL compile failure." + "reason": "Both mention torch.compile, but one is a TensorVariable hasattr error and the other is a multi-controlnet SDXL compile failure." }, { "left": "issue:9171", "right": "issue:9332", "accept": false, - "reason": "Both are loading regressions, but one is from_single_file SD1.5 v_prediction handling and the other is incorrect model loading to identical weights." + "reason": "Different loading bugs: from_single_file regression for SD 1.5 v_prediction versus from_pretrained loading the wrong identical weights." }, { "left": "issue:12369", "right": "issue:13351", "accept": false, - "reason": "Different feature requests in the Wan family, not the same pipeline or change." + "reason": "These are unrelated feature requests around different Wan/Fun ControlNet topics, not a duplicate bug/change." }, { "left": "issue:9148", "right": "issue:9264", - "accept": true, - "reason": "Same underlying request: adding a Flux inpainting pipeline/model support." + "accept": false, + "reason": "Both concern Flux inpainting, but one requests adding the pipeline and the other asks for an inpainting model for flux; not the same artifact." }, { "left": "issue:5628", "right": "issue:5687", "accept": false, - "reason": "Both involve Karras-sigma timestep behavior, but the reported bugs are different enough to keep separate." + "reason": "Both involve Karras sigmas/timesteps, but the concrete bugs differ: denoising offset in LMS scheduler versus duplicate timesteps breaking Img2Img." }, { "left": "issue:4631", "right": "issue:5075", "accept": false, - "reason": "Different scheduler/model-path failures: one is a scalar conversion error, the other a refiner embedding matmul mismatch." + "reason": "Both are matmul/runtime shape errors, but they occur in different scheduler/pipeline contexts and are not clearly the same code-path bug." }, { "left": "issue:6561", "right": "issue:8827", "accept": false, - "reason": "One is a missing config file load error, the other is a usage error with ip-adapter-faceid_sdxl; not the same bug." + "reason": "Both are IP-Adapter loading/usage errors, but one is a missing config file and the other is an error using faceid_sdxl; different failures." }, { "left": "issue:12348", "right": "issue:12529", "accept": false, - "reason": "Both are tensor-size mismatch reports in Wan pipelines, but they involve different models and likely different code paths." + "reason": "Both are tensor-size mismatches in Wan models, but one is resolution-specific in WanImageToVideoPipeline and the other is a tiled-VAE mismatch in Wan2.2 TI2V-5B." }, { "left": "issue:3698", "right": "issue:4912", "accept": false, - "reason": "Both are about conversion to diffusers, but one targets an inpainting model and the other custom SDXL checkpoints." + "reason": "Both are about conversion to diffusers, but one targets inpainting models and the other custom CivitAI SDXL checkpoints; not the same change." }, { "left": "issue:9025", "right": "issue:9038", "accept": false, - "reason": "Same Flax pipeline family, but one asks about from_single_file on TPU and the other about prompt weighting." + "reason": "These are unrelated FlaxStableDiffusionPipeline questions: single-file loading on Kaggle TPU versus prompt weighting." }, { "left": "issue:2674", "right": "issue:3124", "accept": false, - "reason": "Different problems: missing safetensors requirement for LoRA vs model conversion failure." + "reason": "Different model-conversion/LoRA dependency issues; they do not describe the same bug." }, { "left": "issue:10321", "right": "issue:11258", - "accept": false, - "reason": "Both involve LTX loading, but one is an error report about missing model files and the other is a single-file support request; not clearly the same concrete bug." + "accept": true, + "reason": "Both point to the same LTX single-file loading gap: issue:10321 is the missing-model-file error, while issue:11258 is the broader single-file support request." }, { "left": "issue:6794", "right": "issue:7021", "accept": false, - "reason": "Both are import/environment errors, but they concern different symbols and root causes." + "reason": "These are unrelated import/environment errors: one is a training-utils import issue, the other is HF_HOME import configuration." }, { "left": "issue:7417", "right": "issue:7569", "accept": false, - "reason": "Both are Mac test failures, but they fail in different tests and likely different code paths." + "reason": "Both are Mac test failures, but they involve different tests and likely different regressions." }, { "left": "issue:1062", "right": "issue:1185", "accept": false, - "reason": "Both are Dreambooth reproducibility complaints, but one is about repeating paper results and the other about general reproducibility; too broad to merge." + "reason": "Both mention Dreambooth reproducibility, but they are separate reports with no evidence of the same underlying bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The only accepted duplicate-like pair, issue:9148/issue:9264, is well supported by the titles, and the other soft-edge rejections avoid overclaiming equivalence despite topical similarity." + "feedback": "The summary is grounded in the packet and stays conservative. The single accepted edge (issue:10321 \u2194 issue:11258) is plausibly the same LTX single-file support gap, while the rejected pairs are explained in a cautious way without overstating duplication. No mergeability issues apply here since these are issue-to-issue pairs, not PRs." }, "evaluator_used": true, "retried": false, @@ -21181,10 +21187,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -21453,10 +21459,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -21471,162 +21477,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are loose text matches rather than true duplicates. I only see two plausible duplicate families: missing training metadata/config artifacts, and Flux2-klein LoRA adapter loading.", - "confidence": 0.48, - "canonical_issue_reason": "Issue 3694 is the clearest, most detailed report in the strongest duplicate pair: it names the missing artifacts explicitly (`model_index.json` and `config.json`) and includes the training-example context.", + "summary": "Mostly unrelated issue pairs; only the model-loading slowdown reports look plausibly like the same bug. The rest share subsystem keywords but describe different failures, features, or model-specific regressions.", + "confidence": 0.46, + "canonical_issue_reason": "issue:8787 is the broadest representative for the slowdown regression and can cover the more specific SDXL loading/inference complaint in issue:10150.", "canonical_pr_reason": null, - "best_issue_reason": "3694 is the best canonical issue candidate because it gives the most complete failure description and a concrete reproduction path for the metadata/config generation bug.", + "best_issue_reason": "issue:8787 is the best global anchor because it names the generic slowdown symptom, making it the most reusable duplicate target among the candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:228", "right": "issue:2689", "accept": false, - "reason": "Both mention broken links, but one is a specific release asset and the other is a generic image-link report; not the same underlying bug." + "reason": "Both mention broken links, but one is a specific release asset and the other is a generic image-link report; not enough evidence of the same underlying defect." }, { "left": "issue:12037", "right": "issue:12047", "accept": false, - "reason": "Same product family, but one is a model compatibility issue and the other is a GGUF LoRA-fusion failure." + "reason": "Same model family and LoRA terminology, but one is compatibility/self-forcing guidance and the other is GGUF fusion failure; likely distinct code paths." }, { "left": "issue:201", "right": "issue:3406", "accept": false, - "reason": "One is a test item and the other is a real `only_cross_attention=True` runtime bug; not the same change/bug." + "reason": "Both are about attention classes, but the concrete failure differs: a test request versus a specific only_cross_attention bug." }, { "left": "issue:5886", "right": "issue:6561", "accept": false, - "reason": "Feature request for face-model support versus an IP-Adapter config-loading error." + "reason": "IP-Adapter-related, but one asks for face-model support while the other is a missing config.json load error; not the same bug." }, { "left": "issue:10674", "right": "issue:10797", "accept": false, - "reason": "Different pipelines and different offloading/GGUF contexts; no shared concrete code-path." + "reason": "Different pipelines and different offloading/GGUF behavior; too little to treat as one underlying issue." }, { "left": "issue:3382", "right": "issue:3453", "accept": false, - "reason": "Both are training failures, but one is multi-GPU OOM and the other is a BFloat16 CUDA kernel limitation." + "reason": "Both are training OOM/runtime problems, but one is multi-GPU memory pressure and the other is a BFloat16 kernel limitation." }, { "left": "issue:4305", "right": "issue:6008", "accept": false, - "reason": "Unrelated regressions: torch.compile graph breaks versus IP-Adapter/LCM-LoRA behavior." + "reason": "Shared release-regression framing only; one is a torch.compile graph break and the other is IP-Adapter face + LCM LoRA failure." }, { "left": "issue:5061", "right": "issue:5374", "accept": false, - "reason": "Both are CLI/config errors, but they point to different arguments and different failure modes." + "reason": "Both are converter argument errors, but they reference different CLI parameters and likely different call sites." }, { "left": "issue:1594", "right": "issue:1962", "accept": false, - "reason": "Both are video-generation requests, not the same underlying bug or change." + "reason": "Related to video generation, but one is an img2img feature request and the other is general video generation support." }, { "left": "issue:5219", "right": "issue:6427", "accept": false, - "reason": "Both involve conversion/loading, but they are opposite directions and different failure points." + "reason": "Both concern conversion/loading, but one is loading an SDXL safetensors export and the other is converting diffusers to .ckpt; opposite directions and different symptoms." }, { "left": "issue:8649", "right": "issue:8839", "accept": false, - "reason": "Different SD3 bugs: `num_images_per_prompt` handling versus LoRA weight loading." + "reason": "Both involve SD3, but one is a num_images_per_prompt error and the other is a LoRA weight issue." }, { "left": "issue:3728", "right": "issue:7223", "accept": false, - "reason": "Separate new-pipeline requests for different models and tasks." + "reason": "Both are new pipeline proposals, but they are different projects and not the same pipeline change." }, { "left": "issue:8503", "right": "issue:9850", "accept": false, - "reason": "Gradient-checkpointing is the common term, but one is a crash on SD3 and the other is a feature request for frozen models." + "reason": "Gradient checkpointing appears in both, but one is an SD3 crash and the other is a feature request for frozen models; not the same bug." }, { "left": "issue:12358", "right": "issue:9873", "accept": false, - "reason": "Different pipeline proposals with no shared bug/change." + "reason": "Both are pipeline-addition requests, but for different models and different feature scopes." }, { "left": "issue:10302", "right": "issue:10914", "accept": false, - "reason": "Precision-noise during FP8 inference is not the same as unintended CPU offloading." + "reason": "Both mention offloading, but one reports FP8 noise without CPU offloading while the other reports unintended CPU offloading; different problems." }, { "left": "issue:3468", "right": "issue:3694", - "accept": true, - "reason": "Same underlying training-output metadata problem: both report missing `model_index.json`/config artifacts needed for inference or loading." + "accept": false, + "reason": "Both involve missing model_index.json, but the contexts differ enough\u2014Dreambooth output generation versus a finetuning example load path\u2014to avoid merging." }, { "left": "issue:10334", "right": "issue:11229", "accept": false, - "reason": "Both mention NaNs on MPS, but they are different pipelines and likely different mechanisms." + "reason": "Both mention MPS/NaNs, but the first is a Sana MacOS regression and the second is attention slicing on SDXL; different triggers and models." }, { "left": "issue:10462", "right": "issue:9303", "accept": false, - "reason": "Same model family, but one is multi-ID customization and the other is an enhancer/upscaler feature request." + "reason": "One is a customization issue for CogVideoX-5B and the other is a request to add an upscaler/interpolator pipeline; not the same change." }, { "left": "issue:3728", "right": "issue:3826", "accept": false, - "reason": "Different new pipeline ideas: text-to-image versus audio generation." + "reason": "Both are new pipeline requests, but for different modalities and models; only a broad category match." }, { "left": "issue:4686", "right": "issue:4874", "accept": false, - "reason": "Both involve local loading of the upscale pipeline, but the specific failure descriptions differ enough to avoid merging." + "reason": "Both touch StableDiffusionUpscalePipeline local loading, but one is a from_pretrained load failure and the other is a low_res_scheduler-specific bug." }, { "left": "issue:10150", "right": "issue:8787", - "accept": false, - "reason": "Both are slowdown reports, but 10150 adds SDXL inference slowdown and is not clearly the same bug as generic model-loading slowness." + "accept": true, + "reason": "Both report slowdown in model loading/inference, and issue:8787 is the broader generic version of the same performance regression." }, { "left": "issue:4731", "right": "issue:6096", "accept": false, - "reason": "Different torch.compile failures on different models and error types." + "reason": "Both mention torch.compile, but one is a multi-ControlNet SDXL failure and the other is an SVD dynamic output shape exception." }, { "left": "issue:13311", "right": "issue:13547", - "accept": true, - "reason": "Same Flux2-klein LoRA adapter-loading failure: one gives the missing-key/prefix error, the other reports the same loading problem more generally." + "accept": false, + "reason": "Both are FLUX.2-klein LoRA loading problems, but one is a missing key-prefix association and the other is a generic adapter-loading error; likely distinct." }, { "left": "issue:2722", "right": "issue:671", "accept": false, - "reason": "Both are conversion fidelity complaints, but they target different models and different conversion workflows." + "reason": "Both concern converted models not matching, but one is ControlNet UNet conversion and the other is original Stable Diffusion conversion identity; different artifacts and failure modes." } ] }, "evaluator_result": { "accept": true, - "feedback": "The assessment is grounded in the packet and the duplicate calls are conservative. The two accepted pairs are well-supported by the titles, and the remaining pairs are reasonably rejected rather than overstated. Summary does not overclaim beyond the evidence provided." + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:10150 with issue:8787, is plausibly the same slowdown regression based on the titles, and the rejected pairs are generally justified as related only by broad subsystem keywords rather than the same defect. No obvious overstatement." }, "evaluator_used": true, "retried": false, @@ -21725,10 +21731,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -21743,165 +21749,165 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a mixed cluster of unrelated diffusion/model-support issues. The grayscale/single-channel image pair (1280/488) looks like a true duplicate, but the Qwen-Image pair (12367/13065) should be treated conservatively as separate feature requests rather than a duplicate. The rest are same-area but different bugs or asks.", - "confidence": 0.91, - "canonical_issue_reason": "Issue 488 is the clearest concrete bug report in the cluster: it directly names grayscale/single-channel image handling and points to the likely failing path (`numpy_to_pil`).", + "summary": "This is not a single coherent duplicate cluster: most items are unrelated model-support or training-error reports. I only see one likely true duplicate pair around grayscale/single-channel image handling.", + "confidence": 0.78, + "canonical_issue_reason": "No single canonical issue cleanly represents the whole set; the items span many unrelated topics, so a global duplicate target would be unsafe.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 488 is the best canonical issue candidate because it is the most specific and actionable report for the duplicate grayscale/single-channel failure.", + "best_issue_reason": "488 is the clearest representative only for the grayscale/single-channel image bug, but it is not a good global canonical for the full mixed cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4427", "right": "issue:5108", "accept": false, - "reason": "Both are T2I-MultiAdapter reports, but they describe different problems (example not working vs XL exceptions), not the same concrete bug." + "reason": "Same broad T2I-MultiAdapter area, but different example variants and failure modes; not the same concrete bug." }, { "left": "issue:4708", "right": "issue:7021", "accept": false, - "reason": "Different import issues: one is a `diffusers.utils` import failure, the other is an `HF_HOME` import error." + "reason": "Both mention import problems, but one is about diffusers.utils and the other HF_HOME; too different to be duplicates." }, { "left": "issue:1280", "right": "issue:488", "accept": true, - "reason": "Same underlying bug: grayscale / single-channel image handling in diffusion pipelines; 488 is the more precise report of the failing path." + "reason": "Both describe grayscale/single-channel image handling failures in pipelines; same underlying bug." }, { "left": "issue:12367", "right": "issue:13065", "accept": false, - "reason": "Both are Qwen-Image feature requests, but they ask for different capabilities (Eligen LoRA support vs controlnet inpaint+canny), so the bodies are not clearly the same request." + "reason": "Different Qwen-Image feature requests: LoRA support vs controlnet inpaint/canny combination." }, { "left": "issue:1594", "right": "issue:2984", "accept": false, - "reason": "Both concern video generation, but they are different model/paper features rather than the same bug or change." + "reason": "Both are video-related, but they ask for different features and are not the same change." }, { "left": "issue:1305", "right": "issue:1350", "accept": false, - "reason": "Completely different topics: direct inversion image editing vs DiffusionDet object detection." + "reason": "Completely unrelated topics: direct inversion editing versus object detection." }, { "left": "issue:4584", "right": "issue:5849", "accept": false, - "reason": "Same training script family, but different failures (`--resume_from_checkpoint` handling vs missing `args` in accelerate launch)." + "reason": "Same training script family, but one is a checkpoint-resume error and the other is an accelerate entrypoint TypeError." }, { "left": "issue:5168", "right": "issue:5284", "accept": false, - "reason": "Both are control-related extensions, but they target different features: ControlNet-XS vs ControlNet-LLLite." + "reason": "Both are ControlNet variants, but they are distinct model-support requests." }, { "left": "issue:3728", "right": "issue:5576", "accept": false, - "reason": "Different change requests: Shifted Diffusion pipeline vs FreeNoise scheduler." + "reason": "Different scheduler proposals; not the same underlying feature." }, { "left": "issue:8511", "right": "issue:9876", "accept": false, - "reason": "Related to LoRA training memory behavior, but one is about SD3 text encoders not unloading and the other is a broad VRAM-release complaint." + "reason": "Both involve LoRA VRAM usage, but one is specifically about SD3 text encoders not unloading while the other is a broad memory-release complaint." }, { "left": "issue:11182", "right": "issue:9576", "accept": false, - "reason": "Different pipeline loading bugs: unnecessary BF16 download vs `use_safetensors` not propagated to submodules." + "reason": "Different pipeline-loading bugs: unnecessary BF16 downloads versus use_safetensors propagation." }, { "left": "issue:4754", "right": "issue:4849", "accept": false, - "reason": "Textual inversion is the shared area, but one is a missing `model_index.json` after loading and the other is a multi-GPU training question." + "reason": "One is a load-time missing model_index.json error; the other is a multi-GPU training question." }, { "left": "issue:12037", "right": "issue:12052", "accept": false, - "reason": "Both involve Wan 2.2 and LightX2V, but the concrete failures differ: LoRA compatibility vs offloading device-mismatch multiplication." + "reason": "Both involve Wan 2.2 and LightX2V, but one is a LoRA compatibility issue and the other is a device-mismatch offloading bug." }, { "left": "issue:10639", "right": "issue:12412", "accept": false, - "reason": "Different model support requests: Lumina-Image 2.0 vs HunyuanImage-3.0." + "reason": "Separate model support requests for Lumina-Image 2.0 and HunyuanImage-3.0." }, { "left": "issue:2596", "right": "issue:4708", "accept": false, - "reason": "One is documentation pointing to a non-existing helper; the other is a runtime import error." + "reason": "Both mention diffusers.utils, but one is a docs typo and the other is a generic import failure; not enough evidence of the same bug." }, { "left": "issue:3422", "right": "issue:5623", "accept": false, - "reason": "Both mention safety checker, but one asks how to disable it and the other asks for enhanced functionality." + "reason": "Both touch safety checker behavior, but one is a disable request and the other is an enhancement request." }, { "left": "issue:8594", "right": "issue:9628", "accept": false, - "reason": "Same script family, but different failures: tokenizer variable initialization vs FP16 unscale error." + "reason": "Same SD3 dreambooth/LoRA script, but different concrete failures and code paths." }, { "left": "issue:11010", "right": "issue:11408", "accept": false, - "reason": "Different issues entirely: Chroma/Flux architecture support vs compiled-model `set_adapters` support." + "reason": "Flux model support versus compiled-model adapter support; unrelated changes." }, { "left": "issue:3884", "right": "issue:7646", "accept": false, - "reason": "Same training script, but one is a generic error and the other is about `dataloader_num_workers`; not the same bug." + "reason": "Same script name, but one is an execution error and the other is a dataloader_num_workers question." }, { "left": "issue:2132", "right": "issue:4753", "accept": false, - "reason": "Both are Dreambooth LoRA script problems, but the concrete errors are different (`Accelerator.trackers` vs a generic script issue)." + "reason": "Both are dreambooth_lora.py reports, but they describe different errors and no shared root cause is clear." }, { "left": "issue:13162", "right": "issue:13547", "accept": false, - "reason": "Same model family, but one is a support request for Flux.2-klein-4b LoRA and the other is an adapter loading error; not clearly the same bug." + "reason": "Same Flux2-klein LoRA area, but one is support and the other is an adapter-loading error; not the same bug." }, { "left": "issue:2390", "right": "issue:3564", "accept": false, - "reason": "Both are pipeline support requests, but for different adapter types: T2I-Adapter vs Uni-ControlNet." + "reason": "Different control-method support requests: T2I-Adapter versus Uni-ControlNet." }, { "left": "issue:10202", "right": "issue:10594", "accept": false, - "reason": "Both involve FLUX and LoRA/control usage, but one is about not unloading a control LoRA while the other is about multiple controls on FLUX.1-dev." + "reason": "Both mention FLUX and LoRA, but one is about unloading a LoRA and the other about combining multiple controls with LoRA." }, { "left": "issue:1082", "right": "issue:2412", "accept": false, - "reason": "Both are Dreambooth errors, but they are too broad and describe different failure modes." + "reason": "Two different Dreambooth failures with no clear shared code-path." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The summary is mostly grounded, but it overstates at least one soft-edge verdict. The 1280/488 grayscale/single-channel duplicate looks plausible from the titles. However, 12367/13065 is not supported well enough to call a true duplicate: one is a Qwen-Image Eligen LoRA request and the other is Qwen-Image controlnet inpaint + canny support, which are different feature asks on the same model family. That should stay conservative as a non-duplicate unless the bodies clearly show the same bug/request." + "accept": true, + "feedback": "The summary is well grounded in the packet: the cluster is clearly mixed, and the only confidently duplicate-looking pair is 1280 \u2194 488 on grayscale/single-channel image handling. The other pair verdicts stay conservative and are supported by the issue titles/bodies; no PR-merge concerns apply here." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -21998,10 +22004,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -22016,156 +22022,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly unrelated issues that only share broad keywords like training, loading, schedulers, or FLUX/LTX. I don\u2019t see any pair that clearly describes the same concrete bug or change, so all soft edges should be rejected.", - "confidence": 0.9, - "canonical_issue_reason": null, + "summary": "Highly heterogeneous issue cluster; the soft pairs share only loose topical similarity, not the same concrete bug/change. No PRs present.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 6503 is the closest canonical representative because it is an umbrella/tracker-style item about training-script compatibility and has the strongest centrality among the set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:6503 is the closest thing to a cluster anchor because it is a broad tracker with the highest reference activity, but it is still an umbrella issue rather than a duplicate of the others.", + "best_issue_reason": "Issue 6503 is the best issue anchor only by breadth and reference count; the rest are mostly unrelated narrow bugs, API mismatches, or feature requests, so this is not a true duplicate hub.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:11363", "right": "issue:9546", "accept": false, - "reason": "Both mention OOM during validation, but they are different workflows/models: ControlNet example vs Flux Controlnet training." + "reason": "Both mention OOM, but one is a ControlNet inference example and the other is FLUX ControlNet validation during training; different code paths and likely different fixes." }, { "left": "issue:2412", "right": "issue:6073", "accept": false, - "reason": "Both are Dreambooth failures, but one is a generic error and the other is a beginner SDXL README issue; no shared concrete bug." + "reason": "Both are Dreambooth-related errors, but the reports describe different workflows and failure contexts; no evidence of the same underlying bug." }, { "left": "issue:11010", "right": "issue:9643", "accept": false, - "reason": "One is a feature request for FLUX architecture changes, the other is a limitation question about multiple ControlNets." + "reason": "One is a request for Chroma/Flux architecture changes, the other asks about multiple ControlNets in Flux; these are different feature/support questions." }, { "left": "issue:6565", "right": "issue:6741", "accept": false, - "reason": "DPO training instability and a missing CLI arg are different problems in different code paths." + "reason": "Both are Diffusion DPO training issues, but one is about loss spikes/noised samples and the other is a missing CLI argument; different defects." }, { "left": "issue:10321", "right": "issue:11104", "accept": false, - "reason": "Both are LTX-Video related, but one is a missing model file error and the other is broken offload methods." + "reason": "Both involve LTX-Video, but one is a missing-model-file load error while the other says both offload methods are broken; different failure modes." }, { "left": "issue:10460", "right": "issue:9370", "accept": false, - "reason": "These describe different API/signature bugs in different classes, not the same underlying defect." + "reason": "These are separate API signature bugs on different classes with different missing arguments (`mid_block_type` vs `pooled_projection`)." }, { "left": "issue:12221", "right": "issue:9551", "accept": false, - "reason": "One is a support request for DiffSynth ControlNet; the other is a question about using x-labs FLUX ControlNet models." + "reason": "Both are Flux ControlNet support topics, but they ask for different integrations and usage guidance rather than the same bug/change." }, { "left": "issue:2321", "right": "issue:2466", "accept": false, - "reason": "EMA weight handling during inference and checkpoint retention limits are unrelated." + "reason": "One concerns EMA/model copying in inference, the other checkpoint retention behavior during training; not the same underlying issue." }, { "left": "issue:10760", "right": "issue:7210", "accept": false, - "reason": "Import/export exposure bug vs a missing attribute during checkpoint dispatch; different failure points." + "reason": "A missing import/export attribute for a pipeline class is unrelated to a `named_parameters` attribute error on a pipeline object." }, { "left": "issue:11024", "right": "issue:8307", "accept": false, - "reason": "SiglipImageProcessor .to() and StableCascadeUNet missing from the package are different issues." + "reason": "These affect different classes/models (`SiglipImageProcessor` vs `StableCascadeUNet`) and different error conditions." }, { "left": "issue:11432", "right": "issue:9971", "accept": false, - "reason": "A from_pretrained dtype parsing bug is unrelated to DEIS scheduler behavior on FLUX." + "reason": "`from_pretrained(torch_dtype=\"auto\")` behavior is unrelated to DEISMultistepScheduler not working on FLUX." }, { "left": "issue:6069", "right": "issue:6527", "accept": false, - "reason": "Both involve add_noise, but they are different schedulers and likely different edge-case failures." + "reason": "Both are `add_noise` complaints, but they involve different schedulers and different edge conditions, so they are not the same bug." }, { "left": "issue:5227", "right": "issue:7159", "accept": false, - "reason": "Both mention torch.compile and LoRA, but one is loading weights and the other is compiling a fused LoRA path." + "reason": "Both mention `torch.compile`, but one is about `load_lora_weight` and the other about fused LoRA still failing under compile; distinct code paths." }, { "left": "issue:4993", "right": "issue:9223", "accept": false, - "reason": "An import-structure export bug is unrelated to a dependency version conflict." + "reason": "A missing export for `CLIPImageProjection` is unrelated to a dependency version conflict around `pyarrow`." }, { "left": "issue:378", "right": "issue:579", "accept": false, - "reason": "Both are textual inversion issues, but one is about DiagonalGaussianDistribution and the other is a missing config.json." + "reason": "Both are textual inversion issues, but one is about a distribution class and the other about a missing config file; different root causes." }, { "left": "issue:4704", "right": "issue:6709", "accept": false, - "reason": "Both are Custom Diffusion training issues, but the concrete failures and scripts differ." + "reason": "Both are Custom Diffusion training reports, but one is a general training failure and the other is a concept-list-specific bug." }, { "left": "issue:6520", "right": "issue:7538", "accept": false, - "reason": "Blank validation output during adapter training is not the same as a dataset/style training issue." + "reason": "A blank validation output in T2I adapter training is unrelated to a Custom Fill50k style dataset training issue." }, { "left": "issue:8939", "right": "issue:9209", "accept": false, - "reason": "Tokenizer/model loading failure and inability to reach hf-mirror.com are different root causes." + "reason": "One is a tokenizer/model loading path problem; the other is a mirror connectivity failure. Different causes and fixes." }, { "left": "issue:1574", "right": "issue:3613", "accept": false, - "reason": "Both involve Dreambooth memory failures, but the environment and specific problem statements are different." + "reason": "Both are Dreambooth memory problems, but the reports are about different setups and symptoms; not enough to treat as the same concrete bug." }, { "left": "issue:4852", "right": "issue:4911", "accept": false, - "reason": "LoRA loading and custom text_encoder loading in from_single_file are distinct load paths." + "reason": "Loading LoRA weights and loading a custom text encoder in `from_single_file` are different loading pathways and errors." }, { "left": "issue:2177", "right": "issue:2728", "accept": false, - "reason": "AutoencoderKL loading and UNet2DConditionModel safetensors loading are different model-class issues." + "reason": "These are different model classes and different loading failures (`AutoencoderKL` vs `UNet2DConditionModel` with safetensors)." }, { "left": "issue:6162", "right": "issue:7355", "accept": false, - "reason": "SDXL IP-Adapter shape mismatch and Stable Cascade resolution/bfloat16 errors are unrelated." + "reason": "An IP-Adapter SDXL shape mismatch is unrelated to Stable Cascade arbitrary-resolution/bfloat16 generation failures." }, { "left": "issue:4837", "right": "issue:861", "accept": false, - "reason": "Both involve conversion scripts, but one is SDXL-specific and the other is generic ckpt conversion; not the same bug." + "reason": "Both touch the converter script, but one is an SDXL conversion failure and the other is a general ckpt-to-diffusers conversion problem; not clearly the same bug." }, { "left": "issue:6503", "right": "issue:6545", "accept": false, - "reason": "A torch.compile compatibility tracker and an inferencing modularization tracker are different umbrella issues, not duplicates." + "reason": "Both are tracker-style training issues, but they cover different initiatives: torch.compile compatibility versus modularizing inference during/after training." } ] }, @@ -22269,10 +22275,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -22287,169 +22293,172 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of loosely similar model-loading / LoRA / controlnet / training issues, but I don\u2019t see any pair that clearly describes the exact same underlying bug or change. Most soft links are broad-topic similarity rather than duplicates.", - "confidence": 0.68, - "canonical_issue_reason": "Issue #9237 is the most central representative of the repeated Flux/LoRA training theme and has substantial discussion, making it the best canonical issue for this cluster despite the lack of true duplicates.", + "summary": "This cluster is mostly heterogeneous. Only one soft pair looks like a genuine duplicate: the SDXL DreamBooth LoRA resume_from_checkpoint failures in the basic and advanced training scripts. The rest are related at a broad subsystem level but describe different bugs or features.", + "confidence": 0.77, + "canonical_issue_reason": "issue:6482 is the clearest representative of the likely duplicate training-resume bug because it names the exact failing script variant and the concrete FP16 unscale exception.", "canonical_pr_reason": null, - "best_issue_reason": "#9237 best captures the dominant subtheme in this cluster (Flux Dreambooth/LoRA training) and is more representative than the narrower pipeline- or symptom-specific reports.", + "best_issue_reason": "issue:6482 is the best issue to anchor the cluster because it is the most specific and actionable of the likely duplicate pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:7049", "right": "issue:9448", "accept": false, - "reason": "Both are IP-Adapter-related, but one is a tuple/unpack error and the other is a tensor shape attribute error in a different controlnet/inpaint path." + "reason": "Both are IP-Adapter-related, but they report different failures in different pipelines/errors; not the same concrete bug." }, { "left": "issue:10925", "right": "issue:9237", "accept": false, - "reason": "Both concern Flux Dreambooth/LoRA training, but they target different problems: CLIPTextModel finetuning vs a general Flux Dreambooth LoRA training issue." + "reason": "Both mention FLUX DreamBooth/LoRA, but one is a CLIPTextModel finetune issue and the other is a LoRA training issue; too broad." }, { "left": "issue:5558", "right": "issue:7348", "accept": false, - "reason": "Both mention ControlNet/inpaint artifacts, but one is about multiple ControlNets and the other about scheduler behavior; not the same bug." + "reason": "ControlNet inpaint artifacts vs scheduler weird results are different symptoms and likely different code paths." }, { "left": "issue:12613", "right": "issue:12699", "accept": false, - "reason": "Same Wan 2.2 area, but one reports NaNs in latent i2v and the other reports quality degradation after many prompts with a lighting LoRA." + "reason": "Wan 2.2 latent NaNs in i2v and video quality degrading after many prompts are not the same bug." }, { "left": "issue:4305", "right": "issue:7985", "accept": false, - "reason": "Both involve torch.compile, but one is a general graph-break regression and the other is an IP-Adapter-specific compile incompatibility." + "reason": "Both involve torch.compile, but one is a graph break regression and the other is IP-Adapter incompatibility; distinct issues." }, { "left": "issue:7049", "right": "issue:8626", "accept": false, - "reason": "IP-Adapter is shared, but one is a runtime ValueError and the other is a documentation/guidance request, not the same defect." + "reason": "One is a runtime error; the other is a guidance/docs request about multiple IP-Adapter images/masks." }, { "left": "issue:12082", "right": "issue:12565", "accept": false, - "reason": "Both are Wan 2.1 issues, but one is a VAE memory regression after compile and the other is a broader i2v running problem." + "reason": "Both are Wan2.1 issues, but one is VAE memory after compile and the other is a general i2v running problem." }, { "left": "issue:8828", "right": "issue:8839", "accept": false, - "reason": "Both are SD3 training-related, but one is a 16GB training error and the other is a LoRA weight loading error." + "reason": "SD3 training error vs SD3 LoRA weight error are different failures." }, { "left": "issue:10150", "right": "issue:12599", "accept": false, - "reason": "Both concern slowness, but one is SDXL pipeline loading/inference and the other is slow CUDA transfer after loading; different code paths." + "reason": "Both mention slowness, but one is SDXL pipeline loading/inference generally and the other is moving to CUDA after loading; not clearly the same bug." }, { "left": "issue:11642", "right": "issue:9525", "accept": false, - "reason": "Both mention Flux scaling, but one is prompt_embeds_scale in FluxPriorReduxPipeline and the other is lora_scale during Flux loading; different mechanisms." + "reason": "Both say a scale parameter has no effect in Flux, but they concern different parameters and different pipelines." }, { "left": "issue:4303", "right": "issue:4376", "accept": false, - "reason": "Both are SDXL load crashes, but they hit different APIs and different failure modes." + "reason": "Both are SDXL/Flax load crashes, but one is loading SDXL in FlaxStableDiffusionPipeline and the other is load_textual_inversion crashing." }, { "left": "issue:11024", "right": "issue:8182", "accept": false, - "reason": "These are unrelated; one is an SD3 IP-Adapter processor issue, the other is a Logger typo." + "reason": "Completely unrelated errors: SiglipImageProcessor .to() missing vs a Logger typo." }, { "left": "issue:10202", "right": "issue:11464", "accept": false, - "reason": "Both concern Flux Control LoRA, but one is unloading correctly and the other is loading into FluxControlPipeline; not the same bug." + "reason": "Both involve FluxControlPipeline LoRA handling, but one is unloading correctly and the other is inability to load a specific LoRA; not enough to merge as one bug." }, { "left": "issue:1823", "right": "issue:228", "accept": false, - "reason": "Completely unrelated: a comment swap in depth2img versus a broken release link." + "reason": "Swapped comments in depth2img and a broken model link are unrelated." }, { "left": "issue:3019", "right": "issue:3728", "accept": false, - "reason": "Different models and different requests; no shared bug or change." + "reason": "Different model/pipeline feature requests with no common failure mode." }, { "left": "issue:8787", "right": "issue:9525", "accept": false, - "reason": "One is a generic model-loading slowdown, the other is Flux LoRA scaling not taking effect." + "reason": "Generic loading slowdown and a Flux LoRA scale bug are unrelated." }, { "left": "issue:11363", "right": "issue:9732", "accept": false, - "reason": "Both are OOM reports, but one is a controlnet inference example and the other is DreamBooth training; different workloads." + "reason": "ControlNet example CUDA OOM and FLUX DreamBooth training OOM are different workloads and likely different fixes." }, { "left": "issue:4492", "right": "issue:8939", "accept": false, - "reason": "Unrelated issues: a false/False NameError versus a tokenizer loading/path error." + "reason": "A Python boolean typo error and an SD3 tokenizer loading OSError are unrelated." }, { "left": "issue:1350", "right": "issue:2984", "accept": false, - "reason": "Different projects and different problem statements; not duplicates." + "reason": "DiffusionDet object detection and latent video diffusion are different projects/features." }, { "left": "issue:6418", "right": "issue:7773", "accept": false, - "reason": "Both are Stable Video Diffusion, but one is repeated-frame output and the other is a batch-size bug." + "reason": "Both are Stable Video Diffusion issues, but one is identical first-frame output after finetuning and the other is a batch-size>1 generation bug." }, { "left": "issue:6598", "right": "issue:7142", "accept": false, - "reason": "Both compare diffusers output to other UIs, but they involve different pipelines/models and different behaviors." + "reason": "Both compare outputs, but one is generic img2img result mismatch and the other is SDXL-lightning vs webui discrepancy." }, { "left": "issue:4194", "right": "issue:4354", "accept": false, - "reason": "Both are from_single_file-related, but one is a VAE load problem and the other is an incorrect pipeline type return." + "reason": "Both are single-file loading problems, but one is VAE not working and the other is wrong pipeline type returned." }, { "left": "issue:10594", "right": "issue:11109", "accept": false, - "reason": "Both involve FLUX + LoRA + controls, but one is about combining multiple controls and the other about transformer layer mismatch; different bugs." + "reason": "Both mention FLUX and LoRA/control, but one is multiple controls with LoRA and the other is a depth LoRA transformer-layer mismatch." }, { "left": "issue:4061", "right": "issue:4885", "accept": false, - "reason": "Both are SDXL LoRA loading issues, but they affect different pipelines and report different failure modes." + "reason": "Both are SDXL LoRA-loading problems, but they target different pipelines and likely different code paths." }, { "left": "issue:5412", "right": "issue:6482", - "accept": false, - "reason": "Both are resume_from_checkpoint failures in SDXL LoRA training, but the scripts and error signatures differ enough that they are not clearly the same bug." + "accept": true, + "reason": "Same SDXL DreamBooth LoRA checkpoint-resume failure across two script variants; likely one underlying bug in resume handling." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge calls stay conservative overall. The lone accepted pair, issue:5412 and issue:6482, is sufficiently supported by their shared resume_from_checkpoint SDXL DreamBooth LoRA training-failure context. The rest are reasonably rejected as broad subsystem similarities rather than the same bug." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -22545,10 +22554,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -22563,162 +22572,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issue reports with two likely duplicate pairs: MPS img2img corruption and SDXL from_single_file returning the wrong pipeline type. The rest are only broadly similar by subsystem or symptom and should stay separate.", - "confidence": 0.72, - "canonical_issue_reason": null, + "summary": "Mostly unrelated issues were grouped by loose title similarity. Only the `from_single_file` SDXL pipeline-type mismatch pair looks like a true duplicate; the rest are similar only at a broad subsystem level.", + "confidence": 0.71, + "canonical_issue_reason": "issue:6001 is the clearest concrete report of the shared `from_single_file` SDXL loader bug: the refiner is returned as `Img2ImgPipeline` instead of the expected pipeline type.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:6001 is the best canonical anchor because it states the wrong-return-type failure most specifically and is directly actionable.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4141", "right": "issue:4418", "accept": false, - "reason": "Both are in inpaint/controlnet-adjacent areas, but one is about a second ControlNet being ignored and the other is about denoising_start behavior in SDXL inpaint; different bugs." + "reason": "Both are inpaint/control-related, but one is about second ControlNet handling and the other about `denoising_start`; different bugs." }, { "left": "issue:462", "right": "issue:941", - "accept": true, - "reason": "Both describe StableDiffusionImg2ImgPipeline producing broken/noisy outputs on Apple MPS; 941 is a narrower repro of the same pipeline bug." + "accept": false, + "reason": "Both involve MPS img2img behavior, but the symptoms and triggers differ (`0.3.0` regression vs `num_images_per_prompt` noise)." }, { "left": "issue:11143", "right": "issue:11258", "accept": false, - "reason": "Different LTX-Video problems: one is output quality, the other is a single-file loading topic." + "reason": "Different LTX-Video topics: one is quality feedback, the other is a single-file loading request." }, { "left": "issue:4354", "right": "issue:6001", "accept": true, - "reason": "Both report from_single_file returning the wrong SDXL inpaint/refiner pipeline type; same loader return-path bug." + "reason": "Both report `from_single_file` returning the wrong SDXL pipeline class for inpaint/refiner paths, which looks like the same underlying loader bug." }, { "left": "issue:4911", "right": "issue:8435", "accept": false, - "reason": "Both involve from_single_file, but one is a custom text_encoder loading failure and the other is a generic runtime error; not the same concrete defect." + "reason": "Both mention `from_single_file`, but one is about custom text encoders and the other is a generic runtime error; not the same failure." }, { "left": "issue:2560", "right": "issue:3428", "accept": false, - "reason": "Both mention StableDiffusionImg2ImgPipeline, but the titles point to different symptoms and no shared concrete failure mode." + "reason": "Both concern StableDiffusionImg2ImgPipeline behavior, but they describe different output anomalies and likely different code paths." }, { "left": "issue:1005", "right": "issue:1367", "accept": false, - "reason": "Scheduler timestep limits vs saving custom trained_betas configuration are unrelated scheduler issues." + "reason": "Scheduler timestep limits and custom `trained_betas` config saving are separate scheduler issues." }, { "left": "issue:3382", "right": "issue:7757", "accept": false, - "reason": "Both are multi-GPU OOM reports, but for different training scripts and likely different causes." + "reason": "Both are training OOMs, but they target different scripts and settings; too broad to be the same bug." }, { "left": "issue:5623", "right": "issue:7363", "accept": false, - "reason": "General safety checker enhancement vs adding safety checker support for stable_diffusion_xl are related feature areas, not the same bug/change." + "reason": "One is a safety-checker enhancement request, the other a request for SDXL safety checker support; not the same change." }, { "left": "issue:5927", "right": "issue:6243", "accept": false, - "reason": "Both involve IP-Adapter variants, but the feature combinations and failure modes differ." + "reason": "Both mention IP-Adapter, but the configurations and failures are different." }, { "left": "issue:6574", "right": "issue:7629", "accept": false, - "reason": "One is a warning in SVD preprocessing, the other is a PyTorch-version-specific OOM; not the same issue." + "reason": "Both mention SVD, but one is a warning/shape issue and the other is a version-specific OOM; unrelated." }, { "left": "issue:13043", "right": "issue:13298", "accept": false, - "reason": "Different CUDA/device problems: kernel-image support on RTX 5090 vs generator-device handling breaking image generation." + "reason": "Different CUDA/device failures: a specific Blackwell kernel-image issue vs a general generator-device breakage." }, { "left": "issue:3717", "right": "issue:3774", "accept": false, - "reason": "Both concern memory, but one is multi-GPU OOM while the other is zero3 not saving memory in a training script." + "reason": "Both are memory-related training issues, but one is multi-GPU OOM and the other is ZeRO-3 not saving memory." }, { "left": "issue:7229", "right": "issue:7728", "accept": false, - "reason": "Both are advanced training script issues, but one centers on checkpoint saving and the other on an Accelerate error during training." + "reason": "Checkpoint saving problems and an Accelerate error in an advanced DreamBooth script are not the same bug." }, { "left": "issue:6252", "right": "issue:6446", "accept": false, - "reason": "A checkpointed LoRA weight-loading bug and a Colab loading failure are different problems." + "reason": "Both are loading-related, but one is LoRA weights from a checkpoint and the other is a minimal-loading model init failure." }, { "left": "issue:5857", "right": "issue:6008", "accept": false, - "reason": "Both mention LCM LoRA, but one is AnimateDiff SDXL and the other is IP-Adapter FaceId; different code paths." + "reason": "Both involve LCM LoRA, but the affected pipelines and failure modes differ." }, { "left": "issue:5889", "right": "issue:7405", "accept": false, - "reason": "A stable video diffusion topic and a DynamiCrafter new-model request are not duplicates." + "reason": "Completely different models/projects: Stable Video Diffusion vs DynamiCrafter." }, { "left": "issue:13484", "right": "issue:13487", "accept": false, - "reason": "Both are LoRA-loading complaints in Flux/modular pipelines, but one is FAL LoRA loading on Flux 2 dev and the other is loading on text_pipe; not clearly the same bug." + "reason": "Both are LoRA-loading issues in modular/Flux contexts, but the titles point to different components and failure scenarios." }, { "left": "issue:10674", "right": "issue:9971", "accept": false, - "reason": "GGUF support failure in FluxPipeline vs DEIS scheduler not working on FLUX are separate issues." + "reason": "Both are FLUX-related, but one is a GGUF pipeline issue and the other is a scheduler compatibility issue." }, { "left": "issue:10914", "right": "issue:11023", "accept": false, - "reason": "Unexpected CPU offload vs blurry outputs in a custom SDXL pipeline are unrelated." + "reason": "A device offloading problem and a blurry-image custom-pipeline report are unrelated." }, { "left": "issue:10596", "right": "issue:10673", "accept": false, - "reason": "Both are invalid-shape errors, but they involve different pipelines/models and likely different reshape logic." + "reason": "Both are invalid-shape errors, but they occur in different pipelines and likely stem from different shape assumptions." }, { "left": "issue:10833", "right": "issue:8787", "accept": false, - "reason": "A bitsandbytes matrix-shape bug in FluxFillPipeline and a general model-loading slowdown are not the same defect." + "reason": "A bitsandbytes matrix-shape failure and a model-loading slowdown are not the same issue." }, { "left": "issue:10795", "right": "issue:8518", "accept": false, - "reason": "Both involve torch.compile, but one is GGUF via from_single_file and the other is an SD3 compile Unsupported error; different failure points." + "reason": "Both are torch.compile failures, but one is GGUF/from_single_file related while the other is an SD3-specific dynamo unsupported-op error." }, { "left": "issue:5877", "right": "issue:9664", "accept": false, - "reason": "Both are subprocess failures in training scripts, but the affected workflows and likely root causes are different." + "reason": "Both are subprocess training failures, but they involve different scripts and no clear shared root cause." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The two accepted pairs are well-supported by the titles as likely duplicates, while the remaining pairs are mostly different subsystems or distinct failure modes. No soft-edge verdict looks overstated." + "feedback": "The summary is grounded in the packet metadata, and the soft-edge calls are conservative. The one accepted pair (4354/6001) is well supported by the titles as the same `from_single_file` SDXL pipeline-type mismatch bug, while the rest are reasonably rejected as broader subsystem similarities rather than clear duplicates." }, "evaluator_used": true, "retried": false, @@ -22817,10 +22826,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -22835,163 +22844,166 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issues grouped by superficial title similarity. The only close theme is multi-LoRA/Flux loading, but even \u0aa4\u0acd\u0aaf\u0abe\u0a82 the reports differ enough that none should be merged as duplicates. No PRs are present.", - "confidence": 0.93, - "canonical_issue_reason": null, + "summary": "Mostly unrelated issue set; the only strong duplicate-like pair is the multiple-LoRA PEFT error around Flux. Everything else is either a different model/pipeline, a different failure mode, or just a broad subsystem overlap.", + "confidence": 0.39, + "canonical_issue_reason": "issue:11298 is the best canonical issue because it is the clearest, still-open report for the concrete multiple-LoRA/PEFT key-error path and is the most likely umbrella for the closely related Flux LoRA-loading report.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11298 is the clearest, most specific current report in the LoRA-loading subset (open, explicit PEFT key error), but it should not be treated as a true canonical duplicate for the whole cluster.", + "best_issue_reason": "issue:11298 is the strongest global issue candidate: specific, actionable, and aligned with the only plausible duplicate pair in the set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4236", "right": "issue:5075", "accept": false, - "reason": "Both are SDXL refiner problems, but one is text-encoder registration and the other is prompt-embedding shape mismatch; different bugs." + "reason": "Both involve SDXL refiner shape issues, but one is text-encoder registration and the other is prompt-embedding matmul mismatch; not the same bug." }, { "left": "issue:1061", "right": "issue:228", "accept": false, - "reason": "Docs/link update for Textual Inversion is unrelated to a broken release download link." + "reason": "One is a docs/example update request; the other is a broken release link. Different artifact types and different problems." }, { "left": "issue:3026", "right": "issue:3717", "accept": false, - "reason": "Both mention training on limited GPUs, but one is an accelerate launch/runtime error and the other is multi-GPU OOM; not the same bug." + "reason": "Different training/runtime failures: one is an instruct-pix2pix accelerate launch error, the other is multi-GPU OOM." }, { "left": "issue:4483", "right": "issue:5931", "accept": false, - "reason": "Both concern determinism, but they target different pipelines and likely different causes." + "reason": "Both mention determinism, but they target different pipelines and likely different causes; not enough to treat as one bug." }, { "left": "issue:6818", "right": "issue:7033", "accept": false, - "reason": "Both involve IP-Adapter, but one is an unload/runtime error and the other is warning logging; different code paths." + "reason": "Unloading an IP-Adapter and attention-masking warning logs are unrelated failure modes." }, { "left": "issue:7377", "right": "issue:8649", "accept": false, - "reason": "Both mention num_images_per_prompt, but they affect different pipelines/models and report different failure modes." + "reason": "Both mention num_images_per_prompt, but they are different pipelines/model families with different reported breakages." }, { "left": "issue:11907", "right": "issue:9643", "accept": false, - "reason": "Flux scheduler incompatibility is a different problem from Flux multi-ControlNet support." + "reason": "Both are Flux-related, but one is scheduler incompatibility and the other is multiple-ControlNet support; different concrete issues." }, { "left": "issue:4134", "right": "issue:4480", "accept": false, - "reason": "Both are MPS-related, but one is a tensor shape mismatch in ControlNet and the other is an fp16 kernel crash in StableDiffusionPipeline." + "reason": "Both are MPS/Mac issues, but one is a ControlNet tensor-shape mismatch and the other is a kernel crash with fp16 generation." }, { "left": "issue:4993", "right": "issue:5917", "accept": false, - "reason": "Import-structure exposure of CLIPImageProjection is unrelated to checkpoint conversion failing after an upgrade." + "reason": "Import-structure exposure vs checkpoint conversion regression; different code paths and symptoms." }, { "left": "issue:6124", "right": "issue:7513", "accept": false, - "reason": "Both are DreamBooth LoRA reports, but one is an FP16 unscale error and the other is a GPU memory issue; not the same bug." + "reason": "Both concern DreamBooth/LoRA training, but one is an FP16 gradient unscale error and the other is a generic GPU memory issue." }, { "left": "issue:10334", "right": "issue:11342", "accept": false, - "reason": "Both involve Sana on Apple hardware, but one is output corruption/NaNs and the other is pipeline failure; insufficiently specific to merge." + "reason": "Same model family and platform, but the reported failures differ too much (grey images/NaNs vs pipeline failing) to confidently merge." }, { "left": "issue:11010", "right": "issue:9667", "accept": false, - "reason": "Both concern Flux model support/loading, but one is an architecture-change request and the other is a local fp8 loader bug." + "reason": "Both involve Flux-style architecture/model loading, but one is feature/support discussion and the other is an fp8 loading error." }, { "left": "issue:10350", "right": "issue:12501", "accept": false, - "reason": "Both mention VAE decoder, but one is memory usage and the other is a graph break; different issues." + "reason": "Both mention the VAE decoder, but one is memory consumption and the other is a torch graph-break issue; different bugs." }, { "left": "issue:10321", "right": "issue:11143", "accept": false, - "reason": "Both reference LTX-Video, but one is a model-loading file error and the other is a video-quality complaint." + "reason": "Both are about LTX-Video, but one is a missing model file error and the other is video quality complaints." }, { "left": "issue:10180", "right": "issue:11298", - "accept": false, - "reason": "Both are multi-LoRA/Flux-adjacent, but one is loading multiple LoRAs with Flux Control LoRA and the other is a PEFT key error during hotswapping; not clearly the same bug." + "accept": true, + "reason": "Both describe multiple-LoRA handling failing in the Flux/PEFT path, and the titles point to the same underlying key-error/load-management bug." }, { "left": "issue:3357", "right": "issue:6096", "accept": false, - "reason": "Both are torch compile failures, but they are different exception types and likely different unsupported operations." + "reason": "Both are torch.compile issues, but they are different compiler failures with different reported exceptions and code paths." }, { "left": "issue:2674", "right": "issue:5753", "accept": false, - "reason": "One is about safetensors dependency/requirements, the other about SDXL ControlNet with load_lora_weights; unrelated." + "reason": "LoRA-related, but one is a missing dependency/safetensors packaging problem while the other is a pipeline/load_lora_weights incompatibility." }, { "left": "issue:1734", "right": "issue:2156", "accept": false, - "reason": "Both are Dreambooth training performance issues, but one is multi-GPU slowdown and the other is CPU-only training instead of GPU." + "reason": "Different training performance problems: multi-GPU slowness versus running on CPU instead of GPU." }, { "left": "issue:2299", "right": "issue:993", "accept": false, - "reason": "Fine-tuned checkpoint integration and fp16 dtype mismatch are different bugs." + "reason": "Different issues: checkpoint compatibility/support versus a datatype mismatch in fp16." }, { "left": "issue:6709", "right": "issue:6879", "accept": false, - "reason": "Both are Custom Diffusion training bugs, but one is about concept lists and the other is a dtype mismatch." + "reason": "Both are Custom Diffusion, but one is concept-list behavior and the other is a dtype mismatch; not the same bug." }, { "left": "issue:2674", "right": "issue:5604", "accept": false, - "reason": "Both mention LoRA, but one is safetensors availability and the other is unfusing multiple LoRA; different functionality." + "reason": "Both concern LoRA handling, but one is safetensors packaging and the other is inability to unfuse multiple LoRAs; distinct problems." }, { "left": "issue:7889", "right": "issue:8645", "accept": false, - "reason": "Both touch single_file_utils.py, but one reports loader bugs and the other a missing helper function; not the same concrete problem." + "reason": "Same file area, but one is a general single_file_utils bug report and the other is a missing helper function request." }, { "left": "issue:1367", "right": "issue:3250", "accept": false, - "reason": "Scheduler config serialization is unrelated to a pipeline failing to run." + "reason": "Completely different scheduler config save bug versus pipeline execution failure." }, { "left": "issue:10925", "right": "issue:9227", "accept": false, - "reason": "Both involve FLUX DreamBooth/LoRA training, but one targets CLIPTextModel finetuning and the other a runtime problem running the script; not enough evidence they are the same bug." + "reason": "Both are FLUX DreamBooth training reports, but they point to different scripts/models and different failure modes." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The only accepted soft edge, issue:10180 \u2194 issue:11298, is supported by the titles and clearly centers on the same Flux/multiple-LoRA loading failure mode. The rest of the pairwise rejections are appropriately cautious." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -23087,10 +23099,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -23105,156 +23117,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is not a true duplicate cluster: the soft pairs mostly share only broad subsystem keywords (Flux, LoRA, torch.compile, OOM, ControlNet, etc.) but describe different failures, features, or pipelines. I would not merge any of the suggested pairs.", - "confidence": 0.95, - "canonical_issue_reason": "issue:10186 is the clearest self-contained bug report here: a concrete FluxFillPipeline import failure with an explicit exception and straightforward reproduction.", - "canonical_pr_reason": null, - "best_issue_reason": "issue:10186 is the strongest canonical candidate among the set because it is specific, reproducible, and actionable; however, the overall item set is heterogeneous, so it is only a weak representative of the group.", - "best_pr_reason": null, + "summary": "These items are mostly unrelated issues and feature requests across different pipelines, training scripts, and runtime failures. The soft pairs share broad keywords but not the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue is a strong canonical representative because the cluster is heterogeneous; the soft pairs do not point to one shared defect or feature.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "No best issue: any one choice would overfit to a narrow subtopic and ignore the rest of the mixed set.", + "best_pr_reason": "No pull requests are present, so there is no PR candidate to select.", "soft_edge_verdicts": [ { "left": "issue:1617", "right": "issue:2444", "accept": false, - "reason": "Both are DreamBooth-related, but one is a missing-args CLI error and the other is an inpaint script subprocess failure; different bugs and code paths." + "reason": "Different scripts and failures: missing CLI args vs inpaint training subprocess error." }, { "left": "issue:10186", "right": "issue:10833", "accept": false, - "reason": "Same pipeline name, but one is an import/export issue and the other is a quantization matrix-shape runtime bug." + "reason": "Same pipeline name, but one is an import error and the other is a quantization matrix-shape bug." }, { "left": "issue:4731", "right": "issue:7159", "accept": false, - "reason": "Both mention torch.compile, but one is multi-ControlNet on SDXL and the other is fused LoRA; different failing paths." + "reason": "Both involve torch.compile, but they hit different code paths and symptoms." }, { "left": "issue:11033", "right": "issue:11907", "accept": false, - "reason": "Unrelated problems: single-file SD1.5 loading versus Flux scheduler incompatibility." + "reason": "Unet loading from_single_file vs FLUX scheduler compatibility are unrelated bugs." }, { "left": "issue:11408", "right": "issue:9643", "accept": false, - "reason": "One is compiled-model adapter setting failure; the other is a feature/request about multiple ControlNets in Flux." + "reason": "Compiled-model adapter setting and multiple-ControlNet support are different problems." }, { "left": "issue:3565", "right": "issue:3694", "accept": false, - "reason": "OOM during training versus missing model/config files in an example run; not the same bug." + "reason": "OOM during training vs missing model/config files while running an example." }, { "left": "issue:1823", "right": "issue:259", "accept": false, - "reason": "A UI/comment swap issue and a progress-bar fix are unrelated." + "reason": "A comment swap in depth2img is unrelated to a progress bar fix." }, { "left": "issue:2371", "right": "issue:761", "accept": false, - "reason": "Different DreamBooth failures: output content issue versus 8-bit Adam optimizer training crash." + "reason": "DreamBooth output missing objects is not the same as 8-bit Adam optimizer failure." }, { "left": "issue:5432", "right": "issue:5846", "accept": false, - "reason": "Blank validation images and resume-from-checkpoint errors are distinct LoRA training issues." + "reason": "Blank validation images and resume-from-checkpoint errors are different LoRA training issues." }, { "left": "issue:7982", "right": "issue:9953", "accept": false, - "reason": "from_pipe with MultiControlNetModel is unrelated to moving a quantized component to CUDA." + "reason": "from_pipe with MultiControlNetModel and moving a quantized pipeline to CUDA are distinct failures." }, { "left": "issue:10893", "right": "issue:11569", "accept": false, - "reason": "Different feature requests for different models and integrations; not a duplicate." + "reason": "Different feature requests for different models and pipelines." }, { "left": "issue:6823", "right": "issue:6839", "accept": false, - "reason": "Generic ControlNet support request versus from_single_file support for StableVideoDiffusionPipeline." + "reason": "General controlnet usage problem vs StableVideoDiffusion from_single_file support." }, { "left": "issue:11023", "right": "issue:8909", "accept": false, - "reason": "Blurred outputs in a custom pipeline and saving text_encoder_ti checkpoints are separate issues." + "reason": "Blurred inference output and saving text_encoder_ti in training are unrelated." }, { "left": "issue:1565", "right": "issue:2089", "accept": false, - "reason": "GPU selection/OOM on inference versus DreamBooth training OOM with AltDiffusion; different root causes." + "reason": "Both mention OOM, but one is GPU selection behavior and the other is DreamBooth optimizer memory use." }, { "left": "issue:5313", "right": "issue:6809", "accept": false, - "reason": "LoRA fuse reversibility and PEFT-based loading regressions are distinct adapter lifecycle problems." + "reason": "Non-revertible LoRA fuse and PEFT load regressions are different adapter-management bugs." }, { "left": "issue:6561", "right": "issue:8469", "accept": false, - "reason": "Missing IP-Adapter config file versus corrupt face outputs; file-loading and quality bugs differ." + "reason": "Missing IP-Adapter config file and corrupted faces are not the same underlying issue." }, { "left": "issue:10639", "right": "issue:12358", "accept": false, - "reason": "Support requests for different Lumina models; related theme but not the same artifact." + "reason": "Separate feature requests for different Lumina models, not one duplicated request." }, { "left": "issue:11014", "right": "issue:11225", "accept": false, - "reason": "Both are OOM reports, but one is general Colab inference memory and the other is SD3 controlnet validation-step memory." + "reason": "General CUDA OOM on Colab and SD3 controlnet validation OOM are different scenarios." }, { "left": "issue:10937", "right": "issue:12082", "accept": false, - "reason": "torch.compile VAE encode errors and post-compile VAE memory growth are different symptoms and likely different fixes." + "reason": "torch.compile error on vae.encode is different from increased VAE memory after compile." }, { "left": "issue:4206", "right": "issue:4566", "accept": false, - "reason": "ControlNet training breakage in main versus SDXL LoRA checkpoint resume failure are unrelated." + "reason": "Controlnet training on main branch and LoRA checkpoint resume are unrelated." }, { "left": "issue:12120", "right": "issue:9599", "accept": false, - "reason": "A training how-to for distilled Flux models is not the same as asking for a LoRA-only Flux example." + "reason": "Training guidance for distilled FLUX and a request for a LoRA-only FLUX example are not the same bug." }, { "left": "issue:4664", "right": "issue:5516", "accept": false, - "reason": "Both involve LoRA, but one is SDXL pipeline class behavior and the other is t2i-adapter pipeline incompatibility." + "reason": "SDXL LoRA loading mismatch and LoRA with t2i adapters are different integration bugs." }, { "left": "issue:1367", "right": "issue:861", "accept": false, - "reason": "Scheduler configuration persistence and ckpt-to-diffusers conversion are unrelated." + "reason": "Scheduler save_config with trained_betas is unrelated to ckpt-to-diffusers conversion." }, { "left": "issue:6252", "right": "issue:6368", "accept": false, - "reason": "Checkpoint loading failure and unsupported LoRA alpha metadata are different loading/format issues." + "reason": "Checkpoint not loading and unsupported LoRA layer alpha are different checkpoint-format problems." } ] }, @@ -23357,10 +23369,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -23375,162 +23387,165 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft pair appears to be the same underlying bug/change; most are only loosely related by subsystem or model family. I rejected all candidate duplicate edges.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "Mostly unrelated issues with a few loose topical similarities. None of the pairs are strong enough to call duplicates; the SDXL blank-validation pair is related, but not confirmed as the same bug.", + "confidence": 0.77, + "canonical_issue_reason": "Issue 6520 is the best representative of the SDXL blank-validation symptom because it explicitly names `log_validation` and the broader training-adapter context, but the evidence is still too weak to treat it as a confirmed duplicate of 5432.", "canonical_pr_reason": null, - "best_issue_reason": "issue:6586 is the strongest issue candidate overall because it is an open tracker with broad, central scope (`use micro-conditioning for the SDXL trainers`), but the set is too heterogeneous to form a real duplicate cluster.", + "best_issue_reason": "6520 is the most diagnostic issue for the shared SDXL validation-output theme, even though the pair does not clear duplicate threshold.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:6586", "right": "issue:6891", "accept": false, - "reason": "Different tracker topics: SDXL micro-conditioning vs new model card utilities; no shared failing code path." + "reason": "Both are trainer-related tracker issues, but one is about micro-conditioning and the other about model-card saving utilities; different changes." }, { "left": "issue:10016", "right": "issue:5466", "accept": false, - "reason": "Both are loading/shape-mismatch reports, but on different models and paths (`from_single_file` pos_embed vs LoRA down.weight)." + "reason": "Both involve load-time shape mismatches, but they affect different models and different parameters (`pos_embed` vs LoRA `down.weight`)." }, { "left": "issue:2772", "right": "issue:4935", "accept": false, - "reason": "`train_controlnet.py` local dataset validation bug is unrelated to SDXL distributed OOM during controlnet training." + "reason": "One is about local dataset handling in `train_controlnet.py`; the other is an SDXL distributed-training OOM. Same area, different bug." }, { "left": "issue:7180", "right": "issue:7466", "accept": false, - "reason": "Different concrete failures: cpu offload issue in SDXL controlnet img2img vs torch.compile failure with multi-controlnet inpaint." + "reason": "Both touch SDXL ControlNet, but one is CPU offload failure and the other is `torch.compile` with multi-ControlNet inpaint; not the same defect." }, { "left": "issue:4724", "right": "issue:7871", "accept": false, - "reason": "Img2Img checkpoint behavior issue vs blockwise scales with a specific LoRA file; not the same bug." + "reason": "Different problems: SDXL img2img checkpoint behavior versus blockwise scales with a specific LoRA file." }, { "left": "issue:5575", "right": "issue:6790", "accept": false, - "reason": "Hidden-size configuration question in LoRA training is not the same as an unexpected key error in LCM example training." + "reason": "Unrelated: a LoRA hidden-size question versus an LCM training state-dict mismatch." }, { "left": "issue:11225", "right": "issue:4925", "accept": false, - "reason": "Both are OOMs, but on different models/training setups (SD3 validation vs SDXL distributed training)." + "reason": "Both are OOM reports, but they involve different trainers and different validation/training contexts." }, { "left": "issue:1889", "right": "issue:2676", "accept": false, - "reason": "Scheduler variant request/question is not the same as a broader scheduler-naming lookup; no concrete shared bug." + "reason": "Scheduler-related, but one is a feature request and the other is a naming question; not duplicates." }, { "left": "issue:12175", "right": "issue:12216", "accept": false, - "reason": "Both mention Qwen image quality, but one is normal generation and the other image-edit inference; different issue surfaces." + "reason": "Both concern Qwen image quality, but one is generic Qwen-Image quality and the other is Qwen-Image-Edit versus ComfyUI; too different." }, { "left": "issue:11584", "right": "issue:11791", "accept": false, - "reason": "Both involve group offloading, but one is torch.compile incompatibility and the other LoRA loading failure; different code paths." + "reason": "Both mention group offloading, but one is a `torch.compile` incompatibility and the other is LoRA weights failing to load after offloading." }, { "left": "issue:2878", "right": "issue:5873", "accept": false, - "reason": "General PyTorch/ONNX converter problem vs a specific unsupported `scaled_dot_product_attention` export error." + "reason": "Both are ONNX conversion issues, but one is broad PyTorch 2.0 converter discussion and the other is a specific unsupported operator export." }, { "left": "issue:12359", "right": "issue:9576", "accept": false, - "reason": "Documentation bug for `guidance_scale` is unrelated to passing `use_safetensors` to submodules when loading pipelines." + "reason": "Completely different topics: Chroma docs for `guidance_scale` versus `use_safetensors` not being passed to submodules." }, { "left": "issue:3061", "right": "issue:460", "accept": false, - "reason": "Reproducibility issue and ONNX runtime dependency issue are unrelated." + "reason": "Both are loading/reproducibility concerns, but the underlying failures are unrelated." }, { "left": "issue:10315", "right": "issue:9497", "accept": false, - "reason": "Different training failures on different models: cogvideo training error vs Flux Dreambooth matmul shape error." + "reason": "Different training errors for different models: CogVideo training versus Dreambooth Flux matmul shape failure." }, { "left": "issue:4615", "right": "issue:6794", "accept": false, - "reason": "ImportError for `StableDiffusionControlnetPipeline` is unrelated to `cast_training_params` import failure." + "reason": "Both are import/runtime errors, but they involve unrelated symbols and modules." }, { "left": "issue:5432", "right": "issue:6520", "accept": false, - "reason": "Both mention blank validation outputs, but they are different training scripts and likely different root causes." + "reason": "Both are related SDXL blank-validation reports, but the packet does not establish a shared `log_validation` code path or a true duplicate relationship." }, { "left": "issue:1305", "right": "issue:938", "accept": false, - "reason": "Completely different model/pipeline feature requests; no duplicate relationship." + "reason": "Different papers/features entirely: Direct Inversion versus Human Motion Diffusion Model." }, { "left": "issue:2883", "right": "issue:3728", "accept": false, - "reason": "Image-to-video pipeline request vs shifted diffusion pipeline request; different feature work." + "reason": "Different pipeline requests: image-to-video versus shifted diffusion text-to-image." }, { "left": "issue:6224", "right": "issue:6536", "accept": false, - "reason": "TorchDynamo fine-tuning issue and UNet3D out-of-memory are not the same failure mode." + "reason": "Different training/runtime failures: TorchDynamo issue in text-to-image fine-tuning versus UNet3DConditionModel OOM." }, { "left": "issue:10428", "right": "issue:9112", "accept": false, - "reason": "Different models and hardware contexts: Flux on Ascend NPU vs Auraflow tensor shape mismatch." + "reason": "Both are inference tensor-shape errors, but they affect different models and likely different assumptions." }, { "left": "issue:3706", "right": "issue:5576", "accept": false, - "reason": "A new model request (Wuerstchen) is unrelated to a scheduler feature request (FreeNoise)." + "reason": "Different feature requests/models: Wuerstchen pipeline versus FreeNoise scheduler request." }, { "left": "issue:6549", "right": "issue:7766", "accept": false, - "reason": "Warning logs with IP-Adapter+LoRA are unrelated to a usage/how-to question for FaceID Plus." + "reason": "Both mention IP-Adapter, but one is a LoRA warning log and the other is a usage question." }, { "left": "issue:1500", "right": "issue:1611", "accept": false, - "reason": "Library-version incompatibility for a community pipeline is unrelated to a model-specific pipeline regression." + "reason": "Both are old pipeline regressions, but the affected pipelines and breakage descriptions differ." }, { "left": "issue:10863", "right": "issue:9343", "accept": false, - "reason": "Both are loading-model complaints, but one is local safetensors loading and the other is FLUX with `low_cpu_mem_usage=False` and `ignore_mismatched_sizes=True`." + "reason": "Both are load/safetensors-related, but one is a local safetensors file error and the other is a FLUX loading issue with different flags." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is mostly cautious, but the one accepted pair is too strong for the packet: 5432 and 6520 both involve SDXL validation behavior, yet the packet only gives titles and does not explicitly establish a shared `log_validation` code path or a true duplicate relationship. The packet\u2019s deterministic signal also marks that pair as false. A safer phrasing would be that they are related reports, not an accepted duplicate pair." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -23626,10 +23641,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -23644,156 +23659,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is highly heterogeneous: compile/inference crashes, training-script bugs, scheduler issues, and model-specific quality reports. The soft pairs mostly share only broad subsystem vocabulary, not the same concrete bug or change, so none should be merged as duplicates.", - "confidence": 0.91, - "canonical_issue_reason": "issue:5934 is the broadest and most central report in the set (general torch.compile failure) and has enough discussion to act as a fallback umbrella, though the cluster overall is not truly coherent.", + "summary": "This cluster is mostly heterogeneous: several pairs are only loosely related by subsystem (torch.compile, SDXL/SD3 training, schedulers, black-image symptoms) but not the same underlying bug. I would reject all soft edges and treat the issue set as separate reports, with issue 5934 as the broadest representative of the compile-error subset.", + "confidence": 0.89, + "canonical_issue_reason": "issue:5934 is the broadest and most generic report in the set (\u201cError when using torch.compile\u201d), so it best represents the compile-related subset even though it is not a duplicate of the other issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:5934 is the most reusable anchor among these items because it is generic and discussion-rich; it\u2019s the closest thing to a cluster representative, but only as a loose umbrella, not a true duplicate target.", + "best_issue_reason": "issue:5934 is the most central, high-level issue title in the cluster and the best anchor for triage because it cleanly captures the recurring torch.compile failures without overfitting to one narrow stack trace.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4375", "right": "issue:5934", "accept": false, - "reason": "Both mention torch.compile, but one is a compile-time inference crash and the other is a generic compile error; different failure modes and code paths." + "reason": "Both involve torch.compile, but one is a crash during inference and the other is a generic compile error; different concrete failures." }, { "left": "issue:12175", "right": "issue:9904", "accept": false, - "reason": "One is a Qwen Image quality complaint, the other is Flux shape rounding; unrelated symptoms and pipelines." + "reason": "Both mention image quality/shape issues, but they affect different models and different code paths." }, { "left": "issue:3357", "right": "issue:7159", "accept": false, - "reason": "Both involve torch.compile, but the reported errors are different and tied to different LoRA/graph behavior." + "reason": "Both are torch.compile problems, but the reported errors differ (hasattr/TensorVariable vs fused LoRA compile failure)." }, { "left": "issue:3198", "right": "issue:4954", "accept": false, - "reason": "Different textual inversion load failures: safetensors loading vs pickle unpickling; not the same defect." + "reason": "Both are textual inversion loading failures, but one is safetensors loading and the other is an unpickling error." }, { "left": "issue:4724", "right": "issue:5931", "accept": false, - "reason": "Img2Img behavior with an SDXL checkpoint is unrelated to a StableDiffusionPipeline nondeterminism report." + "reason": "One is an SDXL img2img correctness issue; the other is nondeterministic generation. Same area, not the same bug." }, { "left": "issue:3279", "right": "issue:3284", "accept": false, - "reason": "Both are DreamBooth LoRA training issues, but one is checkpoint-resume and the other is an accelerate launch failure; different bugs." + "reason": "Both touch dreambooth LoRA training, but one is checkpoint resume and the other is accelerate launch/runtime setup." }, { "left": "issue:7041", "right": "issue:8552", "accept": false, - "reason": "Both are crashes, but on different pipelines and likely different low-level causes." + "reason": "Both are crashes, but the pipelines and triggering conditions are different." }, { "left": "issue:8649", "right": "issue:8828", "accept": false, - "reason": "One is an SD3 prompt/image-count API regression, the other is a training memory/error report; not the same issue." + "reason": "Both are SD3 training-related, but one is a parameter-handling regression and the other is a resource/training error." }, { "left": "issue:10671", "right": "issue:11023", "accept": false, - "reason": "Deterministic dtype/device lookup in DiffusionPipeline is unrelated to blurred output in a custom LPW pipeline." + "reason": "Different symptoms and likely different causes: dtype/device determinism vs blurry custom-pipeline output." }, { "left": "issue:4935", "right": "issue:5432", "accept": false, - "reason": "One asks about using AttnProcsLayers in an SDXL LoRA script, the other reports blank validation images; different problems." + "reason": "Both concern SDXL LoRA training, but one is about a helper class usage question and the other about blank validation images." }, { "left": "issue:6232", "right": "issue:6527", "accept": false, - "reason": "Both concern add_noise, but they target different schedulers (DPMSolverSinglestep vs EulerDiscrete) and likely different bugs." + "reason": "Same general scheduler area, but different schedulers and likely different add_noise bugs." }, { "left": "issue:8507", "right": "issue:8759", "accept": false, - "reason": "Training text-encoder failure in SD3 is unrelated to an SD3 image-processor cast warning." + "reason": "One is SD3 train_dreambooth text-encoder failure; the other is an image processor warning/cast issue." }, { "left": "issue:6095", "right": "issue:6574", "accept": false, - "reason": "Both are SVD pipeline reports, but one is an Euler-discrete index error and the other is a warning about expected input range; different code paths." + "reason": "Both are SVD pipeline issues, but one is an IndexError in scheduling and the other is a dtype/range warning." }, { "left": "issue:8511", "right": "issue:8839", "accept": false, - "reason": "Unloading SD3 text encoders and an SD3 LoRA weight error are different training issues." + "reason": "Both are SD3 training issues, but the concrete problems differ: text encoders not unloading vs LoRA weight error." }, { "left": "issue:4375", "right": "issue:5227", "accept": false, - "reason": "Both involve torch.compile, but one is an inference crash and the other is a LoRA weight-loading compile failure; not the same bug." + "reason": "Both mention torch.compile, but one is a generic inference crash and the other is a specific LoRA-loading compile incompatibility." }, { "left": "issue:6124", "right": "issue:8732", "accept": false, - "reason": "An FP16 gradient unscale notebook error is unrelated to checkpoint-saving problems in advanced training." + "reason": "Both are training-related, but one is FP16 unscale failure and the other is checkpoint-saving behavior." }, { "left": "issue:6415", "right": "issue:8827", "accept": false, - "reason": "One requests support for FaceID variants, the other reports an error when using an existing FaceID adapter; feature request vs runtime bug." + "reason": "One is a feature-support request for FaceID plus; the other is an actual runtime error with ip-adapter-faceid_sdxl." }, { "left": "issue:5563", "right": "issue:7347", "accept": false, - "reason": "Both are training failures, but they describe different scripts, versions, and symptoms; too broad to treat as one bug." + "reason": "Both are training errors, but the titles are too broad and do not indicate the same concrete failure." }, { "left": "issue:7185", "right": "issue:9515", "accept": false, - "reason": "A general PyTorch version-compatibility question is not the same as a specific SD3 RMSNorm architecture omission." + "reason": "One is a pytorch version support question; the other is a missing RMSNorm implementation detail in SD3 attention." }, { "left": "issue:4985", "right": "issue:7538", "accept": false, - "reason": "An SDXL InstructPix2Pix OOM report and a custom dataset training issue do not describe the same underlying failure." + "reason": "Both are training issues, but one is an OOM and the other is a custom dataset/style training problem." }, { "left": "issue:11674", "right": "issue:11779", "accept": false, - "reason": "A Wan VACE feature request about ref images/control videos is unrelated to a mask-padding bug in the same pipeline family." + "reason": "Feature request vs bug report; they concern different changes in Wan VACE." }, { "left": "issue:5889", "right": "issue:7223", "accept": false, - "reason": "General stable video diffusion discussion and a Latte model request are different change requests." + "reason": "Different model requests: stable video diffusion discussion vs Latte addition request." }, { "left": "issue:4305", "right": "issue:4936", "accept": false, - "reason": "A torch.compile graph-break report is not the same as DreamBooth LoRA resume being broken after 0.19.0." + "reason": "Both are regressions around 0.19.0, but one is a torch.compile graph break and the other is DreamBooth LoRA resume failure." }, { "left": "issue:1556", "right": "issue:1614", "accept": false, - "reason": "Both produce black images, but one is tied to an NSFW/GPU path and the other to autocast with SD 2.1; insufficiently specific to merge." + "reason": "Both produce black images, but they appear to stem from different runtime conditions and code paths." } ] }, @@ -23895,10 +23910,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -23913,156 +23928,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the items share broad diffusion-model themes (training, loading, attention, memory, LoRA) but the soft pairs point to different models, pipelines, and failure modes. I would not merge any of the candidate pairs as duplicates.", - "confidence": 0.91, - "canonical_issue_reason": null, + "summary": "The cluster is highly heterogeneous: the suggested pairs mostly share only broad subsystem/theme similarity (training, loading, attention, memory, imports) rather than the same underlying bug or change. No pull requests are present. I\u2019d treat issue 11223 as the best representative of the closest concrete subcluster, but with low duplicate confidence overall.", + "confidence": 0.69, + "canonical_issue_reason": "issue:11223 is the most concrete and actionable report in the closest apparent subcluster (WAN + LoRA + OOM), with a specific failure mode and enough detail to anchor triage better than the more generic or feature-oriented items.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:11223 is the strongest representative artifact here: it describes a specific reproducible bug, is more detailed than neighboring WAN/LoRA items, and is the clearest candidate for a canonical issue among the set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:7392", "right": "issue:7908", "accept": false, - "reason": "Both mention aspect ratios, but one is a question about training SDXL and the other is a feature request for bucketing; not the same bug." + "reason": "Both concern aspect ratios in training, but one is a how-to/problem report and the other is a feature request for bucketing; not the same bug/change." }, { "left": "issue:2364", "right": "issue:5565", "accept": false, - "reason": "Both are training-script failures, but they affect different scripts and likely different causes." + "reason": "Different training scripts and different failure reports; only a broad \u201ccan\u2019t run training\u201d similarity." }, { "left": "issue:3250", "right": "issue:7306", "accept": false, - "reason": "Different pipelines and symptoms: KDiffusionPipeline startup failure vs float16 handling in LEditsPPPipelineStableDiffusionXL." + "reason": "Different pipelines and different defects: one can\u2019t run a pipeline at all, the other is a float16-specific issue." }, { "left": "issue:7159", "right": "issue:8518", "accept": false, - "reason": "Both involve torch.compile, but the concrete failures differ: fused LoRA compile failure vs a Dynamo unsupported class variable error." + "reason": "Both involve torch.compile, but the model paths and error conditions differ materially; not one fixable code-path bug." }, { "left": "issue:11088", "right": "issue:13105", "accept": false, - "reason": "One is a gguf load error for wan2.1, the other is noisy output at inference time; different stages and bugs." + "reason": "One is a gguf load error and the other is noisy output from a WAN model; different symptoms and likely different root causes." }, { "left": "issue:4213", "right": "issue:4488", "accept": false, - "reason": "Both concern model loading/conversion, but they target different code paths and incompatibilities." + "reason": "Both touch SDXL loading, but they target different APIs and bug conditions; too broad to be the same underlying issue." }, { "left": "issue:4566", "right": "issue:6252", "accept": false, - "reason": "Checkpoint-related LoRA problems, but one is resume-from-checkpoint training and the other is weight loading; not clearly the same defect." + "reason": "Checkpoint resume and LoRA weight loading are related but distinct behaviors with different failure modes." }, { "left": "issue:6112", "right": "issue:9889", "accept": false, - "reason": "Same model family, but one reports noisy output while the other is a segmentation fault; different underlying issues." + "reason": "Both mention sdxl-turbo problems, but one is output corruption/noise while the other is a segfault during inference." }, { "left": "issue:5466", "right": "issue:7003", "accept": false, - "reason": "Both are shape-mismatch errors, but they involve different pipelines, tensor shapes, and likely different adapter loading paths." + "reason": "Different pipelines and different shape-mismatch contexts; not the same concrete bug." }, { "left": "issue:2015", "right": "issue:3826", "accept": false, - "reason": "A generic pipeline question versus a new pipeline proposal; no shared bug or fix." + "reason": "Separate pipeline ideas in different domains; no shared underlying bug or implementation change." }, { "left": "issue:3816", "right": "issue:4356", "accept": false, - "reason": "Both are import-related, but they reference different missing symbols/modules and likely different dependency breakages." + "reason": "Both are import-related errors, but they involve different missing symbols/modules and different fixes." }, { "left": "issue:5540", "right": "issue:6028", "accept": false, - "reason": "Both are OOM reports, but for different models and contexts; symptom-only overlap is not enough." + "reason": "Only the generic symptom overlaps (OOM); the models and circumstances are different." }, { "left": "issue:3674", "right": "issue:3700", "accept": false, - "reason": "Different training-script issues: metadata parsing vs unsupported min_snr argument." + "reason": "Same script, but one is about metadata parsing and the other about an argument not working; distinct issues." }, { "left": "issue:4566", "right": "issue:4985", "accept": false, - "reason": "Different SDXL training problems: checkpoint resume vs InstructPix2Pix OOM." + "reason": "Both are training/LoRA-related, but one is checkpoint resume and the other is an OOM during a different training flow." }, { "left": "issue:4000", "right": "issue:7782", "accept": false, - "reason": "Both involve conversion/loading, but one is ckpt-to-diffusers conversion and the other is safetensors file parsing." + "reason": "Both involve conversion/loading problems, but they concern different file formats and error causes." }, { "left": "issue:10421", "right": "issue:10569", "accept": false, - "reason": "Both mention memory, but one is a CPU leak when moving pipelines to multiple GPUs and the other is high CPU memory use for HunyuanVideo." + "reason": "Both are memory-related, but one is a multi-GPU pipeline leak and the other is high CPU memory use for a specific model." }, { "left": "issue:3422", "right": "issue:7363", "accept": false, - "reason": "Both mention safety checker, but one is a generic disable-safety question and the other is an SDXL-specific support request." + "reason": "Both mention the safety checker, but one is a general question and the other is a specific SDXL request; not the same bug/change." }, { "left": "issue:6749", "right": "issue:8885", "accept": false, - "reason": "Different community-example/install failures with unrelated root causes." + "reason": "Different community examples and distinct setup/runtime failures; no shared underlying defect." }, { "left": "issue:11182", "right": "issue:9525", "accept": false, - "reason": "One is an unnecessary download on BF16 load; the other is LoRA scale having no effect in Flux. Different behaviors." + "reason": "Unnecessary download for BF16 loading and Flux lora_scale behavior are unrelated issues." }, { "left": "issue:11222", "right": "issue:11223", "accept": false, - "reason": "Feature request for more WAN LoRAs vs an OOM bug with WAN LoRas; related area, but not the same issue." + "reason": "Related WAN/LoRA context, but one is a feature request for more LoRAs and the other is an OOM bug under layer-wise upcasting." }, { "left": "issue:12575", "right": "issue:13176", "accept": false, - "reason": "Both touch attention behavior, but one is Chroma backend checks and the other is a UNet attention error; different components." + "reason": "Both are attention-related errors, but they appear to hit different components and failure surfaces." }, { "left": "issue:1704", "right": "issue:929", "accept": false, - "reason": "A general attention-block question versus a specific calculation/optimization discussion; not the same bug." + "reason": "A general question about an attention block is not the same as a specific implementation/performance concern." }, { "left": "issue:1823", "right": "issue:776", "accept": false, - "reason": "Different image-to-image problems with different symptoms and triggers." + "reason": "Different Stable Diffusion bugs: swapped comments versus batch-size behavior in image-to-image." }, { "left": "issue:4708", "right": "issue:5892", "accept": false, - "reason": "Both are import errors, but the missing modules/scope are too generic and not evidence of one underlying defect." + "reason": "Both are generic import errors, but they are not the same missing module/symbol or fix." } ] }, @@ -24165,10 +24180,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -24183,162 +24198,165 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very noisy and spans many unrelated issues; none of the soft-similarity pairs are strong enough to merge. The closest overlap is the FLUX multi-GPU/deepspeed training pair, but even that is not clearly the same bug.", - "confidence": 0.42, - "canonical_issue_reason": "If forced to pick one anchor, issue 9484 is the broadest and most central modern FLUX training report, and it is the closest match to another issue in the set (10026).", + "summary": "The cluster is mostly topical overlap rather than true duplication. The packet provides only soft-similarity signals, so none of the pairs should be promoted to accepted duplicate status; even the closest-looking MultiControlNet pair remains unconfirmed.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 8405 is the clearest anchor for the narrow MultiControlNet/multiple-ControlNet tensor-size symptom, but it should be treated only as the most representative soft match, not a confirmed duplicate target.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 9484 is the best canonical candidate because it is broad, active, and directly overlaps the strongest related complaint in the cluster; the other items are either narrower or clearly about different subsystems.", + "best_issue_reason": "Issue 8405 is the best issue representative because its title states the shared failure mode most directly and broadly, even though the packet does not support treating it as a true duplicate cluster center.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:7168", "right": "issue:8626", "accept": false, - "reason": "Both involve IP-Adapter usage, but one is about getting image embeddings and the other about guidance for multiple images/masks; different problems." + "reason": "Both mention IP-Adapter usage, but the concrete problems differ: image-embed shape handling vs guidance for multiple images/masks." }, { "left": "issue:5628", "right": "issue:6295", "accept": false, - "reason": "Scheduler offset behavior and DPM++ artifacts are related only at a very high level; different concrete bugs and likely different fixes." + "reason": "Scheduler/artifact wording overlaps, but the samplers and failure modes are different." }, { "left": "issue:4954", "right": "issue:7782", "accept": false, - "reason": "Both mention unpickling, but one is textual inversion pickle loading and the other is converting a safetensors file; not the same issue." + "reason": "Both involve unpickling errors, but the contexts are different: textual inversion loading vs safetensors conversion." }, { "left": "issue:6149", "right": "issue:6561", "accept": false, - "reason": "Both are IP-Adapter load errors, but they reference different missing files/models and likely different loading paths." + "reason": "Both are IP-Adapter load errors, but the titles do not indicate the same missing file or same root cause." }, { "left": "issue:5398", "right": "issue:5429", "accept": false, - "reason": "Hacktoberfest task issues for different features; no shared bug or code path." + "reason": "Different hacktoberfest tasks with different code paths: DreamBooth/DeepSpeed vs model support without VAE." }, { "left": "issue:8515", "right": "issue:9392", "accept": false, - "reason": "Duplicate code cleanup and scheduler SNR-shift feature request are unrelated." + "reason": "Duplicate-code cleanup is unrelated to a scheduler feature request." }, { "left": "issue:9063", "right": "issue:9141", "accept": false, - "reason": "Both are FLUX-related, but one is a tokenizer instantiation error and the other is a missing from_single_file API; separate failures." + "reason": "Both are FLUXPipeline-related, but one is tokenizer instantiation and the other is missing from_single_file support." }, { "left": "issue:2459", "right": "issue:6911", "accept": false, - "reason": "Xformers attention in training and 3D class-label handling are different code paths and different bugs." + "reason": "Different APIs and bug classes: xformers attention in a training script vs class-label handling in a 3D UNet." }, { "left": "issue:11214", "right": "issue:6761", "accept": false, - "reason": "Both are unexpected-keyword TypeErrors, but they come from different models/APIs and do not indicate the same fix." + "reason": "Unexpected keyword errors, but on different functions and different components." }, { "left": "issue:6967", "right": "issue:7239", "accept": false, - "reason": "Both are shape mismatches in SDXL-related training, but they occur in different scripts and tensor interfaces; not clearly one underlying bug." + "reason": "Both are SDXL training shape mismatches, but the scripts and mismatched tensors appear different." }, { "left": "issue:10180", "right": "issue:11653", "accept": false, - "reason": "One is about loading multiple LoRAs with Flux Control LoRA, the other about supporting .alpha keys in HiDream LoRAs; different model families and fixes." + "reason": "Different model families and fixes: Flux Control LoRA multi-lora loading vs HiDream .alpha key support." }, { "left": "issue:7982", "right": "issue:9971", "accept": false, - "reason": "MultiControlNet from_pipe support and DEISMultistepScheduler on FLUX are unrelated features/issues." + "reason": "MultiControlNet pipeline conversion and a FLUX scheduler issue are different code paths." }, { "left": "issue:5442", "right": "issue:7223", "accept": false, - "reason": "Different video-generation feature proposals with no shared bug or concrete code path." + "reason": "They are both video-generation topics, but not the same bug or feature request." }, { "left": "issue:10421", "right": "issue:8464", "accept": false, - "reason": "CPU memory leak in multi-GPU pipeline movement and SEGA seed/offload failure are not the same symptom or fix." + "reason": "A CPU memory leak during multi-GPU pipeline moves is not the same as SEGA offload failing when a seed is set." }, { "left": "issue:5529", "right": "issue:8659", "accept": false, - "reason": "Both are training-script problems, but one is a generic bug in train_text_to_image_sdxl.py while the other is specifically checkpoint resume/inference loading." + "reason": "Both involve training scripts, but the failure modes are different: SDXL script bug vs checkpoint resume/inference loading." }, { "left": "issue:10026", "right": "issue:9484", "accept": false, - "reason": "Both are FLUX multi-GPU/deepspeed training reports, but one is ControlNet memory scaling and the other is dreambooth training setup; close but not clearly one exact bug." + "reason": "Both are FLUX multi-GPU/DeepSpeed-related, but one is ControlNet memory behavior and the other is DreamBooth training." }, { "left": "issue:4708", "right": "issue:5023", "accept": false, - "reason": "General import error versus a specific WuerstchenCombinedPipeline import regression; not the same issue." + "reason": "Import errors in different modules/pipelines; no evidence of a shared underlying cause." }, { "left": "issue:12609", "right": "issue:12699", "accept": false, - "reason": "Both involve Wan 2.2, but one is NaNs in latents and the other is quality degradation after repeated prompts with a lighting LoRA; different failure modes." + "reason": "Both mention wan 2.2, but one is NaNs in latents and the other is quality degradation over repeated prompts." }, { "left": "issue:4830", "right": "issue:8405", "accept": false, - "reason": "Both concern multiple ControlNets, but one is a tensor-size bug in MultiControlNet and the other is a broader use-of-multiple-controlnets error; not enough evidence they are the same fix." + "reason": "Very similar MultiControlNet tensor-size wording, but the packet only marks soft similarity and does not support calling this an accepted duplicate." }, { "left": "issue:2091", "right": "issue:7133", "accept": false, - "reason": "Both involve resuming SDXL training, but one is a multi-GPU pipeline resume issue and the other is a text-encoder size mismatch; different concrete failures." + "reason": "Both are resume-from-checkpoint issues, but they affect different scripts and different failure modes." }, { "left": "issue:1976", "right": "issue:2321", "accept": false, - "reason": "Tokenizer file resolution and EMA weight copying are separate bugs in different parts of model loading/inference." + "reason": "A missing tokenizer/config file lookup is different from a model-copying/ema_weights logic issue." }, { "left": "issue:3651", "right": "issue:3901", "accept": false, - "reason": "Multi-GPU dreambooth failure and missing requirements.txt are unrelated." + "reason": "A multi-GPU DreamBooth runtime issue is unrelated to a missing requirements.txt file." }, { "left": "issue:1013", "right": "issue:1195", "accept": false, - "reason": "StableDiffusionMegaPipeline compatibility and fp16 memory-efficient attention are different issues." + "reason": "Pipeline import/versioning issues and fp16 memory-efficient attention issues are different bugs." }, { "left": "issue:10139", "right": "issue:11033", "accept": false, - "reason": "Both are UNet/loading-related, but one is pipeline init using sample_size and the other is from_single_file loading for SD1.5; different code paths." + "reason": "Both concern loading/init, but sample_size handling and from_single_file loading are distinct problems." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "The summary is not conservative enough: the packet marks every soft pair, including issue:4830 vs issue:8405, with `deterministic_accept: false`. The analyst overstates that this is an accepted duplicate pair and that it is the only accepted soft duplicate. The rest of the writeup is broadly reasonable, but it should not elevate any pair to accepted/true-duplicate status without packet support." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -24434,10 +24452,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -24452,156 +24470,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a duplicate set; it mixes many unrelated bugs across different models, schedulers, devices, and training scripts. The soft pairs mostly share only generic crash/training wording, not the same concrete code-path problem.", - "confidence": 0.97, + "summary": "These items are mostly superficially similar bug reports across different models, pipelines, schedulers, and training scripts, not true duplicates. All soft-edge pairs should be rejected.", + "confidence": 0.92, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No single issue is a strong global canonical here; the set is too heterogeneous and the pairwise similarity is mostly generic wording rather than the same bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:7629", "right": "issue:9546", "accept": false, - "reason": "Different failures: SVD OOM on PyTorch 1.13.1 vs Flux ControlNet validation OOM. No shared code path." + "reason": "Both mention OOM, but one is SVD on PyTorch 1.13.1 and the other is Flux ControlNet validation OOM; different model and code path." }, { "left": "issue:6444", "right": "issue:7229", "accept": false, - "reason": "Both involve training/checkpointing, but one is a peft_config attribute error and the other is a checkpoint-saving issue in advanced training." + "reason": "Both involve training/checkpointing, but the failures are different: peft_config attribute error vs checkpoint save issue in advanced training." }, { "left": "issue:10798", "right": "issue:13043", "accept": false, - "reason": "Flux quantization/bitsandbytes device-combo bug vs sage_hub attention kernel support on RTX 5090. Different subsystems." + "reason": "Different problems: BitsAndBytes device-combination bug for Flux quantization vs missing kernel image on RTX 5090 with sage_hub attention." }, { "left": "issue:6382", "right": "issue:8469", "accept": false, - "reason": "IP adapter key error during SDXL LCM distillation vs corrupted faces from an IP-adapter face model. Distinct symptoms and causes." + "reason": "Both involve IP-Adapter, but one is a key error after LCM LoRA training and the other is broken face generation with a different adapter/model." }, { "left": "issue:6117", "right": "issue:6527", "accept": false, - "reason": "Same generic API name, but different schedulers and likely different implementations of add_noise." + "reason": "Scheduler bugs, but they are distinct classes: CMStochasticIterativeScheduler add_noise vs EulerDiscreteScheduler add_noise." }, { "left": "issue:3101", "right": "issue:713", "accept": false, - "reason": "Too generic to treat as the same bug; one is ControlNet example training, the other is an unspecified training error." + "reason": "The titles are too generic and the reports are not clearly the same concrete failure or code path." }, { "left": "issue:11024", "right": "issue:7593", "accept": false, - "reason": "Different IP-adapter failures: missing .to() on SiglipImageProcessor vs missing image_projection_layers on NoneType." + "reason": "Different issues: missing .to() on SiglipImageProcessor vs NoneType/image_projection_layers error in IP-Adapter image embedding prep." }, { "left": "issue:11581", "right": "issue:11651", "accept": false, - "reason": "Both are FLUX-related loading issues, but one is an unexpected dtype argument in CLIPVisionModelWithProjection and the other is a flux-fill-lora load failure." + "reason": "Both are FLUX/IP-related load failures, but one is an unexpected dtype argument and the other is a FluxControl loader problem." }, { "left": "issue:10139", "right": "issue:7982", "accept": false, - "reason": "Pipeline init sample_size handling vs from_pipe with MultiControlNetModel. Not the same bug." + "reason": "Unet sample_size pipeline-init bug is unrelated to MultiControlNetModel.from_pipe behavior." }, { "left": "issue:2132", "right": "issue:3279", "accept": false, - "reason": "Different DreamBooth issues: missing Accelerator.trackers vs resume-from-checkpoint errors." + "reason": "Different failures: Accelerator.trackers missing vs resume-from-checkpoint errors in dreambooth training." }, { "left": "issue:6382", "right": "issue:7722", "accept": false, - "reason": "Both mention IPAdapter, but one is an SDXL LCM training/inference key error and the other is an Animatediff-ControlNet integration error." + "reason": "Both mention IP-Adapter, but the affected pipelines and symptom are different enough that they are not the same bug." }, { "left": "issue:10542", "right": "issue:11104", "accept": false, - "reason": "Different models and failures: Hunyuan Video batch-size regression vs LTX-Video offload methods broken." + "reason": "Different products and failure modes: Hunyuan Video batch-size regression vs LTX-Video offload-method breakage." }, { "left": "issue:10994", "right": "issue:9227", "accept": false, - "reason": "Both are FLUX training problems, but one is a batch-size crash in a LoRA script and the other is a general problem running train_dreambooth_lora_flux.py." + "reason": "Both are FLUX training reports, but one is a batch-size crash and the other is a broader model/script problem." }, { "left": "issue:11050", "right": "issue:2780", "accept": false, - "reason": "Different dtype/AMP issues in different scripts: SD3 ControlNet bf16/fp16 casting vs FP16 gradient unscale in text-to-image LoRA." + "reason": "Different dtype-related bugs in different scripts: SD3 controlnet embed casting vs FP16 gradient unscale error." }, { "left": "issue:3537", "right": "issue:3612", "accept": false, - "reason": "One is a conversion script request; the other is about creating a ControlNet without a model and choosing a preprocessor. Not duplicates." + "reason": "One is a conversion script request, the other is a question about creating ControlNet without a model; not the same issue." }, { "left": "issue:7979", "right": "issue:8614", "accept": false, - "reason": "A malformed config JSON error is not the same as missing config.json for a controlnet single-file model." + "reason": "Both mention config.json, but one is invalid JSON while the other is a missing config file for single-file ControlNet." }, { "left": "issue:10722", "right": "issue:10969", "accept": false, - "reason": "Different runtime errors in different training setups: tensor-size mismatch in DreamBooth prior preservation vs FLUX-Control zero3 weight dimensionality." + "reason": "Different training failures: tensor size mismatch in DreamBooth prior preservation vs a 2-D weight error in FLUX-controlnet zero3." }, { "left": "issue:2361", "right": "issue:3694", "accept": false, - "reason": "Saving checkpoints failure vs missing model_index.json/config.json during finetuning setup. Distinct failure stages." + "reason": "Checkpoint saving failure and missing model_index/config.json are separate example-script problems." }, { "left": "issue:12104", "right": "issue:12771", "accept": false, - "reason": "Generic IndexError vs a specific DPMSolverMultistepScheduler out-of-bounds issue with squaredcos_cap_v2 and Karras sigmas." + "reason": "Both are out-of-bounds errors, but they arise in different contexts and are not clearly the same scheduler bug." }, { "left": "issue:2156", "right": "issue:2203", "accept": false, - "reason": "One is training on CPU instead of GPU; the other is model download path location. Completely different problems." + "reason": "Training uses CPU instead of GPU is unrelated to model download location." }, { "left": "issue:8503", "right": "issue:8828", "accept": false, - "reason": "Both are SD3 training issues, but one is gradient checkpointing-related while the other is a general 16GB training error." + "reason": "Both are SD3 training issues, but one is specifically gradient checkpointing and the other is a general 16GB training error." }, { "left": "issue:12533", "right": "issue:12661", "accept": false, - "reason": "Context-parallelism/CPU-offload hook conflict vs a ChronoEdit crash with context parallelism. Not enough evidence of the same bug." + "reason": "Context parallelism appears in both, but one is a hook conflict and the other is a ChronoEdit crash; not the same concrete bug." }, { "left": "issue:10334", "right": "issue:11315", "accept": false, - "reason": "Both mention MPS, but Sana image corruption/NaNs and HiDream float64 rope are different models and likely different root causes." + "reason": "Both hit MPS, but they are different models and failures: Sana NaNs/gray images vs HiDream rope float64 issue." }, { "left": "issue:9047", "right": "issue:9971", "accept": false, - "reason": "Flux failing on MPS devices is a device/backend issue; DEISMultistepScheduler not working on FLUX is a scheduler issue." + "reason": "Flux on MPS device failure is unrelated to DEISMultistepScheduler behavior on FLUX." } ] }, @@ -24705,10 +24723,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -24723,169 +24741,172 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a noisy mix of unrelated loading, conversion, scheduler, and training-example issues; none of the proposed soft pairs look like safe duplicates.", - "confidence": 0.79, - "canonical_issue_reason": "issue:8359 is the broadest, most central loading/conversion report in the set (from_single_file/custom checkpoints) and is a better anchor than narrower, symptom-specific issues.", + "summary": "The cluster is mostly lexical similarity, not true duplication: it mixes resume-from-checkpoint bugs, model-loading issues, conversion failures, scheduler problems, and adapter/ControlNet reports. Only one pair looks like the same concrete change request/bug in the same script.", + "confidence": 0.88, + "canonical_issue_reason": "issue:3346 is the best anchor because it explicitly asks for resume_from_checkpoint support in `examples/dreambooth/train_dreambooth_lora.py`, and issue:3279 reports the same checkpoint-resume failure in that same script.", "canonical_pr_reason": null, - "best_issue_reason": "issue:8359 best represents the cluster's dominant theme of model-loading incompatibilities and has the clearest, most reusable problem statement.", + "best_issue_reason": "issue:3346 is the clearest representative of the only real duplicate-like pair in the set; the rest of the items are unrelated bugs across different features.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:5016", "right": "issue:5757", "accept": false, - "reason": "Same script family, but one is about SDXL text-encoder training and the other is resume-from-checkpoint errors; not the same concrete bug." + "reason": "Both mention DreamBooth LoRA resume behavior, but one is about SDXL text-encoder training and the other is a checkpoint resume error in a different script/version." }, { "left": "issue:5687", "right": "issue:6295", "accept": false, - "reason": "Both involve schedulers, but the failure modes differ (duplicate timesteps vs artifacts under DPM++ 2M SDE Karras/use_lu_lambdas)." + "reason": "Both are scheduler/artifact complaints, but they describe different schedulers and different failure modes." }, { "left": "issue:2091", "right": "issue:3453", "accept": false, - "reason": "Resume-from-checkpoint multi-GPU pipeline issue vs bf16 CUDA kernel limitation; unrelated code paths." + "reason": "One is a multi-GPU checkpoint resume issue; the other is a BF16 CUDA kernel unsupported error. Different bugs." }, { "left": "issue:12107", "right": "issue:9142", "accept": false, - "reason": "Both mention list/AttributeError, but one is accelerate tracker logging and the other is multi-ControlNet model wiring; different bugs." + "reason": "Both involve lists/custom objects, but one is `accelerator.init_trackers` and the other is multiple ControlNet loading in a different pipeline." }, { "left": "issue:4790", "right": "issue:6226", "accept": false, - "reason": "Both are converter-related, but they are different conversion directions and different downstream failures; not one bug." + "reason": "Same converter script, but the reports point to different downstream failures; not enough evidence they are the same underlying bug." }, { "left": "issue:7286", "right": "issue:8885", "accept": false, - "reason": "Custom Diffusion cross-attention loading and clip-retrieval/modifier-token installation are unrelated problems." + "reason": "Both mention loading-related problems, but they are for different features: Custom Diffusion cross-attention weights vs clip-retrieval/modifier tokens." }, { "left": "issue:6732", "right": "issue:7053", "accept": false, - "reason": "Documentation for trained_betas is not the same as UNet3D class-conditioning behavior." + "reason": "Documentation for `trained_betas` is unrelated to `UNet3DConditionModel` class conditioning." }, { "left": "issue:6226", "right": "issue:9421", "accept": false, - "reason": "A conversion-script loading error is not the same as inaccessible model download/authorization for runwayml/stable-diffusion-inpainting." + "reason": "Converter output compatibility is unrelated to simply failing to access a model on the hub." }, { "left": "issue:3279", "right": "issue:3346", - "accept": false, - "reason": "Both are about checkpoint resume in the same script, but one is a runtime error report and the other is a feature/support request; not enough evidence for one duplicate." + "accept": true, + "reason": "Same training example and same resume-from-checkpoint code path; one reports the error, the other requests the missing support." }, { "left": "issue:2299", "right": "issue:460", "accept": false, - "reason": "ONNX runtime dependency issue vs fine-tuned checkpoint support; different subsystems and symptoms." + "reason": "Checkpoint conversion support and ONNX runtime dependency issues are different problems." }, { "left": "issue:4436", "right": "issue:4453", "accept": false, - "reason": "These are opposite conversion directions and different model formats (CivitAI safetensors with VAE vs SDXL LoRA to safetensors)." + "reason": "Both are conversion topics, but one is safetensors-with-VAE to Diffusers and the other is Diffusers LoRA to A1111 safetensors." }, { "left": "issue:7440", "right": "issue:7879", "accept": false, - "reason": "Missing dataset availability and a separate dataset request are related, but not the same underlying bug/change." + "reason": "One says the pokemon dataset is unavailable; the other asks for a pokemon-blip-captions dataset. Related topic, not the same bug." }, { "left": "issue:9209", "right": "issue:9804", "accept": false, - "reason": "HF mirror connectivity failure vs local ControlNet loading failure; one is networking, the other is model loading." + "reason": "A mirror/network access failure is unrelated to a local `ControlNetModel.from_pretrained` loading bug." }, { "left": "issue:6069", "right": "issue:6117", "accept": false, - "reason": "Both are add_noise scheduler bugs, but they affect different schedulers and different edge cases." + "reason": "Different schedulers and different `add_noise` failures." }, { "left": "issue:4286", "right": "issue:5226", "accept": false, - "reason": "Both mention load_lora_weights, but one is SDXL LoRA loading and the other is the SDXL Adapter pipeline; different pipeline support issues." + "reason": "Both involve `load_lora_weights`, but they are different pipelines and not clearly the same concrete defect." }, { "left": "issue:11922", "right": "issue:9309", "accept": false, - "reason": "Different Flux pipelines/models and different failures; too broad to call duplicates." + "reason": "Different Flux model/pipeline problems; same family name is not enough to merge them." }, { "left": "issue:2161", "right": "issue:4566", "accept": false, - "reason": "Inpainting LoRA training and SDXL resume-from-checkpoint errors are different problems." + "reason": "LoRA training on inpainting models is a different issue from SDXL checkpoint resume errors." }, { "left": "issue:5317", "right": "issue:5341", "accept": false, - "reason": "PEFT API docs and updating example scripts are adjacent maintenance tasks, not one underlying bug." + "reason": "Documentation of PEFT API vs updating example scripts are related but distinct tasks." }, { "left": "issue:6008", "right": "issue:8469", "accept": false, - "reason": "Both are face/IP-Adapter related, but they target different models and symptoms (LCM LoRA interaction vs corrupt faces)." + "reason": "Both are IP-Adapter face issues, but they involve different models and symptoms." }, { "left": "issue:9142", "right": "issue:9643", "accept": false, - "reason": "Both involve multiple ControlNets, but on different model families (Hunyuan DiT vs Flux) and likely different implementations." + "reason": "Both mention multiple ControlNets, but one is a Hunyuan DiT `list` error and the other is a Flux feature/support question." }, { "left": "issue:5886", "right": "issue:8827", "accept": false, - "reason": "A feature request for face model support is not the same as a specific ip-adapter-faceid_sdxl runtime error." + "reason": "Different IP-Adapter face variants and different failure reports." }, { "left": "issue:11432", "right": "issue:7982", "accept": false, - "reason": "torch_dtype='auto' in from_pretrained and from_pipe with MultiControlNetModel are separate API paths and failures." + "reason": "`torch_dtype=\"auto\"` in `from_pretrained` is unrelated to `from_pipe` with `MultiControlNetModel`." }, { "left": "issue:8359", "right": "issue:8900", "accept": false, - "reason": "Same API family, but one is a bug with custom checkpoints and the other is a how-to question for a specific refiner file." + "reason": "Both use `from_single_file`, but one is about custom checkpoints broadly and the other is a specific SDXL refiner safetensors load path." }, { "left": "issue:4974", "right": "issue:6584", "accept": false, - "reason": "General safetensors-to-diffusers conversion question vs a specific DreamShaper v8 inpaint conversion failure." + "reason": "General safetensors-to-Diffusers conversion and a specific DreamShaper inpaint conversion failure are not clearly the same bug." }, { "left": "issue:5997", "right": "issue:7138", "accept": false, - "reason": "Tutorial model-type mismatch and UNet structure mismatch are both SDXL-related, but they are different mismatches at different layers." + "reason": "A tutorial/model-type mismatch is a different problem from SDXL UNet structure mismatch." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: it correctly identifies issue:3279 and issue:3346 as the only clearly duplicate-like pair, and the remaining pairwise judgments stay conservative rather than overclaiming duplication. No soft-edge verdict appears to overstate the evidence." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -24980,10 +25001,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -24998,162 +25019,165 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The candidate edges are all superficial keyword matches; the issues span unrelated bugs, features, and models, so none look like the same underlying problem. No canonical duplicate group emerges from this set.", - "confidence": 0.96, - "canonical_issue_reason": "null", - "canonical_pr_reason": "null", - "best_issue_reason": "null", - "best_pr_reason": "null", + "summary": "This cluster does not contain any conservatively grounded duplicate pair. Most pairs are only superficially related by subsystem or shared keywords, and the 5412/5846 resume_from_checkpoint pair is not strong enough to accept because the packet shows different training scripts with no body-level evidence of the same underlying bug.", + "confidence": 0.91, + "canonical_issue_reason": "No issue in this packet is strong enough to serve as a safe duplicate canonical for the cluster as a whole. If a best-in-group issue must be named, 5412 is the closest anchor for the resume_from_checkpoint topic, but not enough to justify merging 5846 into it.", + "canonical_pr_reason": null, + "best_issue_reason": "5412 is the strongest issue anchor only in the narrow sense that it names the script and failure mode most directly; however, it should not be treated as a confirmed duplicate canonical because 5846 points to a different script and the packet lacks shared-bug evidence.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10026", "right": "issue:5197", "accept": false, - "reason": "Both mention multi-GPU training, but one is Flux ControlNet + DeepSpeed Stage-3 memory usage and the other is SD1.5 training slowdown; different code paths and symptoms." + "reason": "Both are multi-GPU training reports, but the symptoms differ (memory scaling vs slower throughput) and they do not indicate the same bug." }, { "left": "issue:10382", "right": "issue:9331", "accept": false, - "reason": "VAE encoder forward return-type/AttributeError versus FlowMatchEulerDiscreteScheduler index_for_timestep logic; unrelated components." + "reason": "VAE encoder return-type mismatch and FlowMatchEuler index errors are different components and failure modes." }, { "left": "issue:2830", "right": "issue:938", "accept": false, - "reason": "Text/image-to-video question versus a text-to-motion model request; same media domain but different task and pipeline." + "reason": "These are different tasks altogether: image-to-video versus text-to-motion." }, { "left": "issue:4580", "right": "issue:4631", "accept": false, - "reason": "Both are scheduler exceptions with use_karras_sigmas, but they involve different schedulers and different failure modes." + "reason": "Shared use_karras_sigmas alone is not enough; the schedulers and exceptions differ." }, { "left": "issue:7168", "right": "issue:7238", "accept": false, - "reason": "IP-Adapter embed shape troubleshooting versus IP-Adapter masking feature request/bug; not the same issue." + "reason": "One is about embedding tensor shape, the other about masking support; different issues." }, { "left": "issue:2988", "right": "issue:7041", "accept": false, - "reason": "Different runtime failures in schedulers: num_training_steps == num_inference_steps versus segfault when num_inference_steps > 1." + "reason": "Different crash conditions around step counts; not the same defect." }, { "left": "issue:12656", "right": "issue:12771", "accept": false, - "reason": "AuraFlow positional embedding out-of-bounds versus DPMSolverMultistepScheduler out-of-bounds under squaredcos+karras; different model and scheduler bugs." + "reason": "Both are out-of-bounds reports, but in different subsystems with different root causes." }, { "left": "issue:3728", "right": "issue:6407", "accept": false, - "reason": "A new pipeline proposal versus AnyText model announcement/request; no shared underlying bug or change." + "reason": "These are separate feature requests, not duplicate bugs or changes." }, { "left": "issue:1534", "right": "issue:434", "accept": false, - "reason": "Dreambooth notebook error versus notebook incompatibility with version 0.3.0; too generic to be the same defect." + "reason": "Both are training notebook complaints, but the reports are too generic and lack a shared concrete failure." }, { "left": "issue:10180", "right": "issue:12396", "accept": false, - "reason": "Multiple LoRA loading bug versus leaf_level/delete_adapters errors; different adapter management scenarios." + "reason": "Both mention adapters/LoRA, but the actual behaviors differ: loading multiple LoRAs versus delete_adapters/leaf_level errors." }, { "left": "issue:4925", "right": "issue:5997", "accept": false, - "reason": "Distributed ControlNet SDXL OOM versus ControlNet SDXL tutorial type mismatch; unrelated problems." + "reason": "Distributed OOM in training and a tutorial type-mismatch import error are unrelated." }, { "left": "issue:5012", "right": "issue:5432", "accept": false, - "reason": "Text encoder training missing in LoRA script versus blank validation images; different bugs in the training script." + "reason": "Same script family, but one is about training the text encoder and the other is blank validation images." }, { "left": "issue:9237", "right": "issue:9501", "accept": false, - "reason": "Flux Dreambooth LoRA training issue versus delayed model saving during Flux training; different symptoms and likely causes." + "reason": "Both are Flux Dreambooth reports, but the actual problems differ and are not clearly the same code-path bug." }, { "left": "issue:9907", "right": "issue:9936", "accept": false, - "reason": "Maximum sequence length warning versus NCCL timeout in multi-GPU training; not the same code-path problem." + "reason": "A sequence-length warning and an NCCL timeout are distinct symptoms with different likely causes." }, { "left": "issue:10772", "right": "issue:9619", "accept": false, - "reason": "Sana ControlNet support request versus ControlNet pipeline interrupt support; different features and scopes." + "reason": "A Sana ControlNet support request and a ControlNet interrupt feature request are unrelated." }, { "left": "issue:10650", "right": "issue:10866", "accept": false, - "reason": "Lumina shape/reshape runtime error versus Lumina LoRA compatibility with Civitai LoRAs; unrelated failures." + "reason": "One is a Lumina shape/runtime error, the other a LoRA compatibility issue." }, { "left": "issue:10823", "right": "issue:12547", "accept": false, - "reason": "Missing inference files from fine-tuning script versus required --instance_prompt argument for Flux Kontext Dreambooth; different scripts and fixes." + "reason": "Missing inference outputs and a required --instance_prompt argument are unrelated training-script issues." }, { "left": "issue:10937", "right": "issue:7420", "accept": false, - "reason": "Both involve torch.compile, but one is specifically vae.encode and the other is a broad inability to use torch.compile; too broad to merge as one bug." + "reason": "One is a specific vae.encode compile error, the other a broad torch.compile complaint; not enough to treat as one bug." }, { "left": "issue:1304", "right": "issue:3813", "accept": false, - "reason": "General how-to question for text-guided image-to-image versus multi-GPU model-parallel finetuning question; not a duplicate." + "reason": "These are broad how-to questions about different workflows, not duplicate bugs." }, { "left": "issue:4724", "right": "issue:7041", "accept": false, - "reason": "SDXL Img2Img checkpoint behavior versus segfault with num_inference_steps > 1; different pipelines and failure modes." + "reason": "Img2Img checkpoint behavior and an inference-step segfault are unrelated." }, { "left": "issue:5026", "right": "issue:5152", "accept": false, - "reason": "KeyError on text_encoder in 0.21.0 versus W\u00fcrstchen bug in 0.21.3 alpha; different models and issues." + "reason": "A text_encoder KeyError and a Wurstchen alpha bug are different issues." }, { "left": "issue:5412", "right": "issue:5846", "accept": false, - "reason": "Both are resume_from_checkpoint failures, but they occur in different training scripts and there is no evidence of the same concrete bug." + "reason": "Both mention resume_from_checkpoint, but they are different scripts and the packet provides no body evidence that they share the same underlying bug." }, { "left": "issue:11169", "right": "issue:11560", "accept": false, - "reason": "Binary incompatibility numpy import traceback versus a vague import error; insufficient evidence they are the same underlying issue." + "reason": "A specific NumPy binary incompatibility traceback and a generic import error are not enough to establish one underlying bug." }, { "left": "issue:1556", "right": "issue:2107", "accept": false, - "reason": "Black images with CUDA/NSFW warning versus low-resolution glitchy/black images; different triggers and likely causes." + "reason": "Both mention black images, but the reported conditions and likely causes differ." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "The only accepted soft-edge pair is not sufficiently grounded in the packet. 5412 and 5846 do both mention `resume_from_checkpoint`, but they are different training scripts (`train_dreambooth_lora_sdxl.py` vs `train_text_to_image_lora_sdxl.py`), and the packet provides no body evidence showing they share the same underlying bug. This is too strong for a conservative duplicate verdict, so the cluster summary overstates the evidence." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -25251,10 +25275,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -25269,178 +25293,175 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Nearly all soft pairs are distinct issues with different pipelines, APIs, or error modes. The only clear duplicate-style pair is the FLUX IPAdapter feature request cluster.", - "confidence": 0.77, - "canonical_issue_reason": "issue:9825 is the broader, more generic FLUX IPAdapter request; it best represents the shared feature ask behind the FLUX-related pair.", + "summary": "This cluster is highly heterogeneous: most items are unrelated bug reports, feature requests, or training/examples issues. The soft pairs only share broad terminology and should not be deduplicated; none look like the same underlying bug or change.", + "confidence": 0.88, + "canonical_issue_reason": "If one issue must be chosen, issue 11651 is the most concrete standalone bug report: it has a specific FLUXControl LoRA load failure, a clear affected model/pipeline, and the most actionable problem statement. The rest of the cluster is too mixed to form a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9825 is the best global issue candidate because it is the clearest umbrella request for the same FLUX IPAdapter capability.", + "best_issue_reason": "Issue 11651 is the best overall issue candidate because it is a focused, reproducible runtime failure rather than a broad feature request or a vague compatibility report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3124", "right": "issue:5753", "accept": false, - "reason": "Different problems: model conversion failure vs. ControlNet LoRA-loading compatibility." + "reason": "Both are model-loading/runtime problems, but one is a Stable Diffusion conversion failure and the other is an SDXL ControlNet LoRA-loading issue; different code paths and symptoms." }, { "left": "issue:2596", "right": "issue:5603", "accept": false, - "reason": "Different missing-import/doc issues involving unrelated symbols." + "reason": "These concern different documentation/import problems: missing `load_image` in docs versus an `is_safetensors_available` import error." }, { "left": "issue:7377", "right": "issue:9112", "accept": false, - "reason": "Different pipelines and different runtime errors; only a generic inference-failure resemblance." + "reason": "Different pipelines and failures: StableCascade decoder `num_images_per_prompt` handling versus Auraflow tensor shape mismatch." }, { "left": "issue:3250", "right": "issue:861", "accept": false, - "reason": "One is K-Diffusion pipeline runtime, the other is checkpoint conversion." + "reason": "One is a KDiffusion pipeline runtime issue; the other is a checkpoint-to-diffusers conversion problem." }, { "left": "issue:3799", "right": "issue:4806", "accept": false, - "reason": "Both are type-related, but they concern different classes and APIs." + "reason": "These are unrelated typing bugs in different objects: scheduler `from_pretrained` inference versus `UNet2DModel` parameter typing." }, { "left": "issue:4356", "right": "issue:4936", "accept": false, - "reason": "Unrelated module import error vs. DreamBooth resume bug." + "reason": "Different breakages: missing dummy module import on install versus DreamBooth LoRA resume logic." }, { "left": "issue:7223", "right": "issue:9744", "accept": false, - "reason": "Both are video-model addition requests, but for different models and not the same change." + "reason": "Both are model-addition requests, but for different video models and different pipeline surfaces; not the same change." }, { "left": "issue:2132", "right": "issue:2156", "accept": false, - "reason": "Same example family, but the failures are different: trackers attribute error vs. CPU/GPU training path." + "reason": "Both are DreamBooth training issues, but one is an `Accelerator.trackers` attribute error and the other is training falling back to CPU; different failures." }, { "left": "issue:2218", "right": "issue:761", "accept": false, - "reason": "Different DreamBooth training issues with no clear shared code-path." + "reason": "Different training problems: overtraining behavior versus failure with 8-bit Adam optimizer." }, { "left": "issue:10804", "right": "issue:11651", "accept": false, - "reason": "Both mention FLUX LoRA loading, but they target different models/pipelines and error contexts." + "reason": "Both mention FLUX LoRA loading, but one is a ComfyUI-compatible FLUX1.dev LoRA load failure and the other is a FluxControl flux-fill-lora load failure; no evidence of the same root cause." }, { "left": "issue:1197", "right": "issue:4080", "accept": false, - "reason": "Completely different features: OpenAI diffusion pipeline vs. SDXL inpainting." + "reason": "A community OpenAI diffusion pipeline request is unrelated to an SDXL inpainting pipeline request." }, { "left": "issue:2132", "right": "issue:3454", "accept": false, - "reason": "Same script, but one specific accelerator attribute error and one broad latest-version failure are not clearly the same bug." + "reason": "Both are DreamBooth LoRA training failures, but the reported symptoms and regression points differ; no concrete shared bug path is shown." }, { "left": "issue:3250", "right": "issue:746", "accept": false, - "reason": "Different conversion/loading problems for different pipeline types." + "reason": "Different loading/execution paths: StableDiffusionKDiffusionPipeline runtime versus ONNX revision loading." }, { "left": "issue:1919", "right": "issue:4390", "accept": false, - "reason": "Both are missing-dependency complaints, but not the same missing package or example." + "reason": "Both are packaging/dependency complaints, but one is missing datasets for a DreamBooth example and the other is a generic missing requirements report." }, { "left": "issue:3954", "right": "issue:6278", "accept": false, - "reason": "Both involve multi-GPU training, but the concrete issues differ." + "reason": "Both touch multi-GPU training, but one is about LR scheduler step counts and the other is about an example that cannot run on multi-GPU." }, { "left": "issue:2107", "right": "issue:557", "accept": false, - "reason": "Different symptoms and unrelated sampling/image-quality complaints." + "reason": "One reports low-resolution black/glitchy images, while the other compares DDIM results quality; not the same defect." }, { "left": "issue:6224", "right": "issue:8732", "accept": false, - "reason": "TorchDynamo fine-tuning issue vs. checkpoint-saving issue." + "reason": "TorchDynamo fine-tuning failure and checkpoint-saving failure are different training-stage problems." }, { "left": "issue:11627", "right": "issue:9448", "accept": false, - "reason": "Both IP-Adapter related, but different pipelines and different failure modes." + "reason": "Different IP-Adapter issues in different pipelines: SD3.5 InstantX support versus a tuple `.shape` error in ControlNet inpainting." }, { "left": "issue:9403", "right": "issue:9825", - "accept": true, - "reason": "Both request FLUX IPAdapter support; 9825 is the broader umbrella request and matches the same underlying feature." + "accept": false, + "reason": "Both are FLUX/IP-Adapter feature requests, but they target different adapter support and are not the same concrete change." }, { "left": "issue:9576", "right": "issue:9841", "accept": false, - "reason": "Different bugs: safetensors propagation vs. DreamBooth use_dora TypeError." + "reason": "A loading bug with `use_safetensors` propagation and a DreamBooth SDXL LoRA `use_dora=False` type error are unrelated." }, { "left": "issue:2132", "right": "issue:3284", "accept": false, - "reason": "Same example script, but the reported failures are not specific enough to confirm the same root cause." + "reason": "Both mention Accelerate in DreamBooth training, but the specific failures differ: missing `trackers` attribute versus a broader run failure." }, { "left": "issue:11432", "right": "issue:9187", "accept": false, - "reason": "Different loading issues in different model families." + "reason": "These are separate loading issues: `torch_dtype=\"auto\"` on `from_pretrained` versus SDXL IP-Adapter behavior after Kolors update." }, { "left": "issue:3651", "right": "issue:3954", "accept": false, - "reason": "Both multi-GPU-related, but one is inability to run and the other is an LR-step-count bug." + "reason": "Related only by multi-GPU context; one is inability to run DreamBooth multi-GPU, the other is incorrect scheduler step counting." }, { "left": "issue:5799", "right": "issue:6159", "accept": false, - "reason": "Both are unsupported pipeline kwargs, but for different kwargs and pipelines." + "reason": "Both are unsupported keyword-argument issues in inpainting-related pipelines, but they concern different kwargs, pipelines, and code paths." }, { "left": "issue:10883", "right": "issue:12752", "accept": false, - "reason": "Completely different model/update requests." + "reason": "A Marigold model/docs update and a Z-Image img2img/inpainting feature request are unrelated." }, { "left": "issue:5798", "right": "issue:5799", "accept": false, - "reason": "Both are keyword-argument issues, but they affect different pipelines and different parameters." + "reason": "Both are kwarg compatibility bugs, but `callback_on_step_end` in ControlNet and `strength` in inpainting are distinct API gaps." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet, and the soft-edge verdicts are conservative. The only accepted pair, issue:9403 and issue:9825, is supported by the titles as the same FLUX IPAdapter feature request, with 9825 plausibly serving as the broader umbrella issue." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -25535,10 +25556,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -25553,162 +25574,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous set of issue reports with a few plausible duplicates around from_single_file loading, multi-LoRA loading performance, and SDXL checkpoint resume. The rest only share broad subsystem or symptom-level similarity and should stay separate.", - "confidence": 0.72, - "canonical_issue_reason": "Issue 8435 is the broadest from_single_file runtime-error report and the best umbrella among the close matches, though the overall cluster is not a single coherent bug.", + "summary": "Mostly a heterogeneous set of issue reports across different pipelines/features. The only strong duplicate-like pairs are SDXL checkpoint-resume failures and multi-LoRA loading performance complaints; the rest share broad subsystem wording but appear to be different bugs or models.", + "confidence": 0.69, + "canonical_issue_reason": "issue:8659 is the best canonical issue: it is open, detailed, and covers the same SDXL checkpoint-resume/load-failure path as issue:4566, with broader symptom coverage.", "canonical_pr_reason": null, - "best_issue_reason": "8435 is the strongest representative of the repeated from_single_file/load-failure theme and is the most useful issue to anchor the accepted duplicates.", + "best_issue_reason": "issue:8659 is the strongest global issue candidate because it is the clearest representative of one of the true duplicate pairs and is still open for consolidation.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:5395", "right": "issue:5518", "accept": false, - "reason": "Both involve LoRA/PEFT errors, but one is about combining LoRAs and the other is a missing set_lora_layer attribute; different concrete bugs." + "reason": "Both involve LoRA/PEFT, but one is about combining civitai loras and the other about a missing set_lora_layer attribute; different failures." }, { "left": "issue:2616", "right": "issue:5919", "accept": false, - "reason": "Both are training-time attribute errors, but they fail at different code paths and with different missing objects/methods." + "reason": "Different bugs in LoRA saving: one is a NoneType replace error in save_attn_procs, the other is a missing set_lora_layer method." }, { "left": "issue:6131", "right": "issue:6149", "accept": false, - "reason": "Both concern IP-Adapter variants, but the symptoms and targets differ; not enough to show the same underlying bug." + "reason": "Both are IP-Adapter related, but the titles point to different adapter variants and different failure modes." }, { "left": "issue:10514", "right": "issue:10962", "accept": false, - "reason": "Different pipelines and failure modes: Sana resolution binning/sample size versus CogView4 prompt-embed shape issues." + "reason": "Different model-specific shape problems: Sana resolution binning/sample size vs CogView4 prompt-embed shape handling." }, { "left": "issue:9297", "right": "issue:9673", "accept": true, - "reason": "Both report multi-LoRA loading performance problems in the same workflow; linear load-time growth and high CPU use are likely the same inefficiency." + "reason": "Same underlying multi-LoRA loading performance problem; one reports linear load-time growth and the other high CPU usage during the same operation." }, { "left": "issue:11010", "right": "issue:9861", "accept": false, - "reason": "Support for a new Flux architecture and training not updating the transformer are unrelated bugs." + "reason": "Both mention Flux, but one is architecture support and the other is a training-update symptom; not the same bug." }, { "left": "issue:2089", "right": "issue:684", "accept": false, - "reason": "Both mention memory issues, but one is DreamBooth OOM and the other is AMD memory-optimization support; not the same defect." + "reason": "Both mention memory, but one is DreamBooth CUDA OOM on AltDiffusion and the other is AMD memory-optimization support; unrelated root causes." }, { "left": "issue:6460", "right": "issue:9264", "accept": false, - "reason": "One is a loading hang for an existing inpainting model; the other is a request for a Flux inpainting model, not the same bug." + "reason": "One is an inpainting model load hang, the other is a request for a Flux inpainting model; not duplicates." }, { "left": "issue:8543", "right": "issue:8708", "accept": false, - "reason": "Both are DreamBooth example issues, but one is gradient checkpointing and the other is a missing VAE config field; different failures." + "reason": "Different SD3 training issues: gradient checkpointing failure vs missing vae.config.shift_factor in examples." }, { "left": "issue:4402", "right": "issue:6839", "accept": false, - "reason": "Both mention from_single_file, but they affect different pipeline classes and the titles do not establish the same concrete code-path bug." + "reason": "Both are from_single_file issues, but for different pipelines and failure modes; too broad to merge." }, { "left": "issue:455", "right": "issue:457", "accept": false, - "reason": "MPS index.Tensor not implemented and MPS blue artifacts are distinct symptoms, likely different root causes." + "reason": "Both are MPS-related, but one is an aten::index.Tensor NotImplementedError and the other is blue-artifact rendering; different bugs." }, { "left": "issue:11777", "right": "issue:9258", "accept": false, - "reason": "These are unrelated: a HiDream LoRA training question versus Flux LoRA training performance on different OSes." + "reason": "Training question vs performance/debugging question; not the same issue." }, { "left": "issue:2844", "right": "issue:4307", "accept": false, - "reason": "Explicit image-embedding input to StableUnCLIPImg2Img is unrelated to a local_files_only from_pretrained error." + "reason": "Explicit image-embedding runtime error vs local_files_only from_pretrained error; unrelated." }, { "left": "issue:2707", "right": "issue:6226", "accept": false, - "reason": "A syntax error in the converter script is not the same as downstream loading errors in webui." + "reason": "Same script file, but one is a syntax error and the other is a loading regression; not the same concrete bug." }, { "left": "issue:10542", "right": "issue:12738", "accept": false, - "reason": "Both fail for batch size > 1, but they are different pipelines with different assertions and no clear shared bug." + "reason": "Both are batch-size >1 failures, but they affect different pipelines with different assertions." }, { "left": "issue:10302", "right": "issue:12396", "accept": false, - "reason": "FP8 inference noise and leaf_level/delete_adapters errors are unrelated." + "reason": "FP8 inference noise without CPU offloading is unrelated to leaf_level/delete_adapters errors." }, { "left": "issue:4566", "right": "issue:8659", "accept": true, - "reason": "Both are SDXL training/checkpoint-resume failures in the same training script family; 8659 explicitly includes the same resume/load path." + "reason": "Same SDXL checkpoint-resume/load-failure path; 8659 appears to subsume the resume issue and is the better merged target." }, { "left": "issue:4849", "right": "issue:579", "accept": false, - "reason": "Multi-GPU textual inversion training and missing config.json are unrelated." + "reason": "Textual inversion multi-GPU usage vs missing config.json; unrelated." }, { "left": "issue:11363", "right": "issue:13079", "accept": false, - "reason": "Both are CUDA OOM reports, but on different models and scenarios; not a single fixable bug." + "reason": "Both are CUDA OOM reports, but on different Flux inference scenarios; not the same bug." }, { "left": "issue:3481", "right": "issue:7142", "accept": false, - "reason": "These are about differing inference outputs, but the causes and contexts are too different to be duplicates." + "reason": "Different determinism/output-difference complaints in different pipelines; not the same code-path issue." }, { "left": "issue:5128", "right": "issue:8435", - "accept": true, - "reason": "Both are from_single_file loading/runtime failures with local weights; 8435 is the broader generic report and 5128 is a specific manifestation." + "accept": false, + "reason": "Both are from_single_file errors, but the specific pipelines and symptoms differ." }, { "left": "issue:10428", "right": "issue:9096", "accept": false, - "reason": "Different Flux inference problems on different hardware/precision settings; not the same code-path bug." + "reason": "Both mention Flux inference errors, but one is Ascend NPU-specific and the other is torch.half NaNs; different root causes." }, { "left": "issue:11436", "right": "issue:9200", "accept": false, - "reason": "Both are dtype mismatches, but they affect different models/pipelines and are not shown to share the same root cause." + "reason": "Both are dtype mismatch errors, but they occur in different models/pipelines with different tensor types." }, { "left": "issue:2089", "right": "issue:2412", "accept": false, - "reason": "AltDiffusion DreamBooth OOM and a generic DreamBooth error are too broad and not the same underlying failure." + "reason": "General DreamBooth training problems, but one is AltDiffusion OOM and the other is a broader error using dreambooth; not a duplicate." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative about most pairs. The accepted duplicate edges are plausible from the titles alone: 5128\u21948435 share the same from_single_file/local-safetensors failure theme, 4566\u21948659 clearly overlap on SDXL checkpoint resume/load issues, and 9297\u21949673 both describe multi-LoRA loading inefficiency. The remaining verdicts are appropriately cautious and do not overstate sameness." + "feedback": "The summary stays mostly grounded in the packet and the false-pair verdicts are conservative. The two accepted soft pairs are plausible from the titles, and the canonical/open-issue choice is reasonable. No clear overstatement that would force rejection." }, "evaluator_used": true, "retried": false, @@ -25808,10 +25829,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -25826,156 +25847,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated issues about Flux/LoRA loading, training checkpoints, conversion scripts, import errors, feature requests, and memory/VRAM behavior. None of the soft pairs look like true duplicates; most share only broad subsystem words.", - "confidence": 0.93, - "canonical_issue_reason": "No single canonical issue cleanly represents the cluster. The only somewhat central thread is Flux/LoRA loading (e.g. 10512), but the rest of the items are too diverse to treat as one duplicate family.", + "summary": "These items do not form a true duplicate cluster; they span distinct bugs, feature requests, and converter/import/training failures. The soft similarities are mostly topical (LoRA/Flux/training/imports) rather than the same underlying defect, so all soft edges should be rejected.", + "confidence": 0.44, + "canonical_issue_reason": "Issue 9358 is the clearest, most concrete open bug in the set and best represents the Flux/LoRA text-encoder/memory-management subset, even though the overall collection is not a single duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "10512 is the closest concrete bug in the most coherent subtheme (Flux/LoRA loading) and has substantial discussion, but it still does not cover the other unrelated issues.", + "best_issue_reason": "Issue 9358 is the most actionable and well-scoped issue among mostly unrelated entries.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1565", "right": "issue:3094", "accept": false, - "reason": "Both mention GPU OOM, but they are different pipelines and different failure contexts; not the same underlying bug." + "reason": "Both involve OOM, but one is GPU selection/multi-GPU usage and the other is training fine-tuning memory usage; different bugs." }, { "left": "issue:2695", "right": "issue:2991", "accept": false, - "reason": "Both are training/validation failures, but one is a TensorBoard validation issue and the other is mixed-precision behavior in ControlNet; different bugs." + "reason": "Both are training-example validation problems, but they affect different examples and failure modes." }, { "left": "issue:10392", "right": "issue:10512", "accept": false, - "reason": "Both involve Flux/LoRA, but one is a specific Hyper-FLUX.1-dev-8steps-lora scenario and the other is a Quanto Flux LoRA load failure; too different to be duplicates." + "reason": "Both mention Flux/LoRA, but one is a specific FP8/Hyper-FLUX setup and the other is a Quanto loading failure; not the same code path." }, { "left": "issue:10672", "right": "issue:11710", "accept": false, - "reason": "A callback API request and a MagCache feature request are unrelated changes." + "reason": "One is a callback API feature request and the other is a video-cache feature request; unrelated changes." }, { "left": "issue:10963", "right": "issue:11199", "accept": false, - "reason": "Both are import errors, but for different symbols/modules and different root causes." + "reason": "Both are import errors, but for different modules and likely different root causes." }, { "left": "issue:2091", "right": "issue:7197", "accept": false, - "reason": "Both involve resuming training, but one is a multi-GPU checkpoint issue and the other is a KeyError on a specific transform/shape path." + "reason": "Both concern resuming training, but the reported errors and contexts differ enough to be separate bugs." }, { "left": "issue:3191", "right": "issue:4398", "accept": false, - "reason": "Both concern Dreambooth/LoRA outputs, but one is a general breakage report and the other is a black-image bug in SDXL LoRA Dreambooth; not the same concrete issue." + "reason": "Both involve Dreambooth/LoRA, but the symptoms and pipelines are different." }, { "left": "issue:1919", "right": "issue:817", "accept": false, - "reason": "Missing datasets dependency and mixed-precision malfunction are unrelated bugs." + "reason": "Missing dependency vs mixed-precision malfunction; unrelated fixes." }, { "left": "issue:5226", "right": "issue:9109", "accept": false, - "reason": "Load_lora_weights support for a pipeline is unrelated to a torchvision NMS import/operator failure." + "reason": "Different subsystems entirely: LoRA loading in a pipeline versus a torchvision operator import issue." }, { "left": "issue:10421", "right": "issue:11352", "accept": false, - "reason": "A CPU memory leak while moving pipelines across GPUs is not the same as a missing warning when moving 8-bit models to CPU." + "reason": "Memory leak moving pipelines across GPUs is not the same as a missing warning when moving an 8-bit model to CPU." }, { "left": "issue:2904", "right": "issue:3250", "accept": false, - "reason": "Different pipeline failures with different error modes; only broad pipeline-runtime similarity." + "reason": "Different pipelines and different runtime errors." }, { "left": "issue:6790", "right": "issue:8493", "accept": false, - "reason": "Both are model state_dict loading problems, but they affect different models and likely different incompatibilities." + "reason": "Both are state_dict loading issues, but they involve different models and mismatch causes." }, { "left": "issue:10447", "right": "issue:10757", "accept": false, - "reason": "A file-splitting regression in a conversion script and a variable-renaming issue are not the same bug." + "reason": "Same script family, but one is a file-splitting bug and the other is a variable-name cleanup; separate changes." }, { "left": "issue:4790", "right": "issue:4837", "accept": false, - "reason": "Both concern conversion scripts, but one is a round-trip file incompatibility and the other is an SDXL conversion failure; distinct issues." + "reason": "Both touch conversion scripts, but one is round-trip incompatibility and the other is an SDXL conversion failure." }, { "left": "issue:4733", "right": "issue:7310", "accept": false, - "reason": "Both are training-script problems, but one is checkpoint resume and the other is default-settings behavior in a different script." + "reason": "Checkpoint-resume optimizer error vs default training-script behavior; not the same bug." }, { "left": "issue:6176", "right": "issue:9876", "accept": false, - "reason": "Loading LoRA after training and VRAM not being released after training are related themes but different concrete bugs." + "reason": "LoRA loading failure after training versus VRAM not being released after training are different issues." }, { "left": "issue:6524", "right": "issue:6561", "accept": false, - "reason": "A second LoRA load failing is unrelated to an IP-Adapter missing config file error." + "reason": "A second LoRA load breaking behavior is unrelated to a missing IP-Adapter config file." }, { "left": "issue:11225", "right": "issue:11241", "accept": false, - "reason": "Validation OOM and a floating point exception in Flux ControlNet training are different failure modes." + "reason": "Different training failures: validation OOM versus floating-point exception." }, { "left": "issue:11374", "right": "issue:13067", "accept": false, - "reason": "Two feature requests for different models; same broad type, not same change." + "reason": "Separate feature requests for different model families." }, { "left": "issue:11561", "right": "issue:8790", "accept": false, - "reason": "Both mention IP Adapter, but for different pipelines and different requested support paths." + "reason": "Both mention IP Adapter, but one is FluxFill support and the other is SDXL TensorRT integration; different work." }, { "left": "issue:1185", "right": "issue:1944", "accept": false, - "reason": "Both are Dreambooth quality/reproducibility complaints, but they describe different symptoms and likely different causes." + "reason": "Reproducibility issue versus poor generation quality; not the same defect." }, { "left": "issue:5412", "right": "issue:6894", "accept": false, - "reason": "Checkpoint resume failure in one script is different from a LoRA compatibility break in A1111." + "reason": "Both reference train_dreambooth_lora_sdxl, but one is resume_checkpoint failure and the other is A1111 LoRA compatibility." }, { "left": "issue:5892", "right": "issue:6794", "accept": false, - "reason": "Both are import errors, but the missing imports are different and there is no evidence of the same root cause." + "reason": "Generic import error versus a specific failed import from training_utils; insufficiently aligned." }, { "left": "issue:8511", "right": "issue:9358", "accept": false, - "reason": "Both touch text-encoder handling in SD3/Flux LoRA training, but they describe different code paths and symptoms; not safe to merge as duplicates." + "reason": "Both concern text-encoder handling in Flux/SD3 training, but they describe different scripts and different bugs." } ] }, @@ -26077,10 +26098,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -26095,166 +26116,163 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous issues with only one plausible duplicate pair around LoRA/safetensors file selection on Windows. The rest are different models, scripts, or symptom classes despite some broad subsystem overlap.", - "confidence": 0.91, - "canonical_issue_reason": "issue:4665 is the clearest original bug report and includes the concrete Windows/alternate-bin failure scenario; issue:5719 reads more like the code-path fix description.", + "summary": "This cluster is not a true duplicate set: it mixes unrelated issues across Flux inference, Wan OOM/loading, Hunyuan support, SDXL training, LoRA loading, and assorted pipeline bugs. The soft pairs are mostly lexical overlaps rather than the same underlying defect.", + "confidence": 0.93, + "canonical_issue_reason": "No single issue cleanly represents the whole cluster; if a representative is required, issue 12459 is the most concrete and actionable bug report, but it does not cover the rest.", "canonical_pr_reason": null, - "best_issue_reason": "issue:4665 best represents the only likely duplicate cluster in this set and is the most actionable user-facing report.", + "best_issue_reason": "issue 12459 is the strongest standalone candidate because it is a specific, reproducible performance bug with clear scope and current relevance, even though it is not a duplicate of the other items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4368", "right": "issue:8950", "accept": false, - "reason": "Both mention LoRA/OOM-ish training problems, but they are different models and different failure modes (SDXL training vs SD3 LoRA training)." + "reason": "Both mention LoRA/SDXL OOM, but one is a training-memory problem and the other is an SD3 LoRA training bug; different model families and failure modes." }, { "left": "issue:12011", "right": "issue:13079", "accept": false, - "reason": "Both are OOM reports, but one is Wan 2.2 video inference and the other is FLUX.2-klein looping over images; not the same bug." + "reason": "Both are OOM reports, but Wan 2.2 i2v and FLUX.2-klein are different models and the repro conditions differ." }, { "left": "issue:6561", "right": "issue:7238", "accept": false, - "reason": "Different IP-Adapter issues: missing config file vs masking behavior." + "reason": "One is an IP-Adapter checkpoint/file-path loading error; the other is an IP-Adapter masking feature request/bug. Same subsystem, not the same bug." }, { "left": "issue:378", "right": "issue:4849", "accept": false, - "reason": "Textual inversion is the only overlap; one is a distribution error, the other is a multi-GPU usage question." + "reason": "Textual inversion DiagonalGaussianDistribution is a specific runtime issue; the other is a general question about multi-GPU training." }, { "left": "issue:2806", "right": "issue:771", "accept": false, - "reason": "Unrelated topics: pipeline customization printing vs BF16 training failure." + "reason": "Printing customized pipelines and BF16 dreambooth are unrelated problems." }, { "left": "issue:4975", "right": "issue:5606", "accept": false, - "reason": "Both concern LoRA loading, but one is performance inefficiency and the other is a missing-alpha correctness bug." + "reason": "Both are LoRA loading/performance related, but one is about repeated datatype queries and the other about missing alpha metadata." }, { "left": "issue:10470", "right": "issue:10674", "accept": false, - "reason": "Both are Flux inference complaints, but torchao inference and GGUF loading are different backends/code paths." + "reason": "Both are Flux inference failures, but torchao inference and GGUF loading are different backends/code paths." }, { "left": "issue:5575", "right": "issue:8472", "accept": false, - "reason": "A LoRA layer-size configuration question is not the same as pre-training LoRA with other modules simultaneously." + "reason": "One asks about LoRA hidden size configuration; the other asks about simultaneous pre-training with other modules. Not the same defect." }, { "left": "issue:10989", "right": "issue:12147", "accept": false, - "reason": "Both are load failures, but they target different models and different quantization/loading contexts." + "reason": "Both mention LoRA loading failing, but they concern different model families and different causes (quantized Transformers vs Wan2.2 Lightning loading)." }, { "left": "issue:1181", "right": "issue:4241", "accept": false, - "reason": "Training from scratch is unrelated to a question about fine-tuning unidiffusers with LoRA." + "reason": "Training from scratch vs fine-tuning LoRA are different usage questions, not duplicates." }, { "left": "issue:3706", "right": "issue:5474", "accept": false, - "reason": "Different feature requests for different model families; no shared bug or change." + "reason": "Wuerstchen new-model support and Stable Diffusion Reference Only proposal are separate model/pipeline feature requests." }, { "left": "issue:6879", "right": "issue:7239", "accept": false, - "reason": "Different training failures: dtype mismatch in Custom Diffusion vs shape mismatch in SDXL Dreambooth LoRA." + "reason": "Custom Diffusion dtype mismatch and Dreambooth SDXL shape mismatch are different tensor-shape/dtype bugs." }, { "left": "issue:9025", "right": "issue:9032", "accept": false, - "reason": "Same pipeline family and TPU runtime, but one is from_single_file on Kaggle and the other is a minimal example request on Colab; not the same bug." + "reason": "Both are FlaxStableDiffusionPipeline on TPU, but one is from_single_file usage in Kaggle and the other is a minimal example in Colab; not the same bug." }, { "left": "issue:4665", "right": "issue:5719", - "accept": true, - "reason": "These describe the same file-selection bug around LoRA safetensors/bin handling, with 5719 phrasing the fix for the failure reported in 4665." + "accept": false, + "reason": "Both involve LoRA/safetensors file selection on Windows, but one is a Windows alt-bin-file loading failure and the other is a specific preference order for LoRA weights." }, { "left": "issue:4985", "right": "issue:6520", "accept": false, - "reason": "Same general SDXL training area, but one is OOM during InstructPix2Pix training and the other is blank validation output in adapter training." + "reason": "Both are training OOM-ish reports, but one is InstructPix2Pix SDXL training and the other is log_validation producing blank results in adapter training." }, { "left": "issue:3363", "right": "issue:4733", "accept": false, - "reason": "Same script name, but the failures are different: machine-specific failure vs checkpoint resume/optimizer state error." + "reason": "Dreambooth LoRA failing on two machines and resume-from-checkpoint optimizer inf-check errors are different training bugs." }, { "left": "issue:4551", "right": "issue:8939", "accept": false, - "reason": "Different models and symptoms: SDXL Dreambooth crash vs SD3 tokenizer loading error." + "reason": "SDXL dreambooth crash and SD3 tokenizer loading failure are unrelated model/setup issues." }, { "left": "issue:12321", "right": "issue:12412", "accept": false, - "reason": "These are support requests for different HunyuanImage versions, not the same issue." + "reason": "These are support requests for different HunyuanImage model versions, not the same issue." }, { "left": "issue:2439", "right": "issue:3631", "accept": false, - "reason": "Both are inpainting pipeline issues, but one is generic load/run trouble and the other is specifically passing latents." + "reason": "Inpainting pipeline loading/running problems and providing latents to inpainting are related but distinct pipeline bugs." }, { "left": "issue:5529", "right": "issue:5633", "accept": false, - "reason": "Different code areas and different bugs; only a loose training-script similarity." + "reason": "train_text_to_image_sdxl.py bug and a transformer_temporal.py bug are different code paths." }, { "left": "issue:5529", "right": "issue:6230", "accept": false, - "reason": "Same script, but generic bug report vs CUDA OOM; not enough to treat as the same underlying issue." + "reason": "Both mention train_text_to_image_sdxl.py, but one is a generic bug report and the other is specifically CUDA OOM; not enough to treat as the same defect." }, { "left": "issue:1742", "right": "issue:2560", "accept": false, - "reason": "Different reproducibility problems: prompt list inference speed vs list-of-generators output differences." + "reason": "Different generators/benchmarking speed behavior versus img2img output differences with a list of generators." }, { "left": "issue:3984", "right": "issue:4724", "accept": false, - "reason": "Unrelated errors: class label requirement vs SDXL checkpoint behavior." + "reason": "Class-label validation error and SDXL img2img checkpoint behavior are unrelated." }, { "left": "issue:12459", "right": "issue:12565", "accept": false, - "reason": "Both are throughput/running complaints in different model families (Flux vs Wan), so they are not the same bug." + "reason": "Flux.1-Kontext-dev throughput issue and Wan2.1 i2v running issue are separate model/runtime problems." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is well-grounded in the packet: it identifies the one clearly matching duplicate pair (4665/5719) and keeps the rest conservative. The reasoning does not overstate the weaker overlaps, and the accepted pair looks mergeable as a single bugfix PR." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -26350,10 +26368,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -26368,162 +26386,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose collection of diffusion-model loading, dtype, and training issues rather than a single duplicate set. Only a few pairs look like true same-change/same-bug matches (notably Uni-ControlNet, LoRA text-encoder training, and blank/black validation images).", - "confidence": 0.79, - "canonical_issue_reason": "Issue #6130 is the broadest representative of the recurring model/LoRA loading-failure theme, but the overall cluster is still heterogeneous.", + "summary": "This set is mostly heterogeneous: loading, dtype, pipeline support, training, and validation-image bugs. I only see a few true duplicates; most soft pairs are related by subsystem only, not the same underlying issue.", + "confidence": 0.91, + "canonical_issue_reason": "Issue:3564 is the clearest canonical among the true duplicates: it is the broader Uni-ControlNet feature request, and 3569 is the same request in a new-pipeline framing.", "canonical_pr_reason": null, - "best_issue_reason": "#6130 is the best single issue to anchor the cluster because it is the most generally representative of the loading-related failures present here.", + "best_issue_reason": "Issue:3564 is the best issue to anchor the duplicate subset, though the overall collection is too mixed to have one cluster-wide canonical bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4480", "right": "issue:6136", "accept": false, - "reason": "Both are Mac/device inference failures, but one is an MPS float16 kernel crash and the other is SD2.1 NaNs on CPU; different bugs." + "reason": "Both are Mac/cpu generation failures, but one is an MPS fp16 kernel crash and the other is SD2.1 NaNs on CPU." }, { "left": "issue:9443", "right": "issue:9613", "accept": false, - "reason": "Different codepaths and symptoms: SD3 slow-tokenizer instantiation vs a Timesteps.scale attribute error." + "reason": "Different failure points: tokenizer instantiation for SD3 vs missing Timesteps.scale in 0.30.x." }, { "left": "issue:2392", "right": "issue:6506", "accept": false, - "reason": "Prompt_embeds handling in StableDiffusionImg2ImgPipeline is unrelated to StableVideoDiffusion output_type=\"np\" returning a list." + "reason": "One is an img2img prompt_embeds validation error; the other is StableVideoDiffusion output_type formatting." }, { "left": "issue:11436", "right": "issue:5510", "accept": false, - "reason": "Both are dtype mismatch errors, but they occur in different pipelines with different tensor/bias combinations." + "reason": "Both are dtype mismatches, but they occur in different models and code paths with different tensors involved." }, { "left": "issue:5028", "right": "issue:5798", "accept": false, - "reason": "Different unexpected kwargs in different custom pipeline hooks; not the same underlying controlnet bug." + "reason": "Different unexpected-argument regressions in different pipelines: scale vs callback_on_step_end." }, { "left": "issue:11033", "right": "issue:7682", "accept": false, - "reason": "SD1.5 from_single_file loading failure is unrelated to multi-GPU training on smaller GPUs." + "reason": "One is from_single_file SD1.5 loading; the other is a multi-GPU training/runtime problem." }, { "left": "issue:3661", "right": "issue:4213", "accept": false, - "reason": "Both involve model loading/conversion, but one is SD2.x from_ckpt conversion and the other is SDXL from_single_file without accelerate." + "reason": "Both involve model conversion/loading, but one is SD2.x from_ckpt and the other is SDXL from_single_file without accelerate." }, { "left": "issue:10010", "right": "issue:11448", "accept": false, - "reason": "Both are 'not found' model-loading failures, but for different models and repositories; no shared code bug is evident." + "reason": "Both are missing-model errors, but for different pipelines and different absent artifacts/URLs." }, { "left": "issue:10954", "right": "issue:11648", "accept": false, - "reason": "Both mention FLUX/LoRA, but one is a specific load failure and the other is a general how-to question about fp8 transformer LoRA weights." + "reason": "Similar LoRA wording, but one is a loading failure and the other is a general fp8 transformer usage question." }, { "left": "issue:8503", "right": "issue:8649", "accept": false, - "reason": "Both are SD3 issues, but one is gradient-checkpointing crash and the other is a num_images_per_prompt regression." + "reason": "Same SD3 area, but one crashes with gradient checkpointing while the other mishandles num_images_per_prompt." }, { "left": "issue:7146", "right": "issue:7311", "accept": false, - "reason": "SDXL training OOM and checkpoint-saving failure are different problems, not one duplicate bug." + "reason": "Different training bugs: CUDA OOM during SDXL finetuning versus checkpoint save failure." }, { "left": "issue:1604", "right": "issue:3984", "accept": false, - "reason": "Diffusers version breakage in lpw_stable_diffusion.py is unrelated to the num_class_embeds/class_labels validation error." + "reason": "A diffusers version regression in one case, versus a class_labels requirement tied to num_class_embeds in the other." }, { "left": "issue:4183", "right": "issue:4835", "accept": false, - "reason": "Different ControlNet bugs: guidance_scale <= 1 error vs MultiControlNetModel behavior in SDXL img2img." + "reason": "Both ControlNet-related, but one is guidance_scale <= 1 and the other is MultiControlNetModel in SDXL img2img." }, { "left": "issue:3564", "right": "issue:3569", "accept": true, - "reason": "Both are Uni-ControlNet feature requests for the same new pipeline/support addition." + "reason": "Same underlying request: Uni-ControlNet support/new pipeline for the same model family." }, { "left": "issue:7223", "right": "issue:9873", "accept": false, - "reason": "Separate feature requests for different new pipelines (Latte vs OmniGen)." + "reason": "Both are new model pipeline requests, but for entirely different models: Latte video generation vs OmniGen image generation." }, { "left": "issue:3177", "right": "issue:3296", "accept": false, - "reason": "Both are DreamBooth/DeepSpeed training issues, but they affect different failure modes: stage-3 offloading vs text_encoder checkpoint save/load." + "reason": "Same script, but different bugs: DeepSpeed stage-3 offloading versus checkpoint save/load with text_encoder." }, { "left": "issue:4303", "right": "issue:4488", "accept": false, - "reason": "Both are SDXL loading-related, but they target different loading APIs and different failure points." + "reason": "Both SDXL loading issues, but one is a Flax pipeline crash and the other is a conversion utility needing pipeline_class." }, { "left": "issue:9051", "right": "issue:9092", "accept": false, - "reason": "Batch inference support and missing model_index.json are unrelated FLUX issues." + "reason": "One is batch inference support in FluxPipeline; the other is a model_index.json missing-file error." }, { "left": "issue:1005", "right": "issue:861", "accept": false, - "reason": "Timestep-count validation and ckpt-to-diffusers conversion are different problems." + "reason": "Completely different topics: timestep bounds versus checkpoint-to-diffusers conversion." }, { "left": "issue:2210", "right": "issue:3418", "accept": true, - "reason": "Both ask for LoRA text-encoder training support; same feature change." + "reason": "Same feature request: let LoRA training include the text encoder in the training script." }, { "left": "issue:6243", "right": "issue:8790", "accept": false, - "reason": "IP-Adapter Face ID and SDXL TensorRT + IP Adapter are different feature requests, not the same bug/change." + "reason": "Both mention IP-Adapter, but one is Face ID support and the other is SDXL TensorRT integration." }, { "left": "issue:2548", "right": "issue:6130", "accept": false, - "reason": "One is a LoRA training save error and the other is a LoRA loading error; related area, but different failures." + "reason": "Training-time LoRA save failure versus inference-time LoRA loading error; same area, different bugs." }, { "left": "issue:5432", "right": "issue:6666", "accept": true, - "reason": "Both describe blank/black validation images during training and appear to be the same regression." + "reason": "Same validation-image bug: the generated validation images are blank/black during training." }, { "left": "issue:2068", "right": "issue:4250", "accept": false, - "reason": "Generic dtype conversion behavior in from_pretrained is different from incorrect model loading in from_pretrained." + "reason": "Both are model-loading complaints, but 2068 is a dtype-conversion issue while 4250 is an unspecified incorrect-load bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is conservative and grounded in the packet. The accepted soft edges are plausible same-change matches (Uni-ControlNet, LoRA text-encoder training, blank/black validation images), and the rejected pairs are described cautiously without overclaiming shared bugs. The canonical issue choice is also reasonable given the loading/LoRA theme, while still noting heterogeneity." + "feedback": "Grounded overall and appropriately conservative. The summary reflects the packet\u2019s mix, and the accepted true-duplicate pairs look supportable from the issue titles/descriptions: 3564/3569, 2210/3418, and 5432/6666. The rejected soft pairs are generally scoped narrowly enough to avoid overstating duplicates." }, "evaluator_used": true, "retried": false, @@ -26622,10 +26640,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -26640,165 +26658,165 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The clearest duplicate is 10139/10514 around sample_size handling in pipeline init. A couple of other soft pairs look duplicate at the title level (SDXL ControlNet dtype mismatch, SDXL UNet shape mismatch), but most remaining pairs only share a broad subsystem or generic error wording and are not safe to merge.", - "confidence": 0.79, - "canonical_issue_reason": "issue:10139 is the broadest and most reusable representative of the sample_size / pipeline-init bug family, and it is the strongest canonical anchor because 10514 is the clearest matching report.", + "summary": "This cluster is mostly a loose set of issue titles that only share broad themes like pipeline initialization, batch-size handling, dtype mismatches, loading, and OOMs. The only strong soft match is the sample_size-related pair.", + "confidence": 0.66, + "canonical_issue_reason": "issue:10139 is the most root-cause-like report: it directly says sample_size is not fully considered in pipeline init, which is a broader upstream bug description than the more specific downstream symptom in 10514.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10139 is the best overall issue representative because it captures the core pipeline-init sample_size defect in the most general way and has the strongest direct duplicate link in the set.", + "best_issue_reason": "issue:10139 is the best global issue candidate because it is the broadest and most general description of the underlying sample_size/init problem, making it the most reusable canonical reference.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4399", "right": "issue:4418", "accept": false, - "reason": "Different problems: torch.compile with CPU offload versus SDXL inpaint denoising_start behavior." + "reason": "Different failures: torch.compile/cpu offload compatibility vs SDXL inpaint denoising_start behavior." }, { "left": "issue:9200", "right": "issue:9527", - "accept": true, - "reason": "Both titles describe the same ControlNet/main-pipeline dtype mismatch under mixed precision." + "accept": false, + "reason": "Both mention dtype errors, but the titles point to different pipeline/version contexts and do not clearly describe the same concrete bug." }, { "left": "issue:1969", "right": "issue:222", "accept": false, - "reason": "Both are generic runtime errors, but the titles do not show the same underlying bug." + "reason": "Too generic; one is a specific tensor view/stride error, the other is a vague training RuntimeError." }, { "left": "issue:6967", "right": "issue:7138", - "accept": true, - "reason": "Both point to the same SDXL UNet/shape-mismatch family." + "accept": false, + "reason": "Related to SDXL shape/model-structure issues, but one is a training script mismatch and the other is a model structure discrepancy; not enough evidence of one underlying bug." }, { "left": "issue:4724", "right": "issue:7421", "accept": false, - "reason": "Different issues: SDXL img2img checkpoint behavior versus KDiffusion seed handling." + "reason": "Different problems: SDXL img2img checkpoint behavior vs seed handling in a KDiffusion pipeline." }, { "left": "issue:10150", "right": "issue:10223", "accept": false, - "reason": "Slow SDXL loading/inference is unrelated to the LoRA identifier question." + "reason": "Unrelated: slow loading/inference vs a question about obtaining a LoRA identifier." }, { "left": "issue:3933", "right": "issue:4305", "accept": false, - "reason": "Both mention torch.compile graph breaks, but the titles do not establish the same concrete defect or fix." + "reason": "Both involve graph breaks, but one is tied to logging.info and the other to a version-level torch.compile regression; not the same concrete fix." }, { "left": "issue:11223", "right": "issue:12011", "accept": false, - "reason": "Both are Wan OOM reports, but the titles point to different scenarios and not one shared bug." + "reason": "Both are WAN OOM reports, but one is specifically layer-wise upcasting with LoRAs and the other is a general i2v OOM; likely different causes." }, { "left": "issue:4724", "right": "issue:6368", "accept": false, - "reason": "Different failures: SDXL img2img behavior versus unsupported LoRA alpha loading." + "reason": "Different issues: SDXL img2img expected behavior vs unsupported LoRA alpha checkpoint loading." }, { "left": "issue:5617", "right": "issue:6686", "accept": false, - "reason": "Both involve compiled pipelines, but the titles describe different symptoms and code-paths." + "reason": "Both concern compiled models and save/load, but one is a loading failure and the other is lack of post-save performance gains." }, { "left": "issue:4954", "right": "issue:579", "accept": false, - "reason": "Textual inversion unpickling error is not the same as missing config.json." + "reason": "Both are Textual Inversion load errors, but one is an unpickling error and the other is a missing config.json; different failure modes." }, { "left": "issue:3235", "right": "issue:3661", "accept": false, - "reason": "Both are loading-related, but custom diffusion bias loading and from_ckpt SD2 conversion are different problems." + "reason": "Both are checkpoint-loading problems, but they involve different model families and distinct incompatibilities." }, { "left": "issue:8609", "right": "issue:8817", "accept": false, - "reason": "Both mention IP-Adapter, but the titles show different usage contexts and failure modes." + "reason": "Different IP-Adapter/custom pipeline issues without clear evidence of the same code-path bug." }, { "left": "issue:12009", "right": "issue:12147", "accept": false, - "reason": "A from_single_file GGUF load failure is different from a Lightning LoRA loading failure." + "reason": "Both are Wan-related loading issues, but one is GGUF single-file loading and the other is Lightning LoRA loading; not the same bug." }, { "left": "issue:2163", "right": "issue:3885", "accept": false, - "reason": "Training dtype mismatch and VRAM reduction are unrelated." + "reason": "Unrelated: dtype mismatch in training vs a general VRAM-usage question." }, { "left": "issue:10453", "right": "issue:12738", "accept": false, - "reason": "Both fail at batch size > 1, but the titles indicate different models/pipelines and not one shared bug." + "reason": "Both fail above batch size 1, but they concern different pipelines/models and do not clearly share the same underlying implementation bug." }, { "left": "issue:5544", "right": "issue:8649", "accept": false, - "reason": "FreeU disable handling and SD3 num_images_per_prompt handling are unrelated." + "reason": "Different features and failures: FreeU disable error vs num_images_per_prompt handling in SD3." }, { "left": "issue:10139", "right": "issue:10514", "accept": true, - "reason": "Both are about sample_size being mishandled during pipeline init, and 10514 is the more specific 4k/resolution-binning case." + "reason": "Both center on sample_size being mishandled in pipeline setup, with 10514 reading like a concrete manifestation of the broader init bug described in 10139." }, { "left": "issue:11657", "right": "issue:12037", "accept": false, - "reason": "Both involve Wan LoRAs, but the titles describe different compatibility/load issues." + "reason": "Both are Wan LoRA-related, but one is a missing-prefix warning and the other is a compatibility issue with a distilled LoRA type." }, { "left": "issue:5128", "right": "issue:6183", "accept": false, - "reason": "Both use from_single_file, but the titles do not show one concrete shared defect." + "reason": "Both are single-file loading issues, but one is a local safetensors loading problem and the other is an incorrect-shape problem." }, { "left": "issue:3264", "right": "issue:672", "accept": false, - "reason": "These are opposite-direction conversion requests, not the same bug." + "reason": "Both involve conversion/loading of model weights, but they address different directions and different conversion paths." }, { "left": "issue:3706", "right": "issue:3826", "accept": false, - "reason": "Completely different model/pipeline requests." + "reason": "Not duplicates: one is a model introduction request, the other is a new pipeline request." }, { "left": "issue:3591", "right": "issue:4125", "accept": false, - "reason": "Both mention loss curves, but the training setups and symptoms are different." + "reason": "Unrelated training/diagnostics questions about loss curves; no same bug or change." }, { "left": "issue:12108", "right": "issue:9576", "accept": false, - "reason": "Flow-matching scheduler break is unrelated to use_safetensors propagation." + "reason": "Different problems: flow-matching scheduler breakage vs a pipeline loading parameter propagation bug." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The canonical issue choice around 10139/10514 is well grounded, but several soft-edge explanations go beyond what is supported by the packet titles alone. In particular, the true-accept rationale for 3933/4305 asserts a specific logging.info cause that is not evident from 4305\u2019s title, and a few other pair explanations rely on details not directly visible in the provided data. The overall summary is directionally correct, but not conservative enough for a strict accept." + "accept": true, + "feedback": "The analyst\u2019s summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:10139 with issue:10514, is plausibly the strongest match based on the shared sample_size/init theme, and the other pairwise rejections avoid overstating weak title-level similarity." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -26894,10 +26912,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -26912,156 +26930,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set; it mixes unrelated bugs, feature requests, model-loading issues, and training/pipeline questions across many different components. No soft edge looks safe to merge.", - "confidence": 0.96, + "summary": "This cluster is very loose and mostly contains unrelated issues that only share superficial vocabulary around diffusion pipelines, training, or image generation. None of the soft-edge pairs look like the same underlying bug/change, so I would not merge any of them as duplicates.", + "confidence": 0.24, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No single issue is a strong canonical candidate: the set spans distinct bugs and feature requests across SD3, Qwen Image, ControlNet, loaders, training scripts, and unrelated pipeline behavior.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:11135", "right": "issue:3728", "accept": false, - "reason": "Both are new-pipeline proposals, but they target different methods/models (SmoothCache vs Shifted Diffusion) and are not the same change." + "reason": "Both are new-pipeline ideas, but they describe different models and proposals (SmoothCache vs Shifted Diffusion), not the same change." }, { "left": "issue:1161", "right": "issue:3698", "accept": false, - "reason": "Different conversion tasks: Flax-to-PyTorch vs converting an inpainting model to diffusers." + "reason": "Both are conversion-related, but one is Flax-to-PyTorch and the other is inpainting-model conversion; different workflows and targets." }, { "left": "issue:1524", "right": "issue:1604", "accept": false, - "reason": "One is training quality on SD2.0; the other is a version-specific LPW script breakage." + "reason": "One is poor training quality, the other is a version-specific script breakage; not the same bug." }, { "left": "issue:5226", "right": "issue:5753", "accept": false, - "reason": "Both mention load_lora_weights, but they involve different pipelines and failure modes; not the same bug." + "reason": "Both mention LoRA with SDXL variants, but one is AdapterPipeline-specific and the other is ControlNet-specific; different code paths." }, { "left": "issue:6866", "right": "issue:6935", "accept": false, - "reason": "Related to IP-Adapter, but one is ControlNet Inpaint support and the other is Face ID; distinct feature requests." + "reason": "Both are IP-Adapter requests, but one is ControlNet inpaint support and the other is Face ID; distinct functionality." }, { "left": "issue:8588", "right": "issue:8708", "accept": false, - "reason": "Different SD3 setup problems: missing conversion script vs missing shift_factor in dreambooth examples." + "reason": "Both involve SD3 example/script gaps, but one is a missing conversion script and the other is a missing VAE config field in dreambooth examples." }, { "left": "issue:8939", "right": "issue:9044", "accept": false, - "reason": "Both are model-loading errors, but for different models and different missing artifacts." + "reason": "Both are model-loading OSErrors, but they reference different models and missing files, so they are not the same issue." }, { "left": "issue:5005", "right": "issue:8296", "accept": false, - "reason": "Both concern resume behavior, but one is progress-bar truncation and the other is inconsistent global-step handling." + "reason": "Both concern training resume/global step behavior, but one is a progress-bar truncation and the other is inconsistent step accounting across scripts." }, { "left": "issue:2774", "right": "issue:5128", "accept": false, - "reason": "Both involve safetensors/loading, but one is ControlNet conversion and the other is single-file XLPipeline loading." + "reason": "Both are file-conversion/load errors, but they affect different converters and model formats." }, { "left": "issue:1616", "right": "issue:3761", "accept": false, - "reason": "LPW pipeline version support and loading multiple community pipelines are unrelated issues." + "reason": "Both are community pipeline loading issues, but one is LPW pipeline version compatibility and the other is loading multiple pipelines in one process." }, { "left": "issue:6570", "right": "issue:8826", "accept": false, - "reason": "Different runtime failures with different symptoms and code paths." + "reason": "Both are runtime tensor-shape errors, but they occur in different pipelines and have different failure modes." }, { "left": "issue:6141", "right": "issue:6354", "accept": false, - "reason": "Both are ControlNet feature requests, but they ask for different capabilities (parameters vs LoRA support)." + "reason": "One is a request for ControlNet parameters, the other is LoRA support for ControlNet; related area but not the same change." }, { "left": "issue:6141", "right": "issue:9619", "accept": false, - "reason": "Different ControlNet enhancements: extra parameters vs interrupt support." + "reason": "Different ControlNet feature requests: parameter support versus interrupt support." }, { "left": "issue:1572", "right": "issue:1866", "accept": false, - "reason": "Both touch generation behavior, but one is img2img/inpainting pipeline failure and the other is a scheduler strength-edge-case bug." + "reason": "Both involve Stable Diffusion img2img/inpainting, but one is a broken pipeline and the other is scheduler-strength incompatibility." }, { "left": "issue:4186", "right": "issue:7982", "accept": false, - "reason": "Different pipeline problems: SDXL inpaint/refiner vs from_pipe with MultiControlNetModel." + "reason": "Both touch SDXL/ControlNet-style pipelines, but one is an inpaint/refiner runtime issue and the other is a from_pipe/MultiControlNetModel construction issue." }, { "left": "issue:6823", "right": "issue:7350", "accept": false, - "reason": "Both involve ControlNet setup, but one is general inability to use ControlNet and the other is initialization from a LoRA UNet." + "reason": "Both mention ControlNet, but one is a general inability to use it and the other is initializing ControlNet from a LoRA-modified UNet." }, { "left": "issue:10040", "right": "issue:10475", "accept": false, - "reason": "Both complain about poor image quality, but one is grey outputs in SD3 while the other is a broader fine-tuning vs inference quality mismatch." + "reason": "Both report low-quality/gray outputs, but one is Stable Diffusion 3 gray images and the other is a validation-vs-inference quality mismatch during fine-tuning." }, { "left": "issue:12108", "right": "issue:12175", "accept": false, - "reason": "Same model family, but one is a scheduler/flow-matching breakage and the other is an image-quality complaint." + "reason": "Both are Qwen Image-related, but one is a scheduler/flow-matching breakage and the other is generic bad image quality." }, { "left": "issue:10311", "right": "issue:7021", "accept": false, - "reason": "Different import errors with different causes (train_dreambooth script vs HF_HOME environment handling)." + "reason": "Both are import/load errors, but they refer to different environment variables and different failure points." }, { "left": "issue:1704", "right": "issue:3539", "accept": false, - "reason": "Unrelated questions about different concepts." + "reason": "One is a question about an attention block, the other about main_process_first; not duplicates." }, { "left": "issue:1823", "right": "issue:5337", "accept": false, - "reason": "Stable Diffusion comment swap bug vs a general callback explanation question." + "reason": "Different topics entirely: swapped comments in depth2img versus a general question about callback behavior." }, { "left": "issue:11006", "right": "issue:11779", "accept": false, - "reason": "Same Wan ecosystem, but different pipelines and distinct bugs: video output with quantized transformer vs mask padding in VACE." + "reason": "Both involve Wan pipelines, but one is broken video output with quantized transformer and the other is incorrect mask padding in VACE; distinct bugs." }, { "left": "issue:2161", "right": "issue:2185", "accept": false, - "reason": "Different training issues: inpainting LoRA training vs missing config files for naive text-to-image training." + "reason": "One is LoRA training on inpainting, the other missing config files for naive text-to-image training; different issues." }, { "left": "issue:3788", "right": "issue:8414", "accept": false, - "reason": "Two separate new model/pipeline proposals with no shared bug or change." + "reason": "Both are new model/pipeline requests, but they target different models and modalities (visual generation vs text-to-audio)." } ] }, @@ -27163,10 +27181,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -27181,156 +27199,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is very heterogeneous: most items are unrelated model-loading, training, pipeline-feature, or model-specific runtime issues. The soft edges mostly share only broad keywords like \"load\", \"error\", or \"black images\" and do not look like the same underlying bug/change. No PRs are present.", - "confidence": 0.22, - "canonical_issue_reason": "Issue 4250 is the broadest loading-related report in the set, so it is the closest thing to an anchor, but the cluster is too mixed to justify a true duplicate canonical.", + "summary": "No true duplicate cluster emerges here: the soft-linked pairs are mostly superficial title matches across different models, pipelines, training scripts, or error types.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 4250 is the most general and reusable issue title among the list, but it still only partially fits because the rest of the items span many different features and failure modes.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3910", "right": "issue:4061", "accept": false, - "reason": "Both mention LoRA loading, but one is a general LoRA-load problem and the other is an SDXL-specific load error; not the same concrete bug." + "reason": "Both mention LoRA loading, but one is generic and the other is SDXL-specific; likely different model/setup failures." }, { "left": "issue:9112", "right": "issue:9362", "accept": false, - "reason": "Different runtime failures in different contexts: tensor shape mismatch vs index out of bounds." + "reason": "Different error types and model paths: tensor-size mismatch in Auraflow vs an index-out-of-bounds issue." }, { "left": "issue:9187", "right": "issue:9971", "accept": false, - "reason": "Both are FLUX/SDXL-adjacent inference problems, but one is IP-Adapter image encoder handling and the other is a scheduler issue." + "reason": "Unrelated subsystems: IP-Adapter/SDXL image encoder handling vs FLUX scheduler behavior." }, { "left": "issue:4808", "right": "issue:6548", "accept": false, - "reason": "Both involve image pipelines, but one is a SHAP-E integration test and the other is an RGBA img2img latent-prep bug." + "reason": "Both involve img2img-like code, but one is a SHAP-E integration test and the other is an RGBA latent-preprocessing bug." }, { "left": "issue:8787", "right": "issue:9667", "accept": false, - "reason": "One is generic slow loading; the other is a specific FLUX fp8 local-file loading error." + "reason": "One is a performance regression in loading speed; the other is a concrete FLUX fp8 loading error." }, { "left": "issue:3454", "right": "issue:4324", "accept": false, - "reason": "Training-script failure vs UNet forward TypeError; no clear shared code-path or bug." + "reason": "Different failures: a training script regression vs a NoneType error in UNet forward." }, { "left": "issue:1475", "right": "issue:579", "accept": false, - "reason": "Different training setups and different failures: mixed-class DreamBooth/textual inversion behavior vs missing config.json." + "reason": "DreamBooth/textual-inversion training setup issue vs missing config.json for textual inversion models." }, { "left": "issue:10782", "right": "issue:7185", "accept": false, - "reason": "Unrelated topics: Lumina Image issue vs PyTorch version compatibility question." + "reason": "A Lumina Image 2.0 bug and a general PyTorch version compatibility question are not the same issue." }, { "left": "issue:2437", "right": "issue:3891", "accept": false, - "reason": "Both are feature requests for new pipelines, but they are for different models and changes." + "reason": "Both are feature requests for pipelines, but they target completely different pipelines and changes." }, { "left": "issue:2250", "right": "issue:3672", "accept": false, - "reason": "DreamBooth inference failure and concurrent-thread inference failure are different bugs." + "reason": "DreamBooth prompt-generation failure and StableDiffusionPipeline thread-safety failure are different code-path bugs." }, { "left": "issue:11136", "right": "issue:12126", "accept": false, - "reason": "Different model-specific inference errors with no evidence of the same underlying defect." + "reason": "Different pipelines and root causes: SCM timestep restriction vs a cache_context attribute error in wan2.2." }, { "left": "issue:5455", "right": "issue:6371", "accept": false, - "reason": "Notebook/subprocess failure for DreamBooth LoRA vs an assert in controlnet_xs example; not the same issue." + "reason": "Separate example/notebook failures in different training examples with unrelated assertions/process errors." }, { "left": "issue:10722", "right": "issue:11221", "accept": false, - "reason": "Tensor size mismatch during DreamBooth training vs a question about autoencoder non-convergence; different problems." + "reason": "Tensor-shape mismatch during DreamBooth prior preservation is unrelated to autoencoderkl non-convergence." }, { "left": "issue:8894", "right": "issue:9415", "accept": false, - "reason": "Different model/training issues; the shared \"raise error\" wording is too generic to imply the same bug." + "reason": "A training-script runtime error vs a request to add a new parameter are not the same bug." }, { "left": "issue:4446", "right": "issue:7420", "accept": false, - "reason": "Custom text encoder loading from_single_file vs torch.compile support; unrelated code paths." + "reason": "Custom text_encoder loading and torch.compile support are different feature areas with different failures." }, { "left": "issue:11374", "right": "issue:13053", "accept": false, - "reason": "Both are feature requests to add pipeline support, but for different models and implementations." + "reason": "Both are feature requests, but for different model pipelines and no shared concrete change." }, { "left": "issue:7708", "right": "issue:8939", "accept": false, - "reason": "Both are tokenizer/model-loading complaints, but the missing files and model repos differ and the failures are not shown to be the same." + "reason": "Both are tokenizer-loading problems, but they involve different models and likely different missing-file causes." }, { "left": "issue:3250", "right": "issue:5821", "accept": false, - "reason": "Model conversion failure vs CUDA backward/training error; no shared concrete bug." + "reason": "A pipeline runtime failure and a CUDA backward/training issue are not the same underlying bug." }, { "left": "issue:4250", "right": "issue:4888", "accept": false, - "reason": "Both are loading-related, but one is a generic incorrect load report and the other is offline safetensors behavior; too broad to merge." + "reason": "Model conversion failure and offline safetensors loading are distinct workflows." }, { "left": "issue:10863", "right": "issue:9667", "accept": false, - "reason": "Local safetensors loading issue vs loading a FLUX fp8 model with a local transformer file; different artifacts and failure modes." + "reason": "Both involve local model loading, but one is a safetensors-file load error and the other is FLUX fp8 transformer loading." }, { "left": "issue:11182", "right": "issue:9343", "accept": false, - "reason": "BF16 variant causing unnecessary download is not the same as FLUX load failure with mismatched-size handling." + "reason": "One reports an unnecessary download for BF16 variants; the other is a FLUX mismatched-size loading error." }, { "left": "issue:10512", "right": "issue:9108", "accept": false, - "reason": "LoRA loading on FLUX/Quanto vs resuming training with a quantized base model; related theme but not the same bug." + "reason": "Quantization is shared, but one is LoRA load failure and the other is resuming training with a quantized base model." }, { "left": "issue:4185", "right": "issue:6666", "accept": false, - "reason": "Both mention black images, but one is SDXL ControlNet inference and the other is validation-image output; insufficient evidence of a shared root cause." + "reason": "Both mention black images, but one is SDXL ControlNet inference and the other is validation imagery during training." }, { "left": "issue:2856", "right": "issue:6689", "accept": false, - "reason": "Both are conversion-to-diffusers issues, but they concern different models and likely different conversion failures." + "reason": "Both are model-to-diffusers conversion complaints, but the titles point to different models and no shared concrete bug is evident." } ] }, @@ -27433,10 +27451,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -27451,162 +27469,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous; most soft links are only broad topical similarity. Two pairs look like true duplicates: local_files_only loading behavior and generator-list handling.", - "confidence": 0.72, - "canonical_issue_reason": "Issue 4307 is the clearest canonical candidate because it names a concrete loading failure around local_files_only, which matches the strongest duplicate-like theme in the set.", + "summary": "Mostly a mixed bag of unrelated issue reports; only a few soft duplicate mini-clusters stand out: offline/local_files_only loading, CPU-core underutilization, and list-of-generators handling.", + "confidence": 0.64, + "canonical_issue_reason": "No single canonical issue fits the whole set because the reports span many different bugs; the closest umbrella duplicate cluster is the offline/local_files_only loading family, with the broader representative being the auto-pipeline breakage report.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 4307 is the best representative issue here: it is specific, reproducible, and captures the core offline-loading bug more cleanly than the other candidates.", + "best_issue_reason": "Issue 4727 is the strongest representative of the most general, user-facing duplicate family here: offline/local_files_only loading failures affecting pipeline construction, not just one narrow from_pretrained call.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4307", "right": "issue:4727", "accept": true, - "reason": "Both point to the same offline/local_files_only loading bug in pipeline/model loading." + "reason": "Both report the same underlying offline/local_files_only model-loading failure; 4727 is the broader auto-pipeline manifestation of the same bug." }, { "left": "issue:10760", "right": "issue:7516", "accept": false, - "reason": "Different failures: missing ONNX import/class vs missing Diffusers pipeline attribute." + "reason": "Both are import/attribute errors, but they concern different missing pipeline symbols and do not look like the same concrete bug." }, { "left": "issue:1109", "right": "issue:834", - "accept": false, - "reason": "Same broad symptom (CPU underuse) but not the same concrete bug or code path." + "accept": true, + "reason": "Both describe the same CPU underutilization problem; the titles differ, but the underlying complaint is the same performance issue." }, { "left": "issue:10150", "right": "issue:12505", "accept": false, - "reason": "Both are performance-related, but they involve different pipelines and different failure modes." + "reason": "Both mention slowness/throughput, but they target different pipelines and different failure modes." }, { "left": "issue:2984", "right": "issue:5442", "accept": false, - "reason": "Unrelated topics: a video-diffusion model paper/request vs a specific upscale pipeline." + "reason": "These are different video-generation pipeline topics; no evidence they are the same bug or change." }, { "left": "issue:3348", "right": "issue:5504", "accept": false, - "reason": "Both mention LoRA/xformers, but the bugs are different: scale-parameter support vs PEFT LoRA failure." + "reason": "Both involve LoRA/xformers, but one is about the scale parameter and the other about PEFT LoRA loading; too broad to treat as the same bug." }, { "left": "issue:10636", "right": "issue:7418", "accept": false, - "reason": "Different code paths: value_guided_sampling bug vs a community upscale example line issue." + "reason": "Different files and different reported failures; only superficial similarity in being code bugs." }, { "left": "issue:5693", "right": "issue:6176", "accept": false, - "reason": "Both are load errors, but one is VAE loading and the other is LoRA weight loading after DreamBooth training." + "reason": "Both are loading-related errors, but one is a VAE load failure and the other is a LoRA post-training load issue." }, { "left": "issue:6832", "right": "issue:7049", "accept": false, - "reason": "Both involve IP Adapter examples, but the concrete failures differ ('tuple' shape access vs unpacking count mismatch)." + "reason": "Both are IP-Adapter issues, but they are distinct shape/unpack errors in different examples." }, { "left": "issue:1083", "right": "issue:4321", "accept": false, - "reason": "Both are FP16/ONNX output problems, but the backend/pipeline contexts differ and the symptoms are not clearly the same bug." + "reason": "Both concern FP16/ONNX output problems, but the reported symptoms and environments differ materially." }, { "left": "issue:5466", "right": "issue:7770", "accept": false, - "reason": "Both mention tensor-shape mismatch during LoRA loading, but the titles point to different loading scenarios and are not clearly the same defect." + "reason": "Both mention LoRA shape mismatches, but one is SDXL img2img LoRA loading and the other is a broader model+LoRA load mismatch." }, { "left": "issue:12728", "right": "issue:8518", "accept": false, - "reason": "Different models and different torch.compile/Dynamo errors; not the same concrete bug." + "reason": "Both involve torch.compile, but they are different models and different Dynamo failure modes." }, { "left": "issue:2091", "right": "issue:3451", "accept": false, - "reason": "Both are training-related dtype/multi-GPU issues, but they describe different symptoms and root causes." + "reason": "These are unrelated training issues: checkpoint resume/multi-GPU versus fp16 memory use." }, { "left": "issue:9546", "right": "issue:9927", "accept": false, - "reason": "Both concern training stability, but memory exhaustion and HeaderTooLarge are different problems." + "reason": "Both are training-time problems, but one is validation OOM and the other is a header-size error; not the same bug." }, { "left": "issue:10937", "right": "issue:11584", "accept": false, - "reason": "torch.compile/VAE encode errors are unrelated to group-offloading incompatibility." + "reason": "Both mention torch.compile, but one is a VAE encode error and the other is a group-offloading compatibility problem." }, { "left": "issue:11661", "right": "issue:12078", "accept": false, - "reason": "Different Flux-Control example bugs: None max_train_steps handling vs validation input example content." + "reason": "Both are Flux-control training example issues, but one is a None handling bug and the other is validation-input content." }, { "left": "issue:11046", "right": "issue:12459", "accept": false, - "reason": "Same general Flux area, but one is a multi-condition inference feature bug and the other is a batch-throughput issue." + "reason": "Both are Flux inference issues, but one is a functional controlnet/inpainting/IP-adapter setup and the other is a batch throughput complaint." }, { "left": "issue:10010", "right": "issue:11418", "accept": false, - "reason": "Both mention FLUX Fill, but one is a missing model_index.json load failure and the other is about adding an fp8 weights file." + "reason": "Both are about FLUX Fill assets, but one is a missing model_index.json fetch and the other is adding an fp8 safetensor file." }, { "left": "issue:3944", "right": "issue:5160", "accept": false, - "reason": "Different training features: LoRA rank customization vs masking support." + "reason": "Different Dreambooth/LoRA feature requests; same broad area, but not the same bug." }, { "left": "issue:5969", "right": "issue:6136", "accept": false, - "reason": "Both mention NaNs, but one is DDIM/ZSNR-specific and the other is SD2.1 on Mac CPU." + "reason": "Both are NaN reports, but the patch-specific DDIM issue and the Mac CPU SD2.1 issue have different triggers and code paths." }, { "left": "issue:2163", "right": "issue:6518", "accept": false, - "reason": "Both are training/example errors, but dtype mismatch and None division are unrelated." + "reason": "Both are training-script errors, but they involve different root causes: dtype mismatch versus NoneType floor-division." }, { "left": "issue:4974", "right": "issue:861", "accept": false, - "reason": "Both are conversion-related, but converting safetensors to diffusers is not the same bug as converting ckpt to diffusers." + "reason": "Both are model conversion topics, but one is safetensor-to-diffusers and the other is ckpt-to-diffusers; not the same bug." }, { "left": "issue:2068", "right": "issue:2321", "accept": false, - "reason": "Different issues: from_pretrained dtype conversion vs EMA weight assignment during inference." + "reason": "These are both dtype/model-loading discussions, but they report different problems and code paths." }, { "left": "issue:1909", "right": "issue:2560", "accept": true, - "reason": "Both describe the same generator-list handling bug affecting pipeline/scheduler behavior and output consistency." + "reason": "Both point to the same generator-list handling bug in pipelines/schedulers, where passing a list of generators produces incorrect behavior." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays appropriately conservative on the false links. The two accepted pairs are well supported by their titles: issue:4307/issue:4727 both concern local_files_only loading behavior, and issue:1909/issue:2560 both concern list-of-generators handling. The rest of the verdicts avoid overclaiming duplicate status." + "feedback": "The clustering is mostly well-grounded in the packet and the soft-edge accepts are conservative. The accepted pairs (4307/4727, 1109/834, 1909/2560) are credible duplicate-family matches, while the rejections avoid overclaiming on only superficially similar reports. The summary stays appropriately narrow and does not overstate a single umbrella bug across unrelated issues." }, "evaluator_used": true, "retried": false, @@ -27706,10 +27724,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -27724,156 +27742,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is thematically noisy: several Flux/ControlNet/SDXL/training issues share vocabulary, but none of the soft-edge pairs appear to be the same underlying bug or change.", - "confidence": 0.72, - "canonical_issue_reason": "issue:11247 is the clearest actionable open bug in the set and the best representative of the small Flux/controlnet runtime cluster, even though the overall collection is too heterogeneous to form a true duplicate group.", + "summary": "The cluster is broadly heterogeneous; the soft pairs mostly share subsystem vocabulary but not the same concrete bug or change. I would reject all soft edges and avoid selecting a single canonical issue/PR for the whole set.", + "confidence": 0.87, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:11247 is the strongest single issue to anchor on because it is concrete, reproducible, and closer to the Flux-related items than the broader or unrelated reports.", + "best_issue_reason": "No clear best issue: the items span unrelated topics (Flux, SDXL, ControlNet, training scripts, conversion, LoRA, video pipelines), so there is no single representative bug report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3564", "right": "issue:5168", "accept": false, - "reason": "Both are ControlNet feature requests, but Uni-ControlNet and ControlNet-XS are different extensions, not the same bug or change." + "reason": "Both are ControlNet feature requests, but Uni-ControlNet and ControlNet-XS are different variants and likely separate changes." }, { "left": "issue:5917", "right": "issue:9223", "accept": false, - "reason": "One is a checkpoint conversion failure; the other is a dependency/version conflict. They are unrelated." + "reason": "One is a checkpoint conversion failure; the other is a dependency/version conflict about pyarrow. Different problems." }, { "left": "issue:8534", "right": "issue:8557", "accept": false, - "reason": "Both concern SD3 training, but one is about weighting and the other about token limits; different failure modes." + "reason": "Both concern SD3 training, but one is logit-normal weighting and the other is token-limit behavior. Not the same bug." }, { "left": "issue:2083", "right": "issue:3184", "accept": false, - "reason": "Both mention generators, but the scheduler/eta behavior and the generator-list community pipeline issue are distinct." + "reason": "Both mention generators, but one is DDIMScheduler eta behavior and the other is a community pipeline generator-list issue. Different code paths." }, { "left": "issue:3235", "right": "issue:7286", "accept": false, - "reason": "Both involve Custom Diffusion loading, but they target different weight-loading paths and symptoms." + "reason": "Both are about Custom Diffusion weight loading, but the reported failures are different enough to look like separate loader issues." }, { "left": "issue:4790", "right": "issue:5737", "accept": false, - "reason": "Both are conversion-script bugs, but one is a round-trip file compatibility issue and the other is an SDXL conversion failure." + "reason": "Both involve conversion scripts, but they are opposite directions and different model contexts; not one underlying bug." }, { "left": "issue:10057", "right": "issue:11024", "accept": false, - "reason": "FluxTransformer2DModel token/file loading and SiglipImageProcessor .to() support are unrelated code paths." + "reason": "Flux single-file token handling and SiglipImageProcessor .to() are unrelated issues." }, { "left": "issue:12459", "right": "issue:12738", "accept": false, - "reason": "Both mention batch-size problems, but they affect different pipelines and different underlying failures." + "reason": "Both mention batching, but one is a throughput concern and the other is an assertion failure in ZImagePipeline." }, { "left": "issue:2371", "right": "issue:817", "accept": false, - "reason": "Both are DreamBooth example issues, but one is about missing outputs and the other about mixed precision." + "reason": "Both are Dreambooth-related, but one is missing trained objects and the other is mixed-precision training. Different failures." }, { "left": "issue:1524", "right": "issue:1543", "accept": false, - "reason": "Textual inversion training quality and fp16 revision loading are separate problems." + "reason": "Same script family, but poor training quality and fp16 revision handling are not the same defect." }, { "left": "issue:1539", "right": "issue:3799", "accept": false, - "reason": "Both touch typing, but one is a conversion-script type bug and the other is a scheduler type inference bug." + "reason": "These are both type-related, but one is a conversion-script type bug and the other is an inference typing issue. Not duplicates." }, { "left": "issue:2321", "right": "issue:5621", "accept": false, - "reason": "Not copying EMA weights and SDXL ControlNet encoder_hid_proj support are different issues." + "reason": "Model-copy/EMA behavior is unrelated to SDXL ControlNet encoder_hid_proj support." }, { "left": "issue:2081", "right": "issue:5224", "accept": false, - "reason": "Both are JAX/Flax-related, but one is model conversion and the other is a runtime/error case; not the same bug." + "reason": "Both involve JAX/Flax diffusion, but one is conversion from PyTorch weights and the other is a Flax SD 2.1 runtime error." }, { "left": "issue:4402", "right": "issue:7168", "accept": false, - "reason": "One is a from_single_file pipeline misuse issue; the other is about IP-Adapter embedding tensor shape." + "reason": "Both concern SDXL/IP-adapter style usage, but one is from_single_file inpaint and the other is image-embed tensor shape handling." }, { "left": "issue:10191", "right": "issue:9856", "accept": false, - "reason": "GPU selection and distributed port collision are unrelated training/runtime setup problems." + "reason": "GPU selection for training and distributed port conflicts are unrelated." }, { "left": "issue:2991", "right": "issue:3101", "accept": false, - "reason": "Both are ControlNet example errors, but they describe different failures and likely different fixes." + "reason": "Both are ControlNet training example errors, but one is explicitly mixed-precision and the other is a generic failure; insufficient evidence of the same bug." }, { "left": "issue:6258", "right": "issue:7400", "accept": false, - "reason": "Both are Stable Video Diffusion quality failures, but forward chunking degradation and black images are different symptoms." + "reason": "Both are Stable Video Diffusion issues, but forward chunking quality degradation and black-image inference are different symptoms and likely different causes." }, { "left": "issue:5529", "right": "issue:6791", "accept": false, - "reason": "Same training script family, but the reported failures are not clearly the same bug." + "reason": "Same training script, but one report is a bug and the other is a general inability to run. Too vague to merge as one defect." }, { "left": "issue:9717", "right": "issue:9911", "accept": false, - "reason": "A Flux Docker image error and a multi-ControlNet layer-length error are different problems." + "reason": "Flux docker-image setup and multi-ControlNet layer-length mismatch are unrelated." }, { "left": "issue:10782", "right": "issue:12666", "accept": false, - "reason": "Lumina Image 2.0 and Kandinsky 5.0 Video Pipeline are unrelated model/pipeline issues." + "reason": "Lumina Image 2.0 and Kandinsky 5.0 Video Pipeline are different model families and different issues." }, { "left": "issue:1083", "right": "issue:1363", "accept": false, - "reason": "FP16 ONNX incorrect output and a long-prompt-weight scheduler error are different bugs." + "reason": "FP16 ONNX output correctness and long-prompt/LMS scheduler errors are unrelated." }, { "left": "issue:11247", "right": "issue:9637", "accept": false, - "reason": "Flux multi-GPU/controlnet distribution and XFormers attention-mask bf16 handling are different code paths." + "reason": "Multi-GPU Flux controlnet distribution and XFormers attention-mask/bfloat16 behavior are different code paths." }, { "left": "issue:4724", "right": "issue:6283", "accept": false, - "reason": "Both are SDXL img2img-related, but one concerns checkpoint behavior and the other a strength=0 edge case." + "reason": "Both are SDXL img2img-related, but one is checkpoint compatibility and the other is strength=0 behavior in ControlNet img2img." }, { "left": "issue:11657", "right": "issue:13547", "accept": false, - "reason": "Both are LoRA adapter loading issues, but they affect different model families and different loading failures." + "reason": "Both are LoRA loading problems, but they target different transformer families (Wan vs Flux2) and do not look like the same bug." } ] }, @@ -27976,10 +27994,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -27994,156 +28012,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is noisy and spans several unrelated diffusion/training bugs and feature requests. The only repeated theme is \u2018training/inference issues in ControlNet/LoRA/compile flows,\u2019 but none of the soft pairs look like the same underlying bug/change.", - "confidence": 0.79, - "canonical_issue_reason": "Issue 4086 is the most concrete representative of the recurring ControlNet training-script failures, but it is still just one specific SDXL ControlNet training problem rather than a catch-all duplicate target for the whole cluster.", + "summary": "The set is mostly a loose thematic cluster around diffusion training, ControlNet, LoRA, and device/dtype issues, but the soft pairs are not the same underlying bugs. I would not merge any of the candidate duplicates.", + "confidence": 0.68, + "canonical_issue_reason": "Issue 10055 is the best representative of the cluster\u2019s core theme: ControlNet training failing in a specific SD3 training script. The other issues are mostly separate regressions, different model families, or feature requests.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 4086 is the best single representative because it is detailed, concrete, and squarely in the cluster\u2019s main theme of ControlNet training failures; however, it should not be treated as the duplicate of the other items.", + "best_issue_reason": "10055 is the clearest and most central issue to use as a reference point for the controlnet-training failure subset, especially compared with the more peripheral dtype/loading/compile/feature-request items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1099", "right": "issue:1475", "accept": false, - "reason": "Both are DreamBooth-related, but one is about missing class names and the other about training with two image classes; different bugs." + "reason": "Both mention DreamBooth, but one is about missing class names while the other is about using two class types; different failures." }, { "left": "issue:7003", "right": "issue:7770", "accept": false, - "reason": "Both mention tensor-shape problems, but one is an inpainting weight-shape mismatch and the other is a model/LoRA loading mismatch in Python." + "reason": "Both are shape-mismatch style errors, but they occur in different code paths and with different artifacts (inpaint pipeline vs model/LoRA loading)." }, { "left": "issue:3279", "right": "issue:3363", "accept": false, - "reason": "Both concern train_dreambooth_lora.py, but one is resume-from-checkpoint and the other is a broader failure on two machines; not the same bug." + "reason": "Resume-from-checkpoint errors and machine-specific failures are not the same bug." }, { "left": "issue:3405", "right": "issue:5972", "accept": false, - "reason": "One is a ControlNet training process bug; the other is a documentation request about Min-SNR weighting." + "reason": "One is a ControlNet training bug, the other is a documentation issue about Min-SNR weighting; not duplicates." }, { "left": "issue:10055", "right": "issue:10412", "accept": false, - "reason": "Both are SD3 training topics, but one is a ControlNet script failure and the other is excessive VRAM usage in DreamBooth." + "reason": "Both are SD3 training-related, but one is a broken ControlNet training script and the other is excessive VRAM usage in DreamBooth; different problems." }, { "left": "issue:10932", "right": "issue:11584", "accept": false, - "reason": "Both involve torch.compile on HunyuanVideo, but one is recompilation-limit churn and the other is incompatibility with group offloading." + "reason": "Both involve torch.compile on HunyuanVideo, but the reported failures are different (recompile limit vs group offloading incompatibility)." }, { "left": "issue:11222", "right": "issue:11674", "accept": false, - "reason": "These are separate feature requests for different WAN variants; not the same change." + "reason": "Both are WAN-related feature requests, but they ask for different capabilities and are not the same change." }, { "left": "issue:4217", "right": "issue:5128", "accept": false, - "reason": "Both involve safetensors loading, but one is slow loading and the other is an actual load error for XLPipeline." + "reason": "Both concern safetensors/loading, but one is slow loading and the other is a load failure in a different API." }, { "left": "issue:4686", "right": "issue:4697", "accept": false, - "reason": "Both are loading issues, but one is a saved pipeline and the other is a specific LoRA regression after an update." + "reason": "Both are loading issues, but one is a saved pipeline reload problem and the other is a specific LoRA update regression." }, { "left": "issue:9045", "right": "issue:9895", "accept": false, - "reason": "Both are Flux/SD3 device problems, but one is accelerate+peft CUDA transfer failure and the other is a commit-specific CPU/CUDA mismatch in inference." + "reason": "Both are device-placement issues, but they affect different pipelines and arise from different causes." }, { "left": "issue:2988", "right": "issue:4303", "accept": false, - "reason": "Scheduler crash on equal step counts is unrelated to Flax SDXL loading failure." + "reason": "Unrelated bugs: scheduler edge case versus Flax SDXL load crash." }, { "left": "issue:3480", "right": "issue:4876", "accept": false, - "reason": "Both ask for LoRA-related parameters, but one is about scaling layers and the other about standard deviation; different enhancements." + "reason": "Both are LoRA parameter feature requests, but they request different knobs and do not describe the same change." }, { "left": "issue:6879", "right": "issue:6880", "accept": false, - "reason": "The first is a dtype mismatch in Custom Diffusion; the second is a separate retrieve.py issue with no clear shared failure mode." + "reason": "Same broad Custom Diffusion area, but one is a dtype mismatch and the other is about retrieve.py; not the same bug." }, { "left": "issue:4427", "right": "issue:7660", "accept": false, - "reason": "Both say an example does not work, but they are different examples and different code paths." + "reason": "Both mention examples not working, but they are different examples with different functionality." }, { "left": "issue:1100", "right": "issue:4073", "accept": false, - "reason": "Different runtime exceptions in different components: a missing device attribute vs astype on Tensor." + "reason": "Different exceptions in different components: CLIPTextTransformer device access vs DeepFloyd IF safety checker astype." }, { "left": "issue:1543", "right": "issue:660", "accept": false, - "reason": "Both involve fp16/revision handling, but one is textual inversion revision parsing and the other is MPS float16 execution." + "reason": "Both mention fp16/revision, but one is textual inversion-specific and the other is an MPS float16 runtime issue." }, { "left": "issue:1695", "right": "issue:1734", "accept": false, - "reason": "Both are multi-GPU DreamBooth topics, but one is DeepSpeed setup and the other is accelerate performance regression." + "reason": "Both are multi-GPU DreamBooth issues, but one is deepspeed training setup and the other is accelerate performance regression." }, { "left": "issue:9525", "right": "issue:9803", "accept": false, - "reason": "Both are Flux LoRA issues, but one is scale being ignored and the other is loading failure; different bugs." + "reason": "Both are Flux LoRA-loading related, but one is about lora_scale being ignored and the other is a loading failure." }, { "left": "issue:9012", "right": "issue:9496", "accept": false, - "reason": "Both are SD3ControlNetModel problems, but one is weight loading and the other is forward-pass runtime error." + "reason": "Same model family, but one fails when loading weights and the other fails in forward execution; different code paths." }, { "left": "issue:4086", "right": "issue:7982", "accept": false, - "reason": "ControlNet training failure versus a from_pipe MultiControlNetModel component bug; different functionality." + "reason": "Both involve ControlNet, but one is a training failure and the other is a from_pipe / MultiControlNetModel integration issue." }, { "left": "issue:2053", "right": "issue:4993", "accept": false, - "reason": "VRAM regression and missing import structure entry are unrelated." + "reason": "Different regressions entirely: VRAM increase versus missing import structure entry." }, { "left": "issue:5687", "right": "issue:6283", "accept": false, - "reason": "One is duplicate timesteps with Karras sigmas; the other is ControlNet img2img failing at strength=0." + "reason": "Both affect img2img-style behavior, but one is Karras timestep duplication and the other is a strength=0 ControlNet pipeline edge case." }, { "left": "issue:6096", "right": "issue:7159", "accept": false, - "reason": "Both mention torch.compile, but one is a DynamicOutputShapeException in SVD and the other is a LoRA-fused compile failure." + "reason": "Both are torch.compile failures, but they involve different models and different compiler errors." }, { "left": "issue:10382", "right": "issue:8604", "accept": false, - "reason": "These are different dtype/return-type problems in different modules: VAE encoder forward return mismatch versus T5 float16 corruption." + "reason": "Different bugs in different components: VAE forward return type mismatch versus T5 float16 corruption." } ] }, @@ -28246,10 +28264,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -28264,162 +28282,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly a mix of unrelated issues. I\u2019d treat two soft-edge pairs as true duplicates: the LoRA + Dreambooth model-loading breakage, and the missing model_index/config.json loading error. The rest look like distinct bugs or feature requests in nearby subsystems, but not the same underlying problem.", - "confidence": 0.82, - "canonical_issue_reason": "issue:3174 is the clearest representative of the LoRA + Dreambooth loading failure; it names the combined workflow and the breakage mechanism more specifically than the paired duplicate.", + "summary": "This cluster is mostly a grab-bag of unrelated bugs and feature requests. The only clear duplicate-looking pair is the LoRA/Dreambooth loading regression (3174/3191); the rest are different models, error modes, or feature requests.", + "confidence": 0.89, + "canonical_issue_reason": "issue:3174 is the broader, more descriptive report of the LoRA/Dreambooth loading regression and likely subsumes issue:3191.", "canonical_pr_reason": null, - "best_issue_reason": "issue:3174 is the strongest overall issue candidate among the likely duplicate pairs because it is specific, descriptive, and clearly subsumes the paired Dreambooth+LoRA report.", + "best_issue_reason": "issue:3174 is the best canonical issue here because it clearly identifies the regression cause and scope, making it the strongest anchor for the duplicate pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:8832", "right": "issue:9106", "accept": false, - "reason": "Both are runtime failures, but they reference different models and different error modes; not the same bug." + "reason": "Different models and failures: a generic pipeline crash versus a PixArt-Alpha attribute error at time_step=1." }, { "left": "issue:10178", "right": "issue:9112", "accept": false, - "reason": "Both mention tensor/dimension mismatches, but the titles point to different pipelines and distinct shape problems." + "reason": "Both are tensor-size problems, but they affect different components/models and appear to be different code paths." }, { "left": "issue:10782", "right": "issue:12195", "accept": false, - "reason": "Lumina minor issue and a PyTorch custom_op compatibility error are unrelated." + "reason": "Unrelated issues: Lumina minor issue versus an AutoencoderKL custom_op compatibility break with PyTorch 2.3.1." }, { "left": "issue:5894", "right": "issue:5932", "accept": false, - "reason": "Generic invalid buffer size vs. a 4-channel/3-channel input mismatch; different failure causes." + "reason": "Different runtime failures: invalid buffer size versus channel-mismatch in the UNet input." }, { "left": "issue:4389", "right": "issue:5577", "accept": false, - "reason": "Both concern ControlNet loading, but one is a generic load error and the other is a specific state-dict KeyError; not enough to treat as the same bug." + "reason": "Both involve ControlNet loading, but one is a generic load error and the other is a specific from_single_file KeyError; not the same bug." }, { "left": "issue:2784", "right": "issue:7243", "accept": false, - "reason": "A Dreambooth support request and a ResAdapter feature request are unrelated feature topics." + "reason": "Feature requests for different capabilities: Dreambooth on text2video versus ResAdapter support." }, { "left": "issue:4043", "right": "issue:6201", "accept": false, - "reason": "Both involve prompt embeddings, but they describe different API/validation bugs and different code paths." + "reason": "Both concern prompt-embedding handling, but the exact pipeline behavior and failure condition differ." }, { "left": "issue:3803", "right": "issue:5575", "accept": false, - "reason": "Multi-GPU LoRA training sharing and transformer_in hidden size configuration are separate concerns." + "reason": "Multi-GPU LoRA training sharing issue versus a question about hidden-size configuration; not duplicates." }, { "left": "issue:10334", "right": "issue:10914", "accept": false, - "reason": "MacOS/MPS NaNs and unintended CPU offloading are different device-behavior issues." + "reason": "Different symptoms and scopes: MacOS/MPS gray images and CPU NaNs versus unexpected CPU offloading." }, { "left": "issue:10674", "right": "issue:7982", "accept": false, - "reason": "GGUF Flux loading and from_pipe with MultiControlNetModel are unrelated pipeline/component problems." + "reason": "Different problems in different contexts: Flux GGUF loading versus from_pipe with MultiControlNetModel." }, { "left": "issue:2364", "right": "issue:3805", "accept": false, - "reason": "Both are generic execution failures, but they do not identify the same underlying bug or code path." + "reason": "Both are generic run failures, but they are separate execution environments and error reports." }, { "left": "issue:4305", "right": "issue:7420", "accept": false, - "reason": "Both mention torch.compile, but one is a release-caused graph break and the other is a broader inability to use compile; not clearly the same defect." + "reason": "Related to torch.compile, but one is a version-specific graph break regression and the other is a general inability to use compile." }, { "left": "issue:4186", "right": "issue:4788", "accept": false, - "reason": "SDXL inpaint with refiner and SDXL inpaint OOM are different issues." + "reason": "Different failure modes: SDXL inpaint/refiner behavior versus out-of-memory on low VRAM." }, { "left": "issue:9484", "right": "issue:9763", "accept": false, - "reason": "Multigpu dreambooth training and token truncation in Flux ControlNet training are unrelated." + "reason": "Both are Flux training issues, but one is multi-GPU Dreambooth training and the other is token truncation in ControlNet training." }, { "left": "issue:11827", "right": "issue:12459", "accept": false, - "reason": "Same FLUX Kontext family, but one is a safetensors deserialization error and the other is a throughput issue." + "reason": "Completely different issues: safetensors header corruption versus batch inference throughput." }, { "left": "issue:12082", "right": "issue:12529", "accept": false, - "reason": "Compile-time VAE memory growth and a tiled VAE tensor size mismatch are distinct bugs." + "reason": "Different Wan2.1 problems: post-compile VAE memory growth versus tiled VAE tensor-size mismatch." }, { "left": "issue:3694", "right": "issue:5259", - "accept": true, - "reason": "Both describe the same missing model_index.json/config.json-style loading failure when resolving a model directory." + "accept": false, + "reason": "Similar missing-model_index wording, but one is an example-training path problem and the other is a generic directory loading error." }, { "left": "issue:2554", "right": "issue:3450", "accept": false, - "reason": "Accelerate version requirement mismatch and from_ckpt dev-version loading failure are different compatibility problems." + "reason": "Different dependency/version issues: unreleased accelerate requirement versus from_ckpt failing on dev version." }, { "left": "issue:7263", "right": "issue:7722", "accept": false, - "reason": "IP Adapter with attention slicing and Animatediff+ControlNet+IPAdapter interaction are different pipeline bugs." + "reason": "Both mention IP-Adapter, but one is attention-slicing incompatibility and the other is an Animatediff-ControlNet+IPAdapter error." }, { "left": "issue:6333", "right": "issue:8472", "accept": false, - "reason": "Pre-trained LoRA finetuning and questions about simultaneous module pre-training are not the same issue." + "reason": "One is a feature/question about PEFT LoRA fine-tuning; the other is a separate question about simultaneous modules." }, { "left": "issue:10150", "right": "issue:7785", "accept": false, - "reason": "Both mention slowness, but one is about SDXL pipeline loading/inference generally and the other is specifically slow first inference in a particular version." + "reason": "Both are performance complaints, but one is broad SDXL loading/inference slowness and the other is first-inference latency in a specific version." }, { "left": "issue:3174", "right": "issue:3191", "accept": true, - "reason": "These both describe the same LoRA + Dreambooth model-loading breakage." + "reason": "Same underlying LoRA/Dreambooth loading regression; 3174 is the broader report and 3191 is a shorter restatement." }, { "left": "issue:5314", "right": "issue:6372", "accept": false, - "reason": "Two separate pipeline-combination feature requests, not the same bug." + "reason": "Different pipeline integration requests: ControlNet reference inpainting versus SVD + ControlNet." }, { "left": "issue:5630", "right": "issue:7142", "accept": false, - "reason": "SDXL Img2Img scheduler artifacts and SDXL-lightning inference mismatch are different inference-quality issues." + "reason": "Both are SDXL-related quality/performance issues, but one is scheduler noise residue and the other is SDXL-lightning output mismatch with webui." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet and stays mostly conservative. The two accepted duplicate pairs are strongly supported by the titles/bodies: the LoRA+Dreambooth loading breakage, and the missing model_index.json/config.json loading failure. The remaining pairs are treated as distinct issues without overclaiming shared evidence." + "feedback": "The summary is well grounded in the packet, and the duplicate call on issue:3174/issue:3191 is suitably conservative. The other soft-edge verdicts avoid overclaiming and are supported by the issue titles alone." }, "evaluator_used": true, "retried": false, @@ -28518,10 +28536,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -28536,156 +28554,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly false-positive similarity matches rather than true duplicates. The only clear thematic overlap is a loose concentration around Flux/LoRA loading and a few scheduler/controlnet reports, but the paired items describe different bugs, models, or error modes.", - "confidence": 0.91, - "canonical_issue_reason": "issue:10512 is the strongest anchor for the cluster theme: it is an active, well-discussed report about Flux LoRA loading failure, which is the closest shared topic among the LoRA/Flux-related items.", + "summary": "This cluster is mostly a false positive: the items span many distinct FLUX/LoRA/ControlNet/scheduler issues rather than one duplicate bug. The only moderately close pairs are same-theme reports about LoRA loading or Flux training, but they still appear to be different concrete failures or feature requests.", + "confidence": 0.86, + "canonical_issue_reason": "If one issue must anchor the cluster, #10512 is the clearest concrete bug report with substantial discussion and the most directly stated FLUX LoRA loading failure. It is still not a true umbrella duplicate for the rest of the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10512 is the best general triage bucket because it is broad enough to represent Flux-LoRA load regressions and has the most discussion/activity among the nearby reports, though it is not a true duplicate hub for the whole set.", + "best_issue_reason": "#10512 is the strongest representative of the largest apparent theme (Flux LoRA loading problems): it is specific, actively discussed, and describes an actionable runtime failure. The other items are either broader feature requests, different model families, or unrelated errors.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4476", "right": "issue:5997", "accept": false, - "reason": "Both mention ControlNet/SDXL, but one is about poor results on fill50k while the other is a tutorial/model-type mismatch error. Different bug and code path." + "reason": "Both mention ControlNet/SDXL-ish setup problems, but one is about poor results on fill50k and the other is a model-type mismatch in a tutorial; different failure modes and not the same bug." }, { "left": "issue:4852", "right": "issue:4975", "accept": false, - "reason": "One is a LoRA loading failure; the other is a performance issue from repeated datatype queries. Same broad subsystem, not the same underlying bug." + "reason": "Both are LoRA-loading related, but #4852 is a generic loading error while #4975 is an efficiency/performance complaint about repeated datatype queries." }, { "left": "issue:3174", "right": "issue:5606", "accept": false, - "reason": "lora+dreambooth loading breakage and missing-alpha LoRA loading are distinct failure modes with different causes." + "reason": "Both involve LoRA loading, but #3174 is DreamBooth + LoRA model-loading breakage from new LoRA features, while #5606 is specifically missing alpha handling." }, { "left": "issue:1076", "right": "issue:5474", "accept": false, - "reason": "Completely different topics: a generic Stable Diffusion v1.5 question versus a new reference-only pipeline request." + "reason": "These are unrelated across time and topic; #1076 is an early SD 1.5 question, while #5474 proposes a new reference-only pipeline." }, { "left": "issue:3064", "right": "issue:6130", "accept": false, - "reason": "Both are generic LoRA-loading reports, but the titles are too broad to show the same concrete defect; no shared target or file evidence." + "reason": "Both are LoRA loading errors, but the reports differ in specifics and likely cause; no evidence they are the same underlying bug." }, { "left": "issue:4542", "right": "issue:4724", "accept": false, - "reason": "Both are SDXL img2img complaints, but one is about generating more than one image and the other is about unexpected behavior with a base checkpoint." + "reason": "Both concern SDXL img2img output behavior, but one is about producing multiple images and the other about checkpoint compatibility; not the same defect." }, { "left": "issue:11241", "right": "issue:9927", "accept": false, - "reason": "Different training failures: Flux ControlNet floating-point exception versus sdv3 controlnet header-too-large." + "reason": "Both are training-time failures around FLUX/ControlNet-like workflows, but one is a floating-point exception and the other is a header-too-large error; distinct bugs." }, { "left": "issue:1310", "right": "issue:1962", "accept": false, - "reason": "Unrelated requests: k-diffusion pipeline integration versus video generation." + "reason": "These are different high-level pipeline questions: K-Diffusion integration vs video generation. They do not look like duplicate bug reports." }, { "left": "issue:1392", "right": "issue:1962", "accept": false, - "reason": "Stable Diffusion 2 model discussion versus video generation; not the same bug or change." + "reason": "Stable Diffusion 2 adoption vs video generation; same broad ecosystem, but no concrete shared bug or change." }, { "left": "issue:10533", "right": "issue:11922", "accept": false, - "reason": "Both involve FLUX.fill, but one is black-image output in fp16 and the other is a general running problem; too little evidence they share a single defect." + "reason": "Both reference FLUX and fill-related problems, but one is a generated black image in flux fill fp16 and the other is a general running problem for FLUX.1-Fill-dev; too vague to treat as the same bug." }, { "left": "issue:1214", "right": "issue:817", "accept": false, - "reason": "Precision-related failures, but one is CLIPTextModel half-precision behavior and the other is dreambooth mixed precision; not the same code-path problem." + "reason": "Both are mixed-precision/half-precision complaints, but they target different components (CLIPTextModel vs dreambooth example) and different symptoms." }, { "left": "issue:1704", "right": "issue:3824", "accept": false, - "reason": "A generic attention-block question and an inpaint legacy/new pipeline question are unrelated." + "reason": "General questions about attention/inpaint pipelines, but they are unrelated topics and not duplicate bug reports." }, { "left": "issue:3019", "right": "issue:938", "accept": false, - "reason": "Completely different domains: text-guided image editing with SAM versus text-to-motion generation." + "reason": "A SAM text-guided editing request and a motion diffusion model request are completely different features." }, { "left": "issue:4306", "right": "issue:8649", "accept": false, - "reason": "Both are pipeline runtime errors, but on different models and different failure conditions; not enough to treat as one bug." + "reason": "Different issues in different pipelines: reference pipeline list-index error vs SD3 num_images_per_prompt regression." }, { "left": "issue:4888", "right": "issue:7782", "accept": false, - "reason": "Offline safetensors loading and a safetensors conversion unpickling error are unrelated." + "reason": "Both mention safetensors/offline conversion, but one is about loading model safetensors offline and the other is a pickle/unpickling conversion failure." }, { "left": "issue:8874", "right": "issue:9012", "accept": false, - "reason": "Both are weight-loading failures, but they affect different model families and different APIs; no evidence of the same concrete bug." + "reason": "Both are load-weight errors for SD/SD3 models, but one is a single-file CKPT unpickling failure and the other is an SD3ControlNetModel.from_transformer load issue; not the same code path." }, { "left": "issue:10511", "right": "issue:9825", "accept": false, - "reason": "A scheduler redesign request and an IPAdapter support request are separate feature work items." + "reason": "#10511 is a scheduler redesign request, while #9825 is an IPAdapter support request for FLUX pipelines; same ecosystem but different changes." }, { "left": "issue:2988", "right": "issue:6069", "accept": false, - "reason": "Both are scheduler bugs, but they fail in different methods under different conditions; not the same underlying defect." + "reason": "Both concern schedulers, but one is a training/inference step crash and the other is a specific add_noise failure at t=0; not the same bug." }, { "left": "issue:9393", "right": "issue:9729", "accept": false, - "reason": "Both touch FLUX LoRA training/saving, but one is a deepspeed save-model issue and the other is an unexpected T5EncoderModel save type; too different to merge." + "reason": "Both are FLUX LoRA training/save issues, but one is about DeepSpeed failing to save a model and the other is an unexpected T5EncoderModel save-model type error; related area but not clearly duplicate." }, { "left": "issue:10180", "right": "issue:9876", "accept": false, - "reason": "Flux Control LoRA loading versus VRAM release after LoRA training are unrelated problems." + "reason": "#10180 is a Flux Control LoRA/multiple LoRA loading bug, while #9876 is a VRAM reclamation complaint after LoRA training; different problem domain." }, { "left": "issue:10512", "right": "issue:10526", "accept": false, - "reason": "Both involve optimum.quanto and Flux, but one is a LoRA loading failure while the other is an FP8 WeightQBytesTensor constructor TypeError; similar area, different concrete bug." + "reason": "Both mention Flux/Quanto, but #10512 is a LoRA loading failure and #10526 is an optimum.quanto WeightQBytesTensor constructor TypeError in FP8; different underlying failure." }, { "left": "issue:13008", "right": "issue:13351", "accept": false, - "reason": "Different model families and requests: FLUX.2-Klein multi-image LoRA training versus Wan/Flux Fun ControlNet support." + "reason": "One is a future FLUX.2 multi-image LoRA training pipeline feature request, the other is a Wan/Flux controlnet request; not duplicates." }, { "left": "issue:3901", "right": "issue:6278", "accept": false, - "reason": "Missing requirements.txt is an environment/setup issue; the other is a multi-GPU training failure." + "reason": "Completely unrelated: missing requirements.txt vs multi-GPU training issue." }, { "left": "issue:6196", "right": "issue:8233", "accept": false, - "reason": "Both mention depth pipelines, but one is a ControlNetXS SDXL loading error and the other is an fp16 runtime error; not the same bug." + "reason": "Both are depth/fp16-related errors, but one is ControlNetXS Depth SDXL loading and the other is a depth pipeline fp16 error; not enough evidence of the same bug." } ] }, @@ -28787,10 +28805,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -28805,162 +28823,165 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very heterogeneous: it mixes unrelated training, loading, conversion, and inference bugs. None of the soft pairs look like the same underlying issue, so they should stay separate.", - "confidence": 0.47, - "canonical_issue_reason": "issue:10026 is the clearest representative of the narrow multi-GPU training/memory theme in this cluster and has the most concrete, specific bug statement.", + "summary": "The cluster is mostly false-positive similarity: most pairs share a broad subsystem but not the same concrete bug. Per the feedback, the SDXL VAE pair should also stay conservative; `issue:6238` and `issue:6250` are related, but the evidence is only broad VAE/load-path overlap, not enough to treat them as the same underlying defect.", + "confidence": 0.84, + "canonical_issue_reason": "`issue:6250` is the best canonical issue because it is the more specific, actionable report: it names the exact failure (`KeyError`) and the concrete path (`loading SDXL vae from single file`).", "canonical_pr_reason": null, - "best_issue_reason": "issue:10026 is the best single issue to anchor the cluster because it is specific, actionable, and closest to the only recurring theme here: training efficiency/memory behavior.", + "best_issue_reason": "`issue:6250` is the strongest representative for the cluster because it is symptom-specific and tied to a reproducible loading path, while `issue:6238` is broader and less diagnostic.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:12011", "right": "issue:12012", "accept": false, - "reason": "Both are Wan 2.2-related, but one is OOM and the other is a context-state ValueError from apply_first_block_cache; different failure modes." + "reason": "Different failures: WAN i2v OOM vs a context/state `ValueError` from `apply_first_block_cache`." }, { "left": "issue:8511", "right": "issue:8939", "accept": false, - "reason": "LoRA/text-encoder unloading versus tokenizer load failure are unrelated bugs." + "reason": "Different bugs in different areas: SD3 training-script unload behavior vs tokenizer loading for a model repo." }, { "left": "issue:7142", "right": "issue:7785", "accept": false, - "reason": "Inference mismatch with webui versus very slow first inference are different symptoms and likely different causes." + "reason": "Both mention inference, but one is a result mismatch and the other is a first-inference slowdown regression." }, { "left": "issue:5840", "right": "issue:9664", "accept": false, - "reason": "Checkpoint resume/freezing behavior is not the same as a subprocess failure running the training script." + "reason": "Unrelated symptoms: resume/checkpoint freezing issue vs a subprocess failure in a LoRA training script." }, { "left": "issue:3788", "right": "issue:6407", "accept": false, - "reason": "Different new models/pipelines; no shared concrete bug." + "reason": "Both are new pipeline requests, but they are different projects/models and not the same change." }, { "left": "issue:5465", "right": "issue:962", "accept": false, - "reason": "DiffEdit inpaint_strength handling is unrelated to inpaint size support." + "reason": "Both are inpaint-related, but one is about `inpaint_strength` and the other is about supported image size." }, { "left": "issue:10026", "right": "issue:1734", "accept": false, - "reason": "Both mention multi-GPU training, but one is about memory not reducing and the other about performance slowdown; not the same bug." + "reason": "Same broad training area, but one is about memory reduction under ZeRO-3 and the other is about multi-GPU slowness." }, { "left": "issue:10969", "right": "issue:9799", "accept": false, - "reason": "Zero3 weight-dimension failure and tensor-size mismatch are distinct runtime errors." + "reason": "Different runtime errors with no clear shared code path: 2-D weight failure vs tensor-size mismatch." }, { "left": "issue:1672", "right": "issue:4000", "accept": false, - "reason": "Both concern conversion, but wrong output size versus generic ckpt-to-diffusers conversion are not the same defect." + "reason": "Conversion-related, but one reports wrong output size while the other is a general conversion question." }, { "left": "issue:6238", "right": "issue:6250", "accept": false, - "reason": "General SDXL VAE issue versus a specific KeyError loading VAE from single file; too broad to merge confidently." + "reason": "Same SDXL VAE area, but the evidence is only broad loading similarity; `6250` is a specific single-file `KeyError`, which is not enough to assert the same underlying bug as `6238`." }, { "left": "issue:4736", "right": "issue:7310", "accept": false, - "reason": "OOM/invalid cast in an SDXL LoRA script versus defaults not working are different problems." + "reason": "Same training-script family, but OOM/invalid-cast symptoms are not the same bug as defaults not working." }, { "left": "issue:3250", "right": "issue:6323", "accept": false, - "reason": "Pipeline launch failure and TensorRT attribute-setting error are unrelated." + "reason": "Different pipeline areas and failures: K-diffusion pipeline startup vs TensorRT img2img attribute handling." }, { "left": "issue:8588", "right": "issue:8726", "accept": false, - "reason": "Missing conversion script and text encoder errors in SD3 training scripts are different issues." + "reason": "Missing conversion script vs text-encoder errors in SD3 dreambooth scripts; not the same defect." }, { "left": "issue:4086", "right": "issue:5465", "accept": false, - "reason": "ControlNet training failure and DiffEdit inpaint parameter limitation are not the same code-path problem." + "reason": "Different control issues: SDXL ControlNet training failure vs a DiffEdit `inpaint_strength` limitation." }, { "left": "issue:2844", "right": "issue:4303", "accept": false, - "reason": "StableUnCLIP image-embedding input handling and Flax SDXL loading crash are unrelated." + "reason": "Completely different pipeline areas and errors: explicit image embeddings vs SDXL load crash in Flax pipeline." }, { "left": "issue:1774", "right": "issue:350", "accept": false, - "reason": "Both discuss memory, but one is a generic OOM report and the other is a question about expected UNet memory usage." + "reason": "One is a specific GPU OOM report; the other is a generic memory-usage question about UNet." }, { "left": "issue:3379", "right": "issue:746", "accept": false, - "reason": "Dreambooth checkpoint loading and ONNX revision loading are different subsystems and errors." + "reason": "Dreambooth checkpoint loading vs ONNX revision loading are unrelated." }, { "left": "issue:11148", "right": "issue:12689", "accept": false, - "reason": "Adapter loaded flag behavior and checkpoint weight loading are related to LoRA, but they are different concrete bugs." + "reason": "Both are PEFT/LoRA-adjacent, but one is a loaded-flag bug and the other is checkpoint weights not being loaded." }, { "left": "issue:5436", "right": "issue:6540", "accept": false, - "reason": "Both are ControlNet training errors, but one is an example script issue and the other is an FSDP matmul shape error." + "reason": "Both are ControlNet training issues, but the concrete failures and likely code paths differ." }, { "left": "issue:11853", "right": "issue:9187", "accept": false, - "reason": "Both involve IP-Adapter/SDXL, but one is a config conflict when loading together and the other is an update-specific image encoder_folder problem." + "reason": "Both involve IP-Adapter/SDXL, but one is a conflict-size issue and the other is an `image_encoder_folder=None` loading regression." }, { "left": "issue:5575", "right": "issue:6661", "accept": false, - "reason": "LoRA hidden-size configuration is unrelated to an MPS OOM report." + "reason": "Unrelated topics: LoRA layer sizing question vs an MPS OOM report." }, { "left": "issue:4354", "right": "issue:5150", "accept": false, - "reason": "Both mention from_single_file, but one is wrong return type and the other is missing method support." + "reason": "Both are `from_single_file`-related, but they affect different pipelines and different failure modes." }, { "left": "issue:4061", "right": "issue:5693", "accept": false, - "reason": "LoRA load error and VAE load error are different model-loading failures." + "reason": "Both are load errors, but one is LoRA loading in SDXL and the other is VAE loading." }, { "left": "issue:1643", "right": "issue:7629", "accept": false, - "reason": "xformers API mismatch and PyTorch-version-dependent OOM are unrelated." + "reason": "No shared underlying bug: xformers argument mismatch vs version-specific OOM in SVD." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "The report is mostly grounded, but it overstates one soft edge: `issue:6238` vs `issue:6250` is marked accepted even though the packet\u2019s own `deterministic_accept` is false and the available evidence is only broad SDXL/VAE similarity, not enough to assert the same underlying bug. The summary should stay more conservative here." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -29056,10 +29077,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -29074,156 +29095,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose collection of mostly unrelated Diffusers issues, with only superficial overlap around model loading, quantization, or pipeline import errors. None of the soft pairs look like the same underlying bug/change closely enough to merge.", - "confidence": 0.56, - "canonical_issue_reason": "Issue #12599 is the strongest representative of a concrete, user-facing runtime problem and has the clearest discussion signal among the set, but the cluster is too heterogeneous to treat it as a true duplicate root issue.", + "summary": "All proposed soft pairs are topical overlaps only; none appear to describe the same underlying bug or mergeable PR change. The set spans distinct models, pipelines, and failure modes.", + "confidence": 0.92, + "canonical_issue_reason": "No safe canonical issue stands out: the items are mostly unrelated bug reports/features across different pipelines and models, so collapsing them would be too aggressive.", "canonical_pr_reason": null, - "best_issue_reason": "#12599 is the best issue candidate because it is specific, actionable, and describes a well-scoped runtime regression around moving a quantized pipeline component to CUDA.", + "best_issue_reason": "issue:10512 is the strongest standalone issue candidate: it has the most discussion, a concrete loading failure, and enough detail to be actionable even though it is not a duplicate of the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1962", "right": "issue:6460", "accept": false, - "reason": "Both involve pipelines, but one is generic video generation and the other is an endless-loading model load bug; different symptoms and code paths." + "reason": "Both involve video/inpainting loading behavior, but one is about video generation support and the other about endless model loading for a specific checkpoint; different bugs." }, { "left": "issue:6513", "right": "issue:8482", "accept": false, - "reason": "Unrelated model support requests: one is T2IAdapter model coverage, the other is Stable Diffusion 3 support." + "reason": "Different features and models: T2IAdapter support vs Stable Diffusion 3 model support." }, { "left": "issue:4410", "right": "issue:7770", "accept": false, - "reason": "Both mention shape mismatches during loading, but they concern different models and different loading scenarios; too broad to be the same bug." + "reason": "Both mention shape mismatches, but the affected code paths and contexts differ: direct UNet loading vs model+LoRA loading in Python." }, { "left": "issue:10512", "right": "issue:9270", "accept": false, - "reason": "Both mention Flux LoRA, but one is a load failure for Quanto Flux LoRA and the other is incomplete training/support in SimpleTuner; not the same concrete issue." + "reason": "Both concern LoRA, but one is a load failure and the other is incomplete support for a specific Flux training setup; not the same bug." }, { "left": "issue:11824", "right": "issue:9106", "accept": false, - "reason": "Different pipelines and different failures: multi-image FluxKontext input error versus PixArt time_step attribute error." + "reason": "Different pipelines and failure modes: FluxKontext multi-image input error vs PixArt-Alpha time_step=1 attribute error." }, { "left": "issue:1005", "right": "issue:3250", "accept": false, - "reason": "One is a timestep constraint question, the other is a StableDiffusionKDiffusionPipeline runtime failure; no concrete overlap." + "reason": "Unrelated pipeline failures; one is a timestep bound question, the other is a KDiffusion pipeline run failure." }, { "left": "issue:5927", "right": "issue:8790", "accept": false, - "reason": "Both mention IP-Adapter, but the affected setup and failure mode are different; one is a complex combo pipeline, the other is SDXL TensorRT integration." + "reason": "Both mention IP-Adapter, but one is a composite inpaint/reference setup and the other is SDXL TensorRT integration; not the same concrete problem." }, { "left": "issue:8637", "right": "issue:9167", "accept": false, - "reason": "Both are import issues for different pipeline classes; the missing symbols and fixes are not the same." + "reason": "Different missing imports for different pipeline classes." }, { "left": "issue:2596", "right": "issue:6636", "accept": false, - "reason": "Both touch diffusers.utils, but one is docs referencing a nonexistent helper and the other is a missing import symbol; separate problems." + "reason": "Documentation typo/missing symbol vs runtime import error for a different utility constant." }, { "left": "issue:12147", "right": "issue:12529", "accept": false, - "reason": "Same Wan2.2 family, but one is LoRA loading and the other is a tiled VAE tensor mismatch; different components and failures." + "reason": "Both are Wan2.2-related, but one is LoRA loading and the other is tiled VAE tensor mismatch; distinct subsystems and bugs." }, { "left": "issue:8939", "right": "issue:9320", "accept": false, - "reason": "Tokenizer loading for SD3-medium is unrelated to custom diffusion save/load behavior." + "reason": "Tokenizer/config loading for SD3 vs saving/loading finetuned custom diffusion models; different failure points." }, { "left": "issue:8495", "right": "issue:8556", "accept": false, - "reason": "Both are SD3-related load errors, but one is tokenizer JSON parsing and the other is a meta-tensor copy failure; distinct root causes." + "reason": "Both are SD3 loading issues, but one is a tokenizer enum parse error and the other is a meta-tensor copy failure from single-file loading." }, { "left": "issue:12599", "right": "issue:9953", "accept": false, - "reason": "Close in theme, but #12599 is about slow CUDA transfer after loading while #9953 is about an actual error when moving a quantized component; not the same bug." + "reason": "Both concern moving pipelines to CUDA, but one is slowness and the other is an error with quantized components." }, { "left": "issue:378", "right": "issue:4754", "accept": false, - "reason": "Both relate to textual inversion, but one is a DiagonalGaussianDistribution issue and the other is a missing model_index.json after loading; different bugs." + "reason": "Textual inversion appears in both, but one is a model/component class issue and the other is a missing model_index.json when loading a saved artifact." }, { "left": "issue:6283", "right": "issue:7306", "accept": false, - "reason": "Different pipeline behaviors: strength=0 handling in ControlNet img2img versus float16 support in LEdits++." + "reason": "Different SDXL bugs: strength=0 handling in ControlNet img2img vs float16 incompatibility in LEdits++." }, { "left": "issue:1060", "right": "issue:671", "accept": false, - "reason": "Both reference the original SD conversion script, but one is a safety-checker load failure and the other is conversion fidelity not matching the source model." + "reason": "One is a converter safety-checker load issue, the other is model non-identity after conversion; not the same bug." }, { "left": "issue:11584", "right": "issue:11864", "accept": false, - "reason": "Both mention torch.compile, but the failures are in different code paths: group offloading on HunYuanVideo versus a missing torch name in AutoencoderDC.encode." + "reason": "Both involve torch.compile, but one is a group-offloading incompatibility and the other is a fullgraph compile NameError in a different module." }, { "left": "issue:7935", "right": "issue:7936", "accept": false, - "reason": "Both are model loading problems, but one is a missing config on a repo and the other is invalid JSON for a local safetensors file." + "reason": "One is missing config.json for a repo, the other is invalid JSON when loading a local safetensors file." }, { "left": "issue:11839", "right": "issue:12085", "accept": false, - "reason": "Both are gguf loading errors, but they affect different models/pipelines and likely different input validation paths." + "reason": "Both are gguf loading problems, but for different models and likely different input-shape expectations." }, { "left": "issue:5256", "right": "issue:5605", "accept": false, - "reason": "These are feature requests in different training workflows, not the same change." + "reason": "Completely different feature requests: Insta-Flow training support vs per-image caption support for SDXL DreamBooth." }, { "left": "issue:12126", "right": "issue:13258", "accept": false, - "reason": "Different Wan2.2 problems: cache_context switching error versus a missing image-to-video model/pipeline availability issue." + "reason": "Different Wan2.2 reports: a cache_context NoneType inference error vs a missing image-to-video model endpoint." }, { "left": "issue:6832", "right": "issue:8827", "accept": false, - "reason": "Both involve IP-Adapter FaceID, but one is a tuple.shape runtime error in a community pipeline and the other is a broader usage error in SDXL; not enough evidence of the same fix." + "reason": "Both involve IP-Adapter FaceID, but the specific errors and pipelines differ; not enough evidence of the same bug." }, { "left": "issue:6636", "right": "issue:9410", "accept": false, - "reason": "Import errors for different pipeline symbols; unrelated missing exports." + "reason": "Different import failures for different pipeline symbols." }, { "left": "issue:6107", "right": "issue:9187", "accept": false, - "reason": "Both mention IP-Adapter, but one is an environment-specific app issue and the other is an SDXL/Kolors regression with image_encoder_folder=None; different underlying bugs." + "reason": "One is environment-specific IP-Adapter behavior on AWS, the other is an SDXL regression after a Kolors update with image_encoder_folder=None." } ] }, @@ -29326,10 +29347,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -29344,165 +29365,165 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose semantic grouping of training/loading-example issues. The 2466/3802 pair is the strongest near-duplicate match by title and discussion, but the packet only supports same-topic / near-duplicate confidence, not a confirmed duplicate or identical fix path.", - "confidence": 0.83, - "canonical_issue_reason": "Issue 2466 is the best canonical anchor because it has the richest discussion around the checkpoints_total_limit training-example problem and is the closest match to 3802. The evidence supports near-duplicate/same-topic status, but not a fully confirmed duplicate relationship.", + "summary": "The set is mostly heterogeneous and not one duplicate cluster. Only one soft pair looks like a true duplicate: the training-example checkpoint-limit bug. All other pairs share broad subsystem or symptom wording but describe different concrete failures or requests.", + "confidence": 0.94, + "canonical_issue_reason": "issue:2466 is the best canonical issue because it states the checkpoint-limit training-example bug most directly and cleanly; issue:3802 is the same underlying problem with slightly different wording.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 2466 is the best overall issue representative for this cluster because it is the clearest and most discussed instance of the checkpoints_total_limit bug, making it the strongest anchor even though the duplicate/fix-path certainty should stay modest.", + "best_issue_reason": "issue:2466 is the strongest representative in this set: it is a clear, specific bug report with the clearest duplicate relationship to another item, while most other issues are unrelated or only superficially similar.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10472", "right": "issue:9613", "accept": false, - "reason": "Both are runtime errors, but they hit different code paths: LoRA config handling vs timestep scaling in a custom model." + "reason": "Different failures: unexpected lora_bias argument during LoRA inference vs missing Timesteps.scale in 0.30.x. Same broad area, not the same bug." }, { "left": "issue:2784", "right": "issue:5576", "accept": false, - "reason": "One is a feature request for Dreambooth text2video support; the other requests a new scheduler. Same area, different change." + "reason": "Both are feature requests, but they target different changes: Dreambooth for text2video models vs a new FreeNoise scheduler." }, { "left": "issue:4801", "right": "issue:5425", "accept": false, - "reason": "Both involve training examples, but one is a mixed-precision bug and the other is checkpoint saving compatibility." + "reason": "Both mention training examples and checkpoints, but one is mixed-precision behavior in train_unconditional.py and the other is accelerate/PT 2.1 checkpoint-saving compatibility." }, { "left": "issue:2459", "right": "issue:3700", "accept": false, - "reason": "Both are training-script flags, but xformers attention and min_snr are separate behaviors." + "reason": "Different training-script issues: xformers attention not working in LoRA training vs min_snr argument not usable. Similar script family, not the same bug." }, { "left": "issue:4446", "right": "issue:6839", "accept": false, - "reason": "Both touch from_single_file, but one is custom text_encoder loading and the other is StableVideoDiffusionPipeline support." + "reason": "Both involve from_single_file-style loading, but one is custom text_encoder support for StableDiffusionPipeline and the other is StableVideoDiffusionPipeline single-file loading." }, { "left": "issue:1398", "right": "issue:4134", "accept": false, - "reason": "Both relate to MPS/controlnet, but the failures are different: SD2 on Apple Silicon versus a controlnet tensor-size mismatch." + "reason": "Both are Apple/MPS-related, but the concrete problems differ: SD2 on MPS vs ControlNet tensor-size mismatch on M1 Max." }, { "left": "issue:4031", "right": "issue:5128", "accept": false, - "reason": "Both involve local loading, but safety checker loading and safetensors/from_single_file are distinct bugs." + "reason": "Both concern loading from local files, but one is standalone safety checker loading and the other is XLPipeline local safetensors loading; different code paths." }, { "left": "issue:2466", "right": "issue:3802", "accept": true, - "reason": "Strong near-duplicate reports of checkpoints_total_limit not limiting checkpoints in training examples; same topic and likely same fix area, though the packet stops short of confirming an exact duplicate relationship." + "reason": "Same underlying bug: `checkpoints_total_limit` is not being enforced in training examples. The titles and descriptions point to the same checkpoint-rotation problem." }, { "left": "issue:10653", "right": "issue:12599", "accept": false, - "reason": "Both are device-transfer issues, but bitsandbytes 8bit .to() limitations and slow CUDA migration are different problems." + "reason": "Different problems around device movement/performance: bitsandbytes 8-bit module `.to()` CPU restriction vs slow CUDA move after loading." }, { "left": "issue:6382", "right": "issue:7974", "accept": false, - "reason": "Both mention IP-adapter, but one is a training-script key error and the other is an inpaint usage bug." + "reason": "Both mention IP Adapter, but one is an IP-adapter key error in LCM LoRA SDXL training/use and the other is inpaint with IP Adapter face+style." }, { "left": "issue:5607", "right": "issue:7002", "accept": false, - "reason": "Both are runtime shape/buffer failures, but the underlying errors are different." + "reason": "Both are runtime shape/buffer errors, but the reported failures are different: tensor concat mismatch vs invalid buffer size in img2img decoding." }, { "left": "issue:3487", "right": "issue:5558", "accept": false, - "reason": "Both are image-quality complaints, but one is scheduler-specific strength behavior and the other is artifacts with multiple ControlNets." + "reason": "Both involve inpaint output quality, but one is a scheduler/strength quality issue and the other is artifacts with multiple ControlNets." }, { "left": "issue:3101", "right": "issue:6858", "accept": false, - "reason": "Both involve training examples, but the controlNet AttributeError and the missing convert() method are unrelated." + "reason": "Both are controlnet-training related, but the actual errors differ: a training example yield error vs `'str' object has no attribute 'convert'`." }, { "left": "issue:12107", "right": "issue:12157", "accept": false, - "reason": "Both are training-related errors, but init_trackers with custom objects and the torch.device iterable error are distinct." + "reason": "Different accelerator/training errors: custom object list passed to `init_trackers` vs `'torch.device' object is not iterable` with bnb quantization config." }, { "left": "issue:3643", "right": "issue:4631", "accept": false, - "reason": "Both are runtime issues in generation, but they affect different code paths and failure modes." + "reason": "Both are runtime exceptions, but one is an INT_MAX element limit while decoding latents and the other is a scalar conversion failure in DPMSolverSinglestepScheduler." }, { "left": "issue:7133", "right": "issue:7197", "accept": false, - "reason": "Both concern resume/checkpoint behavior, but they describe separate bugs." + "reason": "Both involve resuming training, but one is SDXL Dreambooth LoRA text-encoder size mismatch and the other is a KeyError for 'Shortest Edge' in a different script." }, { "left": "issue:10823", "right": "issue:11232", "accept": false, - "reason": "Both are fine-tuning issues, but missing inference files and noise residuals are not the same defect." + "reason": "Both relate to fine-tuning workflows, but one is missing inference artifacts after training and the other is noise residuals during text-to-image fine-tuning." }, { "left": "issue:5126", "right": "issue:5932", "accept": false, - "reason": "Both mention tensor/channel problems, but the Conv2d argument error and 3-vs-4-channel mismatch are different bugs." + "reason": "Different shape-related errors: a Conv2d forward signature mismatch vs a 4-channel input expectation being violated." }, { "left": "issue:10010", "right": "issue:9454", "accept": false, - "reason": "Both are model-not-found style failures, but they reference different models and artifacts." + "reason": "Both mention a missing model/file, but they are for different models and failure points: FLUX Fill example model_index.json vs CogVideoX Image2Video model not found." }, { "left": "issue:10100", "right": "issue:10639", "accept": false, - "reason": "Both are support requests for new models, but they are for different integrations." + "reason": "Both are feature requests for model support, but they concern different models: ConsisID vs Lumina-Image 2.0." }, { "left": "issue:1013", "right": "issue:4109", "accept": false, - "reason": "Both say a pipeline is not working, but they involve different pipeline families and failure contexts." + "reason": "Both are pipeline-not-working issues, but they are different pipelines and likely different root causes: StableDiffusionMegaPipeline vs StableDiffusionXLInpaintPipeline." }, { "left": "issue:10139", "right": "issue:9576", "accept": false, - "reason": "Both involve pipeline initialization/loading, but sample_size handling and use_safetensors propagation are separate bugs." + "reason": "Different pipeline-loading bugs: `sample_size` handling in pipeline init vs `use_safetensors` not being passed to submodules." }, { "left": "issue:10447", "right": "issue:9719", "accept": false, - "reason": "Different areas entirely: conversion script file-splitting versus ignored disable_progress_bar during loading." + "reason": "Different topics entirely: converter output splitting for large UNet models vs `disable_progress_bar` being ignored during checkpoint loading." }, { "left": "issue:1061", "right": "issue:1823", "accept": false, - "reason": "One is documentation text for an example; the other is a code comment swap." + "reason": "One is documentation/instructions for textual inversion, the other is swapped comments in depth2img. Same area, not the same bug." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The cluster-level conclusion is mostly grounded, but the writeup overstates certainty. The packet supports a strong near-duplicate/same-topic relationship between 2466 and 3802 from their titles and discussion, but it does not confirm a duplicate relationship or a shared fix path. Please soften claims like \"confirmed duplicate target\" and \"same bug and same fix path\" to match the available evidence. The individual soft-edge rejects look conservative; the main issue is the canonical/best-issue certainty." + "accept": true, + "feedback": "The summary is grounded in the packet: the titles support that most pairs are only loosely related, and the 2466/3802 pair is the one clear same-bug match among the listed soft edges. The verdicts are conservative overall, with no obvious overreach from the available evidence." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -29600,10 +29621,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -29618,163 +29639,172 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issues with a loose concentration around Flux/ControlNet/SD3 training, memory, and device/dtype problems. The proposed soft pairs are thematically similar in a few cases, but none look like true duplicates or mergeable PR-equivalents.", - "confidence": 0.44, - "canonical_issue_reason": "issue:10026 is the clearest and most specific report of the recurring Flux multi-GPU memory problem, so it is the best representative issue for the main theme in this set.", + "summary": "Mostly a loose similarity cluster of unrelated diffusers issues across Flux, SDXL, Wan, LoRA, and pipeline-feature requests. Only one pair looks like the same concrete bug family: SDXL `from_single_file` loading errors.", + "confidence": 0.83, + "canonical_issue_reason": "issue:7389 is the clearest representative: it has a concrete SDXL `from_single_file` failure, and it is tightly paired with a near-identical sibling issue on a related SDXL pipeline.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10026 best captures the dominant subtheme (Flux training on multi-GPU with memory not improving under DeepSpeed stage-3) and is the most concrete anchor among the Flux-related reports.", + "best_issue_reason": "issue:7389 is the strongest issue to anchor the cluster because it describes a specific runtime failure in a shared loading path, making it the most useful representative if these are being collapsed.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1188", "right": "issue:2437", "accept": false, - "reason": "Both are pipeline feature requests, but CLIPSeg+SD and T2I-Adapter are different integrations and different underlying changes." + "reason": "Both are pipeline feature requests, but they ask for different pipelines and do not describe the same bug/change." }, { "left": "issue:1486", "right": "issue:2579", "accept": false, - "reason": "DDIMScheduler sample-prediction bug and xFormers misuse/bug are unrelated failure modes." + "reason": "Different scheduler bug vs possible xFormers misuse; no concrete shared failure mode." }, { "left": "issue:1486", "right": "issue:3851", "accept": false, - "reason": "Both involve DDIMScheduler, but one is prediction_type='sample' and the other is epsilon prediction with zero-SNR rescaling; different bugs." + "reason": "Both involve DDIMScheduler, but the prediction type and failure conditions differ." }, { "left": "issue:1949", "right": "issue:3346", "accept": false, - "reason": "Both touch dreambooth training examples, but one is a training error and the other is a resume_from_checkpoint feature request." + "reason": "Both are DreamBooth training issues, but one is tutorial errors and the other is checkpoint resume support." }, { "left": "issue:2673", "right": "issue:5215", "accept": false, - "reason": "Both mention StableDiffusionUpscalePipeline, but one is a config/input verification error and the other is a brightness regression." + "reason": "Both mention StableDiffusionUpscalePipeline, but the reported problems are unrelated." }, { "left": "issue:2800", "right": "issue:3728", "accept": false, - "reason": "A DiffEdit semantic editing issue and a shifted-diffusion pipeline proposal are not the same bug/change." + "reason": "Both are new pipeline requests, but for different methods/models." }, { "left": "issue:4037", "right": "issue:4206", "accept": false, - "reason": "Both concern ControlNet training, but FSDP validation errors and an SDXL main-branch breakage are different code paths." + "reason": "Both are ControlNet training issues, but one is FSDP validation errors and the other is an SDXL main-branch breakage." }, { "left": "issue:4975", "right": "issue:8532", "accept": false, - "reason": "Repeated dtype queries during LoRA loading vs progressive memory increase on LoRA loading are related performance complaints, but not the same underlying bug." + "reason": "Related to LoRA loading, but one is a performance issue and the other is memory growth." }, { "left": "issue:5337", "right": "issue:5457", "accept": false, - "reason": "These are documentation/explanation questions about different callbacks and attention masks, not duplicate issues." + "reason": "Different API documentation questions with no shared bug." }, { "left": "issue:6842", "right": "issue:9167", "accept": false, - "reason": "Import failures for different symbols: Prompt2PromptPipeline vs StableDiffusion3Pipeline." + "reason": "Both are import errors, but they concern different pipeline classes and likely different export gaps." }, { "left": "issue:7441", "right": "issue:8805", "accept": false, - "reason": "Pytest on Python 3.12 and a broken conversion script are unrelated." + "reason": "Completely different problems: Python 3.12 pytest failure vs an SDXL conversion script regression." }, { "left": "issue:8534", "right": "issue:8543", "accept": false, - "reason": "Both are SD3 dreambooth training issues, but wrong logit-normal weighting and gradient checkpointing failure are distinct bugs." + "reason": "Both are SD3 dreambooth training issues, but logit weighting and gradient checkpointing are distinct failures." }, { "left": "issue:8540", "right": "issue:8834", "accept": false, - "reason": "Caching embeddings/VAE representations is a training optimization issue; asking whether SD3 ControlNet code will be released is a feature/release question." + "reason": "Caching/optimization request vs a question about releasing ControlNet training code." }, { "left": "issue:9042", "right": "issue:9873", "accept": false, - "reason": "They are separate model-pipeline addition requests (Flux vs OmniGen), not duplicates." + "reason": "Both are feature requests to add pipelines, but for different models." }, { "left": "issue:9450", "right": "issue:9630", "accept": false, - "reason": "Both are Flux multi-GPU issues, but one is a device mismatch in FluxPipeline and the other is a num_single_layers=0 training/runtime error; not the same bug." + "reason": "Both are Flux multi-GPU issues, but one is a tensor device mismatch and the other is a num_single_layers configuration error." }, { "left": "issue:10671", "right": "issue:9527", "accept": false, - "reason": "Both involve device/dtype handling, but deterministic dtype/device queries and a ControlNet fp32/bf16 mismatch have different roots and fixes." + "reason": "Both touch dtype/device handling, but one is deterministic accessor behavior and the other is a mixed-precision controlnet runtime error." }, { "left": "issue:9309", "right": "issue:9546", "accept": false, - "reason": "FluxControlNetPipeline on Flux 1 schnell and validation-step OOM in Flux ControlNet training are different problems." + "reason": "Both are Flux ControlNet training/runtime issues, but one is a schnell pipeline bug and the other is a validation-step OOM." }, { "left": "issue:11627", "right": "issue:8817", "accept": false, - "reason": "StableDiffusion3Pipeline with a specific IP-Adapter and an IP-Adapter FaceID issue are not the same concrete bug." + "reason": "Different model/pipeline topics: SD3.5 IP-adapter usage vs IP-Adapter FaceID trouble." }, { "left": "issue:10026", "right": "issue:9156", "accept": false, - "reason": "Both are Flux training memory complaints, but one is DeepSpeed stage-3 not reducing memory and the other is plain CUDA OOM during LoRA training." + "reason": "Both are memory-related Flux training issues, but one is DeepSpeed stage-3 memory scaling and the other is LoRA CUDA OOM." }, { "left": "issue:6159", "right": "issue:9109", "accept": false, - "reason": "An SDXL ControlNetXS keyword-argument support issue and a missing torchvision::nms operator are unrelated." + "reason": "Unrelated failures: an unsupported pipeline kwarg vs a missing torchvision operator." }, { "left": "issue:11006", "right": "issue:12565", "accept": false, - "reason": "Both are Wan2.1 i2v reports, but one is a quantized-transformer video corruption bug and the other is a general running issue." + "reason": "Both are Wan2.1 I2V issues, but the concrete symptoms differ and no shared code path is identified." + }, + { + "left": "issue:11627", + "right": "issue:13300", + "accept": false, + "reason": "Different generation setups and failures; no evidence they are the same bug." }, { "left": "issue:12494", "right": "issue:6879", "accept": false, - "reason": "A matmul shape mismatch in latest scripts and a dtype mismatch in Custom Diffusion are different tensor errors." + "reason": "Shape-mismatch matmul error vs dtype mismatch in custom diffusion; different underlying faults." }, { "left": "issue:7389", "right": "issue:7390", - "accept": false, - "reason": "Both are from_single_file errors, but one is StableDiffusionXL and the other is StableDiffusionXLControlNetInpaint; similar area, different pipelines." + "accept": true, + "reason": "Both report `from_single_file` errors on closely related SDXL pipelines, suggesting the same loader bug in a shared code path." }, { "left": "issue:5593", "right": "issue:7982", "accept": false, - "reason": "Both involve from_pipe and ControlNet, but one is ControlNet pipeline conversion and the other is MultiControlNetModel handling; not the same bug." + "reason": "Both involve `from_pipe` and ControlNet, but one is AutoPipeline conversion and the other is MultiControlNetModel support; too distinct to merge confidently." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays mostly conservative. The only strong match identified is the SDXL `from_single_file` pair, which is supported by the titles. The remaining pairs are rejected with cautious reasoning and no clear overclaiming. The summary accurately reflects that this is a loose similarity cluster rather than a coherent single-bug set." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -29871,10 +29901,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -29889,172 +29919,169 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are unrelated one-off issues spanning different models, pipelines, and training paths. Only the two PixArt DreamBooth reports look like the same underlying bug; the rest should stay separate.", - "confidence": 0.91, - "canonical_issue_reason": "issue:5845 is the clearest representative of the only true duplicate pair: it explicitly names DreamBooth training with PixArt Alpha and states the failure directly.", + "summary": "This is a heterogeneous set of loosely keyword-matched issues rather than a true duplicate cluster. Most links are only superficial similarity around LoRA, loading, ControlNet, training, or black-image symptoms; none look like the same concrete bug.", + "confidence": 0.96, + "canonical_issue_reason": "issue:6252 is the clearest concrete bug report in the set: a specific LoRA checkpoint-loading failure with actionable symptoms, making it the best representative of the loosely related loading/LoRA subset.", "canonical_pr_reason": null, - "best_issue_reason": "issue:5845 is the best global issue candidate because it is specific, readable, and matches the shared PixArt DreamBooth failure described by issue:5777.", + "best_issue_reason": "issue:6252 is the most usable issue to anchor triage because it is specific, user-facing, and reproducible, unlike the more generic feature requests or broad compatibility reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:6252", "right": "issue:7793", "accept": false, - "reason": "Both mention LoRA, but one is about weights not loading from a checkpoint and the other about unloading LoRA weights in DeepFloyd. Different operation and likely different code path." + "reason": "Both mention LoRA, but one is checkpoint loading and the other is unloading DeepFloyd LoRA weights; different code paths and failure modes." }, { "left": "issue:2507", "right": "issue:5128", "accept": false, - "reason": "One is a performance regression loading safetensors on GPU; the other is a hard error loading a local .safetensors file via from_single_file. Not the same bug." + "reason": "GPU safetensors load slowness is a performance issue, while the other is a missing import when loading a local .safetensors file." }, { "left": "issue:2988", "right": "issue:4376", "accept": false, - "reason": "Scheduler step-count crash and SDXL inpaint/refiner failure are unrelated subsystems and failure modes." + "reason": "Scheduler step-count crash and SDXL inpaint/refiner crash are unrelated bugs in different pipeline paths." }, { "left": "issue:3910", "right": "issue:4885", "accept": false, - "reason": "Both involve LoRA loading, but one is a general 'certain LoRA will not load' report and the other is specific to StableDiffusionXLControlNetImg2ImgPipeline. Not enough evidence they share one root cause." + "reason": "Generic LoRA load failure versus SDXL ControlNet img2img not loading LoRA weights; same broad area, but not the same concrete bug." }, { "left": "issue:3886", "right": "issue:5425", "accept": false, - "reason": "DDPM training timeout under train_unconditional.py is unrelated to checkpoint-saving compatibility with accelerate/PT 2.1." + "reason": "A distributed watchdog timeout during training is not the same as an accelerate/PT 2.1 checkpoint-saving compatibility issue." }, { "left": "issue:6492", "right": "issue:9239", "accept": false, - "reason": "Both mention black images, but one is generic image generation and the other is specifically ControlNet-Canny. Different contexts and likely different causes." + "reason": "Both involve black images, but one is generic image generation and the other is specifically ControlNet-Canny; too broad to merge." }, { "left": "issue:4591", "right": "issue:7421", "accept": false, - "reason": "Both are determinism/seed issues, but Karlo/unCLIP generator behavior is not the same as StableDiffusionXLKDiffusionPipeline seed handling." + "reason": "Deterministic generation problems in Karlo/unCLIP and seed handling in StableDiffusionXLKDiffusionPipeline are different model/code paths." }, { "left": "issue:4186", "right": "issue:964", "accept": false, - "reason": "Both are inpainting-related, but 4186 is SDXL inpaint with refiner while 964 is a generic inpainting example problem. Too broad to merge." + "reason": "Both are inpainting-related, but one is SDXL inpaint with refiner and the other is an inpainting example failure; not the same bug." }, { "left": "issue:3584", "right": "issue:6073", "accept": false, - "reason": "Both are DreamBooth tutorial problems, but they target different docs/models and do not clearly describe the same bug." + "reason": "These are different DreamBooth tutorial failures with different setups and error contexts, not one underlying defect." }, { "left": "issue:10472", "right": "issue:9731", "accept": false, - "reason": "One is a TypeError from lora_bias during inference; the other is a training how-to for Playground2.5. Different issue type and symptom." + "reason": "The first is a LoRA inference TypeError about lora_bias; the second is a training/how-to request for Playground2.5 style LoRA use." }, { "left": "issue:2695", "right": "issue:5972", "accept": false, - "reason": "ControlNet training validation failure is unrelated to the Min-SNR documentation request." + "reason": "ControlNet training validation crash and a documentation note about Min-SNR weighting are not the same issue." }, { "left": "issue:11962", "right": "issue:13351", "accept": false, - "reason": "Both request support for new models, but they are different model families and capabilities." + "reason": "GGUF support for FLUX.1-Kontext-dev and Wan/Flux.2 ControlNet support requests are different feature requests." }, { "left": "issue:1556", "right": "issue:2507", "accept": false, - "reason": "Black images on CUDA/NSFW warning is unrelated to safetensors loading performance." + "reason": "Black images on CUDA and safetensors GPU load performance are unrelated symptoms." }, { "left": "issue:4885", "right": "issue:9012", "accept": false, - "reason": "Both are ControlNet weight-loading issues, but they concern different model generations and APIs; no clear shared code-path bug." + "reason": "One is LoRA weights not loading in an SDXL ControlNet pipeline; the other is SD3ControlNetModel weight-loading failure; different models and failure points." }, { "left": "issue:5777", "right": "issue:5845", - "accept": true, - "reason": "These appear to be the same underlying PixArt DreamBooth training failure, just phrased differently; both point to DreamBooth on PixArt Alpha not working." + "accept": false, + "reason": "These are both DreamBooth/PixArt training issues, but they concern different model variants and error conditions." }, { "left": "issue:10760", "right": "issue:7082", "accept": false, - "reason": "Import errors, but for entirely different symbols and packages." + "reason": "An attribute missing on an ONNX inpaint pipeline and an import error for CpuOffload are distinct import/runtime problems." }, { "left": "issue:1814", "right": "issue:2614", "accept": false, - "reason": "Changing the CLIP model and changing the pretrained download directory are unrelated features." + "reason": "Changing the CLIP model and changing the pretrained-model download directory are unrelated configuration requests." }, { "left": "issue:2560", "right": "issue:941", "accept": false, - "reason": "Both involve StableDiffusionImg2ImgPipeline, but one is list-of-generators output inconsistency and the other is MPS num_images_per_prompt noise. Different bug reports." + "reason": "Both are determinism/reproducibility complaints, but they affect different pipelines and different trigger conditions." }, { "left": "issue:259", "right": "issue:2880", "accept": false, - "reason": "A progress-bar fix and a JAX Colab runtime failure are unrelated." + "reason": "A progress bar fix is unrelated to a JAX Colab runtime failure." }, { "left": "issue:4840", "right": "issue:5592", "accept": false, - "reason": "Both are checkpoint-loading problems, but they affect different models and serialization formats." + "reason": "Both are checkpoint-loading issues, but they involve different model classes and checkpoint formats, so they are not the same bug." }, { "left": "issue:6823", "right": "issue:9208", "accept": false, - "reason": "Generic 'cannot use controlnet' is too broad to be the same as loading most ControlNet models from_single_file()." + "reason": "General cannot-use-ControlNet versus from_single_file() failing on most ControlNet models are related only at subsystem level, not one concrete defect." }, { "left": "issue:3812", "right": "issue:7708", "accept": false, - "reason": "Both are missing-config style errors, but they refer to different files and model setups." + "reason": "Missing controlnet config values and missing tokenizer/config.json are different missing-file/config problems." }, { "left": "issue:2784", "right": "issue:2984", "accept": false, - "reason": "A feature request for DreamBooth text-to-video is not the same as a paper/announcement issue about latent video diffusion models." + "reason": "A DreamBooth feature request for text-to-video and a latent video diffusion model announcement/problem are not duplicates." }, { "left": "issue:5770", "right": "issue:5829", "accept": false, - "reason": "LCM dataset-access failure and missing unet_time_cond_proj_dim argument are different training-script problems." + "reason": "Dataset access blocking LCM distillation and a missing unet_time_cond_proj_dim argument are unrelated training issues." }, { "left": "issue:5399", "right": "issue:7239", "accept": false, - "reason": "Both are matrix-related runtime errors, but one is dtype mismatch in custom inpainting and the other is a shape mismatch in SDXL LoRA training." + "reason": "A dtype mismatch during custom inpainting inference is different from an SDXL LoRA shape-mismatch during DreamBooth training." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet: it stays close to the issue titles, and the only accepted soft pair (5777/5845) is a plausible PixArt DreamBooth duplicate. The remaining pair judgments are conservative and do not overclaim shared root causes." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -30152,10 +30179,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -30170,168 +30197,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very noisy: it mixes attention-masking/adapter issues with loading, dtype, training, scheduler, and distributed/runtime bugs. I\u2019d treat it as mostly non-duplicate items with only a loose thematic overlap around Diffusers attention/Flux/IP-Adapter behavior.", - "confidence": 0.31, - "canonical_issue_reason": "Issue #10194 is the broadest and most central item around attention masking, which is the only recurring theme that loosely connects a few of the nearby issues. It is still not a strong duplicate anchor because most other items are about unrelated failures.", + "summary": "This cluster is very loose and mostly heterogeneous. A few items orbit attention/masking and FLUX/IP-Adapter support, but most pairs are different bugs or feature requests in adjacent subsystems, so I would not merge any soft edges.", + "confidence": 0.41, + "canonical_issue_reason": "Issue 10194 is the best central representative because it most directly names the shared theme of attention masking in FLUX, which is the closest common thread among the few related items in this cluster.", "canonical_pr_reason": null, - "best_issue_reason": "#10194 is the best single issue to represent the cluster because it is the most general open request in the shared attention/masking area, but it should be used only as a weak umbrella choice, not as proof of duplication.", + "best_issue_reason": "10194 is the broadest, most on-theme issue for the small attention/masking subgroup; it is more representative than the more specific error reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3405", "right": "issue:4037", "accept": false, - "reason": "Both are ControlNet training problems, but one is a generic training bug and the other is specifically FSDP validation errors; different failure modes and likely different fixes." + "reason": "Both are ControlNet training problems, but one is generic and the other is specifically validation errors under FSDP; not the same bug." }, { "left": "issue:3221", "right": "issue:4724", "accept": false, - "reason": "One is about load_attn_procs in an older diffusers version, the other is an SDXL img2img checkpoint behavior issue; same broad area, not the same underlying bug." + "reason": "Both involve loading/pipeline behavior, but they refer to different APIs and failure modes (attn procs vs SDXL img2img checkpoint behavior)." }, { "left": "issue:835", "right": "issue:987", "accept": false, - "reason": "Both mention bad image output, but one is a flax pipeline numpy conversion issue and the other is a speech-to-image green image symptom; not clearly the same code path." + "reason": "Both mention output image defects, but one is a flax pipeline conversion issue and the other is a speech-to-image green output; not the same code path." }, { "left": "issue:1614", "right": "issue:987", "accept": false, - "reason": "Both are image artifact reports, but they involve different pipelines and symptoms (black vs green output), so they are not the same bug." + "reason": "Both are image-output corruption reports, but they are different models and symptoms (blank black vs green image)." }, { "left": "issue:8410", "right": "issue:9825", "accept": false, - "reason": "Both are feature requests around pipeline support, but TextDiffuser and Flux IP-Adapter support are unrelated changes." + "reason": "Both are FLUX-related feature requests, but one asks for native TextDiffuser pipeline support and the other for IP-Adapter support." }, { "left": "issue:4852", "right": "issue:5128", "accept": false, - "reason": "Both are loading-related, but one is LoRA loading and the other is single-file safetensors loading for XLPipeline; different paths and issues." + "reason": "Both are loading errors, but one is LoRA loading and the other is single-file XL pipeline loading; too different to be duplicates." }, { "left": "issue:6250", "right": "issue:9856", "accept": false, - "reason": "One is an SDXL VAE single-file KeyError, the other is a distributed port-in-use error; unrelated runtime problems." + "reason": "KeyError loading an SDXL VAE has nothing to do with a distributed port collision." }, { "left": "issue:10382", "right": "issue:4210", "accept": false, - "reason": "One is a VAE encoder return-type mismatch, the other is a question about attention_mask in UNet2DConditionModel; not duplicates." + "reason": "VAE encoder return-type mismatch is unrelated to UNet attention_mask usage." }, { "left": "issue:11436", "right": "issue:9096", "accept": false, - "reason": "Both involve Flux dtype/numerics, but one is a half vs bfloat16 matmul mismatch and the other is NaNs in half precision; related area, different concrete bug." + "reason": "Both are precision/dtype related, but one is a matmul dtype mismatch in HiDream and the other is NaNs in Flux half precision; not the same defect." }, { "left": "issue:2447", "right": "issue:5897", "accept": false, - "reason": "Both are LoRA-related errors, but one is generic LoRA failure and the other is a device placement bug in a specific training script." + "reason": "Both mention LoRA errors, but one is a broad report and the other is a specific CPU/GPU device mismatch during training." }, { "left": "issue:5829", "right": "issue:9370", "accept": false, - "reason": "Both mention missing arguments in embedding-related code, but one is an LCM training-script crash and the other is a separate model forward signature issue." + "reason": "Both involve missing arguments in model/embedding code, but they are different missing parameters in different components." }, { "left": "issue:10453", "right": "issue:11408", "accept": false, - "reason": "Batch-size limitation in Hunyuan Video and compiled-model set_adapters support are unrelated problems." + "reason": "Hunyuan batch-size limitations and compiled-model set_adapters support are unrelated issues." }, { "left": "issue:3094", "right": "issue:6371", "accept": false, - "reason": "One is CUDA OOM during finetuning; the other is an assertion failure in a controlnet_xs example. Same area, different issue." + "reason": "OOM during text-to-image finetuning is unrelated to an assert error in a controlnet_xs example." }, { "left": "issue:2822", "right": "issue:5881", "accept": false, - "reason": "Training convergence complaints about ControlNet and AnimateDiff quality issues are not the same bug or change." + "reason": "Training convergence quality complaints for ControlNet and Animatediff are not the same underlying bug." }, { "left": "issue:5254", "right": "issue:5465", "accept": false, - "reason": "Both concern pipeline features, but multi-ControlNet support and DiffEdit inpaint_strength behavior are distinct feature requests." + "reason": "Multi-ControlNet support and DiffEdit inpaint_strength limitations are different feature gaps." }, { "left": "issue:10989", "right": "issue:8464", "accept": false, - "reason": "One is a Flux ControlNet LoRA loading problem under quantization, the other is a SEGA cpu-offload seed failure; unrelated failures." + "reason": "Flux ControlNet LoRA loading under quantization and SEGA cpu-offload seed failure are unrelated." }, { "left": "issue:4210", "right": "issue:8645", "accept": false, - "reason": "An attention_mask usage question in UNet2DConditionModel and a missing helper function in single_file_utils are unrelated." + "reason": "A question about UNet attention_mask is not the same as a missing helper function in single_file_utils.py." }, { "left": "issue:4874", "right": "issue:5128", "accept": false, - "reason": "Both mention loading/local usage, but one is a scheduler bug and the other is a single-file load error; different concrete problems." + "reason": "Both are local loading/scheduler related, but one is an upscale pipeline scheduler bug and the other is single-file safetensors loading." }, { "left": "issue:2695", "right": "issue:5997", "accept": false, - "reason": "A ControlNet training validation/reporting bug and an SDXL tutorial type mismatch are not the same underlying issue." + "reason": "ControlNet training validation/TensorBoard behavior is unrelated to the SDXL tutorial model-type mismatch." }, { "left": "issue:3221", "right": "issue:4086", "accept": false, - "reason": "Loading attention processors and SDXL ControlNet not working are too broad and distinct to merge as duplicates." + "reason": "Attn-proc loading in one pipeline is not the same as ControlNet training failing with SDXL base 0.9." }, { "left": "issue:3296", "right": "issue:4389", "accept": false, - "reason": "Dreambooth checkpoint save/load behavior and ControlNetModel load errors are separate code paths." + "reason": "Checkpoint save/load with text_encoder and ControlNetModel loading errors are different problems." }, { "left": "issue:1231", "right": "issue:1363", "accept": false, - "reason": "Both mention LMS schedulers, but Flax scheduler support and an ONNX long-prompt pipeline error are different contexts and failures." + "reason": "Scheduler support in Flax and an ONNX long-prompt/LMS error are related only by scheduler name, not the underlying bug." }, { "left": "issue:5886", "right": "issue:7238", "accept": false, - "reason": "One is a request to support an IP-Adapter face model, the other is IP-Adapter masking; same subsystem, different change." + "reason": "IP-Adapter face model support and IP-Adapter masking are distinct feature requests." }, { "left": "issue:10066", "right": "issue:11432", "accept": false, - "reason": "A self.attn unpacking ValueError and a torch_dtype='auto' loading issue are unrelated." + "reason": "Attention unpacking failure and from_pretrained torch_dtype='auto' issues are different initialization paths." }, { "left": "issue:10194", "right": "issue:7877", "accept": false, - "reason": "Attention masking support in Flux and adding out_dim to Attention are both attention-adjacent, but they are different API/behavior changes." + "reason": "Flux attention masking support and adding out_dim to Attention block are different API changes." }, { "left": "issue:7033", "right": "issue:7168", "accept": false, - "reason": "Warning logs about IP-Adapter attention masking and a 4D tensor embed-shape question are not the same bug." + "reason": "Attention masking warning logs and incorrect IP-Adapter image embeds are separate issues." } ] }, @@ -30433,10 +30460,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -30451,156 +30478,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "False-positive cluster: these items span many unrelated bugs/features across training, loading, docs, and model-specific failures. No PRs are present, and none of the soft pairs look like the same underlying issue.", - "confidence": 0.97, - "canonical_issue_reason": "No canonical issue: the items are not duplicates of one bug; they are a heterogeneous mix of unrelated reports.", + "summary": "This cluster is noisy and heterogeneous: it mixes unrelated training, loading, import, and docs issues across different models and pipelines. None of the soft pairs look like the same underlying bug/change, so all should be rejected.", + "confidence": 0.9, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a safe representative for the cluster because the reports do not share one concrete root cause.", + "best_issue_reason": "No single issue is a strong canonical representative because the cluster spans multiple unrelated problem types; the closest subtheme is model-loading failures, but even those split across different APIs and models.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:2875", "right": "issue:6565", "accept": false, - "reason": "Both are training-loss problems, but one is text-to-image NaNs and the other is DPO diffusion loss spikes/noised samples; different algorithms and failure modes." + "reason": "Both are training-loss problems, but one is generic text-to-image NaNs and the other is DPO diffusion loss spikes/noised samples; different training setups and failure modes." }, { "left": "issue:2947", "right": "issue:4073", "accept": false, - "reason": "Different code paths: missing `load_attn_procs` on UNet vs DeepFloyd IF safety checker calling `astype` on a tensor." + "reason": "Different components and errors: missing `load_attn_procs` on UNet vs safety-checker `astype` on a Tensor." }, { "left": "issue:3111", "right": "issue:3480", "accept": false, - "reason": "One asks for a custom conditioning encoder; the other requests a scale parameter for text-encoder LoRA layers. Related area, but not the same change." + "reason": "One is about using a custom conditioning encoder; the other is a feature request for a LoRA scale parameter. Related area, but not the same change." }, { "left": "issue:4969", "right": "issue:5892", "accept": false, - "reason": "One is a specific `CrossAttention` import failure; the other is only a generic import error with no concrete shared cause." + "reason": "Both mention import errors, but the actual missing symbols are unspecified vs `CrossAttention`; too vague to be the same bug." }, { "left": "issue:3405", "right": "issue:4086", "accept": false, - "reason": "Both mention ControlNet training, but they involve different model families and symptoms; not enough to treat as one bug." + "reason": "Both involve ControlNet training, but one is a general training bug and the other is an SDXL-specific training failure; not the same concrete issue." }, { "left": "issue:1572", "right": "issue:3631", "accept": false, - "reason": "Both involve inpainting, but one is a broken mega pipeline for img2img/inpainting and the other is an exception when passing latents; different bugs." + "reason": "Both touch inpainting, but the first is a pipeline incompatibility and the second is an exception when passing latents; different code paths." }, { "left": "issue:6548", "right": "issue:9558", "accept": false, - "reason": "RGBA latents handling in `prepare_latents` is unrelated to black validation images in TensorBoard." + "reason": "RGBA image misdetected as latents vs black validation image in TensorBoard are unrelated symptoms." }, { "left": "issue:5515", "right": "issue:5619", "accept": false, - "reason": "LCM scheduler initialization failure is a different issue from AutoencodeTiny image-encoding behavior in LCM img2img." + "reason": "Both are LCM-related, but one is scheduler initialization and the other is AutoencodeTiny image encoding in img2img; not the same bug." }, { "left": "issue:11878", "right": "issue:13105", "accept": false, - "reason": "Wan model GGUF loading/runtime failure is not the same as output noise on a different Wan2.1 T2V model." + "reason": "Different Wan model problems: GGUF loading failure for 1.3B vs noisy outputs in Wan2.1 text-to-video." }, { "left": "issue:6256", "right": "issue:7293", "accept": false, - "reason": "Two separate feature requests in different pipelines; not one concrete bug or change." + "reason": "Both are feature requests, but they ask for unrelated pipeline capabilities." }, { "left": "issue:6439", "right": "issue:7119", "accept": false, - "reason": "Both are docs/link problems, but they reference different invalid links and pages." + "reason": "Both are docs link issues, but they concern different links/pages and are not clearly the same broken reference." }, { "left": "issue:6870", "right": "issue:7735", "accept": false, - "reason": "`enable_vae_tiling` missing from img2img/inpaint pipelines is unrelated to a missing `caption_column` dataset field." + "reason": "VAE tiling missing in img2img/inpaint vs missing `caption_column` in custom dataset training are unrelated." }, { "left": "issue:9208", "right": "issue:9804", "accept": false, - "reason": "Both concern loading ControlNet locally, but one is `from_single_file()` for most models and the other is `from_pretrained()` for SDXL ControlNet; different APIs and failures." + "reason": "Both are ControlNet-loading complaints, but they target different APIs (`from_single_file` vs `from_pretrained`) and different model types; not one concrete bug." }, { "left": "issue:10121", "right": "issue:11185", "accept": false, - "reason": "Different models and bugs: FLUX high-resolution generation vs CogView4 image-generation error." + "reason": "Different models and errors: FLUX high-resolution generation vs CogView4 generation error." }, { "left": "issue:6016", "right": "issue:7003", "accept": false, - "reason": "Shape mismatch in model conversion is a different problem from an inpaint pipeline weight-shape mismatch." + "reason": "Both involve tensor shape issues, but one is in the converter and the other is in inpainting model weights; different failure points." }, { "left": "issue:7538", "right": "issue:9731", "accept": false, - "reason": "One is a specific training issue on a custom Fill50k-style dataset; the other is a how-to / usage question for Playground2.5 LoRA training." + "reason": "One is a training issue on a custom Fill50k-style dataset; the other is a how-to/feature question for Playground2.5 LoRA training." }, { "left": "issue:7757", "right": "issue:7960", "accept": false, - "reason": "Both are SDXL multi-GPU training failures, but one is a ControlNet OOM and the other is a general multi-GPU failure; not the same underlying bug." + "reason": "Both are multi-GPU SDXL training failures, but one is a specific ControlNet script OOM and the other is a broader training failure; not enough to merge." }, { "left": "issue:1315", "right": "issue:2596", "accept": false, - "reason": "An import error for `PIL_INTERPOLATION` is unrelated to a documentation link mentioning a non-existing `load_image` helper." + "reason": "Import-time missing symbol vs documentation link to a non-existing helper; different problem classes." }, { "left": "issue:9278", "right": "issue:9501", "accept": false, - "reason": "Both are Flux dreambooth training reports, but one is multigpu training and the other is delayed model saving; different symptoms and code paths." + "reason": "Both involve FLUX DreamBooth training, but one is a multi-GPU training bug and the other is delayed model saving; different concrete symptoms." }, { "left": "issue:2163", "right": "issue:6130", "accept": false, - "reason": "UNet dtype mismatch during text-to-image training is not the same as a LoRA loading error." + "reason": "Training dtype mismatch vs LoRA loading error are different workflows and bugs." }, { "left": "issue:5128", "right": "issue:6196", "accept": false, - "reason": "Local `.safetensors` loading with XLPipeline is a different failure from loading ControlNetXS Depth SDXL." + "reason": "Both are loading issues, but one is local `.safetensors` via XLPipeline and the other is ControlNetXS Depth SDXL loading; different models and code paths." }, { "left": "issue:1944", "right": "issue:730", "accept": false, - "reason": "DreamBooth random generations and DreamBooth learning locally vs Colab are not the same bug." + "reason": "Both are DreamBooth reports, but one is poor prompt learning on SD1.5/2.1-768 and the other is local-vs-Colab training behavior; not the same bug." }, { "left": "issue:10925", "right": "issue:11464", "accept": false, - "reason": "Training FLUX dev with CLIPTextModel is unrelated to loading a FLUX Canny LoRA into FluxControlPipeline." + "reason": "Both mention FLUX dev and LoRA/finetuning, but one is a DreamBooth finetune question and the other is a ControlPipeline loading failure." }, { "left": "issue:4186", "right": "issue:6870", "accept": false, - "reason": "Both mention inpainting, but the refiner bug and missing `enable_vae_tiling` support are distinct issues." + "reason": "Both are inpainting-related, but one is SDXL refiner incompatibility and the other is missing VAE tiling support; different underlying issues." } ] }, @@ -30703,10 +30730,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -30721,156 +30748,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose collection of unrelated Diffusers issues. The only strong duplicate-like pair is the Flux multi-GPU device-placement error around ControlNet/transformer sharding; most other pairs are just same-subsurface or same-model-family, not the same bug.", - "confidence": 0.42, - "canonical_issue_reason": "issue:9450 is the clearest, most concrete bug report in the cluster: a reproducible FluxPipeline multi-GPU device mismatch with an explicit error message. It is also the best anchor for the one near-duplicate ControlNet/Flux multi-GPU report.", + "summary": "This cluster is highly heterogeneous: most items are unrelated bugs or feature requests that only share broad keywords like multi-GPU, loading, or training. None of the proposed soft edges look like the same concrete underlying issue.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue fits the set well; the items span distinct subsystems, model families, and failure modes. If forced to pick a representative, issue 10421 is the broadest multi-GPU-related report, but it still does not cover the rest.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9450 best represents the cluster because it names the exact failing code path and error, and it is the most likely target for consolidation with the closest duplicate candidate.", + "best_issue_reason": "Issue 10421 is the closest broad representative of the multi-GPU/device-placement subgroup, but the overall set is too mixed for a strong dedup canonical.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4356", "right": "issue:5259", "accept": false, - "reason": "Both are loading/import failures, but one is a missing dummy module and the other is missing model_index.json; different failure modes and code paths." + "reason": "Different failure modes: a missing packaged module vs a missing model_index.json in a local model directory." }, { "left": "issue:8956", "right": "issue:9804", "accept": false, - "reason": "Both are model loading problems, but OVStableDiffusionXLPipeline failing to load is not the same as ControlNetModel.from_pretrained failing on a local SDXL ControlNet." + "reason": "Both are load failures, but one is an OpenVINO pipeline model load issue and the other is a local SDXL ControlNet loading bug." }, { "left": "issue:11363", "right": "issue:9895", "accept": false, - "reason": "CUDA OOM in a ControlNet inference example is different from a FluxPipeline cross-device tensor mismatch regression." + "reason": "OOM in a controlnet inference example is not the same bug as Flux device-mismatch after a specific commit." }, { "left": "issue:3643", "right": "issue:7378", "accept": false, - "reason": "These are both memory-related during decoding, but one is an INT_MAX/upsample limitation and the other is higher memory use in a video-to-video decoding strategy." + "reason": "Both mention decoding latents, but one is an INT_MAX kernel limit and the other is a memory-usage complaint about separate decoding." }, { "left": "issue:12009", "right": "issue:12529", "accept": false, - "reason": "Both are Wan2.2-related, but one is a from_single_file GGUF load failure and the other is a tiled VAE tensor-size mismatch." + "reason": "Both are Wan-related, but one is single-file GGUF loading and the other is a tiled VAE tensor-size mismatch in a different path." }, { "left": "issue:3803", "right": "issue:3954", "accept": false, - "reason": "Multi-GPU training issues, but one is model replication/share across GPUs and the other is incorrect LR scheduler step counting." + "reason": "Multi-GPU context only; one is model sharing in LoRA training, the other is incorrect scheduler step counting." }, { "left": "issue:4446", "right": "issue:8900", "accept": false, - "reason": "Both involve from_single_file loading, but custom text encoder support and loading an SDXL refiner file are different problems." + "reason": "Both involve from_single_file, but one is custom text_encoder support and the other is loading an SDXL refiner checkpoint." }, { "left": "issue:7146", "right": "issue:7310", "accept": false, - "reason": "Both are SDXL training issues, but one is CUDA OOM and the other is a defaults/behavior problem in a different script." + "reason": "Both are SDXL training issues, but one is CUDA OOM and the other is a defaults/configuration bug in the LoRA script." }, { "left": "issue:5488", "right": "issue:6870", "accept": false, - "reason": "A demo generation failure is not the same bug as missing enable_vae_tiling support in img2img/inpaint pipelines." + "reason": "A demo code failure is unrelated to missing enable_vae_tiling support in img2img/inpaint pipelines." }, { "left": "issue:7936", "right": "issue:8833", "accept": false, - "reason": "Local safetensors parsing as invalid JSON and missing config.json are different loading errors." + "reason": "A malformed local safetensors load is not the same as a missing config.json for an SD3 model repo." }, { "left": "issue:5570", "right": "issue:5917", "accept": false, - "reason": "A missing compute_snr helper in the package is unrelated to convert_ldm_unet_checkpoint failing after a version upgrade." + "reason": "Missing compute_snr in a package release is a different issue from convert_ldm_unet_checkpoint breaking after an upgrade." }, { "left": "issue:10705", "right": "issue:9450", "accept": false, - "reason": "Both mention FluxPipeline multi-GPU behavior, but one reports slower inference on reuse while the other is an explicit device-placement crash." + "reason": "Both are Flux-related, but one is a reuse/performance slowdown and the other is a multi-GPU device-placement crash." }, { "left": "issue:10893", "right": "issue:7293", "accept": false, - "reason": "A feature request for CogVideoX ControlNet integration is unrelated to an AudioLDM2 pipeline integration request." + "reason": "These are separate feature requests for different pipelines and models, not the same change request." }, { "left": "issue:1251", "right": "issue:8832", "accept": false, - "reason": "RePaint on MPS returning black images and a generic pipe(prompt) crash are different runtime failures." + "reason": "RePaint on MPS returning black images is a different bug from a generic pipe(prompt) crash." }, { "left": "issue:2875", "right": "issue:6821", "accept": false, - "reason": "Training loss becoming NaN and an invalid caption_column configuration error are unrelated." + "reason": "Training NaNs and a caption_column validation error are unrelated training-path problems." }, { "left": "issue:6157", "right": "issue:6218", "accept": false, - "reason": "IP-Adapter not working in ControlNetImg2Img and batched seeds not working are different ControlNet pipeline issues." + "reason": "An ip_adapter integration problem is distinct from a batched-seeds bug in ControlNet/Img2Img." }, { "left": "issue:10421", "right": "issue:11247", "accept": false, - "reason": "A CPU memory leak when moving pipelines across multiple GPUs is not the same as a FluxTransformer2DModel device-distribution error with ControlNet." + "reason": "Both are multi-GPU/device-placement themed, but one reports a CPU memory leak while the other is a controlnet distribution error for FluxTransformer2DModel." }, { "left": "issue:1185", "right": "issue:2465", "accept": false, - "reason": "Both are DreamBooth reproducibility complaints, but one is about reproducing paper results and the other about a training recipe for a blog." + "reason": "Reproducibility complaints vs asking for a reproducible training recipe are related topic-wise, not the same underlying bug." }, { "left": "issue:1572", "right": "issue:2026", "accept": false, - "reason": "A mega pipeline failing for img2img/inpainting and cpu_offload not working in PaintByExample are separate pipeline-specific issues." + "reason": "A pipeline failing for img2img/inpainting is not the same as cpu_offload not working in PaintByExample." }, { "left": "issue:8540", "right": "issue:8555", "accept": false, - "reason": "Both are SD3 training-related feature asks, but caching embeddings/VAE representations is not the same as FP8/text-encoder training availability." + "reason": "One asks for caching embeddings/VAE representations in SD3 training; the other requests FP8 and text-encoder training support." }, { "left": "issue:5447", "right": "issue:5619", "accept": false, - "reason": "Both involve image-to-image behavior, but the scheduler denoising_start bug is unrelated to AutoencodeTiny encoding behavior." + "reason": "Both are img2img-related, but they concern different models and different scheduler/encoding bugs." }, { "left": "issue:4488", "right": "issue:5592", "accept": false, - "reason": "An SDXL checkpoint loading bug without pipeline_class and UniDiffuser failing to load from checkpoints are different model-loading paths." + "reason": "SDXL checkpoint loading without pipeline_class and UniDiffuser checkpoint loading are separate model-loading failures." }, { "left": "issue:4593", "right": "issue:4975", "accept": false, - "reason": "Loading an Automatic1111 LoRA with rank constraints and inefficient datatype queries during LoRA loading are not the same bug." + "reason": "Automatic1111 LoRA rank incompatibility is not the same as repeated datatype queries causing slow loading." }, { "left": "issue:1062", "right": "issue:1944", "accept": false, - "reason": "Both are DreamBooth reproducibility complaints, but one says it cannot repeat paper results and the other reports random results for a prompt; too broad to treat as the same bug." + "reason": "Both mention Dreambooth reproducibility, but one is a generic inability to repeat paper results and the other is a specific prompt/result mismatch." } ] }, @@ -30973,10 +31000,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -30991,162 +31018,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous, but there are a couple of plausible duplicate pairs around `from_single_file` loading failures. The rest look like distinct bugs or model-specific reports rather than the same underlying issue.", - "confidence": 0.68, - "canonical_issue_reason": "Issue 9208 is the strongest representative of the shared loader failure theme: it states the broad ControlNet `from_single_file()` loading problem, and 5577 is a concrete KeyError symptom of that path.", + "summary": "This is a loose similarity cluster of mostly unrelated diffusers issues. The only pair that looks like a likely true duplicate is the from_single_file checkpoint-loading failure pair; the rest share broad subsystem vocabulary but not the same concrete bug.", + "confidence": 0.33, + "canonical_issue_reason": "issue:8359 is the broadest and most representative of the only credible duplicate-like subtopic in this set: single-file/custom checkpoint loading failures.", "canonical_pr_reason": null, - "best_issue_reason": "9208 is the best issue representative because it is broad, specific, and captures the underlying `from_single_file` ControlNet loading failure more cleanly than the narrower symptom reports.", + "best_issue_reason": "issue:8359 is the best overall anchor because it is specific, broadly phrased, and most likely to subsume the closely related from_single_file checkpoint-loading report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:9164", "right": "issue:9227", "accept": false, - "reason": "Both are FLUX training complaints, but one is convergence on an example and the other is a general training/run problem; not the same bug." + "reason": "Same general FLUX training area, but one is a convergence complaint and the other is a runtime/model-loading problem; not the same bug." }, { "left": "issue:12612", "right": "issue:12699", "accept": false, - "reason": "Different WAN models and different symptoms: NaNs in latents vs degraded quality over many prompts." + "reason": "Both mention WAN, but one is NaN latents and the other is quality degradation over many prompts; different failure modes." }, { "left": "issue:2364", "right": "issue:2480", "accept": false, - "reason": "Generic training-script failure vs missing per-checkpoint `unet` folders; unrelated symptoms." + "reason": "Both relate to training scripts, but one is a generic inability to run training and the other is a missing checkpoint folder output; not the same issue." }, { "left": "issue:8359", "right": "issue:9052", "accept": true, - "reason": "Both report `from_single_file` failing on non-standard/custom checkpoints, i.e. the same loader path and same underlying compatibility bug." + "reason": "Both are from_single_file loading failures on nonstandard/custom checkpoints, pointing to the same underlying loader bug." }, { "left": "issue:9237", "right": "issue:9278", "accept": false, - "reason": "Same broad FLUX training area, but one is a general issue and the other is specifically multi-GPU; not clearly the same defect." + "reason": "Both are FLUX dreambooth training reports, but one is a general issue and the other is specifically multigpu; too broad to treat as one bug." }, { "left": "issue:11133", - "right": "issue:9613", + "right": "issue:13258", "accept": false, - "reason": "Different pipelines and different failure modes; no clear shared code-path bug." + "reason": "Different video pipelines/models (CogVideoX vs Wan 2.2 TI2V) with different symptoms; not a shared concrete bug." }, { "left": "issue:5647", "right": "issue:5687", "accept": false, - "reason": "Both touch img2img/inpaint strength edge cases, but they describe different scheduler/pipeline failures and are not clearly the same bug." + "reason": "Both affect img2img/inpaint strength handling, but the specific failure conditions and code paths differ." }, { "left": "issue:9026", "right": "issue:9069", "accept": false, - "reason": "An unexpected keyword argument error and a NumPy type mismatch are unrelated." + "reason": "Completely different errors: one is an unexpected keyword in dispatch, the other is an ndarray type mismatch." }, { "left": "issue:11160", "right": "issue:13258", "accept": false, - "reason": "Both involve video models, but one is inconsistent output and the other is a missing image-to-video pipeline; not the same issue." + "reason": "Both are I2V-related, but they concern different models and different reported problems." }, { "left": "issue:6297", "right": "issue:8939", "accept": false, - "reason": "Dreambooth training trouble vs tokenizer loading error for SD3; different problem domains." + "reason": "Both are loading/training-related, but one is SDXL LoRA training and the other is a tokenizer load error for SD3; not the same bug." }, { "left": "issue:5577", "right": "issue:9208", - "accept": true, - "reason": "Both are ControlNet `from_single_file()` loading failures; 5577 is a specific KeyError manifestation of the broader bug in 9208." + "accept": false, + "reason": "Both involve from_single_file/controlnet loading, but one is a specific missing key and the other is a broad inability to load most ControlNet models." }, { "left": "issue:3101", "right": "issue:6540", "accept": false, - "reason": "Different training errors with no shared concrete failure path." + "reason": "Different training-time failures: ControlNet example error vs FSDP matmul error." }, { "left": "issue:1672", "right": "issue:2659", "accept": false, - "reason": "Wrong output image size after conversion vs batch-size-dependent variation; not the same underlying bug." + "reason": "Wrong image size after conversion is unrelated to batch-size-dependent output variation." }, { "left": "issue:5593", "right": "issue:6069", "accept": false, - "reason": "`AutoPipeline.from_pipe` ControlNet issue vs scheduler `add_noise` failure at `t=0`; unrelated." + "reason": "Different code paths entirely: AutoPipeline.from_pipe with ControlNet vs scheduler.add_noise at t=0." }, { "left": "issue:4255", "right": "issue:5881", "accept": false, - "reason": "Both are quality complaints, but they reference different models and no common code-path bug." + "reason": "Both are subjective quality complaints, not the same bug or code path." }, { "left": "issue:10671", "right": "issue:8909", "accept": false, - "reason": "Deterministic device/dtype lookup issue vs saving `text_encoder_ti`; unrelated." + "reason": "One is a deterministic dtype/device issue in DiffusionPipeline; the other is a training save problem for text_encoder_ti." }, { "left": "issue:1976", "right": "issue:3812", "accept": false, - "reason": "Tokenizer file lookup error vs missing `controlnet` config key; unrelated." + "reason": "Tokenizer file lookup and ControlNet config parsing are unrelated failures." }, { "left": "issue:9768", "right": "issue:9911", "accept": false, - "reason": "NaN/SDPA backend issue vs multi-ControlNet shape mismatch; different failures." + "reason": "Different problems: SDPA/CuDNN NaNs vs multi-ControlNet layer-length mismatch in Flux." }, { "left": "issue:1572", "right": "issue:2439", "accept": false, - "reason": "Both mention inpainting, but one is a mega pipeline limitation and the other is a general inpainting pipeline loading/running issue." + "reason": "Both mention inpainting, but one is a mega pipeline capability issue and the other is a load/run failure; not the same underlying bug." }, { "left": "issue:6250", "right": "issue:9667", "accept": false, - "reason": "SDXL VAE single-file loading vs FLUX fp8 local transformer loading; unrelated." + "reason": "Different models and loaders: SDXL VAE single-file keyerror vs Flux fp8 local transformer loading." }, { "left": "issue:7243", "right": "issue:7405", "accept": false, - "reason": "Different model announcements, not duplicate bugs." + "reason": "Separate new-model requests, not duplicate bugs." }, { "left": "issue:10139", "right": "issue:6870", "accept": false, - "reason": "UNet sample-size init bug vs missing `enable_vae_tiling` on img2img/inpaint pipelines; unrelated." + "reason": "Both touch pipeline initialization/loading, but one is UNet sample size handling and the other is missing enable_vae_tiling; different fixes." }, { "left": "issue:1434", "right": "issue:1478", "accept": false, - "reason": "Same upscale pipeline, but different concrete bugs: OOM at larger sizes vs inability to output more than one image." + "reason": "Both are upscale pipeline problems, but one is OOM and the other is multi-image output support." }, { "left": "issue:3428", "right": "issue:5630", "accept": false, - "reason": "General unexpected img2img result vs scheduler noise leftover in SDXL img2img; not the same defect." + "reason": "Different img2img issues: unexpected results after changing the pipeline vs noise leftover with LMS scheduler." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: issue 9208 is a reasonable canonical representative for the ControlNet `from_single_file()` loader failures, and the two accepted soft pairs both look like the same bug family from the titles alone. The remaining pairs are handled conservatively as distinct or only loosely related, without overclaiming evidence. No PR-pair mergeability concerns apply here since these are issues, not PRs." + "feedback": "The summary is grounded in the packet and stays conservative overall. The only clearly supported duplicate-like pair is the from_single_file checkpoint-loading one, and the other pair calls are mostly defensible as non-duplicates given the titles alone." }, "evaluator_used": true, "retried": false, @@ -31245,10 +31272,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -31263,8 +31290,8 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Loose, heterogeneous cluster of unrelated Diffusers issues (LoRA/inference, import errors, training scripts, video pipelines, and runtime edge cases). No single underlying bug/change stands out as a true duplicate target.", - "confidence": 0.95, + "summary": "These are mostly unrelated support/issues; the soft pairs share only superficial wording and do not look like the same underlying bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -31274,145 +31301,145 @@ "left": "issue:10223", "right": "issue:10472", "accept": false, - "reason": "Different problems: one asks where to get a LoRA ID for inference, the other reports a constructor TypeError on `lora_bias`." + "reason": "Different problems: one asks where to get a LoRA ID, the other reports a lora_bias constructor error during inference." }, { "left": "issue:11199", "right": "issue:11218", "accept": false, - "reason": "Both are import-related, but they concern different missing symbols/modules (`onnx_utils` vs `CogView4LoraLoaderMixin`)." + "reason": "Both are import failures, but for different modules/classes and likely different breakages." }, { "left": "issue:3377", "right": "issue:3539", "accept": false, - "reason": "`SNR weighting optimization` is a training-method question; `main_process_first` is an API usage question." + "reason": "One is about SNR weighting in training, the other about main_process_first semantics; unrelated topics." }, { "left": "issue:5544", "right": "issue:6517", "accept": false, - "reason": "Both are pipeline errors, but one is about FreeU disable handling and the other is a controlnet SDXL import failure." + "reason": "FreeU disable error vs ControlNet SDXL import traceback; different code paths and symptoms." }, { "left": "issue:6157", "right": "issue:6524", "accept": false, - "reason": "Both mention LoRA, but one is about IP-Adapter not working in a controlnet pipeline, the other about reloading LoRA a second time." + "reason": "IP-Adapter in ControlNet img2img versus loading LoRA a second time; not the same defect." }, { "left": "issue:8645", "right": "issue:9141", "accept": false, - "reason": "Different single-file utility problems: missing controlnet model converter vs missing `FluxPipeline.from_single_file()`." + "reason": "Missing single_file_utils helper versus missing FluxPipeline.from_single_file; different APIs." }, { "left": "issue:11298", "right": "issue:8649", "accept": false, - "reason": "`11298` is a LoRA hotswap PEFT key collision; `8649` is an SD3 `num_images_per_prompt` regression." + "reason": "LoRA hot-swapping key error is not the same as SD3 num_images_per_prompt handling." }, { "left": "issue:2052", "right": "issue:4450", "accept": false, - "reason": "Both are textual inversion related, but one is checkpoint resume support and the other is multi-text-inversion example malfunction." + "reason": "Textual inversion resume-from-checkpoint issue differs from multi-text inversion not working." }, { "left": "issue:3884", "right": "issue:4639", "accept": false, - "reason": "Both involve training scripts, but one is `train_text_to_image_lora.py` and the other is a multiprocessing data loader error in `train_instruct_pix2pix_sdxl.py`." + "reason": "train_text_to_image_lora.py error and multiprocessing dataloader error are separate training failures." }, { "left": "issue:10317", "right": "issue:8879", "accept": false, - "reason": "Both are `torch.compile` issues, but one is on StableVideoDiffusionPipeline with different images and the other is SDXL cudagraph failure." + "reason": "Both involve torch.compile, but one is StableVideoDiffusionPipeline with changing images and the other is StableDiffusionXL with cudagraphs." }, { "left": "issue:3694", "right": "issue:8885", "accept": false, - "reason": "`3694` is about missing model/config files in a finetuning example; `8885` is about installing clip-retrieval and loading a modifier token." + "reason": "Pokemon finetuning file/config issue vs clip-retrieval/modifier-token setup problem; unrelated." }, { "left": "issue:3382", "right": "issue:3885", "accept": false, - "reason": "Same training script family, but one is multi-GPU OOM behavior and the other is a general VRAM reduction question." + "reason": "Multi-GPU OOM in text-to-image training is different from general VRAM reduction question." }, { "left": "issue:255", "right": "issue:2844", "accept": false, - "reason": "Both are pipeline runtime errors, but they affect different code paths: image size validation in StableDiffusion vs explicit image embeddings in StableUnCLIPImg2Img." + "reason": "Both runtime errors in pipelines, but one is image-size validation in StableDiffusionPipeline and the other is explicit image embeddings in StableUnCLIPImg2ImgPipeline." }, { "left": "issue:6154", "right": "issue:9856", "accept": false, - "reason": "`6154` is an M2 Mac Mini SDXL LoRA training issue; `9856` is a distributed port conflict." + "reason": "M2 Mac LoRA training issue is unrelated to distributed port conflict." }, { "left": "issue:4638", "right": "issue:4885", "accept": false, - "reason": "Both concern textual inversion/LoRA loading, but one is SDXL textual inversion support and the other is SDXL ControlNet img2img LoRA loading." + "reason": "SDXL textual inversion loading problem versus SDXL ControlNet img2img LoRA loading problem; different loaders." }, { "left": "issue:9320", "right": "issue:9556", "accept": false, - "reason": "Both are LoRA save/load problems, but one is custom diffusion finetune persistence and the other is rslora model save/load behavior." + "reason": "Saved/loaded custom diffusion models and rslora saved-model issues concern different training artifacts." }, { "left": "issue:4306", "right": "issue:5544", "accept": false, - "reason": "Different pipeline bugs: list index error in StableDiffusionReferencePipeline vs FreeU disable handling in SDXL." + "reason": "ReferencePipeline list index error is unrelated to FreeU toggle behavior in SDXL." }, { "left": "issue:10144", "right": "issue:11351", "accept": false, - "reason": "Both are video-related, but one compares Mochi video quality and the other is a datatype question for Wan i2v preprocessing." + "reason": "Mochi video quality complaint is unrelated to Wan i2v processor dtype behavior." }, { "left": "issue:4241", "right": "issue:8485", "accept": false, - "reason": "One asks about supporting LoRA fine-tuning for unidiffusers; the other is a feature request for SD3 LoRA." + "reason": "General question about unidiffusers LoRA support versus SD3 LoRA support request." }, { "left": "issue:4256", "right": "issue:7619", "accept": false, - "reason": "Both are training failures, but one is SDXL Dreambooth and the other is a number-of-channels error in Dreambooth." + "reason": "SDXL Dreambooth training error and Dreambooth number-of-channels error are both training-related but not the same concrete bug." }, { "left": "issue:5603", "right": "issue:9410", "accept": false, - "reason": "Both are import errors, but they involve different missing exports (`is_safetensors_available` vs `StableDiffusion3ControlNetInpaintingPipeline`)." + "reason": "Different missing imports: utility function vs ControlNet inpainting pipeline class." }, { "left": "issue:6112", "right": "issue:7568", "accept": false, - "reason": "`6112` is about SDXL-Turbo output quality; `7568` is about using prompt tokens for interpolation. Different behavior and API usage." + "reason": "SDXL turbo noise generation issue is not the same as prompt-tokens interpolation discussion." }, { "left": "issue:4402", "right": "issue:6610", "accept": false, - "reason": "Both involve SDXL inpaint single-file loading, but one is a `from_single_file()` usage problem and the other is `num_in_channels` having no effect." + "reason": "Single-file SDXL inpaint usage issue differs from num_in_channels being ignored in the same pipeline family." }, { "left": "issue:3461", "right": "issue:3584", "accept": false, - "reason": "Both are DreamBooth-related, but one is a performance/memory comparison and the other is a tutorial failure." + "reason": "DreamBooth performance comparison and tutorial failure are distinct problems." } ] }, @@ -31515,10 +31542,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -31533,156 +31560,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the items span FP8 inference noise, LoRA loading/training, SD3/Flux device and offload errors, OOMs, and unrelated pipeline-specific bugs. None of the soft pairs look like true duplicates of the same concrete bug.", - "confidence": 0.96, - "canonical_issue_reason": "issue:9895 is the most concrete, actionable bug report in the set: a specific FluxPipeline inference device-mismatch regression with a clear error message and likely single code path.", + "summary": "This cluster is highly heterogeneous: most pairs only share superficial topic words (LoRA, FP8, OOM, multi-GPU, black image, etc.) and do not look like the same underlying bug or change.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 5540 is the broadest, most generic report in the set (\u201cInference out of memory\u201d), so it is the closest thing to a representative bucket, though the cluster is not a single duplicate family.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9895 is the best representative because it has the clearest repro, explicit error, and a focused inference-path failure; the rest of the items are too diverse to serve as a stronger cluster anchor.", + "best_issue_reason": "Issue 5540 is the least specific and most reusable as a placeholder canonical issue, but it is still far too generic to truly cover the other reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3222", "right": "issue:6176", "accept": false, - "reason": "Different failure modes: missing training output file vs error loading LoRA weights after DreamBooth training." + "reason": "Both involve LoRA weight loading, but the contexts differ (custom_diffusion training output vs dreambooth_lora loading), so they are not clearly the same bug." }, { "left": "issue:6107", "right": "issue:7049", "accept": false, - "reason": "Both mention IP-Adapter usage, but one is environment-specific and the other is a tensor shape/unpack bug in a basic example; not the same underlying defect." + "reason": "Both mention IP Adapter/Stable Diffusion, but one is environment-specific EC2 behavior and the other is a concrete tensor-unpack error in the basic example." }, { "left": "issue:11014", "right": "issue:2089", "accept": false, - "reason": "Both are OOM-related, but they involve different models, workflows, and contexts; too broad to be duplicates." + "reason": "Both are OOM reports, but one is inference on Colab A100 and the other is training DreamBooth AltDiffusion; different code paths and causes." }, { "left": "issue:7431", "right": "issue:9237", "accept": false, - "reason": "Both are Flux training issues, but one is DoRA/text-encoder training and the other is Flux Dreambooth LoRA training; distinct problems." + "reason": "Both are training-related LoRA reports, but one is DoRA with text encoder and the other is Flux Dreambooth LoRA; not the same underlying failure." }, { "left": "issue:11247", "right": "issue:7682", "accept": false, - "reason": "Both concern multi-GPU usage, but one is FluxTransformer2DModel with controlnet and the other is SD1.5 on multiple GPUs; not the same bug." + "reason": "Both concern multiple GPUs, but the model/pipeline and failure mode differ substantially (Flux+ControlNet distribution vs SD1.5 on weak GPUs)." }, { "left": "issue:8556", "right": "issue:8644", "accept": false, - "reason": "Both involve meta tensor/offload issues in SD3 loading, but one is from_single_file and the other is separately loading the T5 encoder with sequential/model CPU offload; closely related, but not clearly the same defect." + "reason": "Both mention meta-tensor/offload style errors, but one is SD3 from_single_file loading and the other is separately loading T5 with sequential/model CPU offload; too context-specific to merge." }, { "left": "issue:10795", "right": "issue:7159", "accept": false, - "reason": "Both mention torch.compile and GGUF/LoRA, but one is GGUF from_single_file loading and the other is fused LoRA compilation; different code paths." + "reason": "Both involve torch.compile, but the failures are on different model-loading paths (GGUF via from_single_file vs fused LoRA); not the same bug." }, { "left": "issue:5254", "right": "issue:962", "accept": false, - "reason": "Both are inpaint-related, but one is multi-ControlNet support and the other is size limitations in new inpaint pipelines; different changes." + "reason": "Both are inpaint-related, but one is multi-ControlNet support and the other is size limitation in new inpaint pipelines; different issues." }, { "left": "issue:10972", "right": "issue:11702", "accept": false, - "reason": "Both are LoRA loading errors, but they target different model/tooling formats: OneTrainer Flux LoRAs vs AccVid LoRA keys." + "reason": "Both are LoRA loading problems, but they involve different formats and error surfaces (OneTrainer Flux LoRAs vs AccVid LoRA key error)." }, { "left": "issue:4788", "right": "issue:5540", "accept": false, - "reason": "Both are OOM reports, but they are generic and context differs; no evidence of the same code-path issue." + "reason": "Both are OOM-related, but one is SDXL inpaint on 6 GB VRAM and the other is a generic inference OOM report; not the same concrete bug." }, { "left": "issue:8673", "right": "issue:8708", "accept": false, - "reason": "One is missing attention masks for SD3 padding tokens; the other is missing vae.config.shift_factor in dreambooth examples. Different bugs." + "reason": "Both are SD3 issues, but one is missing attention masks and the other is a missing shift_factor config in training examples; unrelated fixes." }, { "left": "issue:2603", "right": "issue:5670", "accept": false, - "reason": "Both are non-square image generation failures, but they are different pipelines (Attend-And-Excite vs PixArtAlpha) and likely separate constraints/bugs." + "reason": "Both concern non-square image generation, but they hit different pipelines and likely different shape-handling code paths." }, { "left": "issue:1100", "right": "issue:1709", "accept": false, - "reason": "One is a CLIPTextTransformer device attribute exception; the other is an ONNX img2img Tensor.astype error. Different frameworks and failures." + "reason": "The errors are completely different (`device` attribute vs `astype` on Tensor), so these are not duplicates." }, { "left": "issue:10302", "right": "issue:9895", "accept": false, - "reason": "Both are inference-time issues, but one is FP8 noise without CPU offloading and the other is a device placement mismatch in FluxPipeline; related theme, not same bug." + "reason": "One is FP8 inference noise without CPU offloading; the other is a device-mismatch regression after a commit. Different bugs." }, { "left": "issue:1614", "right": "issue:2495", "accept": false, - "reason": "Both produce black images, but one is SD2.1 with autocast and the other is MPS UpscalePipeline; not enough to treat as the same defect." + "reason": "Both produce black images, but one is Stable Diffusion 2.1 with autocast and the other is MPS upscale pipeline; too broad to treat as the same bug." }, { "left": "issue:2359", "right": "issue:4241", "accept": false, - "reason": "Different feature requests: latent output_type support vs LoRA support for unidiffusers." + "reason": "These are unrelated feature requests/issues: latent output_type in img2img versus LoRA fine-tuning support for unidiffusers." }, { "left": "issue:11133", "right": "issue:9637", "accept": false, - "reason": "Both mention image generation / attention behavior, but one is a CogVideoX pipeline bug and the other is an xFormers attention-mask failure with bfloat16; not duplicates." + "reason": "Different subsystems and failures: CogVideoX image-to-video bug versus XFormers attention-mask/bfloat handling." }, { "left": "issue:10608", "right": "issue:7441", "accept": false, - "reason": "Multi-GPU pipeline runtime issue vs pytest failing on Python 3.12; unrelated." + "reason": "Multi-GPU pipeline failure and pytest on Python 3.12 are unrelated." }, { "left": "issue:1363", "right": "issue:3631", "accept": false, - "reason": "One is an ONNX scheduler error for long prompt weighting; the other is an inpainting latents exception. Different pipelines and triggers." + "reason": "ONNX LMS scheduler error and inpainting-with-latents exception are different pipeline bugs." }, { "left": "issue:7221", "right": "issue:7616", "accept": false, - "reason": "Both involve bad image output, but one is a noisy model file inference report and the other is low-denoise img2img artifacts; not the same bug." + "reason": "Both mention noisy/artefactual output, but one is generic noise with a specific model and the other is low-denoise artifacts in img2img; not enough to merge." }, { "left": "issue:8511", "right": "issue:8828", "accept": false, - "reason": "Both concern SD3 training, but one is text encoder unloading and the other is a 16GB training error; different failure modes." + "reason": "One is about text encoders not being unloaded during SD3 training scripts; the other is a general SD3 training error due to limited memory." }, { "left": "issue:9187", "right": "issue:9861", "accept": false, - "reason": "One is IP-Adapter broken after Kolors update with image_encoder_folder=None; the other is Flux training not updating transformer weights. Distinct issues." + "reason": "IP Adapter/SDXL image-encoder handling and Flux training not updating the transformer are different code paths and symptoms." }, { "left": "issue:9514", "right": "issue:9667", "accept": false, - "reason": "Both mention Flux FP8 loading, but one is LoRA weights after FP8 quantisation and the other is a local transformer_flux.py loading error; not the same underlying bug." + "reason": "Both are Flux fp8-related, but one is LoRA loading after quantization and the other is loading a local transformer file; too different to be the same underlying bug." }, { "left": "issue:11642", "right": "issue:11989", "accept": false, - "reason": "FluxPriorRedux prompt_embeds_scale has no effect vs a pure black inference image; different behavior and likely different fixes." + "reason": "Prompt scaling having no effect and black inference images are unrelated symptoms with no clear shared code path." } ] }, @@ -31785,10 +31812,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -31803,156 +31830,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous and mostly contains near-miss similarity pairs across different models, loaders, and failure modes. Most soft pairs are not true duplicates; the only broadly representative issue is the open offloading regression report.", - "confidence": 0.87, - "canonical_issue_reason": "Issue 13488 is the broadest, most representative open report in the cluster: it describes a general offloading regression rather than a single model-specific symptom.", + "summary": "The set is highly heterogeneous: most pairs share only a broad subsystem or keyword, not the same concrete bug/change. No PRs are present. The strongest single issue candidate is a broad Flux loading regression, but it is not a clear duplicate anchor for the whole set.", + "confidence": 0.59, + "canonical_issue_reason": "issue:9343 is the broadest open Flux loading failure report and is the closest fit among the Flux-related items.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 13488 is the best global issue anchor because it captures the shared offloading/device-placement theme more generally than the narrower model-specific reports.", + "best_issue_reason": "issue:9343 is the best standalone issue candidate, but the overall collection is too mixed to treat as one duplicate cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:13281", "right": "issue:13488", "accept": false, - "reason": "Both mention offloading/device placement, but 13281 is a torchao-quantized Qwen-image use_stream/device-mismatch case and 13488 is a broader diffusers offloading regression; not the same bug." + "reason": "Both concern offloading, but one is a quantized Qwen image/device-mismatch case and the other is a diffusers release regression; not the same bug." }, { "left": "issue:3764", "right": "issue:6096", "accept": false, - "reason": "Both are torch.compile failures, but one is a backend shutdown error and the other is a dynamic-output-shape exception; different code paths." + "reason": "Both mention torch.compile, but they fail in different pipelines with different exceptions and likely different root causes." }, { "left": "issue:3097", "right": "issue:7480", "accept": false, - "reason": "load_textual_inversion doing nothing and an OOM in DeepFloyd IF are unrelated failures." + "reason": "Textual inversion loading vs DeepFloyd OOM are unrelated failure modes." }, { "left": "issue:6283", "right": "issue:7168", "accept": false, - "reason": "ControlNet img2img strength=0 behavior and IP-Adapter embed shape handling are different issues." + "reason": "ControlNet img2img strength=0 behavior is unrelated to IP-Adapter image embed tensor shape." }, { "left": "issue:2704", "right": "issue:4282", "accept": false, - "reason": "One is a batch-processing request, the other is adding the Multi-ControlNet pipeline; related area but not the same underlying change." + "reason": "One is a request for batch processing, the other is adding Multi-ControlNet support; related area but not the same change." }, { "left": "issue:2780", "right": "issue:3238", "accept": false, - "reason": "FP16 gradient unscale training failure and LoRA safetensor conversion are unrelated." + "reason": "FP16 gradient unscale error in LoRA training is unrelated to LoRA safetensor conversion." }, { "left": "issue:1500", "right": "issue:3761", "accept": false, - "reason": "Community pipeline API incompatibility and loading two community pipelines in one process are different problems." + "reason": "Community pipeline API compatibility vs loading two community pipelines in one process are different issues." }, { "left": "issue:3468", "right": "issue:5565", "accept": false, - "reason": "Missing model_index.json generation and inability to run the SDXL LoRA DreamBooth script are distinct DreamBooth issues." + "reason": "Missing model_index.json generation and failing to run an SDXL DreamBooth script are different DreamBooth problems." }, { "left": "issue:10712", "right": "issue:7210", "accept": false, - "reason": "SD3 prompt_embeds runtime failure and accelerate dispatch missing named_parameters affect different models/code paths." + "reason": "SD3 prompt_embeds runtime error and missing named_parameters on an SDXL pipeline under accelerate are not the same bug." }, { "left": "issue:3065", "right": "issue:4094", "accept": false, - "reason": "Text-encoder LoRA support for DreamBooth and a T2I adapters training script are separate feature requests." + "reason": "DreamBooth LoRA text-encoder tuning request vs adding a T2I adapter training script are different feature requests." }, { "left": "issue:6196", "right": "issue:6250", "accept": false, - "reason": "ControlNetXS loading failure and SDXL VAE single-file KeyError are different loader bugs." + "reason": "ControlNetXS Depth loading and SDXL VAE single-file KeyError are separate loading failures." }, { "left": "issue:10025", "right": "issue:8886", "accept": false, - "reason": "Both are attention-mask issues, but Flux vs Auraflow means different model implementations and likely separate fixes." + "reason": "Both mention attention masks, but they affect different models and code paths (Flux transformer vs Auraflow SDPA)." }, { "left": "issue:6811", "right": "issue:7009", "accept": false, - "reason": "Missing i2vgen-xl model_index.json and poor text-to-video generation quality are not the same issue." + "reason": "Missing model_index.json for i2vgen-xl vs poor results from a text-to-video pipeline are not the same issue." }, { "left": "issue:9306", "right": "issue:9895", "accept": false, - "reason": "CogVideoX wrong-device selection and Flux device mismatch after a commit are related only at a high level, not the same concrete bug." + "reason": "Different models and different device-selection failures; not the same regression." }, { "left": "issue:2161", "right": "issue:5399", "accept": false, - "reason": "LoRA training on inpainting and a dtype mismatch during custom inpainting inference are different failures." + "reason": "LoRA training on inpainting models and dtype mismatch during custom inpainting inference are distinct bugs." }, { "left": "issue:3915", "right": "issue:4520", "accept": false, - "reason": "torch.compile failing on text-to-video and LoRA scale not working with attention slicing are unrelated." + "reason": "torch.compile on text-to-video pipelines is unrelated to LoRA scale behavior with attention slicing." }, { "left": "issue:11208", "right": "issue:8788", "accept": false, - "reason": "SD3 ControlNet inpainting support and VAE tiling on non-power-of-two images are different SD3 limitations." + "reason": "Multi-ControlNet support in SD3 inpainting and SD3 VAE tiling on non-power-of-two images are unrelated." }, { "left": "issue:2554", "right": "issue:460", "accept": false, - "reason": "An accelerate version requirement and an ONNX Runtime dependency issue are unrelated dependency problems." + "reason": "Accelerate version requirement and ONNXRuntime installation mismatch are unrelated environment issues." }, { "left": "issue:10797", "right": "issue:7982", "accept": false, - "reason": "HunyuanVideo group offloading and MultiControlNet from_pipe behavior are different pipeline features." + "reason": "Group offloading on HunyuanVideo with use_stream and from_pipe with MultiControlNetModel are different failures." }, { "left": "issue:5429", "right": "issue:8540", "accept": false, - "reason": "Supporting models without a VAE in examples and SD3 DreamBooth embedding/vae caching are not the same change." + "reason": "Supporting models without a VAE and SD3 DreamBooth embedding/VAE caching are not the same bug or change." }, { "left": "issue:5873", "right": "issue:6685", "accept": false, - "reason": "ONNX export support for scaled_dot_product_attention and absolute output_dir handling in DreamBooth are unrelated." + "reason": "ONNX export of scaled_dot_product_attention is unrelated to DreamBooth output_dir path handling." }, { "left": "issue:12535", "right": "issue:12565", "accept": false, - "reason": "Wan2.2 lightning LoRA loading and a Wan2.1 i2v running issue are different version-specific problems." + "reason": "Both are Wan i2v issues, but they concern different model versions and symptoms." }, { "left": "issue:9343", "right": "issue:9803", "accept": false, - "reason": "Both involve Flux loading, but one is a model-loading flag regression and the other is a LoRA loading failure; not the same bug." + "reason": "Both are Flux-related loading problems, but one is a mismatched-size/low_cpu_mem_usage load error and the other is LoRA loading failure; not clearly the same bug." }, { "left": "issue:5932", "right": "issue:9497", "accept": false, - "reason": "Channel-mismatch in inpainting and Flux DreamBooth matmul shape errors are unrelated." + "reason": "Channel-count mismatch in inference and a Flux DreamBooth mat2 shape error are different underlying errors." } ] }, @@ -32056,10 +32083,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -32074,162 +32101,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly a loose collection of unrelated training/inference/conversion issues; none of the soft pairs look like true duplicates. I picked the most concrete current LoRA/adapter runtime bug as the canonical issue, but overall cluster cohesion is weak.", - "confidence": 0.78, - "canonical_issue_reason": "Issue 11298 is a specific, reproducible LoRA/adapter runtime failure (PEFT key error during hotswapping) and is the clearest representative of the few LoRA-themed reports here.", + "summary": "These are mostly unrelated issue reports; none of the soft pairs look like true duplicates of the same underlying bug/change. The strongest standalone issue in the set is the active, concrete LoRA hotswapping PEFT-key-error report.", + "confidence": 0.84, + "canonical_issue_reason": "issue:11298 is the best canonical candidate: it is open, detailed, and describes a specific reproducible LoRA hotswapping failure with active discussion/inbound attention.", "canonical_pr_reason": null, - "best_issue_reason": "11298 is the best issue candidate because it has a clear error, a concrete user-facing failure mode, and a narrowly defined fix path.", + "best_issue_reason": "issue:11298 is the strongest issue to anchor triage because it has the clearest concrete bug path and the most active discussion among the listed issues.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3761", "right": "issue:861", "accept": false, - "reason": "Different problems: loading two community pipelines in one process vs a ckpt-to-diffusers conversion failure." + "reason": "Different subsystems and failure modes: community pipeline loading vs ckpt-to-diffusers conversion." }, { "left": "issue:4419", "right": "issue:4578", "accept": false, - "reason": "Offset-noise inference mismatch is unrelated to a documentation complaint about VAE usage." + "reason": "One is an inference behavior bug, the other is a documentation complaint about VAE usage." }, { "left": "issue:4468", "right": "issue:9023", "accept": false, - "reason": "Multi-GPU performance discrepancy is not the same as a generic sample-code runtime failure." + "reason": "Both are sample-code related, but one is multi-GPU performance disparity and the other is a runtime failure." }, { "left": "issue:5259", "right": "issue:7936", "accept": false, - "reason": "Both are load/conversion errors, but one is missing model_index.json while the other is an invalid JSON/safetensors parse issue." + "reason": "Different load/convert errors: missing model_index.json versus invalid local safetensors JSON format." }, { "left": "issue:6382", "right": "issue:8626", "accept": false, - "reason": "One is a PEFT/IP-Adapter key error after LCM LoRA training; the other is a guidance request about multiple IP-Adapter images/masks." + "reason": "One is a training/inference key error with IP-Adapter + LoRA; the other is a guidance/docs request about multiple IP-Adapter images/masks." }, { "left": "issue:1539", "right": "issue:4806", "accept": false, - "reason": "Both mention type problems, but they hit different APIs and different validation paths." + "reason": "Both mention parameter typing, but they affect different APIs and code paths, so they are not the same bug." }, { "left": "issue:3487", "right": "issue:3566", "accept": false, - "reason": "Different pipeline behaviors: inpaint strength with EulerAncestral vs controlnet_reference community pipeline results." + "reason": "Both are poor-result reports, but they involve different pipelines and scheduler/controlnet contexts." }, { "left": "issue:4324", "right": "issue:5502", "accept": false, - "reason": "A NoneType-forward error in UNet is unrelated to fp16 support in a training script." + "reason": "A forward-pass tensor/NoneType error is unrelated to fp16 support in a custom diffusion training script." }, { "left": "issue:5886", "right": "issue:6914", "accept": false, - "reason": "Feature request for face model support is not the same as an offload-related FaceID attribute error." + "reason": "One is a feature request for face-model support; the other is an offload-related FaceID runtime bug." }, { "left": "issue:2459", "right": "issue:2674", "accept": false, - "reason": "One is xformers support in a training script; the other is a missing dependency/requirements issue for LoRA." + "reason": "Different LoRA setup issues: xformers integration in training versus missing safetensors requirement." }, { "left": "issue:2412", "right": "issue:4256", "accept": false, - "reason": "Both are DreamBooth-related, but they describe different errors and training setups, not the same bug." + "reason": "Generic dreambooth errors, but they are different training setups and not clearly the same failure." }, { "left": "issue:6482", "right": "issue:8507", "accept": false, - "reason": "Resume-from-checkpoint FP16 unscale failure is unrelated to SD3 text-encoder training failure." + "reason": "Checkpoint resume FP16 unscale failure is unrelated to SD3 text-encoder training failure." }, { "left": "issue:5607", "right": "issue:9362", "accept": false, - "reason": "Different tensor-shape/indexing failures with no sign of the same code path." + "reason": "Different tensor/index errors with no clear shared code path or root cause." }, { "left": "issue:10855", "right": "issue:8440", "accept": false, - "reason": "Gradient accumulation behavior is unrelated to a FromOriginalModelMixin subclass ValueError." + "reason": "Completely different topics: optimizer stepping under gradient accumulation vs a mixin subclass ValueError." }, { "left": "issue:10969", "right": "issue:9546", "accept": false, - "reason": "Zero3 weight-shape failure and validation OOM are different problems." + "reason": "Both are Flux controlnet training issues, but one is a weight-shape failure and the other is validation OOM; not the same bug." }, { "left": "issue:4436", "right": "issue:7782", "accept": false, - "reason": "Both involve conversion, but one is a CIVITAI/VAE conversion question and the other is an invalid-load-key parse error." + "reason": "Both concern conversion/loading, but one is CIVITAI-with-VAE conversion and the other is an invalid safetensors pickle/load error." }, { "left": "issue:10674", "right": "issue:9187", "accept": false, - "reason": "Flux GGUF support failure is unrelated to an IP-Adapter SDXL issue after the Kolors update." + "reason": "GGUF/FluxPipeline incompatibility is unrelated to SDXL IP-Adapter breakage after Kolors update." }, { "left": "issue:4398", "right": "issue:5689", "accept": false, - "reason": "All-black SDXL LoRA outputs and residual-noise DPM++ behavior are different symptoms with different likely causes." + "reason": "Both affect SDXL image quality, but one is all-black outputs in LoRA Dreambooth and the other is residual noise with DPM++ sampling." }, { "left": "issue:10165", "right": "issue:10952", "accept": false, - "reason": "Seed preservation when changing LoRA scale is unrelated to an adapter-name key error in SDXLLongPromptWeightingPipeline." + "reason": "LoRA seed preservation on scale change is a different problem from adapter-name lookup failure when loading LoRA." }, { "left": "issue:6520", "right": "issue:7310", "accept": false, - "reason": "Blank validation output in T2I adapter training is not the same as SDXL LoRA defaults not working." + "reason": "Different training script failures: blank validation output versus defaults not working at all." }, { "left": "issue:8405", "right": "issue:8667", "accept": false, - "reason": "Both are controlnet shape errors, but they involve different models and failure conditions." + "reason": "Both are controlnet tensor issues, but one is multi-controlnet sizing and the other is non-power-of-two image sizing in SD3." }, { "left": "issue:3431", "right": "issue:3487", "accept": false, - "reason": "Both are inpainting/controlnet quality complaints, but they describe distinct pipelines and scenarios." + "reason": "Both are inpainting/controlnet quality problems, but the described pipelines and schedulers differ." }, { "left": "issue:6008", "right": "issue:6107", "accept": false, - "reason": "IP-Adapter+Face failing with LCM LoRA is unrelated to a Colab-vs-EC2 deployment problem." + "reason": "IP-Adapter face/LCM-LoRA incompatibility is unrelated to environment-specific Colab-vs-EC2 runtime behavior." }, { "left": "issue:11247", "right": "issue:11298", "accept": false, - "reason": "Multi-GPU FluxTransformer2DModel distribution with ControlNet is a different issue from PEFT key errors while hotswapping LoRAs." + "reason": "Both mention Flux/controlnet/LoRA adjacent areas, but the failures are different: multi-GPU transformer distribution versus LoRA hotswapping PEFT keys." }, { "left": "issue:10674", "right": "issue:11370", "accept": false, - "reason": "GGUF incompatibility in FluxPipeline is unrelated to a conversion script hitting an invalid pickle load key." + "reason": "Flux GGUF support problem is unrelated to a video conversion script failing to unpickle a file." } ] }, @@ -32332,10 +32359,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -32350,163 +32377,166 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose collection of Diffusers issues across different pipelines, models, and failure modes. The soft-similarity pairs are mostly symptom-level matches rather than the same underlying bug, so none should be merged as duplicates. If one issue has to anchor the set, #11343 is the broadest import-regression report.", - "confidence": 0.42, - "canonical_issue_reason": "No single true duplicate stands out; #11343 is the broadest representative because it concerns a general import regression affecting multiple Diffusers modules.", + "summary": "This cluster is mostly heterogeneous. The only strong duplicate-looking pair is the LPW pipeline compatibility issues; the rest are different bugs, features, or model-specific failures.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 1536 is the best representative because it states the core LPW pipeline compatibility problem in a general way; issue 1616 reads like the version-specific sibling of the same bug.", "canonical_pr_reason": null, - "best_issue_reason": "#11343 is the best global representative because it captures a broad, user-visible regression around importing Diffusers after a dependency update.", + "best_issue_reason": "Issue 1536 is the clearest, most general duplicate target in the set and best captures the shared LPW/diffusers incompatibility theme.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1251", "right": "issue:1614", "accept": false, - "reason": "Both mention black images, but one is MPS RePaintPipeline-specific and the other is SD 2.1 autocast-specific; different triggers and likely different root causes." + "reason": "Both mention black images, but one is RePaint on MPS and the other is SD 2.1 autocast; different pipelines and likely different root causes." }, { "left": "issue:307", "right": "issue:4080", "accept": false, - "reason": "One is a proposal to merge pipelines, the other is a request to add an SDXL inpainting pipeline; not the same bug or change." + "reason": "One is a broad request to merge pipelines; the other asks for an SDXL inpainting pipeline. Not the same bug/change." }, { "left": "issue:4306", "right": "issue:6382", "accept": false, - "reason": "List-index failure in ReferencePipeline and IP-Adapter key error in SDXL training/inference are unrelated code paths." + "reason": "Different failures: reference pipeline list-index error vs IP-Adapter key error in an SDXL/LCM training workflow." }, { "left": "issue:4488", "right": "issue:5621", "accept": false, - "reason": "One is an SDXL checkpoint-loading issue, the other is ControlNet encoder_hid_proj support; both are model-loading topics but not the same defect." + "reason": "Both involve SDXL support, but one is checkpoint loading via from_original_ckpt and the other is ControlNet encoder_hid_proj support; distinct issues." }, { "left": "issue:4504", "right": "issue:4641", "accept": false, - "reason": "Both involve LoRA loading, but one is a state_dict assignment error and the other is native LoRA loading from a local file; different bugs." + "reason": "Both involve LoRA loading, but one is a safetensor state_dict assignment bug and the other is local-file load_lora_weights behavior." }, { "left": "issue:5152", "right": "issue:5524", "accept": false, - "reason": "W\u00fcrstchen alpha bug and an AttributeError in LoRA loading are unrelated." + "reason": "Same model family, but one is a general Wurstchen alpha bug and the other is a specific lora_layer attribute error." }, { "left": "issue:5886", "right": "issue:8469", "accept": false, - "reason": "Feature request for IP-Adapter face model support is not the same as broken face outputs from a specific face model." + "reason": "Feature request for face-model support vs a report of corrupt faces from an existing face model; not the same underlying change." }, { "left": "issue:5889", "right": "issue:6460", "accept": false, - "reason": "Stable video diffusion and endless loading of an inpainting model are different pipeline problems." + "reason": "Unrelated topics: stable video diffusion vs loading stable-diffusion-inpainting hanging." }, { "left": "issue:8440", "right": "issue:8873", "accept": false, - "reason": "Subclass ValueError in FromOriginalModelMixin and xformers performance slowdown are unrelated." + "reason": "One is a mixin/subclass ValueError; the other is xformers performance regression. Different code paths." }, { "left": "issue:4185", "right": "issue:5997", "accept": false, - "reason": "Both touch SDXL ControlNet, but one is black-image inference after rebasing and the other is a tutorial type mismatch; not the same underlying bug." + "reason": "Black-image inference bug with SDXL ControlNet vs a tutorial/model-type mismatch error; not the same bug." }, { "left": "issue:3045", "right": "issue:7483", "accept": false, - "reason": "Both concern LoRA conversion, but one is a generic safetensors-to-diffusers conversion request and the other is an SDXL-base-model-specific script failure." + "reason": "Both are LoRA-to-Diffusers conversion related, but one is a general conversion request and the other is an SDXL-base-specific failure." }, { "left": "issue:12052", "right": "issue:9895", "accept": false, - "reason": "Both are device-mismatch failures, but they occur in different models and code paths (Wan 2.2 offloading vs FluxPipeline after a commit)." + "reason": "Both are device-mismatch/offloading failures, but they affect different pipelines/models and likely different fixes." }, { "left": "issue:4109", "right": "issue:7421", "accept": false, - "reason": "Both are SDXL-related, but one is an inpainting pipeline not working and the other is a seed reproducibility issue in KDiffusionPipeline." + "reason": "Both mention SDXL pipelines, but one is an inpaint pipeline failure and the other is a seed/reproducibility issue in K-Diffusion." }, { "left": "issue:8546", "right": "issue:8579", "accept": false, - "reason": "Missing from-single-file support and broken SD3 LoRA support are separate features/issues." + "reason": "One is missing from-single-file support for SD3; the other is broken SD3 LoRA support. Different features." }, { "left": "issue:11199", "right": "issue:11343", "accept": false, - "reason": "Both involve import problems, but one is a specific onnx_utils import failure while the other is a broader module-import regression after a transformers update." + "reason": "Both are import problems, but one is a specific onnx_utils import and the other is a broad post-transformers-update import failure." }, { "left": "issue:2609", "right": "issue:4460", "accept": false, - "reason": "Canny ControlNet checkpoint support and differing outputs between Stable Diffusion and Diffusers are not the same bug." + "reason": "Canny ControlNet checkpoint request vs a mismatch in outputs between stable diffusion and diffusers; unrelated." }, { "left": "issue:3154", "right": "issue:3223", "accept": false, - "reason": "Both relate to memory handling, but one is failed offloading when RAM is full and the other is RAM not clearing after moving to CUDA." + "reason": "Both are memory/offloading complaints, but one is disk offloading when RAM is full and the other is RAM not clearing after moving to CUDA." }, { "left": "issue:3540", "right": "issue:4667", "accept": false, - "reason": "TextToVideoZero prompt-embed support and from_single_file loading failure are unrelated." + "reason": "Different APIs and failures: prompt_embeds support in TextToVideoZeroPipeline vs from_single_file loading regression." }, { "left": "issue:2156", "right": "issue:3613", "accept": false, - "reason": "Dreambooth using CPU instead of GPU versus Dreambooth OOM with torch 2.0 are opposite training issues, not duplicates." + "reason": "One reports CPU-only training; the other reports GPU OOM in Dreambooth. Opposite symptoms, different issues." }, { "left": "issue:1125", "right": "issue:1925", "accept": false, - "reason": "Stable Diffusion Mega inpaint runtime error and SD 2 inpainting half-precision failure are different pipeline/version-specific problems." + "reason": "Both are inpainting/half-precision related, but one is a runtime error in a specific mega pipeline and the other is SD2 inpainting half-precision not working." }, { "left": "issue:5630", "right": "issue:5703", "accept": false, - "reason": "LMS scheduler noise in SDXL img2img and inability to reconstruct noise with DDIM inverse are different scheduler behaviors." + "reason": "Both concern sampling/output quality, but one is LMS scheduler noise in SDXL img2img and the other is DDIM inverse noise reconstruction." }, { "left": "issue:7053", "right": "issue:9515", "accept": false, - "reason": "UNet3D class conditioning and missing RMSNorm in SD3 joint attention are different model-architecture issues." + "reason": "Different model internals: class conditioning in UNet3DConditionModel vs missing RMSNorm in SD3Transformer2DModel attention." }, { "left": "issue:1524", "right": "issue:3346", "accept": false, - "reason": "Textual inversion training quality and DreamBooth checkpoint resume support are unrelated." + "reason": "Textual inversion training quality vs Dreambooth resume_from_checkpoint support; unrelated training concerns." }, { "left": "issue:1536", "right": "issue:1616", - "accept": false, - "reason": "Both mention LPW pipelines, but one is incompatibility with a new Diffusers version and the other is a community pipeline on 0.8.0; not the same concrete bug." + "accept": true, + "reason": "Both describe the LPW pipeline breaking against newer diffusers versions; same underlying compatibility bug and plausible one-PR fix." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative on the soft-edge pairs. The LPW pair (1536/1616) is the only clearly duplicate-looking match from the titles, and the other rejections do not overclaim sameness based on the available evidence." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -32601,10 +32631,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -32619,156 +32649,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: it mixes unrelated issues across models, pipelines, training scripts, loading paths, and hardware/offload behaviors. I don\u2019t see any soft pair that clearly represents the same underlying bug/change.", - "confidence": 0.95, + "summary": "This cluster is highly heterogeneous; the soft-similarity pairs are mostly keyword matches around training, loading, LoRA, or device movement, but they do not describe the same underlying bug/change. I would reject all candidate duplicate links.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:10421 is the broadest and clearest representative of a concrete runtime/device-management bug in the set, but the cluster is too mixed to treat it as a true duplicate anchor.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:2324", "right": "issue:761", "accept": false, - "reason": "Both involve DreamBooth training, but one is about instance data loading and the other about 8-bit Adam; different failure modes." + "reason": "Both are training-related, but one is about Dreambooth instance data loading and the other about 8-bit Adam optimizer failure; different bug surfaces." }, { "left": "issue:8885", "right": "issue:9421", "accept": false, - "reason": "Unrelated topics: custom diffusion token/clip-retrieval setup vs accessing a specific model repository." + "reason": "One is a Custom Diffusion/clip-retrieval token-loading setup issue, the other is model access permissions for an inpainting repo; unrelated failures." }, { "left": "issue:4389", "right": "issue:6196", "accept": false, - "reason": "Both concern ControlNet loading, but they target different model variants and likely different code paths." + "reason": "Both mention ControlNet loading, but they concern different model classes and error conditions (ControlNetModel vs ControlNetXS Depth SDXL)." }, { "left": "issue:4697", "right": "issue:9187", "accept": false, - "reason": "Different features/models: a specific LoRA load failure vs IP-Adapter breakage after a Kolors update." + "reason": "Different components and regressions: specific LoRA loading after 0.20.0 vs IP-Adapter SDXL breakage after Kolors update." }, { "left": "issue:1717", "right": "issue:4303", "accept": false, - "reason": "Offline cache loading bug vs Flax SDXL loading crash; not the same problem." + "reason": "Offline cache loading failure and Flax SDXL crash are unrelated code paths and error modes." }, { "left": "issue:4340", "right": "issue:5016", "accept": false, - "reason": "Both are SDXL LoRA/text-encoder training issues, but they reference different scripts and symptoms; not enough to call them the same bug." + "reason": "Both involve SDXL text encoders, but they are separate training scripts and different failure reports; not the same concrete bug." }, { "left": "issue:4852", "right": "issue:7793", "accept": false, - "reason": "Generic LoRA-loading error vs unloading LoRA weights in DeepFloyd; different workflows and failures." + "reason": "One is a LoRA loading error, the other is an unload/cleanup error for DeepFloyd LoRA weights; opposite lifecycle stages and different bugs." }, { "left": "issue:3296", "right": "issue:4664", "accept": false, - "reason": "Checkpoint save/load with text_encoder vs SDXL LoRA loading behavior; related domain, but different concrete bugs." + "reason": "Checkpoint save/load with text_encoder vs SDXL LoRA loading on two pipeline class names; not the same issue." }, { "left": "issue:4306", "right": "issue:7851", "accept": false, - "reason": "Reference pipeline list-index error vs concurrency out-of-bounds error; different symptoms and contexts." + "reason": "Reference pipeline list-index error and concurrent-call out-of-bounds error are distinct bugs." }, { "left": "issue:11702", "right": "issue:13001", "accept": false, - "reason": "Both involve model-loading errors, but one is a LoRA key error and the other is a Flux2.Klein GGUF shape mismatch." + "reason": "AccVid LoRA key mismatch and Flux2.Klein GGUF shape mismatch are different models and different loading failures." }, { "left": "issue:10421", "right": "issue:10653", "accept": false, - "reason": "Memory leak when moving pipelines across GPUs vs bitsandbytes 8-bit `.to()` unsupported on a specific transformer model." + "reason": "Both are device-movement issues, but one reports a CPU memory leak moving pipelines across GPUs and the other is a bitsandbytes .to(cpu) unsupported error; not the same bug." }, { "left": "issue:10392", "right": "issue:11418", "accept": false, - "reason": "Different Flux requests: using Flux-dev-fp8 with a LoRA vs asking how to add a specific fp8 safetensors file." + "reason": "Flux fp8 model usage versus adding a Flux fill fp8 safetensors file; similar model family only, not the same failure." }, { "left": "issue:12657", "right": "issue:13140", "accept": false, - "reason": "Quantized Qwen Image Edit failure vs Qwen Edit Plus/Pipelines TypeError; same family, but not the same concrete issue." + "reason": "Qwen image-edit quantization regression and Qwen Edit pipeline NoneType error are different symptoms and likely different code paths." }, { "left": "issue:3454", "right": "issue:3651", "accept": false, - "reason": "Both are DreamBooth training problems, but one is a latest-version compatibility issue and the other is multi-GPU support." + "reason": "train_dreambooth_lora.py regression on latest and multi-GPU Dreambooth failure are separate problems." }, { "left": "issue:11182", "right": "issue:9527", "accept": false, - "reason": "BF16 variant download inefficiency vs dtype mismatch between controlnet fp32 and main pipeline bf16; different bugs." + "reason": "Unnecessary BF16 download and dtype mismatch between ControlNet fp32 and main pipeline bf16 are different issues." }, { "left": "issue:5153", "right": "issue:5442", "accept": false, - "reason": "Distinct research/features: FreeU vs ScaleCrafter, not the same underlying change." + "reason": "FreeU is an algorithm/pipeline feature request, while ScaleCrafter is a different pipeline proposal; no shared bug." }, { "left": "issue:2185", "right": "issue:3871", "accept": false, - "reason": "Training config files missing vs checkpoint resume error; unrelated training setup issues." + "reason": "Missing config files for training and checkpoint-resume training error are unrelated training setup issues." }, { "left": "issue:7133", "right": "issue:7431", "accept": false, - "reason": "Both are DreamBooth/text-encoder training issues, but one is checkpoint resume size mismatch and the other is DoRA-specific training behavior." + "reason": "Both mention text encoders in training, but one is a resume size-mismatch bug and the other is a DoRA training issue; different concrete defects." }, { "left": "issue:11135", "right": "issue:3826", "accept": false, - "reason": "Completely different proposed pipelines: SmoothCache acceleration vs Audio-Journey." + "reason": "Both are new pipeline proposals, but they are entirely different pipelines and changes." }, { "left": "issue:2659", "right": "issue:6949", "accept": false, - "reason": "Both affect generation outputs, but one is batch-size variability and the other is even-step blank images from a specific scheduler." + "reason": "Batch-size variation and even-step blank images are different generation correctness bugs." }, { "left": "issue:12499", "right": "issue:12529", "accept": false, - "reason": "Both are Wan 2.2 shape mismatch reports, but one is I2V conditioning and the other is tiled VAE tensor sizing." + "reason": "Both are Wan 2.2 shape issues, but one is conditioning shape mismatch and the other is tiled VAE tensor-size mismatch; different components." }, { "left": "issue:3871", "right": "issue:6821", "accept": false, - "reason": "Checkpoint resume training error vs an invalid caption_column configuration; different root causes." + "reason": "Checkpoint resume error and caption_column validation error are unrelated training CLI issues." }, { "left": "issue:12011", "right": "issue:12052", "accept": false, - "reason": "Same model family, but one is an OOM and the other is a device-mismatch failure under offloading." + "reason": "Wan 2.2 OOM from large I2V loading and LightX2V offloading device-mismatch failure are different runtime problems." }, { "left": "issue:4194", "right": "issue:5150", "accept": false, - "reason": "Both mention from_single_file, but one is a VAE loading problem and the other is a missing method on a different pipeline." + "reason": "One is a from_single_file loading bug, the other is a missing from_single_file method on a different pipeline class; not the same change." } ] }, @@ -32873,10 +32903,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -32891,174 +32921,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly noisy soft-similarity matches across unrelated Diffusers topics (pipelines, training scripts, schedulers, PEFT, loading/conversion). Only one pair looks like the same underlying bug: the missing LatentDiffusionPipeline symbol. All other soft edges should be rejected.", - "confidence": 0.94, - "canonical_issue_reason": "Issue 50 is the clearest canonical target because it states the missing LatentDiffusionPipeline problem in the most direct, generic form; issue 1055 is the same report with slightly different wording.", + "summary": "This cluster is mostly heterogeneous. Two soft edges look like real duplicates: the missing `LatentDiffusionPipeline` symbol pair and the SDXL diffusers-to-original-format conversion/loading pair. The rest are only loosely related by broad subsystems like DreamBooth, SDXL, schedulers, or video pipelines and should stay separate.", + "confidence": 0.78, + "canonical_issue_reason": "issue:50 is the clearest canonical duplicate anchor: it states the missing `LatentDiffusionPipeline` symbol directly, and issue:1055 is the same bug phrased as \u201cNo LatentDiffusionPipeline on the latest branch.\u201d", "canonical_pr_reason": null, - "best_issue_reason": "Issue 50 is the best issue to anchor the only clear duplicate subgroup in this set; it is older and more canonical than 1055. The rest of the issues are not close enough to form one duplicate cluster.", + "best_issue_reason": "issue:50 is the best overall issue to anchor the cluster because it is the most explicit, self-contained bug report and has the strongest duplicate evidence among the issues shown.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:2830", "right": "issue:7333", "accept": false, - "reason": "Different image-to-video requests: one is about using ModelScope video models, the other about a specific Tuning-Free Noise Rectification paper/approach." + "reason": "Both are about video generation, but one is a capability question for ModelScope while the other is a specific image-to-video method/paper; not the same bug or change." }, { "left": "issue:3706", "right": "issue:6460", "accept": false, - "reason": "A model announcement/request (Wuerstchen) is unrelated to an endless-loading bug when loading runwayml/stable-diffusion-inpainting." + "reason": "Unrelated topics: Wuerstchen model announcement vs. an endless-loading bug while loading a Stable Diffusion inpainting model." }, { "left": "issue:4779", "right": "issue:7333", "accept": false, - "reason": "ModelScope image-to-video pipeline request and Tuning-Free Noise Rectification issue are different model/problem statements." + "reason": "Both mention image/video generation, but one is a pipeline request and the other is a specific image-to-video research issue; no shared concrete failure." }, { "left": "issue:11148", "right": "issue:12012", "accept": false, - "reason": "PEFT adapter flag handling for LoRA loading is unrelated to Wan 2.2 apply_first_block_cache context-state errors." + "reason": "Different code paths: PEFT LoRA adapter state flag handling vs. Wan 2.2 first-block cache context errors." }, { "left": "issue:1363", "right": "issue:2534", "accept": false, - "reason": "Both mention schedulers, but one is an ONNX long-prompt/LMS issue and the other is a DPM vs UniPC/DEIS scheduler error in a different workflow." + "reason": "Both are scheduler-related errors, but the affected pipelines and failure modes differ; too broad to be the same underlying bug." }, { "left": "issue:1013", "right": "issue:3970", "accept": false, - "reason": "StableDiffusionMegaPipeline on Diffusers 6.0 and SDXL prompt embeddings failure are different pipeline/code-path problems." + "reason": "Separate SD pipeline problems: one is StableDiffusionMegaPipeline compatibility, the other is SDXL prompt embeddings not working." }, { "left": "issue:2324", "right": "issue:2465", "accept": false, - "reason": "Dreambooth instance-data bug and a request for a reproducible training recipe are not the same issue." + "reason": "Dreambooth is the shared theme, but one is a concrete data-loading bug and the other is a request for a reproducible training recipe." }, { "left": "issue:4236", "right": "issue:5433", "accept": false, - "reason": "SDXL refiner text-encoder registration and DPM++ artifacts without refiner are distinct SDXL problems." + "reason": "One is SDXL Refiner text-encoder registration, the other is visual artifacts with DPM++ schedulers; not the same defect." }, { "left": "issue:5436", "right": "issue:6092", "accept": false, - "reason": "ControlNet XL training example failure and LoRA example not training anything are different scripts and failure modes." + "reason": "Different training failures in different scripts: ControlNet XL example error vs. LoRA example not training anything." }, { "left": "issue:11006", "right": "issue:12012", "accept": false, - "reason": "Wan 2.1 quantized-transformer video corruption and Wan 2.2 cache-context ValueError are different code paths." + "reason": "Wan 2.x video output corruption with quantized transformer is not the same as the first-block-cache context ValueError." }, { "left": "issue:1565", "right": "issue:8939", "accept": false, - "reason": "GPU selection/OOM behavior is unrelated to a tokenizer-loading error for Stable Diffusion 3 Medium." + "reason": "GPU selection/OOM handling is unrelated to a tokenizer-loading failure for Stable Diffusion 3." }, { "left": "issue:7682", "right": "issue:9780", "accept": false, - "reason": "Multi-GPU SD1.5 runtime issues and a SD3.5 norm_added_k attribute error are unrelated." + "reason": "Multi-GPU SD1.5 runtime behavior is unrelated to an SD3.5 model attribute error during loading." }, { "left": "issue:8277", "right": "issue:8477", "accept": false, - "reason": "Both are consistency-training related, but one is non-convergence and the other is a DistributedDataParallel script error; not the same bug." + "reason": "Both mention consistency training, but one is convergence and the other is a DistributedDataParallel script error; different problems." }, { "left": "issue:4634", "right": "issue:4704", "accept": false, - "reason": "Safe serialization for Custom Diffusion and a failing Custom Diffusion training script are different problems." + "reason": "Both are Custom Diffusion, but one is safe serialization and the other is the training script failing; not the same code-path bug." }, { "left": "issue:5219", "right": "issue:6934", - "accept": false, - "reason": "One is a loading failure for a converted SDXL safetensors model; the other is broken conversion to original SDXL format. Related area, but not the same concrete bug." + "accept": true, + "reason": "These describe the same SDXL diffusers conversion issue from opposite ends: conversion from diffusers to original SDXL format is broken, leading to inability to load the converted safetensors model." }, { "left": "issue:5395", "right": "issue:5628", "accept": false, - "reason": "PEFT LoRA-combining bug and LMS/Karras denoising offset issue are unrelated." + "reason": "Both involve LoRA/DPM-style topics only loosely, but they concern different scheduler/combination behaviors and are not the same bug." }, { "left": "issue:5012", "right": "issue:6297", "accept": false, - "reason": "Both involve SDXL LoRA training, but one is about text-encoder training support and the other is a general 'can\u2019t train on my photos' report." + "reason": "Both are LoRA training issues, but one is about SDXL text-encoder training in a script and the other says a DreamBooth LoRA training version cannot learn from photos." }, { "left": "issue:1261", "right": "issue:3584", "accept": false, - "reason": "Xformers backward-pass failure and a DreamBooth tutorial failure are not the same bug." + "reason": "Xformers backward-pass assertion is unrelated to a DreamBooth tutorial failure." }, { "left": "issue:2361", "right": "issue:6224", "accept": false, - "reason": "Checkpoint-saving crash in finetuning and a TorchDynamo issue during finetuning are different failures." + "reason": "Both are training/checkpointing issues, but one is a checkpoint save crash and the other is a TorchDynamo fine-tuning problem." }, { "left": "issue:1055", "right": "issue:50", "accept": true, - "reason": "Same missing LatentDiffusionPipeline symbol/attribute problem, phrased two ways; this is a clear duplicate pair." + "reason": "Clear duplicate: both report `LatentDiffusionPipeline` missing/not available, with only wording differences." }, { "left": "issue:1261", "right": "issue:4117", "accept": false, - "reason": "Xformers attention-backward failure and SDXL Dreambooth issues are unrelated." + "reason": "Xformers attention backward assertion is not the same as SDXL Dreambooth not working." }, { "left": "issue:3468", "right": "issue:5259", "accept": false, - "reason": "Missing model_index.json during Dreambooth inference and Dreambooth not generating model_index.json are adjacent but not the same concrete report." + "reason": "Dreambooth missing `model_index.json` vs. a generic missing `model_index.json` load error are related by file name but not clearly the same underlying bug from the titles alone." }, { "left": "issue:3984", "right": "issue:785", "accept": false, - "reason": "Class-label conditioning error and PNDMScheduler cost/performance issue are unrelated." + "reason": "Class-label requirement in a model config is unrelated to PNDMScheduler performance cost." }, { "left": "issue:4134", "right": "issue:5607", "accept": false, - "reason": "MPS ControlNet tensor-shape mismatch and a generic tensor-size mismatch in another context are not clearly the same code-path." + "reason": "MPS ControlNet tensor-shape mismatch is different from a general tensor-size mismatch involving another pipeline." }, { "left": "issue:4303", "right": "issue:8939", "accept": false, - "reason": "Flax SDXL load crash and SD3 tokenizer loading failure are different backend/model problems." + "reason": "Both are import/load failures, but one is Flax SDXL loading and the other is tokenizer loading for SD3; different code paths." }, { "left": "issue:4969", "right": "issue:9167", "accept": false, - "reason": "Importing CrossAttention in Diffusers v0.20.1 and importing StableDiffusion3Pipeline are different import/API failures." + "reason": "ImportError for `CrossAttention` and inability to import `StableDiffusion3Pipeline` are distinct missing-symbol issues." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet and stays conservative: only issue 50/1055 is accepted as a duplicate pair, matching the one deterministic accept edge, and the other soft-edge rejections are consistent with the issue titles/body context. No overstatement detected." + "feedback": "Grounded overall. The canonical anchor on issue:50 is supported by the matching LatentDiffusionPipeline missing-symbol reports in issue:1055, and the SDXL diffusers-conversion pair (issue:5219/6934) is a plausible duplicate relation from the titles. The remaining soft-edge rejections are conservative and stay within what the packet supports." }, "evaluator_used": true, "retried": false, @@ -33157,10 +33187,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -33175,156 +33205,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The items are mostly unrelated bug reports and feature requests across different pipelines/features. None of the proposed soft pairs look like the same underlying bug or change, so all soft-edge links should be rejected.", - "confidence": 0.94, - "canonical_issue_reason": "issue:10382 is a precise, concrete bug report with a clear failing code path (VAE encoder forward return type mismatch causing AttributeError), making it the strongest single issue candidate among this set.", + "summary": "These items are mostly unrelated Diffusers issues/feature requests. The proposed soft pairs share broad wording or subsystem names, but not the same concrete bug or change, so none should be merged as duplicates.", + "confidence": 0.18, + "canonical_issue_reason": "Issue 10382 is the most concrete bug report in the set: it describes a specific runtime type mismatch in the VAE encoder forward path with an actionable failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10382 is specific, actionable, and clearly scoped to one bug, which makes it suitable as the representative issue in this cluster.", + "best_issue_reason": "Issue 10382 is the best overall representative because it is specific, reproducible, and clearly bug-oriented, whereas most other items are broader troubleshooting, feature requests, or different failures.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:5150", "right": "issue:5458", "accept": false, - "reason": "One is missing from_single_file support; the other is lora_state_dict ignoring local_files_only. Related area, but different bugs." + "reason": "Both concern loader APIs, but one is about missing from_single_file support and the other is about local_files_only behavior; different bugs." }, { "left": "issue:5515", "right": "issue:6790", "accept": false, - "reason": "LCM pipeline init-without-scheduler versus training state_dict key mismatch are different failure modes and code paths." + "reason": "One is a pipeline init failure without a scheduler, the other is a training state_dict key mismatch; not the same issue." }, { "left": "issue:7223", "right": "issue:7333", "accept": false, - "reason": "Both are video-generation related feature requests, but they target different models/features and are not the same change." + "reason": "Both are video-model related, but one is a feature request to add Latte and the other is about a specific image-to-video method; different changes." }, { "left": "issue:1643", "right": "issue:730", "accept": false, - "reason": "Both involve DreamBooth, but one is an xformers method signature error and the other is local training quality differences; not the same bug." + "reason": "Both mention DreamBooth, but the failures are unrelated: an xformers method signature error versus training quality/behavior on local vs Colab." }, { "left": "issue:8787", "right": "issue:9613", "accept": false, - "reason": "Slow model loading is unrelated to the Timesteps.scale attribute error." + "reason": "Slow model loading is unrelated to the Timesteps.scale attribute error; different code paths and symptoms." }, { "left": "issue:9096", "right": "issue:9768", "accept": false, - "reason": "Both mention NaNs, but the reported environments and likely root causes differ (Flux half precision vs torch 2.5 SDPA/CuDNN)." + "reason": "Both mention NaNs, but one is Flux half-precision inference and the other is torch 2.5 SDPA/CuDNN; not the same bug." }, { "left": "issue:11136", "right": "issue:8788", "accept": false, - "reason": "SanaSprint intermediate timestep support is unrelated to SD3 VAE tiling behavior." + "reason": "These are different pipeline support issues: SanaSprint intermediate timesteps vs SD3 VAE tiling on non-power-of-two images." }, { "left": "issue:4104", "right": "issue:6492", "accept": false, - "reason": "Both mention black outputs, but they appear to be separate image-generation issues with different contexts and symptoms." + "reason": "Both produce black images, but the reported causes and contexts differ and there is no evidence of the same underlying defect." }, { "left": "issue:3325", "right": "issue:4639", "accept": false, - "reason": "CUDA invalid argument during DreamBooth training is unrelated to multiprocessing data-loader errors in a different script." + "reason": "One is a CUDA invalid-argument error in DreamBooth training; the other is a dataloader multiprocessing error in a different script." }, { "left": "issue:12012", "right": "issue:9856", "accept": false, - "reason": "Wan 2.2 context-state error is unrelated to distributed port-in-use errors." + "reason": "A missing context for block cache state is unrelated to a distributed port collision." }, { "left": "issue:10382", "right": "issue:8960", "accept": false, - "reason": "VAE forward return type mismatch is unrelated to gradient accumulation being nonfunctional in a training example." + "reason": "A VAE forward return-type mismatch is unrelated to gradient accumulation not working." }, { "left": "issue:10986", "right": "issue:11709", "accept": false, - "reason": "ComplexDouble tensors on MPS and CausVid LoRA key errors are different bugs affecting different components." + "reason": "ComplexDouble tensors on MPS and a LoRA key error are different failures in different parts of the loading/inference stack." }, { "left": "issue:1643", "right": "issue:4551", "accept": false, - "reason": "DreamBooth xformers signature error is not the same as SDXL dreambooth crash/loading issue." + "reason": "Both involve DreamBooth, but one is an xformers API mismatch and the other is a crash with SDXL; not the same bug." }, { "left": "issue:11062", "right": "issue:11127", "accept": false, - "reason": "Loading a specific LoRA error and a single-file Flux model incompatibility are distinct model-loading problems." + "reason": "Both mention CivitAI/Flux loading, but one is a specific LoRA key error and the other is a single-file incompatibility for a different model." }, { "left": "issue:2081", "right": "issue:2673", "accept": false, - "reason": "JAX conversion failures and StableDiffusionUpscalePipeline config/input verification errors are different issues." + "reason": "Converting PyTorch weights to JAX is unrelated to StableDiffusionUpscalePipeline config validation." }, { "left": "issue:10823", "right": "issue:7146", "accept": false, - "reason": "Fine-tuning script output files and SDXL CUDA OOM are unrelated training issues." + "reason": "Missing inference files from a fine-tuning script and SDXL training OOM are different problems." }, { "left": "issue:4255", "right": "issue:6598", "accept": false, - "reason": "Different-results-quality complaint versus img2img parity complaint; both are output-quality, but not the same bug." + "reason": "Different-results complaints are too broad here; one compares Diffusers to sd-webui generally, the other is a specific img2img parity report." }, { "left": "issue:2161", "right": "issue:3174", "accept": false, - "reason": "Inpainting LoRA training and LoRA/dreambooth loading regression are different workflows and failures." + "reason": "One is about LoRA training on an inpainting model; the other is a loading regression caused by new LoRA features." }, { "left": "issue:10963", "right": "issue:8432", "accept": false, - "reason": "AutoencoderKLWan import failure and FromOriginalVAEMixin import failure are separate import/export issues." + "reason": "Both are import errors around autoencoder-related symbols, but they target different modules and symbols; not clearly the same bug." }, { "left": "issue:11359", "right": "issue:11569", "accept": false, - "reason": "Both concern LTX-Video, but one is a performance feature request and the other is a repo packaging request." + "reason": "Both are LTX-Video feature requests, but one asks for faster inference and the other asks for a diffusers repo release; different requests." }, { "left": "issue:1434", "right": "issue:4788", "accept": false, - "reason": "Both are memory/OOM reports, but they involve different pipelines, image sizes, and contexts." + "reason": "Both are OOM reports, but one is StableDiffusionUpscalePipeline at larger sizes and the other is SDXL inpaint on 6 GB VRAM; not the same failure." }, { "left": "issue:4724", "right": "issue:5558", "accept": false, - "reason": "SDXL img2img behavior with a checkpoint is unrelated to ControlNet inpaint artifacts with multiple ControlNets." + "reason": "Both touch SDXL/control-like workflows, but one is an img2img checkpoint behavior issue and the other is artifacts with multiple ControlNets." }, { "left": "issue:6517", "right": "issue:9055", "accept": false, - "reason": "Both are import errors, but they involve different mixins, modules, and failure points." + "reason": "These are different import errors from different loaders and different symbols; no shared code-path is evident." }, { "left": "issue:5016", "right": "issue:5840", "accept": false, - "reason": "Training text encoder not working and resume-from-checkpoint freezing are different SDXL DreamBooth LoRA issues." + "reason": "Both are SDXL DreamBooth LoRA issues, but one is about training text encoders and the other is about checkpoint resume/freezing; not the same bug." } ] }, @@ -33426,10 +33456,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -33444,156 +33474,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issues with superficial keyword overlap (loading, training, Flux, LoRA, video). I don\u2019t see a true duplicate cluster or a shared code-path change that would justify merging any soft pair.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue fits the cluster cleanly; the items span distinct bugs and feature requests rather than one underlying defect. If forced to pick a representative, the Flux device-mismatch regression issues are the closest family.", + "summary": "These items are mostly distinct reports across different pipelines, training scripts, and feature requests. The soft-similarity pairs are largely broad false positives; even the Flux-related pair is too platform-specific/context-dependent to treat as the same bug.", + "confidence": 0.89, + "canonical_issue_reason": "No canonical issue: the collection is heterogeneous, spanning unrelated bugs and feature requests rather than one underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9895 is the strongest representative if one is needed: it is a concrete, recent FluxPipeline regression with a specific device-placement failure and clear reproduction context.", + "best_issue_reason": "Issue 9895 is the strongest standalone report in the set: it has a precise FluxPipeline device-mismatch regression, a concrete error message, and clear reproduction context.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:2728", "right": "issue:4250", "accept": false, - "reason": "Both are loading failures, but one is UNet2DConditionModel safetensors loading and the other is a broader from_pretrained misload; different code paths and symptoms." + "reason": "Both are model-loading complaints, but one is UNet safetensors loading and the other is a generic from_pretrained issue; too broad to be the same bug." }, { "left": "issue:4892", "right": "issue:5760", "accept": false, - "reason": "Both are video-related, but TokenFlow video editing and video-to-audio synthesis are unrelated problems and changes." + "reason": "Both are video-related diffusion projects, but they concern entirely different systems and goals." }, { "left": "issue:4389", "right": "issue:4911", "accept": false, - "reason": "ControlNetModel loading and custom text_encoder loading in from_single_file are different components and failure modes." + "reason": "Loading failures, but different code paths: ControlNetModel loading vs custom text_encoder loading from_single_file." }, { "left": "issue:222", "right": "issue:3453", "accept": false, - "reason": "Generic training runtime failure versus a specific BFloat16 triu_tril CUDA limitation in text-to-image training; not the same bug." + "reason": "Both are training runtime errors, but the reported failures and causes are unrelated." }, { "left": "issue:7056", "right": "issue:9415", "accept": false, - "reason": "Both mention parameter-related errors, but SDXL LoRA train_text_encoder and CogVideoXTransformer3DModel module parameters are unrelated code paths." + "reason": "Both mention errors after adding parameters, but they target different models/APIs and do not look like one bug." }, { "left": "issue:2132", "right": "issue:3363", "accept": false, - "reason": "Both involve DreamBooth/LoRA, but missing Accelerator.trackers is a dependency/API issue, while 'failed on two machines' is broader and not the same defect." + "reason": "Both involve DreamBooth/LoRA training, but one is an Accelerator attribute error and the other is a machine-specific failure." }, { "left": "issue:9515", "right": "issue:9810", "accept": false, - "reason": "SD3Transformer2DModel RMSNorm omission and LoRA module_to_save() support are different implementation gaps." + "reason": "These are different feature/implementation issues in different subsystems; no shared failing code path." }, { "left": "issue:3450", "right": "issue:5931", "accept": false, - "reason": "from_ckpt loading failure and nondeterministic generation are distinct problems." + "reason": "One is a from_ckpt loading problem, the other is nondeterministic generation; not the same underlying issue." }, { "left": "issue:5605", "right": "issue:6354", "accept": false, - "reason": "These are separate feature requests: per-image captions for SDXL Dreambooth vs LoRA support for ControlNet." + "reason": "Both are LoRA-related feature requests, but they target different capabilities (SDXL DreamBooth captions vs ControlNet support)." }, { "left": "issue:7000", "right": "issue:9096", "accept": false, - "reason": "PyTorch deprecation warning in outputs.py and Flux half-precision NaNs are unrelated." + "reason": "A deprecation warning and a NaN inference bug are different problems, even if both touch Flux/torch internals." }, { "left": "issue:4030", "right": "issue:8435", "accept": false, - "reason": "Stable Diffusion v2.1 embedding runtime error and from_single_file runtime error are not the same code-path failure." + "reason": "Both are runtime errors in Stable Diffusion paths, but the entrypoints and failure modes differ." }, { "left": "issue:10049", "right": "issue:6636", "accept": false, - "reason": "Cannot import MochiPipeline and missing DIFFUSERS_SLOW_IMPORT are different import-symbol issues." + "reason": "Both are import errors, but they are for different symbols and different modules." }, { "left": "issue:10446", "right": "issue:9450", "accept": false, - "reason": "Both involve Flux/multi-GPU, but FluxControlNetPipeline sharding incompatibility and a transformer= device mismatch are different concrete bugs." + "reason": "Both mention Flux and multi-device behavior, but one is FluxControlNet sharding and the other is a FluxPipeline device-placement mismatch." }, { "left": "issue:9856", "right": "issue:9911", "accept": false, - "reason": "Distributed port-in-use conflict and multi-controlnet layer-length mismatch are unrelated." + "reason": "Distributed port conflict vs multi-ControlNet layer-length mismatch; no shared bug." }, { "left": "issue:1922", "right": "issue:2375", "accept": false, - "reason": "Undefined init_git_repo in a notebook and missing safe_open are separate NameError problems." + "reason": "Undefined notebook helper vs missing safe_open import; unrelated failures." }, { "left": "issue:10797", "right": "issue:9187", "accept": false, - "reason": "HunyuanVideoPipeline group offloading and SDXL IP-Adapter image_encoder_folder=None after Kolors update are unrelated regressions." + "reason": "Different pipelines and problems: HunyuanVideo group offloading vs SDXL IP-Adapter image encoder handling." }, { "left": "issue:4566", "right": "issue:7347", "accept": false, - "reason": "Checkpoint resume failure for SDXL LoRA and a general LoRA training failure after update are not the same specific bug." + "reason": "Both are LoRA training/resume complaints, but they describe different regressions and likely different fixes." }, { "left": "issue:4241", "right": "issue:6354", "accept": false, - "reason": "A support request for unidiffusers LoRA fine-tuning and a request for ControlNet LoRA support are different features." + "reason": "A general support question about LoRA on unidiffusers is not the same as adding LoRA support to ControlNet." }, { "left": "issue:2844", "right": "issue:3871", "accept": false, - "reason": "Explicit image-embedding runtime error in StableUnCLIPImg2ImgPipeline and checkpoint-resume training error are unrelated." + "reason": "Explicit image-embedding loading in StableUnCLIP and checkpoint-resume training are unrelated." }, { "left": "issue:2560", "right": "issue:5558", "accept": false, - "reason": "Different-generator output variance and multi-ControlNet artifacts are not the same issue." + "reason": "Deterministic generator behavior and ControlNet inpaint artifacts are different issues." }, { "left": "issue:10428", "right": "issue:9895", "accept": false, - "reason": "Both are Flux inference failures, but Ascend NPU inference error and a CUDA/CPU device mismatch regression are different hardware/path-specific problems." + "reason": "These are the closest pair, but one is an Ascend NPU inference error and the other is a CUDA/CPU regression after a specific commit; too platform-specific to merge as the same bug." }, { "left": "issue:1367", "right": "issue:746", "accept": false, - "reason": "Scheduler config save with custom trained_betas and ONNX revision loading are unrelated." + "reason": "Scheduler config serialization and ONNX revision loading are unrelated." }, { "left": "issue:1166", "right": "issue:2218", "accept": false, - "reason": "TPU save hang in textual inversion and overtrained models during training do not describe the same defect." + "reason": "TPU save hang vs overtrained models; no shared failure mechanism." }, { "left": "issue:2036", "right": "issue:993", "accept": false, - "reason": "OSError loading stable-diffusion-v1-4 and fp16 dtype mismatch are different failure modes." + "reason": "A model load OSError and an fp16 dtype mismatch are distinct problems." } ] }, @@ -33697,10 +33727,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -33715,162 +33745,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a mix of unrelated bugs, feature requests, and pipeline-specific issues. None of the soft-similarity pairs looks like the same underlying defect, so I would not merge any of them as duplicates.", - "confidence": 0.9, - "canonical_issue_reason": "No single canonical issue stands out: the items span different pipelines, models, backends, and failure modes, so there is no safe umbrella duplicate.", + "summary": "No true duplicates here: the soft pairs are only title-similar, but each points to a different bug, feature request, or code path across loaders, pipelines, schedulers, and adapters.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a valid canonical duplicate target; the set is too heterogeneous and the soft pairs do not share the same underlying failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 9063 is the most concrete and actionable report in the set, but only as a representative example; it does not subsume the others.", + "best_issue_reason": "issue:9063 is the clearest standalone bug report in the set, but only as a representative example\u2014not as a real cluster canonical, since the rest are unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:9063", "right": "issue:9089", "accept": false, - "reason": "Both mention Flux, but one is a tokenizer instantiation failure and the other is a missing pipeline class; different root causes and fixes." + "reason": "Both mention FluxPipeline, but one is a tokenizer-loading error and the other is a missing class/import; different root causes." }, { "left": "issue:5510", "right": "issue:6570", "accept": false, - "reason": "Same pipeline family, but one is a dtype mismatch and the other is a tensor-shape mismatch; not the same bug." + "reason": "Both are SDXL inpaint/pix2pix runtime errors, but they involve different tensor-shape/type mismatches and different code paths." }, { "left": "issue:6323", "right": "issue:9660", "accept": false, - "reason": "One is TensorRT guidance_scale assignment, the other is from_pretrained eval-mode behavior; unrelated code paths." + "reason": "One is a TensorRT pipeline attribute issue; the other is from_pretrained not setting eval mode. Not the same bug." }, { "left": "issue:12108", "right": "issue:12458", "accept": false, - "reason": "Both involve Qwen-related work, but one is a scheduler/flow-matching breakage and the other is a batch inference feature request." + "reason": "Both are Qwen-image related, but one is scheduler flow-matching behavior and the other is batch inference/feature request." }, { "left": "issue:10010", "right": "issue:7237", "accept": false, - "reason": "A missing FLUX Fill model_index.json file is unrelated to EDMEulerScheduler lookup failures." + "reason": "FLUX example 404 vs EDMEulerScheduler availability; unrelated failures." }, { "left": "issue:6520", "right": "issue:6749", "accept": false, - "reason": "Training log_validation blank outputs and InstantID result replication are different issues with different surfaces." + "reason": "Training validation blank output vs InstantID community example replication; different scenarios and symptoms." }, { "left": "issue:10472", "right": "issue:7606", "accept": false, - "reason": "Both touch LoRA, but one is an unexpected constructor kwarg and the other is a deprecation/processor API issue." + "reason": "Both touch LoRA, but one is an unexpected lora_bias kwarg and the other is a deprecated processor warning." }, { "left": "issue:11627", "right": "issue:8626", "accept": false, - "reason": "One is a Stable Diffusion 3.5 IP-Adapter pipeline question; the other is guidance/documentation for multi-image IP-Adapter usage." + "reason": "One is an SD3.5 IP-Adapter usage issue; the other is guidance documentation for multiple IP-adapter images/masks." }, { "left": "issue:3728", "right": "issue:7509", "accept": false, - "reason": "A new pipeline request and a ReNoise pipeline proposal are not the same underlying change." + "reason": "A new pipeline request vs a real-image inversion pipeline; not the same feature." }, { "left": "issue:7049", "right": "issue:9202", "accept": false, - "reason": "Both use IP-Adapter, but one errors during unpacking in the basic example and the other is a FaceID Plus V2 compatibility problem in a different pipeline." + "reason": "Both involve IP-Adapter, but one is a value unpacking bug and the other is a specific FaceID Plus V2 incompatibility." }, { "left": "issue:2707", "right": "issue:4790", "accept": false, - "reason": "The converter syntax error and the round-trip conversion incompatibility are different defects, not one duplicate bug." + "reason": "Converter syntax error vs converter output compatibility problem; different issues." }, { "left": "issue:1125", "right": "issue:13300", "accept": false, - "reason": "These concern different pipelines/models and different runtime contexts; they are not the same inpaint/CPU failure." + "reason": "Inpaint pipeline runtime error vs CPU generation failures for SD3.5/FLUX; unrelated code paths." }, { "left": "issue:3948", "right": "issue:4014", "accept": false, - "reason": "Pale images versus watermark pixel artifacts are distinct image-quality problems." + "reason": "Pale images vs watermark pixel artifacts; different output-quality problems." }, { "left": "issue:6218", "right": "issue:6758", "accept": false, - "reason": "Batched seeds in ControlNet/Img2Img and num_images_per_prompt support in InstantID are separate parameter issues." + "reason": "Batched seeds bug in ControlNet/Img2Img vs num_images_per_prompt support in InstantID; not the same defect." }, { "left": "issue:2444", "right": "issue:2707", "accept": false, - "reason": "A training-script exit failure and a converter syntax error are unrelated." + "reason": "Dreambooth inpaint script failure vs converter syntax error; unrelated scripts." }, { "left": "issue:5647", "right": "issue:8551", "accept": false, - "reason": "Both are SDXL inpaint issues, but one is strength==1 handling and the other is device placement with balanced device_map." + "reason": "Inpaint strength==1 bug vs VAE device placement bug under balanced device_map." }, { "left": "issue:940", "right": "issue:941", "accept": false, - "reason": "Both mention MPS, but one is a scheduler crash and the other is a num_images_per_prompt noise bug; different causes." + "reason": "Both are MPS-related, but one is an LMSDiscreteScheduler crash and the other is a num_images_per_prompt noise issue." }, { "left": "issue:10016", "right": "issue:8228", "accept": false, - "reason": "These are different single-file loading failures affecting different components and model shapes." + "reason": "Both are from_single_file loading errors, but one is a shape mismatch for SD3.5M and the other is a CLIPTextModel load failure." }, { "left": "issue:4808", "right": "issue:6218", "accept": false, - "reason": "A failing SHAP-E integration test is not the same as the ControlNet/Img2Img batched-seed bug." + "reason": "SHAP-E integration test failure vs batched seeds bug; different pipelines and symptoms." }, { "left": "issue:7708", "right": "issue:7936", "accept": false, - "reason": "Missing tokenizer config and invalid safetensors JSON are both load failures, but they are not the same root problem." + "reason": "Missing tokenizer/config.json vs invalid JSON in local safetensors; different file-validation failures." }, { "left": "issue:7793", "right": "issue:9012", "accept": false, - "reason": "LoRA unloading and SD3ControlNet weight loading are different operations with different failure paths." + "reason": "LoRA unloading error vs SD3ControlNetModel.from_transformer weight loading error; unrelated loaders." }, { "left": "issue:1633", "right": "issue:2905", "accept": false, - "reason": "One requests optional Karras sigmas in a wrapper pipeline; the other asks to add Karras sigmas to all schedulers. Related theme, but not the same concrete change." + "reason": "One is an optional karras-sigmas wrapper enhancement; the other is a broad scheduler feature request. Same area, not same change." }, { "left": "issue:5628", "right": "issue:5703", "accept": false, - "reason": "Offset behavior with LMS Karras sigmas and DDIM inverse noise reconstruction are distinct scheduler bugs." + "reason": "Karras/LMS denoising offset vs ddim_inverse noise reconstruction; different scheduler behaviors." }, { "left": "issue:11046", "right": "issue:7974", "accept": false, - "reason": "A combined FLUX controlnet/inpainting/IP-adapter issue is not the same as an inpaint-with-IP-Adapter bug." + "reason": "General flux controlnet/inpainting/IP-adapter inference request vs a specific inpaint-with-IP-Adapter bug." }, { "left": "issue:617", "right": "issue:967", "accept": false, - "reason": "NNPACK unsupported hardware warnings and Onnx scheduler operand errors are unrelated." + "reason": "NNPACK hardware warning vs ONNX scheduler incompatibility; unrelated." } ] }, @@ -33973,10 +34003,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -33991,156 +34021,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a grab-bag of unrelated model, training, loading, offloading, and feature-request issues. The soft pairs are similarity-only false positives; there is no clear duplicate cluster or single canonical artifact.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "This cluster is mostly a noisy mix of unrelated issue reports across different models and workflows. The soft pairs only share broad vocabulary (e.g. loading, inference, offloading, batch size) and do not describe the same concrete bug or change, so none should be merged as duplicates.", + "confidence": 0.79, + "canonical_issue_reason": "issue:6545 is the best anchor only because it is an explicit tracker with the most inbound references and the broadest coordination role; the rest are mostly isolated model-specific bugs rather than clear duplicates of one another.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:6545 is the most suitable global issue representative for the cluster due to its tracker status, high discussion activity, and many references; it is the closest thing to a hub, even though the cluster itself does not form a true duplicate set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:12080", "right": "issue:13065", "accept": false, - "reason": "Both are Qwen-Image related, but one is generic image editing inference and the other is a controlnet combination issue; different underlying problems." + "reason": "Different requests: Qwen Image editing inference vs Qwen-Image controlnet inpaint+canny. Same model family, but not the same bug or code path." }, { "left": "issue:4080", "right": "issue:5966", "accept": false, - "reason": "Both mention diffusion pipelines, but these are different feature requests for different models and capabilities." + "reason": "SDXL inpainting pipeline request vs ControlNet for SD 2.1. These are different feature requests for different pipelines/models." }, { "left": "issue:5143", "right": "issue:6418", "accept": false, - "reason": "One is about tiny numerical output differences; the other is a video finetuning collapse to identical frames. Not the same bug." + "reason": "Unet output nondeterminism/precision difference vs stable video diffusion frames repeating. Different symptoms and likely different components." }, { "left": "issue:6594", "right": "issue:7418", "accept": false, - "reason": "Connection/download failure versus a community pipeline code bug; no shared code-path problem." + "reason": "Raw GitHub config download failure vs a community pipeline bug. No shared underlying defect beyond both being example-related." }, { "left": "issue:12319", "right": "issue:13281", "accept": false, - "reason": "Both involve offloading, but one is generic block-level offloading and the other is a torchao-quantized Qwen-Image device mismatch with use_stream=True." + "reason": "Both involve offloading, but one is block_level group offloading generally and the other is torchao-quantized Qwen image with use_stream=True device mismatch. Not the same bug." }, { "left": "issue:1434", "right": "issue:1934", "accept": false, - "reason": "Both are memory-related, but one is OOM in an upscale pipeline and the other is excessive VRAM from cpu_offload; different failure modes." + "reason": "OOM during upscale pipeline vs CPU offload VRAM consumption. Related to memory, but different pipelines and failure modes." }, { "left": "issue:12078", "right": "issue:9476", "accept": false, - "reason": "Both are FLUX training/validation issues, but one is bad validation input in an example and the other is a float/BFloat16 dtype mismatch." + "reason": "Flux Control finetuning validation input issue vs FLUX LoRA validation dtype mismatch. Different training setups and errors." }, { "left": "issue:12157", "right": "issue:12175", "accept": false, - "reason": "One is a training-time torch.device iteration error; the other is a quality complaint. Not the same issue." + "reason": "Quantization-config training device iterable error vs bad generated image quality. One is a runtime exception, the other is a quality complaint." }, { "left": "issue:3937", "right": "issue:6157", "accept": false, - "reason": "Unrelated pipeline/model problems: ZeroscopeXL loading versus ControlNet img2img ip_adapter support." + "reason": "ZeroscopeXL loading issue vs ip_adapter not working in controlnet img2img. Different models and pipeline integration points." }, { "left": "issue:10066", "right": "issue:5607", "accept": false, - "reason": "Different runtime errors in different code paths: unpacking self.attn versus tensor size mismatch." + "reason": "Self.attn unpacking ValueError vs tensor size mismatch in a different runtime path. Similar error class only, not the same bug." }, { "left": "issue:6069", "right": "issue:8668", "accept": false, - "reason": "Scheduler add_noise failing at t=0 is unrelated to SD3 img2img input alignment requirements." + "reason": "Scheduler add_noise failing at t=0 vs SD3 img2img requiring aligned image size. Different code paths and symptoms." }, { "left": "issue:6545", "right": "issue:6891", "accept": false, - "reason": "Both are trackers, but they target different refactors: modular inference flow versus model card utilities." + "reason": "Both are trackers, but they target different maintenance efforts: example inferencing modularization vs model card utilities. Not duplicate work." }, { "left": "issue:2361", "right": "issue:3296", "accept": false, - "reason": "Both involve checkpointing in training scripts, but one is resume/multi-GPU and the other is save/load with text_encoder; not the same bug." + "reason": "Checkpoint save crash in text-to-image finetuning vs checkpoint save/load problems in DreamBooth with text encoder. Different scripts and failure specifics." }, { "left": "issue:10492", "right": "issue:11702", "accept": false, - "reason": "Different LoRA/quantization failures: torchao quantize+fuse versus AccVid LoRA key lookup with diffusers." + "reason": "torchao quantize + LoRA fuse error vs AccVid LoRA key error with diffusers. Both LoRA-related, but not the same bug or model path." }, { "left": "issue:2507", "right": "issue:5606", "accept": false, - "reason": "Safetensors load performance is unrelated to LoRA alpha handling." + "reason": "Safetensors load speed on GPU vs LoRA loading without alpha value. Performance issue vs model metadata handling bug." }, { "left": "issue:2618", "right": "issue:6548", "accept": false, - "reason": "Both involve image input handling, but one is ControlNet tensor input support and the other is RGBA images being misclassified as latents." + "reason": "ControlNet image input type handling vs RGBA image misclassified as latents. Both input-type issues, but different pipelines and failure conditions." }, { "left": "issue:3377", "right": "issue:4400", "accept": false, - "reason": "A training hyperparameter question is unrelated to text_encoder support in a different DreamBooth script." + "reason": "Training-weighting question vs question about text_encoder support. Neither describes the same bug or fix." }, { "left": "issue:3565", "right": "issue:3613", "accept": false, - "reason": "Both mention OOM, but they are different scripts and different training setups; no single code-path bug." + "reason": "Both are memory-related training complaints, but they concern different scripts and different torch/runtime contexts." }, { "left": "issue:5873", "right": "issue:7237", "accept": false, - "reason": "ONNX export support for scaled_dot_product_attention is unrelated to a missing scheduler class." + "reason": "ONNX export of scaled_dot_product_attention vs missing EDMEulerScheduler symbol. Completely different functionality." }, { "left": "issue:3761", "right": "issue:5044", "accept": false, - "reason": "Loading two community pipelines and failing to load Kandinsky 2.2 are separate loading problems." + "reason": "Loading multiple community pipelines in one process vs Kandinsky 2.2 loading failure. Different pipeline-loading problems." }, { "left": "issue:2091", "right": "issue:3885", "accept": false, - "reason": "Checkpoint resume in multi-GPU training is not the same as general VRAM reduction guidance for a training script." + "reason": "Resume-from-checkpoint multi-GPU issue vs VRAM reduction question for training. Similar training context, but not the same underlying bug." }, { "left": "issue:10202", "right": "issue:9178", "accept": false, - "reason": "Both mention FLUX LoRA, but one is unloading behavior and the other is peft target module errors during training." + "reason": "Flux Control LoRA unloading bug vs flux LoRA training target modules error. Different stages: unloading/runtime cleanup vs training setup." }, { "left": "issue:10542", "right": "issue:12459", "accept": false, - "reason": "Different models and symptoms: Hunyuan batch-size breakage versus FLUX batch inference throughput." + "reason": "Hunyuan batch-size regression vs FLUX.1-Kontext batch inference throughput issue. Both concern batching, but not the same defect." }, { "left": "issue:1310", "right": "issue:2015", "accept": false, - "reason": "These are unrelated pipeline ideas, not the same bug or change." + "reason": "K-Diffusion + Stable Diffusion pipeline request vs categorical diffusion pipeline question. Different feature/topic entirely." } ] }, @@ -34242,10 +34272,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -34260,156 +34290,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: mostly unrelated bug reports, feature requests, and training/runtime issues across LoRA, Flux, quantization, pipelines, and misc. inference/training paths. No single underlying duplicate bug/change stands out, so the soft pairs look like superficial textual similarity only.", - "confidence": 0.93, - "canonical_issue_reason": null, + "summary": "This cluster is a heterogeneous mix of diffusion bugs, training errors, quantization issues, and feature requests. The soft pairs are mostly superficial title-similarity matches, not the same underlying bug/change, so none should be merged.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 11651 is the closest concrete runtime-failure report in the recurring Flux/LoRA-loading subset, but the overall cluster is too heterogeneous to have a single true canonical issue.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11651 is the most concrete, reproducible bug report in the set (FluxControl + flux-fill-lora load failure), but it should be treated only as a representative issue, not a true duplicate canonical for the whole cluster.", + "best_issue_reason": "11651 is the best representative issue here because it describes a specific, reproducible failure in the dominant Flux/LoRA theme, even though it does not subsume the other unrelated items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:5923", "right": "issue:9210", "accept": false, - "reason": "Different failure modes: multi-GPU text-to-image fine-tuning hang vs dataset column validation error." + "reason": "Different bugs: multi-GPU training stalls vs a dataset column validation error." }, { "left": "issue:11962", "right": "issue:9264", "accept": false, - "reason": "One is GGUF quantized model support for FLUX.1-Kontext-dev; the other is an inpainting feature request for Flux. Not the same bug/change." + "reason": "Different requests: GGUF quantized-model support vs asking for an inpainting model." }, { "left": "issue:2851", "right": "issue:3803", "accept": false, - "reason": "TPU push_to_hub subprocess issue vs multi-GPU LoRA training data-parallel bug; unrelated code paths." + "reason": "TPU push_to_hub failure and multi-GPU LoRA sharing are unrelated issues." }, { "left": "issue:3613", "right": "issue:6230", "accept": false, - "reason": "Both mention OOM, but one is DreamBooth training and the other is SDXL text-to-image training; different concrete issues." + "reason": "Both are OOM reports, but they are separate training scripts and not the same bug." }, { "left": "issue:10526", "right": "issue:9690", "accept": false, - "reason": "Flux FP8 optimum.quanto constructor TypeError vs example code cleanup; different problem types." + "reason": "A Flux FP8 constructor TypeError is unrelated to code cleanup for an example." }, { "left": "issue:1478", "right": "issue:2673", "accept": false, - "reason": "Upscale pipeline multiple-output handling vs SD upscale config/input verification error; not the same defect." + "reason": "One is about multiple image outputs; the other is an upscale pipeline config/runtime mismatch." }, { "left": "issue:1979", "right": "issue:861", "accept": false, - "reason": "Both are conversion-related, but one is a general SD-to-diffusers conversion problem and the other is a specific ckpt conversion command; insufficiently same underlying bug." + "reason": "Both mention conversion to diffusers, but they target different conversion paths and failures." }, { "left": "issue:11651", "right": "issue:9270", "accept": false, - "reason": "Both involve Flux LoRA support, but one is a FluxControl load failure and the other is incomplete support in SimpleTuner; different integrations and failure points." + "reason": "Related Flux/LoRA area, but one is a FluxControl load failure and the other is broad missing LoRA support." }, { "left": "issue:3238", "right": "issue:6008", "accept": false, - "reason": "LoRA safetensor conversion to diffusers vs IP-Adapter Face failing with LCM LoRA; unrelated." + "reason": "LoRA safetensor conversion is not the same as IP-adapter/LCM LoRA failing." }, { "left": "issue:5877", "right": "issue:6695", "accept": false, - "reason": "Both mention subprocess errors, but the actual training-script contexts and root causes are different and too vague to merge." + "reason": "Generic subprocess errors in different contexts; too vague to be the same bug." }, { "left": "issue:457", "right": "issue:785", "accept": false, - "reason": "MPS blue artifacts vs PNDM scheduler performance cost; completely different issues." + "reason": "Different subsystems: MPS rendering artifacts vs scheduler performance." }, { "left": "issue:3593", "right": "issue:7180", "accept": false, - "reason": "LoRA+xformers second-call failure vs SDXL controlnet CPU offload failure; different code paths and symptoms." + "reason": "LoRA/xformers second-call failure is unrelated to SDXL controlnet CPU offload." }, { "left": "issue:4213", "right": "issue:7592", "accept": false, - "reason": "from_single_file accelerate dependency issue vs DoRA state_dict loading omission; unrelated." + "reason": "Model loading dependency issue vs DoRA state_dict key loading bug." }, { "left": "issue:10514", "right": "issue:9138", "accept": false, - "reason": "Sana resolution binning/sample size support issue vs generic int/str TypeError; not the same bug." + "reason": "Sana resolution binning limit and an int/str TypeError are unrelated." }, { "left": "issue:10472", "right": "issue:11232", "accept": false, - "reason": "LoRA lora_bias constructor error during inference vs noise residuals during text-to-image fine-tuning; different stages and causes." + "reason": "LoRA constructor keyword mismatch is not the same as noise residuals during fine-tuning." }, { "left": "issue:10760", "right": "issue:7228", "accept": false, - "reason": "Missing import attribute for an ONNX pipeline vs traced model missing dtype; different failures." + "reason": "Importing a missing pipeline class is unrelated to a traced model lacking dtype." }, { "left": "issue:10798", "right": "issue:13015", "accept": false, - "reason": "Flux quantization/bitsandbytes device-combination bug vs NPU tensor-device mismatch in QwenImageEditPlus; unrelated." + "reason": "Different device/quantization problems in different models and backends." }, { "left": "issue:11135", "right": "issue:5474", "accept": false, - "reason": "New SmoothCache pipeline proposal vs Stable Diffusion Reference Only pipeline request; different features." + "reason": "Both are pipeline feature requests, but they are distinct techniques and not one change." }, { "left": "issue:11352", "right": "issue:8738", "accept": false, - "reason": "Missing warning when moving 8-bit model to CPU vs SD3Transformer2DModel misconfiguration; unrelated." + "reason": "CPU warning on 8bit model movement is unrelated to SD3Transformer configuration failure." }, { "left": "issue:11569", "right": "issue:13053", "accept": false, - "reason": "Model repo request for ltxv-13b-0.9.7-distilled vs Unipic 3.0 pipeline support; unrelated feature requests." + "reason": "A repo request for a distilled model is not the same as adding Unipic 3.0 pipeline support." }, { "left": "issue:1392", "right": "issue:5474", "accept": false, - "reason": "Generic Stable Diffusion 2 issue vs new Reference Only pipeline proposal; not the same artifact." + "reason": "Generic Stable Diffusion 2 discussion does not match a specific reference-only pipeline request." }, { "left": "issue:1643", "right": "issue:5932", "accept": false, - "reason": "DreamBooth xformers call signature error vs input-channel mismatch in upscale pipeline; different bugs." + "reason": "xformers signature mismatch during DreamBooth and a 4-channel input error are different bugs." }, { "left": "issue:255", "right": "issue:3672", "accept": false, - "reason": "UNet block size validation error vs concurrent-thread inference failure; unrelated." + "reason": "UNet shape mismatch and concurrent thread failure are unrelated." }, { "left": "issue:3700", "right": "issue:6277", "accept": false, - "reason": "Missing min_snr argument support in a training script vs broken gradients in LoRA example; different defects." + "reason": "Unsupported min_snr argument and broken gradients are separate training-example issues." } ] }, @@ -34512,10 +34542,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -34530,166 +34560,163 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous issue cluster around offloading/quantization/loading, with a few clear duplicate pairs (especially inpainting and one textual-inversion save/load pair). The 3442/4450 textual-inversion link is too weak to treat as the same bug from the packet alone.", - "confidence": 0.71, - "canonical_issue_reason": "Issue 10914 is the broadest, clearest description of the cluster\u2019s main theme: unintended CPU offloading. It generalizes better than the more specific quantization- or pipeline-tied reports.", + "summary": "This cluster is thematically loose: several issues concern offloading/quantization/device placement, but the proposed soft pairs are mostly different concrete bugs and should not be merged.", + "confidence": 0.37, + "canonical_issue_reason": "issue:10914 is the broadest representative of the shared offloading/device-placement theme: models being moved to CPU unexpectedly. It subsumes the general category better than the narrower FP8, 8-bit warning, or inefficiency reports.", "canonical_pr_reason": null, - "best_issue_reason": "10914 is the most representative user-facing issue in the set and works best as the umbrella bug for the offloading-related reports.", + "best_issue_reason": "issue:10914 is the best single issue anchor because it is the most general and user-facing statement of the underlying offloading problem.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3781", "right": "issue:5334", "accept": false, - "reason": "Both touch LoRA/training, but one is about loading multiple LoRAs and the other about accelerate.prepare/accumulate needing multiple models; different bugs." + "reason": "LoRA loading regression vs Accelerate training API expectations; different components and failure modes." }, { "left": "issue:3880", "right": "issue:4782", - "accept": true, - "reason": "Both describe the same inpainting defect: unmasked or mask==0 areas being altered when they should stay unchanged." + "accept": false, + "reason": "Both are inpainting-related, but one is degraded unmasked areas and the other is mask==0 changes; not clearly the same bug." }, { "left": "issue:4109", "right": "issue:4354", "accept": false, - "reason": "General SDXL inpaint failure vs from_single_file returning the wrong pipeline type; not the same code-path issue." + "reason": "SDXL inpaint pipeline failure vs wrong pipeline type returned from from_single_file; distinct issues." }, { "left": "issue:4386", "right": "issue:6388", "accept": false, - "reason": "SDXL LoRA conversion and SDXL+LoRA gradient-checkpointing are unrelated." + "reason": "LoRA conversion question vs gradient-checkpointing ignored in SDXL+LoRA training; unrelated bugs." }, { "left": "issue:4554", "right": "issue:8543", "accept": false, - "reason": "Different pipelines and symptoms: Kandinsky 2.2 OOM vs SD3 dreambooth gradient-checkpointing failure." + "reason": "OOM in Kandinsky example vs SD3 gradient-checkpointing failure; different pipelines and symptoms." }, { "left": "issue:5436", "right": "issue:9674", "accept": false, - "reason": "Both are training-example problems, but the concrete failures differ: ControlNet XL example error vs bad SD3 ControlNet result." + "reason": "Training example error vs bad training results; not the same concrete defect." }, { "left": "issue:6801", "right": "issue:9200", "accept": false, - "reason": "Scheduler-loading failure and SDXL ControlNet dtype error are distinct issues." + "reason": "Scheduler loading problem vs SDXL ControlNet dtype error; different code paths." }, { "left": "issue:8588", "right": "issue:8673", "accept": false, - "reason": "Missing SD3 conversion script vs missing SD3 attention masks are unrelated." + "reason": "Missing conversion script vs missing SD3 attention masks; unrelated SD3 issues." }, { "left": "issue:3454", "right": "issue:487", "accept": false, - "reason": "Both are example/training failures, but they involve different models and failure modes with no clear shared code path." + "reason": "Different eras and failures: train script compatibility vs StableDiffusionPipeline on M1 CPU." }, { "left": "issue:9484", "right": "issue:9546", "accept": false, - "reason": "Deepspeed/multigpu training setup vs validation-step OOM are different Flux training problems." + "reason": "FLUX multi-GPU/deepspeed training vs ControlNet validation OOM; separate training problems." }, { "left": "issue:10797", "right": "issue:11370", "accept": false, - "reason": "Group offloading on HunyuanVideoPipeline and a bad conversion file are unrelated." + "reason": "HunyuanVideo group offloading failure vs conversion script parsing error; unrelated." }, { "left": "issue:3280", "right": "issue:4073", "accept": false, - "reason": "Scheduler compatibility in DeepFloyd IF and a Tensor/astype safety-checker error are different bugs." + "reason": "Scheduler incompatibility in DeepFloyd IF vs safety-checker tensor/astype error; distinct subsystems." }, { "left": "issue:3442", "right": "issue:4450", "accept": false, - "reason": "Both are textual-inversion related, but the packet only gives titles and no shared targets/files or direct evidence that they are the same underlying token-dimension bug." + "reason": "Textual inversion embedding dimension mismatch vs multi text inversion not working; related area but not the same bug." }, { "left": "issue:10302", "right": "issue:11872", "accept": false, - "reason": "FP8 inference noise without CPU offloading is not the same as repeated CPU/GPU movement inefficiency." + "reason": "FP8 inference noise without CPU offloading vs repeated CPU/GPU transfers during loops; one is numerical quality, the other performance/offloading behavior." }, { "left": "issue:4183", "right": "issue:4874", "accept": false, - "reason": "ControlNet guidance_scale<=1 error and UpscalePipeline low_res_scheduler bug are unrelated." + "reason": "ControlNet guidance_scale edge case vs low_res_scheduler local-load bug; different triggers and failures." }, { "left": "issue:8463", "right": "issue:8958", "accept": false, - "reason": "Tokenizer/tokenization failure in InstructP2P-SDXL and a PyTorch register_fake compatibility issue are unrelated." + "reason": "InstructP2P-SDXL tokenization failure vs torch.library.register_fake compatibility error; unrelated import/runtime problems." }, { "left": "issue:10337", "right": "issue:9343", "accept": false, - "reason": "Different FLUX loading paths: quantized-transformer IPAdapter failure vs low_cpu_mem_usage/ignore_mismatched_sizes error." + "reason": "Quantized FLUX IPAdapter failure vs FLUX loading with low_cpu_mem_usage/ignore_mismatched_sizes; both FLUX-related but different loading paths." }, { "left": "issue:8744", "right": "issue:9209", "accept": false, - "reason": "Single-file training usage vs HF mirror connectivity are unrelated." + "reason": "Training convenience feature request vs model download connectivity to hf-mirror; not the same issue." }, { "left": "issue:11298", "right": "issue:9637", "accept": false, - "reason": "LoRA hotswapping PEFT key error and XFormers attention-mask/bfloat16 failure are different code paths." + "reason": "LoRA hot-swapping PEFT key error vs xFormers attention-mask/bfloat16 sequence-length bug; different APIs and failure modes." }, { "left": "issue:6636", "right": "issue:9167", "accept": false, - "reason": "Separate import errors involving different missing symbols/classes." + "reason": "Different import errors: DIFFUSERS_SLOW_IMPORT vs StableDiffusion3Pipeline import name issue." }, { "left": "issue:12329", "right": "issue:12529", "accept": false, - "reason": "Wrong Wan model classification vs Wan2.2 VAE tensor-size mismatch are unrelated." + "reason": "Wrong Wan transformer variant loaded vs tiled VAE tensor-size mismatch; unrelated model components." }, { "left": "issue:10914", "right": "issue:11352", "accept": false, - "reason": "Both involve CPU movement, but 11352 is specifically about a missing warning for moving 8-bit quantized models to CPU, not the same unintended-offload bug." + "reason": "Unexpected CPU offloading in general vs missing warning when moving an 8-bit quantized pipeline to CPU; related theme but not the same bug." }, { "left": "issue:6441", "right": "issue:9913", "accept": false, - "reason": "Colab from_pretrained failure and set_lora_device not working are different issues." + "reason": "from_pretrained failing in Colab vs set_lora_device not working; different API paths and symptoms." }, { "left": "issue:3694", "right": "issue:4754", - "accept": true, - "reason": "Both point to the same textual inversion save/load artifact problem: missing model_index.json/config.json after fine-tuning or example execution." + "accept": false, + "reason": "Missing model/config files during example finetuning vs corrupted textual-inversion load error; separate file-handling issues." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The cluster summary is mostly grounded, but at least one soft-edge acceptance looks overstated. In particular, issue:3442 vs issue:4450 is not supported strongly enough by the packet alone: both are textual inversion-related, but the packet only gives titles and no shared targets/files or direct evidence that they are the same token-dimension bug. The other accepted pair (3694 vs 4754) is plausible, but the 3442/4450 verdict is too confident for a soft-edge match, so I would reject on precision grounds." - }, - "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -34786,10 +34813,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -34804,162 +34831,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a duplicate cluster; it\u2019s a heterogeneous set of unrelated issues and feature requests across different models, pipelines, and training paths. All soft-pair candidates should be rejected.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue: the items span distinct bugs/features (Sana 4K loading/device errors, ControlNet requests, tokenizer/compiler issues, Wan GGUF loading, training OOMs, etc.), so none is a safe duplicate anchor.", + "summary": "This cluster is highly heterogeneous: the items mostly cover unrelated bugs, feature requests, or model-specific failures across different pipelines/subsystems. The soft pairs share broad keywords at best, but there is no convincing evidence that any pair is the same underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": "No canonical issue: the set does not cohere around one fixable bug, and the apparent overlaps are only broad subsystem similarity (e.g. Sana, Wan, ControlNet, from_single_file, multi-GPU).", "canonical_pr_reason": null, - "best_issue_reason": "If forced to anchor the set, issue:10520 is the most concrete and actionable bug report, but it only represents one narrow Sana failure mode and is not a true cluster canonical.", + "best_issue_reason": "issue:10520 is the clearest concrete bug report in the set, with a specific runtime traceback and user-reproducible failure, so it is the best representative issue even though it is not a duplicate target for the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1100", "right": "issue:898", "accept": false, - "reason": "Different errors and subsystems: CLIPTextTransformer device attribute bug vs TPU bfloat16 scalar-type support." + "reason": "Different errors in different environments: CLIPTextTransformer device attribute vs TPU bfloat16 scalar type; no shared code-path bug." }, { "left": "issue:1367", "right": "issue:2786", "accept": false, - "reason": "Scheduler config save with custom trained_betas is unrelated to numpy input compatibility in Depth2Img." + "reason": "Unrelated issues: scheduler config serialization vs numpy input handling in Depth2Img; only generic API bug similarity." }, { "left": "issue:2518", "right": "issue:5429", "accept": false, - "reason": "One is xformers support in a training script; the other is support for models without VAE in examples. Not the same bug." + "reason": "Both are example/support requests, but one is xformers support and the other is model-without-VAE support; not the same change." }, { "left": "issue:9763", "right": "issue:9927", "accept": false, - "reason": "Both involve ControlNet/SD3, but one is token truncation and the other is a HeaderTooLarge failure; different concrete issues." + "reason": "Both involve training tokenization limits, but one is Flux ControlNet truncation and the other is SD3 header-size failure; different failure modes." }, { "left": "issue:10514", "right": "issue:10520", "accept": false, - "reason": "Same Sana 4K area, but one is a resolution binning/sample_size limit and the other is a device-mismatch runtime error." + "reason": "Same model family (Sana 4K) but different bugs: resolution binning/sample_size constraint vs cross-device tensor placement error." }, { "left": "issue:2331", "right": "issue:5474", "accept": false, - "reason": "General ControlNet pipeline request vs a specific reference-only SD pipeline proposal; different changes." + "reason": "A general ControlNet pipeline request vs a new reference-only pipeline proposal; not the same underlying feature." }, { "left": "issue:1821", "right": "issue:3487", "accept": false, - "reason": "Different inpaint quality problems: negative_prompt ineffectiveness vs strength behavior with EulerAncestralDiscreteScheduler." + "reason": "Both are inpaint-related, but one is negative_prompt effectiveness and the other is strength/scheduler quality; distinct issues." }, { "left": "issue:12536", "right": "issue:12661", "accept": false, - "reason": "Both mention context parallelism, but one is a sharding divisibility assertion and the other is a ChronoEdit crash; not the same code-path bug." + "reason": "Both mention context parallelism, but one is a sharding-dimension assertion in Wan image-to-video inference and the other is a ChronoEdit crash; no concrete same bug." }, { "left": "issue:1541", "right": "issue:817", "accept": false, - "reason": "Dreambooth checkpoint/model-type loading failure is unrelated to mixed-precision support in the Dreambooth example." + "reason": "Dreambooth loading/model-type detection vs mixed-precision failure; different training issues." }, { "left": "issue:4848", "right": "issue:8786", "accept": false, - "reason": "lpw_stable_diffusion_xl failure is unrelated to SD3 max_sequence_length token handling." + "reason": "Both touch token/length behavior, but one is LPW SDXL support and the other is SD3 max_sequence_length handling; not the same code path." }, { "left": "issue:1261", "right": "issue:684", "accept": false, - "reason": "xformers attention backward assert is a CUDA attention bug, while the other is AMD training memory optimization not working." + "reason": "XFormers backward-pass assertion vs AMD memory optimization not working; different subsystems and symptoms." }, { "left": "issue:5379", "right": "issue:5380", "accept": false, - "reason": "Separate community pipeline requests with no shared bug or implementation path." + "reason": "Distinct community pipeline proposals with no shared bug or implementation target." }, { "left": "issue:12009", "right": "issue:12085", "accept": false, - "reason": "Both are Wan gguf-loading reports, but they affect different transformer classes and show different load failures." + "reason": "Both are GGUF loading failures in Wan models, but they affect different model classes and report different errors; too little evidence of one shared fix." }, { "left": "issue:4014", "right": "issue:4398", "accept": false, - "reason": "SDXL watermark pixel artifacts and all-black images in a SDXL LoRA dreambooth pipeline are different failure modes." + "reason": "SDXL watermark artifacts vs black images in SDXL LoRA dreambooth are different defects, not one underlying bug." }, { "left": "issue:5474", "right": "issue:6372", "accept": false, - "reason": "Different pipeline requests: Stable Diffusion Reference Only vs SVD + ControlNet." + "reason": "Different pipeline proposals: Stable Diffusion Reference Only vs SVD + ControlNet." }, { "left": "issue:11178", "right": "issue:8228", "accept": false, - "reason": "MistoLine single-file ControlNet loading failure is distinct from CLIPTextModel loading failure in from_single_file." + "reason": "Both are from_single_file load failures, but they involve different model components and failure types; not the same bug." }, { "left": "issue:1271", "right": "issue:1919", "accept": false, - "reason": "Notebook dependency installation problems are unrelated to missing datasets in requirements.txt." + "reason": "Both are Dreambooth example-related, but one is dependency installation and the other is missing datasets; unrelated fixes." }, { "left": "issue:2107", "right": "issue:4185", "accept": false, - "reason": "Both mention black/poor images, but one is low-resolution artifacts generally and the other is SDXL ControlNet-specific black images." + "reason": "Both mention black images, but one is low-resolution output and the other is SDXL ControlNet inference; symptom overlap only." }, { "left": "issue:3717", "right": "issue:5923", "accept": false, - "reason": "One is multi-GPU OOM; the other is a stuck execution problem in text-to-image fine-tuning." + "reason": "Both concern multi-GPU training, but one is OOM and the other is stuck execution; different problems." }, { "left": "issue:1865", "right": "issue:3382", "accept": false, - "reason": "Zero-stage3 finetune dimension error is not the same as multi-GPU CUDA OOM." + "reason": "Different training failures: ZeRO stage3 dimension error vs multi-GPU CUDA OOM." }, { "left": "issue:4229", "right": "issue:5923", "accept": false, - "reason": "Different multi-GPU training failures: accelerated Dreambooth error vs stuck text-to-image fine-tuning." + "reason": "Both are multi-GPU training problems, but one is an accelerated-training error and the other is a hang; no shared concrete bug." }, { "left": "issue:5227", "right": "issue:5934", "accept": false, - "reason": "load_lora_weight under torch.compile is a specific LoRA-loading issue; the other is a broader torch.compile error." + "reason": "One is load_lora_weight under torch.compile, the other is a general torch.compile error; overlapping topic but not the same failure." }, { "left": "issue:11819", "right": "issue:9101", "accept": false, - "reason": "Both are training-script requests, but for different models and tasks (FLUX Kontext image editing vs SD3 InstructPix2Pix)." + "reason": "Both ask for training scripts, but for different models/tasks (FLUX Kontext image editing vs SD3 InstructPix2Pix)." }, { "left": "issue:1305", "right": "issue:5205", "accept": false, - "reason": "Direct Inversion paper/example request is unrelated to the repaint + ControlNet + LoRA/text-inversion pipeline discussion." + "reason": "A paper request about Direct Inversion vs a repaint/controlnet pipeline request; no duplicate bug or change." }, { "left": "issue:8503", "right": "issue:9443", "accept": false, - "reason": "Gradient checkpointing crash in SD3 is unrelated to tokenizer instantiation from a slow version." + "reason": "Both are SD3-related, but one is gradient checkpointing crashes and the other is tokenizer slow-version instantiation; distinct issues." } ] }, @@ -35062,10 +35089,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -35080,156 +35107,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a mixed bag of unrelated Diffusers issues across SDXL, ControlNet, LoRA, quantization, safety checking, and training scripts; none of the soft pairs looks like a true duplicate.", - "confidence": 0.93, - "canonical_issue_reason": "No single canonical issue stands out: the items cover different bugs/features across unrelated pipelines, and the closest pairs still diverge by model, pipeline, or failure mode.", + "summary": "This cluster is mostly a loose mix of unrelated feature requests and bug reports around model support, LoRA/ControlNet, quantization, and training/runtime compatibility. None of the soft pairs look like true duplicates; they share only broad subsystem similarity.", + "confidence": 0.82, + "canonical_issue_reason": "issue:8482 is the broadest representative item: it is a general model-support request for a major new family (Stable Diffusion 3), which fits the cluster\u2019s overall support/compatibility theme better than the narrower bug reports.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue for deduping; the set is too heterogeneous to anchor a duplicate cluster.", + "best_issue_reason": "issue:8482 is the best issue anchor because it is a clear, standalone support request with broad scope. It is not a duplicate of the others, but it best represents the cluster\u2019s high-level theme.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4014", "right": "issue:4530", "accept": false, - "reason": "Both mention image artifacts in SDXL, but one is watermark-related and the other is refiner color artifacts; different symptoms and likely different causes." + "reason": "Both mention image artifacts, but one is SDXL watermark artifacts and the other is SDXL refiner color artifacts; different symptoms and fixes." }, { "left": "issue:4213", "right": "issue:4279", "accept": false, - "reason": "One is `from_single_file()` needing accelerate, the other is LoRA loading for SDXL 0.9; different code paths and failures." + "reason": "Both are LoRA-loading problems, but one is `from_single_file()` without accelerate and the other is SDXL 0.9 LoRAs from CivitAI; not the same bug." }, { "left": "issue:5037", "right": "issue:8729", "accept": false, - "reason": "Both involve validation images in training scripts, but the scripts and error behavior differ; not the same concrete bug." + "reason": "Both involve validation images in training scripts, but one is about the second `--validation_image` argument and the other about an unexpected validation image in webdataset SDXL ControlNet training; too different." }, { "left": "issue:3748", "right": "issue:6130", "accept": false, - "reason": "Both are LoRA-loading complaints, but one is a specific float16/float32 safetensor mismatch while the other is a generic load failure." + "reason": "Both are LoRA load failures, but one is a float16/float32 QueryKey mismatch in safetensors and the other is a generic loading error; not clearly the same code-path." }, { "left": "issue:10147", "right": "issue:7199", "accept": false, - "reason": "Multi-view generation support vs. a specific ViewFusion multi-view consistency feature request; related area, not the same change." + "reason": "A multi-view generation support request and a ViewFusion research/feature request are related at a high level, but not the same underlying bug or change." }, { "left": "issue:2331", "right": "issue:3564", "accept": false, - "reason": "General ControlNet pipeline request vs. Uni-ControlNet support; distinct feature requests." + "reason": "Both concern ControlNet support, but one is adding a ControlNet pipeline and the other is Uni-ControlNet support; different feature scopes." }, { "left": "issue:1214", "right": "issue:3661", "accept": false, - "reason": "Half-precision CLIPTextModel issue vs. `from_ckpt` conversion problem for SD 2.x weights; different failures." + "reason": "Half-precision CLIPTextModel failure and `from_ckpt` not working for SD2.x are different compatibility issues." }, { "left": "issue:4031", "right": "issue:4073", "accept": false, - "reason": "Both mention safety checkers, but one is local loading and the other is a DeepFloyd IF tensor/type error; different code paths." + "reason": "Both touch safety checker loading, but one fails loading from a local file while the other is a DeepFloyd IF `astype` tensor error; different paths." }, { "left": "issue:10798", "right": "issue:11215", "accept": false, - "reason": "Both touch FLUX quantization, but one is a device-combination bug and the other is slow inference plus LoRA support; not the same defect." + "reason": "Both are about Flux quantization, but one is a device-combination bitsandbytes bug and the other is slow quantization plus poor LoRA support; not the same concrete problem." }, { "left": "issue:1166", "right": "issue:761", "accept": false, - "reason": "TPU textual inversion saving hang vs. DreamBooth 8-bit Adam failure; unrelated training bugs." + "reason": "TPU textual inversion saving hangs and DreamBooth 8-bit Adam failures are separate training issues." }, { "left": "issue:1434", "right": "issue:4686", "accept": false, - "reason": "Upscale pipeline OOM vs. loading a saved upscale pipeline failing; different failure modes." + "reason": "One is an OOM while upscaling; the other is loading a saved upscale pipeline. Related subsystem, different failure." }, { "left": "issue:4471", "right": "issue:4885", "accept": false, - "reason": "Both are LoRA-loading issues in SDXL pipelines, but one is plain Img2Img and the other is ControlNet Img2Img; not enough evidence of the same underlying bug." + "reason": "Both involve LoRA loading in SDXL pipelines, but the affected pipelines and error modes differ; not enough to treat as the same bug." }, { "left": "issue:5016", "right": "issue:5737", "accept": false, - "reason": "DreamBooth LoRA text encoder training problem vs. SDXL conversion failure; unrelated." + "reason": "SDXL DreamBooth LoRA text encoder training and SDXL conversion-script failure are distinct issues." }, { "left": "issue:5607", "right": "issue:7168", "accept": false, - "reason": "Tensor size mismatch in one case vs. 4D IP-Adapter embeddings in the other; different tensor-shape issues." + "reason": "A tensor-dimension mismatch and IP-Adapter embed shape confusion are not the same underlying bug." }, { "left": "issue:2440", "right": "issue:5399", "accept": false, - "reason": "Both involve inpainting and dtype errors, but one is a load-time float16 issue and the other is a runtime matrix dtype mismatch." + "reason": "Both are inpainting dtype/runtime problems, but one is loading in float16 and the other is a mat1/mat2 dtype mismatch during inference; not the same fix." }, { "left": "issue:11853", "right": "issue:6008", "accept": false, - "reason": "Both mention IP-Adapter/LoRA/ControlNet combinations, but the specific model setups and reported failures are different." + "reason": "Both mention IP-Adapter/LoRA interactions, but they target different model combinations and incompatibilities; not a single duplicate." }, { "left": "issue:4305", "right": "issue:7357", "accept": false, - "reason": "`torch.compile` graph break vs. slow inference despite `torch.compile` and `fuse_loras`; performance-related but not the same bug." + "reason": "One is a specific torch.compile graph break from a diffusers version change; the other is general slow inference despite optimizations. Same area, different issue." }, { "left": "issue:2695", "right": "issue:3405", "accept": false, - "reason": "Both are ControlNet training issues, but one is validation/Tensorboard-specific and the other is a broad training-process bug." + "reason": "Both are ControlNet training problems, but one is a validation/TensorBoard reporting failure and the other is a broader training-process bug; too vague to merge." }, { "left": "issue:12022", "right": "issue:12037", "accept": false, - "reason": "Flash Attention 3 interface compatibility vs. Wan 2.2 LoRA guidance compatibility; different model/runtime problems." + "reason": "Flash-attn interface incompatibility and Wan 2.2 Lightx2v LoRA incompatibility are unrelated concrete problems." }, { "left": "issue:5966", "right": "issue:8482", "accept": false, - "reason": "ControlNet for SD 2.1 vs. support for Stable Diffusion 3 Medium; different feature requests." + "reason": "Both are model/pipeline support requests, but one is ControlNet for SD 2.1 and the other is Stable Diffusion 3 Medium support; different targets." }, { "left": "issue:4880", "right": "issue:5488", "accept": false, - "reason": "A specific ControlNet image-argument TypeError vs. a vague demo-code failure; not the same concrete issue." + "reason": "A ControlNet image-type error and a demo image-generation failure are not the same bug." }, { "left": "issue:5757", "right": "issue:6146", "accept": false, - "reason": "Checkpoint resume error in DreamBooth LoRA SDXL vs. multi-GPU training support; distinct training concerns." + "reason": "Both are training-script issues, but one is checkpoint resume errors and the other is multi-GPU training support; different code paths." }, { "left": "issue:4206", "right": "issue:4985", "accept": false, - "reason": "ControlNet training broken on main vs. SDXL InstructPix2Pix OOM; different pipeline and failure type." + "reason": "ControlNet training broken on main and SDXL InstructPix2Pix OOM are unrelated training failures." }, { "left": "issue:1500", "right": "issue:3221", "accept": false, - "reason": "Different version-compatibility problems affecting different LoRA/attn-loading APIs; not the same bug." + "reason": "Both are version-compatibility problems, but they affect different APIs and releases; not the same defect." } ] }, @@ -35332,10 +35359,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -35350,156 +35377,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous and does not look duplicate-driven; the soft pairs are mostly superficially similar titles within different bugs, pipelines, or model families. All soft edges should be rejected.", - "confidence": 0.9, - "canonical_issue_reason": "No clear canonical duplicate exists because the items span unrelated issues. If one representative issue is needed, issue 9643 is the broadest Flux/controlnet-related report in the set.", + "summary": "The items are a heterogeneous set of unrelated issues; the suggested soft pairs share superficial subsystem keywords but not the same underlying bug or change. No PRs are present.", + "confidence": 0.36, + "canonical_issue_reason": "issue:10573 is the clearest concrete bug report in the set, with a specific failing pipeline and an explicit repro context; it is more representative than the broader or more speculative issues.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 9643 is the most concise, representative issue title among the Flux/controlnet subgroup, but it should be treated only as a loose cluster representative, not a duplicate anchor.", + "best_issue_reason": "issue:10573 is the best overall anchor because it describes a precise, actionable runtime failure rather than a broad question, feature request, or vague symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1672", "right": "issue:1807", "accept": false, - "reason": "Wrong image size after ckpt conversion is unrelated to inference slowdown over repeated generations." + "reason": "Wrong image size after conversion and generation slowdown are different defects with different code paths." }, { "left": "issue:4117", "right": "issue:5565", "accept": false, - "reason": "Both mention Dreambooth/SDXL, but one is about SDXL not working for Dreambooth and the other is a LoRA script failure; different failure modes." + "reason": "SDXL Dreambooth failure and LoRA SDXL script execution issues are separate training problems." }, { "left": "issue:1188", "right": "issue:3891", "accept": false, - "reason": "Both are pipeline requests, but they ask for completely different pipelines and features." + "reason": "A community request for CLIPSeg+SD is unrelated to a Tortoise TTS pipeline request." }, { "left": "issue:2442", "right": "issue:4057", "accept": false, - "reason": "Loss-curve questions for inpainting training are not the same as a blurred-output finetuning bug." + "reason": "Loss-graph expectations for inpainting training and blurred outputs from a finetuned model are not the same bug." }, { "left": "issue:3348", "right": "issue:3551", "accept": false, - "reason": "One is about preserving the scale parameter with xFormers, the other about LoRA disabling xFormers entirely; related area, different bug." + "reason": "One is about LoRA scale support under xFormers, the other about LoRA disabling xFormers entirely; related area, but different behavior." }, { "left": "issue:8834", "right": "issue:9101", "accept": false, - "reason": "A question about SD3 ControlNet training code release is not the same as requesting an InstructPix2Pix training script." + "reason": "Training code availability for SD3 ControlNet and an InstructPix2Pix training script for SD3 are distinct feature requests." }, { "left": "issue:9630", "right": "issue:9861", "accept": false, - "reason": "Training with num_single_layers=0 on a specific multi-GPU setup is a different issue from transformer weights not updating." + "reason": "Multi-GPU/single-GPU training configuration error and transformer not updating during Flux training are different bugs." }, { "left": "issue:1185", "right": "issue:2324", "accept": false, - "reason": "Dreambooth reproducibility problems are not the same as instance data not being read." + "reason": "Dreambooth reproducibility and instance-data loading are separate issues in the training pipeline." }, { "left": "issue:5075", "right": "issue:6879", "accept": false, - "reason": "SDXL refiner prompt-embedding shape mismatch is unrelated to a custom diffusion dtype mismatch." + "reason": "SDXL refiner prompt-embedding shape mismatch and Custom Diffusion dtype mismatch are different failure modes." }, { "left": "issue:10573", "right": "issue:9643", "accept": false, - "reason": "Both involve multiple ControlNets, but they affect different pipelines and failure paths; not the same bug." + "reason": "Padding_mask_crop with multiple ControlNets in SDXL inpaint is not the same as the broader question of Flux multiple-ControlNet support." }, { "left": "issue:3878", "right": "issue:5875", "accept": false, - "reason": "Pix2Pix imagefolder training error is unrelated to ONNX conversion prompt errors for SDXL base." + "reason": "Pix2Pix training error and ONNX conversion prompt error are unrelated." }, { "left": "issue:5395", "right": "issue:6232", "accept": false, - "reason": "A LoRA combination bug is not the same as a scheduler add_noise edge-case." + "reason": "LoRA combination problems with PEFT and a scheduler add_noise bug are unrelated code paths." }, { "left": "issue:6382", "right": "issue:7306", "accept": false, - "reason": "An IP-Adapter key error during/after LCM LoRA training is unrelated to float16 support in LEdits++." + "reason": "IP-Adapter key errors in SDXL LoRA training and float16 failure in LEditsPP are different issues." }, { "left": "issue:7021", "right": "issue:7681", "accept": false, - "reason": "Both are import errors, but they stem from different causes and different modules." + "reason": "HF_HOME import error and an OpenSSL symbol/import failure are distinct environment/import problems." }, { "left": "issue:4646", "right": "issue:6517", "accept": false, - "reason": "ONNX conversion failure for ControlNet is unrelated to the ControlNet SDXL import traceback." + "reason": "ONNX conversion failure for ControlNet and an import failure for SDXL ControlNet img2img are different bugs." }, { "left": "issue:4183", "right": "issue:5953", "accept": false, - "reason": "Guidance-scale <= 1 controlnet generation errors are different from mismatched image/control_image sizes." + "reason": "guidance_scale <= 1 controlnet error and size inconsistency crash are separate ControlNet issues." }, { "left": "issue:10795", "right": "issue:12009", "accept": false, - "reason": "Both involve from_single_file/GGUF, but one is about torch.compile on transformer models and the other about Wan2.2 model loading meta-tensor failure." + "reason": "torch.compile on GGUF-loaded transformer models and Wan2.2 GGUF loading failure are related only by GGUF/from_single_file, not the same defect." }, { "left": "issue:11010", "right": "issue:11045", "accept": false, - "reason": "Flux architecture support and a Flux Schnell Dreambooth crash are different underlying problems." + "reason": "A Flux architecture-change support request and a crash loading Flux Schnell in Dreambooth are different problems." }, { "left": "issue:2507", "right": "issue:4217", "accept": false, - "reason": "Both concern loading speed, but they refer to different safetensors loading scenarios and symptoms." + "reason": "Both mention safetensors loading speed, but they do not establish the same specific bug or code path." }, { "left": "issue:12113", "right": "issue:12329", "accept": false, - "reason": "A channel-mismatch runtime error and a model-version misload (Wan2.1 vs Wan2.2) are distinct bugs." + "reason": "Channel-mismatch runtime error in Wan I2V and incorrect Wan2.1/Wan2.2 model selection are different symptoms." }, { "left": "issue:3826", "right": "issue:5474", "accept": false, - "reason": "These are unrelated new-pipeline requests for different modalities and use cases." + "reason": "Audio-Journey pipeline request and Stable Diffusion Reference Only pipeline request are separate feature proposals." }, { "left": "issue:12575", "right": "issue:13043", "accept": false, - "reason": "Attention-backend checks causing an error is not the same as a missing kernel image on RTX 5090/Blackwell." + "reason": "Attention-backend checks causing an error and Blackwell kernel-image availability are different backend/device issues." }, { "left": "issue:7056", "right": "issue:7133", "accept": false, - "reason": "train_text_encoder parameter failure and checkpoint-resume text-encoder size mismatch are different SDXL training issues." + "reason": "train_text_encoder parameter error and checkpoint resume text-encoder size mismatch are different training issues." }, { "left": "issue:6758", "right": "issue:6964", "accept": false, - "reason": "Lack of num_images_per_prompt support and a 4-channel UNet crash in inpaint are different control-flow bugs." + "reason": "InstantID num_images_per_prompt support and 4-channel UNet incompatibility are unrelated pipeline bugs." } ] }, @@ -35602,10 +35629,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -35620,165 +35647,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose set of unrelated training/loading issues. The safest anchor is the tracker issue:6545, but none of the soft-edge pairs is supported strongly enough to treat as the same bug; issue:1876 vs issue:1979 is only a possible overlap, not a clear duplicate.", - "confidence": 0.84, - "canonical_issue_reason": "issue:6545 is the broad tracker with the strongest reference signal and widest scope, so it is the best representative issue for the cluster.", + "summary": "This cluster is mostly a false-positive similarity bucket: the issues span unrelated bugs across loading, training, compilation, attention, and model-specific pipelines. I found no true duplicate pair among the soft edges.", + "confidence": 0.94, + "canonical_issue_reason": "No single canonical issue stands out; the set is too heterogeneous and the soft pairs mostly share only superficial wording like \u201cerror,\u201d \u201cbug,\u201d or a broad subsystem name.", "canonical_pr_reason": null, - "best_issue_reason": "issue:6545 is the most central and general issue in the set, making it the best cluster anchor.", + "best_issue_reason": "Issue 5934 is the closest representative only because it is a broad, generic torch.compile failure and is the nearest thing to an umbrella bug here, but it still does not cover the rest of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1604", "right": "issue:4936", "accept": false, - "reason": "Both are training-related, but they describe different scripts and failure modes; not the same bug." + "reason": "Both are training-script regressions, but one is lpw_stable_diffusion broken on diffusers 0.9.0 and the other is DreamBooth LoRA resume breakage; different code paths and failure modes." }, { "left": "issue:1876", "right": "issue:1979", "accept": false, - "reason": "There is only weak textual overlap around SD-to-diffusers conversion; the packet shows no explicit shared target, pair evidence, or clear same root cause." + "reason": "Both concern conversion, but one is about checkpoints missing state_dict and the other is a generic SD-to-diffusers conversion failure; not the same underlying bug." }, { "left": "issue:5934", "right": "issue:6096", "accept": false, - "reason": "A generic torch.compile error and an SVD dynamic-shape exception are different code-path failures." + "reason": "Both mention torch.compile, but 6096 is SVD-specific DynamicOutputShapeException while 5934 is a broader compile error; not one concrete fix." }, { "left": "issue:4185", "right": "issue:5689", "accept": false, - "reason": "Both concern image quality, but black images with ControlNet and residual noise with DPM++ are distinct symptoms and likely distinct bugs." + "reason": "Same SDXL/controlnet neighborhood, but one reports black images in ControlNet inference and the other residual noise with DPM++; distinct symptoms and likely distinct causes." }, { "left": "issue:11314", "right": "issue:11321", "accept": false, - "reason": "One is a training-script bug and the other is a README issue; they are not the same underlying defect." + "reason": "One is a fatal mistake in the FLUX ControlNet training script; the other is a README bug. Different artifact and not the same bug." }, { "left": "issue:2015", "right": "issue:5442", "accept": false, - "reason": "These refer to different features with no concrete shared failure mode." + "reason": "Categorical diffusion pipeline versus ScaleCrafter higher-resolution generation; unrelated features with no shared failing code path." }, { "left": "issue:3174", "right": "issue:3585", "accept": false, - "reason": "Both mention LoRA, but the scenarios and failure paths differ too much to merge as one bug." + "reason": "Both involve LoRA/dreambooth, but one is model loading broken by new LoRA features and the other is synthesizing LoRA into SD1.5 failing; not the same issue." }, { "left": "issue:4697", "right": "issue:8885", "accept": false, - "reason": "A specific LoRA loading problem and a Custom Diffusion/clip-retrieval setup issue are unrelated." + "reason": "A specific LoRA loading regression after 0.20.0 versus a Custom Diffusion example install/load-token problem; different failure surfaces." }, { "left": "issue:10891", "right": "issue:9476", "accept": false, - "reason": "Both involve dtype-related errors, but the models and execution paths differ; not enough to treat as one bug." + "reason": "Both are dtype-related, but DCAE internal conversion on bfloat16 and FLUX validation bias/input mismatch are separate bugs." }, { "left": "issue:8182", "right": "issue:858", "accept": false, - "reason": "Both are AttributeErrors, but they involve different missing attributes and unrelated code paths." + "reason": "A logger typo and a Parameter.named_children attribute error are clearly unrelated." }, { "left": "issue:5596", "right": "issue:8732", "accept": false, - "reason": "A captioning/conversion attribute error is not the same as a checkpoint-saving issue." + "reason": "pokemon-blip-captions build failure from a dict.convert call versus checkpoint-saving issues in advanced training; not the same path." }, { "left": "issue:10514", "right": "issue:11208", "accept": false, - "reason": "These are different unsupported-feature reports in different pipelines." + "reason": "Sana 4k resolution-binning/sample_size limitation versus SD3ControlNetInpaintingPipeline MultiControlNet support; different model features and bugs." }, { "left": "issue:6131", "right": "issue:7974", "accept": false, - "reason": "Both involve IP-Adapter, but the pipeline context and failure mode differ." + "reason": "Both touch IP-Adapter, but one is IP Adapter Plus not working generally and the other is inpaint with IP Adapter (face+style); too different to merge." }, { "left": "issue:10282", "right": "issue:9045", "accept": false, - "reason": "A weight-shape mismatch while loading differs from a CUDA conversion issue with accelerate/peft." + "reason": "Meta-tensor weight shape mismatch on load versus StableDiffusion3Pipeline CUDA conversion with accelerate/peft; unrelated underlying problems." }, { "left": "issue:1865", "right": "issue:2412", "accept": false, - "reason": "Different DreamBooth/SD2 errors with no clear shared root cause." + "reason": "Zero Stage3 SD2 dimension error and a generic DreamBooth error are different training failures, not the same bug." }, { "left": "issue:6545", "right": "issue:6552", "accept": false, - "reason": "Both are trackers in example scripts, but one is about modularizing inference and the other about FP16 resume behavior." + "reason": "One is about modularizing inference in example scripts, the other about FP16 resume training; they are separate tracker items." }, { "left": "issue:7197", "right": "issue:9927", "accept": false, - "reason": "Different training-time failures in different contexts; not the same bug." + "reason": "KeyError on resuming SD1.5 text-image training versus HeaderTooLarge in SDv3 ControlNet training; unrelated bugs." }, { "left": "issue:7735", "right": "issue:9971", "accept": false, - "reason": "A dataset column error and a scheduler failure are unrelated." + "reason": "Caption column lookup failure in custom dataset training versus DEISMultistepScheduler not working on FLUX; distinct problems." }, { "left": "issue:11006", "right": "issue:11133", "accept": false, - "reason": "Both involve video pipelines, but the models and symptoms are different." + "reason": "Both are video-pipeline reports, but Wan 2.1 I2V with quantized transformer and CogVideoX I2V are different models and code paths." }, { "left": "issue:3613", "right": "issue:4368", "accept": false, - "reason": "Both mention OOM, but they are separate training setups and not one underlying issue." + "reason": "Both are OOM reports in training, but DreamBooth torch2.0 OOM and SDXL LoRA OOM are not the same concrete bug." }, { "left": "issue:6481", "right": "issue:6741", "accept": false, - "reason": "Different train-script problems: batch-size behavior versus a missing argument." + "reason": "DreamBooth batch-size > 1 failure and missing no_flip arg in Diffusion DPO are unrelated training script issues." }, { "left": "issue:6870", "right": "issue:7945", "accept": false, - "reason": "Different pipeline feature bugs with no shared concrete failure." + "reason": "Missing enable_vae_tiling in SD img2img/inpaint and a community-pipeline combination request are not the same bug." }, { "left": "issue:8955", "right": "issue:9343", "accept": false, - "reason": "A CLIPTextModel state-dict loading error and a FLUX loading-flag issue are different problems." + "reason": "CLIPTextModelWithProjection state_dict loading error versus FLUX loading with low_cpu_mem_usage=False and ignore_mismatched_sizes=True; different load failures." }, { "left": "issue:12003", "right": "issue:12334", "accept": false, - "reason": "A meta-tensor device-transfer failure and an xFormers attention issue are distinct bugs." + "reason": "Meta-tensor copy error versus enable_xformers_memory_efficient_attention bug; no shared concrete code path." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The summary is a bit too strong for the packet, and one soft-edge verdict overstates the evidence. In particular, issue:1876 vs issue:1979 is treated as a clear same-bug duplicate, but the packet only shows weak textual similarity and no explicit shared targets or pair evidence. That should be phrased more conservatively as a possible overlap, not a clear duplicate. Because the evaluator asks for conservative soft-edge judgments grounded in the packet, I would reject this output." - }, - "evaluator_used": true, - "retried": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -35876,10 +35900,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -35894,162 +35918,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is not a true duplicate cluster: it mixes unrelated feature requests, model/support questions, and distinct bugs across different pipelines and error modes. I would not merge any of the soft edges, and I would not nominate a single canonical issue for the whole set.", - "confidence": 0.95, - "canonical_issue_reason": "No clear canonical issue; the items span unrelated topics and the soft pairs do not describe the same underlying bug or change.", + "summary": "This cluster is highly heterogeneous: a mix of unrelated Diffusers feature requests and bug reports across SD, SDXL, Flux, ControlNet, training, and export/loading issues. No pull requests are present, and none of the soft-edge pairs look like true duplicates or mergeable same-fix candidates.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 307 is the broadest umbrella-style request in the set, so it is the least misleading single issue to use as a representative anchor for this mixed cluster.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is representative enough to serve as a safe duplicate target for this cluster.", + "best_issue_reason": "Issue 307 is the most general, cluster-like issue; the rest are mostly specific runtime bugs or narrowly scoped feature requests, so it best serves as a representative issue despite the cluster's weak coherence.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4185", "right": "issue:5621", "accept": false, - "reason": "Both are SDXL ControlNet-related, but one is black images after a rebasing change while the other is lack of encoder_hid_proj support; different problems and fixes." + "reason": "Both involve ControlNet/SDXL, but one is about black images during inference and the other about unsupported `encoder_hid_proj` models; different failures and likely different fixes." }, { "left": "issue:4788", "right": "issue:6028", "accept": false, - "reason": "Both mention OOM, but they involve different models/pipelines (SD XL inpaint vs Kandinsky 3.0) and likely different memory bottlenecks." + "reason": "Both are OOM reports, but for different pipelines/models (SDXL inpaint vs Kandinsky 3.0); not the same underlying bug." }, { "left": "issue:7418", "right": "issue:7991", "accept": false, - "reason": "Both are example-script bugs, but they are in different scripts and contexts (community upscale LDM3D script vs basic training tutorial)." + "reason": "Both are code/tutorial bugs, but they refer to different example files and likely different mistakes." }, { "left": "issue:7441", "right": "issue:9913", "accept": false, - "reason": "Unrelated: python3.12 pytest failure versus a LoRA device-setting issue." + "reason": "Python 3.12 pytest failure is unrelated to `set_lora_device` not working; different subsystem and symptom." }, { "left": "issue:8540", "right": "issue:9101", "accept": false, - "reason": "Both concern SD3 training scripts, but one is dreambooth caching and the other is adding an InstructPix2Pix training script; not the same change." + "reason": "Both are training-script feature requests, but one asks for SD3 Dreambooth caching and the other for an SD3 InstructPix2Pix script; not the same change." }, { "left": "issue:1434", "right": "issue:5540", "accept": false, - "reason": "Both are memory complaints, but they are different pipelines and symptom profiles; not the same concrete bug." + "reason": "Both mention memory, but one is a specific upscale-pipeline OOM and the other is a generic inference OOM report." }, { "left": "issue:3428", "right": "issue:6382", "accept": false, - "reason": "Img2img pipeline behavior change versus an IP-adapter key error during SDXL LCM LoRA training/use; different code paths." + "reason": "These are different problems: an unexpected result when modifying Img2ImgPipeline versus an IP-Adapter key error after LCM LoRA training." }, { "left": "issue:1538", "right": "issue:4109", "accept": false, - "reason": "Different unsupported pipelines/models: StableDiffusionMegaPipeline with SD 2-base versus SDXL inpaint not working." + "reason": "Both are pipeline-specific breakages, but one is about StableDiffusionMegaPipeline with SD 2-base and the other is SDXL inpaint not working; separate code paths." }, { "left": "issue:2163", "right": "issue:3453", "accept": false, - "reason": "Both are training-script runtime errors, but the actual failures differ: UNet dtype mismatch versus CUDA bfloat16 triu/tril not implemented." + "reason": "Both are training-script runtime errors, but one is dtype mismatch in `train_text_to_image.py` and the other is BFloat16 CUDA `triu_tril` support in a different training script." }, { "left": "issue:4037", "right": "issue:4826", "accept": false, - "reason": "Both involve ControlNet, but one is FSDP validation errors during training and the other is multi-ControlNet reference errors; distinct issues." + "reason": "Both involve ControlNet training/reference behavior, but one is FSDP validation errors and the other is MultiControlNet reference errors; different failure modes." }, { "left": "issue:2052", "right": "issue:6252", "accept": false, - "reason": "Both concern checkpoints, but one is textual inversion resume behavior and the other is LoRA weights not loading from a checkpoint." + "reason": "Both involve checkpoint/weights loading, but one is Textual Inversion resume support and the other is LoRA weights not loading from checkpoint; not the same bug." }, { "left": "issue:10281", "right": "issue:8414", "accept": false, - "reason": "A new diffusion scheduler request and a new text-to-audio model request are different features, not the same change." + "reason": "A new scheduler request and a new text-to-audio model request are unrelated feature proposals." }, { "left": "issue:3233", "right": "issue:4453", "accept": false, - "reason": "Both ask about LoRA-to-safetensors conversion, but one is generic conversion and the other is SDXL-specific compatibility with AUTO1111; not the same issue." + "reason": "Both are about LoRA conversion/export, but one is generic trained LoRA to safetensor and the other is SDXL LoRA interoperability with AUTO1111; distinct requests." }, { "left": "issue:4185", "right": "issue:8729", "accept": false, - "reason": "Both are SDXL ControlNet-related, but one is inference black images and the other is an unexpected validation image during training; different bugs." + "reason": "Both are SDXL ControlNet training/inference issues, but black inference images and unexpected validation image generation are different behaviors." }, { "left": "issue:5224", "right": "issue:5607", "accept": false, - "reason": "No concrete overlap beyond being errors; Flax SD 2.1 issue and a tensor size mismatch are unrelated." + "reason": "Flax Stable Diffusion 2.1 error and tensor-size mismatch are unrelated errors in different components." }, { "left": "issue:1127", "right": "issue:1445", "accept": false, - "reason": "Both are ONNX pipeline problems, but they affect different pipelines and failure modes (import traceback vs legacy inpaint attribute error)." + "reason": "Both are ONNX pipeline issues, but one is an import traceback with DML execution provider and the other is an AttributeError in legacy inpainting; different concrete bugs." }, { "left": "issue:10282", "right": "issue:9996", "accept": false, - "reason": "Both are model-loading failures, but one is an x_embedder shape mismatch and the other is a Flux nf4 transformer loading problem; different underlying causes." + "reason": "The model-load shape mismatch and the NF4 transformer-loading issue are separate loading/weight-format problems." }, { "left": "issue:10928", "right": "issue:4356", "accept": false, - "reason": "Completely different failures: missing fp16 weight file versus missing dummy module import." + "reason": "A missing FP16 model file in a local UNet directory is unrelated to a module import error in diffusers." }, { "left": "issue:2783", "right": "issue:3612", "accept": false, - "reason": "Different feature requests: creating an img2img/inpainting ControlNet pipeline versus creating a ControlNet without a model using a specific preprocessor." + "reason": "Both mention ControlNet creation, but one asks for an img2img/inpainting ControlNet pipeline and the other asks how to create a model-free ControlNet with a specific preprocessor; not the same change." }, { "left": "issue:2838", "right": "issue:307", "accept": false, - "reason": "One asks for a ControlNet pipeline variation, the other asks to merge Stable Diffusion pipelines generally; not the same change." + "reason": "One asks about ControlNet pipeline behavior while the other is a broad request to merge Stable Diffusion pipelines; different scope and intent." }, { "left": "issue:2052", "right": "issue:5757", "accept": false, - "reason": "Both are resume-from-checkpoint problems, but they target different training scripts and likely different checkpoint-handling code." + "reason": "Both are checkpoint resume problems, but they affect different scripts and training flows." }, { "left": "issue:6333", "right": "issue:9063", "accept": false, - "reason": "Unrelated: PEFT LoRA finetuning request versus FluxPipeline tokenizer instantiation error." + "reason": "A PEFT LoRA finetuning request and a Flux tokenizer-instantiation error are unrelated." }, { "left": "issue:4353", "right": "issue:5630", "accept": false, - "reason": "Different symptoms and scopes: generic poor SDXL image quality versus LMS scheduler leaving noise in SDXL img2img." + "reason": "Both concern SDXL image quality, but one is a generic poor-quality report and the other is a specific LMS scheduler artifact in Img2Img; not the same bug." }, { "left": "issue:7306", "right": "issue:7480", "accept": false, - "reason": "Different pipelines and problems: float16 unsupported in LEditsPPPipelineStableDiffusionXL versus OOM in DeepFloyd_IF." + "reason": "Both are memory/precision-related, but one is float16 support in LEditsPPPipelineStableDiffusionXL and the other is generic OOM in DeepFloyd IF; different issues." }, { "left": "issue:9047", "right": "issue:9051", "accept": false, - "reason": "Both are Flux-related, but one is MPS device incompatibility and the other is missing batch inference support; not the same bug." + "reason": "Both are Flux-related, but one is MPS device support and the other is batch inference support; separate feature/compatibility requests." } ] }, @@ -36152,10 +36176,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -36170,156 +36194,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: most items are distinct bug reports or feature requests that only share broad model/pipeline keywords. I found no soft pair that clearly describes the same underlying fix; all proposed merges should be rejected.", - "confidence": 0.91, - "canonical_issue_reason": "No single canonical issue cleanly represents this cluster; if forced to choose an anchor, issue #11989 is the broadest symptom report, but the set is too mixed to treat it as a true duplicate hub.", - "canonical_pr_reason": null, - "best_issue_reason": "Issue #11989 is the most generic, broadly phrased report about a black inference image, so it is the closest available representative among these weakly related issues.", - "best_pr_reason": null, + "summary": "This cluster is mostly heterogeneous: the items share broad themes like generation failures, training bugs, or pipeline feature requests, but the soft pairs do not look like the same underlying defect/change. I would not consolidate any of the candidate pairs.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue: the set spans many unrelated models, pipelines, and bug classes, and the apparent overlaps are only superficial symptom-level similarities.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "No global best issue for deduplication purposes; the issues are too diverse to use one as a representative canonical target.", + "best_pr_reason": "No pull requests are present in this cluster.", "soft_edge_verdicts": [ { "left": "issue:9178", "right": "issue:9911", "accept": false, - "reason": "Both are Flux-related, but one is a PEFT target-modules training error and the other is a multi-ControlNet layer-length mismatch; different failures and code paths." + "reason": "Both mention FLUX training/controlnet, but they describe different failures in different workflows; not the same concrete bug." }, { "left": "issue:10533", "right": "issue:11989", "accept": false, - "reason": "Both mention black outputs, but #10533 is specific to FLUX fill fp16 while #11989 is a broader suspected weight-related black image issue; not the same concrete bug." + "reason": "Both mention black images, but one is FLUX fill fp16 and the other is a generic inference black-image report tied to weights; too broad to be the same bug." }, { "left": "issue:3284", "right": "issue:4733", "accept": false, - "reason": "Both involve dreambooth LoRA training, but one fails with accelerate and the other is a checkpoint-resume optimizer inf-check error; distinct problems." + "reason": "Both involve Dreambooth/accelerate, but one is a general run failure and the other is a checkpoint-resume optimizer issue; different failure modes." }, { "left": "issue:1305", "right": "issue:5442", "accept": false, - "reason": "Completely different pipelines/features: Direct Inversion editing versus ScaleCrafter high-resolution generation." + "reason": "These are different pipeline feature requests, not the same change." }, { "left": "issue:2776", "right": "issue:3984", "accept": false, - "reason": "One is a Dreambooth/DeepSpeed runtime-version issue, the other is a class_labels requirement when num_class_embeds > 0; unrelated errors." + "reason": "One is a DeepSpeed/version requirement issue for Dreambooth on 8GB, the other is a class_labels ValueError; unrelated bugs." }, { "left": "issue:5004", "right": "issue:8708", "accept": false, - "reason": "Both are Dreambooth-related, but one is fp16 checkpoint resume and the other is a missing SD3 VAE config field in examples; different root causes." + "reason": "Both are fp16/Dreambooth-related, but one is resume-from-checkpoint in SDXL and the other is a missing VAE shift_factor in SD3 examples; not the same root cause." }, { "left": "issue:5025", "right": "issue:9167", "accept": false, - "reason": "Both are import errors, but they concern different missing symbols in different modules; not the same fix." + "reason": "An import error for randn_tensor and an import problem for StableDiffusion3Pipeline are distinct issues." }, { "left": "issue:5931", "right": "issue:9171", "accept": false, - "reason": "One is nondeterministic StableDiffusionPipeline generation, the other is from_single_file handling of v_prediction SD 1.5 models; unrelated symptoms." + "reason": "Determinism in StableDiffusionPipeline is unrelated to from_single_file handling of v_prediction models." }, { "left": "issue:9148", "right": "issue:9873", "accept": false, - "reason": "These are separate feature requests for different pipelines/models: Flux inpainting versus OmniGen support." + "reason": "Both are feature requests for new pipelines, but they request different models and capabilities." }, { "left": "issue:9665", "right": "issue:9861", "accept": false, - "reason": "One is a DeepSpeed/accelerator.prepare interaction, the other is Flux training not updating transformer weights; different code paths." + "reason": "Both are FLUX training issues, but one is about accelerator.prepare/DeepSpeed config and the other about the transformer not updating; different code-path problems." }, { "left": "issue:11160", "right": "issue:13105", "accept": false, - "reason": "Both are WAN-related, but they target different model sizes/tasks and different symptoms (inconsistent results vs output noise)." + "reason": "Both mention WAN 2.1, but one is inconsistent inference in I2V and the other is noisy output in a T2V model; not clearly the same defect." }, { "left": "issue:1445", "right": "issue:5917", "accept": false, - "reason": "Different areas and failures: ONNX inpaint pipeline legacy attribute error versus checkpoint conversion failure after a version upgrade." + "reason": "Different errors in different tooling paths: ONNX inpaint legacy attribute error vs checkpoint conversion failure after upgrade." }, { "left": "issue:3391", "right": "issue:987", "accept": false, - "reason": "Both describe odd image colors, but they are different pipelines with different inputs and outputs; no shared underlying bug is evident." + "reason": "Both describe bad image output, but they involve different pipelines and different symptom types (overexposure vs green image)." }, { "left": "issue:4885", "right": "issue:6252", "accept": false, - "reason": "Both mention LoRA loading, but one is SDXL ControlNet Img2Img weight loading and the other is checkpoint LoRA retrieval; too broad to merge." + "reason": "Both are about LoRA loading, but one is SDXL ControlNet Img2Img and the other is checkpoint loading generally; not enough evidence of the same bug." }, { "left": "issue:6336", "right": "issue:7722", "accept": false, - "reason": "TensorRT VAE compilation with AutoencoderTiny is unrelated to the animatediff_controlnet/IPAdapter error." + "reason": "These are unrelated pipeline errors affecting different components and use cases." }, { "left": "issue:8922", "right": "issue:8970", "accept": false, - "reason": "Both produce black images, but on different models/backends (SD3 single-file quantized T5 vs Kolors on MPS); no shared fix is apparent." + "reason": "Both produce black images, but they are different models and environments (SD3 single-file vs Kolors on MPS), so this is only symptom-level similarity." }, { "left": "issue:2695", "right": "issue:3812", "accept": false, - "reason": "One is a ControlNet training validation/reporting problem, the other is a missing controlnet config field; different issues." + "reason": "ControlNet training validation/reporting and missing controlnet config are different issues." }, { "left": "issue:1076", "right": "issue:1188", "accept": false, - "reason": "A general Stable Diffusion model question versus a feature request for a CLIPSeg pipeline; not duplicates." + "reason": "These are unrelated community/help requests from different eras and topics." }, { "left": "issue:13008", "right": "issue:13232", "accept": false, - "reason": "Different feature requests for different model families and pipelines; no common code-path." + "reason": "Different feature requests for different models (FLUX LoRA multi-image training vs LTX-2.3 support)." }, { "left": "issue:7237", "right": "issue:7341", "accept": false, - "reason": "EDMEulerScheduler missing symbol and TorchVersion/Version comparison TypeError are unrelated compatibility issues." + "reason": "One is a missing scheduler class, the other is a TorchVersion/Version comparison TypeError; unrelated." }, { "left": "issue:11258", "right": "issue:12329", "accept": false, - "reason": "Different single-file model support topics for different architectures (LTX vs WAN) and different failure modes." + "reason": "Different single-file model-loading issues for different model families (LTX vs Wan)." }, { "left": "issue:1851", "right": "issue:3613", "accept": false, - "reason": "One is multi-GPU being slower than single GPU, the other is out-of-memory with torch 2.0; distinct performance versus memory issues." + "reason": "Both concern Dreambooth performance/resource usage, but one is multi-GPU slowness and the other is OOM with torch 2.0; different problems." }, { "left": "issue:11581", "right": "issue:5798", "accept": false, - "reason": "An IP-Adapter dtype constructor error in FLUX versus an unexpected callback_on_step_end argument in ControlNet; unrelated." + "reason": "Both are TypeErrors, but they occur in different APIs and with different causes; not the same fix." }, { "left": "issue:7306", "right": "issue:7432", "accept": false, - "reason": "Float16 support in LEdits++ XL and general image-result differences versus WebUI are not the same bug." + "reason": "I cannot use float16 in one pipeline is unrelated to generating images that differ from WebUI." } ] }, @@ -36422,10 +36446,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -36440,163 +36464,166 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a loose collection of unrelated GitHub issues with a few superficial thematic overlaps (FLUX Fill, adapters, memory, img2img timesteps), but no duplicate-quality matches among the soft edges.", - "confidence": 0.93, - "canonical_issue_reason": null, + "summary": "This cluster is mostly heterogeneous. The only strong duplicate-like pair is the SDXL img2img timestep/strength failure pair; the rest differ by model, pipeline, or bug class and should stay separate.", + "confidence": 0.74, + "canonical_issue_reason": "issue:9366 is the broader, earlier report for the img2img timestep/strength failure family, so it works best as the canonical issue if these are grouped.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11922 is the broadest FLUX.1-Fill-dev runtime report, so it is the best single umbrella representative of the FLUX Fill subset, though it is not a true duplicate of the others.", + "best_issue_reason": "issue:11092 is the most detailed reproduction of the same img2img timestep/strength code-path problem, making it the best representative issue for triage.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3451", "right": "issue:3885", "accept": false, - "reason": "Both mention VRAM, but one is a specific fp16-vs-fp32 memory regression in train_text_to_image.py and the other is a general VRAM-reduction question." + "reason": "Both mention VRAM usage in training, but one is a specific fp16-vs-fp32 memory regression and the other is a general \u201chow do I reduce VRAM\u201d question, not the same bug." }, { "left": "issue:10337", "right": "issue:10612", "accept": false, - "reason": "Different failure modes: quantized transformers breaking FLUX IPAdapter vs NF4 quantized Flux Fill LoRA loading." + "reason": "Both involve Flux and quantization, but one is IP-Adapter failure and the other is LoRA loading in NF4 quantized Flux Fill; different adapters and code paths." }, { "left": "issue:11418", "right": "issue:11922", "accept": false, - "reason": "One is about adding an fp8 safetensors file; the other is a generic FLUX.1-Fill-dev runtime problem." + "reason": "One is about adding a specific fp8 safetensors file, the other is a vague FLUX.1-Fill-dev runtime problem; too little overlap to treat as the same bug." }, { "left": "issue:11978", "right": "issue:8414", "accept": false, - "reason": "Both are model-addition requests, but for completely different models and code paths." + "reason": "Different models and different feature requests; no shared underlying bug or change." }, { "left": "issue:12666", "right": "issue:13142", "accept": false, - "reason": "Unrelated version-compatibility bugs: WanVACE GGUF loading vs PRXPipeline breaking with transformers 5.1.0." + "reason": "These are unrelated breakages in different pipelines/model stacks; one is a video pipeline missing in a release, the other is a transformers 5.1.0 regression." }, { "left": "issue:1574", "right": "issue:1734", "accept": false, - "reason": "Both are Dreambooth training complaints, but one is a bare-metal 8GB GPU failure and the other is slow multi-GPU training." + "reason": "Both are Dreambooth training issues, but one is a crash on 8GB GPU and the other is a multi-GPU slowdown; different problems." }, { "left": "issue:2107", "right": "issue:2495", "accept": false, - "reason": "Different image-output bugs on different backends: low-resolution glitches vs MPS upscale producing black images." + "reason": "Both mention black/glitchy output, but they affect different pipelines and different conditions, so they do not look like the same bug." }, { "left": "issue:228", "right": "issue:259", "accept": false, - "reason": "Broken model link vs progress-bar behavior; no shared bug." + "reason": "A broken model link and a progress-bar fix are unrelated." }, { "left": "issue:4830", "right": "issue:8667", "accept": false, - "reason": "Both are tensor-shape issues, but they affect different models and scenarios: MultiControlNet vs SD3ControlNetModel on non-power-of-two images." + "reason": "Both are tensor-size issues in controlnet-adjacent code, but one is MultiControlNet and the other is non-power-of-two image sizing in SD3ControlNetModel; not the same defect." }, { "left": "issue:5769", "right": "issue:7185", "accept": false, - "reason": "Mixed-precision training correctness is unrelated to pytorch 2.2 support compatibility." + "reason": "Mixed-precision training correctness and PyTorch version support are different concerns." }, { "left": "issue:7405", "right": "issue:9744", "accept": false, - "reason": "Different model requests: DynamiCrafter video generation vs Mochi video model." + "reason": "These are separate new model additions, not a duplicate bug or change." }, { "left": "issue:1261", "right": "issue:964", "accept": false, - "reason": "Xformers backward-pass assert is not the same as the inpainting example failing." + "reason": "Xformers backward-pass failure and an inpainting example not working are unrelated." }, { "left": "issue:11088", "right": "issue:11878", "accept": false, - "reason": "Both involve Wan and GGUF, but they are different model classes and distinct load failures." + "reason": "Both involve Wan GGUF loading, but they target different model classes and failures; not clearly the same underlying issue." }, { "left": "issue:2420", "right": "issue:3325", "accept": false, - "reason": "Different training failures: xformers memory-efficient attention error vs CUDA invalid argument in Dreambooth." + "reason": "Different training scripts, different failures, and no shared concrete code-path bug." }, { "left": "issue:10147", "right": "issue:11561", "accept": false, - "reason": "Different adapter features: MV-Adapter for multi-view generation vs FluxFillPipeline IP-Adapter loading." + "reason": "Support for MV-Adapter and support for loading IP-Adapter in FluxFill are distinct feature requests." }, { "left": "issue:4080", "right": "issue:6928", "accept": false, - "reason": "Unrelated feature requests: SDXL inpainting pipeline vs a Marigold depthmap colorization option." + "reason": "An SDXL inpainting pipeline request and a Marigold depth colorization option are unrelated changes." }, { "left": "issue:11092", "right": "issue:9366", - "accept": false, - "reason": "Both touch img2img timestep logic, but the pipelines and scheduler contexts differ and the titles do not indicate the same concrete bug." + "accept": true, + "reason": "Both describe img2img failures around step/strength-derived timesteps; they point to the same scheduler/timestep code path and could plausibly be merged into one bug report." }, { "left": "issue:1934", "right": "issue:2053", "accept": false, - "reason": "General VRAM complaints, but one is cpu_offload memory overhead and the other is a version regression consuming more VRAM." + "reason": "Both are memory-related, but one concerns cpu_offload VRAM usage and the other a version regression in overall VRAM consumption." }, { "left": "issue:2480", "right": "issue:3468", "accept": false, - "reason": "Both concern Dreambooth checkpoint artifacts, but missing per-checkpoint unet folders is different from missing model_index.json." + "reason": "Both are Dreambooth checkpoint/export issues, but one is missing per-checkpoint unet folders and the other is missing model_index.json; different outputs and failures." }, { "left": "issue:10866", "right": "issue:13043", "accept": false, - "reason": "Completely different issues: Lumina Image LoRA compatibility vs attention backend kernel support on RTX 5090." + "reason": "Completely different models and failure modes; no duplicate relationship." }, { "left": "issue:4427", "right": "issue:7931", "accept": false, - "reason": "Both mention adapter-like features, but they are different pipeline bugs and not the same code-path failure." + "reason": "A multi-adapter example failure and an IP-Adapter embedding issue in SDXL ControlNet inpaint are different adapter paths." }, { "left": "issue:4542", "right": "issue:8626", "accept": false, - "reason": "SDXL img2img producing only one image is unrelated to guidance for multiple IP adapter images/masks." + "reason": "One is about generating more than one image, the other is documentation/guidance for multiple IP-Adapter images and masks; not the same bug." }, { "left": "issue:10010", "right": "issue:10540", "accept": false, - "reason": "Broken FLUX Fill example URL is not the same as a from_single_file loading regression after FLUX Fill support was added." + "reason": "Both are Flux Fill-related, but one is a missing model_index.json URL and the other is a from_single_file regression after adding Flux Fill support; different problems." }, { "left": "issue:1076", "right": "issue:1392", "accept": false, - "reason": "Generic model-support questions for different Stable Diffusion releases, not the same bug or change." + "reason": "Unrelated issues with no meaningful overlap." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The heterogeneity summary is supported by the packet, and the only accepted soft pair (issue:11092 / issue:9366) is a conservative same-family match around img2img timestep/strength behavior. The other soft-edge rejections are appropriately cautious and do not overstate duplicate evidence." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -36692,10 +36719,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -36710,156 +36737,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a loose collection of unrelated model/pipeline/training issues connected only by broad LoRA/Flux/diffusion wording. None of the soft pairs look like the same concrete bug, so all should stay separate.", - "confidence": 0.41, - "canonical_issue_reason": "issue:13261 is the most substantial and active report here: a detailed, open LoRA/Lokr compatibility failure on FLUX.2-klein-9B, which best represents the cluster\u2019s central LoRA-on-new-models theme.", + "summary": "Mostly unrelated issues; none of the soft pairs look like true duplicates, and there are no PRs in this set.", + "confidence": 0.88, + "canonical_issue_reason": "issue:13261 is the most detailed and actively discussed report, so it is the best single anchor if one issue must represent this mixed set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:13261 has the richest reproduction context and discussion, making it the strongest anchor for this otherwise loosely related set of issues.", + "best_issue_reason": "issue:13261 has the richest context and strongest ongoing discussion, making it the best issue for triage/monitoring among these items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:11581", "right": "issue:9069", "accept": false, - "reason": "Different failures and code paths: IP-Adapter FLUX dtype constructor error vs ndarray type mismatch." + "reason": "Different bugs: FLUX IP-Adapter init dtype argument error vs np.ndarray type mismatch. Same framework family, not the same failure." }, { "left": "issue:1657", "right": "issue:3065", "accept": false, - "reason": "One asks for DreamBooth notebook examples; the other requests a LoRA text-encoder tuning option." + "reason": "Kaggle notebook request is unrelated to adding text-encoder LoRA support in DreamBooth." }, { "left": "issue:4269", "right": "issue:4880", "accept": false, - "reason": "Both are ControlNet validation errors, but the specific bug symptoms and trigger conditions differ." + "reason": "Both are TypeErrors in ControlNet, but one is about check_inputs defaults and the other about missing/invalid image input handling." }, { "left": "issue:4109", "right": "issue:4286", "accept": false, - "reason": "SDXL inpaint import/runtime issue vs LoRA loading failure in SDXL; not the same bug." + "reason": "StableDiffusionXLInpaintPipeline failure is unrelated to SDXL load_lora_weights not working." }, { "left": "issue:7431", "right": "issue:9497", "accept": false, - "reason": "Different training regressions: DoRA/text-encoder issues vs Flux matmul shape error." + "reason": "Different training failures on different model stacks: DoRA/text-encoder issues vs Flux matmul shape error." }, { "left": "issue:13249", "right": "issue:13261", "accept": false, - "reason": "Both concern LoRA on new models, but they involve different model families and distinct failure modes." + "reason": "Both mention LoRA on Flux/Z-Image, but they concern different models and different symptoms; not enough to treat as one bug." }, { "left": "issue:11864", "right": "issue:2375", "accept": false, - "reason": "torch.compile fullgraph NameError is unrelated to the safe_open import NameError." + "reason": "Both are NameError-style failures, but one is torch.compile/fullgraph scoping and the other is a missing safe_open import." }, { "left": "issue:4598", "right": "issue:9109", "accept": false, - "reason": "Generic missing-dependency messaging vs a specific torchvision operator registration error." + "reason": "One is about improving missing dependency errors; the other is a specific torchvision::nms runtime failure." }, { "left": "issue:11978", "right": "issue:9042", "accept": false, - "reason": "Separate feature requests for different pipeline additions, not one shared change." + "reason": "Separate feature requests for different pipelines (Bria 3.2 vs Flux)." }, { "left": "issue:5544", "right": "issue:7681", "accept": false, - "reason": "FreeU disable behavior is unrelated to an OpenSSL/lib import failure." + "reason": "Unrelated issues: FreeU disable handling in SDXL vs a low-level import failure from lib/OpenSSL." }, { "left": "issue:6481", "right": "issue:9393", "accept": false, - "reason": "Different DreamBooth training problems: batch-size handling vs DeepSpeed model-saving failure." + "reason": "Different DreamBooth training problems: batch-size >1 behavior vs DeepSpeed save-model failure." }, { "left": "issue:1566", "right": "issue:4551", "accept": false, - "reason": "Checkpoint-save fp16 crash and SDXL DreamBooth crash are different bugs with different triggers." + "reason": "Both are DreamBooth crashes, but the repros and failure points are different (checkpoint saving with fp16 output vs SDXL base training crash)." }, { "left": "issue:12107", "right": "issue:9613", "accept": false, - "reason": "accelerator.init_trackers custom-object error is unrelated to the Timesteps.scale attribute error." + "reason": "Custom tracker object handling in init_trackers is unrelated to the Timesteps.scale attribute error." }, { "left": "issue:5488", "right": "issue:5509", "accept": false, - "reason": "Demo image generation failure is not the same as np.ndarray/cv2 batch-size validation." + "reason": "Different demo/input-shape problems: one image generation not working vs image/prompt batch-size mismatch." }, { "left": "issue:10591", "right": "issue:8557", "accept": false, - "reason": "SD3 training-script issue vs SD3 token-limit discussion; too different to be duplicates." + "reason": "SD3 LoRA training script problem is too vague to equate with the specific SD3 token-limit training issue." }, { "left": "issue:2856", "right": "issue:4974", "accept": false, - "reason": "Both involve conversion to diffusers, but one is generic SD conversion and the other is safetensor-specific." + "reason": "Both involve conversion to diffusers, but one is a general model conversion problem and the other is about .safetensor conversion; not the same bug." }, { "left": "issue:4265", "right": "issue:6794", "accept": false, - "reason": "Pipeline import error and cast_training_params import error are unrelated failures." + "reason": "Different import errors in different modules: ControlNet pipeline export vs training_utils helper import." }, { "left": "issue:4398", "right": "issue:4530", "accept": false, - "reason": "Black images in an SDXL LORA-dreambooth pipeline are not the same as refiner color artifacts." + "reason": "Both are SDXL image-quality problems, but black outputs from LoRA-dreambooth are not the same as refiner color artifacts." }, { "left": "issue:4925", "right": "issue:9546", "accept": false, - "reason": "Both mention OOM, but they affect different training setups and different steps." + "reason": "Both are OOM reports in ControlNet training, but they hit different models and steps, so they are not clearly the same bug." }, { "left": "issue:9089", "right": "issue:9454", "accept": false, - "reason": "Missing FluxPipeline class and missing CogVideoX Image2Video model are distinct missing-component issues." + "reason": "Both are missing-model issues, but they refer to entirely different models/pipelines (Flux vs CogVideoX)." }, { "left": "issue:2606", "right": "issue:5743", "accept": false, - "reason": "DeepSpeed save_state hang and LCD training stuck at 0% are different training stalls." + "reason": "DeepSpeed save_state hang is unrelated to LCD training stuck at 0%." }, { "left": "issue:13107", "right": "issue:13484", "accept": false, - "reason": "Both are Flux LoRA-loading problems, but they refer to different models and error conditions." + "reason": "Both are Flux LoRA loading problems, but the models and failure modes differ enough that they do not look like one concrete bug." }, { "left": "issue:5023", "right": "issue:7021", "accept": false, - "reason": "Wuerstchen import regression and HF_HOME import error are unrelated." + "reason": "Wuerstchen pipeline import error and HF_HOME import error are unrelated." }, { "left": "issue:5433", "right": "issue:5515", "accept": false, - "reason": "Scheduler artifacts with SDXL DPM++ are not the same as LCM init-without-scheduler failure." + "reason": "Scheduler-related, but one is SDXL visual artifacts with DPM++ and the other is an LCM pipeline initialization failure without a scheduler." } ] }, @@ -36961,10 +36988,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -36979,166 +37006,166 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Conservative read: this cluster is mostly unrelated, and none of the candidate issue pairs is strong enough to accept as a true duplicate/subset edge. There are no PRs in the packet.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "Mostly unrelated issue pairs; only the from_ckpt reports appear to be the same underlying bug. No PRs are present in this cluster.", + "confidence": 0.72, + "canonical_issue_reason": "issue:3595 is the broad root report for from_ckpt loading failures; issue:3661 is the narrower Stable Diffusion 2.x instance of the same loader problem.", "canonical_pr_reason": null, - "best_issue_reason": "issue:3595 is the closest thing to a broad representative because it is a generic from_ckpt loading report, but it should be treated only as a loose thematic anchor, not as a confirmed duplicate hub.", + "best_issue_reason": "issue:3595 is the best representative because it captures the general from_ckpt failure, while issue:3661 is a specific subtype/duplicate.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:5894", "right": "issue:9112", "accept": false, - "reason": "Both are runtime failures, but the concrete symptoms and model contexts differ: invalid buffer size vs tensor dimension mismatch in Auraflow." + "reason": "Both are runtime inference errors, but the titles point to different failures: invalid buffer size vs tensor shape/dtype mismatch in Auraflow." }, { "left": "issue:9164", "right": "issue:9393", "accept": false, - "reason": "Both involve Flux training, but one is about convergence and the other about deepspeed save-model behavior." + "reason": "Both involve Flux DreamBooth training, but one is a convergence problem and the other is a model-saving failure under deepspeed." }, { "left": "issue:7347", "right": "issue:8871", "accept": false, - "reason": "Both are LoRA training problems, but the reported errors and likely code paths are different." + "reason": "Both are LoRA training complaints, but the concrete failures differ: a generic post-update training breakage vs FP16 gradient unscale errors." }, { "left": "issue:3468", "right": "issue:3700", "accept": false, - "reason": "Dreambooth model-index/loading failure and a min_snr script argument are different issues." + "reason": "One is about missing model_index.json after DreamBooth training; the other is about the min_snr argument in a LoRA training script." }, { "left": "issue:10239", "right": "issue:10453", "accept": false, - "reason": "Sequential CPU offload support and batch-size > 1 support are separate HunyuanVideo limitations." + "reason": "Same model family, but different problems: sequential CPU offload support vs batch size > 1 support." }, { "left": "issue:1475", "right": "issue:761", "accept": false, - "reason": "These are different DreamBooth failures: mixed class images vs 8-bit Adam optimizer." + "reason": "DreamBooth-related, but one is about mixing two image classes for training and the other is an 8-bit Adam optimizer failure." }, { "left": "issue:3111", "right": "issue:3418", "accept": false, - "reason": "Custom conditioning encoder support and adding text-encoder support to a training script are related areas, not the same bug." + "reason": "Both mention conditioning/text encoders, but one requests a custom encoder for conditioning and the other asks to add text-encoder handling to a LoRA training script." }, { "left": "issue:1931", "right": "issue:4427", "accept": false, - "reason": "A float16 dtype mismatch and a Multiadapter example failure are unrelated defects." + "reason": "Float16 scalar-type mismatch during SD2 inference is unrelated to the T2I Multiadapter example not working." }, { "left": "issue:10674", "right": "issue:9051", "accept": false, - "reason": "Flux GGUF loading and Flux batch inference are distinct problems." + "reason": "Flux GGUF loading and Flux batch inference are different code paths and different bugs." }, { "left": "issue:4236", "right": "issue:6974", "accept": false, - "reason": "Both touch SDXL text-encoder handling, but one is refiner registration and the other is unloading textual inversion state." + "reason": "One is about SDXL refiner text-encoder registration; the other is about unloading tokenizer/text-encoder components for textual inversion." }, { "left": "issue:4911", "right": "issue:8287", "accept": false, - "reason": "Custom text encoder loading from_single_file and the text_encoder=None path are different code paths." + "reason": "Both involve text_encoder handling, but one is custom text_encoder loading from_single_file and the other is the text_encoder=None path." }, { "left": "issue:5044", "right": "issue:5963", "accept": false, - "reason": "Different Kandinsky versions and different failure modes make this too weak to merge." + "reason": "Different Kandinsky versions and different failure modes: loading failure vs embeds-instead-of-prompts handling." }, { "left": "issue:10674", "right": "issue:7483", "accept": false, - "reason": "Flux GGUF loading and LoRA-to-Diffusers conversion for SDXL are unrelated." + "reason": "Flux GGUF loading is unrelated to LoRA-to-Diffusers conversion for SDXL-base models." }, { "left": "issue:10863", "right": "issue:9914", "accept": false, - "reason": "Local safetensors loading errors and Flux Xlabs LoRA loading are different model-loading issues." + "reason": "Loading local safetensors files and loading trained Flux LoRA weights are not the same underlying bug." }, { "left": "issue:6176", "right": "issue:6560", "accept": false, - "reason": "Both involve LoRA loading, but the concrete failure modes differ and there is no clear shared code path." + "reason": "Both are LoRA-loading related, but the titles suggest different contexts and failure modes; not enough evidence of the same concrete bug." }, { "left": "issue:3174", "right": "issue:4975", "accept": false, - "reason": "One is a model-loading breakage; the other is a performance inefficiency in LoRA loading." + "reason": "One reports LoRA+DreamBooth loading breakage from new features; the other reports inefficient datatype queries during LoRA loading." }, { "left": "issue:3623", "right": "issue:4593", "accept": false, - "reason": "Generic large-model loading trouble is not the same as the specific LoRA rank constraint failure." + "reason": "Unable to load larger models is not the same as LoRA rank being too high for Automatic1111 conversion/loading." }, { "left": "issue:3595", "right": "issue:3661", - "accept": false, - "reason": "They are thematically related to from_ckpt, but the evidence only supports a soft similarity, not a confirmed duplicate/subset relation." + "accept": true, + "reason": "These are the same from_ckpt loading bug, with issue:3661 being a specific Stable Diffusion 2.x manifestation of issue:3595." }, { "left": "issue:6815", "right": "issue:9007", "accept": false, - "reason": "Both can produce black outputs, but the scheduler/pipeline contexts differ and the root cause is not shown to be the same." + "reason": "Both end up with black output, but the specific errors differ: invalid value during cast vs scheduler order-related black images." }, { "left": "issue:11436", "right": "issue:9069", "accept": false, - "reason": "A dtype mismatch and an ndarray type mismatch are unrelated errors." + "reason": "One is a dtype mismatch in Hidream inference; the other is a numpy type mismatch error when expecting np.ndarray." }, { "left": "issue:3876", "right": "issue:7348", "accept": false, - "reason": "ControlNet inpaint demo error and scheduler weird-results are different defects." + "reason": "Both mention StableDiffusionControlNetInpaintPipeline, but one is a demo error and the other is a scheduler-quality regression; not clearly the same bug." }, { "left": "issue:5509", "right": "issue:6506", "accept": false, - "reason": "Image/prompt batch-size validation is unrelated to StableVideoDiffusion output_type handling." + "reason": "Image batch-size mismatch and StableVideoDiffusion returning a list instead of np.ndarray are different API/shape issues." }, { "left": "issue:2577", "right": "issue:7168", "accept": false, - "reason": "DDIMInverseScheduler issues and IP-Adapter embedding shape problems are different subsystems." + "reason": "DDIMInverseScheduler trouble and IP-Adapter image-embed shape confusion are unrelated." }, { "left": "issue:4321", "right": "issue:8922", "accept": false, - "reason": "Both mention black outputs, but one is DirectML FP16 rendering and the other is SD3 quantized T5xxl zero-output behavior." + "reason": "Both produce black outputs, but one is DirectML FP16 ONNX behavior and the other is SD3 single-file quantized-T5 zero-output behavior; different causes." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The summary is not conservative enough for this packet. All listed pairs have deterministic_accept=false, so marking issue:3595 vs issue:3661 as a true duplicate/subset relation overstates the evidence. At most, you can call it a plausible soft similarity, but not accept it as the true edge. The rest of the rejections are broadly grounded, but the one accepted soft edge makes the overall result ungrounded." + "accept": true, + "feedback": "The summary is grounded in the packet: all items are issues, no PRs are present, and the only clearly mergeable/same-bug pair is the from_ckpt reports (3595/3661). The other soft-edge rejections stay conservative and do not overclaim equivalence." }, "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": false, + "retried": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -37234,10 +37261,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -37252,156 +37279,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a duplicate set; the items are mostly unrelated bug reports that only share broad model/pipeline vocabulary. I would not merge any soft pairs.", - "confidence": 0.94, - "canonical_issue_reason": "issue:11464 is the clearest, most concrete Flux-specific failure report (exact pipeline and LoRA loading path), while the rest are broader or different bugs.", + "summary": "This cluster looks mostly like topic-similarity false positives rather than true duplicates. The only mild thematic overlap is around Flux/attention-mask or training/runtime issues, but the paired items still describe different pipelines, symptoms, or code paths. I would keep the Flux attention-mask issue as the best representative of the cluster theme.", + "confidence": 0.72, + "canonical_issue_reason": "issue:10025 is the clearest core Flux transformer attention/mask report and best represents the small thematic overlap in the cluster; the other items are mostly unrelated training, conversion, or pipeline failures.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11464 is the most actionable reference point because it names a specific failing code path and artifact, making it the best representative among these loosely related issues.", + "best_issue_reason": "issue:10025 is the most central and specific issue among the set for the only recurring theme (Flux transformer behavior).", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:7033", "right": "issue:9136", "accept": false, - "reason": "Both involve IP-Adapter, but one is warning logs about attention masking and the other is output quality at some resolutions; different symptoms and likely different fixes." + "reason": "Same broad IP-Adapter area, but one is attention-mask warning logs and the other is output quality at some resolutions; different bug/symptom." }, { "left": "issue:10797", "right": "issue:9971", "accept": false, - "reason": "Different pipelines and failure modes: HunyuanVideo group offloading vs FLUX DEIS scheduler behavior." + "reason": "Different pipelines and failures: HunyuanVideo group offloading vs FLUX scheduler behavior." }, { "left": "issue:1534", "right": "issue:435", "accept": false, - "reason": "Different training notebooks and distinct errors; both Dreambooth/textual inversion related, but not the same bug." + "reason": "Both are training notebook problems, but they are distinct errors in different notebooks and code paths." }, { "left": "issue:3238", "right": "issue:861", "accept": false, - "reason": "Both are model conversion issues, but one is LoRA safetensor conversion and the other is ckpt-to-diffusers conversion; not the same underlying problem." + "reason": "Both involve model conversion, but one is LoRA safetensor conversion and the other is checkpoint-to-diffusers conversion; not the same issue." }, { "left": "issue:11050", "right": "issue:7801", "accept": false, - "reason": "One is a specific dtype-casting bug in SD3 controlnet training; the other is a generic mixed_precision mismatch report." + "reason": "Both touch mixed precision, but one is a specific tensor-casting bug in a training script and the other is a model dtype expectation mismatch." }, { "left": "issue:2107", "right": "issue:4398", "accept": false, - "reason": "Both mention black or glitchy images, but one is a low-resolution generation issue and the other is an SDXL LoRA dreambooth pipeline result issue." + "reason": "Both mention bad/black images, but the first is low-resolution image generation and the second is an SDXL LoRA dreambooth pipeline issue." }, { "left": "issue:3222", "right": "issue:6686", "accept": false, - "reason": "Unrelated: missing custom diffusion weights after training versus Dynamo save/load performance regression." + "reason": "Unrelated problems: missing custom diffusion output file vs Dynamo save/load performance behavior." }, { "left": "issue:11464", "right": "issue:9717", "accept": false, - "reason": "Both are Flux-related, but one is LoRA loading in FluxControlPipeline and the other is a docker image error; different concrete failures." + "reason": "Different scopes: a specific FluxControlPipeline LoRA load failure vs a Flux-dev docker image error." }, { "left": "issue:4476", "right": "issue:8729", "accept": false, - "reason": "Both involve SDXL controlnet training, but one is poor results on fill50k and the other is an unexpected validation image; not the same bug." + "reason": "Both involve SDXL controlnet training, but the symptoms differ: poor results on fill50k vs unexpected validation images in a webdataset script." }, { "left": "issue:6520", "right": "issue:9179", "accept": false, - "reason": "Different symptoms and scopes: blank validation images in adapter training versus nonsensical SDXL+ControlNet training results." + "reason": "Different training scripts and outcomes: blank validation images vs nonsensical SDXL+ControlNet training results." }, { "left": "issue:4255", "right": "issue:7142", "accept": false, - "reason": "Both compare against webui, but they concern different models/pipelines and different quality regressions." + "reason": "Both compare diffusers to webui, but they are separate model/pipeline quality complaints." }, { "left": "issue:5570", "right": "issue:9223", "accept": false, - "reason": "Completely different problems: missing training utility function versus package dependency conflict." + "reason": "Completely unrelated: missing training utility function vs dependency version conflict." }, { "left": "issue:5510", "right": "issue:8668", "accept": false, - "reason": "Different code paths and errors: dtype mismatch in SDXL InstructPix2Pix versus SD3 img2img input-size constraint." + "reason": "Different failure modes in different pipelines: dtype mismatch in SDXL InstructPix2Pix vs SD3 image-size alignment error." }, { "left": "issue:11006", "right": "issue:12037", "accept": false, - "reason": "Both are Wan video-model issues, but one is broken output with quantized transformer and the other is lora compatibility with a different Wan model/version." + "reason": "Both concern Wan video models, but one is broken quantized I2V output and the other is LoRA compatibility with a different Wan version/model setup." }, { "left": "issue:10025", "right": "issue:12071", "accept": false, - "reason": "Both mention Flux transformers, but one is attention masking and the other is changed single-block forward behavior; not enough evidence of the same bug." + "reason": "Both are Flux transformer issues, but one is attention-mask handling and the other is a changed single-block forward behavior; not clearly the same bug." }, { "left": "issue:10954", "right": "issue:11418", "accept": false, - "reason": "Different user problems: LoRA loading failure versus asking how to add an FP8 model file." + "reason": "Different requests: LoRA loading failure vs how to add a specific Flux fill model file." }, { "left": "issue:1185", "right": "issue:2371", "accept": false, - "reason": "Both are Dreambooth-related, but reproducibility and missing objects in output are distinct issues." + "reason": "Both are Dreambooth-related, but reproducibility is not the same as missing training output objects." }, { "left": "issue:12107", "right": "issue:6351", "accept": false, - "reason": "Unrelated runtime errors in different subsystems: tracker initialization versus DDP adapter access." + "reason": "Unrelated runtime issues: tracker initialization with custom objects vs DDP missing disable_adapters." }, { "left": "issue:122", "right": "issue:4101", "accept": false, - "reason": "No meaningful overlap beyond training/conversion context; different errors and workflows." + "reason": "Different errors in different training/conversion paths; no concrete duplicate evidence." }, { "left": "issue:1398", "right": "issue:1537", "accept": false, - "reason": "Shared Apple MPS theme only; one is SD2 on MPS generally, the other is specific scheduler behavior on MPS." + "reason": "Both mention Apple MPS, but the reported bugs are different: SD2 on MPS vs specific scheduler issues on MPS." }, { "left": "issue:1475", "right": "issue:4954", "accept": false, - "reason": "One is about training with multiple class image types, the other is a textual inversion unpickling error." + "reason": "Different Dreambooth/textual inversion issues; one is dataset/class setup, the other is an unpickling error." }, { "left": "issue:1572", "right": "issue:4932", "accept": false, - "reason": "Different pipelines and failures: img2img/inpainting support versus missing pytorch_model.bin after Dreambooth." + "reason": "Different failures in different pipelines: img2img/inpainting support vs missing saved model artifact." }, { "left": "issue:1672", "right": "issue:2036", "accept": false, - "reason": "Different issues: wrong output size after conversion versus an OSError loading a model." + "reason": "Unrelated model-conversion size bug vs OSError loading a specific Stable Diffusion model." }, { "left": "issue:1949", "right": "issue:2552", "accept": false, - "reason": "Both are tutorial/training questions, but they concern different tutorials and different problems." + "reason": "Both are tutorial-related, but they concern different tutorials and different workflows." } ] }, @@ -37505,10 +37532,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -37523,172 +37550,169 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is mostly a mixed bag of unrelated diffusers issues. The only strong duplicate-like pair is the batch-size-dependent generation bug (2659/6222); the rest are at best same-subsystem reports, not the same underlying defect.", - "confidence": 0.87, - "canonical_issue_reason": "issue:6222 is the best canonical candidate because it states the underlying batch-size dependency more directly, which matches the duplicate-like pair with issue:2659.", + "summary": "The set is loosely related by diffusion-model training/loading topics, but the pairwise overlaps are too broad to treat as duplicates. Most items are distinct bugs or feature requests across different pipelines, models, and failure modes.", + "confidence": 0.44, + "canonical_issue_reason": "Issue 9484 is the best single representative of the cluster\u2019s strongest recurring theme: a concrete FLUX training problem. It is also open and specific enough to anchor the small FLUX/LoRA/controlnet subset better than the older, more divergent items.", "canonical_pr_reason": null, - "best_issue_reason": "issue:6222 is the clearest representative of the only true overlap in this set: a batch-size-dependent scheduler/output bug.", + "best_issue_reason": "9484 is the most useful cluster representative because it is current, concrete, and centered on FLUX training/deepspeed behavior rather than a generic docs/import/conversion request.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:2218", "right": "issue:5633", "accept": false, - "reason": "Different problems: training quality/overtraining vs a transformer_temporal bug." + "reason": "Both mention bugs, but one is a training overfitting report and the other is a transformer model bug; they are not the same underlying issue." }, { "left": "issue:2838", "right": "issue:2858", "accept": false, - "reason": "Both mention pipelines, but one is about ControlNet behavior and the other is about img2img seeding; not the same bug." + "reason": "One is about ControlNet pipeline behavior and the other about adding a seeding image to an UnCLIP img2img pipeline; different features and failures." }, { "left": "issue:2991", "right": "issue:6540", "accept": false, - "reason": "Both involve ControlNet training, but the failures are different: mixed-precision example error vs FSDP matmul shape issue." + "reason": "Both are ControlNet training bugs, but mixed precision failure and FSDP matrix-shape failure are different code-path problems." }, { "left": "issue:3065", "right": "issue:4304", "accept": false, - "reason": "One is a feature request for text-encoder LoRA in DreamBooth, the other is SDXL LoRA support; not the same change." + "reason": "Both concern LoRA training support, but one is a text-encoder option for DreamBooth and the other is SDXL text-to-image LoRA support; not the same change." }, { "left": "issue:3112", "right": "issue:6372", "accept": false, - "reason": "Different pipelines and goals: multi-ControlNet inpainting vs SVD+ControlNet support." + "reason": "Multi-ControlNet inpainting and SVD+ControlNet are different pipeline combinations; too broad to merge." }, { "left": "issue:3418", "right": "issue:3480", "accept": false, - "reason": "Related to text-encoder LoRA, but one adds training support while the other asks for a scale parameter; separate changes." + "reason": "One adds text-encoder handling to a training script; the other adds a scale parameter for text-encoder LoRA layers. Related area, but different changes." }, { "left": "issue:3926", "right": "issue:5285", "accept": false, - "reason": "Docs-related, but unrelated topics: an example request vs a translation request." + "reason": "These are both docs tasks, but one is an example request and the other is a translation request; not the same issue." }, { "left": "issue:5178", "right": "issue:5917", "accept": false, - "reason": "Both are generation/conversion bugs, but the concrete failures are different and on different code paths." + "reason": "Different bugs in different components: scheduler resolution dependence versus checkpoint conversion breaking after an upgrade." }, { "left": "issue:5892", "right": "issue:7021", "accept": false, - "reason": "Both are import errors, but the reported causes are different; no clear shared root bug." + "reason": "Both are import errors, but the titles indicate different causes and environments; not enough to \u0441\u0447\u0438\u0442\u0430\u0442\u044c them duplicates." }, { "left": "issue:5917", "right": "issue:5963", "accept": false, - "reason": "Checkpoint conversion failure vs Kandinsky embed input handling; not the same defect." + "reason": "Checkpoint conversion failure and Kandinsky embed-vs-prompt handling are unrelated failures." }, { "left": "issue:7168", "right": "issue:9136", "accept": false, - "reason": "Both mention IP Adapter, but one is a tensor-shape/embed issue and the other is output quality at certain resolutions." + "reason": "Both mention IP-Adapter, but one is about malformed image embeddings and the other about output quality at certain resolutions; different problems." }, { "left": "issue:7237", "right": "issue:9424", "accept": false, - "reason": "Missing scheduler class vs a notebook breakage report; not the same underlying issue." + "reason": "A missing scheduler class and a broken notebook are unrelated issues." }, { "left": "issue:11241", "right": "issue:9484", "accept": false, - "reason": "Both mention FLUX training, but one is a ControlNet floating-point crash and the other is a multi-GPU DreamBooth/deepspeed training request/issue." + "reason": "Both involve FLUX training, but one is a ControlNet floating-point crash and the other is multigpu/deepspeed DreamBooth training; not the same bug." }, { "left": "issue:2549", "right": "issue:4213", "accept": false, - "reason": "Different failures: SD2 UNet/model compatibility vs from_single_file requiring accelerate." + "reason": "Different model-loading failures: SD2 UNet incompatibility versus SDXL from_single_file needing accelerate." }, { "left": "issue:2659", "right": "issue:6222", - "accept": true, - "reason": "Both describe the same batch-size-dependent diffusion output/scheduler bug; 6222 is the clearer root-cause report." + "accept": false, + "reason": "Both mention batch size, but one is image variation/nondeterminism and the other is DDPM scheduler state dependence; too different." }, { "left": "issue:1541", "right": "issue:3296", "accept": false, - "reason": "DreamBooth ckpt/model-type loading vs checkpoint save/load with text_encoder; related area, different bug." + "reason": "Both are DreamBooth checkpoint/loading issues, but model-type detection and save/load with text_encoder are separate bugs." }, { "left": "issue:2363", "right": "issue:6584", "accept": false, - "reason": "Both are conversion questions, but one is LoRA conversion and the other is a specific model conversion problem." + "reason": "Both are conversion questions, but LoRA conversion and DreamShaper inpaint conversion are not the same underlying task." }, { "left": "issue:4852", "right": "issue:8493", "accept": false, - "reason": "Different loading failures on different model classes; not the same code-path problem." + "reason": "Both are load failures, but a generic LoRA load error and SD3Transformer2DModel state_dict mismatch are different problems." }, { "left": "issue:7690", "right": "issue:8359", "accept": false, - "reason": "Tokenizer loading failure vs from_single_file custom checkpoint failure; different root causes." + "reason": "Tokenizer loading and custom checkpoint from_single_file failures are unrelated." }, { "left": "issue:7133", "right": "issue:7138", "accept": false, - "reason": "Both are SDXL mismatch reports, but one is checkpoint resume text-encoder size mismatch and the other is UNet structure mismatch." + "reason": "Both concern SDXL structure mismatches, but one is text-encoder resume size mismatch and the other is UNet structure mismatch; different components." }, { "left": "issue:11791", "right": "issue:12319", "accept": false, - "reason": "Both involve group offloading, but one is LoRA weight loading after offloading and the other is a broader block_level offloading bug; not clearly the same defect." + "reason": "Both involve group offloading/LoRA loading, but one is inability to load LoRA after offloading and the other is block_level offloading being broken; not the same failure mode." }, { "left": "issue:11592", "right": "issue:9613", "accept": false, - "reason": "FluxLoraLoaderMixin delete_adapters regression vs a Timesteps.scale attribute error; unrelated." + "reason": "A FluxLoraLoaderMixin regression and a Timesteps.scale attribute error are different code-path bugs." }, { "left": "issue:6382", "right": "issue:8649", "accept": false, - "reason": "IP adapter key error during SDXL LCM LoRA training vs SD3 num_images_per_prompt behavior; different problems." + "reason": "IP-Adapter key errors during training/inference and SD3 num_images_per_prompt handling are unrelated." }, { "left": "issue:1672", "right": "issue:6382", "accept": false, - "reason": "Checkpoint-to-diffusers image size issue vs IP adapter key error; unrelated." + "reason": "Wrong image size after ckpt conversion and IP-Adapter key errors after LCM LoRA training are different issues." }, { "left": "issue:1931", "right": "issue:4709", "accept": false, - "reason": "Float16 type mismatch in SD2 vs SDXL ControlNet guess mode support; different code paths." + "reason": "Float16 type mismatch in SD2 and SDXL ControlNet guess-mode support are unrelated." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is well grounded in the packet: the only clearly supported duplicate-like pair is 2659/6222, and the rest of the soft-edge rejections stay conservative and consistent with the issue titles. I don\u2019t see overstatement beyond what the packet supports." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -37783,10 +37807,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -37801,156 +37825,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous: dtype/device runtime errors, unrelated pipeline feature requests, and performance/model-support issues. None of the soft pairs looks like the same underlying bug or change with enough specificity to merge.", - "confidence": 0.91, - "canonical_issue_reason": "issue:11436 is the clearest concrete bug report in the set: a reproducible runtime dtype mismatch with a specific failure message and context.", + "summary": "This cluster is mostly a grab bag of unrelated diffusers issues: dtype/device mismatches, pipeline feature requests, loading failures, performance complaints, and example-script bugs. The soft-similarity pairs only share broad vocabulary (e.g. \u201csame dtype\u201d, \u201cSDXL\u201d, \u201cpipeline\u201d) and do not look like the same underlying bug or change.", + "confidence": 0.95, + "canonical_issue_reason": "No single issue clearly unifies the cluster. If one must be picked, #11149 is the most concrete, self-contained bug report, but it is not a good duplicate umbrella for the others.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11436 is the best representative issue here because it is specific, actionable, and describes an actual runtime defect rather than a feature request or vague incompatibility.", + "best_issue_reason": "#11149 is the strongest representative as a precise runtime error with a narrow code path and clear symptom; most other items are either different error classes or feature requests.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3401", "right": "issue:5897", "accept": false, - "reason": "Both are training-time errors, but they concern different failures and code paths: class label validation vs multi-device LoRA training." + "reason": "Both mention tensor/device-related training errors, but one is about missing class labels for class embeddings and the other is a LoRA device placement bug." }, { "left": "issue:4198", "right": "issue:5075", "accept": false, - "reason": "Both mention matmul dtype/shape errors, but the underlying problems are different: a notebook dtype mismatch vs SDXL Refiner prompt-embedding shape mismatch." + "reason": "Both are dtype/matmul failures, but they occur in different contexts: a notebook example versus SDXL refiner prompt-embed shape mismatch." }, { "left": "issue:11978", "right": "issue:12358", "accept": false, - "reason": "These are separate pipeline-addition requests for different models; same category, not the same change." + "reason": "These are separate model pipeline requests for different models, not the same bug or change." }, { "left": "issue:1478", "right": "issue:2495", "accept": false, - "reason": "Both involve StableDiffusionUpscalePipeline, but one is an output-count limitation and the other is an MPS black-image bug." + "reason": "Both involve StableDiffusionUpscalePipeline, but one is about multi-image output support and the other is an MPS black-image rendering bug." }, { "left": "issue:2663", "right": "issue:4250", "accept": false, - "reason": "These are different model-loading bugs with different failure modes; not the same defect." + "reason": "One is a safetensors loading failure for a safety checker; the other is a generic from_pretrained loading issue." }, { "left": "issue:4206", "right": "issue:5997", "accept": false, - "reason": "SDXL ControlNet training being broken and a tutorial type-mismatch error are distinct issues with different scopes." + "reason": "Both are SDXL/controlnet-related, but one is a training breakage and the other is a tutorial/model-type mismatch error." }, { "left": "issue:11436", "right": "issue:720", "accept": false, - "reason": "Both are dtype-related runtime errors, but they occur in different pipelines and code paths, so they are not clearly the same bug." + "reason": "Both are dtype mismatch errors, but the underlying paths differ: Hidream inference matmul dtype mismatch versus traced UNet half/float tracing issue." }, { "left": "issue:4773", "right": "issue:5319", "accept": false, - "reason": "Loading a CivitAI LoRA and speeding up LoRA loading with PEFT are different concerns; one is a bug report, the other a performance request." + "reason": "These are unrelated LoRA-loading concerns: one is CivitAI LoRA loading, the other is a proposal to speed up LoRA loading with PEFT." }, { "left": "issue:7931", "right": "issue:9495", "accept": false, - "reason": "Both mention SDXL/IP-Adapter, but one targets ControlNet inpaint with image embeds and the other PAG behavior; no clear same underlying defect." + "reason": "Both involve SDXL plus IP-Adapter-style conditioning, but one is a ControlNet inpaint parameter bug and the other is a PAG + IPAdapter failure." }, { "left": "issue:8233", "right": "issue:9953", "accept": false, - "reason": "Both touch precision/quantization problems, but one is an fp16 depth-pipeline error and the other is moving a quantized component to CUDA." + "reason": "Both mention fp16/quantization-adjacent behavior, but one is a depth pipeline fp16 error and the other is moving a quantized component to CUDA." }, { "left": "issue:7357", "right": "issue:7785", "accept": false, - "reason": "Both are slowdown reports, but one is about compile/fuse_loras inference speed and the other is a first-inference regression in a newer diffusers release." + "reason": "Both are performance complaints, but one is about torch.compile/fuse_loras on SDXL and the other is slow first inference in diffusers 0.27.x." }, { "left": "issue:3209", "right": "issue:3347", "accept": false, - "reason": "Both mention NaNs, but one is confined to AutoencoderKL on large images while the other is a broader pipeline output issue." + "reason": "Both mention NaNs in diffusion outputs, but one is specifically AutoencoderKL encoder NaNs for large images and the other is a general pipeline producing NaN image arrays." }, { "left": "issue:3540", "right": "issue:3937", "accept": false, - "reason": "These are different pipeline-specific failures: prompt-embedding support in TextToVideoZero vs a ZeroscopeXL loading/usage problem." + "reason": "These are unrelated pipeline issues: prompt embeddings on TextToVideoZeroPipeline versus ZeroscopeXL not working as a normal model." }, { "left": "issue:540", "right": "issue:725", "accept": false, - "reason": "Both concern memory usage, but one is about keeping only UNet on GPU during inference and the other is about device_map/accelerate loading behavior." + "reason": "Both concern memory/offload behavior, but one proposes keeping only UNet on GPU while the other asks for device_map/accelerate support in DiffusionPipeline." }, { "left": "issue:6238", "right": "issue:9844", "accept": false, - "reason": "Both involve SDXL VAE, but the reports are too vague and likely refer to different failure modes; not enough evidence to merge." + "reason": "Both are VAE encoder NaN reports, but they target different model families and are not clearly the same failure mode." }, { "left": "issue:3432", "right": "issue:3565", "accept": false, - "reason": "These are both resource-related training issues, but one is a Half-CPU kernel limitation and the other is CUDA OOM." + "reason": "Both are CUDA/half precision training/inference problems, but one is a CPU half-kernel runtime error and the other is a text-to-image training OOM." }, { "left": "issue:5281", "right": "issue:7658", "accept": false, - "reason": "One is a device-placement error with multiple XL pipelines, the other is a feature question about multiple conditioning in SDXL ControlNet." + "reason": "Both are multi-component SDXL issues, but one is a device placement error with CPU offload and the other is a question about multiple conditioning with ControlNet." }, { "left": "issue:3564", "right": "issue:5284", "accept": false, - "reason": "Both are support requests, but for different ControlNet variants and different implementations." + "reason": "One is a request for Uni-ControlNet support and the other is a request for ControlNet-LLLite support; they are different features." }, { "left": "issue:3761", "right": "issue:6323", "accept": false, - "reason": "Loading two community pipelines and a TensorRT pipeline attribute error are unrelated." + "reason": "One is about loading two community pipelines in the same process; the other is a TensorRT img2img attribute-setting bug." }, { "left": "issue:11149", "right": "issue:11436", "accept": false, - "reason": "Both involve dtype/quantization, but one is an unexpected-argument API error while the other is a runtime matrix dtype mismatch." + "reason": "Both are dtype-related runtime errors, but one is an unexpected keyword argument in quantizer code and the other is a mat1/mat2 dtype mismatch in inference." }, { "left": "issue:1451", "right": "issue:5224", "accept": false, - "reason": "Both are Flax Stable Diffusion issues, but they involve different models and different failures." + "reason": "These are unrelated Flax/JAX Stable Diffusion errors with different model/version contexts." }, { "left": "issue:6159", "right": "issue:7978", "accept": false, - "reason": "A missing denoising_end argument in an SDXL ControlNet pipeline is unrelated to a torchvision operator registration error." + "reason": "One is an unsupported denoising_end argument in an SDXL ControlNet pipeline; the other is a torchvision operator registration failure." }, { "left": "issue:3980", "right": "issue:4460", "accept": false, - "reason": "Both are about output differences, but one is tied to prompt length/tokenization and the other is a broad diffusers-vs-stable-diffusion comparison." + "reason": "Both mention different outputs versus other implementations, but one is a long-prompt tokenization issue and the other is a general stable-diffusion versus diffusers result difference." }, { "left": "issue:3363", "right": "issue:3454", "accept": false, - "reason": "These both reference train_dreambooth_lora.py, but the reports describe different environments and failure modes, so they are not clearly the same bug." + "reason": "Both are train_dreambooth_lora.py failures, but they are described as different machine/environment breakages and do not indicate the same concrete bug." } ] }, @@ -38052,10 +38076,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -38070,156 +38094,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is largely heterogeneous. The only loose commonality is a few Flux-related issues, but they target different examples, models, and code paths. None of the proposed soft pairs look like true duplicates or mergeable into one fix.", - "confidence": 0.84, - "canonical_issue_reason": "Issue 11010 is the broadest Flux-related item in the set and has the most engagement; it is the best thematic anchor, though the cluster overall is not a clean duplicate group.", + "summary": "These items are mostly unrelated Diffusers issues that only share broad model/training/loading vocabulary; none of the soft pairs look like the same underlying bug or change.", + "confidence": 0.94, + "canonical_issue_reason": "No single canonical issue stands out: the set spans distinct bugs across Flux training, loading, checkpointing, memory, and UI/docs behavior rather than one duplicate report.", "canonical_pr_reason": null, - "best_issue_reason": "11010 is the best representative if one issue must anchor the cluster, because it frames a general Flux model-support change rather than a narrow script-specific failure.", + "best_issue_reason": "issue:9227 is the most concrete and actionable report in the set, with a specific Flux training script, model, and failure context; it is the strongest standalone issue even though it does not unify the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1445", "right": "issue:1500", "accept": false, - "reason": "Both are compatibility bugs, but they hit different pipelines and different failure modes; not the same underlying issue." + "reason": "Both are pipeline compatibility problems, but one is an inpaint legacy AttributeError and the other is a community pipeline/version incompatibility; different code paths and fixes." }, { "left": "issue:1866", "right": "issue:5465", "accept": false, - "reason": "Both mention strength values, but one is scheduler behavior in img2img and the other is a DiffEdit pipeline limitation; different code paths." + "reason": "Both mention strength-related behavior, but one is a scheduler/img2img failure at certain strengths and the other is a DiffEdit-only inpaint_strength limitation; not the same bug." }, { "left": "issue:4993", "right": "issue:6283", "accept": false, - "reason": "One is a missing export/import-structure bug, the other is a ControlNet img2img strength=0 behavior issue." + "reason": "CLIPImageProjection import/export structure vs SDXL ControlNet strength=0 behavior are unrelated failure modes." }, { "left": "issue:1672", "right": "issue:867", "accept": false, - "reason": "Image-size conversion bug vs GPU memory retention bug; unrelated symptoms and fixes." + "reason": "Wrong output size after conversion and GPU memory not being released are different bugs with different symptoms and fixes." }, { "left": "issue:4566", "right": "issue:4733", "accept": false, - "reason": "Both are checkpoint-resume training issues, but they affect different training scripts and appear to stem from different resume logic." + "reason": "Both involve checkpoint resume, but one is SDXL LoRA resume support and the other is DreamBooth LoRA optimizer state validation; not the same concrete defect." }, { "left": "issue:5678", "right": "issue:7722", "accept": false, - "reason": "Higher-order autograd attention bug vs Animatediff+IPAdapter integration error; no shared concrete failure." + "reason": "Higher-order derivative errors in attention math and an Animatediff+IPAdapter pipeline error are unrelated." }, { "left": "issue:1404", "right": "issue:1896", "accept": false, - "reason": "Both involve CLIP Guided Stable Diffusion, but one is a feature request/topic and the other is a specific size/type error." + "reason": "Both are CLIP-guided SD issues, but one is about a pipeline/model support question and the other is a size/type runtime error; too different to merge." }, { "left": "issue:1823", "right": "issue:2880", "accept": false, - "reason": "A comment swap in depth2img is unrelated to a JAX Colab startup failure." + "reason": "A swapped comment typo and a JAX Colab runtime failure are not the same bug." }, { "left": "issue:2420", "right": "issue:6277", "accept": false, - "reason": "Both concern LoRA training, but one is xformers attention compatibility and the other is broken gradients; different bugs." + "reason": "xFormers attention enabling in LoRA training and broken gradients are different training-path failures." }, { "left": "issue:3026", "right": "issue:6230", "accept": false, - "reason": "InstructPix2Pix launch error vs SDXL training OOM; no shared underlying defect." + "reason": "Instruct-pix2pix accelerate launch error on one GPU vs SDXL text-to-image OOM are different issues." }, { "left": "issue:3694", "right": "issue:7682", "accept": false, - "reason": "Missing model/config files in a finetuning example is unrelated to multi-GPU performance issues." + "reason": "Missing model/config files in the training example versus multi-GPU SD1.5 execution issues are not the same underlying bug." }, { "left": "issue:5578", "right": "issue:9793", "accept": false, - "reason": "Model-index lookup error vs progress bar update issue; entirely different code paths." + "reason": "Missing model_index.json loading error and a progress-bar update bug are unrelated." }, { "left": "issue:12078", "right": "issue:9227", "accept": false, - "reason": "Both are Flux-related, but one is about example validation input and the other is a FLUX.1-dev training-script problem; not the same bug." + "reason": "Both are Flux-related, but one is an example validation-input problem and the other is a DreamBooth LoRA training failure; different scripts and code paths." }, { "left": "issue:11010", "right": "issue:9270", "accept": false, - "reason": "Both mention Flux, but one is a broad architecture-support request and the other is incomplete LoRA support from a separate toolchain." + "reason": "Both mention Flux support, but one is about architecture changes for Chroma and the other about incomplete LoRA support in SimpleTuner; not a single concrete bug." }, { "left": "issue:1445", "right": "issue:852", "accept": false, - "reason": "Different pipelines and different version-compatibility failures; not the same issue." + "reason": "Both are compatibility crashes, but they affect different pipelines and different API surfaces (legacy inpaint vs MPS device_map handling)." }, { "left": "issue:6183", "right": "issue:9208", "accept": false, - "reason": "Both involve from_single_file loading, but one is an incorrect-shape result and the other is a ControlNet model-loading failure." + "reason": "Both are from_single_file loading issues, but one is incorrect tensor shape and the other is inability to load ControlNet models; distinct failure modes." }, { "left": "issue:6561", "right": "issue:6832", "accept": false, - "reason": "IP-Adapter loading/config-path error vs Face ID tuple/shape error; unrelated failures." + "reason": "An OSError about missing image_encoder config and an AttributeError in IP-Adapter Face ID are different bugs in the same broad area, but not the same fix." }, { "left": "issue:7333", "right": "issue:9873", "accept": false, - "reason": "New pipeline/model feature request vs an OmniGen pipeline addition; same broad area, but not the same underlying change." + "reason": "Image-to-video generation support and OmniGen pipeline addition are different feature requests/issues." }, { "left": "issue:1807", "right": "issue:3124", "accept": false, - "reason": "Performance degradation over generations is unrelated to a conversion failure." + "reason": "A performance degradation over successive generations and a model conversion failure are unrelated." }, { "left": "issue:5719", "right": "issue:8898", "accept": false, - "reason": "Both are loader-path validation issues, but one is LoRA weight file precedence and the other is subfolder path handling; not one concrete bug." + "reason": "Both involve loading-path logic, but one is LoRA weight file precedence and the other is subfolder loading with model files; different conditions and fixes." }, { "left": "issue:2459", "right": "issue:4932", "accept": false, - "reason": "Different training/export bugs with no shared failure path." + "reason": "xFormers enabling in LoRA training and missing pytorch_model.bin output from DreamBooth are separate training/export issues." }, { "left": "issue:3565", "right": "issue:7480", "accept": false, - "reason": "Both are OOM reports, but on different examples and models; too generic to merge." + "reason": "Both are OOM reports, but they concern different examples/models and likely different memory causes." }, { "left": "issue:3975", "right": "issue:5128", "accept": false, - "reason": "A typo in an error message is unrelated to a local safetensors loading failure." + "reason": "A typo in an error message and a failure loading a local safetensors file are not the same bug." }, { "left": "issue:2089", "right": "issue:3647", "accept": false, - "reason": "Both are DreamBooth-related, but one is AltDiffusion OOM and the other is DeepFloyd training; different model-specific issues." + "reason": "Both are training OOMs, but they involve different models and workflows (AltDiffusion DreamBooth vs DeepFloyd DreamBooth)." } ] }, @@ -38323,10 +38347,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -38341,162 +38365,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated bugs, feature requests, and training/inference issues; none of the soft pairs look like true duplicates.", - "confidence": 0.97, - "canonical_issue_reason": "issue:12348 is the strongest representative bug report here: it is concrete, reproducible, and narrowly scoped to a specific tensor-mismatch failure in a named pipeline.", + "summary": "Loose keyword cluster around diffusion/LoRA/Flux/image-video problems, but the proposed pairings are mostly superficial (same terms like tensor mismatch, LoRA, image-to-video, or training) rather than the same underlying bug/change. I would not merge any of the soft edges.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 12348 is the clearest concrete bug report in the set (specific tensor mismatch, reproducible pipeline failure) and is more representative than the broader feature requests or unrelated training/runtime errors.", "canonical_pr_reason": null, - "best_issue_reason": "issue:12348 stands out as the most actionable issue: clear error mode, specific model/pipeline, and a well-defined trigger condition.", + "best_issue_reason": "Issue 12348 is the strongest single issue to anchor triage because it is specific, technically actionable, and not phrased as a broad request or ambiguous complaint.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10823", "right": "issue:8514", "accept": false, - "reason": "Both are training-related, but one is about missing inference artifacts after fine-tuning and the other is an EMAModel parallel fine-tuning error; different failure modes and code paths." + "reason": "Both are training-related, but one is about missing inference artifacts after fine-tuning while the other is an EMAModel parallel fine-tuning error; different code paths and symptoms." }, { "left": "issue:2883", "right": "issue:7333", "accept": false, - "reason": "An image-to-video pipeline feature request is not the same as a noise-rectification research request; they target different changes." + "reason": "One requests an image-to-video pipeline; the other is about a tuning-free noise rectification method for high-fidelity i2v generation. Same domain, different change." }, { "left": "issue:3812", "right": "issue:6870", "accept": false, - "reason": "One is a missing config key for ControlNet, the other is a missing VAE tiling option in img2img/inpaint pipelines; unrelated bugs." + "reason": "Both mention missing pipeline/config options, but they target different flags in different pipelines (`controlnet` config vs `enable_vae_tiling`)." }, { "left": "issue:11408", "right": "issue:9051", "accept": false, - "reason": "set_adapters failing on compiled models and Flux batch inference being unsupported are different API/runtime issues." + "reason": "`set_adapters` on compiled models and Flux batch inference support are unrelated feature/compatibility issues." }, { "left": "issue:11014", "right": "issue:3094", "accept": false, - "reason": "Both mention OOM, but the repros are different workloads: FLUX on A100 vs a DreamBooth finetune example." + "reason": "Both are OOM reports, but they occur in different training scripts and environments; not enough evidence of the same bug." }, { "left": "issue:2806", "right": "issue:6911", "accept": false, - "reason": "Custom pipeline printing behavior and UNetConditional3DModel class labels are unrelated subsystem bugs." + "reason": "Different problems in different components: printing pipeline customization vs class labels not used in a 3D model." }, { "left": "issue:4835", "right": "issue:5395", "accept": false, - "reason": "MultiControlNet img2img behavior and LoRA weight-combination issues are distinct problems affecting different components." + "reason": "MultiControlNet img2img bug and PEFT LoRA-combination bug are separate code paths and separate failure modes." }, { "left": "issue:12034", "right": "issue:12216", "accept": false, - "reason": "Both compare against reference outputs, but they involve different models and different quality regressions." + "reason": "Both compare output quality, but they concern different models and workflows (Wan i2v vs Qwen image editing)." }, { "left": "issue:1188", "right": "issue:2015", "accept": false, - "reason": "These are separate feature requests for different pipeline types; same broad theme but not the same change." + "reason": "A CLIPSeg+Stable Diffusion pipeline request is unrelated to a categorical diffusion pipeline request." }, { "left": "issue:1195", "right": "issue:1538", "accept": false, - "reason": "Memory-efficient attention with fp16 weights and SD2-base pipeline compatibility are different bugs." + "reason": "Memory-efficient attention with fp16 weights is a distinct issue from SD2-base failing in StableDiffusionMegaPipeline." }, { "left": "issue:3393", "right": "issue:4386", "accept": false, - "reason": "Both ask about LoRA conversion, but one is an A1111 safetensors-to-bin conversion question and the other is a different SDXL conversion question." + "reason": "Both involve LoRA conversion, but one asks about A1111 safetensors-to-bin conversion generally and the other is specifically about SDXL; too broad to merge." }, { "left": "issue:6008", "right": "issue:6382", "accept": false, - "reason": "Both mention IP-Adapter and LCM LoRA, but one is an inference-time failure and the other is a training/output key mismatch; not the same underlying bug." + "reason": "Both involve IP-Adapter and LCM LoRA, but one is a runtime compatibility complaint and the other is a training-script key error after distillation." }, { "left": "issue:9343", "right": "issue:9717", "accept": false, - "reason": "FLUX loading with mismatched sizes and a Flux-dev docker image error are different contexts and failure points." + "reason": "Both are FLUX-related, but one is a loading argument mismatch and the other is a docker image error; not the same bug." }, { "left": "issue:1717", "right": "issue:5678", "accept": false, - "reason": "Offline cache loading and higher-order-derivative errors in attention are entirely unrelated." + "reason": "Offline cache loading and higher-order derivative failures are unrelated." }, { "left": "issue:10226", "right": "issue:12348", "accept": false, - "reason": "Both report tensor mismatches, but they involve different models, scripts, and circumstances; training DreamBooth Flux with LoRA is not the same bug as Wan image-to-video resolution-dependent mismatch." + "reason": "Both mention tensor mismatches, but they are different models/pipelines (Dreambooth Flux LoRA vs Wan image-to-video) with no shared code path shown." }, { "left": "issue:1166", "right": "issue:3886", "accept": false, - "reason": "A TPU save hang in textual inversion and a distributed training watchdog timeout are different training failures." + "reason": "TPU save hangs in textual inversion training is unrelated to distributed watchdog timeouts during DDPM training." }, { "left": "issue:4584", "right": "issue:4736", "accept": false, - "reason": "Resume-from-checkpoint failure and OOM/invalid-cast during SDXL LoRA training are different bugs in the same script family." + "reason": "Same script, but one is a checkpoint-resume error and the other is OOM/cast instability; different concrete defects." }, { "left": "issue:11359", "right": "issue:11374", "accept": false, - "reason": "These are separate feature requests for different models and capabilities, not the same change." + "reason": "Two separate feature requests for different models: faster inference for LTX-Video vs adding SkyReels-V2 support." }, { "left": "issue:1188", "right": "issue:1392", "accept": false, - "reason": "A CLIPSeg pipeline request and a Stable Diffusion 2 issue are unrelated." + "reason": "A pipeline request and a general Stable Diffusion 2 issue are not the same bug/change." }, { "left": "issue:3235", "right": "issue:4139", "accept": false, - "reason": "Both involve custom diffusion, but one is a loading failure for biased weights and the other is a precision-casting concern in attention processing." + "reason": "Both are Custom Diffusion-related, but one is about loading biased weights and the other about precision casting in an attention processor." }, { "left": "issue:3280", "right": "issue:3289", "accept": false, - "reason": "Non-default schedulers failing in DeepFloyd IF and super-resolution breaking at non-64x64 inputs are different bugs." + "reason": "Non-default scheduler failure and DeepFloyd super-resolution artifacts are separate pipeline bugs." }, { "left": "issue:4450", "right": "issue:6092", "accept": false, - "reason": "Multi-text inversion not working and LoRA training not doing anything are separate example-script issues." + "reason": "Multi-text-inversion not working and LoRA example not training are different training features and failure modes." }, { "left": "issue:3487", "right": "issue:6870", "accept": false, - "reason": "Poor inpaint results with a specific scheduler and a missing enable_vae_tiling option are not the same defect." + "reason": "Inpaint quality with EulerAncestralDiscreteScheduler and missing `enable_vae_tiling` support are unrelated inpaint issues." }, { "left": "issue:3301", "right": "issue:6894", "accept": false, - "reason": "Both are about LoRA training, but they concern different scripts/models and different breakages; too broad to merge." + "reason": "Both are LoRA training complaints, but they involve different scripts, model scales, and symptoms." }, { "left": "issue:4484", "right": "issue:6639", "accept": false, - "reason": "Weird SDXL Kohya LoRA outputs and a discrepancy between combined vs separate LoRA loading are related to LoRA behavior but describe different bugs and code paths." + "reason": "Weird SDXL Kohya LoRA results and differing outputs from `load_lora_weights` vs manual loading are related to LoRA behavior, but they are not the same concrete bug." } ] }, @@ -38598,10 +38622,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -38616,156 +38640,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is a loose collection of unrelated issues; the soft-similarity pairs mostly share only broad subsystem words (FLUX, IP-Adapter, training, dtype, etc.) and do not describe the same concrete bug or change. No PRs are present.", - "confidence": 0.94, - "canonical_issue_reason": "Issue 8955 is the most concrete, well-documented bug report in the set, with a specific runtime error and detailed context, so it is the best anchor if one must pick a representative issue.", + "summary": "The set is mostly a grab bag of unrelated issues and feature requests; none of the soft-similarity pairs look like true duplicates or mergeable PRs.", + "confidence": 0.88, + "canonical_issue_reason": "No single canonical issue stands out because the items span distinct bugs, models, and feature requests rather than one underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 8955 has the clearest failure mode and strongest reproduction detail, making it the most suitable standalone issue for triage.", + "best_issue_reason": "Issue 8955 is the strongest standalone report: it has the most detailed reproduction and a concrete CLIPTextModelWithProjection state_dict loading failure.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:8955", "right": "issue:9667", "accept": false, - "reason": "Both mention FLUX/loading, but one is a CLIPTextModelWithProjection state-dict load error and the other is an fp8/local transformer file issue; different failure paths." + "reason": "Both involve FLUX loading, but one is a CLIPTextModelWithProjection state_dict error and the other is an fp8/local transformer file loading problem; different failure paths." }, { "left": "issue:6832", "right": "issue:8817", "accept": false, - "reason": "Both involve IP-Adapter FaceID, but one reports a tuple/shape AttributeError and the other is a generic inability to use FaceID; not the same concrete bug." + "reason": "Both mention IP-Adapter FaceID, but the concrete bugs differ: tuple.shape AttributeError vs a generic \"can't use\" report." }, { "left": "issue:3826", "right": "issue:8414", "accept": false, - "reason": "These are different requests for different new audio pipelines/models; same general area, but not the same underlying change." + "reason": "Both are new model/pipeline requests, but they target different audio-generation models and different implementations." }, { "left": "issue:9744", "right": "issue:9873", "accept": false, - "reason": "Both are model-addition requests, but for different models (Mochi vs OmniGen) and different pipelines." + "reason": "Different feature requests: Mochi video model vs OmniGen image generation model." }, { "left": "issue:11045", "right": "issue:8832", "accept": false, - "reason": "One is a FLUX DreamBooth load crash; the other is a generic pipeline image-generation crash. Too vague and not the same code path." + "reason": "One is a FLUX Dreambooth crash, the other is a generic pipe(prompt).images[0] runtime crash; no shared code path." }, { "left": "issue:9051", "right": "issue:9089", "accept": false, - "reason": "One is about FLUX batch inference support, the other about a missing FluxPipeline class; related subsystem but different issues." + "reason": "Batch inference support for FluxPipeline is unrelated to FluxPipeline import/not-found errors." }, { "left": "issue:11581", "right": "issue:11627", "accept": false, - "reason": "Both mention IP-Adapter, but the concrete error and pipeline/model context differ (FLUX dtype kwarg vs SD3.5 pipeline usage)." + "reason": "Both involve adapters, but one is a dtype kwarg TypeError in CLIPVisionModelWithProjection and the other is an SD3 pipeline integration issue." }, { "left": "issue:5465", "right": "issue:9109", "accept": false, - "reason": "DiffEdit inpaint-strength behavior and torchvision::nms import failure are unrelated bugs." + "reason": "DiffEdit inpaint strength behavior is unrelated to the missing torchvision::nms operator error." }, { "left": "issue:11436", "right": "issue:1931", "accept": false, - "reason": "Both are dtype mismatch reports, but they affect different models and pipelines with different root causes." + "reason": "Both are dtype mismatch errors, but they hit different models and code paths (Hidream inference vs SD2 training)." }, { "left": "issue:2645", "right": "issue:5254", "accept": false, - "reason": "Mac/CUDA assertion failure is unrelated to Multi-ControlNet support in the inpaint pipeline." + "reason": "A mac M1 CUDA assertion is a bug; multi-ControlNet support is a feature request." }, { "left": "issue:4419", "right": "issue:7221", "accept": false, - "reason": "Offset-noise inference mismatch is not the same as a noisy output complaint with a specific checkpoint." + "reason": "Offset-noise training/inference mismatch is not the same as a noisy output from a specific checkpoint." }, { "left": "issue:5606", "right": "issue:9673", "accept": false, - "reason": "Loading a LoRA with no alpha is a different bug from high CPU usage while loading multiple LoRAs." + "reason": "LoRA alpha handling and high CPU usage when loading multiple LoRAs are different bugs." }, { "left": "issue:7380", "right": "issue:8414", "accept": false, - "reason": "A one-step diffusion/distillation implementation request is unrelated to a new ConsistencyTTA model request." + "reason": "Both are diffusion-distillation research requests, but they describe different models and are not the same change." }, { "left": "issue:10925", "right": "issue:12459", "accept": false, - "reason": "Dreambooth FLUX finetuning and FLUX.1-Kontext batch throughput are different problems in the same ecosystem." + "reason": "Dreambooth finetuning for FLUX dev CLIPTextModel is distinct from batch inference throughput for FLUX.1-Kontext-dev." }, { "left": "issue:1475", "right": "issue:3765", "accept": false, - "reason": "Training with mixed image classes and ControlNet stacking effects are different training issues." + "reason": "Training with two image classes is unrelated to ControlNet stacking effects during training." }, { "left": "issue:3674", "right": "issue:6277", "accept": false, - "reason": "Metadata JSONL parsing and broken gradients in an example script are unrelated." + "reason": "Metadata.jsonl parsing during fine-tuning is not the same as broken gradients in the example LoRA script." }, { "left": "issue:8514", "right": "issue:9953", "accept": false, - "reason": "Parallel fine-tuning with EMAModel and moving a quantized pipeline to CUDA are different failure modes." + "reason": "EMAModel parallel fine-tuning errors and moving a quantized component to CUDA are different issues." }, { "left": "issue:2459", "right": "issue:6974", "accept": false, - "reason": "XFormers support in a LoRA training script and unload_textual_inversion not cleaning up tokenizer/text encoder state are unrelated." + "reason": "xformers attention not working in one training script is unrelated to unload_textual_inversion not clearing tokenizer_2/text_encoder_2." }, { "left": "issue:1543", "right": "issue:4790", "accept": false, - "reason": "A revision='fp16' parameter issue is not the same as a conversion round-trip compatibility bug." + "reason": "Cannot set revision=fp16 is a separate conversion/configuration bug from round-tripping converted weights between scripts." }, { "left": "issue:1695", "right": "issue:2606", "accept": false, - "reason": "Both involve training/DeepSpeed, but one is multi-GPU training support and the other is save_state hanging; not the same bug." + "reason": "Both involve DeepSpeed training, but one is multi-GPU setup and the other is a save_state hang; not the same concrete bug." }, { "left": "issue:3086", "right": "issue:556", "accept": false, - "reason": "Audio-to-image and seamless texture tiling are completely different features." + "reason": "Input-audio/output-image is unrelated to seamless texture tiling." }, { "left": "issue:4304", "right": "issue:5594", "accept": false, - "reason": "SDXL LoRA support and SDXL LoRA memory savings are related only at a broad training level, not the same change." + "reason": "Support for SDXL Text-to-Image LoRA is a feature request; saving memory during SDXL fine-tuning is an optimization issue." }, { "left": "issue:5002", "right": "issue:9370", "accept": false, - "reason": "Missing use_karras_sigmas on KDPM2 schedulers and missing pooled_projection in CombinedTimestepGuidanceTextProjEmbeddings are unrelated API bugs." + "reason": "Missing use_karras_sigmas in schedulers is unrelated to the missing pooled_projection argument in a different embedding module." }, { "left": "issue:4562", "right": "issue:6689", "accept": false, - "reason": "Both are conversion failures, but they concern different source/target models and likely different conversion paths." + "reason": "Both are conversion failures, but they involve different checkpoints and likely different conversion paths; not enough to treat as one bug." } ] }, @@ -38867,10 +38891,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -38885,166 +38909,166 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose title-similarity bucket of unrelated bugs. The only notably close pair is 2392/6201, which appears to be in the same prompt_embeds/no-prompt validation area, but the packet only gives titles/metadata, so that remains a likely related issue rather than a confirmed duplicate.", - "confidence": 0.62, - "canonical_issue_reason": "issue:6201 is the best representative of the prompt_embeds/no-prompt validation theme, but only at a title/metadata level; it should be treated as a likely related report rather than a proven subsumer of issue:2392.", + "summary": "Mostly heterogeneous issues with no PRs; the only strong duplicate-like pair is the prompt_embeds/prompt handling bug in Stable Diffusion pipelines.", + "confidence": 0.78, + "canonical_issue_reason": "issue:6201 is the better canonical issue for the prompt_embeds bug: it describes the core control-flow problem more generally, while issue:2392 is a narrower img2img manifestation of the same failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "issue:6201 is the strongest standalone issue in the cluster because its title most clearly captures the shared prompt_embeds/no-prompt bug pattern and is the most general representative among the candidates.", + "best_issue_reason": "issue:6201 best represents the one real duplicate pair in this set because it captures the underlying pipeline logic error rather than a single downstream symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:3461", "right": "issue:6967", "accept": false, - "reason": "Both are training-related, but one is a DreamBooth performance/memory complaint and the other is an SDXL prior-preservation shape mismatch; different bugs." + "reason": "Both are training problems, but they concern different scripts and failure modes (performance/vRAM vs prior-preservation shape mismatch)." }, { "left": "issue:6382", "right": "issue:9876", "accept": false, - "reason": "One is an IP-adapter key error after LCM LoRA training; the other is VRAM not being released after LoRA training. Different failure modes and code paths." + "reason": "One is an IP-adapter key error after LCM LoRA training; the other is a VRAM-release concern after LoRA training. Different bugs." }, { "left": "issue:3420", "right": "issue:6974", "accept": false, - "reason": "ControlNet inpaint pipeline integration vs unloading textual inversion components; unrelated implementation bugs." + "reason": "ControlNet/diffusion pipeline integration vs unloading textual inversion components; unrelated code paths." }, { "left": "issue:1251", "right": "issue:2107", "accept": false, - "reason": "Both mention black/glitchy images, but one is an MPS/RePaintPipeline issue and the other is low-resolution image handling. Not the same root cause." + "reason": "Both mention black images, but the triggers are different: RePaint on MPS vs low-resolution image handling." }, { "left": "issue:6435", "right": "issue:9933", "accept": false, - "reason": "Both involve width/height, but one is ControlNet default initialization and the other is missing img2img parameters in SD3. Different API bugs." + "reason": "Both mention width/height, but one is default initialization in ControlNet and the other is a missing API parameter in SD3 img2img." }, { "left": "issue:2392", "right": "issue:6201", - "accept": false, - "reason": "They look related on title alone because both involve prompt_embeds with missing prompt handling, but the packet does not confirm the same code path or exact duplicate bug." + "accept": true, + "reason": "Same underlying prompt handling bug: the pipeline wrongly requires/uses prompt when prompt_embeds (and negative_prompt_embeds) are provided." }, { "left": "issue:2674", "right": "issue:5857", "accept": false, - "reason": "Both are LoRA-related, but one is a safetensors dependency/requirements issue and the other is an AnimateDiff SDXL quality problem. Not duplicates." + "reason": "LoRA/safetensors packaging issue vs LoRA behavior on AnimateDiff SDXL; different problems." }, { "left": "issue:4236", "right": "issue:6686", "accept": false, - "reason": "SDXL Refiner text-encoder registration vs Dynamo compilation after save/load; different behaviors and fixes." + "reason": "SDXL refiner text-encoder registration vs Dynamo compilation save/load performance; not the same bug." }, { "left": "issue:10866", "right": "issue:12147", "accept": false, - "reason": "Both are LoRA loading failures, but they concern different models (Lumina Image 2.0 vs Wan2.2 Lightning) and likely different loaders." + "reason": "Both are LoRA loading issues, but for different models and failure contexts; too broad to treat as one bug." }, { "left": "issue:1310", "right": "issue:1392", "accept": false, - "reason": "K-Diffusion + Stable Diffusion pipeline request vs a generic Stable Diffusion 2 issue; too broad and not the same bug." + "reason": "A K-Diffusion pipeline question is not the same as a Stable Diffusion 2 feature request." }, { "left": "issue:4892", "right": "issue:6726", "accept": false, - "reason": "TokenFlow video editing and CCSR super-resolution are unrelated pipeline/model changes." + "reason": "Different pipelines and tasks: TokenFlow video editing vs content-consistent super-resolution." }, { "left": "issue:8560", "right": "issue:8950", "accept": false, - "reason": "One is SD3 training OOM; the other is an SD3 LoRA training bug. Memory exhaustion is not the same as the reported functional error." + "reason": "Both involve SD3 training, but one is OOM and the other is a LoRA training bug; not the same issue." }, { "left": "issue:222", "right": "issue:3432", "accept": false, - "reason": "Both mention runtime/training failures on CPU half precision, but the reports are too generic and point to different contexts." + "reason": "Generic training RuntimeError vs a specific CPU Half-precision convolution limitation." }, { "left": "issue:10514", "right": "issue:8788", "accept": false, - "reason": "Both involve resolution/size constraints, but Sana resolution binning and SD3 VAE tiling are different model-specific limitations." + "reason": "Different model/tooling constraints: Sana resolution binning sample_size limit vs SD3 VAE tiling on non-power-of-two images." }, { "left": "issue:11370", "right": "issue:7982", "accept": false, - "reason": "Video conversion pickle-load failure vs MultiControlNet from_pipe failure; unrelated bugs." + "reason": "A converter file-load error is unrelated to a from_pipe MultiControlNetModel issue." }, { "left": "issue:2728", "right": "issue:4686", "accept": false, - "reason": "Both are loading-from-pretrained issues, but one is UNet safetensors loading and the other is a locally saved upscale pipeline. Different artifacts and failure points." + "reason": "Safetensors loading for UNet vs loading a saved upscale pipeline; same family of API but different concrete bugs." }, { "left": "issue:7347", "right": "issue:7960", "accept": false, - "reason": "Generic LoRA training failure after update vs SDXL multi-GPU training failure; not enough evidence of the same bug." + "reason": "Both are training failures, but one is a post-update LoRA regression and the other is SDXL multi-GPU training failure." }, { "left": "issue:5509", "right": "issue:5607", "accept": false, - "reason": "Both are tensor/image shape mismatch errors, but the exact mismatched dimensions and triggering paths differ." + "reason": "Both are shape mismatches, but the actual mismatched tensors and inputs are different; not enough to merge as one bug." }, { "left": "issue:3803", "right": "issue:8472", "accept": false, - "reason": "Multi-GPU LoRA training bug vs a general question about pre-training LoRA with other modules; not the same issue." + "reason": "Multi-GPU LoRA training bug vs a general question about simultaneous modules during pre-training." }, { "left": "issue:9527", "right": "issue:9979", "accept": false, - "reason": "One is a dtype mismatch between controlnet and main pipeline; the other is a Flux img2img controlnet channel error. Different code paths." + "reason": "A dtype mismatch with ControlNet fp32/bf16 is not the same as a Flux img2img ControlNet channel error." }, { "left": "issue:3296", "right": "issue:5516", "accept": false, - "reason": "Checkpoint save/load with text_encoder vs LoRA not functioning with t2i adapters; both are LoRA-adjacent but distinct problems." + "reason": "Dreambooth checkpoint save/load with text_encoder vs LoRA behavior with t2i-adapter pipelines; different subsystems." }, { "left": "issue:4932", "right": "issue:6974", "accept": false, - "reason": "Dreambooth not returning pytorch_model.bin vs textual inversion unload not cleaning up tokenizer/text_encoder_2; unrelated save/unload bugs." + "reason": "Dreambooth artifact export issue vs unloading textual inversion components; unrelated." }, { "left": "issue:8955", "right": "issue:9914", "accept": false, - "reason": "Both concern loading errors, but one is a CLIPTextModel state_dict mismatch and the other is an Xlabs LoRA loading proposal. Not the same bug." + "reason": "Both are load-time errors, but they involve different models and failure points (CLIPTextModelWithProjection state_dict vs Flux LoRA loading)." }, { "left": "issue:6839", "right": "issue:9208", "accept": false, - "reason": "from_single_file failures for StableVideoDiffusion vs ControlNet models; same general theme, but different pipelines and loaders." + "reason": "Both mention from_single_file, but one is StableVideoDiffusionPipeline and the other is ControlNet model loading; too broad to combine." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The cluster summary is a bit too confident for the evidence in the packet. In particular, the packet only gives titles/metadata, so claims like \u201cexplicit repro,\u201d \u201cclear shared code-path,\u201d and that issue:6201 \u201csubsumes\u201d issue:2392 are not well grounded. The 2392/6201 pair is plausible on title similarity, but it should be framed more conservatively as a likely related prompt_embeds/no-prompt validation issue rather than a confirmed duplicate. The rest of the soft-edge rejections look appropriately conservative." + "accept": true, + "feedback": "Grounded overall. The summary stays conservative, and the only accepted soft pair (issue:2392 / issue:6201) is well supported by the titles as the same prompt/prompt_embeds handling bug, with 6201 as the broader canonical issue. The rejected pairs are also appropriately cautious and do not overclaim duplication from the packet evidence." }, "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": false, + "retried": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -39139,10 +39163,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -39157,156 +39181,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Heterogeneous cluster of mostly unrelated issues spanning training scripts, LoRA loading, inpainting, schedulers, and feature requests; no clear duplicate core or canonical representative emerges.", - "confidence": 0.98, - "canonical_issue_reason": "No canonical issue: the items are a heterogeneous mix of unrelated bugs/features across different pipelines and training scripts.", + "summary": "The set is heterogeneous and the soft pairs are only superficially similar; none look like true duplicates or mergeable equivalents.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a strong representative because the cluster is broad and incoherent rather than centered on one underlying defect.", + "best_issue_reason": "Issue 8450 is the most concrete and detailed bug report in the only somewhat related inpainting-strength area, but the overall cluster is too mixed to treat it as a true canonical.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1985", "right": "issue:2783", "accept": false, - "reason": "Different feature areas: loading new embeddings in Stable Diffusion vs requesting a controlnet pipeline." + "reason": "Different feature requests: embedding loading vs adding an img2img/inpainting controlnet pipeline." }, { "left": "issue:6131", "right": "issue:8469", "accept": false, - "reason": "Both involve IP-Adapter-related topics, but one is about IP Adapter Plus support and the other is about corrupted faces; not the same bug." + "reason": "Both involve IP-Adapter, but they describe different models and failure modes, not the same bug." }, { "left": "issue:10107", "right": "issue:8543", "accept": false, - "reason": "Both mention gradient checkpointing, but one is about validation behavior in examples and the other is a SD3 dreambooth failure; different code paths." + "reason": "Both mention gradient checkpointing, but one is about validation behavior in training examples and the other is an SD3 dreambooth example failure." }, { "left": "issue:3619", "right": "issue:6911", "accept": false, - "reason": "Unrelated problems: Text2Video-Zero cross-frame attention vs UNetConditional3DModel class-label handling." + "reason": "Video-related, but cross-frame attention failure and class-label handling in a 3D UNet are different code paths." }, { "left": "issue:4307", "right": "issue:4580", "accept": false, - "reason": "Different failures: local_files_only from_pretrained handling vs uniPC schedule with use_karras_sigmas." + "reason": "Unrelated issues: local_files_only loading error vs UniPC Karras schedule exception." }, { "left": "issue:7348", "right": "issue:8450", "accept": false, - "reason": "Both are inpainting-quality complaints, but one is ControlNet inpaint with a specific scheduler and the other is SDXL inpaint at strength=1.0; not clearly the same bug." + "reason": "Both are inpainting-quality bugs, but one is scheduler-specific ControlNet inpaint output and the other is SDXL inpaint noise at strength=1.0; too different to merge confidently." }, { "left": "issue:6809", "right": "issue:9045", "accept": false, - "reason": "Both involve LoRA/PEFT, but one is adapter loading under PEFT and the other is CUDA conversion with accelerate+peft; different concrete issues." + "reason": "Both mention PEFT/LoRA-style loading, but one is adapter loading and the other is CUDA conversion under accelerate+peft." }, { "left": "issue:11136", "right": "issue:11208", "accept": false, - "reason": "Different unsupported-feature reports in different pipelines: SanaSprint SCM timesteps vs SD3ControlNetInpaintingPipeline MultiControlNetModel." + "reason": "Both are unsupported-feature errors, but for different pipelines and different unsupported conditions." }, { "left": "issue:2392", "right": "issue:6734", "accept": false, - "reason": "Different NoneType/prompt-related errors in different pipeline contexts; not the same underlying bug." + "reason": "Different null-handling failures in different pipeline entry points." }, { "left": "issue:4061", "right": "issue:4303", "accept": false, - "reason": "Different SDXL problems: LoRA loading error vs FlaxStableDiffusionPipeline SDXL crash." + "reason": "SDXL LoRA loading vs Flax SDXL loading crash are not the same bug or code path." }, { "left": "issue:3487", "right": "issue:5647", "accept": false, - "reason": "Both involve inpaint strength/scheduler behavior, but the models and failure descriptions differ enough that they do not look like the same concrete bug." + "reason": "Both concern inpaint strength/quality, but one is tied to EulerAncestral scheduler behavior and the other to strength==1 in SDXL inpaint." }, { "left": "issue:12612", "right": "issue:13258", "accept": false, - "reason": "Different Wan model issues: NaNs in latents on 0.35.2 vs missing Wan2.2 image-to-video support." + "reason": "Different Wan model problems: latent NaNs vs a missing image-to-video model." }, { "left": "issue:1574", "right": "issue:2465", "accept": false, - "reason": "Training failure vs a request for a reproducible recipe; not duplicate reports." + "reason": "Dreambooth training failure vs a request for a reproducible recipe; not duplicates." }, { "left": "issue:1350", "right": "issue:2784", "accept": false, - "reason": "Unrelated feature requests: DiffusionDet object detection vs Dreambooth support for text-to-video models." + "reason": "Completely different features: object detection vs text-to-video dreambooth support." }, { "left": "issue:4354", "right": "issue:6615", "accept": false, - "reason": "Both touch from_single_file, but one is wrong pipeline type and the other is local YAML config loading; different defects." + "reason": "Both involve loading from single file/original ckpt, but the failures are different: pipeline type mismatch vs local YAML config handling." }, { "left": "issue:4386", "right": "issue:5594", "accept": false, - "reason": "LoRA conversion question vs SDXL fine-tuning memory usage topic; unrelated." + "reason": "LoRA conversion question vs memory-saving fine-tuning question; no shared bug." }, { "left": "issue:4936", "right": "issue:6392", "accept": false, - "reason": "Different LoRA problems: resume training broken vs load_lora_weights failing for custom LoRA." + "reason": "Resume-training regression and custom LoRA loading failure are separate issues." }, { "left": "issue:7197", "right": "issue:9210", "accept": false, - "reason": "Different training script errors: resume KeyError vs dataset column validation error." + "reason": "Different training errors: shortest-edge resume failure vs invalid image_column value." }, { "left": "issue:11489", "right": "issue:9393", "accept": false, - "reason": "Both are Flux Dreambooth LoRA training issues, but one is an error during training and the other is failure to save the model; not clearly the same code-path bug." + "reason": "Both are Flux LoRA training issues, but one is a training-time error and the other is a model-save problem under deepspeed." }, { "left": "issue:1367", "right": "issue:2904", "accept": false, - "reason": "Different subsystems: scheduler config serialization vs AudioLDM pipeline scheduler attribute error." + "reason": "Scheduler config serialization and AudioLDM scheduler NoneType error are unrelated." }, { "left": "issue:3264", "right": "issue:3537", "accept": false, - "reason": "Both are conversion scripts, but they target different model types and workflows." + "reason": "Different conversion directions and different model types." }, { "left": "issue:4453", "right": "issue:5980", "accept": false, - "reason": "Both mention SDXL and formats, but one is LoRA conversion for A1111 and the other is ControlNet model packaging; not the same issue." + "reason": "Both mention SDXL/LoRA/controlnet artifacts and formats, but one asks about conversion to safetensors and the other about output file type for ControlNet models." }, { "left": "issue:6176", "right": "issue:6392", "accept": false, - "reason": "Both are LoRA-loading complaints, but the reported contexts differ and there is no evidence they share the same failure mode." + "reason": "Both are LoRA loading problems, but the reports point to different scenarios and no shared concrete failure path." }, { "left": "issue:7357", "right": "issue:7420", "accept": false, - "reason": "One is about slow inference with torch.compile/fuse_loras, the other is inability to use torch.compile at all." + "reason": "Performance regression with torch.compile/fuse_loras is different from a basic inability to use torch.compile." } ] }, @@ -39409,10 +39433,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -39427,162 +39451,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous: a mix of unrelated feature requests, training-script failures, loading bugs, and dtype/device issues across different pipelines. Only one pair (4354/4402) looks like the same underlying from_single_file SDXL inpaint bug; the rest are too broad or different in concrete failure mode to merge.", - "confidence": 0.87, - "canonical_issue_reason": "No single issue cleanly represents the whole cluster because the items span unrelated models, scripts, and error classes. If a representative is needed, issue 4354 is the clearest concrete bug report and the only near-duplicate anchor.", + "summary": "This set is mostly a grab-bag of unrelated issues; nearly all soft pairs are only superficially similar and should stay separate. The only likely duplicate pair is the SDXL inpaint `from_single_file()` reports.", + "confidence": 0.78, + "canonical_issue_reason": "issue:4354 is the most concrete, specific bug report in the only likely duplicate pair, and it names the exact failing code path (`StableDiffusionXLInpaintPipeline.from_single_file`).", "canonical_pr_reason": null, - "best_issue_reason": "Issue 4354 is the strongest representative because it is a specific, reproducible API bug with a clear failure mode and directly overlaps the only likely duplicate (4402).", + "best_issue_reason": "issue:4354 is the best canonical issue candidate because it is specific, actionable, and closest to a clear duplicate target; the rest of the items are unrelated feature requests or different failure modes.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:13053", "right": "issue:13067", "accept": false, - "reason": "Both are feature requests for different model adapters (Unipic vs Anima), not the same change." + "reason": "Different feature requests for different models/pipelines: Unipic 3.0 vs Anima." }, { "left": "issue:4885", "right": "issue:8532", "accept": false, - "reason": "Both mention LoRA, but one is a missing load in SDXL ControlNet img2img and the other is a memory growth issue while loading weights." + "reason": "Both mention LoRA loading, but one is SDXL ControlNet img2img and the other is progressive memory growth during LoRA loading; different bugs." }, { "left": "issue:1249", "right": "issue:4073", "accept": false, - "reason": "Different concrete bugs in different components: ONNXPipeline scheduler device access vs DeepFloyd IF safety checker astype handling." + "reason": "Different code paths and failures: ONNX pipeline device attribute error vs DeepFloyd IF safety checker `astype` error." }, { "left": "issue:3454", "right": "issue:6278", "accept": false, - "reason": "Both are training failures, but they involve different scripts and different root causes; not the same bug." + "reason": "Training-script problems in different scenarios: latest-version breakage vs multi-GPU setup failure." }, { "left": "issue:10954", "right": "issue:9341", "accept": false, - "reason": "Both concern FLUX LoRA loading, but one is a generic load failure and the other is specifically about LoRAs containing text_model weights." + "reason": "Both are FLUX LoRA load complaints, but the reported symptoms are not specific enough to say they are the same underlying bug." }, { "left": "issue:4724", "right": "issue:6252", "accept": false, - "reason": "One is SDXL Img2Img behavior with a checkpoint; the other is LoRA weights not loading from a checkpoint. Related area, different bug." + "reason": "Different issues around SDXL img2img behavior vs checkpoint LoRA loading." }, { "left": "issue:5897", "right": "issue:9045", "accept": false, - "reason": "Both are device-related failures, but one is a training LoRA device mismatch and the other is a StableDiffusion3 accelerate/peft CUDA conversion problem." + "reason": "Both involve device mismatch themes, but one is LoRA training and the other is SD3 pipeline `.to(cuda)` with accelerate/peft; not the same bug." }, { "left": "issue:7420", "right": "issue:8589", "accept": false, - "reason": "torch.compile inability and SD3 prior-preservation training are unrelated issues." + "reason": "`torch.compile` failure and prior-preservation support in SD3 DreamBooth are unrelated." }, { "left": "issue:5960", "right": "issue:9927", "accept": false, - "reason": "Both are ControlNet training issues, but the errors and likely code paths differ (uint8 cast warning vs HeaderTooLarge)." + "reason": "Different failure modes: invalid cast warning in image conversion vs `HeaderTooLarge` during SDv3 controlnet training." }, { "left": "issue:10147", "right": "issue:7948", "accept": false, - "reason": "Different feature requests for different adapters/models (MV-Adapter vs MuLan)." + "reason": "Separate feature requests for different adapters/models: MV-Adapter multi-view generation vs MuLan language adaptation." }, { "left": "issue:11830", "right": "issue:8970", "accept": false, - "reason": "Both produce black output, but on different pipelines and platforms with different likely causes." + "reason": "Both produce black output, but on different pipelines and hardware paths (Framepack float16 vs Kolors MPS)." }, { "left": "issue:3975", "right": "issue:794", "accept": false, - "reason": "Both are error-message/UI wording issues, but they are separate messages and not the same fix." + "reason": "A typo in an error message is not the same as a general unsupported-classes error-message improvement." }, { "left": "issue:4354", "right": "issue:4402", "accept": true, - "reason": "Both report the same from_single_file SDXL Inpaint pipeline misbehavior; one says the returned type is wrong, the other says the function cannot be used correctly." + "reason": "Both concern `StableDiffusionXLInpaintPipeline.from_single_file()` misbehaving on the same pipeline; the titles indicate the same underlying construction bug." }, { "left": "issue:5023", "right": "issue:5875", "accept": false, - "reason": "Import error in WuerstchenCombinedPipeline and ONNX conversion prompt error are unrelated." + "reason": "Import error for WuerstchenCombinedPipeline vs ONNX conversion prompt error are different problems." }, { "left": "issue:5633", "right": "issue:7991", "accept": false, - "reason": "A transformer_temporal model bug and a basic training tutorial bug are different scopes and code paths." + "reason": "Different areas: a bug in `transformer_temporal.py` versus a training tutorial bug." }, { "left": "issue:10223", "right": "issue:6382", "accept": false, - "reason": "One asks where to obtain a LoRA id for inference; the other is an IP-adapter key error after SDXL LCM LoRA training. Not the same issue." + "reason": "Unrelated topics: where to get a LoRA ID for inference vs an IP-adapter key error after SDXL LCM LoRA training." }, { "left": "issue:1005", "right": "issue:746", "accept": false, - "reason": "Different subsystems and failures: inference timestep validation vs ONNX revision loading." + "reason": "Different scheduler/model loading problems: timestep count constraint vs ONNX revision loading." }, { "left": "issue:1802", "right": "issue:1931", "accept": false, - "reason": "One is about encoder_hidden_states=None on UNet2DConditionModel; the other is a float16 dtype mismatch in SD2." + "reason": "Related to UNet usage/dtype, but one is `encoder_hidden_states=None` support and the other is a float16/float mismatch; not the same bug." }, { "left": "issue:11006", "right": "issue:12052", "accept": false, - "reason": "Both involve Wan video pipelines, but one is broken output with quantized transformer and the other is a cross-device tensor multiplication failure with offloading." + "reason": "Both involve video/wan pipelines and quantized/offloading behavior, but the failures differ: bad video output vs cross-device tensor multiplication." }, { "left": "issue:10782", "right": "issue:8320", "accept": false, - "reason": "Completely different topics: Lumina Image minor issue vs from_single_file AutoencoderKL __name__ attribute error." + "reason": "Different issues in Lumina Image 2.0 vs `from_single_file` returning an `AutoencoderKL` object without `__name__`." }, { "left": "issue:1082", "right": "issue:761", "accept": false, - "reason": "Both are DreamBooth training failures, but the errors and causes differ (CUBLAS failure vs 8-bit Adam optimizer failure)." + "reason": "Both are DreamBooth training failures, but one is a CUDA CUBLAS execution failure and the other is an 8-bit Adam optimizer failure." }, { "left": "issue:10969", "right": "issue:9156", "accept": false, - "reason": "FLUX controlnet zero3 weight-shape error and Flux LoRA CUDA OOM are different failures." + "reason": "Different FLUX training issues: zero3 controlnet `weight` shape error vs CUDA OOM during LoRA training." }, { "left": "issue:2549", "right": "issue:4236", "accept": false, - "reason": "Different model families and bugs: SD 2.0 UNet behavior vs SDXL Refiner text encoder registration." + "reason": "Different SD2/SDXL model bugs: UNet2DConditionModel not working vs SDXL refiner text encoder registration." }, { "left": "issue:3390", "right": "issue:5719", "accept": false, - "reason": "Both touch local file loading, but one is about loading safetensors generally and the other is a specific LoRA filename precedence bug." + "reason": "One is about loading local safetensors files, the other about LoRA weight-file detection order; same general area but not the same bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The cluster is indeed heterogeneous overall, and the only clearly plausible duplicate-style pair is 4354/4402, which is reasonably identified as the same from_single_file SDXL inpaint issue. The other soft-edge rejections are appropriately cautious and do not overstate equivalence." + "feedback": "The assessment is well grounded in the packet and stays conservative on the soft edges. The only accepted pair, issue:4354 vs issue:4402, is plausibly the same StableDiffusionXLInpaintPipeline.from_single_file bug, and the rest of the verdicts do not overclaim duplicate status from the available titles/bodies. The summary and canonical choice are supported by the evidence provided." }, "evaluator_used": true, "retried": false, @@ -39682,10 +39706,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -39700,169 +39724,172 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly a grab bag of unrelated issues. I don\u2019t see any soft-edge pair that is clearly the same underlying bug; the closest item is the duplicated \"Error in loading Lora\" reports, but that pair was not among the proposed soft edges.", - "confidence": 0.78, - "canonical_issue_reason": "issue:11224 is the best representative because it is the active, higher-discussion report for the exact \"Error in loading Lora\" problem and appears to subsume the earlier duplicate report.", + "summary": "This cluster is mostly thematically related but not a true duplicate set. The only strong duplicate-looking pair is the Kolors black-image reports across ROCm and MPS; most other pairs share a subsystem or symptom but appear to be different bugs/requests.", + "confidence": 0.63, + "canonical_issue_reason": "issue:11209 is the best canonical issue because it matches issue:11224 exactly by title ('Error in loading Lora') and is the earlier report with inbound reference activity.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11224 is the strongest issue candidate overall in this set: it is concrete, open, and has the clearest duplicate signal among the listed items.", + "best_issue_reason": "issue:11209 is the strongest representative of the cluster\u2019s clearest duplicate topic (LoRA loading errors); it is open, earlier, and already referenced by another issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4398", "right": "issue:6666", "accept": false, - "reason": "Both mention black images, but one is SDXL LoRA dreambooth generation and the other validation-image output; too broad to treat as the same bug." + "reason": "Both mention black images, but one is SDXL LoRA-dreambooth generation and the other is validation images; different code paths and likely different bugs." }, { "left": "issue:4680", "right": "issue:692", "accept": false, - "reason": "These are unrelated fine-tuning topics; one is SDXL inpainting, the other is a general request for a fine-tuning example." + "reason": "One is an SDXL inpainting fine-tuning issue and the other is a general request for a fine-tuning example; not the same underlying problem." }, { "left": "issue:4912", "right": "issue:9241", "accept": false, - "reason": "Both involve conversion scripts, but they target different source formats and different workflows; not the same concrete change." + "reason": "Both are model-conversion related, but they target different source formats (CivitAI checkpoints vs XLab LoRA) and different scripts." }, { "left": "issue:5821", "right": "issue:7306", "accept": false, - "reason": "Both are training/runtime errors, but the titles point to different failure modes and different pipelines." + "reason": "Both are training/runtime errors, but they describe different failures (CUDA backward vs float16 in a specific pipeline)." }, { "left": "issue:692", "right": "issue:752", "accept": false, - "reason": "One is a generic example request and the other is a Dreambooth feature request; not a duplicate bug or change." + "reason": "A generic fine-tuning example request is not the same as a feature request for multiple concepts/classes in one Dreambooth model." }, { "left": "issue:7263", "right": "issue:9202", "accept": false, - "reason": "Both involve IP-Adapter-related incompatibilities, but they affect different pipelines and different interactions, so they are not clearly the same bug." + "reason": "Both involve IP-Adapter interactions, but one is attention slicing and the other is group offloading with a specific pipeline; different bug conditions." }, { "left": "issue:8557", "right": "issue:8708", "accept": false, - "reason": "Different SD3 training issues: token-limit behavior versus a missing VAE config field." + "reason": "Different SD3 training issues: token limit versus missing VAE shift_factor configuration." }, { "left": "issue:5529", "right": "issue:7146", "accept": false, - "reason": "Same script, but one is a bug report and the other is a CUDA OOM symptom; not the same underlying defect." + "reason": "Same script family, but one is a generic bug report and the other is a CUDA OOM during fine-tuning; not the same concrete defect." }, { "left": "issue:11033", "right": "issue:3296", "accept": false, - "reason": "Both concern loading/saving in training scripts, but they refer to different code paths and different model setups." + "reason": "Both concern loading/saving in Dreambooth-related code, but one is from_single_file UNet loading and the other is checkpoint save/load with text_encoder." }, { "left": "issue:4808", "right": "issue:5630", "accept": false, - "reason": "Both are img2img-related, but one is a SHAP-E integration test failure and the other is SDXL scheduler noise; different bugs." + "reason": "Both are img2img pipeline problems, but the failures differ (integration test on SHAP-E vs leftover noise with LMS scheduler on SDXL)." }, { "left": "issue:10492", "right": "issue:11209", "accept": false, - "reason": "TorchAO quantize/Lora fuse is a different subsystem from LoRA loading errors." + "reason": "LoRA is mentioned, but this is torchao quantization/fuse behavior versus a LoRA loading error; different issues." }, { "left": "issue:10492", "right": "issue:11224", "accept": false, - "reason": "TorchAO quantize/Lora fuse is a different subsystem from LoRA loading errors." + "reason": "Same as above: quantize/fuse issue is not the same as LoRA loading failure." }, { "left": "issue:1434", "right": "issue:2507", "accept": false, - "reason": "OOM during upscaling and safetensors load performance are unrelated." + "reason": "One is an OOM during upscaling, the other is slow safetensors loading on GPU; unrelated performance problems." }, { "left": "issue:11489", "right": "issue:5274", "accept": false, - "reason": "Both are training errors, but they involve different models, scripts, and likely different dtype failure causes." + "reason": "Both are training errors with mixed precision/dtype themes, but they involve different models, scripts, and failure modes." }, { "left": "issue:10715", "right": "issue:8970", - "accept": false, - "reason": "Both are Kolors black-image reports, but on different backends (ROCm vs MPS) with no evidence of the same code-path bug." + "accept": true, + "reason": "Both report Kolors pipelines producing black images on alternate backends (ROCm/MPS); this looks like the same underlying black-image bug affecting non-CUDA devices." }, { "left": "issue:10743", "right": "issue:9156", "accept": false, - "reason": "ZeRO-3 support request and Flux LoRA OOM are related to memory, but one is a feature request and the other a runtime failure." + "reason": "These are different Flux training concerns: adding ZeRO-3 support versus a CUDA OOM during LoRA training." }, { "left": "issue:1695", "right": "issue:2156", "accept": false, - "reason": "Multiple-GPU/deepspeed training support and CPU-vs-GPU training behavior are different problems." + "reason": "Multi-GPU Deepspeed training and CPU-vs-GPU usage are distinct training setup problems." }, { "left": "issue:11436", "right": "issue:8955", "accept": false, - "reason": "A matmul dtype mismatch in Hidream inference is unrelated to a CLIPTextModel state_dict loading error." + "reason": "A dtype mismatch in Hidream inference is not the same as a CLIPTextModel state_dict loading error." }, { "left": "issue:5575", "right": "issue:9602", "accept": false, - "reason": "Hidden-size configuration in LoRA training and adapter-name lookup failure are different issues." + "reason": "Hidden-size configuration in LoRA training is unrelated to adapter-name lookup failures." }, { "left": "issue:11791", "right": "issue:12673", "accept": false, - "reason": "Both mention group offloading, but one is about loading LoRA weights and the other about a QwenImagePipeline record_stream incompatibility; not the same bug." + "reason": "Both involve group offloading, but one is LoRA weights failing to load and the other is a pipeline compatibility issue with record_stream=True; not clearly the same bug." }, { "left": "issue:6906", "right": "issue:8543", "accept": false, - "reason": "Different Dreambooth example failures with different model families and failure modes." + "reason": "Different Dreambooth example failures: a broken LoRA example versus SD3 gradient checkpointing failure." }, { "left": "issue:7082", "right": "issue:7147", "accept": false, - "reason": "Separate import errors for different symbols; not the same underlying issue." + "reason": "These are unrelated import errors for different symbols from different modules." }, { "left": "issue:9142", "right": "issue:9297", "accept": false, - "reason": "One is a multi-ControlNet loading error, the other is LoRA load-time slowdown; same broad area but not the same bug." + "reason": "One is a controlnet loading AttributeError and the other is linear LoRA load-time slowdown; different issues." }, { "left": "issue:3296", "right": "issue:7619", "accept": false, - "reason": "Both are train_dreambooth.py issues, but one is checkpoint/text_encoder save-load behavior and the other is a channel-mismatch error." + "reason": "Both are Dreambooth.py issues, but one concerns checkpoint save/load with text_encoder and the other a channels mismatch at load time." }, { "left": "issue:2036", "right": "issue:6107", "accept": false, - "reason": "An OSError loading stable-diffusion-v1-4 and an IP-Adapter AWS deployment issue are unrelated." + "reason": "An OSError on a model load is not the same as an IP-Adapter deployment/runtime issue on AWS EC2." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The canonical issue choice for 11209 vs 11224 is supported by the identical titles, earlier creation time, and inbound reference on 11209. The only accepted soft edge (10715/8970) is reasonably justified by the shared black-image symptom on non-CUDA backends. I don\u2019t see unsupported overstatement in the rejected pairs." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -39958,10 +39985,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -39976,156 +40003,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are a mixed set of unrelated bug reports and feature requests. None of the soft pairs look like the same underlying issue; most only share a broad subsystem or model family.", - "confidence": 0.69, - "canonical_issue_reason": "Issue 6054 is the clearest concrete bug report in the set because it names a specific method (`EulerDiscreteScheduler.set_timesteps`) and a specific misuse, making it the best representative issue if one must be chosen.", + "summary": "This is a loose similarity cluster, not a duplicate set: the items span unrelated scheduler bugs, training failures, model-loading issues, and feature requests. I reject every soft edge; none looks like the same underlying bug/change.", + "confidence": 0.94, + "canonical_issue_reason": "No strong canonical issue: the items are heterogeneous and do not share one underlying failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 6054 is the strongest standalone bug candidate: it is specific, actionable, and tied to a precise code path. The rest of the set is mostly broader, different-symptom, or feature-request material.", + "best_issue_reason": "Issue 6054 is the cleanest concrete bug report in the set, but only as a loose anchor; it does not represent the rest of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1672", "right": "issue:4185", "accept": false, - "reason": "Different bugs: one is wrong image size after ckpt conversion; the other is black outputs in SDXL ControlNet inference." + "reason": "Different problems: ckpt-to-diffusers image-size regression vs SDXL ControlNet black images." }, { "left": "issue:10210", "right": "issue:10227", "accept": false, - "reason": "Different failure modes: FluxImg2Img quality issue vs a LoRA feature-support NotImplementedError." + "reason": "Flux img2img near-identity output is unrelated to the LoRA feature limitation error." }, { "left": "issue:1214", "right": "issue:4840", "accept": false, - "reason": "Different model classes and code paths: CLIPTextModel half-precision issue vs FlaxAutoencoderKL checkpoint loading." + "reason": "Half-precision CLIPTextModel failure is unrelated to FlaxAutoencoderKL checkpoint loading." }, { "left": "issue:1363", "right": "issue:4307", "accept": false, - "reason": "Unrelated errors: ONNX long-prompt/LMS scheduler failure vs local_files_only from_pretrained loading error." + "reason": "ONNX scheduler error for long prompts is unrelated to local_files_only from_pretrained failure." }, { "left": "issue:10925", "right": "issue:9514", "accept": false, - "reason": "Both are FLUX-related, but one is DreamBooth/CLIPTextModel training and the other is LoRA weight loading after fp8 quantization." + "reason": "Both mention FLUX, but one is DreamBooth/CLIPTextModel training and the other is fp8 LoRA loading after quantization." }, { "left": "issue:3284", "right": "issue:4704", "accept": false, - "reason": "Different training scripts and symptoms: DreamBooth LoRA with accelerate vs custom diffusion training failure." + "reason": "DreamBooth LoRA accelerate failure and custom diffusion training failure are different training scripts with different failure modes." }, { "left": "issue:11907", "right": "issue:9202", "accept": false, - "reason": "Different integration problems: FLUX scheduler compatibility vs SDXL PAG pipeline with FaceID IP-Adapter." + "reason": "Scheduler incompatibility in Flux is unrelated to the SDXL PAG/IP-Adapter FaceID issue." }, { "left": "issue:1917", "right": "issue:5605", "accept": false, - "reason": "Not duplicates: one asks for more integration tests, the other requests per-image caption support." + "reason": "These are different feature requests: integration testing vs per-image captions." }, { "left": "issue:2674", "right": "issue:771", "accept": false, - "reason": "Different topics: safetensors requirement for LoRA vs BF16 support in DreamBooth." + "reason": "LoRA safetensors packaging issue is unrelated to BF16 DreamBooth behavior." }, { "left": "issue:7243", "right": "issue:7948", "accept": false, - "reason": "Different requested features/models: ResAdapter vs MuLan language adapter support." + "reason": "Different model proposals: ResAdapter vs MuLan language adapter." }, { "left": "issue:2577", "right": "issue:3442", "accept": false, - "reason": "Different bugs: DDIMInverseScheduler not working vs Textual Inversion embedding dimension mismatch." + "reason": "DDIM inverse scheduler problem is unrelated to textual inversion embedding dimension mismatch." }, { "left": "issue:3871", "right": "issue:4728", "accept": false, - "reason": "Both mention training resume, but one is a resume error and the other is an OOM on resume; not the same underlying bug." + "reason": "Both mention resume training, but one is an error on resume and the other is resume OOM; not the same bug." }, { "left": "issue:11023", "right": "issue:7616", "accept": false, - "reason": "Different symptoms and paths: blurred results in a custom pipeline vs low-denoise artifacts in img2img." + "reason": "Both are image-quality complaints, but one is blur in a custom pipeline and the other is low-denoise artifacts." }, { "left": "issue:11127", "right": "issue:11258", "accept": false, - "reason": "Same broad area of single-file model loading, but different model families and different concrete problems." + "reason": "Different single-file model support problems for different model families (Flux vs LTX)." }, { "left": "issue:6323", "right": "issue:9856", "accept": false, - "reason": "Completely unrelated: pipeline attribute assignment error vs distributed port collision." + "reason": "TensorRT guidance_scale attribute error is unrelated to distributed port-collision startup failure." }, { "left": "issue:10928", "right": "issue:3901", "accept": false, - "reason": "Different operational errors: missing fp16 model file vs missing requirements.txt." + "reason": "Missing model weight file and missing requirements.txt are unrelated filesystem/package issues." }, { "left": "issue:6911", "right": "issue:8738", "accept": false, - "reason": "Different model/configuration bugs: class labels ignored in UNetConditional3DModel vs SD3Transformer2DModel configuration failure." + "reason": "Different model-class bugs: class labels in UNetConditional3DModel vs SD3Transformer2DModel configuration." }, { "left": "issue:8309", "right": "issue:8414", "accept": false, - "reason": "Different feature/model proposals: conditional diffusion distillation vs ConsistencyTTA text-to-audio model." + "reason": "Both are distillation-related ideas, but they are distinct model/change requests." }, { "left": "issue:10054", "right": "issue:6054", "accept": false, - "reason": "Both involve EulerDiscreteScheduler, but the concrete bugs differ: wrong `torch.linspace` parameters in tests vs `torch.from_numpy` misuse in `set_timesteps`." + "reason": "Both involve EulerDiscreteScheduler, but one is torch.linspace argument handling in tests and the other is torch.from_numpy misuse in set_timesteps; different bugs." }, { "left": "issue:4756", "right": "issue:5014", "accept": false, - "reason": "Not the same ask: one is a training-style support request, the other is a complaint about poor DreamBooth LoRA results." + "reason": "Training-style support request is unrelated to poor DreamBooth LoRA results." }, { "left": "issue:5254", "right": "issue:7180", "accept": false, - "reason": "Different scope: multi-ControlNet support request vs SDXL ControlNet img2img CPU offload failure." + "reason": "Multi-ControlNet support request is unrelated to a CPU-offload failure in SDXL img2img ControlNet." }, { "left": "issue:11422", "right": "issue:9841", "accept": false, - "reason": "Different training bugs: validation prompt crash in HiDream LoRA vs `use_dora=False` TypeError in SDXL DreamBooth LoRA." + "reason": "Different DreamBooth/LoRA training bugs with different triggers and error types." }, { "left": "issue:2107", "right": "issue:4255", "accept": false, - "reason": "One is a specific low-resolution image artifact bug; the other is a broad comparison to sd-webui quality." + "reason": "Low-resolution glitches are not the same as generally worse results than sd-webui." }, { "left": "issue:3487", "right": "issue:6749", "accept": false, - "reason": "Different issues: poor inpaint results with a scheduler vs inability to reproduce InstantID community example." + "reason": "Inpaint quality with EulerAncestralDiscreteScheduler is unrelated to InstantID community-example replication." } ] }, @@ -40229,10 +40256,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -40247,162 +40274,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose grab-bag of unrelated training/inference bugs across ControlNet, DreamBooth, LoRA, Flux, Qwen, and compile/load paths. The soft pairs share only broad wording and do not look like the same underlying bug or mergeable PR change.", - "confidence": 0.94, - "canonical_issue_reason": "No single canonical issue is safe here: the items span different models, pipelines, and error signatures, so they are not true duplicates of one concrete bug.", + "summary": "This cluster is heterogeneous: the soft-similarity pairs mostly share broad themes like training, ControlNet, LoRA, or torch.compile, but they describe different bugs, models, and code paths. I would not merge any of the proposed pairs as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "If forced to pick a representative, issue:3405 is the broadest ControlNet-training report, but it still only covers one small subset of the cluster and is not a valid umbrella duplicate.", + "best_issue_reason": "issue:4206 is the closest representative of the shared theme because it is a concrete ControlNet training regression with a clear model/context, unlike the more generic or unrelated errors elsewhere in the set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4349", "right": "issue:5515", "accept": false, - "reason": "Both are pipeline failures, but one is about missing text_encoder handling with prompt_embeds and the other is about initializing LCM without a scheduler; different code paths." + "reason": "Different failures: one needs a text_encoder despite prompt_embeds, the other is an LCM init/scheduler loading issue." }, { "left": "issue:3218", "right": "issue:5617", "accept": false, - "reason": "One is torch.compile graph-break behavior in attention code; the other is save/load failure for a compiled ControlNet img2img pipeline. Same subsystem, different bug." + "reason": "Both involve torch.compile, but one is graph-break behavior and the other is save/load after compilation; different code paths." }, { "left": "issue:8879", "right": "issue:9091", "accept": false, - "reason": "Both mention torch.compile, but one is an SDXL cudagraph failure and the other is VAE decoder compile support for SD 1.5; not the same fix." + "reason": "Different problems: cudagraph/torch.compile failure vs VAE decoder compile support." }, { "left": "issue:10066", "right": "issue:6368", "accept": false, - "reason": "A self.attn unpacking ValueError is unrelated to a checkpoint/lora alpha support error; different failure modes and code paths." + "reason": "Unpack error in self.attn is unrelated to the checkpoint/LoRA alpha unsupported error." }, { "left": "issue:10282", "right": "issue:12157", "accept": false, - "reason": "Shape mismatch for x_embedder weights is unrelated to the 'torch.device' not iterable quantization-config error." + "reason": "Shape mismatch loading x_embedder weights is unrelated to the torch.device iteration error in Qwen training." }, { "left": "issue:1883", "right": "issue:540", "accept": false, - "reason": "Both concern memory-related training behavior, but one is VAE gradient checkpointing and the other is keeping only UNet on GPU; separate feature requests." + "reason": "One is VAE gradient checkpointing support; the other is memory placement for UNet." }, { "left": "issue:3379", "right": "issue:7690", "accept": false, - "reason": "Checkpoint loading in DreamBooth and tokenizer loading for SDXL are different components and unrelated bugs." + "reason": "Checkpoint loading failure and tokenizer loading failure are distinct issues." }, { "left": "issue:5433", "right": "issue:9556", "accept": false, - "reason": "SDXL scheduler artifacts without refiner is not the same issue as saved LoRA models misbehaving with rslora." + "reason": "Scheduler artifacts in SDXL are unrelated to saved LoRA model issues with rslora." }, { "left": "issue:9667", "right": "issue:9717", "accept": false, - "reason": "One is a flux fp8 model loading problem with a local transformer file; the other is a Flux-dev Docker image error. Different environments and root causes." + "reason": "Flux fp8 model loading and Flux docker image errors are not the same bug." }, { "left": "issue:1617", "right": "issue:3175", "accept": false, - "reason": "Missing CLI arguments for train_dreambooth.py is unrelated to a multi-face DreamBooth training problem." + "reason": "Missing CLI arguments in dreambooth.py is unrelated to a multi-face training complaint." }, { "left": "issue:4726", "right": "issue:7593", "accept": false, - "reason": "A type annotation issue for callback does not match a NoneType attribute error in IP-Adapter image embedding prep." + "reason": "Callback typing is a different API issue from missing IP-adapter image projection layers." }, { "left": "issue:4885", "right": "issue:7793", "accept": false, - "reason": "Both involve LoRA, but one is SDXL ControlNet img2img failing to load LoRA weights and the other is DeepFloyd LoRA unloading error; different pipelines." + "reason": "LoRA loading in SDXL ControlNet img2img is different from unloading LoRA weights in DeepFloyd." }, { "left": "issue:7197", "right": "issue:9558", "accept": false, - "reason": "Resume-training key error and black validation images in TensorBoard are separate training/visualization bugs." + "reason": "Resume-from-checkpoint KeyError is unrelated to black validation images in TensorBoard." }, { "left": "issue:10178", "right": "issue:10722", "accept": false, - "reason": "A UNet 3D block dimension mismatch is not the same as a DreamBooth prior-preservation tensor size mismatch." + "reason": "A UNet 3D block size mismatch is different from a DreamBooth prior-preservation tensor mismatch." }, { "left": "issue:5972", "right": "issue:9630", "accept": false, - "reason": "One is a documentation issue about Min-SNR weighting; the other is a runtime error in flux-controlnet with num_single_layers=0." + "reason": "Documentation for min-SNR weighting is unrelated to the flux-controlnet num_single_layers runtime error." }, { "left": "issue:10055", "right": "issue:8589", "accept": false, - "reason": "Both are SD3 training topics, but one is a ControlNet training script failure and the other is DreamBooth prior-preservation support; different bugs." + "reason": "Both mention SD3 training, but one is ControlNet training script failure and the other is DreamBooth prior-preservation behavior." }, { "left": "issue:12458", "right": "issue:13140", "accept": false, - "reason": "Qwen-image-edit batch inference request/issue and Qwen Edit pipeline NoneType error are different features and failure modes." + "reason": "Batch inference feature request is unrelated to the NoneType iteration failure in Qwen Edit pipelines." }, { "left": "issue:5925", "right": "issue:6354", "accept": false, - "reason": "ZipLora multi-LoRA inference and ControlNet LoRA support are separate feature requests, not one underlying change." + "reason": "ZipLora multi-LoRA inference and ControlNet LoRA support are different feature requests." }, { "left": "issue:3379", "right": "issue:7133", "accept": false, - "reason": "Both involve checkpoint resumption, but one is a generic intermediate checkpoint load failure and the other is an SDXL text-encoder size mismatch." + "reason": "Checkpoint loading in DreamBooth and SDXL text-encoder size mismatch on resume are different resume/load bugs." }, { "left": "issue:10392", "right": "issue:9178", "accept": false, - "reason": "Flux-dev fp8 + Hyper-FLUX LoRA is not the same bug as Flux LoRA training target-module errors; one is an inference/config combo, the other is training setup." + "reason": "Both concern FLUX LoRA training, but one is an fp8/Hyper-FLUX setup and the other is a peft target-module error." }, { "left": "issue:1617", "right": "issue:5846", "accept": false, - "reason": "A missing-arguments CLI error is unrelated to a resume_from_checkpoint error in a different training script." + "reason": "Missing required args in train_dreambooth.py is unrelated to the resume_from_checkpoint error in an SDXL LoRA script." }, { "left": "issue:2695", "right": "issue:3101", "accept": false, - "reason": "ControlNet validation/TensorBoard failure and a ControlNet example training error are too different to be the same bug." + "reason": "TensorBoard validation failure and a generic ControlNet training example error do not appear to be the same bug." }, { "left": "issue:3405", "right": "issue:4206", "accept": false, - "reason": "Both are ControlNet training issues, but one is generic and the other is SDXL/main-branch-specific; no evidence they share the same concrete code-path bug." + "reason": "Both are ControlNet training issues, but one is generic while the other is specifically SDXL/main-branch breakage; not enough evidence of the same root cause." }, { "left": "issue:13079", "right": "issue:7146", "accept": false, - "reason": "CUDA OOM in FLUX.2-klein looped inference is not the same as SDXL fine-tuning OOM." + "reason": "FLUX.2-klein looped-image CUDA OOM is not the same as SDXL fine-tuning OOM." }, { "left": "issue:2392", "right": "issue:6996", "accept": false, - "reason": "Both are NoneType-related pipeline errors, but they arise from different arguments and different internal checks; not the same underlying bug." + "reason": "Both involve NoneType-related errors, but they occur in different pipeline inputs/paths and are not the same underlying bug." } ] }, @@ -40505,10 +40532,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -40523,156 +40550,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are mostly unrelated diffusers bug reports spanning import regressions, training problems, scheduler/math bugs, and pipeline-specific errors. None of the soft pairs look like the same underlying bug or change, so I would not merge any of them as duplicates.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "No strong duplicate cluster here: the soft pairs mostly share only a broad subsystem (imports, training, ControlNet, compile, MPS) but describe different concrete failures. I would not merge any of the proposed soft edges.", + "confidence": 0.93, + "canonical_issue_reason": "No single issue is a safe canonical representative; the items span unrelated bugs and feature requests rather than one underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No issue is a clearly better global representative because the cluster is heterogeneous and the soft-paired items do not share the same failure mode.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:4241", "right": "issue:752", "accept": false, - "reason": "Different feature requests: unidiffusers LoRA support vs Dreambooth multi-concept training." + "reason": "Different requests: LoRA support for unidiffusers vs multi-concept Dreambooth training." }, { "left": "issue:455", "right": "issue:760", "accept": false, - "reason": "Both mention MPS, but one is an aten::index.Tensor NotImplementedError and the other is incorrect StableDiffusion output on MPS; different failure modes." + "reason": "Different MPS problems: one is a missing op implementation, the other is incorrect StableDiffusion output." }, { "left": "issue:5661", "right": "issue:9200", "accept": false, - "reason": "Both are import/type issues, but one is missing PreTrainedTokenizer and the other is a dtype error in ControlNet pipeline; not the same bug." + "reason": "Import-time NameError for PreTrainedTokenizer vs SDXL ControlNet dtype mismatch." }, { "left": "issue:4530", "right": "issue:6574", "accept": false, - "reason": "Image artifact reports, but one is SDXL refiner visual corruption and the other is SVD pipeline warning about expected tensor range." + "reason": "SDXL refiner image artifacts vs SVD pipeline warning about expected input range; different symptoms and code paths." }, { "left": "issue:4107", "right": "issue:8729", "accept": false, - "reason": "Both are training-quality issues, but one is SDXL LoRA noise/quality degradation and the other is unexpected validation image behavior in ControlNet webdataset training." + "reason": "Both involve training, but one is SDXL LoRA quality degradation while the other is a validation-image issue in ControlNet training." }, { "left": "issue:10227", "right": "issue:10653", "accept": false, - "reason": "Different errors in different code paths: LoRA feature-size constraint vs bitsandbytes 8-bit module cannot be moved to CPU with .to()." + "reason": "Unrelated bugs: LoRA feature-dimension support vs bitsandbytes 8-bit `.to()` to CPU unsupported." }, { "left": "issue:3379", "right": "issue:954", "accept": false, - "reason": "Both are loading-related, but one is Dreambooth intermediate checkpoint loading and the other is login/load failure; unrelated." + "reason": "Dreambooth checkpoint loading vs login/loading failure; no shared underlying bug." }, { "left": "issue:5811", "right": "issue:7136", "accept": false, - "reason": "Both concern callback_on_step_end, but one is a request to add it to all pipelines and the other is a specific pipeline request; not a duplicate bug." + "reason": "General callback API expansion vs a single pipeline asking for callback support; not the same change." }, { "left": "issue:3468", "right": "issue:4754", "accept": false, - "reason": "Both involve saved artifact structure, but one is missing model_index.json after Dreambooth and the other is missing model_index.json after textual inversion; different flows and symptoms." + "reason": "Both are save/load file issues, but one is Dreambooth export metadata and the other is textual inversion model loading." }, { "left": "issue:5558", "right": "issue:9142", "accept": false, - "reason": "Different multi-controlnet issues: artifacts with StableDiffusionControlNetInpaintPipeline vs Hunyuan Dit list/__module__ AttributeError." + "reason": "Different multiple-ControlNet failures on different models and different errors (artifacts vs attribute/type error)." }, { "left": "issue:1185", "right": "issue:1541", "accept": false, - "reason": "Both are Dreambooth-related, but one is reproducibility and the other is model type detection/CKPT loading; not the same defect." + "reason": "Different Dreambooth concerns: reproducibility vs model-type detection/loading checkpoints." }, { "left": "issue:4000", "right": "issue:7785", "accept": false, - "reason": "Both mention conversion/slow inference, but one is ckpt-to-diffusers conversion and the other is first-inference latency in diffusers 0.27.x; unrelated." + "reason": "Converting ckpt format is unrelated to first-inference slowness." }, { "left": "issue:11864", "right": "issue:13104", "accept": false, - "reason": "Both are compile-time name errors, but one is AutoencoderDC missing torch under torch.compile and the other is Logger not defined; different modules and code paths." + "reason": "Separate compile-time NameErrors in different modules: missing `torch` vs missing `logger`." }, { "left": "issue:3296", "right": "issue:4324", "accept": false, - "reason": "Dreambooth checkpoint save/load with text_encoder vs a NoneType iterable error in unet_2d_condition forward; not the same bug." + "reason": "Checkpoint save/load bug in Dreambooth vs a `NoneType` error in UNet forward." }, { "left": "issue:5441", "right": "issue:6139", "accept": false, - "reason": "Both mention attn_processors, but one is a specific training script lacking them and the other is a model object missing the attribute entirely." + "reason": "Related vocabulary (`attn_processors`), but one is a missing attribute in a Wuerstchen prior training script and the other is a model lacking the attribute entirely." }, { "left": "issue:11241", "right": "issue:11749", "accept": false, - "reason": "Both relate to floating-point problems, but one is a training-time Flux ControlNet crash and the other is scheduler timestep precision errors; distinct." + "reason": "Both mention floating point problems, but one is a training crash and the other is scheduler timestep precision." }, { "left": "issue:2668", "right": "issue:6218", "accept": false, - "reason": "Both involve ControlNet/Img2Img behavior, but one misroutes the ControlNet image as an Img2Img image and the other is batched seed handling; different issues." + "reason": "ControlNet image routing bug vs batched seed handling bug; different pipeline behavior." }, { "left": "issue:3765", "right": "issue:6661", "accept": false, - "reason": "One is ControlNet image stacking artifacts during training; the other is MPS OOM at a specific image size. Same broad area, different bug." + "reason": "Stacking artifacts during ControlNet training vs MPS OOM; different failure modes." }, { "left": "issue:10049", "right": "issue:8915", "accept": false, - "reason": "Both are import errors, but MochiPipeline import failure and missing AutoPipelineForText2Image are unrelated symbols/features." + "reason": "Both are import errors, but for different pipeline symbols and unrelated code paths." }, { "left": "issue:10937", "right": "issue:7357", "accept": false, - "reason": "Both mention torch.compile performance/errors, but one is a compile error on vae.encode and the other is slow inference despite compile and fuse_loras." + "reason": "One is a `torch.compile` runtime error on `vae.encode`, the other is a performance complaint despite compile/fuse_loras." }, { "left": "issue:1375", "right": "issue:5022", "accept": false, - "reason": "Different import regressions from different versions and symbols: general diffusers import breakage vs UNet2DModel import failing because PreTrainedTokenizer is undefined." + "reason": "Different import breakages with different root causes and symbols." }, { "left": "issue:175", "right": "issue:988", "accept": false, - "reason": "One questions manual seeding in example code; the other is about FSDP training support. Not the same underlying problem." + "reason": "Manual seed correctness in example code vs FSDP training support; unrelated topics." }, { "left": "issue:11435", "right": "issue:7082", "accept": false, - "reason": "Different missing-import failures: HiDreamImagePipeline from diffusers vs CpuOffload from accelerate.hooks." + "reason": "Different import failures from different packages: `HiDreamImagePipeline` vs `CpuOffload`." }, { "left": "issue:10540", "right": "issue:9822", "accept": false, - "reason": "One is Flux Dev from_single_file failing after Flux Fill support landed; the other is loading SDXL LoRAs into Flux. Related area, but different model-loading behaviors." + "reason": "Flux from_single_file loading regression vs loading SDXL LoRAs into Flux; not the same concrete bug." } ] }, @@ -40774,10 +40801,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -40792,166 +40819,163 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issue set. I only accept two soft duplicate pairs: ControlNet conversion requests/errors (2581 \u2194 2774) and the original SD conversion script bug (6713 \u2194 6899). The remaining pairs differ by symptom, model, or code path and should stay separate.", - "confidence": 0.79, - "canonical_issue_reason": "No single issue cleanly represents the whole mixed set; issue 2581 is the broadest canonical among the accepted duplicates because it captures the underlying ControlNet\u2192Diffusers conversion need.", + "summary": "This cluster is mostly an overmatch of unrelated issues sharing generic ML/diffusers vocabulary. A few pairs are thematically adjacent, but none are clearly the same underlying bug or change.", + "confidence": 0.88, + "canonical_issue_reason": "No single issue cleanly represents a true duplicate set here; the only somewhat related subset is around the Stable Diffusion conversion script, so #6899 is the broadest placeholder if one must be chosen.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 2581 is the best overall issue representative here because it is the broadest, most canonical statement of the shared ControlNet conversion topic.", + "best_issue_reason": "#6899 is the most general, code-path-focused report in the closest subtheme (conversion script), but the cluster overall is too heterogeneous to treat it as a strong canonical issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:2214", "right": "issue:776", "accept": false, - "reason": "Both mention broken behavior, but one is NSFW filtering and the other is batched img2img; different bug classes and code paths." + "reason": "Both are SD-related runtime problems, but one is an NSFW filter regression and the other is an image-to-image batch-size bug; different code paths." }, { "left": "issue:3346", "right": "issue:3454", "accept": false, - "reason": "One is a feature request for resume_from_checkpoint; the other is a generic training-script failure on latest. Not the same bug." + "reason": "Both mention dreambooth LoRA training, but one is about resume_from_checkpoint support and the other is a generic latest-version failure." }, { "left": "issue:11989", "right": "issue:9558", "accept": false, - "reason": "Both report black images, but one is inference output and the other is tensorboard validation display; too likely to be separate causes." + "reason": "Both mention black images, but one is black inference output and the other is black validation images in TensorBoard; likely different causes." }, { "left": "issue:11321", "right": "issue:9178", "accept": false, - "reason": "README documentation bug vs flux LoRA PEFT target-module error; unrelated artifacts." + "reason": "One is a README bug and the other is a Flux LoRA target-modules training error; not the same underlying issue." }, { "left": "issue:2844", "right": "issue:4030", "accept": false, - "reason": "Both are embedding-related runtime errors, but they involve different pipelines and different embedding flows." + "reason": "Both are embedding-related runtime errors, but they affect different pipelines and different failure modes." }, { "left": "issue:2203", "right": "issue:2480", "accept": false, - "reason": "Both touch output directories, but one is model download location and the other is checkpoint folder creation; not the same defect." + "reason": "Both involve dreambooth training directories, but one is model download location and the other is checkpoint folder creation." }, { "left": "issue:455", "right": "issue:993", "accept": false, - "reason": "MPS indexing NotImplementedError vs fp16 dtype mismatch; different hardware/runtime failures." + "reason": "Both are dtype/device stability issues, but one is an MPS index op failure and the other is fp16 type mismatch." }, { "left": "issue:1976", "right": "issue:6435", "accept": false, - "reason": "Incorrect tokenizer/config path and uninitialized ControlNet width/height are unrelated issues." + "reason": "These concern unrelated bugs: a wrong tokenizer file path versus ControlNet default width/height initialization." }, { "left": "issue:4615", "right": "issue:7082", "accept": false, - "reason": "Import errors in different packages/classes; no shared concrete failure mode." + "reason": "Both are import errors, but for unrelated modules and different dependency breakages." }, { "left": "issue:2581", "right": "issue:2774", - "accept": true, - "reason": "Both are about converting ControlNet checkpoints to Diffusers; one is the general support request, the other is a concrete conversion failure on the same workflow." + "accept": false, + "reason": "Both are about ControlNet conversion, but one is a feature request and the other is a specific safetensors conversion error." }, { "left": "issue:5956", "right": "issue:6362", "accept": false, - "reason": "Fine-tuning performance degradation vs dataset-mapping OOM; different symptoms and likely different causes." + "reason": "Both are SDXL fine-tuning complaints, but one is performance degradation and the other is dataset-mapping OOM." }, { "left": "issue:1774", "right": "issue:5540", "accept": false, - "reason": "Both are OOM reports, but one is A6000 training OOM and the other is inference OOM; too broad to merge." + "reason": "Both are OOM reports, but they are generic memory exhaustion symptoms without evidence of the same root cause." }, { "left": "issue:10382", "right": "issue:11574", "accept": false, - "reason": "Both mention VAE internals, but one is a return-type AttributeError and the other is a missing weight KeyError; separate bugs." + "reason": "These are unrelated VAE/weight-loading bugs: a return-type mismatch versus a missing weight key." }, { "left": "issue:2163", "right": "issue:4935", "accept": false, - "reason": "Training dtype mismatch vs a question about AttnProcsLayers usage; not the same underlying problem." + "reason": "Both mention training scripts, but one is dtype mismatch in text-to-image training and the other is a LoRA layer-API question." }, { "left": "issue:5875", "right": "issue:7483", "accept": false, - "reason": "Both mention conversion, but one is ONNX prompt handling and the other is LoRA-to-Diffusers with SDXL base models; different code paths." + "reason": "These are different conversion problems: ONNX prompt issues versus LoRA-to-Diffusers conversion on SDXL base." }, { "left": "issue:6713", "right": "issue:6899", - "accept": true, - "reason": "Same converter script and same failure family; 6899 is the broader report and 6713 is a specific HeaderTooLarge manifestation." + "accept": false, + "reason": "Same utility file, but the titles point to different reports; insufficient evidence they are the same concrete bug." }, { "left": "issue:3674", "right": "issue:4125", "accept": false, - "reason": "Metadata JSONL reading vs LoRA loss-curve expectations; unrelated training concerns." + "reason": "Data/metadata parsing in one issue and loss-curve expectations in the other are not the same bug." }, { "left": "issue:3717", "right": "issue:3803", "accept": false, - "reason": "Both concern multi-GPU training, but one is OOM and the other is model replication/sharing; not clearly the same bug." + "reason": "Both involve multi-GPU training, but one is OOM and the other is a model-sharing/distribution problem." }, { "left": "issue:4031", "right": "issue:6250", "accept": false, - "reason": "Safety checker local-file loading and SDXL VAE single-file loading are distinct loader failures." + "reason": "Both are loading-related, but one is safety-checker local loading and the other is SDXL VAE single-file loading." }, { "left": "issue:7785", "right": "issue:9063", "accept": false, - "reason": "Slow first inference and tokenizer instantiation failure are different issues." + "reason": "Both mention startup/inference friction, but one is slow first inference and the other is tokenizer instantiation failure." }, { "left": "issue:10049", "right": "issue:7082", "accept": false, - "reason": "Importing MochiPipeline is unrelated to importing CpuOffload from accelerate.hooks." + "reason": "Both are import failures, but they involve different packages and unrelated missing symbols." }, { "left": "issue:1802", "right": "issue:1925", "accept": false, - "reason": "UNet encoder_hidden_states handling and fp16 inpainting are different functional problems." + "reason": "One is a UNet encoder_hidden_states handling bug and the other is an fp16 inpainting issue; not the same failure." }, { "left": "issue:3461", "right": "issue:7431", "accept": false, - "reason": "Dreambooth performance complaint vs DoRA training issues; not the same defect." + "reason": "Both are training-performance concerns, but one is DreamBooth efficiency and the other is DoRA text-encoder training issues." }, { "left": "issue:4835", "right": "issue:5647", "accept": false, - "reason": "Both are SDXL pipeline bugs, but one is MultiControlNet img2img and the other is inpaint strength==1; different code paths." + "reason": "Both are pipeline bugs, but one is MultiControlNet handling and the other is SDXL inpaint strength==1 behavior." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The two accepted soft pairs are well-supported by the titles/bodies: 2581\u21942774 (ControlNet conversion) and 6713\u21946899 (the stable diffusion conversion script). The remaining rejections are appropriately cautious and do not overclaim duplicate status from weak thematic similarity." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -41047,10 +41071,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -41065,156 +41089,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated training, inference, conversion, loading, and hardware/compatibility issues across different models and scripts. The soft pairs share only broad subsystem vocabulary, not the same underlying bug or change, so none should be merged as duplicates.", - "confidence": 0.97, + "summary": "The cluster is mostly a grab bag of unrelated Diffusers issues: model-loading, ControlNet, LoRA/LoCon, SDXL training, IP-Adapter, and a few usage questions. None of the soft pairs look like the same underlying bug or change closely enough to merge.", + "confidence": 0.94, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No single issue is a clear canonical for this set; the items are too heterogeneous, though training/inference bugs dominate overall.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10863", "right": "issue:9154", "accept": false, - "reason": "Both involve model file loading, but one is safetensors loading and the other is VAE/PT loading via AutoEncoder; different code paths and failure modes." + "reason": "Both are loading-related, but one is a local safetensors load error and the other is a broken PT/VAE loading path; different artifact types and failure modes." }, { "left": "issue:11363", "right": "issue:12489", "accept": false, - "reason": "Both mention ControlNet/CUDA issues, but one is GPU OOM during an example run and the other is an sm_120 kernel compatibility error; not the same bug." + "reason": "CUDA OOM in a ControlNet example is unrelated to an sm_120 kernel compatibility error; different root causes and fixes." }, { "left": "issue:11383", "right": "issue:11396", "accept": false, - "reason": "Same HiDream theme, but one requests LoRA support and the other asks how to convert a LoRA for ComfyUI; feature request vs conversion question." + "reason": "One asks for HiDream LoRA support; the other asks how to convert an existing HiDream LoRA for ComfyUI. Feature request vs. conversion question." }, { "left": "issue:1290", "right": "issue:1915", "accept": false, - "reason": "Both reference text-to-image workflows, but DreamArtist support and unclip weight unification are separate model-support tasks." + "reason": "DreamArtist community proposal is unrelated to unifying/uploading UnCLIP weights." }, { "left": "issue:1388", "right": "issue:1798", "accept": false, - "reason": "Both are SD2-related training/conversion problems, but they affect different scripts and one is about conversion while the other is missing v_prediction support." + "reason": "Both mention SD2.0 support, but one is about a conversion script update and the other is about v_prediction support in a training script; not the same bug." }, { "left": "issue:1524", "right": "issue:2780", "accept": false, - "reason": "Different training failures: poor SD2.0 quality versus FP16 gradient unscale error in a LoRA script." + "reason": "Poor SD2.0 training quality and FP16 unscale errors are different issues with different symptoms and likely fixes." }, { "left": "issue:1985", "right": "issue:3193", "accept": false, - "reason": "Related to textual inversion embeddings, but one asks for easier loading in the pipeline and the other is specifically batch loading; not the same defect." + "reason": "Both concern embeddings, but one is loading new embeddings in a pipeline and the other is batch-loading textual inversion embeddings; enhancement scope differs." }, { "left": "issue:2784", "right": "issue:3087", "accept": false, - "reason": "Separate feature requests for different model families: text2video Dreambooth support vs LyCORIS (LoCon/LoHA) support." + "reason": "Dreambooth for text2video and LyCORIS support are separate feature requests in the same broad training area." }, { "left": "issue:3539", "right": "issue:6203", "accept": false, - "reason": "These are unrelated questions about process barriers and ControlNet image sizing; no shared bug." + "reason": "main_process_first is a distributed-training question; SDXL ControlNet image-size conditioning is a model-input behavior question." }, { "left": "issue:381", "right": "issue:434", "accept": false, - "reason": "Both mention notebook problems, but they are different notebook versions/sections and different failure contexts." + "reason": "Both reference notebook/training breakage, but they target different docs and versions; not clearly the same bug." }, { "left": "issue:3871", "right": "issue:5197", "accept": false, - "reason": "Checkpoint resume error and multi-GPU slowness are distinct training issues." + "reason": "Checkpoint resume errors and multi-GPU slowdown are distinct training problems." }, { "left": "issue:4241", "right": "issue:4680", "accept": false, - "reason": "Different feature requests: LoRA fine-tuning for unidiffusers versus SDXL inpainting fine-tuning." + "reason": "LoRA support for unidiffusers and SDXL inpainting fine-tuning are different feature requests." }, { "left": "issue:4304", "right": "issue:4386", "accept": false, - "reason": "Both concern SDXL LoRA, but one requests training support and the other asks about converting a safetensors LoRA; not the same change." + "reason": "SDXL LoRA training support and converting an SDXL LoRA safetensor are related, but not the same concrete bug/change." }, { "left": "issue:5504", "right": "issue:6333", "accept": false, - "reason": "PEFT/xformers incompatibility is a runtime bug, while the other is a request to fine-tune a pretrained LoRA; different scopes." + "reason": "PEFT LoRA+xformers incompatibility is a runtime issue; finetuning a pretrained LoRA by PEFT is a different capability request." }, { "left": "issue:5966", "right": "issue:8390", "accept": false, - "reason": "Both mention ControlNet, but one is for Stable Diffusion 2.1 and the other for Stable Cascade training; separate model families and code paths." + "reason": "Stable Diffusion 2.1 ControlNet and Stable Cascade ControlNet training are different models and workflows." }, { "left": "issue:6278", "right": "issue:7960", "accept": false, - "reason": "Both are multi-GPU training complaints, but they concern different scripts/models and different failure characteristics." + "reason": "Both are multi-GPU training failures, but they are for different scripts/models and do not appear to share one code-path bug." }, { "left": "issue:7114", "right": "issue:7311", "accept": false, - "reason": "Both are SDXL training issues, but one is CUDA OOM and the other is checkpoint saving failure; not the same bug." + "reason": "CUDA OOM during SDXL training and failure to save checkpoints are separate failure points." }, { "left": "issue:3175", "right": "issue:5016", "accept": false, - "reason": "Both relate to Dreambooth/LoRA training, but one is multi-face training and the other is text_encoder training support in SDXL LoRA." + "reason": "Multi-face DreamBooth training and SDXL text-encoder training support are different training requests." }, { "left": "issue:7049", "right": "issue:7238", "accept": false, - "reason": "Different IP-Adapter problems: a tuple-unpack error in the basic example versus masking support." + "reason": "IP-Adapter unpacking error and IP-Adapter masking are distinct issues; same subsystem but not same bug." }, { "left": "issue:5619", "right": "issue:5857", "accept": false, - "reason": "Both mention LCM, but one is an AutoEncoder img2img encode bug and the other is poor LCM LoRA behavior in AnimateDiff SDXL." + "reason": "Autoencoder image-encode failure in LCM img2img and weak LCM LoRA performance in AnimateDiff are unrelated." }, { "left": "issue:5577", "right": "issue:7871", "accept": false, - "reason": "Different ControlNet/LoRA loading issues: a from_single_file KeyError versus blockwise scales with a specific SDXL LoRA checkpoint." + "reason": "ControlNet single-file loading KeyError and blockwise-scale LoRA error are different code paths." }, { "left": "issue:10475", "right": "issue:12175", "accept": false, - "reason": "Both are image-quality complaints, but they involve different models (SD3 vs Qwen Image) and are not the same defect." + "reason": "Both complain about image quality, but they involve different models (SD3 vs Qwen Image) and likely different causes." }, { "left": "issue:2312", "right": "issue:2844", "accept": false, - "reason": "Completely different failures: EMA save hook error in text-to-image finetuning versus passing image embeddings to StableUnCLIPImg2ImgPipeline." + "reason": "EMA save-hook failure in text-to-image finetuning and StableUnCLIP image-embedding runtime error are unrelated." }, { "left": "issue:1565", "right": "issue:9856", "accept": false, - "reason": "One is GPU device selection/OOM in training, the other is distributed port collision; unrelated infrastructure issues." + "reason": "GPU selection/OOM and distributed port-in-use errors are different distributed/runtime configuration problems." } ] }, @@ -41317,10 +41341,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -41335,166 +41359,163 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous cluster. The only strong duplicate-looking pair is the SDXL textual inversion/loading pair (5327/8854); the `set_lora_device` vs VRAM-release pair does not have enough title-level evidence to be treated as the same bug, so the cluster should be rejected as a whole for canonicalization.", - "confidence": 0.87, - "canonical_issue_reason": null, + "summary": "The cluster is mostly a set of weakly related, non-duplicate issues. Several pairs share broad themes like LoRA or loading errors, but they describe different code paths, symptoms, or pipelines, so none should be merged as duplicates.", + "confidence": 0.92, + "canonical_issue_reason": "issue:12689 is the clearest concrete bug report in the set: LoRA adapter weights not loading from checkpoints. It is specific, reproducible, and broadly representative of the LoRA-loading failures here without being overly tied to one niche symptom.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:12689 is the best single issue to anchor this cluster because it names a direct loading failure and is easier to match against follow-up reports than the more vague or symptom-only issues.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:12012", "right": "issue:1407", "accept": false, - "reason": "Wan 2.2 first-block cache context error and an ONNX pipeline feature_extractor regression are different pipelines and failure modes." + "reason": "Both mention runtime errors in pipelines, but one is a Wan 2.2 context-state failure and the other is an ONNX SD 2.0 feature_extractor error; different subsystems and root causes." }, { "left": "issue:1524", "right": "issue:2819", "accept": false, - "reason": "Textual inversion training quality complaint vs a network/download failure fetching a community script." + "reason": "Textual inversion training quality vs a download/network failure for a community script are unrelated bugs." }, { "left": "issue:2250", "right": "issue:3468", "accept": false, - "reason": "Dreambooth inference prompt failure vs missing model_index.json after training; related workflow, not the same bug." + "reason": "Dreambooth inference failure and missing model_index.json are different failure modes and likely different fixes." }, { "left": "issue:10337", "right": "issue:10972", "accept": false, - "reason": "Quantized transformers breaking FLUX IPAdapter vs loading OneTrainer Flux LoRA weights; distinct code paths." + "reason": "FLUX IPAdapter quantization failure and OneTrainer Flux LoRA loading failure are both FLUX-related but involve different components and symptoms." }, { "left": "issue:2507", "right": "issue:3198", "accept": false, - "reason": "GPU safetensors load performance vs a failure loading textual inversion weights; different symptoms and causes." + "reason": "GPU safetensors loading speed vs textual inversion loading failure are not the same bug." }, { "left": "issue:1181", "right": "issue:4366", "accept": false, - "reason": "Training from scratch vs fine-tuning a full U-Net; not the same issue." + "reason": "Training from scratch vs an SDXL full U-Net fine-tuning script request are different topics, not duplicates." }, { "left": "issue:5226", "right": "issue:7978", "accept": false, - "reason": "SDXL adapter LoRA-loading issue vs missing torchvision::nms operator; unrelated failures." + "reason": "SDXL adapter LoRA loading incompatibility is unrelated to a missing torchvision nms operator." }, { "left": "issue:7357", "right": "issue:7985", "accept": false, - "reason": "Slow inference with torch.compile/fuse_loras vs IPAdapter incompatibility with torch.compile; same area, different bug." + "reason": "Slow inference with compile/fused LoRAs is a performance concern; IPAdapter not compatible with torch.compile is a different code-path incompatibility." }, { "left": "issue:2439", "right": "issue:3061", "accept": false, - "reason": "Inpainting pipeline load/run problem vs a reproducibility complaint; too broad to be duplicates." + "reason": "Inpainting pipeline loading/running issue vs general reproducibility issue do not describe the same bug." }, { "left": "issue:2893", "right": "issue:6435", "accept": false, - "reason": "ControlNet width/height mix-up vs defaults not initialized; same subsystem, different defect." + "reason": "ControlNet width/height mix-up is a functional bug; default width/height initialization is a separate setup issue." }, { "left": "issue:4782", "right": "issue:8450", "accept": false, - "reason": "Mask-preservation bug vs SDXL inpaint noise at strength=1.0; both inpainting artifacts, but not the same concrete issue." + "reason": "Mask handling in inpainting and SDXL inpaint strength=1.0 noise are different inpainting bugs." }, { "left": "issue:5023", "right": "issue:5892", "accept": false, - "reason": "Specific WuerstchenCombinedPipeline import regression vs generic import error report; insufficient evidence they match." + "reason": "WuerstchenCombinedPipeline import regression and a generic import error are too vague to be the same concrete issue." }, { "left": "issue:5737", "right": "issue:6713", "accept": false, - "reason": "SDXL conversion failure vs HeaderTooLarge during conversion; both conversion-related, but different errors." + "reason": "SDXL conversion failure and HeaderTooLarge during conversion are both conversion-related but not the same underlying error." }, { "left": "issue:9876", "right": "issue:9913", "accept": false, - "reason": "VRAM not being released after LoRA training and `set_lora_device` not working are different title-level bugs; no grounded evidence they share the same function path." + "reason": "VRAM not released after LoRA training and set_lora_device not working are distinct LoRA lifecycle problems." }, { "left": "issue:1261", "right": "issue:1969", "accept": false, - "reason": "xFormers attention backward assert vs a tensor view/stride error; unrelated runtime failures." + "reason": "Xformers backward assertion and tensor view/stride incompatibility are different low-level tensor errors." }, { "left": "issue:1865", "right": "issue:2444", "accept": false, - "reason": "ZeRO stage-3 SD2 finetune dimension error vs Dreambooth inpaint script exit-status failure; different training paths and errors." + "reason": "Zero stage 3 fine-tune dimension error and dreambooth inpaint script exit-status failure are not the same bug." }, { "left": "issue:5327", "right": "issue:8854", - "accept": true, - "reason": "Both are SDXL textual inversion / textual embedding load failures; same feature and same underlying loader bug." + "accept": false, + "reason": "SDXL textual inversion loading error and SDXL textual embedding loading failure are related only broadly; they likely involve different loaders and error causes." }, { "left": "issue:6815", "right": "issue:8832", "accept": false, - "reason": "Black-image cast warning vs a crash on `pipe(prompt).images[0]`; different outcomes." + "reason": "Black image / invalid cast during output conversion is not the same as a crash when calling the pipeline." }, { "left": "issue:4186", "right": "issue:5224", "accept": false, - "reason": "SDXL inpaint/refiner behavior vs Flax Stable Diffusion 2.1 error; different pipelines and model families." + "reason": "SDXL inpaint with refiner and a Flax Stable Diffusion 2.1 error are unrelated models and code paths." }, { "left": "issue:8955", "right": "issue:9497", "accept": false, - "reason": "CLIPTextModelWithProjection state-dict loading error vs Flux Dreambooth mat2 shape error; unrelated code paths." + "reason": "CLIPTextModelWithProjection state-dict loading failure and Dreambooth Flux mat2 shape error are different model-loading/training issues." }, { "left": "issue:2775", "right": "issue:6239", "accept": false, - "reason": "torch._dynamo optimize misuse vs torch FX support for VAE latent sampling; not the same issue." + "reason": "torch._dynamo.optimize misuse and FX support for VAE sampling are different torch-integration problems." }, { "left": "issue:1005", "right": "issue:2786", "accept": false, - "reason": "Timesteps configuration question vs numpy-input incompatibility in Depth2Img; distinct problems." + "reason": "Scheduler timestep limit question and numpy-input incompatibility in Depth2Img are unrelated." }, { "left": "issue:11833", "right": "issue:12689", "accept": false, - "reason": "`set_lora_device` KeyError vs LoRA adapter weights not loaded from checkpoints; both LoRA-related, but different failures." + "reason": "Both are LoRA-related, but one is a KeyError in set_lora_device and the other is missing adapter weights from checkpoints; different failure points." }, { "left": "issue:6383", "right": "issue:9810", "accept": false, - "reason": "Multi-stage learning-rate scheduler support vs `module_to_save()` support in LoRA; unrelated feature requests." + "reason": "Multi-stage learning rates in get_scheduler and LoRA module_to_save support are unrelated feature requests/issues." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The cluster summary is mostly conservative, but one accepted soft-edge verdict is not grounded in the packet. `issue:9876` is about VRAM not being released after LoRA training, while `issue:9913` is about `set_lora_device` not working; the titles do not support calling these the same function-path bug. `issue:5327` vs `issue:8854` looks plausible, but the pair involving `9876/9913` overstates the evidence, so the overall result should be rejected." - }, - "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -41590,10 +41611,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -41608,156 +41629,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the issues span unrelated failures in loading, training, pipelines, hardware backends, and conversion scripts. None of the soft-edge pairs look like the same underlying bug or change.", - "confidence": 0.97, - "canonical_issue_reason": "No single issue is a good canonical: the items are not one duplicate family but many distinct bugs across different pipelines and error modes.", + "summary": "These items are mostly unrelated bug reports across different pipelines, training scripts, and runtime environments. The soft pairs share only generic error wording or the same broad area, not the same underlying defect, so none should be merged as duplicates.", + "confidence": 0.87, + "canonical_issue_reason": "issue:10066 is the most concrete, narrowly scoped bug report in the set, centered on a specific self.attn unpacking failure; it is easier to represent than broader or more environment-specific reports.", "canonical_pr_reason": null, - "best_issue_reason": "No best issue candidate; the most central-looking titles are still too generic and do not cover the rest of the cluster.", + "best_issue_reason": "issue:10066 is a strong issue candidate because it names a precise runtime error and code path, making it highly actionable and representative of a real defect report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10066", "right": "issue:1407", "accept": false, - "reason": "Both are ValueError-style reports, but one is about self.attn unpacking and the other about a missing feature_extractor in an ONNX pipeline; different code paths and symptoms." + "reason": "Both mention ValueError-like failures, but one is a self.attn unpacking bug and the other is an ONNX pipeline feature_extractor regression; different code paths and causes." }, { "left": "issue:3808", "right": "issue:5455", "accept": false, - "reason": "Different problems: DeepFloyd IF stage-2 training failure vs a dreambooth+lora notebook subprocess error; not the same bug." + "reason": "Both are Dreambooth/IF training failures, but one is about DeepFloyd IF stage 2 training support and the other is a subprocess error in a notebook; not the same bug." }, { "left": "issue:2645", "right": "issue:7180", "accept": false, - "reason": "Both mention device/offload-related behavior, but one is a mac M1 CUDA assertion and the other is SDXL ControlNet CPU offload; different environments and failures." + "reason": "Both involve platform/offload issues, but one is a Mac M1 CUDA assertion and the other is an SDXL ControlNet cpu offload failure; different environments and failure modes." }, { "left": "issue:5799", "right": "issue:8494", "accept": false, - "reason": "Inpainting strength keyword handling is unrelated to SD3 force_hooks argument handling; separate API issues." + "reason": "The titles both mention argument handling, but one is an inpainting strength keyword issue and the other is an SD3 force_hooks argument error; unrelated APIs." }, { "left": "issue:11448", "right": "issue:12195", "accept": false, - "reason": "GGUF model loading for a specific pipeline and a torch.library custom_op compatibility bug are unrelated." + "reason": "One is a missing fusing/stable-unclip-2-1-h-img2img model path, the other is a PyTorch 2.3.1 custom_op compatibility break; completely different problems." }, { "left": "issue:2321", "right": "issue:3803", "accept": false, - "reason": "Model copying/EMA weights in inference is unrelated to multi-GPU LoRA training model sharing." + "reason": "One concerns inference model copying and ema_weights, while the other is multi-GPU LoRA training model sharing; same training area but not the same defect." }, { "left": "issue:4073", "right": "issue:455", "accept": false, - "reason": "Both are backend-related tensor issues, but one is a DeepFloyd IF safety checker astype problem and the other is an MPS aten::index.Tensor limitation." + "reason": "Both are tensor/runtime errors on non-CUDA backends, but one is a DeepFloyd safety checker astype issue and the other is an MPS aten::index.Tensor NotImplementedError; distinct code paths." }, { "left": "issue:3296", "right": "issue:4911", "accept": false, - "reason": "Checkpoint save/load with text_encoder and custom text_encoder loading in from_single_file are different loading/training workflows." + "reason": "Both are loading/saving or custom encoder related, but one is dreambooth checkpoint saving with text_encoder and the other is custom text_encoder loading in from_single_file; not clearly the same bug." }, { "left": "issue:6598", "right": "issue:6639", "accept": false, - "reason": "Both concern differing outputs, but one is generic img2img parity and the other is an SDXL LoRA loading-path discrepancy; not the same defect." + "reason": "Both discuss output differences, but one compares img2img results versus A1111 and the other compares two LoRA loading methods in SDXL; different reproduction paths." }, { "left": "issue:7864", "right": "issue:8274", "accept": false, - "reason": "These are distinct video-generation research model requests/issues, not duplicate bug reports." + "reason": "Both are video-generation research items, but they describe different methods and no shared bug or implementation issue." }, { "left": "issue:3978", "right": "issue:7421", "accept": false, - "reason": "MPS img2img failure and seed non-determinism in StableDiffusionXLKDiffusionPipeline are unrelated." + "reason": "Both are determinism/platform-adjacent pipeline issues, but one is MPS img2img failure and the other is seed handling in KDiffusion; not the same bug." }, { "left": "issue:11839", "right": "issue:12009", "accept": false, - "reason": "Both are GGUF/load failures, but they affect different pipelines (FluxKontext vs WanTransformer3D) and different error conditions." + "reason": "Both involve loading newer model formats, but one is FluxKontextPipeline GGUF invalid input size and the other is Wan2.2 GGUF meta tensor copying failure; different loaders and errors." }, { "left": "issue:3346", "right": "issue:5412", "accept": false, - "reason": "Resume-from-checkpoint support in one dreambooth LoRA script is not the same as a training failure in the SDXL variant." + "reason": "Both are checkpoint/resume training issues in dreambooth_lora scripts, but one requests resume_from_checkpoint support and the other reports a specific SDXL resume failure; not enough to treat as one defect." }, { "left": "issue:4985", "right": "issue:6238", "accept": false, - "reason": "Training OOM for InstructPix2Pix SDXL and an SDXL VAE issue are different resource/model problems." + "reason": "Both are SDXL training/runtime issues, but one is an InstructPix2Pix OOM and the other is an SDXL VAE problem; unrelated failure modes." }, { "left": "issue:4389", "right": "issue:6602", "accept": false, - "reason": "Loading ControlNetModel versus loading a pretrained model after fuse_lora/save_pretrained are different loader paths." + "reason": "Both involve model loading errors, but one is ControlNetModel loading and the other is loading after fuse_lora/save_pretrained; separate code paths." }, { "left": "issue:7466", "right": "issue:9091", "accept": false, - "reason": "Both involve torch.compile, but one is multi-ControlNet SDXL inpaint and the other is a VAE decoder compile support issue; separate code paths." + "reason": "Both mention torch.compile, but one is multi-ControlNet SDXL inpaint compilation failure and the other is VAE decoder compile support; not the same concrete bug." }, { "left": "issue:1572", "right": "issue:3182", "accept": false, - "reason": "Both mention pipeline features not working, but img2img/inpainting support and cpu offload with a custom pipeline are distinct issues." + "reason": "Both are pipeline feature issues, but one is stable_diffusion_mega img2img/inpainting behavior and the other is CPU offload with a custom LPW pipeline; different causes." }, { "left": "issue:1976", "right": "issue:8939", "accept": false, - "reason": "Both are loading-path errors, but one is a missing tokenizer/config path and the other is an SD3 tokenizer load failure; different underlying causes." + "reason": "Both are tokenizer/file path loading errors, but one is a missing tokenizer/config.json path for stable-diffusion-v1-5 and the other is SD3 tokenizer loading for a different model family." }, { "left": "issue:2554", "right": "issue:5023", "accept": false, - "reason": "Accelerate version requirement and WuerstchenCombinedPipeline import error are unrelated environment/import issues." + "reason": "One is accelerate version incompatibility for StableDiffusionPipeline; the other is a WuerstchenCombinedPipeline import error in a later version. Same broad release breakage theme, but not the same issue." }, { "left": "issue:6176", "right": "issue:9767", "accept": false, - "reason": "LoRA weight loading after dreambooth training and a cuDNN Frontend error on Flux LoRA are different model/runtime failures." + "reason": "Both concern LoRA-related training/loading, but one is loading LoRA weights after training dreambooth_lora and the other is a FluxImg2ImgPipeLine cuDNN frontend error; different subsystems." }, { "left": "issue:6287", "right": "issue:8805", "accept": false, - "reason": "Unused setup.py dependencies is unrelated to a broken SDXL conversion script." + "reason": "One is about unused setup.py dependencies, the other about a converter script no longer working; no shared bug." }, { "left": "issue:1100", "right": "issue:6351", "accept": false, - "reason": "A CLIPTextTransformer.device attribute warning and a DistributedDataParallel.disable_adapters error are unrelated training/runtime bugs." + "reason": "Both mention attribute errors, but one is a CLIPTextTransformer.device exception and the other is DDP missing disable_adapters during LCM distillation; unrelated." }, { "left": "issue:12574", "right": "issue:13386", "accept": false, - "reason": "WanVideoToVideoPipeline control-argument handling and Qwen image DreamBooth tensor concatenation are completely different bugs." + "reason": "Both are video/training bugs, but one is WanVideoToVideoPipeline control-argument handling and the other is Dreambooth LoRA Qwen image concatenation; different code paths." }, { "left": "issue:4217", "right": "issue:8898", "accept": false, - "reason": "Slow safetensors loading and a from_pretrained subfolder path validation error are separate loading-performance vs path-resolution issues." + "reason": "Both are loading/performance issues, but one is slow safetensors loading and the other is from_pretrained subfolder path validation; not the same defect." } ] }, @@ -41859,10 +41880,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -41877,156 +41898,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: the titles point to many unrelated Diffusers issues across Flux, SDXL, IP-Adapter, LoRA, training, loading, and runtime errors. I do not see any safe duplicate groups here; the soft-similarity scores are superficial.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue fits this cluster; the items describe distinct bugs/features rather than one shared underlying defect.", + "summary": "Mostly unrelated diffusion/modeling issues; the soft pairs share only broad subsystem similarity (IP-Adapter, SDXL, LoRA, FLUX, training scripts) without the same concrete bug or fix path. No duplicate-worthy issue/PR cluster emerges.", + "confidence": 0.91, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:9500 is the strongest representative by breadth and discussion depth, but only as a stand-in for the cluster\u2014not as a duplicate target.", + "best_issue_reason": "issue:9500 is the strongest standalone report here: detailed, open, and tied to a concrete Flux checkpoint-saving failure. It is not, however, clearly a duplicate of the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:6407", "right": "issue:8414", "accept": false, - "reason": "Different requests: AnyText multilingual text generation vs a new text-to-audio consistency model." + "reason": "Different products and problem domains: AnyText text generation/editing vs a new text-to-audio consistency model. Only superficial AI-model similarity." }, { "left": "issue:6793", "right": "issue:7483", "accept": false, - "reason": "Unrelated problems: Hydra serialization failure vs LoRA-to-Diffusers conversion for SDXL-base models." + "reason": "One is Hydra serialization/configuration, the other is LoRA conversion for SDXL base models. No shared bug path." }, { "left": "issue:10705", "right": "issue:11436", "accept": false, - "reason": "Both are inference errors, but on different models and failure modes; no shared bug path." + "reason": "Pipeline reuse slowdown vs a dtype mismatch in Hidream inference. Different failure modes and code paths." }, { "left": "issue:5886", "right": "issue:7168", "accept": false, - "reason": "Both involve IP-Adapter, but one requests face model support while the other is about embedding tensor shape handling." + "reason": "Both mention IP-Adapter, but one requests face-model support and the other asks about tensor shape/embeddings. Not the same bug." }, { "left": "issue:6333", "right": "issue:9914", "accept": false, - "reason": "Different LoRA topics: PEFT pre-trained LoRA finetuning vs loading a trained Flux LoRA with Xlabs." + "reason": "Different LoRA training/loading contexts: PEFT pre-trained LoRA finetuning vs Xlabs LoRA load error. Similar theme only." }, { "left": "issue:11651", "right": "issue:11922", "accept": false, - "reason": "FluxControl flux-fill-lora loading failure is not the same as a FLUX.1-Fill-dev runtime problem." + "reason": "Flux Control LoRA loading problem vs a FLUX.1-Fill-dev runtime issue. Related model family, but not the same concrete defect." }, { "left": "issue:9500", "right": "issue:9726", "accept": false, - "reason": "Both Flux training issues, but one is checkpoint saving and the other is multi-GPU training behavior." + "reason": "Both are Flux training issues, but one is checkpoint saving and the other is multigpu Dreambooth training. Not one underlying bug." }, { "left": "issue:6911", "right": "issue:7441", "accept": false, - "reason": "Different areas: 3D conditional class-label handling vs pytest incompatibility on Python 3.12." + "reason": "Class-label handling in a 3D UNet vs pytest failing on Python 3.12. Completely unrelated." }, { "left": "issue:8504", "right": "issue:8555", "accept": false, - "reason": "Related SD3 training theme, but one is a configuration feature request and the other is FP8/text-encoder availability." + "reason": "Feature request to choose SD3 text encoders vs FP8/text-encoder training availability. Overlapping area, but not the same issue." }, { "left": "issue:5529", "right": "issue:7311", "accept": false, - "reason": "Different training bugs: SDXL text-to-image script issue vs checkpoint saving failure." + "reason": "Bug in SDXL text-to-image training vs checkpoint save failure. Same script family, different concrete problems." }, { "left": "issue:5956", "right": "issue:7146", "accept": false, - "reason": "Both concern fine-tuning performance/resources, but one is degraded SDXL U-Net quality and the other is CUDA OOM." + "reason": "SDXL fine-tuning performance degradation vs CUDA OOM during SDXL fine-tuning. One is quality/perf, the other is memory exhaustion." }, { "left": "issue:6336", "right": "issue:6382", "accept": false, - "reason": "TensorRT AutoencoderTiny compilation error is unrelated to IP-Adapter key errors during SDXL training/use." + "reason": "TensorRT VAE compilation with AutoencoderTiny vs an IP-Adapter key error during SDXL LCM-LoRA training/inference. Not the same code path." }, { "left": "issue:3910", "right": "issue:4488", "accept": false, - "reason": "Different loading defects: certain LoRA files not loading vs SDXL checkpoints requiring pipeline_class." + "reason": "General LoRA loading failure vs SDXL checkpoint loading requiring pipeline_class. Different loading mechanisms and symptoms." }, { "left": "issue:8874", "right": "issue:9171", "accept": false, - "reason": "Both are single-file checkpoint loading issues, but one is CKPT unpickling failure and the other is missing v_prediction support." + "reason": "General CKPT from_single_file unpickling failure vs a regression for SD 1.5 v_prediction handling. Different root causes." }, { "left": "issue:1925", "right": "issue:5504", "accept": false, - "reason": "Different half-precision problems: SD2 inpainting fp16 vs PEFT LoRA incompatibility with xformers." + "reason": "Half-precision inpainting issue vs PEFT LoRA with xformers. Same precision/LoRA ecosystem, but not the same bug." }, { "left": "issue:4840", "right": "issue:7848", "accept": false, - "reason": "AutoencoderKL checkpoint loading in Flax is not the same bug as LPW SDXL decoding." + "reason": "Flax autoencoder checkpoint conversion bug vs community SDXL decoder issue. Both involve decoding/loading, but distinct artifacts and failures." }, { "left": "issue:2554", "right": "issue:785", "accept": false, - "reason": "Different performance/version issues: accelerate version requirement vs scheduler runtime cost." + "reason": "Package/version requirement problem vs scheduler performance cost. No duplicate relationship." }, { "left": "issue:10782", "right": "issue:9488", "accept": false, - "reason": "Different models and symptoms: Lumina Image 2.0 minor issue vs Lumina pipeline failing to generate any image." + "reason": "Different models and symptoms: Lumina Image 2.0 minor issue vs Lumina pipeline failing to generate any image. Too vague to merge." }, { "left": "issue:2947", "right": "issue:3811", "accept": false, - "reason": "Load-attn-procs missing on UNet is a different API issue than LoRA having no effect on generation." + "reason": "Missing UNet load_attn_procs attribute vs LoRA loaded but having no effect. Related to attention/LoRA, but not the same defect." }, { "left": "issue:4257", "right": "issue:4675", "accept": false, - "reason": "Reference-only/outpainting support for SDXL is broader feature work, not the same as ControlNet Reference Only for SD XL." + "reason": "Reference-only/outpainting support request for SDXL vs ControlNet reference-only for SD XL. Similar feature area, but not the same specific request." }, { "left": "issue:4450", "right": "issue:8511", "accept": false, - "reason": "Text inversion example bug is unrelated to SD3 text-encoder unloading." + "reason": "Multi text inversion example not working vs SD3 scripts not unloading text encoders. Different scripts and bugs." }, { "left": "issue:6008", "right": "issue:9187", "accept": false, - "reason": "Both mention IP-Adapter/face-related behavior, but one is LCM LoRA compatibility and the other is SDXL image_encoder_folder=None regression." + "reason": "IP-Adapter plus face failing with LCM LoRA vs IP Adapter broken when image_encoder_folder=None after Kolors update. Both IP-Adapter-related, but distinct failure conditions." }, { "left": "issue:6382", "right": "issue:9136", "accept": false, - "reason": "IP-Adapter key error during training/use is unrelated to output-quality degradation at certain resolutions." + "reason": "IP-Adapter key error in SDXL training/inference vs quality issues at some resolutions. Different symptom class and likely different root cause." }, { "left": "issue:1543", "right": "issue:5502", "accept": false, - "reason": "Different fp16 support failures in separate scripts: textual inversion revision handling vs custom diffusion training script." + "reason": "textual_inversion fp16 revision issue vs custom diffusion training not supporting fp16. Same precision keyword, but different scripts and fixes." } ] }, @@ -42059,12 +42080,14 @@ "issue:1076", "issue:10839", "issue:10954", + "issue:1188", "issue:1214", "issue:12157", "issue:12547", "issue:1310", "issue:1500", "issue:1540", + "issue:1755", "issue:1791", "issue:1919", "issue:2645", @@ -42095,8 +42118,6 @@ "issue:8613", "issue:8668", "issue:8834", - "issue:8874", - "issue:8900", "issue:9450" ], "soft_pairs": [ @@ -42120,183 +42141,183 @@ "issue:8555|issue:8834", "issue:10839|issue:12157", "issue:1214|issue:5502", - "issue:8874|issue:8900", "issue:7000|issue:7950", "issue:1540|issue:1919", - "issue:10392|issue:10954" + "issue:10392|issue:10954", + "issue:1188|issue:1755" ], - "prepared_review_unit_hash": "9d748a850138253a4d07b8c8a437e0b63a517510078f025efb556f39fbbc4b9a", + "prepared_review_unit_hash": "dbf72a4f58b11ad67fa4f42deb9b8de1ea2b8b3424b7c393e40080a31d3659f7", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34242, - "estimated_input_tokens": 8561, - "estimated_eval_tokens": 17378 + "serialized_chars": 34190, + "estimated_input_tokens": 8548, + "estimated_eval_tokens": 17352 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated Diffusers issues spanning different models, platforms, and failure modes. The soft pairs share broad keywords (MPS, LoRA, OOM, conversion, training) but not the same concrete bug, so none should be merged as duplicates.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "The cluster is heterogeneous: most items are distinct model-/pipeline-specific bugs, setup issues, or feature requests. None of the soft pairs look like the same underlying defect, so I rejected them all.", + "confidence": 0.78, + "canonical_issue_reason": "issue:10671 is the best umbrella issue here because it is the most general runtime device/dtype bug and is closest in theme to several other device-mismatch reports, even though the cluster is not actually duplicate-heavy.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:10671 is the strongest representative for the cluster\u2019s loose device/dtype/runtime-error theme; it is still not a true duplicate target for the rest, but it is the clearest canonical issue candidate among this set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:2645", "right": "issue:5617", "accept": false, - "reason": "Both touch ControlNet/Mac-related workflows, but one is a CUDA-on-M1/MPS failure and the other is a save/load problem after torch.compile; different code paths and symptoms." + "reason": "Both involve ControlNet, but one is a Mac/M1 CUDA-support failure and the other is a torch.compile/save_pretrained loading issue; different code paths and failure modes." }, { "left": "issue:10671", "right": "issue:8613", "accept": false, - "reason": "Deterministic dtype/device lookup behavior is unrelated to the caption_column OOM during training." + "reason": "One is a deterministic dtype/device access bug in DiffusionPipeline; the other is an OOM during caption_column training. Different bug class." }, { "left": "issue:6524", "right": "issue:7485", "accept": false, - "reason": "LoRA reloading bug and safety_checker NoneType error are separate pipeline problems." + "reason": "Loading LoRA a second time is unrelated to setting safety_checker to None and hitting a callable error. Not the same defect." }, { "left": "issue:7168", "right": "issue:9450", "accept": false, - "reason": "IP-Adapter embedding shape handling is unrelated to Flux multi-GPU device placement mismatch." + "reason": "IP-Adapter embed shape handling and Flux multi-GPU transformer device placement are different underlying problems." }, { "left": "issue:1500", "right": "issue:4305", "accept": false, - "reason": "Different failures: community pipeline API incompatibility vs torch.compile graph break on a newer diffusers version." + "reason": "A community pipeline being incompatible with a newer diffusers version is not the same as a torch.compile graph break." }, { "left": "issue:1791", "right": "issue:4480", "accept": false, - "reason": "Both involve Mac/MPS, but one is an unsupported pipeline and the other is a float16 kernel crash; not the same bug." + "reason": "Both are Mac/MPS-related, but one is unsupported pipeline/backend behavior and the other is a float16 kernel crash in StableDiffusionPipeline." }, { "left": "issue:10026", "right": "issue:3382", "accept": false, - "reason": "Both mention multi-GPU memory, but they are different training scripts and different failure modes; not clearly the same underlying defect." + "reason": "Both mention multi-GPU memory, but the first is Flux ControlNet + DeepSpeed Stage-3 not reducing memory and the second is a text-to-image training OOM; not the same bug." }, { "left": "issue:12547", "right": "issue:7146", "accept": false, - "reason": "Missing --instance_prompt is a CLI/argument issue, not the same as the SDXL fine-tuning CUDA OOM." + "reason": "Missing --instance_prompt in Dreambooth Flux Kontext I2I is an argument/CLI issue, while the SDXL item is an OOM during training." }, { "left": "issue:2695", "right": "issue:6444", "accept": false, - "reason": "Validation/TensorBoard reporting failure and checkpointing-related peft_config AttributeError are unrelated." + "reason": "TensorBoard validation/report_to failure and missing peft_config during Dreambooth LoRA checkpointing are unrelated training bugs." }, { "left": "issue:4788", "right": "issue:6956", "accept": false, - "reason": "Both are OOM reports, but they affect different pipelines and environments, so they are not the same concrete bug." + "reason": "Both are OOMs, but one is SDXL inpaint on low VRAM and the other is Stable Video Diffusion on SageMaker; different pipelines and constraints." }, { "left": "issue:10334", "right": "issue:10520", "accept": false, - "reason": "Both are Sana-related, but one is gray/NaN output on MPS/CPU and the other is a cross-device tensor runtime error; different issues." + "reason": "Both are Sana-related, but gray/NaN outputs on MacOS MPS/CPU are not the same as an explicit cuda:0 vs cpu device mismatch." }, { "left": "issue:1013", "right": "issue:3574", "accept": false, - "reason": "A pipeline compatibility complaint and a Windows from_ckpt issue are different problems." + "reason": "StableDiffusionMegaPipeline on Diffusers 6.0 and from_ckpt on Windows are unrelated compatibility issues." }, { "left": "issue:1076", "right": "issue:1310", "accept": false, - "reason": "These are too generic and describe different Stable Diffusion usage questions, not a shared bug." + "reason": "A generic Stable Diffusion question is not the same as a K-Diffusion + Stable Diffusion pipeline request." }, { "left": "issue:3967", "right": "issue:7421", "accept": false, - "reason": "Deleted-branch breakage and seed reproducibility in K-Diffusion are unrelated." + "reason": "A deleted SD-XL branch issue and a seed reproducibility bug in StableDiffusionXLKDiffusionPipeline are different problems." }, { "left": "issue:4067", "right": "issue:7057", "accept": false, - "reason": "LoRA scale inconsistency with prompt embeddings is not the same as a checkpoint load failure." + "reason": "LoRA scale inconsistency with prompt embeddings is unrelated to a checkpoint/model loading failure." }, { "left": "issue:4837", "right": "issue:6899", "accept": false, - "reason": "Same conversion script, but the reports are not specific enough to conclude the same SDXL conversion bug." + "reason": "Both mention the SD conversion script, but the titles do not point to the same concrete conversion failure; too little overlap to merge." }, { "left": "issue:5963", "right": "issue:8668", "accept": false, - "reason": "Kandinsky embed-vs-prompt handling and SD3 image-size alignment are different input-validation bugs." + "reason": "Kandinsky 3.0 embed handling and SD3 img2img input-size alignment are different models and different validation paths." }, { "left": "issue:8555", "right": "issue:8834", "accept": false, - "reason": "A feature request for FP8/text encoder training is unrelated to asking whether SD3 ControlNet training code will be released." + "reason": "These are separate feature requests: FP8/text-encoder training vs SD3 ControlNet training code release." }, { "left": "issue:10839", "right": "issue:12157", "accept": false, - "reason": "Dreambooth Flux last-step error and Qwen-Image quantization iterable error are different training/configuration failures." + "reason": "Dreambooth Flux last-step error and Qwen-Image quantization-config device iteration error are unrelated model/training bugs." }, { "left": "issue:1214", "right": "issue:5502", "accept": false, - "reason": "Half-precision CLIPTextModel behavior is not the same as a training script lacking fp16 support." - }, - { - "left": "issue:8874", - "right": "issue:8900", - "accept": false, - "reason": "One is a CKPT unpickling failure in from_single_file, the other is a question about loading a safetensors refiner file." + "reason": "Half-precision behavior in CLIPTextModel and lack of fp16 support in train_custom_diffusion.py are not the same bug." }, { "left": "issue:7000", "right": "issue:7950", "accept": false, - "reason": "A PyTorch deprecation warning is not the same as a broader PyTorch 2.3 compatibility problem." + "reason": "A PyTorch deprecation warning is not the same as a reported incompatibility with PyTorch 2.3.0." }, { "left": "issue:1540", "right": "issue:1919", "accept": false, - "reason": "Missing MODEL_TYPE in Dreambooth/transformers execution and missing datasets in requirements.txt are unrelated setup problems." + "reason": "Missing MODEL_TYPE in Dreambooth/transformers and missing datasets in requirements.txt are unrelated setup problems." }, { "left": "issue:10392", "right": "issue:10954", "accept": false, - "reason": "Both involve FLUX LoRAs, but one is an fp8 training issue and the other is a load failure in a ComfyUI-compatible LoRA; different concrete bugs." + "reason": "Both concern Flux LoRAs, but one is an fp8 training issue with Hyper-FLUX and the other is a ComfyUI-compatible LoRA load failure; not the same defect." + }, + { + "left": "issue:1188", + "right": "issue:1755", + "accept": false, + "reason": "A CLIPSeg + Stable Diffusion pipeline request is unrelated to a generic general-links issue." } ] }, @@ -42315,14 +42336,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:1062", - "issue:1188", - "issue:1755", + "issue:13300", "issue:1961", "issue:2221", "issue:2420", @@ -42370,7 +42390,6 @@ "issue:9515" ], "soft_pairs": [ - "issue:1188|issue:1755", "issue:1961|issue:2221", "issue:2991|issue:6790", "issue:3811|issue:5606", @@ -42394,186 +42413,187 @@ "issue:4911|issue:7286", "issue:4236|issue:8604", "issue:7492|issue:7801", - "issue:6794|issue:6865" + "issue:6794|issue:6865", + "issue:13300|issue:5963" ], - "prepared_review_unit_hash": "fc9cd97ada0df2b3d4703473b9f42b886f418d660c03f7bd014cf8f923854b38", + "prepared_review_unit_hash": "db6a810b63f062b632cc73d55b64bbac131ee1b76fb83c646b714b5195907a58", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 34750, - "estimated_input_tokens": 8688, - "estimated_eval_tokens": 17632 + "serialized_chars": 34274, + "estimated_input_tokens": 8569, + "estimated_eval_tokens": 17394 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated issue reports spanning different features and failure modes. None of the soft-similarity pairs is strong enough to treat as the same underlying bug/change.", - "confidence": 0.47, - "canonical_issue_reason": "issue:3811 is the cleanest representative issue: it has a specific symptom, a clear code path (LoRA loading/generation), and enough discussion to stand in as a concrete example.", + "summary": "This cluster is heterogeneous: most items are unrelated example/pipeline bugs, and the proposed soft links are only surface-level similarity. None of the candidate pairs look like the same concrete bug/change, so I would not merge them. If a representative issue is needed, the broadest user-facing one is the LoRA loading bug.", + "confidence": 0.62, + "canonical_issue_reason": "Issue 3811 is the broadest, most central bug report here: it describes LoRA loading having no effect, which is the closest recurring theme across the LoRA-related tickets in this set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:3811 is the most actionable and well-scoped issue in the set; it is specific enough to serve as a representative bug report, unlike several very generic titles here.", + "best_issue_reason": "Issue 3811 is the best representative because it is concrete, user-facing, and general enough to cover the nearby LoRA/loading regressions better than the very specific or unrelated example issues.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:1188", - "right": "issue:1755", - "accept": false, - "reason": "Different topics entirely: CLIPSeg/Stable Diffusion pipeline request vs an unrelated generic Arabic-titled issue." - }, { "left": "issue:1961", "right": "issue:2221", "accept": false, - "reason": "Same generic title, but no body evidence tying them to the same error; could easily be separate ColossalAI failures." + "reason": "Same vague title, but no evidence they describe the same underlying colossalai failure." }, { "left": "issue:2991", "right": "issue:6790", "accept": false, - "reason": "Different bugs in different scripts: ControlNet mixed-precision failure vs LCM example state_dict mismatch." + "reason": "Different problems: mixed-precision ControlNet example vs LCM checkpoint key mismatch." }, { "left": "issue:3811", "right": "issue:5606", "accept": false, - "reason": "Both involve LoRA loading, but one is no effect on generation and the other is missing alpha handling; not clearly the same bug." + "reason": "Both involve LoRA loading, but one is no-effect generation and the other is missing alpha handling; likely distinct bugs." }, { "left": "issue:4080", "right": "issue:556", "accept": false, - "reason": "One is an SDXL inpainting pipeline request; the other is seamless tiling support. Related domain, not the same change." + "reason": "Different features entirely: SDXL inpainting vs seamless texture tiling." }, { "left": "issue:7801", "right": "issue:9393", "accept": false, - "reason": "Both mention mixed precision/model saving concerns, but one is a generic fp16 mismatch and the other is a FLUX+Deepspeed save-model issue." + "reason": "Generic FP16 mismatch complaint vs a FLUX Dreambooth LoRA save-model issue; not the same bug." }, { "left": "issue:5593", "right": "issue:5963", "accept": false, - "reason": "Both are pipeline/input-handling failures, but they target different models and failure modes." + "reason": "Both pipeline-related, but one is AutoPipeline conversion and the other is passing embeds to Kandinsky 3.0." }, { "left": "issue:4030", "right": "issue:4186", "accept": false, - "reason": "Different SD features and errors: v2.1 embedding runtime error vs SDXL inpaint/refiner behavior." + "reason": "Different SD/v2.1 embedding runtime error vs SDXL inpaint/refiner behavior." }, { "left": "issue:6928", "right": "issue:7962", "accept": false, - "reason": "One requests a display option for a depth-map community pipeline; the other requests an API argument for instruct-pix2pix. Not the same bug." + "reason": "One is a documentation/option request for Marigold, the other is an API enhancement for instruct-pix2pix." }, { "left": "issue:9178", "right": "issue:9420", "accept": false, - "reason": "Both are FLUX/LoRA-related, but one is a training target-module error and the other is a saved transformer load failure after Hyper-SD LoRA." + "reason": "FLUX LoRA PEFT target-module training error is unrelated to a saved transformer failing to load." }, { "left": "issue:4584", "right": "issue:7322", "accept": false, - "reason": "Both are training-script errors, but the symptoms differ: resume checkpoint handling vs SIGKILL during training." + "reason": "Both are training-script errors, but one is resume-from-checkpoint and the other is SIGKILL during SDXL DreamBooth LoRA training." }, { "left": "issue:6351", "right": "issue:6382", "accept": false, - "reason": "Same script family, but one is a DDP attribute error and the other is an IP-Adapter key error after training; distinct code paths." + "reason": "DDP missing disable_adapters and IP-adapter key errors are different failure modes in different code paths." }, { "left": "issue:3748", "right": "issue:3958", "accept": false, - "reason": "Both concern LoRA/CivitAI usage, but one is a dtype mismatch while the other is about CPU offload interaction." + "reason": "Float16/float32 mismatch on LoRA weights is not the same as the enable_model_cpu_offload interaction bug." }, { "left": "issue:7082", "right": "issue:8975", "accept": false, - "reason": "Both are import errors, but they concern different symbols from different modules and are not the same failure." + "reason": "Separate import failures for different symbols from different modules." }, { "left": "issue:2420", "right": "issue:5678", "accept": false, - "reason": "Both touch attention/xFormers-style training behavior, but one is an xformers memory-efficient attention issue and the other is higher-order-derivative behavior with scaled_dot_product_attention." + "reason": "LoRA xFormers attention failure is unrelated to higher-order-derivative attention/autograd behavior." }, { "left": "issue:5873", "right": "issue:9515", "accept": false, - "reason": "ONNX export of scaled_dot_product_attention is unrelated to the SD3 JointAttnProcessor RMSNorm omission." + "reason": "ONNX export support problem is not the same as missing RMSNorm in JointAttnProcessor2_0." }, { "left": "issue:6832", "right": "issue:8626", "accept": false, - "reason": "Both involve IP-Adapter usage, but one is a runtime shape error and the other is documentation/guidance for multiple images/masks." + "reason": "A tuple-shape AttributeError in IP-Adapter Face ID is unrelated to guidance documentation for multiple IP adapter images/masks." }, { "left": "issue:3595", "right": "issue:4086", "accept": false, - "reason": "Different failures: from_ckpt loading vs ControlNet training on SDXL base 0.9." + "reason": "from_ckpt loading failure and SDXL ControlNet training failure are distinct problems." }, { "left": "issue:5890", "right": "issue:8880", "accept": false, - "reason": "One is a tokenizer length warning during inference; the other is a performance regression. Not the same issue." + "reason": "Token length warning during training inference is unrelated to a broader inference speed regression." }, { "left": "issue:4086", "right": "issue:5972", "accept": false, - "reason": "A training failure and a documentation request about Min-SNR weighting are not the same underlying change." + "reason": "Training failure with SDXL ControlNet is not the same as a documentation update for Min-SNR weighting." }, { "left": "issue:1062", "right": "issue:5014", "accept": false, - "reason": "Both are DreamBooth-related quality complaints, but one is about reproducing paper results and the other about LoRA training output quality." + "reason": "Dreambooth paper-result reproducibility and LoRA training results are related only by topic, not by the same bug." }, { "left": "issue:4911", "right": "issue:7286", "accept": false, - "reason": "Different code paths: custom text encoder loading from_single_file vs cross-attention weight loading in Attend-and-Excite." + "reason": "Custom text-encoder loading from_single_file is unrelated to cross-attention weight loading in Custom Diffusion." }, { "left": "issue:4236", "right": "issue:8604", "accept": false, - "reason": "Both mention SDXL-ish model components, but one is refiner text-encoder registration and the other is T5 float16 corruption." + "reason": "SDXL refiner text-encoder registration and T5 float16 corruption are different code paths and symptoms." }, { "left": "issue:7492", "right": "issue:7801", "accept": false, - "reason": "Repository lookup failure vs mixed-precision model dtype mismatch; unrelated." + "reason": "A missing repository URL/model lookup error is unrelated to mixed_precision not producing fp16 weights." }, { "left": "issue:6794", "right": "issue:6865", "accept": false, - "reason": "Importing cast_training_params and checking Image instances in load_image are unrelated fixes." + "reason": "Import error for cast_training_params and missing Image-instance check are separate utility issues." + }, + { + "left": "issue:13300", + "right": "issue:5963", + "accept": false, + "reason": "CPU generation failures in SD3.5/FLUX are unrelated to Kandinsky 3.0 embed-input handling." } ] }, @@ -42592,8 +42612,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -42603,16 +42623,17 @@ "issue:1197", "issue:12609", "issue:13258", - "issue:13300", "issue:1985", "issue:2324", "issue:2599", "issue:378", "issue:4037", "issue:4086", + "issue:4236", "issue:4337", "issue:435", "issue:4405", + "issue:4933", "issue:4964", "issue:5227", "issue:5433", @@ -42620,7 +42641,6 @@ "issue:5737", "issue:5932", "issue:5953", - "issue:5963", "issue:5966", "issue:5967", "issue:6093", @@ -42647,7 +42667,6 @@ "issue:9101" ], "soft_pairs": [ - "issue:13300|issue:5963", "issue:378|issue:435", "issue:5953|issue:8668", "issue:4337|issue:7985", @@ -42670,180 +42689,181 @@ "issue:6441|issue:7057", "issue:7420|issue:8879", "issue:8555|issue:9101", - "issue:8645|issue:8906" + "issue:8645|issue:8906", + "issue:4236|issue:4933" ], - "prepared_review_unit_hash": "23a7203d00f62ba8c68df4f2831b4cb8d1f2eea31b78e4c24c163dad5f0f6ca2", + "prepared_review_unit_hash": "5b8877ed2fd74c3fb58a1a3fcd54a50a4f90457f1742040f6d407a74d85bf89f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33881, - "estimated_input_tokens": 8471, - "estimated_eval_tokens": 17198 + "serialized_chars": 33807, + "estimated_input_tokens": 8452, + "estimated_eval_tokens": 17160 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These pairs are mostly loose topical similarities, not true duplicates. The closest ones share a theme (attention masks, torch.compile, training scripts), but the model/pipeline, failure mode, or code path differs enough that I would not merge them.", - "confidence": 0.84, - "canonical_issue_reason": "issue:8673 is the clearest, most actionable bug report in the set: a specific missing-attention-mask problem in SD3 with a concrete root cause and likely duplicate surface area.", + "summary": "No soft pair looks like the same underlying bug/change; the set is mostly unrelated training, conversion, and compile issues. No pull requests are present. The most representative issue is the SD3 attention-mask bug because it is concrete and actionable.", + "confidence": 0.78, + "canonical_issue_reason": "issue:8673 is the clearest concrete bug report here: missing attention masks in SD3 for padding tokens, with a specific model/component and a plausible targeted fix.", "canonical_pr_reason": null, - "best_issue_reason": "issue:8673 is the best standalone issue here because it is specific, technically clear, and likely to collect related reports without being overly broad.", + "best_issue_reason": "issue:8673 is the strongest standalone issue for triage: specific, reproducible, and directly tied to a code-path bug rather than a broad feature request or vague failure.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:13300", - "right": "issue:5963", - "accept": false, - "reason": "Both are generation failures, but one is CPU inference on FLUX/SD3.5 and the other is Kandinsky embeds-vs-prompts handling; different models and different bug classes." - }, { "left": "issue:378", "right": "issue:435", "accept": false, - "reason": "Both are textual-inversion training bugs, but the exceptions come from different objects/code paths and do not point to the same defect." + "reason": "Both are textual inversion/VAE-related, but they describe different failures and different symptoms; not the same underlying bug." }, { "left": "issue:5953", "right": "issue:8668", "accept": false, - "reason": "Both involve image-size/alignment crashes, but one is ControlNet image/control_image mismatch and the other is SD3 img2img alignment to 16; not the same underlying bug." + "reason": "Both involve image-size handling, but one is ControlNet img2img consistency and the other is SD3 alignment to 16; different code paths and fixes." }, { "left": "issue:4337", "right": "issue:7985", "accept": false, - "reason": "Both mention torch.compile, but they affect different components (Community ControlNetReference vs IPAdapter) and likely different failures." + "reason": "Both mention torch.compile, but one is Community ControlNetReference and the other IPAdapter; too different to merge as one fix." }, { "left": "issue:12609", "right": "issue:13258", "accept": false, - "reason": "Wan 2.2 NaNs in latents and missing TI2V image-to-video support are unrelated problems." + "reason": "Wan 2.2 latent NaNs versus a missing image-to-video model; unrelated problems." }, { "left": "issue:5433", "right": "issue:5737", "accept": false, - "reason": "One is an SDXL artifact quality issue with DPM++ schedulers; the other is a conversion-script failure. Same ecosystem, different code path." + "reason": "SDXL artifacting with schedulers versus SDXL conversion script failure; not the same bug." }, { "left": "issue:4964", "right": "issue:7483", "accept": false, - "reason": "Both concern conversion to diffusers, but one is SDXL model conversion and the other is LoRA-to-Diffusers conversion for SDXL-base models; not the same bug." + "reason": "Both are conversion-related, but one is general SDXL conversion and the other is LoRA-to-Diffusers for SDXL base models; different workflows." }, { "left": "issue:6382", "right": "issue:6790", "accept": false, - "reason": "Both are LCM/SDXL training failures, but one is an IP-adapter key error after training while the other is a state_dict key mismatch during example training." + "reason": "Both concern LCM/SDXL training, but the errors are different state-dict/key problems in different scripts." }, { "left": "issue:10066", "right": "issue:5932", "accept": false, - "reason": "Different runtime errors and different subsystems: unpacking in self.attn versus channel-mismatch in a convolution input." + "reason": "Different runtime failures: unpacking in self.attn versus channel-mismatch in a convolution input." }, { "left": "issue:10025", "right": "issue:8673", "accept": false, - "reason": "Both are about attention masks, but one targets Flux transformer attention and the other SD3 padding-token masking; same theme, not the same bug." + "reason": "Both mention attention masks, but Flux transformer versus SD3 padding-token masking are different model-specific bugs." }, { "left": "issue:10311", "right": "issue:6794", "accept": false, - "reason": "Both are import-related, but they reference different modules and failure points, so they are not duplicates." + "reason": "Both are import errors, but they point to different symbols and scripts; not a single underlying defect." }, { "left": "issue:1197", "right": "issue:1985", "accept": false, - "reason": "OpenAI Diffusion Pipeline is a feature/community request; the other is about loading embeddings in Stable Diffusion. Different requests." + "reason": "A community pipeline request versus a request for easier embedding loading; unrelated feature work." }, { "left": "issue:2324", "right": "issue:2599", "accept": false, - "reason": "Both are DreamBooth training issues, but one is not reading instance data and the other is multi-subject quality degradation; different underlying problems." + "reason": "Both are Dreambooth training issues, but one is instance data loading and the other is multi-subject quality degradation." }, { "left": "issue:4037", "right": "issue:4086", "accept": false, - "reason": "Both touch ControlNet training, but one is an FSDP validation-error issue and the other is a stable-diffusion-xl-base-0.9 setup failure; not the same code-path problem." + "reason": "Both are ControlNet training problems, but one is FSDP validation errors and the other is a general SDXL base-0.9 training failure." }, { "left": "issue:4405", "right": "issue:8887", "accept": false, - "reason": "Textual inversion script for SDXL and SD3 VAE-encode bug are unrelated training issues." + "reason": "Different features and models: SDXL textual inversion script versus SD3 VAE encode bug." }, { "left": "issue:5227", "right": "issue:7466", "accept": false, - "reason": "Both involve torch.compile, but one is LoRA loading and the other is multi-ControlNet inpaint compilation; different components and fixes." + "reason": "Both involve torch.compile, but one is LoRA loading and the other is multi-ControlNet inpainting; different code paths." }, { "left": "issue:5594", "right": "issue:6388", "accept": false, - "reason": "Memory-saving LoRA training and ignored gradient-checkpointing are related to efficiency, but not the same defect." + "reason": "Related SDXL-LoRA optimization topics, but one is memory usage generally and the other is gradient checkpointing being ignored." }, { "left": "issue:5966", "right": "issue:7882", "accept": false, - "reason": "A general ControlNet SD2.1 issue and a missing from_single_file API for T2I-Adapter are separate feature/bug reports." + "reason": "ControlNet for SD2.1 versus T2I-Adapter missing from_single_file; different features and implementations." }, { "left": "issue:5967", "right": "issue:6258", "accept": false, - "reason": "An announcement post and a Stable Video Diffusion quality bug are not duplicates." + "reason": "An announcement versus a pipeline quality bug; no shared underlying defect." }, { "left": "issue:6093", "right": "issue:8547", "accept": false, - "reason": "Push-to-hub repo-creation behavior in a consistency-model training script is unrelated to the latent-consistency scaling bug." + "reason": "SDXL LoRA push_to_hub behavior versus an LCM scaling bug; unrelated." }, { "left": "issue:6441", "right": "issue:7057", "accept": false, - "reason": "Both say something 'doesn't work,' but one is from_pretrained in Colab and the other is a checkpoint loading issue for a specific model; too vague to merge." + "reason": "Both are model-loading problems, but one is from_pretrained in Colab and the other is checkpoint loading; different causes." }, { "left": "issue:7420", "right": "issue:8879", "accept": false, - "reason": "A generic torch.compile issue and a specific torch.compile+cudagraph StableDiffusionXL failure are not clearly the same root cause." + "reason": "Both mention torch.compile, but the failure modes are different and not clearly one mergeable fix." }, { "left": "issue:8555", "right": "issue:9101", "accept": false, - "reason": "FP8/text-encoder training availability for SD3 and an InstructPix2Pix training script for SD3 are different feature requests." + "reason": "SD3 FP8/text encoder training request versus an InstructPix2Pix training script request; unrelated feature requests." }, { "left": "issue:8645", "right": "issue:8906", "accept": false, - "reason": "Missing create_diffusers_controlnet_model_from_ldm helper and missing qk_norm in SD3Transformer2DModel are unrelated API/model questions." + "reason": "Missing helper function in single_file_utils versus missing qk_norm in SD3Transformer2DModel; unrelated implementation gaps." + }, + { + "left": "issue:4236", + "right": "issue:4933", + "accept": false, + "reason": "SDXL refiner text-encoder registration versus Dreambooth output-file saving; different bugs." } ] }, @@ -42862,8 +42882,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -42882,14 +42902,13 @@ "issue:3432", "issue:3481", "issue:3494", - "issue:4236", + "issue:3651", "issue:4306", "issue:4446", "issue:4619", "issue:4631", "issue:4634", "issue:490", - "issue:4933", "issue:5399", "issue:5433", "issue:5510", @@ -42904,6 +42923,7 @@ "issue:720", "issue:7538", "issue:7629", + "issue:7960", "issue:7972", "issue:8514", "issue:8578", @@ -42917,7 +42937,6 @@ "issue:9732" ], "soft_pairs": [ - "issue:4236|issue:4933", "issue:5956|issue:6536", "issue:7629|issue:9309", "issue:10017|issue:7538", @@ -42940,180 +42959,181 @@ "issue:5399|issue:6540", "issue:1896|issue:5510", "issue:4446|issue:8874", - "issue:3432|issue:5846" + "issue:3432|issue:5846", + "issue:3651|issue:7960" ], - "prepared_review_unit_hash": "bc0e8339b019db6a029113b0ac84546ec06d54f38b5a30901978782bbceb3601", + "prepared_review_unit_hash": "f719105a645b609284ca9fe9b132f784c8c39e79d8605eb5c9a52a06c0323a66", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34366, - "estimated_input_tokens": 8592, - "estimated_eval_tokens": 17440 + "serialized_chars": 34335, + "estimated_input_tokens": 8584, + "estimated_eval_tokens": 17424 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set; it mixes many unrelated issues across dataset loading, training, inference, schedulers, serialization, and import/runtime errors. None of the soft pairs look like the same underlying bug or change.", - "confidence": 0.97, - "canonical_issue_reason": "No canonical issue: the items are heterogeneous and do not share one underlying defect.", + "summary": "The cluster is heterogeneous: most items are distinct bug reports that only share broad terms like Flux, LoRA, training, or scheduler. I found no pair that looks like the same underlying issue/change, so all soft edges should be rejected.", + "confidence": 0.91, + "canonical_issue_reason": "No single canonical issue is justified here because the items span separate failures (dataset loading, LoRA/Flux inference, training errors, scheduler bugs, import errors, and memory/offload issues) rather than one duplicate report.", "canonical_pr_reason": null, - "best_issue_reason": "If forced, issue 10381 is the richest standalone report, but it only covers one Flux/quantization/LoRA inference path and is not representative of the cluster as a whole.", + "best_issue_reason": "If a representative issue is needed, issue 5433 is the strongest candidate: it has concrete symptoms, high discussion/inbound activity, and is a well-formed user-facing bug report with clear repro context.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:4236", - "right": "issue:4933", - "accept": false, - "reason": "Different bugs: SDXL refiner text-encoder registration vs Dreambooth not returning pytorch_model.bin after training." - }, { "left": "issue:5956", "right": "issue:6536", "accept": false, - "reason": "Different problems: SD-XL U-Net fine-tuning quality degradation vs UNet3DConditionModel out-of-memory." + "reason": "Different problems: SD-XL fine-tuning quality regression vs UNet3DConditionModel OOM." }, { "left": "issue:7629", "right": "issue:9309", "accept": false, - "reason": "Different models and failures: SVD OOM on PyTorch 1.13.1 vs FluxControlNetPipeline bug on Flux 1 schnell." + "reason": "Unrelated bugs: PyTorch 1.13.1 OOM in SVD vs FluxControlNetPipeline on Flux schnell." }, { "left": "issue:10017", "right": "issue:7538", "accept": false, - "reason": "Different scope: load_dataset() loading issue vs a Custom Fill50k training dataset issue." + "reason": "Both mention dataset/training, but one is a load_dataset() issue and the other is a custom Fill50k training issue." }, { "left": "issue:2272", "right": "issue:3420", "accept": false, - "reason": "Unrelated: ONNX conversion failure vs ControlNet inpaint pipeline not working with Diffusion pipeline." + "reason": "Different code paths: ONNX conversion script failure vs ControlNet inpaint pipeline incompatibility." }, { "left": "issue:10839", "right": "issue:9732", "accept": false, - "reason": "Both are Flux training issues, but one is a last-step training error and the other is an OOM on a specific setup; not the same concrete bug." + "reason": "Both are Flux training reports, but one is a last-step error in Dreambooth LoRA training and the other is CUDA OOM in the README workflow." }, { "left": "issue:3481", "right": "issue:3494", "accept": false, - "reason": "Different scheduler bugs: DDIM step-count consistency vs Multidiffusion corruption across schedulers." + "reason": "Related to generation/schedulers, but the symptoms and pipelines differ: nondeterminism across steps vs corrupted multidiffusion outputs." }, { "left": "issue:2218", "right": "issue:3348", "accept": false, - "reason": "Overtraining during training vs xformers LoRA scale support; not the same issue." + "reason": "Overtrained-model behavior is not the same as xformers/Lora scale-parameter support." }, { "left": "issue:8915", "right": "issue:9410", "accept": false, - "reason": "Different missing imports: AutoPipelineForText2Image vs StableDiffusion3ControlNetInpaintingPipeline." + "reason": "Both are import failures, but for different classes and symbols in different pipelines." }, { "left": "issue:2891", "right": "issue:490", "accept": false, - "reason": "Docs/support mismatch for embeds vs a generic memory issue; unrelated bugs." + "reason": "Docs/support mismatch for StableDiffusionUpscalePipeline embeds is unrelated to the memory issue in older Stable Diffusion." }, { "left": "issue:2312", "right": "issue:4306", "accept": false, - "reason": "Different failures: EMA save hook error vs StableDiffusionReferencePipeline list index out of range." + "reason": "Different failures: EMA save-hook error vs StableDiffusionReferencePipeline list-index error." }, { "left": "issue:8578", "right": "issue:8644", "accept": false, - "reason": "Both mention meta tensors, but they affect different pipelines and triggers (sd_turbo 1.5 vs SD3/T5 offload)." + "reason": "Both involve meta tensors/offload, but they are different setups and code paths (sd_turbo 1.5 vs SD3 T5 encoder offload)." }, { "left": "issue:10180", "right": "issue:9108", "accept": false, - "reason": "Different training problems: multiple LoRAs with Flux Control LoRA vs resuming training with a quantized base model." + "reason": "Different Flux/LoRA training-vs-resume problems; one is loading multiple LoRAs, the other resuming with a quantized base model." }, { "left": "issue:4631", "right": "issue:720", "accept": false, - "reason": "Scheduler scalar-conversion error vs traced UNet half/float dtype mismatch; different code paths." + "reason": "Similar scalar/Half runtime flavor, but they are different scheduler behaviors and different repros." }, { "left": "issue:4619", "right": "issue:7972", "accept": false, - "reason": "Both are dtype mismatch reports, but they occur in different scripts/features (SDXL LoRA training vs LEDITS++)." + "reason": "Both are type-mismatch training issues, but for different scripts and failure modes." }, { "left": "issue:10381", "right": "issue:9096", "accept": false, - "reason": "Flux inference issues, but one is quantization/LoRA/Quanto related and the other is NaNs from torch.half; not the same bug." + "reason": "Both involve Flux inference instability, but one is quantization/LoRA/Quanto-related and the other is torch.half NaNs." }, { "left": "issue:3235", "right": "issue:4634", "accept": false, - "reason": "Custom Diffusion-related, but one is weights-with-biases loading and the other is safe serialization; not clearly the same defect." + "reason": "Custom Diffusion loading vs safe-serialization support are different serialization issues." }, { "left": "issue:6164", "right": "issue:6996", "accept": false, - "reason": "Different bugs: latent output handling in StableDiffusionPipeline vs NoneType added_cond_kwargs check." + "reason": "Different pipeline bugs: latent output handling vs NoneType added_cond_kwargs membership check." }, { "left": "issue:6224", "right": "issue:8514", "accept": false, - "reason": "Different fine-tuning failures: TorchDynamo issue vs EMAModel causing InstructP2P parallel fine-tuning error." + "reason": "TorchDynamo fine-tuning issue and EMAModel/InstructP2P parallel fine-tuning error are separate training failures." }, { "left": "issue:5433", "right": "issue:5546", "accept": false, - "reason": "Related scheduler/theme, but one is visual artifacts without refiner and the other is incorrect step counts with base+refiner ensemble." + "reason": "Both mention SDXL schedulers, but one is visual artifacts without refiner and the other is wrong step counts with Karras sigmas." }, { "left": "issue:5399", "right": "issue:6540", "accept": false, - "reason": "Different runtime errors in different contexts: custom inpainting dtype mismatch vs FSDP controlnet mat2 matrix error." + "reason": "Different dtype/matrix errors in different finetuning/inference paths; not the same bug." }, { "left": "issue:1896", "right": "issue:5510", "accept": false, - "reason": "Different pipelines and failures: CLIP-guided Stable Diffusion size error vs SDXL InstructPix2Pix dtype mismatch." + "reason": "Both are runtime dtype-ish errors, but in unrelated pipelines and with different root causes." }, { "left": "issue:4446", "right": "issue:8874", "accept": false, - "reason": "Different loading problems: custom text_encoder support in from_single_file vs CKPT weights-only unpickling failure." + "reason": "Both concern from_single_file loading, but one is custom text encoder injection and the other is CKPT unpickling failure." }, { "left": "issue:3432", "right": "issue:5846", "accept": false, - "reason": "Different training/runtime issues: Half on CPU conv error vs resume_from_checkpoint failure." + "reason": "Different problems: CPU half-precision conv failure vs resume_from_checkpoint error." + }, + { + "left": "issue:3651", + "right": "issue:7960", + "accept": false, + "reason": "Both are multi-GPU training failures, but they are distinct models and failure modes." } ] }, @@ -43132,8 +43152,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -43151,8 +43171,9 @@ "issue:2414", "issue:2420", "issue:3064", - "issue:3651", "issue:3826", + "issue:4448", + "issue:4460", "issue:4480", "issue:455", "issue:4688", @@ -43176,7 +43197,6 @@ "issue:7480", "issue:7722", "issue:7848", - "issue:7960", "issue:8624", "issue:867", "issue:8738", @@ -43186,7 +43206,6 @@ "issue:993" ], "soft_pairs": [ - "issue:3651|issue:7960", "issue:10030|issue:6996", "issue:1166|issue:4954", "issue:2420|issue:6363", @@ -43209,186 +43228,187 @@ "issue:5617|issue:7420", "issue:5012|issue:5821", "issue:867|issue:993", - "issue:8738|issue:9660" + "issue:8738|issue:9660", + "issue:4448|issue:4460" ], - "prepared_review_unit_hash": "2c9ae8d5a9fb5d7610b8843ec0a1e803f86984f4db399ac1cda965771f885ad2", + "prepared_review_unit_hash": "77d4610275e0df1153393365cdec90e3dface97f457c37c3939858236fabc9eb", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33479, - "estimated_input_tokens": 8370, - "estimated_eval_tokens": 16996 + "serialized_chars": 33565, + "estimated_input_tokens": 8392, + "estimated_eval_tokens": 17040 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issues grouped by superficial keyword overlap; the only clear duplicate pair is the added_cond_kwargs None handling in ControlNet/SDXL pipeline code.", - "confidence": 0.86, - "canonical_issue_reason": "issue:10030 is the clearest root-cause description of the shared added_cond_kwargs None bug in the ControlNet/img2img path.", + "summary": "The cluster is mostly false-positive topical similarity. The only plausible duplicate pair is the ControlNet/added_cond_kwargs None-handling regression; the rest appear to be different bugs, features, or unrelated pipeline/model issues.", + "confidence": 0.72, + "canonical_issue_reason": "issue:10030 is the clearest, most concrete report of the added_cond_kwargs null regression in ControlNet img2img, and it likely covers the symptom described in 6996.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10030 is the best representative issue because it states the underlying pipeline contract mismatch most directly and generally.", + "best_issue_reason": "issue:10030 is the best representative issue: specific pipeline, clear failure mode, and easiest to map to a fix.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:3651", - "right": "issue:7960", - "accept": false, - "reason": "Both are training/multi-GPU problems, but they involve different scripts and failure contexts." - }, { "left": "issue:10030", "right": "issue:6996", "accept": true, - "reason": "Both describe the same underlying bug: code assumes added_cond_kwargs is a dict, but the pipeline passes None." + "reason": "Both describe the same underlying bug: code assumes added_cond_kwargs is a dict, but pipeline ControlNet/img2img can pass None, causing a null-membership/type error." }, { "left": "issue:1166", "right": "issue:4954", "accept": false, - "reason": "Different textual inversion failures: TPU save hang vs a pickling/unpickling error." + "reason": "Both are textual inversion issues, but one is TPU save hang and the other is an UnpicklingError; different failure modes." }, { "left": "issue:2420", "right": "issue:6363", "accept": false, - "reason": "Both are training errors, but one is xformers attention and the other is FP16 gradient unscaling." + "reason": "Different training problems: xformers attention bug vs FP16 gradient unscale error." }, { "left": "issue:3064", "right": "issue:5128", "accept": false, - "reason": "Both mention safetensors/loading, but they are different loading paths and different model types." + "reason": "Both involve .safetensors loading, but one is LoRA loading generally and the other is XLPipeline.from_single_file; not the same bug." }, { "left": "issue:5845", "right": "issue:9516", "accept": false, - "reason": "DreamBooth Pixart training and FLUX joint_attention_kwargs are unrelated bugs." + "reason": "Unrelated model families and failures: PixArt DreamBooth training vs FLUX joint_attention_kwargs propagation." }, { "left": "issue:6894", "right": "issue:7848", "accept": false, - "reason": "One is LoRA loading in A1111, the other is a community example decode problem." + "reason": "Different outputs and code paths: LoRA export compatibility vs lpw SDXL decoding issue." }, { "left": "issue:7168", "right": "issue:7480", "accept": false, - "reason": "IP-Adapter embed shape handling and DeepFloyd_IF OOM are distinct problems." + "reason": "IP-Adapter tensor-shape issue vs DeepFloyd IF OOM; no shared underlying bug." }, { "left": "issue:1704", "right": "issue:528", "accept": false, - "reason": "These are general questions, not the same bug or change." + "reason": "Both are general questions, not the same topic or bug." }, { "left": "issue:4993", "right": "issue:5152", "accept": false, - "reason": "Missing import structure entry for CLIPImageProjection is unrelated to the Wurstchen bug." + "reason": "Import-structure bug for CLIPImageProjection vs Wuerstchen alpha bug; unrelated." }, { "left": "issue:5128", "right": "issue:7722", "accept": false, - "reason": "Local safetensors loading in XLPipeline and Animatediff+IPAdapter errors are different code paths." + "reason": "Local safetensors loading error vs Animatediff ControlNet/IPAdapter usage problem; different code paths." }, { "left": "issue:11370", "right": "issue:9187", "accept": false, - "reason": "Invalid load key during Hunyuan conversion is unrelated to SDXL IP-Adapter behavior after Kolors update." + "reason": "Different failures: invalid pickle/load-key in conversion script vs SDXL IP-Adapter image_encoder_folder=None regression." }, { "left": "issue:4480", "right": "issue:455", "accept": false, - "reason": "Both are MPS-related, but one is an fp16 kernel crash and the other is a specific aten::index.Tensor NotImplementedError." + "reason": "Both are MPS-related, but one is a kernel crash on fp16 and the other is an aten::index.Tensor NotImplementedError." }, { "left": "issue:3826", "right": "issue:5442", "accept": false, - "reason": "Audio pipeline proposal and ScaleCrafter are unrelated." + "reason": "Both are new pipeline proposals, but they are entirely different pipelines and changes." }, { "left": "issue:10795", "right": "issue:10937", "accept": false, - "reason": "Both involve torch.compile, but GGUF loading and vae.encode failures are different bugs." + "reason": "Both mention torch.compile, but one is GGUF loading via from_single_file and the other is vae.encode; different concrete problems." }, { "left": "issue:11232", "right": "issue:12547", "accept": false, - "reason": "Text-to-image fine-tuning noise residuals and missing instance_prompt in Flux DreamBooth are unrelated." + "reason": "Noise residuals during fine-tuning vs missing required --instance_prompt; unrelated training issues." }, { "left": "issue:4688", "right": "issue:6726", "accept": false, - "reason": "Different research/pipeline topics with no shared failure mode." + "reason": "Different research/pipeline topics with no shared bug or code path." }, { "left": "issue:1377", "right": "issue:2414", "accept": false, - "reason": "ImportError for a missing transformers symbol is not the same as NameError: transformers not defined." + "reason": "One is a missing transformers symbol import, the other is a NameError from not importing transformers; different root causes." }, { "left": "issue:2225", "right": "issue:5337", "accept": false, - "reason": "A safety-checker usage question and a callback question are not duplicates." + "reason": "A usage question about safety checker vs a callback explanation question; not duplicates." }, { "left": "issue:11285", "right": "issue:8624", "accept": false, - "reason": "Conversion errors to/from diffusers and an undefined symbol in fused_layer_norm are unrelated." + "reason": "Conversion/value errors versus a runtime shared-library symbol error; not the same issue." }, { "left": "issue:5617", "right": "issue:7420", "accept": false, - "reason": "Both mention torch.compile, but save/load after compile and general compile failure are different issues." + "reason": "Loading a compiled ControlNet pipeline after save_pretrained vs general inability to use torch.compile; different specifics." }, { "left": "issue:5012", "right": "issue:5821", "accept": false, - "reason": "LoRA text encoder training and a CUDA backward/training failure are different problems." + "reason": "LoRA text-encoder training behavior vs a generic CUDA backward failure; unrelated." }, { "left": "issue:867", "right": "issue:993", "accept": false, - "reason": "GPU memory leak and fp16 data type mismatch are unrelated." + "reason": "GPU memory leak in FastAPI vs fp16 data type mismatch; different bugs." }, { "left": "issue:8738", "right": "issue:9660", "accept": false, - "reason": "SD3Transformer2DModel configuration and from_pretrained eval-mode behavior are separate bugs." + "reason": "Model configuration failure in SD3Transformer2DModel vs from_pretrained not setting eval mode; different behavior." + }, + { + "left": "issue:4448", + "right": "issue:4460", + "accept": false, + "reason": "Both discuss output differences, but one is ControlNet outpainting quality and the other is a broad Stable Diffusion vs diffusers comparison; too vague to merge." } ] }, "evaluator_result": { "accept": true, - "feedback": "Grounded and conservative overall. The only accepted soft pair is well-supported by the titles (10030 and 6996 both point to added_cond_kwargs being None), and the rejected pairs are appropriately cautious given the limited evidence in the packet. The summary does not overstate beyond what the issue titles support." + "feedback": "The summary is grounded in the packet and stays conservative. The accepted 10030\u21946996 pair is well supported by the titles, which both point to the same added_cond_kwargs None-handling regression in ControlNet/img2img. The remaining pairs are rejected cautiously without overclaiming duplicate status, and the overall summary appropriately characterizes the rest as likely false-positive topical similarity." }, "evaluator_used": true, "retried": false, @@ -43404,8 +43424,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -43429,10 +43449,10 @@ "issue:4206", "issue:4306", "issue:4353", + "issue:4375", "issue:4389", - "issue:4448", - "issue:4460", "issue:4566", + "issue:4731", "issue:4852", "issue:5028", "issue:5576", @@ -43458,7 +43478,6 @@ "issue:9902" ], "soft_pairs": [ - "issue:4448|issue:4460", "issue:4389|issue:8493", "issue:2156|issue:2480", "issue:3826|issue:5576", @@ -43481,108 +43500,103 @@ "issue:6092|issue:9044", "issue:3174|issue:4852", "issue:7716|issue:7936", - "issue:7629|issue:9613" + "issue:7629|issue:9613", + "issue:4375|issue:4731" ], - "prepared_review_unit_hash": "de8d5ec47f305c72325b3a1f0593f795af19ea9efc43795da47489fc4d9b0bd0", + "prepared_review_unit_hash": "3a0919571159a037211db661eca6a7ceff4ee57570174e4921318d7b479af300", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33217, - "estimated_input_tokens": 8305, - "estimated_eval_tokens": 16866 + "serialized_chars": 33133, + "estimated_input_tokens": 8284, + "estimated_eval_tokens": 16824 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is thematically related around diffusion/Flux/LoRA/controlnet issues, but the soft pairs are mostly different concrete bugs or feature requests rather than true duplicates. I would reject all soft edges.", - "confidence": 0.95, - "canonical_issue_reason": null, + "summary": "This set is heterogeneous: most items are separate bugs, feature requests, or training/loading problems that only share broad ML/Flux/SDXL vocabulary. I would not collapse any of the soft pairs as duplicates.", + "confidence": 0.88, + "canonical_issue_reason": "issue:10972 is the strongest concrete bug report in the broad Flux/LoRA area, with a specific failing code path (loading OneTrainer Flux LoRAs). It is still not a true cluster representative for the rest, which span unrelated pipelines and failure modes.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 10972 is the most concrete standalone bug report in the set, with a specific Flux LoRA loading failure and clearer diagnostic context than the others, but it is not a true canonical for the whole cluster.", + "best_issue_reason": "issue:10972 has the clearest actionable reproduction and a narrowly defined failure mode, making it the best issue anchor among this mixed set.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:4448", - "right": "issue:4460", - "accept": false, - "reason": "Both are about image quality/differences, but one is outpainting quality vs A1111 and the other is general Stable Diffusion vs diffusers mismatch; different concrete problems." - }, { "left": "issue:4389", "right": "issue:8493", "accept": false, - "reason": "Both are loading/model-state errors, but they target different model classes and failure modes: ControlNetModel loading vs SD3Transformer2DModel state_dict mismatch." + "reason": "Different loading failures: ControlNetModel init vs SD3Transformer2DModel state_dict mismatch. Not the same code path." }, { "left": "issue:2156", "right": "issue:2480", "accept": false, - "reason": "Dreambooth-related, but one is CPU-vs-GPU training behavior and the other is checkpoint directory output structure; not the same bug." + "reason": "One is GPU-vs-CPU training execution, the other is checkpoint directory layout/output saving. Separate bugs." }, { "left": "issue:3826", "right": "issue:5576", "accept": false, - "reason": "These are unrelated feature requests: a new audio pipeline vs a new diffusion scheduler." + "reason": "A new pipeline request is not the same as a new scheduler request; different artifact and scope." }, { "left": "issue:7311", "right": "issue:9501", "accept": false, - "reason": "Both involve training/checkpoint saving, but one is an SDXL checkpoint save failure and the other is delayed saving during Flux DreamBooth training; different code paths." + "reason": "Both involve training/save behavior, but one is a checkpoint save failure and the other is delayed model saving in Flux. Not one bug." }, { "left": "issue:7871", "right": "issue:8900", "accept": false, - "reason": "One is a blockwise-scale error with a specific LoRA, the other is about loading an SDXL refiner from_single_file; not the same issue." + "reason": "Blockwise scale handling in a LoRA file is unrelated to loading an SDXL refiner from_single_file." }, { "left": "issue:1398", "right": "issue:867", "accept": false, - "reason": "Both mention GPU-related behavior, but one is Apple Silicon/MPS support and the other is GPU memory not releasing in FastAPI." + "reason": "MPS/Apple Silicon compatibility and FastAPI GPU memory retention are unrelated runtime issues." }, { "left": "issue:3975", "right": "issue:7871", "accept": false, - "reason": "A typo in an error message is unrelated to a functional blockwise-scales LoRA error." + "reason": "A typo in an error message is unrelated to a LoRA/blockwise-scale loading failure." }, { "left": "issue:7296", "right": "issue:9902", "accept": false, - "reason": "PEFT/DoRA dependency pinning is unrelated to a Conda version pinning problem." + "reason": "PEFT dependency pinning and a stuck conda version are distinct packaging/versioning problems." }, { "left": "issue:3101", "right": "issue:4206", "accept": false, - "reason": "Both are ControlNet training failures, but one is a training example error and the other is a main-branch SDXL training regression; not enough to treat as one bug." + "reason": "Both mention ControlNet training, but one is a training example error and the other is a broken SDXL main-branch behavior." }, { "left": "issue:2371", "right": "issue:2465", "accept": false, - "reason": "Dreambooth appears in both, but one is missing generated objects in output while the other asks for a reproducible recipe blog; different concerns." + "reason": "Dreambooth output artifacts missing vs asking for a reproducible recipe are not duplicate bug reports." }, { "left": "issue:4566", "right": "issue:7350", "accept": false, - "reason": "Checkpoint resume for SDXL LoRA training is different from initializing ControlNet from a UNet with LoRA." + "reason": "Checkpoint resume in SDXL LoRA training is unrelated to initializing ControlNet from a UNet with LoRA." }, { "left": "issue:12299", @@ -43594,67 +43608,73 @@ "left": "issue:5028", "right": "issue:5799", "accept": false, - "reason": "Different API problems: an unexpected 'scale' keyword in a reference pipeline vs a 'strength' keyword question for inpainting." + "reason": "Different API/runtime errors in different pipelines: hacked DownBlock2D scale kwarg vs inpainting strength kwarg." }, { "left": "issue:10194", "right": "issue:11145", "accept": false, - "reason": "Both mention Flux, but one is about attention masking support and the other is about Flux Fill LoRA/dreambooth finetuning support; distinct feature requests." + "reason": "One is a Flux attention-masking feature request; the other is Flux Fill LoRA/DreamBooth support. Related theme, not same change." }, { "left": "issue:2264", "right": "issue:4306", "accept": false, - "reason": "A seed-setting error in InstructPix2Pix is unrelated to a list-index error in StableDiffusionReferencePipeline." + "reason": "InstructPix2Pix seed handling and ReferencePipeline list-index failure are different pipeline bugs." }, { "left": "issue:10972", "right": "issue:11062", "accept": false, - "reason": "Both are LoRA loading failures, but they involve different LoRA sources/models and lack evidence of the same underlying code-path bug." + "reason": "Both are LoRA loading failures, but for different model families and error modes; not clearly the same defect." }, { "left": "issue:2331", "right": "issue:6372", "accept": false, - "reason": "A general ControlNet pipeline request is not the same as an SVD+ControlNet feature request." + "reason": "Adding a ControlNet pipeline and SVD+ControlNet support are different feature requests." }, { "left": "issue:6093", "right": "issue:9204", "accept": false, - "reason": "These are about different training scripts and behaviors; one is push_to_hub repo creation, the other is a Flux DreamBooth training script request." + "reason": "Push-to-hub repo creation in one SDXL training script is unrelated to the FLUX.1 DreamBooth training script issue." }, { "left": "issue:4353", "right": "issue:6258", "accept": false, - "reason": "Different quality complaints in different models: SDXL image quality vs Stable Video Diffusion forward chunking degradation." + "reason": "Both involve image quality, but one is SDXL generation quality and the other is forward-chunking degradation in SVD." }, { "left": "issue:6092", "right": "issue:9044", "accept": false, - "reason": "A LoRA example not training is unrelated to a missing diffusion_pytorch_model.bin file when loading SDXL base." + "reason": "A LoRA example not training and a missing diffusion_pytorch_model.bin file are unrelated load/train problems." }, { "left": "issue:3174", "right": "issue:4852", "accept": false, - "reason": "Both involve LoRA loading, but one is a regression from new LoRA features breaking lora+dreambooth loading and the other is a generic load error with no matching failure details." + "reason": "Both mention LoRA loading, but one is a broader regression from new LoRA features while the other is a standalone load error. Not enough to merge." }, { "left": "issue:7716", "right": "issue:7936", "accept": false, - "reason": "Missing config.json and invalid JSON in a local safetensors load are different loader failures." + "reason": "Missing config.json in a directory and invalid JSON when loading a safetensors file are different file-format/path errors." }, { "left": "issue:7629", "right": "issue:9613", "accept": false, - "reason": "One is an SVD OOM issue on PyTorch 1.13.1, the other is a Timesteps.scale attribute error in a custom diffusion model on 0.30.x." + "reason": "OOM on PyTorch 1.13.1 and a missing scale attribute on Timesteps are unrelated compatibility bugs." + }, + { + "left": "issue:4375", + "right": "issue:4731", + "accept": false, + "reason": "Both involve torch.compile, but one is a general inference crash and the other is multi-ControlNet SDXL compile failure. Different code paths." } ] }, @@ -43673,8 +43693,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -43684,6 +43704,7 @@ "issue:10591", "issue:10797", "issue:10928", + "issue:1141", "issue:1486", "issue:1552", "issue:1720", @@ -43698,11 +43719,9 @@ "issue:3452", "issue:409", "issue:4125", - "issue:4375", "issue:4376", "issue:4427", "issue:4468", - "issue:4731", "issue:4773", "issue:4975", "issue:5125", @@ -43719,6 +43738,7 @@ "issue:6602", "issue:6632", "issue:6722", + "issue:699", "issue:7146", "issue:7267", "issue:7480", @@ -43727,7 +43747,6 @@ "issue:9803" ], "soft_pairs": [ - "issue:4375|issue:4731", "issue:1922|issue:5661", "issue:2210|issue:6722", "issue:10016|issue:6183", @@ -43750,186 +43769,190 @@ "issue:1720|issue:2442", "issue:1865|issue:1946", "issue:10591|issue:6632", - "issue:10928|issue:7267" + "issue:10928|issue:7267", + "issue:1141|issue:699" ], - "prepared_review_unit_hash": "71a2368542d2f60823ba91330ea7a1364fca986416b8e8b87a8bd3a3b1dd4bf6", + "prepared_review_unit_hash": "294acd5f937eb8e73fefe529554ec801070048f94330098cb8d056e2ae54679c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33354, - "estimated_input_tokens": 8339, - "estimated_eval_tokens": 16934 + "serialized_chars": 33332, + "estimated_input_tokens": 8333, + "estimated_eval_tokens": 16922 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated Diffusers issues that only share broad themes like loading, LoRA, SDXL, or inference failures. I do not see a single underlying bug/change that ties these items together, so the soft links should be rejected.", - "confidence": 0.91, + "summary": "The cluster is mostly heterogeneous. Most pairings are clearly different bugs or feature requests from the titles alone. The only near-match is the two `from_single_file` shape-mismatch reports, but with title-only evidence and different models/pipelines, it should stay uncertain rather than accepted.", + "confidence": 0.81, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No true duplicate canonical issue stands out; if forced to pick a representative, issue 6183 is the closest because it is a concrete SDXL `from_single_file` shape-mismatch report, but the cluster as a whole is too heterogeneous to treat as one bug.", + "best_issue_reason": "issue:6183 is the clearest representative of the `from_single_file` shape-mismatch theme, but the packet does not justify treating it as a confirmed duplicate canonical.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:4375", - "right": "issue:4731", - "accept": false, - "reason": "Both are Torch compile/inference failures, but one is generic crash during inference and the other is a multi-ControlNet SDXL-specific compile failure; not the same concrete bug." - }, { "left": "issue:1922", "right": "issue:5661", "accept": false, - "reason": "Both mention notebook/API breakage, but one is an undefined helper in a training example and the other is a missing `PreTrainedTokenizer` symbol in a release; different issues." + "reason": "Both involve missing names/imports, but they are different symbols in different contexts: notebook helper vs tokenizer typing/import issue." }, { "left": "issue:2210", "right": "issue:6722", "accept": false, - "reason": "Both are feature requests around text encoders, not the same defect or change request." + "reason": "Both mention text encoders, but one is about training LoRA on the text encoder and the other is about using SigLIP as the text encoder." }, { "left": "issue:10016", "right": "issue:6183", "accept": false, - "reason": "Both involve SDXL single-file loading and shape mismatches, but the exact models and failure modes differ; not enough to treat as one bug." + "reason": "Close on `from_single_file` and shape mismatch, but the packet only gives titles and the pipelines/models differ (SD3.5M vs SDXL), so this is too uncertain to merge as a duplicate." }, { "left": "issue:1552", "right": "issue:5447", "accept": false, - "reason": "Both concern scheduler/pipeline behavior, but one is attention slicing on SD2 and the other is denoising_start with 2nd-order schedulers on SDXL img2img; different code paths." + "reason": "Different failure modes in different pipeline/scheduler paths: attention slicing vs `denoising_start` with 2nd-order schedulers." }, { "left": "issue:4125", "right": "issue:6362", "accept": false, - "reason": "Both are training-loss or fine-tuning discussions, but one is an expected loss curve question and the other is dataset mapping OOM on SDXL; unrelated." + "reason": "One is about whether a loss curve looks expected; the other is an OOM during dataset mapping. Different problems." }, { "left": "issue:3296", "right": "issue:5125", "accept": false, - "reason": "One is checkpoint save/load with text encoder during training; the other is an XFormers CUDA error during attention processing. Different failures." + "reason": "Checkpoint save/load with text encoder is unrelated to an XFormers CUDA invalid-configuration error." }, { "left": "issue:10428", "right": "issue:10470", "accept": false, - "reason": "Both are Flux inference issues, but one is on Ascend NPU and the other is torchao inference not working; different backends and likely different bugs." + "reason": "Both are Flux-related, but Ascend NPU inference and torchao inference are different environments and code paths." }, { "left": "issue:4376", "right": "issue:6527", "accept": false, - "reason": "Both mention pipeline/scheduler problems, but one is textual inversion on SDXL and the other is Euler add_noise behavior; not the same issue." + "reason": "StableDiffusionXL textual inversion crash and EulerDiscreteScheduler `add_noise` bug are different subsystems." }, { "left": "issue:1486", "right": "issue:4468", "accept": false, - "reason": "A scheduler bug versus a multi-GPU performance discrepancy; these are not the same underlying problem." + "reason": "DDIMScheduler prediction type behavior is unrelated to multi-GPU vs single-GPU sample-code performance." }, { "left": "issue:4773", "right": "issue:4975", "accept": false, - "reason": "Both are LoRA-related, but one is loading CivitAI LoRA and the other is a loading inefficiency caused by repeated datatype queries; different concerns." + "reason": "Both are about LoRA loading, but one is a CivitAI loading issue and the other is inefficient datatype querying." }, { "left": "issue:10470", "right": "issue:10797", "accept": false, - "reason": "Both mention inference, but one is Flux torchao inference and the other is HunyuanVideo group offloading; different models and mechanisms." + "reason": "Flux torchao inference and HunyuanVideo group offloading affect different pipelines and failure modes." }, { "left": "issue:6385", "right": "issue:7146", "accept": false, - "reason": "Both are memory-related, but one is StableVideoDiffusionPipeline VRAM usage and the other is SDXL fine-tuning OOM; not a duplicate." + "reason": "Both are memory complaints, but StableVideoDiffusionPipeline VRAM usage and SDXL fine-tuning OOM are separate issues." }, { "left": "issue:1925", "right": "issue:4427", "accept": false, - "reason": "Both are example/model usage failures, but one is SD2 inpainting half-precision and the other is a T2I multi-adapter example; different code paths." + "reason": "Inpainting half-precision failure and T2I Multiadapter example breakage are different features and bugs." }, { "left": "issue:3224", "right": "issue:3452", "accept": false, - "reason": "Learning-rate scaling duplicate in a LoRA script versus a generic DreamBooth training-on-low-VRAM question; no same bug." + "reason": "Learning-rate scaling duplication is unrelated to low-VRAM Dreambooth training." }, { "left": "issue:409", "right": "issue:785", "accept": false, - "reason": "Both involve attention/scheduler internals, but the titles point to different API eras and different performance/feature issues; not the same defect." + "reason": "Missing `enable_attention_slicing` vs PNDMScheduler performance cost are not the same defect." }, { "left": "issue:6176", "right": "issue:6602", "accept": false, - "reason": "Both are LoRA save/load errors, but one is loading LoRA after dreambooth_lora training and the other is loading a pretrained model after fuse_lora/save_pretrained; different workflows." + "reason": "Both involve LoRA/save-load workflows, but they hit different code paths: loading after Dreambooth-LoRA training vs loading after `fuse_lora()`/`save_pretrained()`." }, { "left": "issue:6280", "right": "issue:7784", "accept": false, - "reason": "One is about a missing `from_single_file` method on a ControlNet pipeline; the other is a missing `upcast_vae` attribute on an upscale pipeline. Different API gaps." + "reason": "Missing `from_single_file` on one pipeline and missing `upcast_vae` on another are unrelated API issues." }, { "left": "issue:6382", "right": "issue:7480", "accept": false, - "reason": "Both are training/memory-related, but one is an IP-adapter key error after LCM LoRA training and the other is DeepFloyd IF OOM; unrelated." + "reason": "IP-adapter key error after LCM LoRA training and DeepFloyd IF OOM are unrelated." }, { "left": "issue:5606", "right": "issue:9803", "accept": false, - "reason": "Both are LoRA loading failures, but one is no alpha value handling and the other is Flux LoRA loading failure; not enough evidence of the same bug." + "reason": "Both are LoRA loading failures, but one is about missing alpha handling and the other is a Flux-specific loading failure." }, { "left": "issue:1720", "right": "issue:2442", "accept": false, - "reason": "M1 Pro finetuning errors versus an inpainting-model loss-graph question; different topics and no shared defect." + "reason": "M1 finetuning errors and an expected-loss-curve question for inpainting are different topics." }, { "left": "issue:1865", "right": "issue:1946", "accept": false, - "reason": "A Zero-Stage-3 finetuning dimension error is unrelated to an xformers CUDA invalid-argument error." + "reason": "Zero-stage-3 finetuning dimension errors and XFormers CUDA invalid-argument errors come from different subsystems." }, { "left": "issue:10591", "right": "issue:6632", "accept": false, - "reason": "One is about a wrong SD3 LoRA training script, the other asks whether `save_state` is necessary; different problems." + "reason": "One reports a problem in an SD3 LoRA training script; the other asks whether `save_state` is necessary. Not a duplicate bug." }, { "left": "issue:10928", "right": "issue:7267", "accept": false, - "reason": "Both involve loading data or model files, but one is missing `diffusion_pytorch_model.fp16.bin` in a UNet directory and the other is broken local data loading in train_controlnet.py; different failures." + "reason": "Missing a specific UNet weight file and broken local dataset loading in `train_controlnet.py` are different file/loading failures." + }, + { + "left": "issue:1141", + "right": "issue:699", + "accept": false, + "reason": "Local file support in custom pipelines and negative prompts are unrelated feature requests." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "Most of the negative pairwise judgments look conservative and are supported by the titles. However, the one accepted duplicate pair is a bit too strong for the evidence in the packet: issue:10016 and issue:6183 both involve `from_single_file` and shape mismatches, but the packet only gives titles, and the models/pipelines differ (SD3.5M vs SDXL) with no body-level confirmation that this is the same bug. I\u2019d keep that pair soft/uncertain rather than accepted." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -43942,17 +43965,18 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10016", "issue:10315", "issue:10526", + "issue:10866", "issue:11178", - "issue:1141", "issue:11702", + "issue:12047", "issue:13107", "issue:1434", "issue:1823", @@ -43976,7 +44000,6 @@ "issue:6443", "issue:6812", "issue:6821", - "issue:699", "issue:7079", "issue:7133", "issue:7228", @@ -43996,7 +44019,6 @@ "issue:9914" ], "soft_pairs": [ - "issue:1141|issue:699", "issue:1823|issue:547", "issue:228|issue:547", "issue:3379|issue:945", @@ -44019,180 +44041,181 @@ "issue:10526|issue:9514", "issue:10016|issue:11178", "issue:9767|issue:9914", - "issue:10315|issue:8477" + "issue:10315|issue:8477", + "issue:10866|issue:12047" ], - "prepared_review_unit_hash": "a80468f76c2c42c047ac62383d34c56823d87a5135be69d49bc2bf6c5df0480e", + "prepared_review_unit_hash": "1d405cb13478283e2f32598d4bd739ab6a373c78e1c8b8adff640d75d35e3ce6", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33470, - "estimated_input_tokens": 8368, - "estimated_eval_tokens": 16992 + "serialized_chars": 33509, + "estimated_input_tokens": 8378, + "estimated_eval_tokens": 17012 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The candidate pairs are mostly title-similarity matches across different models, scripts, or failure modes; none look like the same concrete bug/change, so all soft edges should be rejected.", - "confidence": 0.89, - "canonical_issue_reason": null, + "summary": "No soft pair looks like the same underlying bug/change. The set is a grab bag of unrelated loading, training, OOM, scheduler, and LoRA issues that only share broad diffusers/modeling vocabulary.", + "confidence": 0.97, + "canonical_issue_reason": "issue:10016 is the best representative only in the sense that it is a precise, concrete bug report with a clear failure mode and reproducible load path; the cluster as a whole is not a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10016 is the clearest standalone bug report in the set, but the items are too heterogeneous to form a true canonical issue.", + "best_issue_reason": "issue:10016 has the clearest scoped symptom and strongest bug-report quality among the issues, making it the least-bad representative if one must be chosen.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:1141", - "right": "issue:699", - "accept": false, - "reason": "Local-file/custom-pipeline support and negative prompts are unrelated features; no shared bug path." - }, { "left": "issue:1823", "right": "issue:547", "accept": false, - "reason": "Swapped comments in depth2img and docs for local loading are different kinds of issues." + "reason": "Swapped comments in depth2img is unrelated to documentation about loading stable diffusion locally." }, { "left": "issue:228", "right": "issue:547", "accept": false, - "reason": "Broken release link is unrelated to local-file loading documentation." + "reason": "Broken model link is a release artifact problem, not a documentation issue." }, { "left": "issue:3379", "right": "issue:945", "accept": false, - "reason": "Dreambooth checkpoint loading and ONNX conversion are different code paths and problems." + "reason": "Dreambooth checkpoint loading and ONNX custom checkpoint conversion are different code paths and failures." }, { "left": "issue:3452", "right": "issue:3658", "accept": false, - "reason": "Both mention training, but 4GB VRAM/OOM concerns differ from the upsampler noise-latent bug." + "reason": "Low-VRAM Dreambooth training request is not the same as the additional-noise-latents bug in upsampler scripts." }, { "left": "issue:4080", "right": "issue:8482", "accept": false, - "reason": "SDXL inpainting support and SD3-medium support are unrelated model features." + "reason": "Adding SDXL inpainting support is a feature request, not the same as SD3.5 support." }, { "left": "issue:434", "right": "issue:435", "accept": false, - "reason": "Both are notebook breakages, but they involve different notebooks and different errors." + "reason": "Training notebook breakage and AutoencoderKLOutput.sample error are related only by notebook context, not the same bug." }, { "left": "issue:4724", "right": "issue:7695", "accept": false, - "reason": "SDXL img2img checkpoint behavior and ip_adapter mask warnings in attention processors are different issues." + "reason": "SDXL img2img behavior and ip_adapter_masks warnings in attention are different problems." }, { "left": "issue:5515", "right": "issue:8464", "accept": false, - "reason": "LCM scheduler initialization and SEGA CPU-offload seed failures are not the same bug." + "reason": "LCM scheduler initialization failure and SEGA seed/offload failure are different pipeline bugs." }, { "left": "issue:6812", "right": "issue:6821", "accept": false, - "reason": "A LoRA dtype mismatch and an invalid caption_column argument are unrelated training errors." + "reason": "LoRA Q/K/V dtype mismatch and caption_column validation error are unrelated training-script issues." }, { "left": "issue:7133", "right": "issue:9358", "accept": false, - "reason": "SDXL DreamBooth resume size mismatch and Flux text-encoder reinitialization are different failure modes." + "reason": "Checkpoint resume size mismatch in SDXL DreamBooth LoRA is not the same as redundant text-encoder reinit in Flux LoRA training." }, { "left": "issue:6092", "right": "issue:9164", "accept": false, - "reason": "Both are LoRA training complaints, but they target different example scripts/models and different symptoms." + "reason": "LoRA example not training and dog example not converging are both training complaints but not the same concrete defect." }, { "left": "issue:5044", "right": "issue:8874", "accept": false, - "reason": "Kandinsky load failure and CKPT unpickling failure in from_single_file are distinct loading problems." + "reason": "Kandinsky 2.2 load failure and CKPT unpickling failure are different load mechanisms and errors." }, { "left": "issue:1434", "right": "issue:7480", "accept": false, - "reason": "OOM in StableDiffusionUpscalePipeline and OOM in DeepFloyd IF are separate pipelines and resource issues." + "reason": "Upscaling OOM in StableDiffusionUpscalePipeline is unrelated to DeepFloyd IF memory usage." }, { "left": "issue:2163", "right": "issue:4316", "accept": false, - "reason": "UNet dtype mismatch in text-to-image training and a DreamBooth validation error are different training-script bugs." + "reason": "UNet dtype mismatch in text-to-image training is not the same as validation-option failure in DreamBooth." }, { "left": "issue:11702", "right": "issue:13107", "accept": false, - "reason": "Both are LoRA key errors, but they involve different models and load paths, so not the same underlying bug." + "reason": "AccVid LoRA key errors and Flux Klein LoRA missing keys are different model/LoRA compatibility issues." }, { "left": "issue:2696", "right": "issue:3223", "accept": false, - "reason": "CUDA-not-enabled installation error and GPU memory not clearing are unrelated." + "reason": "Torch CUDA not compiled and RAM not cleared after moving pipeline to cuda are unrelated environment/runtime issues." }, { "left": "issue:7079", "right": "issue:7228", "accept": false, - "reason": "Transformer2DModel patch/norm-type error and traced-model missing dtype attribute are different code-path failures." + "reason": "Transformer2DModel patch/norm error and traced-model missing dtype are distinct model API bugs." }, { "left": "issue:4446", "right": "issue:8589", "accept": false, - "reason": "Custom text_encoder loading and SD3 DreamBooth prior-preservation training are not the same issue." + "reason": "Custom text_encoder loading in from_single_file is not the same as SD3 DreamBooth prior-preservation training." }, { "left": "issue:6443", "right": "issue:7354", "accept": false, - "reason": "Scheduler incompatibility in SAGPipeline and PNDMScheduler/DDPMPipeline mismatch are related only at a broad level, not the same concrete bug." + "reason": "SAGPipeline scheduler compatibility and PNDMScheduler incompatibility with DDPMPipeline are different scheduler/pipeline mismatches." }, { "left": "issue:10526", "right": "issue:9514", "accept": false, - "reason": "FP8 construction TypeError in optimum.quanto and post-quantization LoRA loading failure in Flux-dev are different problems." + "reason": "optimum.quanto FP8 tensor construction error and LoRA loading after fp8 quantisation are related only broadly to quantization." }, { "left": "issue:10016", "right": "issue:11178", "accept": false, - "reason": "Both are from_single_file load failures, but they affect different model families and different shape-mismatch contexts." + "reason": "Both are from_single_file loading errors, but one is SD3.5M pos_embed shape mismatch and the other is MistoLine ControlNet loading; different models and failure modes." }, { "left": "issue:9767", "right": "issue:9914", "accept": false, - "reason": "Flux img2img cuDNN frontend runtime error and Xlabs LoRA loading error are different failures." + "reason": "FluxImg2Img cuDNN frontend error and Xlabs-trained LoRA loading fix proposal are not the same underlying bug." }, { "left": "issue:10315", "right": "issue:8477", "accept": false, - "reason": "CogVideo training error and a distributed consistency-training script error are unrelated." + "reason": "CogVideo training error and distributed consistency-script error are unrelated training failures." + }, + { + "left": "issue:10866", + "right": "issue:12047", + "accept": false, + "reason": "Lumina Image 2.0 LoRA incompatibility and Wan2.2 GGUF LoRA fusion failure are different model-family issues." } ] }, @@ -44211,14 +44234,14 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10231", "issue:1037", "issue:10639", - "issue:10866", "issue:11136", "issue:1141", "issue:12047", @@ -44258,6 +44281,7 @@ "issue:771", "issue:7931", "issue:8874", + "issue:8885", "issue:8939", "issue:9052", "issue:9558", @@ -44265,7 +44289,6 @@ "issue:9873" ], "soft_pairs": [ - "issue:10866|issue:12047", "issue:4236|issue:771", "issue:1407|issue:3984", "issue:4398|issue:6492", @@ -44288,180 +44311,181 @@ "issue:1537|issue:4134", "issue:1541|issue:4488", "issue:11136|issue:13079", - "issue:1974|issue:5447" + "issue:1974|issue:5447", + "issue:10231|issue:8885" ], - "prepared_review_unit_hash": "c24c316be13cfdf57dc16f621263e6edc230fe0ec16efe7a8eeb840ca8fc03de", + "prepared_review_unit_hash": "6ebdcacae71824e7bb291e9e25f334f14a002b52a213f814f692497948484cc9", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33998, - "estimated_input_tokens": 8500, - "estimated_eval_tokens": 17256 + "serialized_chars": 34596, + "estimated_input_tokens": 8649, + "estimated_eval_tokens": 17554 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster does not form a true duplicate set. The items are mostly unrelated pipeline/model-specific bugs, feature requests, and training issues; the suggested soft pairs only share broad vocabulary like \u201cblack images,\u201d \u201cfrom_single_file,\u201d or \u201csupport,\u201d not the same concrete failure mode.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "This looks like a false-positive cluster: the items span unrelated issues across tutorials, training scripts, pipeline support, model loading, and device/runtime errors. None of the soft pairs appear to be the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits this set; the items are thematically broad but not duplicate-convergent, so the cluster should be split rather than anchored to one issue.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "None of the issues is a strong global canonical for the whole set because they address different code paths and product areas; the safest triage is no global issue canonical.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:10866", - "right": "issue:12047", - "accept": false, - "reason": "Different problems: Lumina Image 2.0 LoRA compatibility vs Wan2.2 GGUF LoRA fusing failure." - }, { "left": "issue:4236", "right": "issue:771", "accept": false, - "reason": "Different pipeline bugs: SDXL refiner text-encoder registration vs BF16 + Dreambooth training." + "reason": "Both mention SD/model incompatibility, but one is an SDXL refiner registration issue and the other is a BF16+Dreambooth problem; different bugs." }, { "left": "issue:1407", "right": "issue:3984", "accept": false, - "reason": "Unrelated errors: ONNX feature_extractor regression vs class_labels requirement for class-conditioned models." + "reason": "One is an ONNX pipeline missing feature_extractor, the other is a class_labels validation error; unrelated failure modes." }, { "left": "issue:4398", "right": "issue:6492", "accept": false, - "reason": "Both mention black output, but one is SDXL LORA-dreambooth and the other is generic image generation; no shared code path." + "reason": "Both describe black images, but one is an SDXL LoRA-dreambooth pipeline and the other is generic image generation; too broad to treat as the same bug." }, { "left": "issue:2107", "right": "issue:6815", "accept": false, - "reason": "Both involve black images, but one is low-resolution rendering artifacts and the other is tensorboard validation display casting warnings." + "reason": "Both involve bad low-resolution/black outputs, but one is about sub-512 image resolution glitches and the other is a tensorboard casting warning; not the same code path." }, { "left": "issue:2444", "right": "issue:4639", "accept": false, - "reason": "Different training-script failures: Dreambooth inpaint script error vs multiprocessing data-loader error in a different SDXL script." + "reason": "Both are training-script errors, but one is dreambooth inpaint subprocess failure and the other is SDXL instruct-pix2pix dataloader multiprocessing; different causes." }, { "left": "issue:1037", "right": "issue:1141", "accept": false, - "reason": "Different feature requests: adding a community pipeline vs adding local-file support in custom pipelines." + "reason": "A community pipeline request and local-file support for custom pipelines are different feature requests." }, { "left": "issue:10639", "right": "issue:9873", "accept": false, - "reason": "Different model additions: Lumina-Image 2.0 vs OmniGen; not the same pipeline or bug." + "reason": "Both are model/pipeline additions, but Lumina-Image 2.0 and OmniGen are distinct model integrations." }, { "left": "issue:2156", "right": "issue:2465", "accept": false, - "reason": "Different topics: Dreambooth example uses CPU instead of GPU vs asking for a reproducible training recipe." + "reason": "One reports CPU-only Dreambooth training; the other asks for a reproducible recipe. Same topic, not the same bug." }, { "left": "issue:2781", "right": "issue:5919", "accept": false, - "reason": "Different distributed/LoRA issues: gloo CUDA device attribute error vs missing set_lora_layer on Linear." + "reason": "Distributed training _cuda_setDevice error and LoRA layer attribute error are unrelated runtime issues." }, { "left": "issue:5557", "right": "issue:9729", "accept": false, - "reason": "Unrelated errors: unexpected use_reentrant kwarg vs Flux LoRA save-model type error." + "reason": "Both mention unexpected arguments, but one is use_reentrant and the other is save_model on T5EncoderModel; different API mismatches." }, { "left": "issue:7418", "right": "issue:8939", "accept": false, - "reason": "Different failures: a bug in a community upscale LDM3D pipeline file vs tokenizer loading for SD3 medium." + "reason": "The first is a bug in a community upscale script; the second is a tokenizer loading failure for SD3. Different subsystems." }, { "left": "issue:6832", "right": "issue:7931", "accept": false, - "reason": "Different IP-Adapter/ControlNet issues: tuple.shape error vs ip_adapter_image_embeds incompatibility." + "reason": "Both involve IP-Adapter-like image handling, but one is a tuple/shape crash and the other is ControlNet inpaint with ip_adapter_image_embeds; not the same defect." }, { "left": "issue:12047", "right": "issue:12329", "accept": false, - "reason": "Different Wan2.2 bugs: LoRA fusion failure vs single-file transformer loading as Wan2.1." + "reason": "Both concern Wan model variants, but one is LoRA fusion failing on Wan2.2 GGUF and the other is incorrect transformer version selection; separate bugs." }, { "left": "issue:5520", "right": "issue:7591", "accept": false, - "reason": "Both mention HeaderTooLarge, but one is deserialization generally and the other is local_files_only on from_single_file; insufficient evidence of same bug." + "reason": "Both mention HeaderTooLarge, but one is generic deserialization and the other is from_single_file with local_files_only; not enough to merge." }, { "left": "issue:5619", "right": "issue:9558", "accept": false, - "reason": "Different black-image contexts: LCM img2img AutoencodeTiny failure vs tensorboard validation display." + "reason": "Both end up with black images, but one is an AutoencodeTiny LCM img2img encode bug and the other is tensorboard validation display; different failure points." }, { "left": "issue:8874", "right": "issue:9052", "accept": false, - "reason": "Related single-file loading topic, but one is CKPT weights-only unpickling failure and the other is failures on certain Civitai checkpoints; likely different root causes." + "reason": "Both are from_single_file checkpoint-loading problems, but one is a PyTorch unpickling/weights-only failure and the other is civit checkpoint compatibility; likely different root causes." }, { "left": "issue:6870", "right": "issue:7492", "accept": false, - "reason": "Different problems: missing enable_vae_tiling in SD img2img/inpaint vs model repository not found for t2iadapter." + "reason": "One is missing enable_vae_tiling on img2img/inpaint, the other is a repository lookup failure for a T2I adapter URL; unrelated." }, { "left": "issue:3182", "right": "issue:3420", "accept": false, - "reason": "Both are custom/community pipeline issues, but one is CPU offload incompatibility and the other is a ControlNet inpaint pipeline mismatch." + "reason": "Both are pipeline incompatibilities, but one is CPU offload with a custom LPW pipeline and the other is a ControlNet inpaint pipeline vs diffusion pipeline mismatch." }, { "left": "issue:6230", "right": "issue:6791", "accept": false, - "reason": "Different training script failures: CUDA OOM in text-to-image SDXL training vs cannot run the same script at all." + "reason": "Both are SDXL training script problems, but one is CUDA OOM and the other is a script that cannot run at all; not the same bug." }, { "left": "issue:1537", "right": "issue:4134", "accept": false, - "reason": "Both mention MPS, but one is scheduler support on Apple MPS and the other is a ControlNet tensor-shape mismatch on M1 Max." + "reason": "Apple MPS scheduler behavior and ControlNet tensor-size mismatch are different device/model issues." }, { "left": "issue:1541", "right": "issue:4488", "accept": false, - "reason": "Different loading failures: Dreambooth cannot determine model type from CKPT vs SDXL checkpoint loading without pipeline_class." + "reason": "Dreambooth CKPT model-type detection and SDXL checkpoint loading without pipeline_class are separate loading problems." }, { "left": "issue:11136", "right": "issue:13079", "accept": false, - "reason": "Different resource/feature issues: SanaSprintPipeline timestep support vs FLUX.2-klein CUDA OOM in a loop." + "reason": "A SanaSprint timestep limitation and a FLUX.2-klein CUDA OOM loop issue are unrelated." }, { "left": "issue:1974", "right": "issue:5447", "accept": false, - "reason": "Different scheduler bugs: Euler reproducibility with generator lists vs SDXL img2img denoising_start on second-order schedulers." + "reason": "Euler generator-list reproducibility and SDXL img2img denoising_start with 2nd-order schedulers do not describe the same bug." + }, + { + "left": "issue:10231", + "right": "issue:8885", + "accept": false, + "reason": "Ubuntu tutorial ImportError and custom diffusion dependency/token loading issues are unrelated." } ] }, @@ -44485,7 +44509,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10231", "issue:10281", "issue:1049", "issue:11088", @@ -44511,10 +44534,12 @@ "issue:457", "issue:4892", "issue:4974", + "issue:5004", "issue:5475", "issue:5605", "issue:5628", "issue:5873", + "issue:6093", "issue:6232", "issue:660", "issue:6928", @@ -44524,7 +44549,7 @@ "issue:7168", "issue:7752", "issue:7882", - "issue:8885", + "issue:7978", "issue:9161", "issue:9186", "issue:9227", @@ -44534,7 +44559,6 @@ "issue:9942" ], "soft_pairs": [ - "issue:10231|issue:8885", "issue:3740|issue:4213", "issue:11088|issue:12047", "issue:1049|issue:2747", @@ -44557,180 +44581,188 @@ "issue:7080|issue:7142", "issue:9576|issue:9673", "issue:5628|issue:6232", - "issue:2785|issue:6928" + "issue:2785|issue:6928", + "issue:5004|issue:6093", + "issue:4213|issue:7978" ], - "prepared_review_unit_hash": "6fc31cbb82642ca849d6d5d5f8d93b73c095d109ef5fc49c05ace673abddbb12", + "prepared_review_unit_hash": "41e3de3109a8faa1f4108017684a9728dc826fe6acc47a344858c8ea39e7cddc", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 33613, - "estimated_input_tokens": 8404, - "estimated_eval_tokens": 17064 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34418, + "estimated_input_tokens": 8605, + "estimated_eval_tokens": 17466 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: the items span unrelated bugs, feature requests, docs, and pipeline/model-specific issues. The soft pairs only share superficial wording (e.g., \"support\", \"pipeline\", \"load\", \"convert\") and do not point to the same underlying defect or change.", - "confidence": 0.97, - "canonical_issue_reason": null, - "canonical_pr_reason": null, - "best_issue_reason": null, - "best_pr_reason": null, + "summary": "This cluster is a loose grab-bag of unrelated issues and feature requests; the soft edges are driven by generic title overlap, not shared bugs or mergeable changes. No pair looks like a true duplicate.", + "confidence": 0.99, + "canonical_issue_reason": "No single canonical issue: the items span unrelated schedulers, pipeline requests, training-script bugs, conversion problems, and backend/device-specific failures.", + "canonical_pr_reason": "No PRs are present in this cluster.", + "best_issue_reason": "No strong global best issue; if forced, issue:12574 is the most concrete bug report, but it does not represent the rest of the cluster.", + "best_pr_reason": "No PRs are present in this cluster.", "soft_edge_verdicts": [ - { - "left": "issue:10231", - "right": "issue:8885", - "accept": false, - "reason": "Different problems: tutorial import error on Ubuntu vs custom pipeline local-file/modifier-token install/load issues." - }, { "left": "issue:3740", "right": "issue:4213", "accept": false, - "reason": "Both are loading/conversion-related, but one is legacy attention conversion with accelerate device_map, the other is SDXL from_single_file without accelerate." + "reason": "Different problems: legacy attention conversion with device_map=sequential vs SDXL from_single_file loading without accelerate." }, { "left": "issue:11088", "right": "issue:12047", "accept": false, - "reason": "Different model paths and failures: wan2.1 GGUF load error vs Wan2.2 GGUF LoRA fusion failure." + "reason": "Both mention Wan/GGUF, but one is a load error and the other is LoRA fusion failure on a different model/version and code path." }, { "left": "issue:1049", "right": "issue:2747", "accept": false, - "reason": "Docs issues only; one is a specific Euler scheduler docs update, the other is a vague generic docs request." + "reason": "Both are docs-related, but one is specifically Euler scheduler docs and the other is a generic docs update request." }, { "left": "issue:1141", "right": "issue:5605", "accept": false, - "reason": "Custom pipeline local-file support is unrelated to SDXL Dreambooth per-image caption support." + "reason": "Unrelated feature requests: local files in custom pipelines vs per-image caption support for SDXL dreambooth." }, { "left": "issue:1614", "right": "issue:4321", "accept": false, - "reason": "Both produce bad images, but on different backends and symptoms: autocast black images vs DirectML FP16 black images in ONNX." + "reason": "Both show black images, but on different backends and causes: autocast on SD 2.1 vs ONNX FP16 DirectML." }, { "left": "issue:4002", "right": "issue:7882", "accept": false, - "reason": "Both are missing from_single_file, but for different pipelines (StableDiffusionInpaintPipeline vs t2i-adapter pipeline)." + "reason": "Both mention from_single_file, but they target different pipeline classes and likely different registration gaps." }, { "left": "issue:5873", "right": "issue:7053", "accept": false, - "reason": "Unrelated: ONNX export failure for scaled_dot_product_attention vs UNet3DConditionModel class-conditioning bug." + "reason": "Completely different bugs: ONNX export of scaled_dot_product_attention vs UNet3DConditionModel class conditioning." }, { "left": "issue:12547", "right": "issue:12574", "accept": false, - "reason": "Both mention training scripts, but one is a missing CLI arg for Dreambooth Flux I2I while the other is a pipeline bug in WanVideoToVideoPipeline/VACE control args." + "reason": "Different training scripts, models, and missing arguments; not the same underlying failure mode." }, { "left": "issue:7168", "right": "issue:9448", "accept": false, - "reason": "Both involve adapter image tensors, but the actual failures differ: getting 4D IP-Adapter embeds vs tuple.shape AttributeError in ControlNet inpaint pipeline." + "reason": "Both involve tuple/shape errors around IP-Adapter-style inputs, but they are different pipelines and different code paths." }, { "left": "issue:4892", "right": "issue:5475", "accept": false, - "reason": "Different video-generation features: TokenFlow pipeline vs LaVie pipeline proposal." + "reason": "Different pipeline support requests: TokenFlow video editing vs LaVie video generation." }, { "left": "issue:9161", "right": "issue:9227", "accept": false, - "reason": "Both are Flux training issues, but one is distributed bugs in the script and the other is a model-specific FLUX.1-dev run problem." + "reason": "Same script name, but one reports distributed training bugs while the other is a model-specific run issue; not a single concrete bug." }, { "left": "issue:1730", "right": "issue:7752", "accept": false, - "reason": "Completely different topics: model conversion to CKPT/webui compatibility vs a training hyperparameter question." + "reason": "Unrelated topics: model conversion to webui vs a training hyperparameter question." }, { "left": "issue:10281", "right": "issue:11135", "accept": false, - "reason": "Both are feature requests, but FreeScale scheduler and SmoothCache new pipeline are unrelated changes." + "reason": "Different requests entirely: a new diffusion scheduler vs a new pipeline proposal." }, { "left": "issue:12722", "right": "issue:13079", "accept": false, - "reason": "Different user scenarios: Kaggle GPU usage question vs FLUX.2-klein loop-induced CUDA OOM." + "reason": "Unrelated: Kaggle usage question for qwen-image vs CUDA OOM in FLUX.2-klein." }, { "left": "issue:3045", "right": "issue:4974", "accept": false, - "reason": "Both are about LoRA/model conversion, but one is converting safetensors to diffusers and the other is converting a safetensor model generally; not the same concrete request." + "reason": "Both concern conversion to diffusers, but one is LoRA conversion and the other is general model conversion." }, { "left": "issue:7882", "right": "issue:9186", "accept": false, - "reason": "Both mention adapters, but one is missing from_single_file for T2I Adapter while the other is a Differential Diffusion + ControlNet pipeline feature request." + "reason": "Different feature areas: missing from_single_file for t2i adapter vs differential diffusion with ControlNet." }, { "left": "issue:457", "right": "issue:660", "accept": false, - "reason": "Different MPS bugs: blue artifacts on MPS vs fp16/revision loading on M1." + "reason": "Both are MPS-related, but one reports blue artifacts and the other is about running fp16 on M1; not the same bug." }, { "left": "issue:11978", "right": "issue:13232", "accept": false, - "reason": "A new Bria 3.2 pipeline request is unrelated to LTX-2.3 support." + "reason": "Different model-support requests: Bria 3.2 vs LTX-2.3." }, { "left": "issue:12149", "right": "issue:9942", "accept": false, - "reason": "Both are installation/version complaints, but one is a source-vs-pip alignment issue for diffusers 0.34.0 and the other is inability to install diffusers>=0.32.0dev." + "reason": "Both are packaging/version issues, but one is source-code mismatch after install and the other is install failure." }, { "left": "issue:7080", "right": "issue:7142", "accept": false, - "reason": "Different inference/training mismatches: SDXL control net training vs wandb output drift, not the same bug." + "reason": "Different model/pipeline contexts: SDXL controlnet training inference mismatch vs SDXL-lightning inference mismatch." }, { "left": "issue:9576", "right": "issue:9673", "accept": false, - "reason": "Both are loading-related, but one is use_safetensors propagation to submodules and the other is high CPU usage when loading multiple LoRAs." + "reason": "Different loading bugs: safetensors propagation to submodules vs high CPU usage when loading multiple LoRAs." }, { "left": "issue:5628", "right": "issue:6232", "accept": false, - "reason": "Both touch schedulers, but LMS Karras denoising offset and DPMSolverSinglestep add_noise failures are different code paths." + "reason": "Scheduler bugs, but in different schedulers and different behaviors; not one shared defect." }, { "left": "issue:2785", "right": "issue:6928", "accept": false, - "reason": "Both are pipeline-related feature requests, but M1/M2 support for text-to-video is unrelated to disabling depthmap colorization in Marigold." + "reason": "Unrelated feature requests: M1/M2 support for text-to-video vs disabling depthmap colorization in Marigold." + }, + { + "left": "issue:5004", + "right": "issue:6093", + "accept": false, + "reason": "Different SDXL training-script issues: checkpoint resume at fp16 vs push_to_hub repo creation behavior." + }, + { + "left": "issue:4213", + "right": "issue:7978", + "accept": false, + "reason": "Completely unrelated issues: from_single_file with SDXL vs missing torchvision::nms." } ] }, @@ -44749,19 +44781,22 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:1037", "issue:10447", + "issue:10550", + "issue:10594", "issue:10650", "issue:1125", "issue:12396", "issue:12459", "issue:12533", "issue:12536", + "issue:1434", "issue:1811", "issue:1931", "issue:2107", @@ -44770,27 +44805,24 @@ "issue:3182", "issue:4104", "issue:4185", - "issue:4213", "issue:4368", "issue:4402", "issue:4476", "issue:4605", "issue:4726", "issue:4806", - "issue:5004", + "issue:5215", "issue:5399", "issue:5509", "issue:5630", "issue:5719", "issue:5897", - "issue:6093", "issue:6164", "issue:6362", "issue:7003", "issue:7267", "issue:7354", "issue:7660", - "issue:7978", "issue:8464", "issue:8729", "issue:8792", @@ -44804,8 +44836,6 @@ "issue:9856" ], "soft_pairs": [ - "issue:5004|issue:6093", - "issue:4213|issue:7978", "issue:5630|issue:8792", "issue:2107|issue:4104", "issue:4402|issue:8900", @@ -44827,187 +44857,192 @@ "issue:4185|issue:4476", "issue:1931|issue:6164", "issue:7354|issue:7660", - "issue:12533|issue:12536" + "issue:12533|issue:12536", + "issue:10550|issue:10594", + "issue:1434|issue:5215" ], - "prepared_review_unit_hash": "42fa3f7d422c1aab0d9da6be5bf96c5db6cffe1fa11209d8744535fc0b4de5ab", + "prepared_review_unit_hash": "397b5c2b4b5662c734651142296b856d881e407c875590851799c93f9607e703", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34613, - "estimated_input_tokens": 8654, - "estimated_eval_tokens": 17564 + "serialized_chars": 34685, + "estimated_input_tokens": 8672, + "estimated_eval_tokens": 17600 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is heterogeneous: pipeline feature requests, model-loading issues, training bugs, distributed/context-parallelism errors, and unrelated runtime failures. None of the soft pairs look like the same underlying bug or change, so all candidate duplicate links should be rejected.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue is safe: the items span unrelated subsystems and failure modes. If one must be chosen as the closest anchor, issue 12533 is the most concrete and broadly scoped around context-parallelism/offload interactions, but it is not a true duplicate hub for the whole set.", + "summary": "Only one soft pair is a clear duplicate: the two issues about preferring .safetensors/.bin weights when both exist. The rest are topic-adjacent at best but describe different pipelines, symptoms, or code paths.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 1811 is the cleaner canonical because it states the exact file-selection bug in `from_pretrained` and is phrased as the underlying behavior to fix; 5719 is the same bug from the LoRA-loading path and is narrower.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 12533 is the strongest umbrella-style issue among the nearby context-parallelism reports because it names the concrete hook/offload conflict and has the most general framing. Even so, the cluster overall is not cohesive enough to justify merging the others into it.", + "best_issue_reason": "Issue 1811 is the best representative overall: specific, actionable, and clearly about a single loader bug rather than a broad symptom report.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:5004", - "right": "issue:6093", - "accept": false, - "reason": "Different training-script problems: fp16 resume checkpointing vs push_to_hub repo creation behavior." - }, - { - "left": "issue:4213", - "right": "issue:7978", - "accept": false, - "reason": "Unrelated issues: from_single_file with SDXL/accelerate versus a torchvision nms runtime dependency error." - }, { "left": "issue:5630", "right": "issue:8792", "accept": false, - "reason": "Both concern image quality, but they describe different situations and likely different causes: SDXL img2img scheduler leftovers vs migration quality degradation." + "reason": "Both involve SDXL image quality, but one is about LMS scheduler leftovers in img2img and the other is a general migration-quality complaint; not the same concrete bug." }, { "left": "issue:2107", "right": "issue:4104", "accept": false, - "reason": "Both mention black/glitchy outputs, but one is low-resolution image artifacts and the other is a cast warning leading to black output; not clearly the same bug." + "reason": "Both mention black/glitchy images, but 2107 is resolution-specific while 4104 is tied to an invalid-cast warning; likely different failure modes." }, { "left": "issue:4402", "right": "issue:8900", "accept": false, - "reason": "Both are from_single_file loading problems, but one is SDXL inpaint misuse and the other is loading an SDXL refiner file; too different in code path and symptom." + "reason": "Both use `from_single_file`, but they concern different SDXL models and different loading problems; not one shared bug." }, { "left": "issue:12459", "right": "issue:9717", "accept": false, - "reason": "Different contexts: FLUX batch throughput issue versus a docker image runtime error." + "reason": "Throughput in batch inference and a docker image launch error are unrelated." }, { "left": "issue:4726", "right": "issue:4806", "accept": false, - "reason": "Type-related issues, but one is StableDiffusionPipeline callback typing and the other is UNet2DModel parameter typing; not the same defect." + "reason": "Both are type-related API issues, but they affect different symbols and likely different fixes." }, { "left": "issue:5897", "right": "issue:8826", "accept": false, - "reason": "Different failure modes: device placement mismatch during LoRA training versus an empty tensor list in cat()." + "reason": "LoRA training device-mismatch and an empty-list `torch.cat` error are distinct runtime problems." }, { "left": "issue:10650", "right": "issue:7003", "accept": false, - "reason": "Both are shape errors, but for different models/pipelines and incompatible tensor dimensions; not the same underlying bug." + "reason": "Both are shape mismatch errors, but they occur on different models/pipelines and point to different expected tensor layouts." }, { "left": "issue:8464", "right": "issue:9637", "accept": false, - "reason": "Both involve runtime/distributed-style failures, but one is SEGA cpu-offload seeding and the other is XFormers attention-mask/bfloat16 sequence-length constraints." + "reason": "CPU offload plus seed handling is unrelated to xFormers attention-mask/bfloat shape constraints." }, { "left": "issue:10447", "right": "issue:7267", "accept": false, - "reason": "Both are data-loading/script issues, but one is a converter splitting UNet weights and the other is broken local data loading in train_controlnet.py." + "reason": "A conversion script splitting UNet weights is unrelated to broken local data loading in a training script." }, { "left": "issue:6362", "right": "issue:9523", "accept": false, - "reason": "Both mention training/mapping, but one is SDXL fine-tuning OOM during dataset mapping and the other is controlnet training mapping errors across sd3/sdxl/flux." + "reason": "Both mention dataset mapping, but one is an OOM during fine-tuning and the other is an error in specific controlnet training scripts; too broad to be the same bug." }, { "left": "issue:2673", "right": "issue:3182", "accept": false, - "reason": "Inpaint/upscale pipeline error versus a custom LPW pipeline failing with cpu offload; different pipeline implementations and failure points." + "reason": "Upscale pipeline config/image input errors are unrelated to custom pipeline CPU offload incompatibility." }, { "left": "issue:4368", "right": "issue:8729", "accept": false, - "reason": "LoRA training OOM and unexpected validation image generation in controlnet training are unrelated symptoms." + "reason": "LoRA training OOM and an unexpected validation image in a controlnet training script are different issues." }, { "left": "issue:12396", "right": "issue:9856", "accept": false, - "reason": "Adapter deletion with leaf_level versus distributed port-in-use startup failure; unrelated subsystems." + "reason": "Adapter deletion/leaf-level errors are unrelated to distributed port conflicts." }, { "left": "issue:5399", "right": "issue:5509", "accept": false, - "reason": "Both are inference-time errors, but one is dtype mismatch in custom inpainting and the other is image/prompt batch-size handling." + "reason": "A dtype mismatch in custom inpainting is not the same as image batch-size validation for numpy/cv2 inputs." }, { "left": "issue:1037", "right": "issue:2838", "accept": false, - "reason": "Community pipeline request versus a controlnet pipeline feature/request issue; not the same bug." + "reason": "A request to add a pipeline and a request to support a controlnet pipeline without control are not duplicates." }, { "left": "issue:1125", "right": "issue:4605", "accept": false, - "reason": "Different inpainting-related problems: a Stable Diffusion Mega pipeline runtime error versus a long-prompt-weighting inpaint model error." + "reason": "Both mention inpaint-related errors, but the pipelines and reported failures differ; not enough evidence of one shared bug." }, { "left": "issue:8839", "right": "issue:9767", "accept": false, - "reason": "Both involve LoRA on newer model families, but one is an SD3 weight error and the other is a FluxImg2Img cuDNN frontend failure." + "reason": "LoRA weight errors on SD3 and a cuDNN frontend error on Flux Img2Img are unrelated." }, { "left": "issue:1811", "right": "issue:5719", - "accept": false, - "reason": "Both concern load-order behavior for weights, but one is generic from_pretrained file selection and the other is LoRA weight file discovery order." + "accept": true, + "reason": "Both report the same loader-selection bug: when both `.bin` and `.safetensors` are present, the code should prefer one instead of treating them as multiple files." }, { "left": "issue:4185", "right": "issue:4476", "accept": false, - "reason": "Both involve controlnet/SDXL image quality, but one is black images after rebasing and the other is failed expected results on fill50k; different manifestations and likely causes." + "reason": "Both touch SDXL ControlNet, but one is about black images at inference and the other about unexpected results on a training dataset; different bugs." }, { "left": "issue:1931", "right": "issue:6164", "accept": false, - "reason": "Both are dtype/output-type issues, but one is float16 vs float mismatch in SD2 and the other is latent output handling in StableDiffusionPipeline." + "reason": "Float16 dtype handling and latent output-type handling are separate pipeline behaviors." }, { "left": "issue:7354", "right": "issue:7660", "accept": false, - "reason": "DDPMPipeline scheduler compatibility and trace example failure are unrelated." + "reason": "Scheduler compatibility with DDPMPipeline is unrelated to a trace example failure." }, { "left": "issue:12533", "right": "issue:12536", "accept": false, - "reason": "Both mention context parallelism, but one is a hooks conflict with CPU offload while the other is a tensor-size divisibility assertion in WanImageToVideoPipeline; not the same code-path problem." + "reason": "Both involve context parallelism, but one is a hook conflict with CPU offload and the other is a mesh-size divisibility assertion; different code paths." + }, + { + "left": "issue:10550", + "right": "issue:10594", + "accept": false, + "reason": "Loading LoRA into a quantized base model and using multiple controls with LoRA on FLUX are different problems." + }, + { + "left": "issue:1434", + "right": "issue:5215", + "accept": false, + "reason": "Upscale OOM at larger resolutions and brightness shifts for dark images are unrelated symptoms." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and the soft-edge calls are conservative. The only accepted pair, 1811/5719, is reasonably supported as the same loader-selection bug, while the rest are plausibly different bugs/code paths based on the titles and metadata provided." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -45019,15 +45054,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10026", "issue:10287", - "issue:10550", - "issue:10594", "issue:10804", "issue:10952", "issue:11024", @@ -45038,7 +45071,6 @@ "issue:12299", "issue:12529", "issue:13162", - "issue:1434", "issue:1544", "issue:1925", "issue:202", @@ -45056,7 +45088,6 @@ "issue:4488", "issue:4801", "issue:4840", - "issue:5215", "issue:5344", "issue:5545", "issue:5577", @@ -45067,14 +45098,16 @@ "issue:7057", "issue:7210", "issue:7441", + "issue:7690", "issue:7985", "issue:861", + "issue:8873", "issue:9089", - "issue:9164" + "issue:9164", + "issue:9576", + "issue:9804" ], "soft_pairs": [ - "issue:10550|issue:10594", - "issue:1434|issue:5215", "issue:1925|issue:2036", "issue:202|issue:4001", "issue:2068|issue:4488", @@ -45096,180 +45129,182 @@ "issue:2462|issue:4080", "issue:5344|issue:9089", "issue:6894|issue:9164", - "issue:10287|issue:11024" + "issue:10287|issue:11024", + "issue:7690|issue:9804", + "issue:8873|issue:9576" ], - "prepared_review_unit_hash": "e4742c8548f5d09559756cad74452b722d63f004c7f06e4ace471cf3b57992bc", + "prepared_review_unit_hash": "ce107ad420b6e6857d0967ab40dd6ae86d63c3ca535c01bff77bbdf66914afe2", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33320, - "estimated_input_tokens": 8330, - "estimated_eval_tokens": 16916 + "serialized_chars": 33274, + "estimated_input_tokens": 8319, + "estimated_eval_tokens": 16894 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is noisy and heterogeneous: it mixes unrelated reports about LoRA loading, ControlNet, VAE/memory, compilation, docs, tests, and example scripts. None of the soft pairs look like the same underlying bug or change, so all soft edges should be rejected.", + "summary": "This cluster is heterogeneous: it mixes unrelated docs, training, loading, ControlNet, LoRA, scheduler, and compile/memory issues across different model families. None of the soft pairs look like the same underlying bug or change.", "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue: the items span distinct bug reports and requests with different models, pipelines, and failure modes.", + "canonical_issue_reason": "No strong canonical issue exists because the cluster is not cohesive. If forced to pick one, #10287 is the broadest, but it is too vague to represent the others as a true duplicate anchor.", "canonical_pr_reason": null, - "best_issue_reason": "If forced to pick one representative issue, issue:10550 is the most concrete and well-scoped, but it is not a true duplicate target for the rest of the cluster.", + "best_issue_reason": "#10287 is the most generic/broad item in the set, but it is still a weak representative because the rest of the cluster spans many unrelated failures.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:10550", - "right": "issue:10594", - "accept": false, - "reason": "Both mention LoRA, but one is quantized-base loading and the other is multi-Control FLUX usage; different failure paths." - }, - { - "left": "issue:1434", - "right": "issue:5215", - "accept": false, - "reason": "Both are about upscale pipelines, but one is OOM while the other is output brightness drift; different bugs." - }, { "left": "issue:1925", "right": "issue:2036", "accept": false, - "reason": "Different models and symptoms: half-precision in SD2 inpainting vs an OSError loading SD v1-4." + "reason": "Both mention model loading problems, but one is SD2 inpainting half-precision behavior and the other is a stable-diffusion-v1-4 OSError; different failures and code paths." }, { "left": "issue:202", "right": "issue:4001", "accept": false, - "reason": "These are unrelated repository maintenance tasks: a test addition vs a README demo link." + "reason": "A test request for unet_blocks is unrelated to a README/demo documentation update." }, { "left": "issue:2068", "right": "issue:4488", "accept": false, - "reason": "One is dtype conversion in from_pretrained; the other is SDXL checkpoint loading via from_original_stable_diffusion_ckpt." + "reason": "Different bugs: dtype conversion in from_pretrained vs SDXL checkpoint loading without pipeline_class." }, { "left": "issue:3595", "right": "issue:4186", "accept": false, - "reason": "Both involve loading/inpainting wording, but one is from_ckpt conversion and the other is SDXL inpaint with refiner behavior." + "reason": "convert-from-ckpt failure is not the same as SDXL inpaint/refiner behavior." }, { "left": "issue:4265", "right": "issue:6823", "accept": false, - "reason": "Import error for a specific SDXL ControlNet pipeline is not the same as a general 'cannot use controlnet' report." + "reason": "One is a specific import error for StableDiffusionXLControlNetPipeline; the other is a general 'cannot use controlnet' report." }, { "left": "issue:4453", "right": "issue:861", "accept": false, - "reason": "LoRA export/conversion for SDXL is unrelated to generic ckpt-to-diffusers conversion." + "reason": "LoRA conversion for SDXL/A1111 is different from generic ckpt-to-diffusers conversion." }, { "left": "issue:5783", "right": "issue:6234", "accept": false, - "reason": "Both reference train_text_to_image_sdxl.py, but one is EMA behavior and the other is a broader example-script issue." + "reason": "EMA training behavior and a generic example issue are not the same bug." }, { "left": "issue:7057", "right": "issue:7441", "accept": false, - "reason": "A model checkpoint loading problem is not the same as pytest failing on Python 3.12." + "reason": "Model checkpoint loading failure vs pytest failing on Python 3.12 are unrelated." }, { "left": "issue:1544", "right": "issue:2234", "accept": false, - "reason": "Scheduler API absence and xformers finetuning failure are distinct training-time issues." + "reason": "Missing get_velocity on DDPMScheduler is a scheduler API issue; xformers finetuning is a separate performance/training problem." }, { "left": "issue:10026", "right": "issue:12082", "accept": false, - "reason": "Different domains: Flux ControlNet multi-GPU memory scaling vs WAN VAE memory after compile." + "reason": "Both are GPU-memory-related, but one is Flux ControlNet + DeepSpeed training and the other is Wan VAE memory after compile; different stacks." }, { "left": "issue:11651", "right": "issue:13162", "accept": false, - "reason": "Both mention FLUX/LoRA, but one is loading flux-fill-lora through FluxControl and the other is support for a different FLUX.2 model." + "reason": "Both involve Flux LoRA support, but one is loading flux-fill-lora with FluxControl while the other is a request for Flux.2-klein-4b support; not the same concrete bug." }, { "left": "issue:12299", "right": "issue:12529", "accept": false, - "reason": "Both are tensor-size mismatches, but they affect different models and different code paths (SD3 embeddings vs Wan2.2 tiled VAE)." + "reason": "Different models and errors: SD3 tensor mismatch with negative prompt embeddings vs Wan2.2 tiled VAE size mismatch." }, { "left": "issue:5545", "right": "issue:5577", "accept": false, - "reason": "InpaintPipeline usage problem and ControlNet single-file key mismatch are unrelated failures." + "reason": "StableDiffusionInpaintPipeline issue and ControlNetModel.from_single_file KeyError are distinct loading paths." }, { "left": "issue:4840", "right": "issue:7985", "accept": false, - "reason": "Flax checkpoint loading and torch.compile compatibility are separate backend/runtime issues." + "reason": "FlaxAutoencoderKL checkpoint loading and IPAdapter torch.compile incompatibility are unrelated." }, { "left": "issue:3886", "right": "issue:4801", "accept": false, - "reason": "DDP collective timeout during unconditional training is not the same as a mixed-precision example bug." + "reason": "A distributed watchdog timeout is not the same as a mixed-precision example bug." }, { "left": "issue:10804", "right": "issue:10952", "accept": false, - "reason": "Both involve LoRA loading, but one is a FLUX1.dev ComfyUI-compatible LoRA failure and the other is an adapter-name mismatch in SDXL long-prompt weighting." + "reason": "Both are LoRA-loading related, but they fail in different pipelines with different adapter-name behavior; not clearly the same bug." }, { "left": "issue:5545", "right": "issue:7210", "accept": false, - "reason": "General inpaint pipeline problems and an accelerate dispatch attribute error are different issues." + "reason": "Inpaint pipeline trouble and missing named_parameters on StableDiffusionXLPipeline are separate issues." }, { "left": "issue:11687", "right": "issue:11700", "accept": false, - "reason": "These are adjacent feature requests for Pruna, but one is docs-only and the other is implementation integration." + "reason": "One is a docs request for adding Pruna; the other is implementation work for loading through from_pretrained/pipeline." }, { "left": "issue:2462", "right": "issue:4080", "accept": false, - "reason": "IterableDataset finetuning support and adding an SDXL inpainting pipeline are unrelated requests." + "reason": "IterableDataset training support and adding an SDXL inpainting pipeline are unrelated changes." }, { "left": "issue:5344", "right": "issue:9089", "accept": false, - "reason": "Missing dependency omegaconf and missing FluxPipeline class are different setup/library problems." + "reason": "Missing omegaconf dependency and missing FluxPipeline symbol are different problems." }, { "left": "issue:6894", "right": "issue:9164", "accept": false, - "reason": "Both concern LoRA/model usage, but one is an A1111 incompatibility regression and the other is training convergence of a Flux example." + "reason": "Both are training-example LoRA complaints, but they target different models and symptoms, so they are not the same underlying bug." }, { "left": "issue:10287", "right": "issue:11024", "accept": false, - "reason": "A documentation example issue is not the same as a SiglipImageProcessor .to() API bug in SD3IPAdapterMixin." + "reason": "A generic docs issue is unrelated to the SiglipImageProcessor .to() method bug." + }, + { + "left": "issue:7690", + "right": "issue:9804", + "accept": false, + "reason": "Tokenizer loading failure for SDXL base and local SDXL ControlNet loading failure are different code paths." + }, + { + "left": "issue:8873", + "right": "issue:9576", + "accept": false, + "reason": "xformers runtime slowdown and use_safetensors not propagating to submodules are unrelated." } ] }, @@ -45288,8 +45323,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -45305,9 +45340,12 @@ "issue:2052", "issue:2161", "issue:2163", + "issue:228", "issue:2577", "issue:2674", "issue:2774", + "issue:2880", + "issue:3064", "issue:381", "issue:3871", "issue:3948", @@ -45318,6 +45356,7 @@ "issue:5043", "issue:5465", "issue:5575", + "issue:5606", "issue:5607", "issue:5670", "issue:579", @@ -45325,26 +45364,20 @@ "issue:6130", "issue:6196", "issue:7592", - "issue:7690", "issue:7782", "issue:7793", "issue:8469", "issue:8493", "issue:8667", "issue:8828", - "issue:8873", "issue:8874", "issue:8950", "issue:9410", - "issue:9576", "issue:9611", "issue:9767", - "issue:9804", "issue:9895" ], "soft_pairs": [ - "issue:7690|issue:9804", - "issue:8873|issue:9576", "issue:9611|issue:9895", "issue:2161|issue:5575", "issue:8828|issue:8950", @@ -45366,180 +45399,182 @@ "issue:10863|issue:8493", "issue:1061|issue:1475", "issue:1867|issue:5465", - "issue:1949|issue:2163" + "issue:1949|issue:2163", + "issue:228|issue:2880", + "issue:3064|issue:5606" ], - "prepared_review_unit_hash": "44f5a46d58e762808147c425f0a004f06cdb6941077cc07fec4e8383dd28a255", + "prepared_review_unit_hash": "06af1b7014c88312b5554e7984df0172577f1794b48808e9362914996898dd3c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34082, - "estimated_input_tokens": 8521, - "estimated_eval_tokens": 17298 + "serialized_chars": 33945, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly superficial text-similarity matches; the paired issues concern different pipelines, models, or failure modes, so they should not be deduplicated.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "This cluster is mostly a grab bag of unrelated diffusers bugs and feature requests across training, loading, inference, and notebooks. None of the soft pairs look like true duplicates; they mostly share only a subsystem name or a generic error pattern.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue stands out because the items are not one underlying bug; they span distinct models, pipelines, and failure modes.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 10722 is the most concrete standalone bug report in the set: it has a specific runtime error, clear training context, and enough detail to be actionable.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:7690", - "right": "issue:9804", - "accept": false, - "reason": "Both are loading-related, but one is a tokenizer load failure and the other is local ControlNet SDXL loading; different code paths and bugs." - }, - { - "left": "issue:8873", - "right": "issue:9576", - "accept": false, - "reason": "One is an xformers performance complaint, the other is a safetensors flag propagation bug; unrelated issues." - }, { "left": "issue:9611", "right": "issue:9895", "accept": false, - "reason": "Both involve Flux, but one is image batching and the other is a device-placement regression from a specific commit; not the same bug." + "reason": "Both are Flux-related, but one is an image batching error and the other is a device-placement regression after a commit; different code paths." }, { "left": "issue:2161", "right": "issue:5575", "accept": false, - "reason": "Both mention LoRA, but one is inpainting-model training and the other is a hidden-size configuration question." + "reason": "Both mention LoRA training, but one is about inpainting-model support and the other is about hidden-size configuration for a layer." }, { "left": "issue:8828", "right": "issue:8950", "accept": false, - "reason": "Both are SD3 training reports, but one is a general 16GB training error and the other is a specific SD3 LoRA training bug." + "reason": "Both involve SD3 training, but one is a generic 16GB training error and the other is a LoRA training bug; not the same failure." }, { "left": "issue:10722", "right": "issue:5607", "accept": false, - "reason": "Both are tensor size mismatch errors, but they occur in different training scripts and describe different shape failures." + "reason": "Both are tensor-size mismatch errors, but the shapes and training scenarios differ, so they do not clearly point to one underlying bug." }, { "left": "issue:1534", "right": "issue:381", "accept": false, - "reason": "Both refer to notebooks, but one is a DreamBooth notebook error and the other is an inference-pipeline notebook section error." + "reason": "One is a DreamBooth notebook error and the other is a custom inference pipeline notebook issue; unrelated notebook paths." }, { "left": "issue:2774", "right": "issue:579", "accept": false, - "reason": "One is ControlNet safetensors conversion, the other is a missing config.json in textual inversion; different failure modes." + "reason": "ControlNet safetensors conversion and missing config.json in textual inversion are different loading problems." }, { "left": "issue:5043", "right": "issue:5925", "accept": false, - "reason": "Both are LoRA-related feature requests, but one asks to add finetuning and the other adds multi-LoRA inference support." + "reason": "These are separate feature requests: adding LoRA finetuning for Wuerstchen vs adding ZipLora support." }, { "left": "issue:5670", "right": "issue:8667", "accept": false, - "reason": "Different pipelines and different bugs: PixArtAlpha non-square generation versus SD3ControlNet tensor mismatch on non-power-of-two images." + "reason": "Both mention non-square image handling, but they target different pipelines/models and likely different shape assumptions." }, { "left": "issue:11046", "right": "issue:8469", "accept": false, - "reason": "Both involve IP-Adapter-related generation, but one is a Flux controlnet/inpainting pipeline request and the other is a corrupt-face output bug on SD15." + "reason": "One is a complex Flux pipeline composition request/error, the other is corrupted faces from a specific IP-Adapter model; not the same bug." }, { "left": "issue:3948", "right": "issue:4398", "accept": false, - "reason": "Both are image-quality complaints, but pale images and all-black SDXL LoRA outputs are not the same underlying defect." + "reason": "Both concern bad generated images, but pale images and all-black images are distinct symptoms with different likely causes." }, { "left": "issue:2577", "right": "issue:2674", "accept": false, - "reason": "One is DDIMInverseScheduler not working, the other is a LoRA dependency/requirements issue; unrelated." + "reason": "DDIMInverseScheduler usage and a LoRA dependency/requirements issue are unrelated." }, { "left": "issue:6130", "right": "issue:6196", "accept": false, - "reason": "Both are loading errors, but one is generic LoRA loading and the other is ControlNetXS Depth SDXL loading." + "reason": "Both are loading failures, but one is generic LoRA loading and the other is ControlNetXS Depth SDXL loading; different artifacts and code paths." }, { "left": "issue:1988", "right": "issue:3871", "accept": false, - "reason": "Different training problems: fp16/bf16 dtype mismatch during image generation versus checkpoint resume failure." + "reason": "One is an fp16/bf16 dtype mismatch during image generation in a trainer example; the other is checkpoint-resume trouble in text-to-image training." }, { "left": "issue:5025", "right": "issue:9410", "accept": false, - "reason": "Import errors for different symbols and pipelines; no shared bug beyond generic import failure." + "reason": "An import error for randn_tensor and an inability to import a specific SD3 ControlNet inpainting pipeline are different import failures." }, { "left": "issue:2052", "right": "issue:4566", "accept": false, - "reason": "Both concern resume-from-checkpoint behavior, but they are different scripts (textual inversion vs SDXL LoRA) and not clearly the same bug." + "reason": "Both mention resuming from checkpoint, but they affect different training scripts and model families." }, { "left": "issue:4086", "right": "issue:7592", "accept": false, - "reason": "Different feature areas and failures: SDXL ControlNet training versus DoRA state_dict loading." + "reason": "ControlNet training on SDXL and DoRA state_dict loading are separate problems involving different models and failure points." }, { "left": "issue:7782", "right": "issue:8874", "accept": false, - "reason": "Both are model-file loading errors, but one is a bad safetensors conversion and the other is CKPT single-file loading with weights-only unpickling failure." + "reason": "Both involve loading from files/checkpoints, but one is a bad safetensors conversion and the other is a CKPT unpickling weights-only failure." }, { "left": "issue:7793", "right": "issue:9767", "accept": false, - "reason": "One is unloading LoRA weights error in DeepFloyd; the other is a cuDNN frontend runtime error in Flux LoRA inference." + "reason": "One is unloading LoRA weights in DeepFloyd, the other is a cuDNN frontend error during Flux LoRA inference; not the same bug." }, { "left": "issue:10863", "right": "issue:8493", "accept": false, - "reason": "Both are load-time errors, but they affect different artifacts and classes: local safetensors loading versus SD3Transformer2DModel state_dict loading." + "reason": "Both are state_dict loading errors, but one is a local safetensors file load failure and the other is SD3Transformer2DModel key mismatch." }, { "left": "issue:1061", "right": "issue:1475", "accept": false, - "reason": "Textual inversion example documentation update is unrelated to DreamBooth/textual inversion training with two classes of images." + "reason": "These are different training/tutorial issues: a docs update for textual inversion vs training with two class types in DreamBooth/textual inversion." }, { "left": "issue:1867", "right": "issue:5465", "accept": false, - "reason": "Both involve strength-like parameters, but img2img scheduler behavior and DiffEdit inpaint_strength limits are different pipeline bugs." + "reason": "Both involve image-to-image strength behavior, but img2img scheduler behavior and DiffEdit inpaint_strength support are distinct issues." }, { "left": "issue:1949", "right": "issue:2163", "accept": false, - "reason": "Both are training-related runtime errors, but one is a DreamBooth tutorial failure and the other is a dtype mismatch in text-to-image training." + "reason": "Both are training example errors, but one is a custom-model tutorial problem and the other is a UNet dtype mismatch in train_text_to_image." + }, + { + "left": "issue:228", + "right": "issue:2880", + "accept": false, + "reason": "A broken release link and a Colab notebook that does not run are unrelated." + }, + { + "left": "issue:3064", + "right": "issue:5606", + "accept": false, + "reason": "Both mention LoRA and safetensors, but one is loading a .safetensors LoRA and the other is handling missing alpha metadata." } ] }, @@ -45558,8 +45593,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -45570,12 +45605,11 @@ "issue:12348", "issue:12536", "issue:1988", - "issue:228", "issue:2596", + "issue:2603", "issue:2669", - "issue:2880", + "issue:2673", "issue:2991", - "issue:3064", "issue:3412", "issue:3910", "issue:4031", @@ -45583,10 +45617,10 @@ "issue:4134", "issue:4697", "issue:5025", + "issue:5227", "issue:5458", "issue:5595", "issue:5596", - "issue:5606", "issue:5862", "issue:5905", "issue:6162", @@ -45595,6 +45629,7 @@ "issue:7114", "issue:7185", "issue:7267", + "issue:7420", "issue:7556", "issue:7636", "issue:7799", @@ -45612,8 +45647,6 @@ "issue:9670" ], "soft_pairs": [ - "issue:228|issue:2880", - "issue:3064|issue:5606", "issue:3412|issue:7556", "issue:5905|issue:8482", "issue:9378|issue:9402", @@ -45635,180 +45668,182 @@ "issue:5862|issue:9515", "issue:6791|issue:7114", "issue:12348|issue:12536", - "issue:11046|issue:8450" + "issue:11046|issue:8450", + "issue:5227|issue:7420", + "issue:2603|issue:2673" ], - "prepared_review_unit_hash": "8dd0cf15464f070325b0fed84600fdc8d36e1508f2f7757bdac809f0bef1de6d", + "prepared_review_unit_hash": "9c074aa2cdb8b52b973fb9cfcf8443d375d7f64e3af92d1901a76e481311fb1a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33361, - "estimated_input_tokens": 8341, - "estimated_eval_tokens": 16938 + "serialized_chars": 33465, + "estimated_input_tokens": 8367, + "estimated_eval_tokens": 16990 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated issues. The soft pairs share generic error wording (shape mismatch, loading failure, docs typo, etc.) but usually involve different models, pipelines, or root causes, so there is no safe canonical duplicate here.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "These items are mostly one-off runtime, loading, or documentation issues across different pipelines and models. None of the soft pairs look like true duplicates of the same concrete bug/change; they mostly share a subsystem or symptom type rather than the same code-path failure.", + "confidence": 0.38, + "canonical_issue_reason": "No clear duplicate cluster emerges, so the best representative issue is issue 12348: it\u2019s a concrete, reproducible inference tensor-mismatch report for WanImageToVideoPipeline and is one of the most specific runtime-bug entries in the set.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 12348 is the strongest standalone issue in this set because it describes an actionable pipeline failure with a clear symptom and scope. It is more representative than the vague feature requests or generic error reports.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:228", - "right": "issue:2880", - "accept": false, - "reason": "Different problems: broken release download link vs JAX Colab not running." - }, - { - "left": "issue:3064", - "right": "issue:5606", - "accept": false, - "reason": "Both relate to LoRA loading, but one is safetensors loading and the other is missing alpha handling; not the same bug." - }, { "left": "issue:3412", "right": "issue:7556", "accept": false, - "reason": "Unrelated docs updates for different topics." + "reason": "Both are docs-related, but one is a stable-diffusion docs update and the other is Animatediff scheduler docs; different scope and no shared bug/change." }, { "left": "issue:5905", "right": "issue:8482", "accept": false, - "reason": "Different feature requests: diffusion-GAN support vs SD3 model support." + "reason": "Different model requests: diffusion-GAN versus Stable Diffusion 3 support. Same broad repo area, not the same underlying change." }, { "left": "issue:9378", "right": "issue:9402", "accept": false, - "reason": "Both Flux ControlNet related, but one requests Xlabs support and the other adds img2img/inpaint pipelines." + "reason": "Both involve Flux ControlNet, but one requests Xlabs ControlNet support and the other adds img2img/inpaint pipelines; separate features." }, { "left": "issue:7799", "right": "issue:9115", "accept": false, - "reason": "Different pipelines and failures: InstructPix2Pix + IP-Adapter vs AnimateDiff-Lightning/SDXL motion adapter." + "reason": "Different pipelines and failure modes; both are image-generation errors but not the same concrete bug." }, { "left": "issue:4031", "right": "issue:7267", "accept": false, - "reason": "Both are loading/training issues, but one is safety checker local-file loading and the other is train_controlnet local data loading." + "reason": "Unrelated loading problems: safety checker local-file loading versus train_controlnet local data loading." }, { "left": "issue:3910", "right": "issue:4697", "accept": false, - "reason": "Both LoRA-related, but one is generic load failure and the other is a specific post-0.20.0 regression." + "reason": "Both mention LoRA loading, but the reports are too generic and likely different failure causes/regressions; not enough to treat as the same bug." }, { "left": "issue:5595", "right": "issue:6435", "accept": false, - "reason": "Same size-ish symptom, but one is invalid height/width input validation and the other is ControlNet default dimensions initialization." + "reason": "One is a validation error for non-divisible dimensions; the other is a ControlNet default width/height initialization problem." }, { "left": "issue:9227", "right": "issue:9309", "accept": false, - "reason": "Different Flux training vs Flux ControlNet inference failures." + "reason": "Both are Flux-related, but one is a DreamBooth LoRA training issue and the other is a FluxControlNetPipeline runtime bug." }, { "left": "issue:2669", "right": "issue:5596", "accept": false, - "reason": "Both are AttributeError reports, but on different objects and workflows; no shared code path shown." + "reason": "Different attribute errors in different preprocessing/build paths; no clear shared code-path." }, { "left": "issue:2991", "right": "issue:4037", "accept": false, - "reason": "Both mention ControlNet, but one is mixed-precision example failure and the other is FSDP validation errors." + "reason": "Both are ControlNet training issues, but mixed-precision example failure and FSDP validation errors are distinct problems." }, { "left": "issue:9341", "right": "issue:9670", "accept": false, - "reason": "Different LoRA backend problems: Flux text_model LoRA loading vs PEFT-required error." + "reason": "Both concern LoRA loading, but one is about Flux LoRA key handling and the other is a missing PEFT backend requirement." }, { "left": "issue:4134", "right": "issue:7636", "accept": false, - "reason": "Both ControlNet-related, but one is an MPS tensor size mismatch and the other is a usage question about SDXL tile model." + "reason": "A device-specific tensor size mismatch on MPS versus a how-to/support request for the SDXL tile ControlNet model." }, { "left": "issue:10520", "right": "issue:9045", "accept": false, - "reason": "Device-mismatch errors in different pipelines and setups; not enough evidence of the same bug." + "reason": "Both are device-placement issues, but they hit different models/pipelines and different code paths." }, { "left": "issue:1988", "right": "issue:6162", "accept": false, - "reason": "Different mismatch types and models: fp16/bf16 trainer bias/input type error vs SDXL IP-Adapter shape mismatch." + "reason": "Both are generation-time shape/type mismatches, but one is fp16/bf16 input-bias type mismatch and the other is an IP-Adapter shape mismatch." }, { "left": "issue:11185", "right": "issue:11824", "accept": false, - "reason": "Different models and failures: CogView4 image generation error vs FluxKontext multi-image input error." + "reason": "Different models and different error contexts; both are generic generation failures only at the title level." }, { "left": "issue:2596", "right": "issue:5025", "accept": false, - "reason": "Both are docs/import name issues, but one is a documentation typo and the other is an actual missing utility export." + "reason": "Both are diffusers.utils import/documentation problems, but they refer to different symbols and different underlying breakages." }, { "left": "issue:3910", "right": "issue:8237", "accept": false, - "reason": "Both LoRA-loading issues, but they concern different failure modes and model variants." + "reason": "Both are LoRA-loading complaints, but one is a generic load failure and the other is a missing SDXL text-projection key." }, { "left": "issue:5458", "right": "issue:7185", "accept": false, - "reason": "Different problems: local_files_only behavior in LoRA loading vs PyTorch 2.2 compatibility question." + "reason": "Local-files-only behavior in LoRA loading is unrelated to a broader PyTorch version support question." }, { "left": "issue:5862", "right": "issue:9515", "accept": false, - "reason": "Both about transformer internals, but one is missing gradient checkpointing support and the other is missing RMSNorm in a processor." + "reason": "Different Transformer internals: gradient checkpointing support versus missing RMSNorm in a specific attention processor." }, { "left": "issue:6791", "right": "issue:7114", "accept": false, - "reason": "Same training script, but one cannot run at all while the other is specifically CUDA OOM." + "reason": "Same training script, but one report says it can't run while the other is specifically CUDA OOM; not the same bug." }, { "left": "issue:12348", "right": "issue:12536", "accept": false, - "reason": "Both WanImageToVideoPipeline issues, but one is a resolution-dependent tensor mismatch while the other is a context-parallel sharding divisibility assertion." + "reason": "Both are WanImageToVideoPipeline tensor-size complaints, but one is a resolution-dependent mismatch and the other is a context-parallelism mesh divisibility assertion; related theme, not clearly the same bug." }, { "left": "issue:11046", "right": "issue:8450", "accept": false, - "reason": "Different Flux pipeline questions: multi-component inference request vs SDXL inpainting noise bug." + "reason": "Both involve Flux/SD-style inference, but one is a multi-feature Flux pipeline request and the other is an SDXL inpainting noise bug." + }, + { + "left": "issue:5227", + "right": "issue:7420", + "accept": false, + "reason": "Both mention torch.compile, but one is specifically load_lora_weight incompatibility and the other is a general inability to use torch.compile." + }, + { + "left": "issue:2603", + "right": "issue:2673", + "accept": false, + "reason": "Different pipeline issues: non-square image support versus an upscale pipeline config/image input error." } ] }, @@ -45827,8 +45862,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -45845,9 +45880,7 @@ "issue:1647", "issue:2185", "issue:2210", - "issue:2603", "issue:2645", - "issue:2673", "issue:2689", "issue:2841", "issue:3111", @@ -45862,7 +45895,6 @@ "issue:4837", "issue:5170", "issue:5226", - "issue:5227", "issue:5337", "issue:5387", "issue:5593", @@ -45870,19 +45902,20 @@ "issue:6415", "issue:6610", "issue:6713", + "issue:6731", "issue:6758", "issue:7214", "issue:7229", - "issue:7420", "issue:7441", "issue:7592", "issue:7974", + "issue:8541", "issue:8615", - "issue:9237" + "issue:9237", + "issue:9454", + "issue:9729" ], "soft_pairs": [ - "issue:5227|issue:7420", - "issue:2603|issue:2673", "issue:7229|issue:9237", "issue:10575|issue:7214", "issue:10972|issue:11209", @@ -45905,192 +45938,194 @@ "issue:4080|issue:8615", "issue:4354|issue:4667", "issue:4647|issue:7441", - "issue:5226|issue:6758" + "issue:5226|issue:6758", + "issue:6731|issue:9729", + "issue:8541|issue:9454" ], - "prepared_review_unit_hash": "2342a25d78a53bb7944f1f73e9e70f7448c8bc7c85dbe8a0eb30b6d2aa860ab3", + "prepared_review_unit_hash": "5c075ccbe10b6b4c3a5a3c0523b8f20a2423b25379637f1192185327d5f75c6a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 33348, - "estimated_input_tokens": 8337, - "estimated_eval_tokens": 16930 + "serialized_chars": 33452, + "estimated_input_tokens": 8363, + "estimated_eval_tokens": 16982 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly noisy and heterogeneous, with one clear duplicate subgroup around unloading/removing LoRA weights after loading. Most other soft pairs are only superficially similar by subsystem or wording.", - "confidence": 0.84, - "canonical_issue_reason": "issue:3689 is the clearest representative of the LoRA unload/remove-after-load bug request and has the strongest discussion signal among the matching subgroup.", + "summary": "This set is mostly heterogeneous issues with only a couple of likely duplicate subclusters. The clearest matches are the LoRA-loading reports and the LoRA-removal reports; most other soft edges are only broad subsystem similarity and should stay separate.", + "confidence": 0.79, + "canonical_issue_reason": "issue:10972 is the strongest issue canonical in this set because it is the most specific LoRA-loading failure and has two near-identical follow-up reports (11209, 11224) pointing to the same problem class.", "canonical_pr_reason": null, - "best_issue_reason": "issue:3689 is the best overall issue anchor because it states the concrete LoRA removal problem directly and aligns with the only obvious duplicate pair in the set.", + "best_issue_reason": "issue:10972 is the best representative issue overall: concrete bug title, clear failure mode, and it anchors the only strong duplicate-like pairings in the set.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:5227", - "right": "issue:7420", - "accept": false, - "reason": "Both mention torch.compile, but 5227 is specifically about load_lora_weights with torch.compile while 7420 is a generic compile failure; different failure paths." - }, - { - "left": "issue:2603", - "right": "issue:2673", - "accept": false, - "reason": "Different problems: non-square image generation in Attend-And-Excite vs config/image input validation in StableDiffusionUpscalePipeline." - }, { "left": "issue:7229", "right": "issue:9237", "accept": false, - "reason": "Unrelated training/checkpoint save issue versus a Flux dreambooth LoRA training issue; same broad area only." + "reason": "Both mention training issues, but they concern different workflows and failure modes; not the same underlying bug." }, { "left": "issue:10575", "right": "issue:7214", "accept": false, - "reason": "Both involve DPMSolverMultistepScheduler, but one is an IndexError out-of-bounds case and the other is a karras_sigmas regression; not the same concrete bug." + "reason": "Same scheduler family, but one is an IndexError and the other is a karras_sigmas regression; too different to merge." }, { "left": "issue:10972", "right": "issue:11209", - "accept": false, - "reason": "10972 is specific to OneTrainer Flux LoRAs, while 11209 is only a generic LoRA loading error; not enough evidence they are the same bug." + "accept": true, + "reason": "Both are LoRA loading failures; 11209 looks like a generic report of the same underlying loading bug seen in 10972." }, { "left": "issue:10972", "right": "issue:11224", - "accept": false, - "reason": "Same concern as above: generic LoRA load failure versus a specific Flux LoRA format issue." + "accept": true, + "reason": "Same LoRA-loading failure class as 10972, with 11224 appearing to be another report of the same bug." }, { "left": "issue:2185", "right": "issue:3347", "accept": false, - "reason": "Training config-file omissions and NaN image output are unrelated." + "reason": "Training config files missing vs NaN image output are unrelated issues." }, { "left": "issue:2645", "right": "issue:4701", "accept": false, - "reason": "Both are CUDA-related, but they affect different components and error modes; not the same underlying defect." + "reason": "Both are CUDA-related errors, but they occur in different code paths and are not the same bug." }, { "left": "issue:2841", "right": "issue:3724", "accept": false, - "reason": "A scheduler sampling-temperature feature request and a watermarking module idea are unrelated." + "reason": "One is a scheduler feature request and the other is a watermarking feature idea; unrelated." }, { "left": "issue:12319", "right": "issue:12673", "accept": false, - "reason": "Both mention group offloading, but the reported failures are different (block_level vs record_stream=True compatibility) and likely separate code paths." + "reason": "Both involve group offloading, but the reported problems are different enough that they do not look like the same bug." }, { "left": "issue:4837", "right": "issue:6713", "accept": false, - "reason": "Same converter script, but different reported failures and model contexts; not clearly the same bug." + "reason": "Both mention the same conversion script, but the failures differ (SDXL conversion vs HeaderTooLarge)." }, { "left": "issue:6415", "right": "issue:7974", "accept": false, - "reason": "Different IP-Adapter feature requests/bugs with no concrete overlap beyond the general adapter area." + "reason": "Different IP-Adapter requests/bugs; not the same concrete issue." }, { "left": "issue:12329", "right": "issue:12565", "accept": false, - "reason": "Version-detection/loading of Wan2.2 versus a broader Wan2.1 i2v running issue are not the same defect." + "reason": "Same Wan image-to-video area, but one is a model-loading/version mismatch and the other is a running issue; not clearly duplicate." }, { "left": "issue:5226", - "right": "issue:6758", + "right": "issue:7592", "accept": false, - "reason": "Both are pipeline capability issues, but one is about load_lora_weights on SDXL Adapter and the other is about num_images_per_prompt on InstantID; different problems." + "reason": "Both are LoRA/adapter loading problems, but they concern different mechanisms (load_lora_weights vs DoRA state_dict keys)." }, { "left": "issue:5647", "right": "issue:6610", "accept": false, - "reason": "Both concern SDXL inpainting, but one is a strength==1 bug and the other is from_single_file num_in_channels handling; different code paths." + "reason": "Both hit SDXL inpaint pipeline paths, but the symptoms and triggers differ." }, { "left": "issue:11103", "right": "issue:1647", "accept": false, - "reason": "Unrelated question topics with no shared bug or feature change." + "reason": "A repo-selection question and a color-correction question are unrelated." }, { "left": "issue:1304", "right": "issue:5337", "accept": false, - "reason": "A training question and a callback explanation question are unrelated." + "reason": "Training guidance vs callback semantics; no shared bug/change." }, { "left": "issue:2210", "right": "issue:3111", "accept": false, - "reason": "Training the text encoder for LoRA and using a custom conditioning encoder are adjacent concepts but not the same underlying request." + "reason": "Both are about conditioning/encoder ideas, but they are different feature requests." }, { "left": "issue:2645", "right": "issue:5593", "accept": false, - "reason": "Different pipeline errors; only the general pipeline domain overlaps." + "reason": "ControlNet CUDA failure vs AutoPipeline ControlNet wrapping failure; different issues." }, { "left": "issue:2689", "right": "issue:5387", "accept": false, - "reason": "Both mention broken links, but they refer to different documentation assets and are not the same issue." + "reason": "Both are broken-link reports, but they point to different links/doc pages." }, { "left": "issue:3689", "right": "issue:5170", "accept": true, - "reason": "These are the same underlying request/bug: removing or unloading LoRA weights after loading them." + "reason": "These describe the same underlying need to remove/unload LoRA weights after loading them." }, { "left": "issue:4080", "right": "issue:8615", "accept": false, - "reason": "Inpainting support for SDXL and fuse_lora support for SD3 are unrelated changes." + "reason": "Different feature requests in different model families; not duplicates." }, { "left": "issue:4354", "right": "issue:4667", "accept": false, - "reason": "Both involve from_single_file, but they target different models and different failures." + "reason": "Both involve from_single_file, but the pipeline class and failure mode differ." }, { "left": "issue:4647", "right": "issue:7441", "accept": false, - "reason": "Unrelated model-architecture bug versus test runner compatibility issue." + "reason": "Model architecture bug vs test/runtime compatibility issue; unrelated." }, { "left": "issue:5226", "right": "issue:6758", "accept": false, - "reason": "Different pipeline-specific capability bugs; no concrete shared code-path problem." + "reason": "One is a LoRA loading bug and the other is a missing parameter support issue; not the same bug." + }, + { + "left": "issue:6731", + "right": "issue:9729", + "accept": false, + "reason": "Unhandled pipeline class vs unexpected saved model type in Flux training; unrelated." + }, + { + "left": "issue:8541", + "right": "issue:9454", + "accept": false, + "reason": "Both are model-not-found errors, but they refer to different models and repos." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet, and the stated soft-edge verdicts are conservative overall. The lone accepted pair (3689/5170) is strongly supported by the titles, and the other rejections avoid overclaiming despite broad subsystem similarity." + "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted soft edges are defensible from the titles alone, especially the identical LoRA-loading reports and the unload/remove LoRA pair. The rejected edges are generally conservative and do not overclaim duplicates. No mergeability issues apply here because this is issue clustering rather than a PR pair." }, "evaluator_used": true, "retried": false, @@ -46106,8 +46141,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -46130,6 +46165,7 @@ "issue:2163", "issue:2234", "issue:2244", + "issue:3250", "issue:3382", "issue:3468", "issue:4185", @@ -46139,8 +46175,9 @@ "issue:4702", "issue:5442", "issue:5919", + "issue:6258", + "issue:6335", "issue:6407", - "issue:6731", "issue:6914", "issue:6949", "issue:7330", @@ -46148,21 +46185,17 @@ "issue:7793", "issue:7985", "issue:8532", - "issue:8541", "issue:8626", "issue:8887", + "issue:8922", "issue:9069", "issue:9112", - "issue:9454", "issue:9523", "issue:954", - "issue:9729", "issue:9744", "issue:9857" ], "soft_pairs": [ - "issue:6731|issue:9729", - "issue:8541|issue:9454", "issue:3382|issue:4639", "issue:4185|issue:6949", "issue:11584|issue:7985", @@ -46184,180 +46217,182 @@ "issue:10472|issue:9069", "issue:4340|issue:4398", "issue:1717|issue:954", - "issue:11436|issue:11824" + "issue:11436|issue:11824", + "issue:6258|issue:8922", + "issue:3250|issue:6335" ], - "prepared_review_unit_hash": "31579ea0dd409308243e28e6f8bd0d099e4214ab98dca920a747fcc1feaeff0b", + "prepared_review_unit_hash": "53b004aa764076ea5fb0e5a6f3149fed800b5356128772bef7f0a9b890f3fe10", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34494, - "estimated_input_tokens": 8624, - "estimated_eval_tokens": 17504 + "serialized_chars": 34451, + "estimated_input_tokens": 8613, + "estimated_eval_tokens": 17482 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a grab-bag of unrelated issue reports across different models, pipelines, and failure modes (Flux/SDXL/LoRA loading, compile/offload, dtype mismatches, OOMs, docs/example requests). No pair looks like the same underlying bug/change.", - "confidence": 0.99, + "summary": "This cluster is heterogeneous: it mixes unrelated issues across training, inference, LoRA, torch.compile, model-loading, and feature requests. None of the soft edges look like true duplicates.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No single issue is a good canonical representative because the items do not share one underlying bug or change request. They span many distinct models, code paths, and symptoms.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:6731", - "right": "issue:9729", - "accept": false, - "reason": "Different problems: unsupported pipeline class vs Flux LoRA save-model type error." - }, - { - "left": "issue:8541", - "right": "issue:9454", - "accept": false, - "reason": "Different models and failures: SD3 repository lookup vs CogVideoX model missing." - }, { "left": "issue:3382", "right": "issue:4639", "accept": false, - "reason": "One is multi-GPU OOM in text-to-image training; the other is a dataloader multiprocessing error in instruct-pix2pix SDXL." + "reason": "Different failures: multi-GPU text-to-image OOM vs multiprocessing data-loader error in a different script." }, { "left": "issue:4185", "right": "issue:6949", "accept": false, - "reason": "Both affect image output, but one is ControlNet black images and the other is DPMSolverSinglestep even-step blank images; different code paths." + "reason": "Both are image-quality problems, but one is SDXL ControlNet black images and the other is DPMSolver even-step blank outputs; different code paths and symptoms." }, { "left": "issue:11584", "right": "issue:7985", "accept": false, - "reason": "Both mention torch.compile, but one is group offloading on HunyuanVideo and the other is IP-Adapter incompatibility; not the same bug." + "reason": "Different torch.compile incompatibilities in different pipelines/models; not the same underlying bug." }, { "left": "issue:1692", "right": "issue:5919", "accept": false, - "reason": "Unrelated errors: pickling a lambda vs missing set_lora_layer on Linear." + "reason": "Pickling a lambda and missing set_lora_layer on Linear are unrelated errors." }, { "left": "issue:5442", "right": "issue:6407", "accept": false, - "reason": "Distinct pipeline-add requests: ScaleCrafter vs AnyText." + "reason": "One is a pipeline implementation request, the other is a request for a different model; not duplicates." }, { "left": "issue:6914", "right": "issue:8626", "accept": false, - "reason": "One is a runtime AttributeError in IP-Adapter-FaceID; the other is guidance/docs for multiple IP-adapter images/masks." + "reason": "One is an offload/runtime AttributeError in IP-Adapter-FaceID, the other is documentation/guidance for multiple IP-Adapter inputs." }, { "left": "issue:10795", "right": "issue:11864", "accept": false, - "reason": "Different failures: torch.compile with GGUF from_single_file vs AutoencoderDC.encode missing torch import." + "reason": "GGUF from_single_file torch.compile failure vs AutoencoderDC fullgraph name resolution error; unrelated code paths." }, { "left": "issue:13300", "right": "issue:7355", "accept": false, - "reason": "Different models and symptoms: CPU generation failures for SD3.5/Flux vs Stable Cascade bfloat16/arbitrary-resolution error." + "reason": "Different models and failures: CPU generation errors on SD3.5/FLUX vs Stable Cascade bfloat16 arbitrary-resolution issue." }, { "left": "issue:12104", "right": "issue:9112", "accept": false, - "reason": "Different runtime errors: index-out-of-bounds vs tensor shape mismatch." + "reason": "Index out of bounds and tensor-size mismatch are different runtime bugs." }, { "left": "issue:13079", "right": "issue:13107", "accept": false, - "reason": "OOM in a FLUX.2-klein loop is unrelated to Flux Klein LoRA missing-key errors." + "reason": "CUDA OOM in a loop and LoRA missing-key loading errors are unrelated." }, { "left": "issue:10198", "right": "issue:10540", "accept": false, - "reason": "Both are Flux-related, but one is a weight mismatch and the other is from_single_file loading breaking after Flux Fill support; not the same concrete bug." + "reason": "Flux weight mismatch and Flux Dev from_single_file loading failure are different symptoms and likely different root causes." }, { "left": "issue:7793", "right": "issue:8532", "accept": false, - "reason": "Both involve LoRA weights, but unloading errors and progressive memory growth are different lifecycle bugs." + "reason": "Both involve LoRA weights, but unloading errors and progressive memory growth are different problems; too broad to merge." }, { "left": "issue:1976", "right": "issue:2234", "accept": false, - "reason": "Incorrect tokenizer file lookup vs xformers memory-efficient attention training failure are unrelated." + "reason": "Incorrect tokenizer file lookup and xformers finetuning failure are unrelated." }, { "left": "issue:2163", "right": "issue:7330", "accept": false, - "reason": "Both are training-time issues, but UNet dtype mismatch and FP16 gradient unscale errors arise from different mechanisms." + "reason": "Training dtype mismatch and FP16 unscale error are different failures in different parts of the training loop." }, { "left": "issue:2244", "right": "issue:3468", "accept": false, - "reason": "Dreambooth CPU half-precision inference error is unrelated to missing model_index.json output." + "reason": "Dreambooth CPU half-precision runtime error vs missing model_index.json are unrelated." }, { "left": "issue:11978", "right": "issue:9744", "accept": false, - "reason": "Separate feature requests for different models: Bria 3.2 vs Mochi Video." + "reason": "Both are feature requests, but for different models and different additions." }, { "left": "issue:4702", "right": "issue:8887", "accept": false, - "reason": "New VQ-VAE example request vs SD3 VAE encode bug; not duplicates." + "reason": "VQ-VAE example request and SD3 VAE encode bug are not the same change." }, { "left": "issue:9523", "right": "issue:9857", "accept": false, - "reason": "General error-mapping issue across training scripts vs a specific FLUX controlnet tensor-size mismatch." + "reason": "Broad controlnet training error-mapping issue vs a specific FLUX controlnet embedding-size mismatch; not a duplicate." }, { "left": "issue:10472", "right": "issue:9069", "accept": false, - "reason": "Completely different errors: unexpected lora_bias kwarg vs np.ndarray type mismatch." + "reason": "Unexpected lora_bias keyword and np.ndarray type mismatch are unrelated errors." }, { "left": "issue:4340", "right": "issue:4398", "accept": false, - "reason": "Both are SDXL LoRA related, but one is a train_text_encoder training error and the other is all-black inference images in a different pipeline." + "reason": "SDXL LoRA training error with text encoder and black-image dreambooth inference are different bugs." }, { "left": "issue:1717", "right": "issue:954", "accept": false, - "reason": "Offline cached-model loading failure vs login/load failure are unrelated." + "reason": "Offline cached-model loading and post-login loading failure are different authentication/cache problems." }, { "left": "issue:11436", "right": "issue:11824", "accept": false, - "reason": "Different models and code paths: Hidream dtype mismatch vs FluxKontextPipeline multi-image input error." + "reason": "Hidream dtype mismatch vs FluxKontext multi-image input error are unrelated model-specific bugs." + }, + { + "left": "issue:6258", + "right": "issue:8922", + "accept": false, + "reason": "Forward-chunking quality degradation and quantized-T5 zero-output are different issues." + }, + { + "left": "issue:3250", + "right": "issue:6335", + "accept": false, + "reason": "StableDiffusionKDiffusionPipeline runtime failure and ControlNet output-size control bug are unrelated." } ] }, @@ -46376,8 +46411,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -46385,14 +46420,16 @@ "issue:1092", "issue:10954", "issue:11464", + "issue:11819", + "issue:12120", "issue:12195", "issue:1388", "issue:1717", + "issue:222", "issue:2324", "issue:2822", "issue:2947", "issue:3112", - "issue:3250", "issue:3468", "issue:3623", "issue:3824", @@ -46414,8 +46451,6 @@ "issue:5897", "issue:6093", "issue:6203", - "issue:6258", - "issue:6335", "issue:6839", "issue:7041", "issue:7377", @@ -46426,13 +46461,10 @@ "issue:8668", "issue:8708", "issue:8900", - "issue:8922", "issue:9142", "issue:9270" ], "soft_pairs": [ - "issue:6258|issue:8922", - "issue:3250|issue:6335", "issue:1388|issue:4366", "issue:2822|issue:6093", "issue:5488|issue:8668", @@ -46454,180 +46486,182 @@ "issue:2947|issue:9142", "issue:8567|issue:8708", "issue:10954|issue:11464", - "issue:4183|issue:7377" + "issue:4183|issue:7377", + "issue:11819|issue:12120", + "issue:222|issue:4117" ], - "prepared_review_unit_hash": "88c68c2a0a057692fb33ae41d93b1fad516e28d1d1fdf47d8521d18f1a20b6c7", + "prepared_review_unit_hash": "16f6dba662713b03d61034b9a7b8f9ed2c9871c8af168628a8e9677425a92a78", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 34118, - "estimated_input_tokens": 8530, - "estimated_eval_tokens": 17316 + "serialized_chars": 33472, + "estimated_input_tokens": 8368, + "estimated_eval_tokens": 16992 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose set of Diffusers issues around FLUX/SDXL/other pipeline-loading and training edge cases. The strongest center is the FLUX LoRA load-failure thread, but the soft pairs are only topically similar, not the same concrete bug.", - "confidence": 0.64, - "canonical_issue_reason": "issue:10804 is the clearest representative of the FLUX1.dev LoRA load failure: it has the most detailed repro, general wording, and appears to be the base report that later similar complaints build on.", + "summary": "No PRs are present. Most soft pairs are clearly unrelated by title; the only mild theme overlap is among FLUX/LoRA reports, but they describe different pipelines, model variants, or training vs loading problems, so they should stay separate.", + "confidence": 0.76, + "canonical_issue_reason": "issue:10954 is the most general and current report of the FLUX1.dev LoRA loading failure, so it is the best umbrella duplicate target among the FLUX-related issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10804 is the best overall issue anchor because it is specific enough to the underlying failure mode yet broad enough to represent the FLUX LoRA loading complaints in this set.", + "best_issue_reason": "Open, generic, and directly about the core symptom (FLUX1.dev LoRA fails to load), which makes it the best issue to anchor related follow-ups.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:6258", - "right": "issue:8922", - "accept": false, - "reason": "Both are Stable Video Diffusion model-quality/loading complaints, but one is about forward chunking degrading output and the other is about single-file SD3 loading producing zero/black images; different pipelines and failure modes." - }, - { - "left": "issue:3250", - "right": "issue:6335", - "accept": false, - "reason": "Both involve pipeline execution, but one is about StableDiffusionKDiffusionPipeline startup/runtime problems while the other is specifically about ControlNet img2img output sizing behavior." - }, { "left": "issue:1388", "right": "issue:4366", "accept": false, - "reason": "These are both SD2-related feature requests, but one is about adapting a conversion script for SD 2.0 and the other is about full U-Net text-to-image fine-tuning; not the same bug or change." + "reason": "Different topics: SD 2.0 conversion script vs SD-XL full U-Net fine-tuning." }, { "left": "issue:2822", "right": "issue:6093", "accept": false, - "reason": "Training issues, but one concerns slower convergence for ControlNet training and the other concerns push_to_hub behavior in an SDXL consistency-lora script; unrelated concrete problems." + "reason": "Training convergence complaint vs push_to_hub behavior; not the same bug." }, { "left": "issue:5488", "right": "issue:8668", "accept": false, - "reason": "Both mention generation failures, but one is a demo-code one-image generation problem and the other is SD3 img2img failing on misaligned input dimensions." + "reason": "Demo image-generation failure is unrelated to SD3 img2img input-size alignment." }, { "left": "issue:5226", "right": "issue:5862", "accept": false, - "reason": "One is about load_lora_weights in StableDiffusionXLAdapterPipeline, the other about gradient checkpointing support in Transformer2DModel; different code paths and fixes." + "reason": "LoRA loading in SDXL adapter pipeline vs gradient checkpointing support in Transformer2DModel." }, { "left": "issue:10804", "right": "issue:9270", "accept": false, - "reason": "Both concern FLUX LoRA support, but 10804 is a specific comfy-ui-compatible LoRA load failure while 9270 is a broader incomplete support report for Flux Dev from SimpleTuner; similar subsystem, not the same bug." + "reason": "Both mention FLUX LoRA, but one is a load failure and the other is incomplete training-support for SimpleTuner." }, { "left": "issue:1092", "right": "issue:4520", "accept": false, - "reason": "One is about Euler/attention slicing and the other about LoRA scale not working with attention slicing; related terminology but not the same underlying issue." + "reason": "Euler/attention slicing issue is not the same as LoRA scale with attention slicing." }, { "left": "issue:3112", "right": "issue:5474", "accept": false, - "reason": "A multi-ControlNet inpainting request versus a proposed new Stable Diffusion Reference Only pipeline; different feature requests." + "reason": "Multi-ControlNet inpainting vs a new reference-only pipeline proposal; different changes." }, { "left": "issue:3824", "right": "issue:6203", "accept": false, - "reason": "One is a question about inpaint pipeline variants, the other about SDXL ControlNet image-size conditioning; not the same defect." + "reason": "Legacy-vs-new inpaint pipeline question is unrelated to SDXL ControlNet image-size conditioning." }, { "left": "issue:3937", "right": "issue:4194", "accept": false, - "reason": "ZeroscopeXL behaving like a normal model vs VAE problems with from_single_file; both are loading/model-usage issues, but the concrete failures differ." + "reason": "ZeroscopeXL pipeline behavior vs from_single_file VAE loading; different failures." }, { "left": "issue:4332", "right": "issue:7644", "accept": false, - "reason": "xformers support for Kandinsky2.2 and non-power-of-two Stable Cascade image sizing are unrelated pipeline-specific issues." + "reason": "xformers on Kandinsky2.2 vs Stable Cascade non-power-of-two image handling." }, { "left": "issue:4727", "right": "issue:5545", "accept": false, - "reason": "Auto pipeline local_files_only behavior versus a StableDiffusionInpaintPipeline problem; both involve loading, but not the same bug." + "reason": "local_files_only auto-pipeline breakage is unrelated to an inpaint pipeline problem." }, { "left": "issue:5647", "right": "issue:7041", "accept": false, - "reason": "Both are runtime failures, but one is a StableDiffusionXL inpaint strength==1 edge case and the other is a segmentation fault when num_inference_steps > 1; different failure mechanisms." + "reason": "Different failure modes and conditions in different pipelines; not the same underlying bug." }, { "left": "issue:4340", "right": "issue:5897", "accept": false, - "reason": "SDXL LoRA training with train_text_encoder and a device-mismatch error in train_text_to_image_lora.py are both training problems, but they are not the same code-path or fix." + "reason": "Both are LoRA training errors, but they involve different scripts and different concrete failures." }, { "left": "issue:6839", "right": "issue:8900", "accept": false, - "reason": "Both ask about from_single_file with video/image diffusion models, but one is Stable Video Diffusion loading and the other is SDXL refiner loading; different model families and symptoms." + "reason": "Both use from_single_file, but one is Stable Video Diffusion and the other is SDXL refiner; distinct loading paths." }, { "left": "issue:1717", "right": "issue:3623", "accept": false, - "reason": "Offline HF cache loading versus larger-model loading are both environment/loading complaints, but not the same issue." + "reason": "Offline cached-model loading is unrelated to generic large-model loading issues." }, { "left": "issue:3468", "right": "issue:4117", "accept": false, - "reason": "Dreambooth model_index.json generation and SDXL Dreambooth support are related training topics, but one is an output artifact bug and the other is broader SDXL training behavior." + "reason": "Missing model_index.json is different from SDXL Dreambooth not working." }, { "left": "issue:2324", "right": "issue:817", "accept": false, - "reason": "Both are Dreambooth training issues, but one is instance data not being read and the other is mixed precision not working; distinct bugs." + "reason": "Instance-data loading bug vs mixed-precision issue in Dreambooth; unrelated." }, { "left": "issue:12195", "right": "issue:7681", "accept": false, - "reason": "Both are import/runtime failures, but one is a torch.library custom_op compatibility issue and the other is an OpenSSL/lib X509 flag import error; unrelated causes." + "reason": "PyTorch custom_op incompatibility is unrelated to a lib/X509 import error." }, { "left": "issue:2947", "right": "issue:9142", "accept": false, - "reason": "An old load_attn_procs AttributeError versus Hunyuan DiT multiple-ControlNet loading failure; different models and APIs." + "reason": "Missing load_attn_procs on UNet and multi-ControlNet __module__ error are different code paths." }, { "left": "issue:8567", "right": "issue:8708", "accept": false, - "reason": "SD3 training objective correctness question versus missing vae.config.shift_factor in SD3 Dreambooth examples; related to SD3 but not the same underlying bug." + "reason": "Training objective question for SD3 is not the same as a missing shift_factor in examples." }, { "left": "issue:10954", "right": "issue:11464", "accept": false, - "reason": "Both are FLUX LoRA load failures, but one is a comfy-ui-compatible FLUX1.dev LoRA and the other is a specific FLUX.1-Canny-dev-lora in FluxControlPipeline; different model variants and pipeline context." + "reason": "Both are FLUX LoRA loading reports, but they target different model variants/pipelines and don't clearly share one concrete defect." }, { "left": "issue:4183", "right": "issue:7377", "accept": false, - "reason": "ControlNet guidance_scale<=1 error versus StableCascadeDecoderPipeline num_images_per_prompt>1; both are parameter edge cases, but different pipelines and code paths." + "reason": "ControlNet guidance-scale bug is unrelated to Stable Cascade num_images_per_prompt handling." + }, + { + "left": "issue:11819", + "right": "issue:12120", + "accept": false, + "reason": "Different FLUX LoRA-training questions: image-editing Kontext vs distilled flux-schnell." + }, + { + "left": "issue:222", + "right": "issue:4117", + "accept": false, + "reason": "Generic training runtime error is not the same as SDXL Dreambooth failure." } ] }, @@ -46646,8 +46680,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -46658,24 +46692,22 @@ "issue:11182", "issue:11256", "issue:11489", - "issue:11819", "issue:12022", - "issue:12120", "issue:12535", "issue:1807", "issue:1985", - "issue:222", "issue:2244", "issue:2296", "issue:2392", "issue:2462", "issue:2806", "issue:2844", + "issue:2904", + "issue:3040", "issue:3175", "issue:3774", "issue:3802", "issue:3954", - "issue:4117", "issue:4306", "issue:4480", "issue:4566", @@ -46684,6 +46716,7 @@ "issue:5862", "issue:6107", "issue:6110", + "issue:6324", "issue:6382", "issue:6518", "issue:6548", @@ -46697,12 +46730,11 @@ "issue:8309", "issue:8588", "issue:8950", + "issue:9142", "issue:9156", "issue:9673" ], "soft_pairs": [ - "issue:11819|issue:12120", - "issue:222|issue:4117", "issue:7223|issue:8309", "issue:5577|issue:8228", "issue:5226|issue:6110", @@ -46724,180 +46756,182 @@ "issue:4480|issue:760", "issue:1807|issue:2296", "issue:1985|issue:2462", - "issue:2806|issue:730" + "issue:2806|issue:730", + "issue:2904|issue:9142", + "issue:3040|issue:6324" ], - "prepared_review_unit_hash": "d92bb649cea2da2bcb2d0d0e44a2156c78408e468e1ec797c9de9526f05ddc68", + "prepared_review_unit_hash": "9623ca72643682c7c86154ce99cf572f0bb769191964d1c7a06ec6da4047aff3", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34345, - "estimated_input_tokens": 8587, - "estimated_eval_tokens": 17430 + "serialized_chars": 34515, + "estimated_input_tokens": 8629, + "estimated_eval_tokens": 17514 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly a heterogeneous mix of unrelated bugs, feature requests, and training-script questions. The soft-similarity links are generally too weak for duplicate merging; none look like the same concrete issue or change.", - "confidence": 0.93, - "canonical_issue_reason": "Issue 10210 is the clearest, most concrete bug report in the set and the best representative anchor, but the overall cluster is too mixed to treat any single issue as a true canonical duplicate target.", + "summary": "This set is highly heterogeneous: it mixes unrelated issues about img2img behavior, DreamBooth/LoRA training, model loading, scheduler math, MPS/flash-attn compatibility, and feature requests. None of the soft edges looks like the same underlying bug or change, so all are false positives.", + "confidence": 0.95, + "canonical_issue_reason": "No canonical issue: the items span many unrelated bugs/features and do not form a single duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 10210 is the most actionable end-user bug report: a specific img2img symptom with a clear expected-vs-actual failure mode, unlike the more speculative feature requests and broad training errors.", + "best_issue_reason": "No single issue is a good global representative for deduping because there is no coherent shared problem across the set.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:11819", - "right": "issue:12120", - "accept": false, - "reason": "Both mention FLUX training, but one is image-editing with FLUX Kontext and the other is training a LoRA for distilled FLUX; different use cases and failure modes." - }, - { - "left": "issue:222", - "right": "issue:4117", - "accept": false, - "reason": "Generic runtime error vs SDXL DreamBooth failure; too broad and no shared concrete code path." - }, { "left": "issue:7223", "right": "issue:8309", "accept": false, - "reason": "Latte video-generation integration request and conditional diffusion distillation are unrelated features." + "reason": "One is a video-generation model integration request; the other is a diffusion distillation feature request." }, { "left": "issue:5577", "right": "issue:8228", "accept": false, - "reason": "Both are model-loading errors, but one is ControlNet from_single_file key mismatch and the other is CLIPTextModel loading failure; different bugs." + "reason": "Both involve `from_single_file`, but one is a ControlNet key error and the other is a CLIPTextModel loading failure." }, { "left": "issue:5226", "right": "issue:6110", "accept": false, - "reason": "Both involve load_lora_weights, but one is adapter compatibility and the other is offline loading behavior; not the same defect." + "reason": "Both mention `load_lora_weights`, but one is an SDXL adapter compatibility bug and the other is an offline-loading problem." }, { "left": "issue:10210", "right": "issue:6548", "accept": false, - "reason": "Both concern img2img inputs, but 10210 is FluxImg2ImgPipeline producing near-identical outputs, while 6548 is RGBA images being misread as latents in SD/SDXL img2img." + "reason": "Both are img2img-related, but one reports near-identity output in Flux and the other is an RGBA latent-handling bug in SD/SDXL." }, { "left": "issue:3175", "right": "issue:3774", "accept": false, - "reason": "DreamBooth multi-face training and ZeRO-3 memory savings are different training concerns." + "reason": "Different training problems: multi-face DreamBooth behavior vs. Zero3 memory savings not appearing." }, { "left": "issue:10722", "right": "issue:7644", "accept": false, - "reason": "Prior-preservation tensor shape mismatch vs stable-cascade failure on non-power-of-two images; unrelated errors." + "reason": "A tensor-shape mismatch during DreamBooth training is not the same as Stable Cascade failing on non-power-of-two images." }, { "left": "issue:4566", "right": "issue:8950", "accept": false, - "reason": "SDXL LoRA checkpoint resume issue vs SD3 LoRA training bug; different model families and symptoms." + "reason": "Both are LoRA training issues, but one is checkpoint resume and the other is an SD3 training bug." }, { "left": "issue:6107", "right": "issue:6382", "accept": false, - "reason": "IP-Adapter AWS/runtime compatibility vs a key error during LCM LoRA training and later IP-Adapter use; not the same issue." + "reason": "Both mention IP-Adapter, but one is an environment-specific app load failure and the other is a training-to-inference key mismatch." }, { "left": "issue:10518", "right": "issue:8588", "accept": false, - "reason": "A script bug in sd3_lora_colab versus a missing SD3 conversion script; different problems." + "reason": "One reports a problem in a training script; the other says a conversion script is missing entirely." }, { "left": "issue:11489", "right": "issue:9156", "accept": false, - "reason": "Flux LoRA training error vs Flux LoRA CUDA OOM; one is correctness/runtime failure, the other is resource exhaustion." + "reason": "A generic Flux LoRA training error is not the same as a Flux LoRA CUDA OOM." }, { "left": "issue:2244", "right": "issue:2844", "accept": false, - "reason": "DreamBooth CPU half-precision inference failure vs explicit image-embedding handling in StableUnCLIP; unrelated pipeline paths." + "reason": "Different pipelines and different failures: DreamBooth Half-precision CPU norm vs. explicit image-embedding handling in UnCLIP." }, { "left": "issue:4306", "right": "issue:8305", "accept": false, - "reason": "Reference pipeline list-index error vs scheduler eta range issue; different code paths." + "reason": "A StableDiffusionReferencePipeline index error is unrelated to a scheduler `eta` range issue." }, { "left": "issue:5862", "right": "issue:7589", "accept": false, - "reason": "Gradient checkpointing support in Transformer2DModel vs GLIGEN not working; not the same bug." + "reason": "Gradient-checkpointing support in Transformer2DModel is not the same as GLIGEN not working." }, { "left": "issue:3802", "right": "issue:3954", "accept": false, - "reason": "Checkpoint limit argument not working vs wrong LR scheduler step count; both training examples, but distinct behaviors." + "reason": "One is about checkpoint limiting; the other is about incorrect LR scheduler step counts in multi-GPU examples." }, { "left": "issue:11182", "right": "issue:9673", "accept": false, - "reason": "BF16 variant triggering an unnecessary download vs high CPU usage when loading multiple LoRAs; different loading issues." + "reason": "An unnecessary BF16 download is unrelated to high CPU usage when loading multiple LoRAs." }, { "left": "issue:10967", "right": "issue:11256", "accept": false, - "reason": "Two independent feature requests for different methods/models; not duplicates." + "reason": "These are separate feature requests for different models/methods, not the same integration." }, { "left": "issue:12022", "right": "issue:12535", "accept": false, - "reason": "Flash-attention interface compatibility vs Wan2.2 LoRA loading failure; unrelated." + "reason": "Flash-attn interface incompatibility is unrelated to Wan2.2 Lightning LoRA loading." }, { "left": "issue:2392", "right": "issue:6518", "accept": false, - "reason": "Prompt-embeds validation error in img2img vs NoneType floor-division in LCM distillation; different failures." + "reason": "An img2img prompt_embeds validation error is unrelated to a `NoneType // int` training crash." }, { "left": "issue:4480", "right": "issue:760", "accept": false, - "reason": "Both involve MPS, but one is a kernel crash with float16 and the other is unexpected output in an older release; not the same concrete bug." + "reason": "Both are MPS-related, but one is a kernel crash with float16 and the other is incorrect output in an older version." }, { "left": "issue:1807", "right": "issue:2296", "accept": false, - "reason": "Inference slowdown over generations vs xformers attention not working on T4; unrelated performance/runtime topics." + "reason": "Inference slowdown over time is not the same issue as xformers attention failing on T4." }, { "left": "issue:1985", "right": "issue:2462", "accept": false, - "reason": "Embedding loading UX request vs IterableDataset finetuning support; different features." + "reason": "Loading embeddings and IterableDataset finetuning are different feature requests." }, { "left": "issue:2806", "right": "issue:730", "accept": false, - "reason": "Customized pipeline printing vs DreamBooth learning behavior; no shared code path or symptom." + "reason": "A pipeline-printing customization bug is unrelated to DreamBooth learning differently locally vs Colab." + }, + { + "left": "issue:2904", + "right": "issue:9142", + "accept": false, + "reason": "AudioLDM scheduler initialization failure and Hunyuan Dit multi-ControlNet loading are different code paths." + }, + { + "left": "issue:3040", + "right": "issue:6324", + "accept": false, + "reason": "Both mention `__module__`, but the failures come from different bad input types and different contexts." } ] }, @@ -46916,8 +46950,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -46936,8 +46970,6 @@ "issue:2364", "issue:2548", "issue:2786", - "issue:2904", - "issue:3040", "issue:3645", "issue:3660", "issue:3799", @@ -46952,11 +46984,14 @@ "issue:4892", "issue:4935", "issue:499", + "issue:5150", "issue:5295", "issue:5840", + "issue:5862", "issue:593", "issue:6110", - "issue:6324", + "issue:6196", + "issue:6602", "issue:6801", "issue:6894", "issue:7053", @@ -46965,14 +47000,11 @@ "issue:7392", "issue:7466", "issue:7785", - "issue:9142", "issue:9187", "issue:9267", "issue:9421" ], "soft_pairs": [ - "issue:2904|issue:9142", - "issue:3040|issue:6324", "issue:3799|issue:4726", "issue:4488|issue:4667", "issue:5840|issue:6894", @@ -46994,180 +47026,182 @@ "issue:1061|issue:593", "issue:10772|issue:11547", "issue:1264|issue:3645", - "issue:3660|issue:4724" + "issue:3660|issue:4724", + "issue:5150|issue:5862", + "issue:6196|issue:6602" ], - "prepared_review_unit_hash": "a4234551e3769dd0043888328d22bf9843ffcd73e9ac610a5f4ce1ca8599249b", + "prepared_review_unit_hash": "6f85aad33b48bd49fae1750d45fb681163addeb8237a485a32de106162fed483", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34119, - "estimated_input_tokens": 8530, - "estimated_eval_tokens": 17316 + "serialized_chars": 33994, + "estimated_input_tokens": 8499, + "estimated_eval_tokens": 17254 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster looks heterogeneous: it mixes unrelated model-loading, training, scheduler, and feature-request issues that only share broad vocabulary. None of the soft edges look like true duplicate pairs, so there is no clear canonical issue for the cluster.", - "confidence": 0.94, - "canonical_issue_reason": "No single issue meaningfully represents the cluster; the items span multiple unrelated bugs/features rather than one duplicate family.", - "canonical_pr_reason": "No PRs are present in this cluster.", - "best_issue_reason": "No issue is a strong global representative because the cluster is not centered on one underlying bug or workflow.", - "best_pr_reason": "No PRs are present, so none are suitable as a global PR representative.", + "summary": "This cluster is a weakly related grab-bag of mostly unrelated diffusers issues across loading, training, scheduler, ControlNet, LoRA, and pipeline support. None of the proposed soft pairs look like the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue cleanly represents the cluster: the items cover many different code paths and features, so any one issue would be a misleading canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "There is no strong global best issue here; the cluster is not cohesive enough for one issue to serve as a reliable representative.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:2904", - "right": "issue:9142", - "accept": false, - "reason": "Different pipelines, models, and failures: AudioLDM scheduler NoneType vs Hunyuan DiT multiple-controlnet loading error." - }, - { - "left": "issue:3040", - "right": "issue:6324", - "accept": false, - "reason": "Same general error class, but the concrete bug differs: list-type vs bool-type __module__ failure in different contexts." - }, { "left": "issue:3799", "right": "issue:4726", "accept": false, - "reason": "Unrelated type-hint issues: scheduler return type inference vs callback annotation in StableDiffusionPipeline." + "reason": "Different problems: scheduler type inference vs a callback typing issue in a pipeline call signature." }, { "left": "issue:4488", "right": "issue:4667", "accept": false, - "reason": "Both concern single-file loading, but one is SDXL ckpt import behavior and the other is StableDiffusionPipeline v0.20.0 failure; not the same concrete bug." + "reason": "Both mention from_single_file, but one is SDXL checkpoint loading via ckpt conversion and the other is a generic v0.20.0 failure for StableDiffusionPipeline; not the same bug." }, { "left": "issue:5840", "right": "issue:6894", "accept": false, - "reason": "Different symptoms and environments: checkpoint resume freezing in SDXL LoRA training vs A1111 LoRA compatibility regression." + "reason": "Both concern SDXL LoRA training, but one is resume/freezing behavior and the other is A1111 compatibility after export; different failure modes." }, { "left": "issue:7303", "right": "issue:7392", "accept": false, - "reason": "Scheduler zero_snr support and SDXL aspect-ratio training are unrelated problems." + "reason": "Scheduler zero-SNR support is unrelated to SDXL multi-aspect-ratio training." }, { "left": "issue:2364", "right": "issue:499", "accept": false, - "reason": "Both mention training scripts, but they are different scripts with different failure modes; too broad to treat as one bug." + "reason": "Generic training-script failure versus unconditional training script not working; too broad and not clearly the same concrete bug." }, { "left": "issue:4868", "right": "issue:9267", "accept": false, - "reason": "Different attribute errors in different code paths: missing __getattr__ vs missing rope attribute in transformer_flux." + "reason": "Different attribute errors in different components: a super().__getattr__ issue versus missing rope in FLUX transformer code." }, { "left": "issue:6801", "right": "issue:7785", "accept": false, - "reason": "Scheduler loading regression and slow first inference are different performance/compatibility issues." + "reason": "Scheduler loading error versus slow first inference; different symptoms and code paths." }, { "left": "issue:6110", "right": "issue:9187", "accept": false, - "reason": "Offline LoRA loading and SDXL IP-Adapter image encoder handling are separate features/problems." + "reason": "Offline LoRA loading and SDXL IP-Adapter image_encoder_folder behavior are unrelated." }, { "left": "issue:2326", "right": "issue:2548", "accept": false, - "reason": "Both involve LoRA saving/export, but one is conversion to safetensors/ckpt and the other is training save failure; not clearly the same bug." + "reason": "Both involve LoRA, but one is converting LoRA weights to safetensors/ckpt and the other is saving finetuning LoRA; not the same bug." }, { "left": "issue:4229", "right": "issue:4639", "accept": false, - "reason": "Both are training failures, but they affect different scripts and different subsystems (accelerate multi-GPU vs dataloader multiprocessing)." + "reason": "Both are multi-GPU training errors, but they affect different scripts and failure modes." }, { "left": "issue:10180", "right": "issue:12078", "accept": false, - "reason": "Different Flux Control LoRA topics: loading multiple LoRAs vs an example validation-input problem." + "reason": "Different feature areas: loading multiple LoRAs with Flux Control LoRA versus a Flux Control finetuning validation-input example problem." }, { "left": "issue:4206", "right": "issue:7466", "accept": false, - "reason": "Both are controlnet-related, but one is generic SDXL controlnet training breakage and the other is torch.compile failure in multi-controlnet inpaint." + "reason": "ControlNet training with SDXL versus torch.compile failing for multi-ControlNet in SDXL inpaint; related subsystem, but not the same bug." }, { "left": "issue:7053", "right": "issue:7185", "accept": false, - "reason": "Class-conditioning behavior in UNet3DConditionModel is unrelated to PyTorch version support." + "reason": "Unet3D class conditioning bug is unrelated to PyTorch 2.2 support concerns." }, { "left": "issue:12012", "right": "issue:12499", "accept": false, - "reason": "Both are Wan 2.2 issues, but one is apply_first_block_cache context handling and the other is I2V condition shape mismatch." + "reason": "apply_first_block_cache context-state error is unrelated to Wan 2.2 I2V condition shape mismatch." }, { "left": "issue:10040", "right": "issue:9421", "accept": false, - "reason": "Grey output images and inability to access a specific inpainting model are unrelated." + "reason": "Grey images in SD3 and inability to access an inpainting model are different issues." }, { "left": "issue:2786", "right": "issue:4935", "accept": false, - "reason": "Depth2Img numpy/transformers compatibility and a LoRA-layer usage question are different concerns." + "reason": "A numpy/transformers pipeline compatibility bug is unrelated to a question about AttnProcsLayers usage in an SDXL LoRA training script." }, { "left": "issue:11178", "right": "issue:12029", "accept": false, - "reason": "Both are model loading problems, but they involve different models and likely different loader failures." + "reason": "MistoLine ControlNet loading and X-Omni-En model loading are different model-specific loading issues." }, { "left": "issue:4892", "right": "issue:5295", "accept": false, - "reason": "Distinct video-generation/editing projects with no shared bug." + "reason": "TokenFlow video editing pipeline and LAVIE video generation are different pipeline proposals, not duplicates." }, { "left": "issue:1061", "right": "issue:593", "accept": false, - "reason": "Docs update for textual inversion on SD1.5 is unrelated to repeated-image generation behavior." + "reason": "Textual inversion doc update versus repeated images in high-res stable diffusion; unrelated." }, { "left": "issue:10772", "right": "issue:11547", "accept": false, - "reason": "Both request ControlNet support, but for different base models and implementations; not the same change." + "reason": "Both request ControlNet support, but for different model families (Sana vs Flux fill dev); not the same concrete change." }, { "left": "issue:1264", "right": "issue:3645", "accept": false, - "reason": "Hard-coded scheduler checkpoint and a controlnet inpaint training/help request are unrelated." + "reason": "Hard-coded scheduler issue in dreambooth training is unrelated to a ControlNet inpaint training/help question." }, { "left": "issue:3660", "right": "issue:4724", "accept": false, - "reason": "torch.compile/Lora/xformers incompatibility is a different failure mode from SDXL img2img behavior with a checkpoint." + "reason": "torch.compile with LoRA/xformers is unrelated to SDXL img2img behavior with a base checkpoint." + }, + { + "left": "issue:5150", + "right": "issue:5862", + "accept": false, + "reason": "from_single_file missing for UpscalePipeline is unrelated to Transformer2DModel gradient checkpointing support." + }, + { + "left": "issue:6196", + "right": "issue:6602", + "accept": false, + "reason": "Two different loading/saving failures: ControlNetXS Depth SDXL loading versus loading after fuse_lora/save_pretrained." } ] }, @@ -47186,8 +47220,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -47209,24 +47243,23 @@ "issue:2599", "issue:2618", "issue:2707", + "issue:3124", "issue:3221", "issue:3325", "issue:3391", "issue:3420", "issue:347", "issue:3660", + "issue:4080", "issue:4185", "issue:4325", "issue:4664", "issue:5103", - "issue:5150", "issue:5379", "issue:5442", "issue:5544", - "issue:5862", - "issue:6196", + "issue:5905", "issue:6492", - "issue:6602", "issue:6722", "issue:6911", "issue:8532", @@ -47241,8 +47274,6 @@ "issue:9997" ], "soft_pairs": [ - "issue:5150|issue:5862", - "issue:6196|issue:6602", "issue:8543|issue:8726", "issue:4664|issue:5544", "issue:1271|issue:347", @@ -47264,180 +47295,182 @@ "issue:11700|issue:9279", "issue:1350|issue:5442", "issue:1616|issue:5379", - "issue:2618|issue:3391" + "issue:2618|issue:3391", + "issue:3124|issue:6911", + "issue:4080|issue:5905" ], - "prepared_review_unit_hash": "d91d10237e4a713e08feb2b2447280af9e33ff742c967bc3a39b42a7cf22e60a", + "prepared_review_unit_hash": "139feee643d0f44638d1e32d1348622dc3311a7c6fe5e52c6768cacf7d6599ff", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33869, - "estimated_input_tokens": 8468, - "estimated_eval_tokens": 17192 + "serialized_chars": 33237, + "estimated_input_tokens": 8310, + "estimated_eval_tokens": 16876 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a mixed bag of unrelated bug reports and feature requests; the soft pairs are mostly superficial similarities around pipelines, LoRAs, or black-image symptoms rather than the same underlying defect.", - "confidence": 0.86, - "canonical_issue_reason": "No single canonical issue fits the whole set because the items span multiple unrelated subsystems. If a representative bug must be chosen, issue 12689 is the most concrete and actionable loading-failure report.", + "summary": "This cluster is mostly a loose topical bundle of diffusion, LoRA, pipeline, and training issues rather than true duplicates. None of the soft pairs look like the same underlying bug/change, so all are rejected.", + "confidence": 0.78, + "canonical_issue_reason": "issue:12689 is the most concrete, bug-focused report in the set: a clear loading failure for LoRA adapter weights from checkpoints with an actionable symptom statement.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 12689 is the best standalone representative: it describes a concrete LoRA-loading failure with direct user impact and clear actionability.", + "best_issue_reason": "issue:12689 is the best representative issue overall because it is specific, reproducible, and about a core failure mode (LoRA checkpoint loading) rather than a broad feature request or vague symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:5150", - "right": "issue:5862", - "accept": false, - "reason": "Different problems: missing from_single_file support vs gradient checkpointing support." - }, - { - "left": "issue:6196", - "right": "issue:6602", - "accept": false, - "reason": "Both are loading-related, but they hit different code paths and different model/component failures." - }, { "left": "issue:8543", "right": "issue:8726", "accept": false, - "reason": "Same broad SD3 dreambooth area, but one is gradient checkpointing and the other is text-encoder errors; not the same bug." + "reason": "Both involve SD3 text-encoder/training problems, but one is a gradient-checkpointing failure and the other is text encoder errors in training scripts; different concrete bugs." }, { "left": "issue:4664", "right": "issue:5544", "accept": false, - "reason": "Both mention SDXL, but one is LoRA loading and the other is FreeU disable behavior." + "reason": "Both mention SDXL/LoRA-related pipeline behavior, but the symptoms differ: one is a LoRA loading inconsistency, the other is a FreeU disable error." }, { "left": "issue:1271", "right": "issue:347", "accept": false, - "reason": "One is a Dreambooth notebook dependency-install problem; the other is a huggingface-hub version resolution issue." + "reason": "Both are dependency/install issues, but they concern different environments and failure modes; not the same underlying packaging bug." }, { "left": "issue:2707", "right": "issue:3325", "accept": false, - "reason": "Unrelated scripts and failures: a syntax error in conversion code vs a CUDA runtime error during Dreambooth training." + "reason": "A syntax error in a conversion script and a CUDA runtime error during DreamBooth training are unrelated code paths." }, { "left": "issue:4185", "right": "issue:6492", "accept": false, - "reason": "Both describe black images, but the reported contexts and likely causes are too different to treat as the same bug." + "reason": "Both mention black images, but one is specifically SDXL ControlNet inference while the other is a generic image-generation symptom with unknown cause." }, { "left": "issue:3221", "right": "issue:3660", "accept": false, - "reason": "Different APIs and failure modes: load_attn_procs regression vs torch.compile interactions with LoRA/xformers." + "reason": "These are different problems: attn processor loading in one version versus torch.compile compatibility with LoRA/xformers." }, { "left": "issue:3420", "right": "issue:6911", "accept": false, - "reason": "These concern different models and bugs; one is ControlNet inpainting integration, the other class-label handling in a 3D UNet." + "reason": "Different models and failures: a ControlNet inpaint pipeline issue versus class-label handling in a 3D UNet." }, { "left": "issue:8532", "right": "issue:9914", "accept": false, - "reason": "Both involve LoRA loading, but one is a memory increase while the other is a loading/compatibility error for a specific Flux/Xlabs format." + "reason": "Both involve LoRA loading, but one is memory growth during loading and the other is a specific Flux/Xlabs loading error; not the same bug." }, { "left": "issue:12535", "right": "issue:12689", "accept": false, - "reason": "Both are LoRA-loading issues, but they refer to different model families and different failure modes, not one shared bug." + "reason": "Both are LoRA-loading complaints, but they target different model families and different failure modes, so they should not be merged." }, { "left": "issue:1197", "right": "issue:905", "accept": false, - "reason": "Separate community-pipeline feature requests with no shared concrete change or bug." + "reason": "These are separate community pipeline requests with different use cases; no shared underlying bug or implementation change." }, { "left": "issue:234", "right": "issue:2549", "accept": false, - "reason": "Different issues: half-precision runtime behavior vs an SD2-based UNet model not working." + "reason": "Half-precision instability and an SD 2.0 UNet failure are not the same issue, even if both relate to Stable Diffusion models." }, { "left": "issue:4325", "right": "issue:5103", "accept": false, - "reason": "Both mention black images, but one is SDXL casting/runtime warning and the other is zero-SNR DDIM behavior; not the same root cause." + "reason": "Both produce black images, but the reported triggers differ (SDXL warning/cast path vs zero-SNR DDIM), so the underlying cause is not clearly the same." }, { "left": "issue:6722", "right": "issue:8555", "accept": false, - "reason": "Related to text encoders, but one is using SigLIP as an encoder and the other is SD3 FP8/training availability." + "reason": "A feature request to use SigLIP as text encoder is not the same as SD3 text-encoder training availability/fp8 support." }, { "left": "issue:9134", "right": "issue:9997", "accept": false, - "reason": "One reports a compatibility issue with LoRAs trained by SimpleTuner; the other is a feature request for supporting new Flux LoRAs." + "reason": "Both mention Flux LoRAs, but one is a specific incompatibility with SimpleTuner-trained LoRAs and the other is a broader support request for new LoRAs." }, { "left": "issue:1100", "right": "issue:1544", "accept": false, - "reason": "Different AttributeErrors in different components: CLIPTextTransformer.device vs DDPMScheduler.get_velocity." + "reason": "Different attribute errors in different components: CLIPTextTransformer device access versus DDPMScheduler get_velocity." }, { "left": "issue:12195", "right": "issue:13142", "accept": false, - "reason": "Different library regressions affecting different components: AutoencoderKL/torch.library vs PRXPipeline/transformers." + "reason": "Both are version-compatibility regressions, but they involve different packages and distinct failing APIs." }, { "left": "issue:2465", "right": "issue:2599", "accept": false, - "reason": "A recipe request is not the same as a multi-subject training quality bug." + "reason": "A request for a reproducible training recipe is not the same as a report that multi-subject DreamBooth training yields worse quality." }, { "left": "issue:12575", "right": "issue:9953", "accept": false, - "reason": "Different problems: attention backend checks error vs moving a quantized component to CUDA." + "reason": "These concern different failures: attention-backend checks in Chroma versus moving a quantized component to CUDA." }, { "left": "issue:11700", "right": "issue:9279", "accept": false, - "reason": "Both are feature requests around loading, but they target different capabilities and code paths." + "reason": "One is a Pruna integration request for from_pretrained/pipeline loading, the other is dynamic LoRA loading with torch.compile; separate feature changes." }, { "left": "issue:1350", "right": "issue:5442", "accept": false, - "reason": "Unrelated community-pipeline requests." + "reason": "Completely different domains: object detection (DiffusionDet) versus higher-resolution image generation (ScaleCrafter)." }, { "left": "issue:1616", "right": "issue:5379", "accept": false, - "reason": "Different pipeline feature requests with no shared bug or change." + "reason": "Different community pipelines with no shared bug or implementation target." }, { "left": "issue:2618", "right": "issue:3391", "accept": false, - "reason": "Different symptom and pipeline: ControlNet float-tensor image input vs DeepFloyd super-resolution overexposure." + "reason": "A ControlNet image-input type issue and a DeepFloyd overexposure bug are unrelated pipeline problems." + }, + { + "left": "issue:3124", + "right": "issue:6911", + "accept": false, + "reason": "Stable Diffusion model conversion failure and a 3D UNet class-label bug are different code paths entirely." + }, + { + "left": "issue:4080", + "right": "issue:5905", + "accept": false, + "reason": "An SDXL inpainting pipeline request and a diffusion-GAN model request are separate feature additions, not duplicates." } ] }, @@ -47456,8 +47489,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -47469,13 +47502,13 @@ "issue:1363", "issue:1616", "issue:2163", + "issue:2192", "issue:2250", "issue:2707", "issue:2838", - "issue:3124", + "issue:3177", "issue:3648", "issue:3881", - "issue:4080", "issue:4520", "issue:4835", "issue:5233", @@ -47485,18 +47518,18 @@ "issue:5753", "issue:5783", "issue:5862", - "issue:5905", "issue:5966", "issue:6323", "issue:6351", "issue:6427", "issue:6549", "issue:6561", - "issue:6911", + "issue:6791", "issue:6934", "issue:6974", "issue:7549", "issue:7592", + "issue:7619", "issue:8233", "issue:8493", "issue:8518", @@ -47511,8 +47544,6 @@ "issue:9979" ], "soft_pairs": [ - "issue:3124|issue:6911", - "issue:4080|issue:5905", "issue:7592|issue:8493", "issue:11584|issue:8518", "issue:6549|issue:6561", @@ -47534,180 +47565,182 @@ "issue:4520|issue:6974", "issue:5233|issue:9084", "issue:5578|issue:5862", - "issue:5783|issue:7549" + "issue:5783|issue:7549", + "issue:2192|issue:6791", + "issue:3177|issue:7619" ], - "prepared_review_unit_hash": "d2b882c5fa900984fbce4d50b3e3935d887b69802ee177f5ac285a8691628a10", + "prepared_review_unit_hash": "ab8a689ba648803f9f46af465d02c934ebbd0f4633a8776e6a4802bbf9bb1b01", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34129, - "estimated_input_tokens": 8533, - "estimated_eval_tokens": 17322 + "serialized_chars": 34202, + "estimated_input_tokens": 8551, + "estimated_eval_tokens": 17358 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: the items span many unrelated bugs across loading, training, ControlNet, schedulers, and model conversion, with only superficial title overlap. All soft edges should be rejected.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue cleanly covers the cluster; the only reasonable representative is issue 5753, but the set is too heterogeneous to treat as one bug.", + "summary": "This is not a true duplicate cluster: the issues span many unrelated bugs/features across model loading, training scripts, ControlNet/Flux/Wan/SD3, compilation, and file/path errors. The soft-link suggestions are mostly driven by superficial title similarity, not the same underlying problem.", + "confidence": 0.94, + "canonical_issue_reason": "No single issue is a safe canonical target; the items describe different code paths and failure modes, so collapsing them would be incorrect.", "canonical_pr_reason": null, - "best_issue_reason": "issue 5753 is the closest representative of the recurring pipeline/adapter interaction theme, but it still does not match most other items.", + "best_issue_reason": "None is a strong global representative because the set is heterogeneous rather than a near-duplicate cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:3124", - "right": "issue:6911", - "accept": false, - "reason": "Both mention model conversion/model classes, but one is SD model conversion and the other is 3D class-label handling; different code paths and bugs." - }, - { - "left": "issue:4080", - "right": "issue:5905", - "accept": false, - "reason": "SDXL inpainting pipeline request vs diffusion-GAN model request; unrelated feature asks." - }, { "left": "issue:7592", "right": "issue:8493", "accept": false, - "reason": "DoRA state_dict loading issue vs SD3Transformer2DModel state_dict mismatch; both are loading bugs but for different mechanisms and models." + "reason": "Both involve state_dict loading, but they target different models and failure modes (DoRA key coverage vs SD3Transformer2DModel mismatch)." }, { "left": "issue:11584", "right": "issue:8518", "accept": false, - "reason": "torch.compile with group offloading on HunyuanVideo vs SD3 torch.compile unsupported call; different models and failure modes." + "reason": "Both mention torch.compile, but one is a group offloading incompatibility and the other is an SD3 dynamo unsupported-call error." }, { "left": "issue:6549", "right": "issue:6561", "accept": false, - "reason": "LoRA alpha warning vs missing IP-Adapter config.json; both loading-related, but not the same defect." + "reason": "Different LoRA/IP-Adapter loading problems: one is a lora_weight logging issue, the other is a missing config.json file path error." }, { "left": "issue:8732", "right": "issue:8909", "accept": false, - "reason": "Checkpoint saving issue in advanced training vs saving text_encoder_ti in dreambooth LoRA; similar area, different save logic and reproduction." + "reason": "Both are training/save related, but one is checkpoint saving in SD1.5 advanced training and the other is text_encoder_ti saving in a Dreambooth LoRA script." }, { "left": "issue:4835", "right": "issue:5606", "accept": false, - "reason": "MultiControlNet img2img bug vs LoRA alpha handling bug; only broad pipeline-level similarity." + "reason": "ControlNet img2img with MultiControlNetModel is unrelated to LoRA loading with no alpha value." }, { "left": "issue:6351", "right": "issue:8832", "accept": false, - "reason": "DDP missing disable_adapters in SDXL LCM distillation vs a generic pipe(prompt) crash; not the same code path." + "reason": "One is a DDP adapter attribute error in a training script; the other is a generic runtime crash on pipeline image generation." }, { "left": "issue:8958", "right": "issue:9534", "accept": false, - "reason": "PyTorch version compatibility error vs FluxMultiControlNetModel config attribute bug; unrelated runtime issues." + "reason": "These are different PyTorch/Flux problems: a torch.library API mismatch versus FluxMultiControlNetModel missing config." }, { "left": "issue:10928", "right": "issue:8233", "accept": false, - "reason": "Missing fp16 weight file while loading a UNet vs depth pipeline fp16 error; both mention fp16 but different models and causes." + "reason": "Both mention fp16/model handling, but one is a missing model file and the other is a depth pipeline fp16 error." }, { "left": "issue:2163", "right": "issue:2707", "accept": false, - "reason": "UNet dtype mismatch during training vs a syntax error in a conversion script; not a duplicate bug." + "reason": "Different training/conversion failures: UNet dtype mismatch during text-to-image training vs a syntax error in a conversion script." }, { "left": "issue:6427", "right": "issue:6934", "accept": false, - "reason": "Both are conversion problems, but one is diffusers-to-ckpt and the other diffusers-to-original-SDXL; different converters and failures." + "reason": "Both are conversion bugs, but they refer to different converters and outputs (.ckpt vs original SDXL format) without evidence of the same defect." }, { "left": "issue:11247", "right": "issue:9979", "accept": false, - "reason": "Flux transformer multi-GPU/controlnet distribution vs flux img2img controlnet channel mismatch; both Flux+ControlNet, but different issues." + "reason": "Both involve Flux controlnet/multi-GPU-ish topics, but one is transformer distribution across GPUs and the other is an img2img controlnet channels error." }, { "left": "issue:3648", "right": "issue:3881", "accept": false, - "reason": "Both are training advice questions, not the same bug or change." + "reason": "These are generic how-to questions about training size/epochs, not the same concrete bug." }, { "left": "issue:5753", "right": "issue:6323", "accept": false, - "reason": "SDXL ControlNet load_lora_weights failure vs TensorRT img2img guidance_scale attribute error; different pipelines and error surfaces." + "reason": "Different ControlNet/Lora integration issues: one is load_lora_weights failure, the other is a TensorRT img2img guidance_scale attribute problem." }, { "left": "issue:2250", "right": "issue:5509", "accept": false, - "reason": "Dreambooth inference not generating images vs batch-size mismatch for np.ndarray/cv2 image input; unrelated inference bugs." + "reason": "Dreambooth prompt-generation failure is unrelated to image batch-size validation for np.ndarray/cv2 inputs." }, { "left": "issue:1363", "right": "issue:1616", "accept": false, - "reason": "ONNX LPW scheduler error vs community LPW pipeline release/support issue; related topic, but not the same defect." + "reason": "Both concern LPW/community pipelines, but one is an ONNX scheduler error and the other is a general Diffusers version/support question." }, { "left": "issue:9422", "right": "issue:9927", "accept": false, - "reason": "ControlNet SDXL training NaNs vs HeaderTooLarge during sdv3 controlnet training; different failure modes." + "reason": "Different training instability bugs: one is NaNs during custom ControlNet SDXL backprop, the other is HeaderTooLarge when training controlnet with SD3." }, { "left": "issue:11351", "right": "issue:13258", "accept": false, - "reason": "Video processor dtype question vs missing Wan2.2 TI2V model entry; not the same bug." + "reason": "Both mention Wan video models, but one is a datatype question for a processor and the other is a missing Wan2.2 TI2V model/package issue." }, { "left": "issue:2838", "right": "issue:5966", "accept": false, - "reason": "Request to support a controlnet pipeline variant vs request for ControlNet on SD 2.1; both are feature requests, but distinct models/use cases." + "reason": "One asks for ControlNet pipeline behavior; the other is a request for Stable Diffusion 2.1 ControlNet support." }, { "left": "issue:4520", "right": "issue:6974", "accept": false, - "reason": "LoRA scale with attention slicing vs unloading textual inversion not removing tokenizer_2/text_encoder_2; different adapter-management bugs." + "reason": "Both are adapter-related, but one is a LoRA scale + attention slicing bug and the other is textual inversion unload missing tokenizer/text encoder cleanup." }, { "left": "issue:5233", "right": "issue:9084", "accept": false, - "reason": "Callback step value issue vs SD3 inpaint EulerA scheduler ordering bug; different API surfaces and fixes." + "reason": "Different scheduler/callback semantics: wrong callback step value vs SD3 inpaint EulerA scale_model_input ordering." }, { "left": "issue:5578", "right": "issue:5862", "accept": false, - "reason": "Missing model_index.json when loading a random module vs lack of gradient checkpointing support in Transformer2DModel; not duplicates." + "reason": "A missing model_index.json load error is unrelated to Transformer2DModel gradient checkpointing support." }, { "left": "issue:5783", "right": "issue:7549", "accept": false, - "reason": "EMA not working in train_text_to_image_sdxl vs potential UNet EMA bug; both mention EMA, but one is a training script behavior and the other a suspected model-level bug." + "reason": "Both are SDXL training issues, but one is EMA not working in the training script and the other is a suspected UNet EMA bug; not clearly the same defect." + }, + { + "left": "issue:2192", + "right": "issue:6791", + "accept": false, + "reason": "Different training-script issues: missing CLI arguments in one script vs an SDXL training script that cannot run." + }, + { + "left": "issue:3177", + "right": "issue:7619", + "accept": false, + "reason": "DeepSpeed stage-3 offloading in Dreambooth is unrelated to a channel-number mismatch in advanced Dreambooth training." } ] }, @@ -47726,27 +47759,28 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10026", "issue:10550", + "issue:11133", "issue:11447", "issue:11702", "issue:1205", + "issue:12359", "issue:1249", "issue:1552", "issue:1974", "issue:2091", - "issue:2192", "issue:2462", "issue:255", "issue:2581", "issue:2991", "issue:3019", "issue:3103", - "issue:3177", "issue:3191", "issue:3445", "issue:3540", @@ -47763,7 +47797,6 @@ "issue:6157", "issue:6382", "issue:6407", - "issue:6791", "issue:6811", "issue:6949", "issue:7033", @@ -47772,17 +47805,15 @@ "issue:730", "issue:761", "issue:7616", - "issue:7619", "issue:7948", "issue:8732", + "issue:8738", "issue:8873", "issue:9454", "issue:9546", "issue:9763" ], "soft_pairs": [ - "issue:2192|issue:6791", - "issue:3177|issue:7619", "issue:5687|issue:7616", "issue:2091|issue:8732", "issue:1205|issue:4974", @@ -47805,186 +47836,188 @@ "issue:2581|issue:4000", "issue:4892|issue:5889", "issue:3191|issue:3445", - "issue:4993|issue:5023" + "issue:4993|issue:5023", + "issue:10026|issue:8738", + "issue:11133|issue:12359" ], - "prepared_review_unit_hash": "bed2962fdfb6394f4e047bbe4c7f2fbf88a52407415deab9be03a7066ef2f37a", + "prepared_review_unit_hash": "d3f74e51f642801bf80d5f4d59b5e49d9571652274ea83024ba44c94eeb9244b", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34193, - "estimated_input_tokens": 8549, - "estimated_eval_tokens": 17354 + "serialized_chars": 34226, + "estimated_input_tokens": 8557, + "estimated_eval_tokens": 17370 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a keyword-similarity bucket, not a duplicate set. The issues span LoRA loading, checkpoint/offline loading, training scripts, conversion, scheduler behavior, and unrelated pipeline bugs; none appear to describe the same concrete bug/change.", - "confidence": 0.29, - "canonical_issue_reason": null, + "summary": "The cluster is largely heterogeneous: it mixes unrelated model-loading, scheduler, documentation, and training bugs. The only loose thematic concentration is around Flux/ControlNet training, but the specific failures differ enough that none of the soft pairs should be merged as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "issue:9546 is the best umbrella issue for the Flux ControlNet train-example memory-related reports in this set; it is broader and more representative than the more specific token-truncation or validation/OOM variants.", "canonical_pr_reason": null, - "best_issue_reason": "issue:3191 is the broadest representative of the LoRA/DreamBooth training failures in the set, but it is not a safe canonical duplicate target because the cluster is heterogeneous.", + "best_issue_reason": "issue:9546 has the clearest cluster-central topic and the most discussion among the Flux ControlNet training issues, so it is the best candidate to anchor that subtopic even though the set is not a true duplicate cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:2192", - "right": "issue:6791", - "accept": false, - "reason": "Both concern training script arguments/running, but one is missing CLI flags in a LoRA script and the other is a separate SDXL script runtime failure; no same code-path bug." - }, - { - "left": "issue:3177", - "right": "issue:7619", - "accept": false, - "reason": "Both are DreamBooth training problems, but one is DeepSpeed stage-3 offload breakage and the other is a channel mismatch error; different failure modes." - }, { "left": "issue:5687", "right": "issue:7616", "accept": false, - "reason": "Both involve generation artifacts, but Karras-sigma duplicate timesteps in img2img is unrelated to low-denoise artifacts with Playground 2.5." + "reason": "Both mention bad img2img behavior, but one is a scheduler duplicate-timestep bug and the other is low-denoise artifacts in a specific model; not the same concrete failure." }, { "left": "issue:2091", "right": "issue:8732", "accept": false, - "reason": "Both involve checkpoint handling, but one is multi-GPU resume training and the other is checkpoint saving; distinct workflows and bugs." + "reason": "Checkpoint-resume failures and checkpoint-saving issues are related only broadly; they describe different training paths and different breakages." }, { "left": "issue:1205", "right": "issue:4974", "accept": false, - "reason": "Both are conversion-related, but one is SD-to-ONNX conversion and the other is safetensor-to-diffusers conversion; different converter paths." + "reason": "These are opposite conversion directions (SD to ONNX vs safetensors to diffusers) and are not the same change or bug." }, { "left": "issue:2462", "right": "issue:5966", "accept": false, - "reason": "Both are about training/support in the ecosystem, but one is IterableDataset finetuning support and the other is ControlNet for SD 2.1; not the same issue." + "reason": "IterableDataset training support and ControlNet SD2.1 support are unrelated feature/usage topics." }, { "left": "issue:6811", "right": "issue:9454", "accept": false, - "reason": "Both are model-not-found / download failures, but they target different models and URLs with no shared underlying bug." + "reason": "Both are 'model not found' reports, but for different models and likely different repository/config issues." }, { "left": "issue:7079", "right": "issue:8873", "accept": false, - "reason": "One is a Transformer2DModel patch/norm error; the other is xformers performance regression. Different problems." + "reason": "Transformer patch/norm errors and xformers performance are distinct problems with no shared code-path evidence." }, { "left": "issue:1249", "right": "issue:1974", "accept": false, - "reason": "Both hit schedulers, but one is an OnnxPipeline ndarray/device error and the other is generator-list reproducibility in Euler schedulers." + "reason": "ONNX pipeline ndarray device errors and Euler scheduler generator-list handling are unrelated." }, { "left": "issue:11447", "right": "issue:8873", "accept": false, - "reason": "Offline checkpoint shard loading and xformers slowdown are unrelated issues." + "reason": "Offline checkpoint shard loading and xformers speed regressions are unrelated subsystems and failures." }, { "left": "issue:255", "right": "issue:3103", "accept": false, - "reason": "Both concern dimensions, but one is a StableDiffusionPipeline unet block error and the other is inpainting height/width behavior; not the same bug." + "reason": "Both concern image dimensions, but one is a Stable Diffusion unet_blocks error and the other is an inpainting height/width topic; not enough to be the same bug." }, { "left": "issue:3019", "right": "issue:6407", "accept": false, - "reason": "A text-guided editing feature request and a multilingual visual text generation model request are separate feature requests." + "reason": "These are different research-feature requests for text/image editing, not the same underlying issue." }, { "left": "issue:5432", "right": "issue:6949", "accept": false, - "reason": "Validation-image blankness in LoRA SDXL training and even-step blank images in a scheduler are not the same code-path." + "reason": "Both mention blank outputs, but one is blank validation images in a LoRA training script and the other is a scheduler-specific blank-image bug." }, { "left": "issue:7255", "right": "issue:7948", "accept": false, - "reason": "Both are model/pipeline feature requests, but for different methods (PeRFlow vs MuLan) and no shared bug/change." + "reason": "Completely different model/proposal topics; no shared bug or change." }, { "left": "issue:2991", "right": "issue:761", "accept": false, - "reason": "Mixed-precision ControlNet example failure is unrelated to DreamBooth 8-bit Adam training failure." + "reason": "Mixed-precision ControlNet example failure and 8-bit Adam DreamBooth failure are different training configuration bugs." }, { "left": "issue:9546", "right": "issue:9763", "accept": false, - "reason": "Both are Flux ControlNet training issues, but one is validation OOM and the other is prompt truncation at >77 tokens; different bugs." + "reason": "Both are Flux ControlNet training issues, but one is validation-step OOM while the other is token truncation beyond 77 tokens; different concrete bugs." }, { "left": "issue:5023", "right": "issue:5152", "accept": false, - "reason": "Both concern Wuerstchen, but one is an import error in 0.21.0 and the other is a separate bug in 0.21.3 alpha." + "reason": "Both are W\u00fcrstchen-related, but one is an import error in 0.21.0 and the other is a different bug in 0.21.3 alpha; not clearly the same issue." }, { "left": "issue:10550", "right": "issue:11702", "accept": false, - "reason": "Both mention LoRA loading, but one is loading LoRA into a quantized base model and the other is an AccVid LoRA key error; different models and failure points." + "reason": "Both involve LoRA loading, but one is quantized-base-model loading and the other is an AccVid-specific key error; different code paths." }, { "left": "issue:3540", "right": "issue:6157", "accept": false, - "reason": "Prompt embedding support for TextToVideoZeroPipeline and IP-Adapter behavior in ControlNetImg2ImgPipeline are unrelated." + "reason": "TextToVideoZero prompt-embedding support and ControlNetImg2Img IP-Adapter behavior are unrelated pipeline bugs." }, { "left": "issue:1552", "right": "issue:730", "accept": false, - "reason": "Stable Diffusion 2 attention slicing and DreamBooth local-vs-Colab learning differences are not the same bug." + "reason": "Both are DreamBooth/training-related, but the reported failures and conditions differ substantially." }, { "left": "issue:6382", "right": "issue:7033", "accept": false, - "reason": "IP-adapter key error during LCM LoRA training/use and attention masking warning logs are different issues; one is a hard error, the other is logging." + "reason": "A training/loading key error with IP-Adapter and IP-Adapter attention masking warnings are not the same bug." }, { "left": "issue:2581", "right": "issue:4000", "accept": false, - "reason": "Both are about conversion, but one is ControlNet checkpoint conversion and the other is SD 2.0 ckpt-to-diffusers conversion; separate tasks." + "reason": "ControlNet conversion support and SD2.0 ckpt-to-diffusers conversion are adjacent but not the same conversion task or failure." }, { "left": "issue:4892", "right": "issue:5889", "accept": false, - "reason": "TokenFlow pipeline request and Stable Video Diffusion request are different model/pipeline additions." + "reason": "TokenFlow and Stable Video Diffusion are distinct video-generation features with no concrete duplicate evidence." }, { "left": "issue:3191", "right": "issue:3445", "accept": false, - "reason": "Both are LoRA-related training issues, but one is a general DreamBooth+LoRA breakage and the other is a specific text-encoder monkey-patching failure." + "reason": "DreamBooth+LoRA broken and text-encoder monkey-patching not working are related only at a very high level; different failure modes." }, { "left": "issue:4993", "right": "issue:5023", "accept": false, - "reason": "CLIPImageProjection export/import-structure issue and WuerstchenCombinedPipeline import error are related to packaging but involve different symbols and breakages." + "reason": "CLIPImageProjection import-structure exposure and W\u00fcrstchenCombinedPipeline import error are separate import issues affecting different symbols." + }, + { + "left": "issue:10026", + "right": "issue:8738", + "accept": false, + "reason": "Flux ControlNet memory behavior and SD3Transformer2DModel configuration are different training/model-setup issues." + }, + { + "left": "issue:11133", + "right": "issue:12359", + "accept": false, + "reason": "CogVideoX image-to-video pipeline bug and Chroma guidance_scale documentation bug are unrelated." } ] }, @@ -48003,24 +48036,21 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10026", "issue:10040", "issue:10428", "issue:1100", "issue:11006", "issue:11104", - "issue:11133", "issue:11136", "issue:11359", "issue:11410", "issue:11830", "issue:12120", - "issue:12359", "issue:1536", "issue:1695", "issue:1802", @@ -48030,6 +48060,7 @@ "issue:2495", "issue:2695", "issue:3213", + "issue:4030", "issue:4186", "issue:4321", "issue:4752", @@ -48037,11 +48068,14 @@ "issue:4936", "issue:5227", "issue:5274", + "issue:6090", "issue:6258", "issue:6385", "issue:6540", "issue:6549", + "issue:6594", "issue:6858", + "issue:7002", "issue:7731", "issue:7950", "issue:7974", @@ -48050,7 +48084,6 @@ "issue:8518", "issue:8551", "issue:858", - "issue:8738", "issue:8879", "issue:8958", "issue:9630", @@ -48058,8 +48091,6 @@ "issue:9979" ], "soft_pairs": [ - "issue:10026|issue:8738", - "issue:11133|issue:12359", "issue:6540|issue:6858", "issue:11359|issue:12120", "issue:11830|issue:4321", @@ -48082,196 +48113,195 @@ "issue:1100|issue:858", "issue:4892|issue:7731", "issue:10428|issue:9979", - "issue:1695|issue:2089" + "issue:1695|issue:2089", + "issue:4030|issue:6594", + "issue:6090|issue:7002" ], - "prepared_review_unit_hash": "397aef36b9ac5b78c68a0a90ea834764082b5e73eb129998d2db70bd654b4e87", + "prepared_review_unit_hash": "45ad2ebb03a8073b9415e5615cc95178bc39ae6ba7620445d8e7f4e343e4092a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34608, - "estimated_input_tokens": 8652, - "estimated_eval_tokens": 17560 + "serialized_chars": 34795, + "estimated_input_tokens": 8699, + "estimated_eval_tokens": 17654 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous; most pairs are only superficially similar by model/pipeline name or error class. The only near-duplicate-looking pair is the `lpw_stable_diffusion` compatibility regression.", - "confidence": 0.84, - "canonical_issue_reason": "Issue 1536 is the closest anchor because it is the clearest concrete bug report and the one soft pair that plausibly points to the same underlying `lpw_stable_diffusion` regression as issue 3213.", + "summary": "This cluster is mostly a loose collection of unrelated issues that only share broad symptom patterns (black/grey images, pipeline errors, training/runtime exceptions). None of the soft pairs look like safe duplicates under a strict same-code-path standard.", + "confidence": 0.37, + "canonical_issue_reason": "issue:10040 is the most general and representative visible-output corruption report in the set, with decent discussion and references, so it is the best canonical stand-in for the image-artifact subset.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 1536 is the strongest representative issue: it names a specific broken custom pipeline and a version-compatibility failure mode, which makes it the cleanest triage anchor.", + "best_issue_reason": "issue:10040 is concrete, recent, and broadly framed enough to anchor triage for the image-generation artifact reports; it is more representative than the highly specific one-off errors.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:10026", - "right": "issue:8738", - "accept": false, - "reason": "Both mention Flux/ControlNet-style setup problems, but one is a memory scaling complaint and the other is an SD3 configuration issue; different bugs and code paths." - }, - { - "left": "issue:11133", - "right": "issue:12359", - "accept": false, - "reason": "Completely different topics: CogVideoX I2V runtime bug vs Chroma documentation for `guidance_scale`." - }, { "left": "issue:6540", "right": "issue:6858", "accept": false, - "reason": "Both are ControlNet training issues, but the failures differ (`mat2 must be a matrix` vs `'str' object has no attribute convert'`) and point to separate root causes." + "reason": "Different failures: FSDP matrix-shape error in ControlNet fine-tuning vs a string/convert AttributeError during controlnet training." }, { "left": "issue:11359", "right": "issue:12120", "accept": false, - "reason": "A speed feature request for LTX-Video is unrelated to a Flux distilled-LoRA training question." + "reason": "Feature request about LTX-Video speed vs a how-to question about training LoRA on a distilled Flux model; not the same bug." }, { "left": "issue:11830", "right": "issue:4321", "accept": false, - "reason": "Both are black-output reports, but they involve different pipelines, backends, and failure contexts; not enough to be the same bug." + "reason": "Both mention black output, but the pipelines and backends differ (FramePack float16 vs ONNX StableDiffusion DirectML), so this is too broad to merge as one bug." }, { "left": "issue:1802", "right": "issue:4186", "accept": false, - "reason": "`encoder_hidden_states=None` handling in UNet is unrelated to SDXL inpaint/refiner behavior." + "reason": "Different code paths and symptoms: encoder_hidden_states=None on unet_2d_condition vs SDXL inpaint/refiner behavior." }, { "left": "issue:2107", "right": "issue:2179", "accept": false, - "reason": "One reports low-resolution glitchy/black images, the other a Flax scheduler black-image issue; similar symptom, different underlying problem." + "reason": "Low-resolution glitch/black images vs FlaxDDPMScheduler black images; same broad symptom, but not clearly the same underlying defect." }, { "left": "issue:6385", "right": "issue:8551", "accept": false, - "reason": "VRAM usage in Stable Video Diffusion is unrelated to SDXL inpaint device placement with `device_map=\"balanced\"`." + "reason": "VRAM usage issue vs device placement bug for image latents; unrelated underlying problems." }, { "left": "issue:11136", "right": "issue:9889", "accept": false, - "reason": "Different models and failures: SanaSprint timestep support vs an SDXL-turbo segmentation fault." + "reason": "SanaSprint timestep support limitation vs SDXL-turbo segfault; different models and different failure modes." }, { "left": "issue:10040", "right": "issue:6258", "accept": false, - "reason": "Grey-image generation is not the same as forward-chunking quality degradation, even if both affect output quality." + "reason": "Grey image generation in SD3 vs quality degradation from forward chunking in Stable Video Diffusion; not the same bug." }, { "left": "issue:2695", "right": "issue:9630", "accept": false, - "reason": "ControlNet validation/TensorBoard handling is unrelated to a Flux-ControlNet `num_single_layers=0` runtime error." + "reason": "Tensorboard validation failure during training vs flux-controlnet num_single_layers handling on single GPU; unrelated." }, { "left": "issue:2495", "right": "issue:4321", "accept": false, - "reason": "Both mention black images under FP16, but one is MPS upscale and the other is DirectML ONNX; too different to merge." + "reason": "Both produce black images, but they are distinct pipelines/backends (MPS upscale vs ONNX DirectML), so they should not be merged." }, { "left": "issue:11410", "right": "issue:7950", "accept": false, - "reason": "A tensor-unpacking error in attention code is unrelated to a PyTorch version compatibility issue." + "reason": "Attention unpacking ValueError vs PyTorch 2.3 compatibility issue; not the same concrete code-path." }, { "left": "issue:6549", "right": "issue:7974", "accept": false, - "reason": "IP-Adapter + LoRA warning logs are not the same as an inpaint+IP-Adapter functional bug." + "reason": "Warning-log typo/reporting issue vs IP-Adapter inpaint bug; different symptoms and fixes." }, { "left": "issue:4936", "right": "issue:8472", "accept": false, - "reason": "DreamBooth resume breakage is different from a general question about pre-training LoRA with other modules." + "reason": "Dreambooth resume regression vs a question about training LoRA with other modules; not duplicate bug reports." }, { "left": "issue:5227", "right": "issue:8879", "accept": false, - "reason": "Both involve `torch.compile`, but one is about `load_lora_weight` and the other about cudagraph failure in SDXL; separate failures." + "reason": "LoRA loading with torch.compile vs torch.compile+cudagraph failure on SDXL; related theme, but not the same underlying defect." }, { "left": "issue:11006", "right": "issue:11104", "accept": false, - "reason": "Different video pipelines and different symptoms; no shared concrete code-path bug." + "reason": "Wan 2.1 I2V quantized-transformer video corruption vs LTX-Video offload methods broken; different pipelines and failure mechanisms." }, { "left": "issue:1536", "right": "issue:3213", - "accept": true, - "reason": "Both point to the same `lpw_stable_diffusion` pipeline breakage/compatibility regression, with 3213 reading like a \u043f\u043e\u0432\u0442\u043e\u0440ed report of the same underlying issue." + "accept": false, + "reason": "Both concern lpw_stable_diffusion, but one is a compatibility report and the other is a later breakage report; insufficient evidence they are the same bug." }, { "left": "issue:4752", "right": "issue:5274", "accept": false, - "reason": "Both are dtype mismatch symptoms, but they occur in different pipelines and training/inference paths." + "reason": "SDXL bias/input dtype mismatch in inference vs mixed-precision training dtype error; similar dtype flavor, but different code paths and likely different fixes." }, { "left": "issue:8182", "right": "issue:8958", "accept": false, - "reason": "A typo in logger usage is unrelated to the `torch.library.register_fake` API absence." + "reason": "Logger API typo vs torch.library.register_fake missing in PyTorch 2.4; completely unrelated." }, { "left": "issue:8518", "right": "issue:8879", "accept": false, - "reason": "Both are `torch.compile` problems, but the reported unsupported call on SD3 is not the same bug as cudagraph failure on SDXL." + "reason": "SD3 torch.compile unsupported UserDefinedClassVariable vs SDXL torch.compile+cudagraph failure; different models and compiler failure sites." }, { "left": "issue:1100", "right": "issue:858", "accept": false, - "reason": "Different attribute errors on unrelated objects; no shared code path." + "reason": "CLIPTextTransformer device attribute issue vs Parameter.named_children AttributeError; different API breakages." }, { "left": "issue:4892", "right": "issue:7731", "accept": false, - "reason": "TokenFlow and StreamingT2V are different video-generation features, not the same bug." + "reason": "Distinct research/pipeline features (TokenFlow vs StreamingT2V), not the same bug." }, { "left": "issue:10428", "right": "issue:9979", "accept": false, - "reason": "One is a Flux inference error on Ascend NPU, the other a Flux img2img ControlNet channel mismatch; distinct issues." + "reason": "Both mention Flux, but one is an Ascend NPU inference error and the other is a controlnet channels mismatch; different problems." }, { "left": "issue:1695", "right": "issue:2089", "accept": false, - "reason": "DreamBooth multi-GPU/DeepSpeed training and AltDiffusion OOM are different training problems." + "reason": "Dreambooth multi-GPU/deepspeed training request vs a CUDA OOM in AltDiffusion DreamBooth; unrelated." + }, + { + "left": "issue:4030", + "right": "issue:6594", + "accept": false, + "reason": "Stable Diffusion embedding runtime error vs network failure fetching a remote config file; not duplicate issues." + }, + { + "left": "issue:6090", + "right": "issue:7002", + "accept": false, + "reason": "Channel-count mismatch in a model layer vs an invalid buffer size memory error; different root causes." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "Grounded in the packet: the representative issue choice and the cluster summary are supported by titles alone, and the rejected soft pairs are framed conservatively. The `lpw_stable_diffusion` pair is the only plausible near-duplicate and the verdict does not overstate the evidence." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -48283,8 +48313,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -48301,6 +48331,9 @@ "issue:10940", "issue:11014", "issue:11062", + "issue:11215", + "issue:11464", + "issue:11648", "issue:11853", "issue:2234", "issue:2265", @@ -48312,7 +48345,6 @@ "issue:3566", "issue:3595", "issue:3886", - "issue:4030", "issue:4398", "issue:5064", "issue:5075", @@ -48321,24 +48353,20 @@ "issue:5258", "issue:5433", "issue:5630", - "issue:6090", "issue:6154", "issue:6297", - "issue:6594", "issue:6741", - "issue:7002", "issue:7168", "issue:7516", "issue:7982", "issue:8832", "issue:9136", "issue:9208", + "issue:9525", "issue:9558", "issue:987" ], "soft_pairs": [ - "issue:4030|issue:6594", - "issue:6090|issue:7002", "issue:7168|issue:8832", "issue:2265|issue:3886", "issue:10492|issue:11062", @@ -48360,180 +48388,182 @@ "issue:10512|issue:10804", "issue:1061|issue:2880", "issue:1070|issue:259", - "issue:10940|issue:7516" + "issue:10940|issue:7516", + "issue:11215|issue:9525", + "issue:11464|issue:11648" ], - "prepared_review_unit_hash": "b4d644473028e3a70025666dc187b5d9a22ff7629e40a80f0330e08d3f33831f", + "prepared_review_unit_hash": "cfceafada8fdff137a5554f74a1b7eff8f5057a5e8bdacf9ffb372ba3c516474", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33466, - "estimated_input_tokens": 8367, - "estimated_eval_tokens": 16990 + "serialized_chars": 33213, + "estimated_input_tokens": 8304, + "estimated_eval_tokens": 16864 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: LoRA-loading problems, SDXL/pipeline bugs, training OOMs, docs requests, and import/runtime errors. None of the proposed soft pairs look like the same underlying bug or change.", - "confidence": 0.94, - "canonical_issue_reason": "Issue 10512 is the clearest representative of the recurring LoRA-loading/theme in this set, with a detailed and active bug report about a concrete failure path.", + "summary": "This cluster is mostly heterogeneous: LoRA/Flux loading and scaling issues appear alongside unrelated training, scheduler, docs, and pipeline bugs. None of the soft pairs look like true duplicates; they\u2019re only loosely similar by subsystem.", + "confidence": 0.96, + "canonical_issue_reason": "Issue 9525 is the clearest concrete bug report in the densest recurring theme (Flux/LoRA behavior), but the overall cluster is too mixed to treat it as a real duplicate anchor.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 10512 is the strongest standalone issue to anchor the cluster because it is specific, actionable, and closest to the most repeated functional area (LoRA loading).", + "best_issue_reason": "Issue 9525 is the most actionable standalone report: specific title, clear symptom, and a focused code-path (Flux LoRA scale handling).", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:4030", - "right": "issue:6594", - "accept": false, - "reason": "Different failures: one is a Stable Diffusion v2.1 embedding runtime error, the other is a network/download connection error." - }, - { - "left": "issue:6090", - "right": "issue:7002", - "accept": false, - "reason": "Both are runtime failures, but the concrete bugs differ: channel-mismatch convolution input vs invalid buffer size/memory blowup." - }, { "left": "issue:7168", "right": "issue:8832", "accept": false, - "reason": "One is specifically about incorrect IP-Adapter image embedding tensors; the other is a generic pipeline crash at image generation." + "reason": "Different problems: IP-Adapter image embedding tensor shape vs a generic pipeline crash on `pipe(prompt)`." }, { "left": "issue:2265", "right": "issue:3886", "accept": false, - "reason": "Different code paths and symptoms: ONNXRuntime incompatibility in a training script vs distributed training watchdog timeout." + "reason": "Different failures: ONNXRuntime incompatibility in a training script vs distributed collective timeout during ddpm training." }, { "left": "issue:10492", "right": "issue:11062", "accept": false, - "reason": "Both mention LoRA, but one is about torchao quantize/fuse errors and the other is about loading a CivitAI LoRA; not the same bug." + "reason": "Related to quantization/LoRA, but one is a torchao quantize fuse error and the other is a CivitAI LoRA load error in a different setup." }, { "left": "issue:11014", "right": "issue:3382", "accept": false, - "reason": "Both mention OOM, but one is a general CUDA OOM in inference and the other is multi-GPU training OOM; too broad to treat as duplicates." + "reason": "Both mention OOM, but the contexts differ: inference on Colab A100 vs multi-GPU training memory usage." }, { "left": "issue:10223", "right": "issue:4398", "accept": false, - "reason": "A question about where to obtain a LoRA ID is unrelated to black-image output from the SDXL LoRA-dreambooth pipeline." + "reason": "Completely different topics: a question about a LoRA ID for inference vs all-black images in an SDXL LoRA dreambooth pipeline." }, { "left": "issue:5128", "right": "issue:9208", "accept": false, - "reason": "Same API family, but one is loading a local .safetensors into XLPipeline and the other is loading controlnet models from single file; not clearly the same defect." + "reason": "Both involve loading files/models, but one is a local `.safetensors` load and the other is ControlNet loading from single-file checkpoints." }, { "left": "issue:5215", "right": "issue:5630", "accept": false, - "reason": "Both are image-quality complaints in SDXL pipelines, but brightness shift and leftover noise are different bugs." + "reason": "Both are image-quality regressions, but one is brightness shift in upscale and the other is leftover noise in SDXL img2img with LMS." }, { "left": "issue:3566", "right": "issue:987", "accept": false, - "reason": "Different pipelines and outputs: controlnet_reference unexpected results vs speech-to-image green output." + "reason": "Both report unexpected output colors/artifacts, but the pipelines and symptoms are different (ControlNet reference vs speech-to-image green output)." }, { "left": "issue:5630", "right": "issue:9558", "accept": false, - "reason": "SDXL img2img scheduler noise is unrelated to a black validation image in tensorboard." + "reason": "Different manifestations: noise leftover in SDXL img2img vs black validation images in tensorboard." }, { "left": "issue:11853", "right": "issue:9136", "accept": false, - "reason": "One is a loading/config conflict when combining ControlNet and IP-Adapter; the other is a quality issue at certain resolutions." + "reason": "Not the same bug: one is a controlnet+IP-Adapter dimension conflict, the other is a general quality issue at some resolutions." }, { "left": "issue:6154", "right": "issue:6741", "accept": false, - "reason": "Different training issues: M2 Mac LoRA training failure vs a missing argparse field in Diffusion DPO." + "reason": "Different training problems: SDXL LoRA training on M2 Mac vs a missing `no_flip` argument in a Diffusion DPO script." }, { "left": "issue:3595", "right": "issue:7982", "accept": false, - "reason": "Both involve conversion/loading helpers, but from_ckpt failure and from_pipe with MultiControlNet are different concrete problems." + "reason": "Both are `from_*`/loading-related, but one is `from_ckpt` failure and the other is `from_pipe` with MultiControlNetModel." }, { "left": "issue:2234", "right": "issue:6297", "accept": false, - "reason": "Both are training problems, but xformers finetuning failure and SDXL LoRA photo-training failure are not the same bug." + "reason": "Both concern training, but one is xformers memory-efficient attention and the other is SDXL LoRA training on photos." }, { "left": "issue:5258", "right": "issue:5433", "accept": false, - "reason": "Both mention SDXL/refiner, but one is ControlNet inpainting with refiner and the other is DPM++ artifacts without refiner." + "reason": "Different SDXL/refiner issues: controlnet inpaint with refiner vs DPM++ artifacts without refiner." }, { "left": "issue:10165", "right": "issue:10866", "accept": false, - "reason": "Both are LoRA-related, but seed preservation when changing scale is unrelated to compatibility/loading of a Lumina Image 2.0 LoRA." + "reason": "Both mention LoRA, but one is seed preservation when changing scale and the other is a Lumina/Civitai compatibility problem." }, { "left": "issue:5064", "right": "issue:5075", "accept": false, - "reason": "Different refiner issues: out-of-space during inference vs a tensor shape mismatch when using prompt embeddings." + "reason": "Both involve SDXL refiner usage, but one is out-of-space inference and the other is a prompt-embedding shape mismatch." }, { "left": "issue:1013", "right": "issue:3149", "accept": false, - "reason": "Different versions and bug types: a pipeline not working in Diffusers 6.0 vs non-deterministic images in 0.14.0." + "reason": "Different core bugs: a specific StableDiffusionMegaPipeline compatibility issue vs nondeterministic image generation." }, { "left": "issue:1049", "right": "issue:3412", "accept": false, - "reason": "Both are docs issues, but they target unrelated documentation pages and changes." + "reason": "Both are docs-related, but they request different documentation updates (Euler schedulers vs stable diffusion next steps)." }, { "left": "issue:10512", "right": "issue:10804", "accept": false, - "reason": "Both are FLUX/LoRA load failures, but the titles suggest different model/format contexts (Quanto Flux vs ComfyUI-compatible FLUX1.dev) without enough evidence of the same bug." + "reason": "Related to FLUX LoRA loading, but one is Quanto Flux LoRA load failure and the other is a ComfyUI-compatible FLUX1.dev LoRA load failure." }, { "left": "issue:1061", "right": "issue:2880", "accept": false, - "reason": "One is a documentation update request; the other is a JAX Colab runtime failure." + "reason": "Different content entirely: a text inversion docs update vs a JAX Colab runtime failure." }, { "left": "issue:1070", "right": "issue:259", "accept": false, - "reason": "Different Stable Diffusion issues: inpainting bug vs progress-bar behavior." + "reason": "Different pipelines and symptoms: inpainting bug in new SD vs progress bar behavior in Stable Diffusion." }, { "left": "issue:10940", "right": "issue:7516", "accept": false, - "reason": "Separate import problems with different missing symbols/modules: triton.ops missing vs DepthNormalEstimationPipeline absent." + "reason": "Unrelated import/runtime issues: missing `triton.ops` vs missing `DepthNormalEstimationPipeline` symbol." + }, + { + "left": "issue:11215", + "right": "issue:9525", + "accept": false, + "reason": "Both touch LoRA and quantization/Flux, but one is about slow quantization and ineffective LoRA support, not the same as `lora_scale` having no effect." + }, + { + "left": "issue:11464", + "right": "issue:11648", + "accept": false, + "reason": "Both are Flux/LoRA-related, but one is loading a specific Canny-dev LoRA in FluxControlPipeline and the other is fp8 transformer LoRA weight handling." } ] }, @@ -48552,14 +48582,11 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:11215", - "issue:11464", - "issue:11648", "issue:1185", "issue:11964", "issue:1304", @@ -48589,6 +48616,7 @@ "issue:383", "issue:4064", "issue:4080", + "issue:4117", "issue:4376", "issue:4868", "issue:4933", @@ -48597,18 +48625,17 @@ "issue:528", "issue:540", "issue:5595", + "issue:6158", "issue:660", "issue:668", + "issue:684", "issue:7428", "issue:776", "issue:905", "issue:9178", - "issue:9525", "issue:9637" ], "soft_pairs": [ - "issue:11215|issue:9525", - "issue:11464|issue:11648", "issue:1185|issue:2599", "issue:11964|issue:9178", "issue:1304|issue:2093", @@ -48630,180 +48657,182 @@ "issue:3175|issue:3803", "issue:3583|issue:7428", "issue:3811|issue:4868", - "issue:383|issue:528" + "issue:383|issue:528", + "issue:4080|issue:6158", + "issue:4117|issue:684" ], - "prepared_review_unit_hash": "e643d7b93ed64606e27771f57a20eaf89f606aca476285e49aea5133d35c97ec", + "prepared_review_unit_hash": "d09b840b81656e8e5645999b5000d223f5c3a87dea35c61653624fd27316622b", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33305, - "estimated_input_tokens": 8327, - "estimated_eval_tokens": 16910 + "serialized_chars": 32722, + "estimated_input_tokens": 8181, + "estimated_eval_tokens": 16618 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly Flux/LoRA-related user reports, but they split into distinct bugs: LoRA loading/effectiveness, fp8/quantized model loading, training target-module errors, and a separate quantization-slow complaint. No PRs are present.", - "confidence": 0.83, - "canonical_issue_reason": "issue:9525 is the cleanest, most generic Flux LoRA bug report: LoRA loading has no effect. It best represents the common failure mode behind the Flux LoRA-loading reports without adding extra unrelated symptoms.", + "summary": "This cluster is a broad false positive: it mixes Dreambooth/LoRA training bugs, pipeline feature requests, platform-specific errors, and unrelated questions. None of the soft-edge pairs look like the same underlying issue or change.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 4117 is the most substantial bug report in the training-related subset and the closest thing to a cluster anchor, but it is still distinct from the other Dreambooth/LoRA reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9525 is the best global issue candidate because it states the core Flux/LoRA symptom directly and broadly; the other Flux issues are either more specific load errors, training setup problems, or mixed with performance concerns.", + "best_issue_reason": "Issue 4117 is the strongest representative artifact: detailed repro, clear failure mode, and the broadest relevance among the training-bug issues in this mixed set.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:11215", - "right": "issue:9525", - "accept": false, - "reason": "Related Flux LoRA complaints, but 11215 also mixes in quantization slowdown and frames the problem more broadly; not the same concrete bug as `lora_scale` having no effect." - }, - { - "left": "issue:11464", - "right": "issue:11648", - "accept": false, - "reason": "Both involve Flux LoRA loading, but one is a ControlPipeline load failure and the other is fp8 transformer-model handling; different failure modes." - }, { "left": "issue:1185", "right": "issue:2599", "accept": false, - "reason": "Both are Dreambooth training complaints, but one is about reproducibility and the other about multi-subject quality degradation." + "reason": "Both concern Dreambooth results, but one is about reproducibility and the other about multi-subject quality degradation; different bugs." }, { "left": "issue:11964", "right": "issue:9178", "accept": false, - "reason": "Both are Flux LoRA issues, but one is a KeyError on missing weights while the other is a PEFT target-module configuration error during training." + "reason": "Both mention Flux LoRA, but one is a load-time missing-weights KeyError and the other is a training target-module error." }, { "left": "issue:1304", "right": "issue:2093", "accept": false, - "reason": "One is a training how-to for text-guided image-to-image SD, the other is a generic image-to-image diffusion-model topic." + "reason": "These are both image-to-image related, but one is a how-to/training question and the other is a generic model-topic issue." }, { "left": "issue:1305", "right": "issue:5153", "accept": false, - "reason": "Different research/features entirely: Direct Inversion vs FreeU." + "reason": "Direct Inversion and FreeU are unrelated research/features; no shared bug or change path." }, { "left": "issue:1500", "right": "issue:2053", "accept": false, - "reason": "Different problems: pipeline incompatibility with a specific version vs increased VRAM usage." + "reason": "One is a community pipeline incompatibility, the other is a VRAM regression; different symptoms and code paths." }, { "left": "issue:1552", "right": "issue:1807", "accept": false, - "reason": "One is an attention-slicing compatibility bug; the other is a runtime slowdown over repeated generations." + "reason": "Attention slicing breakage is not the same as gradually slowing inference; different performance issues." }, { "left": "issue:1572", "right": "issue:4933", "accept": false, - "reason": "Different areas: img2img/inpainting pipeline support vs Dreambooth checkpoint export." + "reason": "Img2img/inpainting pipeline failure is unrelated to Dreambooth checkpoint export problems." }, { "left": "issue:2218", "right": "issue:5014", "accept": false, - "reason": "Training quality complaints, but one is generic overfitting and the other is not getting expected results from LoRA training." + "reason": "Both describe poor training outcomes, but one is generic overtraining and the other is specifically LoRA results not showing up." }, { "left": "issue:2361", "right": "issue:4376", "accept": false, - "reason": "Different crash sites: checkpoint saving vs textual inversion loading." + "reason": "Checkpoint-save crashing and textual inversion crashing are different failures in different scripts/pipelines." }, { "left": "issue:2437", "right": "issue:4064", "accept": false, - "reason": "Both are feature requests for pipelines, but they target different pipeline families and are not the same change." + "reason": "Both are pipeline feature requests, but for different pipeline types and different functionality." }, { "left": "issue:2462", "right": "issue:540", "accept": false, - "reason": "IterableDataset training support and GPU memory optimization are unrelated bugs/features." + "reason": "IterableDataset support and GPU memory usage are unrelated training concerns." }, { "left": "issue:2645", "right": "issue:660", "accept": false, - "reason": "Both mention M1/MPS/CUDA issues, but they are different execution-path failures." + "reason": "Both involve Apple hardware, but one is a ControlNet CUDA assertion and the other is fp16/MPS execution." }, { "left": "issue:2689", "right": "issue:776", "accept": false, - "reason": "Broken image link is unrelated to the img2img batch-size bug." + "reason": "A broken image link is not the same as an img2img batch-size bug." }, { "left": "issue:2776", "right": "issue:5595", "accept": false, - "reason": "Different errors: DeepSpeed version/runtime issue vs invalid image dimensions." + "reason": "DeepSpeed/ZeRO runtime failure is unrelated to invalid image size dimensions." }, { "left": "issue:2783", "right": "issue:905", "accept": false, - "reason": "Similar broad area of img2img/inpainting support, but not the same concrete pipeline request." + "reason": "Both mention img2img/inpainting, but one is a ControlNet pipeline request and the other is a broader community pipeline question." }, { "left": "issue:2785", "right": "issue:4080", "accept": false, - "reason": "Different feature requests: text-to-video M1/M2 support vs SDXL inpainting pipeline." + "reason": "M1/M2 text-to-video support and SDXL inpainting are different feature requests for different pipelines." }, { "left": "issue:2844", "right": "issue:668", "accept": false, - "reason": "Completely different runtime failures in different workflows." + "reason": "StableUnCLIP image-embedding passing and Dreambooth on Colab are unrelated runtime errors." }, { "left": "issue:2991", "right": "issue:9637", "accept": false, - "reason": "Both involve attention/mixed-precision/XFormers edge cases, but they are different code paths and error conditions." + "reason": "Both touch attention/mixed precision, but they affect different mechanisms: ControlNet mixed-precision vs XFormers attention-mask/bfloat16 edge case." }, { "left": "issue:3175", "right": "issue:3803", "accept": false, - "reason": "Dreambooth multi-face training vs multi-GPU LoRA training are separate issues." + "reason": "Multi-face Dreambooth training and multi-GPU LoRA synchronization are different training problems." }, { "left": "issue:3583", "right": "issue:7428", "accept": false, - "reason": "Different requested features: Mixture of Diffusers vs Layer Diffusion." + "reason": "These are separate feature requests for different diffusion variants; no same implementation target." }, { "left": "issue:3811", "right": "issue:4868", "accept": false, - "reason": "LoRA having no effect and a `super().__getattr__` AttributeError are unrelated bugs." + "reason": "LoRA load having no effect on generation is not the same as a generic AttributeError in __getattr__." }, { "left": "issue:383", "right": "issue:528", "accept": false, - "reason": "General design/question threads, but about different configuration topics." + "reason": "Both are conceptual questions about model classes, but they are about different frameworks and different topics." + }, + { + "left": "issue:4080", + "right": "issue:6158", + "accept": false, + "reason": "SDXL inpainting and animatediff SDXL support are separate pipeline features." + }, + { + "left": "issue:4117", + "right": "issue:684", + "accept": false, + "reason": "Dreambooth failing on SDXL is unrelated to AMD training memory optimization issues." } ] }, @@ -48822,19 +48851,20 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:12369", + "issue:12457", "issue:12480", "issue:12481", "issue:12482", "issue:12483", "issue:2453", - "issue:4080", - "issue:4117", "issue:4152", + "issue:4241", "issue:4274", "issue:4316", "issue:4340", @@ -48852,14 +48882,12 @@ "issue:5432", "issue:5577", "issue:5932", - "issue:6158", "issue:6218", "issue:6297", "issue:6388", "issue:6423", "issue:6439", "issue:6791", - "issue:684", "issue:6865", "issue:7002", "issue:7009", @@ -48877,8 +48905,6 @@ "issue:986" ], "soft_pairs": [ - "issue:4080|issue:6158", - "issue:4117|issue:684", "issue:4152|issue:7877", "issue:4274|issue:5371", "issue:4316|issue:5432", @@ -48901,192 +48927,194 @@ "issue:12481|issue:12483", "issue:12482|issue:12483", "issue:12480|issue:12483", - "issue:2453|issue:6423" + "issue:2453|issue:6423", + "issue:2453|issue:4241", + "issue:12369|issue:12457" ], - "prepared_review_unit_hash": "c8ed7f079ef742f1f147f05736d0215a944d3ec5c5c7630eb5f76aecaa3c1dd7", + "prepared_review_unit_hash": "4e391ebdd7279f255cdf9d73a62083af3a4483abf58579417381147f78b3473e", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 33713, - "estimated_input_tokens": 8429, - "estimated_eval_tokens": 17114 + "serialized_chars": 33140, + "estimated_input_tokens": 8285, + "estimated_eval_tokens": 16826 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are unrelated, but four October 2025 issues are clear duplicates of the same CLIPTextModel/offload_state_dict compatibility failure; the dependency-conflict report is the best representative.", - "confidence": 0.97, - "canonical_issue_reason": "Choose issue 12483: it states the underlying dependency mismatch clearly (diffusers 0.35.1 vs transformers 4.57.0) and subsumes the nearly identical CLIPTextModel '__init__'/'unexpected kwarg offload_state_dict' reports in 12480/12481/12482.", + "summary": "This cluster is mostly unrelated issue reports, except for a tight duplicate set about `offload_state_dict` / `CLIPTextModel` / the diffusers-transformers dependency conflict. I would use issue 12483 as the canonical issue for that duplicate group.", + "confidence": 0.95, + "canonical_issue_reason": "12483 is the most complete and highest-level report in the duplicate set: it frames the problem as a diffusers 0.35.1 vs transformers 4.57.0 dependency conflict causing the `offload_state_dict` TypeError, while 12480-12482 are narrower near-duplicate titles of the same failure.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 12483 is the strongest cluster anchor because it is the most general, diagnostic description of the same bug and is easiest to triage against the duplicate runtime TypeError reports.", + "best_issue_reason": "12483 is the best triage anchor because it captures the underlying compatibility problem, not just the surface CLIPTextModel constructor error.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:4080", - "right": "issue:6158", - "accept": false, - "reason": "Different features: SDXL inpainting pipeline vs Animatediff SDXL support request." - }, - { - "left": "issue:4117", - "right": "issue:684", - "accept": false, - "reason": "Different problems: Dreambooth on SDXL vs AMD training-memory optimization failure." - }, { "left": "issue:4152", "right": "issue:7877", "accept": false, - "reason": "Unrelated attention customization request vs new out_dim support request; same subsystem only." + "reason": "Both concern attention APIs, but one asks for mixed attention kwargs per processor and the other for an `out_dim` option on Attention; different feature requests." }, { "left": "issue:4274", "right": "issue:5371", "accept": false, - "reason": "Different pipelines/features: BLIP diffusion addition vs BLIP inpainting pipeline support." + "reason": "Both mention BLIP, but one requests BLIP Diffusion support and the other BLIP inpainting pipeline support; separate additions." }, { "left": "issue:4316", "right": "issue:5432", "accept": false, - "reason": "Both are training-script issues, but one is validation-option error and the other is blank validation images; not the same bug." + "reason": "Both are training-script validation issues, but they affect different scripts and symptoms, so they are not the same bug." }, { "left": "issue:4340", "right": "issue:8472", "accept": false, - "reason": "Both involve LoRA training, but one is an SDXL training error and the other is a general question about simultaneous modules." + "reason": "One is an SDXL LoRA training error and the other is a general question about training LoRA with other modules; not the same concrete problem." }, { "left": "issue:4405", "right": "issue:6388", "accept": false, - "reason": "Feature request for SDXL textual inversion vs a gradient-checkpointing bug in SDXL LoRA training." + "reason": "Textual inversion script request and gradient-checkpointing ignored in SDXL LoRA training are unrelated changes." }, { "left": "issue:4460", "right": "issue:7142", "accept": false, - "reason": "Different symptom and context: output mismatch in diffusion inference vs SDXL-lightning inference mismatch." + "reason": "Both compare diffusers output against another implementation, but they involve different models and failure modes." }, { "left": "issue:4727", "right": "issue:5577", "accept": false, - "reason": "Different loading paths: local_files_only auto pipeline breakage vs ControlNet from_single_file KeyError." + "reason": "One is about `local_files_only` auto pipelines; the other is a `from_single_file` ControlNet key error. Different code paths." }, { "left": "issue:4773", "right": "issue:8532", "accept": false, - "reason": "Loading CivitAI LoRA vs progressive memory increase while loading LoRA weights; same broad topic only." + "reason": "Both involve LoRA loading, but one is about loading a specific CivitAI LoRA and the other about progressive memory growth; different bugs." }, { "left": "issue:4806", "right": "issue:5330", "accept": false, - "reason": "UNet2DModel type annotation issue vs class_labels being ignored in forward; related area but not the same defect." + "reason": "Both touch UNet2DModel, but one is a parameter type issue and the other is `class_labels` being ignored in forward; not the same defect." }, { "left": "issue:5004", "right": "issue:6439", "accept": false, - "reason": "Training resume/fp16 bug vs docs link 404; unrelated." + "reason": "Checkpoint-resume training failure and a broken docs link are unrelated." }, { "left": "issue:5150", "right": "issue:5405", "accept": false, - "reason": "from_single_file missing on an upscaler pipeline vs SDXL from_single_file not overwriting weights; different code paths." + "reason": "Both mention `from_single_file`, but one is a missing method on StableDiffusionUpscalePipeline and the other is weights not being overwritten in SDXL loading." }, { "left": "issue:5392", "right": "issue:986", "accept": false, - "reason": "General training question vs unrelated demo API question." + "reason": "A training question for latent diffusion and a butterfly demo question are unrelated." }, { "left": "issue:5932", "right": "issue:7002", "accept": false, - "reason": "Both are runtime shape/buffer errors, but they involve different pipelines and failure modes." + "reason": "Both are runtime errors in pipelines, but they describe different symptoms and likely different root causes." }, { "left": "issue:6218", "right": "issue:7592", "accept": false, - "reason": "Batch seeding bug in ControlNet/img2img vs DoRA state_dict loading bug; not the same underlying issue." + "reason": "Batch seed handling in pipelines vs incomplete DoRA state dict loading are unrelated." }, { "left": "issue:6297", "right": "issue:9393", "accept": false, - "reason": "Different training workflows and failures: SDXL LoRA photos training vs flux dreambooth LoRA save-model problem." + "reason": "Both are LoRA training problems, but they involve different models, tooling, and failure points." }, { "left": "issue:6791", "right": "issue:7146", "accept": false, - "reason": "train_text_to_image_sdxl.py failing to run vs CUDA OOM during SDXL fine-tuning; different root causes." + "reason": "Same training script, but one report says it cannot run while the other is specifically CUDA OOM; not the same issue." }, { "left": "issue:6865", "right": "issue:9645", "accept": false, - "reason": "Both are import/loading concerns, but one is an image-instance check bug and the other is a missing symbol import." + "reason": "A load_image type-check bug and an import error for a loader mixin are unrelated." }, { "left": "issue:7009", "right": "issue:9568", "accept": false, - "reason": "Text-to-video quality/loading issue vs CogVideoX autoencoder reconstruction failure; unrelated." + "reason": "Different model families and different reconstruction/generation failures." }, { "left": "issue:9611", "right": "issue:9767", "accept": false, - "reason": "Both touch Flux image pipelines, but one is batching and the other is cuDNN frontend LoRA execution error; not the same bug." + "reason": "Both involve FluxImg2ImgPipeline, but one is an image batching error and the other is a cuDNN frontend error during LoRA use; too different to merge." }, { "left": "issue:12481", "right": "issue:12483", "accept": true, - "reason": "Same underlying bug: identical CLIPTextModel unexpected 'offload_state_dict' TypeError caused by the diffusers/transformers compatibility mismatch." + "reason": "Same `offload_state_dict` TypeError family; 12481 is just another phrasing of the same diffusers/transformers incompatibility." }, { "left": "issue:12482", "right": "issue:12483", "accept": true, - "reason": "Same underlying bug: identical CLIPTextModel unexpected 'offload_state_dict' TypeError from the same version conflict." + "reason": "Same underlying bug: `CLIPTextModel` receives unexpected `offload_state_dict`, matching the same compatibility problem." }, { "left": "issue:12480", "right": "issue:12483", "accept": true, - "reason": "Same underlying bug: the title and failure match the other CLIPTextModel/offload_state_dict compatibility reports." + "reason": "Same root cause and error path; 12480 is a duplicate report of the `offload_state_dict` constructor failure." }, { "left": "issue:2453", "right": "issue:6423", "accept": false, - "reason": "Completely unrelated: contribution question vs inability to reproduce fill50k results." + "reason": "General contribution question and a specific reproducibility issue are unrelated." + }, + { + "left": "issue:2453", + "right": "issue:4241", + "accept": false, + "reason": "A contribution question and a LoRA fine-tuning feature request are not duplicates." + }, + { + "left": "issue:12369", + "right": "issue:12457", + "accept": false, + "reason": "Both are CineScale-related, but the titles are too broad to confirm the same bug or change; insufficient evidence for a duplicate merge." } ] }, "evaluator_result": { "accept": true, - "feedback": "The clustering is well grounded in the packet. The four October 2025 issues are highly similar and the choice of issue 12483 as the canonical anchor is conservative and supported by the titles. The non-matching soft pairs are also rejected with mostly cautious, title-based reasoning. No overstatement stands out." + "feedback": "The summary is grounded in the packet: 12369/12457 is a separate CineScale pair, and 12480/12481/12482/12483 are a clear duplicate cluster around the same `offload_state_dict` / `CLIPTextModel` constructor failure. The soft-edge rejections are conservative and the accepted duplicates are well supported by the near-identical titles and high similarity scores." }, "evaluator_used": true, "retried": false, @@ -49102,15 +49130,14 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:11519", "issue:11560", "issue:12369", - "issue:12457", "issue:1755", "issue:1962", "issue:200", @@ -49120,6 +49147,8 @@ "issue:2453", "issue:2462", "issue:2838", + "issue:3385", + "issue:3427", "issue:3576", "issue:3593", "issue:3698", @@ -49156,8 +49185,6 @@ "issue:9410" ], "soft_pairs": [ - "issue:2453|issue:4241", - "issue:12369|issue:12457", "issue:3698|issue:4445", "issue:4241|issue:4876", "issue:4241|issue:6423", @@ -49184,216 +49211,205 @@ "issue:2453|issue:2838", "issue:5932|issue:7619", "issue:5274|issue:5425", - "issue:1962|issue:2195" + "issue:1962|issue:2195", + "issue:3385|issue:3427" ], - "prepared_review_unit_hash": "ca08df602f31c3b685cfd4aaefd84991b483cc971f9869f177a1d8f9b39ad59a", + "prepared_review_unit_hash": "a23ea9e51211db08377febf02ae14ebd6430d169449931f451d33af42a611df0", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 29, - "serialized_chars": 34067, - "estimated_input_tokens": 8517, - "estimated_eval_tokens": 17290 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 34301, + "estimated_input_tokens": 8576, + "estimated_eval_tokens": 17408 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issue set with a few clear duplicate families. The strongest matches are around SDXL/img2img denoising offsets, the DiagonalGaussianDistribution latent_dist error, Dreambooth channel-mismatch, and FLUX/MPS float64 support.", - "confidence": 0.92, - "canonical_issue_reason": "issue:5628 is the best canonical issue among the related bug reports: it gives a specific, well-diagnosed SDXL img2img scheduler/denoising offset problem and sits at the center of the accepted denoising-start/end duplicate family.", + "summary": "Mostly a heterogeneous cluster of unrelated feature requests and bug reports. The only strong duplicate family I see is the DreamBooth channel-mismatch error pair (5932/7619), with a second textual-inversion/DiagonalGaussianDistribution pair (378/384) also looking like the same underlying bug. The other soft pairs are broader similarity matches but not true duplicates.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 5932 is the clearest concrete bug report in the set and has the strongest duplicate match (7619) with a reproducible channel-mismatch traceback.", "canonical_pr_reason": null, - "best_issue_reason": "issue:5628 is the strongest overall issue to represent the cluster's real duplicate-bug content because it is precise, actionable, and directly overlaps with the other accepted scheduler/denoising reports.", + "best_issue_reason": "5932 is the best issue to keep as the canonical representative of its bug family; most other items are feature/support requests rather than the same defect.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:2453", - "right": "issue:4241", - "accept": false, - "reason": "Both are support-style requests, but one is about contributing and the other is about LoRA fine-tuning support; not the same underlying problem." - }, - { - "left": "issue:12369", - "right": "issue:12457", - "accept": true, - "reason": "Both ask for CineScale support; the titles differ mainly in specificity, but the feature request is the same." - }, { "left": "issue:3698", "right": "issue:4445", "accept": false, - "reason": "Different topics: converting the inpainting model to diffusers vs. how to finetune a LoRA model." + "reason": "Different asks: converting an inpainting model to diffusers vs guidance on LoRA finetuning." }, { "left": "issue:4241", "right": "issue:4876", "accept": false, - "reason": "One is a request for LoRA fine-tuning support, the other is about exposing a LoRA layer std argument; related area, but not the same bug/change." + "reason": "Both mention LoRA, but one is about fine-tuning support and the other is about exposing a layer-std argument; not the same change." }, { "left": "issue:4241", "right": "issue:6423", "accept": false, - "reason": "LoRA fine-tuning support request vs. reproducibility of fill50k results; different issues." + "reason": "Unrelated feature request vs reproduction issue for fill50k results." }, { "left": "issue:4445", "right": "issue:4876", "accept": false, - "reason": "General LoRA finetuning question vs. a specific parameter request for LoRA layers; not the same underlying change." + "reason": "Finetuning how-to vs LoRA initialization parameter request; different topics." }, { "left": "issue:4445", "right": "issue:692", "accept": false, - "reason": "A finetuning question is not the same as asking for a diffusion-model fine-tuning example." + "reason": "General fine-tuning example request is not the same as a LoRA finetuning question." }, { "left": "issue:3576", "right": "issue:3593", "accept": false, - "reason": "Both mention xformers, but they involve different features and symptoms (Textual Inversion TypeError vs. LoRA second-call failure)." + "reason": "Both involve xformers, but one is Textual Inversion and the other is LoRA/PyTorch 2.0 second-call failure; different code paths." }, { "left": "issue:4913", "right": "issue:6572", "accept": false, - "reason": "Different feature requests: one for ControlNet OpenPose SDXL, the other for a ControlNetXS SDXL inpaint pipeline." + "reason": "Different pipeline/model requests: ControlNet openpose for SDXL vs ControlNetXS inpaint pipeline." }, { "left": "issue:1755", "right": "issue:4609", "accept": false, - "reason": "No meaningful overlap; one is a generic/non-English issue and the other is a model availability request." + "reason": "No meaningful overlap beyond generic support-request noise; unrelated subjects." }, { "left": "issue:1755", "right": "issue:2311", "accept": false, - "reason": "Unrelated issues with no shared bug or change." + "reason": "Generic unrelated issue vs UniPC scheduler request." }, { "left": "issue:4922", "right": "issue:692", "accept": false, - "reason": "Android demo request vs. diffusion finetuning example request; unrelated." + "reason": "Android demo request is unrelated to diffusion fine-tuning examples." }, { "left": "issue:4750", "right": "issue:4978", "accept": false, - "reason": "Both involve inpaint preprocessing, but they request different preprocessors and do not clearly describe the same concrete bug/change." + "reason": "Both touch inpainting preprocessors, but they ask for different specific preprocessors and behaviors; not clearly the same change." }, { "left": "issue:200", "right": "issue:202", "accept": false, - "reason": "Both are test-related issues, but they target different files and test additions; not duplicates." + "reason": "Different test additions for different files/classes." }, { "left": "issue:11560", "right": "issue:9410", "accept": false, - "reason": "Both mention import errors, but one is generic and the other is a specific pipeline import failure; insufficient evidence they are the same bug." + "reason": "Generic import error is too vague to treat as the same as the SD3 ControlNet inpainting import failure." }, { "left": "issue:9031", "right": "issue:9032", "accept": false, - "reason": "Different FlaxStableDiffusionPipeline questions: safety checker disabling vs. a minimal Colab TPU example." + "reason": "Different FlaxStableDiffusionPipeline questions: safety checker vs TPU colab setup." }, { "left": "issue:4716", "right": "issue:4747", "accept": false, - "reason": "Different SDXL topics: long-prompt handling vs. missing model_index.json after loading a fine-tuned LoRA." - }, - { - "left": "issue:5628", - "right": "issue:5630", - "accept": true, - "reason": "Both describe SDXL/img2img denoising/scheduler issues with leftover noise or offset start/end behavior; same underlying code-path is plausible." + "reason": "SDXL long-prompt handling vs missing model_index.json after loading a fine-tuned LoRA; unrelated issues." }, { "left": "issue:378", "right": "issue:384", "accept": true, - "reason": "These are the same textual inversion / DiagonalGaussianDistribution latent_dist bug, with one issue phrased as the resulting AttributeError." + "reason": "Same Textual Inversion / DiagonalGaussianDistribution bug family; 384 reads like the concrete error manifestation of 378." }, { "left": "issue:1962", "right": "issue:3902", "accept": false, - "reason": "Both are about video generation, but they refer to different capabilities and likely different code paths." + "reason": "Both are video-related, but one is a stable-diffusion video-generation request and the other is a Control-A-Video request." }, { "left": "issue:9038", "right": "issue:9039", "accept": false, - "reason": "Different FlaxStableDiffusionPipeline requests: prompt weighting vs. loading LoRA weights." + "reason": "Prompt weighting and LoRA loading are separate API requests, not one duplicate bug." }, { "left": "issue:2462", "right": "issue:5947", "accept": false, - "reason": "IterableDataset finetuning support vs. a diffusion-DPO policy objective discussion; unrelated." + "reason": "IterableDataset finetuning support and Diffusion-DPO policy discussion are unrelated." }, { "left": "issue:11519", "right": "issue:12369", "accept": false, - "reason": "MAGI-1 support request and CineScale support request are unrelated features." + "reason": "Different model support requests: MAGI-1 vs Wan 2.1 CineScale up-to-4K." }, { "left": "issue:5447", "right": "issue:5628", - "accept": true, - "reason": "Both concern denoising_start/end behavior in SDXL img2img with scheduler-specific offsets; same bug family." + "accept": false, + "reason": "Both concern denoising start/end behavior, but they target different scheduler configurations and symptoms; too broad to merge as one bug." }, { "left": "issue:9047", "right": "issue:9224", - "accept": true, - "reason": "The FLUX-on-MPS failure is directly consistent with the hardcoded torch.float64-on-MPS limitation." + "accept": false, + "reason": "MPS failure vs a specific float64-on-Metal limitation; related theme, but not clearly the same underlying defect from the available evidence." }, { "left": "issue:2453", "right": "issue:2838", "accept": false, - "reason": "Different support questions: contributing guidance vs. ControlNet pipeline behavior." + "reason": "Contribution questions and ControlNet pipeline behavior are unrelated." }, { "left": "issue:5932", "right": "issue:7619", "accept": true, - "reason": "Both report the same channel-mismatch runtime error in Dreambooth training, just with different wording and model context." + "reason": "Same concrete DreamBooth channel-count mismatch; 7619 is essentially the same error class as 5932." }, { "left": "issue:5274", "right": "issue:5425", "accept": false, - "reason": "Different training/runtime problems: mixed_precision dtype mismatch vs. checkpoint-saving compatibility with accelerate/PT 2.1." + "reason": "Different training/checkpointing problems; no shared concrete bug." }, { "left": "issue:1962", "right": "issue:2195", "accept": false, - "reason": "Video generation with stable diffusion vs. Tune A Video are related only at a very high level, not the same change or bug." + "reason": "Both mention video, but they are different feature requests with different targets." + }, + { + "left": "issue:3385", + "right": "issue:3427", + "accept": false, + "reason": "Distinct model-support requests for different JAX integrations." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is grounded in the packet and stays mostly conservative. The accepted soft-edge pairs are defensible from the titles/bodies, especially the SDXL img2img denoising family, the DiagonalGaussianDistribution error pair, the Dreambooth channel-mismatch pair, and the FLUX/MPS float64 pair. The non-accepted pairs are also appropriately cautious and generally distinguish feature/support requests from distinct bugs or requests." + "feedback": "The summary is grounded in the packet and the soft-edge verdicts are generally conservative. The main duplicate family callout (5932/7619) is well supported, and the other pairs are mostly reasonable false negatives. Minor caution: avoid wording that implies extra evidence not shown in the packet (for example, a specific 'traceback') unless it appears in the issue text." }, "evaluator_used": true, "retried": false, @@ -49423,6 +49439,7 @@ "issue:11630", "issue:11878", "issue:12097", + "issue:12499", "issue:12567", "issue:12732", "issue:12738", @@ -49439,9 +49456,7 @@ "issue:2838", "issue:2984", "issue:338", - "issue:3385", "issue:339", - "issue:3427", "issue:3698", "issue:3699", "issue:4141", @@ -49464,7 +49479,6 @@ "issue:9825" ], "soft_pairs": [ - "issue:3385|issue:3427", "issue:7972|issue:8241", "issue:671|issue:743", "issue:3698|issue:5162", @@ -49488,186 +49502,187 @@ "issue:12732|issue:12738", "issue:12567|issue:13105", "issue:11432|issue:11555", - "issue:2195|issue:2984" + "issue:2195|issue:2984", + "issue:12097|issue:12499" ], - "prepared_review_unit_hash": "29cd06f3b212e1d96687e95116ce25112a781b518326c2a24e191fb6aba26ffa", + "prepared_review_unit_hash": "2b7e286bfe17b38d35cb4426e584e10db1be9f6a8c5a1e29b7e4d046c3546054", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 33227, - "estimated_input_tokens": 8307, - "estimated_eval_tokens": 16870 + "serialized_chars": 32732, + "estimated_input_tokens": 8183, + "estimated_eval_tokens": 16622 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issue titles with only broad topical overlap; no soft pair looks like the exact same bug or change, so this cluster does not support a strong duplicate canonical.", - "confidence": 0.41, - "canonical_issue_reason": "No single canonical issue stands out: the items span many different models, pipelines, and failure modes, and the overlaps are mostly broad topic similarity rather than duplicate reports.", + "summary": "This cluster is mostly a grab-bag of unrelated Diffusers issues and feature requests; none of the soft pairs look like true duplicates. The most representative issue is the broad FLUX pipeline request.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 9042 is the broadest, most central request in the FLUX subgroup, and it has nearby follow-up interest (e.g. FLUX inpainting/IPAdapter requests), so it is the best anchor for this cluster.", "canonical_pr_reason": null, - "best_issue_reason": "No clear best issue for canonical triage; several candidates are broad feature requests, but none cleanly subsumes the rest as the same underlying defect or change.", + "best_issue_reason": "Issue 9042 is the best overall representative because it is a clear, high-signal feature request with broader relevance than the more niche or noisy reports in the set.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:3385", - "right": "issue:3427", - "accept": false, - "reason": "Both are JAX-related feature requests, but they target different models and different work; not the same change." - }, { "left": "issue:7972", "right": "issue:8241", "accept": false, - "reason": "Both mention LEDITS++, but one is a type-mismatch bug and the other is a model-size concern; different underlying problems." + "reason": "Same model family (LEDITS++) but different bugs: type mismatch vs abnormal model size." }, { "left": "issue:671", "right": "issue:743", "accept": false, - "reason": "Both concern conversion, but one reports non-identical outputs after conversion while the other cannot convert a ckpt at all; different bugs." + "reason": "Both concern conversion, but one is about non-identical outputs while the other is a conversion failure; different concrete problems." }, { "left": "issue:3698", "right": "issue:5162", "accept": false, - "reason": "Inpainting-model conversion and folder-to-safetensors conversion are unrelated tasks." + "reason": "Different requests: converting an inpainting model to Diffusers vs converting a folder checkpoint to safetensors." }, { "left": "issue:2838", "right": "issue:5947", "accept": false, - "reason": "ControlNet pipeline behavior and Diffusion-DPO policy objectives are different topics." + "reason": "ControlNet pipeline behavior request and Diffusion-DPO policy discussion are unrelated." }, { "left": "issue:1755", "right": "issue:5480", "accept": false, - "reason": "Generic discussion and EDM implementation request are not the same issue." + "reason": "Generic title vs EDM implementation request; no sign of the same underlying issue." }, { "left": "issue:9042", "right": "issue:9264", "accept": false, - "reason": "Adding the base Flux pipeline and adding a Flux inpainting model are related but distinct feature requests." + "reason": "Both are FLUX-related, but one asks for the base pipeline and the other for an inpainting model; separate feature work." }, { "left": "issue:11630", "right": "issue:11878", "accept": false, - "reason": "Single-file loading support and GGUF 1.3B failure are different WanVACE issues." + "reason": "Support for from_single_file and a GGUF 1.3B runtime failure are different problems." }, { "left": "issue:201", "right": "issue:338", "accept": false, - "reason": "Both are test-related, but they cover different test classes and different scheduler behavior." + "reason": "Different tests for different components; not the same failure or change." }, { "left": "issue:201", "right": "issue:339", "accept": false, - "reason": "Same as above: different tests and different target behavior." + "reason": "Different test coverage targets (attention class vs Karras VE)." }, { "left": "issue:2221", "right": "issue:2268", "accept": false, - "reason": "Both mention ColossalAI, but the reports are too generic to establish the same concrete bug." + "reason": "Both mention ColossalAI errors, but the titles are too generic to establish the same root cause." }, { "left": "issue:10049", "right": "issue:11560", "accept": false, - "reason": "Importing MochiPipeline and a generic import error are not enough to prove the same failure." + "reason": "MochiPipeline import failure and a generic import error are not enough to identify the same bug." }, { "left": "issue:4913", "right": "issue:5926", "accept": false, - "reason": "Different model requests with no shared bug or change." + "reason": "Different model/package requests entirely: controlnet-openpose-sdxl-1.0 vs Kandinsky 3.0." }, { "left": "issue:13232", "right": "issue:13351", "accept": false, - "reason": "LTX-2.3 support and Wan/Flux ControlNet support are unrelated feature requests." + "reason": "LTX-2.3 support and Wan/Flux controlnet support are distinct feature requests." }, { "left": "issue:1158", "right": "issue:1160", "accept": false, - "reason": "One is test interference in config tests; the other is a hang in a local custom pipeline test." + "reason": "Test interference in config tests and a hanging custom-pipeline test are different issues." }, { "left": "issue:9042", "right": "issue:9825", "accept": false, - "reason": "Flux pipeline support and IPAdapter support for Flux are separate enhancements, not one merged change." + "reason": "FLUX base pipeline support vs IPAdapter support for FLUX pipelines; related area but not the same change." }, { "left": "issue:4835", "right": "issue:5112", "accept": false, - "reason": "Same pipeline family, but one bug is MultiControlNet behavior and the other is a float-to-size construction error." + "reason": "Same pipeline class, but one bug is MultiControlNet behavior and the other is a torch.Size float-shape error." }, { "left": "issue:6042", "right": "issue:6572", "accept": false, - "reason": "SDXL-turbo distillation and ControlNetXS inpainting are different features." + "reason": "SDXL-turbo distillation example request and ControlNetXS inpaint pipeline request are different features." }, { "left": "issue:2830", "right": "issue:4609", "accept": false, - "reason": "Image/text-to-video with ModelScope and a T2I-Adapter style model are unrelated requests." + "reason": "A video-generation question and a model availability request are unrelated." }, { "left": "issue:3699", "right": "issue:4141", "accept": false, - "reason": "CUDA-on-macOS import error and second-controlnet handling are completely different issues." + "reason": "Different bugs: CUDA-on-M2 assertion vs ControlNet inpaint not respecting the second controlnet." }, { "left": "issue:12097", "right": "issue:12896", "accept": false, - "reason": "VRAM OOM on WAN 2.2 and MPS-vs-CUDA behavior are different device/runtime problems." + "reason": "WAN 2.2 VRAM OOM at the end and an MPS-specific failure are different platform/runtime problems." }, { "left": "issue:12732", "right": "issue:12738", "accept": false, - "reason": "Both are Z-Image related, but one is a generic inference error and the other is a specific batch-size assertion failure." + "reason": "Both are Z-Image pipeline errors, but one is generic inference failure and the other is a batch-size assertion; not clearly the same bug." }, { "left": "issue:12567", "right": "issue:13105", "accept": false, - "reason": "Both mention noise, but one is tied to Ulysses Attention and the other to a Wan T2V model output issue; not the same bug." + "reason": "Output noise under Ulysses Attention vs WAN 2.1 T2V noise are different model/code-path reports." }, { "left": "issue:11432", "right": "issue:11555", "accept": false, - "reason": "torch_dtype=auto and device_map=auto are different from-pretrained behaviors." + "reason": "torch_dtype='auto' not working and device_map='auto' support are different API concerns." }, { "left": "issue:2195", "right": "issue:2984", "accept": false, - "reason": "Both are video-generation topics, but one is a feature request and the other is a specific long-video model proposal." + "reason": "Tune A Video and latent video diffusion long-video generation are different feature requests." + }, + { + "left": "issue:12097", + "right": "issue:12499", + "accept": false, + "reason": "WAN 2.2 VRAM OOM and WAN 2.2 I2V shape mismatch are different failure modes." } ] }, @@ -49686,8 +49701,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -49699,12 +49714,11 @@ "issue:11653", "issue:11821", "issue:11878", - "issue:12097", - "issue:12499", "issue:12606", "issue:1755", "issue:1962", "issue:2015", + "issue:2195", "issue:2453", "issue:2635", "issue:321", @@ -49741,7 +49755,6 @@ "issue:9220" ], "soft_pairs": [ - "issue:12097|issue:12499", "issue:8493|issue:8738", "issue:11315|issue:11436", "issue:4269|issue:4615", @@ -49767,198 +49780,199 @@ "issue:4913|issue:6042", "issue:1755|issue:4042", "issue:5017|issue:6423", - "issue:5017|issue:7303" + "issue:5017|issue:7303", + "issue:2195|issue:3838" ], - "prepared_review_unit_hash": "0083607b53d13a14e3c12eebcdba86fa4df31e64cc6b60cee6b00a98e1ca1f9e", + "prepared_review_unit_hash": "03903b07fdff4db092d807e9722b29e500d3d55f5665fe844bc48547628aecda", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 34211, - "estimated_input_tokens": 8553, - "estimated_eval_tokens": 17362 + "serialized_chars": 33657, + "estimated_input_tokens": 8415, + "estimated_eval_tokens": 17086 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly a grab bag of unrelated issues across different models, pipelines, and training scripts; none of the soft pairs look like true duplicates under a strict same-bug standard.", - "confidence": 0.94, - "canonical_issue_reason": "issue:8493 is the most concrete, well-scoped bug report in the set (SD3Transformer2DModel state_dict loading failure) and has enough diagnostic detail to serve as the best anchor issue.", + "summary": "This set does not form a single duplicate cluster; it contains many unrelated issue reports. All soft-pair candidates are false positives and should be kept separate.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "It is the clearest representative of a specific actionable defect rather than a feature request or vague usage question.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:12097", - "right": "issue:12499", - "accept": false, - "reason": "Both are Wan 2.2 reports, but one is TI2V end-of-run OOM and the other is an I2V condition-shape mismatch; different failure modes and code paths." - }, { "left": "issue:8493", "right": "issue:8738", "accept": false, - "reason": "Both mention SD3Transformer2DModel, but one is a state_dict loading error and the other is a configuration issue; not the same underlying bug." + "reason": "Both concern SD3Transformer2DModel, but one is a state_dict loading error and the other is a configuration issue; similar subsystem, not the same bug." }, { "left": "issue:11315", "right": "issue:11436", "accept": false, - "reason": "HiDream-related, but one is RoPE float64 on MPS and the other is a Half vs BFloat16 inference dtype mismatch; distinct defects." + "reason": "Both are HiDream dtype-related, but one is a float64 RoPE/MPS failure and the other is a Half vs BFloat16 matmul mismatch; different code paths and symptoms." }, { "left": "issue:4269", "right": "issue:4615", "accept": false, - "reason": "A TypeError in check_inputs is unrelated to an import-name failure for the ControlNet pipeline." + "reason": "Different problems: a TypeError from default args in check_inputs versus an import-name failure for ControlNetPipeline." }, { "left": "issue:6287", "right": "issue:8803", "accept": false, - "reason": "Unused dependencies in setup.py and a hidden_states duplicate assignment are unrelated changes/bugs." + "reason": "One is dependency cleanup in setup.py, the other is a duplicate assignment in attention_processor.py; no duplicate bug/change." }, { "left": "issue:11436", "right": "issue:11653", "accept": false, - "reason": "One is a runtime dtype mismatch during inference; the other is support for .alpha LoRA keys. Different problems." + "reason": "One reports a runtime dtype mismatch in HiDream inference; the other requests support for .alpha LoRA keys. Different issues." }, { "left": "issue:2453", "right": "issue:5017", "accept": false, - "reason": "General contribution question vs a request for control mode in ControlNet; not duplicates." + "reason": "General contribution question versus a feature request for control mode in ControlNet; unrelated." }, { "left": "issue:10760", "right": "issue:12606", "accept": false, - "reason": "Both are import-related, but one is a missing attribute in an ONNX pipeline and the other is a wrong import in a conversion script." + "reason": "Both mention imports, but they refer to different scripts and different import failures." }, { "left": "issue:11821", "right": "issue:11878", "accept": false, - "reason": "Both are WanVACE issues, but one is prompt-list support and the other is GGUF model loading for 1.3B; different concrete issues." + "reason": "WanVACE prompt-list support and WanVACE GGUF 1.3B loading are distinct issues." }, { "left": "issue:10120", "right": "issue:8560", "accept": false, - "reason": "Both are SD3 memory problems, but one is DreamBooth memory consumption and the other is training OOM on a 4090; not the same bug report." + "reason": "Both mention SD3 training memory/OOM, but one is DreamBooth+SD3 memory consumption and the other is a 4090 OOM during SD3 training; not the same concrete bug." }, { "left": "issue:1076", "right": "issue:321", "accept": false, - "reason": "Stable-diffusion-v1-5 question vs Blended Diffusion; unrelated topics." + "reason": "Completely different topics: stable-diffusion-v1-5 discussion versus blended diffusion." }, { "left": "issue:3728", "right": "issue:7408", "accept": false, - "reason": "Different proposed pipelines (Shifted Diffusion vs FRESCO), not the same underlying change." + "reason": "Different proposed pipelines/models: Shifted Diffusion versus FRESCO." }, { "left": "issue:4225", "right": "issue:4406", "accept": false, - "reason": "ControlNet state_dict loading errors are different from a safetensor header deserialization failure." + "reason": "ControlNet state_dict loading errors are not the same as a safetensor header deserialization failure." }, { "left": "issue:2015", "right": "issue:6572", "accept": false, - "reason": "Categorical diffusion pipeline request vs ControlNetXS SDXL inpaint pipeline request; distinct pipeline proposals." + "reason": "Categorical diffusion pipeline and ControlNetXS-SDXL-Inpaint-Pipeline are unrelated feature requests." }, { "left": "issue:321", "right": "issue:322", "accept": false, - "reason": "Different research methods (Blended Diffusion vs V-Diffusion), not duplicates." + "reason": "Different diffusion methods/models: Blended Diffusion versus V-Diffusion." }, { "left": "issue:3788", "right": "issue:7408", "accept": false, - "reason": "UniControl and FRESCO are separate pipeline requests with different methods and use cases." + "reason": "UniControl and FRESCO are separate pipeline requests with different functionality." }, { "left": "issue:6444", "right": "issue:6482", "accept": false, - "reason": "Both involve SDXL LoRA training scripts, but one is a peft_config/checkpointing bug and the other is a resume_from_checkpoint FP16 unscale error." + "reason": "Both are training-script failures, but they are different errors triggered by different conditions." }, { "left": "issue:4232", "right": "issue:5284", "accept": false, - "reason": "Instruct-pix2pix with controlnet condition is a different feature/implementation request than ControlNet-LLLite support." + "reason": "Instruct-pix2pix via ControlNet condition is not the same change as ControlNet-LLLite support." }, { "left": "issue:6008", "right": "issue:6393", "accept": false, - "reason": "LCM LoRA interaction with IP adapter face is unrelated to private HF link download failures in from_single_file." + "reason": "IP-Adapter/LCM LoRA failure and private HF download failure through from_single_file are unrelated." }, { "left": "issue:9218", "right": "issue:9220", "accept": false, - "reason": "Both are PAG-related, but one is a guess-mode compatibility request and the other is an undefined-variable bug in pipeline_pag_sd; not the same issue." + "reason": "PAG guess-mode support request versus an undefined-variable bug in PAG pipeline code; not the same bug." }, { "left": "issue:1755", "right": "issue:2635", "accept": false, - "reason": "The Arabic-titled issue and the scheduler-identification question are unrelated." + "reason": "General discussion thread versus a scheduler naming question; no duplicate relationship." }, { "left": "issue:8477", "right": "issue:8894", "accept": false, - "reason": "Both are training-script errors, but they concern different scripts and failure details; too broad to treat as the same bug." + "reason": "Both involve training scripts, but they reference different scripts and likely different distributed-training failures." }, { "left": "issue:1962", "right": "issue:3838", "accept": false, - "reason": "Both are about video generation, but one is a general stable-diffusion question and the other is a specific Video Composer pipeline request." + "reason": "Video generation with stable diffusion and Video Composer are distinct topics." }, { "left": "issue:4913", "right": "issue:6042", "accept": false, - "reason": "ControlNet OpenPose SDXL vs SDXL-Turbo distilling example; different models/tasks." + "reason": "ControlNet OpenPose SDXL model discussion is unrelated to an SDXL-turbo distillation example request." }, { "left": "issue:1755", "right": "issue:4042", "accept": false, - "reason": "The Arabic-titled issue is unrelated to combining reference-only with inpainting and depth control." + "reason": "General thread versus a question about combining reference-only, inpainting, and depth control; not duplicates." }, { "left": "issue:5017", "right": "issue:6423", "accept": false, - "reason": "Control mode for ControlNet is unrelated to a fill50k reproducibility question." + "reason": "Control mode request for ControlNet and inability to reproduce fill50k results are unrelated." }, { "left": "issue:5017", "right": "issue:7303", "accept": false, - "reason": "Control mode request and zero_snr scheduler support are different feature requests with no shared concrete bug/change." + "reason": "Control mode for ControlNet and enabling zero_snr via from_pretrained are unrelated." + }, + { + "left": "issue:2195", + "right": "issue:3838", + "accept": false, + "reason": "Tune A Video and Video Composer are different video-generation projects/requests." } ] }, @@ -49997,9 +50011,8 @@ "issue:13065", "issue:13232", "issue:1976", - "issue:2195", + "issue:3113", "issue:3698", - "issue:3838", "issue:3861", "issue:3902", "issue:4042", @@ -50032,7 +50045,6 @@ "issue:9147" ], "soft_pairs": [ - "issue:2195|issue:3838", "issue:6095|issue:6096", "issue:5770|issue:6288", "issue:3698|issue:6161", @@ -50055,180 +50067,181 @@ "issue:4789|issue:6393", "issue:3902|issue:4042", "issue:10737|issue:9141", - "issue:10049|issue:12606" + "issue:10049|issue:12606", + "issue:3113|issue:3902" ], - "prepared_review_unit_hash": "b302fdb2f36d404528c17bddd9e56a22f1815d979b0f2f0051fe827c6ea4affb", + "prepared_review_unit_hash": "6f9542f464186bdc421a43910a3a8f58f746caa86b79dce5b1859ba48dfe395e", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33340, - "estimated_input_tokens": 8335, - "estimated_eval_tokens": 16926 + "serialized_chars": 32878, + "estimated_input_tokens": 8220, + "estimated_eval_tokens": 16696 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly separate bug reports and feature requests around different pipelines/models, not duplicates of one another. The soft-similarity pairs generally share a subsystem or theme but not the same concrete failure or change.", - "confidence": 0.89, - "canonical_issue_reason": "issue:8488 is the clearest concrete bug report in the set: a specific StableDiffusion3Pipeline `from_pretrained` exception with enough detail to represent the loading-failure cluster.", + "summary": "This is not a true duplicate cluster: the soft links mostly connect superficially similar diffusers issues (imports, Flux/SD3, controlnet, model support) that actually describe different bugs, features, or code paths.", + "confidence": 0.88, + "canonical_issue_reason": "No single canonical issue stands out because the set is heterogeneous and none of the soft pairs share the same concrete failure mode.", "canonical_pr_reason": null, - "best_issue_reason": "issue:8488 is the strongest global representative because it is specific, reproducible, and about a core code path rather than a vague request or broad topic.", + "best_issue_reason": "No global best issue for deduplication; if forced, issue:9141 is the cleanest self-contained bug report, but it is not representative of the rest of the set.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:2195", - "right": "issue:3838", - "accept": false, - "reason": "Both are video-related, but one is about Tune-A-Video and the other about Video Composer; different models and different problems." - }, { "left": "issue:6095", "right": "issue:6096", "accept": false, - "reason": "Same SVD area, but one is an IndexError in scheduling and the other is a torch.compile DynamicOutputShapeException; not the same bug." + "reason": "Both involve SVD, but one is an IndexError in scheduling_euler_discrete and the other is a torch.compile DynamicOutputShapeException; different code paths." }, { "left": "issue:5770", "right": "issue:6288", "accept": false, - "reason": "LCM distillation dataset access vs. LoRA learning rate tuning are unrelated training issues." + "reason": "One is about dataset access preventing LCM distillation; the other is about the default learning rate being too small for LCM LoRA training." }, { "left": "issue:3698", "right": "issue:6161", "accept": false, - "reason": "Inpainting model conversion to diffusers is unrelated to the sigma-scheduling bug." + "reason": "Inpainting model conversion to diffusers is unrelated to the sigma-scheduling bug about discarding the next-to-last sigma." }, { "left": "issue:11560", "right": "issue:8975", "accept": false, - "reason": "Both mention import errors, but they are about different symbols and likely different root causes." + "reason": "Both mention import errors, but one is generic and the other is specifically about LoraLoaderMixin import removal; not enough to treat as the same bug." }, { "left": "issue:12369", "right": "issue:13232", "accept": false, - "reason": "Both are feature/support requests for different models, not one underlying change." + "reason": "Different model requests entirely: Wan 2.1 CineScale up to 4K vs LTX-2.3 support." }, { "left": "issue:10893", "right": "issue:13065", "accept": false, - "reason": "Both involve controlnet, but for different models and different requested capabilities." + "reason": "Both are control-related feature requests, but they target different model families and different controlnet capabilities." }, { "left": "issue:10460", "right": "issue:9026", "accept": false, - "reason": "Missing `mid_block_type` on UNet2DModel is unrelated to the `load_checkpoint_and_dispatch` keyword error." + "reason": "Missing documented UNet2DModel argument is unrelated to the load_checkpoint_and_dispatch unexpected force_hook argument error." }, { "left": "issue:5023", "right": "issue:9147", "accept": false, - "reason": "Both are import/autopipeline-related, but they concern different pipelines and failure modes." + "reason": "WuerstchenCombinedPipeline import failure and Flux autopipeline error are distinct pipeline/import issues." }, { "left": "issue:9001", "right": "issue:9112", "accept": false, - "reason": "Both are Auraflow issues, but aspect-ratio support and a tensor-shape mismatch are not the same bug." + "reason": "Aspect ratio support request is not the same as a tensor-shape mismatch during Auraflow inference." }, { "left": "issue:4241", "right": "issue:5017", "accept": false, - "reason": "LoRA fine-tuning for unidiffusers and ControlNet control mode are unrelated feature requests." + "reason": "LoRA fine-tuning for unidiffusers and control mode for controlnet are different feature requests." }, { "left": "issue:5227", "right": "issue:8518", "accept": false, - "reason": "Both touch torch.compile, but they are different code paths, models, and exception types." + "reason": "Both touch torch.compile, but one is about load_lora_weight failing under compile and the other is an unsupported SD3 compile trace; different failure modes." }, { "left": "issue:10186", "right": "issue:10204", "accept": false, - "reason": "ImportError for `FluxFillPipeline` and inpainting quality degradation are unrelated issues." + "reason": "FluxFillPipeline import error and FluxFillPipeline inpainting quality degradation are unrelated issues." }, { "left": "issue:8488", "right": "issue:8565", "accept": false, - "reason": "Both are SD3 pipeline issues, but one is a `from_pretrained` exception and the other is a `load_lora_weights` IndexError." + "reason": "StableDiffusion3Pipeline from_pretrained exception and load_lora_weights IndexError affect different APIs and likely different fixes." }, { "left": "issue:8503", "right": "issue:8511", "accept": false, - "reason": "Gradient checkpointing crashes and text-encoder unloading in SD3 Dreambooth/LoRA are different defects." + "reason": "Gradient checkpointing crash in SD3 is unrelated to the script not unloading text encoders." }, { "left": "issue:3861", "right": "issue:4250", "accept": false, - "reason": "Both mention `from_pretrained`, but the titles indicate different load/usage problems with no clear shared root cause." + "reason": "Both mention from_pretrained, but one is about the API form/naming and the other about incorrect model loading behavior; not the same bug." }, { "left": "issue:5318", "right": "issue:5341", "accept": false, - "reason": "Both are PEFT-related documentation/example tasks, but they are separate content updates rather than one bug or change." + "reason": "These are separate PEFT documentation/example updates, not the same underlying issue." }, { "left": "issue:1160", "right": "issue:1976", "accept": false, - "reason": "A hanging custom pipeline test and a tokenizer/config file lookup error are unrelated." + "reason": "A hanging pytest test is not the same as an incorrect tokenizer/config.json file lookup." }, { "left": "issue:10658", "right": "issue:12624", "accept": false, - "reason": "`provider_options` in `InferenceSession` and `strength` for `wanimagetovideopipeline` are unrelated API requests." + "reason": "onnxruntime provider_options and WanImageToVideoPipeline strength are unrelated API requests." }, { "left": "issue:4524", "right": "issue:5118", "accept": false, - "reason": "AnimateDiff support and ControlNet reference-only usage are different feature requests." + "reason": "AnimateDiff integration and ControlNet reference_only usage are different topics." }, { "left": "issue:4789", "right": "issue:6393", "accept": false, - "reason": "Both involve download/load failures, but one is DeepFloyd IF and the other is private HF link handling via `from_single_file`." + "reason": "Both involve downloads, but one is DeepFloyd IF download failure and the other is private HF link handling via from_single_file; different causes." }, { "left": "issue:3902", "right": "issue:4042", "accept": false, - "reason": "Control-A-Video and combining reference-only with inpainting/depth control are different feature questions." + "reason": "control-a-video and combining reference-only with inpainting/depth control are distinct feature requests." }, { "left": "issue:10737", "right": "issue:9141", "accept": false, - "reason": "Both are Flux pipeline issues, but one is missing `image_encoder` during encode and the other is missing `from_single_file()`." + "reason": "Both concern FluxPipeline, but one lacks image_encoder in encode_image and the other lacks from_single_file; different APIs." }, { "left": "issue:10049", "right": "issue:12606", "accept": false, - "reason": "A MochiPipeline import problem and a wrong import in a conversion script are different code paths and failures." + "reason": "Cannot import MochiPipeline is unrelated to wrong import behavior in scripts/convert_sd3_to_diffusers.py." + }, + { + "left": "issue:3113", + "right": "issue:3902", + "accept": false, + "reason": "Training-free layout control with cross-attention guidance is a different control idea from control-a-video." } ] }, @@ -50247,8 +50260,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -50272,7 +50285,6 @@ "issue:2737", "issue:2822", "issue:2830", - "issue:3113", "issue:3140", "issue:338", "issue:3467", @@ -50288,6 +50300,8 @@ "issue:5034", "issue:5168", "issue:5314", + "issue:5436", + "issue:5794", "issue:6572", "issue:6866", "issue:6901", @@ -50301,7 +50315,6 @@ "issue:9825" ], "soft_pairs": [ - "issue:3113|issue:3902", "issue:2195|issue:2830", "issue:2822|issue:3140", "issue:12060|issue:13101", @@ -50326,198 +50339,199 @@ "issue:4913|issue:5168", "issue:10467|issue:10526", "issue:4750|issue:6572", - "issue:10817|issue:8503" + "issue:10817|issue:8503", + "issue:5436|issue:5794" ], - "prepared_review_unit_hash": "c1dbfaad0f5f37cfad1177e82161ccb7047416b25e0f882c4d8f0ab6d2cef063", + "prepared_review_unit_hash": "1757c7506c0119bbed7a2d973e045a082e689e5364e86f624cc7f32b24ee284f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 33100, - "estimated_input_tokens": 8275, - "estimated_eval_tokens": 16806 + "serialized_chars": 33651, + "estimated_input_tokens": 8413, + "estimated_eval_tokens": 17082 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly heterogeneous. Only two soft pairs look like true duplicates: the inpainting fp16/half-precision load failure and the FlowMatchEulerDiscreteScheduler timestep precision/indexing bug. The rest are adjacent feature requests or broad subsystem overlaps, not the same underlying issue.", - "confidence": 0.77, - "canonical_issue_reason": "issue:11749 is the strongest canonical issue candidate: it is a narrowly scoped runtime bug with a concrete failing code path, and it has a close duplicate-like companion in issue:9331.", + "summary": "Mostly unrelated Diffusers issues; only a few soft pairs look like true duplicates. No PRs are present, and there is no single umbrella issue for the whole set.", + "confidence": 0.78, + "canonical_issue_reason": "issue:2440 is the clearest duplicate target in the only strong issue family (float16 inpainting load errors), but the overall set is too heterogeneous to have one true canonical.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11749 is the best global issue representative because it is specific, technical, and likely to anchor a real duplicate cluster without being a broad feature request.", + "best_issue_reason": "issue:2440 is the best standalone anchor for triage: specific, reproducible, and directly matches the strongest near-duplicate pair.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:3113", - "right": "issue:3902", - "accept": false, - "reason": "Both concern video/control-related ideas, but they describe different feature requests and not the same bug or change." - }, { "left": "issue:2195", "right": "issue:2830", "accept": false, - "reason": "Both ask for video-generation capability, but the model/use-case differs and there is no shared concrete bug." + "reason": "Different tasks: video generation request vs a ModelScope video-use question." }, { "left": "issue:2822", "right": "issue:3140", "accept": false, - "reason": "One is a training-quality complaint and the other is an import error; same area, different failure mode." + "reason": "Training slowdown/error in ControlNet is unrelated to an import failure." }, { "left": "issue:12060", "right": "issue:13101", "accept": false, - "reason": "The second item is too vague to establish the same underlying issue as the DiT-block question." + "reason": "A question about DiT block availability is unrelated to a malformed huggingface issue." }, { "left": "issue:1925", "right": "issue:2440", "accept": true, - "reason": "Both report inpainting model loading failing in float16/half precision, which points to the same underlying code-path bug." + "reason": "Both report inpainting models failing in float16/half precision; same underlying loading/runtime bug." }, { "left": "issue:1594", "right": "issue:3728", "accept": false, - "reason": "These are unrelated feature requests in the video/text-to-image space, not the same issue." + "reason": "Img2Img video-generation discussion and a shifted-diffusion text-to-image pipeline request are different changes." }, { "left": "issue:8503", "right": "issue:8507", "accept": false, - "reason": "Both are SD3-related, but one is a gradient-checkpointing crash and the other is a train_text_encoder failure; not the same bug." + "reason": "Same SD3 area, but gradient-checkpointing crashes and train_text_encoder failures are different bugs." }, { "left": "issue:10421", "right": "issue:12754", "accept": false, - "reason": "Both involve pipeline movement/precision, but one is a CPU memory leak and the other is unintended float32 conversion." + "reason": "Multi-GPU memory leak and from_pipe dtype conversion are different problems and code paths." }, { "left": "issue:10950", "right": "issue:9680", "accept": false, - "reason": "Docs broken versus SDXL runtime failure are unrelated." + "reason": "Broken docs for Accelerate inference and broken SDXL unconditional generation are unrelated." }, { "left": "issue:10421", "right": "issue:12599", "accept": false, - "reason": "Both concern moving pipelines to CUDA, but the symptoms differ: memory leak versus slowness." + "reason": "CUDA move slowness is not the same bug as a memory leak when moving pipelines across GPUs." }, { "left": "issue:4305", "right": "issue:5034", "accept": false, - "reason": "torch.compile graph breaks and progress bar display issues are unrelated bugs." + "reason": "torch.compile graph breaks and checkpoint-loading progress-bar glitches are unrelated." }, { "left": "issue:9148", "right": "issue:9825", "accept": false, - "reason": "Both are FLUX pipeline feature requests, but they target different capabilities." + "reason": "Flux inpainting pipeline support and FLUX IP-Adapter support are separate feature requests." }, { "left": "issue:3467", "right": "issue:4733", "accept": false, - "reason": "Both involve training resume behavior, but the reported failures and code paths differ." + "reason": "Resume-step bug in ControlNet training is different from the DreamBooth LoRA optimizer-state error." }, { "left": "issue:5314", "right": "issue:9264", "accept": false, - "reason": "Different inpainting-related feature requests for different models; not the same change." + "reason": "ControlNet reference inpainting and Flux inpainting are distinct pipeline requests." }, { "left": "issue:2830", "right": "issue:3838", "accept": false, - "reason": "Both are video-generation requests, but they are distinct feature asks with no shared concrete bug." + "reason": "A generic video-generation question and a Video Composer request are not the same concrete issue." }, { "left": "issue:11749", "right": "issue:9331", "accept": true, - "reason": "Both describe the same FlowMatchEulerDiscreteScheduler timestep/indexing precision problem, just from different angles." + "reason": "Both describe the same FlowMatchEulerDiscreteScheduler timestep/indexing precision bug." }, { "left": "issue:4972", "right": "issue:5009", - "accept": false, - "reason": "Both are Wuerstchen-related, but one is example code failing and the other is a generic error report; too little evidence of sameness." + "accept": true, + "reason": "Both are wuerstchen runtime/example failures and appear to be the same underlying issue." }, { "left": "issue:11560", "right": "issue:9062", "accept": false, - "reason": "A generic import error is not enough to conclude it is the same issue as the StableAudioPipeline missing import." + "reason": "A vague import error is too broad to treat as the same issue as StableAudioPipeline import failure." }, { "left": "issue:272", "right": "issue:338", "accept": false, - "reason": "A feature request for inpainting support and a test request for LMS Discrete are not duplicates." + "reason": "K-LMS inpainting support and LMS-discrete test coverage are different tasks." }, { "left": "issue:10460", "right": "issue:6901", "accept": false, - "reason": "Missing UNet2DModel argument and missing FlaxDDIMScheduler method are different API gaps." + "reason": "Missing mid_block_type on UNet2DModel is unrelated to missing clip_sample on FlaxDDIMScheduler." }, { "left": "issue:1594", "right": "issue:3902", "accept": false, - "reason": "Both mention video, but one is an img2img feature request and the other is a control-a-video request." + "reason": "Img2Img for video generation and control-a-video are different feature requests." }, { "left": "issue:2737", "right": "issue:6866", "accept": false, - "reason": "Different feature requests for different pipelines/schedulers; no shared bug." + "reason": "CLIP guidance support and IP-Adapter support are separate feature changes." }, { "left": "issue:4913", "right": "issue:5168", "accept": false, - "reason": "Both are ControlNet-related support requests, but for different model variants and capabilities." + "reason": "ControlNet-openpose SDXL and ControlNet-XS support are related only broadly, not the same concrete change." }, { "left": "issue:10467", "right": "issue:10526", "accept": false, - "reason": "One is an example-code issue and the other is a specific optimum.quanto TypeError; not the same problem." + "reason": "An FP8 example-code artifact issue and a WeightQBytesTensor constructor TypeError are different bugs." }, { "left": "issue:4750", "right": "issue:6572", "accept": false, - "reason": "Both mention inpainting, but they are different pipeline/support requests." + "reason": "Inpaint + LaMa preprocessor and ControlNetXS-SDXL-Inpaint-Pipeline are different feature requests." }, { "left": "issue:10817", "right": "issue:8503", "accept": false, - "reason": "Missing SD3 control nets and SD3 gradient-checkpointing crashes are separate issues." + "reason": "SD3 ControlNet support and SD3 gradient-checkpointing crashes are unrelated." + }, + { + "left": "issue:5436", + "right": "issue:5794", + "accept": false, + "reason": "ControlNet XL training error and missing tokenizer/config.json are different problems." } ] }, "evaluator_result": { "accept": true, - "feedback": "The analysis is grounded in the packet and stays mostly conservative. The two accepted soft pairs are plausibly duplicates from the titles alone: issue:1925/2440 on inpainting fp16/float16 load failures, and issue:11749/9331 on FlowMatchEulerDiscreteScheduler timestep/indexing precision. The rejections are generally cautious and fit the evidence available. The summary also avoids overclaiming beyond the packet." + "feedback": "The summary is grounded in the packet: it correctly characterizes the set as mostly heterogeneous, notes there are no PRs, and only elevates a small number of duplicate-looking issue pairs. The accepted soft-edge verdicts are conservative and align with the issue titles, especially 1925\u21942440, 11749\u21949331, and 4972\u21945009. No clear evidence of overstatement beyond modest inference from titles." }, "evaluator_used": true, "retried": false, @@ -50533,8 +50547,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -50557,6 +50571,7 @@ "issue:3431", "issue:3451", "issue:3566", + "issue:3958", "issue:4063", "issue:4376", "issue:4611", @@ -50571,8 +50586,7 @@ "issue:5318", "issue:5320", "issue:5339", - "issue:5436", - "issue:5794", + "issue:5395", "issue:5824", "issue:5926", "issue:6681", @@ -50587,7 +50601,6 @@ "issue:993" ], "soft_pairs": [ - "issue:5436|issue:5794", "issue:3431|issue:3566", "issue:5318|issue:5320", "issue:4611|issue:4779", @@ -50611,192 +50624,193 @@ "issue:10460|issue:10472", "issue:12239|issue:9825", "issue:4376|issue:5824", - "issue:2618|issue:2822" + "issue:2618|issue:2822", + "issue:3958|issue:5395" ], - "prepared_review_unit_hash": "3e952720894fe1af74bda620d66e3abd7aa7b91454a105affafbd8643ed5a279", + "prepared_review_unit_hash": "8a47365430f4e77200019c22e313de53f69138a2b123deb6dff7d8c6f8653ba4", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 33220, - "estimated_input_tokens": 8305, - "estimated_eval_tokens": 16866 + "serialized_chars": 33214, + "estimated_input_tokens": 8304, + "estimated_eval_tokens": 16864 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most soft matches are broad topical overlaps rather than true duplicates. I only accept the clearly same-bug pairs around fp16 dtype mismatch, export_to_video failures, and SDXL ControlNet LoRA loading.", - "confidence": 0.71, - "canonical_issue_reason": "issue:993 is the best representative of the clear fp16/autocast duplicate pair with issue:775: it states the concrete data-type mismatch more directly and generally.", + "summary": "The set is mostly unrelated issues grouped by loose thematic similarity. I only consider two soft pairs as true duplicates: the fp16 dtype mismatch reports and the StableDiffusionXLControlNetImg2Img LoRA-loading reports.", + "confidence": 0.84, + "canonical_issue_reason": "No single issue cleanly represents the whole heterogeneous set; if one canonical issue must be chosen, #4885 is the clearest concrete bug report and anchors the strongest duplicate pair (#5109) around the same pipeline/LoRA-loading failure.", "canonical_pr_reason": null, - "best_issue_reason": "issue:993 is the most actionable issue to keep as the representative for the fp16 dtype-mismatch bug; the other candidate is just a less specific variant of the same report.", + "best_issue_reason": "#4885 is the best issue candidate because it is specific, reproducible, and narrowly scoped to a concrete pipeline bug, making it a strong triage anchor for the matching report in #5109.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:5436", - "right": "issue:5794", - "accept": false, - "reason": "Both are SDXL-related setup problems, but one is an example error and the other is a missing tokenizer/config file; different failure modes." - }, { "left": "issue:3431", "right": "issue:3566", "accept": false, - "reason": "Both mention ControlNet quality, but one is about inpainting+ControlNet results and the other about a reference community pipeline; not the same bug." + "reason": "Both are ControlNet-related, but they describe different problems and different pipelines; too broad to be the same bug." }, { "left": "issue:5318", "right": "issue:5320", "accept": false, - "reason": "PEFT-related, but one is tutorials/docs and the other is a specific API enhancement for `fuse_lora`; not duplicates." + "reason": "One is docs/tutorials, the other is a PEFT API change; related topic, but not the same change request." }, { "left": "issue:4611", "right": "issue:4779", "accept": false, - "reason": "Both are video/modelscope themed, but one is an integration/announcement and the other is a pipeline feature request; not the same change or bug." + "reason": "Both mention ModelScope/video models, but they target different features and are not clearly the same underlying request." }, { "left": "issue:6681", "right": "issue:7178", - "accept": true, - "reason": "Both are about `export_to_video` failing; the first title is just vague while the second describes the same broken behavior more explicitly." + "accept": false, + "reason": "Same function name, but the reports are too generic to confirm the same concrete export bug." }, { "left": "issue:12221", "right": "issue:13351", "accept": false, - "reason": "Both are ControlNet support requests, but for different models/features (DiffSynth vs Wan 2.2/Flux.2); not the same item." + "reason": "Different model families and different ControlNet support requests; not the same bug or change." }, { "left": "issue:2163", "right": "issue:2466", "accept": false, - "reason": "Training dtype mismatch versus checkpoint retention limit; unrelated issues." + "reason": "Training dtype mismatch vs checkpoint retention behavior; unrelated issues." }, { "left": "issue:6688", "right": "issue:9303", "accept": false, - "reason": "Different feature requests for different video models/pipelines; no shared concrete bug or change." + "reason": "Different feature additions in the video space; no shared code-path bug is evident." }, { "left": "issue:5075", "right": "issue:5258", "accept": false, - "reason": "Both involve SDXL/refiner behavior, but one is a prompt-embedding shape mismatch and the other is a ControlNet inpaint/refiner setup; different problems." + "reason": "Both involve SDXL/refiner, but one is a prompt-embedding shape error and the other is an inpaint/refiner pipeline issue." }, { "left": "issue:2213", "right": "issue:3451", "accept": false, - "reason": "TPU distributed hang versus fp16 memory regression; not the same bug." + "reason": "TPU distributed setup hang vs fp16 memory regression; unrelated failures." }, { "left": "issue:10817", "right": "issue:8507", "accept": false, - "reason": "Missing SD3 control nets versus a DreamBooth text-encoder training failure; different code paths." + "reason": "Both are SD3-related, but one is missing auto pipeline support and the other is a DreamBooth training failure." }, { "left": "issue:2985", "right": "issue:5926", "accept": false, - "reason": "Different model generations (Kandinsky 2.1 vs 3.0), so these are separate feature requests." + "reason": "Different Kandinsky versions and different feature scopes; not duplicates." }, { "left": "issue:10950", "right": "issue:9045", "accept": false, - "reason": "Docs/inference section breakage versus CUDA conversion failure with accelerate+peft; unrelated." + "reason": "Documentation breakage vs runtime device-conversion failure; not the same underlying issue." }, { "left": "issue:5054", "right": "issue:5824", "accept": false, - "reason": "One is a missing pipeline attribute, the other is a bad default in `get_add_time_ids()`; different defects." + "reason": "Both are SDXL, but they concern different bugs in different methods." }, { "left": "issue:4885", "right": "issue:5109", "accept": true, - "reason": "Both report `StableDiffusionXLControlNetImg2ImgPipeline` failing around `load_lora_weights`; same concrete code-path and likely same fix." + "reason": "Same pipeline and same LoRA-loading path; one report says weights do not load, the other says the load call raises, which plausibly describes the same bug." }, { "left": "issue:4063", "right": "issue:5205", "accept": false, - "reason": "Textual inversion unloading versus a broader repaint/controlnet/lora usage scenario; not the same underlying bug." + "reason": "Textual inversion unload vs inpainting/controlnet usage; different behaviors and code paths." }, { "left": "issue:4888", "right": "issue:5339", "accept": false, - "reason": "Offline safetensors loading and inability to load a local safetensor are related only at a very high level; not enough to treat as duplicates." + "reason": "Both concern local/offline safetensor loading, but the failure modes are too broad and not clearly the same bug." }, { "left": "issue:775", "right": "issue:993", "accept": true, - "reason": "Both describe the same fp16/data-type mismatch in Stable Diffusion, with and without autocast; clear duplicate bug reports." + "reason": "Both describe the same fp16/dtype mismatch class of Stable Diffusion failure, with nearly identical symptom framing." }, { "left": "issue:321", "right": "issue:938", "accept": false, - "reason": "Completely different research/model requests (Blended Diffusion vs Human Motion Diffusion Model)." + "reason": "Completely different model domains: diffusion images vs text-to-motion." }, { "left": "issue:11315", "right": "issue:11653", "accept": false, - "reason": "MPS float64 rope failure versus LoRA `.alpha` key support for HiDream; different bugs in the same model family." + "reason": "HiDream RoPE/MPS numeric issue vs LoRA key compatibility; different underlying problems." }, { "left": "issue:12221", "right": "issue:9825", "accept": false, - "reason": "Different support requests for different pipeline features (DiffSynth ControlNet vs FLUX IPAdapters)." + "reason": "Both are feature requests around adapters/controlnets, but they target different model families and capabilities." }, { "left": "issue:10460", "right": "issue:10472", "accept": false, - "reason": "One is a missing documented constructor arg, the other is an unexpected LoRA keyword error; not the same issue." + "reason": "A missing model init argument vs an unexpected LoRA keyword argument; unrelated bugs." }, { "left": "issue:12239", "right": "issue:9825", "accept": false, - "reason": "Support request for InfiniteTalk versus FLUX IPAdapters; separate features." + "reason": "InfiniteTalk support vs FLUX IPAdapter support; different features." }, { "left": "issue:4376", "right": "issue:5824", "accept": false, - "reason": "Textual inversion crash in SDXL versus a default-value bug in `get_add_time_ids()`; different root causes." + "reason": "Both are SDXL issues, but one is textual inversion loading and the other is a default-dimension bug." }, { "left": "issue:2618", "right": "issue:2822", "accept": false, - "reason": "ControlNet image-input support versus training convergence speed; not duplicates." + "reason": "Input type handling for ControlNet vs training convergence performance; not the same issue." + }, + { + "left": "issue:3958", + "right": "issue:5395", + "accept": false, + "reason": "Both mention LoRA/Civitai/PEFT, but one is cpu-offload interaction and the other is combining LoRAs; different problems." } ] }, "evaluator_result": { "accept": true, - "feedback": "The output is grounded in the packet and the soft-edge verdicts are mostly conservative. The accepted pairs are the clearest same-bug matches in the set: fp16 dtype mismatch (775/993), export_to_video failures (6681/7178), and SDXL ControlNet LoRA loading (4885/5109). The rejected pairs are generally justified by distinct titles or different code paths, and the summary does not overstate the evidence." + "feedback": "The summary is grounded in the packet and stays mostly conservative: it avoids collapsing the heterogeneous set into one theme, and the two accepted soft pairs are reasonably supported by closely matching titles/symptoms. No soft-edge verdict looks overstated relative to the available evidence." }, "evaluator_used": true, "retried": false, @@ -50812,8 +50826,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -50840,16 +50854,16 @@ "issue:2904", "issue:3565", "issue:3788", - "issue:3958", "issue:460", "issue:4789", + "issue:4912", "issue:5008", "issue:5317", "issue:5318", - "issue:5395", "issue:5480", "issue:579", "issue:5835", + "issue:6161", "issue:6234", "issue:6243", "issue:7012", @@ -50867,7 +50881,6 @@ "issue:989" ], "soft_pairs": [ - "issue:3958|issue:5395", "issue:1261|issue:2444", "issue:5317|issue:5318", "issue:1755|issue:5008", @@ -50891,193 +50904,197 @@ "issue:7191|issue:7490", "issue:3565|issue:7012", "issue:1961|issue:579", - "issue:2775|issue:989" + "issue:2775|issue:989", + "issue:4912|issue:6161" ], - "prepared_review_unit_hash": "f12b444320355bcb5e81fa0d128f124793a185202a474fd8c691ecf3d1df03c7", + "prepared_review_unit_hash": "71afc8ffa10a11591e795aaa98ab6d70558a2dbb0dc22e60cf9e046313620518", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33814, - "estimated_input_tokens": 8454, - "estimated_eval_tokens": 17164 + "serialized_chars": 33821, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated diffusers issues (feature requests, bug reports, docs/tests, and model-specific problems). None of the soft pairs look like true duplicates or the same concrete change.", - "confidence": 0.96, - "canonical_issue_reason": "No single issue is a reliable canonical here: the cluster is too heterogeneous to represent one underlying bug or feature.", - "canonical_pr_reason": "No PRs are present in this cluster.", - "best_issue_reason": "No clear best issue overall. If forced, 10658 is one of the cleaner self-contained feature requests, but the set is too mixed to privilege any one issue as a canonical target.", - "best_pr_reason": "No PRs are present, so there is no best PR candidate.", + "summary": "Most candidate pairs are only superficially similar and do not describe the same bug or change. The only likely duplicate is the missing-`datasets` dependency report pair (1463/1919).", + "confidence": 0.72, + "canonical_issue_reason": "Issue 1919 is the better canonical for the `datasets` dependency problem because it identifies the root cause in `requirements.txt`, while 1463 mainly reports the resulting import error.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1919 is the clearest actionable issue in this set: it states the missing dependency/fix path directly and serves well as the duplicate target for the related import-error report.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:3958", - "right": "issue:5395", - "accept": false, - "reason": "Both mention LoRA/PEFT, but one is about CPU offload and the other about combining civitai LoRAs; different underlying problems." - }, { "left": "issue:1261", "right": "issue:2444", "accept": false, - "reason": "Xformers CUDA backward assert vs a dreambooth script subprocess failure; same broad area, not the same bug." + "reason": "Different failures: xformers backward kernel assert vs a DreamBooth script exit-status error." }, { "left": "issue:5317", "right": "issue:5318", "accept": false, - "reason": "Documentation tasks in different scopes: PEFT API docs vs tutorials. Not a duplicate bug/change." + "reason": "Same broad PEFT area, but one is API docs and the other is tutorials; not the same bug/change." }, { "left": "issue:1755", "right": "issue:5008", "accept": false, - "reason": "No meaningful overlap: one is a generic/non-English issue title, the other is a TryOnDiffusion implementation request." + "reason": "Unrelated topics; no shared underlying bug or implementation." }, { "left": "issue:12338", "right": "issue:13477", "accept": false, - "reason": "AutoencoderDC VAE slicing/decoding bug and an ErnieImage locals() callback scope bug are unrelated." + "reason": "VAE slicing/decoding bug vs a callback `locals()` scope bug; different code paths." }, { "left": "issue:7390", "right": "issue:8359", "accept": false, - "reason": "Both involve from_single_file, but one is SDXL ControlNet inpaint and the other custom checkpoints; likely different failure modes." + "reason": "Both mention `from_single_file`, but they target different pipeline/custom-checkpoint scenarios and failures." }, { "left": "issue:11561", "right": "issue:9825", "accept": false, - "reason": "Both are IP-Adapter/FLUX-related, but one targets FluxFillPipeline specifically and the other FLUX pipelines generally; not the same concrete change." + "reason": "Both involve IP-Adapter/FLUX support, but one is FluxFill-specific and the other is a broader FLUX-pipeline feature request." }, { "left": "issue:1594", "right": "issue:3788", "accept": false, - "reason": "Img2Img for video generation and UniControl pipeline implementation are different feature requests." + "reason": "One asks for img2img video generation; the other requests a new pipeline implementation." }, { "left": "issue:1255", "right": "issue:1375", "accept": false, - "reason": "Missing DPMSolverMultistepScheduler import vs a broader import break caused by PR #1365; not the same issue." + "reason": "Both are import-related, but one is a missing scheduler export and the other is a broader import regression from a PR." }, { "left": "issue:1363", "right": "issue:1407", "accept": false, - "reason": "Different ONNX pipeline failures: LMS discrete scheduler long-prompt error vs missing feature_extractor ValueError." + "reason": "Different ONNX problems: long-prompt/LMS scheduler error vs missing `feature_extractor` in an SD2.0 ONNX pipeline." }, { "left": "issue:4789", "right": "issue:7793", "accept": false, - "reason": "DeepFloyd download failure and LoRA unload error are unrelated." + "reason": "Download failure for DeepFloyd IF vs LoRA unload error; unrelated underlying bugs." }, { "left": "issue:2904", "right": "issue:5835", "accept": false, - "reason": "AudioLDM scheduler NoneType error vs using pretrained MusicLDM pipeline; different pipelines and symptoms." + "reason": "AudioLDM scheduler attribute error vs general MusicLDM pipeline usage; not the same concrete issue." }, { "left": "issue:1463", "right": "issue:1919", - "accept": false, - "reason": "Both mention datasets, but one is a missing module error and the other is a requirements.txt omission in the dreambooth example." + "accept": true, + "reason": "Both point to the same missing `datasets` dependency; 1919 names the root cause in `requirements.txt` and 1463 reports the import symptom." }, { "left": "issue:10658", "right": "issue:9610", "accept": false, - "reason": "Add provider_options to InferenceSession vs add load_lora_weights to SD3 img2img; unrelated feature requests." + "reason": "ONNXRuntime provider options vs adding LoRA loading to a Stable Diffusion 3 img2img pipeline; different changes." }, { "left": "issue:1083", "right": "issue:460", "accept": false, - "reason": "FP16 incorrect ONNX output and CUDA-vs-CPU onnxruntime dependency are different ONNX bugs." + "reason": "Both are ONNX-related, but one is FP16 output correctness and the other is runtime/provider installation requirements." }, { "left": "issue:10873", "right": "issue:9450", "accept": false, - "reason": "Different pipelines and failures: SanaPipeline issue vs FluxPipeline multi-GPU device mismatch." + "reason": "Different pipeline bugs: SanaPipeline issue vs FluxPipeline multi-GPU device placement error." }, { "left": "issue:1117", "right": "issue:2377", "accept": false, - "reason": "Both are test requests, but they cover different utilities and behaviors; not the same change." + "reason": "Test-request for multiple negative prompts vs tests for EMAModel; unrelated scope." }, { "left": "issue:5008", "right": "issue:5480", "accept": false, - "reason": "TryOnDiffusion two-UNet implementation and EDM implementation request are distinct model features." + "reason": "One is a specific TryOnDiffusion implementation issue; the other is a request for EDM support." }, { "left": "issue:5480", "right": "issue:7380", "accept": false, - "reason": "Both are diffusion-method requests, but EDM and one-step distillation are different algorithms and would not merge into one PR." + "reason": "Both are model/pipeline feature requests, but they are for different diffusion methods and not the same change." }, { "left": "issue:6234", "right": "issue:8803", "accept": false, - "reason": "Train script issue for SDXL and an attention_processor duplicate assignment bug are unrelated." + "reason": "Different problems: SDXL text-to-image example issue vs a duplicate assignment in attention processor code." }, { "left": "issue:6243", "right": "issue:7243", "accept": false, - "reason": "IP-Adapter Face ID and ResAdapter are different adapter features." + "reason": "IP-Adapter Face ID support request vs ResAdapter implementation request; different features." }, { "left": "issue:7191", "right": "issue:7490", "accept": false, - "reason": "A vague import problem and a backward-pass problem are too unspecific and not the same underlying bug." + "reason": "Generic import problem vs a backward-pass problem; too little overlap to be the same bug." }, { "left": "issue:3565", "right": "issue:7012", "accept": false, - "reason": "CUDA OOM during training vs v_prediction causing weird images; different symptoms and code paths." + "reason": "CUDA OOM during text-to-image training vs weird images with `v_prediction`; different failure modes." }, { "left": "issue:1961", "right": "issue:579", "accept": false, - "reason": "ColossalAI error and missing config.json for textual inversion are unrelated." + "reason": "Different missing-package/import issues and different affected workflows; not the same report." }, { "left": "issue:2775", "right": "issue:989", "accept": false, - "reason": "torch._dynamo.optimize non-function error and weight-dimension assertion failure are different runtime bugs." + "reason": "`torch._dynamo.optimize` non-function error is unrelated to the `weight` must be 2-D runtime error." + }, + { + "left": "issue:4912", + "right": "issue:6161", + "accept": false, + "reason": "Custom SDXL checkpoint conversion request vs sigma-discard behavior; unrelated changes." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only positive duplicate call, 1463/1919, is well supported by the titles and issue descriptions (missing `datasets` package vs missing dependency in `requirements.txt`). The other pair verdicts are appropriately cautious given the limited evidence in the packet. " + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -51089,8 +51106,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -51116,19 +51133,19 @@ "issue:3623", "issue:3728", "issue:3817", + "issue:4232", "issue:4250", "issue:4524", "issue:4702", "issue:4837", - "issue:4912", "issue:4969", "issue:5034", + "issue:5168", "issue:541", "issue:5757", "issue:5998", "issue:6016", "issue:6136", - "issue:6161", "issue:7146", "issue:7322", "issue:7350", @@ -51144,7 +51161,6 @@ "issue:9825" ], "soft_pairs": [ - "issue:4912|issue:6161", "issue:2722|issue:3623", "issue:12412|issue:9825", "issue:10897|issue:9825", @@ -51168,192 +51184,196 @@ "issue:3728|issue:3817", "issue:4969|issue:9248", "issue:10100|issue:12239", - "issue:2673|issue:2799" + "issue:2673|issue:2799", + "issue:4232|issue:5168" ], - "prepared_review_unit_hash": "92926e0fabcf799e530bc2d64fe3daed15d4b47b27d0a08ff80ce0a4782bf159", + "prepared_review_unit_hash": "2d489d75bf7a2c841a4cb79fb786fce21dfb645a6eef81ee9bab2511cb2e6557", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33644, - "estimated_input_tokens": 8411, - "estimated_eval_tokens": 17078 + "serialized_chars": 33610, + "estimated_input_tokens": 8403, + "estimated_eval_tokens": 17062 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated feature requests and bug reports that only share broad Diffusers terminology. I found no soft pair that clearly looks like the same underlying bug/change, so all suggested merges are rejected.", - "confidence": 0.54, - "canonical_issue_reason": "Issue 8488 is the strongest single anchor: a concrete, high-engagement StableDiffusion3Pipeline initialization failure with substantial discussion. Even so, the cluster is too heterogeneous for it to serve as a true duplicate hub.", + "summary": "The cluster is mostly a collection of unrelated support requests and bug reports. The only close pair is the bf16/half type-mismatch reports, but the available evidence is still only soft similarity rather than a confirmed duplicate/root-cause match.", + "confidence": 0.64, + "canonical_issue_reason": "Issue 10891 is the broadest bfloat16-conversion bug report in the set, so it is the best tentative canonical issue, but the packet does not prove that 11117 is the same root cause.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 8488 is the best representative issue in this set because it is detailed, specific, and has the most discussion around a core runtime failure. It is still only a weak global representative because most other items are unrelated.", + "best_issue_reason": "Issue 10891 is the strongest issue candidate for grouping the bf16/half mismatch symptom, though only as a tentative umbrella for a related report rather than a confirmed duplicate.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:4912", - "right": "issue:6161", - "accept": false, - "reason": "Checkpoint conversion support and sigma-selection behavior are different problems in different areas." - }, { "left": "issue:2722", "right": "issue:3623", "accept": false, - "reason": "ControlNet conversion mismatch and large-model loading are unrelated bugs." + "reason": "Both are model-loading/conversion complaints, but they describe different failures and no shared concrete bug is evident." }, { "left": "issue:12412", "right": "issue:9825", "accept": false, - "reason": "Separate model support requests for different pipelines and features." + "reason": "Separate feature requests for different model/pipeline support; same broad area, not the same change." }, { "left": "issue:10897", "right": "issue:9825", "accept": false, - "reason": "Different model support requests; no shared concrete code-path." + "reason": "Different requested model support items with no shared code-path or bug." }, { "left": "issue:4837", "right": "issue:6016", "accept": false, - "reason": "Both are conversion-script issues, but they describe different failures and likely different fixes." + "reason": "Same conversion script, but the reported failures differ and do not clearly point to one underlying defect." }, { "left": "issue:10883", "right": "issue:11519", "accept": false, - "reason": "Different model update/support requests, not the same change." + "reason": "Different model support/update requests; no common bug or merged change." }, { "left": "issue:1594", "right": "issue:2195", "accept": false, - "reason": "Both touch video generation, but one asks for img2img support and the other for Tune-A-Video." + "reason": "Both concern video generation, but one is img2img video support and the other is Tune-A-Video; not the same artifact." }, { "left": "issue:7146", "right": "issue:9732", "accept": false, - "reason": "Both are OOM reports, but for different training scripts and workflows." + "reason": "Both mention OOM during training, but they are different training scripts/models and likely different memory issues." }, { "left": "issue:2676", "right": "issue:4524", "accept": false, - "reason": "Scheduler naming question versus AnimateDiff support request; unrelated." + "reason": "A scheduler compatibility question vs. an AnimateDiff integration request; unrelated changes." }, { "left": "issue:8994", "right": "issue:9680", "accept": false, - "reason": "Generation stuck at 0 and broken unconditional generation are different symptoms and likely different causes." + "reason": "Both describe generation problems, but the symptoms and likely code paths are too different to merge as one bug." }, { "left": "issue:10891", "right": "issue:11117", "accept": false, - "reason": "Both mention bfloat16/half issues, but they affect different pipelines/components and are not clearly the same bug." + "reason": "The titles suggest a related bf16/half mismatch, but the packet does not provide body evidence tying them to one confirmed duplicate or shared root cause; treat this as soft similarity only." }, { "left": "issue:4702", "right": "issue:9101", "accept": false, - "reason": "Separate training-script requests for different use cases." + "reason": "Both ask for training scripts, but they target different models and training setups." }, { "left": "issue:6136", "right": "issue:7417", "accept": false, - "reason": "Both are Mac NaN reports, but one is inference output and the other is a LoRA fuse test failure." + "reason": "Both involve NaN-related behavior, but one is Mac CPU inference and the other is a LoRA fuse test; not the same bug." }, { "left": "issue:5757", "right": "issue:7322", "accept": false, - "reason": "Checkpoint resume errors and SIGKILL during training are different failure modes." + "reason": "Both are training errors, but they concern different scripts, versions, and failure modes." }, { "left": "issue:11117", "right": "issue:8488", "accept": false, - "reason": "Both are SD3 pipeline errors, but one is a type-mismatch runtime error and the other is a from_pretrained exception." + "reason": "These are different StableDiffusion3Pipeline failures; one is a bf16 mismatch, the other is a from_pretrained exception." }, { "left": "issue:8604", "right": "issue:8837", "accept": false, - "reason": "T5 float16 corruption and SlicedAttnProcessor bug are unrelated." + "reason": "Different bugs in different components: corrupted output from T5 fp16 conversion vs. a sliced attention processor issue." }, { "left": "issue:1156", "right": "issue:222", "accept": false, - "reason": "Both are vague training failures, but there is no concrete shared bug or path." + "reason": "Generic training failures with insufficient evidence of a shared underlying defect." }, { "left": "issue:3537", "right": "issue:541", "accept": false, - "reason": "ControlNet conversion script and SageMaker support are unrelated." + "reason": "A ControlNet conversion script request and a SageMaker support issue are unrelated." }, { "left": "issue:3221", "right": "issue:7350", "accept": false, - "reason": "Different APIs and failure modes: attention-proc loading versus ControlNet initialization from LoRA UNet." + "reason": "Different ControlNet/LoRA loading problems with distinct failure contexts." }, { "left": "issue:3178", "right": "issue:5998", "accept": false, - "reason": "Both mention VersatileDiffusion, but the reported bugs are distinct and not mergeable as one fix." + "reason": "Same pipeline name, but one is a generic bug report and the other is a specific TyperError; not enough to treat as one change." }, { "left": "issue:4250", "right": "issue:5034", "accept": false, - "reason": "Model loading bug and progress-bar display issue are not the same underlying problem." + "reason": "Different symptoms: incorrect from_pretrained loading vs. a progress-bar display issue." }, { "left": "issue:3728", "right": "issue:3817", "accept": false, - "reason": "Different proposed pipelines with no shared bug or implementation change." + "reason": "Both are new pipeline requests, but they concern different projects and capabilities." }, { "left": "issue:4969", "right": "issue:9248", "accept": false, - "reason": "ImportError for CrossAttention and install/upgrade guidance are unrelated." + "reason": "An import error report and a pip install command request are not the same bug." }, { "left": "issue:10100", "right": "issue:12239", "accept": false, - "reason": "Different model support requests with no shared concrete code-path." + "reason": "Separate support requests for different models; only superficial similarity." }, { "left": "issue:2673", "right": "issue:2799", "accept": false, - "reason": "Both involve pipeline/checkpoint mismatch themes, but the specific errors and fixes are different." + "reason": "One is an upscale pipeline usage error; the other is a request to raise a clearer mismatch error, not the same underlying bug." + }, + { + "left": "issue:4232", + "right": "issue:5168", + "accept": false, + "reason": "Both involve ControlNet-related work, but one is instruct-pix2pix training/inference and the other is ControlNet-XS support." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "The writeup is mostly conservative on the unrelated pairs, but it overreaches on issue:10891 vs issue:11117. The packet does not provide body evidence linking them, and the soft-pair metadata marks that pair as deterministic_accept=false. Based on the available information, this should be treated as at most a plausible soft similarity, not a confirmed duplicate/root-cause match." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -51366,8 +51386,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -51404,7 +51424,6 @@ "issue:4849", "issue:4912", "issue:4978", - "issue:5168", "issue:541", "issue:6008", "issue:7569", @@ -51421,7 +51440,6 @@ "issue:9936" ], "soft_pairs": [ - "issue:4232|issue:5168", "issue:6008|issue:9223", "issue:1633|issue:2737", "issue:761|issue:911", @@ -51447,194 +51465,185 @@ "issue:10737|issue:9450", "issue:4390|issue:4598" ], - "prepared_review_unit_hash": "4bf73f63b140eebd89cf0bb421479a51d6995a2b952f5ff65f24d0bfa465b8f7", + "prepared_review_unit_hash": "31b130d70af874938b81ba1a3deedb1e5309014c667679f62aa568353af0ec65", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 33873, - "estimated_input_tokens": 8469, - "estimated_eval_tokens": 17194 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33086, + "estimated_input_tokens": 8272, + "estimated_eval_tokens": 16800 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is not one duplicate cluster; it contains many unrelated issue reports across different models, pipelines, devices, and feature requests. Most soft edges are false positives from title similarity, with at most one borderline MPS-output pair that looks plausibly duplicate-like.", - "confidence": 0.92, - "canonical_issue_reason": null, + "summary": "The cluster is only loosely related by diffusion pipelines and transformer/loader issues, but the paired items are not the same underlying bug or change. Most are different pipelines, different hardware backends, or distinct feature requests, so none of the soft edges should be merged.", + "confidence": 0.31, + "canonical_issue_reason": "Issue #12754 is the clearest self-contained API bug in the set and is broadly representative of the pipeline-loading/casting theme, but the cluster overall is heterogeneous rather than a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good global canonical here; the items are too heterogeneous to represent one underlying bug. If forced to choose a representative issue, none fits cleanly.", + "best_issue_reason": "Issue #12754 is the strongest representative issue because it describes a specific, actionable regression in core pipeline behavior with clear expected-vs-actual semantics.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:4232", - "right": "issue:5168", - "accept": false, - "reason": "Different requests: instruct-pix2pix/controlnet training vs ControlNet-XS support. Same broad area, not the same bug." - }, { "left": "issue:6008", "right": "issue:9223", "accept": false, - "reason": "One is an IP-Adapter/LCM LoRA runtime failure; the other is a dependency/version conflict around pyarrow. Unrelated." + "reason": "Different problems: IP-Adapter/LCM face integration failure vs dependency/version conflict around pyarrow; unrelated symptoms and code paths." }, { "left": "issue:1633", "right": "issue:2737", "accept": false, - "reason": "Both mention scheduler features, but one is karras sigmas in k_diffusion and the other is CLIP guidance in DPMSolver. Different changes." + "reason": "Both mention scheduler/features, but one is karras sigmas in a wrapper pipeline and the other is CLIP guidance support; different changes and APIs." }, { "left": "issue:761", "right": "issue:911", "accept": false, - "reason": "DreamBooth issues, but one is 8-bit Adam training failure and the other is a bad config regression. Different root problems." + "reason": "Both are DreamBooth-related, but one is an 8-bit Adam training failure and the other is a bad config regression; not the same bug." }, { "left": "issue:11374", "right": "issue:12695", "accept": false, - "reason": "Both are model support requests, but for different models (SkyReels-V2 vs Hunyuan v1.5). Not duplicates." + "reason": "SkyReels-V2 support and Hunyuan v1.5 support are separate model-integration feature requests, not the same change." }, { "left": "issue:3171", "right": "issue:4496", "accept": false, - "reason": "LPW prompt support for controlnet vs prompt fusion are different features and failure modes." + "reason": "LPW prompt support for ControlNet and prompt fusion are different prompt-handling features with no shared concrete bug." }, { "left": "issue:4450", "right": "issue:4849", "accept": false, - "reason": "Multi-text inversion example broken vs multi-GPU training for textual inversion. Different scope and cause." + "reason": "Multi textual inversion example failure and multi-GPU textual inversion training are unrelated issues despite both involving textual inversion." }, { "left": "issue:4232", "right": "issue:4978", "accept": false, - "reason": "Different control/preprocessor features: instruct-pix2pix with controlnet condition vs inpaint_global_harmonious preprocessor usage." + "reason": "Instruct-pix2pix with ControlNet vs inpaint_global_harmonious preprocessor usage are different pipeline/preprocessor topics." }, { "left": "issue:12599", "right": "issue:13279", "accept": false, - "reason": "Both are loading-related, but one is CUDA transfer slowness and the other is an AutoencoderRAE compatibility error with older transformers." + "reason": "Extremely slow CUDA move after loading vs AutoencoderRAE loading error with older transformers are different loading/runtime problems." }, { "left": "issue:4912", "right": "issue:541", "accept": false, - "reason": "Checkpoint conversion to diffusers vs SageMaker deployment support. No shared bug." + "reason": "Custom SDXL checkpoint conversion and SageMaker support are separate workflow/environment requests, not the same bug or change." }, { "left": "issue:10795", "right": "issue:12728", "accept": false, - "reason": "Both involve torch.compile, but one is GGUF from_single_file transformer compilation and the other is Wan 2.2 channels_last/Dynamo unsupported behavior. Different code paths." + "reason": "torch.compile on GGUF-loaded transformer models and Wan 2.2 compile/channels_last failures are distinct compiler/model-path issues." }, { "left": "issue:10737", "right": "issue:9063", "accept": false, - "reason": "Flux pipeline issues, but one is encode_image missing image_encoder and the other is tokenizer instantiation from a slow version." + "reason": "Missing image_encoder in FluxPipeline encode_image and tokenizer instantiation failure in FluxPipeline are different failures in different components." }, { "left": "issue:10150", "right": "issue:12754", "accept": false, - "reason": "SDXL loading/inference slowness vs from_pipe dtype conversion to float32. Related theme, different concrete bug." + "reason": "SDXL slow loading/inference is a performance complaint, while from_pipe float32 conversion is a dtype regression; not the same bug." }, { "left": "issue:1305", "right": "issue:4063", "accept": false, - "reason": "Completely different topics: image editing method paper vs unloading a textual inversion checkpoint." + "reason": "Direct Inversion and unloading a textual inversion checkpoint are different features in the inversion/textual-inversion area, not duplicates." }, { "left": "issue:1263", "right": "issue:1375", "accept": false, - "reason": "Both mention import problems, but one is missing StableDiffusionSafetyChecker and the other is a regression from PR #1365. Not the same issue." + "reason": "Importing StableDiffusionSafetyChecker and a PR breaking diffusers import are different import failures with different root causes." }, { "left": "issue:3725", "right": "issue:4302", "accept": false, - "reason": "Both are LoRA-related, but one is general A1111 LoRA support and the other is a specific SDXL LoRA load failure. Too different to merge." + "reason": "A1111 LoRA support failure and SDXL LoRA loading compatibility are both LoRA-related, but they are distinct compatibility problems." }, { "left": "issue:1188", "right": "issue:3043", "accept": false, - "reason": "Different community pipeline requests: CLIPSeg+Stable Diffusion vs ODISE." + "reason": "CLIPSeg x Stable Diffusion pipeline request and ODISE pipeline request are separate model integrations, not one change." }, { "left": "issue:4480", "right": "issue:7569", "accept": false, - "reason": "Both involve Mac/MPS, but one is a kernel crash during generation and the other is a FreeU test failure. Different bug class." + "reason": "Both are Mac/MPS-related, but one is a kernel crash during generation and the other is a failing test for FreeU; different concrete bugs." }, { "left": "issue:4213", "right": "issue:8435", "accept": false, - "reason": "Both mention from_single_file, but 4213 is specifically about needing accelerate for SD-XL; 8435 is a generic runtime error with no matching concrete cause." + "reason": "Both concern from_single_file, but one is specifically SD-XL without accelerate and the other is a generic runtime error; too broad to treat as the same bug." }, { "left": "issue:3325", "right": "issue:7646", "accept": false, - "reason": "Different training scripts and failures: DreamBooth CUDA invalid argument vs dataloader_num_workers behavior in text-to-image LoRA training." + "reason": "CUDA invalid-argument during DreamBooth training and dataloader_num_workers questions in a LoRA script are not the same underlying issue." }, { "left": "issue:457", "right": "issue:760", - "accept": true, - "reason": "Both describe Stable Diffusion output corruption on MPS, with very similar symptoms ('blue artifacts' / 'unexpected output') and the same backend/context. Plausibly the same underlying bug." + "accept": false, + "reason": "Both are MPS-related, but one reports blue artifacts and the other unexpected StableDiffusionPipeline output; different symptoms and likely causes." }, { "left": "issue:11642", "right": "issue:9861", "accept": false, - "reason": "Flux prompt scaling has no effect vs Flux training not updating transformer weights. Distinct bugs." + "reason": "prompt_embeds_scale having no effect in FluxPriorReduxPipeline and Flux training not updating the transformer are unrelated Flux issues." }, { "left": "issue:9161", "right": "issue:9936", "accept": false, - "reason": "Both are Flux multi-GPU training problems, but one is distributed bugs in train_dreambooth_lora_flux.py and the other is an NCCL timeout in train_controlnet_flux.py. Not the same failure." + "reason": "Distributed bugs in train_dreambooth_lora_flux.py and NCCL timeout in train_controlnet_flux.py are different multi-GPU training failures." }, { "left": "issue:10737", "right": "issue:9450", "accept": false, - "reason": "Both are Flux-related, but one is an encode_image initializer bug and the other is a multi-GPU device placement error when passing transformer=. Different code paths." + "reason": "Both are FluxPipeline issues, but one is a missing image encoder and the other is a multi-GPU device placement error; different code paths." }, { "left": "issue:4390", "right": "issue:4598", "accept": false, - "reason": "Missing requirements vs uninformative error messages for missing dependencies. The latter is about diagnostics, not the dependency issue itself." + "reason": "Missing requirements and uninformative missing-dependency errors are both dependency-related, but one is a packaging issue and the other is an error-message improvement." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The clustering summary is grounded in the packet: the set is clearly heterogeneous, and the soft-edge rejections are conservative rather than overstated. The only accepted edge, issue:457 vs issue:760, is plausibly duplicate-like based on shared MPS output-corruption symptoms. No PR-mergeability concerns apply here." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -51646,8 +51655,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -51674,7 +51683,6 @@ "issue:322", "issue:3348", "issue:3406", - "issue:3451", "issue:3630", "issue:4186", "issue:4232", @@ -51691,13 +51699,13 @@ "issue:8507", "issue:8511", "issue:8780", - "issue:8939", "issue:9025", "issue:9039", - "issue:9063", "issue:9142", "issue:938", "issue:9539", + "issue:9576", + "issue:9660", "issue:9895" ], "soft_pairs": [ @@ -51714,6 +51722,7 @@ "issue:9025|issue:9039", "issue:2579|issue:3348", "issue:8507|issue:8511", + "issue:9576|issue:9660", "issue:3171|issue:6141", "issue:4445|issue:8780", "issue:11133|issue:13034", @@ -51723,188 +51732,180 @@ "issue:5002|issue:6901", "issue:1367|issue:2843", "issue:200|issue:3406", - "issue:322|issue:938", - "issue:8939|issue:9063", - "issue:3451|issue:7646" + "issue:322|issue:938" ], - "prepared_review_unit_hash": "59c655620b7cd9d9ad83b8e06f1166dedf336e7c57adc9b32860eb5e7bacce79", + "prepared_review_unit_hash": "438da3e1339759bac4cb6cbdb67149cdb094ba0ae85cb06d299780f120d4d038", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 34309, - "estimated_input_tokens": 8578, - "estimated_eval_tokens": 17412 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33164, + "estimated_input_tokens": 8291, + "estimated_eval_tokens": 16838 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set. The items span many unrelated issues across different pipelines, schedulers, training scripts, loading paths, and model families; the proposed soft edges mostly share only generic error wording or a broad subsystem name.", - "confidence": 0.95, - "canonical_issue_reason": "No single canonical issue fits this cluster: the items are too heterogeneous to represent one underlying bug or change.", + "summary": "This cluster is not a true duplicate set: it mixes many unrelated issues across different models, pipelines, training scripts, and years. None of the soft pairs look like the same concrete bug/change, so all should stay separate.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue: the items span unrelated topics (scheduler APIs, LoRA loading, offloading, ControlNet, video pipelines, training bugs, and model conversion) with no shared underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue for the whole cluster; at most, a few offloading-related items form a loose subtheme, but they are still not the same concrete failure.", + "best_issue_reason": "No single issue is a good cluster representative because the cluster is heterogeneous and does not center on one actionable problem.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:9539", "right": "issue:9895", "accept": false, - "reason": "Both mention device-mismatch failures, but they are different pipelines and contexts (CogVideoX image-to-video float8 issue vs FluxPipeline offloading regression)." + "reason": "Both involve device/type failures in video pipelines, but one is Float8 index_select on CogVideoX and the other is a Flux offloading device-mismatch regression; different code paths and fixes." }, { "left": "issue:3630", "right": "issue:4747", "accept": false, - "reason": "Both are loading/runtime errors, but one is an unknown UNet2DModel error and the other is a missing model_index.json when loading fine-tuned SDXL LoRA." + "reason": "Both are runtime/load errors, but one is an unknown UNet2DModel issue and the other is missing model_index.json after loading an SDXL LoRA; not the same bug." }, { "left": "issue:10952", "right": "issue:9142", "accept": false, - "reason": "These are different adapter/loading bugs: missing LoRA adapter name in SDXLLongPromptWeightingPipeline vs multi-controlnet type error in Hunyuan DiT." - }, - { - "left": "issue:11050", - "right": "issue:11182", - "accept": false, - "reason": "One is a bf16/fp16 dtype-casting bug in a training script; the other is an unnecessary BF16 variant download during model loading." + "reason": "Both mention adapter/model loading failures, but one is SDXL LoRA adapter-name validation and the other is Hunyuan DiT multiple-ControlNet loading; different mechanisms." }, { "left": "issue:1082", "right": "issue:1608", "accept": false, - "reason": "Both are training failures, but the concrete problems differ: cublasLtMatmul execution failure vs training stopping at 500 steps on T4." + "reason": "Both are training-related failures, but one is a Dreambooth CUDA matmul crash and the other is training stopping at 500 steps; no shared root cause." }, { "left": "issue:2722", "right": "issue:743", "accept": false, - "reason": "Both involve conversion, but they are not the same bug: ControlNet UNet mismatch after conversion vs general ckpt-to-diffusers conversion failure." + "reason": "Both concern model conversion/loading, but one is ControlNet UNet mismatch after conversion and the other is generic ckpt-to-diffusers conversion failure; too broad to merge." }, { "left": "issue:4186", "right": "issue:5258", "accept": false, - "reason": "Both mention SDXL inpaint/refiner, but one is inpainting not working with refiner and the other is a ControlNet inpaint/refiner combination." + "reason": "Both mention SDXL inpaint/refiner, but one is inpainting with refiner and the other is ControlNet inpaint with refiner; different pipelines and failure modes." }, { "left": "issue:4232", "right": "issue:5118", "accept": false, - "reason": "These are different ControlNet-related questions/features (instruct-pix2pix with controlnet condition vs reference_only usage)." + "reason": "Both relate to controlnet/instruct-pix2pix style usage, but one is training/inference code and the other is a reference_only usage question; not the same bug." }, { "left": "issue:2195", "right": "issue:4949", "accept": false, - "reason": "Unrelated video-model topics: Tune-A-Video vs ProPainter in/outpainting for video." + "reason": "These are unrelated video topics: Tune-A-Video versus ProPainter in/outpainting; no common defect." }, { "left": "issue:2439", "right": "issue:3103", "accept": false, - "reason": "Both are inpainting-related, but one is loading/running the inpainting pipeline and the other is a height/width parameter issue." + "reason": "Both involve inpainting, but one is loading/running the inpainting pipeline and the other is the height/width parameter behavior; different issues." }, { - "left": "issue:8507", - "right": "issue:8511", + "left": "issue:11050", + "right": "issue:11182", "accept": false, - "reason": "Both touch SD3 training scripts, but the failures differ: text encoder training fails vs text encoders not being unloaded." + "reason": "One is a missing dtype cast in an SD3 training script; the other is an unnecessary BF16 variant download. Different code paths and fixes." }, { "left": "issue:9025", "right": "issue:9039", "accept": false, - "reason": "Different Flax pipeline operations: from_single_file on Kaggle TPU VM vs load_lora_weights." + "reason": "Both are Flax pipeline questions, but one is from_single_file on TPU and the other is load_lora_weights in FlaxStableDiffusionPipeline; separate features." }, { "left": "issue:2579", "right": "issue:3348", "accept": false, - "reason": "Both reference xFormers, but one is a possible xFormers bug/misuse and the other is a feature request for LoRA scale support with xFormers." + "reason": "Both mention xFormers and LoRA, but one is a possible xFormers bug/misuse and the other is missing scale parameter support for LoRA with xFormers; not the same defect." + }, + { + "left": "issue:8507", + "right": "issue:8511", + "accept": false, + "reason": "Closely related SD3 training topics, but one is train_dreambooth_sd3.py failing with --train_text_encoder and the other is text encoders not being unloaded; distinct fixes." + }, + { + "left": "issue:9576", + "right": "issue:9660", + "accept": false, + "reason": "Both are pipeline-loading regressions, but one is use_safetensors not propagated to submodules and the other is modules not switched to eval mode; different behaviors." }, { "left": "issue:3171", "right": "issue:6141", "accept": false, - "reason": "These are separate ControlNet feature requests: LPW prompt support vs control_mode/resize_mode parameters." + "reason": "Both concern ControlNet parameters, but one is LPW prompt handling and the other is control_mode/resize_mode support; unrelated feature gaps." }, { "left": "issue:4445", "right": "issue:8780", "accept": false, - "reason": "A how-to question about LoRA finetuning is not the same as a dtype mismatch bug between model and input data." + "reason": "Both are vague model/data-type usage problems, but one asks how to finetune LoRA and the other reports model/input dtype mismatch; not the same bug." }, { "left": "issue:11133", "right": "issue:13034", "accept": false, - "reason": "Different pipelines and failures: CogVideoX image-to-video bug vs WanPipeline height/width behavior." + "reason": "Both are Wan-related, but one is a cogvideox image-to-video bug and the other is WanPipeline height/width handling; different pipelines and symptoms." }, { "left": "issue:1076", "right": "issue:322", "accept": false, - "reason": "Completely unrelated: Stable Diffusion v1-5 discussion vs V-Diffusion model topic." + "reason": "Completely unrelated topics: stable-diffusion-v1-5 discussion versus V-Diffusion." }, { "left": "issue:12052", "right": "issue:13488", "accept": false, - "reason": "Both involve offloading, but the reports are not the same concrete bug: Wan 2.2 LightX2V device-mismatch failure vs a broader diffusers-0.38.0.dev0 offloading regression." + "reason": "Both mention offloading, but one is a concrete tensor-device mismatch in Wan 2.2 LightX2V offloading and the other is a broad diffusers 0.38.0.dev0 offloading regression; too broad to treat as one bug." }, { "left": "issue:3094", "right": "issue:7646", "accept": false, - "reason": "Different training concerns: OOM during finetuning vs dataloader_num_workers behavior in a LoRA training script." + "reason": "Both are training script questions, but one is CUDA OOM on a finetuning example and the other is dataloader_num_workers behavior; different issues." }, { "left": "issue:5002", "right": "issue:6901", "accept": false, - "reason": "Both are scheduler API gaps, but they affect different methods and classes: use_karras_sigmas vs clip_sample." + "reason": "Both are scheduler API gaps, but one is missing use_karras_sigmas and the other is missing clip_sample in FlaxDDIMScheduler; separate implementations." }, { "left": "issue:1367", "right": "issue:2843", "accept": false, - "reason": "Saving scheduler config with custom trained_betas is unrelated to multi-GPU checkpoint saving." + "reason": "Both involve saving configuration/checkpoints, but one is scheduler config with custom trained_betas and the other is multi-GPU save checkpoint; not the same bug." }, { "left": "issue:200", "right": "issue:3406", "accept": false, - "reason": "A resnet test request is unrelated to an attention-class bug with only_cross_attention=True." + "reason": "Not the same problem: one is a test request for resnet class, the other is Attention class behavior with only_cross_attention=True." }, { "left": "issue:322", "right": "issue:938", "accept": false, - "reason": "Different model-family topics with no shared concrete bug or fix." - }, - { - "left": "issue:8939", - "right": "issue:9063", - "accept": false, - "reason": "Both are tokenizer/loading errors, but they involve different models and failure modes, so they are not the same issue." - }, - { - "left": "issue:3451", - "right": "issue:7646", - "accept": false, - "reason": "Both are training-script questions, but fp16 memory usage and dataloader_num_workers are unrelated problems." + "reason": "Completely unrelated projects/models: Katherine's V-Diffusion and Human Motion Diffusion Model." } ] }, @@ -51923,8 +51924,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -51951,18 +51952,15 @@ "issue:1552", "issue:2392", "issue:2830", + "issue:3451", "issue:3624", "issue:3728", - "issue:3884", "issue:3944", "issue:4269", "issue:4686", - "issue:4839", "issue:5128", "issue:5205", - "issue:5573", "issue:5605", - "issue:5997", "issue:6357", "issue:6503", "issue:660", @@ -51970,14 +51968,19 @@ "issue:713", "issue:7243", "issue:7333", + "issue:7646", "issue:775", "issue:8500", "issue:8565", "issue:8588", "issue:8759", + "issue:8939", + "issue:9063", "issue:9166" ], "soft_pairs": [ + "issue:8939|issue:9063", + "issue:3451|issue:7646", "issue:3624|issue:4269", "issue:3944|issue:5605", "issue:10883|issue:7243", @@ -51999,188 +52002,186 @@ "issue:12009|issue:12037", "issue:2830|issue:5205", "issue:11117|issue:8565", - "issue:10336|issue:10873", - "issue:5573|issue:5997", - "issue:3884|issue:4839" + "issue:10336|issue:10873" ], - "prepared_review_unit_hash": "e2f02a6ef58f75fab5ea4be8bda3f514975d3841a8f9151cac6f8e12f33ce581", + "prepared_review_unit_hash": "335d6c27591f1b6471e461ab0ee7a519143b4e4fe66bb6d6b2aebb090ac94109", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34063, - "estimated_input_tokens": 8516, - "estimated_eval_tokens": 17288 + "serialized_chars": 34427, + "estimated_input_tokens": 8607, + "estimated_eval_tokens": 17470 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a false-positive similarity set: the items span unrelated models, pipelines, training scripts, and API bugs. I only accept the pairs that look like the same fp16/attention-slicing code-path issue.", - "confidence": 0.88, - "canonical_issue_reason": "No single issue is a safe canonical for the whole set: the items span many unrelated bugs, features, and model families.", + "summary": "This cluster is mostly a grab bag of unrelated Diffusers issues spanning different models, pipelines, training scripts, and feature requests. Only the attention-slicing reports look likely to be the same underlying bug; the rest are too model- or feature-specific to merge as duplicates.", + "confidence": 0.9, + "canonical_issue_reason": "Issue #1405 is the broadest and most representative report here: it describes the core attention-slicing breakage on main, which also subsumes the more specific SD2 report.", "canonical_pr_reason": null, - "best_issue_reason": "No issue cleanly represents the whole cluster for deduping; any single choice would collapse unrelated topics.", + "best_issue_reason": "Issue #1405 is the best single issue to keep as the cluster representative because it is the most general statement of the duplicate-worthy bug and points at the shared code path.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:8939", + "right": "issue:9063", + "accept": false, + "reason": "Both are tokenizer-loading failures, but for different pipelines/models and different error modes; not the same concrete bug." + }, + { + "left": "issue:3451", + "right": "issue:7646", + "accept": false, + "reason": "One is fp16 memory usage in training, the other is a dataloader worker question; unrelated." + }, { "left": "issue:3624", "right": "issue:4269", "accept": false, - "reason": "Both mention TypeError, but they concern different pipeline paths and different failure causes." + "reason": "Both mention TypeError, but they hit different validation paths and different APIs." }, { "left": "issue:3944", "right": "issue:5605", "accept": false, - "reason": "Same DreamBooth area, but one is about LoRA rank customization and the other about per-image captions; different changes." + "reason": "Both are Dreambooth-related feature requests, but one is LoRA rank customization and the other is per-image captions; distinct changes." }, { "left": "issue:10883", "right": "issue:7243", "accept": false, - "reason": "Both are model/pipeline issues, but they refer to different models and different problem types." + "reason": "Different pipelines/models and different problems; no shared underlying bug." }, { "left": "issue:10658", "right": "issue:11700", "accept": false, - "reason": "Both touch loading/integration, but one is about onnxruntime session options and the other about Pruna pipeline loading; not the same bug." + "reason": "One asks for provider_options in ONNX Runtime session creation; the other requests Pruna integration for from_pretrained/pipeline." }, { "left": "issue:10144", "right": "issue:12666", "accept": false, - "reason": "Both are video-related, but they involve different models and unrelated failure modes." + "reason": "Both are video-model issues, but they concern different model families and different problems." }, { "left": "issue:8500", "right": "issue:8588", "accept": false, - "reason": "Both are SD3-related, but one is a token-limit question and the other is a missing conversion script." + "reason": "Both are SD3-related, but one is about token limits and the other about a missing conversion script." }, { "left": "issue:1082", "right": "issue:713", "accept": false, - "reason": "Training errors are too generic here; the reported failures and contexts are not the same." - }, - { - "left": "issue:10460", - "right": "issue:10848", - "accept": false, - "reason": "Both are API mismatch problems, but they concern different functions and different call sites." + "reason": "Too generic to treat as the same bug; one is a specific CUDA matmul failure in Dreambooth and the other is a broad training error." }, { "left": "issue:1478", "right": "issue:4686", "accept": false, - "reason": "Same pipeline class, but one is about multi-image output and the other about loading a saved pipeline; different bugs." + "reason": "Different StableDiffusionUpscalePipeline failures: output count vs loading a saved pipeline." }, { "left": "issue:10653", "right": "issue:10710", "accept": false, - "reason": "Completely different subjects: quantized model movement vs DDUF support." + "reason": "Completely different topics: 8-bit module device transfer vs DDUF format support." }, { "left": "issue:3728", "right": "issue:7333", "accept": false, - "reason": "Both are diffusion research/pipeline topics, but they are different requested methods and not the same change." + "reason": "Both are video-generation related, but they are different pipelines/feature requests rather than the same bug." + }, + { + "left": "issue:10460", + "right": "issue:10848", + "accept": false, + "reason": "Missing UNet argument and an unexpected keyword in IP-Adapter loading are separate API issues." }, { "left": "issue:1405", "right": "issue:1552", "accept": true, - "reason": "These both point to the same attention-slicing bug: `enable_attention_slicing()` failing in practice, with SD2 as a concrete manifestation." + "reason": "These describe the same attention-slicing breakage; #1552 is the SD2-specific manifestation of the broader failure reported in #1405." }, { "left": "issue:660", "right": "issue:775", - "accept": true, - "reason": "Both describe the same fp16-on-MPS failure class; one is a specific M1 report, the other a broader fp16/autocast complaint." + "accept": false, + "reason": "Both involve fp16, but one is MPS-specific revision/dtype behavior and the other is a broader autocast-related problem." }, { "left": "issue:1099", "right": "issue:2392", "accept": false, - "reason": "Different DreamBooth/pipeline errors with no shared failing code path." + "reason": "Unrelated Dreambooth class-name issue vs img2img prompt_embeds validation error." }, { "left": "issue:8759", "right": "issue:9166", "accept": false, - "reason": "Both involve image processing, but one is an SD3 cast warning and the other is overlay behavior; not the same bug." + "reason": "Both touch image_processor, but they concern different methods and symptoms." }, { "left": "issue:10608", "right": "issue:11396", "accept": false, - "reason": "Unrelated: multi-GPU HunyuanVideo pipeline execution vs LoRA format conversion question." + "reason": "Different domains entirely: multi-GPU pipeline execution vs LoRA format conversion." }, { "left": "issue:6503", "right": "issue:6891", "accept": false, - "reason": "Both are training-script tracker items, but they target different utility updates and code changes." + "reason": "Torch.compile training compatibility and model-card saving are unrelated training-script topics." }, { "left": "issue:5128", "right": "issue:6357", "accept": false, - "reason": "Different problems: loading a local safetensors file vs API/local result discrepancies." + "reason": "Loading a local safetensors file is unrelated to API-vs-local result differences." }, { "left": "issue:12009", "right": "issue:12037", "accept": false, - "reason": "Same Wan 2.2 family, but one is a GGUF loading/meta-tensor issue and the other is LoRA compatibility; not mergeable as one fix." + "reason": "Both are Wan 2.2 issues, but one is single-file GGUF loading and the other is LoRA compatibility; different code paths." }, { "left": "issue:2830", "right": "issue:5205", "accept": false, - "reason": "Different use cases and pipelines; no shared underlying bug." + "reason": "No shared underlying bug; one asks about video generation, the other is a StableDiffusion repaint/ControlNet feature request." }, { "left": "issue:11117", "right": "issue:8565", "accept": false, - "reason": "Both are SD3 pipeline issues, but one is a dtype mismatch runtime error and the other is a LoRA loading IndexError." + "reason": "Same model family, but different failures: dtype mismatch at runtime vs an IndexError in LoRA loading." }, { "left": "issue:10336", "right": "issue:10873", "accept": false, - "reason": "Both are Sana-related, but the reported failures are different and do not clearly share one code path." - }, - { - "left": "issue:5573", - "right": "issue:5997", - "accept": false, - "reason": "ControlNet-related, but one is weight initialization timing and the other is an SDXL tutorial/model-type error." - }, - { - "left": "issue:3884", - "right": "issue:4839", - "accept": false, - "reason": "Both mention LoRA, but one is a training-script error and the other is a `lora_scale` type issue; different fixes." + "reason": "Both are Sana-related, but activation explosion/clamping and a specific SanaPipeline issue are not the same defect." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet and stays conservative: it treats the cluster as mostly unrelated and only accepts the two genuinely plausible same-bug pairs (attention slicing; fp16/MPS). The reject reasons are cautious and aligned with the titles, without overclaiming shared root causes." + "feedback": "The representative choice and cluster summary are grounded in the packet, and the soft-edge verdicts stay conservative overall. The only accepted duplicate pair (#1405/#1552) is well supported by the shared attention-slicing failure, while the rest are reasonably rejected as model-, pipeline-, or feature-specific rather than clearly the same bug." }, "evaluator_used": true, "retried": false, @@ -52196,8 +52197,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -52222,15 +52223,17 @@ "issue:2689", "issue:3162", "issue:3838", + "issue:3884", "issue:4000", - "issue:4159", "issue:4210", "issue:4349", "issue:4356", - "issue:4390", "issue:462", "issue:4692", "issue:4724", + "issue:4839", + "issue:5573", + "issue:5997", "issue:6439", "issue:6790", "issue:7000", @@ -52245,11 +52248,12 @@ "issue:9343", "issue:9415", "issue:9461", - "issue:9496", "issue:9648", "issue:993" ], "soft_pairs": [ + "issue:5573|issue:5997", + "issue:3884|issue:4839", "issue:10111|issue:8876", "issue:1195|issue:993", "issue:4356|issue:7000", @@ -52272,188 +52276,186 @@ "issue:10045|issue:7681", "issue:10950|issue:8850", "issue:10392|issue:9461", - "issue:11229|issue:9096", - "issue:10382|issue:9496", - "issue:4159|issue:4390" + "issue:11229|issue:9096" ], - "prepared_review_unit_hash": "570589de6fe35337daac7e5bffdba4eadd8dedbe56adcb254cba19aaaabd0c8c", + "prepared_review_unit_hash": "1e8291432fdbd5c4db84ecf1419a27a89d2669f7b0c7f55651c3000140005fe6", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33696, - "estimated_input_tokens": 8424, - "estimated_eval_tokens": 17104 + "serialized_chars": 34326, + "estimated_input_tokens": 8582, + "estimated_eval_tokens": 17420 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The proposed links are mostly generic keyword matches, not true duplicates. The items span many unrelated bugs, feature requests, and docs issues, so no single canonical duplicate stands out.", - "confidence": 0.96, - "canonical_issue_reason": "No clear canonical issue; the set is heterogeneous and the soft-edge candidates do not share the same underlying bug.", + "summary": "This cluster is highly heterogeneous; the soft pairs mostly share only broad subsystem keywords and do not look like the same underlying bug or change.", + "confidence": 0.87, + "canonical_issue_reason": "issue:10512 is the strongest representative issue in the cluster: it is a clear, highly active LoRA-loading failure report with concrete Flux/Quanto context and more discussion than the other similar-looking LoRA issues.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue representative: none of the issues cleanly captures a single duplicate cluster.", + "best_issue_reason": "issue:10512 is the best issue to keep as the representative bug report because it is specific, well-scoped, and has the most user activity among the candidate issue reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5573", + "right": "issue:5997", + "accept": false, + "reason": "Both are ControlNet/SDXL-adjacent, but one is a performance/init-time issue and the other is a model-type mismatch in a tutorial; different bugs." + }, + { + "left": "issue:3884", + "right": "issue:4839", + "accept": false, + "reason": "One is a training script failure and the other is a type error around lora_scale; no evidence they are the same concrete defect." + }, { "left": "issue:10111", "right": "issue:8876", "accept": false, - "reason": "Both mention FLUX, but one is a general FLUX pipeline bug report and the other is Pixart Sigma microconditioning; different models and failure modes." + "reason": "Different pipelines and different symptoms; this is only shared FLUX/Pipeline terminology, not the same bug." }, { "left": "issue:1195", "right": "issue:993", "accept": false, - "reason": "Both involve fp16/attention, but one is about memory efficient attention with fp16 weights and the other is a datatype mismatch in Stable Diffusion fp16; not the same bug." + "reason": "Both mention fp16/dtype problems, but one is about memory-efficient attention and the other is a general stable diffusion dtype mismatch." }, { "left": "issue:4356", "right": "issue:7000", "accept": false, - "reason": "Import error for a missing dummy module vs a deprecation warning in outputs.py; unrelated problems." + "reason": "A module import breakage vs a PyTorch deprecation warning; unrelated issues." }, { "left": "issue:6790", "right": "issue:7782", "accept": false, - "reason": "State_dict key mismatch during training vs safetensors unpickling error during conversion; different code paths and root causes." + "reason": "Unexpected state_dict keys during training is unrelated to a safetensors/UnpicklingError during conversion." }, { "left": "issue:4210", "right": "issue:4692", "accept": false, - "reason": "Question about attention_mask usage in UNet vs a memory issue in UNet2DConditionModel; not the same underlying defect." + "reason": "One asks how to use attention_mask; the other is a memory issue. No shared bug." }, { "left": "issue:10952", "right": "issue:8826", "accept": false, - "reason": "LoRA adapter-name mismatch when loading vs torch.cat with an empty tensor list; distinct failures." + "reason": "LoRA adapter-name loading error vs torch.cat empty-list runtime error; different code paths and failures." }, { "left": "issue:11675", "right": "issue:11790", "accept": false, - "reason": "Both are LoRA-loading reports, but they concern different models/scripts and different load-time errors; too broad to merge." + "reason": "Both are LoRA-loading complaints, but the titles point to different contexts and no concrete evidence shows the same root cause." }, { "left": "issue:10772", "right": "issue:13065", "accept": false, - "reason": "A controlnet support request vs a Qwen-Image controlnet inpaint/canny issue; unrelated." + "reason": "A ControlNet support request for Sana vs a Qwen-Image controlnet request; these are different feature requests." }, { "left": "issue:2689", "right": "issue:6439", "accept": false, - "reason": "Both mention broken links, but they point to different assets/docs and different pages." + "reason": "Both are broken links, but they are different docs issues with no indication of the same link or page." }, { "left": "issue:9343", "right": "issue:9415", "accept": false, - "reason": "FLUX load mismatch with ignore_mismatched_sizes vs a CogVideoXTransformer3DModel parameter-module error; different models and bugs." + "reason": "FLUX loading with mismatched sizes is unrelated to adding a new parameter module in CogVideoXTransformer3DModel." }, { "left": "issue:10512", "right": "issue:9461", "accept": false, - "reason": "LoRA loading failure vs a vague Flux inference issue; not enough evidence they are the same concrete bug." + "reason": "One is a specific LoRA load failure; the other is an \u0905\u0938\u094d\u092a\u0937\u094d\u091f/unfinished FLUX inference issue title with no matching bug detail." }, { "left": "issue:462", "right": "issue:4724", "accept": false, - "reason": "Both are img2img-related, but one is an Apple MPS regression and the other is an SDXL checkpoint behavior issue; different contexts." + "reason": "Both involve img2img pipelines, but one is an Apple MPS regression and the other is an SDXL checkpoint behavior mismatch." }, { "left": "issue:1594", "right": "issue:3838", "accept": false, - "reason": "Img2Img for video generation vs Video Composer feature request; related domain, but not the same change or bug." + "reason": "A feature request for img2img video generation is not the same as a Video Composer request." }, { "left": "issue:10382", "right": "issue:9648", "accept": false, - "reason": "VAE forward return-type mismatch causing AttributeError vs VAE tiling causing memory overflow; unrelated symptoms." + "reason": "A VAE forward return-type AttributeError is different from a VAE tiling memory overflow." }, { "left": "issue:10950", "right": "issue:7210", "accept": false, - "reason": "Docs link broken vs accelerate dispatch hitting named_parameters on a pipeline; different issue types." + "reason": "A broken documentation section and a runtime accelerate dispatch error are unrelated." }, { "left": "issue:1538", "right": "issue:3162", "accept": false, - "reason": "Model incompatibility with stable-diffusion-2-base vs missing learned_classifier_free_sampling_embeddings; different failures." + "reason": "StableDiffusionMegaPipeline checkpoint compatibility and a classifier-free-sampling embedding error are different problems." }, { "left": "issue:13137", "right": "issue:13221", "accept": false, - "reason": "Both mention LoRA support/loading, but they concern different models and error classes; too broad to treat as duplicates." + "reason": "Both are LoRA support/load issues, but they concern different model families and different error descriptions; not enough to treat as the same bug." }, { "left": "issue:1478", "right": "issue:4349", "accept": false, - "reason": "Multi-image output limitation vs missing text_encoder despite prompt_embeds; not the same bug." + "reason": "Both are upscale-pipeline issues, but one is multi-image output handling and the other is text_encoder/prompt_embeds handling." }, { "left": "issue:4000", "right": "issue:8723", "accept": false, - "reason": "Checkpoint conversion problem vs SD3ControlNetModel initialization error; unrelated." + "reason": "Checkpoint-to-diffusers conversion is not the same as SD3ControlNetModel initialization failure." }, { "left": "issue:10045", "right": "issue:7681", "accept": false, - "reason": "General pipeline import problem vs environment-specific OpenSSL/lib import failure; not the same underlying issue." + "reason": "Both are import-related, but one is a generic pipeline import problem and the other is a specific OpenSSL/lib symbol failure." }, { "left": "issue:10950", "right": "issue:8850", "accept": false, - "reason": "Broken docs section vs broken community pipelines mirror; both are broken links, but different targets." + "reason": "Broken docs section vs broken community pipelines mirror; different artifacts and failures." }, { "left": "issue:10392", "right": "issue:9461", "accept": false, - "reason": "Flux-dev-fp8/Lora interaction issue vs a vague Flux inference report; insufficient overlap to call them duplicates." + "reason": "Both mention FLUX, but one is a specific fp8/lora setup and the other is an unclear FLUX inference issue; not the same bug." }, { "left": "issue:11229", "right": "issue:9096", "accept": false, - "reason": "Both mention NaNs, but one is attention slicing on SDXL/MPS and the other is Flux half-precision inference; different code paths." - }, - { - "left": "issue:10382", - "right": "issue:9496", - "accept": false, - "reason": "VAE return-type mismatch vs SD3ControlNet forward-function error; distinct model components and failures." - }, - { - "left": "issue:4159", - "right": "issue:4390", - "accept": false, - "reason": "High VRAM usage vs missing requirements; no duplicate relation." + "reason": "Both are NaN reports, but they affect different models, devices, and mechanisms (attention slicing on MPS vs half-precision Flux inference)." } ] }, @@ -52472,13 +52474,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10025", - "issue:10282", + "issue:10382", "issue:10514", "issue:10710", "issue:1083", @@ -52487,10 +52489,7 @@ "issue:11275", "issue:11283", "issue:11709", - "issue:12009", - "issue:12113", "issue:12126", - "issue:12348", "issue:12529", "issue:1475", "issue:1534", @@ -52503,7 +52502,9 @@ "issue:3405", "issue:3728", "issue:3891", + "issue:4159", "issue:4314", + "issue:4390", "issue:4542", "issue:5429", "issue:5462", @@ -52524,9 +52525,12 @@ "issue:7968", "issue:7972", "issue:8790", - "issue:9023" + "issue:9023", + "issue:9496" ], "soft_pairs": [ + "issue:10382|issue:9496", + "issue:4159|issue:4390", "issue:12126|issue:12529", "issue:11023|issue:11275", "issue:10025|issue:7968", @@ -52537,7 +52541,6 @@ "issue:5606|issue:6368", "issue:1083|issue:2941", "issue:5429|issue:5462", - "issue:10282|issue:12009", "issue:10514|issue:10710", "issue:1534|issue:668", "issue:5769|issue:5772", @@ -52549,191 +52552,187 @@ "issue:3728|issue:7405", "issue:3405|issue:4314", "issue:2618|issue:3140", - "issue:3250|issue:5523", - "issue:12113|issue:12348" + "issue:3250|issue:5523" ], - "prepared_review_unit_hash": "3474ed79761d74cb654f393987ca9d65b565cea9c70759c28dbe421b22f2583a", + "prepared_review_unit_hash": "676958d4fa95aeec2134876c4af9a28e8c5d649a6aa20780b61314bbb55f68df", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33942, - "estimated_input_tokens": 8486, - "estimated_eval_tokens": 17228 + "serialized_chars": 33664, + "estimated_input_tokens": 8416, + "estimated_eval_tokens": 17088 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous; only one soft pair looks like a true duplicate: the WanImageToVideoPipeline resolution/tensor-mismatch reports. The rest are superficial lexical matches across different bugs, models, or code paths.", - "confidence": 0.87, - "canonical_issue_reason": "issue:12348 is the broadest and most representative Wan report: it is open, describes a resolution-dependent tensor mismatch, and can subsume the more specific channel-count failure in 12113.", - "canonical_pr_reason": null, - "best_issue_reason": "issue:12348 is the best anchor for deduping because it captures the underlying WanImageToVideoPipeline resolution mismatch without narrowing to one exact tensor shape error.", - "best_pr_reason": null, + "summary": "These are mostly false-positive similarities: the items span unrelated bugs, feature requests, and model-specific failures, so there is no real duplicate cluster to collapse.", + "confidence": 0.96, + "canonical_issue_reason": "No clear canonical issue exists here because the candidates are not the same underlying bug or change; they cover different subsystems and failure modes.", + "canonical_pr_reason": "No pull requests are present in this set.", + "best_issue_reason": "No single issue is a good global representative for the cluster; if forced to pick a standalone diagnostic report, #10382 is the most detailed, but it does not match the rest.", + "best_pr_reason": "No pull requests are present in this set.", "soft_edge_verdicts": [ + { + "left": "issue:10382", + "right": "issue:9496", + "accept": false, + "reason": "Both mention forward/type errors in VAE-related code, but they are different models and different failure paths." + }, + { + "left": "issue:4159", + "right": "issue:4390", + "accept": false, + "reason": "VRAM blow-up during DiffEdit is unrelated to missing requirements." + }, { "left": "issue:12126", "right": "issue:12529", "accept": false, - "reason": "Both are Wan-related, but one is a cache_context meta-tensor switch bug and the other is a tiled-VAE tensor-size mismatch; different failing code paths." + "reason": "Both are Wan2.2 issues, but one is a cache_context NoneType error and the other is a tiled VAE tensor mismatch." }, { "left": "issue:11023", "right": "issue:11275", "accept": false, - "reason": "Different models and failures: blurred images in a custom SDXL pipeline vs a VAE NoneType addition error in ResnetUpsampleBlock2D." + "reason": "Both are image-quality/runtime problems, but they involve different pipelines and distinct root causes." }, { "left": "issue:10025", "right": "issue:7968", "accept": false, - "reason": "Both mention attention, but one is Flux transformer attention masking and the other is UNetMotionModel attention internals; not the same bug." + "reason": "They only share the word attention; one is about Flux transformer attention masks, the other about an UNet motion-module attention component." }, { "left": "issue:1565", "right": "issue:1608", "accept": false, - "reason": "One is GPU selection / device placement, the other is training stopping at 500 steps; unrelated issues." + "reason": "Both concern training on GPUs, but one is GPU selection/OOM and the other is a training stop at 500 steps." }, { "left": "issue:5510", "right": "issue:7972", "accept": false, - "reason": "Both are type-mismatch flavored, but they affect different pipelines and likely different casting paths; too broad to treat as one bug." + "reason": "Both are type-mismatch reports, but they affect different pipelines and different tensor/bias handling." }, { "left": "issue:10920", "right": "issue:9023", "accept": false, - "reason": "A scheduler inf-at-first-step bug is not the same as a vague sample code failure." + "reason": "Both are generic runtime failures, but the scheduler inf bug is unrelated to sample code execution failures." }, { "left": "issue:11283", "right": "issue:11709", "accept": false, - "reason": "Both are Wan loading errors, but one is header deserialization and the other is a CausVid LoRA key error; not the same underlying failure." + "reason": "A header deserialization error and a LoRA key error are different loading problems with no shared code path shown." }, { "left": "issue:5606", "right": "issue:6368", "accept": false, - "reason": "Both mention LoRA alpha, but one is missing-alpha handling during load and the other is unsupported-alpha checkpoint parsing; different bugs." + "reason": "Both involve LoRA alpha handling, but one is missing alpha value and the other is unsupported checkpoint layer naming." }, { "left": "issue:1083", "right": "issue:2941", "accept": false, - "reason": "FP16 ONNX incorrect output and EulerAncestral scheduler results are different issues in different subsystems." + "reason": "FP16 ONNX output correctness is unrelated to EulerAncestralDiscreteScheduler behavior in Stable Diffusion." }, { "left": "issue:5429", "right": "issue:5462", "accept": false, - "reason": "Support for models without VAE in examples is unrelated to adding img2img support for LCMs." - }, - { - "left": "issue:10282", - "right": "issue:12009", - "accept": false, - "reason": "Both are Wan model loading problems, but one is an x_embedder shape mismatch and the other is a meta-tensor copy failure for GGUF; not the same code-path problem." + "reason": "One is about supporting models without VAE in examples, the other is about adding img2img for LCMs." }, { "left": "issue:10514", "right": "issue:10710", "accept": false, - "reason": "Sana resolution-binning/sample_size support and DDUF format support are unrelated." + "reason": "Sana resolution binning support is unrelated to DDUF format support." }, { "left": "issue:1534", "right": "issue:668", "accept": false, - "reason": "Both involve DreamBooth training, but the notebook error and T4 runtime/OOM issue are distinct." + "reason": "Both are DreamBooth notebook/training issues, but they do not describe the same failure mode." }, { "left": "issue:5769", "right": "issue:5772", "accept": false, - "reason": "Mixed precision training correctness and dataset/webdataset speed are unrelated concerns." + "reason": "Mixed-precision training correctness and dataset performance are unrelated." }, { "left": "issue:4542", "right": "issue:7377", "accept": false, - "reason": "Both involve num_images_per_prompt/batching, but they are different pipelines and different concrete failures." + "reason": "Both are multi-image generation problems, but they affect different pipelines and different internal components." }, { "left": "issue:1475", "right": "issue:5877", "accept": false, - "reason": "Training on two image classes vs a subprocess error in DreamBooth/textual inversion are not the same bug." + "reason": "DreamBooth/textual inversion class-image handling is unrelated to subprocess errors in those training scripts." }, { "left": "issue:6869", "right": "issue:7178", "accept": false, - "reason": "Both mention export_to_video, but one is a specific unpacking exception and the other is a generic non-working report; insufficiently concrete to merge." + "reason": "Both mention export_to_video, but one is a concrete unpacking ValueError and the other is a generic malfunction report; not enough to treat as the same bug." }, { "left": "issue:3891", "right": "issue:7408", "accept": false, - "reason": "Different new pipeline requests: Tortoise TTS vs FRESCO." + "reason": "These are separate pipeline feature requests for different models (Tortoise TTS vs FRESCO)." }, { "left": "issue:7243", "right": "issue:8790", "accept": false, - "reason": "ResAdapter and SDXL TensorRT + IP Adapter are different features, not duplicate bugs." + "reason": "ResAdapter and SDXL TensorRT/IP-Adapter are different model-integration topics." }, { "left": "issue:3728", "right": "issue:7405", "accept": false, - "reason": "Separate new-model/pipeline requests for Shifted Diffusion and DynamiCrafter." + "reason": "They are different new-model requests: Shifted Diffusion vs DynamiCrafter." }, { "left": "issue:3405", "right": "issue:4314", "accept": false, - "reason": "ControlNet training bug and interrupted-download write-protection are unrelated." + "reason": "ControlNet training errors and interrupted-download write protection are unrelated." }, { "left": "issue:2618", "right": "issue:3140", "accept": false, - "reason": "Accepting an image-input bug and an import error for the same pipeline would overreach; these are different failure modes." + "reason": "Passing torch.FloatTensor image inputs to ControlNet is unrelated to the import error for StableDiffusionControlNetPipeline." }, { "left": "issue:3250", "right": "issue:5523", "accept": false, - "reason": "Unable to run a specific pipeline and inconsistent results across runs are not the same underlying bug." - }, - { - "left": "issue:12113", - "right": "issue:12348", - "accept": true, - "reason": "Both report WanImageToVideoPipeline failures that vary with resolution and manifest as tensor/channel mismatches; 12113 looks like a specific instance of the broader issue in 12348." + "reason": "A pipeline execution failure and inconsistent results are too generic and do not indicate the same underlying bug." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative: it treats the cluster as mostly heterogeneous and only elevates the WanImageToVideoPipeline pair (12113/12348), which is supported by the titles as a resolution-dependent tensor/channel mismatch. The other soft-edge rejections are phrased cautiously and do not overstate duplicate evidence." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -52745,8 +52744,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -52764,6 +52763,8 @@ "issue:11342", "issue:11489", "issue:11749", + "issue:12113", + "issue:12348", "issue:12369", "issue:12752", "issue:3159", @@ -52773,7 +52774,6 @@ "issue:5335", "issue:5710", "issue:6095", - "issue:6287", "issue:6363", "issue:6439", "issue:6463", @@ -52781,7 +52781,6 @@ "issue:7180", "issue:7303", "issue:7418", - "issue:7441", "issue:7456", "issue:7562", "issue:7629", @@ -52800,6 +52799,7 @@ "issue:9933" ], "soft_pairs": [ + "issue:12113|issue:12348", "issue:7180|issue:7562", "issue:10050|issue:12752", "issue:10054|issue:9933", @@ -52822,191 +52822,190 @@ "issue:11489|issue:9729", "issue:3159|issue:6439", "issue:10937|issue:11749", - "issue:11223|issue:11283", - "issue:6287|issue:7441" + "issue:11223|issue:11283" ], - "prepared_review_unit_hash": "581fd296433be1614f244c64bddbd68b2d409047783e2d922b7c9602474382ea", + "prepared_review_unit_hash": "df8e04adb95524e6a787484ec613da6c4312ec78fb5f68566ec88c7e93b00f13", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34121, - "estimated_input_tokens": 8531, - "estimated_eval_tokens": 17318 + "serialized_chars": 34264, + "estimated_input_tokens": 8566, + "estimated_eval_tokens": 17388 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The packet does not support any true duplicate merges with high confidence. The listed pairs are mostly thematic similarities across broad model/pipeline areas, but the evidence shown here is too weak and non-specific to treat them as the same underlying bug or change. No PRs are present.", - "confidence": 0.23, - "canonical_issue_reason": "There is no single canonical issue for this cluster because the items span unrelated bugs, feature requests, and docs/questions. The soft pairs are not conservatively established duplicates from the packet alone.", + "summary": "Mostly heterogeneous issue set, with a handful of likely duplicates around WanImageToVideoPipeline tensor mismatches, SDXL ControlNet CPU offload, lpw_stable_diffusion_xl output corruption, Flux LoRA training, and HunyuanVideo bitsandbytes device handling.", + "confidence": 0.64, + "canonical_issue_reason": "Issue 12113 is the clearest concrete bug report in the set, and 12348 reads like a broader follow-up of the same WanImageToVideoPipeline resolution-specific tensor mismatch.", "canonical_pr_reason": null, - "best_issue_reason": "If one issue must be used as the most concrete representative, issue 7562 is the strongest standalone candidate because it has a specific pipeline, a clear failure mode, and a focused reproducer. That said, it is not justified as a canonical duplicate anchor for the whole cluster.", + "best_issue_reason": "12113 has the most specific repro/error message and is the best representative among the true duplicate-like pairs.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12113", + "right": "issue:12348", + "accept": true, + "reason": "Same WanImageToVideoPipeline tensor/channel mismatch at certain resolutions; 12348 is the broader symptom report." + }, { "left": "issue:7180", "right": "issue:7562", - "accept": false, - "reason": "Related ControlNet/offload area, but the packet only shows thematic overlap and no shared filenames, targets, or concrete proof of the same bug." + "accept": true, + "reason": "Both describe SDXL ControlNet Img2Img breaking under CPU offload; 7562 is the multicontrolnet-specific form of the same bug." }, { "left": "issue:10050", "right": "issue:12752", "accept": false, - "reason": "Both involve img2img/inpainting themes, but one is a capability question and the other is a pipeline request for a different model family." + "reason": "Different requests for different pipelines/models: img2img KDiffusion support vs Z-Image img2img/inpainting support." }, { "left": "issue:10054", "right": "issue:9933", "accept": false, - "reason": "A scheduler linspace bug and an img2img API parameter issue are different code-path problems." + "reason": "Scheduler torch.linspace parameter bug vs missing width/height parameters in SD3 img2img are unrelated." }, { "left": "issue:3880", "right": "issue:5710", "accept": false, - "reason": "Inpainting degradation and scheduler initialization missing sigmas are unrelated failures." + "reason": "Inpainting quality degradation and missing sigmas initialization are different code paths." }, { "left": "issue:4368", "right": "issue:8560", "accept": false, - "reason": "Both are OOM reports, but the model/training contexts differ and the packet gives no evidence of the same root cause." + "reason": "Both are OOM reports, but for different models and training setups (SDXL LoRA vs SD3 training)." }, { "left": "issue:6363", "right": "issue:6695", "accept": false, - "reason": "Training gradient scaling error versus subprocess startup failure; no shared bug path is shown." + "reason": "train_text_to_image_lora AMP unscale error is unrelated to the subprocess CalledProcessError report." }, { "left": "issue:6463", "right": "issue:7456", "accept": false, - "reason": "A missing pipeline request is unrelated to a documentation/code typo about an exception path." + "reason": "Missing audio_diffusion pipeline and a missing raise statement are not the same bug." }, { "left": "issue:7303", "right": "issue:9397", "accept": false, - "reason": "Scheduler-related titles are too broad here; zero_snr creation and SGMUniform support are different changes." + "reason": "Both mention schedulers, but one is about zero_snr creation and the other about Hyper LoRA support; not the same change." }, { "left": "issue:7681", "right": "issue:9034", "accept": false, - "reason": "Import/OpenSSL failure and an unnecessary dependency requirement are unrelated." + "reason": "OpenSSL import failure and an unexpected sentencepiece dependency are unrelated issues." }, { "left": "issue:6095", "right": "issue:7629", "accept": false, - "reason": "Different scheduler/runtime problems: an IndexError in one case and version-specific OOM in the other." + "reason": "Euler discrete scheduling IndexError is distinct from PyTorch-version-specific OOM in SVD." }, { "left": "issue:10573", "right": "issue:7931", "accept": false, - "reason": "Same pipeline family, but the feature interactions differ: multicontrolnet padding crop versus IP-Adapter embeds." + "reason": "Both are ControlNet inpaint issues, but one is padding_mask_crop with multiple controlnets and the other is IP-Adapter embeds." }, { "left": "issue:10500", "right": "issue:10653", - "accept": false, - "reason": "Similar HunyuanVideo/bitsandbytes context, but the packet does not establish the same concrete failure path." + "accept": true, + "reason": "Same HunyuanVideo bitsandbytes/device mismatch class; both point to 8bit model CPU/GPU placement problems." }, { "left": "issue:8850", "right": "issue:9680", "accept": false, - "reason": "One is a community pipeline mirror/infrastructure issue, the other is an SDXL generation bug." + "reason": "Community pipelines mirror breakage is unrelated to SDXL unconditional generation." }, { "left": "issue:11023", "right": "issue:7848", - "accept": false, - "reason": "Both mention lpw_stable_diffusion_xl, but the packet only supports broad similarity, not a confirmed duplicate bug." + "accept": true, + "reason": "Both point to lpw_stable_diffusion_xl producing degraded/blurred output due to the same custom-pipeline decode issue." }, { "left": "issue:5335", "right": "issue:7418", "accept": false, - "reason": "A deployment question and a code bug in a community pipeline are not the same issue." + "reason": "Local deployment/network access question is unrelated to a bug in a community upscale pipeline." }, { "left": "issue:9025", "right": "issue:9031", "accept": false, - "reason": "Same pipeline class, but the problems are different: from_single_file on TPU versus disabling the safety checker." + "reason": "One is about from_single_file on TPU VM; the other is about disabling the safety checker." }, { "left": "issue:4616", "right": "issue:7636", "accept": false, - "reason": "A LoRA conversion script bug and a controlnet usage question are different topics." + "reason": "SDXL LoRA conversion script failure is not the same as how to use the SDXL tile ControlNet model." }, { "left": "issue:11229", "right": "issue:11342", "accept": false, - "reason": "Both mention MPS, but they affect different pipelines/models and do not show the same failure mode." + "reason": "Both involve MPS, but they are different pipelines with different failure modes." }, { "left": "issue:10639", "right": "issue:12369", "accept": false, - "reason": "Unrelated support requests for different model families and capabilities." + "reason": "Different feature requests for different models: Lumina-Image 2.0 support vs Wan 2.1 4K scaling." }, { "left": "issue:11489", "right": "issue:9729", - "accept": false, - "reason": "The titles suggest the same script family, but the packet does not provide enough concrete overlap to confirm the same bug." + "accept": true, + "reason": "Both are the same Flux LoRA advanced training failure; 9729 states the concrete T5EncoderModel save-model error." }, { "left": "issue:3159", "right": "issue:6439", "accept": false, - "reason": "Two documentation-related issues, but they concern different docs and different problems." + "reason": "Docs typo in a code example and a 404 link in another guide are unrelated documentation issues." }, { "left": "issue:10937", "right": "issue:11749", "accept": false, - "reason": "torch.compile/VAE encode errors and scheduler floating-point issues are distinct bugs." + "reason": "torch.compile on vae.encode and FlowMatchEulerDiscreteScheduler floating-point timestep errors are different bugs." }, { "left": "issue:11223", "right": "issue:11283", "accept": false, - "reason": "OOM with layer-wise upcasting and a WAN header deserialization error are not the same failure." - }, - { - "left": "issue:6287", - "right": "issue:7441", - "accept": false, - "reason": "Dependency cleanup and a pytest Python 3.12 failure are unrelated." + "reason": "WAN layer-wise upcasting OOM with LoRAs and WAN header deserialization are separate failures." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The summary is too confident for the evidence in the packet. Every listed pair is only supported by soft similarity, with deterministic_accept=false and no shared filenames/targets, so the analyst should not mark any of them as true duplicates without stronger grounding. In particular, the accepted pairs (7180/7562, 10500/10653, 11023/7848, 11489/9729) are plausible thematic matches but not conservatively established as mergeable duplicate issues from the packet alone. The overall claim that there are 'a few true duplicate-like pairs' overstates the evidence." + "accept": true, + "feedback": "The summary is grounded in the provided packet and stays appropriately conservative. The accepted soft-edge pairs are all plausibly same-bug clusters based on the titles, and the rejected pairs are sensible distinctions rather than overclaims. No mergeability issue since these are issue pairs, not PR pairs." }, "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": false, + "retried": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -53054,6 +53053,7 @@ "issue:539", "issue:6142", "issue:6162", + "issue:6287", "issue:6368", "issue:6506", "issue:6725", @@ -53061,18 +53061,18 @@ "issue:7177", "issue:7219", "issue:7243", + "issue:7441", "issue:8373", "issue:8410", "issue:8613", "issue:8863", "issue:8885", "issue:8975", - "issue:9450", "issue:9551", - "issue:9731", - "issue:9904" + "issue:9731" ], "soft_pairs": [ + "issue:6287|issue:7441", "issue:6725|issue:9731", "issue:3085|issue:539", "issue:1941|issue:2728", @@ -53095,181 +53095,180 @@ "issue:11063|issue:8373", "issue:4615|issue:6142", "issue:10757|issue:8975", - "issue:10282|issue:6368", - "issue:9450|issue:9904" + "issue:10282|issue:6368" ], - "prepared_review_unit_hash": "7a97b3129e8fe568ab27b24e78bd6f9940646fd85bd3feed05bcc44d6b16a943", + "prepared_review_unit_hash": "e869ef94fe3c996985d3c4df189116059808e808228051c7fc310bf625040f38", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34070, - "estimated_input_tokens": 8518, - "estimated_eval_tokens": 17292 + "serialized_chars": 33984, + "estimated_input_tokens": 8496, + "estimated_eval_tokens": 17248 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft duplicate edges look like the same underlying bug or change. The set is a mix of unrelated bug reports and feature requests that only share broad subsystems (often Flux, loading, schedulers, or pipeline support), not a concrete fix target.", - "confidence": 0.93, - "canonical_issue_reason": "If a single issue must represent the set, issue 9450 is the strongest candidate because it is a concrete, reproducible bug report with a clear FluxPipeline device-mismatch symptom. That said, the items are not true duplicates of one another.", + "summary": "This cluster is mostly a false-positive mix of unrelated Diffusers bugs, feature requests, and environment/loading errors. None of the soft pairs look like the same underlying bug or change, so I reject all of them.", + "confidence": 0.92, + "canonical_issue_reason": "No true duplicate set emerges here; if a representative issue must be chosen, issue:7219 is the most substantive and active standalone request in the group, but it is not a duplicate of the others.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 9450 is the most actionable issue in the batch: it names the exact failure mode, context, and code path, making it a better representative bug report than the more request-like or vague items.", + "best_issue_reason": "Issue:7219 has the highest discussion activity and is a broad, concrete feature request, so it is the best single anchor for this mixed cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6287", + "right": "issue:7441", + "accept": false, + "reason": "Dependency cleanup/setup.py and Python 3.12 pytest breakage are unrelated issues." + }, { "left": "issue:6725", "right": "issue:9731", "accept": false, - "reason": "Different problems: one is generating images with LDMPipeline, the other is training a LoRA/style workflow for Playground2.5." + "reason": "Different requests: LDMPipeline image generation vs Playground2.5 LoRA training guidance." }, { "left": "issue:3085", "right": "issue:539", "accept": false, - "reason": "Different API regressions: cache-folder argument removal vs snapshot_download allow_patterns compatibility." + "reason": "Both are API/version friction, but they concern different functions and different failure modes." }, { "left": "issue:1941", "right": "issue:2728", "accept": false, - "reason": "Different loading failures: custom UnCLIP pipeline loading versus safetensors loading for UNet2DConditionModel." + "reason": "Custom pipeline loading for UnCLIP vs safetensors loading for UNet2DConditionModel are distinct code paths." }, { "left": "issue:3623", "right": "issue:4446", "accept": false, - "reason": "One is a large-model loading/memory issue; the other is about custom text_encoder support in from_single_file." + "reason": "General large-model loading problem vs custom text_encoder support in from_single_file; not the same bug." }, { "left": "issue:7243", "right": "issue:9551", "accept": false, - "reason": "Different feature requests in the same broad area: ResAdapter support vs Flux ControlNet usage." + "reason": "Separate model-support requests: ResAdapter vs Flux ControlNet usage." }, { "left": "issue:6162", "right": "issue:8863", "accept": false, - "reason": "Both mention shape-related failures, but the concrete bugs differ: IP-Adapter mismatch vs tuple lacking a shape attribute." + "reason": "IP-Adapter SDXL shape mismatch and tuple.shape AttributeError are different runtime failures." }, { "left": "issue:4648", "right": "issue:7177", "accept": false, - "reason": "Different checkpoint/config loading paths and different symptoms; not the same bug." + "reason": "Missing config.json in a directory is not the same as StableCascadeDecoderPipeline.from_pretrained failing." }, { "left": "issue:6506", "right": "issue:6839", "accept": false, - "reason": "Output-type return bug versus from_single_file loading failure for StableVideoDiffusionPipeline." + "reason": "One is an output_type bug, the other is from_single_file loading support for Stable Video Diffusion." }, { "left": "issue:11519", "right": "issue:8410", "accept": false, - "reason": "Separate model support requests: MAGI-1 versus TextDiffuser pipelines." + "reason": "Both are feature requests for new models, but they target different pipelines/models entirely." }, { "left": "issue:1433", "right": "issue:1445", "accept": false, - "reason": "Related ONNX area, but the concrete failures differ: missing config attribute vs legacy inpaint pipeline attribute error." + "reason": "Different ONNX-related AttributeErrors in different pipeline classes." }, { "left": "issue:3019", "right": "issue:4609", "accept": false, - "reason": "Unrelated feature requests: text-guided image editing with SAM versus providing a T2I-Adapter model." + "reason": "Unrelated community model requests: text-guided editing with SAM vs t2iadapter_style." }, { "left": "issue:11023", "right": "issue:8613", "accept": false, - "reason": "Different issues: blurry outputs in a custom pipeline versus OOM during training/input processing." + "reason": "Playground v2.5 blur/custom pipeline issue and a caption_column OOM are unrelated." }, { "left": "issue:3901", "right": "issue:4159", "accept": false, - "reason": "A missing requirements.txt file is unrelated to a DiffEdit VRAM exhaustion problem." + "reason": "Missing requirements.txt and DiffEdit VRAM usage are completely different problems." }, { "left": "issue:10057", "right": "issue:8885", "accept": false, - "reason": "Different custom-loading/token problems in different workflows; not the same underlying bug." + "reason": "FluxTransformer2DModel.from_single_file token loading and Custom Diffusion modifier-token installation are different issues." }, { "left": "issue:10321", "right": "issue:10398", "accept": false, - "reason": "One is a missing model file error for LTX-Video; the other is target modules not found for a base model." + "reason": "Missing model files in LTX-Video vs missing target modules in a base model are separate loading/training errors." }, { "left": "issue:2311", "right": "issue:2635", "accept": false, - "reason": "Scheduler-related, but one asks for UniPC scheduler support while the other maps DPM++ names to scheduler types." + "reason": "UniPC scheduler request and scheduler-name mapping question are related only broadly, not the same issue." }, { "left": "issue:1037", "right": "issue:1290", "accept": false, - "reason": "Separate community-pipeline requests for different models." + "reason": "Different community pipeline proposals; no shared bug or concrete implementation target." }, { "left": "issue:3043", "right": "issue:7219", "accept": false, - "reason": "Both are feature requests, but for different additions: ODISE vs SUPIR Upscaler." + "reason": "Different feature requests for different models/pipelines (ODISE vs SUPIR)." }, { "left": "issue:1207", "right": "issue:2044", "accept": false, - "reason": "Completely different runtime errors: CUDA symbol lookup failure vs DDP missing dtype." + "reason": "CUDA runtime symbol failure and DistributedDataParallel dtype attribute error are unrelated runtime/environment issues." }, { "left": "issue:11063", "right": "issue:8373", "accept": false, - "reason": "Different concerns: attention-mask padding logic versus an error-message wording issue." + "reason": "Incorrect padding in prepare_attention_mask and a generic incorrect error message are not the same bug." }, { "left": "issue:4615", "right": "issue:6142", "accept": false, - "reason": "Import error for a pipeline class versus a NameError for start_training." + "reason": "Import error for ControlNet pipeline and undefined start_training are unrelated." }, { "left": "issue:10757", "right": "issue:8975", "accept": false, - "reason": "A variable rename cleanup is unrelated to an import error for LoraLoaderMixin." + "reason": "A variable-name fix in a converter script is unrelated to the LoraLoaderMixin import error." }, { "left": "issue:10282", "right": "issue:6368", "accept": false, - "reason": "Different model-loading failures: x_embedder weight shape mismatch vs unsupported LoRA checkpoint layer." - }, - { - "left": "issue:9450", - "right": "issue:9904", - "accept": false, - "reason": "Both are FluxPipeline issues, but one is a multi-GPU device mismatch and the other is silent output-shape rounding; different bugs." + "reason": "Shape mismatch while loading x_embedder weights and unsupported lora alpha checkpoint layers are different checkpoint-loading problems." } ] }, @@ -53288,8 +53287,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -53333,16 +53332,17 @@ "issue:6688", "issue:7142", "issue:785", - "issue:8485", "issue:8588", - "issue:8834", "issue:8880", "issue:8918", "issue:9258", + "issue:9450", "issue:9895", + "issue:9904", "issue:9958" ], "soft_pairs": [ + "issue:9450|issue:9904", "issue:1207|issue:1544", "issue:1539|issue:2321", "issue:1617|issue:2221", @@ -53367,193 +53367,192 @@ "issue:6278|issue:6518", "issue:10010|issue:10675", "issue:4436|issue:5300", - "issue:6234|issue:8918", - "issue:8485|issue:8834" + "issue:6234|issue:8918" ], - "prepared_review_unit_hash": "25afe53a6796405da625e2428e512fc46c7af08f233f00f31fe020495a682297", + "prepared_review_unit_hash": "697d230343d49a19bd74747e0d7691fe2fd04a48c6e665e7630b448c5e754244", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 34278, - "estimated_input_tokens": 8570, - "estimated_eval_tokens": 17396 + "serialized_chars": 34382, + "estimated_input_tokens": 8596, + "estimated_eval_tokens": 17448 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a heterogeneous mix of unrelated issues that only share broad Diffusers/Flux wording. I do not see any pair that clearly describes the same underlying bug or change, so I would reject all soft-edge duplicate candidates.", - "confidence": 0.89, - "canonical_issue_reason": "issue:10111 is the broadest Flux-pipeline bug report and the closest thing to a cluster anchor, but it is still too generic to serve as a true duplicate target for the other issues.", + "summary": "This cluster is mostly a noisy mix of unrelated issues, with a few Flux-related reports that are close in theme but not the same underlying bug. I would not merge any of the soft-edge pairs as duplicates.", + "confidence": 0.35, + "canonical_issue_reason": "issue:10111 is the broadest, most umbrella Flux bug report in the set, so it is the best single issue anchor for this noisy cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9895 is the most concrete, actionable issue in the set: it names a specific FluxPipeline regression, a failing error, and a suspect commit.", + "best_issue_reason": "issue:10111 best represents the dominant Flux-related portion of the cluster without overfitting to one narrow symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:9450", + "right": "issue:9904", + "accept": false, + "reason": "Both are FluxPipeline bugs, but one is a multi-GPU device placement error and the other is silent shape rounding; different code paths and fixes." + }, { "left": "issue:1207", "right": "issue:1544", "accept": false, - "reason": "Different scheduler/runtime failures: CUDA symbol resolution vs missing DDPMScheduler.get_velocity." + "reason": "One is a CUDA runtime symbol loading failure, the other is a missing scheduler method; unrelated failures." }, { "left": "issue:1539", "right": "issue:2321", "accept": false, - "reason": "Unrelated problems: conversion-script type handling vs EMA weight copying in inference." + "reason": "Conversion script type mismatch vs inference EMA/model copying semantics; not the same bug." }, { "left": "issue:1617", "right": "issue:2221", "accept": false, - "reason": "Different training/runtime errors with no shared code-path or symptom." + "reason": "Argument parsing error and a ColossalAI runtime error are distinct problems." }, { "left": "issue:10010", "right": "issue:4648", "accept": false, - "reason": "Both are file-not-found style reports, but one is a remote Flux Fill model URL 404 and the other is a local config.json lookup failure." + "reason": "Both mention missing files, but they refer to different assets and setup failures; not one underlying defect." }, { "left": "issue:1290", "right": "issue:541", "accept": false, - "reason": "Community model request vs SageMaker deployment topic; not the same bug/change." + "reason": "Community project request versus SageMaker support question; no shared bug/change." }, { "left": "issue:11023", "right": "issue:12995", "accept": false, - "reason": "Different models and questions: blurry SDXL custom pipeline output vs Cosmos-Predict2.5 video behavior." + "reason": "Different pipelines and different questions: blurry output in a custom pipeline vs video generation/reconstruction behavior." }, { "left": "issue:10111", "right": "issue:10675", "accept": false, - "reason": "Generic Flux pipeline bug report vs a specific scheduler max_shift configuration discrepancy." + "reason": "Related to Flux, but one is a general bug report and the other is a scheduler max_shift configuration discrepancy." }, { "left": "issue:10796", "right": "issue:11396", "accept": false, - "reason": "Unrelated docs/how-to questions for different models and workflows." + "reason": "Different model families and different documentation/format questions." }, { "left": "issue:197", "right": "issue:200", "accept": false, - "reason": "Delete dead code vs add tests for resnet; not the same underlying issue." + "reason": "Dead code cleanup request versus a test task; not duplicates." }, { "left": "issue:2560", "right": "issue:462", "accept": false, - "reason": "Different pipeline regressions with different symptoms and platforms." + "reason": "Different pipelines and symptoms: generator-list nondeterminism vs MPS release breakage." }, { "left": "issue:3043", "right": "issue:3891", "accept": false, - "reason": "Separate feature requests for different pipelines/models." + "reason": "Separate feature requests for entirely different pipelines/models." }, { "left": "issue:3043", "right": "issue:6688", "accept": false, - "reason": "Separate feature requests; no evidence they are the same change." + "reason": "Both are feature additions, but for unrelated models and code paths." }, { "left": "issue:6234", "right": "issue:8588", "accept": false, - "reason": "Both involve example/script gaps, but they target different scripts and model families." + "reason": "Different training/conversion script problems in different model families." }, { "left": "issue:6374", "right": "issue:9258", "accept": false, - "reason": "LoRA weight-loading question vs Flux LoRA training speed comparison; different issues." + "reason": "Chinese-language LoRA weight-loading question vs training speed investigation; unrelated." }, { "left": "issue:8880", "right": "issue:9895", "accept": false, - "reason": "Performance regression vs device-placement error; different concrete failures." + "reason": "Performance regression vs device mismatch error; different symptoms and fixes." }, { "left": "issue:1160", "right": "issue:1166", "accept": false, - "reason": "Both mention hangs, but they are different tests/training flows and different failure points." + "reason": "Both hang, but in different tests and contexts, so not the same underlying bug." }, { "left": "issue:3706", "right": "issue:4609", "accept": false, - "reason": "New model addition vs providing a specific existing model; not duplicates." + "reason": "Different model/pipeline addition requests." }, { "left": "issue:6141", "right": "issue:9958", "accept": false, - "reason": "Different feature requests in different components (ControlNet params vs VAE tiling)." + "reason": "ControlNet parameter support and VAE temporal decoder tiling are unrelated feature requests." }, { "left": "issue:10226", "right": "issue:10994", "accept": false, - "reason": "Both are Flux LoRA training problems, but the titles point to different failure modes and are not clearly the same bug." + "reason": "Both are Flux LoRA training issues, but one is a tensor-size mismatch and the other is a batch-size crash; not clearly the same bug." }, { "left": "issue:1677", "right": "issue:785", "accept": false, - "reason": "Incorrect pipeline call type vs PNDM scheduler performance cost; unrelated." + "reason": "Wrong data type invocation versus scheduler performance cost; unrelated." }, { "left": "issue:5881", "right": "issue:7142", "accept": false, - "reason": "Separate output-quality complaints for different models and setups." + "reason": "Both compare output quality, but they concern different pipelines/models and different mismatch causes." }, { "left": "issue:6278", "right": "issue:6518", "accept": false, - "reason": "Same script family, but one is a multi-GPU failure and the other is a NoneType division crash; not clearly the same bug." + "reason": "Same script family, but one is a multi-GPU runtime issue and the other is a NoneType arithmetic crash; different root causes." }, { "left": "issue:10010", "right": "issue:10675", "accept": false, - "reason": "Both mention Flux, but one is a missing model file and the other is a scheduler config mismatch." + "reason": "A missing remote model file error is not the same as a Flux scheduler configuration mismatch." }, { "left": "issue:4436", "right": "issue:5300", "accept": false, - "reason": "Conversion workflow vs network connectivity complaint; not the same issue." + "reason": "Conversion workflow versus internet-connection behavior; unrelated topics." }, { "left": "issue:6234", "right": "issue:8918", "accept": false, - "reason": "Different training scripts and different broken behavior." - }, - { - "left": "issue:8485", - "right": "issue:8834", - "accept": false, - "reason": "Support request for SD3 LoRA vs question about SD3 ControlNet training release; different feature asks." + "reason": "Different training scripts and different failure reports; not one duplicate bug." } ] }, @@ -53578,6 +53577,7 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10054", + "issue:10231", "issue:10855", "issue:10920", "issue:11014", @@ -53597,7 +53597,6 @@ "issue:2830", "issue:2984", "issue:3073", - "issue:3963", "issue:3974", "issue:3984", "issue:4178", @@ -53606,10 +53605,7 @@ "issue:4605", "issue:4611", "issue:4647", - "issue:4665", "issue:5108", - "issue:5320", - "issue:5341", "issue:5557", "issue:5731", "issue:6572", @@ -53619,14 +53615,19 @@ "issue:7513", "issue:7686", "issue:8465", + "issue:8485", + "issue:8624", "issue:8790", + "issue:8834", "issue:9096", "issue:9350", "issue:9799", "issue:9895" ], "soft_pairs": [ + "issue:8485|issue:8834", "issue:1974|issue:2083", + "issue:10231|issue:8624", "issue:11014|issue:7513", "issue:9350|issue:9799", "issue:4213|issue:5108", @@ -53647,182 +53648,180 @@ "issue:1617|issue:5557", "issue:11178|issue:8465", "issue:3984|issue:7686", - "issue:4178|issue:4194", - "issue:3963|issue:4665", - "issue:5320|issue:5341" + "issue:4178|issue:4194" ], - "prepared_review_unit_hash": "3dc36db2b2a8afb91881cb1a35852acf89b181675b1ad4628b3a1608adf1d255", + "prepared_review_unit_hash": "d124da7a5250eb9215e89584d0b73d1ef09b434b645c5fcebd3d34ba20ec2dae", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33804, - "estimated_input_tokens": 8451, - "estimated_eval_tokens": 17158 + "serialized_chars": 33772, + "estimated_input_tokens": 8443, + "estimated_eval_tokens": 17142 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is not a duplicate set; it\u2019s a grab-bag of unrelated issues spanning schedulers, loading, training, memory, and API/version mismatches. All soft pairs are superficial similarity matches and should be rejected.", - "confidence": 0.95, - "canonical_issue_reason": "No clear canonical issue: the listed items are unrelated, so none should represent the cluster.", + "summary": "The set is a broad grab-bag of unrelated Diffusers issues. None of the soft pairs is close enough to treat as the same underlying bug or change; most are only superficially similar by subsystem or symptom.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 12483 is the strongest standalone representative: it is a concrete, well-scoped runtime compatibility bug with a clear version interaction and actionable reproduction context.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a safe global representative because the items do not share the same underlying bug or change.", + "best_issue_reason": "Issue 12483 is the best global issue candidate because it is specific, technically grounded, and more diagnostic than the more symptom-level reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:8485", + "right": "issue:8834", + "accept": false, + "reason": "SD3 LoRA support request vs SD3 ControlNet training-code request; related domain but different feature asks." + }, { "left": "issue:1974", "right": "issue:2083", "accept": false, - "reason": "Both mention schedulers and generators, but one is about Euler schedulers with a list of generators and the other about DDIMScheduler with eta>0; different code paths and failure modes." + "reason": "Both involve schedulers and generators, but one is about Euler schedulers with generator lists and the other is DDIM eta handling; distinct bugs." + }, + { + "left": "issue:10231", + "right": "issue:8624", + "accept": false, + "reason": "Ubuntu tutorial ImportError vs SD3 fused layer norm undefined-symbol error; different failure modes and code paths." }, { "left": "issue:11014", "right": "issue:7513", "accept": false, - "reason": "Both are memory-related, but one is a general CUDA OOM report and the other is an SDXL LoRA DreamBooth training issue on a specific GPU; not the same bug." + "reason": "Generic CUDA OOM complaint vs specific SDXL LoRA DreamBooth on 3060 memory issue; not the same defect." }, { "left": "issue:9350", "right": "issue:9799", "accept": false, - "reason": "Different problems: one is a deprecation warning about txt_ids shape, the other is a tensor size mismatch runtime error." + "reason": "txt_ids deprecation warning vs tensor shape mismatch runtime error; unrelated symptoms and likely unrelated fixes." }, { "left": "issue:4213", "right": "issue:5108", "accept": false, - "reason": "Both involve XL pipelines, but one is about from_single_file without accelerate and the other is a T2I Multiadapter XL exception; unrelated failures." + "reason": "from_single_file loading issue without accelerate vs T2I Multiadapter XL exceptions; different pipelines and error classes." }, { "left": "issue:1350", "right": "issue:3073", "accept": false, - "reason": "These are separate feature requests for different model families (DiffusionDet vs Consistency models), not the same change." + "reason": "DiffusionDet project discussion vs consistency-models feature request; not a duplicate bug/change." }, { "left": "issue:12436", "right": "issue:12483", "accept": false, - "reason": "Related ecosystem/version symptoms, but one is a CLIPTextModel kwarg error and the other is a broader diffusers/transformers dependency conflict; not the same concrete bug." + "reason": "CLIPTextModel unexpected kwarg is a downstream symptom, while 12483 is a broader transformers/diffusers version conflict; similar area but not same issue." }, { "left": "issue:2830", "right": "issue:4611", "accept": false, - "reason": "Different topics: a question about using a modelscope video model versus an issue about VideoComposer being merged with Modelscope." + "reason": "Modelscope video usage question vs VideoComposer merge topic; different content and no shared bug." }, { "left": "issue:5731", "right": "issue:7456", "accept": false, - "reason": "One is about a missing trainable logvar scalar in AutoencoderKL, the other is a missing raise statement; no shared bug." + "reason": "AutoencoderKL missing logvar scalar vs missing raise statement; unrelated implementation issues." }, { "left": "issue:6572", "right": "issue:8790", "accept": false, - "reason": "Both are SDXL-related, but one is about a ControlNetXS inpaint pipeline and the other about SDXL TensorRT + IP Adapter; different problems." + "reason": "ControlNetXS-SDXL-Inpaint pipeline request vs SDXL TensorRT/IP-Adapter request; different features, not one bug." }, { "left": "issue:2676", "right": "issue:3974", "accept": false, - "reason": "A scheduler-support question and a safetensors loading issue in an inpaint pipeline are unrelated." + "reason": "Scheduler-name question vs safetensors support bug in a specific pipeline; no same underlying defect." }, { "left": "issue:4647", "right": "issue:6958", "accept": false, - "reason": "Both mention missing mid_block args, but they concern different models and different parameters (num_layers vs use_linear_projection)." + "reason": "UNet2DModel mid_block num_layers plumbing bug vs unet_motion_model use_linear_projection omission; different model classes and code paths." }, { "left": "issue:1539", "right": "issue:1976", "accept": false, - "reason": "Conversion script type handling and a tokenizer/config file lookup error are unrelated." + "reason": "Conversion script type bug vs tokenizer config file path error; unrelated." }, { "left": "issue:2161", "right": "issue:4605", "accept": false, - "reason": "One is a LoRA training question on an inpainting model; the other is a pipeline error with long prompt weighting and inpaint models; not the same underlying defect." + "reason": "LoRA training on inpainting model request vs long-prompt-weighting pipeline error; different pipelines and failure conditions." }, { "left": "issue:10920", "right": "issue:9096", "accept": false, - "reason": "Different numerical instability reports in different schedulers/models: DDIMInverseScheduler infs versus Flux half-precision NaNs." + "reason": "DDIMInverseScheduler inf values from alpha/sample setting vs Flux half-precision NaNs; both numeric instability but different mechanisms." }, { "left": "issue:10054", "right": "issue:7183", "accept": false, - "reason": "Both are scheduler issues, but one is a torch.linspace argument bug in EulerDiscreteScheduler and the other is a stale _class_name display bug." + "reason": "EulerDiscreteScheduler linspace parameter bug vs scheduler class-name display bug; different scheduler internals." }, { "left": "issue:10855", "right": "issue:12761", "accept": false, - "reason": "Unrelated: gradient accumulation/optimizer stepping versus a backward gradient-count mismatch." + "reason": "Gradient accumulation step-frequency question vs templated attention backward gradient-count mismatch; unrelated training vs autograd bug." }, { "left": "issue:2984", "right": "issue:3073", "accept": false, - "reason": "Both are model requests, but they target different architectures and capabilities; not one underlying change." + "reason": "Latent video diffusion model discussion vs consistency models request; both research-related but not the same change." }, { "left": "issue:11540", "right": "issue:9895", "accept": false, - "reason": "Both involve FluxPipeline, but one is about VAE slicing noise/support and the other is a device placement regression; distinct bugs." + "reason": "VAE slicing causing noise / unsupported in Flux img2img vs Flux device-mismatch regression; both Flux but different defects." }, { "left": "issue:1617", "right": "issue:5557", "accept": false, - "reason": "A missing-arguments CLI error for dreambooth training is unrelated to a use_reentrant keyword-argument error." + "reason": "Missing required args in training script vs unexpected use_reentrant keyword; different CLI/API errors." }, { "left": "issue:11178", "right": "issue:8465", "accept": false, - "reason": "Both are from_single_file loading problems, but they involve different models and different root causes (missing config.json vs loading MistoLine ControlNet)." + "reason": "from_single_file loading MistoLine ControlNet issue vs StableCascadeUNet.from_single_file config.json error; similar loader surface, different models and missing files." }, { "left": "issue:3984", "right": "issue:7686", "accept": false, - "reason": "Different validation errors in different UNet contexts: class_labels requirement versus cross_attention_dim requirement." + "reason": "class_labels requirement in one config path vs cross_attention_dim requirement in UNet2DModel; both validation errors but not the same bug." }, { "left": "issue:4178", "right": "issue:4194", "accept": false, - "reason": "LoRA unloading with XFormers and VAE loading with from_single_file are different subsystems and failures." - }, - { - "left": "issue:3963", - "right": "issue:4665", - "accept": false, - "reason": "Both mention Windows, but one is about auto-downloading weights in Gradio and the other about safetensors vs alternative bin files." - }, - { - "left": "issue:5320", - "right": "issue:5341", - "accept": false, - "reason": "A PEFT API enhancement and an example-script update are related by library area, but not the same bug or change." + "reason": "LoRA unload with XFormers vs from_single_file VAE loading problem; different pipeline components and failure modes." } ] }, @@ -53841,8 +53840,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -53867,22 +53866,22 @@ "issue:3467", "issue:3537", "issue:3699", + "issue:3963", "issue:4407", "issue:4418", "issue:4648", + "issue:4665", "issue:4702", "issue:4830", - "issue:4935", "issue:5022", "issue:5118", - "issue:5476", + "issue:5320", + "issue:5341", "issue:5877", "issue:5968", "issue:6560", "issue:7003", - "issue:7043", "issue:7191", - "issue:7350", "issue:7418", "issue:743", "issue:8234", @@ -53896,6 +53895,8 @@ "issue:9958" ], "soft_pairs": [ + "issue:3963|issue:4665", + "issue:5320|issue:5341", "issue:11448|issue:8637", "issue:1524|issue:5877", "issue:3699|issue:4418", @@ -53917,182 +53918,180 @@ "issue:2052|issue:3467", "issue:2437|issue:3043", "issue:3537|issue:4702", - "issue:4407|issue:4648", - "issue:4935|issue:7350", - "issue:5476|issue:7043" + "issue:4407|issue:4648" ], - "prepared_review_unit_hash": "c1ed9e669b5323c7b90af3400433b1690ae0f11bca8d7fcb4218b067a75f5e06", + "prepared_review_unit_hash": "7f18cfb823bbe1d0af7bd2c86b4a5f71b46756509c370aa84501121c7485c64f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34112, - "estimated_input_tokens": 8528, - "estimated_eval_tokens": 17312 + "serialized_chars": 34048, + "estimated_input_tokens": 8512, + "estimated_eval_tokens": 17280 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated issues that only share broad surface themes via keyword similarity (imports, loading, shape mismatches, training scripts). I did not find any soft pair that looks like the same underlying bug/change.", - "confidence": 0.86, - "canonical_issue_reason": "issue:9171 is the strongest representative: a concrete, regression-style from_single_file loading bug with clear reproduction context and substantial discussion.", + "summary": "This set is not a true duplicate cluster: the soft pairs span unrelated bugs, features, and docs/examples across different pipelines and components. All soft edges should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": "issue:10016 is the clearest self-contained bug report, but it is only a representative anchor; the rest of the items are heterogeneous and not duplicates of one another.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9171 is the most actionable and broadly representative issue in the set; it describes a specific pipeline loading regression rather than a vague feature request.", + "best_issue_reason": "issue:10016 is the strongest standalone issue to surface as a representative of the set because it is concrete and actionable, even though it is not a duplicate hub.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3963", + "right": "issue:4665", + "accept": false, + "reason": "Different failure modes: auto-download on Windows vs safetensors/alternate bin-file handling." + }, + { + "left": "issue:5320", + "right": "issue:5341", + "accept": false, + "reason": "One is a PEFT API enhancement request; the other is updating example scripts, not the same bug." + }, { "left": "issue:11448", "right": "issue:8637", "accept": false, - "reason": "Both are import/load failures, but they affect different pipelines and different missing symbols; not the same bug." + "reason": "Completely different issues: missing model path/name vs import error for a pipeline class." }, { "left": "issue:1524", "right": "issue:5877", "accept": false, - "reason": "Textual inversion training quality and subprocess failures in training are different problems." + "reason": "Training quality problem vs subprocess/checkpoint-resume failure; different root causes and scripts." }, { "left": "issue:3699", "right": "issue:4418", "accept": false, - "reason": "CUDA-on-OSX assertion and SDXL inpaint denoising_start behavior are unrelated." + "reason": "OSX/CUDA platform error vs SDXL inpaint denoising_start behavior; unrelated code paths." }, { "left": "issue:3140", "right": "issue:7191", "accept": false, - "reason": "Both mention import issues, but one is a missing pipeline import and the other is an unspecified strange import problem with different context." + "reason": "Importing a specific pipeline class is not the same as an unspecified import problem." }, { "left": "issue:3112", "right": "issue:5118", "accept": false, - "reason": "Both are ControlNet-related, but one is multi-controlnet inpainting and the other is reference_only usage; different code paths." + "reason": "Multi-ControlNet inpainting and reference_only ControlNet usage are different features." }, { "left": "issue:10016", "right": "issue:7003", "accept": false, - "reason": "Both involve tensor shape mismatches on loading, but the model types and failure causes differ." + "reason": "Both are shape mismatches, but in different pipelines and tensors (SD3.5M pos_embed vs inpaint conv weights)." }, { "left": "issue:10210", "right": "issue:8809", "accept": false, - "reason": "Img2img output similarity to input and feedforward sequence-length inconsistency are not the same bug." + "reason": "Img2img output similarity is unrelated to feedforward output inconsistency with sequence length." }, { "left": "issue:4830", "right": "issue:8234", "accept": false, - "reason": "Both are tensor-shape/typing errors in different subsystems; not the same root cause." + "reason": "Different tensor errors: MultiControlNet size mismatch vs ip_adapter_scale dict multiplication." }, { "left": "issue:5968", "right": "issue:9101", "accept": false, - "reason": "ONNX conversion for video pipelines and an SD3 training-script request are unrelated." + "reason": "ONNX conversion request for video pipelines vs a training script request for SD3; not the same bug/change." }, { "left": "issue:7418", "right": "issue:8837", "accept": false, - "reason": "A bug in a specific community pipeline and a SlicedAttnProcessor bug are different code paths." + "reason": "Community upscale script bug and SlicedAttnProcessor bug are separate code paths." }, { "left": "issue:6560", "right": "issue:9171", "accept": false, - "reason": "Both involve from_single_file loading regressions, but one is a LoRA shape-size issue and the other is SD1.5 v_prediction handling; not the same bug." + "reason": "LoRA shape-size regression and SD1.5 v_prediction loading regression are different loader issues." }, { "left": "issue:1060", "right": "issue:743", "accept": false, - "reason": "Both concern conversion scripts, but one is a safety-checker load problem and the other is a general ckpt conversion failure." + "reason": "Safety-checker loading failure in one converter vs ckpt-to-diffusers conversion failure in another." }, { "left": "issue:5022", "right": "issue:9147", "accept": false, - "reason": "Different import/autopipeline failures in different model families; only the symptom type overlaps." + "reason": "Import-time tokenizer error vs Flux autopipeline error; unrelated subsystems." }, { "left": "issue:10511", "right": "issue:10883", "accept": false, - "reason": "Scheduler redesign and Marigold update are unrelated feature requests." + "reason": "Scheduler redesign for SD3/Flux is unrelated to a Marigold feature update." }, { "left": "issue:10886", "right": "issue:10987", "accept": false, - "reason": "Both are diffusion model topics, but one is a feature idea and the other is a broader research request; not the same change." + "reason": "Both are diffusion-model feature ideas, but they target different domains and are not the same change." }, { "left": "issue:10887", "right": "issue:11700", "accept": false, - "reason": "Importing a trainer class and adding Pruna integration are unrelated." + "reason": "Missing StableDiffusionXLTrainer import and Pruna integration are unrelated." }, { "left": "issue:11410", "right": "issue:9933", "accept": false, - "reason": "An attention_processor unpacking error and missing width/height params in SD3 img2img are different bugs." + "reason": "Attention processor unpack error vs missing width/height parameters in SD3 img2img; different bugs." }, { "left": "issue:12367", "right": "issue:9958", "accept": false, - "reason": "Qwen-Image LoRA support and AutoencoderKLTemporalDecoder tiling are unrelated feature requests." + "reason": "Qwen-Image Lora support request and AutoencoderKLTemporalDecoder tiling support are unrelated features." }, { "left": "issue:2052", "right": "issue:3467", "accept": false, - "reason": "Both are resume-related, but one is checkpoint resume in textual inversion and the other is ControlNet step resumption; different implementations." + "reason": "Both involve resuming training, but one is textual inversion checkpoint resume and the other is ControlNet step accounting." }, { "left": "issue:2437", "right": "issue:3043", "accept": false, - "reason": "T2I-Adapter pipeline and ODISE are different pipeline additions, not duplicates." + "reason": "Different pipeline additions: T2I-Adapter vs ODISE." }, { "left": "issue:3537", "right": "issue:4702", "accept": false, - "reason": "ControlNet conversion tooling and a VQ-VAE training example are unrelated." + "reason": "ControlNet conversion script and VQ-VAE training example are unrelated requests." }, { "left": "issue:4407", "right": "issue:4648", "accept": false, - "reason": "Local hub_download storage question and missing config.json loading error are different problems." - }, - { - "left": "issue:4935", - "right": "issue:7350", - "accept": false, - "reason": "Both touch LoRA/ControlNet, but one is a training utility question and the other is ControlNet initialization from a LoRA UNet; not the same bug." - }, - { - "left": "issue:5476", - "right": "issue:7043", - "accept": false, - "reason": "DreamBooth xformers compatibility and custom captions are unrelated training concerns." + "reason": "Local hub_download storage question vs missing config.json in a ControlNet directory; different file-loading problems." } ] }, @@ -54111,8 +54110,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -54122,10 +54121,8 @@ "issue:10893", "issue:11221", "issue:11468", - "issue:11555", "issue:11574", "issue:12113", - "issue:12754", "issue:1290", "issue:1530", "issue:1919", @@ -54134,16 +54131,16 @@ "issue:2552", "issue:2661", "issue:2662", - "issue:3085", "issue:3162", "issue:3647", "issue:3788", "issue:4321", - "issue:4529", "issue:4830", + "issue:4935", "issue:4975", "issue:5037", "issue:5233", + "issue:5476", "issue:556", "issue:5737", "issue:5791", @@ -54154,6 +54151,8 @@ "issue:6234", "issue:6287", "issue:6666", + "issue:7043", + "issue:7350", "issue:7405", "issue:7460", "issue:8647", @@ -54166,6 +54165,8 @@ "issue:9894" ], "soft_pairs": [ + "issue:4935|issue:7350", + "issue:5476|issue:7043", "issue:6234|issue:6287", "issue:1082|issue:3647", "issue:10003|issue:9479", @@ -54188,195 +54189,202 @@ "issue:5037|issue:6666", "issue:2091|issue:2661", "issue:2091|issue:2662", - "issue:5791|issue:5897", - "issue:3085|issue:4529", - "issue:11555|issue:12754" + "issue:5791|issue:5897" ], - "prepared_review_unit_hash": "9672b655a77f386e59cf9d653266298069d6f18dc0b8fcd0cdc53ff75f0a0ae6", + "prepared_review_unit_hash": "83a26a3b0ee1ca2fa0fc5d0628cf99eebe887d7ba2bd70e9e16ec4c1ce7c5bf5", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34337, - "estimated_input_tokens": 8585, - "estimated_eval_tokens": 17426 + "serialized_chars": 34417, + "estimated_input_tokens": 8605, + "estimated_eval_tokens": 17466 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft pair looks like the same underlying bug/change; these are mostly unrelated issues across different components, feature requests, and example docs. All soft edges should be rejected.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue fits this cluster: the items span distinct bugs, feature requests, and example/documentation problems rather than one duplicate theme.", + "summary": "Most items are unrelated bug reports or feature requests. Only two soft pairs look like true duplicates: the learned_classifier_free_sampling_embeddings errors, and the checkpoint write-error reports.", + "confidence": 0.88, + "canonical_issue_reason": "Issue 3162 is the clearest canonical target for the learned_classifier_free_sampling_embeddings bug because it states the exact runtime error directly and matches the paired report closely.", "canonical_pr_reason": null, - "best_issue_reason": "issue:12754 is the strongest standalone issue candidate because it is detailed, open, and has the most discussion/references, but it does not represent the rest of the cluster.", + "best_issue_reason": "Issue 9171 is the strongest standalone report overall: it is a regression, names the exact entrypoint and model condition, and is more actionable than the vaguer reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4935", + "right": "issue:7350", + "accept": false, + "reason": "Both are about LoRA/ControlNet, but one is a usage question and the other is a different initialization failure." + }, + { + "left": "issue:5476", + "right": "issue:7043", + "accept": false, + "reason": "Different topics: xformers loading compatibility vs DreamBooth custom captions." + }, { "left": "issue:6234", "right": "issue:6287", "accept": false, - "reason": "One is a broken training example script, the other is an overbroad dependency list in setup.py; different fixes and failure modes." + "reason": "One is a training-script issue; the other is about packaging/dependencies. Not the same bug." }, { "left": "issue:1082", "right": "issue:3647", "accept": false, - "reason": "Dreambooth CUDA matmul crash vs training with DeepFloyd; same area, but not the same concrete bug." + "reason": "Both mention DreamBooth, but the CUDA matmul crash and DeepFloyd training are different problems." }, { "left": "issue:10003", "right": "issue:9479", "accept": false, - "reason": "Scheduler bug at 1000 inference steps vs inpaint pipeline ignoring the mask; unrelated pipeline behavior." + "reason": "Different pipelines and symptoms: scheduler bug at 1000 steps vs inpaint mask leakage." }, { "left": "issue:2185", "right": "issue:2552", "accept": false, - "reason": "Missing config files for a training script vs a question about an inpaint tutorial; not the same issue." + "reason": "Unrelated training-config omission vs an inpaint tutorial question." }, { "left": "issue:1530", "right": "issue:3162", - "accept": false, - "reason": "Both mention learned classifier-free sampling embeddings, but one is a missing field in VQDiffusion and the other is an expected-but-not-passed error in a different context." + "accept": true, + "reason": "Same underlying bug: missing/expected learned_classifier_free_sampling_embeddings causing the same runtime error." }, { "left": "issue:6218", "right": "issue:9181", "accept": false, - "reason": "ControlNet/img2img batched seeds issue vs a strange noise dot artifact; different symptoms and code paths." + "reason": "Both are ControlNet-related, but batched seeds and strange noise artifacts are different issues." }, { "left": "issue:4830", "right": "issue:5233", "accept": false, - "reason": "MultiControlNet tensor shape mismatch vs wrong step value in callback parameters; not the same bug." + "reason": "Tensor-size mismatch in MultiControlNet is unrelated to callback step numbering." }, { "left": "issue:11574", "right": "issue:12113", "accept": false, - "reason": "KeyError for a VAE weight key vs WanImageToVideo channel mismatch; different models and failures." + "reason": "Different failures: a VAE state/key error vs a 3D convolution channel mismatch in WanImageToVideoPipeline." }, { "left": "issue:5737", "right": "issue:6016", "accept": false, - "reason": "Two separate converter failures for SDXL/torch.Size handling; both conversion-related but distinct error conditions." + "reason": "Both involve the converter script, but the SDXL conversion failure and the torch.Size() issue are distinct symptoms." }, { "left": "issue:3788", "right": "issue:7405", "accept": false, - "reason": "New pipeline proposal vs a new model proposal; feature requests, not duplicates." + "reason": "Two separate model/pipeline requests: UniControl vs DynamiCrafter." }, { "left": "issue:10382", "right": "issue:9370", "accept": false, - "reason": "VAE forward return-type mismatch causing AttributeError vs a missing pooled_projection argument; different API breakages." + "reason": "Different code paths: VAE encoder return-type mismatch vs missing pooled_projection argument." }, { "left": "issue:11221", "right": "issue:11468", "accept": false, - "reason": "Autoencoder training non-convergence vs a float8 quantization question; unrelated topics." + "reason": "Non-convergence in autoencoder training is unrelated to float8_e4m3_tensor quantization discussion." }, { "left": "issue:4975", "right": "issue:8831", "accept": false, - "reason": "Performance problem in LoRA loading vs unsupported LoRA layers continuation; different aspects of LoRA support." + "reason": "Both mention LoRA, but inefficiency in loading and unsupported-layer continuation are not the same bug." }, { "left": "issue:10893", "right": "issue:9894", "accept": false, - "reason": "ControlNet integration request for CogVideoX vs DC-AE integration; separate feature requests." + "reason": "Separate feature requests for different components: CogVideoX ControlNet vs DC-AE integration." }, { "left": "issue:5989", "right": "issue:7460", "accept": false, - "reason": "safetensors bug vs AdaLayerNorm bug; different subsystems." + "reason": "A safetensors bug is unrelated to an AdaLayerNorm bug." }, { "left": "issue:1919", "right": "issue:3647", "accept": false, - "reason": "Dreambooth example dependency packaging vs training with DeepFloyd; unrelated." + "reason": "Dependency requirements for a DreamBooth example are unrelated to DeepFloyd DreamBooth training." }, { "left": "issue:1290", "right": "issue:556", "accept": false, - "reason": "Community project announcement vs seamless tiling feature request; no shared bug/change." + "reason": "DreamArtist community project vs seamless texture tiling are different topics." }, { "left": "issue:4321", "right": "issue:8970", "accept": false, - "reason": "Black images on DirectML ONNX FP16 vs black images on MPS for Kolors; same symptom, but different backends/models and likely different causes." + "reason": "Both are black-image reports, but they affect different backends/models and do not look like one concrete bug." }, { "left": "issue:8647", "right": "issue:9171", "accept": false, - "reason": "ControlNet img2img NoneType iteration error vs single-file v_prediction regression; different code paths." + "reason": "Different problems: a NoneType iteration error in ControlNet Img2Img vs an SD 1.5 v_prediction regression." }, { "left": "issue:5037", "right": "issue:6666", "accept": false, - "reason": "Second validation image argument error vs black validation images; both validation-related but not the same failure." + "reason": "Validation-image argument handling is not the same as black validation images." }, { "left": "issue:2091", "right": "issue:2661", "accept": false, - "reason": "Resume-from-checkpoint multi-GPU training issue vs checkpoint write error; different problems." + "reason": "Resuming multi-GPU training from checkpoint is a different issue from checkpoint save/write failures." }, { "left": "issue:2091", "right": "issue:2662", "accept": false, - "reason": "Resume training pipeline issue vs generic checkpoint write error; not the same root cause." + "reason": "Same as above: checkpoint resuming and checkpoint write errors are distinct." }, { "left": "issue:5791", "right": "issue:5897", "accept": false, - "reason": "Dreambooth prior preservation/SNR gamma question vs LoRA device mismatch error; different training bugs." - }, - { - "left": "issue:3085", - "right": "issue:4529", - "accept": false, - "reason": "Cache folder argument removal vs unexpected network_alpha in LoRA loader; unrelated API changes." + "reason": "DreamBooth prior preservation/SNR gamma discussion is unrelated to the LoRA device-mismatch error." }, { - "left": "issue:11555", - "right": "issue:12754", - "accept": false, - "reason": "device_map=\"auto\" support request vs from_pipe default dtype conversion bug; both pipeline-loading topics but different concrete issues." + "left": "issue:2661", + "right": "issue:2662", + "accept": true, + "reason": "Same underlying checkpoint write-error bug; one is just a more specific report of the same failure." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "Grounded in the packet overall, and the soft-edge calls stay conservative. The two accepted pairs are plausible duplicates and look mergeable as single bugfixes; the rejected pairs are mostly supported by distinct titles/symptoms rather than overclaimed matches." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -54400,34 +54408,34 @@ "issue:11182", "issue:11298", "issue:1140", + "issue:11555", "issue:11561", "issue:11653", "issue:11749", "issue:12359", "issue:12695", + "issue:12754", "issue:1290", "issue:13279", "issue:13350", "issue:13416", "issue:1434", - "issue:1934", "issue:1962", "issue:2891", + "issue:3085", "issue:3431", "issue:3455", "issue:3891", "issue:4149", - "issue:4159", "issue:4363", "issue:4436", + "issue:4529", "issue:4782", "issue:4949", "issue:5485", - "issue:5576", "issue:5607", "issue:5889", "issue:6132", - "issue:6493", "issue:6688", "issue:7080", "issue:7219", @@ -54442,6 +54450,8 @@ "issue:9744" ], "soft_pairs": [ + "issue:3085|issue:4529", + "issue:11555|issue:12754", "issue:9227|issue:9464", "issue:11093|issue:8850", "issue:7390|issue:8900", @@ -54464,198 +54474,193 @@ "issue:4149|issue:5607", "issue:11182|issue:13279", "issue:3891|issue:6688", - "issue:3431|issue:4782", - "issue:5576|issue:6493", - "issue:1934|issue:4159" + "issue:3431|issue:4782" ], - "prepared_review_unit_hash": "50fde4ab86591c9f3626d656459011a77142e60a2b0cc18a55663b31f72af9be", + "prepared_review_unit_hash": "d53ab0228cfe0533529eafe2c0103c5af9760dd78a95dc9a5e8dececc0621451", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 32986, - "estimated_input_tokens": 8247, - "estimated_eval_tokens": 16750 + "serialized_chars": 33078, + "estimated_input_tokens": 8270, + "estimated_eval_tokens": 16796 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a set of unrelated model/pipeline feature requests and bug reports that only share broad domain vocabulary. The only clear duplicate pair is the FreeNoise request vs. the FreeNoise pipeline issue; everything else is too generic or targets different concrete problems.", + "summary": "Mostly a false-positive similarity cluster: the items share common Diffusers/model-support vocabulary, but they describe different models, pipelines, schedulers, or bug classes rather than the same underlying issue. No pull requests are present.", "confidence": 0.92, - "canonical_issue_reason": "Issue #5576 is the clearest original request: it explicitly asks to implement FreeNoise as a new diffusion scheduler. It is more descriptive than the later '[Pipeline] FreeNoise' title and best anchors the duplicate pair.", + "canonical_issue_reason": "Issue 11010 is the most substantive and broadly scoped item in the set, with the richest discussion and a concrete feature-support framing, so it is the best anchor if one issue must represent this heterogeneous cluster.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #5576 is the best representative because it states the feature and scope plainly, making it the most useful canonical issue for FreeNoise.", + "best_issue_reason": "11010 is the clearest representative of the cluster\u2019s general \u201cadd/support a new model/pipeline\u201d theme, while most others are narrower one-off bugs or unrelated support requests.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3085", + "right": "issue:4529", + "accept": false, + "reason": "Both involve loading/API changes, but one is about a removed cache-folder argument and the other is a LoRA `network_alpha` keyword error; different code paths." + }, + { + "left": "issue:11555", + "right": "issue:12754", + "accept": false, + "reason": "Both touch pipeline loading behavior, but `device_map=\"auto\"` support and `from_pipe` dtype conversion are separate problems." + }, { "left": "issue:9227", "right": "issue:9464", "accept": false, - "reason": "Both are about train_dreambooth_lora_flux, but one is a general run problem and the other is an IndexError tied to prodigy + train_text_encoder; different concrete bugs." + "reason": "Both mention FLUX DreamBooth LoRA training, but one is a convergence/example issue and the other is an `IndexError` with Prodigy/text-encoder training." }, { "left": "issue:11093", "right": "issue:8850", "accept": false, - "reason": "TorchAO example breakage and community pipelines mirror breakage are unrelated failures in different areas." + "reason": "Both say something is broken, but they refer to different artifacts: TorchAO example vs community pipelines mirror." }, { "left": "issue:7390", "right": "issue:8900", "accept": false, - "reason": "Both mention from_single_file loading, but one is an SDXL ControlNet inpaint error and the other is a refiner loading question; not the same bug." + "reason": "Both are `from_single_file` loading errors, but they target different SDXL pipelines/models and likely different failure modes." }, { "left": "issue:4363", "right": "issue:4436", "accept": false, - "reason": "Both involve SDXL/VAE safetensors handling, but one asks how to load a specific base model and the other is about converting CIVITAI safetensors; different tasks." + "reason": "Both are about SDXL safetensors loading/conversion, but one is a usage question and the other is a conversion workflow issue." }, { "left": "issue:5889", "right": "issue:9744", "accept": false, - "reason": "Stable video diffusion and Mochi video model support are different model integrations." + "reason": "Both are video-model related, but Stable Video Diffusion support and Mochi support are distinct model additions." }, { "left": "issue:3891", "right": "issue:7219", "accept": false, - "reason": "Two separate feature requests for different pipelines (Tortoise TTS vs SUPIR upscaler)." + "reason": "Both are requests to add a pipeline, but Tortoise TTS and SUPIR are unrelated models and code paths." }, { "left": "issue:1434", "right": "issue:2891", "accept": false, - "reason": "OOM during upscaling is a runtime memory issue; missing embed support in docs is a capability/documentation mismatch." + "reason": "Same pipeline family, but one is an OOM problem and the other is a docs/API mismatch about embeds." }, { "left": "issue:11298", "right": "issue:11653", "accept": false, - "reason": "Both are LoRA-related, but one is a PEFT key error during hotswapping and the other is support for '.alpha' keys in HiDream LoRAs." + "reason": "Both involve LoRA metadata/key handling, but `peft` key errors during hotswapping and `.alpha` keys in HiDream LoRAs are different bugs." }, { "left": "issue:1140", "right": "issue:1290", "accept": false, - "reason": "Hypernetworks and DreamArtist are different community methods/features, not the same underlying request." + "reason": "DreamArtist may be related to hypernetworks, but these are not clearly the same request or bug; one is generic hypernetwork support and the other is a specific community model proposal." }, { "left": "issue:1962", "right": "issue:4949", "accept": false, - "reason": "Video generation with Stable Diffusion and ProPainter in/outpainting for video are different feature areas." + "reason": "Both concern video generation, but stable-diffusion video generation and ProPainter in/outpainting are separate features." }, { "left": "issue:11010", "right": "issue:12359", "accept": false, - "reason": "One is a request to support the Chroma model family; the other is a documentation bug about guidance_scale." + "reason": "Both mention Chroma, but one is a feature/support request and the other is a documentation bug about `guidance_scale`." }, { "left": "issue:11749", "right": "issue:9090", "accept": false, - "reason": "Both are scheduler timestep bugs, but they affect different schedulers and symptoms; not the same concrete defect." + "reason": "Different schedulers and different timestep bugs; these are not the same underlying defect." }, { "left": "issue:9164", "right": "issue:9464", "accept": false, - "reason": "Both touch train_dreambooth_lora_flux, but one is a convergence problem and the other is an IndexError with specific options." + "reason": "Both are FLUX training issues, but one is non-convergence in an example and the other is a runtime `IndexError`." }, { "left": "issue:5485", "right": "issue:7080", "accept": false, - "reason": "Slow ControlNet batch inference and wandb output divergence during training are different issues." + "reason": "Both mention ControlNet/batch workflows, but one is slow batch inference and the other is a wandb output divergence issue." }, { "left": "issue:10100", "right": "issue:11561", "accept": false, - "reason": "ConsisID support and FluxFillPipeline IP-Adapter loading are unrelated model-support requests." + "reason": "Different feature requests for different pipelines: ConsisID support vs loading IP-Adapter in FluxFill." }, { "left": "issue:10772", "right": "issue:12695", "accept": false, - "reason": "Sana ControlNet support and Hunyuan v1.5 support are separate model requests." + "reason": "Different model support requests: Sana ControlNet vs Hunyuan v1.5." }, { "left": "issue:6132", "right": "issue:6688", "accept": false, - "reason": "LOADABLE_CLASSES proposal and MotionCntrl support are unrelated." + "reason": "One is about PyTorch loadable classes, the other is a request for MotionCntrl support; no shared underlying bug/change." }, { "left": "issue:13350", "right": "issue:13416", "accept": false, - "reason": "Both mention FLUX.2, but one is about batch inference usage and the other about negative prompts; different questions." + "reason": "Both are usage questions for new models, but they concern different models and different prompt/batch behavior." }, { "left": "issue:3455", "right": "issue:7670", "accept": false, - "reason": "Use pretrained instruct pix2pix during training and how to use IDDPM are unrelated." + "reason": "Unrelated topics: instruct-pix2pix training vs how to use IDDPM." }, { "left": "issue:4149", "right": "issue:5607", "accept": false, - "reason": "Both are tensor-shape runtime errors, but they occur in different pipelines and with different failure modes." + "reason": "Both are tensor mismatch errors, but the affected pipelines and failure contexts are different." }, { "left": "issue:11182", "right": "issue:13279", "accept": false, - "reason": "Unnecessary BF16 variant download and AutoencoderRAE loading error with older transformers are different bugs." + "reason": "Different loading issues: unnecessary BF16 download vs an `AutoencoderRAE`/older-transformers compatibility error." }, { "left": "issue:3891", "right": "issue:6688", "accept": false, - "reason": "Tortoise TTS pipeline request and MotionCntrl support are separate feature requests." + "reason": "Different community feature requests for unrelated models: Tortoise TTS vs MotionCntrl." }, { "left": "issue:3431", "right": "issue:4782", "accept": false, - "reason": "Poor ControlNet inpainting results and mask==0 pixels being modified are different inpainting problems." - }, - { - "left": "issue:5576", - "right": "issue:6493", - "accept": true, - "reason": "Both refer to FreeNoise; one is the original request to implement it as a diffusion scheduler and the other is the pipeline issue for the same feature." - }, - { - "left": "issue:1934", - "right": "issue:4159", - "accept": false, - "reason": "CPU offload memory growth and DiffEdit enormous VRAM requirements are distinct memory-related issues." + "reason": "Both are inpainting-related, but poor ControlNet results and mask-overwrite behavior are distinct bugs." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "Grounded and conservative overall. The summary stays within the packet evidence, and the single accepted pair (FreeNoise issue vs pipeline issue) is well supported. The remaining soft-edge rejections are cautious and do not overclaim duplicates." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -54684,16 +54689,16 @@ "issue:13062", "issue:13429", "issue:1790", + "issue:1934", "issue:2299", - "issue:2319", "issue:2324", "issue:2361", "issue:2737", "issue:2755", "issue:2891", "issue:3570", + "issue:4159", "issue:4236", - "issue:4524", "issue:4876", "issue:4913", "issue:5034", @@ -54702,16 +54707,16 @@ "issue:5197", "issue:5258", "issue:5463", + "issue:5576", "issue:5656", - "issue:590", "issue:5981", "issue:6117", "issue:6132", "issue:6357", + "issue:6493", "issue:713", "issue:7168", "issue:7312", - "issue:8410", "issue:8511", "issue:8526", "issue:8593", @@ -54721,6 +54726,8 @@ "issue:9460" ], "soft_pairs": [ + "issue:5576|issue:6493", + "issue:1934|issue:4159", "issue:10639|issue:12752", "issue:2737|issue:7312", "issue:2755|issue:3570", @@ -54742,182 +54749,180 @@ "issue:1790|issue:2324", "issue:8526|issue:8593", "issue:9012|issue:9460", - "issue:12666|issue:13062", - "issue:4524|issue:8410", - "issue:2319|issue:590" + "issue:12666|issue:13062" ], - "prepared_review_unit_hash": "f800bfa61959149e6f2ffe3f7382ba7c34f0236d37aee98c918c120cc8b4f574", + "prepared_review_unit_hash": "13093cc1bbefd21464212334889e8ca69b93df4918557b9afc124edc8ebaf740", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 32766, - "estimated_input_tokens": 8192, - "estimated_eval_tokens": 16640 + "serialized_chars": 32744, + "estimated_input_tokens": 8186, + "estimated_eval_tokens": 16628 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are a heterogeneous grab bag of unrelated model, pipeline, training, and scheduler topics. None of the suggested soft pairs look like the same underlying bug or change, so all should be rejected.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue fits: the items span unrelated subsystems and distinct feature requests/bugs rather than one duplicate cluster.", - "canonical_pr_reason": "No pull requests are present in this cluster.", - "best_issue_reason": "No best issue candidate stands out because the set is not a coherent issue family; any choice would be arbitrary.", - "best_pr_reason": null, + "summary": "No soft pair looks like the same concrete bug/change; they mostly share broad keywords or subsystem names but target different pipelines, schedulers, training scripts, or model integrations. I\u2019d use issue:12574 as the best canonical issue because it is a specific, well-scoped bug report with a clear failure mode.", + "confidence": 0.84, + "canonical_issue_reason": "issue:12574 is the strongest canonical issue candidate: it describes a concrete runtime failure, names the affected pipeline and underlying VACE control-argument handling, and is more actionable than the broader feature/request items.", + "canonical_pr_reason": "No PRs are present in the set.", + "best_issue_reason": "issue:12574 is the best overall issue representative because it is specific, technically grounded, and easy to map to a concrete fix path.", + "best_pr_reason": "No PRs are present, so there is no best PR candidate.", "soft_edge_verdicts": [ + { + "left": "issue:5576", + "right": "issue:6493", + "accept": false, + "reason": "Both mention FreeNoise, but one is a scheduler request and the other is a pipeline entry; they are related features, not the same underlying change." + }, + { + "left": "issue:1934", + "right": "issue:4159", + "accept": false, + "reason": "Both are VRAM-related complaints, but they concern different pipelines and memory footprints, so they are not the same bug." + }, { "left": "issue:10639", "right": "issue:12752", "accept": false, - "reason": "Different model requests: Lumina-Image 2.0 support vs Z-Image img2img/inpainting pipelines." + "reason": "These are separate model-support requests for different pipelines; same broad theme, not one concrete issue." }, { "left": "issue:2737", "right": "issue:7312", "accept": false, - "reason": "Different scheduler/finetuning topics with no shared code-path or bug." + "reason": "Different functionality: scheduler CLIP guidance support versus AlignProp/direct reward finetuning." }, { "left": "issue:2755", "right": "issue:3570", "accept": false, - "reason": "Both involve ckpt loading, but one is float16 support and the other is disabling downloads; not the same bug." + "reason": "Both involve checkpoint loading, but one is float16 support and the other is disabling downloads; different fixes." }, { "left": "issue:4876", "right": "issue:5656", "accept": false, - "reason": "A LoRA layer std argument request is unrelated to adding a LoRA loader pipeline." + "reason": "LoRA layer std as an argument is unrelated to adding a LoRA loader." }, { "left": "issue:5118", "right": "issue:6132", "accept": false, - "reason": "ControlNet reference_only usage question is unrelated to LOADABLE_CLASSES support." + "reason": "ControlNet reference_only usage and loadable classes are unrelated code paths." }, { "left": "issue:10817", "right": "issue:8511", "accept": false, - "reason": "Different SD3 problems: missing control nets in auto_pipeline vs text encoder unloading in training scripts." + "reason": "Both are SD3-related, but one is auto_pipeline missing control nets and the other is Dreambooth/LoRA encoder unloading." }, { "left": "issue:10514", "right": "issue:12574", "accept": false, - "reason": "Unrelated model-specific failures: Sana resolution binning vs WanVideo/VACE control-argument handling." + "reason": "Different models and failure modes: Sana resolution binning/sample size versus WanVideoToVideo control-argument handling." }, { "left": "issue:2891", "right": "issue:7168", "accept": false, - "reason": "Different functionality: upscale embeds support vs IP-Adapter image embed shape handling." + "reason": "One is about pipeline embed support; the other is about IP-Adapter image embeddings tensor shape." }, { "left": "issue:11973", "right": "issue:9161", "accept": false, - "reason": "Different training-script bugs: required instance_prompt argument vs distributed bugs." + "reason": "Separate training-script bugs in different Flux scripts; not the same failure." }, { "left": "issue:2299", "right": "issue:5034", "accept": false, - "reason": "Different issues around checkpoints; one is fine-tuned checkpoint support, the other is progress-bar display." + "reason": "Fast-stable-diffusion checkpoint support and progress-bar display are unrelated." }, { "left": "issue:2361", "right": "issue:5197", "accept": false, - "reason": "Checkpoint-save crash in finetuning vs multi-GPU slowdown are distinct training issues." + "reason": "Checkpoint-saving crashes and multi-GPU slowdown are different training issues." }, { "left": "issue:4913", "right": "issue:8790", "accept": false, - "reason": "Different topics: a ControlNet model artifact/request vs SDXL TensorRT + IP-Adapter integration." + "reason": "OpenPose ControlNet support and SDXL TensorRT/IP-Adapter integration are different features." }, { "left": "issue:5981", "right": "issue:6117", "accept": false, - "reason": "Both are scheduler bugs, but they concern different schedulers and different methods." + "reason": "Two distinct scheduler bugs affecting different algorithms and methods." }, { "left": "issue:1156", "right": "issue:713", "accept": false, - "reason": "Generic training failures from different eras and with different symptoms; not the same bug." + "reason": "Both are vague training problems, but there is no evidence they describe the same root cause." }, { "left": "issue:5463", "right": "issue:6357", "accept": false, - "reason": "Local variable crash vs API/local result differences are different failure modes and code paths." + "reason": "An UnboundLocalError in one path is not the same as mismatched API vs local results." }, { "left": "issue:5118", "right": "issue:5168", "accept": false, - "reason": "ControlNet reference_only usage question is not the same as ControlNet-XS support." + "reason": "Reference_only usage and ControlNet-XS support are separate ControlNet topics." }, { "left": "issue:4236", "right": "issue:5258", "accept": false, - "reason": "SDXL refiner text-encoder registration and ControlNet inpaint with refiner are related only broadly to SDXL, not the same bug." + "reason": "SDXL refiner text-encoder registration and ControlNet inpaint with refiner are different pipeline integration problems." }, { "left": "issue:12358", "right": "issue:13429", "accept": false, - "reason": "Different model pipeline requests: Lumina-DiMOO vs JoyAI-Image-Edit." + "reason": "Both request new pipeline support, but for different models and different changes." }, { "left": "issue:1790", "right": "issue:2324", "accept": false, - "reason": "General training question vs Dreambooth instance-data bug; no duplicate evidence." + "reason": "A general training question is not the same as Dreambooth not reading instance data." }, { "left": "issue:8526", "right": "issue:8593", "accept": false, - "reason": "Different runtime/backend problems: SD3 MPS attention scaling vs SD3 CPU bfloat16 noise." + "reason": "Different SD3 issues on different platforms: MPS scale handling versus CPU bfloat16 noise." }, { "left": "issue:9012", "right": "issue:9460", "accept": false, - "reason": "Different ControlNet problems: loading SD3ControlNetModel weights vs allowing a list of controlnets in the pipeline." + "reason": "Loading SD3ControlNet weights and assigning multiple controlnets to the pipeline are separate problems." }, { "left": "issue:12666", "right": "issue:13062", "accept": false, - "reason": "Kandinsky pipeline missing in a release vs unsupported onnxruntime-windowsml backend; unrelated." - }, - { - "left": "issue:4524", - "right": "issue:8410", - "accept": false, - "reason": "AnimateDiff and TextDiffuser are different pipeline feature requests." - }, - { - "left": "issue:2319", - "right": "issue:590", - "accept": false, - "reason": "A typo in a loader method is unrelated to a Karras scheduler formula typo." + "reason": "Kandinsky video pipeline availability is unrelated to onnxruntime-windowsml support policy." } ] }, @@ -54948,9 +54953,7 @@ "issue:10952", "issue:1099", "issue:11223", - "issue:11412", "issue:11435", - "issue:11448", "issue:11721", "issue:12078", "issue:12221", @@ -54963,6 +54966,7 @@ "issue:2213", "issue:2294", "issue:2310", + "issue:2319", "issue:2646", "issue:3094", "issue:3113", @@ -54971,17 +54975,18 @@ "issue:3350", "issue:4042", "issue:4232", - "issue:4241", + "issue:4524", "issue:4619", "issue:4815", "issue:4846", "issue:4907", "issue:5545", - "issue:5656", + "issue:590", "issue:6095", "issue:6688", "issue:6866", "issue:7138", + "issue:8410", "issue:8604", "issue:8832", "issue:8939", @@ -54991,6 +54996,8 @@ "issue:9926" ], "soft_pairs": [ + "issue:4524|issue:8410", + "issue:2319|issue:590", "issue:2646|issue:6095", "issue:1305|issue:3113", "issue:6866|issue:9511", @@ -55012,24 +55019,22 @@ "issue:12458|issue:12738", "issue:2213|issue:3094", "issue:2310|issue:4042", - "issue:319|issue:321", - "issue:4241|issue:5656", - "issue:11412|issue:11448" + "issue:319|issue:321" ], - "prepared_review_unit_hash": "e5718a7dd8f900eeab6e711acfe3c8822e3e2448b706034ccccf0ede858fde94", + "prepared_review_unit_hash": "bb4200927a2fcf77743490fa6c99da774e1f9235caa3746edf739d6598b7e969", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33794, + "serialized_chars": 33793, "estimated_input_tokens": 8449, "estimated_eval_tokens": 17154 }, @@ -55038,156 +55043,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly noisy: the items share generic language around support, crashes, imports, and batching, but they do not describe the same underlying bug or change. I did not find any soft pair that should be merged as duplicates.", - "confidence": 0.31, - "canonical_issue_reason": "issue:12458 is the most active and broadly framed issue in the set, so it is the best representative anchor among otherwise unrelated reports.", + "summary": "This set is mostly a loose collection of unrelated feature requests and bug reports; the soft pairs are only superficially similar by title, not true duplicates. I would not merge any of the candidate pairs.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:12458 has the strongest discussion/inbound activity and a general feature/bug framing, making it the most suitable single issue to represent this cluster.", + "best_issue_reason": "Issue #8410 is the broadest umbrella request (native Diffusers pipeline support), but the items are too heterogeneous for any issue to serve as a true canonical duplicate target.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4524", + "right": "issue:8410", + "accept": false, + "reason": "Both are feature requests about adding pipeline support, but for different pipelines (AnimateDiff vs TextDiffuser); not the same underlying change." + }, + { + "left": "issue:2319", + "right": "issue:590", + "accept": false, + "reason": "Both mention typos, but they are in different functions and different code paths, so they are not duplicates." + }, { "left": "issue:2646", "right": "issue:6095", "accept": false, - "reason": "Both mention IndexError, but they affect different pipelines and scheduler paths, so they are not the same bug." + "reason": "Both are IndexError reports, but one is tiled VAE and the other is SVD scheduling; distinct bugs." }, { "left": "issue:1305", "right": "issue:3113", "accept": false, - "reason": "These are different research feature requests for different methods; similar paper-style wording is not enough to merge them." + "reason": "Both are image-editing/control-related papers, but they describe different methods and desired features." }, { "left": "issue:6866", "right": "issue:9511", "accept": false, - "reason": "One requests IP-Adapter support and the other requests multi-controlnet batching; same pipeline family, but different changes." + "reason": "Same pipeline family, but one requests IP-Adapter support and the other batching for multi-controlnet; different problems." }, { "left": "issue:4619", "right": "issue:7138", "accept": false, - "reason": "A training dtype/bias mismatch is unrelated to an SDXL UNet structure mismatch." + "reason": "SDXL-related, but one is a training dtype mismatch and the other is a model structure mismatch; not the same bug." }, { "left": "issue:10147", "right": "issue:12221", "accept": false, - "reason": "Both are support requests, but for different adapters/control methods and different model workflows." + "reason": "Different requested integrations/features: MV-Adapter multi-view generation vs DiffSynth ControlNet support." }, { "left": "issue:11223", "right": "issue:13079", "accept": false, - "reason": "Both involve OOM reports, but on different models, inputs, and code paths." + "reason": "Both are OOM reports, but on different models and scenarios; no shared concrete code-path is evident." }, { "left": "issue:3350", "right": "issue:5545", "accept": false, - "reason": "These concern different pipelines and different failure modes, not the same underlying issue." + "reason": "Different pipelines/models and different failure modes; too broad to be duplicates." }, { "left": "issue:8604", "right": "issue:8989", "accept": false, - "reason": "One is a T5 fp16 corruption bug; the other is a Stable Audio sequential offloading bug." + "reason": "Different bugs in different systems: T5 fp16 corruption vs Stable Audio offloading behavior." }, { "left": "issue:10100", "right": "issue:6688", "accept": false, - "reason": "Both ask for model support, but ConsisID and MotionCntrl are distinct integrations." + "reason": "Both request support for different model families (ConsisID vs MotionCntrl); not the same change." }, { "left": "issue:10231", "right": "issue:11435", "accept": false, - "reason": "A tutorial ImportError on Ubuntu is not the same as a missing HiDreamImagePipeline export/import issue." + "reason": "Both are import-related, but they concern different missing symbols and likely different dependency issues." }, { "left": "issue:10848", "right": "issue:10952", "accept": false, - "reason": "These are different adapter-loading failures with different APIs and error causes." + "reason": "Both involve adapters, but one is a bad function signature and the other is adapter-name handling during LoRA load." }, { "left": "issue:4232", "right": "issue:4846", "accept": false, - "reason": "An instruct-pix2pix/controlnet training request and a dreambooth training issue are unrelated." + "reason": "One is instruct-pix2pix/controlnet training/inference; the other is a DreamBooth training result issue." }, { "left": "issue:8939", "right": "issue:9850", "accept": false, - "reason": "Tokenizer loading failure and gradient-checkpointing behavior are unrelated bugs." + "reason": "Completely different topics: tokenizer loading for SD3 vs gradient checkpointing support." }, { "left": "issue:11721", "right": "issue:8832", "accept": false, - "reason": "Both are crashes, but one is on model load and the other occurs during inference execution." + "reason": "Both are crashes, but on different pipelines and likely different root causes." }, { "left": "issue:1099", "right": "issue:2294", "accept": false, - "reason": "Dreambooth class-name setup and a video inpaint import error are not the same problem." + "reason": "DreamBooth class-name issue vs missing StableDiffusionVideoInpaintPipeline import; unrelated." }, { "left": "issue:4815", "right": "issue:4907", "accept": false, - "reason": "Latents-only input support and an SDXL inpainting warning are different inpainting concerns." + "reason": "Latents-only input support vs an SDXL inpainting warning; not the same underlying bug or feature." }, { "left": "issue:12078", "right": "issue:12995", "accept": false, - "reason": "These are unrelated pipeline/example questions about different models and behaviors." + "reason": "Different products and scenarios: Flux finetuning validation input vs Cosmos-Predict2.5 video behavior." }, { "left": "issue:12367", "right": "issue:9926", "accept": false, - "reason": "A LoRA support request and a gguf quantization/speed request are different changes." + "reason": "One asks for a specific LoRA feature, the other for gguf quantization/speed support; unrelated." }, { "left": "issue:12458", "right": "issue:12738", "accept": false, - "reason": "Both involve batching, but they target different pipelines and likely different implementation fixes." + "reason": "Both mention batching, but they are different pipelines/models and different failure contexts." }, { "left": "issue:2213", "right": "issue:3094", "accept": false, - "reason": "TPU distributed hang and CUDA OOM during finetuning are different training issues." + "reason": "Same training script name, but one is TPU distributed hang and the other is CUDA OOM; distinct issues." }, { "left": "issue:2310", "right": "issue:4042", "accept": false, - "reason": "Both mention inpainting, but one is about LDM/Inpainting cleanup and the other about combining reference-only with controls." + "reason": "Different compositional inpainting requests; no shared concrete bug or code path." }, { "left": "issue:319", "right": "issue:321", "accept": false, - "reason": "These are distinct diffusion methods, not duplicate reports." - }, - { - "left": "issue:4241", - "right": "issue:5656", - "accept": false, - "reason": "A LoRA fine-tuning support request and a LoRA loader feature request are related but not the same change." - }, - { - "left": "issue:11412", - "right": "issue:11448", - "accept": false, - "reason": "Both are missing-model/path failures, but for different repositories and pipeline names." + "reason": "Separate research-model support requests (VQ-Diffusion vs Blended Diffusion), not duplicates." } ] }, @@ -55215,10 +55220,10 @@ "issue:10550", "issue:10817", "issue:10928", - "issue:11222", "issue:11275", "issue:11279", - "issue:11283", + "issue:11412", + "issue:11448", "issue:11555", "issue:11709", "issue:12567", @@ -55239,6 +55244,7 @@ "issue:3467", "issue:4061", "issue:4073", + "issue:4241", "issue:4269", "issue:4286", "issue:4580", @@ -55249,18 +55255,19 @@ "issue:5314", "issue:5573", "issue:5594", + "issue:5656", "issue:5927", "issue:6688", "issue:7789", "issue:7842", - "issue:852", "issue:8649", "issue:9026", "issue:9138", - "issue:940", "issue:9637" ], "soft_pairs": [ + "issue:4241|issue:5656", + "issue:11412|issue:11448", "issue:11275|issue:9026", "issue:3467|issue:5573", "issue:4580|issue:5002", @@ -55283,24 +55290,22 @@ "issue:4750|issue:5927", "issue:11555|issue:9138", "issue:1934|issue:3209", - "issue:1896|issue:2471", - "issue:11222|issue:11283", - "issue:852|issue:940" + "issue:1896|issue:2471" ], - "prepared_review_unit_hash": "7aefc0d1a3ce264910f57c323c40fd229c6ca6ebd9f2435a30f923fd040edff5", + "prepared_review_unit_hash": "955edf8cf0c54dc2a0310e293e081a07b9d95d253039ad0cd6c5a401938bd314", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34019, + "serialized_chars": 34020, "estimated_input_tokens": 8505, "estimated_eval_tokens": 17266 }, @@ -55309,169 +55314,166 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft pairs look like true duplicates; the set spans unrelated bugs, feature requests, and backend-specific failures. I would not merge any of the proposed pairs.", - "confidence": 0.96, - "canonical_issue_reason": "Issue #11279 is the broadest open thread in the set and the closest thing to a representative issue, but it is still not a duplicate of the soft-pair candidates.", + "summary": "Mostly unrelated issues across different features/subsystems. The only plausible duplicate pair is the SDXL LoRA-loading reports (#4061 and #4286); the rest are too different in concrete error, API, or code path to merge as duplicates.", + "confidence": 0.76, + "canonical_issue_reason": "#4286 is the clearest concrete report for the SDXL LoRA-loading failure (`load_lora_weights` not working in SDXL), while #4061 is a shorter, more generic symptom report.", "canonical_pr_reason": null, - "best_issue_reason": "#11279 is the best global issue candidate because it is open and comparatively broad; the rest are much more specific and do not share a single underlying bug.", + "best_issue_reason": "#4286 is the best representative issue because it names the specific failing API and model family, making it the most actionable canonical duplicate target.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4241", + "right": "issue:5656", + "accept": false, + "reason": "Both mention LoRA, but one is about fine-tuning unidiffusers and the other is about adding a LoRA loader to community pipelines; different change requests." + }, + { + "left": "issue:11412", + "right": "issue:11448", + "accept": false, + "reason": "Both are model-not-found/404 reports, but for different repositories/models and different loader paths." + }, { "left": "issue:11275", "right": "issue:9026", "accept": false, - "reason": "Different failures: VAE NoneType addition vs an unexpected kwarg in checkpoint dispatch." + "reason": "A VAE addition bug and an unexpected keyword argument in checkpoint dispatch are unrelated code paths." }, { "left": "issue:3467", "right": "issue:5573", "accept": false, - "reason": "Both mention ControlNet, but one is about wrong resume step and the other about long weight initialization." + "reason": "Both are ControlNet training issues, but one is resume-step accounting and the other is initialization timing; not the same bug." }, { "left": "issue:4580", "right": "issue:5002", "accept": false, - "reason": "Scheduler/Karras-sigmas related, but these are different scheduler classes and different concrete bugs." + "reason": "Both involve Karras sigmas, but one is a UniPC schedule exception and the other is missing scheduler arguments in different classes." }, { "left": "issue:13043", "right": "issue:13176", "accept": false, - "reason": "Different attention-backend/device errors with no shared code-path shown." + "reason": "Different attention failures with different backends and failure modes; not the same underlying bug." }, { "left": "issue:7789", "right": "issue:7842", "accept": false, - "reason": "One is an MPS float32 requirement; the other is a DeepFloyd shape mismatch on non-divisible image sizes." + "reason": "One is an MPS float32 requirement, the other is an image-size tensor mismatch in DeepFloyd stage 2." }, { "left": "issue:10928", "right": "issue:13176", "accept": false, - "reason": "Missing model file error vs an attention runtime error; unrelated failures." + "reason": "A missing checkpoint file error and an attention runtime error are unrelated." }, { "left": "issue:11709", "right": "issue:12567", "accept": false, - "reason": "LoRA key-loading issue vs Ulysses attention noise generation; different subsystems and symptoms." + "reason": "LoRA key-loading failure versus Ulysses attention noise generation; different components and symptoms." }, { "left": "issue:1633", "right": "issue:5594", "accept": false, - "reason": "Karras sigma support in a scheduler wrapper vs SDXL LoRA memory usage; unrelated changes." + "reason": "A scheduler option request and SDXL LoRA memory-saving request are different features, not one bug." }, { "left": "issue:2437", "right": "issue:6688", "accept": false, - "reason": "Separate feature requests for different pipelines, not the same change." + "reason": "Two different pipeline feature requests with no shared concrete implementation bug." }, { "left": "issue:10550", "right": "issue:11279", "accept": false, - "reason": "Quantized LoRA loading problem vs partial model initialization behavior; not the same bug." + "reason": "Quantized LoRA loading versus initializing new layers in partially pretrained models are different loading/training scenarios." }, { "left": "issue:2560", "right": "issue:3209", "accept": false, - "reason": "Determinism bug with generator lists vs NaNs in AutoencoderKL on large images." + "reason": "Different failures: generator-list nondeterminism versus AutoencoderKL NaNs on large images." }, { "left": "issue:1358", "right": "issue:2462", "accept": false, - "reason": "Safetensors support vs IterableDataset training support; unrelated." + "reason": "Safetensors support and IterableDataset finetuning support are unrelated enhancement requests." }, { "left": "issue:4619", "right": "issue:4839", "accept": false, - "reason": "Mixed-precision training type mismatch vs a lora_scale type error." + "reason": "Training dtype mismatch and lora_scale type errors are distinct bugs in different parts of the LoRA path." }, { "left": "issue:3289", "right": "issue:4073", "accept": false, - "reason": "DeepFloyd super-resolution size bug vs a safety-checker astype attribute error." + "reason": "DeepFloyd super-resolution resizing bug versus a safety-checker attribute error; different code paths." }, { "left": "issue:4061", "right": "issue:4286", - "accept": false, - "reason": "Both are SDXL LoRA loading complaints, but the titles do not indicate the same concrete failure." - }, - { - "left": "issue:3140", - "right": "issue:4269", - "accept": false, - "reason": "Import error for the pipeline class vs a TypeError inside check_inputs." + "accept": true, + "reason": "Both report SDXL LoRA loading failing, centered on the same underlying `load_lora_weights`/LoRA-in-SDXL path." }, { "left": "issue:10817", "right": "issue:8649", "accept": false, - "reason": "Different SD3 problems: missing ControlNets vs num_images_per_prompt handling." + "reason": "Missing SD3 control nets and `num_images_per_prompt` behavior are separate SD3 pipeline issues." }, { "left": "issue:10191", "right": "issue:9637", "accept": false, - "reason": "GPU selection/training device issue vs an XFormers attention-mask/bfloat16 kernel failure." + "reason": "GPU device selection in training is unrelated to the xFormers attention-mask bfloat16 kernel bug." }, { "left": "issue:2310", "right": "issue:5314", "accept": false, - "reason": "Both are inpainting-related, but one is about LDM/Lama cleaning and the other is a ControlNet reference pipeline." + "reason": "Both mention inpainting, but one is LDM/LAMA cleaner and the other is a ControlNet reference inpainting pipeline request." }, { "left": "issue:4750", "right": "issue:5927", "accept": false, - "reason": "Overlapping inpaint/Lama wording, but these are distinct composite pipeline requests." + "reason": "Both reference inpaint/lama concepts, but one is a preprocessor request and the other is a broader combined-feature request." }, { "left": "issue:11555", "right": "issue:9138", "accept": false, - "reason": "Device_map support question vs an unrelated int/str comparison TypeError." + "reason": "`device_map=\"auto\"` support and an int-vs-str TypeError are not the same issue." }, { "left": "issue:1934", "right": "issue:3209", "accept": false, - "reason": "Memory offload regression vs NaNs on large images; different code paths." + "reason": "CPU offload memory usage and AutoencoderKL NaNs are unrelated runtime problems." }, { "left": "issue:1896", "right": "issue:2471", "accept": false, - "reason": "Different pipeline-specific size errors, not the same bug." - }, - { - "left": "issue:11222", - "right": "issue:11283", - "accept": false, - "reason": "Feature request for more WAN LoRAs vs a Wan header deserialization error." - }, - { - "left": "issue:852", - "right": "issue:940", - "accept": false, - "reason": "Both are MPS crashes, but they affect different APIs and failure modes." + "reason": "Different pipeline size-handling failures with different repros and code paths." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The only duplicate-like pair supported by the packet is indeed #4061/#4286 (both SDXL LoRA loading failures), and the rest are reasonably rejected as different bugs, features, or code paths. No evidence of overclaiming." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -55491,10 +55493,10 @@ "issue:10111", "issue:10636", "issue:1106", + "issue:11222", + "issue:11283", "issue:11644", "issue:12195", - "issue:12574", - "issue:13477", "issue:1543", "issue:2185", "issue:2310", @@ -55509,7 +55511,6 @@ "issue:460", "issue:4646", "issue:4750", - "issue:4801", "issue:5225", "issue:5395", "issue:5480", @@ -55522,10 +55523,10 @@ "issue:7049", "issue:746", "issue:7695", - "issue:817", "issue:8320", "issue:8432", "issue:8493", + "issue:852", "issue:8604", "issue:8742", "issue:8752", @@ -55533,11 +55534,14 @@ "issue:911", "issue:9297", "issue:9358", + "issue:940", "issue:9464", "issue:9822", "issue:9930" ], "soft_pairs": [ + "issue:11222|issue:11283", + "issue:852|issue:940", "issue:1106|issue:3221", "issue:2437|issue:6243", "issue:460|issue:746", @@ -55560,195 +55564,196 @@ "issue:3073|issue:5480", "issue:4445|issue:5656", "issue:5777|issue:6359", - "issue:1543|issue:746", - "issue:4801|issue:817", - "issue:12574|issue:13477" + "issue:1543|issue:746" ], - "prepared_review_unit_hash": "1c64e866af1b7e0f52adbdc34bee5a0a39e20f8e7afd7b6040894041f399062a", + "prepared_review_unit_hash": "5f83ef21c52ce1f20e695337ac935a7e7609e8a267ba6ec91b9ac529510f54a6", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33873, - "estimated_input_tokens": 8469, - "estimated_eval_tokens": 17194 + "serialized_chars": 33790, + "estimated_input_tokens": 8448, + "estimated_eval_tokens": 17152 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose collection of mostly unrelated issues across different pipelines/features; none of the soft pairs look like true duplicates or the same code change. The closest thematic group is FLUX-related training/inference bugs, but even those are distinct failures.", - "confidence": 0.79, - "canonical_issue_reason": "issue:10111 is the broadest umbrella title and best fits the cluster\u2019s mixed FLUX/pipeline-bug theme, even though the set is not a tight duplicate group.", + "summary": "Most items are unrelated one-off bug reports or feature requests. The only plausible duplicate-like pair is the LaMa/inpaint preprocessing issues; everything else is too semantically different to merge.", + "confidence": 0.78, + "canonical_issue_reason": "issue:4750 is the clearest representative of the LaMa/inpaint preprocessing topic; issue:2310 appears to be the older, vaguer counterpart.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10111 is the most representative and reusable triage anchor because it captures the broad FLUX pipeline bug category without being tied to one narrow repro.", + "best_issue_reason": "issue:4750 is the strongest issue to anchor this cluster because it is the most specific, concrete report among the few near-matches.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11222", + "right": "issue:11283", + "accept": false, + "reason": "Feature request for more WAN LoRAs vs a deserialization error in WAN loading; different problems." + }, + { + "left": "issue:852", + "right": "issue:940", + "accept": false, + "reason": "Both mention MPS crashes, but the scheduler and failure mode are different enough to treat as separate bugs." + }, { "left": "issue:1106", "right": "issue:3221", "accept": false, - "reason": "Different subsystems and failure modes: PNDM extra UNet calls vs load_attn_procs behavior in diffusers 0.16.0." + "reason": "PNDM repeatedly invoking UNet is unrelated to load_attn_procs failing in diffusers 0.16.0." }, { "left": "issue:2437", "right": "issue:6243", "accept": false, - "reason": "T2I-Adapter pipeline request and IP-Adapter Face ID issue are unrelated feature areas, not the same bug." + "reason": "T2I-Adapter pipeline request vs IP-Adapter Face ID request; different features." }, { "left": "issue:460", "right": "issue:746", "accept": false, - "reason": "Both are ONNX-related but one is runtime dependency selection and the other is revision loading for img2img; different problems." + "reason": "Both are ONNX-related, but one is about dependency selection and the other is an offline revision loading failure." }, { "left": "issue:5395", "right": "issue:9297", "accept": false, - "reason": "Both involve LoRAs, but one is PEFT combining civitai LoRAs and the other is linear load-time scaling for multiple LoRAs." + "reason": "Both involve LoRAs, but one is about broken PEFT merging and the other about linear load-time slowdown." }, { "left": "issue:3975", "right": "issue:5989", "accept": false, - "reason": "A typo in an error message and a safetensors.torch bug are not the same underlying issue." + "reason": "Typo in an error message is unrelated to a safetensors.torch bug." }, { "left": "issue:4210", "right": "issue:5225", "accept": false, - "reason": "UNet attention_mask usage question and SDXL VRAM exhaustion are unrelated." + "reason": "Attention_mask usage in UNet is unrelated to SDXL VRAM exhaustion during forward pass." }, { "left": "issue:8432", "right": "issue:8742", "accept": false, - "reason": "ImportError for FromOriginalVAEMixin and removal of LoRAAttnProcessor2_0 are distinct API compatibility issues." + "reason": "ImportError in FromOriginalVAEMixin and removal of LoRAAttnProcessor2_0 are different compatibility issues." }, { "left": "issue:7049", "right": "issue:7695", "accept": false, - "reason": "IP-Adapter unpacking error and ignored ip_adapter_masks in AttnProcessor2_0 are different code paths and symptoms." + "reason": "Both touch IP-Adapter, but one is an unpacking error in the basic example and the other is ignored cross_attention_kwargs." }, { "left": "issue:460", "right": "issue:4646", "accept": false, - "reason": "ONNX CPU/CUDA dependency selection is different from ControlNet-to-ONNX conversion failure." + "reason": "ONNX runtime dependency issue vs ControlNet-to-ONNX conversion failure; not the same bug." }, { "left": "issue:9358", "right": "issue:9464", "accept": false, - "reason": "Both are FLUX training issues, but one is redundant text-encoder reinit and the other is an IndexError with prodigy + train_text_encoder." + "reason": "Both are FLUX training script issues, but one is redundant reinitialization and the other is a list index error." }, { "left": "issue:11644", "right": "issue:8493", "accept": false, - "reason": "LTXConditionPipeline init-from-pretrained failure is unrelated to SD3Transformer2DModel state_dict loading." + "reason": "LTXConditionPipeline init failure and SD3Transformer2DModel state_dict loading are separate model-loading bugs." }, { "left": "issue:10111", "right": "issue:9822", "accept": false, - "reason": "Generic FLUX pipeline bugs and loading SDXL LoRAs into Flux are related only by theme, not the same bug." + "reason": "Broad FLUX pipeline bug report vs specifically loading SDXL LoRAs into Flux; not the same concrete change." }, { "left": "issue:12195", "right": "issue:8320", "accept": false, - "reason": "PyTorch 2.3.1 torch.library.custom_op failure is unrelated to the __name__ attribute error in from_single_file." + "reason": "PyTorch 2.3.1 custom_op failure in AutoencoderKL is unrelated to the from_single_file __name__ attribute error." }, { "left": "issue:2185", "right": "issue:911", "accept": false, - "reason": "Missing config files for naive T2I training and Dreambooth bad config are both config-related but not the same issue." + "reason": "Missing training config files and Dreambooth returning bad config are different configuration issues." }, { "left": "issue:3576", "right": "issue:3643", "accept": false, - "reason": "XFormers/Textual Inversion TypeError and latent decoding INT_MAX overflow are different runtime failures." + "reason": "XFormers/Textual Inversion TypeError is unrelated to latent upsample INT_MAX decoding errors." }, { "left": "issue:10636", "right": "issue:8837", "accept": false, - "reason": "Bug in value_guided_sampling.py and a sliced attention processor bug are different pipeline components." + "reason": "value_guided_sampling.py bug and SlicedAttnProcessor bug are different code paths." }, { "left": "issue:8604", "right": "issue:9930", "accept": false, - "reason": "Corrupted image from T5 float16 conversion is not the same as PAG adaptive scale behavior." + "reason": "T5 float16 corruption and PAG adaptive scale bug are unrelated." }, { "left": "issue:6110", "right": "issue:8752", "accept": false, - "reason": "Offline LoRA loading and offline from_single_file SDXL loading are both offline-related, but they affect different loaders and failures." + "reason": "Both are offline-related, but one concerns loading LoRA weights and the other from_single_file SDXL loading." }, { "left": "issue:2310", "right": "issue:4750", - "accept": false, - "reason": "LDM inpainting/LAMA cleaner and inpaint + LAMA preprocessor are different feature requests/issues." + "accept": true, + "reason": "Both point to the same LaMa/inpaint preprocessing area; 4750 is the clearer concrete report and 2310 appears to describe the same functionality more vaguely." }, { "left": "issue:3073", "right": "issue:5480", "accept": false, - "reason": "Consistency models request and EDM implementation request are separate model-feature requests." + "reason": "Consistency models request and EDM request are separate model-feature requests." }, { "left": "issue:4445", "right": "issue:5656", "accept": false, - "reason": "How to finetune LoRA and adding a LoRA loader to community pipelines are not the same bug/change." + "reason": "General LoRA finetuning question vs adding a LoRA loader to community pipelines; different scope." }, { "left": "issue:5777", "right": "issue:6359", "accept": false, - "reason": "PixArt DreamBooth training error and EulerDiscreteScheduler velocity/add_noise bug are unrelated." + "reason": "PixArt dreambooth training error and EulerDiscreteScheduler velocity/get_noise issue are unrelated." }, { "left": "issue:1543", "right": "issue:746", "accept": false, - "reason": "textual_inversion fp16 revision issue and ONNX img2img revision issue are different loading paths." - }, - { - "left": "issue:4801", - "right": "issue:817", - "accept": false, - "reason": "Mixed precision bug in train_unconditional.py and DreamBooth mixed precision issue are similar category but different examples and code paths." - }, - { - "left": "issue:12574", - "right": "issue:13477", - "accept": false, - "reason": "WanVideoToVideoPipeline control-argument handling and ErnieImage locals() scope bug are unrelated." + "reason": "textual_inversion revision handling and ONNX img2img revision loading are different loader bugs." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative: the only accepted pair is a plausible duplicate-like LaMa/inpaint preprocessing match, while the rest are appropriately rejected as different bugs or feature requests. The canonical issue choice is also defensible based on the clearer title of issue:4750." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -55768,7 +55773,9 @@ "issue:10003", "issue:11225", "issue:11437", + "issue:12574", "issue:1274", + "issue:13477", "issue:1642", "issue:1785", "issue:2306", @@ -55781,6 +55788,7 @@ "issue:4480", "issue:4697", "issue:4789", + "issue:4801", "issue:4839", "issue:4912", "issue:5160", @@ -55795,14 +55803,13 @@ "issue:6357", "issue:6418", "issue:6749", - "issue:6818", "issue:6823", "issue:7041", "issue:709", "issue:730", "issue:7306", "issue:7417", - "issue:7766", + "issue:817", "issue:8509", "issue:852", "issue:8604", @@ -55810,11 +55817,11 @@ "issue:898", "issue:911", "issue:9134", - "issue:9178", - "issue:9479", - "issue:9867" + "issue:9178" ], "soft_pairs": [ + "issue:4801|issue:817", + "issue:12574|issue:13477", "issue:3453|issue:898", "issue:9134|issue:9178", "issue:6109|issue:6196", @@ -55836,182 +55843,180 @@ "issue:730|issue:911", "issue:7306|issue:8604", "issue:2772|issue:6357", - "issue:4314|issue:4789", - "issue:9479|issue:9867", - "issue:6818|issue:7766" + "issue:4314|issue:4789" ], - "prepared_review_unit_hash": "f57983df598675599f35e166614c24cddfef92d54189ed3b6304408084c364cd", + "prepared_review_unit_hash": "37b8d342bf856b99cc7f56f174a706c9abf6199a043e0514dbf6765934d1b7a5", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33333, - "estimated_input_tokens": 8334, - "estimated_eval_tokens": 16924 + "serialized_chars": 33346, + "estimated_input_tokens": 8337, + "estimated_eval_tokens": 16930 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is noisy and mostly splits into several unrelated bug themes; there is no single defensible canonical duplicate for all items. The strongest concrete bug report is the num_inference_steps crash/segfault pair, but even that is not a safe merge target as a duplicate set.", - "confidence": 0.78, - "canonical_issue_reason": "No single item cleanly serves as a canonical representative for the whole cluster because the items span many different pipelines, features, and failure modes.", + "summary": "This cluster is a loose collection of mostly unrelated Diffusers bug reports and feature requests around training, inference, LoRA/ControlNet, and platform-specific failures. None of the soft pairs look like strong same-bug duplicates, so I would keep the issues separated.", + "confidence": 0.91, + "canonical_issue_reason": "issue:10003 is the clearest, concrete runtime bug in a core Stable Diffusion scheduler/inference path, and it is more specific than the broader training/example reports in the set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:7041 is the clearest concrete runtime failure in the set and the most plausible anchor among the num_inference_steps reports, but the cluster overall is still too heterogeneous for a true canonical duplicate.", + "best_issue_reason": "issue:10003 is the best representative issue because it describes a reproducible scheduler failure with a precise trigger, making it the most actionable anchor for the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4801", + "right": "issue:817", + "accept": false, + "reason": "Both mention mixed precision in training examples, but they are different scripts and likely different failure modes; too broad to treat as the same bug." + }, + { + "left": "issue:12574", + "right": "issue:13477", + "accept": false, + "reason": "Unrelated components and symptoms: WanVideo/VACE control-arg handling versus an ErnieImage callback scope issue." + }, { "left": "issue:3453", "right": "issue:898", "accept": false, - "reason": "Both mention BFloat16, but the failing code paths and environments differ; this is a shared symptom, not the same bug." + "reason": "Different backend/hardware failures: CUDA BF16 triu/tril support versus TPU BF16 scalar-type support." }, { "left": "issue:9134", "right": "issue:9178", "accept": false, - "reason": "Both are Flux/LoRA-related, but one is about bad training outputs and the other is a PEFT target-module error." + "reason": "Both are Flux LoRA-related, but one concerns SimpleTuner-trained LoRAs and the other PEFT target-module errors; not the same concrete bug." }, { "left": "issue:6109", "right": "issue:6196", "accept": false, - "reason": "Same ControlNet-XS/depth area, but one is a feature/request-style issue and the other is a concrete load failure." + "reason": "Related topic area, but one is a ControlNet-XS depth feature/usage issue and the other is a loading error for a specific SDXL model." }, { "left": "issue:3585", "right": "issue:852", "accept": false, - "reason": "These are different problems: LoRA synthesis failure versus an MPS device_map keyword crash." + "reason": "Different problems: LoRA synthesis into SD 1.5 versus an MPS device_map keyword crash." }, { "left": "issue:4480", "right": "issue:7417", "accept": false, - "reason": "Both touch Mac/MPS/fp16, but one is a generation crash and the other is a LoRA fuse test failure." + "reason": "Both are Mac-related, but one is an MPS kernel crash during generation and the other is a test failure around LoRA fuse NaNs." }, { "left": "issue:5633", "right": "issue:8837", "accept": false, - "reason": "Only the broad attention/transformer area overlaps; the specific bugs are different." + "reason": "Different code paths and symptoms in temporal transformer versus sliced attention processor logic." }, { "left": "issue:6823", "right": "issue:8509", "accept": false, - "reason": "Both mention ControlNet, but they describe different features and no shared concrete failure." + "reason": "One is a generic inability to use ControlNet; the other is a ControlNet++ feature request." }, { "left": "issue:1274", "right": "issue:1785", "accept": false, - "reason": "Both involve num_images_per_prompt, but the affected pipelines and regression contexts differ." + "reason": "Both mention num_images_per_prompt, but the pipelines and failure contexts differ enough that this is not clearly one bug." }, { "left": "issue:11225", "right": "issue:11437", "accept": false, - "reason": "Both are training/validation issues, but one is OOM and the other is missing outputs, so the code-path problem is not the same." + "reason": "Validation OOM versus missing outputs/checkpoints-images during training are distinct issues, even if both occur in training flows." }, { "left": "issue:3325", "right": "issue:4839", "accept": false, - "reason": "These are unrelated runtime/type errors; the shared theme is only that they occur during model usage." + "reason": "Different failures in different features: DreamBooth CUDA invalid argument versus lora_scale type error." }, { "left": "issue:10003", "right": "issue:7041", "accept": false, - "reason": "Both involve num_inference_steps, but one is a scheduler bug at 1000 steps and the other is a segfault for >1 step; not enough to merge." + "reason": "Both involve inference-step crashes, but the described triggers and symptoms are different enough that they should not be merged without stronger evidence." }, { "left": "issue:1642", "right": "issue:709", "accept": false, - "reason": "A missing scheduler forward pass and a documentation docstring issue are not the same bug." + "reason": "Scheduler implementation bug versus documentation typo; clearly unrelated." }, { "left": "issue:2306", "right": "issue:3630", "accept": false, - "reason": "Both are vague UNet2DModel errors, but the actual failures described are different and too underspecified to treat as duplicates." + "reason": "Generic runtime error versus unknown UNet2DModel error; no evidence of the same underlying defect." }, { "left": "issue:4697", "right": "issue:6749", "accept": false, - "reason": "A specific LoRA-loading regression and an InstantID replication issue are different problems." + "reason": "A specific LoRA load regression and an InstantID reproduction issue are not the same bug." }, { "left": "issue:4912", "right": "issue:5162", "accept": false, - "reason": "Both are conversion-related, but they target different conversions and different user workflows." + "reason": "Both are conversion-related, but one asks for SDXL checkpoint conversion and the other for folder-to-safetensors conversion; different requests." }, { "left": "issue:5160", "right": "issue:5594", "accept": false, - "reason": "DreamBooth/LoRA is only a broad area match; masking and memory usage are separate issues." + "reason": "DreamBooth/LoRA masking feature request versus SDXL LoRA memory-saving request; not the same change." }, { "left": "issue:5285", "right": "issue:5925", "accept": false, - "reason": "Documentation translation and ZipLora feature work are unrelated." + "reason": "Docs translation work and a ZipLora feature request are unrelated." }, { "left": "issue:5300", "right": "issue:6418", "accept": false, - "reason": "Internet connectivity and identical video frames are unrelated bug reports." + "reason": "Internet-connection complaint versus stable video diffusion output collapse; no shared bug surface." }, { "left": "issue:730", "right": "issue:911", "accept": false, - "reason": "Both are DreamBooth-related, but one is local-vs-Colab learning behavior and the other is a bad config regression." + "reason": "Both concern DreamBooth, but one is a training-behavior report and the other is a bad config issue; too different to merge." }, { "left": "issue:7306", "right": "issue:8604", "accept": false, - "reason": "Both mention float16, but they affect different pipelines and have different failure modes." + "reason": "Both mention float16, but they affect different pipelines/components and have different symptoms." }, { "left": "issue:2772", "right": "issue:6357", "accept": false, - "reason": "A local dataset path validation bug and API/local result differences are not the same issue." + "reason": "Local dataset validation in train_controlnet.py and API-vs-local result differences are separate problems." }, { "left": "issue:4314", "right": "issue:4789", "accept": false, - "reason": "Interrupted-download write protection and inability to download a model are different download-path failures." - }, - { - "left": "issue:9479", - "right": "issue:9867", - "accept": false, - "reason": "Both are inpainting mask-leak bugs, but they are reported on different pipelines and cannot be safely merged without stronger evidence." - }, - { - "left": "issue:6818", - "right": "issue:7766", - "accept": false, - "reason": "An unload error and a how-to question about FaceID Plus are not duplicate bugs." + "reason": "Interrupted-download write-protection and inability to download DeepFloyd IF are different download issues." } ] }, @@ -56052,7 +56057,6 @@ "issue:13488", "issue:13506", "issue:1377", - "issue:1572", "issue:2149", "issue:2150", "issue:2311", @@ -56068,23 +56072,26 @@ "issue:4912", "issue:4964", "issue:5091", - "issue:5344", "issue:5576", "issue:6008", - "issue:6427", + "issue:6818", + "issue:7766", "issue:8518", "issue:8588", "issue:8759", "issue:9026", "issue:9166", - "issue:917", "issue:9241", "issue:9258", + "issue:9479", "issue:9602", + "issue:9867", "issue:987", "issue:993" ], "soft_pairs": [ + "issue:9479|issue:9867", + "issue:6818|issue:7766", "issue:10873|issue:11721", "issue:3073|issue:5576", "issue:10147|issue:10896", @@ -56107,198 +56114,193 @@ "issue:1377|issue:305", "issue:13043|issue:13506", "issue:3097|issue:993", - "issue:3769|issue:987", - "issue:5344|issue:6427", - "issue:1572|issue:917" + "issue:3769|issue:987" ], - "prepared_review_unit_hash": "81880cc43a55e13b438a1d65eee1c5a1c108cf98ade18ac09dfccf81b45ca21a", + "prepared_review_unit_hash": "6382d9f62cf7d186a0d18036a312e6707871b4fbbd802cee250ca9a2cc1c131c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34290, - "estimated_input_tokens": 8573, - "estimated_eval_tokens": 17402 + "serialized_chars": 34322, + "estimated_input_tokens": 8581, + "estimated_eval_tokens": 17418 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issue reports with broad topical overlap only. The only plausible duplicate-like pair is the two SanaPipeline load-crash reports; the rest look like different bugs, features, or model-specific requests.", - "confidence": 0.88, - "canonical_issue_reason": "issue:11721 is the clearest, most general statement of the SanaPipeline load failure, and it subsumes the more model-specific report in issue:10873.", + "summary": "This cluster is highly heterogeneous: it mixes unrelated feature requests, conversion scripts, pipeline-specific bugs, and dependency/runtime errors across many different models and code paths. No single issue or PR meaningfully represents the whole set, and none of the soft-edge pairs are close enough to be true duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:11721 is the strongest canonical issue candidate: concise, clearly bug-focused, and broad enough to represent the SanaPipeline load crash without being model-specific.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:9479", + "right": "issue:9867", + "accept": false, + "reason": "Both are inpaint/mask-related symptoms, but they affect different pipelines (StableDiffusion vs Flux) and do not point to the same concrete bug." + }, + { + "left": "issue:6818", + "right": "issue:7766", + "accept": false, + "reason": "One is an unload/runtime error with IP-Adapter; the other is a FaceID usage question. Different problem and code path." + }, { "left": "issue:10873", "right": "issue:11721", - "accept": true, - "reason": "Both describe a SanaPipeline crash during load; the second is a general version of the first and they plausibly point to the same load-path bug." + "accept": false, + "reason": "These Sana issues are not the same: one is model/pipeline-specific breakage, the other is a generic load crash." }, { "left": "issue:3073", "right": "issue:5576", "accept": false, - "reason": "Different requests: adding consistency models versus implementing FreeNoise scheduler." + "reason": "Both are requests, but one asks for consistency models and the other for a FreeNoise scheduler; separate features." }, { "left": "issue:10147", "right": "issue:10896", "accept": false, - "reason": "Different model support requests for unrelated pipelines/use cases." + "reason": "Different support requests for different models/tasks: multi-view generation vs expressive portrait animation." }, { "left": "issue:10569", "right": "issue:10932", "accept": false, - "reason": "Both mention HunyuanVideo, but one is CPU memory usage and the other is torch.compile recompile-limit behavior; different failures." + "reason": "A CPU memory issue for HunyuanVideo is unrelated to the torch.compile recompile_limit problem." }, { "left": "issue:10472", "right": "issue:8759", "accept": false, - "reason": "Unrelated bugs: LoRA lora_bias init error versus SD3 image postprocessing warning." + "reason": "Different failures: LoRA `lora_bias` init error versus an SD3 image processor warning." }, { "left": "issue:2311", "right": "issue:5576", "accept": false, - "reason": "Both are scheduler-related, but they request different schedulers and different changes." + "reason": "Both mention schedulers, but they request different ones (UniPC vs FreeNoise) and are not the same change." }, { "left": "issue:4731", "right": "issue:8518", "accept": false, - "reason": "Both involve torch.compile, but they are different concrete failures on different code paths." + "reason": "Both involve torch.compile, but they are distinct failures in different pipelines and code paths." }, { "left": "issue:9026", "right": "issue:9166", "accept": false, - "reason": "Different subsystems: checkpoint offloading keyword error versus image overlay behavior." + "reason": "One is a changed offloading API keyword error; the other is unexpected overlay behavior. Not the same bug." }, { "left": "issue:12108", "right": "issue:13488", "accept": false, - "reason": "Both touch pipeline/runtime compatibility, but one is flow-matching schedulers and the other is offloading regression; not the same bug." + "reason": "Both touch flow/offloading behavior, but one is a scheduler-parameter pipeline break and the other is a dev-release offloading regression." }, { "left": "issue:10848", "right": "issue:9602", "accept": false, - "reason": "Different adapter-loading problems with different symptoms and APIs." + "reason": "Both are adapter/load-related, but one is a `load_model_dict_into_meta(device)` keyword error and the other is an adapter-name state mismatch." }, { "left": "issue:2444", "right": "issue:2775", "accept": false, - "reason": "Completely different failures: training script exit error versus torch._dynamo optimize misuse." + "reason": "Different runtime failures in different contexts: a training script exit vs a torch._dynamo optimize error." }, { "left": "issue:5091", "right": "issue:9258", "accept": false, - "reason": "Different performance/offload topics; no shared concrete bug." + "reason": "Multi-GPU/offload usage and Windows-vs-Linux performance are unrelated issues." }, { "left": "issue:1255", "right": "issue:305", "accept": false, - "reason": "Both are import errors, but for different modules and missing symbols." + "reason": "Both are import errors, but for different symbols/modules and different dependency problems." }, { "left": "issue:4808", "right": "issue:6008", "accept": false, - "reason": "Different pipeline/test failure versus IP-Adapter+LCM LoRA regression." + "reason": "A SHAP-E integration test failure is unrelated to the IP-Adapter plus face/LCM LoRA regression." }, { "left": "issue:2149", "right": "issue:2150", "accept": false, - "reason": "Related texture-painting topic, but one is a workflow request and the other is a specific img2img model compatibility request." + "reason": "They are related in theme, but one is a workflow request and the other is img2img texture projection support; not the same concrete change." }, { "left": "issue:4000", "right": "issue:4964", "accept": false, - "reason": "Both concern conversion, but one is SD2 ckpt-to-diffusers and the other is SDXL model conversion; different artifacts and formats." + "reason": "Both are conversion requests, but they target different model families and formats (SD2 ckpt vs SDXL)." }, { "left": "issue:10076", "right": "issue:9241", "accept": false, - "reason": "Different conversion directions and model types: HF-to-original format versus XLab LoRA-to-diffusers LoRA." + "reason": "Both mention conversion scripts, but they convert different things: HF-to-original format vs XLab LoRA-to-diffusers LoRA." }, { "left": "issue:11000", "right": "issue:8588", "accept": false, - "reason": "Missing argument in a pipeline call versus a missing conversion script; not the same bug." + "reason": "A missing `shift` argument in WanImageToVideoPipeline is unrelated to a missing SD3 conversion script." }, { "left": "issue:1161", "right": "issue:4912", "accept": false, - "reason": "Both are conversion-related, but Flax-to-PyTorch and custom CivitAI checkpoint-to-diffusers are different tasks." + "reason": "Flax-to-PyTorch conversion and SDXL checkpoint-to-diffusers conversion are different tasks." }, { "left": "issue:1377", "right": "issue:305", "accept": false, - "reason": "Both are import errors, but they target different missing symbols and likely different dependency versions." + "reason": "Different import errors from different packages/symbols; no shared underlying bug is evident." }, { "left": "issue:13043", "right": "issue:13506", "accept": false, - "reason": "Different attention-backend issues: unsupported GPU kernel on Blackwell versus ring-attention failure." + "reason": "These are distinct attention-backend issues: Blackwell kernel support vs ring-attention incompatibility." }, { "left": "issue:3097", "right": "issue:993", "accept": false, - "reason": "Different bugs: textual inversion appears to do nothing versus fp16 datatype mismatch." + "reason": "Textual inversion doing nothing and fp16 dtype mismatch are separate bugs with different causes." }, { "left": "issue:3769", "right": "issue:987", "accept": false, - "reason": "Different output-quality issues in different pipelines; no evidence of a shared root cause." - }, - { - "left": "issue:5344", - "right": "issue:6427", - "accept": false, - "reason": "Different conversion/dependency issues: missing omegaconf versus broken diffusers-to-ckpt conversion." - }, - { - "left": "issue:1572", - "right": "issue:917", - "accept": false, - "reason": "Different pipeline problems: stable_diffusion_mega img2img/inpainting support versus ONNX inpainting error." + "reason": "Different pipeline defects: a quartered panorama output versus a green speech-to-image output." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The SanaPipeline pair (issue:10873 / issue:11721) is plausibly the same load crash, while the other pairwise rejections are appropriately cautious given the titles. No overstatement stands out, and this is not a PR-pair case." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -56327,6 +56329,7 @@ "issue:11555", "issue:1185", "issue:12995", + "issue:1572", "issue:1617", "issue:1708", "issue:1866", @@ -56342,28 +56345,29 @@ "issue:4349", "issue:4676", "issue:5008", + "issue:5344", "issue:5565", "issue:5840", "issue:6028", - "issue:6101", + "issue:6427", "issue:6688", "issue:713", "issue:7223", - "issue:7348", "issue:7380", "issue:7460", "issue:7789", "issue:7793", "issue:7836", "issue:8393", - "issue:8918", + "issue:917", "issue:9170", "issue:9306", - "issue:9424", "issue:9778", "issue:9895" ], "soft_pairs": [ + "issue:5344|issue:6427", + "issue:1572|issue:917", "issue:5565|issue:9778", "issue:1866|issue:3116", "issue:5008|issue:7380", @@ -56385,182 +56389,180 @@ "issue:1617|issue:713", "issue:199|issue:201", "issue:4349|issue:4676", - "issue:5840|issue:8393", - "issue:8918|issue:9424", - "issue:6101|issue:7348" + "issue:5840|issue:8393" ], - "prepared_review_unit_hash": "da911b4331b525f013c62ab3ef4a655330e6a99d8d7ef32fc327e630fe286aa2", + "prepared_review_unit_hash": "b653d926b6f1c3cc482122ad4dcd6cd7b4bb406d2ea9653a60241d49f2541942", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 32940, - "estimated_input_tokens": 8235, - "estimated_eval_tokens": 16726 + "serialized_chars": 32845, + "estimated_input_tokens": 8212, + "estimated_eval_tokens": 16680 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is very heterogeneous: most soft pairs share only broad subsystem words, not the same concrete bug or change, so none of the soft edges should be merged. If a representative issue is needed, the broadest/openest item is the device-map feature request.", - "confidence": 0.36, - "canonical_issue_reason": "issue:11555 is the broadest, most general pipeline-level request in the set and is still open/active; it best serves as a lightweight representative, though the cluster itself is not a true duplicate set.", + "summary": "This cluster is mostly a set of unrelated Diffusers issues that only share broad keywords like pipelines, training, or model support. The proposed soft edges are all superficial similarities rather than the same concrete bug/change.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:11555 has the clearest reusable problem statement (pipeline-wide device_map support) and the widest potential applicability among these unrelated issues.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5344", + "right": "issue:6427", + "accept": false, + "reason": "One is a missing dependency error; the other is a model conversion bug. Different failure modes and code paths." + }, + { + "left": "issue:1572", + "right": "issue:917", + "accept": false, + "reason": "Both mention inpainting, but one is a pipeline support gap and the other is an ONNX runtime error. Not the same bug." + }, { "left": "issue:5565", "right": "issue:9778", "accept": false, - "reason": "Both are tutorial/script breakage reports, but they reference different tutorials and failure contexts; not the same underlying bug." + "reason": "Both are run/tutorial related, but they concern different scripts and likely different setup problems." }, { "left": "issue:1866", "right": "issue:3116", "accept": false, - "reason": "Both mention scheduler behavior in img2img, but they concern different schedulers and different failure modes." + "reason": "Both are scheduler-related img2img bugs, but they involve different schedulers and distinct incorrect behaviors." }, { "left": "issue:5008", "right": "issue:7380", "accept": false, - "reason": "One is a specific TryOnDiffusion implementation issue, the other is a new algorithm implementation request; not duplicates." + "reason": "One is a specific architecture implementation request, the other is a separate distillation method request." }, { "left": "issue:7460", "right": "issue:7836", "accept": false, - "reason": "AdaLayerNorm bug vs CLIP overfitting in a training example are unrelated." + "reason": "AdaLayerNorm bug and CLIP overfitting are unrelated issues in different parts of training/inference." }, { "left": "issue:7789", "right": "issue:7793", "accept": false, - "reason": "Both involve deepfloyd, but one is an mps float32 compatibility issue and the other is LoRA unloading error; different code paths." + "reason": "Different DeepFloyd problems: float32/MPS behavior versus LoRA unload errors." }, { "left": "issue:10100", "right": "issue:7223", "accept": false, - "reason": "Different feature requests for different model families (ConsisID vs Latte)." + "reason": "Both are feature requests for adding models, but they are different models and separate implementations." }, { "left": "issue:1013", "right": "issue:3162", "accept": false, - "reason": "StableDiffusionMegaPipeline regression vs missing learned_classifier_free_sampling_embeddings are different bugs." + "reason": "Both are compatibility-style errors, but one is a pipeline breakage and the other is a missing embedding argument." }, { "left": "issue:1062", "right": "issue:1708", "accept": false, - "reason": "Dreambooth reproducibility and YAML config passing are distinct training UX problems." + "reason": "Dreambooth reproducibility/training quality is unrelated to config-file passing for inference." }, { "left": "issue:3073", "right": "issue:3891", "accept": false, - "reason": "Add consistency models vs add Tortoise TTS are unrelated pipeline requests." + "reason": "Both are add-a-pipeline requests, but for completely different model families and use cases." }, { "left": "issue:3073", "right": "issue:6688", "accept": false, - "reason": "Different requested additions in different model areas; no shared concrete change." + "reason": "Same broad feature-request category, but not the same model or underlying change." }, { "left": "issue:10674", "right": "issue:9895", "accept": false, - "reason": "Both are FluxPipeline-related device problems, but the GGUF failure and cuda/cpu mismatch stem from different scenarios." + "reason": "Both involve FluxPipeline, but one is GGUF support and the other is a device placement regression." }, { "left": "issue:2202", "right": "issue:3974", "accept": false, - "reason": "OpenVINO support request vs safetensors loading bug are unrelated." + "reason": "OpenVINO support request versus a safetensors loading bug in a different pipeline." }, { "left": "issue:6028", "right": "issue:9170", "accept": false, - "reason": "Both are memory-related, but they affect different models and failure conditions." + "reason": "Both mention high memory use, but one is an OOM report and the other asks about GPU requirements for SDXL+ControlNet." }, { "left": "issue:10514", "right": "issue:11555", "accept": false, - "reason": "Sana sample_size/use_resolution_binning limitation is not the same as a general device_map support request." + "reason": "Sample-size/binning limitation and device_map auto support are unrelated pipeline features." }, { "left": "issue:10321", "right": "issue:10520", "accept": false, - "reason": "LTX-Video missing weight files vs Sana cross-device tensor placement error are different bugs." + "reason": "One is a missing model file error; the other is a cross-device tensor placement error." }, { "left": "issue:1185", "right": "issue:3301", "accept": false, - "reason": "Reproducibility complaint vs LoRA training failure are not the same issue." + "reason": "Both are Dreambooth-related, but one is reproducibility and the other is a training script failure." }, { "left": "issue:11033", "right": "issue:12995", "accept": false, - "reason": "Unet from_single_file loading failure is unrelated to the Cosmos-Predict2.5 pipeline behavior question." + "reason": "Single-file SD1.5 loading and Cosmos-Predict2.5 video behavior are unrelated issues." }, { "left": "issue:11041", "right": "issue:9306", "accept": false, - "reason": "Both involve device placement/offloading, but they concern different pipelines and distinct execution paths." + "reason": "Both mention device/offloading behavior, but they are different pipelines and different specific faults." }, { "left": "issue:1617", "right": "issue:713", "accept": false, - "reason": "Generic missing-arguments CLI error vs generic training error are too broad and not clearly the same bug." + "reason": "A missing-arguments CLI error is not the same as a generic training error." }, { "left": "issue:199", "right": "issue:201", "accept": false, - "reason": "One is a dead-code cleanup request, the other is a test request; not duplicates." + "reason": "One is a dead-code cleanup request; the other is a test request." }, { "left": "issue:4349", "right": "issue:4676", "accept": false, - "reason": "Missing text_encoder handling vs TAESD latent brightness are different pipeline defects." + "reason": "Text-encoder requirement bug and dark TAESD latents are different pipeline issues." }, { "left": "issue:5840", "right": "issue:8393", "accept": false, - "reason": "Checkpoint resume/freezing issue vs slow shard loading are different problems." - }, - { - "left": "issue:8918", - "right": "issue:9424", - "accept": false, - "reason": "Both mention broken training notebooks/scripts, but they are different model versions and likely different regressions." - }, - { - "left": "issue:6101", - "right": "issue:7348", - "accept": false, - "reason": "Both are controlnet-related quality issues, but one is a performance regression and the other is a scheduler artifact bug." + "reason": "Checkpoint resume/freezing bug is unrelated to slow shard loading." } ] }, @@ -56588,7 +56590,6 @@ "issue:10940", "issue:11010", "issue:11178", - "issue:11220", "issue:11387", "issue:11864", "issue:11901", @@ -56616,24 +56617,27 @@ "issue:5108", "issue:5197", "issue:5224", - "issue:5317", - "issue:5320", "issue:5595", "issue:6095", + "issue:6101", "issue:6370", "issue:6374", "issue:6393", "issue:6695", + "issue:7348", "issue:801", "issue:852", + "issue:8918", "issue:9001", "issue:9034", + "issue:9424", "issue:945", "issue:962", - "issue:9756", "issue:9822" ], "soft_pairs": [ + "issue:8918|issue:9424", + "issue:6101|issue:7348", "issue:4186|issue:5108", "issue:2185|issue:2712", "issue:11387|issue:11864", @@ -56655,182 +56659,180 @@ "issue:11010|issue:12108", "issue:1363|issue:945", "issue:1274|issue:962", - "issue:11901|issue:12147", - "issue:11220|issue:9756", - "issue:5317|issue:5320" + "issue:11901|issue:12147" ], - "prepared_review_unit_hash": "db5be41429dac7aa5aa173866ba1b6395ac6b3da96d2c6e3982039891b968df6", + "prepared_review_unit_hash": "782fa1d436508aa7521e3956fe728f04ba44564b32cad4145d2acc8fb37285d2", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33616, - "estimated_input_tokens": 8404, - "estimated_eval_tokens": 17064 + "serialized_chars": 33756, + "estimated_input_tokens": 8439, + "estimated_eval_tokens": 17134 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very noisy: the titles span unrelated bugs, feature requests, and docs issues. None of the soft pairs look like the same underlying bug/change, so all should stay separate.", - "confidence": 0.9, - "canonical_issue_reason": "issue:10025 is the strongest single anchor in the set: it is clearly bug-oriented, has the highest engagement (discussion and inbound references), and sits in the recurring Flux/transformer subset that appears most central here.", + "summary": "This is a loose topical cluster of Diffusers issues around model loading, Flux/Wan/Qwen support, schedulers, and training/runtime bugs. The soft pairs are mostly superficial similarity matches rather than true duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:10025 is the best representative issue for triage because it is concrete, well-discussed, and more likely to be the kind of actionable bug a duplicate cluster would center on than the more request-like or niche items.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:8918", + "right": "issue:9424", + "accept": false, + "reason": "Both mention broken training scripts, but one is SD3 Dreambooth and the other is a specific notebook regression; different code paths and fixes." + }, + { + "left": "issue:6101", + "right": "issue:7348", + "accept": false, + "reason": "Both involve ControlNet/inpaint behavior, but one is a performance complaint and the other is a scheduler output bug; not the same defect." + }, { "left": "issue:4186", "right": "issue:5108", "accept": false, - "reason": "Both mention Stable Diffusion XL/multi-adapter style features, but one is an inpaint+refiner behavior issue and the other is a T2I multiadapter exception; different code paths and failure modes." + "reason": "Both are XL pipeline problems, but one is SDXL inpaint+refiner and the other is T2I multi-adapter exceptions; different features and failure modes." }, { "left": "issue:2185", "right": "issue:2712", "accept": false, - "reason": "One is missing training config files, the other is a missing config key (`norm_num_groups`); both are config-related but not the same underlying defect." + "reason": "Both are config-related, but one is missing training config files and the other is missing a specific normalization config key; unrelated issues." }, { "left": "issue:11387", "right": "issue:11864", "accept": false, - "reason": "AutoencoderDC RMSNorm type mismatch is a model-architecture bug, while the compile failure is a torch.compile scoping issue; unrelated fixes." + "reason": "One is an AutoencoderDC normalization implementation issue; the other is a torch.compile scoping error in encode. Different underlying bugs." }, { "left": "issue:6095", "right": "issue:9001", "accept": false, - "reason": "An SVD scheduler index error and an Auraflow aspect-ratio feature request are different problems with no shared failing path." + "reason": "Both are scheduler-related, but one is an IndexError in Euler discrete scheduling for SVD and the other is aspect-ratio handling in Auraflow; not the same change." }, { "left": "issue:490", "right": "issue:801", "accept": false, - "reason": "Both involve memory/attention behavior in image pipelines, but the reported issue shapes differ: general diffusers memory use versus enable_attention_slicing not being respected in inpaint/img2img." + "reason": "Both discuss memory/attention behavior in image pipelines, but one is generic memory usage and the other is attention slicing in inpaint/img2img; too broad to merge." }, { "left": "issue:3682", "right": "issue:4630", "accept": false, - "reason": "Loading LoRAs and modifying LoRA weights after loading are related topics, but they are distinct feature requests with different operations and expected APIs." + "reason": "Both concern LoRAs, but one is loading weights and the other is changing LoRA weight after load; distinct functionality." }, { "left": "issue:1335", "right": "issue:852", "accept": false, - "reason": "These are both constructor/backend compatibility issues, but one is ONNX pipeline initialization and the other is an MPS `device_map` crash; not the same bug." + "reason": "Both are constructor/runtime compatibility issues, but one is ONNX pipeline init and the other is MPS device_map handling; unrelated code paths." }, { "left": "issue:2412", "right": "issue:3647", "accept": false, - "reason": "Both are about Dreambooth training, but one is a generic Dreambooth error and the other is specifically DeepFloyd training; too broad to merge." + "reason": "Both are Dreambooth/training-related, but one is a generic Dreambooth error and the other is specific to DeepFloyd support; not duplicates." }, { "left": "issue:4808", "right": "issue:6393", "accept": false, - "reason": "SHAP-E integration test failure and private HF download failure via `from_single_file` are unrelated." + "reason": "Both involve loading-related failures, but one is an integration test for SHAP-E img2img and the other is private HF link download via from_single_file; different problems." }, { "left": "issue:10940", "right": "issue:9034", "accept": false, - "reason": "`triton.ops` import failure is a dependency/import problem; sentencepiece requirement for AutoPipelineForText2Image is a packaging/dependency request. Not the same bug." + "reason": "Both mention import/requirements issues, but one is missing triton.ops and the other is an unnecessary sentencepiece requirement; not the same bug." }, { "left": "issue:1231", "right": "issue:5224", "accept": false, - "reason": "K-LMS scheduler in Flax and a Flax Stable Diffusion 2.1 error are both Flax-related but refer to different pipeline/scheduler issues." + "reason": "Both mention schedulers/Flax, but one is a K-LMS scheduler feature request and the other is a Stable Diffusion 2.1 error; different contexts." }, { "left": "issue:3382", "right": "issue:5197", "accept": false, - "reason": "Both concern multi-GPU training, but one is a CUDA OOM bug and the other is a performance regression; different underlying problems." + "reason": "Both are multi-GPU training performance issues, but one is CUDA OOM in a script and the other is slower-than-single-GPU throughput; not a duplicate fix." }, { "left": "issue:10025", "right": "issue:9822", "accept": false, - "reason": "Attention mask handling in Flux transformer and loading SDXL LoRAs into Flux are separate model/loading behaviors." + "reason": "Both mention Flux, but one is about attention masks and the other is LoRA loading into Flux; separate functionality." }, { "left": "issue:1530", "right": "issue:4390", "accept": false, - "reason": "Missing VQDiffusion sampling embeddings and generic missing requirements are not the same issue." + "reason": "Both are missing-configuration style reports, but one is a specific VQDiffusion pipeline field and the other is missing requirements in general; not the same issue." }, { "left": "issue:3015", "right": "issue:6370", "accept": false, - "reason": "ControlNet format conversion and using diffusers LoRA in AUTOMATIC1111 are different interoperability requests." + "reason": "Both mention converting/loading LoRA or ControlNet assets, but one is ControlNet format conversion and the other is using diffusers LoRA in A1111; different tasks." }, { "left": "issue:4747", "right": "issue:6374", "accept": false, - "reason": "SDXL LoRA load failing with missing `model_index.json` and a Chinese-language Qwen LoRA loading question are not the same failure." + "reason": "Both concern loading fine-tuned/LoRA assets, but one is a missing model_index.json after load and the other is a Qwen LoRA weight question; unrelated." }, { "left": "issue:5595", "right": "issue:6695", "accept": false, - "reason": "A height/width divisibility validation error is unrelated to a subprocess failure when starting training." + "reason": "Both are runtime errors, but one is a shape divisibility ValueError and the other is a subprocess failure while starting training; no shared root cause." }, { "left": "issue:11178", "right": "issue:12423", "accept": false, - "reason": "MistoLine ControlNet loading via `from_single_file` and QwenImageTransformer2DModel `from_single_file` are separate model-loading cases." + "reason": "Both are single-file/model loading issues, but one is MistoLine ControlNet and the other is QwenImageTransformer2DModel.from_single_file; different models and loaders." }, { "left": "issue:11010", "right": "issue:12108", "accept": false, - "reason": "Both reference Chroma/Flux and flow-matching context, but one is about supporting a Flux-based architecture and the other is a scheduler-parameter break; they are not the same concrete bug/change." + "reason": "Both relate to Flux/Chroma and flow-style models, but one is support for Chroma architecture changes and the other is a scheduler parameterization break; not mergeable as one bug." }, { "left": "issue:1363", "right": "issue:945", "accept": false, - "reason": "ONNX long-prompt weight + LMS scheduler error and adding a custom checkpoint for ONNX conversion are different ONNX-related issues." + "reason": "One is an ONNX scheduler interaction error; the other is adding a custom checkpoint model for ONNX conversion. Different issue classes." }, { "left": "issue:1274", "right": "issue:962", "accept": false, - "reason": "Both are SD pipeline behavior bugs, but one is `num_images_per_prompt` regression and the other is inpaint pipeline size support." + "reason": "Both are SD pipeline shape/batch handling issues, but one is num_images_per_prompt regression and the other is inpaint size support; not the same defect." }, { "left": "issue:11901", "right": "issue:12147", "accept": false, - "reason": "Both involve Wan2.1/Wan2.2 LoRA loading, but they target different model versions and failure modes; too risky to treat as one bug." - }, - { - "left": "issue:11220", - "right": "issue:9756", - "accept": false, - "reason": "Documentation page behavior and a missing notebook file are unrelated documentation issues." - }, - { - "left": "issue:5317", - "right": "issue:5320", - "accept": false, - "reason": "Both are PEFT-related feature requests, but documenting the API and adding `adapter_name` to `fuse_lora` are distinct changes and not one concrete fix." + "reason": "Both are Wan LoRA loading reports, but they target different model generations and likely different loaders/failure points; too weak for duplicate triage." } ] }, @@ -56860,6 +56862,7 @@ "issue:10520", "issue:10590", "issue:10891", + "issue:11220", "issue:11436", "issue:11569", "issue:11644", @@ -56867,10 +56870,8 @@ "issue:11710", "issue:1889", "issue:2053", - "issue:2150", "issue:2635", "issue:2676", - "issue:2783", "issue:2905", "issue:3199", "issue:3300", @@ -56879,10 +56880,11 @@ "issue:4265", "issue:4303", "issue:4923", + "issue:5317", + "issue:5320", "issue:5327", "issue:5509", "issue:5558", - "issue:6069", "issue:6101", "issue:6423", "issue:6452", @@ -56892,7 +56894,6 @@ "issue:7751", "issue:7936", "issue:7939", - "issue:8323", "issue:8435", "issue:8494", "issue:8593", @@ -56901,9 +56902,12 @@ "issue:9303", "issue:9306", "issue:9667", + "issue:9756", "issue:9876" ], "soft_pairs": [ + "issue:11220|issue:9756", + "issue:5317|issue:5320", "issue:10891|issue:11436", "issue:8494|issue:8593", "issue:5509|issue:9225", @@ -56926,188 +56930,186 @@ "issue:2053|issue:4159", "issue:7293|issue:8780", "issue:10302|issue:9667", - "issue:5558|issue:6101", - "issue:2150|issue:2783", - "issue:6069|issue:8323" + "issue:5558|issue:6101" ], - "prepared_review_unit_hash": "6b690f7b210c6797b53f210753177c3b601b3764e77744cd295e1646f351f237", + "prepared_review_unit_hash": "be3273e6bf170122352579858d72eb7dcc497874dfc619c5fc7f82c53b65c72d", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34189, - "estimated_input_tokens": 8548, - "estimated_eval_tokens": 17352 + "serialized_chars": 34180, + "estimated_input_tokens": 8545, + "estimated_eval_tokens": 17346 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate pairs look strong enough to merge; the soft edges are mostly broad thematic similarities (dtype/device/scheduler/loader issues) but each pair describes a different concrete bug or feature request.", - "confidence": 0.92, - "canonical_issue_reason": "issue:9306 is the most concrete, narrowly scoped runtime bug here, with a specific pipeline and a clear device-selection symptom.", + "summary": "This cluster is heterogeneous: it mixes unrelated device-routing bugs, dtype/bfloat16 mismatches, feature requests, scheduler questions, docs issues, and model-loading failures. None of the soft edges look like true duplicates or the same fixable code path.", + "confidence": 0.96, + "canonical_issue_reason": "issue:8780 is the broadest generic symptom in the set ('Model and input data type is not same'), so it is the closest weak umbrella for the dtype-related subset, but the overall cluster is not cohesive enough to treat as one duplicate thread.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9306 is the strongest standalone issue candidate overall because it is specific, reproducible, and actionable.", + "best_issue_reason": "issue:8780 is the most representative fallback because it is generic and open-ended, but it still only covers a subset of the issues and is not a strong global canonical for the whole cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11220", + "right": "issue:9756", + "accept": false, + "reason": "Docs page not working and missing notebook are unrelated documentation/navigation problems, not the same underlying bug." + }, + { + "left": "issue:5317", + "right": "issue:5320", + "accept": false, + "reason": "Both are PEFT-related, but one is documentation and the other is an API enhancement request; different changes." + }, { "left": "issue:10891", "right": "issue:11436", "accept": false, - "reason": "Both are bfloat16-related, but one is a DCAE type-conversion bug and the other is a HiDream matmul dtype mismatch; different code paths." + "reason": "Both involve bfloat16, but they hit different components and failure modes: DCAE internal conversions vs Hidream matmul dtype mismatch." }, { "left": "issue:8494", "right": "issue:8593", "accept": false, - "reason": "One is about an unexpected force_hooks argument in SD3 loading; the other is a CPU bfloat16 noise issue. Not the same bug." + "reason": "Unexpected force_hooks kwarg and CPU bfloat16 noise are different SD3 issues with no shared code path." }, { "left": "issue:5509", "right": "issue:9225", "accept": false, - "reason": "Both involve image/latent shape mismatches, but they hit different pipelines and different failure modes." + "reason": "Both are image-shape related, but one is prompt/image batch sizing and the other is RGBA latent sizing; different bugs." }, { "left": "issue:10191", "right": "issue:9876", "accept": false, - "reason": "GPU device selection during training is unrelated to LoRA VRAM release after training." + "reason": "GPU device selection during training and VRAM not being released after LoRA training are different resource-management problems." }, { "left": "issue:1889", "right": "issue:2635", "accept": false, - "reason": "Both mention DPM++ schedulers, but one asks about a variant and the other asks naming/lookup; not the same defect or change." + "reason": "A scheduler variant request and a scheduler-name mapping question are related only broadly, not the same issue or fix." }, { "left": "issue:7223", "right": "issue:9303", "accept": false, - "reason": "These are separate feature requests for different video-generation models." + "reason": "Separate feature requests for different video models; not the same underlying change." }, { "left": "issue:10590", "right": "issue:10891", "accept": false, - "reason": "Both mention bfloat16, but one is Sana tiled VAE decode and the other is DCAE type conversion; different failures." + "reason": "Both mention bfloat16, but one is Sana tiled VAE decode and the other is DCAE internal type conversion; different code paths." }, { "left": "issue:2905", "right": "issue:6452", "accept": false, - "reason": "Scheduler sigmas support and masked_image_latents input are unrelated changes." + "reason": "Adding Karras sigmas to schedulers and adding masked_image_latents are unrelated feature requests." }, { "left": "issue:4923", "right": "issue:6423", "accept": false, - "reason": "AudioLDM2 transcript support and fill50k result reproduction are unrelated issues." + "reason": "AudioLDM2 transcript support and fill50k reproduction are unrelated issues." }, { "left": "issue:10520", "right": "issue:9306", "accept": false, - "reason": "Both are device-mismatch symptoms, but they involve different pipelines and likely different root causes." + "reason": "Both are device-placement errors, but they concern different pipelines and likely different bugs." }, { "left": "issue:10227", "right": "issue:11653", "accept": false, - "reason": "Both concern LoRA handling, but one is a feature limitation on feature sizes and the other is alpha-key support." + "reason": "Both concern LoRA support, but one is a feature limitation on feature sizes and the other is .alpha-key compatibility; not the same fix." }, { "left": "issue:11569", "right": "issue:11710", "accept": false, - "reason": "A repo request for LTXV is unrelated to MagCache support." + "reason": "Different feature requests for different models/projects; no shared code defect." }, { "left": "issue:11644", "right": "issue:8435", "accept": false, - "reason": "Both are loading errors, but for different APIs and different pipelines." + "reason": "LTXConditionPipeline init failure and from_single_file runtime error are different loading paths and symptoms." }, { "left": "issue:2676", "right": "issue:702", "accept": false, - "reason": "Scheduler naming/support questions are not the same as reverse-sampling support." + "reason": "Scheduler naming question and reverse sampling proposal are about the same area but not the same underlying change." }, { "left": "issue:3300", "right": "issue:4265", "accept": false, - "reason": "General Hugging Face breakage is not the same as a specific import error for one pipeline." + "reason": "A general 'Diffusers on HuggingFace are broken' report and an import error for a specific pipeline are not the same bug." }, { "left": "issue:4303", "right": "issue:7751", "accept": false, - "reason": "These concern different model conversion/loading workflows." + "reason": "Flax SDXL loading crash and animatediff conversion errors are unrelated." }, { "left": "issue:7936", "right": "issue:7939", "accept": false, - "reason": "Both are local-loading complaints, but one is an invalid JSON parse on safetensors and the other is a failure to load a model from the same folder." + "reason": "Both are loading failures, but one is invalid JSON in a safetensors file and the other is loading from the same folder; different causes." }, { "left": "issue:3975", "right": "issue:5327", "accept": false, - "reason": "A typo in an error message is unrelated to a textual inversion loading crash." + "reason": "A typo in an error message and SDXL textual inversion loading failure are unrelated." }, { "left": "issue:3199", "right": "issue:3300", "accept": false, - "reason": "Custom pipelines broken and Hugging Face release broken are too broad to be the same bug." + "reason": "Custom pipelines broken and HF-hosted diffusers broken are too broad and refer to different problems." }, { "left": "issue:2053", "right": "issue:4159", "accept": false, - "reason": "Both mention VRAM, but one is a regression report and the other is a specific DiffEdit memory blowup." + "reason": "Both mention VRAM, but one is a version-wide regression and the other is DiffEdit's high memory use; not the same issue." }, { "left": "issue:7293", "right": "issue:8780", "accept": false, - "reason": "An AudioLDM2 feature request is unrelated to a generic dtype mismatch report." + "reason": "AudioLDM2 model integration request and a generic dtype mismatch error are unrelated." }, { "left": "issue:10302", "right": "issue:9667", "accept": false, - "reason": "FP8 inference noise and loading a local flux transformer file are different issues." + "reason": "Both involve FP8/transformer-flux context, but one reports inference noise without CPU offloading and the other a local file loading error; different fixes." }, { "left": "issue:5558", "right": "issue:6101", "accept": false, - "reason": "Artefacts with multiple ControlNets and slow ControlNet inpaint performance are different problems." - }, - { - "left": "issue:2150", - "right": "issue:2783", - "accept": false, - "reason": "A texture-projection compatibility request is not the same as creating an img2img ControlNet pipeline." - }, - { - "left": "issue:6069", - "right": "issue:8323", - "accept": false, - "reason": "A scheduler add_noise edge case and an unrecognized scheduler file are unrelated." + "reason": "Same pipeline family, but one is output artefacts with multiple ControlNets and the other is performance regression; not the same concrete bug." } ] }, @@ -57131,15 +57133,14 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10066", "issue:10591", "issue:10772", "issue:10855", - "issue:12037", "issue:12147", "issue:13279", "issue:1343", "issue:1961", + "issue:2150", "issue:2220", "issue:2312", "issue:2331", @@ -57147,12 +57148,12 @@ "issue:2424", "issue:2453", "issue:2644", + "issue:2783", "issue:2843", "issue:3480", "issue:3689", "issue:3694", "issue:3740", - "issue:4447", "issue:4453", "issue:4724", "issue:4782", @@ -57162,6 +57163,7 @@ "issue:5197", "issue:5829", "issue:6055", + "issue:6069", "issue:6090", "issue:6288", "issue:6357", @@ -57169,7 +57171,7 @@ "issue:6570", "issue:7210", "issue:7895", - "issue:8465", + "issue:8323", "issue:8477", "issue:8514", "issue:8543", @@ -57181,6 +57183,8 @@ "issue:9979" ], "soft_pairs": [ + "issue:2150|issue:2783", + "issue:6069|issue:8323", "issue:6570|issue:9225", "issue:10772|issue:9511", "issue:10855|issue:8543", @@ -57202,182 +57206,180 @@ "issue:1961|issue:6055", "issue:10591|issue:7895", "issue:1343|issue:2424", - "issue:5829|issue:6288", - "issue:4447|issue:8465", - "issue:10066|issue:12037" + "issue:5829|issue:6288" ], - "prepared_review_unit_hash": "ee3527d1de87be1deca4c3f8940a012b8076fb9f5de8986b671c1e3574e01b91", + "prepared_review_unit_hash": "69294bcaec44989ad1b27bbceaba42f1a8c202fed85d6977b4c741cfaaea570a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33813, - "estimated_input_tokens": 8454, - "estimated_eval_tokens": 17164 + "serialized_chars": 33639, + "estimated_input_tokens": 8410, + "estimated_eval_tokens": 17076 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated issues that only share broad subsystem keywords (SDXL, LoRA, ControlNet, accelerate, training). None of the soft pairs looks like the same underlying bug/change closely enough to merge.", - "confidence": 0.97, - "canonical_issue_reason": "issue:6570 is a concrete, well-scoped runtime bug report with a specific pipeline and reproducible tensor-shape failure, making it the best representative issue in this otherwise heterogeneous cluster.", + "summary": "This is a noisy, heterogeneous cluster of unrelated issue reports across training scripts, pipeline features, scheduler/runtime errors, and dependency/loading problems. None of the soft edges look like the same underlying bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:6570 is the most actionable and technically specific issue here; it has a clear error and target code path, unlike the broader feature requests and vague problem reports.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2150", + "right": "issue:2783", + "accept": false, + "reason": "Both are ControlNet-adjacent, but one is about texture projection with img2img and the other is a request for an img2img/inpainting ControlNet pipeline; different feature work." + }, + { + "left": "issue:6069", + "right": "issue:8323", + "accept": false, + "reason": "Different failure modes: a scheduler math bug at t=0 versus a Chinese-language scheduler file recognition issue." + }, { "left": "issue:6570", "right": "issue:9225", "accept": false, - "reason": "Both involve SDXL InstructPix2Pix shape issues, but one is a general tensor mismatch and the other is specifically rgba latent sizing; similar symptom, not the same concrete bug." + "reason": "Both involve SDXL InstructPix2Pix, but one is a tensor-size mismatch in the pipeline and the other is a latent-size issue specifically with RGBA images; related but not the same concrete bug." }, { "left": "issue:10772", "right": "issue:9511", "accept": false, - "reason": "Different features and models: Sana ControlNet support vs SDXL ControlNet inpaint batching. Same broad domain, not one bug/change." + "reason": "Sana ControlNet support request and multi-ControlNet batching for SDXL inpaint are separate feature requests for different pipelines." }, { "left": "issue:10855", "right": "issue:8543", "accept": false, - "reason": "Gradient accumulation optimizer stepping bug is unrelated to SD3 dreambooth gradient checkpointing failure." + "reason": "Gradient accumulation stepping every iteration is an optimizer loop bug, while the SD3 dreambooth issue is about gradient checkpointing failure; unrelated training problems." }, { "left": "issue:3694", "right": "issue:5197", "accept": false, - "reason": "Pokemon example path/config issue is unrelated to multi-GPU slowdown during text-to-image finetuning." + "reason": "One is a missing model/config path during example fine-tuning; the other is multi-GPU training being slower than single GPU. Different symptoms and root causes." }, { "left": "issue:2331", "right": "issue:2390", "accept": false, - "reason": "ControlNet pipeline support and T2I-Adapter support are separate feature requests with different implementation targets." + "reason": "Both are pipeline feature requests, but ControlNet and T2I-Adapter are distinct integrations with different code paths." }, { "left": "issue:2453", "right": "issue:4923", "accept": false, - "reason": "How to contribute is a meta question; AudioLDM 2 transcript/TTS is a specific feature request." + "reason": "A general contribution question is unrelated to an AudioLDM 2 feature request." }, { "left": "issue:4724", "right": "issue:8994", "accept": false, - "reason": "SDXL img2img checkpoint behavior and generation steps stuck at 0 are different failure modes and code paths." + "reason": "Both mention generation behavior, but one is SDXL img2img not working as expected and the other is steps stuck at 0; not the same bug." }, { "left": "issue:3740", "right": "issue:7210", "accept": false, - "reason": "Both mention accelerate/device dispatch, but one is legacy attention block conversion and the other is missing named_parameters on a pipeline object." + "reason": "Different accelerate/device_map issues: legacy attention conversion with sequential dispatch versus missing named_parameters on a pipeline object." }, { "left": "issue:4782", "right": "issue:4907", "accept": false, - "reason": "Mask bleed in inpainting is not the same as an SDXL new-inpainting-model warning." + "reason": "Inpainting mask leakage versus a warning about a new inpainting model under SDXL; different problems." }, { "left": "issue:4874", "right": "issue:6357", "accept": false, - "reason": "Local low_res_scheduler bug and API-vs-local result differences are too broad to be the same underlying defect." + "reason": "Upscale pipeline local-loading bug and API-vs-local output differences are not the same issue." }, { "left": "issue:12147", "right": "issue:13279", "accept": false, - "reason": "Wan2.2 Lightning LoRA loading failure is unrelated to AutoencoderRAE loading with older transformers." + "reason": "Different model-loading failures: Wan2.2 Lightning LoRA loading versus AutoencoderRAE loading with older transformers." }, { "left": "issue:8477", "right": "issue:8514", "accept": false, - "reason": "Both are distributed-training errors, but the scripts and failure causes differ; not the same concrete bug." + "reason": "Both are distributed training errors, but one is a DDP error in a consistency script and the other is an EMAModel/InstructP2P parallel fine-tuning failure; not the same code path." }, { "left": "issue:6090", "right": "issue:9979", "accept": false, - "reason": "Both are channel mismatch errors, but they occur in different models/pipelines and do not indicate the same fixable code path." + "reason": "Both are tensor/channel mismatch errors, but they occur in different pipelines and likely different model-input contracts." }, { "left": "issue:2220", "right": "issue:2644", "accept": false, - "reason": "Multi-GPU dreambooth problems and whether two 12GB GPUs are sufficient are not the same issue." + "reason": "Both ask about multi-GPU DreamBooth training, but one is general problems and the other is whether training on two 12GB GPUs is supported; not the same concrete bug." }, { "left": "issue:2312", "right": "issue:2843", "accept": false, - "reason": "EMA save hook error and multi-GPU checkpoint saving are related only loosely by saving; not the same bug." + "reason": "EMA hook save error versus multi-GPU checkpoint saving; both are saving-related but distinct failure points and contexts." }, { "left": "issue:3480", "right": "issue:3689", "accept": false, - "reason": "Text-encoder LoRA scale support and unloading LoRA weights are different LoRA API requests." + "reason": "A request for a text-encoder LoRA scale parameter is unrelated to unloading LoRA weights after loading." }, { "left": "issue:4453", "right": "issue:6399", "accept": false, - "reason": "LoRA conversion for A1111 compatibility and LoRA visibility in UIs are related ecosystem concerns, but not one concrete bug." + "reason": "Both concern LoRA compatibility/visibility, but one is SDXL LoRA conversion for A1111 while the other is pytorch_lora_weights visibility in UIs; different interoperability problems." }, { "left": "issue:9069", "right": "issue:9293", "accept": false, - "reason": "NumPy type mismatch and missing Float8 storage object are unrelated type/runtime errors." + "reason": "A NumPy type mismatch versus a missing Float8 storage backend are unrelated dependency/type errors." }, { "left": "issue:1961", "right": "issue:6055", "accept": false, - "reason": "ColossalAI error and generic accelerate error do not point to the same code path or defect." + "reason": "A ColossalAI-specific error and a generic accelerate error are too underspecified and appear unrelated." }, { "left": "issue:10591", "right": "issue:7895", "accept": false, - "reason": "SD3 LoRA training script issue and Flax attention bug are separate stacks and problems." + "reason": "SD3 LoRA training-script problem and Flax memory-efficient attention bug are different frameworks and code paths." }, { "left": "issue:1343", "right": "issue:2424", "accept": false, - "reason": "Enable xformers by default is a feature/policy issue; flake8-comprehension is code-style tooling." + "reason": "Enabling xformers by default is a feature/configuration change; flake8-comprehension checks are linting, not the same change." }, { "left": "issue:5829", "right": "issue:6288", "accept": false, - "reason": "Missing unet_time_cond_proj_dim argument in LCM training is unrelated to the default learning rate discussion." - }, - { - "left": "issue:4447", - "right": "issue:8465", - "accept": false, - "reason": "Restricted model access and missing config.json in from_single_file are entirely different loading failures." - }, - { - "left": "issue:10066", - "right": "issue:12037", - "accept": false, - "reason": "These are different model-specific attention/LoRA compatibility issues with no clear shared fix." + "reason": "LCM training script missing an argument versus default LR being too small; both are training-script issues but not the same bug." } ] }, @@ -57401,10 +57403,11 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10066", "issue:10100", "issue:1064", - "issue:1083", "issue:11907", + "issue:12037", "issue:1546", "issue:1604", "issue:1831", @@ -57412,7 +57415,6 @@ "issue:2044", "issue:2177", "issue:2340", - "issue:2495", "issue:2614", "issue:2646", "issue:2761", @@ -57422,6 +57424,7 @@ "issue:4014", "issue:4210", "issue:4376", + "issue:4447", "issue:4593", "issue:4638", "issue:5981", @@ -57436,13 +57439,12 @@ "issue:7002", "issue:709", "issue:7219", - "issue:7415", + "issue:8465", "issue:851", "issue:8588", "issue:9063", "issue:9141", "issue:9366", - "issue:9476", "issue:9511", "issue:9619", "issue:968", @@ -57451,6 +57453,8 @@ "issue:9972" ], "soft_pairs": [ + "issue:4447|issue:8465", + "issue:10066|issue:12037", "issue:5981|issue:5989", "issue:4014|issue:6295", "issue:1064|issue:1546", @@ -57472,104 +57476,114 @@ "issue:1941|issue:2177", "issue:2761|issue:709", "issue:3978|issue:6283", - "issue:6632|issue:8588", - "issue:1083|issue:2495", - "issue:7415|issue:9476" + "issue:6632|issue:8588" ], - "prepared_review_unit_hash": "52f89865d38894306a2fb7da4ada06bcf4760d2b64b7541bb5d2c8752c76fa34", + "prepared_review_unit_hash": "7118101572ceeff53b2af18f3bd4c1cd9824633c08a299ad15880063128fd060", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33783, - "estimated_input_tokens": 8446, - "estimated_eval_tokens": 17148 + "serialized_chars": 33854, + "estimated_input_tokens": 8464, + "estimated_eval_tokens": 17184 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are broad-theme matches only and should stay separate. The only strong duplicate-looking pair is the SDXL textual inversion failure: \"StableDiffusionXLPipeline crashes when calling load_textual_inversion\" and \"load_textual_inversion doesn't work on SDXL\".", + "summary": "This set is mostly unrelated issues spanning different pipelines, schedulers, and feature requests. The only clear duplicate pair is the SDXL textual inversion / load_textual_inversion failure, where both reports describe the same underlying bug.", "confidence": 0.89, - "canonical_issue_reason": "issue:4638 is the clearest representative: it names the exact operation (`load_textual_inversion`) and the affected model family (SDXL), making the underlying bug easier to recognize and deduplicate.", + "canonical_issue_reason": "issue:4638 is the most general and direct statement of the bug (\u201cload_textual_inversion doesn't work on SDXL\u201d), making it a good canonical representative for the duplicate SDXL textual inversion reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:4638 is the best global issue representative because it is concise, specific, and broad enough to cover the duplicate SDXL textual inversion failure without overfitting to a single crash report.", + "best_issue_reason": "issue:4638 best represents the shared underlying bug because it is concise, specific to SDXL, and broad enough to cover the crash report in issue:4376.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4447", + "right": "issue:8465", + "accept": false, + "reason": "Different problems: one is a restricted-model access/auth issue, the other is a from_single_file config loading error." + }, + { + "left": "issue:10066", + "right": "issue:12037", + "accept": false, + "reason": "Both mention model incompatibility, but they are different model architectures and failure modes, not the same bug." + }, { "left": "issue:5981", "right": "issue:5989", "accept": false, - "reason": "Both are generic bug reports, but one is about a scheduler and the other about safetensors; no shared concrete code path." + "reason": "Unrelated: scheduler bug vs safetensors.torch bug." }, { "left": "issue:4014", "right": "issue:6295", "accept": false, - "reason": "Both mention artifacts, but they involve different pipelines/settings and likely different root causes." + "reason": "Both involve image artifacts, but the titles point to different schedulers and likely different root causes." }, { "left": "issue:1064", "right": "issue:1546", "accept": false, - "reason": "One is a scheduler add_noise error, the other is a truncation-warning/batched-input failure; unrelated failure modes." + "reason": "Different code paths: LMSDiscreteScheduler.add_noise() error vs batched-input truncation warning handling." }, { "left": "issue:2044", "right": "issue:851", "accept": false, - "reason": "Both involve dtype errors, but one is a DDP attribute error and the other is a float/half type mismatch." + "reason": "Both are dtype-related errors, but they are different runtime contexts and not clearly the same defect." }, { "left": "issue:2614", "right": "issue:968", "accept": false, - "reason": "Changing the download directory and slow model loading are different problems." + "reason": "Download-directory configuration and model-load slowness are distinct issues." }, { "left": "issue:9511", "right": "issue:9619", "accept": false, - "reason": "These are distinct ControlNet feature requests: multi-controlnet batching vs interrupt support." + "reason": "Different ControlNet requests: batching for inpaint vs interrupt support." }, { "left": "issue:1831", "right": "issue:2340", "accept": false, - "reason": "Different pipelines and bugs: Dreambooth precision mismatch vs tiny SD safety-checker error." + "reason": "Different pipelines and failures: Dreambooth class-image precision mismatch vs safety-checker error in tiny SD testing." }, { "left": "issue:7002", "right": "issue:9972", "accept": false, - "reason": "Both mention buffer-size explosions, but they affect different products/pipelines and are not clearly the same bug." + "reason": "Both are huge-buffer symptoms, but the affected pipelines and reported behavior differ too much to treat as one bug." }, { "left": "issue:4210", "right": "issue:7000", "accept": false, - "reason": "A usage question about attention_mask is unrelated to a PyTorch deprecation warning." + "reason": "A forward-signature/attention_mask question is unrelated to a PyTorch deprecation warning." }, { "left": "issue:6113", "right": "issue:6689", "accept": false, - "reason": "One asks how to use a HF repo, the other reports a conversion failure; not the same underlying issue." + "reason": "Different tasks: loading control models from a repo vs converting a specific checkpoint to Diffusers." }, { "left": "issue:4376", "right": "issue:4638", "accept": true, - "reason": "Both report SDXL `load_textual_inversion` failing, with one describing a crash and the other stating it does not work at all; same underlying bug." + "reason": "Same underlying bug: SDXL load_textual_inversion fails/crashes in both reports." }, { "left": "issue:10100", @@ -57581,79 +57595,67 @@ "left": "issue:2646", "right": "issue:6238", "accept": false, - "reason": "Tiled VAE IndexError and SDXL VAE issues are too vague and distinct to treat as duplicates." + "reason": "Both mention VAE, but one is tiled-VAE IndexError and the other is an SDXL VAE issue with different symptoms." }, { "left": "issue:3933", "right": "issue:4593", "accept": false, - "reason": "Graph compile/logging and LoRA rank-loading failures are unrelated." + "reason": "Graph compile logging issue and LoRA rank mismatch are unrelated." }, { "left": "issue:11907", "right": "issue:9366", "accept": false, - "reason": "Both involve schedulers, but the pipelines and failure conditions differ materially." + "reason": "Both involve schedulers, but one is a Flux pipeline incompatibility and the other is an img2img strength/step-count edge case." }, { "left": "issue:9141", "right": "issue:9904", "accept": false, - "reason": "Missing `.from_single_file()` and silent image-shape rounding are separate FluxPipeline issues." + "reason": "Both are FluxPipeline issues, but one is missing from_single_file() and the other is image-shape rounding." }, { "left": "issue:1604", "right": "issue:3191", "accept": false, - "reason": "A version-specific script breakage and a Dreambooth+LoRA report are not the same bug." + "reason": "Different failures in different features: lpw_stable_diffusion breakage vs Dreambooth+LoRA breakage." }, { "left": "issue:9063", "right": "issue:9895", "accept": false, - "reason": "Tokenizer instantiation failure and cross-device tensor placement are different FluxPipeline problems." + "reason": "Tokenizer instantiation failure and device-mismatch inference error are distinct bugs." }, { "left": "issue:1941", "right": "issue:2177", "accept": false, - "reason": "Custom pipeline loading for UnCLIP and AutoencoderKL loading are unrelated." + "reason": "Custom pipeline loading for UnCLIP is unrelated to AutoencoderKL loading failure." }, { "left": "issue:2761", "right": "issue:709", "accept": false, - "reason": "Both are documentation-related, but they concern different APIs and different docs work." + "reason": "Both are documentation-related, but they target different APIs and documentation gaps." }, { "left": "issue:3978", "right": "issue:6283", "accept": false, - "reason": "MPS img2img failure and SDXL ControlNet strength=0 behavior are different pipeline-specific bugs." + "reason": "Different img2img failures on different pipelines and platforms." }, { "left": "issue:6632", "right": "issue:8588", "accept": false, - "reason": "Lora training save_state question and missing SD3 conversion script are unrelated." - }, - { - "left": "issue:1083", - "right": "issue:2495", - "accept": false, - "reason": "FP16 ONNX incorrect output and MPS black image are different backend-specific bugs." - }, - { - "left": "issue:7415", - "right": "issue:9476", - "accept": false, - "reason": "Both involve bfloat16, but one is TPU AutoencoderKL behavior and the other is Flux training validation; not the same code path." + "reason": "A training-script question and a missing conversion script are not the same issue." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet, and the soft-edge verdicts stay conservative. The only accepted duplicate-like pair, issue:4376 vs issue:4638, is supported by both titles pointing to SDXL `load_textual_inversion`, while the rest are reasonably kept separate based on differing pipelines, APIs, or failure modes." + "feedback": "The output is grounded in the packet and stays conservative. The only accepted duplicate pair, issue:4376 vs issue:4638, is well supported by the titles, and the other pair rejections are phrased cautiously without overstating equivalence." }, "evaluator_used": true, "retried": false, @@ -57677,21 +57679,20 @@ "issue:10016", "issue:10336", "issue:10671", + "issue:1083", "issue:11396", "issue:11432", "issue:1161", "issue:12052", "issue:12728", - "issue:12783", - "issue:13105", "issue:13446", "issue:1434", "issue:1704", "issue:2331", "issue:2402", + "issue:2495", "issue:3249", "issue:3406", - "issue:3915", "issue:4232", "issue:4410", "issue:4450", @@ -57708,9 +57709,9 @@ "issue:6176", "issue:6460", "issue:6548", - "issue:660", "issue:6666", "issue:7294", + "issue:7415", "issue:7574", "issue:7575", "issue:759", @@ -57719,11 +57720,14 @@ "issue:8809", "issue:9148", "issue:9303", + "issue:9476", "issue:9637", "issue:9680", "issue:9792" ], "soft_pairs": [ + "issue:1083|issue:2495", + "issue:7415|issue:9476", "issue:11432|issue:13446", "issue:1434|issue:4674", "issue:8809|issue:9637", @@ -57745,182 +57749,180 @@ "issue:6460|issue:9148", "issue:2402|issue:4880", "issue:2331|issue:4232", - "issue:6162|issue:7575", - "issue:12783|issue:13105", - "issue:3915|issue:660" + "issue:6162|issue:7575" ], - "prepared_review_unit_hash": "b652476af8e2324a7cc2f0f5d5202d41963700d3485d5437e23da0c37b4d5e33", + "prepared_review_unit_hash": "54f24c825176cd54f4100c5f3923694e30b493a11d77ed2b678047835c1f2e45", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34211, - "estimated_input_tokens": 8553, - "estimated_eval_tokens": 17362 + "serialized_chars": 34278, + "estimated_input_tokens": 8570, + "estimated_eval_tokens": 17396 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous and the soft pairs are only superficially similar; none look like true duplicates or mergeable PR equivalents.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue: these items cover unrelated bugs/features across model loading, training, attention kernels, device/dtype placement, and pipeline-specific behavior.", + "summary": "No duplicate cluster emerges here: the items span many unrelated bugs/features across loading, dtype, device placement, memory, training, and conversion. The soft pairs are mostly keyword-similar but describe different code paths or different model/pipeline families, so none should be merged.", + "confidence": 0.97, + "canonical_issue_reason": "No canonical issue: the set is heterogeneous and the paired items do not share the same underlying bug or repro path.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10671 is the clearest self-contained runtime bug report and the most broadly phrased issue, but it is still not representative enough to unify this cluster.", + "best_issue_reason": "No single best issue for deduplication: the collection is not a coherent duplicate group, so any one issue would be an arbitrary representative.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1083", + "right": "issue:2495", + "accept": false, + "reason": "Both are inference/output correctness issues, but one is FP16 ONNX and the other is an MPS upscale black-image bug; different backends and failure modes." + }, + { + "left": "issue:7415", + "right": "issue:9476", + "accept": false, + "reason": "Both involve mixed precision/bfloat16, but one is a TPU AutoencoderKL issue and the other is Flux LoRA validation dtype mismatch; different components and triggers." + }, { "left": "issue:11432", "right": "issue:13446", "accept": false, - "reason": "Different problems: torch_dtype='auto' handling vs custom_pipeline trust verification." + "reason": "One is about torch_dtype='auto' on from_pretrained, the other about trust checks for custom_pipeline; unrelated API behaviors." }, { "left": "issue:1434", "right": "issue:4674", "accept": false, - "reason": "Both mention memory, but one is upscaler OOM and the other is a width/height switching leak." + "reason": "OOM during upscaling is not the same as a width/height memory leak across calls; different symptom and likely different fix." }, { "left": "issue:8809", "right": "issue:9637", "accept": false, - "reason": "Different code paths: feedforward sequence-length inconsistency vs XFormers attention-mask/bfloat issue." + "reason": "Both mention tensor/shape-related behavior, but one is feedforward output variance with sequence length and the other is XFormers attention-mask/bfloat16 failure." }, { "left": "issue:10016", "right": "issue:4410", "accept": false, - "reason": "Both are shape mismatches on loading, but they involve different parameters and likely different model components." + "reason": "Both are load-time shape mismatches, but for different models and parameters: SD3.5M pos_embed vs UNet conv_in.weight." }, { "left": "issue:6176", "right": "issue:7574", "accept": false, - "reason": "Different state-dict/loading failures: LoRA reload after dreambooth training vs a missing key error." + "reason": "Both are missing/mismatched weights, but one is LoRA loading after DreamBooth and the other is a specific checkpoint key error; not the same bug." }, { "left": "issue:6666", "right": "issue:8729", "accept": false, - "reason": "Both touch validation images, but the symptoms and training scripts are different." + "reason": "Validation images being black is not the same as an unexpected validation image in a webdataset controlnet script; different pipeline stage and outcome." }, { "left": "issue:5368", "right": "issue:5897", "accept": false, - "reason": "One is fp16 Q/K/V dtype mismatch; the other is CPU/CUDA device mismatch during LoRA training." + "reason": "Both are dtype/device mismatches, but one is FP16 attention QKV dtype and the other is LoRA training across CPU/CUDA; different code paths." }, { "left": "issue:4450", "right": "issue:7294", "accept": false, - "reason": "Different training bugs: multi-text-inversion behavior vs incorrect text-encoder grad settings." + "reason": "One is multi-text inversion not working, the other is a wrong text-encoder grad setting in Custom Diffusion; related area but distinct issues." }, { "left": "issue:6132", "right": "issue:9303", "accept": false, - "reason": "These are unrelated requests: loadable class support vs a new Flux inpainting pipeline." + "reason": "Adding LOADABLE_CLASSES and adding a VEnhancer model are unrelated feature requests; no shared underlying bug/change." }, { "left": "issue:12052", "right": "issue:12728", "accept": false, - "reason": "Same model family, but one is offloading device mismatch and the other is torch.compile/channels_last/Dynamo unsupported behavior." + "reason": "Both touch Wan 2.2, but one is offloading device-mismatch during multiplication and the other is torch.compile/channels_last with Dynamo unsupported behavior; distinct failures." }, { "left": "issue:5647", "right": "issue:5799", "accept": false, - "reason": "Similar word 'strength', but one is an inpaint edge-case bug and the other is API usage/documentation around the argument." + "reason": "Strength handling in SDXL inpainting is different from a general keyword-argument question about inpainting; not the same concrete bug." }, { "left": "issue:1161", "right": "issue:759", "accept": false, - "reason": "Unrelated topics: Flax-to-PyTorch conversion vs textual inversion saving behavior." + "reason": "Flax-to-PyTorch conversion and optional full-model saving for textual inversion are unrelated model/export topics." }, { "left": "issue:4724", "right": "issue:6548", "accept": false, - "reason": "Both are img2img-related, but one is checkpoint compatibility and the other is RGBA latent handling." + "reason": "Both are img2img-related, but one concerns SDXL checkpoint behavior and the other misclassifies RGBA input as latents in prepare_latents; different defects." }, { "left": "issue:8469", "right": "issue:9680", "accept": false, - "reason": "Different pipelines and failures: broken IP-Adapter faces vs SDXL unconditional generation." + "reason": "Corrupt IP-Adapter faces and broken SDXL unconditional generation are different generation bugs with no clear shared code-path." }, { "left": "issue:3249", "right": "issue:6164", "accept": false, - "reason": "Different issues: an invalid AttnUpBlock2D argument vs latent output handling in StableDiffusionPipeline." + "reason": "AttnUpBlock2D upsample_size argument support is unrelated to StableDiffusionPipeline latent output handling." }, { "left": "issue:10336", "right": "issue:10671", "accept": false, - "reason": "Different symptoms and subsystems: Sana activation explosion vs DiffusionPipeline dtype/device determinism." + "reason": "Sana activation clamping/explosion and DiffusionPipeline dtype/device determinism are different runtime problems." }, { "left": "issue:11396", "right": "issue:9792", "accept": false, - "reason": "A user conversion question and a dataset cache-loading bug are unrelated." + "reason": "A LoRA conversion/how-to question is unrelated to dataset cache mapping failure." }, { "left": "issue:1704", "right": "issue:3406", "accept": false, - "reason": "One is a generic attention question; the other is a specific only_cross_attention bug." + "reason": "A generic attention-block question is not the same as a bug where Attention fails for only_cross_attention=True." }, { "left": "issue:6460", "right": "issue:9148", "accept": false, - "reason": "Model loading hang vs a request to add a new inpainting pipeline." + "reason": "Endless loading for a specific inpainting model is unrelated to a request for adding a Flux inpainting pipeline." }, { "left": "issue:2402", "right": "issue:4880", "accept": false, - "reason": "Environment CUDA availability error vs ControlNet image input type validation." + "reason": "CUDA-not-enabled environment error and ControlNet image type validation are different classes of problems." }, { "left": "issue:2331", "right": "issue:4232", "accept": false, - "reason": "Related domain, but one is a ControlNet pipeline feature request and the other is instruct-pix2pix training/inference code." + "reason": "Adding a ControlNet pipeline and training/inference code for instruct-pix2pix with ControlNet are different feature requests, not duplicate bugs." }, { "left": "issue:6162", "right": "issue:7575", "accept": false, - "reason": "Different shape-mismatch bugs in different components (IP-Adapters vs Transformer2DModel/adanorm)." - }, - { - "left": "issue:12783", - "right": "issue:13105", - "accept": false, - "reason": "Both report noisy output, but one is Sage Attention on sm90 and the other is a Wan2.1 model output issue." - }, - { - "left": "issue:3915", - "right": "issue:660", - "accept": false, - "reason": "Different failures: torch.compile on text-to-video pipelines vs MPS fp16 dtype/revision handling." + "reason": "IP-Adapter SDXL shape mismatch and Transformer2DModel adanorm shape error are both shape-related but involve different modules and repros." } ] }, @@ -57952,6 +57954,8 @@ "issue:11569", "issue:12110", "issue:12633", + "issue:12783", + "issue:13105", "issue:1463", "issue:1566", "issue:1672", @@ -57964,11 +57968,11 @@ "issue:3045", "issue:3087", "issue:3139", - "issue:337", "issue:3467", "issue:3647", "issue:3689", "issue:384", + "issue:3915", "issue:3944", "issue:4331", "issue:460", @@ -57979,21 +57983,20 @@ "issue:5518", "issue:5574", "issue:5575", + "issue:660", "issue:6790", "issue:7054", - "issue:7243", "issue:7454", "issue:7629", "issue:8560", - "issue:8565", "issue:861", "issue:8613", - "issue:917", "issue:9216", - "issue:9264", - "issue:9362" + "issue:9264" ], "soft_pairs": [ + "issue:12783|issue:13105", + "issue:3915|issue:660", "issue:2695|issue:3467", "issue:3689|issue:7454", "issue:5574|issue:5575", @@ -58015,189 +58018,180 @@ "issue:8560|issue:8613", "issue:10672|issue:11569", "issue:1463|issue:384", - "issue:5000|issue:5518", - "issue:8565|issue:9362", - "issue:7243|issue:9216", - "issue:337|issue:917" + "issue:5000|issue:5518" ], - "prepared_review_unit_hash": "0d69a17d5a868f233515c2cf3b97eab1bdc180bb45edd1718015ba6cdc61fe0e", + "prepared_review_unit_hash": "dcc6f1f8f5d4dd7ca2f204e6f07be128d40fea0c53d536ed09c54333fa655885", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 34020, - "estimated_input_tokens": 8505, - "estimated_eval_tokens": 17266 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33220, + "estimated_input_tokens": 8305, + "estimated_eval_tokens": 16866 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated bugs, feature requests, environment issues, and training/model-specific reports. There is no strong duplicate core across the set.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue stands out; the items span distinct subsystems and failure modes (schedulers, LoRA loading/fusing, training OOMs, conversion scripts, and missing deps).", + "summary": "The soft pairs are mostly weak title-level similarities across different models, schedulers, or training features. I don\u2019t see any pair that is safely the same underlying bug/change, so all soft edges should be rejected.", + "confidence": 0.92, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:12633 is the most concrete, code-path-specific bug report here and the closest fit for a duplicate anchor, but only for scheduler-timestep issues\u2014not the whole cluster.", + "best_issue_reason": "issue:12633 is the clearest standalone bug report here: it names a specific scheduler path, a precise mismatch condition (`timestep_spacing='linspace'`), and should be easy to triage.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12783", + "right": "issue:13105", + "accept": false, + "reason": "Both mention noisy/confetti output, but one is about Sage Attention sm90 and the other about a Wan model outputting noise; different failure contexts and likely different fixes." + }, + { + "left": "issue:3915", + "right": "issue:660", + "accept": false, + "reason": "`torch.compile` on text-to-video pipelines is unrelated to MPS fp16 image generation; shared wording is only about PyTorch/runtime settings." + }, { "left": "issue:2695", "right": "issue:3467", "accept": false, - "reason": "Different bugs: tensorboard validation crash vs resuming at the wrong step with gradient accumulation." + "reason": "Both concern training workflows and validation/resume behavior, but one is Tensorboard validation failure in ControlNet training and the other is wrong resume step with gradient accumulation." }, { "left": "issue:3689", "right": "issue:7454", "accept": false, - "reason": "Both mention LoRA, but one is about unloading weights after load and the other is a generic support request." + "reason": "Both are LoRA-related, but one asks how to unload weights after loading and the other is a generic feature request for LoRA support; not the same bug/change." }, { "left": "issue:5574", "right": "issue:5575", "accept": false, - "reason": "Both are LoRA training questions, but they ask about different parameters and are not the same bug/change." + "reason": "Both are LoRA training questions, but one is about training part of UNet attention parameters and the other about hidden size for `transformer_in`; distinct configuration issues." }, { "left": "issue:1931", "right": "issue:2440", "accept": false, - "reason": "Both involve float16, but one is a dtype mismatch during Stable Diffusion 2 usage and the other is an inpainting model load error." + "reason": "Both involve float16-related failures, but one is a scalar type mismatch in Stable Diffusion 2 and the other is an inpainting model load error in float16; different code paths." }, { "left": "issue:2266", "right": "issue:460", "accept": false, - "reason": "Different support/performance problems: CPU offload slowdown vs ONNX runtime dependency selection." + "reason": "Both relate to performance/runtime on specific hardware setups, but one is sequential CPU offload slowdown and the other is ONNX runtime backend selection; not the same issue." }, { "left": "issue:3045", "right": "issue:3139", "accept": false, - "reason": "Conversion of LoRA safetensors to diffusers is not the same as asking how to use a safetensors file for generation." + "reason": "One is about converting LoRA safetensors to diffusers, the other about using a safetensors file for generation; adjacent topics but different user needs." }, { "left": "issue:6790", "right": "issue:7054", "accept": false, - "reason": "Both concern LoRA/LCM internals, but one is a state_dict key mismatch and the other is input state_dict mutation." + "reason": "Both mention LoRA/LCM training state handling, but one is a missing key in a training state_dict and the other is input state_dict being emptied by `load_lora_weights`; different bugs." }, { "left": "issue:1062", "right": "issue:4331", "accept": false, - "reason": "Dreambooth result mismatch and training-script result differences are related only broadly, not the same concrete issue." + "reason": "Both reference DreamBooth results, but one is a general inability to reproduce paper results and the other is a mismatch between two training scripts; not the same defect." }, { "left": "issue:10382", "right": "issue:12110", "accept": false, - "reason": "Different code paths: VAE encoder return-type mismatch vs UNet1D GaussianFourier initialization bug." + "reason": "One reports a VAE encoder return-type mismatch causing AttributeError; the other is a UNet1DModel GaussianFourier initialization bug. Different modules and failures." }, { "left": "issue:10695", "right": "issue:12633", "accept": false, - "reason": "Both are scheduler timestep bugs, but they affect different schedulers and different calculations; not mergeable as one fix." + "reason": "Both concern DDIM scheduler timestep handling, but one is DDIMInverseScheduler previous timestep calculation and the other is DDIMScheduler spacing mismatch; related area, not the same bug." }, { "left": "issue:1566", "right": "issue:3647", "accept": false, - "reason": "Both involve Dreambooth, but one is a checkpoint-save crash and the other is training with DeepFloyd." + "reason": "Both touch DreamBooth training, but one crashes after saving with fp16 output and the other is training with DeepFloyd; different scenarios and failure modes." }, { "left": "issue:2984", "right": "issue:3087", "accept": false, - "reason": "A model-paper/project entry is unrelated to a LoRA model-support request." + "reason": "Latent Video Diffusion Models and LyCORIS support are separate feature requests with no shared concrete bug or code path." }, { "left": "issue:3944", "right": "issue:4912", "accept": false, - "reason": "Both are about LoRA/customization, but one asks for custom network rank in training while the other asks about converting CivitAI SDXL checkpoints." + "reason": "Both ask for more LoRA/customization support, but one is network rank customization in Dreambooth and the other is converting custom SDXL checkpoints; different requests." }, { "left": "issue:9216", "right": "issue:9264", "accept": false, - "reason": "One is a custom multi-model pipeline idea; the other is a request for an inpainting model for Flux." + "reason": "A general multi-model pipeline idea and a Flux inpainting feature request are unrelated feature requests." }, { "left": "issue:2549", "right": "issue:4692", "accept": false, - "reason": "Both mention UNet2DConditionModel, but one is a functional failure on SD2.0 while the other is a memory concern." + "reason": "Both mention UNet2DConditionModel, but one says the model doesn't work for SD 2.0 and the other asks about memory usage; not the same bug." }, { "left": "issue:1672", "right": "issue:861", "accept": false, - "reason": "Both are conversion-related, but one reports wrong output size after a conversion script and the other is a general checkpoint-to-diffusers conversion failure." + "reason": "Both are conversion-related, but one reports wrong image size after ckpt-to-diffusers conversion and the other cannot convert a ckpt model at all; different failure points." }, { "left": "issue:11223", "right": "issue:7629", "accept": false, - "reason": "Both are OOM reports, but they concern different models and runtime environments; not the same underlying bug." + "reason": "Both are OOM reports, but one is layer-wise upcasting with WAN and LoRAs, while the other is SVD on PyTorch 1.13.1 versus 2.x; different root causes." }, { "left": "issue:1463", "right": "issue:5402", "accept": false, - "reason": "Different missing-package/import errors: datasets vs compute_snr." + "reason": "One is a missing `datasets` dependency, the other is an import error for `compute_snr`; both are import issues but unrelated packages/functions." }, { "left": "issue:8560", "right": "issue:8613", "accept": false, - "reason": "Both are OOM issues, but one is SD3 training on a 4090 and the other is caption_column memory usage during dataset processing." + "reason": "Both are OOM reports, but one is SD3 training on a 4090 and the other is caption-column OOM during dataset processing; different stages and triggers." }, { "left": "issue:10672", "right": "issue:11569", "accept": false, - "reason": "A callback support request is unrelated to a repo request for a specific model release." + "reason": "A pipeline callback feature request and a repo request for a specific model are unrelated enhancement requests." }, { "left": "issue:1463", "right": "issue:384", "accept": false, - "reason": "Different import/runtime failures: missing datasets package vs missing latent_dist attribute." + "reason": "Both are missing-module style errors, but one is absent `datasets` and the other is a `DiagonalGaussianDistribution` attribute error; different problems." }, { "left": "issue:5000", "right": "issue:5518", "accept": false, - "reason": "Both involve LoRA behavior, but one is unfusing regression and the other is a ModuleList attribute error in a different LoRA path." - }, - { - "left": "issue:8565", - "right": "issue:9362", - "accept": false, - "reason": "Both are IndexError reports, but they arise in different pipelines and with different off-by-one contexts." - }, - { - "left": "issue:7243", - "right": "issue:9216", - "accept": false, - "reason": "ResAdapter is a specific adapter-model issue, while the other is a generic multi-model pipeline proposal." - }, - { - "left": "issue:337", - "right": "issue:917", - "accept": false, - "reason": "These are both old image-generation bugs, but one is img2img/text-guided inpainting and the other is an ONNX inpainting error; not the same bug." + "reason": "Both involve LoRA-related regressions, but one is unfusing after fuse/unfuse and the other is a `ModuleList` missing `time`; not the same code path." } ] }, @@ -58226,20 +58220,16 @@ "issue:10893", "issue:10987", "issue:11010", - "issue:11109", "issue:11374", - "issue:11592", - "issue:12009", "issue:12149", - "issue:12396", "issue:12575", "issue:12666", - "issue:12887", "issue:13067", "issue:1909", "issue:2806", "issue:319", "issue:322", + "issue:337", "issue:4390", "issue:4609", "issue:5166", @@ -58255,22 +58245,29 @@ "issue:6586", "issue:6685", "issue:7223", + "issue:7243", "issue:7312", "issue:7417", "issue:7428", "issue:8482", "issue:8543", - "issue:8667", + "issue:8565", "issue:8880", "issue:8918", "issue:8994", + "issue:917", + "issue:9216", "issue:9258", + "issue:9362", "issue:938", "issue:9393", "issue:9665", "issue:9732" ], "soft_pairs": [ + "issue:8565|issue:9362", + "issue:7243|issue:9216", + "issue:337|issue:917", "issue:5593|issue:7417", "issue:11010|issue:12575", "issue:1909|issue:2806", @@ -58291,183 +58288,180 @@ "issue:10893|issue:13067", "issue:12149|issue:12666", "issue:4609|issue:938", - "issue:5966|issue:6412", - "issue:12009|issue:12887", - "issue:11592|issue:12396", - "issue:11109|issue:8667" + "issue:5966|issue:6412" ], - "prepared_review_unit_hash": "384169c2a9c3c2d45d7c1b6129ad104ff80757b969beef2eba82c16ad4047874", + "prepared_review_unit_hash": "0390b8ccd35da16fb178492bab8f56a29b5aa3b628966d6a6cfed5629a1b870f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33399, - "estimated_input_tokens": 8350, - "estimated_eval_tokens": 16956 + "serialized_chars": 33304, + "estimated_input_tokens": 8326, + "estimated_eval_tokens": 16908 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No strong duplicate cluster emerges; most items are unrelated feature requests or distinct bugs. The only borderline case is the Flux Lora `delete_adapters` area, but it is still too specific/different to merge with confidence.", - "confidence": 0.74, - "canonical_issue_reason": "No clear canonical issue for the cluster: the items span many unrelated model-support requests, training bugs, and regressions, with no pair clearly describing the same underlying problem.", - "canonical_pr_reason": "No PRs were provided.", - "best_issue_reason": "issue:11592 is the most concrete and actionable standalone bug report: it names a specific regression in `FluxLoraLoaderMixin.delete_adapters` and is close to the only somewhat related follow-up issue.", + "summary": "This cluster is very heterogeneous: it mixes unrelated feature requests, model-support asks, training bugs, and API issues. None of the soft pairs look like the same underlying bug/change, so all should be rejected. The closest representative issue is the broad multi-view generation feature request, but even that is only a loose thematic fit, not a true duplicate anchor.", + "confidence": 0.31, + "canonical_issue_reason": "Issue 10147 is the broadest open feature request in the set and best matches the general 'add model support' theme, though the cluster is not a true duplicate group.", + "canonical_pr_reason": null, + "best_issue_reason": "10147 is the most generic, representative open issue among mostly model-support requests; it can serve as a loose anchor even though the set is not a coherent duplicate cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:8565", + "right": "issue:9362", + "accept": false, + "reason": "Both mention IndexError, but they refer to different pipelines and failure modes; not the same bug." + }, + { + "left": "issue:7243", + "right": "issue:9216", + "accept": false, + "reason": "Both are about new pipeline/model composition, but they are different feature requests and code changes." + }, + { + "left": "issue:337", + "right": "issue:917", + "accept": false, + "reason": "Both concern inpainting, but one is an img2img/text-guided bug and the other is an ONNX inpainting error; not the same issue." + }, { "left": "issue:5593", "right": "issue:7417", "accept": false, - "reason": "Different problems: `AutoPipeline.from_pipe` with ControlNet pipelines versus a Mac-specific LoRA fuse NaN failure." + "reason": "ControlNet AutoPipeline failure and a Mac LoRA-fuse test failure are unrelated code paths." }, { "left": "issue:11010", "right": "issue:12575", "accept": false, - "reason": "Both are Chroma/Flux-related, but one is a model architecture request and the other is an attention-backend check error; not the same bug." + "reason": "A Flux/Chroma model-support request is not the same as an attention-backend check error." }, { "left": "issue:1909", "right": "issue:2806", "accept": false, - "reason": "Generator API mismatch versus custom pipeline printing; distinct behaviors and code paths." + "reason": "Generator handling in schedulers and pipeline printing/customization are different API problems." }, { "left": "issue:10147", "right": "issue:7223", "accept": false, - "reason": "Separate feature requests for multi-view generation and video generation support." + "reason": "Both are feature requests for new generation capabilities, but MV-Adapter multi-view support and Latte video support are distinct model integrations." }, { "left": "issue:10887", "right": "issue:11374", "accept": false, - "reason": "Import error for `StableDiffusionXLTrainer` is unrelated to SkyReels-V2 support integration." + "reason": "Importing StableDiffusionXLTrainer and CogVideoX ControlNet integration are unrelated." }, { "left": "issue:10987", "right": "issue:8482", "accept": false, - "reason": "Generic spatio-temporal models versus SD3-medium support; different requests." + "reason": "A general spatio-temporal model request and SD3 model support are different features." }, { "left": "issue:4390", "right": "issue:5166", "accept": false, - "reason": "Missing dependencies in setup versus a VQVAE scaling-factor bug; not the same issue." + "reason": "Missing requirements and a missing VQVAE scaling factor are different bugs." }, { "left": "issue:5371", "right": "issue:7428", "accept": false, - "reason": "Different feature requests: BLIP inpainting support versus layer diffusion support." + "reason": "BLIP inpainting support and layer diffusion are different feature requests." }, { "left": "issue:6073", "right": "issue:9732", "accept": false, - "reason": "A beginner DreamBooth setup error is not the same as Flux DreamBooth CUDA OOM during training." + "reason": "Both involve DreamBooth guides, but one is a beginner setup error and the other is CUDA OOM during Flux training." }, { "left": "issue:6287", "right": "issue:8918", "accept": false, - "reason": "Unused dependencies in `setup.py` is unrelated to the SD3 DreamBooth script being broken." + "reason": "Setup dependency cleanup and a broken SD3 dreambooth script are unrelated." }, { "left": "issue:6685", "right": "issue:8543", "accept": false, - "reason": "Absolute `output_dir` support and gradient-checkpointing failure are different DreamBooth issues." + "reason": "Absolute output_dir path handling and SD3 gradient checkpointing are different bugs." }, { "left": "issue:8880", "right": "issue:9258", "accept": false, - "reason": "Both mention speed, but one is a version regression and the other is a Windows-vs-Linux performance gap; not the same change or bug." + "reason": "Both are performance complaints, but they concern different workloads and environments." }, { "left": "issue:6545", "right": "issue:6586", "accept": false, - "reason": "A tracker for modularizing inference is unrelated to micro-conditioning support for SDXL trainers." + "reason": "A training/inference modularization tracker and SDXL micro-conditioning tracker are different changes." }, { "left": "issue:319", "right": "issue:322", "accept": false, - "reason": "Two different model families (`VQ-Diffusion` vs `V-Diffusion`), not duplicates." + "reason": "Both are early diffusion model requests, but VQ-Diffusion and V-Diffusion are distinct model families." }, { "left": "issue:5605", "right": "issue:7312", "accept": false, - "reason": "Per-image caption support for DreamBooth is unrelated to AlignProp direct reward finetuning support." + "reason": "Per-image caption support for DreamBooth and AlignProp support are different training features." }, { "left": "issue:9393", "right": "issue:9665", "accept": false, - "reason": "Both involve DeepSpeed, but one is a save-model failure and the other is `accelerator.prepare()` behavior; different concrete bugs." + "reason": "Both involve DeepSpeed, but one is a save-model failure and the other is accelerator.prepare() behavior." }, { "left": "issue:5931", "right": "issue:8994", "accept": false, - "reason": "Nondeterministic generation and steps stuck at 0 are distinct symptoms with different likely causes." + "reason": "Nondeterministic generation and generation steps stuck at 0 are different bugs." }, { "left": "issue:10893", "right": "issue:13067", "accept": false, - "reason": "Separate feature requests for CogVideoX ControlNet and Anima support." + "reason": "CogVideoX ControlNet integration and Anima support are separate model feature requests." }, { "left": "issue:12149", "right": "issue:12666", "accept": false, - "reason": "Package/source version misalignment is not the same as missing Kandinsky 5.0 video pipeline support." + "reason": "Both mention release mismatch or missing pipelines, but they are different versioning/support problems." }, { "left": "issue:4609", "right": "issue:938", "accept": false, - "reason": "Different model requests: a T2I adapter model versus a human motion diffusion model." + "reason": "A specific T2I-Adapter model request and a human motion diffusion model request are unrelated." }, { "left": "issue:5966", "right": "issue:6412", "accept": false, - "reason": "ControlNet model support and ControlNet annotator ONNX export are related only broadly, not the same issue." - }, - { - "left": "issue:12009", - "right": "issue:12887", - "accept": false, - "reason": "Both are `from_single_file` loading failures, but they concern different models and error paths." - }, - { - "left": "issue:11592", - "right": "issue:12396", - "accept": false, - "reason": "Related `delete_adapters` area, but one is a regression report and the other is a specific `leaf_level` interaction; too specific to treat as the same underlying bug without stronger evidence." - }, - { - "left": "issue:11109", - "right": "issue:8667", - "accept": false, - "reason": "Different mismatch failures in different model components: depth LoRA transformer layers versus SD3ControlNet image-size handling." + "reason": "ControlNet for SD2.1 and ControlNet annotators to ONNX are different parts of the stack and different changes." } ] }, @@ -58497,13 +58491,15 @@ "issue:10639", "issue:1083", "issue:11093", + "issue:11109", "issue:11223", "issue:11229", "issue:11315", - "issue:11536", - "issue:11540", "issue:11592", + "issue:12009", + "issue:12396", "issue:12567", + "issue:12887", "issue:13286", "issue:1539", "issue:2177", @@ -58513,14 +58509,12 @@ "issue:3379", "issue:3537", "issue:3799", - "issue:4210", "issue:4376", "issue:4949", "issue:4969", "issue:5455", "issue:5510", "issue:5576", - "issue:5661", "issue:5689", "issue:6186", "issue:656", @@ -58530,6 +58524,7 @@ "issue:8482", "issue:8504", "issue:8641", + "issue:8667", "issue:8790", "issue:9069", "issue:9166", @@ -58540,6 +58535,9 @@ "issue:9496" ], "soft_pairs": [ + "issue:12009|issue:12887", + "issue:11592|issue:12396", + "issue:11109|issue:8667", "issue:11592|issue:13286", "issue:3097|issue:4376", "issue:3537|issue:7368", @@ -58561,182 +58559,186 @@ "issue:10542|issue:11093", "issue:5510|issue:9225", "issue:10147|issue:8790", - "issue:10639|issue:8641", - "issue:11536|issue:11540", - "issue:4210|issue:5661" + "issue:10639|issue:8641" ], - "prepared_review_unit_hash": "9c420272417b3d1bf0d568577c5302aee15dec57bdeafb15ca2ba715967ade4e", + "prepared_review_unit_hash": "815f57ac2a638473d81dbe5c930554ca5953b3704818ed92ed88c9189d5a1fbe", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 32877, - "estimated_input_tokens": 8220, - "estimated_eval_tokens": 16696 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33615, + "estimated_input_tokens": 8404, + "estimated_eval_tokens": 17064 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All soft-edge pairs are false positives: they share broad model/runtime themes, but each pair describes a different underlying bug, feature request, or code path. No duplicate cluster is supported here.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "All proposed soft edges are superficial title-level similarities rather than the same underlying bug or feature change. The set mixes unrelated models, pipelines, export/type issues, and feature requests, so every candidate pair is rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No true canonical issue stands out; the cluster is heterogeneous and the soft matches only share broad wording.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11540 is the most detailed and actionable standalone bug report in the set, but it is not a duplicate target for the others.", + "best_issue_reason": "issue:11229 is the best standalone representative because it is a concrete, reproducible bug report with clear SDXL/MPS context and actionable symptoms.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12009", + "right": "issue:12887", + "accept": false, + "reason": "Both involve from_single_file loading, but they target different models and failure modes (Wan2.2 GGUF meta-tensor copy error vs z-image-turbo load errors)." + }, + { + "left": "issue:11592", + "right": "issue:12396", + "accept": false, + "reason": "Both mention delete_adapters, but 11592 is a FluxLoraLoaderMixin regression while 12396 is a separate leaf_level + delete_adapters error path." + }, + { + "left": "issue:11109", + "right": "issue:8667", + "accept": false, + "reason": "Different bugs: depth LoRA transformer-layer mismatch vs SD3ControlNet image-size mismatch." + }, { "left": "issue:11592", "right": "issue:13286", "accept": false, - "reason": "FluxLoRA adapter deletion regression vs torchao quantization support request; unrelated failures." + "reason": "Unrelated problems: adapter deletion regression vs unsupported torchao quantization version." }, { "left": "issue:3097", "right": "issue:4376", "accept": false, - "reason": "Both involve textual inversion, but one is a no-op bug and the other is an SDXL crash; different symptoms and likely different causes." + "reason": "Both touch textual inversion, but one is a no-op report and the other is a pipeline crash in SDXL; different concrete failures." }, { "left": "issue:3537", "right": "issue:7368", "accept": false, - "reason": "ControlNet conversion script vs GLIGEN training support; different features entirely." + "reason": "A controlnet conversion script request and GLIGEN training support are different changes with no shared bug." }, { "left": "issue:3799", "right": "issue:7535", "accept": false, - "reason": "Scheduler type inference issue vs a Tensor type-hint mismatch; not the same bug." + "reason": "Both are typing-related, but one is scheduler type inference and the other is a FloatTensor type-hint incompatibility." }, { "left": "issue:5576", "right": "issue:6186", "accept": false, - "reason": "FreeNoise scheduler request vs i2vgen-xl implementation request; separate feature work." + "reason": "A new scheduler request and i2vgen-xl support are unrelated feature requests." }, { "left": "issue:11223", "right": "issue:12567", "accept": false, - "reason": "WAN LoRA OOM with layer-wise upcasting vs Ulysses Attention producing noise; distinct runtime problems." + "reason": "Different symptoms and code paths: WAN LoRA OOM with layer-wise upcasting vs noise from Ulysses Attention." }, { "left": "issue:10003", "right": "issue:9362", "accept": false, - "reason": "DDIM/DDPM step-count bug vs an index out-of-bounds error; different code paths." + "reason": "Scheduler bugs at 1000 steps and an index-out-of-bounds error are not the same underlying issue." }, { "left": "issue:2830", "right": "issue:4949", "accept": false, - "reason": "Modelscope image/text-to-video question vs ProPainter in/outpainting request; different model use cases." + "reason": "Both are video-related questions, but one asks about image/text-to-video with ModelScope and the other is about ProPainter in/outpainting." }, { "left": "issue:3171", "right": "issue:3537", "accept": false, - "reason": "LPW prompt handling in ControlNet vs a ControlNet conversion script; not the same issue." + "reason": "LPW prompt support for ControlNet and a ControlNet conversion script are different requests." }, { "left": "issue:8482", "right": "issue:8504", "accept": false, - "reason": "General SD3-Medium support vs selecting which text encoders to train; related area, but different change requests." + "reason": "SD3-medium core support and choosing text encoders for SD3-medium are related only broadly; they are not the same concrete change." }, { "left": "issue:2177", "right": "issue:3379", "accept": false, - "reason": "AutoencoderKL loading failure vs DreamBooth checkpoint loading failure; unrelated bugs." + "reason": "Different loading failures: pretrained AutoencoderKL vs Dreambooth intermediate checkpoint loading." }, { "left": "issue:9370", "right": "issue:9496", "accept": false, - "reason": "Missing pooled_projection argument in one forward path vs SD3ControlNetModel forward error; different concrete failures." + "reason": "Both are SD3 forward-path errors, but one is a missing pooled_projection argument and the other is a different ControlNet forward failure." }, { "left": "issue:11229", "right": "issue:11315", "accept": false, - "reason": "MPS NaNs from attention slicing vs MPS breakage from float64 rope implementation; same platform, different root causes." + "reason": "Both mention MPS, but NaNs from attention slicing and float64 rope breaking on MPS have different causes and fixes." }, { "left": "issue:1083", "right": "issue:1539", "accept": false, - "reason": "FP16 ONNX incorrect output vs conversion-script int typing bug; unrelated." + "reason": "An FP16 ONNX correctness issue and a conversion-script type-int bug are unrelated." }, { "left": "issue:4969", "right": "issue:656", "accept": false, - "reason": "ImportError for CrossAttention vs DreamBooth training error on 16GB GPU; no shared underlying defect." + "reason": "CrossAttention import failure and a Dreambooth training error on 16GB GPUs are different issues." }, { "left": "issue:5455", "right": "issue:7606", "accept": false, - "reason": "DreamBooth+lora notebook subprocess failure vs deprecated Kandinsky LoRA processor; different problems." + "reason": "A subprocess failure in a notebook and a deprecated LoRA processor warning are not the same bug." }, { "left": "issue:9069", "right": "issue:9298", "accept": false, - "reason": "NumPy type mismatch vs missing Float8 storage object; unrelated compatibility issues." + "reason": "These are different type/storage problems: ndarray type mismatch vs missing Float8 storage object." }, { "left": "issue:5689", "right": "issue:9166", "accept": false, - "reason": "Residual noise in SDXL DPM++ outputs vs unexpected overlay behavior; similar output complaints, but different code paths." + "reason": "Residual noise in DPM++ output and unexpected apply_overlay behavior are separate issues in different subsystems." }, { "left": "issue:10542", "right": "issue:11093", "accept": false, - "reason": "Hunyuan Video batch-size regression vs TorchAO example breakage; not the same bug." + "reason": "Hunyuan Video batch-size breakage and a broken TorchAO example are unrelated." }, { "left": "issue:5510", "right": "issue:9225", "accept": false, - "reason": "Half/float bias mismatch in SDXL InstructPix2Pix vs latent size mismatch with RGBA images; distinct pipeline errors." + "reason": "Half/float bias mismatch in SDXL InstructPix2Pix and RGBA latent size mismatch are different pipeline bugs." }, { "left": "issue:10147", "right": "issue:8790", "accept": false, - "reason": "MV-Adapter multi-view support vs SDXL TensorRT + IP-Adapter; broad adapter theme, but different implementation targets." + "reason": "A multi-view generation feature request and SDXL TensorRT + IP Adapter support are unrelated requests." }, { "left": "issue:10639", "right": "issue:8641", "accept": false, - "reason": "Lumina-Image 2.0 support vs Lumina-T2X integration; related model family but separate feature requests." - }, - { - "left": "issue:11536", - "right": "issue:11540", - "accept": false, - "reason": "Flux prompt-adherence regression vs VAE slicing noise / Flux image2image support; different bugs, not one fixable change." - }, - { - "left": "issue:4210", - "right": "issue:5661", - "accept": false, - "reason": "UNet attention_mask usage question vs missing PreTrainedTokenizer import; unrelated issues." + "reason": "Support for Lumina-Image 2.0 and integrating Lumina-T2X are different model additions." } ] }, @@ -58769,6 +58771,8 @@ "issue:11109", "issue:11283", "issue:11412", + "issue:11536", + "issue:11540", "issue:11709", "issue:11779", "issue:1185", @@ -58785,31 +58789,31 @@ "issue:3441", "issue:3944", "issue:4005", + "issue:4210", "issue:4331", "issue:4349", "issue:4631", "issue:4871", "issue:5002", "issue:5171", - "issue:5593", + "issue:5661", "issue:6242", "issue:6423", "issue:6446", "issue:7219", "issue:725", - "issue:7569", "issue:8320", "issue:8885", "issue:9026", "issue:9293", "issue:9424", "issue:9516", - "issue:9731", - "issue:9768", "issue:9902", "issue:9930" ], "soft_pairs": [ + "issue:11536|issue:11540", + "issue:4210|issue:5661", "issue:6446|issue:8885", "issue:1101|issue:725", "issue:10202|issue:11109", @@ -58831,182 +58835,180 @@ "issue:4871|issue:5171", "issue:9026|issue:9293", "issue:4631|issue:5002", - "issue:1185|issue:4331", - "issue:5593|issue:7569", - "issue:9731|issue:9768" + "issue:1185|issue:4331" ], - "prepared_review_unit_hash": "4c5ffd83e2a1960fafe673212a7ac6524ee244d8c5ee5b62f06a318eb8598ebc", + "prepared_review_unit_hash": "a010149ec0c445d12d31f1ffff3d6ea2b3deb49df48f777c6c1a016347182efa", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33876, - "estimated_input_tokens": 8469, - "estimated_eval_tokens": 17194 + "serialized_chars": 34022, + "estimated_input_tokens": 8506, + "estimated_eval_tokens": 17268 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All proposed soft edges are false positives. The items span many unrelated bugs/features across Flux, Wan, LoRA training, schedulers, loading, and test failures, so this is not a true duplicate cluster.", - "confidence": 0.96, - "canonical_issue_reason": "No clear canonical issue: the set is too heterogeneous and the soft-linked pairs do not describe the same underlying bug.", + "summary": "The set is heterogeneous: Flux, Wan, LoRA, scheduler, loading, and training issues are mixed together. None of the soft pairs look like true duplicates of the same underlying bug/change, so all should stay separate.", + "confidence": 0.22, + "canonical_issue_reason": "issue:11536 is the clearest broad, user-facing bug report in the dominant Flux theme, with concrete behavior and discussion; it is more representative than the narrower Flux/loading variants.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a strong global representative because the items cover multiple unrelated problem areas rather than one duplicated defect.", + "best_issue_reason": "issue:11536 is the best single issue to anchor this cluster because it is specific enough to be actionable yet broad enough to represent the main Flux-related regression pattern seen here.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11536", + "right": "issue:11540", + "accept": false, + "reason": "Both are Flux-related, but one is prompt adherence regression and the other is VAE slicing/noise behavior; different code paths and symptoms." + }, + { + "left": "issue:4210", + "right": "issue:5661", + "accept": false, + "reason": "UNet attention_mask usage question is unrelated to the missing PreTrainedTokenizer import error." + }, { "left": "issue:6446", "right": "issue:8885", "accept": false, - "reason": "Different failures: one is a Stable Diffusion 2-1 loading problem, the other is a Colab/clip-retrieval/custom diffusion setup issue." + "reason": "Stable Diffusion Colab loading problem is unrelated to clip-retrieval install/modifier-token issues." }, { "left": "issue:1101", "right": "issue:725", "accept": false, - "reason": "A scheduler feature request and a memory-footprint/device-mapping feature are related only at a broad diffusion-library level, not the same change." + "reason": "Scheduler addition request and memory-footprint/loading infrastructure request are different features." }, { "left": "issue:10202", "right": "issue:11109", "accept": false, - "reason": "Both mention LoRA, but one is about unloading Flux Control LoRA correctly while the other is a depth LoRA transformer layer mismatch." + "reason": "Flux Control LoRA unloading bug is not the same as depth LoRA transformer layer mismatch." }, { "left": "issue:10010", "right": "issue:11412", "accept": false, - "reason": "Both are 404/loading-path problems, but for different model families and different missing URLs." + "reason": "Both mention 404s, but they point to different model repos/examples and likely different missing artifacts." }, { "left": "issue:6242", "right": "issue:8320", "accept": false, - "reason": "Unrelated: a request for training logs versus a from_single_file attribute error on AutoencoderKL." + "reason": "Dataset log request is unrelated to the from_single_file AutoencoderKL attribute error." }, { "left": "issue:9424", "right": "issue:9902", "accept": false, - "reason": "Notebook breakage with latest code is not the same as Conda version pinning/stuck behavior." + "reason": "Notebook breakage after updates is not the same as Conda version pinning/stuck-version behavior." }, { "left": "issue:11088", "right": "issue:11709", "accept": false, - "reason": "Different code paths and errors: Wan transformer GGUF load failure versus CausVid LoRA key loading failure." + "reason": "Wan2.1 GGUF deserialization/load error is different from CausVid LoRA key loading failure." }, { "left": "issue:12110", "right": "issue:13477", "accept": false, - "reason": "A UNet1D GaussianFourier initialization bug is unrelated to an ErnieImage callback locals-scope bug." + "reason": "UNet1DModel initialization bug and ErnieImage callback scope bug are unrelated code paths." }, { "left": "issue:12071", "right": "issue:9516", "accept": false, - "reason": "One concerns Flux Kontext transformer forward behavior changes; the other is missing propagation of joint_attention_kwargs in FLUX." + "reason": "Both are FLUX/transformer-adjacent, but one is changed forward behavior and the other is missing joint_attention_kwargs propagation." }, { "left": "issue:2210", "right": "issue:3944", "accept": false, - "reason": "Training text encoder with LoRA is a different request from customizing LoRA network rank." + "reason": "Training text encoder with LoRA is a feature request, not the same as choosing network rank for Dreambooth LoRA training." }, { "left": "issue:1672", "right": "issue:1677", "accept": false, - "reason": "Different pipeline problems: wrong output size after ckpt-to-diffusers conversion versus wrong data type when calling the pipeline." + "reason": "Wrong output size after ckpt conversion is unrelated to pipeline call data type validation." }, { "left": "issue:1934", "right": "issue:3441", "accept": false, - "reason": "Both are memory-related, but one is cpu_offload VRAM consumption and the other is higher VRAM usage with PyTorch 2 under specific conditions." + "reason": "Both involve memory usage, but one is cpu_offload VRAM blowup and the other is PyTorch2/xformers VRAM regression." }, { "left": "issue:2392", "right": "issue:4349", "accept": false, - "reason": "Both involve prompt_embeds, but one is img2img validation behavior while the other is an upscale pipeline requiring a text_encoder unexpectedly." + "reason": "Both involve prompt_embeds-related validation, but they affect different pipelines and failure modes." }, { "left": "issue:10165", "right": "issue:9930", "accept": false, - "reason": "Seed preservation when changing LoRA scale is unrelated to PAG adaptive scale behavior." + "reason": "LoRA scale breaking seed preservation is not the same as PAG adaptive-scale behavior." }, { "left": "issue:3073", "right": "issue:7219", "accept": false, - "reason": "A request to add consistency models is not the same as a request to add the SUPIR upscaler." + "reason": "Requesting Consistency Models and adding a SUPIR upscaler are different feature requests." }, { "left": "issue:4005", "right": "issue:6423", "accept": false, - "reason": "Aspect-ratio handling in inpaint pipeline is a different bug from reproducing fill50k results." + "reason": "Inpaint aspect-ratio bug is unrelated to inability to reproduce fill50k results." }, { "left": "issue:11283", "right": "issue:11779", "accept": false, - "reason": "Wan header deserialization failure and Wan VACE mask padding are separate bugs in different stages." + "reason": "Wan header deserialization failure is a loading/format issue, not the VACE mask-padding bug." }, { "left": "issue:10428", "right": "issue:12653", "accept": false, - "reason": "Ascend NPU Flux inference errors are not the same as a PRX rope float64/MPS-NPU dtype bug." + "reason": "Ascend NPU Flux inference error and PRX rope float64/MPS-NPU issue are different underlying device/dtype problems." }, { "left": "issue:4871", "right": "issue:5171", "accept": false, - "reason": "Usage question for from_single_file is not the same as an unexpected safety_checker keyword argument error." + "reason": "How to run from_single_file is unrelated to the unexpected safety_checker keyword error." }, { "left": "issue:9026", "right": "issue:9293", "accept": false, - "reason": "Different compatibility issues: an unexpected force_hook argument versus missing Float8 storage support." + "reason": "Unexpected force_hook argument and missing Float8 storage object are separate loading/runtime failures." }, { "left": "issue:4631", "right": "issue:5002", "accept": false, - "reason": "DPMSolverSinglestep scalar conversion failure is unrelated to missing use_karras_sigmas arguments in KDPM2 schedulers." + "reason": "Scheduler scalar-conversion crash with Karras sigmas is distinct from missing use_karras_sigmas support." }, { "left": "issue:1185", "right": "issue:4331", "accept": false, - "reason": "General DreamBooth reproducibility complaints do not match the specific difference between two training scripts." - }, - { - "left": "issue:5593", - "right": "issue:7569", - "accept": false, - "reason": "AutoPipeline.from_pipe with ControlNet is unrelated to a Mac test failure for FreeU." - }, - { - "left": "issue:9731", - "right": "issue:9768", - "accept": false, - "reason": "A training/how-to question for Playground2.5 is unrelated to a torch 2.5 SDPA NaN backend error." + "reason": "Dreambooth reproducibility complaints are related at a high level, but these reference different training scripts and causes." } ] }, @@ -59041,9 +59043,7 @@ "issue:1158", "issue:11584", "issue:12239", - "issue:12359", "issue:12458", - "issue:12575", "issue:1263", "issue:12695", "issue:13176", @@ -59051,7 +59051,6 @@ "issue:13277", "issue:13506", "issue:1524", - "issue:2737", "issue:305", "issue:3537", "issue:3884", @@ -59059,10 +59058,10 @@ "issue:4005", "issue:4061", "issue:4354", - "issue:5079", "issue:5328", "issue:5371", "issue:5390", + "issue:5593", "issue:6055", "issue:6079", "issue:6132", @@ -59071,15 +59070,20 @@ "issue:7210", "issue:7219", "issue:7389", + "issue:7569", "issue:7751", "issue:8323", "issue:8594", "issue:8850", "issue:8939", "issue:8989", - "issue:9111" + "issue:9111", + "issue:9731", + "issue:9768" ], "soft_pairs": [ + "issue:5593|issue:7569", + "issue:9731|issue:9768", "issue:6132|issue:7219", "issue:10057|issue:10873", "issue:11275|issue:11584", @@ -59101,182 +59105,180 @@ "issue:3884|issue:8594", "issue:4002|issue:4005", "issue:4354|issue:5390", - "issue:10672|issue:12695", - "issue:12359|issue:12575", - "issue:2737|issue:5079" + "issue:10672|issue:12695" ], - "prepared_review_unit_hash": "8c499632b0f60c5fb7b5b22f076b2ceb4c0b2e56746ad255ef1acb762398655c", + "prepared_review_unit_hash": "a3c70d113fe58593ad76fdfb5978f786ebe5cbf4e3415a6822b3cfc91b6f94e2", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33652, - "estimated_input_tokens": 8413, - "estimated_eval_tokens": 17082 + "serialized_chars": 33656, + "estimated_input_tokens": 8414, + "estimated_eval_tokens": 17084 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a real duplicate set: it mixes unrelated bugs, feature requests, and documentation issues across different pipelines/models. The soft links look like superficial text similarity rather than the same underlying code path, so none should be merged.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "No true duplicate cluster here: these are mostly unrelated issues across different pipelines, features, and failure modes. All soft edges look like false positives.", + "confidence": 0.99, + "canonical_issue_reason": "No issue is a good canonical representative: the items span unrelated bugs/features across different pipelines, scripts, and backends.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10057 is the oldest and most concrete API/loading bug, so it is the closest thing to a representative issue, but the overall cluster is too heterogeneous for a true canonical duplicate.", + "best_issue_reason": "No single issue stands out as a global best match; the cluster is heterogeneous and the soft pairs are all false positives.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5593", + "right": "issue:7569", + "accept": false, + "reason": "Different problems: AutoPipeline ControlNet handling vs a Mac FreeU test failure." + }, + { + "left": "issue:9731", + "right": "issue:9768", + "accept": false, + "reason": "Unrelated: Playground2.5 LoRA training guidance vs a CuDNN SDPA NaN backend bug." + }, { "left": "issue:6132", "right": "issue:7219", "accept": false, - "reason": "LOADABLE_CLASSES support vs a SUPIR upscaler request; unrelated feature areas." + "reason": "Different scope: serialization LOADABLE_CLASSES vs adding a new upscaler model." }, { "left": "issue:10057", "right": "issue:10873", "accept": false, - "reason": "Different problems in different pipelines: `from_single_file` token/loading issue vs SanaPipeline model-specific failure." + "reason": "Different pipelines and symptoms: Flux from_single_file token issue vs Sana pipeline issue." }, { "left": "issue:11275", "right": "issue:11584", "accept": false, - "reason": "VAE `ResnetUpsampleBlock2D` NoneType error vs torch.compile/group offloading incompatibility; different code paths." + "reason": "VAE NoneType addition bug vs torch.compile/group offloading incompatibility." }, { "left": "issue:1158", "right": "issue:1524", "accept": false, - "reason": "Test isolation bug vs poor training quality report; not the same defect." + "reason": "Test interference in config vs poor textual inversion training quality." }, { "left": "issue:10302", "right": "issue:8989", "accept": false, - "reason": "FP8 inference noise vs sequential offloading bug; both performance-related but not the same issue." + "reason": "FP8 inference noise vs sequential offloading bug on Stable Audio." }, { "left": "issue:6079", "right": "issue:7389", "accept": false, - "reason": "Generic scalar half/float mismatch vs StableDiffusionXLPipeline.from_single_file error; different symptoms and fixes." + "reason": "Generic dtype mismatch vs a from_single_file loading error; not the same bug." }, { "left": "issue:7751", "right": "issue:8939", "accept": false, - "reason": "Animatediff motion-module conversion failure vs tokenizer loading error; unrelated workflows." + "reason": "Animatediff conversion errors vs SD3 tokenizer loading failure." }, { "left": "issue:10590", "right": "issue:11137", "accept": false, - "reason": "Sana tiled-VAE BF16 failure vs Latte dtype mismatch; both dtype-related but different pipelines and failures." + "reason": "Different pipeline failures: Sana BF16 tiled VAE decode vs Latte dtype mismatch." }, { "left": "issue:12239", "right": "issue:13191", "accept": false, - "reason": "Two separate support requests for different models; no shared underlying bug." + "reason": "Support requests for different models; not the same concrete change." }, { "left": "issue:1263", "right": "issue:305", "accept": false, - "reason": "Both are import-related, but one is a specific safety checker import path issue and the other is a generic diffusers import error." + "reason": "Both are import errors, but different modules and root causes." }, { "left": "issue:3537", "right": "issue:6141", "accept": false, - "reason": "A conversion script request vs ControlNet parameter support; different changes entirely." + "reason": "ControlNet conversion script vs adding ControlNet parameters; unrelated changes." }, { "left": "issue:5371", "right": "issue:6331", "accept": false, - "reason": "BLIP inpainting support request vs SAG for inpainting; same broad area, different features." + "reason": "Different inpainting feature requests: BLIP support vs SAG for inpainting." }, { "left": "issue:8323", "right": "issue:9111", "accept": false, - "reason": "Scheduler file recognition vs missing `scheduler` attribute on BitDiffusion; different bugs." + "reason": "Scheduler file recognition vs missing scheduler attribute; different failures." }, { "left": "issue:10542", "right": "issue:8850", "accept": false, - "reason": "Hunyuan batch-size regression vs community pipeline mirror issue; unrelated." + "reason": "Hunyuan batch-size regression vs community pipeline mirror breakage." }, { "left": "issue:12458", "right": "issue:13277", "accept": false, - "reason": "Qwen-image-edit batch inference request vs QwenImagePipeline batch failure with Ulysses SP; related keywords but not the same concrete bug." + "reason": "Both mention batch inputs, but one is a feature/request and the other a QwenImagePipeline failure." }, { "left": "issue:6055", "right": "issue:7210", "accept": false, - "reason": "Generic accelerate error vs `named_parameters` missing on StableDiffusionXLPipeline with accelerate dispatch; not the same root cause." + "reason": "Too broad: generic accelerate error vs a specific named_parameters dispatch issue." }, { "left": "issue:4061", "right": "issue:5328", "accept": false, - "reason": "LoRA load error on SDXL vs LoRA models all failing in a later report; too broad and not clearly the same bug." + "reason": "Different LoRA complaints on different models; no clear shared code-path." }, { "left": "issue:13176", "right": "issue:13506", "accept": false, - "reason": "General UNet attention error vs Sage attention/ring-attention incompatibility; different attention backends and failures." + "reason": "Different attention backends and different error contexts; not one bug." }, { "left": "issue:3884", "right": "issue:8594", "accept": false, - "reason": "Training script error in text-to-image LoRA vs unbound tokenizer variable in DreamBooth LoRA SD3; different scripts and bugs." + "reason": "Different training scripts and failure sites; not mergeable as one fix." }, { "left": "issue:4002", "right": "issue:4005", "accept": false, - "reason": "Missing `from_single_file` API vs aspect-ratio preservation; unrelated Stable Diffusion inpaint issues." + "reason": "One is missing from_single_file support, the other is aspect-ratio behavior." }, { "left": "issue:4354", "right": "issue:5390", "accept": false, - "reason": "`from_single_file` returning wrong pipeline type vs unwanted text encoder download with `local_files_only=True`; different loading problems." + "reason": "Different SDXL issues: wrong pipeline type vs unwanted text-encoder download." }, { "left": "issue:10672", "right": "issue:12695", "accept": false, - "reason": "Callback hook support request vs Hunyuan v1.5 support request; separate feature requests." - }, - { - "left": "issue:12359", - "right": "issue:12575", - "accept": false, - "reason": "Documentation bug about `guidance_scale` vs attention-backend-check runtime error; not duplicates." - }, - { - "left": "issue:2737", - "right": "issue:5079", - "accept": false, - "reason": "Scheduler feature support vs LoRA loading support; distinct enhancements." + "reason": "Callback API support request vs Hunyuan v1.5 model support; unrelated." } ] }, @@ -59295,8 +59297,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -59308,17 +59310,17 @@ "issue:11033", "issue:12022", "issue:12085", + "issue:12359", + "issue:12575", "issue:12605", "issue:13104", - "issue:1962", - "issue:2202", "issue:2364", "issue:2392", + "issue:2737", "issue:2769", "issue:3064", "issue:3580", "issue:3585", - "issue:3817", "issue:4014", "issue:4043", "issue:4122", @@ -59326,7 +59328,7 @@ "issue:4496", "issue:4646", "issue:5028", - "issue:5168", + "issue:5079", "issue:5233", "issue:5433", "issue:5465", @@ -59350,6 +59352,8 @@ "issue:9936" ], "soft_pairs": [ + "issue:12359|issue:12575", + "issue:2737|issue:5079", "issue:3064|issue:3585", "issue:4496|issue:843", "issue:4646|issue:8604", @@ -59372,188 +59376,186 @@ "issue:2769|issue:4453", "issue:3580|issue:5028", "issue:2392|issue:4043", - "issue:6497|issue:7408", - "issue:1962|issue:3817", - "issue:2202|issue:5168" + "issue:6497|issue:7408" ], - "prepared_review_unit_hash": "2993e376186525023dcd361154ef6e26f751818eb15fe7423b5759bb8e8dd7b8", + "prepared_review_unit_hash": "8451a17120db01b1a01760c9a461f275faff27aa1c85b6e8480accceac0a67a2", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33825, - "estimated_input_tokens": 8457, - "estimated_eval_tokens": 17170 + "serialized_chars": 33922, + "estimated_input_tokens": 8481, + "estimated_eval_tokens": 17218 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly false-positive similarity across unrelated bug reports in diffusers (Flux, SDXL, training, attention, GGUF, docs). I found no true duplicate pairs and no PRs.", - "confidence": 0.18, - "canonical_issue_reason": "No single canonical issue: the set is heterogeneous and the soft matches are mostly title-level resemblance rather than the same underlying bug.", + "summary": "The soft pairs are mostly broad thematic matches at best, but not true duplicates: they span different models, pipelines, training scripts, and failure modes. I would reject all soft edges and treat this as a heterogeneous set rather than one dedupe cluster.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:10674 is the clearest self-contained bug report and the best cluster anchor, but it still does not duplicate the other issues.", + "best_issue_reason": "issue:10674 is the closest representative only because it is a concrete FluxPipeline runtime/loading bug, but the overall set is too heterogeneous to have a real canonical issue.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12359", + "right": "issue:12575", + "accept": false, + "reason": "One is a Chroma docs bug about guidance_scale; the other is a Chroma attention-backend runtime error. Different problem and fix." + }, + { + "left": "issue:2737", + "right": "issue:5079", + "accept": false, + "reason": "Scheduler CLIP guidance support and LoRA loading support are unrelated features/issues." + }, { "left": "issue:3064", "right": "issue:3585", "accept": false, - "reason": "Both involve LoRA-related loading/training, but one is a .safetensors loading issue and the other is a failed synthesis/use case on SD1.5; different bugs." + "reason": "Both mention LoRA, but one is about loading .safetensors LoRAs and the other about a synthesized LoRA model failing. Different bug paths." }, { "left": "issue:4496", "right": "issue:843", "accept": false, - "reason": "Prompt fusion is a feature/request, while prompt syntax documentation is a docs issue; not the same bug or change." + "reason": "Prompt fusion feature request is unrelated to prompt syntax documentation." }, { "left": "issue:4646", "right": "issue:8604", "accept": false, - "reason": "ControlNet-to-ONNX conversion failure and T5 fp16 corruption are distinct model/export problems." + "reason": "ControlNet ONNX conversion failure vs T5 fp16 corruption are different code paths and failure modes." }, { "left": "issue:5465", "right": "issue:9001", "accept": false, - "reason": "Different pipelines and failure modes: DiffEdit inpaint_strength handling vs Auraflow aspect-ratio support." + "reason": "DiffEdit inpaint_strength behavior and Auraflow aspect ratio handling are unrelated." }, { "left": "issue:12605", "right": "issue:9904", "accept": false, - "reason": "One is backend-specific attn_mask dropping, the other is Flux image-shape rounding; unrelated behavior." + "reason": "Attention-mask backend bug and Flux image-shape rounding are distinct issues." }, { "left": "issue:10111", "right": "issue:10591", "accept": false, - "reason": "Both are vague 'bugs' in training/pipeline scripts, but they concern different models and code paths." + "reason": "Both are general bug reports, but FLUX pipeline bugs and SD3 LoRA training-script bugs are not the same underlying issue." }, { "left": "issue:10302", "right": "issue:9514", "accept": false, - "reason": "Both mention FP8/Flux, but one is inference noise without CPU offloading and the other is LoRA weight loading after quantization; not the same bug." + "reason": "FP8 inference noise without CPU offloading is not the same as LoRA loading after FP8 quantization." }, { "left": "issue:5540", "right": "issue:9170", "accept": false, - "reason": "Generic out-of-memory reports are not the same as a GPU-memory requirement question; different underlying issue." + "reason": "Both are memory-related, but generic OOM and a specific SDXL+ControlNet memory requirement report are not duplicates." }, { "left": "issue:6486", "right": "issue:8708", "accept": false, - "reason": "Embedding retraction occurring twice and a missing shift_factor in SD3 examples are unrelated DreamBooth bugs." + "reason": "Dreambooth embedding retraction bug and missing SD3 shift_factor in examples are unrelated training issues." }, { "left": "issue:8535", "right": "issue:8593", "accept": false, - "reason": "XFormers attention support and bfloat16 CPU noise are different SD3 issues." + "reason": "xFormers attention enablement and bfloat16 CPU noise are different SD3 problems." }, { "left": "issue:10674", "right": "issue:12085", "accept": false, - "reason": "Both mention GGUF loading, but they affect different models/pipelines and likely different code paths." + "reason": "Both involve GGUF loading, but they affect different model classes and likely different loader paths; too broad to merge as one bug." }, { "left": "issue:4122", "right": "issue:5233", "accept": false, - "reason": "One is scheduler timestep calculation, the other is callback step parameter value; related theme but not the same defect." + "reason": "Timesteps mismatch in SDXL pipelines and callback step-value bug are related only at a very high level." }, { "left": "issue:9063", "right": "issue:9904", "accept": false, - "reason": "Tokenizer instantiation failure and silent image-shape rounding are unrelated FluxPipeline bugs." + "reason": "Tokenizer instantiation failure and image-shape rounding are unrelated FluxPipeline issues." }, { "left": "issue:2364", "right": "issue:5951", "accept": false, - "reason": "Both are training issues, but they refer to different scripts/models and different failures." + "reason": "Unable to run training script is far too vague and does not match the PixArt training issue specifically." }, { "left": "issue:9163", "right": "issue:9936", "accept": false, - "reason": "Tensor bugs in train_dreambooth_lora_flux and NCCL timeouts in train_controlnet_flux are separate problems." + "reason": "Both are Flux training issues, but tensor bugs and NCCL timeout in multigpu training are not the same concrete failure." }, { "left": "issue:11033", "right": "issue:13104", "accept": false, - "reason": "from_single_file loading failure and a missing logger definition are different errors." + "reason": "SD1.5 from_single_file loading failure and a logger-not-defined error are unrelated." }, { "left": "issue:10315", "right": "issue:7056", "accept": false, - "reason": "CogVideo training error and SDXL LoRA fine-tuning with train_text_encoder are unrelated training bugs." + "reason": "CogVideo training error and SDXL LoRA fine-tuning/train_text_encoder error are different models and code paths." }, { "left": "issue:12022", "right": "issue:12605", "accept": false, - "reason": "FlashAttention-3 interface incompatibility and attn_mask being ignored are distinct attention/backend bugs." + "reason": "Latest flash-attention interface incompatibility and attn_mask being ignored are separate attention backend bugs." }, { "left": "issue:4014", "right": "issue:5433", "accept": false, - "reason": "Both mention visual artifacts, but the causes differ: watermark artifacts vs DPM++/SDXL scheduler artifacts." + "reason": "Both mention image artifacts, but watermark artifacts and DPM++/no-refiner artifacts are different causes." }, { "left": "issue:2769", "right": "issue:4453", "accept": false, - "reason": "Model round-trip corruption and SDXL LoRA conversion guidance are not the same issue." + "reason": "Round-tripping ckpt/safetensors conversion and converting SDXL LoRA to safetensors are different workflows." }, { "left": "issue:3580", "right": "issue:5028", "accept": false, - "reason": "XFormers TypeError and an unexpected 'scale' kwarg in a hacked downblock forward are separate code-path failures." + "reason": "XFormers TypeError and hacked DownBlock scale-arg TypeError are different runtime errors." }, { "left": "issue:2392", "right": "issue:4043", "accept": false, - "reason": "Both concern prompt embeddings, but one is an img2img prompt=None validation error and the other is an SDXL requirement for pooled embeddings." + "reason": "Img2Img prompt_embeds validation and SDXL pooled_prompt_embeds requirement are related API areas but not the same bug." }, { "left": "issue:6497", "right": "issue:7408", "accept": false, - "reason": "Different projects/models entirely: DragNUWA versus FRESCO." - }, - { - "left": "issue:1962", - "right": "issue:3817", - "accept": false, - "reason": "Video generation with Stable Diffusion and editable dance generation from music are different feature requests." - }, - { - "left": "issue:2202", - "right": "issue:5168", - "accept": false, - "reason": "OpenVINO pipeline support and ControlNet-XS support are distinct feature requests." + "reason": "DragNUWA and FRESCO are unrelated pipeline/model issues." } ] }, @@ -59572,8 +59574,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -59588,11 +59590,12 @@ "issue:12239", "issue:13191", "issue:1358", - "issue:1405", - "issue:2340", + "issue:1962", + "issue:2202", "issue:2440", "issue:2905", "issue:3040", + "issue:3817", "issue:4061", "issue:4471", "issue:4736", @@ -59600,14 +59603,13 @@ "issue:5002", "issue:5043", "issue:5162", + "issue:5168", "issue:5256", "issue:5285", "issue:5325", "issue:539", - "issue:5395", "issue:5515", "issue:6094", - "issue:6333", "issue:6486", "issue:660", "issue:6749", @@ -59627,6 +59629,8 @@ "issue:9420" ], "soft_pairs": [ + "issue:1962|issue:3817", + "issue:2202|issue:5168", "issue:6486|issue:7600", "issue:7300|issue:7970", "issue:4736|issue:9210", @@ -59648,182 +59652,180 @@ "issue:12221|issue:13191", "issue:1358|issue:5162", "issue:5043|issue:5285", - "issue:6749|issue:6906", - "issue:5395|issue:6333", - "issue:1405|issue:2340" + "issue:6749|issue:6906" ], - "prepared_review_unit_hash": "2b02c3838a6fa329f6420f9e785de76b45488120f7ee5bdc7f6476ec308b32b0", + "prepared_review_unit_hash": "ce327b187f9c4d94f93e81a7f855bacf08eb825f789c9fcf21fa48edbfc9eeff", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33373, - "estimated_input_tokens": 8344, - "estimated_eval_tokens": 16944 + "serialized_chars": 33300, + "estimated_input_tokens": 8325, + "estimated_eval_tokens": 16906 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly unrelated issues that only share broad recurring themes like loading, LoRA, schedulers, or training examples. I did not find any true duplicate pair among the soft candidates. No PRs are present.", - "confidence": 0.95, - "canonical_issue_reason": "issue:10512 is the most substantial and discussion-heavy report in the set, with a concrete reproducible loading failure; it is the best single representative only because the cluster is otherwise heterogeneous.", + "summary": "This cluster is not a true duplicate set: it mixes unrelated feature requests, training/runtime bugs, and documentation/support items across many different pipelines and model families. All soft edges are false positives.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits the cluster because the items are semantically diverse and do not describe the same underlying bug or feature request.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10512 is the strongest issue to anchor on if one must pick a canonical item: it has the richest troubleshooting context and highest engagement, while the rest are mostly unrelated feature requests or distinct bugs.", + "best_issue_reason": "If one issue must stand in as a loose representative, issue 10147 is the broadest current open feature request, but it still does not cover the rest of the cluster well.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1962", + "right": "issue:3817", + "accept": false, + "reason": "Generic video generation with Stable Diffusion vs editable dance generation from music; different use cases and code paths." + }, + { + "left": "issue:2202", + "right": "issue:5168", + "accept": false, + "reason": "OpenVINO pipeline support and ControlNet-XS support are separate subsystem/model feature requests." + }, { "left": "issue:6486", "right": "issue:7600", "accept": false, - "reason": "Both are DreamBooth-related, but one is about embedding retraction happening twice and the other about missing accelerator.accumulate on text_encoder; different bug paths." + "reason": "Double embedding retraction in DreamBooth advanced vs not applying accelerator.accumulate in text_encoder; distinct training bugs." }, { "left": "issue:7300", "right": "issue:7970", "accept": false, - "reason": "Both mention memory usage, but one is about set_default_attn_processor in UNet and the other about enable_model_cpu_offload; not the same issue." + "reason": "Both mention memory, but one is about set_default_attn_processor() behavior and the other is CPU offload RAM growth; different bugs." }, { "left": "issue:4736", "right": "issue:9210", "accept": false, - "reason": "One is a training OOM/invalid-value problem in SDXL LoRA training; the other is a CLI argument validation error for image_column." + "reason": "Train LoRA OOM/invalid-cast failure vs dataset image_column validation error; unrelated failure modes." }, { "left": "issue:1101", "right": "issue:5256", "accept": false, - "reason": "Unrelated requests: adding a scheduler versus adding Insta-Flow training support." + "reason": "Scheduler feature addition vs Insta-Flow training support; separate changes." }, { "left": "issue:4061", "right": "issue:4471", "accept": false, - "reason": "Both are LoRA-loading issues in SDXL, but they affect different pipelines and fail for different reasons; not clearly the same bug." + "reason": "Both are SDXL LoRA load problems, but they target different pipeline APIs and specific load failures." }, { "left": "issue:5002", "right": "issue:539", "accept": false, - "reason": "Different API gaps entirely: missing use_karras_sigmas on schedulers versus snapshot_download argument compatibility." + "reason": "Missing scheduler arguments vs snapshot_download allow_patterns keyword error; unrelated API compatibility issues." }, { "left": "issue:5515", "right": "issue:8323", "accept": false, - "reason": "LCM pipeline initialization failure without a scheduler is unrelated to scheduler file recognition/identification." + "reason": "LCM pipeline initialization failure without scheduler vs scheduler file recognition error; different underlying problems." }, { "left": "issue:2440", "right": "issue:660", "accept": false, - "reason": "Different environments and failure modes: float16 inpainting load error versus MPS fp16 revision loading." + "reason": "Inpainting float16 load error vs MPS fp16 revision behavior; different environments and failure causes." }, { "left": "issue:3040", "right": "issue:6094", "accept": false, - "reason": "The first is a docs-copy/paste AttributeError, the second is a version regression in TextToVideoZeroPipeline after a diffusers update." + "reason": "Doc copy-paste AttributeError vs TextToVideoZero regression after a version update; unrelated issues." }, { "left": "issue:4752", "right": "issue:5325", "accept": false, - "reason": "SDXL mixed-precision tensor mismatch versus packaging type information distribution; no shared underlying bug." + "reason": "SDXL dtype mismatch in pipeline code vs packaging type information distribution; no shared bug." }, { "left": "issue:11581", "right": "issue:9293", "accept": false, - "reason": "Both are loading/type-related, but one is a CLIPVisionModel dtype kwarg problem in FLUX IP-Adapter loading and the other is a missing Float8 storage backend." + "reason": "FLUX IP-Adapter dtype init error vs missing Float8 storage object; both dtype-related but different components and code paths." }, { "left": "issue:7311", "right": "issue:8894", "accept": false, - "reason": "Different training scripts and failure classes; no evidence they share the same root cause." + "reason": "Checkpoint save failure in train_text_to_image_sdxl.py vs a separate training script error after a new commit." }, { "left": "issue:6774", "right": "issue:7611", "accept": false, - "reason": "Mixed-precision xformers dtype mismatch is unrelated to a tensor size mismatch in unet2dconditional." + "reason": "AttentionProcessor mixed-precision mismatch vs UNet tensor size mismatch; distinct runtime errors." }, { "left": "issue:10100", "right": "issue:8410", "accept": false, - "reason": "Both are feature requests to add model/pipeline support, but they target different models and code paths." + "reason": "ConsisID support and native TextDiffuser support are different model integrations." }, { "left": "issue:10147", "right": "issue:12239", "accept": false, - "reason": "Different community-model support requests: MV-Adapter multi-view generation versus InfiniteTalk." + "reason": "MV-Adapter multi-view generation vs InfiniteTalk support; separate feature requests." }, { "left": "issue:1037", "right": "issue:2905", "accept": false, - "reason": "Completely different requests: adding a community pipeline versus adding Karras sigmas to schedulers." + "reason": "Community pipeline for Stable Diffusion Telephone vs Karras sigmas scheduler support; unrelated additions." }, { "left": "issue:10512", "right": "issue:9420", "accept": false, - "reason": "Both involve FLUX LoRA loading/saving, but one is about a Quanto-quantized LoRA not loading and the other about a saved Hyper-SD LoRA model failing to load; too different to be the same bug." + "reason": "Quanto Flux LoRA load failure vs transformer load failure after FLUX+Hyper-SD save; related to FLUX, but different concrete bugs." }, { "left": "issue:11005", "right": "issue:8588", "accept": false, - "reason": "A docstring/default mismatch is unrelated to a missing conversion script." + "reason": "Docstring/default mismatch in pipeline_wan_i2v.py vs missing SD3 conversion script; unrelated." }, { "left": "issue:12221", "right": "issue:13191", "accept": false, - "reason": "Separate model-support feature requests for DiffSynth ControlNet and Elastic-DiT." + "reason": "DiffSynth ControlNet support vs Elastic-DiT support; different model support requests." }, { "left": "issue:1358", "right": "issue:5162", "accept": false, - "reason": "Safetensors format support and a folder-to-safetensors conversion script are related in topic but not the same change or bug." + "reason": "Safetensors support in the library vs a conversion script to safetensors; related topic, but not the same issue." }, { "left": "issue:5043", "right": "issue:5285", "accept": false, - "reason": "A training feature request for W\u00fcrstchen is unrelated to adding Japanese documentation translation." + "reason": "W\u00fcrstchen LoRA finetuning vs Japanese docs translation; entirely different scopes." }, { "left": "issue:6749", "right": "issue:6906", "accept": false, - "reason": "Both are broken examples, but they concern different pipelines and failure modes; not the same underlying bug." - }, - { - "left": "issue:5395", - "right": "issue:6333", - "accept": false, - "reason": "One is a specific PEFT LoRA-combination bug, the other is a request to finetune a pre-trained LoRA by PEFT; not duplicates." - }, - { - "left": "issue:1405", - "right": "issue:2340", - "accept": false, - "reason": "Attention slicing regression and tiny SD safety-checker error are distinct issues affecting different parts of the pipeline." + "reason": "InstantID community example reproducibility vs broken LoRA DreamBooth example; different examples and failures." } ] }, @@ -59842,13 +59844,11 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10350", - "issue:10382", "issue:10412", "issue:10636", "issue:10672", @@ -59856,17 +59856,17 @@ "issue:11555", "issue:11839", "issue:12107", - "issue:12108", "issue:1214", "issue:12470", "issue:12505", - "issue:12575", "issue:13067", + "issue:1405", "issue:1655", "issue:1791", "issue:1801", "issue:1896", "issue:1949", + "issue:2340", "issue:2392", "issue:2466", "issue:2728", @@ -59878,10 +59878,12 @@ "issue:4534", "issue:457", "issue:5368", + "issue:5395", "issue:541", "issue:5502", "issue:5595", "issue:6094", + "issue:6333", "issue:6363", "issue:6452", "issue:646", @@ -59897,6 +59899,8 @@ "issue:9799" ], "soft_pairs": [ + "issue:5395|issue:6333", + "issue:1405|issue:2340", "issue:8744|issue:8939", "issue:10412|issue:11225", "issue:1791|issue:457", @@ -59918,182 +59922,180 @@ "issue:1214|issue:851", "issue:2728|issue:2947", "issue:7606|issue:9799", - "issue:11555|issue:12505", - "issue:12108|issue:12575", - "issue:10350|issue:10382" + "issue:11555|issue:12505" ], - "prepared_review_unit_hash": "a53883ed0111273e5b7606b780cfadec16f8ad4beeff065b29dec78bcf61752e", + "prepared_review_unit_hash": "8a4af21d276fa053c960aa9898d16d18803a9c841864d78335b9d3240e6e33a8", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34124, - "estimated_input_tokens": 8531, - "estimated_eval_tokens": 17318 + "serialized_chars": 34022, + "estimated_input_tokens": 8506, + "estimated_eval_tokens": 17268 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated bug/feature issues that only look similar at the title level (memory, dtype, pipeline support, loading, and API errors). None of the proposed soft pairs look like true duplicates or one concrete code-path change.", - "confidence": 0.68, - "canonical_issue_reason": "Issue 10382 is the clearest concrete bug report with a specific code-path and failure mode, but the cluster overall is too heterogeneous to have a strong single canonical issue.", + "summary": "These items are mostly a heterogeneous set of unrelated bugs/features across different pipelines and training paths. I don\u2019t see any soft pair that clearly describes the same underlying issue, so all soft edges should be rejected.", + "confidence": 0.79, + "canonical_issue_reason": "There is no strong duplicate cluster here; if one issue must be used as an anchor, issue 10412 is the most detailed and self-contained report, but it does not subsume the others.", "canonical_pr_reason": null, - "best_issue_reason": "10382 is the best representative by specificity and diagnosability; several other issues are broader, more speculative, or about different subsystems entirely.", + "best_issue_reason": "Issue 10412 is the strongest standalone issue in the set because it has the clearest problem statement and most context; the rest are mostly unrelated or too broad to serve as a better canonical anchor.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5395", + "right": "issue:6333", + "accept": false, + "reason": "Both involve LoRA/PEFT, but one is about combining civitai LoRAs and the other about finetuning a pretrained LoRA; different problems and code paths." + }, + { + "left": "issue:1405", + "right": "issue:2340", + "accept": false, + "reason": "Attention slicing regression and a safety-checker failure are unrelated bugs; same project area, not the same issue." + }, { "left": "issue:8744", "right": "issue:8939", "accept": false, - "reason": "Both are about model loading/training workflows, but one is a training-setup request and the other is a tokenizer-loading failure for SD3; different failures and code paths." + "reason": "Loading from single-file training support and a tokenizer-loading error for SD3 are different failures." }, { "left": "issue:10412", "right": "issue:11225", "accept": false, - "reason": "Both mention high VRAM use, but one is DreamBooth training on SD3.5-Large and the other is ControlNet validation OOM; these are separate scenarios and likely different fixes." + "reason": "Both are OOM reports, but they concern different training examples, models, and triggering steps." }, { "left": "issue:1791", "right": "issue:457", "accept": false, - "reason": "Both involve MPS, but one is pipeline support on Mac/MPS and the other is blue-artifact rendering; different symptoms and likely unrelated root causes." + "reason": "Both mention MPS, but one is UnCLIP unsupported on Mac/MPS and the other is image artifact corruption; not the same bug." }, { "left": "issue:1896", "right": "issue:2392", "accept": false, - "reason": "Both are prompt/input handling errors, but one is CLIP-guided SD size/dict mismatch and the other is img2img with prompt_embeds rejecting None prompt; not the same bug." + "reason": "Different runtime errors in different pipelines: one is a size/type error in CLIP-guided SD, the other is prompt_embeds validation in img2img." }, { "left": "issue:3799", "right": "issue:8373", "accept": false, - "reason": "These are both type/error-message issues, but one is about inferred return types from from_pretrained and the other is an incorrect user-facing error message; no shared concrete defect." + "reason": "Type inference for `from_pretrained` and an incorrect error message are unrelated." }, { "left": "issue:5595", "right": "issue:6363", "accept": false, - "reason": "Both are training/runtime errors involving dimensions or fp16, but one is invalid image size divisibility and the other is FP16 gradient unscale failure; different code paths." + "reason": "One is an image-size divisibility validation error; the other is an fp16 gradient unscale failure during LoRA training." }, { "left": "issue:6094", "right": "issue:8552", "accept": false, - "reason": "Both concern pipeline crashes, but TextToVideoZero breaking after a version update and a segmentation fault in inference are not the same bug." + "reason": "Both involve TextToVideoZeroPipeline, but one is a version regression and the other is a segmentation fault during inference; too different to merge." }, { "left": "issue:1655", "right": "issue:3767", "accept": false, - "reason": "One is pipeline initialization failure and the other is FrozenDict mutability; unrelated issues despite both being low-level library behavior." + "reason": "Pipeline initialization failure and FrozenDict mutability semantics are unrelated." }, { "left": "issue:4492", "right": "issue:4534", "accept": false, - "reason": "Both show runtime errors in generated workflows, but one is a JSON/Python boolean NameError and the other is an MPS validation crash; clearly different defects." + "reason": "A JSON/boolean NameError and an SDXL DreamBooth MPS validation crash are different bugs." }, { "left": "issue:11839", "right": "issue:12107", "accept": false, - "reason": "One is a GGUF input-size problem for FluxKontextPipeline, the other is accelerator.init_trackers rejecting a custom object; no common underlying bug." + "reason": "GGUF loading for FluxKontextPipeline and `accelerator.init_trackers` with custom objects are unrelated." }, { "left": "issue:1949", "right": "issue:2466", "accept": false, - "reason": "Both relate to example/training scripts, but one is a general dreambooth tutorial error and the other is checkpoint limiting not working; different functionality." + "reason": "DreamBooth tutorial training errors and checkpoint limit enforcement are different example-script issues." }, { "left": "issue:10672", "right": "issue:8482", "accept": false, - "reason": "One asks for callback_on_step_end support across pipelines, while the other is support for a specific model; feature request overlap is too broad to be a duplicate." + "reason": "Callback support for pipelines and SD3-medium model support are not the same change or bug." }, { "left": "issue:12470", "right": "issue:13067", "accept": false, - "reason": "A quantizer API enhancement and a feature request to add Anima support are unrelated changes." + "reason": "Quantizer API additions and adding Anima model support are unrelated feature requests." }, { "left": "issue:1801", "right": "issue:646", "accept": false, - "reason": "Both are attribute/API errors, but for different objects and different missing methods; not the same bug." + "reason": "`getattr` misuse and a missing `set_format` method on PNDMScheduler are different attribute errors." }, { "left": "issue:3698", "right": "issue:541", "accept": false, - "reason": "One is a request to convert an inpainting model to diffusers, the other is a SageMaker issue; no shared defect or change." + "reason": "Converting an inpainting model and running Diffusers on SageMaker are unrelated requests." }, { "left": "issue:5368", "right": "issue:5502", "accept": false, - "reason": "Both involve fp16, but one is a dtype mismatch during attention and the other is train_custom_diffusion.py not supporting fp16; different fix targets." + "reason": "Both mention fp16, but one is a dtype mismatch in attention and the other is lack of fp16 support in a training script." }, { "left": "issue:6452", "right": "issue:8780", "accept": false, - "reason": "Both mention model/input format mismatches, but one is adding masked_image_latents to a ControlNet inpaint pipeline and the other is a general dtype mismatch complaint." + "reason": "Adding masked image latents to ControlNet inpaint and a generic model/input dtype mismatch are not the same issue." }, { "left": "issue:10636", "right": "issue:8604", "accept": false, - "reason": "Both mention bugs in model/sampling code, but value_guided_sampling.py and T5 fp16 corruption are different components and failure modes." + "reason": "A bug in `value_guided_sampling.py` and corrupted output from converting the T5 encoder to float16 are unrelated." }, { "left": "issue:1214", "right": "issue:851", "accept": false, - "reason": "Both are dtype/half-precision problems, but CLIPTextModel half precision and a generic float-vs-half mismatch are not enough to conclude the same underlying bug." + "reason": "Both are half-precision/dtype complaints, but the reports are too vague and likely stem from different code paths to treat as duplicates." }, { "left": "issue:2728", "right": "issue:2947", "accept": false, - "reason": "Both are model-loading-related attribute/model-class issues, but safetensors loading and missing load_attn_procs are different API problems." + "reason": "Cannot load safetensors in `from_pretrained` and missing `load_attn_procs` are distinct model-loading/API issues." }, { "left": "issue:7606", "right": "issue:9799", "accept": false, - "reason": "Different model families and different error shapes; one is a deprecated LoRA method issue, the other is a tensor-size mismatch." + "reason": "A deprecated LoRA processor in Kandinsky 2.2 and a tensor-size mismatch runtime error are unrelated." }, { "left": "issue:11555", "right": "issue:12505", "accept": false, - "reason": "One is a general request for device_map=\"auto\" support in pipelines, the other is a tgate error in img2img pipelines; different features and failure conditions." - }, - { - "left": "issue:12108", - "right": "issue:12575", - "accept": false, - "reason": "Both involve pipeline/backend checks, but one is a scheduler flow-matching breakage and the other is attention backend checks causing an error; not the same fix." - }, - { - "left": "issue:10350", - "right": "issue:10382", - "accept": false, - "reason": "Both are VAE-related, but one is decoder memory usage in SD2.1 and the other is an encoder return-type mismatch causing AttributeError; distinct bugs." + "reason": "`device_map=\"auto\"` support and a tgate error in img2img pipelines are different topics." } ] }, @@ -60112,13 +60114,15 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10100", "issue:10178", + "issue:10350", + "issue:10382", "issue:10512", "issue:10887", "issue:10896", @@ -60128,13 +60132,12 @@ "issue:11687", "issue:11779", "issue:11922", - "issue:1197", - "issue:1205", + "issue:12108", + "issue:12575", "issue:12624", "issue:12738", "issue:13277", "issue:1358", - "issue:1440", "issue:2195", "issue:2266", "issue:2838", @@ -60143,7 +60146,6 @@ "issue:4014", "issue:4269", "issue:4406", - "issue:459", "issue:4790", "issue:5538", "issue:5628", @@ -60167,6 +60169,8 @@ "issue:9690" ], "soft_pairs": [ + "issue:12108|issue:12575", + "issue:10350|issue:10382", "issue:10178|issue:6570", "issue:10887|issue:11569", "issue:11687|issue:12624", @@ -60188,188 +60192,186 @@ "issue:6513|issue:6866", "issue:8577|issue:9101", "issue:3960|issue:4406", - "issue:10512|issue:9690", - "issue:1197|issue:459", - "issue:1205|issue:1440" + "issue:10512|issue:9690" ], - "prepared_review_unit_hash": "042ba5bfd332d5668102fda34260992ddaea44c2314d71d5b0c00a8bf338bd30", + "prepared_review_unit_hash": "e3a9ff55bc6ae657bc6cf82d94f64eede679eda4ff85755b879a8425caefa803", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33940, - "estimated_input_tokens": 8485, - "estimated_eval_tokens": 17226 + "serialized_chars": 34160, + "estimated_input_tokens": 8540, + "estimated_eval_tokens": 17336 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly unrelated feature requests and bug reports; the only likely duplicate is the safetensors/LoRA HeaderTooLarge pair. No PRs are present.", - "confidence": 0.79, - "canonical_issue_reason": "Issue #4406 is the clearest representative of the only true duplicate-like pair: it states the full safetensors HeaderTooLarge error explicitly and is more specific than #3960.", + "summary": "This cluster is mostly heterogeneous: feature requests, unrelated pipeline bugs, and a few load/shape/runtime errors. There is no cluster-wide duplicate set; the only clear duplicate-looking pair is the safetensors HeaderTooLarge LoRA reports.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue cleanly represents the whole cluster because the items span many unrelated bugs/features. If one issue must be chosen, issue:4406 is the best canonical for the only true duplicate pair because it states the exact safetensors HeaderTooLarge failure most explicitly.", "canonical_pr_reason": null, - "best_issue_reason": "#4406 is the best issue anchor for this cluster because it has the most explicit, canonical error signature; otherwise the remaining issues are too heterogeneous to share one representative.", + "best_issue_reason": "issue:4406 is the strongest issue candidate: it is concrete, specific, and matches another near-identical report (issue:3960) about the same LoRA safetensors HeaderTooLarge error.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12108", + "right": "issue:12575", + "accept": false, + "reason": "Different failures in Chroma/Qwen image pipelines: flow-matching scheduler behavior vs attention-backend checks." + }, + { + "left": "issue:10350", + "right": "issue:10382", + "accept": false, + "reason": "Both are VAE-related, but one is decoder memory use and the other is an encoder return-type AttributeError." + }, { "left": "issue:10178", "right": "issue:6570", "accept": false, - "reason": "Both are tensor shape errors, but in different pipelines and with different failure contexts." + "reason": "Different tensor-shape bugs in different modules/models; only superficial size-mismatch similarity." }, { "left": "issue:10887", "right": "issue:11569", "accept": false, - "reason": "Import/export support bug vs feature request for a model repo; unrelated." + "reason": "Import error for StableDiffusionXLTrainer vs a feature request for a model repo; unrelated." }, { "left": "issue:11687", "right": "issue:12624", "accept": false, - "reason": "Docs/framework addition request vs pipeline API change request; not the same change." + "reason": "Docs/framework addition request vs a new pipeline parameter request; not the same bug or change." }, { "left": "issue:5849", "right": "issue:6374", "accept": false, - "reason": "Trainer script argument error vs LoRA weight loading issue; different bugs." + "reason": "Training-script argparse TypeError vs LoRA weight loading question; different code paths and symptoms." }, { "left": "issue:1358", "right": "issue:2838", "accept": false, - "reason": "General safetensors support request vs ControlNet pipeline behavior request; unrelated." + "reason": "Generic safetensors support request vs a ControlNet pipeline feature request; unrelated." }, { "left": "issue:7267", "right": "issue:9680", "accept": false, - "reason": "Local data loading breakage vs unconditional generation regression; different code paths." + "reason": "Broken local data loading in train_controlnet.py vs SDXL unconditional generation bug; different problems." }, { "left": "issue:4790", "right": "issue:6016", "accept": false, - "reason": "Both involve conversion scripts, but the reported failures are different." + "reason": "Both involve conversion utilities, but one is round-trip file compatibility and the other is a torch.Size formatting/error case." }, { "left": "issue:10986", "right": "issue:11779", "accept": false, - "reason": "MPS ComplexDouble conversion bug vs Wan VACE mask padding bug; unrelated." + "reason": "MPS complex-dtype conversion error in transformer pipeline vs incorrect mask padding in Wan VACE; different pipelines and bugs." }, { "left": "issue:7936", "right": "issue:8968", "accept": false, - "reason": "Local safetensors JSON parsing failure vs generic safetensors loading support; not the same defect." + "reason": "Invalid JSON/header parsing for a local safetensors file vs general pipeline safetensors loading support." }, { "left": "issue:2195", "right": "issue:2883", "accept": false, - "reason": "Video feature request topics overlap only broadly, not as the same change." + "reason": "Both are video-related feature requests, but they target different pipeline additions and are not the same change." }, { "left": "issue:2266", "right": "issue:4269", "accept": false, - "reason": "Performance regression from CPU offload vs input validation TypeError; different issues." + "reason": "Sequential CPU offload slowdown vs ControlNet input-validation TypeError; different concrete issues." }, { "left": "issue:5538", "right": "issue:5731", "accept": false, - "reason": "Upscale pipeline latent handling vs AutoencoderKL logvar implementation; different underlying problems." + "reason": "Upscale pipeline latent-usage question vs missing trainable logvar in AutoencoderKL decoder; unrelated code paths." }, { "left": "issue:8624", "right": "issue:8939", "accept": false, - "reason": "Shared SD3 model context, but one is a missing symbol and the other is tokenizer loading." + "reason": "Undefined CUDA symbol in Stable Diffusion 3 setup vs tokenizer-loading OSError; different runtime failures." }, { "left": "issue:12738", "right": "issue:13277", "accept": false, - "reason": "Both mention batch inputs, but they affect different pipelines and failure modes." + "reason": "Both mention batching, but one is ZImage batch-size assertion and the other is QwenImage Ulysses SP batch handling." }, { "left": "issue:4014", "right": "issue:5824", "accept": false, - "reason": "Watermark artifacts vs SDXL add_time_ids default issue; unrelated." + "reason": "SDXL watermark artifacts vs incorrect text_encoder_projection_dim default; unrelated bugs." }, { "left": "issue:5628", "right": "issue:7640", "accept": false, - "reason": "Scheduler offset behavior vs from_single_file dimension error; different bugs." + "reason": "Scheduler denoising range offset vs from_single_file dimension runtime error; different root causes." }, { "left": "issue:10100", "right": "issue:10896", "accept": false, - "reason": "Two separate model-support feature requests; not one underlying change." + "reason": "Two different model-support requests (ConsisID vs SkyReels-A1); not the same feature." }, { "left": "issue:10954", "right": "issue:11922", "accept": false, - "reason": "Both FLUX-related, but one is a LoRA load failure and the other is a running problem for a different model." + "reason": "FLUX LoRA loading failure vs FLUX.1-Fill-dev runtime problem; same family name but different concrete bugs." }, { "left": "issue:6513", "right": "issue:6866", "accept": false, - "reason": "Different adapter/pipeline support requests, not the same feature." + "reason": "Different adapter integrations: T2IAdapter model support vs IP-Adapter support for SDXL ControlNet inpaint." }, { "left": "issue:8577", "right": "issue:9101", "accept": false, - "reason": "Different SD3 feature requests: differential diffusion vs InstructPix2Pix training support." + "reason": "Differential diffusion feature for SD3 vs an InstructPix2Pix training script for SD3; different work items." }, { "left": "issue:3960", "right": "issue:4406", "accept": true, - "reason": "Same safetensors/HeaderTooLarge loading failure; the titles and problem statements align closely enough to be duplicates." + "reason": "Both report the same safetensors HeaderTooLarge problem while loading a LoRA; 4406 is the clearer description." }, { "left": "issue:10512", "right": "issue:9690", "accept": false, - "reason": "LoRA load problem vs FP8 example cleanup request; unrelated." - }, - { - "left": "issue:1197", - "right": "issue:459", - "accept": false, - "reason": "Both are community ideas, but they describe different pipelines and goals." - }, - { - "left": "issue:1205", - "right": "issue:1440", - "accept": false, - "reason": "Model conversion to ONNX vs deleting cached models; unrelated." + "reason": "LoRA loading failure in Flux/Quanto vs FP8 example-code cleanup; unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is conservative and grounded in the packet. The only accepted soft edge (#3960/#4406) is well supported by the matching HeaderTooLarge safetensors error, and the other pairs are appropriately rejected without overclaiming. No PRs are present, so the PR-related fields are fine as null." + "feedback": "The summary is well grounded in the packet and stays conservative. The one accepted soft-edge pair, issue:3960 and issue:4406, is supported by the matching safetensors HeaderTooLarge LoRA reports. The remaining pair verdicts are appropriately non-committal/negative given the evidence. No overstatement detected." }, "evaluator_used": true, "retried": false, @@ -60385,22 +60387,23 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10054", "issue:10336", "issue:10542", "issue:10639", "issue:11519", "issue:1181", + "issue:1197", + "issue:1205", "issue:12060", "issue:12239", - "issue:12606", "issue:13430", "issue:1350", + "issue:1440", "issue:1594", "issue:193", "issue:1962", @@ -60410,12 +60413,12 @@ "issue:318", "issue:3325", "issue:3444", - "issue:3623", "issue:3698", "issue:3891", "issue:3975", "issue:4014", "issue:4232", + "issue:459", "issue:4674", "issue:4788", "issue:4949", @@ -60426,7 +60429,6 @@ "issue:5886", "issue:5960", "issue:6136", - "issue:671", "issue:6935", "issue:7569", "issue:7661", @@ -60440,6 +60442,8 @@ "issue:9680" ], "soft_pairs": [ + "issue:1197|issue:459", + "issue:1205|issue:1440", "issue:12060|issue:9599", "issue:193|issue:3975", "issue:4674|issue:4788", @@ -60461,182 +60465,180 @@ "issue:4014|issue:8723", "issue:6136|issue:7569", "issue:5886|issue:7661", - "issue:10336|issue:9450", - "issue:3623|issue:671", - "issue:10054|issue:12606" + "issue:10336|issue:9450" ], - "prepared_review_unit_hash": "90f81c14965303532eba22f0515bd5f34603d608a866e0e3ab61f4d20b07f430", + "prepared_review_unit_hash": "cb1ccf974fcd160bf941fb2d7d55ac5fcdf55ccb9febcffd67704bf1b56c5532", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33327, - "estimated_input_tokens": 8332, - "estimated_eval_tokens": 16920 + "serialized_chars": 33161, + "estimated_input_tokens": 8291, + "estimated_eval_tokens": 16838 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated bug reports, model-support requests, training issues, and conversion/tooling questions. I do not see any true duplicate pair among the soft edges, so all candidate links should be rejected.", - "confidence": 0.21, - "canonical_issue_reason": "If a single issue must anchor this noisy cluster, #193 ('Pipeline error') is the broadest and most generic report, but it is not a strong true canonical because the rest of the items span many unrelated problems.", + "summary": "This cluster is heterogeneous: it mixes unrelated model-support requests, training/inference feature asks, and distinct bug reports across many different pipelines. The soft pairs are mostly only superficially similar by wording and should all be rejected as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "#193 is the least specific, most catch-all issue in the set, so it is the only plausible representative anchor; however, it is still not a real duplicate target for the surrounding items.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1197", + "right": "issue:459", + "accept": false, + "reason": "Different requests: OpenAI Diffusion Pipeline vs Incremental Diffusion. Similar theme, not the same feature or bug." + }, + { + "left": "issue:1205", + "right": "issue:1440", + "accept": false, + "reason": "ONNX conversion request and deleting cached models are unrelated." + }, { "left": "issue:12060", "right": "issue:9599", "accept": false, - "reason": "Both are FLUX/DiT-adjacent, but one asks whether a DiT block exists and the other asks for a LoRA-only finetune example. Different requests, not the same underlying issue." + "reason": "A question about DiT blocks is not the same as asking for a LoRA-only FLUX finetune example." }, { "left": "issue:193", "right": "issue:3975", "accept": false, - "reason": "#193 is a generic pipeline failure report, while #3975 is a typo in an error message. Different code paths and fixes." + "reason": "Generic pipeline error report vs a typo in an error message; different problems." }, { "left": "issue:4674", "right": "issue:4788", "accept": false, - "reason": "One is a memory leak when changing width/height across calls; the other is an out-of-memory report for SDXL inpaint on 6 GB VRAM. Similar theme, different bug." + "reason": "Memory leak when changing width/height is distinct from SDXL inpaint OOM on 6 GB VRAM." }, { "left": "issue:8504", "right": "issue:9101", "accept": false, - "reason": "Both involve SD3 training, but one requests choosing which text encoders to train and the other requests an InstructPix2Pix training script. Not the same change." + "reason": "Selecting text encoders for SD3 training and an InstructPix2Pix SD3 training script are separate feature requests." }, { "left": "issue:10639", "right": "issue:12239", "accept": false, - "reason": "These are support requests for completely different models: Lumina-Image 2.0 vs InfiniteTalk." + "reason": "Support for Lumina-Image 2.0 and support for InfiniteTalk are different model integrations." }, { "left": "issue:10542", "right": "issue:9680", "accept": false, - "reason": "Different model families and failures: Hunyuan Video batch-size breakage versus SDXL unconditional generation issues." + "reason": "Hunyuan Video batch-size regression is unrelated to SDXL unconditional generation being broken." }, { "left": "issue:11519", "right": "issue:13430", "accept": false, - "reason": "Two separate model support requests (MAGI-1 vs JoyAI-Image-Edit), with no shared concrete bug or code path." + "reason": "Support requests for MAGI-1 and JoyAI-Image-Edit target different models." }, { "left": "issue:1181", "right": "issue:3698", "accept": false, - "reason": "Training a model from scratch is a different feature from converting an inpainting model to diffusers." + "reason": "Training from scratch is not the same as converting an inpainting model to diffusers." }, { "left": "issue:2210", "right": "issue:5160", "accept": false, - "reason": "Both mention LoRA, but one is about training the text encoder and the other about masking behavior in Dreambooth/LoRA. Different functionality." + "reason": "LoRA text-encoder training and Dreambooth/LoRA masking are different training features." }, { "left": "issue:2332", "right": "issue:4232", "accept": false, - "reason": "A Dreambooth inference error and an instruct-pix2pix/controlnet training+inference request are not the same underlying problem." + "reason": "Dreambooth inference error vs instruct-pix2pix ControlNet training/inference; not the same bug." }, { "left": "issue:4964", "right": "issue:7782", "accept": false, - "reason": "Converting SDXL models to diffusers is unrelated to a safetensors conversion/unpickling error." + "reason": "SDXL model conversion and safetensors conversion failure are unrelated." }, { "left": "issue:1350", "right": "issue:3087", "accept": false, - "reason": "DiffusionDet object detection and LyCORIS support are entirely different features." + "reason": "DiffusionDet object detection support and LyCORIS/LoCon/LoHA support are different domains." }, { "left": "issue:1594", "right": "issue:4949", "accept": false, - "reason": "Both touch video generation, but one is img2img for video generation and the other is ProPainter in/outpainting. Different models and tasks." + "reason": "Video img2img generation and ProPainter in/outpainting are not the same change." }, { "left": "issue:1962", "right": "issue:3444", "accept": false, - "reason": "Stable-diffusion video generation and ImageBind/StableUnCLIP audio-to-image integration are unrelated modalities and code paths." + "reason": "Video generation with stable diffusion is unrelated to audio-to-image integration with ImageBind." }, { "left": "issue:318", "right": "issue:3891", "accept": false, - "reason": "Both are TTS-related, but Grad-TTS and Tortoise TTS are different pipelines and support requests." + "reason": "Grad-TTS and Tortoise TTS are separate TTS pipeline requests." }, { "left": "issue:5687", "right": "issue:5710", "accept": false, - "reason": "Both involve scheduler/sigma behavior, but one is duplicate timesteps breaking Img2Img at low strength and the other is missing self.sigmas during initialization. Related area, distinct bug." + "reason": "One is a low-strength img2img timestep bug; the other is a scheduler initialization bug. Different code paths." }, { "left": "issue:6935", "right": "issue:7882", "accept": false, - "reason": "IP-Adapter face ID support and missing from_single_file on the T2I adapter pipeline are different feature gaps." + "reason": "IP-Adapter face ID support and missing from_single_file on T2I-Adapter are unrelated." }, { "left": "issue:3325", "right": "issue:5960", "accept": false, - "reason": "Different training failures: a CUDA invalid-argument error in Dreambooth vs a NaN cast warning in ControlNet SDXL training." + "reason": "Dreambooth CUDA invalid argument and SDXL ControlNet NaN/uint8 cast warning are different failures." }, { "left": "issue:4014", "right": "issue:8723", "accept": false, - "reason": "SDXL watermark artifacts and SD3ControlNetModel initialization errors are separate problems." + "reason": "Watermark pixel artifacts in SDXL are unrelated to SD3ControlNetModel initialization errors." }, { "left": "issue:6136", "right": "issue:7569", "accept": false, - "reason": "Both mention Mac, but one reports NaNs in StableDiffusionPipeline on CPU and the other is a FreeU test failure. Different code paths." + "reason": "SD2.1 NaNs on Mac and a FreeU test failure on Mac are not the same issue." }, { "left": "issue:5886", "right": "issue:7661", "accept": false, - "reason": "IP-Adapter face model support and making AnimateDiff work with an IP-Adapter-loaded UNet are different feature requests, not one bug." + "reason": "IP-Adapter face model support and enabling animate diff with IP-Adapter are distinct feature requests." }, { "left": "issue:10336", "right": "issue:9450", "accept": false, - "reason": "Sana activation explosion/clamping and FluxPipeline multi-GPU device mismatch are unrelated bugs." - }, - { - "left": "issue:3623", - "right": "issue:671", - "accept": false, - "reason": "Unable to load larger models versus converted models not being identical are different issues: one is resource/loading, the other conversion correctness." - }, - { - "left": "issue:10054", - "right": "issue:12606", - "accept": false, - "reason": "Scheduler test parameters to torch.linspace and a wrong script import are unrelated problems." + "reason": "Sana activation clamping and Flux multi-GPU device mismatch are different bugs in different models." } ] }, @@ -60655,15 +60657,14 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10054", "issue:10210", - "issue:10336", "issue:10513", - "issue:10514", "issue:10650", "issue:10800", "issue:10932", @@ -60672,7 +60673,7 @@ "issue:11352", "issue:12396", "issue:12504", - "issue:1791", + "issue:12606", "issue:2203", "issue:2228", "issue:2364", @@ -60680,6 +60681,7 @@ "issue:3032", "issue:3085", "issue:319", + "issue:3623", "issue:3861", "issue:3973", "issue:4674", @@ -60696,9 +60698,9 @@ "issue:5889", "issue:5897", "issue:6028", + "issue:671", "issue:7232", "issue:730", - "issue:760", "issue:7611", "issue:7799", "issue:7991", @@ -60709,6 +60711,8 @@ "issue:987" ], "soft_pairs": [ + "issue:3623|issue:671", + "issue:10054|issue:12606", "issue:11352|issue:12504", "issue:319|issue:938", "issue:5743|issue:5897", @@ -60730,182 +60734,180 @@ "issue:10513|issue:12396", "issue:3032|issue:3861", "issue:4936|issue:7991", - "issue:9392|issue:9485", - "issue:1791|issue:760", - "issue:10336|issue:10514" + "issue:9392|issue:9485" ], - "prepared_review_unit_hash": "609d78588254a0ceb644b0f60521b59d28a4586387aa7997fb181be2ee47ba88", + "prepared_review_unit_hash": "a7b7b4a63d68d8dbe4096c6cf04d11861be708f7cb3f2b028c0ee62024f21410", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33372, - "estimated_input_tokens": 8343, - "estimated_eval_tokens": 16942 + "serialized_chars": 33386, + "estimated_input_tokens": 8347, + "estimated_eval_tokens": 16950 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is almost entirely a set of loose soft-similarity false positives: most pairs share only broad terms like pipeline, shape, memory, training, or scheduler, but describe different bugs, features, or APIs. No soft edge looks like the same underlying defect or change.", - "confidence": 0.96, - "canonical_issue_reason": "If a single issue must anchor the cluster, issue 5889 is the broadest/highest-traffic general thread, but the cluster is not actually duplicate-coherent, so there is no true canonical duplicate target here.", + "summary": "The soft links are mostly false positives: these issues span unrelated pipelines, scheduler behavior, quantization/offload, shape mismatches, and docs/training edge cases. No pair looks like the same underlying bug or change.", + "confidence": 0.94, + "canonical_issue_reason": "issue:10210 is the most concrete and widely discussed user-facing bug report here, so it is the least arbitrary representative of this heterogeneous set.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 5889 is the most generic and visible topic among the set, but it still does not match the others as the same bug; it is only the least-bad anchor for an otherwise unrelated cluster.", + "best_issue_reason": "As a cluster representative, issue:10210 is the strongest fit by specificity and discussion activity, but it should not be treated as a true duplicate anchor for the whole group.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3623", + "right": "issue:671", + "accept": false, + "reason": "Both are model loading/behavior complaints, but one is about large-model loading failures and the other about converted-model fidelity; different underlying problems." + }, + { + "left": "issue:10054", + "right": "issue:12606", + "accept": false, + "reason": "Scheduler test parameter bug vs wrong script import; unrelated code paths." + }, { "left": "issue:11352", "right": "issue:12504", "accept": false, - "reason": "Different problems: 8-bit CPU move/warning handling vs a FreeU ComplexHalf warning." + "reason": "CPU move warning for 8bit quantized pipelines vs freeU ComplexHalf warning; different functionality and failure modes." }, { "left": "issue:319", "right": "issue:938", "accept": false, - "reason": "Completely different models/domains: VQ-Diffusion vs Human Motion Diffusion." + "reason": "Completely different domains: VQ-Diffusion model request vs human motion diffusion model request." }, { "left": "issue:5743", "right": "issue:5897", "accept": false, - "reason": "Training stalled at 0% is unrelated to a CUDA/CPU device mismatch in LoRA training." + "reason": "Training stuck at 0% versus device-mismatch in LoRA training; not the same bug." }, { "left": "issue:5372", "right": "issue:5824", "accept": false, - "reason": "Safety checker usage question vs a default value bug in get_add_time_ids()." + "reason": "Both mention SDXL pipeline internals, but one asks how to use safety_checker and the other reports a bad default in get_add_time_ids()." }, { "left": "issue:2364", "right": "issue:4704", "accept": false, - "reason": "Both involve training failures, but they are different scripts and different failure modes." + "reason": "Generic training-script failure vs Custom Diffusion training failure; too broad and no shared concrete failure mode." }, { "left": "issue:4674", "right": "issue:490", "accept": false, - "reason": "Both are memory-related, but one is a width/height cache leak and the other is a general memory complaint." + "reason": "Memory leak when changing width/height across calls is not the same as general low-memory complaints." }, { "left": "issue:2228", "right": "issue:3861", "accept": false, - "reason": "API/loading confusion around from_pretrained naming, but not the same concrete failure." + "reason": "Loading-to-GPU issue vs API naming question about from_pretrained; different underlying issue." }, { "left": "issue:4835", "right": "issue:8930", "accept": false, - "reason": "Both are ControlNet img2img issues, but one is MultiControlNet integration and the other is pose adherence." + "reason": "ControlNet img2img bug with MultiControlNetModel vs pose-control adherence issues; both about control images but not the same concrete defect." }, { "left": "issue:5889", "right": "issue:7232", "accept": false, - "reason": "Stable video diffusion vs a feature request for multimodal diffusion transformers." + "reason": "Stable video diffusion feature request/announcement vs support for multimodal diffusion transformers; not duplicates." }, { "left": "issue:539", "right": "issue:987", "accept": false, - "reason": "snapshot_download argument incompatibility vs an unrelated green-image output bug." + "reason": "Different unrelated TypeError reports from different APIs and pipelines." }, { "left": "issue:7611", "right": "issue:7799", "accept": false, - "reason": "Both are shape/matrix mismatch errors, but they occur in different pipelines and code paths." + "reason": "Shape mismatch in unet2dconditional vs matmul mismatch with IP-Adapter; both are tensor-shape errors but distinct code paths." }, { "left": "issue:3085", "right": "issue:3973", "accept": false, - "reason": "Both are keyword-argument/API issues, but the affected functions and missing parameters are different." + "reason": "Removed return_cache_folder argument vs unexpected vae_path keyword; both are parameter/API mismatches but not the same one." }, { "left": "issue:2947", "right": "issue:4692", "accept": false, - "reason": "Missing method on UNet2DConditionModel vs a separate memory concern." + "reason": "Missing load_attn_procs attribute vs generic UNet memory concerns; unrelated defects." }, { "left": "issue:10650", "right": "issue:10962", "accept": false, - "reason": "Both mention shape issues, but Lumina reshape failure and Cogview4 prompt-embed handling are different bugs." + "reason": "Lumina reshape runtime error vs Cogview4 prompt-embed shape issues; similar symptom class only, not the same bug." }, { "left": "issue:10800", "right": "issue:10932", "accept": false, - "reason": "Quantization/offload dtype failure vs torch.compile recompile-limit exhaustion; unrelated code paths." + "reason": "Quantized model plus sequential CPU offload crash vs torch.compile recompile limit; different mechanisms and failures." }, { "left": "issue:10210", "right": "issue:11006", "accept": false, - "reason": "Output-similarity bug in FLUX img2img vs broken video output in Wan 2.1 with quantized transformer." + "reason": "FluxImg2Img output too similar to input is not the same as broken Wan 2.1 I2V output with a quantized transformer." }, { "left": "issue:5044", "right": "issue:6028", "accept": false, - "reason": "Kandinsky 2.2 load failure vs Kandinsky 3.0 out-of-memory; different versions and failure classes." + "reason": "Kandinsky 2.2 load failure vs Kandinsky 3.0 CUDA OOM; different models and different failure modes." }, { "left": "issue:2203", "right": "issue:730", "accept": false, - "reason": "Download location issue vs DreamBooth training quality/local-vs-Colab behavior; not the same bug." + "reason": "DreamBooth download location question vs DreamBooth training quality issue; unrelated." }, { "left": "issue:10513", "right": "issue:12396", "accept": false, - "reason": "Colab FLUX Schnell errors vs adapter deletion/leaf_level errors; unrelated subsystems." + "reason": "Colab FLUX errors vs leaf_level/delete_adapters errors; different subsystems and symptoms." }, { "left": "issue:3032", "right": "issue:3861", "accept": false, - "reason": "A missing from_pretrained attribute on a ControlNet pipeline is not the same as a generic Pipeline.from_pretrained question." + "reason": "ControlNet class lacking from_pretrained vs generic Pipeline.from_pretrained question; not the same defect." }, { "left": "issue:4936", "right": "issue:7991", "accept": false, - "reason": "DreamBooth LoRA resume regression vs a basic training tutorial bug; different workflows and failures." + "reason": "Dreambooth LoRA resume regression vs basic training tutorial bug; different workflows and root causes." }, { "left": "issue:9392", "right": "issue:9485", "accept": false, - "reason": "Both concern schedulers, but one is SNR shift feature work and the other is device placement policy." - }, - { - "left": "issue:1791", - "right": "issue:760", - "accept": false, - "reason": "Both involve MPS, but one is lack of support and the other is unexpected output behavior." - }, - { - "left": "issue:10336", - "right": "issue:10514", - "accept": false, - "reason": "Sana activation explosion/clamping vs a 4k resolution-binning sample_size restriction; different problems." + "reason": "Scheduler SNR shift design question vs request to move scheduler computation to GPU; related area but not the same underlying bug/change." } ] }, @@ -60932,13 +60934,13 @@ "issue:1013", "issue:10202", "issue:10334", + "issue:10336", "issue:10411", "issue:10490", + "issue:10514", "issue:10529", "issue:10596", "issue:10671", - "issue:10760", - "issue:10767", "issue:10817", "issue:1083", "issue:10873", @@ -60948,8 +60950,8 @@ "issue:12052", "issue:12321", "issue:12459", - "issue:13034", "issue:13506", + "issue:1791", "issue:2089", "issue:4210", "issue:4332", @@ -60968,10 +60970,10 @@ "issue:6783", "issue:6794", "issue:6928", + "issue:760", "issue:7793", "issue:7842", "issue:834", - "issue:8410", "issue:8589", "issue:8612", "issue:8647", @@ -60979,6 +60981,8 @@ "issue:9972" ], "soft_pairs": [ + "issue:1791|issue:760", + "issue:10336|issue:10514", "issue:11148|issue:9953", "issue:1083|issue:834", "issue:4616|issue:8647", @@ -61000,182 +61004,180 @@ "issue:10490|issue:12321", "issue:10529|issue:6928", "issue:10596|issue:6794", - "issue:10671|issue:10873", - "issue:10760|issue:13034", - "issue:10767|issue:8410" + "issue:10671|issue:10873" ], - "prepared_review_unit_hash": "843ab9db4f9fb84229e715c9c680e340b513238700bd6065da051dde54f9fac7", + "prepared_review_unit_hash": "e7fe80716a3256933b8fd6dee5645cd0d5dc9aa76344ba43b25233f710506ea9", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33774, - "estimated_input_tokens": 8444, - "estimated_eval_tokens": 17144 + "serialized_chars": 33866, + "estimated_input_tokens": 8467, + "estimated_eval_tokens": 17190 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are a loose topical collection around Diffusers bugs and feature requests, not a duplicate cluster. The soft pairs mostly share broad subsystem words only; none look like the same concrete bug or change.", + "summary": "This set is not a duplicate cluster: the items span many unrelated bugs, feature requests, and model-specific issues. None of the soft-similarity pairs are the same underlying problem closely enough to merge.", "confidence": 0.97, - "canonical_issue_reason": "issue:10671 is the strongest representative because it describes a concrete, core pipeline bug with clear behavior, reproduction, and some cross-reference activity.", + "canonical_issue_reason": "No single canonical issue stands out because the issues are heterogeneous and do not describe one shared bug or feature request.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10671 is the best issue to anchor the cluster triage because it is specific, broadly relevant, and well-scoped compared with the more niche or feature-like reports.", + "best_issue_reason": "No issue is a good global canonical for the set; the topics are too diverse (Sana, Flux, PEFT, schedulers, training, MPS, etc.).", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1791", + "right": "issue:760", + "accept": false, + "reason": "Both mention MPS, but one is UnCLIPPipeline support and the other is unexpected StableDiffusionPipeline output in an older release; different code paths and symptoms." + }, + { + "left": "issue:10336", + "right": "issue:10514", + "accept": false, + "reason": "Both are Sana-related, but one reports exploding activations/clamping while the other is a 4k resolution-binning sample_size limitation; not the same bug." + }, { "left": "issue:11148", "right": "issue:9953", "accept": false, - "reason": "Different problems: PEFT LoRA adapter state handling vs moving a quantized component to CUDA." + "reason": "Both involve loading/unloading adapters, but one is a PEFT state flag bug and the other is an error moving a pipeline with a quantized component to CUDA." }, { "left": "issue:1083", "right": "issue:834", "accept": false, - "reason": "Unrelated: FP16 ONNX correctness issue vs general CPU under-utilization." + "reason": "FP16 ONNX incorrect output is not the same as CPU underutilization; different functionality and failure mode." }, { "left": "issue:4616", "right": "issue:8647", "accept": false, - "reason": "Different code paths and failures: SDXL LoRA conversion script vs ControlNet img2img NoneType error." + "reason": "One is an SDXL LoRA conversion script issue, the other is a ControlNet img2img runtime NoneType error; unrelated." }, { "left": "issue:11886", "right": "issue:12459", "accept": false, - "reason": "Different FLUX topics: a pipeline issue vs batch inference throughput, not the same bug." + "reason": "Both are FLUX/FluxKontext-related, but one is a pipeline issue and the other is a batch throughput concern; not the same concrete bug." }, { "left": "issue:10411", "right": "issue:6518", "accept": false, - "reason": "Different training scripts and errors: lora-weight training usage question vs NoneType floor-division in a distillation script." + "reason": "Different training scripts and different errors: one asks how to call a consistency-stiffness LoRA training script, the other is a NoneType floor-division failure in a different distillation script." }, { "left": "issue:10817", "right": "issue:8589", "accept": false, - "reason": "Different concerns: missing SD3 ControlNets in auto-pipeline vs SD3 DreamBooth prior-preservation training." + "reason": "SD3 ControlNet missing in auto_pipeline is a support gap; SD3 DreamBooth prior preservation is a training limitation. Different features, different fixes." }, { "left": "issue:6091", "right": "issue:6222", "accept": false, - "reason": "Both are scheduler/decoder-related but the bugs differ: memory usage with size args vs DDPM batch-size dependency." + "reason": "Both are scheduler/memory-related, but one concerns a consistency decoder size argument and the other DDPM batch-size dependency; not the same bug." }, { "left": "issue:12052", "right": "issue:13506", "accept": false, - "reason": "Different failures: Wan 2.2 offloading device-mismatch vs Sage attention incompatibility with ring-attention." + "reason": "Both involve attention/offloading on newer models, but one is a device-mismatch multiplication failure in Wan 2.2 and the other is Sage attention failing with ring-attention; distinct code paths." }, { "left": "issue:7793", "right": "issue:7842", "accept": false, - "reason": "Different DeepFloyd issues: LoRA unloading error vs tensor-size mismatch on non-8-divisible input images." + "reason": "Both mention LoRA/deepfloyd, but one is unloading LoRA weights causing an error and the other is a tensor-size mismatch on non-8-divisible input dimensions; not the same issue." }, { "left": "issue:6783", "right": "issue:8612", "accept": false, - "reason": "Different model/runtime issues: PixArt-XL generation corruption vs Pixart-sigma mixed-precision incompatibility." + "reason": "Both are PixArt/Pixart-sigma generation quality or precision issues, but one is messed-up generations and the other is mixed-precision incompatibility; different problems." }, { "left": "issue:5545", "right": "issue:5595", "accept": false, - "reason": "Both are pipeline input problems, but one is StableDiffusionInpaintPipeline behavior and the other is invalid height/width validation." + "reason": "Both are Stable Diffusion pipeline errors, but one is inpainting pipeline behavior and the other is a height/width divisibility validation error." }, { "left": "issue:5646", "right": "issue:5689", "accept": false, - "reason": "Related scheduler family, but not the same bug: stability issues on a specific model vs residual noise in SDXL DPM++ output." + "reason": "Both concern diffusion scheduler behavior, but one is broad stability issues on a specific model and the other is residual noise in SDXL from DPM++; not the same concrete defect." }, { "left": "issue:4210", "right": "issue:4332", "accept": false, - "reason": "Different attention-mask/xformers concerns: UNet forward API question vs kandinsky2.2 xformers failure." + "reason": "Attention_mask usage in UNet2DConditionModel is unrelated to xformers failing in kandinsky2.2 pipeline." }, { "left": "issue:4534", "right": "issue:4631", "accept": false, - "reason": "Different runtime bugs: SDXL dreambooth validation on MPS vs DPMSolverSinglestepScheduler Karras sigma scalar error." + "reason": "Both are runtime errors in generation/training, but one is SDXL dreambooth on MPS with a dimension mismatch and the other is a DPMSolverSinglestepScheduler scalar conversion error." }, { "left": "issue:2089", "right": "issue:5951", "accept": false, - "reason": "Both training-related, but DreamBooth OOM and PixArt training are not the same underlying issue." + "reason": "DreamBooth OOM with AltDiffusion is a memory problem; Pixart training is a broad training issue with no matching failure mode." }, { "left": "issue:1013", "right": "issue:5026", "accept": false, - "reason": "Different release-era import/pipeline issues: StableDiffusionMegaPipeline broken in Diffusers 6.0 vs KeyError for text_encoder in 0.21.0." + "reason": "StableDiffusionMegaPipeline not working in Diffusers 6.0 and KeyError:'text_encoder' in 0.21.0 are version-specific but describe different failures in different code paths." }, { "left": "issue:10202", "right": "issue:11046", "accept": false, - "reason": "Different FLUX/LoRA and FLUX/controlnet-inpainting-plus-ip-adapter failures; no same code-path evidence." + "reason": "A Flux Control LoRA unload bug is unrelated to FLUX pipeline inference with ControlNet/inpainting/IP-Adapter." }, { "left": "issue:10334", "right": "issue:9972", "accept": false, - "reason": "Different macOS issues with different symptoms: Sana MPS/CPU NaNs vs CogX 10TB buffer request." + "reason": "Sana MacOS grey/NaN outputs are unrelated to CogX requesting an absurd 10TB buffer; different pipeline and failure mechanism." }, { "left": "issue:10490", "right": "issue:12321", "accept": false, - "reason": "Feature requests for different models/platform support, not the same bug or change." + "reason": "'support Infinity' and HunyuanImage-2.1 support are both feature requests, but for different models/features and not the same request." }, { "left": "issue:10529", "right": "issue:6928", "accept": false, - "reason": "Different feature gaps: DDIMInverseScheduler missing linspace spacing vs depthmap colorization option." + "reason": "Missing linspace timestep spacing in DDIMInverseScheduler is unrelated to disabling depthmap colorization in Marigold." }, { "left": "issue:10596", "right": "issue:6794", "accept": false, - "reason": "Different errors in different areas: FluxControlPipeline shape mismatch vs cast_training_params import failure." + "reason": "Invalid shape in FluxControlPipeline is a runtime shape bug; cast_training_params import error is a separate training-utils API issue." }, { "left": "issue:10671", "right": "issue:10873", "accept": false, - "reason": "Different bugs: generic DiffusionPipeline dtype/device determinism vs SanaPipeline model-specific issue." - }, - { - "left": "issue:10760", - "right": "issue:13034", - "accept": false, - "reason": "Different import vs geometry issues: missing ONNX legacy pipeline attribute vs WanPipeline height/width handling." - }, - { - "left": "issue:10767", - "right": "issue:8410", - "accept": false, - "reason": "Different requests: FlashVideo integration discussion vs native TextDiffuser pipeline support." + "reason": "Deterministic dtype/device access in DiffusionPipeline and SanaPipeline issues are both pipeline bugs, but they are unrelated subsystems with different symptoms." } ] }, @@ -61194,11 +61196,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10760", + "issue:10767", "issue:1101", "issue:1117", "issue:11351", @@ -61211,6 +61215,7 @@ "issue:12645", "issue:12666", "issue:12754", + "issue:13034", "issue:1305", "issue:13191", "issue:13377", @@ -61229,26 +61234,25 @@ "issue:2866", "issue:2883", "issue:2972", - "issue:2984", "issue:3010", "issue:3073", - "issue:3093", "issue:3159", - "issue:3817", "issue:3838", "issue:3843", "issue:4386", "issue:4390", - "issue:4630", "issue:5285", "issue:5877", "issue:6928", "issue:7365", + "issue:8410", "issue:940", "issue:9471", "issue:9744" ], "soft_pairs": [ + "issue:10760|issue:13034", + "issue:10767|issue:8410", "issue:1101|issue:5285", "issue:1117|issue:3010", "issue:11351|issue:12666", @@ -61270,182 +61274,180 @@ "issue:2761|issue:3159", "issue:2866|issue:4386", "issue:2883|issue:3838", - "issue:2972|issue:7365", - "issue:2984|issue:3817", - "issue:3093|issue:4630" + "issue:2972|issue:7365" ], - "prepared_review_unit_hash": "db62e4ac4224ef34ad26744c1dfd29e1d56daa4c2f884dbfb63e816b27c6be39", + "prepared_review_unit_hash": "2ba797a0bc4b539409cf083cb29a971b8bdeb64ba5acd865a2f115cbf2df2450", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33007, - "estimated_input_tokens": 8252, - "estimated_eval_tokens": 16760 + "serialized_chars": 33004, + "estimated_input_tokens": 8251, + "estimated_eval_tokens": 16758 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a broad mix of unrelated feature requests, bug reports, and docs/tests issues. None of the soft pairs look like true duplicates; they mostly share only superficial theme overlap (e.g. model support, MPS, LoRA, video pipelines).", - "confidence": 0.97, - "canonical_issue_reason": "No clear canonical issue: the items do not converge on one underlying bug or change, and the apparent overlaps are only loose topical similarities.", + "summary": "All proposed soft-edge pairs are weakly similar at best and mostly unrelated; they span different bugs, features, docs, models, and pipelines. I would not merge any pair as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue stands out: this set is heterogeneous, with unrelated model-support requests, bugs, docs issues, and pipeline-specific regressions.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good global representative here because the cluster is heterogeneous rather than one duplicate family.", + "best_issue_reason": "No global best issue candidate: none of these issues clearly subsumes the others or represents the same underlying problem across the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10760", + "right": "issue:13034", + "accept": false, + "reason": "Different failures in different pipelines: import-time missing attribute vs WanPipeline height/width handling." + }, + { + "left": "issue:10767", + "right": "issue:8410", + "accept": false, + "reason": "FlashVideo integration request is unrelated to native TextDiffuser pipeline support." + }, { "left": "issue:1101", "right": "issue:5285", "accept": false, - "reason": "DPM-Solver scheduler request vs Japanese docs translation; unrelated." + "reason": "Scheduler feature request and Japanese docs translation are unrelated." }, { "left": "issue:1117", "right": "issue:3010", "accept": false, - "reason": "Tests for multiple negative prompts vs distributed multi-GPU inference docs; different scope." + "reason": "Negative-prompt tests and distributed inference docs address different areas." }, { "left": "issue:11351", "right": "issue:12666", "accept": false, - "reason": "Wan i2v float32 dtype question vs Kandinsky video pipeline availability; different problems." + "reason": "Wan video processor dtype question is unrelated to Kandinsky 5.0 video pipeline availability." }, { "left": "issue:11352", "right": "issue:12599", "accept": false, - "reason": "8-bit quantized model CPU move warning vs slow CUDA transfer after loading; different code paths." + "reason": "One is missing a warning on CPU move for 8-bit models; the other is slow CUDA transfer after loading." }, { "left": "issue:11519", "right": "issue:13191", "accept": false, - "reason": "MAGI-1 support request vs Elastic-DiT support request; both model requests but not the same model/change." + "reason": "Two separate model support requests (MAGI-1 vs Elastic-DiT) with no shared bug/change." }, { "left": "issue:12108", "right": "issue:12645", "accept": false, - "reason": "Scheduler-enabled breakage in Qwen Image/Chroma vs Qwen-Image speed comparison; bug vs performance complaint." + "reason": "Flow-matching scheduler breakage is a functional bug; the other is a speed comparison complaint." }, { "left": "issue:12338", "right": "issue:9471", "accept": false, - "reason": "AutoencoderDC VAE slicing bug vs file-extension inference ValueError; unrelated." + "reason": "AutoencoderDC VAE slicing bug is unrelated to a file-extension/resource-type parsing error." }, { "left": "issue:12358", "right": "issue:9744", "accept": false, - "reason": "Lumina-DiMOO pipeline request vs Mochi video model request; different models." + "reason": "Different model support requests: Lumina-DiMOO vs Mochi video model." }, { "left": "issue:12754", "right": "issue:13377", "accept": false, - "reason": "from_pipe dtype conversion bug vs QwenImage CFG masking bug; separate behaviors." + "reason": "Both mention pipeline behavior, but one is float32 casting in from_pipe and the other is CFG suppression with negative_prompt_embeds." }, { "left": "issue:1305", "right": "issue:3073", "accept": false, - "reason": "Direct Inversion paper request vs Consistency Models request; different features." + "reason": "Direct Inversion method request is unrelated to adding consistency models." }, { "left": "issue:1398", "right": "issue:940", "accept": false, - "reason": "General Stable Diffusion 2 on MPS issue vs LMSDiscreteScheduler crash on MPS; related platform but not same bug." + "reason": "Both touch MPS, but one is Stable Diffusion 2 support and the other is an LMSDiscreteScheduler crash." }, { "left": "issue:1530", "right": "issue:1919", "accept": false, - "reason": "Missing VQDiffusion embedding field vs missing DreamBooth dependency; unrelated missing items." + "reason": "Missing VQDiffusion pipeline embedding field and missing DreamBooth requirements are unrelated." }, { "left": "issue:1541", "right": "issue:5877", "accept": false, - "reason": "DreamBooth model-type detection/load failure vs subprocess error in textual inversion/DreamBooth; different failure modes." + "reason": "Both involve DreamBooth, but one is CKPT/model-type detection and the other is subprocess failures in training scripts." }, { "left": "issue:202", "right": "issue:2377", "accept": false, - "reason": "Tests for unet_blocks vs tests for EMAModel; both tests, but different targets." + "reason": "Unet blocks tests and EMAModel tests are unrelated test additions." }, { "left": "issue:2250", "right": "issue:2603", "accept": false, - "reason": "DreamBooth prompt generation failure vs Attend-And-Excitte non-square image issue; different pipelines." + "reason": "Different DreamBooth/Attend-And-Excite inference problems with different symptoms and code paths." }, { "left": "issue:2538", "right": "issue:6928", "accept": false, - "reason": "Enable train_text_encoder for DreamBooth LoRA vs disable depthmap colorization; unrelated option requests." + "reason": "Train-text-encoder option and Marigold depthmap colorization are separate feature requests." }, { "left": "issue:2550", "right": "issue:3843", "accept": false, - "reason": "ControlNet-to-diffusers conversion failure vs missing global_step key; different bugs." + "reason": "ControlNet model conversion and missing global_step key are different code-path issues." }, { "left": "issue:2674", "right": "issue:4390", "accept": false, - "reason": "Specific safetensors dependency problem vs generic missing requirements; not the same issue." + "reason": "A missing safetensors dependency in LoRA workflows is not the same as a generic missing requirements report." }, { "left": "issue:2761", "right": "issue:3159", "accept": false, - "reason": "UNet1DModel docs request vs typo in docs code; different documentation issues." + "reason": "Documentation request for UNet1DModel and a typo fix in docs are different doc issues." }, { "left": "issue:2866", "right": "issue:4386", "accept": false, - "reason": "LoRA bin-to-safetensors conversion enhancement vs question about converting SDXL LoRA; related topic but not the same change." + "reason": "Both involve LoRA conversion, but one is bin-to-safetensors conversion and the other is an SDXL safetensor LoRA question." }, { "left": "issue:2883", "right": "issue:3838", "accept": false, - "reason": "Image-to-video pipeline request vs Video Composer request; different model/pipeline additions." + "reason": "Image-to-video pipeline and Video Composer are broad feature requests, not the same concrete change." }, { "left": "issue:2972", "right": "issue:7365", "accept": false, - "reason": "Loop variable adjustment bug vs pooled_prompt_embeds overwrite bug; unrelated code paths." - }, - { - "left": "issue:2984", - "right": "issue:3817", - "accept": false, - "reason": "Latent video diffusion model request vs editable dance generation from music; different research/pipeline asks." - }, - { - "left": "issue:3093", - "right": "issue:4630", - "accept": false, - "reason": "Modify noise directly vs modify LoRA weights after loading; different objects and APIs." + "reason": "The 0.01 adjustment-loop bug is unrelated to pooled_prompt_embeds being overwritten." } ] }, @@ -61464,11 +61466,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:2984", + "issue:3093", "issue:3111", "issue:3113", "issue:3171", @@ -61477,6 +61481,7 @@ "issue:3406", "issue:3444", "issue:3728", + "issue:3817", "issue:4052", "issue:4264", "issue:4311", @@ -61484,6 +61489,7 @@ "issue:4386", "issue:4524", "issue:459", + "issue:4630", "issue:4868", "issue:4877", "issue:4912", @@ -61503,7 +61509,6 @@ "issue:6732", "issue:692", "issue:7147", - "issue:7237", "issue:7264", "issue:7377", "issue:7440", @@ -61511,14 +61516,13 @@ "issue:8410", "issue:8451", "issue:8495", - "issue:8593", "issue:8792", "issue:9001", - "issue:9454", - "issue:9793", - "issue:9902" + "issue:9793" ], "soft_pairs": [ + "issue:2984|issue:3817", + "issue:3093|issue:4630", "issue:3111|issue:4912", "issue:3111|issue:7264", "issue:3113|issue:5442", @@ -61541,188 +61545,186 @@ "issue:5931|issue:9793", "issue:6093|issue:8451", "issue:6295|issue:8792", - "issue:7147|issue:7377", - "issue:7237|issue:9902", - "issue:8593|issue:9454" + "issue:7147|issue:7377" ], - "prepared_review_unit_hash": "059084dbaa137bb8b9ddfc7751f3d26d5110f2398be80f5e8379e71197e91552", + "prepared_review_unit_hash": "3e021c1dca14cd8227729ef04a265755017fba3d368aa51187a5a1eb77645736", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33184, - "estimated_input_tokens": 8296, - "estimated_eval_tokens": 16848 + "serialized_chars": 33185, + "estimated_input_tokens": 8297, + "estimated_eval_tokens": 16850 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a noisy mix of unrelated Diffusers feature requests, runtime bugs, docs issues, and model-specific questions. None of the proposed soft pairs look like true duplicates of the same underlying change or bug.", - "confidence": 0.98, - "canonical_issue_reason": "No clear canonical issue: the items span many unrelated subsystems and problem types, so one issue would not safely represent the cluster as a duplicate set.", + "summary": "This cluster is mostly a set of unrelated feature requests/questions across different Diffusers pipelines and utilities; the soft-similarity links are not strong enough to treat any pair as duplicates.", + "confidence": 0.34, + "canonical_issue_reason": "issue:4524 is the strongest representative issue here: it has the highest discussion activity and inbound references, and it is a concrete feature request rather than a narrow one-off question.", "canonical_pr_reason": null, - "best_issue_reason": "issue:4524 is the most substantial and discussion-rich issue here, but it is still just a representative feature request rather than a true canonical for these heterogeneous items.", + "best_issue_reason": "issue:4524 is the best single issue to anchor the cluster because it is prominent, substantive, and clearly scoped to a real product enhancement.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2984", + "right": "issue:3817", + "accept": false, + "reason": "Unrelated topics: long-video diffusion model research vs editable dance generation from music." + }, + { + "left": "issue:3093", + "right": "issue:4630", + "accept": false, + "reason": "Noise manipulation during generation is different from changing LoRA weights after loading." + }, { "left": "issue:3111", "right": "issue:4912", "accept": false, - "reason": "Custom conditioning encoder support and SDXL checkpoint conversion are different tasks." + "reason": "Custom conditioning encoder support is unrelated to converting SDXL checkpoints." }, { "left": "issue:3111", "right": "issue:7264", "accept": false, - "reason": "One is about conditioning encoders; the other is about serializing/loading custom Transformers models." + "reason": "Custom conditioning encoder support is a different problem from serializing/loading custom Transformers models." }, { "left": "issue:3113", "right": "issue:5442", "accept": false, - "reason": "Both are pipeline ideas, but they target different methods and code paths." + "reason": "Cross-attention layout control and ScaleCrafter are different feature requests for different code paths." }, { "left": "issue:3171", "right": "issue:5070", "accept": false, - "reason": "LPW prompt support for ControlNet is unrelated to the only_mid_control/sd_locked example flags." + "reason": "Both mention ControlNet-style prompting, but one is about LPW prompts and the other about specific control flags in an example; not the same bug/change." }, { "left": "issue:3221", "right": "issue:3338", "accept": false, - "reason": "An attention processor bug and a slow import-time issue are not the same defect." + "reason": "A version-specific attention processor loading bug is unrelated to slow package import time." }, { "left": "issue:3406", "right": "issue:4311", "accept": false, - "reason": "only_cross_attention behavior and local_files_only pipeline-class lookup are separate bugs." + "reason": "only_cross_attention behavior is unrelated to the local_files_only pipeline_class bug." }, { "left": "issue:3406", "right": "issue:5062", "accept": false, - "reason": "A typo in attention code is not the same as the only_cross_attention logic failure." + "reason": "Attention behavior for only_cross_attention is different from a typo in attention processor naming." }, { "left": "issue:3444", "right": "issue:3728", "accept": false, - "reason": "ImageBind/audio-to-image integration and Shifted Diffusion are different pipeline requests." + "reason": "Audio-to-image integration is unrelated to shifted diffusion text-to-image support." }, { "left": "issue:4052", "right": "issue:7440", "accept": false, - "reason": "Missing from_single_file support is unrelated to the Pokemon dataset availability issue." + "reason": "Missing from_single_file import and a missing dataset are unrelated issues." }, { "left": "issue:4264", "right": "issue:7456", "accept": false, - "reason": "Missing example images and a missing raise statement are unrelated." + "reason": "Missing example images is not the same as a missing raise statement." }, { "left": "issue:4312", "right": "issue:5371", "accept": false, - "reason": "SAG support for inpainting/img2img and BLIP inpainting support are different feature requests." + "reason": "SAG for inpainting/img2img and BLIP inpainting support are different pipeline feature requests." }, { "left": "issue:4386", "right": "issue:4877", "accept": false, - "reason": "SDXL LoRA conversion and exposing noise_init_strength are unrelated." + "reason": "LoRA conversion for SDXL is unrelated to exposing noise_init_strength as an input argument." }, { "left": "issue:4524", "right": "issue:6616", "accept": false, - "reason": "AnimateDiff support and classifier-guidance implementation are different features." + "reason": "AnimateDiff pipeline support and classifier guidance are different concrete features affecting different generation paths." }, { "left": "issue:459", "right": "issue:692", "accept": false, - "reason": "Incremental Diffusion and a fine-tuning example are not the same change." + "reason": "Incremental diffusion and an example for fine-tuning are not the same underlying issue." }, { "left": "issue:4868", "right": "issue:8495", "accept": false, - "reason": "A super().__getattr__ AttributeError and an SD3 tokenizer enum parsing error are distinct failures." + "reason": "A Python AttributeError in one path is unrelated to an SD3 tokenizer enum deserialization error." }, { "left": "issue:5130", "right": "issue:6732", "accept": false, - "reason": "A DDPMScheduler question and undocumented trained_betas are different issues." + "reason": "A scheduler usage question is not the same as undocumented trained_betas documentation." }, { "left": "issue:5254", "right": "issue:9001", "accept": false, - "reason": "Multi-ControlNet inpainting and Auraflow aspect ratios are unrelated pipeline features." + "reason": "Multi-ControlNet inpaint support and Auraflow aspect ratios are separate pipeline enhancements." }, { "left": "issue:5284", "right": "issue:8410", "accept": false, - "reason": "ControlNet-LLLite and native TextDiffuser pipeline support are different capabilities." + "reason": "ControlNet-LLLite support and native TextDiffuser support are different model/pipeline additions." }, { "left": "issue:5643", "right": "issue:6113", "accept": false, - "reason": "Using IP-Adapter ControlNet and loading models from sd_control_collection are separate usage questions." + "reason": "How to use IP-Adapter ControlNet is unrelated to loading models from the sd_control_collection repo." }, { "left": "issue:5931", "right": "issue:9793", "accept": false, - "reason": "Generation nondeterminism and a stuck progress bar are unrelated bugs." + "reason": "Generation nondeterminism and a progress bar update issue are different bugs." }, { "left": "issue:6093", "right": "issue:8451", "accept": false, - "reason": "push_to_hub repo-creation behavior and training-script freezing details do not match." + "reason": "push_to_hub behavior in one training script is unrelated to missing training details for frozen UNet parameters." }, { "left": "issue:6295", "right": "issue:8792", "accept": false, - "reason": "Artifacts with a specific scheduler and general migration quality degradation are not the same bug." + "reason": "Artifacts with a specific sampler setting are not the same as general quality degradation migrating from A1111 to Diffusers." }, { "left": "issue:7147", "right": "issue:7377", "accept": false, - "reason": "A missing import and a num_images_per_prompt runtime error are different StableCascade issues." - }, - { - "left": "issue:7237", - "right": "issue:9902", - "accept": false, - "reason": "A missing scheduler class and a conda version pin problem are unrelated." - }, - { - "left": "issue:8593", - "right": "issue:9454", - "accept": false, - "reason": "SD3 bf16 CPU noise and a CogVideoX model lookup failure are unrelated." + "reason": "Import failure for StableCascadeDecoderPipeline is different from a runtime error when num_images_per_prompt > 1." } ] }, @@ -61741,8 +61743,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -61761,10 +61763,8 @@ "issue:11363", "issue:11570", "issue:12338", - "issue:12605", "issue:12656", "issue:12981", - "issue:13377", "issue:1815", "issue:3177", "issue:3748", @@ -61775,11 +61775,11 @@ "issue:4631", "issue:5313", "issue:5710", - "issue:6385", - "issue:6506", + "issue:7237", "issue:7322", "issue:782", "issue:7897", + "issue:8593", "issue:8831", "issue:9026", "issue:9051", @@ -61788,13 +61788,17 @@ "issue:9298", "issue:9331", "issue:9417", + "issue:9454", "issue:9460", "issue:9479", "issue:9496", "issue:954", - "issue:9637" + "issue:9637", + "issue:9902" ], "soft_pairs": [ + "issue:7237|issue:9902", + "issue:8593|issue:9454", "issue:9026|issue:9298", "issue:9331|issue:9417", "issue:12338|issue:9637", @@ -61816,191 +61820,189 @@ "issue:11223|issue:11363", "issue:11010|issue:9479", "issue:11258|issue:12981", - "issue:10382|issue:12656", - "issue:12605|issue:13377", - "issue:6385|issue:6506" + "issue:10382|issue:12656" ], - "prepared_review_unit_hash": "f733a8ea210f9687ebc872d46380639854543e5abc6a4ec78c100b6d4726c83b", + "prepared_review_unit_hash": "c2203eb4bf353f7c8ed3d9fdf922062bd14ae063a68d9a285902cb69d4061c65", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33583, - "estimated_input_tokens": 8396, - "estimated_eval_tokens": 17048 + "serialized_chars": 33359, + "estimated_input_tokens": 8340, + "estimated_eval_tokens": 16936 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous: the issues span many unrelated pipelines, schedulers, quantization, and training bugs. Only one pair looks like a true duplicate on the same code path (`FlowMatchEulerDiscreteScheduler.index_for_timestep`). No PRs are present.", - "confidence": 0.78, - "canonical_issue_reason": "issue:9331 is the cleanest anchor for the one strong duplicate pair because it names the exact scheduler and method directly, without extra symptom-specific wording.", + "summary": "This cluster is mostly noisy and spans many unrelated bug reports and feature requests. The inpainting trio is the strongest thematic group, with #3880 as the best umbrella issue. For soft edges, only the DPMSolverSinglestepScheduler pair looks like a real duplicate-style match; the FlowMatch `index_for_timestep` pair should be rejected because one is a bug report and the other is a performance suggestion.", + "confidence": 0.79, + "canonical_issue_reason": "#3880 is the clearest broad inpainting-defect report and works best as the umbrella for related masked-area degradation complaints, including #9479 and the later FluxFill inpainting quality issue.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9331 is the best representative issue in this set because it most clearly states the shared `index_for_timestep` scheduler bug and can subsume issue:10584 as the more specific symptom report.", + "best_issue_reason": "#3880 is the most representative standalone bug in the set: it describes a concrete inpainting degradation problem with broad applicability.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:7237", + "right": "issue:9902", + "accept": false, + "reason": "Unrelated: missing scheduler class vs a conda version pin/stuck version issue." + }, + { + "left": "issue:8593", + "right": "issue:9454", + "accept": false, + "reason": "Different bugs: SD3 CPU bfloat16 noise vs CogVideoX image2video model loading." + }, { "left": "issue:9026", "right": "issue:9298", "accept": false, - "reason": "Different failures: an unexpected `force_hook` kwarg in dispatch vs a missing `Float8_e4m3fnStorage` object." + "reason": "Different failure modes: unexpected `force_hook` arg vs missing `Float8_e4m3fnStorage`." }, { "left": "issue:9331", "right": "issue:9417", "accept": false, - "reason": "Same method family, but one is a correctness bug report and the other is a performance optimization request; not the same underlying issue." + "reason": "Same function name, but one is a correctness bug report and the other is a performance suggestion; not the same underlying defect." }, { "left": "issue:12338", "right": "issue:9637", "accept": false, - "reason": "Different subsystems: VAE slicing/decoding multiple images vs xFormers attention-mask handling under bf16." + "reason": "Different runtime issues: AutoencoderDC slicing decode bug vs xFormers attention-mask/bfloat16 failure." }, { "left": "issue:5313", "right": "issue:8831", "accept": false, - "reason": "Both involve LoRA, but one is about non-revertible fuse behavior and the other is about unsupported-layer continuation; not the same bug." + "reason": "Both are LoRA-related, but they describe different problems: non-revertible fuse vs unsupported layers." }, { "left": "issue:1815", "right": "issue:3177", "accept": false, - "reason": "Both mention DeepSpeed training, but they describe different failures in different scripts and stages." + "reason": "Both involve DeepSpeed, but the reported training/offloading failures are different." }, { "left": "issue:11329", "right": "issue:11570", "accept": false, - "reason": "Quantization is common, but this is a memory leak versus a `.to()` casting failure when loading LoRAs." + "reason": "Shared quantization context, but memory leak and `.to()` cast breakage are distinct bugs." }, { "left": "issue:4229", "right": "issue:7322", "accept": false, - "reason": "Both are training-related OOM/SIGKILL reports, but the contexts and failure modes are different." + "reason": "Both are training failures, but the contexts and failure causes differ." }, { "left": "issue:10584", "right": "issue:9331", - "accept": true, - "reason": "Both describe the same `FlowMatchEulerDiscreteScheduler.index_for_timestep` problem; one reports missing timestep indices and the other names the same method/scheduler failure." + "accept": false, + "reason": "Related scheduler area, but the specific reports do not clearly describe the same concrete bug." }, { "left": "issue:10447", "right": "issue:10928", "accept": false, - "reason": "Related to UNet model files, but one is about the converter splitting weights and the other is about a missing fp16 file on load." + "reason": "Model file handling is only a broad theme; conversion splitting and missing fp16 weights are different issues." }, { "left": "issue:10204", "right": "issue:10833", "accept": false, - "reason": "Different FluxFill problems: quality degradation across iterations vs a bitsandbytes matrix-shape multiplication error." + "reason": "Same pipeline family, but iterative inpainting degradation and quantization shape mismatch are different bugs." }, { "left": "issue:11109", "right": "issue:9178", "accept": false, - "reason": "Both involve LoRA/model-shape issues, but one is a transformer layer mismatch and the other is a PEFT target-module error." + "reason": "Different LoRA problems: transformer layer mismatch vs PEFT target-module errors." }, { "left": "issue:10737", "right": "issue:9051", "accept": false, - "reason": "Both are FluxPipeline issues, but one is a missing `image_encoder` path and the other is lack of batch inference support." + "reason": "Both are FluxPipeline issues, but missing image encoder and lack of batch inference are different problems." }, { "left": "issue:7897", "right": "issue:9170", "accept": false, - "reason": "One is multi-GPU training not working, the other is high GPU-memory demand; not the same concrete bug." + "reason": "Both mention resource pressure, but they are not the same underlying bug." }, { "left": "issue:3765", "right": "issue:3880", "accept": false, - "reason": "Both concern image quality/artifacts, but one is ControlNet training stacking effects and the other is inpainting preserving masked regions." + "reason": "Different tasks and symptoms: ControlNet training stacking effects vs inpainting mask degradation." }, { "left": "issue:9460", "right": "issue:9496", "accept": false, - "reason": "Same component, but one requests list assignment support and the other reports a forward-function error; too different to merge as one bug." + "reason": "Feature request to assign multiple ControlNet models vs a forward-function error." }, { "left": "issue:4631", "right": "issue:5710", - "accept": false, - "reason": "Same scheduler class, but one is a `use_karras_sigmas` runtime error and the other is missing `self.sigmas` during initialization; not clearly the same bug." + "accept": true, + "reason": "Same scheduler class and same sigma initialization/code-path problem; these look like the same underlying bug." }, { "left": "issue:782", "right": "issue:954", "accept": false, - "reason": "Different auth/loading failures: model download broke after `use_auth_token` removal vs load failure after successful login." + "reason": "Both are auth/download-related, but they are not the same failure." }, { "left": "issue:3748", "right": "issue:4520", "accept": false, - "reason": "Different LoRA issues: dtype mismatch while loading safetensors vs attention-slicing ignoring `lora_scale`." + "reason": "Both involve LoRA, but dtype mismatch on load and attention-slicing scale behavior are different bugs." }, { "left": "issue:11223", "right": "issue:11363", "accept": false, - "reason": "Both mention OOM, but they are unrelated pipelines and scenarios." + "reason": "OOM reports in different contexts without evidence of the same root cause." }, { "left": "issue:11010", "right": "issue:9479", "accept": false, - "reason": "One is a request for a new Flux-based model architecture, the other is an inpainting masking bug." + "reason": "Architecture-support request vs a stable-diffusion inpainting bug; not the same issue." }, { "left": "issue:11258", "right": "issue:12981", "accept": false, - "reason": "Both are LTX model-loading topics, but one is single-file support and the other is GGUF loading for LTX-2." + "reason": "Same LTX family, but different versions/formats and different loading problems." }, { "left": "issue:10382", "right": "issue:12656", "accept": false, - "reason": "Completely different bugs: VAE return-type mismatch vs AuraFlow positional embedding out-of-bounds." - }, - { - "left": "issue:12605", - "right": "issue:13377", - "accept": false, - "reason": "Different attention/control-flow bugs in different pipelines; not the same code path." - }, - { - "left": "issue:6385", - "right": "issue:6506", - "accept": false, - "reason": "Same pipeline family, but one is excessive VRAM usage and the other is an output type mismatch." + "reason": "Completely unrelated: VAE return-type mismatch vs AuraFlow positional embedding out-of-bounds." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. It correctly treats the cluster as mostly heterogeneous, identifies the single strongest duplicate-style pair (`10584`/`9331`) without overclaiming broader duplication, and notes that no PRs are present. The soft-edge rejections are appropriately cautious and do not overstate evidence." + "accept": false, + "feedback": "Rejecting because one soft-edge verdict overstates the evidence: issue #9331 and #9417 share a function name, but the packet shows a bug report versus a performance suggestion, not the same underlying defect. The rest of the pairing judgments are mostly conservative, and the inpainting umbrella choice is plausible, but this accept is too strong for the available evidence." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -62013,19 +62015,20 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10111", "issue:10231", "issue:10897", - "issue:11555", "issue:11592", "issue:1185", "issue:12108", + "issue:12605", "issue:12653", + "issue:13377", "issue:1417", "issue:1708", "issue:1857", @@ -62041,10 +62044,10 @@ "issue:3769", "issue:4390", "issue:4480", - "issue:4492", - "issue:5402", "issue:6112", "issue:6256", + "issue:6385", + "issue:6506", "issue:6533", "issue:6901", "issue:7082", @@ -62053,7 +62056,6 @@ "issue:775", "issue:834", "issue:8410", - "issue:8551", "issue:8593", "issue:8604", "issue:8832", @@ -62068,6 +62070,8 @@ "issue:984" ], "soft_pairs": [ + "issue:12605|issue:13377", + "issue:6385|issue:6506", "issue:2906|issue:7082", "issue:6256|issue:9610", "issue:2507|issue:834", @@ -62090,195 +62094,196 @@ "issue:2984|issue:3444", "issue:9171|issue:9719", "issue:4390|issue:6901", - "issue:2532|issue:3769", - "issue:4492|issue:5402", - "issue:11555|issue:8551" + "issue:2532|issue:3769" ], - "prepared_review_unit_hash": "e3f7d9a6fbf11d7010f9f5a2258bf37bf6007e33b9f2483d357dad441ed6d991", + "prepared_review_unit_hash": "c1d21b56e926fe5d4f8834a7b6db046c18097fa27bab2dbb1a57cd85373e9f60", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34059, - "estimated_input_tokens": 8515, - "estimated_eval_tokens": 17286 + "serialized_chars": 34126, + "estimated_input_tokens": 8532, + "estimated_eval_tokens": 17320 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate cluster emerges here: the items cover many unrelated bugs, regressions, and feature requests. All soft pairs look like false positives rather than the same underlying change or defect.", - "confidence": 0.96, - "canonical_issue_reason": "#10111 is the broadest umbrella-style issue, but there is no real duplicate hub; the rest of the items are distinct problems and should not be merged into it.", + "summary": "This cluster is mostly heterogeneous: the soft matches are broad topical similarities, not true duplicates. The only clear duplicate-like pair is the StableDiffusionPanoramaPipeline get_views / quartered-image bug.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 2532 is the best root-cause anchor: it names the faulty `get_views` method, while 3769 is the downstream symptom report for the same panorama bug.", "canonical_pr_reason": null, - "best_issue_reason": "#10111 is the closest thing to a representative issue because it is generic and pipeline-level, but it is still too vague to be a precise canonical for this set.", + "best_issue_reason": "Issue 2532 is the most actionable representative in the set because it points to the underlying code path rather than only the visual failure mode.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12605", + "right": "issue:13377", + "accept": false, + "reason": "Both are attention/mask-related FLUX bugs, but they affect different code paths: backend attention dispatch vs CFG being skipped when `mask` is `None`." + }, + { + "left": "issue:6385", + "right": "issue:6506", + "accept": false, + "reason": "Same broad StableVideoDiffusion area, but one is VRAM usage and the other is output type mismatch; not the same bug." + }, { "left": "issue:2906", "right": "issue:7082", "accept": false, - "reason": "Different import failures: TextualInversionLoaderMixin missing PreTrainedTokenizer vs CpuOffload import from accelerate.hooks." + "reason": "Both are import/type-related failures, but they concern unrelated symbols (`PreTrainedTokenizer` vs `CpuOffload`) and different subsystems." }, { "left": "issue:6256", "right": "issue:9610", "accept": false, - "reason": "Different feature requests: step-begin callbacks in pipelines vs adding load_lora_weights to one specific img2img file." + "reason": "Both are pipeline API feature requests, but they ask for different callbacks/methods and do not describe the same change." }, { "left": "issue:2507", "right": "issue:834", "accept": false, - "reason": "Performance complaints in different areas: safetensors GPU loading vs CPU under-utilization; not the same bug." + "reason": "Both mention performance, but one is safetensors load speed and the other is CPU underutilization during runtime; different problems." }, { "left": "issue:11592", "right": "issue:8832", "accept": false, - "reason": "Unrelated regressions: FluxLoraLoaderMixin delete_adapters vs a generic crash at pipe(prompt).images[0]." + "reason": "One is a FLUX LoRA adapter regression; the other is a generic crash when calling `pipe(prompt)`. No concrete shared code path." }, { "left": "issue:3728", "right": "issue:6533", "accept": false, - "reason": "Different requests: new shifted-diffusion pipeline vs Marigold monocular depth estimation pipeline." + "reason": "Both are requests for new pipelines, but they target different models/tasks and are not the same feature." }, { "left": "issue:1934", "right": "issue:984", "accept": false, - "reason": "Different code paths: cpu_offload VRAM usage vs F.interpolate breaking for large batch sizes." + "reason": "Unrelated performance/memory complaints in different code paths; no evidence of a shared underlying defect." }, { "left": "issue:7492", "right": "issue:9209", "accept": false, - "reason": "Repository lookup failure for a model path vs inability to connect to hf-mirror.com; different networking/model resolution issues." + "reason": "Both involve loading models, but one is a missing repo error and the other is a network/mirror connectivity issue." }, { "left": "issue:12108", "right": "issue:9460", "accept": false, - "reason": "Both mention pipeline/model integration, but one is a flow-matching scheduler bug and the other is ControlNet list assignment support." + "reason": "Different pipelines and different failures: flow-matching scheduler breakage vs assigning multiple ControlNet models." }, { "left": "issue:12108", "right": "issue:12653", "accept": false, - "reason": "Different defects: flow-matching scheduler behavior vs PRX rope float64 causing MPS/NPU issues." + "reason": "Completely different problems: scheduler compatibility vs RoPE dtype causing MPS/NPU issues." }, { "left": "issue:10111", "right": "issue:9516", "accept": false, - "reason": "#10111 is a vague umbrella, but #9516 is a specific joint_attention_kwargs propagation bug in FLUX transformers." + "reason": "Both mention FLUX, but one is a vague multi-bug report and the other is a specific missing `joint_attention_kwargs` pass-through; not the same bug." }, { "left": "issue:2784", "right": "issue:7380", "accept": false, - "reason": "Different topics: Dreambooth support for text2video vs one-step diffusion/distillation implementation." + "reason": "Both are feature requests, but they propose different diffusion capabilities and are not duplicates." }, { "left": "issue:4480", "right": "issue:8604", "accept": false, - "reason": "Both involve float16/MPS-related image problems, but one is a kernel crash in StableDiffusionPipeline and the other is T5 float16 corruption." + "reason": "Both are fp16/MPS-related image corruption reports, but they involve different models/components and different failure modes." }, { "left": "issue:1185", "right": "issue:1708", "accept": false, - "reason": "Dreambooth reproducibility vs passing a YAML config for inference; not the same underlying problem." + "reason": "Dreambooth is the only shared topic; one is reproducibility and the other is config-file usage. Not the same issue." }, { "left": "issue:337", "right": "issue:775", "accept": false, - "reason": "Both are old generation bugs, but the symptoms and reported triggers differ; not clearly the same defect." + "reason": "Both are generation bugs, but one concerns img2img/inpainting behavior and the other fp16 without autocast." }, { "left": "issue:10231", "right": "issue:9023", "accept": false, - "reason": "Different runtime failures: tutorial ImportError on Ubuntu 20.04 vs a general sample code execution failure." + "reason": "Both are generic run failures, but they are too unspecific and do not point to the same underlying problem." }, { "left": "issue:6112", "right": "issue:8593", "accept": false, - "reason": "Both produce noise outputs, but they concern different models and environments (SDXL-Turbo vs SD3 on CPU bf16)." + "reason": "Both produce noise, but on different models and precisions; the bug reports do not indicate the same root cause." }, { "left": "issue:9320", "right": "issue:9719", "accept": false, - "reason": "Finetuned model save/load regression vs progress-bar suppression during checkpoint loading; different behaviors." + "reason": "Saved/loaded finetuned weights in custom diffusion is unrelated to a progress-bar suppression regression." }, { "left": "issue:10897", "right": "issue:8410", "accept": false, - "reason": "Different feature requests for different model families: SkyReels-A1 support vs native TextDiffuser pipelines." + "reason": "Both are feature requests for new model support, but they are about different pipelines/tasks." }, { "left": "issue:1417", "right": "issue:1857", "accept": false, - "reason": "CPU offloading implementation complaint vs UnCLIP pipeline test coverage; unrelated issues." + "reason": "CPU offloading implementation concerns are unrelated to UnCLIP test coverage." }, { "left": "issue:2984", "right": "issue:3444", "accept": false, - "reason": "Different pipeline work: latent video diffusion models vs ImageBind + StableUnCLIPImg2Img audio-to-image integration." + "reason": "Both request new pipeline/model support, but they are for different modalities and concrete changes." }, { "left": "issue:9171", "right": "issue:9719", "accept": false, - "reason": "One is a v_prediction regression in from_single_file, the other is disable_progress_bar being ignored during shard loading." + "reason": "A single-file loading regression for SD 1.5 `v_prediction` is unrelated to `disable_progress_bar` being ignored." }, { "left": "issue:4390", "right": "issue:6901", "accept": false, - "reason": "Missing requirements/setup issue vs a missing clip_sample implementation in FlaxDDIMScheduler; not the same bug." + "reason": "Missing requirements vs missing `clip_sample` implementation: both are 'missing' in name only, not the same defect." }, { "left": "issue:2532", "right": "issue:3769", - "accept": false, - "reason": "Different panorama pipeline bugs: get_views method behavior vs quartered image output." - }, - { - "left": "issue:4492", - "right": "issue:5402", - "accept": false, - "reason": "A Python literal NameError vs a missing compute_snr import; unrelated failures." - }, - { - "left": "issue:11555", - "right": "issue:8551", - "accept": false, - "reason": "Both involve device_map, but one asks for auto support and the other reports wrong-device latent transfer with balanced mapping." + "accept": true, + "reason": "These describe the same StableDiffusionPanoramaPipeline bug: a `get_views` defect causing the quartered/incorrect panorama output." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The evaluation is grounded in the packet and stays conservative: the rejected soft pairs are only matched on broad topical similarity, while issue:2532 vs issue:3769 is a plausible same-bug pair based on the panorama/get_views titles. No overstatement that would require rejection." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -62290,8 +62295,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -62302,13 +62307,12 @@ "issue:10833", "issue:10883", "issue:10893", - "issue:11045", "issue:11144", "issue:11283", "issue:11352", "issue:11547", + "issue:11555", "issue:11674", - "issue:11721", "issue:12110", "issue:12157", "issue:12221", @@ -62316,19 +62320,19 @@ "issue:12547", "issue:13221", "issue:13261", - "issue:1963", "issue:2016", "issue:2319", "issue:2532", "issue:3576", "issue:4080", "issue:4141", + "issue:4492", "issue:4639", "issue:5006", "issue:5028", "issue:5118", "issue:5128", - "issue:5392", + "issue:5402", "issue:5441", "issue:5463", "issue:5749", @@ -62337,6 +62341,7 @@ "issue:6095", "issue:6281", "issue:7629", + "issue:8551", "issue:9031", "issue:9038", "issue:9551", @@ -62344,6 +62349,8 @@ "issue:9619" ], "soft_pairs": [ + "issue:4492|issue:5402", + "issue:11555|issue:8551", "issue:1231|issue:2016", "issue:10883|issue:9551", "issue:5118|issue:5866", @@ -62365,182 +62372,180 @@ "issue:4639|issue:5441", "issue:2319|issue:2532", "issue:3576|issue:5028", - "issue:10833|issue:11352", - "issue:11045|issue:11721", - "issue:1963|issue:5392" + "issue:10833|issue:11352" ], - "prepared_review_unit_hash": "93b3b302d559552fcdb96574ebe6cc4f4bbc1377546a395de6c78d0982ee9892", + "prepared_review_unit_hash": "8ad9cc2dee261c6ba8d915d03a182bfc4aea49119f86cbb1e0b62b93c6ac57b8", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33387, - "estimated_input_tokens": 8347, - "estimated_eval_tokens": 16950 + "serialized_chars": 33512, + "estimated_input_tokens": 8378, + "estimated_eval_tokens": 17012 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated diffusers issues and feature requests, mostly from different pipelines/models. The soft pairs share only broad keywords like \"controlnet\", \"flax\", \"index error\", or \"quantization\", but not the same concrete bug or change path, so none should be merged as duplicates.", - "confidence": 0.92, - "canonical_issue_reason": "No single issue cleanly represents a duplicate cluster here; the items are too heterogeneous.", + "summary": "The cluster is heterogeneous: it mixes unrelated feature requests, import errors, training-script bugs, and pipeline-specific regressions. None of the soft pairs look like the same underlying issue or change, so all soft edges are rejected.", + "confidence": 0.95, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:10833 is the most concrete, bug-focused report in the set, with a specific failure mode and enough detail to serve as a representative anchor if one must be chosen.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4492", + "right": "issue:5402", + "accept": false, + "reason": "Different problems: a JSON/boolean typo (`false` vs `False`) versus a missing `compute_snr` import. No shared code path." + }, + { + "left": "issue:11555", + "right": "issue:8551", + "accept": false, + "reason": "Both mention `device_map`, but one is a broad feature question and the other is a specific VAE device-placement bug in an inpaint pipeline." + }, { "left": "issue:1231", "right": "issue:2016", "accept": false, - "reason": "Both are flax-related, but one is a scheduler request and the other is a sampling regression; different code paths and symptoms." + "reason": "Both are Flax-related, but one is a scheduler request and the other is a sampling regression. Not the same bug/change." }, { "left": "issue:10883", "right": "issue:9551", "accept": false, - "reason": "Marigold update/feature request vs. Flux ControlNet usage question; no shared underlying bug or change." + "reason": "Different feature requests in different model families; no shared concrete implementation target." }, { "left": "issue:5118", "right": "issue:5866", "accept": false, - "reason": "Both mention ControlNet, but one asks about reference_only usage and the other is about AnimateDiff + ControlNet behavior; not the same issue." + "reason": "Both touch ControlNet usage, but one is about reference-only support and the other is AnimateDiff+ControlNet integration. Different code paths." }, { "left": "issue:5749", "right": "issue:6095", "accept": false, - "reason": "Both are IndexError reports, but they affect different pipelines and likely different failing indices/code paths." + "reason": "Both are IndexErrors, but they affect different pipelines and likely different scheduler logic. Too broad to merge." }, { "left": "issue:10100", "right": "issue:12221", "accept": false, - "reason": "ConsisID support and DiffSynth ControlNet support are separate feature requests for different integrations." + "reason": "Both are feature requests, but for unrelated models/integrations. No duplicate underlying change." }, { "left": "issue:10893", "right": "issue:9619", "accept": false, - "reason": "Both are ControlNet feature requests, but one is CogVideoX integration and the other is interrupt support in StableDiffusionControlNetPipeline." + "reason": "One requests CogVideoX ControlNet support; the other requests interrupt support for StableDiffusionControlNetPipeline. Unrelated changes." }, { "left": "issue:13221", "right": "issue:13261", "accept": false, - "reason": "Both concern LoRA on newer models, but they target different model families and different failure modes." + "reason": "Both mention LoRA, but one is a general support issue and the other is a specific FLUX.2-klein-9B LoRA/Lokr compatibility problem." }, { "left": "issue:4141", "right": "issue:6281", "accept": false, - "reason": "One is a bug in multi-ControlNet inpainting behavior; the other is an import error for MultiControlNetModel." + "reason": "One is a MultiControlNet behavior bug; the other is an import error for `MultiControlNetModel`. Related area, but not the same defect." }, { "left": "issue:11144", "right": "issue:9576", "accept": false, - "reason": "Flax dtype initialization bug vs. safetensors propagation bug during pipeline loading; unrelated paths." + "reason": "Different concerns: dtype initialization in Flax UNet versus `use_safetensors` propagation when loading pipelines." }, { "left": "issue:11283", "right": "issue:11674", "accept": false, - "reason": "Wan deserialization crash vs. a feature request for Wan VACE inputs; not the same bug/change." + "reason": "Same Wan ecosystem, but one is a deserialization failure and the other is a feature request for VACE inputs. Not the same issue." }, { "left": "issue:5128", "right": "issue:5463", "accept": false, - "reason": "Loading a local safetensors file and an UnboundLocalError on image_latents are different problems." + "reason": "Unrelated errors: loading a local safetensors file versus an `image_latents` uninitialized variable." }, { "left": "issue:12157", "right": "issue:12547", "accept": false, - "reason": "Different training-time errors in different workflows/models; no shared concrete root cause." + "reason": "Both are training-time errors, but caused by different scripts and different root causes (`torch.device` iterable bug vs missing `--instance_prompt`)." }, { "left": "issue:10111", "right": "issue:10518", "accept": false, - "reason": "Both are bug reports, but they concern different pipelines and distinct failure modes." + "reason": "Different pipelines and different bug reports; no evidence they share the same root cause." }, { "left": "issue:10760", "right": "issue:12110", "accept": false, - "reason": "Import error for a legacy pipeline class vs. UNet1DModel initialization bug; unrelated." + "reason": "Different code paths and error types: import failure for an ONNX pipeline class versus UNet1D initialization bug." }, { "left": "issue:4080", "right": "issue:5947", "accept": false, - "reason": "Pipeline feature request vs. policy/research objective; not a duplicate bug or change." + "reason": "A new SDXL inpainting pipeline request is unrelated to a Diffusion-DPO policy objective request." }, { "left": "issue:9031", "right": "issue:9038", "accept": false, - "reason": "Same Flax pipeline family, but one asks to disable safety checker and the other asks for prompt weighting." + "reason": "Both are Flax feature questions, but one is about disabling the safety checker and the other about prompt weighting. Separate features." }, { "left": "issue:5006", "right": "issue:7629", "accept": false, - "reason": "Both mention OOM, but they are different models, versions, and triggering conditions." + "reason": "Both involve OOM, but in different models, versions, and circumstances. Not a single concrete bug." }, { "left": "issue:10893", "right": "issue:11547", "accept": false, - "reason": "Both relate to Flux/ControlNet, but one is CogVideoX ControlNet integration and the other is Flux Fill dev support." + "reason": "Both reference ControlNet, but they target different model families and different feature requests." }, { "left": "issue:4639", "right": "issue:5441", "accept": false, - "reason": "Different training scripts and unrelated errors; no shared code path." + "reason": "Different training scripts and failures: dataloader multiprocessing versus missing `attn_processors` on WuerstchenPrior." }, { "left": "issue:2319", "right": "issue:2532", "accept": false, - "reason": "A typo in a method doc/code name is not the same as a panorama get_views bug." + "reason": "A typo fix in one method is unrelated to a panorama pipeline `get_views` bug." }, { "left": "issue:3576", "right": "issue:5028", "accept": false, - "reason": "Different TypeErrors in different pipelines and mechanisms; not mergeable as one fix." + "reason": "Both are TypeErrors, but caused by different mechanisms and pipeline implementations. Not mergeable as one fix." }, { "left": "issue:10833", "right": "issue:11352", "accept": false, - "reason": "Both involve bitsandbytes/quantization, but one is a matrix shape failure and the other is a missing warning when moving to CPU." - }, - { - "left": "issue:11045", - "right": "issue:11721", - "accept": false, - "reason": "Both are load crashes, but on different models/pipelines with different likely causes." - }, - { - "left": "issue:1963", - "right": "issue:5392", - "accept": false, - "reason": "One is a latent unconditional generation issue, the other is a how-to request about training unconditional latent diffusion; not duplicates." + "reason": "Both involve bitsandbytes quantization, but one is a matrix-shape runtime failure and the other is a missing CPU-move warning. Different outcomes and fixes." } ] }, @@ -62567,25 +62572,26 @@ "issue:10302", "issue:10467", "issue:10757", - "issue:11422", + "issue:11045", + "issue:11721", "issue:12308", "issue:1304", "issue:1335", "issue:1434", "issue:1519", "issue:1672", + "issue:1963", "issue:3301", "issue:3624", - "issue:3651", "issue:4331", "issue:4349", "issue:4436", "issue:463", "issue:4880", "issue:5112", + "issue:5392", "issue:5465", "issue:5474", - "issue:5523", "issue:5873", "issue:6054", "issue:6087", @@ -62606,7 +62612,6 @@ "issue:8997", "issue:9147", "issue:9297", - "issue:9393", "issue:9397", "issue:9505", "issue:9713", @@ -62614,6 +62619,8 @@ "issue:9941" ], "soft_pairs": [ + "issue:11045|issue:11721", + "issue:1963|issue:5392", "issue:1519|issue:785", "issue:8523|issue:8997", "issue:3301|issue:4331", @@ -62635,182 +62642,180 @@ "issue:5465|issue:7049", "issue:7138|issue:8744", "issue:9147|issue:9941", - "issue:7682|issue:9297", - "issue:11422|issue:9393", - "issue:3651|issue:5523" + "issue:7682|issue:9297" ], - "prepared_review_unit_hash": "3b84d893568930b3cab7d794b3b4da2bb9830a88854e53d0e4bfa38551991053", + "prepared_review_unit_hash": "248dc76b6833d784100e959c26b24b49eb928e510e30b122b36e0c0ff77f927c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34614, - "estimated_input_tokens": 8654, - "estimated_eval_tokens": 17564 + "serialized_chars": 34590, + "estimated_input_tokens": 8648, + "estimated_eval_tokens": 17552 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is not a duplicate cluster; the items cover many unrelated bugs, feature requests, and environment-specific failures across training, inference, conversion, schedulers, and pipelines. All soft-edge pairs should be rejected.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "This cluster is mostly a grab bag of unrelated issues across different pipelines, training scripts, schedulers, conversion utilities, and model families. I do not see a true duplicate set here, so all soft pairs are rejected.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical issue stands out because the items are not the same underlying bug; they cover distinct failures and feature requests across different code paths.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 11045 is the most concrete, self-contained bug report in the set: it names a specific crash, a specific model, and a specific training script, making it a good representative issue if one must be chosen.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11045", + "right": "issue:11721", + "accept": false, + "reason": "Both are load-time crashes, but they involve different pipelines/models (Flux Schnell vs SanaPipeline) and different failure contexts." + }, + { + "left": "issue:1963", + "right": "issue:5392", + "accept": false, + "reason": "Same broad topic of unconditional latent diffusion, but one is a generation issue and the other is a how-to/training question; not the same bug." + }, { "left": "issue:1519", "right": "issue:785", "accept": false, - "reason": "No: noisy batch outputs on Apple M2 is an inference correctness bug, while PNDMScheduler cost is a performance issue in a different code path." + "reason": "No overlap in underlying problem: noisy batch outputs on Apple M2 vs scheduler performance cost." }, { "left": "issue:8523", "right": "issue:8997", "accept": false, - "reason": "No: generic CUDA error in a playground app does not match a Kolors scheduler integration error; different models and failure modes." + "reason": "Different failures in different models/code paths; only a superficial \u201cCUDA error / scheduler error\u201d similarity." }, { "left": "issue:3301", "right": "issue:4331", "accept": false, - "reason": "No: one is a broken DreamBooth LoRA training script, the other reports differing results between DreamBooth and DreamBooth-LoRA; not the same bug." + "reason": "Both involve DreamBooth/LoRA training, but one is a generic failure report and the other is a behavioral discrepancy between two scripts." }, { "left": "issue:10757", "right": "issue:8939", "accept": false, - "reason": "No: variable naming in a conversion script is unrelated to tokenizer loading failure for SD3." + "reason": "Unrelated: variable-name fix in a conversion script vs tokenizer loading failure for an SD3 model." }, { "left": "issue:1672", "right": "issue:6087", "accept": false, - "reason": "No: wrong image size after ckpt-to-diffusers conversion is unrelated to saving/loading LoRA config and lora_alpha inference mismatch." + "reason": "Different subsystems and symptoms: checkpoint conversion image size mismatch vs LoRA config persistence causing wrong lora_alpha." }, { "left": "issue:10302", "right": "issue:10467", "accept": false, - "reason": "No: FP8 inference noise is a runtime precision issue, while the FLUX example issue appears to be an example/code-generation artifact." + "reason": "FP8 inference noise is a runtime quality bug; the other appears to be about example-code/file-name output and is not the same issue." }, { "left": "issue:1335", "right": "issue:3624", "accept": false, - "reason": "No: missing vae_encoder in ONNX pipeline init is a constructor/API mismatch, not the same as getattr attribute-name type error." + "reason": "Different TypeErrors in different code paths; no evidence they share the same underlying defect." }, { "left": "issue:4880", "right": "issue:6054", "accept": false, - "reason": "No: ControlNet image argument validation is unrelated to an EulerDiscreteScheduler torch.from_numpy misuse." + "reason": "Different pipeline bugs: ControlNet image validation vs Euler scheduler timestep handling." }, { "left": "issue:5474", "right": "issue:9713", "accept": false, - "reason": "No: a proposed new reference-only pipeline is a feature request, not a controlnet-training implementation proposal." + "reason": "A proposed new pipeline vs a controlnet-training implementation request; not duplicate bugs." }, { "left": "issue:5873", "right": "issue:9902", "accept": false, - "reason": "No: ONNX export support for scaled_dot_product_attention is unrelated to a conda version being stuck." + "reason": "ONNX export failure is unrelated to the conda version/package-management issue." }, { "left": "issue:5112", "right": "issue:6283", "accept": false, - "reason": "No: SDXL ControlNet img2img dtype/shape error is different from strength=0 behavior in the same pipeline family." + "reason": "Both are SDXL ControlNet img2img issues, but one is a float/shape TypeError and the other is a strength=0 edge-case; not clearly the same bug." }, { "left": "issue:6095", "right": "issue:8565", "accept": false, - "reason": "No: SVD scheduling index error and StableDiffusion3 LoRA load IndexError are different pipelines and different failure points." + "reason": "Both mention IndexError, but they affect different pipelines and code paths (SVD scheduling vs SD3 LoRA loading)." }, { "left": "issue:6162", "right": "issue:7611", "accept": false, - "reason": "No: IP-Adapter shape mismatch in SDXL is not the same as a generic UNet tensor size mismatch." + "reason": "Both are tensor/shape mismatch reports, but they concern different models and different components; too broad to merge." }, { "left": "issue:1434", "right": "issue:4349", "accept": false, - "reason": "No: upscale OOM from large images is a memory issue, while the other is a missing text_encoder dependency despite prompt_embeds." + "reason": "Same upscale pipeline family, but one is an OOM scaling issue and the other is a missing text_encoder compatibility bug." }, { "left": "issue:6323", "right": "issue:9505", "accept": false, - "reason": "No: TensorRT guidance_scale attribute assignment problem is unrelated to a checkpoint key missing during upscaling." + "reason": "Completely different problems: a TensorRT attribute-setting issue vs a checkpoint key mismatch when upscaling." }, { "left": "issue:12308", "right": "issue:9397", "accept": false, - "reason": "No: flash-attn kernel support for non-Hopper GPUs is an implementation/support request, while SGMUniform scheduler support for Hyper LoRAs is a different feature request." + "reason": "Different feature requests in unrelated areas: flash-attn support vs scheduler support for Hyper LoRAs." }, { "left": "issue:1304", "right": "issue:463", "accept": false, - "reason": "No: a how-to question about text-guided img2img is not the same as a general super-resolution diffusion model request." + "reason": "A training-how-to question and a super-resolution model request are not duplicates." }, { "left": "issue:4436", "right": "issue:8708", "accept": false, - "reason": "No: converting CIVITAI safetensors to diffusers is unrelated to missing vae.config.shift_factor in SD3 dreambooth examples." + "reason": "Both involve VAE/diffusers context, but one is conversion from civitai safetensors and the other is a missing SD3 training config field." }, { "left": "issue:5465", "right": "issue:7049", "accept": false, - "reason": "No: DiffEdit inpaint_strength limitation is a parameter-handling issue; the IP adapter example error is a separate unpacking/shape bug." + "reason": "Different pipeline bugs with different failure modes; no shared concrete code-path problem." }, { "left": "issue:7138", "right": "issue:8744", "accept": false, - "reason": "No: SDXL UNet structure mismatch and loading from single file for training are different problems." + "reason": "SDXL UNet structure mismatch vs single-file loading for training are unrelated." }, { "left": "issue:9147", "right": "issue:9941", "accept": false, - "reason": "No: FLUX autopipeline error is unrelated to a generic stable diffusion colab failure." + "reason": "Flux autopipeline error and a generic Colab stable-diffusion error do not describe the same defect." }, { "left": "issue:7682", "right": "issue:9297", "accept": false, - "reason": "No: multi-GPU SD1.5 execution problems are different from linear LoRA load-time slowdown." - }, - { - "left": "issue:11422", - "right": "issue:9393", - "accept": false, - "reason": "No: validation prompt crashes HiDream LoRA training, while the other issue is about saving Flux1 Dreambooth LoRA models with deepspeed." - }, - { - "left": "issue:3651", - "right": "issue:5523", - "accept": false, - "reason": "No: multi-GPU DreamBooth training failure is not the same as nondeterministic results across runs." + "reason": "One is a multi-GPU runtime issue; the other is a LoRA loading performance complaint." } ] }, @@ -62829,12 +62834,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10025", + "issue:10282", "issue:10334", "issue:10336", "issue:10712", @@ -62844,7 +62850,9 @@ "issue:1100", "issue:11050", "issue:11137", + "issue:11169", "issue:11222", + "issue:11422", "issue:11519", "issue:11721", "issue:11724", @@ -62856,18 +62864,14 @@ "issue:2294", "issue:2444", "issue:2606", - "issue:2838", - "issue:2844", "issue:3338", - "issue:3679", - "issue:4005", + "issue:3651", "issue:4043", "issue:4116", "issue:4349", "issue:4399", - "issue:4779", - "issue:5009", "issue:5034", + "issue:5523", "issue:5897", "issue:5934", "issue:5947", @@ -62878,13 +62882,17 @@ "issue:8885", "issue:9248", "issue:9374", + "issue:9393", "issue:9471", "issue:9516", "issue:9958" ], "soft_pairs": [ + "issue:11422|issue:9393", + "issue:3651|issue:5523", "issue:1100|issue:2444", "issue:10952|issue:9471", + "issue:10282|issue:11169", "issue:10896|issue:11519", "issue:2294|issue:3338", "issue:4116|issue:7185", @@ -62903,183 +62911,180 @@ "issue:11137|issue:13300", "issue:11222|issue:11779", "issue:12195|issue:9248", - "issue:6866|issue:9958", - "issue:2838|issue:4005", - "issue:2844|issue:5009", - "issue:3679|issue:4779" + "issue:6866|issue:9958" ], - "prepared_review_unit_hash": "dbd9e95fe38b9ae64149961d03b8477c3fd5d8346ce66b29c39af70505c4a5b9", + "prepared_review_unit_hash": "c86a7d74b193c3fe1e73c30cef2ed60c371e92a32cedf320ed033856c0be3390", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33771, - "estimated_input_tokens": 8443, - "estimated_eval_tokens": 17142 + "serialized_chars": 33945, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is noisy: it mixes unrelated support requests, model-specific bugs, scheduler issues, and training/inference errors. None of the soft pairs look like the same underlying bug/change, so I would not collapse any of them as duplicates.", - "confidence": 0.79, - "canonical_issue_reason": "Issue 10025 is the most generic and central title here, directly naming the Flux transformer attention-mask problem and serving as the best anchor for the small attention/transformer subset.", + "summary": "These are mostly soft-similarity false positives: the linked issues cover unrelated bugs, feature requests, environment/import problems, and different model pipelines. None of the candidate pairs look like the same underlying defect/change, so no duplicates should be merged.", + "confidence": 0.98, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 10025 is the best overall representative because it is broad, concrete, and closest to the recurring transformer/attention-mask theme; the rest are mostly unrelated point bugs or feature requests.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11422", + "right": "issue:9393", + "accept": false, + "reason": "Both mention LoRA/training, but one is a validation-prompt crash in HiDream LoRA training while the other is a Flux DreamBooth LoRA save failure; different pipelines and failure modes." + }, + { + "left": "issue:3651", + "right": "issue:5523", + "accept": false, + "reason": "Multi-GPU train_dreambooth failure is unrelated to run-to-run inconsistency; one is a launch/runtime bug, the other is nondeterminism." + }, { "left": "issue:1100", "right": "issue:2444", "accept": false, - "reason": "Different failures: CLIPTextTransformer device attribute access vs a training script exiting nonzero. Same general area of errors, but not the same bug." + "reason": "CLIPTextTransformer device attribute error is a model-object bug, while the other is a generic subprocess exit in inpaint training; no same code path." }, { "left": "issue:10952", "right": "issue:9471", "accept": false, - "reason": "LoRA adapter-name mismatch in SDXL long-prompt weighting is unrelated to a CoreML file-extension/resource-type parsing error." + "reason": "LoRA adapter-name lookup failure is unrelated to file-extension inference on a model path." + }, + { + "left": "issue:10282", + "right": "issue:11169", + "accept": false, + "reason": "Checkpoint shape mismatch for x_embedder.weight is unrelated to the NumPy binary-incompatibility import error." }, { "left": "issue:10896", "right": "issue:11519", "accept": false, - "reason": "Both are support requests, but for different models (SkyReels-A1 vs MAGI-1) and different underlying work." + "reason": "These are separate model-support requests for different models, not the same change." }, { "left": "issue:2294", "right": "issue:3338", "accept": false, - "reason": "Import error for a missing pipeline symbol is not the same as slow diffusers import performance." + "reason": "Missing pipeline import and slow package import are different issues with different symptoms and fixes." }, { "left": "issue:4116", "right": "issue:7185", "accept": false, - "reason": "Embedding handling in StableDiffusionXLImg2ImgPipeline is unrelated to a question about PyTorch version support." + "reason": "SDXL embedding-passing bug is unrelated to PyTorch version support question." }, { "left": "issue:5947", "right": "issue:8482", "accept": false, - "reason": "Diffusion-DPO policy discussion is not the same as requesting Stable Diffusion 3 medium support." + "reason": "Policy-objective discussion is unrelated to adding support for a specific model." }, { "left": "issue:9374", "right": "issue:9958", "accept": false, - "reason": "AutoPipeline support for Flux pipelines and AutoencoderKLTemporalDecoder tiling are different feature requests in different code paths." + "reason": "AutoPipeline FLUX support and Autoencoder tiling support touch different features and code paths." }, { "left": "issue:4043", "right": "issue:4349", "accept": false, - "reason": "Both mention prompt embeddings, but one is an SDXL API requirement and the other is a pipeline assumption about text_encoder presence; not the same concrete bug." + "reason": "Both involve embeddings in SDXL-family pipelines, but one is about requiring pooled embeddings and the other about omitting text_encoder; not the same concrete bug." }, { "left": "issue:11721", "right": "issue:8885", "accept": false, - "reason": "SanaPipeline load crash is unrelated to a Custom Diffusion example dependency/install issue." + "reason": "SanaPipeline load crash is unrelated to clip-retrieval installation/modifier-token loading." }, { "left": "issue:10334", "right": "issue:10336", "accept": false, - "reason": "Both are Sana-related, but one reports NaNs/gray images on MacOS and the other reports exploding activations and clamping; too different to merge." + "reason": "MacOS/MPS gray images and CPU NaNs versus activation explosion/clamping are related at best, but not clearly the same defect." }, { "left": "issue:11907", "right": "issue:12771", "accept": false, - "reason": "Both involve schedulers, but one is a general Flux compatibility complaint while the other is a specific out-of-bounds bug for squaredcos_cap_v2 with Karras sigmas." + "reason": "Scheduler incompatibility in FLUX with two schedulers is broader and different from the specific DPM solver out-of-bounds bug with Karras sigmas." }, { "left": "issue:2606", "right": "issue:5897", "accept": false, - "reason": "DeepSpeed save_state hang during LoRA training is a different failure mode from a CUDA/CPU device mismatch during training." + "reason": "DeepSpeed save_state hang and tensor device mismatch are distinct training failures." }, { "left": "issue:5034", "right": "issue:6351", "accept": false, - "reason": "Incorrect progress-bar display while loading checkpoints is unrelated to DistributedDataParallel lacking disable_adapters." + "reason": "Checkpoint progress-bar behavior is unrelated to DDP missing disable_adapters." }, { "left": "issue:4399", "right": "issue:5934", "accept": false, - "reason": "PyTorch compile plus CPU offload incompatibility is not the same as a broader compile error report." + "reason": "Both mention torch.compile, but one is specifically compile plus CPU offload, while the other is a general compile error." }, { "left": "issue:10025", "right": "issue:9516", "accept": false, - "reason": "Both touch FLUX transformer internals, but one is about attention masks and the other about joint_attention_kwargs not being forwarded; different code-path issues." + "reason": "Both are FLUX-transformer keyword/attention topics, but one is about attention-mask support and the other about joint_attention_kwargs not being forwarded; not the same bug." }, { "left": "issue:10712", "right": "issue:11050", "accept": false, - "reason": "RuntimeError with StableDiffusion3 prompt_embeds is distinct from a training-time dtype casting omission for prompt_embeds/pooled_prompt_embeds." + "reason": "RuntimeError using prompt_embeds in SD3 inference is different from bf16/fp16 dtype casting in a training script." }, { "left": "issue:10954", "right": "issue:11724", "accept": false, - "reason": "FLUX LoRA load failure is unrelated to Chroma attention masking in the transformer." + "reason": "A FLUX LoRA loading failure is unrelated to Chroma attention-mask behavior in the transformer." }, { "left": "issue:11137", "right": "issue:13300", "accept": false, - "reason": "Latte dtype mismatch and CPU generation failures for SD3.5/FLUX are different models and different symptoms." + "reason": "A Latte dtype mismatch and CPU failures in SD3.5/FLUX are different models and likely different root causes." }, { "left": "issue:11222", "right": "issue:11779", "accept": false, - "reason": "WAN LoRA support request is not the same as a WAN VACE mask-padding bug." + "reason": "WAN LoRA support request and Wan VACE mask-padding bug are different scope: feature request vs concrete pipeline bug." }, { "left": "issue:12195", "right": "issue:9248", "accept": false, - "reason": "PyTorch custom_op incompatibility and a generic install/upgrade note are not the same bug." + "reason": "PyTorch custom_op compatibility failure is unrelated to a pip install/upgrade instruction issue." }, { "left": "issue:6866", "right": "issue:9958", "accept": false, - "reason": "IP-Adapter support for an SDXL ControlNet inpaint pipeline is unrelated to tiling support in AutoencoderKLTemporalDecoder." - }, - { - "left": "issue:2838", - "right": "issue:4005", - "accept": false, - "reason": "A request to support ControlNet without control is unrelated to StableDiffusionInpaintPipeline aspect-ratio preservation." - }, - { - "left": "issue:2844", - "right": "issue:5009", - "accept": false, - "reason": "Explicitly passing image embeddings to StableUnCLIPImg2ImgPipeline is a different problem from a Wuerstchen error." - }, - { - "left": "issue:3679", - "right": "issue:4779", - "accept": false, - "reason": "A feature request for watermark-free text-to-video models is not the same as a ModelScope image-to-video pipeline request." + "reason": "IP-Adapter support for an SDXL ControlNet inpaint pipeline is unrelated to AutoencoderKLTemporalDecoder tiling support." } ] }, @@ -63105,14 +63110,11 @@ "nodes": [ "issue:10057", "issue:10227", - "issue:10231", "issue:10492", "issue:10514", "issue:10705", "issue:10952", "issue:11570", - "issue:1261", - "issue:1314", "issue:13191", "issue:13232", "issue:1536", @@ -63120,10 +63122,16 @@ "issue:177", "issue:181", "issue:2712", + "issue:2838", + "issue:2844", "issue:3284", + "issue:3679", "issue:3799", + "issue:4005", "issue:4542", "issue:4672", + "issue:4779", + "issue:5009", "issue:5168", "issue:5284", "issue:5325", @@ -63135,9 +63143,7 @@ "issue:6096", "issue:6141", "issue:6444", - "issue:6548", "issue:6593", - "issue:6657", "issue:6809", "issue:6870", "issue:7629", @@ -63147,12 +63153,14 @@ "issue:8429", "issue:8801", "issue:8834", - "issue:8975", "issue:9648", "issue:9904", "issue:9972" ], "soft_pairs": [ + "issue:2838|issue:4005", + "issue:2844|issue:5009", + "issue:3679|issue:4779", "issue:10057|issue:7982", "issue:4542|issue:6593", "issue:4672|issue:8834", @@ -63173,193 +63181,187 @@ "issue:181|issue:3284", "issue:2712|issue:6444", "issue:10705|issue:9904", - "issue:1536|issue:5645", - "issue:6548|issue:6657", - "issue:1261|issue:1314", - "issue:10231|issue:8975" + "issue:1536|issue:5645" ], - "prepared_review_unit_hash": "296a9219e20cb923ec97e17fbf409cadb50603d83b87d5b03e660b78985c5358", + "prepared_review_unit_hash": "edccaa218385e273f50043cfb56194ffdec3ad597eb6e07443c8d4c78c49f831", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34064, - "estimated_input_tokens": 8516, - "estimated_eval_tokens": 17288 + "serialized_chars": 33822, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly heterogeneous: almost all soft pairs are distinct bugs or feature requests that only share a broad subsystem. The only plausible duplicate is the xformers backward-pass failure pair (1261/1314).", - "confidence": 0.88, - "canonical_issue_reason": "issue:1261 is the best canonical anchor because it names the concrete xformers backward-pass assert and is the older, more specific report; issue:1314 reads like a platform-specific repro of the same failure.", + "summary": "These items do not form a duplicate cluster: they span many unrelated bugs, feature requests, and model-support issues across different pipelines and code paths. All soft edges should be rejected.", + "confidence": 0.99, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:1261 is the strongest global issue candidate: it is specific, actionable, and likely the duplicate target for the one accepted soft edge.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2838", + "right": "issue:4005", + "accept": false, + "reason": "Different requests: ControlNet pipeline behavior vs inpaint aspect-ratio preservation." + }, + { + "left": "issue:2844", + "right": "issue:5009", + "accept": false, + "reason": "Different runtime failures in different models/pipelines; no shared code-path bug." + }, + { + "left": "issue:3679", + "right": "issue:4779", + "accept": false, + "reason": "A model announcement/request vs an image-to-video pipeline feature request." + }, { "left": "issue:10057", "right": "issue:7982", "accept": false, - "reason": "Different problems: a FluxTransformer2DModel.from_single_file token issue vs a from_pipe/MultiControlNetModel compatibility bug." + "reason": "Different APIs and failures: from_single_file token issue vs MultiControlNet from_pipe bug." }, { "left": "issue:4542", "right": "issue:6593", "accept": false, - "reason": "One is about StableDiffusionXLImg2ImgPipeline only producing one image; the other is a request to add load_lora_weights support." + "reason": "One is about generating multiple images; the other is LoRA loading support for SDXL img2img." }, { "left": "issue:4672", "right": "issue:8834", "accept": false, - "reason": "VRAM reduction for sdxl_lora training is unrelated to a request for SD3 ControlNet training code." + "reason": "Training VRAM reduction vs a request to release SD3 ControlNet training code." }, { "left": "issue:6870", "right": "issue:9648", "accept": false, - "reason": "Both mention enable_vae_tiling, but one reports a missing option in img2img/inpaint while the other reports memory overflow when it is enabled; not the same concrete bug." + "reason": "Missing enable_vae_tiling support vs memory overflow when using it; related topic but not the same bug." }, { "left": "issue:5399", "right": "issue:8429", "accept": false, - "reason": "Custom inpainting dtype mismatch is unrelated to the I2VGen-XL inference script malfunction." + "reason": "Different pipelines and symptoms: dtype mismatch in inpainting vs I2VGen-XL script issues." }, { "left": "issue:10492", "right": "issue:6809", "accept": false, - "reason": "torchao quantize/lora fuse error and PEFT LoRA adapter-loading regression are different failure modes with different fixes." + "reason": "Different LoRA/quantization problems with different failure modes and code paths." }, { "left": "issue:5168", "right": "issue:8410", "accept": false, - "reason": "ControlNet-XS support and native TextDiffuser support are separate feature requests." + "reason": "Support requests for different model families; not the same underlying change." }, { "left": "issue:5284", "right": "issue:8801", "accept": false, - "reason": "ControlNet-LLLite support is unrelated to Kolors support." + "reason": "ControlNet-LLLite support and Kolors support are separate model integrations." }, { "left": "issue:6141", "right": "issue:7877", "accept": false, - "reason": "ControlNet control_mode/resize_mode support is not the same as adding out_dim to Attention blocks." + "reason": "Unrelated features: ControlNet parameters vs Attention out_dim argument." }, { "left": "issue:5963", "right": "issue:9972", "accept": false, - "reason": "Kandinsky embeds-vs-prompts failure is unrelated to a CogX MacOS buffer issue." + "reason": "Kandinsky embeds-vs-prompts runtime error is unrelated to CogX memory allocation on macOS." }, { "left": "issue:10952", "right": "issue:11570", "accept": false, - "reason": "Both concern LoRA loading, but one is an adapter-name mismatch and the other is a bitsandbytes .to() cast problem; different code paths." + "reason": "Both involve LoRA loading, but they are different bugs in different paths and not one mergeable fix." }, { "left": "issue:172", "right": "issue:177", "accept": false, - "reason": "Progress-bar options and documentation for subfolder are unrelated." + "reason": "Progress bar options vs documentation for a model-loading argument." }, { "left": "issue:13191", "right": "issue:13232", "accept": false, - "reason": "Elastic-DiT support and LTX-2.3 support are separate model-support requests." + "reason": "Support requests for different models; no shared concrete bug/change." }, { "left": "issue:3799", "right": "issue:5325", "accept": false, - "reason": "A DDPMScheduler type inference bug is unrelated to packaging type information distribution." + "reason": "Type inference bug vs packaging/type-distribution issue; unrelated." }, { "left": "issue:5857", "right": "issue:5881", "accept": false, - "reason": "LCM LoRA not working well on AnimateDiff SDXL is a specific functional bug; the other is a general quality complaint about animatediff output." + "reason": "AnimateDiff/LCM-specific incompatibility vs a broad quality complaint about AnimateDiff." }, { "left": "issue:6096", "right": "issue:7629", "accept": false, - "reason": "torch.compile dynamic shape failure and PyTorch 1.13.1 OOM are different issues." + "reason": "Both mention SVD, but one is a torch.compile shape exception and the other is an OOM issue on an older PyTorch version." }, { "left": "issue:10227", "right": "issue:10514", "accept": false, - "reason": "LoRA feature-size NotImplementedError and Sana resolution-binning sample_size support are unrelated." + "reason": "Different model-specific errors: LoRA feature-size limitation vs Sana resolution-binning/sample_size support." }, { "left": "issue:181", "right": "issue:3284", "accept": false, - "reason": "Image writing failure in train_unconditional.py is unrelated to an accelerate failure in train_dreambooth_lora.py." + "reason": "Different training scripts and failures; no common underlying bug." }, { "left": "issue:2712", "right": "issue:6444", "accept": false, - "reason": "Missing norm_num_groups config and CLIPTextModel peft_config AttributeError are different bugs." + "reason": "Missing config field warning vs CLIPTextModel peft_config AttributeError are separate problems." }, { "left": "issue:10705", "right": "issue:9904", "accept": false, - "reason": "Pipeline reuse slowdown is unrelated to silent output-shape rounding." + "reason": "FluxPipeline performance degradation when reusing the pipeline vs silent image-shape rounding." }, { "left": "issue:1536", "right": "issue:5645", "accept": false, - "reason": "Custom pipeline incompatibility with new diffusers and broken LCM support are different problems." - }, - { - "left": "issue:6548", - "right": "issue:6657", - "accept": false, - "reason": "RGBA image misdetected as latents is unrelated to frames/channels being swapped in UNet3DConditionModel." - }, - { - "left": "issue:1261", - "right": "issue:1314", - "accept": true, - "reason": "Both report xformers backward-pass failures during loss/backprop; the Windows report appears to be a repro of the same underlying attention-backward bug." - }, - { - "left": "issue:10231", - "right": "issue:8975", - "accept": false, - "reason": "A tutorial ImportError on Ubuntu 20.04 is not the same as the specific LoraLoaderMixin import failure." + "reason": "Different custom-pipeline incompatibilities and different feature areas." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The analysis is conservative and grounded in the packet. The only accepted soft edge, issue:1261 \u2194 issue:1314, is plausibly the same xformers backward-pass bug based on the titles and is stated cautiously. The remaining pairwise rejections are also consistent with the issue titles and do not overstate duplication." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -63377,13 +63379,12 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10139", + "issue:10231", "issue:10382", - "issue:10489", "issue:10722", "issue:10952", - "issue:11536", "issue:11644", - "issue:12110", + "issue:1261", "issue:12610", "issue:1314", "issue:1657", @@ -63413,6 +63414,8 @@ "issue:5989", "issue:6068", "issue:6460", + "issue:6548", + "issue:6657", "issue:6809", "issue:7012", "issue:7591", @@ -63420,11 +63423,14 @@ "issue:8612", "issue:8708", "issue:8874", + "issue:8975", "issue:929", - "issue:9477", - "issue:9895" + "issue:9477" ], "soft_pairs": [ + "issue:6548|issue:6657", + "issue:1261|issue:1314", + "issue:10231|issue:8975", "issue:10382|issue:12610", "issue:5630|issue:6068", "issue:1314|issue:5678", @@ -63446,182 +63452,186 @@ "issue:1657|issue:4405", "issue:4578|issue:8708", "issue:5989|issue:9477", - "issue:10139|issue:11644", - "issue:11536|issue:9895", - "issue:10489|issue:12110" + "issue:10139|issue:11644" ], - "prepared_review_unit_hash": "1b363dae1ddeeab09b4ee1668205ee10a96317a57102e3e2d764a060d11aa1f1", + "prepared_review_unit_hash": "d5e12cd7b7fd1967f64a3d4fdfe94a9aeae1dcf7a6da5c92a78cbd8eb7ecf744", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 33565, - "estimated_input_tokens": 8392, - "estimated_eval_tokens": 17040 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34609, + "estimated_input_tokens": 8653, + "estimated_eval_tokens": 17562 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated pipeline-init bugs, training/runtime errors, docs/example requests, and adapter/loading issues. None of the soft pairs look like the same underlying bug or change.", - "confidence": 0.84, - "canonical_issue_reason": "10139 is the clearest concrete, self-contained bug report in the set and is more representative than the many unrelated or lower-signal issues, though there is no strong single canonical duplicate here.", + "summary": "All soft pairs are false positives: they share broad subsystem terms at most, but they describe different concrete bugs, different code paths, or different failure modes. No duplicate-worthy canonical issue or PR stands out from this set.", + "confidence": 0.95, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "10139 is the best global issue candidate because it describes a specific code-path defect with an actionable fix surface; the rest of the cluster is too fragmented to serve as a better canonical issue.", + "best_issue_reason": "No single issue is a good canonical target here; the issues span unrelated failures (loading, attention/backward, LoRA, schedulers, docs, and training/runtime errors) rather than one underlying bug.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6548", + "right": "issue:6657", + "accept": false, + "reason": "Different bug classes: img2img RGBA latent handling vs UNet3D frame/channel layout." + }, + { + "left": "issue:1261", + "right": "issue:1314", + "accept": false, + "reason": "Both involve xformers/backward, but one is CUDA stride assert and the other is a Windows loss-backward error; not the same failure." + }, + { + "left": "issue:10231", + "right": "issue:8975", + "accept": false, + "reason": "Both are import-related, but one is a tutorial ImportError on Ubuntu and the other is a missing LoraLoaderMixin import." + }, { "left": "issue:10382", "right": "issue:12610", "accept": false, - "reason": "Different failures: VAE forward return-type mismatch vs Quanto/group-offload device mismatch." + "reason": "Unrelated code paths: VAE forward return-type mismatch vs Quanto/group offload device mismatch." }, { "left": "issue:5630", "right": "issue:6068", "accept": false, - "reason": "Both are SDXL-related, but one is LMS noise quality and the other is DDIM sampling correctness." + "reason": "Different scheduler issues with different symptoms and algorithms (LMS noise leftover vs DDIM incorrect samples)." }, { "left": "issue:1314", "right": "issue:5678", "accept": false, - "reason": "xformers/backward error on Windows is unrelated to higher-order autograd issues in scaled_dot_product_attention." + "reason": "Both touch attention/backward, but one is xformers on Windows and the other is higher-order-derivative failure in scaled_dot_product_attention." }, { "left": "issue:170", "right": "issue:444", "accept": false, - "reason": "Unexpected eta kwarg is a scheduler API mismatch; the other is an inference-step indexing bug." + "reason": "Unrelated errors: unexpected eta keyword vs index error at certain inference-step counts." }, { "left": "issue:201", "right": "issue:929", "accept": false, - "reason": "Test request for attention class is unrelated to a duplicated square-root computation question." + "reason": "One is a test issue, the other is an attention scaling implementation question; not the same bug." }, { "left": "issue:2518", "right": "issue:4002", "accept": false, - "reason": "Missing xformers support in an example is unrelated to missing from_single_file support in a pipeline." + "reason": "Different missing features: xformers support in an example script vs missing from_single_file on a pipeline." }, { "left": "issue:4913", "right": "issue:5866", "accept": false, - "reason": "Both mention ControlNet, but one is a model release/topic and the other is an AnimateDiff+ControlNet integration request." + "reason": "Both mention ControlNet, but one is a specific SDXL openpose model issue and the other is AnimateDiff + ControlNet integration." }, { "left": "issue:5598", "right": "issue:5989", "accept": false, - "reason": "A _modify_text_encoder bug is unrelated to a safetensors.torch bug." + "reason": "Different components: a text encoder modification bug vs a safetensors.torch bug." }, { "left": "issue:5846", "right": "issue:7012", "accept": false, - "reason": "Checkpoint resume failure and v_prediction image quality are different code paths and symptoms." + "reason": "Resume-from-checkpoint training error is unrelated to v_prediction image quality issues." }, { "left": "issue:1742", "right": "issue:1909", "accept": false, - "reason": "Different problems: prompt-list speed variance vs generator-list/scheduler API mismatch." + "reason": "Performance timing for repeated prompts is unrelated to generator-argument support in schedulers." }, { "left": "issue:4042", "right": "issue:6460", "accept": false, - "reason": "A usage/composition question is not the same as endless loading when importing an inpainting model." + "reason": "Different problems: combining pipelines/features vs endless loading while loading an inpainting model." }, { "left": "issue:10952", "right": "issue:6809", "accept": false, - "reason": "Both involve LoRA loading, but the adapter-name error and PEFT-loading regression are distinct bugs." + "reason": "Both are LoRA-related, but one is adapter-name resolution in a pipeline and the other is PEFT loading regressions." }, { "left": "issue:5485", "right": "issue:8612", "accept": false, - "reason": "ControlNet batch slowness is unrelated to PixArt mixed-precision incompatibility." + "reason": "Batch ControlNet inference slowdown is unrelated to PixArt-Sigma mixed-precision incompatibility." }, { "left": "issue:10722", "right": "issue:7606", "accept": false, - "reason": "DreamBooth tensor-shape mismatch is unrelated to a deprecated LoRA processor in Kandinsky 2.2." + "reason": "Different training/runtime failures: DreamBooth tensor-size mismatch vs deprecated Kandinsky LoRA processor API." }, { "left": "issue:1925", "right": "issue:3915", "accept": false, - "reason": "Half-precision inpainting failure and torch.compile text-to-video issues are different defects." + "reason": "Unrelated contexts: SD2 inpainting half precision vs torch.compile on text-to-video pipelines." }, { "left": "issue:3975", "right": "issue:7591", "accept": false, - "reason": "A typo in an error message is unrelated to a local_files_only HeaderTooLarge loading failure." + "reason": "A typo in an error message is not the same as a HeaderTooLarge failure in from_single_file local loading." }, { "left": "issue:3643", "right": "issue:5607", "accept": false, - "reason": "INT_MAX decoding-latents failure is a different issue from a generic tensor-size mismatch." + "reason": "Different tensor-shape failures in different pipelines; no shared underlying bug." }, { "left": "issue:4954", "right": "issue:8874", "accept": false, - "reason": "Both are loading/unpickling problems, but textual inversion checkpoint handling and from_single_file CKPT loading are not the same bug." + "reason": "Both involve unpickling, but one is textual inversion loading and the other is ckpt from_single_file loading with weights-only failure." }, { "left": "issue:1657", "right": "issue:4405", "accept": false, - "reason": "A notebook request is not the same as creating a textual inversion training script." + "reason": "A question about a Kaggle notebook is unrelated to creating an SDXL textual inversion training script." }, { "left": "issue:4578", "right": "issue:8708", "accept": false, - "reason": "Doc wording about VAE usage differs from a missing shift_factor in SD3 DreamBooth examples." + "reason": "Different documentation omissions: VAE not used in docs vs missing vae.config.shift_factor in DreamBooth examples." }, { "left": "issue:5989", "right": "issue:9477", "accept": false, - "reason": "A safetensors.torch bug and a GatheredParameters callable error are unrelated." + "reason": "Different failures entirely: safetensors.torch bug vs GatheredParameters callable error." }, { "left": "issue:10139", "right": "issue:11644", "accept": false, - "reason": "Both are init-time pipeline bugs, but they affect different pipelines and different failure modes." - }, - { - "left": "issue:11536", - "right": "issue:9895", - "accept": false, - "reason": "Prompt adherence regression in FluxPipeline is unrelated to a device-mismatch error after a commit." - }, - { - "left": "issue:10489", - "right": "issue:12110", - "accept": false, - "reason": "SanaPipeline example issues and UNet1DModel GaussianFourier initialization bugs are unrelated." + "reason": "Both concern pipeline init, but they affect different pipelines with different initialization failures and no shared concrete code path." } ] }, @@ -63649,9 +63659,12 @@ "issue:10314", "issue:10446", "issue:10450", + "issue:10489", "issue:1100", + "issue:11536", "issue:11721", "issue:11973", + "issue:12110", "issue:1524", "issue:1922", "issue:2310", @@ -63673,9 +63686,7 @@ "issue:5670", "issue:579", "issue:5927", - "issue:6363", "issue:6565", - "issue:6812", "issue:7276", "issue:7303", "issue:7466", @@ -63686,15 +63697,16 @@ "issue:8874", "issue:9112", "issue:9163", - "issue:9227", - "issue:9358", "issue:9516", "issue:953", "issue:9637", "issue:9778", - "issue:9876" + "issue:9876", + "issue:9895" ], "soft_pairs": [ + "issue:11536|issue:9895", + "issue:10489|issue:12110", "issue:3624|issue:5171", "issue:10239|issue:10314", "issue:11721|issue:8874", @@ -63716,182 +63728,180 @@ "issue:2783|issue:7303", "issue:7276|issue:9778", "issue:8482|issue:8577", - "issue:9637|issue:9876", - "issue:9227|issue:9358", - "issue:6363|issue:6812" + "issue:9637|issue:9876" ], - "prepared_review_unit_hash": "fe384c84f685b6713f69120f6c4477bcf231615f7e1db123edf36478e6e843d6", + "prepared_review_unit_hash": "9bacd8f8eef7352c3c9c367aa8bbc71c9ebff4b1f1b0aae1e5461725fb49d5ba", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34070, - "estimated_input_tokens": 8518, - "estimated_eval_tokens": 17292 + "serialized_chars": 34056, + "estimated_input_tokens": 8514, + "estimated_eval_tokens": 17284 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is not a true duplicate cluster: the items span many unrelated pipelines, models, and training failures. All soft edges were rejected because they describe different bugs or feature requests, not the same underlying change.", + "summary": "No true duplicate cluster emerges here: the proposed pairs mostly share broad subsystems or similar error shapes, but they describe different bugs, features, or code paths. I would reject all soft-edge matches.", "confidence": 0.96, - "canonical_issue_reason": "Issue 9227 is the strongest single representative if one must be chosen: it is a detailed, current, high-signal report with active discussion about a concrete FLUX training/runtime failure. That said, the surrounding items are not duplicates of it.", + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 9227 is the best overall issue to anchor the cluster because it is recent, substantive, and specific, while most other items are separate bug reports across different subsystems.", + "best_issue_reason": "issue:10446 is the most concrete standalone bug report with a clear subsystem target, but it is not a duplicate of the other items.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11536", + "right": "issue:9895", + "accept": false, + "reason": "Both involve FluxPipeline, but one is prompt adherence and the other is a device-mismatch regression; different failures." + }, + { + "left": "issue:10489", + "right": "issue:12110", + "accept": false, + "reason": "SanaPipeline example/load issues are unrelated to a UNet1DModel GaussianFourier initialization bug." + }, { "left": "issue:3624", "right": "issue:5171", "accept": false, - "reason": "Both are TypeErrors, but the causes differ: one is an invalid getattr attribute name, the other is an unexpected `safety_checker` constructor argument." + "reason": "Generic getattr TypeError is not the same as SDXL rejecting the safety_checker argument." }, { "left": "issue:10239", "right": "issue:10314", "accept": false, - "reason": "Same pipeline family, but one reports lack of `enable_sequential_cpu_offload()` support and the other reports NaN generation; these are distinct failure modes." + "reason": "HunyuanVideo offload-support limitation and NaN generation are different pipeline problems." }, { "left": "issue:11721", "right": "issue:8874", "accept": false, - "reason": "Both mention load-time failures, but one is SanaPipeline crashing on load and the other is a CKPT single-file unpickling/weights-only error." + "reason": "SanaPipeline crash-on-load and single-file CKPT unpickling failure are distinct load paths." }, { "left": "issue:5670", "right": "issue:9112", "accept": false, - "reason": "Different bugs: non-square image generation in PixArtAlphaPipeline versus a tensor shape mismatch in Auraflow inference." + "reason": "Non-square image support in PixArtAlpha is unrelated to Auraflow tensor-shape mismatch." }, { "left": "issue:2310", "right": "issue:5927", "accept": false, - "reason": "These are unrelated inpaint/composition requests: LDM inpainting/LaMA cleaner versus IP-Adapter with Reference AdaIN/attn and Inpaint+Lama." + "reason": "LDM inpainting/LaMA cleaner and IP-Adapter + reference inpaint are different workflows." }, { "left": "issue:2332", "right": "issue:2830", "accept": false, - "reason": "One is an error after DreamBooth training; the other is a question about generating video with a ModelScope video model." + "reason": "Dreambooth inference error and a video-model usage question are not the same bug." }, { "left": "issue:3029", "right": "issue:3591", "accept": false, - "reason": "Custom-resolution ControlNet support and an unintuitive ControlNet loss curve are different issues." + "reason": "Custom ControlNet resolution support is not the same as an unclear loss curve during training." }, { "left": "issue:3109", "right": "issue:953", "accept": false, - "reason": "Both ask about smaller images, but they are separate requests with different target sizes and contexts." + "reason": "Both ask about smaller images, but one is a concrete generation request and the other is general guidance." }, { "left": "issue:3661", "right": "issue:5344", "accept": false, - "reason": "`from_ckpt` failing on SD 2.x weights is not the same as missing the `omegaconf` dependency." + "reason": "from_ckpt SD2 weight conversion failure is unrelated to a missing omegaconf dependency." }, { "left": "issue:1100", "right": "issue:5109", "accept": false, - "reason": "`CLIPTextTransformer` missing `.device` and LoRA loading failing in SDXL ControlNet img2img are different code paths and errors." + "reason": "CLIPTextTransformer.device and load_lora_weights exception in SDXL ControlNet img2img are different code paths." }, { "left": "issue:1524", "right": "issue:2466", "accept": false, - "reason": "Poor textual inversion training quality is unrelated to checkpoint limit behavior not limiting saved checkpoints." + "reason": "Poor textual inversion training quality is unrelated to checkpoint retention behavior." }, { "left": "issue:2712", "right": "issue:579", "accept": false, - "reason": "One is a missing `norm_num_groups` config field, the other is a missing `config.json` file; these are separate config-loading problems." + "reason": "Missing norm_num_groups in config and missing config.json are different configuration errors." }, { "left": "issue:4125", "right": "issue:6565", "accept": false, - "reason": "A LoRA loss-curve question is not the same as DPO diffusion training instability and noisy samples." + "reason": "LoRA loss-curve expectations and DPO diffusion instability are not the same underlying bug." }, { "left": "issue:7538", "right": "issue:7735", "accept": false, - "reason": "Custom Fill50k-style training issues differ from a missing `caption_column` named `text`." + "reason": "Custom Fill50k training issue and missing caption_column are different dataset/training failures." }, { "left": "issue:10446", "right": "issue:7466", "accept": false, - "reason": "Multi-GPU/model-sharding incompatibility with FluxControlNet and torch.compile failure for multi-ControlNet SDXL inpaint are different problems." + "reason": "FluxControlNetPipeline model-sharding incompatibility is distinct from multi-ControlNet torch.compile failure." }, { "left": "issue:10450", "right": "issue:9516", "accept": false, - "reason": "A structural mismatch with `FluxSingleTransformerBlock` is not the same as `joint_attention_kwargs` not being passed through FLUX transformers." + "reason": "Transformer block implementation differences and missing joint_attention_kwargs forwarding are separate issues." }, { "left": "issue:11973", "right": "issue:9163", "accept": false, - "reason": "The DreamBooth LoRA script missing `--instance_prompt` is an argparse usage issue, while the other report is about tensor bugs in the same script." + "reason": "Missing required instance_prompt in the script is not the same as tensor bugs in train_dreambooth_lora_flux.py." }, { "left": "issue:1922", "right": "issue:490", "accept": false, - "reason": "An undefined notebook helper (`init_git_repo`) is unrelated to a general memory usage complaint." + "reason": "Undefined init_git_repo in a notebook is unrelated to memory usage differences between frameworks." }, { "left": "issue:2783", "right": "issue:7303", "accept": false, - "reason": "Creating an img2img/inpainting ControlNet pipeline and using `zero_snr` when constructing a scheduler are unrelated requests." + "reason": "Request to create an img2img ControlNet pipeline and scheduler zero_snr support are different topics." }, { "left": "issue:7276", "right": "issue:9778", "accept": false, - "reason": "Downloaded pretrained-model network errors are unrelated to a tutorial run bug." + "reason": "Network errors while using a downloaded pretrained model are not the same as a tutorial bug." }, { "left": "issue:8482", "right": "issue:8577", "accept": false, - "reason": "General SD3 model support and differential diffusion with SD3 are distinct feature requests." + "reason": "General SD3 support and Differential Diffusion with SD3 are distinct feature requests." }, { "left": "issue:9637", "right": "issue:9876", "accept": false, - "reason": "An xFormers attention-mask/bfloat16 sequence-length bug is different from VRAM not being released after LoRA training." - }, - { - "left": "issue:9227", - "right": "issue:9358", - "accept": false, - "reason": "Both concern `train_dreambooth_lora_flux.py`, but one is a broad runtime problem with FLUX.1-dev and the other is a specific redundant text-encoder reinitialization issue." - }, - { - "left": "issue:6363", - "right": "issue:6812", - "accept": false, - "reason": "FP16 gradient unscale errors in LoRA training and Q/K/V dtype mismatch errors are different training failures." + "reason": "xFormers attention-mask bfloat bug and LoRA VRAM release concern are unrelated." } ] }, @@ -63916,12 +63926,9 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10896", - "issue:1099", "issue:11033", - "issue:11486", "issue:11824", "issue:12501", - "issue:1260", "issue:1290", "issue:13001", "issue:13298", @@ -63946,24 +63953,29 @@ "issue:6055", "issue:6073", "issue:6141", + "issue:6363", "issue:6446", "issue:6520", "issue:6530", + "issue:6812", "issue:7441", "issue:746", "issue:7460", "issue:7675", "issue:7751", "issue:8551", - "issue:8828", "issue:8939", + "issue:9227", "issue:9309", "issue:9320", + "issue:9358", "issue:9397", "issue:9630", "issue:9825" ], "soft_pairs": [ + "issue:9227|issue:9358", + "issue:6363|issue:6812", "issue:1363|issue:746", "issue:1546|issue:6055", "issue:5605|issue:7675", @@ -63985,182 +63997,180 @@ "issue:1604|issue:3551", "issue:6446|issue:9320", "issue:8551|issue:8939", - "issue:1857|issue:259", - "issue:1099|issue:1260", - "issue:11486|issue:8828" + "issue:1857|issue:259" ], - "prepared_review_unit_hash": "06fc67c7c47daed60243db07bab16fbdd3cc1058c8b4e7cb87dda77200249624", + "prepared_review_unit_hash": "2bd8658dc2620f321f9e9e58c702f68210a15f9034cc41c52d14a5a90bfe1d6b", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33352, - "estimated_input_tokens": 8338, - "estimated_eval_tokens": 16932 + "serialized_chars": 33495, + "estimated_input_tokens": 8374, + "estimated_eval_tokens": 17004 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a duplicate set; it\u2019s a broad grab bag of unrelated Diffusers issues spanning model loading, training scripts, scheduler bugs, feature requests, and platform-specific failures. The soft-similarity pairs mostly share generic words like \"error\" or \"support\" but do not describe the same underlying bug/change.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "This cluster is mostly a set of unrelated diffusers issues, not a true duplicate set. The soft-edge candidates are false positives driven by generic wording like \u201cerror\u201d, \u201cbug\u201d, and \u201csupport\u201d rather than the same underlying code-path bug or feature request.", + "confidence": 0.96, + "canonical_issue_reason": "No issue is a clean canonical duplicate target because the cluster is heterogeneous. If one placeholder must be chosen, issue 13298 is the broadest open runtime-bug report with some inbound references, while most others are narrower feature requests or unrelated failures.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 13298 is the best single representative only as a placeholder: it is open, broadly framed, and actionable as a runtime-generation bug. That said, the set overall does not share one underlying defect.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:9227", + "right": "issue:9358", + "accept": false, + "reason": "Both mention train_dreambooth_lora_flux, but one is a model/run failure and the other is redundant text-encoder reinitialization; different bugs, not one duplicate." + }, + { + "left": "issue:6363", + "right": "issue:6812", + "accept": false, + "reason": "Both are LoRA training errors, but one is FP16 unscale failure and the other is Q/K/V dtype mismatch; distinct failure modes." + }, { "left": "issue:1363", "right": "issue:746", "accept": false, - "reason": "Both mention ONNX-related workflows, but one is a long-prompt scheduler error in a community pipeline and the other is an img2img ONNX revision loading problem; different code paths and failures." + "reason": "Different backends and symptoms: ONNX long-prompt weight error with LMS scheduler vs img2img ONNX revision loading problem." }, { "left": "issue:1546", "right": "issue:6055", "accept": false, - "reason": "Batched truncation warning causing an error is unrelated to a generic accelerate error; no shared bug." + "reason": "Truncation warning on batched inputs is unrelated to a generic accelerate error." }, { "left": "issue:5605", "right": "issue:7675", "accept": false, - "reason": "SDXL DreamBooth per-image caption support and MLX quantized linear support are separate feature requests for different subsystems." + "reason": "One is SDXL DreamBooth caption support, the other is MLX quantized linear support; both are feature requests but for different components." }, { "left": "issue:11033", "right": "issue:13001", "accept": false, - "reason": "SD1.5 Unet from_single_file loading and Flux2.Klein GGUF shape mismatch are different model formats/loaders and not the same bug." + "reason": "Both are model-loading issues, but SD1.5 from_single_file loading and Flux2.Klein GGUF shape mismatch are different model families and code paths." }, { "left": "issue:6520", "right": "issue:9630", "accept": false, - "reason": "Blank validation results in adapter training is unrelated to a flux-controlnet single-GPU/multi-GPU num_single_layers crash." + "reason": "Different scripts and symptoms: blank validation output in adapter training vs num_single_layers handling in flux-controlnet on single GPU." }, { "left": "issue:12501", "right": "issue:13298", "accept": false, - "reason": "A VAE decoder graph break and a CUDA generator-device bug are distinct runtime/device issues." + "reason": "Both mention runtime/device behavior, but graph breaks in VAE decoder and CUDA generator device breakage are not the same bug." }, { "left": "issue:4804", "right": "issue:4830", "accept": false, - "reason": "Dictionary mutation in LoRA loading and MultiControlNet tensor-size mismatch are separate failures in different components." + "reason": "LoraLoaderMixin dict-iteration error and MultiControlNet tensor-size error affect different subsystems." }, { "left": "issue:3279", "right": "issue:7751", "accept": false, - "reason": "Checkpoint resume errors in DreamBooth LoRA and Animatediff conversion errors are unrelated training/conversion problems." + "reason": "Checkpoint resume errors in DreamBooth LoRA are unrelated to animatediff motion-module conversion failures." }, { "left": "issue:5402", "right": "issue:6073", "accept": false, - "reason": "Missing compute_snr import and a beginner DreamBooth setup error do not describe the same issue." + "reason": "Missing compute_snr import is a package/API issue; the SDXL DreamBooth README error is a separate user-facing training problem." }, { "left": "issue:10896", "right": "issue:9825", "accept": false, - "reason": "SkyReels-A1 support and FLUX IPAdapter support are different feature additions." + "reason": "Both are support requests, but SkyReels-A1 portrait animation and FLUX IPAdapter support are different feature asks." }, { "left": "issue:11824", "right": "issue:9309", "accept": false, - "reason": "Multi-image FluxKontextPipeline errors and FluxControlNetPipeline snel bug are different pipeline-specific bugs." + "reason": "Different FLUX pipelines and distinct failures: multi-image Kontext input vs FluxControlNet schnell bug." }, { "left": "issue:1290", "right": "issue:3086", "accept": false, - "reason": "DreamArtist community pipeline work and an audio-to-image request are unrelated." + "reason": "DreamArtist community pipeline and audio-to-image are unrelated topics." }, { "left": "issue:3089", "right": "issue:7460", "accept": false, - "reason": "A DDPM scheduler bug and an AdaLayerNorm bug are different code-path defects." + "reason": "DDPM scheduler bug and AdaLayerNorm bug are different internal code-path issues." }, { "left": "issue:3427", "right": "issue:6530", "accept": false, - "reason": "makeavid-sd-jax and SDXL ControlNet K-diffusion are unrelated model/pipeline topics." + "reason": "Different projects/features: makeavid-sd-jax vs SDXL K-diffusion ControlNet." }, { "left": "issue:3612", "right": "issue:541", "accept": false, - "reason": "ControlNet creation/preprocessor support and SageMaker deployment questions are not the same bug or change." + "reason": "ControlNet preprocessor configuration and SageMaker usage are unrelated." }, { "left": "issue:5570", "right": "issue:6520", "accept": false, - "reason": "Missing compute_snr in a package release and blank validation output during adapter training are different problems." + "reason": "Missing compute_snr packaging issue is not the same as blank log_validation output during training." }, { "left": "issue:5712", "right": "issue:7441", "accept": false, - "reason": "Windows/Python 3.12 install failure and pytest-on-python3.12 failure are platform-related but distinct issues." + "reason": "Both involve Python 3.12, but one is pip install on Windows and the other is pytest execution; different problems." }, { "left": "issue:6141", "right": "issue:9397", "accept": false, - "reason": "ControlNet control_mode/resize_mode support and Hyper-LoRA scheduler support are separate feature requests." + "reason": "Control mode/resize mode support for ControlNet and SGMUniform scheduler support for Hyper LoRAs are separate feature requests." }, { "left": "issue:1604", "right": "issue:3551", "accept": false, - "reason": "A diffusers version compatibility break in lpw_stable_diffusion.py and LoRA disabling xFormers are different regressions." + "reason": "Broken lpw stable diffusion on diffusers 0.9.0 is unrelated to LoRA disabling xFormers." }, { "left": "issue:6446", "right": "issue:9320", "accept": false, - "reason": "Colab minimal-loading failure for SD 2-1 and custom diffusion save/load issues are unrelated model-loading/training bugs." + "reason": "Minimal-loading Colab model load failure and custom-diffusion save/load problems are different training/loading paths." }, { "left": "issue:8551", "right": "issue:8939", "accept": false, - "reason": "An SDXL inpaint device-map latent placement bug and a tokenizer-loading OSError for SD3 are different failures." + "reason": "VAE device_map balanced latent placement bug and tokenizer loading failure for SD3 are unrelated components." }, { "left": "issue:1857", "right": "issue:259", "accept": false, - "reason": "Poor test coverage for UnCLIP is not the same as a progress-bar fix." - }, - { - "left": "issue:1099", - "right": "issue:1260", - "accept": false, - "reason": "DreamBooth class-name handling and a missing DPMSolverMultistepScheduler import are unrelated issues." - }, - { - "left": "issue:11486", - "right": "issue:8828", - "accept": false, - "reason": "xFormers memory-efficient attention in a training script and an SD3 16GB training error are different training problems." + "reason": "Poor UnCLIP testing and progress bar fix are not the same underlying bug or change." } ] }, @@ -64185,29 +64195,28 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10045", + "issue:10231", "issue:10569", "issue:10800", "issue:10952", + "issue:1099", "issue:11199", - "issue:1161", + "issue:11486", "issue:11700", - "issue:1181", "issue:1207", + "issue:1260", "issue:12624", "issue:2132", "issue:2646", "issue:3010", - "issue:3338", "issue:3401", "issue:3412", "issue:4229", "issue:4376", "issue:4438", "issue:444", - "issue:4445", "issue:4534", "issue:4605", - "issue:4615", "issue:4907", "issue:4933", "issue:5118", @@ -64220,11 +64229,13 @@ "issue:6530", "issue:7389", "issue:761", + "issue:7682", "issue:7722", "issue:782", "issue:7856", "issue:8234", "issue:8801", + "issue:8828", "issue:9027", "issue:9403", "issue:9495", @@ -64233,11 +64244,14 @@ "issue:9825" ], "soft_pairs": [ + "issue:1099|issue:1260", + "issue:11486|issue:8828", "issue:10569|issue:10800", "issue:2646|issue:444", "issue:4376|issue:4438", "issue:3401|issue:5743", "issue:9602|issue:9660", + "issue:10231|issue:7682", "issue:10045|issue:11199", "issue:1207|issue:2132", "issue:3010|issue:3412", @@ -64254,189 +64268,186 @@ "issue:4933|issue:782", "issue:6442|issue:6445", "issue:10952|issue:7856", - "issue:4605|issue:7389", - "issue:1161|issue:4445", - "issue:1181|issue:4445", - "issue:3338|issue:4615" + "issue:4605|issue:7389" ], - "prepared_review_unit_hash": "79c26fa95534836a286d0c953794aa88806acc514bcd438cdf313e632249b85b", + "prepared_review_unit_hash": "b9ff7671ca36617e6b649aa9c2930a84c78fb35d914a3ca5ba6812c099ab566d", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33462, - "estimated_input_tokens": 8366, - "estimated_eval_tokens": 16988 + "serialized_chars": 34031, + "estimated_input_tokens": 8508, + "estimated_eval_tokens": 17272 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft pairs are mostly superficial text matches, not true duplicates: they cover different pipelines, different error modes, and different feature requests. No pull requests are present.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue: the items span unrelated bugs/features across imports, adapters, training, and pipelines.", + "summary": "This is a noisy cluster of mostly unrelated diffusers issues linked only by broad title similarity; I don\u2019t see any true duplicate pairs or shared fixes among the soft edges.", + "confidence": 0.18, + "canonical_issue_reason": "issue:10045 is the broadest, most umbrella-style problem report in the set, so it works best as a loose cluster anchor.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9602 is the cleanest, most concrete bug report with a narrow adapter-name error, so it is the best practical triage anchor among these issues.", + "best_issue_reason": "issue:10045 is the best representative issue because its generic import/pipeline wording is the closest match to the cluster\u2019s superficial theme.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1099", + "right": "issue:1260", + "accept": false, + "reason": "Both are import/runtime failures, but they are different APIs and different failure modes (DreamBooth class-name issue vs scheduler import error)." + }, + { + "left": "issue:11486", + "right": "issue:8828", + "accept": false, + "reason": "Different request and bug: xformers training-script support vs SD3 training memory error." + }, { "left": "issue:10569", "right": "issue:10800", "accept": false, - "reason": "Different problems: one is general CPU memory usage for HunyuanVideo, the other is a specific BnB int4 + sequential offload type error." + "reason": "Both involve HunyuanVideo CPU usage, but one is high memory consumption and the other is a BnB int4 + offload quantization type error." }, { "left": "issue:2646", "right": "issue:444", "accept": false, - "reason": "Both mention IndexError, but the contexts differ: tiled VAE vs. num_inference_steps scheduling behavior." + "reason": "Unrelated failures: tiled VAE IndexError vs inference-step indexing bug." }, { "left": "issue:4376", "right": "issue:4438", "accept": false, - "reason": "Both are SDXL pipeline errors, but the titles indicate different call paths and likely different failures." + "reason": "Same pipeline family, but different crash contexts and no shared concrete code path identified." }, { "left": "issue:3401", "right": "issue:5743", "accept": false, - "reason": "One is about class_labels requirements for class embeddings; the other is a Latent Consistency Distillation training stall." + "reason": "Class-label validation error is unrelated to LCD training stuck at 0%." }, { "left": "issue:9602", "right": "issue:9660", "accept": false, - "reason": "Adapter-name mismatch versus from_pretrained failing to set modules to eval; distinct bugs and code paths." + "reason": "Adapter-name lookup failure is different from from_pretrained leaving modules in train mode." + }, + { + "left": "issue:10231", + "right": "issue:7682", + "accept": false, + "reason": "ImportError in tutorial on Ubuntu is unrelated to multi-GPU memory/throughput behavior." }, { "left": "issue:10045", "right": "issue:11199", "accept": false, - "reason": "Importing pipelines is too broad to treat as a duplicate of a specific onnx_utils import failure." + "reason": "Both mention imports, but one is generic pipeline importing and the other is a specific onnx_utils import failure." }, { "left": "issue:1207", "right": "issue:2132", "accept": false, - "reason": "CUDA runtime symbol resolution error versus Accelerator.trackers attribute error; unrelated failures." + "reason": "CUDA runtime symbol error and Accelerator trackers attribute error are different environment/library issues." }, { "left": "issue:3010", "right": "issue:3412", "accept": false, - "reason": "One is distributed inference docs; the other is a separate stable diffusion docs update request." + "reason": "Both are docs-related, but they are not the same document or fix." }, { "left": "issue:5118", "right": "issue:6530", "accept": false, - "reason": "Different subject matter: ControlNet reference_only usage versus SDXL K-diffusion with ControlNet." + "reason": "ControlNet reference_only usage question is not the same as SDXL K-diffusion + ControlNet error." }, { "left": "issue:5371", "right": "issue:9403", "accept": false, - "reason": "Blip inpainting pipe support request versus Flux IPAdapter support; different features." + "reason": "Different feature requests: BLIP inpainting support vs Xlabs IPAdapter support for Flux." }, { "left": "issue:6093", "right": "issue:9027", "accept": false, - "reason": "Pruna push_to_hub behavior in one training script versus generic argument-name consistency; not the same bug." + "reason": "Push_to_hub repo-creation bug is unrelated to argument-name consistency discussion." }, { "left": "issue:7722", "right": "issue:8234", "accept": false, - "reason": "Animatediff + IPAdapter integration error versus ip_adapter_scale dict multiplication type error." + "reason": "Both touch IPAdapter, but one is Animatediff/ControlNet integration and the other is ip_adapter_scale type handling." }, { "left": "issue:8801", "right": "issue:9825", "accept": false, - "reason": "Kolors support request versus FLUX IPAdapters support request; separate model families and feature asks." + "reason": "Different model families and feature requests: Kolors support vs Flux IPAdapter support." }, { "left": "issue:8234", "right": "issue:9495", "accept": false, - "reason": "A scaling type error with IPAdapter Plus versus SDXL PAG with IPAdapter not working; different failures." + "reason": "Both involve IPAdapter, but one is a dict scaling TypeError and the other is SDXL PAG compatibility." }, { "left": "issue:4907", "right": "issue:5799", "accept": false, - "reason": "Inpainting warning with SDXL versus adding a strength keyword to inpainting; related domain, but not the same issue." + "reason": "Inpainting warning with SDXL is not the same as adding a strength argument to inpainting." }, { "left": "issue:4229", "right": "issue:4534", "accept": false, - "reason": "Multi-GPU accelerated training error versus SDXL dreambooth validation crash on mps; distinct environments and failure modes." + "reason": "Different training contexts and errors: multi-GPU Accelerate issue vs MPS validation crash." }, { "left": "issue:2646", "right": "issue:761", "accept": false, - "reason": "Both involve training/runtime errors, but tiled VAE IndexError and 8bit Adam DreamBooth failure are separate bugs." + "reason": "Tiled VAE IndexError is unrelated to DreamBooth + 8-bit Adam failure." }, { "left": "issue:11700", "right": "issue:12624", "accept": false, - "reason": "Pruna integration for loading pipelines versus requesting a strength parameter on wanimage-to-video; unrelated feature requests." + "reason": "Pruna integration request and WaniImageToVideoPipeline strength request are unrelated feature asks." }, { "left": "issue:4933", "right": "issue:782", "accept": false, - "reason": "Dreambooth output artifact issue versus model download broken after use_auth_token removal; different subsystems." + "reason": "DreamBooth output artifact problem is unrelated to model download auth-token removal." }, { "left": "issue:6442", "right": "issue:6445", "accept": false, - "reason": "Unscale FP16 gradients error versus caption_column validation error; unrelated training script issues." + "reason": "Both are training-time ValueErrors, but from different validation paths and different user inputs." }, { "left": "issue:10952", "right": "issue:7856", "accept": false, - "reason": "Adapter-name lookup failure versus tokenizer token-collision warning; not the same bug." + "reason": "Adapter-name mismatch during LoRA loading is unrelated to tokenizer multi-vector token collision." }, { "left": "issue:4605", "right": "issue:7389", "accept": false, - "reason": "LongPromptWeighting inpaint model error versus SDXL from_single_file loading error; different pipeline operations." - }, - { - "left": "issue:1161", - "right": "issue:4445", - "accept": false, - "reason": "Flax-to-PyTorch conversion request versus general LoRA finetuning how-to question; unrelated." - }, - { - "left": "issue:1181", - "right": "issue:4445", - "accept": false, - "reason": "Training from scratch request versus LoRA finetuning question; different feature/support topics." - }, - { - "left": "issue:3338", - "right": "issue:4615", - "accept": false, - "reason": "Import slowdown versus missing StableDiffusionControlnetPipeline import; both import-related but not the same underlying issue." + "reason": "Different pipeline errors: long-prompt-weighting inpaint issue vs from_single_file loading error." } ] }, @@ -64469,14 +64480,14 @@ "issue:11342", "issue:11437", "issue:11584", + "issue:1161", "issue:11790", + "issue:1181", "issue:12096", "issue:13104", "issue:13249", "issue:1367", "issue:1388", - "issue:1522", - "issue:1817", "issue:1934", "issue:1964", "issue:2294", @@ -64484,11 +64495,12 @@ "issue:2507", "issue:2774", "issue:2995", - "issue:3085", "issue:3209", - "issue:3249", + "issue:3338", "issue:3391", "issue:4071", + "issue:4445", + "issue:4615", "issue:4675", "issue:4789", "issue:4912", @@ -64509,6 +64521,9 @@ "issue:993" ], "soft_pairs": [ + "issue:1161|issue:4445", + "issue:1181|issue:4445", + "issue:3338|issue:4615", "issue:10016|issue:7770", "issue:2507|issue:557", "issue:2995|issue:7572", @@ -64530,182 +64545,186 @@ "issue:11437|issue:9546", "issue:10336|issue:11342", "issue:10518|issue:8534", - "issue:10952|issue:13104", - "issue:1522|issue:3249", - "issue:1817|issue:3085" + "issue:10952|issue:13104" ], - "prepared_review_unit_hash": "6eb7f5dfb14cf0afad99b84357cf81c055942613b63447b2fa1ae3c786a45ff8", + "prepared_review_unit_hash": "b7729bb3d1e76de661d7a9e2a74a9d6ee7a20e645a3f0d63749409fd5f0d6572", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 33081, - "estimated_input_tokens": 8271, - "estimated_eval_tokens": 16798 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33811, + "estimated_input_tokens": 8453, + "estimated_eval_tokens": 17162 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is not a single duplicate cluster; the soft links are mostly superficial title similarity across unrelated bugs (loading, LoRA, offloading, training, memory, and pipeline issues). No pair is close enough to merge as the same underlying change or defect.", - "confidence": 0.93, - "canonical_issue_reason": null, + "summary": "This cluster is noisy and spans several unrelated Diffusers bug families: LoRA/model loading, tensor shape mismatches, training/checkpointing, offloading/compile issues, and assorted pipeline-specific regressions. The soft pairs are mostly shared-subsystem similarity, not the same underlying bug, so they should stay separate.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 11286 is the broadest, most generic LoRA-loading report in the set, so it is the best umbrella issue if one must anchor this cluster; most other items are not true duplicates of it.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "11286 is the most representative user-facing report among the LoRA-loading subset, but it is still only a loose umbrella and not a precise duplicate match for the rest of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1161", + "right": "issue:4445", + "accept": false, + "reason": "Flax-to-PyTorch conversion is unrelated to a LoRA finetuning question." + }, + { + "left": "issue:1181", + "right": "issue:4445", + "accept": false, + "reason": "Training from scratch vs. how to finetune LoRA are different requests." + }, + { + "left": "issue:3338", + "right": "issue:4615", + "accept": false, + "reason": "Import performance regression is not the same as a missing pipeline import error." + }, { "left": "issue:10016", "right": "issue:7770", "accept": false, - "reason": "Both mention loading/shapes, but 10016 is an empty-state-dict single-file SD3.5M positional embedding mismatch, while 7770 is a model+LoRA tensor shape mismatch in Python; different code paths and failure modes." + "reason": "Both mention shape issues, but SD3.5 single-file pos_embed mismatch and model+LoRA tensor mismatch are different code paths." }, { "left": "issue:2507", "right": "issue:557", "accept": false, - "reason": "2507 is a safetensors loading performance complaint, while 557 is about DDIM output quality; same library area, not the same bug." + "reason": "Safetensors load speed and DDIM quality are unrelated problems." }, { "left": "issue:2995", "right": "issue:7572", "accept": false, - "reason": "2995 is a matrix multiplication shape mismatch in a model path, whereas 7572 is a dtype mismatch between input and bias; different root causes." + "reason": "Matrix shape mismatch and dtype mismatch are different failure modes." }, { "left": "issue:1388", "right": "issue:4675", "accept": false, - "reason": "1388 is about adapting a conversion script for SD 2.0, while 4675 is ControlNet Reference Only for SDXL; unrelated features." + "reason": "SD 2.0 conversion scripting and SDXL ControlNet reference-only support are different changes." }, { "left": "issue:4912", "right": "issue:9003", "accept": false, - "reason": "4912 is SDXL checkpoint conversion, while 9003 requests adjustable safety checker strength; no shared defect." + "reason": "Checkpoint conversion to diffusers is unrelated to safety_checker strength control." }, { "left": "issue:11584", "right": "issue:12096", "accept": false, - "reason": "Both involve group offloading, but 11584 is a torch.compile incompatibility on hunyuanvideo_frampack and 12096 is WanVACEPipeline failing with apply_group_offloading; different pipelines and failure specifics." + "reason": "Both involve offloading/compile behavior, but they target different models, pipelines, and failure modes." }, { "left": "issue:1934", "right": "issue:557", "accept": false, - "reason": "1934 concerns CPU offload memory usage, while 557 concerns DDIM generation quality; not the same issue." + "reason": "CPU offload memory usage is unrelated to DDIM output quality." }, { "left": "issue:11286", "right": "issue:11790", "accept": false, - "reason": "Both are LoRA loading problems, but the reports are not specific enough to show the same failing code path or identical adapter handling bug." + "reason": "Both are LoRA loading complaints, but they are different scripts/models and report different concrete errors." }, { "left": "issue:11127", "right": "issue:13249", "accept": false, - "reason": "11127 is a single-file Flux model compatibility issue, while 13249 says LoRA stopped working for Z-Image; different models and loading flows." + "reason": "CivitAI Flux single-file loading and Z-Image LoRA breakage are different model families and bugs." }, { "left": "issue:1964", "right": "issue:4071", "accept": false, - "reason": "1964 is a PyTorch 2.0 support planning issue, while 4071 requests Inpaint-only+lama support; unrelated." + "reason": "PyTorch support planning is unrelated to inpaint-only+lama feature support." }, { "left": "issue:3391", "right": "issue:4789", "accept": false, - "reason": "3391 is overexposed output in DeepFloyd super-resolution, while 4789 is a download failure for DeepFloyd IF; one is image quality, the other acquisition." + "reason": "Overexposed super-resolution output and download failure for DeepFloyd IF are different issues." }, { "left": "issue:6705", "right": "issue:9393", "accept": false, - "reason": "6705 is an Accelerate/FSDP checkpoint save error during T2I training, while 9393 is a Deepspeed Flux LoRA model-save problem; both are save-related but not the same bug." + "reason": "Both involve saving during training, but accelerate+FSDP ckpt saving and deepspeed Flux LoRA model saving are not the same bug." }, { "left": "issue:7706", "right": "issue:7799", "accept": false, - "reason": "7706 is CUDA OOM when loading a custom pipeline, while 7799 is a matmul shape mismatch in StableDiffusionInstructPix2Pix with IP-Adapter; different runtime failures." + "reason": "Custom pipeline OOM and IP-Adapter matmul shape mismatch are unrelated." }, { "left": "issue:2774", "right": "issue:4964", "accept": false, - "reason": "2774 is a controlnet safetensors-to-diffusers conversion error, while 4964 is a general SDXL model conversion request; not the same conversion problem." + "reason": "ControlNet safetensors conversion and SDXL model conversion are different conversion paths and errors." }, { "left": "issue:5325", "right": "issue:993", "accept": false, - "reason": "5325 is about packaging type information distribution, while 993 is a fp16 dtype mismatch in stable diffusion; unrelated." + "reason": "Packaging type metadata distribution and fp16 dtype mismatch are unrelated." }, { "left": "issue:1367", "right": "issue:2312", "accept": false, - "reason": "1367 is about saving a scheduler config with custom trained_betas, while 2312 is an EMA save hook error in text-to-image finetuning; both touch saving, but different objects and failures." + "reason": "Scheduler config saving with trained_betas and EMA hook save failure are different code paths." }, { "left": "issue:2294", "right": "issue:6823", "accept": false, - "reason": "2294 is an ImportError for StableDiffusionVideoInpaintPipeline, while 6823 is a generic controlnet usage problem; not the same missing-symbol issue." + "reason": "Missing video inpaint import and inability to use ControlNet are unrelated." }, { "left": "issue:3209", "right": "issue:7450", "accept": false, - "reason": "3209 reports NaNs in AutoencoderKL encoder outputs for large images, while 7450 reports AutoencoderKL multi-thread issues; same component, different bug." + "reason": "Large-image NaNs in AutoencoderKL and multithreading issues are different bugs." }, { "left": "issue:11437", "right": "issue:9546", "accept": false, - "reason": "11437 says training produces no outputs/validation images, while 9546 is Flux ControlNet validation running out of memory; one is missing artifacts, the other OOM." + "reason": "Missing training outputs and validation-step OOM are different symptoms and fixes." }, { "left": "issue:10336", "right": "issue:11342", "accept": false, - "reason": "10336 is Sana activations exploding/clamping, while 11342 is Sana-sprint failing on MPS; different symptoms and likely different fixes." + "reason": "Both mention Sana, but activation/clamping explosions on one side and MPS failure on the other are different problems." }, { "left": "issue:10518", "right": "issue:8534", "accept": false, - "reason": "10518 flags a possible mistake in an SD3 LoRA training example script, while 8534 is wrong logit-normal weighting in SD3 dreambooth training; both are SD3 training issues but not the same defect." + "reason": "A wrong example script and an SD3 logit-normal weighting bug are not the same underlying issue." }, { "left": "issue:10952", "right": "issue:13104", "accept": false, - "reason": "10952 is an adapter-name mismatch when loading LoRA in SDXLLongPromptWeightingPipeline, while 13104 is a generic 'Logger is not defined' error; unrelated." - }, - { - "left": "issue:1522", - "right": "issue:3249", - "accept": false, - "reason": "1522 is a UNet forward signature mismatch for encoder_hidden_states, while 3249 is AttnUpBlock2D not accepting upsample_size; both are API signature issues but different arguments and code paths." - }, - { - "left": "issue:1817", - "right": "issue:3085", - "accept": false, - "reason": "1817 is get_scheduler() rejecting num_cycles in Dreambooth, while 3085 is a cache-folder argument removal in config handling; distinct API breakages." + "reason": "LoRA adapter-name mismatch and an undefined logger error are unrelated." } ] }, @@ -64724,8 +64743,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -64733,19 +64752,21 @@ "issue:1037", "issue:10514", "issue:10675", - "issue:10772", "issue:10833", "issue:10994", "issue:1101", "issue:11342", "issue:13034", + "issue:1522", + "issue:1817", "issue:1961", "issue:2098", "issue:2268", "issue:2299", - "issue:2390", "issue:2645", "issue:2747", + "issue:3085", + "issue:3249", "issue:3391", "issue:3487", "issue:3631", @@ -64763,7 +64784,6 @@ "issue:5689", "issue:5703", "issue:6141", - "issue:6243", "issue:6517", "issue:6818", "issue:725", @@ -64772,13 +64792,14 @@ "issue:7836", "issue:785", "issue:8523", - "issue:8834", "issue:8898", "issue:9397", "issue:9799", "issue:9822" ], "soft_pairs": [ + "issue:1522|issue:3249", + "issue:1817|issue:3085", "issue:1961|issue:2268", "issue:2299|issue:5592", "issue:2645|issue:4134", @@ -64800,182 +64821,180 @@ "issue:5335|issue:785", "issue:5689|issue:5703", "issue:10833|issue:13034", - "issue:4305|issue:5687", - "issue:10772|issue:8834", - "issue:2390|issue:6243" + "issue:4305|issue:5687" ], - "prepared_review_unit_hash": "f1814d658ed93bbe6133d40e22efbb7b2cfc4e6a0e5b9b1a9851aa8979e0d462", + "prepared_review_unit_hash": "57a3ba0cb7f56e91dac2d4306132fed908e1d41328bea999e1ecf171e25ddd9f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33646, - "estimated_input_tokens": 8412, - "estimated_eval_tokens": 17080 + "serialized_chars": 33796, + "estimated_input_tokens": 8449, + "estimated_eval_tokens": 17154 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous set of Diffusers issues spanning unrelated models, schedulers, docs, and runtime errors. The soft pairs mostly share only broad subsystem keywords, not the same concrete bug or feature change. No PRs are present.", - "confidence": 0.9, - "canonical_issue_reason": "Issue 10045 is the broadest generic pipeline-import report and is the closest thing to an umbrella title, but the cluster is too mixed to support a strict duplicate canonical.", + "summary": "No strong duplicate cluster here: the items span unrelated bug reports, feature requests, docs, and scheduler/model-specific issues. All soft-edge pairs should be rejected because they do not describe the same underlying defect or change.", + "confidence": 0.89, + "canonical_issue_reason": "issue:10045 is the broadest umbrella title in the set ('Problem with importing pipelines'), but this is only a weak anchor; the overall set is too heterogeneous to justify a true canonical duplicate issue.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 10045 is the best representative only because it is generic and broad; it does not actually cover the unrelated feature requests and bug reports in the rest of the set.", + "best_issue_reason": "issue:10045 is the most generic and potentially reusable target, but it is still not a good global representative because the rest of the items are largely unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1522", + "right": "issue:3249", + "accept": false, + "reason": "Both are TypeError-style API mismatch reports, but for different arguments and different blocks; not the same bug." + }, + { + "left": "issue:1817", + "right": "issue:3085", + "accept": false, + "reason": "Both mention removed/unsupported scheduler-related args, but one is Dreambooth num_cycles and the other is cache folder; unrelated failures." + }, { "left": "issue:1961", "right": "issue:2268", "accept": false, - "reason": "Both mention ColossalAI, but they are generic, separate reports with no evidence of the same failure or code path." + "reason": "Both are vague ColossalAI-related errors, but the reports do not indicate the same code path or fix." }, { "left": "issue:2299", "right": "issue:5592", "accept": false, - "reason": "Both involve checkpoints, but they concern different models and different loading problems." + "reason": "Different model-loading contexts and failure modes; one is fine-tuned checkpoints in Diffusers, the other is UniDiffuser checkpoint loading." }, { "left": "issue:2645", "right": "issue:4134", "accept": false, - "reason": "Both are ControlNet-on-Mac issues, but one is a CUDA-availability assertion and the other is a tensor-shape mismatch." + "reason": "Both are MPS/CUDA device issues in ControlNet, but one is a CUDA assertion and the other is a tensor shape mismatch; not the same bug." }, { "left": "issue:2747", "right": "issue:7375", "accept": false, - "reason": "Both are docs-related, but they are clearly different documentation tasks." + "reason": "One is a generic docs update request; the other is a specific docs note about FreeInit outputs." }, { "left": "issue:4445", "right": "issue:725", "accept": false, - "reason": "A how-to question about LoRA finetuning is unrelated to adding device-map/accelerate support." + "reason": "A how-to question about LoRA finetuning is not the same as a memory-footprint feature request for device map/accelerate." }, { "left": "issue:5429", "right": "issue:9397", "accept": false, - "reason": "These are different adapter/scheduler feature requests with no shared bug or change." + "reason": "Both involve model/scheduler support additions, but they target different model families and different missing capabilities." }, { "left": "issue:6141", "right": "issue:7312", "accept": false, - "reason": "Different feature requests in unrelated parts of the pipeline stack." + "reason": "One requests ControlNet control/resize modes; the other is about AlignProp direct reward finetuning. Different features." }, { "left": "issue:10514", "right": "issue:11342", "accept": false, - "reason": "Both are Sana-related, but one is a resolution-binning/sample-size limitation and the other is an MPS runtime failure." + "reason": "Both are Sana pipeline issues, but one is a resolution-binning/sample_size constraint and the other is an MPS failure; distinct bugs." }, { "left": "issue:3901", "right": "issue:4390", "accept": false, - "reason": "Both mention requirements, but one is missing requirements.txt and the other is a generic missing-requirements report." + "reason": "Both are missing-file/environment setup errors, but one is missing requirements.txt and the other is generic missing requirements; not the same incident." }, { "left": "issue:6818", "right": "issue:8523", "accept": false, - "reason": "IP-Adapter unload/model-component reconstruction is unrelated to a generic playground CUDA error." + "reason": "Both mention runtime/back-end errors, but one is IP-Adapter unload construction and the other is a playground CUDA error; unrelated." }, { "left": "issue:2098", "right": "issue:3631", "accept": false, - "reason": "Both are runtime exceptions, but they affect different API calls and do not point to the same underlying bug." + "reason": "A torch.randn argument mismatch is unrelated to the inpainting latents exception." }, { "left": "issue:3391", "right": "issue:3487", "accept": false, - "reason": "Different pipelines and different failure modes: overexposure vs poor results with a specific scheduler." + "reason": "Both are image-quality complaints, but one is overexposure in super-resolution and the other is poor results with a specific scheduler/strength combo." }, { "left": "issue:10045", "right": "issue:6517", "accept": false, - "reason": "Both concern import failures, but 6517 is a specific ControlNet SDXL import traceback and not the same bug as the generic umbrella issue." + "reason": "A generic pipeline import problem is not the same as a specific ControlNet SDXL import traceback." }, { "left": "issue:10994", "right": "issue:9799", "accept": false, - "reason": "Both are Flux tensor-shape/runtime problems, but the titles do not identify the same code path or defect." + "reason": "Different tensor-shape crashes in different training/inference paths; not the same underlying bug." }, { "left": "issue:1037", "right": "issue:1101", "accept": false, - "reason": "Different feature additions: a community pipeline vs a scheduler." + "reason": "Both are feature requests for community schedulers/pipelines, but they are different additions." }, { "left": "issue:10675", "right": "issue:9822", "accept": false, - "reason": "Flux max_shift config and loading SDXL LoRAs into Flux are different problems." + "reason": "Flux scheduler config max_shift and loading SDXL LoRAs into Flux are unrelated issues." }, { "left": "issue:4554", "right": "issue:7836", "accept": false, - "reason": "OOM in an example and CLIP overfitting are unrelated." + "reason": "One is an OOM on a specific example; the other is CLIP training overfitting. No shared bug." }, { "left": "issue:4665", "right": "issue:8898", "accept": false, - "reason": "Both are model-loading issues, but one is Windows safetensors/bin selection and the other is subfolder path resolution." + "reason": "Both concern loading/safetensors behavior, but one is Windows alt-bin selection and the other is subfolder path validation; different defects." }, { "left": "issue:5335", "right": "issue:785", "accept": false, - "reason": "Completely unrelated deployment/network question vs scheduler performance issue." + "reason": "A deployment/repo-access question is unrelated to scheduler performance." }, { "left": "issue:5689", "right": "issue:5703", "accept": false, - "reason": "Both involve noise/schedulers, but they target different algorithms and different behaviors." + "reason": "Residual noise with DPM++ SDXL is different from ddim_inverse noise reconstruction; distinct scheduler behavior." }, { "left": "issue:10833", "right": "issue:13034", "accept": false, - "reason": "Both are shape-related runtime bugs, but in different pipelines and with different manifestations." + "reason": "Different model/pipeline-specific shape or quantization issues; no evidence they share the same failure path." }, { "left": "issue:4305", "right": "issue:5687", "accept": false, - "reason": "torch.compile graph breaks and duplicate timesteps are unrelated defects." - }, - { - "left": "issue:10772", - "right": "issue:8834", - "accept": false, - "reason": "A support request for Sana ControlNet and a question about releasing SD3 ControlNet training code are not duplicates." - }, - { - "left": "issue:2390", - "right": "issue:6243", - "accept": false, - "reason": "Both are adapter support requests, but they concern different adapters and different requested integrations." + "reason": "Both involve timesteps/schedulers, but one is a torch.compile graph break and the other is duplicate timesteps with Karras sigmas; different problems." } ] }, @@ -64994,13 +65013,14 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10287", "issue:10511", + "issue:10772", "issue:10958", "issue:1188", "issue:12239", @@ -65010,7 +65030,7 @@ "issue:1790", "issue:2026", "issue:2098", - "issue:2203", + "issue:2390", "issue:2462", "issue:2465", "issue:3176", @@ -65018,7 +65038,6 @@ "issue:322", "issue:3221", "issue:3366", - "issue:3694", "issue:4071", "issue:4148", "issue:4306", @@ -65032,12 +65051,12 @@ "issue:5605", "issue:5607", "issue:5670", + "issue:6243", "issue:6331", "issue:6661", - "issue:7012", "issue:7480", - "issue:817", "issue:8801", + "issue:8834", "issue:8885", "issue:8989", "issue:9112", @@ -65048,6 +65067,8 @@ "issue:9930" ], "soft_pairs": [ + "issue:10772|issue:8834", + "issue:2390|issue:6243", "issue:4148|issue:6331", "issue:5070|issue:5605", "issue:5168|issue:8801", @@ -65069,182 +65090,180 @@ "issue:3176|issue:3181", "issue:10287|issue:9237", "issue:1785|issue:5607", - "issue:2098|issue:4306", - "issue:2203|issue:817", - "issue:3694|issue:7012" + "issue:2098|issue:4306" ], - "prepared_review_unit_hash": "015503ad746abde62e3a8e98b428ec9218210984b6be762b557a686e13602f4f", + "prepared_review_unit_hash": "21ea6d7c0eb555f692c1c387a546e5521d6c80c037b9ce2289c1cf3f061355ab", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33242, - "estimated_input_tokens": 8311, - "estimated_eval_tokens": 16878 + "serialized_chars": 33068, + "estimated_input_tokens": 8267, + "estimated_eval_tokens": 16790 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All proposed soft pairs look like false positives: they span different pipelines, models, or unrelated support topics, so there is no real duplicate cluster here.", - "confidence": 0.97, - "canonical_issue_reason": "Issue 9237 is the strongest standalone report in the set: it is concrete, highly discussed, and has inbound references, so it is the best representative issue if one must pick an anchor.", + "summary": "This cluster is a loose grab-bag of mostly unrelated issues. The soft pairs share only broad subsystem words (ControlNet, Dreambooth, TensorRT, OOM, inpainting, etc.) but not the same concrete bug or change, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "If forced to pick one anchor, issue 10511 is the broadest open product request, but the set does not have a real common underlying problem, so no strong canonical issue exists.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 9237 is the most substantive and actionable issue among the items, with the clearest bug context and the most discussion.", + "best_issue_reason": "Issue 10511 is the most general and forward-looking open request in the set, making it the least arbitrary issue to anchor the cluster, though it is not actually a duplicate hub.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10772", + "right": "issue:8834", + "accept": false, + "reason": "Both mention ControlNet/SD3, but one is asking for Sana ControlNet support and the other for SD3 controlnet training code release; different artifacts and goals." + }, + { + "left": "issue:2390", + "right": "issue:6243", + "accept": false, + "reason": "T2I-Adapter support and IP-Adapter Face ID are different adapter types and different feature requests." + }, { "left": "issue:4148", "right": "issue:6331", "accept": false, - "reason": "Both mention inpainting, but one is an SDXL model-config error and the other is a SAG feature request; different problems." + "reason": "One is an SDXL inpainting config error, the other is a request for SAG inpainting support; not the same bug." }, { "left": "issue:5070", "right": "issue:5605", "accept": false, - "reason": "ControlNet example tweak vs SDXL Dreambooth caption support are unrelated changes." + "reason": "ControlNet example parameter changes and SDXL DreamBooth per-image caption support are unrelated training/example features." }, { "left": "issue:5168", "right": "issue:8801", "accept": false, - "reason": "Different model support requests (ControlNet-XS vs Kolors), not the same bug or change." + "reason": "ControlNet-XS support and Kolors support target different model families and changes." }, { "left": "issue:1566", "right": "issue:5124", "accept": false, - "reason": "Dreambooth fp16 checkpoint crash and checkpoint fine-tuning usage question are unrelated." + "reason": "DreamBooth fp16 checkpoint crash and a general fine-tuning question about safetensors are not the same issue." }, { "left": "issue:10958", "right": "issue:9112", "accept": false, - "reason": "Different pipelines and failures: CogView4 meta-tensor copy error vs Auraflow tensor-size mismatch." + "reason": "Both are inference errors, but they affect different pipelines and failure modes (meta tensor copy vs tensor size mismatch)." }, { "left": "issue:1790", "right": "issue:2465", "accept": false, - "reason": "Both are training questions, but one asks how training works and the other asks for a reproducible Dreambooth recipe; not duplicates." + "reason": "These are both training questions, but one is a generic 'how training works' request and the other asks for a reproducible DreamBooth recipe." }, { "left": "issue:5595", "right": "issue:5670", "accept": false, - "reason": "Both are shape-related runtime errors, but they affect different pipelines and error paths." + "reason": "Both concern size/shape problems, but one is about image dimensions needing to be divisible by 8 and the other is a PixArtAlpha non-square image runtime error." }, { "left": "issue:8885", "right": "issue:9763", "accept": false, - "reason": "Custom Diffusion install/loader issue is unrelated to Flux ControlNet token truncation during training." + "reason": "Custom Diffusion dependency/install/token-loading problems are not the same as Flux ControlNet token truncation during training." }, { "left": "issue:2026", "right": "issue:4724", "accept": false, - "reason": "Different pipelines and failures: PaintByExample cpu offload vs SDXL img2img checkpoint behavior." + "reason": "PaintByExample cpu_offload and SDXL Img2Img unexpected behavior are separate pipeline issues." }, { "left": "issue:4071", "right": "issue:4675", "accept": false, - "reason": "Both are feature requests around special conditioning, but they target different capabilities and code paths." + "reason": "Inpaint-only+lama support and SDXL ControlNet reference-only are different feature requests." }, { "left": "issue:5118", "right": "issue:9216", "accept": false, - "reason": "Reference-only ControlNet usage question is unrelated to a custom multi-model pipeline proposal." + "reason": "A question about ControlNet reference_only usage is unrelated to a custom pipeline that combines any number of models." }, { "left": "issue:6661", "right": "issue:7480", "accept": false, - "reason": "Both involve OOMs, but on different hardware/model setups and different workflows." + "reason": "Both mention OOM, but one is MPS training on specific image sizes and the other is DeepFloyd IF inference memory usage." }, { "left": "issue:12501", "right": "issue:9856", "accept": false, - "reason": "VAE graph-break issue and distributed port collision are unrelated." + "reason": "A VAE graph-break bug and a distributed port conflict are entirely different failures." }, { "left": "issue:3221", "right": "issue:3366", "accept": false, - "reason": "Both touch LoRA/attention-related areas, but one is a load_attn_procs version problem and the other is an autocast regression." + "reason": "load_attn_procs failing in diffusers 0.16.0 and LoRALinearLayer breaking autocast are different code paths and symptoms." }, { "left": "issue:8989", "right": "issue:9930", "accept": false, - "reason": "Stable Audio sequential offloading and PAG adaptive scale are different features with different failures." + "reason": "Sequential offloading in Stable Audio and PAG adaptive scale are unrelated issues." }, { "left": "issue:10511", "right": "issue:12239", "accept": false, - "reason": "Scheduler redesign for SD3/Flux and InfiniteTalk support are unrelated requests." + "reason": "FlowMatchEuler scheduler redesign for SD3/Flux and InfiniteTalk support are different model-specific requests." }, { "left": "issue:1188", "right": "issue:322", "accept": false, - "reason": "A new CLIPSeg+Stable Diffusion pipeline request is unrelated to V-Diffusion from Katherine." + "reason": "CLIPSeg x Stable Diffusion pipeline support and V-Diffusion are unrelated pipeline requests." }, { "left": "issue:2462", "right": "issue:4071", "accept": false, - "reason": "IterableDataset training support and inpaint-only/lama support are different feature requests." + "reason": "IterableDataset training support and inpaint-only+lama support are different areas and not the same underlying change." }, { "left": "issue:3176", "right": "issue:3181", "accept": false, - "reason": "Both are TensorRT example problems, but the reported errors and likely code paths differ; not enough to treat as one bug." + "reason": "Both are TensorRT-related, but one is a demo error and the other is a different example TypeError; not enough evidence of the same fix." }, { "left": "issue:10287", "right": "issue:9237", "accept": false, - "reason": "Documentation example issue vs Flux Dreambooth LoRA training bug are unrelated." + "reason": "Documentation example issues and Flux DreamBooth LoRA training problems are unrelated." }, { "left": "issue:1785", "right": "issue:5607", "accept": false, - "reason": "UnCLIP num_images_per_prompt bug and a tensor-size mismatch error are different failures." + "reason": "UnCLIPPipeline num_images_per_prompt bug and a tensor size mismatch in another path are different failures." }, { "left": "issue:2098", "right": "issue:4306", "accept": false, - "reason": "Torch.randn argument error and StableDiffusionReferencePipeline index error are unrelated." - }, - { - "left": "issue:2203", - "right": "issue:817", - "accept": false, - "reason": "Dreambooth download-path question and mixed-precision bug are unrelated." - }, - { - "left": "issue:3694", - "right": "issue:7012", - "accept": false, - "reason": "Missing config/model files in the pokemon example and v_prediction image quality issues are different problems." + "reason": "Torch.randn argument errors and StableDiffusionReferencePipeline list index errors are unrelated." } ] }, @@ -65271,7 +65290,6 @@ "issue:10428", "issue:10448", "issue:11050", - "issue:11275", "issue:1141", "issue:11648", "issue:11938", @@ -65283,7 +65301,7 @@ "issue:1537", "issue:201", "issue:2161", - "issue:2326", + "issue:2203", "issue:2377", "issue:2440", "issue:2518", @@ -65293,11 +65311,11 @@ "issue:2784", "issue:3043", "issue:3073", + "issue:3694", "issue:3986", "issue:4213", "issue:4488", "issue:4593", - "issue:4764", "issue:4846", "issue:4912", "issue:5515", @@ -65307,10 +65325,11 @@ "issue:6463", "issue:6561", "issue:6742", + "issue:7012", "issue:7519", "issue:7661", "issue:7841", - "issue:8463", + "issue:817", "issue:8759", "issue:9047", "issue:9241", @@ -65318,6 +65337,8 @@ "issue:9899" ], "soft_pairs": [ + "issue:2203|issue:817", + "issue:3694|issue:7012", "issue:10428|issue:9299", "issue:10448|issue:11648", "issue:1141|issue:4912", @@ -65339,182 +65360,180 @@ "issue:2737|issue:3986", "issue:2772|issue:4593", "issue:3043|issue:3073", - "issue:4213|issue:6321", - "issue:11275|issue:8463", - "issue:2326|issue:4764" + "issue:4213|issue:6321" ], - "prepared_review_unit_hash": "22e9d1d571e0f312e13db9ea740eb82a07d2993a14b022c10c2a4af99c174fcb", + "prepared_review_unit_hash": "74e5c77475202ba1ac523cf314847a70f112ceb45ca1ccdbcd677c5cd0ea4e74", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33552, - "estimated_input_tokens": 8388, - "estimated_eval_tokens": 17032 + "serialized_chars": 33587, + "estimated_input_tokens": 8397, + "estimated_eval_tokens": 17050 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a set of unrelated issues from very different areas (Flux/NPU, DDUF mmap, training dtype, VAE bugs, local files, LoRA conversion/loading, attention backends, scheduler/runtime errors, etc.). The soft edges look like title-only similarity false positives rather than duplicates.", + "summary": "This cluster is mostly a grab bag of unrelated issues with superficial lexical overlap only. None of the soft pairs look like the same underlying bug/change, so all soft edges should be rejected.", "confidence": 0.97, - "canonical_issue_reason": null, + "canonical_issue_reason": "No single canonical duplicate target stands out here; the issues span unrelated subsystems, hardware backends, training scripts, and feature requests.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:12575 is the most concrete self-contained bug report in the set, with a specific subsystem ([Chroma]), a clear failure mode, and enough detail to serve as the best representative issue.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2203", + "right": "issue:817", + "accept": false, + "reason": "Different problems: model download location vs mixed-precision training failure." + }, + { + "left": "issue:3694", + "right": "issue:7012", + "accept": false, + "reason": "One is about missing model/config files during finetuning; the other is about bad outputs with v-prediction." + }, { "left": "issue:10428", "right": "issue:9299", "accept": false, - "reason": "Flux on Ascend NPU vs Flux CUDA graph position embeddings; different hardware and problem types." + "reason": "Both mention Flux, but one is an Ascend NPU inference error and the other is about CUDAGraphs for position embeddings." }, { "left": "issue:10448", "right": "issue:11648", "accept": false, - "reason": "DDUF mmap loading vs LoRA weight loading on fp8 transformer models; unrelated feature requests." + "reason": "DDUF mmap loading is unrelated to loading LoRA weights into an FP8 transformer model." }, { "left": "issue:1141", "right": "issue:4912", "accept": false, - "reason": "Local custom pipeline files vs converting CivitAI SDXL checkpoints; both about loading but different workflows." + "reason": "Local-file support in custom pipelines is a different concern from converting CivitAI SDXL checkpoints." }, { "left": "issue:1343", "right": "issue:2518", "accept": false, - "reason": "Default xformers enablement vs missing xformers support in one training script; different scope and fix." + "reason": "General xformers default enablement is not the same as an example script missing xformers support." }, { "left": "issue:201", "right": "issue:2377", "accept": false, - "reason": "Attention class tests vs EMAModel tests; unrelated testing tasks." + "reason": "These are different test requests for different components: attention class vs EMAModel." }, { "left": "issue:2161", "right": "issue:2552", "accept": false, - "reason": "LoRA training on inpainting model vs inpainting pipeline tutorial question; one bug/feature, one documentation/support issue." + "reason": "Training LoRA on inpainting models is distinct from asking for tutorial/docs about StableDiffusionInpaintPipeline." }, { "left": "issue:5731", "right": "issue:6463", "accept": false, - "reason": "AutoencoderKL logvar implementation bug vs missing audio_diffusion pipeline; unrelated components." + "reason": "AutoencoderKL logvar behavior is unrelated to the missing audio_diffusion pipeline." }, { "left": "issue:6561", "right": "issue:7661", "accept": false, - "reason": "Missing IP-Adapter config file vs animate diff with IP-Adapter-loaded UNet; different failure modes." + "reason": "Missing IP-Adapter config.json is a loading error; the other is about enabling animate diff with an IP-adapter-loaded UNet." }, { "left": "issue:7519", "right": "issue:7841", "accept": false, - "reason": "xFormers scale kwarg warning vs SDXL red-dot visualization artifact; not the same bug." + "reason": "An ignored xformers keyword is not the same bug as SDXL images showing red dots when enlarged." }, { "left": "issue:9241", "right": "issue:9899", "accept": false, - "reason": "XLab LoRA conversion script vs converting a transformer (not UNet) to checkpoint; different conversion targets." + "reason": "Converting XLab LoRA to diffusers LoRA is unrelated to converting a transformer to a checkpoint." }, { "left": "issue:5515", "right": "issue:5553", "accept": false, - "reason": "LCM pipeline missing scheduler vs DiffusionPipeline loading custom model; both load-related but distinct code paths." + "reason": "LCM scheduler initialization failure and custom model loading failure are different code-paths." }, { "left": "issue:12575", "right": "issue:13506", "accept": false, - "reason": "Chroma attention backend check error vs Sage attention with ring-attention; different backend compatibility issues." + "reason": "Both concern attention backends, but one is a Chroma backend-check error and the other is Sage attention failing with ring-attention; not the same bug." }, { "left": "issue:1519", "right": "issue:1537", "accept": false, - "reason": "Noisy batch output on Apple M2 vs MPS scheduler behavior; both Apple-related but different symptoms and fixes." + "reason": "Apple M2 batch-image noise and Apple MPS scheduler issues are different symptoms and likely different root causes." }, { "left": "issue:1477", "right": "issue:2440", "accept": false, - "reason": "CUDA cublas GEMM invalid value vs inpainting load float16 error; unrelated runtime failures." + "reason": "A CUDA cublas batched GEMM error is not the same as failing to load an inpainting model in float16." }, { "left": "issue:11938", "right": "issue:9047", "accept": false, - "reason": "RMSNorm registration bug vs Flux not working on MPS; different modules and platforms." + "reason": "RMSNorm registration bug and Flux not working on MPS are unrelated." }, { "left": "issue:11050", "right": "issue:8759", "accept": false, - "reason": "bf16/fp16 prompt embed dtype casting in training vs SD3 image_processor NaN cast warning; different tensor paths." + "reason": "Mixed-precision casting in a training script is different from a runtime warning in SD3 image postprocessing." }, { "left": "issue:2784", "right": "issue:4846", "accept": false, - "reason": "Dreambooth for text2video request vs strange Dreambooth results; one feature request, one training quality report." + "reason": "A feature request for text-to-video dreambooth support is not the same as a report of strange dreambooth training results." }, { "left": "issue:4488", "right": "issue:6742", "accept": false, - "reason": "SDXL checkpoint loading without pipeline_class vs checkpointing changes DownBlockMotion behavior; unrelated loading vs execution bug." + "reason": "SDXL checkpoint-loading behavior and DownBlockMotion checkpointing differences are different issues." }, { "left": "issue:2737", "right": "issue:3986", "accept": false, - "reason": "CLIP guidance for DPMSolverMultistepScheduler vs mesh output for ShapEPipeline; different pipeline features." + "reason": "Scheduler support for CLIP guidance is unrelated to adding mesh output in ShapEPipeline." }, { "left": "issue:2772", "right": "issue:4593", "accept": false, - "reason": "Local dataset handling in train_controlnet.py vs LoRA rank too high for Automatic1111 imports; unrelated training vs LoRA compatibility." + "reason": "Local dataset handling in train_controlnet.py and LoRA rank mismatch loading are different bugs." }, { "left": "issue:3043", "right": "issue:3073", "accept": false, - "reason": "Add ODISE vs add consistency models; distinct model additions." + "reason": "Add ODISE and add consistency models are separate feature requests." }, { "left": "issue:4213", "right": "issue:6321", "accept": false, - "reason": "from_single_file() SDXL loading without accelerate vs CLIP model not detected; unrelated loading dependencies." - }, - { - "left": "issue:11275", - "right": "issue:8463", - "accept": false, - "reason": "VAE ResnetUpsampleBlock2D NoneType addition error vs InstructP2P-SDXL tokenizer NoneType error; different failing components." - }, - { - "left": "issue:2326", - "right": "issue:4764", - "accept": false, - "reason": "Converting LoRA to safetensors/ckpt vs loading an OFT-generated LoRA; both LoRA-related but distinct formats and failures." + "reason": "SDXL from_single_file without accelerate and missing openai/clip-vit-large-patch14 are unrelated loading issues." } ] }, @@ -65533,22 +65552,22 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10653", "issue:10757", + "issue:11275", "issue:1188", "issue:11922", - "issue:11989", "issue:12653", "issue:13015", "issue:1895", "issue:1969", "issue:1992", "issue:201", + "issue:2326", "issue:2571", "issue:321", "issue:3222", @@ -65556,7 +65575,6 @@ "issue:3422", "issue:3487", "issue:3582", - "issue:3771", "issue:3829", "issue:3974", "issue:4171", @@ -65564,6 +65582,7 @@ "issue:4194", "issue:4304", "issue:4436", + "issue:4764", "issue:4913", "issue:5150", "issue:5168", @@ -65576,18 +65595,20 @@ "issue:6530", "issue:6928", "issue:7061", - "issue:7080", "issue:7421", "issue:7568", "issue:7877", "issue:7985", "issue:8234", + "issue:8463", "issue:9062", "issue:9511", "issue:9611", "issue:9889" ], "soft_pairs": [ + "issue:11275|issue:8463", + "issue:2326|issue:4764", "issue:3377|issue:4304", "issue:3422|issue:6928", "issue:3829|issue:7421", @@ -65609,182 +65630,180 @@ "issue:4194|issue:4436", "issue:4913|issue:6530", "issue:1895|issue:1969", - "issue:3222|issue:4185", - "issue:3771|issue:7080", - "issue:10653|issue:11989" + "issue:3222|issue:4185" ], - "prepared_review_unit_hash": "d407189abfc2980d03ed7c0d91badf5ffedba3b44600aa6439c5c30b37f2bc65", + "prepared_review_unit_hash": "75054b666d0b614e537ad451613fc6a10e3749663cebd12f9993504e18dfb37c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33975, - "estimated_input_tokens": 8494, - "estimated_eval_tokens": 17244 + "serialized_chars": 33836, + "estimated_input_tokens": 8459, + "estimated_eval_tokens": 17174 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft pairs are mostly false positives: they share broad themes like diffusion, ControlNet, ONNX, or device handling, but the concrete bugs/feature requests are different. No single duplicate cluster emerges.", - "confidence": 0.95, - "canonical_issue_reason": "No clear canonical issue stands out; the items span unrelated bugs, feature requests, docs issues, and training/inference failures rather than one underlying defect.", + "summary": "The list is a grab bag of unrelated bug reports and feature requests. The soft-similarity pairs are mostly superficial keyword matches; none look like the same underlying issue or change.", + "confidence": 0.88, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No issue is a strong global representative of a duplicate set, so there is no suitable best canonical issue.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11275", + "right": "issue:8463", + "accept": false, + "reason": "Both are runtime errors, but they affect different code paths: VAE ResnetUpsampleBlock2D addition vs InstructP2P-SDXL tokenizer handling." + }, + { + "left": "issue:2326", + "right": "issue:4764", + "accept": false, + "reason": "Both mention LoRA conversion/loading, but one is about conversion to safetensors/ckpt and the other about loading a specific OFT-generated LoRA file." + }, { "left": "issue:3377", "right": "issue:4304", "accept": false, - "reason": "Both concern training, but one is about SNR weighting and the other about adding SDXL LoRA support; different requests." + "reason": "Training-related topics, but one asks about SNR weighting optimization and the other is a request for SDXL LoRA support." }, { "left": "issue:3422", "right": "issue:6928", "accept": false, - "reason": "Unrelated features: disabling the safety checker vs adding a depthmap colorization option." + "reason": "Both are about pipeline options, but disabling the safety checker is unrelated to adding a depthmap colorization toggle." }, { "left": "issue:3829", "right": "issue:7421", "accept": false, - "reason": "Both touch determinism/reproducibility, but one is ONNX text-to-image output variance and the other is seed handling in K-Diffusion; different code paths." + "reason": "Both involve determinism/seed behavior, but one is ONNX EulerAncestral output variance and the other is KDiffusion pipeline seed handling." }, { "left": "issue:4171", "right": "issue:7061", "accept": false, - "reason": "One is a homepage render bug, the other an incorrect docstring; not the same underlying issue." + "reason": "These are documentation issues, but they concern different pages and different defects: homepage reference rendering vs an incorrect community pipeline docstring." }, { "left": "issue:5150", "right": "issue:6383", "accept": false, - "reason": "Missing from_single_file support is unrelated to scheduler multi-stage learning rates." + "reason": "One is missing a pipeline method, the other is about scheduler learning-rate support; not the same bug." }, { "left": "issue:5968", "right": "issue:6412", "accept": false, - "reason": "Both mention ONNX, but one is video conversion scripts and the other is ControlNet annotators; different deliverables." + "reason": "Both are conversion/export requests, but text/video ONNX conversion and ControlNet annotator ONNX support are distinct features." }, { "left": "issue:7568", "right": "issue:9889", "accept": false, - "reason": "Prompt-token interpolation is unrelated to an sdxl-turbo segmentation fault." + "reason": "Both involve inference behavior, but interpolation token handling is unrelated to an sdxl-turbo segmentation fault under a specific torch version." }, { "left": "issue:7985", "right": "issue:8234", "accept": false, - "reason": "Both involve IPAdapter, but one is torch.compile incompatibility and the other is a dict-vs-tensor scaling error; different bugs." + "reason": "IPAdapter-related issues, but one is a torch.compile incompatibility and the other is a scale-argument type error with IPAdapter Plus." }, { "left": "issue:2571", "right": "issue:5849", "accept": false, - "reason": "Different failures: missing ProjectConfiguration import vs a training script main() signature error." + "reason": "Both are import/CLI failures, but they stem from different causes: accelerate ProjectConfiguration import breakage vs a missing args parameter in an SDXL LoRA training entrypoint." }, { "left": "issue:12653", "right": "issue:13015", "accept": false, - "reason": "Both are device/dtype-related, but one is MPS rope float64 handling and the other is an NPU tensor placement error; not the same fix." + "reason": "Both are device/dtype portability problems, but one is a float64 rope reference breaking MPS and the other is an NPU tensor device mismatch in QwenImageEditPlus." }, { "left": "issue:1188", "right": "issue:321", "accept": false, - "reason": "A CLIPSeg+Stable Diffusion pipeline request is unrelated to the Blended Diffusion feature request." + "reason": "Different feature requests: adding a CLIPSeg+Stable Diffusion pipeline versus a community blended diffusion request." }, { "left": "issue:1992", "right": "issue:201", "accept": false, - "reason": "A JAX attention refactor proposal and an attention test request are not duplicates." + "reason": "One is a JAX attention refactor question, the other is a tests request for an attention class; not the same underlying change." }, { "left": "issue:3487", "right": "issue:3582", "accept": false, - "reason": "Both mention inpaint strength, but the reported symptoms and pipelines differ; not clearly one underlying bug." + "reason": "Both involve inpaint strength behavior, but one reports poor results with a scheduler and the other says strength=1 does not work in ControlNet inpaint." }, { "left": "issue:3974", "right": "issue:5168", "accept": false, - "reason": "Support for `.safetensors` in a ControlNet inpaint pipeline is different from requesting ControlNet-XS support." + "reason": "Both mention ControlNet, but one is safetensors loading support and the other is ControlNet-XS support; different features." }, { "left": "issue:6141", "right": "issue:9511", "accept": false, - "reason": "ControlNet parameter support vs multi-controlnet batching are different enhancement areas." + "reason": "Both are about ControlNet options, but control_mode/resize_mode support is unrelated to multi-controlnet batching." }, { "left": "issue:11922", "right": "issue:9611", "accept": false, - "reason": "A general FLUX.1-Fill-dev running problem is not the same as a specific batching error in FluxImg2Img/FluxInpaint." + "reason": "Both are Flux pipeline problems, but one is a generic FLUX.1-Fill-dev run issue and the other is a specific batching error in img2img/inpaint pipelines." }, { "left": "issue:5605", "right": "issue:7877", "accept": false, - "reason": "Per-image caption support for dreambooth is unrelated to adding an out_dim argument to Attention." + "reason": "Both request pipeline/block extensions, but per-image caption support for SDXL DreamBooth is unrelated to adding an out_dim argument to Attention." }, { "left": "issue:10757", "right": "issue:9062", "accept": false, - "reason": "A variable rename in a conversion script is unrelated to an import error for StableAudioPipeline." + "reason": "Both are about names/imports, but one is a variable rename in a conversion script and the other is an import failure for StableAudioPipeline." }, { "left": "issue:4194", "right": "issue:4436", "accept": false, - "reason": "Both involve VAE/conversion, but one is a from_single_file inference issue and the other is a CIVITAI safetensors conversion request." + "reason": "Both are conversion/loading issues, but from_single_file VAE behavior and converting CIVITAI safetensors with VAE are different problems." }, { "left": "issue:4913", "right": "issue:6530", "accept": false, - "reason": "Different ControlNet topics: openpose SDXL support vs K-diffusion with ControlNet." + "reason": "Both mention ControlNet/SDXL, but one is a model/pipeline request and the other is about K-diffusion support with ControlNet." }, { "left": "issue:1895", "right": "issue:1969", "accept": false, - "reason": "EMA updates in multi-GPU training are unrelated to a tensor view/stride runtime error." + "reason": "Both are runtime/training errors, but EMA multi-GPU update logic and tensor view/stride compatibility are unrelated." }, { "left": "issue:3222", "right": "issue:4185", "accept": false, - "reason": "Custom diffusion training output missing a weights file is unrelated to black inference images in SDXL ControlNet." - }, - { - "left": "issue:3771", - "right": "issue:7080", - "accept": false, - "reason": "A broken CKPT-to-Diffusers conversion path is different from a training-time wandb inference-output mismatch." - }, - { - "left": "issue:10653", - "right": "issue:11989", - "accept": false, - "reason": "bitsandbytes CPU transfer limitations are unrelated to a black-output inference problem." + "reason": "Both involve inpainting/control-related behavior, but missing custom diffusion output files and SDXL ControlNet black-image inference are not the same bug." } ] }, @@ -65803,18 +65822,20 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10194", "issue:10311", "issue:1062", + "issue:10653", "issue:10705", "issue:10891", "issue:10893", "issue:11218", + "issue:11989", "issue:12141", "issue:12412", "issue:13053", @@ -65831,6 +65852,7 @@ "issue:2883", "issue:3336", "issue:3591", + "issue:3771", "issue:3826", "issue:3975", "issue:4217", @@ -65843,21 +65865,20 @@ "issue:4826", "issue:5002", "issue:5166", - "issue:5412", "issue:5605", "issue:5719", - "issue:5984", "issue:696", - "issue:7191", + "issue:7080", "issue:7408", "issue:834", "issue:8752", "issue:9063", "issue:911", - "issue:9181", "issue:9188" ], "soft_pairs": [ + "issue:3771|issue:7080", + "issue:10653|issue:11989", "issue:2645|issue:4826", "issue:4782|issue:4815", "issue:10194|issue:9188", @@ -65879,182 +65900,180 @@ "issue:1643|issue:696", "issue:460|issue:834", "issue:5002|issue:5166", - "issue:1831|issue:4551", - "issue:7191|issue:9181", - "issue:5412|issue:5984" + "issue:1831|issue:4551" ], - "prepared_review_unit_hash": "207ddd6ef456154d3a23be8f15f6afdb2df220f257fe39739f5b91bed903186f", + "prepared_review_unit_hash": "7cf54367f6b8de04ea1cb2e501b0db05c9dd61343c6f903acdb353aa5e9b3698", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33456, - "estimated_input_tokens": 8364, - "estimated_eval_tokens": 16984 + "serialized_chars": 33643, + "estimated_input_tokens": 8411, + "estimated_eval_tokens": 17078 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very noisy: the items span many unrelated bugs and feature requests, and none of the soft-similarity pairs look like true duplicates of the same underlying change or defect.", - "confidence": 0.37, - "canonical_issue_reason": "issue:9063 is the best single anchor only because it is the most concrete Flux-related bug report in the set; the other Flux items are broader or different requests (attention masking, outpainting, pipeline reuse) rather than the same failure.", + "summary": "This cluster is not a true duplicate set; it mixes many unrelated feature requests and bugs across Dreambooth, ControlNet, schedulers, Flux, video pipelines, and loading/import errors. Most soft pairs share only very broad subsystem words and should not be merged.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 10194 is the broadest, still-open request in the set and best serves as a loose representative for the Flux-related subset, though the cluster overall is heterogeneous.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9063 is the most suitable global representative for this cluster because it has a specific, reproducible technical problem and is closer to the only loose thematic overlap in the set (Flux pipeline issues).", + "best_issue_reason": "10194 is the most general and current issue among the items; it is a reasonable anchor only because no tighter canonical duplicate exists.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3771", + "right": "issue:7080", + "accept": false, + "reason": "Both mention inference/output problems, but they concern different pipelines and failure modes; not the same bug." + }, + { + "left": "issue:10653", + "right": "issue:11989", + "accept": false, + "reason": "An 8-bit .to(cpu) limitation and a black-image rendering issue are unrelated symptoms in different code paths." + }, { "left": "issue:2645", "right": "issue:4826", "accept": false, - "reason": "Both are ControlNet-related, but one is a macOS CUDA assertion and the other is a MultiControlNet reference error; different code paths and fixes." + "reason": "Both are ControlNet-related, but one is a Mac/CUDA assertion and the other is a MultiControlNet/reference error; different bugs." }, { "left": "issue:4782", "right": "issue:4815", "accept": false, - "reason": "Both mention inpainting, but one is a mask correctness bug and the other is a feature request for latents-only inputs; not the same issue." + "reason": "Inpainting mask corruption and latents-only input support are different change requests, not the same underlying issue." }, { "left": "issue:10194", "right": "issue:9188", "accept": false, - "reason": "Both are Flux-related, but attention masking support and outpainting are different features/problems." + "reason": "Both are Flux-related feature requests, but attention masking and outpainting are distinct capabilities." }, { "left": "issue:10311", "right": "issue:11218", "accept": false, - "reason": "Both are import errors, but they involve different scripts and different missing symbols/packages." + "reason": "These are separate import errors involving different modules and missing symbols." }, { "left": "issue:10893", "right": "issue:13053", "accept": false, - "reason": "These are separate model integration feature requests for different pipelines/models." + "reason": "Unrelated model-support feature requests for different pipelines." }, { "left": "issue:1594", "right": "issue:2883", "accept": false, - "reason": "Both are about video generation, but one asks for img2img video generation and the other for an image-to-video pipeline; not the same concrete change." + "reason": "Both concern video generation, but one asks for img2img support and the other for an image-to-video pipeline; not the same request." }, { "left": "issue:1976", "right": "issue:911", "accept": false, - "reason": "A tokenizer/config file lookup error is unrelated to a bad Dreambooth config report." + "reason": "Both are config-related, but one is a tokenizer file lookup error and the other is a Dreambooth bad-config report." }, { "left": "issue:2311", "right": "issue:3336", "accept": false, - "reason": "Different scheduler features with different algorithms and requested behavior." + "reason": "Different scheduler feature requests; they are not the same algorithm or code-path change." }, { "left": "issue:3826", "right": "issue:7408", "accept": false, - "reason": "Different new pipeline proposals with no shared code-path or bug." + "reason": "Both are new pipeline requests, but for entirely different pipelines and use cases." }, { "left": "issue:1062", "right": "issue:1730", "accept": false, - "reason": "Both touch Dreambooth, but one is about reproducibility and the other about CKPT conversion/webui compatibility." + "reason": "Dreambooth-related, but one is about reproducing paper results and the other about CKPT conversion/webui inference." }, { "left": "issue:10705", "right": "issue:9063", "accept": false, - "reason": "Both mention Flux, but one is a pipeline reuse slowdown and the other is a tokenizer instantiation/import-style error." + "reason": "Pipeline reuse slowdown and Flux tokenizer instantiation errors are unrelated." }, { "left": "issue:12412", "right": "issue:13191", "accept": false, - "reason": "Separate feature requests for different model families." + "reason": "Separate support requests for different model families; no shared bug or change." }, { "left": "issue:2875", "right": "issue:3591", "accept": false, - "reason": "Both mention training metrics, but one is a text-to-image training NaN bug and the other is a ControlNet loss-curve question." + "reason": "Training loss instability and ControlNet loss-curve confusion are different problems." }, { "left": "issue:4376", "right": "issue:8752", "accept": false, - "reason": "Both involve loading/model import behavior, but the failures are different: textual inversion crash vs offline from_single_file loading." + "reason": "Both are SDXL pipeline issues, but one is textual inversion crashing and the other is offline from_single_file loading." }, { "left": "issue:4217", "right": "issue:4773", "accept": false, - "reason": "Both involve model loading, but one is safetensors load performance and the other is LoRA loading support." + "reason": "Both involve loading, but one is safetensors load time and the other is CivitAI LoRA loading; different issues." }, { "left": "issue:10891", "right": "issue:12141", "accept": false, - "reason": "Both are dtype/precision issues, but they affect different pipelines and different conversion paths." + "reason": "Both mention bfloat16, but they affect different models/pipelines and likely different dtype handling paths." }, { "left": "issue:2841", "right": "issue:5605", "accept": false, - "reason": "A scheduler temperature feature request is unrelated to SDXL Dreambooth per-image caption support." + "reason": "Unrelated feature requests: scheduler sampling temperature vs SDXL Dreambooth captions." }, { "left": "issue:3975", "right": "issue:5719", "accept": false, - "reason": "Both are about loading-related messaging, but one is a typo fix and the other is file-selection logic for LoRA weights." + "reason": "A typo in an error message and LoRA file-selection logic are unrelated." }, { "left": "issue:1643", "right": "issue:696", "accept": false, - "reason": "Both are Dreambooth-related, but one is an xformers API mismatch and the other is CUDA OOM; different root causes." + "reason": "Both are Dreambooth reports, but one is an xFormers API mismatch and the other is CUDA OOM." }, { "left": "issue:460", "right": "issue:834", "accept": false, - "reason": "ONNX runtime dependency selection is unrelated to CPU under-utilization." + "reason": "One is an ONNX runtime dependency problem; the other is a generic CPU utilization complaint." }, { "left": "issue:5002", "right": "issue:5166", "accept": false, - "reason": "Different scheduler/VQ-VAE missing-parameter issues with no shared code-path." + "reason": "Different scheduler/VQ-VAE issues with no shared concrete bug." }, { "left": "issue:1831", "right": "issue:4551", "accept": false, - "reason": "Both are Dreambooth training issues, but one is class-image precision mismatch and the other is an SDXL crash." - }, - { - "left": "issue:7191", - "right": "issue:9181", - "accept": false, - "reason": "The titles are both about odd pipeline behavior, but one is an import problem and the other is a visual noise artifact in ControlNet." - }, - { - "left": "issue:5412", - "right": "issue:5984", - "accept": false, - "reason": "Both are SDXL Dreambooth LoRA training issues, but one is checkpoint resumption failure and the other is fp16 RAM usage." + "reason": "Both involve Dreambooth, but one is class-image precision mismatch and the other is an SDXL Dreambooth crash." } ] }, @@ -66073,19 +66092,17 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10227", - "issue:1033", "issue:10710", "issue:10928", "issue:1106", "issue:1156", "issue:1478", - "issue:1572", "issue:1657", "issue:1805", "issue:1909", @@ -66104,11 +66121,12 @@ "issue:4638", "issue:4704", "issue:4727", - "issue:4806", "issue:5079", + "issue:5412", "issue:5476", "issue:5616", "issue:5953", + "issue:5984", "issue:6093", "issue:6201", "issue:6372", @@ -66116,18 +66134,21 @@ "issue:6513", "issue:684", "issue:713", + "issue:7191", "issue:7276", "issue:7390", "issue:7454", "issue:8228", - "issue:8373", "issue:8495", "issue:8588", "issue:8671", "issue:9078", + "issue:9181", "issue:9241" ], "soft_pairs": [ + "issue:7191|issue:9181", + "issue:5412|issue:5984", "issue:7390|issue:8228", "issue:1156|issue:684", "issue:10227|issue:10710", @@ -66149,189 +66170,190 @@ "issue:8588|issue:8671", "issue:1478|issue:5953", "issue:4704|issue:7276", - "issue:3963|issue:4727", - "issue:1033|issue:1572", - "issue:4806|issue:8373" + "issue:3963|issue:4727" ], - "prepared_review_unit_hash": "aab8b3e1f96b11aa947584cfb7b3b454318c0b9215537b3ca9081374cd1a2a3c", + "prepared_review_unit_hash": "0bd595aef088b0df683159f90193218d8302542a550451a593d6cc804a983d2a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33745, - "estimated_input_tokens": 8437, - "estimated_eval_tokens": 17130 + "serialized_chars": 33741, + "estimated_input_tokens": 8436, + "estimated_eval_tokens": 17128 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate cluster emerges here: the soft-edge pairs are mostly generic token matches across unrelated issues. All soft edges should be rejected.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "Most pairs are only superficially similar; they concern different models, pipelines, or failure modes. The only clear duplicate is the multiple-generator/DDIMScheduler issue pair 1909\u21942083.", + "confidence": 0.92, + "canonical_issue_reason": "issue:1909 is the best canonical issue: it states the broad root problem (pipelines accept a list of generators, but schedulers do not), while issue:2083 is a narrower DDIMScheduler/eta>0 manifestation.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:1909 is the strongest global representative because it is the most general and actionable description of the shared bug.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:7191", + "right": "issue:9181", + "accept": false, + "reason": "Different problems: an import failure versus a ControlNet pipeline artifact/noise issue." + }, + { + "left": "issue:5412", + "right": "issue:5984", + "accept": false, + "reason": "Both are DreamBooth/SDXL-related, but one is resume_from_checkpoint failure and the other is fp16 RAM usage." + }, { "left": "issue:7390", "right": "issue:8228", "accept": false, - "reason": "Both mention loading problems, but one is about SDXL ControlNet inpaint from_single_file and the other about CLIPTextModel loading in a conversion path; different failure modes and code paths." + "reason": "Both mention from_single_file, but they describe different load failures on different model components and paths." }, { "left": "issue:1156", "right": "issue:684", "accept": false, - "reason": "Both are training-related, but one is a generic training failure and the other is AMD-specific memory optimization issues; not the same bug." + "reason": "Generic training failure vs AMD memory-optimization problems; no shared concrete bug." }, { "left": "issue:10227", "right": "issue:10710", "accept": false, - "reason": "LoRA NotImplementedError vs DDUF support question; unrelated functionality." + "reason": "LoRA feature-support error and DDUF format support are unrelated." }, { "left": "issue:5616", "right": "issue:6427", "accept": false, - "reason": "Both concern model conversion/loading errors, but the SDXL safetensors load error and diffusers-to-ckpt conversion bug are different problems." + "reason": "Both are model conversion/loading complaints, but they target different workflows and failure points." }, { "left": "issue:10928", "right": "issue:8495", "accept": false, - "reason": "Both are format/load errors, but one is missing fp16 UNet weights and the other is a tokenizer enum parsing failure in SD3." + "reason": "Missing fp16 weight file versus SD3 tokenizer enum parsing error; unrelated." }, { "left": "issue:1106", "right": "issue:2480", "accept": false, - "reason": "PNDM repeatedly calling UNet is a scheduler/runtime issue; the DreamBooth checkpoint folder issue is a training-output path problem." + "reason": "Repeated UNet calls in PNDM is unrelated to a DreamBooth script checkpoint folder issue." }, { "left": "issue:1657", "right": "issue:5476", "accept": false, - "reason": "Kaggle DreamBooth examples request vs whether xformers-trained weights can load without xformers; unrelated." + "reason": "Kaggle examples request versus xformers portability question; not the same bug or change." }, { "left": "issue:1805", "right": "issue:322", "accept": false, - "reason": "Request for a scalable diffusion pipeline vs a V-Diffusion feature request; not the same change." + "reason": "Different feature requests for different diffusion approaches." }, { "left": "issue:1909", "right": "issue:2083", - "accept": false, - "reason": "Both involve generators, but one is a general API mismatch between pipelines and schedulers, while the other is a specific DDIMScheduler eta failure; not the same bug." + "accept": true, + "reason": "Same underlying bug: multiple generators are passed through pipelines, but DDIMScheduler/eta>0 handling only supports a single generator." }, { "left": "issue:2214", "right": "issue:2769", "accept": false, - "reason": "NSFW filter issue vs model round-trip conversion corruption; unrelated." + "reason": "NSFW filter regression is unrelated to conversion round-tripping breaking a model." }, { "left": "issue:2277", "right": "issue:6093", "accept": false, - "reason": "Wrong attention_head_dim length in UNet config vs push_to_hub behavior in SDXL consistency-model LoRA training; different components." + "reason": "UNet config length validation issue versus push_to_hub behavior in a training script." }, { "left": "issue:2447", "right": "issue:713", "accept": false, - "reason": "LoRA loading error vs generic training error; too vague to be the same underlying bug." + "reason": "LoRA-specific error versus generic training error; too little concrete overlap." }, { "left": "issue:3113", "right": "issue:4042", "accept": false, - "reason": "Cross-attention layout control feature request vs combining reference-only with inpainting/depth control; related area but different requested capability." + "reason": "Both involve conditioning/control features, but they are distinct feature requests with different goals." }, { "left": "issue:3564", "right": "issue:6372", "accept": false, - "reason": "Uni-ControlNet support and SVD + ControlNet are different pipeline feature requests." + "reason": "Uni-ControlNet support and SVD+ControlNet are different pipeline features." }, { "left": "issue:4349", "right": "issue:6201", "accept": false, - "reason": "Upscale pipeline missing text_encoder when prompt_embeds are supplied vs stable diffusion prompt-encoding fallback bug; different prompt-handling paths." + "reason": "Both touch prompt_embeds handling, but they affect different pipelines and different logic paths." }, { "left": "issue:4638", "right": "issue:9078", "accept": false, - "reason": "SDXL textual inversion loading issue vs Flux LoRA loading issue; both are loading-related but for different model families and code paths." + "reason": "Textual inversion on SDXL and Flux LoRA loading are unrelated." }, { "left": "issue:5079", "right": "issue:6513", "accept": false, - "reason": "LyCORIS/LoRA loading support request vs T2IAdapter model support request; unrelated features." + "reason": "LoRA loading support versus T2IAdapter model support are separate feature gaps." }, { "left": "issue:7454", "right": "issue:9241", "accept": false, - "reason": "General LoRA support request vs a script to convert XLab LoRA to diffusers LoRA; one is feature support, the other is a conversion tool request." + "reason": "General LoRA support request versus a LoRA conversion script; not the same concrete change." }, { "left": "issue:8588", "right": "issue:8671", "accept": false, - "reason": "Missing SD3 conversion script vs redundant code in a training script; one is a missing file request, the other is code cleanup." + "reason": "Missing SD3 conversion script and redundant DreamBooth code are unrelated." }, { "left": "issue:1478", "right": "issue:5953", "accept": false, - "reason": "Both are pipeline size/output issues, but one is about multiple image outputs in upscaling and the other is a ControlNet image/control_image size mismatch crash." + "reason": "Upscale pipeline output-count bug versus ControlNet image-size mismatch crash." }, { "left": "issue:4704", "right": "issue:7276", "accept": false, - "reason": "Custom diffusion training script failure vs DreamBooth example network errors; different scripts and failure causes." + "reason": "Custom diffusion training failure versus DreamBooth example network errors; different scripts and failure modes." }, { "left": "issue:3963", "right": "issue:4727", "accept": false, - "reason": "Gradio auto-download issue on Windows vs auto pipelines breaking with local_files_only; both relate to loading, but not the same bug." - }, - { - "left": "issue:1033", - "right": "issue:1572", - "accept": false, - "reason": "stable_diffusion_mega IndexError vs img2img/inpainting unsupported for stable_diffusion_mega; different symptoms and likely causes." - }, - { - "left": "issue:4806", - "right": "issue:8373", - "accept": false, - "reason": "UNet2DModel parameter typing bug vs an incorrect error message; unrelated." + "reason": "Auto-download on Windows and local_files_only auto-pipeline handling are different download-related issues." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The conclusions are well grounded in the packet and stay conservative. The 1909\u21942083 pair is clearly the strongest duplicate-like match from the titles, and the other rejected pairs mostly do appear to involve different bugs, workflows, or components. I don\u2019t see any overstatement that would require rejection." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -66343,25 +66365,22 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:1033", "issue:10448", - "issue:10470", "issue:10520", "issue:10798", "issue:10954", "issue:11033", - "issue:11093", "issue:11307", - "issue:11314", "issue:12037", "issue:12052", "issue:12423", "issue:1246", - "issue:12905", "issue:13034", "issue:1315", "issue:13162", @@ -66375,6 +66394,7 @@ "issue:4480", "issue:4506", "issue:4726", + "issue:4806", "issue:5016", "issue:5574", "issue:5685", @@ -66383,6 +66403,7 @@ "issue:668", "issue:7322", "issue:8264", + "issue:8373", "issue:8407", "issue:8495", "issue:8500", @@ -66398,6 +66419,8 @@ "issue:9942" ], "soft_pairs": [ + "issue:1033|issue:1572", + "issue:4806|issue:8373", "issue:10520|issue:8264", "issue:1572|issue:964", "issue:10798|issue:13034", @@ -66419,182 +66442,180 @@ "issue:8500|issue:8535", "issue:8708|issue:9686", "issue:8723|issue:9460", - "issue:10448|issue:10954", - "issue:11314|issue:12905", - "issue:10470|issue:11093" + "issue:10448|issue:10954" ], - "prepared_review_unit_hash": "e3283d60e83964006b3bb3dc6bb1a0659ae7aad2925a1751b7e78fe84ec6b39e", + "prepared_review_unit_hash": "98a1cf8a7810117c20f10f474855b72d7a240e4d6c7909d70e6294384b2a98a5", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33563, - "estimated_input_tokens": 8391, - "estimated_eval_tokens": 17038 + "serialized_chars": 33045, + "estimated_input_tokens": 8262, + "estimated_eval_tokens": 16780 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous collection of Diffusers issues spanning unrelated models, pipelines, schedulers, offloading, and training examples. None of the soft-similarity pairs look like true duplicates of the same concrete bug/change.", - "confidence": 0.96, - "canonical_issue_reason": "No clear canonical issue emerges from the cluster; the items are mostly unrelated. If one must be chosen as the strongest standalone report, issue 12052 is the most concrete and reproducible.", + "summary": "This set is mostly a grab-bag of unrelated issues that only share superficial terms like offloading, loading, device mismatch, or model names. I do not see any true duplicate pairs; all soft links look too weak and cross different code paths or different models.", + "confidence": 0.88, + "canonical_issue_reason": "Issue #12052 is the strongest anchor if one must be chosen: it is concrete, well-scoped, and describes a specific device-mismatch failure in a clearly identified Wan 2.2 + LightX2V offloading path.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 12052 is the best standalone issue candidate because it describes a specific device-mismatch failure in a clearly identified code path (Wan 2.2 + LightX2V offloading) with actionable reproduction details.", + "best_issue_reason": "#12052 is the most actionable and diagnostic issue in the set, with a clear repro context and failure mode; it is a better representative than the more generic or title-only reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1033", + "right": "issue:1572", + "accept": false, + "reason": "Same broad Stable Diffusion Mega area, but different failure reports: an IndexError versus img2img/inpainting not working. Too weak to be the same bug." + }, + { + "left": "issue:4806", + "right": "issue:8373", + "accept": false, + "reason": "One is a type annotation/API issue on UNet2DModel; the other is an incorrect error message. Different bug classes and code paths." + }, { "left": "issue:10520", "right": "issue:8264", "accept": false, - "reason": "Both are runtime problems, but one is a CUDA/CPU device mismatch in Sana 4K and the other is a training process stopping at 0%; different bugs." + "reason": "Runtime device mismatch in Sana 4K is unrelated to a training job stopping at 0%. No clear shared root cause." }, { "left": "issue:1572", "right": "issue:964", "accept": false, - "reason": "Both mention inpainting, but 1572 is about `stable_diffusion_mega` img2img/inpainting support while 964 is a generic inpainting example failure; not the same root cause." + "reason": "Both mention inpainting, but one is a StableDiffusion Mega img2img/inpainting failure and the other is a generic inpainting example issue. Not enough evidence for duplication." }, { "left": "issue:10798", "right": "issue:13034", "accept": false, - "reason": "Flux quantization/device-combination bug vs WanPipeline height/width bug; unrelated models and code paths." + "reason": "Completely different topics: Flux quantization/device combinations versus WanPipeline height/width handling." }, { "left": "issue:12037", "right": "issue:13162", "accept": false, - "reason": "Both involve LoRA compatibility, but they target different Wan/Flux models and different failure modes." + "reason": "Both involve LoRA support, but for different models and different compatibility problems. Not the same concrete bug." }, { "left": "issue:1315", "right": "issue:1375", "accept": false, - "reason": "Both are import regressions, but one is a missing symbol (`PIL_INTERPOLATION`) and the other is a broader package import break from a PR; insufficient evidence they are the same bug." + "reason": "Import error from a missing symbol is not the same as a PR allegedly breaking imports; different symptom and likely different root cause." }, { "left": "issue:11033", "right": "issue:12423", "accept": false, - "reason": "Both are `from_single_file` issues, but they concern different model classes and different loading failures." + "reason": "Both are from_single_file loading issues, but for different models (SD1.5 UNet vs QwenImageTransformer2DModel). Different code paths." }, { "left": "issue:4453", "right": "issue:5574", "accept": false, - "reason": "LoRA conversion/export question vs training a subset of UNet attention parameters; different topics and no shared bug." + "reason": "LoRA conversion for SDXL safetensors is unrelated to training a subset of UNet attention parameters with LoRA." }, { "left": "issue:9902", "right": "issue:9942", "accept": false, - "reason": "Conda version pinning/stuck version vs pip install failure for a dev release; unrelated package-management issues." + "reason": "A conda version/install problem is not the same as a pip install failure for a diffusers dev version." }, { "left": "issue:5685", "right": "issue:8407", "accept": false, - "reason": "Both mention schedulers, but one is a vague scheduler concern and the other is a specific `set_timesteps()` kwargs compatibility bug." + "reason": "Generic scheduler concerns versus a specific UniPC set_timesteps kwargs compatibility bug. Too broad to merge." }, { "left": "issue:4375", "right": "issue:8880", "accept": false, - "reason": "Torch 2 compile crash vs inference speed regression; different symptoms and likely different causes." + "reason": "Torch 2 compile crash during inference and inference slowdown are different symptoms; no concrete same bug path." }, { "left": "issue:4480", "right": "issue:4726", "accept": false, - "reason": "Mac MPS float16 kernel crash vs incorrect `callback` type annotation; unrelated bugs." + "reason": "MPS kernel crash on Mac M1 with fp16 is unrelated to a callback type issue." }, { "left": "issue:5016", "right": "issue:7322", "accept": false, - "reason": "SDXL dreambooth LoRA text-encoder training issue vs training job SIGKILL; different failure modes." + "reason": "Training text encoders in SDXL LoRA is a functionality bug; SIGKILL during training is a resource/crash symptom. Not the same issue." }, { "left": "issue:4141", "right": "issue:4399", "accept": false, - "reason": "ControlNet inpaint/second-controlnet behavior vs PyTorch 2 compile with CPU offload incompatibility; different subsystems and bugs." + "reason": "Second controlnet handling and torch compile with CPU offload are separate problems in the broader pipeline ecosystem." }, { "left": "issue:1246", "right": "issue:668", "accept": false, - "reason": "Both are DreamBooth training errors, but one is FP16 gradient unscale and the other is a Tesla T4 runtime error; not the same bug." + "reason": "Both are dreambooth runtime problems, but one is an FP16 gradient/unscale issue and the other is a Colab Tesla T4 runtime error. Not the same bug." }, { "left": "issue:4506", "right": "issue:6201", "accept": false, - "reason": "Prompt variable overwrite vs wrongly re-encoding the prompt when embeddings are already supplied; related area but different logic errors." + "reason": "Both involve prompt embeddings, but one says a variable is overwritten while the other is about not re-encoding when prompt is absent. Different concrete behaviors." }, { "left": "issue:8495", "right": "issue:8593", "accept": false, - "reason": "Tokenizer deserialization error in SD3 vs BF16 CPU generation noise; different code paths and symptoms." + "reason": "SD3 tokenizer parsing error and SD3 bfloat16 CPU noise are unrelated issues." }, { "left": "issue:11307", "right": "issue:12052", "accept": false, - "reason": "Both are offloading-related, but they involve different models and distinct block-level vs tensor-device mismatch failures." + "reason": "Both mention offloading and device mismatches, but they affect different models and different tensor operations. Not mergeable as one PR-level fix." }, { "left": "issue:1934", "right": "issue:6398", "accept": false, - "reason": "Both mention memory usage, but one is CPU offload VRAM consumption and the other is an unspecified memory leak; too vague to treat as the same bug." + "reason": "Memory usage with cpu_offload and a vague memory leak report are too generic and not clearly the same bug." }, { "left": "issue:8500", "right": "issue:8535", "accept": false, - "reason": "SD3 token-limit question vs SD3ControlNet pipeline API support; unrelated." + "reason": "Token limit questions and enabling xformers memory-efficient attention are unrelated." }, { "left": "issue:8708", "right": "issue:9686", "accept": false, - "reason": "Missing `vae.config.shift_factor` in SD3 dreambooth examples vs SD3 ControlNet pooled projections bug; different issues." + "reason": "Missing SD3 dreambooth config field and a ControlNet pooled_projections bug are different features and code paths." }, { "left": "issue:8723", "right": "issue:9460", "accept": false, - "reason": "Model construction failure from transformer vs adding list assignment support for ControlNet models; related API area but not the same bug." + "reason": "SD3ControlNetModel initialization failure versus supporting a list of controlnets in the pipeline are different change requests." }, { "left": "issue:10448", "right": "issue:10954", "accept": false, - "reason": "DDUF mmap loading feature request vs FLUX LoRA load failure; unrelated." - }, - { - "left": "issue:11314", - "right": "issue:12905", - "accept": false, - "reason": "Flux controlnet training script mistake vs wrong Z_Image guidance-scale implementation; different features and bugs." - }, - { - "left": "issue:10470", - "right": "issue:11093", - "accept": false, - "reason": "TorchAO inference not working vs TorchAO example broken; same general theme, but the available evidence does not show the same concrete code-path problem." + "reason": "DDUF mmap loading and FLUX LoRA loading are unrelated model/loading concerns." } ] }, @@ -66613,19 +66634,23 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10227", + "issue:10470", "issue:10596", "issue:10950", + "issue:11093", "issue:11127", "issue:11221", + "issue:11314", "issue:11555", "issue:11648", "issue:12738", + "issue:12905", "issue:13073", "issue:13104", "issue:13286", @@ -66649,24 +66674,22 @@ "issue:4282", "issue:4312", "issue:525", - "issue:5894", "issue:6146", "issue:6357", "issue:6372", - "issue:6700", "issue:7054", "issue:7079", - "issue:7640", "issue:7960", "issue:8429", "issue:8493", "issue:8894", "issue:9387", "issue:969", - "issue:9690", - "issue:9972" + "issue:9690" ], "soft_pairs": [ + "issue:11314|issue:12905", + "issue:10470|issue:11093", "issue:11555|issue:13286", "issue:12738|issue:13073", "issue:11221|issue:7960", @@ -66688,182 +66711,180 @@ "issue:1439|issue:525", "issue:1805|issue:321", "issue:7054|issue:8493", - "issue:11648|issue:9690", - "issue:6700|issue:7640", - "issue:5894|issue:9972" + "issue:11648|issue:9690" ], - "prepared_review_unit_hash": "7258275fde6be64cb609adcca47bdffc014cdbf1ebd6e9fb9d8b33dc2b2924f2", + "prepared_review_unit_hash": "8e485a89b2dbb69cb53e763e7dde7b517ae2478940327a2aac0735639afd6132", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33077, - "estimated_input_tokens": 8270, - "estimated_eval_tokens": 16796 + "serialized_chars": 32993, + "estimated_input_tokens": 8249, + "estimated_eval_tokens": 16754 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated bugs, feature requests, and docs/training issues across different pipelines and code paths. The soft-similarity pairs overlap only superficially, so none should be merged as duplicates.", - "confidence": 0.97, - "canonical_issue_reason": "No canonical issue: the set does not form one coherent duplicate cluster, and the titles point to distinct problems/features.", + "summary": "The cluster is mostly a loose thematic grouping around Flux/torchao, LoRA, and ZImage issues, but the soft pairs do not look like true duplicates; they describe different failure modes, pipelines, or docs/example problems rather than one shared bug.", + "confidence": 0.74, + "canonical_issue_reason": "issue:10470 is the broadest concrete bug report in the set, centered on Flux/torchao inference failure; the other issues are either narrower variants (LoRA, fp8, guidance, batch size) or clearly different subsystems.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good global dedupe representative here; none of the candidate pairs describe the same underlying defect.", + "best_issue_reason": "issue:10470 is the best representative issue because it captures the widest underlying Flux/torchao inference problem without being tied to a very specific symptom or example cleanup task.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11314", + "right": "issue:12905", + "accept": false, + "reason": "Both are Z-Image/Flux-style issues, but one is a controlnet training-script mistake and the other is a guidance-scale implementation bug; different code paths and fixes." + }, + { + "left": "issue:10470", + "right": "issue:11093", + "accept": false, + "reason": "Both mention torchao/Flux examples, but one is inference not working and the other is a broken example; too vague to treat as the same concrete bug." + }, { "left": "issue:11555", "right": "issue:13286", "accept": false, - "reason": "Different concerns: device_map auto support vs torchao quantization compatibility; same area, but not the same bug." + "reason": "device_map='auto' support is an API/feature request, while torchao>=0.16.0 quantization not supported is a version-compatibility bug; not the same issue." }, { "left": "issue:12738", "right": "issue:13073", "accept": false, - "reason": "Both are ZImage-related, but one is a batch-size assertion and the other is a guidance_scale limitation in a different pipeline." + "reason": "Batch-size assertion failure and guidance_scale>1 support are different runtime conditions and different failures." }, { "left": "issue:11221", "right": "issue:7960", "accept": false, - "reason": "Training autoencoderkl non-convergence is unrelated to SDXL multi-GPU training failure." + "reason": "Non-convergence in autoencoderkl training and SDXL multi-GPU training failure are unrelated training problems." }, { "left": "issue:2646", "right": "issue:2663", "accept": false, - "reason": "Tiled VAE index error and safety checker loading failure are separate code paths and failure modes." + "reason": "Tiled VAE IndexError and safety-checker loading from safetensors are separate components and separate bugs." }, { "left": "issue:6146", "right": "issue:8894", "accept": false, - "reason": "Different training scripts and different errors; not the same underlying problem." + "reason": "Both are training-script complaints, but they concern different scripts and failure modes, not one shared defect." }, { "left": "issue:1918", "right": "issue:2439", "accept": false, - "reason": "DDIM scheduler example issue is unrelated to inpainting pipeline load/run problems." + "reason": "DDIM scheduler behavior for a ddpm example is unrelated to inpainting pipeline loading/running issues." }, { "left": "issue:2581", "right": "issue:4086", "accept": false, - "reason": "Checkpoint conversion support and ControlNet training not working are different tasks and bugs." + "reason": "ControlNet checkpoint conversion and SDXL ControlNet training are different problems with different workflows." }, { "left": "issue:2712", "right": "issue:3694", "accept": false, - "reason": "Missing config key warning is unrelated to file-location errors during finetuning setup." + "reason": "Missing config keys during model init and missing model_index/config files during finetuning are different loading/configuration errors." }, { "left": "issue:272", "right": "issue:4312", "accept": false, - "reason": "Both mention inpainting, but these are separate feature requests for different sampling/guidance methods." + "reason": "K-LMS support for inpainting and a multi-ControlNet pipeline request are distinct feature requests." }, { "left": "issue:6372", "right": "issue:9387", "accept": false, - "reason": "SVD+ControlNet and MimicMotion are different pipelines with no shared defect evidence." + "reason": "SVD+ControlNet and MimicMotion are unrelated pipeline feature requests." }, { "left": "issue:10227", "right": "issue:11555", "accept": false, - "reason": "LoRA feature-size support error is unrelated to device_map auto support for pipelines." + "reason": "LoRA loading into fp8/quantized modules is a concrete model-compatibility error, while device_map='auto' is a pipeline placement feature request." }, { "left": "issue:10950", "right": "issue:8429", "accept": false, - "reason": "Broken docs and a broken inference script are not the same issue." + "reason": "Broken Accelerate docs and a malfunctioning I2VGen-XL inference script are unrelated." }, { "left": "issue:1398", "right": "issue:1519", "accept": false, - "reason": "Both involve Apple Silicon/MPS, but one is model compatibility and the other is batched-image noise." + "reason": "Both involve Apple Silicon/MPS, but one is SD2 support and the other is noisy batch outputs on M2; different bugs." }, { "left": "issue:4031", "right": "issue:6357", "accept": false, - "reason": "Safety-checker local file loading failure is distinct from API-vs-local output differences." + "reason": "Local safety-checker loading failure is not the same as API vs local result differences." }, { "left": "issue:11127", "right": "issue:13104", "accept": false, - "reason": "Single-file FLUX loading problem is unrelated to a missing logger definition." + "reason": "A CivitAI Flux single-file loading problem and an undefined logger error are unrelated issues." }, { "left": "issue:3120", "right": "issue:4282", "accept": false, - "reason": "ControlNet training collapse to same output is not the same as adding a Multi-ControlNet pipeline." + "reason": "Always-same-result ControlNet training and adding a Multi-ControlNet pipeline are different topics." }, { "left": "issue:2500", "right": "issue:969", "accept": false, - "reason": "Meta-tensor copy failures and multi-CUDA accelerate test failures are separate infrastructure issues." + "reason": "Meta tensor copy errors and multi-GPU accelerator test failures are not the same underlying bug." }, { "left": "issue:10596", "right": "issue:7079", "accept": false, - "reason": "Both are shape/model errors, but they affect different modules and no concrete shared bug is shown." + "reason": "FluxControlPipeline shape errors and Transformer2DModel patch/norm_type errors are different model-path failures." }, { "left": "issue:1439", "right": "issue:525", "accept": false, - "reason": "Transformers attribute error and local image loading notebook issue are unrelated." + "reason": "FeatureExtractionMixin import error and local-disk image loading in a notebook are unrelated." }, { "left": "issue:1805", "right": "issue:321", "accept": false, - "reason": "A request for a scalable diffusion pipeline is not the same as blended diffusion." + "reason": "Scalable Diffusion pipeline proposal and Blended Diffusion are different feature ideas." }, { "left": "issue:7054", "right": "issue:8493", "accept": false, - "reason": "State-dict mutation in LoRA loading and SD3Transformer state-dict load errors are different problems." + "reason": "State-dict mutation in LoraLoaderMixin and SD3Transformer2DModel load errors are different loading bugs." }, { "left": "issue:11648", "right": "issue:9690", "accept": false, - "reason": "LoRA loading with fp8 transformer and FLUX FP8 example cleanup are not the same bug." - }, - { - "left": "issue:6700", - "right": "issue:7640", - "accept": false, - "reason": "temb-passing TypeErrors and smaller-dimension RuntimeErrors are separate code-path failures." - }, - { - "left": "issue:5894", - "right": "issue:9972", - "accept": false, - "reason": "Both mention buffer sizes, but the contexts and root causes are too different to treat as duplicates." + "reason": "LoRA weight loading with fp8 transformer models and Flux FP8 example cleanup are not the same bug/change." } ] }, @@ -66882,8 +66903,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -66895,10 +66916,7 @@ "issue:12359", "issue:12367", "issue:13477", - "issue:1363", - "issue:1536", "issue:1643", - "issue:1941", "issue:1976", "issue:2177", "issue:2375", @@ -66918,11 +66936,13 @@ "issue:5334", "issue:5462", "issue:5821", + "issue:5894", "issue:6418", + "issue:6700", "issue:7061", - "issue:7341", "issue:7450", "issue:7490", + "issue:7640", "issue:785", "issue:8803", "issue:8805", @@ -66932,11 +66952,13 @@ "issue:941", "issue:954", "issue:9844", - "issue:9902", "issue:9926", - "issue:9958" + "issue:9958", + "issue:9972" ], "soft_pairs": [ + "issue:6700|issue:7640", + "issue:5894|issue:9972", "issue:2375|issue:785", "issue:2177|issue:954", "issue:2783|issue:5462", @@ -66958,189 +66980,180 @@ "issue:1643|issue:2420", "issue:10150|issue:9844", "issue:1976|issue:3799", - "issue:7061|issue:8876", - "issue:7341|issue:9902", - "issue:1363|issue:1536", - "issue:1941|issue:954" + "issue:7061|issue:8876" ], - "prepared_review_unit_hash": "42bb3943413e71a76d117debf897837d00f002416e6367d675c7d334e8c43db5", + "prepared_review_unit_hash": "58841e656c59ab82b0ff87f6657cd9572473638ea440050b46b95581d6d6e098", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 33953, - "estimated_input_tokens": 8489, - "estimated_eval_tokens": 17234 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33099, + "estimated_input_tokens": 8275, + "estimated_eval_tokens": 16806 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose set of unrelated issue reports across different pipelines, features, and environments. The suggested soft pairs share only broad subsystem similarity and do not look like the same underlying bug or change.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue stands out: the items are heterogeneous, and the soft pairs are not true duplicates.", + "summary": "This cluster is a mixed bag of unrelated issues; the soft pairs mostly share only a broad subsystem or symptom word, not the same underlying bug. I would not merge any of the candidate pairs as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "issue:3672 is the strongest representative issue in this set: it has the highest discussion/inbound-reference activity and describes a concrete, well-scoped StableDiffusionPipeline concurrency failure.", "canonical_pr_reason": null, - "best_issue_reason": "If one issue must serve as the cluster representative, issue:3970 is the most concrete and broadly framed inference bug, but it is only a weak stand-in for this set.", + "best_issue_reason": "issue:3672 is the most substantive and high-signal issue here, so it is the best overall issue representative for the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6700", + "right": "issue:7640", + "accept": false, + "reason": "Different failures: attention-processor temb TypeErrors vs SDXL from_single_file shape/dimension runtime error. Same general area, not the same bug." + }, + { + "left": "issue:5894", + "right": "issue:9972", + "accept": false, + "reason": "Both mention buffer sizes, but 5894 is a generic invalid-buffer error while 9972 is a MacOS-specific huge allocation/request issue. Not the same code-path problem." + }, { "left": "issue:2375", "right": "issue:785", "accept": false, - "reason": "Different failures: a missing safe_open NameError versus CPU core underutilization; no shared code path." + "reason": "safe_open NameError and PNDM scheduler performance are unrelated." }, { "left": "issue:2177", "right": "issue:954", "accept": false, - "reason": "Pretrained AutoencoderKL loading failure is unrelated to an auth/login loading problem." + "reason": "AutoencoderKL loading failure vs login/load issue; no shared bug mechanism." }, { "left": "issue:2783", "right": "issue:5462", "accept": false, - "reason": "Both mention img2img-style workflows, but one is a ControlNet pipeline request and the other is an LCM img2img feature request." + "reason": "Both are img2img-related feature requests, but they target different pipeline additions and are not the same change." }, { "left": "issue:2961", "right": "issue:3455", "accept": false, - "reason": "Training example request versus using pretrained instruct-pix2pix in training; not the same bug or fix." + "reason": "Training-example request for StableUnCLIP vs a separate instruct-pix2pix training request; different features." }, { "left": "issue:3116", "right": "issue:941", "accept": false, - "reason": "Both involve img2img noise, but they target different schedulers/settings and different device behavior." + "reason": "Both report noise in img2img, but one is scheduler/inference-step specific and the other is MPS/num_images_per_prompt specific. Distinct bugs." }, { "left": "issue:4148", "right": "issue:4312", "accept": false, - "reason": "SDXL inpainting config error is not the same as a request to support SAG for inpainting/img2img." + "reason": "SDXL inpainting config error vs SAG support request for inpainting/img2img. Not the same underlying issue." }, { "left": "issue:5283", "right": "issue:6418", "accept": false, - "reason": "Completely different domains: media track merging versus stable video diffusion frame collapse." + "reason": "Completely different topics: media track merging vs stable video diffusion frame collapse." }, { "left": "issue:5334", "right": "issue:8803", "accept": false, - "reason": "Accelerate multi-model training support is unrelated to a duplicated assignment bug in attention processing." + "reason": "Accelerate multi-model training API request vs attention_processor hidden_states duplicate assignment bug. Unrelated." }, { "left": "issue:5821", "right": "issue:7490", "accept": false, - "reason": "Both reference backward/training, but one is a CUDA backward failure and the other is a generic backward problem with no clear match." + "reason": "Both mention backward, but one is a CUDA training failure and the other is a generic backward problem. Not enough evidence of the same bug." }, { "left": "issue:8803", "right": "issue:8805", "accept": false, - "reason": "One is a code bug in attention_processor.py; the other is a broken SDXL conversion script." + "reason": "Attention processor hidden_states bug vs SDXL conversion script failure; different code paths." }, { "left": "issue:1109", "right": "issue:2507", "accept": false, - "reason": "CPU core usage and safetensors GPU load speed are different performance complaints with different causes." + "reason": "CPU core utilization issue vs safetensors GPU load slowness; unrelated." }, { "left": "issue:10085", "right": "issue:10475", "accept": false, - "reason": "LyCORIS batch inference runtime error is unrelated to SD3 fine-tuning/inference quality mismatch." + "reason": "LyCORIS batch-inference/skip_guidance_layers runtime error vs SD3 inference quality comparison during finetuning. Different problem classes." }, { "left": "issue:3672", "right": "issue:7450", "accept": false, - "reason": "Both are threading-related, but they involve different components and symptoms; not enough evidence of the same bug." + "reason": "Both are threading-related, but one is StableDiffusionPipeline concurrency and the other is AutoencoderKL multithreading. Not the same concrete bug." }, { "left": "issue:3970", "right": "issue:4061", "accept": false, - "reason": "SDXL prompt embeddings not working and SDXL LoRA load error are different pipeline issues." + "reason": "SDXL prompt embeddings bug vs SDXL LoRA loading error; same model family, different failure modes." }, { "left": "issue:8970", "right": "issue:8997", "accept": false, - "reason": "Same model family, but black-image-on-MPS and EDMDPMSolverMultistepScheduler usage error are distinct problems." + "reason": "Both involve Kolors, but one is black images on MPS and the other is a scheduler usage error. Distinct issues." }, { "left": "issue:12359", "right": "issue:13477", "accept": false, - "reason": "A documentation bug for guidance_scale is unrelated to an ErnieImage callback locals() scope bug." + "reason": "Documentation bug for Chroma guidance_scale vs ErnieImage callback locals() scope bug; unrelated." }, { "left": "issue:10672", "right": "issue:12367", "accept": false, - "reason": "A request to support callback_on_step_end is not the same as a request for Qwen-Image Eligen LoRA support." + "reason": "Callback-on-step-end support request vs Qwen-Image Eligen LoRA support request; different feature work." }, { "left": "issue:9926", "right": "issue:9958", "accept": false, - "reason": "gguf quantize/speed support and AutoencoderKLTemporalDecoder tiling are different feature requests." + "reason": "gguf quantize/speedup support and AutoencoderKLTemporalDecoder tiling are unrelated feature requests." }, { "left": "issue:1643", "right": "issue:2420", "accept": false, - "reason": "The xformers signature mismatch in DreamBooth is not the same as the xformers issue in LoRA training." + "reason": "Both mention xformers, but one is a DreamBooth method-signature error and the other is a LoRA training example issue. Not the same bug." }, { "left": "issue:10150", "right": "issue:9844", "accept": false, - "reason": "Slow SDXL loading/inference and NaNs from the SDXL VAE encoder are different symptoms and likely different fixes." + "reason": "Slow loading/inference for SDXL pipelines vs NaNs from SDXL VAE encoder; different symptoms and likely different causes." }, { "left": "issue:1976", "right": "issue:3799", "accept": false, - "reason": "Incorrect file lookup for tokenizer/config.json is unrelated to a typing/inference issue in DDPMScheduler.from_pretrained." + "reason": "Incorrect tokenizer file lookup vs incorrect inferred type for scheduler.from_pretrained; unrelated." }, { "left": "issue:7061", "right": "issue:8876", "accept": false, - "reason": "A docstring mistake is not the same as missing resolution microconditioning in the PixArt Sigma pipeline." - }, - { - "left": "issue:7341", - "right": "issue:9902", - "accept": false, - "reason": "TorchVersion/Version comparison error and conda version stuck are unrelated environment/versioning issues." - }, - { - "left": "issue:1363", - "right": "issue:1536", - "accept": false, - "reason": "An ONNX long-prompt/LMS scheduler error does not match custom pipeline incompatibility with a new diffusers version." - }, - { - "left": "issue:1941", - "right": "issue:954", - "accept": false, - "reason": "Unable to load a custom UnCLIP pipeline is not the same as failing to load after login." + "reason": "Both are pipeline/doc/config style issues, but they concern different pipelines and different concrete fixes." } ] }, @@ -67159,8 +67172,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -67180,15 +67193,14 @@ "issue:12620", "issue:12905", "issue:13073", + "issue:1363", + "issue:1536", "issue:1672", - "issue:1704", - "issue:1992", + "issue:1941", "issue:2560", "issue:2988", - "issue:3176", "issue:3391", "issue:3452", - "issue:3887", "issue:3908", "issue:4159", "issue:4520", @@ -67198,22 +67210,26 @@ "issue:5465", "issue:590", "issue:6091", - "issue:6359", "issue:6906", + "issue:7341", "issue:7428", "issue:7722", "issue:7773", "issue:8237", "issue:8317", "issue:8589", - "issue:8637", "issue:8708", "issue:9477", "issue:9495", + "issue:954", "issue:967", + "issue:9902", "issue:9936" ], "soft_pairs": [ + "issue:7341|issue:9902", + "issue:1363|issue:1536", + "issue:1941|issue:954", "issue:10194|issue:12367", "issue:1672|issue:6091", "issue:3452|issue:6906", @@ -67234,183 +67250,180 @@ "issue:1082|issue:4839", "issue:11000|issue:12620", "issue:11486|issue:8589", - "issue:1249|issue:967", - "issue:1704|issue:1992", - "issue:3176|issue:6359", - "issue:3887|issue:8637" + "issue:1249|issue:967" ], - "prepared_review_unit_hash": "42a60f7f9299122106f4e9254ddc657d5b7a3f97e1101a94937eae279172097f", + "prepared_review_unit_hash": "ca75944e1c39f868de19073887a41b7c3b78c6858ba411d041630c1d76655bbe", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33807, - "estimated_input_tokens": 8452, - "estimated_eval_tokens": 17160 + "serialized_chars": 33870, + "estimated_input_tokens": 8468, + "estimated_eval_tokens": 17192 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items do not form a single duplicate cluster; they span unrelated feature requests and bugs across training, pipelines, schedulers, LoRA, video, and tooling. All soft-edge pairs look like false positives from superficial text overlap.", - "confidence": 0.96, + "summary": "This cluster is a set of largely unrelated issue reports; none of the soft pairs look like true duplicates or the same fix/change. Most are only loosely related by broad themes like training scripts, schedulers, or guidance-scale support.", + "confidence": 0.95, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:7341", + "right": "issue:9902", + "accept": false, + "reason": "Different problems: a TorchVersion/Version comparison TypeError vs a Conda version pinning/stuck-version report." + }, + { + "left": "issue:1363", + "right": "issue:1536", + "accept": false, + "reason": "Both involve custom/compatibility issues, but one is ONNX long-prompt weight + LMS scheduler, the other is lpw_stable_diffusion compatibility with newer diffusers." + }, + { + "left": "issue:1941", + "right": "issue:954", + "accept": false, + "reason": "Unrelated failures: custom pipeline loading for UnCLIP vs a login/load-after-login issue." + }, { "left": "issue:10194", "right": "issue:12367", "accept": false, - "reason": "Different requests in different areas: Flux attention masking vs Qwen-Image Eligen LoRA support." + "reason": "Different feature requests for different models/pipelines: Flux attention masking vs Qwen-Image Eligen LoRA support." }, { "left": "issue:1672", "right": "issue:6091", "accept": false, - "reason": "One is an image-size regression after ckpt conversion; the other is a consistency decoder memory issue." + "reason": "Both mention size, but one is image size after conversion and the other is a consistency decoder memory issue with a size argument." }, { "left": "issue:3452", "right": "issue:6906", "accept": false, - "reason": "Both mention DreamBooth/LoRA, but one is about low-VRAM training and the other says the LoRA example is broken; not the same bug." + "reason": "DreamBooth/training-related, but one is a low-VRAM question and the other is a broken LoRA DreamBooth example." }, { "left": "issue:8317", "right": "issue:8708", "accept": false, - "reason": "Both involve VAE-related training examples, but one is a missing generator argument in decode() and the other is missing shift_factor in SD3 examples." + "reason": "Both concern SD3/VAE details, but one is a missing generator arg in AsymmetricAutoencoderKL decode and the other is missing shift_factor in training examples." }, { "left": "issue:4159", "right": "issue:5465", "accept": false, - "reason": "Different pipelines and failure modes: VRAM blowup in DiffEdit vs inpaint_strength behavior in DiffEdit." + "reason": "Different DiffEdit bugs: excessive VRAM usage vs inpaint_strength only working at 0.8." }, { "left": "issue:8237", "right": "issue:9477", "accept": false, - "reason": "Different LoRA/load issues: missing key loading vs a GatheredParameters callable error." + "reason": "Different issues: SDXL LoRA key loading vs a GatheredParameters callable error." }, { "left": "issue:4520", "right": "issue:5028", "accept": false, - "reason": "Both mention LoRA scale, but one is an Attention Slicing incompatibility and the other is a specific reference pipeline forward signature error." + "reason": "Both mention scale/LoRA, but the failing code paths are different: attention slicing with lora_scale vs SDXL reference pipeline hacked block forward args." }, { "left": "issue:7722", "right": "issue:9495", "accept": false, - "reason": "Different combined-feature bugs: Animatediff+ControlNet+IPAdapter vs SDXL PAG+IPAdapter." + "reason": "Both involve IPAdapter, but they are different pipelines and likely different bugs." }, { "left": "issue:12540", "right": "issue:13073", "accept": false, - "reason": "Same model family (ZImage/Sana), but one is a pipeline compatibility issue and the other is a specific guidance_scale limitation." + "reason": "Both are guidance-related, but one is SanaPipeline embeddings support and the other is ZImageControlNetPipeline guidance_scale behavior." }, { "left": "issue:11973", "right": "issue:9936", "accept": false, - "reason": "Training script argument parsing error vs multi-GPU NCCL timeout; unrelated symptoms and code paths." + "reason": "Different training issues: missing instance_prompt arg vs NCCL timeout in multi-GPU controlnet training." }, { "left": "issue:2988", "right": "issue:590", "accept": false, - "reason": "Scheduler crash at a step-count edge case vs a typo in the Karras noise formula." + "reason": "Both scheduler-related, but one is a crash at equal train/inference steps and the other is a Karras noise formula typo." }, { "left": "issue:3391", "right": "issue:4557", "accept": false, - "reason": "DeepFloyd overexposure bug is unrelated to the __getattr__ AttributeError." + "reason": "Completely unrelated: overexposure in DeepFloyd super-resolution vs a __getattr__ AttributeError." }, { "left": "issue:2560", "right": "issue:3908", "accept": false, - "reason": "Different problems: generator-list nondeterminism in img2img vs Streamlit model caching." + "reason": "Different domains: deterministic outputs with generator lists vs Streamlit model caching." }, { "left": "issue:10128", "right": "issue:7428", "accept": false, - "reason": "A support question about fastercache is unrelated to a general question about layer diffusion support." + "reason": "Both are feature requests, but for unrelated features: fastercache vs layer diffusion." }, { "left": "issue:10144", "right": "issue:7773", "accept": false, - "reason": "Mochi output quality question is unrelated to Stable Video Diffusion batch-size behavior." + "reason": "Different video-generation quality issues in different models/pipelines." }, { "left": "issue:10591", "right": "issue:12905", "accept": false, - "reason": "Different SD3/Z-image topics: LoRA training script issue vs wrong guidance scale implementation." + "reason": "Both are 'wrong' behavior reports, but they concern different components: SD3 LoRA training script vs Z_Image guidance scale implementation." }, { "left": "issue:10653", "right": "issue:11833", "accept": false, - "reason": "Different device-management bugs: 8bit bitsandbytes .to(cpu) unsupported vs KeyError in set_lora_device." + "reason": "Different device-management bugs: bitsandbytes 8bit .to(cpu) limitation vs set_lora_device KeyError." }, { "left": "issue:1082", "right": "issue:4839", "accept": false, - "reason": "Dreambooth CUDA matmul failure is not the same as a type error around lora_scale." + "reason": "Unrelated errors: DreamBooth CUDA matmul failure vs a lora_scale type error." }, { "left": "issue:11000", "right": "issue:12620", "accept": false, - "reason": "Pipeline argument mismatch vs missing release assets; completely unrelated." + "reason": "No overlap in bug/change: missing shift argument vs missing release assets." }, { "left": "issue:11486", "right": "issue:8589", "accept": false, - "reason": "Training-script option request vs SD3 DreamBooth prior-preservation bug; not the same change." + "reason": "Different training concerns: enabling xformers memory-efficient attention vs SD3 prior preservation support." }, { "left": "issue:1249", "right": "issue:967", "accept": false, - "reason": "Different compatibility bugs in different pipelines: ONNX ndarray device error vs ONNX scheduler operand/type issue." - }, - { - "left": "issue:1704", - "right": "issue:1992", - "accept": false, - "reason": "A vague attention-block question is unrelated to JAX attention refactoring." - }, - { - "left": "issue:3176", - "right": "issue:6359", - "accept": false, - "reason": "TensorRT demo error and EulerDiscreteScheduler missing get_velocity are different subsystems and failure modes." - }, - { - "left": "issue:3887", - "right": "issue:8637", - "accept": false, - "reason": "SD pipeline output difference is unrelated to an import error for PixArtSigmaPipeline." + "reason": "Both are ONNX/scheduler related, but the reported failures and likely fixes are different." } ] }, @@ -67429,8 +67442,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -67440,19 +67453,19 @@ "issue:12633", "issue:1274", "issue:13243", - "issue:1574", + "issue:1704", "issue:1730", "issue:1896", - "issue:2068", + "issue:1992", "issue:2306", "issue:2775", - "issue:2784", + "issue:3176", "issue:3363", "issue:3420", "issue:3584", - "issue:3591", "issue:3765", "issue:3769", + "issue:3887", "issue:3915", "issue:4480", "issue:4542", @@ -67463,8 +67476,6 @@ "issue:4964", "issue:5226", "issue:5390", - "issue:5412", - "issue:5480", "issue:5546", "issue:5570", "issue:5628", @@ -67472,6 +67483,7 @@ "issue:5989", "issue:6068", "issue:6234", + "issue:6359", "issue:6529", "issue:6548", "issue:6593", @@ -67480,10 +67492,14 @@ "issue:7490", "issue:7665", "issue:7836", + "issue:8637", "issue:8994", "issue:9933" ], "soft_pairs": [ + "issue:1704|issue:1992", + "issue:3176|issue:6359", + "issue:3887|issue:8637", "issue:4542|issue:6548", "issue:4680|issue:6529", "issue:4702|issue:7368", @@ -67504,183 +67520,180 @@ "issue:3420|issue:3915", "issue:7490|issue:7665", "issue:12633|issue:13243", - "issue:4480|issue:8994", - "issue:1574|issue:5412", - "issue:2068|issue:3591", - "issue:2784|issue:5480" + "issue:4480|issue:8994" ], - "prepared_review_unit_hash": "942fa351799e8d89d98bc863d97a22b067201f9492b97a882c4b0b0228fe2703", + "prepared_review_unit_hash": "be6a7961ce506ec8bbe7b6ee40e2b4236b37bbd0cebd534752b6d5abd86a65ff", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33767, - "estimated_input_tokens": 8442, - "estimated_eval_tokens": 17140 + "serialized_chars": 33721, + "estimated_input_tokens": 8431, + "estimated_eval_tokens": 17118 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster does not form a single duplicate set: the items span unrelated documentation, training, pipeline, and scheduler bugs. All soft edges are too broad or only share a subsystem, not the same concrete issue.", - "confidence": 0.92, - "canonical_issue_reason": "No safe canonical issue for duplicate triage: the set is heterogeneous and the apparent similarities are mostly broad topic overlap, not the same underlying bug.", - "canonical_pr_reason": "No pull requests are present in this cluster.", - "best_issue_reason": "issue:12633 is the strongest standalone report here: it names a precise scheduler mismatch, the affected API, and the triggering setting (`timestep_spacing='linspace'`), making it the most actionable issue even though it is not a duplicate anchor.", + "summary": "The soft pairs are mostly broad thematic matches rather than true duplicates; I would reject every edge. The strongest standalone issue in this set is the SDXL/DDIM sampling bug, which is the most concrete and reproducible report.", + "confidence": 0.86, + "canonical_issue_reason": "issue:6068 is the most technically precise and self-contained bug report: it names the exact scheduler, the SDXL code path, and the incorrect-sampling symptom with substantial discussion.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6068 is the best representative issue because it is specific, reproducible, and directly tied to a single code path rather than a vague feature/question.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1704", + "right": "issue:1992", + "accept": false, + "reason": "Both mention attention, but one is a general question and the other is a JAX refactor topic; not the same bug/change." + }, + { + "left": "issue:3176", + "right": "issue:6359", + "accept": false, + "reason": "TensorRT demo failure and a missing scheduler method are unrelated code paths." + }, + { + "left": "issue:3887", + "right": "issue:8637", + "accept": false, + "reason": "A quality/effect complaint about SD pipelines is unrelated to an import error for PixArtSigmaPipeline." + }, { "left": "issue:4542", "right": "issue:6548", "accept": false, - "reason": "Both involve img2img behavior, but one is about multi-image output limits and the other about RGBA inputs being misread as latents; different bugs." + "reason": "One is about SDXL img2img generating only one image, the other is an RGBA latent-handling bug; different failure modes." }, { "left": "issue:4680", "right": "issue:6529", "accept": false, - "reason": "SDXL inpainting/fine-tuning request vs Fooocus inpaint topic; same general area, not the same code-path problem." + "reason": "General SDXL inpainting fine-tuning and a Fooocus inpaint request are not the same concrete issue." }, { "left": "issue:4702", "right": "issue:7368", "accept": false, - "reason": "A VQ-VAE example script request and a GLIGEN training question are unrelated changes." + "reason": "Training a VQ-VAE example and asking about GLIGEN training are different model/workflow requests." }, { "left": "issue:4841", "right": "issue:6234", "accept": false, - "reason": "ControlNet image-size issue vs a text-to-image example script issue; no shared concrete defect." + "reason": "ControlNet image-size trouble and a text-to-image SDXL training script issue do not share the same bug." }, { "left": "issue:4964", "right": "issue:5390", "accept": false, - "reason": "Model conversion to diffusers and unwanted text-encoder download with `local_files_only=True` are distinct problems." + "reason": "Model conversion to Diffusers and unwanted text-encoder downloads with local_files_only are unrelated." }, { "left": "issue:5226", "right": "issue:6593", "accept": false, - "reason": "Both mention `load_lora_weights`, but one is an adapter pipeline incompatibility and the other is an img2img feature request; not the same bug." + "reason": "Both touch LoRA support, but they target different pipeline classes and different implementation paths." }, { "left": "issue:5689", "right": "issue:6068", "accept": false, - "reason": "Different scheduler/sample-quality defects: residual noise in DPM++ vs incorrect DDIM samples on SDXL." + "reason": "Both are scheduler-quality complaints, but DPM++ residual noise and DDIM incorrect samples are distinct scheduler bugs." }, { "left": "issue:5989", "right": "issue:7836", "accept": false, - "reason": "A `safetensors.torch` bug and CLIP overfitting in a training example are unrelated." + "reason": "A safetensors.torch bug and CLIP example overfitting are unrelated." }, { "left": "issue:6598", "right": "issue:9933", "accept": false, - "reason": "Different img2img issues: nondeterministic results vs missing width/height parameters in SD3 img2img." + "reason": "Img2img parity differences and missing width/height parameters are different pipeline issues." }, { "left": "issue:3363", "right": "issue:3584", "accept": false, - "reason": "Both are DreamBooth training failures, but they refer to different scripts/environments and likely different root causes." + "reason": "Both involve DreamBooth, but one is a LoRA training failure and the other is a tutorial failure; not the same bug." }, { "left": "issue:2775", "right": "issue:5570", "accept": false, - "reason": "Torch Dynamo optimization error vs missing `compute_snr` symbol; unrelated." + "reason": "A torch._dynamo optimization error and a missing compute_snr utility are unrelated." }, { "left": "issue:5546", "right": "issue:5628", "accept": false, - "reason": "Both concern Karras sigma handling, but one is step-count mismatch in ensemble usage and the other is denoising-range offset in LMS; not the same concrete bug." + "reason": "Both involve Karras sigmas, but one is wrong step counts in base+refiner while the other is denoising range offset in LMS; different code paths." }, { "left": "issue:1274", "right": "issue:1896", "accept": false, - "reason": "`num_images_per_prompt` failure in SD pipelines vs a CLIP-guided Stable Diffusion shape/type error; unrelated." + "reason": "A num_images_per_prompt failure and a CLIP-guided Stable Diffusion dict-size error are unrelated." }, { "left": "issue:10231", "right": "issue:10287", "accept": false, - "reason": "Both are documentation/example complaints, but one is an Ubuntu import error in a tutorial and the other is a generic HF docs issue; too weak to merge." + "reason": "Both are documentation/example complaints, but the first is a specific Ubuntu import error and the second is a generic docs issue." }, { "left": "issue:2306", "right": "issue:4747", "accept": false, - "reason": "INT_MAX error and missing `model_index.json` after loading a fine-tuned SDXL LoRA are different failure modes." + "reason": "An INT_MAX runtime error and a missing model_index.json load error are different problems." }, { "left": "issue:1185", "right": "issue:1730", "accept": false, - "reason": "DreamBooth reproducibility and converting DreamBooth to CKPT for webui are separate concerns." + "reason": "DreamBooth reproducibility and CKPT conversion/webui incompatibility are different topics." }, { "left": "issue:3765", "right": "issue:3769", "accept": false, - "reason": "Both describe image artifacts, but one is ControlNet training stacking effects and the other is panorama quartering; not the same underlying bug." + "reason": "Both mention image artifacts during training/inference, but controlnet stacking and panorama quartering are distinct issues." }, { "left": "issue:3420", "right": "issue:3915", "accept": false, - "reason": "Different areas and symptoms: ControlNet inpaint pipeline compatibility vs `torch.compile` on text-to-video pipelines." + "reason": "ControlNet inpaint pipeline incompatibility and torch.compile text-to-video failure are unrelated." }, { "left": "issue:7490", "right": "issue:7665", "accept": false, - "reason": "The titles are too generic to establish the same defect; no shared code path or symptom details." + "reason": "A generic backward problem and an instruct-pix2pix issue are too vague and distinct to treat as the same bug." }, { "left": "issue:12633", "right": "issue:13243", "accept": false, - "reason": "Both are scheduler bugs, but one is DDIM timestep-list mismatch and the other is FlowMatch sigma-shift duplication; different implementations and fixes." + "reason": "Both are scheduler timestep bugs, but one is DDIM linspace stepping and the other is FlowMatchEuler sigma shifting; different implementations." }, { "left": "issue:4480", "right": "issue:8994", "accept": false, - "reason": "MPS float16 kernel crash vs generation steps stuck at 0; unrelated." - }, - { - "left": "issue:1574", - "right": "issue:5412", - "accept": false, - "reason": "Both concern DreamBooth training, but one is a bare-metal 8GB GPU failure and the other is checkpoint resume failure in SDXL LoRA training." - }, - { - "left": "issue:2068", - "right": "issue:3591", - "accept": false, - "reason": "A dtype conversion bug in `from_pretrained` and confusing ControlNet loss curves are unrelated." - }, - { - "left": "issue:2784", - "right": "issue:5480", - "accept": false, - "reason": "Separate feature requests (DreamBooth for text2video vs EDM implementation), not duplicates." + "reason": "An MPS float16 kernel crash and generation steps stuck at 0 are not the same failure." } ] }, @@ -67712,18 +67725,18 @@ "issue:11700", "issue:13311", "issue:13547", + "issue:1574", + "issue:2068", "issue:2213", - "issue:2614", "issue:2772", + "issue:2784", "issue:2891", "issue:2985", - "issue:3019", "issue:3040", - "issue:3113", "issue:3381", "issue:3584", + "issue:3591", "issue:3954", - "issue:4314", "issue:4353", "issue:4354", "issue:4471", @@ -67731,7 +67744,8 @@ "issue:4880", "issue:4913", "issue:5128", - "issue:5205", + "issue:5412", + "issue:5480", "issue:5567", "issue:5619", "issue:5714", @@ -67739,7 +67753,6 @@ "issue:5925", "issue:5931", "issue:6201", - "issue:6616", "issue:692", "issue:7661", "issue:7922", @@ -67754,6 +67767,9 @@ "issue:9610" ], "soft_pairs": [ + "issue:1574|issue:5412", + "issue:2068|issue:3591", + "issue:2784|issue:5480", "issue:11581|issue:13547", "issue:2985|issue:4913", "issue:4471|issue:5714", @@ -67774,183 +67790,180 @@ "issue:4354|issue:7946", "issue:7922|issue:8495", "issue:1082|issue:3584", - "issue:1161|issue:692", - "issue:2614|issue:4314", - "issue:3019|issue:5205", - "issue:3113|issue:6616" + "issue:1161|issue:692" ], - "prepared_review_unit_hash": "88d3837b25dd8e31336eec6970550b579911db0875cafb83d9c5b3e1d68cbd2d", + "prepared_review_unit_hash": "51c84a21d239bc3b2f9ac0797910c78b2714d9e3fa6a77dc3ff6aff6ede848d1", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34019, - "estimated_input_tokens": 8505, - "estimated_eval_tokens": 17266 + "serialized_chars": 34041, + "estimated_input_tokens": 8511, + "estimated_eval_tokens": 17278 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The items do not form a single duplicate cluster; the proposed soft-edge pairs are mostly unrelated bugs or feature requests that only share broad subsystem words. I would reject all soft-edge links and avoid naming a canonical issue/PR for this set.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "This set is highly heterogeneous: it mixes DreamBooth training failures, LoRA/model-loading bugs, pipeline feature requests, and unrelated scheduler/dtype/docs issues. There is no true duplicate cluster here, so I would not merge any of the soft-edge pairs.", + "confidence": 0.86, + "canonical_issue_reason": "No single issue cleanly represents the whole set; if forced to pick an umbrella, issue 11659 is the broadest still-open LoRA-loading report, but it does not subsume the many unrelated training, TPU, scheduler, and docs bugs.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 11659 is the closest global representative because it is generic, open, and centered on LoRA loading, which is the only theme that recurs in a few nearby reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1574", + "right": "issue:5412", + "accept": false, + "reason": "Both are DreamBooth-related, but they describe different failures in different scripts/situations (8GB GPU training instability vs resume_from_checkpoint in SDXL LoRA)." + }, + { + "left": "issue:2068", + "right": "issue:3591", + "accept": false, + "reason": "One is a dtype-conversion bug in from_pretrained; the other is a ControlNet loss-curve question. Different problems entirely." + }, + { + "left": "issue:2784", + "right": "issue:5480", + "accept": false, + "reason": "These are separate feature requests: DreamBooth text2video support vs implementing EDM." + }, { "left": "issue:11581", "right": "issue:13547", "accept": false, - "reason": "Both mention LoRA/FLUX loading, but one is a CLIPVisionModel dtype constructor error and the other is a missing Flux2Transformer2DModel LoRA-key match; different failure points." + "reason": "Both involve FLUX/LoRA loading, but one is a CLIPVisionModel dtype constructor error and the other is a Flux2 adapter key mismatch. Not the same code-path bug." }, { "left": "issue:2985", "right": "issue:4913", "accept": false, - "reason": "Kandinsky 2.1 and controlnet-openpose-sdxl-1.0 are unrelated models/features, not the same bug." + "reason": "Kandinsky 2.1 release discussion vs a specific ControlNet OpenPose SDXL model request; unrelated." }, { "left": "issue:4471", "right": "issue:5714", "accept": false, - "reason": "One is SDXL Img2Img LoRA weight loading; the other is a PEFT model support warning path. Different code paths and symptoms." + "reason": "SDXL img2img LoRA weight loading and maybe_raise_or_warn(PeftModel) are different defects in different parts of the stack." }, { "left": "issue:5905", "right": "issue:5925", "accept": false, - "reason": "Separate feature requests: diffusion-GAN model support vs ZipLora multi-LoRA inference." + "reason": "Both are model/inference feature requests, but for different features (diffusion-GAN vs ZipLoRA)." }, { "left": "issue:3040", "right": "issue:4880", "accept": false, - "reason": "The first is a docs/copy-paste AttributeError about list objects; the second is a ControlNet input validation error. Not the same issue." + "reason": "One is a docs/example AttributeError about a list object; the other is ControlNet single-image input validation. Different issues." }, { "left": "issue:11700", "right": "issue:9610", "accept": false, - "reason": "Pruna integration for from_pretrained/pipeline is unrelated to adding load_lora_weights in an img2img pipeline." + "reason": "Pruna integration for loading models and adding load_lora_weights to SD3 img2img are separate enhancements." }, { "left": "issue:5619", "right": "issue:6201", "accept": false, - "reason": "Both concern pipeline encode logic, but one is AutoencodeTiny/LCM img2img image encoding and the other is prompt_embeds handling when prompt is absent; not the same concrete bug." + "reason": "AutoencodeTiny encoding failure and prompt_embeds handling in StableDiffusion are different pipeline bugs." }, { "left": "issue:5567", "right": "issue:9331", "accept": false, - "reason": "Unet.config being overwritten by accelerator.prepare is unrelated to FlowMatchEulerDiscreteScheduler index_for_timestep behavior." + "reason": "accelerator.prepare overwriting unet.config is unrelated to FlowMatchEulerDiscreteScheduler index handling." }, { "left": "issue:10311", "right": "issue:8742", "accept": false, - "reason": "DreamBooth import error in train script is unrelated to a removed LoRAAttnProcessor2_0 class compatibility break." + "reason": "Import error in train_dreambooth.py and removal of LoRAAttnProcessor2_0 are not the same bug or change." }, { "left": "issue:7661", "right": "issue:7974", "accept": false, - "reason": "Both involve IP-Adapter, but one is about enabling animate diff pipeline construction and the other is an inpaint bug; different failures." + "reason": "Both mention IP-Adapter, but one is about creating an AnimateDiff pipeline with IP-Adapter-loaded UNet and the other is an inpaint bug. Too different to merge." }, { "left": "issue:8537", "right": "issue:9224", "accept": false, - "reason": "Different dtype/platform incompatibilities: torch.float8_e4m3fn missing vs hardcoded float64 on MPS." + "reason": "Both are dtype/platform compatibility issues, but one is missing float8 on torch and the other is float64 unsupported on MPS. Distinct failures." }, { "left": "issue:2891", "right": "issue:3381", "accept": false, - "reason": "Unsupported embeds in StableDiffusionUpscalePipeline docs vs discoloured batch inpainting output are unrelated." + "reason": "StableDiffusionUpscalePipeline embeds support/docs mismatch and inpaint discoloration are unrelated." }, { "left": "issue:2772", "right": "issue:5128", "accept": false, - "reason": "ControlNet local dataset handling and XLPipeline.from_single_file safetensors loading are different problems." + "reason": "ControlNet local dataset validation and XLPipeline.from_single_file safetensors loading are different pipeline issues." }, { "left": "issue:3954", "right": "issue:4753", "accept": false, - "reason": "Learning-rate scheduler step count in multi-GPU training is unrelated to a train_dreambooth_lora.py bug." + "reason": "Learning-rate scheduler step counting and train_dreambooth_lora.py failures are separate training bugs." }, { "left": "issue:4353", "right": "issue:5931", "accept": false, - "reason": "Poor SDXL image quality is not the same as StableDiffusionPipeline non-determinism." + "reason": "Poor SDXL image quality and non-deterministic StableDiffusionPipeline outputs are different symptom classes." }, { "left": "issue:11659", "right": "issue:13311", "accept": false, - "reason": "General LoRA load issue and a specific Flux2Transformer2DModel prefix/key mismatch are not the same bug." + "reason": "These are related by LoRA loading, but one is a generic load issue and the other is a specific Flux2 Transformer key-prefix mismatch; not the same concrete bug." }, { "left": "issue:2213", "right": "issue:898", "accept": false, - "reason": "TPU hang with no distributed environment is unrelated to unsupported BFloat16 on TPU." + "reason": "TPU hang in a training script vs unsupported BFloat16 on TPU are distinct TPU problems." }, { "left": "issue:4354", "right": "issue:7946", "accept": false, - "reason": "Wrong pipeline type from from_single_file is unrelated to custom_pipeline caching." + "reason": "Wrong pipeline type from_single_file and custom_pipeline caching are different loading/caching bugs." }, { "left": "issue:7922", "right": "issue:8495", "accept": false, - "reason": "Missing ORTStableDiffusionXLPipeline attribute and SD3 tokenizer enum parse failure are unrelated load errors." + "reason": "Missing ORTStableDiffusionXLPipeline attribute and SD3 tokenizer enum parse failure are unrelated." }, { "left": "issue:1082", "right": "issue:3584", "accept": false, - "reason": "DreamBooth CUDA execution failure is not the same as a failed DreamBooth tutorial; too broad and symptom-level only." + "reason": "DreamBooth CUDA execution failure and a failed DreamBooth tutorial are not the same underlying bug." }, { "left": "issue:1161", "right": "issue:692", "accept": false, - "reason": "Converting Flax to PyTorch is unrelated to asking for a fine-tuning example." - }, - { - "left": "issue:2614", - "right": "issue:4314", - "accept": false, - "reason": "Changing pretrained-model download directory and interrupted-download write protection are separate file/download behaviors." - }, - { - "left": "issue:3019", - "right": "issue:5205", - "accept": false, - "reason": "Segment-anything text-guided editing request and Repaint+ControlNet+LoRA request are different feature asks." - }, - { - "left": "issue:3113", - "right": "issue:6616", - "accept": false, - "reason": "Training-free layout control with cross-attention guidance and classifier-guidance implementation are distinct guidance methods." + "reason": "Converting Flax to PyTorch and asking for a fine-tuning example are unrelated." } ] }, @@ -67982,23 +67995,23 @@ "issue:11885", "issue:1207", "issue:12175", - "issue:12334", "issue:12501", "issue:13067", "issue:1367", + "issue:2614", "issue:2669", - "issue:2763", "issue:2799", "issue:2905", + "issue:3019", "issue:3045", + "issue:3113", "issue:3174", - "issue:3218", "issue:3233", "issue:3699", "issue:3728", "issue:3960", + "issue:4314", "issue:4578", - "issue:487", "issue:5009", "issue:5026", "issue:5205", @@ -68009,21 +68022,23 @@ "issue:590", "issue:6112", "issue:6232", + "issue:6616", "issue:7185", "issue:725", "issue:7370", "issue:7389", "issue:7600", "issue:7841", - "issue:801", "issue:8759", "issue:917", "issue:938", "issue:9858", - "issue:9946", "issue:9953" ], "soft_pairs": [ + "issue:2614|issue:4314", + "issue:3019|issue:5205", + "issue:3113|issue:6616", "issue:3699|issue:5763", "issue:3728|issue:7370", "issue:4578|issue:6112", @@ -68044,193 +68059,187 @@ "issue:2905|issue:725", "issue:10800|issue:8759", "issue:3045|issue:3233", - "issue:12501|issue:9953", - "issue:12334|issue:9946", - "issue:2763|issue:3218", - "issue:487|issue:801" + "issue:12501|issue:9953" ], - "prepared_review_unit_hash": "b404da7f0fc0f6fa8aeb5e402901ebebf10b636576c3cad077e735e24118b85c", + "prepared_review_unit_hash": "e34fccbbd156bbd56ada0e37d26a3a51fefaf80a28fb288edcd080b0efce31d8", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33687, - "estimated_input_tokens": 8422, - "estimated_eval_tokens": 17100 + "serialized_chars": 33046, + "estimated_input_tokens": 8262, + "estimated_eval_tokens": 16780 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly heterogeneous. Only the two xformers-memory-efficient-attention issues look like the same underlying bug; the rest are different models, features, or unrelated error types.", - "confidence": 0.83, - "canonical_issue_reason": "issue:12334 is the clearer, more descriptive report of the xformers memory-efficient attention bug, so it is the better canonical issue than the shorthand duplicate.", + "summary": "The set is largely heterogeneous; most pairs are only loosely related by subsystem or model family, not the same underlying bug/change. No soft edge is strong enough to merge on the evidence shown.", + "confidence": 0.68, + "canonical_issue_reason": "issue:11885 is the best representative because it is the most concrete, API-specific bug report (`load_lora_weights` on FLUX.1) and the closest to any other overlapping report in the set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:12334 is the best representative issue for the only convincing duplicate pair in this cluster; its title is explicit and user-facing.", + "best_issue_reason": "issue:11885 has the clearest actionable failure description and the most likely duplicate-target value among the listed issues; the rest are broader feature requests or unrelated bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2614", + "right": "issue:4314", + "accept": false, + "reason": "Both involve downloads, but one is about changing the download directory and the other about interrupted-download write protection; different bugs." + }, + { + "left": "issue:3019", + "right": "issue:5205", + "accept": false, + "reason": "Both are image-editing-related feature requests, but they target different capabilities and code paths." + }, + { + "left": "issue:3113", + "right": "issue:6616", + "accept": false, + "reason": "Both mention guidance, but one is layout control and the other classifier guidance implementation; not the same change." + }, { "left": "issue:3699", "right": "issue:5763", "accept": false, - "reason": "Different failures: CUDA/runtime install error vs. safetensors save behavior." + "reason": "Completely different problems: CUDA on macOS versus safetensors serialization." }, { "left": "issue:3728", "right": "issue:7370", "accept": false, - "reason": "Different requests for different pipelines/models; no shared bug." + "reason": "Both are new-model/pipeline support requests, but for different architectures and tasks." }, { "left": "issue:4578", "right": "issue:6112", "accept": false, - "reason": "Doc/VAE usage issue vs. image quality problem in SDXL-turbo." + "reason": "Doc wording about VAE usage is unrelated to an SDXL-turbo image quality complaint." }, { "left": "issue:5205", "right": "issue:938", "accept": false, - "reason": "Different domains and bugs: Stable Diffusion inpainting/ControlNet vs. text-to-motion model." + "reason": "Different application areas: Stable Diffusion repaint/control versus text-to-motion generation." }, { "left": "issue:5334", "right": "issue:7600", "accept": false, - "reason": "Related training infrastructure, but not the same concrete bug path." + "reason": "Related to training utilities, but one is an Accelerate API enhancement and the other a DreamBooth script misuse bug." }, { "left": "issue:2799", "right": "issue:917", "accept": false, - "reason": "Checkpoint/pipeline mismatch error is unrelated to ONNX inpainting failure." + "reason": "Mismatch-checkpoint validation and ONNX inpainting are unrelated." }, { "left": "issue:11651", "right": "issue:11885", "accept": false, - "reason": "Both involve FLUX LoRA loading, but one is specific to FluxControl and the other is a broader load_lora_weights bug; not enough to treat as the same bug." + "reason": "Both concern FLUX LoRA loading, but one is FluxControl-specific and the other is a broader `load_lora_weights` bug; not enough evidence they are the same concrete failure." }, { "left": "issue:3174", "right": "issue:3960", "accept": false, - "reason": "Different LoRA loading/model-loading issues; one is a feature regression, the other a header-size error." + "reason": "One is a LoRA/DreamBooth loading regression and the other is a header-size error; different symptoms and likely different causes." }, { "left": "issue:5578", "right": "issue:7841", "accept": false, - "reason": "Missing model_index.json error is unrelated to the SDXL red-dot image quality report." + "reason": "A download interruption write-protection issue is unrelated to the SDXL viewing artifact." }, { "left": "issue:5710", "right": "issue:6232", "accept": false, - "reason": "Same scheduler class, but different bugs in initialization vs. add_noise behavior." + "reason": "Same scheduler class, but one is missing `sigmas` at initialization and the other is an `add_noise` malfunction under some settings; too weak to treat as the same bug." }, { "left": "issue:11569", "right": "issue:13067", "accept": false, - "reason": "Feature request for a repo vs. support request for a different model; not duplicates." + "reason": "A feature request for a specific repo/model is not the same as a general request to add support for Anima." }, { "left": "issue:1367", "right": "issue:590", "accept": false, - "reason": "Custom trained_betas config save bug is unrelated to Karras noise formula typo." + "reason": "Scheduler config saving with custom betas and a Karras formula typo are distinct issues." }, { "left": "issue:5009", "right": "issue:7389", "accept": false, - "reason": "Wuerstchen error and SDXL from_single_file error are different pipeline issues." + "reason": "Wuerstchen runtime error and SDXL single-file loading error are separate model/pipeline failures." }, { "left": "issue:5026", "right": "issue:7185", "accept": false, - "reason": "KeyError on text_encoder is unrelated to pytorch version support." + "reason": "A KeyError on `text_encoder` is unrelated to a question about PyTorch 2.2 support." }, { "left": "issue:1207", "right": "issue:2669", "accept": false, - "reason": "Both are AttributeError reports, but they point to different root causes and code paths." + "reason": "CUDA runtime symbol resolution and a `NoneType.replace` AttributeError are unrelated failures." }, { "left": "issue:10204", "right": "issue:12175", "accept": false, - "reason": "Different models and symptoms: FluxFill multi-iteration degradation vs. Qwen Image quality." + "reason": "Both mention poor output quality, but for different models/pipelines and without evidence of the same root cause." }, { "left": "issue:11489", "right": "issue:9858", "accept": false, - "reason": "Flux LoRA training error vs. KeyError: 'train' are not the same bug." + "reason": "Both are training-related, but one is a Flux LoRA training error and the other is a generic `KeyError: 'train'`; not the same bug." }, { "left": "issue:2905", "right": "issue:725", "accept": false, - "reason": "Scheduler feature request vs. pipeline device-map/accelerate loading feature; unrelated." + "reason": "Scheduler Karras-sigma support and DiffusionPipeline device-map/accelerate memory reduction are unrelated enhancements." }, { "left": "issue:10800", "right": "issue:8759", "accept": false, - "reason": "Quantization/offload type mismatch vs. invalid uint8 cast warning in image processing." + "reason": "Quantization/offload dtype mismatch and SD3 image-processing cast warnings are different code paths." }, { "left": "issue:3045", "right": "issue:3233", "accept": false, - "reason": "Opposite directions of LoRA conversion; not the same change or bug." + "reason": "These are inverse conversion requests (LoRA to diffusers vs LoRA to safetensors), not the same task or bug." }, { "left": "issue:12501", "right": "issue:9953", "accept": false, - "reason": "Torch graph break in VAE decoder vs. quantized component CUDA move error." - }, - { - "left": "issue:12334", - "right": "issue:9946", - "accept": true, - "reason": "Both report the same xformers memory-efficient attention bug; one is the generic bug report and the other is the same feature entry." - }, - { - "left": "issue:2763", - "right": "issue:3218", - "accept": false, - "reason": "Both touch ControlNet/torch.compile performance, but the failures are different (VRAM usage vs. graph breaks)." - }, - { - "left": "issue:487", - "right": "issue:801", - "accept": false, - "reason": "M1 CPU incompatibility is unrelated to attention slicing not being respected." + "reason": "A VAE graph-break issue and a quantized-component CUDA move error are unrelated." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is well grounded in the packet and the soft-edge calls are conservative overall. The only accepted pair, issue:12334 vs issue:9946, is plausibly the same xformers memory-efficient attention bug, and the rest are reasonably treated as non-duplicates based on the titles. No overclaiming stands out." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -68251,6 +68260,7 @@ "issue:10869", "issue:11307", "issue:11436", + "issue:12334", "issue:12752", "issue:1284", "issue:13430", @@ -68258,7 +68268,9 @@ "issue:1925", "issue:2392", "issue:2495", + "issue:2763", "issue:2908", + "issue:3218", "issue:3451", "issue:3461", "issue:3583", @@ -68270,21 +68282,17 @@ "issue:4264", "issue:4305", "issue:4349", - "issue:4480", + "issue:487", "issue:5108", - "issue:5368", "issue:5371", - "issue:5557", "issue:6242", "issue:6392", "issue:6393", "issue:6790", - "issue:7197", "issue:7199", "issue:7333", - "issue:7438", "issue:7509", - "issue:7574", + "issue:801", "issue:8274", "issue:8410", "issue:8485", @@ -68294,9 +68302,13 @@ "issue:9101", "issue:967", "issue:9792", - "issue:9895" + "issue:9895", + "issue:9946" ], "soft_pairs": [ + "issue:12334|issue:9946", + "issue:2763|issue:3218", + "issue:487|issue:801", "issue:11307|issue:11436", "issue:2495|issue:4349", "issue:3583|issue:5371", @@ -68317,190 +68329,190 @@ "issue:8535|issue:8539", "issue:12752|issue:13430", "issue:3974|issue:8410", - "issue:4086|issue:5108", - "issue:7197|issue:7574", - "issue:4480|issue:7438", - "issue:5368|issue:5557" + "issue:4086|issue:5108" ], - "prepared_review_unit_hash": "23fde56d912e80f32c99fbf8a41ae3d71e08a72a539ac6930532bc3c01d27a06", + "prepared_review_unit_hash": "e093339d571fded8bf65ae361087f11dc971f2f28d72a44fc2ba82e5c301c02f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33969, - "estimated_input_tokens": 8493, - "estimated_eval_tokens": 17242 + "serialized_chars": 33946, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated runtime errors, feature requests, training issues, and research-paper pipeline proposals. None of the soft pairs look like the same underlying bug or change, so there is no defensible duplicate merge point.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "Mostly unrelated issue set; the only plausible duplicate is the xformers memory-efficient-attention report pair. All other soft edges look like different bugs or feature requests in nearby subsystems.", + "confidence": 0.91, + "canonical_issue_reason": "issue:9946 is the cleanest, most direct title for the xformers attention problem and reads like the broader canonical report versus the more symptom-focused issue:12334.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:9946 is the best representative issue because it names the exact API/problem area and is the broader, more canonical phrasing for the only near-duplicate pair.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12334", + "right": "issue:9946", + "accept": true, + "reason": "Same xformers attention-enable API and same general bug area; these look like duplicate reports of the same problem." + }, + { + "left": "issue:2763", + "right": "issue:3218", + "accept": false, + "reason": "Both touch ControlNet, but one is VRAM growth with cudnn.benchmark and the other is torch.compile graph breaks; different failure modes." + }, + { + "left": "issue:487", + "right": "issue:801", + "accept": false, + "reason": "Both involve image pipelines, but M1 CPU failure and attention slicing not being honored are different bugs." + }, { "left": "issue:11307", "right": "issue:11436", "accept": false, - "reason": "Same project area, but one is block-level offloading and the other is a dtype mismatch; different root causes." + "reason": "HiDream group offloading at block level is not the same as the Half/BFloat16 dtype mismatch during inference." }, { "left": "issue:2495", "right": "issue:4349", "accept": false, - "reason": "Both mention upscale pipelines, but black-image-on-MPS and missing-text-encoder with prompt_embeds are distinct failures." + "reason": "Both are upscale-pipeline issues, but black output on MPS and missing text_encoder despite prompt_embeds are different code-path problems." }, { "left": "issue:3583", "right": "issue:5371", "accept": false, - "reason": "One is Mixture of Diffusers support, the other BLIP inpainting support; different features." + "reason": "Unrelated feature requests: mixture-of-diffusers support vs BLIP inpainting pipe support." }, { "left": "issue:4069", "right": "issue:4263", "accept": false, - "reason": "Textual inversion unloading and LoRA unloading are different adapter lifecycles and bugs." + "reason": "Both are unloading-related, but textual inversion unloading and LoRA unloading are different mechanisms and bugs." }, { "left": "issue:4213", "right": "issue:6393", "accept": false, - "reason": "Both involve from_single_file, but one is an accelerate dependency issue and the other is private-link download failure." + "reason": "Both mention from_single_file, but one is missing accelerate and the other is private HF link download failure; different root causes." }, { "left": "issue:4264", "right": "issue:8708", "accept": false, - "reason": "Missing example images and missing SD3 VAE config fields are unrelated." + "reason": "Missing example images and missing vae.config.shift_factor are unrelated documentation/training issues." }, { "left": "issue:6242", "right": "issue:9792", "accept": false, - "reason": "A request for dataset logs is not the same as a dataset caching/mapping error." + "reason": "Dataset log request vs cache-loading failure when mapping a dataset are not the same issue." }, { "left": "issue:6392", "right": "issue:6790", "accept": false, - "reason": "Custom LoRA loading failure and an LCM training state_dict mismatch are different code paths." + "reason": "Custom LoRA loading failure and unexpected state_dict key in LCM training are different problems." }, { "left": "issue:7333", "right": "issue:7509", "accept": false, - "reason": "Different research pipelines/papers; no concrete shared bug." + "reason": "Both are video/inversion papers, but they target different algorithms and pipelines." }, { "left": "issue:8485", "right": "issue:9101", "accept": false, - "reason": "SD3 LoRA support and SD3 InstructPix2Pix training are separate feature requests." + "reason": "Support for SD3 LoRA and an InstructPix2Pix training script for SD3 are separate feature requests." }, { "left": "issue:3451", "right": "issue:3461", "accept": false, - "reason": "Both are performance/memory complaints, but they concern different scripts and symptoms." + "reason": "Both concern memory/performance, but text-to-image fp16 memory usage and Dreambooth throughput/VRAM are different reports." }, { "left": "issue:1802", "right": "issue:967", "accept": false, - "reason": "UNet encoder_hidden_states handling and ONNX scheduler compatibility are unrelated bugs." + "reason": "UNet encoder_hidden_states handling and ONNX scheduler incompatibility are unrelated issues." }, { "left": "issue:1925", "right": "issue:2908", "accept": false, - "reason": "Stable Diffusion 2 inpainting fp16 issues and ControlNet training shape/mixed-precision errors are different problems." + "reason": "Stable-diffusion inpainting fp16 problems and ControlNet training shape/mixed-precision errors are different bugs." }, { "left": "issue:1033", "right": "issue:2392", "accept": false, - "reason": "Stable-diffusion-mega IndexError and img2img prompt_embeds validation are separate pipeline bugs." + "reason": "IndexError in stable_diffusion_mega and img2img prompt_embeds validation failure are unrelated." }, { "left": "issue:7199", "right": "issue:8274", "accept": false, - "reason": "These are different research pipelines with no shared implementation bug." + "reason": "Different research pipelines: ViewFusion multiview consistency vs FIFO-Diffusion infinite video generation." }, { "left": "issue:1284", "right": "issue:4305", "accept": false, - "reason": "Batch-size slowdown and torch.compile graph breaks are not the same issue." + "reason": "Batch-size slowdown and torch.compile graph breaks are unrelated performance vs compiler issues." }, { "left": "issue:10869", "right": "issue:9895", "accept": false, - "reason": "Meta-tensor .item() failure and FluxPipeline CPU/CUDA device mismatch are distinct runtime errors." + "reason": "Meta-tensor .item() failure and Flux device-placement mismatch are different runtime errors." }, { "left": "issue:8535", "right": "issue:8539", "accept": false, - "reason": "Both are SD3-related, but xformers attention enablement and cross-attention setting failures are different bugs." + "reason": "Both are SD3-related, but xformers attention enablement and cross-attention configuration are not the same bug." }, { "left": "issue:12752", "right": "issue:13430", "accept": false, - "reason": "Z-Image pipeline support and JoyAI-Image-Edit support are different model requests." + "reason": "Z-Image img2img/inpainting pipeline request and JoyAI-Image-Edit support are different model support requests." }, { "left": "issue:3974", "right": "issue:8410", "accept": false, - "reason": "ControlNetInpaint safetensors support and native TextDiffuser support are unrelated feature requests." + "reason": "ControlNet inpaint .safetensors support and TextDiffuser native pipeline support are unrelated feature requests." }, { "left": "issue:4086", "right": "issue:5108", "accept": false, - "reason": "SDXL ControlNet training failure and T2I Multiadapter XL exceptions are different model-path issues." - }, - { - "left": "issue:7197", - "right": "issue:7574", - "accept": false, - "reason": "Both are KeyErrors, but they refer to different missing keys in different training/loading scenarios." - }, - { - "left": "issue:4480", - "right": "issue:7438", - "accept": false, - "reason": "MPS fp16 kernel crash and XPU generation stall are different backend-specific failures." - }, - { - "left": "issue:5368", - "right": "issue:5557", - "accept": false, - "reason": "Mixed-precision dtype mismatch and unexpected use_reentrant argument are unrelated errors." + "reason": "SDXL ControlNet training failure and T2I Multiadapter XL exceptions are different pipeline/training problems." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative: the xformers pair (12334/9946) is the only clearly plausible duplicate, and the other pairs are reasonably treated as different bugs or feature requests. No overstatement detected." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -68517,17 +68529,14 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10470", "issue:10658", "issue:10767", "issue:10967", "issue:11307", "issue:11437", "issue:11687", - "issue:12157", "issue:12501", "issue:12610", - "issue:12645", "issue:1407", "issue:1704", "issue:2370", @@ -68545,28 +68554,33 @@ "issue:3418", "issue:3454", "issue:3698", + "issue:4480", "issue:460", "issue:5043", "issue:5062", "issue:5368", "issue:5395", + "issue:5557", "issue:5997", "issue:6250", "issue:6321", "issue:6858", "issue:6879", "issue:6958", + "issue:7197", + "issue:7438", + "issue:7574", "issue:8744", "issue:8837", - "issue:8880", "issue:905", "issue:9138", "issue:9227", - "issue:9291", - "issue:9361", "issue:9558" ], "soft_pairs": [ + "issue:7197|issue:7574", + "issue:4480|issue:7438", + "issue:5368|issue:5557", "issue:5997|issue:8744", "issue:12501|issue:12610", "issue:2370|issue:2780", @@ -68588,189 +68602,186 @@ "issue:3406|issue:6958", "issue:3418|issue:5043", "issue:3698|issue:905", - "issue:6858|issue:9138", - "issue:12157|issue:12645", - "issue:9291|issue:9361", - "issue:10470|issue:8880" + "issue:6858|issue:9138" ], - "prepared_review_unit_hash": "76ec59708c47ecea42fd6ff8bd6a3032ccb717d512c2d617d29c949d90daeda2", + "prepared_review_unit_hash": "5bc90661fdb1ad888b9839d3cd22dd37a7c03bce93242d3a0ff81538cce3b391", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 33933, - "estimated_input_tokens": 8484, - "estimated_eval_tokens": 17224 + "serialized_chars": 33344, + "estimated_input_tokens": 8336, + "estimated_eval_tokens": 16928 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is heterogeneous: most items are distinct bugs or feature requests across unrelated areas (training scripts, VAE/checkpointing, Flux/LoRA, ONNX, docs). None of the soft pairs appears to be the same underlying issue/change.", - "confidence": 0.91, - "canonical_issue_reason": null, + "summary": "This is not a true duplicate cluster: the items span many unrelated bugs and feature requests across training, checkpointing, attention/dtype errors, ONNX, and docs. The soft pairs are mostly superficial keyword matches rather than the same underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": "No canonical issue fits the cluster well; the issues are too heterogeneous, so picking one representative would be misleading.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No single issue is a good global representative because the cluster lacks a common underlying bug or feature change.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:7197", + "right": "issue:7574", + "accept": false, + "reason": "Both mention checkpoint-like errors, but the failing keys and scenarios differ; not the same bug." + }, + { + "left": "issue:4480", + "right": "issue:7438", + "accept": false, + "reason": "MPS kernel crash on Mac and XPU generation stall are different platform/runtime problems." + }, + { + "left": "issue:5368", + "right": "issue:5557", + "accept": false, + "reason": "One is a mixed-precision dtype mismatch; the other is an unexpected keyword argument error." + }, { "left": "issue:5997", "right": "issue:8744", "accept": false, - "reason": "Both involve model/tooling setup, but one is an SDXL tutorial error and the other is about loading single-file models for training; different failures and code paths." + "reason": "ONNX tutorial model-type mismatch and single-file training support are unrelated." }, { "left": "issue:12501", "right": "issue:12610", "accept": false, - "reason": "Graph-break behavior in VAE decoder is unrelated to a device-mismatch error caused by Quanto + group offload." + "reason": "Graph break in VAE decoder vs device mismatch from quanto/group offload are different code paths." }, { "left": "issue:2370", "right": "issue:2780", "accept": false, - "reason": "General LoRA/GPU-capacity question vs. a specific FP16 unscale runtime error; not the same bug." + "reason": "A GPU memory/capacity question and an FP16 unscale error are not the same issue." }, { "left": "issue:2783", "right": "issue:3698", "accept": false, - "reason": "An img2img/inpainting controlnet feature request and converting an inpainting model to diffusers are related topic-wise, but not the same concrete change." + "reason": "Both touch inpainting, but one requests a pipeline feature and the other model conversion; not one bug/change." }, { "left": "issue:2646", "right": "issue:6250", "accept": false, - "reason": "Both mention VAE, but one is a tiled-VAE IndexError and the other is an SDXL single-file loading KeyError; different root causes." + "reason": "Tiled VAE indexing failure and SDXL VAE single-file KeyError are distinct loading/runtime bugs." }, { "left": "issue:10767", "right": "issue:10967", "accept": false, - "reason": "Different model integrations (FlashVideo vs LaVie/Cinemo); not the same feature or bug." + "reason": "Both are model-integration requests, but for different models and different tasks." }, { "left": "issue:2616", "right": "issue:2843", "accept": false, - "reason": "A save_attn_procs NoneType error is unrelated to multi-GPU checkpoint saving." + "reason": "NoneType in save_attn_procs and multi-GPU checkpoint saving errors are different failure modes." }, { "left": "issue:3379", "right": "issue:6321", "accept": false, - "reason": "Intermediate checkpoint loading in Dreambooth is unrelated to CLIP detection during model loading." + "reason": "Checkpoint loading failure and missing CLIP detection are unrelated." }, { "left": "issue:5368", "right": "issue:6879", "accept": false, - "reason": "Both are dtype-related, but one is an attention Q/K/V mismatch under mixed precision and the other is a Custom Diffusion matmul dtype mismatch; too different to merge." + "reason": "Both are dtype-related, but they fail in different subsystems with different triggers." }, { "left": "issue:11307", "right": "issue:9227", "accept": false, - "reason": "HiDream block-level group offloading and FLUX Dreambooth training are different code paths and errors." + "reason": "Group offloading device mismatch and FLUX dreambooth training issues are different problems." }, { "left": "issue:2644", "right": "issue:3454", "accept": false, - "reason": "Training on two 12GB GPUs vs. a script failing on latest versions are not the same issue." + "reason": "A general training capability question is not the same as a specific script regression." }, { "left": "issue:5395", "right": "issue:8837", "accept": false, - "reason": "Combining LoRAs with PEFT and a SlicedAttnProcessor bug are unrelated." + "reason": "LoRA combination bug and sliced attention processor bug do not share the same code path." }, { "left": "issue:10658", "right": "issue:11687", "accept": false, - "reason": "Adding provider_options to InferenceSession is unrelated to a docs request for pruna." + "reason": "ONNX provider options and a docs request for pruna are unrelated." }, { "left": "issue:11437", "right": "issue:9558", "accept": false, - "reason": "Missing outputs during training and a black validation image in TensorBoard are different symptoms and likely different causes." + "reason": "Missing validation outputs and black tensorboard validation images are different output/visualization issues." }, { "left": "issue:1407", "right": "issue:460", "accept": false, - "reason": "Both are ONNX-related, but one is an SD 2.0 pipeline feature_extractor error and the other is a CUDA/CPU onnxruntime requirement issue." + "reason": "Feature_extractor regression in an ONNX pipeline and CUDA-vs-CPU onnxruntime selection are different issues." }, { "left": "issue:1704", "right": "issue:5062", "accept": false, - "reason": "A general question about attention blocks is not the same as a processor typo fix." + "reason": "A general question about attention blocks and a processor typo fix are not duplicates." }, { "left": "issue:2661", "right": "issue:3279", "accept": false, - "reason": "Checkpoint write failures are distinct from resume-from-checkpoint errors." + "reason": "Checkpoint write errors and resume-from-checkpoint errors are related only loosely; not the same bug." }, { "left": "issue:2662", "right": "issue:3279", "accept": false, - "reason": "A generic checkpoint write error is not the same as checkpoint resume failure." + "reason": "Same as above: saving write error versus resume failure are distinct." }, { "left": "issue:3406", "right": "issue:6958", "accept": false, - "reason": "Different attention configuration issues in different modules; no clear shared concrete bug." + "reason": "only_cross_attention behavior and use_linear_projection in a motion model are separate attention issues." }, { "left": "issue:3418", "right": "issue:5043", "accept": false, - "reason": "Adding text-encoder support to a LoRA script and adding LoRA finetuning for Wurstchen are separate feature requests." + "reason": "Adding text-encoder LoRA support and adding W\u00fcrstchen LoRA fine-tuning are different feature requests." }, { "left": "issue:3698", "right": "issue:905", "accept": false, - "reason": "An inpainting model conversion request is not the same as asking for an img2img/inpainting pipeline." + "reason": "Inpainting model conversion and a community image2image/inpainting pipeline request are related only broadly, not the same concrete change." }, { "left": "issue:6858", "right": "issue:9138", "accept": false, - "reason": "ControlNet image conversion error vs. a generic int/string comparison TypeError; unrelated." - }, - { - "left": "issue:12157", - "right": "issue:12645", - "accept": false, - "reason": "Training-time quantization config iteration error is unrelated to inference speed complaints." - }, - { - "left": "issue:9291", - "right": "issue:9361", - "accept": false, - "reason": "Both are FLUX/LoRA-related, but one is key compatibility for Kohya-ss weights and the other is LoRA having no visible effect at inference; related area, not clearly the same bug." - }, - { - "left": "issue:10470", - "right": "issue:8880", - "accept": false, - "reason": "Flux torchao inference failing and a broad inference-speed regression are different problems." + "reason": "A .convert attribute error in controlnet training and an int-vs-str comparison type error are unrelated." } ] }, @@ -68796,17 +68807,19 @@ "nodes": [ "issue:10003", "issue:10194", + "issue:10470", "issue:10520", "issue:1064", "issue:10869", "issue:11104", "issue:11144", + "issue:12157", + "issue:12645", "issue:12656", "issue:1543", "issue:2052", "issue:2068", "issue:2310", - "issue:2392", "issue:2661", "issue:2662", "issue:2772", @@ -68823,27 +68836,28 @@ "issue:4975", "issue:5000", "issue:5062", - "issue:5325", "issue:5395", - "issue:5824", "issue:5884", - "issue:6176", "issue:7136", - "issue:7300", "issue:756", "issue:7695", - "issue:7735", "issue:775", "issue:7773", "issue:8469", "issue:8594", "issue:8644", "issue:8850", + "issue:8880", "issue:9109", "issue:929", + "issue:9291", + "issue:9361", "issue:9599" ], "soft_pairs": [ + "issue:12157|issue:12645", + "issue:9291|issue:9361", + "issue:10470|issue:8880", "issue:4229|issue:8594", "issue:11144|issue:9109", "issue:3350|issue:4727", @@ -68865,189 +68879,186 @@ "issue:11104|issue:8644", "issue:2772|issue:2786", "issue:10003|issue:12656", - "issue:5000|issue:5395", - "issue:5824|issue:7735", - "issue:2392|issue:5325", - "issue:6176|issue:7300" + "issue:5000|issue:5395" ], - "prepared_review_unit_hash": "3e230292c804964b72406866f7e1c0823ada2192536f45b22a1c8330e144d7be", + "prepared_review_unit_hash": "066217b66d8afeb9b42573212255184c88d10b6538bf126e3cf7fd5c90d0d23b", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34080, - "estimated_input_tokens": 8520, - "estimated_eval_tokens": 17296 + "serialized_chars": 33956, + "estimated_input_tokens": 8489, + "estimated_eval_tokens": 17234 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a false-positive mix of unrelated bug reports and feature requests across different pipelines, models, and error modes. No pair looks like the same underlying issue, so there is no strong duplicate canonical beyond a reference anchor.", + "summary": "These items are not duplicates: they cover many unrelated models, pipelines, and failure modes. None of the soft pairs look safe to merge into one bug/change, so all are rejected.", "confidence": 0.95, - "canonical_issue_reason": "Issue 11144 is the most detailed open report, so it is the best anchor if one must be chosen, but it does not actually subsume the rest of this heterogeneous cluster.", + "canonical_issue_reason": "issue:10470 is the strongest representative if one must be chosen: it is a concrete, self-contained bug report with clear repro context and follow-up discussion; the rest are too heterogeneous to serve as a shared canonical duplicate.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 11144 is the most substantive and specific open issue in the set, making it the most useful reference point despite not being a true duplicate hub.", + "best_issue_reason": "issue:10470 is the best standalone issue in this set because it is specific, reproducible, and well-scoped, unlike the other reports which mostly describe different subsystems or feature requests.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:12157", + "right": "issue:12645", + "accept": false, + "reason": "Quantization-config training error vs inference-speed complaint; different failure modes and code paths." + }, + { + "left": "issue:9291", + "right": "issue:9361", + "accept": false, + "reason": "Both mention FLUX LoRA, but one is key compatibility and the other is missing LoRA effect at inference; not the same bug." + }, + { + "left": "issue:10470", + "right": "issue:8880", + "accept": false, + "reason": "torchao inference breakage vs a later-version speed regression; correctness bug and performance regression are distinct." + }, { "left": "issue:4229", "right": "issue:8594", "accept": false, - "reason": "Different failure modes and code paths: multi-GPU accelerated training error vs UnboundLocalError in a specific DreamBooth LoRA script." + "reason": "Different scripts and errors: multi-GPU accelerate failure vs UnboundLocalError in train_dreambooth_lora_sd3.py." }, { "left": "issue:11144", "right": "issue:9109", "accept": false, - "reason": "Unrelated problems: Flax dtype initialization vs missing torchvision::nms operator." + "reason": "Flax UNet dtype initialization issue is unrelated to missing torchvision::nms operator." }, { "left": "issue:3350", "right": "issue:4727", "accept": false, - "reason": "Animov-512x error report vs auto-pipeline local_files_only mirror handling; not the same bug." + "reason": "Animov model errors vs local_files_only auto-pipeline loading; no shared underlying defect." }, { "left": "issue:10520", "right": "issue:10869", "accept": false, - "reason": "Both are runtime errors, but one is cuda/cpu device mismatch and the other is meta-tensor .item() misuse; distinct causes." + "reason": "Both are device-related, but one is cuda/cpu mismatch in Sana and the other is meta-tensor .item() failure." }, { "left": "issue:2310", "right": "issue:3112", "accept": false, - "reason": "LDM inpainting/LaMA cleaner issue vs a multi-ControlNet inpainting feature request; different scope." + "reason": "Inpainting cleaner issue vs multi-ControlNet inpainting feature request; not the same problem." }, { "left": "issue:8469", "right": "issue:8850", "accept": false, - "reason": "Corrupt IP-Adapter face outputs vs broken community pipeline mirrors; unrelated." + "reason": "Broken faces in IP-Adapter output vs community pipelines mirror broken are unrelated issues." }, { "left": "issue:1543", "right": "issue:775", "accept": false, - "reason": "Both mention fp16, but one is a revision/loading problem and the other is autocast/half-precision behavior; not the same defect." + "reason": "fp16/revision loading problem vs general fp16 without autocast issue; overlapping topic, but different concrete bugs." }, { "left": "issue:1064", "right": "issue:3951", "accept": false, - "reason": "Scheduler add_noise error vs UniDiffuser NaNs; different components and symptoms." + "reason": "Scheduler add_noise error vs NaNs in demo code; different symptoms and mechanisms." }, { "left": "issue:2786", "right": "issue:3348", "accept": false, - "reason": "NumPy input compatibility bug vs LoRA scale support with xformers; unrelated changes." + "reason": "numpy/transformers incompatibility in Depth2Img vs LoRA scale support with xformers; unrelated." }, { "left": "issue:2068", "right": "issue:4975", "accept": false, - "reason": "Model dtype conversion correctness bug vs LoRA loading inefficiency; same theme but different problem." + "reason": "One is incorrect dtype conversion behavior, the other is repeated datatype-query inefficiency; related area, different bug." }, { "left": "issue:2052", "right": "issue:2661", "accept": false, - "reason": "Checkpoint resume failure vs checkpoint save write error; different directions of the workflow." + "reason": "Checkpoint resume failure vs checkpoint write error; distinct checkpointing problems." }, { "left": "issue:2052", "right": "issue:2662", "accept": false, - "reason": "Resume-from-checkpoint bug vs checkpoint write error; not the same underlying issue." + "reason": "Same as above: resume-from-checkpoint and save/write failure are different issues." }, { "left": "issue:10194", "right": "issue:9599", "accept": false, - "reason": "Flux attention masking feature request vs FLUX LoRA-only finetune example request; separate requests." + "reason": "Flux attention masking support vs LoRA-only finetune example request; different feature requests." }, { "left": "issue:3984", "right": "issue:756", "accept": false, - "reason": "Class labels validation error vs a deprecation-tuple error; different validation paths." + "reason": "class_labels requirement vs deprecation-tuple error; unrelated API failures." }, { "left": "issue:4403", "right": "issue:7695", "accept": false, - "reason": "Half-precision LayerNorm kernel missing vs ip_adapter_masks being ignored; different code paths." + "reason": "Half-precision LayerNorm kernel missing vs ignored ip_adapter_masks; different code paths." }, { "left": "issue:4841", "right": "issue:7773", "accept": false, - "reason": "ControlNet image-size issue vs Stable Video Diffusion batch-size bug; not the same bug." + "reason": "ControlNet image-size bug vs Stable Video Diffusion batch-size generation bug; not the same underlying defect." }, { "left": "issue:5062", "right": "issue:929", "accept": false, - "reason": "Attention typo fix vs an attention math question; no shared defect." + "reason": "Typo in attention processor vs redundant sqrt calculation; both in attention code, but not the same bug." }, { "left": "issue:5884", "right": "issue:7136", "accept": false, - "reason": "IP-Adapter support across many pipelines vs callback_on_step_end for LDM3D; separate feature work." + "reason": "IP-Adapter support expansion vs callback_on_step_end for LDM3D; separate feature requests." }, { "left": "issue:11104", "right": "issue:8644", "accept": false, - "reason": "Both involve offload/meta tensors, but they affect different models and failure points." + "reason": "Both involve offload/meta-tensor problems, but they affect different models and loading flows." }, { "left": "issue:2772", "right": "issue:2786", "accept": false, - "reason": "Local dataset handling in train_controlnet.py vs NumPy input compatibility in a pipeline; unrelated." + "reason": "Local dataset validation in ControlNet training vs numpy/transformers incompatibility in Depth2Img; unrelated." }, { "left": "issue:10003", "right": "issue:12656", "accept": false, - "reason": "Scheduler bug at 1000 inference steps vs AuraFlow positional embedding OOB; completely different issues." + "reason": "DDIMScheduler 1000-step bug vs AuraFlow positional embedding OOB; different models and failures." }, { "left": "issue:5000", "right": "issue:5395", "accept": false, - "reason": "Broken LoRA unfusing vs combining civitai LoRAs with PEFT; related area but not the same concrete bug." - }, - { - "left": "issue:5824", - "right": "issue:7735", - "accept": false, - "reason": "SDXL add_time_ids default dimension bug vs missing caption column in training data; unrelated." - }, - { - "left": "issue:2392", - "right": "issue:5325", - "accept": false, - "reason": "Img2img prompt_embeds prompt validation error vs packaging type information distribution; no overlap." - }, - { - "left": "issue:6176", - "right": "issue:7300", - "accept": false, - "reason": "DreamBooth LoRA load error vs high GPU memory after set_default_attn_processor(); different symptoms and code paths." + "reason": "Broken unfuse-after-fuse LoRA regression vs combining Civitai LoRAs with PEFT; same broad subsystem, but different concrete bugs." } ] }, @@ -69066,24 +69077,23 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10025", - "issue:10303", "issue:10412", "issue:10675", "issue:1076", "issue:11135", "issue:11148", - "issue:1305", "issue:1398", "issue:1524", "issue:1642", "issue:1734", "issue:1866", + "issue:2392", "issue:2984", "issue:3089", "issue:319", @@ -69100,10 +69110,11 @@ "issue:4557", "issue:4665", "issue:4753", - "issue:5046", "issue:5054", + "issue:5325", "issue:5510", "issue:5798", + "issue:5824", "issue:6079", "issue:6452", "issue:6486", @@ -69111,8 +69122,8 @@ "issue:6783", "issue:6967", "issue:7370", - "issue:7460", "issue:760", + "issue:7735", "issue:8414", "issue:8534", "issue:9106", @@ -69120,6 +69131,8 @@ "issue:9460" ], "soft_pairs": [ + "issue:5824|issue:7735", + "issue:2392|issue:5325", "issue:11135|issue:8414", "issue:1642|issue:3089", "issue:1866|issue:337", @@ -69141,182 +69154,180 @@ "issue:3418|issue:6452", "issue:3748|issue:660", "issue:3788|issue:7370", - "issue:4438|issue:5054", - "issue:10303|issue:7460", - "issue:1305|issue:5046" + "issue:4438|issue:5054" ], - "prepared_review_unit_hash": "f7d3c77aa49e8595a29c5e6ca546e683f054dee5e5232a27f8cb400df9720879", + "prepared_review_unit_hash": "b6d37ff1854f1de11ce7b2f5dd6934a5089c0f7c19647c8b7a864735eb4b9417", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33388, - "estimated_input_tokens": 8347, - "estimated_eval_tokens": 16950 + "serialized_chars": 33538, + "estimated_input_tokens": 8385, + "estimated_eval_tokens": 17026 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated bug reports, feature requests, and model-support requests across different pipelines, schedulers, and training scripts. None of the suggested soft pairs look like the same underlying issue/change, so all should stay separate.", - "confidence": 0.88, - "canonical_issue_reason": "issue:11148 is the most concrete, well-scoped bug report in the set, with a specific failing behavior and clear reproduction context.", + "summary": "This cluster is mostly a loose mix of unrelated Diffusers issues that only share broad model/pipeline vocabulary; the soft pairs are superficial and should not be merged as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "issue:5824 is the best representative because it is a concrete, well-scoped SDXL pipeline bug with a clear code-path target; the rest of the cluster is too heterogeneous to serve as a better canonical issue.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11148 is a strong representative issue because it names the exact PEFT adapter state bug and its downstream impact on enable/disable adapters.", + "best_issue_reason": "issue:5824 is the strongest issue to anchor triage: specific, technically detailed, and centered on a reproducible pipeline behavior problem rather than a vague or generic report.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5824", + "right": "issue:7735", + "accept": false, + "reason": "SDXL `get_add_time_ids()` default-value bug vs. a custom dataset caption-column error; different failure modes and code paths." + }, + { + "left": "issue:2392", + "right": "issue:5325", + "accept": false, + "reason": "`prompt_embeds` validation bug in img2img vs. packaging type distribution; unrelated symptoms and fixes." + }, { "left": "issue:11135", "right": "issue:8414", "accept": false, - "reason": "Different requests for different model/pipeline additions: SmoothCache for diffusion transformers vs ConsistencyTTA text-to-audio model support." + "reason": "Two separate proposal-style items for different acceleration/model concepts, not the same bug or change." }, { "left": "issue:1642", "right": "issue:3089", "accept": false, - "reason": "Both are scheduler bugs, but they concern different schedulers and different failure modes; not the same concrete code path." + "reason": "Different scheduler implementations with distinct missing/incorrect behavior; not one shared defect." }, { "left": "issue:1866", "right": "issue:337", "accept": false, - "reason": "Both touch img2img/inpainting behavior, but one is a DPMSolver strength bug and the other is a broader old inpainting issue." + "reason": "img2img strength handling in DPMSolverSinglestepScheduler vs. inpainting guidance behavior; only loosely related by pipeline type." }, { "left": "issue:4086", "right": "issue:444", "accept": false, - "reason": "Unrelated: SDXL ControlNet training failure vs a num_inference_steps indexing error." + "reason": "ControlNet training issue vs. timestep IndexError; different parts of the stack and different bugs." }, { "left": "issue:4213", "right": "issue:6783", "accept": false, - "reason": "Different problems in different areas: SD-XL single-file loading without accelerate vs corrupted PixArt-XL generations." + "reason": "from_single_file() / accelerate dependency issue vs. PixArt generation corruption; not the same underlying problem." }, { "left": "issue:1398", "right": "issue:760", "accept": false, - "reason": "Both mention MPS, but they target different model generations and different symptoms; too broad to treat as one bug." + "reason": "Both mention MPS, but one is SD2 support on Apple Silicon and the other is unexpected SD1.5 output on an older diffusers version." }, { "left": "issue:1524", "right": "issue:4753", "accept": false, - "reason": "Different training scripts and different failure reports; no clear shared underlying defect." + "reason": "Different DreamBooth/textual inversion training problems with different scripts and failure modes." }, { "left": "issue:4337", "right": "issue:4665", "accept": false, - "reason": "Completely different subsystems: torch.compile/ControlNet vs Windows safetensors file selection." + "reason": "torch.compile failure with a specific community ControlNet vs. Windows safetensors file selection; unrelated issues." }, { "left": "issue:4557", "right": "issue:9106", "accept": false, - "reason": "Both are attribute errors, but they point to different objects and contexts with no evidence of the same root cause." + "reason": "Different AttributeError cases in different pipelines/models; no shared concrete code path." }, { "left": "issue:5510", "right": "issue:6079", "accept": false, - "reason": "Both are dtype mismatch errors, but they occur in different pipelines and code paths; not enough to merge." + "reason": "Both are dtype mismatch errors, but they occur in different pipelines and likely different layers, so they are not the same bug." }, { "left": "issue:6486", "right": "issue:8534", "accept": false, - "reason": "Different DreamBooth training bugs: double embedding retraction vs wrong logit-normal weighting." + "reason": "Dreambooth embedding retraction bug vs. SD3 logit-normal weighting; unrelated training logic." }, { "left": "issue:10412", "right": "issue:6967", "accept": false, - "reason": "One is a VRAM usage complaint, the other is a shape-mismatch training failure; not the same issue." + "reason": "Training memory blowup in SD3.5 DreamBooth vs. SDXL prior-preservation shape mismatch; different failure classes." }, { "left": "issue:5798", "right": "issue:9166", "accept": false, - "reason": "ControlNet callback keyword handling and image overlay behavior are unrelated." + "reason": "ControlNet callback argument compatibility vs. image_processor overlay behavior; different APIs and defects." }, { "left": "issue:1734", "right": "issue:3284", "accept": false, - "reason": "Both involve DreamBooth/accelerate, but one reports severe slowdown while the other is a training failure; different bugs." + "reason": "Both concern DreamBooth/accelerate performance or execution, but one is multi-GPU slowness and the other is a script failure; not the same bug." }, { "left": "issue:10025", "right": "issue:10675", "accept": false, - "reason": "Both mention Flux, but one is about attention masking and the other about scheduler max_shift configuration." + "reason": "Flux attention-mask request vs. Flux scheduler config max_shift discrepancy; different components and changes." }, { "left": "issue:1076", "right": "issue:319", "accept": false, - "reason": "Different generations and different topics: stable-diffusion-v1-5 vs VQ-Diffusion." + "reason": "Stable Diffusion v1.5 discussion vs. VQ-Diffusion model request; no duplicate relationship." }, { "left": "issue:11148", "right": "issue:9460", "accept": false, - "reason": "PEFT adapter state handling and SD3 ControlNet model-list support are separate features/bugs with no shared code-path." + "reason": "PEFT adapter state bug vs. ControlNet list support; unrelated features in different subsystems." }, { "left": "issue:2984", "right": "issue:7370", "accept": false, - "reason": "Both are model-support requests, but they concern different research models and different modalities." + "reason": "A finished paper/model entry vs. a request to support another model architecture; not the same change." }, { "left": "issue:3418", "right": "issue:6452", "accept": false, - "reason": "Different feature requests in different pipelines: text-encoder LoRA training vs masked_image_latents input for ControlNet inpaint." + "reason": "Training script enhancement request vs. adding an inpaint pipeline input; different feature requests." }, { "left": "issue:3748", "right": "issue:660", "accept": false, - "reason": "Both involve dtype/device issues, but one is a LoRA safetensor loading mismatch and the other is an MPS fp16 runtime question." + "reason": "LoRA safetensor dtype mismatch vs. MPS fp16 revision usage; related only by dtype/platform, not the same bug." }, { "left": "issue:3788", "right": "issue:7370", "accept": false, - "reason": "Separate pipeline/model support proposals for different architectures; not the same underlying change." + "reason": "Two separate new-model/pipeline support requests for different architectures." }, { "left": "issue:4438", "right": "issue:5054", "accept": false, - "reason": "Both are SDXL errors, but one is a generic pipeline error and the other is a missing final_offload_hook attribute; no clear duplication." - }, - { - "left": "issue:10303", - "right": "issue:7460", - "accept": false, - "reason": "Different bugs in different components: attention permutation vs AdaLayerNorm behavior." - }, - { - "left": "issue:1305", - "right": "issue:5046", - "accept": false, - "reason": "Both reference diffusion-editing papers, but they are distinct methods and support requests." + "reason": "Both are SDXL errors, but one is a generic pipeline runtime error and the other is a missing `final_offload_hook` attribute; different code paths." } ] }, @@ -69341,14 +69352,14 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10054", + "issue:10303", "issue:10659", "issue:10674", "issue:1106", "issue:11398", - "issue:11627", "issue:11642", "issue:12299", - "issue:12895", + "issue:1305", "issue:1310", "issue:1405", "issue:1805", @@ -69369,17 +69380,17 @@ "issue:462", "issue:4868", "issue:4933", + "issue:5046", "issue:5570", - "issue:5607", "issue:5743", "issue:5897", "issue:6008", "issue:6584", "issue:6809", "issue:7276", + "issue:7460", "issue:7492", "issue:7497", - "issue:7686", "issue:801", "issue:9164", "issue:9358", @@ -69389,6 +69400,8 @@ "issue:9974" ], "soft_pairs": [ + "issue:10303|issue:7460", + "issue:1305|issue:5046", "issue:1805|issue:3043", "issue:3391|issue:4868", "issue:5570|issue:6008", @@ -69410,189 +69423,190 @@ "issue:2068|issue:3861", "issue:1405|issue:801", "issue:5897|issue:7492", - "issue:4219|issue:6584", - "issue:11627|issue:12895", - "issue:5607|issue:7686" + "issue:4219|issue:6584" ], - "prepared_review_unit_hash": "eb7bad0fe32357c63d9601dcd33ace44c555a19be4b8b5da7a2ee482dc177cde", + "prepared_review_unit_hash": "eaeb18a6c5fc137cfc218785e7aa0f2a4ed0d04b88b2981122327984653ee218", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33384, - "estimated_input_tokens": 8346, - "estimated_eval_tokens": 16948 + "serialized_chars": 33221, + "estimated_input_tokens": 8306, + "estimated_eval_tokens": 16868 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a set of mostly unrelated issues that only look similar at the title/wording level. None of the soft pairs appear to describe the same underlying bug or change, so all soft edges should be rejected.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "The set is mostly a grab bag of unrelated bugs/features. I only see one plausible duplicate pair around `enable_attention_slicing()` regressions; the rest target different models, pipelines, or failure modes.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 1405 is the broadest and earliest report of the attention-slicing regression, so it is the best umbrella representative for the one plausible duplicate pair.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "1405 best captures the shared code-path problem (`enable_attention_slicing()` not being honored) and is the most general reproduction among the issues here.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10303", + "right": "issue:7460", + "accept": false, + "reason": "Different components and failures: attention tensor permutation in `head_to_batch_dim` vs AdaLayerNorm behavior." + }, + { + "left": "issue:1305", + "right": "issue:5046", + "accept": false, + "reason": "Both are image-editing related, but they describe different methods/proposals (`Direct Inversion` vs `DragonDiffusion`) rather than the same bug or change." + }, { "left": "issue:1805", "right": "issue:3043", "accept": false, - "reason": "Different feature requests: scalable diffusion pipeline vs ODISE integration; no shared bug or code path." + "reason": "Unrelated requests: a scalable diffusion pipeline vs adding ODISE." }, { "left": "issue:3391", "right": "issue:4868", "accept": false, - "reason": "Unrelated problems: DeepFloyd super-resolution overexposure vs a superclass __getattr__ AttributeError." + "reason": "Different failure modes and areas: overexposure in DeepFloyd SR vs an `__getattr__` AttributeError." }, { "left": "issue:5570", "right": "issue:6008", "accept": false, - "reason": "Different failures in different areas: missing compute_snr package symbol vs IP-Adapter/LCM LoRA regression." + "reason": "Unrelated package/API issues: missing `compute_snr` vs IP-Adapter/LCM-LoRA inference failure." }, { "left": "issue:10054", "right": "issue:9974", "accept": false, - "reason": "Unrelated scheduler test bug vs ControlNet SD3 parameter support request." + "reason": "Different scheduler/config problems: wrong `torch.linspace` parameters in EulerDiscreteScheduler vs SD3 controlnet expected-parameter handling." }, { "left": "issue:1310", "right": "issue:1990", "accept": false, - "reason": "Stable Diffusion/K-Diffusion behavior bug vs a pipeline proposal; not the same issue." + "reason": "Both are pipeline ideas, but they are distinct proposals for different workflows and do not describe the same change." }, { "left": "issue:2331", "right": "issue:3043", "accept": false, - "reason": "ControlNet pipeline request vs ODISE support request; same broad domain but different changes." + "reason": "Both mention pipelines, but one is ControlNet support and the other is ODISE; these are separate feature requests." }, { "left": "issue:3113", "right": "issue:7497", "accept": false, - "reason": "Layout control via cross-attention guidance vs fine-grained subject attribute control; distinct proposals." + "reason": "Different research/feature requests with different goals and code paths." }, { "left": "issue:12299", "right": "issue:9443", "accept": false, - "reason": "Both concern SD3, but one is a tensor mismatch with negative prompt embeddings and the other is tokenizer instantiation; different bugs." + "reason": "Distinct SD3 issues: tensor mismatch with negative prompt embeddings vs tokenizer slow/fast instantiation." }, { "left": "issue:4542", "right": "issue:6809", "accept": false, - "reason": "SDXL img2img multi-image generation bug vs LoRA adapter loading under PEFT; unrelated code paths." + "reason": "Different bugs: SDXL img2img multi-image generation vs LoRA adapter loading under PEFT." }, { "left": "issue:7276", "right": "issue:9548", "accept": false, - "reason": "Dreambooth network/model download error vs Flux Dreambooth LoRA training issue; not the same failure." + "reason": "Different Dreambooth/Flux problems with different symptoms; not the same underlying bug." }, { "left": "issue:4933", "right": "issue:954", "accept": false, - "reason": "Dreambooth checkpoint export problem vs a generic login/loading issue; no concrete overlap." + "reason": "Completely different areas: missing model artifact after training vs login/load failure." }, { "left": "issue:9164", "right": "issue:9358", "accept": false, - "reason": "Both involve Flux Dreambooth LoRA, but one is convergence/training behavior and the other is redundant text encoder reinitialization; different bugs." + "reason": "Both involve Flux Dreambooth LoRA, but one is convergence on a sample prompt and the other is redundant text-encoder reinitialization; not clearly the same bug." }, { "left": "issue:1106", "right": "issue:2459", "accept": false, - "reason": "Both mention repeated attention/model usage, but one is PNDM inference behavior and the other is xFormers in a training script; not the same bug." + "reason": "Different code paths: PNDM repeatedly calling UNet vs xFormers attention in a training script." }, { "left": "issue:2532", "right": "issue:2579", "accept": false, - "reason": "Panorama pipeline get_views bug vs a vague xFormers bug/misuse report; insufficient concrete overlap." + "reason": "Panorama `get_views` bug and a possible xFormers issue are unrelated." }, { "left": "issue:389", "right": "issue:462", "accept": false, - "reason": "README example breakage vs Apple MPS img2img regression; different contexts and code paths." + "reason": "README example breakage is not the same as the Apple MPS img2img regression." }, { "left": "issue:4551", "right": "issue:5743", "accept": false, - "reason": "DreamBooth SDXL crash vs latent consistency distillation training stuck at 0%; unrelated training failures." + "reason": "Different training failures: DreamBooth crash on SDXL base vs latent consistency distillation stuck at 0%." }, { "left": "issue:10659", "right": "issue:10674", "accept": false, - "reason": "Both are user-facing generation problems, but ConsisID output quality and FluxPipeline GGUF support are different pipelines and bugs." + "reason": "Different models and symptoms: poor ConsisID output vs FluxPipeline/GGUF incompatibility." }, { "left": "issue:11398", "right": "issue:11642", "accept": false, - "reason": "Different Flux bugs: missing config/attn processor support vs prompt_embeds_scale having no effect." + "reason": "Both are Flux-related, but one is missing config/set_default_attn_processor support and the other is `prompt_embeds_scale` having no effect; separate issues." }, { "left": "issue:2068", "right": "issue:3861", "accept": false, - "reason": "Same API area, but one is dtype conversion in from_pretrained and the other is a typo/alternate from_pretrained naming question." + "reason": "`from_pretrained` dtype conversion behavior is unrelated to the `Pipeline.from_pretrained` naming/question." }, { "left": "issue:1405", "right": "issue:801", - "accept": false, - "reason": "Both involve attention slicing, but one is main-branch breakage and the other is inpainting/img2img not honoring it; not clearly the same defect." + "accept": true, + "reason": "Both report the same underlying regression: `enable_attention_slicing()` is not being respected in pipeline inference, with one broader report and one that calls out inpainting/img2img specifically." }, { "left": "issue:5897", "right": "issue:7492", "accept": false, - "reason": "Device-mismatch LoRA training error vs a missing Hugging Face repository URL; unrelated root causes." + "reason": "LoRA device-mismatch training error and a Hugging Face repository-not-found error are unrelated." }, { "left": "issue:4219", "right": "issue:6584", "accept": false, - "reason": "ControlNet checkpoint conversion support vs DreamShaper inpaint model conversion; related area but different requests." - }, - { - "left": "issue:11627", - "right": "issue:12895", - "accept": false, - "reason": "Both SD3-related, but one is an IP-Adapter integration issue and the other is a num_images_per_prompt crash with batched prompts; different bugs." - }, - { - "left": "issue:5607", - "right": "issue:7686", - "accept": false, - "reason": "Tensor size mismatch during runtime vs a UNet constructor ValueError about cross_attention_dim; not the same problem." + "reason": "Both concern conversion tooling, but they are different checkpoint/model cases and do not clearly indicate the same bug or change." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative on the soft-edge pairs. The only accepted duplicate, issue:1405 vs issue:801, is well supported by the titles and the summary does not overclaim beyond that. The other pairwise rejections are plausible and appropriately cautious." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -69617,9 +69631,11 @@ "issue:10897", "issue:1106", "issue:11220", + "issue:11627", "issue:1188", "issue:12071", "issue:12149", + "issue:12895", "issue:1298", "issue:13034", "issue:1675", @@ -69640,24 +69656,24 @@ "issue:4042", "issue:4052", "issue:4069", - "issue:4125", "issue:4303", "issue:4542", "issue:477", "issue:5374", "issue:5405", + "issue:5607", "issue:6113", - "issue:6157", - "issue:6288", "issue:6793", "issue:7428", "issue:7562", + "issue:7686", "issue:8236", "issue:8801", - "issue:9092", "issue:9822" ], "soft_pairs": [ + "issue:11627|issue:12895", + "issue:5607|issue:7686", "issue:10382|issue:6793", "issue:2695|issue:2712", "issue:3958|issue:7562", @@ -69679,182 +69695,180 @@ "issue:2856|issue:6113", "issue:3112|issue:4042", "issue:3565|issue:8236", - "issue:4052|issue:5405", - "issue:4125|issue:6288", - "issue:6157|issue:9092" + "issue:4052|issue:5405" ], - "prepared_review_unit_hash": "3b4ccf2ac8cb4077293cf85c3ff6b9d81a73a1b2bc94693627b2367af8d0d5f7", + "prepared_review_unit_hash": "7b4f5ef4b06c7779bbb31c1cd229db1e4a9c6f9e93e1c8f0116504caf3c3ecd0", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33123, - "estimated_input_tokens": 8281, - "estimated_eval_tokens": 16818 + "serialized_chars": 33314, + "estimated_input_tokens": 8329, + "estimated_eval_tokens": 16914 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate or mergeable issue/PR pair is supported here; the soft-similarity pairs mostly share a subsystem keyword but describe different bugs, feature requests, or docs/training questions.", + "summary": "This cluster is a loose collection of unrelated diffusers issues spanning model support requests, training bugs, pipeline regressions, docs, and scheduler/config problems. The proposed soft pairs are mostly superficial title similarity and do not describe the same underlying defect or change.", "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11627", + "right": "issue:12895", + "accept": false, + "reason": "Both mention StableDiffusion3Pipeline, but one is a specific InstantX IP-Adapter support request and the other is a num_images_per_prompt batched-prompt crash; different bugs/changes." + }, + { + "left": "issue:5607", + "right": "issue:7686", + "accept": false, + "reason": "Both are runtime errors, but the first is a tensor shape mismatch in a pipeline and the second is a UNet config/cross-attention initialization error; unrelated code paths." + }, { "left": "issue:10382", "right": "issue:6793", "accept": false, - "reason": "VAE forward return-type AttributeError is unrelated to Hydra config serialization failure." + "reason": "One reports VAE forward return-type AttributeError, the other is a Hydra serialization problem; not the same issue." }, { "left": "issue:2695", "right": "issue:2712", "accept": false, - "reason": "Both concern training/configuration, but one is a TensorBoard validation failure and the other is a missing config field; different root causes." + "reason": "ControlNet validation/tensorboard behavior and missing norm_num_groups config are different training/configuration problems." }, { "left": "issue:3958", "right": "issue:7562", "accept": false, - "reason": "Both mention CPU offload, but one is about LoRA loading and the other is a multicontrolnet failure in SDXL img2img." + "reason": "Both involve offload and adapters, but one is LoRA with enable_model_cpu_offload and the other is multicontrolnet offload failure; not the same concrete bug." }, { "left": "issue:10010", "right": "issue:9822", "accept": false, - "reason": "A missing FLUX Fill model_index.json for an example is not the same problem as loading SDXL LoRAs into Flux." + "reason": "FLUX Fill model_index.json 404 vs loading SDXL LoRAs into Flux are unrelated compatibility/model-loading issues." }, { "left": "issue:10112", "right": "issue:7428", "accept": false, - "reason": "Detail-Daemon issue is unrelated to a feature request for layer diffusion support." + "reason": "The first is an unclear diffusers discussion issue, while the second is a request to support layer diffusion; no shared defect." }, { "left": "issue:10752", "right": "issue:13034", "accept": false, - "reason": "FP16 gradient unscale bug is unrelated to WanPipeline height/width handling." + "reason": "FP16 gradient unscale failure and WanPipeline height/width handling are different bugs." }, { "left": "issue:10855", "right": "issue:12071", "accept": false, - "reason": "Gradient accumulation causing optimizer.step() every step is unrelated to a FLUX transformer forward-behavior change." + "reason": "Gradient accumulation causing optimizer.step every step is unrelated to Flux Kontext transformer forward behavior changes." }, { "left": "issue:10897", "right": "issue:8801", "accept": false, - "reason": "Two separate model support requests with different targets; not the same bug or change." + "reason": "Both are feature requests for different models (SkyReels-A1 vs Kolors); not duplicates." }, { "left": "issue:1106", "right": "issue:2163", "accept": false, - "reason": "Repeated UNet calls in PNDM is a scheduler behavior issue, not the dtype mismatch seen in text-to-image training." + "reason": "PNDM repeatedly calling UNet and a dtype mismatch during text-to-image training are separate problems." }, { "left": "issue:11220", "right": "issue:12149", "accept": false, - "reason": "Broken documentation page and pip/source version mismatch are different issues." + "reason": "Broken docs page and pip/source version mismatch are distinct release/documentation issues." }, { "left": "issue:1188", "right": "issue:2800", "accept": false, - "reason": "A pipeline feature request for CLIPSeg x Stable Diffusion is distinct from DiffEdit image-editing support." + "reason": "CLIPSeg x Stable Diffusion pipeline request and DiffEdit pipeline request are different additions." }, { "left": "issue:1298", "right": "issue:4542", "accept": false, - "reason": "Broken SD img2img after a change is not the same as SDXL Img2Img producing only one image." + "reason": "Both concern img2img pipelines, but one is a regression in Stable Diffusion img2img and the other is SDXL img2img output count limitation; different failures." }, { "left": "issue:1675", "right": "issue:5374", "accept": false, - "reason": "Model conversion state-dict errors do not match an unexpected checkpoint_path argument error." + "reason": "Model conversion state-dict error and unexpected checkpoint_path argument are unrelated conversion/API issues." }, { "left": "issue:176", "right": "issue:4069", "accept": false, - "reason": "Missing Glide pipeline and unloading a textual inversion checkpoint are unrelated concerns." + "reason": "Glide pipeline missing and unloading a textual inversion checkpoint are not the same bug." }, { "left": "issue:1949", "right": "issue:4303", "accept": false, - "reason": "Dreambooth tutorial training errors are not the same as a Flax pipeline crash loading SDXL." + "reason": "DreamBooth tutorial training errors and FlaxStableDiffusionPipeline loading SDXL are different subsystems and failures." }, { "left": "issue:2153", "right": "issue:477", "accept": false, - "reason": "Float16 incompatibility on GTX 1660 is unrelated to the PNDM off-by-one scheduler bug." + "reason": "GPU float16 compatibility and a PNDM off-by-one scheduler error are unrelated." }, { "left": "issue:2377", "right": "issue:338", "accept": false, - "reason": "EMAModel test coverage request and LMS Discrete test request are different test additions." + "reason": "Both ask for tests, but for different components (EMAModel vs LMS Discrete); not duplicates." }, { "left": "issue:2377", "right": "issue:339", "accept": false, - "reason": "EMAModel tests and Karras VE tests are separate test tasks." + "reason": "Same as above: different test targets, no shared underlying bug/change." }, { "left": "issue:2856", "right": "issue:6113", "accept": false, - "reason": "Converting an SD model to diffusers is not the same as using sd_control_collection models in diffusers." + "reason": "Model conversion failure and how to use a control-model repo in diffusers are distinct user problems." }, { "left": "issue:3112", "right": "issue:4042", "accept": false, - "reason": "Multi-ControlNet inpainting is not the same as combining reference-only, inpainting, and depth control." + "reason": "Both touch ControlNet composition, but one requests multi-ControlNet inpainting and the other asks how to combine reference-only with inpainting and depth; related area, not the same change." }, { "left": "issue:3565", "right": "issue:8236", "accept": false, - "reason": "CUDA OOM during text-to-image training is unrelated to the multi-GPU step-count scheduler bug." + "reason": "OOM in text-to-image training and wrong LR scheduler step count on multi-GPU are different training issues." }, { "left": "issue:4052", "right": "issue:5405", "accept": false, - "reason": "Missing from_single_file attribute and SDXL weights not overwriting are different single-file loading problems." - }, - { - "left": "issue:4125", - "right": "issue:6288", - "accept": false, - "reason": "A loss-curve question during LoRA fine-tuning is not the same as an overly small default learning rate issue." - }, - { - "left": "issue:6157", - "right": "issue:9092", - "accept": false, - "reason": "IP-Adapter not working in ControlNet img2img is unrelated to a missing model_index.json in FluxPipeline." + "reason": "Missing from_single_file API and sdxl.from_single_file not overwriting weights are different bugs." } ] }, @@ -69901,33 +69915,35 @@ "issue:351", "issue:3566", "issue:3767", + "issue:4125", "issue:4134", "issue:4376", "issue:4414", "issue:460", "issue:4880", "issue:5372", - "issue:5414", "issue:5761", "issue:5798", - "issue:5857", "issue:6068", "issue:6090", "issue:6091", + "issue:6157", + "issue:6288", "issue:6463", - "issue:6548", "issue:7428", "issue:7619", "issue:7802", "issue:7872", - "issue:7895", "issue:8589", "issue:8708", "issue:8837", + "issue:9092", "issue:9551", "issue:9895" ], "soft_pairs": [ + "issue:4125|issue:6288", + "issue:6157|issue:9092", "issue:6463|issue:8708", "issue:7428|issue:7802", "issue:8589|issue:8837", @@ -69950,195 +69966,196 @@ "issue:10767|issue:9551", "issue:1811|issue:3064", "issue:2769|issue:351", - "issue:4414|issue:6091", - "issue:5414|issue:7895", - "issue:5857|issue:6548" + "issue:4414|issue:6091" ], - "prepared_review_unit_hash": "f3f1fa8e8ac28431b9f6bba5149f6f6f72e61345c0027dce7bf0f9739b37b06b", + "prepared_review_unit_hash": "988ab67530a5032173ad2bac14be7735e4c21350b3f59907b597d3bf5257357f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34367, - "estimated_input_tokens": 8592, - "estimated_eval_tokens": 17440 + "serialized_chars": 34295, + "estimated_input_tokens": 8574, + "estimated_eval_tokens": 17404 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly unrelated issue reports spanning different Diffusers subsystems. The only clear duplicate-like pair is the two checkpoint write-error reports (#2661/#2662); the broader/general one is the best canonical issue candidate.", - "confidence": 0.83, - "canonical_issue_reason": "Issue #2662 has the broader title, more discussion, and appears to describe the same checkpoint-saving write error as #2661, making it the best canonical issue in this set.", + "summary": "This cluster is mostly heterogeneous Diffusers issues with only a couple of plausible duplicate pairs. The strongest duplicate is the checkpoint write-error pair (2661/2662); a second likely duplicate is the DreamBooth channel-mismatch pair (6090/7619).", + "confidence": 0.91, + "canonical_issue_reason": "No single cluster-wide canonical issue fits well because the items span many unrelated bugs and feature requests; among the clear duplicates, issue 2662 is the broader representative for the checkpoint write-error bug.", "canonical_pr_reason": null, - "best_issue_reason": "#2662 is the most suitable global issue representative because it is the more general and more active version of the duplicate checkpoint-saving write-error report.", + "best_issue_reason": "Issue 2662 is the best issue representative because it uses the broadest wording for the checkpoint save/write-error problem and is the cleanest anchor for deduping its pair.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4125", + "right": "issue:6288", + "accept": false, + "reason": "Both are LoRA-training questions, but one is about loss curves and the other about default learning rate; not the same bug/change." + }, + { + "left": "issue:6157", + "right": "issue:9092", + "accept": false, + "reason": "ControlNet/IP-Adapter behavior vs a missing model_index.json error are different problems." + }, { "left": "issue:6463", "right": "issue:8708", "accept": false, - "reason": "Different topics: missing audio_diffusion pipeline vs missing SD3 VAE shift_factor in DreamBooth examples." + "reason": "Unrelated topics: missing audio_diffusion pipeline vs SD3 dreambooth VAE shift_factor." }, { "left": "issue:7428", "right": "issue:7802", "accept": false, - "reason": "One is a feature/support request for layer diffusion; the other is a scheduler support request. Not the same change." + "reason": "Feature request for layer diffusion support is unrelated to an OLSS scheduler request." }, { "left": "issue:8589", "right": "issue:8837", "accept": false, - "reason": "Different bugs in different areas: DreamBooth prior preservation vs SlicedAttnProcessor behavior." + "reason": "DreamBooth prior-preservation bug and SlicedAttnProcessor bug are different code paths." }, { "left": "issue:2661", "right": "issue:3296", "accept": false, - "reason": "Both involve checkpoint-related workflows, but one is a write error and the other is DreamBooth save/load with text_encoder; not the same concrete bug." + "reason": "Both mention checkpointing, but 3296 is a DreamBooth text_encoder save/load issue, not the same write-error bug as 2661." }, { "left": "issue:2662", "right": "issue:3296", "accept": false, - "reason": "Same as above: checkpoint write error versus DreamBooth checkpoint save/load failure are related but distinct." + "reason": "Checkpoint write error vs DreamBooth text_encoder checkpoint save/load failure are distinct failures." }, { "left": "issue:3428", "right": "issue:3566", "accept": false, - "reason": "Both mention unexpected results in pipelines, but they concern different pipelines and different functionality." + "reason": "Different pipelines and symptoms; this is not the same underlying bug." }, { "left": "issue:4880", "right": "issue:5798", "accept": false, - "reason": "Both are ControlNet pipeline issues, but one is an input-type validation error and the other is an unexpected callback keyword error." + "reason": "One is a ControlNet image-argument TypeError, the other is a callback keyword-argument API mismatch." }, { "left": "issue:11945", "right": "issue:9895", "accept": false, - "reason": "Different failure modes and environments: floating point exception with nightly PyTorch/CUDA versus FluxPipeline device mismatch." + "reason": "Nightly PyTorch/CUDA floating-point exception vs FluxPipeline device mismatch are different runtime errors." }, { "left": "issue:1109", "right": "issue:460", "accept": false, - "reason": "CPU utilization issue is unrelated to ONNX runtime backend selection." + "reason": "CPU core utilization and ONNX CUDA/CPU runtime selection are unrelated." }, { "left": "issue:4414", "right": "issue:5761", "accept": false, - "reason": "A sample request for consistency models is not the same as a complaint about the cost of the consistency decoder." + "reason": "Consistency-model sample request and consistency-decoder cost discussion are not the same issue." }, { "left": "issue:11410", "right": "issue:7872", "accept": false, - "reason": "A specific attention_processor unpacking error is not the same as a general version/package report." + "reason": "A specific attention __call__ unpacking ValueError is unrelated to a version-tag issue." }, { "left": "issue:1541", "right": "issue:3767", "accept": false, - "reason": "DreamBooth MODEL_TYPE loading failure is unrelated to FrozenDict mutability." + "reason": "DreamBooth model-type detection failure is unrelated to FrozenDict mutability." }, { "left": "issue:11148", "right": "issue:13279", "accept": false, - "reason": "Different PEFT/transformers loading issues with different symptoms and code paths." + "reason": "PEFT adapter state handling and AutoencoderRAE loading with older transformers are different bugs." }, { "left": "issue:3116", "right": "issue:6068", "accept": false, - "reason": "Both are sampling quality bugs, but they involve different schedulers, pipelines, and failure conditions." + "reason": "KDPM2DiscreteScheduler image2image noise issue is not the same as SDXL DDIM sample correctness." }, { "left": "issue:6090", "right": "issue:7619", - "accept": false, - "reason": "Both mention channel mismatches, but one is a specific runtime shape error and the other is a DreamBooth training issue; not clearly the same bug." + "accept": true, + "reason": "Both report the same underlying channel-mismatch problem (8 channels expected, 4 provided) in DreamBooth/related diffusion code." }, { "left": "issue:1195", "right": "issue:1611", "accept": false, - "reason": "Memory-efficient attention with fp16 weights is unrelated to the StableDiffusionMegaPipeline regression." + "reason": "fp16 memory-efficient attention and StableDiffusionMegaPipeline regressions are not the same failure." }, { "left": "issue:1519", "right": "issue:4134", "accept": false, - "reason": "Different Apple/MPS problems: noisy batch outputs versus ControlNet tensor-size mismatch." + "reason": "Apple MPS issues in both, but noisy batch outputs and ControlNet tensor-size mismatch are different bugs." }, { "left": "issue:1323", "right": "issue:1674", "accept": false, - "reason": "Flax dtype-loading bug and Dreambooth scheduling/DPMSolver loading issue are different problems." + "reason": "Loading Flax models without dtype and DreamBooth scheduling/config loading are unrelated." }, { "left": "issue:4376", "right": "issue:5372", "accept": false, - "reason": "Textual inversion crash in SDXL and safety_checker usage question are unrelated." + "reason": "Textual inversion crash in SDXL and safety_checker usage questions are different issues." }, { "left": "issue:10767", "right": "issue:9551", "accept": false, - "reason": "FlashVideo <> Diffusers and Flux controlnet usage are not the same underlying issue." + "reason": "Flux-related community pipeline topics, but not the same concrete bug or change." }, { "left": "issue:1811", "right": "issue:3064", "accept": false, - "reason": "Weight-file selection during from_pretrained is unrelated to loading a safetensors LoRA." + "reason": "Weight-selection behavior in from_pretrained vs loading a safetensors LoRA are distinct problems." }, { "left": "issue:2769", "right": "issue:351", "accept": false, - "reason": "Model conversion corruption and direct CUDA loading are different issues." + "reason": "Checkpoint conversion corruption and CUDA loading behavior are unrelated." }, { "left": "issue:4414", "right": "issue:6091", "accept": false, - "reason": "Consistency model sampling request is unrelated to consistency decoder memory size behavior." - }, - { - "left": "issue:5414", - "right": "issue:7895", - "accept": false, - "reason": "Broken PEFT set_adapters(weights) and Flax memory-efficient attention are separate bugs." - }, - { - "left": "issue:5857", - "right": "issue:6548", - "accept": false, - "reason": "LCM LoRA behavior in AnimateDiff SDXL is unrelated to RGBA image handling in prepare_latents." + "reason": "Consistency-model sampling and consistency-decoder memory-size handling are different concerns." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: the cluster is heterogeneous, and 2662 is a reasonable broader representative for the checkpoint write-error duplicate pair. The soft-edge verdicts are mostly conservative, and the accepted 6090/7619 pair is defensible from the titles as a DreamBooth channel-mismatch duplicate." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -70173,28 +70190,28 @@ "issue:2578", "issue:3222", "issue:3537", - "issue:381", "issue:4014", "issue:4297", - "issue:435", "issue:4478", "issue:4534", "issue:4593", "issue:4932", - "issue:4957", "issue:5008", "issue:5043", - "issue:5064", "issue:5079", + "issue:5414", "issue:562", + "issue:5857", "issue:5968", "issue:6518", + "issue:6548", "issue:6616", "issue:6749", "issue:7210", "issue:7589", "issue:7666", "issue:7757", + "issue:7895", "issue:7972", "issue:8463", "issue:8615", @@ -70204,6 +70221,8 @@ "issue:984" ], "soft_pairs": [ + "issue:5414|issue:7895", + "issue:5857|issue:6548", "issue:4478|issue:7972", "issue:4593|issue:6749", "issue:1037|issue:5043", @@ -70225,182 +70244,180 @@ "issue:3222|issue:4932", "issue:7757|issue:8894", "issue:1064|issue:562", - "issue:2578|issue:984", - "issue:381|issue:435", - "issue:4957|issue:5064" + "issue:2578|issue:984" ], - "prepared_review_unit_hash": "b10dabf29000dbf5e81b8f7365383634f1b5a4c04e0b9b8a088b353f1bfbe7aa", + "prepared_review_unit_hash": "96705845007fa804b8dbda05f86e590393b292c0ca264025814e6ccc07a552c7", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33601, - "estimated_input_tokens": 8401, - "estimated_eval_tokens": 17058 + "serialized_chars": 33587, + "estimated_input_tokens": 8397, + "estimated_eval_tokens": 17050 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster does not contain a single duplicate set; the soft pairs are mostly superficial title matches across unrelated bugs, feature requests, and scripts. I would reject all proposed edges.", - "confidence": 0.96, + "summary": "These items are not duplicates of one another; the cluster is a loose lexical-similarity bucket containing unrelated issues across different pipelines, schedulers, training scripts, and environment bugs. All soft edges should be rejected.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:12195 is the strongest standalone report: it is concrete, version-specific, and clearly describes a reproducible runtime regression in AutoencoderKL with PyTorch 2.3.1.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5414", + "right": "issue:7895", + "accept": false, + "reason": "PEFT adapter weighting vs Flax attention are unrelated subsystems and failure modes." + }, + { + "left": "issue:5857", + "right": "issue:6548", + "accept": false, + "reason": "An LCM LoRA quality issue is unrelated to an img2img latent-vs-image RGBA bug." + }, { "left": "issue:4478", "right": "issue:7972", "accept": false, - "reason": "Both are SDXL runtime/type-mismatch complaints, but they hit different code paths and different features (from_pretrained bias/input types vs LEDITS++)." + "reason": "SDXL initialization dtype mismatch is not the same as LEDITS++ type mismatch." }, { "left": "issue:4593", "right": "issue:6749", "accept": false, - "reason": "LoRA rank-loading failure and an InstantID replication issue are different models and different bugs." + "reason": "LoRA rank-loading error is unrelated to an InstantID reproduction issue." }, { "left": "issue:1037", "right": "issue:5043", "accept": false, - "reason": "A community-pipeline addition request is unrelated to a Wurstchen LoRA finetuning request." + "reason": "A community pipeline request is unrelated to a Wurstchen LoRA feature request." }, { "left": "issue:10518", "right": "issue:8671", "accept": false, - "reason": "One is a bug in an SD3 LoRA training script; the other is a code-cleanup note in a different SD3 script." + "reason": "A broken SD3 Colab script is not the same as redundant code in train_dreambooth_lora_sd3.py." }, { "left": "issue:10640", "right": "issue:7210", "accept": false, - "reason": "Both involve multi-GPU/model wrapping, but the failures are different: DataParallel on FluxInpaint vs accelerate dispatch on SDXL." + "reason": "DataParallel on FluxInpaintPipeline is unrelated to accelerate.load_checkpoint_and_dispatch on SDXL." }, { "left": "issue:12149", "right": "issue:12195", "accept": false, - "reason": "Packaging/source-code mismatch and a torch.custom_op compatibility error are unrelated environment problems." + "reason": "Source/package version mismatch is a different problem from the torch.library.custom_op regression." }, { "left": "issue:1344", "right": "issue:1755", "accept": false, - "reason": "The second item is not a substantive related bug report; there is no shared underlying issue." + "reason": "Multi-GPU attention support and a non-English generic issue are unrelated." }, { "left": "issue:3537", "right": "issue:5968", "accept": false, - "reason": "These are both conversion utilities, but one is ControlNet conversion and the other is Text2Video/Img2Video ONNX export." + "reason": "A ControlNet conversion script request is unrelated to an ONNX video conversion script request." }, { "left": "issue:5008", "right": "issue:6616", "accept": false, - "reason": "A TryOnDiffusion implementation proposal is not the same change as a classifier-guidance proposal." + "reason": "TryOnDiffusion implementation discussion is unrelated to adding classifier guidance." }, { "left": "issue:7666", "right": "issue:8786", "accept": false, - "reason": "Custom pipeline loading from a single file and SD3 token-length behavior are different problems." + "reason": "from_single_file support for a custom pipeline is unrelated to SD3 token-length behavior." }, { "left": "issue:4014", "right": "issue:4297", "accept": false, - "reason": "Both mention artifacts/watermarks, but one is SDXL watermark pixel artifacts and the other is batched img2img fishnet artifacts; not the same bug." + "reason": "General SDXL watermark artifacts and batched img2img fishnet artifacts are different concrete bugs." }, { "left": "issue:11208", "right": "issue:7589", "accept": false, - "reason": "Unsupported MultiControlNetModel in an SD3 inpainting pipeline is unrelated to GLIGEN layout synthesis failures." + "reason": "MultiControlNet support in SD3 inpainting is unrelated to GLIGEN layout-to-image failure." }, { "left": "issue:11839", "right": "issue:12995", "accept": false, - "reason": "FluxKontext gguf input-size errors and Cosmos-Predict2.5 video semantics are different issues." + "reason": "GGUF input-size loading failure is unrelated to a Cosmos-Predict2.5 pipeline usage question." }, { "left": "issue:1755", "right": "issue:1990", "accept": false, - "reason": "No meaningful overlap; the pair is not the same bug or feature." + "reason": "A generic issue title is unrelated to a depth-to-image variation pipeline proposal." }, { "left": "issue:2361", "right": "issue:4534", "accept": false, - "reason": "Checkpoint-saving crashes in text-to-image finetuning and a validation crash on MPS are different failure points and scripts." + "reason": "Checkpoint saving crash in a finetuning script is unrelated to SDXL validation on mps." }, { "left": "issue:5079", "right": "issue:8615", "accept": false, - "reason": "Loading LoRA weights and fusing LoRA are distinct features, not the same underlying change." + "reason": "LyCORIS loading support and SD3 fuse_lora support are different adapter/model features." }, { "left": "issue:6518", "right": "issue:8463", "accept": false, - "reason": "A NoneType floor-division error in distillation and a NoneType tokenize error in InstructP2P-SDXL are unrelated." + "reason": "LCM distillation NoneType error is unrelated to a tokenizer NoneType error in InstructP2P-SDXL." }, { "left": "issue:1742", "right": "issue:1974", "accept": false, - "reason": "Inference speed variability and Euler scheduler generator handling are different bugs." + "reason": "Inference speed variability for repeated prompts is unrelated to Euler scheduler generator-list handling." }, { "left": "issue:3222", "right": "issue:4932", "accept": false, - "reason": "Both mention missing saved weights, but custom diffusion and Dreambooth are different training flows and artifacts." + "reason": "Missing custom diffusion weights after training is unrelated to Dreambooth not returning pytorch_model.bin." }, { "left": "issue:7757", "right": "issue:8894", "accept": false, - "reason": "A multi-GPU OOM in train_controlnet_sdxl is unrelated to an error in train_cm_ct_unconditional.py." + "reason": "ControlNet SDXL multi-GPU OOM is unrelated to a train_cm_ct_unconditional script error." }, { "left": "issue:1064", "right": "issue:562", "accept": false, - "reason": "These are scheduler-related, but one reports add_noise errors while the other reports inpaint incompatibility; not the same concrete bug." + "reason": "LMSDiscreteScheduler.add_noise error and StableDiffusionInpaintPipeline/LMS incompatibility are related in theme but not clearly the same concrete bug." }, { "left": "issue:2578", "right": "issue:984", "accept": false, - "reason": "Editable install problems and F.interpolate large-batch failures are unrelated." - }, - { - "left": "issue:381", - "right": "issue:435", - "accept": false, - "reason": "Different notebook errors in different tutorial sections; no shared code-path bug." - }, - { - "left": "issue:4957", - "right": "issue:5064", - "accept": false, - "reason": "Both are disk-space complaints, but they occur in different pipelines and scenarios and do not indicate one underlying bug." + "reason": "Editable install failure is unrelated to an F.interpolate large-batch interpolation bug." } ] }, @@ -70426,13 +70443,10 @@ "nodes": [ "issue:10290", "issue:10737", - "issue:10767", - "issue:10860", "issue:10953", "issue:11497", "issue:1161", "issue:1214", - "issue:12221", "issue:12547", "issue:1314", "issue:1875", @@ -70445,13 +70459,17 @@ "issue:3689", "issue:3717", "issue:3767", + "issue:381", "issue:3876", "issue:4210", + "issue:435", "issue:4665", "issue:4701", "issue:4876", + "issue:4957", "issue:4969", "issue:4985", + "issue:5064", "issue:5124", "issue:514", "issue:5162", @@ -70469,10 +70487,11 @@ "issue:7939", "issue:8228", "issue:9089", - "issue:9611", - "issue:9810" + "issue:9611" ], "soft_pairs": [ + "issue:381|issue:435", + "issue:4957|issue:5064", "issue:3279|issue:5124", "issue:5465|issue:9611", "issue:10737|issue:9089", @@ -70494,191 +70513,186 @@ "issue:1875|issue:6164", "issue:3717|issue:4985", "issue:3383|issue:4665", - "issue:10290|issue:10953", - "issue:10767|issue:12221", - "issue:10860|issue:9810" + "issue:10290|issue:10953" ], - "prepared_review_unit_hash": "601ed74a335bfb023b638e43c12b4fe666d453c4fa9e897e3b7740ae86530550", + "prepared_review_unit_hash": "4082e174959371c7b94cbb5ff009b06c23791ece841ca2d014b229f86af8f44a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33166, - "estimated_input_tokens": 8292, - "estimated_eval_tokens": 16840 + "serialized_chars": 33278, + "estimated_input_tokens": 8320, + "estimated_eval_tokens": 16896 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a set of unrelated issue reports that only share broad model/training/loading themes. The Flux/ControlNet torch.compile pair (10290/10953) is only weakly similar from titles alone, so it should be treated as related-but-uncertain, not a confirmed duplicate.", - "confidence": 0.75, - "canonical_issue_reason": "issue:10290 is the clearest user-facing report of the Flux+ControlNet torch.compile failure; issue:10953 may be a related fix hint, but the packet only provides titles, so duplicate status is not established.", + "summary": "This cluster is not a duplicate set; it\u2019s a bag of unrelated issues that only share superficial vocabulary around pipelines, training, Flux, ControlNet, and loading. I would reject all soft-edge merges and avoid selecting a canonical issue/PR for the cluster.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:10290 best represents the cluster because it states the concrete runtime symptom and affected component more directly than the terse transpose/permute note.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:381", + "right": "issue:435", + "accept": false, + "reason": "Different bugs: a notebook section error vs a textual inversion AutoencoderKLOutput attribute error." + }, + { + "left": "issue:4957", + "right": "issue:5064", + "accept": false, + "reason": "Both mention space issues, but one is disk space during ControlNet+inpainting-SDXL and the other is out-of-space using a refiner for inference; different workflows and failure modes." + }, { "left": "issue:3279", "right": "issue:5124", "accept": false, - "reason": "Both are training/checkpoint topics, but one is checkpoint resume and the other is safetensor fine-tuning; different bugs." + "reason": "Checkpoint resume failure in Dreambooth LoRA is unrelated to a question about finetuning a .safetensor checkpoint." }, { "left": "issue:5465", "right": "issue:9611", "accept": false, - "reason": "DiffEdit inpaint_strength handling and Flux image batching are unrelated failure modes." + "reason": "Different pipelines and symptoms: DiffEdit inpaint_strength limitation vs Flux image batching error." }, { "left": "issue:10737", "right": "issue:9089", "accept": false, - "reason": "A missing image_encoder in encode_image and a missing FluxPipeline symbol are different code-path issues." + "reason": "One is missing image_encoder in FluxPipeline.encode_image; the other is FluxPipeline not found at import time." }, { "left": "issue:1161", "right": "issue:5162", "accept": false, - "reason": "Model conversion from Flax to PyTorch is not the same as converting a checkpoint folder to safetensors." + "reason": "Model conversion from Flax to PyTorch is unrelated to a script for converting a folder checkpoint to safetensors." }, { "left": "issue:1314", "right": "issue:4701", "accept": false, - "reason": "xformers backward on Windows and AutoencoderKL float32 CUDA errors are distinct problems." + "reason": "Both are CUDA/precision-related, but one is a Windows xformers backward error and the other is AutoencoderKL decoder differentiation at float32." }, { "left": "issue:2331", "right": "issue:5284", "accept": false, - "reason": "A general ControlNet pipeline request and ControlNet-LLLite support are different feature requests." + "reason": "A general ControlNet pipeline request vs ControlNet-LLLite support; same broad area, but not the same change or bug." }, { "left": "issue:3689", "right": "issue:4876", "accept": false, - "reason": "Unloading LoRA weights and adding a std argument for LoRA layers are separate requests." + "reason": "LoRA unload behavior is unrelated to making LoRA std configurable." }, { "left": "issue:3876", "right": "issue:5558", "accept": false, - "reason": "Both mention ControlNet inpaint, but one is a demo error and the other is multi-ControlNet artifacts; not the same bug." + "reason": "Both involve ControlNet inpainting, but one is a demo error and the other is artifacts with multiple ControlNets; not the same concrete defect." }, { "left": "issue:4969", "right": "issue:514", "accept": false, - "reason": "CrossAttention import failure is unrelated to an attention slicing runtime error." + "reason": "Different import/error contexts: missing CrossAttention vs enable_attention_slicing() throwing an error." }, { "left": "issue:4969", "right": "issue:7177", "accept": false, - "reason": "CrossAttention import failure and StableCascadeDecoderPipeline.from_pretrained error are different issues." + "reason": "Missing CrossAttention import is unrelated to StableCascadeDecoderPipeline.from_pretrained failure." }, { "left": "issue:11497", "right": "issue:12547", "accept": false, - "reason": "Dataset preprocessing cache behavior and a missing --instance_prompt argument are unrelated training-script problems." + "reason": "Unrelated training-script issues: cache depending on irrelevant args vs missing required --instance_prompt argument." }, { "left": "issue:5646", "right": "issue:6598", "accept": false, - "reason": "Scheduler instability versus img2img output differences are not the same concrete bug." + "reason": "Scheduler stability on a specific model is unrelated to img2img result differences vs A1111." }, { "left": "issue:1214", "right": "issue:3767", "accept": false, - "reason": "Half-precision CLIPTextModel failure is unrelated to FrozenDict mutability." + "reason": "Half-precision CLIPTextModel failure is unrelated to FrozenDict mutability semantics." }, { "left": "issue:1922", "right": "issue:4210", "accept": false, - "reason": "An undefined notebook helper and attention_mask usage in UNet forward are unrelated." + "reason": "Notebook undefined function bug is unrelated to a question about attention_mask in UNet2DConditionModel.forward." }, { "left": "issue:2192", "right": "issue:6277", "accept": false, - "reason": "Missing CLI arguments in one training script and broken gradients in another are different issues." + "reason": "Missing dataset_name/caption_column arguments in one script is unrelated to broken gradients in a LoRA training example." }, { "left": "issue:3403", "right": "issue:776", "accept": false, - "reason": "TorchDynamo pipeline loading and image-to-image batch_size>1 failures are different code paths." + "reason": "TorchDynamo pipeline loading issues are not the same as batch_size>1 img2img breakage." }, { "left": "issue:6112", "right": "issue:6149", "accept": false, - "reason": "SDXL-turbo noise and ip-adapter load errors are unrelated." + "reason": "Noisy SDXL-turbo output is unrelated to an ip-adapter-full-face_sd15 load error." }, { "left": "issue:7939", "right": "issue:8228", "accept": false, - "reason": "Both concern loading models, but one is a generic same-folder load failure and the other is a specific from_single_file CLIPTextModel error." + "reason": "Both are loading-related, but one is loading a model from the same folder and the other is from_single_file failing to load CLIPTextModel; different code paths." }, { "left": "issue:1875", "right": "issue:6164", "accept": false, - "reason": "fp16 training type mismatch and latent output handling are distinct bugs." + "reason": "fp16 training type mismatch in Imagic is unrelated to latent output handling in StableDiffusionPipeline." }, { "left": "issue:3717", "right": "issue:4985", "accept": false, - "reason": "Both mention OOM, but they are different training scenarios and scripts." + "reason": "GPU OOM on multi-GPU hardware is unrelated to SDXL InstructPix2Pix training OOM; different scenario and likely different scaling cause." }, { "left": "issue:3383", "right": "issue:4665", "accept": false, - "reason": "Partially downloaded pipeline loading and Windows safetensors/bin selection are separate loading issues." + "reason": "Partial pipeline download handling is unrelated to Windows use_safetensors behavior with alternate .bin files." }, { "left": "issue:10290", "right": "issue:10953", "accept": false, - "reason": "The titles suggest a possible torch.compile-related connection, but the packet only gives titles and there is not enough evidence to confirm the same underlying bug or fix path." - }, - { - "left": "issue:10767", - "right": "issue:12221", - "accept": false, - "reason": "FlashVideo vs Diffusers and DiffSynth ControlNet support are different integration requests." - }, - { - "left": "issue:10860", - "right": "issue:9810", - "accept": false, - "reason": "LoRA usability in ComfyUI and module_to_save() support are related broadly, but not the same concrete bug." + "reason": "Torch.compile incompatibility for Flux+ControlNet is unrelated to a code-style/performance suggestion to use permute instead of transpose." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "Most soft-edge rejections are conservative and grounded in the packet. However, the accepted 10290/10953 pair is overstated: the packet only gives titles, and while they may be related, there is not enough evidence to confidently say they describe the same bug or the same torch.compile compatibility path. Please downgrade that verdict to conservative uncertainty rather than accept." - }, - "evaluator_used": true, - "retried": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -70698,14 +70712,16 @@ "nodes": [ "issue:10057", "issue:10165", - "issue:10226", "issue:10239", + "issue:10767", + "issue:10860", "issue:1111", "issue:11247", "issue:11592", "issue:11864", "issue:12108", "issue:12157", + "issue:12221", "issue:12338", "issue:1574", "issue:1790", @@ -70718,21 +70734,18 @@ "issue:4080", "issue:4366", "issue:4506", - "issue:4530", "issue:5225", "issue:5284", "issue:5395", "issue:5463", "issue:5489", "issue:6091", - "issue:6295", "issue:6335", "issue:6398", "issue:6452", "issue:6530", "issue:660", "issue:7009", - "issue:7138", "issue:7191", "issue:7228", "issue:7640", @@ -70742,9 +70755,12 @@ "issue:8879", "issue:8900", "issue:9370", - "issue:953" + "issue:953", + "issue:9810" ], "soft_pairs": [ + "issue:10767|issue:12221", + "issue:10860|issue:9810", "issue:2402|issue:660", "issue:4080|issue:6452", "issue:7640|issue:8879", @@ -70766,182 +70782,180 @@ "issue:2264|issue:3481", "issue:3655|issue:7191", "issue:4080|issue:4366", - "issue:10165|issue:8464", - "issue:10226|issue:7138", - "issue:4530|issue:6295" + "issue:10165|issue:8464" ], - "prepared_review_unit_hash": "3015dc1086d65351a88ef8a4d77d8e5f2fd796b3a1efd31800b85f8085e28329", + "prepared_review_unit_hash": "4853dfb235d9bf16ee3e99343ac6149984f6e2fa23a6153ddc95050ce37e2fb8", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33378, - "estimated_input_tokens": 8345, - "estimated_eval_tokens": 16946 + "serialized_chars": 33322, + "estimated_input_tokens": 8331, + "estimated_eval_tokens": 16918 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very heterogeneous: it mixes unrelated model-loading, training, memory, controlnet, and utility bugs. None of the soft pairs look like true duplicates, so I rejected them all. If forced to anchor the set, the most concrete issue is the FluxTransformer2DModel.from_single_file report, but it is not a real representative for the whole cluster.", - "confidence": 0.88, - "canonical_issue_reason": "Issue 10057 is the most concrete, reproducible bug report in the set (FluxTransformer2DModel.from_single_file token/loading path), but the surrounding items are too diverse for a true canonical duplicate.", + "summary": "This cluster is mostly noisy: the items share broad Diffusers/Flux/ControlNet/LoRA themes, but they describe different code paths, platforms, and failure modes rather than the same underlying bug or change.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 10057 is the most concrete, narrowly scoped bug report in the set (a specific FluxTransformer2DModel.from_single_file token problem) and is a reasonable cluster head, though the cluster itself is not truly coherent.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 10057 is the best single anchor because it is narrowly scoped and actionable; however, the cluster does not have a strong shared underlying bug, so this is only a weak representative.", + "best_issue_reason": "10057 is the clearest actionable issue: it names a specific model class and method, implies a reproducible loading path, and is more bug-like than the feature requests or broad compatibility reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10767", + "right": "issue:12221", + "accept": false, + "reason": "Both mention diffusers integrations, but one is a FlashVideo/diffusers topic and the other is a request to support DiffSynth Controlnet; different concrete work." + }, + { + "left": "issue:10860", + "right": "issue:9810", + "accept": false, + "reason": "One is a ComfyUI compatibility failure for trained LoRA weights, the other is a module_to_save() support request; related area, not the same bug." + }, { "left": "issue:2402", "right": "issue:660", "accept": false, - "reason": "Different backends/platforms: CUDA-not-compiled vs MPS fp16 on M1. Not the same bug." + "reason": "These are different backend/platform problems: CUDA-not-enabled vs MPS fp16 behavior on M1." }, { "left": "issue:4080", "right": "issue:6452", "accept": false, - "reason": "Both touch inpainting/controlnet, but one is a new SDXL inpaint pipeline and the other is a masked_image_latents API request. Not duplicates." + "reason": "Both concern inpainting, but one asks for a new SDXL inpainting pipeline while the other asks for a masked_image_latents input on ControlNet inpaint; not the same change." }, { "left": "issue:7640", "right": "issue:8879", "accept": false, - "reason": "Different failures: from_single_file shape/dimension mismatch vs torch.compile+cudagraph on SDXL." + "reason": "Both involve StableDiffusionXL, but the failures are different: from_single_file dimension handling vs torch.compile/cudagraph issues." }, { "left": "issue:11592", "right": "issue:8900", "accept": false, - "reason": "LoRA adapter deletion regression and SDXL refiner loading guidance are unrelated." + "reason": "Unrelated topics: a FluxLoraLoaderMixin regression versus a question about loading an SDXL refiner checkpoint." }, { "left": "issue:5395", "right": "issue:5489", "accept": false, - "reason": "Same broad PEFT area, but one is a specific LoRA-combination bug and the other is an integration announcement. Not the same concrete issue." + "reason": "One is a specific bug combining Civitai LoRAs with PEFT; the other is a broad PEFT x diffusers integration umbrella issue." }, { "left": "issue:12108", "right": "issue:12157", "accept": false, - "reason": "Both Qwen-Image related, but one is scheduler/flow-matching breakage and the other is a bnb_quantization_config_path training error." + "reason": "Different problem types: inference breakage with flow-matching schedulers vs a training-time quantization-config error." }, { "left": "issue:1574", "right": "issue:1790", "accept": false, - "reason": "Dreambooth training bug vs a general training question. No duplicate relation." + "reason": "A Dreambooth GPU failure is not the same as a general question about how training works." }, { "left": "issue:4506", "right": "issue:5463", "accept": false, - "reason": "Different variables and code paths: prompt_embeds overwrite vs image_latents UnboundLocalError." + "reason": "Both are variable-scope bugs, but they affect different variables, code paths, and symptoms; not the same defect." }, { "left": "issue:6091", "right": "issue:6398", "accept": false, - "reason": "Consistency decoder size-argument memory issue and a vague memory leak report are not the same bug." + "reason": "One is a specific Consistency Decoder memory issue tied to size arguments; the other is an unspecific memory-leak report." }, { "left": "issue:6335", "right": "issue:7881", "accept": false, - "reason": "ControlNet img2img size control and lpw_stable_diffusion from_single_file failure are unrelated." + "reason": "Different pipeline problems: output-size handling in ControlNet Img2Img vs from_single_file loading for lpw_stable_diffusion." }, { "left": "issue:10239", "right": "issue:7009", "accept": false, - "reason": "Different pipelines and symptoms: cpu offload support vs poor generated results." + "reason": "Both are video-related, but one is about sequential CPU offload support and the other about poor results from a specific TextToVideoSD model load." }, { "left": "issue:11864", "right": "issue:12338", "accept": false, - "reason": "Same class name, but different failures: torch.compile name error vs VAE slicing/multi-image decoding bug." + "reason": "Same AutoencoderDC area, but different bugs: torch.compile fullgraph name resolution vs VAE slicing when decoding multiple images." }, { "left": "issue:3651", "right": "issue:3683", "accept": false, - "reason": "Multi-GPU Dreambooth training and Multi-ControlNet save_pretrained are unrelated." + "reason": "Multi-GPU Dreambooth training and Multi-ControlNet save_pretrained() are unrelated changes." }, { "left": "issue:7228", "right": "issue:7781", "accept": false, - "reason": "Traced-model dtype attribute error vs PIL image serialization error. Different subsystems." + "reason": "A traced-model dtype attribute error and a PIL serialization error are distinct failure modes." }, { "left": "issue:10057", "right": "issue:11247", "accept": false, - "reason": "Both mention FluxTransformer2DModel, but one is a from_single_file token/loading issue and the other is multi-GPU/controlnet distribution." + "reason": "Both mention FluxTransformer2DModel, but one is a from_single_file token issue and the other is multi-GPU/controlnet distribution; different code paths." }, { "left": "issue:5225", "right": "issue:9370", "accept": false, - "reason": "VRAM usage problem vs missing pooled_projection argument. Not the same defect." + "reason": "A VRAM/memory limitation is not the same as a missing pooled_projection argument in a forward method." }, { "left": "issue:1111", "right": "issue:953", "accept": false, - "reason": "Checkpoint content/compliance discussion vs image size guidance. Completely unrelated." + "reason": "These are unrelated: a checkpoint dataset constraint versus guidance on generating smaller images." }, { "left": "issue:5284", "right": "issue:6530", "accept": false, - "reason": "ControlNet-LLLite feature request vs SDXL k-diffusion controlnet bug. Different topics." + "reason": "ControlNet-LLLite support is a feature request; SDXL K-diffusion with ControlNet is a separate pipeline compatibility issue." }, { "left": "issue:2264", "right": "issue:3481", "accept": false, - "reason": "Seed-setting error in one pipeline vs expected image variation across inference steps in another." + "reason": "Both involve seeding/determinism, but one is an exception when setting the seed and the other is expected image variation across inference steps." }, { "left": "issue:3655", "right": "issue:7191", "accept": false, - "reason": "Multi-GPU Dreambooth failure and a generic import problem are unrelated." + "reason": "Both are import-related, but the titles do not indicate the same module, backend, or failure mechanism." }, { "left": "issue:4080", "right": "issue:4366", "accept": false, - "reason": "New SDXL inpainting pipeline request vs fine-tuning full U-Net script; not duplicates." + "reason": "An SDXL inpainting pipeline request is not the same as fine-tuning the SDXL text-to-image script for a full U-Net." }, { "left": "issue:10165", "right": "issue:8464", "accept": false, - "reason": "Both mention seeds, but one is LoRA scale changing seed preservation and the other is cpu offload with seed." - }, - { - "left": "issue:10226", - "right": "issue:7138", - "accept": false, - "reason": "Both involve Flux/SDXL shape mismatch themes, but the models and failure modes differ; not the same concrete bug." - }, - { - "left": "issue:4530", - "right": "issue:6295", - "accept": false, - "reason": "Different artifact reports in different samplers/settings; too broad to be duplicates." + "reason": "Both mention LoRA/seed behavior, but one is about preserving seeds when changing LoRA scale and the other is a SEGA CPU-offload seed failure." } ] }, @@ -70965,6 +70979,7 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10226", "issue:10381", "issue:10447", "issue:10453", @@ -70981,13 +70996,12 @@ "issue:2437", "issue:2669", "issue:3455", - "issue:3658", "issue:4139", "issue:4149", "issue:4353", "issue:4406", "issue:4410", - "issue:4975", + "issue:4530", "issue:4991", "issue:5004", "issue:5070", @@ -70998,9 +71012,11 @@ "issue:5783", "issue:6016", "issue:6037", + "issue:6295", "issue:6496", "issue:6518", "issue:6713", + "issue:7138", "issue:7147", "issue:7450", "issue:7600", @@ -71010,10 +71026,11 @@ "issue:8612", "issue:8918", "issue:8939", - "issue:9090", "issue:9450" ], "soft_pairs": [ + "issue:10226|issue:7138", + "issue:4530|issue:6295", "issue:10453|issue:12540", "issue:13015|issue:13298", "issue:12321|issue:12437", @@ -71035,182 +71052,180 @@ "issue:5685|issue:7450", "issue:10743|issue:10925", "issue:2669|issue:6518", - "issue:1101|issue:5070", - "issue:3658|issue:7600", - "issue:4975|issue:9090" + "issue:1101|issue:5070" ], - "prepared_review_unit_hash": "e6a5cc977cb1be6b1caddb9523defc4592a33bbb7332c4b52fc454401087f56b", + "prepared_review_unit_hash": "c8817a99115052b8d555e037601514fe083dae8d44827a311e03bb8954e6cc69", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33648, - "estimated_input_tokens": 8412, - "estimated_eval_tokens": 17080 + "serialized_chars": 34196, + "estimated_input_tokens": 8549, + "estimated_eval_tokens": 17354 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous. The strongest common thread is Flux/device-placement inference bugs, but most pair candidates are only superficially similar and should not be merged.", - "confidence": 0.78, - "canonical_issue_reason": "issue:9450 is the clearest, most concrete Flux bug report in the set: a specific multi-GPU tensor device mismatch with a reproducible trigger (`transformer=`).", + "summary": "This cluster is mostly heterogeneous, with a small Flux/device-mismatch subgroup but many unrelated training, conversion, scheduler, and feature-request issues. No soft pairs look like safe duplicates.", + "confidence": 0.36, + "canonical_issue_reason": "issue:9450 is the clearest and broadest Flux device-mismatch report in the set, so it is the best anchor for the small overlapping subgroup.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9450 is the best representative issue because it isolates one concrete code-path failure and has the most discussion around an actionable bug, unlike broader mixed reports such as 10381.", + "best_issue_reason": "issue:9450 is the strongest triage representative: explicit FluxPipeline device mismatch, concrete error text, and broader applicability than the narrower Flux/LoRA variants.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10226", + "right": "issue:7138", + "accept": false, + "reason": "Both mention model-shape mismatch, but one is Dreambooth Flux+LoRA and the other is SDXL UNet structure mismatch; different models and likely different bugs." + }, + { + "left": "issue:4530", + "right": "issue:6295", + "accept": false, + "reason": "Both are artifact-quality complaints, but the symptoms and likely causes differ: SDXL refiner color artifacts vs DPM++ 2M SDE Karras artifacts." + }, { "left": "issue:10453", "right": "issue:12540", "accept": false, - "reason": "Different bugs and models: Hunyuan Video batch-size limitation vs SanaPipeline embedding integration failure." + "reason": "Different pipelines and failures: Hunyuan Video batch-size limitation vs SanaCombinedTimestepGuidanceEmbeddings incompatibility with SanaPipeline." }, { "left": "issue:13015", "right": "issue:13298", "accept": false, - "reason": "Both are device-mismatch errors, but one is QwenImageEditPlus on NPU and the other is a generic CUDA generator issue; not the same bug." + "reason": "Both are device-mismatch errors, but one is QwenImageEditPlus on NPU and the other is Flux image generation with a CUDA generator; not the same concrete code path." }, { "left": "issue:12321", "right": "issue:12437", "accept": false, - "reason": "Separate feature requests for different models (HunyuanImage-2.1 vs OmniSVG)." + "reason": "Separate feature requests for different models: HunyuanImage-2.1 support vs OmniSVG support." }, { "left": "issue:1647", "right": "issue:3455", "accept": false, - "reason": "Unrelated topics: color-correction .pt usage question vs instruct-pix2pix training behavior." + "reason": "A usage question about color-correction .pt files is unrelated to training InstructPix2Pix with a pretrained model." }, { "left": "issue:2203", "right": "issue:761", "accept": false, - "reason": "Model download location question vs 8-bit Adam training failure; different failure modes." + "reason": "Different problems: download location/path handling vs DreamBooth failure with 8-bit Adam." }, { "left": "issue:2437", "right": "issue:6037", "accept": false, - "reason": "Both are pipeline additions, but for unrelated architectures and use cases." + "reason": "Different pipeline additions: T2I-Adapter vs InstaFlow." }, { "left": "issue:4139", "right": "issue:8612", "accept": false, - "reason": "Different precision-related problems in different components/models, not the same code-path." + "reason": "Both involve precision, but they affect different models and code paths: custom diffusion attention processor casting vs PixArt-sigma mixed-precision inference." }, { "left": "issue:4149", "right": "issue:5509", "accept": false, - "reason": "Both involve batch-shape mismatches, but in different pipelines and input types; not clearly the same bug." + "reason": "Both are batch-dimension mismatches, but they occur in different pipelines and inputs; not enough evidence of one underlying bug." }, { "left": "issue:4991", "right": "issue:5405", "accept": false, - "reason": "Training dataloader preparation bug vs single-file SDXL weight overwrite bug; unrelated." + "reason": "Training setup bug vs single-file SDXL weight-overwrite issue; unrelated failures." }, { "left": "issue:5004", "right": "issue:7600", "accept": false, - "reason": "Both are DreamBooth-related, but one is checkpoint resume at fp16 and the other is missing accumulate context on text_encoder." + "reason": "Checkpoint-resume fp16 bug and missing accelerate.accumulate on text_encoder are distinct training issues." }, { "left": "issue:5271", "right": "issue:6496", "accept": false, - "reason": "Different DreamBooth enhancements: learning-rate control vs multi-person example support." + "reason": "Different DreamBooth requests: separate learning rates vs multiple people/faces example support." }, { "left": "issue:5783", "right": "issue:8918", "accept": false, - "reason": "Separate SDXL EMA issue vs SD3 DreamBooth script breakage." + "reason": "SDXL EMA handling and a broken SD3 DreamBooth script are unrelated training bugs." }, { "left": "issue:7147", "right": "issue:7856", "accept": false, - "reason": "Import error for StableCascadeDecoderPipeline vs tokenizer multi-vector token conflict; unrelated." + "reason": "Import error for StableCascadeDecoderPipeline vs tokenizer multi-vector token collision; different subsystems." }, { "left": "issue:10381", "right": "issue:9450", "accept": false, - "reason": "Both mention Flux/device issues, but 10381 mixes quantization, LoRA, and Quanto/ZeroGPU concerns; it is not the same concrete multi-GPU bug as 9450." + "reason": "Both are Flux/device-related, but 10381 centers on quantization/LoRA/Quanto and zero-GPU spaces, while 9450 is a multi-GPU transformer device-placement issue." }, { "left": "issue:4410", "right": "issue:6016", "accept": false, - "reason": "Both are conversion/load errors, but they point to different failure shapes and code paths." + "reason": "Both involve conversion/loading shape problems, but the specific failures differ: meta-device conv_in.weight mismatch vs a torch.Size() issue in the converter." }, { "left": "issue:10447", "right": "issue:8939", "accept": false, - "reason": "Converter splitting UNet into two files vs tokenizer loading/path error; unrelated." + "reason": "Different issues in different tools: UNet split-file conversion breakage vs SD3 tokenizer load failure." }, { "left": "issue:4353", "right": "issue:7798", "accept": false, - "reason": "Different model quality/bug reports with no shared underlying code-path." + "reason": "Unrelated quality bug reports for different pipelines: SDXL refiner artifacts vs PixArtAlphaPipeline bug." }, { "left": "issue:4406", "right": "issue:6713", "accept": false, - "reason": "Both mention HeaderTooLarge, but one is a generic safetensor deserialization error and the other is converter-script specific; not enough to treat as the same bug." + "reason": "Same error class appears in the title, but one is a generic safetensor deserialization issue and the other is specific to the conversion script; not enough evidence they are the same underlying bug." }, { "left": "issue:5685", "right": "issue:7450", "accept": false, - "reason": "Scheduler concerns vs AutoencoderKL multi-threading; different subsystems." + "reason": "Both mention scheduler or threading concerns, but they are different issues: scheduler problems vs AutoencoderKL multi-threading." }, { "left": "issue:10743", "right": "issue:10925", "accept": false, - "reason": "Both are FLUX training related, but one asks for ZeRO-3 support and the other is a DreamBooth fine-tuning issue." + "reason": "Flux zero-3 training support and Dreambooth finetuning FLUX CLIPTextModel are related to Flux training but are not the same change or bug." }, { "left": "issue:2669", "right": "issue:6518", "accept": false, - "reason": "Both are NoneType-related exceptions, but they arise from different scripts and operations." + "reason": "Different NoneType errors in different examples/scripts; similar exception shape but not the same root cause." }, { "left": "issue:1101", "right": "issue:5070", "accept": false, - "reason": "Scheduler feature request vs ControlNet example flag changes; not duplicates." - }, - { - "left": "issue:3658", - "right": "issue:7600", - "accept": false, - "reason": "Both DreamBooth training issues, but they affect different parts of the training loop and are not the same bug." - }, - { - "left": "issue:4975", - "right": "issue:9090", - "accept": false, - "reason": "LoRA loading inefficiency vs PNDMScheduler timestep repetition; unrelated." + "reason": "DPM-Solver scheduler addition and a ControlNet example tweak are unrelated." } ] }, @@ -71229,8 +71244,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -71245,25 +71260,25 @@ "issue:2784", "issue:3116", "issue:3381", + "issue:3658", "issue:3951", "issue:4171", "issue:444", "issue:4453", "issue:4664", - "issue:4788", "issue:4885", + "issue:4975", "issue:5062", "issue:5079", - "issue:5480", "issue:5485", "issue:5798", "issue:5969", "issue:6176", "issue:6454", - "issue:6616", "issue:6790", "issue:696", "issue:7454", + "issue:7600", "issue:7690", "issue:7798", "issue:7801", @@ -71273,7 +71288,7 @@ "issue:9044", "issue:9063", "issue:9084", - "issue:9170", + "issue:9090", "issue:9181", "issue:9204", "issue:9332", @@ -71284,6 +71299,8 @@ "issue:9665" ], "soft_pairs": [ + "issue:3658|issue:7600", + "issue:4975|issue:9090", "issue:5079|issue:7454", "issue:6176|issue:9505", "issue:6454|issue:7801", @@ -71305,182 +71322,180 @@ "issue:2276|issue:696", "issue:5798|issue:9181", "issue:3951|issue:5969", - "issue:4171|issue:4453", - "issue:4788|issue:9170", - "issue:5480|issue:6616" + "issue:4171|issue:4453" ], - "prepared_review_unit_hash": "1f8225e370546104d67dd2d6d8fc4ec3161e68f396d777c67e16863466b86373", + "prepared_review_unit_hash": "e8cd540d4d8989f0a00452487d50bc525660a2e1de90339e8475d414de1565de", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33702, - "estimated_input_tokens": 8426, - "estimated_eval_tokens": 17108 + "serialized_chars": 33814, + "estimated_input_tokens": 8454, + "estimated_eval_tokens": 17164 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very heterogeneous: most soft pairs are only weakly text-similar and describe different pipelines, bugs, or feature requests. I would not merge any of the suggested pairs as duplicates. The closest thing to a canonical issue is the scheduler/indexing bug around num_inference_steps.", - "confidence": 0.29, - "canonical_issue_reason": "issue:444 is the clearest concrete bug report in the set and is the most general scheduler/index error description; it best matches the only nearby duplicate-like report (issue:1033).", + "summary": "This cluster is highly heterogeneous: most items are isolated bugs/feature requests across different pipelines, schedulers, and training scripts. I don\u2019t see any soft pair that is clearly the same underlying bug or change, so no merges are warranted.", + "confidence": 0.9, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:444 is the best representative for a duplicate cluster because it describes a repeatable, specific runtime error with broader applicability than the other isolated reports.", + "best_issue_reason": "If one issue must anchor the cluster, #9332 is the most substantial and detailed, but it is still not a true duplicate hub for the rest of the items.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3658", + "right": "issue:7600", + "accept": false, + "reason": "Both are DreamBooth-related, but one is about extra noise latents in upsampler scripts and the other about missing accelerator.accumulate on text_encoder; different bugs." + }, + { + "left": "issue:4975", + "right": "issue:9090", + "accept": false, + "reason": "LoRA loading inefficiency and repeated PNDM timesteps are unrelated subsystems and failure modes." + }, { "left": "issue:5079", "right": "issue:7454", "accept": false, - "reason": "Different requests: loading LyCORIS via load_lora_weights vs generic LoRA support. Not the same bug/change." + "reason": "One requests load_lora_weights support for LyCORIS, the other is a generic LORA support request; similar area, not the same change." }, { "left": "issue:6176", "right": "issue:9505", "accept": false, - "reason": "Both involve LoRA/upscaling workflows, but the failure modes are unrelated: loading LoRA after training vs a missing key during upscaling." + "reason": "DreamBooth LoRA load failure and an upscaler KeyError involve different pipelines and error paths." }, { "left": "issue:6454", "right": "issue:7801", "accept": false, - "reason": "Training gradient scaling error vs mixed_precision not applying; related area, but not the same concrete problem." + "reason": "Both mention fp16, but one is an AMP gradient unscale error and the other is about mixed_precision not producing fp16 weights." }, { "left": "issue:6790", "right": "issue:9204", "accept": false, - "reason": "Different model/training failures: LCM state_dict key mismatch vs a FLUX DreamBooth training script issue." + "reason": "State-dict key mismatch in LCM training is not the same as the FLUX DreamBooth training-script issue." }, { "left": "issue:7798", "right": "issue:9084", "accept": false, - "reason": "Different pipelines and symptoms: PixArtAlphaPipeline bug vs SD3 inpaint scheduler step-order problem." + "reason": "PixArtAlphaPipeline bug and SD3 inpaint EulerA scheduler behavior are different code paths." }, { "left": "issue:11642", "right": "issue:9044", "accept": false, - "reason": "Both mention loading/model behavior, but one is prompt_embeds_scale having no effect and the other is a missing tokenizer/model file error." + "reason": "Prompt embedding scaling not taking effect and SDXL model-file load failure are unrelated problems." }, { "left": "issue:9484", "right": "issue:9665", "accept": false, - "reason": "Both are training/deepspeed-related, but one asks for multigpu FLUX DreamBooth while the other reports accelerator.prepare() behavior under a DeepSpeed config." + "reason": "Both involve training/DeepSpeed, but one is FLUX DreamBooth multigpu usage and the other is accelerator.prepare interaction; not the same concrete bug." }, { "left": "issue:9350", "right": "issue:9362", "accept": false, - "reason": "Both are index-related, but they refer to different tensors, different code paths, and different error conditions." + "reason": "A deprecation warning about txt_ids shape is not the same as an index-out-of-bounds runtime error." }, { "left": "issue:1033", "right": "issue:444", "accept": false, - "reason": "Both are IndexError reports, but they are not clearly the same code path; one is stable_diffusion_mega-specific and the other is a num_inference_steps pattern bug." + "reason": "Both are IndexErrors in diffusion contexts, but the described conditions and pipelines differ too much to treat as one bug." }, { "left": "issue:11000", "right": "issue:9027", "accept": false, - "reason": "Missing shift argument in a video pipeline vs argument-name consistency feedback; not the same bug/change." + "reason": "Missing shift argument in WanImageToVideoPipeline is unrelated to argument-name consistency complaints." }, { "left": "issue:1992", "right": "issue:5062", "accept": false, - "reason": "Unrelated topics: JAX attention refactoring discussion vs a typo in attention code." + "reason": "JAX attention refactoring discussion and a processor typo are not duplicate bugs." }, { "left": "issue:2202", "right": "issue:2784", "accept": false, - "reason": "General OpenVINO support question vs DreamBooth support request for text2video models; not duplicates." + "reason": "OpenVINO support progress and DreamBooth support for text2video are different feature requests." }, { "left": "issue:8535", "right": "issue:8557", "accept": false, - "reason": "Both are SD3-related, but one is xformers memory-efficient attention and the other is token limit in training." + "reason": "xformers attention support in SD3 and SD3 training token-limit issues are separate concerns." }, { "left": "issue:4664", "right": "issue:4885", "accept": false, - "reason": "Both concern SDXL LoRA loading, but one is a pipeline naming/loading discrepancy and the other is ControlNet img2img not loading weights; not clearly the same bug." + "reason": "Both are LoRA-loading reports, but they target different pipelines and symptoms; not clearly one underlying defect." }, { "left": "issue:1284", "right": "issue:5485", "accept": false, - "reason": "Batch size slowdown vs ControlNet batch inference taking forever are too broad and not the same underlying issue." + "reason": "General batch-size slowdown is not the same as ControlNet batch inference taking forever." }, { "left": "issue:11432", "right": "issue:9332", "accept": false, - "reason": "Both are from_pretrained/loading-related, but one is torch_dtype='auto' behavior and the other is identical weights loading the wrong specified model." + "reason": "Both concern from_pretrained behavior, but one is torch_dtype=\"auto\" handling and the other is loading the wrong/identical weights; different bugs." }, { "left": "issue:7690", "right": "issue:9063", "accept": false, - "reason": "Both are model/tokenizer loading errors for SDXL/Flux, but they are different missing-resource and tokenizer-construction failures." + "reason": "Tokenizer-loading failures in different models/pipelines are not the same concrete issue." }, { "left": "issue:3116", "right": "issue:3381", "accept": false, - "reason": "Different pipeline bugs: KDPM2DiscreteScheduler latent noise in img2img vs discolored inpaint batch inference/docs issue." + "reason": "A scheduler latent-noise bug in image2image is not the same as discolored inpaint batch inference output." }, { "left": "issue:2276", "right": "issue:696", "accept": false, - "reason": "Dreambooth/xformers installation problem vs CUDA OOM training issue; not the same defect." + "reason": "xformers installation problems and DreamBooth CUDA OOM are unrelated." }, { "left": "issue:5798", "right": "issue:9181", "accept": false, - "reason": "Both involve ControlNet, but one is an unexpected callback keyword argument and the other is a visual noise-dot artifact." + "reason": "callback_on_step_end API error and strange noise dots in ControlNet do not share a single code-path fix." }, { "left": "issue:3951", "right": "issue:5969", "accept": false, - "reason": "NaN outputs in a demo vs DDIM eta=1 NaNs with ZSNR patch; superficially similar symptom, but different code paths and triggers." + "reason": "Both mention NaNs, but they arise from different models/settings and are not clearly the same defect." }, { "left": "issue:4171", "right": "issue:4453", "accept": false, - "reason": "Documentation/reference page rendering issue vs SDXL LoRA conversion question; unrelated." - }, - { - "left": "issue:4788", - "right": "issue:9170", - "accept": false, - "reason": "Both are memory-related, but one is SDXL inpaint OOM on 6 GB VRAM and the other is a general question about whether SDXL+ControlNet needs >36 GB." - }, - { - "left": "issue:5480", - "right": "issue:6616", - "accept": false, - "reason": "EDM implementation request vs classifier-guidance implementation request; different feature work." + "reason": "Documentation rendering and LoRA conversion are completely unrelated." } ] }, @@ -71499,18 +71514,16 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10016", "issue:10287", "issue:10518", - "issue:10917", "issue:11041", "issue:11093", - "issue:11214", "issue:11387", "issue:11497", "issue:11907", @@ -71529,22 +71542,24 @@ "issue:3420", "issue:4497", "issue:4676", + "issue:4788", "issue:4880", "issue:5226", "issue:5379", "issue:5399", + "issue:5480", "issue:5515", "issue:6054", "issue:6167", "issue:6548", + "issue:6616", "issue:6700", "issue:7220", "issue:759", "issue:7770", "issue:7836", "issue:8543", - "issue:8887", - "issue:9027", + "issue:9170", "issue:9279", "issue:9291", "issue:9448", @@ -71554,6 +71569,8 @@ "issue:9895" ], "soft_pairs": [ + "issue:4788|issue:9170", + "issue:5480|issue:6616", "issue:7836|issue:8543", "issue:5399|issue:7770", "issue:10287|issue:9548", @@ -71575,182 +71592,180 @@ "issue:11041|issue:9895", "issue:10016|issue:12423", "issue:4676|issue:6548", - "issue:10518|issue:12905", - "issue:10917|issue:8887", - "issue:11214|issue:9027" + "issue:10518|issue:12905" ], - "prepared_review_unit_hash": "f48f969aac5aae7457c77238d9dad407d749944da09bc32a275d3e2a123e1283", + "prepared_review_unit_hash": "1afda699b9ed473b16009c6f9775dd78dfee54259b3406076f5aad34f63cc846", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34295, - "estimated_input_tokens": 8574, - "estimated_eval_tokens": 17404 + "serialized_chars": 34265, + "estimated_input_tokens": 8567, + "estimated_eval_tokens": 17390 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is a loose collection of unrelated issue reports across different models, pipelines, schedulers, and training scripts. None of the soft-edge pairs look like true duplicates or the same underlying fix.", + "summary": "This cluster is heterogeneous: most items are unrelated pipeline/example/model bugs, so I do not see any true duplicate pairs to merge. The only reasonable representative issue is the specific SD3.5M from_single_file shape-mismatch crash, but the rest of the set spans distinct code paths and features.", "confidence": 0.92, - "canonical_issue_reason": "Issue #10016 is the most concrete and self-contained bug report: a specific `from_single_file` shape mismatch on SD3.5M with a clear error target (`pos_embed.pos_embed`).", + "canonical_issue_reason": "Issue 10016 is the most concrete and actionable report: it names a specific model, a specific API (`from_single_file`), and a precise failure (`mismatched shape for pos_embed.pos_embed`).", "canonical_pr_reason": null, - "best_issue_reason": "Issue #10016 is the best standalone representative because it describes a precise, reproducible loading failure rather than a broad or ambiguous problem.", + "best_issue_reason": "Issue 10016 is the best single representative because it is narrowly scoped, reproducible, and clearly describes the affected loading path and error.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4788", + "right": "issue:9170", + "accept": false, + "reason": "Both are memory-related, but one is SDXL inpaint OOM and the other is a ControlNet+SDXL VRAM requirement question; different symptom and likely different root cause." + }, + { + "left": "issue:5480", + "right": "issue:6616", + "accept": false, + "reason": "These are separate feature requests: EDM implementation vs classifier-guidance implementation." + }, { "left": "issue:7836", "right": "issue:8543", "accept": false, - "reason": "Both are training-example complaints, but one is about CLIP overfitting and the other about SD3 gradient checkpointing failure; different bugs and fixes." + "reason": "Different example bugs in different training scripts; overfitting in CLIP training is not the same as gradient checkpointing failure in SD3 dreambooth." }, { "left": "issue:5399", "right": "issue:7770", "accept": false, - "reason": "One is a dtype mismatch in custom inpainting inference; the other is a model/LoRA tensor shape mismatch during loading. Same broad area, not the same bug." + "reason": "Both mention loading/inference errors, but one is a dtype mismatch in custom inpainting inference and the other is a model+LoRA shape mismatch; not the same bug." }, { "left": "issue:10287", "right": "issue:9548", "accept": false, - "reason": "A documentation example issue is unrelated to a Flux DreamBooth LoRA training regression." + "reason": "A docs-example issue and a Flux dreambooth LoRA training regression are unrelated." }, { "left": "issue:12895", "right": "issue:9448", "accept": false, - "reason": "Both mention pipeline runtime errors, but the concrete failures differ: `NoneType.ndim` with batched prompts versus IP-Adapter tuple shape handling." + "reason": "Different pipelines and failures: batched prompts with `num_images_per_prompt` versus IP-Adapter tuple/shape handling in ControlNet inpaint." }, { "left": "issue:1566", "right": "issue:2091", "accept": false, - "reason": "Checkpoint-saving fp16 crash and checkpoint-resume multi-GPU issues are different training-path problems." + "reason": "One is fp16 checkpoint saving in Dreambooth; the other is resuming multi-GPU training from checkpoint. Same general area, but not the same concrete bug." }, { "left": "issue:11093", "right": "issue:9680", "accept": false, - "reason": "TorchAO example breakage is not the same as SDXL unconditional generation being broken; different code paths and symptoms." + "reason": "Different broken examples/pipelines with no shared code-path failure." }, { "left": "issue:170", "right": "issue:2083", "accept": false, - "reason": "Both involve `eta`, but one is an unsupported scheduler argument error and the other is a multi-generator/DDIM edge case." + "reason": "Both involve eta, but one is an unexpected kwarg crash and the other is a multiple-generator/DDIMScheduler limitation; not the same underlying bug." }, { "left": "issue:4497", "right": "issue:759", "accept": false, - "reason": "Dynamic thresholding and textual inversion save behavior are unrelated features." + "reason": "Dynamic thresholding and optional full-model saving for textual inversion are unrelated." }, { "left": "issue:9279", "right": "issue:9291", "accept": false, - "reason": "Dynamic LoRA loading with `torch.compile` is unrelated to FLUX LoRA key compatibility." + "reason": "Dynamic LoRA loading with `torch.compile` and FLUX LoRA key compatibility are different LoRA-related issues with different code paths." }, { "left": "issue:11497", "right": "issue:9731", "accept": false, - "reason": "Dataset cache invalidation in SD3 ControlNet training is not the same as a Playground2.5 LoRA usage question." + "reason": "Dataset preprocessing cache invalidation and a Playground2.5 LoRA training question are unrelated." }, { "left": "issue:5226", "right": "issue:6167", "accept": false, - "reason": "Both touch StableDiffusionXLAdapterPipeline, but one is about `load_lora_weights` and the other about `load_ip_adapter`; different integration points." + "reason": "`load_lora_weights` support and `load_ip_adapter` addition are different adapter-loading features." }, { "left": "issue:11387", "right": "issue:12338", "accept": false, - "reason": "AutoencoderDC RMSNorm type selection is unrelated to the VAE slicing multi-image decoding bug." + "reason": "Both touch AutoencoderDC, but one is a normalization-layer bug and the other is a slicing/multi-image decoding bug." }, { "left": "issue:2495", "right": "issue:3116", "accept": false, - "reason": "A black-image MPS issue is not the same as latent noise appearing in KDPM2DiscreteScheduler img2img behavior." + "reason": "MPS black-image output and KDPM2DiscreteScheduler latent noise are distinct failures." }, { "left": "issue:5379", "right": "issue:5515", "accept": false, - "reason": "A community SpeechToImage pipeline and an LCM pipeline scheduler initialization bug are unrelated." + "reason": "A community pipeline request and an LCM pipeline scheduler initialization failure are unrelated." }, { "left": "issue:3420", "right": "issue:7220", "accept": false, - "reason": "ControlNetInpaint failing in a diffusion pipeline context is different from `.unload_ip_adapter()` returning an error." + "reason": "Both mention ControlNet/IP-Adapter plumbing, but one is a diffusion-pipeline compatibility issue and the other is an `.unload_ip_adapter()` error; different bugs." }, { "left": "issue:1842", "right": "issue:6054", "accept": false, - "reason": "Unexpected `eta` keyword errors and `torch.from_numpy` misuse in `set_timesteps()` are distinct scheduler bugs." + "reason": "Both mention scheduler/eta behavior, but one is a generic unexpected-eta kwarg crash and the other is a `torch.from_numpy` misuse in EulerDiscreteScheduler." }, { "left": "issue:4880", "right": "issue:6700", "accept": false, - "reason": "Single-ControlNet image input validation and attention processor `temb` argument passing are different TypeError causes." + "reason": "Single-ControlNet image input validation and `temb` forwarding to attention processors are unrelated TypeError sources." }, { "left": "issue:11907", "right": "issue:12096", "accept": false, - "reason": "Flux scheduler incompatibility and WanVACE group-offloading failure are separate model/runtime issues." + "reason": "Flux scheduler incompatibility and WanVACEPipeline group-offloading failure are different pipelines and different code paths." }, { "left": "issue:11041", "right": "issue:9895", "accept": false, - "reason": "WAN2.1 group-offloading errors and a FluxPipeline device-placement regression are not the same bug." + "reason": "Wan2.1 group-offloading errors and Flux device-placement errors are unrelated." }, { "left": "issue:10016", "right": "issue:12423", "accept": false, - "reason": "Both are `from_single_file` loading reports, but they involve different models and different failure modes." + "reason": "Both are `from_single_file` loading reports, but they affect different model families and different tensor shapes; not the same bug." }, { "left": "issue:4676", "right": "issue:6548", "accept": false, - "reason": "TAESD latent darkness and RGBA image handling in `prepare_latents` are different latent-processing bugs." + "reason": "TAESD latents being too dark and RGBA images being misread as latents are different preprocessing/encoding problems." }, { "left": "issue:10518", "right": "issue:12905", "accept": false, - "reason": "An SD3 LoRA training script mistake is unrelated to Z_Image guidance-scale implementation." - }, - { - "left": "issue:10917", - "right": "issue:8887", - "accept": false, - "reason": "A question about the lumina-2.0 script is not the same as an SD3 training-script VAE encode bug." - }, - { - "left": "issue:11214", - "right": "issue:9027", - "accept": false, - "reason": "An unexpected `encoder_hidden_states` keyword error is not the same as a general naming-consistency discussion." + "reason": "A LoRA training example script issue and a guidance-scale implementation bug are unrelated." } ] }, @@ -71776,19 +71791,17 @@ "nodes": [ "issue:10223", "issue:10281", - "issue:10302", + "issue:10917", "issue:11127", + "issue:11214", "issue:11387", "issue:11859", - "issue:12052", "issue:1444", "issue:1543", - "issue:1672", "issue:1695", "issue:1979", "issue:2083", "issue:2265", - "issue:2722", "issue:3177", "issue:3249", "issue:3583", @@ -71817,6 +71830,8 @@ "issue:6906", "issue:6921", "issue:7456", + "issue:8887", + "issue:9027", "issue:9134", "issue:9248", "issue:9485", @@ -71824,6 +71839,8 @@ "issue:9996" ], "soft_pairs": [ + "issue:10917|issue:8887", + "issue:11214|issue:9027", "issue:1444|issue:5084", "issue:1979|issue:4000", "issue:2083|issue:579", @@ -71845,182 +71862,180 @@ "issue:4331|issue:5014", "issue:10281|issue:11859", "issue:1695|issue:3177", - "issue:5026|issue:5875", - "issue:1672|issue:2722", - "issue:10302|issue:12052" + "issue:5026|issue:5875" ], - "prepared_review_unit_hash": "92e2bea864cd221ed1a2bfd45c914dedea94d1f728f9ce6604250646220ebddc", + "prepared_review_unit_hash": "e75f45c1481c4fca8fe2f0cb46b5eda2a818a749c0f6317fdd0f8550506c6762", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33660, - "estimated_input_tokens": 8415, - "estimated_eval_tokens": 17086 + "serialized_chars": 33531, + "estimated_input_tokens": 8383, + "estimated_eval_tokens": 17022 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft-edge pair looks like the same underlying bug or feature change. Most pairs only share a broad subsystem name (scheduler, Flux, DreamBooth, conversion, inpainting) but differ in the concrete failure mode or requested behavior, so all candidate duplicates should be rejected.", - "confidence": 0.94, + "summary": "This cluster is heterogeneous: it mixes scheduler feature requests, training-script bugs, import/load errors, and model-specific runtime issues. No soft pair looks like the same underlying defect, so all soft edges should be rejected.", + "confidence": 0.96, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 9485 is the broadest scheduler-related anchor in the set, but it is still only a loose representative; the cluster is too mixed for a true duplicate canonical.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10917", + "right": "issue:8887", + "accept": false, + "reason": "Different models and code paths: lumina-2.0 script correctness vs SD3 training VAE encode." + }, + { + "left": "issue:11214", + "right": "issue:9027", + "accept": false, + "reason": "A specific UVit2DModel TypeError is not the same as a general API naming-consistency discussion." + }, { "left": "issue:1444", "right": "issue:5084", "accept": false, - "reason": "Both are import errors, but for different classes and symbols: StableDiffusionSafetyChecker vs FlaxStableDiffusionXLPipeline." + "reason": "Both are import errors, but for different symbols and pipeline families; not the same bug." }, { "left": "issue:1979", "right": "issue:4000", "accept": false, - "reason": "Both concern SD-to-diffusers conversion, but one is a general conversion failure while the other is specifically SD 2.0 ckpt conversion." + "reason": "Both concern conversion, but one is generic SD conversion and the other is specifically SD2 ckpt conversion." }, { "left": "issue:2083", "right": "issue:579", "accept": false, - "reason": "Unrelated bugs: DDIMScheduler/generator handling vs missing config.json in textual inversion." + "reason": "DDIM/generator behavior is unrelated to a missing config.json in textual inversion." }, { "left": "issue:3583", "right": "issue:6706", "accept": false, - "reason": "Mixture of Diffusers pipeline support and 'cads support' are not the same change request." + "reason": "Both are feature requests, but Mixture of Diffusers and cads support are unrelated changes." }, { "left": "issue:3730", "right": "issue:4148", "accept": false, - "reason": "Both touch inpainting, but one is a blending timestep bug and the other is an SDXL config error." + "reason": "Inpainting blending timestep bug is different from SDXL inpainting config validation." }, { "left": "issue:4303", "right": "issue:6921", "accept": false, - "reason": "Different problems: loading SDXL in Flax vs SDXL weights deviating from official weights." + "reason": "Loader crash for SDXL in Flax is not the same as SDXL weight-shipping deviation." }, { "left": "issue:4554", "right": "issue:6906", "accept": false, - "reason": "Kandinsky OOM and DreamBooth LoRA breakage are separate issues with no shared code-path evidence." + "reason": "OOM in a Kandinsky example and a broken DreamBooth LoRA example are separate issues." }, { "left": "issue:4648", "right": "issue:7456", "accept": false, - "reason": "Missing config.json in ControlNet vs a missing raise statement are unrelated." + "reason": "Missing config.json and a missing raise statement are unrelated defects." }, { "left": "issue:5595", "right": "issue:6731", "accept": false, - "reason": "One is an input size validation error; the other is an unhandled pipeline class error." + "reason": "Image-size validation and unhandled pipeline-class errors are different failures." }, { "left": "issue:5603", "right": "issue:6142", "accept": false, - "reason": "ImportError for is_safetensors_available is unrelated to NameError for start_training." + "reason": "An import error and a NameError in training setup are not the same underlying bug." }, { "left": "issue:6535", "right": "issue:9485", "accept": false, - "reason": "Both mention schedulers, but one asks for scheduler swapping in SVD and the other for moving scheduler work to GPU/cuda." + "reason": "Both mention schedulers, but one asks to change the scheduler and the other asks for device placement in scheduler code." }, { "left": "issue:6727", "right": "issue:9248", "accept": false, - "reason": "A dtype mismatch during attention is unrelated to a pip installation command." + "reason": "A dtype mismatch in attention is unrelated to pip install / upgrade guidance." }, { "left": "issue:9134", "right": "issue:9677", "accept": false, - "reason": "Both involve Flux, but one is LoRA training quality and the other is keeping text encoders in fp32." + "reason": "Both mention Flux, but one is a LoRA training issue and the other is a text-encoder precision policy change." }, { "left": "issue:3249", "right": "issue:5028", "accept": false, - "reason": "Different TypeErrors in different blocks: AttnUpBlock2D upsample_size vs hacked_DownBlock2D_forward scale kwarg." + "reason": "Block-argument mismatch bugs, but on different blocks and different keywords; not one concrete code path." }, { "left": "issue:10223", "right": "issue:11387", "accept": false, - "reason": "Dreambooth inference LoRA-ID question and RMSNorm2d fix are entirely different topics." + "reason": "A LoRA inference-ID question and an AutoencoderDC normalization fix are unrelated." }, { "left": "issue:1543", "right": "issue:2265", "accept": false, - "reason": "Cannot set fp16 revision in textual inversion is unrelated to ONNXRuntime incompatibility." + "reason": "Textual inversion fp16 revision and ONNXRuntime incompatibility are separate problems." }, { "left": "issue:4122", "right": "issue:5687", "accept": false, - "reason": "Both are scheduler timing bugs, but they affect different schedulers and failure modes." + "reason": "Both are scheduler timestep bugs, but they affect different schedulers and different failure modes." }, { "left": "issue:11127", "right": "issue:9996", "accept": false, - "reason": "Both are Flux loading issues, but one is about single-file model loading and the other about nf4 transformer loading." + "reason": "Flux single-file loading failure and NF4 transformer loading failure are different model-loading issues." }, { "left": "issue:4331", "right": "issue:5014", "accept": false, - "reason": "Different DreamBooth complaints: script output differences vs lack of results from LoRA training." + "reason": "Different DreamBooth/LoRA training symptoms; not the same bug." }, { "left": "issue:10281", "right": "issue:11859", "accept": false, - "reason": "Both are scheduler feature requests, but for different schedulers with no evidence they are the same request." + "reason": "Both are scheduler feature requests, but for different scheduler algorithms." }, { "left": "issue:1695", "right": "issue:3177", "accept": false, - "reason": "Both mention DreamBooth + DeepSpeed, but one is multi-GPU training and the other is stage-3 offloading being broken." + "reason": "Both mention DreamBooth with DeepSpeed, but one is multi-GPU training and the other is stage-3 offload breakage." }, { "left": "issue:5026", "right": "issue:5875", "accept": false, - "reason": "A KeyError in 0.21.0 and an ONNX prompt conversion error are unrelated." - }, - { - "left": "issue:1672", - "right": "issue:2722", - "accept": false, - "reason": "Both are conversion issues, but one is wrong image size and the other is ControlNet UNet mismatch." - }, - { - "left": "issue:10302", - "right": "issue:12052", - "accept": false, - "reason": "Both involve inference/offloading, but one is FP8 noise without CPU offload and the other is cross-device tensor multiplication failure in Wan 2.2." + "reason": "A text_encoder KeyError and an ONNX conversion prompt error are unrelated." } ] }, @@ -72039,23 +72054,27 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10097", + "issue:10302", "issue:10412", "issue:10526", "issue:11050", "issue:1205", + "issue:12052", "issue:1231", "issue:12412", "issue:13301", + "issue:1672", "issue:1896", "issue:1918", "issue:2351", "issue:2444", + "issue:2722", "issue:2900", "issue:3009", "issue:3029", @@ -72066,7 +72085,6 @@ "issue:4107", "issue:4366", "issue:4708", - "issue:4736", "issue:4839", "issue:5016", "issue:5160", @@ -72075,7 +72093,6 @@ "issue:5515", "issue:5605", "issue:5760", - "issue:5849", "issue:5968", "issue:6142", "issue:6372", @@ -72089,11 +72106,11 @@ "issue:835", "issue:8556", "issue:8732", - "issue:9527", - "issue:9857", - "issue:9858" + "issue:9527" ], "soft_pairs": [ + "issue:1672|issue:2722", + "issue:10302|issue:12052", "issue:10097|issue:6439", "issue:1205|issue:542", "issue:2351|issue:5166", @@ -72115,182 +72132,180 @@ "issue:3140|issue:6142", "issue:4366|issue:5968", "issue:11050|issue:9527", - "issue:2444|issue:4839", - "issue:9857|issue:9858", - "issue:4736|issue:5849" + "issue:2444|issue:4839" ], - "prepared_review_unit_hash": "22314ffef33258c825c0db25f06e39f970313ab0a436ddc45c654ff3ad388438", + "prepared_review_unit_hash": "4652251a26a2018d18fe593e067cc0c241c8927cc17b538ed300a94690f04a29", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33460, - "estimated_input_tokens": 8365, - "estimated_eval_tokens": 16986 + "serialized_chars": 33483, + "estimated_input_tokens": 8371, + "estimated_eval_tokens": 16998 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is broadly heterogeneous: docs, scheduler, model-support, conversion, and training bugs. Most soft pairs are not the same underlying issue; none are strong PR duplicates, and there are no PRs in the set.", - "confidence": 0.79, - "canonical_issue_reason": "No single canonical issue stands out because the items span unrelated problems; if one must be chosen, the most concrete bug report is the SD3 controlnet dtype-casting issue (11050).", + "summary": "This cluster is not a duplicate set; the items span unrelated docs, model-support, conversion, scheduler, training, and dtype issues. All soft pairs are too broad or only superficially similar, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 11050 is the best representative bug report: it names a specific file, a concrete dtype-casting defect, and has a clear fix surface.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1672", + "right": "issue:2722", + "accept": false, + "reason": "Both involve conversion, but one is an output-size bug in ckpt-to-diffusers while the other is a ControlNet UNet mismatch after conversion; different failures and fixes." + }, + { + "left": "issue:10302", + "right": "issue:12052", + "accept": false, + "reason": "Both mention FP8/offloading, but one is inference noise and the other is a tensor device-mismatch crash; not the same code-path problem." + }, { "left": "issue:10097", "right": "issue:6439", "accept": false, - "reason": "Both are docs-related, but one is a missing optimization-doc section and the other is a broken fp16 link; different fixes and not the same bug." + "reason": "Both are docs-related, but one is a missing optimization section and the other is a broken fp16 link; unrelated documentation issues." }, { "left": "issue:1205", "right": "issue:542", "accept": false, - "reason": "Both mention ONNX, but one is about SD model conversion and the other about UNet export; different code paths and likely separate issues." + "reason": "Both are ONNX export topics, but one is converting full Stable Diffusion models and the other is a UNet export issue; too different to be one bug." }, { "left": "issue:2351", "right": "issue:5166", "accept": false, - "reason": "Unrelated scheduler API gap vs VQ-VAE scaling-factor bug; no shared underlying defect." + "reason": "A missing scheduler method versus a missing VQ-VAE scaling factor in a pipeline; different components and fixes." }, { "left": "issue:3009", "right": "issue:3029", "accept": false, - "reason": "Both concern ControlNet training, but one is a general training-improvement request and the other is a custom-resolution question; not the same concrete bug." + "reason": "General ControlNet training advice is not the same as a custom-resolution question." }, { "left": "issue:3221", "right": "issue:7548", "accept": false, - "reason": "Different problems: attention processor loading in one case, AnimateDiff flicker in the other." + "reason": "One is an attention-procs loading API issue; the other is AnimateDiff flicker output quality. Same subsystem, different problems." }, { "left": "issue:5160", "right": "issue:5605", "accept": false, - "reason": "Both are DreamBooth/LoRA topics, but one is masking and the other is per-image captions; distinct feature requests." + "reason": "Masking support and per-image caption support are distinct training features." }, { "left": "issue:10526", "right": "issue:8556", "accept": false, - "reason": "Different failures and code paths: optimum.quanto tensor construction vs SD3 from_single_file meta-tensor copy error." + "reason": "Different models and error modes: optimum.quanto FP8 TypeError versus meta-tensor load failure in SD3 from_single_file." }, { "left": "issue:3960", "right": "issue:6713", "accept": false, - "reason": "They share the HeaderTooLarge error string, but one is LoRA-related and the other is a Stable Diffusion conversion script; too little evidence of the same underlying bug." + "reason": "Both mention HeaderTooLarge, but one is about LoRA and the other about the conversion script; insufficient evidence of the same underlying bug." }, { "left": "issue:4107", "right": "issue:5016", "accept": false, - "reason": "Both involve SDXL LoRA, but one reports degraded quality/noise and the other cannot train text encoders; separate problems." + "reason": "LoRA noise/reduced quality is unrelated to SDXL text encoder training not being enabled." }, { "left": "issue:4708", "right": "issue:7177", "accept": false, - "reason": "Generic import failure vs StableCascadeDecoderPipeline.from_pretrained error; not the same concrete issue." + "reason": "Generic import error versus StableCascade decoder from_pretrained failure; not the same issue." }, { "left": "issue:6372", "right": "issue:6497", "accept": false, - "reason": "Both are new model/pipeline support requests, but they target different models and functionality." + "reason": "SVD+ControlNet support and DragNUWA support are separate model/pipeline feature requests." }, { "left": "issue:10412", "right": "issue:8732", "accept": false, - "reason": "Different training issues: VRAM usage on SD3.5-LoRA vs checkpoint-saving problems in SD1.5 advanced training." + "reason": "VRAM overuse in SD3.5 DreamBooth training is unrelated to checkpoint-saving issues in SD1.5 advanced training." }, { "left": "issue:1918", "right": "issue:5515", "accept": false, - "reason": "Both involve schedulers, but one is a DDIM example issue and the other is LCM initialization without a scheduler; not the same bug." + "reason": "Both scheduler-related, but one is a DDIM example bug and the other is LCM pipeline initialization without a scheduler; different failure points." }, { "left": "issue:1896", "right": "issue:660", "accept": false, - "reason": "CLIP-guided SD size/type error is unrelated to the MPS fp16 revision issue." + "reason": "A CLIP-guided Stable Diffusion size/type error is not the same as an MPS fp16 runtime issue." }, { "left": "issue:2900", "right": "issue:3585", "accept": false, - "reason": "Both mention LoRA merging, but one asks how to merge/save weights while the other reports a failed LoRA-synthesizing model; not clearly the same code-path bug." + "reason": "Both mention LoRA merging, but one asks how to merge/save LoRA weights and the other reports a failure using a synthesized-LoRA model; not clearly the same bug." }, { "left": "issue:5760", "right": "issue:7864", "accept": false, - "reason": "Completely different research/model topics: video-to-audio synthesis vs camera motion transfer." + "reason": "Different research/demo projects in video generation; no shared concrete bug or change." }, { "left": "issue:1231", "right": "issue:835", "accept": false, - "reason": "Flax appears in both, but one is a scheduler request and the other is a numpy_to_pil output bug." + "reason": "A Flax scheduler issue and a Flax numpy_to_pil failure are unrelated." }, { "left": "issue:12412", "right": "issue:13301", "accept": false, - "reason": "Different model support requests for different pipelines; no shared underlying change." + "reason": "Support requests for different model families and pipelines; not duplicates." }, { "left": "issue:3140", "right": "issue:6142", "accept": false, - "reason": "ImportError for StableDiffusionControlNetPipeline is unrelated to a NameError for start_training." + "reason": "StableDiffusionControlNetPipeline import failure versus missing start_training symbol; unrelated errors." }, { "left": "issue:4366", "right": "issue:5968", "accept": false, - "reason": "Different features: SDXL full-UNet fine-tuning vs Text2Video/Img2Video ONNX conversion." + "reason": "Full U-Net SDXL fine-tuning and video-to-ONNX conversion are unrelated tasks." }, { "left": "issue:11050", "right": "issue:9527", "accept": false, - "reason": "Both are dtype-related ControlNet problems, but one is a specific SD3 training-script casting bug and the other is a broader fp32/bf16 mismatch report; not enough to treat as one issue." + "reason": "Both are dtype-related, but one is a missing cast in a specific training script and the other is a controlnet/mainpipe precision mismatch; not the same concrete fix." }, { "left": "issue:2444", "right": "issue:4839", "accept": false, - "reason": "Dreambooth inpaint launch failure and LoRA type error are unrelated." - }, - { - "left": "issue:9857", - "right": "issue:9858", - "accept": false, - "reason": "One is a FLUX controlnet embedding-size mismatch; the other is a KeyError on 'train'. Different failures." - }, - { - "left": "issue:4736", - "right": "issue:5849", - "accept": false, - "reason": "Different LoRA training issues: one is OOM/invalid cast in SDXL LoRA training, the other is an accelerate-launch signature error." + "reason": "DreamBooth inpaint script failure and lora_scale type error are different problems." } ] }, @@ -72309,8 +72324,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -72326,7 +72341,6 @@ "issue:1335", "issue:1949", "issue:2538", - "issue:2695", "issue:2844", "issue:2947", "issue:3418", @@ -72334,10 +72348,9 @@ "issue:4297", "issue:4390", "issue:4456", - "issue:4478", - "issue:4534", "issue:4584", "issue:4619", + "issue:4736", "issue:4767", "issue:4932", "issue:4969", @@ -72355,15 +72368,18 @@ "issue:6794", "issue:7079", "issue:7220", - "issue:8241", "issue:8410", "issue:8803", "issue:9648", + "issue:9857", + "issue:9858", "issue:988", "issue:989", "issue:9911" ], "soft_pairs": [ + "issue:9857|issue:9858", + "issue:4736|issue:5849", "issue:10147|issue:10462", "issue:10446|issue:9911", "issue:12124|issue:13062", @@ -72385,182 +72401,180 @@ "issue:1082|issue:989", "issue:5846|issue:5849", "issue:6794|issue:7079", - "issue:4584|issue:4619", - "issue:2695|issue:4534", - "issue:4478|issue:8241" + "issue:4584|issue:4619" ], - "prepared_review_unit_hash": "1bfdb31b67b7e9c12354c2048dc75f84a44f044b7dcf9cecd1d273e9a810d440", + "prepared_review_unit_hash": "7e3763374682d3b746f41d6117337b990a92604555e7a6bdbe8c476b7f992982", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 34388, - "estimated_input_tokens": 8597, - "estimated_eval_tokens": 17450 + "serialized_chars": 33746, + "estimated_input_tokens": 8437, + "estimated_eval_tokens": 17130 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated Diffusers issues that share generic words like \"multi\", \"support\", or \"error\". None of the soft pairs look like the same underlying bug or change, so I reject all merges.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue: the items span unrelated feature requests, training-script errors, pipeline bugs, scheduler behavior, and platform-specific failures, so any one issue would be a poor duplicate target for the rest.", + "summary": "This cluster is not a true duplicate set: it mixes unrelated feature requests, training-example issues, pipeline/runtime bugs, and hardware/backend compatibility problems across many different code paths.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a safe canonical representative; the items are thematically broad but technically unrelated.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a strong global representative because the cluster is not cohesive; if forced to pick by breadth/activity, issue 10446 is detailed, but it still does not match most of the other items.", + "best_issue_reason": "No global best issue is suitable because the cluster does not describe one underlying bug or change.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:9857", + "right": "issue:9858", + "accept": false, + "reason": "One is a FLUX ControlNet embedding size mismatch, the other is a KeyError: 'train'; different failures." + }, + { + "left": "issue:4736", + "right": "issue:5849", + "accept": false, + "reason": "OOM/cast errors in SDXL LoRA training vs accelerate launch missing args; not the same bug." + }, { "left": "issue:10147", "right": "issue:10462", "accept": false, - "reason": "Different features and code paths: MV-Adapter multi-view generation vs CogVideoX multi-ID customization." + "reason": "Multi-view MV-Adapter support request vs CogVideoX multi-ID customization; different features." }, { "left": "issue:10446", "right": "issue:9911", "accept": false, - "reason": "Both mention Flux and multi-* settings, but one is model sharding incompatibility and the other is multi-ControlNet layer-length mismatch." + "reason": "FluxControlNetPipeline model-sharding incompatibility vs multi-ControlNet layer-length mismatch; different code paths." }, { "left": "issue:12124", "right": "issue:13062", "accept": false, - "reason": "Training dataloader shuffling in qwen-image is unrelated to an onnxruntime-windowsml support decision." + "reason": "Dataloader shuffle question for qwen-image training vs unsupported onnxruntime-windowsml backend." }, { "left": "issue:2947", "right": "issue:4456", "accept": false, - "reason": "Both are loader-related attribute errors, but they occur in different functions and stem from different missing-attribute paths." + "reason": "UNet missing load_attn_procs attribute vs ModuleList indexing error in LoRA loading; unrelated loader issues." }, { "left": "issue:4969", "right": "issue:5300", "accept": false, - "reason": "Importing CrossAttention is unrelated to the package unexpectedly connecting to the internet." + "reason": "CrossAttention import error vs package connecting to internet; different problems." }, { "left": "issue:5538", "right": "issue:8803", "accept": false, - "reason": "Pipeline latent handling and a duplicate assignment in attention processing are different bugs." + "reason": "Upscale pipeline latent usage question vs duplicate assignment in attention processor; unrelated." }, { "left": "issue:4932", "right": "issue:7220", "accept": false, - "reason": "Dreambooth checkpoint output and IP-Adapter unloading on ControlNet inpaint are unrelated." + "reason": "Dreambooth output file issue vs unload_ip_adapter error in a ControlNet inpaint pipeline; not the same bug." }, { "left": "issue:4767", "right": "issue:5254", "accept": false, - "reason": "Scheduler prediction_type support and multi-ControlNet inpainting are different feature requests." + "reason": "Scheduler prediction_type support vs multi-ControlNet support in inpaint pipeline; different changes." }, { "left": "issue:4977", "right": "issue:5592", "accept": false, - "reason": "Removing checkpoints in one example is unrelated to loading UniDiffuser from checkpoints." + "reason": "Checkpoint removal in an example script vs UniDiffuser checkpoint loading failure; different artifacts." }, { "left": "issue:5628", "right": "issue:6159", "accept": false, - "reason": "One is a scheduler offset bug; the other is missing denoising_end support in a specific pipeline." + "reason": "LMS scheduler denoising offset vs missing denoising_end support in ControlNetXSPipeline; different scheduler/pipeline issues." }, { "left": "issue:4297", "right": "issue:6295", "accept": false, - "reason": "Both report artifacts, but they involve different pipelines and different suspected causes." + "reason": "SDXL batched img2img artifacts vs DPM++ 2M SDE Karras artifacts; similar symptom but not the same concrete bug." }, { "left": "issue:11561", "right": "issue:8410", "accept": false, - "reason": "FluxFill IP-Adapter loading and TextDiffuser native support are separate feature requests." + "reason": "FluxFill IP-Adapter support vs native TextDiffuser support; different feature requests." }, { "left": "issue:1251", "right": "issue:3769", "accept": false, - "reason": "Black output on mps in RePaint and quartered output in Panorama are different pipeline/device bugs." + "reason": "MPS black image in RePaintPipeline vs panorama quartered output; distinct pipeline bugs." }, { "left": "issue:1335", "right": "issue:4390", "accept": false, - "reason": "A missing vae_encoder constructor arg is not the same as a generic missing requirements issue." + "reason": "Missing vae_encoder argument vs missing requirements; unrelated import/setup issues." }, { "left": "issue:2538", "right": "issue:3418", "accept": false, - "reason": "Both concern LoRA training and text encoder options, but they target different scripts and different requested behavior." + "reason": "Enabling train_text_encoder in one script vs adding text-encoder part in another training script; related theme but not the same change." }, { "left": "issue:5334", "right": "issue:988", "accept": false, - "reason": "Accelerate multi-model preparation during training is not the same problem as torch FSDP support for examples." + "reason": "Accelerate multi-model handling request vs FSDP training examples; broad training infrastructure, but not the same issue." }, { "left": "issue:12113", "right": "issue:9648", "accept": false, - "reason": "A channel-mismatch error in WanImageToVideoPipeline is unrelated to VAE tiling memory overflow." + "reason": "WanImageToVideo input channel mismatch vs VAE tiling memory overflow; different root causes." }, { "left": "issue:1949", "right": "issue:2844", "accept": false, - "reason": "Dreambooth tutorial training errors and explicit image-embedding passing in StableUnCLIP are different workflows." + "reason": "Dreambooth tutorial training errors vs explicit image embeddings in StableUnCLIPImg2ImgPipeline; unrelated." }, { "left": "issue:1082", "right": "issue:989", "accept": false, - "reason": "A CUBLAS matmul execution failure is not the same as a 2-D weight shape error." + "reason": "Dreambooth cublasLtMatmul failure vs generic weight dimensionality error; not the same bug." }, { "left": "issue:5846", "right": "issue:5849", "accept": false, - "reason": "Both hit train_text_to_image_lora_sdxl.py, but one is resume-from-checkpoint behavior and the other is an accelerate entrypoint signature error." + "reason": "resume_from_checkpoint error vs accelerate launch missing args; both training-script issues but different failures." }, { "left": "issue:6794", "right": "issue:7079", "accept": false, - "reason": "An import error for cast_training_params is unrelated to a Transformer2DModel patch/norm_type bug." + "reason": "Import error for cast_training_params vs Transformer2DModel patch/norm_type error; unrelated." }, { "left": "issue:4584", "right": "issue:4619", "accept": false, - "reason": "Both are SDXL training-script failures, but one is resume_from_checkpoint and the other is input/bias dtype mismatch." - }, - { - "left": "issue:2695", - "right": "issue:4534", - "accept": false, - "reason": "A Tensorboard validation/reporting issue in ControlNet training is different from an MPS validation crash in SDXL Dreambooth." - }, - { - "left": "issue:4478", - "right": "issue:8241", - "accept": false, - "reason": "An SDXL dtype mismatch on load is unrelated to an abnormal LEDITS++ model size report." + "reason": "resume_from_checkpoint error in SDXL LoRA training vs half/float bias mismatch; different training bugs." } ] }, @@ -72579,8 +72593,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -72600,7 +72614,7 @@ "issue:1546", "issue:2202", "issue:2471", - "issue:2984", + "issue:2695", "issue:3087", "issue:3689", "issue:4030", @@ -72608,8 +72622,9 @@ "issue:4269", "issue:4332", "issue:4368", + "issue:4478", + "issue:4534", "issue:4615", - "issue:5046", "issue:5079", "issue:5245", "issue:5327", @@ -72622,6 +72637,7 @@ "issue:7420", "issue:7432", "issue:7572", + "issue:8241", "issue:8827", "issue:8837", "issue:8930", @@ -72629,11 +72645,11 @@ "issue:9227", "issue:9479", "issue:9531", - "issue:9813", - "issue:9858", - "issue:9927" + "issue:9813" ], "soft_pairs": [ + "issue:2695|issue:4534", + "issue:4478|issue:8241", "issue:1546|issue:4269", "issue:7432|issue:7572", "issue:11661|issue:9227", @@ -72656,188 +72672,186 @@ "issue:4030|issue:5327", "issue:4107|issue:4368", "issue:10100|issue:12358", - "issue:2202|issue:3087", - "issue:2984|issue:5046", - "issue:9858|issue:9927" + "issue:2202|issue:3087" ], - "prepared_review_unit_hash": "232dd75c4956bd5b8a251db53b5250bdc94158be871aa1b111f7ca873913e183", + "prepared_review_unit_hash": "8494d4b350226c1c3a9a992e67b248c0c9c4c5e40da3fcfdf075a1d52798904d", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33971, - "estimated_input_tokens": 8493, - "estimated_eval_tokens": 17242 + "serialized_chars": 34075, + "estimated_input_tokens": 8519, + "estimated_eval_tokens": 17294 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is a collection of mostly unrelated feature requests and bug reports across different pipelines, adapters, and training scripts. None of the proposed soft pairs looks like the same underlying issue, so I would not merge any of them as duplicates.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "The set is broadly heterogeneous; the soft pairs are mostly superficial topic matches rather than true duplicates. No PRs are present.", + "confidence": 0.22, + "canonical_issue_reason": "If a single issue must be chosen, issue 12783 is the most concrete, reproducible open bug report in the noisy-output/attention area, but the overall set is not a single duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11117 is the strongest standalone bug report: it has a concrete runtime error, a specific pipeline, and a clear failure mode, so it is the best representative issue if one must pick a single item.", + "best_issue_reason": "Issue 12783 is the strongest standalone issue: it has a clear symptom, current impact, and enough detail to be actionable.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2695", + "right": "issue:4534", + "accept": false, + "reason": "Both are training/validation crashes, but they affect different pipelines and error paths." + }, + { + "left": "issue:4478", + "right": "issue:8241", + "accept": false, + "reason": "Different models and different failures: dtype mismatch versus abnormal model size." + }, { "left": "issue:1546", "right": "issue:4269", "accept": false, - "reason": "Different pipelines and different failures: SD batched truncation handling vs ControlNet input validation TypeError." + "reason": "Both involve input validation, but one is truncation/batched input behavior and the other is a TypeError in check_inputs." }, { "left": "issue:7432", "right": "issue:7572", "accept": false, - "reason": "Both are image-quality/runtime problems, but the root causes differ: WebUI parity vs dtype mismatch in a convolution/bias path." + "reason": "One is about image quality differences; the other is a concrete dtype mismatch error." }, { "left": "issue:11661", "right": "issue:9227", "accept": false, - "reason": "Same broad Flux area, but one is a None-handling bug in an example script and the other is a model-specific training failure with a different symptom." + "reason": "Different training scripts and different bugs; only a broad LoRA/training theme overlaps." }, { "left": "issue:8837", "right": "issue:8989", "accept": false, - "reason": "Different subsystems and symptoms: SlicedAttnProcessor bug vs Stable Audio sequential offloading." + "reason": "Both concern performance/attention-related behavior, but they are distinct code paths and symptoms." }, { "left": "issue:10933", "right": "issue:11396", "accept": false, - "reason": "IP-Adapter load error and LoRA format-conversion question are unrelated." + "reason": "IP-Adapter loading error versus a LoRA conversion question; not the same underlying bug." }, { "left": "issue:11117", "right": "issue:12895", "accept": false, - "reason": "Both touch StableDiffusion3Pipeline, but one is a Half/BFloat16 tensor-type crash and the other is a NoneType.ndim crash in batched prompt handling." + "reason": "Different Stable Diffusion 3 failures: dtype mismatch versus NoneType batching crash." }, { "left": "issue:11779", "right": "issue:9479", "accept": false, - "reason": "Both involve masking, but they are different pipelines and different bugs: Wan VACE mask padding vs inpaint changing unmasked regions." + "reason": "Both involve masking, but one is Wan VACE mask padding and the other is inpaint overwriting unmasked regions." }, { "left": "issue:11901", "right": "issue:13137", "accept": false, - "reason": "Different model families and adapter-loading paths: Wan2.1 LoRA loading vs Flux 2 LoKr weights." + "reason": "Model-specific LoRA loading failures with different models and error modes." }, { "left": "issue:12567", "right": "issue:12783", "accept": false, - "reason": "Both report noisy outputs, but the attention implementations and likely code paths are different (Ulysses vs Sage Attention)." + "reason": "Both mention noisy output from attention optimizations, but the implementations and failure details differ." }, { "left": "issue:1305", "right": "issue:3087", "accept": false, - "reason": "A paper/reference request is not the same as adding LyCORIS model support." + "reason": "Direct inversion support request and LyCORIS support are unrelated feature requests." }, { "left": "issue:3689", "right": "issue:5079", "accept": false, - "reason": "Related LoRA area, but unload-after-load behavior is not the same bug as adding LyCORIS loading support." + "reason": "Related to LoRA APIs, but one is unloading weights and the other is loading LyCORIS." }, { "left": "issue:4332", "right": "issue:7420", "accept": false, - "reason": "Different acceleration paths and failures: xformers support in one pipeline vs torch.compile usability." + "reason": "xformers usage in a specific pipeline versus torch.compile enablement are unrelated." }, { "left": "issue:4615", "right": "issue:5245", "accept": false, - "reason": "An import-name error is not the same as a pipenv import-stage freeze." + "reason": "ImportError in diffusers versus a pipenv import-stage freeze; not the same issue." }, { "left": "issue:5245", "right": "issue:7191", "accept": false, - "reason": "Both mention import problems, but one is a freeze in pipenv and the other is a separate import issue with no matching failure mode." + "reason": "Generic import problems, but no shared code path or concrete failure mode." }, { "left": "issue:6439", "right": "issue:9531", "accept": false, - "reason": "Docs link 404 and SDXL sigma scheduling are unrelated." + "reason": "Docs link 404 and SDXL sigma tuning are unrelated." }, { "left": "issue:8827", "right": "issue:9813", "accept": false, - "reason": "Both are IP-Adapter related, but one is a specific FaceID-SDXL loading error while the other is a batch feature-mixing bug in embed preparation; not enough to call them the same bug." + "reason": "Both are IP-Adapter-related, but one is a FaceID loading error and the other is batch feature mixing." }, { "left": "issue:5573", "right": "issue:7138", "accept": false, - "reason": "ControlNet weight initialization slowness is different from SDXL UNet structure mismatch." + "reason": "ControlNet initialization slowness versus SDXL UNet structure mismatch are different problems." }, { "left": "issue:6109", "right": "issue:8930", "accept": false, - "reason": "Both concern ControlNet, but they are different pipelines and distinct behavior failures." + "reason": "Both involve ControlNet-like conditioning, but they target different pipelines and bugs." }, { "left": "issue:2471", "right": "issue:5607", "accept": false, - "reason": "Different tensor-shape failures in different pipelines; no evidence they share the same root cause." + "reason": "Both mention shape-related failures, but the underlying pipelines and tensor mismatch causes are different." }, { "left": "issue:4030", "right": "issue:5327", "accept": false, - "reason": "Different models and errors: SD v2.1 embedding runtime issue vs SDXL textual inversion load failure." + "reason": "Embedding/textual-inversion related, but one is a runtime embedding error and the other is an SDXL import/load issue." }, { "left": "issue:4107", "right": "issue:4368", "accept": false, - "reason": "One is a quality degradation/noise report, the other is a CUDA OOM training issue." + "reason": "Noise/quality degradation and OOM during LoRA training are not the same bug." }, { "left": "issue:10100", "right": "issue:12358", "accept": false, - "reason": "Both are feature requests for new model support, but for different models and pipelines." + "reason": "Both are feature requests for new pipelines, but they are for different models and changes." }, { "left": "issue:2202", "right": "issue:3087", "accept": false, - "reason": "OpenVINO pipeline support and LyCORIS model support are unrelated feature requests." - }, - { - "left": "issue:2984", - "right": "issue:5046", - "accept": false, - "reason": "Different research/model requests with no shared code-path bug." - }, - { - "left": "issue:9858", - "right": "issue:9927", - "accept": false, - "reason": "Both are training-time issues, but the reported failures are unrelated: a KeyError versus a HeaderTooLarge error." + "reason": "OpenVINO pipeline support and LyCORIS support are unrelated requests." } ] }, @@ -72856,8 +72870,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -72877,23 +72891,21 @@ "issue:1957", "issue:2266", "issue:2507", - "issue:2532", - "issue:2696", "issue:2775", + "issue:2984", "issue:3026", "issue:3585", "issue:3937", "issue:4478", "issue:4551", - "issue:460", "issue:4605", "issue:4618", + "issue:5046", "issue:5110", "issue:5274", "issue:5594", "issue:5678", "issue:5714", - "issue:5824", "issue:5829", "issue:6335", "issue:6520", @@ -72907,9 +72919,13 @@ "issue:7640", "issue:7836", "issue:9032", - "issue:9171" + "issue:9171", + "issue:9858", + "issue:9927" ], "soft_pairs": [ + "issue:2984|issue:5046", + "issue:9858|issue:9927", "issue:10029|issue:10398", "issue:2266|issue:2507", "issue:4618|issue:5110", @@ -72931,182 +72947,180 @@ "issue:10010|issue:9032", "issue:10782|issue:12149", "issue:11581|issue:11885", - "issue:11592|issue:9171", - "issue:2532|issue:5824", - "issue:2696|issue:460" + "issue:11592|issue:9171" ], - "prepared_review_unit_hash": "0dde96cf0f996fb090dbc0fd1642deba10036275dbb097c63ba2cd2a3ff06e70", + "prepared_review_unit_hash": "1b35112be0d07dfa1a98171a776bdca0a35c0e5fbb44dd84777b95834a99c4e3", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33974, - "estimated_input_tokens": 8494, - "estimated_eval_tokens": 17244 + "serialized_chars": 33862, + "estimated_input_tokens": 8466, + "estimated_eval_tokens": 17188 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set looks like a collection of mostly unrelated issues rather than one duplicate cluster. The soft-edge pairs share broad themes at most, but they describe different errors, models, or feature requests and should not be merged.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "All proposed soft links are false positives; the items are a heterogeneous mix of unrelated bugs, features, and regressions rather than one duplicate cluster.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical duplicate: the issues span unrelated code paths and product areas.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "issue:11581 is the most concrete standalone bug report in the set, with a specific runtime error and likely actionable fix.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2984", + "right": "issue:5046", + "accept": false, + "reason": "Different topics: a video diffusion paper/proposal vs a drag-style manipulation feature request." + }, + { + "left": "issue:9858", + "right": "issue:9927", + "accept": false, + "reason": "Both are training-related errors, but the failure modes differ completely (`KeyError: 'train'` vs `HeaderTooLarge`)." + }, { "left": "issue:10029", "right": "issue:10398", "accept": false, - "reason": "Both are PEFT/LoRA-related, but one is an AttributeError on Linear.base_layer and the other is a missing target-modules/base-model mismatch; different failure points." + "reason": "Both involve PEFT/adapter loading, but one is a missing `base_layer` attribute and the other is missing target modules in the base model." }, { "left": "issue:2266", "right": "issue:2507", "accept": false, - "reason": "Both concern performance, but one is sequential CPU offload slowdown while the other is safetensors loading being slower on GPU; not the same bug." + "reason": "Both mention performance, but one is sequential CPU offload slowdown and the other is safetensors GPU loading being slower than CPU." }, { "left": "issue:4618", "right": "issue:5110", "accept": false, - "reason": "One asks how to use a specific SDXL/controlnet model combo; the other is a Hugging Face repo-id validation error. Different problems." + "reason": "Completely unrelated: SDXL ControlNet usage vs Hugging Face repo-id validation failure." }, { "left": "issue:11223", "right": "issue:11674", "accept": false, - "reason": "OOM with LoRAs during WAN layer-wise upcasting is unrelated to a feature request for ref images and multiple control videos." + "reason": "Different requests: an OOM bug with LoRAs under layer-wise upcasting vs a feature request for Wan VACE inputs." }, { "left": "issue:10010", "right": "issue:7836", "accept": false, - "reason": "FLUX Fill 404/model download failure is unrelated to a CLIP training example overfitting report." + "reason": "Unrelated bugs: a FLUX Fill model download failure vs CLIP training overfitting." }, { "left": "issue:1957", "right": "issue:5274", "accept": false, - "reason": "LR scheduler initialization fix is unrelated to a mixed-precision dtype mismatch during unconditional image generation training." + "reason": "Different training problems: LR scheduler initialization vs mixed-precision dtype mismatch." }, { "left": "issue:5594", "right": "issue:6529", "accept": false, - "reason": "Memory-saving for SDXL LoRA fine-tuning and SDXL Fooocus Inpaint are different topics, not the same defect." + "reason": "Different concerns: SDXL LoRA memory-saving discussion vs Fooocus inpaint support." }, { "left": "issue:5714", "right": "issue:7185", "accept": false, - "reason": "PeftModel support in maybe_raise_or_warn is unrelated to whether diffusers supports PyTorch 2.2.0." + "reason": "Unrelated issues: PeftModel handling in `maybe_raise_or_warn` vs PyTorch 2.2 compatibility." }, { "left": "issue:3026", "right": "issue:3585", "accept": false, - "reason": "Accelerate launch error in instruct_pix2pix and a failure using a LoRA-synthesizing model are different code paths and symptoms." + "reason": "Different code paths and symptoms: instruct-pix2pix launch failure vs LoRA synthesis into SD1.5 failing." }, { "left": "issue:7168", "right": "issue:7640", "accept": false, - "reason": "IP-Adapter image-embed tensor-shape confusion is not the same as the from_single_file dimension RuntimeError in SDXL." + "reason": "Both are image-generation issues, but IP-Adapter embedding shape handling is not the same as `from_single_file` dimension errors." }, { "left": "issue:10640", "right": "issue:6794", "accept": false, - "reason": "FluxInpaintPipeline DataParallel multi-GPU error is unrelated to an import failure for cast_training_params." + "reason": "Different subsystems: FluxInpaint/DataParallel GPU error vs import failure for `cast_training_params`." }, { "left": "issue:3937", "right": "issue:6567", "accept": false, - "reason": "ZeroscopeXL not working as a normal model and an SD 1.5 text_config_dict warning are not the same bug." + "reason": "Unrelated: ZeroscopeXL model loading behavior vs an SD 1.5 import warning about `text_config_dict`." }, { "left": "issue:4478", "right": "issue:4605", "accept": false, - "reason": "SDXL input/bias type mismatch and a long-prompt-weighting inpaint error are different failures." + "reason": "Both are SD model errors, but one is a dtype/bias mismatch in SDXL loading and the other is an inpaint pipeline error." }, { "left": "issue:2775", "right": "issue:6520", "accept": false, - "reason": "torch._dynamo.optimize on a non-function object is unrelated to blank validation results during adapter training." + "reason": "Different failures: a Torch Dynamo optimization error vs blank validation outputs during adapter training." }, { "left": "issue:4551", "right": "issue:5829", "accept": false, - "reason": "DreamBooth crashing on SDXL base and missing unet_time_cond_proj_dim in LCM scripts are distinct issues." + "reason": "Different training crashes: SDXL DreamBooth failure vs missing `unet_time_cond_proj_dim` in LCM scripts." }, { "left": "issue:11148", "right": "issue:12501", "accept": false, - "reason": "A PEFT adapter-loaded flag regression is unrelated to the VAE decoder next(iter(..)) graph break." + "reason": "Both are bugs, but one is PEFT adapter state tracking and the other is a VAE decoder graph-break issue." }, { "left": "issue:6335", "right": "issue:6598", "accept": false, - "reason": "Control of output size in img2img is not the same as differing outputs with the same parameters." + "reason": "Different img2img problems: output-size control vs differing results for identical parameters." }, { "left": "issue:5678", "right": "issue:7575", "accept": false, - "reason": "Higher-order derivative errors from scaled_dot_product_attention and an adanorm shape error are different code-path problems." + "reason": "Distinct model/math issues: higher-order derivative behavior in attention vs a shape error in Transformer2DModel with adanorm." }, { "left": "issue:10010", "right": "issue:9032", "accept": false, - "reason": "FLUX Fill model download failure is unrelated to a FlaxStableDiffusionPipeline minimum-working-example question." + "reason": "Completely unrelated: FLUX Fill example download failure vs a Flax/TPU setup question." }, { "left": "issue:10782", "right": "issue:12149", "accept": false, - "reason": "A Lumina minor issue is not the same as pip-installed diffusers being out of sync with source code." + "reason": "Different scopes: a Lumina Image 2.0 minor issue vs package/source version mismatch." }, { "left": "issue:11581", "right": "issue:11885", "accept": false, - "reason": "Both involve FLUX adapter loading, but one is a CLIPVisionModelWithProjection dtype constructor error and the other is a load_lora_weights regression; too different to merge." + "reason": "Both concern FLUX adapters, but one is an IP-Adapter constructor dtype error and the other is a LoRA loading regression." }, { "left": "issue:11592", "right": "issue:9171", "accept": false, - "reason": "delete_adapters regression and from_single_file v_prediction handling are unrelated regressions." - }, - { - "left": "issue:2532", - "right": "issue:5824", - "accept": false, - "reason": "Panorama get_views behavior and SDXL add_time_ids default issues affect different pipelines and functions." - }, - { - "left": "issue:2696", - "right": "issue:460", - "accept": false, - "reason": "Torch not compiled with CUDA after installing diffusers is unrelated to ONNX Runtime CPU/CUDA dependency selection." + "reason": "Unrelated regressions: FluxLoraLoaderMixin adapter deletion vs SD 1.5 `v_prediction` handling in `from_single_file`." } ] }, @@ -73125,8 +73139,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -73134,7 +73148,8 @@ "issue:11127", "issue:1214", "issue:1530", - "issue:2737", + "issue:2532", + "issue:2696", "issue:3089", "issue:3171", "issue:4000", @@ -73144,6 +73159,7 @@ "issue:4414", "issue:447", "issue:4542", + "issue:460", "issue:4605", "issue:4675", "issue:4753", @@ -73151,15 +73167,13 @@ "issue:5054", "issue:5152", "issue:5372", - "issue:5821", + "issue:5824", "issue:5981", "issue:5989", "issue:6159", "issue:6183", "issue:6258", - "issue:6513", "issue:6790", - "issue:7000", "issue:7009", "issue:7041", "issue:7210", @@ -73179,6 +73193,8 @@ "issue:9953" ], "soft_pairs": [ + "issue:2532|issue:5824", + "issue:2696|issue:460", "issue:3089|issue:5989", "issue:3171|issue:447", "issue:4000|issue:8373", @@ -73200,182 +73216,180 @@ "issue:8900|issue:9209", "issue:1530|issue:5002", "issue:4414|issue:9023", - "issue:4605|issue:962", - "issue:5821|issue:7000", - "issue:2737|issue:6513" + "issue:4605|issue:962" ], - "prepared_review_unit_hash": "1aed2ab87bdea17038bc9304137be68e806a8d5e645aca93a7e379ffadb28f51", + "prepared_review_unit_hash": "610c32b0b96ca39c913a7142487a73d6d99df01272d740ef5a84f1c7454fe4ed", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33409, - "estimated_input_tokens": 8353, - "estimated_eval_tokens": 16962 + "serialized_chars": 33464, + "estimated_input_tokens": 8366, + "estimated_eval_tokens": 16988 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This looks like a heterogeneous false-positive similarity cluster, not a true duplicate set. The items span unrelated bugs in loading, precision, schedulers, training, networking, and pipeline behavior, so no soft pair is convincing.", - "confidence": 0.91, - "canonical_issue_reason": "Issue:6183 is the most concrete, narrowly scoped bug report in the set (a specific StableDiffusionXL from_single_file shape problem), but the cluster is too mixed to have a true canonical duplicate hub.", + "summary": "This is not a true duplicate cluster: the soft pairs mostly share only broad diffusion-model vocabulary or the same subsystem, but they describe different bugs, features, or backend/platform failures. I would not merge any of the proposed pairs.", + "confidence": 0.98, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue:6183 is the best representative because it has a clear repro and code path, unlike the broader or more ambiguous reports elsewhere in the cluster.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2532", + "right": "issue:5824", + "accept": false, + "reason": "Both involve Stable Diffusion pipeline internals, but one is a panorama get_views bug and the other is an SDXL add_time_ids default issue; different code paths and symptoms." + }, + { + "left": "issue:2696", + "right": "issue:460", + "accept": false, + "reason": "CUDA installation/runtime error vs ONNX Runtime backend dependency selection; these are separate environment/backend problems." + }, { "left": "issue:3089", "right": "issue:5989", "accept": false, - "reason": "Both mention scheduler bugs, but they are different failures in different components (DDPM scheduler vs safetensors.torch)." + "reason": "DDPM scheduler logic bug and safetensors.torch bug are unrelated components with different failure modes." }, { "left": "issue:3171", "right": "issue:447", "accept": false, - "reason": "Prompt-related, but one is LPW controlnet behavior and the other is truncation detection/token return; not the same bug." + "reason": "LPW prompt support for ControlNet vs prompt truncation detection/return values; different feature requests." }, { "left": "issue:4000", "right": "issue:8373", "accept": false, - "reason": "One is about converting a ckpt to diffusers format, the other is an incorrect error message; no shared code-path bug." + "reason": "Model ckpt conversion problem vs an incorrect error message; not the same underlying bug." }, { "left": "issue:4080", "right": "issue:4675", "accept": false, - "reason": "Both involve SDXL-related features, but one requests an inpainting pipeline and the other ControlNet reference-only support." + "reason": "SDXL inpainting pipeline request vs ControlNet Reference Only support; separate features." }, { "left": "issue:5054", "right": "issue:5372", "accept": false, - "reason": "Different pipeline API issues: missing final_offload_hook vs safety_checker usage; not duplicate failures." + "reason": "Missing final_offload_hook attribute and safety_checker usage are different StableDiffusionXL API issues." }, { "left": "issue:6790", "right": "issue:7592", "accept": false, - "reason": "Both are state_dict loading problems, but they affect different models/features (LCM example vs DoRA) and likely different fixes." + "reason": "Both are state_dict loading problems, but one is an LCM example key mismatch and the other is DoRA key loading; not the same concrete bug." }, { "left": "issue:7041", "right": "issue:8792", "accept": false, - "reason": "A segfault at >1 steps is not the same as image-quality regression during migration." + "reason": "Segfault at num_inference_steps > 1 is a runtime crash; image-quality degradation during migration is a separate quality regression." }, { "left": "issue:11127", "right": "issue:9856", "accept": false, - "reason": "Single-file FLUX loading and distributed port-collision errors are unrelated." + "reason": "Single-file FLUX model loading failure vs distributed port collision; unrelated errors." }, { "left": "issue:8644", "right": "issue:9953", "accept": false, - "reason": "Both involve offload/device placement, but one is T5 encoder meta-tensor offload and the other is moving a quantized component to CUDA; different code paths." + "reason": "Both mention device/offload behavior, but one is T5 encoder sequential offload meta-tensor failure and the other is moving a quantized component to CUDA; different code paths." }, { "left": "issue:4403", "right": "issue:6159", "accept": false, - "reason": "LayerNorm half-precision failure and unsupported denoising_end are different pipeline/runtime issues." + "reason": "Half-precision LayerNorm failure vs missing denoising_end argument support; unrelated." }, { "left": "issue:6183", "right": "issue:8373", "accept": false, - "reason": "Incorrect tensor shape from_single_file is unrelated to an incorrect-format error message." + "reason": "Incorrect tensor shape from from_single_file is not the same as an incorrect error message." }, { "left": "issue:10940", "right": "issue:7210", "accept": false, - "reason": "Missing triton.ops import and missing named_parameters on a pipeline are separate environment/API problems." + "reason": "Missing triton.ops import and missing named_parameters under accelerate are different integration issues." }, { "left": "issue:1214", "right": "issue:4139", "accept": false, - "reason": "Both are precision-themed, but one is CLIPTextModel half-precision failure and the other is casting a custom attention processor." + "reason": "Both concern precision, but CLIPTextModel half-precision support and custom attention processor casting are not the same bug." }, { "left": "issue:4542", "right": "issue:7355", "accept": false, - "reason": "Multi-image img2img limitation and Stable Cascade arbitrary-resolution bf16 errors are not the same bug." + "reason": "Cannot generate more than one image in SDXL img2img vs Stable Cascade arbitrary-resolution bfloat16 errors; unrelated." }, { "left": "issue:4753", "right": "issue:8236", "accept": false, - "reason": "Dreambooth LoRA script trouble and wrong scheduler step count are unrelated training issues." + "reason": "train_dreambooth_lora.py problem vs wrong multi-GPU LR scheduler step count; different training bugs." }, { "left": "issue:9279", "right": "issue:9292", "accept": false, - "reason": "Both mention LoRA, but one is dynamic loading with torch.compile and the other is key incompatibility for FLUX LoRAs." + "reason": "Dynamic LoRA loading with torch.compile and FLUX LoRA key incompatibility are separate issues." }, { "left": "issue:5152", "right": "issue:5981", "accept": false, - "reason": "Different scheduler/model bugs; no concrete shared failure mode." + "reason": "W\u00fcrstchen alpha bug vs DPMSolver scheduler bug; no concrete overlap." }, { "left": "issue:6258", "right": "issue:7009", "accept": false, - "reason": "Forward chunking quality degradation in video pipeline is not the same as poor results from a text-to-video checkpoint." + "reason": "Forward chunking degrading quality in Stable Video Diffusion is not the same as poor results from a specific text-to-video checkpoint." }, { "left": "issue:8900", "right": "issue:9209", "accept": false, - "reason": "Single-file loading guidance and mirror connection failure are unrelated." + "reason": "Loading SDXL refiner from_single_file vs failing to connect to hf-mirror.com are unrelated." }, { "left": "issue:1530", "right": "issue:5002", "accept": false, - "reason": "Both are missing-argument/API compatibility reports, but they concern different pipeline/scheduler symbols and fixes." + "reason": "Missing learned_classifier_free_sampling_embeddings and missing use_karras_sigmas are different API gaps in different schedulers/pipelines." }, { "left": "issue:4414", "right": "issue:9023", "accept": false, - "reason": "Consistency Model sampling support and a generic sample-code failure are not the same underlying bug." + "reason": "Consistency Model sample request vs generic sample code failure; too broad to be the same bug." }, { "left": "issue:4605", "right": "issue:962", "accept": false, - "reason": "Inpaint-model error in long-prompt weighting and lack of non-512 inpaint size support are different pipeline issues." - }, - { - "left": "issue:5821", - "right": "issue:7000", - "accept": false, - "reason": "A CUDA backward/training failure and a PyTorch deprecation warning are not duplicates." - }, - { - "left": "issue:2737", - "right": "issue:6513", - "accept": false, - "reason": "CLIP guidance support for a scheduler and missing T2IAdapter model support are unrelated feature requests." + "reason": "Long prompt weighting with an inpaint model error vs inpaint pipelines only supporting 512x512 are different inpaint limitations." } ] }, @@ -73394,8 +73408,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -73415,6 +73429,7 @@ "issue:1919", "issue:1922", "issue:2098", + "issue:2737", "issue:2838", "issue:305", "issue:3057", @@ -73423,23 +73438,22 @@ "issue:4186", "issue:4222", "issue:4249", - "issue:4585", "issue:459", "issue:4768", - "issue:4977", - "issue:5034", "issue:5110", "issue:5120", "issue:5485", "issue:5605", "issue:5689", + "issue:5821", "issue:5989", "issue:6042", "issue:6055", "issue:6313", + "issue:6513", "issue:6616", "issue:6686", - "issue:7147", + "issue:7000", "issue:7490", "issue:7798", "issue:8552", @@ -73449,6 +73463,8 @@ "issue:9974" ], "soft_pairs": [ + "issue:5821|issue:7000", + "issue:2737|issue:6513", "issue:4222|issue:5605", "issue:5989|issue:7798", "issue:6042|issue:8790", @@ -73470,182 +73486,180 @@ "issue:4094|issue:4768", "issue:5485|issue:8552", "issue:7490|issue:9331", - "issue:10066|issue:9974", - "issue:4585|issue:7147", - "issue:4977|issue:5034" + "issue:10066|issue:9974" ], - "prepared_review_unit_hash": "3dd242566142cd7933828c0dd2864c8b79aa4a7516e44274e46a617cb87695e4", + "prepared_review_unit_hash": "8d6779f32c05601def4a185e5d303425b8903681d6a5000663b3d6c0dc473079", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33895, - "estimated_input_tokens": 8474, - "estimated_eval_tokens": 17204 + "serialized_chars": 33910, + "estimated_input_tokens": 8478, + "estimated_eval_tokens": 17212 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No convincing duplicate cluster here: the soft pairs span unrelated models, examples, import errors, training bugs, and feature requests. All soft edges are rejected.", - "confidence": 0.68, - "canonical_issue_reason": "issue:11489 is the most actionable concrete bug report in the set: a reproducible Flux LoRA training failure with active discussion. It is the best anchor even though the rest are largely unrelated.", + "summary": "This cluster is largely heterogeneous: it mixes unrelated bugs, feature requests, and support questions across different pipelines and versions. None of the soft-edge pairs look like true duplicates of the same underlying bug/change.", + "confidence": 0.96, + "canonical_issue_reason": "No issue cleanly represents the whole set because the cluster is not cohesive. If one must be chosen, issue 11489 is the most concrete, actionable bug report (a specific Flux LoRA training failure) rather than a broad support question or a generic error report.", "canonical_pr_reason": null, - "best_issue_reason": "Same as canonical: 11489 is the clearest current code-path bug report and most suitable representative issue among this mixed batch.", + "best_issue_reason": "Issue 11489 is the best single triage candidate: it describes a specific, reproducible training error with a clear context. The rest are too mixed in scope to serve as a strong canonical representative.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5821", + "right": "issue:7000", + "accept": false, + "reason": "Different problems: CUDA/backward training failure vs a PyTorch deprecation warning." + }, + { + "left": "issue:2737", + "right": "issue:6513", + "accept": false, + "reason": "Both are feature requests, but for different capabilities: CLIP guidance vs T2IAdapter model support." + }, { "left": "issue:4222", "right": "issue:5605", "accept": false, - "reason": "Low-resolution dataset training question vs per-image caption support request for SDXL dreambooth; different problem and scope." + "reason": "Low-resolution dataset training question is unrelated to per-image caption support." }, { "left": "issue:5989", "right": "issue:7798", "accept": false, - "reason": "safetensors.torch bug vs PixArtAlphaPipeline bug; unrelated subsystems and symptoms." + "reason": "Unrelated errors in different components: safetensors serialization vs PixArtAlphaPipeline bug." }, { "left": "issue:6042", "right": "issue:8790", "accept": false, - "reason": "SDXL-turbo distilling example request vs SDXL TensorRT + IP Adapter issue; different workflows and code paths." + "reason": "Different topics: SDXL-turbo distillation example vs SDXL TensorRT/IP-Adapter integration." }, { "left": "issue:11489", "right": "issue:11661", "accept": false, - "reason": "Flux LoRA training error in one script vs args.max_train_steps None bug in flux-control; both Flux-related but distinct failures in different examples." + "reason": "Both mention Flux training, but the concrete failures differ and are not the same code-path bug." }, { "left": "issue:10198", "right": "issue:9516", "accept": false, - "reason": "Flux.1-dev weight mismatch vs joint_attention_kwargs not passed to FLUX transformers; different bug class and code path." + "reason": "Weight mismatch is not the same as missing joint_attention_kwargs propagation." }, { "left": "issue:1909", "right": "issue:2098", "accept": false, - "reason": "Scheduler generator mismatch vs torch.randn argument error; unrelated API issues." + "reason": "Generator API mismatch is unrelated to the randn argument error." }, { "left": "issue:4186", "right": "issue:6686", "accept": false, - "reason": "SDXL inpaint/refiner malfunction vs Dynamo save/load performance regression; different symptoms and fixes." + "reason": "Inpaint/refiner malfunction is unrelated to Dynamo save/load performance regression." }, { "left": "issue:6313", "right": "issue:6616", "accept": false, - "reason": "Null-Text Inversion pipeline request vs classifier-guidance implementation request; separate feature ideas, not duplicates." + "reason": "Different requested features: null-text inversion pipeline vs classifier-guidance implementation." }, { "left": "issue:1709", "right": "issue:6055", "accept": false, - "reason": "ONNX img2img astype AttributeError vs accelerate error; unrelated runtime failures." + "reason": "ONNX Tensor.astype error is unrelated to an accelerate issue." }, { "left": "issue:3624", "right": "issue:5110", "accept": false, - "reason": "getattr attribute-name TypeError vs HFValidationError on repo id format; different validation paths." + "reason": "AttributeError from getattr is different from HF repo-id validation failure." }, { "left": "issue:1919", "right": "issue:5120", "accept": false, - "reason": "Missing datasets dependency in dreambooth example vs prior-preservation training bug; different root causes and fixes." + "reason": "Missing dependency in dreambooth docs is not the same as a prior-preservation training bug." }, { "left": "issue:10650", "right": "issue:12995", "accept": false, - "reason": "Lumina tensor shape error vs Cosmos-Predict2.5 pipeline behavior question; unrelated topics." + "reason": "Lumina shape invalid error is unrelated to Cosmos-Predict2.5 pipeline behavior." }, { "left": "issue:11169", "right": "issue:12645", "accept": false, - "reason": "NumPy binary incompatibility import failure vs inference speed complaint; not the same bug." + "reason": "Binary incompatibility import failure is unrelated to Qwen-Image speed concerns." }, { "left": "issue:12126", "right": "issue:12565", "accept": false, - "reason": "wan2.2 cache_context switching error vs Wan2.1 i2v running issue; different model versions and failure modes." + "reason": "Both are Wan issues, but different versions and different failures; not the same concrete bug." }, { "left": "issue:1444", "right": "issue:305", "accept": false, - "reason": "Both are import errors, but one is missing StableDiffusionSafetyChecker and the other is a generic diffusers import problem; too broad and likely different causes." + "reason": "Missing StableDiffusionSafetyChecker import is not the same as the generic diffusers import error." }, { "left": "issue:1922", "right": "issue:5689", "accept": false, - "reason": "Undefined init_git_repo in notebook vs DPM++ residual noise in SDXL; unrelated code paths." + "reason": "Undefined notebook helper is unrelated to DPM++ residual noise in SDXL." }, { "left": "issue:2838", "right": "issue:459", "accept": false, - "reason": "ControlNet pipeline usage question vs incremental diffusion concept; not the same issue." + "reason": "ControlNet pipeline behavior request is unrelated to incremental diffusion." }, { "left": "issue:3057", "right": "issue:4249", "accept": false, - "reason": "Conversion script import failure vs prompt_embeds input validation bug; different modules and behaviors." + "reason": "Conversion-script import failure is unrelated to check_inputs prompt_embeds validation." }, { "left": "issue:4094", "right": "issue:4768", "accept": false, - "reason": "Training script for T2I adapters vs saving multiple adapter models; related area but feature request vs checkpoint format support, not one concrete bug." + "reason": "Adding a T2I adapter training script is not the same as saving multiple adapters in one checkpoint." }, { "left": "issue:5485", "right": "issue:8552", "accept": false, - "reason": "ControlNet batch inference slowness vs TextToVideoZero segmentation fault; different severity and code paths." + "reason": "Batch inference slowness is unrelated to a TextToVideoZeroPipeline segfault." }, { "left": "issue:7490", "right": "issue:9331", "accept": false, - "reason": "Generic backward problem vs FlowMatchEulerDiscreteScheduler index_for_timestep bug; insufficiently specific overlap." + "reason": "Generic backward problem is unrelated to FlowMatchEulerDiscreteScheduler index_for_timestep." }, { "left": "issue:10066", "right": "issue:9974", "accept": false, - "reason": "self.attn unpacking ValueError vs controlnet_sd3 expected-parameters request; different components and error types." - }, - { - "left": "issue:4585", - "right": "issue:7147", - "accept": false, - "reason": "SDXL MultiControlNet documentation inconsistency vs StableCascadeDecoderPipeline import error; unrelated issues." - }, - { - "left": "issue:4977", - "right": "issue:5034", - "accept": false, - "reason": "Removing checkpoints in training script vs progress-bar misdisplay during checkpoint loading; different bugs in separate examples." + "reason": "Different bugs: attention unpacking error vs missing expected parameters in controlnet_sd3." } ] }, @@ -73664,13 +73678,14 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10111", "issue:10147", + "issue:10382", "issue:1055", "issue:1083", "issue:10952", @@ -73693,32 +73708,33 @@ "issue:3717", "issue:4528", "issue:4557", + "issue:4585", "issue:4972", + "issue:4977", + "issue:5034", "issue:5037", "issue:5046", - "issue:5112", "issue:5395", "issue:5432", "issue:5442", "issue:5763", "issue:5875", - "issue:6094", - "issue:6176", - "issue:6517", "issue:6548", "issue:6839", + "issue:7133", + "issue:7147", "issue:720", "issue:7243", "issue:835", "issue:8801", - "issue:8930", - "issue:9034", "issue:9090", "issue:9387", "issue:9686", "issue:9846" ], "soft_pairs": [ + "issue:4585|issue:7147", + "issue:4977|issue:5034", "issue:1055|issue:4557", "issue:4528|issue:5763", "issue:11659|issue:13162", @@ -73727,6 +73743,7 @@ "issue:1574|issue:3717", "issue:11160|issue:12097", "issue:10111|issue:9686", + "issue:10382|issue:7133", "issue:1083|issue:835", "issue:10952|issue:11732", "issue:1158|issue:2480", @@ -73739,183 +73756,180 @@ "issue:4972|issue:720", "issue:1572|issue:2799", "issue:5037|issue:5432", - "issue:10147|issue:8801", - "issue:6094|issue:6176", - "issue:6517|issue:9034", - "issue:5112|issue:8930" + "issue:10147|issue:8801" ], - "prepared_review_unit_hash": "96ef67b4de12b85d804738ef5e2337ac4bce3c4f60f5d1ffecaadabf5dae311a", + "prepared_review_unit_hash": "3e1e7b4b33cd46eb79c4d6c27810457b17a03f3fd8329c7675f5359d40e0c3d8", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33938, - "estimated_input_tokens": 8485, - "estimated_eval_tokens": 17226 + "serialized_chars": 33871, + "estimated_input_tokens": 8468, + "estimated_eval_tokens": 17192 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated bug reports and feature requests across pipelines, LoRA/adapters, ONNX, schedulers, and training scripts. None of the soft pairs look like true duplicates of the same concrete issue, so there is no strong cluster-level canonical item.", - "confidence": 0.97, - "canonical_issue_reason": "No clear canonical issue: the items are too heterogeneous, and the title overlap is mostly superficial.", + "summary": "This cluster is highly heterogeneous: most items are unrelated feature requests or distinct bug reports, and none of the soft pairs look like true duplicates of the same concrete issue. The only loose theme is LoRA/loading-related reports, but even those point to different failures and code paths.", + "confidence": 0.92, + "canonical_issue_reason": "issue:11659 is the broadest active issue in the set, but the cluster is not a genuine duplicate group; it only loosely matches the recurring LoRA-loading theme.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue cleanly represents the set; 10111 is broad, but still only covers a small slice of the cluster.", + "best_issue_reason": "issue:11659 is the most generic and umbrella-like issue title in the set, so it is the least bad representative if one must be chosen.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4585", + "right": "issue:7147", + "accept": false, + "reason": "Different problems: SDXL MultiControlNet documentation/usage vs missing import for StableCascadeDecoderPipeline." + }, + { + "left": "issue:4977", + "right": "issue:5034", + "accept": false, + "reason": "Both mention checkpoints, but one is about distributed checkpoint directories and the other about progress bar display during loading." + }, { "left": "issue:1055", "right": "issue:4557", "accept": false, - "reason": "Missing pipeline class vs. AttributeError in getattr handling; different failures and code paths." + "reason": "Different failures: missing LatentDiffusionPipeline vs AttributeError on `super().__getattr__`." }, { "left": "issue:4528", "right": "issue:5763", "accept": false, - "reason": "LoRA quality/effectiveness issue vs. save_pretrained safetensors serialization bug." + "reason": "LoRA quality/output issue vs `save_pretrained` not writing safetensors; different code paths." }, { "left": "issue:11659", "right": "issue:13162", "accept": false, - "reason": "Generic LoRA load failure vs. request for FLUX.2-klein-4b LoRA support; not the same bug." + "reason": "Both are LoRA-related, but one is a generic load failure and the other is a model-specific support request for Flux.2-klein-4b." }, { "left": "issue:5046", "right": "issue:7243", "accept": false, - "reason": "Two separate model/pipeline feature proposals (DragonDiffusion vs ResAdapter), not one underlying change." + "reason": "Different pipeline proposals: DragonDiffusion vs ResAdapter; not the same bug or change." }, { "left": "issue:5395", "right": "issue:9090", "accept": false, - "reason": "PEFT LoRA combination bug vs. PNDM timestep repetition bug; unrelated subsystems." + "reason": "LoRA combination/PEFT bug vs scheduler timestep repetition bug; unrelated failures." }, { "left": "issue:1574", "right": "issue:3717", "accept": false, - "reason": "DreamBooth on 8GB GPU failing vs. multi-GPU OOM; same broad symptom, different context and cause." + "reason": "Different resource issues in different contexts: bare-metal 8GB Dreambooth training vs multi-GPU OOM on 3090Ti." }, { "left": "issue:11160", "right": "issue:12097", "accept": false, - "reason": "WAN 2.1 inference inconsistency vs. WAN2.2 TI2V end-of-run OOM; different problems." + "reason": "Different Wan versions and different symptoms: inference inconsistency vs VRAM OOM." }, { "left": "issue:10111", "right": "issue:9686", "accept": false, - "reason": "Both mention pipeline bugs, but one is a generic FLUX report and the other is an SD3 ControlNet internal bug." + "reason": "FLUX pipeline bugs vs SD3 ControlNet pooled projection bug; different subsystems and symptoms." + }, + { + "left": "issue:10382", + "right": "issue:7133", + "accept": false, + "reason": "VAE encoder return-type AttributeError vs SDXL Dreambooth LoRA text-encoder size mismatch; not the same bug." }, { "left": "issue:1083", "right": "issue:835", "accept": false, - "reason": "FP16 ONNX incorrect output vs. flax numpy_to_pil failure; unrelated pipelines and error modes." + "reason": "FP16 ONNX output correctness vs flax pipeline `numpy_to_pil` failure; unrelated." }, { "left": "issue:10952", "right": "issue:11732", "accept": false, - "reason": "Adapter-name mismatch when loading LoRA vs. unpacking error in load_lora_weights; distinct failure signatures." + "reason": "Both are LoRA load errors, but the adapter-name mismatch and tuple-unpack error come from different failure modes." }, { "left": "issue:1158", "right": "issue:2480", "accept": false, - "reason": "Pytest test interference vs. DreamBooth checkpoint directory layout; not the same bug." + "reason": "Test interference in `test_config.py` vs Dreambooth checkpoint folder layout; unrelated." }, { "left": "issue:1205", "right": "issue:3045", "accept": false, - "reason": "Converting SD model to ONNX vs. converting LoRA safetensors to diffusers; different conversion flows." + "reason": "Different conversion tasks: SD model to ONNX vs LoRA safetensors to diffusers." }, { "left": "issue:1445", "right": "issue:5875", "accept": false, - "reason": "Different ONNX pipeline import/prompt errors in separate pipeline classes." + "reason": "Both are ONNX-related, but they describe different pipeline errors and different conversion scenarios." }, { "left": "issue:2288", "right": "issue:3065", "accept": false, - "reason": "Add InstructPix2Pix training script vs. add text-encoder LoRA tuning option; separate training features." + "reason": "Training-script implementation request vs adding a LoRA tuning option; separate feature requests." }, { "left": "issue:3391", "right": "issue:6548", "accept": false, - "reason": "Overexposure in DeepFloyd super-resolution vs. RGBA image misread as latents in img2img; different code paths." + "reason": "Different image-processing bugs: overexposure in DeepFloyd SR vs RGBA being mistaken for latents in img2img." }, { "left": "issue:5442", "right": "issue:9387", "accept": false, - "reason": "Two distinct pipeline feature requests (ScaleCrafter vs. MimicMotion), not duplicates." + "reason": "Distinct pipeline features: ScaleCrafter vs MimicMotion." }, { "left": "issue:6839", "right": "issue:9846", "accept": false, - "reason": "StableVideoDiffusion from_single_file support vs. FluxControlNetModel from_single_file support; different models." + "reason": "Different missing feature requests: from_single_file for StableVideoDiffusion vs FluxControlNetModel." }, { "left": "issue:4972", "right": "issue:720", "accept": false, - "reason": "Wuerstchen example runtime failure vs. traced UNet dtype mismatch; unrelated issues." + "reason": "Example-code failure in Wuerstchen vs traced UNet dtype mismatch; different issues." }, { "left": "issue:1572", "right": "issue:2799", "accept": false, - "reason": "StableDiffusion mega img2img/inpainting failure vs. model-checkpoint/pipeline mismatch validation; not the same bug." + "reason": "One is a pipeline capability bug for img2img/inpainting, the other is a checkpoint-vs-pipeline mismatch guard." }, { "left": "issue:5037", "right": "issue:5432", "accept": false, - "reason": "Validation image handling error vs. blank validation images; similar area, different defect." + "reason": "Second validation image argument error vs blank validation images; related area but different concrete bugs." }, { "left": "issue:10147", "right": "issue:8801", "accept": false, - "reason": "Feature requests for different models (MV-Adapter vs. Kolors), no shared underlying change." - }, - { - "left": "issue:6094", - "right": "issue:6176", - "accept": false, - "reason": "TextToVideoZeroPipeline regression vs. LoRA loading after DreamBooth training; unrelated failures." - }, - { - "left": "issue:6517", - "right": "issue:9034", - "accept": false, - "reason": "Import error for SDXL ControlNet pipeline vs. missing sentencepiece dependency; different root causes." - }, - { - "left": "issue:5112", - "right": "issue:8930", - "accept": false, - "reason": "SDXL ControlNet img2img tensor-shape error vs. poor pose-conditioning fidelity; different bug classes." + "reason": "Different model support requests: MV-Adapter multi-view generation vs Kolors support." } ] }, @@ -73934,22 +73948,18 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10107", "issue:10749", - "issue:10869", "issue:10950", "issue:10962", "issue:11010", - "issue:11022", - "issue:11035", "issue:11642", "issue:12574", - "issue:13203", "issue:1742", "issue:1823", "issue:2221", @@ -73969,25 +73979,32 @@ "issue:4680", "issue:4752", "issue:5044", + "issue:5112", "issue:5274", "issue:5510", "issue:579", "issue:5989", "issue:6087", + "issue:6094", "issue:6142", "issue:6164", + "issue:6176", + "issue:6517", "issue:7535", "issue:785", "issue:8469", "issue:8557", "issue:8752", - "issue:8826", + "issue:8930", "issue:8939", + "issue:9034", "issue:9309", - "issue:9958", "issue:9979" ], "soft_pairs": [ + "issue:6094|issue:6176", + "issue:6517|issue:9034", + "issue:5112|issue:8930", "issue:3765|issue:6087", "issue:10749|issue:11010", "issue:1823|issue:3622", @@ -74008,190 +74025,190 @@ "issue:4213|issue:8752", "issue:10107|issue:8557", "issue:10950|issue:8469", - "issue:9309|issue:9979", - "issue:11022|issue:13203", - "issue:10869|issue:8826", - "issue:11035|issue:9958" + "issue:9309|issue:9979" ], - "prepared_review_unit_hash": "6f43500fade916dab2a01352200319cec587a482d6f0c35270c99d64f6061905", + "prepared_review_unit_hash": "b387869dd2c746764aa5aa6ccbf505e7290bc0b6bb94a35f1a5ded313dcd6446", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33453, - "estimated_input_tokens": 8364, - "estimated_eval_tokens": 16984 + "serialized_chars": 33676, + "estimated_input_tokens": 8419, + "estimated_eval_tokens": 17094 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These soft links are all superficial title overlaps; the items cover unrelated bugs, feature requests, and docs issues, so none of the candidate pairs should be treated as duplicates.", - "confidence": 0.96, - "canonical_issue_reason": "issue:10869 is the clearest concrete runtime bug report in the set, but there is no strong true canonical because the cluster is heterogeneous.", + "summary": "Mostly unrelated issue set spanning docs, training, dependency, and many distinct pipeline bugs. The only plausible duplicate pair is the two Flux ControlNet reports; the rest should stay separate.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 9979 is the most concrete and actionable bug report in the only plausible duplicate subgroup, with an explicit channels error and clearer repro value.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10869 is the most specific and actionable issue to use as a representative anchor if one must be chosen.", + "best_issue_reason": "Issue 9979 is the best anchor for deduping within this cluster because it names the concrete Flux ControlNet failure mode, while the broader set has no single shared root cause.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6094", + "right": "issue:6176", + "accept": false, + "reason": "Different pipelines and failures: TextToVideoZero regression after a version update vs LoRA weight loading after DreamBooth training." + }, + { + "left": "issue:6517", + "right": "issue:9034", + "accept": false, + "reason": "One is an import traceback for an SDXL ControlNet module; the other is an optional dependency requirement for AutoPipelineForText2Image." + }, + { + "left": "issue:5112", + "right": "issue:8930", + "accept": false, + "reason": "Different bug shapes: a runtime tensor shape/type error vs a control-image adherence quality problem." + }, { "left": "issue:3765", "right": "issue:6087", "accept": false, - "reason": "ControlNet training artifact bug vs LoRA config save/load bug; different code paths and failure modes." + "reason": "Training image stacking effects in ControlNet is unrelated to saving/loading LoRA config and wrong lora_alpha." }, { "left": "issue:10749", "right": "issue:11010", "accept": false, - "reason": "GGUF support for Lumina2 pipeline is unrelated to Chroma/Flux architecture changes." + "reason": "Separate feature requests for different models and capabilities: GGUF support in Lumina2 vs architecture support for Chroma/Flux." }, { "left": "issue:1823", "right": "issue:3622", "accept": false, - "reason": "Swapped comments in depth2img docs/code is unrelated to an import error in a tutorial." + "reason": "A comments-swapped docs/content issue is unrelated to a tutorial import error." }, { "left": "issue:2228", "right": "issue:785", "accept": false, - "reason": "GPU loading issue for StableDiffusionPipeline is not the same as PNDMScheduler performance overhead." + "reason": "Loading the pipeline onto GPU and PNDMScheduler runtime cost are different problems." }, { "left": "issue:228", "right": "issue:3622", "accept": false, - "reason": "Broken release link is a documentation/release artifact problem, not the tutorial import error." + "reason": "Broken release link and tutorial import failure are unrelated." }, { "left": "issue:3624", "right": "issue:6142", "accept": false, - "reason": "Different NameError/TypeError failures in different contexts; no shared underlying bug is evident." + "reason": "Different exceptions in different code paths: getattr TypeError vs missing start_training NameError." }, { "left": "issue:3698", "right": "issue:4680", "accept": false, - "reason": "Converting an inpainting model to diffusers is a conversion request, not SDXL fine-tuning." + "reason": "Model conversion request vs fine-tuning SDXL inpainting discussion; not the same bug or change." }, { "left": "issue:4436", "right": "issue:5989", "accept": false, - "reason": "CIVITAI safetensors-to-diffusers conversion is unrelated to a safetensors.torch bug report." + "reason": "Converting CIVITAI weights to diffusers is unrelated to a safetensors.torch bug report." }, { "left": "issue:4752", "right": "issue:7535", "accept": false, - "reason": "A specific SDXL dtype mismatch is not the same as a type-hint compatibility issue." + "reason": "Runtime half/float mismatch in SDXL is not the same as a type-hint incompatibility issue." }, { "left": "issue:10962", "right": "issue:12574", "accept": false, - "reason": "CogView4 prompt-embed shape handling and WanVideo VACE control-arg handling are different pipeline bugs." + "reason": "Different model pipelines and failure modes: prompt_embed shape handling in CogView4 vs missing control-arg handling in WanVideoToVideo." }, { "left": "issue:11642", "right": "issue:8939", "accept": false, - "reason": "FluxPriorRedux prompt scaling issue is unrelated to tokenizer loading failure for SD3 models." + "reason": "Prompt scale not taking effect in FluxPriorRedux is unrelated to tokenizer loading/dependency resolution." }, { "left": "issue:3284", "right": "issue:3805", "accept": false, - "reason": "Dreambooth LoRA training failing with accelerate is not the same as a Colab subprocess failure." + "reason": "An accelerate training-script failure is not the same as a Colab CalledProcessError." }, { "left": "issue:5510", "right": "issue:6164", "accept": false, - "reason": "SDXL instruct-pix2pix dtype mismatch differs from StableDiffusionPipeline latent output handling." + "reason": "These are different pipeline regressions: SDXL instruct-pix2pix dtype mismatch vs latent output handling in StableDiffusionPipeline." }, { "left": "issue:1742", "right": "issue:2392", "accept": false, - "reason": "Inference speed variance and prompt_embeds validation error are distinct problems." + "reason": "Inference speed variation from prompt list content is unrelated to prompt_embeds validation raising NoneType errors." }, { "left": "issue:2221", "right": "issue:579", "accept": false, - "reason": "ColossalAI integration error is unrelated to missing config.json during textual inversion." + "reason": "ColossalAI integration errors and missing config.json in textual inversion are distinct issues." }, { "left": "issue:2646", "right": "issue:5044", "accept": false, - "reason": "Tiled VAE IndexError and Kandinsky 2.2 load failure are separate model/runtime issues." + "reason": "Tiled VAE IndexError is unrelated to Kandinsky 2.2 loading failure." }, { "left": "issue:4125", "right": "issue:5274", "accept": false, - "reason": "LoRA loss-curve question is not the same as half/float dtype mismatch during training." + "reason": "A question about LoRA loss curves is not the same as a dtype mismatch training error." }, { "left": "issue:4213", "right": "issue:8752", "accept": false, - "reason": "Both mention from_single_file, but one is about missing accelerate and the other about offline loading." + "reason": "Both mention from_single_file, but one is about missing accelerate while the other is about offline server loading; different root causes." }, { "left": "issue:10107", "right": "issue:8557", "accept": false, - "reason": "Gradient checkpointing during validation in examples is unrelated to SD3 token-limit training." + "reason": "Training example validation checkpointing is unrelated to SD3 token-limit constraints." }, { "left": "issue:10950", "right": "issue:8469", "accept": false, - "reason": "Broken inference docs and corrupt IP-Adapter face outputs are not the same underlying bug." + "reason": "Documentation brokenness and broken face outputs are not the same underlying bug." }, { "left": "issue:9309", "right": "issue:9979", - "accept": false, - "reason": "Flux control-net bug and Flux img2img controlnet channel mismatch are separate pipeline issues." - }, - { - "left": "issue:11022", - "right": "issue:13203", - "accept": false, - "reason": "Quanto support for WAN 2.1 transformation and Zimage LoRA support are unrelated feature/bug reports." - }, - { - "left": "issue:10869", - "right": "issue:8826", - "accept": false, - "reason": "Meta-tensor .item() failure is not the same as an empty-list torch.cat runtime error." - }, - { - "left": "issue:11035", - "right": "issue:9958", - "accept": false, - "reason": "from_single_file symlink/download support differs from AutoencoderKLTemporalDecoder tiling support." + "accept": true, + "reason": "Both point to Flux ControlNet pipeline failures on Flux 1.x, and the explicit channels error in 9979 plausibly describes the same underlying code-path bug reported more broadly in 9309." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: almost all pairs are clearly different by title/content, and the only plausible duplicate subgroup is the Flux ControlNet pair (9309/9979). The soft verdicts stay conservative overall, and the acceptance of that one pair is appropriately qualified as plausible rather than certain." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -74209,13 +74226,14 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10672", - "issue:1083", + "issue:10869", "issue:10883", "issue:10967", + "issue:11022", + "issue:11035", "issue:11222", "issue:1161", "issue:11630", - "issue:1205", "issue:12147", "issue:122", "issue:13203", @@ -74234,10 +74252,8 @@ "issue:4059", "issue:5168", "issue:5368", - "issue:5567", "issue:5576", "issue:5757", - "issue:5824", "issue:5984", "issue:599", "issue:6054", @@ -74255,9 +74271,13 @@ "issue:9216", "issue:941", "issue:9551", - "issue:9637" + "issue:9637", + "issue:9958" ], "soft_pairs": [ + "issue:11022|issue:13203", + "issue:10869|issue:8826", + "issue:11035|issue:9958", "issue:11222|issue:11630", "issue:2780|issue:756", "issue:6399|issue:6958", @@ -74279,182 +74299,186 @@ "issue:1161|issue:1798", "issue:5168|issue:7370", "issue:1451|issue:1896", - "issue:12147|issue:13203", - "issue:5567|issue:5824", - "issue:1083|issue:1205" + "issue:12147|issue:13203" ], - "prepared_review_unit_hash": "01149a2102396f4f52c01c166518b33f7cd4a2d506d476fd5551a1c5cd51b174", + "prepared_review_unit_hash": "bbe9fcf2405c2b90bec4cdb0f38a94c9e1952b7fe18f6ab81678a9113f1035e0", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 24, - "serialized_chars": 33993, - "estimated_input_tokens": 8499, - "estimated_eval_tokens": 17254 + "soft_pair_count": 25, + "serialized_chars": 34224, + "estimated_input_tokens": 8556, + "estimated_eval_tokens": 17368 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All candidate pairs are superficial matches; none look like the same underlying bug or feature change, so this cluster should not be merged as duplicates.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue: the items are thematically diverse and the proposed links only share broad words like \"support\", \"pipeline\", or \"error\".", + "summary": "These items are a mixed bag of unrelated bugs and feature requests; the soft pairs mostly share surface keywords or the same subsystem, but not the same underlying change. I would not merge any of the candidate pairs as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": "No single issue is a true canonical duplicate for this set because the cluster is heterogeneous. If one must be chosen, issue #11630 is the most actionable standalone request thanks to its explicit target API and concrete feature ask.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a strong global representative because none of the soft pairs are true duplicates or the same concrete code-path problem.", + "best_issue_reason": "#11630 is the strongest issue to keep as the representative item: it is specific, implementation-oriented, and names the exact missing support request for WanVACETransformer3DModel.from_single_file.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11022", + "right": "issue:13203", + "accept": false, + "reason": "Different models and different problems: WAN Quanto conversion support vs Zimage LoRA support." + }, + { + "left": "issue:10869", + "right": "issue:8826", + "accept": false, + "reason": "Meta tensor .item() failure is unrelated to torch.cat on an empty list." + }, + { + "left": "issue:11035", + "right": "issue:9958", + "accept": false, + "reason": "from_single_file download symlink handling is unrelated to AutoencoderKLTemporalDecoder tiling." + }, { "left": "issue:11222", "right": "issue:11630", "accept": false, - "reason": "Both concern Wan-related features, but one is generic LoRA coverage and the other is single-file loading for WanVACE; different requests and code paths." + "reason": "Both are WAN-related, but one is LoRA coverage and the other is from_single_file model loading; different feature requests." }, { "left": "issue:2780", "right": "issue:756", "accept": false, - "reason": "The first is an FP16 gradient unscale failure in training, while the second is a deprecation-tuple/value parsing error; unrelated bugs." + "reason": "FP16 gradient unscale error and deprecation tuple misuse are unrelated training/runtime issues." }, { "left": "issue:6399", "right": "issue:6958", "accept": false, - "reason": "LoRA visibility in UIs is unrelated to a motion model's use_linear_projection setting in mid_block." + "reason": "LoRA visibility in UI is unrelated to a missing use_linear_projection setting in unet_motion_model." }, { "left": "issue:852", "right": "issue:941", "accept": false, - "reason": "Both mention MPS, but one is a device_map crash on load and the other is noisy img2img output with num_images_per_prompt." + "reason": "Both mention MPS, but they are distinct bugs in different code paths and symptoms." }, { "left": "issue:2107", "right": "issue:3431", "accept": false, - "reason": "Low-resolution black images and poor inpainting+ControlNet quality are different symptoms in different pipelines." + "reason": "Low-resolution black images is a different problem from poor inpainting+ControlNet results." }, { "left": "issue:5757", "right": "issue:5984", "accept": false, - "reason": "Checkpoint-resume errors and fp16 variant RAM usage on Colab are separate training-script issues." + "reason": "Checkpoint resume errors and fp16 RAM usage are separate training issues." }, { "left": "issue:5576", "right": "issue:8309", "accept": false, - "reason": "A new scheduler request and a conditional diffusion distillation feature are not the same change." + "reason": "Different feature requests: a FreeNoise scheduler versus conditional diffusion distillation." }, { "left": "issue:7238", "right": "issue:7661", "accept": false, - "reason": "Both involve IP-Adapter, but masking behavior and enabling AnimateDiff pipeline composition are distinct problems." + "reason": "IP-Adapter masking and enabling AnimateDiff with IP-Adapter are related only by subsystem, not the same bug." }, { "left": "issue:10672", "right": "issue:7877", "accept": false, - "reason": "Callback hooks for pipelines and an Attention block out_dim argument are unrelated API changes." + "reason": "callback_on_step_end support and Attention out_dim support are unrelated API requests." }, { "left": "issue:122", "right": "issue:2856", "accept": false, - "reason": "A missing exceptions module during training is unrelated to converting an SD model to diffusers." + "reason": "Import error during training is unrelated to converting an SD model to diffusers." }, { "left": "issue:2548", "right": "issue:3296", "accept": false, - "reason": "Both are saving/loading problems, but one is LoRA finetuning save failure and the other is DreamBooth checkpoint handling with text_encoder." + "reason": "Both touch saving, but they concern different scripts and failure modes." }, { "left": "issue:599", "right": "issue:757", "accept": false, - "reason": "Deepspeed integration and textual inversion attention splicing are separate features." + "reason": "Deepspeed integration and attention splicing for textual inversion are unrelated changes." }, { "left": "issue:6054", "right": "issue:8826", "accept": false, - "reason": "A scheduler timesteps bug and a torch.cat empty-list runtime error are different code paths." + "reason": "Scheduler from_numpy misuse is unrelated to torch.cat on an empty list." }, { "left": "issue:10883", "right": "issue:9216", "accept": false, - "reason": "A Marigold update and a custom multi-model pipeline are unrelated changes." + "reason": "A Marigold update request is unrelated to a custom multi-model pipeline concept." }, { "left": "issue:10967", "right": "issue:9551", "accept": false, - "reason": "Integrating LaVie/Cinemo and using x-labs Flux ControlNet models are different model-support requests." + "reason": "Integrating LaVie/Cinemo is unrelated to using x-labs Flux ControlNet models." }, { "left": "issue:2676", "right": "issue:4059", "accept": false, - "reason": "A scheduler-naming question is not the same as a ControlNet reference pipeline bug or the proposed inpaint pipeline implementation." + "reason": "Scheduler naming question is unrelated to the ControlNet ReferencePipeline bug/question." }, { "left": "issue:3087", "right": "issue:3974", "accept": false, - "reason": "LyCORIS model support and .safetensors support in ControlNet inpaint are unrelated issues." + "reason": "LyCORIS support and ControlNet inpaint safetensors handling are different requests." }, { "left": "issue:5368", "right": "issue:9637", "accept": false, - "reason": "Both are attention/mixed-precision errors, but one is a dtype mismatch in fp16 and the other is an XFormer attention-mask edge case with bfloat16 and sequence length." + "reason": "Both involve attention dtypes, but one is fp16 QKV dtype mismatch and the other is xFormers mask/bfloat/sequence-length behavior." }, { "left": "issue:1161", "right": "issue:1798", "accept": false, - "reason": "Converting Flax models to PyTorch is different from adding v_prediction support to a Flax DreamBooth script." + "reason": "General Flax-to-PyTorch conversion is different from adding v_prediction support to a Flax training script." }, { "left": "issue:5168", "right": "issue:7370", "accept": false, - "reason": "ControlNet-XS support and OpenSora STDiT support are different model integration requests." + "reason": "ControlNet-XS support and OpenSora STDiT support are unrelated model integrations." }, { "left": "issue:1451", "right": "issue:1896", "accept": false, - "reason": "A FlaxStableDiffusionPipeline example failure and a CLIP-guided SD dict-size error do not describe the same bug." + "reason": "Flax pipeline example failure is unrelated to the CLIP-guided size/type error." }, { "left": "issue:12147", "right": "issue:13203", "accept": false, - "reason": "Wan2.2 Lightning LoRA loading and Zimage LoRA support are different model families and different issues." - }, - { - "left": "issue:5567", - "right": "issue:5824", - "accept": false, - "reason": "An accelerator.prepare config overwrite bug is unrelated to the text_encoder_projection_dim default in get_add_time_ids()." - }, - { - "left": "issue:1083", - "right": "issue:1205", - "accept": false, - "reason": "FP16 ONNX incorrect output and converting Stable Diffusion to ONNX are not the same problem." + "reason": "Wan2.2 Lightning LoRA loading failure is unrelated to Zimage LoRA support." } ] }, @@ -74473,19 +74497,19 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10057", - "issue:10066", "issue:10120", + "issue:1083", "issue:1106", "issue:1109", - "issue:1158", "issue:11859", "issue:11978", + "issue:1205", "issue:12412", "issue:1556", "issue:1811", @@ -74493,7 +74517,6 @@ "issue:2822", "issue:3405", "issue:3683", - "issue:3886", "issue:3891", "issue:4149", "issue:4186", @@ -74505,9 +74528,11 @@ "issue:5128", "issue:5258", "issue:542", + "issue:5567", "issue:5595", "issue:5628", "issue:5646", + "issue:5824", "issue:5989", "issue:6347", "issue:6368", @@ -74519,7 +74544,6 @@ "issue:8667", "issue:8873", "issue:8970", - "issue:9471", "issue:9645", "issue:9664", "issue:9763", @@ -74528,6 +74552,8 @@ "issue:9957" ], "soft_pairs": [ + "issue:5567|issue:5824", + "issue:1083|issue:1205", "issue:1106|issue:542", "issue:11859|issue:9957", "issue:1811|issue:4665", @@ -74549,182 +74575,180 @@ "issue:4773|issue:5128", "issue:6368|issue:9793", "issue:2822|issue:4269", - "issue:5628|issue:5646", - "issue:1158|issue:3886", - "issue:10066|issue:9471" + "issue:5628|issue:5646" ], - "prepared_review_unit_hash": "b14754ec025f4f2d1b81032644825b47e3d6cd3d70f37661d56c122a5269b581", + "prepared_review_unit_hash": "812310177adcdb845fbdc99efa7a6e99fc8e6961ec0f4d305921ab1c387940d7", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33790, - "estimated_input_tokens": 8448, - "estimated_eval_tokens": 17152 + "serialized_chars": 33684, + "estimated_input_tokens": 8421, + "estimated_eval_tokens": 17098 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Highly heterogeneous cluster: the soft links are mostly false positives across unrelated bugs, feature requests, and model/pipeline topics. No PRs are present, and I would not merge any of the proposed soft pairs.", - "confidence": 0.74, - "canonical_issue_reason": "Issue 9763 is the most substantive standalone report in the set, with the richest reproduction details and discussion around a concrete Flux ControlNet token-truncation bug. The rest of the cluster is too mixed to form a cleaner canonical center.", + "summary": "All candidates appear to be distinct issues rather than true duplicates. The cluster is noisy: many items mention similar subsystems (Flux/SDXL/MPS/schedulers/loading), but the concrete failure modes and targets differ.", + "confidence": 0.93, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "9763 is the strongest representative issue if one must be chosen: detailed, specific, and actively discussed. It has far more diagnostic value than the shorter feature requests or loosely related error reports nearby.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5567", + "right": "issue:5824", + "accept": false, + "reason": "Different bugs: one is accelerator.prepare overwriting unet.config, the other is an SDXL time-ids default value problem." + }, + { + "left": "issue:1083", + "right": "issue:1205", + "accept": false, + "reason": "Both involve ONNX, but one reports incorrect FP16 ONNX output while the other is about converting Stable Diffusion to ONNX." + }, { "left": "issue:1106", "right": "issue:542", "accept": false, - "reason": "Different problems: PNDM calling UNet repeatedly vs an ONNX export issue for UNet. Only broad subsystem overlap." + "reason": "Unrelated code paths: repeated UNet calls in PNDM vs a UNet ONNX export issue." }, { "left": "issue:11859", "right": "issue:9957", "accept": false, - "reason": "Both are scheduler feature requests, but they describe different schedulers and different changes, not the same bug or PR-sized fix." + "reason": "Both are scheduler ideas, but they are different scheduler implementations and not the same change." }, { "left": "issue:1811", "right": "issue:4665", "accept": false, - "reason": "Both involve weight-file selection, but one is about downloading both file types and the other is a Windows safetensors failure with alternative bin files. Not the same concrete issue." + "reason": "Related to weight-file selection, but one is about downloading both bin and safetensors, the other about use_safetensors failing on Windows with alternate bin files." }, { "left": "issue:3405", "right": "issue:3683", "accept": false, - "reason": "ControlNet training bug vs implementing save_pretrained for Multi-ControlNet; unrelated change requests." + "reason": "Training bug vs serialization feature request; not the same underlying issue." }, { "left": "issue:5989", "right": "issue:6899", "accept": false, - "reason": "One is a safetensors.torch bug, the other is a converter script bug. Different code paths and symptoms." + "reason": "One is a safetensors.torch bug, the other is a converter-script bug; different failure points." }, { "left": "issue:9645", "right": "issue:9942", "accept": false, - "reason": "ImportError for a missing mixin vs installation/package version failure. Not the same defect." + "reason": "Import error for a loader mixin vs pip installation/versioning failure; distinct problems." }, { "left": "issue:5258", "right": "issue:7931", "accept": false, - "reason": "Both touch SDXL controlnet/inpaint, but one is refiner interaction and the other is IP-Adapter embeds handling. Distinct bugs." + "reason": "Different pipeline features: SDXL ControlNet inpaint with refiner vs IP-Adapter image_embeds support." }, { "left": "issue:10120", "right": "issue:6536", "accept": false, - "reason": "Different OOM reports in different training contexts and model classes; not the same underlying memory bug." + "reason": "Both are memory-related, but one is DreamBooth/SD3 memory consumption and the other is UNet3DConditionModel OOM during fine-tuning." }, { "left": "issue:1109", "right": "issue:1556", "accept": false, - "reason": "CPU core utilization question vs black images on GPU/CUDA with NSFW warning. Unrelated." + "reason": "CPU utilization complaint vs black images with NSFW warning; unrelated symptoms and code paths." }, { "left": "issue:11978", "right": "issue:12412", "accept": false, - "reason": "Both are model-support requests, but for different models and pipelines. Not a single mergeable change." + "reason": "Different model-support requests: Bria 3.2 pipeline vs HunyuanImage-3.0 support." }, { "left": "issue:8526", "right": "issue:8970", "accept": false, - "reason": "Both mention MPS, but one is ineffective scale passing in SD3 and the other is black output in Kolors. Different bugs." + "reason": "Both mention MPS, but one is ineffective joint_attention_kwargs scale on SD3 and the other is black images in Kolors." }, { "left": "issue:4186", "right": "issue:4530", "accept": false, - "reason": "Same family/refiner theme, but one is functional inpaint failure and the other is visual artifacts. Not the same issue." + "reason": "Both involve SDXL refiner, but one is inpaint-with-refiner failure and the other is artifacting in generated images." }, { "left": "issue:2514", "right": "issue:4217", "accept": false, - "reason": "Both are loading slowdowns, but one is cached pipeline loading and the other is safetensors-specific load time. Too broad to unify." + "reason": "Both concern loading speed, but one is cached pipeline load time and the other is safetensors load time." }, { "left": "issue:3891", "right": "issue:6347", "accept": false, - "reason": "Different pipeline additions (Tortoise TTS vs GlueGen). No shared underlying change." + "reason": "Different model additions: Tortoise TTS pipeline vs GlueGen." }, { "left": "issue:4149", "right": "issue:8667", "accept": false, - "reason": "Both are tensor mismatch reports, but for different pipelines and different triggers. Not the same bug." + "reason": "Different tensor-mismatch triggers and models: SemanticStableDiffusion batch handling vs SD3ControlNet non-power-of-two image sizes." }, { "left": "issue:8873", "right": "issue:9664", "accept": false, - "reason": "Performance complaint with xformers vs a training subprocess error. No concrete overlap." + "reason": "Performance complaint with xformers on A100 vs a training subprocess failure; no shared bug." }, { "left": "issue:5595", "right": "issue:7686", "accept": false, - "reason": "Height/width validation error vs cross_attention_dim configuration error. Different validation paths." + "reason": "Different validation errors: image size divisibility vs missing cross_attention_dim for a UNet block." }, { "left": "issue:10057", "right": "issue:9763", "accept": false, - "reason": "Both mention Flux and tokens, but one is a from_single_file loading/token issue and the other is training truncation at 77 tokens. Different concrete failures." + "reason": "Both mention Flux and token limits, but one is a from_single_file loading/token issue while the other is training truncation beyond 77 tokens." }, { "left": "issue:4773", "right": "issue:5128", "accept": false, - "reason": "Both are loading-related, but one is CivitAI LoRA loading and the other is local .safetensors via XLPipeline.from_single_file. Not the same bug." + "reason": "Different loading scenarios: CivitAI LoRA loading vs local .safetensors loading via XLPipeline.from_single_file." }, { "left": "issue:6368", "right": "issue:9793", "accept": false, - "reason": "Checkpoint layer-support error vs progress bar not updating. Unrelated." + "reason": "Checkpoint/lora alpha support error vs progress bar update problem; unrelated." }, { "left": "issue:2822", "right": "issue:4269", "accept": false, - "reason": "Training convergence difference vs TypeError in check_inputs; different failure modes and code paths." + "reason": "Training convergence issue vs a TypeError in check_inputs; not the same bug." }, { "left": "issue:5628", "right": "issue:5646", "accept": false, - "reason": "Both are scheduler-related, but they involve different schedulers, different symptoms, and different models. Too far apart to merge." - }, - { - "left": "issue:1158", - "right": "issue:3886", - "accept": false, - "reason": "Test interference vs distributed training timeout. Unrelated." - }, - { - "left": "issue:10066", - "right": "issue:9471", - "accept": false, - "reason": "Attention unpacking ValueError vs file-extension inference error. Different root causes." + "reason": "Both scheduler-related, but they describe different schedulers and different failure modes." } ] }, @@ -74748,6 +74772,7 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10066", "issue:10315", "issue:1055", "issue:10653", @@ -74756,12 +74781,10 @@ "issue:10749", "issue:10757", "issue:10782", - "issue:10887", "issue:10962", "issue:11005", - "issue:11022", + "issue:1158", "issue:11779", - "issue:11901", "issue:12299", "issue:1895", "issue:2026", @@ -74769,6 +74792,7 @@ "issue:3761", "issue:3805", "issue:3816", + "issue:3886", "issue:4314", "issue:4492", "issue:4840", @@ -74789,14 +74813,16 @@ "issue:9052", "issue:9142", "issue:9181", - "issue:9241", "issue:9422", + "issue:9471", "issue:962", "issue:9704", "issue:9850", "issue:9867" ], "soft_pairs": [ + "issue:1158|issue:3886", + "issue:10066|issue:9471", "issue:10653|issue:10659", "issue:5002|issue:5907", "issue:1895|issue:4492", @@ -74818,182 +74844,180 @@ "issue:1055|issue:3816", "issue:10674|issue:10749", "issue:10757|issue:7147", - "issue:10782|issue:11005", - "issue:10887|issue:9241", - "issue:11022|issue:11901" + "issue:10782|issue:11005" ], - "prepared_review_unit_hash": "f9cc45d24642c393f3f98a14e3714ae6c792ac8b202591c13626822814fdbaec", + "prepared_review_unit_hash": "a947f8f55feea4732520c119f92b283071c86f474d4c85b0cfe0dcf92ee21aea", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33464, - "estimated_input_tokens": 8366, - "estimated_eval_tokens": 16988 + "serialized_chars": 33564, + "estimated_input_tokens": 8391, + "estimated_eval_tokens": 17038 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set; it mixes many unrelated issues across training, conversion, scheduler APIs, pipeline support, and masking/loading bugs. No soft edge pair is safe to merge.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue is suitable: the items are heterogeneous and do not describe one underlying bug or change.", - "canonical_pr_reason": "No PRs are present in this cluster.", - "best_issue_reason": "No issue is a strong global representative because the cluster is too broad and internally unrelated.", - "best_pr_reason": "No PRs are present in this cluster.", + "summary": "These items are mostly unrelated bug reports and feature requests from different pipelines, schedulers, and tooling paths. None of the soft pairs look like the same underlying issue, so all candidate duplicate links should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": "issue:10066 is the most concrete, actionable bug report in the set (specific traceback, clear failure mode), but there is no strong evidence that any other item is a true duplicate of it.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "issue:10066 is the best representative issue only because it is specific and well-scoped; however, the cluster overall does not form a coherent duplicate set.", + "best_pr_reason": "No pull requests are present, so there is no suitable canonical PR.", "soft_edge_verdicts": [ + { + "left": "issue:1158", + "right": "issue:3886", + "accept": false, + "reason": "Different failures: test interference vs distributed training collective timeout; not the same bug." + }, + { + "left": "issue:10066", + "right": "issue:9471", + "accept": false, + "reason": "Unpack error in attention code vs file-extension inference error; unrelated code paths." + }, { "left": "issue:10653", "right": "issue:10659", "accept": false, - "reason": "Bitsandbytes CPU offload error for HunyuanVideoTransformer3DModel vs poor ConsisID output quality; different bugs and code paths." + "reason": "bitsandbytes 8-bit device-move limitation vs poor output quality in ConsisID; different problems." }, { "left": "issue:5002", "right": "issue:5907", "accept": false, - "reason": "Both concern scheduler arguments, but one is missing `use_karras_sigmas` and the other is a `prediction_type` docs/API mismatch in a different scheduler." + "reason": "Both mention scheduler arguments/docs, but one is missing an argument while the other disagrees with docs; not the same concrete bug." }, { "left": "issue:1895", "right": "issue:4492", "accept": false, - "reason": "EMA multi-GPU update bug vs `false` NameError in config/script parsing; unrelated." + "reason": "Multi-GPU EMA update issue vs JSON-style lowercase boolean NameError; unrelated." }, { "left": "issue:9142", "right": "issue:9704", "accept": false, - "reason": "Multiple controlnets loading error vs cuDNN frontend execution-plan failure; distinct failures." + "reason": "Multiple controlnets attribute error vs cuDNN execution-plan error; separate runtime failures." }, { "left": "issue:4840", "right": "issue:9052", "accept": false, - "reason": "FlaxAutoencoderKL checkpoint loading bug vs `from_single_file` failing on certain Civit checkpoints; both loading-related but different conversion paths." + "reason": "Checkpoint loading mismatch in FlaxAutoencoderKL vs single-file conversion failing on certain checkpoints; different loaders and symptoms." }, { "left": "issue:5737", "right": "issue:9850", "accept": false, - "reason": "SDXL conversion failure vs gradient checkpointing with frozen model; unrelated functionality." + "reason": "SDXL conversion failure vs gradient checkpointing with frozen model; no shared bug." }, { "left": "issue:2026", "right": "issue:2891", "accept": false, - "reason": "cpu_offload failure in PaintByExample vs docs/support mismatch for embeds in UpscalePipeline; not the same bug." + "reason": "cpu_offload issue vs embeds support/documentation mismatch; different pipeline behaviors." }, { "left": "issue:3761", "right": "issue:4314", "accept": false, - "reason": "Community pipelines loading conflict vs interrupted-download write-protection; unrelated." + "reason": "Community pipeline loading conflict vs interrupted-download write protection; unrelated filesystem/runtime issues." }, { "left": "issue:3805", "right": "issue:5984", "accept": false, - "reason": "Colab CalledProcessError vs fp16 memory/RAM issue in dreambooth; different symptoms and root causes." + "reason": "Colab subprocess failure vs fp16 RAM usage problem; not the same failure mode." }, { "left": "issue:5016", "right": "issue:5984", "accept": false, - "reason": "SDXL LoRA text-encoder training limitation vs fp16 RAM issue; unrelated training problems." + "reason": "Text encoder training not working vs RAM blow-up from fp16 variant; different training issues." }, { "left": "issue:10962", "right": "issue:12299", "accept": false, - "reason": "Prompt-embedding shape issue in Cogview4 vs negative-prompt embedding tensor mismatch in SD3; both embedding-related but not the same concrete bug." + "reason": "Prompt-embedding shape issue in Cogview4 vs negative prompt embedding mismatch in SD3; similar theme but different pipelines and bugs." }, { "left": "issue:5630", "right": "issue:9181", "accept": false, - "reason": "LMS/Karras denoising leftover noise vs a noise dot in ControlNet pipeline; different pipelines and symptoms." + "reason": "LMS/Karras denoising offset vs ControlNet noise dot artifact; different scheduler/pipeline symptoms." }, { "left": "issue:5628", "right": "issue:8792", "accept": false, - "reason": "LMS scheduler offset with Karras sigmas vs general image-quality degradation after migration; too broad and not the same defect." + "reason": "LMS denoising-end/start offset with Karras sigmas vs image quality degradation migrating from A1111; not the same concrete defect." }, { "left": "issue:801", "right": "issue:962", "accept": false, - "reason": "Attention slicing behavior in inpaint/img2img vs new inpaint pipelines only supporting 512x512; unrelated pipeline limitations." + "reason": "Attention slicing not respected in inpaint/img2img vs inpaint pipelines not supporting non-512 sizes; different functional gaps." }, { "left": "issue:9023", "right": "issue:9422", "accept": false, - "reason": "Generic sample-code runtime failure vs NaNs during custom ControlNet SDXL training; unrelated." + "reason": "Sample code failure vs NaNs during ControlNet-SDXL training; too broad and not the same bug." }, { "left": "issue:10315", "right": "issue:8514", "accept": false, - "reason": "CogVideo training error vs EMAModel/InstructP2P parallel fine-tuning issue; different models and failure modes." + "reason": "CogVideo training error vs EMAModel parallel fine-tuning error; different training stacks and failure causes." }, { "left": "issue:11779", "right": "issue:9867", "accept": false, - "reason": "Wan VACE mask padding bug vs FluxInpaintPipeline overwriting pixels outside the mask; both mask-related but different code paths and root causes." + "reason": "Mask padding bug in Wan VACE vs FluxInpaint overwriting pixels outside mask; both are inpainting-related but distinct defects." }, { "left": "issue:6427", "right": "issue:7621", "accept": false, - "reason": "Model conversion .diffusers<->.ckpt issue vs SDXL InstructPix2Pix cosxl_edit incompatibility; unrelated." + "reason": "diffusers-to-.ckpt conversion problem vs InstructPix2Pix with cosxl_edit compatibility issue; unrelated conversion vs inference bugs." }, { "left": "issue:1055", "right": "issue:3816", "accept": false, - "reason": "Missing `LatentDiffusionPipeline` vs missing `transformers.CLIPImageProcessor`; both import errors but for different symbols and packages." + "reason": "Missing LatentDiffusionPipeline import availability vs missing CLIPImageProcessor attribute in transformers; different dependency/API issues." }, { "left": "issue:10674", "right": "issue:10749", "accept": false, - "reason": "Flux GGUF support failure vs Lumina2 GGUF support request; same general theme, but different pipelines/models and not one concrete bug." + "reason": "Flux GGUF support failure vs request to add GGUF support in Lumina2; one is a bug report, the other a feature request for a different pipeline." }, { "left": "issue:10757", "right": "issue:7147", "accept": false, - "reason": "Variable rename in conversion script vs missing `StableCascadeDecoderPipeline` import; unrelated." + "reason": "Variable naming cleanup in a converter script vs missing StableCascadeDecoderPipeline import; unrelated." }, { "left": "issue:10782", "right": "issue:11005", "accept": false, - "reason": "Lumina Image 2.0 minor issue vs WAN pipeline default/docstring discrepancy; unrelated." - }, - { - "left": "issue:10887", - "right": "issue:9241", - "accept": false, - "reason": "Missing `StableDiffusionXLTrainer` import vs XLab LoRA conversion script request; unrelated." - }, - { - "left": "issue:11022", - "right": "issue:11901", - "accept": false, - "reason": "Quanto support for WAN 2.1 transformation vs loading a specific LoRA into a WAN2.1 pipeline; different model-loading/quantization problems." + "reason": "Lumina Image minor issue vs docstring/default discrepancy in wan i2v args; different scopes and code paths." } ] }, @@ -75017,11 +75041,14 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10887", + "issue:11022", "issue:1117", "issue:11214", "issue:11321", "issue:11833", "issue:11867", + "issue:11901", "issue:12071", "issue:1379", "issue:1440", @@ -75047,26 +75074,25 @@ "issue:337", "issue:3412", "issue:3444", - "issue:3487", - "issue:3730", "issue:3788", "issue:3951", "issue:4057", "issue:4405", "issue:4580", "issue:5026", - "issue:6331", "issue:6548", - "issue:6949", "issue:7119", "issue:713", "issue:7392", "issue:7798", "issue:8887", + "issue:9241", "issue:9320", "issue:938" ], "soft_pairs": [ + "issue:10887|issue:9241", + "issue:11022|issue:11901", "issue:1117|issue:2613", "issue:11214|issue:12071", "issue:11321|issue:7798", @@ -75088,182 +75114,180 @@ "issue:3065|issue:4405", "issue:3176|issue:3951", "issue:3291|issue:7392", - "issue:3444|issue:3788", - "issue:3487|issue:6949", - "issue:3730|issue:6331" + "issue:3444|issue:3788" ], - "prepared_review_unit_hash": "53ecdb58c29f4505923a7934c80445acf441520c9c837bd204d0e8492fb8828f", + "prepared_review_unit_hash": "63f69cc433ab9aec4c9b3b337513b4a16b8b78f1713b81a9de23a0be84f89d9a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33458, - "estimated_input_tokens": 8365, - "estimated_eval_tokens": 16986 + "serialized_chars": 33484, + "estimated_input_tokens": 8371, + "estimated_eval_tokens": 16998 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is a heterogeneous mix of unrelated bugs, feature requests, docs, and questions. None of the soft pairs look like true duplicates of the same underlying issue, so no canonical item stands out.", - "confidence": 0.97, - "canonical_issue_reason": "No clear canonical issue: the items span many unrelated topics and failure modes, so there is no single issue that represents the cluster well.", + "summary": "This cluster is highly heterogeneous: the suggested pairs span unrelated bugs, feature requests, docs, and model/pipeline-specific issues. None of the soft pairs look like the same underlying change or defect, so all should be rejected.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 11214 is the most concrete, self-contained bug report in the set (specific runtime TypeError with a clear code path). It is the best representative issue only in the weak sense that the cluster does not form a real duplicate group.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a strong global anchor for this cluster because the entries are not a duplicate set and cover unrelated subsystems and request types.", + "best_issue_reason": "Issue 11214 is the most suitable single issue to anchor triage because it is open, specific, and actionable; the other items are mostly unrelated or more like feature requests/docs than duplicate bug reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10887", + "right": "issue:9241", + "accept": false, + "reason": "Different topics: SDXLTrainer import error vs XLab-to-diffusers LoRA conversion script." + }, + { + "left": "issue:11022", + "right": "issue:11901", + "accept": false, + "reason": "Both mention WAN 2.1, but one is Quanto transformation support and the other is LoRA loading on MPS; different failure modes and code paths." + }, { "left": "issue:1117", "right": "issue:2613", "accept": false, - "reason": "Test coverage request for negative prompts vs feature request for multiple LoRA layers; different change scope." + "reason": "Negative prompts tests vs adding multiple LoRA layers; unrelated scope." }, { "left": "issue:11214", "right": "issue:12071", "accept": false, - "reason": "Different model/code paths: UVit2DModel encoder_hidden_states TypeError vs flux kontext transformer forward behavior change." + "reason": "Different runtime bugs in different models/components: UVit2DModel.forward keyword handling vs Flux Kontext single-block behavior." }, { "left": "issue:11321", "right": "issue:7798", "accept": false, - "reason": "README bug for flux controlnet training vs PixArtAlphaPipeline bug; unrelated issues." + "reason": "Flux ControlNet README bug is documentation, while PixArtAlphaPipeline bug is model/runtime behavior." }, { "left": "issue:11833", "right": "issue:9320", "accept": false, - "reason": "LoRA device KeyError vs custom diffusion save/load bug; distinct model-management failures." + "reason": "LoRA device placement KeyError is a runtime state bug; train_custom_diffusion save/load is a separate persistence issue." }, { "left": "issue:11867", "right": "issue:8887", "accept": false, - "reason": "A question about control SD3 vs a specific wrong VAE encode training bug; not the same problem." + "reason": "Question about SD3 control vs SD3 VAE encode bug; not the same defect." }, { "left": "issue:1379", "right": "issue:3062", "accept": false, - "reason": "Repeated operation in Transformer2DModel vs a docstring correctness question; one is code behavior, the other documentation." + "reason": "Repeated operation in Transformer2DModel vs a docstring question about encoder_hidden_states; one is code behavior, the other docs." }, { "left": "issue:1440", "right": "issue:2614", "accept": false, - "reason": "Deleting cached model vs changing pretrained-model download directory; unrelated cache/config topics." + "reason": "Deleting cached model vs changing pretrained-model download directory; unrelated utilities." }, { "left": "issue:1611", "right": "issue:5026", "accept": false, - "reason": "Both are regressions, but they affect different pipelines and error signatures, so they are not the same bug." + "reason": "StableDiffusionMegaPipeline regression vs KeyError 'text_encoder' in 0.21.0; different pipeline breakages." }, { "left": "issue:1730", "right": "issue:4057", "accept": false, - "reason": "Converted ckpt compatibility problem vs blurred finetuned inpainting output; different symptoms and likely causes." + "reason": "CKPT conversion/webui compatibility vs blurred finetuned inpainting output; different problems." }, { "left": "issue:1805", "right": "issue:938", "accept": false, - "reason": "General diffusion pipeline request vs text-to-motion model proposal; different feature areas." + "reason": "New scalable diffusion pipeline proposal vs human motion diffusion model proposal; distinct features." }, { "left": "issue:1896", "right": "issue:337", "accept": false, - "reason": "CLIP-guided Stable Diffusion size/dict error vs img2img/text-guided inpainting problem; different failing code paths." + "reason": "CLIP-guided Stable Diffusion size error vs img2img/inpainting issue; different code paths." }, { "left": "issue:1985", "right": "issue:2838", "accept": false, - "reason": "Loading new embeddings vs asking for a ControlNet pipeline variant; unrelated functionality." + "reason": "Embedding-loading feature request vs ControlNet pipeline behavior question; not the same change." }, { "left": "issue:1990", "right": "issue:2015", "accept": false, - "reason": "Two separate pipeline proposals for different tasks; not the same underlying change." + "reason": "Two separate pipeline proposals (Depth2ImgVariation vs categorical diffusion); unrelated." }, { "left": "issue:2312", "right": "issue:4580", "accept": false, - "reason": "EMA save-hook error during training vs UniPC scheduler exception with karras sigmas; different components." + "reason": "EMA save hook error in training vs UniPC schedule exception with karras sigmas; unrelated scheduler/training bugs." }, { "left": "issue:2351", "right": "issue:6548", "accept": false, - "reason": "Missing get_velocity on a scheduler vs RGBA latents handling in img2img; separate bugs." + "reason": "Missing get_velocity in DEISMultistepScheduler vs RGBA image misclassified as latents; different defects." }, { "left": "issue:2412", "right": "issue:713", "accept": false, - "reason": "Dreambooth-specific training error vs a generic training error report; too broad and not the same failure." + "reason": "Dreambooth error vs generic training error; too broad and not the same underlying issue." }, { "left": "issue:2689", "right": "issue:7119", "accept": false, - "reason": "Broken image link vs invalid docs link; both documentation issues but not duplicates." + "reason": "Broken image link vs invalid docs link; both are link issues but not the same artifact or fix." }, { "left": "issue:3017", "right": "issue:3412", "accept": false, - "reason": "Different documentation requests for different topics; not the same doc change." + "reason": "Distributed inference docs request vs stable diffusion next-steps docs update; separate documentation tasks." }, { "left": "issue:3065", "right": "issue:4405", "accept": false, - "reason": "DreamBooth LoRA text-encoder tuning vs SDXL textual inversion training script; different examples and objectives." + "reason": "DreamBooth LoRA text-encoder tuning vs SDXL textual inversion training script; distinct example scripts and goals." }, { "left": "issue:3176", "right": "issue:3951", "accept": false, - "reason": "TensorRT demo error vs UniDiffuser NaNs; unrelated models and failure modes." + "reason": "TensorRT demo error vs UniDiffuser NaNs; different subsystems and failure symptoms." }, { "left": "issue:3291", "right": "issue:7392", "accept": false, - "reason": "Both mention aspect ratios, but one is about DeepFloyd and the other SDXL training; not the same bug." + "reason": "Deepfloyd aspect ratios vs SDXL multiple aspect ratio training; similar theme but different models and requests." }, { "left": "issue:3444", "right": "issue:3788", "accept": false, - "reason": "Audio-to-image integration request vs a new controllable generation pipeline proposal; distinct features." - }, - { - "left": "issue:3487", - "right": "issue:6949", - "accept": false, - "reason": "Different scheduler-related generation issues with different symptoms; not one concrete bug." - }, - { - "left": "issue:3730", - "right": "issue:6331", - "accept": false, - "reason": "Wrong timestep in inpainting blending vs SAG for inpainting; related area but different underlying problems." + "reason": "Audio-to-image integration request vs UniControl new pipeline proposal; unrelated feature work." } ] }, @@ -75287,6 +75311,8 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:3487", + "issue:3730", "issue:3765", "issue:3805", "issue:3826", @@ -75314,13 +75340,13 @@ "issue:6132", "issue:6141", "issue:6242", + "issue:6331", "issue:6354", "issue:6359", "issue:6439", - "issue:6481", "issue:6524", - "issue:6725", "issue:6928", + "issue:6949", "issue:6958", "issue:7341", "issue:7405", @@ -75330,13 +75356,13 @@ "issue:8540", "issue:8588", "issue:8906", - "issue:9350", "issue:9396", - "issue:9488", "issue:962", "issue:9958" ], "soft_pairs": [ + "issue:3487|issue:6949", + "issue:3730|issue:6331", "issue:3765|issue:5956", "issue:3805|issue:4872", "issue:3826|issue:7405", @@ -75358,182 +75384,180 @@ "issue:6242|issue:7879", "issue:6354|issue:9958", "issue:6359|issue:7341", - "issue:6439|issue:9396", - "issue:6481|issue:9350", - "issue:6725|issue:9488" + "issue:6439|issue:9396" ], - "prepared_review_unit_hash": "a383357d5227787a3ce26f983d867bdb83d40905dd4f1501393abfcf8cf4cae7", + "prepared_review_unit_hash": "7efa34e0864eda513d2472fe3917d227767703a800986a7b661e9305764e2a3f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33331, - "estimated_input_tokens": 8333, - "estimated_eval_tokens": 16922 + "serialized_chars": 33302, + "estimated_input_tokens": 8326, + "estimated_eval_tokens": 16908 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated feature requests, docs issues, training bugs, and model/pipeline-specific problems. None of the soft-similarity pairs look like the same underlying bug or change, so all should be rejected as duplicates.", + "summary": "All candidate soft edges are false positives: the cluster is a mix of unrelated issues spanning inpainting, schedulers, training, docs, datasets, and new pipeline/model requests.", "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3487", + "right": "issue:6949", + "accept": false, + "reason": "Both concern generation quality, but one is about inpaint strength with EulerAncestral and the other is blank outputs from DPMSolverSinglestep at even steps; different scheduler bugs." + }, + { + "left": "issue:3730", + "right": "issue:6331", + "accept": false, + "reason": "Inpainting time-step blending bug vs a SAG feature/request for inpainting; not the same underlying problem." + }, { "left": "issue:3765", "right": "issue:5956", "accept": false, - "reason": "Both mention SD-XL training/image degradation, but one is about ControlNet stacking artifacts and the other about U-Net fine-tuning performance; different failure modes and likely different fixes." + "reason": "Both mention training quality, but one is ControlNet stacking artifacts during training and the other is degraded SDXL U-Net fine-tuning performance; different issues." }, { "left": "issue:3805", "right": "issue:4872", "accept": false, - "reason": "Colab CalledProcessError is an execution/setup failure, while the other is GPU out-of-memory during SDXL finetuning; unrelated causes." + "reason": "Both involve Colab failures, but one is a generic CalledProcessError and the other is CUDA OOM in SDXL finetuning; different failure modes." }, { "left": "issue:3826", "right": "issue:7405", "accept": false, - "reason": "Both are new model/pipeline requests, but for different audio/video workflows and no shared code-path problem." + "reason": "Both are new model/pipeline proposals, but they target completely different modalities: audio diffusion vs video diffusion." }, { "left": "issue:3891", "right": "issue:6132", "accept": false, - "reason": "Tortoise TTS pipeline request is unrelated to the nn.Module LOADABLE_CLASSES issue." + "reason": "Tortoise TTS pipeline request vs adding LOADABLE_CLASSES support in PyTorch serialization; unrelated changes." }, { "left": "issue:4075", "right": "issue:4803", "accept": false, - "reason": "A docstring typo is not the same as a missing requirements file error." + "reason": "Minor ShapE docstring typo vs a missing requirements file error; unrelated." }, { "left": "issue:4186", "right": "issue:962", "accept": false, - "reason": "SDXL inpaint with refiner and old inpaint-size limitations are different pipeline bugs in different generations." + "reason": "Both are inpainting-related, but one is SDXL inpaint with refiner and the other is a 512x512 size limitation in older inpaint pipelines; different bugs." }, { "left": "issue:4213", "right": "issue:6055", "accept": false, - "reason": "SDXL from_single_file without accelerate and a generic accelerate error are not the same concrete bug." + "reason": "SD-XL from_single_file() needing accelerate vs a generic accelerate error; too vague and not the same bug." }, { "left": "issue:4524", "right": "issue:5866", "accept": false, - "reason": "AnimateDiff pipeline request and AnimateDiff+ControlNet are related only by subsystem, not the same underlying change." + "reason": "AnimateDiff library request vs AnimateDiff + ControlNet feature request; related area but different change requests." }, { "left": "issue:4648", "right": "issue:8906", "accept": false, - "reason": "Missing config.json for ControlNet checkpoint loading is unrelated to a missing qk_norm attribute in SD3Transformer2DModel." + "reason": "Missing config.json during loading vs absence of qk_norm in SD3Transformer2DModel; unrelated." }, { "left": "issue:4668", "right": "issue:6524", "accept": false, - "reason": "Weird LoRA images and second-load LoRA malfunction describe different LoRA failure modes." + "reason": "Weird images from Kohya-style LoRA vs LoRA breaking on second load; different LoRA problems." }, { "left": "issue:4672", "right": "issue:8540", "accept": false, - "reason": "VRAM reduction for SDXL LoRA training is not the same as caching embeddings/VAE representations for SD3 dreambooth." + "reason": "SDXL LoRA VRAM reduction request vs SD3 dreambooth embedding/VAE caching; different training optimizations." }, { "left": "issue:4680", "right": "issue:5980", "accept": false, - "reason": "SDXL inpainting fine-tuning and ControlNet SDXL output file format are unrelated." + "reason": "SDXL inpainting fine-tuning vs ControlNet SDXL model export format question; unrelated." }, { "left": "issue:4991", "right": "issue:6958", "accept": false, - "reason": "Missing train_dataloader in accelerate.prepare() and use_linear_projection/mid_block mismatch in unet_motion_model are different training/model bugs." + "reason": "Missing train_dataloader in accelerate.prepare() vs use_linear_projection absent in mid_block; different code paths." }, { "left": "issue:5258", "right": "issue:8241", "accept": false, - "reason": "SDXL ControlNet Inpaint with refiner and LEDITS++ abnormal model size are unrelated issues." + "reason": "SDXL ControlNet inpaint with refiner vs abnormal LEDITS++ model size; unrelated model issues." }, { "left": "issue:5285", "right": "issue:6928", "accept": false, - "reason": "Docs translation request and Marigold depthmap colorization option are unrelated feature requests." + "reason": "Docs translation request vs Marigold depthmap colorization option; unrelated." }, { "left": "issue:5598", "right": "issue:5981", "accept": false, - "reason": "_modify_text_encoder bug and DPMSolverMultistepScheduler bug touch different components and behaviors." + "reason": "_modify_text_encoder bug vs DPMSolverMultistepScheduler bug; different components." }, { "left": "issue:5783", "right": "issue:8588", "accept": false, - "reason": "EMA not working in an SDXL training script and a missing SD3 conversion script are unrelated." + "reason": "EMA not working in SDXL training script vs missing SD3 conversion script; unrelated." }, { "left": "issue:6141", "right": "issue:8390", "accept": false, - "reason": "ControlNet control_mode/resize_mode support and Stable Cascade ControlNet training are different feature requests." + "reason": "ControlNet parameter support request vs Stable Cascade ControlNet training issue; different change scopes." }, { "left": "issue:6242", "right": "issue:7879", "accept": false, - "reason": "Request for fill50k logs and a pokemon captions dataset are not duplicate bugs or changes." + "reason": "Dataset log request vs a specific captions dataset request; not the same issue." }, { "left": "issue:6354", "right": "issue:9958", "accept": false, - "reason": "Add LoRA support for ControlNet and AutoencoderKLTemporalDecoder tiling are unrelated feature requests." + "reason": "LoRA support for ControlNet vs AutoencoderKLTemporalDecoder tiling support; unrelated." }, { "left": "issue:6359", "right": "issue:7341", "accept": false, - "reason": "EulerDiscreteScheduler velocity/add_noise bug is unrelated to a TorchVersion vs Version comparison error." + "reason": "EulerDiscreteScheduler velocity/add_noise bug vs a TorchVersion/Version comparison TypeError; different bugs." }, { "left": "issue:6439", "right": "issue:9396", "accept": false, - "reason": "A broken fp16 doc link and Korean docs translation are unrelated documentation items." - }, - { - "left": "issue:6481", - "right": "issue:9350", - "accept": false, - "reason": "Dreambooth batch-size bug and txt_ids deprecation warning concern different training code paths." - }, - { - "left": "issue:6725", - "right": "issue:9488", - "accept": false, - "reason": "LDMPipeline image generation request and Lumina pipeline failure are unrelated pipeline issues." + "reason": "Broken fp16 doc link vs Korean docs translation; unrelated documentation tasks." } ] }, @@ -75558,7 +75582,6 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10191", - "issue:11208", "issue:12012", "issue:13073", "issue:1323", @@ -75573,22 +75596,21 @@ "issue:3338", "issue:3908", "issue:4043", - "issue:4297", "issue:4356", "issue:4615", "issue:4646", "issue:5281", - "issue:5433", "issue:5619", "issue:5770", "issue:6351", + "issue:6481", + "issue:6725", "issue:6742", "issue:6949", "issue:7054", "issue:746", "issue:7592", "issue:7636", - "issue:7660", "issue:8320", "issue:8405", "issue:8429", @@ -75597,8 +75619,10 @@ "issue:9007", "issue:9112", "issue:9178", + "issue:9350", "issue:9361", "issue:9471", + "issue:9488", "issue:9495", "issue:9602", "issue:9846", @@ -75607,6 +75631,8 @@ "issue:9962" ], "soft_pairs": [ + "issue:6481|issue:9350", + "issue:6725|issue:9488", "issue:6949|issue:9007", "issue:7636|issue:8405", "issue:8320|issue:9846", @@ -75629,188 +75655,186 @@ "issue:5770|issue:6351", "issue:1323|issue:4615", "issue:3338|issue:4356", - "issue:4646|issue:746", - "issue:11208|issue:7660", - "issue:4297|issue:5433" + "issue:4646|issue:746" ], - "prepared_review_unit_hash": "19c76a77ffe060e958ac1c481654d14a1c82ac4b0cc79a944ed51124758d8667", + "prepared_review_unit_hash": "eab199b693b10864f8f15951ab77c8ccf5a9f117c66c91f08176f7ab1b4688d2", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34528, - "estimated_input_tokens": 8632, - "estimated_eval_tokens": 17520 + "serialized_chars": 34509, + "estimated_input_tokens": 8628, + "estimated_eval_tokens": 17512 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the soft-similarity pairs mostly share only broad subsystem keywords (ControlNet, LoRA, scheduler, training, import/load) but not the same concrete bug or change. I would not merge any of the candidate pairs, and there is no clear canonical issue/PR for the whole set.", - "confidence": 0.96, + "summary": "No soft-edge pair looks like the same underlying bug/change; the set is mostly unrelated issue reports. There are no PRs in the batch. If a representative issue must be chosen, issue 13073 is the clearest narrowly-scoped bug report.", + "confidence": 0.39, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 13073 is the most concrete and narrowly scoped bug report here, with a specific pipeline and behavior, so it is the best anchor if one issue must represent the set.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6481", + "right": "issue:9350", + "accept": false, + "reason": "Different problems: batch-size-related training failure vs a deprecation warning about 3D txt_ids tensors." + }, + { + "left": "issue:6725", + "right": "issue:9488", + "accept": false, + "reason": "Both are generation issues, but they involve different pipelines and unrelated failure modes." + }, { "left": "issue:6949", "right": "issue:9007", "accept": false, - "reason": "Both are DPMSolverSinglestepScheduler artifacts, but one is blank images at even steps and the other is black output with 3rd-order settings; different failure modes and likely different fixes." + "reason": "Same scheduler family, but different concrete symptoms and conditions: blank images at even steps vs black output with order 3." }, { "left": "issue:7636", "right": "issue:8405", "accept": false, - "reason": "Both mention multiple ControlNets, but 7636 is a usage question for SDXL tile ControlNet while 8405 is a tensor-size error with multiple controlnets; not the same bug." + "reason": "Both mention ControlNet, but one is about using an SDXL tile model and the other about multi-ControlNet tensor shape errors." }, { "left": "issue:8320", "right": "issue:9846", "accept": false, - "reason": "8320 is a from_single_file attribute error on AutoencoderKL, while 9846 is a feature request for FluxControlNetModel support in from_single_file; not duplicate." + "reason": "Both concern from_single_file, but they are different models and different missing/invalid behavior." }, { "left": "issue:8960", "right": "issue:9850", "accept": false, - "reason": "Different training concerns: gradient accumulation is nonfunctional versus enabling gradient checkpointing with a frozen model." + "reason": "Training-example gradient accumulation bug is unrelated to frozen-model gradient checkpointing support." }, { "left": "issue:9178", "right": "issue:9361", "accept": false, - "reason": "Both involve Flux LoRA, but one is a PEFT target-modules training error and the other is LoRA having no effect at inference; different code paths." + "reason": "Training-time PEFT target-module errors and inference-time missing LoRA effect are different bugs." }, { "left": "issue:9902", "right": "issue:9962", "accept": false, - "reason": "Unrelated: one is a Conda version pinning/stuck issue, the other is an unexpected use_cuda_graph keyword argument." + "reason": "Completely unrelated: conda version pinning vs an unexpected keyword argument error." }, { "left": "issue:7636", "right": "issue:9495", "accept": false, - "reason": "Both are advanced conditioning-related questions, but 7636 is about using SDXL ControlNet tile models and 9495 is SDXL PAG with IPAdapter; different problems." + "reason": "Different SDXL features and failures: ControlNet tile usage vs PAG with IPAdapter." }, { "left": "issue:12012", "right": "issue:9471", "accept": false, - "reason": "Completely different failures: Wan 2.2 context-state ValueError versus CoreML file-extension/resource-type inference error." + "reason": "Unrelated: Wan 2.2 cache-context error vs a CoreML file-extension inference error." }, { "left": "issue:8429", "right": "issue:9112", "accept": false, - "reason": "Different models and errors: I2VGen-XL inference script issues versus an Auraflow tensor shape mismatch." + "reason": "Different models and failures: I2VGen-XL script malfunction vs Auraflow tensor-size mismatch." }, { "left": "issue:3116", "right": "issue:3908", "accept": false, - "reason": "Scheduler noise in image2image is unrelated to caching a model in Streamlit." + "reason": "Scheduler latent-noise bug and Streamlit caching problem are unrelated." }, { "left": "issue:7054", "right": "issue:7592", "accept": false, - "reason": "Both concern adapter loading, but one empties the passed state_dict and the other misses DoRA keys; distinct underlying bugs." + "reason": "Both are adapter-loading issues, but one mutates an input state_dict and the other misses DoRA keys; not the same bug." }, { "left": "issue:10191", "right": "issue:9602", "accept": false, - "reason": "GPU-device selection during training is unrelated to adapter-name validation in a loaded model." + "reason": "GPU device selection during training is unrelated to an adapter-name validation error." }, { "left": "issue:1608", "right": "issue:1946", "accept": false, - "reason": "Training stopping at 500 steps on T4 is unrelated to a CUDA xformers invalid-argument error." + "reason": "Training halting at 500 steps is unrelated to xformers CUDA invalid-argument failures." }, { "left": "issue:4043", "right": "issue:5619", "accept": false, - "reason": "SDXL prompt-embed/pooled-embed requirements are unrelated to AutoencodeTiny failing for LCM img2img encode." + "reason": "Different pipelines and different missing-input problems; no same code path." }, { "left": "issue:13073", "right": "issue:13300", "accept": false, - "reason": "Different issues: ZImageControlNet guidance_scale support versus CPU inference failures for SD3.5/FLUX." + "reason": "Both are generation bugs, but they affect different pipelines/models and different failure conditions." }, { "left": "issue:6742", "right": "issue:9850", "accept": false, - "reason": "Both mention checkpointing, but 6742 is behavior divergence in DownBlockMotion and 9850 is enabling checkpointing with frozen models; not same bug." + "reason": "Checkpointing-related, but one is a motion block behavior difference and the other is enabling checkpointing with a frozen model." }, { "left": "issue:1495", "right": "issue:2578", "accept": false, - "reason": "Importing init_git_repo and editable-install problems are separate packaging/import failures." + "reason": "Import failure for init_git_repo is unrelated to editable-install behavior." }, { "left": "issue:1672", "right": "issue:3222", "accept": false, - "reason": "Wrong image size after ckpt-to-diffusers conversion is unrelated to missing custom diffusion weight output." + "reason": "Wrong image size after conversion is unrelated to missing custom diffusion weight files after training." }, { "left": "issue:5281", "right": "issue:8523", "accept": false, - "reason": "Different device/runtime issues: CPU offload across XL pipelines versus a generic playground CUDA error." + "reason": "Both are CUDA/device errors, but they involve different workflows and no shared concrete bug." }, { "left": "issue:5770", "right": "issue:6351", "accept": false, - "reason": "Dataset access preventing LCM distill run is unrelated to a DistributedDataParallel disable_adapters attribute error." + "reason": "Dataset-access problem and DDP missing-disable_adapters error are unrelated." }, { "left": "issue:1323", "right": "issue:4615", "accept": false, - "reason": "Flax dtype-loading behavior and a missing StableDiffusionControlnetPipeline import are unrelated." + "reason": "Flax dtype-loading failure is unrelated to a missing StableDiffusionControlnetPipeline import." }, { "left": "issue:3338", "right": "issue:4356", "accept": false, - "reason": "Import slowness and a missing dummy module are different packaging/runtime problems." + "reason": "Import-performance complaint and missing dummy module error are different issues." }, { "left": "issue:4646", "right": "issue:746", "accept": false, - "reason": "Both involve ONNX/export/loading, but one is ControlNet conversion failure and the other is an img2img ONNX revision loading issue." - }, - { - "left": "issue:11208", - "right": "issue:7660", - "accept": false, - "reason": "MultiControlNet support for SD3ControlNetInpaintingPipeline is unrelated to a trace example not working." - }, - { - "left": "issue:4297", - "right": "issue:5433", - "accept": false, - "reason": "Both report visual artifacts, but one is batched SDXL img2img fishnet artifacts and the other is DPM++/SDXL artifacts without refiner; different reproduction conditions and likely different causes." + "reason": "Both touch ONNX, but one is a conversion failure and the other is an ONNX-revision loading problem." } ] }, @@ -75841,15 +75865,13 @@ "issue:1076", "issue:1083", "issue:10869", - "issue:10940", + "issue:11208", "issue:11241", "issue:11519", "issue:11574", "issue:12025", - "issue:12060", "issue:12299", "issue:12319", - "issue:12373", "issue:12489", "issue:12505", "issue:12535", @@ -75868,22 +75890,26 @@ "issue:3401", "issue:4059", "issue:4104", + "issue:4297", "issue:4551", "issue:4631", "issue:4646", + "issue:5433", "issue:5953", "issue:6533", "issue:7223", "issue:7370", "issue:7629", + "issue:7660", "issue:7895", - "issue:7935", "issue:8552", "issue:9007", "issue:9516", "issue:968" ], "soft_pairs": [ + "issue:11208|issue:7660", + "issue:4297|issue:5433", "issue:1534|issue:193", "issue:1083|issue:4646", "issue:12319|issue:13488", @@ -75905,182 +75931,180 @@ "issue:12505|issue:12754", "issue:12535|issue:13279", "issue:10596|issue:12025", - "issue:1076|issue:4059", - "issue:10940|issue:7935", - "issue:12060|issue:12373" + "issue:1076|issue:4059" ], - "prepared_review_unit_hash": "1bdeb3c57fb1db68b50df539db1093f5f080438b3c053626a44bdf2b6a7ced3a", + "prepared_review_unit_hash": "95603e57c7c592b4ccb56835e4bf76e64a9b012a3581c86a05d74a0bc9e3a423", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33824, - "estimated_input_tokens": 8456, - "estimated_eval_tokens": 17168 + "serialized_chars": 33844, + "estimated_input_tokens": 8461, + "estimated_eval_tokens": 17178 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: the items span unrelated feature requests, model-loading errors, scheduler bugs, and offloading regressions. The soft edges are mostly title-level similarity and should be rejected.", - "confidence": 0.97, - "canonical_issue_reason": null, - "canonical_pr_reason": null, - "best_issue_reason": null, - "best_pr_reason": null, + "summary": "This cluster is highly heterogeneous: it mixes unrelated bug reports, feature requests, and model/pipeline support questions. None of the soft-similarity pairs look like true duplicates of the same underlying issue/change, so I would not merge any of them.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical duplicate stands out because the items span unrelated problems across different pipelines, schedulers, loading paths, and feature requests. If a representative issue must be chosen, #11208 is the strongest standalone bug report with broad impact and the most discussion.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "#11208 is the best standalone issue to anchor on: it is a concrete, user-facing bug with clear scope and higher engagement than most others here.", + "best_pr_reason": "No pull requests are present in this cluster.", "soft_edge_verdicts": [ + { + "left": "issue:11208", + "right": "issue:7660", + "accept": false, + "reason": "Different problems: unsupported MultiControlNetModel in one specific pipeline vs a trace example failing to work." + }, + { + "left": "issue:4297", + "right": "issue:5433", + "accept": false, + "reason": "Both mention visual artifacts, but the failure modes and triggering conditions differ (batched img2img watermarks vs DPM++/no-refiner artifacts)." + }, { "left": "issue:1534", "right": "issue:193", "accept": false, - "reason": "Both are generic runtime problems, but one is a Dreambooth notebook error and the other is an unspecific pipeline error; no shared concrete bug." + "reason": "Dreambooth notebook training error is unrelated to a generic pipeline error report." }, { "left": "issue:1083", "right": "issue:4646", "accept": false, - "reason": "ONNX wrong-output bug vs ONNX conversion failure; related area, but different failure modes and likely different fixes." + "reason": "Incorrect FP16 ONNX output is a different bug from ControlNet ONNX conversion failure." }, { "left": "issue:12319", "right": "issue:13488", "accept": false, - "reason": "Both mention offloading, but one is a block-level offloading regression and the other is a version-wide breakage claim; too broad to be the same bug." + "reason": "Both concern offloading, but one is a block_level regression and the other is a version-wide breakage; not the same concrete code-path." }, { "left": "issue:2437", "right": "issue:6533", "accept": false, - "reason": "Two separate feature requests for different pipelines; no duplicate problem or mergeable change." + "reason": "A T2I-Adapter pipeline request is unrelated to the Marigold pipeline request." }, { "left": "issue:4631", "right": "issue:9007", "accept": false, - "reason": "Same scheduler family, but one is a scalar-conversion exception and the other is black output with order 3; different concrete bugs." + "reason": "Same scheduler family, but one is a Tensor-to-scalar runtime error and the other is black output; too different to treat as one bug." }, { "left": "issue:5953", "right": "issue:7629", "accept": false, - "reason": "ControlNet image/control size crash vs PyTorch 1.13.1 OOM in SVD; unrelated issues." + "reason": "ControlNet image/control size crash and SVD OOM on PyTorch 1.13 are unrelated." }, { "left": "issue:10869", "right": "issue:12299", "accept": false, - "reason": "Both are tensor-shape/metadata problems, but one is meta-tensor item() usage and the other is SD3 prompt-embedding mismatch; different code paths." + "reason": "Meta-tensor item() failure and SD3 tensor mismatch with negative prompts are different shape/embedding bugs." }, { "left": "issue:11241", "right": "issue:8552", "accept": false, - "reason": "Floating-point exception in Flux ControlNet training vs segmentation fault in TextToVideoZero inference; not the same bug." + "reason": "Flux ControlNet training floating-point exception is unrelated to TextToVideoZeroPipeline segmentation fault." }, { "left": "issue:2153", "right": "issue:4104", "accept": false, - "reason": "Float16 GPU incompatibility vs black image output/cast warning; different symptoms and likely unrelated fixes." + "reason": "GTX 1660 float16 runtime issues are not the same as black-image invalid cast output." }, { "left": "issue:1934", "right": "issue:2763", "accept": false, - "reason": "VRAM/offload memory complaint vs ControlNet cuDNN benchmark memory increase; not the same underlying issue." + "reason": "Both mention VRAM, but cpu_offload memory bloat and cudnn.benchmark ControlNet VRAM growth are distinct issues." }, { "left": "issue:3401", "right": "issue:4551", "accept": false, - "reason": "Class-label validation error during training vs SDXL Dreambooth crash; separate model/training problems." + "reason": "class_labels / num_class_embeds validation is unrelated to a SDXL Dreambooth crash." }, { "left": "issue:7223", "right": "issue:7370", "accept": false, - "reason": "Different video-model feature requests (Latte vs STDiT/OpenSora support); not duplicates." + "reason": "These are different feature requests for different video models (Latte vs OpenSora STDiT)." }, { "left": "issue:10653", "right": "issue:12489", "accept": false, - "reason": "8-bit `.to()`/device-move limitation vs CUDA sm_120 kernel compatibility error; unrelated failures." + "reason": "Bitsandbytes .to()/device movement problem is unrelated to CUDA sm_120 kernel compatibility." }, { "left": "issue:10281", "right": "issue:11519", "accept": false, - "reason": "Two unrelated feature requests for different models/schedulers (FreeScale vs MAGI-1)." + "reason": "A new scheduler request and a MAGI-1 support request are different feature asks." }, { "left": "issue:1440", "right": "issue:968", "accept": false, - "reason": "Model cache deletion request vs slow model loading; different concerns." + "reason": "Deleting cached models is unrelated to model-loading slowness." }, { "left": "issue:1633", "right": "issue:2858", "accept": false, - "reason": "Karras sigmas support vs seeding image for unclip img2img; separate feature requests." + "reason": "Karras sigmas support and seeding image support are separate pipeline enhancements." }, { "left": "issue:7895", "right": "issue:9516", "accept": false, - "reason": "Flax memory-efficient attention bug vs FLUX joint_attention_kwargs not forwarded; different implementations and fixes." + "reason": "Flax memory-efficient attention bug and FLUX joint_attention_kwargs wiring are different code paths." }, { "left": "issue:10282", "right": "issue:11574", "accept": false, - "reason": "Shape mismatch while loading x_embedder weights vs missing VAE key; both load-time errors, but not the same missing tensor." + "reason": "Shape mismatch on x_embedder.weight and missing VAE key are different loading errors." }, { "left": "issue:12505", "right": "issue:12754", "accept": false, - "reason": "tgate img2img error vs `from_pipe` dtype conversion behavior; unrelated pipeline issues." + "reason": "tgate img2img error and from_pipe dtype conversion are unrelated issues." }, { "left": "issue:12535", "right": "issue:13279", "accept": false, - "reason": "Wan2.2 LoRA load failure vs AutoencoderRAE loading with older transformers; different models and incompatibilities." + "reason": "Wan2.2 LoRA loading failure and AutoencoderRAE compatibility with older transformers are distinct model-loading problems." }, { "left": "issue:10596", "right": "issue:12025", "accept": false, - "reason": "FluxControlPipeline shape error vs Cosmos VAE encoder API error; not the same code path." + "reason": "FluxControlPipeline shape error and Cosmos VAE Encoder API error are unrelated." }, { "left": "issue:1076", "right": "issue:4059", "accept": false, - "reason": "General Stable Diffusion v1.5 question vs ControlNet reference pipeline bug/question; unrelated." - }, - { - "left": "issue:10940", - "right": "issue:7935", - "accept": false, - "reason": "Missing optional dependency (`triton.ops`) vs missing model config file; different loading failures." - }, - { - "left": "issue:12060", - "right": "issue:12373", - "accept": false, - "reason": "Question about whether a DiT block exists vs proposal for an Unlearn API; not the same issue or change." + "reason": "A general stable-diffusion-v1-5 question is unrelated to a ControlNet reference pipeline bug/question." } ] }, @@ -76112,12 +76136,14 @@ "issue:10659", "issue:1070", "issue:10920", + "issue:10940", "issue:10950", "issue:1195", "issue:12022", - "issue:12216", + "issue:12060", "issue:12221", "issue:12319", + "issue:12373", "issue:13430", "issue:1811", "issue:2097", @@ -76129,11 +76155,9 @@ "issue:3698", "issue:3876", "issue:4720", - "issue:4767", "issue:5546", "issue:5607", "issue:5777", - "issue:5907", "issue:5927", "issue:599", "issue:6073", @@ -76143,6 +76167,7 @@ "issue:7264", "issue:7538", "issue:7732", + "issue:7935", "issue:7950", "issue:8277", "issue:8824", @@ -76150,10 +76175,11 @@ "issue:9109", "issue:9161", "issue:9171", - "issue:9464", - "issue:9558" + "issue:9464" ], "soft_pairs": [ + "issue:10940|issue:7935", + "issue:12060|issue:12373", "issue:12221|issue:13430", "issue:5546|issue:7732", "issue:599|issue:725", @@ -76175,192 +76201,187 @@ "issue:3698|issue:7264", "issue:2097|issue:2967", "issue:6073|issue:7056", - "issue:10411|issue:8277", - "issue:4767|issue:5907", - "issue:12216|issue:9558" + "issue:10411|issue:8277" ], - "prepared_review_unit_hash": "2e04e9500349a2c1eafa44cc7df930c0a8de3ed5c65ddf2c8c8e1dea128d844e", + "prepared_review_unit_hash": "3403a71205823bee90be2bfeba5605c50ba8bb0029884278968e1f99b8531b20", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34146, - "estimated_input_tokens": 8537, - "estimated_eval_tokens": 17330 + "serialized_chars": 34152, + "estimated_input_tokens": 8538, + "estimated_eval_tokens": 17332 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are unrelated Diffusers issues/feature requests across different models, schedulers, training scripts, and docs. Only the two UNet1D reports look like the same underlying runtime problem; the rest should stay separate.", - "confidence": 0.79, - "canonical_issue_reason": "issue:2967 is the clearest representative of the only plausible duplicate pair, with the most explicit UNet1D runtime-failure framing.", + "summary": "This set does not contain any clear duplicate issue groups; the suggested soft pairs are mostly superficial title similarity across different bugs, features, and docs/training topics. I\u2019d anchor the cluster on a concrete, well-scoped bug report rather than try to merge unrelated items.", + "confidence": 0.89, + "canonical_issue_reason": "Issue #12319 is a concrete, clearly scoped bug report with strong discussion/inbound references, making it the best anchor among mostly unrelated items.", "canonical_pr_reason": null, - "best_issue_reason": "issue:2967 is the strongest anchor for triage because it describes the UNet1D failure more concretely than the other UNet1D report and is the better duplicate representative.", + "best_issue_reason": "Issue #12319 is the most suitable representative: specific failure mode, actionable title, and higher engagement than the other issues.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10940", + "right": "issue:7935", + "accept": false, + "reason": "Both mention loading/import problems, but one is a Triton import error and the other is a missing config.json model-repo issue; different root causes and fixes." + }, + { + "left": "issue:12060", + "right": "issue:12373", + "accept": false, + "reason": "A general question about DiT blocks versus an API proposal for unlearn; unrelated intent and code change surface." + }, { "left": "issue:12221", "right": "issue:13430", "accept": false, - "reason": "Both are feature/support requests, but they target different models and changes: DiffSynth ControlNet support vs JoyAI-Image-Edit support." + "reason": "Different feature requests for different model support; same broad theme of contribution requests, but not the same underlying bug/change." }, { "left": "issue:5546", "right": "issue:7732", "accept": false, - "reason": "Both involve SDXL/refiner/controlnet areas, but one is a scheduler step-count bug and the other is a runtime error with ControlNet+Refiner." + "reason": "Both involve schedulers/SDXL-ish behavior, but one is a step-count bug in base+refiner ensembles and the other is a ControlNet/refiner runtime error; not the same issue." }, { "left": "issue:599", "right": "issue:725", "accept": false, - "reason": "Related to model loading/memory, but one asks for Deepspeed integration and the other for device-map/accelerate loading support." + "reason": "Both are memory/deployment related, but one requests DeepSpeed support and the other device-map/accelerate loading; separate changes." }, { "left": "issue:3876", "right": "issue:6964", "accept": false, - "reason": "Both touch inpainting/controlnet, but they are different failures: a demo error vs incompatibility with 4-channel UNets." + "reason": "Both are inpainting-related, but they describe different pipeline failures and model-channel compatibility problems; not the same code-path bug." }, { "left": "issue:1070", "right": "issue:4720", "accept": false, - "reason": "Both are inpainting-related, but the concrete bugs differ: an inpainting version regression vs AutoencoderTiny compatibility." + "reason": "Older inpainting bug versus AutoencoderTiny incompatibility; same general area but different component and failure mechanism." }, { "left": "issue:7538", "right": "issue:8824", "accept": false, - "reason": "Both are dataset/training questions, but they concern different workflows: Fill50k-style training vs instruct-pix2pix dataset creation." + "reason": "Different training/data-prep questions for different pipelines; no shared concrete bug or patch target." }, { "left": "issue:2437", "right": "issue:5927", "accept": false, - "reason": "These are separate feature requests for different adapter types and pipeline combinations; same broad area, not the same change." + "reason": "A feature request for a new pipeline versus a usage/problem report combining existing components; not duplicates." }, { "left": "issue:7950", "right": "issue:8832", "accept": false, - "reason": "Different problems entirely: a PyTorch/Diffusers compatibility issue vs a generic pipeline crash when indexing images." + "reason": "Framework/library incompatibility versus a generic runtime crash when calling a pipeline; too vague and different root causes." }, { "left": "issue:10514", "right": "issue:10659", "accept": false, - "reason": "Different model-specific issues: Sana resolution binning limits vs ConsisID poor generation quality." + "reason": "Sana resolution-binning limitation versus ConsisID output quality complaint; different models and problems." }, { "left": "issue:10920", "right": "issue:9171", "accept": false, - "reason": "Different scheduler/loader bugs with different triggers and code paths; not the same underlying defect." + "reason": "Both are scheduler/pipeline regression reports, but one is DDIMInverseScheduler infs and the other is from_single_file prediction_type handling; different code paths." }, { "left": "issue:12022", "right": "issue:9109", "accept": false, - "reason": "Unrelated dependency/interface failures: flash-attention 3 API incompatibility vs missing torchvision::nms operator." + "reason": "Flash-attention interface mismatch is unrelated to a torchvision nms operator missing error; different dependencies and fixes." }, { "left": "issue:9161", "right": "issue:9464", "accept": false, - "reason": "Both are about training Flux LoRA, but the concrete failures differ: distributed bugs vs an IndexError with prodigy and train_text_encoder." + "reason": "Both affect Flux DreamBooth LoRA training, but one reports distributed bugs and the other a prodigy/train_text_encoder IndexError; not the same concrete bug." }, { "left": "issue:2392", "right": "issue:3347", "accept": false, - "reason": "One is an input-validation error around prompt_embeds, the other is NaN image output; different symptoms and likely different causes." + "reason": "Prompt_embeds validation error in img2img versus NaN image output from a diffusion pipeline; different symptoms and likely different causes." }, { "left": "issue:10282", "right": "issue:5607", "accept": false, - "reason": "Both are tensor-shape failures, but the titles indicate different code paths and different mismatched dimensions." + "reason": "Both are tensor shape/runtime failures, but the affected shapes and code paths differ; insufficient evidence of the same underlying bug." }, { "left": "issue:10450", "right": "issue:10591", "accept": false, - "reason": "Architecture comparison for Flux blocks vs an sd3 LoRA training-script bug; not the same bug or change." + "reason": "SingleStreamBlock implementation mismatch versus sd3 LoRA training script issues; same model family but not the same change or bug." }, { "left": "issue:10950", "right": "issue:12319", "accept": false, - "reason": "Documentation breakage and group offloading behavior are unrelated issues." + "reason": "Docs broken for accelerate inference versus a group offloading runtime bug; different artifact types and fixes." }, { "left": "issue:1195", "right": "issue:1811", "accept": false, - "reason": "Different components: memory-efficient attention with fp16 weights vs from_pretrained downloading both weight formats." + "reason": "Memory-efficient attention fp16 support versus from_pretrained choosing between bin and safetensors; unrelated loading/attention problems." }, { "left": "issue:3622", "right": "issue:5777", "accept": false, - "reason": "A tutorial import error and a PixArt DreamBooth training error are unrelated." + "reason": "Tutorial import error versus PixArt DreamBooth training failure; different surfaces and root causes." }, { "left": "issue:3698", "right": "issue:7264", "accept": false, - "reason": "Different feature areas: model conversion to Diffusers vs serialization/loading of custom transformers models." + "reason": "Inpainting model conversion request versus custom transformers model serialization/loading problem; not the same issue." }, { "left": "issue:2097", "right": "issue:2967", - "accept": true, - "reason": "Both titles describe the same UNet1D failure to run, with no meaningful distinction in component or symptom from the titles." + "accept": false, + "reason": "Both are UNet1D-related, but one is a general problem using UNet1D and the other cannot get a simple UNet1D to run; not enough evidence they are the same concrete bug." }, { "left": "issue:6073", "right": "issue:7056", "accept": false, - "reason": "A beginner DreamBooth SDXL setup error is not the same as an LoRA fine-tuning error involving train_text_encoder." + "reason": "Beginner DreamBooth SDXL setup error versus a LoRA fine-tuning error with train_text_encoder; different scripts and failure modes." }, { "left": "issue:10411", "right": "issue:8277", "accept": false, - "reason": "Training-script usage question vs convergence problem; different issue types and likely different causes." - }, - { - "left": "issue:4767", - "right": "issue:5907", - "accept": false, - "reason": "Scheduler prediction_type support question vs DDPMScheduler docs/API mismatch; not the same bug." - }, - { - "left": "issue:12216", - "right": "issue:9558", - "accept": false, - "reason": "One is about Qwen-Image-Edit quality compared to ComfyUI, the other about black validation images in tensorboard." + "reason": "How-to training question versus a convergence problem in consistency training; different nature and no shared underlying bug." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "Grounded and conservative overall. The UNet1D pair is a plausible duplicate anchor, and the other soft-edge rejections stay appropriately cautious given the packet evidence." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -76381,12 +76402,12 @@ "issue:10653", "issue:11448", "issue:11872", + "issue:12216", "issue:12359", "issue:13446", "issue:1556", "issue:1931", "issue:2068", - "issue:2201", "issue:2444", "issue:2534", "issue:255", @@ -76403,30 +76424,32 @@ "issue:4605", "issue:4665", "issue:4724", + "issue:4767", "issue:487", "issue:5216", "issue:5334", "issue:5553", "issue:5578", + "issue:5907", "issue:6176", "issue:6457", - "issue:6574", "issue:6811", "issue:6880", "issue:6958", - "issue:7056", "issue:7421", "issue:7519", "issue:7589", "issue:8393", - "issue:8477", "issue:8588", "issue:8943", "issue:9188", + "issue:9558", "issue:9599", "issue:984" ], "soft_pairs": [ + "issue:4767|issue:5907", + "issue:12216|issue:9558", "issue:3251|issue:6958", "issue:3574|issue:4665", "issue:4724|issue:8393", @@ -76448,182 +76471,180 @@ "issue:9188|issue:9599", "issue:10653|issue:11872", "issue:255|issue:487", - "issue:3799|issue:3861", - "issue:7056|issue:8477", - "issue:2201|issue:6574" + "issue:3799|issue:3861" ], - "prepared_review_unit_hash": "ea54051711cf193b157a7d892c19be2603c5bfe6d8b91b329fe78a05cd130b97", + "prepared_review_unit_hash": "e175dfc83d20daa516310a9746639bc3ad9784a09678529d67708cd8028fd34d", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34101, - "estimated_input_tokens": 8526, - "estimated_eval_tokens": 17308 + "serialized_chars": 34053, + "estimated_input_tokens": 8514, + "estimated_eval_tokens": 17284 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issues spanning model loading, scheduler behavior, training errors, docs, and feature requests. The soft pairs are only superficially similar; none look like true duplicates or one-change matches.", - "confidence": 0.89, - "canonical_issue_reason": "issue:5553 is the broadest loading-related report in the set and the best anchor for a loose cluster of model/pipeline loading problems, though most other items are still distinct.", + "summary": "These are mostly unrelated issue topics; the soft-similarity links are too broad to treat as duplicates, so all candidate merges should be rejected.", + "confidence": 0.93, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:5553 is the most generally applicable issue here; it covers a central DiffusionPipeline loading failure rather than a narrow symptom or documentation-only report.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4767", + "right": "issue:5907", + "accept": false, + "reason": "Both mention prediction_type, but one is about EulerAncestralDiscreteScheduler support and the other is a DDPMScheduler docs/argument mismatch; different bugs and different code paths." + }, + { + "left": "issue:12216", + "right": "issue:9558", + "accept": false, + "reason": "Qwen-Image-Edit quality regression vs black validation images in TensorBoard; one is model output quality, the other is logging/display." + }, { "left": "issue:3251", "right": "issue:6958", "accept": false, - "reason": "Different code paths and symptoms: ControlNet shuffle CFG-side behavior vs a UNet motion model configuration gap." + "reason": "Different models and failures: ControlNet Shuffle CFG handling vs a unet_motion_model mid_block projection mismatch." }, { "left": "issue:3574", "right": "issue:4665", "accept": false, - "reason": "One is a Windows from_ckpt problem; the other is safetensors selection on Windows with alternative bin files. Related platform, but not the same bug." + "reason": "Both involve Windows loading problems, but one is from_ckpt behavior and the other is use_safetensors resolution when alternate bin files exist; not the same underlying bug." }, { "left": "issue:4724", "right": "issue:8393", "accept": false, - "reason": "Img2img checkpoint compatibility is unrelated to slow checkpoint shard loading." + "reason": "Img2Img output mismatch with an SDXL checkpoint vs slow checkpoint shard loading; unrelated symptoms and fixes." }, { "left": "issue:1556", "right": "issue:3948", "accept": false, - "reason": "Black images with NSFW warning and pale images are different rendering symptoms with no clear shared root cause." + "reason": "Black images with an NSFW warning is a safety-checker/output issue, while pale images is a color/quality complaint; not the same defect." }, { "left": "issue:2444", "right": "issue:6880", "accept": false, - "reason": "Different scripts and failures: dreambooth_inpaint exits with an error, while custom diffusion retrieve.py is a separate utility problem." + "reason": "Different scripts and errors: dreambooth_inpaint training failure vs custom diffusion retrieve.py behavior." }, { "left": "issue:1931", "right": "issue:984", "accept": false, - "reason": "Float16 dtype mismatch in Stable Diffusion 2 is not the same as F.interpolate failing for large batch sizes." + "reason": "Float16 datatype mismatch during Stable Diffusion 2 loading is unrelated to F.interpolate breaking for large batch sizes." }, { "left": "issue:5216", "right": "issue:7589", "accept": false, - "reason": "Inpainting tensor iteration with a scheduler issue is distinct from GLIGEN layout-to-image not working in general." + "reason": "Both are GLIGEN-related, but one is a concrete in-painting 0-d tensor TypeError with a specific scheduler and the other is a general layout-to-image failure." }, { "left": "issue:5553", "right": "issue:6176", "accept": false, - "reason": "General custom model loading failure vs loading LoRA weights after dreambooth_lora training; different stages and causes." + "reason": "Custom model loading failure and LoRA weight loading after DreamBooth training are distinct pipeline stages and failure modes." }, { "left": "issue:12359", "right": "issue:13446", "accept": false, - "reason": "Documentation wording around guidance_scale is unrelated to the custom_pipeline trust/security check." + "reason": "A documentation bug for guidance_scale is not the same as a security/trust-check issue for custom_pipeline." }, { "left": "issue:1037", "right": "issue:280", "accept": false, - "reason": "Community pipeline addition request vs scheduler addition request; same area, different feature." + "reason": "Different feature requests: adding a pipeline vs adding a scheduler." }, { "left": "issue:11448", "right": "issue:8943", "accept": false, - "reason": "Missing fusing/stable-unclip pipeline file is not the same as a libtorch CUDA shared-library installation failure." + "reason": "Missing stable-unclip model file and missing libtorch CUDA shared library are unrelated loading/environment problems." }, { "left": "issue:2068", "right": "issue:3238", "accept": false, - "reason": "dtype conversion bug in from_pretrained is unrelated to converting LoRA safetensors to diffusers format." + "reason": "Datatype conversion in from_pretrained is unrelated to converting LoRA safetensors to diffusers format." }, { "left": "issue:2534", "right": "issue:4605", "accept": false, - "reason": "Scheduler switching error and long-prompt/inpaint pipeline error are different concrete failures." + "reason": "Scheduler switch errors and long-prompt-weighting inpaint errors are different pipeline failures." }, { "left": "issue:2761", "right": "issue:4391", "accept": false, - "reason": "Missing docs for UNet1DModel.extra_in_channels does not match a wrong default value/documentation issue in SDXL pipeline." + "reason": "Both touch documentation/defaults, but for different models and parameters; not the same issue." }, { "left": "issue:4264", "right": "issue:8588", "accept": false, - "reason": "Missing example images and missing conversion script are both docs/repo maintenance issues, but not the same artifact." + "reason": "Missing example images and a missing conversion script are both absent artifacts, but they are different files and requests." }, { "left": "issue:5334", "right": "issue:6457", "accept": false, - "reason": "Accelerate API support for multiple models and SDXL-LoRA gradient accumulation are training-related but not the same bug." + "reason": "Accelerate API handling of multiple models is not the same as multi-GPU gradient accumulation behavior for SDXL-LoRA." }, { "left": "issue:5578", "right": "issue:6811", "accept": false, - "reason": "Both mention model_index.json, but one is a generic mistaken path/file lookup and the other is a specific missing file in a model repo." + "reason": "Both mention model_index.json, but one is a local/random-module loading error and the other is a specific model repo URL missing the file." }, { "left": "issue:7421", "right": "issue:7519", "accept": false, - "reason": "Seed reproducibility in KDiffusion is unrelated to xFormers ignoring cross_attention_kwargs['scale']." + "reason": "Seed reproducibility in StableDiffusionXLKDiffusionPipeline is unrelated to xFormers ignoring cross_attention_kwargs['scale']." }, { "left": "issue:9188", "right": "issue:9599", "accept": false, - "reason": "Flux outpainting feature request and FLUX LoRA-only finetune example request are different changes." + "reason": "Both are FLUX-related, but outpainting and LoRA-only finetuning are different feature requests." }, { "left": "issue:10653", "right": "issue:11872", "accept": false, - "reason": "Unsupported .to() on an 8-bit loaded module is a correctness limitation, while repeated CPU/GPU moves are an efficiency problem." + "reason": "One is an unsupported .to() move for an 8-bit bitsandbytes module; the other is inefficient repeated CPU/GPU offloading loops." }, { "left": "issue:255", "right": "issue:487", "accept": false, - "reason": "UNet size divisibility error on StableDiffusionPipeline is distinct from M1 CPU incompatibility." + "reason": "An image size divisibility bug in StableDiffusionPipeline is not the same as M1 CPU incompatibility." }, { "left": "issue:3799", "right": "issue:3861", "accept": false, - "reason": "Incorrect inferred type for DDPMScheduler.from_pretrained is not the same as the Pipeline.from_pretrained naming/behavior question." - }, - { - "left": "issue:7056", - "right": "issue:8477", - "accept": false, - "reason": "SDXL LoRA fine-tuning with train_text_encoder and a distributed training script DDP error are separate training failures." - }, - { - "left": "issue:2201", - "right": "issue:6574", - "accept": false, - "reason": "Tokenizer padding differences and an SVD pipeline FutureWarning are unrelated pipeline behaviors." + "reason": "Inferred type issues for DDPMScheduler.from_pretrained and a Pipeline.from_pretrained naming question are unrelated." } ] }, @@ -76653,13 +76674,13 @@ "issue:10210", "issue:10475", "issue:10518", - "issue:10917", "issue:1127", "issue:11432", "issue:1207", "issue:12781", "issue:13249", "issue:1750", + "issue:2201", "issue:2402", "issue:2880", "issue:2891", @@ -76675,19 +76696,19 @@ "issue:4668", "issue:5215", "issue:5392", - "issue:5693", "issue:590", "issue:5989", - "issue:6250", "issue:6256", "issue:6354", "issue:6452", + "issue:6574", "issue:7054", + "issue:7056", "issue:7142", - "issue:7365", "issue:7368", "issue:7575", "issue:7889", + "issue:8477", "issue:8514", "issue:8700", "issue:8785", @@ -76697,6 +76718,8 @@ "issue:989" ], "soft_pairs": [ + "issue:7056|issue:8477", + "issue:2201|issue:6574", "issue:3954|issue:590", "issue:2891|issue:5215", "issue:4198|issue:989", @@ -76718,182 +76741,180 @@ "issue:3978|issue:4480", "issue:10085|issue:7142", "issue:3366|issue:3450", - "issue:3481|issue:4668", - "issue:5693|issue:6250", - "issue:10917|issue:7365" + "issue:3481|issue:4668" ], - "prepared_review_unit_hash": "c39b46bc5b4f125665b21e17a1d2d6765c0951a74c8330410ef1e0a5e4ed5175", + "prepared_review_unit_hash": "e0578cf35171d0a437b239d3fd68177d64a634ae8d76951ce4afd5339ab7a117", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33736, - "estimated_input_tokens": 8434, - "estimated_eval_tokens": 17124 + "serialized_chars": 33905, + "estimated_input_tokens": 8477, + "estimated_eval_tokens": 17210 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of mostly unrelated diffusers issues; the soft links are mostly title-level similarity, not shared root causes. No pair should be merged as duplicates.", - "confidence": 0.96, - "canonical_issue_reason": "Issue #10111 is the broadest umbrella title in the set and the best anchor for the FLUX-related pipeline bug reports, even though the overall cluster is heterogeneous.", + "summary": "This cluster is heterogeneous: the issues span unrelated pipelines, model families, training scripts, dtype/runtime errors, and feature requests. The soft-similarity pairs are not duplicate candidates for the same underlying bug/change.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 10210 is the most concrete, narrowly scoped bug report in the set (specific pipeline, clear symptom, reproducible behavior), so it is the best single issue representative if a canonical issue must be chosen.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #10111 is the most suitable representative because it is the most general and centrally named around the FLUX pipeline, while the other items are narrower or clearly unrelated.", + "best_issue_reason": "Issue 10210 is the strongest representative overall because it describes a precise, user-facing failure mode rather than a broad or speculative report.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:7056", + "right": "issue:8477", + "accept": false, + "reason": "Different failures in different training paths: SDXL LoRA fine-tuning error vs a Consistency Training script DDP error." + }, + { + "left": "issue:2201", + "right": "issue:6574", + "accept": false, + "reason": "Tokenizer padding behavior is unrelated to an SVD pipeline warning about expected input range." + }, { "left": "issue:3954", "right": "issue:590", "accept": false, - "reason": "Different scheduler bugs: multi-GPU training step counting vs a typo in the Karras noise formula." + "reason": "Multi-GPU training-step counting is unrelated to a typo in a Karras scheduler noise formula." }, { "left": "issue:2891", "right": "issue:5215", "accept": false, - "reason": "Both involve StableDiffusionUpscalePipeline, but one is an API/doc mismatch and the other is a brightness-quality regression." + "reason": "Docs/API mismatch for upscale embeds is not the same as a brightness regression on dark source images." }, { "left": "issue:4198", "right": "issue:989", "accept": false, - "reason": "Generic runtime errors, but they point to different failures: dtype mismatch in a notebook vs a 2-D weight assertion." + "reason": "A dtype mismatch in a notebook is unrelated to a linear-layer weight shape error." }, { "left": "issue:2880", "right": "issue:3081", "accept": false, - "reason": "Both are Colab failures, but they concern different notebooks and different workflows/backends." + "reason": "Different Colab breakages in different projects; no shared code path or symptom." }, { "left": "issue:5392", "right": "issue:7368", "accept": false, - "reason": "Both are training questions, but they target different models and feature requests, not the same bug or change." + "reason": "Training an unconditional latent diffusion model and training GLIGEN are different requests, not the same bug." }, { "left": "issue:6354", "right": "issue:9610", "accept": false, - "reason": "Both mention LoRA support, but they are different concrete pipeline additions for different models and use cases." + "reason": "Adding LoRA support for ControlNet is a different change from adding load_lora_weights to SD3 img2img." }, { "left": "issue:8514", "right": "issue:9415", "accept": false, - "reason": "Different model-specific errors with no shared code path or fix." + "reason": "EMAModel parallel fine-tuning failure and CogVideoX parameter-module errors affect different subsystems and code paths." }, { "left": "issue:10210", "right": "issue:7054", "accept": false, - "reason": "One is a FluxImg2Img output-quality bug; the other is a LoRA loader mutating the input state_dict." + "reason": "FluxImg2Img output collapse is unrelated to load_lora_weights mutating a passed state_dict." }, { "left": "issue:7575", "right": "issue:8863", "accept": false, - "reason": "Both are shape/attribute errors, but they describe different underlying failures in different components." + "reason": "Transformer2DModel adanorm shape mismatch is not the same as a generic tuple-has-no-shape AttributeError." }, { "left": "issue:1127", "right": "issue:2402", "accept": false, - "reason": "Different backend/runtime issues: ONNX/DML provider traceback vs PyTorch CUDA-not-enabled assertion." + "reason": "ONNX/DML import traceback is unrelated to Torch being built without CUDA." }, { "left": "issue:1207", "right": "issue:1750", "accept": false, - "reason": "Both are symbol/loading errors, but the symbols, environments, and root causes are different." + "reason": "Both are symbol/linker errors, but they refer to different missing CUDA-related symbols on different environments." }, { "left": "issue:10111", "right": "issue:7889", "accept": false, - "reason": "FLUX pipeline bugs are unrelated to single-file checkpoint loading bugs." + "reason": "A vague FLUX pipeline bug summary is not the same underlying issue as single_file_utils checkpoint-loading bugs." }, { "left": "issue:10518", "right": "issue:12781", "accept": false, - "reason": "Completely unrelated: an SD3 LoRA colab script issue vs a Discord login problem." + "reason": "An SD3 example script problem is unrelated to a Discord login/access issue." }, { "left": "issue:4217", "right": "issue:5989", "accept": false, - "reason": "Related to safetensors broadly, but one is load-time performance and the other is a library bug." + "reason": "Slow safetensors loading is not the same as a safetensors.torch bug." }, { "left": "issue:6256", "right": "issue:6452", "accept": false, - "reason": "Both are feature requests in pipelines, but they are distinct API additions with different targets." + "reason": "These are separate feature requests for different pipeline callbacks versus masked-image-latents support." }, { "left": "issue:8700", "right": "issue:8785", "accept": false, - "reason": "Both request PAG support, but for different pipelines/models and not the same change." + "reason": "Both mention PAG, but they target different model families and different pipeline additions." }, { "left": "issue:11432", "right": "issue:13249", "accept": false, - "reason": "Unrelated: torch_dtype auto-loading behavior vs a LoRA regression for Z-Image." + "reason": "torch_dtype='auto' loading behavior is unrelated to a LoRA regression on Z-Image." }, { "left": "issue:10204", "right": "issue:10475", "accept": false, - "reason": "Both are image-quality complaints, but they concern different pipelines and different contexts." + "reason": "FluxFill inpainting drift over iterations is different from SD3 fine-tuning inference quality mismatch." }, { "left": "issue:3978", "right": "issue:4480", "accept": false, - "reason": "Both involve MPS, but one is img2img not working and the other is a float16 kernel crash." + "reason": "Both are MPS-related, but one is img2img not working and the other is a float16 kernel crash; not the same bug." }, { "left": "issue:10085", "right": "issue:7142", "accept": false, - "reason": "Different inference problems: LyCORIS batch inference with skip_guidance_layers vs SDXL-lightning output mismatch." + "reason": "LyCORIS batch-inference/skip_guidance_layers runtime error is unrelated to SDXL-lightning inference differing from webui." }, { "left": "issue:3366", "right": "issue:3450", "accept": false, - "reason": "Different bugs: autocast being blocked by LoRALinearLayer vs a from_ckpt regression." + "reason": "LoRALinearLayer autocast behavior is unrelated to from_ckpt loading failures." }, { "left": "issue:3481", "right": "issue:4668", "accept": false, - "reason": "Both are image-quality complaints, but one is scheduler-step determinism and the other is LoRA-style weird outputs." - }, - { - "left": "issue:5693", - "right": "issue:6250", - "accept": false, - "reason": "Both involve VAE loading, but the failure modes and loading paths differ." - }, - { - "left": "issue:10917", - "right": "issue:7365", - "accept": false, - "reason": "Unrelated: a Lumina-2.0 script question vs pooled_prompt_embeds being overwritten." + "reason": "Different problems: DDIM step-count sensitivity versus weird images from a Kohya-style LoRA." } ] }, @@ -76918,6 +76939,7 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10798", + "issue:10917", "issue:11437", "issue:12078", "issue:12097", @@ -76929,29 +76951,28 @@ "issue:1633", "issue:1704", "issue:2359", - "issue:2500", - "issue:2501", "issue:2578", "issue:2780", - "issue:2838", - "issue:2858", + "issue:3026", "issue:3287", "issue:3481", "issue:4598", - "issue:4923", "issue:5006", "issue:5043", "issue:5112", "issue:5433", "issue:5485", - "issue:5605", + "issue:5693", "issue:5769", + "issue:6250", "issue:6506", "issue:6535", "issue:6790", "issue:6812", + "issue:6821", "issue:7210", "issue:725", + "issue:7365", "issue:7421", "issue:7535", "issue:8233", @@ -76966,6 +76987,8 @@ "issue:9539" ], "soft_pairs": [ + "issue:5693|issue:6250", + "issue:10917|issue:7365", "issue:1273|issue:1704", "issue:2359|issue:725", "issue:2578|issue:4598", @@ -76985,184 +77008,182 @@ "issue:3481|issue:7421", "issue:9306|issue:9539", "issue:10798|issue:12110", + "issue:3026|issue:6821", "issue:1353|issue:1377", - "issue:1633|issue:3287", - "issue:2500|issue:2501", - "issue:2838|issue:4923", - "issue:2858|issue:5605" + "issue:1633|issue:3287" ], - "prepared_review_unit_hash": "cc5c33c3a44361a4dcb4d1555c5d5eb50a681a271b26a46a8df58bac5c57b707", + "prepared_review_unit_hash": "463174fec3258b59d97d408e29ebe7896641ac4bfd5c3ba6963e76d90ebc7f2b", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33591, - "estimated_input_tokens": 8398, - "estimated_eval_tokens": 17052 + "serialized_chars": 33480, + "estimated_input_tokens": 8370, + "estimated_eval_tokens": 16996 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated bug reports, feature requests, questions, and model-specific runtime issues across different pipelines and versions. None of the soft pairs look like true duplicates of the same concrete bug/change.", - "confidence": 0.91, - "canonical_issue_reason": "Issue 12110 is the most concrete, self-contained bug report in the set, with a specific code path (UNet1DModel GaussianFourier time projection) and a clear failure mode. The cluster itself is not a true duplicate group, so this is only a representative anchor.", + "summary": "This cluster is a loose mix of unrelated diffusers issues spanning training, inference, loading, and model-specific bugs. None of the soft pairs look like true duplicates or the same underlying change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 12110 is the best triage representative because it is narrowly scoped, technically specific, and more actionable than the many broad questions, feature requests, or loosely related runtime complaints here.", + "best_issue_reason": "issue:12110 is the most concrete, clearly scoped bug report in the set, but it should be treated only as an anchor; the surrounding items are not duplicates of it.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5693", + "right": "issue:6250", + "accept": false, + "reason": "Both concern VAE loading, but one is a generic load error and the other is an SDXL single-file KeyError; different failure modes and code paths." + }, + { + "left": "issue:10917", + "right": "issue:7365", + "accept": false, + "reason": "A script-correctness question about Lumina-2.0 is unrelated to a specific pooled_prompt_embeds overwrite bug." + }, { "left": "issue:1273", "right": "issue:1704", "accept": false, - "reason": "Both are questions, but one is about Dreambooth mean-loss calculation and the other is a general attention-block question; no shared bug or code path." + "reason": "These are separate conceptual questions with no shared concrete bug or fix." }, { "left": "issue:2359", "right": "issue:725", "accept": false, - "reason": "Img2Img latent output support and DiffusionPipeline device_map/accelerate loading are different pipeline concerns and different fixes." + "reason": "One is about latent output_type behavior in img2img; the other is about device maps and loading memory footprint." }, { "left": "issue:2578", "right": "issue:4598", "accept": false, - "reason": "Editable install breakage and missing-dependency error messaging are unrelated problems." + "reason": "Editable install failure and missing-dependency error messaging are different problems." }, { "left": "issue:5769", "right": "issue:8671", "accept": false, - "reason": "Mixed-precision training correctness question and redundant SD3 dreambooth code cleanup are not the same bug/change." + "reason": "Mixed-precision training correctness and redundant code cleanup are not the same bug/change." }, { "left": "issue:7535", "right": "issue:8612", "accept": false, - "reason": "Type-hint compatibility and mixed-precision inference incompatibility are distinct issues." + "reason": "Type-hint incompatibility is unrelated to PixArt-Sigma mixed-precision inference issues." }, { "left": "issue:5433", "right": "issue:8792", "accept": false, - "reason": "Both mention image quality, but one is a scheduler/refiner artifact issue and the other is migration quality degradation; different root causes." + "reason": "Both are image-quality complaints, but they describe different pipelines, settings, and likely different root causes." }, { "left": "issue:6506", "right": "issue:6535", "accept": false, - "reason": "StableVideoDiffusion output_type behavior and scheduler-switch support are separate feature/bug requests." + "reason": "A return-type bug in StableVideoDiffusionPipeline is unrelated to scheduler customization support." }, { "left": "issue:6790", "right": "issue:9166", "accept": false, - "reason": "LCM state_dict key mismatch and image_processor overlay behavior are unrelated." + "reason": "Unexpected state_dict keys during LCM training has nothing to do with image_processor.apply_overlay behavior." }, { "left": "issue:11437", "right": "issue:12078", "accept": false, - "reason": "Training outputs/validation artifacts not being created is different from a bad example validation input in Flux control finetuning." + "reason": "Both mention validation/output during training, but one is missing artifacts and the other is a problematic example validation input." }, { "left": "issue:2780", "right": "issue:6812", "accept": false, - "reason": "Both are LoRA/fp16-related, but one is an unscale FP16 gradients error and the other is a QKV dtype mismatch; not the same concrete failure." + "reason": "Both are LoRA training errors, but the FP16 unscale error and Q/K/V dtype mismatch are distinct failures." }, { "left": "issue:12097", "right": "issue:13105", "accept": false, - "reason": "Wan2.2 end-of-run VRAM OOM and Wan2.1 output noise are different symptoms on different model versions." + "reason": "Different Wan models with different symptoms: end-of-run OOM versus output noise." }, { "left": "issue:5112", "right": "issue:5485", "accept": false, - "reason": "A torch.Size float-conversion crash in ControlNet img2img is unrelated to slow batch inference." + "reason": "A shape/type error in SDXL ControlNet Img2Img is not the same as ControlNet batch inference slowness." }, { "left": "issue:5043", "right": "issue:725", "accept": false, - "reason": "LoRA finetuning support for Wurstchen and pipeline loading/device-map memory reduction are unrelated requests." + "reason": "A feature request for W\u00fcrstchen LoRA finetuning is unrelated to device-map memory loading support." }, { "left": "issue:8534", "right": "issue:8673", "accept": false, - "reason": "Both are SD3-related, but one is wrong logit-normal weighting in training and the other is missing attention masks; different bugs." + "reason": "Both are SD3-related, but wrong logit-normal weighting and missing attention masks are separate training issues." }, { "left": "issue:5006", "right": "issue:8233", "accept": false, - "reason": "SDXL end-of-training BF16 OOM and depth-pipeline fp16 errors are separate runtime issues." + "reason": "These are different precision-related problems in different pipelines: BF16 end-of-training OOM versus an fp16 depth pipeline error." }, { "left": "issue:7210", "right": "issue:8752", "accept": false, - "reason": "Missing named_parameters with accelerate dispatch and offline from_single_file loading are different loading-path failures." + "reason": "Missing named_parameters with accelerate dispatch and offline from_single_file loading are different pipeline-loading issues." }, { "left": "issue:3481", "right": "issue:7421", "accept": false, - "reason": "Both touch reproducibility, but one is inference-step dependence in DDIM and the other is seed not working in KDiffusion; not the same bug." + "reason": "Different determinism complaints in different pipelines; not the same reproducible bug." }, { "left": "issue:9306", "right": "issue:9539", "accept": false, - "reason": "Wrong device selection on CogVideoXPipeline.from_pretrained and Float8 index_select_cuda errors in a different CogVideoX pipeline are not the same concrete issue." + "reason": "Both are CogVideoX issues, but wrong device selection and float8 index_select errors are distinct." }, { "left": "issue:10798", "right": "issue:12110", "accept": false, - "reason": "Flux quantization/device-combination behavior with bitsandbytes is unrelated to UNet1DModel GaussianFourier initialization." + "reason": "Flux quantization/device-combination issues are unrelated to UNet1DModel GaussianFourier initialization." + }, + { + "left": "issue:3026", + "right": "issue:6821", + "accept": false, + "reason": "An accelerate-launch error in instruct_pix2pix is unrelated to caption_column validation." }, { "left": "issue:1353", "right": "issue:1377", "accept": false, - "reason": "A pipeline-composition question and a transformers import error are unrelated." + "reason": "Combining a training script with LDMPipline and a missing transformers import are unrelated." }, { "left": "issue:1633", "right": "issue:3287", "accept": false, - "reason": "Optional karras sigmas in a wrapper pipeline and support for improved DDPM are different feature requests." - }, - { - "left": "issue:2500", - "right": "issue:2501", - "accept": false, - "reason": "Issue 2500 is one specific meta-tensor failure, while 2501 is a broad multi-test regression report; insufficient evidence they are the same root cause." - }, - { - "left": "issue:2838", - "right": "issue:4923", - "accept": false, - "reason": "ControlNet pipeline behavior and AudioLDM2 transcript support are unrelated." - }, - { - "left": "issue:2858", - "right": "issue:5605", - "accept": false, - "reason": "Seeding image support for UnCLIP img2img and per-image captions for SDXL Dreambooth are distinct feature requests." + "reason": "Both are scheduler-related feature requests, but they target different algorithms and changes; not the same concrete PR-sized fix." } ] }, @@ -77189,7 +77210,6 @@ "issue:10729", "issue:10932", "issue:10950", - "issue:11047", "issue:11092", "issue:1125", "issue:11885", @@ -77198,6 +77218,10 @@ "issue:1677", "issue:1889", "issue:2098", + "issue:2500", + "issue:2501", + "issue:2838", + "issue:2858", "issue:3052", "issue:3065", "issue:3149", @@ -77207,10 +77231,9 @@ "issue:3811", "issue:3951", "issue:4213", - "issue:4353", - "issue:4530", "issue:4584", "issue:4638", + "issue:4923", "issue:5225", "issue:5313", "issue:5371", @@ -77219,9 +77242,7 @@ "issue:5656", "issue:579", "issue:5890", - "issue:5903", "issue:5947", - "issue:5969", "issue:6136", "issue:6722", "issue:7053", @@ -77229,13 +77250,15 @@ "issue:8463", "issue:8577", "issue:911", - "issue:9218", "issue:9397", "issue:9496", "issue:9610", "issue:962" ], "soft_pairs": [ + "issue:2500|issue:2501", + "issue:2838|issue:4923", + "issue:2858|issue:5605", "issue:3052|issue:3638", "issue:3065|issue:5371", "issue:3811|issue:5313", @@ -77256,183 +77279,180 @@ "issue:7053|issue:7730", "issue:10950|issue:11092", "issue:3149|issue:3481", - "issue:3951|issue:6136", - "issue:4353|issue:5903", - "issue:4530|issue:5969", - "issue:11047|issue:9218" + "issue:3951|issue:6136" ], - "prepared_review_unit_hash": "1794fc43fdec50a5402660c6ebe0ed6ee5fb9b3c088e9b949a294e5935d0b432", + "prepared_review_unit_hash": "cb565fa66f429ce16021b0d1ca3a56751bff93a491aeb4e9be51f8a528604f4d", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33677, - "estimated_input_tokens": 8420, - "estimated_eval_tokens": 17096 + "serialized_chars": 33263, + "estimated_input_tokens": 8316, + "estimated_eval_tokens": 16888 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated runtime bugs, feature requests, docs issues, and LoRA/scheduler/pipeline tickets. None of the soft pairs look like true duplicates or the same concrete change.", - "confidence": 0.94, - "canonical_issue_reason": "No single canonical issue stands out because the items do not form one coherent underlying bug or feature area.", - "canonical_pr_reason": "No pull requests are present in the cluster.", - "best_issue_reason": "No best representative issue: the set spans many unrelated code paths, so any one issue would be arbitrary.", - "best_pr_reason": "No pull requests are present in the cluster.", + "summary": "This set is heterogeneous: most pairings are broad keyword/theme matches, not true duplicates. The soft edges span unrelated bugs, feature requests, and docs changes; none look like the same underlying fixable change. No PRs are present.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:3811 is the strongest standalone bug report here: it describes a concrete, user-visible LoRA loading failure with a clear symptom and likely actionable code path, even though it is not a duplicate of the other items.", + "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2500", + "right": "issue:2501", + "accept": false, + "reason": "Specific meta-tensor copy failure vs broad multi-test failures; different root causes and scope." + }, + { + "left": "issue:2838", + "right": "issue:4923", + "accept": false, + "reason": "ControlNet pipeline behavior request is unrelated to AudioLDM2 transcript/text-to-speech support." + }, + { + "left": "issue:2858", + "right": "issue:5605", + "accept": false, + "reason": "Seed image support for unCLIP img2img and per-image captions for DreamBooth are different features." + }, { "left": "issue:3052", "right": "issue:3638", "accept": false, - "reason": "Both touch diffusion-related features, but one is a control request for Versatile Diffusion and the other is a new Prompt-Free Diffusion pipeline; different artifacts and code paths." + "reason": "Versatile Diffusion control request and a new prompt-free diffusion pipeline are not the same change." }, { "left": "issue:3065", "right": "issue:5371", "accept": false, - "reason": "DreamBooth LoRA text-encoder tuning and BLIP inpainting support are separate feature requests with no shared bug/change." + "reason": "DreamBooth LoRA text-encoder tuning and BLIP inpainting support are unrelated feature requests." }, { "left": "issue:3811", "right": "issue:5313", "accept": false, - "reason": "Both involve LoRA, but one reports LoRA having no effect and the other reports fuse/unfuse not reverting; different failure modes." + "reason": "Both mention LoRA, but one is loading having no effect and the other is about non-revertible fuse; different code paths." }, { "left": "issue:5225", "right": "issue:9496", "accept": false, - "reason": "SDXL forward-pass VRAM usage and SD3ControlNet forward errors are unrelated model-specific bugs." + "reason": "SDXL memory exhaustion and SD3ControlNet forward errors are distinct model-specific bugs." }, { "left": "issue:5594", "right": "issue:6722", "accept": false, - "reason": "Memory-saving for SDXL fine-tuning is not the same as using SigLIP as a text encoder; different feature areas." + "reason": "SDXL LoRA fine-tuning memory usage vs using SigLIP as a text encoder are unrelated topics." }, { "left": "issue:5605", "right": "issue:9397", "accept": false, - "reason": "Per-image caption support for DreamBooth and SGMUniform scheduler support for Hyper LoRAs are distinct requests." + "reason": "DreamBooth per-image captions and SGMUniform scheduler support for Hyper LoRAs are different requests." }, { "left": "issue:5947", "right": "issue:8577", "accept": false, - "reason": "Diffusion-DPO policy discussion and Differential Diffusion for SD3 are different methods, not the same underlying change." + "reason": "Diffusion-DPO policy objective and Differential Diffusion in SD3 do not share the same underlying bug/change." }, { "left": "issue:10932", "right": "issue:12338", "accept": false, - "reason": "A torch.compile recompile-limit issue in HunyuanVideo is unrelated to AutoencoderDC VAE slicing during multi-image decoding." + "reason": "torch._dynamo recompile_limit on HunyuanVideo compile is unrelated to AutoencoderDC slicing multiple-image decoding." }, { "left": "issue:11885", "right": "issue:13311", "accept": false, - "reason": "Both mention FLUX LoRA loading, but they target different model families and different symptoms (missing effect vs missing keys prefix)." + "reason": "Both are Flux LoRA-loading issues, but they target different model generations and symptoms; not the same concrete bug." }, { "left": "issue:1677", "right": "issue:2098", "accept": false, - "reason": "Wrong input type to StableDiffusionPipeline and invalid torch.randn arguments are separate runtime errors." + "reason": "Wrong input type to StableDiffusionPipeline and invalid torch.randn arguments are different runtime errors." }, { "left": "issue:4584", "right": "issue:5890", "accept": false, - "reason": "Checkpoint resume failure in an SDXL LoRA training script is not the same as a tokenizer-length warning during text-to-image inference." + "reason": "resume_from_checkpoint in an SDXL LoRA training script is unrelated to a token-length warning in a different training script." }, { "left": "issue:579", "right": "issue:911", "accept": false, - "reason": "Both mention config problems, but textual inversion missing config.json and DreamBooth bad config are different workflows and fixes." + "reason": "Missing textual inversion config.json and DreamBooth bad config are different configuration failures." }, { "left": "issue:10729", "right": "issue:8463", "accept": false, - "reason": "The _hf_hook AttributeError in hook removal and the NoneType.tokenize error in InstructP2P-SDXL are unrelated code paths." + "reason": "_hf_hook deletion AttributeError and NoneType.tokenize in InstructP2P-SDXL are unrelated failures." }, { "left": "issue:4213", "right": "issue:4638", "accept": false, - "reason": "SDXL single-file loading without accelerate and SDXL textual inversion loading are different loading mechanisms and bugs." + "reason": "from_single_file() without accelerate and load_textual_inversion on SDXL are separate issues." }, { "left": "issue:1125", "right": "issue:962", "accept": false, - "reason": "Stable Diffusion Mega inpaint runtime error and inpaint size limitations are distinct issues, even though both involve inpainting." + "reason": "Inpaint pipeline runtime error and size support limitation are related to inpainting but not the same bug." }, { "left": "issue:1889", "right": "issue:3336", "accept": false, - "reason": "These are different scheduler proposals: DPM++ SDE variant versus DPM-Solver Multistep Inverse Scheduler." + "reason": "Different scheduler variants in the same family, but not the same concrete change or bug." }, { "left": "issue:5656", "right": "issue:9610", "accept": false, - "reason": "Both are LoRA-loader requests, but they target different pipelines and likely separate implementations; not the same concrete change." + "reason": "Adding a LoRA loader to community pipelines vs adding load_lora_weights to a specific SD3 img2img pipeline are different feature requests." }, { "left": "issue:7053", "right": "issue:7730", "accept": false, - "reason": "UNet3DConditionModel class conditioning and adding request timeouts are completely unrelated." + "reason": "UNet3D class conditioning and request timeout handling are unrelated issues." }, { "left": "issue:10950", "right": "issue:11092", "accept": false, - "reason": "A broken docs section and an empty-timesteps img2img bug are not the same issue." + "reason": "Broken docs and empty timesteps in SDXL-Turbo img2img are unrelated." }, { "left": "issue:3149", "right": "issue:3481", "accept": false, - "reason": "Non-deterministic images and changing outputs with different inference steps are not the same bug; the latter can be expected behavior." + "reason": "Both discuss determinism, but one is a version-wide nondeterminism regression while the other is expected step-dependent image differences." }, { "left": "issue:3951", "right": "issue:6136", "accept": false, - "reason": "Both mention NaNs, but they affect different pipelines/environments and do not share a concrete root cause." - }, - { - "left": "issue:4353", - "right": "issue:5903", - "accept": false, - "reason": "General SDXL quality complaints and edge artifacts from non-32-divisible resolutions are different symptoms and fixes." - }, - { - "left": "issue:4530", - "right": "issue:5969", - "accept": false, - "reason": "Color artifacts in SDXL refiner and DDIM NaNs with ZSNR patch are unrelated numerical/quality issues." - }, - { - "left": "issue:11047", - "right": "issue:9218", - "accept": false, - "reason": "Missing PAG inpaint model support and PAG with guess mode are separate feature requests, not the same bug." + "reason": "NaNs in UniDiffuser demo and NaNs in StableDiffusionPipeline on Mac CPU are different models/code paths." } ] }, @@ -77458,13 +77478,13 @@ "nodes": [ "issue:10191", "issue:10334", + "issue:11047", "issue:11148", "issue:11275", "issue:11486", "issue:12104", "issue:1314", "issue:13286", - "issue:1611", "issue:1709", "issue:1919", "issue:1923", @@ -77476,36 +77496,39 @@ "issue:2844", "issue:3381", "issue:3481", - "issue:3494", "issue:4149", + "issue:4353", "issue:4361", + "issue:4530", "issue:4631", "issue:4668", "issue:4913", "issue:5374", "issue:5578", + "issue:5903", + "issue:5969", "issue:6092", - "issue:6320", "issue:6359", "issue:7021", - "issue:7142", "issue:7636", "issue:7974", "issue:8441", "issue:8493", "issue:8509", "issue:8514", - "issue:852", "issue:8609", "issue:8894", "issue:8960", "issue:9154", - "issue:9298", + "issue:9218", "issue:9343", "issue:9768", "issue:9841" ], "soft_pairs": [ + "issue:4353|issue:5903", + "issue:4530|issue:5969", + "issue:11047|issue:9218", "issue:11486|issue:9841", "issue:1919|issue:1923", "issue:10191|issue:11275", @@ -77526,183 +77549,180 @@ "issue:2370|issue:6092", "issue:7021|issue:8441", "issue:4361|issue:8894", - "issue:10334|issue:9768", - "issue:1611|issue:852", - "issue:3494|issue:7142", - "issue:6320|issue:9298" + "issue:10334|issue:9768" ], - "prepared_review_unit_hash": "8942fadb094ba00f656192366f96953b10001e5f17f40a7cbb8ffca2c612c01e", + "prepared_review_unit_hash": "a43f9acef4788b8ed0fe44fb59e665d97f29e961bc8cfa144006a5d3a2d0a772", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34222, - "estimated_input_tokens": 8556, - "estimated_eval_tokens": 17368 + "serialized_chars": 34129, + "estimated_input_tokens": 8533, + "estimated_eval_tokens": 17322 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a noisy mix of unrelated training/loading/runtime issues. None of the soft pairs look like true duplicates or the same code change, so all are rejected.", - "confidence": 0.89, - "canonical_issue_reason": "issue:9841 is a concrete, reproducible training runtime bug with a clear failure mode, making it the strongest representative issue in this set.", + "summary": "This cluster is a mixed bag of unrelated diffusers bugs/features across training, loading, scheduler, and environment issues. None of the proposed soft pairs looks like the same underlying bug or change, so no duplicate merges are justified.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue cleanly subsumes the cluster; the items span distinct problems rather than one shared bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:9841 is the best issue to anchor triage because it is specific, actionable, and clearly a bug rather than a broad feature request or vague report.", + "best_issue_reason": "Issue 11148 is one of the clearest, most concrete bug reports in the set, but it still is not representative enough to serve as a true umbrella issue.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4353", + "right": "issue:5903", + "accept": false, + "reason": "Both concern image quality, but 4353 is a broad SDXL quality complaint while 5903 is specifically about edge artifacts from non-divisible resolutions." + }, + { + "left": "issue:4530", + "right": "issue:5969", + "accept": false, + "reason": "One reports SDXL refiner color artifacts; the other is a scheduler NaN regression with ZSNR patch. Different failure modes." + }, + { + "left": "issue:11047", + "right": "issue:9218", + "accept": false, + "reason": "Both mention PAG, but 11047 is about a missing controlnet inpaint model in the pipeline, while 9218 is about PAG with guess mode. Not the same bug." + }, { "left": "issue:11486", "right": "issue:9841", "accept": false, - "reason": "Feature request for xformers support vs a TypeError in DreamBooth SDXL LoRA training; different problem and code path." + "reason": "One asks for xformers support in training; the other is a DreamBooth SDXL LoRA TypeError around use_dora=False. Different code paths." }, { "left": "issue:1919", "right": "issue:1923", "accept": false, - "reason": "Both are setup/dependency problems, but one is missing datasets for DreamBooth and the other is missing CUDA libs in Docker/bitsandbytes; not the same underlying bug." + "reason": "Unrelated setup problems: missing Python dependency for a DreamBooth example versus missing CUDA libraries in a Dockerfile." }, { "left": "issue:10191", "right": "issue:11275", "accept": false, - "reason": "GPU device selection issue vs VAE ResnetUpsampleBlock2D NoneType addition; unrelated failures." + "reason": "GPU device selection in training is unrelated to the VAE NoneType addition error in ResnetUpsampleBlock2D." }, { "left": "issue:1314", "right": "issue:2844", "accept": false, - "reason": "Windows xformers backward error vs explicit image embeddings error in StableUnCLIPImg2Img; different pipeline and symptom." + "reason": "xformers backward error on Windows is unrelated to explicitly passing image embeddings to StableUnCLIPImg2ImgPipeline." }, { "left": "issue:9154", "right": "issue:9343", "accept": false, - "reason": "AutoEncoder/PT weight loading broken vs FLUX low_cpu_mem_usage/load_mismatched_sizes issue; separate model-loading problems." + "reason": "These are different loading bugs: PT VAE loading versus FLUX loading with low_cpu_mem_usage=False and ignore_mismatched_sizes=True." }, { "left": "issue:2044", "right": "issue:2068", "accept": false, - "reason": "DDP dtype attribute error during training vs from_pretrained dtype not converting model weights; related theme, but not the same bug." + "reason": "Both involve dtype wording, but one is a DDP wrapper missing dtype and the other is from_pretrained not converting model dtype. Not the same issue." }, { "left": "issue:3481", "right": "issue:4149", "accept": false, - "reason": "Scheduler/inference-step determinism complaint vs semantic pipeline batch dimension mismatch; different defects." + "reason": "Determinism across inference steps is a different problem from batch prompt/negative-prompt dimension mismatch." }, { "left": "issue:7636", "right": "issue:8609", "accept": false, - "reason": "ControlNet SDXL tile usage question vs custom_pipeline/IP-Adapter incompatibility; not duplicate issues." + "reason": "ControlNet SDXL tile usage and custom_pipeline HD painter with IP-Adapter are separate pipeline integration questions." }, { "left": "issue:8493", "right": "issue:8960", "accept": false, - "reason": "SD3Transformer2DModel state_dict loading error vs gradient accumulation not working in a training example; unrelated." + "reason": "State-dict loading for SD3Transformer2DModel is unrelated to gradient accumulation being nonfunctional in the basic training example." }, { "left": "issue:1709", "right": "issue:3381", "accept": false, - "reason": "ONNX img2img Tensor.astype error vs inpaint batch discoloration/docs issue; different pipelines and failures." + "reason": "ONNX tensor astype error and inpaint batch discoloration are different bugs with different symptoms and code paths." }, { "left": "issue:1931", "right": "issue:4631", "accept": false, - "reason": "Float16 type mismatch in Stable Diffusion 2 vs DPMSolverSinglestepScheduler scalar conversion error; distinct runtime paths." + "reason": "Float16 mismatch in Stable Diffusion 2 is unrelated to DPMSolverSinglestepScheduler scalar conversion with Karras sigmas." }, { "left": "issue:2020", "right": "issue:5374", "accept": false, - "reason": "Conversion script error vs unexpected checkpoint_path argument; different scripts and causes." + "reason": "Conversion-script failure and unexpected checkpoint_path argument are separate API/script issues." }, { "left": "issue:4668", "right": "issue:7974", "accept": false, - "reason": "Kohya-style LoRA weird images vs inpaint with IP Adapter bug; both visual quality complaints but not the same bug." + "reason": "Weird LoRA images and inpaint with IP-Adapter are different generation scenarios and likely different root causes." }, { "left": "issue:4913", "right": "issue:8509", "accept": false, - "reason": "ControlNet OpenPose SDXL model item vs ControlNet++ topic; no evidence of the same underlying issue." + "reason": "OpenPose SDXL model topic versus ControlNet++ are not the same concrete bug or fix." }, { "left": "issue:5578", "right": "issue:6359", "accept": false, - "reason": "Missing model_index.json error vs EulerDiscreteScheduler get_velocity/v_prediction error; unrelated loading vs scheduler API issue." + "reason": "Missing model_index.json and missing get_velocity on EulerDiscreteScheduler are unrelated loader versus scheduler API problems." }, { "left": "issue:11148", "right": "issue:8514", "accept": false, - "reason": "PEFT adapter flag not set vs EMAModel parallel fine-tuning error; different components and failure modes." + "reason": "PEFT adapter state handling and EMAModel parallel fine-tuning error are distinct training failures." }, { "left": "issue:12104", "right": "issue:13286", "accept": false, - "reason": "IndexError on empty dimension vs unsupported torchao >= 0.16.0 quantization; unrelated." + "reason": "Out-of-bounds indexing and torchao quantization support are unrelated errors." }, { "left": "issue:2370", "right": "issue:6092", "accept": false, - "reason": "LoRA training on a 24GB GPU vs example not training anything; too broad and not the same defect." + "reason": "A question about whether LoRA training fits on 24 GB GPU is not the same as a LoRA example that does not train anything." }, { "left": "issue:7021", "right": "issue:8441", "accept": false, - "reason": "HF_HOME import error vs create_dynamic_module recursion with relative HF_HOME; different symptoms and code paths." + "reason": "Both involve HF_HOME, but one is an import error and the other is infinite recursion from relative paths. Different bugs." }, { "left": "issue:4361", "right": "issue:8894", "accept": false, - "reason": "Safetensors shared-tensor saving issue vs training script error in train_cm_ct_unconditional.py; not the same bug." + "reason": "Shared-tensors safetensors save issue and train_cm_ct_unconditional failure are unrelated." }, { "left": "issue:10334", "right": "issue:9768", "accept": false, - "reason": "MacOS/MPS gray images and CPU NaNs vs torch 2.5 CuDNN SDPA NaN error; both NaN/image-quality issues, but different platforms and likely different root causes." - }, - { - "left": "issue:1611", - "right": "issue:852", - "accept": false, - "reason": "StableDiffusionMegaPipeline regression vs MPS unexpected device_map keyword error; unrelated compatibility issues." - }, - { - "left": "issue:3494", - "right": "issue:7142", - "accept": false, - "reason": "Multidiffusion corrupted images with schedulers vs SDXL-lightning inference mismatch with webui; similar outcome but different feature areas and causes." - }, - { - "left": "issue:6320", - "right": "issue:9298", - "accept": false, - "reason": "Remote model lookup failure vs missing Float8 storage object; separate loading/serialization problems." + "reason": "MacOS/MPS greys and CPU NaNs in Sana are not the same as a torch 2.5 CuDNN SDPA NaN backend issue." } ] }, @@ -77738,6 +77758,7 @@ "issue:12533", "issue:12540", "issue:13506", + "issue:1611", "issue:1791", "issue:2107", "issue:2262", @@ -77745,36 +77766,38 @@ "issue:2844", "issue:2961", "issue:3171", + "issue:3494", "issue:3883", "issue:3948", "issue:4080", "issue:4134", "issue:4159", - "issue:4194", "issue:4324", "issue:4403", "issue:455", "issue:4589", - "issue:4648", "issue:4720", "issue:5225", - "issue:5371", "issue:5593", - "issue:6001", "issue:6157", + "issue:6320", "issue:6807", + "issue:7142", "issue:7456", "issue:7535", - "issue:7802", "issue:7842", "issue:8234", "issue:843", + "issue:852", "issue:8577", - "issue:8943", "issue:8950", + "issue:9298", "issue:9511" ], "soft_pairs": [ + "issue:1611|issue:852", + "issue:3494|issue:7142", + "issue:6320|issue:9298", "issue:1791|issue:455", "issue:2674|issue:4159", "issue:4324|issue:8234", @@ -77795,183 +77818,180 @@ "issue:2262|issue:3883", "issue:3171|issue:843", "issue:4080|issue:4589", - "issue:4080|issue:8577", - "issue:4194|issue:6001", - "issue:4648|issue:8943", - "issue:5371|issue:7802" + "issue:4080|issue:8577" ], - "prepared_review_unit_hash": "be1c4caac7e0d5f7e782eba1c2451a5622c6275795f8624e00728c5b3125078b", + "prepared_review_unit_hash": "6708d58f6623f008b9179aa31c4591cc03d45ef10a61989056ec32584fd7a94b", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 32922, - "estimated_input_tokens": 8231, - "estimated_eval_tokens": 16718 + "serialized_chars": 32953, + "estimated_input_tokens": 8239, + "estimated_eval_tokens": 16734 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated issues that only share superficial wording around pipelines, offloading, or general runtime bugs. The soft pairs are false positives, not duplicates of the same underlying bug.", - "confidence": 0.97, - "canonical_issue_reason": "Issue 11162 is the broadest representative of the small WanImageToVideo offloading subset, but it is not a true cluster-wide canonical because the items are otherwise unrelated.", + "summary": "These items are mostly broad topical neighbors, not true duplicates. The soft pairs share themes like pipelines, offloading, MPS, VRAM, or Wan/Sana, but each points to a different bug, compatibility issue, or feature request.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue: the candidate pairs are heterogeneous and only loosely related by subsystem/theme, not by the same underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": "11162 is the most general issue here and the closest thing to a representative topic, but only for the Wan/offloading subgroup, not the full set.", + "best_issue_reason": "issue:11162 is the most concrete and representative item here, centered on a specific WanImageToVideoPipeline offloading problem; it is actionable, but not the same bug as the others.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1611", + "right": "issue:852", + "accept": false, + "reason": "Both mention older model/runtime compatibility on MPS, but one is a pipeline/model regression and the other is a device_map keyword error; different failure modes." + }, + { + "left": "issue:3494", + "right": "issue:7142", + "accept": false, + "reason": "Both are image-quality complaints, but one is Multidiffusion corruption across schedulers and the other is SDXL-lightning output mismatch vs webui." + }, + { + "left": "issue:6320", + "right": "issue:9298", + "accept": false, + "reason": "Remote model lookup failure and Float8 storage lookup error are unrelated bugs." + }, { "left": "issue:1791", "right": "issue:455", "accept": false, - "reason": "Different platform/backend failures: UnCLIP on Mac/MPS vs aten::index.Tensor on MPS." + "reason": "Both involve MPS, but one is a pipeline not supported on Mac and the other is an aten::index Tensor backend error; not the same bug." }, { "left": "issue:2674", "right": "issue:4159", "accept": false, - "reason": "LoRA dependency/packaging complaint vs DiffEdit VRAM usage problem; unrelated bugs." + "reason": "One is a LoRA packaging/requirements issue; the other is DiffEdit extreme VRAM usage. Same broad area, different problem." }, { "left": "issue:4324", "right": "issue:8234", "accept": false, - "reason": "Different runtime errors in different code paths: NoneType iteration in UNet vs ip_adapter_scale dict multiplication." + "reason": "A NoneType iteration crash in UNet forward is unrelated to ip_adapter_scale dict/Tensor multiplication." }, { "left": "issue:4403", "right": "issue:6157", "accept": false, - "reason": "Torch half-precision LayerNorm limitation vs ip_adapter not working in ControlNet img2img; not the same defect." + "reason": "One is a torch half-precision LayerNorm backend limitation; the other is ip_adapter not working in a controlnet img2img pipeline." }, { "left": "issue:4720", "right": "issue:7535", "accept": false, - "reason": "AutoencoderTiny incompatibility with img2img/inpainting vs a type-hint mismatch issue." + "reason": "AutoencoderTiny pipeline incompatibility is not the same as a FloatTensor type-hint mismatch." }, { "left": "issue:5593", "right": "issue:6807", "accept": false, - "reason": "AutoPipeline.from_pipe ControlNet conversion failure vs test collection failures on MPS." + "reason": "AutoPipeline.from_pipe failing on ControlNet pipelines is unrelated to MPS test collection failures." }, { "left": "issue:11661", "right": "issue:8950", "accept": false, - "reason": "Train-step argument handling in flux-control example vs SD3 LoRA training bug; different examples and failure modes." + "reason": "Training-step None handling in flux-control examples is unrelated to SD3 LoRA training bugs." }, { "left": "issue:2844", "right": "issue:5225", "accept": false, - "reason": "StableUnCLIP image-embedding runtime error vs SDXL forward-pass VRAM exhaustion; not duplicates." + "reason": "StableUnCLIP image embedding runtime error and SDXL forward-pass VRAM exhaustion are different issues." }, { "left": "issue:4134", "right": "issue:7842", "accept": false, - "reason": "MPS tensor size mismatch in ControlNet vs DeepFloyd stage-2 input-size divisibility crash; distinct pipelines and causes." + "reason": "MPS ControlNet tensor-size mismatch and DeepFloyd size-mismatch on non-multiple-of-8 images are separate shape bugs." }, { "left": "issue:12047", "right": "issue:13506", "accept": false, - "reason": "Wan2.2 LoRA fusion failure vs sage attention/ring-attention incompatibility; different components and symptoms." + "reason": "LoRA fusion on Wan2.2 GGUF and Sage attention/ring-attention failures are unrelated." }, { "left": "issue:10790", "right": "issue:9511", "accept": false, - "reason": "VRAM usage in Flux ControlNet CPU offload vs multi-controlnet batching for SDXL inpaint; different features." + "reason": "Both concern ControlNet, but one is CPU offload VRAM usage and the other is multi-ControlNet batching support for a specific pipeline." }, { "left": "issue:11000", "right": "issue:7456", "accept": false, - "reason": "WanImageToVideoPipeline missing shift argument vs a generic missing raise statement; no direct overlap." + "reason": "Missing shift argument in WanImageToVideoPipeline and a generic missing raise statement are not the same bug." }, { "left": "issue:11136", "right": "issue:12540", "accept": false, - "reason": "SanaSprint timestep support limitation vs SanaCombinedTimestepGuidanceEmbeddings incompatibility; same family but different bugs." + "reason": "SanaSprint intermediate timesteps support and SanaCombinedTimestepGuidanceEmbeddings incompatibility are different Sana pipeline issues." }, { "left": "issue:11162", "right": "issue:11163", "accept": false, - "reason": "Offloading-related Wan pipeline issue vs broken latent math when preparing latents; separate failure modes." + "reason": "Both are WanImageToVideoPipeline bugs, but one is about offloading support and the other is broken latent-preparation math; different code paths." }, { "left": "issue:12012", "right": "issue:12533", "accept": false, - "reason": "Wan first-block cache context error vs hooks conflict between context parallelism and CPU offload; related area but not the same bug." + "reason": "First is a Wan 2.2 first-block-cache context error; second is a hooks conflict between context parallelism and CPU offload." }, { "left": "issue:124", "right": "issue:2961", "accept": false, - "reason": "Dataset creation documentation request vs StableUnCLIP training example request; both docs-related but different asks." + "reason": "A dataset creation documentation request and a StableUnCLIP training example request are related to docs, not the same ask." }, { "left": "issue:2107", "right": "issue:3948", "accept": false, - "reason": "Low-resolution glitch/black image bug vs pale image output bug; different rendering issues." + "reason": "Low-resolution glitchy/black images and pale generated images are different rendering/color issues." }, { "left": "issue:2262", "right": "issue:3883", "accept": false, - "reason": "LoRA usage question vs realistic vision unsupported in dreambooth; unrelated user support topics." + "reason": "A safetensors LoRA usage question is unrelated to a realistic-vision dreambooth compatibility issue." }, { "left": "issue:3171", "right": "issue:843", "accept": false, - "reason": "LPW prompt support request for controlnet vs prompt syntax documentation; different documentation requests." + "reason": "LPW prompt support for controlnet and prompt syntax documentation are different documentation/features." }, { "left": "issue:4080", "right": "issue:4589", "accept": false, - "reason": "SDXL inpainting pipeline request vs ControlNet XL pipeline request; both feature requests but not the same change." + "reason": "One requests SDXL inpainting support; the other concerns XL ControlNet img2img/inpaint pipeline variants. Related area, but not the same change." }, { "left": "issue:4080", "right": "issue:8577", "accept": false, - "reason": "SDXL inpainting pipeline request vs differential diffusion support for SD3; different models and features." - }, - { - "left": "issue:4194", - "right": "issue:6001", - "accept": false, - "reason": "from_single_file VAE loading problem vs refiner returning the wrong pipeline class; distinct single-file loading bugs." - }, - { - "left": "issue:4648", - "right": "issue:8943", - "accept": false, - "reason": "Missing config.json in a local directory vs missing libtorch_cuda_cu.so during install; unrelated environment/setup problems." - }, - { - "left": "issue:5371", - "right": "issue:7802", - "accept": false, - "reason": "BLIP inpainting pipeline support request vs OLSS scheduler support request; different feature requests." + "reason": "Inpainting pipeline request and Differential Diffusion for SD3 are distinct features." } ] }, @@ -77990,52 +78010,52 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10191", "issue:1061", - "issue:10887", "issue:11962", - "issue:1439", "issue:175", - "issue:1801", "issue:1896", "issue:2024", - "issue:2202", "issue:2392", "issue:2439", "issue:2772", "issue:3194", "issue:351", "issue:4122", + "issue:4194", "issue:4213", "issue:4529", "issue:4557", "issue:4593", + "issue:4648", "issue:4885", "issue:5112", - "issue:5284", + "issue:5371", "issue:5463", "issue:5703", "issue:5972", "issue:5989", + "issue:6001", "issue:6320", "issue:6793", "issue:7041", "issue:7054", "issue:7191", - "issue:7264", "issue:7616", "issue:7681", "issue:775", "issue:7757", + "issue:7802", "issue:8523", "issue:8546", "issue:8551", "issue:8556", + "issue:8943", "issue:9134", "issue:9181", "issue:9248", @@ -78045,6 +78065,9 @@ "issue:9941" ], "soft_pairs": [ + "issue:4194|issue:6001", + "issue:4648|issue:8943", + "issue:5371|issue:7802", "issue:5989|issue:9248", "issue:6320|issue:8551", "issue:3194|issue:4557", @@ -78065,183 +78088,180 @@ "issue:4593|issue:5972", "issue:8546|issue:8556", "issue:1896|issue:2439", - "issue:1061|issue:175", - "issue:10887|issue:7264", - "issue:1439|issue:1801", - "issue:2202|issue:5284" + "issue:1061|issue:175" ], - "prepared_review_unit_hash": "ff66da48bc30d95214508367c34d37c21ea6a8bec907fb3896e76e0b60eefba0", + "prepared_review_unit_hash": "4afa09ffa1cb0d6bdf218283e3a611f41c2aeef4bfaa615d5dfb18a848f33ce7", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34167, - "estimated_input_tokens": 8542, - "estimated_eval_tokens": 17340 + "serialized_chars": 34193, + "estimated_input_tokens": 8549, + "estimated_eval_tokens": 17354 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is noisy and heterogeneous: it mixes unrelated CUDA/device, import, scheduler, LoRA, documentation, and feature-request issues. None of the soft pairs look like the same underlying bug or change, so all are rejected.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue is a safe choice; the items span many unrelated bugs and feature requests rather than one duplicate thread.", + "summary": "This cluster is mostly a false positive: the items span many unrelated bugs, feature requests, and docs/training questions across different pipelines and subsystems. No pull requests are present. A single SD3 from_single_file pair is superficially close, but still looks like a feature-gap report versus a concrete loading error, not the same underlying defect.", + "confidence": 0.91, + "canonical_issue_reason": "issue:8551 is the most concrete, actionable bug report in the set: it names a specific pipeline, a specific device_map setting, and a clear incorrect device-transfer behavior for VAE image latents.", "canonical_pr_reason": null, - "best_issue_reason": "No strong global representative exists; any issue pick would be arbitrary because the cluster is not centered on one concrete problem.", + "best_issue_reason": "issue:8551 is the strongest standalone issue for triage because it is narrowly scoped, reproducible, and tied to a specific code path rather than a broad support request or vague runtime error.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4194", + "right": "issue:6001", + "accept": false, + "reason": "Both mention from_single_file, but one is a VAE-loading failure and the other is a refiner returning the wrong pipeline class; different symptoms and likely different fixes." + }, + { + "left": "issue:4648", + "right": "issue:8943", + "accept": false, + "reason": "Missing config.json in a model directory is unrelated to a torch shared-object import/linker failure during installation." + }, + { + "left": "issue:5371", + "right": "issue:7802", + "accept": false, + "reason": "One is a request for BLIP inpainting support, the other is a scheduler support request; different features and code paths." + }, { "left": "issue:5989", "right": "issue:9248", "accept": false, - "reason": "Both are install/package-related, but one is a safetensors bug and the other is an upgrade command issue; not the same underlying problem." + "reason": "A safetensors.torch bug and a pip install upgrade request are not the same underlying issue." }, { "left": "issue:6320", "right": "issue:8551", "accept": false, - "reason": "Remote model lookup failure vs VAE device-map latent placement bug; different code paths and symptoms." + "reason": "Remote-model lookup failure for Emu2 is unrelated to SDXL inpaint image-latent device placement with device_map='balanced'." }, { "left": "issue:3194", "right": "issue:4557", "accept": false, - "reason": "bitsandbytes metadata/install failure is unrelated to the __getattr__ AttributeError." + "reason": "bitsandbytes package metadata missing is unrelated to a super().__getattr__ AttributeError." }, { "left": "issue:9613", "right": "issue:9731", "accept": false, - "reason": "A specific timesteps API regression and a general how-to/training question are not duplicates." + "reason": "A Timesteps.scale attribute error in 0.30.x is unrelated to a question about training LoRA on a custom dataset." }, { "left": "issue:4885", "right": "issue:5112", "accept": false, - "reason": "Both touch SDXL ControlNet img2img, but one is LoRA loading and the other is a float/shape construction error; different bugs." + "reason": "LoRA weights not loading in one pipeline and a torch.Size float-type error in another are different bugs." }, { "left": "issue:5703", "right": "issue:9181", "accept": false, - "reason": "DDIM inverse reconstruction vs a visual noise dot artifact are not the same failure." + "reason": "DDIM inverse noise reconstruction and a ControlNet noise artifact are different problems and code paths." }, { "left": "issue:7191", "right": "issue:9331", "accept": false, - "reason": "Generic import trouble is unrelated to the FlowMatchEuler scheduler index logic bug." + "reason": "A strange import problem is not the same as a scheduler index_for_timestep bug." }, { "left": "issue:7681", "right": "issue:9941", "accept": false, - "reason": "An OpenSSL/import environment error is not the same as a general Colab runtime failure report." + "reason": "An OpenSSL/lib attribute import failure is unrelated to a generic stable diffusion colab runtime error." }, { "left": "issue:2772", "right": "issue:5463", "accept": false, - "reason": "Train script dataset handling and an UnboundLocalError in image latents are different code-path bugs." + "reason": "Local dataset validation in train_controlnet.py is unrelated to an UnboundLocalError on image_latents." }, { "left": "issue:4122", "right": "issue:7041", "accept": false, - "reason": "Both are inference-related, but wrong timestep count and segmentation fault are distinct issues." + "reason": "SDXL timestep-count mismatch in DPMSolverMultistepScheduler is not the same as a segmentation fault when num_inference_steps > 1." }, { "left": "issue:7616", "right": "issue:8523", "accept": false, - "reason": "Low-denoise artifacts in Playground 2.5 are not the same as a CUDA error in Playground." + "reason": "Low-denoise img2img artifacts and a CUDA error in playground are too broad and do not indicate the same defect." }, { "left": "issue:10191", "right": "issue:7757", "accept": false, - "reason": "GPU selection going to the wrong device is different from multi-GPU OOM behavior." + "reason": "Forcing the wrong GPU device and multi-GPU OOM are distinct training issues with different failure modes." }, { "left": "issue:11962", "right": "issue:9134", "accept": false, - "reason": "A GGUF model support request and Flux LoRA training issues are unrelated." + "reason": "GGUF quantized model support for FLUX.1-Kontext-dev is unrelated to issues with FLUX LoRAs trained by SimpleTuner." }, { "left": "issue:2392", "right": "issue:6793", "accept": false, - "reason": "prompt_embeds validation in img2img is unrelated to hydra serialization failures." + "reason": "prompt_embeds handling in img2img is unrelated to Hydra serialization failures." }, { "left": "issue:2024", "right": "issue:351", "accept": false, - "reason": "A corrupted notebook file and a CUDA loading issue are unrelated." + "reason": "A corrupted notebook file and a CUDA-loading concern are unrelated." }, { "left": "issue:4213", "right": "issue:775", "accept": false, - "reason": "Missing accelerate for from_single_file is not the same as fp16/autocast behavior." + "reason": "from_single_file without accelerate and fp16/autocast behavior are different issues." }, { "left": "issue:4529", "right": "issue:7054", "accept": false, - "reason": "Both concern LoRA loading, but one is an unexpected keyword and the other mutates the passed state_dict; not the same bug." + "reason": "An unexpected network_alpha keyword error and state_dict mutation in load_lora_weights are separate LoRA-loading bugs." }, { "left": "issue:4593", "right": "issue:5972", "accept": false, - "reason": "LoRA rank conversion failure vs documentation about Min-SNR weighting are unrelated." + "reason": "Automatic1111 LoRA rank limits and Min-SNR training documentation are unrelated." }, { "left": "issue:8546", "right": "issue:8556", "accept": false, - "reason": "Both mention SD3/from_single_file, but one is missing support and the other is a meta-tensor load error; not the same fix." + "reason": "Both involve SD3 from_single_file, but one is a missing-support feature request while the other is a specific meta-tensor copy error; not enough evidence they are the same bug." }, { "left": "issue:1896", "right": "issue:2439", "accept": false, - "reason": "A shape/type error in CLIP-guided diffusion and an inpainting pipeline loading issue are different failures." + "reason": "A CLIP-guided Stable Diffusion size/type error is unrelated to an inpainting pipeline loading issue." }, { "left": "issue:1061", "right": "issue:175", "accept": false, - "reason": "Documentation update for a notebook and a question about torch.manual_seed are unrelated." - }, - { - "left": "issue:10887", - "right": "issue:7264", - "accept": false, - "reason": "Importing a missing trainer symbol is not the same as serializing/loading a custom transformers model." - }, - { - "left": "issue:1439", - "right": "issue:1801", - "accept": false, - "reason": "A missing transformers attribute and a bad getattr argument type are different attribute errors." - }, - { - "left": "issue:2202", - "right": "issue:5284", - "accept": false, - "reason": "Two feature requests in different areas (OpenVINO vs ControlNet-LLLite) are not duplicates." + "reason": "A Textual Inversion example update request is unrelated to whether torch.manual_seed is used correctly in training examples." } ] }, @@ -78260,25 +78280,24 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10198", "issue:1033", - "issue:10448", - "issue:11036", - "issue:11351", + "issue:10887", "issue:11418", - "issue:11580", - "issue:11648", "issue:1304", "issue:1323", + "issue:1439", "issue:1566", "issue:175", + "issue:1801", "issue:2024", "issue:2068", + "issue:2202", "issue:2775", "issue:2883", "issue:3184", @@ -78293,6 +78312,7 @@ "issue:5037", "issue:525", "issue:5256", + "issue:5284", "issue:5432", "issue:5515", "issue:5685", @@ -78301,6 +78321,7 @@ "issue:5968", "issue:6359", "issue:7054", + "issue:7264", "issue:7267", "issue:7408", "issue:7548", @@ -78308,13 +78329,15 @@ "issue:8645", "issue:8970", "issue:8975", - "issue:8997", "issue:9096", "issue:9147", "issue:9224", "issue:9241" ], "soft_pairs": [ + "issue:10887|issue:7264", + "issue:1439|issue:1801", + "issue:2202|issue:5284", "issue:2883|issue:7408", "issue:4888|issue:5432", "issue:5968|issue:9241", @@ -78335,183 +78358,180 @@ "issue:5256|issue:5905", "issue:1033|issue:3184", "issue:7267|issue:8578", - "issue:10198|issue:11418", - "issue:11648|issue:8997", - "issue:10448|issue:11580", - "issue:11036|issue:11351" + "issue:10198|issue:11418" ], - "prepared_review_unit_hash": "3b71930541d7cd85d174a53f84733f6bc9406fc38a408a636dfe01882be7cd19", + "prepared_review_unit_hash": "26827a5cc0618eb64a53f670ba8dcde3253688219b261b67d1766d3d250ccb3d", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33497, - "estimated_input_tokens": 8375, - "estimated_eval_tokens": 17006 + "serialized_chars": 33515, + "estimated_input_tokens": 8379, + "estimated_eval_tokens": 17014 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items do not form a coherent duplicate cluster; the soft pairs are mostly weak keyword matches across different models, pipelines, and bug types, so all candidate links should be rejected.", + "summary": "This cluster is heterogeneous: mostly unrelated bug reports, feature requests, and API questions across different models, pipelines, and training scripts. I do not see a credible duplicate core to unify them.", "confidence": 0.96, - "canonical_issue_reason": "No single issue is a clear canonical target here; the set is heterogeneous and the pairings are only superficially similar.", + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No issue cleanly represents the cluster overall; even the closest FLUX/fp8 items are about different underlying requests or failures.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10887", + "right": "issue:7264", + "accept": false, + "reason": "Different bugs: one is an import error for `StableDiffusionXLTrainer`, the other is pipeline serialization/loading of custom transformers models." + }, + { + "left": "issue:1439", + "right": "issue:1801", + "accept": false, + "reason": "Unrelated errors: missing `FeatureExtractionMixin` in transformers vs `getattr()` receiving a non-string attribute name." + }, + { + "left": "issue:2202", + "right": "issue:5284", + "accept": false, + "reason": "Different feature requests for different subsystems: OpenVINO support vs ControlNet-LLLite support." + }, { "left": "issue:2883", "right": "issue:7408", "accept": false, - "reason": "Both are pipeline-related, but one is a general image-to-video pipeline request and the other is a specific FRESCO pipeline topic; different underlying change." + "reason": "Both are pipeline requests, but for different models/use cases; not the same underlying change." }, { "left": "issue:4888", "right": "issue:5432", "accept": false, - "reason": "One is about offline safetensors loading, the other about blank validation images; unrelated bugs." + "reason": "Different issues: offline safetensors loading vs blank validation images in the SDXL LoRA training script." }, { "left": "issue:5968", "right": "issue:9241", "accept": false, - "reason": "Both mention conversion scripts, but the conversion targets are different and not the same code path." + "reason": "Both involve conversion tooling, but they target different formats and model workflows; not the same code-path problem." }, { "left": "issue:6359", "right": "issue:8645", "accept": false, - "reason": "Different missing API/functionality reports in different modules; not the same defect." + "reason": "Separate bugs: scheduler `get_velocity`/`add_noise` behavior vs a missing utility function in `single_file_utils.py`." }, { "left": "issue:7054", "right": "issue:8975", "accept": false, - "reason": "One is a mutation bug in LoRA loading, the other is an import error for the mixin; different failure modes." + "reason": "Different failures: mutating the passed LoRA state dict vs an import error for `LoraLoaderMixin`." }, { "left": "issue:1323", "right": "issue:2068", "accept": false, - "reason": "Both involve dtype, but one is a flax loading requirement and the other is about dtype conversion in from_pretrained; not the same issue." + "reason": "Related dtype theme, but different root causes: missing dtype on Flax model load vs `from_pretrained` not converting dtype." }, { "left": "issue:1566", "right": "issue:5829", "accept": false, - "reason": "Both are training-script crashes, but the causes are unrelated and in different LCM/Dreambooth paths." + "reason": "Different training crashes: fp16 checkpoint save failure in Dreambooth vs missing `unet_time_cond_proj_dim` for LCM scripts." }, { "left": "issue:8970", "right": "issue:9224", "accept": false, - "reason": "Both involve MPS, but one is black-image output and the other is an unsupported float64 dtype; distinct bugs." + "reason": "Both mention MPS, but one is a black-image output bug and the other is an unsupported hardcoded float64 dtype." }, { "left": "issue:3700", "right": "issue:5037", "accept": false, - "reason": "Different training-script problems with different arguments and code paths." + "reason": "Distinct script issues: `min_snr` argument handling vs a validation-image path bug." }, { "left": "issue:7548", "right": "issue:9096", "accept": false, - "reason": "Both concern bad outputs, but one is flicker in Animate Diff and the other is NaNs in Flux half precision; not the same bug." + "reason": "Different inference problems: AnimateDiff flicker vs Flux half-precision NaNs." }, { "left": "issue:5515", "right": "issue:5685", "accept": false, - "reason": "One is a concrete missing-scheduler initialization failure; the other is a broad scheduler discussion, not a matching defect." + "reason": "Scheduler-related, but not the same bug; one is missing scheduler initialization, the other is a broad scheduler-issues report." }, { "left": "issue:1304", "right": "issue:3377", "accept": false, - "reason": "A how-to training question and an SNR weighting question are not duplicates." + "reason": "A training how-to question vs an SNR-weighting optimization question; no shared bug or change." }, { "left": "issue:175", "right": "issue:3207", "accept": false, - "reason": "Completely different topics: seed usage in example code vs attention implementation behavior." + "reason": "Unrelated topics: seed usage in example training code vs attention implementation compatibility with torch < 2.0." }, { "left": "issue:2024", "right": "issue:525", "accept": false, - "reason": "Different notebooks and different failure modes; no shared underlying bug." + "reason": "Different notebook problems: corrupted notebook file vs loading images from local disk." }, { "left": "issue:2775", "right": "issue:4210", "accept": false, - "reason": "A torch._dynamo runtime error and an attention_mask API question are unrelated." + "reason": "Separate concerns: `torch._dynamo.optimize` misuse vs `attention_mask` usage in `UNet2DConditionModel`." }, { "left": "issue:3730", "right": "issue:4312", "accept": false, - "reason": "One is a timestep bug in inpainting blending; the other is a feature request about SAG support." + "reason": "Both touch inpainting, but one is a timestep blending bug and the other is a feature request for SAG in inpainting/img2img." }, { "left": "issue:4708", "right": "issue:9147", "accept": false, - "reason": "Both mention import/error contexts, but they concern different FLUX/autopipeline code paths." + "reason": "Import error in `diffusers.utils` vs Flux autopipeline error; different code paths and symptoms." }, { "left": "issue:5256", "right": "issue:5905", "accept": false, - "reason": "Both ask for new model support, but for different models and different training stacks." + "reason": "Both are feature requests, but for different models/training support; not the same underlying change." }, { "left": "issue:1033", "right": "issue:3184", "accept": false, - "reason": "Different pipelines and different errors; only broad stable-diffusion context overlaps." + "reason": "Different failures in different pipelines: Stable Diffusion Mega index error vs `lpw_stable_diffusion` generator-list handling." }, { "left": "issue:7267", "right": "issue:8578", "accept": false, - "reason": "Local data loading failure vs meta-tensor copy failure; unrelated bugs." + "reason": "Unrelated bugs: local data loading in `train_controlnet.py` vs meta-tensor copy failure with `sd_turbo 1.5`." }, { "left": "issue:10198", "right": "issue:11418", "accept": false, - "reason": "Both reference FLUX and weights, but one is a mismatch problem and the other is how to add an fp8 checkpoint." - }, - { - "left": "issue:11648", - "right": "issue:8997", - "accept": false, - "reason": "Different models and different problem types; not the same underlying issue." - }, - { - "left": "issue:10448", - "right": "issue:11580", - "accept": false, - "reason": "A file-loading/mmap feature request is unrelated to FP8 FLUX support." - }, - { - "left": "issue:11036", - "right": "issue:11351", - "accept": false, - "reason": "Different model-specific questions about latent conditioning vs processor dtype; not duplicates." + "reason": "Flux weight mismatch vs how to add a Flux fp8 safetensors file; one is a bug, the other is a usage question." } ] }, @@ -78530,16 +78550,21 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10052", + "issue:10448", "issue:10715", "issue:11023", + "issue:11036", + "issue:11351", "issue:11436", + "issue:11580", "issue:1160", + "issue:11648", "issue:12613", "issue:12632", "issue:12752", @@ -78556,7 +78581,6 @@ "issue:3767", "issue:4305", "issue:4419", - "issue:4480", "issue:4559", "issue:4605", "issue:4826", @@ -78567,23 +78591,22 @@ "issue:5245", "issue:5433", "issue:5574", - "issue:5630", "issue:5897", - "issue:6016", "issue:6079", "issue:6183", "issue:6359", - "issue:6560", "issue:6661", "issue:7053", "issue:7591", - "issue:7842", "issue:7921", "issue:8233", - "issue:8450", + "issue:8997", "issue:9731" ], "soft_pairs": [ + "issue:11648|issue:8997", + "issue:10448|issue:11580", + "issue:11036|issue:11351", "issue:1160|issue:5245", "issue:2737|issue:4559", "issue:3612|issue:7921", @@ -78604,183 +78627,180 @@ "issue:12613|issue:13350", "issue:6183|issue:7591", "issue:6661|issue:9731", - "issue:11436|issue:12632", - "issue:4480|issue:7842", - "issue:5630|issue:8450", - "issue:6016|issue:6560" + "issue:11436|issue:12632" ], - "prepared_review_unit_hash": "2e1b5ec9a9258f1f0475cc19bfb968e986a2119623393f86ff58f4fe110b6752", + "prepared_review_unit_hash": "ed3eb69a81a13dde289569229efa7d499b55e06b20f700c7a80bd4f7130d52c0", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33576, - "estimated_input_tokens": 8394, - "estimated_eval_tokens": 17044 + "serialized_chars": 33314, + "estimated_input_tokens": 8329, + "estimated_eval_tokens": 16914 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly false-positive similarity: the issues span unrelated bugs, feature requests, loader failures, dtype mismatches, and pipeline-specific rendering problems. None of the soft pairs look like the same underlying defect/change.", - "confidence": 0.87, - "canonical_issue_reason": null, + "summary": "These items are not a true duplicate cluster; most soft pairs only overlap on generic terms like dtype, pipeline, or training, but describe different bugs or feature requests.", + "confidence": 0.95, + "canonical_issue_reason": "No clear canonical issue: the set is heterogeneous (loaders, schedulers, training, image artifacts, dtype mismatches, feature requests) and does not center on one underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No single issue is a strong representative because there is no shared concrete failure mode across the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11648", + "right": "issue:8997", + "accept": false, + "reason": "Different problems: LoRA loading for an fp8 transformer model vs a scheduler error in Kolors." + }, + { + "left": "issue:10448", + "right": "issue:11580", + "accept": false, + "reason": "Unrelated topics: DDUF mmap loading vs FLUX fp8 support." + }, + { + "left": "issue:11036", + "right": "issue:11351", + "accept": false, + "reason": "Different code paths: latent-condition operations vs Wan i2v processor dtype behavior." + }, { "left": "issue:1160", "right": "issue:5245", "accept": false, - "reason": "Both describe hangs, but one is a pytest custom-pipeline freeze and the other is an import-stage freeze in pipenv; different code paths and symptoms." + "reason": "Both mention freezes, but one is a pytest hang and the other is an import-stage pipenv freeze; not the same bug." }, { "left": "issue:2737", "right": "issue:4559", "accept": false, - "reason": "Both are prompt/scheduler-adjacent feature requests, but CLIP guidance support and long weighted prompt support are distinct changes." + "reason": "Different feature requests: CLIP guidance support vs long weighted prompts." }, { "left": "issue:3612", "right": "issue:7921", "accept": false, - "reason": "Both mention ControlNet/IP-Adapter controls, but one asks for a reference ControlNet setup and the other for guidance start/end parameters; not the same fix." + "reason": "ControlNet creation/preprocessor request is not the same as IP-Adapter control_guidance_start/end usage." }, { "left": "issue:4921", "right": "issue:5433", "accept": false, - "reason": "Both concern SDXL image quality, but one is refiner erasing details and the other is DPM++ artifacts without the refiner; different underlying behavior." + "reason": "Both involve SDXL/refiner quality, but one is detail erasure and the other is scheduler artifacts without the refiner." }, { "left": "issue:3363", "right": "issue:5897", "accept": false, - "reason": "Both are training failures, but one is a machine-dependent failure in Dreambooth LoRA training and the other is a concrete cpu/cuda device mismatch." + "reason": "Both are training failures, but they are different scripts and different root causes." }, { "left": "issue:4826", "right": "issue:8233", "accept": false, - "reason": "Both are pipeline errors, but ControlNet Reference/MultiControlNet errors and Depth pipeline fp16 errors are unrelated failures." + "reason": "Different pipelines and errors: MultiControlNet reference issues vs fp16 depth pipeline failure." }, { "left": "issue:1677", "right": "issue:2228", "accept": false, - "reason": "Wrong dtype when calling StableDiffusionPipeline is a different bug from not loading the pipe onto GPU during from_pretrained." + "reason": "Wrong data type on pipeline call is not the same as failing to load the pipe onto GPU." }, { "left": "issue:13256", "right": "issue:5022", "accept": false, - "reason": "A packaging/version-compatibility complaint is not the same as a missing name import error in diffusers." + "reason": "One is dependency/version compatibility, the other is an import NameError in UNet2DModel." }, { "left": "issue:10052", "right": "issue:6359", "accept": false, - "reason": "A syntax error on ctx and a missing scheduler method/get_velocity issue are unrelated." + "reason": "Syntax error on ctx is unrelated to scheduler velocity/add_noise behavior." }, { "left": "issue:10715", "right": "issue:11023", "accept": false, - "reason": "Both produce bad images, but black images on ROCm for Kolors and blurry images in a custom SDXL pipeline are different model/path failures." + "reason": "Different image-quality failures on different models and hardware; not one underlying bug." }, { "left": "issue:12752", "right": "issue:13429", "accept": false, - "reason": "Both are image-edit related, but one requests img2img/inpainting support for Z-Image and the other asks to add a specific model; not the same change." + "reason": "Feature request for Z-Image img2img/inpainting is not the same as adding JoyAI-Image-Edit." }, { "left": "issue:1717", "right": "issue:3767", "accept": false, - "reason": "Offline cache loading and FrozenDict mutability are unrelated issues." + "reason": "Offline cache loading and FrozenDict immutability are unrelated." }, { "left": "issue:2518", "right": "issue:5070", "accept": false, - "reason": "Missing xformers support in an example and adding only_mid_control/sd_locked flags are separate example changes." + "reason": "Different example features: xformers support vs ControlNet flags." }, { "left": "issue:4305", "right": "issue:5143", "accept": false, - "reason": "torch.compile graph breaks and tiny output differences from the same input are different problems." + "reason": "torch.compile graph breaks are a different issue from tiny numerical output differences." }, { "left": "issue:4419", "right": "issue:7053", "accept": false, - "reason": "Offset noise not used in SDXL inference and UNet3D class conditioning are different feature gaps." + "reason": "SDXL offset-noise inference behavior is unrelated to UNet3DConditionModel class conditioning." }, { "left": "issue:4605", "right": "issue:6079", "accept": false, - "reason": "An inpaint/long-prompt pipeline error is not the same as a half-vs-float scalar type mismatch." + "reason": "Inpaint long-prompt error is not the same as a generic Half-vs-Float dtype mismatch." }, { "left": "issue:5143", "right": "issue:5574", "accept": false, - "reason": "Numerical non-determinism in UNet output is unrelated to LoRA training of attention parameters." + "reason": "Numerical output variance is unrelated to LoRA training of a subset of UNet attention parameters." }, { "left": "issue:12613", "right": "issue:13350", "accept": false, - "reason": "A nan-in-latent bug in Wan 2.2 i2v is unrelated to a how-to question for FLUX.2 batch inference." + "reason": "Wan i2v NaNs and FLUX.2 batch inference are different model/runtime issues." }, { "left": "issue:6183", "right": "issue:7591", "accept": false, - "reason": "Shape mismatch from from_single_file and a local_files_only HeaderTooLarge load error are different loader issues." + "reason": "Both use from_single_file, but one is a shape bug and the other is a local_files_only HeaderTooLarge error." }, { "left": "issue:6661", "right": "issue:9731", "accept": false, - "reason": "OOM during MPS training at certain image sizes is not the same as a general Playground2.5 LoRA training usage question." + "reason": "OOM on MPS training at certain image sizes is different from how to train a Playground2.5 LoRA." }, { "left": "issue:11436", "right": "issue:12632", "accept": false, - "reason": "Both are dtype mismatch errors, but they involve different models and different execution contexts; too broad to treat as one bug." - }, - { - "left": "issue:4480", - "right": "issue:7842", - "accept": false, - "reason": "MPS kernel crash with float16 and a tensor size mismatch in DeepFloyd stage 2 are unrelated failures." - }, - { - "left": "issue:5630", - "right": "issue:8450", - "accept": false, - "reason": "Both mention noisy SDXL outputs, but one is an LMS scheduler artifact and the other is noise at strength=1.0 in inpainting; not the same code path." - }, - { - "left": "issue:6016", - "right": "issue:6560", - "accept": false, - "reason": "A conversion script torch.Size() issue and a LoRA shape-size regression are not the same defect." + "reason": "Both are dtype mismatches, but they occur in different models and code paths (Hidream inference vs SDXL img2img)." } ] }, @@ -78811,16 +78831,12 @@ "issue:10768", "issue:10772", "issue:1083", - "issue:11047", "issue:11145", "issue:11148", - "issue:11561", "issue:1160", - "issue:1195", "issue:12108", "issue:12308", "issue:12540", - "issue:13430", "issue:1544", "issue:1961", "issue:1974", @@ -78833,11 +78849,11 @@ "issue:2941", "issue:3101", "issue:3116", - "issue:3162", "issue:3184", "issue:3799", "issue:4305", "issue:4332", + "issue:4480", "issue:4767", "issue:487", "issue:5046", @@ -78846,14 +78862,21 @@ "issue:5570", "issue:5578", "issue:5595", + "issue:5630", "issue:5760", "issue:5951", + "issue:6016", + "issue:6560", "issue:660", + "issue:7842", + "issue:8450", "issue:8989", - "issue:9220", "issue:984" ], "soft_pairs": [ + "issue:4480|issue:7842", + "issue:5630|issue:8450", + "issue:6016|issue:6560", "issue:1974|issue:3184", "issue:1064|issue:1961", "issue:2549|issue:2728", @@ -78874,193 +78897,187 @@ "issue:10010|issue:5578", "issue:10514|issue:12540", "issue:10768|issue:11145", - "issue:10772|issue:12308", - "issue:11047|issue:9220", - "issue:11561|issue:13430", - "issue:1195|issue:3162" + "issue:10772|issue:12308" ], - "prepared_review_unit_hash": "fe68975629a7a4754a30fc6ab3c75654bd67a690d4480053699ab4ebd3bc2534", + "prepared_review_unit_hash": "72774378900d22730f36a1956abdd3f7b80d9eb20d91e9e8624e03200965e06a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33748, - "estimated_input_tokens": 8437, - "estimated_eval_tokens": 17130 + "serialized_chars": 34059, + "estimated_input_tokens": 8515, + "estimated_eval_tokens": 17286 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated Diffusers issues were grouped by weak title similarity. Only the generator-list reproducible batching reports look like the same underlying bug; the rest are distinct bugs or feature requests across different pipelines/schedulers/models.", - "confidence": 0.86, - "canonical_issue_reason": "Issue 1974 is the broadest representative of the one plausible duplicate pair: both it and 3184 describe failures when passing a list of generators for reproducible batch generation.", + "summary": "This cluster is mostly a heterogeneous mix of Diffusers bugs/features across different models, schedulers, and platforms. I don\u2019t see any true duplicate pairs among the soft candidates; they are mostly thematically similar at a very high level but not the same underlying bug/change.", + "confidence": 0.78, + "canonical_issue_reason": "issue:2799 is the broadest, most generally phrased bug report in the set and is the best anchor if one issue must represent the cluster; the others are mostly narrower model-, pipeline-, or platform-specific failures.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 1974 is the best issue representative because it states the core shared bug more generally, while 3184 is a pipeline-specific manifestation of the same generator-list handling problem.", + "best_issue_reason": "issue:2799 is the most reusable representative of this mixed set because it targets a general checkpoint/pipeline mismatch problem rather than a narrow one-off failure.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4480", + "right": "issue:7842", + "accept": false, + "reason": "Both are runtime failures, but one is an M1 fp16/MPS crash and the other is a DeepFloyd image-size mismatch; different code paths and root causes." + }, + { + "left": "issue:5630", + "right": "issue:8450", + "accept": false, + "reason": "Both mention bad image output, but one is scheduler behavior in SDXL img2img and the other is inpainting strength=1.0 noise; not the same defect." + }, + { + "left": "issue:6016", + "right": "issue:6560", + "accept": false, + "reason": "Both involve size/shape errors, but one is conversion code and the other is LoRA loading/regression; different failing operations." + }, { "left": "issue:1974", "right": "issue:3184", - "accept": true, - "reason": "Same underlying bug: list-of-generators batch reproducibility is broken in both cases; one is a specific community pipeline manifestation." + "accept": false, + "reason": "Both mention generator lists, but one is Euler scheduler reproducibility and the other is a community pipeline input issue; not the same underlying bug." }, { "left": "issue:1064", "right": "issue:1961", "accept": false, - "reason": "Both are generic errors, but they involve different code paths and no shared concrete bug is evident." + "reason": "These are unrelated: an LMSDiscreteScheduler add_noise error versus a ColossalAI integration error." }, { "left": "issue:2549", "right": "issue:2728", "accept": false, - "reason": "Both concern UNet2DConditionModel, but one is about SD 2.0 model compatibility and the other about safetensors loading; distinct failures." + "reason": "One is an SD2-based UNet incompatibility, the other is safetensors loading in from_pretrained; same model family, different failures." }, { "left": "issue:2941", "right": "issue:4767", "accept": false, - "reason": "Related scheduler family, but one is undesired SD pipeline results and the other is prediction_type support; not the same concrete bug." + "reason": "Both touch EulerAncestralDiscreteScheduler, but one reports poor image results in SD and the other asks for prediction_type=sample support; not a single concrete bug." }, { "left": "issue:10636", "right": "issue:8989", "accept": false, - "reason": "Different subsystems: value_guided_sampling vs sequential offloading in Stable Audio." + "reason": "Different components entirely: value_guided_sampling.py versus Stable Audio sequential offloading." }, { "left": "issue:487", "right": "issue:660", "accept": false, - "reason": "Both mention Apple/MPS/FP16, but one is a general StableDiffusionPipeline CPU issue and the other is a float16-on-MPS configuration issue." + "reason": "Both are Apple/M1 float16-related, but one is CPU support and the other is MPS fp16 behavior; too different to treat as duplicates." }, { "left": "issue:11148", "right": "issue:12108", "accept": false, - "reason": "PEFT LoRA adapter state handling is unrelated to scheduler flow-matching failures in Qwen Image/Chroma pipelines." + "reason": "PEFT adapter state tracking versus flow-matching scheduler breakage in Qwen/Chroma pipelines; unrelated bugs." }, { "left": "issue:4332", "right": "issue:984", "accept": false, - "reason": "xformers support in Kandinsky2.2 is unrelated to a large-batch interpolation bug." + "reason": "xformers support in Kandinsky2.2 is unrelated to F.interpolate breaking for large batch sizes." }, { "left": "issue:5046", "right": "issue:5760", "accept": false, - "reason": "These are unrelated research/proposal issues with no shared bug or code path." + "reason": "A no-code feature proposal for DragonDiffusion versus a video-to-audio synthesis paper entry; not duplicates." }, { "left": "issue:5274", "right": "issue:5436", "accept": false, - "reason": "Different training failures with different symptoms and likely different root causes." + "reason": "fp16 mixed-precision training error versus a ControlNet XL training-example failure; different scenarios and error sources." }, { "left": "issue:1160", "right": "issue:5595", "accept": false, - "reason": "A pytest hang is unrelated to a height/width divisibility validation error." + "reason": "A hanging pytest versus a height/width divisibility validation error; unrelated." }, { "left": "issue:1083", "right": "issue:3116", "accept": false, - "reason": "FP16 ONNX output correctness and KDPM2DiscreteScheduler image2image noise are different problems." + "reason": "FP16 ONNX output correctness and KDPM2DiscreteScheduler img2img noise are different bugs." }, { "left": "issue:4305", "right": "issue:5570", "accept": false, - "reason": "torch.compile graph breaks and a missing compute_snr helper are unrelated." + "reason": "torch.compile graph breaks versus a missing compute_snr utility in packaging; unrelated code paths." }, { "left": "issue:2221", "right": "issue:2412", "accept": false, - "reason": "Both say 'colossalai error,' but that is too generic to establish the same underlying bug." + "reason": "Both are generic 'error' titles, but one is ColossalAI-related and the other is DreamBooth-related; no shared concrete bug." }, { "left": "issue:1544", "right": "issue:3799", "accept": false, - "reason": "Missing get_velocity and incorrect inferred type are separate scheduler API/type issues." + "reason": "A missing get_velocity method at runtime versus an incorrect type inference issue; one is API availability, the other is typing." }, { "left": "issue:3101", "right": "issue:5951", "accept": false, - "reason": "ControlNet training example errors and PixArt training are unrelated." + "reason": "ControlNet example training error versus PixArt training discussion; too broad to be the same issue." }, { "left": "issue:2439", "right": "issue:2799", "accept": false, - "reason": "The inpainting pipeline loading issue is not clearly the same as the explicit checkpoint/pipeline mismatch validation bug." + "reason": "Inpainting pipeline loading/running problems are not the same as explicitly raising on checkpoint/pipeline mismatch, even if they can overlap conceptually." }, { "left": "issue:10010", "right": "issue:5578", "accept": false, - "reason": "Both mention model_index.json, but one is a missing FLUX Fill model file and the other is a different OSError path." + "reason": "Both mention model_index.json, but one is a missing remote HF file and the other is a local-folder OSError; different root causes." }, { "left": "issue:10514", "right": "issue:12540", "accept": false, - "reason": "Both are Sana issues, but one is resolution binning/sample_size and the other is combined timestep guidance embeddings; different bugs." + "reason": "One is Sana resolution binning/sample_size support, the other is SanaCombinedTimestepGuidanceEmbeddings compatibility with SanaPipeline; distinct features/bugs." }, { "left": "issue:10768", "right": "issue:11145", "accept": false, - "reason": "Negative prompt support for Hunyuan video is unrelated to Flux Fill LoRA/dreambooth support." + "reason": "Hunyuan negative-prompt support and Flux Fill LoRA/dreambooth finetuning are unrelated model capabilities." }, { "left": "issue:10772", "right": "issue:12308", "accept": false, - "reason": "Sana ControlNet support and flash-attn GPU kernel support are different feature requests." - }, - { - "left": "issue:11047", - "right": "issue:9220", - "accept": false, - "reason": "Both touch PAG, but one is missing a controlnet inpaint model and the other is an undefined variable in a PAG pipeline." - }, - { - "left": "issue:11561", - "right": "issue:13430", - "accept": false, - "reason": "Both are model-support requests, but for different models and different integration work." - }, - { - "left": "issue:1195", - "right": "issue:3162", - "accept": false, - "reason": "Memory-efficient attention with fp16 weights and missing learned_classifier_free_sampling_embeddings are unrelated." + "reason": "Sana ControlNet support and flash-attn kernel support on non-Hopper GPUs are separate feature requests." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The assessment is grounded in the packet and stays conservative overall. The only accepted soft-edge pair, issue:1974 with issue:3184, is a plausible same-family generator-list batching bug based on the titles alone. The remaining pairs are rejected without overstating shared root causes, and the summary appropriately frames the matches as mostly title-similarity noise." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -79072,31 +79089,32 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:1083", + "issue:11047", + "issue:11561", + "issue:1195", "issue:12037", "issue:1205", "issue:12053", "issue:1284", "issue:1323", + "issue:13430", "issue:1486", "issue:1535", "issue:1909", "issue:1931", "issue:201", - "issue:2331", "issue:2414", "issue:2844", "issue:2857", + "issue:3162", "issue:3198", "issue:3441", - "issue:3494", "issue:3660", - "issue:3980", "issue:4075", "issue:4217", "issue:4308", @@ -79106,14 +79124,12 @@ "issue:4789", "issue:4911", "issue:5034", - "issue:542", "issue:5515", "issue:5598", "issue:5633", "issue:5645", "issue:6028", "issue:660", - "issue:6688", "issue:6801", "issue:6914", "issue:7428", @@ -79122,11 +79138,15 @@ "issue:8393", "issue:8485", "issue:8555", + "issue:9220", "issue:9297", "issue:9479", "issue:9813" ], "soft_pairs": [ + "issue:11047|issue:9220", + "issue:11561|issue:13430", + "issue:1195|issue:3162", "issue:12037|issue:12053", "issue:1205|issue:4436", "issue:1284|issue:1486", @@ -79148,189 +79168,186 @@ "issue:1284|issue:5034", "issue:1323|issue:4789", "issue:2414|issue:6801", - "issue:3441|issue:3660", - "issue:3494|issue:3980", - "issue:1083|issue:542", - "issue:2331|issue:6688" + "issue:3441|issue:3660" ], - "prepared_review_unit_hash": "9d91dff09bba063c1b2443236ebfe3351808c1b138b321d75fadb18bacb22245", + "prepared_review_unit_hash": "56c82977db5134f76302a91f0e29cd5de1b32b0dc1752c57a3e860b109aa007a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33625, - "estimated_input_tokens": 8407, - "estimated_eval_tokens": 17070 + "serialized_chars": 33665, + "estimated_input_tokens": 8417, + "estimated_eval_tokens": 17090 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft edge looks like a true duplicate. The candidates span different bugs/features and broad subsystems (conversion, schedulers, LoRA loading, performance, docs, and pipeline behavior), so this set does not form a single duplicate cluster.", + "summary": "This cluster is heterogeneous: it mixes unrelated feature requests, docs/test issues, load-time regressions, and model-specific bugs across many subsystems. None of the soft pairs look like the same underlying defect or change, so there are no duplicate merges here.", "confidence": 0.98, - "canonical_issue_reason": null, + "canonical_issue_reason": "No single canonical issue fits this cluster well; the items are too diverse and do not share one concrete bug or feature request.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 9813 is the most detailed and actionable standalone bug report in the set, but it is still not a good duplicate anchor for the cluster as a whole.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11047", + "right": "issue:9220", + "accept": false, + "reason": "Both mention PAG, but one is about missing ControlNet PAG inpaint support and the other is an undefined variable in a PAG pipeline; different problems and likely different code paths." + }, + { + "left": "issue:11561", + "right": "issue:13430", + "accept": false, + "reason": "Both are feature requests around image-edit pipelines, but they target different models and APIs (FluxFill IP-Adapter loading vs JoyAI-Image-Edit support)." + }, + { + "left": "issue:1195", + "right": "issue:3162", + "accept": false, + "reason": "fp16 memory-efficient attention failures are unrelated to missing classifier-free-sampling embeddings; different bug classes." + }, { "left": "issue:12037", "right": "issue:12053", "accept": false, - "reason": "Different model families and failure modes: WanTransformer3DModel compatibility vs Flux1.Dev Kohya LoRA text-encoder support." + "reason": "Different model families and failures: WanTransformer3D/Lightx2v LoRAs versus Flux1.Dev Kohya LoRA text-encoder layer support." }, { "left": "issue:1205", "right": "issue:4436", "accept": false, - "reason": "Both involve conversion, but one is ONNX export and the other is converting a CIVITAI safetensors checkpoint to diffusers; not the same bug." + "reason": "ONNX conversion for SD models is a different workflow from converting CIVITAI safetensors with a VAE into diffusers." }, { "left": "issue:1284", "right": "issue:1486", "accept": false, - "reason": "Batch-size slowdown is a performance concern, while DDIMScheduler prediction_type='sample' is a scheduler correctness bug." + "reason": "Longer sample time with bigger batches is a performance complaint, not the DDIMScheduler prediction_type='sample' correctness bug." }, { "left": "issue:1535", "right": "issue:2844", "accept": false, - "reason": "Image Variation Pipeline feature request is unrelated to explicit image-embedding runtime errors in StableUnCLIPImg2ImgPipeline." + "reason": "Image Variation Pipeline support and explicit image-embedding errors in StableUnCLIPImg2Img are separate pipeline issues." }, { "left": "issue:1909", "right": "issue:4483", "accept": false, - "reason": "Generator-list handling in pipelines is unrelated to SDXL generator determinism." + "reason": "Both involve generators, but one is about list-of-generators API support while the other is about SDXL nondeterminism; not the same bug." }, { "left": "issue:201", "right": "issue:4416", "accept": false, - "reason": "Test attention class vs T2I Adapter tests are different test targets, not the same issue." + "reason": "These are unrelated test tasks for different components, not the same underlying change." }, { "left": "issue:2857", "right": "issue:7428", "accept": false, - "reason": "UniDiffuser support request is unrelated to layer diffusion support request." + "reason": "Generic UniDiffuser support and layer diffusion support are different feature requests with different model targets." }, { "left": "issue:3198", "right": "issue:4911", "accept": false, - "reason": "Textual inversion safetensors loading differs from custom text_encoder loading in from_single_file." + "reason": "Safetensors textual inversion loading and custom text_encoder loading from_single_file are distinct loading paths." }, { "left": "issue:4075", "right": "issue:4308", "accept": false, - "reason": "Docstring typo in ShapEImg2ImgPipeline is unrelated to a guidance_scale default/documentation mismatch in SDXL." + "reason": "Both are docstring issues, but they concern different pipelines and different parameters." }, { "left": "issue:4217", "right": "issue:7785", "accept": false, - "reason": "Both mention slow loading, but one is safetensors load time generally and the other is very slow first inference in diffusers 0.27.x." + "reason": "Slow safetensors loading and very slow first inference are performance problems in different stages of the workflow." }, { "left": "issue:5598", "right": "issue:5633", "accept": false, - "reason": "_modify_text_encoder bug and transformer_temporal.py bug affect different code paths." + "reason": "Different internal modules and different bugs; no clear shared code-path defect." }, { "left": "issue:7428", "right": "issue:7670", "accept": false, - "reason": "Layer diffusion support request is unrelated to the IDDPM usage question." + "reason": "Layer diffusion support and IDDPM support are unrelated feature requests." }, { "left": "issue:8393", "right": "issue:9297", "accept": false, - "reason": "Checkpoint shard loading slowdown and multiple-LoRA load-time linear growth are different performance problems." + "reason": "Both mention load time, but one is about checkpoint shards and the other about loading multiple LoRAs; different objects and causes." }, { "left": "issue:8485", "right": "issue:8555", "accept": false, - "reason": "SD3 LoRA support request and SD3 FP8/text-encoder training request are distinct feature asks." + "reason": "Both are about SD3, but one asks for LoRA support while the other asks for FP8/text-encoder training; distinct capabilities." }, { "left": "issue:6914", "right": "issue:9813", "accept": false, - "reason": "IP-Adapter-FaceID offload AttributeError is a different bug from batch feature mixing in prepare_ip_adapter_image_embeds." + "reason": "Both touch IP-Adapter, but one is an offload-related shape AttributeError and the other is batch feature mixing in prepare_ip_adapter_image_embeds; not the same concrete bug." }, { "left": "issue:6028", "right": "issue:9479", "accept": false, - "reason": "Kandinsky 3.0 OOM is a memory-capacity issue; inpaint mask leakage is a correctness bug in a different pipeline." + "reason": "Kandinsky 3.0 OOM and StableDiffusionInpaintPipeline masking errors are unrelated issues." }, { "left": "issue:5515", "right": "issue:5645", "accept": false, - "reason": "LCM pipeline init-without-scheduler failure and new core LCM support being broken are related only loosely, but not the same concrete bug." + "reason": "Both involve LCM, but one is missing scheduler initialization and the other is a broader core support breakage; not a single mergeable fix." }, { "left": "issue:1931", "right": "issue:660", "accept": false, - "reason": "Both are float16-related, but one is a Stable Diffusion 2 dtype mismatch and the other is an MPS fp16 revision/runtime issue." + "reason": "Both are dtype-related, but one is SD2 float16 mismatch and the other is an MPS fp16 revision issue; different environments and root causes." }, { "left": "issue:1284", "right": "issue:5034", "accept": false, - "reason": "Batch-size slowdown is unrelated to checkpoint loading progress-bar display corruption." + "reason": "Batch-size slowdown and progress-bar display bugs are unrelated." }, { "left": "issue:1323", "right": "issue:4789", "accept": false, - "reason": "Flax dtype loading and DeepFloyd IF download failure are unrelated." + "reason": "Flax dtype-loading and DeepFloyd IF download failures are separate problems." }, { "left": "issue:2414", "right": "issue:6801", "accept": false, - "reason": "Missing transformers import when loading from hub is unrelated to scheduler loading errors in diffusers 0.26.0." + "reason": "A missing transformers import and scheduler loading failure are unrelated." }, { "left": "issue:3441", "right": "issue:3660", "accept": false, - "reason": "Higher VRAM usage under PyTorch 2 without xformers is a different problem from torch.compile incompatibility with LoRA/xformers." - }, - { - "left": "issue:3494", - "right": "issue:3980", - "accept": false, - "reason": "Multidiffusion corrupted images with schedulers is not the same as prompt-length differences vs A1111." - }, - { - "left": "issue:1083", - "right": "issue:542", - "accept": false, - "reason": "FP16 ONNX incorrect output and UNet export-to-ONNX are both ONNX-related, but they describe different export/precision issues and are not clearly the same bug." - }, - { - "left": "issue:2331", - "right": "issue:6688", - "accept": false, - "reason": "A ControlNet pipeline feature request is unrelated to a MotionCntrl feature request." + "reason": "PyTorch 2 VRAM regressions without xformers and torch.compile incompatibilities with LoRA/xformers are related only broadly, not the same code-path bug." } ] }, @@ -79349,14 +79366,14 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10658", "issue:10743", - "issue:10914", + "issue:1083", "issue:12149", "issue:12499", "issue:12565", @@ -79364,38 +79381,38 @@ "issue:1655", "issue:1755", "issue:1790", + "issue:2331", "issue:2412", "issue:2480", "issue:2783", "issue:2991", "issue:3336", "issue:3452", + "issue:3494", "issue:3593", "issue:3698", "issue:3805", "issue:3944", + "issue:3980", "issue:4241", - "issue:4551", "issue:4701", "issue:4972", "issue:499", "issue:5050", "issue:5120", "issue:5166", + "issue:542", "issue:543", - "issue:5577", "issue:5687", "issue:5761", "issue:5772", "issue:5905", "issue:6141", - "issue:6709", - "issue:6793", + "issue:6688", "issue:6928", "issue:7053", "issue:7240", "issue:730", - "issue:8613", "issue:9090", "issue:9227", "issue:9241", @@ -79403,6 +79420,9 @@ "issue:9902" ], "soft_pairs": [ + "issue:3494|issue:3980", + "issue:1083|issue:542", + "issue:2331|issue:6688", "issue:2412|issue:5120", "issue:5166|issue:7240", "issue:1655|issue:499", @@ -79423,183 +79443,180 @@ "issue:3698|issue:5905", "issue:3944|issue:4241", "issue:5050|issue:9241", - "issue:12499|issue:12565", - "issue:4551|issue:6709", - "issue:10914|issue:8613", - "issue:5577|issue:6793" + "issue:12499|issue:12565" ], - "prepared_review_unit_hash": "625fdfc72d443d7299f0e4c0460aaefb660be3c4dbbe39b14355ab32118f95a7", + "prepared_review_unit_hash": "948532922f418b44cab882ce6974ff22b4963a0278aa289950c5eabfa8296079", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 32737, - "estimated_input_tokens": 8185, - "estimated_eval_tokens": 16626 + "serialized_chars": 32619, + "estimated_input_tokens": 8155, + "estimated_eval_tokens": 16566 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the paired items are mostly unrelated feature requests, training issues, and version/configuration bugs. I would not merge any soft pair as duplicates. There are no PRs in the set; issue:10914 is the best single issue anchor only because it is a concrete open bug report, not because it matches the others closely.", - "confidence": 0.93, - "canonical_issue_reason": "issue:10914 is the most concrete open bug report here, but the cluster is too mixed to support a strong duplicate canonical beyond that.", + "summary": "This cluster is a noisy mix of unrelated issue reports and feature requests; none of the soft-linked pairs look like the same underlying bug or change, so there is no clear canonical duplicate target.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:10914 is the most actionable open issue, though it still does not represent the rest of the set well.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3494", + "right": "issue:3980", + "accept": false, + "reason": "Both concern generation quality, but one is about Multidiffusion/scheduler corruption and the other about long-prompt behavior in A1111; different bugs and code paths." + }, + { + "left": "issue:1083", + "right": "issue:542", + "accept": false, + "reason": "Both mention ONNX, but one reports incorrect FP16 output while the other is an export issue for UNet; not the same failure mode." + }, + { + "left": "issue:2331", + "right": "issue:6688", + "accept": false, + "reason": "Different feature requests: adding a ControlNet pipeline vs adding MotionCtrl support." + }, { "left": "issue:2412", "right": "issue:5120", "accept": false, - "reason": "Both mention DreamBooth training, but one is a generic error and the other is a prior-preservation regression; different bug causes and code paths." + "reason": "Both are DreamBooth-related, but one is a generic error and the other is a prior-preservation regression; not clearly the same bug." }, { "left": "issue:5166", "right": "issue:7240", "accept": false, - "reason": "Different model/pipeline problems: unconditional LDM VQVAE scaling vs low-quality output in ImageNet conditional LDM conversion." + "reason": "Different pipelines/models and different symptoms: missing VQ-VAE scaling in unconditional LDM vs low-quality output when converting ImageNet conditional LDM." }, { "left": "issue:1655", "right": "issue:499", "accept": false, - "reason": "One is an initialization failure for StableDiffusionImageVariationPipeline, the other is an unconditional training script failure." + "reason": "Pipeline initialization failure vs example training script failure; too different to be the same issue." }, { "left": "issue:1655", "right": "issue:543", "accept": false, - "reason": "Initialization bug vs DDP stride warning; not the same underlying issue." + "reason": "Pipeline init problem and a DDP stride warning are unrelated." }, { "left": "issue:10658", "right": "issue:6928", "accept": false, - "reason": "ONNX Runtime provider_options support and Marigold depthmap colorization are unrelated." + "reason": "Completely different topics: ONNX Runtime provider options vs Marigold depthmap colorization setting." }, { "left": "issue:10743", "right": "issue:9227", "accept": false, - "reason": "Both involve FLUX training, but one asks for ZeRO-3 support and the other reports a run failure with a specific model; not the same bug." + "reason": "Both are FLUX training-related, but one asks for ZeRO-3 support and the other reports a runtime problem with a specific FLUX model; not the same concrete bug." }, { "left": "issue:3805", "right": "issue:730", "accept": false, - "reason": "Colab subprocess failure vs DreamBooth local/Colab behavior; too broad and not the same defect." + "reason": "Colab execution error vs DreamBooth learning behavior differs; these are not the same problem." }, { "left": "issue:4701", "right": "issue:5761", "accept": false, - "reason": "AutoencoderKL float32 CUDA differentiation error is unrelated to the consistency decoder cost discussion." + "reason": "AutoencoderKL CUDA/autograd error vs consistency decoder cost question; unrelated." }, { "left": "issue:5687", "right": "issue:9090", "accept": false, - "reason": "Both mention repeated timesteps, but they are different schedulers and different failure modes, so not a single concrete bug." + "reason": "Both mention repeated timesteps, but one is Karras-sigma/img2img low-strength behavior and the other is a PNDM scheduler timestep bug; too different to merge." }, { "left": "issue:5772", "right": "issue:7053", "accept": false, - "reason": "Dataset backend performance question vs UNet3DConditionModel class conditioning bug; unrelated." + "reason": "Webdataset performance question vs UNet3D class-conditioning bug; unrelated." }, { "left": "issue:1608", "right": "issue:2991", "accept": false, - "reason": "Training stopping at 500 steps on T4 is not the same as a mixed-precision ControlNet example failure." + "reason": "Training stopping at 500 steps on T4 and a mixed-precision ControlNet example failure are different issues." }, { "left": "issue:3593", "right": "issue:4972", "accept": false, - "reason": "LoRA/xformers second-call failure and Wuerstchen example runtime failure are different code paths and symptoms." + "reason": "LoRA+xformers second-call failure vs wuerstchen example failing to run; different code paths and symptoms." }, { "left": "issue:1790", "right": "issue:2480", "accept": false, - "reason": "General training question vs missing checkpoint folder behavior in DreamBooth; not duplicates." + "reason": "A general 'how does training work' question is not the same as a specific checkpoint-folder bug in DreamBooth." }, { "left": "issue:12149", "right": "issue:9902", "accept": false, - "reason": "Both are version-related complaints, but one is pip/source alignment and the other is Conda pinning; not the same issue." + "reason": "Both are version-related complaints, but one is diffusers package/source mismatch and the other is a conda version pinning issue; unrelated." }, { "left": "issue:1755", "right": "issue:3336", "accept": false, - "reason": "Non-technical/general issue vs scheduler feature request; no duplicate relationship." + "reason": "No meaningful overlap; one is a generic/non-English issue title and the other is a scheduler feature request." }, { "left": "issue:2783", "right": "issue:6141", "accept": false, - "reason": "Pipeline creation request vs extra ControlNet parameters; one is much broader and not the same change." + "reason": "Both are ControlNet-related feature requests, but one asks for an img2img/inpainting pipeline and the other for control_mode/resize_mode parameters; different changes." }, { "left": "issue:3452", "right": "issue:988", "accept": false, - "reason": "Low-VRAM DreamBooth training and FSDP support are different training concerns." + "reason": "DreamBooth low-VRAM training vs FSDP support for examples are separate enablement topics." }, { "left": "issue:3698", "right": "issue:5905", "accept": false, - "reason": "Inpainting model conversion and diffusion-GAN model request are unrelated model-adoption topics." + "reason": "Different model support requests: inpainting model conversion vs adding diffusion-GAN." }, { "left": "issue:3944", "right": "issue:4241", "accept": false, - "reason": "LoRA rank customization in DreamBooth vs asking for LoRA fine-tuning support in unidiffusers; different requests." + "reason": "Both mention LoRA, but one is about network rank in DreamBooth training and the other about LoRA fine-tuning for unidiffusers; not the same change." }, { "left": "issue:5050", "right": "issue:9241", "accept": false, - "reason": "Config-change request in LoRA vs conversion script for XLab LoRA; not the same underlying task." + "reason": "LoRA config change request vs a conversion script from XLab LoRA to Diffusers LoRA; different artifacts and goals." }, { "left": "issue:12499", "right": "issue:12565", "accept": false, - "reason": "Related Wan I2V area, but one is a specific condition-shape mismatch and the other is a broad running issue for a different version." - }, - { - "left": "issue:4551", - "right": "issue:6709", - "accept": false, - "reason": "SDXL DreamBooth crash vs Custom Diffusion concept-list bug; both are training issues but not the same defect." - }, - { - "left": "issue:10914", - "right": "issue:8613", - "accept": false, - "reason": "Unexpected CPU offload and an OOM from caption_column are different runtime/resource problems." - }, - { - "left": "issue:5577", - "right": "issue:6793", - "accept": false, - "reason": "ControlNet single-file loading KeyError and Hydra serialization failure are unrelated failures." + "reason": "Both concern Wan i2v, but they target different versions and different symptoms (shape mismatch vs general running issue)." } ] }, @@ -79618,23 +79635,20 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10010", "issue:10178", "issue:10518", - "issue:10640", "issue:10803", - "issue:10986", - "issue:11088", + "issue:10914", "issue:11314", "issue:11315", "issue:11342", "issue:11412", - "issue:11448", "issue:11700", "issue:1399", "issue:1931", @@ -79647,32 +79661,38 @@ "issue:3468", "issue:3724", "issue:4386", + "issue:4551", "issue:459", + "issue:5577", "issue:5907", "issue:5917", "issue:6113", "issue:6686", + "issue:6709", "issue:672", + "issue:6793", "issue:7002", "issue:7380", - "issue:7483", "issue:7636", "issue:7716", "issue:7773", "issue:8241", "issue:8532", "issue:8588", + "issue:8613", "issue:8671", "issue:8780", "issue:8803", "issue:8805", "issue:8943", - "issue:9167", "issue:9471", "issue:9670", "issue:9719" ], "soft_pairs": [ + "issue:4551|issue:6709", + "issue:10914|issue:8613", + "issue:5577|issue:6793", "issue:1399|issue:5907", "issue:7636|issue:8241", "issue:10178|issue:7002", @@ -79693,183 +79713,180 @@ "issue:8532|issue:9670", "issue:9471|issue:9719", "issue:3468|issue:7716", - "issue:10518|issue:11314", - "issue:10640|issue:7483", - "issue:10986|issue:11088", - "issue:11448|issue:9167" + "issue:10518|issue:11314" ], - "prepared_review_unit_hash": "09a24a4931bf05547f1acd790835ab6ed988667546e8dddac0b5822146af957f", + "prepared_review_unit_hash": "73629a617fbad027a44c3ee53317db4ed7b37014395bff7257d1d48adfa7b5a0", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34091, - "estimated_input_tokens": 8523, - "estimated_eval_tokens": 17302 + "serialized_chars": 34060, + "estimated_input_tokens": 8515, + "estimated_eval_tokens": 17286 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes model-loading 404s, MPS/dtype/runtime bugs, conversion-script regressions, and training-example issues. None of the soft pairs look like true duplicates or mergeable fixes; they mostly share only broad subsystem keywords.", - "confidence": 0.27, - "canonical_issue_reason": "issue:1931 is the clearest, most concrete standalone bug report in the set: a specific runtime dtype mismatch with an explicit error message and reproducible inference path.", + "summary": "This cluster is heterogeneous: most items are unrelated model/script/runtime bugs that only share shallow text similarity. None of the soft pairs look like the same underlying issue or a mergeable duplicate PR.", + "confidence": 0.72, + "canonical_issue_reason": "issue:10914 is the broadest, most representative runtime symptom in the set (unexpected CPU offloading), so it is the least specific anchor among otherwise unrelated reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:1931 has the strongest canonical-triage qualities here: concise title, direct failure mode, and broad user impact on a core Stable Diffusion pipeline path.", + "best_issue_reason": "issue:10914 is the best single representative issue because it describes a general cross-model runtime behavior rather than a narrow model- or script-specific failure.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4551", + "right": "issue:6709", + "accept": false, + "reason": "Different failures and code paths: SDXL DreamBooth crash vs. custom diffusion concept-list training bug." + }, + { + "left": "issue:10914", + "right": "issue:8613", + "accept": false, + "reason": "Both involve resource symptoms, but one is unintended CPU offloading and the other is an OOM caused by caption-column training input handling." + }, + { + "left": "issue:5577", + "right": "issue:6793", + "accept": false, + "reason": "ControlNet checkpoint loading KeyError vs. Hydra serialization failure; unrelated subsystems." + }, { "left": "issue:1399", "right": "issue:5907", "accept": false, - "reason": "Both are API-compatibility complaints, but one is a model block constructor error and the other is a scheduler argument mismatch; different code paths and fixes." + "reason": "Unexpected UNet constructor arg in VersatileDiffusion vs. scheduler docs/API mismatch; not the same bug." }, { "left": "issue:7636", "right": "issue:8241", "accept": false, - "reason": "One is a usage question about SDXL tile ControlNet, the other is about an abnormal model size; no shared underlying bug." + "reason": "A usage question about SDXL tile ControlNet vs. an abnormal model size report; different problems." }, { "left": "issue:10178", "right": "issue:7002", "accept": false, - "reason": "Both mention size-related failures, but one is a 3D block dimension mismatch and the other is an img2img buffer-size blowup; not the same defect." + "reason": "3D block dimension mismatch vs. img2img invalid buffer size; different error contexts." }, { "left": "issue:10010", "right": "issue:8943", "accept": false, - "reason": "A missing Hugging Face model file and a CUDA shared-library install error are unrelated failure modes." + "reason": "Missing FLUX model_index.json from a repo URL vs. installation-time CUDA library error; unrelated." }, { "left": "issue:10803", "right": "issue:7380", "accept": false, - "reason": "A feature proposal for SANA variants and an implementation request for one-step diffusion are different changes, not duplicates." + "reason": "A proposal for SANARubber/SANA flexibility vs. a one-step diffusion implementation request; not the same change." }, { "left": "issue:11412", "right": "issue:7773", "accept": false, - "reason": "A 404 on a model repo and a batch-size generation bug are unrelated." + "reason": "404 for a specific Marigold model repo vs. a Stable Video Diffusion batch-size bug; different artifacts and code paths." }, { "left": "issue:2843", "right": "issue:6686", "accept": false, - "reason": "Both touch multi-GPU/optimization topics, but one is about checkpoint saving and the other about Dynamo performance after save/load; different bugs." + "reason": "Multi-GPU checkpoint saving vs. Dynamo compilation performance after save/load; different behaviors." }, { "left": "issue:3724", "right": "issue:459", "accept": false, - "reason": "A watermarking feature request and an incremental diffusion idea are separate enhancement proposals." + "reason": "Watermarking module idea vs. incremental diffusion concept; both feature ideas, but not the same request." }, { "left": "issue:4386", "right": "issue:672", "accept": false, - "reason": "A question about SDXL LoRA conversion and a ckpt-format conversion request are related only by broad conversion tooling, not the same issue." + "reason": "SDXL LoRA conversion question vs. converting diffusers weights to original CompVis ckpt format; related theme, but not the same concrete task." }, { "left": "issue:5917", "right": "issue:6113", "accept": false, - "reason": "One is a regression in a conversion utility after upgrade; the other is a general how-to question about using a repo model." + "reason": "Checkpoint conversion regression after upgrade vs. how to use a control repo in diffusers; not the same underlying bug." }, { "left": "issue:8588", "right": "issue:8805", "accept": false, - "reason": "Both are conversion-script problems, but they concern different scripts and different missing/incorrect behavior." + "reason": "Missing SD3 conversion script vs. an SDXL conversion script that no longer works; both conversion-related, but different scripts and failures." }, { "left": "issue:8671", "right": "issue:8803", "accept": false, - "reason": "A cleanup/refactor note in train_dreambooth_lora_sd3.py and a duplicate assignment in attention_processor.py are different files and issues." + "reason": "Redundant code in a SD3 training script vs. duplicate assignment in attention_processor.py; unrelated code locations." }, { "left": "issue:2606", "right": "issue:3177", "accept": false, - "reason": "Both involve DeepSpeed training, but one is a save_state hang and the other is stage-3 offloading behavior; not the same bug." + "reason": "DeepSpeed save_state hang vs. DreamBooth ZeRO-3 offloading being broken; both training-related, but different failure modes." }, { "left": "issue:11315", "right": "issue:11342", "accept": false, - "reason": "Both are MPS-related, but they affect different models and have different root causes (float64 rope vs pipeline failure)." + "reason": "Both are MPS issues, but one is a float64 RoPE implementation bug in HiDream and the other is a Sana-sprint pipeline failure." }, { "left": "issue:1931", "right": "issue:2340", "accept": false, - "reason": "A float16/float mismatch in Stable Diffusion 2 is not the same as a tiny-pipeline safety-checker error." + "reason": "Float16 dtype mismatch in SD2 vs. tiny pipeline safety-checker error; unrelated code paths." }, { "left": "issue:11700", "right": "issue:8780", "accept": false, - "reason": "A request to add pruna integration is unrelated to a generic model/input dtype mismatch error." + "reason": "Pruna integration request vs. a model/input dtype mismatch bug; different topics." }, { "left": "issue:2841", "right": "issue:2905", "accept": false, - "reason": "Both are scheduler-related feature requests, but one asks for temperature sampling and the other for Karras sigmas; separate changes." + "reason": "Both touch schedulers, but one requests sampling temperature and the other requests Karras sigmas; separate feature changes." }, { "left": "issue:8532", "right": "issue:9670", "accept": false, - "reason": "Progressive memory growth while loading LoRA weights is unrelated to a missing PEFT backend dependency error." + "reason": "LoRA memory growth during loading vs. PEFT backend requirement failure; different loading problems." }, { "left": "issue:9471", "right": "issue:9719", "accept": false, - "reason": "A file-extension inference error and an ignored progress-bar flag are different loading behaviors with different fixes." + "reason": "Multiple file extensions/config inference error vs. progress bar being ignored during shard loading; unrelated loader behaviors." }, { "left": "issue:3468", "right": "issue:7716", "accept": false, - "reason": "Both mention missing model metadata files, but one is about Dreambooth not generating model_index.json and the other is about a directory missing config.json; not the same underlying bug." + "reason": "Missing model_index.json after DreamBooth export vs. missing config.json in a different directory; similar symptom, but different artifacts and flows." }, { "left": "issue:10518", "right": "issue:11314", "accept": false, - "reason": "Two separate training-script complaints for different models and scripts; the underlying faults are not shown to be the same." - }, - { - "left": "issue:10640", - "right": "issue:7483", - "accept": false, - "reason": "A multi-GPU FluxInpaintPipeline DataParallel runtime error is unrelated to an SDXL LoRA conversion-script incompatibility." - }, - { - "left": "issue:10986", - "right": "issue:11088", - "accept": false, - "reason": "An MPS complex-dtype conversion bug and a wan2.1 gguf load error are different code paths and failure modes." - }, - { - "left": "issue:11448", - "right": "issue:9167", - "accept": false, - "reason": "A missing model repository and an import failure for StableDiffusion3Pipeline are distinct issues." + "reason": "Wrong SD3 training script vs. Flux ControlNet training script bug; different scripts and model families." } ] }, @@ -79894,7 +79911,11 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10281", + "issue:10640", + "issue:10986", + "issue:11088", "issue:1109", + "issue:11448", "issue:11989", "issue:12108", "issue:1344", @@ -79926,22 +79947,22 @@ "issue:5670", "issue:5689", "issue:6079", - "issue:6162", - "issue:6452", "issue:6725", + "issue:7483", "issue:7644", "issue:7735", "issue:7962", "issue:8464", - "issue:8482", "issue:8488", - "issue:8813", - "issue:9495", + "issue:9167", "issue:9856", "issue:9861", "issue:9957" ], "soft_pairs": [ + "issue:10640|issue:7483", + "issue:10986|issue:11088", + "issue:11448|issue:9167", "issue:1375|issue:1495", "issue:11989|issue:9861", "issue:5670|issue:7644", @@ -79962,183 +79983,180 @@ "issue:4080|issue:7962", "issue:4171|issue:5246", "issue:4484|issue:5433", - "issue:5606|issue:7735", - "issue:6162|issue:9495", - "issue:6452|issue:7962", - "issue:8482|issue:8813" + "issue:5606|issue:7735" ], - "prepared_review_unit_hash": "b36e50724d01650523a783481184427bb7432cc6a7cc48c966d4adc61097ed61", + "prepared_review_unit_hash": "8ca0713ac1b6fe26f93b95c4c1d53f13c0aca63effe7286b759b9ed754c6c1e6", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33499, - "estimated_input_tokens": 8375, - "estimated_eval_tokens": 17006 + "serialized_chars": 34120, + "estimated_input_tokens": 8530, + "estimated_eval_tokens": 17316 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous issue reports with no pull requests. The soft pairs are generally not duplicates; the closest one is the SDXL download behavior pair, but it still looks like different loading paths/trigger conditions rather than the same bug.", - "confidence": 0.79, - "canonical_issue_reason": "Issue 5433 is the strongest representative issue: it is a concrete, well-described SDXL inference bug with substantial discussion and inbound references.", - "canonical_pr_reason": null, - "best_issue_reason": "Issue 5433 is the best overall issue candidate because it is the most substantive and broadly representative bug report in the set.", - "best_pr_reason": null, + "summary": "This cluster is heterogeneous: the suggested pairs are only loosely text-similar and do not share the same concrete bug or change. I reject all soft edges.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits this cluster; it mixes unrelated bugs, feature requests, import errors, scheduler issues, and training/runtime problems.", + "canonical_pr_reason": "No PRs are present in this cluster.", + "best_issue_reason": "No issue is a good global representative because the items do not converge on one underlying defect or feature.", + "best_pr_reason": "No PR is present to evaluate as a representative change.", "soft_edge_verdicts": [ + { + "left": "issue:10640", + "right": "issue:7483", + "accept": false, + "reason": "Different problems: FluxInpaintPipeline DataParallel multi-GPU error vs SDXL LoRA conversion failure." + }, + { + "left": "issue:10986", + "right": "issue:11088", + "accept": false, + "reason": "MPS ComplexDouble conversion bug and wan2.1 GGUF load error are unrelated." + }, + { + "left": "issue:11448", + "right": "issue:9167", + "accept": false, + "reason": "Missing pipeline path vs missing StableDiffusion3Pipeline import; different issues." + }, { "left": "issue:1375", "right": "issue:1495", "accept": false, - "reason": "Both are import-related, but they describe different failures and different symbols: a regression from PR #1365 vs missing init_git_repo in hub_utils." + "reason": "Both are import-related, but they fail on different symbols and root causes." }, { "left": "issue:11989", "right": "issue:9861", "accept": false, - "reason": "Different problems: black inference images tied to weights vs Flux training not updating the transformer model." + "reason": "Black output images and Flux training not updating the transformer are different bugs." }, { "left": "issue:5670", "right": "issue:7644", "accept": false, - "reason": "Both are shape/size-related runtime failures, but one is PixArtAlpha non-square image generation and the other is Stable Cascade non-power-of-two handling." + "reason": "Non-square PixArtAlpha failure is not the same as Stable Cascade non-power-of-two image failure." }, { "left": "issue:2659", "right": "issue:6725", "accept": false, - "reason": "One reports batch-size-dependent output drift; the other is a generic request to generate images with LDMPipeline." + "reason": "Batch-size variation in generated images is unrelated to a request to generate images with LDMPipeline." }, { "left": "issue:4580", "right": "issue:8488", "accept": false, - "reason": "Different code paths: UniPC/Karras scheduler exception vs StableDiffusion3Pipeline from_pretrained loading exception." + "reason": "UniPC use_karras_sigmas exception and StableDiffusion3 from_pretrained exception are different code paths." }, { "left": "issue:1867", "right": "issue:337", "accept": false, - "reason": "Both involve img2img, but one is a low-strength scheduler limitation and the other is a text-guided inpainting problem." + "reason": "Both touch img2img, but one is a low-strength scheduler bug and the other is a text-guided inpainting problem." }, { "left": "issue:2163", "right": "issue:2213", "accept": false, - "reason": "Different failures in training examples: dtype mismatch vs TPU distributed hang." + "reason": "Training dtype mismatch in one script vs TPU distributed hang in another are unrelated." }, { "left": "issue:4708", "right": "issue:6079", "accept": false, - "reason": "Import error in diffusers.utils is unrelated to a Half-vs-Float tensor type mismatch." + "reason": "Import error for diffusers.utils is not the same as a Half-vs-Float tensor type mismatch." }, { "left": "issue:12108", "right": "issue:13446", "accept": false, - "reason": "One is a scheduler compatibility break in Qwen/Chroma pipelines; the other is a trust-check/security issue for custom_pipeline." + "reason": "Flow-matching scheduler breakage and custom_pipeline trust checks are unrelated." }, { "left": "issue:8464", "right": "issue:9856", "accept": false, - "reason": "Different operational issues: SEGA + cpu offload + seed vs distributed port collision." + "reason": "Seed handling with CPU offload is unrelated to a distributed port-in-use error." }, { "left": "issue:10281", "right": "issue:9957", "accept": false, - "reason": "Both are scheduler requests, but FreeScale and BDIA-DDIM are distinct proposals and not the same underlying change." + "reason": "Different scheduler feature requests: FreeScale vs BDIA-DDIM." }, { "left": "issue:2851", "right": "issue:3717", "accept": false, - "reason": "TPU push_to_hub failure is unrelated to multi-GPU OOM." + "reason": "TPU push_to_hub limitation and multi-GPU OOM are different runtime issues." }, { "left": "issue:3585", "right": "issue:5110", "accept": false, - "reason": "LoRA synthesis failure and HFValidationError repo-id formatting are different issues." + "reason": "LoRA synthesis failure and HFValidationError repo-id formatting are unrelated." }, { "left": "issue:1109", "right": "issue:4752", "accept": false, - "reason": "CPU core utilization and SDXL half/float mismatch are unrelated." + "reason": "CPU core utilization complaint and SDXL dtype mismatch do not match." }, { "left": "issue:3566", "right": "issue:5689", "accept": false, - "reason": "Different symptoms and components: controlnet_reference unexpected output vs DPM++ residual noise in SDXL without refiner." + "reason": "ControlNet reference artifacts and residual noise in SDXL DPM++ are different symptoms and likely different causes." }, { "left": "issue:1344", "right": "issue:4042", "accept": false, - "reason": "One is a multi-GPU support bug in attention.py; the other is a feature request around reference-only plus inpainting/depth control." + "reason": "Single-GPU attention.py limitation is unrelated to combining reference-only with inpainting/depth control." }, { "left": "issue:4021", "right": "issue:5390", "accept": false, - "reason": "Both concern SDXL loading downloads, but one is from_single_file unnecessarily fetching CLIP while the other is local_files_only not preventing text-encoder download; same area, not clearly the same bug." + "reason": "Both involve downloads, but one is unnecessary CLIP download from_single_file and the other is local_files_only text-encoder fetch behavior." }, { "left": "issue:4080", "right": "issue:7962", "accept": false, - "reason": "Add SDXL inpainting pipeline vs add cross_attention_kwargs to instruct-pix2pix call method are separate feature requests." + "reason": "SDXL inpainting pipeline request and cross_attention_kwargs for instruct-pix2pix are separate feature asks." }, { "left": "issue:4171", "right": "issue:5246", "accept": false, - "reason": "Documentation/page rendering bug is unrelated to a missing CLIPImageProjection attribute error." + "reason": "Reference page rendering bug and missing CLIPImageProjection attribute are unrelated." }, { "left": "issue:4484", "right": "issue:5433", "accept": false, - "reason": "Both mention SDXL image quality, but one is Kohya LoRA weird results and the other is DPM++ artifacts without refiner; too different to merge." + "reason": "SDXL Kohya LoRA quality issues and DPM++ residual noise without refiner are not the same bug." }, { "left": "issue:5606", "right": "issue:7735", "accept": false, - "reason": "LoRA alpha handling and caption_column dataset parsing are unrelated." - }, - { - "left": "issue:6162", - "right": "issue:9495", - "accept": false, - "reason": "Both involve IPAdapter on SDXL, but one is a shape-mismatch bug and the other is PAG + IPAdapter not working; not the same concrete failure." - }, - { - "left": "issue:6452", - "right": "issue:7962", - "accept": false, - "reason": "Different feature requests for different pipelines and inputs." - }, - { - "left": "issue:8482", - "right": "issue:8813", - "accept": false, - "reason": "Stable Diffusion 3 model support and Differential Diffusion with Kolors are different model/pipeline requests." + "reason": "LoRA alpha handling and caption_column dataset validation are unrelated." } ] }, @@ -80164,20 +80182,16 @@ "nodes": [ "issue:10239", "issue:1061", - "issue:10671", "issue:12022", "issue:1477", "issue:1633", "issue:1831", - "issue:2179", "issue:2266", "issue:2276", "issue:2437", "issue:2577", "issue:2782", "issue:2843", - "issue:3029", - "issue:3481", "issue:3580", "issue:3612", "issue:4349", @@ -80191,9 +80205,10 @@ "issue:5760", "issue:5881", "issue:5897", - "issue:5997", "issue:6087", "issue:6132", + "issue:6162", + "issue:6452", "issue:6709", "issue:7009", "issue:7333", @@ -80201,19 +80216,24 @@ "issue:7415", "issue:7728", "issue:776", + "issue:7962", + "issue:8482", "issue:8708", + "issue:8813", "issue:8885", "issue:8939", "issue:9209", "issue:9398", + "issue:9495", "issue:9523", - "issue:9660", "issue:9719" ], "soft_pairs": [ + "issue:6162|issue:9495", + "issue:6452|issue:7962", + "issue:8482|issue:8813", "issue:8885|issue:9209", "issue:6709|issue:8939", - "issue:10671|issue:9660", "issue:2276|issue:3580", "issue:4414|issue:4468", "issue:5897|issue:7415", @@ -80232,182 +80252,180 @@ "issue:2843|issue:6087", "issue:4349|issue:7357", "issue:1061|issue:776", - "issue:1633|issue:3612", - "issue:2179|issue:3481", - "issue:3029|issue:5997" + "issue:1633|issue:3612" ], - "prepared_review_unit_hash": "2cfc073cb0ce3d525744f2e7721c55a44e9fe0ad882c240c9e484e717dd11c95", + "prepared_review_unit_hash": "cb746f5b39cacd37974468ec0c9fe099bd77d08fe7eab44e6e147721ae650b09", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 34041, - "estimated_input_tokens": 8511, - "estimated_eval_tokens": 17278 + "serialized_chars": 33864, + "estimated_input_tokens": 8466, + "estimated_eval_tokens": 17188 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a broad set of mostly unrelated Diffusers issues spanning pipelines, schedulers, training scripts, docs, and backend/runtime errors. The soft pairs look like similarity false-positives rather than true duplicates; all should be rejected.", - "confidence": 0.92, - "canonical_issue_reason": "No single underlying bug/change unifies these items; if a fallback representative is needed, issue 10671 is the broadest core-pipeline bug, but it does not actually match the rest of the set.", + "summary": "This cluster is not a real duplicate set: the issues span unrelated bugs and feature requests across pipelines, schedulers, training scripts, model loading, and docs. All soft pairs are superficial matches only, so every edge should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": "No strong canonical issue exists because the cluster is heterogeneous; if forced to anchor one, issue 10239 is the most concrete and recent offload-related report, but it does not match the rest well.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 10671 is the closest generic representative because it concerns a fundamental DiffusionPipeline device/dtype behavior, but the cluster is too heterogeneous for a strong canonical issue.", + "best_issue_reason": "Issue 10239 is the most actionable and specific report in the set, but only as a loose triage anchor rather than a true duplicate representative.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6162", + "right": "issue:9495", + "accept": false, + "reason": "Both mention SDXL/IP-Adapter, but one is a shape-mismatch bug and the other is a PAG incompatibility; different failure modes and likely different fixes." + }, + { + "left": "issue:6452", + "right": "issue:7962", + "accept": false, + "reason": "Different pipelines and requests: ControlNet inpaint masked_image_latents versus adding cross_attention_kwargs to Instruct-Pix2Pix." + }, + { + "left": "issue:8482", + "right": "issue:8813", + "accept": false, + "reason": "One is general SD3-medium support, the other is a Differential Diffusion/Kolors issue; unrelated model/pipeline problems." + }, { "left": "issue:8885", "right": "issue:9209", "accept": false, - "reason": "Different problems: dependency/install and modifier-token loading vs network/mirror connectivity while downloading a model." + "reason": "Custom Diffusion/clip-retrieval setup issues are unrelated to an hf-mirror connectivity failure." }, { "left": "issue:6709", "right": "issue:8939", "accept": false, - "reason": "Custom Diffusion concept-list training bug is unrelated to a tokenizer-loading error for SD3." - }, - { - "left": "issue:10671", - "right": "issue:9660", - "accept": false, - "reason": "Both are pipeline-core issues, but one is about dtype/device access determinism and the other about from_pretrained not setting eval mode; different bugs and fixes." + "reason": "Custom Diffusion training with concept lists is a different bug from SD3 tokenizer loading." }, { "left": "issue:2276", "right": "issue:3580", "accept": false, - "reason": "One is xformers installation failure in Colab; the other is a runtime TypeError when using XFormers." + "reason": "One is an install problem for xformers in Colab, the other is a runtime TypeError using XFormers; not the same bug." }, { "left": "issue:4414", "right": "issue:4468", "accept": false, - "reason": "Consistency Model sample request vs multi-GPU performance discrepancy; not the same bug or change." + "reason": "A consistency-model sample request and a multi-GPU performance complaint are unrelated." }, { "left": "issue:5897", "right": "issue:7415", "accept": false, - "reason": "Training device-mismatch in LoRA code vs AutoencoderKL bf16 behavior on TPU v4 are distinct backend/precision issues." + "reason": "LoRA device-mismatch during training and a bfloat16 TPU VAE issue are different code paths and symptoms." }, { "left": "issue:7728", "right": "issue:9523", "accept": false, - "reason": "Different training paths and failure modes: advanced SDXL DreamBooth accelerate error vs ControlNet training mapping error." + "reason": "Both are training-script errors, but they concern different scripts and distinct mapping/Accelerate failures." }, { "left": "issue:1477", "right": "issue:1831", "accept": false, - "reason": "Low-level CUDA cublas GEMM error is not the same issue as Dreambooth class-image precision mismatch." + "reason": "CUDA cublas invalid-value and DreamBooth class-image precision mismatch are unrelated bugs." }, { "left": "issue:2577", "right": "issue:5628", "accept": false, - "reason": "Different schedulers and symptoms: DDIMInverseScheduler not working vs LMS Karras-sigma denoising offset." + "reason": "Both involve schedulers, but DDIMInverseScheduler failure and LMS/Karras denoising offset are different concrete issues." }, { "left": "issue:5760", "right": "issue:7333", "accept": false, - "reason": "Unrelated application domains: video-to-audio synthesis vs image-to-video generation." + "reason": "Video-to-audio synthesis and image-to-video generation are different projects and tasks." }, { "left": "issue:8939", "right": "issue:9719", "accept": false, - "reason": "Tokenizer loading failure for SD3 is unrelated to disable_progress_bar being ignored during checkpoint loading." + "reason": "Tokenizer loading for SD3-medium versus progress-bar behavior while loading shards are unrelated." }, { "left": "issue:2266", "right": "issue:2782", "accept": false, - "reason": "Both mention sequential offloading, but one is a slowdown and the other is an error; likely different code paths and fixes." + "reason": "Both mention sequential offload, but one is slowdown and the other is a generic offloading error; not the same bug." }, { "left": "issue:2437", "right": "issue:6132", "accept": false, - "reason": "A new T2I-Adapter pipeline request is not the same as adding LOADABLE_CLASSES support." + "reason": "A new T2I-Adapter pipeline request is unrelated to adding LOADABLE_CLASSES." }, { "left": "issue:4677", "right": "issue:7009", "accept": false, - "reason": "Documentation example error in text-guided inpainting is unrelated to poor generation quality in TextToVideoSDPipeline." + "reason": "A documentation script error and poor TextToVideoSDPipeline results are different issues." }, { "left": "issue:4767", "right": "issue:5465", "accept": false, - "reason": "Scheduler prediction_type support and DiffEdit inpaint_strength behavior are separate pipeline/scheduler issues." + "reason": "EulerAncestralDiscreteScheduler sample prediction support and DiffEdit inpaint_strength limits are separate feature/bug reports." }, { "left": "issue:5630", "right": "issue:5881", "accept": false, - "reason": "Both concern output quality, but they target different pipelines and different root causes." + "reason": "Both are quality complaints, but they involve different pipelines/models and different underlying causes." }, { "left": "issue:8708", "right": "issue:9398", "accept": false, - "reason": "Missing SD3 dreambooth config field vs missing pooled_projection argument in a model forward call are different bugs." + "reason": "Missing SD3 dreambooth config and a missing pooled_projection argument in a different embedding module are not the same defect." }, { "left": "issue:10239", "right": "issue:12022", "accept": false, - "reason": "Sequential CPU offload support for HunyuanVideoPipeline is unrelated to flash_attention_3 interface incompatibility." + "reason": "HunyuanVideoPipeline sequential CPU offload support and flash_attention_3 interface compatibility are unrelated problems." }, { "left": "issue:2843", "right": "issue:6087", "accept": false, - "reason": "Checkpoint handling in multi-GPU training is not the same as failing to save/load LoraConfig correctly." + "reason": "Multi-GPU checkpoint saving and LoRAConfig save/load producing wrong lora_alpha are distinct concerns." }, { "left": "issue:4349", "right": "issue:7357", "accept": false, - "reason": "Missing text_encoder handling with prompt_embeds is a functional bug, while the other is an inference speed/performance issue." + "reason": "Missing text_encoder handling and slow inference with torch.compile/fuse_loras are different bugs." }, { "left": "issue:1061", "right": "issue:776", "accept": false, - "reason": "Docs update for textual inversion instructions is unrelated to img2img batch-size breakage." + "reason": "A textual inversion example instruction update is unrelated to an image-to-image batch-size bug." }, { "left": "issue:1633", "right": "issue:3612", "accept": false, - "reason": "Karras-sigmas support in a k-diffusion wrapper is a scheduler option request, not a ControlNet construction/preprocessor question." - }, - { - "left": "issue:2179", - "right": "issue:3481", - "accept": false, - "reason": "FlaxDDPMScheduler black images and DDIMScheduler step-dependent differences are distinct scheduler bugs." - }, - { - "left": "issue:3029", - "right": "issue:5997", - "accept": false, - "reason": "Custom resolution support for ControlNet is unrelated to an SDXL tutorial model-type instantiation error." + "reason": "Karras sigmas in the k_diffusion wrapper and ControlNet/reference_adain+attn guidance are unrelated feature requests." } ] }, @@ -80448,24 +80466,24 @@ "issue:1304", "issue:13107", "issue:205", + "issue:2179", "issue:2437", "issue:2480", "issue:2737", - "issue:2763", "issue:2786", - "issue:3043", + "issue:3029", "issue:307", "issue:3455", + "issue:3481", "issue:3958", "issue:4111", "issue:4116", "issue:4264", - "issue:4282", "issue:459", "issue:5313", "issue:5523", "issue:5731", - "issue:6132", + "issue:5997", "issue:6313", "issue:6320", "issue:6548", @@ -80481,6 +80499,8 @@ "issue:9216" ], "soft_pairs": [ + "issue:2179|issue:3481", + "issue:3029|issue:5997", "issue:307|issue:459", "issue:4111|issue:4116", "issue:4264|issue:5731", @@ -80502,188 +80522,186 @@ "issue:1141|issue:2737", "issue:1273|issue:3455", "issue:1304|issue:205", - "issue:2437|issue:6313", - "issue:2763|issue:4282", - "issue:3043|issue:6132" + "issue:2437|issue:6313" ], - "prepared_review_unit_hash": "d6043f6acace690aae3ca5e6641f48604762ebfb95f9e6ca7f2f56c6f498af38", + "prepared_review_unit_hash": "026aa9258adac1c8e8440ca03c720a904f6dbd6117045feafe392211603a125a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33849, - "estimated_input_tokens": 8463, - "estimated_eval_tokens": 17182 + "serialized_chars": 34050, + "estimated_input_tokens": 8513, + "estimated_eval_tokens": 17282 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are false positives from broad keyword overlap rather than the same underlying bug/change. Only the ControlNet single-file loading reports look like a real duplicate cluster.", - "confidence": 0.87, - "canonical_issue_reason": "issue:8614 is the broadest description of the shared failure mode: ControlNet single-file loading breaks because config.json is missing. issue:11178 is a specific MistoLine example of that same problem.", + "summary": "Most items are unrelated feature requests or bug reports across different pipelines. The only near-duplicate pair is the ControlNet single-file loading/config.json reports.", + "confidence": 0.79, + "canonical_issue_reason": "issue:8614 is the broadest and most general statement of the ControlNet single-file loading problem, while issue:11178 is a specific symptom for one model.", "canonical_pr_reason": null, - "best_issue_reason": "issue:8614 is the best duplicate sink because it states the generic loader/config.json problem without narrowing to one model, making it the most reusable canonical issue.", + "best_issue_reason": "issue:8614 best represents the shared underlying bug because it describes the missing config.json / single-file ControlNet loading problem at the right abstraction level.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2179", + "right": "issue:3481", + "accept": false, + "reason": "Different problems: FlaxDDPMScheduler black images vs DDIM step-count sensitivity and inconsistent outputs." + }, + { + "left": "issue:3029", + "right": "issue:5997", + "accept": false, + "reason": "Both involve ControlNet tutorials, but one is custom resolution and the other is a model-type mismatch; not the same bug." + }, { "left": "issue:307", "right": "issue:459", "accept": false, - "reason": "Both are broad diffusion ideas, but one is a pipeline-merging request and the other is incremental diffusion; not the same change." + "reason": "One is about merging Stable Diffusion pipelines; the other is an incremental diffusion feature request." }, { "left": "issue:4111", "right": "issue:4116", "accept": false, - "reason": "Same pipeline family, but one is a VRAM/load issue and the other is an embeddings-input bug; different failures." + "reason": "Same pipeline family, but different issues: VRAM/memory loading vs embeddings not working." }, { "left": "issue:4264", "right": "issue:5731", "accept": false, - "reason": "SDXL example-image omissions and AutoencoderKL logvar mismatch are unrelated problems." + "reason": "Missing example images is unrelated to the AutoencoderKL trainable logvar implementation issue." }, { "left": "issue:6616", "right": "issue:7380", "accept": false, - "reason": "Classifier-guidance support and one-step diffusion distillation are distinct features, not the same implementation." + "reason": "Classifier guidance support and one-step diffusion distillation are different changes/problems." }, { "left": "issue:6639", "right": "issue:7432", "accept": false, - "reason": "LoRA loading inconsistency in SDXL and matching WebUI output are different concrete issues." + "reason": "Both concern different outputs, but one is LoRA loading order in SDXL and the other is general inference differences vs WebUI." }, { "left": "issue:11779", "right": "issue:12574", "accept": false, - "reason": "Both are Wan/VACE-related, but one is mask padding and the other is missing control-argument handling; different code paths." + "reason": "Both are Wan/VACE-related, but one is mask padding and the other is missing control-argument handling; distinct failure modes." }, { "left": "issue:2480", "right": "issue:5523", "accept": false, - "reason": "Checkpoint directory creation and inconsistent run results are unrelated." + "reason": "Dreambooth checkpoint directory bug is unrelated to generic run-to-run inconsistency." }, { "left": "issue:10566", "right": "issue:12071", "accept": false, - "reason": "Different transformer models and different behaviors; no shared bug." + "reason": "Different models and code paths: CogVideoX forward work vs flux kontext single-block forward behavior." }, { "left": "issue:10673", "right": "issue:12025", "accept": false, - "reason": "AuraFlow reshape failure and Cosmos VAE invalid API call are unrelated." + "reason": "AuraFlow shape mismatch and Cosmos VAE encoder API error are unrelated." }, { "left": "issue:7389", "right": "issue:7640", "accept": false, - "reason": "Both mention SDXL/from_single_file errors, but they appear to be different shape/loading failures and not one concrete bug." + "reason": "Both mention SDXL from_single_file, but one is a generic load error and the other is a dimensions-specific runtime error." }, { "left": "issue:11178", "right": "issue:8614", "accept": true, - "reason": "Same underlying ControlNet single-file loading failure around missing config.json; one is a specific model instance of the broader bug." + "reason": "Both describe the same ControlNet single-file loading failure centered on missing config.json / from_single_file support; 8614 is the broader root report." }, { "left": "issue:2786", "right": "issue:6548", "accept": false, - "reason": "Both are img2img input-handling bugs, but one is numpy/transformers compatibility and the other is RGBA image misclassified as latents." + "reason": "NumPy input incompatibility with transformers-4.27 is different from RGBA image handling in prepare_latents." }, { "left": "issue:11627", "right": "issue:8539", "accept": false, - "reason": "SD3 IP-Adapter support and SD3 cross-attention issues are different topics." + "reason": "StableDiffusion3 with IP-Adapter and SD3 cross-attention issues are separate model-integration problems." }, { "left": "issue:3958", "right": "issue:5313", "accept": false, - "reason": "LoRA CPU offload behavior and non-revertible fuse are separate LoRA-related issues." + "reason": "Different LoRA-related problems: cpu offload interaction vs non-revertible fuse behavior." }, { "left": "issue:12012", "right": "issue:12536", "accept": false, - "reason": "Wan first-block cache context setup and context-parallel sharding size constraints are different bugs." + "reason": "Both Wan-related, but one is context/caching state management and the other is context-parallel sharding size divisibility." }, { "left": "issue:6320", "right": "issue:8956", "accept": false, - "reason": "Emu2 remote-model resolution and OVStableDiffusionXLPipeline loading are unrelated." + "reason": "Emu2 remote model lookup and OVStableDiffusionXL model loading are unrelated." }, { "left": "issue:10050", "right": "issue:9216", "accept": false, - "reason": "An img2img KDiffusion equivalent and a multi-model pipeline are different feature requests." + "reason": "Img2img KDiffusion equivalent question is unrelated to a custom multi-model pipeline proposal." }, { "left": "issue:11010", "right": "issue:13107", "accept": false, - "reason": "Both mention Flux, but one is architecture support for Chroma and the other is LoRA missing-key errors." + "reason": "Flux Chroma architecture support and Flux Klein LoRA missing-key errors are different issues." }, { "left": "issue:1141", "right": "issue:2737", "accept": false, - "reason": "Local-file support for custom pipelines is unrelated to CLIP guidance in a scheduler." + "reason": "Local files in custom pipelines is unrelated to CLIP guidance support for a scheduler." }, { "left": "issue:1273", "right": "issue:3455", "accept": false, - "reason": "Dreambooth loss calculation and pretrained instruct-pix2pix training are not the same issue." + "reason": "Dreambooth mean-loss question and using pretrained instruct-pix2pix in its own training are not duplicates." }, { "left": "issue:1304", "right": "issue:205", "accept": false, - "reason": "Training an img2img model and adding interpolation example scripts are different requests." + "reason": "Training img2img stable diffusion and adding interpolation example scripts are unrelated." }, { "left": "issue:2437", "right": "issue:6313", "accept": false, - "reason": "T2I-Adapter pipeline support and null-text inversion pipeline implementation are separate features." - }, - { - "left": "issue:2763", - "right": "issue:4282", - "accept": false, - "reason": "ControlNet VRAM regression with cudnn.benchmark and Multi-ControlNet support are different concerns." - }, - { - "left": "issue:3043", - "right": "issue:6132", - "accept": false, - "reason": "ODISE support and LOADABLE_CLASSES support are unrelated feature requests." + "reason": "T2I-Adapter pipeline request is unrelated to Null-Text Inversion pipeline implementation." } ] }, "evaluator_result": { "accept": true, - "feedback": "The verdicts are conservative and grounded in the packet. The only accepted duplicate pair, issue:11178 and issue:8614, is well supported by the titles and descriptions around ControlNet single-file loading and missing config.json. The remaining pairs are rejected cautiously with reasonable distinctions rather than overstating evidence. The summary also stays within the evidence by framing the rest as keyword overlap." + "feedback": "Grounded overall. The canonical issue choice is supported by the packet: issue:8614 is the broader ControlNet single-file/config.json report, and issue:11178 is a narrower symptom of the same loading problem. The other soft-edge verdicts are phrased conservatively as non-duplicates and do not overclaim beyond the titles/body summaries shown." }, "evaluator_used": true, "retried": false, @@ -80704,15 +80722,15 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10178", "issue:10191", "issue:10311", "issue:10446", "issue:10447", "issue:10823", "issue:10963", - "issue:2722", + "issue:2763", "issue:2961", + "issue:3043", "issue:3065", "issue:3191", "issue:3223", @@ -80721,6 +80739,7 @@ "issue:3390", "issue:3604", "issue:3960", + "issue:4282", "issue:4665", "issue:4756", "issue:4801", @@ -80734,8 +80753,8 @@ "issue:5432", "issue:5654", "issue:5873", + "issue:6132", "issue:6157", - "issue:6176", "issue:6354", "issue:6359", "issue:656", @@ -80744,7 +80763,6 @@ "issue:7079", "issue:770", "issue:8241", - "issue:8405", "issue:8824", "issue:8939", "issue:9047", @@ -80754,6 +80772,8 @@ "issue:9926" ], "soft_pairs": [ + "issue:2763|issue:4282", + "issue:3043|issue:6132", "issue:3065|issue:7043", "issue:4756|issue:5432", "issue:6354|issue:9926", @@ -80762,7 +80782,6 @@ "issue:10191|issue:10446", "issue:5399|issue:8241", "issue:10447|issue:10823", - "issue:2722|issue:6176", "issue:10311|issue:10963", "issue:2961|issue:3604", "issue:3191|issue:3960", @@ -80776,181 +80795,180 @@ "issue:5112|issue:6996", "issue:514|issue:656", "issue:5873|issue:9424", - "issue:8824|issue:8939", - "issue:10178|issue:8405" + "issue:8824|issue:8939" ], - "prepared_review_unit_hash": "84cbfcbb12e549a2c7ff6d5613cf4113659f9d3ab9aacc47a2e952be521e8ebc", + "prepared_review_unit_hash": "857a934ef5d37ecbb58948ec81ffed5f70b50ced45f8d3a4b23d01a82e1c7007", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34172, - "estimated_input_tokens": 8543, - "estimated_eval_tokens": 17342 + "serialized_chars": 34099, + "estimated_input_tokens": 8525, + "estimated_eval_tokens": 17306 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: the items span unrelated training, inference, conversion, LoRA, scheduler, and device-placement topics. None of the proposed soft pairs look like the same underlying bug or change, so I would not collapse them into a single canonical artifact.", - "confidence": 0.95, - "canonical_issue_reason": null, + "summary": "No duplicate cluster here: the items span unrelated training, import, pipeline, conversion, and scheduler issues. The soft pairs share only broad subsystem similarity, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": "If one issue had to anchor the set, #10191 is the clearest self-contained bug report with a specific expected-vs-actual regression (GPU device selection during training), but it does not represent the rest of the heterogeneous cluster.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "#10191 is the best single issue to keep as a representative because it is concrete, well-scoped, and describes a clear reproducible training-device bug.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2763", + "right": "issue:4282", + "accept": false, + "reason": "ControlNet VRAM regression vs a Multi-ControlNet feature request; different problem types." + }, + { + "left": "issue:3043", + "right": "issue:6132", + "accept": false, + "reason": "ODISE addition vs LOADABLE_CLASSES support; unrelated features." + }, { "left": "issue:3065", "right": "issue:7043", "accept": false, - "reason": "Both involve DreamBooth/LoRA training, but one is a feature request for text-encoder LoRA tuning and the other is about custom captions; related area, not the same bug or change." + "reason": "DreamBooth LoRA option request vs custom captions support; related area, but different change requests." }, { "left": "issue:4756", "right": "issue:5432", "accept": false, - "reason": "Both mention SDXL training scripts, but one asks for a training example/style support and the other reports blank validation images; different problems." + "reason": "Both are DreamBooth/SDXL training-related, but one is about adding a training example and the other is a validation-image bug." }, { "left": "issue:6354", "right": "issue:9926", "accept": false, - "reason": "ControlNet LoRA support and GGUF quantize/speed support are unrelated feature requests." + "reason": "ControlNet LoRA support vs GGUF quantize/speed-up support; unrelated enhancements." }, { "left": "issue:6359", "right": "issue:7079", "accept": false, - "reason": "Scheduler v-prediction velocity support versus Transformer2DModel patch/norm_type tensor error are different code paths and failures." + "reason": "Scheduler velocity/add_noise bug vs Transformer2DModel patch/norm_type error; different code paths." }, { "left": "issue:9047", "right": "issue:9178", "accept": false, - "reason": "Both are Flux-related, but one is an MPS runtime issue and the other is a LoRA PEFT target-module error; not the same defect." + "reason": "MPS support for Flux vs FLUX LoRA PEFT target-module errors; distinct issues." }, { "left": "issue:10191", "right": "issue:10446", "accept": false, - "reason": "GPU selection being ignored during training is not the same as FluxControlNetPipeline model-sharding incompatibility." + "reason": "GPU device selection during training vs FluxControlNetPipeline multi-GPU/model-sharding incompatibility; not the same bug." }, { "left": "issue:5399", "right": "issue:8241", "accept": false, - "reason": "Different symptoms and likely different modules: dtype mismatch in custom inpainting versus LEDITS++ model size anomaly." + "reason": "Custom inpainting dtype mismatch vs LEDITS++ model size question; unrelated." }, { "left": "issue:10447", "right": "issue:10823", "accept": false, - "reason": "Both touch conversion/training workflows, but one is about split UNet checkpoint files and the other about missing inference files from fine-tuning; not a single bug." - }, - { - "left": "issue:2722", - "right": "issue:6176", - "accept": false, - "reason": "ControlNet conversion mismatch and loading LoRA weights after DreamBooth training are distinct issues." + "reason": "Converter splitting UNet into two files vs finetuning script output files for inference; similar output packaging theme, but different scripts and failures." }, { "left": "issue:10311", "right": "issue:10963", "accept": false, - "reason": "Both are import errors, but they concern different symbols and subsystems; no concrete shared failure mode." + "reason": "Dreambooth import error vs AutoencoderKLWan import error; both are import issues, but for different symbols and modules." }, { "left": "issue:2961", "right": "issue:3604", "accept": false, - "reason": "StableUnCLIP training example request versus loss not decreasing during finetuning are different concerns." + "reason": "Request for a StableUnCLIP training example vs training loss not decreasing; one is documentation/example, the other a training-quality bug." }, { "left": "issue:3191", "right": "issue:3960", "accept": false, - "reason": "DreamBooth+LoRA broken and HeaderTooLarge for LoRA are unrelated issues." + "reason": "DreamBooth+LoRA broken vs HeaderTooLarge for LoRA; different failure modes." }, { "left": "issue:3223", "right": "issue:6157", "accept": false, - "reason": "CUDA RAM not clearing after moving pipeline to CUDA is unrelated to IP-Adapter support in ControlNetImg2ImgPipeline." + "reason": "Memory not cleared after moving pipeline to CUDA vs IP-Adapter not working in ControlNetImg2ImgPipeline; unrelated pipeline bugs." }, { "left": "issue:3325", "right": "issue:4801", "accept": false, - "reason": "Train_dreambooth CUDA invalid argument and mixed-precision bug in train_unconditional.py are separate failures." + "reason": "CUDA invalid-argument in train_dreambooth.py vs mixed-precision bug in train_unconditional.py; different scripts and errors." }, { "left": "issue:3377", "right": "issue:5654", "accept": false, - "reason": "Both discuss SNR weighting, but one is a general question and the other is a specific feature request for v-prediction examples; not the same actionable bug/change." + "reason": "General question about SNR weighting vs a request to add Min-SNR weight for v-prediction; related topic, but not the same underlying issue." }, { "left": "issue:3390", "right": "issue:4665", "accept": false, - "reason": "Local safetensors loading and Windows use_safetensors behavior with alternative bin files are different loading issues." + "reason": "Local safetensors loading question vs Windows use_safetensors fallback bug; different scenarios." }, { "left": "issue:4969", "right": "issue:5061", "accept": false, - "reason": "ImportError for CrossAttention versus argument error in the conversion script are unrelated." + "reason": "CrossAttention import error vs config_files argument error; separate API breakages." }, { "left": "issue:5025", "right": "issue:770", "accept": false, - "reason": "randn_tensor import error and Colab training startup failure are not the same underlying problem." + "reason": "randn_tensor import error vs Colab training startup failure; unrelated." }, { "left": "issue:5079", "right": "issue:9397", "accept": false, - "reason": "LoRA loading support for Lyrocis and Hyper LoRA scheduler support are different feature requests." + "reason": "LoRA loading support request vs SGMUniform scheduler support for Hyper LoRAs; both mention LoRA, but not the same change." }, { "left": "issue:5112", "right": "issue:6996", "accept": false, - "reason": "ControlNetImg2Img tensor-size float error and NoneType added_cond_kwargs error are separate runtime issues." + "reason": "Size/type error in SDXL ControlNet img2img vs NoneType added_cond_kwargs error; different runtime paths." }, { "left": "issue:514", "right": "issue:656", "accept": false, - "reason": "Attention slicing error and DreamBooth 16GB GPU training error are not the same bug." + "reason": "Attention slicing error vs 16GB DreamBooth training error; too broad and not the same bug." }, { "left": "issue:5873", "right": "issue:9424", "accept": false, - "reason": "ONNX export failure and a broken notebook with the latest version are unrelated." + "reason": "ONNX export failure vs broken DreamBooth notebook; unrelated." }, { "left": "issue:8824", "right": "issue:8939", "accept": false, - "reason": "Custom instruct-pix2pix dataset setup and tokenizer loading for SD3 are unrelated." - }, - { - "left": "issue:10178", - "right": "issue:8405", - "accept": false, - "reason": "Both mention tensor/dimension mismatches, but one is a UNet 3D blocks dimension-size mismatch and the other is a multiple-ControlNet tensor-size error; too different to merge." + "reason": "Custom dataset guidance for instruct-pix2pix vs tokenizer load failure for SD3; different topics and failure modes." } ] }, @@ -80969,11 +80987,12 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10178", "issue:1070", "issue:11000", "issue:11006", @@ -80991,7 +81010,6 @@ "issue:3032", "issue:3103", "issue:3287", - "issue:3468", "issue:3551", "issue:3767", "issue:3880", @@ -81008,12 +81026,12 @@ "issue:5777", "issue:5890", "issue:617", - "issue:6321", "issue:6481", "issue:671", "issue:7228", "issue:7264", "issue:7904", + "issue:8405", "issue:8535", "issue:8556", "issue:8809", @@ -81024,6 +81042,7 @@ "issue:9962" ], "soft_pairs": [ + "issue:10178|issue:8405", "issue:12895|issue:2392", "issue:3103|issue:917", "issue:4964|issue:671", @@ -81047,193 +81066,192 @@ "issue:1746|issue:2444", "issue:1821|issue:3032", "issue:1821|issue:987", - "issue:5646|issue:5685", - "issue:3468|issue:6321" + "issue:5646|issue:5685" ], - "prepared_review_unit_hash": "d607374e04700579993f0c55c1decde2b5f7fa44931de637eb67a161d41493fe", + "prepared_review_unit_hash": "13487e560d80b1a96e0631a63fc2e0b266ebcf0878e61e78e21778d3ee452e73", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34296, - "estimated_input_tokens": 8574, - "estimated_eval_tokens": 17404 + "serialized_chars": 34266, + "estimated_input_tokens": 8567, + "estimated_eval_tokens": 17390 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous: the items span unrelated bugs, feature requests, and training/model-loading issues. The only near-duplicate pair is the SD3 xformers enablement report. No PRs were provided.", - "confidence": 0.91, - "canonical_issue_reason": "Issue 8535 is the best canonical for the only true duplicate-like theme here: it directly names the SD3 xformers enablement problem and is the more specific representative of that pair.", + "summary": "This cluster is mostly heterogeneous and does not look like a set of true duplicates. The only strong duplicate-like pair is the two xFormers memory-efficient-attention issues; the rest are related only at a broad subsystem level or are clearly different bugs/features.", + "confidence": 0.83, + "canonical_issue_reason": "issue:8535 is the best canonical issue because it is the clearest, most specific description of the only near-duplicate bug/change in the cluster: xFormers memory-efficient attention enablement for SD3.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 11006 is the strongest standalone issue overall: it has a concrete Wan 2.1 I2V failure mode, a specific pipeline, and a clear bug report that looks actionable on its own.", + "best_issue_reason": "issue:8535 is the strongest representative issue in this set: concrete, code-path-specific, and paired with a very likely duplicate (issue:9946).", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10178", + "right": "issue:8405", + "accept": false, + "reason": "Both mention size mismatches, but one is a 3D UNet block issue and the other is a multiple-ControlNet tensor sizing problem; different code paths." + }, { "left": "issue:12895", "right": "issue:2392", "accept": false, - "reason": "Both are prompt/embed-related runtime errors, but they affect different pipelines and different failure modes; not the same underlying bug." + "reason": "Both involve None/prompt handling, but they occur in different pipelines and under different APIs; not the same bug." }, { "left": "issue:3103", "right": "issue:917", "accept": false, - "reason": "Both mention inpainting, but one is about height/width handling and the other about ONNX inpainting errors; too different to be duplicates." + "reason": "Both are inpainting-related, but one is about height/width handling and the other is an ONNX inpainting error; too different." }, { "left": "issue:4964", "right": "issue:671", "accept": false, - "reason": "Both concern model conversion, but one is about SDXL conversion generally and the other about original Stable Diffusion conversion outputs not matching; different bugs." + "reason": "Both are conversion-related, but they target different model families and conversion flows; not a single duplicate bug." }, { "left": "issue:11000", "right": "issue:9962", "accept": false, - "reason": "Both are argument/signature errors, but they involve different APIs and different missing kwargs; not the same code-path problem." + "reason": "Missing shift argument vs unexpected use_cuda_graph keyword are unrelated API mismatches." }, { "left": "issue:11470", "right": "issue:13124", "accept": false, - "reason": "Training script issues, but one is about tensorboard log images and the other about FP16 gradient unscale during validation; unrelated failure modes." + "reason": "These are separate training-script problems with different symptoms and triggers." }, { "left": "issue:2737", "right": "issue:3287", "accept": false, - "reason": "Both are scheduler-related feature requests, but one is CLIP guidance support and the other is improved DDPM support; not the same change." + "reason": "A scheduler support request and a DDPM support request are different feature asks, not the same change." }, { "left": "issue:4589", "right": "issue:5336", "accept": false, - "reason": "Both touch ControlNet img2img/inpaint, but one is about XL pipeline support and the other about a missing strength parameter; different requests." + "reason": "Both concern ControlNet img2img/inpaint, but one is about missing XL pipeline variants and the other about adding a strength parameter; different fixes." }, { "left": "issue:4648", "right": "issue:5777", "accept": false, - "reason": "One is a missing config.json load error, the other is a DreamBooth training failure; no shared underlying bug." + "reason": "Missing config.json and a DreamBooth PixArt training error are unrelated." }, { "left": "issue:7904", "right": "issue:8535", "accept": false, - "reason": "Both mention scheduler/xformers-adjacent behavior, but one is sigmas mutating after forward and the other is xformers enablement in SD3; different problems." + "reason": "Scheduler sigmas mutating after forward is unrelated to enabling xFormers memory-efficient attention." }, { "left": "issue:4418", "right": "issue:5890", "accept": false, - "reason": "One is a denoising_start behavior bug in SDXL inpaint; the other is a tokenizer length warning during text-to-image inference; unrelated." + "reason": "Inpaint denoising_start behavior and a token-length warning during text-to-image training are not the same bug." }, { "left": "issue:4446", "right": "issue:4871", "accept": false, - "reason": "Both involve from_single_file/custom loading, but one is about a custom text encoder and the other about how to run SDXL from_single_file; not the same defect." + "reason": "One is about custom text_encoder support in from_single_file; the other is a general how-to question for from_single_file." }, { "left": "issue:2579", "right": "issue:3551", "accept": false, - "reason": "Both mention xFormers, but one is a possible misuse/bug report and the other is LoRA disabling xFormers; different code paths." + "reason": "Generic xFormers misuse/bug speculation is not the same as LoRA disabling xFormers." }, { "left": "issue:6481", "right": "issue:9674", "accept": false, - "reason": "Training batch-size behavior in DreamBooth is unrelated to SD3 ControlNet training quality." + "reason": "Batch-size training failure and SD3 ControlNet quality issues are different problems." }, { "left": "issue:7228", "right": "issue:8556", "accept": false, - "reason": "Both are load-time errors, but one concerns traced model dtype and the other meta tensor copying in SD3 from_single_file; different causes." + "reason": "Both are loading-time errors, but one is a missing dtype attribute on traced models and the other is a meta-tensor copy failure in SD3 from_single_file." }, { "left": "issue:1070", "right": "issue:3880", "accept": false, - "reason": "Both are inpainting complaints, but one is a generic new-version bug report and the other is specifically degraded unmasked areas; not clearly the same bug." + "reason": "Both are inpainting complaints, but the titles do not point to the same concrete bug; one is generic and the other is specifically about unmasked-area degradation." }, { "left": "issue:11006", "right": "issue:8809", "accept": false, - "reason": "Wan I2V broken output is not the same issue as feedforward outputs changing with sequence length; different model behavior." + "reason": "A Wan I2V video-output bug with quantized transformer is unrelated to a feedforward sequence-length inconsistency." }, { "left": "issue:1817", "right": "issue:1944", "accept": false, - "reason": "DreamBooth scheduler argument error versus random generation results; one is a crash, the other is a quality/conditioning complaint." + "reason": "DreamBooth scheduler argument handling and random DreamBooth outputs are different issues." }, { "left": "issue:3767", "right": "issue:617", "accept": false, - "reason": "FrozenDict mutability and NNPACK unsupported hardware are unrelated." + "reason": "FrozenDict mutability and NNPACK hardware support are unrelated." }, { "left": "issue:4912", "right": "issue:7264", "accept": false, - "reason": "Both are about custom models, but one is SDXL checkpoint conversion and the other is serializing/loading custom transformers in DiffusionPipeline; not the same bug." + "reason": "Model conversion to diffusers and serialization/loading of custom transformers models are different code paths." }, { "left": "issue:8535", "right": "issue:9946", "accept": true, - "reason": "These are effectively the same xformers memory-efficient-attention issue, both centered on enabling that functionality in the pipeline, with nearly identical wording." + "reason": "These are essentially the same xFormers memory-efficient-attention issue; the titles describe the same concrete path and could be merged." }, { "left": "issue:1746", "right": "issue:2444", "accept": false, - "reason": "Both involve training/inference scripts, but the reported errors and scripts are different and do not point to the same defect." + "reason": "A Flax pipeline error and a DreamBooth inpaint script failure are unrelated." }, { "left": "issue:1821", "right": "issue:3032", "accept": false, - "reason": "Negative prompt ineffectiveness in inpainting and a missing from_pretrained attribute are unrelated." + "reason": "Negative_prompt ineffectiveness in inpainting and a missing from_pretrained attribute are unrelated." }, { "left": "issue:1821", "right": "issue:987", "accept": false, - "reason": "Inpaint negative-prompt behavior and speech-to-image green output are different issues." + "reason": "The inpainting negative_prompt complaint and a speech-to-image green-image problem are different bugs." }, { "left": "issue:5646", "right": "issue:5685", "accept": false, - "reason": "Both mention schedulers, but one is a specific stability comparison on a model and the other is a vague potential scheduler issue; not enough to treat as one bug." - }, - { - "left": "issue:3468", - "right": "issue:6321", - "accept": false, - "reason": "One is missing model_index.json after DreamBooth, the other is clip-vit-large-patch14 detection; different loading/training problems." + "reason": "Both discuss schedulers generally, but they are not the same concrete scheduler bug or fix." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative on the soft-edge pairs. The single accepted duplicate pair (issue:8535/issue:9946) is well supported by the titles, and the rest of the verdicts do not overstate similarity. No PR-pair concerns apply because no PRs were provided." + "feedback": "The summary is grounded in the packet and stays conservative. The only clearly supported duplicate-like pair is issue:8535 / issue:9946, and the other soft-edge rejections are appropriately cautious given the titles and issue contexts." }, "evaluator_used": true, "retried": false, @@ -81249,8 +81267,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -81265,6 +81283,7 @@ "issue:2851", "issue:3097", "issue:3405", + "issue:3468", "issue:4232", "issue:4376", "issue:4480", @@ -81283,6 +81302,7 @@ "issue:6090", "issue:6109", "issue:6280", + "issue:6321", "issue:6639", "issue:6686", "issue:6807", @@ -81290,9 +81310,7 @@ "issue:7415", "issue:7562", "issue:7589", - "issue:7690", "issue:7891", - "issue:7939", "issue:8526", "issue:898", "issue:9106", @@ -81303,6 +81321,7 @@ "issue:9911" ], "soft_pairs": [ + "issue:3468|issue:6321", "issue:11574|issue:11833", "issue:1434|issue:3097", "issue:2722|issue:5917", @@ -81325,181 +81344,180 @@ "issue:10302|issue:9690", "issue:3405|issue:6109", "issue:1249|issue:5509", - "issue:2851|issue:898", - "issue:7690|issue:7939" + "issue:2851|issue:898" ], - "prepared_review_unit_hash": "aac22b934a476bf1f2338ec1b29e0f41923269109f89b4deb5e244990fcb4733", + "prepared_review_unit_hash": "ee7d17e6ec0c2dc038e633e2557e5f12cb6b9cfebea683175cf1313480582345", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33714, - "estimated_input_tokens": 8429, - "estimated_eval_tokens": 17114 + "serialized_chars": 33727, + "estimated_input_tokens": 8432, + "estimated_eval_tokens": 17120 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose collection of unrelated diffusers issues that only share superficial wording around pipelines, ControlNet, LoRA, or device/offload behavior. None of the soft-edge pairs looks like the same underlying bug or change, so they should stay separate.", - "confidence": 0.91, - "canonical_issue_reason": null, + "summary": "This is a heterogeneous set of unrelated issues rather than a true duplicate cluster. All proposed soft pairs are different concrete bugs or feature requests and should be kept separate.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue is justified: the artifacts span unrelated topics (LoRA loading, ControlNet, TPU/MPS/FP8, conversion, training, and pipeline API gaps).", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No best representative issue stands out because the cluster is not a single underlying bug family.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3468", + "right": "issue:6321", + "accept": false, + "reason": "Different problems: missing `model_index.json` for DreamBooth vs CLIP model not being detected." + }, { "left": "issue:11574", "right": "issue:11833", "accept": false, - "reason": "Both mention LoRA-related errors, but one is a VAE checkpoint KeyError (`vae.encoder.conv_out.conv.weight`) and the other is `set_lora_device` reporting KeyError after LoRA load. Different code paths and failure modes." + "reason": "Both mention KeyError, but one is a VAE weight lookup failure and the other is a LoRA device-setting error." }, { "left": "issue:1434", "right": "issue:3097", "accept": false, - "reason": "OOM during upscaling is a memory-capacity problem, while `load_textual_inversion` doing nothing is an adapter-loading behavior bug. Unrelated symptoms and subsystems." + "reason": "OOM during upscaling is unrelated to `load_textual_inversion` doing nothing." }, { "left": "issue:2722", "right": "issue:5917", "accept": false, - "reason": "ControlNet conversion mismatch and `convert_ldm_unet_checkpoint` failing after a version upgrade are both conversion-related, but they concern different converters and different model artifacts." + "reason": "Both involve conversion code, but one is ControlNet UNet mismatch after conversion and the other is `convert_ldm_unet_checkpoint` breaking after an upgrade." }, { "left": "issue:4232", "right": "issue:5866", "accept": false, - "reason": "Instruct-pix2pix with controlnet condition vs AnimateDiff + ControlNet are different pipelines and different requested behaviors. Not the same bug." + "reason": "Different feature requests/bugs: instruct-pix2pix with ControlNet vs AnimateDiff + ControlNet." }, { "left": "issue:4376", "right": "issue:5753", "accept": false, - "reason": "Both involve `load_textual_inversion`/LoRA-style loading, but one is an SDXL crash on textual inversion and the other is `load_lora_weights` failing on SDXL ControlNet. Different APIs and error classes." + "reason": "Both are SDXL/adapter-loading related, but one is textual inversion crashing and the other is LoRA loading failing in ControlNet." }, { "left": "issue:5515", "right": "issue:9332", "accept": false, - "reason": "LCM pipeline scheduler initialization failure is unrelated to `StableDiffusionPipeline.from_pretrained()` loading the wrong weights. No shared concrete code-path." + "reason": "LCM scheduler initialization failure is not the same as `from_pretrained()` loading identical weights." }, { "left": "issue:4530", "right": "issue:6686", "accept": false, - "reason": "SDXL refiner image artifacts are an output-quality issue, while Dynamo compilation save/load losing performance is a compilation/cache behavior issue. Different problems." + "reason": "Image artifacts in SDXL refiner are unrelated to Dynamo compile performance after save/load." }, { "left": "issue:4826", "right": "issue:7562", "accept": false, - "reason": "Multi-ControlNet errors in ControlNet Reference and `enable_model_cpu_offload` failing on SDXL ControlNet Img2Img are different feature combinations and different failure points." + "reason": "Both mention MultiControlNet, but the failures are different: reference ControlNet errors vs CPU offload failure in an SDXL img2img pipeline." }, { "left": "issue:10154", "right": "issue:6639", "accept": false, - "reason": "`load_lora_weights` attribute error vs LoRA loading producing different outputs than two-step loading. One is a crash, the other is a correctness mismatch." + "reason": "One is an attribute error from `load_lora_weights`; the other is output differences between loading LoRA in one step vs two steps." }, { "left": "issue:4646", "right": "issue:7240", "accept": false, - "reason": "ControlNet-to-ONNX conversion failure and low-quality ImageNet conditional LDM conversion output are both conversion topics, but they are not the same bug." + "reason": "ControlNet ONNX conversion failure is unrelated to low-quality output from conditional LDM conversion." }, { "left": "issue:5152", "right": "issue:9106", "accept": false, - "reason": "W\u00fcrstchen alpha bug and PixArt-Alpha `time_step=1` attribute error are different models and different failure modes." + "reason": "W\u00fcrstchen alpha bug and PixArt-alpha attribute error are different models and different failures." }, { "left": "issue:4649", "right": "issue:7589", "accept": false, - "reason": "`text_embeds` NoneType error in an SDXL/added-conditions path is unrelated to GLIGEN layout-to-image synthesis not working." + "reason": "`NoneType` membership error in `added_cond_kwargs` is not the same as GLIGEN not working." }, { "left": "issue:6280", "right": "issue:9846", "accept": false, - "reason": "Missing `from_single_file` on SDXL ControlNet Img2Img and missing `from_single_file` on FluxControlNetModel are the same API gap in a broad sense, but they are separate model families and separate feature requests; too broad to merge as one bug." + "reason": "Both ask for `from_single_file`, but they target different classes/pipelines and are separate API gaps." }, { "left": "issue:4480", "right": "issue:8526", "accept": false, - "reason": "MPS half-precision kernel crash on Mac M1 is unrelated to SD3 MPS `joint_attention_kwargs` scale being ineffective. Different backend issues and different symptoms." + "reason": "MPS kernel crash with fp16 is different from `joint_attention_kwargs['scale']` being ineffective without PEFT." }, { "left": "issue:2091", "right": "issue:5124", "accept": false, - "reason": "Checkpoint resume in multi-GPU training and a question about fine-tuning a safetensor checkpoint are not the same underlying bug." + "reason": "Checkpoint resume with multi-GPU training is unrelated to how to fine-tune a `.safetensor` checkpoint." }, { "left": "issue:9539", "right": "issue:9911", "accept": false, - "reason": "CogVideoX FP8 CUDA `index_select` failure is a specific dtype/kernel limitation, while Flux multi-ControlNet with differing layer lengths is a model-shape handling issue." + "reason": "Float8 CUDA operator support in CogVideoX is unrelated to Flux MultiControlNet layer-length mismatch." }, { "left": "issue:7415", "right": "issue:7891", "accept": false, - "reason": "TPU bfloat16 AutoencoderKL corruption and generic CPU/CUDA device mismatch are different hardware/device errors with different root causes." + "reason": "TPU bfloat16 AutoencoderKL issues are not the same as mixed CPU/CUDA device placement errors." }, { "left": "issue:4480", "right": "issue:6807", "accept": false, - "reason": "Mac M1 MPS runtime crash and test collection failures on MPS are both MPS-related, but one is an inference crash and the other is a contributor test collection issue." + "reason": "A runtime crash on MPS is not the same as test collection failures on MPS." }, { "left": "issue:5607", "right": "issue:6090", "accept": false, - "reason": "These are both tensor-shape runtime errors, but the reported mismatched dimensions and model contexts are different; no evidence of the same failing code path." + "reason": "Both are shape/channel mismatches, but they describe different tensor dimensions and likely different code paths." }, { "left": "issue:10302", "right": "issue:9690", "accept": false, - "reason": "FP8 inference noise report and FLUX FP8 example code cleanup are not the same bug; one is a model-quality concern, the other is documentation/example maintenance." + "reason": "FP8 inference noise is not the same as FP8 example-code cleanup." }, { "left": "issue:3405", "right": "issue:6109", "accept": false, - "reason": "ControlNet training bug and ControlNet-xs with depth control are different training/model-variant issues with no clear overlap." + "reason": "Generic ControlNet training bug vs a specific ControlNet-xs depth-control issue." }, { "left": "issue:1249", "right": "issue:5509", "accept": false, - "reason": "ONNX pipeline `numpy.ndarray` missing `.device` and image batch-size mismatch for np/cv2 inputs are distinct input-handling bugs." + "reason": "OnnxPipeline ndarray `.device` attribute error is different from image/prompt batch-size mismatch." }, { "left": "issue:2851", "right": "issue:898", "accept": false, - "reason": "TPU `push_to_hub` failure is about training workflow on TPU, while the bfloat16 scalar-type exception is a lower-level dtype/device issue. Not the same change." - }, - { - "left": "issue:7690", - "right": "issue:7939", - "accept": false, - "reason": "Tokenizer loading failure for SDXL base and loading a model from the same folder are different loading problems with different assets and likely different causes." + "reason": "TPU push_to_hub training limitation is unrelated to unsupported BFloat16 scalar type on TPU." } ] }, @@ -81518,8 +81536,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -81558,12 +81576,12 @@ "issue:3428", "issue:3480", "issue:3769", - "issue:3902", "issue:5170", "issue:5336", "issue:6958", "issue:725", - "issue:7497", + "issue:7690", + "issue:7939", "issue:8659", "issue:9052", "issue:9331", @@ -81572,6 +81590,7 @@ "issue:9772" ], "soft_pairs": [ + "issue:7690|issue:7939", "issue:10421|issue:12505", "issue:10855|issue:12633", "issue:10913|issue:11675", @@ -81594,187 +81613,186 @@ "issue:3162|issue:6958", "issue:3280|issue:3391", "issue:3480|issue:5170", - "issue:3480|issue:5336", - "issue:3902|issue:7497" + "issue:3480|issue:5336" ], - "prepared_review_unit_hash": "a6e6cf8a000d40b80684b552e08db33d9c1d07ab6e1629f59fa433e5e94ecd31", + "prepared_review_unit_hash": "8cd42ff87cd6c4c43d42c0a3709726f0f1b90616309fbc3b566ce9524a628bcd", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33164, - "estimated_input_tokens": 8291, - "estimated_eval_tokens": 16838 + "serialized_chars": 33213, + "estimated_input_tokens": 8304, + "estimated_eval_tokens": 16864 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issue titles; the only plausible duplicate pair is the two CivitAI `from_single_file` loading failures, with one being a specific Flux checkpoint instance of the other.", - "confidence": 0.93, - "canonical_issue_reason": "#9052 is the broader, cleaner description of the shared bug: `from_single_file` fails on certain CivitAI checkpoints. #11127 reads like a specific example of that loader failure.", + "summary": "Mostly a heterogeneous set of unrelated issues; the only strong duplicate-like pair is the CivitAI/from_single_file loading failure.", + "confidence": 0.71, + "canonical_issue_reason": "issue:9052 is the clearest, most general statement of the shared bug class (from_single_file failing on certain CivitAI checkpoints), so it is the best duplicate target.", "canonical_pr_reason": null, - "best_issue_reason": "#9052 is the best representative issue because it states the underlying loading problem more generally and can subsume the more specific checkpoint report.", + "best_issue_reason": "issue:9052 is specific enough to be actionable while still broad enough to subsume issue:11127's more specific Flux/CivitAI example.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:7690", + "right": "issue:7939", + "accept": false, + "reason": "Both are loading-related, but one is a tokenizer load failure and the other is a model-from-folder load failure; too different to treat as the same bug." + }, { "left": "issue:10421", "right": "issue:12505", "accept": false, - "reason": "Different problems: GPU pipeline migration memory leak vs img2img `tgate` runtime error." + "reason": "Different problems in different code paths: GPU memory leak during pipeline placement vs tgate errors in img2img pipelines." }, { "left": "issue:10855", "right": "issue:12633", "accept": false, - "reason": "Unrelated training/scheduler behaviors: gradient accumulation optimizer stepping vs DDIM timestep mismatch." + "reason": "Gradient accumulation optimizer stepping is unrelated to DDIM timestep list mismatches." }, { "left": "issue:10913", "right": "issue:11675", "accept": false, - "reason": "Both mention loading, but one is dataset-from-disk behavior and the other is pretrained LoRA weight loading." + "reason": "Dataset-from-disk loading and pretrained LoRA weight loading are unrelated failures." }, { "left": "issue:11000", "right": "issue:11214", "accept": false, - "reason": "Different model APIs and failures: missing `shift` arg vs unexpected `encoder_hidden_states` keyword." + "reason": "Both mention missing arguments, but they affect different models and signatures; not the same underlying bug." }, { "left": "issue:1101", "right": "issue:3418", "accept": false, - "reason": "Separate feature requests for different components: scheduler addition vs LoRA training script changes." + "reason": "Scheduler feature request vs LoRA training-script enhancement; different changes." }, { "left": "issue:11127", "right": "issue:9052", "accept": true, - "reason": "Both report `from_single_file` failures on CivitAI checkpoints; #11127 is a specific Flux-model instance of the broader loader bug." + "reason": "Both report from_single_file failing on certain CivitAI checkpoints; 11127 is a specific example of the broader loading bug described by 9052." }, { "left": "issue:1117", "right": "issue:202", "accept": false, - "reason": "Tests for multiple negative prompts vs tests for `unet_blocks` are unrelated." + "reason": "Different test additions for different areas, not the same defect or change." }, { "left": "issue:11470", "right": "issue:8659", "accept": false, - "reason": "Different SDXL training-script concerns: tensorboard image logging vs checkpoint resume/inference loading." + "reason": "Training-script logging/images and checkpoint-resume/inference loading are separate issues." }, { "left": "issue:1181", "right": "issue:725", "accept": false, - "reason": "Generic training-from-scratch request vs model-loading memory footprint improvement." + "reason": "Model-from-scratch training support and memory-footprint loading improvements are unrelated." }, { "left": "issue:1197", "right": "issue:939", "accept": false, - "reason": "Community pipeline request vs using multiple community pipes in a list are distinct features." + "reason": "General community pipeline request vs using multiple community pipes in a list; not the same issue." }, { "left": "issue:11989", "right": "issue:9331", "accept": false, - "reason": "Black output image issue vs scheduler index lookup bug are different code paths." + "reason": "Black inference images and a scheduler index_for_timestep bug do not describe the same failure mode." }, { "left": "issue:12437", "right": "issue:9772", "accept": false, - "reason": "Both are support requests, but for different models/features (OmniSVG vs ControlNetPlus Union)." + "reason": "Different feature-support requests for different models/components." }, { "left": "issue:1407", "right": "issue:1821", "accept": false, - "reason": "Different pipeline regressions: ONNX feature_extractor error vs weak negative_prompt effect in inpainting." + "reason": "Onnx feature_extractor error and inpaint negative_prompt behavior are unrelated." }, { "left": "issue:1478", "right": "issue:3769", "accept": false, - "reason": "Different image-shape/output bugs in different pipelines." + "reason": "Both are output-shape/image-generation problems, but they affect different pipelines and concrete bugs." }, { "left": "issue:177", "right": "issue:244", "accept": false, - "reason": "Unrelated documentation tasks." + "reason": "Documentation tasks for different topics; not duplicates." }, { "left": "issue:1857", "right": "issue:2871", "accept": false, - "reason": "Different pipeline bugs: test coverage for UnCLIP vs latent output handling in inpainting." + "reason": "Poor test coverage for UnCLIP and latent output handling in inpainting are separate concerns." }, { "left": "issue:2392", "right": "issue:3428", "accept": false, - "reason": "Both involve img2img, but the concrete failures differ and don\u2019t look like the same bug." + "reason": "Img2Img prompt_embeds validation is a concrete API bug; the other is a broader unexpected-result report with no clear same code-path." }, { "left": "issue:2462", "right": "issue:2737", "accept": false, - "reason": "Different feature requests in training data loading vs scheduler guidance support." + "reason": "Different feature requests in different subsystems." }, { "left": "issue:2782", "right": "issue:968", "accept": false, - "reason": "Offloading error vs model loading slowness are not the same underlying issue." + "reason": "Sequential offloading error is not the same as slow model loading, even though both relate to loading/memory." }, { "left": "issue:3162", "right": "issue:6958", "accept": false, - "reason": "Different missing-argument/model-architecture issues." + "reason": "Different model/configuration bugs with no shared code-path evidence." }, { "left": "issue:3280", "right": "issue:3391", "accept": false, - "reason": "Both are DeepFloyd-related, but one is scheduler compatibility and the other is image overexposure." + "reason": "Both involve DeepFloyd IF, but one is scheduler compatibility and the other is overexposure in the super-resolution pipeline." }, { "left": "issue:3480", "right": "issue:5170", "accept": false, - "reason": "LoRA scaling support request vs removing LoRA after load are different behaviors." + "reason": "LoRA scaling and LoRA removal after loading are different features." }, { "left": "issue:3480", "right": "issue:5336", "accept": false, - "reason": "Different parameter requests for different pipelines." - }, - { - "left": "issue:3902", - "right": "issue:7497", - "accept": false, - "reason": "Unrelated feature requests." + "reason": "Both add pipeline parameters, but they target different parameters and pipelines, so they are not the same change." } ] }, "evaluator_result": { "accept": true, - "feedback": "The canonical pair choice is grounded in the packet: #11127 and #9052 both concern `from_single_file` loading failures with CivitAI checkpoints, with #11127 reading as a specific instance of the broader #9052 bug. The rest of the soft-edge rejections are conservative and supported by the titles; no PR-mergeability issue applies here." + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:11127 vs issue:9052, is reasonably supported by the titles as the same from_single_file / CivitAI checkpoint loading bug class. The rejected pairs are described at a high level without overclaiming evidence." }, "evaluator_used": true, "retried": false, @@ -81790,13 +81808,12 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:11973", - "issue:13256", "issue:1540", "issue:1709", "issue:1941", @@ -81808,6 +81825,7 @@ "issue:2966", "issue:3379", "issue:3591", + "issue:3902", "issue:4109", "issue:4446", "issue:4453", @@ -81831,7 +81849,7 @@ "issue:7178", "issue:7450", "issue:7485", - "issue:7708", + "issue:7497", "issue:8380", "issue:8909", "issue:9138", @@ -81844,6 +81862,7 @@ "issue:988" ], "soft_pairs": [ + "issue:3902|issue:7497", "issue:4109|issue:6001", "issue:4446|issue:6880", "issue:4453|issue:988", @@ -81866,181 +81885,180 @@ "issue:2966|issue:4985", "issue:1941|issue:3379", "issue:11973|issue:9227", - "issue:7178|issue:9320", - "issue:13256|issue:7708" + "issue:7178|issue:9320" ], - "prepared_review_unit_hash": "414f232d87fe1cde6feb8f1c2f25e0d13198654173d0fec04bf97e9cca814a98", + "prepared_review_unit_hash": "975c67863f16f5fd8b18a6f5f5c489c6be2058449427b39b8559ac482bffb533", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33678, - "estimated_input_tokens": 8420, - "estimated_eval_tokens": 17096 + "serialized_chars": 33575, + "estimated_input_tokens": 8394, + "estimated_eval_tokens": 17044 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous; the proposed soft pairs do not look like the same underlying bug or change. Most are unrelated training, pipeline, model-loading, or environment issues, so no pair should be merged as duplicates.", - "confidence": 0.78, - "canonical_issue_reason": "issue:11973 is the most specific and actionable report: a concrete CLI failure in a named training script with a clear missing-argument error.", + "summary": "This cluster is highly heterogeneous: most items are standalone bugs across different pipelines, scripts, and error modes. None of the soft pairs look like true duplicates or mergeable same-fix cases, so I would keep them separate.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:11973 is the best anchor for the cluster because it is narrowly scoped, reproducible, and easy to triage; the other issues are broader or about different code paths.", + "best_issue_reason": "Issue 9227 is the most concrete and actionable recent report in the set; it has a specific FLUX/Dreambooth failure, clear reproduction context, and an active discussion.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3902", + "right": "issue:7497", + "accept": false, + "reason": "Both mention control-related concepts, but one is about control-a-video and the other is a subject-attribute control request; different bugs and workflows." + }, { "left": "issue:4109", "right": "issue:6001", "accept": false, - "reason": "Both mention SDXL inpaint pipelines, but one is a general pipeline failure and the other is a from_single_file refiner-type mixup; different bugs." + "reason": "Both involve StableDiffusionXLInpaintPipeline, but one is a general failure and the other is a from_single_file refiner type-return bug; not the same code path." }, { "left": "issue:4446", "right": "issue:6880", "accept": false, - "reason": "Custom text_encoder loading vs. custom diffusion retrieve script; unrelated failure modes and code paths." + "reason": "Custom text_encoder loading in from_single_file and Custom Diffusion retrieve.py are unrelated failures in different features." }, { "left": "issue:4453", "right": "issue:988", "accept": false, - "reason": "LoRA conversion/export issue versus an old FSDP training question; same broad area but not the same bug." + "reason": "LoRA export/compatibility with A1111 and FSDP training are different problems; same broad ecosystem only." }, { "left": "issue:5112", "right": "issue:7485", "accept": false, - "reason": "Different pipelines and errors: SDXL ControlNet shape/type issue vs setting safety_checker to None causing callable error." + "reason": "One is an SDXL ControlNet shape conversion error, the other is a safety_checker callable error; different failure locations." }, { "left": "issue:5963", "right": "issue:7168", "accept": false, - "reason": "Kandinsky embeds-vs-prompts behavior and IP-Adapter image embeds handling are distinct API/data-shape problems." + "reason": "Both mention embeds, but Kandinsky prompt/embeds handling and IP-Adapter image embeds are separate APIs and bugs." }, { "left": "issue:5977", "right": "issue:9756", "accept": false, - "reason": "Saving LoRA key names and missing notebook file are unrelated." + "reason": "Saving LoRA keys and a missing notebook file are unrelated." }, { "left": "issue:6162", "right": "issue:6832", "accept": false, - "reason": "IP-Adapter shape mismatch on SDXL and Face ID community pipeline tuple/shape error are different implementations and symptoms." + "reason": "Both are IP-Adapter related, but one is an SDXL generation shape mismatch and the other is a Face ID community pipeline tuple-shape error; not the same defect." }, { "left": "issue:6518", "right": "issue:9138", "accept": false, - "reason": "A training script NoneType floor-division bug is unrelated to a generic int/string comparison error." + "reason": "Different scripts and different type errors; no shared underlying bug." }, { "left": "issue:7102", "right": "issue:7450", "accept": false, - "reason": "ControlNet multi-prompt bug and AutoencoderKL multithreading issue affect different components and code paths." + "reason": "ControlNet multi-prompt behavior and AutoencoderKL threading/memory issues are unrelated." }, { "left": "issue:8380", "right": "issue:9210", "accept": false, - "reason": "A forward-signature default-value change and an image_column validation error are not the same bug." + "reason": "ControlNet forward default-arg cleanup and a dataset column validation error are not the same issue." }, { "left": "issue:2091", "right": "issue:2213", "accept": false, - "reason": "Checkpoint resume/multi-GPU training issue vs TPU distributed-environment hang; different runtime setups and failures." + "reason": "Both touch training/distributed execution, but one is checkpoint resume on multi-GPU and the other is TPU hang/no distributed environment; different root causes." }, { "left": "issue:2799", "right": "issue:5465", "accept": false, - "reason": "Checkpoint/pipeline mismatch validation and DiffEdit inpaint_strength limitation are separate correctness checks." + "reason": "Checkpoint/pipeline mismatch validation and DiffEdit inpaint_strength handling are unrelated." }, { "left": "issue:1540", "right": "issue:2412", "accept": false, - "reason": "Missing MODEL_TYPE in transformers/dreambooth execution is different from a generic dreambooth error report." + "reason": "Both are Dreambooth-related, but the missing MODEL_TYPE startup failure and a generic Dreambooth error are too vague to treat as the same bug." }, { "left": "issue:9350", "right": "issue:9637", "accept": false, - "reason": "txt_ids tensor deprecation and xFormers attention-mask/bfloat16 failure are unrelated tensor-shape/backend issues." + "reason": "A deprecation warning for txt_ids shape and an xFormers attention-mask/bfloat16 failure are different issues." }, { "left": "issue:6565", "right": "issue:8909", "accept": false, - "reason": "DPO diffusion training instability is not the same as saving text_encoder_ti in a LoRA training script." + "reason": "DPO training instability and text_encoder_ti saving problems are unrelated training bugs." }, { "left": "issue:3591", "right": "issue:6565", "accept": false, - "reason": "ControlNet loss-curve confusion is a different training behavior issue than DPO diffusion loss spikes." + "reason": "ControlNet loss-curve confusion is not the same as DPO diffusion loss spikes/noised samples." }, { "left": "issue:4674", "right": "issue:6028", "accept": false, - "reason": "DiffusionPipeline width/height memory leak and Kandinsky 3.0 OOM are both memory-related but not the same bug." + "reason": "A DiffusionPipeline width/height memory leak and Kandinsky OOM are different memory issues in different code paths." }, { "left": "issue:1709", "right": "issue:2265", "accept": false, - "reason": "ONNX img2img tensor.astype error and ONNXRuntime incompatibility in unconditional training are distinct compatibility issues." + "reason": "Both reference ONNXRuntime, but the img2img tensor.astype error and train_unconditional incompatibility are distinct problems." }, { "left": "issue:4483", "right": "issue:6783", "accept": false, - "reason": "Nondeterministic SDXL generator arguments and PixArt-XL generations being messed up are different model/output problems." + "reason": "Generator nondeterminism in SDXL and PixArt generations being messed up are not the same concrete bug." }, { "left": "issue:2966", "right": "issue:4985", "accept": false, - "reason": "Multi-GPU InstructPix2Pix training trouble and SDXL InstructPix2Pix OOM are different failure modes." + "reason": "Multi-GPU train_instruct_pix2pix behavior and SDXL InstructPix2Pix OOM are different failure modes." }, { "left": "issue:1941", "right": "issue:3379", "accept": false, - "reason": "UnCLIP custom pipeline loading and Dreambooth intermediate checkpoint loading are unrelated loading/resume issues." + "reason": "UnCLIP custom pipeline loading and Dreambooth intermediate checkpoint loading are unrelated." }, { "left": "issue:11973", "right": "issue:9227", "accept": false, - "reason": "Both are Flux DreamBooth training reports, but one is a missing required CLI argument in kontext and the other is a model-specific training problem; not the same concrete bug." + "reason": "Both are Flux Dreambooth training reports, but one is a missing required CLI argument and the other is a model-specific training failure; not the same underlying bug." }, { "left": "issue:7178", "right": "issue:9320", "accept": false, - "reason": "export_to_video malfunction and custom diffusion save/load issues are unrelated." - }, - { - "left": "issue:13256", - "right": "issue:7708", - "accept": false, - "reason": "A package version-compatibility complaint and a missing tokenizer/config.json file are different problems." + "reason": "Video export behavior and custom diffusion model save/load issues are unrelated." } ] }, @@ -82059,18 +82077,18 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10302", "issue:10467", - "issue:10638", "issue:10803", "issue:10804", "issue:11169", "issue:13035", + "issue:13256", "issue:1407", "issue:1784", "issue:2185", @@ -82100,12 +82118,12 @@ "issue:6445", "issue:6517", "issue:6533", - "issue:6801", "issue:7012", "issue:7183", "issue:7243", "issue:7330", "issue:7569", + "issue:7708", "issue:7751", "issue:817", "issue:8612", @@ -82114,6 +82132,7 @@ "issue:9619" ], "soft_pairs": [ + "issue:13256|issue:7708", "issue:4219|issue:4488", "issue:4456|issue:4868", "issue:2250|issue:3381", @@ -82136,181 +82155,180 @@ "issue:10302|issue:9461", "issue:4720|issue:8612", "issue:5178|issue:7569", - "issue:6368|issue:6517", - "issue:10638|issue:6801" + "issue:6368|issue:6517" ], - "prepared_review_unit_hash": "b2eed5d0907c61693a5e62d1883a85cd090a4fb171d9e594114c3ae077b028f1", + "prepared_review_unit_hash": "b23dfef83c5acc298d2f2b6768bd4276443e09217a2b4f39028918dd29c1ed98", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34306, - "estimated_input_tokens": 8577, - "estimated_eval_tokens": 17410 + "serialized_chars": 34359, + "estimated_input_tokens": 8590, + "estimated_eval_tokens": 17436 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very noisy: most items are unrelated issues that only share broad terms like scheduler, ControlNet, LoRA, or precision. None of the proposed soft pairs look like true duplicates.", - "confidence": 0.61, - "canonical_issue_reason": "issue:10302 is the clearest, broadest representative of the only recurring theme here: precision-related inference quality/regression. It is also open and well-scoped.", + "summary": "This cluster is not a duplicate set: it groups many unrelated issues across inference, training, schedulers, LoRA, ControlNet, and dependency/import problems. The soft edges are mostly superficial text similarity and should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:10302 is the best single issue to anchor this cluster because it is specific, actionable, and closest to the small precision/inference subtheme present among otherwise unrelated reports.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:13256", + "right": "issue:7708", + "accept": false, + "reason": "Both mention tokenizer/package compatibility, but one is a vague version-compatibility note and the other is a missing tokenizer/config file in SDXL; different root causes." + }, { "left": "issue:4219", "right": "issue:4488", "accept": false, - "reason": "One is about ControlNet checkpoint conversion; the other is about SDXL checkpoint loading. Same broad area, different bug." + "reason": "ControlNet checkpoint conversion support vs SDXL original-checkpoint loading behavior; not the same bug or change." }, { "left": "issue:4456", "right": "issue:4868", "accept": false, - "reason": "Both are loader/runtime attribute errors, but they describe different failing code paths and different root causes." + "reason": "Different loader/runtime failures: one is a LoRA ModuleList indexing error, the other is a generic __getattr__ AttributeError." }, { "left": "issue:2250", "right": "issue:3381", "accept": false, - "reason": "Dreambooth prompt-generation failure vs inpaint batch discoloration/docs issue; not the same defect." + "reason": "Dreambooth inference not generating vs discolored inpaint batch output; both are inference issues but affect different pipelines and symptoms." }, { "left": "issue:2185", "right": "issue:5794", "accept": false, - "reason": "Missing training config files vs missing tokenizer config for SDXL; related ecosystem, but different artifacts and problems." + "reason": "Missing training config files vs missing tokenizer/config.json for SDXL; unrelated file-availability problems." }, { "left": "issue:1407", "right": "issue:5875", "accept": false, - "reason": "ONNX pipeline feature_extractor error vs SDXL ONNX prompt conversion error; not a duplicate bug." + "reason": "ONNX pipeline feature_extractor error from a prior PR vs SDXL-to-ONNX prompt error; different pipeline and failure mode." }, { "left": "issue:5628", "right": "issue:7183", "accept": false, - "reason": "Scheduler denoising offset behavior and scheduler class-name reporting are unrelated issues." + "reason": "LMS/Karras denoising-end offset bug vs scheduler class-name metadata mismatch; not the same underlying scheduler defect." }, { "left": "issue:4037", "right": "issue:7751", "accept": false, - "reason": "ControlNet FSDP validation errors vs Animatediff conversion errors; different pipelines and failure modes." + "reason": "ControlNet FSDP validation errors vs AnimateDiff conversion errors; unrelated training vs conversion paths." }, { "left": "issue:6363", "right": "issue:6445", "accept": false, - "reason": "FP16 unscale gradient error is unrelated to caption_column validation." + "reason": "FP16 gradient unscale error during LoRA training vs dataset caption_column validation error; completely different causes." }, { "left": "issue:2213", "right": "issue:2410", "accept": false, - "reason": "TPU distributed hang vs missing flax/pytorch model files; both training-related, but not the same bug." + "reason": "TPU distributed-environment hang in training script vs missing output artifacts in flax training; different training stages and symptoms." }, { "left": "issue:10467", "right": "issue:10804", "accept": false, - "reason": "Example-code filename/build artifact issue vs LoRA load failure; different FLUX problems." + "reason": "Different FLUX-related problems: one mentions example-code/compiler artifact text, the other a LoRA load failure; no shared concrete bug." }, { "left": "issue:7012", "right": "issue:7330", "accept": false, - "reason": "Weird images from v_prediction vs FP16 unscale error; distinct training/inference failures." + "reason": "Weird images with v_prediction vs FP16 gradient unscale error; one is model quality, the other is training precision handling." }, { "left": "issue:10803", "right": "issue:7243", "accept": false, - "reason": "A new model/pipeline request vs a different new adapter request; not a shared underlying change." + "reason": "A new SANA/ResAdapter pipeline request vs a domain-consistent resolution adapter proposal; different feature requests." }, { "left": "issue:11169", "right": "issue:13035", "accept": false, - "reason": "Binary incompatibility during import vs missing MT5Tokenizer import; both import failures, different causes." + "reason": "Numpy binary incompatibility when importing schedulers vs missing MT5Tokenizer from transformers; different dependency/import failures." }, { "left": "issue:1784", "right": "issue:2737", "accept": false, - "reason": "Multiple timesteps support request vs CLIP guidance support request; same scheduler family, but separate features." + "reason": "Scheduler support for multiple timesteps vs CLIP guidance support for one scheduler; related only at a high level, not the same change." }, { "left": "issue:3788", "right": "issue:6533", "accept": false, - "reason": "UniControl pipeline request vs Marigold pipeline request; unrelated model additions." + "reason": "New UniControl pipeline request vs Marigold depth-estimation pipeline request; separate feature additions." }, { "left": "issue:4139", "right": "issue:817", "accept": false, - "reason": "Both mention precision, but one is a custom attention processor cast issue and the other is a Dreambooth mixed-precision problem." + "reason": "Custom diffusion attention precision casting vs dreambooth mixed-precision not working; similar theme but different code paths and targets." }, { "left": "issue:5017", "right": "issue:9186", "accept": false, - "reason": "Control mode feature request vs Differential Diffusion with ControlNet pipeline; same subsystem, but different change requests." + "reason": "Request for ControlNet control mode vs differential diffusion with ControlNet pipeline; both are ControlNet-related but not the same feature." }, { "left": "issue:5957", "right": "issue:9619", "accept": false, - "reason": "LoRACompatibleConv padding-mode support and ControlNet interrupt support are unrelated." + "reason": "LoRACompatibleConv padding_mode support vs interrupt support in ControlNet pipeline; unrelated functionality." }, { "left": "issue:5932", "right": "issue:6176", "accept": false, - "reason": "Channel-mismatch inference error vs LoRA weight loading after Dreambooth training; different failure points." + "reason": "Channel-mismatch during SDXL inference vs LoRA weight loading after DreamBooth LoRA training; different failure points." }, { "left": "issue:10302", "right": "issue:9461", "accept": false, - "reason": "FP8 inference noise and vague Flux inference issue are not clearly the same bug." + "reason": "FP8 inference noise bug vs vague Flux inference report; too little evidence and no matching concrete symptom." }, { "left": "issue:4720", "right": "issue:8612", "accept": false, - "reason": "AutoencoderTiny pipeline incompatibility vs Pixart-sigma mixed-precision inference; different models and defects." + "reason": "AutoencoderTiny incompatibility with img2img/inpainting vs PixArt-sigma mixed-precision inference issue; different models and root causes." }, { "left": "issue:5178", "right": "issue:7569", "accept": false, - "reason": "DPM-SDE resolution-change bug vs Mac test failure for FreeU; no shared underlying issue." + "reason": "DPM SDE resolution-change bug vs Mac-specific FreeU test failure; unrelated subsystems and environments." }, { "left": "issue:6368", "right": "issue:6517", "accept": false, - "reason": "Unsupported LoRA alpha checkpoint vs ControlNet img2img import failure; unrelated." - }, - { - "left": "issue:10638", - "right": "issue:6801", - "accept": false, - "reason": "Running a specific SDXL scheduler example and loading a scheduler in diffusers 0.26.0 are related only at a broad level, not the same concrete bug." + "reason": "Unsupported LoRA checkpoint layer alpha vs controlnet SDXL img2img import error; different loaders and error types." } ] }, @@ -82329,12 +82347,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10035", + "issue:10638", "issue:10963", "issue:1099", "issue:11047", @@ -82346,7 +82365,6 @@ "issue:2097", "issue:2674", "issue:3901", - "issue:4206", "issue:4266", "issue:4578", "issue:4631", @@ -82363,6 +82381,7 @@ "issue:6399", "issue:6529", "issue:6574", + "issue:6801", "issue:6949", "issue:7061", "issue:7221", @@ -82376,13 +82395,13 @@ "issue:8502", "issue:8551", "issue:8613", - "issue:8850", "issue:9336", "issue:9477", "issue:9568", "issue:9889" ], "soft_pairs": [ + "issue:10638|issue:6801", "issue:10963|issue:12606", "issue:1099|issue:1255", "issue:2674|issue:3901", @@ -82405,181 +82424,180 @@ "issue:10035|issue:9568", "issue:7950|issue:9889", "issue:4266|issue:7239", - "issue:6574|issue:7695", - "issue:4206|issue:8850" + "issue:6574|issue:7695" ], - "prepared_review_unit_hash": "bed2c4d2e7937859507771abd93d95e229cf8ddb939da8f81f0ca5b32d347c0c", + "prepared_review_unit_hash": "6231154b344d7df6b6b3601b402f288a972efbed14d5d02611b365d0c4c6a7f2", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33294, - "estimated_input_tokens": 8324, - "estimated_eval_tokens": 16904 + "serialized_chars": 33314, + "estimated_input_tokens": 8329, + "estimated_eval_tokens": 16914 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is not a true duplicate cluster; it spans many unrelated bugs, docs issues, training problems, import errors, and model-specific regressions. All soft-edge pairs are only loosely similar by subsystem or model family, not the same underlying change or failure.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue: the items are heterogeneous and do not describe one shared bug.", + "summary": "The items are mostly unrelated bug reports spanning different models, pipelines, schedulers, and training scripts. The soft-edge pairs look like title-level similarity only, not the same underlying defect, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue is a clear canonical representative for this cluster because the set is heterogeneous and does not describe one underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": "If forced to choose a representative issue, 11885 is the most detailed and diagnostic report, but it still does not represent the rest of the set.", + "best_issue_reason": "issue:11885 is the strongest standalone issue candidate because it has the most detailed report and a concrete, reproducible LoRA-loading failure in FLUX.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10638", + "right": "issue:6801", + "accept": false, + "reason": "Both mention schedulers, but one is about running a specific SDXL script and the other about loading a scheduler; different failure modes and code paths." + }, { "left": "issue:10963", "right": "issue:12606", "accept": false, - "reason": "Both are import-related, but one is a missing exported class and the other is a wrong script import path; different fixes." + "reason": "Import failure for AutoencoderKLWan is unrelated to a wrong-import issue in a conversion script." }, { "left": "issue:1099", "right": "issue:1255", "accept": false, - "reason": "Different missing symbols in different classes; same generic import-error pattern, not the same bug." + "reason": "Dreambooth class-name issue and missing DPMSolverMultistepScheduler import are different components and bugs." }, { "left": "issue:2674", "right": "issue:3901", "accept": false, - "reason": "One is a missing dependency for LoRA support; the other is a missing requirements file. Not the same issue." + "reason": "LoRA/safetensors requirement mismatch is unrelated to a missing requirements.txt file." }, { "left": "issue:4578", "right": "issue:7221", "accept": false, - "reason": "Documentation omission versus noisy image generation; unrelated problems." + "reason": "Documentation complaint about VAE usage does not match a noisy-image inference problem." }, { "left": "issue:5919", "right": "issue:9477", "accept": false, - "reason": "Different runtime failures in different code paths; no clear shared root cause." + "reason": "Missing set_lora_layer on Linear and a GatheredParameters callability error are different runtime failures." }, { "left": "issue:6093", "right": "issue:6399", "accept": false, - "reason": "One is a push-to-hub repo-creation bug in a training script, the other is LoRA visibility in UIs." + "reason": "push_to_hub behavior in a training script is unrelated to LoRA visibility in UI tools." }, { "left": "issue:7061", "right": "issue:7418", "accept": false, - "reason": "Docstring mismatch versus a code bug in a community pipeline; not the same defect." + "reason": "Incorrect docstring and a bug in a community pipeline file are not the same defect." }, { "left": "issue:8451", "right": "issue:9336", "accept": false, - "reason": "Both mention training internals, but one concerns frozen UNet params and the other requires_grad on IPAdapter layers." + "reason": "Unet-freezing/training-script complaint is unrelated to IPAdapter gradients being left on." }, { "left": "issue:11885", "right": "issue:13162", "accept": false, - "reason": "Same model family, but one reports a LoRA loading bug and the other requests LoRA support; different underlying change." + "reason": "Both involve Flux and LoRA, but one is a loading bug and the other is support for a different model variant; not the same concrete change." }, { "left": "issue:4631", "right": "issue:6949", "accept": false, - "reason": "Both involve the scheduler, but the symptoms and triggers differ enough that they are not the same bug." + "reason": "Both involve DPMSolverSinglestepScheduler, but one is a scalar-conversion runtime error and the other produces blank images at even step counts." }, { "left": "issue:2097", "right": "issue:6162", "accept": false, - "reason": "UNet1D usage problem versus SDXL IP-Adapter shape mismatch; unrelated code paths." + "reason": "UNet1D usage and IP-Adapter SDXL shape mismatch are unrelated model-specific issues." }, { "left": "issue:7735", "right": "issue:8613", "accept": false, - "reason": "Caption-column validation error versus OOM during caption-based training; different failure modes." + "reason": "Missing caption column and OOM while using caption_column are different dataset/training problems." }, { "left": "issue:5553", "right": "issue:5917", "accept": false, - "reason": "Custom model loading failure versus a checkpoint conversion regression after upgrade; not the same issue." + "reason": "Custom model loading failure and convert_ldm_unet_checkpoint regression are distinct conversion/loading issues." }, { "left": "issue:11047", "right": "issue:8502", "accept": false, - "reason": "Different pipeline/model support requests; no shared concrete bug." + "reason": "ControlNet PAG inpaint support and SD3 512x512 generation are different feature/issues." }, { "left": "issue:1519", "right": "issue:5485", "accept": false, - "reason": "Apple M2 noisy outputs versus slow ControlNet batch inference; same broad area only." + "reason": "Apple M2 noisy outputs and slow ControlNet batch inference are different performance/correctness bugs." }, { "left": "issue:4701", "right": "issue:7575", "accept": false, - "reason": "Autoencoder decoder CUDA/dtype error versus Transformer2DModel shape error; distinct bugs." + "reason": "AutoencoderKL CUDA differentiation error and Transformer2DModel adanorm shape error are separate code paths." }, { "left": "issue:5598", "right": "issue:9477", "accept": false, - "reason": "A text-encoder modification bug versus a callable error in GatheredParameters; no common code failure." + "reason": "Text-encoder modification bug and GatheredParameters callability error are unrelated." }, { "left": "issue:5605", "right": "issue:6529", "accept": false, - "reason": "Dreambooth per-image caption support request versus an SDXL inpaint topic; unrelated." + "reason": "SDXL dreambooth caption support and Fooocus inpaint are not the same bug or change." }, { "left": "issue:5743", "right": "issue:8551", "accept": false, - "reason": "Training stuck at 0% versus VAE latent device placement in inpaint; different root causes." + "reason": "Training stuck at 0% in LCD and VAE device placement in inpaint are different problems." }, { "left": "issue:10035", "right": "issue:9568", "accept": false, - "reason": "Both concern CogVideoX VAE/autoencoder, but memory regression and reconstruction failure are different bugs." + "reason": "Both are CogVideoX-related, but one is a memory regression and the other is a reconstruction failure; different symptoms and likely different fixes." }, { "left": "issue:7950", "right": "issue:9889", "accept": false, - "reason": "Torch/Diffusers incompatibility versus a segmentation fault on a specific torch build; not the same defect." + "reason": "General PyTorch/diffusers incompatibility and a specific SDXL-turbo segmentation fault are not the same issue." }, { "left": "issue:4266", "right": "issue:7239", "accept": false, - "reason": "Mac SDXL broadcast issue versus Dreambooth LoRA matmul shape mismatch; different models and failures." + "reason": "Both mention SDXL training errors, but one is a broadcast-compatibility problem on Mac and the other is a matrix-shape mismatch in Dreambooth LoRA." }, { "left": "issue:6574", "right": "issue:7695", "accept": false, - "reason": "SVD pipeline warning versus IP-Adapter mask handling warning; same area of attention code but not one bug." - }, - { - "left": "issue:4206", - "right": "issue:8850", - "accept": false, - "reason": "ControlNet SDXL training breakage versus community pipeline mirror broken; unrelated problems." + "reason": "A warning about SVD pipeline tensor range and ignored ip_adapter_masks are unrelated." } ] }, @@ -82598,8 +82616,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -82609,7 +82627,6 @@ "issue:11497", "issue:11839", "issue:12610", - "issue:1405", "issue:1746", "issue:1964", "issue:2044", @@ -82619,10 +82636,10 @@ "issue:3182", "issue:3287", "issue:3401", - "issue:3854", "issue:3958", "issue:3970", "issue:3975", + "issue:4206", "issue:4241", "issue:4255", "issue:4269", @@ -82649,10 +82666,12 @@ "issue:8543", "issue:861", "issue:8615", + "issue:8850", "issue:9241", "issue:9926" ], "soft_pairs": [ + "issue:4206|issue:8850", "issue:10596|issue:11839", "issue:1964|issue:5429", "issue:2044|issue:2966", @@ -82676,187 +82695,186 @@ "issue:2392|issue:5335", "issue:4255|issue:8393", "issue:10303|issue:7968", - "issue:11497|issue:8507", - "issue:1405|issue:3854" + "issue:11497|issue:8507" ], - "prepared_review_unit_hash": "75f4b330573ec9a0ba7c71b4b54c9a769b4050a7739239596282e7403376c814", + "prepared_review_unit_hash": "7218ee6030fe919c6541a128a10c546ad9d6a2f1fec9b1225eeccdf96d21d628", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34064, - "estimated_input_tokens": 8516, - "estimated_eval_tokens": 17288 + "serialized_chars": 34053, + "estimated_input_tokens": 8514, + "estimated_eval_tokens": 17284 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: the items span many unrelated Diffusers bugs, feature requests, and training/runtime errors. No soft-edge pair looks like the same underlying bug or change.", - "confidence": 0.93, + "summary": "This set is a loose collection of mostly unrelated issues; there is no strong duplicate cluster or single canonical bug/change to consolidate around.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4206", + "right": "issue:8850", + "accept": false, + "reason": "Different problems: SDXL controlnet training breakage vs community-pipelines mirror availability." + }, { "left": "issue:10596", "right": "issue:11839", "accept": false, - "reason": "Both mention shape-related failures, but they affect different pipelines and different failure modes (FluxControlPipeline invalid shape vs FluxKontextPipeline gguf input-size error)." + "reason": "Both are Flux pipeline shape/input errors, but they target different pipelines and failure modes." }, { "left": "issue:1964", "right": "issue:5429", "accept": false, - "reason": "PyTorch 2.0 support planning is unrelated to adding support for models without a VAE in examples." + "reason": "PyTorch 2.0 support request vs support for models without VAE; unrelated feature areas." }, { "left": "issue:2044", "right": "issue:2966", "accept": false, - "reason": "A DDP dtype attribute error is a different bug from a multi-GPU training example problem." + "reason": "DDP dtype attribute error vs multi-GPU example training issue; not the same bug." }, { "left": "issue:2786", "right": "issue:861", "accept": false, - "reason": "One is a numpy/transformers compatibility issue in Depth2Img; the other is ckpt-to-diffusers conversion. Different code paths." + "reason": "Depth2Img numpy/transformers compatibility issue vs checkpoint conversion failure." }, { "left": "issue:3287", "right": "issue:4241", "accept": false, - "reason": "General improved DDPM support and LoRA fine-tuning support are separate feature requests, not the same change." + "reason": "General DDPM model support vs LoRA fine-tuning support; different changes." }, { "left": "issue:3287", "right": "issue:5079", "accept": false, - "reason": "Improved DDPM support and loading LyCORIS weights are different model/adapter features." + "reason": "OpenAI improved DDPM support vs loading a specific Lyrocis LoRA format." }, { "left": "issue:3970", "right": "issue:4353", "accept": false, - "reason": "Prompt embeddings not working in SDXL is a concrete functional bug; poor generation quality is a broad, different complaint." + "reason": "Prompt embeddings not working vs poor image quality; same model family, different issue." }, { "left": "issue:3975", "right": "issue:7914", "accept": false, - "reason": "A typo in an error message is unrelated to a device_map/balanced textual inversion runtime failure." + "reason": "Error-message typo vs textual inversion failure under device_map='balanced'." }, { "left": "issue:4241", "right": "issue:8615", "accept": false, - "reason": "Both concern LoRA, but one is about fine-tuning unidiffusers and the other about fuse_lora on SD3; not the same bug or patch." + "reason": "LoRA fine-tuning support request vs fuse_lora support on SD3; related theme but not the same concrete change." }, { "left": "issue:4269", "right": "issue:5171", "accept": false, - "reason": "ControlNet input validation TypeError and SDXL safety_checker constructor mismatch are different pipeline API issues." + "reason": "ControlNet check_inputs TypeError vs SDXL init rejecting safety_checker kwarg." }, { "left": "issue:4383", "right": "issue:5980", "accept": false, - "reason": "ControlNet model deployment guidance is unrelated to whether generated models are .pth or .safetensors." + "reason": "How to use a trained ControlNet in A1111 vs file format question for SDXL ControlNet models." }, { "left": "issue:4438", "right": "issue:5372", "accept": false, - "reason": "A generic SDXL pipeline error is not the same as specifically using safety_checker in SDXL." + "reason": "Generic SDXL pipeline error vs question about using safety_checker; not the same defect." }, { "left": "issue:4993", "right": "issue:5570", "accept": false, - "reason": "Missing CLIPImageProjection import structure and missing compute_snr in training_utils are separate packaging/import problems." + "reason": "Missing CLIPImageProjection export vs missing compute_snr helper in training_utils." }, { "left": "issue:6161", "right": "issue:9241", "accept": false, - "reason": "Scheduler sigma handling is unrelated to converting XLab LoRA formats via a script." + "reason": "Sigma-selection algorithm issue vs LoRA conversion script request; unrelated." }, { "left": "issue:6486", "right": "issue:8543", "accept": false, - "reason": "Embedding retraction happening twice and SD3 dreambooth gradient checkpointing failure are different training bugs." + "reason": "Dreambooth embedding retraction duplicated vs SD3 dreambooth gradient checkpointing failure." }, { "left": "issue:8482", "right": "issue:9926", "accept": false, - "reason": "Support for SD3-medium and gguf quantize/speed-up support are distinct feature requests." + "reason": "Support for SD3-medium model vs GGUF quantize/speed support; different feature requests." }, { "left": "issue:1746", "right": "issue:579", "accept": false, - "reason": "A Flax Stable Diffusion error and a missing config.json in textual inversion are unrelated subsystems." + "reason": "Flax pipeline error vs missing config.json in textual inversion; unrelated." }, { "left": "issue:3401", "right": "issue:5368", "accept": false, - "reason": "Class labels required for num_class_embeds and mixed-precision Q/K/V dtype mismatch are different validation/runtime issues." + "reason": "num_class_embeds/class_labels validation vs fp16 attention dtype mismatch." }, { "left": "issue:3182", "right": "issue:3958", "accept": false, - "reason": "Both involve offload, but one is lpw_stable_diffusion with model CPU offload and the other is LoRA with CPU offload; different concrete failure paths." + "reason": "lpw_stable_diffusion CPU offload bug vs LoRA offload issue; both offload-related but distinct code paths." }, { "left": "issue:11436", "right": "issue:12610", "accept": false, - "reason": "Half-vs-bfloat16 matmul dtype mismatch in Hidream is not the same as weights-on-CPU / mat1-on-GPU device mismatch with Quanto plus group offload." + "reason": "Half/BFloat16 matmul dtype mismatch vs CPU/GPU device mismatch under group offload; different underlying causes." }, { "left": "issue:2392", "right": "issue:5335", "accept": false, - "reason": "Prompt_embeds NoneType validation error is unrelated to deployment behind a regional Hugging Face block." + "reason": "prompt_embeds handling bug vs local deployment/network access question." }, { "left": "issue:4255", "right": "issue:8393", "accept": false, - "reason": "Image quality complaints and slow checkpoint shard loading are unrelated." + "reason": "Image quality comparison vs slow checkpoint shard loading; unrelated." }, { "left": "issue:10303", "right": "issue:7968", "accept": false, - "reason": "A tensor permutation bug in Attention.head_to_batch_dim is different from general attention behavior in UNetMotionModel's motion module." + "reason": "Attention tensor permutation bug vs attention implementation in the motion module; not enough evidence of the same defect." }, { "left": "issue:11497", "right": "issue:8507", "accept": false, - "reason": "Dataset preprocessing cache key behavior and SD3 train_text_encoder failure are separate training issues." - }, - { - "left": "issue:1405", - "right": "issue:3854", - "accept": false, - "reason": "Broken attention slicing and a locked from_ckpt load path are different bugs." + "reason": "SD3 ControlNet dataset cache-key bug vs SD3 DreamBooth text-encoder training failure; same project area, different bugs." } ] }, @@ -82875,8 +82893,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -82884,6 +82902,7 @@ "issue:11437", "issue:11874", "issue:13221", + "issue:1405", "issue:1486", "issue:1990", "issue:2331", @@ -82893,6 +82912,7 @@ "issue:3643", "issue:3699", "issue:3808", + "issue:3854", "issue:3975", "issue:4116", "issue:4117", @@ -82906,7 +82926,6 @@ "issue:5284", "issue:5485", "issue:5866", - "issue:5917", "issue:6093", "issue:6222", "issue:6242", @@ -82918,7 +82937,6 @@ "issue:6958", "issue:7142", "issue:7197", - "issue:7300", "issue:7440", "issue:7640", "issue:9350", @@ -82930,6 +82948,7 @@ "issue:9901" ], "soft_pairs": [ + "issue:1405|issue:3854", "issue:1486|issue:4414", "issue:1990|issue:2437", "issue:4116|issue:6958", @@ -82944,7 +82963,6 @@ "issue:2331|issue:6533", "issue:4325|issue:5233", "issue:7142|issue:9889", - "issue:5917|issue:7300", "issue:3699|issue:970", "issue:11874|issue:13221", "issue:2332|issue:4846", @@ -82955,178 +82973,178 @@ "issue:5485|issue:6222", "issue:6250|issue:7197" ], - "prepared_review_unit_hash": "f2c03fab49bee0624b13f554a55e5f915b3be8d48d0562469efd6b274ed30821", + "prepared_review_unit_hash": "52f786f536c724305ac1875e45025a5444e25454277012fcbbfa0ad610fcb7c3", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33711, - "estimated_input_tokens": 8428, - "estimated_eval_tokens": 17112 + "serialized_chars": 33666, + "estimated_input_tokens": 8417, + "estimated_eval_tokens": 17090 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated issues that only weakly overlap by shared keywords like Dreambooth, ControlNet, LoRA, or generic error terms. I found no true duplicate pairs among the soft edges, so there is no meaningful canonical issue/PR here.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "This set is mostly a loose topical collection of unrelated diffusion/training issues, not a true duplicate cluster. No soft pair looks like the same underlying bug or change.", + "confidence": 0.95, + "canonical_issue_reason": "No strong duplicate anchor emerges; issue:9476 is the most detailed, specific, and current open bug report among the set, so it is the best fallback representative.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue:9476 has the clearest concrete failure mode, traceback, and active status, making it the most actionable issue to keep as the cluster representative.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1405", + "right": "issue:3854", + "accept": false, + "reason": "Different failures: attention slicing regression vs a ckpt-loading hang/lock; only broad pipeline/runtime overlap." + }, { "left": "issue:1486", "right": "issue:4414", "accept": false, - "reason": "Different problems: DDIMScheduler prediction behavior vs a consistency model sampling proposal." + "reason": "One is a scheduler bug with prediction_type='sample'; the other is a new pipeline proposal/sample request. Not the same bug." }, { "left": "issue:1990", "right": "issue:2437", "accept": false, - "reason": "Separate feature requests for different pipelines (Depth2ImgVariation vs T2I-Adapter/ControlNet)." + "reason": "Depth2ImgVariation proposal and ControlNet pipeline request are separate feature ideas." }, { "left": "issue:4116", "right": "issue:6958", "accept": false, - "reason": "Different code paths and symptoms: SDXL img2img embeddings vs unet_motion_model projection settings." + "reason": "Img2Img embedding passing failure vs a motion model configuration mismatch; different code paths and symptoms." }, { "left": "issue:5016", "right": "issue:6288", "accept": false, - "reason": "One is a text-encoder training failure; the other is a learning-rate tuning complaint." + "reason": "Text-encoder training failure in SDXL LoRA vs a learning-rate tuning complaint for LCM LoRA; not the same defect." }, { "left": "issue:5017", "right": "issue:6412", "accept": false, - "reason": "Different ControlNet requests: control mode vs annotators exported to ONNX." + "reason": "Control mode request vs ONNX annotators request; unrelated feature asks." }, { "left": "issue:6242", "right": "issue:7440", "accept": false, - "reason": "Unrelated dataset/logging discussion; no shared bug or code path." + "reason": "A request for logs vs dataset availability issue; no shared bug/change." }, { "left": "issue:9704", "right": "issue:9901", "accept": false, - "reason": "Distinct runtime errors in different attention/backend paths; not the same bug." + "reason": "Different runtime errors in different components: cuDNN execution plan failure vs tuple/shape error with AttnProcessor." }, { "left": "issue:11437", "right": "issue:9476", "accept": false, - "reason": "Different training failures: missing outputs/validation images vs BF16 input/bias mismatch." + "reason": "Training outputs not being created vs a flux validation dtype mismatch; separate training problems." }, { "left": "issue:6570", "right": "issue:9350", "accept": false, - "reason": "Different SDXL issues: tensor shape mismatch vs txt_ids deprecation warning." + "reason": "Tensor size mismatch in SDXL InstructPix2Pix vs a deprecation warning about txt_ids shape; not the same issue." }, { "left": "issue:11329", "right": "issue:9530", "accept": false, - "reason": "Both mention memory, but one is a bitsandbytes leak and the other is torch.compile/UNet GPU memory freeing." + "reason": "Both mention memory, but one is a bitsandbytes quantization leak and the other is GPU memory retention after torch.compile; different causes." }, { "left": "issue:3643", "right": "issue:7640", "accept": false, - "reason": "Different failures: latent upsample INT_MAX limit vs from_single_file dimension/runtime error." + "reason": "Both touch latent decoding/dimensions, but the errors are distinct: INT_MAX overflow vs from_single_file dimension/runtime failure." }, { "left": "issue:2331", "right": "issue:6533", "accept": false, - "reason": "Separate feature requests for ControlNet and Marigold pipelines." + "reason": "ControlNet pipeline addition vs Marigold pipeline proposal; separate pipeline requests." }, { "left": "issue:4325", "right": "issue:5233", "accept": false, - "reason": "Black-image/NaN rendering issue vs wrong callback step parameter." + "reason": "Black-image/NaN casting issue vs incorrect step value passed to callback; unrelated bugs." }, { "left": "issue:7142", "right": "issue:9889", "accept": false, - "reason": "Both are inference problems, but one is output mismatch and the other is a segmentation fault." - }, - { - "left": "issue:5917", - "right": "issue:7300", - "accept": false, - "reason": "Unrelated UNet conversion regression vs GPU memory usage after changing attention processor." + "reason": "Both are SDXL inference-related, but one is output divergence and the other is a segfault; not the same code-path bug." }, { "left": "issue:3699", "right": "issue:970", "accept": false, - "reason": "Mac/CUDA availability error is unrelated to generic integration test failures." + "reason": "A platform/CUDA availability error vs failing integration tests; no shared underlying defect." }, { "left": "issue:11874", "right": "issue:13221", "accept": false, - "reason": "Both involve LoRA, but one is exclude_modules behavior and the other is Zimage support." + "reason": "Different LoRA-related requests/bugs: exclude_modules matching vs Zimage LoRA support." }, { "left": "issue:2332", "right": "issue:4846", "accept": false, - "reason": "Different Dreambooth reports: inference-after-training error vs strange training result." + "reason": "Inference-after-training error vs strange training result; same general workflow but not the same concrete failure." }, { "left": "issue:3233", "right": "issue:6093", "accept": false, - "reason": "Unrelated LoRA export request vs push_to_hub repo-creation behavior." + "reason": "Exporting trained LoRA to safetensors vs push_to_hub repo creation behavior; unrelated functionality." }, { "left": "issue:3808", "right": "issue:4117", "accept": false, - "reason": "Different Dreambooth/model-specific failures: DeepFloyd IF stage 2 vs SDXL Dreambooth." + "reason": "Different Dreambooth problems on different model families and stages; not one bug." }, { "left": "issue:5284", "right": "issue:5866", "accept": false, - "reason": "ControlNet-LLLite support request is unrelated to AnimateDiff + ControlNet." + "reason": "ControlNet-LLLite support request vs AnimateDiff+ControlNet support request; separate feature asks." }, { "left": "issue:3975", "right": "issue:497", "accept": false, - "reason": "Both are typos, but in different error messages and code locations." + "reason": "Both are typo fixes, but they concern different error messages and methods." }, { "left": "issue:5485", "right": "issue:6222", "accept": false, - "reason": "Different problems: ControlNet batch inference performance vs DDPM batch-size dependency." + "reason": "Batch inference performance issue vs scheduler batch-size dependency; different failure modes." }, { "left": "issue:6250", "right": "issue:7197", "accept": false, - "reason": "Both mention KeyError, but the failing contexts are unrelated (SDXL VAE load vs text-image resume)." + "reason": "Different KeyError contexts: SDXL VAE single-file loading vs training resume with shortest-edge preprocessing." } ] }, @@ -83228,10 +83246,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -83497,10 +83515,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -83770,10 +83788,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -84039,10 +84057,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -84309,10 +84327,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -84579,10 +84597,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -84850,10 +84868,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -85125,10 +85143,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -85395,10 +85413,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -85666,10 +85684,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -85684,162 +85702,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These soft pairs are mostly superficially similar by subsystem or wording, but they do not describe the same underlying bug/change. I would not merge any of them as duplicates.", - "confidence": 0.94, - "canonical_issue_reason": "issue:8594 is the most concrete, actionable bug report in the set: it names a specific script, a specific runtime error, and a likely code-path failure. None of the other issues is a clearer duplicate target.", + "summary": "The cluster is highly heterogeneous: the paired issues share only broad topical similarity (mostly pipeline/training/runtime errors) and do not appear to describe the same underlying bug or change. No canonical issue/PR stands out.", + "confidence": 0.93, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:8594 is the strongest standalone issue to anchor the cluster because it is specific, reproducible, and points to a direct fix location in training code.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:1462", "right": "issue:562", "accept": false, - "reason": "Both are Stable Diffusion inpaint issues, but one is an unassigned-variable bug with multiple images/masks and the other is scheduler compatibility; different failures and fixes." + "reason": "Both are inpainting-related, but one is a multi-image/multi-mask unassigned-variable bug and the other is an LMSDiscreteScheduler compatibility issue; different failure modes and code paths." }, { "left": "issue:9171", "right": "issue:9768", "accept": false, - "reason": "One is a from_single_file regression for SD 1.5 v_prediction models, the other is an SDPA/CuDNN NaN backend error; unrelated code paths." + "reason": "Both concern image quality/regression in generation, but one is an SDXL from_single_file v_prediction loading issue and the other is a torch 2.5 CuDNN SDPA NaN problem; not the same bug." }, { "left": "issue:3886", "right": "issue:8594", "accept": false, - "reason": "Both are training-script failures, but they affect different scripts and symptoms: distributed timeout vs UnboundLocalError in SD3 LoRA training." + "reason": "Both are training crashes, but one is a distributed collective timeout and the other is an UnboundLocalError in an SD3 dreambooth script; unrelated root causes." }, { "left": "issue:125", "right": "issue:2548", "accept": false, - "reason": "Accelerate launch argument parsing and LoRA checkpoint saving are different problems with no shared bug mechanism." + "reason": "Both involve training/accelerate usage, but one is an accelerate launch argument error and the other is a LoRA saving failure; different problems." }, { "left": "issue:12896", "right": "issue:13105", "accept": false, - "reason": "WAN 2.2 MPS inference failure and WAN 2.1 T2V output noise are distinct model/runtime issues." + "reason": "Both are Wan model inference issues, but one is an MPS-specific transformer switch/cache_context error and the other is output noise in a different model/version; not the same code-path bug." }, { "left": "issue:3910", "right": "issue:5458", "accept": false, - "reason": "One is a generic LoRA load failure, the other is a local_files_only behavior bug in lora_state_dict; different root causes." + "reason": "Both mention LoRA loading, but one says certain LoRAs will not load while the other is about local_files_only not being respected; different defects." }, { "left": "issue:2439", "right": "issue:6238", "accept": false, - "reason": "Inpainting pipeline loading/running vs SDXL VAE issue are separate components and failure modes." + "reason": "Both are pipeline-related, but one is inpainting load/run failure and the other is an SDXL VAE issue; too broad to be the same bug." }, { "left": "issue:4504", "right": "issue:7936", "accept": false, - "reason": "Both concern loading safetensors, but one is an internal variable assignment bug and the other is invalid JSON in a local file; not the same issue." + "reason": "Both concern safetensors loading, but one is a local variable assignment error in LoRA loading and the other is invalid JSON parsing for a local safetensors file; different failure points." }, { "left": "issue:8723", "right": "issue:8960", "accept": false, - "reason": "SD3ControlNetModel initialization and gradient accumulation in a training example are unrelated." + "reason": "Both involve SD3/training behavior, but one is SD3 ControlNet initialization and the other is gradient accumulation in a basic training example; not mergeable as one fix." }, { "left": "issue:1013", "right": "issue:3887", "accept": false, - "reason": "StableDiffusionMegaPipeline breakage in Diffusers 6.0 and a generic SD pipeline output-difference report are not the same defect." + "reason": "Both are about pipeline output differences, but one is StableDiffusionMegaPipeline on Diffusers 6.0 and the other is a generic SD pipeline effect mismatch; insufficiently specific and likely unrelated." }, { "left": "issue:10198", "right": "issue:10675", "accept": false, - "reason": "Flux weight mismatch and Flux scheduler max_shift configuration are different model/config mismatches." + "reason": "Both are Flux-related config issues, but one is a weight mismatch and the other is a scheduler max_shift difference; distinct configuration problems." }, { "left": "issue:10591", "right": "issue:9686", "accept": false, - "reason": "SD3 LoRA training script problems and SD3 ControlNet pooled projection bug are separate code paths." + "reason": "Both touch model training/inference bugs, but one is an SD3 LoRA training-script issue and the other is an SD3 ControlNet pooled projections bug; different components." }, { "left": "issue:10962", "right": "issue:8832", "accept": false, - "reason": "Prompt-embed shape handling in CogView4 and a generic pipe(prompt) crash are not the same bug." + "reason": "Both mention shape/runtime failures, but one is Cogview4 prompt-embed shape handling and the other is a generic pipe(prompt).images[0] crash; not clearly the same bug." }, { "left": "issue:13232", "right": "issue:13301", "accept": false, - "reason": "These are both feature/support requests for different pipelines, not duplicate reports of the same change." + "reason": "One is a support request for LTX-2.3, the other is a feature request for PixArtAlpha modular pipeline support; not the same underlying issue." }, { "left": "issue:2015", "right": "issue:6533", "accept": false, - "reason": "Categorical diffusion pipeline and Marigold depth estimation are different pipeline requests with no shared implementation target." + "reason": "These are unrelated pipeline feature requests: categorical diffusion vs. Marigold monocular depth estimation." }, { "left": "issue:3026", "right": "issue:3803", "accept": false, - "reason": "Single-GPU accelerate launch error in instruct_pix2pix is unrelated to multi-GPU LoRA model sharing." + "reason": "Both are training-script related, but one is an instruct_pix2pix accelerate-launch error and the other is multi-GPU LoRA sharing; different scripts and failures." }, { "left": "issue:3418", "right": "issue:4366", "accept": false, - "reason": "Request to add text-encoder support in one LoRA script and SD-XL full U-Net finetuning are different training-script enhancements." + "reason": "Both are training script enhancements, but one asks for text-encoder support in a LoRA script and the other is full U-Net fine-tuning for SDXL; not the same change." }, { "left": "issue:4337", "right": "issue:6727", "accept": false, - "reason": "torch.compile failure in Community ControlNetReference and a Q/K/V dtype mismatch are different runtime errors." + "reason": "Both are runtime errors, but one is torch.compile with ControlNetReference and the other is a QKV dtype mismatch in attention; unrelated code paths." }, { "left": "issue:5458", "right": "issue:9670", "accept": false, - "reason": "local_files_only not respected and PEFT backend required are distinct LoRA-loading issues." + "reason": "Both involve LoRA loading/backends, but one is about local_files_only handling and the other is a PEFT backend requirement; different problems." }, { "left": "issue:1083", "right": "issue:945", "accept": false, - "reason": "FP16 ONNX wrong output and inability to add a custom checkpoint for ONNX conversion are different ONNX-related problems." + "reason": "Both concern ONNX conversion/output correctness, but one is FP16 ONNX incorrect output and the other is adding a custom checkpoint to ONNX conversion; distinct issues." }, { "left": "issue:11644", "right": "issue:8503", "accept": false, - "reason": "LTXConditionPipeline init failure and SD3 gradient checkpointing crash are separate model/training issues." + "reason": "Both are initialization/training failures, but one is LTXConditionPipeline init-from-pretrained and the other is SD3 gradient checkpointing crash; not the same underlying bug." }, { "left": "issue:2035", "right": "issue:4014", "accept": false, - "reason": "Tiled upscaling artifacts and SDXL watermark pixel artifacts are both artifact reports, but from different code paths and causes." + "reason": "Both mention poor output quality, but one is tiled upscaling artifacts and the other is SDXL watermark pixel artifacts; different features and fixes." }, { "left": "issue:12126", "right": "issue:13477", "accept": false, - "reason": "wan2.2 transformer/cache_context error and ErnieImage locals() scope bug are unrelated failures." + "reason": "Both are pipeline bugs with object-scope style failures, but one is wan2.2 transformer/cache_context switching and the other is ErnieImage callback locals() scope; unrelated components." }, { "left": "issue:5124", "right": "issue:945", "accept": false, - "reason": "Fine-tuning a checkpoint and converting a custom checkpoint to ONNX are different tasks, not duplicate bug reports." + "reason": "Both reference checkpoints/safetensors/ONNX workflows, but one is how to fine tune a checkpoint .safetensor and the other is converting a custom checkpoint to ONNX; not the same bug." }, { "left": "issue:1574", "right": "issue:1708", "accept": false, - "reason": "Dreambooth training failing on 8GB GPU and a YAML config question for inference are not the same issue." + "reason": "One is Dreambooth training failing on 8GB bare metal Linux, the other is a question about passing a YAML config for inference; different topics and not a duplicate." } ] }, @@ -85941,10 +85959,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -86211,10 +86229,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -86481,10 +86499,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -86754,10 +86772,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -87024,10 +87042,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -87293,10 +87311,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -87563,10 +87581,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -87833,10 +87851,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, @@ -88105,10 +88123,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -88374,10 +88392,10 @@ "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, @@ -88560,8 +88578,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -88592,7 +88610,7 @@ "issue:4767", "issue:4875", "issue:4933", - "issue:5017", + "issue:4935", "issue:5103", "issue:5544", "issue:5546", @@ -88604,7 +88622,7 @@ "issue:6686", "issue:6964", "issue:7220", - "issue:8390", + "issue:7300", "issue:8451", "issue:8873", "issue:9051", @@ -88637,180 +88655,180 @@ "issue:2758|issue:5103", "issue:3354|issue:4424", "issue:4616|issue:5544", - "issue:5017|issue:8390" + "issue:4935|issue:7300" ], - "prepared_review_unit_hash": "d91328b382efea49b415c8d66fffd250f04960b2e6728d60632df50436a541c7", + "prepared_review_unit_hash": "1cf5569ba6cc86016275ab76992dfbe706b164d8cfc26d426f62136daab57c40", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33314, - "estimated_input_tokens": 8329, - "estimated_eval_tokens": 16914 + "serialized_chars": 33417, + "estimated_input_tokens": 8355, + "estimated_eval_tokens": 16966 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are mostly unrelated; the soft pairs only share superficial keywords like ControlNet, LoRA, training, or black images, not the same underlying bug/change.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "This cluster is a loose similarity bucket, not a true duplicate set: the items span unrelated training bugs, scheduler/timestep issues, memory/performance regressions, and feature requests across different pipelines and models. None of the soft pairs look like the same underlying bug/change.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits the cluster; the only commonality is superficial topic overlap in titles, not a shared code-path or bug.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "No globally best issue for deduplication. The closest tracker-style issues (for example 6586) are still too narrow and do not represent the rest of the heterogeneous set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:5997", "right": "issue:9907", "accept": false, - "reason": "Different problems: SDXL ControlNet tutorial/model-type mismatch vs Flux controlnet sequence-length warning." + "reason": "Different problems: a ControlNet SDXL tutorial type-mismatch versus a Flux controlnet sequence-length warning." }, { "left": "issue:2579", "right": "issue:3441", "accept": false, - "reason": "xFormers misuse/bug report versus PyTorch 2 VRAM regression; different symptoms and code paths." + "reason": "Both mention xFormers/PyTorch performance, but one is a misuse/bug report and the other is a VRAM regression under PyTorch 2." }, { "left": "issue:3441", "right": "issue:4159", "accept": false, - "reason": "Both mention memory, but one is PyTorch2 VRAM under specific conditions and the other is DiffEdit needing huge VRAM." + "reason": "Not the same bug: higher VRAM with PyTorch 2 is distinct from DiffEdit\u2019s extreme VRAM requirement." }, { "left": "issue:6101", "right": "issue:6964", "accept": false, - "reason": "ControlNet inpaint performance regression is not the same as 4-channel UNet compatibility failure." + "reason": "Different code paths: ControlNet inpaint performance versus incompatibility with 4-channel UNets." }, { "left": "issue:10194", "right": "issue:12120", "accept": false, - "reason": "Flux attention masking feature request is unrelated to LoRA training on distilled Flux models." + "reason": "A feature request for Flux attention masking is unrelated to a question about LoRA training on distilled Flux models." }, { "left": "issue:11221", "right": "issue:9799", "accept": false, - "reason": "AutoencoderKL non-convergence during training is unrelated to a tensor shape mismatch runtime error." + "reason": "Training non-convergence for AutoencoderKL is unrelated to a tensor shape mismatch runtime error." }, { "left": "issue:4875", "right": "issue:9051", "accept": false, - "reason": "Both involve batch behavior, but one is a scheduler crash on batch-size changes and the other is FluxPipeline lacking batch inference support." + "reason": "Scheduler crash when batch size changes is not the same as lack of batch inference support in FluxPipeline." }, { "left": "issue:6552", "right": "issue:6586", "accept": false, - "reason": "Training resume FP16 tracker and SDXL micro-conditioning tracker are separate training topics." + "reason": "Both are trackers, but they cover different training fixes: FP16 resume issues vs SDXL micro-conditioning." }, { "left": "issue:11321", "right": "issue:8451", "accept": false, - "reason": "README bug for Flux ControlNet training is not the same as freezing UNet parameters / missing sd1.x training script." + "reason": "A README bug for Flux ControlNet training is unrelated to documentation about frozen UNet parameters / missing SD1.x training script." }, { "left": "issue:11569", "right": "issue:12624", "accept": false, - "reason": "Repo request for a model checkpoint is unrelated to adding a strength parameter to WanImageToVideoPipeline." + "reason": "A repo-request feature asks for a model repository; the other requests adding a strength parameter to a video pipeline." }, { "left": "issue:2875", "right": "issue:3209", "accept": false, - "reason": "Both mention NaNs, but one is training loss divergence and the other is AutoencoderKL encoder NaNs on large images." + "reason": "NaN training loss and AutoencoderKL encoder NaNs are different failures with different triggers." }, { "left": "issue:4767", "right": "issue:5621", "accept": false, - "reason": "Scheduler sample support and SDXL ControlNet encoder_hid_proj support are different concrete issues." + "reason": "Scheduler prediction_type support is unrelated to SDXL ControlNet encoder_hid_proj support." }, { "left": "issue:5546", "right": "issue:6686", "accept": false, - "reason": "Wrong step count in base+refiner ensemble is unrelated to lack of compile performance improvement after save/load." + "reason": "Wrong step counts in schedulers is a different problem from Dynamo compilation performance not persisting after save/load." }, { "left": "issue:9297", "right": "issue:9670", "accept": false, - "reason": "Linear LoRA load time regression is not the same as PEFT-backend-required load failure." + "reason": "Linear LoRA load-time slowdown is unrelated to the PEFT-backend-required load failure." }, { "left": "issue:11092", "right": "issue:11136", "accept": false, - "reason": "Both concern timesteps, but they affect different pipelines/models and have different failure modes." + "reason": "Both involve timesteps, but they affect different pipelines/models and different failure modes." }, { "left": "issue:1009", "right": "issue:1486", "accept": false, - "reason": "Textual inversion weight decay bug is unrelated to DDIMScheduler prediction_type='sample' behavior." + "reason": "Textual inversion weight-decay bug is unrelated to a DDIMScheduler prediction_type='sample' bug." }, { "left": "issue:3348", "right": "issue:3960", "accept": false, - "reason": "LoRA xformers scale support and HeaderTooLarge are entirely different kinds of bugs." + "reason": "Support for xFormers scale in LoRA models is not the same as a HeaderTooLarge error." }, { "left": "issue:4933", "right": "issue:7220", "accept": false, - "reason": "Dreambooth output file problem is unrelated to unload_ip_adapter() failing in ControlNet inpaint." + "reason": "Dreambooth missing pytorch_model.bin after training is unrelated to unload_ip_adapter() failing in ControlNet inpaint." }, { "left": "issue:11083", "right": "issue:8873", "accept": false, - "reason": "DDP stride mismatch warning is a different issue from xformers being slower than no xformers." + "reason": "DDP stride-mismatch warning and xFormers being slower on A100 are different performance/debug issues." }, { "left": "issue:11978", "right": "issue:12369", "accept": false, - "reason": "These are separate model/pipeline feature requests for different systems." + "reason": "These are separate feature requests for different video/image models and pipelines." }, { "left": "issue:2758", "right": "issue:5103", "accept": false, - "reason": "Both show black images, but one is ROCm torch.compile-related and the other is zero-SNR DDIM-related." + "reason": "Both mention black images, but one is ROCm torch.compile warning behavior and the other is zero-SNR DDIM sampling." }, { "left": "issue:3354", "right": "issue:4424", "accept": false, - "reason": "Training-script fp16 casting bug and SDXL VAE upcast bug are different dtype-handling problems in different scripts." + "reason": "Unet being cast to fp16 in a LoRA training script is distinct from unconditional VAE upcasting in SDXL Instruct Pix2Pix." }, { "left": "issue:4616", "right": "issue:5544", "accept": false, - "reason": "LoRA safetensor conversion failure is unrelated to disabling FreeU raising an error." + "reason": "SDXL LoRA conversion failing is unrelated to disabling FreeU when it is not enabled." }, { - "left": "issue:5017", - "right": "issue:8390", + "left": "issue:4935", + "right": "issue:7300", "accept": false, - "reason": "Control mode request for ControlNet and Stable Cascade controlnet training are not the same underlying change." + "reason": "A question about AttnProcsLayers in an SDXL LoRA script is unrelated to high GPU memory after set_default_attn_processor()." } ] }, @@ -88829,16 +88847,14 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10039", "issue:1009", "issue:10112", - "issue:10266", - "issue:10337", "issue:10472", "issue:11169", "issue:11497", @@ -88861,6 +88877,7 @@ "issue:4916", "issue:4923", "issue:5016", + "issue:5017", "issue:5271", "issue:5474", "issue:5515", @@ -88874,6 +88891,7 @@ "issue:6700", "issue:6742", "issue:7859", + "issue:8390", "issue:8488", "issue:8644", "issue:8803", @@ -88884,6 +88902,7 @@ "issue:964" ], "soft_pairs": [ + "issue:5017|issue:8390", "issue:5474|issue:9216", "issue:5593|issue:6700", "issue:5710|issue:9007", @@ -88906,181 +88925,180 @@ "issue:3282|issue:5897", "issue:10039|issue:8803", "issue:1009|issue:3405", - "issue:10112|issue:7859", - "issue:10266|issue:10337" + "issue:10112|issue:7859" ], - "prepared_review_unit_hash": "98bb2d28689a032e68831b116f3424e3137281871eb1381c5e06b44433bb3dcf", + "prepared_review_unit_hash": "3d108e0afe08035696f4506efef07f0b7412a82c9ca3a191113b2a0206b7eb08", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34396, - "estimated_input_tokens": 8599, - "estimated_eval_tokens": 17454 + "serialized_chars": 34356, + "estimated_input_tokens": 8589, + "estimated_eval_tokens": 17434 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: most items only share broad vocabulary like schedulers, pipelines, attention, or xFormers, but they describe different bugs or feature requests. I would reject all soft edges; none look like the same underlying change or defect.", - "confidence": 0.24, - "canonical_issue_reason": "issue:10039 is the most concrete, narrowly scoped bug report with an explicit file/function target (`AttnUpBlock2D` in `unet_2d_blocks.py`) and a clear missing-argument failure mode.", + "summary": "The soft links mostly connect unrelated diffusers issues across different subsystems (pipelines, schedulers, training scripts, environment/import errors, and feature requests). I do not see any pair that clearly describes the same underlying bug/change, so all soft edges should be rejected.", + "confidence": 0.91, + "canonical_issue_reason": "Issue:10039 is the sharpest concrete code bug in the set (a specific UNet attention call in unet_2d_blocks.py), so it is the best available anchor, though the items are not actually one duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10039 is the best representative issue because it is precise, actionable, and clearly tied to a single code-path fix; the rest are broader, vaguer, or clearly different problems.", + "best_issue_reason": "Issue:10039 is the most concrete and code-path-specific item here; the others are mostly broader feature requests or unrelated runtime/environment failures.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5017", + "right": "issue:8390", + "accept": false, + "reason": "ControlNet control mode request vs Stable Cascade ControlNet training bug; different problem and scope." + }, { "left": "issue:5474", "right": "issue:9216", "accept": false, - "reason": "Both are pipeline-related ideas, but one is a specific reference-only pipeline request and the other is a generic multi-model pipeline concept; different changes." + "reason": "Two different pipeline feature requests for different models; not the same change." }, { "left": "issue:5593", "right": "issue:6700", "accept": false, - "reason": "Both mention pipeline/attention internals, but one is about `AutoPipeline.from_pipe` with ControlNet and the other is a `temb` argument TypeError in attention processors; unrelated bugs." + "reason": "AutoPipeline conversion failure and attention temb TypeError are different code paths." }, { "left": "issue:5710", "right": "issue:9007", "accept": false, - "reason": "Same scheduler family, but one is missing `sigmas` on init and the other is black output at order 3; different failure modes and likely different code paths." + "reason": "Scheduler initialization bug vs black output with order=3; distinct scheduler defects." }, { "left": "issue:2306", "right": "issue:3643", "accept": false, - "reason": "Both involve runtime size limits, but one is an `INT_MAX` error and the other is an NHWC upsample tensor-size limit while decoding latents; not the same defect." + "reason": "Both mention INT_MAX, but one is a generic size error and the other is a specific latent decoding limit; not clearly the same bug." }, { "left": "issue:10472", "right": "issue:11497", "accept": false, - "reason": "One is a LoRA `lora_bias` constructor argument error, the other is dataset cache invalidation from unrelated args; clearly different areas." + "reason": "LoRA bias constructor error vs preprocessing cache dependence on unrelated args; unrelated bugs." }, { "left": "issue:11169", "right": "issue:13176", "accept": false, - "reason": "Binary incompatibility on import versus an attention error in UNet2DConditionModel; unrelated issues." + "reason": "Binary incompatibility on import vs UNet attention error; different failure modes." }, { "left": "issue:12358", "right": "issue:13301", "accept": false, - "reason": "Both are pipeline feature requests, but they target different pipelines and different additions; not the same change." + "reason": "Separate pipeline support requests for different models; not duplicates." }, { "left": "issue:8488", "right": "issue:8644", "accept": false, - "reason": "Both relate to Stable Diffusion 3 loading, but one is a `from_pretrained` exception and the other is a separate T5/meta-tensor offload bug; different code paths." + "reason": "Both involve SD3 loading, but one is a generic from_pretrained exception and the other is a specific offload/meta-tensor bug." }, { "left": "issue:5515", "right": "issue:6742", "accept": false, - "reason": "One is an LCM scheduler initialization failure; the other is a motion block checkpointing discrepancy. Same broad domain, different bugs." + "reason": "LCM pipeline without scheduler and DownBlockMotion checkpointing behavior are unrelated." }, { "left": "issue:1314", "right": "issue:4916", "accept": false, - "reason": "Both are training-time failures, but one is xformers backward on Windows and the other is gradient checkpointing + ZeRO3; not the same underlying issue." + "reason": "Windows xformers backward failure vs checkpointing/ZeRO3 training failure; different interactions." }, { "left": "issue:1537", "right": "issue:941", "accept": false, - "reason": "Both are MPS-related, but they affect different schedulers and different symptoms (`KDPM2` scheduler issues vs noisy `num_images_per_prompt` output)." + "reason": "MPS scheduler issue vs img2img noise with num_images_per_prompt; separate bugs." }, { "left": "issue:1540", "right": "issue:579", "accept": false, - "reason": "Dreambooth/transformers missing `MODEL_TYPE` is unrelated to Textual Inversion missing `config.json`; different subsystems and failures." + "reason": "Missing MODEL_TYPE in Dreambooth/transformers vs missing config.json in textual inversion; different config-loading issues." }, { "left": "issue:1918", "right": "issue:964", "accept": false, - "reason": "Both are example failures, but one is a DDIM scheduler issue for a specific dataset example and the other is inpainting example failure; insufficient evidence of sameness." + "reason": "Different example failures (DDIM scheduler vs inpainting example); not the same underlying defect." }, { "left": "issue:6412", "right": "issue:9186", "accept": false, - "reason": "One is a request to port ControlNet annotators to ONNX; the other is about differential diffusion with ControlNet. Feature request vs pipeline behavior bug." + "reason": "ControlNet annotator ONNX request vs differential diffusion with ControlNet pipeline; different feature areas." }, { "left": "issue:5016", "right": "issue:6366", "accept": false, - "reason": "Both are Dreambooth/LoRA training scripts, but one is about text encoder training and the other about checkpoint reload correctness; different concrete bugs." + "reason": "SDXL LoRA text encoder training bug vs checkpoint restore bug; same script family but different concrete failures." }, { "left": "issue:4923", "right": "issue:5271", "accept": false, - "reason": "AudioLDM 2 transcript support is a feature request; different learning rates for text encoders and UNet is a training configuration feature. Not the same issue." + "reason": "AudioLDM2 transcript feature request vs dreambooth learning-rate feature request; unrelated." }, { "left": "issue:1262", "right": "issue:1896", "accept": false, - "reason": "Rectangle-image training and CLIP-guided SD size/dict TypeError are unrelated." + "reason": "Rectangle-image training question vs CLIP-guided Stable Diffusion size/type error; unrelated." }, { "left": "issue:2579", "right": "issue:3660", "accept": false, - "reason": "Both mention xFormers/compile, but one is a likely xFormers bug or misuse and the other is torch.compile incompatibility with LoRA or xFormers; too different to merge." + "reason": "Ambiguous xFormers misuse/bug report vs torch.compile incompatibility with LoRA/xformers; not the same bug." }, { "left": "issue:4593", "right": "issue:5678", "accept": false, - "reason": "LoRA rank-loading failure and higher-order-derivative attention error are distinct problems in different code paths." + "reason": "LoRA rank validation error vs higher-order-derivative attention bug; different code paths." }, { "left": "issue:3282", "right": "issue:5897", "accept": false, - "reason": "One is missing `memory_efficient_attention_forward` operator support; the other is a CPU/GPU device mismatch during LoRA training. Different root causes." + "reason": "xFormers operator dispatch failure vs device mismatch in LoRA training; unrelated runtime errors." }, { "left": "issue:10039", "right": "issue:8803", "accept": false, - "reason": "Missing output sample size in `AttnUpBlock2D` and duplicate `hidden_states` assignment in attention processor are different attention bugs in different files." + "reason": "Both touch attention internals, but one is a missing output sample size in AttnUpBlock2D and the other is a duplicate hidden_states assignment; different bugs." }, { "left": "issue:1009", "right": "issue:3405", "accept": false, - "reason": "Weight decay bug in textual inversion and a controlnet training bug are unrelated training issues." + "reason": "Textual inversion weight decay bug vs ControlNet training bug; unrelated." }, { "left": "issue:10112", "right": "issue:7859", "accept": false, - "reason": "One is a vague discussion about diffusers, the other is a request to add FlashFace implementation; not the same underlying change." - }, - { - "left": "issue:10266", - "right": "issue:10337", - "accept": false, - "reason": "UniPC + FlowMatch index error and FLUX IPAdapter failure with quantized transformers are different runtime problems." + "reason": "Diffuse 'Detail-Daemon' issue vs FlashFace implementation request; different topics." } ] }, @@ -89099,11 +89117,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10266", + "issue:10337", "issue:10381", "issue:10591", "issue:10782", @@ -89119,11 +89139,9 @@ "issue:1994", "issue:201", "issue:2192", - "issue:2392", "issue:2412", "issue:2785", "issue:2907", - "issue:3040", "issue:3045", "issue:3706", "issue:3761", @@ -89153,6 +89171,7 @@ "issue:9810" ], "soft_pairs": [ + "issue:10266|issue:10337", "issue:10381|issue:8604", "issue:10591|issue:11790", "issue:10591|issue:9810", @@ -89175,181 +89194,180 @@ "issue:12012|issue:12610", "issue:1746|issue:2412", "issue:4616|issue:5441", - "issue:6250|issue:7574", - "issue:2392|issue:3040" + "issue:6250|issue:7574" ], - "prepared_review_unit_hash": "4beec597f964f0da3e7060432171baa7b3f02723f320cc8c6b827f95d819fefd", + "prepared_review_unit_hash": "7fad6dd7b23982ca8f04ee448c5b955a00db496e3ae7c36cda04533472a66136", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 32992, - "estimated_input_tokens": 8248, - "estimated_eval_tokens": 16752 + "serialized_chars": 32858, + "estimated_input_tokens": 8215, + "estimated_eval_tokens": 16686 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All proposed edges are superficial keyword matches; none appear to describe the same underlying bug or change. The items span unrelated areas (LoRA training/loading, prompt handling, scheduler behavior, offload, file loading, dtype conversion), so this is not a clean duplicate cluster.", + "summary": "All candidate pairs look like distinct issues or feature requests across different models, scripts, and failure modes. The shared keywords are too broad to treat any pair as the same underlying bug/change, so every soft edge is rejected.", "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue fits well: the set is heterogeneous and the soft pairs do not share a concrete failure mode or code path.", + "canonical_issue_reason": "No clear canonical issue: the items span unrelated topics and none of the pairings convincingly collapse to one bug or change.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue: the cluster is too mixed to have one representative bug report without losing important distinctions.", + "best_issue_reason": "No single issue is a safe duplicate hub for this set; picking one would overfit to superficial keyword overlap.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10266", + "right": "issue:10337", + "accept": false, + "reason": "Different pipelines and failures: UniPC/FlowMatch index error vs FLUX IPAdapter quantization failure." + }, { "left": "issue:10381", "right": "issue:8604", "accept": false, - "reason": "Different problems: Flux/Quanto quantization + LoRA inference error vs T5 text-encoder float16 corruption." + "reason": "Both mention Flux and quantization, but one is a Flux LoRA/Quanto inference issue and the other is a T5 float16 corruption bug." }, { "left": "issue:10591", "right": "issue:11790", "accept": false, - "reason": "Both are LoRA-related, but one is an SD3 training-script issue and the other is a weight-loading failure from a different script/path." + "reason": "Both are LoRA training-script complaints, but they target different scripts and different failure points." }, { "left": "issue:10591", "right": "issue:9810", "accept": false, - "reason": "Training-script bug vs missing LoRA `module_to_save()` support; related ecosystem, but not the same bug." + "reason": "SD3 training-script bug vs a general LoRA module_to_save limitation; not the same issue." }, { "left": "issue:10782", "right": "issue:12083", "accept": false, - "reason": "Both mention model-specific issues, but Lumina Image 2.0 and Qwen-Image long-prompt errors are unrelated failure cases." + "reason": "Different models and symptoms: Lumina Image 2.0 minor issue vs Qwen-Image long-prompt error." }, { "left": "issue:10901", "right": "issue:9178", "accept": false, - "reason": "Different models and symptoms: HunyuanVideo negative-prompt output bug vs Flux LoRA PEFT target-module error." + "reason": "HunyuanVideo negative_prompt behavior and FLUX LoRA PEFT target-module errors are unrelated code paths." }, { "left": "issue:1101", "right": "issue:2785", "accept": false, - "reason": "Scheduler feature request is unrelated to M1/M2 text-to-video support." + "reason": "Add a scheduler vs add M1/M2 support; separate feature requests." }, { "left": "issue:1117", "right": "issue:201", "accept": false, - "reason": "Multiple negative prompt tests vs attention class tests; same area of testing, but not the same issue." + "reason": "Both are test-related, but they cover different behaviors and do not describe the same test failure/change." }, { "left": "issue:1994", "right": "issue:5079", "accept": false, - "reason": "S3 path support in `from_pretrained` is unrelated to LyCORIS LoRA loading support." + "reason": "AWS/from_pretrained storage support is unrelated to loading LyCORIS weights." }, { "left": "issue:2192", "right": "issue:6230", "accept": false, - "reason": "Missing DreamBooth LoRA CLI args vs SDXL out-of-memory during training; different bugs." + "reason": "Missing CLI arguments in one training script vs CUDA OOM in another; distinct problems." }, { "left": "issue:2907", "right": "issue:3761", "accept": false, - "reason": "CPU offload called twice vs loading two community pipelines in one process; both are lifecycle issues but different code paths." + "reason": "Repeated CPU offload and loading multiple community pipelines are different lifecycle bugs." }, { "left": "issue:3045", "right": "issue:5989", "accept": false, - "reason": "LoRA conversion script issue vs generic `safetensors.torch` bug; no concrete overlap shown." + "reason": "LoRA conversion utility issue vs a safetensors.torch bug; no concrete overlap." }, { "left": "issue:3706", "right": "issue:7525", "accept": false, - "reason": "Wuerstchen model announcement vs discrete scheduler typo/feature request; not duplicates." + "reason": "A new model announcement and a discrete scheduler issue are not the same change." }, { "left": "issue:4194", "right": "issue:6441", "accept": false, - "reason": "Single-file VAE loading bug vs Colab `from_pretrained` issue; different environments and failure modes." + "reason": "Single-file VAE loading failure vs Google Colab from_pretrained failure; different contexts and symptoms." }, { "left": "issue:4468", "right": "issue:6686", "accept": false, - "reason": "Multi-GPU sample-code performance discrepancy vs Dynamo save/load performance regression; related only at a very broad level." + "reason": "Multi-GPU performance discrepancy and Dynamo save/load performance regression are different issues." }, { "left": "issue:5135", "right": "issue:6176", "accept": false, - "reason": "LoRA quality degradation with xformers vs failure loading LoRA weights after training; distinct symptoms and likely causes." + "reason": "LoRA quality degradation with xformers is unrelated to loading LoRA weights after training." }, { "left": "issue:5489", "right": "issue:7157", "accept": false, - "reason": "General PEFT integration alert vs LCM-LoRA unexpected keys when loading with peft; not the same concrete bug." + "reason": "Broad PEFT integration work vs unexpected keys when loading LCM LoRA; not the same concrete bug." }, { "left": "issue:5538", "right": "issue:8588", "accept": false, - "reason": "Upscale pipeline latent usage question vs missing SD3 conversion script; unrelated." + "reason": "Up-scaler latent usage bug vs missing SD3 conversion script; different files and failures." }, { "left": "issue:5777", "right": "issue:8567", "accept": false, - "reason": "PixArt DreamBooth training error vs question about SD3 training objective; different topics entirely." + "reason": "PixArt dreambooth training error vs SD3 training-objective question; not the same underlying problem." }, { "left": "issue:11653", "right": "issue:8579", "accept": false, - "reason": "Both are LoRA support problems, but HiDream `.alpha` key compatibility and SD3 LoRA non-functional are different model-specific issues." + "reason": "Both involve LoRA support, but one is about .alpha keys in HiDream LoRAs and the other is SD3 LoRA functionality." }, { "left": "issue:12012", "right": "issue:12610", "accept": false, - "reason": "Wan 2.2 first-block cache context error vs Quanto+group-offload device mismatch; both involve execution state, but not the same bug." + "reason": "Context-state error with Wan 2.2 caching vs Quanto/group-offload device mismatch; different mechanisms." }, { "left": "issue:1746", "right": "issue:2412", "accept": false, - "reason": "Flax Stable Diffusion error vs DreamBooth error; too vague and not tied to the same code path." + "reason": "Flax pipeline error and a generic dreambooth error are too broad to be the same bug." }, { "left": "issue:4616", "right": "issue:5441", "accept": false, - "reason": "SDXL LoRA conversion script failure vs Wuerstchen prior script missing attn processors; both affect LoRA tooling but are different scripts/problems." + "reason": "SDXL LoRA conversion failure vs WuerstchenPrior missing attn_processors; different models and scripts." }, { "left": "issue:6250", "right": "issue:7574", "accept": false, - "reason": "SDXL VAE single-file `KeyError` vs checkpoint key `middle_block_out.0.weight`; both are loading-key issues but not the same missing key or loader bug." - }, - { - "left": "issue:2392", - "right": "issue:3040", - "accept": false, - "reason": "Img2Img `prompt_embeds` validation bug vs copy-pasted docs causing list `__module__` AttributeError; unrelated." + "reason": "Different KeyErrors in different loading paths; no evidence they share the same root cause." } ] }, @@ -89368,8 +89386,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -89384,10 +89402,12 @@ "issue:13488", "issue:2163", "issue:2370", + "issue:2392", "issue:2410", "issue:2581", "issue:2610", "issue:2988", + "issue:3040", "issue:3149", "issue:3194", "issue:3250", @@ -89409,11 +89429,9 @@ "issue:5178", "issue:5339", "issue:5365", - "issue:5628", "issue:5957", "issue:6457", "issue:6870", - "issue:7240", "issue:8296", "issue:8429", "issue:8494", @@ -89423,6 +89441,7 @@ "issue:9613" ], "soft_pairs": [ + "issue:2392|issue:3040", "issue:11872|issue:13488", "issue:2610|issue:4480", "issue:2988|issue:444", @@ -89445,188 +89464,190 @@ "issue:1339|issue:3149", "issue:2370|issue:4756", "issue:4232|issue:8509", - "issue:4904|issue:5957", - "issue:5628|issue:7240" + "issue:4904|issue:5957" ], - "prepared_review_unit_hash": "e01bb86a98505ef718767972be5f9e94d93b0b5e54460f96a20267d9add49360", + "prepared_review_unit_hash": "b73dab6ce2f0d2000536b83d611b2944cc00c108ddd680f9984afb8eb0984f05", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33908, - "estimated_input_tokens": 8477, - "estimated_eval_tokens": 17210 + "serialized_chars": 33972, + "estimated_input_tokens": 8493, + "estimated_eval_tokens": 17242 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate group emerges from this cluster: the issues are mostly unrelated bugs that only share broad subsystems like offloading, LoRA loading, schedulers, or training scripts. All soft edges should stay separate.", - "confidence": 0.95, - "canonical_issue_reason": "No canonical issue selected: the items do not share one underlying defect; they span distinct bugs and feature gaps.", + "summary": "The set is mostly heterogeneous and the soft edges are largely false positives from superficial title similarity. The only likely true duplicate pair is the seed-determinism reports (#1339 and #3149).", + "confidence": 0.89, + "canonical_issue_reason": "No single canonical issue fits the full cluster: it mixes unrelated bugs across flux, LoRA loading, offloading, training scripts, conversion tools, and scheduler behavior. If a subcluster representative were needed, #1339 is the broadest determinism report.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue is appropriate for deduplication because there is no single representative bug across this set.", + "best_issue_reason": "No globally best issue for the whole cluster; the items are too unrelated for one issue to represent them well. #1339 is only the best representative for the determinism subcluster, not the cluster overall.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2392", + "right": "issue:3040", + "accept": false, + "reason": "Both involve prompt-related errors, but one is an img2img prompt_embeds ValueError and the other is a docs-copy/paste AttributeError; different bugs." + }, { "left": "issue:11872", "right": "issue:13488", "accept": false, - "reason": "Both involve offloading, but one is a performance inefficiency and the other is a version regression; different underlying bugs." + "reason": "Both mention offloading, but one is a loop inefficiency from repeated CPU/GPU transfers and the other is a regression breaking offloading in a specific dev version." }, { "left": "issue:2610", "right": "issue:4480", "accept": false, - "reason": "Both are Mac-related errors, but one is a local build clang issue and the other is an MPS half-precision crash." + "reason": "Mac-related failures, but one is a local setup/clang problem and the other is an MPS float16 kernel crash during generation." }, { "left": "issue:2988", "right": "issue:444", "accept": false, - "reason": "Both mention scheduler/step edge cases, but they describe different failure conditions and symptoms." + "reason": "Both are scheduler/step-count problems, but one crashes when training and inference steps are equal while the other is an index error for specific step values." }, { "left": "issue:10460", "right": "issue:8494", "accept": false, - "reason": "Different missing/unsupported arguments in different APIs; not the same code-path defect." + "reason": "Different models and failures: a missing documented UNet argument versus an unexpected keyword argument in SD3 offloading/hooks." }, { "left": "issue:3250", "right": "issue:5178", "accept": false, - "reason": "Both are runtime failures, but one is a pipeline loading error and the other is a scheduler resolution-change bug." + "reason": "Both are pipeline/runtime issues, but one is about StableDiffusionKDiffusionPipeline not running and the other is a DPM-SDE resolution-change failure." }, { "left": "issue:3450", "right": "issue:347", "accept": false, - "reason": "One is a from_ckpt runtime issue; the other is a dependency installation/version resolution problem." + "reason": "One is a from_ckpt runtime regression; the other is an installation dependency resolution failure. Not the same bug." }, { "left": "issue:5339", "right": "issue:8752", "accept": false, - "reason": "Both concern loading local/offline models, but the reported errors and contexts are different." + "reason": "Both are loading problems, but one is a local safetensor loading complaint and the other is an offline from_single_file/SDXL resolution issue." }, { "left": "issue:6457", "right": "issue:8296", "accept": false, - "reason": "Both touch training/resume behavior, but one is about multi-GPU gradient accumulation and the other about checkpoint global-step bookkeeping." + "reason": "Both touch training bookkeeping, but one is gradient accumulation on multi-GPU SDXL-LoRA and the other is checkpoint global-step confusion across scripts." }, { "left": "issue:10428", "right": "issue:8429", "accept": false, - "reason": "Both are inference-script problems, but for different models and different failures." + "reason": "Different domains and code paths: Ascend NPU Flux inference versus an I2VGen-XL inference-script malfunction." }, { "left": "issue:3958", "right": "issue:4279", "accept": false, - "reason": "Both involve CivitAI LoRAs, but one is an offload interaction and the other is an SDXL LoRA loading failure." + "reason": "Both mention CivitAI LoRAs, but one is tied to enable_model_cpu_offload and the other to SDXL 0.9 loading failure; not clearly the same concrete bug." }, { "left": "issue:3194", "right": "issue:6870", "accept": false, - "reason": "Completely unrelated: bitsandbytes package metadata vs missing VAE tiling support." + "reason": "Completely different failures: bitsandbytes package metadata missing versus enable_vae_tiling missing in specific pipelines." }, { "left": "issue:4753", "right": "issue:9163", "accept": false, - "reason": "Both are training-script bugs, but they are different scripts with different tensor errors." + "reason": "Both are training-script bugs, but they affect different scripts and different tensor issues; not the same underlying defect." }, { "left": "issue:2581", "right": "issue:4488", "accept": false, - "reason": "Both concern checkpoint conversion/loading, but they target different model families and different workflows." + "reason": "One is about converting ControlNet checkpoints to diffusers; the other is SDXL checkpoint loading from single file in offline mode." }, { "left": "issue:4101", "right": "issue:4837", "accept": false, - "reason": "Both are conversion failures, but one is ControlNet conversion and the other is SDXL original-checkpoint conversion." + "reason": "Both involve conversion to diffusers, but one is a custom ControlNet conversion failure and the other is an SDXL conversion failure; too different to merge." }, { "left": "issue:12338", "right": "issue:13386", "accept": false, - "reason": "Both are tensor-related bugs, but one is VAE slicing decoding and the other is prior-preservation concatenation in a training script." + "reason": "Different subsystems and symptoms: VAE slicing/AutoencoderDC decoding versus a Qwen image Dreambooth prior-preservation concatenation crash." }, { "left": "issue:2163", "right": "issue:2410", "accept": false, - "reason": "Different training pipelines and different failures: dtype mismatch vs missing output artifacts." + "reason": "One is a dtype mismatch in a training example; the other is missing output artifacts in a Flax training script. Not the same issue." }, { "left": "issue:4376", "right": "issue:4664", "accept": false, - "reason": "Different SDXL loading paths and different feature failures: textual inversion vs LoRA loading." + "reason": "Both concern SDXL loading, but one is textual inversion crashing and the other is LoRA loading on a differently named pipeline class." }, { "left": "issue:5365", "right": "issue:9613", "accept": false, - "reason": "Both mention timestep-related errors, but the failures and affected code paths are different." + "reason": "Both are inference-time errors, but one is a num_inference_steps issue in txt2video and the other is a missing scale attribute on a custom Timesteps object." }, { "left": "issue:11732", "right": "issue:13137", "accept": false, - "reason": "Both involve LoRA loading, but one is a generic unpacking regression and the other is a LoKr-specific Flux 2 error." + "reason": "Both mention load_lora_weights, but one is a generic unpacking ValueError and the other is a Flux 2 Klein LoKr-specific loading error." }, { "left": "issue:1339", "right": "issue:3149", - "accept": false, - "reason": "Both mention nondeterminism, but the reports are too broad and do not point to the same concrete bug." + "accept": true, + "reason": "Same underlying symptom: nondeterministic outputs despite the same seed. #3149 is a version-specific restatement of #1339's determinism problem." }, { "left": "issue:2370", "right": "issue:4756", "accept": false, - "reason": "Unrelated training questions/features: GPU feasibility vs adding an SDXL training script." + "reason": "Different topics: hardware feasibility for LoRA training versus a request for an SDXL dreambooth-lora example script." }, { "left": "issue:4232", "right": "issue:8509", "accept": false, - "reason": "Different requests around controlnet/instruct-pix2pix and a separate controlnet++ issue." + "reason": "Both are feature requests around control conditions, but they concern different models and capabilities, not the same change." }, { "left": "issue:4904", "right": "issue:5957", "accept": false, - "reason": "Unrelated feature requests in different components: GLIGEN modularization vs LoRA-compatible convolution padding behavior." - }, - { - "left": "issue:5628", - "right": "issue:7240", - "accept": false, - "reason": "Scheduler offset behavior is unrelated to low-quality output during LDM conversion." + "reason": "Unrelated changes: modularizing GLIGEN versus fixing LoRACompatibleConv padding_mode behavior." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: the cluster is clearly heterogeneous, and the only soft edge with direct support from the titles is #1339/#3149. The other pair verdicts stay conservative and avoid overclaiming duplicate status from superficial title similarity. No mergeability issues since there are no PR pairs." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -89638,8 +89659,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -89664,7 +89685,6 @@ "issue:2495", "issue:2534", "issue:2800", - "issue:3730", "issue:4043", "issue:4149", "issue:4307", @@ -89674,6 +89694,7 @@ "issue:490", "issue:4911", "issue:5009", + "issue:5628", "issue:5984", "issue:6154", "issue:6297", @@ -89682,7 +89703,7 @@ "issue:6880", "issue:696", "issue:7066", - "issue:7165", + "issue:7240", "issue:7333", "issue:7616", "issue:8264", @@ -89693,6 +89714,7 @@ "issue:9858" ], "soft_pairs": [ + "issue:5628|issue:7240", "issue:2035|issue:490", "issue:2534|issue:5009", "issue:4043|issue:460", @@ -89715,181 +89737,180 @@ "issue:5984|issue:6297", "issue:11104|issue:13286", "issue:8264|issue:8880", - "issue:1811|issue:4307", - "issue:3730|issue:7165" + "issue:1811|issue:4307" ], - "prepared_review_unit_hash": "e39619a96f7a55603501ac538162db633932c5086090e05dafdb5020cb54acd9", + "prepared_review_unit_hash": "912c17f86674bc2552a5bfa5450ed278fd7d061316dacd9379815c372c1ed6e6", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33988, - "estimated_input_tokens": 8497, - "estimated_eval_tokens": 17250 + "serialized_chars": 34079, + "estimated_input_tokens": 8520, + "estimated_eval_tokens": 17296 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is not a true duplicate cluster: the issues span many unrelated Diffusers bugs/features (Flux, training notebooks, upscaling, ONNX, quantization, batching, etc.). None of the soft pairs look like the same underlying problem.", - "confidence": 0.96, - "canonical_issue_reason": "No canonical issue: the items cover distinct pipelines, errors, and feature requests rather than one shared bug.", + "summary": "This cluster is a loose grab-bag of unrelated Diffusers bugs/features (Flux, LoRA training, schedulers, offload, quantization, pipeline shape mismatches). None of the soft candidate pairs look like the same underlying bug or change, so all should be kept separate.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 9611 is the most concrete, self-contained regression report in the set: it names the exact Flux pipelines, the batching failure, and a version. The rest of the items are too disparate to serve as a better cluster anchor.", "canonical_pr_reason": null, - "best_issue_reason": "No single best issue; the pool is heterogeneous, so there is no representative duplicate target to anchor on.", + "best_issue_reason": "Issue 9611 is the strongest standalone issue to retain for follow-up because it is specific, reproducible, and actionable, with a clear pipeline scope and failure mode.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5628", + "right": "issue:7240", + "accept": false, + "reason": "Different problems: scheduler denoising offset vs low-quality ImageNet LDM conversion output." + }, { "left": "issue:2035", "right": "issue:490", "accept": false, - "reason": "Both are memory/quality complaints around upscaling vs general memory usage, but they describe different pipelines and failure modes." + "reason": "Both mention upscaling, but one is artifact quality and the other is an OOM/memory blowout; not the same bug." }, { "left": "issue:2534", "right": "issue:5009", "accept": false, - "reason": "Scheduler-selection error and a Wuerstchen error are unrelated model/path failures." + "reason": "Scheduler-switching error and Wuerstchen error are unrelated failures in different code paths." }, { "left": "issue:4043", "right": "issue:460", "accept": false, - "reason": "SDXL pooled-embed requirement is a model-input constraint, not the ONNX CPU/CUDA dependency issue." + "reason": "SDXL prompt-embed tensor requirements and ONNX Runtime dependency selection are unrelated issues." }, { "left": "issue:4668", "right": "issue:7616", "accept": false, - "reason": "Both mention bad images, but one is LoRA training artifacts and the other is low-denoise img2img artifacts; different causes and code paths." + "reason": "Both are image-artifact reports, but they involve different models, pipelines, and trigger conditions." }, { "left": "issue:10526", "right": "issue:10958", "accept": false, - "reason": "FP8 quantization TypeError and meta-tensor copy failure are distinct initialization/weight-loading problems." + "reason": "Flux FP8 quantization TypeError and CogView4 meta-tensor copy failure are distinct bugs in different subsystems." }, { "left": "issue:10518", "right": "issue:11083", "accept": false, - "reason": "A notebook script bug and a DDP stride-mismatch warning are unrelated." + "reason": "Training-notebook bug vs DDP stride-mismatch warning; no shared concrete code path." }, { "left": "issue:1434", "right": "issue:4692", "accept": false, - "reason": "Upscaling OOM and UNet memory concerns are both memory-related but not the same concrete bug." + "reason": "Both are memory-related, but one is StableDiffusionUpscalePipeline OOM and the other is a general UNet memory concern." }, { "left": "issue:1534", "right": "issue:696", "accept": false, - "reason": "Dreambooth notebook error and generic Dreambooth OOM are different training issues." + "reason": "Dreambooth notebook error and training OOM are different failure modes, not the same underlying bug." }, { "left": "issue:1643", "right": "issue:2459", "accept": false, - "reason": "Both involve xformers calls, but they refer to different training scripts and likely different API mismatches." + "reason": "Both reference xformers attention, but the reports are for different scripts and are not clearly the same concrete defect." }, { "left": "issue:6732", "right": "issue:9756", "accept": false, - "reason": "Documentation for a scheduler parameter and a missing notebook file are unrelated." + "reason": "A scheduler documentation issue and a missing notebook file are unrelated." }, { "left": "issue:11160", "right": "issue:11222", "accept": false, - "reason": "WAN I2V inference inconsistency and support for more WAN LoRAs are different feature/problem requests." + "reason": "WAN inference inconsistency and WAN LoRA support are different issues." }, { "left": "issue:6154", "right": "issue:9858", "accept": false, - "reason": "SDXL training on M2 and a KeyError on 'train' are unrelated notebook/script failures." + "reason": "SDXL LoRA training on M2 Mac and a KeyError:'train' report do not describe the same bug." }, { "left": "issue:10594", "right": "issue:11886", "accept": false, - "reason": "Multiple-control LoRA usage and FluxKontextPipeline issues are different pipeline-specific bugs." + "reason": "Both are Flux-related, but one is multi-control + LoRA behavior and the other is a FluxKontextPipeline issue." }, { "left": "issue:1556", "right": "issue:2495", "accept": false, - "reason": "Black images on CUDA and black images on MPS in upscale pipeline are superficially similar but platform-specific and not clearly the same bug." + "reason": "Black-image reports on CUDA vs MPS and different pipelines are not the same underlying failure." }, { "left": "issue:4911", "right": "issue:6880", "accept": false, - "reason": "Custom text encoder loading and Custom Diffusion retrieve.py are different loading/training issues." + "reason": "Custom text-encoder loading and Custom Diffusion retrieve.py are separate features/issues." }, { "left": "issue:7066", "right": "issue:8609", "accept": false, - "reason": "Batch img2img usage and custom hd_painter/IP-Adapter incompatibility are unrelated." + "reason": "Multi-image img2img support and a custom pipeline/IP-Adapter integration issue are unrelated." }, { "left": "issue:10121", "right": "issue:9611", "accept": false, - "reason": "High-resolution Flux1-dev generation and Flux batching errors are different Flux problems." + "reason": "Both are Flux issues, but one is high-resolution generation failure and the other is image batching in img2img/inpaint; different code paths." }, { "left": "issue:2800", "right": "issue:7333", "accept": false, - "reason": "DiffEdit semantic editing and an image-to-video rectification paper/topic are not duplicates." + "reason": "DiffEdit semantic editing and a tuning-free noise-rectification I2V topic are not the same bug/change." }, { "left": "issue:4149", "right": "issue:6570", "accept": false, - "reason": "SemanticStableDiffusion batch dimension mismatch and SDXL InstructPix2Pix tensor-size mismatch are separate shape bugs." + "reason": "Both mention tensor shape mismatches, but they affect different pipelines and conditioning flows." }, { "left": "issue:5984", "right": "issue:6297", "accept": false, - "reason": "Colab RAM issue in an SDXL LoRA notebook and inability to train on photos are distinct training complaints." + "reason": "Both touch train_dreambooth_lora_sdxl, but one is a RAM issue and the other is a training-quality/problem report." }, { "left": "issue:11104", "right": "issue:13286", "accept": false, - "reason": "Broken LTX-Video offload methods and torchao quantization support are different subsystems." + "reason": "LTX-Video offload breakage and torchao quantization support are unrelated." }, { "left": "issue:8264", "right": "issue:8880", "accept": false, - "reason": "Training stalled at 0% and inference slowdown across versions are different performance/stability issues." + "reason": "Training stuck at 0% and inference speed regression are different symptoms in different workflows." }, { "left": "issue:1811", "right": "issue:4307", "accept": false, - "reason": "Weight file selection behavior and local_files_only loading errors are different from_pretrained problems." - }, - { - "left": "issue:3730", - "right": "issue:7165", - "accept": false, - "reason": "Wrong inpainting blending timestep and JAX dynamic thresholding scale are unrelated algorithmic issues." + "reason": "Weight-download behavior and local_files_only error are separate from_pretrained issues with different causes." } ] }, @@ -89908,8 +89929,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -89926,11 +89947,11 @@ "issue:1367", "issue:1403", "issue:1486", - "issue:2319", "issue:3089", "issue:3431", "issue:3573", "issue:3593", + "issue:3730", "issue:4104", "issue:4484", "issue:4668", @@ -89945,7 +89966,7 @@ "issue:5703", "issue:6366", "issue:7061", - "issue:709", + "issue:7165", "issue:7440", "issue:7575", "issue:8323", @@ -89963,6 +89984,7 @@ "issue:9677" ], "soft_pairs": [ + "issue:3730|issue:7165", "issue:5153|issue:938", "issue:5412|issue:6366", "issue:5537|issue:9180", @@ -89985,190 +90007,189 @@ "issue:10695|issue:7061", "issue:10697|issue:7440", "issue:1403|issue:4668", - "issue:1486|issue:3089", - "issue:2319|issue:709" + "issue:1486|issue:3089" ], - "prepared_review_unit_hash": "f28184cd2d96dff11a98fd1613b81ee8d015c28a020308ac51698e9b32dc364a", + "prepared_review_unit_hash": "6551cf81b36b2e3c2db7e73d8da0241b171015c1c8212577d31f7c63356de059", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33846, - "estimated_input_tokens": 8462, - "estimated_eval_tokens": 17180 + "serialized_chars": 33824, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly unrelated single issues spanning different models, schedulers, and training scripts. The only plausible duplicate pair is the SDXL LoRA checkpoint-resume/load bug (5412 and 6366).", - "confidence": 0.68, - "canonical_issue_reason": "issue:6366 is the clearest representative of the overlapping SDXL DreamBooth LoRA checkpoint-loading/resume failure and covers both affected scripts.", + "summary": "Only one soft edge is a clear same-bug match: the SDXL DreamBooth LoRA checkpoint-resume/loading reports (5412 \u2194 6366). The rest are better treated as distinct issues; several are in the same general area (schedulers, LoRA/training, Flux/WAN) but do not look like the same underlying bug or change.", + "confidence": 0.84, + "canonical_issue_reason": "issue:6366 is the best canonical issue because it describes the broader checkpoint-loading/resume failure across both SDXL DreamBooth LoRA training scripts, making it the most representative root report.", "canonical_pr_reason": null, - "best_issue_reason": "issue:6366 has the broadest, most explicit description of the underlying checkpoint bug and best matches the accepted duplicate pair.", + "best_issue_reason": "issue:6366 is the strongest global issue candidate since it is broader and more actionable than the narrower resume-from-checkpoint report while covering the shared failure mode.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3730", + "right": "issue:7165", + "accept": false, + "reason": "Both are scheduler-related, but they describe different bugs in different implementations and do not look like the same fix." + }, { "left": "issue:5153", "right": "issue:938", "accept": false, - "reason": "Different features and domains: FreeU in diffusion U-Net vs a community human-motion pipeline." + "reason": "Different topics entirely: a U-Net optimization idea versus a community text-to-motion pipeline request." }, { "left": "issue:5412", "right": "issue:6366", "accept": true, - "reason": "Both report SDXL DreamBooth LoRA checkpoint/resume loading failures in the same training code path." + "reason": "These appear to report the same checkpoint-resume/loading problem in SDXL DreamBooth LoRA training, with 6366 being the broader description." }, { "left": "issue:5537", "right": "issue:9180", "accept": false, - "reason": "Notebook/template failure vs a pipeline execution-device attribute error; not the same bug." + "reason": "Different code paths and symptoms: a notebook not working versus a pipeline attribute error." }, { "left": "issue:5703", "right": "issue:8323", "accept": false, - "reason": "Different scheduler problems with no clear shared code path or symptom." + "reason": "Both are scheduler-adjacent, but the reported problems are distinct and not the same bug." }, { "left": "issue:9101", "right": "issue:9241", "accept": false, - "reason": "Training-script request vs LoRA conversion script; unrelated changes." + "reason": "Different functionality: an InstructPix2Pix training script versus a LoRA conversion script." }, { "left": "issue:5125", "right": "issue:7575", "accept": false, - "reason": "Both are runtime errors, but they affect different components and failure modes (xformers attention vs Transformer2DModel/adanorm)." + "reason": "Distinct failures in different model/components: xFormers CUDA error versus a Transformer2DModel shape error." }, { "left": "issue:3431", "right": "issue:4484", "accept": false, - "reason": "Both mention bad results with ControlNet/LoRA, but the models and failure descriptions are too different to be the same bug." + "reason": "Both concern poor image quality, but they involve different pipelines/models and different failure modes." }, { "left": "issue:4104", "right": "issue:8450", "accept": false, - "reason": "Generic black/noisy output vs a specific SDXL inpaint strength=1.0 issue; not enough evidence of one underlying defect." + "reason": "Both mention bad image output, but the titles point to different concrete problems and likely different fixes." }, { "left": "issue:3573", "right": "issue:3593", "accept": false, - "reason": "One is a local upscaler class_labels bug; the other is LoRA+xformers second-call failure." + "reason": "Local upscaler class-label handling is unrelated to a LoRA xformers second-call failure." }, { "left": "issue:4975", "right": "issue:5135", "accept": false, - "reason": "Performance regression in LoRA loading vs quality degradation during inference; different problems." + "reason": "Performance regression in LoRA loading versus LoRA quality degradation are related only loosely, not the same bug." }, { "left": "issue:8503", "right": "issue:8960", "accept": false, - "reason": "Gradient checkpointing crash vs nonfunctional gradient accumulation; unrelated training mechanics." + "reason": "Different training-time issues: gradient checkpointing crash versus gradient accumulation not working." }, { "left": "issue:1231", "right": "issue:1367", "accept": false, - "reason": "Scheduler implementation issue in Flax vs saving custom scheduler config; different code paths." + "reason": "Scheduler-related, but the titles indicate different bugs and different affected behavior." }, { "left": "issue:12609", "right": "issue:13311", "accept": false, - "reason": "WAN NaNs in latents vs Flux2 LoRA key-prefix mismatch; different model families and symptoms." + "reason": "Different model families and different failures: NaNs in latents versus LoRA key/prefix mismatch." }, { "left": "issue:5126", "right": "issue:5225", "accept": false, - "reason": "Conv2d forward signature error vs SDXL memory/VRAM limit issue; unrelated." + "reason": "A Conv2d forward signature error is not the same as an SDXL VRAM exhaustion issue." }, { "left": "issue:10540", "right": "issue:8898", "accept": false, - "reason": "Flux single-file loading regression vs from_pretrained subfolder path validation; not the same bug." + "reason": "Both are loading issues, but one is a from_single_file Flux regression and the other is from_pretrained path validation." }, { "left": "issue:11041", "right": "issue:9539", "accept": false, - "reason": "WAN group-offloading error vs CogVideoX float8 index_select CUDA limitation; different failures." + "reason": "Different models and code paths: WAN group offloading versus CogVideoX float8 indexing." }, { "left": "issue:9178", "right": "issue:9677", "accept": false, - "reason": "LoRA PEFT target-module error vs keeping text encoders in fp32; separate Flux training concerns." + "reason": "Both are Flux/LoRA-adjacent, but one is a PEFT target-module error and the other is about text encoder precision." }, { "left": "issue:10121", "right": "issue:9225", "accept": false, - "reason": "High-resolution Flux1-dev generation issue vs RGBA latent-size mismatch in SDXL InstructPix2Pix; different pipelines." + "reason": "Different pipelines and different symptoms: Flux high-resolution generation versus SDXL InstructPix2Pix latent-size mismatch." }, { "left": "issue:1037", "right": "issue:1140", "accept": false, - "reason": "Community pipeline request vs community hypernetwork topic; not duplicates." + "reason": "A community pipeline request and a community hypernetwork issue are not the same bug or change." }, { "left": "issue:10695", "right": "issue:7061", "accept": false, - "reason": "DDIMInverseScheduler timestep bug vs incorrect docstring; unrelated." + "reason": "A scheduler timestep calculation bug and a docstring correction are unrelated." }, { "left": "issue:10697", "right": "issue:7440", "accept": false, - "reason": "Random transform mismatch in training vs dataset availability issue; not the same underlying bug." + "reason": "Training-transform inconsistency and dataset availability are different kinds of problems." }, { "left": "issue:1403", "right": "issue:4668", "accept": false, - "reason": "Safety checker output mismatch vs Kohya-style LoRA weird images; too different to merge." + "reason": "Safety checker output differences are not the same as LoRA-generated weird images." }, { "left": "issue:1486", "right": "issue:3089", "accept": false, - "reason": "Different scheduler bugs (DDIM prediction_type sample vs DDPM bug) with no clear shared fix." - }, - { - "left": "issue:2319", - "right": "issue:709", - "accept": false, - "reason": "Both are docstring/typo-related, but they concern different methods and are not the same issue." + "reason": "Both are scheduler bugs, but they concern different schedulers and different calculations." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The conclusions are grounded in the packet and conservative overall. The only accepted soft pair, issue:5412 with issue:6366, is well supported by their titles describing the same SDXL DreamBooth LoRA checkpoint/resume loading failure. The remaining pair rejections are also appropriately cautious given the mismatch in titles and domains. The summary does not overclaim beyond the available evidence." + "accept": false, + "feedback": "Mostly conservative, but the writeup overstates the evidence in a few places. The summary says all non-matching pairs are \"cross-topic\" false positives, yet several are actually same-domain but different bugs (e.g., scheduler-related or LoRA/training-related), so that wording is too strong. A few pair reasons also rely on specifics not visible in the packet, so the grounding is not fully verifiable here. The 5412\u21946366 edge is the only clearly supported same-bug match; the rest should be described more narrowly as distinct issues rather than cross-topic false positives." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -90181,14 +90202,15 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:11144", "issue:1188", "issue:2276", + "issue:2319", "issue:2560", "issue:2663", "issue:2716", @@ -90216,13 +90238,13 @@ "issue:6112", "issue:6363", "issue:6640", + "issue:709", "issue:7168", "issue:761", "issue:7708", "issue:7766", "issue:785", "issue:7985", - "issue:801", "issue:835", "issue:8744", "issue:8885", @@ -90231,11 +90253,11 @@ "issue:9439", "issue:9539", "issue:9610", - "issue:964", "issue:9717", "issue:9813" ], "soft_pairs": [ + "issue:2319|issue:709", "issue:2663|issue:3761", "issue:2716|issue:3585", "issue:2776|issue:6363", @@ -90258,181 +90280,180 @@ "issue:2803|issue:579", "issue:4332|issue:7168", "issue:5967|issue:6112", - "issue:7708|issue:8744", - "issue:801|issue:964" + "issue:7708|issue:8744" ], - "prepared_review_unit_hash": "c167a4bc79730abadaa4f0e5897c6b38730ffdb25046523e5f8c3bfcc06a2da1", + "prepared_review_unit_hash": "b96c26914beb6997c5471ffc2fd6d5616c88c5d60abffab0e8e1b56c58b8b78c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33697, - "estimated_input_tokens": 8425, - "estimated_eval_tokens": 17106 + "serialized_chars": 33712, + "estimated_input_tokens": 8428, + "estimated_eval_tokens": 17112 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set; it groups many superficially similar diffusion/LoRA/IP-Adapter/turbo/loading topics that are actually different bugs, feature requests, and support questions. No soft edge looks like the same concrete underlying issue.", - "confidence": 0.93, - "canonical_issue_reason": "No strong canonical issue exists here because the items span unrelated problem types and code paths. If forced, issue 5226 is one of the clearest concrete bug reports, but it does not subsume the rest of the cluster.", + "summary": "No true duplicate group emerges here; the soft pairs mostly share broad diffusion/LoRA themes but describe different failure modes, code paths, or feature requests. I\u2019d treat this set as mostly unrelated issues with one representative bug report chosen for triage.", + "confidence": 0.29, + "canonical_issue_reason": "issue:9171 is the most concrete, actionable bug report in the set: it names a specific regression, exact API (`StableDiffusionPipeline.from_single_file()`), model family, and failure condition (`v_prediction`).", "canonical_pr_reason": null, - "best_issue_reason": "Issue 5226 is the most self-contained and actionable bug report among the set, but the cluster is too heterogeneous for it to serve as a true representative canonical issue.", + "best_issue_reason": "issue:9171 is the best overall issue candidate because it is narrowly scoped, reproducible, and tied to a clear code path; it\u2019s a stronger triage anchor than the more generic questions/feature requests.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2319", + "right": "issue:709", + "accept": false, + "reason": "Different problems: a typo in a method implementation vs a documentation/docstring issue." + }, { "left": "issue:2663", "right": "issue:3761", "accept": false, - "reason": "Both are loading-related, but one is a safety-checker file handling bug and the other is a multi-community-pipeline process issue; different code paths." + "reason": "One is a safety-checker safetensors loading failure; the other is loading two community pipelines in the same process." }, { "left": "issue:2716", "right": "issue:3585", "accept": false, - "reason": "A generic checkpoint-loading question is not the same as a specific LoRA synthesis failure." + "reason": "Checkpoint-loading help request vs LoRA synthesis failure; no shared bug mechanism." }, { "left": "issue:2776", "right": "issue:6363", "accept": false, - "reason": "Both concern training, but one is DeepSpeed/ZeRO setup and the other is FP16 gradient unscaling; unrelated failures." + "reason": "DreamBooth/DeepSpeed version compatibility on 8GB hardware vs FP16 gradient unscale error in LoRA training." }, { "left": "issue:3933", "right": "issue:7985", "accept": false, - "reason": "Both mention torch.compile, but one is a logging-side graph break while the other is an IP-Adapter incompatibility." + "reason": "Both mention compile-related issues, but one is logging breaking graph compile and the other is IP-Adapter incompatibility with `torch.compile`." }, { "left": "issue:4310", "right": "issue:4453", "accept": false, - "reason": "SDXL artifact quality and LoRA conversion advice are unrelated." + "reason": "SDXL scanline artifacts are an image-quality bug; the other is a conversion/how-to question about LoRA safetensors." }, { "left": "issue:5205", "right": "issue:6640", "accept": false, - "reason": "Both mention LoRA, but they target different pipelines and different functionality." + "reason": "ControlNet/Repaint with LoRA-textual inversion is a pipeline-composition request; Unet3DConditional LoRA is a different model integration task." }, { "left": "issue:5925", "right": "issue:9610", "accept": false, - "reason": "Both are LoRA-related feature requests, but they are different requests on different pipelines and not the same bug/change." + "reason": "ZipLora multi-LoRA enhancement request vs adding `load_lora_weights` to an SD3 img2img pipeline; related area, different change." }, { "left": "issue:761", "right": "issue:835", "accept": false, - "reason": "DreamBooth optimizer failure and flax numpy_to_pil output handling are unrelated." + "reason": "Different code paths: 8-bit Adam training failure vs flax pipeline image conversion failure." }, { "left": "issue:4593", "right": "issue:8885", "accept": false, - "reason": "Both involve loading LoRA/custom-diffusion assets, but one is a rank constraint error and the other is a Colab installation/modifier-token problem." + "reason": "A1111 LoRA rank constraint error is unrelated to clip-retrieval installation and modifier-token loading in the custom diffusion example." }, { "left": "issue:9178", "right": "issue:9439", "accept": false, - "reason": "Both are Flux-related, but one is a PEFT target-module training error and the other is inconsistent loading behavior." + "reason": "Both are Flux-related, but one is a LoRA PEFT target-module training error and the other is inconsistent outputs when loading the model differently." }, { "left": "issue:9539", "right": "issue:9717", "accept": false, - "reason": "One is a Float8 CUDA operator failure in CogVideoX; the other is a Flux docker-image error." + "reason": "Float8 CUDA op limitation in CogVideoX is unrelated to a Flux-dev docker image error." }, { "left": "issue:1188", "right": "issue:319", "accept": false, - "reason": "Different feature requests for different models; no shared underlying bug." + "reason": "Two separate feature requests for different model/pipeline additions, not the same change." }, { "left": "issue:2276", "right": "issue:3593", "accept": false, - "reason": "xformers installation on Colab is not the same as xformers misbehavior during a second LoRA pipeline call." + "reason": "xformers installation on a notebook vs second-call failure when enabling xformers with LoRA; different failure surface." }, { "left": "issue:2560", "right": "issue:3347", "accept": false, - "reason": "Different output-correctness issues in different pipelines with no clear shared root cause." + "reason": "List-of-generators nondeterminism in img2img is not the same as NaN image output from text prompts." }, { "left": "issue:3073", "right": "issue:5046", "accept": false, - "reason": "Separate model requests; one asks for consistency models, the other is a DragonDiffusion announcement." + "reason": "Consistency models request and DragonDiffusion request are distinct model-feature proposals." }, { "left": "issue:3672", "right": "issue:785", "accept": false, - "reason": "Thread-safety failure and scheduler performance cost are unrelated." + "reason": "Concurrent-thread failure vs scheduler performance cost; not the same bug." }, { "left": "issue:5226", "right": "issue:5516", "accept": false, - "reason": "Both are LoRA pipeline problems, but the pipelines and failure modes differ; not enough evidence of the same code-path bug." + "reason": "Both involve LoRA in adapter-like pipelines, but the concrete incompatibilities and affected pipelines differ." }, { "left": "issue:7766", "right": "issue:9813", "accept": false, - "reason": "A usage question about FaceID Plus is not the same as a batch-processing feature-mixing bug." + "reason": "Usage questions about IP-Adapter FaceID Plus vs a specific `prepare_ip_adapter_image_embeds` batch-mixing bug." }, { "left": "issue:11144", "right": "issue:9171", "accept": false, - "reason": "Flax dtype initialization and from_single_file v_prediction handling are different issues." + "reason": "Different models and failure modes: Flax UNet dtype initialization vs single-file loading regression for SD 1.5 v-prediction." }, { "left": "issue:2803", "right": "issue:579", "accept": false, - "reason": "Docs build failure on WSL2 and missing config.json for textual inversion are unrelated." + "reason": "Local docs build failure on WSL2 is unrelated to a missing `config.json` for textual inversion." }, { "left": "issue:4332", "right": "issue:7168", "accept": false, - "reason": "xformers on Kandinsky2.2 and IP-Adapter image-embed shape issues are different bugs." + "reason": "xformers support for Kandinsky 2.2 vs incorrect IP-Adapter image-embed tensor shape; different bugs." }, { "left": "issue:5967", "right": "issue:6112", "accept": false, - "reason": "An announcement and a downstream quality complaint about sdxl-turbo are not duplicate reports." + "reason": "An announcement and a user report of noisy output are not the same underlying issue." }, { "left": "issue:7708", "right": "issue:8744", "accept": false, - "reason": "Missing tokenizer config during model loading is unrelated to single-file training support." - }, - { - "left": "issue:801", - "right": "issue:964", - "accept": false, - "reason": "Both are inpainting-related, but one is about attention slicing behavior and the other is an example not working; insufficient evidence of the same bug." + "reason": "Missing tokenizer config for an SDXL base model vs a request to use single-file loading for advanced model training." } ] }, @@ -90451,8 +90472,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -90492,19 +90513,20 @@ "issue:4841", "issue:5138", "issue:5300", - "issue:5977", "issue:6535", "issue:6574", - "issue:6958", "issue:7009", "issue:7147", "issue:761", "issue:776", "issue:7889", + "issue:801", "issue:8588", - "issue:9044" + "issue:9044", + "issue:964" ], "soft_pairs": [ + "issue:801|issue:964", "issue:1062|issue:761", "issue:6535|issue:6574", "issue:4014|issue:457", @@ -90527,181 +90549,180 @@ "issue:3967|issue:4578", "issue:4171|issue:5300", "issue:447|issue:4496", - "issue:4708|issue:5300", - "issue:5977|issue:6958" + "issue:4708|issue:5300" ], - "prepared_review_unit_hash": "863f46c1b802f7e1304862ef9e21343e0f2b783294468379ab7cf9b817cf49fd", + "prepared_review_unit_hash": "961bd2a7d1a27d42463530b8f2f3b39bbe507b89c9f650d031e936ddb37fa390", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33020, - "estimated_input_tokens": 8255, - "estimated_eval_tokens": 16766 + "serialized_chars": 33012, + "estimated_input_tokens": 8253, + "estimated_eval_tokens": 16762 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster does not look like a true duplicate set: the items span unrelated bugs, feature requests, docs issues, and model-specific problems. The suggested soft pairs mostly share only generic terms (e.g. black images, docs, GGUF, training) and do not appear to describe the same underlying change or failure mode.", - "confidence": 0.97, - "canonical_issue_reason": "issue:10795 is the strongest standalone bug report here: it is detailed, open, and tied to a concrete code path (`from_single_file` + GGUF + `torch.compile`).", - "canonical_pr_reason": null, - "best_issue_reason": "issue:10795 is the most actionable and specific issue in the set, with enough detail to serve as the best representative if one must be chosen.", + "summary": "The cluster is not a true duplicate set; it contains many unrelated issue reports spanning different models, features, and bugs, with only superficial title similarity from the automatic soft matcher. No soft pair should be merged.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue fits this cluster: the items describe distinct problems across unrelated pipelines/models (GGUF loading, scheduler changes, watermark artifacts, docs, training, etc.).", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "issue:10795 is the strongest standalone issue to represent a cluster member because it is detailed, reproducible, and points to a specific code path (`from_single_file` + GGUF + `torch.compile`), but it is not a true canonical for the whole set.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:801", + "right": "issue:964", + "accept": false, + "reason": "Both are image-generation bugs, but one is about attention slicing and the other is about inpainting/image-to-image batch handling; different failure modes and likely different fixes." + }, { "left": "issue:1062", "right": "issue:761", "accept": false, - "reason": "Both are DreamBooth training complaints, but one is about reproducing paper results and the other about 8-bit Adam failure; different bugs." + "reason": "Both are DreamBooth/training related, but one is about reproducing paper results and the other is an 8-bit Adam training failure; not the same bug." }, { "left": "issue:6535", "right": "issue:6574", "accept": false, - "reason": "One is a feature request to change the scheduler; the other is a warning about SVD input range. Different problems." + "reason": "Both concern Stable Video Diffusion scheduling/behavior, but one asks about swapping the scheduler and the other reports a warning about input range; distinct issues." }, { "left": "issue:4014", "right": "issue:457", "accept": false, - "reason": "Both mention visual artifacts, but the pipelines, device contexts, and artifact types differ; not the same bug." + "reason": "Both mention visual artifacts, but one is SDXL watermark pixel artifacts and the other is MPS blue artifacts; different hardware/path and symptom." }, { "left": "issue:12047", "right": "issue:12085", "accept": false, - "reason": "Both involve GGUF, but one is LoRA fusion on Wan2.2 and the other is model loading on WanVACE; distinct code paths." + "reason": "Both are GGUF/lightx2v related, but one is LoRA fusion on Wan2.2 GGUF and the other is model loading for WanVACETransformer3DModel; different operations and code paths." }, { "left": "issue:13274", "right": "issue:9044", "accept": false, - "reason": "Completely different failures: AMD recursion error vs missing SDXL weight file." + "reason": "Both are model-loading/runtime errors, but one is an AMD recursion error on stable-audio-open and the other is an OSError about missing SDXL weight files; unrelated root causes." }, { "left": "issue:10795", "right": "issue:12096", "accept": false, - "reason": "`torch.compile` with GGUF loading and `apply_group_offloading` are separate runtime paths and failure modes." + "reason": "Both involve new model/pipeline support, but one is GGUF + `torch.compile` for transformer models and the other is group offloading in WanVACEPipeline; not the same defect." }, { "left": "issue:10209", "right": "issue:7009", "accept": false, - "reason": "Different model families and symptoms; one is a fine-tuning error, the other is poor generation quality on a text-to-video pipeline." + "reason": "The titles are different model-specific problems (`cogvideox-1.5-5b` and TextToVideoSDPipeline with ali-vilab model); no evidence they are the same bug." }, { "left": "issue:11083", "right": "issue:8588", "accept": false, - "reason": "DDP stride mismatch warning is unrelated to a missing conversion script." + "reason": "One is a DDP stride-mismatch warning during runtime; the other is a missing docs conversion script. Different subsystem and issue type." }, { "left": "issue:11389", "right": "issue:12437", "accept": false, - "reason": "Scheduler refactor request vs feature support request for a model; no shared bug/change." + "reason": "Scheduler API refactor request and OmniSVG support request are unrelated feature requests." }, { "left": "issue:11569", "right": "issue:12120", "accept": false, - "reason": "Both are model-usage questions, but one asks for a repo and the other asks about LoRA training; not duplicates." + "reason": "Both are about model/distilled-flux/ltxv usage, but one requests a repo for a specific distilled model and the other asks how to train a LoRA for distilled Flux; different asks." }, { "left": "issue:1614", "right": "issue:2179", "accept": false, - "reason": "Both mention black outputs, but they are on different frameworks/backends and likely different root causes." + "reason": "Both report black outputs, but one is Stable Diffusion 2.1 with autocast and the other is FlaxDDPMScheduler; different implementations and likely different bugs." }, { "left": "issue:1823", "right": "issue:2871", "accept": false, - "reason": "Depth2img comment swap is unrelated to latent output handling in inpainting." + "reason": "One is swapped comments in depth2img docs/code, the other is incorrect handling of `output_type='latent'` in inpainting; unrelated." }, { "left": "issue:1857", "right": "issue:776", "accept": false, - "reason": "Poor testing of UnCLIP is not the same as batch_size>1 image-to-image breakage." + "reason": "UnCLIP pipeline testing quality and batch_size>1 image-to-image breakage are unrelated issues." }, { "left": "issue:2416", "right": "issue:4841", "accept": false, - "reason": "Training slowdown with xformers vs ControlNet image-size issue are unrelated." + "reason": "Training slowdown with xformers and a ControlNet image-size issue are different performance/shape problems." }, { "left": "issue:2596", "right": "issue:2761", "accept": false, - "reason": "Both are documentation issues, but they concern different symbols and topics." + "reason": "Both are documentation-related, but one is a missing `load_image` reference and the other is documentation for `UNet1DModel.extra_in_channels`; different docs gaps." }, { "left": "issue:2803", "right": "issue:7147", "accept": false, - "reason": "Docs build on WSL2 and import failure for StableCascade are unrelated." + "reason": "WSL2 docs build failure and missing `StableCascadeDecoderPipeline` import are unrelated environment/import problems." }, { "left": "issue:3097", "right": "issue:3540", "accept": false, - "reason": "Different pipelines and API surfaces; no evidence of the same defect." + "reason": "Both concern prompt embeddings/conditioning, but one says `load_textual_inversion` does nothing while the other is `prompt_embeds`/`negative_prompt_embeds` not working in TextToVideoZeroPipeline; different pipeline behavior." }, { "left": "issue:3455", "right": "issue:5138", "accept": false, - "reason": "One is a training workflow request, the other is moving a helper function to training utils." + "reason": "One is a training recipe request for instruct-pix2pix; the other is a training utility refactor request. Not the same bug/change." }, { "left": "issue:3565", "right": "issue:7889", "accept": false, - "reason": "OOM during example training is not the same as `load_single_file_model_checkpoint` bugs." + "reason": "Text-to-image training OOM and `load_single_file_model_checkpoint` bugs are not the same failure mode or code path." }, { "left": "issue:3967", "right": "issue:4578", "accept": false, - "reason": "Branch/dependency breakage vs documentation typo; different issues." + "reason": "One is an SDXL branch/dependency issue; the other is a docs problem about VAE usage. No shared underlying bug." }, { "left": "issue:4171", "right": "issue:5300", "accept": false, - "reason": "Homepage reference-page rendering vs unwanted internet access; unrelated." + "reason": "Reference page rendering and unexpected internet connectivity are unrelated documentation/UI vs networking/configuration issues." }, { "left": "issue:447", "right": "issue:4496", "accept": false, - "reason": "Prompt truncation detection and prompt fusion are different prompt-related features, not the same bug." + "reason": "Prompt truncation detection and prompt fusion are both prompt-related, but they solve different problems and would not be merged into one fix." }, { "left": "issue:4708", "right": "issue:5300", "accept": false, - "reason": "Import error in `diffusers.utils` is unrelated to network connectivity behavior." - }, - { - "left": "issue:5977", - "right": "issue:6958", - "accept": false, - "reason": "LoRA key-saving behavior and motion model block configuration are unrelated problems." + "reason": "Import error in `diffusers.utils` and unexpected internet connections are unrelated." } ] }, @@ -90720,14 +90741,12 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:1037", - "issue:10569", - "issue:10901", "issue:11422", "issue:1750", "issue:2272", @@ -90747,6 +90766,7 @@ "issue:5661", "issue:5824", "issue:5924", + "issue:5977", "issue:6054", "issue:6196", "issue:6218", @@ -90754,6 +90774,7 @@ "issue:6323", "issue:6336", "issue:6791", + "issue:6958", "issue:699", "issue:7168", "issue:7378", @@ -90775,6 +90796,7 @@ "issue:9813" ], "soft_pairs": [ + "issue:5977|issue:6958", "issue:6196|issue:9495", "issue:7985|issue:9336", "issue:6336|issue:7378", @@ -90797,181 +90819,180 @@ "issue:5510|issue:5924", "issue:7421|issue:7946", "issue:11422|issue:8589", - "issue:5449|issue:6791", - "issue:10569|issue:10901" + "issue:5449|issue:6791" ], - "prepared_review_unit_hash": "03195c772ee8828c098f0a5e05daf81163ee48b210e9b40840d776c9c120a278", + "prepared_review_unit_hash": "f159f6c7f636f6a212a247969ed0e432ac9b96f261290920c72f3c1b4fab555c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34063, - "estimated_input_tokens": 8516, - "estimated_eval_tokens": 17288 + "serialized_chars": 34050, + "estimated_input_tokens": 8513, + "estimated_eval_tokens": 17282 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a single duplicate cluster; the issues span unrelated pipelines, models, training scripts, conversion tools, and runtime errors. All soft-edge pairs should be rejected.", - "confidence": 0.98, - "canonical_issue_reason": "No canonical issue: the items are heterogeneous and do not share one underlying bug or change path.", + "summary": "These items are mostly unrelated Diffusers issues spanning training, pipelines, compilation, schedulers, and utilities. None of the soft pairs look like the same underlying bug or change, so all soft edges are rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No clear best issue for duplicate anchoring because there is no cohesive problem cluster; the reports are mostly unrelated.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5977", + "right": "issue:6958", + "accept": false, + "reason": "Different bugs: LoRA checkpoint key naming vs a missing `use_linear_projection` setting in `unet_motion_model`." + }, { "left": "issue:6196", "right": "issue:9495", "accept": false, - "reason": "Different bugs in different IP-Adapter/ControlNet-related paths: ControlNetXS Depth SDXL loading vs SDXL PAG with IPAdapter." + "reason": "Different pipelines and failures: ControlNetXS depth loading vs SDXL PAG with IP-Adapter." }, { "left": "issue:7985", "right": "issue:9336", "accept": false, - "reason": "Both mention IP-Adapter, but one is a torch.compile compatibility issue and the other is requires_grad behavior in attention layers." + "reason": "Both involve IP-Adapter/compile, but one is a torch.compile incompatibility and the other is training parameters remaining `requires_grad=True`." }, { "left": "issue:6336", "right": "issue:7378", "accept": false, - "reason": "TensorRT VAE compilation with AutoencoderTiny is unrelated to AnimateDiff video-to-video memory usage." + "reason": "Unrelated: TensorRT VAE compilation with AutoencoderTiny vs AnimateDiff video-to-video latent decoding memory usage." }, { "left": "issue:5619", "right": "issue:9350", "accept": false, - "reason": "AutoencodeTiny img2img encoding bug and txt_ids deprecation are different code paths and symptoms." + "reason": "Different issues: AutoencodeTiny img2img encode behavior vs a deprecation warning about 3D `txt_ids` tensors." }, { "left": "issue:1750", "right": "issue:2578", "accept": false, - "reason": "Debian VM symbol resolution failure is unrelated to editable-install import/setup problems." + "reason": "Different failure modes: Debian VM symbol lookup error vs editable-install packaging problem." }, { "left": "issue:5645", "right": "issue:6275", "accept": false, - "reason": "LCM core support being broken is not the same as incomplete T2I-Adapter model coverage." + "reason": "LCM core behavior being broken is not the same as T2I-Adapters not supporting all official models." }, { "left": "issue:2272", "right": "issue:6323", "accept": false, - "reason": "Checkpoint-to-ONNX conversion failure and TensorRT img2img guidance_scale assignment error are distinct problems." + "reason": "Different code paths: checkpoint-to-ONNX conversion vs TensorRT img2img `guidance_scale` attribute handling." }, { "left": "issue:5661", "right": "issue:9034", "accept": false, - "reason": "Missing PreTrainedTokenizer symbol and an unexpected sentencepiece requirement are separate dependency/import issues." + "reason": "Missing `PreTrainedTokenizer` import is unrelated to an AutoPipeline pulling in `sentencepiece`." }, { "left": "issue:9309", "right": "issue:9398", "accept": false, - "reason": "Both touch Flux/embedding code, but the reported failures are different: ControlNetPipeline bug vs missing pooled_projection argument." + "reason": "FluxControlNet on `schnell` and a missing `pooled_projection` argument are separate model/runtime issues." }, { "left": "issue:7168", "right": "issue:9813", "accept": false, - "reason": "One is a user question about IP-Adapter image embeds, the other is a specific batch feature-mixing bug in prepare_ip_adapter_image_embeds." + "reason": "Both are IP-Adapter related, but one is about incorrect image embeds and the other about batch feature mixing in `prepare_ip_adapter_image_embeds`." }, { "left": "issue:4375", "right": "issue:8518", "accept": false, - "reason": "Both involve torch.compile, but they are different crashes in different models and code paths." + "reason": "Both mention `torch.compile`, but they describe different crashes in different models/code paths." }, { "left": "issue:2774", "right": "issue:4888", "accept": false, - "reason": "ControlNet safetensors conversion failure is unrelated to offline safetensors loading behavior." + "reason": "ControlNet safetensors conversion failure is unrelated to fully offline safetensors model loading." }, { "left": "issue:4297", "right": "issue:6218", "accept": false, - "reason": "Both are batch-related img2img issues, but fishnet artifacts/watermarks are not the same as batched seed handling." + "reason": "Different batching bugs: SDXL img2img artifacts vs batched seeds handling in ControlNet/img2img pipelines." }, { "left": "issue:5233", "right": "issue:5824", "accept": false, - "reason": "Wrong callback step value and a bad default for text_encoder_projection_dim are unrelated parameter bugs." + "reason": "Callback step value mismatch is distinct from the SDXL `text_encoder_projection_dim` default issue." }, { "left": "issue:5447", "right": "issue:7848", "accept": false, - "reason": "Denoising_start with 2nd-order schedulers is a different issue from lpw_stable_diffusion_xl decoding." + "reason": "Different pipelines and symptoms: SDXL img2img scheduler/denoising-start bug vs a community LPW decoding issue." }, { "left": "issue:7629", "right": "issue:8609", "accept": false, - "reason": "SVD OOM on PyTorch 1.13.1 and custom_pipeline/IP_adapter incompatibility are unrelated." + "reason": "Unrelated: SVD OOM on PyTorch 1.13.1 vs custom pipeline/IP-Adapter incompatibility." }, { "left": "issue:6054", "right": "issue:8879", "accept": false, - "reason": "EulerDiscreteScheduler torch.from_numpy misuse and torch.compile+cudagraph failure are different runtime problems." + "reason": "Euler scheduler NumPy conversion misuse is not the same as a torch.compile + cudagraph failure." }, { "left": "issue:1037", "right": "issue:699", "accept": false, - "reason": "A community pipeline request is not the same as adding negative prompt support." + "reason": "Community pipeline addition and negative prompt support are different feature requests." }, { "left": "issue:4176", "right": "issue:4912", "accept": false, - "reason": "Mandatory authentication and SDXL checkpoint conversion are unrelated feature/UX requests." + "reason": "Mandatory authentication is unrelated to converting custom CivitAI SDXL checkpoints to diffusers." }, { "left": "issue:5510", "right": "issue:5924", "accept": false, - "reason": "Dtype mismatch in InstructPix2Pix is unrelated to a VAE memory spike." + "reason": "Different runtime problems: dtype mismatch in SDXL InstructPix2Pix vs VAE memory spike." }, { "left": "issue:7421", "right": "issue:7946", "accept": false, - "reason": "Seed handling in KDiffusion and custom_pipeline caching are separate issues." + "reason": "Seed handling in K-Diffusion is unrelated to custom pipeline caching." }, { "left": "issue:11422", "right": "issue:8589", "accept": false, - "reason": "HiDream LoRA validation-prompt crash and SD3 DreamBooth prior-preservation support are different training concerns." + "reason": "Different training bugs: HiDream LoRA validation prompt crash vs SD3 DreamBooth prior preservation support." }, { "left": "issue:5449", "right": "issue:6791", "accept": false, - "reason": "Same example script, but one is a multiprocessing pickling bug while the other is a vague can't-run report; not clearly the same defect." - }, - { - "left": "issue:10569", - "right": "issue:10901", - "accept": false, - "reason": "High CPU memory usage in HunyuanVideo and wrong video output with negative_prompt are unrelated HunyuanVideo problems." + "reason": "Pickling `preprocess_train` in a training script is not the same as `train_text_to_image_sdxl.py` failing to run." } ] }, @@ -90990,17 +91011,20 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10100", "issue:10512", + "issue:10569", + "issue:10901", "issue:11109", "issue:11144", "issue:1814", "issue:2228", + "issue:2459", + "issue:2722", "issue:2782", "issue:2783", "issue:2822", @@ -91022,7 +91046,6 @@ "issue:6176", "issue:6354", "issue:6393", - "issue:6560", "issue:6593", "issue:7053", "issue:7312", @@ -91030,14 +91053,12 @@ "issue:7770", "issue:7982", "issue:8451", - "issue:8801", "issue:8834", "issue:8871", "issue:9038", "issue:9052", "issue:9332", "issue:9511", - "issue:9628", "issue:9648", "issue:988", "issue:9930", @@ -91045,6 +91066,7 @@ "issue:9958" ], "soft_pairs": [ + "issue:10569|issue:10901", "issue:1814|issue:2782", "issue:2822|issue:988", "issue:2858|issue:4222", @@ -91067,181 +91089,180 @@ "issue:5963|issue:9648", "issue:4314|issue:6393", "issue:11144|issue:7982", - "issue:6560|issue:9628", - "issue:10100|issue:8801" + "issue:2459|issue:2722" ], - "prepared_review_unit_hash": "5be6558fbd33993a526d918e9928d3a52158038ea2c733b060cdc8e3fcdd033b", + "prepared_review_unit_hash": "12179fb52e47941d74b87dadc66c911e4c2605465a3e4058a36ac68e95973a3a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33561, - "estimated_input_tokens": 8391, - "estimated_eval_tokens": 17038 + "serialized_chars": 33650, + "estimated_input_tokens": 8413, + "estimated_eval_tokens": 17082 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The items are mostly unrelated feature requests and bug reports; none of the soft pairs look like the same underlying issue, so I would not consolidate them into a duplicate cluster.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "This cluster is not a real duplicate set: it mixes unrelated bugs and feature requests across LoRA loading, HunyuanVideo, ControlNet, schedulers, Flax/JAX, and training scripts. No soft pair looks like the same underlying issue closely enough to merge.", + "confidence": 0.97, + "canonical_issue_reason": "No true duplicate canonical stands out here; if a single representative is needed, issue 9332 is the clearest standalone bug report because it is open, detailed, and reproducible.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 9332 is the strongest self-contained report in the set: it describes a concrete loading failure with enough detail to be actionable.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10569", + "right": "issue:10901", + "accept": false, + "reason": "Different HunyuanVideo problems: CPU memory use vs wrong output when using negative_prompt." + }, { "left": "issue:1814", "right": "issue:2782", "accept": false, - "reason": "Different problems: changing CLIP models versus a sequential offloading error in pipeline loading." + "reason": "Unrelated requests: CLIP model support vs a sequential offloading error." }, { "left": "issue:2822", "right": "issue:988", "accept": false, - "reason": "Training convergence quality issue is not the same as an FSDP training-support request." + "reason": "Different training concerns: convergence quality vs FSDP training setup." }, { "left": "issue:2858", "right": "issue:4222", "accept": false, - "reason": "Img2img seeding support and low-resolution dataset training advice are unrelated." + "reason": "Different img2img/training contexts: seeding image support vs low-resolution dataset training." }, { "left": "issue:3418", "right": "issue:4756", "accept": false, - "reason": "Both concern training scripts, but they request different script features for different workflows." + "reason": "Both involve LoRA training, but one asks for text-encoder training code and the other for SDXL style training support; not the same bug or change." }, { "left": "issue:3808", "right": "issue:7770", "accept": false, - "reason": "DeepFloyd IF training failure is a different code path from a LoRA tensor shape mismatch during loading." + "reason": "Different failure modes: DeepFloyd IF stage 2 training vs tensor shape mismatch when loading model and LoRA." }, { "left": "issue:4194", "right": "issue:5339", "accept": false, - "reason": "Both mention loading, but one is a VAE-from-single-file bug and the other is a local safetensor load problem." + "reason": "Both are loading-related, but one is VAE handling in from_single_file and the other is local safetensor loading failure; different code paths." }, { "left": "issue:4578", "right": "issue:7519", "accept": false, - "reason": "Documentation complaint versus an xFormers attention processor behavior issue." + "reason": "Documentation issue vs xformers attention-processor behavior; not the same bug." }, { "left": "issue:6354", "right": "issue:9511", "accept": false, - "reason": "LoRA support for ControlNet and multi-controlnet batching are distinct feature requests." + "reason": "Feature request for ControlNet LoRA support vs batching support for SDXL ControlNet inpaint; different scope and problem." }, { "left": "issue:9052", "right": "issue:9332", "accept": false, - "reason": "Both are model-loading bugs, but the failing paths and symptoms differ enough that they do not look like one defect." + "reason": "Both are model-loading bugs, but one is from_single_file failing on some checkpoints and the other is from_pretrained loading the wrong weights; different API paths and symptoms." }, { "left": "issue:8871", "right": "issue:9930", "accept": false, - "reason": "FP16 gradient unscale error during LoRA training is unrelated to PAG adaptive scale behavior." + "reason": "Different training bugs: FP16 unscale error during LoRA training vs PAG adaptive scale behavior." }, { "left": "issue:5110", "right": "issue:6593", "accept": false, - "reason": "HF repo-id validation failure is not the same issue as adding load_lora_weights support to an img2img pipeline." + "reason": "Repo-id validation error is unrelated to adding load_lora_weights support in SDXL img2img." }, { "left": "issue:6176", "right": "issue:9941", "accept": false, - "reason": "LoRA weight loading after DreamBooth training and a Colab runtime error are different reports." + "reason": "LoRA weight loading after DreamBooth training vs a general Colab runtime error; not the same underlying bug." }, { "left": "issue:10512", "right": "issue:11109", "accept": false, - "reason": "Quanto Flux LoRA load failure is unrelated to a depth LoRA transformer-layer mismatch." + "reason": "Different LoRA issues: Quanto Flux LoRA loading vs depth LoRA transformer-layer mismatch." }, { "left": "issue:2783", "right": "issue:4675", "accept": false, - "reason": "New pipeline creation request versus SDXL reference-only ControlNet support are different changes." + "reason": "Different ControlNet feature requests: a new img2img/inpainting pipeline vs SDXL reference-only support." }, { "left": "issue:4922", "right": "issue:8834", "accept": false, - "reason": "Android demo request is unrelated to SD3 ControlNet training-code release." + "reason": "Unrelated requests: Android demo vs SD3 ControlNet training code release." }, { "left": "issue:4969", "right": "issue:8451", "accept": false, - "reason": "CrossAttention import failure does not match a question about frozen UNet parameters and missing training script details." + "reason": "ImportError for CrossAttention is unrelated to training-script documentation about frozen UNet parameters." }, { "left": "issue:7053", "right": "issue:9038", "accept": false, - "reason": "Class-conditioning support in UNet3D is a different code path from Flax prompt-weight handling." + "reason": "Different model families and behaviors: UNet3D class conditioning vs prompt weighting in FlaxStableDiffusionPipeline." }, { "left": "issue:7312", "right": "issue:9958", "accept": false, - "reason": "Direct reward finetuning support and VAE temporal-decoder tiling are unrelated changes." + "reason": "AlignProp direct reward finetuning is unrelated to AutoencoderKLTemporalDecoder tiling support." }, { "left": "issue:2228", "right": "issue:5628", "accept": false, - "reason": "Pipeline GPU-loading failure is not the same as LMS scheduler Karras-sigma offset behavior." + "reason": "Pipeline GPU placement issue vs scheduler denoising offset bug; different code paths." }, { "left": "issue:5963", "right": "issue:9648", "accept": false, - "reason": "Kandinsky embeds-vs-prompts failure and VAE tiling memory overflow are different bugs." + "reason": "Kandinsky embeds/prompt handling is unrelated to VAE tiling memory overflow." }, { "left": "issue:4314", "right": "issue:6393", "accept": false, - "reason": "Interrupted download write-protection and private-link from_single_file failure are different download issues." + "reason": "Interrupted-download write protection vs private HF link failure in from_single_file." }, { "left": "issue:11144", "right": "issue:7982", "accept": false, - "reason": "Incorrect Flax UNet dtypes and from_pipe MultiControlNet handling are unrelated problems." - }, - { - "left": "issue:6560", - "right": "issue:9628", - "accept": false, - "reason": "LoRA shape-size regression and FP16 unscale error are separate training failures." + "reason": "Flax dtype initialization bug vs from_pipe with MultiControlNetModel; different subsystems and symptoms." }, { - "left": "issue:10100", - "right": "issue:8801", + "left": "issue:2459", + "right": "issue:2722", "accept": false, - "reason": "Support requests for different models (ConsisID vs Kolors) are not duplicates." + "reason": "LoRA xformers training issue vs ControlNet conversion mismatch; not the same underlying problem." } ] }, @@ -91260,11 +91281,12 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10100", "issue:10194", "issue:10427", "issue:1084", @@ -91286,14 +91308,12 @@ "issue:2785", "issue:3296", "issue:3381", - "issue:3428", "issue:3551", "issue:3570", "issue:3740", "issue:4073", "issue:4236", "issue:4349", - "issue:4446", "issue:4488", "issue:4615", "issue:4977", @@ -91304,17 +91324,20 @@ "issue:5710", "issue:6154", "issue:6256", - "issue:746", + "issue:6560", "issue:7498", "issue:7548", - "issue:8228", "issue:8405", "issue:843", "issue:8514", + "issue:8801", "issue:8832", + "issue:9628", "issue:9804" ], "soft_pairs": [ + "issue:6560|issue:9628", + "issue:10100|issue:8801", "issue:10194|issue:12695", "issue:10427|issue:12308", "issue:1084|issue:843", @@ -91336,182 +91359,180 @@ "issue:2560|issue:3551", "issue:4488|issue:4977", "issue:5628|issue:5710", - "issue:7548|issue:8832", - "issue:3428|issue:746", - "issue:4446|issue:8228" + "issue:7548|issue:8832" ], - "prepared_review_unit_hash": "f22434898d40be67495ae5a3df641d838fca7e4a486f12c5c77eea5c20e15a74", + "prepared_review_unit_hash": "87c38422b7ae7943f37372f5751ec540802d94cd9e8ae897c9321a82bea14cb0", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33766, - "estimated_input_tokens": 8442, - "estimated_eval_tokens": 17140 + "serialized_chars": 33712, + "estimated_input_tokens": 8428, + "estimated_eval_tokens": 17112 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft-edge pair looks like the same underlying bug/change; the set is mostly a loose collection of related-but-distinct issues across different models, pipelines, and failure modes. No PRs are present.", - "confidence": 0.83, - "canonical_issue_reason": "issue:12505 is the most concrete, technically specific open bug report in the set, with a narrow failure mode that makes it the cleanest anchor.", + "summary": "This cluster is mostly a grab bag of unrelated feature requests and bug reports around different diffusion models, pipelines, training scripts, and loading paths. The soft pairs share only broad keywords like \"support\", \"load\", or \"error\", not the same concrete bug/change, so none should be merged as duplicates.", + "confidence": 0.9, + "canonical_issue_reason": "issue:9804 is the clearest concrete, self-contained bug report in the set: it names a specific API (`ControlNetModel.from_pretrained`), a specific failure mode, and a specific model class (SDXL ControlNet local loading).", "canonical_pr_reason": null, - "best_issue_reason": "issue:12505 is the best representative issue because it describes a focused, reproducible code-path problem rather than a broad feature request or vague compatibility report.", + "best_issue_reason": "issue:9804 is the best representative issue because it is specific, actionable, and tied to a precise code path; several others are either broader feature requests or more ambiguous reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6560", + "right": "issue:9628", + "accept": false, + "reason": "Both involve LoRA/training stability, but they describe different scripts, symptoms, and failure modes; not the same underlying bug." + }, + { + "left": "issue:10100", + "right": "issue:8801", + "accept": false, + "reason": "Both are model-support requests, but for different architectures and different work items." + }, { "left": "issue:10194", "right": "issue:12695", "accept": false, - "reason": "Different model support requests: Flux attention masking vs Hunyuan v1.5 support." + "reason": "Both are \"support\" requests, but they target unrelated features/models and do not point to one shared change." }, { "left": "issue:10427", "right": "issue:12308", "accept": false, - "reason": "Different features in different areas: AuraFlow GGUF loading vs flash-attn GPU kernel support." + "reason": "One is GGUF loading for AuraFlow; the other is flash-attn kernel support on non-Hopper GPUs. Different problems." }, { "left": "issue:1084", "right": "issue:843", "accept": false, - "reason": "Both are documentation-related, but one is about weight reuse across pipelines and the other is prompt syntax docs." + "reason": "Both are documentation-related, but they cover different topics and would not be fixed by one PR." }, { "left": "issue:12505", "right": "issue:9804", "accept": false, - "reason": "Different SDXL loading bugs: tgate error in img2img pipelines vs local ControlNet loading failure." + "reason": "Both mention pipeline/loading errors, but one is a tgate/img2img bug and the other is ControlNet local loading; different code paths." }, { "left": "issue:2785", "right": "issue:6256", "accept": false, - "reason": "Both are pipeline feature requests, but they target unrelated capabilities (M1/M2 support vs callback hook)." + "reason": "M1/M2 support and callback API support are unrelated feature requests." }, { "left": "issue:3570", "right": "issue:4615", "accept": false, - "reason": "Unrelated loading/import failures: checkpoint download suppression vs missing ControlNet pipeline import." + "reason": "One is about disabling downloads when loading checkpoints; the other is an import/export symbol issue. Different bugs." }, { "left": "issue:3740", "right": "issue:4236", "accept": false, - "reason": "Both mention encoder/attention registration, but one is accelerate device_map conversion and the other is SDXL refiner text-encoder registration." + "reason": "Different failures: device_map conversion with accelerate vs SDXL refiner text-encoder registration." }, { "left": "issue:1566", "right": "issue:3296", "accept": false, - "reason": "Different training checkpoint bugs: fp16 output crash after saving vs checkpoint save/load failure with text_encoder." + "reason": "Both involve checkpoint save/load in Dreambooth, but the specific failure conditions differ enough that they are not the same bug." }, { "left": "issue:2772", "right": "issue:4349", "accept": false, - "reason": "Different pipeline concerns: dataset path validation in ControlNet training vs missing text_encoder in UpscalePipeline." + "reason": "A local dataset validation issue in a training script is not the same as missing text_encoder handling in an upscale pipeline." }, { "left": "issue:3381", "right": "issue:5399", "accept": false, - "reason": "Both are inpainting-related, but one is discoloration in batch inference and the other is a dtype mismatch in custom models." + "reason": "Both touch inpainting/dtype behavior, but one is batch discoloration and a docs concern, while the other is a custom-model matmul dtype mismatch." }, { "left": "issue:10954", "right": "issue:13203", "accept": false, - "reason": "Both are LoRA-loading issues, but they involve different model families and likely different code paths." + "reason": "Both are LoRA-related, but they concern different models and different load failures; too broad to merge." }, { "left": "issue:11644", "right": "issue:8514", "accept": false, - "reason": "Different failure modes: LTXConditionPipeline init error vs EMAModel parallel fine-tuning error." + "reason": "Different pipeline/training errors with different triggers and components; no shared fix is evident." }, { "left": "issue:11830", "right": "issue:13105", "accept": false, - "reason": "Both are output-quality problems, but they affect different video models and symptoms (black output vs noise)." + "reason": "Both are output-quality problems, but one is black output under float16 and the other is noise in a different model; not the same bug." }, { "left": "issue:1407", "right": "issue:1546", "accept": false, - "reason": "Different regression types: ONNX feature_extractor error vs truncation warning causing batched-input failure." + "reason": "An ONNX feature_extractor regression and a truncation warning batch-input error are distinct issues." }, { "left": "issue:2716", "right": "issue:5693", "accept": false, - "reason": "Generic loading problems, but one is checkpoint loading and the other is VAE loading." + "reason": "General checkpoint-loading help and a VAE loading error are not the same underlying issue." }, { "left": "issue:4073", "right": "issue:8405", "accept": false, - "reason": "Both are runtime tensor errors, but they arise in different subsystems (safety checker vs multiple ControlNets)." + "reason": "Safety-checker dtype error vs multi-ControlNet tensor-size mismatch: different subsystems and failure modes." }, { "left": "issue:6154", "right": "issue:7498", "accept": false, - "reason": "Both are Apple Silicon training issues, but they are different scripts, models, and hardware generations." + "reason": "Both are Apple-silicon training issues, but they affect different models/scripts and do not share one concrete bug." }, { "left": "issue:1677", "right": "issue:5325", "accept": false, - "reason": "No substantive overlap: pipeline dtype misuse vs packaging type distribution." + "reason": "Wrong data type when calling a pipeline is unrelated to packaging type information distribution." }, { "left": "issue:2560", "right": "issue:3551", "accept": false, - "reason": "Different bugs: generator-list determinism in img2img vs LoRA/xFormers interaction." + "reason": "Different regressions: generator-list determinism vs LoRA disabling xFormers." }, { "left": "issue:4488", "right": "issue:4977", "accept": false, - "reason": "Different code paths entirely: SDXL checkpoint conversion/loading vs checkpoint cleanup in an example script." + "reason": "Checkpoint conversion/loading behavior and checkpoint cleanup in a training example are unrelated." }, { "left": "issue:5628", "right": "issue:5710", "accept": false, - "reason": "Both scheduler-related, but one is Karras sigma offset behavior and the other is missing sigmas initialization." + "reason": "The former is a denoising offset bug; the latter is missing sigma initialization. Related area, but not the same code-path problem." }, { "left": "issue:7548", "right": "issue:8832", "accept": false, - "reason": "Different output problems: AnimateDiff flicker vs a generic pipeline crash on image indexing." - }, - { - "left": "issue:3428", - "right": "issue:746", - "accept": false, - "reason": "Different img2img issues: unexpected behavior when changing the pipeline vs ONNX revision loading." - }, - { - "left": "issue:4446", - "right": "issue:8228", - "accept": false, - "reason": "Both involve from_single_file text-encoder handling, but one is custom text_encoder support and the other is a CLIPTextModel load regression." + "reason": "AnimateDiff flicker and a pipeline crash on image access are different issues." } ] }, @@ -91530,8 +91551,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -91543,22 +91564,22 @@ "issue:11828", "issue:12437", "issue:1655", - "issue:1896", "issue:2410", "issue:2444", "issue:2659", "issue:2784", + "issue:3428", "issue:3455", "issue:3679", "issue:3944", "issue:3974", "issue:4232", "issue:4405", + "issue:4446", "issue:4484", "issue:4649", "issue:4672", "issue:4803", - "issue:4972", "issue:5392", "issue:5485", "issue:5510", @@ -91569,22 +91590,23 @@ "issue:6335", "issue:6441", "issue:6685", - "issue:713", "issue:7177", "issue:7456", + "issue:746", "issue:7658", + "issue:8228", "issue:8434", "issue:8450", - "issue:8507", "issue:8708", "issue:9067", "issue:9223", "issue:9398", "issue:9476", - "issue:9933", - "issue:9972" + "issue:9933" ], "soft_pairs": [ + "issue:3428|issue:746", + "issue:4446|issue:8228", "issue:11828|issue:9223", "issue:2659|issue:5485", "issue:4484|issue:8450", @@ -91606,189 +91628,180 @@ "issue:6441|issue:7177", "issue:6685|issue:8708", "issue:7456|issue:9398", - "issue:7658|issue:8434", - "issue:8507|issue:9972", - "issue:1896|issue:4972", - "issue:2444|issue:713" + "issue:7658|issue:8434" ], - "prepared_review_unit_hash": "19249aa485389ea85ee4c6729f2fe72cc4d689fbee74548fe19ab5edcb756534", + "prepared_review_unit_hash": "a2c11e95ea470399ef8111decd0b76798838384ee4cdb2520cf89fc565e5e74e", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 34435, - "estimated_input_tokens": 8609, - "estimated_eval_tokens": 17474 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33668, + "estimated_input_tokens": 8417, + "estimated_eval_tokens": 17090 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated pipeline bugs, training-script errors, dependency/install issues, and feature requests. I would not merge any of the soft pairs as duplicates.", - "confidence": 0.93, - "canonical_issue_reason": "No single canonical issue stands out; the items span unrelated bugs and feature requests across different pipelines, training scripts, and environments.", + "summary": "This cluster is heterogeneous: it mixes unrelated bug reports, feature requests, training issues, install problems, and pipeline-specific regressions. The soft pairs mostly share only broad Diffusers vocabulary, not the same underlying bug or change, so none should be merged.", + "confidence": 0.94, + "canonical_issue_reason": "No single canonical issue stands out; the items cover many unrelated pipelines, models, and request types, so this cluster should not be collapsed to one issue.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a good global representative because the cluster is not centered on one underlying problem.", + "best_issue_reason": "Issue 10381 is the broadest concrete bug report in the set (Flux quantization/LoRA inference), but it is still not a true umbrella for the rest of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3428", + "right": "issue:746", + "accept": false, + "reason": "Different problems: one is an unexpected img2img pipeline result, the other is an ONNX revision loading issue." + }, + { + "left": "issue:4446", + "right": "issue:8228", + "accept": false, + "reason": "Both mention from_single_file/text encoder loading, but one is about custom text_encoder override support while the other is a CLIPTextModel load failure." + }, { "left": "issue:11828", "right": "issue:9223", "accept": false, - "reason": "Unrelated: one is a ControlNet pipeline type bug, the other is a Python package dependency conflict." + "reason": "Completely unrelated: a type error in ControlNet conditioning arguments versus a dependency/version conflict on pyarrow." }, { "left": "issue:2659", "right": "issue:5485", "accept": false, - "reason": "Different problems: batch-size nondeterminism vs ControlNet batch inference performance." + "reason": "Different symptoms and scope: batch-size-related output variation versus ControlNet batch inference slowness." }, { "left": "issue:4484", "right": "issue:8450", "accept": false, - "reason": "Both mention odd image output, but they affect different SDXL/inpaint code paths and are not the same bug." + "reason": "Both concern image quality, but one is about SDXL Kohya LoRA training artifacts and the other is an inpainting noise regression at strength=1.0." }, { "left": "issue:10381", "right": "issue:10446", "accept": false, - "reason": "Both are Flux-related, but one is quantization/LoRA inference and the other is multi-GPU sharding with ControlNet." + "reason": "Different code paths: Flux quantization/LoRA inference error versus multi-GPU/model-sharding incompatibility with FluxControlNetPipeline." }, { "left": "issue:10893", "right": "issue:11700", "accept": false, - "reason": "Different features: CogVideoX ControlNet integration vs Pruna loading integration." + "reason": "Both are feature requests, but they target different integrations: CogvideoX ControlNet support versus Pruna loading integration." }, { "left": "issue:11032", "right": "issue:12437", "accept": false, - "reason": "Completely unrelated: TRT quantization backend support vs OmniSVG support." + "reason": "Unrelated requests: TensorRT backend support versus OmniSVG model support." }, { "left": "issue:11828", "right": "issue:9476", "accept": false, - "reason": "Different errors in different models/pipelines; no shared concrete code path." + "reason": "Different failures: a ControlNet argument type bug versus a Flux LoRA validation dtype mismatch." }, { "left": "issue:1655", "right": "issue:617", "accept": false, - "reason": "One is a pipeline initialization failure, the other is an NNPACK hardware warning in AutoencoderKL." + "reason": "Different issues entirely: pipeline initialization failure versus an NNPACK hardware warning from AutoencoderKL." }, { "left": "issue:2410", "right": "issue:2444", "accept": false, - "reason": "Separate training-script failures with different causes and scripts." + "reason": "Different training script failures: missing saved model artifacts versus a subprocess exit error in dreambooth inpaint." }, { "left": "issue:2784", "right": "issue:3974", "accept": false, - "reason": "Feature request for text2video Dreambooth vs ControlNet safetensors loading issue." + "reason": "Different requests: Dreambooth support for text2video models versus safetensors support for ControlNet inpaint." }, { "left": "issue:3455", "right": "issue:4405", "accept": false, - "reason": "Different example/training requests: instruct-pix2pix training behavior vs textual inversion for SDXL." + "reason": "Different training features: pretrained instruct-pix2pix usage versus a textual inversion training script for SDXL." }, { "left": "issue:3679", "right": "issue:5926", "accept": false, - "reason": "Unrelated model announcements/requests in different model families." + "reason": "Different model requests: specific text2video model additions versus Kandinsky 3.0 support." }, { "left": "issue:3944", "right": "issue:4672", "accept": false, - "reason": "Both are LoRA training topics, but one is rank customization and the other is VRAM reduction; not the same bug." + "reason": "Not the same change: network-rank configurability for LoRA training versus reducing VRAM usage." }, { "left": "issue:4232", "right": "issue:5392", "accept": false, - "reason": "Different scope: instruct-pix2pix with ControlNet vs unconditional latent diffusion training." + "reason": "Different tasks: instruct-pix2pix with ControlNet conditioning versus training an unconditional latent diffusion model." }, { "left": "issue:4649", "right": "issue:5510", "accept": false, - "reason": "Similar SDXL family, but different runtime failures and different pipeline code paths." + "reason": "Different runtime errors in different pipelines: NoneType added_cond_kwargs versus dtype mismatch in SDXL InstructPix2Pix." }, { "left": "issue:4803", "right": "issue:6050", "accept": false, - "reason": "Both are install-related, but one is a missing requirements file and the other is a pip build/install failure." + "reason": "Different installation failures: missing requirements file versus pip install subprocess-exited-with-error." }, { "left": "issue:5518", "right": "issue:9067", "accept": false, - "reason": "No shared bug: a LoRA attribute error vs a missing-module import error." + "reason": "Different import/runtime problems: ModuleList missing time attribute versus missing diffusers.hub_utils module." }, { "left": "issue:6335", "right": "issue:9933", "accept": false, - "reason": "Both mention size parameters, but they are different pipelines with different missing/ignored arguments." + "reason": "Related only at a high level; these are different pipeline API issues on different models and parameters, not the same bug." }, { "left": "issue:6441", "right": "issue:7177", "accept": false, - "reason": "Different from_pretrained failures in unrelated contexts." + "reason": "Both involve from_pretrained, but one is a Google Colab issue and the other is a StableCascadeDecoderPipeline-specific load error." }, { "left": "issue:6685", "right": "issue:8708", "accept": false, - "reason": "Dreambooth-related, but one is output_dir path handling and the other is a missing SD3 VAE config field." + "reason": "Different Dreambooth topics: absolute output_dir path handling versus a missing SD3 VAE shift_factor in examples." }, { "left": "issue:7456", "right": "issue:9398", "accept": false, - "reason": "A missing raise statement is not the same as a missing pooled_projection argument." + "reason": "Different bugs: a missing raise statement versus a missing pooled_projection argument in a projection embedding module." }, { "left": "issue:7658", "right": "issue:8434", "accept": false, - "reason": "Both involve multiple control inputs, but one asks if it is possible and the other is specifically about XSPipeline support." - }, - { - "left": "issue:8507", - "right": "issue:9972", - "accept": false, - "reason": "Completely unrelated: SD3 Dreambooth text encoder failure vs CogX macOS memory allocation bug." - }, - { - "left": "issue:1896", - "right": "issue:4972", - "accept": false, - "reason": "Different example failures with different error signatures and code paths." - }, - { - "left": "issue:2444", - "right": "issue:713", - "accept": false, - "reason": "Too generic to be duplicate; one is a specific inpaint training script failure, the other is a broad training error report." + "reason": "Both mention multiple ControlNets, but one is a question about whether multiple conditioning is possible and the other is a feature request for specific XSPipelines; not the same concrete bug or change." } ] }, @@ -91807,27 +91820,26 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10186", - "issue:10350", "issue:10428", "issue:11092", "issue:1158", "issue:11581", "issue:11872", - "issue:12436", "issue:12574", "issue:12610", - "issue:13140", "issue:13298", "issue:1695", "issue:1807", "issue:1842", "issue:1851", + "issue:1896", + "issue:2444", "issue:2507", "issue:2554", "issue:2799", @@ -91836,10 +91848,10 @@ "issue:3740", "issue:3887", "issue:3967", - "issue:4356", "issue:4598", "issue:4789", "issue:488", + "issue:4972", "issue:5009", "issue:5152", "issue:5372", @@ -91853,15 +91865,19 @@ "issue:6196", "issue:6392", "issue:6869", + "issue:713", "issue:7210", "issue:7312", "issue:782", - "issue:8551", + "issue:8507", "issue:8708", - "issue:8994", - "issue:9052" + "issue:9052", + "issue:9972" ], "soft_pairs": [ + "issue:8507|issue:9972", + "issue:1896|issue:4972", + "issue:2444|issue:713", "issue:3431|issue:557", "issue:1842|issue:5009", "issue:11872|issue:12574", @@ -91882,183 +91898,180 @@ "issue:3887|issue:3967", "issue:5977|issue:8708", "issue:6158|issue:7312", - "issue:6392|issue:9052", - "issue:10350|issue:8551", - "issue:4356|issue:8994", - "issue:12436|issue:13140" + "issue:6392|issue:9052" ], - "prepared_review_unit_hash": "4d1b48a0fd8662511e1cf9a5fb139db346232f9ecbbb719c6325ee44b2e6890b", + "prepared_review_unit_hash": "c01acc4ef5bf30985055068e89781a52231d6be9f2d5e66f8f181c7109bed718", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33945, - "estimated_input_tokens": 8487, - "estimated_eval_tokens": 17230 + "serialized_chars": 33769, + "estimated_input_tokens": 8443, + "estimated_eval_tokens": 17142 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster does not look like a duplicate set; it is a heterogeneous mix of unrelated Diffusers issues spanning imports, memory/offload, training, checkpoints, and pipeline-specific bugs. None of the soft pairs appear to share the same underlying bug/change.", - "confidence": 0.98, - "canonical_issue_reason": "No clear canonical issue: the items cover different subsystems and failure modes, so there is no single underlying bug to deduplicate around.", + "summary": "This set is a grab bag of unrelated diffusers issues; none of the soft-similarity pairs look like the same underlying bug or change. All soft edges should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 6068 is the strongest representative bug report: concrete, reproducible, and high-signal (SDXL/DDIM incorrect sampling) with substantial discussion.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a strong global representative because the cluster is too diverse and lacks a dominant shared problem.", + "best_issue_reason": "Issue 6068 is the best overall issue candidate because it describes a specific code-path defect with clear scope and impact, making it a strong canonical reference.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:8507", + "right": "issue:9972", + "accept": false, + "reason": "Different failures: SD3 DreamBooth training vs CogX/macOS buffer allocation; no shared bug path." + }, + { + "left": "issue:1896", + "right": "issue:4972", + "accept": false, + "reason": "Different pipelines and errors; CLIP-guided SD size/type issue is unrelated to wuerstchen example startup failure." + }, + { + "left": "issue:2444", + "right": "issue:713", + "accept": false, + "reason": "Generic training error is too broad to equate with a specific DreamBooth inpaint subprocess failure." + }, { "left": "issue:3431", "right": "issue:557", "accept": false, - "reason": "Different problems: inpainting+ControlNet quality vs DDIM quality regression; not the same bug." + "reason": "Different symptoms and contexts: ControlNet inpainting quality vs DDIM quality comparison." }, { "left": "issue:1842", "right": "issue:5009", "accept": false, - "reason": "Both are generic errors, but the contexts differ (`eta` argument vs wuerstchen error) and there is no evidence of the same code path." + "reason": "Unexpected `eta` kwarg is a scheduler API mismatch, not the same as a generic wuerstchen error." }, { "left": "issue:11872", "right": "issue:12574", "accept": false, - "reason": "Both mention device/offload behavior, but one is CPU offloading inefficiency and the other is a WanVideo/VACE control-argument failure; different bugs." + "reason": "Performance/offloading inefficiency is not the same as WanVideo control-argument handling failure." }, { "left": "issue:1851", "right": "issue:3175", "accept": false, - "reason": "Both are Dreambooth-related, but one is multi-GPU slowdown and the other is training multi-face support; not duplicates." + "reason": "Both are training-related, but they describe different multi-GPU cases and different scripts." }, { "left": "issue:5372", "right": "issue:7210", "accept": false, - "reason": "Different issues: safety_checker usage in SDXL vs missing `named_parameters` with accelerate dispatch." + "reason": "Safety checker usage question is unrelated to missing `named_parameters` under accelerate dispatch." }, { "left": "issue:10428", "right": "issue:11092", "accept": false, - "reason": "Flux inference on Ascend NPU vs empty timesteps in SDXL-Turbo img2img; unrelated pipeline failures." + "reason": "Flux inference on Ascend NPU is a hardware/runtime error; empty timesteps in SDXL-Turbo is a scheduler edge case." }, { "left": "issue:2799", "right": "issue:488", "accept": false, - "reason": "Checkpoint/pipeline mismatch error vs grayscale `numpy_to_pil` failure; different functionality." + "reason": "Checkpoint/pipeline mismatch error is unrelated to grayscale `numpy_to_pil` handling." }, { "left": "issue:1158", "right": "issue:5523", "accept": false, - "reason": "Test interference in `test_config.py` vs inconsistent generation results; no shared underlying bug." + "reason": "Test interference and inconsistent results are both broad, but not the same concrete bug." }, { "left": "issue:4789", "right": "issue:782", "accept": false, - "reason": "Model download failure and broken auth-token download are both download-related, but not the same concrete issue." + "reason": "DeepFloyd download failure is unrelated to the auth-token API removal issue." }, { "left": "issue:12610", "right": "issue:13298", "accept": false, - "reason": "Both involve device placement, but one is Quanto+group offload mismatch and the other is generator-on-cuda breaking image generation; different root causes." + "reason": "Both mention device mismatch, but one is quant/offload weight placement and the other is generator device handling; not the same bug." }, { "left": "issue:6068", "right": "issue:6196", "accept": false, - "reason": "Different SDXL sampler/V-prediction bug vs ControlNetXS Depth SDXL loading error." + "reason": "Incorrect DDIM sampling in SDXL is a different failure from ControlNetXS Depth SDXL loading." }, { "left": "issue:2554", "right": "issue:3740", "accept": false, - "reason": "Both involve accelerate/version/device mapping, but one is an accelerator version requirement and the other is legacy attention conversion with `device_map=\"sequential\"`." + "reason": "Unreleased accelerate dependency is unrelated to legacy attention block conversion with sequential device_map." }, { "left": "issue:5152", "right": "issue:6869", "accept": false, - "reason": "W\u00fcrstchen-specific bug vs video export unpacking error; unrelated code paths." + "reason": "W\u00fcrstchen alpha bug and `export_to_video` unpacking error are different code paths." }, { "left": "issue:10186", "right": "issue:11581", "accept": false, - "reason": "Both are import/loading failures around FLUX, but one is missing `FluxFillPipeline` import and the other is an IP-Adapter dtype constructor error; different bugs." + "reason": "ImportError for `FluxFillPipeline` is not the same as an IP Adapter dtype-constructor error in FLUX." }, { "left": "issue:1807", "right": "issue:2507", "accept": false, - "reason": "Both discuss performance, but one is inference slowing over generations and the other is safetensors loading speed; not the same defect." + "reason": "Both are performance complaints, but one is generation slowdown and the other is safetensors load speed." }, { "left": "issue:4598", "right": "issue:5570", "accept": false, - "reason": "Informative dependency errors vs missing `compute_snr` symbol in training utils; different problem statements and fixes." + "reason": "Missing dependency messaging is a UX/error-reporting issue, not the same as missing `compute_snr` in training utils." }, { "left": "issue:1695", "right": "issue:5951", "accept": false, - "reason": "Both are training-related, but one is multi-GPU Dreambooth/deepspeed and the other is Pixart training; not duplicates." + "reason": "DreamBooth multi-GPU training and Pixart training are different workloads with different failure modes." }, { "left": "issue:3887", "right": "issue:3967", "accept": false, - "reason": "General SD effect differences vs SDXL not working after branch deletion; no shared bug." + "reason": "Quality/regression complaint is unrelated to a branch/deleted-code runtime failure." }, { "left": "issue:5977", "right": "issue:8708", "accept": false, - "reason": "LoRA key-saving behavior vs missing `vae.config.shift_factor` in SD3 training examples; unrelated." + "reason": "LoRA checkpoint key-saving and SD3 DreamBooth `shift_factor` missing are different training issues." }, { "left": "issue:6158", "right": "issue:7312", "accept": false, - "reason": "Animatediff SDXL feature request vs `named_parameters` missing on pipeline with accelerate; not the same issue." + "reason": "Animatediff SDXL support request is unrelated to AlignProp reward finetuning support." }, { "left": "issue:6392", "right": "issue:9052", "accept": false, - "reason": "Custom LoRA loading failure vs `from_single_file` failing on certain checkpoints; different loading paths." - }, - { - "left": "issue:10350", - "right": "issue:8551", - "accept": false, - "reason": "Both touch VAE memory/device handling, but one is VAE decoder memory use in SD2.1 and the other is SDXL inpaint latent device placement with balanced device_map." - }, - { - "left": "issue:4356", - "right": "issue:8994", - "accept": false, - "reason": "Missing dummy module in 0.19.2 vs generation steps stuck at 0; unrelated errors." - }, - { - "left": "issue:12436", - "right": "issue:13140", - "accept": false, - "reason": "Constructor keyword errors, but for different models and causes (`offload_state_dict` in CLIPTextModel vs NoneType in Qwen Edit pipelines)." + "reason": "Custom LoRA loading and single-file civit checkpoint conversion are both loading flows, but not the same failure." } ] }, @@ -92077,13 +92090,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10039", - "issue:10282", + "issue:10350", "issue:1101", "issue:11329", "issue:1140", @@ -92091,6 +92104,8 @@ "issue:11489", "issue:11833", "issue:1197", + "issue:12436", + "issue:13140", "issue:2044", "issue:2596", "issue:3300", @@ -92098,7 +92113,7 @@ "issue:3880", "issue:3944", "issue:4183", - "issue:4403", + "issue:4356", "issue:4419", "issue:4835", "issue:4876", @@ -92109,7 +92124,6 @@ "issue:5897", "issue:5954", "issue:6087", - "issue:6517", "issue:6709", "issue:6725", "issue:6732", @@ -92117,21 +92131,23 @@ "issue:7142", "issue:7264", "issue:7312", - "issue:7572", - "issue:7651", "issue:8477", "issue:8532", + "issue:8551", "issue:8707", "issue:8780", "issue:8805", "issue:8826", "issue:8939", "issue:8958", - "issue:9090", + "issue:8994", "issue:9602", "issue:9856" ], "soft_pairs": [ + "issue:10350|issue:8551", + "issue:4356|issue:8994", + "issue:12436|issue:13140", "issue:8477|issue:8707", "issue:9602|issue:9856", "issue:8826|issue:8958", @@ -92152,183 +92168,180 @@ "issue:7264|issue:8780", "issue:5224|issue:6709", "issue:4835|issue:5109", - "issue:5954|issue:7142", - "issue:7651|issue:9090", - "issue:4403|issue:7572", - "issue:10282|issue:6517" + "issue:5954|issue:7142" ], - "prepared_review_unit_hash": "b367987fe00b9b2bd5168d53d827756ad64f52b1130d661278365501234442da", + "prepared_review_unit_hash": "48f7874eea34b261181d1563aea2a75eb9beabfd75d003e983fd593703d7757c", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34475, - "estimated_input_tokens": 8619, - "estimated_eval_tokens": 17494 + "serialized_chars": 34445, + "estimated_input_tokens": 8612, + "estimated_eval_tokens": 17480 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft-similarity links are mostly false positives: they connect broadly related Diffusers/transformer topics, but not the same concrete bug or change. I found no safe duplicate merges among the candidate pairs.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue stands out; the items span unrelated bugs, feature requests, docs problems, and pipeline-specific regressions, so none safely subsumes the rest.", + "summary": "This is not a true duplicate cluster: it groups many unrelated issues across schedulers, LoRA training, VAE/device-map bugs, pipeline failures, and docs. All soft pairs are only superficially similar and should stay separate.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue fits this cluster; the items are heterogeneous and do not describe the same underlying bug or change.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a strong global representative because the cluster is heterogeneous rather than one underlying defect.", + "best_issue_reason": "No strong best issue candidate either; the cluster is too mixed for one issue to safely represent the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10350", + "right": "issue:8551", + "accept": false, + "reason": "Both involve VAE decoding, but one is high memory use in SD2.1 and the other is a device-map latent transfer bug in SDXL inpaint; different failure modes." + }, + { + "left": "issue:4356", + "right": "issue:8994", + "accept": false, + "reason": "0.19.2 missing dummy module import is unrelated to generation steps stuck at 0." + }, + { + "left": "issue:12436", + "right": "issue:13140", + "accept": false, + "reason": "Different errors and code paths: CLIPTextModel init keyword mismatch vs Qwen pipeline NoneType iteration." + }, { "left": "issue:8477", "right": "issue:8707", "accept": false, - "reason": "Both are training errors, but one is a DistributedDataParallel failure and the other is an SDXL/Flux training script error; different code paths." + "reason": "Both training-related, but one is a DDP error in a research script and the other is an SDXL text-to-image training failure; not the same bug." }, { "left": "issue:9602", "right": "issue:9856", "accept": false, - "reason": "Adapter-name validation and distributed port conflicts are unrelated failures." + "reason": "Adapter registration error and distributed port conflict are unrelated." }, { "left": "issue:8826", "right": "issue:8958", "accept": false, - "reason": "A torch.cat empty-list error is not the same as a missing torch.library.register_fake API." + "reason": "torch.cat on an empty list is a different runtime issue from torch.library.register_fake missing." }, { "left": "issue:11424", "right": "issue:11833", "accept": false, - "reason": "VAE group_norm decode failure and set_lora_device KeyError are different subsystems." + "reason": "VAE group_norm decode failure is unrelated to a LoRA device-setting KeyError." }, { "left": "issue:10039", "right": "issue:8805", "accept": false, - "reason": "Missing AttnUpBlock2D sample size and an SDXL conversion-script regression are different bugs." + "reason": "Forgotten output sample size in a UNet block is not the same as the SDXL conversion script regression." }, { "left": "issue:1101", "right": "issue:341", "accept": false, - "reason": "These are separate scheduler feature requests, not the same change." + "reason": "Both are scheduler requests, but DPM-Solver and K-LMS in-paint are distinct features, not one bug/change." }, { "left": "issue:1140", "right": "issue:1197", "accept": false, - "reason": "Two unrelated community pipeline requests." + "reason": "Separate community feature requests with no shared code path." }, { "left": "issue:3880", "right": "issue:6087", "accept": false, - "reason": "Inpainting degradation and LoRA config save/load corruption are distinct issues." + "reason": "Inpainting degradation and LoRAConfig save/load metadata issues are unrelated." }, { "left": "issue:3944", "right": "issue:4876", "accept": false, - "reason": "Both concern LoRA parameters, but one asks for network rank customization and the other for std initialization; different feature requests." + "reason": "Both mention LoRA, but one requests network-rank customization for training and the other asks for a std argument for LoRA layers; different changes." }, { "left": "issue:4183", "right": "issue:6725", "accept": false, - "reason": "ControlNet guidance-scale bug and generic LDMPipeline generation request are not the same problem." + "reason": "ControlNet guidance_scale bug and LDMPipeline generation request are unrelated." }, { "left": "issue:4419", "right": "issue:6732", "accept": false, - "reason": "Offset-noise inference mismatch is different from a scheduler doc omission." + "reason": "Offset-noise inference mismatch is unrelated to documentation for trained_betas." }, { "left": "issue:5079", "right": "issue:7312", "accept": false, - "reason": "LoRA loading support and AlignProp reward finetuning are unrelated changes." + "reason": "Loading a specific LoRA type and AlignProp support are different features." }, { "left": "issue:5897", "right": "issue:6812", "accept": false, - "reason": "Both are LoRA training errors, but one is device placement and the other is dtype mismatch; different fixes." + "reason": "Both are LoRA training errors, but device mismatch and dtype mismatch are different underlying bugs." }, { "left": "issue:11489", "right": "issue:8939", "accept": false, - "reason": "Flux LoRA training failure and tokenizer loading failure are unrelated." + "reason": "Flux LoRA training failure and tokenizer load failure for SD3 are unrelated." }, { "left": "issue:11329", "right": "issue:8532", "accept": false, - "reason": "Both mention memory, but one is a bitsandbytes leak and the other is LoRA-weight loading growth; not the same bug." + "reason": "Memory leak in bitsandbytes quantization and progressive memory increase when loading LoRA weights are different code paths." }, { "left": "issue:2044", "right": "issue:5335", "accept": false, - "reason": "DDP dtype attribute access and local deployment blocked by Hugging Face are unrelated." + "reason": "DDP dtype attribute error has nothing to do with local deployment behind blocked Hugging Face access." }, { "left": "issue:2596", "right": "issue:3300", "accept": false, - "reason": "Docs referencing a missing function is not the same as HF-hosted package breakage." + "reason": "Documentation typo about load_image is unrelated to the broader 'Diffusers on HuggingFace are broken' report." }, { "left": "issue:7264", "right": "issue:8780", "accept": false, - "reason": "Custom-model serialization and model/input dtype mismatch are different issues." + "reason": "Serialization/loading of custom transformers models is unrelated to a model/input dtype mismatch." }, { "left": "issue:5224", "right": "issue:6709", "accept": false, - "reason": "A Flax SD 2.1 error and custom diffusion concept-list training bug are separate problems." + "reason": "Flax SD2.1 error and custom diffusion concept-list training bug are different issues." }, { "left": "issue:4835", "right": "issue:5109", "accept": false, - "reason": "MultiControlNet img2img behavior and load_lora_weights exception are different code paths." + "reason": "Same pipeline family, but MultiControlNetModel img2img failure and load_lora_weights exception are distinct concrete problems." }, { "left": "issue:5954", "right": "issue:7142", "accept": false, - "reason": "Kandinsky22 inference error and SDXL-lightning output discrepancy are not the same bug." - }, - { - "left": "issue:7651", - "right": "issue:9090", - "accept": false, - "reason": "Img2img timestep calculation and repeated PNDM timesteps are scheduler-related but different concrete defects." - }, - { - "left": "issue:4403", - "right": "issue:7572", - "accept": false, - "reason": "Half-precision LayerNorm failure and half/float bias-type mismatch are different runtime errors." - }, - { - "left": "issue:10282", - "right": "issue:6517", - "accept": false, - "reason": "An x_embedder weight shape mismatch is unrelated to an import traceback in SDXL ControlNet img2img." + "reason": "Kandinsky22 inference error and SDXL-lightning output mismatch are unrelated." } ] }, @@ -92347,34 +92360,30 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10121", + "issue:10282", "issue:10591", "issue:10895", "issue:11302", "issue:11436", - "issue:11749", "issue:11938", - "issue:1323", - "issue:13386", "issue:175", "issue:181", "issue:2052", - "issue:2053", "issue:2276", - "issue:2294", "issue:2332", "issue:2676", "issue:3159", "issue:3380", - "issue:3441", "issue:3453", "issue:3593", "issue:4292", + "issue:4403", "issue:4585", "issue:4615", "issue:4709", @@ -92384,6 +92393,7 @@ "issue:5315", "issue:5799", "issue:6162", + "issue:6517", "issue:6634", "issue:6732", "issue:7009", @@ -92391,17 +92401,23 @@ "issue:7341", "issue:7365", "issue:7562", + "issue:7572", + "issue:7651", "issue:7931", "issue:8451", "issue:8534", "issue:8551", "issue:8975", "issue:9051", + "issue:9090", "issue:9181", "issue:9822", "issue:9844" ], "soft_pairs": [ + "issue:7651|issue:9090", + "issue:4403|issue:7572", + "issue:10282|issue:6517", "issue:11302|issue:11436", "issue:2052|issue:5124", "issue:8551|issue:9844", @@ -92422,183 +92438,180 @@ "issue:7009|issue:9822", "issue:10121|issue:6162", "issue:4585|issue:8975", - "issue:2276|issue:4872", - "issue:11749|issue:13386", - "issue:1323|issue:2294", - "issue:2053|issue:3441" + "issue:2276|issue:4872" ], - "prepared_review_unit_hash": "34958917997c7c6eca3f6eadf4d007b1568ac66422054770062d92c4fac19295", + "prepared_review_unit_hash": "7cbb23eec8e072cb22f86750396be1e8d45bac3f5c8d5ce87335124b3ab30e83", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 34115, - "estimated_input_tokens": 8529, - "estimated_eval_tokens": 17314 + "serialized_chars": 34199, + "estimated_input_tokens": 8550, + "estimated_eval_tokens": 17356 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose collection of unrelated Diffusers issues with only superficial title overlap. None of the soft pairs look like true duplicates, so all should be rejected. There are no PRs in the cluster; if a representative issue is needed, issue #11436 is the clearest concrete bug report.", - "confidence": 0.91, - "canonical_issue_reason": "No single canonical duplicate stands out: the issues span different models, scripts, schedulers, and failure modes, and the soft pairs are not same-bug matches.", + "summary": "This cluster is not a true duplicate set: it mixes unrelated issues across inference, training scripts, schedulers, docs, import errors, and model-specific bugs. No soft-edge pair looks like the same underlying defect/change.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 7651 is the most central and well-developed bug report here: it targets a concrete pipeline timing defect, has the highest discussion activity, and is more representative than the narrower one-off reports.", "canonical_pr_reason": null, - "best_issue_reason": "#11436 is the most concrete, self-contained bug report with a specific runtime error and enough context to serve as a representative issue.", + "best_issue_reason": "7651 is the best standalone issue to keep as the cluster anchor because it describes a clear, reproducible core pipeline bug rather than a peripheral docs/import/training variant.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:7651", + "right": "issue:9090", + "accept": false, + "reason": "Both mention timesteps, but 7651 is about img2img timestep calculation while 9090 is a PNDMScheduler repeated-timestep bug; different code paths and symptoms." + }, + { + "left": "issue:4403", + "right": "issue:7572", + "accept": false, + "reason": "Both are dtype-related, but one is LayerNorm half-kernel support and the other is input/bias dtype mismatch; not the same bug." + }, + { + "left": "issue:10282", + "right": "issue:6517", + "accept": false, + "reason": "A model weight shape mismatch during loading is unrelated to a pipeline import failure in SDXL ControlNet img2img." + }, { "left": "issue:11302", "right": "issue:11436", "accept": false, - "reason": "Different HiDream problems: saving a quantized model vs an inference dtype mismatch. Not the same code path." + "reason": "One is about saving a quantized HiDream model; the other is an inference dtype mismatch in HiDream. Same model family, different failure modes." }, { "left": "issue:2052", "right": "issue:5124", "accept": false, - "reason": "Textual inversion checkpoint resume bug vs a generic question about fine-tuning a .safetensor checkpoint. Different issues." + "reason": "Checkpoint resume behavior in textual inversion is not the same as a question about fine-tuning from a .safetensors checkpoint." }, { "left": "issue:8551", "right": "issue:9844", "accept": false, - "reason": "One is a device placement bug in SDXL inpaint VAE offload; the other is NaNs from the SDXL VAE encoder. Different failures." + "reason": "Device placement for VAE latents with balanced device_map is unrelated to SDXL VAE NaN outputs." }, { "left": "issue:3593", "right": "issue:7562", "accept": false, - "reason": "LoRA+xformers second-call failure is unrelated to multicontrolnet CPU offload on SDXL img2img." + "reason": "LoRA+xformers second-call failure is a different problem from multi-ControlNet CPU offload failure in img2img." }, { "left": "issue:10591", "right": "issue:8451", "accept": false, - "reason": "Different training scripts and concerns: SD3 LoRA script bug vs missing/frozen SD1.x training details." + "reason": "These concern different training-script issues: SD3 LoRA script correctness versus sd1.x training details and frozen parameters." }, { "left": "issue:10895", "right": "issue:7365", "accept": false, - "reason": "Deprecation handling is unrelated to pooled_prompt_embeds being overwritten in a pipeline input bug." + "reason": "Deprecation handling is unrelated to pooled_prompt_embeds being overwritten by prompt_embeds[0]." }, { "left": "issue:11938", "right": "issue:9051", "accept": false, - "reason": "RMSNorm module registration bug vs Flux batch inference limitation. Not the same underlying change." + "reason": "RMSNorm submodule registration is a model-initialization bug; Flux batch inference support is a separate feature/behavior issue." }, { "left": "issue:175", "right": "issue:3159", "accept": false, - "reason": "Seed usage in example training code vs a documentation typo. Different scope and problem." + "reason": "A training-code seeding question is not the same as a documentation typo." }, { "left": "issue:181", "right": "issue:3453", "accept": false, - "reason": "Image-writing failure in train_unconditional.py is unrelated to the BF16 triu/tril error in text-to-image training." + "reason": "Image writing failure in unconditional training is distinct from a bfloat16 CUDA kernel limitation in text-to-image training." }, { "left": "issue:2332", "right": "issue:2676", "accept": false, - "reason": "Dreambooth inference error vs a question about scheduler names/support. Not duplicate bug reports." + "reason": "Dreambooth inference errors and scheduler-name lookup questions are unrelated." }, { "left": "issue:3380", "right": "issue:7165", "accept": false, - "reason": "Hard-coded embedding size bug is unrelated to dynamic thresholding scale in JAX." + "reason": "Hard-coded embedding size and JAX dynamic-thresholding scale are different components with different failure modes." }, { "left": "issue:4292", "right": "issue:5799", "accept": false, - "reason": "M1 inpainting crash and the inpainting strength keyword issue are different behaviors and likely different fixes." + "reason": "M1 inpainting crashes are not the same as a strength keyword-argument API issue." }, { "left": "issue:4615", "right": "issue:9181", "accept": false, - "reason": "ImportError for ControlNet pipeline vs a visual artifact ('noise dot') in StableDiffusionControlNetPipeline. Different issues." + "reason": "Importing a missing pipeline class is unrelated to a ControlNet noise-artifact bug." }, { "left": "issue:4709", "right": "issue:7931", "accept": false, - "reason": "Unsupported guess mode in SDXL ControlNet is unrelated to IP-Adapter shape mismatch in inpaint." + "reason": "Guess mode support in SDXL ControlNet is different from IP-Adapter embed handling in the inpaint pipeline." }, { "left": "issue:4753", "right": "issue:8534", "accept": false, - "reason": "Dreambooth LoRA bug vs SD3 dreambooth logit-normal weighting bug. Different scripts and code paths." + "reason": "Both touch training scripts, but they target different scripts and different bugs." }, { "left": "issue:5315", "right": "issue:6634", "accept": false, - "reason": "Both touch DDIMInverseScheduler, but one is a step-function bug and the other is a clipping-option question; not the same concrete defect." + "reason": "Both involve DDIMInverseScheduler, but one is a step-function bug and the other is about use_clipped_model_output; not clearly the same defect." }, { "left": "issue:6732", "right": "issue:7341", "accept": false, - "reason": "Scheduler documentation for trained_betas is unrelated to the TorchVersion/Version comparison TypeError." + "reason": "Missing documentation for trained_betas is unrelated to a TorchVersion/Version comparison TypeError." }, { "left": "issue:7009", "right": "issue:9822", "accept": false, - "reason": "Poor results in a text-to-video pipeline are unrelated to loading SDXL LoRAs into Flux." + "reason": "Poor results from a text-to-video pipeline are not the same as loading SDXL LoRAs into Flux." }, { "left": "issue:10121", "right": "issue:6162", "accept": false, - "reason": "Flux high-resolution generation issue is unrelated to SDXL IP-Adapter shape mismatch." + "reason": "Flux high-resolution generation limitations and SDXL IP-Adapter shape mismatch are different model/pipeline bugs." }, { "left": "issue:4585", "right": "issue:8975", "accept": false, - "reason": "SDXL MultiControlNet documentation/runtime mismatch is unrelated to the LoraLoaderMixin import error." + "reason": "A documentation inconsistency about MultiControlNet is unrelated to an import error for LoraLoaderMixin." }, { "left": "issue:2276", "right": "issue:4872", "accept": false, - "reason": "xformers installation failure in a notebook and CUDA OOM during SDXL finetuning are different problems." - }, - { - "left": "issue:11749", - "right": "issue:13386", - "accept": false, - "reason": "Scheduler timestep floating-point errors are unrelated to qwen_image prior-preservation tensor concatenation crashes." - }, - { - "left": "issue:1323", - "right": "issue:2294", - "accept": false, - "reason": "Flax dtype-loading bug vs missing StableDiffusionVideoInpaintPipeline import. Different modules and failure modes." - }, - { - "left": "issue:2053", - "right": "issue:3441", - "accept": false, - "reason": "Both mention VRAM, but one is a broad regression and the other is a PyTorch2/xformers-specific memory increase. Not the same bug." + "reason": "xformers installation failure and CUDA OOM during finetuning are separate problems." } ] }, @@ -92617,8 +92630,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -92628,14 +92641,17 @@ "issue:10722", "issue:1080", "issue:11497", - "issue:11581", - "issue:11659", + "issue:11749", + "issue:1323", + "issue:13386", "issue:1734", "issue:1988", + "issue:2053", + "issue:2294", "issue:2618", - "issue:2659", "issue:3029", "issue:3103", + "issue:3441", "issue:3717", "issue:4139", "issue:4815", @@ -92648,10 +92664,8 @@ "issue:5605", "issue:5737", "issue:5860", - "issue:6091", "issue:6109", "issue:617", - "issue:6232", "issue:6335", "issue:6392", "issue:6399", @@ -92659,7 +92673,6 @@ "issue:6527", "issue:6732", "issue:6921", - "issue:8323", "issue:8503", "issue:8556", "issue:8579", @@ -92672,6 +92685,9 @@ "issue:9680" ], "soft_pairs": [ + "issue:11749|issue:13386", + "issue:1323|issue:2294", + "issue:2053|issue:3441", "issue:5860|issue:8850", "issue:11497|issue:8503", "issue:3029|issue:5436", @@ -92692,183 +92708,180 @@ "issue:10722|issue:1734", "issue:5395|issue:8989", "issue:10054|issue:6527", - "issue:10085|issue:10150", - "issue:2659|issue:6091", - "issue:6232|issue:8323", - "issue:11581|issue:11659" + "issue:10085|issue:10150" ], - "prepared_review_unit_hash": "561165ff28bfba7273771834ed06ba4f6314489c0dae0daac9a4aba3762cfffb", + "prepared_review_unit_hash": "4088aa53ebc6ffa72f0787d4e512a2435c167f83bbb0eca5435dd7bc2b23614e", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33799, - "estimated_input_tokens": 8450, - "estimated_eval_tokens": 17156 + "serialized_chars": 33852, + "estimated_input_tokens": 8463, + "estimated_eval_tokens": 17182 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly unrelated issue reports; none of the soft pairs look like true duplicates or the same underlying bug/change. I would anchor the cluster on the most concrete, well-scoped report: the SD3 preprocessing-cache bug.", - "confidence": 0.87, - "canonical_issue_reason": "issue:11497 is the most concrete and actionable report in the set, with a specific subsystem (dataset preprocessing cache) and a clear failure mode tied to unrelated arguments.", + "summary": "All candidate pairs are weak title-level similarities rather than true duplicates. The set spans many unrelated bugs across schedulers, SD3/SDXL training, ControlNet, LoRA, and pipeline loading; none of the soft pairs appear to describe the same concrete code path.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 11497 is the most broadly phrased, concrete, and actionable bug report in the set: it identifies a specific SD3 ControlNet preprocessing/cache dependency problem that is clearly reproducible and could anchor related training-script issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:11497 has the clearest bug description and strongest triage value as a representative issue for this set.", + "best_issue_reason": "Issue 11497 is the best standalone issue to keep because it is specific enough to triage, has a clear failure mechanism, and is broader than the more narrowly scoped or vaguer reports around it.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11749", + "right": "issue:13386", + "accept": false, + "reason": "Different domains and failures: scheduler floating-point timesteps vs DreamBooth prior-preservation tensor concatenation." + }, + { + "left": "issue:1323", + "right": "issue:2294", + "accept": false, + "reason": "Unrelated problems: Flax dtype loading vs missing export/import for a video inpaint pipeline." + }, + { + "left": "issue:2053", + "right": "issue:3441", + "accept": false, + "reason": "Both mention VRAM, but one is a version regression and the other is a PyTorch 2/xformers-specific memory issue." + }, { "left": "issue:5860", "right": "issue:8850", "accept": false, - "reason": "Different problems: Tiny Autoencoder/ControlNet pipeline failure vs community pipeline mirror being broken. Same broad area, not the same bug." + "reason": "Different features: Tiny Autoencoder+ControlNet failure vs broken community pipeline mirrors." }, { "left": "issue:11497", "right": "issue:8503", "accept": false, - "reason": "One is cache invalidation in SD3 ControlNet scripts; the other is a gradient checkpointing crash in SD3 training. Distinct code paths and symptoms." + "reason": "Unrelated SD3 training-script cache bug vs gradient-checkpointing crash." }, { "left": "issue:3029", "right": "issue:5436", "accept": false, - "reason": "Both concern ControlNet examples, but one is about custom resolution input handling and the other is an SDXL training example error. Not the same underlying defect." + "reason": "Both touch ControlNet, but one is custom resolution support and the other is an SDXL training example error." }, { "left": "issue:3717", "right": "issue:4849", "accept": false, - "reason": "Multi-GPU OOM is unrelated to textual inversion multi-GPU usage; same topic of GPUs, different issue class and failure mode." + "reason": "OOM during multi-GPU training is not the same as a general question about using multiple GPUs for textual inversion." }, { "left": "issue:6921", "right": "issue:9680", "accept": false, - "reason": "SDXL shipped weights deviating from official weights is a model artifact issue, while unconditional generation being broken is a runtime pipeline bug." + "reason": "Different issues: shipped SDXL weights diverging from official weights vs broken unconditional generation." }, { "left": "issue:8556", "right": "issue:8579", "accept": false, - "reason": "from_single_file meta-tensor loading failure is a model-loading issue; SD3 LoRA support non-functional is a LoRA integration issue. Different code paths." + "reason": "Different SD3 failures: meta-tensor loading error vs nonfunctional LoRA support." }, { "left": "issue:3103", "right": "issue:4815", "accept": false, - "reason": "Inpainting output dimension handling is not the same as allowing latents-only inputs. Related pipeline area, but distinct feature/bug." + "reason": "Both involve inpainting, but one is output height/width handling and the other is latent-only input support." }, { "left": "issue:4139", "right": "issue:6392", "accept": false, - "reason": "Custom diffusion attention precision casting vs load_lora_weights failing on a custom LoRA are separate LoRA-related problems, not the same bug." + "reason": "Different code paths: custom diffusion attention precision casting vs custom LoRA loading failure." }, { "left": "issue:1988", "right": "issue:617", "accept": false, - "reason": "fp16/bf16 image generation type mismatch is unrelated to NNPACK hardware initialization warnings." + "reason": "Unrelated failures: fp16/bf16 generation dtype mismatch vs NNPACK hardware support warning." }, { "left": "issue:6109", "right": "issue:6335", "accept": false, - "reason": "ControlNet-xs with depth control and ControlNet img2img output sizing are different pipeline behaviors and failure modes." + "reason": "Different ControlNet topics: depth-based ControlNet-xs behavior vs img2img output size parameter handling." }, { "left": "issue:1080", "right": "issue:5334", "accept": false, - "reason": "A usage question about running DreamBooth from Python is not the same as an Accelerate API requirement for multiple models during training." + "reason": "A usage question about launching DreamBooth from Python is not the same as Accelerate needing multiple models passed to prepare/accumulate." }, { "left": "issue:2618", "right": "issue:4841", "accept": false, - "reason": "ControlNet image input type handling vs image size issue are distinct problems." + "reason": "Different problems: FloatTensor image input acceptance vs ControlNet image-size handling." }, { "left": "issue:5557", "right": "issue:6482", "accept": false, - "reason": "Unexpected keyword use_reentrant in checkpointing is not the same as FP16 gradient unscale failure during resume." + "reason": "Unrelated errors: unexpected use_reentrant kwarg vs resume-from-checkpoint FP16 unscale failure." }, { "left": "issue:5605", "right": "issue:8615", "accept": false, - "reason": "Per-image caption support for SDXL DreamBooth is a training-data feature request; fuse_lora on SD3 is a model/adapter capability request." + "reason": "Different feature requests: per-image captions for SDXL DreamBooth vs fuse_lora support for SD3." }, { "left": "issue:5737", "right": "issue:9664", "accept": false, - "reason": "SDXL conversion script failure and DreamBooth subprocess failure are different scripts and different failure causes." + "reason": "Different code paths: conversion script failure for SDXL vs DreamBooth LoRA training subprocess failure." }, { "left": "issue:6399", "right": "issue:6732", "accept": false, - "reason": "LoRA visibility in UIs is an integration/packaging symptom; documented trained_betas is a scheduler API docs issue." + "reason": "Unrelated: LoRA visibility in UIs vs scheduler documentation for trained_betas." }, { "left": "issue:8708", "right": "issue:9467", "accept": false, - "reason": "Missing VAE shift_factor in SD3 DreamBooth examples is unrelated to CogVideoX num_frames guard rails." + "reason": "Different guardrails: missing SD3 VAE shift_factor in training examples vs num_frames validation for CogVideoX I2V." }, { "left": "issue:10722", "right": "issue:1734", "accept": false, - "reason": "Prior-preservation tensor shape mismatch during DreamBooth training is not the same as multi-GPU training being slower." + "reason": "Different symptoms and likely causes: prior-preservation tensor mismatch vs slow multi-GPU DreamBooth training." }, { "left": "issue:5395", "right": "issue:8989", "accept": false, - "reason": "Combining LoRAs from civitai is unrelated to sequential offloading behavior in Stable Audio." + "reason": "Unrelated: PEFT LoRA-merging bug vs Stable Audio sequential offloading bug." }, { "left": "issue:10054", "right": "issue:6527", "accept": false, - "reason": "Wrong torch.linspace parameters in EulerDiscreteScheduler testing is different from an add_noise bug in the same scheduler family." + "reason": "Same scheduler family but different concrete bugs: wrong torch.linspace parameters in tests vs add_noise behavior." }, { "left": "issue:10085", "right": "issue:10150", "accept": false, - "reason": "LyCORIS/skip_guidance_layers runtime error is unrelated to SDXL pipeline load/inference slowness." - }, - { - "left": "issue:2659", - "right": "issue:6091", - "accept": false, - "reason": "Batch-size-dependent image variation is a reproducibility issue; consistency decoder memory with size argument is a memory/performance issue." - }, - { - "left": "issue:6232", - "right": "issue:8323", - "accept": false, - "reason": "DPMSolverSinglestepScheduler add_noise bug and generic scheduler-file recognition issue are not the same defect." - }, - { - "left": "issue:11581", - "right": "issue:11659", - "accept": false, - "reason": "IP Adapter loading fails due to an unexpected dtype argument on CLIPVisionModelWithProjection; the LoRA load issue is unrelated." + "reason": "Different problems: LyCORIS/batch inference/skip_guidance_layers runtime error vs slow SDXL pipeline loading/inference." } ] }, @@ -92887,8 +92900,8 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ @@ -92896,11 +92909,12 @@ "issue:10121", "issue:10302", "issue:10315", - "issue:10798", "issue:11163", "issue:11540", + "issue:11581", "issue:11657", "issue:11659", + "issue:2659", "issue:2838", "issue:2851", "issue:3026", @@ -92918,12 +92932,13 @@ "issue:5984", "issue:6091", "issue:6230", + "issue:6232", "issue:6598", "issue:6610", - "issue:6921", "issue:7210", "issue:7322", "issue:7355", + "issue:8323", "issue:8450", "issue:8707", "issue:8710", @@ -92931,8 +92946,6 @@ "issue:898", "issue:9106", "issue:9163", - "issue:9200", - "issue:9224", "issue:9292", "issue:9361", "issue:939", @@ -92941,6 +92954,9 @@ "issue:9944" ], "soft_pairs": [ + "issue:2659|issue:6091", + "issue:6232|issue:8323", + "issue:11581|issue:11659", "issue:10120|issue:6230", "issue:11163|issue:9106", "issue:11657|issue:11659", @@ -92962,182 +92978,186 @@ "issue:6610|issue:8450", "issue:9163|issue:9464", "issue:9292|issue:9361", - "issue:5984|issue:7322", - "issue:6921|issue:9200", - "issue:10798|issue:9224" + "issue:5984|issue:7322" ], - "prepared_review_unit_hash": "d2ca034298c77efcd7b2f6548660febb4f387c15821a711d6713abe98e92942e", + "prepared_review_unit_hash": "dfa7bcde0d1cfb1e382b4550d220e8657496e10b9658d7d2c579e46c3f03c763", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 33553, - "estimated_input_tokens": 8389, - "estimated_eval_tokens": 17034 + "soft_pair_count": 25, + "serialized_chars": 33915, + "estimated_input_tokens": 8479, + "estimated_eval_tokens": 17214 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated diffusers issues across Flux, SDXL, Wan, UniDiffuser, training scripts, and utility/errors. None of the soft pairs look like the same underlying bug, so all should stay separate.", - "confidence": 0.9, - "canonical_issue_reason": "No single canonical issue: the items span multiple distinct pipelines, models, and failure modes rather than one shared bug.", + "summary": "This is a loose thematic cluster of diffusion/model bugs, not a true duplicate set. The soft pairs mostly share only broad subsystem similarity, so none should be merged as duplicates.", + "confidence": 0.61, + "canonical_issue_reason": "issue:11540 is the clearest, most concrete bug report in the set: it names an exact pipeline regression and a second closely related pipeline limitation, making it the best representative issue even though the cluster itself is heterogeneous.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10302 is the most concrete active bug report here: it names one reproducible Flux inference noise problem with a specific trigger (FP8 without CPU offloading).", + "best_issue_reason": "issue:11540 is the strongest standalone issue for triage because it is specific, reproducible, and actionable, with an explicit symptom and affected APIs.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2659", + "right": "issue:6091", + "accept": false, + "reason": "Different problems: batch-size nondeterminism in SD vs consistency decoder memory usage with a size argument." + }, + { + "left": "issue:6232", + "right": "issue:8323", + "accept": false, + "reason": "Both are scheduler-related only in the broadest sense; one is a DPMSolver add_noise bug, the other is an unreadable scheduler filename issue." + }, + { + "left": "issue:11581", + "right": "issue:11659", + "accept": false, + "reason": "Both involve LoRA/IP-adapter style loading on FLUX/Wan, but the concrete failures and code paths differ." + }, { "left": "issue:10120", "right": "issue:6230", "accept": false, - "reason": "Both mention memory/OOM, but they are different training scenarios and models (DreamBooth SD3 vs SDXL text-to-image)." + "reason": "Dreambooth SD3 memory consumption is not the same as SDXL training out-of-memory." }, { "left": "issue:11163", "right": "issue:9106", "accept": false, - "reason": "Different code paths and failures: Wan latent prep math vs PixArt time_step attribute error." + "reason": "Wan image-to-video latent math bug is unrelated to PixArt time_step attribute errors." }, { "left": "issue:11657", "right": "issue:11659", "accept": false, - "reason": "Both are LoRA-related, but one is a Wan key-prefix mismatch warning and the other is a vague general load issue; not clearly the same bug." + "reason": "Both mention LoRA loading, but one is a silent Wan LoRA key mismatch and the other is a generic load issue with different context." }, { "left": "issue:5270", "right": "issue:6091", "accept": false, - "reason": "A notebook memory leak is not the same as a Consistency Decoder size-argument memory issue." + "reason": "Notebook memory leak and consistency decoder size-related memory issue are different symptoms and code paths." }, { "left": "issue:10315", "right": "issue:8707", "accept": false, - "reason": "Different training failures in different pipelines (CogVideo vs SDXL text-to-image)." + "reason": "CogVideo training error and SDXL text-to-image training error are unrelated training failures." }, { "left": "issue:2851", "right": "issue:3026", "accept": false, - "reason": "Unrelated problems: TPU push_to_hub failure vs single-GPU instruct_pix2pix launch error." + "reason": "TPU push_to_hub failure is unrelated to instruct-pix2pix single-GPU launch errors." }, { "left": "issue:4484", "right": "issue:6598", "accept": false, - "reason": "Both report bad outputs, but one is SDXL Kohya LoRA weirdness and the other is img2img parity with A1111; not the same concrete bug." + "reason": "Weird SDXL Kohya LoRA results and img2img differences vs A1111 are different model/workflow behaviors." }, { "left": "issue:10302", "right": "issue:11540", "accept": false, - "reason": "Both involve Flux noise, but the triggers differ (FP8 without offloading vs VAE slicing); likely separate code paths." + "reason": "FP8 noise without CPU offloading is a precision/offloading issue, not the same as VAE slicing noise or missing slicing support." }, { "left": "issue:3194", "right": "issue:7210", "accept": false, - "reason": "bitsandbytes installation metadata error is unrelated to the missing named_parameters attribute on SDXL pipelines." + "reason": "bitsandbytes packaging error is unrelated to accelerate dispatch on StableDiffusionXLPipeline." }, { "left": "issue:8759", "right": "issue:9944", "accept": false, - "reason": "Different SD3/SD3.5 issues: invalid cast warning vs loading size mismatch." + "reason": "SD3 image_processor NaN warning and SD3.5 size mismatch are different failure modes." }, { "left": "issue:10121", "right": "issue:7355", "accept": false, - "reason": "High-resolution Flux1-Dev generation failure is not the same as Stable Cascade arbitrary-resolution/bfloat16 errors." + "reason": "Flux high-resolution generation failure does not match Stable Cascade arbitrary-resolution+bfloat16 errors." }, { "left": "issue:2838", "right": "issue:939", "accept": false, - "reason": "ControlNet pipeline behavior is unrelated to using multiple community pipes in a list." + "reason": "ControlNet pipeline support question is unrelated to using multiple community pipes in a list." }, { "left": "issue:3097", "right": "issue:3576", "accept": false, - "reason": "Textual inversion no-op vs XFormers TypeError are different failures, even if they touch the same feature area." + "reason": "Textual inversion doing nothing is not the same as the xformers TypeError during textual inversion." }, { "left": "issue:3631", "right": "issue:898", "accept": false, - "reason": "Inpainting with latents and a TPU bfloat16 scalar-type exception are separate problems." + "reason": "Inpainting latents exception and TPU bfloat16 scalar type exception are different pipelines and errors." }, { "left": "issue:3951", "right": "issue:5592", "accept": false, - "reason": "NaNs in UniDiffuser demo output is not the same as failing to load UniDiffuser checkpoints." + "reason": "UniDiffuser NaNs in demo code and load-from-checkpoints failure are distinct issues." }, { "left": "issue:3975", "right": "issue:4618", "accept": false, - "reason": "A typo in an error message is unrelated to using a DreamShaper XL model with ControlNet." + "reason": "Safetensors error-message typo is unrelated to ControlNet usage guidance." }, { "left": "issue:5884", "right": "issue:8710", "accept": false, - "reason": "Feature requests for IP-Adapter and PAG support are distinct changes, not duplicates." + "reason": "IP-Adapter support expansion and PAG support request are different feature requests, not the same change." }, { "left": "issue:5984", "right": "issue:9664", "accept": false, - "reason": "Both involve train_dreambooth_lora_sdxl.py, but one is a RAM issue and the other is a generic subprocess failure." + "reason": "Colab system RAM issue in SDXL LoRA training and a subprocess error in a different training script are not the same bug." }, { "left": "issue:6610", "right": "issue:8450", "accept": false, - "reason": "Both are inpainting-related, but one is about num_in_channels being ignored and the other is about strength=1.0 noise." + "reason": "from_single_file num_in_channels ignored and SDXL inpainting strength=1.0 noise are different inpainting bugs." }, { "left": "issue:9163", "right": "issue:9464", "accept": false, - "reason": "Both are train_dreambooth_lora_flux.py issues, but one is a generic tensor bug and the other is a specific IndexError with prodigy/train_text_encoder." + "reason": "Tensor bugs in FLUX dreambooth LoRA training and a Prodigy/train_text_encoder index error are separate training failures." }, { "left": "issue:9292", "right": "issue:9361", "accept": false, - "reason": "LoRA key compatibility warnings and LoRA having no effect are related in theme, but not the same concrete failure." + "reason": "X-Labs FLUX LoRA key compatibility and FLUXPipeline LoRA effect missing are related only at a high level; the concrete incompatibilities differ." }, { "left": "issue:5984", "right": "issue:7322", "accept": false, - "reason": "Different training failures: fp16 Colab RAM issue vs SIGKILL during a newer diffusers training run." - }, - { - "left": "issue:6921", - "right": "issue:9200", - "accept": false, - "reason": "SDXL weight mismatch and a StableDiffusionXLControlNetPipeline dtype error are unrelated issues." - }, - { - "left": "issue:10798", - "right": "issue:9224", - "accept": false, - "reason": "Flux quantization with bitsandbytes is a different problem from MPS not supporting hardcoded float64." + "reason": "SDXL LoRA Colab RAM issue and dreambooth_lora_sdxl SIGKILL training error are both resource-related but not the same underlying bug." } ] }, @@ -93156,13 +93176,13 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:10695", "issue:10710", + "issue:10798", "issue:1084", "issue:1106", "issue:11092", @@ -93172,7 +93192,6 @@ "issue:11790", "issue:12373", "issue:12606", - "issue:12761", "issue:1730", "issue:2299", "issue:2414", @@ -93200,16 +93219,19 @@ "issue:6636", "issue:6686", "issue:6783", + "issue:6921", "issue:776", "issue:7939", "issue:8373", - "issue:8667", "issue:8956", "issue:9112", - "issue:9460", + "issue:9200", + "issue:9224", "issue:9844" ], "soft_pairs": [ + "issue:6921|issue:9200", + "issue:10798|issue:9224", "issue:2299|issue:4977", "issue:1106|issue:2552", "issue:11790|issue:12606", @@ -93231,182 +93253,180 @@ "issue:11092|issue:9112", "issue:2716|issue:6176", "issue:4031|issue:5553", - "issue:4043|issue:5081", - "issue:8667|issue:9460", - "issue:10695|issue:12761" + "issue:4043|issue:5081" ], - "prepared_review_unit_hash": "f9d0c517c100d151d2dd2d6506b5b989510ac278b86c9e933893fcea426695bd", + "prepared_review_unit_hash": "7acc606cf2659159f5658406d6d68c8fac9b689fb10bead4be9d1049bdbb0667", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33265, - "estimated_input_tokens": 8317, - "estimated_eval_tokens": 16890 + "serialized_chars": 33186, + "estimated_input_tokens": 8297, + "estimated_eval_tokens": 16850 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is not a duplicate set: it mixes unrelated bugs, feature requests, docs questions, and training/loading issues across different subsystems. No soft-pair looks like the same underlying change or defect, so all soft edges should be rejected.", - "confidence": 0.97, - "canonical_issue_reason": "issue:10695 is the most concrete, code-path-specific bug report in the set (a scheduler timestep calculation defect) and is a plausible anchor if one issue must be chosen.", + "summary": "This cluster is heterogeneous and mostly false-positive grouping: it mixes unrelated loading, training, inference, docs, and feature-request issues. If a representative issue is needed, the broadest loader-related report is the best fit, but there is no strong true duplicate center.", + "confidence": 0.93, + "canonical_issue_reason": "issue:5553 is the broadest and most generic loader-failure report in the set, so it is the closest weak representative. That said, the cluster is not a cohesive duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10695 is the strongest representative issue because it describes a specific, actionable bug in a core algorithm rather than a broad question or feature request.", + "best_issue_reason": "issue:5553 best matches the widest subset of titles about model/pipeline loading failures, making it the most plausible cluster anchor despite low semantic cohesion.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6921", + "right": "issue:9200", + "accept": false, + "reason": "Official SDXL weight mismatch vs ControlNet dtype error; different bug classes and code paths." + }, + { + "left": "issue:10798", + "right": "issue:9224", + "accept": false, + "reason": "Bitsandbytes flux quantization/device-combo bug vs hardcoded float64 on MPS; both precision-related, but not the same issue." + }, { "left": "issue:2299", "right": "issue:4977", "accept": false, - "reason": "Different problems: model conversion/checkpoint compatibility vs a training-example checkpoint-saving behavior." + "reason": "Checkpoint conversion/compatibility problem vs example cleanup about checkpoint removal; unrelated changes." }, { "left": "issue:1106", "right": "issue:2552", "accept": false, - "reason": "Unrelated topics: PNDM UNet call frequency vs inpainting tutorial/documentation." + "reason": "PNDM repeatedly invoking UNet vs inpainting model tutorial question; different subjects entirely." }, { "left": "issue:11790", "right": "issue:12606", "accept": false, - "reason": "Different bug areas: LoRA weight loading from training output vs a wrong import in a conversion script." + "reason": "LoRA weight loading failure vs wrong import in an SD3 conversion script; distinct failures." }, { "left": "issue:251", "right": "issue:776", "accept": false, - "reason": "Both concern image generation quality, but the reported defects are different: latent fragility vs img2img batch-size behavior." + "reason": "Latent distribution fragility vs batch_size>1 img2img breakage; both image generation issues, but not the same bug." }, { "left": "issue:4186", "right": "issue:6783", "accept": false, - "reason": "Different models and failures: SDXL inpaint/refiner behavior vs PixArt generation corruption." + "reason": "SDXL inpaint/refiner failure vs PixArt generations corrupted; different models and failure modes." }, { "left": "issue:4399", "right": "issue:5763", "accept": false, - "reason": "Unrelated code paths: PyTorch compile/offload interaction vs safetensors serialization on save." + "reason": "torch.compile plus CPU offload incompatibility vs save_pretrained not writing safetensors; unrelated subsystems." }, { "left": "issue:10710", "right": "issue:8373", "accept": false, - "reason": "A format-support question is not the same as an incorrect error-message bug." + "reason": "Feature request for DDUF support vs incorrect error message; not the same underlying change." }, { "left": "issue:6686", "right": "issue:9844", "accept": false, - "reason": "Different symptoms and layers: compilation performance after save/load vs VAE encoder NaNs." + "reason": "Dynamo compile perf regression after save/load vs VAE encoder NaNs; different modules and symptoms." }, { "left": "issue:1084", "right": "issue:307", "accept": false, - "reason": "Docs request about weight reuse is not the same as an architectural pipeline-merging feature request." + "reason": "Documentation around weight reuse vs merging pipelines into one; related theme, but not one concrete bug/change." }, { "left": "issue:11374", "right": "issue:12373", "accept": false, - "reason": "Completely different asks: adding SkyReels support vs an Unlearn API." + "reason": "SkyReels-V2 integration request vs Unlearn API request; separate feature proposals." }, { "left": "issue:11447", "right": "issue:11448", "accept": false, - "reason": "Different bugs: offline shard-file resolution vs a missing model path." + "reason": "Offline checkpoint shard lookup bug vs missing fusing/stable-unclip model path; both loading-related, but different concrete problems." }, { "left": "issue:1730", "right": "issue:2442", "accept": false, - "reason": "Unrelated: checkpoint conversion failing in webui vs a question about loss curves for inpainting finetuning." + "reason": "Model conversion to CKPT/webui compatibility vs training loss-graph question; different workflows and issues." }, { "left": "issue:2414", "right": "issue:6636", "accept": false, - "reason": "Different import failures affecting different modules and root causes." + "reason": "Missing transformers import during hub load vs missing DIFFUSERS_SLOW_IMPORT symbol; distinct import failures." }, { "left": "issue:4031", "right": "issue:4874", "accept": false, - "reason": "Both involve loading-related behavior, but one is a safety-checker local-file failure and the other is a low_res_scheduler bug in upscale pipeline loading." + "reason": "Local safety-checker load failure vs low_res_scheduler bug when loading from local; different components." }, { "left": "issue:4476", "right": "issue:5412", "accept": false, - "reason": "Different phases and failures: inference quality on SDXL+ControlNet vs checkpoint resume training failure." + "reason": "Inference quality issue with SDXL+ControlNet vs training resume_from_checkpoint failure; unrelated code paths." }, { "left": "issue:4641", "right": "issue:4697", "accept": false, - "reason": "Both are LoRA-loading issues, but one is about local native LoRA weights with load_lora_weights and the other is a specific LoRa after an update; no clear same root cause." + "reason": "Local-file LoRA loading regression vs specific LoRA failing after an update; same broad area, but too unspecific to treat as the same bug." }, { "left": "issue:5135", "right": "issue:6008", "accept": false, - "reason": "Unrelated: LoRA quality degradation/xformers suspicion vs IP-Adapter Plus Face failing with an LCM LoRA." + "reason": "LoRA quality degradation possibly from xformers vs IP-Adapter Plus Face failing with LCM LoRA; different features and symptoms." }, { "left": "issue:7939", "right": "issue:8956", "accept": false, - "reason": "Both are model-loading complaints, but they target different pipelines/models and different error conditions." + "reason": "Same-folder model load failure vs OVStableDiffusionXLPipeline model load failure; both loading issues, but not the same path or error." }, { "left": "issue:11092", "right": "issue:9112", "accept": false, - "reason": "Different inference bugs in different models: empty timesteps in SDXL-Turbo img2img vs Auraflow tensor-size mismatch." + "reason": "SDXL Img2Img empty timesteps bug vs Auraflow tensor-size mismatch; unrelated pipelines and errors." }, { "left": "issue:2716", "right": "issue:6176", "accept": false, - "reason": "A general 'how to load checkpoint' question is not the same as a LoRA weight loading error after training." + "reason": "General question about loading checkpoints vs error loading LoRA after DreamBooth training; not a duplicate." }, { "left": "issue:4031", "right": "issue:5553", "accept": false, - "reason": "Different loading failures: stand-alone safety checker from local file vs a custom model load failure." + "reason": "Safety checker local-file load failure vs custom model load failure; both are loader bugs, but different components and failure modes." }, { "left": "issue:4043", "right": "issue:5081", "accept": false, - "reason": "Both touch SDXL text-embedding handling, but one is an API requirement for pooled embeddings and the other is an LPW XL matmul shape mismatch." - }, - { - "left": "issue:8667", - "right": "issue:9460", - "accept": false, - "reason": "Different SD3ControlNet problems: non-power-of-two image tensor mismatch vs supporting a list of ControlNet models." - }, - { - "left": "issue:10695", - "right": "issue:12761", - "accept": false, - "reason": "Completely different bugs: DDIMInverseScheduler timestep logic vs a backward-gradient count mismatch." + "reason": "Missing pooled prompt embeddings for SDXL vs LPW XL matrix-shape mismatch; both shape-related, but different model/input requirements." } ] }, @@ -93430,15 +93450,15 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10695", "issue:10768", "issue:10772", "issue:10893", - "issue:1214", + "issue:12761", "issue:2201", + "issue:2392", "issue:2550", "issue:272", - "issue:2763", - "issue:2908", "issue:2991", "issue:3181", "issue:3403", @@ -93449,14 +93469,13 @@ "issue:3682", "issue:3876", "issue:3935", + "issue:3951", "issue:4001", "issue:4118", - "issue:4185", "issue:4304", "issue:4306", "issue:4675", "issue:4804", - "issue:4826", "issue:5043", "issue:5365", "issue:5794", @@ -93470,18 +93489,22 @@ "issue:7813", "issue:8390", "issue:8503", - "issue:8659", + "issue:8667", "issue:8855", "issue:8898", "issue:8930", "issue:9225", + "issue:9460", "issue:9539", "issue:9756" ], "soft_pairs": [ + "issue:8667|issue:9460", + "issue:10695|issue:12761", "issue:10768|issue:10772", "issue:10893|issue:8390", "issue:2201|issue:6598", + "issue:2392|issue:3951", "issue:2550|issue:5875", "issue:272|issue:4304", "issue:3181|issue:3624", @@ -93499,183 +93522,180 @@ "issue:6335|issue:8930", "issue:3643|issue:5365", "issue:3403|issue:7286", - "issue:2991|issue:3467", - "issue:1214|issue:2908", - "issue:4185|issue:8659", - "issue:2763|issue:4826" + "issue:2991|issue:3467" ], - "prepared_review_unit_hash": "f53323b89da89130a4e8806b9dea0e9b6968fe2910d88a8a9a99125f465ad58e", + "prepared_review_unit_hash": "89b9f6f46c6c1d3cb06f448d113ff949f78f4cd9062df905a25a02529c244dd6", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33220, - "estimated_input_tokens": 8305, - "estimated_eval_tokens": 16866 + "serialized_chars": 33323, + "estimated_input_tokens": 8331, + "estimated_eval_tokens": 16918 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: it mixes unrelated feature requests, documentation tasks, and distinct bug reports across different models/pipelines. The soft links are mostly broad topical similarity (e.g. ControlNet, loading, mixed precision) rather than the same concrete issue.", - "confidence": 0.92, - "canonical_issue_reason": null, - "canonical_pr_reason": null, - "best_issue_reason": "No single canonical issue cleanly represents the cluster; if one must be chosen, issue:8930 is the most substantive and detailed open item, but it is still not a duplicate umbrella for the rest.", - "best_pr_reason": null, + "summary": "This cluster is heterogeneous: the soft pairs mostly connect unrelated bugs/features across different pipelines, schedulers, loaders, and docs. None look like true duplicates of the same underlying change or defect.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue is a good canonical here; the items span unrelated topics, so duplicate triage should stay separate.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "No issue stands out as a global best representative because the cluster does not share one concrete bug or feature area.", + "best_pr_reason": "No pull requests are present in this cluster.", "soft_edge_verdicts": [ + { + "left": "issue:8667", + "right": "issue:9460", + "accept": false, + "reason": "Different problems: SD3ControlNet tensor-size mismatch vs adding list support for multiple ControlNet models." + }, + { + "left": "issue:10695", + "right": "issue:12761", + "accept": false, + "reason": "Scheduler timestep calculation bug vs custom attention backward gradient-count error; unrelated code paths." + }, { "left": "issue:10768", "right": "issue:10772", "accept": false, - "reason": "Different feature requests for different models: negative prompting in Hunyuan video vs ControlNet support for Sana." + "reason": "Negative-prompt support question for Hunyuan video vs Sana ControlNet support request." }, { "left": "issue:10893", "right": "issue:8390", "accept": false, - "reason": "Both mention ControlNet, but one asks for CogVideoX integration and the other asks about Stable Cascade training; different model families and changes." + "reason": "CogVideoX ControlNet integration request vs Stable Cascade ControlNet training request; different features/models." }, { "left": "issue:2201", "right": "issue:6598", "accept": false, - "reason": "Tokenization padding behavior vs img2img output differences are separate bugs with different code paths." + "reason": "Tokenizer padding behavior mismatch vs img2img prompt_embeds reproducibility; not the same defect." + }, + { + "left": "issue:2392", + "right": "issue:3951", + "accept": false, + "reason": "Img2img prompt_embeds validation error vs UniDiffuser NaNs; different pipelines and symptoms." }, { "left": "issue:2550", "right": "issue:5875", "accept": false, - "reason": "Both are conversion-related, but one is ControlNet-to-diffusers conversion and the other is SDXL-to-ONNX prompt handling; not the same failure." + "reason": "ControlNet-to-diffusers conversion issue vs ONNX conversion prompt error; different conversion paths." }, { "left": "issue:272", "right": "issue:4304", "accept": false, - "reason": "Inpainting scheduler support vs SDXL LoRA training are unrelated feature requests." + "reason": "K-LMS inpainting support vs SDXL LoRA training support; unrelated feature requests." }, { "left": "issue:3181", "right": "issue:3624", "accept": false, - "reason": "Generic TypeError wording is not enough; these affect different examples and different underlying causes." + "reason": "TensorRT example TypeError vs generic getattr attribute-name TypeError; different failures." }, { "left": "issue:3578", "right": "issue:4118", "accept": false, - "reason": "Both involve AutoencoderKL loading, but one is checkpoint loading failure and the other is a missing from_single_file attribute; different APIs/issues." + "reason": "AutoencoderKL checkpoint loading failure vs missing from_single_file attribute; related component, different bug." }, { "left": "issue:3682", "right": "issue:725", "accept": false, - "reason": "LoRA weight loading and pipeline device-map/accelerate support are different features." + "reason": "LoRA loading weights vs device_map/accelerate loading memory footprint; separate loading concerns." }, { "left": "issue:3935", "right": "issue:4001", "accept": false, - "reason": "Both are docs tasks, but they target different documentation changes with no shared bug or implementation." + "reason": "Docstrings for feed-forward chunking vs README demo-space docs; documentation tasks are unrelated." }, { "left": "issue:4675", "right": "issue:8390", "accept": false, - "reason": "ControlNet for SDXL vs Stable Cascade ControlNet training are different model targets and different work items." + "reason": "SDXL reference-only ControlNet feature vs Stable Cascade ControlNet training; different model families." }, { "left": "issue:4804", "right": "issue:9225", "accept": false, - "reason": "Dictionary-iteration error in LoRA loading vs latent-shape mismatch in SDXL InstructPix2Pix are unrelated bugs." + "reason": "LoRA loader iteration bug vs rgba latent size mismatch; different runtime issues." }, { "left": "issue:5043", "right": "issue:7736", "accept": false, - "reason": "LoRA finetuning for Wurstchen and supported callbacks are unrelated feature requests." + "reason": "W\u00fcrstchen LoRA finetuning request vs supported callbacks API; not the same change." }, { "left": "issue:5794", "right": "issue:8898", "accept": false, - "reason": "Both are loading/path problems, but they concern different missing-file/validation scenarios and different call sites." + "reason": "Missing tokenizer/config.json in SDXL base vs from_pretrained subfolder path resolution error; different loading bugs." }, { "left": "issue:7237", "right": "issue:9756", "accept": false, - "reason": "A missing scheduler class is unrelated to a missing notebook file." + "reason": "Missing EDMEulerScheduler class vs missing notebook file; unrelated." }, { "left": "issue:7813", "right": "issue:8855", "accept": false, - "reason": "Tensor timesteps question and attention-kwargs propagation are different API concerns." + "reason": "Question about passing timesteps as tensors vs joint_attention_kwargs not reaching the AttentionProcessor." }, { "left": "issue:8503", "right": "issue:9539", "accept": false, - "reason": "Different model/runtime failures: SD3 gradient checkpointing crash vs CogVideoX float8 CUDA indexing error." + "reason": "SD3 gradient checkpointing crash vs CogVideoX Float8 index_select error; different model/runtime issues." }, { "left": "issue:3876", "right": "issue:4306", "accept": false, - "reason": "Different pipelines and different runtime failures: ControlNet inpaint demo error vs ReferencePipeline list-index error." + "reason": "ControlNet inpaint demo error vs StableDiffusionReferencePipeline list-index error; different pipelines." }, { "left": "issue:6335", "right": "issue:8930", "accept": false, - "reason": "Both are ControlNet img2img issues, but one is output-size control and the other is pose-image adherence; not the same bug." + "reason": "Output-size control in ControlNet img2img vs pose-conditioning not being respected; both ControlNet but not the same bug." }, { "left": "issue:3643", "right": "issue:5365", "accept": false, - "reason": "Latent upsample INT_MAX error and txt2video inference-step handling are unrelated." + "reason": "Latent decode INT_MAX overflow vs txt2video num_inference_steps error; unrelated symptoms." }, { "left": "issue:3403", "right": "issue:7286", "accept": false, - "reason": "TorchDynamo/pipeline loading and custom diffusion cross-attention weight loading are different code paths." + "reason": "TorchDynamo pipeline-loading issue vs custom diffusion cross-attention weight loading; different components." }, { "left": "issue:2991", "right": "issue:3467", "accept": false, - "reason": "Mixed-precision example failure and checkpoint-resume step offset are distinct training bugs." - }, - { - "left": "issue:1214", - "right": "issue:2908", - "accept": false, - "reason": "Half-precision CLIPTextModel and array-shape/mixed-precision training-script errors are not the same issue." - }, - { - "left": "issue:4185", - "right": "issue:8659", - "accept": false, - "reason": "SDXL black-image inference after rebasing and checkpoint resume/load failures are different problems." - }, - { - "left": "issue:2763", - "right": "issue:4826", - "accept": false, - "reason": "VRAM usage regression with cudnn.benchmark and MultiControlNet errors are separate ControlNet issues." + "reason": "Mixed-precision ControlNet example failure vs wrong-step resume with gradient accumulation; distinct training bugs." } ] }, @@ -93700,19 +93720,20 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:1205", + "issue:1214", "issue:1271", "issue:1617", - "issue:1821", - "issue:2097", "issue:2326", "issue:234", "issue:235", "issue:2364", "issue:2596", "issue:2606", + "issue:2763", "issue:2782", "issue:2880", "issue:2905", + "issue:2908", "issue:3081", "issue:3222", "issue:3224", @@ -93721,11 +93742,11 @@ "issue:3591", "issue:3740", "issue:3884", + "issue:4185", "issue:4616", "issue:4709", - "issue:4726", + "issue:4826", "issue:4936", - "issue:5075", "issue:5538", "issue:5605", "issue:5630", @@ -93739,15 +93760,18 @@ "issue:7300", "issue:7415", "issue:7483", - "issue:7535", "issue:7600", "issue:8226", "issue:8551", + "issue:8659", "issue:9001", "issue:9610", "issue:970" ], "soft_pairs": [ + "issue:1214|issue:2908", + "issue:4185|issue:8659", + "issue:2763|issue:4826", "issue:1205|issue:2782", "issue:2905|issue:6354", "issue:4616|issue:970", @@ -93768,183 +93792,180 @@ "issue:1271|issue:2364", "issue:3740|issue:7483", "issue:234|issue:3573", - "issue:3591|issue:8226", - "issue:1821|issue:3222", - "issue:2097|issue:5075", - "issue:4726|issue:7535" + "issue:3591|issue:8226" ], - "prepared_review_unit_hash": "fff9bce5586f29253b3e368d3d80be172e93d953fc093f46d9a76b7ecdec1d34", + "prepared_review_unit_hash": "719d2a95506d5dc90be907cf843f56898f638baf0b3af74c3735b15c78ddb77f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33229, - "estimated_input_tokens": 8308, - "estimated_eval_tokens": 16872 + "serialized_chars": 33801, + "estimated_input_tokens": 8451, + "estimated_eval_tokens": 17158 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a broad mix of unrelated Stable Diffusion / Diffusers issues; the soft-similarity links are mostly superficial keyword matches, not duplicate bugs or changes.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "The cluster is heterogeneous: most items are distinct bugs or feature requests spanning conversion, training scripts, ControlNet, schedulers, and docs. There is only loose thematic overlap around precision/memory/runtime issues, but no true duplicate pair stands out.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 234 is the broadest representative of the half-precision/runtime bug theme and is the closest fit as a canonical issue, though the overall set is not a tight duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 234 is the most general user-facing failure report here and overlaps thematically with several precision-related complaints better than the narrower, script-specific issues.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1214", + "right": "issue:2908", + "accept": false, + "reason": "Both mention mixed/half precision, but one is CLIPTextModel inference and the other is a ControlNet training-script shape issue; different code paths." + }, + { + "left": "issue:4185", + "right": "issue:8659", + "accept": false, + "reason": "SDXL/ControlNet context overlaps, but one is black-image inference and the other is checkpoint resume/load behavior in a training script." + }, + { + "left": "issue:2763", + "right": "issue:4826", + "accept": false, + "reason": "Both are ControlNet-related, but one is VRAM usage with cudnn benchmarking and the other is MultiControlNet reference handling errors; not the same bug." + }, { "left": "issue:1205", "right": "issue:2782", "accept": false, - "reason": "One is about converting a model to ONNX, the other is a sequential offloading error; different code paths and symptoms." + "reason": "One is SD-to-ONNX conversion and the other is a sequential offloading error; different features and failure modes." }, { "left": "issue:2905", "right": "issue:6354", "accept": false, - "reason": "Scheduler feature request versus ControlNet LoRA support request; unrelated changes." + "reason": "A scheduler feature request and a ControlNet LoRA support request are unrelated change requests." }, { "left": "issue:4616", "right": "issue:970", "accept": false, - "reason": "LoRA conversion failure for SDXL is unrelated to integration tests failing." + "reason": "LoRA conversion for SDXL and failed integration tests are unrelated; no shared concrete code-path bug." }, { "left": "issue:5630", "right": "issue:9001", "accept": false, - "reason": "SDXL img2img scheduler noise issue and Auraflow aspect ratios are different pipelines/problems." + "reason": "Both concern pipeline behavior, but one is LMS noise in SDXL Img2Img and the other is Auraflow aspect ratios; different models and bugs." }, { "left": "issue:6162", "right": "issue:6336", "accept": false, - "reason": "IP-Adapter embedding shape mismatch is not the same as TensorRT VAE compilation with AutoencoderTiny." + "reason": "IP-Adapter shape mismatch in SDXL is unrelated to TensorRT VAE compilation with AutoencoderTiny." }, { "left": "issue:2606", "right": "issue:3884", "accept": false, - "reason": "DeepSpeed save_state hang in LoRA training is distinct from a generic training-script error." + "reason": "Both involve training scripts failing, but one is DeepSpeed save_state hang and the other is a different LoRA training-script error." }, { "left": "issue:1617", "right": "issue:2326", "accept": false, - "reason": "Dreambooth CLI argument error and LoRA conversion error affect different scripts and failure modes." + "reason": "Both are conversion/training utility issues, but the first is missing required CLI args and the second is LoRA safetensors/ckpt conversion output errors." }, { "left": "issue:4936", "right": "issue:6520", "accept": false, - "reason": "Dreambooth resume regression and adapter validation output issue are different training behaviors." + "reason": "Both are DreamBooth/adapter training reports, but one is resume broken in Dreambooth LoRA and the other is blank validation output in T2I adapter training." }, { "left": "issue:7168", "right": "issue:8551", "accept": false, - "reason": "IP-Adapter embed shape handling is unrelated to SDXL inpaint device placement with balanced device_map." + "reason": "Both mention image embeddings/latents, but one is about IP-Adapter embedding tensor shape and the other is VAE device placement in SDXL inpaint." }, { "left": "issue:4709", "right": "issue:6164", "accept": false, - "reason": "ControlNet guess mode support and latent output handling are separate pipeline features." + "reason": "One is SDXL ControlNet guess-mode support and the other is latent output handling in StableDiffusionPipeline; feature gap vs bug in different paths." }, { "left": "issue:235", "right": "issue:2880", "accept": false, - "reason": "CUDA porting trouble and a JAX Colab runtime issue are not the same bug." + "reason": "Both mention runtime portability/cuda-like environment issues, but one is a standalone Python CUDA problem and the other is a JAX colab execution failure." }, { "left": "issue:2596", "right": "issue:3081", "accept": false, - "reason": "Missing documentation symbol and non-functional Colab notebook are different issues." + "reason": "Docs referencing a missing utility function and a broken Textual Inversion Colab are separate issues." }, { "left": "issue:3222", "right": "issue:7300", "accept": false, - "reason": "Missing output artifact in custom diffusion training is unrelated to UNet attention memory growth." + "reason": "Missing custom diffusion weights after training and high GPU memory after changing attention processor are different bugs." }, { "left": "issue:3224", "right": "issue:7600", "accept": false, - "reason": "Learning-rate scaling duplication in DreamBooth LoRA is not the same as missing accelerator.accumulate on text encoder." + "reason": "Learning-rate scaling duplication in DreamBooth LoRA and missing accelerator.accumulate on text_encoder are unrelated training-script defects." }, { "left": "issue:3453", "right": "issue:7415", "accept": false, - "reason": "BFloat16 CUDA matmul failure in text-to-image training is distinct from TPU v4 AutoencoderKL bfloat16 issues." + "reason": "Both are precision-related, but one is a CUDA bfloat16 matmul/triu issue in text-to-image training and the other is an AutoencoderKL TPU bfloat16 bug." }, { "left": "issue:5538", "right": "issue:6287", "accept": false, - "reason": "Pipeline latent usage question and setup.py dependency cleanup are unrelated." + "reason": "A pipeline implementation question and setup.py dependency cleanup are unrelated." }, { "left": "issue:5605", "right": "issue:9610", "accept": false, - "reason": "SDXL Dreambooth caption support request and SD3 img2img LoRA loader request are different features." + "reason": "Per-image caption support for SDXL DreamBooth and adding load_lora_weights to SD3 img2img are distinct feature requests." }, { "left": "issue:1271", "right": "issue:2364", "accept": false, - "reason": "GCP notebook dependency installation problem and an unable-to-run training script report are too general and not the same underlying failure." + "reason": "Both are install/run problems, but one is dependency installation on GCP notebooks and the other is a generic training-script launch failure." }, { "left": "issue:3740", "right": "issue:7483", "accept": false, - "reason": "Legacy attention block conversion with device_map is unrelated to SDXL LoRA conversion script failure." + "reason": "Both involve conversion scripts, but one is legacy attention block conversion with accelerate device_map and the other is LoRA-to-Diffusers for SDXL 1.0 base models." }, { "left": "issue:234", "right": "issue:3573", "accept": false, - "reason": "Half-precision Stable Diffusion runtime issue is not the same as a class_labels requirement in the upscaler pipeline." + "reason": "Half-precision pipeline failure versus a local upscaler class_labels requirement; different pipelines and error causes." }, { "left": "issue:3591", "right": "issue:8226", "accept": false, - "reason": "ControlNet loss-curve confusion and VectorQuantizer loss are different model/training topics." - }, - { - "left": "issue:1821", - "right": "issue:3222", - "accept": false, - "reason": "Negative prompt ineffectiveness in inpaint pipeline is unrelated to missing custom diffusion weights after training." - }, - { - "left": "issue:2097", - "right": "issue:5075", - "accept": false, - "reason": "UNet1D usage problem and SDXL refiner prompt-embedding shape mismatch are unrelated." - }, - { - "left": "issue:4726", - "right": "issue:7535", - "accept": false, - "reason": "Callback type hint complaint and FloatTensor vs Tensor typing are different API/type issues." + "reason": "ControlNet loss-curve confusion is a training/UX question, while VectorQuantizer Loss is a separate model-loss issue." } ] }, @@ -93975,15 +93996,20 @@ "issue:11144", "issue:11749", "issue:1273", + "issue:1821", "issue:2068", + "issue:2097", "issue:2228", "issue:2250", "issue:2578", + "issue:3222", "issue:3539", "issue:3970", "issue:4618", "issue:4676", + "issue:4726", "issue:4972", + "issue:5075", "issue:5150", "issue:5162", "issue:540", @@ -93993,30 +94019,28 @@ "issue:5931", "issue:5968", "issue:5998", - "issue:6242", "issue:6321", "issue:6439", - "issue:6441", "issue:6446", "issue:6561", "issue:6685", "issue:6811", "issue:6966", "issue:7049", - "issue:7054", + "issue:7535", "issue:7616", "issue:7762", - "issue:7782", "issue:7939", "issue:7945", "issue:851", "issue:9069", "issue:9187", - "issue:9613", - "issue:9660", - "issue:9670" + "issue:9613" ], "soft_pairs": [ + "issue:1821|issue:3222", + "issue:2097|issue:5075", + "issue:4726|issue:7535", "issue:11144|issue:9187", "issue:4676|issue:7049", "issue:5931|issue:7945", @@ -94037,183 +94061,180 @@ "issue:3970|issue:4618", "issue:5150|issue:6685", "issue:5405|issue:6439", - "issue:5630|issue:7616", - "issue:6242|issue:6441", - "issue:7054|issue:9660", - "issue:7782|issue:9670" + "issue:5630|issue:7616" ], - "prepared_review_unit_hash": "61bb48d77ea0e20a4bc010be1475c83122603e79ff51d884cfa129363f246c99", + "prepared_review_unit_hash": "0f24b2018100e0bacf406d1f8c28ff20ac695639c1be2a064658e7ce2d8dc53b", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33152, - "estimated_input_tokens": 8288, - "estimated_eval_tokens": 16832 + "serialized_chars": 33194, + "estimated_input_tokens": 8299, + "estimated_eval_tokens": 16854 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very noisy: the items mostly share broad Hugging Face / diffusion-loading language, but they describe different bugs, models, and code paths. I would not merge any of the proposed soft edges.", - "confidence": 0.89, - "canonical_issue_reason": "issue:2068 is the broadest representative of the cluster\u2019s recurring theme: loading-time API behavior for diffusion models, especially around `from_pretrained` type handling.", + "summary": "This cluster is mostly a set of unrelated issue reports that only share superficial diffusers/model-loading vocabulary. I do not see any true duplicate pairs among the soft edges.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 10016 is the most concrete, self-contained bug report in the set and the best representative of the cluster\u2019s strongest theme (model loading / single-file shape mismatch).", "canonical_pr_reason": null, - "best_issue_reason": "Best available anchor for the cluster, but only loosely: it captures the general model-loading / conversion behavior class better than the narrower, model-specific reports.", + "best_issue_reason": "Issue 10016 is the best standalone issue to keep because it describes a specific reproducible failure and is more actionable than the many questions, feature requests, or broad symptom reports here.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1821", + "right": "issue:3222", + "accept": false, + "reason": "Negative prompt behavior in inpaint is unrelated to missing custom diffusion output files." + }, + { + "left": "issue:2097", + "right": "issue:5075", + "accept": false, + "reason": "UNet1D usage problems and SDXL refiner prompt-embedding shape mismatch are different code paths and failures." + }, + { + "left": "issue:4726", + "right": "issue:7535", + "accept": false, + "reason": "A callback type-hint issue is unrelated to a tensor type-hint incompatibility." + }, { "left": "issue:11144", "right": "issue:9187", "accept": false, - "reason": "Different bugs: one is Flax UNet dtype initialization, the other is IP-Adapter SDXL image encoder folder handling after a Kolors update." + "reason": "Flax UNet dtype initialization and SDXL IP-Adapter image encoder handling are distinct bugs." }, { "left": "issue:4676", "right": "issue:7049", "accept": false, - "reason": "Unrelated problems: TAESD latent brightness versus an IP-Adapter unpacking error in the example code." + "reason": "TAESD latent brightness and IP-Adapter unpacking errors are unrelated runtime problems." }, { "left": "issue:5931", "right": "issue:7945", "accept": false, - "reason": "Determinism in StableDiffusionPipeline is not the same as combining community pipelines for image generation." + "reason": "Generation nondeterminism is not the same as a community-pipeline composition request." }, { "left": "issue:1111", "right": "issue:540", "accept": false, - "reason": "Completely different topics: checkpoint licensing/training data versus memory usage on Stable Diffusion." + "reason": "A checkpoint licensing/data question is unrelated to memory-placement behavior for Stable Diffusion." }, { "left": "issue:2228", "right": "issue:6446", "accept": false, - "reason": "Both involve loading, but one is GPU placement for a pipeline and the other is Colab minimal-loading behavior; not the same concrete bug." + "reason": "Pipeline GPU placement/loading issues are similar at a broad level but not the same concrete failure." }, { "left": "issue:5162", "right": "issue:5968", "accept": false, - "reason": "Different conversion requests: checkpoint-to-safetensors versus text/video-to-ONNX export." + "reason": "A checkpoint-to-safetensors converter request is unrelated to video-to-ONNX export." }, { "left": "issue:5630", "right": "issue:7762", "accept": false, - "reason": "Both concern output quality, but the underlying issues and pipelines differ; not a duplicate bug." + "reason": "SDXL img2img noise leftover and generic bad results are too broad to be the same bug." }, { "left": "issue:5998", "right": "issue:9069", "accept": false, - "reason": "A Typer argument error in a pipeline loader is unrelated to a NumPy type mismatch." + "reason": "VersatileDiffusion argument mismatch and NumPy type error are different incompatibilities." }, { "left": "issue:5919", "right": "issue:6966", "accept": false, - "reason": "Both are attribute errors, but on different objects and code paths (`set_lora_layer` vs `attn_processors`)." + "reason": "Missing LoRA layer support on Linear and missing attn_processors after forward pass are different model-internal issues." }, { "left": "issue:6561", "right": "issue:6811", "accept": false, - "reason": "Both are missing-file errors, but they concern different repos and different expected files (`config.json` vs `model_index.json`)." + "reason": "IP-Adapter local config-file lookup and i2vgen-xl model_index.json not found are separate missing-file/load-path failures." }, { "left": "issue:10016", "right": "issue:7939", "accept": false, - "reason": "SD3.5M single-file shape mismatch is unrelated to loading a model from the same folder." + "reason": "Single-file SD3.5M shape mismatch is not the same as failing to load a model from the same folder." }, { "left": "issue:2068", "right": "issue:851", "accept": false, - "reason": "Both mention dtype, but one is about `from_pretrained` not converting types while the other is a separate Float-vs-Half runtime error." + "reason": "from_pretrained dtype conversion complaint and a Half-vs-Float mismatch are related only loosely, not the same bug." }, { "left": "issue:2578", "right": "issue:6321", "accept": false, - "reason": "Editable install failure is unrelated to a missing CLIP model detection issue." + "reason": "Editable install failure and CLIP model detection failure are unrelated environment/setup problems." }, { "left": "issue:2250", "right": "issue:4972", "accept": false, - "reason": "Dreambooth prompt-generation failure and Wuerstchen example failure are different workflows and failures." + "reason": "Dreambooth prompt generation failure is not the same as a wuerstchen example runtime failure." }, { "left": "issue:1055", "right": "issue:1099", "accept": false, - "reason": "A missing pipeline class and a missing Dreambooth class name are different API availability issues." + "reason": "Missing LatentDiffusionPipeline and missing dreambooth class name are different missing-API issues." }, { "left": "issue:11749", "right": "issue:9613", "accept": false, - "reason": "Both touch scheduler/timesteps internals, but one is floating-point precision and the other is a missing `scale` attribute after a version change." + "reason": "Floating-point timestep errors and a missing scale attribute on Timesteps are different scheduler bugs." }, { "left": "issue:1273", "right": "issue:3539", "accept": false, - "reason": "Training-loss question and `main_process_first` usage question are unrelated." + "reason": "A mean-loss calculation question is not the same as a question about main_process_first semantics." }, { "left": "issue:3970", "right": "issue:4618", "accept": false, - "reason": "Prompt embeddings not working is not the same as asking how to use a specific SDXL checkpoint with ControlNet." + "reason": "SDXL prompt embeddings not working and using a specific checkpoint with ControlNet are not the same underlying problem." }, { "left": "issue:5150", "right": "issue:6685", "accept": false, - "reason": "Missing `from_single_file` support and unsupported absolute output paths are different feature/validation issues." + "reason": "Missing from_single_file support and unsupported absolute output paths in Advanced Dreambooth are unrelated." }, { "left": "issue:5405", "right": "issue:6439", "accept": false, - "reason": "Single-file weight overwrite behavior and a docs 404 are unrelated." + "reason": "SDXL from_single_file overwriting weights and a broken docs link are not the same issue." }, { "left": "issue:5630", "right": "issue:7616", "accept": false, - "reason": "Both are image-quality complaints, but one is scheduler noise in SDXL img2img and the other is low-denoise artifacts in a different setup." - }, - { - "left": "issue:6242", - "right": "issue:6441", - "accept": false, - "reason": "A request for dataset logs is unrelated to a Colab `from_pretrained` failure." - }, - { - "left": "issue:7054", - "right": "issue:9660", - "accept": false, - "reason": "LoRA state-dict mutation and module eval-mode handling are separate loading-behavior bugs." - }, - { - "left": "issue:7782", - "right": "issue:9670", - "accept": false, - "reason": "Invalid safetensors conversion input and a missing PEFT backend for LoRA loading are different errors." + "reason": "Both concern SDXL img2img quality, but one is scheduler-specific noise and the other is low-denoise artifacts; not identical enough." } ] }, @@ -94258,9 +94279,6 @@ "issue:2761", "issue:3171", "issue:347", - "issue:3728", - "issue:4263", - "issue:4363", "issue:4701", "issue:4773", "issue:4919", @@ -94268,25 +94286,31 @@ "issue:547", "issue:5574", "issue:5578", - "issue:5989", + "issue:6242", "issue:6297", "issue:6412", + "issue:6441", "issue:6506", + "issue:7054", "issue:7147", "issue:7183", "issue:7322", "issue:7450", "issue:759", - "issue:7864", + "issue:7782", "issue:7946", "issue:8906", - "issue:8918", "issue:9515", "issue:9516", "issue:9648", + "issue:9660", + "issue:9670", "issue:9836" ], "soft_pairs": [ + "issue:6242|issue:6441", + "issue:7054|issue:9660", + "issue:7782|issue:9670", "issue:8906|issue:9515", "issue:11088|issue:12082", "issue:1363|issue:1918", @@ -94307,75 +94331,90 @@ "issue:11547|issue:13065", "issue:12060|issue:9836", "issue:2761|issue:547", - "issue:3171|issue:6412", - "issue:3728|issue:7864", - "issue:4263|issue:8918", - "issue:4363|issue:5989" + "issue:3171|issue:6412" ], - "prepared_review_unit_hash": "9a7ce842e34eaedef82f4bae43a8896979a603ccc8d932325a545d5dbf5a54a6", + "prepared_review_unit_hash": "c42fc934543e499c8cbf2f27516992defc4010e728707c25499ffbb8a5287854", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33699, - "estimated_input_tokens": 8425, - "estimated_eval_tokens": 17106 + "serialized_chars": 33821, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous set of unrelated model, pipeline, memory, and training issues. The only plausible duplicate pair is the SD3Transformer2DModel attention/normalization reports (8906/9515); the rest do not share the same concrete bug path.", - "confidence": 0.81, - "canonical_issue_reason": "issue:9515 is the clearest candidate in the only plausible duplicate pair, because it names the SD3Transformer2DModel/JointAttnProcessor2_0 normalization gap directly, while 8906 is a shorter and less explicit question about qk_norm.", + "summary": "The items are a heterogeneous set of unrelated bug reports and feature/support questions connected only by shallow title similarity. None of the soft-paired artifacts look like the same underlying issue, so all soft edges should be rejected and there is no good global canonical issue/PR for this cluster.", + "confidence": 0.98, + "canonical_issue_reason": "No single issue is a valid canonical representative: the cluster mixes unrelated topics (loading errors, OOMs, controlnet, LoRA, schedulers, docs, features) with no evidence of one shared bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:12113 is the strongest standalone bug report in this set: it identifies a specific WanImageToVideoPipeline channel-mismatch runtime error and has inbound references, making it more concrete than the other issue titles available here.", + "best_issue_reason": "No issue is globally suitable as the cluster anchor because the soft links are lexical-only and do not indicate one underlying defect shared by the group.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:6242", + "right": "issue:6441", + "accept": false, + "reason": "Unrelated topics: a dataset log request versus a from_pretrained/Colab loading failure." + }, + { + "left": "issue:7054", + "right": "issue:9660", + "accept": false, + "reason": "Different bugs: LoRA state_dict mutation versus modules not being set to eval mode." + }, + { + "left": "issue:7782", + "right": "issue:9670", + "accept": false, + "reason": "Unrelated loading/conversion error versus LoRA backend requirement error." + }, { "left": "issue:8906", "right": "issue:9515", - "accept": true, - "reason": "Both point to the same SD3Transformer2DModel attention/normalization implementation gap in the same code path." + "accept": false, + "reason": "Different SD3 transformer concerns: absence of qk_norm versus RMSNorm in JointAttnProcessor2_0." }, { "left": "issue:11088", "right": "issue:12082", "accept": false, - "reason": "Different components and failures: GGUF load error for Wan2.1 transformer vs VAE memory increase after compile." + "reason": "Different model/load issues: GGUF load error versus VAE memory increase after compile." }, { "left": "issue:1363", "right": "issue:1918", "accept": false, - "reason": "Both mention schedulers, but the pipelines and concrete errors differ; no clear same underlying bug." + "reason": "Both mention schedulers, but one is an ONNX long-prompt error and the other is a DDIM example issue; not the same bug." }, { "left": "issue:4701", "right": "issue:7450", "accept": false, - "reason": "AutoencoderKL CUDA differentiation error is unrelated to multi-thread issues." + "reason": "Both are AutoencoderKL-related, but one is a CUDA differentiation error and the other is a multithreading issue." }, { "left": "issue:6297", "right": "issue:7322", "accept": false, - "reason": "Same training script family, but different failure modes; SIGKILL/OOM is not the same bug as inability to train on photos." + "reason": "Different training failures: SDXL Dreambooth training behavior versus SIGKILL during a specific script/version combo." }, { "left": "issue:1883", "right": "issue:759", "accept": false, - "reason": "Gradient checkpointing for VAE and optional full-model saving for textual inversion are unrelated changes." + "reason": "Gradient checkpointing for VAE is unrelated to optional full-model saving for textual inversion." }, { "left": "issue:4773", @@ -94387,113 +94426,92 @@ "left": "issue:5574", "right": "issue:9516", "accept": false, - "reason": "Different topics: LoRA training attention params vs FLUX joint_attention_kwargs propagation." + "reason": "Different LoRA/attention topics: training partial UNet attention parameters versus FLUX joint_attention_kwargs propagation." }, { "left": "issue:5578", "right": "issue:7946", "accept": false, - "reason": "Missing model_index.json loading error is unrelated to custom pipeline caching." + "reason": "Model_index.json loading failure is unrelated to custom_pipeline caching." }, { "left": "issue:11363", "right": "issue:12900", "accept": false, - "reason": "Both are OOM reports, but for different models/pipelines and likely different memory hotspots." + "reason": "Both are OOM reports, but for different models and scenarios: ControlNet inference versus Qwen-Image at 2048px." }, { "left": "issue:6506", "right": "issue:9648", "accept": false, - "reason": "Output type mismatch in StableVideoDiffusionPipeline is unrelated to VAE tiling memory overflow." + "reason": "Different memory/output issues: StableVideoDiffusion output type bug versus VAE tiling causing overflow." }, { "left": "issue:1403", "right": "issue:2026", "accept": false, - "reason": "Safety checker output differences are unrelated to cpu_offload behavior in PaintByExample." + "reason": "Safety checker output mismatch is unrelated to cpu_offload not working in PaintByExample." }, { "left": "issue:1541", "right": "issue:1730", "accept": false, - "reason": "Both involve Dreambooth/CKPT loading, but one is model-type detection and the other is a converted-model load issue in webui." + "reason": "Related to Dreambooth/CKPT loading, but one is MODEL_TYPE detection and the other is a converted-model compatibility question." }, { "left": "issue:1969", "right": "issue:5425", "accept": false, - "reason": "Tensor view/stride runtime error is unrelated to checkpoint-saving compatibility issues." + "reason": "Tensor view-stride runtime error is unrelated to checkpoint saving compatibility with accelerate/PT 2.1." }, { "left": "issue:7147", "right": "issue:7183", "accept": false, - "reason": "Importing StableCascadeDecoderPipeline and scheduler class-name metadata are different bugs." + "reason": "Importing StableCascadeDecoderPipeline is unrelated to scheduler.config reporting the wrong class name." }, { "left": "issue:11878", "right": "issue:12113", "accept": false, - "reason": "Both are Wan-related, but one is GGUF 1.3B model loading while the other is a 3D input channel mismatch in a pipeline." + "reason": "Both involve Wan models, but one is a GGUF 1.3B load problem and the other is a channel-mismatch runtime error in WanImageToVideoPipeline." }, { "left": "issue:1070", "right": "issue:347", "accept": false, - "reason": "Stable diffusion inpainting bug is unrelated to installation dependency resolution failure." + "reason": "Stable diffusion inpainting bug is unrelated to installing diffusers/huggingface-hub dependency failures." }, { "left": "issue:11547", "right": "issue:13065", "accept": false, - "reason": "Different controlnet feature requests for different models; not the same underlying change." + "reason": "Different ControlNet/Qwen-Image feature requests; one is Flux fill dev support, the other is Qwen-Image inpaint+canny support." }, { "left": "issue:12060", "right": "issue:9836", "accept": false, - "reason": "A question about DiT block existence is not the same as a feature request to record layer_id." + "reason": "A question about DiT block definitions is unrelated to recording layer_id for DiT models." }, { "left": "issue:2761", "right": "issue:547", "accept": false, - "reason": "Separate documentation requests with different scopes." + "reason": "Documentation for UNet1DModel.extra_in_channels is not the same as general stable diffusion local loading docs." }, { "left": "issue:3171", "right": "issue:6412", "accept": false, - "reason": "LPW prompt support for controlnet and ControlNet annotators to ONNX are unrelated." - }, - { - "left": "issue:3728", - "right": "issue:7864", - "accept": false, - "reason": "Different new-pipeline proposals for different tasks; not the same change." - }, - { - "left": "issue:4263", - "right": "issue:8918", - "accept": false, - "reason": "LoRA unloading and SD3 DreamBooth training being broken are different problems." - }, - { - "left": "issue:4363", - "right": "issue:5989", - "accept": false, - "reason": "Loading an SDXL VAE checkpoint and a safetensors.torch bug are not the same issue." + "reason": "LPW prompt support for ControlNet is unrelated to converting ControlNet annotators to ONNX." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "Rejecting because the summary overstates evidence for issue:12113. The packet only shows the title/body length and inbound references; it does not support claims like \"reproducible\" or \"stack-trace-level detail.\" The rest of the verdicts look conservative and grounded, including the SD3Transformer2DModel pair as a plausible duplicate." - }, - "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -94527,21 +94545,20 @@ "issue:1922", "issue:2234", "issue:2439", - "issue:244", - "issue:2459", - "issue:280", "issue:3287", "issue:3574", "issue:3643", + "issue:3728", "issue:3853", - "issue:3885", "issue:3908", "issue:4236", "issue:4250", + "issue:4263", + "issue:4363", "issue:4468", "issue:5064", "issue:5395", - "issue:5567", + "issue:5989", "issue:6297", "issue:6366", "issue:660", @@ -94554,12 +94571,15 @@ "issue:7945", "issue:7955", "issue:8541", - "issue:8594", "issue:8637", + "issue:8918", "issue:9450", "issue:9792" ], "soft_pairs": [ + "issue:3728|issue:7864", + "issue:4263|issue:8918", + "issue:4363|issue:5989", "issue:5395|issue:7945", "issue:7773|issue:9792", "issue:3643|issue:7891", @@ -94580,193 +94600,187 @@ "issue:10782|issue:11055", "issue:11083|issue:8637", "issue:11447|issue:8541", - "issue:1197|issue:3287", - "issue:244|issue:280", - "issue:2459|issue:5567", - "issue:3885|issue:8594" + "issue:1197|issue:3287" ], - "prepared_review_unit_hash": "a9c00d254c2f836af20b0bba78695888d1d8efeb8da602ea59437cab86054318", + "prepared_review_unit_hash": "00e3ba985f5ffa6e960c6ecf232c975893f98bb5f55315623eb1435c75940e11", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 34255, - "estimated_input_tokens": 8564, - "estimated_eval_tokens": 17384 + "serialized_chars": 33538, + "estimated_input_tokens": 8385, + "estimated_eval_tokens": 17026 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous issue reports that share only broad subsystem overlap (generation, training, loading, device placement). The only plausibly duplicate pair is the Flux/optimum.quanto quantization-related reports.", - "confidence": 0.44, - "canonical_issue_reason": "Issue 10381 is the best umbrella representative: it is the broadest Flux/quantization report here and the only one that plausibly covers another item in the set.", + "summary": "This is not a coherent duplicate cluster: the items span many unrelated bugs, docs, and feature requests. All soft edges look like false positives rather than the same underlying issue.", + "confidence": 0.94, + "canonical_issue_reason": "No canonical issue: the list spans unrelated bugs, docs, training, loading, scheduler, and performance reports, so there isn't one issue that safely represents a duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "10381 is the strongest global issue representative because it captures the most general and potentially shared Flux/quanto failure mode among the listed items.", + "best_issue_reason": "No global best issue: none of the items is a strong representative for the full set because the topics diverge too much.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3728", + "right": "issue:7864", + "accept": false, + "reason": "Different requests: a new text-to-image pipeline versus MotionMaster camera-motion transfer." + }, + { + "left": "issue:4263", + "right": "issue:8918", + "accept": false, + "reason": "Different bugs: LoRA unloading versus SD3 training script breakage." + }, + { + "left": "issue:4363", + "right": "issue:5989", + "accept": false, + "reason": "Both involve loading failures, but one is a model-loading question and the other is a safetensors library bug; not the same code-path." + }, { "left": "issue:5395", "right": "issue:7945", "accept": false, - "reason": "Both are image-generation-related, but one is a PEFT/Lora composition bug and the other is a community pipeline combination request; different problems and code paths." + "reason": "PEFT/LoRA combination bug versus a community-pipeline combination request." }, { "left": "issue:7773", "right": "issue:9792", "accept": false, - "reason": "Stable Video Diffusion batch-size generation failure is unrelated to dataset cache mapping." + "reason": "Different failure modes: Stable Video Diffusion batch-size issue versus dataset cache loading." }, { "left": "issue:3643", "right": "issue:7891", "accept": false, - "reason": "One is a latent-decoding INT_MAX/upsample limit, the other is a CPU/CUDA device mismatch; not the same bug." + "reason": "Different runtime problems: latent upsampling tensor-size limit versus cross-device tensor placement." }, { "left": "issue:6297", "right": "issue:6366", "accept": false, - "reason": "Both concern SDXL LoRA training, but one is a general training failure and the other is checkpoint reload logic; not the same concrete defect." + "reason": "Different training issues: can't train on photos versus checkpoint restore/load logic." }, { "left": "issue:1865", "right": "issue:2234", "accept": false, - "reason": "Zero-stage3 dimension error and xformers memory-efficient attention finetuning failure are distinct training issues." + "reason": "Different paths: ZeRO stage-3 finetuning dimension error versus xformers attention finetuning failure." }, { "left": "issue:2439", "right": "issue:660", "accept": false, - "reason": "Inpainting pipeline loading/running is unrelated to MPS fp16 revision behavior." + "reason": "Inpainting pipeline loading versus MPS fp16/revision handling; not the same bug." }, { "left": "issue:10920", "right": "issue:7785", "accept": false, - "reason": "DDIMInverseScheduler inf values and slow first inference in diffusers 0.27.x are different failures." + "reason": "Scheduler inf-values bug versus slow first-inference performance regression." }, { "left": "issue:11872", "right": "issue:9450", "accept": false, - "reason": "CPU offloading inefficiency is a performance regression; the Flux multi-GPU report is a device placement error." + "reason": "Both mention GPU movement, but one is offloading inefficiency and the other is a device mismatch when passing a transformer." }, { "left": "issue:4236", "right": "issue:5064", "accept": false, - "reason": "Text encoder registration in SDXL Refiner is unrelated to out-of-space inference behavior." + "reason": "Refiner text-encoder registration issue versus out-of-space during inference." }, { "left": "issue:7864", "right": "issue:7955", "accept": false, - "reason": "These are different model/pipeline proposals with no shared bug or change." + "reason": "Different community models: camera motion transfer versus identity-preserving portrait generation." }, { "left": "issue:1922", "right": "issue:3574", "accept": false, - "reason": "Undefined notebook helper and Windows from_ckpt loading issue are unrelated." + "reason": "Notebook undefined-function bug versus Windows checkpoint conversion/loading issue." }, { "left": "issue:3853", "right": "issue:7591", "accept": false, - "reason": "Both mention HeaderTooLarge, but one is LoRA weight loading and the other is single-file loading with local_files_only; insufficiently same concrete bug." + "reason": "Similar error text, but the loading paths differ enough that this is not clearly the same underlying bug." }, { "left": "issue:3908", "right": "issue:4250", "accept": false, - "reason": "Streamlit caching and from_pretrained loading are different concerns." + "reason": "Streamlit caching behavior versus incorrect from_pretrained model loading." }, { "left": "issue:10381", "right": "issue:10526", - "accept": true, - "reason": "Both are Flux + optimum.quanto / quantization inference failures, and the titles point to the same underlying FP8/quantized loading path." + "accept": false, + "reason": "Related Flux/Quanto area, but one is a broad inference report with LoRA/Zero GPU spaces while the other is a specific WeightQBytesTensor TypeError; not clearly one bug." }, { "left": "issue:12433", "right": "issue:13256", "accept": false, - "reason": "Documentation/version-compatibility guidance and a generic package version question are not the same issue." + "reason": "Docs/version-compatibility complaint versus a generic package compatibility claim; not the same issue." }, { "left": "issue:1284", "right": "issue:4468", "accept": false, - "reason": "Longer sample time with larger batch size is not the same as multi-GPU performance differences." + "reason": "Batch-size slowdown versus multi-GPU performance difference." }, { "left": "issue:10928", "right": "issue:7492", "accept": false, - "reason": "Missing local shard file and repository-not-found are different model-loading failures." + "reason": "Missing local shard file versus repository-not-found on the hub." }, { "left": "issue:10782", "right": "issue:11055", "accept": false, - "reason": "Lumina minor issue and unconditional training colorization are unrelated." + "reason": "Lumina minor issue versus unconditional training producing colorized images." }, { "left": "issue:11083", "right": "issue:8637", "accept": false, - "reason": "DDP stride-mismatch warning and missing PixArtSigmaPipeline import are distinct issues." + "reason": "DDP stride mismatch warning versus missing PixArtSigmaPipeline import." }, { "left": "issue:11447", "right": "issue:8541", "accept": false, - "reason": "Offline shard resolution and SD3 repository lookup failure are different loading problems." + "reason": "Offline shard-file lookup failure versus model repository not found." }, { "left": "issue:1197", "right": "issue:3287", "accept": false, - "reason": "Community pipeline request and improved DDPM support request are separate feature proposals." - }, - { - "left": "issue:244", - "right": "issue:280", - "accept": false, - "reason": "Formatter contributor docs and DEIS scheduler addition are unrelated." - }, - { - "left": "issue:2459", - "right": "issue:5567", - "accept": false, - "reason": "xformers enablement in a training script and accelerator.prepare overwriting unet.config are different bugs." - }, - { - "left": "issue:3885", - "right": "issue:8594", - "accept": false, - "reason": "VRAM usage for a training script and an UnboundLocalError in SD3 training are different problems." + "reason": "OpenAI Diffusion Pipeline proposal versus OpenAI improved DDPM support request." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative overall. The only accepted soft pair, issue 10381 vs 10526, is plausibly the same Flux/optimum.quanto FP8/quantized-loading failure family. The other pair rejections are generally justified by the titles and do not overclaim mergeability or duplication." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -94790,15 +94804,15 @@ "issue:10860", "issue:10963", "issue:11062", - "issue:11321", - "issue:11343", "issue:12610", "issue:13001", + "issue:244", + "issue:2459", + "issue:280", "issue:2841", - "issue:2941", "issue:2966", - "issue:3431", "issue:3781", + "issue:3885", "issue:4057", "issue:4418", "issue:4453", @@ -94809,6 +94823,7 @@ "issue:5160", "issue:5225", "issue:5442", + "issue:5567", "issue:5710", "issue:5763", "issue:5846", @@ -94818,8 +94833,7 @@ "issue:6949", "issue:7593", "issue:7752", - "issue:7836", - "issue:8637", + "issue:8594", "issue:8759", "issue:8786", "issue:9025", @@ -94832,6 +94846,9 @@ "issue:9953" ], "soft_pairs": [ + "issue:244|issue:280", + "issue:2459|issue:5567", + "issue:3885|issue:8594", "issue:4057|issue:7752", "issue:4453|issue:4680", "issue:4788|issue:5225", @@ -94852,183 +94869,180 @@ "issue:2966|issue:6444", "issue:1066|issue:3781", "issue:10710|issue:9138", - "issue:10860|issue:10963", - "issue:11321|issue:7836", - "issue:11343|issue:8637", - "issue:2941|issue:3431" + "issue:10860|issue:10963" ], - "prepared_review_unit_hash": "b503f0b16f3ec773c369565fa88034930ad5a6445a6b6196ea1f7c05f8fa9ef1", + "prepared_review_unit_hash": "cfaee5345e68204eaa6e92b65d46a4433a018aa16c32314342c145e785b63db4", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33476, - "estimated_input_tokens": 8369, - "estimated_eval_tokens": 16994 + "serialized_chars": 33523, + "estimated_input_tokens": 8381, + "estimated_eval_tokens": 17018 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is highly heterogeneous: the items span unrelated bugs, feature requests, docs, training scripts, and loading/runtime failures. None of the proposed soft pairs look like true duplicates or one mergeable change.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "This cluster is heterogeneous: many items are in adjacent Diffusers areas (LoRA, schedulers, pipelines, quantization), but the paired issues do not cleanly describe the same underlying bug or change. I would not merge any of the suggested soft pairs.", + "confidence": 0.82, + "canonical_issue_reason": "Issue 10227 is the clearest self-contained bug report in the LoRA subgroup, with a concrete exception and an actionable technical description. That said, the overall cluster is too mixed for a true duplicate canonical.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 10227 is the strongest single representative issue because it is precise, technical, and easy to reproduce/search against, but it does not subsume the other cluster members.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:244", + "right": "issue:280", + "accept": false, + "reason": "Formatter-docs guidance and a DEIS scheduler feature request are unrelated." + }, + { + "left": "issue:2459", + "right": "issue:5567", + "accept": false, + "reason": "xformers attention in a LoRA training script and an accelerator/config overwrite bug are different code-paths." + }, + { + "left": "issue:3885", + "right": "issue:8594", + "accept": false, + "reason": "VRAM reduction advice for training is not the same bug as an UnboundLocalError in an SD3 training script." + }, { "left": "issue:4057", "right": "issue:7752", "accept": false, - "reason": "Both are about training/inpainting, but one reports blurred finetuned inpainting output while the other asks about class-image counts; different problems." + "reason": "Finetuned inpainting blur and a training hyperparameter question are unrelated." }, { "left": "issue:4453", "right": "issue:4680", "accept": false, - "reason": "Different topics: LoRA export/conversion to A1111 format versus SDXL inpainting finetuning behavior." + "reason": "LoRA conversion to safetensors and SDXL inpainting fine-tuning are different problems." }, { "left": "issue:4788", "right": "issue:5225", "accept": false, - "reason": "Both mention SDXL memory, but one is about inpaint OOM on 6GB VRAM and the other about forward-pass OOM on 24GB; not the same bug." + "reason": "Both involve memory limits, but they are separate OOM reports in different contexts and not the same bug." }, { "left": "issue:5150", "right": "issue:9025", "accept": false, - "reason": "One is a missing from_single_file implementation for an upscale pipeline; the other is a usage question about FlaxStableDiffusionPipeline on TPU." + "reason": "Both mention from_single_file, but they target different pipeline classes and do not show the same missing implementation bug." }, { "left": "issue:5160", "right": "issue:9241", "accept": false, - "reason": "Dreambooth/LoRA masking is a training-feature request, while the other is a script to convert XLab LoRAs; different scope and code path." + "reason": "Dreambooth/LoRA masking and an XLab-to-Diffusers conversion script are unrelated." }, { "left": "issue:5442", "right": "issue:6313", "accept": false, - "reason": "ScaleCrafter pipeline support and Null-Text Inversion are unrelated pipeline features." + "reason": "ScaleCrafter and Null-Text Inversion are distinct pipeline features with no shared failure mode." }, { "left": "issue:6533", "right": "issue:9387", "accept": false, - "reason": "Marigold depth estimation and MimicMotion are different pipelines with no shared underlying bug." + "reason": "Marigold depth estimation and MimicMotion are separate pipelines." }, { "left": "issue:7593", "right": "issue:9267", "accept": false, - "reason": "Both involve Flux-related internals, but one is an IP-Adapter image-embed NoneType error and the other is a missing rope attribute; distinct failures." + "reason": "IP-Adapter image embed preparation error and missing rope attribute in Flux are different failures." }, { "left": "issue:8759", "right": "issue:8786", "accept": false, - "reason": "Both are SD3/tokenization-related warnings, but one is an image cast warning and the other is a max_sequence_length token-count issue." + "reason": "A runtime warning in SD3 image postprocessing is not the same as max_sequence_length having no effect on token count." }, { "left": "issue:4418", "right": "issue:5763", "accept": false, - "reason": "Both mention save/load behavior, but one is SDXL inpaint denoising_start and the other is safe_serialization/save_pretrained output format." + "reason": "SDXL inpaint denoising_start behavior and safetensors serialization are unrelated." }, { "left": "issue:5846", "right": "issue:9523", "accept": false, - "reason": "Resume-from-checkpoint in an SDXL LoRA example is unrelated to mapping errors in SD3/SDXL/Flux controlnet training scripts." + "reason": "Checkpoint resume error in one training script and mapping errors across several training scripts are not the same concrete bug." }, { "left": "issue:10085", "right": "issue:9051", "accept": false, - "reason": "Both mention Flux batch inference, but one is a LyCORIS/skip_guidance_layers runtime error and the other is a broader unsupported-batch-inference limitation; not the same concrete bug." + "reason": "Batch inference/skip_guidance_layers with LyCORIS is more specific than a general FluxPipeline batch-inference limitation; not the same underlying issue." }, { "left": "issue:11062", "right": "issue:13001", "accept": false, - "reason": "Different loading failures: a missing AutoencoderKLWan import versus a shape mismatch when loading a Flux2.Klein GGUF." + "reason": "A CivitAI LoRA loading error and a Flux2.Klein GGUF shape mismatch are distinct model-loading failures." }, { "left": "issue:5710", "right": "issue:6949", "accept": false, - "reason": "Both are scheduler bugs, but one lacks self.sigmas at init and the other produces blank images on even step numbers; separate defects." + "reason": "Both touch DPMSolverSinglestepScheduler, but missing sigmas initialization and even-step blank images are different symptoms without a clear same-root-cause match." }, { "left": "issue:10227", "right": "issue:9138", "accept": false, - "reason": "A LoRA feature-support limitation is not the same as a generic TypeError comparing int and str." + "reason": "A LoRA feature-size NotImplementedError and an int/str TypeError are unrelated." }, { "left": "issue:12610", "right": "issue:9953", "accept": false, - "reason": "Both involve quantized/offloaded pipelines, but one is a Quanto+group-offload device mismatch and the other is moving a quantized pipeline to CUDA; different code paths." + "reason": "Both involve quantized pipelines/device movement, but one is a Quanto+group-offload matmul device mismatch and the other is a broader cuda-move error; too imprecise to merge as the same fix." }, { "left": "issue:2841", "right": "issue:459", "accept": false, - "reason": "Sampling temperature in schedulers is unrelated to the broader incremental diffusion request." + "reason": "Sampling temperature support and incremental diffusion are different feature requests." }, { "left": "issue:2966", "right": "issue:6444", "accept": false, - "reason": "Multi-GPU training script problems and checkpointing-related CLIPTextModel peft_config errors are not the same issue." + "reason": "Multi-GPU training issues in different scripts with different stack traces are not the same bug." }, { "left": "issue:1066", "right": "issue:3781", "accept": false, - "reason": "Fork doc-builder behavior and inability to load multiple LoRAs are unrelated." + "reason": "Docs generation in forks and loading multiple LoRAs are unrelated." }, { "left": "issue:10710", "right": "issue:9138", "accept": false, - "reason": "A DDUF format support question is unrelated to a TypeError comparing int and str." + "reason": "A question about DDUF support and a TypeError report are unrelated." }, { "left": "issue:10860", "right": "issue:10963", "accept": false, - "reason": "LoRA compatibility in ComfyUI and an AutoencoderKLWan import error are distinct model-loading problems." - }, - { - "left": "issue:11321", - "right": "issue:7836", - "accept": false, - "reason": "A Flux ControlNet README bug and a CLIP training overfitting report do not share an underlying code bug." - }, - { - "left": "issue:11343", - "right": "issue:8637", - "accept": false, - "reason": "Transformers import breakage and a missing PixArtSigmaPipeline export are both import issues, but not the same missing-module bug." - }, - { - "left": "issue:2941", - "right": "issue:3431", - "accept": false, - "reason": "EulerAncestralDiscreteScheduler output quality and inpainting+ControlNet poor results are different pipeline-quality problems." + "reason": "A LoRA compatibility/loading problem and a missing AutoencoderKLWan import are different issues." } ] }, @@ -95053,6 +95067,8 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10638", + "issue:11321", + "issue:11343", "issue:1156", "issue:12666", "issue:12881", @@ -95061,13 +95077,14 @@ "issue:1798", "issue:2233", "issue:2645", + "issue:2941", "issue:2961", "issue:3103", "issue:3249", "issue:3296", + "issue:3431", "issue:3467", "issue:3871", - "issue:3967", "issue:4340", "issue:4598", "issue:4615", @@ -95082,26 +95099,26 @@ "issue:6481", "issue:6632", "issue:6726", - "issue:6958", "issue:7057", - "issue:7300", - "issue:7350", "issue:7400", "issue:7438", - "issue:7456", + "issue:7836", "issue:7950", "issue:7991", "issue:8507", + "issue:8637", "issue:8708", "issue:8788", "issue:9090", "issue:9370", - "issue:9467", "issue:9793", "issue:9844", "issue:9889" ], "soft_pairs": [ + "issue:11321|issue:7836", + "issue:11343|issue:8637", + "issue:2941|issue:3431", "issue:4598|issue:5002", "issue:4615|issue:7950", "issue:5166|issue:9370", @@ -95122,183 +95139,180 @@ "issue:10638|issue:9844", "issue:12666|issue:12881", "issue:1798|issue:5429", - "issue:2233|issue:2961", - "issue:3967|issue:6958", - "issue:7300|issue:7350", - "issue:7456|issue:9467" + "issue:2233|issue:2961" ], - "prepared_review_unit_hash": "7af77ff7c54f7ebf6b7e97ec481296069695becd3aea6a390111d0ff62b3cad4", + "prepared_review_unit_hash": "42a9660ae60509780225b7aeaeb4e4783b50608f9a5a512b6ad75c031a7a7e1e", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33656, - "estimated_input_tokens": 8414, - "estimated_eval_tokens": 17084 + "serialized_chars": 33720, + "estimated_input_tokens": 8430, + "estimated_eval_tokens": 17116 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a loose set of mostly unrelated Diffusers bug reports spanning training, pipelines, schedulers, and platform issues. None of the soft edges look like true duplicates, so they should all be kept separate.", - "confidence": 0.86, - "canonical_issue_reason": "issue:8507 is the best anchor: it is open, recent, and has a concrete, actionable SD3 training failure with a clear code path.", + "summary": "This cluster is mostly a loose grouping of unrelated issue reports that only share broad diffusers/training/inference vocabulary. None of the soft pairs look like the same underlying bug.", + "confidence": 0.39, + "canonical_issue_reason": "issue:7991 is the best representative of the cluster\u2019s broad training/tutorial-bug theme and is specific enough to serve as a canonical anchor.", "canonical_pr_reason": null, - "best_issue_reason": "issue:8507 is the strongest representative available despite the cluster being heterogeneous; it is specific enough to triage and still active.", + "best_issue_reason": "issue:7991 is the most generally applicable and actionable issue in the set; the others are either narrower, more version-specific, or about different pipelines altogether.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:11321", + "right": "issue:7836", + "accept": false, + "reason": "One is a FLUX ControlNet README problem; the other is a CLIP training overfitting example bug. Different topic and failure mode." + }, + { + "left": "issue:11343", + "right": "issue:8637", + "accept": false, + "reason": "Both are import errors, but they involve different packages/pipelines and likely different dependency breakages." + }, + { + "left": "issue:2941", + "right": "issue:3431", + "accept": false, + "reason": "EulerAncestral scheduler output quality is unrelated to Inpainting+ControlNet poor results; different code paths." + }, { "left": "issue:4598", "right": "issue:5002", "accept": false, - "reason": "Missing dependency error handling is unrelated to a scheduler argument mismatch." + "reason": "Missing dependency error messaging is not the same as schedulers missing a Karras argument." }, { "left": "issue:4615", "right": "issue:7950", "accept": false, - "reason": "ImportError for a pipeline class and a PyTorch compatibility regression are different failure modes." + "reason": "ControlNet import error versus PyTorch 2.3 incompatibility; one is a symbol/export issue, the other a version compatibility issue." }, { "left": "issue:5166", "right": "issue:9370", "accept": false, - "reason": "VQVAE scaling-factor omission is not the same bug as a missing pooled_projection argument." + "reason": "VQ-VAE scaling factor in unconditional LDM is unrelated to missing pooled_projection in timestep guidance embeddings." }, { "left": "issue:6481", "right": "issue:7057", "accept": false, - "reason": "Batch-size-specific Dreambooth failure and a generic checkpoint issue are different code paths." + "reason": "Dreambooth batch-size bug and a failing checkpoint load are different training failures." }, { "left": "issue:6632", "right": "issue:8708", "accept": false, - "reason": "Save-state questions for LoRA training are unrelated to missing SD3 VAE shift_factor support." + "reason": "save_state question for LoRA scripts does not match missing vae.config.shift_factor in SD3 dreambooth examples." }, { "left": "issue:7400", "right": "issue:9889", "accept": false, - "reason": "Black-image output in stable video diffusion is not the same as an SDXL-turbo segfault." + "reason": "Black images in stable video diffusion and an SDXL-turbo segmentation fault are distinct inference/runtime problems." }, { "left": "issue:7438", "right": "issue:9793", "accept": false, - "reason": "XPU generation stalling and a progress bar not updating are distinct symptoms and likely different bugs." + "reason": "XPU generation stall is not the same as a progress bar refresh bug." }, { "left": "issue:5475", "right": "issue:6726", "accept": false, - "reason": "These are different model/paper requests, not the same underlying bug or change." + "reason": "These are different model papers/features: LaVie video generation versus CCSR super-resolution." }, { "left": "issue:3467", "right": "issue:3871", "accept": false, - "reason": "Both mention resume behavior, but one is ControlNet step accounting and the other is a generic text-to-image checkpoint error." + "reason": "ControlNet resume-step accounting with gradient accumulation is different from a generic checkpoint resume error in text-to-image training." }, { "left": "issue:7991", "right": "issue:9090", "accept": false, - "reason": "A tutorial bug and a PNDM timestep duplication issue are unrelated." + "reason": "Training tutorial bug and PNDMScheduler repeated timesteps are unrelated subsystems." }, { "left": "issue:1323", "right": "issue:579", "accept": false, - "reason": "Missing dtype on Flax model loading is unrelated to a missing config.json in textual inversion." + "reason": "Flax model dtype loading and missing config.json during textual inversion are different loader failures." }, { "left": "issue:1572", "right": "issue:3103", "accept": false, - "reason": "Inpainting support failures and height/width handling are different issues in the same broad area." + "reason": "stable_diffusion_mega img2img/inpainting support and inpainting height/width handling are not the same bug." }, { "left": "issue:4340", "right": "issue:8507", "accept": false, - "reason": "Both involve train_text_encoder, but they affect different training scripts/models and are not the same concrete bug." + "reason": "SDXL LoRA train_text_encoder error and SD3 dreambooth train_text_encoder failure are different model-specific training issues." }, { "left": "issue:5693", "right": "issue:8788", "accept": false, - "reason": "VAE load errors and SD3 VAE tiling support are different problems." + "reason": "VAE load error versus SD3 VAE tiling support are separate VAE problems." }, { "left": "issue:1156", "right": "issue:3296", "accept": false, - "reason": "A broad training failure and checkpoint save/load with text_encoder are not the same bug." + "reason": "A generic 'training doesn't work' report and checkpoint save/load failure with text_encoder are not clearly the same defect." }, { "left": "issue:3249", "right": "issue:487", "accept": false, - "reason": "An AttnUpBlock2D argument mismatch is unrelated to M1 CPU pipeline failure." + "reason": "AttnUpBlock2D upsample_size API mismatch is unrelated to StableDiffusionPipeline not working on M1 CPU." }, { "left": "issue:2645", "right": "issue:5365", "accept": false, - "reason": "Mac/CUDA ControlNet failure and txt2video num_inference_steps handling are different issues." + "reason": "ControlNet on M1 with CUDA assertion and txt2video num_inference_steps error are different platform and scheduler issues." }, { "left": "issue:10638", "right": "issue:9844", "accept": false, - "reason": "Scheduler scheduling_dpmsolver_multistep.py running issue and SDXL VAE NaNs do not share the same code path." + "reason": "SDXL schedule script problem is not the same as SDXL VAE encoder producing NaNs." }, { "left": "issue:12666", "right": "issue:12881", "accept": false, - "reason": "Missing Kandinsky 5.0 video pipeline and prompt2prompt prompt replacement are unrelated." + "reason": "Kandinsky 5.0 Video Pipeline missing from a release is unrelated to a prompt2prompt word-replacement bug." }, { "left": "issue:1798", "right": "issue:5429", "accept": false, - "reason": "v_prediction support for dreambooth is not the same as supporting models without VAE in other examples." + "reason": "Flax dreambooth v_prediction support and a generic request to support models without VAE are different feature gaps." }, { "left": "issue:2233", "right": "issue:2961", "accept": false, - "reason": "Textual inversion validation/accumulation issues are unrelated to a StableUnCLIP training example request." - }, - { - "left": "issue:3967", - "right": "issue:6958", - "accept": false, - "reason": "Deleted-branch SDXL breakage and unet_motion_model projection wiring are unrelated." - }, - { - "left": "issue:7300", - "right": "issue:7350", - "accept": false, - "reason": "GPU memory growth after attn processor swap is not the same as ControlNet initialization from LoRA UNet." - }, - { - "left": "issue:7456", - "right": "issue:9467", - "accept": false, - "reason": "A missing raise statement and a missing num_frames guard rail are different fixes and different failure contexts." + "reason": "Textual inversion validation/gradient accumulation bugs do not match the StableUnCLIP training example request." } ] }, @@ -95323,20 +95337,18 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10045", - "issue:10287", "issue:10829", "issue:11045", "issue:11232", "issue:11486", "issue:1572", - "issue:1717", "issue:181", - "issue:2177", "issue:2440", "issue:2444", "issue:2673", "issue:2737", "issue:3698", + "issue:3967", "issue:4264", "issue:4374", "issue:4400", @@ -95353,25 +95365,30 @@ "issue:5647", "issue:6109", "issue:6801", + "issue:6958", "issue:7043", "issue:7080", "issue:7185", "issue:725", + "issue:7300", + "issue:7350", + "issue:7456", "issue:759", "issue:7686", "issue:8511", "issue:8543", - "issue:8593", - "issue:8786", - "issue:9023", "issue:9161", "issue:9225", "issue:9237", "issue:9366", + "issue:9467", "issue:9534", "issue:9568" ], "soft_pairs": [ + "issue:3967|issue:6958", + "issue:7300|issue:7350", + "issue:7456|issue:9467", "issue:11232|issue:8511", "issue:4580|issue:5109", "issue:2673|issue:4686", @@ -95392,25 +95409,22 @@ "issue:5004|issue:8543", "issue:5079|issue:725", "issue:6109|issue:7080", - "issue:7185|issue:9568", - "issue:8593|issue:8786", - "issue:10287|issue:9023", - "issue:1717|issue:2177" + "issue:7185|issue:9568" ], - "prepared_review_unit_hash": "bdd03514b448389c591c7348037f84708a6eaf09601e410ab14c390429bbd6f8", + "prepared_review_unit_hash": "7b345a93006340096cf504667af476def1b89cf4a1dc37d0dcd1f065dcee9f80", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33556, + "serialized_chars": 33553, "estimated_input_tokens": 8389, "estimated_eval_tokens": 17034 }, @@ -95419,156 +95433,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is only loosely related: most items are diffusion pipeline/training/loading bugs, but the soft pairs are not the same underlying defects. They span distinct code paths such as schedulers, ControlNet, DreamBooth/LoRA scripts, model loading, and docs/examples.", - "confidence": 0.86, - "canonical_issue_reason": "No exact duplicate issue clearly dominates. If one representative is needed, issue 9237 is the best anchor because it is the most detailed and discussion-heavy training-related report in the set, but the overall cluster is still heterogeneous.", - "canonical_pr_reason": null, - "best_issue_reason": "Issue 9237 is the strongest general representative: it has rich repro/discussion and sits near several other training-script reports, making it more central than the narrower loading/runtime or docs issues.", - "best_pr_reason": null, + "summary": "These items are a loose set of unrelated Diffusers bugs/feature requests. The soft pairs mostly share broad subsystem words (pipeline, ControlNet, DreamBooth, SDXL) but not the same concrete failure or fix path, so I would not merge any of them as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "No clear canonical issue: the collection is heterogeneous and the proposed soft pairs do not describe the same underlying bug.", + "canonical_pr_reason": "No pull requests are present in the cluster.", + "best_issue_reason": "No single issue stands out as the best representative because the items are not a duplicate set and cover distinct problems.", + "best_pr_reason": "No pull requests are present in the cluster.", "soft_edge_verdicts": [ + { + "left": "issue:3967", + "right": "issue:6958", + "accept": false, + "reason": "Both involve SDXL/unet-related failures, but one is a branch/deployment complaint and the other is a specific mid_block config mismatch in unet_motion_model. Different root causes." + }, + { + "left": "issue:7300", + "right": "issue:7350", + "accept": false, + "reason": "One reports high GPU memory after set_default_attn_processor(); the other is about initializing ControlNet from a LoRA-tuned UNet. Same area, different bugs." + }, + { + "left": "issue:7456", + "right": "issue:9467", + "accept": false, + "reason": "Both mention missing guard-rail/raise-style errors, but they are different products and code paths: a generic missing raise statement vs CogVideoX num_frames validation." + }, { "left": "issue:11232", "right": "issue:8511", "accept": false, - "reason": "Both are training-related, but one is about text-to-image fine-tuning noise residuals and the other about SD3 text encoder unloading; different bugs and fixes." + "reason": "Training-related only at a high level; one is about noise residuals during text-to-image fine-tuning, the other about SD3 scripts not unloading text encoders. Not the same defect." }, { "left": "issue:4580", "right": "issue:5109", "accept": false, - "reason": "UniPC scheduler karras-sigma exception and SDXL ControlNet img2img load_lora_weights exception are separate code paths." + "reason": "unipc/Karras scheduler exception and SDXL ControlNet img2img load_lora_weights exception are unrelated failures." }, { "left": "issue:2673", "right": "issue:4686", "accept": false, - "reason": "Both mention StableDiffusionUpscalePipeline, but one is a pipeline usage/config failure and the other is loading a locally saved pipeline from_pretrained." + "reason": "Both touch StableDiffusionUpscalePipeline, but one is a runtime config/image verification failure and the other is loading a saved pipeline with from_pretrained. Different issue shape." }, { "left": "issue:5567", "right": "issue:9534", "accept": false, - "reason": "Different failures: accelerator.prepare clobbering unet.config versus FluxMultiControlNetModel missing config attribute." + "reason": "Both mention config attributes, but one is accelerator.prepare overwriting unet.config and the other is FluxMultiControlNetModel lacking config. Not the same code-path bug." }, { "left": "issue:11045", "right": "issue:9161", "accept": false, - "reason": "Flux Schnell model-loading crash in one script versus distributed-training bugs in another; not the same defect." + "reason": "Both are Flux/DreamBooth LoRA training reports, but one is a model load crash and the other is distributed-training bugs. Not one concrete bug." }, { "left": "issue:4806", "right": "issue:7686", "accept": false, - "reason": "One is an incorrect parameter type in UNet2DModel, the other is a missing cross_attention_dim validation for CrossAttnDownBlock2D." + "reason": "Both concern UNet2DModel initialization, but one is an incorrect parameter type and the other is a missing cross_attention_dim requirement. Different validation errors." }, { "left": "issue:11486", "right": "issue:9237", "accept": false, - "reason": "This is a feature request to enable xformers memory-efficient attention, not the same issue as the Flux DreamBooth/LoRA training bug." + "reason": "One requests enabling xformers in the training script; the other is a broad Flux DreamBooth LoRA training issue. No shared concrete failure." }, { "left": "issue:5647", "right": "issue:9366", "accept": false, - "reason": "SDXL inpaint strength==1 behavior and DPMSolver img2img step/strength edge cases are different bugs." + "reason": "Both involve strength-related img2img behavior, but 5647 is a specific inpaint strength==1 bug while 9366 is a scheduler/step-count edge case. Not mergeable as one fix." }, { "left": "issue:10045", "right": "issue:6801", "accept": false, - "reason": "Pipeline import problems and scheduler loading problems are distinct components and failure modes." + "reason": "Importing pipelines and loading schedulers are different initialization problems, even though both happen during model loading." }, { "left": "issue:10829", "right": "issue:7043", "accept": false, - "reason": "DreamBooth LoRA SDXL OOM during inference is unrelated to custom captions in DreamBooth training." + "reason": "OOM during final inference in SDXL DreamBooth LoRA is unrelated to custom captions support in Dreambooth." }, { "left": "issue:1572", "right": "issue:2440", "accept": false, - "reason": "A pipeline not working for img2img/inpainting is not the same as failing to load pretrained AutoencoderKL weights." + "reason": "Both mention inpainting, but one is a mega pipeline compatibility issue and the other is float16 model loading. Not the same bug." }, { "left": "issue:181", "right": "issue:2444", "accept": false, - "reason": "Unconditional training image-write failure and DreamBooth inpaint script exit-status failure are different issues." + "reason": "Train_unconditional image writing failure and dreambooth_inpaint subprocess exit error are unrelated training-script issues." }, { "left": "issue:2737", "right": "issue:4400", "accept": false, - "reason": "A scheduler feature request for CLIP guidance is unrelated to text_encoder support in a training script." + "reason": "CLIP guidance scheduler support and text_encoder support in an inpaint LoRA script are different feature requests." }, { "left": "issue:3698", "right": "issue:759", "accept": false, - "reason": "Model conversion to diffusers and optional full-model saving for textual inversion are unrelated changes." + "reason": "Converting an inpainting model to diffusers and making full-model saving optional for textual inversion are unrelated tasks." }, { "left": "issue:4264", "right": "issue:4453", "accept": false, - "reason": "Missing SDXL example images is documentation/assets, while the other is LoRA conversion compatibility." + "reason": "Missing example images for SDXL ensemble of experts and SDXL LoRA conversion to safetensors are unrelated documentation/tooling issues." }, { "left": "issue:4374", "right": "issue:4841", "accept": false, - "reason": "ControlNet pipeline issue and ControlNet image-size issue are too different to be the same bug." + "reason": "Both involve ControlNet and image size, but one is an inpaint pipeline issue and the other is a generic ControlNet image-size problem. Too different to treat as duplicates." }, { "left": "issue:4830", "right": "issue:9225", "accept": false, - "reason": "MultiControlNet tensor-size mismatch versus RGBA latent-size mismatch in InstructPix2Pix are separate shape bugs." + "reason": "Wrong tensor size in MultiControlNet and latent-size mismatch with rgba images are both shape issues, but they affect different pipelines and failure mechanisms." }, { "left": "issue:5004", "right": "issue:8543", "accept": false, - "reason": "Checkpoint resume at fp16 and SD3 gradient-checkpointing failure are different training problems." + "reason": "Checkpoint resume at fp16 and gradient checkpointing failure are both training reliability issues, but they are separate bugs." }, { "left": "issue:5079", "right": "issue:725", "accept": false, - "reason": "LyCORIS loading support and device_map/accelerate memory-footprint support are unrelated feature requests." + "reason": "LoRA loading support and device-map/accelerate memory reduction are different model-loading enhancements." }, { "left": "issue:6109", "right": "issue:7080", "accept": false, - "reason": "ControlNet-xs depth support and SDXL control-net training inference logging drift are not the same underlying issue." + "reason": "ControlNet-xs depth support and SDXL ControlNet training inference output drift from wandb are unrelated ControlNet problems." }, { "left": "issue:7185", "right": "issue:9568", "accept": false, - "reason": "PyTorch version support question and CogVideoX autoencoder reconstruction failure are unrelated." - }, - { - "left": "issue:8593", - "right": "issue:8786", - "accept": false, - "reason": "BF16 noise on CPU and max_sequence_length token-count behavior in SD3 are distinct bugs." - }, - { - "left": "issue:10287", - "right": "issue:9023", - "accept": false, - "reason": "Documentation example issue and sample code runtime failure are not the same defect." - }, - { - "left": "issue:1717", - "right": "issue:2177", - "accept": false, - "reason": "Offline cache loading failure and AutoencoderKL pretrained loading failure affect different loading paths and components." + "reason": "PyTorch 2.2 compatibility question and CogVideoX autoencoder reconstruction failure have no meaningful overlap beyond being user reports." } ] }, @@ -95593,23 +95607,22 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10165", + "issue:10287", "issue:10950", - "issue:1101", "issue:11321", - "issue:1141", "issue:11547", "issue:12308", "issue:1246", "issue:1335", "issue:1358", "issue:1530", + "issue:1717", "issue:2024", "issue:2163", + "issue:2177", "issue:2265", "issue:2359", - "issue:2616", "issue:2803", - "issue:2856", "issue:3403", "issue:3612", "issue:3627", @@ -95617,13 +95630,11 @@ "issue:3933", "issue:4236", "issue:4256", - "issue:4562", "issue:4615", "issue:4677", "issue:4975", "issue:5009", "issue:5245", - "issue:525", "issue:5327", "issue:5466", "issue:6238", @@ -95635,13 +95646,19 @@ "issue:7147", "issue:7267", "issue:7968", + "issue:8593", "issue:8673", + "issue:8786", + "issue:9023", "issue:9239", "issue:9599", "issue:9613", "issue:9619" ], "soft_pairs": [ + "issue:8593|issue:8786", + "issue:10287|issue:9023", + "issue:1717|issue:2177", "issue:10950|issue:7267", "issue:2803|issue:4615", "issue:1246|issue:3627", @@ -95662,183 +95679,180 @@ "issue:5245|issue:7147", "issue:2163|issue:2265", "issue:4256|issue:5009", - "issue:10165|issue:9613", - "issue:1101|issue:1141", - "issue:2616|issue:525", - "issue:2856|issue:4562" + "issue:10165|issue:9613" ], - "prepared_review_unit_hash": "2519f148bff5ede399b823f745792b5399c9b84d38567b9551050fbc978f7b89", + "prepared_review_unit_hash": "840801755755459cce1d76b2084bdca31a95f2b4b09af0feb15ec958b1950286", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33562, - "estimated_input_tokens": 8391, - "estimated_eval_tokens": 17038 + "serialized_chars": 33521, + "estimated_input_tokens": 8381, + "estimated_eval_tokens": 17018 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a grab-bag of unrelated issues and feature requests; the soft pairs do not appear to describe the same underlying bug or change.", - "confidence": 0.97, - "canonical_issue_reason": "No canonical issue: the items span many distinct bugs, docs problems, and feature requests rather than one duplicate family.", + "summary": "This cluster is highly heterogeneous: it mixes unrelated docs, training, loading, and runtime issues across different pipelines/models. There is no convincing duplicate group; if one issue must be picked as the broadest representative, #10287 is the most generic docs/example report.", + "confidence": 0.17, + "canonical_issue_reason": "#10287 is the broadest and most generic issue here, framed around an example-code/documentation problem rather than a narrow model- or pipeline-specific bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:10165 is a clear, self-contained bug report and the cleanest single representative issue here, but it is not a canonical duplicate target for the cluster.", + "best_issue_reason": "#10287 is the best overall issue representative because its scope is broad and documentation-focused, but it still should not be treated as a duplicate anchor for the rest of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:8593", + "right": "issue:8786", + "accept": false, + "reason": "Different SD3 bugs: one is bfloat16 noise on CPU, the other is a max_sequence_length token-count issue." + }, + { + "left": "issue:10287", + "right": "issue:9023", + "accept": false, + "reason": "Both mention example/sample code, but they are distinct documentation/runtime failures with no shared code path shown." + }, + { + "left": "issue:1717", + "right": "issue:2177", + "accept": false, + "reason": "Offline cache loading vs pretrained AutoencoderKL loading are different model-loading bugs." + }, { "left": "issue:10950", "right": "issue:7267", "accept": false, - "reason": "Both are about broken docs/training examples, but they concern different examples and failure modes." + "reason": "Doc section brokenness and local training-data loading failure are unrelated issues." }, { "left": "issue:2803", "right": "issue:4615", "accept": false, - "reason": "WSL2 docs build problems and an import error for a controlnet pipeline are unrelated." + "reason": "Docs build on WSL2 and an import error for a ControlNet pipeline are separate problems." }, { "left": "issue:1246", "right": "issue:3627", "accept": false, - "reason": "Different training scripts with different runtime errors; no shared underlying bug is evident." + "reason": "Different example failures: FP16 gradient unscale in Dreambooth vs retrieve.py/custom diffusion error." }, { "left": "issue:1335", "right": "issue:1530", "accept": false, - "reason": "Missing constructor arg in ONNX pipeline and missing embedding field in VQDiffusion are different API issues." + "reason": "These are different missing-argument/missing-field bugs in different pipelines, not the same underlying defect." }, { "left": "issue:5327", "right": "issue:5466", "accept": false, - "reason": "Both mention SDXL loading, but one is textual inversion loading and the other is LoRA shape mismatch; different code paths." + "reason": "SDXL textual inversion loading and LoRA weight-shape mismatch are distinct loading paths." }, { "left": "issue:6238", "right": "issue:6921", "accept": false, - "reason": "Both touch SDXL, but one is a VAE issue and the other is about shipped weights deviating from official weights." + "reason": "An SDXL VAE issue is not the same as shipped weights deviating from official weights." }, { "left": "issue:11321", "right": "issue:9239", "accept": false, - "reason": "A README bug for Flux controlnet training is not the same as black outputs from ControlNet-Canny inference." + "reason": "README/documentation bug vs black-image runtime output from ControlNet-Canny are different." }, { "left": "issue:11547", "right": "issue:9599", "accept": false, - "reason": "Feature request for Flux controlnet support and a request for a LoRA-only FLUX finetune example are different asks." + "reason": "Feature request for Flux controlnet support and request for a LoRA-only example are not the same change." }, { "left": "issue:12308", "right": "issue:9619", "accept": false, - "reason": "Flash-attn GPU kernel support and ControlNet interrupt support are unrelated feature requests." + "reason": "Flash-attn kernel support for non-Hopper GPUs is unrelated to adding interrupt support in a ControlNet pipeline." }, { "left": "issue:1358", "right": "issue:699", "accept": false, - "reason": "Safetensors support and negative prompts are unrelated features." + "reason": "Safetensors support and negative-prompt support are unrelated feature requests." }, { "left": "issue:2024", "right": "issue:646", "accept": false, - "reason": "A corrupted notebook file and a scheduler attribute error are unrelated." + "reason": "Corrupted notebook file and PNDMScheduler API missing attribute are different classes of bugs." }, { "left": "issue:2359", "right": "issue:3612", "accept": false, - "reason": "Img2img latent output support and ControlNet creation/preprocessor guidance are different pipeline features." + "reason": "Img2img latent output type and ControlNet creation/preprocessor questions do not share the same code path." }, { "left": "issue:3403", "right": "issue:7061", "accept": false, - "reason": "TorchDynamo pipeline loading and a wrong docstring are not the same problem." + "reason": "Pipeline loading with TorchDynamo and an incorrect docstring are unrelated." }, { "left": "issue:3933", "right": "issue:4975", "accept": false, - "reason": "Graph compile failing on logging and LoRA loading inefficiency are different issues." + "reason": "Graph compile/logging failure and LoRA loading inefficiency are separate concerns." }, { "left": "issue:7968", "right": "issue:8673", "accept": false, - "reason": "Both mention attention, but they affect different models and different bugs; no same code-path evidence." + "reason": "Both are attention-related, but they affect different models and different mechanisms (motion module vs SD3 text padding masks)." }, { "left": "issue:3661", "right": "issue:6321", "accept": false, - "reason": "Stable Diffusion checkpoint conversion and a missing CLIP model detection issue are unrelated." + "reason": "Checkpoint conversion for SD2 weights and missing CLIP detection are unrelated loading issues." }, { "left": "issue:4236", "right": "issue:4677", "accept": false, - "reason": "SDXL refiner text encoder registration and an incorrect inpainting doc script are unrelated." + "reason": "SDXL refiner text-encoder registration and an inpainting doc script error are not the same bug." }, { "left": "issue:5245", "right": "issue:7147", "accept": false, - "reason": "Pipenv import freezing and a missing StableCascade decoder pipeline import are different problems." + "reason": "A pipenv import freeze and a missing pipeline import are different failures." }, { "left": "issue:2163", "right": "issue:2265", "accept": false, - "reason": "Training dtype mismatch and ONNXRuntime incompatibility are different training/runtime issues." + "reason": "UNet dtype mismatch during text-to-image training and ONNXRuntime incompatibility are separate example/runtime issues." }, { "left": "issue:4256", "right": "issue:5009", "accept": false, - "reason": "SDXL Dreambooth training error and a Wuerstchen error are unrelated." + "reason": "Different model-specific training/runtime errors: SDXL Dreambooth vs Wuerstchen." }, { "left": "issue:10165", "right": "issue:9613", "accept": false, - "reason": "LoRA scale seed preservation and a Timesteps.scale attribute error are unrelated bugs." - }, - { - "left": "issue:1101", - "right": "issue:1141", - "accept": false, - "reason": "Scheduler support and local-file pipeline support are different feature requests." - }, - { - "left": "issue:2616", - "right": "issue:525", - "accept": false, - "reason": "LoRA save_attn_procs NoneType error and loading images from local disk are unrelated." - }, - { - "left": "issue:2856", - "right": "issue:4562", - "accept": false, - "reason": "Both are model-conversion issues, but they target different source formats and failure modes; not the same concrete bug." + "reason": "LoRA seed preservation on scale change and a Timesteps.scale attribute error on a custom diffusion model are unrelated bugs." } ] }, @@ -95863,14 +95877,18 @@ "cluster_id": "cluster-50-3014", "nodes": [ "issue:10017", + "issue:1101", "issue:11035", "issue:11136", + "issue:1141", "issue:11447", "issue:12060", "issue:12195", "issue:12695", "issue:13225", "issue:181", + "issue:2616", + "issue:2856", "issue:3026", "issue:3174", "issue:3209", @@ -95878,13 +95896,12 @@ "issue:3808", "issue:4061", "issue:4134", - "issue:4178", "issue:4263", - "issue:4354", + "issue:4562", "issue:4764", - "issue:4912", "issue:5070", "issue:5079", + "issue:525", "issue:5313", "issue:5368", "issue:5466", @@ -95900,17 +95917,17 @@ "issue:7049", "issue:7197", "issue:7619", - "issue:782", "issue:7851", "issue:7877", - "issue:7962", "issue:7991", "issue:8604", - "issue:861", "issue:8732", "issue:9613" ], "soft_pairs": [ + "issue:1101|issue:1141", + "issue:2616|issue:525", + "issue:2856|issue:4562", "issue:3174|issue:5313", "issue:5070|issue:6928", "issue:5079|issue:7877", @@ -95931,183 +95948,180 @@ "issue:11447|issue:12195", "issue:12060|issue:13225", "issue:181|issue:3026", - "issue:4061|issue:4263", - "issue:4178|issue:4354", - "issue:4912|issue:7962", - "issue:782|issue:861" + "issue:4061|issue:4263" ], - "prepared_review_unit_hash": "708fb40eeb3a6dbb86a69d161f2e7b7ebdd44ae8cf96e1385b7161477a4fb502", + "prepared_review_unit_hash": "d59060458951c43e9b477976cff9a3990d0d48325648a8793e363d55580992bf", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33563, - "estimated_input_tokens": 8391, - "estimated_eval_tokens": 17038 + "serialized_chars": 33437, + "estimated_input_tokens": 8360, + "estimated_eval_tokens": 16976 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No true duplicate cluster here: the soft pairs are mostly same-subsystem but clearly different bugs, feature requests, or questions. I would not merge any of these issues into a single canonical ticket.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "This cluster is a loose collection of unrelated diffusers issues; none of the soft-edge pairs look like true duplicates or the same fix. The shared wording is usually only at the subsystem level (LoRA, SDXL, training, timesteps), which is too broad to merge.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue stands out: the items span many unrelated bugs/features, so this is not a coherent duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good canonical representative; the set is heterogeneous and the soft links are too weak to indicate one underlying bug.", + "best_issue_reason": "No issue is a strong global representative because the cluster is heterogeneous; at most they share the broader diffusers domain, not one underlying problem.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1101", + "right": "issue:1141", + "accept": false, + "reason": "Different requests: adding a scheduler vs supporting local files in custom pipelines." + }, + { + "left": "issue:2616", + "right": "issue:525", + "accept": false, + "reason": "Both mention loading/saving workflows, but one is a LoRA save error and the other is local-disk image loading in notebooks." + }, + { + "left": "issue:2856", + "right": "issue:4562", + "accept": false, + "reason": "Both are model-conversion failures, but they concern different conversion paths, formats, and error modes." + }, { "left": "issue:3174", "right": "issue:5313", "accept": false, - "reason": "Both involve LoRA, but one is model loading breakage and the other is non-revertible fuse behavior; different failure modes." + "reason": "Both involve LoRA, but one is loading broken by new features and the other is non-revertible fuse behavior." }, { "left": "issue:5070", "right": "issue:6928", "accept": false, - "reason": "ControlNet training-example flags vs depthmap colorization option; unrelated feature request." + "reason": "Unrelated: ControlNet training flags vs a Marigold depthmap colorization option." }, { "left": "issue:5079", "right": "issue:7877", "accept": false, - "reason": "LoRA loading support for a specific model family vs adding an Attention block argument; not the same bug." + "reason": "Loading a specific LoRA style is unrelated to adding an attention out_dim argument." }, { "left": "issue:5689", "right": "issue:6921", "accept": false, - "reason": "Residual-noise sampler issue vs shipped SDXL weights deviating from official weights; different root causes." + "reason": "Both touch SDXL quality, but one is a sampler noise issue and the other is a shipped-weights mismatch." }, { "left": "issue:6520", "right": "issue:6949", "accept": false, - "reason": "Blank validation output during adapter training vs DPMSolver blank images at even steps; distinct code paths." + "reason": "Blank validation output is not the same bug as even-step blank images from a specific scheduler." }, { "left": "issue:3432", "right": "issue:7012", "accept": false, - "reason": "CPU Half convolution runtime error vs v_prediction producing weird images; not the same defect." + "reason": "CPU half-precision convolution failure is unrelated to v_prediction image artifacts." }, { "left": "issue:10017", "right": "issue:6520", "accept": false, - "reason": "Dataset loading problem vs validation callback issue; no shared underlying bug." + "reason": "Dataset loading with load_dataset() is unrelated to blank validation results during adapter training." }, { "left": "issue:11136", "right": "issue:9613", "accept": false, - "reason": "SCM timestep support limitation vs missing Timesteps.scale attribute; separate compatibility issues." + "reason": "Different timesteps-related failures with different models and errors; not the same code path." }, { "left": "issue:3808", "right": "issue:7619", "accept": false, - "reason": "DeepFloyd IF stage-2 training failure vs Dreambooth channel-mismatch error; different pipelines and symptoms." + "reason": "Both are training issues, but they affect different pipelines and different underlying failures." }, { "left": "issue:4134", "right": "issue:7851", "accept": false, - "reason": "MPS ControlNet tensor-shape mismatch vs concurrent pipeline index-out-of-bounds; unrelated." + "reason": "MPS tensor-shape mismatch in ControlNet is unrelated to a concurrent-pipeline index error." }, { "left": "issue:3209", "right": "issue:7000", "accept": false, - "reason": "Autoencoder NaNs on large images vs a deprecation warning; not the same bug." + "reason": "NaN encoder outputs and a PyTorch deprecation warning are not the same bug." }, { "left": "issue:4764", "right": "issue:5466", "accept": false, - "reason": "LoRA weight format incompatibility vs SDXL image-to-image LoRA shape mismatch; similar area, different concrete issue." + "reason": "Both are LoRA load errors, but they involve different file types/models and different incompatibility causes." }, { "left": "issue:5923", "right": "issue:8732", "accept": false, - "reason": "Multi-GPU training hang vs checkpoint-save problem; separate training failures." + "reason": "Multi-GPU training hang is unrelated to checkpoint-saving trouble." }, { "left": "issue:7991", "right": "issue:8604", "accept": false, - "reason": "Basic tutorial bug vs float16-corrupted T5 encoder output; unrelated." + "reason": "A tutorial bug is unrelated to corrupt images from float16 T5 conversion." }, { "left": "issue:5368", "right": "issue:7049", "accept": false, - "reason": "Mixed-precision dtype mismatch vs IP-Adapter unpacking error; different call sites and causes." + "reason": "Mixed-precision dtype mismatch is unrelated to an IP-Adapter unpacking error." }, { "left": "issue:5577", "right": "issue:7197", "accept": false, - "reason": "from_single_file ControlNet key error vs resume-training shortest-edge key error; not the same failure." + "reason": "Both are KeyErrors, but they are different missing keys in different loading/resume scenarios." }, { "left": "issue:11035", "right": "issue:12695", "accept": false, - "reason": "Download symlink handling vs Hunyuan model support request; unrelated." + "reason": "A symlink/download option request is unrelated to Hunyuan v1.5 support." }, { "left": "issue:11447", "right": "issue:12195", "accept": false, - "reason": "Offline shard-file loading bug vs PyTorch custom_op compatibility break; different dependencies." + "reason": "Offline shard-file handling and a Torch custom_op compatibility failure are unrelated." }, { "left": "issue:12060", "right": "issue:13225", "accept": false, - "reason": "General question about DiT availability vs a request for RAE+DiT checkpoint support; not duplicate reports." + "reason": "A question about whether DiT blocks exist is not the same as support for published RAE+DiT checkpoints." }, { "left": "issue:181", "right": "issue:3026", "accept": false, - "reason": "Image-writing failure in unconditional training vs instruct-pix2pix accelerate launch error; distinct scripts and errors." + "reason": "Image-writing failure in a training script is unrelated to a one-GPU accelerate launch error." }, { "left": "issue:4061", "right": "issue:4263", "accept": false, - "reason": "LoRA load error vs LoRA unload broken; opposite operations and likely different fixes." - }, - { - "left": "issue:4178", - "right": "issue:4354", - "accept": false, - "reason": "LoRA unloading with XFormers vs wrong pipeline type from_single_file; unrelated issues." - }, - { - "left": "issue:4912", - "right": "issue:7962", - "accept": false, - "reason": "Checkpoint conversion request vs adding cross_attention_kwargs to a pipeline call; different feature areas." - }, - { - "left": "issue:782", - "right": "issue:861", - "accept": false, - "reason": "Model download auth-token regression vs checkpoint-to-diffusers conversion failure; unrelated." + "reason": "Both involve LoRA, but one is loading LoRA in SDXL and the other is unloading/fusing behavior." } ] }, @@ -96139,22 +96153,20 @@ "issue:11343", "issue:12173", "issue:12369", - "issue:1284", "issue:1604", "issue:1923", "issue:1957", "issue:2265", "issue:2412", "issue:2459", - "issue:2830", "issue:3120", "issue:3452", "issue:3570", - "issue:3679", "issue:3748", "issue:3954", + "issue:4178", "issue:4279", - "issue:4414", + "issue:4354", "issue:4438", "issue:4615", "issue:4771", @@ -96170,6 +96182,8 @@ "issue:7370", "issue:7606", "issue:7736", + "issue:782", + "issue:7962", "issue:8464", "issue:8493", "issue:8502", @@ -96181,6 +96195,9 @@ "issue:9822" ], "soft_pairs": [ + "issue:4178|issue:4354", + "issue:4912|issue:7962", + "issue:782|issue:861", "issue:1923|issue:2412", "issue:3120|issue:5803", "issue:4279|issue:5518", @@ -96202,182 +96219,186 @@ "issue:4804|issue:962", "issue:1009|issue:2459", "issue:10772|issue:9186", - "issue:11343|issue:8915", - "issue:1284|issue:4414", - "issue:2830|issue:3679" + "issue:11343|issue:8915" ], - "prepared_review_unit_hash": "698724ddb089ca07f289ae0d6a638ff1b3d34a2d40e454b0defa5b1f372a8648", + "prepared_review_unit_hash": "eb0a81000588768e2afab9e550c7635bb720195dc6d7dda60d7b044549596cbe", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 24, - "serialized_chars": 33625, - "estimated_input_tokens": 8407, - "estimated_eval_tokens": 17070 + "soft_pair_count": 25, + "serialized_chars": 33986, + "estimated_input_tokens": 8497, + "estimated_eval_tokens": 17250 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous; the soft-similarity pairs mostly share generic words like weight, import, support, or loading but describe different bugs, features, or code paths. I would not merge any of the proposed pairs.", - "confidence": 0.96, - "canonical_issue_reason": "Issue #9332 is the most concrete, broadly impactful bug report in the set: it has a clear loading-path failure, detailed reproduction, and describes incorrect model weights being loaded.", + "summary": "The items are mostly unrelated issues; the soft-similarity links look like broad topical overlap rather than true duplicates. I would not merge any of the proposed soft edges.", + "confidence": 0.86, + "canonical_issue_reason": "issue:9332 is the strongest standalone issue to anchor the set: it has the most detailed reproduction, a concrete loading symptom, and a broad but specific code path around model loading.", "canonical_pr_reason": null, - "best_issue_reason": "#9332 is the strongest representative issue here because it is a specific, user-visible defect with a clear runtime symptom and likely broader impact than the feature requests or environment-specific reports.", + "best_issue_reason": "issue:9332 is the best representative issue overall because it is detailed, actionable, and centered on a concrete bug rather than a vague feature request or environment-specific complaint.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4178", + "right": "issue:4354", + "accept": false, + "reason": "Both mention LoRA/pipeline behavior, but one is about unloading weights with XFormers and the other is about a wrong pipeline type from `from_single_file`; different bugs and code paths." + }, + { + "left": "issue:4912", + "right": "issue:7962", + "accept": false, + "reason": "One is about converting SDXL CivitAI checkpoints to diffusers, the other about adding `cross_attention_kwargs` to Instruct-Pix2Pix; this is a feature request vs a conversion problem." + }, + { + "left": "issue:782", + "right": "issue:861", + "accept": false, + "reason": "These are separate loading/conversion failures: model download auth handling vs checkpoint-to-diffusers conversion." + }, { "left": "issue:1923", "right": "issue:2412", "accept": false, - "reason": "Different problems: missing CUDA libraries in a Docker/bitsandbytes setup vs an error while using Dreambooth training." + "reason": "Missing CUDA libraries in a Docker/bitsandbytes setup is unrelated to a DreamBooth training error." }, { "left": "issue:3120", "right": "issue:5803", "accept": false, - "reason": "Not the same bug: one says ControlNet always produces the same output regardless of input, the other is about reproducing black images in SD." + "reason": "ControlNet training producing unchanged outputs is a distinct bug from reproducing black images in SD generation." }, { "left": "issue:4279", "right": "issue:5518", "accept": false, - "reason": "Both involve LoRA loading, but the concrete failures differ: SDXL LoRAs cannot be loaded vs a ModuleList attribute error in a different loader path." + "reason": "Both involve LoRA loading, but one is SDXL CivitAI loading failure and the other is an `AttributeError` involving `ModuleList`; not the same defect." }, { "left": "issue:8464", "right": "issue:9332", "accept": false, - "reason": "Different code paths and symptoms: SEGA with cpu offload fails when seeding, while #9332 is about from_pretrained loading identical weights." + "reason": "SEGA cpu-offload seed failure is a different pipeline/runtime bug from `from_pretrained()` loading the wrong weights." }, { "left": "issue:10198", "right": "issue:9822", "accept": false, - "reason": "Related to weights/Flux only at a high level; one is a weight mismatch bug, the other is a request about loading SDXL LoRAs into Flux." + "reason": "Both mention Flux/SDXL weights, but one is a Flux.1-dev weight mismatch and the other is loading SDXL LoRAs into Flux; related area, different concrete problem." }, { "left": "issue:10897", "right": "issue:7370", "accept": false, - "reason": "Separate feature requests for different video models (SkyReels-A1 vs OpenSora STDiT), not the same underlying change." + "reason": "These are separate support requests for different video/image models, not the same underlying change." }, { "left": "issue:11220", "right": "issue:8502", "accept": false, - "reason": "Documentation page not working as expected vs a runtime generation issue for SD3 medium; different artifacts and failure modes." + "reason": "Documentation page malfunction is not the same as an SD3 medium 512x512 generation issue." }, { "left": "issue:12173", "right": "issue:12369", "accept": false, - "reason": "Both are Wan-related, but one is i2i latent upscale and the other is a CineScale/4K feature request; not the same bug or change." + "reason": "Both are Wan-related, but one is latent upscale i2i behavior and the other is a 4K CineScale support request; different capabilities and likely different fixes." }, { "left": "issue:3452", "right": "issue:6703", "accept": false, - "reason": "Unrelated issues: low-VRAM Dreambooth training vs a textual inversion token being weakened by negative prompt handling." + "reason": "DreamBooth 4GB VRAM training concerns are unrelated to a textual inversion token being weakened by negative prompts." }, { "left": "issue:3570", "right": "issue:861", "accept": false, - "reason": "Different workflows: disabling downloads when loading a checkpoint vs converting a ckpt to diffusers." + "reason": "Disabling downloads when loading checkpoints is a different loading-path issue from checkpoint conversion to diffusers." }, { "left": "issue:3748", "right": "issue:7606", "accept": false, - "reason": "One is a float16/float32 mismatch loading a LoRA safetensor; the other is a deprecated LoRA processor API in Kandinsky." + "reason": "A float16/float32 mismatch while loading LoRA safetensors is unrelated to a deprecated Kandinsky LoRA processor class." }, { "left": "issue:4438", "right": "issue:5824", "accept": false, - "reason": "Both touch SDXL, but one is a generic pipeline error and the other is a specific bad default in get_add_time_ids(); not the same concrete bug." + "reason": "A generic SDXL pipeline error and a wrong `text_encoder_projection_dim` default are not clearly the same bug." }, { "left": "issue:4615", "right": "issue:5084", "accept": false, - "reason": "Different import errors for different pipeline classes/backends: ControlNet pipeline import vs FlaxStableDiffusionXLPipeline import." + "reason": "These are import errors for different pipeline symbols and different backends (ControlNet vs Flax SDXL)." }, { "left": "issue:4912", "right": "issue:7736", "accept": false, - "reason": "A checkpoint-conversion feature request is not the same as a request for supported callbacks." + "reason": "Custom checkpoint conversion and adding callbacks are unrelated changes." }, { "left": "issue:6218", "right": "issue:7102", "accept": false, - "reason": "Both involve ControlNet, but one is batched seeds in ControlNet/Img2Img and the other is multi-ControlNet plus multi-prompt behavior." + "reason": "Both involve ControlNet, but one is batched seeds in ControlNet/Img2Img and the other is multi-ControlNet with multi-prompt behavior; not the same concrete bug." }, { "left": "issue:1604", "right": "issue:2265", "accept": false, - "reason": "Different scripts and incompatibilities: lpw_stable_diffusion with diffusers 0.9.0 vs train_unconditional with ONNXRuntime." + "reason": "A broken LPW script on diffusers 0.9.0 is unrelated to ONNXRuntime incompatibility in the unconditional training script." }, { "left": "issue:1957", "right": "issue:3954", "accept": false, - "reason": "Both mention LR scheduling, but one is an init fix and the other is an incorrect step count in multi-GPU training." + "reason": "Both touch LR scheduling, but one is scheduler initialization in unconditional training and the other is step counting under multi-GPU; different failure modes." }, { "left": "issue:4771", "right": "issue:8493", "accept": false, - "reason": "Generic fine-tuned diffusion model loading vs a specific SD3Transformer2DModel state_dict mismatch; not the same issue." + "reason": "Loading a fine-tuned diffusion model is too broad to match the specific SD3Transformer2DModel state_dict error." }, { "left": "issue:4804", "right": "issue:962", "accept": false, - "reason": "Different bugs entirely: dictionary mutation during LoRA loading vs inpaint pipelines not supporting non-512 sizes." + "reason": "A LoRA loader dictionary mutation error is unrelated to inpaint pipeline size limitations." }, { "left": "issue:1009", "right": "issue:2459", "accept": false, - "reason": "Training-script issues in different places: a weight decay bug in textual inversion vs xformers attention not working in text-to-image LoRA training." + "reason": "Textual inversion weight decay and LoRA xformers support in a different training script are separate training issues." }, { "left": "issue:10772", "right": "issue:9186", "accept": false, - "reason": "Not the same change: a request for Sana Controlnet support vs a Differential Diffusion ControlNet pipeline issue/request." + "reason": "A feature request for Sana ControlNet support is not the same as a Differential Diffusion ControlNet pipeline issue." }, { "left": "issue:11343", "right": "issue:8915", "accept": false, - "reason": "Both are import errors, but they affect different modules and likely different dependency breakages." - }, - { - "left": "issue:1284", - "right": "issue:4414", - "accept": false, - "reason": "Performance complaint about larger batch size slowing sampling vs a sample example for Consistency Model; unrelated." - }, - { - "left": "issue:2830", - "right": "issue:3679", - "accept": false, - "reason": "Different text-to-video topics: a question about image/text-to-video with a modelscope model vs a post about new watermark-free models." + "reason": "Both are import failures, but the affected modules and root causes differ: a transformers update breaking diffusers imports versus a missing `AutoPipelineForText2Image` import." } ] }, @@ -96409,18 +96430,21 @@ "issue:11885", "issue:11962", "issue:11989", + "issue:1284", "issue:2822", - "issue:2838", + "issue:2830", "issue:2905", "issue:3113", "issue:3347", "issue:3366", "issue:3570", "issue:3674", + "issue:3679", "issue:3787", "issue:4158", "issue:4217", "issue:4398", + "issue:4414", "issue:4419", "issue:4724", "issue:4977", @@ -96436,12 +96460,9 @@ "issue:7057", "issue:7168", "issue:7519", - "issue:752", - "issue:7735", "issue:7945", "issue:8440", "issue:8534", - "issue:8759", "issue:8922", "issue:9042", "issue:9546", @@ -96450,6 +96471,8 @@ "issue:9958" ], "soft_pairs": [ + "issue:1284|issue:4414", + "issue:2830|issue:3679", "issue:4398|issue:4977", "issue:5953|issue:9546", "issue:6348|issue:9958", @@ -96471,182 +96494,180 @@ "issue:8534|issue:9686", "issue:10511|issue:11135", "issue:11022|issue:11885", - "issue:2822|issue:4217", - "issue:2838|issue:752", - "issue:7735|issue:8759" + "issue:2822|issue:4217" ], - "prepared_review_unit_hash": "6e4f0010f2723f8b84afd471e39e25a0cb2e9d6061490e094047e8c087a5aefd", + "prepared_review_unit_hash": "7c7ed2ff413f174889504efdfdc92fd8f3dc7edaac42389a500a7e9352faa32f", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33735, - "estimated_input_tokens": 8434, - "estimated_eval_tokens": 17124 + "serialized_chars": 33479, + "estimated_input_tokens": 8370, + "estimated_eval_tokens": 16996 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a false positive mix of unrelated issues across schedulers, ControlNet, LoRA training, Flux/SD3 features, and image-quality bugs. The only clearly related pair is the SD3 black/zero-image reports (8759 and 8922); everything else should be kept separate.", - "confidence": 0.96, - "canonical_issue_reason": "issue:8759 is the strongest issue candidate because it is concrete, technical, and already has a duplicate reference to issue:8922 for the same SD3 black-image symptom.", + "summary": "This cluster is a grab bag of unrelated Diffusers issues and feature requests across different pipelines, schedulers, model families, and utility bugs. None of the soft pairs look like the same underlying defect or change, so they should stay separate.", + "confidence": 0.98, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:8759 is the best representative issue: it describes a specific runtime warning in the SD3 image-processing path and anchors the duplicate black-image report cluster better than the broader or more speculative items.", + "best_issue_reason": "issue:10382 is the clearest standalone bug report here: it names a specific module, a concrete return-type mismatch, and an AttributeError symptom, making it the strongest triage anchor among otherwise unrelated items.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1284", + "right": "issue:4414", + "accept": false, + "reason": "Batch-size slowdown and Consistency Model sampling are different performance/problem domains with no shared bug path." + }, + { + "left": "issue:2830", + "right": "issue:3679", + "accept": false, + "reason": "A question about image/text-to-video use is unrelated to announcing new text-to-video models." + }, { "left": "issue:4398", "right": "issue:4977", "accept": false, - "reason": "Both mention generation/training problems, but one is an all-black image bug and the other is checkpoint cleanup in an example script; different underlying issues." + "reason": "Black SDXL LoRA outputs and removing checkpoints in a training example are unrelated issues." }, { "left": "issue:5953", "right": "issue:9546", "accept": false, - "reason": "One is a crash from mismatched image/control_image sizes, the other is Flux ControlNet training OOM during validation; same subsystem, different bug." + "reason": "Control image size mismatch crash and Flux validation OOM are different controlnet/training problems." }, { "left": "issue:6348", "right": "issue:9958", "accept": false, - "reason": "Both touch autoencoder support, but one asks for 3D image support and the other for temporal decoder tiling; these are distinct features." + "reason": "3D medical/video AutoencoderKL support and temporal-decoder tiling are different feature requests for different classes." }, { "left": "issue:7168", "right": "issue:7945", "accept": false, - "reason": "IP-Adapter image embedding shape confusion and combining community pipelines are unrelated." + "reason": "IP-Adapter image-embed tensor-shape troubleshooting is unrelated to a new community pipeline proposal." }, { "left": "issue:4158", "right": "issue:7168", "accept": false, - "reason": "from_single_file hanging and IP-Adapter embed shape issues are different code paths and symptoms." + "reason": "from_single_file stalling and IP-Adapter embedding shape issues are separate code paths and symptoms." }, { "left": "issue:2905", "right": "issue:5256", "accept": false, - "reason": "Scheduler-wide Karras sigmas support and Insta-Flow training support are separate changes." + "reason": "Scheduler-wide Karras sigmas support and Insta-Flow training support are unrelated feature requests." }, { "left": "issue:3570", "right": "issue:7057", "accept": false, - "reason": "A download-toggle problem while loading ckpt and a model checkpoint that doesn't work are too vague and do not point to the same concrete bug." + "reason": "Model download disabling and a checkpoint load failure are both loading-related but not the same concrete bug." }, { "left": "issue:4419", "right": "issue:7519", "accept": false, - "reason": "Offset-noise inference behavior and ignored cross_attention_kwargs scale are different pipeline behaviors." + "reason": "Offset-noise inference behavior and XFormers scale-kwarg handling are different attention/inference concerns." }, { "left": "issue:10382", "right": "issue:9930", "accept": false, - "reason": "A VAE encoder return-type AttributeError and a PAG adaptive-scale bug are unrelated." + "reason": "VAE encoder return-type mismatch with AttributeError is unrelated to PAG adaptive scale behavior." }, { "left": "issue:3347", "right": "issue:3787", "accept": false, - "reason": "NaN image output and an import error from partially initialized diffusers are entirely different failures." + "reason": "NaN image generation and a circular import ImportError are distinct failure modes." }, { "left": "issue:3674", "right": "issue:6363", "accept": false, - "reason": "Metadata JSONL parsing and FP16 gradient unscaling are unrelated training issues." + "reason": "metadata.jsonl parsing and FP16 gradient unscale errors occur in different training stages and code paths." }, { "left": "issue:5410", "right": "issue:5630", "accept": false, - "reason": "Both affect image quality, but latent upscaling blur and LMS leftover noise in SDXL img2img are different scheduler/pipeline problems." + "reason": "Blurry latent upscale and leftover noise in SDXL Img2Img are both quality complaints, but they point to different pipeline/scheduler bugs." }, { "left": "issue:10573", "right": "issue:11989", "accept": false, - "reason": "Multiple-ControlNet padding_mask_crop handling and a pure-black inference image are not the same bug." + "reason": "Multiple-controlnet padding_mask_crop failure and a black image/weight suspicion are unrelated symptoms." }, { "left": "issue:11962", "right": "issue:9042", "accept": false, - "reason": "GGUF quantized FLUX.1-Kontext-dev support and adding a Flux pipeline are separate feature requests." + "reason": "GGUF support for FLUX.1-Kontext-dev and adding the Flux pipeline are different model-support requests." }, { "left": "issue:3113", "right": "issue:5153", "accept": false, - "reason": "Training-free layout control and FreeU are different algorithmic techniques with different implementation goals." + "reason": "Cross-attention layout control and FreeU are unrelated diffusion technique feature requests." }, { "left": "issue:3366", "right": "issue:4724", "accept": false, - "reason": "Autocast being blocked by LoRALinearLayer and an SDXL Img2Img pipeline checkpoint issue are unrelated." + "reason": "Autocast breakage in LoRALinearLayer and SDXL Img2Img misbehavior with a checkpoint are different bugs." }, { "left": "issue:5977", "right": "issue:8440", "accept": false, - "reason": "Saving unet.unet LoRA keys and FromOriginalModelMixin subclass ValueError do not describe the same failure mode." + "reason": "Saving unwanted LoRA keys and FromOriginalModelMixin subclass ValueError are separate implementation issues." }, { "left": "issue:6001", "right": "issue:8922", "accept": false, - "reason": "Refiner loading returning the wrong pipeline class and SD3 quantized single-file black outputs are different bugs." + "reason": "Incorrect pipeline class returned from from_single_file and black output from quantized T5 are different problems." }, { "left": "issue:8534", "right": "issue:9686", "accept": false, - "reason": "Wrong logit-normal weighting in SD3 training and ControlNet pooled-projection behavior are distinct SD3 issues." + "reason": "Wrong SD3 dreambooth weighting and SD3 ControlNet pooled projection bugs are separate SD3 issues in different parts of the stack." }, { "left": "issue:10511", "right": "issue:11135", "accept": false, - "reason": "Scheduler redesign for SD3/Flux and a new SmoothCache pipeline are not the same change." + "reason": "Scheduler redesign for SD3/Flux and a new SmoothCache pipeline are unrelated proposals." }, { "left": "issue:11022", "right": "issue:11885", "accept": false, - "reason": "Quanto support for WAN 2.1 and a FLUX.1 LoRA-loading bug are unrelated." + "reason": "Quanto support for WAN 2.1 and a FLUX.1 LoRA loading bug are unrelated model-specific topics." }, { "left": "issue:2822", "right": "issue:4217", "accept": false, - "reason": "Slow ControlNet training convergence and slow safetensors loading are different performance concerns." - }, - { - "left": "issue:2838", - "right": "issue:752", - "accept": false, - "reason": "A ControlNet pipeline behavior request and multi-concept Dreambooth support are not the same bug/change." - }, - { - "left": "issue:7735", - "right": "issue:8759", - "accept": false, - "reason": "A missing dataset caption column and an SD3 black-image/image_processor cast warning are unrelated." + "reason": "Training convergence slowness and safetensors loading slowness are different performance issues." } ] }, @@ -96682,6 +96703,7 @@ "issue:2276", "issue:2462", "issue:2552", + "issue:2838", "issue:2966", "issue:3073", "issue:3811", @@ -96691,14 +96713,12 @@ "issue:4109", "issue:4321", "issue:4491", - "issue:4665", "issue:4872", "issue:5108", "issue:5135", "issue:5567", "issue:5574", "issue:5604", - "issue:5794", "issue:5835", "issue:6008", "issue:6154", @@ -96710,16 +96730,18 @@ "issue:730", "issue:752", "issue:7706", + "issue:7735", "issue:7946", "issue:7982", "issue:8429", "issue:8472", - "issue:9134", - "issue:9291", + "issue:8759", "issue:9667", "issue:9686" ], "soft_pairs": [ + "issue:2838|issue:752", + "issue:7735|issue:8759", "issue:2966|issue:6154", "issue:6320|issue:9667", "issue:10937|issue:11077", @@ -96741,182 +96763,180 @@ "issue:6435|issue:9686", "issue:10085|issue:10712", "issue:2276|issue:6774", - "issue:4061|issue:4491", - "issue:4665|issue:5794", - "issue:9134|issue:9291" + "issue:4061|issue:4491" ], - "prepared_review_unit_hash": "96895a1cec456ff36fa9615c2cb20ed1d1d9c0ea80d01cede2593337df9cec4c", + "prepared_review_unit_hash": "fdac0d2b1b31a8f526a5e1421f455f178ccd15d288d34557170575b69e7064a7", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 33378, - "estimated_input_tokens": 8345, - "estimated_eval_tokens": 16946 + "serialized_chars": 32973, + "estimated_input_tokens": 8244, + "estimated_eval_tokens": 16744 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No real duplicate cluster here: the items span unrelated bugs, feature requests, and training/inference issues across different pipelines and components. All soft pairs look like false positives based on broad topical similarity only.", - "confidence": 0.93, - "canonical_issue_reason": "There is no single canonical issue for this set because the issues are heterogeneous and do not describe the same underlying defect.", + "summary": "This cluster is not a true duplicate set; it mixes unrelated issues across training, LoRA, ControlNet, SDXL, VAE, and pipeline utilities. All soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "If one issue must be chosen as the most representative bug report, issue 10712 is the most concrete and self-contained runtime failure report among the set.", + "best_issue_reason": "Issue 10712 is the cleanest concrete bug report in the set, but it is not a canonical duplicate target because the rest of the cluster is unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2838", + "right": "issue:752", + "accept": false, + "reason": "Different requests: ControlNet pipeline behavior vs DreamBooth multi-concept support." + }, + { + "left": "issue:7735", + "right": "issue:8759", + "accept": false, + "reason": "Training dataset caption-column error vs SD3 image postprocessing warning; unrelated code paths." + }, { "left": "issue:2966", "right": "issue:6154", "accept": false, - "reason": "Both are training-script issues, but one is multi-GPU training behavior and the other is SDXL LoRA on an M2 Mac; different environments and failure modes." + "reason": "Multi-GPU instruct-pix2pix training issue vs SDXL LoRA training on M2 Mac; different failures." }, { "left": "issue:6320", "right": "issue:9667", "accept": false, - "reason": "Both involve model loading, but one is remote-model lookup failure in Emu2 and the other is local FLUX fp8 loading with a custom transformer file; not the same bug." + "reason": "Remote model loading failure vs local flux fp8 transformer file loading; different artifacts and symptoms." }, { "left": "issue:10937", "right": "issue:11077", "accept": false, - "reason": "One is a torch.compile VAE encode error, the other is a VAE conversion-script question; different code paths and problem types." + "reason": "torch.compile VAE encode error vs VAE conversion-script question; not the same bug." }, { "left": "issue:3811", "right": "issue:5604", "accept": false, - "reason": "Both are LoRA-related, but one says loaded LoRA has no effect on generation while the other is about unfusing multiple LoRAs; distinct behaviors." + "reason": "Both LoRA-related, but one is ineffective LoRA loading and the other is unfusing multiple LoRA; different problems." }, { "left": "issue:1161", "right": "issue:1883", "accept": false, - "reason": "Model conversion from Flax to PyTorch is unrelated to enabling gradient checkpointing for the VAE." + "reason": "Model conversion workflow vs VAE gradient checkpointing; unrelated issues." }, { "left": "issue:2462", "right": "issue:752", "accept": false, - "reason": "IterableDataset support for finetuning is not the same as Dreambooth multi-concept/class support." + "reason": "IterableDataset training support request vs DreamBooth multi-concept support; different feature asks." }, { "left": "issue:3073", "right": "issue:4064", "accept": false, - "reason": "A request to add consistency models is unrelated to adding PSLD upscaler/deblur/inpainting pipelines." + "reason": "Consistency-model request vs PSLD pipeline request; different features." }, { "left": "issue:4321", "right": "issue:5835", "accept": false, - "reason": "Black images in an ONNX DirectML pipeline are unrelated to using the pretrained MusicLDM pipeline." + "reason": "Black images in ONNX DirectML vs pretrained MusicLDM usage; unrelated." }, { "left": "issue:11363", "right": "issue:8429", "accept": false, - "reason": "Both mention inference examples, but one is ControlNet CUDA OOM and the other is an I2VGen-XL official script malfunction; different pipelines and symptoms." + "reason": "ControlNet example OOM vs I2VGen-XL script malfunction; different pipelines and failures." }, { "left": "issue:5135", "right": "issue:8472", "accept": false, - "reason": "LoRA quality degradation during generation is not the same as questions about jointly pretraining LoRA with other modules." + "reason": "Both LoRA-adjacent, but one is quality degradation and the other is about simultaneous pretraining with other modules." }, { "left": "issue:5567", "right": "issue:7982", "accept": false, - "reason": "Accelerator.prepare overwriting unet.config is a training-object mutation bug, while from_pipe with MultiControlNetModel is pipeline construction/copy behavior." + "reason": "Accelerate overwriting unet.config vs MultiControlNet from_pipe failure; unrelated code paths." }, { "left": "issue:7180", "right": "issue:7706", "accept": false, - "reason": "CPU offload failure in SDXL ControlNet img2img is unrelated to custom pipeline CUDA OOM." + "reason": "ControlNet cpu-offload failure vs custom pipeline OOM; both memory-related but not the same bug." }, { "left": "issue:1851", "right": "issue:730", "accept": false, - "reason": "Multi-GPU slowdown in Dreambooth is a performance issue, while DreamBooth learning locally vs Colab is a training-quality/environment issue." + "reason": "Multi-GPU slowdown vs local DreamBooth not learning; different symptoms and causes." }, { "left": "issue:4109", "right": "issue:7946", "accept": false, - "reason": "StableDiffusionXLInpaintPipeline not working is unrelated to custom_pipeline caching." + "reason": "SDXL inpaint pipeline failure vs custom_pipeline caching; unrelated." }, { "left": "issue:5574", "right": "issue:6008", "accept": false, - "reason": "Training a part of UNet attention parameters with LoRA is unrelated to IP-Adapter Plus Face failing with an LCM LoRA." + "reason": "LoRA parameter-training question vs IP-Adapter/LCM LoRA failure; different concrete issues." }, { "left": "issue:1608", "right": "issue:4872", "accept": false, - "reason": "Stopping at 500 steps on T4 and SDXL finetuning CUDA OOM are both resource/training issues, but they are different failure modes with different triggers." + "reason": "Training halts at 500 steps vs CUDA OOM during finetuning; both on T4, but distinct failures." }, { "left": "issue:2552", "right": "issue:4057", "accept": false, - "reason": "Tutorial questions for StableDiffusionInpaintPipeline are not the same as blurred output from a finetuned inpainting model." + "reason": "Tutorial question vs blurred output in finetuned inpainting model; not a duplicate." }, { "left": "issue:5108", "right": "issue:6274", "accept": false, - "reason": "T2I-Adapter Multiadapter XL exceptions and multiple-adapter breakage are related broadly, but not enough evidence they are the same concrete bug." + "reason": "Both involve multiple adapters, but one is T2I Multiadapter XL exceptions and the other is generic multiple-adapter breakage; not enough to be the same bug." }, { "left": "issue:6435", "right": "issue:9686", "accept": false, - "reason": "Default width/height initialization in ControlNet is unrelated to the SD3 ControlNet pooled projections bug." + "reason": "ControlNet default dimensions init bug vs SD3 controlnet pooled projections bug; different failures." }, { "left": "issue:10085", "right": "issue:10712", "accept": false, - "reason": "LyCORIS batch inference with skip_guidance_layers and SD3 prompt_embeds runtime errors are different pipeline/runtime paths." + "reason": "Runtime errors, but one is LyCORIS batch inference with skip_guidance_layers and the other is SD3 prompt_embeds; different paths." }, { "left": "issue:2276", "right": "issue:6774", "accept": false, - "reason": "xformers installation problems in a training notebook are unrelated to an AttentionProcessor dtype mismatch at runtime." + "reason": "xformers installation issue vs xformers mixed-precision dtype mismatch; unrelated." }, { "left": "issue:4061", "right": "issue:4491", "accept": false, - "reason": "SDXL LoRA load error and SDXL refiner broken in latest diffusers are both SDXL-related but describe different components and failures." - }, - { - "left": "issue:4665", - "right": "issue:5794", - "accept": false, - "reason": "Windows use_safetensors behavior with alternate bin files is unrelated to missing tokenizer/config.json in an SDXL model repo." - }, - { - "left": "issue:9134", - "right": "issue:9291", - "accept": false, - "reason": "Flux LoRA issues from SimpleTuner and kohya-ss key incompatibility are both FLUX LoRA topics, but one is a training/output issue and the other is a key-compatibility/loading issue." + "reason": "SDXL LoRA loading error vs SDXL refiner broken in latest diffusers; same area, different bug." } ] }, @@ -96942,7 +96962,6 @@ "nodes": [ "issue:10026", "issue:10277", - "issue:10500", "issue:1070", "issue:11127", "issue:12009", @@ -96961,20 +96980,19 @@ "issue:2907", "issue:3199", "issue:3213", - "issue:3624", "issue:3631", "issue:3884", "issue:4282", + "issue:4665", "issue:4686", "issue:4854", "issue:4923", - "issue:5509", "issue:5619", + "issue:5794", "issue:5829", "issue:5967", "issue:6240", "issue:6793", - "issue:6869", "issue:7102", "issue:7801", "issue:785", @@ -96983,6 +97001,8 @@ "issue:7946", "issue:8407", "issue:8898", + "issue:9134", + "issue:9291", "issue:9479", "issue:9674", "issue:9680", @@ -96990,6 +97010,8 @@ "issue:9889" ], "soft_pairs": [ + "issue:4665|issue:5794", + "issue:9134|issue:9291", "issue:12009|issue:13001", "issue:1214|issue:2244", "issue:7801|issue:9674", @@ -97011,182 +97033,180 @@ "issue:2858|issue:4923", "issue:2312|issue:3631", "issue:10026|issue:11127", - "issue:10277|issue:8407", - "issue:3624|issue:5509", - "issue:10500|issue:6869" + "issue:10277|issue:8407" ], - "prepared_review_unit_hash": "e3ff469e0f1d574d636ed7d5ecbeb7c92299a21dd0fb0fa29500325f16b9996d", + "prepared_review_unit_hash": "90d631ecee4595e31983f28669bc9853b93eb6c9e8735b1b8933a10e70e86388", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 33847, - "estimated_input_tokens": 8462, - "estimated_eval_tokens": 17180 + "serialized_chars": 33786, + "estimated_input_tokens": 8447, + "estimated_eval_tokens": 17150 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items span many unrelated pipeline, scheduler, training, and model-loading problems. None of the soft pairs look like the same concrete bug/change, so they should stay separate.", - "confidence": 0.92, - "canonical_issue_reason": "Issue 9704 is the strongest standalone bug report in the set: it is concrete, reproducible, and has the most substantial discussion around a specific runtime failure.", - "canonical_pr_reason": null, - "best_issue_reason": "Issue 9704 is the best representative issue because it documents a well-scoped failure mode with strong triage value and clear error symptoms.", - "best_pr_reason": null, + "summary": "This cluster is highly heterogeneous: the paired issues share only broad keywords or subsystems, not the same concrete bug. All soft edges should be rejected.", + "confidence": 0.91, + "canonical_issue_reason": "No single issue is a good canonical for the whole set; the items span unrelated failures in loading, training, inference, schedulers, and pipeline utilities.", + "canonical_pr_reason": "No PRs are present in this cluster.", + "best_issue_reason": "If forced to pick one representative, issue 13077 is the most concrete and well-engaged bug report, but it still does not represent the rest of the cluster well.", + "best_pr_reason": "No PRs are present, so there is no best PR candidate.", "soft_edge_verdicts": [ + { + "left": "issue:4665", + "right": "issue:5794", + "accept": false, + "reason": "Both concern loading assets, but one is a Windows safetensors/bin selection issue and the other is a missing tokenizer config in SDXL; different bugs." + }, + { + "left": "issue:9134", + "right": "issue:9291", + "accept": false, + "reason": "Both mention FLUX LoRAs, but one is a training issue with SimpleTuner and the other is key-format incompatibility from kohya-ss; not the same failure." + }, { "left": "issue:12009", "right": "issue:13001", "accept": false, - "reason": "Both involve GGUF loading, but they are different model families and different failures: Wan2.2 meta-tensor copy vs Flux2.Klein shape mismatch." + "reason": "Both are GGUF loading problems, but they involve different models and different errors: meta-tensor copy failure vs shape mismatch." }, { "left": "issue:1214", "right": "issue:2244", "accept": false, - "reason": "Both mention half precision, but one is CLIPTextModel fp16 behavior and the other is a Dreambooth CPU normalization error; different code paths." + "reason": "Both involve half-precision, but one is CLIPTextModel precision behavior and the other is a Dreambooth CPU norm implementation error; not the same code path." }, { "left": "issue:7801", "right": "issue:9674", "accept": false, - "reason": "Mixed-precision configuration not applying to the model is unrelated to bad SD3 ControlNet training results." + "reason": "One says fp16 mixed precision does not actually make the model fp16; the other is poor SD3 ControlNet training results. Different issues." }, { "left": "issue:7945", "right": "issue:9680", "accept": false, - "reason": "One is a feature request about combining community pipelines; the other is an SDXL unconditional generation bug." + "reason": "One is a feature request about combining community pipelines; the other is SDXL unconditional generation being broken." }, { "left": "issue:7946", "right": "issue:8898", "accept": false, - "reason": "Custom pipeline caching and from_pretrained subfolder path validation are different loading/caching problems." + "reason": "Custom pipeline caching and subfolder from_pretrained loading are unrelated bugs." }, { "left": "issue:1791", "right": "issue:2578", "accept": false, - "reason": "Mac/MPS UnCLIP support is unrelated to editable-mode installation failure." + "reason": "Mac/MPS support for UnCLIP and editable-install packaging failure are unrelated." }, { "left": "issue:3199", "right": "issue:3213", "accept": false, - "reason": "Both touch custom pipelines, but one is a broad custom-pipeline breakage report and the other is specifically lpw_stable_diffusion regression; not the same bug." + "reason": "Both mention broken pipelines, but one is a general custom-pipeline breakage and the other is a specific LPW pipeline regression." }, { "left": "issue:4282", "right": "issue:7102", "accept": false, - "reason": "A Multi-ControlNet feature request is not the same as a Multi-ControlNet/Multi-Prompt bug in an existing pipeline." + "reason": "Adding Multi-ControlNet is a feature request; the other is a bug with multicontrolnet and multi-prompt handling." }, { "left": "issue:5619", "right": "issue:5829", "accept": false, - "reason": "LCM img2img AutoencodeTiny encoding failure and missing unet_time_cond_proj_dim in train scripts are separate issues." + "reason": "Both are LCM-related, but one is an AutoencodeTiny img2img encode bug and the other is a missing train-script config argument." }, { "left": "issue:7851", "right": "issue:9479", "accept": false, - "reason": "Concurrent-call index errors are unrelated to inpainting changing unmasked regions." + "reason": "Concurrent pipeline-call indexing errors and inpainting mask leakage are different bugs." }, { "left": "issue:12689", "right": "issue:13077", "accept": false, - "reason": "LoRA checkpoint adapter loading and shallow-copy transformer weights are different weight-handling bugs." + "reason": "One is checkpoint LoRA adapter weights not loading; the other is transformer weights being shallow-copied in ZImageControlNet. Different underlying problems." }, { "left": "issue:1070", "right": "issue:4854", "accept": false, - "reason": "Both are inpainting-related, but one concerns a new-version bug and the other only processing the first image; not the same failure." + "reason": "Both are inpainting-related, but one is a new-version bug and the other only affects the first image in a batch; not the same symptom or fix." }, { "left": "issue:6240", "right": "issue:9704", "accept": false, - "reason": "device_map='auto' loading errors are unrelated to a cuDNN frontend execution-plan failure." + "reason": "device_map auto loading errors and a cuDNN frontend graph execution failure are unrelated." }, { "left": "issue:4686", "right": "issue:6793", "accept": false, - "reason": "Pipeline loading from a saved folder and Hydra serialization failure are distinct problems." + "reason": "Local pipeline reload failure and Hydra serialization failure are different concerns." }, { "left": "issue:1616", "right": "issue:2907", "accept": false, - "reason": "LPW pipeline compatibility on an older diffusers version is unrelated to calling enable_model_cpu_offload twice." + "reason": "LPW pipeline compatibility and calling enable_model_cpu_offload twice are unrelated." }, { "left": "issue:2447", "right": "issue:3884", "accept": false, - "reason": "Generic LoRA training errors and a specific train_text_to_image_lora.py failure are not clearly the same bug." + "reason": "Both are LoRA-related, but one is a generic LoRA error and the other is a specific training-script failure." }, { "left": "issue:5967", "right": "issue:9889", "accept": false, - "reason": "An announcement post and an inference segfault are not duplicate bug reports." + "reason": "An announcement about Stable Diffusion Turbo is not the same as a segfault in sdxl-turbo inference." }, { "left": "issue:1546", "right": "issue:785", "accept": false, - "reason": "Batched truncation-warning failure and PNDM scheduler performance cost are different issues." + "reason": "Truncation-warning batching error and PNDMScheduler performance cost are unrelated." }, { "left": "issue:2858", "right": "issue:4923", "accept": false, - "reason": "Seed image support in UnCLIP img2img is unrelated to the AudioLDM2 transcript feature request." + "reason": "UnCLIP seeding-image feature request and AudioLDM 2 transcript/text-to-speech feature request are different models and changes." }, { "left": "issue:2312", "right": "issue:3631", "accept": false, - "reason": "EMA save-hook failure and inpainting latent-input exception are separate code paths." + "reason": "EMA save-hook error and inpainting-latents exception are different training/inference failures." }, { "left": "issue:10026", "right": "issue:11127", "accept": false, - "reason": "Flux ControlNet memory usage under DeepSpeed and single-file loading for a Flux model are unrelated issues." + "reason": "Multi-GPU DeepSpeed memory behavior and a single-file Flux model load failure are unrelated." }, { "left": "issue:10277", "right": "issue:8407", "accept": false, - "reason": "Fluxfill Euler scheduler support and UniPC set_timesteps kwargs support are different scheduler-specific bugs." - }, - { - "left": "issue:3624", - "right": "issue:5509", - "accept": false, - "reason": "A getattr attribute-type error is unrelated to image batch-size validation for ndarray/cv2 inputs." - }, - { - "left": "issue:10500", - "right": "issue:6869", - "accept": false, - "reason": "BitsAndBytes device placement mismatch and export_to_video unpacking errors are different failures in different subsystems." + "reason": "Using Euler in fluxfill and UniPC scheduler kwargs support are separate scheduler issues." } ] }, @@ -97205,16 +97225,16 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10337", + "issue:10500", "issue:10526", "issue:10590", "issue:1127", - "issue:11570", "issue:1214", "issue:1271", "issue:13015", @@ -97222,18 +97242,16 @@ "issue:13386", "issue:1785", "issue:1895", - "issue:2161", - "issue:2326", "issue:2495", "issue:2644", "issue:3094", "issue:3181", "issue:3209", "issue:3280", + "issue:3624", "issue:3683", "issue:3752", "issue:3787", - "issue:3958", "issue:4014", "issue:4122", "issue:4134", @@ -97243,9 +97261,9 @@ "issue:4874", "issue:499", "issue:5000", - "issue:5328", "issue:5353", "issue:5463", + "issue:5509", "issue:5546", "issue:5596", "issue:5646", @@ -97253,6 +97271,7 @@ "issue:6639", "issue:6661", "issue:6807", + "issue:6869", "issue:7450", "issue:7548", "issue:8609", @@ -97260,6 +97279,8 @@ "issue:9530" ], "soft_pairs": [ + "issue:3624|issue:5509", + "issue:10500|issue:6869", "issue:1214|issue:2644", "issue:3094|issue:5596", "issue:1271|issue:3181", @@ -97282,195 +97303,186 @@ "issue:5353|issue:8609", "issue:5646|issue:7450", "issue:1785|issue:5000", - "issue:6639|issue:9530", - "issue:3958|issue:5328", - "issue:10337|issue:11570", - "issue:2161|issue:2326" + "issue:6639|issue:9530" ], - "prepared_review_unit_hash": "2cc66c4a225f1baf88c2a4e8925e17228e995da81d8449e7288e7f22916487ba", + "prepared_review_unit_hash": "7aa38c32a591aba342a00f970d874c3ed720b5396aa393f79954f1eb02bdf338", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 26, - "serialized_chars": 34680, - "estimated_input_tokens": 8670, - "estimated_eval_tokens": 17596 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33945, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: most items are unrelated bug reports that only share broad subsystem vocabulary (Flux, LoRA, MPS, SDXL, schedulers, etc.). I found no soft pair that clearly represents the same concrete bug/change, so no merges are warranted.", - "confidence": 0.23, - "canonical_issue_reason": "If a single issue must be chosen, 10337 is the most concrete and narrowly scoped report, but the cluster is too mixed to treat it as a true canonical duplicate target.", + "summary": "This cluster is heterogeneous: the items cover unrelated model families, devices, and failure modes. None of the soft pairs looks like the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a valid duplicate canonical here; the titles span distinct bugs across different pipelines and error types. If forced to pick a representative, issue:10500 is the clearest runtime failure report, but it does not subsume the rest.", "canonical_pr_reason": null, - "best_issue_reason": "10337 is the clearest representative of a specific reproducible bug report; the others are either similarly broad, more generic, or clearly about different failure modes.", + "best_issue_reason": "issue:10500 is the best standalone issue to keep as a representative because it is concrete and well-scoped, but it is not a good merge target for the other issues.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3624", + "right": "issue:5509", + "accept": false, + "reason": "Different failures: generic getattr TypeError vs image/prompt batch-size mismatch." + }, + { + "left": "issue:10500", + "right": "issue:6869", + "accept": false, + "reason": "Both are runtime errors, but one is a device-mismatch crash and the other is an export_to_video unpacking bug." + }, { "left": "issue:1214", "right": "issue:2644", "accept": false, - "reason": "Half-precision CLIPTextModel failure and DreamBooth GPU-count question are different problems." + "reason": "Half-precision CLIPTextModel bug vs a multi-GPU training question; not the same underlying problem." }, { "left": "issue:3094", "right": "issue:5596", "accept": false, - "reason": "Both mention pokemon examples, but one is OOM during finetuning and the other is a dict/convert installation error." + "reason": "OOM during finetuning vs dict.convert error while building captions; unrelated code paths." }, { "left": "issue:1271", "right": "issue:3181", "accept": false, - "reason": "Dreambooth notebook dependency installation and TensorRT example TypeError are unrelated." + "reason": "Dependency installation problem vs TensorRT example TypeError; different subsystems." }, { "left": "issue:2495", "right": "issue:6661", "accept": false, - "reason": "Both involve MPS, but black-image output in upscaling and OOM during training are different code paths." + "reason": "MPS black-image bug vs MPS OOM during training; same platform, different issues." }, { "left": "issue:4122", "right": "issue:5546", "accept": false, - "reason": "Both are scheduler-step complaints, but one is a DPMSolver timesteps bug and the other is a base+refiner Karras-sigmas ensemble issue." + "reason": "Both mention scheduler step counts, but they target different pipeline setups and likely different bugs." }, { "left": "issue:4874", "right": "issue:5463", "accept": false, - "reason": "A local low_res_scheduler bug and an UnboundLocalError on image_latents are distinct failures." + "reason": "Local low_res_scheduler bug vs unbound image_latents variable; not the same defect." }, { "left": "issue:10526", "right": "issue:9461", "accept": false, - "reason": "Flux quantized WeightQBytesTensor construction error and vague Flux inference issue are not the same concrete bug." + "reason": "Flux FP8 quantization TypeError vs a generic Flux inference issue; no shared concrete failure." }, { "left": "issue:1895", "right": "issue:499", "accept": false, - "reason": "EMA multi-GPU update behavior and unconditional training script failure are different issues." + "reason": "EMA multi-GPU update bug vs example training script failure; distinct behavior and scope." }, { "left": "issue:3683", "right": "issue:6278", "accept": false, - "reason": "Saving Multi-ControlNet and multi-GPU LCM distillation are unrelated." + "reason": "save_pretrained support request vs multi-GPU training run failure; unrelated changes." }, { "left": "issue:3787", "right": "issue:4598", "accept": false, - "reason": "A circular/partial import traceback and missing-dependency error messaging are different problems." + "reason": "Partially initialized import traceback vs missing-dependency error messaging; different problem classes." }, { "left": "issue:4697", "right": "issue:6661", "accept": false, - "reason": "LoRA loading after 0.20.0 and MPS OOM during training are unrelated." + "reason": "LoRA loading regression vs MPS training OOM; not the same underlying bug." }, { "left": "issue:13015", "right": "issue:13386", "accept": false, - "reason": "NPU device mismatch in QwenImageEditPlus and tensor concatenation crash with prior preservation are different bugs." + "reason": "NPU same-device tensor error vs prior-preservation concatenation crash; different code paths." }, { "left": "issue:1127", "right": "issue:1335", "accept": false, - "reason": "Both mention OnnxStableDiffusionPipeline, but import-time DML traceback and constructor missing vae_encoder are not the same code-path problem." + "reason": "ONNX import traceback vs constructor missing vae_encoder; both ONNX-related, but not the same bug." }, { "left": "issue:3209", "right": "issue:7548", "accept": false, - "reason": "AutoencoderKL NaNs on large images and AnimateDiff flicker are different symptoms and likely different causes." + "reason": "AutoencoderKL NaNs vs flicker in AnimateDiff outputs; different symptoms and mechanisms." }, { "left": "issue:3280", "right": "issue:3752", "accept": false, - "reason": "DeepFloyd IF scheduler compatibility and DreamBooth inference failure are not the same bug." + "reason": "Scheduler incompatibility in DeepFloyd IF vs finetuned DeepFloyd IF inference issue; too broad to merge." }, { "left": "issue:4014", "right": "issue:4438", "accept": false, - "reason": "SDXL watermark artifacts and a generic SDXL pipeline error are unrelated." + "reason": "Watermark artifacts vs generic SDXL error; no concrete shared failure." }, { "left": "issue:4134", "right": "issue:6807", "accept": false, - "reason": "ControlNet tensor-size mismatch on MPS and test collection failures on MPS are different issues." + "reason": "ControlNet tensor-shape mismatch on MPS vs MPS test collection failure; different layers of failure." }, { "left": "issue:6661", "right": "issue:6807", "accept": false, - "reason": "Training OOM on MPS and contributor test collection failures on MPS are not the same underlying bug." + "reason": "Training OOM on MPS vs test collection failure on MPS; platform overlap only." }, { "left": "issue:10337", "right": "issue:10590", "accept": false, - "reason": "Flux IPAdapter quantization failure and Sana BFloat16 tiled VAE decode crash are different model/features." + "reason": "Quantized Flux IPAdapter failure vs Sana BF16 tiled VAE decode issue; different models and failure modes." }, { "left": "issue:5353", "right": "issue:8609", "accept": false, - "reason": "FreeU usage question and custom pipeline/IP-Adapter incompatibility are unrelated." + "reason": "Usage question about FreeU vs custom_pipeline/IP_adapter incompatibility; not duplicate bugs." }, { "left": "issue:5646", "right": "issue:7450", "accept": false, - "reason": "Scheduler stability on a specific model and AutoencoderKL multithreading issues are different code paths." + "reason": "Scheduler stability concerns vs AutoencoderKL multithreading issues; unrelated subsystems." }, { "left": "issue:1785", "right": "issue:5000", "accept": false, - "reason": "UnCLIP num_images_per_prompt bug and LoRA fuse/unfuse regression are unrelated." + "reason": "UnCLIP num_images_per_prompt bug vs LoRA fuse/unfuse regression; different features." }, { "left": "issue:6639", "right": "issue:9530", "accept": false, - "reason": "Different LoRA-loading outputs and GPU-memory freeing after torch.compile are unrelated concerns." - }, - { - "left": "issue:3958", - "right": "issue:5328", - "accept": false, - "reason": "Civitai LoRA with CPU offload and all Civitai LoRAs failing are broadly related but not specific enough to treat as the same bug without stronger evidence." - }, - { - "left": "issue:10337", - "right": "issue:11570", - "accept": false, - "reason": "Both involve quantization, but IPAdapter on Flux and load_lora_weights on bitsandbytes text encoders are different concrete failures." - }, - { - "left": "issue:2161", - "right": "issue:2326", - "accept": false, - "reason": "LoRA training inpainting support and LoRA conversion to safetensors/ckpt are different tasks and failures." + "reason": "Different SDXL LoRA-loading outputs vs GPU memory freeing after torch.compile; unrelated concerns." } ] }, @@ -97489,36 +97501,40 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ - "issue:1014", + "issue:10337", "issue:10526", "issue:10533", "issue:10550", "issue:10848", + "issue:11570", "issue:1249", - "issue:1261", "issue:1552", "issue:1918", "issue:1944", "issue:1957", "issue:2041", + "issue:2161", "issue:2201", "issue:2244", + "issue:2326", "issue:2402", "issue:2507", "issue:2560", "issue:2722", "issue:3454", "issue:3908", + "issue:3958", "issue:4427", "issue:4692", "issue:4888", "issue:5109", "issue:5112", + "issue:5328", "issue:579", "issue:5805", "issue:6028", @@ -97532,18 +97548,17 @@ "issue:746", "issue:7485", "issue:7851", - "issue:8469", "issue:8565", "issue:9170", - "issue:9366", "issue:9523", "issue:954", - "issue:9731", - "issue:9813", "issue:9927", "issue:9941" ], "soft_pairs": [ + "issue:3958|issue:5328", + "issue:10337|issue:11570", + "issue:2161|issue:2326", "issue:4692|issue:9170", "issue:6218|issue:7485", "issue:2402|issue:6028", @@ -97565,189 +97580,186 @@ "issue:2244|issue:5805", "issue:4427|issue:7354", "issue:10526|issue:10533", - "issue:5112|issue:7851", - "issue:9366|issue:9731", - "issue:8469|issue:9813", - "issue:1014|issue:1261" + "issue:5112|issue:7851" ], - "prepared_review_unit_hash": "763a70d4378e441fe70bd1f785188adacb7d8b5b3f1f7da3e00a46fdecb0ccfe", + "prepared_review_unit_hash": "cb6d2980c060a9270c51cb3184981042a526baee3842bcc7720d70cf21ffcd1a", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 34515, - "estimated_input_tokens": 8629, - "estimated_eval_tokens": 17514 + "serialized_chars": 34220, + "estimated_input_tokens": 8555, + "estimated_eval_tokens": 17366 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose grab bag of unrelated diffusers issues; the proposed soft edges are mostly broad keyword matches (LoRA, ControlNet, scheduler, offline, memory) rather than the same underlying bug, so none should be merged.", - "confidence": 0.97, - "canonical_issue_reason": "No clear canonical issue: the items span many unrelated bugs and features (Dreambooth, ControlNet, LoRA, schedulers, IP-Adapter, ONNX, Windows, offline loading, memory/OOM), so none is a safe duplicate anchor.", + "summary": "This cluster is not a true duplicate set; it mixes many distinct bugs around quantization, LoRA/IP-Adapter loading, schedulers, offline loading, ControlNet, and generic runtime errors. I would not merge any soft pairs.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue: the items cover several unrelated failure modes, so any representative would be misleading as a duplicate anchor.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a strong global representative because the cluster is not centered on one code-path failure; the titles point to distinct problems across different pipelines and scripts.", + "best_issue_reason": "issue:10337 is the clearest representative of the quantized-adapter loading theme, but only as a loose topical anchor\u2014not a duplicate target for the whole cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3958", + "right": "issue:5328", + "accept": false, + "reason": "Both mention LoRA/Civitai and offload-related failures, but one is about model CPU offload and the other is a broader LoRA regression; not the same concrete bug." + }, + { + "left": "issue:10337", + "right": "issue:11570", + "accept": false, + "reason": "Different code paths and symptoms: IPAdapter + quantized transformers vs load_lora_weights with bitsandbytes quantized text_encoder." + }, + { + "left": "issue:2161", + "right": "issue:2326", + "accept": false, + "reason": "Training LoRA on an inpainting model is unrelated to converting a LoRA to safetensors/ckpt." + }, { "left": "issue:4692", "right": "issue:9170", "accept": false, - "reason": "Both mention memory, but one is a general UNet memory concern and the other is a specific SDXL+ControlNet VRAM requirement; different problems." + "reason": "Both are memory-related, but one is a UNet memory question and the other is SDXL+ControlNet VRAM requirements; not the same bug." }, { "left": "issue:6218", "right": "issue:7485", "accept": false, - "reason": "Batched seed handling in ControlNet pipelines is unrelated to the safety_checker NoneType callable error." + "reason": "Batched seeds in ControlNet pipelines and a NoneType safety_checker error are separate pipeline issues." }, { "left": "issue:2402", "right": "issue:6028", "accept": false, - "reason": "CUDA-not-enabled setup failure is not the same as a Kandinsky 3.0 out-of-memory runtime error." + "reason": "CUDA not enabled is an environment/setup error, while Kandinsky 3.0 OOM is a runtime memory exhaustion issue." }, { "left": "issue:4888", "right": "issue:6110", "accept": false, - "reason": "Both involve offline loading, but one is generic safetensors model loading and the other is LoRA weight loading; not the same bug." + "reason": "Both involve offline loading, but one is generic safetensors model loading and the other is LoRA weights specifically; too broad to treat as one bug." }, { "left": "issue:1552", "right": "issue:2201", "accept": false, - "reason": "Attention slicing incompatibility is unrelated to tokenizer padding differences." + "reason": "Attention slicing failure and tokenizer padding differences are different functional areas." }, { "left": "issue:2041", "right": "issue:3454", "accept": false, - "reason": "Both are Dreambooth-related, but one is Windows compatibility and the other is breakage on latest code; different failure modes." + "reason": "Windows compatibility for Dreambooth is not the same as train_dreambooth_lora.py failing on latest versions." }, { "left": "issue:6517", "right": "issue:9941", "accept": false, - "reason": "A specific import error for SDXL ControlNet Img2Img does not match a generic Colab runtime failure." + "reason": "Import failure for a specific ControlNet SDXL module is unrelated to a general Colab runtime failure." }, { "left": "issue:1957", "right": "issue:6297", "accept": false, - "reason": "LR scheduler initialization fix is unrelated to Dreambooth SDXL training on photos." + "reason": "LR scheduler init in unconditional training is unrelated to Dreambooth SDXL training not working on photos." }, { "left": "issue:5109", "right": "issue:8565", "accept": false, - "reason": "Both are LoRA-loading issues, but they hit different pipelines and different exceptions; not the same concrete code-path bug." + "reason": "Both are LoRA-loading errors, but they hit different pipelines and different failure modes (exception vs IndexError)." }, { "left": "issue:1944", "right": "issue:579", "accept": false, - "reason": "Dreambooth producing random outputs is unrelated to a missing config.json in textual inversion." + "reason": "Dreambooth output quality and missing config.json in textual inversion are unrelated." }, { "left": "issue:746", "right": "issue:954", "accept": false, - "reason": "ONNX revision loading and login/download access are different kinds of load failures." + "reason": "Onnx revision loading and post-login loading failures are different problems." }, { "left": "issue:9523", "right": "issue:9927", "accept": false, - "reason": "Training-script error mapping across SD3/SDXL/Flux is not the same as an HTTP header-too-large failure in SDv3 controlnet training." + "reason": "ControlNet training mapping errors and HeaderTooLarge during SD3 ControlNet training are distinct errors." }, { "left": "issue:2507", "right": "issue:3908", "accept": false, - "reason": "Safetensors GPU loading performance is unrelated to caching a model in a Streamlit demo." + "reason": "Safetensors GPU load performance and Streamlit model caching are unrelated." }, { "left": "issue:2560", "right": "issue:2722", "accept": false, - "reason": "Different outputs from generator lists is unrelated to ControlNet model conversion mismatches." + "reason": "List-of-generators output differences and ControlNet UNet conversion mismatch are different bugs." }, { "left": "issue:1249", "right": "issue:1918", "accept": false, - "reason": "A numpy device attribute error in OnnxPipeline is unrelated to a DDIM scheduler issue." + "reason": "An OnnxPipeline numpy/device AttributeError is unrelated to a DDIM scheduler example issue." }, { "left": "issue:6222", "right": "issue:6335", "accept": false, - "reason": "DDPM batch-size dependence and output-size control in ControlNet Img2Img are distinct issues." + "reason": "Batch-size dependence in DDPM scheduler and ControlNet img2img output sizing are separate pipeline behaviors." }, { "left": "issue:10550", "right": "issue:10848", "accept": false, - "reason": "LoRA loading into a quantized base model is unrelated to IP-Adapter meta-loading with an unexpected device argument." + "reason": "Quantized-base-model LoRA loading and load_model_dict_into_meta device keyword handling are different loading paths." }, { "left": "issue:2722", "right": "issue:5805", "accept": false, - "reason": "ControlNet conversion mismatch and XLA float/half type mismatch are different bugs." + "reason": "ControlNet conversion mismatch and an XLA float/half type error are not the same failure." }, { "left": "issue:2244", "right": "issue:5805", "accept": false, - "reason": "Dreambooth CPU half-precision normalization failure is unrelated to the SDXL XLA type mismatch." + "reason": "CPU half-precision normalization failure and XLA input/bias type mismatch are different runtime errors." }, { "left": "issue:4427", "right": "issue:7354", "accept": false, - "reason": "A broken Multiadapter example is not the same as PNDMScheduler incompatibility with DDPMPipeline." + "reason": "A broken MultiAdapter example and PNDM/DDPM compatibility are unrelated." }, { "left": "issue:10526", "right": "issue:10533", "accept": false, - "reason": "Both are Flux-related, but one is an FP8 quantization TypeError and the other is black-image output in Flux Fill fp16; different root causes." + "reason": "Different Flux problems: one is an optimum.quanto FP8 tensor construction TypeError, the other is black-image generation in fp16." }, { "left": "issue:5112", "right": "issue:7851", "accept": false, - "reason": "A torch.Size float/shape error is unrelated to an out-of-bounds index during concurrent pipeline calls." - }, - { - "left": "issue:9366", - "right": "issue:9731", - "accept": false, - "reason": "Scheduler step/strength failure is unrelated to training a LoRA on Playground2.5." - }, - { - "left": "issue:8469", - "right": "issue:9813", - "accept": false, - "reason": "Both mention IP-Adapter, but one is corrupt faces from a specific model and the other is batch feature mixing in embedding preparation; not the same concrete bug." - }, - { - "left": "issue:1014", - "right": "issue:1261", - "accept": false, - "reason": "A vague upgrade problem is far too broad and does not match the xformers backward-pass assertion." + "reason": "A torch.Size float-item error in ControlNet img2img and an out-of-bounds index under concurrent calls are unrelated." } ] }, @@ -97766,14 +97778,15 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ "issue:10010", "issue:10052", "issue:10120", + "issue:1014", "issue:10202", "issue:10204", "issue:10490", @@ -97786,13 +97799,9 @@ "issue:10796", "issue:1088", "issue:10895", - "issue:10932", - "issue:10963", - "issue:11047", - "issue:11343", "issue:11383", "issue:11939", - "issue:12195", + "issue:1261", "issue:12712", "issue:1805", "issue:1831", @@ -97812,15 +97821,20 @@ "issue:6602", "issue:7312", "issue:7802", + "issue:8469", "issue:8950", "issue:9115", - "issue:9223", "issue:9309", + "issue:9366", "issue:9374", "issue:9420", - "issue:9637" + "issue:9731", + "issue:9813" ], "soft_pairs": [ + "issue:9366|issue:9731", + "issue:8469|issue:9813", + "issue:1014|issue:1261", "issue:2091|issue:5951", "issue:4566|issue:4728", "issue:5254|issue:6336", @@ -97841,190 +97855,180 @@ "issue:1076|issue:1805", "issue:10796|issue:11383", "issue:1088|issue:2444", - "issue:10895|issue:11939", - "issue:10932|issue:9223", - "issue:10932|issue:9637", - "issue:10963|issue:11343", - "issue:11047|issue:12195" + "issue:10895|issue:11939" ], - "prepared_review_unit_hash": "f993bbdc9581e411037ff437379ad339442b81c41a0c9f033dc5a4885f2484af", + "prepared_review_unit_hash": "652ccde4e8922bb0dae2eed2e1debbb516eded724fb6b2b6cd67abef2d6e15f1", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 33948, - "estimated_input_tokens": 8487, - "estimated_eval_tokens": 17230 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33220, + "estimated_input_tokens": 8305, + "estimated_eval_tokens": 16866 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a true duplicate cluster: the paired items are mostly unrelated issues and feature requests spanning different models, pipelines, and failure modes.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue: the items span unrelated bugs and feature requests across different pipelines/models.", + "summary": "This cluster is a loose mix of unrelated Diffusers issues and feature requests, with several Flux-related items but no clear duplicate group. The soft pairs are mostly broad topical similarities rather than the same concrete bug or change, so none should be merged.", + "confidence": 0.91, + "canonical_issue_reason": "No single issue is a reliable canonical duplicate target here: the items span different pipelines, models, and failure modes. If forced to pick a representative issue, issue:9374 is the broadest recent Flux API-support request, but it is not a true duplicate anchor.", "canonical_pr_reason": null, - "best_issue_reason": "If one must anchor the set, issue 10204 is the most detailed open bug report, but it only covers FluxFill inpainting quality and does not unify the cluster.", + "best_issue_reason": "issue:9374 is the best representative of the recent Flux subgroup because it covers multiple Flux pipelines in one API-support request, but the cluster overall is too heterogeneous for a strong canonical choice.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:9366", + "right": "issue:9731", + "accept": false, + "reason": "Different problem types: scheduler failure in SDXL image-to-image vs a training/how-to style request for Playground2.5 Lora use." + }, + { + "left": "issue:8469", + "right": "issue:9813", + "accept": false, + "reason": "Both touch IP-Adapter, but one reports corrupt face outputs and the other a batch-embedding mixing bug in a helper function; not the same code-path bug." + }, + { + "left": "issue:1014", + "right": "issue:1261", + "accept": false, + "reason": "Upgrade-related regression report vs xformers CUDA backward assertion; unrelated symptoms and subsystems." + }, { "left": "issue:2091", "right": "issue:5951", "accept": false, - "reason": "Resume-from-checkpoint training bug vs a Pixart training topic; different models and failure modes." + "reason": "Checkpoint resume in multi-GPU text-to-image training vs PixArt training question; different scripts and issues." }, { "left": "issue:4566", "right": "issue:4728", "accept": false, - "reason": "Both mention resuming training, but one is SDXL LoRA checkpoint resume and the other is OOM on UNet resume; not the same bug." + "reason": "Both involve resuming training, but one is SDXL LoRA checkpoint resume and the other is GPU OOM on resume in text_to_image; not the same bug." }, { "left": "issue:5254", "right": "issue:6336", "accept": false, - "reason": "Multi-ControlNet feature request vs a TensorRT VAE compilation error; unrelated code paths." + "reason": "Multi-ControlNet support request for an inpaint pipeline vs TensorRT VAE compilation error with AutoencoderTiny; unrelated changes." }, { "left": "issue:3694", "right": "issue:5064", "accept": false, - "reason": "Missing model/config files during fine-tuning vs disk space exhaustion during inference; different problems." + "reason": "Missing model/config files during finetuning example setup vs out-of-space during refiner inference; different failures." }, { "left": "issue:4361", "right": "issue:6602", "accept": false, - "reason": "Both involve save/load behavior, but one is a safetensors shared-tensors saving regression and the other is a fuse_lora()/save_pretrained() load error." + "reason": "Both are save/load related, but shared-tensor safetensors serialization and fuse_lora/save_pretrained loading are distinct mechanisms and bugs." }, { "left": "issue:1831", "right": "issue:4534", "accept": false, - "reason": "Both are dreambooth-related, but one is class-image precision mismatch and the other is an MPS validation crash." + "reason": "Dreambooth class-image precision mismatch vs SDXL Dreambooth validation crash on MPS; different pipelines and failure modes." }, { "left": "issue:10010", "right": "issue:9420", "accept": false, - "reason": "Both are FLUX-related, but one is a missing model_index.json download and the other is loading a saved FLUX+Hyper-SD transformer." + "reason": "Both mention FLUX model loading, but one is a missing example URL/model_index.json and the other is a saved model that cannot be reloaded after FLUX+Hyper-SD LoRA; not the same fix." }, { "left": "issue:10052", "right": "issue:10596", "accept": false, - "reason": "Syntax error on ctx is unrelated to an invalid-shape runtime error in FluxControlPipeline." + "reason": "Syntax error on ctx vs invalid shape in FluxControlPipeline; completely different bugs." }, { "left": "issue:10120", "right": "issue:8950", "accept": false, - "reason": "Different SD3 training issues: memory consumption in dreambooth vs a LoRA training bug." + "reason": "Memory consumption in Dreambooth+SD3 vs SD3 LoRA training bug; not the same underlying issue." }, { "left": "issue:10202", "right": "issue:9309", "accept": false, - "reason": "FluxControl LoRA unload bug vs a FluxControlNetPipeline run failure on Flux 1 schnell; not the same concrete defect." + "reason": "Flux Control LoRA unloading state bug vs FluxControlNetPipeline failure on Flux schnell; related area but different concrete problems." }, { "left": "issue:10204", "right": "issue:6258", "accept": false, - "reason": "Both report quality degradation, but they are different pipelines with different causes and symptoms." + "reason": "Both describe quality degradation, but one is FluxFill inpainting over iterations and the other is Stable Video Diffusion forward chunking; different pipelines and causes." }, { "left": "issue:10490", "right": "issue:7802", "accept": false, - "reason": "Different scheduler feature requests: Infinity support vs OLSS Scheduler support." + "reason": "Infinity support request vs OLSS scheduler support request; separate feature requests." }, { "left": "issue:1061", "right": "issue:3622", "accept": false, - "reason": "Doc update request for Textual Inversion vs an import error in a tutorial; unrelated." + "reason": "Docs/tutorial update request vs an import error in a tutorial; one is editorial, the other is a runtime issue." }, { "left": "issue:10638", "right": "issue:9115", "accept": false, - "reason": "SDXL scheduler script problem vs AnimateDiff-Lightning/SDXL motion adapter error; different code paths." + "reason": "DPMSolverMultistepScheduler example failure vs AnimateDiff-Lightning/SDXL motion adapter error; different pipelines and bugs." }, { "left": "issue:10672", "right": "issue:7312", "accept": false, - "reason": "Callback-on-step-end support vs direct-reward finetuning support; both are feature requests but not the same change." + "reason": "callback_on_step_end support across pipelines vs AlignProp direct reward finetuning support; both feature requests, but not the same change." }, { "left": "issue:10672", "right": "issue:9374", "accept": false, - "reason": "General callback support request vs AutoPipeline support for Flux pipelines; distinct API additions." + "reason": "Both are API-support requests, but one is a callback hook and the other is AutoPipeline support for Flux pipelines; no shared concrete implementation target." }, { "left": "issue:10673", "right": "issue:12712", "accept": false, - "reason": "AuraFlow shape mismatch vs missing FLUX.2-dev-bnb-4bit model; unrelated." + "reason": "AuraFlow tensor reshape error vs missing FLUX.2-dev-bnb-4bit model; unrelated issues." }, { "left": "issue:1076", "right": "issue:1805", "accept": false, - "reason": "Stable Diffusion v1.5 question vs a request to add a scalable diffusion pipeline; not duplicates." + "reason": "Stable-diffusion-v1-5 question vs adding a scalable diffusion pipeline; not the same underlying request." }, { "left": "issue:10796", "right": "issue:11383", "accept": false, - "reason": "Docs for HunyuanVideo LoRA vs HiDream LoRA support; different models and requests." + "reason": "Both ask about LoRA docs/support, but for different models (HunyuanVideo vs HiDream); not duplicates." }, { "left": "issue:1088", "right": "issue:2444", "accept": false, - "reason": "Different dreambooth training scripts and different errors; no shared underlying bug." + "reason": "Both are Dreambooth training problems, but different scripts and errors (flax vs inpaint subprocess failure)." }, { "left": "issue:10895", "right": "issue:11939", "accept": false, - "reason": "General deprecation handling vs updating flux_controlnet training code; too broad to be the same change." - }, - { - "left": "issue:10932", - "right": "issue:9223", - "accept": false, - "reason": "HunyuanVideo compile/recompile issue vs a dependency version conflict; unrelated." - }, - { - "left": "issue:10932", - "right": "issue:9637", - "accept": false, - "reason": "Compilation/recompile-limit problem vs xFormers attention-mask BF16 failure; different stack and fix." - }, - { - "left": "issue:10963", - "right": "issue:11343", - "accept": false, - "reason": "Missing AutoencoderKLWan import vs broader diffusers import failures after a transformers update; not the same root cause." - }, - { - "left": "issue:11047", - "right": "issue:12195", - "accept": false, - "reason": "ControlNet PAG inpaint model support gap vs AutoencoderKL failing on torch.library.custom_op; unrelated." + "reason": "Deprecation cleanup request vs Flux ControlNet training code update; different scopes and fix paths." } ] }, @@ -98043,15 +98047,18 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:10932", + "issue:10963", "issue:11047", "issue:11055", "issue:1117", "issue:11220", + "issue:11343", "issue:11362", "issue:11418", "issue:11547", @@ -98060,6 +98067,7 @@ "issue:1197", "issue:1205", "issue:1214", + "issue:12195", "issue:12308", "issue:1249", "issue:1255", @@ -98071,13 +98079,8 @@ "issue:1358", "issue:1363", "issue:1375", - "issue:1482", - "issue:1486", - "issue:1535", - "issue:1536", "issue:1810", "issue:1869", - "issue:2532", "issue:305", "issue:3213", "issue:3300", @@ -98087,17 +98090,22 @@ "issue:4071", "issue:4646", "issue:4702", - "issue:4969", "issue:8317", "issue:8557", "issue:9188", + "issue:9223", "issue:9224", "issue:9239", "issue:9438", + "issue:9637", "issue:967", "issue:9958" ], "soft_pairs": [ + "issue:10932|issue:9223", + "issue:10932|issue:9637", + "issue:10963|issue:11343", + "issue:11047|issue:12195", "issue:11047|issue:9239", "issue:11055|issue:8557", "issue:1117|issue:338", @@ -98119,189 +98127,192 @@ "issue:1310|issue:1869", "issue:1358|issue:4071", "issue:1363|issue:3213", - "issue:1375|issue:305", - "issue:1482|issue:4969", - "issue:1486|issue:2532", - "issue:1535|issue:1536" + "issue:1375|issue:305" ], - "prepared_review_unit_hash": "af5657629091cf9a10b10b069c2f7d72514aaa9821c35ae57590519692830af8", + "prepared_review_unit_hash": "956b1c830d4bb02cbba2e5ee88622d46a2528b9845a8f50175fc0b6b2de4e265", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 33104, - "estimated_input_tokens": 8276, - "estimated_eval_tokens": 16808 + "soft_pair_count": 26, + "serialized_chars": 33590, + "estimated_input_tokens": 8398, + "estimated_eval_tokens": 17052 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: the soft links mostly reflect superficial keyword overlap (ControlNet, Flux, ONNX, pipelines, tests) rather than the same underlying bug or change. I would not merge any of the soft edges as duplicates.", - "confidence": 0.83, - "canonical_issue_reason": null, + "summary": "This cluster is a broad false positive: the items span unrelated bugs, docs, model-loading, and pipeline issues across different code paths. There is no single underlying defect or change to treat as a canonical duplicate.", + "confidence": 0.99, + "canonical_issue_reason": "No clear canonical issue: the cluster mixes unrelated topics (compilation limits, import errors, controlnet/inpaint, training artifacts, docs, GGUF shape mismatches, etc.) without a shared code path.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 1869 is the most discussion-heavy and broadly framed item, but it still does not represent the rest of this mixed cluster well enough to serve as a true canonical.", + "best_issue_reason": "No issue in this set is a good global representative, because any choice would misrepresent the rest of the cluster as the same bug.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:10932", + "right": "issue:9223", + "accept": false, + "reason": "Different problems: torch._dynamo recompilation in HunyuanVideo vs dependency version conflicts around pyarrow/datasets." + }, + { + "left": "issue:10932", + "right": "issue:9637", + "accept": false, + "reason": "Unrelated code paths: HunyuanVideo compile/recompile_limit vs XFormers attention-mask bfloat16 edge case." + }, + { + "left": "issue:10963", + "right": "issue:11343", + "accept": false, + "reason": "Both are import failures, but one is a missing symbol and the other is a transformers-version compatibility break; not the same bug." + }, + { + "left": "issue:11047", + "right": "issue:12195", + "accept": false, + "reason": "Different model/component and failure: missing ControlNet pag inpaint model vs AutoencoderKL custom_op on older PyTorch." + }, { "left": "issue:11047", "right": "issue:9239", "accept": false, - "reason": "Both mention ControlNet, but one is about a missing inpaint model in the pipeline and the other about black output with ControlNet-Canny; different bugs." + "reason": "ControlNet model availability issue is unrelated to black-image output from ControlNet-Canny." }, { "left": "issue:11055", "right": "issue:8557", "accept": false, - "reason": "Unconditional image generation colorization and SD3 training token limits are unrelated failure modes." + "reason": "Training colorization artifact for unconditional image generation is not the same as SD3 token-limit training behavior." }, { "left": "issue:1117", "right": "issue:338", "accept": false, - "reason": "Multiple negative prompts tests and LMS Discrete tests cover different features." + "reason": "Both mention negatives/schedulers in tests, but one is multiple negative prompts and the other is LMS Discrete tests; different scopes." }, { "left": "issue:1117", "right": "issue:339", "accept": false, - "reason": "Different scheduler coverage: negative prompts vs Karras VE." + "reason": "Different scheduler test targets: negative prompt behavior vs Karras VE tests." }, { "left": "issue:11220", "right": "issue:9438", "accept": false, - "reason": "Both are docs issues, but they concern different pages and different broken links/behavior." + "reason": "Docs page not working is not the same as a quicktour page 404 caused by model_name in the URL." }, { "left": "issue:11362", "right": "issue:8317", "accept": false, - "reason": "Different model APIs: return_dict handling on HiDreamImageTransformer2DModel vs missing generator in AsymmetricAutoencoderKL decode." + "reason": "Both involve Autoencoder-style models, but one is an unexpected keyword in HiDreamImageTransformer2DModel and the other is a missing generator arg in AsymmetricAutoencoderKL." }, { "left": "issue:11418", "right": "issue:13484", "accept": false, - "reason": "Adding a Flux fp8 weight file is not the same as failing LoRA loading on Flux 2 dev." + "reason": "Adding a Flux fill model file is not the same as LoRA loading failures for Flux 2 dev FAL LoRAs." }, { "left": "issue:11547", "right": "issue:9188", "accept": false, - "reason": "Different Flux features: controlnet support vs outpainting." + "reason": "ControlNet support request for Flux fill dev is different from Flux outpainting feature request." }, { "left": "issue:11830", "right": "issue:9224", "accept": false, - "reason": "Both involve dtype issues, but one is black output in Framepack with fp16 and the other is MPS incompatibility with hardcoded float64." + "reason": "Black output in framepack with float16 is unrelated to hardcoded float64 on MPS." }, { "left": "issue:11839", "right": "issue:13001", "accept": false, - "reason": "Both are GGUF loading problems, but they target different Flux models/pipelines and do not clearly describe the same concrete bug." + "reason": "Both are GGUF loading issues, but they describe different pipelines/models and different errors (invalid input size vs shape mismatch)." }, { "left": "issue:1197", "right": "issue:1290", "accept": false, - "reason": "Two different community pipeline proposals with no shared code-path issue." + "reason": "Two community pipeline proposals with different concepts: OpenAI Diffusion vs DreamArtist." }, { "left": "issue:1197", "right": "issue:3724", "accept": false, - "reason": "OpenAI diffusion pipeline and watermarking module are unrelated feature ideas." + "reason": "OpenAI Diffusion pipeline proposal is unrelated to a watermarking module discussion." }, { "left": "issue:1205", "right": "issue:4646", "accept": false, - "reason": "ONNX conversion is the only overlap; Stable Diffusion conversion and ControlNet conversion are distinct problems." + "reason": "ONNX conversion of SD models is unrelated to controlnet-to-onnx conversion failure." }, { "left": "issue:1214", "right": "issue:967", "accept": false, - "reason": "Half-precision CLIPTextModel failure and ONNX scheduler incompatibility are different bugs." + "reason": "Half-precision CLIPTextModel issue is not the same as scheduler incompatibility in OnnxStableDiffusionPipeline." }, { "left": "issue:12308", "right": "issue:9958", "accept": false, - "reason": "Flash-attn support on non-Hopper GPUs and AutoencoderKLTemporalDecoder tiling are unrelated." + "reason": "flash-attn support for non-Hopper GPUs is unrelated to AutoencoderKLTemporalDecoder tiling support." }, { "left": "issue:1249", "right": "issue:1810", "accept": false, - "reason": "A runtime device attribute error in ONNX pipelines is not the same as NumPy deprecated types." + "reason": "OnnxPipeline numpy/device error is not the same as deprecated numpy type usage." }, { "left": "issue:1255", "right": "issue:3300", "accept": false, - "reason": "Importing one scheduler failing is not the same as a broad claim that Diffusers on HuggingFace are broken." + "reason": "Import error for one scheduler symbol is unrelated to a broad report that HuggingFace-hosted diffusers are broken." }, { "left": "issue:1304", "right": "issue:4702", "accept": false, - "reason": "Training text-guided image-to-image diffusion and training a VQ-VAE are different tasks." + "reason": "Training text-guided image-to-image diffusion is different from an example script request for training a VQ-VAE." }, { "left": "issue:1310", "right": "issue:1869", "accept": false, - "reason": "Both are community-oriented pipeline ideas, but K-Diffusion + Stable Diffusion and UnCLIP interpolations are distinct changes." + "reason": "K-Diffusion + Stable Diffusion pipeline is not the same underlying change as UnCLIP image/text interpolations." }, { "left": "issue:1358", "right": "issue:4071", "accept": false, - "reason": "Safetensors support and inpaint-only+lama support are unrelated feature requests." + "reason": "Safetensors support is unrelated to inpaint-only + LaMa support." }, { "left": "issue:1363", "right": "issue:3213", "accept": false, - "reason": "Both involve lpw_stable_diffusion, but one is an ONNX/LMS scheduler error and the other is a broader pipeline breakage; not clearly the same bug." + "reason": "ONNX long-prompt/LMS discrete scheduler error is unrelated to lpw_stable_diffusion being broken." }, { "left": "issue:1375", "right": "issue:305", "accept": false, - "reason": "Both are import errors, but one is explicitly caused by PR #1365 and the other is a separate generic ImportError report." - }, - { - "left": "issue:1482", - "right": "issue:4969", - "accept": false, - "reason": "Stable Diffusion upscale pipeline and CrossAttention import failure are unrelated." - }, - { - "left": "issue:1486", - "right": "issue:2532", - "accept": false, - "reason": "DDIMScheduler prediction_type='sample' and StableDiffusionPanoramaPipeline get_views are different code paths." - }, - { - "left": "issue:1535", - "right": "issue:1536", - "accept": false, - "reason": "Image Variation Pipeline and lpw_stable_diffusion compatibility are separate pipeline issues." + "reason": "A PR breaking imports may resemble import issues generally, but these refer to different failures and likely different fixes." } ] }, @@ -98320,11 +98331,15 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:1482", + "issue:1486", + "issue:1535", + "issue:1536", "issue:1546", "issue:1566", "issue:1594", @@ -98348,12 +98363,9 @@ "issue:2375", "issue:2390", "issue:2416", - "issue:2462", - "issue:2507", + "issue:2532", "issue:2616", "issue:2618", - "issue:2663", - "issue:272", "issue:2775", "issue:2800", "issue:2908", @@ -98364,9 +98376,8 @@ "issue:3674", "issue:3694", "issue:3817", - "issue:4400", - "issue:459", "issue:463", + "issue:4969", "issue:543", "issue:5462", "issue:5824", @@ -98375,6 +98386,9 @@ "issue:861" ], "soft_pairs": [ + "issue:1482|issue:4969", + "issue:1486|issue:2532", + "issue:1535|issue:1536", "issue:1546|issue:543", "issue:1566|issue:3279", "issue:1594|issue:3444", @@ -98396,189 +98410,186 @@ "issue:2359|issue:5462", "issue:2375|issue:3694", "issue:2390|issue:3087", - "issue:2416|issue:2618", - "issue:2462|issue:459", - "issue:2507|issue:2663", - "issue:272|issue:4400" + "issue:2416|issue:2618" ], - "prepared_review_unit_hash": "f2f8a460df9e31f4f088173277a42717fee1ff0ff104aea6c5f641202eaec58a", + "prepared_review_unit_hash": "d6521d9c1c077224b5caa8a6f6d634d498d10b0578ba5331ab174554837dd125", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 33651, - "estimated_input_tokens": 8413, - "estimated_eval_tokens": 17082 + "serialized_chars": 33722, + "estimated_input_tokens": 8431, + "estimated_eval_tokens": 17118 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate cluster emerges here: the soft pairs are mostly broad topical matches within Stable Diffusion, Dreambooth, conversion, or training scripts, but they describe different bugs, features, or error modes. All soft edges should be rejected.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "These items are a loose collection of unrelated Diffusers issues spanning different pipelines, bugs, and feature requests. The soft pairs mostly share only superficial subsystem overlap, not the same concrete problem, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 3087 is the broadest, highest-activity request in the set and is the best representative if one issue must be chosen, but the cluster itself does not show a true duplicate theme.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 3087 has the strongest discussion signal and a broad community-facing feature scope compared with the other mostly narrow bug reports and questions.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:1482", + "right": "issue:4969", + "accept": false, + "reason": "Different topics: a Stable Diffusion upscale pipeline request vs an import error for CrossAttention." + }, + { + "left": "issue:1486", + "right": "issue:2532", + "accept": false, + "reason": "DDIMScheduler prediction_type bug and panorama get_views bug are unrelated code paths." + }, + { + "left": "issue:1535", + "right": "issue:1536", + "accept": false, + "reason": "General image variation pipeline request vs lpw_stable_diffusion compatibility break." + }, { "left": "issue:1546", "right": "issue:543", "accept": false, - "reason": "Different failures: truncation/batched-input error vs DDP stride warning." + "reason": "Batched truncation warning error is not the same as DDP grad-stride warning." }, { "left": "issue:1566", "right": "issue:3279", "accept": false, - "reason": "Both Dreambooth-related, but one is fp16 checkpoint-save crash and the other is checkpoint-resume errors." + "reason": "Both involve Dreambooth checkpoints, but one crashes after saving fp16 output and the other fails on resume from checkpoint." }, { "left": "issue:1594", "right": "issue:3444", "accept": false, - "reason": "Img2img/video generation request is unrelated to audio-to-image integration." + "reason": "Video img2img request and audio-to-image integration are different generative tasks." }, { "left": "issue:1594", "right": "issue:3817", "accept": false, - "reason": "Different feature requests: video img2img vs dance-generation editing." + "reason": "Video generation from img2img and music-to-dance generation are distinct features." }, { "left": "issue:1604", "right": "issue:782", "accept": false, - "reason": "Different bugs: lpw_stable_diffusion breakage after diffusers upgrade vs model download/auth-token regression." + "reason": "LPW pipeline break in diffusers 0.9.0 is unrelated to the model-download auth_token regression." }, { "left": "issue:1647", "right": "issue:1704", "accept": false, - "reason": "Unrelated questions; one about color-correction weights, the other about attention blocks." + "reason": "A color-correction usage question does not match an attention block question." }, { "left": "issue:1692", "right": "issue:2616", "accept": false, - "reason": "Different runtime errors in different scripts: pickle/lambda vs NoneType replace on save_attn_procs." + "reason": "Pickling a lambda and NoneType.replace during save_attn_procs are different failures." }, { "left": "issue:1704", "right": "issue:199", "accept": false, - "reason": "A general attention question is not the same as deleting dead code." + "reason": "A question about attention blocks is not a dead-code cleanup issue." }, { "left": "issue:1851", "right": "issue:2371", "accept": false, - "reason": "Performance regression on multi-GPU vs missing Dreambooth outputs are different problems." + "reason": "Multi-GPU slowdown in Dreambooth training is unrelated to missing objects in Dreambooth output." }, { "left": "issue:1876", "right": "issue:861", "accept": false, - "reason": "Both involve conversion, but one is missing state_dict while the other is a generic ckpt-to-diffusers conversion failure." + "reason": "Both mention conversion, but one is about missing state_dict in checkpoints and the other about a ckpt-to-diffusers conversion failure." }, { "left": "issue:1909", "right": "issue:3162", "accept": false, - "reason": "Scheduler generator API mismatch is unrelated to missing learned_classifier_free_sampling_embeddings." + "reason": "Generator list handling in pipelines and missing learned_classifier_free_sampling_embeddings are unrelated API issues." }, { "left": "issue:1962", "right": "issue:2800", "accept": false, - "reason": "Video generation request and DiffEdit semantic editing are different features." + "reason": "General video generation request and DiffEdit semantic editing are different use cases." }, { "left": "issue:1969", "right": "issue:2775", "accept": false, - "reason": "Distinct runtime errors: tensor view/stride issue vs torch._dynamo on non-function object." + "reason": "Tensor view stride error and torch._dynamo.optimize misuse are separate runtime errors." }, { "left": "issue:2068", "right": "issue:834", "accept": false, - "reason": "Model dtype conversion bug is unrelated to CPU under-utilization." + "reason": "Model dtype conversion bug is not the same as CPU under-utilization." }, { "left": "issue:2163", "right": "issue:2908", "accept": false, - "reason": "Different example scripts and failures: text-to-image dtype mismatch vs controlnet shape/mixed-precision errors." + "reason": "Both are training-script errors, but they affect different scripts and different mixed-precision/shaping failures." }, { "left": "issue:2192", "right": "issue:3674", "accept": false, - "reason": "Missing CLI arguments and metadata.jsonl parsing are separate issues." + "reason": "Missing CLI args in dreambooth_lora and metadata.jsonl parsing in text_to_image_lora are separate issues." }, { "left": "issue:2225", "right": "issue:463", "accept": false, - "reason": "Safety-checker usage question is unrelated to super-resolution diffusion support." + "reason": "Safety checker usage outside a pipeline is not the same as a super-resolution model request." }, { "left": "issue:2319", "right": "issue:5824", "accept": false, - "reason": "Different methods and bugs: typo in original-ckpt loader vs SDXL add_time_ids default mismatch." + "reason": "A typo in an original-stable-diffusion loader method is unrelated to SDXL time-id default behavior." }, { "left": "issue:2359", "right": "issue:5462", "accept": false, - "reason": "Both mention img2img/latent output, but they target different pipelines and change requests." + "reason": "Allowing latent output_type in img2img is different from adding img2img support for LCMs." }, { "left": "issue:2375", "right": "issue:3694", "accept": false, - "reason": "Safe_open import error is unrelated to missing model_index/config during training." + "reason": "safe_open NameError and missing model_index/config.json are unrelated file-loading failures." }, { "left": "issue:2390", "right": "issue:3087", "accept": false, - "reason": "Separate feature requests for different model types: T2I-Adapter vs LyCORIS." + "reason": "T2I-Adapter support and LyCORIS support are different model-family feature requests." }, { "left": "issue:2416", "right": "issue:2618", "accept": false, - "reason": "Performance slowdown with xformers is unrelated to ControlNet float-tensor image input support." - }, - { - "left": "issue:2462", - "right": "issue:459", - "accept": false, - "reason": "IterableDataset support for finetuning is unrelated to incremental diffusion." - }, - { - "left": "issue:2507", - "right": "issue:2663", - "accept": false, - "reason": "GPU loading slowdown is not the same as model load failure when safetensors are present." - }, - { - "left": "issue:272", - "right": "issue:4400", - "accept": false, - "reason": "K-LMS inpainting support is unrelated to text_encoder support in Dreambooth LoRA." + "reason": "xformers training slowdown and ControlNet FloatTensor image input are unrelated problems." } ] }, @@ -98597,11 +98608,15 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:2462", + "issue:2507", + "issue:2663", + "issue:272", "issue:2755", "issue:2763", "issue:2785", @@ -98626,31 +98641,30 @@ "issue:3461", "issue:3481", "issue:3551", - "issue:3566", "issue:3574", - "issue:3580", - "issue:3624", "issue:3774", "issue:3817", "issue:4071", "issue:4103", + "issue:4400", "issue:4450", "issue:4476", + "issue:459", "issue:4949", "issue:5002", - "issue:5075", "issue:5112", "issue:5128", - "issue:557", "issue:5592", "issue:5989", "issue:6218", "issue:6560", "issue:6732", - "issue:7183", "issue:938" ], "soft_pairs": [ + "issue:2462|issue:459", + "issue:2507|issue:2663", + "issue:272|issue:4400", "issue:2755|issue:5989", "issue:2763|issue:5112", "issue:2785|issue:4071", @@ -98671,183 +98685,180 @@ "issue:3411|issue:4450", "issue:3444|issue:3817", "issue:3461|issue:6560", - "issue:3551|issue:3774", - "issue:3566|issue:557", - "issue:3580|issue:5075", - "issue:3624|issue:7183" + "issue:3551|issue:3774" ], - "prepared_review_unit_hash": "7b64416cdfdb3c7beff3bc4ba1e1da742fd2c7aee9f82e54f2571da4788076ee", + "prepared_review_unit_hash": "3fe7ab1870604fdb75593c2450b9f1a175a8777dcd43013222b87dbc46ae5398", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 32950, - "estimated_input_tokens": 8238, - "estimated_eval_tokens": 16732 + "serialized_chars": 32811, + "estimated_input_tokens": 8203, + "estimated_eval_tokens": 16662 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of mostly unrelated diffusers issues spanning Stable Diffusion loading, ControlNet, schedulers, LoRA, offload, and assorted feature requests. I do not see any true duplicate pair among the soft candidates.", - "confidence": 0.94, - "canonical_issue_reason": "issue:2755 is the most concrete, narrowly scoped bug report in the set, but the cluster is too heterogeneous for a strong canonical duplicate.", + "summary": "This cluster is not a duplicate set: it mixes many unrelated Stable Diffusion/diffusers bugs, feature requests, and API questions across loading, schedulers, ControlNet, LoRA, pipelines, and docs. None of the soft pairs look like the same underlying change or defect.", + "confidence": 0.93, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:2755 is a good representative only in the sense that it is a clear, actionable code-path bug; it is not a good cluster-wide duplicate anchor because the rest of the items are unrelated.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:2462", + "right": "issue:459", + "accept": false, + "reason": "Different topics: iterable dataset fine-tuning support vs a general diffusion model concept request." + }, + { + "left": "issue:2507", + "right": "issue:2663", + "accept": false, + "reason": "Both mention safetensors, but one is a performance issue on GPU loading and the other is a model-loading failure for a specific safety checker file." + }, + { + "left": "issue:272", + "right": "issue:4400", + "accept": false, + "reason": "Inpainting K-LMS scheduler support is unrelated to DreamBooth inpaint LoRA text-encoder support." + }, { "left": "issue:2755", "right": "issue:5989", "accept": false, - "reason": "Different bugs: StableDiffusion ckpt float16 loading vs a generic safetensors.torch bug." + "reason": "One is a Stable Diffusion float16 loading bug; the other is a generic safetensors.torch bug with no clear same code path." }, { "left": "issue:2763", "right": "issue:5112", "accept": false, - "reason": "Both mention memory/model behavior, but one is ControlNet VRAM with cudnn.benchmark and the other is an SDXL ControlNet img2img float-size error." + "reason": "ControlNet VRAM/cudnn behavior is a different problem from an SDXL ControlNet dtype/shape error." }, { "left": "issue:2785", "right": "issue:4071", "accept": false, - "reason": "M1/M2 text-to-video support request is unrelated to inpaint-only+lama support." + "reason": "Text-to-video M1/M2 support and inpaint-only lama support are separate feature requests." }, { "left": "issue:2800", "right": "issue:938", "accept": false, - "reason": "DiffEdit semantic image editing is unrelated to human motion diffusion." + "reason": "DiffEdit semantic image editing and human motion diffusion are unrelated features." }, { "left": "issue:2871", "right": "issue:6732", "accept": false, - "reason": "Inpainting latent output handling is unrelated to scheduler documentation." + "reason": "Stable Diffusion inpainting latent output handling is unrelated to scheduler documentation." }, { "left": "issue:2883", "right": "issue:4949", "accept": false, - "reason": "Image-to-video pipeline and ProPainter video in/outpainting are different features." + "reason": "Image-to-video and ProPainter video in/outpainting are different video-generation problems." }, { "left": "issue:2907", "right": "issue:5592", "accept": false, - "reason": "CPU offload called twice causing pipeline failures is unrelated to UniDiffuser checkpoint loading." + "reason": "Model CPU offload being called twice is unrelated to checkpoint loading for UniDiffuser." }, { "left": "issue:3029", "right": "issue:4476", "accept": false, - "reason": "ControlNet custom resolution question is unrelated to SDXL+ControlNet fill50k output mismatch." + "reason": "ControlNet custom resolution question is not the same as an SDXL ControlNet fill50k result mismatch." }, { "left": "issue:3085", "right": "issue:5002", "accept": false, - "reason": "Cache-folder argument removal and missing scheduler argument are different API breakages." + "reason": "Cache-folder argument removal and missing scheduler argument are unrelated API breakages." }, { "left": "issue:3116", "right": "issue:3481", "accept": false, - "reason": "KDPM2DiscreteScheduler latent-noise bug in img2img is unrelated to DDIM step-dependent image differences." + "reason": "KDPM2DiscreteScheduler image2image latent noise bug is not the same as DDIM step-dependent image dissimilarity." }, { "left": "issue:3139", "right": "issue:3283", "accept": false, - "reason": "Using safetensors for image generation and converting LoRA safetensors to .bin are not the same issue." + "reason": "How to use safetensors for generation vs how to convert LoRA safetensors to .bin are different usage questions." }, { "left": "issue:3142", "right": "issue:3223", "accept": false, - "reason": "Safety checker placement is unrelated to RAM not being cleared after moving pipeline to cuda." + "reason": "Safety checker placement and RAM not clearing after moving pipeline to CUDA are separate memory/inference issues." }, { "left": "issue:3182", "right": "issue:6218", "accept": false, - "reason": "Both involve pipeline behavior, but one is custom_pipeline CPU offload and the other is batched seeds in ControlNet/Img2Img." + "reason": "LPW Stable Diffusion CPU offload incompatibility is not the same as batched seeds failing in ControlNet pipelines." }, { "left": "issue:3251", "right": "issue:6732", "accept": false, - "reason": "ControlNet Shuffle CFG-side behavior is unrelated to scheduler docs." + "reason": "ControlNet Shuffle behavior is unrelated to scheduler documentation." }, { "left": "issue:3366", "right": "issue:3574", "accept": false, - "reason": "LoRALinearLayer autocast regression and from_ckpt Windows problems are different bugs." + "reason": "LoRALinearLayer autocast regression and Windows from_ckpt loading issues are different bugs." }, { "left": "issue:3403", "right": "issue:5128", "accept": false, - "reason": "Pipeline loading with TorchDynamo is not the same as failing to load a local .safetensors via XLPipeline.from_single_file." + "reason": "TorchDynamo pipeline loading and local safetensors loading failure are unrelated loading paths." }, { "left": "issue:3406", "right": "issue:4103", "accept": false, - "reason": "Attention only_cross_attention bug is unrelated to AutoencoderKL scaling_factor handling." + "reason": "Attention class only_cross_attention bug and AutoencoderKL scaling_factor handling are separate model components." }, { "left": "issue:3411", "right": "issue:4450", "accept": false, - "reason": "Training test coverage for DreamBooth LoRA is unrelated to multi-text inversion not working." + "reason": "DreamBooth LoRA training test coverage and multi-text inversion failure are different training/example issues." }, { "left": "issue:3444", "right": "issue:3817", "accept": false, - "reason": "ImageBind + StableUnCLIP audio2image integration and editable dance generation from music are different feature requests." + "reason": "ImageBind/StableUnCLIP audio-to-image integration and editable dance generation from music are distinct projects." }, { "left": "issue:3461", "right": "issue:6560", "accept": false, - "reason": "Dreambooth performance/vRAM complaint is unrelated to LoRA shape-size regression." + "reason": "DreamBooth script performance/memory comparison is unrelated to LoRA shape-size regression." }, { "left": "issue:3551", "right": "issue:3774", "accept": false, - "reason": "LoRA deactivating xFormers and zero3 GPU-memory savings are separate training/runtime issues." - }, - { - "left": "issue:3566", - "right": "issue:557", - "accept": false, - "reason": "controlnet_reference unexpected results do not match generic DDIM quality complaints." - }, - { - "left": "issue:3580", - "right": "issue:5075", - "accept": false, - "reason": "XFormers TypeError and SDXL refiner prompt-embedding matmul shape mismatch are different failure modes." - }, - { - "left": "issue:3624", - "right": "issue:7183", - "accept": false, - "reason": "getattr TypeError is unrelated to scheduler _class_name reporting the wrong class." + "reason": "LoRA disabling xFormers and zero3 not saving GPU memory are different optimization/runtime problems." } ] }, @@ -98866,11 +98877,14 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:3566", + "issue:3580", + "issue:3624", "issue:3658", "issue:3677", "issue:3688", @@ -98893,12 +98907,9 @@ "issue:4349", "issue:4353", "issue:4354", - "issue:4363", "issue:4436", - "issue:4445", - "issue:4453", "issue:4456", - "issue:4468", + "issue:5075", "issue:5079", "issue:5215", "issue:5281", @@ -98907,19 +98918,22 @@ "issue:5408", "issue:5410", "issue:5476", - "issue:5956", + "issue:557", "issue:6529", "issue:6685", "issue:7061", + "issue:7183", "issue:7392", "issue:7418", - "issue:7454", "issue:7509", "issue:8320", "issue:8440", "issue:8637" ], "soft_pairs": [ + "issue:3566|issue:557", + "issue:3580|issue:5075", + "issue:3624|issue:7183", "issue:3658|issue:5403", "issue:3677|issue:3688", "issue:3677|issue:5476", @@ -98941,189 +98955,186 @@ "issue:4241|issue:7392", "issue:4349|issue:5215", "issue:4353|issue:5410", - "issue:4354|issue:6685", - "issue:4363|issue:4453", - "issue:4445|issue:7454", - "issue:4468|issue:5956" + "issue:4354|issue:6685" ], - "prepared_review_unit_hash": "65dbcc0d4aab70490b08f526d38821671a87d7035bd39cddd10cf0cb5c9b6e1b", + "prepared_review_unit_hash": "99e2951d7779c2e22f044bca4f66adcf6ff86d43094a8304e3ca28c085471ba8", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 33201, - "estimated_input_tokens": 8301, - "estimated_eval_tokens": 16858 + "serialized_chars": 33304, + "estimated_input_tokens": 8326, + "estimated_eval_tokens": 16908 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated issues about LoRA, SDXL, pipelines, docs, training scripts, and import/runtime errors. None of the soft pairs appear to be the same underlying bug or change, so there is no good duplicate canonical here.", - "confidence": 0.97, + "summary": "No true duplicate pairs stand out here: the items span unrelated bugs, feature requests, and docs issues across different pipelines and code paths. All soft edges look like superficial title similarity only.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:3566", + "right": "issue:557", + "accept": false, + "reason": "Different topics: a community controlnet_reference pipeline issue vs a DDIM quality comparison; no shared bug path." + }, + { + "left": "issue:3580", + "right": "issue:5075", + "accept": false, + "reason": "Both mention inference/runtime errors, but one is an xFormers TypeError and the other is an SDXL refiner shape mismatch; distinct failures." + }, + { + "left": "issue:3624", + "right": "issue:7183", + "accept": false, + "reason": "Unrelated errors: getattr TypeError vs scheduler class-name metadata reporting." + }, { "left": "issue:3658", "right": "issue:5403", "accept": false, - "reason": "Different problems: DreamBooth upsampler noise-latent bug vs T2I adapter training script bug." + "reason": "Dreambooth upsampler training bug vs T2I adapter training script question; different scripts and symptoms." }, { "left": "issue:3677", "right": "issue:3688", "accept": false, - "reason": "General Kandinsky feature request vs a specific LoRA/SPDA optimization issue." + "reason": "General Kandinsky feature request vs a specific LoRA/PT 2.0 optimization issue; not the same change." }, { "left": "issue:3677", "right": "issue:5476", "accept": false, - "reason": "Kandinsky support request is unrelated to xformers-based DreamBooth weight loading." + "reason": "Kandinsky support request vs a DreamBooth/xformers compatibility question; unrelated scope." }, { "left": "issue:3693", "right": "issue:7061", "accept": false, - "reason": "Different doc issues: Kandinsky pipeline docstrings vs imagic community pipeline inference docstring." + "reason": "Both are doc-related, but one is Kandinsky docstrings and the other is imagic pipeline inference docs; different files and problems." }, { "left": "issue:3750", "right": "issue:5408", "accept": false, - "reason": "KDiffusion LoRA loading support vs a bad-inference report for kohya-style LoRA." + "reason": "KDiffusion LoRA-loading bug vs bad inference with kohya-style LoRA; separate issues in different paths." }, { "left": "issue:3788", "right": "issue:7509", "accept": false, - "reason": "UniControl new-pipeline request is unrelated to ReNoise inversion pipeline work." + "reason": "New pipeline request (UniControl) vs ReNoise pipeline proposal; clearly different additions." }, { "left": "issue:3809", "right": "issue:7418", "accept": false, - "reason": "DDP stride-mismatch warning vs a bug in a specific community upscale LDM3D script." + "reason": "DDP stride-mismatch warning vs a specific bug in an ldm3d community script; not the same issue." }, { "left": "issue:3816", "right": "issue:4456", "accept": false, - "reason": "Transformers CLIPImageProcessor import error is unrelated to a ModuleList indexing error in LoRA loading." + "reason": "Different import/load failures: missing CLIPImageProcessor vs ModuleList attribute error in LoRA loading." }, { "left": "issue:3946", "right": "issue:8637", "accept": false, - "reason": "`from_ckpt` naming complaint vs PixArtSigmaPipeline import failure; different API issues." + "reason": "Both involve naming/import mechanics, but one is about from_ckpt naming and the other an import failure for PixArtSigmaPipeline; unrelated." }, { "left": "issue:3967", "right": "issue:8320", "accept": false, - "reason": "SDXL branch-deletion breakage vs `from_single_file` AutoencoderKL `__name__` attribute error." + "reason": "SDXL branch/deletion regression vs from_single_file __name__ error; distinct regressions." }, { "left": "issue:3986", "right": "issue:5079", "accept": false, - "reason": "ShapE mesh-output feature request vs LyCORIS loading support." + "reason": "Support for mesh output in ShapEPipeline vs loading LyCORIS weights; different features." }, { "left": "issue:4001", "right": "issue:5320", "accept": false, - "reason": "README demo link/docs request vs PEFT `fuse_lora` adapter_name API change." + "reason": "README/demo docs request vs PEFT API change for fuse_lora; not the same artifact or bug." }, { "left": "issue:4043", "right": "issue:5281", "accept": false, - "reason": "SDXL pooled-embed requirement vs multi-pipeline CPU offload device mismatch." + "reason": "SDXL prompt/pool embedding requirement vs multi-pipeline CPU offload device mismatch; separate SDXL bugs." }, { "left": "issue:4069", "right": "issue:4354", "accept": false, - "reason": "Textual inversion unload behavior is unrelated to SDXL inpaint type returned by from_single_file." + "reason": "Textual inversion unload behavior vs from_single_file returning wrong pipeline type; different code paths." }, { "left": "issue:4071", "right": "issue:6529", "accept": false, - "reason": "Inpaint-only+lama support request vs SDXL Fooocus Inpaint feature request." + "reason": "Inpaint-only+lama support request vs SDXL Fooocus Inpaint discussion; one is a feature request, the other a pipeline-specific topic." }, { "left": "issue:4171", "right": "issue:4436", "accept": false, - "reason": "Docs home-page rendering bug vs CIVITAI safetensors/VAE conversion issue." + "reason": "Docs rendering issue on the reference page vs CIVITAI safetensors conversion; unrelated." }, { "left": "issue:4194", "right": "issue:8440", "accept": false, - "reason": "Single-file VAE loading issue is not the same as FromOriginalModelMixin subclass ValueError." + "reason": "from_single_file VAE loading issue vs FromOriginalModelMixin subclass ValueError; different mechanics." }, { "left": "issue:4222", "right": "issue:7392", "accept": false, - "reason": "Low-resolution LDM training advice vs SDXL multi-aspect-ratio training question." + "reason": "General LDM training on low-resolution data vs SDXL multi-aspect-ratio training question; not a duplicate bug report." }, { "left": "issue:4241", "right": "issue:7392", "accept": false, - "reason": "LoRA fine-tuning support question vs SDXL multi-aspect-ratio training question; only loosely related training topics." + "reason": "LoRA fine-tuning support for unidiffusers vs SDXL aspect-ratio training; different requests." }, { "left": "issue:4349", "right": "issue:5215", "accept": false, - "reason": "UpscalePipeline missing text_encoder despite prompt_embeds vs dark-image brightness shift in upscale pipeline." + "reason": "Upscale pipeline missing text_encoder despite prompt_embeds vs brightness shift on dark images; separate failures." }, { "left": "issue:4353", "right": "issue:5410", "accept": false, - "reason": "Generic SDXL quality complaint vs blurry latent-upscale behavior; not the same concrete bug." + "reason": "General SDXL poor-quality report vs latent upscale blur; too broad and not the same concrete bug." }, { "left": "issue:4354", "right": "issue:6685", "accept": false, - "reason": "from_single_file pipeline-type issue vs advanced DreamBooth absolute output_dir path handling." - }, - { - "left": "issue:4363", - "right": "issue:4453", - "accept": false, - "reason": "How to load an SDXL base file vs converting SDXL LoRA to AUTO1111 safetensors; different tasks." - }, - { - "left": "issue:4445", - "right": "issue:7454", - "accept": false, - "reason": "How to finetune a LoRA model vs generic LoRA support request; too broad and not a duplicate." - }, - { - "left": "issue:4468", - "right": "issue:5956", - "accept": false, - "reason": "Multi-GPU sample performance discrepancy vs SD-XL U-Net fine-tuning degradation; different symptoms and likely causes." + "reason": "Wrong pipeline type from_single_file vs DreamBooth absolute output_dir paths; unrelated bugs." } ] }, @@ -99142,11 +99153,15 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:4363", + "issue:4445", + "issue:4453", + "issue:4468", "issue:4504", "issue:4593", "issue:4634", @@ -99174,28 +99189,27 @@ "issue:5777", "issue:5783", "issue:5791", - "issue:5805", - "issue:5905", - "issue:5925", - "issue:5947", + "issue:5956", "issue:5972", - "issue:6079", "issue:6388", "issue:6412", "issue:6496", "issue:6791", "issue:7177", "issue:7405", + "issue:7454", "issue:770", "issue:8537", "issue:8742", "issue:8771", "issue:8803", "issue:8831", - "issue:9101", - "issue:9511" + "issue:9101" ], "soft_pairs": [ + "issue:4363|issue:4453", + "issue:4445|issue:7454", + "issue:4468|issue:5956", "issue:4504|issue:5328", "issue:4593|issue:8831", "issue:4634|issue:4638", @@ -99216,183 +99230,180 @@ "issue:5429|issue:6496", "issue:5760|issue:7405", "issue:5783|issue:8803", - "issue:5791|issue:5972", - "issue:5805|issue:6079", - "issue:5905|issue:5947", - "issue:5925|issue:9511" + "issue:5791|issue:5972" ], - "prepared_review_unit_hash": "f8e4aa4c75a8ae3f74281b2aa5f75058c6a1b4dbe1d0213a8fef1396b535aed6", + "prepared_review_unit_hash": "d844052030b6421b647b23f64565c2fb872ce65e2fbd33a067a374a0d92dbda7", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 32782, - "estimated_input_tokens": 8196, - "estimated_eval_tokens": 16648 + "serialized_chars": 32787, + "estimated_input_tokens": 8197, + "estimated_eval_tokens": 16650 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are mostly unrelated; the only commonality is superficial overlap around LoRA/SDXL/training terminology. None of the soft pairs look like true duplicates or the same concrete change.", + "summary": "This cluster is mostly a loose collection of SDXL/LoRA/training/loading issues, not true duplicates. The strongest single issue is the concrete LoRA safetensor loading bug, but the rest are different bugs or feature requests in nearby areas.", "confidence": 0.86, - "canonical_issue_reason": "No single canonical issue fits the whole set: the items span distinct bugs, feature requests, and model-specific failures. If forced to pick a representative, the closest broad theme is LoRA loading, but that still only covers a subset.", + "canonical_issue_reason": "issue:4504 is the most concrete, reproducible bug report in the set: a specific LoRA safetensor load path failing with a Python exception and enough detail to diagnose.", "canonical_pr_reason": null, - "best_issue_reason": "No strong global best issue for the full set. The closest representative is issue:4641 for LoRA loading, but it is not a valid canonical for the entire heterogeneous batch.", + "best_issue_reason": "issue:4504 is the best representative issue overall because it is a precise code-path failure rather than a broad question or feature request.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:4363", + "right": "issue:4453", + "accept": false, + "reason": "Both mention SDXL and safetensors, but one is about loading a checkpoint and the other about converting LoRA weights for A1111; different tasks and failure modes." + }, + { + "left": "issue:4445", + "right": "issue:7454", + "accept": false, + "reason": "A how-to question about finetuning LoRA is not the same as a broad LoRA support feature request." + }, + { + "left": "issue:4468", + "right": "issue:5956", + "accept": false, + "reason": "Both involve performance, but one is multi-GPU sample-code performance and the other is degraded SD-XL U-Net fine-tuning quality/performance; not the same bug." + }, { "left": "issue:4504", "right": "issue:5328", "accept": false, - "reason": "Both mention LoRA failures, but one is a specific safetensor state_dict error and the other is a broad Civitai regression report; not the same bug." + "reason": "Both are LoRA loading complaints, but 4504 is a specific state_dict exception while 5328 is a broad regression report about CivitAI LoRAs failing; too vague to call the same underlying bug." }, { "left": "issue:4593", "right": "issue:8831", "accept": false, - "reason": "Different problems: rank/shape mismatch when loading Automatic1111 LoRAs vs unsupported-layer continuation work." + "reason": "Both concern LoRA layer compatibility, but they describe different errors and different code paths." }, { "left": "issue:4634", "right": "issue:4638", "accept": false, - "reason": "Custom Diffusion safe serialization and SDXL textual inversion loading are different code paths and failures." + "reason": "Custom diffusion safe serialization and SDXL textual inversion loading are unrelated loader issues." }, { "left": "issue:4641", "right": "issue:5328", "accept": false, - "reason": "Both involve LoRA loading, but 4641 is local-file LoRA weights and 5328 reports all Civitai LoRAs failing; too broad and not clearly the same defect." + "reason": "Both are LoRA loading problems, but one is a local-file load_lora_weights issue and the other is a broad \u201call failing\u201d report; insufficient evidence they are the same bug." }, { "left": "issue:4667", "right": "issue:8537", "accept": false, - "reason": "Both use from_single_file, but one is a generic pipeline regression and the other is an SD3 float8 attribute error." + "reason": "Both are from from_single_file, but they target different pipelines and different root causes." }, { "left": "issue:4672", "right": "issue:9101", "accept": false, - "reason": "Training VRAM reduction for SDXL LoRA is unrelated to an InstructPix2Pix training script request for SD3." + "reason": "Training-memory reduction for SDXL LoRA is unrelated to adding an InstructPix2Pix training script for SD3." }, { "left": "issue:4675", "right": "issue:6412", "accept": false, - "reason": "ControlNet reference-only feature request is unrelated to annotators-to-ONNX." + "reason": "ControlNet Reference Only for SD XL and ControlNet annotators to ONNX are distinct features." }, { "left": "issue:4681", "right": "issue:4877", "accept": false, - "reason": "Transformer positional embeddings and LoRA noise_init_strength are unrelated changes." + "reason": "Transformer2DModel positional embeddings and making noise_init_strength configurable are unrelated changes." }, { "left": "issue:4708", "right": "issue:5061", "accept": false, - "reason": "Import error in diffusers.utils is unrelated to a convert_original_stable_diffusion_to_diffusers argument error." + "reason": "An import error in diffusers.utils is not the same as an argument error in a conversion script." }, { "left": "issue:4724", "right": "issue:5081", "accept": false, - "reason": "Both are SDXL issues, but one is an Img2Img pipeline behavior bug and the other is an LPW tensor shape mismatch." + "reason": "Both involve SDXL inference, but one is an img2img checkpoint behavior issue and the other is a tensor-shape mismatch in LPW XL." }, { "left": "issue:4764", "right": "issue:6791", "accept": false, - "reason": "LoRA loading from an OFT source is unrelated to the SDXL training script failing to run." + "reason": "A LoRA weight compatibility problem is unrelated to a training script failing to run." }, { "left": "issue:5022", "right": "issue:8771", "accept": false, - "reason": "An import-time tokenizer name error is not the same as removing LoRAAttnProcessor and breaking downstream dependencies." + "reason": "One is a missing-name import error; the other is about removing LoRAAttnProcessor and downstream dependency breakage." }, { "left": "issue:5084", "right": "issue:8742", "accept": false, - "reason": "Flax SDXL import failure is unrelated to LoRAAttnProcessor2_0 removal." + "reason": "FlaxStableDiffusionXLPipeline import failure and LoRAAttnProcessor2_0 removal are different compatibility problems." }, { "left": "issue:5128", "right": "issue:5777", "accept": false, - "reason": "from_single_file loading for XLPipeline and a PixArt dreambooth training error are different issues." + "reason": "Loading a local safetensors file is unrelated to a PixArt DreamBooth training error." }, { "left": "issue:5138", "right": "issue:6388", "accept": false, - "reason": "compute_snr refactor request and gradient-checkpointing being ignored are unrelated training topics." + "reason": "Moving compute_snr to training_utils is a refactor request, not the same as gradient checkpointing being ignored." }, { "left": "issue:525", "right": "issue:770", "accept": false, - "reason": "Local-disk image loading in a notebook and Colab training startup failure are not the same bug." + "reason": "Local-disk image loading in a notebook and Colab startup failures are different training-environment issues." }, { "left": "issue:5374", "right": "issue:7177", "accept": false, - "reason": "Unexpected checkpoint_path argument error is unrelated to StableCascadeDecoderPipeline.from_pretrained failing." + "reason": "An unexpected keyword argument and a StableCascade from_pretrained error are different failure modes." }, { "left": "issue:5429", "right": "issue:6496", "accept": false, - "reason": "Both touch Dreambooth-related examples, but one is a VAE support feature request and the other is a multi-person example request." + "reason": "Support for non-DreamBooth examples without VAE is not the same as a multiple-person DreamBooth example request." }, { "left": "issue:5760", "right": "issue:7405", "accept": false, - "reason": "Video-to-audio synthesis work and DynamiCrafter video diffusion are distinct new-model requests." + "reason": "These are entirely different research-model issues with no shared bug context." }, { "left": "issue:5783", "right": "issue:8803", "accept": false, - "reason": "EMA not working in an SDXL training script is unrelated to a duplicate hidden_states assignment bug." + "reason": "EMA not working in an SDXL training script is unrelated to a duplicate assignment in attention_processor.py." }, { "left": "issue:5791", "right": "issue:5972", "accept": false, - "reason": "Dreambooth hyperparameter interaction bug and a training-docs Min-SNR weighting issue are not the same change." - }, - { - "left": "issue:5805", - "right": "issue:6079", - "accept": false, - "reason": "Both are dtype mismatch errors, but they occur in different contexts and do not clearly share the same root cause." - }, - { - "left": "issue:5905", - "right": "issue:5947", - "accept": false, - "reason": "Diffusion-GAN model request and Diffusion-DPO policy discussion are unrelated." - }, - { - "left": "issue:5925", - "right": "issue:9511", - "accept": false, - "reason": "ZipLora multi-LoRA inference and multi-controlnet batching are different feature areas." + "reason": "A training-parameter interaction report is not the same as a documentation note about Min-SNR weighting." } ] }, @@ -99411,13 +99422,18 @@ "review_unit_count": 364, "status": "reviewed", "reason": null, - "source": "cache", - "cache_hit": true, + "source": "llm", + "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-50-3014", "nodes": [ + "issue:5805", + "issue:5905", + "issue:5925", + "issue:5947", "issue:5977", "issue:5980", + "issue:6079", "issue:6091", "issue:6092", "issue:6093", @@ -99455,11 +99471,15 @@ "issue:9220", "issue:9448", "issue:9477", + "issue:9511", "issue:9674", "issue:9873", "issue:9974" ], "soft_pairs": [ + "issue:5805|issue:6079", + "issue:5905|issue:5947", + "issue:5925|issue:9511", "issue:5977|issue:6093", "issue:5980|issue:6093", "issue:6091|issue:6753", @@ -99483,166 +99503,184 @@ "issue:9032|issue:9039", "issue:9069|issue:9974" ], - "prepared_review_unit_hash": "4d8190fb89cbca0295ac88ca8ccadc6848c8e94500eb625e0b8989f2dcea514c", + "prepared_review_unit_hash": "301c6d5f3c950188f57f2ae8bef5cee252ee2a960a218d888190524d243aa413", "component_budget": { "node_count": 3014, "item_count": 3014, - "soft_pair_count": 8928, - "serialized_chars": 4764584, - "estimated_input_tokens": 1191146, - "estimated_eval_tokens": 2382548 + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 }, "budget": { - "node_count": 42, - "item_count": 42, - "soft_pair_count": 22, - "serialized_chars": 29249, - "estimated_input_tokens": 7313, - "estimated_eval_tokens": 14882 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33362, + "estimated_input_tokens": 8341, + "estimated_eval_tokens": 16938 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are a loose theme cluster around diffusion/pipeline/training topics, but the soft pairs are mostly unrelated or only share a broad subsystem name. I found no true duplicate pair to merge.", - "confidence": 0.97, - "canonical_issue_reason": "No single issue is a clear duplicate target; the set spans unrelated bugs, feature requests, and model-specific training questions.", + "summary": "This is not a true duplicate cluster: the issues span many unrelated bugs and feature requests across SDXL, SD3, LoRA, ControlNet, IP-Adapter, Flax, and training scripts. None of the soft pairs looks like the same underlying change or defect.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue fits the cluster because the items are heterogeneous and do not share one root cause or code path.", "canonical_pr_reason": null, - "best_issue_reason": "issue:8540 is the broadest open SD3 training follow-up in the set, but it still does not subsume the other issues as duplicates.", + "best_issue_reason": "No global best issue is appropriate here; any single issue would misrepresent the mixed set of unrelated reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:5805", + "right": "issue:6079", + "accept": false, + "reason": "Both are dtype-mismatch runtime errors, but they occur in different pipelines/contexts and do not clearly point to the same bug." + }, + { + "left": "issue:5905", + "right": "issue:5947", + "accept": false, + "reason": "One is a request to add a diffusion-GAN model; the other is a policy/objective request for Diffusion-DPO. Different changes." + }, + { + "left": "issue:5925", + "right": "issue:9511", + "accept": false, + "reason": "ZipLoRA multi-LoRA inference and multi-ControlNet batching are unrelated features in different subsystems." + }, { "left": "issue:5977", "right": "issue:6093", "accept": false, - "reason": "Different LoRA save-file keys vs push_to_hub repo-creation behavior; unrelated bugs." + "reason": "Saving extra `unet.unet` keys and push_to_hub repo creation are different LoRA training issues with different fixes." }, { "left": "issue:5980", "right": "issue:6093", "accept": false, - "reason": "ControlNet checkpoint format question vs SDXL LoRA hub-push issue; different topics." + "reason": "ControlNet SDXL output file format question is unrelated to the consistency-model LoRA push_to_hub bug." }, { "left": "issue:6091", "right": "issue:6753", "accept": false, - "reason": "Consistency decoder memory with size argument vs VAE configuration problem; not the same code path." + "reason": "Consistency decoder memory with size argument and VAE configuration are different problems." }, { "left": "issue:6092", "right": "issue:9674", "accept": false, - "reason": "LoRA example not training vs SD3 controlnet poor results; different failure modes and models." + "reason": "LoRA example not training and SD3 ControlNet producing bad results are separate training-path issues." }, { "left": "issue:6203", "right": "issue:6388", "accept": false, - "reason": "SDXL ControlNet image-size conditioning question vs gradient checkpointing being ignored in SDXL-LoRA; unrelated." + "reason": "SDXL ControlNet image-size conditioning and ignored gradient checkpointing in SDXL+LoRA are unrelated." }, { "left": "issue:6240", "right": "issue:7871", "accept": false, - "reason": "AutoPipeline device_map error vs blockwise scales error in a specific SDXL LoRA example; different bugs." + "reason": "AutoPipeline `device_map=\"auto\"` error and blockwise-scale LoRA error affect different code paths." }, { "left": "issue:6256", "right": "issue:7962", "accept": false, - "reason": "Both are pipeline feature requests, but for different APIs and call signatures; not the same change." + "reason": "Both are pipeline API feature requests, but for different callbacks/kwargs and not the same change." }, { "left": "issue:6313", "right": "issue:9873", "accept": false, - "reason": "Null-Text Inversion pipeline implementation vs OmniGen pipeline addition; different pipelines and scope." + "reason": "Null-text inversion pipeline and OmniGen pipeline are distinct model/pipeline requests." }, { "left": "issue:6457", "right": "issue:8543", "accept": false, - "reason": "Multi-GPU gradient accumulation for SDXL-LoRA vs SD3 dreambooth gradient checkpointing failure; distinct training issues." + "reason": "Multi-GPU gradient accumulation for SDXL-LoRA and SD3 dreambooth gradient-checkpointing are separate training bugs." }, { "left": "issue:7112", "right": "issue:7440", "accept": false, - "reason": "IP-Adapter image embed dimension issue vs dataset availability problem; unrelated." + "reason": "IP-Adapter image embedding dimensions and dataset availability are unrelated." }, { "left": "issue:7220", "right": "issue:9448", "accept": false, - "reason": "Both involve IP-Adapter in inpaint pipelines, but the reported errors are different and likely from different code paths." + "reason": "Both involve IP-Adapter in a ControlNet inpaint pipeline, but one is unload-time cleanup and the other is a shape error during use; different code paths." }, { "left": "issue:7438", "right": "issue:8645", "accept": false, - "reason": "XPU generation stall vs missing create_diffusers_controlnet_model_from_ldm helper; unrelated." + "reason": "XPU generation stall and missing `create_diffusers_controlnet_model_from_ldm` are unrelated." }, { "left": "issue:7469", "right": "issue:9220", "accept": false, - "reason": "AutoPipeline parameter naming issue vs PAG pipeline undefined variable; different pipelines and bugs." + "reason": "AutoPipeline variable naming and PAG pipeline undefined variable are different bugs in different code." }, { "left": "issue:7798", "right": "issue:7836", "accept": false, - "reason": "PixArtAlpha pipeline bug vs CLIP training overfitting example; no overlap." + "reason": "PixArtAlphaPipeline bug and CLIP training overfitting are unrelated." }, { "left": "issue:7841", "right": "issue:8731", "accept": false, - "reason": "Red-dot rendering artifact vs batched generation slowness; different symptoms." + "reason": "Visual artifact in SDXL output and batched text-to-image slowness are different symptoms and fixes." }, { "left": "issue:8320", "right": "issue:9477", "accept": false, - "reason": "from_single_file AutoencoderKL __name__ error vs GatheredParameters callable error; unrelated." + "reason": "Single-file AutoencoderKL `__name__` error and `GatheredParameters` callable error are unrelated." }, { "left": "issue:8485", "right": "issue:8540", "accept": false, - "reason": "SD3 LoRA support request vs SD3 dreambooth caching/representation optimization; related theme, but not the same fix." + "reason": "SD3 LoRA support and SD3 dreambooth embedding/VAE caching are separate feature requests." }, { "left": "issue:8500", "right": "issue:8534", "accept": false, - "reason": "SD3 token-limit question vs wrong logit-normal weighting in SD3 dreambooth; different problems." + "reason": "SD3 token-limit question and wrong logit-normal weighting are different SD3 training concerns." }, { "left": "issue:8504", "right": "issue:8540", "accept": false, - "reason": "Choosing SD3 text encoders vs caching SD3 embeddings/VAE outputs; separate training features." + "reason": "Choosing text encoders for SD3-medium and caching embeddings/VAE representations are distinct enhancements." }, { "left": "issue:8546", "right": "issue:8649", "accept": false, - "reason": "Missing from-single-file support vs num_images_per_prompt regression; unrelated SD3 issues." + "reason": "Missing SD3 from-single-file support and `num_images_per_prompt` regression are unrelated issues." }, { "left": "issue:9032", "right": "issue:9039", "accept": false, - "reason": "FlaxStableDiffusionPipeline TPU/MWE setup vs Flax LoRA loading; different user needs and code paths." + "reason": "Both are Flax pipeline questions, but TPU minimal example and LoRA loading are not the same underlying bug/change." }, { "left": "issue:9069", "right": "issue:9974", "accept": false, - "reason": "Numpy type error vs SD3 ControlNet parameter expectations; unrelated." + "reason": "A NumPy type error and adding expected parameters to `controlnet_sd3` are unrelated." } ] }, @@ -107374,7 +107412,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 1, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -107469,10 +107507,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -107709,7 +107747,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 2, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -107801,10 +107839,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -108020,7 +108058,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 3, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -108111,10 +108149,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -108327,7 +108365,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 4, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -108415,10 +108453,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -108607,7 +108645,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 5, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -108696,10 +108734,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -108891,7 +108929,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 6, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -108982,10 +109020,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -109192,7 +109230,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 7, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -109282,10 +109320,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -109492,7 +109530,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 8, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -109582,10 +109620,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -109786,7 +109824,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 9, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -109875,10 +109913,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -110070,7 +110108,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 10, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -110158,10 +110196,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -110347,7 +110385,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 11, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -110437,10 +110475,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -110644,7 +110682,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 12, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -110732,10 +110770,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -110921,7 +110959,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 13, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -111017,10 +111055,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -111254,7 +111292,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 14, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -111343,10 +111381,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -111541,7 +111579,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 15, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -111628,10 +111666,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -111805,7 +111843,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 16, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -111893,10 +111931,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -112082,7 +112120,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 17, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -112169,10 +112207,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -112361,7 +112399,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 18, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -112452,10 +112490,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -112659,7 +112697,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 19, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -112747,10 +112785,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -112942,7 +112980,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 20, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -113032,10 +113070,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -113233,7 +113271,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 21, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -113322,10 +113360,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -113517,7 +113555,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 22, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -113603,10 +113641,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -113786,7 +113824,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 23, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -113872,10 +113910,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -114049,7 +114087,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 24, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -114139,10 +114177,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -114340,7 +114378,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 25, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -114434,10 +114472,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -114665,7 +114703,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 26, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -114754,10 +114792,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -114949,7 +114987,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 27, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -115037,10 +115075,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -115226,7 +115264,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 28, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -115312,10 +115350,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -115495,7 +115533,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 29, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -115585,10 +115623,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -115786,7 +115824,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 30, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -115874,10 +115912,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -116069,7 +116107,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 31, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -116157,10 +116195,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -116346,7 +116384,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 32, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -116432,10 +116470,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -116615,7 +116653,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 33, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -116702,10 +116740,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -116891,7 +116929,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 34, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -116979,10 +117017,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -117174,7 +117212,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 35, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -117262,10 +117300,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -117451,7 +117489,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 36, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -117538,10 +117576,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -117727,7 +117765,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 37, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -117820,10 +117858,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -118045,7 +118083,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 38, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -118139,10 +118177,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -118364,7 +118402,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 39, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -118463,10 +118501,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -118718,7 +118756,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 40, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -118838,10 +118876,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 33, @@ -119309,7 +119347,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 41, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -119401,10 +119439,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -119620,7 +119658,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 42, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -119709,10 +119747,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -119910,7 +119948,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 43, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -119999,10 +120037,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -120017,8 +120055,8 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a heterogeneous set of unrelated docs, CI, test, and model changes. None of the soft-edge pairs look like the same underlying bug or feature change, so I would not merge any of them as duplicates.", - "confidence": 0.97, + "summary": "This is a heterogeneous set of unrelated PRs spanning docs fixes, CI tweaks, model changes, and feature work. The soft pairs are mostly superficial title/file similarity rather than the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.91, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -120028,157 +120066,157 @@ "left": "pull_request:5499", "right": "pull_request:6998", "accept": false, - "reason": "Both are documentation edits, but they fix different link/callout issues in different pages." + "reason": "Both are docs link/callout edits, but they fix different pages and different link formatting issues." }, { "left": "pull_request:55", "right": "pull_request:59", "accept": false, - "reason": "Both touch resnet code, but they are separate architecture refactors for different model families and code paths." + "reason": "Both touch resnet code, but one merges glide resnet into the general resnet while the other finalizes a different resnet variant; not the same change." }, { "left": "pull_request:5506", "right": "pull_request:5823", "accept": false, - "reason": "One fixes PEFT/xFormers scale passing; the other adds adapter_names support to fuse_lora. Different bugs and changes." + "reason": "Both involve LoRA/PEFT attention code, but one fixes xformers scale handling and the other adds adapter_names to fuse_lora; different code paths." }, { "left": "pull_request:555", "right": "pull_request:824", "accept": false, - "reason": "Flax VAE addition and a Flax test update are not the same underlying change." + "reason": "Both are Flax-related model utilities, but one adds Flax VAE support and the other adds a Flax test and utility cleanup; unrelated changes." }, { "left": "pull_request:5698", "right": "pull_request:6198", "accept": false, - "reason": "LCM img2img behavior fix versus benchmarking/reporting changes are unrelated." + "reason": "Both mention LCM, but one fixes img2img behavior in the pipeline and the other adds benchmark reporting; different functionality." }, { "left": "pull_request:5707", "right": "pull_request:5727", "accept": false, - "reason": "One adds tests, the other adds training scripts; they do not fix the same code-path problem." + "reason": "One is a fix for LCM loading across pipelines; the other adds LCM training scripts and README. Not the same bug/change." }, { "left": "pull_request:5745", "right": "pull_request:6475", "accept": false, - "reason": "Both are doc-type fixes, but for different classes/models and different files." + "reason": "Both are ConsistencyDecoder/SVD docs-only tweaks, but they address different documentation details." }, { "left": "pull_request:5745", "right": "pull_request:7015", "accept": false, - "reason": "These are unrelated documentation corrections in different areas of the codebase." + "reason": "A doc type/example fix versus a single-file loading doc/example fix; different artifacts and problems." }, { "left": "pull_request:5818", "right": "pull_request:7324", "accept": false, - "reason": "Both are test-related, but they target different pipelines and different test suites." + "reason": "Both are CI/test workflow related, but one updates PR test matrix setup while the other updates Stable Cascade tests; not the same issue." }, { "left": "pull_request:5855", "right": "pull_request:6669", "accept": false, - "reason": "Both edit README text, but one is a typo fix and the other updates repository counts/links." + "reason": "Both are README/doc errata, but they change different example/community docs and different mistakes." }, { "left": "pull_request:5983", "right": "pull_request:6002", "accept": false, - "reason": "SVD docs cleanup and stable video diffusion guidance-scale logic are not the same bug/change." + "reason": "Both are SVD docs/pipeline fixes, but one edits documentation while the other changes guidance-scale logic in the pipeline." }, { "left": "pull_request:6314", "right": "pull_request:7472", "accept": false, - "reason": "These are separate broken-link fixes in different docs." + "reason": "Both are typo/doc fixes, but they target different docs and unrelated wording/link issues." }, { "left": "pull_request:6433", "right": "pull_request:8426", "accept": false, - "reason": "Community pipeline addition and CI workflow dependency changes are unrelated." + "reason": "One adds a community pipeline entry; the other changes workflow dependency installation. No shared underlying bug." }, { "left": "pull_request:6462", "right": "pull_request:7472", "accept": false, - "reason": "Both are docs/link tweaks, but they fix different pages and different link targets." + "reason": "Different docs updates in different files; one fixes a paper link, the other a broken examples link." }, { "left": "pull_request:6837", "right": "pull_request:8851", "accept": false, - "reason": "Different documentation additions for different models/pipelines." + "reason": "Both are docs additions, but one is an i2vgen XL docs dependency/import tweak and the other adds AuraFlow docs and code." }, { "left": "pull_request:7181", "right": "pull_request:7325", "accept": false, - "reason": "Both are typo sweeps, but they touch different sets of files and are not the same concrete change." + "reason": "Both are typo/doc cleanups, but they touch many different files and unrelated text fixes." }, { "left": "pull_request:7319", "right": "pull_request:9066", "accept": false, - "reason": "These add docs for different models/schedulers; same area, but not the same underlying feature." + "reason": "Both add scheduler/model docs, but one introduces EDM scheduler pages and the other adds Stable Cascade UNet docs." }, { "left": "pull_request:74", "right": "pull_request:9", "accept": false, - "reason": "Grad-TTS mid-block addition is a later architectural change, not the same as the original Grad-TTS model/pipeline introduction." + "reason": "Both concern Grad-TTS/resnet-era model work, but one adds a mid-block to Grad-TTS while the other introduces the Grad-TTS model/pipeline itself." }, { "left": "pull_request:7425", "right": "pull_request:8851", "accept": false, - "reason": "Unrelated docs fixes for different tutorial/model pages." + "reason": "Both are docs-related, but one fixes LoRA tutorial text while the other adds AuraFlow docs and implementation." }, { "left": "pull_request:7709", "right": "pull_request:8403", "accept": false, - "reason": "Both mention Tailscale, but they modify different workflows for different CI behavior." + "reason": "Both touch CI runner/workflow infrastructure, but one adds tailscale to push tests and the other updates the tailscale action version in SSH runner." }, { "left": "pull_request:7797", "right": "pull_request:8932", "accept": false, - "reason": "These are unrelated test-suite changes for different models." + "reason": "Both reduce test coverage/runtime by adjusting tiny test models, but they are for different test suites and different model classes." }, { "left": "pull_request:7867", "right": "pull_request:9024", "accept": false, - "reason": "Both adjust CI cache/setup, but they are different workflow-wide infra changes, not one concrete fix." + "reason": "Both modify workflow cache settings, but they target different CI changes and 9024 spans many workflows plus dependency updates." }, { "left": "pull_request:798", "right": "pull_request:972", "accept": false, - "reason": "Both are typo fixes, but they correct different typos in different files." + "reason": "Both are small docstring/readme fixes, but they address different typos in different files." }, { "left": "pull_request:798", "right": "pull_request:975", "accept": false, - "reason": "Docstring typo fix and a docs title capitalization fix are unrelated." + "reason": "Different documentation typo fixes in different files; no shared underlying change." }, { "left": "pull_request:8470", "right": "pull_request:9266", "accept": false, - "reason": "Different README/community documentation updates with no shared bug or change." + "reason": "One rewrites a research project README; the other adds an entry to community projects docs. Same broad area, but not the same change." }, { "left": "pull_request:8600", "right": "pull_request:9107", "accept": false, - "reason": "Both touch the same training script, but they fix different bugs: deletion of tokenizers vs loading-hook key filtering." + "reason": "Both are SD3 DreamBooth-related, but one fixes text-encoder cleanup in two scripts while the other fixes a loading hook state-dict filter; different bugs." } ] }, @@ -120194,7 +120232,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 44, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -120281,10 +120319,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -120470,7 +120508,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 45, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -120558,10 +120596,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -120747,7 +120785,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 46, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -120834,10 +120872,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -121017,7 +121055,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 47, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -121105,10 +121143,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -121300,7 +121338,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 48, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -121388,10 +121426,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -121583,7 +121621,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 49, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -121672,10 +121710,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -121690,168 +121728,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly false-positive similarity: the pairs span unrelated docs edits, test xfails, pipeline bug fixes, and separate LoRA/Flux/controlnet changes. I don\u2019t see a true duplicate cluster here.", - "confidence": 0.96, + "summary": "This cluster is a grab bag of unrelated docs, test, CI, and a few LoRA/Flux/controlnet fixes. The soft pairs are mostly title/file-similarity false positives rather than the same underlying bug or change, so I would not merge any of them as duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a clear duplicate hub; the items are heterogeneous and mostly only share superficial wording or subsystem overlap.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is a strong global canonical choice because there is no single underlying bug/change across the set.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:382", "right": "pull_request:388", "accept": false, - "reason": "Both are docs-only pipeline pages, but they document different pipelines and different content." + "reason": "Both are docs-only, but they update different pipeline pages for different models; no shared bug or change." }, { "left": "pull_request:3923", "right": "pull_request:5236", "accept": false, - "reason": "One fixes a missing torch import in stable diffusion docs; the other is a Korean docs formatting cleanup. Unrelated." + "reason": "Both are documentation fixes, but for unrelated pages/languages and different content issues." }, { "left": "pull_request:3987", "right": "pull_request:4707", "accept": false, - "reason": "Audio Diffusion code snippet fix vs. custom diffusion training README/script adjustments; different change sets and bugs." + "reason": "Audio Diffusion snippet fix and Custom Diffusion README rewrite are unrelated examples/docs changes." }, { "left": "pull_request:4574", "right": "pull_request:6342", "accept": false, - "reason": "Separate documentation updates in different areas; no shared concrete bug or edit." + "reason": "One updates InstructPix2Pix README examples; the other adds an installation note. Different issues." }, { "left": "pull_request:5003", "right": "pull_request:5298", "accept": false, - "reason": "Both touch LoRA internals, but one fixes fuse/unfuse state handling and the other fixes torch.compile behavior for LoRA convs; different code-path problems." + "reason": "Both touch LoRA code, but one fixes fuse/unfuse logic while the other fixes torch.compile conv behavior; different code-path bugs." }, { "left": "pull_request:5051", "right": "pull_request:7068", "accept": false, - "reason": "Broken-link docs fix and typo cleanup are not the same underlying change." + "reason": "Broken docs link vs typo/docstring cleanup; same broad docs area, but not the same change." }, { "left": "pull_request:5157", "right": "pull_request:5833", "accept": false, - "reason": "ControlNet LoRA conversion logic vs. PixArt docs expansion; unrelated." + "reason": "ControlNet LoRA name conversion fix and single-file loading import fix are different loader paths and bugs." }, { "left": "pull_request:7050", "right": "pull_request:7118", "accept": false, - "reason": "Both are typo/doc cleanup PRs, but they edit different docs and fix different wording issues." + "reason": "Both are typo/doc tweaks, but in different examples/docs sections with no shared underlying defect." }, { "left": "pull_request:8490", "right": "pull_request:9760", "accept": false, - "reason": "README_sd3 content correction vs. top-level README wording cleanup; no shared concrete bug." + "reason": "DreamBooth SD3 README script-name correction and top-level README wording cleanup are unrelated docs edits." }, { "left": "pull_request:8642", "right": "pull_request:9100", "accept": false, - "reason": "Different typo fixes in different training docs/code; not the same issue." + "reason": "Stable Diffusion 3 return-type typo and text-to-image help-text typo are separate docs fixes." }, { "left": "pull_request:8860", "right": "pull_request:9760", "accept": false, - "reason": "No diff evidence for 8860 and, even if it is a README update, it is not clearly the same change as the top-level README edit." + "reason": "One is a README_sd3 update, the other is a broad README wording tweak; not the same change." }, { "left": "pull_request:9070", "right": "pull_request:9083", "accept": false, - "reason": "A new Flux img2img pipeline and single-file support for Flux Transformer are distinct features." + "reason": "Adding a Flux img2img pipeline and adding single-file support for FluxTransformer2DModel are distinct features." }, { "left": "pull_request:9086", "right": "pull_request:9754", "accept": false, - "reason": "Both are Flux training-related, but one adds DreamBooth/LoRA scripts and the other adds a quantized research workflow; not the same concrete fix." + "reason": "Both are Flux-related, but one adds DreamBooth LoRA training scripts while the other adds a quantized research training setup; different concrete work." }, { "left": "pull_request:4352", "right": "pull_request:5470", "accept": false, - "reason": "Different pipeline families and different fixes; only broad subsystem overlap." + "reason": "Kandinsky CPU offload/xformers support and ControlNet inpaint pipeline/docs fixes are unrelated." }, { "left": "pull_request:5129", "right": "pull_request:5202", "accept": false, - "reason": "Multiple-LoRA fuse/unfuse support and Kohya checkpoint mapping are related to LoRA, but they solve different problems in different paths." + "reason": "One adds multi-LoRA fuse/unfuse support; the other adds Kohya checkpoint mapping entries. Different functionality." }, { "left": "pull_request:1229", "right": "pull_request:1508", "accept": false, - "reason": "Both edit the DreamBooth README, but one changes install guidance and the other clarifies num_class_images; not duplicates." + "reason": "Both edit DreamBooth README text, but they address different guidance topics and not the same issue." }, { "left": "pull_request:13352", "right": "pull_request:13536", "accept": false, - "reason": "Same workflow file, but one adds checkout and the other changes permissions/instructions; separate CI workflow changes." + "reason": "One adds a checkout step; the other changes workflow permissions/instructions. Same file, different workflow behavior." }, { "left": "pull_request:2892", "right": "pull_request:4373", "accept": false, - "reason": "Stable Diffusion Upscale input validation vs. SDXL resize ordering; different pipeline bugs." + "reason": "Upcaler input validation for embeds and SDXL resize/default-size handling are separate pipeline bugs." }, { "left": "pull_request:3334", "right": "pull_request:5015", "accept": false, - "reason": "Docstring wording cleanup vs. broken docs link; unrelated documentation fixes." + "reason": "Docstring formatting cleanup and broken-link fix are unrelated documentation fixes." }, { "left": "pull_request:4188", "right": "pull_request:5420", "accept": false, - "reason": "Different follow-up docs/examples PRs with no shared underlying bug or code path." + "reason": "ControlNet SDXL dataset fingerprinting and basic training doc example correction are different problems." }, { "left": "pull_request:4242", "right": "pull_request:5833", "accept": false, - "reason": "Single-file VAE loading support vs. ControlNet import-path fix in single-file loading; distinct changes." + "reason": "Allowing VAE in from_single_file and fixing ControlNet import paths are different loader issues." }, { "left": "pull_request:4242", "right": "pull_request:5834", "accept": false, - "reason": "Single-file VAE loading support vs. ControlNet import-path fix in single-file loading; distinct changes." + "reason": "VAE loading support versus single-file ControlNet import correction are different code paths." }, { "left": "pull_request:4749", "right": "pull_request:4859", "accept": false, - "reason": "DreamBooth LoRA checkpointing vs. SDXL inpaint fast-test input tweak; unrelated." + "reason": "SDXL LoRA checkpointing and SDXL inpaint fast-test setup are unrelated changes." }, { "left": "pull_request:4833", "right": "pull_request:5157", "accept": false, - "reason": "Both are LoRA-related code changes, but one fixes unfuse behavior and the other fixes Kohya/controlnet name conversion; different bugs." + "reason": "LoRA fuse/unfuse internals and ControlNet LoRA conversion mappings are different bugs in different areas." }, { "left": "pull_request:6215", "right": "pull_request:7632", "accept": false, - "reason": "Both mention VAE, but one adjusts PixArt test initialization and the other fixes DreamBooth VAE path handling; not the same issue." + "reason": "PixArt test initialization and DreamBooth VAE path handling are unrelated changes." }, { "left": "pull_request:11469", "right": "pull_request:12418", "accept": false, - "reason": "Both xfail tests, but they target different test files and different version-specific failures." + "reason": "Both are test xfails, but for different pipelines and different failure modes; not the same fix." } ] }, @@ -121867,7 +121905,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 50, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -121955,10 +121993,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -121973,163 +122011,172 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: the PRs span unrelated CI, docs, LoRA, Flux, Wan, Qwen, scheduler, and test changes. The soft edges are mostly superficial similarity rather than the same underlying bug or feature.", - "confidence": 0.97, + "summary": "Most of the cluster is unrelated feature/docs/test work. The only grounded duplicate-like pair is the tokenizers CI workflow follow-up; the Sana layerwise-casting skip pair is not the same fix and should be rejected.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items do not share one underlying bug or change, so none of the PRs cleanly represents the whole cluster.", + "canonical_pr_reason": "PR 10132 is the base tokenizers workflow fix; PR 10497 is a narrow follow-up correcting that same install step.", "best_issue_reason": null, - "best_pr_reason": "No single PR is a good global representative; the cluster is too heterogeneous for a meaningful best-PR choice.", + "best_pr_reason": "No PR is a strong global representative for this heterogeneous cluster; 10132 is only the best local anchor because it owns the only grounded duplicate pair.", "soft_edge_verdicts": [ { "left": "pull_request:11716", "right": "pull_request:12714", "accept": false, - "reason": "Different LoRA problems: return_metadata handling vs Flux2 NaN test; not the same code path." + "reason": "Both are LoRA test/code changes, but one fixes Flux metadata handling and the other fixes a Flux2 NaN test; different bugs/code paths." }, { "left": "pull_request:12551", "right": "pull_request:12647", "accept": false, - "reason": "Both touch CI skips, but they target different tests and different failures." + "reason": "Sana test skip vs autoencoder_dc layerwise-casting skip in a different test file; not the same bug or mergeable fix." }, { "left": "pull_request:12526", "right": "pull_request:13058", "accept": false, - "reason": "Different Wan features: Animate pipeline vs LTX2 condition pipeline; unrelated changes." + "reason": "Base Qwen-Image support vs Qwen Image Animate pipeline; separate feature additions." }, { "left": "pull_request:10359", "right": "pull_request:9696", "accept": false, - "reason": "Unrelated CI/test-workflow changes, not the same underlying fix." + "reason": "A Sana test cleanup and a PR workflow change; unrelated." }, { "left": "pull_request:4862", "right": "pull_request:5470", "accept": false, - "reason": "Different ControlNet bugs: multi-controlnet img2img sizing vs upscale safety checker." + "reason": "SDXL MultiControlNet height/width handling vs inpaint pipeline docs/fixes; different issues." }, { "left": "pull_request:6898", "right": "pull_request:6939", "accept": false, - "reason": "Both standardize model cards, but for different scripts and content; not one duplicate change." + "reason": "Both standardize model cards, but they target different example scripts and aren\u2019t one concrete change." }, { "left": "pull_request:10750", "right": "pull_request:9638", "accept": false, - "reason": "Different Flux work: ControlNet img2img support vs Xlabs ControlNets." + "reason": "Flux ControlNet img2img support vs Xlabs ControlNet support; separate implementations." }, { "left": "pull_request:7638", "right": "pull_request:7717", "accept": false, - "reason": "Both are docs/style cleanups, but they fix different files and issues." + "reason": "Wrong-link fix vs docstring style cleanup; unrelated." }, { "left": "pull_request:11025", "right": "pull_request:11353", "accept": false, - "reason": "Different Wan additions: LoRA conversion support vs FLF2V pipeline/docs support." + "reason": "Wan LoRA conversion vs FLF2V pipeline support; separate features." }, { "left": "pull_request:12055", "right": "pull_request:12164", "accept": false, - "reason": "Qwen-Image base support and Qwen Image Edit support are related but distinct features." + "reason": "Qwen-Image base package support vs Qwen Image Edit support; different feature sets." }, { "left": "pull_request:1920", "right": "pull_request:2398", "accept": false, - "reason": "Different documentation updates for unrelated topics." + "reason": "Different docs edits in unrelated areas." }, { "left": "pull_request:2588", "right": "pull_request:4396", "accept": false, - "reason": "Different changes: LoRA training support vs prompt-encoding dtype fallback." + "reason": "Text-encoder LoRA support vs encode_prompt fallback behavior; different problems." }, { "left": "pull_request:4272", "right": "pull_request:4276", "accept": false, - "reason": "Optional import cleanup and pipeline import smoke test are not the same fix." + "reason": "Missing optional imports vs import test; related topic, not the same change." }, { "left": "pull_request:4621", "right": "pull_request:6564", "accept": false, - "reason": "Different adapter logic vs IP Adapter path handling; unrelated code-path problems." + "reason": "Multi-adapter validation vs IP-Adapter path handling; different subsystems." }, { "left": "pull_request:6564", "right": "pull_request:8825", "accept": false, - "reason": "Path generation fix in IP Adapter vs adding ControlNet/IP-Adapter support to K-Diffusion." + "reason": "IP-Adapter path fix vs SDXL K-Diffusion ControlNet/IP-Adapter support; not the same bug." }, { "left": "pull_request:7118", "right": "pull_request:9207", "accept": false, - "reason": "Different docs fixes in different pages; not the same change." + "reason": "IP-Adapter docs typos vs CPU-offloading docs; unrelated." }, { "left": "pull_request:7131", "right": "pull_request:9207", "accept": false, - "reason": "Same here: separate documentation fixes, not duplicates." + "reason": "Separate documentation edits in different areas." }, { "left": "pull_request:7932", "right": "pull_request:9696", "accept": false, - "reason": "LoRA test expectation changes and PR-test workflow changes are unrelated." + "reason": "LoRA test expectation tweaks vs PR workflow update; unrelated." }, { "left": "pull_request:6898", "right": "pull_request:6907", "accept": false, - "reason": "Both are model-card standardization efforts, but on different examples and with different edits." + "reason": "Different model-card standardizations for different example families; too broad for one duplicate." }, { "left": "pull_request:1229", "right": "pull_request:1513", "accept": false, - "reason": "Both update DreamBooth docs, but they add different instructions for different setup steps." + "reason": "Dreambooth README setup changes for different instructions." }, { "left": "pull_request:2595", "right": "pull_request:3128", "accept": false, - "reason": "Scheduler config switching and img2img timestep slicing solve different bugs." + "reason": "Scheduler config switching vs img2img timestep selection; separate bugs." }, { "left": "pull_request:4862", "right": "pull_request:4886", "accept": false, - "reason": "Different pipeline bugs in ControlNet img2img vs StableDiffusionUpscale safety checking." + "reason": "MultiControlNet shape handling vs upscale safety-checker input processing; different bugs." }, { "left": "pull_request:6448", "right": "pull_request:7738", "accept": false, - "reason": "Different LoRA test/device fixes; same area, but not the same underlying issue." + "reason": "Two LoRA test/code adjustments, but they fix different assertions and paths." }, { "left": "pull_request:10717", "right": "pull_request:10867", "accept": false, - "reason": "Single IP-Adapter support vs multi-IP-Adapter support; related feature area, but not one mergeable change." + "reason": "Flux img2img IP-Adapter support vs multi-IP-Adapter support; separate changes." + }, + { + "left": "pull_request:10132", + "right": "pull_request:10497", + "accept": true, + "reason": "Same tokenizers install line in the same workflow; 10497 is a direct correction to 10132's install target." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, + "evaluator_result": { + "accept": false, + "feedback": "Reject: one accepted soft edge is not grounded. PR 12551 is a Sana test skip, while PR 12647 is an autoencoder_dc layerwise-casting skip in a different test file; they do not look like a single mergeable bugfix. The summary also overstates that this is the same Sana CI flake family. The tokenizers workflow follow-up (10132/10497) is grounded, but the cluster-level claim that only two soft edges are real duplicates is too strong given the unsupported second acceptance." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -122138,7 +122185,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 51, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -122225,10 +122272,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -122243,162 +122290,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a grab bag of unrelated PRs: docs cleanups, test-only changes, and a few distinct runtime fixes/features across Flux, Sana, HunyuanVideo, XLA, and LTX. I don\u2019t see any true duplicate pair here, so no soft edge should be accepted.", - "confidence": 0.96, + "summary": "This set is mostly unrelated PRs that only look similar by title or broad subsystem. I found no true duplicate pairs; the soft edges are all false positives across docs, typos, feature work, and separate bug fixes.", + "confidence": 0.24, "canonical_issue_reason": null, - "canonical_pr_reason": "If forced to pick one canonical PR, PR 10115 is the cleanest standalone bug fix: it has an explicit issue target, a narrow code-path correction in embeddings.py, and no mixed doc/feature churn.", + "canonical_pr_reason": "PR 10498 is the strongest representative PR here: it has a clear concrete runtime change, touches multiple code paths in a single feature area (PyTorch/XLA support), and is linked to an explicit issue target.", "best_issue_reason": null, - "best_pr_reason": "PR 10115 is the best representative of a real defect fix rather than a docs-only or broad feature PR; it is specific, merged, and directly addresses a runtime bug.", + "best_pr_reason": "PR 10498 is the best overall PR candidate because it is substantive, self-contained, and clearly scoped to one underlying change, unlike the many small docs/typo PRs in the set.", "soft_edge_verdicts": [ { "left": "pull_request:12983", "right": "pull_request:13121", "accept": false, - "reason": "Different LTX-2 changes: single-file loading support vs a separate inference bug fix for num_videos_per_prompt/CFG." + "reason": "Different LTX-2 changes: one adds single-file support, the other fixes batching/CFG inference. Same area, not the same bug or change." }, { "left": "pull_request:5979", "right": "pull_request:6303", "accept": false, - "reason": "Both are large new pipeline/script efforts, but they target different models and features, not the same underlying change." + "reason": "Completely different work: DDGAN pipeline/scheduler vs ADD training scripts. No shared code-path problem." }, { "left": "pull_request:10192", "right": "pull_request:10259", "accept": false, - "reason": "Both are Flux-related, but one adds a Control/Inpaint pipeline and the other changes LoRA loading behavior; distinct code paths and goals." + "reason": "Both Flux-related, but one adds a Control/Inpaint pipeline while the other adds LoRA loading support for Flux Control/Fill. Different features." }, { "left": "pull_request:10340", "right": "pull_request:10431", "accept": false, - "reason": "Both touch Sana, but one fixes 2K model config and the other fixes 1K positional embedding behavior; separate bugs." + "reason": "Both Sana fixes, but they address different configuration/positional-embedding issues and different model variants." }, { "left": "pull_request:10115", "right": "pull_request:8481", "accept": false, - "reason": "FluxPosEmbed rotary embedding bug fix vs an image processor validation-message change; unrelated artifacts." + "reason": "Unrelated bugs in different subsystems: Flux rotary embeddings vs image processor input validation." }, { "left": "pull_request:1020", "right": "pull_request:1194", "accept": false, - "reason": "General docs/training updates vs CI workflow changes for documentation; not the same underlying change." + "reason": "Docs update vs GitHub Actions workflow update; not the same underlying change." }, { "left": "pull_request:1020", "right": "pull_request:2788", "accept": false, - "reason": "Both alter docs navigation, but one is training docs content and the other is toctree deduplication; different edits." + "reason": "Both touch docs structure, but one updates prose while the other deduplicates the training section tree. Distinct changes." }, { "left": "pull_request:10222", "right": "pull_request:8600", "accept": false, - "reason": "PyTorch/XLA support in ControlNet is unrelated to Dreambooth SD3 cleanup." + "reason": "PyTorch/XLA runtime support in ControlNet is unrelated to DreamBooth SD3 training cleanup." }, { "left": "pull_request:10230", "right": "pull_request:10318", "accept": false, - "reason": "Docs adding LoRA mixin references vs Flux LoRA integration tests; documentation and test additions are not the same change." + "reason": "Docs list expansion vs Flux LoRA integration tests. Related topic, but not the same change or bug." }, { "left": "pull_request:10232", "right": "pull_request:9734", "accept": false, - "reason": "Minor Sana docs correction vs SD3 doc/test/model update; different scope and code paths." + "reason": "Separate Sana documentation correction vs broader SD3 doc/test update." }, { "left": "pull_request:10253", "right": "pull_request:9196", "accept": false, - "reason": "Contributed note in autoencoder docs vs PixArt attention-processor API change; unrelated." + "reason": "Contributor-note docs change vs PixArt attention processor feature; unrelated." }, { "left": "pull_request:10278", "right": "pull_request:11634", "accept": false, - "reason": "Both edit torchao docs, but one fixes an example and the other fixes a reference link; not the same concrete change." + "reason": "Same doc file, but one fixes the example and the other adds a link reference. Not the same issue." }, { "left": "pull_request:10358", "right": "pull_request:11319", "accept": false, - "reason": "Both add community projects entries, but they are different projects and different PRs." + "reason": "Both add community projects, but they are different project entries with no shared bug/change." }, { "left": "pull_request:10376", "right": "pull_request:11546", "accept": false, - "reason": "HunyuanVideo original LoRA conversion support vs FramePack LoRA loader registration; different model support work." + "reason": "Both LoRA-related, but one converts original HunyuanVideo LoRAs and the other adds FramePack LoRA loading support. Different code paths." }, { "left": "pull_request:10394", "right": "pull_request:9116", "accept": false, - "reason": "Peft tutorial link correction vs Stable Audio dead-link fix; both docs-only but unrelated." + "reason": "Both are link fixes, but they fix different docs and different links." }, { "left": "pull_request:1041", "right": "pull_request:531", "accept": false, - "reason": "Early test fixes for stable diffusion/latent diffusion vs GPU tensor-device test adjustments; no shared bug." + "reason": "Different test fixes in different areas and eras; no shared defect." }, { "left": "pull_request:10434", "right": "pull_request:11847", "accept": false, - "reason": "Both are docs link/example fixes, but they touch different pages and different broken references." + "reason": "Docs link repair across many pages vs a single-file example correction. Not the same underlying change." }, { "left": "pull_request:10498", "right": "pull_request:11204", "accept": false, - "reason": "PyTorch/XLA runtime support across pipelines vs SDXL XLA training implementation; different layers of the stack and different goals." + "reason": "Both involve PyTorch/XLA, but one adds inference/runtime support across pipelines while the other adds an SDXL training example. Different problems." }, { "left": "pull_request:10531", "right": "pull_request:11851", "accept": false, - "reason": "SD3 negative-prompt doc updates vs a schedulers doc note about _keep_in_fp32_modules; unrelated documentation topics." + "reason": "Negative-prompt docs for Flux vs a note about `_keep_in_fp32_modules`; unrelated documentation changes." }, { "left": "pull_request:10548", "right": "pull_request:11789", "accept": false, - "reason": "HunyuanVideo batching bug fix vs HunyuanVideo Framepack device-test adjustments; one is runtime logic, the other is test maintenance." + "reason": "HunyuanVideo batch-size bug fix vs Framepack device-test adjustments. Different artifacts and different issues." }, { "left": "pull_request:1077", "right": "pull_request:1412", "accept": false, - "reason": "Scheduler test refactor vs SD2 inpainting integration tests; both are tests but not the same change." + "reason": "Scheduler test refactor vs SD2 inpainting integration tests. Different changes." }, { "left": "pull_request:10802", "right": "pull_request:11623", "accept": false, - "reason": "A typo fix in a controlnet training example vs a typo fix in Flux pipeline docs; different files and unrelated text." + "reason": "Both typo fixes, but in different files and with different wording; not a duplicate change." }, { "left": "pull_request:10819", "right": "pull_request:11770", "accept": false, - "reason": "Adding missing LoRA docs entries vs broader LoRA docs cleanup and code/doc restructuring; not the same fix." + "reason": "Both touch LoRA docs, but one adds missing entries while the other performs broader cleanup and code/doc refactors. Not the same fix." }, { "left": "pull_request:10819", "right": "pull_request:11851", "accept": false, - "reason": "LoRA docs entry additions vs schedulers documentation note; different subsystems and different content." + "reason": "LoRA docs additions vs an fp32-module documentation note; unrelated." }, { "left": "pull_request:10888", "right": "pull_request:9860", "accept": false, - "reason": "AutoPipeline registration for SD3 ControlNet vs SD3.5 ControlNet example support; one is API wiring, the other is example/model support." + "reason": "Both concern SD3 ControlNet, but one wires AutoPipeline support while the other updates the training example for SD3.5. Different scopes and code paths." } ] }, @@ -122414,7 +122461,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 52, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -122510,10 +122557,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -122528,210 +122575,210 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated docs cleanups, typo fixes, small feature additions, and modular-pipeline changes. I don\u2019t see a single underlying bug/change family that would justify merging them as duplicates.", - "confidence": 0.14, + "summary": "This cluster is mostly a false-positive pileup: unrelated docs typos, feature additions, modular-pipeline refactors, and a few genuine code fixes. None of the suggested soft edges look like the same underlying change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 11407 is the broadest maintenance/typo sweep and the closest thing to a representative PR in this set, but the cluster is too heterogeneous to form a true duplicate family.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "11407 is the most representative of the recurring docs/cleanup theme, but it still doesn\u2019t capture the unrelated feature, pipeline, and test-removal PRs in the cluster.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:10938", "right": "pull_request:11983", "accept": false, - "reason": "Different docs pages with different fixes: evaluation guidance vs installation command wording." + "reason": "Both are docs-only, but one updates evaluation guidance and the other fixes installation instructions; unrelated changes." }, { "left": "pull_request:10938", "right": "pull_request:9760", "accept": false, - "reason": "Both touch docs, but one updates evaluation.md and the other edits the root README; no same underlying change." + "reason": "Evaluation-doc update vs README heading formatting; same file type, different content and intent." }, { "left": "pull_request:11096", "right": "pull_request:12193", "accept": false, - "reason": "Both are README/doc updates, but for different models and different instructions; not a duplicate change." + "reason": "Flux TPU README performance wording vs Qwen DreamBooth script name correction; unrelated model/docs edits." }, { "left": "pull_request:11236", "right": "pull_request:11237", "accept": false, - "reason": "Unrelated edits: SanaSprint heading rename versus dtype-map formatting in loading docs." + "reason": "One renames a Sana pipeline doc heading, the other fixes a dtype-map example; different docs and issues." }, { "left": "pull_request:11248", "right": "pull_request:11983", "accept": false, - "reason": "One fixes a Flux ControlNet README typo; the other updates installation instructions. Different artifacts and problems." + "reason": "Flux ControlNet README code fix vs installation guide update; no shared bug or code path." }, { "left": "pull_request:11248", "right": "pull_request:9760", "accept": false, - "reason": "Different README files and unrelated content; no shared bug or change." + "reason": "A README code formatting fix and a general README title tweak are not the same change." }, { "left": "pull_request:11249", "right": "pull_request:11251", "accept": false, - "reason": "Both are ControlNet-related, but they fix different pipelines and different code paths, so they are not mergeable as one PR." + "reason": "Both touch ControlNet, but one adds XLab controlnet support in Flux image-to-image while the other adds SD3 multi-controlnet inpainting support; distinct features and code paths." }, { "left": "pull_request:11303", "right": "pull_request:11983", "accept": false, - "reason": "Autoencoder class-name correction in model docs vs installation.md command fix; unrelated." + "reason": "Autoencoder docs class-name fix vs installation docs; unrelated doc corrections." }, { "left": "pull_request:11303", "right": "pull_request:9760", "accept": false, - "reason": "Different documentation targets and different corrections; not the same underlying issue." + "reason": "Autoencoder doc correction vs README formatting cleanup; different topics." }, { "left": "pull_request:11407", "right": "pull_request:12265", "accept": false, - "reason": "Typo cleanup sweep versus modular pipeline/runtime fixes; too broad and not the same change." + "reason": "Both are typo sweeps, but they target different file sets and unrelated edits; not the same underlying fix." }, { "left": "pull_request:11416", "right": "pull_request:12265", "accept": false, - "reason": "Docs/comments typo cleanup is not the same as modular code changes across pipelines and hooks." + "reason": "Docs/comments typo cleanup vs broader typo cleanup across code and notebooks; too general to be the same change." }, { "left": "pull_request:11476", "right": "pull_request:12265", "accept": false, - "reason": "String/comment typo fixes and test updates do not match the modular pipeline/code changes in 12265." + "reason": "Another typo-fix sweep, but the affected files and exact edits differ; not mergeable as one PR." }, { "left": "pull_request:1162", "right": "pull_request:1555", "accept": false, - "reason": "Both involve community pipeline docs/code, but one fixes links and the other fixes cache-dir handling; different problems." + "reason": "Community links cleanup in docs vs a cache-dir bug in dynamic module loading; different problems." }, { "left": "pull_request:1162", "right": "pull_request:2257", "accept": false, - "reason": "Community pipeline link cleanup is unrelated to adding the UnCLIP text interpolation pipeline." + "reason": "Docs link fix vs adding a new community pipeline feature; unrelated." }, { "left": "pull_request:1162", "right": "pull_request:3063", "accept": false, - "reason": "Link/docs cleanup versus pipeline __setattr__ behavior fix; different code paths." + "reason": "Docs link fix vs pipeline __setattr__ bug fix; different code paths and intents." }, { "left": "pull_request:1162", "right": "pull_request:3266", "accept": false, - "reason": "Community README links and dynamic-module version loading are separate issues." + "reason": "Docs link fix vs dynamic module version-loading bug; not the same issue." }, { "left": "pull_request:11746", "right": "pull_request:12560", "accept": false, - "reason": "License-year update across docs is unrelated to modular pipeline refactors and node-utils changes." + "reason": "License-year refresh vs modular pipeline refactor; unrelated maintenance work." }, { "left": "pull_request:11838", "right": "pull_request:11879", "accept": false, - "reason": "Both concern deprecated pipelines, but one adds warnings in docs while the other removes tests; not the same concrete change." + "reason": "Deprecated-pipeline docs warnings vs removing tests for deprecated pipelines; related theme, but not the same change." }, { "left": "pull_request:11847", "right": "pull_request:12068", "accept": false, - "reason": "Wan single-file example fix is unrelated to Qwen docs/example corrections." + "reason": "Wan single_file example fix vs Qwen docs corrections; different models and docs." }, { "left": "pull_request:11847", "right": "pull_request:12487", "accept": false, - "reason": "Specific Wan example cleanup versus a broad broken-links fix; different underlying problems." + "reason": "Wan example fix vs broad broken-link cleanup across many docs; separate doc tasks." }, { "left": "pull_request:11857", "right": "pull_request:12337", "accept": false, - "reason": "Flux docs reference updates and xformers enablement touch the same area, but they fix different issues and code paths." + "reason": "Flux docs reference update vs a Flux xformers runtime fix; one is docs-only, the other changes code behavior." }, { "left": "pull_request:11875", "right": "pull_request:11958", "accept": false, - "reason": "Kontext auto-pipeline work and adding the F5 TTS pipeline are entirely different features." + "reason": "Kontext auto-pipeline addition vs F5 TTS pipeline addition; different features and codebases." }, { "left": "pull_request:11983", "right": "pull_request:12890", "accept": false, - "reason": "Installation.md command correction and Wan docs cleanup are unrelated docs changes." + "reason": "Installation command correction vs removing hfoptions markup in wan docs; unrelated docs edits." }, { "left": "pull_request:1218", "right": "pull_request:98", "accept": false, - "reason": "Both are conversion scripts, but for different model families and different conversion logic." + "reason": "Different conversion scripts for different model families; not the same migration or bug fix." }, { "left": "pull_request:12183", "right": "pull_request:12984", "accept": false, - "reason": "QwenImage docs button addition and Flux2 Klein docs additions are unrelated model docs updates." + "reason": "QwenImage LoRA-button docs update vs Flux2 Klein docs addition; unrelated documentation changes." }, { "left": "pull_request:12247", "right": "pull_request:12307", "accept": false, - "reason": "AutoGuidance validation fix and modular-pipeline updates are not the same bug or change." + "reason": "AutoGuidance validation fix vs a broad modular cleanup PR; different scope and code." }, { "left": "pull_request:12247", "right": "pull_request:13115", "accept": false, - "reason": "AutoGuidance validation and modular Flux block naming/import fixes are separate changes." + "reason": "Specific AutoGuidance validation bug vs generic modular pipeline fixes; not the same underlying problem." }, { "left": "pull_request:12360", "right": "pull_request:12902", "accept": false, - "reason": "Chroma docs wording and distributed inference device_map usage are unrelated documentation fixes." + "reason": "Chroma docs wording update vs distributed inference device_map fix; unrelated." }, { "left": "pull_request:12360", "right": "pull_request:13135", "accept": false, - "reason": "Chroma guidance docs and LTX2 image-to-video docstring corrections are different pages and issues." + "reason": "Chroma docs update vs LTX2 docstring fix; different pipelines and edits." }, { "left": "pull_request:12487", "right": "pull_request:12716", "accept": false, - "reason": "Both fix links, but on different docs and with different target URLs; not a duplicate PR." + "reason": "General broken-link cleanup vs Flux2 link correction; both docs, but not the same concrete issue." }, { "left": "pull_request:12560", "right": "pull_request:13148", "accept": false, - "reason": "Modular pipeline code cleanup and modular docs expansion are related only broadly, not the same concrete change." + "reason": "Different modular-doc/refactor updates touching different structures; too broad to be duplicates." }, { "left": "pull_request:1269", "right": "pull_request:1527", "accept": false, - "reason": "Both touch stable-diffusion inpaint-related code, but one is a refactor and the other changes input handling semantics." + "reason": "Stable Diffusion refactor vs inpaint tensor-input support; distinct pipeline behavior changes." }, { "left": "pull_request:12731", "right": "pull_request:12797", "accept": false, - "reason": "Bria FIBO docs and Kandinsky docs are separate model pages with different corrections." + "reason": "Bria Fibo minor docs fixes vs Kandinsky docs fixes; unrelated model documentation." } ] }, @@ -122747,7 +122794,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 53, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -122839,10 +122886,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -122857,8 +122904,8 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a true duplicate cluster: the items span unrelated docs edits, pipeline/model additions, scheduler bug fixes, and test refactors across many subsystems. None of the soft edges look like the same underlying change or bug.", - "confidence": 0.98, + "summary": "This is not a true duplicate cluster: it mixes unrelated docs fixes, test refactors, scheduler/code bugfixes, and conversion-script changes that only share broad maintenance language. No single PR clearly represents the same underlying change across the cluster.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -122868,181 +122915,181 @@ "left": "pull_request:12883", "right": "pull_request:13135", "accept": false, - "reason": "Different docs-only fixes: a torchao typo in quantization docs vs LTX2 image-to-video docstring corrections." + "reason": "Both are small docs edits, but they target different pipelines and different content; no shared bug or change." }, { "left": "pull_request:12984", "right": "pull_request:13158", "accept": false, - "reason": "Different documentation updates: Flux2 Klein API docs/source change vs QwenImage layered pipeline docs." + "reason": "Unrelated documentation additions for different pipelines; same docs area is too broad to merge as duplicates." }, { "left": "pull_request:12984", "right": "pull_request:13567", "accept": false, - "reason": "Unrelated scope: pipeline docs/code for Flux2 Klein vs repository agent instruction docs." + "reason": "Klein docs vs repo agent-doc updates are unrelated changes." }, { "left": "pull_request:13253", "right": "pull_request:13368", "accept": false, - "reason": "Different test refactors for different model families: Z-Image transformer vs AutoencoderKL." + "reason": "Different test refactors for different model families; superficial CI-style similarity only." }, { "left": "pull_request:13253", "right": "pull_request:13369", "accept": false, - "reason": "Different test refactors for different model families: Z-Image transformer vs AutoencoderDC." + "reason": "Different model test suites and different refactor goals; not the same underlying bug/change." }, { "left": "pull_request:13254", "right": "pull_request:13371", "accept": false, - "reason": "Different test refactors: LTX transformer tests vs AutoencoderKLWan tests." + "reason": "LTX transformer tests and WAN autoencoder tests are unrelated refactors." }, { "left": "pull_request:13315", "right": "pull_request:13371", "accept": false, - "reason": "Different models and test suites: Sana transformer tests vs AutoencoderKLWan tests." + "reason": "Different test modules for different architectures; no shared concrete fix." }, { "left": "pull_request:13335", "right": "pull_request:13371", "accept": false, - "reason": "Different models and test suites: Cosmos transformer tests vs AutoencoderKLWan tests." + "reason": "Cosmos transformer test refactor vs WAN autoencoder test refactor are separate maintenance changes." }, { "left": "pull_request:1341", "right": "pull_request:178", "accept": false, - "reason": "Different changes: add Stable Diffusion Repaint pipeline vs Stable Diffusion pipeline test updates." + "reason": "Stable Diffusion RePaint implementation vs stable-diffusion pipeline test update are unrelated." }, { "left": "pull_request:136", "right": "pull_request:43", "accept": false, - "reason": "Packaging manifest fix vs adding the model card template are related only loosely, but they are distinct changes and not one bug fix." + "reason": "Manifest packaging fix and model-card-template addition are different changes." }, { "left": "pull_request:1380", "right": "pull_request:3074", "accept": false, - "reason": "Different underlying issues: Transformer2DModel runtime bug vs docstring-only correction." + "reason": "One fixes a Transformer2DModel code path, the other only corrects a docstring; not duplicate fixes." }, { "left": "pull_request:1386", "right": "pull_request:1505", "accept": false, - "reason": "Different scheduler changes: DDIM/Euler v-prediction support vs broader v-prediction support in other schedulers." + "reason": "Both relate to v-prediction, but they touch different scheduler sets and appear as separate feature expansions, not one duplicate PR." }, { "left": "pull_request:1408", "right": "pull_request:61", "accept": false, - "reason": "Different conversion scripts and targets: Stable Diffusion sample-size conversion vs LatentDiffusionUncond conversion script." + "reason": "Different conversion scripts for different model families; no shared bug or change." }, { "left": "pull_request:1408", "right": "pull_request:63", "accept": false, - "reason": "Different conversion scripts and targets: Stable Diffusion sample-size conversion vs BDDM conversion script." + "reason": "Sample-size conversion fix vs BDDM conversion script are unrelated converter changes." }, { "left": "pull_request:1418", "right": "pull_request:1763", "accept": false, - "reason": "Different subsystems: Versatile Diffusion test/API fixes vs attention-mask handling in attention.py." + "reason": "Versatile Diffusion tests/docs fix vs attention-mask behavior fix in attention.py are separate issues." }, { "left": "pull_request:1487", "right": "pull_request:1627", "accept": false, - "reason": "Docs changes for xformers vs an implementation change enabling xformers support in modeling_utils." + "reason": "Docs-only xformers updates versus actual model helper changes; same topic, different code-path and intent." }, { "left": "pull_request:1487", "right": "pull_request:3549", "accept": false, - "reason": "Both docs-only, but for unrelated topics: xformers docs vs qualitative evaluation / prompt dataset docs." + "reason": "Both are docs PRs, but for different subjects and pipelines; not the same change." }, { "left": "pull_request:1510", "right": "pull_request:98", "accept": false, - "reason": "Different scopes: scheduler/pipeline conversion improvements vs a conversion script for NCSN++ models." + "reason": "Different conversion-tooling PRs for different model types and scripts; not duplicates." }, { "left": "pull_request:1582", "right": "pull_request:2415", "accept": false, - "reason": "RePaint scheduler bug fix vs pipeline_utils import/version guard; unrelated code paths." + "reason": "RePaint scheduler fix and pipeline_utils import/version handling are unrelated." }, { "left": "pull_request:1582", "right": "pull_request:2795", "accept": false, - "reason": "RePaint scheduler bug fix vs LoRA training script behavior; different features entirely." + "reason": "RePaint bugfix vs LoRA training-script tweak are unrelated changes." }, { "left": "pull_request:1583", "right": "pull_request:2415", "accept": false, - "reason": "Legacy inpaint pipeline fix vs pipeline_utils import/version guard; different problems." + "reason": "Stable Diffusion inpaint-legacy bugfix vs pipeline_utils version logic are unrelated." }, { "left": "pull_request:1583", "right": "pull_request:2795", "accept": false, - "reason": "Legacy inpaint pipeline bug fix vs LoRA training script adjustment; not the same issue." + "reason": "Different codebases and different problems; no duplicate overlap." }, { "left": "pull_request:1601", "right": "pull_request:3183", "accept": false, - "reason": "Different bugs in different areas: classifier-free guidance handling vs DreamBooth LoRA checkpoint registration." + "reason": "Pipeline classifier-free-guidance bugfix vs DreamBooth LoRA checkpointing fix are unrelated." }, { "left": "pull_request:168", "right": "pull_request:299", "accept": false, - "reason": "Major Stable Diffusion pipeline addition vs PNDM timestep calculation hotfix; not mergeable as one change." + "reason": "Stable Diffusion pipeline introduction and a later PNDM hotfix are not the same underlying change." }, { "left": "pull_request:168", "right": "pull_request:904", "accept": false, - "reason": "Stable Diffusion pipeline addition vs inpainting pipeline implementation; related domain, but different features and code paths." + "reason": "Broad Stable Diffusion pipeline work vs inpainting implementation are separate features." }, { "left": "pull_request:1701", "right": "pull_request:3079", "accept": false, - "reason": "Different pipelines and purposes: RePaint API/test updates vs attend-and-excite fast test tuning." + "reason": "RePaint API/test cleanup and attend-and-excite fast-test tuning are unrelated test/pipeline changes." }, { "left": "pull_request:1761", "right": "pull_request:1769", "accept": false, - "reason": "Both UnCLIP-related, but one adds CPU offloading and the other only adjusts offload tests; not the same concrete change." + "reason": "Both concern UnCLIP, but one adds CPU offloading in code while the other only adjusts test settings and thresholds." }, { "left": "pull_request:178", "right": "pull_request:5", "accept": false, - "reason": "Stable diffusion pipeline test update vs adding the Latent Diffusion UNet model; unrelated changes." + "reason": "Stable Diffusion pipeline test update vs latent-diffusion UNet implementation are unrelated." }, { "left": "pull_request:178", "right": "pull_request:971", "accept": false, - "reason": "Stable Diffusion pipeline test update vs a community bit diffusion example/README addition; different artifacts." + "reason": "Pipeline test changes and community bit-diffusion example/docs are unrelated." }, { "left": "pull_request:190", "right": "pull_request:3", "accept": false, - "reason": "Making Stable Diffusion import work without scipy is unrelated to the hub-import/config loading changes in PR 3." + "reason": "SciPy-import guard in a pipeline is unrelated to the from-hub configuration/import infrastructure fix." } ] }, @@ -123058,7 +123105,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 54, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -123150,10 +123197,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -123168,192 +123215,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a heterogeneous mix of unrelated docs tweaks, typo fixes, test updates, and a few feature additions; none of the proposed soft pairs look like true duplicate PRs.", - "confidence": 0.98, - "canonical_issue_reason": "No issue is present in the cluster.", - "canonical_pr_reason": "No single PR is a good canonical for duplication: the items span unrelated changes across docs, examples, tests, and core code.", - "best_issue_reason": "No issue is present in the cluster.", - "best_pr_reason": "No clear best PR exists; any anchor would be arbitrary because the cluster is not one underlying bug/change.", + "summary": "This is mostly a grab-bag of unrelated docs, typo, and test PRs that only look superficially similar. None of the provided soft edges are true duplicate/change matches; the only obvious near-duplicate in the broader set is the pair of identical Moreh textual inversion PRs, but that pair is not one of the listed soft edges.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 3554 is the best representative PR here because it is the earlier copy of the only near-identical duplicate-style change in the set (3554/3555), and there are no issue artifacts to anchor on.", + "best_issue_reason": null, + "best_pr_reason": "PR 3554 is the strongest standalone PR candidate for the cluster: it is the earliest clear duplicate-like submission and is more representative than the many unrelated docs/typo/test PRs around it.", "soft_edge_verdicts": [ { "left": "pull_request:2077", "right": "pull_request:3068", "accept": false, - "reason": "Different scope: one adds validation-image support to textual inversion training, the other is mostly docs cleanup/expansion for textual inversion and loaders." + "reason": "Textual-inversion validation-image support vs docs finishing textual inversion; different change scope and code path." }, { "left": "pull_request:2077", "right": "pull_request:3554", "accept": false, - "reason": "Not the same change set: 2077 is a small example feature update, while 3554 is a much broader forked contribution touching multiple training scripts and setup files." + "reason": "Validation-image feature work is unrelated to the Moreh textual inversion environment/script bundle." }, { "left": "pull_request:2077", "right": "pull_request:3555", "accept": false, - "reason": "Same as 3554: similar title wording, but the actual edited files and change scope are different from 2077." + "reason": "Validation-image feature work is unrelated to the Moreh textual inversion environment/script bundle." }, { "left": "pull_request:2086", "right": "pull_request:428", "accept": false, - "reason": "Both are docs-related, but they document different features: LoRA support vs general using-diffusers/custom pipeline docs." + "reason": "LoRA docs addition is not the same as the broad docs restructuring in Using Diffusers." }, { "left": "pull_request:21", "right": "pull_request:72", "accept": false, - "reason": "Different code paths and purposes: model tests for UNet/grad_tts utils vs 1D up/downsample block tests." + "reason": "Different test additions for different model blocks; they do not fix the same underlying behavior." }, { "left": "pull_request:2118", "right": "pull_request:715", "accept": false, - "reason": "Unrelated fixes: a misspelled import-utils entry vs guarding testing-utils imports when PyTorch is absent." + "reason": "One fixes an import-utils spelling entry; the other guards testing-utils imports when PyTorch is absent." }, { "left": "pull_request:2137", "right": "pull_request:2715", "accept": false, - "reason": "Typos only make these look similar; one is a small loaders.py doc fix, the other is a wide typo sweep across many model/pipeline docs." + "reason": "Loaders typo/doc cleanup vs a broad typo sweep across many modules; not the same concrete change." }, { "left": "pull_request:2137", "right": "pull_request:3118", "accept": false, - "reason": "Completely different changes: loaders.py wording fixes vs a ControlNet pipeline API breaking-change fix." + "reason": "Loaders typo/doc cleanup vs ControlNet default-argument fix; unrelated code paths." }, { "left": "pull_request:2138", "right": "pull_request:948", "accept": false, - "reason": "Both are tiny wording fixes, but in different docs/code areas with no shared underlying bug." + "reason": "Docs typo in loaders vs comment typo in DDPM pipeline; unrelated artifacts." }, { "left": "pull_request:2175", "right": "pull_request:3474", "accept": false, - "reason": "Separate documentation additions: diffusion pipeline components vs the new AttnProcessor API page." + "reason": "Diffusion pipeline docs update vs attention-processor docs page; separate documentation additions." }, { "left": "pull_request:2193", "right": "pull_request:2952", "accept": false, - "reason": "Related to the same SAG feature, but not the same concrete change: 2193 adds the pipeline; 2952 only updates docs and links." + "reason": "SAG pipeline introduction vs later SAG docs update; the latter is a follow-up, not the same PR-worthy change." }, { "left": "pull_request:22", "right": "pull_request:72", "accept": false, - "reason": "Different artifacts and concerns: a logging typo in modeling_utils vs new 1D layer tests." + "reason": "Different test targets and different code areas; not the same bug or change." }, { "left": "pull_request:2208", "right": "pull_request:910", "accept": false, - "reason": "Different documentation topics: xFormers installation guidance vs Stable Diffusion inpainting license/docs update." + "reason": "xFormers installation docs vs inpainting pipeline launch docs; unrelated content." }, { "left": "pull_request:2278", "right": "pull_request:299", "accept": false, - "reason": "No overlap in underlying change: a large video diffusion draft vs a scheduler math refactor." + "reason": "Video Stable Diffusion project work vs a PNDM scheduler hotfix; completely different changes." }, { "left": "pull_request:2290", "right": "pull_request:3971", "accept": false, - "reason": "Unrelated docs fixes in different areas: InstructPix2Pix snippet formatting vs SDXL documentation workflow/link changes." + "reason": "Pix2Pix doc formatting fix vs SDXL docs/workflow changes; unrelated." }, { "left": "pull_request:2320", "right": "pull_request:260", "accept": false, - "reason": "Different typos in unrelated files: stable-diffusion checkpoint conversion vs an examples README spelling fix." + "reason": "Stable Diffusion checkpoint-loading typo fix vs README typo fix; different files and behavior." }, { "left": "pull_request:2320", "right": "pull_request:353", "accept": false, - "reason": "Not the same bug or change; one fixes an inference config filename typo, the other corrects a UNet block error message." + "reason": "Checkpoint-loading typo fix vs UNet block error-message typo; unrelated." }, { "left": "pull_request:2320", "right": "pull_request:407", "accept": false, - "reason": "Unrelated fixes in separate code paths: conversion helper typo vs attention-slicing method name." + "reason": "Checkpoint-loading typo fix vs attention-slicing typo fix; different code paths." }, { "left": "pull_request:2320", "right": "pull_request:798", "accept": false, - "reason": "Different file and behavior: ckpt conversion typo vs docstring formatting in UNet2DCondition." + "reason": "Checkpoint-loading typo fix vs UNet2DCondition docstring typo; unrelated." }, { "left": "pull_request:2329", "right": "pull_request:3389", "accept": false, - "reason": "Both touch tests/workflows, but they are for different efforts: push-test refactor vs attention-refactor integration testing." + "reason": "Workflow refactor vs attention-refactor integration-test PR; not the same concrete change." }, { "left": "pull_request:2433", "right": "pull_request:3549", "accept": false, - "reason": "Different doc content: an img2img SDEdit note vs evaluation guidance about prompt datasets." + "reason": "SDEdit note in img2img docs vs qualitative evaluation docs; unrelated documentation changes." }, { "left": "pull_request:2446", "right": "pull_request:421", "accept": false, - "reason": "Separate doc fixes: InstructPix2Pix example code vs scheduler conceptual docs." + "reason": "Pix2Pix docs snippet correction vs scheduler docs typo fix; separate topics." }, { "left": "pull_request:2446", "right": "pull_request:4275", "accept": false, - "reason": "Different documentation pages and issues; both are docs, but not the same concrete change." + "reason": "Pix2Pix docs snippet correction vs AutoPipeline docs fix; unrelated docs pages." }, { "left": "pull_request:2539", "right": "pull_request:3063", "accept": false, - "reason": "Different bugs: a doc default value correction in Pix2Pix Zero vs a pipeline __setattr__ None-handling bug." + "reason": "Pix2Pix Zero doc default-value correction vs pipeline __setattr__ behavior fix; not the same bug." }, { "left": "pull_request:2540", "right": "pull_request:3572", "accept": false, - "reason": "Both mention ControlNet, but one corrects docs text and the other changes the inpainting example and narrative." + "reason": "ControlNet docs correction vs inpainting ControlNet example rewrite; different docs changes." }, { "left": "pull_request:2566", "right": "pull_request:407", "accept": false, - "reason": "Different typos in unrelated areas: tensorboard tracker usage vs attention slicing method naming." + "reason": "TensorBoard tracker typo fix vs attention slicing typo fix; unrelated behavior." }, { "left": "pull_request:2566", "right": "pull_request:798", "accept": false, - "reason": "No shared underlying issue; training script tracker selection vs a UNet docstring typo." + "reason": "TensorBoard tracker typo fix vs a docstring typo in UNet2DCondition; unrelated." }, { "left": "pull_request:2587", "right": "pull_request:3474", "accept": false, - "reason": "Translated docs rollout vs attention-processor API docs; unrelated changes." + "reason": "Chinese translation docs PR vs attention-processor docs page; different documentation sets." }, { "left": "pull_request:2604", "right": "pull_request:4856", "accept": false, - "reason": "Both are link fixes, but they target different pages and destinations, so they are not duplicates." + "reason": "Colab link fix in index vs SDXL API-to-guide link fix; not the same change." }, { "left": "pull_request:2624", "right": "pull_request:3474", "accept": false, - "reason": "Pipeline from_pretrained docs addition is unrelated to the new AttnProcessor documentation page." + "reason": "Pipeline cache_dir doc addition vs attention-processor docs page; unrelated docs content." } ] }, @@ -123369,7 +123416,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 55, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -123467,10 +123514,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -123485,222 +123532,222 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a grab bag of unrelated doc/typo/test PRs. None of the soft-edge candidates look like the same underlying change, so all should be rejected.", - "confidence": 0.93, + "summary": "This cluster is mostly a pile of unrelated small documentation edits, test tweaks, and a few independent code fixes. The soft matches are largely title/topic similarity rather than the same underlying change, so they should stay separate.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR emerges here: the items span unrelated documentation edits, test adjustments, and code fixes, and the soft-pair candidates do not share a concrete bug or change.", + "canonical_pr_reason": "No single canonical PR stands out: the items span unrelated docs nits, test adjustments, and code fixes, with only superficial similarity in wording or subsystem.", "best_issue_reason": null, - "best_pr_reason": "PR 3460 is the strongest representative: it fixes a concrete inpaint pipeline code path, adds the needed typing/import updates, and includes test coverage.", + "best_pr_reason": "No PR is a good global duplicate representative here; the strongest-looking pairs are still different files or different concrete behaviors, so there is no shared mergeable change to anchor the cluster.", "soft_edge_verdicts": [ { "left": "pull_request:2639", "right": "pull_request:2869", "accept": false, - "reason": "Different docs pages with separate punctuation edits; not the same change." + "reason": "Both are tiny docs edits, but they update different pages with different wording fixes; same style, not the same change." }, { "left": "pull_request:2639", "right": "pull_request:2911", "accept": false, - "reason": "Unrelated documentation updates in different API pages." + "reason": "Different documentation files and unrelated sentence punctuation fixes." }, { "left": "pull_request:2639", "right": "pull_request:2912", "accept": false, - "reason": "Different docs files; both are minor punctuation fixes, not one underlying issue." + "reason": "Different docs page; only the PR titles are similarly generic." }, { "left": "pull_request:2639", "right": "pull_request:2913", "accept": false, - "reason": "Unrelated docs pages; only share a typo-style edit." + "reason": "One edits basic_training.mdx, the other stable_unclip.mdx; unrelated documentation changes." }, { "left": "pull_request:2639", "right": "pull_request:2922", "accept": false, - "reason": "Different API docs pages and separate wording fixes." + "reason": "Different docs targets with no shared underlying bug or content." }, { "left": "pull_request:2639", "right": "pull_request:2924", "accept": false, - "reason": "Unrelated docs edits in different pipeline pages." + "reason": "Different docs pages; both are minor prose fixes only." }, { "left": "pull_request:2639", "right": "pull_request:2925", "accept": false, - "reason": "Different scheduler overview docs; not the same bug or change." + "reason": "Same generic docs-update pattern, but separate files and separate edits." }, { "left": "pull_request:2639", "right": "pull_request:2926", "accept": false, - "reason": "Separate docs pages with independent typo cleanup." + "reason": "Different scheduler docs page; not the same underlying change." }, { "left": "pull_request:2639", "right": "pull_request:2927", "accept": false, - "reason": "Different docs files; same style of edit is not enough." + "reason": "Unrelated docs edits in different API pages." }, { "left": "pull_request:2639", "right": "pull_request:2929", "accept": false, - "reason": "Unrelated documentation updates." + "reason": "Different documentation file and different typo cleanup." }, { "left": "pull_request:2639", "right": "pull_request:2930", "accept": false, - "reason": "Different scheduler doc pages, same punctuation cleanup only." + "reason": "Same superficial 'update docs' theme, but distinct file and wording." }, { "left": "pull_request:2639", "right": "pull_request:2931", "accept": false, - "reason": "Different docs file and content; not the same issue." + "reason": "Different scheduler docs page; no shared concrete fix." }, { "left": "pull_request:2639", "right": "pull_request:2932", "accept": false, - "reason": "Separate docs pages with minor prose cleanup." + "reason": "Different docs page; both are punctuation/formatting nits only." }, { "left": "pull_request:2639", "right": "pull_request:2933", "accept": false, - "reason": "Different docs pages; no shared underlying bug." + "reason": "Separate docs pages with unrelated copy fixes." }, { "left": "pull_request:2639", "right": "pull_request:2935", "accept": false, - "reason": "Different docs files; both are small wording fixes." + "reason": "Different docs file and different sentence-level correction." }, { "left": "pull_request:2639", "right": "pull_request:2936", "accept": false, - "reason": "Unrelated docs updates in different scheduler pages." + "reason": "Different docs page; no shared code path or content change." }, { "left": "pull_request:2639", "right": "pull_request:2937", "accept": false, - "reason": "Different docs pages; same typo-class edit only." + "reason": "Different scheduler doc, only similarly small prose edits." }, { "left": "pull_request:2639", "right": "pull_request:2938", "accept": false, - "reason": "Different docs pages with independent wording fixes." + "reason": "Different docs page with unrelated wording cleanup." }, { "left": "pull_request:2639", "right": "pull_request:2939", "accept": false, - "reason": "Unrelated documentation edits, not one concrete change." + "reason": "Different API docs page; not the same underlying issue." }, { "left": "pull_request:2701", "right": "pull_request:3458", "accept": false, - "reason": "ONNXRuntime support vs inpaint LoRA support; completely different code paths." + "reason": "One adds ONNX Runtime ROCm support; the other adds inpaint LoRA/cross-attention support. Different features, different code paths." }, { "left": "pull_request:2701", "right": "pull_request:3460", "accept": false, - "reason": "ONNXRuntime import support is unrelated to the inpaint pipeline change." + "reason": "ONNXRuntime backend support is unrelated to inpaint LoRA scale support in Stable Diffusion." }, { "left": "pull_request:2715", "right": "pull_request:568", "accept": false, - "reason": "Both touch typos, but they are different files and different cleanup scopes." + "reason": "Both are broad typo fixes, but they touch different files and different kinds of edits; not the same concrete change." }, { "left": "pull_request:2760", "right": "pull_request:3318", "accept": false, - "reason": "Different textual inversion-related work: one adds quantization distillation docs/code, the other is an IF textual inversion WIP." + "reason": "Both mention textual inversion, but one is a quantization/distillation example and the other is a WIP IF textual inversion feature; different implementations and goals." }, { "left": "pull_request:2767", "right": "pull_request:3359", "accept": false, - "reason": "LoRA training script arg cleanup vs disabling safety checker in another script; not the same fix." + "reason": "One fixes `use_peft` logic in the LoRA trainer; the other disables safety checker in a different training script." }, { "left": "pull_request:2792", "right": "pull_request:3449", "accept": false, - "reason": "Skipping MPS tests is unrelated to the workflow rework and ONNX test removal." + "reason": "Both involve tests/CI on MPS, but one skips specific tests while the other reworks broader MPS and ONNX test workflows; not one change." }, { "left": "pull_request:2808", "right": "pull_request:3755", "accept": false, - "reason": "Different flaky-test relaxations in different test modules; not the same failure." + "reason": "Both relax tests, but they cover different models and different failure modes." }, { "left": "pull_request:2839", "right": "pull_request:4472", "accept": false, - "reason": "Different README updates for different example areas." + "reason": "Different README updates with no shared bug or change set." }, { "left": "pull_request:2873", "right": "pull_request:3744", "accept": false, - "reason": "Docs phrasing cleanup vs workflow token rename; unrelated changes." + "reason": "Both are docs cleanups, but one changes phrasing in contributing docs and the other updates a workflow secret name." }, { "left": "pull_request:2894", "right": "pull_request:824", "accept": false, - "reason": "Adding a flax requirement is unrelated to the flax pipeline test addition." + "reason": "Adding a Flax requirement is unrelated to adding a Flax test harness/change." }, { "left": "pull_request:299", "right": "pull_request:309", "accept": false, - "reason": "PNDM scheduler hotfix vs README link fix; no shared code path." + "reason": "The hotfix in PNDM scheduler code is unrelated to the README notebook link fix." }, { "left": "pull_request:3012", "right": "pull_request:3820", "accept": false, - "reason": "A comment typo in textual inversion vs an LDM3D pipeline/doc change; unrelated." + "reason": "A comment typo fix is not the same as the LDM3D pipeline/docs update." }, { "left": "pull_request:3063", "right": "pull_request:3345", "accept": false, - "reason": "Pipeline __setattr__ bugfix vs README pipeline-class edit; different artifacts and problems." + "reason": "The pipeline `__setattr__` bugfix is unrelated to a README example class rename." }, { "left": "pull_request:3132", "right": "pull_request:3345", "accept": false, - "reason": "Score SDE pipeline type annotation fix vs README example change; unrelated." + "reason": "A type annotation fix in the Score SDE pipeline does not match a README example change." }, { "left": "pull_request:3132", "right": "pull_request:972", "accept": false, - "reason": "Type typo in score SDE pipeline vs torch_dtype README typo; not the same change." + "reason": "Unrelated: score_sde pipeline typing versus a Stable Diffusion README typo fix." }, { "left": "pull_request:3132", "right": "pull_request:975", "accept": false, - "reason": "Pipeline type fix vs docs title capitalization; unrelated." + "reason": "Different docs/config file and different kind of change; no duplicate basis." } ] }, @@ -123716,7 +123763,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 56, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -123807,10 +123854,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -123825,180 +123872,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated docs/typo updates plus a few distinct code fixes; none of the soft edges look like the same underlying change.", + "summary": "The soft links are mostly superficial overlaps (docs/typos/shared filenames) rather than the same underlying bug or change. I would reject all candidate duplicate pairs.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No true duplicate PR stands out across this heterogeneous set. If one PR must anchor the cluster, 5707 is the most substantive, concrete, and well-tested code fix (LoRA loading behavior across pipelines), unlike the many isolated documentation edits.", + "canonical_pr_reason": "No issue artifacts are present. If a PR must be chosen as the anchor, pull_request:3460 is the strongest representative because it is a merged, self-contained behavioral fix with accompanying tests and a clear concrete code path.", "best_issue_reason": null, - "best_pr_reason": "5707 is the best representative PR: it addresses a specific bug path, touches real runtime code, and includes broad test coverage. It is more substantial and canonical than the many small docs/typo PRs in this set.", + "best_pr_reason": "pull_request:3460 is the best PR anchor among this set: it has a focused functional change, tests, and a clear problem statement, unlike the mostly docs-only or loosely related fixes.", "soft_edge_verdicts": [ { "left": "pull_request:3183", "right": "pull_request:957", "accept": false, - "reason": "Different bugs in different scripts: DreamBooth checkpoint registration vs. a scheduler init parameter cleanup in text-to-image training." + "reason": "Both are bugfix PRs, but they fix different code paths: DreamBooth LoRA checkpointing vs DDPMScheduler args in text-to-image training." }, { "left": "pull_request:3345", "right": "pull_request:5568", "accept": false, - "reason": "One is a docs-only pipeline-class example fix; the other fixes custom pipeline loading logic in code. Not the same change." + "reason": "One is a README example class typo; the other fixes custom pipeline loading logic. Same broad area, not the same bug." }, { "left": "pull_request:3446", "right": "pull_request:4352", "accept": false, - "reason": "A warning-message wording fix in pipeline_utils is unrelated to Kandinsky combined pipeline feature additions." + "reason": "A warning-message wording fix is unrelated to Kandinsky combined-pipeline functionality." }, { "left": "pull_request:3458", "right": "pull_request:5250", "accept": false, - "reason": "Inpaint LoRA/cross-attention support is unrelated to adding a DDPO docs entry." + "reason": "Inpaint LoRA scale support and DDPO documentation are unrelated changes." }, { "left": "pull_request:3460", "right": "pull_request:5250", "accept": false, - "reason": "This is a concrete inpaint pipeline API change; the other is a docs-only DDPO page addition." + "reason": "Same as above: inpaint LoRA support vs DDPO docs, no shared underlying change." }, { "left": "pull_request:3577", "right": "pull_request:4626", "accept": false, - "reason": "Both are docs, but they fix different link targets in different pages and are not the same underlying issue." + "reason": "Both touch docs links, but they update different pages and different broken links." }, { "left": "pull_request:3641", "right": "pull_request:4352", "accept": false, - "reason": "Inpainting latent handling bugfix is unrelated to Kandinsky combined pipeline offload/xformers additions." + "reason": "Initial-latents inpainting fix is unrelated to Kandinsky combined-pipeline method additions." }, { "left": "pull_request:3709", "right": "pull_request:4853", "accept": false, - "reason": "VAE image processor docs addition is unrelated to an actual image_processor width bug fix in code." + "reason": "Adding image-processor docs is unrelated to the image-processor width bug fix." }, { "left": "pull_request:3710", "right": "pull_request:3821", "accept": false, - "reason": "Both add docs pages, but for different languages and different content; no shared bug or change." + "reason": "Different documentation content and locale: EN image-processor docs vs Korean translation of another guide." }, { "left": "pull_request:3710", "right": "pull_request:5286", "accept": false, - "reason": "The English image processor doc addition and the Japanese docs addition are separate localization/doc tasks." + "reason": "These are different localized docs additions, not the same artifact or bug." }, { "left": "pull_request:3786", "right": "pull_request:3894", "accept": false, - "reason": "PR template addition is unrelated to Flax UNet dropout implementation." + "reason": "PR template addition is unrelated to Flax UNet dropout behavior." }, { "left": "pull_request:3786", "right": "pull_request:4172", "accept": false, - "reason": "A PR template file is unrelated to Flax model type-casting changes." + "reason": "PR template addition is unrelated to Flax type-casting changes." }, { "left": "pull_request:3819", "right": "pull_request:5454", "accept": false, - "reason": "A README FAQ update is unrelated to a schedulers docs wording correction." + "reason": "An FAQ README edit and a schedulers docs wording fix are unrelated." }, { "left": "pull_request:3923", "right": "pull_request:5983", "accept": false, - "reason": "Both touch docs about reproducibility/usage, but one adds a torch import and the other changes an SVD tutorial link/formatting; not the same fix." + "reason": "Torch import fix in Stable Diffusion docs is unrelated to SVD documentation formatting." }, { "left": "pull_request:3971", "right": "pull_request:4721", "accept": false, - "reason": "SDXL docs/build workflow fixes and Korean DreamBooth docs edits are unrelated changes." + "reason": "SDXL docs/workflow fixes and Korean DreamBooth docs changes are unrelated." }, { "left": "pull_request:3971", "right": "pull_request:6339", "accept": false, - "reason": "The SDXL docs/build workflow change is unrelated to the AnimateDiff doc example update." + "reason": "Docs build / SDXL doc fixes are unrelated to AnimateDiff docs wording updates." }, { "left": "pull_request:3985", "right": "pull_request:5640", "accept": false, - "reason": "SDXL doc refinements are unrelated to AnimateDiff documentation nits." + "reason": "SDXL documentation tweaks and AnimateDiff documentation tweaks are different changes." }, { "left": "pull_request:4012", "right": "pull_request:5640", "accept": false, - "reason": "Dance Diffusion doc corrections are unrelated to AnimateDiff docs." + "reason": "Dance Diffusion doc wording changes are unrelated to AnimateDiff doc nits." }, { "left": "pull_request:408", "right": "pull_request:710", "accept": false, - "reason": "General typo cleanup across docs is unrelated to fixing a malformed docstring in two model files." + "reason": "General docs typo cleanup is unrelated to the specific docstring formatting fix." }, { "left": "pull_request:4275", "right": "pull_request:4965", "accept": false, - "reason": "AutoPipeline docs corrections are unrelated to Wuerstchen docs formatting changes." + "reason": "AutoPipeline docs corrections and W\u00fcrstchen docs formatting are different documentation changes." }, { "left": "pull_request:4320", "right": "pull_request:5505", "accept": false, - "reason": "ONNX runtime docs wording fixes are unrelated to ControlNet API naming fixes plus loader updates." + "reason": "ONNX docs wording and ControlNet naming / loader changes are not the same fix." }, { "left": "pull_request:4350", "right": "pull_request:6339", "accept": false, - "reason": "T2I adapter doc precision fix is unrelated to AnimateDiff doc updates." + "reason": "T2I adapter docs dtype fix and AnimateDiff docs updates are unrelated." }, { "left": "pull_request:4472", "right": "pull_request:6342", "accept": false, - "reason": "DreamBooth README edits and an installation note for Python 3.12 on Windows are separate documentation changes." + "reason": "SDXL README cleanup and installation.md note are different documentation updates." }, { "left": "pull_request:4596", "right": "pull_request:6311", "accept": false, - "reason": "An SDXL conversion script is unrelated to the Diffusion Fast tutorial docs." + "reason": "SDXL conversion script and diffusion-fast tutorial are completely different additions." }, { "left": "pull_request:4746", "right": "pull_request:6592", "accept": false, - "reason": "These are different inpainting efforts: one is a large experimental WIP training/inference branch, the other is a new inpainting training script." + "reason": "Both mention inpainting, but one is a broad WIP SDXL inpainting training refactor and the other adds a separate inpainting training script; not the same concrete change." }, { "left": "pull_request:4822", "right": "pull_request:5505", "accept": false, - "reason": "Both touch loaders.py, but one fixes a LoRA warning string while the other fixes custom pipeline loading and API naming; not the same bug." + "reason": "Both touch loaders/docs, but they address different typos and different functionality." }, { "left": "pull_request:4869", "right": "pull_request:5707", "accept": false, - "reason": "Combined-pipeline tests are not the same as the LCM loader/pipeline changes; shared area only, not the same change." + "reason": "Combined-pipeline tests and LCM tests are different test additions for different features." }, { "left": "pull_request:4887", "right": "pull_request:6787", "accept": false, - "reason": "Removing a latents argument from a Kandinsky pipeline is unrelated to a marigold depth-estimation input normalization bug." + "reason": "Kandinsky prior-emb2emb latent-input removal and Marigold input-range normalization are unrelated bugs." } ] }, @@ -124014,7 +124061,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 57, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -124105,10 +124152,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -124123,180 +124170,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab-bag of unrelated small PRs across docs, tests, training scripts, pipelines, and scheduler/model fixes. There is no clear duplicate pair or single underlying bug/change tying them together, so all soft-edge candidates should be rejected.", - "confidence": 0.98, + "summary": "A noisy cluster of mostly unrelated docs, tests, and feature PRs across different subsystems; the shared similarity is mostly generic wording, not the same underlying bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items are heterogeneous and do not share one concrete code-path problem or change.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No single best PR; the cluster mixes unrelated docs, tests, feature additions, and bug fixes, so none is a strong representative duplicate target.", + "best_pr_reason": "PR 6847 is the strongest single representative here because it is the most substantial and heavily reviewed change, but it still addresses a distinct IP-Adapter masking feature rather than the same change as the others.", "soft_edge_verdicts": [ { "left": "pull_request:5", "right": "pull_request:555", "accept": false, - "reason": "UNet-for-LDM feature addition vs Flax VAE import/model addition; different artifacts and code paths." + "reason": "Different model components and scopes: UNet for latent diffusion vs Flax VAE implementation/fix." }, { "left": "pull_request:5", "right": "pull_request:889", "accept": false, - "reason": "New UNet model implementation vs adding DiffusionPipeline components docs/API; unrelated changes." + "reason": "UNet model addition vs diffusion pipeline API/docs components exposure; not the same code-path change." }, { "left": "pull_request:5051", "right": "pull_request:7181", "accept": false, - "reason": "Separate docs typo fixes in unrelated files; no shared underlying bug." + "reason": "Both are typo/docs cleanup, but they fix different text across different files and are not one concrete duplicate change." }, { "left": "pull_request:5184", "right": "pull_request:5707", "accept": false, - "reason": "Scheduler add_noise test coverage vs LCM loader/pipeline changes; not the same change or bug." + "reason": "Scheduler test augmentation vs LCM feature/test additions; different test targets and intent." }, { "left": "pull_request:5267", "right": "pull_request:7150", "accept": false, - "reason": "README formatting/autodoc docs update vs callback example code snippet fix; unrelated docs work." + "reason": "Unrelated documentation edits in different pages and contexts." }, { "left": "pull_request:5332", "right": "pull_request:6322", "accept": false, - "reason": "DreamBooth LoRA description typo vs training script attention/scheduler changes; different issues." + "reason": "DreamBooth LoRA typo fix vs training-script attention/scheduler changes; different fixes." }, { "left": "pull_request:5454", "right": "pull_request:7150", "accept": false, - "reason": "Schedulers docs link fix vs callback docs example update; unrelated documentation edits." + "reason": "Broken scheduler docs link vs callback example formatting; unrelated documentation issues." }, { "left": "pull_request:5533", "right": "pull_request:5686", "accept": false, - "reason": "Different test suites for different pipelines (SDXL mixture vs Shap-E); not duplicate failures." + "reason": "Different fast-test updates for different pipelines; not the same bug or mergeable PR." }, { "left": "pull_request:555", "right": "pull_request:574", "accept": false, - "reason": "Both touch VAE Flax, but one adds the model and the other fixes a decode assignment; not the same bug/change." + "reason": "Same general Flax VAE area, but 555 adds the model/subfolder plumbing while 574 tweaks decode behavior; not the same concrete fix." }, { "left": "pull_request:5640", "right": "pull_request:6466", "accept": false, - "reason": "AnimateDiff doc nits vs SVD doc cleanup; unrelated doc-only changes." + "reason": "AnimateDiff doc nits vs SVD doc edits; unrelated docs." }, { "left": "pull_request:5707", "right": "pull_request:5723", "accept": false, - "reason": "LCM tests vs LCM docs; same topic but not the same underlying issue or code-path." + "reason": "LCM test/loader changes vs LCM documentation; different artifacts and code paths." }, { "left": "pull_request:5855", "right": "pull_request:7150", "accept": false, - "reason": "Community README typo vs callback example fix; unrelated docs." + "reason": "Community README typo fix vs callback example formatting; unrelated docs changes." }, { "left": "pull_request:5901", "right": "pull_request:6847", "accept": false, - "reason": "ControlNet IP-Adapter support vs IP-Adapter attention masking; different features and merge surfaces." + "reason": "Different IP-Adapter work: ControlNet img2img integration vs attention masking feature; not one concrete PR." }, { "left": "pull_request:5901", "right": "pull_request:6941", "accept": false, - "reason": "Different ControlNet pipeline variants and distinct IP-Adapter support work; not one concrete fix." + "reason": "Different pipelines and features: ControlNet img2img IP-Adapter support vs SDXL ControlNet inpaint support." }, { "left": "pull_request:5982", "right": "pull_request:6174", "accept": false, - "reason": "SDXL Turbo docs addition vs LoRA training docs typo; unrelated." + "reason": "SDXL Turbo docs vs LoRA docs typo; unrelated documentation fixes." }, { "left": "pull_request:6002", "right": "pull_request:6475", "accept": false, - "reason": "SVD guidance-scale logic fix vs SVD VAE type docs correction; different problems." + "reason": "Stable Video Diffusion guidance-scale code fix vs VAE type documentation; different layers entirely." }, { "left": "pull_request:6207", "right": "pull_request:6401", "accept": false, - "reason": "Training script cleanup vs LoRA PEFT test simplification; different files and purposes." + "reason": "Removing deprecated training-script code vs LoRA PEFT test cleanup; different targets." }, { "left": "pull_request:6215", "right": "pull_request:6217", "accept": false, - "reason": "PixArt test initialization fix vs Stable Video Diffusion torch.compile fix; unrelated test/behavior changes." + "reason": "PixArt test init changes vs Stable Video Diffusion torch.compile fix; unrelated test/code paths." }, { "left": "pull_request:6314", "right": "pull_request:6453", "accept": false, - "reason": "Separate paper-link typo fixes for different docs pages; not duplicates." + "reason": "Two separate docs typo/link fixes for different papers and pages." }, { "left": "pull_request:6360", "right": "pull_request:7963", "accept": false, - "reason": "Padding mask crop support for inpaint pipelines vs custom timesteps/sigmas for controlnet inpaint/img2img; related area but different feature work." + "reason": "Both touch inpaint pipelines, but one adds padding_mask_crop while the other adds custom timesteps/sigmas; different features." }, { "left": "pull_request:6419", "right": "pull_request:6752", "accept": false, - "reason": "WIP in-context-learning community pipeline vs PromptDiffusion pipeline; unrelated community examples." + "reason": "WIP prompt-diffusion community pipeline vs finalized PromptDiffusion community pipeline; not the same artifact." }, { "left": "pull_request:6433", "right": "pull_request:8427", "accept": false, - "reason": "New community pipeline vs workflow debug logging; completely unrelated." + "reason": "Community pipeline addition vs workflow debug logging; no shared bug/change." }, { "left": "pull_request:6475", "right": "pull_request:7747", "accept": false, - "reason": "SVD VAE type documentation fix vs generic VAE tiling test assertion fix; not the same issue." + "reason": "SVD VAE type doc fix vs generic VAE tiling test fix; unrelated." }, { "left": "pull_request:6511", "right": "pull_request:7932", "accept": false, - "reason": "Checkpointing with torch.compile in training scripts vs LoRA test expectation updates; unrelated." + "reason": "Torch.compile checkpointing support vs LoRA test expectation updates; different code paths." }, { "left": "pull_request:6633", "right": "pull_request:7353", "accept": false, - "reason": "Checkpoint config reading fix vs adding FromSingleFileMixin to a pipeline; different code paths." + "reason": "Checkpoint YAML loading fix vs ControlNet SDXL img2img loader mixin wiring; unrelated." }, { "left": "pull_request:6760", "right": "pull_request:6847", "accept": false, - "reason": "Adjusting slow IP-Adapter tests vs implementing IP-Adapter attention masking; tests are not the feature bug itself." + "reason": "Both involve IP-Adapter, but one adjusts test expectations while the other adds attention masking; not the same concrete change." }, { "left": "pull_request:6775", "right": "pull_request:8851", "accept": false, - "reason": "DDPMScheduler docs parameter addition vs AuraFlow docs/model additions; unrelated." + "reason": "Scheduler docs parameter addition vs AuraFlow docs/model/pipeline addition; unrelated." }, { "left": "pull_request:6787", "right": "pull_request:6904", "accept": false, - "reason": "Marigold input range bug vs load_image PIL support; different utilities and fixes." + "reason": "Marigold input-range bug fix vs load_image PIL support; different utility bugs." } ] }, @@ -124312,7 +124359,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 58, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -124402,10 +124449,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -124420,181 +124467,184 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No true duplicates here: the soft pairs are mostly superficial title/file overlap across different bugs, features, docs, CI, and test-only changes.", - "confidence": 0.96, + "summary": "This cluster is mostly a set of unrelated false-positive similarity pairs across different subsystems (memory offloading, WAN, Flux, SD3 training, docs/tests, CI, etc.). There is no single coherent duplicate target for the whole cluster; only one pair looks like a plausible same-bug docstring fix.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR fits this cluster; the items are heterogeneous and do not share one underlying change.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No single PR is a good global representative; the similarities are too broad or incidental to justify one canonical PR.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:6961", "right": "pull_request:7797", "accept": false, - "reason": "One is single-file loader/test scaffolding; the other reduces DDPM fast-test model size. Different changes, different bug classes." + "reason": "Different work: single-file loader test scaffolding vs DDPM fast-test model downsizing." }, { "left": "pull_request:6961", "right": "pull_request:7803", "accept": false, - "reason": "Same as above: unrelated single-file loader work vs DDIM fast-test reduction." + "reason": "Different work: single-file loader tests vs DDIM fast-test model downsizing." }, { "left": "pull_request:6961", "right": "pull_request:7804", "accept": false, - "reason": "Unrelated: single-file loader/test changes versus AMUSED fast-test model downsizing." + "reason": "Different work: single-file loader tests vs Amused fast-test model downsizing." }, { "left": "pull_request:6961", "right": "pull_request:7833", "accept": false, - "reason": "Single-file loader/tests versus AudioLDM test size reduction; only both are test-related in a broad sense." + "reason": "Different work: single-file loader tests vs AudioLDM fast-test model downsizing." }, { "left": "pull_request:710", "right": "pull_request:798", - "accept": false, - "reason": "Both fix docstring formatting, but they touch different UNet classes and are separate typo/doc cleanup changes." + "accept": true, + "reason": "Same underlying docstring typo fix (`r\"\"\"`) in closely related UNet model files; these could plausibly be combined." }, { "left": "pull_request:7150", "right": "pull_request:9266", "accept": false, - "reason": "Different docs topics: callback example update vs community projects list entry." + "reason": "Unrelated docs edits: callback example cleanup vs community projects list update." }, { "left": "pull_request:737", "right": "pull_request:839", "accept": false, - "reason": "Both remove use_auth_token in examples, but they are separate docs/example cleanups, not one underlying bug." + "reason": "Both remove `use_auth_token`, but one is broad repo cleanup and the other is a DreamBooth README-only tweak." }, { "left": "pull_request:7410", "right": "pull_request:7919", "accept": false, - "reason": "FreeInit behavior fix is unrelated to offload behavior under device_map; different subsystems and failure modes." + "reason": "Different bugs: FreeInit progress-bar/refactor changes vs offload behavior when `hf_device_map` is enabled." }, { "left": "pull_request:7638", "right": "pull_request:8303", "accept": false, - "reason": "Both are documentation typos, but one is a CONTRIBUTING link update and the other is philosophical wording; no shared bug." + "reason": "Both docs typos, but unrelated files and wording fixes with no shared concrete change." }, { "left": "pull_request:7977", "right": "pull_request:8642", "accept": false, - "reason": "Both are typo/doc fixes in different Stable Diffusion docs/pipeline files; not the same change." + "reason": "Different docs typos in different Stable Diffusion 3 pages; not the same change." }, { "left": "pull_request:806", "right": "pull_request:971", "accept": false, - "reason": "DreamBooth EMA training work is unrelated to the community README/community pipeline PR." + "reason": "DreamBooth EMA training change vs adding a community bit-diffusion example." }, { "left": "pull_request:822", "right": "pull_request:91", "accept": false, - "reason": "Flax scheduler API update is a broad scheduler change; the older scheduler API cleanup is much larger and unrelated in scope." + "reason": "Both are scheduler/API related, but they fix different generation stacks and different concrete code paths." }, { "left": "pull_request:8478", "right": "pull_request:9408", "accept": false, - "reason": "HunyuanDiT ControlNet/pipeline addition vs Flux ControlNet pipelines; same broad area, but different models and concrete code paths." + "reason": "Different feature additions: HunyuanDiT ControlNet vs Flux ControlNet img2img/inpaint." }, { "left": "pull_request:8498", "right": "pull_request:8642", "accept": false, - "reason": "Different doc typos in different Stable Diffusion 3 docs; no shared underlying bug." + "reason": "Different typo fixes in different Stable Diffusion 3 docs files; not the same artifact." }, { "left": "pull_request:8591", "right": "pull_request:8755", "accept": false, - "reason": "General SD3 training utility refactor vs a specific text-encoder training bug in DreamBooth LoRA; not mergeable as one fix." + "reason": "Different SD3 training changes: utility refactor vs text-encoder training bug fix." }, { "left": "pull_request:8631", "right": "pull_request:9473", "accept": false, - "reason": "Single-file T5 checkpoint return bug versus Flux DreamBooth optimizer LR fix; unrelated code paths." + "reason": "Different fixes in different single-file / Flux training contexts; no shared concrete bug." }, { "left": "pull_request:8775", "right": "pull_request:9266", "accept": false, - "reason": "A README Colab link update and a community projects table update are separate documentation edits." + "reason": "README Colab link update vs community projects table update; unrelated docs changes." }, { "left": "pull_request:9266", "right": "pull_request:9553", "accept": false, - "reason": "Community projects listing update vs distributed inference device-map tip; different docs pages and purposes." + "reason": "Different docs updates: community project listing vs distributed inference device-map tip." }, { "left": "pull_request:9591", "right": "pull_request:9592", "accept": false, - "reason": "Both are docstring/doc cleanup PRs, but for different files and APIs; no evidence of one underlying issue." + "reason": "Unrelated documentation refactors in different modules." }, { "left": "pull_request:9699", "right": "pull_request:9832", "accept": false, - "reason": "New SSH runner option vs adding quantization nightly tests; CI-related but not the same change." + "reason": "Different CI additions: runner availability vs quantization nightly test job." }, { "left": "pull_request:12004", "right": "pull_request:12562", "accept": false, - "reason": "Wan2.2 model support/convert script work is unrelated to the context-parallel crash fix in a Wan model." + "reason": "Same Wan family, but one adds Wan2.2 support while the other fixes a CP crash in a specific Wan model." }, { "left": "pull_request:11081", "right": "pull_request:11682", "accept": false, - "reason": "Both concern group offloading, but one adds record_stream and the other adds disk/NVMe offload support; distinct features and code paths." + "reason": "Same group-offloading area, but one adds `record_stream` and the other adds disk/NVMe offload; distinct changes." }, { "left": "pull_request:7550", "right": "pull_request:8992", "accept": false, - "reason": "Different community pipeline additions: SDXL Differential Diffusion vs HunyuanDiT differential diffusion." + "reason": "Both are community diffusion pipelines, but they target different models and different features." }, { "left": "pull_request:6868", "right": "pull_request:6946", "accept": false, - "reason": "Both touch IP-Adapter loading, but one introduces image_embeds call support and the other low_cpu_mem_usage loading; different changes." + "reason": "Both touch IP-Adapter loading, but one adds image-embed plumbing and the other adds low-CPU-memory loading support." }, { "left": "pull_request:12551", "right": "pull_request:13039", "accept": false, - "reason": "CI skipping for Sana layerwise casting tests vs Wan CPU upcasting fix; unrelated pipelines and issues." + "reason": "Different Wan-related fixes: CI skip for Sana layerwise casting vs CPU upcasting fix in Wan Animate." }, { "left": "pull_request:12970", "right": "pull_request:12988", "accept": false, - "reason": "Both are context-parallel related, but one changes Qwen-Image CP compatibility and the other removes a divisibility assertion for a different crash condition." + "reason": "Both are context-parallel fixes, but one is Qwen-Image-specific and the other is a general seq-len divisibility crash fix." }, { "left": "pull_request:11835", "right": "pull_request:12069", "accept": false, - "reason": "Both are test-failure fixes, but for different pipelines (float16 CUDA vs AudioLDM2 transformers-version handling)." + "reason": "Different test fixes in different pipelines; one is float16 CUDA tests, the other is AudioLDM2 cache-position handling." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The false-edge verdicts are well supported by the cited diffs, and the lone accepted pair (710/798) is plausibly the same docstring typo fix across related UNet files. No pair is overstated as mergeable beyond what the diffs show." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -124603,7 +124653,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 59, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -124691,10 +124741,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -124709,150 +124759,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose lexical bucket of unrelated PRs across pipelines, LoRA, docs, CI, and examples. I don\u2019t see any pair that fixes the same concrete bug/change, so the soft edges should all be rejected.", + "summary": "No duplicate cluster emerges here: the items span unrelated fixes across pipelines, docs, CI, and LoRA utilities. Even the similar-looking pairs usually touch different concrete code paths or different products, so they should stay separate.", "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 11786 is the broadest cross-cutting change in the set, touching CI plus several test files, so it is the best single representative if one must anchor the cluster.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:10581", "right": "pull_request:10717", "accept": false, - "reason": "Different pipelines and different adapter integration work; same broad IP-Adapter theme, but not the same bug or change." + "reason": "Different pipelines and fixes: SD3 inpaint IP-Adapter support vs Flux img2img negative embed ordering. Same broad area, not the same bug/change." }, { "left": "pull_request:3446", "right": "pull_request:4714", "accept": false, - "reason": "Unrelated fixes: a pipeline-loading warning message vs a missing-transformers import path bug." + "reason": "Both are small error-message/import adjustments, but they fix different issues in different code paths and are not mergeable as one change." }, { "left": "pull_request:4034", "right": "pull_request:4666", "accept": false, - "reason": "One adds LoRA unload utility support, the other adds SDXL ControlNet LoRA support; different code paths and behavior." + "reason": "LoRA unload utility vs SDXL ControlNet LoRA support are distinct features in different subsystems." }, { "left": "pull_request:4572", "right": "pull_request:4632", "accept": false, - "reason": "Same example script, but different concrete bugs: network_alpha naming vs checkpointing/casting issues." + "reason": "Both touch an SDXL training script, but one fixes network_alpha naming while the other fixes checkpointing/casting behavior; not the same underlying bug." }, { "left": "pull_request:4629", "right": "pull_request:4783", "accept": false, - "reason": "One is a docs/feature expansion around long weighted prompts; the other is a prompt-embedding correctness note. Not the same change." + "reason": "Docs/navigation cleanup vs SDXL prompt-embedding semantics. Overlap is only on SDXL docs, not the concrete change." }, { "left": "pull_request:4761", "right": "pull_request:5407", "accept": false, - "reason": "Different pipelines and issues: Paint by Example initialization vs mixed-precision training in unconditional generation." + "reason": "Paint-by-example encoder sizing fix is unrelated to unconditional training mixed-precision handling." }, { "left": "pull_request:5506", "right": "pull_request:6816", "accept": false, - "reason": "A PEFT+xformers attention fix is not the same as a DDP training-script fix, even though both mention LoRA/PEFT." + "reason": "Both are PEFT/LoRA-related, but one fixes xformers interaction in attention layers and the other fixes an SDXL Dreambooth training script. Different bugs." }, { "left": "pull_request:4944", "right": "pull_request:6946", "accept": false, - "reason": "General IP-Adapter support vs a loading-time optimization for IP-Adapter; related area, but distinct functionality." + "reason": "General IP-Adapter support work vs a specific low_cpu_mem_usage loading enhancement. Same subsystem, different change." }, { "left": "pull_request:10318", "right": "pull_request:9257", "accept": false, - "reason": "Flux LoRA test coverage vs a DreamBooth Flux script bug fix; different artifacts and bug surfaces." + "reason": "Flux LoRA integration tests vs Dreambooth Flux prompt/tokenization bug fix are unrelated." }, { "left": "pull_request:10911", "right": "pull_request:11786", "accept": false, - "reason": "A Lumina2 LoRA NaN test fix and a CI marker/workflow fix are unrelated." + "reason": "Lumina2 LoRA NaN test fix vs CI marker/workflow changes; no shared bug or code path." }, { "left": "pull_request:1215", "right": "pull_request:2062", "accept": false, - "reason": "DDIM scheduler generator handling and Paint by Example CPU offload are separate pipeline bugs." + "reason": "DDIM generator/eta bug fix is unrelated to paint-by-example CPU offload support." }, { "left": "pull_request:12714", "right": "pull_request:13313", "accept": false, - "reason": "Flux2 LoRA NaN test adjustments vs Kohya-to-Diffusers LoRA conversion/loading logic; different problems." + "reason": "Flux2 LoRA NaN test fix vs Klein LoRA conversion/loading support are different problems." }, { "left": "pull_request:13365", "right": "pull_request:13536", "accept": false, - "reason": "Both touch the Claude review workflow, but one enables fork review and the other changes PR-opening/editing behavior; not the same change." + "reason": "Both alter the Claude review workflow, but one enables fork reviews and the other allows opening PRs; separate behaviors and changes." }, { "left": "pull_request:2892", "right": "pull_request:4099", "accept": false, - "reason": "Upscaler input-embed validation and SDXL prompt-embed handling are distinct embedding bugs in different pipelines." + "reason": "Stable Diffusion upscaler input validation vs SDXL prompt embed handling are different pipeline bugs." }, { "left": "pull_request:2915", "right": "pull_request:4666", "accept": false, - "reason": "A test tolerance tweak for stable diffusion text inversion is not the same as enabling LoRA for SDXL ControlNet." + "reason": "A flaky test tolerance adjustment is unrelated to SDXL ControlNet LoRA support." }, { "left": "pull_request:3968", "right": "pull_request:4083", "accept": false, - "reason": "Broad SDXL cleanup/docs work vs low-precision VAE support; shared ecosystem, different code changes." + "reason": "Large SDXL improvement/docs sweep vs low-precision VAE support. They share a theme but not the same concrete change." }, { "left": "pull_request:4498", "right": "pull_request:5750", "accept": false, - "reason": "One is adding SDXL reference-attention functionality, the other is a documentation-only reference intro." + "reason": "One adds SDXL reference-image functionality, the other is only a technology intro doc. Not the same fix/change." }, { "left": "pull_request:4666", "right": "pull_request:5615", "accept": false, - "reason": "Adding SDXL ControlNet LoRA support and reverting SDXL adapter LoRA loading are opposite, but not the same fix." + "reason": "SDXL ControlNet LoRA enablement vs a revert in the T2I adapter pipeline; different code paths and goals." }, { "left": "pull_request:5015", "right": "pull_request:6330", "accept": false, - "reason": "Broken doc link fix vs SVD documentation rendering fix; unrelated documentation issues." + "reason": "Broken link fix in docs vs SVD video-rendering doc fix; both docs-only but unrelated." }, { "left": "pull_request:5217", "right": "pull_request:5852", "accept": false, - "reason": "PEFT import typo and PR-test workflow silent-behavior fix are unrelated CI/runtime issues." + "reason": "PEFT import typo fix vs CI silent-behavior fix in workflow config; different defects." }, { "left": "pull_request:7118", "right": "pull_request:8571", "accept": false, - "reason": "Minor docs typos vs a broad style/errata sweep; same category, but not a single duplicate change." + "reason": "Docs typos in IP-Adapter vs broad errata/style cleanup; not the same underlying issue." + }, + { + "left": "pull_request:7458", + "right": "pull_request:7893", + "accept": false, + "reason": "Python-3.12 test skipping vs Dockerfile Python 3.10 upgrade. Different maintenance changes." + }, + { + "left": "pull_request:9175", + "right": "pull_request:9770", + "accept": false, + "reason": "Flux ControlNet union support vs a specific Xlabs ControlNet inpainting adjustment. Same area, different capability." }, { "left": "pull_request:9506", "right": "pull_request:9676", "accept": false, - "reason": "A docs/import cleanup in one file and an EntryNotFoundError import fix in benchmarks are unrelated." + "reason": "Docs import/path fix and benchmark import fix are unrelated." }, { "left": "pull_request:11353", "right": "pull_request:11588", "accept": false, - "reason": "Wan FLF2V feature support and Wan LoRA I2V loading logic are different features, not the same code-path bug." + "reason": "Wan-FLF2V feature support vs Wan LoRA I2V support. Same model family, different concrete features and code paths." } ] }, @@ -124868,7 +124930,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 60, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -124956,10 +125018,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -124974,162 +125036,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a grab bag of unrelated PRs across docs, tests, pipelines, and LoRA loaders. A few pairs share files or subsystems, but they target different concrete bugs or feature changes, so none should be merged as duplicates.", - "confidence": 0.94, + "summary": "This is a heterogeneous cluster with only superficial overlap: most soft edges share a broad subsystem, file path, or test/docs context, but they do not fix the same concrete bug or implement the same change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR cleanly represents this cluster; the items span unrelated docs, tests, pipeline additions, and LoRA fixes.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "No global PR is a good canonical representative because the soft pairs are mostly superficial overlaps, not the same underlying change.", "soft_edge_verdicts": [ { "left": "pull_request:6573", "right": "pull_request:7016", "accept": false, - "reason": "Both touch IP-Adapter code, but 6573 adds multi-IP-Adapter support while 7016 refactors embed handling and skips loading the image encoder; different changes and code paths." + "reason": "Both touch IP-Adapter code, but one adds multi-adapter support while the other refactors embedding prep and image-encoder loading; different changes, not one bug." }, { "left": "pull_request:10359", "right": "pull_request:9206", "accept": false, - "reason": "Both are test-only fixes, but they address different failing test suites and different failure modes." + "reason": "Both are test fixes, but they target different failures in different pipelines and xformers paths." }, { "left": "pull_request:11582", "right": "pull_request:11726", "accept": false, - "reason": "Same Wan area, but 11582 is the Wan VACE pipeline release and 11726 changes Wan LoRA conversion logic; not the same fix." + "reason": "Wan VACE pipeline work vs Wan LoRA conversion support are related only by project name, not the same code-path problem." }, { "left": "pull_request:1496", "right": "pull_request:2734", "accept": false, - "reason": "Both update Habana docs, but one adds the page and the other updates version/benchmark text; separate doc changes." + "reason": "Both are Habana docs, but one adds documentation and the other updates benchmark/version numbers; not duplicates." }, { "left": "pull_request:2001", "right": "pull_request:483", "accept": false, - "reason": "One is a ColossalAI example fix, the other is a broad typo cleanup/action; no shared bug." + "reason": "A ColossalAI example typo fix is unrelated to the repo-wide typo cleanup/action PR." }, { "left": "pull_request:2001", "right": "pull_request:978", "accept": false, - "reason": "Both are minor text/example edits, but they affect different examples and different codepaths." + "reason": "Example typo fix vs general typo cleanup across docs/code; different scope and intent." }, { "left": "pull_request:3716", "right": "pull_request:4695", "accept": false, - "reason": "ParaDiGMS and TextToVideoZeroSDXL are unrelated pipelines with distinct implementations." + "reason": "Both add pipelines, but they implement different pipelines with different code paths and features." }, { "left": "pull_request:4370", "right": "pull_request:4862", "accept": false, - "reason": "One fixes dummy import naming, the other fixes ControlNet SDXL img2img sizing; unrelated." + "reason": "Dummy import naming cleanup is unrelated to the SDXL ControlNet MultiControlNetModel bug fix." }, { "left": "pull_request:5275", "right": "pull_request:6795", "accept": false, - "reason": "Both involve SDXL-related examples, but one adds an SDXL Flax example and the other changes DreamBooth LoRA micro-conditioning; different features." + "reason": "SDXL JAX/TPU example and SDXL micro-conditioning training script are separate features, not one change." }, { "left": "pull_request:13029", "right": "pull_request:13120", "accept": false, - "reason": "Modular docs quickstart and modular loading tests are separate documentation/test work, not the same change." + "reason": "Modular docs quickstart and modular loading tests address different artifacts and problems." }, { "left": "pull_request:2223", "right": "pull_request:888", "accept": false, - "reason": "Semantic Guidance and CycleDiffusion are different pipeline additions." + "reason": "Both add pipelines, but CycleDiffusion and semantic guidance are distinct pipeline implementations." }, { "left": "pull_request:2959", "right": "pull_request:3675", "accept": false, - "reason": "A UNet3D test tweak and a UniDiffuser test fix are unrelated test issues." + "reason": "These are unrelated test tweaks for different models and failure modes." }, { "left": "pull_request:3446", "right": "pull_request:4373", "accept": false, - "reason": "Pipeline warning-message wording and SDXL resize logic are different problems." + "reason": "Pipeline loading warning-message cleanup is unrelated to the SDXL resize bug." }, { "left": "pull_request:3641", "right": "pull_request:4373", "accept": false, - "reason": "Inpainting latent handling and SDXL resize behavior are separate code-path fixes." + "reason": "Inpainting latent handling and SDXL resize are different code paths and issues." }, { "left": "pull_request:4714", "right": "pull_request:5788", "accept": false, - "reason": "Missing-transformers error handling and LCM timestep validation are unrelated scheduler/import fixes." + "reason": "Both improve error messages, but for different subsystems and failures." }, { "left": "pull_request:530", "right": "pull_request:689", "accept": false, - "reason": "Both touch tests, but one adjusts img2img assertions and the other changes attention implementation for slow tests; not the same issue." + "reason": "Both are test-related, but they fix different flaky/slow test issues." }, { "left": "pull_request:5506", "right": "pull_request:7246", "accept": false, - "reason": "PEFT/xFormers attention logic and VAE tiling test-device handling are unrelated." + "reason": "PEFT/xformers attention fix and VAE tiling device fix are unrelated." }, { "left": "pull_request:6265", "right": "pull_request:6757", "accept": false, - "reason": "GLIGEN refactor and k-diffusion sigma device placement are different pipeline internals." + "reason": "GLIGEN refactor and SDXL k-diffusion sigma handling are separate changes." }, { "left": "pull_request:6864", "right": "pull_request:7113", "accept": false, - "reason": "A PEFT docs note about torch.compile and Docker Slack notifications are wholly unrelated." + "reason": "A docs note about torch.compile and a Docker CI Slack notification are unrelated." }, { "left": "pull_request:8510", "right": "pull_request:9473", "accept": false, - "reason": "Both are DreamBooth training-script fixes, but they address different optimization/lr issues in different scripts and models." + "reason": "Both involve DreamBooth scripts, but they fix different learning-rate behaviors in different Flux vs non-Flux scripts." }, { "left": "pull_request:10911", "right": "pull_request:11124", "accept": false, - "reason": "Two separate test fixes for different models and failures." + "reason": "Lumina2 LoRA fuse_nan test and CogView4 test update are unrelated test changes." }, { "left": "pull_request:12584", "right": "pull_request:13229", "accept": false, - "reason": "SANA-Video pipeline launch and adding an LTX VAE are related by project, but they are distinct feature additions, not one duplicate change." + "reason": "Both concern Sana-Video, but one adds the pipeline while the other swaps in an LTX VAE; they are not the same change." }, { "left": "pull_request:10206", "right": "pull_request:11595", "accept": false, - "reason": "Both change LoRA loader code, but one fixes unload for Flux Control and the other fixes load_lora_weights behavior; opposite directions and different bugs." + "reason": "Both touch LoRA loading/unloading in Flux, but they fix different functions and different behaviors in the load/unload paths." }, { "left": "pull_request:10365", "right": "pull_request:9386", "accept": false, - "reason": "Torch/XLA support for Aura Flow and PAG inpaint support are unrelated pipeline features." + "reason": "Torch/XLA support for AuraFlow and PAG support for SD inpaint are unrelated pipeline additions." }, { "left": "pull_request:10648", "right": "pull_request:10885", "accept": false, - "reason": "One is a DDPM docs update for sigmoid scheduling, the other adds CogVideoX scheduler docs; different documentation topics." + "reason": "Both are docs PRs, but they document unrelated scheduler features." } ] }, @@ -125145,7 +125207,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 61, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -125233,10 +125295,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -125251,162 +125313,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of small docs, test, and code-fix PRs. The soft edges mostly reflect broad textual similarity or shared subsystems, but they do not appear to be the same underlying bug or change.", - "confidence": 0.97, + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs typo fixes, test adjustments, and distinct bug fixes across many pipelines and examples.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: the items span unrelated docs typo fixes, test cleanups, and distinct pipeline/runtime bugs, so there is no safe duplicate anchor for the whole cluster.", + "canonical_pr_reason": "No single canonical PR: the items are heterogeneous and the soft edges are only superficial title-level similarities.", "best_issue_reason": null, - "best_pr_reason": "PR 6848 is the strongest single representative: it has an explicit issue target and a concrete, narrowly scoped bug fix across the LCM distillation scripts. Even so, it is only representative, not a cluster-wide duplicate.", + "best_pr_reason": "PR 6848 is the strongest standalone code fix with an explicit issue target and a clear, concrete change, but it still does not represent the rest of the cluster.", "soft_edge_verdicts": [ { "left": "pull_request:11248", "right": "pull_request:11891", "accept": false, - "reason": "Both are tiny docs fixes, but they touch different guides and different typos in unrelated pipelines." + "reason": "Both are README/doc typo fixes, but they touch different examples and fix different text errors; not the same change." }, { "left": "pull_request:11303", "right": "pull_request:11891", "accept": false, - "reason": "One fixes an autoencoder doc class name; the other fixes a pipeline method typo in a different guide." + "reason": "Different docs files and different corrections: one fixes an autoencoder class name, the other fixes a method call typo." }, { "left": "pull_request:11473", "right": "pull_request:12018", "accept": false, - "reason": "Both are quantization docs edits, but they fix different wording/labels in different sections and are not the same change." + "reason": "Both are quantization docs edits, but the specific wording/label fixes are unrelated and do not share the same underlying bug." }, { "left": "pull_request:11755", "right": "pull_request:11835", "accept": false, - "reason": "One fixes an LTX cpu-offload test issue; the other fixes float16 CUDA test behavior. Same test area, different bugs." + "reason": "One fixes LTX cpu-offload support in a pipeline; the other changes float16 test behavior. Different code paths and purposes." }, { "left": "pull_request:11891", "right": "pull_request:12193", "accept": false, - "reason": "Different README typo fixes in different examples; no shared bug or code path." + "reason": "A Chroma docs typo fix and a Qwen DreamBooth script name correction are unrelated." }, { "left": "pull_request:12360", "right": "pull_request:12464", "accept": false, - "reason": "Both are docs-related, but one rewrites Chroma guidance text while the other fixes Marigold syntax/docs generation." + "reason": "Both are docs-oriented, but they target different pipeline docs and different issues; one is Chroma docs, the other is Marigold docs/code syntax cleanup." }, { "left": "pull_request:2004", "right": "pull_request:2029", "accept": false, - "reason": "Both are typo/cleanup PRs, but they address unrelated code/docs issues in different files." + "reason": "DreamBooth README wording changes are unrelated to code-side validation-message typo fixes in pipelines and schedulers." }, { "left": "pull_request:2339", "right": "pull_request:3471", "accept": false, - "reason": "DreamBooth docs appear in both, but one is checkpointing docs/tests and the other is a torch.compile note; different fixes." + "reason": "Both touch DreamBooth docs/script help text, but they fix different issues: checkpointing behavior vs a torch.compile note." }, { "left": "pull_request:2536", "right": "pull_request:652", "accept": false, - "reason": "A major VAE gradient-checkpointing feature/test change does not match a small Unet test tolerance tweak." + "reason": "Enabling VAE gradient checkpointing is a feature/code fix; lowering a test tolerance is a separate testing change." }, { "left": "pull_request:2559", "right": "pull_request:3226", "accept": false, - "reason": "Both touch docs for controlnet, but one is a link-format fix and the other adds new ControlNet v1.1 documentation." + "reason": "Both are docs formatting updates around ControlNet, but they are separate documentation fixes with different targets." }, { "left": "pull_request:2976", "right": "pull_request:5015", "accept": false, - "reason": "Both are docs link-navigation fixes, but they update different pages and different broken links." + "reason": "Both are link-navigation docs edits, but they correct different pages and different broken links." }, { "left": "pull_request:3128", "right": "pull_request:3214", "accept": false, - "reason": "Same broad img2img area, but one fixes timestep slicing across several pipelines while the other fixes attention head reshaping." + "reason": "Both are bug fixes, but one repairs timestep slicing across img2img pipelines while the other fixes attention reshape logic; not the same underlying bug." }, { "left": "pull_request:3506", "right": "pull_request:4248", "accept": false, - "reason": "Both involve training scripts, but one fixes checkpoint resume step scaling and the other fixes tracker config handling." + "reason": "Checkpoint-resume step scaling and accelerator tracker config handling are distinct training-script bugs." }, { "left": "pull_request:3642", "right": "pull_request:4457", "accept": false, - "reason": "Both involve community TensorRT inpaint material, but one adds the pipeline and the other makes minor corrections; not the same change." + "reason": "One adds a TensorRT inpainting pipeline; the other makes minor fixes to the existing community inpaint example. Same area, different change." }, { "left": "pull_request:4211", "right": "pull_request:4797", "accept": false, - "reason": "One is SDXL docstring cleanup, the other is LoRA loader logic for SDXL blocks; unrelated fix types." + "reason": "A docstring/example fix for SDXL inpainting is unrelated to the LoRA loader block-mapping change." }, { "left": "pull_request:4595", "right": "pull_request:6311", "accept": false, - "reason": "The WIP XL conversion effort and the fast diffusion tutorial add different content with no shared bug/change." + "reason": "A WIP conversion project and a new fast-diffusion tutorial are unrelated documentation/features." }, { "left": "pull_request:4707", "right": "pull_request:4934", "accept": false, - "reason": "Custom diffusion README edits and a new T2I-Adapter training feature are unrelated additions." + "reason": "Custom diffusion README training instructions and T2I-Adapter training docs/script are different example features, not one bug." }, { "left": "pull_request:5041", "right": "pull_request:5707", "accept": false, - "reason": "Both affect SDXL tests, but one adds refiner-only coverage while the other adds LCM tests across multiple pipelines." + "reason": "Both involve tests, but one adds SDXL refiner tests while the other adds LCM-related tests and loader fixes; different scopes and code paths." }, { "left": "pull_request:5807", "right": "pull_request:7627", "accept": false, - "reason": "A community long-prompt embedding enhancement is not the same issue as the Playground img2img latent-prep fix." + "reason": "Community LPW prompt-duplication support and Playground img2img latent normalization are unrelated fixes." }, { "left": "pull_request:5844", "right": "pull_request:7910", "accept": false, - "reason": "One adds a tutorial example; the other adds API docs for image processors. Same docs domain, different artifacts." + "reason": "A custom-pipeline overview example and image-processor API docs are different documentation additions." }, { "left": "pull_request:6002", "right": "pull_request:6389", "accept": false, - "reason": "Same Stable Video Diffusion file, but one fixes guidance-scale semantics and the other only changes the noise_aug_strength type." + "reason": "The first changes Stable Video Diffusion guidance-scale behavior; the second only corrects a parameter type annotation/docs text." }, { "left": "pull_request:6216", "right": "pull_request:7747", "accept": false, - "reason": "Both are test-related, but one patches PEFT LoRA key handling and the other fixes a VAE tiling assertion." + "reason": "A PEFT-specific LoRA test compatibility fix and a VAE tiling test fix are unrelated." }, { "left": "pull_request:6279", "right": "pull_request:6848", "accept": false, - "reason": "Both touch LCM distillation scripts, but they fix different bugs: guidance-scale embedding on multi-GPU vs target U-Net initialization." + "reason": "Both touch LCM distillation scripts, but they fix different bugs: guidance-scale embeddings vs target U-Net initialization." }, { "left": "pull_request:6680", "right": "pull_request:6962", "accept": false, - "reason": "Different test failures: DPM slow-test setup versus PixArt slow-test generator handling." + "reason": "A DPM-related slow-test adjustment and PixArt slow-test fixes are separate test issues." }, { "left": "pull_request:8272", "right": "pull_request:9100", "accept": false, - "reason": "Both are grammatical errata, but they change unrelated messages in different code paths and files." + "reason": "These are unrelated typo/grammar fixes in different files and different user-facing messages." } ] }, @@ -125422,7 +125484,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 62, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -125510,10 +125572,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -125699,7 +125761,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 63, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -125786,10 +125848,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -125975,7 +126037,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 64, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -126061,10 +126123,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -126244,7 +126306,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 65, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -126331,10 +126393,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -126520,7 +126582,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 66, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -126609,10 +126671,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -126804,7 +126866,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 67, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -126898,10 +126960,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -127135,7 +127197,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 68, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -127228,10 +127290,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -127447,7 +127509,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 69, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -127541,10 +127603,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -127766,7 +127828,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 70, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -127867,10 +127929,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -128134,7 +128196,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 71, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -128223,10 +128285,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -128418,7 +128480,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 72, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -128508,10 +128570,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -128709,7 +128771,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 73, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -128800,10 +128862,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -129007,7 +129069,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 74, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -129096,10 +129158,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -129297,7 +129359,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 75, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -129388,10 +129450,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -129601,7 +129663,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 76, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -129691,10 +129753,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -129898,7 +129960,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 77, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -129988,10 +130050,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -130189,7 +130251,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 78, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -130283,10 +130345,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -130511,7 +130573,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 79, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -130599,10 +130661,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -130794,7 +130856,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 80, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -130880,10 +130942,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -131063,7 +131125,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 81, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -131151,10 +131213,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -131340,7 +131402,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 82, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -131427,10 +131489,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -131610,7 +131672,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 83, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -131697,10 +131759,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -131886,7 +131948,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 84, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -131973,10 +132035,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -132162,7 +132224,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 85, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -132251,10 +132313,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -132446,7 +132508,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 86, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -132535,10 +132597,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -132742,7 +132804,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 87, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -132832,10 +132894,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -133033,7 +133095,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 88, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -133121,10 +133183,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -133310,7 +133372,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 89, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -133403,10 +133465,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -133622,7 +133684,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 90, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -133716,10 +133778,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -133947,7 +134009,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 91, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -134038,10 +134100,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -134245,7 +134307,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 92, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -134341,10 +134403,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -134584,7 +134646,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 93, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -134680,10 +134742,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -134698,216 +134760,216 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated docs fixes, CI/test refactors, and modular pipeline changes across different subsystems. I don\u2019t see a single underlying bug/change that ties the artifacts together, so I would not merge any of the soft-edge pairs.", - "confidence": 0.93, + "summary": "This cluster is a mixed bag of unrelated docs, test, and code cleanup PRs. The soft links are all superficial topical overlap, not the same underlying bug or change.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "No pull request cleanly represents a duplicate cluster here; the items span unrelated fixes rather than one shared code-path change.", + "canonical_pr_reason": "If a representative PR is needed, pull_request:12307 is the broadest substantive code change, but it still does not anchor the rest of the items as duplicates.", "best_issue_reason": null, - "best_pr_reason": "No single PR is a good global canonical for this set because the cluster is not centered on one concrete fix or feature.", + "best_pr_reason": "pull_request:12307 is the closest thing to a cluster anchor because it is a multi-file modular pipeline fix with real code impact; however, the cluster is too heterogeneous for a true duplicate canonical.", "soft_edge_verdicts": [ { "left": "pull_request:12247", "right": "pull_request:12272", "accept": false, - "reason": "Both are modular-pipeline fixes, but they target different bugs in different files: AutoGuidance validation vs Flux default height/width handling." + "reason": "Different fixes in different code paths: AutoGuidance validation vs FLUX modular pipeline defaults." }, { "left": "pull_request:1227", "right": "pull_request:2175", "accept": false, - "reason": "Both touch docs, but one adds multi-modal examples while the other adds a generic components docs entry; not the same change." + "reason": "Both are docs, but one adds modular docs and the other updates API docs; no shared change." }, { "left": "pull_request:1227", "right": "pull_request:2587", "accept": false, - "reason": "Unrelated docs additions in different sections and languages; no shared bug or feature." + "reason": "Unrelated documentation updates for different sections and languages." }, { "left": "pull_request:1227", "right": "pull_request:2624", "accept": false, - "reason": "One is multi-modal examples docs, the other is a cache_dir doc addition; different documentation targets." + "reason": "No same bug or feature; docs for community examples vs cache_dir docs." }, { "left": "pull_request:1227", "right": "pull_request:429", "accept": false, - "reason": "Both are documentation PRs, but they cover unrelated topics and code areas." + "reason": "Different documentation targets and different content areas." }, { "left": "pull_request:1227", "right": "pull_request:500", "accept": false, - "reason": "These are separate docs updates for different APIs and concepts; not duplicates." + "reason": "Both are docs changes, but about unrelated API/config topics." }, { "left": "pull_request:1230", "right": "pull_request:1751", "accept": false, - "reason": "Both address test/device behavior, but one fixes MPS generator fast tests while the other fixes UnCLIP generator propagation." + "reason": "Different runtime issues: MPS test setup vs UnCLIP generator handling." }, { "left": "pull_request:1230", "right": "pull_request:2362", "accept": false, - "reason": "Different test fixes in different pipelines; only broad CI/test similarity." + "reason": "Different test fixes for different pipelines and different failure modes." }, { "left": "pull_request:1230", "right": "pull_request:3079", "accept": false, - "reason": "Both are speed/test tweaks, but they address different pipeline tests and failure modes." + "reason": "Both touch tests, but they target different pipelines and unrelated flakiness/performance issues." }, { "left": "pull_request:12307", "right": "pull_request:12980", "accept": false, - "reason": "Same file appears, but the changes are unrelated: broad modular pipeline adjustments vs a Melllon param/docstring fix." + "reason": "Shared modular-pipeline area, but one is broad pipeline wiring and the other is Mellon param/docstring cleanup." }, { "left": "pull_request:12343", "right": "pull_request:12479", "accept": false, - "reason": "A VAE slicing bug fix and an environment constant casing fix are unrelated changes." + "reason": "Different bugs in different files: VAE slicing logic vs an environment constant parsing fix." }, { "left": "pull_request:12387", "right": "pull_request:12416", "accept": false, - "reason": "Both are support/docs around new features, but they implement different functionality for different models/subsystems." + "reason": "Different feature work: kernels flash-attention support vs QwenImage Edit Plus modular support." }, { "left": "pull_request:12487", "right": "pull_request:13040", "accept": false, - "reason": "Both are docs fixes, but one is link cleanup and the other is CLI warning grammar; not the same issue." + "reason": "Both are documentation/grammar fixes, but they affect unrelated pages and issues." }, { "left": "pull_request:12487", "right": "pull_request:13076", "accept": false, - "reason": "Different documentation errors in different pages; no shared concrete problem." + "reason": "Unrelated docs fixes: broken links vs quantization syntax correction." }, { "left": "pull_request:12487", "right": "pull_request:13181", "accept": false, - "reason": "One is broken links cleanup, the other is a torchrun command-order fix; unrelated." + "reason": "Different docs topics and different corrections." }, { "left": "pull_request:12552", "right": "pull_request:13040", "accept": false, - "reason": "Both touch wording, but they fix different docs pages and different typos/phrasing issues." + "reason": "Typos in different docs files; no single underlying problem." }, { "left": "pull_request:12552", "right": "pull_request:13076", "accept": false, - "reason": "Separate docs typo/syntax fixes in unrelated pages." + "reason": "Separate doc typo cleanup from a quantization config syntax fix." }, { "left": "pull_request:12552", "right": "pull_request:13181", "accept": false, - "reason": "No shared underlying documentation bug beyond both being minor doc edits." + "reason": "Different docs pages and unrelated edits." }, { "left": "pull_request:12618", "right": "pull_request:13294", "accept": false, - "reason": "A CI install typo and a flaky-test mitigation are different problems in different areas." + "reason": "CI workflow typo fix is unrelated to WanVACE flaky test adjustments." }, { "left": "pull_request:12675", "right": "pull_request:13135", "accept": false, - "reason": "Both are docstring/docs corrections, but for different pipelines and different mistakes." + "reason": "Different docstring/docs fixes for different pipelines." }, { "left": "pull_request:12731", "right": "pull_request:12890", "accept": false, - "reason": "Unrelated docs cleanups: Bria Fibo wording vs Wan hfoptions removal." + "reason": "Both edit docs, but FIBO guidance changes are unrelated to Wan docs cleanup." }, { "left": "pull_request:12757", "right": "pull_request:12890", "accept": false, - "reason": "Different docs areas and edits; one formats kernels references, the other edits Wan docs structure." + "reason": "Different docs topics: attention backend formatting vs Wan pipeline options cleanup." }, { "left": "pull_request:12775", "right": "pull_request:13158", "accept": false, - "reason": "Z-Image docs and QwenImageLayered docs are different model families." + "reason": "Z-Image docs addition is unrelated to QwenImageLayered docs." }, { "left": "pull_request:12775", "right": "pull_request:13253", "accept": false, - "reason": "Docs for Z-Image vs a test refactor for Z-Image Transformer are not the same concrete change." + "reason": "Docs addition vs test refactor; not the same change." }, { "left": "pull_request:12775", "right": "pull_request:13567", "accept": false, - "reason": "Z-Image docs and .ai agent guidance are unrelated." + "reason": "Z-Image docs are unrelated to agent docs / pipeline guidance updates." }, { "left": "pull_request:1282", "right": "pull_request:2836", "accept": false, - "reason": "Both are link fixes in docs, but they correct different broken links in different guides." + "reason": "Both are docs link fixes, but they concern different guides and different target links." }, { "left": "pull_request:12890", "right": "pull_request:12971", "accept": false, - "reason": "Different docs restructuring tasks in different pages." + "reason": "Wan markdown cleanup vs distributed inference section reordering are unrelated docs changes." }, { "left": "pull_request:12890", "right": "pull_request:13400", "accept": false, - "reason": "Wan docs cleanup and profiling README acknowledgment addition are unrelated." + "reason": "Different documentation pages and different content updates." }, { "left": "pull_request:12980", "right": "pull_request:13115", "accept": false, - "reason": "A Melllon param/docstring fix and Flux modular export changes are unrelated despite both touching modular pipeline code." + "reason": "Both are modular-pipeline related, but one is Mellon docstring cleanup and the other is Flux export/init changes." }, { "left": "pull_request:12984", "right": "pull_request:13382", "accept": false, - "reason": "Flux2 docs/code addition and modular conversion documentation are different changes." + "reason": "Flux2 Klein docs/code addition is unrelated to auto-docstring template documentation." }, { "left": "pull_request:12984", "right": "pull_request:13499", "accept": false, - "reason": "Flux2 Klein docs and a speed/memory docs mention are unrelated." + "reason": "Different docs/code areas: Flux2 Klein pipeline vs speed-memory optimization guidance." }, { "left": "pull_request:13082", "right": "pull_request:13368", "accept": false, - "reason": "Both are test refactors, but they target different model families and different test code paths." + "reason": "Wan model test refactor vs AutoencoderKL test refactor are different model families and test goals." }, { "left": "pull_request:13082", "right": "pull_request:13369", "accept": false, - "reason": "Different autoencoder test refactors; not the same underlying bug." + "reason": "Same general test-refactor theme, but different autoencoder targets and code paths." }, { "left": "pull_request:13174", "right": "pull_request:13344", "accept": false, - "reason": "Attention-backend tests and GLM image transformer tests are different test additions for different subsystems." + "reason": "Attention-backend test additions are unrelated to GLM image transformer tests." } ] }, @@ -134923,7 +134985,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 94, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -135012,10 +135074,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -135030,174 +135092,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a duplicate set; it mixes unrelated PRs across docs, version bumps, pipeline additions, scheduler/conversion changes, and test tweaks. No soft edge looks like the same underlying bug or mergeable change.", + "summary": "This cluster is a false-positive collection: the PRs share broad themes like diffusion pipelines, schedulers, tests, or docs, but none of the soft-edge pairs fix the same concrete bug or change.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a good canonical representative because the items span many unrelated topics and code paths.", + "canonical_pr_reason": "No single PR is a valid canonical representative; the cluster spans unrelated feature, typo, doc, test, and release changes.", "best_issue_reason": null, - "best_pr_reason": "No global best PR: the cluster is heterogeneous, so any one PR would misrepresent the rest.", + "best_pr_reason": "No PR is a strong global best choice because the set is heterogeneous; any pick would be arbitrary and misleading.", "soft_edge_verdicts": [ { "left": "pull_request:1319", "right": "pull_request:421", "accept": false, - "reason": "Different issues: loading schedulers from a subfolder vs fixing scheduler docs/typos." + "reason": "Scheduler-loading fix vs scheduler docs typo; unrelated change scopes." }, { "left": "pull_request:1330", "right": "pull_request:908", "accept": false, - "reason": "Unrelated changes: inpaint scaling in training/examples/pipeline vs community mega pipeline inpaint wiring." + "reason": "Training/inpaint scaling fix vs community pipeline inpaint routing; different problems." }, { "left": "pull_request:1341", "right": "pull_request:874", "accept": false, - "reason": "Different features: adding Stable Diffusion Repaint vs generic community pipeline README/docs." + "reason": "New repaint pipeline feature vs generic README example; not the same change." }, { "left": "pull_request:1373", "right": "pull_request:1849", "accept": false, - "reason": "Same file, but different fixes: versatile diffusion model expansion vs forwarding cross_attention_kwargs." + "reason": "Versatile Diffusion feature expansion vs one attention-kwargs bugfix; different code paths." }, { "left": "pull_request:1394", "right": "pull_request:831", "accept": false, - "reason": "Both bump versions, but to different releases and times; not the same concrete change." + "reason": "Different release version bumps; not duplicates." }, { "left": "pull_request:1401", "right": "pull_request:848", "accept": false, - "reason": "Related to height/width handling, but one is optional args across many pipelines and the other is Flax-only latent shape correction." + "reason": "Broad optional height/width API change vs Flax shape bug in one pipeline." }, { "left": "pull_request:1408", "right": "pull_request:98", "accept": false, - "reason": "Both touch conversion scripts, but one fixes sample size conversion while the other is a different ncsnpp conversion script." + "reason": "Stable Diffusion converter fix vs NCSNPP converter script; unrelated scripts." }, { "left": "pull_request:1412", "right": "pull_request:3389", "accept": false, - "reason": "Integration tests for SD2 inpainting vs an unrelated attention refactor test harness." + "reason": "Integration tests for SD2 inpainting vs attention-refactor integration tests; unrelated." }, { "left": "pull_request:1412", "right": "pull_request:72", "accept": false, - "reason": "Different test targets and eras: SD2 inpainting integration tests vs 1D up/downsample unit tests." + "reason": "SD2 inpaint tests vs 1D up/downsample tests; different test areas." }, { "left": "pull_request:1423", "right": "pull_request:168", "accept": false, - "reason": "Flax Stable Diffusion 2 support vs the original Stable Diffusion pipeline introduction." + "reason": "Stable Diffusion 2 Flax support vs original Stable Diffusion pipeline introduction." }, { "left": "pull_request:1437", "right": "pull_request:260", "accept": false, - "reason": "Both are typos, but in different docs/code paths and not the same bug." + "reason": "Pipeline typo fix vs README typo fix; unrelated files and bugs." }, { "left": "pull_request:1437", "right": "pull_request:407", "accept": false, - "reason": "Different typo fixes in the same file, but unrelated code paths and no shared bug." + "reason": "Different typos in the same file, but not the same concrete issue." }, { "left": "pull_request:1437", "right": "pull_request:798", "accept": false, - "reason": "Separate typo/docstring fixes; not the same underlying change." + "reason": "Pipeline typo vs UNet docstring typo; unrelated." }, { "left": "pull_request:1467", "right": "pull_request:3066", "accept": false, - "reason": "Slow-test threshold tweaks in different test files and contexts; not a single bug." + "reason": "Both relax slow tests, but they touch different assertions and test cases." }, { "left": "pull_request:1490", "right": "pull_request:641", "accept": false, - "reason": "Version/deprecation cleanup vs a dependency pin; unrelated maintenance changes." + "reason": "Version/deprecation bump vs temporary dependency pin; not the same fix." }, { "left": "pull_request:1496", "right": "pull_request:154", "accept": false, - "reason": "Documentation for Habana Gaudi vs the original stable-diffusion conversion script; unrelated." + "reason": "Habana docs vs Stable Diffusion conversion script; unrelated." }, { "left": "pull_request:1496", "right": "pull_request:1605", "accept": false, - "reason": "Habana docs vs Versatile Diffusion upcast_attention support; different areas entirely." + "reason": "Habana docs vs Versatile Diffusion attention option; unrelated." }, { "left": "pull_request:15", "right": "pull_request:395", "accept": false, - "reason": "README output path typo vs resnet variable renaming; unrelated edits." + "reason": "README output-path tweak vs variable renaming in resnet; unrelated." }, { "left": "pull_request:1510", "right": "pull_request:1989", "accept": false, - "reason": "Both conversion-related, but one adds scheduler/conversion improvements while the other adds safetensors loading support." + "reason": "Both touch the conversion script, but they address different enhancements." }, { "left": "pull_request:1558", "right": "pull_request:948", "accept": false, - "reason": "Different typo fixes in different pipeline files; not the same underlying issue." + "reason": "Stable Diffusion typo cleanup vs DDPM comment typo; unrelated." }, { "left": "pull_request:1576", "right": "pull_request:3068", "accept": false, - "reason": "ORT integration examples vs textual inversion docs; unrelated." + "reason": "ORT example integration vs textual inversion docs; different changes." }, { "left": "pull_request:1576", "right": "pull_request:3166", "accept": false, - "reason": "ORT training example integration vs a docs-only textual inversion loader example." + "reason": "ORT example integration vs textual inversion loader docs fix; unrelated." }, { "left": "pull_request:1576", "right": "pull_request:3554", "accept": false, - "reason": "ORT integration for examples vs a Moreh textual inversion fork; different codepaths." + "reason": "ORT example integration vs Moreh textual inversion adaptation; unrelated." }, { "left": "pull_request:1576", "right": "pull_request:3555", "accept": false, - "reason": "ORT integration for examples vs a different Moreh textual inversion fork; unrelated." + "reason": "ORT example integration vs Moreh textual inversion adaptation; unrelated." }, { "left": "pull_request:1627", "right": "pull_request:316", "accept": false, - "reason": "xformers enablement in modeling utils vs a README contribution edit; unrelated." + "reason": "xformers enable support vs README contribution text; unrelated." }, { "left": "pull_request:1640", "right": "pull_request:3417", "accept": false, - "reason": "Automatic xformers enabling change vs an unCLIP scheduler deprecation note; different fixes." + "reason": "xformers auto-enable policy vs unCLIP scheduler warning note; different concerns." }, { "left": "pull_request:1727", "right": "pull_request:3241", "accept": false, - "reason": "ONNX img2img preprocessing bug fix vs a test tolerance adjustment for UnCLIP." + "reason": "ONNX img2img preprocessing fix vs UnCLIP attention-slicing test tolerance; unrelated." } ] }, @@ -135213,7 +135275,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 95, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -135313,10 +135375,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -135331,234 +135393,234 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated small docs/typo/feature PRs; the soft links look like lexical similarity rather than duplicates. No soft edge appears to be the same underlying change or bug.", + "summary": "This cluster is a mix of unrelated documentation fixes, example-script tweaks, and a few real code changes across different subsystems. The soft-similarity links do not indicate duplicate bugs or mergeable PRs, so no pair should be collapsed.", "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 2009 is the broadest and most central code change here: it adds textual inversion loading across Stable Diffusion pipelines, touches many relevant files, and has by far the most review/discussion activity.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 2009 is the best representative PR for the cluster because it is the most substantial, most widely applicable change, and most clearly anchored in the Stable Diffusion loading path.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:178", "right": "pull_request:2278", "accept": false, - "reason": "Unrelated: stable diffusion pipeline test update vs a large video diffusion/community project PR." + "reason": "Stable Diffusion test update vs large video-stable-diffusion feature work; unrelated changes." }, { "left": "pull_request:178", "right": "pull_request:309", "accept": false, - "reason": "Unrelated: stable diffusion test fixture update vs a README link fix." + "reason": "Stable Diffusion test data update vs README link fix; different scope." }, { "left": "pull_request:178", "right": "pull_request:72", "accept": false, - "reason": "Unrelated: stable diffusion test update vs 1D up/downsample tests." + "reason": "Pipeline test update vs 1D up/downsample tests; different code paths." }, { "left": "pull_request:1838", "right": "pull_request:2009", "accept": false, - "reason": "Related theme, but not the same change: 1838 adds logging images to textual inversion; 2009 adds textual inversion loading support across Stable Diffusion pipelines." + "reason": "Textual inversion logging/docs work vs loader API support for embeddings; not the same bug/change." }, { "left": "pull_request:1885", "right": "pull_request:260", "accept": false, - "reason": "Unrelated: dreambooth training typo fix vs README typo fix." + "reason": "DreamBooth loop variable fix vs README typo fix; unrelated." }, { "left": "pull_request:1885", "right": "pull_request:353", "accept": false, - "reason": "Unrelated: dreambooth typo fix vs UNet block typo fix." + "reason": "Training script typo vs UNet block error-message typo; different files and issues." }, { "left": "pull_request:1885", "right": "pull_request:407", "accept": false, - "reason": "Unrelated: dreambooth training typo fix vs attention slicing typo fix." + "reason": "DreamBooth training typo vs attention-slicing typo; unrelated subsystems." }, { "left": "pull_request:1885", "right": "pull_request:798", "accept": false, - "reason": "Unrelated: dreambooth CLI typo fix vs UNet docstring typo fix." + "reason": "DreamBooth training typo vs docstring typo; not duplicate work." }, { "left": "pull_request:1885", "right": "pull_request:972", "accept": false, - "reason": "Unrelated: dreambooth typo fix vs README torch_dtype typo fix." + "reason": "Training loop typo vs README torch_dtype typo; different artifacts." }, { "left": "pull_request:190", "right": "pull_request:688", "accept": false, - "reason": "Different fixes in different layers: scipy import guard vs Flax/PyTorch import handling." + "reason": "PyTorch import guard for Stable Diffusion vs Flax-without-PyTorch import handling; distinct compatibility fixes." }, { "left": "pull_request:1932", "right": "pull_request:3285", "accept": false, - "reason": "Different problems: DDIM scheduler conversion/loading behavior vs disk offload support." + "reason": "DDIM scheduler conversion fix vs disk offload support; unrelated functionality." }, { "left": "pull_request:1965", "right": "pull_request:2006", "accept": false, - "reason": "Same file, but different merger issues: import path cleanup vs config/device/map/state-dict handling." + "reason": "Both touch checkpoint merger, but one is import cleanup and the other is broader merger behavior fixes; not the same concrete change." }, { "left": "pull_request:1995", "right": "pull_request:2539", "accept": false, - "reason": "Unrelated: img2img strength range typo vs pix2pix-zero guidance doc default." + "reason": "Img2img strength range typo vs pix2pix-zero guidance default doc fix; unrelated." }, { "left": "pull_request:1995", "right": "pull_request:260", "accept": false, - "reason": "Unrelated: value-range typo vs README typo." + "reason": "Img2img parameter validation vs README typo fix; unrelated." }, { "left": "pull_request:1995", "right": "pull_request:353", "accept": false, - "reason": "Unrelated: value-range typo vs UNet block typo." + "reason": "Img2img validation message vs UNet block typo; different code paths." }, { "left": "pull_request:2006", "right": "pull_request:869", "accept": false, - "reason": "Unrelated: checkpoint merger fix vs community README/import cleanup." + "reason": "Checkpoint merger behavior fix vs community README/import bug cleanup; unrelated." }, { "left": "pull_request:2006", "right": "pull_request:925", "accept": false, - "reason": "Unrelated: checkpoint merger fix vs prompt weighting bug in LPW community pipeline." + "reason": "Checkpoint merger behavior fix vs prompt-padding bug in LPW community pipeline; different problem." }, { "left": "pull_request:2009", "right": "pull_request:3587", "accept": false, - "reason": "Unrelated: textual inversion loader feature vs community CLIP-guided images-mixing pipeline." + "reason": "Textual inversion loader support vs community pipeline README/example addition; unrelated." }, { "left": "pull_request:2077", "right": "pull_request:2760", "accept": false, - "reason": "Different scope: validation images for textual inversion vs an Intel quantization/distillation example." + "reason": "Validation-image support in textual inversion vs new DFQ example project; not duplicate." }, { "left": "pull_request:2078", "right": "pull_request:2139", "accept": false, - "reason": "Related LoRA topic, but distinct changes: TensorBoard logging vs LoRA scale/alpha docs." + "reason": "TensorBoard image logging in LoRA training vs LoRA alpha/scale docs section; different change type and target." }, { "left": "pull_request:2085", "right": "pull_request:3470", "accept": false, - "reason": "Same broader DreamBooth docs area, but different content: random crop note vs IF DreamBooth docs." + "reason": "DreamBooth crop option docs vs IF DreamBooth docs; unrelated documentation." }, { "left": "pull_request:2086", "right": "pull_request:3458", "accept": false, - "reason": "Docs on LoRA support vs inpaint LoRA scale support are not the same underlying change." + "reason": "LoRA docs addition vs inpaint LoRA-scale support implementation; different code and intent." }, { "left": "pull_request:2086", "right": "pull_request:3460", "accept": false, - "reason": "Docs on LoRA support vs inpaint LoRA scale support with tests are different change types and different code paths." + "reason": "LoRA docs addition vs inpaint LoRA-scale support implementation/tests; not the same change." }, { "left": "pull_request:2103", "right": "pull_request:3187", "accept": false, - "reason": "Unrelated: adding LoRA tag metadata vs x4 upscaler model offload." + "reason": "Tagging model cards with lora vs x4 upscaler offload support; unrelated." }, { "left": "pull_request:2103", "right": "pull_request:3458", "accept": false, - "reason": "Unrelated: LoRA tag metadata vs inpaint LoRA support." + "reason": "LoRA tag metadata vs inpaint cross-attention support; different purpose." }, { "left": "pull_request:2103", "right": "pull_request:3460", "accept": false, - "reason": "Unrelated: LoRA tag metadata vs inpaint LoRA support with tests." + "reason": "LoRA tag metadata vs inpaint support implementation/tests; unrelated." }, { "left": "pull_request:2134", "right": "pull_request:948", "accept": false, - "reason": "Unrelated typo fixes in different README/pipeline docs." + "reason": "README typo fix in DreamBooth vs pipeline_ddpm comment typo; unrelated docs-only fixes." }, { "left": "pull_request:2138", "right": "pull_request:3839", "accept": false, - "reason": "Unrelated: docs typo vs scheduler add_noise behavior fix." + "reason": "Docs wording fix vs SDE VE add_noise behavior fix; different artifacts and code paths." }, { "left": "pull_request:2144", "right": "pull_request:3030", "accept": false, - "reason": "Different code paths: transformer output projection bug vs panorama CFG handling bug." + "reason": "Transformer 2D projection dimension bug vs panorama CFG bug; different models and failure modes." }, { "left": "pull_request:2151", "right": "pull_request:4320", "accept": false, - "reason": "Unrelated typo/docstring fix vs ONNX docs typo fix." + "reason": "EMAModel state_dict typo fix vs ONNX docs typo; unrelated." }, { "left": "pull_request:2154", "right": "pull_request:260", "accept": false, - "reason": "Unrelated: CLI env variable typo vs README typo." + "reason": "CLI environment version variable typo vs README typo; different files and issues." }, { "left": "pull_request:2154", "right": "pull_request:353", "accept": false, - "reason": "Unrelated typo fixes in different files and subsystems." + "reason": "CLI env version typo vs UNet block typo; unrelated." }, { "left": "pull_request:2154", "right": "pull_request:407", "accept": false, - "reason": "Unrelated typo fixes in different files and subsystems." + "reason": "CLI env version typo vs attention-slicing typo; unrelated." }, { "left": "pull_request:2154", "right": "pull_request:798", "accept": false, - "reason": "Unrelated typo fixes in different files and subsystems." + "reason": "CLI env version typo vs docstring typo; unrelated." }, { "left": "pull_request:2175", "right": "pull_request:3388", "accept": false, - "reason": "Unrelated docs tweaks: DiffusionPipeline components vs installation instructions." + "reason": "DiffusionPipeline docs tweak vs installation docs adding transformers; unrelated docs." }, { "left": "pull_request:2181", "right": "pull_request:2540", "accept": false, - "reason": "Different docs topics: Flax scheduler guide vs ControlNet section correction." + "reason": "Flax scheduler docs section vs ControlNet docs correction; unrelated documentation." }, { "left": "pull_request:2181", "right": "pull_request:421", "accept": false, - "reason": "Different docs topics: Flax scheduler guide vs general scheduler docs typo fix." + "reason": "Flax scheduler docs addition vs scheduler docs typo cleanup; different scope and content." } ] }, @@ -135574,7 +135636,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 96, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -135664,10 +135726,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -135682,174 +135744,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very heterogeneous: mostly unrelated docs, tests, CI, and feature PRs. The soft pairs share at most a broad topic or one filename, but they do not look like the same underlying bug or change.", - "confidence": 0.97, + "summary": "The cluster is highly heterogeneous: unrelated docs fixes, test updates, dependency/CI changes, and a few feature PRs. None of the soft pairs look like the same concrete bug or change, so all should be rejected.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a credible canonical for this cluster because the items span unrelated changes; even the closest pairs are only superficial file- or topic-level matches.", + "canonical_pr_reason": "No single PR is a good canonical representative; the items span unrelated maintenance and feature work rather than one duplicate change.", "best_issue_reason": null, - "best_pr_reason": "No PR is a strong global representative. The closest thematic overlap is the torch2.0 docs pair, but it is still not duplicate-level and the rest of the cluster is unrelated.", + "best_pr_reason": "No PR is a strong global duplicate target because there is no shared underlying bug or change across the cluster.", "soft_edge_verdicts": [ { "left": "pull_request:2208", "right": "pull_request:4022", "accept": false, - "reason": "Both are docs updates, but they affect different translations/files and unrelated content." + "reason": "xFormers docs update vs Korean translation overhaul; different documentation changes." }, { "left": "pull_request:2216", "right": "pull_request:888", "accept": false, - "reason": "Different feature additions: DifFace pipeline vs CycleDiffusion pipeline." + "reason": "DifFace pipeline addition vs CycleDiffusion pipeline addition; different features and code paths." }, { "left": "pull_request:2271", "right": "pull_request:268", "accept": false, - "reason": "Both touch tests, but they fix different test suites and different expected values." + "reason": "Different test maintenance: one adjusts slow test expectations, the other updates model/pipeline expected outputs across several components." }, { "left": "pull_request:2290", "right": "pull_request:3680", "accept": false, - "reason": "One is a pix2pix doc fix, the other is a link fix in a different docs page." + "reason": "Pix2Pix doc formatting fix vs A1111 LoRA loader-link fix; unrelated docs changes." }, { "left": "pull_request:2333", "right": "pull_request:406", "accept": false, - "reason": "Different documentation fixes in different files; only superficial typo/format similarity." + "reason": "Ethical guidelines wording change vs toctree typo fix; unrelated docs edits." }, { "left": "pull_request:2362", "right": "pull_request:3400", "accept": false, - "reason": "MPS skip decorator vs dependency addition are unrelated test/setup changes." + "reason": "MPS skip decorator/test changes vs dependency addition for tests; different code and intent." }, { "left": "pull_request:2381", "right": "pull_request:979", "accept": false, - "reason": "Dummy import handling in stable diffusion vs a missing import in README are unrelated." + "reason": "Stable Diffusion dummy-import guard vs README missing import; unrelated fixes in different files." }, { "left": "pull_request:2396", "right": "pull_request:408", "accept": false, - "reason": "Scheduler test cleanup vs general docs typo fixes are not the same change." + "reason": "Scheduler test warning cleanup vs documentation typo fixes; not the same change." }, { "left": "pull_request:2405", "right": "pull_request:830", "accept": false, - "reason": "Both are release/version-related, but one reverts a release and the other sets a release version." + "reason": "Revert of release bump vs release version bump; opposite versioning actions, not duplicates." }, { "left": "pull_request:2488", "right": "pull_request:4515", "accept": false, - "reason": "Attend-and-excite pipeline validation refactor vs SDXL DreamBooth validation placement are different code paths." + "reason": "Pipeline mixin refactor for num_images_per_prompt vs moving pipeline.to() for validation; different training/runtime behavior." }, { "left": "pull_request:2498", "right": "pull_request:3099", "accept": false, - "reason": "One changes batch test behavior; the other changes a pipeline default argument." + "reason": "Attend-and-excite test timeout adjustment vs changing a pipeline default parameter; unrelated test vs API change." }, { "left": "pull_request:2503", "right": "pull_request:2764", "accept": false, - "reason": "Same docs file, but they update different parts of the torch2.0 guide and solve different documentation issues." + "reason": "Different torch2.0 docs edits: benchmark/title wording vs installation guidance; same file family but not the same change." }, { "left": "pull_request:2530", "right": "pull_request:3755", "accept": false, - "reason": "EMAModel test suite addition vs a single tolerance relaxation are not the same change." + "reason": "New EMAModel test suite vs relaxing a flaky test tolerance; distinct test changes." }, { "left": "pull_request:2533", "right": "pull_request:4740", "accept": false, - "reason": "Both mention safetensors, but one is a transient test/CI change and the other is a dependency pin." + "reason": "Temporary CI/test setup for safetensors rc1 vs a later safetensors rc1 sanity-check WIP; different dependency-testing changes." }, { "left": "pull_request:2557", "right": "pull_request:4112", "accept": false, - "reason": "Different DreamBooth enhancements: intermediate logging vs LoRA rank support." + "reason": "DreamBooth intermediate logging vs adding rank support to LoRA training; separate training-script enhancements." }, { "left": "pull_request:2563", "right": "pull_request:703", "accept": false, - "reason": "Removing a CUDA cache call in EMA is unrelated to CI cache testing." + "reason": "Removing cuda cache emptying in EMA step vs CI cache testing workflow; unrelated fixes." }, { "left": "pull_request:2587", "right": "pull_request:3388", "accept": false, - "reason": "Chinese docs translation vs installation docs update are different documentation changes." + "reason": "Korean docs translation update vs adding transformers to installation docs; different documentation work." }, { "left": "pull_request:2590", "right": "pull_request:483", "accept": false, - "reason": "CI dependency-table check vs typo-check infrastructure are unrelated workflow changes." + "reason": "CI dependency-table check addition vs typo-fix workflow/docs changes; different maintenance areas." }, { "left": "pull_request:2604", "right": "pull_request:4794", "accept": false, - "reason": "Both are docs fixes, but they concern different pages and different content." + "reason": "Colab link fix in docs vs compel syntax/example update; unrelated docs edits." }, { "left": "pull_request:2624", "right": "pull_request:3388", "accept": false, - "reason": "Docs for cache_dir in pipeline_utils vs install-doc changes are unrelated." + "reason": "cache_dir docs addition in pipeline utils vs installation docs adding transformers; different documentation targets." }, { "left": "pull_request:2629", "right": "pull_request:2631", "accept": false, - "reason": "StableDiffusionSAG non-square support vs DreamBooth main-process checks are different code paths." + "reason": "StableDiffusionSAG non-square image support vs DreamBooth main-process checks; unrelated code paths." }, { "left": "pull_request:2637", "right": "pull_request:2639", "accept": false, - "reason": "Two separate quicktour/basic-training doc edits, not the same underlying change." + "reason": "Quicktour wording cleanup vs basic training tutorial import addition; different docs pages and fixes." }, { "left": "pull_request:2637", "right": "pull_request:3310", "accept": false, - "reason": "Quicktour wording cleanup vs stable_diffusion tutorial edits are unrelated docs changes." + "reason": "Quicktour wording cleanup vs stable_diffusion tutorial import fix; unrelated docs changes." }, { "left": "pull_request:2639", "right": "pull_request:2642", "accept": false, - "reason": "Basic training docs vs loading docs; same docs area but different content." + "reason": "Basic training tutorial import fix vs loading docs wording change; separate docs edits." }, { "left": "pull_request:2639", "right": "pull_request:2647", "accept": false, - "reason": "Basic training docs vs schedulers docs are unrelated page edits." + "reason": "Basic training tutorial import fix vs schedulers docs grammar correction; unrelated documentation changes." }, { "left": "pull_request:2639", "right": "pull_request:2684", "accept": false, - "reason": "Basic training docs vs custom pipeline overview wording change are unrelated." + "reason": "Basic training tutorial import fix vs custom pipeline overview punctuation; different docs locations and fixes." }, { "left": "pull_request:2639", "right": "pull_request:2685", "accept": false, - "reason": "Basic training docs vs KerasCV docs are different documentation fixes." + "reason": "Basic training tutorial import fix vs KerasCV docs punctuation fix; unrelated docs changes." } ] }, @@ -135865,7 +135927,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 97, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -135965,10 +136027,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -135983,234 +136045,234 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of tiny docs/typo PRs plus a few unrelated test/code changes; I don\u2019t see a single duplicate bug/change theme tying them together.", + "summary": "Mostly unrelated documentation typo/phrasing PRs, plus a few separate test/code changes; there is no single underlying duplicate change to collapse.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR subsumes the cluster. If a representative is needed, #2873 is the broadest generic docs-phrasing cleanup across multiple files.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "#2873 is the best representative only in the loose sense that it\u2019s a small, generic documentation wording fix; otherwise the items are too disparate for a true canonical PR.", + "best_pr_reason": "PR 910 is the broadest, most substantive docs update in the set (multi-file inpainting guidance), so it is the best anchor if one must pick a representative PR, but the cluster is not a true duplicate group.", "soft_edge_verdicts": [ { "left": "pull_request:2639", "right": "pull_request:2686", "accept": false, - "reason": "Both are doc edits, but one updates basic training and the other unconditional generation; different pages and changes." + "reason": "Different docs pages and different tweaks: basic_training imports `os`, while unconditional_image_generation only adds a period." }, { "left": "pull_request:2639", "right": "pull_request:2687", "accept": false, - "reason": "Different docs pages and independent wording fixes; not the same underlying change." + "reason": "Separate tutorial docs with unrelated wording fixes; not the same underlying change." }, { "left": "pull_request:2639", "right": "pull_request:2688", "accept": false, - "reason": "Unconditional training vs img2img docs; unrelated content and fixes." + "reason": "basic_training import fix vs img2img section punctuation/formatting edits; different changes." }, { "left": "pull_request:2639", "right": "pull_request:2690", "accept": false, - "reason": "Different docs topic and separate minor prose fix." + "reason": "Unrelated docs pages and edits; one is a code snippet import, the other a sentence punctuation fix." }, { "left": "pull_request:2639", "right": "pull_request:2730", "accept": false, - "reason": "One is a docs tweak, the other is training docs; no shared bug/change." + "reason": "Different training docs and different edits; no shared concrete change." }, { "left": "pull_request:2639", "right": "pull_request:2740", "accept": false, - "reason": "These touch different training docs and different edits; not duplicates." + "reason": "basic_training typo/import cleanup is unrelated to the dreambooth/text_inversion docs revisions in 2740." }, { "left": "pull_request:2639", "right": "pull_request:2741", "accept": false, - "reason": "Basic training docs vs textual inversion docs; unrelated." + "reason": "Different files and different typo fixes; not one merged change." }, { "left": "pull_request:2639", "right": "pull_request:2742", "accept": false, - "reason": "Different training docs and different wording fixes; not the same PR." + "reason": "basic_training import edit vs dreambooth wording fix; separate doc updates." }, { "left": "pull_request:2639", "right": "pull_request:2746", "accept": false, - "reason": "Basic training docs vs fp16 optimization docs; separate content." + "reason": "Different docs and topics; one is tutorial code, the other fp16 prose cleanup." }, { "left": "pull_request:2639", "right": "pull_request:2749", "accept": false, - "reason": "Different docs pages and unrelated grammar edits." + "reason": "Different docs pages; no evidence of a shared underlying bug/change." }, { "left": "pull_request:2639", "right": "pull_request:2751", "accept": false, - "reason": "Basic training docs vs text inversion docs; no shared concrete change." + "reason": "Separate training docs typo fixes; too broad and unrelated to be duplicates." }, { "left": "pull_request:2639", "right": "pull_request:2752", "accept": false, - "reason": "Different docs pages and prose-only fixes; not duplicates." + "reason": "Different conceptual docs and different edits; no same change." }, { "left": "pull_request:2639", "right": "pull_request:2862", "accept": false, - "reason": "Tutorial docs vs evaluation docs; unrelated edits." + "reason": "basic_training import cleanup is unrelated to evaluation.mdx wording fixes." }, { "left": "pull_request:2639", "right": "pull_request:2863", "accept": false, - "reason": "Different API docs pages with separate wording fixes." + "reason": "Different docs pages; typo fixes only, not the same underlying change." }, { "left": "pull_request:2639", "right": "pull_request:2864", "accept": false, - "reason": "Different docs pages; one is tutorial cleanup, the other overview punctuation." + "reason": "Separate docs updates in different files; no duplicate bug/change." }, { "left": "pull_request:2639", "right": "pull_request:2865", "accept": false, - "reason": "AltDiffusion docs vs basic training docs; unrelated changes." + "reason": "Unrelated doc pages and edits; only broad docs similarity." }, { "left": "pull_request:2639", "right": "pull_request:2870", "accept": false, - "reason": "Different docs areas and separate minor edits." + "reason": "Different docs pages; one is a snippet/import edit, the other stable_diffusion_safe wording cleanup." }, { "left": "pull_request:2639", "right": "pull_request:2923", "accept": false, - "reason": "Basic training docs vs versatile diffusion docs; no common code-path or content change." + "reason": "Different training tutorial files and separate wording fixes." }, { "left": "pull_request:2642", "right": "pull_request:3310", "accept": false, - "reason": "Both are docs, but they edit different pages and different topics." + "reason": "loading.mdx variant wording change vs stable_diffusion.mdx PIL import fix; unrelated docs updates." }, { "left": "pull_request:2647", "right": "pull_request:3310", "accept": false, - "reason": "Schedulers docs vs stable_diffusion tutorial; unrelated grammar fixes." + "reason": "Schedulers docs punctuation vs stable_diffusion tutorial import fix; not one change." }, { "left": "pull_request:2684", "right": "pull_request:3310", "accept": false, - "reason": "Custom pipeline overview vs stable diffusion tutorial; separate doc updates." + "reason": "custom_pipeline_overview punctuation fix is unrelated to stable_diffusion.md import correction." }, { "left": "pull_request:2742", "right": "pull_request:650", "accept": false, - "reason": "Both mention DreamBooth, but one fixes prose in the docs and the other changes install instructions and adds requirements; not the same change." + "reason": "Same general DreamBooth area, but different files and different changes; not a duplicate PR." }, { "left": "pull_request:2746", "right": "pull_request:983", "accept": false, - "reason": "fp16 docs typo fix vs empty debug PR; no shared underlying bug/change." + "reason": "fp16 prose cleanup vs an empty debug PR; no shared concrete change." }, { "left": "pull_request:2764", "right": "pull_request:2766", "accept": false, - "reason": "Torch 2.0 docs update vs MPS test fixes; different artifacts and purposes." + "reason": "Docs update about torch 2.0 vs a test fix for MPS outputs; different artifact types and changes." }, { "left": "pull_request:2764", "right": "pull_request:910", "accept": false, - "reason": "Unrelated docs: torch 2.0 benchmark vs in-painting pipeline docs." + "reason": "Torch 2.0 docs benchmark update and inpainting docs overhaul are unrelated changes." }, { "left": "pull_request:2766", "right": "pull_request:2792", "accept": false, - "reason": "Both are test-related, but they fix different MPS issues in different test files and code paths." + "reason": "Both touch MPS-related tests, but they fix different test suites and different failures; not the same bug." }, { "left": "pull_request:2788", "right": "pull_request:3472", "accept": false, - "reason": "Both touch training docs, but one deduplicates the docs nav and the other adds an IF training section." + "reason": "Training toctree cleanup is unrelated to adding IF content in DreamBooth docs." }, { "left": "pull_request:2815", "right": "pull_request:2823", "accept": false, - "reason": "Same file, but one updates checkpoint/docs content while the other rewrites the usage example; different concrete changes." + "reason": "Both edit stable_unclip.mdx, but 2815 adds background/checkpoints while 2823 rewrites the example code; separate doc updates." }, { "left": "pull_request:2817", "right": "pull_request:3821", "accept": false, - "reason": "UNet1DModel docs vs a Korean translated pipeline guide; unrelated." + "reason": "UNet1DModel docs addition vs a full Korean translation page; not the same change." }, { "left": "pull_request:286", "right": "pull_request:642", "accept": false, - "reason": "README link fix in stable diffusion README vs examples README link fix; similar style, different files and targets." + "reason": "Different README/link fixes in different files and contexts; not duplicates." }, { "left": "pull_request:2869", "right": "pull_request:3119", "accept": false, - "reason": "Paint-by-example docs typo vs repaint interface update; different pipelines and changes." + "reason": "Different pipeline docs and different wording/interface updates; not the same underlying change." }, { "left": "pull_request:2873", "right": "pull_request:3572", "accept": false, - "reason": "Generic phrasing cleanup vs controlnet inpaint doc example correction; different docs and issues." + "reason": "General phrasing cleanup vs controlnet inpainting doc correction; unrelated edits." }, { "left": "pull_request:2894", "right": "pull_request:3894", "accept": false, - "reason": "A requirements file addition vs a Flax dropout feature; not the same underlying change." + "reason": "New flax requirements file vs Flax UNet dropout implementation; completely different changes." }, { "left": "pull_request:2911", "right": "pull_request:3119", "accept": false, - "reason": "Stable diffusion image-variation docs punctuation fix vs repaint docs interface change; unrelated." + "reason": "Stable-diffusion image_variation punctuation fix is unrelated to repaint interface docs." }, { "left": "pull_request:2912", "right": "pull_request:3119", "accept": false, - "reason": "ControlNet docs punctuation fix vs repaint API/doc update; different artifacts." + "reason": "ControlNet docs punctuation fix vs repaint interface update; different changes." }, { "left": "pull_request:2913", "right": "pull_request:3119", "accept": false, - "reason": "Stable unCLIP docs punctuation fix vs repaint docs interface change; unrelated." + "reason": "Stable-unCLIP docs punctuation/final newline fix is not the same as repaint docs." }, { "left": "pull_request:2919", "right": "pull_request:4715", "accept": false, - "reason": "A test-speedup change in UNet tests vs renaming community test scripts; different goals and files." + "reason": "MPS test speedup and example script renames are different kinds of changes with no shared bug." } ] }, @@ -136226,7 +136288,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 98, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -136315,10 +136377,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -136333,168 +136395,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This looks like a heterogeneous bucket of mostly unrelated small docs, test, and bugfix PRs. None of the soft pairs appears to be the same underlying change, so all soft edges should be rejected.", - "confidence": 0.96, + "summary": "This cluster is highly heterogeneous: it mixes unrelated documentation edits, example-script fixes, test adjustments, and core pipeline/scheduler bug fixes. There isn\u2019t a single underlying bug or change that unifies the items into one duplicate set.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "If a representative PR is needed, PR 5117 is the broadest anchor: it spans many training scripts and fixes a concrete resume-step/logging path, while most other items are isolated docs/comment/test edits.", + "canonical_pr_reason": "No single PR serves as a canonical representative because the cluster spans many unrelated changes across docs, tests, examples, and core code paths.", "best_issue_reason": null, - "best_pr_reason": "PR 3641 is the strongest standalone code fix here: it addresses a specific inpainting initial-latents bug in two pipeline implementations with clear behavioral impact.", + "best_pr_reason": "No PR is a good global canonical choice; the items do not share one concrete fix or feature, so any PR would misrepresent the cluster.", "soft_edge_verdicts": [ { "left": "pull_request:296", "right": "pull_request:979", "accept": false, - "reason": "Auth-token test/workflow fix vs README fp16/import update; same general project but unrelated changes." + "reason": "Different changes: auth-token test/workflow fix vs missing import + dtype README update." }, { "left": "pull_request:299", "right": "pull_request:356", "accept": false, - "reason": "PNDM timestep logic bug vs stable-diffusion fine-tuning example; different code paths and problems." + "reason": "PNDM scheduler hotfix is unrelated to the stable-diffusion fine-tuning example PR." }, { "left": "pull_request:2994", "right": "pull_request:5157", "accept": false, - "reason": "ControlNet training type-cast fix vs LoRA/Kohya name-conversion fix; distinct bugs in different subsystems." + "reason": "ControlNet training type-casting fix and LoRA name-conversion fix touch different code paths and bugs." }, { "left": "pull_request:3012", "right": "pull_request:948", "accept": false, - "reason": "Both are typo/comment cleanups, but they touch unrelated files and changes; not the same change." + "reason": "Both are typo/comment fixes, but in different pipelines and unrelated files; not the same change." }, { "left": "pull_request:3030", "right": "pull_request:3979", "accept": false, - "reason": "Panorama CFG behavior fix vs SDXL/xformers attention check; different pipeline bugs." + "reason": "Panorama CFG bug fix is unrelated to the SDXL attention processor/type-check bug." }, { "left": "pull_request:3079", "right": "pull_request:3099", "accept": false, - "reason": "Test-speedup changes vs a pipeline default-value change; one is test-only, the other changes API behavior." + "reason": "One speeds up tests; the other changes a pipeline default value. Different artifacts and purposes." }, { "left": "pull_request:3132", "right": "pull_request:5332", "accept": false, - "reason": "Type annotation typo vs explanatory comment typo; unrelated small edits." + "reason": "A type-hint fix in score_sde_ve is unrelated to a DreamBooth comment typo." }, { "left": "pull_request:3166", "right": "pull_request:4794", "accept": false, - "reason": "Textual inversion loader docs/example vs compel syntax docs; both docs, but different features and intent." + "reason": "Textual inversion docs/example tweak vs compel syntax fix; both docs, but different issues." }, { "left": "pull_request:3169", "right": "pull_request:3663", "accept": false, - "reason": "Habana optimization docs update vs compel/textual-inversion docs; unrelated documentation topics." + "reason": "Habana docs update and Compel textual-inversion docs are separate documentation topics." }, { "left": "pull_request:3183", "right": "pull_request:4729", "accept": false, - "reason": "DreamBooth LoRA checkpoint registration bug vs progress-bar update bug; separate runtime issues." + "reason": "DreamBooth LoRA checkpointing bug and a community script progress-bar bug are unrelated." }, { "left": "pull_request:3226", "right": "pull_request:3572", "accept": false, - "reason": "ControlNet v1.1 docs vs inpainting docs correction; both docs, but not the same change." + "reason": "ControlNet v1.1 docs addition is not the same as correcting the inpainting ControlNet docs example." }, { "left": "pull_request:3388", "right": "pull_request:3795", "accept": false, - "reason": "Install docs adding transformers vs overview-page pipeline additions; unrelated docs updates." + "reason": "Installation docs change vs overview-page pipeline additions; unrelated documentation scopes." }, { "left": "pull_request:3470", "right": "pull_request:4203", "accept": false, - "reason": "New IF DreamBooth docs section vs a typo fix in the DreamBooth README; same area but different scope and content." + "reason": "Shared README file, but one adds IF training docs while the other only fixes a typo." }, { "left": "pull_request:3472", "right": "pull_request:5250", "accept": false, - "reason": "IF DreamBooth docs vs DDPO docs; both new docs, but unrelated subjects." + "reason": "DreamBooth IF docs and DDPO docs are separate new documentation sections for different workflows." }, { "left": "pull_request:3472", "right": "pull_request:5723", "accept": false, - "reason": "IF DreamBooth docs vs LCM docs; different model families and documentation goals." + "reason": "DreamBooth IF docs and LCM docs describe different models and usage patterns." }, { "left": "pull_request:3506", "right": "pull_request:3641", "accept": false, - "reason": "ControlNet checkpoint resume-step scaling bug vs inpainting initial-latents bug; different code paths." + "reason": "Checkpoint resume-step fix for ControlNet training is unrelated to inpainting initial-latents handling." }, { "left": "pull_request:3525", "right": "pull_request:4203", "accept": false, - "reason": "README cleanup/marketing text vs DreamBooth README typo fix; unrelated README edits." + "reason": "README update is broad/tainted content, while the other is a typo fix; not the same underlying change." }, { "left": "pull_request:3525", "right": "pull_request:4503", "accept": false, - "reason": "README cleanup/marketing text vs LoRA support clarification; different documentation concerns." + "reason": "README update and LoRA docs clarification are different documentation changes." }, { "left": "pull_request:356", "right": "pull_request:904", "accept": false, - "reason": "Large text-to-image fine-tuning example vs inpainting pipeline implementation; completely different scope." + "reason": "Stable diffusion fine-tuning repo/docs are unrelated to the inpainting pipeline implementation." }, { "left": "pull_request:3577", "right": "pull_request:5639", "accept": false, - "reason": "Broken-link docs fix vs AnimateDiff Motion LoRAs docs; unrelated documentation changes." + "reason": "Broken README links and AnimateDiff Motion LoRA docs are unrelated documentation updates." }, { "left": "pull_request:3680", "right": "pull_request:4721", "accept": false, - "reason": "Loader-method link fix vs Korean DreamBooth translation edits; not the same artifact or bug." + "reason": "Loader-method link fix and Korean DreamBooth translation are different docs changes." }, { "left": "pull_request:3744", "right": "pull_request:4058", "accept": false, - "reason": "Docs-upload token rename vs SDXL docs formatting/content correction; unrelated." + "reason": "Docs upload token fix is unrelated to SDXL documentation content changes." }, { "left": "pull_request:3755", "right": "pull_request:3842", "accept": false, - "reason": "Relaxing tolerances in different tests is only a superficial similarity; they target different flaky tests." + "reason": "Both relax tests, but they cover different models/components and are not the same flaky test fix." }, { "left": "pull_request:3798", "right": "pull_request:5117", "accept": false, - "reason": "Both touch training scripts, but one is comment cleanup and the other fixes resume/logging logic across many scripts; not the same bug/change." + "reason": "Both touch training scripts, but one is comment cleanup and the other fixes resume-step logging/logic across many scripts." }, { "left": "pull_request:3821", "right": "pull_request:4285", "accept": false, - "reason": "Translated docs addition vs ONNX/OpenVINO docs; different documentation sets." + "reason": "Translated documentation addition is unrelated to ONNX/OpenVINO/SDXL docs." }, { "left": "pull_request:388", "right": "pull_request:5", "accept": false, - "reason": "Early latent-diffusion docs vs original UNet model implementation; historically and functionally distinct." + "reason": "A docs/pipeline page for latent diffusion is unrelated to adding the original UNet model implementation." } ] }, @@ -136510,7 +136572,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 99, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -136601,10 +136663,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -136619,7 +136681,7 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These PRs are not duplicates of one another; the cluster mixes unrelated docs typo fixes, SDXL pipeline bugfixes, and feature additions across different subsystems. No single canonical artifact cleanly represents the whole set.", + "summary": "This cluster is heterogeneous: it mixes unrelated docs typo cleanups, SDXL runtime bugfixes, feature additions, and refactors. I do not see a single underlying issue/change spanning the items, so the soft matches should be rejected.", "confidence": 0.96, "canonical_issue_reason": null, "canonical_pr_reason": null, @@ -136630,175 +136692,175 @@ "left": "pull_request:3979", "right": "pull_request:4796", "accept": false, - "reason": "Both touch SDXL/upscale code, but 3979 fixes an `isinstance` check in SDXL pipelines while 4796 fixes VAE upcasting/type mismatches across SDXL pipelines; related area, different bug." + "reason": "Both touch SDXL pipeline float/dtype code, but one is a small attention-processor check in upscale/XL pipelines while the other fixes broader SDXL latent upcasting behavior across multiple pipelines." }, { "left": "pull_request:3979", "right": "pull_request:4797", "accept": false, - "reason": "3979 is a small attention-processor check fix; 4797 is about SDXL LoRA block remapping/loading. Different code paths and failure modes." + "reason": "Different bugs: attention-processor detection in a pipeline vs LoRA block remapping in loaders." }, { "left": "pull_request:3979", "right": "pull_request:5078", "accept": false, - "reason": "3979 changes attention processor dtype logic; 5078 fixes SDXL text encoder training gradients. Unrelated bugs." + "reason": "Unrelated fixes: SDXL VAE/attention handling vs text-encoder training gradient-checkpointing setup." }, { "left": "pull_request:406", "right": "pull_request:408", "accept": false, - "reason": "Both are docs typo fixes, but one corrects a toctree key and the other fixes multiple typos across docs; not the same underlying change." + "reason": "Both are docs typo cleanup, but they fix different files and different text; not the same change." }, { "left": "pull_request:407", "right": "pull_request:948", "accept": false, - "reason": "Both are tiny typo fixes in pipeline comments, but they affect different files and different text; not a duplicate change." + "reason": "One fixes an attention-slicing API typo; the other fixes a DDPM comment typo." }, { "left": "pull_request:4082", "right": "pull_request:4267", "accept": false, - "reason": "4082 changes a DreamBooth requirement version; 4267 fixes an SDXL docs image link. Different docs and unrelated issues." + "reason": "Requirement pin change vs SDXL docs image-link correction; no shared bug or code path." }, { "left": "pull_request:4082", "right": "pull_request:4582", "accept": false, - "reason": "Requirement pin change versus a ControlNet SDXL docstring correction; no shared bug or code path." + "reason": "Dependency/version fix vs ControlNet SDXL docstring correction; unrelated." }, { "left": "pull_request:4082", "right": "pull_request:5013", "accept": false, - "reason": "4082 is a package requirement fix; 5013 fixes dtype handling in SDXL InstructPix2Pix latents. Different problem domains." + "reason": "Package requirement adjustment vs SDXL instruct-pix2pix dtype bug fix; different artifacts and problems." }, { "left": "pull_request:4106", "right": "pull_request:4595", "accept": false, - "reason": "4106 adds the Tortoise TTS pipeline; 4595 is a separate conversion-script work-in-progress. Feature addition vs unrelated script effort." + "reason": "A large Tortoise TTS feature addition vs an unfinished converter script for SDXL; not the same change." }, { "left": "pull_request:4203", "right": "pull_request:5267", "accept": false, - "reason": "Both are docs-heavy PRs, but 4203 fixes DreamBooth README typos while 5267 updates docs README formatting guidance. Different documentation areas and intent." + "reason": "DreamBooth README typo fix vs docs README formatting guidance; different docs and edits." }, { "left": "pull_request:4203", "right": "pull_request:5855", "accept": false, - "reason": "DreamBooth README typo fixes versus a community README import typo; same broad doc hygiene but not the same artifact or issue." + "reason": "DreamBooth README typo vs community README import typo; both docs-only but unrelated." }, { "left": "pull_request:421", "right": "pull_request:564", "accept": false, - "reason": "421 fixes a scheduler docs typo; 564 rewrites Flax scheduler docs/API details across multiple files. The scope and content differ substantially." + "reason": "Scheduler docs typo cleanup vs Flax scheduler API/documentation expansion; different scope and content." }, { "left": "pull_request:4262", "right": "pull_request:4763", "accept": false, - "reason": "4262 is an SDXL DreamBooth LoRA training fix; 4763 is a script to convert SDXL LoRA safetensors for A1111. Related SDXL tooling, but different concrete changes." + "reason": "SDXL DreamBooth LoRA training fixes vs a LoRA-to-safetensors conversion script; separate functionality." }, { "left": "pull_request:4267", "right": "pull_request:6339", "accept": false, - "reason": "4267 fixes an SDXL docs image link; 6339 updates AnimateDiff docs/example code. Different pipelines and unrelated fixes." + "reason": "Different docs fixes for different features: SDXL image swap vs AnimateDiff docs." }, { "left": "pull_request:4275", "right": "pull_request:5831", "accept": false, - "reason": "4275 corrects AutoPipeline docs code examples; 5831 only adds a docs section title. Not the same bug/change." + "reason": "Autopipeline docs correction vs a docs tree title fix; unrelated doc edits." }, { "left": "pull_request:4285", "right": "pull_request:5286", "accept": false, - "reason": "4285 expands ONNX/OpenVINO docs; 5286 adds Japanese docs. Both docs-related but completely different work." + "reason": "ONNX/OpenVINO documentation update vs adding Japanese docs; not the same underlying change." }, { "left": "pull_request:4330", "right": "pull_request:6342", "accept": false, - "reason": "4330 corrects ControlNet SDXL README header text; 6342 adds an installation note for Python 3.12 on Windows. Different documentation issues." + "reason": "ControlNet README header correction vs installation docs note; unrelated documentation fixes." }, { "left": "pull_request:4350", "right": "pull_request:4696", "accept": false, - "reason": "4350 fixes an fp16 type in T2I-Adapter docs; 4696 adds T2I-Adapter-XL models and docs. Same subsystem, but one is a doc tweak and the other is a feature expansion." + "reason": "Base T2I-Adapter docs fp16 fix vs adding T2I-Adapter-XL support; one is a doc tweak, the other is a feature/code addition." }, { "left": "pull_request:4350", "right": "pull_request:6314", "accept": false, - "reason": "4350 is a docs dtype example fix; 6314 fixes a broken paper link. Both are docs-only but not the same underlying issue." + "reason": "Different docs issues in the same broad area, but one is dtype in adapter docs and the other is a paper-link typo." }, { "left": "pull_request:4380", "right": "pull_request:4457", "accept": false, - "reason": "4380 is assorted training/docs minor fixes; 4457 updates a TensorRT inpaint pipeline example and README import usage. Different examples and code." + "reason": "Minor docs fixes across training guides vs a TensorRT inpaint pipeline README/code fix; not the same change." }, { "left": "pull_request:4437", "right": "pull_request:5828", "accept": false, - "reason": "4437 fixes consistency-model sampling bugs in UNet/pipeline code; 5828 only adds `@torch.no_grad()` to a VAE test. Not the same change." + "reason": "Consistency-model sampling bugfix vs a VAE test memory annotation; different code paths." }, { "left": "pull_request:4461", "right": "pull_request:6265", "accept": false, - "reason": "4461 adds a pipeline-class argument to a conversion script; 6265 refactors GLIGEN out of stable-diffusion namespaces. Separate refactor/CLI changes." + "reason": "Converter script argument addition vs GLIGEN refactor; unrelated changes." }, { "left": "pull_request:4472", "right": "pull_request:4503", "accept": false, - "reason": "4472 updates an SDXL DreamBooth README; 4503 clarifies LoRA docs and supported checkpoints. Different docs and topics." + "reason": "DreamBooth README edit vs LoRA docs clarification; different documents and content." }, { "left": "pull_request:4479", "right": "pull_request:6122", "accept": false, - "reason": "4479 fixes typos across several pipeline docstrings; 6122 is a large README typo sweep. Both are typo cleanups, but not the same concrete artifact." + "reason": "Pipeline typo fixes in code vs a broad docs typo sweep; same general theme but not the same bug/change." }, { "left": "pull_request:4503", "right": "pull_request:5267", "accept": false, - "reason": "4503 clarifies LoRA/SDXL support; 5267 fixes README markdown code fences. Different documentation concerns." + "reason": "LoRA SDXL support clarification vs docs README code-block markup changes; unrelated." }, { "left": "pull_request:4563", "right": "pull_request:6433", "accept": false, - "reason": "4563 adds Prompt2Prompt to community pipelines; 6433 adds GlueGen. Both community pipeline additions, but distinct features and code paths." + "reason": "Prompt2Prompt community pipeline addition vs Gluegen community pipeline addition; separate features." }, { "left": "pull_request:4582", "right": "pull_request:5063", "accept": false, - "reason": "4582 fixes a ControlNet SDXL docstring example; 5063 changes guidance_rescale docstrings and related example text across many pipelines. Related docstring area, but not one specific bug." + "reason": "ControlNet SDXL docstring fix vs guidance_rescale docstring updates across multiple pipelines; different doc issues." }, { "left": "pull_request:4612", "right": "pull_request:4626", "accept": false, - "reason": "4612 fixes links in controlling-generation docs; 4626 updates text-to-image training docs links. Different docs pages and link targets." + "reason": "Link-path correction in controlling_generation vs link updates in text2image docs; related topic but different files and fixes." }, { "left": "pull_request:4721", "right": "pull_request:5831", "accept": false, - "reason": "4721 is a Korean dreambooth docs cleanup; 5831 only adds a title in the English toctree. Not the same change." + "reason": "Korean DreamBooth docs cleanup vs English docs tree title fix; unrelated localization/docs maintenance." } ] }, @@ -136814,7 +136876,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 100, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -136904,10 +136966,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -136922,180 +136984,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: most items are unrelated small bugfix/docs PRs across different subsystems. I found no convincing duplicate pairs; all soft edges should be rejected.", - "confidence": 0.96, + "summary": "This cluster is a grab bag of largely unrelated PRs across docs, tests, examples, and pipeline fixes. The soft links look like broad keyword similarity rather than the same underlying bug or change, so I rejected all candidate duplicate edges.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a true duplicate hub here; if one must be picked, PR 6847 is the strongest anchor because it is a substantial feature PR with code, tests, docs, and an explicit issue target, unlike the many isolated typo/doc/test fixes.", + "canonical_pr_reason": "No clear canonical PR: the cluster spans unrelated fixes across docs, tests, training scripts, and pipelines.", "best_issue_reason": null, - "best_pr_reason": "PR 6847 is the best representative of the cluster because it is the most substantial and self-contained change, with clear user-facing scope and coverage across implementation, docs, and tests.", + "best_pr_reason": "PR 6847 is the most substantial standalone change here, with broad code/test/docs impact and high discussion, but it still is not a safe duplicate anchor.", "soft_edge_verdicts": [ { "left": "pull_request:4741", "right": "pull_request:6796", "accept": false, - "reason": "Unrelated fixes in different areas: AudioLDM2 dummy imports vs torchvision import gating in training utilities." + "reason": "AudioLDM2 dummy import fix vs torchvision import gating; unrelated bugs." }, { "left": "pull_request:4796", "right": "pull_request:5745", "accept": false, - "reason": "SDXL pipeline type-mismatch fix vs a documentation-only consistency decoder type correction; not the same bug or code path." + "reason": "SDXL type-mismatch fix vs a docstring type correction; different artifacts and problems." }, { "left": "pull_request:4796", "right": "pull_request:6475", "accept": false, - "reason": "SDXL pipeline casting bug vs Stable Video Diffusion VAE type doc update; different pipelines and different changes." + "reason": "Pipeline float16 upcasting issue vs VAE type doc update; not the same code-path bug." }, { "left": "pull_request:4796", "right": "pull_request:6647", "accept": false, - "reason": "Runtime SDXL type-mismatch fix vs a scheduler test expectation update; no shared underlying bug." + "reason": "SDXL pipeline dtype issue vs a scheduler test expectation update; unrelated." }, { "left": "pull_request:4847", "right": "pull_request:6105", "accept": false, - "reason": "Prompt Free Diffusion community example vs SDE Drag pipeline addition; different features and code paths." + "reason": "Prompt Free Diffusion example vs SDE Drag pipeline addition; different community examples." }, { "left": "pull_request:4862", "right": "pull_request:5698", "accept": false, - "reason": "ControlNet img2img MultiControlNet bug vs LoRA/xFormers attention processor bug; both touch attention-related code but fix different failures." + "reason": "MultiControlNet height/width bug vs LCM img2img fix; different pipelines and failures." }, { "left": "pull_request:4862", "right": "pull_request:6424", "accept": false, - "reason": "ControlNet img2img height/width bug vs PixArt weight conversion bug; unrelated subsystems." + "reason": "ControlNet img2img fix vs PixArt weight conversion bug; unrelated changes." }, { "left": "pull_request:4946", "right": "pull_request:5727", "accept": false, - "reason": "Docs overview entry for training scripts vs adding LCM scripts; not the same change." + "reason": "Training overview docs vs LCM script additions; not the same change." }, { "left": "pull_request:498", "right": "pull_request:925", "accept": false, - "reason": "Attention slicing method typo vs prompt weighting truncation fix; different functionality and failure modes." + "reason": "Attention-slicing disable fix vs lpw token padding logic; unrelated community/pipeline bugs." }, { "left": "pull_request:5041", "right": "pull_request:5184", "accept": false, - "reason": "SDXL refiner-only test coverage vs scheduler add_noise loop tests; both are tests but not the same issue." + "reason": "SDXL refiner-only tests vs scheduler consistency tests; different test targets." }, { "left": "pull_request:5065", "right": "pull_request:5505", "accept": false, - "reason": "Attention processor typo fix vs ControlNet naming typo fix; unrelated documentation/code corrections." + "reason": "Attention-processor typo/docs cleanup vs ControlNet naming typo; same subsystem label but different fixes." }, { "left": "pull_request:5066", "right": "pull_request:5505", "accept": false, - "reason": "Attention processor type-check typo fixes vs ControlNet pipeline class-name typo; different artifacts and issues." + "reason": "Attention-processor typo cleanup vs ControlNet naming typo; unrelated concrete changes." }, { "left": "pull_request:5207", "right": "pull_request:5652", "accept": false, - "reason": "T2I adapter README filename correction vs PixArt pipeline class-name correction; unrelated." + "reason": "T2I adapter README filename fix vs PixArt pipeline class-name fix; different files and issues." }, { "left": "pull_request:5236", "right": "pull_request:5745", "accept": false, - "reason": "Korean docs formatting cleanup vs ConsistencyDecoder doc type fix; different doc pages and distinct fixes." + "reason": "Korean docs formatting cleanup vs ConsistencyDecoder doc type fix; unrelated." }, { "left": "pull_request:5236", "right": "pull_request:7015", "accept": false, - "reason": "Korean docs formatting cleanup vs single-file loader doc/example adjustments; not the same underlying change." + "reason": "Korean docs formatting cleanup vs single-file SDXL doc example fix; unrelated docs changes." }, { "left": "pull_request:5286", "right": "pull_request:5844", "accept": false, - "reason": "Japanese docs page creation vs adding a Japanese SDXL reference example; related language area but different deliverables." + "reason": "Japanese docs addition vs an English docs reference example; not the same change." }, { "left": "pull_request:5407", "right": "pull_request:7096", "accept": false, - "reason": "Mixed-precision unconditional training fix vs ControlNet training enhancements; different training scripts and bugs." + "reason": "Mixed-precision training fix for unconditional generation vs ControlNet training enhancements; unrelated examples." }, { "left": "pull_request:5478", "right": "pull_request:5844", "accept": false, - "reason": "Japanese docs workflow/page setup vs adding a Japanese SDXL reference in English docs; not duplicates." + "reason": "Japanese docs workflow/content addition vs an English docs example addition; different documentation tasks." }, { "left": "pull_request:5695", "right": "pull_request:6215", "accept": false, - "reason": "PixArt mask-feature condition fix vs PixArt test initialization stabilization; one is a runtime bug, the other a test adjustment." + "reason": "PixArt mask-feature condition bug vs PixArt test seeding/expected slice updates; not the same underlying issue." }, { "left": "pull_request:5697", "right": "pull_request:6009", "accept": false, - "reason": "PEFT/xFormers attention processor bugfix vs TensorRT constructor argument fix; unrelated code paths." + "reason": "PEFT LoRA/xformers attention bug fix vs TensorRT super-args error; different code paths." }, { "left": "pull_request:5744", "right": "pull_request:6330", "accept": false, - "reason": "API docs typo cleanup vs SVD documentation rendering/content fix; both docs, but not the same change." + "reason": "API docs typo sweep vs SVD video-rendering doc fix; unrelated documentation edits." }, { "left": "pull_request:5828", "right": "pull_request:6680", "accept": false, - "reason": "Memory test annotation fix vs DPM-related slow test failure adjustments; different tests and problems." + "reason": "Consistency Decoder VAE test memory fix vs DPM scheduler slow-test fix; unrelated tests." }, { "left": "pull_request:5828", "right": "pull_request:7335", "accept": false, - "reason": "Memory test annotation fix vs controlnet SDXL config test fix; unrelated test maintenance." + "reason": "VAE test memory fix vs ControlNet SDXL config test; different modules and failures." }, { "left": "pull_request:5831", "right": "pull_request:6339", "accept": false, - "reason": "Docs title fix in toctree vs AnimateDiff docs update; different documentation changes." + "reason": "Generic docs title fix vs AnimateDiff docs fix; not the same artifact." }, { "left": "pull_request:591", "right": "pull_request:827", "accept": false, - "reason": "Flax safety checker save/load refactor vs broader PT/Flax API alignment; related framework area but not the same concrete change." + "reason": "Flax safety-checker save/load refactor vs PT/Flax API alignment; both Flax-related but different problems." }, { "left": "pull_request:6021", "right": "pull_request:7916", "accept": false, - "reason": "DreamBooth LoRA SDXL variant argument vs Ascend NPU support for SDXL training; different features." + "reason": "Dreambooth LoRA variant argument vs Ascend NPU support for SDXL; unrelated feature work." }, { "left": "pull_request:6140", "right": "pull_request:6847", "accept": false, - "reason": "IP-Adapter API doc page vs IP-Adapter attention masking feature; docs and feature are related but not the same PR-worthy fix." + "reason": "IP-Adapter API docs page vs IP-Adapter attention masking feature; docs versus code change, not one bug." }, { "left": "pull_request:6140", "right": "pull_request:6941", "accept": false, - "reason": "IP-Adapter API doc page vs SDXL ControlNet Inpaint IP-Adapter support; documentation and implementation are distinct." + "reason": "IP-Adapter docs page vs ControlNet Inpaint IP-Adapter support; same topic area but different changes." } ] }, @@ -137111,7 +137173,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 101, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -137207,10 +137269,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -137225,210 +137287,216 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: mostly unrelated docs, tests, import/path refactors, and independent feature or bugfix PRs. I do not see any safe duplicate/merge pairs among the soft edges.", - "confidence": 0.99, + "summary": "This set is highly heterogeneous: it contains unrelated docs-only fixes, test tweaks, dependency/import cleanup, scheduler additions, and separate IP-Adapter/controlnet changes. There isn\u2019t a single underlying bug or change tying the cluster together.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is canonical here; the items do not share one underlying bug or change, so any representative PR would be arbitrary.", + "canonical_pr_reason": "No single canonical PR fits: the artifacts span unrelated documentation edits, tests, refactors, and feature work across different subsystems.", "best_issue_reason": null, - "best_pr_reason": "No global best PR: the set spans unrelated changes across docs, tests, pipelines, loaders, and utilities, so none is a meaningful duplicate anchor.", + "best_pr_reason": "No best PR for deduplication; the items do not appear to be duplicates or mergeable into one concrete fix/change.", "soft_edge_verdicts": [ { "left": "pull_request:62", "right": "pull_request:848", "accept": false, - "reason": "Unrelated fixes: latent diffusion pipeline cleanup vs Flax stable diffusion width/height handling." + "reason": "Both are old pipeline fixes, but they address different code paths: one changes latent diffusion decoding, the other fixes Flax latent shape calculation." }, { "left": "pull_request:6205", "right": "pull_request:6314", "accept": false, - "reason": "Both touch T2I-Adapter docs, but they fix different links in different parts of the page." + "reason": "Both edit the same docs file, but one is a blog URL correction and the other fixes a paper link typo; unrelated documentation changes." }, { "left": "pull_request:6209", "right": "pull_request:7746", "accept": false, - "reason": "Example README test-script docs vs VAE test consistency are unrelated." + "reason": "One fixes an example README script, the other fixes a VAE test; same project area but not the same underlying bug." }, { "left": "pull_request:6244", "right": "pull_request:6847", "accept": false, - "reason": "IP-Adapter/gligen refactor is not the same change as attention masking support." + "reason": "Both touch IP-Adapter-related code, but 6244 is an embedding/refactor PR while 6847 adds attention masking support; different features." }, { "left": "pull_request:6244", "right": "pull_request:6941", "accept": false, - "reason": "Different IP-Adapter-related code paths: refactor vs SDXL ControlNet inpaint support." + "reason": "Both involve IP-Adapter support in ControlNet-ish pipelines, but they implement different pipelines/features and are not the same fix." }, { "left": "pull_request:626", "right": "pull_request:715", "accept": false, - "reason": "ONNX typing fix and PyTorch-import guard are separate compatibility fixes." + "reason": "One is an ONNX typing fix, the other is a PyTorch import guard; unrelated maintenance changes." }, { "left": "pull_request:6276", "right": "pull_request:8427", "accept": false, - "reason": "Community pipeline feature PR vs GitHub Actions workflow debug print are unrelated." + "reason": "One adds a community IP-Adapter FaceID pipeline, the other only prints env vars in a workflow; no shared bug/change." }, { "left": "pull_request:6293", "right": "pull_request:6847", "accept": false, - "reason": "SDXL ControlNet img2img IP-Adapter support is a different pipeline from masking docs/impl." + "reason": "Both are IP-Adapter related, but one adds IP-Adapter to SDXL ControlNet img2img and the other adds attention masking; different concrete changes." }, { "left": "pull_request:6293", "right": "pull_request:6941", "accept": false, - "reason": "These are different ControlNet pipelines and different IP-Adapter integrations." + "reason": "These target different ControlNet pipelines and different integration points for IP-Adapter support; not the same code-path fix." }, { "left": "pull_request:6339", "right": "pull_request:7779", "accept": false, - "reason": "AnimateDiff docs fix vs AutoPipeline docstring fix are unrelated." + "reason": "One updates AnimateDiff docs/example code, the other fixes AutoPipeline docstrings; unrelated documentation work." }, { "left": "pull_request:6669", "right": "pull_request:8775", "accept": false, - "reason": "README stats updates and Colab-link updates are separate documentation edits." + "reason": "Both are README updates, but one is repo stats wording and the other is a Colab link in a research README; unrelated." }, { "left": "pull_request:6754", "right": "pull_request:8882", "accept": false, - "reason": "Diffusion-DPO loss correction is unrelated to IP-Adapter loader doc text." + "reason": "The DPO training loss fix is unrelated to an IP-Adapter loader docstring correction." }, { "left": "pull_request:6754", "right": "pull_request:8896", "accept": false, - "reason": "Training-loss fix and single-file utility config plumbing are unrelated." + "reason": "The DPO training loss fix and the single-file controlnet utility call fix are different training/runtime issues." }, { "left": "pull_request:6762", "right": "pull_request:8422", "accept": false, - "reason": "Kandinsky test flakiness tweak is unrelated to subfolder-loading test coverage." + "reason": "One relaxes/flakes a Kandinsky test, the other adds subfolder-loading test coverage; different test concerns." }, { "left": "pull_request:6765", "right": "pull_request:8882", "accept": false, - "reason": "Hub cache path normalization is not the same as an IP-Adapter loader doc fix." + "reason": "Path-handling cache update in hub_utils is unrelated to an IP-Adapter loader docstring fix." }, { "left": "pull_request:6765", "right": "pull_request:8896", "accept": false, - "reason": "Hub cache path normalization and single-file utility change are separate path-handling fixes." + "reason": "Cache-path handling and single-file controlnet config wiring address different runtime paths." }, { "left": "pull_request:6766", "right": "pull_request:8882", "accept": false, - "reason": "Constants path cleanup is unrelated to IP-Adapter docs." + "reason": "Constants path normalization is unrelated to the IP-Adapter loader docstring change." }, { "left": "pull_request:6766", "right": "pull_request:8896", "accept": false, - "reason": "Constants path cleanup and single-file utility change are different code paths." + "reason": "A constants.py path update does not match the single-file controlnet utility fix." }, { "left": "pull_request:6767", "right": "pull_request:8882", "accept": false, - "reason": "Flax model path handling and IP-Adapter loader docs are unrelated." + "reason": "Flax model path handling and an IP-Adapter loader docstring correction are unrelated." }, { "left": "pull_request:6767", "right": "pull_request:8896", "accept": false, - "reason": "Flax pretrained-path handling is a different fix from controlnet single-file utility changes." + "reason": "Different modules and different failures: Flax loading vs controlnet single-file config conversion." }, { "left": "pull_request:6768", "right": "pull_request:8882", "accept": false, - "reason": "Configuration-utils path normalization is unrelated to IP-Adapter loader docs." + "reason": "Configuration utils path handling and IP-Adapter loader docs are unrelated changes." }, { "left": "pull_request:6768", "right": "pull_request:8896", "accept": false, - "reason": "Configuration-utils path handling and single-file utility changes are separate issues." + "reason": "These touch different utilities and fix different issues." }, { "left": "pull_request:6796", "right": "pull_request:8491", "accept": false, - "reason": "Torchvision import guard and DualTransformer2D import path fix are unrelated." + "reason": "One adds a torchvision availability guard; the other fixes a model import path. Same broad import area, but not the same bug." }, { "left": "pull_request:6808", "right": "pull_request:7746", "accept": false, - "reason": "MPS test-collection fix and VAE test dtype fix are different test issues." + "reason": "Testing utility device normalization is unrelated to the VAE test consistency fix." }, { "left": "pull_request:6837", "right": "pull_request:8675", "accept": false, - "reason": "I2VGenXL docs deps and fast diffusion benchmark note are unrelated documentation changes." + "reason": "One updates I2VGenXL docs dependencies; the other changes a fast-diffusion tutorial note. Both docs, but unrelated content." }, { "left": "pull_request:6871", "right": "pull_request:7747", "accept": false, - "reason": "Adding VAE slicing methods is not the same as fixing the VAE tiling test assertion." + "reason": "VAE slicing/tiling feature work is different from a test assertion fix for tiled VAE decoding." + }, + { + "left": "pull_request:6941", + "right": "pull_request:6762", + "accept": false, + "reason": "ControlNet SDXL IP-Adapter support and a Kandinsky test flake fix are unrelated." }, { "left": "pull_request:6962", "right": "pull_request:7564", "accept": false, - "reason": "PixArt slow-test adjustments and assorted pipeline test fixes do not target the same bug." + "reason": "PixArt slow test fixes are unrelated to the broader Animatediff/test compatibility cleanup." }, { "left": "pull_request:7015", "right": "pull_request:9015", "accept": false, - "reason": "SDXL single-file docs and PIA example docs are different examples and fixes." + "reason": "Different doc examples for different pipelines; no shared underlying bug or change." }, { "left": "pull_request:7120", "right": "pull_request:7319", "accept": false, - "reason": "EDM scheduler implementation and EDM scheduler docs are separate changes, not one bugfix." + "reason": "EDM scheduler code addition versus docs-only scheduler pages are not the same change." }, { "left": "pull_request:7206", "right": "pull_request:8630", "accept": false, - "reason": "Training-script tagging cleanup is unrelated to SD3 DreamBooth LoRA text-encoder support." + "reason": "Training-script tagging across many examples is unrelated to SD3 DreamBooth LoRA text-encoder training." }, { "left": "pull_request:7226", "right": "pull_request:7780", "accept": false, - "reason": "IP-Adapter API/docs expansion is different from later masking/face-id documentation edits." + "reason": "Both touch IP-Adapter docs, but 7226 adds API/embedding docs while 7780 edits masking wording/example text; not the same fix." }, { "left": "pull_request:7259", "right": "pull_request:7360", "accept": false, - "reason": "TCD intro docs and TCD scheduler doc comments/types are different documentation tasks." + "reason": "TCD intro docs and TCD scheduler docstring cleanup are different artifacts with different purposes." }, { "left": "pull_request:7272", "right": "pull_request:7302", "accept": false, - "reason": "Custom Diffusion PEFT loader bug and text-encoder gradient bug are distinct fixes." + "reason": "Custom Diffusion PEFT loader fix and the training-script gradient-index bug are different code paths and different issues." } ] }, @@ -137444,7 +137512,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 102, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -137535,10 +137603,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -137553,186 +137621,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: it mixes docs-only edits, test-only fixes, workflow tweaks, and several unrelated model/pipeline feature PRs across different subsystems. I do not see any pair that safely looks like the same underlying bug or change.", - "confidence": 0.96, + "summary": "Loose cluster of mostly unrelated small docs, tests, workflow, and feature PRs; no soft pair looks like a true duplicate of the same underlying change.", + "confidence": 0.16, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a safe canonical for this cluster; the items span unrelated topics and code paths rather than one duplicated fix or feature.", + "canonical_pr_reason": "If forced to pick one PR as the closest umbrella representative, PR 9008 is the broadest errata/cleanup item; however, the cluster is too heterogeneous for a real canonical PR.", "best_issue_reason": null, - "best_pr_reason": "PR 9408 is the strongest standalone change in scope and quality (explicit issue target, substantial implementation, and tests), but it is not a good canonical anchor for the whole cluster.", + "best_pr_reason": "PR 9008 is the best single representative only because it is the broadest generic cleanup/docs PR in the set, not because it matches the others as a duplicate fix.", "soft_edge_verdicts": [ { "left": "pull_request:7384", "right": "pull_request:8851", "accept": false, - "reason": "Both are docs PRs, but they touch different pages/features (FreeInit outputs vs AuraFlow docs) and are not the same change." + "reason": "Both are docs-related, but one adds FreeInit outputs and the other adds AuraFlow docs; different features." }, { "left": "pull_request:7425", "right": "pull_request:8737", "accept": false, - "reason": "A small docs output-image fix is unrelated to a broad workflow/benchmark cleanup PR." + "reason": "Missing output image docs vs a large workflow/repo cleanup PR; not the same change." }, { "left": "pull_request:7425", "right": "pull_request:9066", "accept": false, - "reason": "Different documentation topics and files; no shared bug or duplicated change." + "reason": "Both touch docs, but one is a tutorial image fix and the other adds Stable Cascade docs; unrelated." }, { "left": "pull_request:7464", "right": "pull_request:8498", "accept": false, - "reason": "One fixes an example import path, the other fixes a doc typo; unrelated changes." + "reason": "Small LDM3D import fix vs an SD3 docstring typo; different artifacts and code paths." }, { "left": "pull_request:7530", "right": "pull_request:8946", "accept": false, - "reason": "A training autocast bugfix is not the same as a test assertion adjustment." + "reason": "Training autocast bugfix vs a test expectation change; unrelated." }, { "left": "pull_request:7627", "right": "pull_request:8720", "accept": false, - "reason": "Both affect img2img/testing around latents, but one fixes pipeline latent handling while the other updates precision-sensitive test expectations." + "reason": "ControlNet img2img latent handling vs slow test precision updates; not the same bug." }, { "left": "pull_request:7719", "right": "pull_request:7918", "accept": false, - "reason": "Workflow secret wiring and a Flax attention reshape fix are unrelated code paths." + "reason": "Workflow secret fix vs a Flax attention reshape bugfix; completely different changes." }, { "left": "pull_request:7746", "right": "pull_request:9627", "accept": false, - "reason": "A VAE test dtype fix and a LoRA warning-capture test change are different test issues." + "reason": "VAE test dtype adjustment vs LoRA warning-test reordering; unrelated test fixes." }, { "left": "pull_request:7746", "right": "pull_request:9683", "accept": false, - "reason": "A VAE test dtype fix and a CogVideoX integration test rename/unskip are unrelated." + "reason": "VAE test dtype fix vs CogVideoX test rename/unskip; different test issues." }, { "left": "pull_request:7827", "right": "pull_request:9765", "accept": false, - "reason": "Docstring CPU-offload wording and SD3 pipeline docstring corrections are not the same change." + "reason": "Both are doc changes, but one fixes CPU offload wording and the other fixes SD3 pipeline docstrings; not the same change." }, { "left": "pull_request:798", "right": "pull_request:948", "accept": false, - "reason": "A docstring typo in UNet is unrelated to a DDPM comment typo." + "reason": "Unet docstring typo vs DDPM comment typo; unrelated." }, { "left": "pull_request:8427", "right": "pull_request:9765", "accept": false, - "reason": "A workflow debug-print addition and SD3 docstring fixes are different areas and changes." + "reason": "Workflow debug print vs SD3 docstring typo; no shared bug." }, { "left": "pull_request:8470", "right": "pull_request:9553", "accept": false, - "reason": "A README/project rename update is unrelated to distributed inference documentation." + "reason": "Different documentation updates for different topics; not one underlying change." }, { "left": "pull_request:8479", "right": "pull_request:9408", "accept": false, - "reason": "Both touch export/init surfaces, but they implement different features in different model families (HunyuanDiT controlnet vs Flux ControlNet pipelines)." + "reason": "Both add controlnet-related features, but for different models and code paths; too broad to merge as one PR." }, { "left": "pull_request:8481", "right": "pull_request:8836", "accept": false, - "reason": "An image_processor error-message fix is unrelated to a sliced attention loop bugfix." + "reason": "Image processor error message vs sliced attention loop off-by-one fix; unrelated." }, { "left": "pull_request:8497", "right": "pull_request:9107", "accept": false, - "reason": "A scheduler spelling fix is not the same as a DreamBooth SD3 loading-hook bugfix." + "reason": "Scheduler spelling fix vs SD3 loading-hook logic bug; different artifacts." }, { "left": "pull_request:8838", "right": "pull_request:9196", "accept": false, - "reason": "A transformer SD3 forward-chunking helper and a PixArt attention-processor helper are different model classes and changes." + "reason": "SD3 forward chunking support vs PixArt attention-processor helper; different model features." }, { "left": "pull_request:8845", "right": "pull_request:9100", "accept": false, - "reason": "Typo fixes in PAG tests and a typo fix in an example argument help string are not the same artifact." + "reason": "PAG test typo vs a LoRA example typo; unrelated test/docs edits." }, { "left": "pull_request:8845", "right": "pull_request:9683", "accept": false, - "reason": "PAG test typo fixes are unrelated to CogVideoX image-to-video integration test changes." + "reason": "PAG test typo vs CogVideoX test rename/unskip; not the same issue." }, { "left": "pull_request:8951", "right": "pull_request:9765", "accept": false, - "reason": "A DreamBooth SD3 deletion-condition fix is unrelated to SD3 pipeline docstring corrections." + "reason": "SD3 training code deletion logic bug vs SD3 pipeline docstring typo; code vs docs." }, { "left": "pull_request:8984", "right": "pull_request:9765", "accept": false, - "reason": "An ONNX conversion feature for SD3 is a different change from a pipeline docstring fix." + "reason": "SD3 ONNX conversion support vs a docstring fix; different scope entirely." }, { "left": "pull_request:9008", "right": "pull_request:9077", "accept": false, - "reason": "These are both typo/errata-style PRs, but they target different docs and examples with no shared bug." + "reason": "Broad errata/docs cleanup vs isolated typo fixes; not one duplicate change." }, { "left": "pull_request:9008", "right": "pull_request:9739", "accept": false, - "reason": "General typo/errata cleanup is not the same as adding SD3 single-file loading docs and code." + "reason": "General errata cleanup vs SD3 docs typo and single-file checkpoint docs; unrelated." }, { "left": "pull_request:9152", "right": "pull_request:9266", "accept": false, - "reason": "Both are docs, but one is distributed inference guidance and the other is the community projects table." + "reason": "Distributed inference tutorial expansion vs community projects list update; different docs additions." }, { "left": "pull_request:9408", "right": "pull_request:9571", "accept": false, - "reason": "Both are Flux ControlNet-related, but one adds new img2img/inpaint pipelines while the other adds control guidance scheduling to existing pipelines; different code-path problems." + "reason": "Flux ControlNet pipeline additions vs Flux ControlNet guidance-window parameters; same subsystem, but different concrete features." }, { "left": "pull_request:9408", "right": "pull_request:9658", "accept": false, - "reason": "A core Flux ControlNet feature PR and a community FLUX distilled ControlNet example are not the same underlying change." + "reason": "Flux ControlNet pipeline feature vs a community example script; not the same underlying change." }, { "left": "pull_request:948", "right": "pull_request:972", "accept": false, - "reason": "A DDPM comment typo and a README torch_dtype typo are unrelated docs fixes." + "reason": "DDPM comment typo vs README argument-name typo; unrelated." }, { "left": "pull_request:12086", "right": "pull_request:12387", "accept": false, - "reason": "Flux modular I2I support and flash-attention-kernels backend support are entirely different features." + "reason": "Flux modular I2I support vs FlashAttention-kernels support; different features." }, { "left": "pull_request:12006", "right": "pull_request:12909", "accept": false, - "reason": "Both involve WAN, but one adds a 5B I2V pipeline and the other fixes WAN 2.1 context-parallel behavior; different versions and code paths." + "reason": "Wan 5B I2V feature addition vs a WAN context-parallel fix; related area, but not the same bug or merged change." } ] }, @@ -137748,7 +137816,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 103, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -137834,10 +137902,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -137852,163 +137920,166 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: mostly unrelated documentation, test, LoRA, quantization, and pipeline fixes across different subsystems. I do not see a true duplicate group here, so all soft-edge candidates should be rejected.", - "confidence": 0.98, + "summary": "This cluster is mostly a grab bag of unrelated PRs across docs, quantization, Wan/Qwen, checkpoints, tests, and hardware-specific fixes. The only plausible duplicate-like pair is the two LoRA torch.compile fixes, which address the same underlying compile-compatibility problem in the LoRA conv path.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR is a good canonical for the whole cluster; the items are too heterogeneous, with only a small LoRA compile-fix subcluster overlapping.", "best_issue_reason": null, - "best_pr_reason": "No single PR is a strong canonical representative; the items span unrelated changes, and the soft links are only loose topical similarity rather than the same underlying bug or feature.", + "best_pr_reason": "No global best PR for the cluster because there is no single underlying bug/change tying the items together.", "soft_edge_verdicts": [ { "left": "pull_request:6946", "right": "pull_request:9073", "accept": false, - "reason": "Different features: IP-Adapter low_cpu_mem_usage support vs SparseControlNet single-file loading." + "reason": "Different features in different areas: IP-Adapter low_cpu_mem_usage loading vs SparseControlNet single-file loading support." }, { "left": "pull_request:6868", "right": "pull_request:7189", "accept": false, - "reason": "Both touch adapter/embed plumbing, but they fix different problems and code paths; not the same concrete bug." + "reason": "Both are IP-Adapter related, but one adds image_embeds input support and the other fixes plus-version embedding handling; not the same concrete change." }, { "left": "pull_request:9213", "right": "pull_request:9964", "accept": false, - "reason": "Different quantization backends and goals: bitsandbytes support vs GGUF single-file support." + "reason": "Different quantization backends and changes: bitsandbytes quantization support vs GGUF support." }, { "left": "pull_request:12416", "right": "pull_request:13280", "accept": false, - "reason": "Unrelated subsystems: QwenImage Edit Plus modular support vs FlashAttention-4 backend support." + "reason": "Unrelated subsystems: QwenImage Edit Plus modular pipeline support vs FlashAttention-4 support." }, { "left": "pull_request:12074", "right": "pull_request:12324", "accept": false, - "reason": "Wan LoRA second-transformer loading is distinct from the Wan2.2 VACE conversion/model support PR." + "reason": "Both touch Wan, but one adds second-transformer LoRA loading and the other adds Wan2.2 VACE-Fun conversion/model support; different changes." }, { "left": "pull_request:2367", "right": "pull_request:3652", "accept": false, - "reason": "Both involve checkpointing, but one adds a total-limit option while the other refactors checkpoint naming/layout." + "reason": "Both mention checkpoints, but one adds a checkpoint limit argument and the other refactors checkpoint naming; not the same bug or change." }, { "left": "pull_request:7632", "right": "pull_request:9061", "accept": false, - "reason": "DreamBooth VAE path fix is unrelated to sharded checkpoint loading with variants." + "reason": "Different problems: DreamBooth VAE loading path issue vs sharded-checkpoint loading with variants." }, { "left": "pull_request:2467", "right": "pull_request:823", "accept": false, - "reason": "Latent output type support is unrelated to the negative-prompt type mismatch fix." + "reason": "Different pipeline bugs: latent output type addition vs negative-prompt type mismatch handling." }, { "left": "pull_request:3299", "right": "pull_request:4761", "accept": false, - "reason": "Different areas: LoRA training docs/example cleanup vs Paint-by-Example image encoder fix." + "reason": "Docs/example sync for LoRA training vs Paint-by-Example image encoder fix; unrelated." }, { "left": "pull_request:4315", "right": "pull_request:5298", - "accept": false, - "reason": "Both are LoRA compile-related, but they address different conv forward paths and are not the same bug." + "accept": true, + "reason": "Same underlying LoRA torch.compile compatibility problem in the conv path; one fixes the no-LoRA branch and the other fixes the LoRA-enabled branch, so they plausibly belong in one PR." }, { "left": "pull_request:4508", "right": "pull_request:5948", "accept": false, - "reason": "from_pretrained use_onnx plumbing is unrelated to the deprecated-variant warning condition." + "reason": "Different pipeline_utils issues: use_onnx propagation vs variant warning logic." }, { "left": "pull_request:4795", "right": "pull_request:4862", "accept": false, - "reason": "ControlNet torch.compile fix vs MultiControlNet image-shape handling are different issues." + "reason": "Different ControlNet bugs: torch.compile condition handling vs SDXL ControlNet img2img shape handling for MultiControlNetModel." }, { "left": "pull_request:5738", "right": "pull_request:5958", "accept": false, - "reason": "LoRA delete-adapter feature and load_lora support refactor are distinct changes." + "reason": "Different LoRA features: delete adapters vs load_lora support refactor." }, { "left": "pull_request:6217", "right": "pull_request:7786", "accept": false, - "reason": "Stable Video Diffusion VAE compile fix vs num_videos_per_prompt behavior fix are different bugs." + "reason": "Same Stable Video Diffusion area but different issues: VAE compile compatibility vs num_videos_per_prompt behavior." }, { "left": "pull_request:823", "right": "pull_request:826", "accept": false, - "reason": "Negative-prompt type checking and DreamBooth fp16 loss casting are unrelated." + "reason": "Different bugs in different code paths: negative-prompt type checking vs DreamBooth fp16 loss casting." }, { "left": "pull_request:4899", "right": "pull_request:5127", "accept": false, - "reason": "ControlNet LoRA support and MultiAdapter support are different feature additions." + "reason": "Related adapter/controlnet ecosystem, but one is ControlNet LoRA support and the other is Multi-Adapter support; not the same change." }, { "left": "pull_request:10230", "right": "pull_request:10544", "accept": false, - "reason": "Different docs additions for different topics; not the same underlying change." + "reason": "Both docs PRs, but they cover unrelated topics: LoRA loader docs vs ParaAttention optimization docs." }, { "left": "pull_request:10318", "right": "pull_request:11595", "accept": false, - "reason": "Both are Flux LoRA-related, but one adds tests and the other fixes a loader bug; not the same concrete change." + "reason": "Flux LoRA integration tests vs a LoRA loading fix; same broad domain but different concrete changes." }, { "left": "pull_request:11479", "right": "pull_request:12451", "accept": false, - "reason": "Intel Gaudi doc update vs doc-builder Dockerfile dependency changes are unrelated." + "reason": "Unrelated docs/config changes: Intel Gaudi wording update vs doc-builder Dockerfile packages." }, { "left": "pull_request:11843", "right": "pull_request:12541", "accept": false, - "reason": "Wan VACE exclude_modules tests are unrelated to the latest few-step Wan LoRA conversion fix." + "reason": "Both Wan LoRA related, but one is tests for exclude_modules and the other is support for latest few-step Wan LoRA; different changes." }, { "left": "pull_request:12681", "right": "pull_request:13042", "accept": false, - "reason": "Different test fixes in different subsystems; no shared underlying bug." + "reason": "Different CI/test fixes in unrelated areas: CPU pipeline tests vs QwenImage prompt isolation." }, { "left": "pull_request:12782", "right": "pull_request:13211", "accept": false, - "reason": "HunyuanVideo offloading tests and group-offloading runtime fix are not the same artifact." + "reason": "Both involve offloading tests/fixes, but for different models and different bugs; not the same code-path problem." }, { "left": "pull_request:2172", "right": "pull_request:3080", "accept": false, - "reason": "Validation-in-multi-GPU and save_pretrained None-handling are unrelated." + "reason": "Different training/pipeline concerns: multi-GPU validation behavior vs save_pretrained handling of None entries." }, { "left": "pull_request:2464", "right": "pull_request:3339", "accept": false, - "reason": "Training-script xformers cleanup vs urllib3 pinning are unrelated maintenance changes." + "reason": "Unrelated changes: xformers handling in a LoRA training script vs urllib3 dependency pinning." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: it correctly characterizes the cluster as mostly heterogeneous and only treats the LoRA torch.compile pair as plausibly mergeable for the same bugfix. The accepted soft-edge verdict on 4315/5298 is supported by the shared file and matching compile-fix theme, and the false verdicts avoid overclaiming duplicates." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -138017,7 +138088,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 104, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -138103,10 +138174,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -138121,156 +138192,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated PRs; the soft-similarity pairs share broad subsystems or testing/doc themes, but they are not the same concrete bug or change.", - "confidence": 0.96, + "summary": "This cluster is mostly a set of unrelated PRs that only weakly resemble one another by broad keywords like tests, docs, IP-Adapter, or model-family names. I found no true duplicate pair under a strict same-bug/same-change standard.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 10261 is the strongest representative: it\u2019s a broad, central Flux IP-Adapter feature with multiple codepaths touched and explicit issue links, unlike the others which are smaller or more isolated.", + "canonical_pr_reason": "PR 7016 is the strongest representative of the only recurring technical theme here (IP-Adapter plumbing): it is broad, well-reviewed, and its changes are reused by later, narrower IP-Adapter follow-ups.", "best_issue_reason": null, - "best_pr_reason": "PR 10261 is the best overall PR candidate for the cluster because it is the most substantial cross-cutting change and has the clearest linkage to tracked work.", + "best_pr_reason": "PR 7016 is the best global representative because it has the widest functional surface area and the clearest downstream relevance among the items; the rest are mostly isolated fixes, docs, or tests.", "soft_edge_verdicts": [ { "left": "pull_request:2983", "right": "pull_request:3847", "accept": false, - "reason": "One is documentation for installing test dependencies; the other adds a dependency-check test workflow and related files. Related area, but different change." + "reason": "Docs for installing test deps vs adding a dependency-check test workflow; different changes and different failure modes." }, { "left": "pull_request:3127", "right": "pull_request:4886", "accept": false, - "reason": "Both touch safety checker behavior, but one fixes img2img processing and the other fixes upscale pipeline postprocessing. Different code paths." + "reason": "Both touch safety checker behavior, but one fixes img2img processing and the other fixes upscale safety-checker input handling; not the same code path." }, { "left": "pull_request:6346", "right": "pull_request:7096", "accept": false, - "reason": "PR 6346 is a training-example follow-up for LoRA state dict handling; PR 7096 is ControlNet training README/script enhancements. Not the same bug or feature." + "reason": "Training-example follow-up vs ControlNet training enhancements; unrelated scripts and functionality." }, { "left": "pull_request:6864", "right": "pull_request:7328", "accept": false, - "reason": "One adds docs for torch.compile with PEFT; the other changes PyPI publish dependencies. Same general tooling space, but unrelated changes." + "reason": "One is a docs note about torch.compile with LoRA fusion; the other is a packaging workflow tweak adding torch to a publish step." }, { "left": "pull_request:7103", "right": "pull_request:7156", "accept": false, - "reason": "ControlNet image validation fix vs stalebot label normalization. No shared underlying bug." + "reason": "ControlNet input validation bug vs stale-bot label normalization; no shared underlying bug." }, { "left": "pull_request:7246", "right": "pull_request:7458", "accept": false, - "reason": "VAE tiling test device fix and Python 3.12 dynamo test skipping are unrelated test-maintenance changes." + "reason": "VAE tiling test-device fix vs skipping dynamo tests on Python 3.12; unrelated test maintenance." }, { "left": "pull_request:7932", "right": "pull_request:8797", "accept": false, - "reason": "Both are test adjustments, but one updates LoRA expected values and the other changes sharded checkpoint test behavior. Different problems." + "reason": "LoRA test expectation update vs sharded checkpoint test fixes; different test areas and causes." }, { "left": "pull_request:5477", "right": "pull_request:5559", "accept": false, - "reason": "Both concern latent consistency img2img, but one adds a community pipeline and the other adds the official pipeline implementation. Not the same PR-worthy change." + "reason": "Related family name, but one adds a community pipeline and the other adds an img2img pipeline; different features and code paths." }, { "left": "pull_request:5010", "right": "pull_request:6787", "accept": false, - "reason": "Gligen export/API fix vs Marigold input normalization bug; no common underlying code issue." + "reason": "GLIGEN import/export fix vs Marigold input-range correction; unrelated subsystems." }, { "left": "pull_request:6355", "right": "pull_request:7613", "accept": false, - "reason": "Widget-format cleanup for one advanced training script vs broader model-card standardization. Different scope and behavior." + "reason": "Both are advanced-training docs/card tweaks, but they fix different scripts and different metadata problems." }, { "left": "pull_request:6828", "right": "pull_request:7820", "accept": false, - "reason": "Both mention IP-Adapter, but they fix different model/pipeline integrations: UNetMotionModel internals vs InstructPix2Pix pipeline wiring." + "reason": "Both mention IP-Adapter, but one changes UNetMotionModel internals and the other wires IP-Adapter into InstructPix2Pix pipelines; not one concrete bug." }, { "left": "pull_request:4083", "right": "pull_request:4098", "accept": false, - "reason": "Low-precision VAE support for SDXL is unrelated to adding SDXL inpainting support, even though both touch SDXL pipelines." + "reason": "SDXL VAE precision support vs adding SDXL inpainting; these are separate features and not mergeable as one fix." }, { "left": "pull_request:9175", "right": "pull_request:9542", "accept": false, - "reason": "Flux Union ControlNet support and Flux LoRA text-encoder support are separate features for different mechanisms." + "reason": "Union ControlNet support vs Kohya Flux LoRA text-encoder conversion; completely different model/tooling changes." }, { "left": "pull_request:12024", "right": "pull_request:12041", "accept": false, - "reason": "Wan follow-up changes and VAE patchify fixes are distinct code-path fixes, not one underlying bug." + "reason": "Wan2.2 follow-up pipeline/model fix vs VAE patchify correction; different bugs in different components." }, { "left": "pull_request:7016", "right": "pull_request:8901", "accept": false, - "reason": "IP-Adapter embed refactor across many pipelines vs adding Kolors IP-Adapter support. Same subsystem, but clearly different feature work." + "reason": "Both are IP-Adapter-related, but 7016 is a broad loader/API refactor while 8901 adds Kolors support with its own image-encoder requirements; not the same change." }, { "left": "pull_request:10261", "right": "pull_request:10717", "accept": false, - "reason": "Both are Flux IP-Adapter work, but 10261 adds core support while 10717 extends img2img/control pipelines. Adjacent, not the same concrete change." + "reason": "Both concern Flux IP-Adapter, but one is the core support PR and the other adds img2img/control-pipeline wiring; separate code paths." }, { "left": "pull_request:10366", "right": "pull_request:11756", "accept": false, - "reason": "SANA offload/gpu-test cleanup vs SANA test fixes. Same project area, but different bugs and edits." + "reason": "Sana transformer/offload fix vs SANA test cleanup; unrelated." }, { "left": "pull_request:10854", "right": "pull_request:11124", "accept": false, - "reason": "Hunyuan DiT test module rename and CogView4 test tokenizer update are unrelated test changes." + "reason": "Hunyuan DiT test-module rename vs CogView4 test update; different test targets." }, { "left": "pull_request:1269", "right": "pull_request:192", "accept": false, - "reason": "Stable Diffusion refactor/finalization vs README parameter correction. Completely different change types." + "reason": "Stable Diffusion refactor cleanup vs README example parameter update; unrelated scope." }, { "left": "pull_request:1527", "right": "pull_request:192", "accept": false, - "reason": "Inpaint tensor-input support vs README example parameter update. No shared bug or patch." + "reason": "Inpaint tensor-input support vs README example parameter update; one is code behavior, the other is documentation." }, { "left": "pull_request:2915", "right": "pull_request:5093", "accept": false, - "reason": "Textual inversion tolerance change and moving slow tests to nightly are both test housekeeping, but not the same issue." + "reason": "Text inversion tolerance test tweak vs moving many slow tests to nightly; separate test-policy changes." }, { "left": "pull_request:3316", "right": "pull_request:3707", "accept": false, - "reason": "Patch-release doc workflow branch filter vs secret-handling in doc workflows. Both CI/docs, but different operational fixes." + "reason": "Both are documentation workflow changes, but one changes release branch matching and the other switches secrets/workflow triggering; distinct fixes." }, { "left": "pull_request:3850", "right": "pull_request:4845", "accept": false, - "reason": "AudioDiffusion test fixture fix and SD inpaint get_dummy_inputs overhaul are unrelated test fixes." + "reason": "AudioDiffusion test adjustments vs Stable Diffusion inpaint input handling; unrelated pipelines." }, { "left": "pull_request:4989", "right": "pull_request:4999", "accept": false, - "reason": "Wuerstchen num_images_per_prompt bug and compel-usage docs/behavior update are separate changes in the same area, not one concrete fix." + "reason": "Two Wuerstchen PRs, but one fixes num_images_per_prompt and the other fixes compel usage; different bugs and merge intents." } ] }, @@ -138286,7 +138357,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 105, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -138374,10 +138445,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -138392,162 +138463,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: most items are unrelated docs, training, loader, or pipeline PRs that only share broad keywords. No soft pair looks like a true duplicate or mergeable same-change candidate.", - "confidence": 0.94, + "summary": "This cluster is heterogeneous: it mixes unrelated docs, loader fixes, pipeline features, and example scripts across different subsystems. The soft links are title/subsystem similarity only, not the same underlying bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a canonical duplicate target; the items span unrelated fixes/features across different subsystems and models.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is globally best for the cluster because there is no coherent underlying bug/change tying these artifacts together.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:5030", "right": "pull_request:5702", "accept": false, - "reason": "Both are bug fixes, but they address different code paths: model offload vs AnimateDiff prompt handling." + "reason": "Different bugs: model offload key handling vs AnimateDiff prompt handling." }, { "left": "pull_request:6009", "right": "pull_request:6099", "accept": false, - "reason": "Both mention superclass arguments, but the PRs fix different pipelines and different constructor mismatches." + "reason": "Different TensorRT and TextToVideoZero constructor fixes; unrelated code paths." }, { "left": "pull_request:6191", "right": "pull_request:8490", "accept": false, - "reason": "Both are README edits for different examples; no shared bug or feature change." + "reason": "Both are README updates for different examples, not the same change." }, { "left": "pull_request:6499", "right": "pull_request:7226", "accept": false, - "reason": "Both are docs/API updates for different adapter types; not the same underlying change." + "reason": "Different docs/API work: PEFT docs vs IP-Adapter embedding support." }, { "left": "pull_request:8688", "right": "pull_request:8691", "accept": false, - "reason": "Both touch LoRA loading, but one fixes DoRA conversion and the other fixes vanilla fine-tuned LoRA loading." + "reason": "Both touch LoRA loading, but one fixes DoRA conversion and the other vanilla loading logic; not the same bug." }, { "left": "pull_request:8916", "right": "pull_request:8985", "accept": false, - "reason": "Both are test-only changes, but they cover different models and different test adjustments." + "reason": "Different targets: adding Hunyuan DiT tests vs resizing Lumina test models." }, { "left": "pull_request:11726", "right": "pull_request:12679", "accept": false, - "reason": "Both add LoRA support, but for different models and separate implementation paths." + "reason": "Both are LoRA-related, but Wan LoRA conversion and ChronoEdit multi-LoRA support are separate features." }, { "left": "pull_request:4062", "right": "pull_request:5057", "accept": false, - "reason": "One adds Shap-E mesh output, the other adds clip-skip support for SDXL; unrelated changes." + "reason": "Shap-E mesh output vs SDXL clip-skip support are unrelated pipeline features." }, { "left": "pull_request:4944", "right": "pull_request:6828", "accept": false, - "reason": "Both involve IP-Adapter work, but they fix different internals and different model integration points." + "reason": "IP-Adapter support and UNetMotionModel image-conditioning fixes are different implementations." }, { "left": "pull_request:10454", "right": "pull_request:11428", "accept": false, - "reason": "Both are Hunyuan-related, but one fixes an attention-mask dimension bug while the other adds Framepack functionality." + "reason": "Hunyuan Video attention-mask dim fix vs Framepack pipeline addition are not the same code-path problem." }, { "left": "pull_request:11485", "right": "pull_request:12249", "accept": false, - "reason": "Both add new pipelines, but they are entirely different models/features with no shared bug or code path." + "reason": "Both add pipelines, but Flex.2-preview and Step1X-Edit are distinct models and codebases." }, { "left": "pull_request:10205", "right": "pull_request:10378", "accept": false, - "reason": "Both involve training/LoRA scripts, but they solve different training setup issues in different pipelines." + "reason": "InstructPix2Pix LoRA refactor and SD3 LoRA training distributed support are unrelated training scripts." }, { "left": "pull_request:10205", "right": "pull_request:11059", "accept": false, - "reason": "One is a training-script LoRA feature, the other is a Wan LoRA conversion utility; different changes." + "reason": "Different LoRA work: InstructPix2Pix training refactor vs Wan LoRA conversion support." }, { "left": "pull_request:10378", "right": "pull_request:9993", "accept": false, - "reason": "Both edit the SD3 DreamBooth LoRA script, but they fix different checkpoint-saving behaviors." + "reason": "Same file family, but different SD3 checkpoint-saving fixes and distinct save-path logic." }, { "left": "pull_request:10544", "right": "pull_request:9244", "accept": false, - "reason": "Docs for ParaAttention are unrelated to the Flux single-file pipeline loading change." + "reason": "ParaAttention docs vs Flux single-file pipeline support are unrelated changes." }, { "left": "pull_request:12451", "right": "pull_request:13100", "accept": false, - "reason": "Different areas entirely: Dockerfile package install vs modular pipeline documentation/state changes." + "reason": "Dockerfile dependency update vs ModularPipeline docs/runtime refactor are unrelated." }, { "left": "pull_request:2479", "right": "pull_request:4396", "accept": false, - "reason": "Both are prompt/LoRA-related, but one adds text-encoder LoRA training support and the other fixes prompt encoding without a text encoder instance." + "reason": "Text-encoder LoRA training support vs prompt-encoding fallback logic are not the same bug." }, { "left": "pull_request:4910", "right": "pull_request:6330", "accept": false, - "reason": "These are separate documentation fixes for different guides and different examples." + "reason": "Inpainting docs typo fix vs SVD docs rendering fix are separate documentation edits." }, { "left": "pull_request:5157", "right": "pull_request:5470", "accept": false, - "reason": "One fixes LoRA key conversion, the other fixes ControlNet inpaint docs/code; no common bug." + "reason": "Different ControlNet/LoRA conversion fixes; one is kohya conversion parsing, the other is SDXL inpaint docs/code cleanup." }, { "left": "pull_request:6241", "right": "pull_request:6904", "accept": false, - "reason": "6241 lacks diff evidence, and load_image PIL support is a distinct utility change from the described runtime error." + "reason": "Runtime error report vs load_image PIL support fix; no shared underlying change." }, { "left": "pull_request:6972", "right": "pull_request:7158", "accept": false, - "reason": "One is an IP-Adapter unload test fix; the other is an LCM benchmark cleanup." + "reason": "IP-Adapter unload test correction vs LCM benchmark fix are unrelated." }, { "left": "pull_request:7476", "right": "pull_request:9311", "accept": false, - "reason": "Both mention ControlNet/SDXL, but one fixes torch.compile multi-controlnet logic and the other adds CFG cutoff callbacks." + "reason": "Different SDXL ControlNet issues: multi-controlnet compile behavior vs CFG cutoff callbacks." }, { "left": "pull_request:9017", "right": "pull_request:9542", "accept": false, - "reason": "Both add LoRA support, but for different model families and different conversion/loading logic." + "reason": "AuraFlow LoRA support and Flux LoRA text-encoder handling are separate model integrations." }, { "left": "pull_request:7550", "right": "pull_request:9040", "accept": false, - "reason": "Both are differential diffusion community pipelines, but they target different models and use cases." + "reason": "Different community pipelines for different models (SDXL differential diffusion vs HunyuanDiT differential diffusion)." }, { "left": "pull_request:10292", "right": "pull_request:9367", "accept": false, - "reason": "Both touch Flux auto-pipeline registration, but they add different pipeline variants and are not the same concrete change." + "reason": "Both touch AutoPipeline and Flux, but one adds Flux Control and the other adds Flux inpaint/img2img/controlnet routing; not the same concrete fix." } ] }, @@ -138563,7 +138634,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 106, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -138650,10 +138721,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -138668,162 +138739,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated PRs across core features, docs, tests, CI, and LoRA/quantization work. The soft pairs mostly show superficial similarity from shared filenames or adjacent subsystems, but they do not describe the same concrete bug or change.", - "confidence": 0.96, + "summary": "This cluster is a loose set of superficially similar but mostly unrelated PRs across docs, tests, schedulers, LoRA, pipelines, and CI. I would not treat any soft edge as a true duplicate/merge candidate; all should be rejected.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR: the cluster spans unrelated features and fixes across different subsystems, so there is no representative PR that subsumes the rest.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "No PR is a good global representative for this heterogeneous cluster; picking one would incorrectly imply the rest are duplicates of the same change.", "soft_edge_verdicts": [ { "left": "pull_request:5938", "right": "pull_request:7944", "accept": false, - "reason": "Both involve adapter/conditioning support, but one adds ControlNet LoRA plumbing and the other adds PAG support; different features and code paths." + "reason": "Different changes: ControlNet LoRA support vs PAG support. They only share broad loader/pipeline plumbing, not the same bug or change." }, { "left": "pull_request:10581", "right": "pull_request:9881", "accept": false, - "reason": "Different adapter features: SD3 inpaint IP-Adapter support vs IP-Adapter XFormers attention processor. Related area, not the same fix." + "reason": "Different code paths: SD3 inpaint IP-Adapter support vs IP-Adapter XFormers attention processor support. Same broad area, but not one mergeable fix." }, { "left": "pull_request:11716", "right": "pull_request:11802", "accept": false, - "reason": "Both touch LoRA loader utilities, but one fixes metadata handling for non-diffusers Flux LoRAs while the other removes deprecated helper methods." + "reason": "Flux LoRA metadata/output handling vs lora deprecation cleanup. The second is API cleanup, not the same underlying bug." }, { "left": "pull_request:11835", "right": "pull_request:12418", "accept": false, - "reason": "Both are test/CI-related, but one fixes float16 CUDA test behavior and the other xfails a specific legacy download test due to a transformers import issue." + "reason": "One fixes float16 CUDA test behavior; the other xfails a legacy-variant download test due to a transformers import issue. Unrelated test failures." }, { "left": "pull_request:4417", "right": "pull_request:5807", "accept": false, - "reason": "Both are SDXL-related docs/community changes, but one adds Zero123 support and the other fixes num_images_per_prompt in an LPW SDXL community pipeline." + "reason": "Zero123 community pipeline support vs LPW SDXL prompt-embedding duplication. Different features and unrelated code paths." }, { "left": "pull_request:2129", "right": "pull_request:2476", "accept": false, - "reason": "Both add pipeline arguments, but one is about allowing LoRA from pipeline and the other is Attend-and-Excite image-size support; not the same bug or change." + "reason": "LoRA passthrough in pipelines vs Attend-and-Excite image-size support. Both touch pipelines, but they fix different problems." }, { "left": "pull_request:8797", "right": "pull_request:9379", "accept": false, - "reason": "Both adjust test behavior, but one fixes sharding tests and the other fixes fast GPU tests for Flux; unrelated test failures." + "reason": "Sharded checkpoint test fixes vs Flux LoRA training/test GPU fixes. Shared theme is tests, but not the same defect." }, { "left": "pull_request:6828", "right": "pull_request:7186", "accept": false, - "reason": "Both involve IP-Adapter-style conditioning, but one updates UNetMotionModel IP-adapter handling and the other moves Face ID support to core; distinct implementations." + "reason": "UNetMotionModel IP-Adapter handling update vs moving IP-Adapter Face ID into core. Related ecosystem, but different concrete functionality." }, { "left": "pull_request:10136", "right": "pull_request:11428", "accept": false, - "reason": "Same broad HunyuanVideo area, but one introduces the base Hunyuan Video pipeline and the other adds the Framepack variant; separate features." + "reason": "Hunyuan Video base pipeline/model integration vs Hunyuan Video Framepack extension. Same family, different pipelines and changesets." }, { "left": "pull_request:3001", "right": "pull_request:3020", "accept": false, - "reason": "Both are scheduler additions, but one adds Karras support to multistep schedulers and the other adds stochastic DPM-SDE sampling; different algorithms." + "reason": "Both are scheduler-related, but one adds Karras handling to DPMSolverMultistep while the other adds a stochastic DPM++ SDE sampler. Different algorithms." }, { "left": "pull_request:2956", "right": "pull_request:3001", "accept": false, - "reason": "Both mention Karras/discrete schedulers, but one updates EulerDiscreteScheduler docs/code while the other targets DPMSolver/DEIS multistep schedulers." + "reason": "Both mention Karras sigmas, but they modify different schedulers and different implementations. Not the same concrete code-path fix." }, { "left": "pull_request:10199", "right": "pull_request:9070", "accept": false, - "reason": "One is a small RF inversion eta decay tweak; the other adds a Flux img2img pipeline. No shared underlying bug." + "reason": "RF inversion eta decay tweak in a community pipeline vs introducing the Flux img2img pipeline. Completely different functionality." }, { "left": "pull_request:10211", "right": "pull_request:9015", "accept": false, - "reason": "Both are example/docs fixes, but one corrects a CogView3Plus checkpoint name and the other fixes a PIA example; unrelated content." + "reason": "Checkpoint-id example fix for CogView3Plus vs PIA example/docs update. Both are example/doc corrections, not the same artifact or bug." }, { "left": "pull_request:10270", "right": "pull_request:11716", "accept": false, - "reason": "Both touch embeddings/LoRA-related areas, but the first fixes a CogVideoX positional embedding issue while the second fixes Flux LoRA metadata handling." + "reason": "CogVideoX embedding initialization fix vs Flux LoRA metadata return fix. No shared underlying issue." }, { "left": "pull_request:10286", "right": "pull_request:10481", "accept": false, - "reason": "Both are Sana-related defaults/docs, but one changes the pretrained model in docs and the other changes pipeline clean_caption behavior." + "reason": "Sana README checkpoint path update vs Sana pipeline default clean_caption change. One is docs, the other is runtime behavior." }, { "left": "pull_request:10316", "right": "pull_request:10825", "accept": false, - "reason": "Docs-only changes for different pipelines: one removes a generator arg from LTX Video docs, the other adds checkpoint-loading guidance for text2image." + "reason": "LTX video docs tweak vs text2image training docs update. Both are documentation-only and unrelated topics." }, { "left": "pull_request:10524", "right": "pull_request:10825", "accept": false, - "reason": "Both are docs edits, but one fixes HunyuanVideo checkpoint IDs and the other adds training/checkpoint instructions; different topics." + "reason": "Hunyuan Video docs checkpoint correction vs text2image training docs addition. Different docs for different pipelines." }, { "left": "pull_request:10578", "right": "pull_request:11660", "accept": false, - "reason": "Both are LoRA loader work, but one adds 4-bit Flux LoRA loading support and the other supports Omi HiDream LoRAs; distinct conversion paths." + "reason": "Flux 4bit LoRA loading support vs HiDream LoRA conversion support. Same loader subsystem, but different model families and code paths." }, { "left": "pull_request:10648", "right": "pull_request:11545", "accept": false, - "reason": "One is a scheduler docs update (sigmoid beta schedule), the other adds missing Flux img2img VAE slicing/tiling methods; unrelated." + "reason": "DDPM docs update for sigmoid schedule vs Flux img2img VAE slicing/tiling methods. One is docs-only, the other is pipeline behavior." }, { "left": "pull_request:1104", "right": "pull_request:796", "accept": false, - "reason": "Both touch CI/testing setup, but one adds slow MPS tests and the other adds Apple M1 tests; not the same change." + "reason": "Both touch CI/M1 testing, but one adds slow MPS tests while the other sets up Apple M1 tests. Separate infrastructure changes." }, { "left": "pull_request:11058", "right": "pull_request:11099", "accept": false, - "reason": "Both are CI/test workflow changes, but one makes CUDA-only tests device-agnostic and the other gates PR GPU tests on styling; different objectives." + "reason": "Device-agnostic test cleanup vs conditioning PR GPU tests on styling/quality checks. Different test infrastructure fixes." }, { "left": "pull_request:1129", "right": "pull_request:1765", "accept": false, - "reason": "Both modify DreamBooth Flax code, but one fixes sample batch size and the other is a broader Flax fixes PR with unrelated parser/training changes." + "reason": "Both involve DreamBooth Flax, but one fixes sample batch sizing while the other is a broader Flax fixes PR. Not the same bug." }, { "left": "pull_request:12018", "right": "pull_request:12464", "accept": false, - "reason": "Both are docs fixes, but one renames a quantization link label while the other fixes Marigold docs syntax; no shared bug." + "reason": "Quantization docs link rename vs Marigold docs syntax cleanup. Completely unrelated documentation changes." }, { "left": "pull_request:12170", "right": "pull_request:12275", "accept": false, - "reason": "Both are LoRA/quantization-related, but one supports more Qwen LoRAs and the other expands TorchAO config support; different features." + "reason": "Qwen LoRA conversion support vs TorchAO config-class support. Different feature areas and unrelated changes." }, { "left": "pull_request:1230", "right": "pull_request:1701", "accept": false, - "reason": "Both are older test/pipeline fixes, but one targets MPS generator handling in DDIM/DDPM tests and the other is RePaint API conforming and fast tests." + "reason": "MPS generator fast-test fix vs RePaint fast tests/API conformance. Both are tests/pipeline cleanup, but not the same issue." } ] }, @@ -138839,7 +138910,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 107, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -138927,10 +138998,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -138945,8 +139016,8 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a heterogeneous mix of unrelated docs fixes, training examples, pipeline additions, and small bugfixes; I don\u2019t see a true duplicate cluster here. All soft edges should be rejected.", - "confidence": 0.96, + "summary": "This is a heterogeneous set of unrelated diffusers PRs spanning docs, example scripts, pipeline additions, refactors, and scheduler/loaders fixes. I don\u2019t see a single underlying bug or change tying the items together, so none of the soft edges should be merged as duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -138956,151 +139027,151 @@ "left": "pull_request:12856", "right": "pull_request:13056", "accept": false, - "reason": "Same model family, but one adds Z-Image Omni Base implementation and the other adds Z-Image LoRA training docs/code; different changes and not one underlying fix." + "reason": "Same model family, but one adds Z-Image implementation docs and the other adds a Z-Image LoRA training example; different artifacts and changes." }, { "left": "pull_request:1408", "right": "pull_request:1416", "accept": false, - "reason": "Both touch conversion tooling, but they fix different bugs: sample-size handling vs ONNX safety-checker/feature-extractor wiring." + "reason": "PR 1408 fixes sample-size conversion; PR 1416 fixes ONNX conversion/inference behavior and safety checker handling. Different code paths." }, { "left": "pull_request:1436", "right": "pull_request:688", "accept": false, - "reason": "Both concern Flax loading, but one fixes `from_pt` messaging/behavior and the other fixes importing without PyTorch; separate issues." + "reason": "Both touch Flax loading, but 1436 is a small wording/import cleanup while 688 adds broader Flax-without-PyTorch support. Not the same concrete fix." }, { "left": "pull_request:1605", "right": "pull_request:1849", "accept": false, - "reason": "Both are Versatile Diffusion tweaks, but one adds `upcast_attention` and the other forwards `cross_attention_kwargs`; distinct code-path changes." + "reason": "One adds upcast_attention plumbing, the other passes cross_attention_kwargs through Versatile Diffusion attention. Different features in the same area." }, { "left": "pull_request:168", "right": "pull_request:1927", "accept": false, - "reason": "Stable Diffusion pipeline code vs documentation guide rewrite; not the same bug or change." + "reason": "PR 168 adds Stable Diffusion pipeline code; PR 1927 rewrites a Stable Diffusion docs guide. Not duplicate changes." }, { "left": "pull_request:1879", "right": "pull_request:3389", "accept": false, - "reason": "The latter is an integration-test run for the attention refactor, not the refactor itself; related context, but not a duplicate change." + "reason": "1879 is the attention-file refactor; 3389 is a draft integration-test run for that refactor. Related topic, but not the same change." }, { "left": "pull_request:2103", "right": "pull_request:3945", "accept": false, - "reason": "Both mention DreamBooth LoRA, but one adds a model tag and the other adds a rank parameter to training; separate enhancements." + "reason": "Both edit DreamBooth LoRA scripts, but one only adds a model tag while the other introduces a rank argument. Distinct changes." }, { "left": "pull_request:2400", "right": "pull_request:874", "accept": false, - "reason": "Both touch community docs, but they introduce different examples and purposes." + "reason": "Both affect community docs, but they document different community examples and different behaviors. Too broad to be the same issue." }, { "left": "pull_request:2446", "right": "pull_request:3398", "accept": false, - "reason": "Different docs pages and different fixes: an InstructPix2Pix snippet correction vs a Stable Diffusion typo fix." + "reason": "One fixes an instruct-pix2pix code snippet; the other corrects a typo in stable_diffusion.mdx. Different doc fixes." }, { "left": "pull_request:2693", "right": "pull_request:4856", "accept": false, - "reason": "Unrelated documentation fixes on different pages: an image-link correction vs an API cross-link fix." + "reason": "One fixes an inpaint doc image link; the other fixes an SDXL guide link. Unrelated documentation issues." }, { "left": "pull_request:2974", "right": "pull_request:4457", "accept": false, - "reason": "Both are TensorRT community pipeline items, but one is txt2img and the other is inpaint; different concrete pipelines." + "reason": "Both involve TensorRT/community pipeline docs, but one adds an SD/txt2img pipeline while the other fixes an inpaint pipeline. Different pipelines." }, { "left": "pull_request:3680", "right": "pull_request:5499", "accept": false, - "reason": "One fixes a loader-method link, the other broadly fixes many docs links; not the same artifact or bug." + "reason": "3680 is a specific loader-method link fix; 5499 is a broad links cleanup across many API docs. Not the same bug." }, { "left": "pull_request:3755", "right": "pull_request:5289", "accept": false, - "reason": "Both relax test tolerances, but for different models/tests and different failure modes." + "reason": "Both relax test tolerances, but they are in different test modules for different pipelines. Separate flaky-test fixes." }, { "left": "pull_request:392", "right": "pull_request:402", "accept": false, - "reason": "Different documentation restructuring efforts; no shared underlying issue." + "reason": "Both are docs overhauls, but one is a conceptual contribution page and the other is a toctree/intro restructure. Not the same underlying change." }, { "left": "pull_request:3945", "right": "pull_request:4934", "accept": false, - "reason": "Different training scripts/features: DreamBooth LoRA rank vs T2I-Adapter training docs/script." + "reason": "3945 adds a DreamBooth LoRA rank parameter; 4934 adds T2I-Adapter training docs/scripts. Different training workflows." }, { "left": "pull_request:3971", "right": "pull_request:5180", "accept": false, - "reason": "One fixes SDXL docs and the other adjusts doc-builder markdown formatting; unrelated." + "reason": "3971 is an SDXL docs/build workflow fix; 5180 is a Wuerstchen citation/doc-builder fix. Different docs problems." }, { "left": "pull_request:4211", "right": "pull_request:5275", "accept": false, - "reason": "Different SDXL documentation/example additions with no shared concrete bug." + "reason": "4211 fixes SDXL inpainting docstrings; 5275 adds an SDXL research example. Different purpose and code." }, { "left": "pull_request:4258", "right": "pull_request:4503", "accept": false, - "reason": "Both are LoRA/SDXL docs, but they address different text in different files." + "reason": "Both mention SDXL/LoRA docs, but one revises an SDXL README note and the other clarifies LoRA/Kohya support. Different documentation changes." }, { "left": "pull_request:4343", "right": "pull_request:6305", "accept": false, - "reason": "Separate features in separate areas: SDXL LoRA rank argument vs DDPMScheduler zero-SNR rescaling." + "reason": "4343 adds a DreamBooth LoRA rank argument; 6305 adds zero-terminal-SNR rescaling to DDPMScheduler. Different concrete code paths." }, { "left": "pull_request:4498", "right": "pull_request:4763", "accept": false, - "reason": "One adds SDXL reference/adain support; the other adds LoRA conversion tooling. Different changes." + "reason": "4498 adds an SDXL reference pipeline; 4763 adds a LoRA-to-safetensors conversion script. Completely different functionality." }, { "left": "pull_request:4582", "right": "pull_request:5367", "accept": false, - "reason": "Different doc fixes: ControlNet SDXL docstring vs SDXL tutorial pipeline construction." + "reason": "4582 fixes a ControlNet SDXL docstring example; 5367 fixes SDXL tutorial pipeline creation. Different documentation targets." }, { "left": "pull_request:4595", "right": "pull_request:6289", "accept": false, - "reason": "Completely unrelated work: an XL-to-stable-diffusion conversion WIP vs AnimateDiff SparseCntrl." + "reason": "4595 is an SDXL conversion-pipeline WIP; 6289 adds AnimateDiff SparseCntrl. Different features and codebases." }, { "left": "pull_request:4659", "right": "pull_request:6757", "accept": false, - "reason": "Different code paths entirely: SDXL long weighted prompt pipeline vs k-diffusion sigma device handling." + "reason": "4659 adds an SDXL long-weighted-prompt pipeline; 6757 moves sigmas to device in k-diffusion. Unrelated changes." }, { "left": "pull_request:4661", "right": "pull_request:6757", "accept": false, - "reason": "Different changes in unrelated subsystems; no plausible merge-as-one duplicate." + "reason": "4661 also adds the long-weighted-prompt community pipeline, while 6757 is a scheduler/device fix. No duplicate bug or change." }, { "left": "pull_request:4795", "right": "pull_request:5157", "accept": false, - "reason": "Both involve ControlNet/LoRA internals, but they fix different bugs in different functions and should not be merged as duplicates." + "reason": "4795 fixes torch.compile behavior in ControlNet; 5157 fixes LoRA naming/conversion logic in loaders. Different code paths and symptoms." } ] }, @@ -139116,7 +139187,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 108, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -139203,10 +139274,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -139221,156 +139292,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a duplicate set: it contains many unrelated PRs across docs, tests, loaders, pipelines, and model support. All soft edges look like broad superficial similarity rather than the same underlying bug/change.", - "confidence": 0.97, + "summary": "These PRs are mostly unrelated; the soft-similarity pairs share broad themes (docs, tests, IP-Adapter, LoRA, scheduling) but not the same concrete bug or change.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a valid canonical duplicate target here; the items split into many distinct fixes and feature additions with only shallow overlap.", + "canonical_pr_reason": "No true duplicate cluster emerges here. If forced to pick one representative PR, 10867 is the strongest anchor because it has an explicit issue target and a substantial multi-file code change; the rest are distinct fixes, docs, or tests.", "best_issue_reason": null, - "best_pr_reason": "No PR cleanly represents the whole cluster as a duplicate group; the safest triage is to leave the cluster unmerged because it is heterogeneous.", + "best_pr_reason": "10867 is the best single PR to represent the set only in the weak sense that it is the most substantive and clearly scoped change, but it is not a duplicate of the others.", "soft_edge_verdicts": [ { "left": "pull_request:5039", "right": "pull_request:5982", "accept": false, - "reason": "Both are SDXL-related docs, but one is about textual inversion and the other is SDXL Turbo docs; different changes." + "reason": "Both are docs, but one is SDXL textual inversion and the other is SDXL Turbo docs; different topics and deliverables." }, { "left": "pull_request:5698", "right": "pull_request:6528", "accept": false, - "reason": "Both touch IP-Adapter paths, but they fix different pipelines and different bugs." + "reason": "Different bugs in different pipelines: LCM img2img vs ControlNet img2img IP-Adapter handling." }, { "left": "pull_request:6163", "right": "pull_request:6291", "accept": false, - "reason": "Both are LoRA tests, but they address different test failures and assertions, not the same bug." + "reason": "Both are LoRA tests, but one adds xformers-attention coverage while the other changes the unfusion tolerance check; not the same underlying fix." }, { "left": "pull_request:6311", "right": "pull_request:7319", "accept": false, - "reason": "Both add docs pages, but for different topics: diffusion fast tutorial vs EDM scheduler docs." + "reason": "Both touch docs navigation, but one adds a fast diffusion tutorial and the other documents EDM schedulers." }, { "left": "pull_request:6875", "right": "pull_request:7301", "accept": false, - "reason": "Both adjust expected test values, but for unrelated components and failures." + "reason": "Different test fixes in different subsystems: IP-Adapter output slice vs VAE scaling constant." }, { "left": "pull_request:7029", "right": "pull_request:9015", "accept": false, - "reason": "Both are docs/example cleanups, but one fixes AnimateLCM docs and the other fixes a PIA example." + "reason": "Both are documentation-only, but one fixes AnimateLCM doc links/alt text while the other updates a PIA example." }, { "left": "pull_request:7050", "right": "pull_request:7181", "accept": false, - "reason": "Both are typo/docs cleanup PRs, but they touch different docs and code paths." + "reason": "Typos/examples cleanup versus training resume-log behavior; unrelated changes." }, { "left": "pull_request:11516", "right": "pull_request:11572", "accept": false, - "reason": "Both mention LTX-Video, but one is model/docs support and the other is non-diffusers LoRA conversion support." + "reason": "Same model family, but one is an LTX Video release/docs update and the other is LoRA format support; different concrete changes." }, { "left": "pull_request:4115", "right": "pull_request:5511", "accept": false, - "reason": "Both concern SDXL denoising schedules, but they fix different scheduler edge cases and code paths." + "reason": "Related area, but not the same bug: one adds discrete timestep support for SDXL ensemble-of-experts, the other fixes a second-order scheduler edge case." }, { "left": "pull_request:10589", "right": "pull_request:10867", "accept": false, - "reason": "One adds IP-Adapter support for SD3 img2img; the other adds multi IP-Adapter support for Flux. Different pipelines and mechanisms." + "reason": "Both involve IP-Adapter support, but they target different pipelines and capabilities (SD3 img2img vs Flux multi-adapter support)." }, { "left": "pull_request:12269", "right": "pull_request:13169", "accept": false, - "reason": "One is Flux modular pipeline support; the other is a Flux Klein PEFT LoRA format fix. Different problems." + "reason": "Different Flux features: modular Kontext support versus PEFT LoRA format conversion." }, { "left": "pull_request:7550", "right": "pull_request:9423", "accept": false, - "reason": "Both are community pipeline additions, but they implement different pipelines and different features." + "reason": "Both are differential diffusion community pipelines, but for different models/pipelines (SDXL vs Kolors)." }, { "left": "pull_request:6999", "right": "pull_request:8856", "accept": false, - "reason": "Both are img2img-related bug fixes, but they fix different input-handling issues in different pipelines." + "reason": "Unrelated img2img/input handling bugs in different pipelines: Stable Video Diffusion tensor normalization vs ControlNet/SDXL image batch duplication." }, { "left": "pull_request:10941", "right": "pull_request:11716", "accept": false, - "reason": "Both touch Flux loader logic, but one is copy/fix plumbing and the other is metadata-return behavior for non-diffusers LoRA loading." + "reason": "One is a copy/fix pass across Flux pipelines; the other is a LoRA loader metadata fix. Different code paths and outcomes." }, { "left": "pull_request:12736", "right": "pull_request:12973", "accept": false, - "reason": "One enables regional compilation for Z-Image; the other adds GLM-Image model/pipeline support. Different features." + "reason": "Different models and features: regional compilation for Z-Image versus adding GLM-Image model support." }, { "left": "pull_request:2994", "right": "pull_request:5117", "accept": false, - "reason": "One fixes type casting in ControlNet training; the other fixes resume-step logging across training scripts." + "reason": "ControlNet training type-casting and training resume logging are separate issues." }, { "left": "pull_request:11684", "right": "pull_request:12275", "accept": false, - "reason": "One adds GGUF single-file loading support; the other adds TorchAO quantization support. Different subsystems." + "reason": "Different subsystems: GGUF single-file checkpoint loading vs TorchAO quantization config support." }, { "left": "pull_request:11481", "right": "pull_request:12315", "accept": false, - "reason": "Both are test-only PRs, but one renames LoRA tests and the other changes scheduler setup across LoRA tests." + "reason": "A test rename is not the same change as collapsing LoRA tests to a single scheduler." }, { "left": "pull_request:11767", "right": "pull_request:12782", "accept": false, - "reason": "Both are offloading/test-related, but they target different models and distinct test assumptions." + "reason": "Implementation fix for group offloading with disk versus a test tweak for HunyuanVideo offloading." }, { "left": "pull_request:12490", "right": "pull_request:13330", "accept": false, - "reason": "One fixes an Audioldm2 test config; the other is a torchao test/workflow fix. No shared bug." + "reason": "Different test issues: CLAP config setup in AudioLDM2 vs TorchAO test adjustments." }, { "left": "pull_request:12681", "right": "pull_request:13323", "accept": false, - "reason": "One fixes CI pipeline CPU tests; the other adjusts device filtering in common pipeline tests. Different failures." + "reason": "Both are CI/test fixes, but they address different failures and code paths." }, { "left": "pull_request:130", "right": "pull_request:1579", "accept": false, - "reason": "Both mention examples, but one is a README update and the other adds Intel optimization examples." + "reason": "Both touch examples docs, but one adds an accelerate setup step and the other adds Intel optimization examples." }, { "left": "pull_request:2009", "right": "pull_request:473", "accept": false, - "reason": "One adds textual inversion loading; the other is an inpainting sample fix. Unrelated code changes." + "reason": "Textual inversion loader support and a Stable Diffusion inpaint sample fix are unrelated." }, { "left": "pull_request:2325", "right": "pull_request:365", "accept": false, - "reason": "One is an ONNX converter script; the other is a VAE type-hint change. No underlying duplicate." + "reason": "ONNX VAE converter support and VAE type-hint cleanup are different changes with no shared bug." } ] }, @@ -139386,7 +139457,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 109, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -139475,10 +139546,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -139493,168 +139564,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose bag of mostly unrelated maintenance/docs/test PRs. The soft edges are driven by superficial similarity (shared subsystems, filenames, or repeated phrases) rather than the same underlying bug or change, so none should be merged as duplicates.", - "confidence": 0.95, + "summary": "This is a loose similarity cluster, not a true duplicate set. The PRs span unrelated fixes across docs, CI, tests, pipelines, and model loaders; none of the soft edges look like the same concrete bug/change.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR stands out as a duplicate target. If one representative is needed, PR 4502 is the most complete and merged SDXL-related change in the set, but the cluster as a whole is not a true duplicate group.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 4502 is the strongest standalone PR here: merged, substantial, and with inbound references. It is still not a duplicate anchor for the other PRs in this cluster.", "soft_edge_verdicts": [ { "left": "pull_request:3766", "right": "pull_request:4301", "accept": false, - "reason": "Both touch Kandinsky-related docs/code, but one updates a conversion script and the other edits documentation; different artifacts and different changes." + "reason": "Kandinsky UNet conversion script vs Kandinsky docs update; same broad area, but not the same bug or change." }, { "left": "pull_request:4397", "right": "pull_request:4862", "accept": false, - "reason": "SDXL docstring wording fix vs ControlNet MultiControlNet bug fix are unrelated code paths and problems." + "reason": "SDXL docstring wording vs a ControlNet img2img bug fix; unrelated code paths." }, { "left": "pull_request:4499", "right": "pull_request:4666", "accept": false, - "reason": "The SDXL reference-attn example PRs are community example/docs work; the other is SDXL ControlNet LoRA support. Not the same bug or change." + "reason": "SDXL reference-attn community example/docs vs SDXL ControlNet LoRA support; different features and code paths." }, { "left": "pull_request:4500", "right": "pull_request:4666", "accept": false, - "reason": "Same as above: an example/docs PR versus an SDXL ControlNet LoRA implementation change." + "reason": "Same as above: SDXL reference example/docs vs ControlNet LoRA handling are not the same underlying change." }, { "left": "pull_request:4501", "right": "pull_request:4666", "accept": false, - "reason": "Example/docs addition for SDXL reference guidance, not the same concrete code-path as SDXL ControlNet LoRA support." + "reason": "Same as above: a reference-image SDXL example is unrelated to enabling LoRA in ControlNet." }, { "left": "pull_request:4502", "right": "pull_request:4666", "accept": false, - "reason": "Merged reference-attn/adain SDXL example PR, but it is not the same underlying change as enabling LoRA for SDXL ControlNets." + "reason": "Same as above: different SDXL features, different files, different fix scopes." }, { "left": "pull_request:4761", "right": "pull_request:6215", "accept": false, - "reason": "Both are test fixes, but for different components (PaintByExample vs PixArt) and unrelated failures." + "reason": "Paint-by-example image encoder fix vs PixArt test seeding/expected-values update; different bugs." }, { "left": "pull_request:5898", "right": "pull_request:6578", "accept": false, - "reason": "Both are docs edits, but one rewrites unconditional generation docs and the other fixes ControlNet heading formatting; no shared underlying bug." + "reason": "Two separate documentation edits with different content and scope; not the same change." }, { "left": "pull_request:6231", "right": "pull_request:7166", "accept": false, - "reason": "Both involve SDXL LoRA training, but one fixes fp16 gradient precision and the other fixes a text-encoder loading bug; separate issues." + "reason": "SDXL LoRA training fp16 gradient fix vs SDXL DreamBooth LoRA loading fix; distinct training/loading bugs." }, { "left": "pull_request:7932", "right": "pull_request:9536", "accept": false, - "reason": "SDXL LoRA test fixture updates and a ChatGLMTokenizer padding-side fix are unrelated." + "reason": "LoRA test expected-value update vs ChatGLM tokenizer padding-side API tweak; unrelated." }, { "left": "pull_request:10376", "right": "pull_request:11655", "accept": false, - "reason": "Both touch LoRA pipeline utilities, but one adds HunyuanVideo LoRA conversion and the other adds bitsandbytes 8-bit handling; different code paths." + "reason": "Both touch LoRA loading, but one adds HunyuanVideo original-format support and the other adds bnb 8-bit handling for Flux Control LoRA; different concrete changes." }, { "left": "pull_request:10359", "right": "pull_request:10911", "accept": false, - "reason": "Two unrelated test-cleanup PRs for different model families; not the same defect." + "reason": "Different test fixes in different model suites; no shared underlying bug." }, { "left": "pull_request:10589", "right": "pull_request:9881", "accept": false, - "reason": "One adds IP-Adapter support to SD3 Img2Img, the other adds an XFormers attention processor for IP-Adapter; related area but not the same concrete change." + "reason": "SD3 Img2Img IP-Adapter support vs IP-Adapter XFormers attention processor support; related subsystem, but not one fix." }, { "left": "pull_request:11786", "right": "pull_request:12618", "accept": false, - "reason": "CI marker refactor and uv install typo fix are distinct workflow issues." + "reason": "Big GPU marker CI changes vs uv install typo fix; completely separate CI issues." }, { "left": "pull_request:3268", "right": "pull_request:3337", "accept": false, - "reason": "Both are image postprocessing refactors, but they target different scopes (img2img refactor vs broad pipeline refactor) and are not the same PR-worthy change." + "reason": "Img2Img postprocessing refactor vs broader postprocessing refactor for other pipelines; same theme, but not the same change." }, { "left": "pull_request:4132", "right": "pull_request:5038", "accept": false, - "reason": "Making accelerate optional in checkpoint conversion is unrelated to making LoRA optional arguments explicit." + "reason": "Optional accelerate handling in checkpoint conversion vs explicit LoRA scale argument passing; unrelated code paths." }, { "left": "pull_request:5506", "right": "pull_request:7058", "accept": false, - "reason": "Both are LoRA-related, but one fixes xformers/PEFT scaling behavior and the other avoids in-place mutation of passed state dicts; different bugs." + "reason": "PEFT/xformers LoRA call-signature fix vs copying state dict before load_lora_weights; different defects." }, { "left": "pull_request:8679", "right": "pull_request:9423", "accept": false, - "reason": "Both add differential diffusion community pipelines, but for different model families (SD3 vs Kolors) and separate implementations." + "reason": "Two different differential-diffusion community pipelines for different models; not the same underlying bug or mergeable change." }, { "left": "pull_request:2393", "right": "pull_request:3716", "accept": false, - "reason": "Both add new pipelines and update docs, but Panorama and ParaDiGMS are distinct features with different code paths." + "reason": "Both add pipelines, but panorama and ParaDiGMS are different features with different implementations." }, { "left": "pull_request:8992", "right": "pull_request:9268", "accept": false, - "reason": "Differential diffusion for HunyuanDiT vs Flux soft inpainting via differential diffusion are different pipelines and changes." + "reason": "Different differential-diffusion community pipelines for HunyuanDiT vs Flux; not duplicates." }, { "left": "pull_request:10312", "right": "pull_request:9257", "accept": false, - "reason": "Attention reshape docs/fix and a Flux dreambooth prompt-tokenization bug are unrelated." + "reason": "Attention head-to-batch bug fix vs Flux DreamBooth prompt/tokenization change; unrelated." }, { "left": "pull_request:11113", "right": "pull_request:9473", "accept": false, - "reason": "AutoencoderKL train-script bugfix and Flux dreambooth LoRA LR handling fix are separate training-script issues." + "reason": "AutoencoderKL train-script fix vs Flux DreamBooth LoRA optimizer-lr fix; separate training-script bugs." }, { "left": "pull_request:11892", "right": "pull_request:12551", "accept": false, - "reason": "Removing k-diffusion dependency and skipping Sana CI tests are unrelated maintenance changes." + "reason": "Removing k-diffusion dependency vs skipping Sana CI tests; unrelated maintenance." }, { "left": "pull_request:12378", "right": "pull_request:13099", "accept": false, - "reason": "Code fix for custom-code loading vs docs guide for AutoModel custom models; same topic, but not the same underlying change." + "reason": "Custom-code loading logic bug fix vs documentation for custom models; docs only, not the same code change." }, { "left": "pull_request:12551", "right": "pull_request:13296", "accept": false, - "reason": "Skipping Sana tests in CI and updating release-test transformer versions are different CI changes." + "reason": "Skipping Sana CI tests vs release-test transformer version update; different CI maintenance tasks." }, { "left": "pull_request:1945", "right": "pull_request:823", "accept": false, - "reason": "UNet 2D block tests and negative-prompt type-mismatch fixes address different test and runtime behaviors." + "reason": "UNet block tests vs negative-prompt type mismatch fix; unrelated test coverage vs runtime bug." } ] }, @@ -139670,7 +139741,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 110, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -139759,10 +139830,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -139777,162 +139848,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated feature PRs, bug fixes, docs-only changes, and separate training/example updates across loaders, ControlNet, SDXL, LoRA, and community pipelines. I do not see one underlying duplicate bug/change that ties the cluster together.", - "confidence": 0.96, + "summary": "This is a heterogeneous cluster of unrelated pull requests across ControlNet/SDXL fixes, LoRA loader/training changes, conversion scripts, docs typos, and new pipeline features. No soft edge looks like the same underlying bug or change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR: the cluster contains many unrelated changes (docs, loaders, training, conversion, and new pipelines) rather than one underlying fix.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "No clear best PR for deduping this cluster; the items are not variants of one change, so any choice would be arbitrary.", "soft_edge_verdicts": [ { "left": "pull_request:3462", "right": "pull_request:4262", "accept": false, - "reason": "Both touch DreamBooth LoRA training, but they fix different problems: original-UNet robustness vs SDXL training/multiple fixes. Not the same change." + "reason": "Both are DreamBooth/LoRA training PRs, but one improves orig UNet robustness while the other fixes fp16 gradient/unscale behavior; different bugs." }, { "left": "pull_request:3766", "right": "pull_request:4461", "accept": false, - "reason": "Different scripts and goals: Kandinsky conversion script updates vs adding a pipeline_class_name option to the Stable Diffusion conversion script." + "reason": "Same conversion-script area, but Kandinsky UNet config updates vs. adding a pipeline_class_name argument for SD conversion are unrelated changes." }, { "left": "pull_request:4010", "right": "pull_request:4237", "accept": false, - "reason": "Both are SDXL-related, but one fixes sequential CPU offload and the other fixes prompt encoding dtype across pipelines. Different code paths." + "reason": "Both touch SDXL img2img/offload code, but 4010 fixes sequential CPU offload while 4237 fixes prompt embedding dtype in encode_prompt." }, { "left": "pull_request:4010", "right": "pull_request:4238", "accept": false, - "reason": "Offload handling in SDXL img2img is unrelated to the ControlNet SDXL inference fix." + "reason": "Same broad SDXL area, but 4010 is about sequential offload in img2img and 4238 is about ControlNet inference/offload sequencing; not the same bug." }, { "left": "pull_request:4034", "right": "pull_request:5738", "accept": false, - "reason": "Same broad loader area, but one adds LoRA unload and the other adds adapter deletion. Distinct APIs and behavior." + "reason": "Both modify loaders.py, but 4034 adds LoRA unloading utilities/docs and 5738 adds adapter deletion; distinct features." }, { "left": "pull_request:4078", "right": "pull_request:6147", "accept": false, - "reason": "Both involve single-file / conversion paths, but one enforces accelerate for from_single_file and the other adds SDXL refiner/inpainting support. Different bugs." + "reason": "Both involve from_single_file/convert_from_ckpt, but 4078 forces accelerate while 6147 adds SDXL inpainting/refiner single-file support; different code-path problems." }, { "left": "pull_request:4132", "right": "pull_request:5833", "accept": false, - "reason": "Optional accelerate in checkpoint conversion is unrelated to a ControlNet single-file import fix." + "reason": "Both concern single-file loading, but 4132 makes accelerate optional and 5833 fixes ControlNet import paths; unrelated issues." }, { "left": "pull_request:4132", "right": "pull_request:5834", "accept": false, - "reason": "Different issues: accelerate optionality vs relative-import correction for ControlNet single-file loading." + "reason": "Both concern single-file loading, but 4132 changes accelerate handling and 5834 fixes relative imports for ControlNet classes; not duplicates." }, { "left": "pull_request:4201", "right": "pull_request:4847", "accept": false, - "reason": "These are separate community pipelines (FABRIC vs Prompt Free Diffusion) with no shared underlying bug or change." + "reason": "FABRIC community example/docs vs. Prompt Free Diffusion community example; different pipelines and purposes." }, { "left": "pull_request:4856", "right": "pull_request:5015", "accept": false, - "reason": "Both are docs link fixes, but they patch different broken links in different files; not the same artifact." + "reason": "Both are docs-link fixes, but they correct different broken links on different pages." }, { "left": "pull_request:4862", "right": "pull_request:6147", "accept": false, - "reason": "MultiControlNet img2img shape handling is unrelated to SDXL inpainting single-file/refiner support." + "reason": "4862 fixes MultiControlNetModel img2img shape handling, while 6147 adds SDXL refiner support in single-file loading; different bugs." }, { "left": "pull_request:4886", "right": "pull_request:7156", "accept": false, - "reason": "StableDiffusionUpscale safety checker fix and stalebot label handling are unrelated code paths and domains." + "reason": "StableDiffusionUpscalePipeline safety-checker fix vs. stalebot label normalization; completely unrelated." }, { "left": "pull_request:5003", "right": "pull_request:7018", "accept": false, - "reason": "Both are LoRA-related, but one fixes fuse/unfuse mechanics while the other fixes fp16 gradient unscaling in training. Different bugs." + "reason": "Both are LoRA-related, but 5003 fixes fuse/unfuse internals while 7018 fixes gradient unscaling in advanced LoRA training; not the same change." }, { "left": "pull_request:5035", "right": "pull_request:5085", "accept": false, - "reason": "Both touch loaders.py, but one removes a forced transformers import and the other fixes LoRA offloading compatibility. Different concerns." + "reason": "Both touch loaders.py, but 5035 removes a hard transformers dependency and 5085 prevents offloading breakage for incompatible LoRA checkpoints; separate issues." }, { "left": "pull_request:5117", "right": "pull_request:6119", "accept": false, - "reason": "Both are training-script fixes, but one corrects resume logging and the other fixes fp16 gradient unscaling. Not the same underlying issue." + "reason": "Training-log resume fixes across scripts vs. fp16 gradient unscaling for LoRA training; different underlying problems." }, { "left": "pull_request:5386", "right": "pull_request:7325", "accept": false, - "reason": "Both are typo cleanup PRs, but they affect different documentation/code areas and are not a single duplicate change." + "reason": "Both are typo/doc cleanup, but they affect different files and languages; not the same artifact or bug." }, { "left": "pull_request:5979", "right": "pull_request:6419", "accept": false, - "reason": "DDGAN pipeline/scheduler work is unrelated to the prompt-diffusion in-context-learning community pipeline." + "reason": "DDGAN pipeline/scheduler addition vs. prompt-diffusion in-context learning example; unrelated feature work." }, { "left": "pull_request:6021", "right": "pull_request:6997", "accept": false, - "reason": "DreamBooth advanced variant support and LoRA support for UNetSpatioTemporalConditionModel are different features in different components." + "reason": "DreamBooth LoRA advanced variant support vs. LoRA support for UNetSpatioTemporalConditionModel; different models and changes." }, { "left": "pull_request:6144", "right": "pull_request:6259", "accept": false, - "reason": "Same training script, but one changes checkpoint LoRA state handling and the other fixes autocast for validation inference. Distinct bugs." + "reason": "Both in text-to-image LoRA training, but one fixes saving unwrapped LoRA weights and the other fixes autocast-related runtime errors during validation; distinct bugs." }, { "left": "pull_request:6721", "right": "pull_request:7790", "accept": false, - "reason": "AnimateDiff SDXL and SDXL ControlNet Reference are separate pipeline additions, not the same concrete change." + "reason": "AnimateDiff SDXL pipeline vs. SDXL masked im2im community pipeline; different features, not the same underlying change." }, { "left": "pull_request:8769", "right": "pull_request:8791", "accept": false, - "reason": "Both modify AutoencoderKL, but one adds attentionless VAE support and the other adds tiling options. Separate features." + "reason": "Both modify AutoencoderKL, but one adds attentionless VAE support and the other adds tiling support for SD3; separate changes." }, { "left": "pull_request:8825", "right": "pull_request:9893", "accept": false, - "reason": "A k-diffusion SDXL pipeline extension with ControlNet/IP-Adapter is unrelated to a community SDXL ControlNet reference pipeline." + "reason": "K-diffusion SDXL ControlNet/IP-Adapter support vs. a community reference pipeline for SDXL ControlNet; different pipeline code paths." + }, + { + "left": "pull_request:7371", + "right": "pull_request:8616", + "accept": false, + "reason": "Both are LoRA feature work, but 7371 adds DoRA support and 8616 adds LoRA fusion for SD3; not the same concrete fix." }, { "left": "pull_request:10723", "right": "pull_request:9175", "accept": false, - "reason": "Both involve ControlNet Union, but one adds per-control-type scaling in union models while the other adds Flux Union ControlNet support. Different model families and code paths." + "reason": "Both mention ControlNet Union, but 10723 adds per-control-type scaling while 9175 adds Flux ControlNet Union support; different model families and functionality." }, { "left": "pull_request:5448", "right": "pull_request:5979", "accept": false, - "reason": "Latent Consistency Models and DDGAN are separate model families with unrelated implementation changes." + "reason": "LCM pipeline/scheduler vs. DDGAN pipeline/scheduler; both are new pipelines but not the same underlying feature." }, { "left": "pull_request:9463", "right": "pull_request:9659", "accept": false, - "reason": "PAG support for SD Img2Img and SD3 LoRA support are unrelated features, even though both extend pipelines." + "reason": "PAG support for SD Img2Img vs. LoRA support for SD3 Img2Img; unrelated pipeline enhancements." } ] }, @@ -139948,7 +140025,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 111, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -140035,10 +140112,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -140053,8 +140130,8 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a set of superficially similar but mostly unrelated PRs across different features (Qwen-Image, Wan LoRA, IP-Adapter, docs, tests, and example fixes). None of the soft edges look like true duplicates of the same concrete change.", - "confidence": 0.98, + "summary": "This cluster is highly heterogeneous: it mixes unrelated PRs across model additions, docs, loaders, training examples, and test-only fixes. None of the soft pairs look like the same underlying bug or change, so I reject all soft edges.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -140064,151 +140141,151 @@ "left": "pull_request:12215", "right": "pull_request:12225", "accept": false, - "reason": "Different Qwen-Image features: one adds ControlNet support, the other adds an edit-inpainting pipeline." + "reason": "Different Qwen-Image features: ControlNet support vs edit inpainting pipeline." }, { "left": "pull_request:11532", "right": "pull_request:12040", "accept": false, - "reason": "Both touch LoRA conversion code, but they target different model families and different changes (HiDream vs Wan lightx2v/2nd denoiser)." + "reason": "Different model families and changes: HiDream LoRA support vs Wan lightx2v/2nd-denoiser LoRA support." }, { "left": "pull_request:10576", "right": "pull_request:11843", "accept": false, - "reason": "Unrelated test additions for different subsystems: 8-bit BnB+LoRA loading vs Wan VACE exclude_modules coverage." + "reason": "Unrelated test coverage for different subsystems and models." }, { "left": "pull_request:11238", "right": "pull_request:9968", "accept": false, - "reason": "One is an SD3 ControlNet validation fix; the other is a broader ControlNet refactor across many modules." + "reason": "One is a specific SD3 ControlNet validation fix; the other is a broad ControlNet refactor." }, { "left": "pull_request:1282", "right": "pull_request:2002", "accept": false, - "reason": "Docs link fix vs adding LoRA training support are different changes with no shared bug." + "reason": "Docs link correction versus a new LoRA training feature; not the same change." }, { "left": "pull_request:243", "right": "pull_request:888", "accept": false, - "reason": "Different pipelines entirely: RePaint inpainting vs CycleDiffusion." + "reason": "Different pipelines: RePaint inpainting vs CycleDiffusion." }, { "left": "pull_request:4370", "right": "pull_request:5833", "accept": false, - "reason": "Both are import-related, but one fixes dummy SDXL imports and the other fixes single-file loading imports for ControlNet." + "reason": "Unrelated import/name cleanup vs single-file loading import fix." }, { "left": "pull_request:4499", "right": "pull_request:4763", "accept": false, - "reason": "SDXL reference-attn docs/example work is unrelated to the SDXL LoRA conversion script." + "reason": "Different SDXL work: reference-attn example vs LoRA conversion utility script." }, { "left": "pull_request:4500", "right": "pull_request:4763", "accept": false, - "reason": "Same as above: reference-adain support is a different feature from LoRA conversion tooling." + "reason": "Different SDXL work: reference-attn example vs LoRA conversion utility script." }, { "left": "pull_request:4501", "right": "pull_request:4763", "accept": false, - "reason": "Another SDXL reference-attn variant, not the same change as the LoRA conversion script." + "reason": "Different SDXL work: reference-attn example vs LoRA conversion utility script." }, { "left": "pull_request:7924", "right": "pull_request:8825", "accept": false, - "reason": "Different problems: an IP-Adapter conditional kwargs bug in SDXL ControlNet Inpaint vs adding ControlNet/IP-Adapter support to K-Diffusion SDXL." + "reason": "Different code paths: ControlNet Inpaint SDXL IP-Adapter bug fix vs adding ControlNet/IP-Adapter support to SDXL K-Diffusion." }, { "left": "pull_request:8777", "right": "pull_request:8981", "accept": false, - "reason": "Dummy attention processor classes vs adding LoraLoaderMixin to init files are unrelated chore changes." + "reason": "Dummy attention processor classes vs adding LoraLoaderMixin to package inits." }, { "left": "pull_request:5900", "right": "pull_request:7820", "accept": false, - "reason": "General IP-Adapter support expansion vs a specific InstructPix2Pix IP-Adapter fix; not the same code-path bug." + "reason": "Broad IP-Adapter support across many pipelines vs a narrow InstructPix2Pix-specific fix." }, { "left": "pull_request:11726", "right": "pull_request:12074", "accept": false, - "reason": "Both are Wan LoRA related, but one fixes conversion for more Wan LoRAs while the other adds second-transformer loading and LightX2V support." + "reason": "Both touch Wan LoRA conversion, but they solve different concrete problems and are not one mergeable change." }, { "left": "pull_request:10359", "right": "pull_request:11159", "accept": false, - "reason": "Test cleanup in LoRA layer tests vs quantization dtype fixes; different scope and different failure modes." + "reason": "Different test-fix areas: LoRA layer test cleanup vs quantization dtype/test updates." }, { "left": "pull_request:12490", "right": "pull_request:12681", "accept": false, - "reason": "A CLAPConfig test adjustment is unrelated to the later CI-only pipeline CPU test fixes." + "reason": "AudioLDM2 ClapConfig test tweak is unrelated to the broader CI CPU test fixes." }, { "left": "pull_request:4944", "right": "pull_request:7016", "accept": false, - "reason": "Related to IP-Adapter internals, but one is the original support work and the other is a later refactor/extension across many pipelines." + "reason": "IP-Adapter feature work vs later refactor/embedding API changes; not the same underlying change." }, { "left": "pull_request:1012", "right": "pull_request:1147", "accept": false, - "reason": "DreamBooth text-encoder train mode fix vs Flax examples loading text encoder from subfolder are different issues." + "reason": "Different frameworks and bugs: PyTorch DreamBooth text-encoder train mode vs Flax examples loading text encoder from subfolder." }, { "left": "pull_request:10481", "right": "pull_request:11142", "accept": false, - "reason": "Sana clean_caption default change is unrelated to the Sana-Sprint converter/model checkpoint script update." + "reason": "Different Sana fixes: clean_caption default change vs conversion-script checkpoint/scheduler updates." }, { "left": "pull_request:10547", "right": "pull_request:9206", "accept": false, - "reason": "Offload test configuration for CogVideoX/CogView3 vs broken xformers test fixes; different test failures." + "reason": "Different test failures and models; only both are test fixes." }, { "left": "pull_request:1283", "right": "pull_request:888", "accept": false, - "reason": "Versatile Diffusion addition is a separate model release from CycleDiffusion." + "reason": "Different pipeline additions: Versatile Diffusion vs CycleDiffusion." }, { "left": "pull_request:2489", "right": "pull_request:3397", "accept": false, - "reason": "Core ML docs addition is unrelated to the text-to-image fine-tuning doc link fix." + "reason": "Unrelated docs additions; no shared bug or implementation change." }, { "left": "pull_request:3589", "right": "pull_request:5750", "accept": false, - "reason": "README reference-tricks documentation vs a failed/closed SD reference-only tech intro; not the same concrete change." + "reason": "Reference-tricks README notes are not the same as adding a reference-only SDXL introduction." }, { "left": "pull_request:4070", "right": "pull_request:4749", "accept": false, - "reason": "SDXL docs cleanup is unrelated to DreamBooth LoRA checkpointing changes." + "reason": "SDXL docs cleanup and SDXL DreamBooth LoRA checkpointing are separate issues." }, { "left": "pull_request:4706", "right": "pull_request:6225", "accept": false, - "reason": "Different LoRA bugs: missing fc alpha defaults in loader logic vs parsing correct alphas in PEFT training scripts." + "reason": "Related LoRA alpha theme, but one is loader internals and the other is example config parsing; not the same concrete fix." } ] }, @@ -140224,7 +140301,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 112, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -140311,10 +140388,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -140329,156 +140406,156 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly unrelated pull requests that only overlap at a broad subsystem level (docs, tests, loaders, schedulers, pipelines). None of the soft pairs look like the same underlying bug or change, so all soft edges should be rejected.", - "confidence": 0.88, + "summary": "This cluster is a noisy topical-similarity group, not a true duplicate set: it mixes unrelated docs cleanups, test adjustments, feature additions, and separate bug fixes across many subsystems. No single PR cleanly represents a shared underlying change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "There is no single canonical PR for this set: the items span many distinct features and fixes rather than one duplicate change. If forced to pick a representative PR, #9827 is the most substantive cross-cutting bug fix, but it is still not a duplicate of the others.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "#9827 is the strongest standalone candidate because it contains a concrete runtime bug fix plus docs/tests around device_map handling. It is the most representative of a real fix in this mixed set, though it does not unify the cluster.", + "best_pr_reason": "No single PR is a good global representative; the cluster is too heterogeneous for any one PR to serve as the canonical change.", "soft_edge_verdicts": [ { "left": "pull_request:5539", "right": "pull_request:7411", "accept": false, - "reason": "Both are typo/docs cleanup PRs, but they fix different files and different wording issues; not the same change." + "reason": "Both are minor cleanup/docs-oriented PRs, but they touch different files and unrelated changes (DreamBooth typos vs broad typo/docs sweep)." }, { "left": "pull_request:5608", "right": "pull_request:7103", "accept": false, - "reason": "Both touch LoRA-related code, but one fixes empty network alpha handling and the other fixes text-encoder patch removal; different bugs." + "reason": "Both fix bugs, but one is a PEFT/LoRA empty alpha bug while the other is a ControlNet image validation bug; different code paths and symptoms." }, { "left": "pull_request:9473", "right": "pull_request:9646", "accept": false, - "reason": "Same Flux Dreambooth area, but one changes T5 LR handling for Prodigy while the other makes LoRA target modules configurable; separate changes." + "reason": "Both affect Flux Dreambooth scripts, but one fixes optimizer LR handling for T5 and the other adds configurable LoRA target modules; not the same bug/change." }, { "left": "pull_request:12091", "right": "pull_request:12288", "accept": false, - "reason": "Both add new model families, but they are different products/pipelines entirely, not the same underlying feature." + "reason": "Both add new model/pipeline exports, but they introduce different products (Dream vs InfiniteTalk) and unrelated implementations." }, { "left": "pull_request:9040", "right": "pull_request:9268", "accept": false, - "reason": "Both are differential-diffusion community pipelines, but for different base models and different implementations." + "reason": "Both are differential-diffusion community pipelines, but they target different models and separate implementations; not mergeable as one fix." }, { "left": "pull_request:3135", "right": "pull_request:4223", "accept": false, - "reason": "Both involve ControlNet training scripts, but one enforces resolution divisibility and the other fixes SDXL training call/details; not the same bug." + "reason": "Both touch ControlNet training, but one enforces resolution divisibility and the other fixes SDXL training-script call/import behavior; distinct issues." }, { "left": "pull_request:12021", "right": "pull_request:12416", "accept": false, - "reason": "Both are model/pipeline enablement PRs, but for different architectures and code paths." + "reason": "Both are model/pipeline integration PRs, but they add different features for different architectures (LTX attention backends vs QwenImage Edit Plus modular support)." }, { "left": "pull_request:11159", "right": "pull_request:11715", "accept": false, - "reason": "One fixes GGUF/BnB dtype behavior in single-file loading; the other only adjusts test version gating. Not the same change." + "reason": "Both are quantization test-related, but one fixes dtype handling in single-file loading and the other adds bitsandbytes version guards for compile tests." }, { "left": "pull_request:1215", "right": "pull_request:742", "accept": false, - "reason": "Both modify DDIM noise generation, but one targets CPU generator handling and the other fixes dtype for eta>0; related area, different bugs." + "reason": "Same scheduler file, but the fixes address different DDIM noise-generation bugs (CPU-generator path vs dtype correctness)." }, { "left": "pull_request:12277", "right": "pull_request:12551", "accept": false, - "reason": "Both disable tests in CI, but for different pipelines and reasons; not a shared code-path fix." + "reason": "Both skip tests in CI, but they target different pipelines and different flaky behaviors; not the same underlying defect." }, { "left": "pull_request:12882", "right": "pull_request:13056", "accept": false, - "reason": "One adds a community inference pipeline, the other adds a DreamBooth training example; different artifacts and goals." + "reason": "Both are Z-Image related, but one adds a community differential img2img pipeline and the other adds DreamBooth training docs/script; separate deliverables." }, { "left": "pull_request:2976", "right": "pull_request:4215", "accept": false, - "reason": "Documentation-only changes in different sections and topics; not the same issue." + "reason": "Both are docs link/path fixes, but they affect unrelated documentation pages and do not share the same bug." }, { "left": "pull_request:4481", "right": "pull_request:6303", "accept": false, - "reason": "Different large WIP additions in the diffusion ecosystem, but unrelated models/scripts and no shared bug or feature." + "reason": "Both are large new pipeline/algorithm PRs, but EDM implementation and ADD training scripts are entirely different changes." }, { "left": "pull_request:5943", "right": "pull_request:7113", "accept": false, - "reason": "Transformer gradient checkpointing and Docker Slack notifications are unrelated changes." + "reason": "Completely unrelated: transformer gradient checkpointing versus Docker Slack notifications." }, { "left": "pull_request:6213", "right": "pull_request:6845", "accept": false, - "reason": "Both are cleanup refactors in video-related code, but they affect different pipelines and code paths." + "reason": "Both are refactors/cleanup in video-related code, but they concern different pipelines and different code paths." }, { "left": "pull_request:6217", "right": "pull_request:6330", "accept": false, - "reason": "One fixes torch.compile compatibility in SVD; the other is a docs rendering tweak. Different layers entirely." + "reason": "One fixes torch.compile behavior in Stable Video Diffusion, the other is a docs update for SVD rendering; not the same change." }, { "left": "pull_request:6448", "right": "pull_request:7932", "accept": false, - "reason": "Both touch LoRA tests, but they address different test-side side effects and expected values; not the same fix." + "reason": "Both are LoRA test fixes, but they address different regressions in different tests and outputs." }, { "left": "pull_request:7277", "right": "pull_request:7712", "accept": false, - "reason": "Typo/deprecation cleanup versus import fixes; no shared underlying bug." + "reason": "Both are general cleanup PRs, but they modify unrelated files and issues (typos/docs vs imports)." }, { "left": "pull_request:7353", "right": "pull_request:7590", "accept": false, - "reason": "A pipeline mixin update and a Python version-range packaging change are unrelated." + "reason": "These are unrelated: adding FromSingleFileMixin support to a ControlNet pipeline versus expanding Python version support in setup.py." }, { "left": "pull_request:8630", "right": "pull_request:9139", "accept": false, - "reason": "Different DreamBooth domains: SD3 text-encoder training versus Flux training-script bug fixes." + "reason": "Both touch DreamBooth docs/training, but one adds SD3 text-encoder training and the other is a Flux DreamBooth bugfix/update set." }, { "left": "pull_request:11081", "right": "pull_request:11605", "accept": false, - "reason": "Both involve group offloading internals, but one adds record_stream support and the other makes it torch.compile-safe; separate code-path issues." + "reason": "Both involve group offloading, but one adds record_stream support while the other makes it compatible with torch.compile; different underlying fixes." }, { "left": "pull_request:7825", "right": "pull_request:9379", "accept": false, - "reason": "One fixes device-map/offload tests; the other fixes fast GPU tests in Flux. Same broad testing theme, not the same bug." + "reason": "Both are test adjustments, but one fixes device-map/offload tests and the other fixes Flux GPU test inputs/device handling." }, { "left": "pull_request:9821", "right": "pull_request:9827", "accept": false, - "reason": "Closely related device_map work, but 9821 adds the utility/check behavior while 9827 applies it to LoRA loading; one is foundational, the other is a caller fix." + "reason": "Related theme, but not duplicates: 9821 changes the general device_map detection utility, while 9827 uses that behavior to fix LoRA loading with device maps." }, { "left": "pull_request:10494", "right": "pull_request:9536", "accept": false, - "reason": "Completely unrelated: LoRA test install setup versus ChatGLM tokenizer padding behavior." + "reason": "One changes tokenizers installation in CI, the other fixes a tokenizer API in Kolors; unrelated code and symptom." } ] }, @@ -140494,7 +140571,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 113, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -140584,10 +140661,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -140602,178 +140679,181 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous. The only clear overlaps are the WAN VACE bf16/fp16 mask-cast fix pair and the train_text_to_image_lora text-encoder finetuning pair; the rest are separate bugs, features, or docs/tests work.", - "confidence": 0.76, + "summary": "This cluster is mostly a title-similarity false positive set: the PRs span unrelated fixes across LoRA offloading, WAN dtype handling, DDIM/other schedulers, docs-only updates, and various test tweaks rather than one shared bug/change.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR represents the whole cluster; if one representative is needed, 12143 is the best fit because it contains a concrete code-path fix for the WAN VACE dtype bug and related test-side cleanup.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "12143 is the strongest overall representative: it is a compact, merged fix for a specific runtime bug in `pipeline_wan_vace.py` with adjacent test support, whereas most other items are unrelated.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:12142", - "right": "pull_request:12143", - "accept": true, - "reason": "Same WAN VACE mask dtype bug in `pipeline_wan_vace.py`; both change the same line, and 12143 just adds test cleanup around the same fix." + "right": "pull_request:12914", + "accept": false, + "reason": "Different code paths: WAN VACE mask dtype fix vs. a Cosmos doc typo." }, { "left": "pull_request:12143", "right": "pull_request:12914", "accept": false, - "reason": "Different changes: WAN VACE bf16/fp16 mask casting vs a Cosmos docs typo fix." + "reason": "Different code paths: WAN VACE mask dtype fix vs. a Cosmos doc typo." }, { "left": "pull_request:1215", "right": "pull_request:819", "accept": false, - "reason": "Both are DDIM-related, but they fix different problems: generator/device handling for eta>0 vs Windows int64 timestep conversion." + "reason": "Both touch DDIM, but one fixes eta/generator noise handling and the other fixes timestep dtype on Windows; separate bugs." }, { "left": "pull_request:1588", "right": "pull_request:3614", "accept": false, - "reason": "Unrelated test changes in different areas; one is FP16 common-test handling, the other is a Stable Diffusion test tweak." + "reason": "One is a test compatibility tweak for FP16, the other is a documentation/test cleanup; not the same bug." }, { "left": "pull_request:2584", "right": "pull_request:4862", "accept": false, - "reason": "Different pipelines and bugs: ControlNet img2img vs SDXL ControlNet img2img MultiControlNet handling." + "reason": "Different features/bugs: a community ControlNet pipeline addition vs. SDXL MultiControlNet img2img sizing fix." }, { "left": "pull_request:2861", "right": "pull_request:2919", "accept": false, - "reason": "Completely different scope: StableUnCLIP img2img test coverage vs a 3D UNet test speed-up." + "reason": "Stable UnCLIP image_embeds handling vs. shrinking a 3D UNet test for speed; unrelated." }, { "left": "pull_request:2994", "right": "pull_request:4749", "accept": false, - "reason": "Different code paths and purposes: ControlNet training dtype casting vs SDXL DreamBooth LoRA checkpoint loading." + "reason": "ControlNet training type-casting fix vs. SDXL LoRA checkpoint loading fix; different training scripts and failure modes." }, { "left": "pull_request:4215", "right": "pull_request:6578", "accept": false, - "reason": "Both are docs touch-ups, but they target different ControlNet documentation fixes." + "reason": "Docs path update for ControlNet vs. heading/formatting fix in ControlNet docs; not the same change." }, { "left": "pull_request:7797", "right": "pull_request:8422", "accept": false, - "reason": "Different test work: DDPM fast-test model shrink vs a subfolder-loading test suite." + "reason": "A DDPM fast-test size reduction vs. a generic subfolder-loading test suite; unrelated test maintenance." }, { "left": "pull_request:7803", "right": "pull_request:8422", "accept": false, - "reason": "Different test work: DDIM fast-test model shrink vs subfolder-loading coverage." + "reason": "A DDIM fast-test size reduction vs. a generic subfolder-loading test suite; unrelated test maintenance." }, { "left": "pull_request:7804", "right": "pull_request:8422", "accept": false, - "reason": "Different test work: Amused fast-test shrink vs subfolder-loading coverage." + "reason": "Amused fast-test size reduction vs. a generic subfolder-loading test suite; unrelated test maintenance." }, { "left": "pull_request:9099", "right": "pull_request:9175", "accept": false, - "reason": "Different features: SD3 ControlNet inpainting vs Flux Union ControlNet support." + "reason": "Both concern ControlNet/Flux support, but one adds SD3 controlnet inpainting while the other adds Union ControlNet for Flux; distinct features." }, { "left": "pull_request:11896", "right": "pull_request:12790", "accept": false, - "reason": "Both are LoRA-related, but one fixes hook disabling while loading LoRAs and the other normalizes ZImage LoRA keys." + "reason": "Different LoRA issues: hook removal/offloading when loading LoRAs vs. LoRA key normalization for ZImage models." }, { "left": "pull_request:6946", "right": "pull_request:7186", "accept": false, - "reason": "Same broad IP-Adapter area, but different changes: low_cpu_mem_usage loading vs moving Face ID to core." + "reason": "Related IP-Adapter area, but one adds low_cpu_mem_usage loading and the other moves Face ID to core with model/embedding changes; not one concrete fix." }, { "left": "pull_request:10730", "right": "pull_request:9601", "accept": false, - "reason": "Unrelated fixes: hook unwrapping during offload removal vs VAE dtype handling in SD3 ControlNet training." + "reason": "Unwrap hooks removal in pipeline utils vs. VAE dtype control in SD3 training; unrelated problems." }, { "left": "pull_request:13100", "right": "pull_request:13410", "accept": false, - "reason": "Both are modular-pipeline docs, but they are different documentation changes and not the same artifact." + "reason": "Modular pipeline docs update vs. agent docs/skill restructuring; different documentation targets and intent." }, { "left": "pull_request:1327", "right": "pull_request:2928", "accept": false, - "reason": "Unrelated: custom local pipeline loading vs controlnet training local-dataset loading." + "reason": "Custom pipeline local loading convenience vs. ControlNet local dataset loading; different features." + }, + { + "left": "pull_request:1327", + "right": "pull_request:3446", + "accept": false, + "reason": "Custom pipeline local loading vs. a pipeline loading warning-message wording fix; not the same bug." }, { "left": "pull_request:2002", "right": "pull_request:816", "accept": false, - "reason": "Different topics: LoRA text-encoder support in an example vs deepspeed FP16 training compatibility." + "reason": "LoRA training example support vs. deepspeed FP16 loss casting fix; unrelated training concerns." }, { "left": "pull_request:2479", "right": "pull_request:3691", "accept": false, - "reason": "Both involve text-encoder LoRA, but one adds training-script support and the other switches the attention processor class; different changes." + "reason": "LoRA text-encoder training support vs. choosing LoRAAttnProcessor2_0 for PT 2.0; distinct LoRA enhancements." }, { "left": "pull_request:2479", "right": "pull_request:3912", - "accept": true, - "reason": "Same underlying `train_text_to_image_lora.py` text-encoder finetuning feature; 3912 looks like the docs/test follow-up to the code introduced in 2479." + "accept": false, + "reason": "Both touch the LoRA text-to-image script, but one adds CLI support while the other adds text-encoder finetuning/docs; could not be merged as one concrete bug fix." }, { "left": "pull_request:2557", "right": "pull_request:3016", "accept": false, - "reason": "Different training features: DreamBooth logging vs Min-SNR/gradient-accumulation docs for ControlNet flax training." + "reason": "DreamBooth logging vs. ControlNet Min-SNR/gradient-accumulation docs and flax training updates; unrelated." }, { "left": "pull_request:3128", "right": "pull_request:4886", "accept": false, - "reason": "Different bugs in different pipelines: img2img timestep selection vs StableDiffusionUpscale safety-checker input handling." + "reason": "Img2img timestep selection bug vs. upscale safety-checker input processing; different pipelines and defects." }, { "left": "pull_request:4070", "right": "pull_request:4428", "accept": false, - "reason": "Both are SDXL docs, but 4070 is a minor wording/link fix and 4428 is a broader documentation reorganization." + "reason": "Minor SDXL doc link/wording fix vs. broad SDXL docs reorganization; not the same change." }, { "left": "pull_request:4862", "right": "pull_request:6085", "accept": false, - "reason": "Different code paths: ControlNet SDXL img2img height/width handling vs scheduler `add_noise` indexing fixes." + "reason": "ControlNet img2img MultiControlNet shape handling vs. scheduler add_noise indexing bug; unrelated." }, { "left": "pull_request:5157", "right": "pull_request:7018", "accept": false, - "reason": "Both touch LoRA tooling, but one is ControlNet LoRA conversion logic and the other is DreamBooth LoRA gradient scaling." + "reason": "LoRA conversion name-mapping fix for ControlNet/LyCORIS vs. advanced LoRA gradient unscaling fix; different layers of the stack." }, { "left": "pull_request:6147", "right": "pull_request:7166", "accept": false, - "reason": "Different loading bugs: single-file SDXL refiner loading vs SDXL DreamBooth LoRA text-encoder loading." + "reason": "Single-file SDXL inpainting/refiner loading vs. SDXL LoRA checkpoint loading; related ecosystem, but different concrete fixes." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded and conservative. It correctly identifies the two genuinely mergeable overlaps in the packet: the WAN VACE dtype-mask fix pair (12142/12143) and the train_text_to_image_lora text-encoder finetuning pair (2479/3912). The other pairwise rejections are supported by the diffs and titles, and the representative-PR choice is reasonable for a heterogeneous cluster." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -140782,7 +140862,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 114, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -140869,10 +140949,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -140887,162 +140967,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the PRs span unrelated LoRA, pipeline, quantization, docs, workflow, and feature work. The soft edges mostly reflect broad topic or filename overlap, not the same underlying bug/change, so they should all be rejected.", - "confidence": 0.94, + "summary": "The cluster is noisy and largely heterogeneous. The soft pairs mostly share broad subsystem terms (LoRA, SDXL, single-file, pipelines, tests) but do not look like the same underlying bug/change, so none should be merged as duplicates.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 12996 is the strongest standalone representative here: merged, explicit issue target, substantial implementation plus docs, and a coherent end-to-end feature change.", + "canonical_pr_reason": "No clear canonical PR: the items span unrelated fixes and features across different pipelines, loaders, tests, and docs.", "best_issue_reason": null, - "best_pr_reason": "PR 12996 is the best overall PR candidate if one must be chosen, because it is the most complete and well-scoped change in the set; the others are smaller or more narrowly test/doc/workflow fixes.", + "best_pr_reason": "PR 10397 is the strongest standalone representative in the set because it is a merged, concrete Flux LoRA fix with user-facing impact, but it still is not a true duplicate anchor for the whole cluster.", "soft_edge_verdicts": [ { "left": "pull_request:7295", "right": "pull_request:9244", "accept": false, - "reason": "Different problems: single-file UNet loader cleanup vs adding Flux single-file support. Same loader area, but not the same bug/change." + "reason": "Both touch single-file loading, but one is a UNet loader cleanup and the other adds Flux pipeline single-file support; different models and code paths." }, { "left": "pull_request:8631", "right": "pull_request:8902", "accept": false, - "reason": "One returns the model from a checkpoint helper; the other fixes sharded checkpoint filename handling. Distinct single-file loader issues." + "reason": "One fixes a return value for T5 checkpoint conversion; the other changes sharded checkpoint sibling detection. Same utility area, different bugs." }, { "left": "pull_request:8755", "right": "pull_request:9807", "accept": false, - "reason": "Unrelated domains: SD3 DreamBooth text-encoder training fix vs AutoencoderKL LoRA adapter support." + "reason": "SD3 Dreambooth LoRA text-encoder training and AutoencoderKL LoRA adapter support are unrelated changes in different subsystems." }, { "left": "pull_request:10396", "right": "pull_request:10397", "accept": false, - "reason": "Both LoRA-related, but 10396 fixes state-dict expansion indexing while 10397 fixes LoRA unloading/reset behavior for Flux. Different code paths." + "reason": "Both are Flux LoRA-related, but 10396 fixes state-dict expansion indexing while 10397 changes unload/reset behavior and docs; not the same concrete bug." }, { "left": "pull_request:1250", "right": "pull_request:243", "accept": false, - "reason": "Both are inpainting/community pipelines, but one adds CLIPSeg+StableDiffusionInpainting and the other adds RePaint. Not the same change." + "reason": "Community CLIPSeg+inpainting example and RePaint inpainting pipeline are different features with no shared code-path bug." }, { "left": "pull_request:2034", "right": "pull_request:3850", "accept": false, - "reason": "Separate test fixes for different pipelines (DiT vs audio diffusion), with different setup and symptoms." + "reason": "These are separate test fixes for DiT and audio diffusion; same test-fix pattern but different components." }, { "left": "pull_request:2034", "right": "pull_request:689", "accept": false, - "reason": "DiT test-shape update vs a generic attention performance/speed fix. No shared bug." + "reason": "DiT test adjustments are unrelated to the attention performance refactor in the slow-tests PR." }, { "left": "pull_request:4795", "right": "pull_request:6864", "accept": false, - "reason": "ControlNet torch.compile fix vs a docs note about torch.compile for PEFT. Documentation is not the same as the model bug fix." + "reason": "One fixes ControlNet torch-compile behavior; the other is documentation about torch.compile usage with PEFT." }, { "left": "pull_request:8688", "right": "pull_request:8755", "accept": false, - "reason": "DoRA conversion utility fix is unrelated to SD3 DreamBooth text-encoder training logic." + "reason": "DoRA conversion logic and SD3 Dreambooth text-encoder training are different LoRA-related areas and different bugs." }, { "left": "pull_request:12983", "right": "pull_request:13237", "accept": false, - "reason": "Both touch single_file_utils, but one adds LTX2 single-file mappings and the other adds Flux.2 Klein presets. Different model families and behavior." + "reason": "LTX2 single-file support and FLUX.2 Klein single-file mapping are different model families and separate additions." }, { "left": "pull_request:7476", "right": "pull_request:8856", "accept": false, - "reason": "Both concern img2img/control image handling, but one is SDXL multi-controlnet compile behavior and the other is a batch duplication error fix across several pipelines." + "reason": "MultiControlNet SDXL img2img compile handling and generic img2img image-batch duplication are different bug fixes." }, { "left": "pull_request:7476", "right": "pull_request:8859", "accept": false, - "reason": "Different bugs: ControlNet multi-controlnet control-keep logic vs AuraFlow default resolution/compile compatibility." + "reason": "These affect different pipelines and failure modes: ControlNet shape handling versus AuraFlow compile compatibility." }, { "left": "pull_request:11485", "right": "pull_request:12225", "accept": false, - "reason": "Separate feature additions: Flex.2 pipeline vs Qwen-Image-Edit inpainting pipeline." + "reason": "Flex.2 preview pipeline addition and Qwen-Image edit inpainting pipeline are separate feature PRs for different models." }, { "left": "pull_request:10911", "right": "pull_request:11755", "accept": false, - "reason": "Lumina2 LoRA fuse test fix is unrelated to LTX latent upsample cpu-offload behavior." + "reason": "Lumina2 LoRA fuse test coverage and LTX latent upsample CPU offload behavior are unrelated test changes." }, { "left": "pull_request:12416", "right": "pull_request:13089", "accept": false, - "reason": "QwenImage Edit Plus modular support is a feature add; NAG guider update is a different model-control change, not the same underlying change." + "reason": "Both mention QwenImage modular code, but one adds Edit Plus support while the other adds a new guider and broader pipeline updates; not the same change." }, { "left": "pull_request:4862", "right": "pull_request:5922", "accept": false, - "reason": "Different pipelines and bugs: SDXL ControlNet img2img height/width handling vs MultiAdapter preprocessing in a community example." + "reason": "SDXL MultiControlNet img2img shape handling and MultiAdapter inpainting preprocessing are different adapter paths and different bugs." }, { "left": "pull_request:9268", "right": "pull_request:9423", "accept": false, - "reason": "Both are differential-diffusion community examples, but for different base models (Flux vs Kolors) and separate code paths." + "reason": "Both are differential-diffusion examples, but they target different pipelines (Flux vs Kolors) and are separate community implementations." }, { "left": "pull_request:4408", "right": "pull_request:4774", "accept": false, - "reason": "SDXL single-file inpainting support vs adding negative size/crop conditions in SDXL. Related subsystem, but not the same fix." + "reason": "SDXL from_single_file inpainting support and negative size/crop conditioning are different user-facing features." }, { "left": "pull_request:8389", "right": "pull_request:9696", "accept": false, - "reason": "Benchmark workflow infrastructure changes vs PR-test workflow input validation changes. Distinct CI workflows." + "reason": "Benchmark workflow maintenance and PR-test workflow input validation are unrelated CI changes." }, { "left": "pull_request:4481", "right": "pull_request:4847", "accept": false, - "reason": "EDM pipeline/scheduler implementation is unrelated to a community prompt-free diffusion example." + "reason": "EDM pipeline/scheduler implementation and Prompt Free Diffusion community example are distinct additions with no shared bug." }, { "left": "pull_request:4498", "right": "pull_request:6197", "accept": false, - "reason": "Stable Diffusion XL reference example/docs vs DeepCache support. Same broad ecosystem, different functionality." + "reason": "An SDXL reference example and DeepCache support for Stable Diffusion are not the same underlying change." }, { "left": "pull_request:5915", "right": "pull_request:6884", "accept": false, - "reason": "IP-Adapter Plus support is a model capability addition; the other only tightens scale validation in set_ip_adapter_scale." + "reason": "IP-Adapter Plus support and scale-length validation in IP-Adapter are related but separate issues, not one duplicate fix." }, { "left": "pull_request:11159", "right": "pull_request:11522", "accept": false, - "reason": "Quantization dtype handling for single-file loading vs an AudioLDM2 transformers compatibility fix. Unrelated code paths." + "reason": "Mixed-int8/quantization dtype fixes and AudioLDM2 transformers-main compatibility are unrelated changes." }, { "left": "pull_request:11588", "right": "pull_request:12496", "accept": false, - "reason": "Wan I2V LoRA support and Wan I2V condition-shape mismatch are different issues in the same pipeline, not one duplicate bug." + "reason": "WAN I2V LoRA support and WAN I2V condition-shape mismatch are different code paths and different bugs." }, { "left": "pull_request:12919", "right": "pull_request:12996", "accept": false, - "reason": "Both touch attention dispatch, but 12919 adds NPU attention support while 12996 adds Ulysses Anything Attention for distributed inference. Different backends and purposes." + "reason": "NPU attention enablement and Ulysses Anything Attention are both attention-related, but they solve different parallel/back-end problems and should not be merged as duplicates." } ] }, @@ -141058,7 +141138,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 115, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -141146,10 +141226,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -141164,168 +141244,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a mixed bag of unrelated PRs spanning docs, tests, LoRA plumbing, single-file loading, Flux/Wan/SD3/Sana features, and assorted bug fixes. The soft pairs share broad themes or files but not the same concrete bug/change, so none should be merged as duplicates.", - "confidence": 0.95, + "summary": "This cluster is a grab-bag of mostly unrelated PRs across docs, tests, LoRA loading, single-file support, and several different model families. The apparent similarities are usually just broad subsystem overlap, not the same underlying bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No single PR is a good canonical representative for the cluster; 11804 is the broadest substantive code change, but the overall set is too heterogeneous to treat as one duplicate family.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:10581", "right": "pull_request:10728", "accept": false, - "reason": "Different features and code paths: SD3 inpaint IP-Adapter support vs Flux quantized IP-Adapter loader behavior." + "reason": "Both touch Flux-adjacent code, but one adds IP-Adapter support for SD3 inpaint and the other is a Flux quantization/IP-Adapter tweak; different pipelines and different problems." }, { "left": "pull_request:12091", "right": "pull_request:12526", "accept": false, - "reason": "Unrelated model additions: Dream 7B pipeline vs Wan2.2 Animate pipeline." + "reason": "Both are large model-addition PRs, but Dream 7B and Wan2.2 Animate are different models/pipelines with unrelated code paths." }, { "left": "pull_request:4632", "right": "pull_request:6464", "accept": false, - "reason": "Both touch training scripts, but they fix different checkpoint/save issues in different Dreambooth LoRA paths." + "reason": "Both are training-script bug fixes, but they target different Dreambooth LoRA scripts and different checkpoint/save issues." }, { "left": "pull_request:5057", "right": "pull_request:5127", "accept": false, - "reason": "Same SDXL family, but clip-skip support and multi-adapter support are distinct changes with different bugs." + "reason": "SDXL clip-skip support and MultiAdapter support are separate features touching some of the same pipeline files but fixing different behavior." }, { "left": "pull_request:11546", "right": "pull_request:11804", "accept": false, - "reason": "Both involve LoRAs, but one adds Hunyuan FramePack loading and the other adds group-offloading support for dynamic LoRA loading/unloading." + "reason": "Both involve LoRA loading infrastructure, but one is a Hunyuan/FramePack PEFT mapping addition and the other is group-offloading dynamic load/unload support." }, { "left": "pull_request:5185", "right": "pull_request:5417", "accept": false, - "reason": "Related LoRA scaling area, but 5185 fixes PEFT CI behavior while 5417 fixes unscale/scale handling across pipelines." + "reason": "Both relate to LoRA scaling, but 5185 is a PEFT CI fix while 5417 changes scale/unscale behavior across many pipelines; not the same bug." }, { "left": "pull_request:7550", "right": "pull_request:9268", "accept": false, - "reason": "Different community pipelines and models; both mention differential diffusion but address separate implementations." + "reason": "Both are differential-diffusion community pipelines, but SDXL img2img and Flux soft inpainting are different implementations and change sets." }, { "left": "pull_request:8679", "right": "pull_request:9597", "accept": false, - "reason": "Both are community differential-diffusion examples, but one is SD3 img2img and the other is Auraflow; not the same change." + "reason": "Both are community diffusion examples, but one is SD3 differential img2img and the other is Auraflow differential diffusion; unrelated pipelines." }, { "left": "pull_request:11823", "right": "pull_request:12040", "accept": false, - "reason": "Different LoRA conversion targets: Flux Kontext vs Wan/lightx2v support." + "reason": "Both add LoRA conversion support, but for different targets: Fal Kontext vs Wan/lightx2v. Similar subsystem, different conversion logic and model families." }, { "left": "pull_request:9367", "right": "pull_request:9507", "accept": false, - "reason": "Auto-pipeline registration for Flux variants vs a specific Flux ControlNet bugfix are not the same concrete change." + "reason": "Both are Flux-related, but one extends auto-pipeline registrations and the other fixes Flux ControlNet batching/control modes; distinct code paths." }, { "left": "pull_request:9175", "right": "pull_request:9445", "accept": false, - "reason": "ControlNet support for Flux vs a Flux-with-CFG community example; different artifacts and purposes." + "reason": "Both are Flux features, but Union ControlNet support and Flux with CFG are separate pipeline capabilities, not one concrete fix." }, { "left": "pull_request:11025", "right": "pull_request:11823", "accept": false, - "reason": "Both extend LoRA conversion, but for different model families and conversion logic." + "reason": "Both touch LoRA conversion utilities, but they support different external formats/models (Wan vs Fal Kontext) and do not fix the same conversion bug." }, { "left": "pull_request:12737", "right": "pull_request:12988", "accept": false, - "reason": "Both involve context parallelism, but one fixes FLUX.2 tensor layout and the other removes a shard divisibility assert; different bugs." + "reason": "Both are context-parallel related, but one fixes Flux.2 context-parallel settings and the other changes generic sharding assertions; not the same issue." }, { "left": "pull_request:10005", "right": "pull_request:10332", "accept": false, - "reason": "Docs for missing pipelines vs a model-loading shape error fix; unrelated." + "reason": "One adds missing Flux pipeline docs; the other fixes a shape attribute in loading errors. Same repo area only by coincidence." }, { "left": "pull_request:10005", "right": "pull_request:8675", "accept": false, - "reason": "Both are docs changes, but they cover different topics and files." + "reason": "Flux pipeline docs versus a general fast-diffusion docs tip update are unrelated documentation changes." }, { "left": "pull_request:10099", "right": "pull_request:10977", "accept": false, - "reason": "A typo fix in xfail decorators vs save/load optional-component test updates; unrelated." + "reason": "One is a typo fix in xfail conditions; the other adds save/load optional-component tests across multiple pipelines." }, { "left": "pull_request:10102", "right": "pull_request:9843", "accept": false, - "reason": "One tests skip_guidance_layers in SD3, the other fixes an SD3 LoRA test; different issues." + "reason": "Both are test additions, but one covers SD3 skip_guidance_layers behavior and the other updates SD3 LoRA test data and expectations." }, { "left": "pull_request:10109", "right": "pull_request:10648", "accept": false, - "reason": "Torch XLA support in Stable Audio vs scheduler docs for sigmoid; not the same change." + "reason": "Stable Audio torch_xla support and DDPM sigmoid-doc updates are different features with no shared bug." }, { "left": "pull_request:10115", "right": "pull_request:10783", "accept": false, - "reason": "Flux positional embedding bugfix vs Dreambooth fp16 validation bug; different code paths." + "reason": "FluxPosEmbed bugfix versus Dreambooth fp16 unscaling bug are unrelated code paths." }, { "left": "pull_request:10115", "right": "pull_request:11439", "accept": false, - "reason": "Flux embedding fix vs Hi-Dream validation cleanup; unrelated." + "reason": "FluxPosEmbed fix and Hi-Dream validation cleanup address different problems in different model families." }, { "left": "pull_request:10179", "right": "pull_request:10278", "accept": false, - "reason": "Community example fixes in different areas: rf inversion doc/example vs TorchAO docs example." + "reason": "A community example doc fix for rf-inversion is unrelated to a TorchAo Flux quantization example fix." }, { "left": "pull_request:10179", "right": "pull_request:10517", "accept": false, - "reason": "Example documentation fix vs Sana 4K pipeline bug; unrelated." + "reason": "A doc/example correction and a Sana 4K pipeline bug report are not the same underlying issue." }, { "left": "pull_request:10183", "right": "pull_request:10364", "accept": false, - "reason": "Single-file support for AutoencoderDC vs torch_xla support in Sana; different features." + "reason": "AutoencoderDC single-file support and Sana torch_xla support are unrelated features." }, { "left": "pull_request:10183", "right": "pull_request:11271", "accept": false, - "reason": "Both are single-file related, but they add support for different models and model-selection logic." + "reason": "Both touch single-file utilities, but one adds AutoencoderDC support and the other fixes LTX 0.9.5 model inference/config mapping." }, { "left": "pull_request:10188", "right": "pull_request:11595", "accept": false, - "reason": "Both touch Flux LoRA loading, but they fix different bugs in different parts of the loader." + "reason": "Both concern Flux LoRA loading, but 10188 handles mixed normal/expanding LoRAs while 11595 fixes load_lora_weights base-weight selection; different concrete bugs." }, { "left": "pull_request:1020", "right": "pull_request:1221", "accept": false, - "reason": "General docs update vs weight-reuse docs example; not duplicates." + "reason": "Both are README/docs updates, but one updates training docs broadly and the other adds a specific weight-reuse section." } ] }, @@ -141341,7 +141421,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 116, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -141436,10 +141516,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -141454,204 +141534,204 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a broad grab-bag of unrelated PRs: docs typos, test refactors, small pipeline fixes, and feature additions across many different components. I do not see a single underlying bug or change theme that ties them together as duplicates.", - "confidence": 0.98, + "summary": "The cluster is a loose collection of mostly unrelated docs/tests/chore PRs that only share broad terminology (e.g. Mochi, LoRA, docs, tests) or superficial filename overlap. I do not see one underlying bug/change tying them together, so the soft edges should all be rejected.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No PR here is a clear canonical representative; the PRs span unrelated subsystems and change types, so choosing one would be arbitrary.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No single PR is a good global representative because the cluster is not centered on one concrete fix or feature.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:10205", "right": "pull_request:11292", "accept": false, - "reason": "Both touch LoRA conversion code, but one is for InstructPix2Pix PEFT support and the other for broader SDXL LoRA mapping; different concrete changes." + "reason": "Different changes: InstructPix2Pix LoRA PEFT refactor vs SDXL LoRA conversion utility; only broad LoRA theme overlaps." }, { "left": "pull_request:10220", "right": "pull_request:9683", "accept": false, - "reason": "Both are test renames, but they target different models and test suites (Mochi vs CogVideoX)." + "reason": "Unrelated test renames in different pipelines (Mochi vs CogVideoX image2video)." }, { "left": "pull_request:10222", "right": "pull_request:11105", "accept": false, - "reason": "Different problems in different areas: ControlNet XLA step marking vs AutoencoderKLWan gradient checkpointing." + "reason": "Different problems: XLA step marking in ControlNet pipeline vs gradient checkpointing support for AutoencoderKLWan." }, { "left": "pull_request:10232", "right": "pull_request:10404", "accept": false, - "reason": "Both are docs cleanups, but one is a Sana attribution/link fix and the other is a broad typo pass across many pipeline docs." + "reason": "Both touch docs, but one is a Sana attribution/correction and the other is a broad typo pass across many pipeline docs." }, { "left": "pull_request:10246", "right": "pull_request:11545", "accept": false, - "reason": "Different APIs and code paths: missing attention processors docs vs adding VAE slicing/tiling methods to Flux img2img." + "reason": "Different scope: adding missing attention processors docs/code vs adding VAE slicing/tiling methods to Flux img2img." }, { "left": "pull_request:10250", "right": "pull_request:8836", "accept": false, - "reason": "Separate bugs in different model components: a MidResTemporalBlock1D downsample mistake vs SlicedAttnProcessor loop bounds." + "reason": "Different code paths: MidResTemporalBlock1D downsample bug vs SlicedAttnProcessor loop fix." }, { "left": "pull_request:10253", "right": "pull_request:8838", "accept": false, - "reason": "A documentation attribution change is unrelated to adding disable_forward_chunking to SD3 transformer." + "reason": "Unrelated chores: contribution note in docs vs adding disable_forward_chunking to SD3 transformer." }, { "left": "pull_request:10255", "right": "pull_request:10275", "accept": false, - "reason": "Mochi LoRA loader comment/docstring cleanup is unrelated to licensing header fixes in Mochi/LTX files." + "reason": "Comment/docstring cleanup in Mochi LoRA loader is unrelated to licensing-header fixes across Mochi and LTX." }, { "left": "pull_request:10255", "right": "pull_request:10585", "accept": false, - "reason": "Different Mochi-related changes: a copied-from comment fix vs a VAE annotation correction." + "reason": "Different Mochi follow-ups: LoRA loader comment fix vs VAE annotation correction." }, { "left": "pull_request:10255", "right": "pull_request:11439", "accept": false, - "reason": "Loader comment cleanup and HiDream validation bug fix address different code paths and models." + "reason": "Different Mochi-related fixes: loader docstring cleanup vs Hi-Dream LoRA validation bug." }, { "left": "pull_request:10255", "right": "pull_request:11600", "accept": false, - "reason": "A code comment/docstring fix is unrelated to a CI filename correction." + "reason": "Different area entirely: Mochi LoRA loader docs vs CI filename display fix." }, { "left": "pull_request:10268", "right": "pull_request:11105", "accept": false, - "reason": "Single-file Mochi transformer support is unrelated to AutoencoderKLWan gradient checkpointing." + "reason": "Single-file support for Mochi Transformer is unrelated to WAN autoencoder gradient checkpointing." }, { "left": "pull_request:1027", "right": "pull_request:874", "accept": false, - "reason": "Both are example docs, but they cover different examples and different kinds of additions." + "reason": "Both are example/community docs, but one adds a DDPM dataloader option and the other adds a generic community pipeline inference example." }, { "left": "pull_request:10271", "right": "pull_request:9808", "accept": false, - "reason": "Both are test refactors, but they target different test suites and different model families." + "reason": "Both are tests, but one is TorchAO serialization refactoring and the other is a broad VAE test refactor across many models." }, { "left": "pull_request:10278", "right": "pull_request:10419", "accept": false, - "reason": "Unrelated docs changes: a TorchAO example correction vs scheduler docstring parameter fixes." + "reason": "Different docs fixes: TorchAO example correction vs scheduler docstrings/parameters update." }, { "left": "pull_request:10278", "right": "pull_request:9655", "accept": false, - "reason": "Different documentation domains: TorchAO example usage vs xDiT image/path updates." + "reason": "Different documentation topics: TorchAO example vs xDiT image-path updates." }, { "left": "pull_request:10299", "right": "pull_request:10825", "accept": false, - "reason": "Both are docs, but one updates GGUF usage and the other adds training checkpoint guidance." + "reason": "Different docs topics: GGUF example cleanup vs training text2image checkpoint-loading guidance." }, { "left": "pull_request:10299", "right": "pull_request:10858", "accept": false, - "reason": "A GGUF example tweak is not the same change as correcting the pipelines overview table entry." + "reason": "GGUF example change is not the same issue as the pipelines overview category rename for I2VGen-XL." }, { "left": "pull_request:10319", "right": "pull_request:10419", "accept": false, - "reason": "Docstring wording fix in HunyuanVideo is unrelated to scheduler parameter documentation fixes." + "reason": "Docstring typo in HunyuanVideo pipeline is unrelated to scheduler parameter documentation." }, { "left": "pull_request:10319", "right": "pull_request:9072", "accept": false, - "reason": "A pipeline docstring correction is unrelated to a grammar-only docs typo fix." + "reason": "One is a HunyuanVideo docstring correction; the other is a grammar fix in ControlNet docs." }, { "left": "pull_request:10319", "right": "pull_request:9655", "accept": false, - "reason": "Different docs content and scope: HunyuanVideo docstring vs xDiT image path updates." + "reason": "HunyuanVideo docstring fix and xDiT doc image-path updates are unrelated." }, { "left": "pull_request:10323", "right": "pull_request:11325", "accept": false, - "reason": "Quantization link cleanup and HiDream docstring fixes are unrelated doc changes." + "reason": "Quantization link cleanup is unrelated to HiDream docstring replacement/doc examples." }, { "left": "pull_request:10332", "right": "pull_request:11747", "accept": false, - "reason": "A shape-mismatch error-message fix is unrelated to adding missing HiDream license headers." + "reason": "Shape error message fix in model loading is unrelated to missing HiDream license headers." }, { "left": "pull_request:10359", "right": "pull_request:10494", "accept": false, - "reason": "One is a broad test cleanup pass; the other is a CI dependency/install tweak." + "reason": "Broad test fixes in multiple files vs CI tokenizers install change; not the same bug." }, { "left": "pull_request:10365", "right": "pull_request:11105", "accept": false, - "reason": "Different code paths and models: AuraFlow XLA step marking vs Wan autoencoder gradient checkpointing." + "reason": "Both mention XLA/torch_xla in different pipeline code paths, but they fix different pipelines and behaviors." }, { "left": "pull_request:10394", "right": "pull_request:10632", "accept": false, - "reason": "Both are docs fixes, but they touch different guides and different kinds of errors." + "reason": "Docs link fix in PEFT inference tutorial vs image-path correction in para attention docs." }, { "left": "pull_request:10394", "right": "pull_request:9066", "accept": false, - "reason": "A PEFT tutorial link fix is unrelated to adding a StableCascade UNet docs page." + "reason": "PEFT doc link correction is unrelated to adding the StableCascadeUNet docs page." }, { "left": "pull_request:1041", "right": "pull_request:1727", "accept": false, - "reason": "Different test/pipeline issues from different eras and subsystems; not the same bug." + "reason": "Very different issues: older Stable Diffusion test fixes vs ONNX img2img preprocessing bug." }, { "left": "pull_request:1041", "right": "pull_request:1729", "accept": false, - "reason": "The failing tests addressed are different: general failing tests vs latent diffusion CPU/GPU handling." + "reason": "Different test areas and failure modes: stable diffusion test updates vs latent diffusion CPU/GPU test adjustments." }, { "left": "pull_request:10419", "right": "pull_request:10858", "accept": false, - "reason": "Scheduler docstring parameter fixes are unrelated to a pipelines overview table correction." + "reason": "Scheduler docstring parameter fixes are unrelated to the pipelines overview table category rename." }, { "left": "pull_request:10429", "right": "pull_request:8836", "accept": false, - "reason": "Ascend NPU rotary embedding dtype handling is a different bug than the SlicedAttnProcessor loop fix." + "reason": "Ascend NPU rotary embedding dtype fix is unrelated to sliced attention processor loop handling." }, { "left": "pull_request:1044", "right": "pull_request:178", "accept": false, - "reason": "A new music spectrogram diffusion pipeline is unrelated to stable diffusion test coverage." + "reason": "Music Spectrogram diffusion pipeline addition is unrelated to adding stable diffusion pipeline tests." } ] }, @@ -141667,7 +141747,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 117, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -141760,10 +141840,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -141778,8 +141858,8 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a mixed bag of unrelated docs, tests, chore fixes, and feature PRs. The soft pairs share broad topics but not the same concrete bug or change, so none look like true duplicates or mergeable together.", - "confidence": 0.94, + "summary": "The cluster is highly heterogeneous: it mixes unrelated doc-only edits, test cleanups, pipeline bug fixes, and model/loading feature work. None of the soft pairs look like the same underlying bug or change, so there is no true duplicate set here.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -141789,115 +141869,115 @@ "left": "pull_request:10458", "right": "pull_request:10888", "accept": false, - "reason": "One only adds slow/nightly test markers; the other adds SD3 ControlNet to AutoPipeline. Different change targets and code paths." + "reason": "One only adds test markers for SD3 LoRA integration tests; the other adds SD3 ControlNet to AutoPipeline. Different code paths and purposes." }, { "left": "pull_request:10464", "right": "pull_request:10534", "accept": false, - "reason": "NPU FLUX adaptation touches attention and pipeline code; RMSNorm NPU adaptation is a separate normalization fix." + "reason": "Both mention NPU, but one is FLUX NPU adaptation across several files and the other is an RMSNorm-specific normalization change. Not the same bug/change." }, { "left": "pull_request:10491", "right": "pull_request:9683", "accept": false, - "reason": "CI path fix for a common test file is unrelated to the CogVideoX I2V test rename/unskip." + "reason": "CI path fix for a misnamed test file vs. CogVideoX I2V test renaming/unskipping. Different targets and unrelated changes." }, { "left": "pull_request:10504", "right": "pull_request:11355", "accept": false, - "reason": "Both remove/adjust test prints and names, but they affect different test suites and are not the same bug." + "reason": "Both touch tests, but one removes debug prints and the other adjusts bitsandbytes test behavior/naming. Not the same underlying issue." }, { "left": "pull_request:10504", "right": "pull_request:9598", "accept": false, - "reason": "Removing test prints is unrelated to adding version notes in Flux LoRA integration tests." + "reason": "Test-print cleanup is unrelated to adding version notes in Flux LoRA integration tests." }, { "left": "pull_request:10505", "right": "pull_request:11355", "accept": false, - "reason": "Same broad 'test cleanup' theme, but the actual files and edits differ and do not fix one shared problem." + "reason": "General test print removal vs. bitsandbytes test consistency updates; different files and fixes." }, { "left": "pull_request:10505", "right": "pull_request:9598", "accept": false, - "reason": "Different test-cleanup/doc-note changes with no shared code-path or bug." + "reason": "Removing prints from tests is unrelated to documenting Flux LoRA test version requirements." }, { "left": "pull_request:10531", "right": "pull_request:10648", "accept": false, - "reason": "Negative-prompt docs for SD3/Flux pipelines are unrelated to DDPMScheduler docs adding sigmoid support." + "reason": "Both are docs changes, but one documents negative prompts for Flux/SD3 pipelines while the other documents DDPM sigmoid scheduling. Different functionality." }, { "left": "pull_request:10531", "right": "pull_request:10885", "accept": false, - "reason": "Both are docs, but one updates pipeline docs and the other adds new scheduler docs; no common underlying change." + "reason": "Flux negative-prompt docs and CogVideoX scheduler docs are separate documentation updates for different APIs." }, { "left": "pull_request:10544", "right": "pull_request:11743", "accept": false, - "reason": "ParaAttention documentation is unrelated to a new shortened Flux training script." + "reason": "ParaAttention documentation is unrelated to adding a shortened Flux training script with quantization." }, { "left": "pull_request:10585", "right": "pull_request:11778", "accept": false, - "reason": "Mochi VAE annotation fix is unrelated to Wan VACE mask padding logic." + "reason": "Mochi VAE annotation fix and Wan VACE mask-padding fix affect different pipelines and different bugs." }, { "left": "pull_request:1063", "right": "pull_request:1437", "accept": false, - "reason": "Both are tiny typo fixes, but they touch different variable names and code paths years apart." + "reason": "Both are typo/variable-name fixes, but in different attention/pipeline code paths and unrelated issues." }, { "left": "pull_request:10630", "right": "pull_request:11755", "accept": false, - "reason": "Updating Flax test slices is unrelated to adding CPU offload support for LTX latent upsample." + "reason": "A failing Flax scheduler slice update is unrelated to fixing CPU offload behavior for LTX latent upsample." }, { "left": "pull_request:10632", "right": "pull_request:9765", "accept": false, - "reason": "ParaAttention image-path doc fix and SD3 docstring typo fix are unrelated documentation edits." + "reason": "Docs image-path correction for ParaAttention vs. SD3 pipeline docstring typo fix; unrelated documentation items." }, { "left": "pull_request:10633", "right": "pull_request:9987", "accept": false, - "reason": "Flux IP-Adapter docs example is unrelated to the SD3.5 IP-Adapter pipeline integration feature work." + "reason": "Flux IP-Adapter docs example and SD3.5 IP-Adapter pipeline integration are related only thematically, not the same concrete change." }, { "left": "pull_request:10648", "right": "pull_request:11693", "accept": false, - "reason": "DDPM scheduler docs change is unrelated to bitsandbytes compile docs." + "reason": "Both are docs, but DDPM sigmoid scheduler docs and bitsandbytes compile docs cover different features and code paths." }, { "left": "pull_request:10648", "right": "pull_request:11731", "accept": false, - "reason": "DDPM scheduler docs and memory/offload tips address different documentation topics." + "reason": "Scheduler docs vs. memory/offload compile tip; same broad optimization/docs area, but not the same change." }, { "left": "pull_request:10684", "right": "pull_request:10802", "accept": false, - "reason": "Hunyuan-Video input-check typo fix is unrelated to a controlnet comment typo fix." + "reason": "Hunyuan-Video input validation typo and a controlnet training comment fix are unrelated typo fixes in different files." }, { "left": "pull_request:10783", "right": "pull_request:11439", "accept": false, - "reason": "fp16 unscaling bug fix in dreambooth LoRA is unrelated to Hi-Dream validation cleanup." + "reason": "Different training scripts and different bugs: fp16 unscaling fix vs. Hi-Dream validation cleanup." }, { "left": "pull_request:10819", @@ -141909,67 +141989,67 @@ "left": "pull_request:10819", "right": "pull_request:11737", "accept": false, - "reason": "LoRA docs entries are unrelated to distributed Hi-Dream LoRA training support." + "reason": "Docs-only LoRA additions vs. DeepSpeed support in Hi-Dream LoRA training are different changes." }, { "left": "pull_request:10819", "right": "pull_request:11768", "accept": false, - "reason": "LoRA docs entries and hook-removal logic fixes do not address the same issue." + "reason": "LoRA docs updates do not match the hook-removal logic fix in loaders." }, { "left": "pull_request:10845", "right": "pull_request:10990", "accept": false, - "reason": "Flux T5 training bug fix is separate from Flux quantized LoRA support, despite both involving Flux." + "reason": "Flux LoRA training bugfix and Flux quantized LoRA support are different feature areas and code paths." }, { "left": "pull_request:10858", "right": "pull_request:9655", "accept": false, - "reason": "Overview-table wording fix is unrelated to xDiT image-link repair." + "reason": "API overview table wording fix is unrelated to xDiT image-path documentation repair." }, { "left": "pull_request:10885", "right": "pull_request:11693", "accept": false, - "reason": "Scheduler docs addition is unrelated to bitsandbytes compile documentation." + "reason": "CogVideoX scheduler docs and bitsandbytes compile docs are unrelated docs changes." }, { "left": "pull_request:10938", "right": "pull_request:11248", "accept": false, - "reason": "Evaluation doc update and controlnet README fix are unrelated docs changes." + "reason": "Evaluation docs update and a README typo fix are unrelated documentation edits." }, { "left": "pull_request:10938", "right": "pull_request:11303", "accept": false, - "reason": "Evaluation guidance update is unrelated to Allegro autoencoder doc correction." + "reason": "Evaluation guidance and Allegro autoencoder docs are separate documentation topics." }, { "left": "pull_request:10938", "right": "pull_request:12193", "accept": false, - "reason": "Evaluation docs and Qwen README launcher rename are unrelated." + "reason": "Evaluation docs and Qwen README path fix do not target the same issue." }, { "left": "pull_request:10938", "right": "pull_request:9553", "accept": false, - "reason": "Evaluation docs tip and distributed inference docs are different topics." + "reason": "Evaluation conceptual docs are unrelated to distributed inference device-map documentation." }, { "left": "pull_request:10947", "right": "pull_request:11550", "accept": false, - "reason": "Both are single-file-related, but one adds Sana transformer support and the other adds HiDream GGUF/single-file support; different models and loaders." + "reason": "Both are single-file/loading related, but one adds Sana transformer support and the other adds HiDream GGUF/single-file support. Different models and implementations." }, { "left": "pull_request:10975", "right": "pull_request:10991", "accept": false, - "reason": "Wan I2V pipeline bug fixes and Wan single-file support are distinct changes, not one shared fix." + "reason": "Wan pipeline bug fixes and Wan single-file support are distinct changes with different goals and code paths." } ] }, @@ -141985,7 +142065,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 118, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -142079,10 +142159,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -142097,54 +142177,54 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster does not look like a duplicate set; it mixes unrelated docs, tests, CI, and feature PRs. All soft edges were rejected because they do not fix the same concrete bug/change, even when they touch the same file or subsystem.", + "summary": "This is a heterogeneous set of mostly unrelated PRs. The soft links are superficial overlaps in docs, tests, or broad subsystem names, not the same underlying bug or change.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "10991 is the strongest representative PR: it adds Wan single-file support across loaders, models, tests, and docs, giving it the clearest self-contained scope among the set.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "10991 is the best standalone representative if one PR must anchor the cluster, because it is feature-complete and internally coherent; the others are mostly tiny typo/test/CI/doc fixes or separate features.", "soft_edge_verdicts": [ { "left": "pull_request:10977", "right": "pull_request:11434", "accept": false, - "reason": "Unrelated test additions: save/load optional components vs an import fix." + "reason": "Different test fixes in unrelated files: save/load optional-components coverage vs an import adjustment in common test utilities." }, { "left": "pull_request:10977", "right": "pull_request:9683", "accept": false, - "reason": "Different pipeline test changes with no shared bug or code path." + "reason": "Both are test-related, but they target different pipelines and different behaviors; not the same bug or change." }, { "left": "pull_request:10978", "right": "pull_request:9627", "accept": false, - "reason": "LoRA conversion bugfix vs a test warning-capture change." + "reason": "LoRA conversion logic for OneTrainer Flux is unrelated to an SDXL test timing/warning fix." }, { "left": "pull_request:10991", "right": "pull_request:11053", "accept": false, - "reason": "Both involve Wan docs/pipelines, but they add different features (single-file loading vs video-to-video)." + "reason": "Same Wan docs file, but one adds single-file loading while the other adds a video-to-video pipeline; distinct features and code paths." }, { "left": "pull_request:1103", "right": "pull_request:60", "accept": false, - "reason": "A README URL fix is unrelated to a scheduler implementation change." + "reason": "A README URL/license fix is unrelated to a scheduler initialization rewrite." }, { "left": "pull_request:11037", "right": "pull_request:11740", "accept": false, - "reason": "CI dependency setup vs GGUF compile/offload tests are different changes." + "reason": "CI dependency plumbing for bnb tests is unrelated to adding GGUF compile/offload tests." }, { "left": "pull_request:11037", "right": "pull_request:11953", "accept": false, - "reason": "CI dependency setup vs HiDream test slice updates are unrelated." + "reason": "Workflow dependency changes do not match a HiDream test-slice update." }, { "left": "pull_request:1104", @@ -142156,139 +142236,139 @@ "left": "pull_request:11113", "right": "pull_request:9646", "accept": false, - "reason": "AutoencoderKL training fix vs Flux LoRA target-module configurability are different bugs/features." + "reason": "AutoencoderKL training-script bug fix and Flux Dreambooth LoRA target-module configurability are separate changes." }, { "left": "pull_request:11204", "right": "pull_request:9754", "accept": false, - "reason": "PyTorch/XLA SDXL training work vs Flux quantized training are separate efforts." + "reason": "PyTorch/XLA SDXL training support is unrelated to Flux quantized training example work." }, { "left": "pull_request:11212", "right": "pull_request:11634", "accept": false, - "reason": "General MPS docs updates vs a TorchAO doc link fix are not the same change." + "reason": "Both are docs-only, but one is an MPS badge update and the other is a torchao link update; not the same issue." }, { "left": "pull_request:11248", "right": "pull_request:11303", "accept": false, - "reason": "A README typo fix and an Allegro doc class-name fix are unrelated." + "reason": "A Flux README typo fix and an autoencoder docs symbol correction are unrelated documentation edits." }, { "left": "pull_request:11248", "right": "pull_request:12193", "accept": false, - "reason": "Different README fixes for different examples, no shared bug." + "reason": "Different README corrections for different examples; no shared underlying bug." }, { "left": "pull_request:113", "right": "pull_request:169", "accept": false, - "reason": "Scheduler initialization fix vs score-SDE documentation are unrelated." + "reason": "Score SDE VE scheduler initialization and PNDM JAXification are different scheduler changes." }, { "left": "pull_request:113", "right": "pull_request:400", "accept": false, - "reason": "Scheduler code fix vs a docs/pipeline page update are not duplicates." + "reason": "The doc PR describes Score SDE VE, but the code fix is a scheduler init change; not the same concrete change." }, { "left": "pull_request:11303", "right": "pull_request:12193", "accept": false, - "reason": "Different documentation typo fixes in different example READMEs." + "reason": "Autoencoder docs class-name fix and Qwen README filename fix are unrelated docs updates." }, { "left": "pull_request:11325", "right": "pull_request:12259", "accept": false, - "reason": "HiDream docstring cleanup vs VAE scale-factor docs are unrelated." + "reason": "HiDream docstring cleanup and distributed-inference VAE scale-factor correction are separate docs fixes." }, { "left": "pull_request:11434", "right": "pull_request:11789", "accept": false, - "reason": "Import fix vs HunyuanVideo device-test changes are different issues." + "reason": "Import cleanup in test utilities is unrelated to HunyuanVideo device-test behavior." }, { "left": "pull_request:11434", "right": "pull_request:12070", "accept": false, - "reason": "Import fix vs Qwen test slice update are unrelated." + "reason": "Common test import fix does not match Qwen inference slice updates." }, { "left": "pull_request:11439", "right": "pull_request:11600", "accept": false, - "reason": "Hi-Dream validation bugfix vs CI failure-filename fix are unrelated." + "reason": "HiDream LoRA validation logic and CI failure-file naming are unrelated." }, { "left": "pull_request:11439", "right": "pull_request:12343", "accept": false, - "reason": "Different bugs: Hi-Dream validation cleanup vs VAE slicing decode behavior." + "reason": "HiDream validation cleanup is unrelated to VAE slicing decode behavior." }, { "left": "pull_request:1147", "right": "pull_request:2479", "accept": false, - "reason": "Flax text-encoder loading fix vs LoRA training support are different changes." + "reason": "Flax text-encoder loading-from-subfolder and LoRA support for the clip text encoder are different training-script changes." }, { "left": "pull_request:11473", "right": "pull_request:11623", "accept": false, - "reason": "TorchAO doc typo fix vs a Flux code comment typo fix are unrelated." + "reason": "torchao documentation typo fix and a Flux pipeline typo fix are separate doc/code edits." }, { "left": "pull_request:11473", "right": "pull_request:11634", "accept": false, - "reason": "Same docs file, but different fixes: quantization naming vs external doc link." + "reason": "Same torchao doc file, but different content fixes: quantization shorthand typo vs external link update." }, { "left": "pull_request:11473", "right": "pull_request:12487", "accept": false, - "reason": "TorchAO docs typo and broad broken-link cleanup are not the same underlying issue." + "reason": "torchao docs typo fix is not the same as a broad broken-links sweep." }, { "left": "pull_request:11473", "right": "pull_request:12552", "accept": false, - "reason": "Different typo/doc cleanup targets with no shared change." + "reason": "torchao quantization-doc wording and Kandinsky5 typo fixes are unrelated documentation changes." }, { "left": "pull_request:11478", "right": "pull_request:11953", "accept": false, - "reason": "HiDream transformer test enablement vs HiDream fast test slices are separate test additions." + "reason": "HiDream transformer tests and HiDream image fast-test slices are both for the model family, but they are different test additions, not one duplicate change." }, { "left": "pull_request:11485", "right": "pull_request:11602", "accept": false, - "reason": "Different new pipelines (Flex.2 vs Sana Sprint img2img) despite shared registration files." + "reason": "A new Flex.2 pipeline and a Sana Sprint img2img pipeline are entirely different features." }, { "left": "pull_request:11529", "right": "pull_request:11983", "accept": false, - "reason": "No evidence of the same bug; one is a docs update with no diff preview, the other changes installation instructions." + "reason": "Unspecified docs edit in inpaint.md is unrelated to an installation command fix." }, { "left": "pull_request:11537", "right": "pull_request:11697", "accept": false, - "reason": "UNet torch.compile tests vs TorchAO quantization compile/offload tests are different code paths." + "reason": "UNet torch.compile test coverage is unrelated to TorchAO/bitsandbytes compile-offloading tests." }, { "left": "pull_request:11558", "right": "pull_request:11954", "accept": false, - "reason": "Layerwise casting/group offloading tests vs Hunyuan Video slice tests are unrelated." + "reason": "Layerwise casting plus group offloading tests are unrelated to Hunyuan Video test-slice updates." } ] }, @@ -142304,7 +142384,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 119, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -142397,10 +142477,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -142415,8 +142495,8 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab-bag of unrelated documentation, CI, test-gating, and small code-fix PRs. The soft pairs mostly share a file path or broad subsystem, but they address different concrete problems and should not be deduplicated.", - "confidence": 0.97, + "summary": "This cluster is mostly superficial similarity rather than true duplication: the paired PRs span unrelated docs, test tweaks, and bugfixes across different code paths. I would reject all soft edges.", + "confidence": 0.93, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -142426,187 +142506,187 @@ "left": "pull_request:1157", "right": "pull_request:848", "accept": false, - "reason": "Different bugs: Flax report_to/push_to_hub handling vs Flax latent shape/height-width fix." + "reason": "Both are Flax-related, but one fixes training arg handling/report_to/push_to_hub while the other fixes latent shape from width/height; different bugs." }, { "left": "pull_request:11600", "right": "pull_request:12228", "accept": false, - "reason": "CI filename fix vs a LoRA tutorial typo; same general area only by coincidence." + "reason": "CI filename display fix vs LoRA docs typo; no shared code path or underlying bug." }, { "left": "pull_request:11613", "right": "pull_request:11979", "accept": false, - "reason": "Both touch test modeling common code, but one changes CI report filenames and the other adds torch version guards." + "reason": "Both touch torch.compile tests, but 11613 reverts CI/test changes while 11979 only adds version guards; different fixes." }, { "left": "pull_request:1162", "right": "pull_request:1360", "accept": false, - "reason": "README/community links cleanup is unrelated to adding a new community pipeline implementation." + "reason": "README/community link cleanup vs adding a K-Diffusion community pipeline; unrelated changes." }, { "left": "pull_request:1162", "right": "pull_request:925", "accept": false, - "reason": "Link cleanup vs LPW prompt truncation logic; not the same change." + "reason": "Community links/docs cleanup and LPW token padding logic are unrelated." }, { "left": "pull_request:11634", "right": "pull_request:12018", "accept": false, - "reason": "Both are docs link edits in quantization docs, but they fix different links in different pages." + "reason": "Both are quantization docs edits, but they fix different link/text wording issues in separate pages." }, { "left": "pull_request:1167", "right": "pull_request:120", "accept": false, - "reason": "Both modify README, but one adds an Intel GPU note and the other adds example sections." + "reason": "README note about Intel XPU vs main README examples update; not the same underlying change." }, { "left": "pull_request:1169", "right": "pull_request:1601", "accept": false, - "reason": "Same Stable Diffusion pipeline file, but one is an MPS scheduler/device precision fix and the other fixes CFG control flow." + "reason": "Different pipeline bugs: MPS scheduler dtype handling vs classifier-free guidance branching." }, { "left": "pull_request:1171", "right": "pull_request:3132", "accept": false, - "reason": "Both are typo fixes, but they are in different pipelines and different text." + "reason": "Typo cleanup in community pipeline docs vs a score_sde scheduler type annotation fix; unrelated." }, { "left": "pull_request:11715", "right": "pull_request:11979", "accept": false, - "reason": "Both adjust compilation-related tests, but one gates bitsandbytes tests and the other gates torch version for compile tests." + "reason": "Bitsandbytes test version annotations are unrelated to torch.compile test version gating." }, { "left": "pull_request:1173", "right": "pull_request:1416", "accept": false, - "reason": "Both concern ONNX conversion/inference, but they target different concrete issues and cannot be merged into one fix." + "reason": "Both are ONNX-related, but one targets scheduler compatibility/safety checker and the other fixes conversion/inference feature_extractor handling; distinct concrete bugs." }, { "left": "pull_request:11847", "right": "pull_request:12259", "accept": false, - "reason": "Single-file Wan example docs vs distributed inference VAE scale-factor docs; unrelated." + "reason": "Wan single-file docs example vs distributed inference VAE scale-factor docs; unrelated." }, { "left": "pull_request:1186", "right": "pull_request:848", "accept": false, - "reason": "DreamBooth Flax report_to/push_to_hub fix is unrelated to the Flax width/height bug." + "reason": "Dreambooth Flax report_to/push_to_hub setup vs Flax pipeline width/height bug; different code paths." }, { "left": "pull_request:1187", "right": "pull_request:848", "accept": false, - "reason": "Extra copy-pasta cleanup in DreamBooth Flax is not the same bug as the latent shape issue." + "reason": "Dreambooth Flax cleanup vs latent shape handling in a different Flax pipeline; not the same bug." }, { "left": "pull_request:11953", "right": "pull_request:12070", "accept": false, - "reason": "Different fast-test slice updates for different pipelines; not the same underlying issue." + "reason": "Both update test slices, but for different models (HiDream vs Qwen) and different expected outputs." }, { "left": "pull_request:11953", "right": "pull_request:12775", "accept": false, - "reason": "HiDream test slice update and Z-Image docs addition are unrelated." + "reason": "HiDream test slices vs Z-Image docs addition; unrelated." }, { "left": "pull_request:11983", "right": "pull_request:12827", "accept": false, - "reason": "Installation.md wording fix vs distributed_inference syntax correction; separate docs changes." + "reason": "Both touch docs syntax, but one is uv install wording and the other is shell block formatting; no shared bug." }, { "left": "pull_request:120", "right": "pull_request:16", "accept": false, - "reason": "Both edit README.md, but they are from different eras and add different content." + "reason": "Both edit README.md, but one adds examples while the other fixes wording; not the same change." }, { "left": "pull_request:120", "right": "pull_request:1960", "accept": false, - "reason": "Main README examples addition and ColossalAI README install instructions are unrelated." + "reason": "Main README examples update vs ColossalAI README install instructions; unrelated." }, { "left": "pull_request:12018", "right": "pull_request:12902", "accept": false, - "reason": "Quantization overview link wording vs context-parallel device_map usage; different docs problems." + "reason": "Quantization docs wording vs distributed inference device_map usage; different documentation fixes." }, { "left": "pull_request:1211", "right": "pull_request:41", "accept": false, - "reason": "Both are comment/docstring cleanups, but they touch different files and different commentary." + "reason": "Flax usage comments vs resnet comment cleanup; unrelated." }, { "left": "pull_request:12129", "right": "pull_request:12228", "accept": false, - "reason": "NPU attention typo fix vs LoRA tutorial punctuation/typo fix; unrelated." + "reason": "NPU attention typo fix vs LoRA docs typo; different subsystems and purposes." }, { "left": "pull_request:12129", "right": "pull_request:12675", "accept": false, - "reason": "Different typo fixes in different docs/code paths." + "reason": "NPU attention kernel typo fix vs Sana video docs typo; unrelated." }, { "left": "pull_request:1218", "right": "pull_request:154", "accept": false, - "reason": "Same conversion script family, but one adds scheduler options and the other is the initial script creation." + "reason": "Both concern conversion scripts, but 154 adds the original SD converter and 1218 extends it with new scheduler options; not the same bug." }, { "left": "pull_request:12193", "right": "pull_request:12890", "accept": false, - "reason": "Qwen README filename update vs Wan markdown cleanup; unrelated docs changes." + "reason": "Qwen Dreambooth filename/docs tweak vs Wan docs cleanup; unrelated." }, { "left": "pull_request:12197", "right": "pull_request:12775", "accept": false, - "reason": "Flux Kontext inpainting docs and Z-Image docs are distinct features." + "reason": "Flux Kontext inpainting docs vs Z-Image docs; unrelated features." }, { "left": "pull_request:1221", "right": "pull_request:2764", "accept": false, - "reason": "Weight reuse docs and torch 2 benchmark docs address completely different topics." + "reason": "Weight reuse README docs vs torch 2.0 benchmark docs; no shared change." }, { "left": "pull_request:12228", "right": "pull_request:12675", "accept": false, - "reason": "Two separate typo/doc fixes in different tutorials; not the same underlying issue." + "reason": "Two independent typo/doc fixes in different pages; not the same underlying issue." }, { "left": "pull_request:12247", "right": "pull_request:12980", "accept": false, - "reason": "AutoGuidance validation logic bug and Mellon param/docstring work are unrelated." + "reason": "AutoGuidance validation logic bug vs Mellon param/docstring refactor; unrelated." }, { "left": "pull_request:12259", "right": "pull_request:12360", "accept": false, - "reason": "Distributed inference VAE scale factor docs vs Chroma guidance docs; no shared bug." + "reason": "Distributed inference VAE scale-factor docs vs Chroma guidance docs rewrite; different documentation topics." }, { "left": "pull_request:12259", "right": "pull_request:12883", "accept": false, - "reason": "Different documentation typos in different quantization/training pages." + "reason": "Distributed inference docs fix vs torchao quantization typo fix; unrelated." } ] }, @@ -142622,7 +142702,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 120, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -142713,10 +142793,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -142731,180 +142811,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate cluster emerges here: the PRs span unrelated docs updates, test additions, typo fixes, and separate feature work across different subsystems. I rejected all soft edges because none clearly describe the same underlying bug or mergeable change.", - "confidence": 0.97, + "summary": "This cluster is mostly unrelated PRs spanning docs typos, test updates, and separate feature additions. There is no clear duplicate issue/PR at the cluster level; the only loosely related items are in the Versatile Diffusion / modular docs areas, but they still address different changes.", + "confidence": 0.96, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No single PR is a good canonical representative for this set because the items are heterogeneous and mostly unrelated. If forced, the most substantive standalone change is PR 1283, but it is not a duplicate target for the others.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:1227", "right": "pull_request:1487", "accept": false, - "reason": "Both are docs changes, but they touch different pages and different topics; not the same change or bug." + "reason": "Both are docs PRs, but one adds multi-modal examples while the other expands Stable Diffusion xformers docs; different scope and change intent." }, { "left": "pull_request:1227", "right": "pull_request:2433", "accept": false, - "reason": "Different documentation areas: multi-modal examples overview vs SDEdit note in img2img docs." + "reason": "Different docs topics: multi-modal examples vs an SDEdit note for img2img." }, { "left": "pull_request:1229", "right": "pull_request:2002", "accept": false, - "reason": "One is a DreamBooth README install note, the other is a broad LoRA fine-tuning support PR; not the same underlying change." + "reason": "One is a small DreamBooth README install note, the other is a broad LoRA fine-tuning support PR touching many files." }, { "left": "pull_request:1256", "right": "pull_request:824", "accept": false, - "reason": "A pipeline test change and a Flax test addition are unrelated test work, not the same bug fix." + "reason": "One adds a negative_prompt test in PyTorch pipelines; the other is a Flax scheduler/pipeline test change. Not the same bug or change." }, { "left": "pull_request:12675", "right": "pull_request:13040", "accept": false, - "reason": "Different docs fixes: Sana video docs vs CLI warning grammar." + "reason": "Both are typo/wording fixes, but in unrelated docs/code paths: Sana video docs vs fp16_safetensors CLI warnings." }, { "left": "pull_request:12675", "right": "pull_request:13181", "accept": false, - "reason": "Different docs pages and subjects; no shared concrete bug or change." + "reason": "Sana video docs typo versus torchrun argument order docs; unrelated edits." }, { "left": "pull_request:12681", "right": "pull_request:13293", "accept": false, - "reason": "CPU test fixes for different pipelines, with different code paths and failures." + "reason": "Both relate to tests/CI, but they fix different failing pipelines and different code paths." }, { "left": "pull_request:1269", "right": "pull_request:473", "accept": false, - "reason": "Shared filename aside, one is a stable diffusion refactor and the other is an inpainting sample fix; not the same issue." + "reason": "Shared Stable Diffusion area, but 1269 is a refactor across multiple pipelines while 473 is a specific inpainting sample fix." }, { "left": "pull_request:12731", "right": "pull_request:13283", "accept": false, - "reason": "Separate docs edits for different topics; no common underlying fix." + "reason": "Both are docs, but one is Bria Fibo guidance and the other updates FlashAttention-3 support wording; no shared underlying change." }, { "left": "pull_request:12775", "right": "pull_request:13344", "accept": false, - "reason": "Z-Image docs addition vs GLM Image model tests are unrelated." + "reason": "Z-Image docs versus GLM Image transformer tests; unrelated artifacts and purposes." }, { "left": "pull_request:1283", "right": "pull_request:1373", "accept": false, - "reason": "Both concern Versatile Diffusion, but one adds the model and the other extends it with additional pipelines/features; not the same mergeable change." + "reason": "Both are Versatile Diffusion-related, but 1283 adds the base model integration while 1373 extends it with additional pipelines; not the same concrete change." }, { "left": "pull_request:1283", "right": "pull_request:21", "accept": false, - "reason": "Completely different areas: Versatile Diffusion model integration vs early generic model tests." + "reason": "Versatile Diffusion model integration versus an early Stable Diffusion test PR; unrelated." }, { "left": "pull_request:12856", "right": "pull_request:13253", "accept": false, - "reason": "One adds Z-Image implementation/docs, the other refactors Z Image transformer tests; related subsystem, but not the same change." + "reason": "One adds the z-image-omni-base implementation, the other refactors Z Image transformer tests. Related subsystem, but different underlying change." }, { "left": "pull_request:12883", "right": "pull_request:13499", "accept": false, - "reason": "Both mention torchao/quantization, but they update different docs with different purposes and text." + "reason": "Both mention quantization-related docs, but one fixes a torchao typo and the other adds general torchao mentions in speed/memory docs." }, { "left": "pull_request:1291", "right": "pull_request:2209", "accept": false, - "reason": "Stable Diffusion notebook link fix vs img2img argument-name fix are unrelated." + "reason": "Stable Diffusion notebook link fix versus an img2img argument-name fix; different examples and bugs." }, { "left": "pull_request:12914", "right": "pull_request:13134", "accept": false, - "reason": "A typo fix in an example and a typing cleanup in a transformer model are not the same bug." + "reason": "Cosmos pipeline doc typo versus a transformer typing annotation fix; unrelated changes." }, { "left": "pull_request:12971", "right": "pull_request:13283", "accept": false, - "reason": "Documentation reordering vs an attention-backend support note; no shared code-path change." + "reason": "Distributed inference docs reordering versus attention backend support wording; different documentation issues." }, { "left": "pull_request:12984", "right": "pull_request:13099", "accept": false, - "reason": "Different docs for Flux2 Klein vs AutoModel custom-code support." + "reason": "Flux2 Klein docs addition versus AutoModel custom-code guide; separate docs/features." }, { "left": "pull_request:13029", "right": "pull_request:13410", "accept": false, - "reason": "Both are modular/agent docs, but they are distinct document rewrites and not the same concrete change." + "reason": "The modular quickstart guide and the AI agent modular-doc restructuring are different documentation efforts, not the same change." }, { "left": "pull_request:131", "right": "pull_request:316", "accept": false, - "reason": "README syntax-highlighting cleanup is unrelated to the contributions/installation README rewrite." + "reason": "Both touch README.md, but one adds syntax highlighting and the other adds installation/contribution content; not duplicates." }, { "left": "pull_request:131", "right": "pull_request:72", "accept": false, - "reason": "README formatting and 1D up/downsample tests are unrelated changes." + "reason": "README formatting change versus 1D up/downsample test coverage; unrelated." }, { "left": "pull_request:1311", "right": "pull_request:178", "accept": false, - "reason": "AltDiffusion test/dependency work is unrelated to Stable Diffusion pipeline tests." + "reason": "AltDiffusion test/dependency work versus an early Stable Diffusion pipeline test; different models and goals." }, { "left": "pull_request:1311", "right": "pull_request:3400", "accept": false, - "reason": "Adding omegaconf and AltDiffusion tests are separate changes, not the same bug fix." + "reason": "Shared dependency/setup files, but 1311 adds AltDiffusion tests while 3400 adds omegaconf for tests; not the same bug/change." }, { "left": "pull_request:13120", "right": "pull_request:13192", "accept": false, - "reason": "Two different modular-pipeline tests covering different invariants." + "reason": "Both add modular pipeline tests, but one checks load-id tagging and the other checks saved index consistency; different behaviors." }, { "left": "pull_request:13158", "right": "pull_request:13567", "accept": false, - "reason": "A QwenImage docs addition and agent-docs updates are unrelated." + "reason": "Qwen-Image docs versus AI agent docs; no overlap in underlying change." }, { "left": "pull_request:1319", "right": "pull_request:580", "accept": false, - "reason": "Flax scheduler subfolder loading fix vs a Flax scheduler test PR are not the same change." + "reason": "Both are Flax-related, but one fixes scheduler subfolder loading and the other is an initial Flax scheduler test/change set." }, { "left": "pull_request:13283", "right": "pull_request:13400", "accept": false, - "reason": "Attention backend support docs vs profiling guide README edits are unrelated." + "reason": "Attention backend support note versus profiling guide acknowledgements; unrelated docs changes." }, { "left": "pull_request:1373", "right": "pull_request:1605", "accept": false, - "reason": "Extended Versatile Diffusion feature work and an upcast_attention parameter addition are different changes, even if they touch the same area." + "reason": "Versatile Diffusion expansion versus an upcast_attention parameter addition in one helper file; different concrete changes." } ] }, @@ -142920,7 +143000,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 121, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -143012,10 +143092,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -143030,186 +143110,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a loose cluster of Diffusers PRs with only superficial similarity across stable-diffusion docs, training scripts, and pipeline bugfixes. The soft edges mostly connect broad topical overlap rather than the same concrete change.", - "confidence": 0.14, + "summary": "This cluster is mostly a grab-bag of unrelated diffusers PRs; the soft links are driven by broad naming/subsystem similarity rather than the same concrete fix. I would not merge any of the suggested soft edges.", + "confidence": 0.31, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a defensible canonical here: the items span unrelated fixes and docs updates, so any one PR would be an arbitrary representative rather than a true duplicate target.", + "canonical_pr_reason": "No issue artifacts are present. If a PR must represent the cluster, #1415 is the broadest merged stable-diffusion code fix and the most central example among the candidates.", "best_issue_reason": null, - "best_pr_reason": "PR 1415 is the broadest representative of the shared stable-diffusion pipeline area, but it is still too generic to serve as a safe duplicate canonical for the rest of the set.", + "best_pr_reason": "#1415 is the best representative PR because it is a merged, multi-file stable-diffusion fix touching several pipelines/tests, whereas the others are narrower one-off docs/tests/script changes.", "soft_edge_verdicts": [ { "left": "pull_request:1415", "right": "pull_request:1849", "accept": false, - "reason": "Different fixes in the same file family: 1415 is a broad stable-diffusion cleanup, while 1849 specifically threads cross_attention_kwargs through Versatile Diffusion." + "reason": "Both touch attention-related code, but #1415 is a broad stable-diffusion cleanup and #1849 is a separate Versatile Diffusion cross_attention_kwargs fix. Different bug/change." }, { "left": "pull_request:1415", "right": "pull_request:356", "accept": false, - "reason": "1415 is a pipeline bugfix batch; 356 is the initial text-to-image fine-tuning script/docs. Different change types and targets." + "reason": "#1415 fixes stable-diffusion pipeline behavior; #356 adds training example docs/script. Not the same underlying change." }, { "left": "pull_request:1437", "right": "pull_request:3118", "accept": false, - "reason": "A typo fix in a deprecation string is unrelated to ControlNet prepare_image default-argument behavior." + "reason": "A typo fix in one pipeline file is unrelated to a ControlNet API-default fix. Different scope and code path." }, { "left": "pull_request:1476", "right": "pull_request:3472", "accept": false, - "reason": "Both are docs, but one is about training dependencies and the other adds DreamBooth IF documentation; not the same change." + "reason": "Training dependency docs and DreamBooth IF docs are separate documentation updates, not the same change." }, { "left": "pull_request:1487", "right": "pull_request:3388", "accept": false, - "reason": "One adds xformers docs, the other updates install instructions to include Transformers. Different documentation issues." + "reason": "xformers docs and install-docs changes are unrelated documentation fixes." }, { "left": "pull_request:1510", "right": "pull_request:154", "accept": false, - "reason": "Both touch the conversion script, but 154 introduces it and 1510 expands schedulers/docs around it; not the same bug or change set." + "reason": "They both mention the conversion script, but #154 introduces it and #1510 later extends the whole stable-diffusion stack. Not the same concrete change." }, { "left": "pull_request:154", "right": "pull_request:299", "accept": false, - "reason": "Initial conversion-script addition versus a PNDM scheduler hotfix are unrelated." + "reason": "The original conversion script addition and a PNDM scheduler hotfix are unrelated." }, { "left": "pull_request:154", "right": "pull_request:904", "accept": false, - "reason": "Conversion script creation and Stable Diffusion inpainting rewrite address different code paths." + "reason": "Conversion-script creation vs inpainting pipeline rewrite are different changes." }, { "left": "pull_request:1561", "right": "pull_request:2257", "accept": false, - "reason": "Both community PRs, but one adds a mega pipeline and the other adds UnCLIP text interpolation; different features." + "reason": "Both are community pipeline PRs, but they implement different pipelines with different behavior." }, { "left": "pull_request:1564", "right": "pull_request:2557", "accept": false, - "reason": "DreamBooth default-accelerate docs and intermediate logging in training code are unrelated changes." + "reason": "Docs-only default accelerate config vs DreamBooth logging/training-script changes are unrelated." }, { "left": "pull_request:1576", "right": "pull_request:2077", "accept": false, - "reason": "ORT integration for training examples is unrelated to allowing validation images in textual inversion." + "reason": "ORT integration for training examples and textual-inversion validation support are separate feature changes." }, { "left": "pull_request:1576", "right": "pull_request:375", "accept": false, - "reason": "ORT example integration and Colab links in textual inversion are different documentation/example updates." + "reason": "ORT integration and Colab-link docs are not the same underlying change." }, { "left": "pull_request:1586", "right": "pull_request:2629", "accept": false, - "reason": "Both mention non-square support, but they apply to different pipelines and different implementations; not the same concrete fix." + "reason": "Both address non-square images, but for different pipelines and different fixes. Too broad to be one PR." }, { "left": "pull_request:1587", "right": "pull_request:2209", "accept": false, - "reason": "Paint-by-Example width handling and an img2img argument-name doc fix are unrelated." + "reason": "Default image width logic in Paint by Example is unrelated to an img2img docs argument-name correction." }, { "left": "pull_request:1591", "right": "pull_request:75", "accept": false, - "reason": "Both involve attention.py, but 1591 fixes slice upcasting while 75 fixes Glide constructor behavior; different bugs." + "reason": "Shared file does not make them duplicates: one fixes sliced-attention upcasting, the other changes Glide attention initialization logic." }, { "left": "pull_request:1601", "right": "pull_request:832", "accept": false, - "reason": "One fixes classifier-free guidance branching in diffusion pipelines; the other fixes NSFW handling in Flax Stable Diffusion." + "reason": "Classifier-free-guidance bug in Stable/Alt Diffusion is unrelated to a Flax NSFW masking fix." }, { "left": "pull_request:1601", "right": "pull_request:957", "accept": false, - "reason": "Classifier-free guidance control flow and a DDPMScheduler tensor_format bug are unrelated." + "reason": "A diffusion-pipeline control-flow bug and a scheduler-argument cleanup in training code are unrelated." }, { "left": "pull_request:1625", "right": "pull_request:178", "accept": false, - "reason": "K-diffusion default size handling and a stable-diffusion test update are not the same change." + "reason": "k-diffusion height/width defaults and a stable-diffusion test update do not fix the same problem." }, { "left": "pull_request:1640", "right": "pull_request:942", "accept": false, - "reason": "Auto-enabling xformers in training scripts and an MPS dtype fix in LMS scheduler address different problems." + "reason": "Auto-enabling xformers in training scripts is unrelated to the MPS float64 scheduler issue." }, { "left": "pull_request:1668", "right": "pull_request:3472", "accept": false, - "reason": "DreamBooth checkpointing/state restore and IF DreamBooth docs are unrelated." + "reason": "DreamBooth checkpointing and DreamBooth IF docs are different artifacts and different changes." }, { "left": "pull_request:168", "right": "pull_request:2278", "accept": false, - "reason": "The original Stable Diffusion pipeline introduction is not a duplicate of a later video Stable Diffusion project." + "reason": "The original Stable Diffusion pipeline intro and a later video-stable-diffusion project are not the same change." }, { "left": "pull_request:168", "right": "pull_request:309", "accept": false, - "reason": "A core pipeline PR and a README link fix are not the same underlying change." + "reason": "Base pipeline introduction vs a README link fix are unrelated." }, { "left": "pull_request:1687", "right": "pull_request:3472", "accept": false, - "reason": "Checkpointing across training scripts versus adding IF guidance to DreamBooth docs are different updates." + "reason": "Checkpointing across training scripts and IF DreamBooth docs address different needs." }, { "left": "pull_request:1727", "right": "pull_request:3449", "accept": false, - "reason": "ONNX img2img preprocessing fixes and CI workflow rework are unrelated." + "reason": "ONNX img2img preprocessing fix is unrelated to a workflow/test rework PR." }, { "left": "pull_request:1729", "right": "pull_request:62", "accept": false, - "reason": "Test gating for CPU/GPU behavior is unrelated to the latent-diffusion-uncond pipeline fix." + "reason": "A test-only LDM CPU/GPU adjustment and an LDM unconditional pipeline fix are not the same bug." }, { "left": "pull_request:1756", "right": "pull_request:1763", "accept": false, - "reason": "Both touch attention masks, but one adds mask support to U-Clip while the other removes an unused mask from Versatile Diffusion; different code paths." + "reason": "Both mention attention masks, but one adds mask handling for UCLIP while the other removes an unused mask in Versatile Diffusion. Opposite and different code paths." }, { "left": "pull_request:1763", "right": "pull_request:1849", "accept": false, - "reason": "Versatile Diffusion attention-mask cleanup is not the same as threading cross_attention_kwargs through the model." + "reason": "Unused attention-mask cleanup and cross_attention_kwargs plumbing are distinct Versatile Diffusion fixes." }, { "left": "pull_request:1769", "right": "pull_request:1967", "accept": false, - "reason": "UnCLIP CPU offload test adjustments and a WIP CPU offload example in Stable Diffusion are not duplicates." + "reason": "UnCLIP CPU-offload test changes and a WIP CPU-offload example are unrelated." }, { "left": "pull_request:1778", "right": "pull_request:2428", "accept": false, - "reason": "Textual inversion hardcoded-name cleanup and removing author names from docs are different editorial changes." + "reason": "Removing hardcoded names in training scripts is unrelated to removing author names in docs." } ] }, @@ -143225,7 +143305,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 122, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -143317,10 +143397,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -143335,186 +143415,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a duplicate set; it mixes unrelated docs edits, new pipelines, and isolated bug/typo fixes across different code paths. No soft-edge pair looks mergeable into a single PR.", - "confidence": 0.97, + "summary": "This cluster is mostly a loose set of unrelated docs, typo, example, and feature PRs; there are no true duplicates here. The most substantial code change is the textual inversion loader PR, but it is not a duplicate of the other items.", + "confidence": 0.82, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items are heterogeneous and do not share the same underlying bug or change.", + "canonical_pr_reason": "PR 2009 is the strongest representative because it is the largest cross-cutting code change in the set, with broad loader/pipeline impact and substantial review, unlike the many small docs/typo fixes.", "best_issue_reason": null, - "best_pr_reason": "No best PR for deduping; there is no single change that meaningfully represents the rest.", + "best_pr_reason": "PR 2009 is the best overall representative of the cluster: it is the most substantive, code-centric change and the closest thing to a central artifact among mostly unrelated small PRs.", "soft_edge_verdicts": [ { "left": "pull_request:178", "right": "pull_request:296", "accept": false, - "reason": "Both touch Stable Diffusion tests, but one adds a pipeline slice-update test while the other only fixes auth token usage in CI/tests." + "reason": "Both mention Stable Diffusion tests, but one adds a test while the other fixes auth tokens in CI; different problems and code paths." }, { "left": "pull_request:178", "right": "pull_request:388", "accept": false, - "reason": "Different scope: stable diffusion test data update vs an unconditional latent diffusion pipeline/docs addition." + "reason": "Unrelated artifacts: a Stable Diffusion test update vs. an Unconditional Latent Diffusion docs/pipeline item." }, { "left": "pull_request:1788", "right": "pull_request:533", "accept": false, - "reason": "Same file, but different changes: prediction_type='sample' handling bugfix vs reverting type hints in DDIM." + "reason": "Same file, but different changes: a real DDIM scheduler bug fix vs. a revert of type hints. Not the same underlying issue." }, { "left": "pull_request:191", "right": "pull_request:421", "accept": false, - "reason": "Import-fix PR for SciPy/transformers availability is unrelated to a scheduler docs typo cleanup." + "reason": "Import fix for scipy/dummy objects vs. scheduler documentation cleanup; unrelated." }, { "left": "pull_request:192", "right": "pull_request:888", "accept": false, - "reason": "README parameter alignment for Stable Diffusion is unrelated to adding the CycleDiffusion pipeline." + "reason": "Stable Diffusion README parameter tweak vs. adding CycleDiffusion pipeline; different scope and problem." }, { "left": "pull_request:1920", "right": "pull_request:2489", "accept": false, - "reason": "Different docs work: translating docs vs adding Core ML documentation." + "reason": "Both are docs additions, but they cover different topics and files with no shared bug or change." }, { "left": "pull_request:1932", "right": "pull_request:3215", "accept": false, - "reason": "One fixes DDIMPipeline scheduler conversion; the other adds 8-bit/offload handling in pipeline_utils." + "reason": "DDIM loading behavior fix vs. 8-bit loading checks in pipeline_utils; unrelated code paths." }, { "left": "pull_request:1951", "right": "pull_request:2839", "accept": false, - "reason": "ColossalAI dependency/API update is unrelated to updating the examples README table." + "reason": "ColossalAI example update vs. examples README update; related only at a high level, not the same change." }, { "left": "pull_request:1960", "right": "pull_request:879", "accept": false, - "reason": "Different README updates in different sections; no shared bug or code path." + "reason": "Different README/community docs edits with no shared bug or concrete fix." }, { "left": "pull_request:1995", "right": "pull_request:3063", "accept": false, - "reason": "Img2img strength validation typo is unrelated to __setattr__ config handling for None values." + "reason": "Different pipeline issues: strength range validation typo vs. __setattr__ None-handling bug." }, { "left": "pull_request:1995", "right": "pull_request:3099", "accept": false, - "reason": "Different issues: strength range message typo vs attend-and-excite default attn_res." + "reason": "Separate stable diffusion pipeline changes; one is a validation-message fix, the other changes a default attention resolution." }, { "left": "pull_request:1996", "right": "pull_request:2733", "accept": false, - "reason": "Different conversion features: safetensors output vs fp16 support for original-ckpt conversion." + "reason": "Both touch conversion scripts, but one adds safetensors output and the other adds fp16 support for a different conversion direction." }, { "left": "pull_request:2006", "right": "pull_request:879", "accept": false, - "reason": "Checkpoint merger bugfix is unrelated to README table edits." + "reason": "Community merger logic fix vs. README table formatting; unrelated." }, { "left": "pull_request:2009", "right": "pull_request:3439", "accept": false, - "reason": "Both involve textual inversion, but one adds embeddings loading to pipelines and the other extends the loader API to accept state dicts; distinct changes." + "reason": "Both involve textual inversion, but one adds loading embeddings to pipelines and the other enables state-dict input in the loader; different features." }, { "left": "pull_request:2022", "right": "pull_request:912", "accept": false, - "reason": "k-diffusion version gating and package pinning are related only by dependency management, not the same bug/change." + "reason": "Dependency/version gating for k-diffusion vs. a project version bump; not the same issue." }, { "left": "pull_request:2040", "right": "pull_request:2393", "accept": false, - "reason": "Two different pipeline features: InstructPix2Pix vs MultiDiffusion Panorama." + "reason": "Both add major pipelines, but they are different pipelines with different implementations and docs." }, { "left": "pull_request:2085", "right": "pull_request:2433", "accept": false, - "reason": "Separate docs updates for DreamBooth cropping and SDEdit notes in img2img." + "reason": "Separate docs tweaks for different training/pipeline topics." }, { "left": "pull_request:2101", "right": "pull_request:3863", "accept": false, - "reason": "Pix2Pix example docs/code update is unrelated to Habana optimization docs." + "reason": "Pix2Pix example update vs. Habana optimization docs; unrelated." }, { "left": "pull_request:2118", "right": "pull_request:3941", "accept": false, - "reason": "Typo fix in import_utils is unrelated to loading-variants documentation." + "reason": "Import spelling fix vs. docs/code address fix; unrelated." }, { "left": "pull_request:2118", "right": "pull_request:869", "accept": false, - "reason": "Import-utils naming typo fix and community README/import bug are different changes." + "reason": "Both touch imports at a broad level, but they fix different modules and problems." }, { "left": "pull_request:2138", "right": "pull_request:2404", "accept": false, - "reason": "Docs typo cleanup is unrelated to fixing the AttnProcessor2_0 symbol in code." + "reason": "Different typo fixes in unrelated docs/code paths." }, { "left": "pull_request:2138", "right": "pull_request:3800", "accept": false, - "reason": "Different typo fixes in unrelated docs/code areas." + "reason": "Docs typo vs. panorama code typo; not the same artifact." }, { "left": "pull_request:2151", "right": "pull_request:260", "accept": false, - "reason": "EMAModel load_state_dict fix is unrelated to a README spelling fix." + "reason": "EMAModel load_state_dict bug fix vs. README typo; unrelated." }, { "left": "pull_request:2151", "right": "pull_request:353", "accept": false, - "reason": "Different typo bugs in unrelated modules." + "reason": "Both are small fixes, but they affect different code paths and error conditions." }, { "left": "pull_request:2151", "right": "pull_request:407", "accept": false, - "reason": "EMAModel state handling is unrelated to attention-slicing API typo." + "reason": "Different fixes in different modules: EMAModel state loading vs. attention slicing behavior." }, { "left": "pull_request:2151", "right": "pull_request:798", "accept": false, - "reason": "Different docstring/logic typo fixes in separate modules." + "reason": "Different typos in unrelated modules; no shared underlying bug." }, { "left": "pull_request:2151", "right": "pull_request:972", "accept": false, - "reason": "EMAModel load_state_dict bug is unrelated to the torch_dtype README typo." + "reason": "EMAModel bug fix vs. README parameter typo; unrelated." }, { "left": "pull_request:2181", "right": "pull_request:2788", "accept": false, - "reason": "Flax scheduler docs addition is unrelated to deduplicating the training docs TOC." + "reason": "Docs-only changes in different sections: Flax scheduler usage vs. training TOC deduplication." }, { "left": "pull_request:2181", "right": "pull_request:580", "accept": false, - "reason": "Docs-only Flax scheduler section vs an early Flax scheduler test implementation; not the same change." + "reason": "Docs on Flax scheduler usage vs. an experimental Flax scheduler test PR; not the same change." } ] }, @@ -143530,7 +143610,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 123, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -143621,10 +143701,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -143639,180 +143719,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: it mixes unrelated docs, example, test, and pipeline PRs. The soft-edge pairs are only superficially similar (same file, same broad area, or typo/docs tweaks), so all should be rejected.", - "confidence": 0.95, + "summary": "These PRs are only superficially similar (mostly typos, docs, or test tweaks) and do not form a true duplicate cluster around one underlying bug/change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No canonical PR: the cluster mixes unrelated documentation, test, and code changes with no single underlying fix or feature.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "No single PR is a strong representative because the items are heterogeneous and not mergeable into one coherent change.", "soft_edge_verdicts": [ { "left": "pull_request:2197", "right": "pull_request:3839", "accept": false, - "reason": "Different bugs: one fixes an inpaint pipeline typo/argument naming issue, the other changes SDE-VE noise handling behavior." + "reason": "Different bugs in different areas: inpaint pipeline argument typo vs SDE VE noise handling." }, { "left": "pull_request:2209", "right": "pull_request:4317", "accept": false, - "reason": "Both touch docs/example wording, but they fix unrelated problems in different examples and code paths." + "reason": "Docs argument-name fix in img2img example vs DreamBooth validation option fix; unrelated." }, { "left": "pull_request:2257", "right": "pull_request:874", "accept": false, - "reason": "Shared README file only; these add different community pipeline entries and unrelated README content." + "reason": "Both touch community docs, but one adds UnCLIP interpolation and the other adds a generic inference example." }, { "left": "pull_request:2278", "right": "pull_request:309", "accept": false, - "reason": "Completely different changes: a large video-stable-diffusion feature PR vs a link fix in README." + "reason": "Video Stable Diffusion pipeline addition vs README link fix; not the same change." }, { "left": "pull_request:2278", "right": "pull_request:386", "accept": false, - "reason": "Unrelated scopes: new video/community pipeline work vs stable diffusion docs/pipeline introduction." + "reason": "Video Stable Diffusion community pipeline vs stable diffusion API docs; unrelated." }, { "left": "pull_request:2287", "right": "pull_request:2629", "accept": false, - "reason": "Different training-script logging/argument changes versus SAG pipeline support for non-square images." + "reason": "W&B logging for unconditional image generation vs SAG non-square support; different code paths and goals." }, { "left": "pull_request:2290", "right": "pull_request:642", "accept": false, - "reason": "Both are docs-related typos/example fixes, but for different pipelines and unrelated code paths." + "reason": "Pix2Pix docs indentation fix vs a docs link update for an unrelated training notebook." }, { "left": "pull_request:2320", "right": "pull_request:4546", "accept": false, - "reason": "Both are typo fixes, but in different files and unrelated conversion/script contexts." + "reason": "Different typo fixes in different conversion/script files; not the same underlying issue." }, { "left": "pull_request:2320", "right": "pull_request:948", "accept": false, - "reason": "Different typos in different pipelines; no shared bug or merged code path." + "reason": "Checkpoint conversion filename typo vs a DDPM comment typo; unrelated." }, { "left": "pull_request:2333", "right": "pull_request:3680", "accept": false, - "reason": "Different documentation fixes in different markdown pages." + "reason": "Ethical guidelines formatting cleanup vs loader-method link fix; unrelated docs changes." }, { "left": "pull_request:2362", "right": "pull_request:2766", "accept": false, - "reason": "Both are test/MPS-related, but they affect different test helpers and distinct test files; not the same fix." + "reason": "Both concern MPS/test behavior, but one adds a skip decorator while the other changes expected slices for torch 2.0." }, { "left": "pull_request:2362", "right": "pull_request:796", "accept": false, - "reason": "Both mention MPS/testing, but one adds a skip decorator and the other adds Apple M1 tests; separate changes." + "reason": "MPS skip decorator vs Apple M1 test setup; different fixes." }, { "left": "pull_request:2367", "right": "pull_request:2557", "accept": false, - "reason": "Shared dreambooth training file only; one adds checkpoint-limit plumbing, the other adds intermediate logging." + "reason": "Both DreamBooth training scripts, but one adds checkpoint limits and the other adds intermediate logging." }, { "left": "pull_request:2383", "right": "pull_request:889", "accept": false, - "reason": "Different features: pix2pix-zero helper method vs generic pipeline components API." + "reason": "Pix2Pix Zero edit-vector utility vs DiffusionPipeline components support; unrelated." }, { "left": "pull_request:2395", "right": "pull_request:2762", "accept": false, - "reason": "Same broad Pix2Pix Zero area, but one adds embedding-generation docs and the other documents multi-controlnet conditioning; unrelated changes." + "reason": "Pix2Pix Zero docs on embeddings vs ControlNet multi-conditioning docs; unrelated features." }, { "left": "pull_request:2396", "right": "pull_request:2530", "accept": false, - "reason": "One fixes scheduler test warnings, the other adds EMAModel tests; no shared underlying bug." + "reason": "Scheduler test warning cleanup vs EMAModel test suite; unrelated." }, { "left": "pull_request:2396", "right": "pull_request:934", "accept": false, - "reason": "Both are test adjustments, but for different failures and different model paths." + "reason": "UniPC scheduler test fixes vs MPS failing tests; different scopes." }, { "left": "pull_request:2398", "right": "pull_request:3473", "accept": false, - "reason": "Both are Pix2Pix Zero docs, but one adds inversion examples and the other adds textual inversion inference docs; different topics." + "reason": "Pix2Pix Zero inversion docs vs textual inversion inference docs; same broad topic, different pipelines and content." }, { "left": "pull_request:2398", "right": "pull_request:3852", "accept": false, - "reason": "Same docs area only; one is a pipeline guide, the other is a contributor note in a separate page." + "reason": "Pix2Pix Zero docs vs a contributor note in paradigms docs; unrelated." }, { "left": "pull_request:2400", "right": "pull_request:869", "accept": false, - "reason": "Shared community README file, but they add different community pipeline entries/content and fix different things." + "reason": "Both touch community README, but one adds image interpolation and the other fixes README/import behavior." }, { "left": "pull_request:2433", "right": "pull_request:3388", "accept": false, - "reason": "Different docs pages: one adds SDEdit context to img2img docs, the other changes installation instructions." + "reason": "SDEdit note in img2img docs vs installation docs update; unrelated." }, { "left": "pull_request:2446", "right": "pull_request:3971", "accept": false, - "reason": "Unrelated docs fixes in different pipelines and files." + "reason": "Pix2Pix docs snippet fix vs SD XL docs/workflow fixes; unrelated." }, { "left": "pull_request:2487", "right": "pull_request:3549", "accept": false, - "reason": "Both are docs, but one adds a JAX/Flax guide and the other updates evaluation docs; not the same change." + "reason": "Stable Diffusion JAX/Flax guide vs evaluation docs update; different documentation topics." }, { "left": "pull_request:2487", "right": "pull_request:3894", "accept": false, - "reason": "Different docs versus model code change; no shared bug or feature." + "reason": "Docs guide vs Flax dropout implementation change; not the same underlying change." }, { "left": "pull_request:2487", "right": "pull_request:4766", "accept": false, - "reason": "Only share the docs tree; one adds a JAX/Flax guide while the other is a draft Pix2Pix Zero guide with code and API updates." + "reason": "Both docs, but one is a JAX/Flax how-to and the other is a Pix2Pix Zero guide; unrelated." }, { "left": "pull_request:2498", "right": "pull_request:3079", "accept": false, - "reason": "Same test file, but one addresses timeout behavior and the other speeds up fast tests by changing model setup/inference steps." + "reason": "Both touch attend-and-excite tests, but one addresses timeout by shrinking batch coverage while the other speeds up fast tests by changing model/input settings." }, { "left": "pull_request:2503", "right": "pull_request:2766", "accept": false, - "reason": "Different targets: torch 2.0 docs/benchmark wording versus MPS test expectations." + "reason": "PyTorch 2.0 benchmark/docs change vs MPS test expectation fix; unrelated." }, { "left": "pull_request:2533", "right": "pull_request:3168", "accept": false, - "reason": "Unrelated WIP testing changes: one alters workflow/dependency pinning, the other changes docs text for a doc-builder test." + "reason": "RC1 testing workflow/package pinning vs doc-builder testing WIP; unrelated." } ] }, @@ -143828,7 +143908,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 124, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -143923,10 +144003,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -143941,210 +144021,210 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These PRs are mostly unrelated small docs/test/code fixes that only share broad keywords like typos, docs, or warnings. No soft-edge pair looks like the same concrete bug or mergeable change.", - "confidence": 0.97, + "summary": "No true duplicate cluster emerges here: the soft pairs are mostly unrelated docs/typo/test/code changes across different files and bugs, so all candidate edges should be rejected.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 2536 is the strongest representative: it is a substantive code change with broad impact, higher discussion/review activity, and a clear technical purpose compared with the many tiny typo/docs fixes in the cluster.", + "canonical_pr_reason": "No single PR is a valid canonical duplicate target; the items are heterogeneous and do not resolve to one underlying change. If forced to pick a standalone representative, PR #2845 is the clearest concrete bug fix, but it is not a duplicate hub.", "best_issue_reason": null, - "best_pr_reason": "PR 2536 is the best single PR anchor for this cluster because it is the most substantial and review-heavy change, making it a more stable representative than the many one-line documentation edits.", + "best_pr_reason": "PR #2845 is the strongest standalone PR in the set: it has an explicit issue target, a concrete code-path fix, and a focused behavioral change rather than a minor doc/test typo.", "soft_edge_verdicts": [ { "left": "pull_request:2536", "right": "pull_request:3789", "accept": false, - "reason": "Both mention gradient checkpointing, but one enables it for VAE and the other adds 3D U-Net support; different model paths and different fixes." + "reason": "Different features: VAE gradient checkpointing vs 3D U-Net gradient checkpointing." }, { "left": "pull_request:2539", "right": "pull_request:3941", "accept": false, - "reason": "Unrelated documentation edits: one fixes a default value comment, the other fixes loading docs/import text." + "reason": "Unrelated doc tweaks in different pipelines; one changes a default doc value, the other adds import guidance." }, { "left": "pull_request:2540", "right": "pull_request:2677", "accept": false, - "reason": "Both touch ControlNet docs, but one is a wording/section correction and the other adds a new training docs entry and page." + "reason": "Different documentation sections: ControlNet API docs vs training docs entry." }, { "left": "pull_request:2543", "right": "pull_request:651", "accept": false, - "reason": "Both are warnings-related, but one is about model CPU offload messaging and the other is a scheduler deprecation warning; different code paths." + "reason": "Both mention warnings, but they target different subsystems and different deprecation issues." }, { "left": "pull_request:2559", "right": "pull_request:2677", "accept": false, - "reason": "Same broad docs area, but one fixes a ControlNet docs format link and the other adds ControlNet training documentation." + "reason": "Different docs pages and different edits; no shared underlying bug/change." }, { "left": "pull_request:2559", "right": "pull_request:421", "accept": false, - "reason": "Both are docs tweaks, but they concern different pages and unrelated content changes." + "reason": "Both are docs, but about different topics and files; not the same change." }, { "left": "pull_request:2559", "right": "pull_request:4275", "accept": false, - "reason": "Unrelated docs work: ControlNet docs formatting versus AutoPipeline docs corrections." + "reason": "Different docs fixes; no concrete overlap beyond being documentation edits." }, { "left": "pull_request:2559", "right": "pull_request:4721", "accept": false, - "reason": "Different documentation targets and different language/localization changes; not the same change." + "reason": "Different language/docs updates with no shared bug or text change." }, { "left": "pull_request:2563", "right": "pull_request:4145", "accept": false, - "reason": "One removes an empty-cache call in EMA; the other is a draft UnCLIP pipeline demo/change and not the same bug." + "reason": "EMA cache cleanup is unrelated to the UnCLIP pipeline placeholder change." }, { "left": "pull_request:2566", "right": "pull_request:4546", "accept": false, - "reason": "Both are typo fixes, but in different files and unrelated features: tensorboard logging vs ControlNet script text." + "reason": "Separate typo fixes in unrelated files and code paths." }, { "left": "pull_request:2566", "right": "pull_request:948", "accept": false, - "reason": "One fixes a tensorboard tracker call in training; the other is a comment typo in DDPM pipeline code." + "reason": "Training tensorboard tracker typo vs a pipeline comment typo; not the same fix." }, { "left": "pull_request:2598", "right": "pull_request:4022", "accept": false, - "reason": "Paint-by-example docs versus Korean translation updates; different files, scope, and content." + "reason": "Both are docs, but one is PaintByExample docs and the other is Korean translation work." }, { "left": "pull_request:2598", "right": "pull_request:4203", "accept": false, - "reason": "Different docs pages and unrelated typos; not the same underlying change." + "reason": "Different documentation targets and content changes; not duplicates." }, { "left": "pull_request:2602", "right": "pull_request:641", "accept": false, - "reason": "Both test huggingface-hub pins, but they pin different versions in different releases and are not the same concrete change." + "reason": "Both pin huggingface-hub in setup.py, but to different versions for different test releases; not the same change." }, { "left": "pull_request:2637", "right": "pull_request:3163", "accept": false, - "reason": "Both are docs edits, but one is a quicktour typo fix and the other is author credit for custom diffusion; unrelated." + "reason": "Quicktour wording fix vs custom diffusion attribution/update; unrelated docs edits." }, { "left": "pull_request:2637", "right": "pull_request:3961", "accept": false, - "reason": "Unrelated docs punctuation/wording edits on different pages." + "reason": "Different doc pages and different text corrections." }, { "left": "pull_request:2639", "right": "pull_request:2748", "accept": false, - "reason": "One adds an import in a training tutorial, the other fixes a wording typo in torch 2.0 benchmarking docs." + "reason": "Basic training docs vs torch 2.0 optimization docs; no shared underlying issue." }, { "left": "pull_request:2642", "right": "pull_request:3163", "accept": false, - "reason": "Different docs fixes in different guides; no shared bug or change." + "reason": "Loading docs wording fix vs custom diffusion attribution/code update; unrelated." }, { "left": "pull_request:2642", "right": "pull_request:3961", "accept": false, - "reason": "Different docs pages and unrelated edits." + "reason": "Different docs topics and edits." }, { "left": "pull_request:2647", "right": "pull_request:3163", "accept": false, - "reason": "Scheduler docs wording versus custom diffusion author credit; unrelated." + "reason": "Schedulers docs wording fix vs custom diffusion attribution/update; no duplicate basis." }, { "left": "pull_request:2647", "right": "pull_request:3961", "accept": false, - "reason": "Different docs pages and different content changes." + "reason": "Different docs pages and textual changes only." }, { "left": "pull_request:2677", "right": "pull_request:2958", "accept": false, - "reason": "One adds training docs for ControlNet, the other adds a ControlNet inpaint pipeline; related topic but different code/documentation changes." + "reason": "Training docs expansion vs adding a controlnet inpaint pipeline; different artifacts and goals." }, { "left": "pull_request:2678", "right": "pull_request:3942", "accept": false, - "reason": "Both touch tests, but one speeds up EMA serialization tests and the other fixes numpy test handling in pipeline comparison code." + "reason": "Both touch tests, but one changes EMA serialization setup and the other fixes xformers NP test handling." }, { "left": "pull_request:2684", "right": "pull_request:3163", "accept": false, - "reason": "A punctuation fix in custom pipeline overview is unrelated to author-credit changes in a different training doc." + "reason": "Custom pipeline overview wording vs custom diffusion credit/update; unrelated." }, { "left": "pull_request:2684", "right": "pull_request:3961", "accept": false, - "reason": "Different docs pages and unrelated edits." + "reason": "Different docs pages with no shared bug or edit." }, { "left": "pull_request:2685", "right": "pull_request:3961", "accept": false, - "reason": "Kerascv docs typo versus consistency models contributor formatting; not the same change." + "reason": "KerasCV docs punctuation fix vs contributor credit change; unrelated." }, { "left": "pull_request:2693", "right": "pull_request:2836", "accept": false, - "reason": "Both are docs fixes, but for different guides and different links; not the same underlying issue." + "reason": "Inpaint image-link fix vs DreamBooth LoRA link fix; same broad docs area but different issues." }, { "left": "pull_request:2766", "right": "pull_request:3923", "accept": false, - "reason": "One updates MPS test expectations; the other adds an import in stable diffusion docs." + "reason": "Both add/import torch in docs/tests contexts, but they fix different files and failures." }, { "left": "pull_request:2767", "right": "pull_request:3581", "accept": false, - "reason": "LoRA training script argument rename versus a community example import cleanup; different code paths and purposes." + "reason": "LoRA training args refactor vs bit_diffusion import cleanup; no shared code-path problem." }, { "left": "pull_request:2812", "right": "pull_request:4022", "accept": false, - "reason": "Stable unCLIP docs/code work is unrelated to a Korean translation update." + "reason": "Stable unCLIP docs update vs Korean translation update; not the same change." }, { "left": "pull_request:2817", "right": "pull_request:3710", "accept": false, - "reason": "UNet1DModel docs and image processor docs are separate documentation changes with different subject matter." + "reason": "UNet1DModel docs for extra_in_channels vs image processor docs entry; unrelated." }, { "left": "pull_request:2845", "right": "pull_request:4853", "accept": false, - "reason": "One fixes None handling in StableUnCLIP image embeddings, the other fixes width/height handling in the image processor; different bugs." + "reason": "StableUnCLIP image_embeds handling vs image_processor width bug; completely different code paths." }, { "left": "pull_request:2861", "right": "pull_request:3847", "accept": false, - "reason": "A pipeline behavior test and a dependency-check workflow are unrelated test additions." + "reason": "StableUnCLIP test for image_embeds=None vs dependency-check workflow/tests; unrelated." } ] }, @@ -144160,7 +144240,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 125, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -144252,10 +144332,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -144270,186 +144350,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab-bag of mostly unrelated docs typo fixes, documentation additions, and a few distinct code changes. None of the soft pairs look like the same underlying bug or change, so I would not merge any of them as duplicates.", + "summary": "This cluster is mostly a grab bag of unrelated docs typos, link fixes, and a few separate code changes across different subsystems. None of the soft pairs look like the same underlying bug or change, so all should be rejected as duplicates.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 4796 is the best representative if one must be chosen: it is a substantive, concrete bug fix with an explicit issue target and broader code impact than the many one-line doc/typo edits.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 4796 is the strongest canonical representative overall: it fixes a specific SDXL pipeline type-mismatch issue, touches multiple pipelines, and is clearly more substantive than the surrounding documentation-only PRs.", + "best_pr_reason": "PR 4796 is the most substantial and concrete fix in the set, but it is still not representative of the others; the cluster is not a true duplicate family.", "soft_edge_verdicts": [ { "left": "pull_request:2869", "right": "pull_request:3165", "accept": false, - "reason": "Both are tiny docs edits, but they touch different pages and different topics; not the same change." + "reason": "Both are tiny docs edits, but they touch different pages and unrelated wording; not the same change." }, { "left": "pull_request:2897", "right": "pull_request:3690", "accept": false, - "reason": "StableUnCLIP docs example vs Kandinsky pipeline example fix; different pipelines and unrelated content." + "reason": "Stable UnCLIP docs/example addition and a Kandinsky pipeline example fix are unrelated changes." }, { "left": "pull_request:2911", "right": "pull_request:3165", "accept": false, - "reason": "Independent punctuation/docs cleanup in different docs files; no shared bug or change." + "reason": "Different documentation pages and different typo fixes; no shared underlying bug." }, { "left": "pull_request:2912", "right": "pull_request:3165", "accept": false, - "reason": "ControlNet docs punctuation fix vs custom diffusion docs spacing/additions; unrelated edits." + "reason": "ControlNet docs punctuation fix vs a custom diffusion docs edit; unrelated content." }, { "left": "pull_request:2912", "right": "pull_request:4564", "accept": false, - "reason": "One is a docs anchor punctuation tweak, the other updates SDXL ControlNet repo/model names and code examples; different problems." + "reason": "A docs punctuation fix is not the same as updating SDXL ControlNet model/repo names." }, { "left": "pull_request:2958", "right": "pull_request:3121", "accept": false, - "reason": "ControlNet inpaint export/addition vs controlnet global pooling change; separate code paths and features." + "reason": "ControlNet inpaint pipeline addition and global pooling in ControlNet model are separate code changes." }, { "left": "pull_request:2974", "right": "pull_request:897", "accept": false, - "reason": "Both touch community README, but one adds a TensorRT pipeline and the other adds a speech-to-image example; distinct features." + "reason": "Community README/TensorRT pipeline work is not the same as adding a speech-to-image example." }, { "left": "pull_request:3023", "right": "pull_request:3691", "accept": false, - "reason": "Different attention/LoRA-related internals; one is Torch 2.0 attention plumbing, the other is text-encoder LoRA processor selection." + "reason": "Both relate to attention/torch 2.0, but they fix different code paths and different bugs." }, { "left": "pull_request:3063", "right": "pull_request:848", "accept": false, - "reason": "Pipeline __setattr__ bug fix vs Flax width/height latent shape fix; unrelated code paths." + "reason": "Pipeline __setattr__ None-handling and Flax width/height handling are unrelated bugs." }, { "left": "pull_request:3118", "right": "pull_request:4333", "accept": false, - "reason": "ControlNet default-argument compatibility fix vs DreamBooth validation scheduler change; different issues." + "reason": "ControlNet default-arg fix and DreamBooth validation-scheduler change are different features/issues." }, { "left": "pull_request:3118", "right": "pull_request:948", "accept": false, - "reason": "Different pipeline fixes in unrelated files; no shared concrete bug." + "reason": "A ControlNet API fix and a DDPM comment typo are clearly unrelated." }, { "left": "pull_request:3123", "right": "pull_request:3132", "accept": false, - "reason": "A textual inversion JAX typo fix and a ScoreSdeVe typing fix are unrelated." + "reason": "Textual inversion doc typo and ScoreSdeVe type hint fix are unrelated." }, { "left": "pull_request:3132", "right": "pull_request:4546", "accept": false, - "reason": "Different typos in different code areas; no common underlying defect." + "reason": "ScoreSdeVe type hints and a ControlNet converter docstring typo do not share a concrete bug." }, { "left": "pull_request:3132", "right": "pull_request:948", "accept": false, - "reason": "ScoreSdeVe constructor type annotation vs DDPM comment typo; not the same change." + "reason": "Different files, different subsystems, and no common fix." }, { "left": "pull_request:3215", "right": "pull_request:4696", "accept": false, - "reason": "8-bit/offload handling vs T2I-Adapter-XL feature work; different technical problems." + "reason": "8-bit loading checks in pipeline_utils and T2I-Adapter-XL model additions are unrelated." }, { "left": "pull_request:3226", "right": "pull_request:3542", "accept": false, - "reason": "ControlNet v1.1 docs cleanup vs ControlNet LoRA support code changes; related subsystem but different concrete changes." + "reason": "ControlNet v1.1 docs and ControlNet LoRA support are different changes, not one bug fix." }, { "left": "pull_request:3262", "right": "pull_request:4564", "accept": false, - "reason": "IF name/model-card update vs SDXL ControlNet repo-name update; different model families and edits." + "reason": "Updating IF naming to XL and updating the SDXL ControlNet repo name are separate documentation/model-name changes." }, { "left": "pull_request:3266", "right": "pull_request:4796", "accept": false, - "reason": "Community pipeline version-loading fix vs SDXL pipeline type-mismatch fix; separate bugs." + "reason": "Community pipeline version loading and SDXL type-mismatch fixes are different bugs in different code paths." }, { "left": "pull_request:3271", "right": "pull_request:4934", "accept": false, - "reason": "Docs links for training scripts vs adding a new T2I-Adapter training script; not the same change." + "reason": "Linking training docs is not the same as adding the T2I-Adapter training script and docs." }, { "left": "pull_request:3334", "right": "pull_request:4910", "accept": false, - "reason": "Docstring cleanup in pipeline utils vs inpainting example typo fix; unrelated." + "reason": "Docstring cleanup and an inpainting example variable fix are unrelated." }, { "left": "pull_request:3345", "right": "pull_request:4531", "accept": false, - "reason": "README pipeline class correction vs SD reference pipeline indexing fix; different files and behavior." + "reason": "README pipeline-class correction and reference-pipeline indexing fix are different issues." }, { "left": "pull_request:3370", "right": "pull_request:4564", "accept": false, - "reason": "PyTorch 2.0 docs update vs ControlNet SDXL repo-name docs/code update; distinct documentation topics." + "reason": "PyTorch 2.0 docs updates and SDXL ControlNet repo-name updates are unrelated documentation changes." }, { "left": "pull_request:3388", "right": "pull_request:4227", "accept": false, - "reason": "Installation docs update vs textual inversion README import fix; unrelated docs changes." + "reason": "Installation docs update and a missing import in textual inversion are different fixes." }, { "left": "pull_request:3398", "right": "pull_request:5323", "accept": false, - "reason": "Stable Diffusion typo in a tutorial vs adapter docs link correction; different documentation targets." + "reason": "Stable Diffusion typo fix and adapter-doc link fix are unrelated docs edits." }, { "left": "pull_request:3417", "right": "pull_request:5671", "accept": false, - "reason": "unCLIP scheduler note vs dependency-test workflow debug logging; unrelated changes." + "reason": "UnCLIP scheduler note and a dependency test workflow tweak are not the same change." }, { "left": "pull_request:3525", "right": "pull_request:4258", "accept": false, - "reason": "README marketing/example edits vs SDXL DreamBooth note rewrite; different docs and purpose." + "reason": "README marketing/content changes and an SDXL hyperparameter note are unrelated." }, { "left": "pull_request:3549", "right": "pull_request:5844", "accept": false, - "reason": "Evaluation docs addition about PartiPrompts vs Japanese SDXL reference example; unrelated conceptual docs." + "reason": "Evaluation-doc prose update and a Japanese SDXL reference addition are different docs changes." }, { "left": "pull_request:3563", "right": "pull_request:3690", "accept": false, - "reason": "VQ pipeline-related block fix vs Kandinsky pipeline example fix; different code paths." + "reason": "A VQ pipeline block fix and a Kandinsky example fix are unrelated code paths." }, { "left": "pull_request:3568", "right": "pull_request:4626", "accept": false, - "reason": "Broken tutorial links fix vs text2image docs link updates; similar theme but distinct pages and edits." + "reason": "Broken internal links in tutorials and text2image training-link fixes are not the same underlying issue." } ] }, @@ -144465,7 +144545,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 126, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -144553,10 +144633,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -144571,168 +144651,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: mostly unrelated docs cleanups, small bug fixes, and feature PRs. None of the suggested soft pairs appear to be true duplicates or mergeable into a single PR. If a representative PR is needed, PR 4099 is the strongest concrete bug-fix candidate.", - "confidence": 0.18, + "summary": "This cluster is a mixed bag of unrelated docs, workflow, test, and code-fix PRs. The soft pairs mostly match on broad keywords like \"docs\", \"fix\", or the same subsystem, but they do not describe the same underlying bug or change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 4099 is the closest thing to a canonical target: it fixes a concrete SDXL prompt-embedding bug across several related pipelines and has an explicit issue reference.", + "canonical_pr_reason": "No single canonical PR fits: the set spans many independent fixes across different files and subsystems, so there is no shared underlying change to canonicalize.", "best_issue_reason": null, - "best_pr_reason": "PR 4099 is the best representative overall because it is a substantive code-path fix rather than docs-only churn, and it impacts a cohesive family of SDXL pipelines.", + "best_pr_reason": "PR 4099 is the strongest standalone bug-fix PR in the set because it has an explicit issue target, a concrete code-path correction, and test updates; however, it is not a duplicate of the other items.", "soft_edge_verdicts": [ { "left": "pull_request:3577", "right": "pull_request:5411", "accept": false, - "reason": "Both are documentation-related, but they fix different problems: broken README links vs a broken training-doc link and related pipeline docs. Not the same change." + "reason": "Both touch docs, but 3577 fixes broken README links while 5411 fixes a broken textual inversion training link and related code/docs; different content and purpose." }, { "left": "pull_request:3592", "right": "pull_request:5065", "accept": false, - "reason": "Both touch attention processor code, but one removes debug prints while the other fixes a typo/logic issue in a broader attention-processor-related patch. Different bugs, not one mergeable fix." + "reason": "3592 removes debug prints in attention_processor.py; 5065 is a broad LoRA/attention-processor typo fix with many docs/tests/file changes. Same area, not the same bug." }, { "left": "pull_request:3625", "right": "pull_request:4740", "accept": false, - "reason": "These are unrelated workflow changes: one is a doc-builder test tweak, the other is a safetensors dependency pin sanity check. No shared underlying bug." + "reason": "A doc-builder workflow test PR and a safetensors dependency pin are unrelated changes with no shared code path." }, { "left": "pull_request:3685", "right": "pull_request:5911", "accept": false, - "reason": "Different domains and fixes: StableDiffusion thread inference support versus IP-Adapter face-model support. Not the same code-path problem." + "reason": "3685 adds thread inference support to StableDiffusionPipeline; 5911 adds IP-Adapter Full support. Different features and code paths." }, { "left": "pull_request:3710", "right": "pull_request:4285", "accept": false, - "reason": "One adds image-processor docs, the other adds OpenVINO/ONNX runtime docs. Same broad docs area, but not the same artifact or issue." + "reason": "Both are documentation PRs, but they add different pages for different features (image processor vs ONNX/OpenVINO docs)." }, { "left": "pull_request:3755", "right": "pull_request:3942", "accept": false, - "reason": "Both are test-related, but one relaxes a flaky tolerance and the other adjusts xformers-related NumPy test handling. Different failures." + "reason": "Both adjust tests, but they target different test files and different failures; not the same failing test or bug." }, { "left": "pull_request:3786", "right": "pull_request:3994", "accept": false, - "reason": "A PR template addition is unrelated to adding an act_fn parameter to a UNet block. No shared bug or change." + "reason": "A PR template addition is unrelated to adding an act_fn parameter to OutValueFunctionBlock." }, { "left": "pull_request:382", "right": "pull_request:910", "accept": false, - "reason": "Both are docs PRs, but they target different pipelines and content eras. No evidence of the same underlying change." + "reason": "These are separate documentation updates for different pipelines and eras; no shared underlying change." }, { "left": "pull_request:3839", "right": "pull_request:4320", "accept": false, - "reason": "One fixes scheduler noise injection behavior, the other is a typo/docs heading cleanup. Completely different." + "reason": "3839 fixes scheduler noise handling in code, while 4320 is a documentation typo fix. Different artifact types and problems." }, { "left": "pull_request:3852", "right": "pull_request:5844", "accept": false, - "reason": "Contributor-note docs for paradigms are unrelated to adding a Japanese SDXL reference example. Same docs area, different content." + "reason": "Both are docs, but one adds a contributor note to paradigms docs and the other adds a Japanese SDXL reference example. Unrelated." }, { "left": "pull_request:3903", "right": "pull_request:5550", "accept": false, - "reason": "Kandinsky2.2 feature implementation versus a doc checkpoint correction. Not the same bug or change." + "reason": "3903 is a large Kandinsky2.2 implementation PR; 5550 only corrects a checkpoint URL in the docs. Not the same change." }, { "left": "pull_request:392", "right": "pull_request:420", "accept": false, - "reason": "Both are early docs reorganizations, but they modify different sections and files for different purposes. Not duplicates." + "reason": "General docs/contribution content versus optimization-section cleanup; different pages and objectives." }, { "left": "pull_request:3941", "right": "pull_request:4741", "accept": false, - "reason": "A docs code-snippet fix for loading guidance is unrelated to fixing dummy imports for AudioLDM2." + "reason": "3941 fixes a broken loading-doc code snippet; 4741 fixes dummy imports for AudioLDM2. Different bugs and files." }, { "left": "pull_request:3941", "right": "pull_request:5745", "accept": false, - "reason": "Both are documentation/code snippet corrections, but one is about loading variants and the other is a consistency-decoder example type fix. Different issues." + "reason": "A docs code-snippet/import fix is unrelated to a consistency-decoder docstring type correction." }, { "left": "pull_request:3962", "right": "pull_request:4437", "accept": false, - "reason": "A checkpoint URL doc correction is unrelated to consistency-model sampling bug fixes in UNet code." + "reason": "3962 changes a doc URL; 4437 changes model sampling behavior for consistency models. Not the same issue." }, { "left": "pull_request:3971", "right": "pull_request:4965", "accept": false, - "reason": "Workflow/docs build changes and W\u00fcrstchen doc formatting are not the same underlying problem." + "reason": "Both are documentation-only, but they update different docs pages for different features; no duplicate change." }, { "left": "pull_request:3979", "right": "pull_request:4267", "accept": false, - "reason": "Both mention SDXL/upscale docs or code, but one fixes attention-processor selection logic and the other only swaps image links in docs. Not the same change." + "reason": "3979 fixes SDXL pipeline code; 4267 only swaps images in SDXL docs. Code bug vs documentation asset update." }, { "left": "pull_request:3979", "right": "pull_request:4729", "accept": false, - "reason": "The upscale pipeline attention-path fix is unrelated to a progress-bar update in a community example." + "reason": "Both are runtime code fixes, but they affect different pipelines and different symptoms; they are not mergeable as one fix." }, { "left": "pull_request:3979", "right": "pull_request:5013", "accept": false, - "reason": "Different SDXL bugs: attention processor handling versus dtype/upcasting in InstructPix2Pix SDXL image latents. Not mergeable as one fix." + "reason": "3979 targets SDXL upscale attention handling, while 5013 fixes SDXL InstructPix2Pix image latent dtype. Same family, different concrete bug." }, { "left": "pull_request:3987", "right": "pull_request:4113", "accept": false, - "reason": "Audio Diffusion doc snippet correction is unrelated to the SDXL inpaint dtype fix." + "reason": "Audio Diffusion doc snippet fix versus SDXL inpaint dtype fix; unrelated." }, { "left": "pull_request:4022", "right": "pull_request:5549", "accept": false, - "reason": "Korean translation update and Portuguese translation bootstrap are different localization efforts, not duplicates." + "reason": "Different translation rollouts (Korean vs initial Portuguese) with separate doc tree/content changes." }, { "left": "pull_request:406", "right": "pull_request:710", "accept": false, - "reason": "A toctree typo fix and a docstring formatting fix are unrelated changes, even though both are documentation-adjacent." + "reason": "406 fixes a toctree typo; 710 fixes docstring formatting for issue #709. No shared bug." }, { "left": "pull_request:4070", "right": "pull_request:5367", "accept": false, - "reason": "Both are SDXL docs, but one fixes doc links and the other fixes pipeline construction in a tutorial. Different concrete issues." + "reason": "4070 makes minor SDXL docs wording/link changes; 5367 fixes SDXL tutorial pipeline creation code. Different artifact and scope." }, { "left": "pull_request:4082", "right": "pull_request:5367", "accept": false, - "reason": "A dependency/version correction for SDXL requirements is unrelated to fixing the SDXL tutorial pipeline constructor." + "reason": "A dependency file tweak for SDXL training is unrelated to the tutorial pipeline creation fix." }, { "left": "pull_request:4099", "right": "pull_request:4397", "accept": false, - "reason": "Both relate to SDXL, but one fixes prompt embedding handling in code while the other only changes a docstring default value. Not the same bug." + "reason": "4099 fixes prompt-embedding handling in SDXL pipelines; 4397 only updates an SDXL docstring default. Different code paths and severity." }, { "left": "pull_request:4099", "right": "pull_request:4886", "accept": false, - "reason": "These are different StableDiffusionUpscale/SDXL bugs: prompt-embedding repetition versus safety-checker preprocessing. Same broad area, not same fix." + "reason": "Both are pipeline bug fixes, but one corrects prompt embedding duplication in SDXL while the other fixes safety-checker preprocessing in UpscalePipeline. Not the same underlying bug." } ] }, @@ -144748,7 +144828,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 127, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -144841,10 +144921,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -144859,192 +144939,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a loose collection of mostly unrelated PRs, many around SDXL/docs. The soft edges are largely superficial topic overlaps (same subsystem or wording) rather than the same concrete fix/change, so almost all should be rejected.", - "confidence": 0.81, + "summary": "This cluster is mostly a loose collection of SDXL-related docs, tests, and feature PRs rather than true duplicates. I found no strong duplicate PR pairs to merge, and the best representative PR is the broad T2I-Adapter-XL feature addition.", + "confidence": 0.86, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #4502 is the strongest representative PR in this set: it is merged, substantial, and a complete SDXL feature addition with broad discussion/review. It is more substantive than the small doc/test fixes and more clearly defines a concrete change than the other candidates.", + "canonical_pr_reason": "PR 4696 is the most central and substantive item: it adds T2I-Adapter-XL support across code, docs, and tests, with extensive review activity and broad impact.", "best_issue_reason": null, - "best_pr_reason": "PR #4502 is the best single PR choice because it is a finished, feature-complete SDXL change with significant scope and review history, making it the most representative artifact among this mixed set.", + "best_pr_reason": "PR 4696 is the strongest canonical representative because it is the largest, most complete SDXL-related change in the set and covers a concrete feature rather than a small docs-only tweak.", "soft_edge_verdicts": [ { "left": "pull_request:4196", "right": "pull_request:5982", "accept": false, - "reason": "Both are SDXL docs, but one is a generic docs improvement and the other is SDXL Turbo documentation; different changes and not the same underlying bug/fix." + "reason": "Both are docs PRs, but one improves SDXL pipeline docs and the other adds SDXL Turbo docs; different artifacts and different change scopes." }, { "left": "pull_request:421", "right": "pull_request:642", "accept": false, - "reason": "Both are docs edits, but they fix unrelated documentation issues in different files and on different topics." + "reason": "Both touch documentation, but they fix unrelated docs pages in different areas of the project." }, { "left": "pull_request:4216", "right": "pull_request:4457", "accept": false, - "reason": "Both touch the TensorRT inpaint example, but they address different changes: one is timestep handling, the other is a separate import/class fix." + "reason": "They both involve the TensorRT inpaint example, but 4216 changes timestep handling/strength behavior while 4457 is a separate minor example/README cleanup." }, { "left": "pull_request:4238", "right": "pull_request:4717", "accept": false, - "reason": "One is SDXL ControlNet inference code, the other is a docs formatting fix in ControlNet training docs; not the same code-path problem." + "reason": "One fixes SDXL ControlNet inference code; the other is a docs formatting fix. Not the same underlying bug or change." }, { "left": "pull_request:4258", "right": "pull_request:5454", "accept": false, - "reason": "These are unrelated docs edits in different areas; the shared theme is only generic documentation cleanup." + "reason": "Both are docs-only changes, but they concern unrelated documentation pages and link updates." }, { "left": "pull_request:4258", "right": "pull_request:6438", "accept": false, - "reason": "A README note rewrite and an installation note update are separate docs changes with no shared underlying issue." + "reason": "Different docs targets: one rewrites SDXL DreamBooth notes, the other updates installation guidance." }, { "left": "pull_request:4267", "right": "pull_request:6647", "accept": false, - "reason": "One fixes doc image links, the other updates an SDXL k-diffusion test expectation; different artifacts and different problems." + "reason": "One fixes an SDXL docs image reference; the other updates an SDXL k-diffusion test baseline. Different kinds of changes." }, { "left": "pull_request:4276", "right": "pull_request:6638", "accept": false, - "reason": "A pipeline import test and a from_single_file config-file test are unrelated test additions." + "reason": "Both are tests, but they validate different behaviors in different code paths and are not the same fix." }, { "left": "pull_request:4285", "right": "pull_request:4696", "accept": false, - "reason": "These are different documentation features: one adds ONNX/OpenVINO SDXL docs, the other adds T2I-Adapter-XL model support." + "reason": "ONNX/OpenVINO documentation work is unrelated to T2I-Adapter-XL model and pipeline implementation." }, { "left": "pull_request:4315", "right": "pull_request:6217", "accept": false, - "reason": "Both are torch-compile related, but they fix different code paths: LoRA conv forwarding versus Stable Video Diffusion VAE compilation handling." + "reason": "Both mention torch compile, but they fix different components and failure modes in different pipelines." }, { "left": "pull_request:4350", "right": "pull_request:4965", "accept": false, - "reason": "A T2I-Adapter docs dtype fix and a Wuerstchen formatting doc fix are unrelated." + "reason": "Both are docs edits, but they target unrelated topics: T2I-Adapter fp16 usage vs Wuerstchen formatting." }, { "left": "pull_request:4350", "right": "pull_request:6205", "accept": false, - "reason": "The adapter dtype docs fix is unrelated to the t2i blog URL correction." + "reason": "Same broad T2I-adapter area, but different docs and different issues: dtype usage vs blog link correction." }, { "left": "pull_request:4351", "right": "pull_request:6669", "accept": false, - "reason": "Both edit README.md, but one changes an image URL and the other updates repository counts/text; not the same change." + "reason": "Both touch README.md, but they change unrelated sections: image URL vs repository stats text." }, { "left": "pull_request:4380", "right": "pull_request:5309", "accept": false, - "reason": "Both are doc cleanup, but they target different pages and different issues." + "reason": "Both are minor docs fixes, but they address different documentation pages and content errors." }, { "left": "pull_request:4380", "right": "pull_request:5369", "accept": false, - "reason": "These are separate minor docs fixes in different guides; no shared concrete bug." + "reason": "These are separate docs cleanups in different guides, not the same underlying change." }, { "left": "pull_request:4499", "right": "pull_request:6197", "accept": false, - "reason": "The reference-attn SDXL PR and DeepCache support are different features; sharing SDXL/community context is not enough." + "reason": "One adds SDXL reference-attn/adain support; the other adds DeepCache support. Different features and code paths." }, { "left": "pull_request:4500", "right": "pull_request:6197", "accept": false, - "reason": "Same as above: unrelated changes despite similar community-pipeline context." + "reason": "Same as above: SDXL reference pipeline work is unrelated to DeepCache support." }, { "left": "pull_request:4501", "right": "pull_request:6197", "accept": false, - "reason": "Same as above: different feature areas and code paths." + "reason": "SDXL reference-attn/adain support and DeepCache support are distinct feature additions." }, { "left": "pull_request:4502", "right": "pull_request:6197", "accept": false, - "reason": "The merged reference-attn/adain SDXL feature is unrelated to DeepCache pipeline support." + "reason": "The merged SDXL reference PR is not the same change as DeepCache support." }, { "left": "pull_request:4582", "right": "pull_request:6647", "accept": false, - "reason": "One fixes a ControlNet SDXL docstring example, the other updates an SDXL k-diffusion test baseline; not the same bug." + "reason": "One corrects a ControlNet SDXL docstring; the other updates an SDXL test expectation. Not the same bug." }, { "left": "pull_request:4612", "right": "pull_request:4965", "accept": false, - "reason": "A controlling-generation link fix and a Wuerstchen formatting fix are unrelated docs changes." + "reason": "They are unrelated docs updates for different pipeline families." }, { "left": "pull_request:4626", "right": "pull_request:5236", "accept": false, - "reason": "Both are docs, but one is a text2image link update and the other is a Korean unconditional-generation formatting fix; no shared underlying issue." + "reason": "Both are link/docs fixes, but in different languages and different pages with no shared concrete change." }, { "left": "pull_request:4626", "right": "pull_request:6440", "accept": false, - "reason": "These are different local-link fixes in different docs sections, not one duplicate change." + "reason": "Both edit links, but they address different docs and different local-link problems." }, { "left": "pull_request:4629", "right": "pull_request:6757", "accept": false, - "reason": "A long-weighted-prompt pipeline PR and a k-diffusion sigma device move are unrelated code changes." + "reason": "Long weighted prompt pipeline docs/code and SDXL k-diffusion sigma placement are unrelated." }, { "left": "pull_request:4652", "right": "pull_request:5013", "accept": false, - "reason": "A LoRA test serialization tweak and an SDXL InstructPix2Pix dtype fix are different problems." + "reason": "One fixes LoRA test serialization; the other fixes SDXL InstructPix2Pix latent dtype handling. Different issues." }, { "left": "pull_request:4652", "right": "pull_request:5104", "accept": false, - "reason": "The LoRA test change and timestep-bias training feature are unrelated." + "reason": "A LoRA test fix and a new timestep-bias training feature are not the same change." }, { "left": "pull_request:4652", "right": "pull_request:6962", "accept": false, - "reason": "LoRA SDXL tests and PixArt slow tests are entirely different test suites and bugs." + "reason": "These are tests for different pipelines and failure modes, so they are not duplicates." }, { "left": "pull_request:4659", "right": "pull_request:6721", "accept": false, - "reason": "Replacing the long-weighted-prompt PR is unrelated to AnimateDiff SDXL support." + "reason": "A weighted-prompt pipeline/docs replacement is unrelated to AnimateDiff SDXL support." }, { "left": "pull_request:4661", "right": "pull_request:6721", "accept": false, - "reason": "One is the merged long-weighted-prompt pipeline PR, the other is an AnimateDiff SDXL pipeline; different features." + "reason": "Same reasoning: weighted-prompt pipeline work is not the same as AnimateDiff SDXL." }, { "left": "pull_request:4666", "right": "pull_request:4712", "accept": false, - "reason": "An SDXL ControlNet LoRA enablement PR and a tests_fetcher utility PR do not fix the same issue." + "reason": "SDXL ControlNet LoRA support is a pipeline feature; tests_fetcher is unrelated infrastructure work." } ] }, @@ -145060,7 +145140,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 128, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -145151,10 +145231,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -145169,180 +145249,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated docs, test, typo, and small bugfix PRs; the soft edges are largely false positives, with no strong duplicate-theme binding the set together.", - "confidence": 0.91, + "summary": "This cluster is not a duplicate set: it mixes unrelated docs tweaks, training-example updates, workflow/config changes, and several distinct code fixes. The only similarity is superficial (small typo/docstring-style edits or shared broad files like `setup.py` / `loaders.py`), not the same underlying bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 4796 is the best anchor: it is a concrete SDXL runtime bug fix with an explicit issue target and touches the actual affected code paths, unlike the many docs/test/typo-only PRs.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 4796 is the strongest representative overall because it addresses a specific code-path defect and has the clearest bug-fix signal; the rest are mostly incidental edits or unrelated maintenance.", + "best_pr_reason": "No single PR cleanly represents the cluster, because the items do not describe one concrete bug or feature. The closest matches are still materially different in intent and code path, so there is no good canonical PR for duplicate triage.", "soft_edge_verdicts": [ { "left": "pull_request:4684", "right": "pull_request:5952", "accept": false, - "reason": "Different docs changes: Korean translation nav rewrite vs English pipeline overview entry." + "reason": "Both are docs-related, but one is a Korean translation sitemap update while the other adds a pipeline list entry for Kandinsky 3; unrelated content and purpose." }, { "left": "pull_request:4696", "right": "pull_request:5207", "accept": false, - "reason": "Both mention T2I-Adapter, but one adds XL model support and code while the other only fixes a README path." + "reason": "Different changes entirely: T2I-Adapter-XL model support vs correcting a filename in the T2I adapter README." }, { "left": "pull_request:4717", "right": "pull_request:5144", "accept": false, - "reason": "Unrelated docs fixes in different areas: ControlNet Tip tag vs BLIP sample-code parenthesis." + "reason": "One fixes a missing `` in ControlNet docs; the other corrects a sample code resize call in BLIP Diffusion docs. Different documents and issues." }, { "left": "pull_request:4740", "right": "pull_request:6018", "accept": false, - "reason": "Both are doc/setup-related, but one pins safetensors while the other is a doc-builder test edit." + "reason": "Both touch setup/docs around packaging, but one pins `safetensors` for a test sanity check while the other is a throwaway doc-builder test edit; not the same change." }, { "left": "pull_request:4748", "right": "pull_request:5996", "accept": false, - "reason": "Same file but different changes: release instructions vs package metadata/license cleanup." + "reason": "Both edit `setup.py`, but one changes dependency/version-checking behavior and the other edits package metadata/cleanup text. Same file, different intent." }, { "left": "pull_request:4796", "right": "pull_request:5303", "accept": false, - "reason": "Different artifacts and problems: SDXL pipeline dtype bug vs DeepFloyd IF docstring comments." + "reason": "Different targets: SDXL pipeline dtype fixes versus DeepFloyd IF docstring cleanup. No shared bug or code path." }, { "left": "pull_request:4797", "right": "pull_request:5201", "accept": false, - "reason": "Both LoRA-related, but they fix different code paths and bug classes." + "reason": "Both are LoRA-related, but one fixes SDXL LoRA block mapping while the other changes xformers/LoRA processor removal behavior across multiple modules. Distinct fixes." }, { "left": "pull_request:4822", "right": "pull_request:5217", "accept": false, - "reason": "Both are small loader typos, but they fix different text and import errors." + "reason": "Both touch `loaders.py`, but one fixes a user-facing LoRA warning string and the other fixes a PEFT import typo. Separate problems." }, { "left": "pull_request:4856", "right": "pull_request:5986", "accept": false, - "reason": "Both fix docs links, but to different pages and unrelated content." + "reason": "One fixes an internal cross-doc link in SDXL API docs; the other changes a video URL in SVD docs. Same broad docs category only." }, { "left": "pull_request:4856", "right": "pull_request:6604", "accept": false, - "reason": "API link fix vs adding missing autodoc entries; not the same change." + "reason": "One is a link correction in SDXL docs; the other adds missing autodoc members for AutoencoderKL. Different documentation tasks." }, { "left": "pull_request:4869", "right": "pull_request:5818", "accept": false, - "reason": "Combined pipeline tests vs PR workflow changes; unrelated test additions." + "reason": "Combined pipeline tests are unrelated to a workflow PR that restructures PR test matrix setup. Different subsystems and goals." }, { "left": "pull_request:4934", "right": "pull_request:6569", "accept": false, - "reason": "Both are training-example PRs, but they implement different features for different scripts." + "reason": "T2I adapter training script/docs versus modularizing InstructPix2Pix SDXL validation during training; not the same feature or code path." }, { "left": "pull_request:4946", "right": "pull_request:5250", "accept": false, - "reason": "Both update training docs, but for different features: T2I-Adapters vs DDPO." + "reason": "Both update training docs indexes, but one adds T2I-Adapters and the other adds DDPO. Distinct training topics." }, { "left": "pull_request:4952", "right": "pull_request:5180", "accept": false, - "reason": "Docstring wording fix vs citation block formatting fix; different docs changes." + "reason": "One fixes a pipeline offload docstring; the other repairs Wuerstchen docs citation formatting. No shared underlying issue." }, { "left": "pull_request:5013", "right": "pull_request:6647", "accept": false, - "reason": "SDXL dtype runtime fix vs k-diffusion test expectation update; not the same bug." + "reason": "Both relate to SDXL testing/dtype behavior, but one fixes image-latent dtype handling in InstructPix2Pix SDXL and the other only updates a test expectation for k-diffusion outputs. Different code paths." }, { "left": "pull_request:5051", "right": "pull_request:6620", "accept": false, - "reason": "Different docs typos in different pipelines/scripts." + "reason": "Wuerstchen docs typo fixes are unrelated to a script-argument example change in the SDXL LoRA conversion script." }, { "left": "pull_request:5051", "right": "pull_request:6998", "accept": false, - "reason": "Different docs fixes in unrelated areas." + "reason": "Wuerstchen docs typo cleanup versus a ControlNet/IP-Adapter callout formatting fix; different docs and content." }, { "left": "pull_request:5051", "right": "pull_request:7050", "accept": false, - "reason": "Different docs typo cleanup in separate example families." + "reason": "Wuerstchen docs typos and text-to-image example typos are both doc fixes, but they are unrelated files and edits." }, { "left": "pull_request:5065", "right": "pull_request:5961", "accept": false, - "reason": "Attention-processor typo fix vs PixArt docstring typo; unrelated." + "reason": "The first is a large Wuerstchen-related code/docs/model PR; the second is a tiny PixArt docstring default-value correction. No overlap in the underlying change." }, { "left": "pull_request:5083", "right": "pull_request:6433", "accept": false, - "reason": "Issue-template note vs new community pipeline addition." + "reason": "Issue-template guidance for community pipelines is unrelated to adding the Gluegen community pipeline itself." }, { "left": "pull_request:5085", "right": "pull_request:5186", "accept": false, - "reason": "LoRA offloading bug fix vs FreeU proof-of-concept; different functionality." + "reason": "LoRA offloading/import handling fixes are not the same as a FreeU proof-of-concept in UNet blocks; different mechanisms and scope." }, { "left": "pull_request:509", "right": "pull_request:531", "accept": false, - "reason": "Both are test fixes, but for different models and failures." + "reason": "Both are test helpers, but one adds image-loading utilities for SD tests while the other fixes numeric expectations on GPU. Different test failures." }, { "left": "pull_request:5097", "right": "pull_request:5679", "accept": false, - "reason": "Float16 CUDA test updates vs xformers determinism tests; different test concerns." + "reason": "Both are test tweaks for xformers/determinism, but one adds float16 inference coverage and the other revises attn-processor determinism assertions. Not the same bug." }, { "left": "pull_request:5144", "right": "pull_request:6604", "accept": false, - "reason": "BLIP Diffusion sample syntax fix vs AutoencoderKL API docs entries." + "reason": "BLIP Diffusion sample-code parenthesis fix versus AutoencoderKL API docs additions. Different artifacts and issues." }, { "left": "pull_request:5180", "right": "pull_request:6998", "accept": false, - "reason": "Doc-builder citation formatting vs IP-Adapter callout formatting." + "reason": "Wuerstchen citation/doc builder cleanup does not match an IP-Adapter callout formatting correction." }, { "left": "pull_request:5207", "right": "pull_request:5740", "accept": false, - "reason": "README path fix vs RealFill distributed-training code." + "reason": "A README filename correction for T2I adapters is unrelated to a RealFill distributed-training code fix." }, { "left": "pull_request:5298", "right": "pull_request:6511", "accept": false, - "reason": "Different compile-related fixes in different components; not one shared bug." + "reason": "Both involve compile/LoRA-ish training contexts, but one fixes LoRA conv forward behavior and the other makes SDXL DreamBooth checkpointing compatible with `torch.compile`. Different components." }, { "left": "pull_request:5332", "right": "pull_request:5505", "accept": false, - "reason": "Both are typo fixes, but in unrelated example and API naming." + "reason": "DreamBooth LoRA comment typo versus ControlNet API naming fixes; no shared bug or mergeable code change." } ] }, @@ -145358,7 +145438,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 129, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -145448,10 +145528,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -145466,174 +145546,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a loose topical mix of unrelated docs, tests, and code PRs. The soft edges mostly share broad vocabulary (typos, PixArt, IP-Adapter, LCM, Flax) but not the same concrete bug or change.", - "confidence": 0.97, + "summary": "This cluster is heterogeneous: it mixes unrelated docs typo cleanups, scheduler/model fixes, and distinct feature work. None of the soft-paired PRs look like the same underlying bug/change, so all soft edges should be rejected.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR cleanly represents this cluster because the items are not duplicates of one another; if forced, PR 5901 is the most substantive concrete code change, but it is still unrelated to most others.", + "canonical_pr_reason": "No single PR is a good canonical representative because the cluster spans multiple unrelated themes (docs typos, IP-Adapter, PixArt, LoRA, ControlNet, tests, and schedulers).", "best_issue_reason": null, - "best_pr_reason": "PR 5901 (IP-Adapter for StableDiffusionControlNetImg2ImgPipeline) is the strongest standalone, user-facing code change in the set, with a clear code path and functional impact.", + "best_pr_reason": "No PR is globally suitable as a duplicate canonical: the items are too diverse and do not converge on one concrete code-path or change request.", "soft_edge_verdicts": [ { "left": "pull_request:5385", "right": "pull_request:5552", "accept": false, - "reason": "Different changes in unrelated areas: CI copy-check signature handling vs FreeU disable logic." + "reason": "Different fixes in different areas: copy-signature CI utility vs FreeU disable logic. No shared underlying bug." }, { "left": "pull_request:5386", "right": "pull_request:7050", "accept": false, - "reason": "Both are typo/doc cleanup, but they target different files and fixes, not the same underlying change." + "reason": "Both are typo-related, but they fix different docs/examples in different files and do not address the same issue." }, { "left": "pull_request:5386", "right": "pull_request:7131", "accept": false, - "reason": "Separate documentation typo fixes in different sections; not one concrete duplicated edit." + "reason": "Separate docs cleanup PRs with different file sets and different edits; not the same concrete change." }, { "left": "pull_request:5411", "right": "pull_request:7015", "accept": false, - "reason": "Broken training-doc link fix vs single-file load example fix; unrelated doc issues." + "reason": "Broken training doc link fix vs from_single_file example/doc update. Different problems." }, { "left": "pull_request:5506", "right": "pull_request:6437", "accept": false, - "reason": "PEFT/xformers runtime fix vs CI test exclusion change; different problems entirely." + "reason": "PEFT/xformers runtime fix vs CI job exclusion change. One is code behavior, the other is test workflow." }, { "left": "pull_request:555", "right": "pull_request:936", "accept": false, - "reason": "Flax VAE addition vs test file split/refactor; no shared bug or feature." + "reason": "Flax VAE implementation addition vs splitting Stable Diffusion tests. Entirely different work." }, { "left": "pull_request:5615", "right": "pull_request:6947", "accept": false, - "reason": "SDXL adapter LoRA revert vs T2I-Adapter model-card standardization; unrelated." + "reason": "SDXL adapter LoRA revert vs T2I-Adapter model card standardization. Unrelated." }, { "left": "pull_request:5626", "right": "pull_request:7564", "accept": false, - "reason": "Both touch tests, but they fix different test failures in different subsystems." + "reason": "Both touch tests, but they fix different test failures in different subsystems and are not one underlying bug." }, { "left": "pull_request:564", "right": "pull_request:580", "accept": false, - "reason": "Scheduler behavior fix vs initial scheduler test draft; not the same concrete change." + "reason": "Related Flax scheduler area, but one is a scheduler fix and the other is initial scheduler test scaffolding; not the same change." }, { "left": "pull_request:5672", "right": "pull_request:5842", "accept": false, - "reason": "PixArt non-square support vs 512px resolution bug; related model family but different bugs and code paths." + "reason": "Both concern PixArt, but one adds non-square image support and the other fixes a 512px resolution issue. Different bugs." }, { "left": "pull_request:5723", "right": "pull_request:7384", "accept": false, - "reason": "LCM docs addition vs AnimateDiff FreeInit docs; unrelated documentation work." + "reason": "Unrelated docs additions: LCM docs vs AnimateDiff FreeInit docs." }, { "left": "pull_request:5741", "right": "pull_request:6308", "accept": false, - "reason": "LoRA unload optimization vs dummy-component test fix; different code paths and goals." + "reason": "LoRA unload speedup vs dummy-component test fix. Different problems." }, { "left": "pull_request:5744", "right": "pull_request:5898", "accept": false, - "reason": "Large API-doc typo sweep vs a single spacing/doc rewrite; not the same change." + "reason": "General API docs typo cleanup vs a specific unconditional image generation space/link wording fix. Not the same issue." }, { "left": "pull_request:5788", "right": "pull_request:6184", "accept": false, - "reason": "LCM error-message change vs LoRA offline handling; no common bug." + "reason": "LCM scheduler error handling vs LoRA offline weight lookup. Different code paths and bugs." }, { "left": "pull_request:5814", "right": "pull_request:7857", "accept": false, - "reason": "PixArt-Alpha docs vs PixArt-Sigma docs; same theme, but distinct documentation and pipeline additions." + "reason": "PixArt docs about 8-bit inference vs PixArt-Sigma docs/pipeline addition. Different documentation/features." }, { "left": "pull_request:5818", "right": "pull_request:6588", "accept": false, - "reason": "Workflow/test-matrix update vs SDXL k-diffusion slow-test expectation update; unrelated." + "reason": "Test workflow update vs SDXL k-diffusion slow-test expectation update. Not the same bug." }, { "left": "pull_request:5823", "right": "pull_request:6437", "accept": false, - "reason": "LoRA adapter_names feature vs CI exclusion for non-PEFT tests; different concerns." + "reason": "Feature work adding adapter_names to fuse_lora vs CI exclusion for PEFT tests. Different goals." }, { "left": "pull_request:5828", "right": "pull_request:6209", "accept": false, - "reason": "Consistency Decoder test tweak vs text-to-image README script fix; unrelated test/doc maintenance." + "reason": "Consistency decoder memory test fix vs text-to-image README script fix. No shared underlying issue." }, { "left": "pull_request:5828", "right": "pull_request:6808", "accept": false, - "reason": "Consistency Decoder test tweak vs MPS test-collection fix; different failures." + "reason": "Different test-related fixes: one is a VAE test, the other MPS collection handling." }, { "left": "pull_request:5901", "right": "pull_request:6140", "accept": false, - "reason": "Feature implementation vs API documentation; not the same underlying change." + "reason": "Code feature for ControlNet Img2Img IP-Adapter vs standalone IP-Adapter API docs. Docs and implementation are not duplicates." }, { "left": "pull_request:5901", "right": "pull_request:6244", "accept": false, - "reason": "Both mention IP-Adapter, but one is ControlNet Img2Img integration and the other is a broader embeddings refactor; not mergeable as one fix." + "reason": "Both involve IP-Adapter, but one integrates it into ControlNet Img2Img while the other refactors embeddings across models. Not one concrete fix." }, { "left": "pull_request:5901", "right": "pull_request:6293", "accept": false, - "reason": "Same general IP-Adapter theme, but different pipeline variants (SD vs SDXL) and separate code paths." + "reason": "Similar feature family, but SD v1 ControlNet Img2Img vs SDXL ControlNet Img2Img are distinct pipelines and changes." }, { "left": "pull_request:5901", "right": "pull_request:6875", "accept": false, - "reason": "Core feature addition vs a test expectation update; not the same bug/change." + "reason": "Implementation PR vs test expectation update for IP-Adapter. Different artifacts and not mergeable as one change." }, { "left": "pull_request:5922", "right": "pull_request:7103", "accept": false, - "reason": "MultiAdapter inpainting fix vs ControlNet image validation bug; unrelated pipeline issues." + "reason": "MultiAdapter inpainting fix vs ControlNet input validation fix. Different bug reports." }, { "left": "pull_request:5961", "right": "pull_request:7827", "accept": false, - "reason": "PixArt docstring typo vs CPU offload docstring fix; different files and concerns." + "reason": "PixArt docstring typo vs ControlNet CPU offload docstring fix. Different files and issues." }, { "left": "pull_request:5983", "right": "pull_request:6389", "accept": false, - "reason": "SVD documentation cleanup vs Stable Video Diffusion type annotation fix; unrelated." + "reason": "SVD docs cleanup vs Stable Video Diffusion parameter type fix. Different change types and code paths." }, { "left": "pull_request:5986", "right": "pull_request:6314", "accept": false, - "reason": "Video link replacement vs T2I-Adapter paper link fix; unrelated doc edits." + "reason": "Broken video link fix vs T2I-Adapter paper link typo. Unrelated docs issues." } ] }, @@ -145649,7 +145729,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 130, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -145740,10 +145820,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -145758,186 +145838,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly unrelated docs, test updates, and separate bug fixes. The soft edges only show loose thematic overlap (same subsystem or same file in a few cases), not the same concrete change, so none should be merged as duplicates.", - "confidence": 0.97, + "summary": "These items are mostly unrelated one-off docs, examples, tests, or small bug fixes. I don\u2019t see any true duplicate PRs among the soft pairs; several are in the same subsystem, but they address different concrete problems or files.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a clear canonical duplicate target; the items do not share one underlying bug/change. If forced to pick the most representative standalone fix, PR 6293 is the strongest concrete code-path change with tests.", + "canonical_pr_reason": "PR 6085 is a clear, substantive bug fix in core scheduler behavior with concrete code-path changes and broad impact; it is the strongest representative of the set.", "best_issue_reason": null, - "best_pr_reason": "PR 6293 is the most substantive standalone fix in the set: it adds IP-Adapter support to the SDXL ControlNet img2img pipeline and includes direct code-path changes plus tests. It is still not a good canonical for the whole cluster.", + "best_pr_reason": "PR 6085 is the best standalone PR in this cluster because it fixes an actual runtime bug in `add_noise` across multiple scheduler implementations, rather than a docs/test-only or narrow example change.", "soft_edge_verdicts": [ { "left": "pull_request:5986", "right": "pull_request:6462", "accept": false, - "reason": "Both are docs-only, but they fix different links on different pages; not the same underlying change." + "reason": "Both are docs fixes, but one updates a video link and the other corrects a paper reference for aMUSEd; different content and no shared bug." }, { "left": "pull_request:5986", "right": "pull_request:7937", "accept": false, - "reason": "Unrelated documentation edits: a video link fix vs adding a VideoProcessor docs header." + "reason": "Unrelated documentation changes: a broken SVD video link versus adding a VideoProcessor docs header." }, { "left": "pull_request:6021", "right": "pull_request:6305", "accept": false, - "reason": "Different changes in different places: adding a variant CLI arg vs adding rescale_zero_snr support to DDPMScheduler." + "reason": "Both touch training/config args, but one adds `variant` to a DreamBooth LoRA script and the other adds zero-SNR rescaling to DDPMScheduler; different code paths." }, { "left": "pull_request:6021", "right": "pull_request:6670", "accept": false, - "reason": "Both pass a variant argument, but they target different scripts and different code paths." + "reason": "Both involve `variant`, but one is a training script arg and the other updates checkpoint_merger loading; not the same underlying change." }, { "left": "pull_request:6085", "right": "pull_request:7103", "accept": false, - "reason": "Separate scheduler add_noise bug vs ControlNet input validation bug; same broad area only." + "reason": "Different bugs: scheduler `add_noise` step indexing versus ControlNet image validation logic." }, { "left": "pull_request:6106", "right": "pull_request:6592", "accept": false, - "reason": "Different scripts and different fixes; one is a doc/example pipe example update, the other is a new inpainting training script WIP." + "reason": "One fixes a generated example in a training script; the other is a new inpainting training script WIP. No duplicate bug/change." }, { "left": "pull_request:6122", "right": "pull_request:7325", "accept": false, - "reason": "Both are typo/docs sweeps, but they touch different docs and code areas with no single bug/change in common." + "reason": "Both are typo/docs cleanup, but they affect different files and unrelated content areas." }, { "left": "pull_request:6140", "right": "pull_request:6760", "accept": false, - "reason": "One adds IP-Adapter API docs; the other adjusts IP-Adapter slow test expectations. Not the same change." + "reason": "Adding IP-Adapter API docs is unrelated to updating slow-test expectations for IP-Adapter." }, { "left": "pull_request:6184", "right": "pull_request:7200", "accept": false, - "reason": "LoRA offline weight lookup fix is unrelated to transformer norm-type validation." + "reason": "LoRA offline weight-name handling is unrelated to Transformer2D layernorm/patch-input validation." }, { "left": "pull_request:62", "right": "pull_request:869", "accept": false, - "reason": "These are from different eras and different scopes; pipeline behavior fix vs community README/import cleanup." + "reason": "Very different changes: latent diffusion pipeline behavior versus community README/import bug fixes." }, { "left": "pull_request:6209", "right": "pull_request:7747", "accept": false, - "reason": "README example script fix vs VAE tiling test fix; different artifacts and bugs." + "reason": "One fixes a README test script; the other adjusts a VAE tiling test assertion. Same docs/tests area, but not the same bug." }, { "left": "pull_request:6209", "right": "pull_request:8470", "accept": false, - "reason": "A training README correction is unrelated to updating a research-project README." + "reason": "Example README fix versus research-project README rewrite; unrelated documentation changes." }, { "left": "pull_request:6244", "right": "pull_request:6760", "accept": false, - "reason": "Broad IP-Adapter refactor vs test expectation adjustment; not mergeable as one fix." + "reason": "IP-Adapter refactor in model/pipeline code versus slow-test expectation updates; the latter is test maintenance, not the same code-path bug." }, { "left": "pull_request:6245", "right": "pull_request:6247", "accept": false, - "reason": "They overlap on one Wuerstchen file, but one fixes fp16 LoRA training args and the other adds a dreambooth example/documentation; different changes." + "reason": "Same Wuerstchen area, but one fixes fp16 training/LoRA config and the other adds a dreambooth LoRA example; different changes." }, { "left": "pull_request:6282", "right": "pull_request:6592", "accept": false, - "reason": "Dreambooth runtime cleanup vs adding a new inpainting training script; not the same bug or feature." + "reason": "Dreambooth runtime-error cleanup is unrelated to adding a new inpainting training script." }, { "left": "pull_request:6293", "right": "pull_request:6528", "accept": false, - "reason": "Both involve IP-Adapter ControlNet, but one adds SDXL img2img support while the other fixes a missing code path in ControlNet img2img; distinct fixes." + "reason": "Both IP-Adapter-related, but one adds IP-Adapter support to SDXL ControlNet Img2Img while the other fixes missing code in ControlNet img2img. Different pipelines and bugs." }, { "left": "pull_request:6293", "right": "pull_request:6760", "accept": false, - "reason": "Pipeline feature work vs test-value updates; the latter is just test maintenance." + "reason": "Feature addition versus slow-test threshold adjustment; not the same underlying issue." }, { "left": "pull_request:6337", "right": "pull_request:6912", "accept": false, - "reason": "retrieve_latents usage in a TensorRT img2img example is unrelated to denoising_start truthiness handling." + "reason": "One fixes `retrieve_latents` usage in a TensorRT img2img example; the other fixes truthiness checks for `denoising_start` across several pipelines." }, { "left": "pull_request:6604", "right": "pull_request:6837", "accept": false, - "reason": "AutoencoderKL docs API exposure vs i2vgen docs dependency ordering; unrelated docs changes." + "reason": "Missing API docs for AutoencoderKL versus docs-only dependency order tweaks in I2VGenXL; unrelated." }, { "left": "pull_request:6604", "right": "pull_request:7425", "accept": false, - "reason": "Missing API doc entry vs a PEFT tutorial formatting/content fix; no shared bug." + "reason": "AutoencoderKL API docs versus a PEFT inference tutorial fix; different documentation targets." }, { "left": "pull_request:6680", "right": "pull_request:7719", "accept": false, - "reason": "Slow test scheduler config tweak vs CI workflow auth key addition; different concerns." + "reason": "Slow test failure fix for DPM scheduler versus adding a Tailscale auth key to workflow failure handling; unrelated." }, { "left": "pull_request:6760", "right": "pull_request:6875", "accept": false, - "reason": "Both touch IP-Adapter tests, but they update different expected outputs in different test cases." + "reason": "Both update IP-Adapter tests, but they cover different expected outputs and scenarios; not a single duplicate bug." }, { "left": "pull_request:6775", "right": "pull_request:8675", "accept": false, - "reason": "DDPMScheduler parameter docs vs a fast-diffusion tutorial note; unrelated." + "reason": "A DDPMScheduler parameter docs change is unrelated to a fast-diffusion note/caching docs update." }, { "left": "pull_request:6787", "right": "pull_request:7156", "accept": false, - "reason": "Marigold input normalization fix vs stale bot label normalization; different code paths." + "reason": "Marigold input normalization bug fix versus stalebot label handling; completely different code paths." }, { "left": "pull_request:6795", "right": "pull_request:6997", "accept": false, - "reason": "Dreambooth SDXL micro-conditioning args vs adding LoRA support to a spatio-temporal UNet; unrelated changes." + "reason": "DreamBooth SDXL micro-conditioning vs adding LoRA support to UNetSpatioTemporalConditionModel; different features." }, { "left": "pull_request:6808", "right": "pull_request:7747", "accept": false, - "reason": "MPS test-collection helper fix vs VAE tiling test fix; no common underlying bug." + "reason": "MPS test-collection fix in testing utilities versus a VAE tiling assertion fix; unrelated test issues." }, { "left": "pull_request:6871", "right": "pull_request:9070", "accept": false, - "reason": "VAE tiling/slicing in img2img and inpaint is unrelated to adding a FLUX img2img pipeline." + "reason": "Adding VAE tiling/slicing to SD img2img/inpaint is unrelated to a new FLUX img2img pipeline." }, { "left": "pull_request:6875", "right": "pull_request:7868", "accept": false, - "reason": "IP-Adapter expected-output update vs offload test assertion change; different test failures." + "reason": "IP-Adapter output expectation updates versus offload test assertions; different test behaviors and failures." }, { - "left": "pull_request:7068", - "right": "pull_request:7291", + "left": "pull_request:6912", + "right": "pull_request:7068", "accept": false, - "reason": "Both are typo/documentation fixes, but they target different files and different wording; not the same issue." + "reason": "Different bug classes: denoising_start truthiness versus typo/doc cleanup across loaders and pipeline examples." } ] }, @@ -145953,7 +146033,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 131, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -146043,10 +146123,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -146061,180 +146141,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab-bag of unrelated small PRs (docs typos, test fixes, scheduler additions, training-script features, CI/config changes). I don\u2019t see a single underlying bug or change that would justify a canonical duplicate anchor.", + "summary": "A heterogeneous set of unrelated PRs (docs typos, tests, CI, scheduler additions, training feature work, and bug fixes). The soft pairs mostly share superficial keywords or filenames but do not fix the same concrete problem, so none should be clustered as duplicates.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR fits this cluster: the items are not one underlying bug/change but many independent PRs across docs, tests, CI, and features.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 7530 is the strongest standalone representative because it fixes a concrete runtime bug, has an explicit issue target, and is substantially more substantive than the docs/test-only changes.", "soft_edge_verdicts": [ { "left": "pull_request:7120", "right": "pull_request:7360", "accept": false, - "reason": "Different changes: adding an EDM DPM scheduler vs doc-only TCD scheduler wording/type fixes." + "reason": "Different scheduler work: one adds an EDM DPM solver scheduler, the other tweaks TCD scheduler docs/signatures. Same broad area, not the same change." }, { "left": "pull_request:7150", "right": "pull_request:7594", "accept": false, - "reason": "Same docs file, but different callback example fixes in different sections; not the same bug or change." + "reason": "Both touch callback docs, but one updates a code example and the other fixes a separate example bug. Superficial overlap only." }, { "left": "pull_request:7156", "right": "pull_request:9021", "accept": false, - "reason": "Unrelated: stalebot label handling vs SD3 ControlNet transformer layer loading logic." + "reason": "Unrelated changes: stalebot label handling vs SD3 ControlNet layer-loading logic." }, { "left": "pull_request:7164", "right": "pull_request:7564", "accept": false, - "reason": "Different subsystems: Docker/ONNX test environment pinning vs pipeline test/deprecation handling." + "reason": "CI Dockerfile pinning vs pipeline scheduler deprecation handling; no shared bug or code path." }, { "left": "pull_request:7206", "right": "pull_request:7235", "accept": false, - "reason": "Related training area, but one adds tags across many scripts while the other adds DoRA support to one script." + "reason": "Both are training-script related, but one adds tags across many scripts while the other adds DoRA support to one script. Different goals and diffs." }, { "left": "pull_request:7233", "right": "pull_request:8702", "accept": false, - "reason": "Different features: DoRA training in SDXL DreamBooth vs an SD3 checkpoint conversion script." + "reason": "DoRA training option in DreamBooth LoRA vs SD3 checkpoint conversion script. Completely different code paths." }, { "left": "pull_request:7259", "right": "pull_request:7384", "accept": false, - "reason": "Both are docs additions, but one is a new TCD intro page and the other adds FreeInit outputs to AnimateDiff docs." + "reason": "Separate docs additions: TCD intro page vs FreeInit outputs in AnimateDiff docs." }, { "left": "pull_request:7291", "right": "pull_request:7451", "accept": false, - "reason": "UNet doc typo fix vs broad docs typo sweep; not the same concrete issue." + "reason": "A documentation default-value typo fix is unrelated to broad docs typo cleanup across many files." }, { "left": "pull_request:7335", "right": "pull_request:8908", "accept": false, - "reason": "One is a controlnet SDXL config test fix; the other is a request-caching skip/annotation fix in model tests." + "reason": "ControlNet SDXL config test/docs ordering fix vs a request-caching skip annotation. Different test failures." }, { "left": "pull_request:7411", "right": "pull_request:9077", "accept": false, - "reason": "Both are typo/documentation sweeps, but they touch many unrelated files and address different text issues." + "reason": "One is a general typo cleanup across schedulers/docs, the other mixes unrelated typo fixes across several files; no single underlying bug." }, { "left": "pull_request:7473", "right": "pull_request:8272", "accept": false, - "reason": "Same general area of error-message cleanup, but different code paths and unrelated affected modules." + "reason": "Both fix incorrect `raise` statements, but they target different files and different input-validation sites; not the same concrete bug." }, { "left": "pull_request:7495", "right": "pull_request:8470", "accept": false, - "reason": "README link cleanup vs updating a research-project README to a different project; no shared bug/change." + "reason": "README link cleanup vs replacing a research-project README with a different project description. Not the same change." }, { "left": "pull_request:7530", "right": "pull_request:7903", "accept": false, - "reason": "Autocast behavior in training scripts vs a video processor test/validation PR; unrelated code paths." + "reason": "Autocast behavior in training scripts vs a video processor input-validation test harness. Different subsystems and bugs." }, { "left": "pull_request:7530", "right": "pull_request:8064", "accept": false, - "reason": "Autocast fix for training scripts vs Marigold pipeline implementation/test work; not the same change." + "reason": "Autocast bug fix vs Marigold pipeline feature/test addition. No shared code path." }, { "left": "pull_request:7542", "right": "pull_request:7977", "accept": false, - "reason": "CPU offload test typo vs a docs typo in fp16 optimization guide; unrelated." + "reason": "CPU offload test typo vs docs typo in an unrelated optimization guide." }, { "left": "pull_request:7542", "right": "pull_request:8498", "accept": false, - "reason": "Test typo fix vs stable-diffusion-3 docs typo; different files and problems." + "reason": "Test typo fix vs a docs string/prompt typo fix. No underlying duplicate." }, { "left": "pull_request:7564", "right": "pull_request:7966", "accept": false, - "reason": "Pipeline deprecation/test fixes vs PixArt Sigma test expectation updates; different concrete failures." + "reason": "Scheduler deprecation/test updates vs PixArt Sigma test expectation updates. Both are test-related but different failures." }, { "left": "pull_request:7564", "right": "pull_request:8720", "accept": false, - "reason": "Pipeline test/deprecation fixes vs slow-test precision updates; both tests, but different failures and code paths." + "reason": "Pipeline deprecation/test adjustments vs precision-threshold updates in slow pipeline tests. Not the same bug." }, { "left": "pull_request:7590", "right": "pull_request:9060", "accept": false, - "reason": "Setup.py Python version range change vs removing a stray print in Flux transformer." + "reason": "Python packaging version-range change is unrelated to removing a debug print in a model forward pass." }, { "left": "pull_request:7632", "right": "pull_request:8926", "accept": false, - "reason": "Different DreamBooth issues: VAE loading path bug vs checkpoint save logic for text encoders." + "reason": "Both concern DreamBooth LoRA scripts, but one fixes VAE path loading and the other fixes checkpoint saving for advanced SDXL LoRA. Different issues." }, { "left": "pull_request:7747", "right": "pull_request:9627", "accept": false, - "reason": "VAE tiling test precision fix vs a DoRA warning/test ordering fix; unrelated." + "reason": "VAE tiling test fix vs DoRA LoRA warning test ordering. Different test cases and behaviors." }, { "left": "pull_request:7747", "right": "pull_request:9683", "accept": false, - "reason": "VAE tiling test fix vs CogVideoX test rename/unskip; different tests and failures." + "reason": "VAE tiling test fix vs CogVideoX integration test rename/unskip. Not duplicates." }, { "left": "pull_request:7767", "right": "pull_request:8498", "accept": false, - "reason": "PixArt pipeline cleanup vs a docs typo fix in SD3; no shared bug or code path." + "reason": "PixArt pipeline refactor/removal vs a single docs typo fix. No shared bug/change." }, { "left": "pull_request:7832", "right": "pull_request:9318", "accept": false, - "reason": "CI cache mount path update vs nightly test token secret update; both CI but different infrastructure issues." + "reason": "CI cache mount path update vs nightly hub token secret rename. Both CI-only, but different infra problems." }, { "left": "pull_request:7916", "right": "pull_request:9242", "accept": false, - "reason": "Ascend NPU support in one training script vs xpu testing utilities support; different functionality and modules." + "reason": "Ascend NPU support in SDXL training is unrelated to xpu test backend support." }, { "left": "pull_request:7966", "right": "pull_request:8720", "accept": false, - "reason": "Different slow-test maintenance: PixArt Sigma-specific expectations vs broad pipeline precision adjustments." + "reason": "PixArt Sigma test expectation changes vs broader precision-related slow test fixes. Similar test domain, but not the same concrete change." }, { "left": "pull_request:810", "right": "pull_request:822", "accept": false, - "reason": "The older PR removes/adapts a scheduler API call, while the later PR restores proper scale/init noise behavior; related file, different fixes." + "reason": "These are related to the same Flax scheduler API transition, but they address different stages of the update and different files; not the same duplicate change." }, { "left": "pull_request:810", "right": "pull_request:910", "accept": false, - "reason": "Flax scheduler API adaptation vs in-painting docs update; unrelated." + "reason": "Flax scheduler API adaptation vs initial inpainting docs update. Completely unrelated." } ] }, @@ -146250,7 +146330,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 132, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -146339,10 +146419,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -146357,168 +146437,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: mostly unrelated PRs spanning Flux, SD3, Stable Diffusion, CI, docs, tests, and examples. None of the soft-edge pairs look like the same underlying change, so I would not consolidate any of them.", + "summary": "This is a mixed set of unrelated PRs across Flux, SD3, ControlNet, CI, docs, and tests. The soft edges are only broad topical similarities or shared filenames, not the same concrete bug/change, so they should all be rejected.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR represents a duplicate set here; the items cover distinct feature work, bug fixes, docs, CI, and test-only changes across different code paths.", + "canonical_pr_reason": "No single PR cleanly represents this cluster; the items are independent changes in different subsystems, so there is no meaningful canonical PR.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "No global best PR stands out because the cluster is not one duplicate family; picking any single PR would misrepresent the rest.", "soft_edge_verdicts": [ { "left": "pull_request:828", "right": "pull_request:864", "accept": false, - "reason": "Different changes: one adjusts Flax pipeline tests, the other adds a Flax scheduler helper and scheduler tests." + "reason": "Different Flax test/scheduler changes; not the same bug or fix." }, { "left": "pull_request:8290", "right": "pull_request:9307", "accept": false, - "reason": "Both mention embeddings-related areas, but one is a broad HunyuanDIT doc/CI refactor and the other is a rotary-embedding implementation change." + "reason": "Shared embeddings file, but unrelated HunyuanDIT refactor vs rotary-embedding fix." }, { "left": "pull_request:832", "right": "pull_request:957", "accept": false, - "reason": "Unrelated bug fix in Flax NSFW handling versus a text-to-image scheduler API cleanup." + "reason": "Unrelated bugs in Flax nsfw handling vs text-to-image scheduler args." }, { "left": "pull_request:8422", "right": "pull_request:8696", "accept": false, - "reason": "Subfolder-loading test additions versus marking a few Marigold tests flaky; different purposes and code paths." + "reason": "Subfolder-loading tests vs Marigold flaky-test annotations; different concerns." }, { "left": "pull_request:8496", "right": "pull_request:8755", "accept": false, - "reason": "SD3 warning-log behavior versus DreamBooth SD3 text-encoder training logic; not the same bug." + "reason": "Transformer warning-log tweak vs SD3 dreambooth training bug; unrelated." }, { "left": "pull_request:8583", "right": "pull_request:8838", "accept": false, - "reason": "Both touch SD3 transformer code, but one adds XFormers support while the other adds forward-chunking disable support." + "reason": "Both touch SD3 transformer, but one adds XFormers support and the other forward chunking." }, { "left": "pull_request:8600", "right": "pull_request:9980", "accept": false, - "reason": "DreamBooth script cleanup versus LoRA conversion mapping changes; different features and failure modes." + "reason": "Dreambooth script cleanup vs LoRA conversion mapping; different code paths." }, { "left": "pull_request:8627", "right": "pull_request:9251", "accept": false, - "reason": "SD3 ControlNet docs/example work versus a CI workflow trigger change; no shared underlying bug." + "reason": "SD3 ControlNet docs/code vs CI trigger change; unrelated." }, { "left": "pull_request:8838", "right": "pull_request:9249", "accept": false, - "reason": "Transformer utility addition versus a full SD3 ControlNet training example; related domain, but not the same change." + "reason": "SD3 transformer utility vs SD3 ControlNet example; not the same change." }, { "left": "pull_request:8851", "right": "pull_request:9282", "accept": false, - "reason": "Both are docs updates, but one adds AuraFlow docs while the other only updates the pipeline overview table." + "reason": "AuraFlow docs vs pipeline table update; different doc changes." }, { "left": "pull_request:889", "right": "pull_request:936", "accept": false, - "reason": "Pipeline component API addition versus splitting Stable Diffusion tests into separate files; unrelated." + "reason": "Both touch pipeline utilities/tests, but address different features and bugs." }, { "left": "pull_request:8913", "right": "pull_request:8926", "accept": false, - "reason": "Checkpoint loading from subfolders versus DreamBooth checkpoint saving; opposite directions and distinct bugs." + "reason": "Checkpoint loading fix vs checkpoint saving fix; different failures." }, { "left": "pull_request:9005", "right": "pull_request:9699", "accept": false, - "reason": "Both change runner config, but one rewires workflow groups broadly while the other only adds a new selectable SSH runner." + "reason": "Runner config refactor vs adding a new runner option; not the same issue." }, { "left": "pull_request:9021", "right": "pull_request:9838", "accept": false, - "reason": "SD3 ControlNet layer-loading fix versus Flux DreamBooth memory/repetition logic; different models and issues." + "reason": "SD3ControlNet num_layers bug vs Flux training memory fix; unrelated." }, { "left": "pull_request:9060", "right": "pull_request:9286", "accept": false, - "reason": "Removing a stray debug print is unrelated to exporting a new class in __init__.py." + "reason": "Removing a debug print vs exporting FluxMultiControlNetModel; unrelated." }, { "left": "pull_request:9070", "right": "pull_request:9754", "accept": false, - "reason": "New Flux img2img pipeline versus a Flux LoRA quantization training project; not the same code-path problem." + "reason": "Flux img2img pipeline addition vs Flux quantized training example; different artifacts." }, { "left": "pull_request:9196", "right": "pull_request:9598", "accept": false, - "reason": "PixArt attention-processor API addition versus documentation note for Flux LoRA tests; unrelated." + "reason": "PixArt attention API change vs Flux LoRA test notes; unrelated." }, { "left": "pull_request:9355", "right": "pull_request:9832", "accept": false, - "reason": "Workflow dispatch for fast GPU tests versus adding quantization nightly tests; both CI-related but different changes." + "reason": "Both are CI-related, but one adds manual dispatch and the other adds quantization tests." }, { "left": "pull_request:9379", "right": "pull_request:9442", "accept": false, - "reason": "Flux fast-test fixes versus a CI workaround for deprecated attention blocks; not the same underlying bug." + "reason": "Fast GPU test fixes for different failures; not one concrete bug." }, { "left": "pull_request:9385", "right": "pull_request:9571", "accept": false, - "reason": "Different Flux ControlNet work: one adds Xlabs ControlNet support, the other adds guidance start/end controls to existing pipelines." + "reason": "Different controlnet implementations and different feature additions." }, { "left": "pull_request:9385", "right": "pull_request:9620", "accept": false, - "reason": "Xlabs ControlNet integration versus adding interrupt support to ControlNet pipelines; separate features." + "reason": "Xlabs ControlNet support vs interrupt handling; separate problems." }, { "left": "pull_request:9385", "right": "pull_request:9658", "accept": false, - "reason": "Different artifacts and goals: Xlabs ControlNet support versus a community demo for distilled Flux ControlNet." + "reason": "Xlabs ControlNet support vs Flux distilled ControlNet example; unrelated." }, { "left": "pull_request:9385", "right": "pull_request:9860", "accept": false, - "reason": "Xlabs ControlNet support versus SD3.5 ControlNet example support; distinct models and code paths." + "reason": "Different model families: Xlabs Flux ControlNet vs SD3.5 ControlNet example." }, { "left": "pull_request:4847", "right": "pull_request:6752", "accept": false, - "reason": "Prompt Free Diffusion community example and PromptDiffusion research pipeline are different projects, not the same change." + "reason": "Prompt Free Diffusion community example vs PromptDiffusion research project; distinct projects." }, { "left": "pull_request:10122", "right": "pull_request:9257", "accept": false, - "reason": "Flux attention-mask fix versus DreamBooth Flux prompt/tokenization bug fix; separate code paths and symptoms." + "reason": "Flux attention-mask fix vs Flux Dreambooth training bug; different code paths." }, { "left": "pull_request:1527", "right": "pull_request:3298", "accept": false, - "reason": "Stable Diffusion inpaint tensor input support versus arbitrary aspect ratio support in IFSuperResolution; unrelated pipeline changes." + "reason": "Stable Diffusion inpaint tensor input fix vs IF super-resolution aspect-ratio support; unrelated." } ] }, @@ -146534,7 +146614,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 133, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -146622,10 +146702,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -146640,168 +146720,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: most soft links are just broad subsystem overlap (docs, CI, Dockerfiles, model additions, or test refactors) rather than the same concrete bug or change. I would not merge any of these pairs as duplicates.", + "summary": "All soft edges look like false positives: they mostly share broad subsystems or filenames, but the underlying changes are different bugs, features, or docs/CI updates.", "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No single pull request is a credible canonical anchor for this cluster because the linked PRs span unrelated changes across docs, CI, models, pipelines, and tests.", + "canonical_pr_reason": "No single canonical PR: the soft links span unrelated fixes across CI, docs, model additions, tests, and refactors, so none is a safe cluster representative.", "best_issue_reason": null, - "best_pr_reason": "No PR is a strong global representative; the cluster does not share one underlying fix/change. If forced, none of the PRs is suitable as canonical.", + "best_pr_reason": "No PR in this set is globally representative of the whole cluster; the items are heterogeneous and should stay separate.", "soft_edge_verdicts": [ { "left": "pull_request:1564", "right": "pull_request:3042", "accept": false, - "reason": "Both touch DreamBooth-related docs/requirements, but one adds a default accelerate config and the other raises an accelerate minimum version. Different changes." + "reason": "Both touch training docs, but one adds a default accelerate config note and the other raises accelerate minimum versions for example requirements; different changes." }, { "left": "pull_request:5188", "right": "pull_request:5393", "accept": false, - "reason": "One adds compute_snr to training utils; the other updates training docs/examples to use HFApi. Same area, not the same bug/change." + "reason": "Both affect training examples, but one centralizes compute_snr and the other updates HFApi usage in tutorials/examples; not the same fix." }, { "left": "pull_request:7634", "right": "pull_request:8404", "accept": false, - "reason": "One adds a high-res fix UNet/pipeline; the other adds Latte video generation. Completely different features." + "reason": "High-res fix for SD vs adding Latte video model support; only overlap is generic model registration/docs." }, { "left": "pull_request:7893", "right": "pull_request:9721", "accept": false, - "reason": "Both are Dockerfile maintenance, but one upgrades Python versions and the other pins torch versions. Different infra issues." + "reason": "Both are Dockerfile environment changes, but one upgrades Python while the other pins torch versions; unrelated fixes." }, { "left": "pull_request:11588", "right": "pull_request:11639", "accept": false, - "reason": "Both involve Wan video code, but one extends LoRA I2V support and the other standardizes VAE encode sampling. Different concrete code paths." + "reason": "Same Wan area, but one expands LoRA weights for I2V and the other standardizes VAE sampling mode in video-to-video; different code paths." }, { "left": "pull_request:10132", "right": "pull_request:9426", "accept": false, - "reason": "Both are CI workflow edits, but one installs tokenizers for a PEFT test job and the other updates artifact uploader versions. Not the same fix." + "reason": "Both touch CI, but one installs tokenizers for a backend test job and the other updates artifact uploader versions; separate changes." }, { "left": "pull_request:10597", "right": "pull_request:9987", "accept": false, - "reason": "One only updates an SD3 docs checkpoint id; the other adds SD3.5 IP-Adapter pipeline integration. Different work items." + "reason": "Both mention SD3, but one updates a docs model ID while the other integrates SD3.5 IP-Adapter pipeline support; different functionality." }, { "left": "pull_request:10148", "right": "pull_request:8702", "accept": false, - "reason": "OmniGen addition vs SD3 conversion script. Unrelated functionality." + "reason": "OmniGen feature addition versus an SD3 conversion script; completely different artifacts and code paths." }, { "left": "pull_request:10197", "right": "pull_request:8978", "accept": false, - "reason": "Both affect CI, but one changes TPU runner config and the other is a large example/notebook update. Different concerns." + "reason": "Both are broad workflow/example updates, but one changes TPU CI setup and the other updates DreamBooth/Imagic examples; unrelated." }, { "left": "pull_request:10200", "right": "pull_request:8722", "accept": false, - "reason": "One refactors ControlNet Union; the other refactors embeddings internals. Broad refactors, but not the same underlying change." + "reason": "Different subsystems: ControlNet union refactor versus embeddings refactor; no shared concrete bug or change." }, { "left": "pull_request:10271", "right": "pull_request:9524", "accept": false, - "reason": "TorchAO test refactor vs CogVideoX VAE conv_cache refactor. No shared bug or code path." + "reason": "One refactors TorchAO serialization tests; the other refactors CogVideoX VAE internals; unrelated test/model changes." }, { "left": "pull_request:10370", "right": "pull_request:10850", "accept": false, - "reason": "Style/compat cleanup in torchao quantizer vs adding diffusers hook helper utilities. Different changes." + "reason": "TorchAO quantizer style cleanup versus hook helper utilities; different purposes and code paths." }, { "left": "pull_request:10547", "right": "pull_request:11052", "accept": false, - "reason": "Both touch tests, but one fixes offload tests for specific transformers and the other adds a CUDA compatibility test decorator. Not duplicates." + "reason": "Both are test-related, but one fixes offload tests for specific models and the other adds compute-capability gating; distinct issues." }, { "left": "pull_request:10554", "right": "pull_request:9077", "accept": false, - "reason": "One fixes a logic bug in SD3 LoRA loading; the other is a typo/docs cleanup PR. Not the same issue." + "reason": "SD3 LoRA example fix versus generic typo fixes; not the same underlying bug." }, { "left": "pull_request:10554", "right": "pull_request:9739", "accept": false, - "reason": "One fixes the transformer-vs-unet state-dict filter bug; the other updates SD3 docs/example text. Different scope." + "reason": "Both touch SD3-related docs/examples, but one fixes a training script state-dict filter and the other adds single-file SD3.5 docs; different changes." }, { "left": "pull_request:1137", "right": "pull_request:1527", "accept": false, - "reason": "Legacy inpaint safety-checker dtype fix vs allowing tensor image/mask inputs in inpaint. Same family, but different bugs." + "reason": "Same inpaint family, but one fixes safety-checker dtype handling and the other allows tensor image/mask inputs; separate bugs." }, { "left": "pull_request:11813", "right": "pull_request:11818", "accept": false, - "reason": "One is Kontext training docs/code; the other removes syncs in inference denoising loops. Different code paths." + "reason": "Kontext training changes versus a runtime sync optimization in Flux denoising; unrelated." }, { "left": "pull_request:12062", "right": "pull_request:12450", "accept": false, - "reason": "Both are CI-wide changes, but one enables all GPUs and the other revisits dependency installation. Not one concrete bug." + "reason": "Both CI-related, but one enables all GPUs in workflows and the other revisits installation steps; different operational fixes." }, { "left": "pull_request:1219", "right": "pull_request:397", "accept": false, - "reason": "DDIM docs vs Quicktour docs. Both documentation-only, but unrelated content." + "reason": "DDIM docs versus Quicktour docs; unrelated documentation." }, { "left": "pull_request:1219", "right": "pull_request:415", "accept": false, - "reason": "DDIM docs vs training overview/docs. Different documentation updates." + "reason": "DDIM docs versus training overview/docs; unrelated documentation updates." }, { "left": "pull_request:12360", "right": "pull_request:12797", "accept": false, - "reason": "Chroma docs fix vs Kandinsky docs fixes. Unrelated model docs." + "reason": "Different docs for different models: Chroma guidance wording versus Kandinsky minor fixes." }, { "left": "pull_request:12637", "right": "pull_request:13146", "accept": false, - "reason": "Sana video copy cleanup vs MT5Tokenizer type fix in PAG HunyuanDiT. Different models and issues." + "reason": "Sana video transformer copy cleanup versus tokenizer type fix in PAG HunyuanDiT; unrelated code changes." }, { "left": "pull_request:1424", "right": "pull_request:397", "accept": false, - "reason": "Stable Diffusion 2 docs vs Quicktour. Different docs packages." + "reason": "Stable Diffusion 2 docs versus Quicktour docs; different documentation additions." }, { "left": "pull_request:1424", "right": "pull_request:415", "accept": false, - "reason": "Stable Diffusion 2 docs vs training overview docs. Not the same change." + "reason": "Stable Diffusion 2 docs versus training overview/docs; unrelated documentation." }, { "left": "pull_request:1468", "right": "pull_request:2655", "accept": false, - "reason": "Saving trained betas in schedulers/config vs LoRA saving/loading fixes. Both persistence-related, but distinct bugs." + "reason": "Saving trained betas across schedulers versus LoRA save/load behavior; different config and loader problems." }, { "left": "pull_request:159", "right": "pull_request:718", "accept": false, - "reason": "Both are typo cleanup PRs, but they touch different files and eras; no shared underlying bug/change." + "reason": "Both are typo/wording fixes, but they target different files and unrelated text changes; too broad to merge as one change." } ] }, @@ -146817,7 +146897,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 134, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -146906,10 +146986,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -146924,174 +147004,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate cluster here: the items are mostly unrelated docs, tests, CI, and small bugfix PRs. The soft edges are superficial theme matches only; none fix the same concrete code path.", - "confidence": 0.96, + "summary": "This set is highly heterogeneous: mostly unrelated docs, tests, and core code changes. I don\u2019t see any soft-edge pair that clearly reflects the same underlying bug or change, so I would reject all of them as duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR is a clear canonical representative; the items span unrelated subsystems and change types.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 2327 is the strongest standalone core bugfix-style change, but it is still not a duplicate representative for the rest of the set.", "soft_edge_verdicts": [ { "left": "pull_request:1945", "right": "pull_request:2327", "accept": false, - "reason": "Tests for UNet 2D blocks vs. a UNet input-length validation change; different underlying changes and scopes." + "reason": "Tests for UNet blocks vs. a UNet input-validation fix; related area, but not the same concrete change." }, { "left": "pull_request:1967", "right": "pull_request:3168", "accept": false, - "reason": "A runtime `torch.cuda.empty_cache()` tweak in Stable Diffusion vs. a docs-builder workflow change; unrelated." + "reason": "CPU offload example code vs. doc-builder workflow; unrelated." }, { "left": "pull_request:1982", "right": "pull_request:531", "accept": false, - "reason": "Both are test-only PRs, but they target different test files and different behaviors; not the same bug/change." + "reason": "Both are test tweaks, but they exercise different components and are not the same bug." }, { "left": "pull_request:2138", "right": "pull_request:4239", "accept": false, - "reason": "A docs typo fix vs. a DreamBooth LoRA text-encoder-related code change; no shared defect." + "reason": "A doc typo fix vs. a DreamBooth/text-encoder code change; unrelated." }, { "left": "pull_request:2271", "right": "pull_request:3400", "accept": false, - "reason": "Slow test expected-value updates vs. adding `omegaconf` dependencies; unrelated." + "reason": "Slow-test numeric updates vs. adding a dependency; unrelated." }, { "left": "pull_request:2322", "right": "pull_request:3262", "accept": false, - "reason": "A large inpaint example/update PR vs. a DeepFloyd IF rename/docs/API update; different pipelines and code paths." + "reason": "Large example/inpainting update vs. IF model renaming; different features." }, { "left": "pull_request:2345", "right": "pull_request:4187", "accept": false, - "reason": "xFormers training warning in example scripts vs. DPMSolver timestep indexing logic; unrelated fixes." + "reason": "xFormers warning in training scripts vs. scheduler timestep indexing; unrelated bugs." }, { "left": "pull_request:2398", "right": "pull_request:4187", "accept": false, - "reason": "Pix2Pix Zero docs addition vs. scheduler indexing logic; different artifacts and purposes." + "reason": "Pix2Pix Zero docs vs. DPMSolver scheduler internals; unrelated." }, { "left": "pull_request:2553", "right": "pull_request:555", "accept": false, - "reason": "Flax CFG flag support across Stable Diffusion pipelines vs. adding a Flax VAE model; not the same issue." + "reason": "Flax CFG flag work vs. adding Flax VAE support; different changes." }, { "left": "pull_request:2759", "right": "pull_request:774", "accept": false, - "reason": "A broad test-running/CI PR vs. a small test fix; not one underlying change." + "reason": "Broad test-run/CI changes vs. a specific test fix; not the same change." }, { "left": "pull_request:2759", "right": "pull_request:909", "accept": false, - "reason": "A broad CI/test orchestration PR vs. a specific inpaint test update; unrelated." + "reason": "Broad test-run/CI changes vs. a specific inpaint test fix; not the same change." }, { "left": "pull_request:3003", "right": "pull_request:3006", "accept": false, - "reason": "Both touch schedulers, but one is dynamic-thresholding behavior/docs and the other is variance handling in DDPM; different concrete bugs." + "reason": "Both touch schedulers, but one is dynamic thresholding and the other is variance clamping; different bugs." }, { "left": "pull_request:3292", "right": "pull_request:3536", "accept": false, - "reason": "Training input perturbation for text-to-image vs. DeepFloyd IF super-resolution PIL normalization; unrelated." + "reason": "Input perturbation training flag vs. IF super-resolution normalization; unrelated." }, { "left": "pull_request:3607", "right": "pull_request:5754", "accept": false, - "reason": "Attention/temb fix vs. style-only formatting cleanup; no shared bug." + "reason": "Attention implementation fix vs. formatting-only cleanup; unrelated." }, { "left": "pull_request:3637", "right": "pull_request:3916", "accept": false, - "reason": "Both are broad branch syncs, but they are not the same change set or underlying bug; too coarse to merge as duplicates." + "reason": "Both are branch-sync/meta PRs with broad diffs, but not the same underlying bug or change." }, { "left": "pull_request:3642", "right": "pull_request:5895", "accept": false, - "reason": "A community TensorRT inpainting pipeline vs. Stable Video Diffusion; different features entirely." + "reason": "TensorRT inpainting pipeline vs. Stable Video Diffusion addition; different features." }, { "left": "pull_request:3653", "right": "pull_request:5622", "accept": false, - "reason": "A lint-only cleanup vs. a large optimization/docs/examples PR; not the same fix." + "reason": "Lint cleanup vs. optimization/task-ops PR; unrelated." }, { "left": "pull_request:3766", "right": "pull_request:4887", "accept": false, - "reason": "Kandinsky UNet conversion script updates vs. removing a latent input from a Kandinsky prior pipeline; same family, but different concrete code paths and changes." + "reason": "Kandinsky conversion-script changes vs. removing a prior_emb2emb latent input; related project, different code paths." }, { "left": "pull_request:386", "right": "pull_request:390", "accept": false, - "reason": "Stable Diffusion pipeline docs vs. Apple Silicon MPS optimization docs; unrelated documentation topics." + "reason": "Stable Diffusion docs vs. MPS optimization docs; different documentation topics." }, { "left": "pull_request:388", "right": "pull_request:404", "accept": false, - "reason": "Unconditional latent diffusion docs vs. fp16 optimization docs; different documentation changes." + "reason": "Different docs pages and scopes; not the same change." }, { "left": "pull_request:388", "right": "pull_request:417", "accept": false, - "reason": "Unconditional latent diffusion docs vs. general inference pipeline docs; not the same artifact or change." + "reason": "Both are inference docs, but they cover different pipeline guides and rewrites." }, { "left": "pull_request:3957", "right": "pull_request:6405", "accept": false, - "reason": "Attention-name correction in model code vs. a docs-builder workflow tweak; unrelated." + "reason": "Attn-name correction work vs. doc-builder workflow testing; unrelated." }, { "left": "pull_request:397", "right": "pull_request:415", "accept": false, - "reason": "Quicktour docs vs. training docs; adjacent documentation, but not the same page or change." + "reason": "Quicktour docs vs. training docs; both docs, but not the same change." }, { "left": "pull_request:397", "right": "pull_request:418", "accept": false, - "reason": "Quicktour intro docs vs. API docs for DiffusionPipeline; different documentation sections and goals." + "reason": "Quicktour docs vs. DiffusionPipeline API docs; different documentation artifacts." }, { "left": "pull_request:402", "right": "pull_request:453", "accept": false, - "reason": "Intro section rework vs. a single installation typo fix; related docs area, but not the same underlying change." + "reason": "Intro-section reorg vs. a one-line installation typo fix; not the same underlying change." }, { "left": "pull_request:415", "right": "pull_request:418", "accept": false, - "reason": "Training docs vs. DiffusionPipeline API docs; different documentation targets." + "reason": "Training docs vs. DiffusionPipeline API docs; different scopes." }, { "left": "pull_request:4239", "right": "pull_request:6410", "accept": false, - "reason": "DreamBooth LoRA text_encoder fix vs. RL value-guided sampling module registration; unrelated code paths." + "reason": "DreamBooth text-encoder fixes vs. RL module registration; unrelated code paths." } ] }, @@ -147107,7 +147187,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 135, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -147196,10 +147276,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -147214,175 +147294,178 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated PRs; the soft links are driven by superficial title/file similarity (docs, tests, or the same broad subsystem) rather than the same concrete bug or change. I would not merge any of the soft-edge pairs.", - "confidence": 0.97, + "summary": "This cluster is mostly heterogeneous: it mixes unrelated docs, examples, tests, and core bugfix PRs. I only see one plausible near-duplicate pair in the HunyuanVideo attention-mask fixes; everything else should be kept separate.", + "confidence": 0.62, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items span unrelated fixes, docs, tests, utilities, schedulers, and pipeline changes, so there isn\u2019t a single underlying duplicate change to anchor the cluster.", + "canonical_pr_reason": "PR 10454 is the strongest single representative only because it has the clearest explicit bugfix/issue linkage in the one repeated code path (HunyuanVideo attention mask); the rest of the cluster does not form a coherent duplicate set.", "best_issue_reason": null, - "best_pr_reason": "No PR is a safe global representative because the cluster is heterogeneous and the apparent links are only superficial.", + "best_pr_reason": "PR 10454 is the best standalone PR to anchor this cluster: it is a concrete fix with an explicit issue target and a narrowly scoped code-path change.", "soft_edge_verdicts": [ { "left": "pull_request:4294", "right": "pull_request:5916", "accept": false, - "reason": "Both are rename-like changes, but they touch different files and unrelated semantics (SDXL rename/docs vs DreamBooth output_dir default)." + "reason": "Both are rename/cleanup docs or args changes, but they touch different SDXL vs DreamBooth surfaces and solve unrelated problems." }, { "left": "pull_request:4347", "right": "pull_request:5707", "accept": false, - "reason": "Different changes: one adds a step index counter in schedulers/pipeline flow, the other is LCM test additions." + "reason": "Both involve scheduler behavior, but one adds a step-index counter and the other adds LCM tests; not the same fix." }, { "left": "pull_request:5552", "right": "pull_request:6150", "accept": false, - "reason": "Same broad model codebase, but unrelated fixes: FreeU disable logic vs IP-Adapter conversion support." + "reason": "Both are core loader/model changes, but one fixes FreeU disable logic while the other adds IP-Adapter conversion support." }, { "left": "pull_request:6007", "right": "pull_request:6864", "accept": false, - "reason": "Both are docs, but they cover different topics and code paths; no shared underlying bug/change." + "reason": "Both are docs-only, but they cover unrelated topics: contribution guidance vs PEFT/torch.compile notes." }, { "left": "pull_request:6140", "right": "pull_request:6643", "accept": false, - "reason": "Both add API docs, but for different components (IP-Adapter vs UViT2D); not the same change." + "reason": "Both add docs entries, but IP-Adapter and UViT2D are different features with no shared bug/change." }, { "left": "pull_request:6215", "right": "pull_request:8386", "accept": false, - "reason": "Both are test-only changes, but they fix unrelated test setup/shape issues in different suites." + "reason": "Both adjust test initialization, but they target different models and different test harness behavior." }, { "left": "pull_request:6754", "right": "pull_request:8978", "accept": false, - "reason": "One is a diffusion-DPO loss fix; the other is a large DreamBooth/Imagic notebook and example update. Unrelated." + "reason": "Different research/example updates: diffusion DPO loss math vs DreamBooth notebook/example overhaul." }, { "left": "pull_request:6765", "right": "pull_request:8978", "accept": false, - "reason": "Path/cache utility cleanup vs example notebook updates; no common bug or code path." + "reason": "Cache-path utility cleanup and DreamBooth notebook/example rewrite are unrelated changes." }, { "left": "pull_request:6766", "right": "pull_request:8978", "accept": false, - "reason": "Constants path handling vs example notebook updates; unrelated changes." + "reason": "Constants path handling and DreamBooth notebook/example rewrite are unrelated." }, { "left": "pull_request:6767", "right": "pull_request:8978", "accept": false, - "reason": "Flax loading path fixes vs example notebook updates; different code paths." + "reason": "Flax model path handling and DreamBooth notebook/example rewrite are unrelated fixes." }, { "left": "pull_request:6768", "right": "pull_request:8978", "accept": false, - "reason": "Configuration path handling vs example notebook updates; unrelated." + "reason": "Configuration path handling and DreamBooth notebook/example rewrite are unrelated." }, { "left": "pull_request:6800", "right": "pull_request:8990", "accept": false, - "reason": "Import-utility availability helper vs adding a new PAG pipeline; different concerns entirely." + "reason": "Import availability plumbing and adding a new PAG ControlNet pipeline are different changes." }, { "left": "pull_request:7247", "right": "pull_request:8696", "accept": false, - "reason": "One fixes config-checking tests; the other marks flaky Marigold tests. Both are tests, but not the same failure." + "reason": "One fixes single-file config checking, the other marks Marigold tests flaky; different code paths and goals." }, { "left": "pull_request:7627", "right": "pull_request:9506", "accept": false, - "reason": "Different issues: Playground img2img latent handling vs documentation path/import cleanup." + "reason": "Both touch img2img-ish docs/code, but one fixes latent scaling in ControlNet/SDXL while the other is a docs import/path update." }, { "left": "pull_request:8322", "right": "pull_request:8572", "accept": false, - "reason": "Both are broad errata/docs cleanup, but they are distinct edits with different files and scopes." + "reason": "Both are broad errata/docs sweep PRs, but they cover different files and different cleanup sets." }, { "left": "pull_request:8478", "right": "pull_request:9620", "accept": false, - "reason": "Separate Hunyuan/ControlNet work: one adds a model/pipeline, the other adds interrupt support to ControlNet execution." + "reason": "HunyuanDiT ControlNet/pipeline addition is unrelated to ControlNet interrupt support." }, { "left": "pull_request:3160", "right": "pull_request:3476", "accept": false, - "reason": "Same feature family (Karras sigmas), but applied to different schedulers and not the same concrete bug/change." + "reason": "Both add Karras sigmas support, but they are separate schedulers and not the same concrete change." }, { "left": "pull_request:10454", "right": "pull_request:10482", - "accept": false, - "reason": "Both touch HunyuanVideo attention masking, but one fixes mask dimensional broadcasting while the other targets NaNs on older PyTorch; related area, different fixes." + "accept": true, + "reason": "Both modify the same HunyuanVideo attention-mask code path to fix masking-related failures; they are close enough to be merged into one PR." }, { "left": "pull_request:11516", "right": "pull_request:12021", "accept": false, - "reason": "LTX release/docs/pipeline integration vs attention-backend support for LTX; separate changes." + "reason": "LTX Video release/docs/model wiring is unrelated to LTX attention-backend support." }, { "left": "pull_request:1038", "right": "pull_request:1147", "accept": false, - "reason": "Both affect Flax example scripts, but one is a dtype fix and the other changes text-encoder loading from subfolder." + "reason": "Same Flax example area, but different bugs: dtype selection vs loading the text encoder from the correct subfolder." }, { "left": "pull_request:11469", "right": "pull_request:12530", "accept": false, - "reason": "Both are test xfails, but they target different pipelines and different version-related failures." + "reason": "Both add xfails, but they target different failing tests and different upstream version constraints." }, { "left": "pull_request:6484", "right": "pull_request:6999", "accept": false, - "reason": "Same Stable Video Diffusion file, but one fixes GPU generators and the other fixes tensor image preprocessing; distinct bugs." + "reason": "Same SVD pipeline file, but different bugs: GPU-generated tensors vs image-as-tensor preprocessing." }, { "left": "pull_request:6229", "right": "pull_request:7941", "accept": false, - "reason": "Unrelated test changes: adding a CI dependency vs marking a test class slow." + "reason": "Adding PEFT to CI and marking a single test class slow are unrelated test/CI changes." }, { "left": "pull_request:7632", "right": "pull_request:9601", "accept": false, - "reason": "Both concern DreamBooth training, but one fixes VAE loading path logic and the other changes VAE dtype handling." + "reason": "Both involve DreamBooth/SD3 training, but one fixes VAE path handling and the other makes VAE dtype conditional on mixed precision." }, { "left": "pull_request:11273", "right": "pull_request:11433", "accept": false, - "reason": "Notification bot timeout change vs test decorator rename/fix; no shared code path." + "reason": "Notification bot timeout cleanup is unrelated to a test decorator rename/fix." }, { "left": "pull_request:1527", "right": "pull_request:1913", "accept": false, - "reason": "Both involve image/mask handling in pipelines, but they address different pipelines and opposite concerns (supporting tensors vs input validation)." + "reason": "Both are pipeline input-handling changes, but stable diffusion inpaint tensor support and img2img input validation are different bugs." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, 10454/10482, is plausibly mergeable because both edit the same HunyuanVideo attention-mask code path; the rest of the pairwise rejections are consistent with distinct bugs/features rather than duplicates." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -147391,7 +147474,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 136, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -147479,10 +147562,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -147497,162 +147580,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose similarity group of mostly unrelated small PRs spanning docs, tests, CI, and pipeline fixes. None of the soft-edge pairs look like the same underlying bug or change, so no merge-worthy duplicate is present.", + "summary": "The candidate pairs are mostly superficial similarities across unrelated docs, tests, CI, and pipeline fixes. I don\u2019t see any pair that fixes the same concrete bug/change, so no duplicates are accepted and there is no meaningful canonical PR/issue for the cluster.", "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "pull_request:4855 is the strongest standalone PR in the set because it fixes a concrete core loading/offloading path across multiple pipelines, but it is still unrelated to the other items and should not be treated as a duplicate anchor.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:1554", "right": "pull_request:1968", "accept": false, - "reason": "Docs-only correction for UNet text shape vs a CPU offload behavior fix; different subsystem and bug." + "reason": "Docs-only correction for UNet text shapes vs a CPU offload behavior fix; different files and problems." }, { "left": "pull_request:3589", "right": "pull_request:5111", "accept": false, - "reason": "Community README note about scheduler quality vs adding generator support to K-DPM2/Karras code paths; not the same change." + "reason": "Reference-tricks README update vs adding karras sigmas to KDPM2 schedulers; unrelated changes." }, { "left": "pull_request:3798", "right": "pull_request:5948", "accept": false, - "reason": "Comment cleanup in training scripts vs a from_pretrained warning condition fix; unrelated." + "reason": "Comment/typo cleanup across training scripts vs a warning-condition fix in from_pretrained; not the same bug." }, { "left": "pull_request:4451", "right": "pull_request:6261", "accept": false, - "reason": "Moving tests to nightly is test scheduling, while 6261 refactors Attend-and-Excite pipeline imports; different intent and code path." + "reason": "Moving tests to nightly is a test-runner change; moving Attend-and-Excite out of stable_diffusion is a code refactor." }, { "left": "pull_request:4855", "right": "pull_request:5788", "accept": false, - "reason": "Offload-hook handling in loaders vs an LCM timestep validation error; no shared underlying bug." + "reason": "Offload-hook handling with textual inversion side loading vs an LCM timesteps error check; unrelated code paths." }, { "left": "pull_request:5946", "right": "pull_request:6448", "accept": false, - "reason": "Example test refactor vs LoRA test assertion fixes; both are tests, but not the same bug or change." + "reason": "Example-test refactor vs LoRA test assertion fixes; both test-related but different underlying issues." }, { "left": "pull_request:10086", "right": "pull_request:8467", "accept": false, - "reason": "Stable Diffusion 3 skip-guidance-layer batching bug vs SEGA device placement fix; different pipelines and failure modes." + "reason": "StableDiffusion3 skip-guidance batch-size bug vs SEGA device-placement fix; different pipelines and failures." }, { "left": "pull_request:10966", "right": "pull_request:11366", "accept": false, - "reason": "CogView4 input validation vs Hidream transformer output shape change; unrelated artifacts." + "reason": "CogView4 input validation vs Hidream transformer output-shape change; unrelated functionality." }, { "left": "pull_request:11089", "right": "pull_request:11841", "accept": false, - "reason": "Wan docs memory optimization update vs batch inference docs addition; distinct documentation changes." + "reason": "Wan docs memory-optimization content vs generic batch-generation docs; different documentation topics." }, { "left": "pull_request:11614", "right": "pull_request:12450", "accept": false, - "reason": "Docker CI security-related script fix vs broad CI installation cleanup; different workflows and problems." + "reason": "Docker CI security fix vs CI installation cleanup; both CI but not the same issue." }, { "left": "pull_request:1229", "right": "pull_request:152", "accept": false, - "reason": "Dreambooth dependency/install note vs adding local folder dataset support; not the same underlying feature or bug." + "reason": "DreamBooth README installation note vs local-folder training support; unrelated training features." }, { "left": "pull_request:12315", "right": "pull_request:12493", "accept": false, - "reason": "LoRA test simplification vs LTX 0.9.8 pipeline/docs/model update; unrelated." + "reason": "LoRA test simplification vs LTX-Video 0.9.8 model/docs updates; no shared bug or change." }, { "left": "pull_request:1468", "right": "pull_request:3298", "accept": false, - "reason": "Scheduler config serialization/trained betas vs arbitrary aspect ratio support in IF superresolution; different code paths." + "reason": "Saving trained betas in schedulers vs arbitrary aspect-ratio support in IF super-resolution; different subsystems." }, { "left": "pull_request:1913", "right": "pull_request:3536", "accept": false, - "reason": "Input type validation in img2img pipelines vs PIL normalization bug in IF superresolution; separate issues." + "reason": "Img2img input-type validation vs PIL normalization for IF super-resolution; separate bugs and pipelines." }, { "left": "pull_request:1945", "right": "pull_request:2706", "accept": false, - "reason": "New UNet block tests vs EMA save/restore during unconditional training; no shared bug." + "reason": "Adding UNet block tests vs preserving/restoring EMA weights in training; tests vs runtime training logic." }, { "left": "pull_request:2928", "right": "pull_request:3298", "accept": false, - "reason": "ControlNet local dataset loading vs IF superresolution aspect ratio work; unrelated training/pipeline changes." + "reason": "ControlNet local-dataset loading vs IF super-resolution aspect-ratio support; unrelated pipeline features." }, { "left": "pull_request:2968", "right": "pull_request:788", "accept": false, - "reason": "Post-processing indentation/branch fix vs PIL resampling warning fix; different img2img concerns." + "reason": "Post-processing indentation/control-flow fix vs PIL resampling deprecation warnings; different bug fixes." }, { "left": "pull_request:5035", "right": "pull_request:5792", "accept": false, - "reason": "Optional transformers import removal vs import-path cleanup in example scripts; unrelated maintenance." + "reason": "Dropping forced transformers dependency vs import-path fixes in examples; unrelated packaging/import issues." }, { "left": "pull_request:6432", "right": "pull_request:8510", "accept": false, - "reason": "README learning-rate edit vs scheduler math fix in advanced LoRA training; not the same concrete problem." + "reason": "README learning-rate example update vs LR scheduler math fixes in advanced training scripts; not the same change." }, { "left": "pull_request:6904", "right": "pull_request:8937", "accept": false, - "reason": "load_image PIL support vs AuraFlow long-prompt handling; unrelated functionality." + "reason": "load_image PIL support vs long-prompt handling in AuraFlow; unrelated code paths." }, { "left": "pull_request:7044", "right": "pull_request:8761", "accept": false, - "reason": "UNet refactor re-addition vs Hunyuan embedding refactor; both refactors, but different models and code paths." + "reason": "Re-adding UNet refactor plumbing vs Hunyuan-DiT embedding refactor; different model internals." }, { "left": "pull_request:7455", "right": "pull_request:8512", "accept": false, - "reason": "Remove distutils broadly vs deprecated mask helper removal; different maintenance tasks." + "reason": "Removing distutils dependencies vs removing deprecated prepare_mask_and_masked_image; unrelated cleanup changes." }, { "left": "pull_request:11243", "right": "pull_request:12315", "accept": false, - "reason": "Musubi Wan LoRA conversion support vs LoRA test scheduler simplification; one adds conversion code, the other edits tests." + "reason": "Musubi Wan LoRA conversion support vs LoRA test scheduler simplification; one is feature support, the other test maintenance." }, { "left": "pull_request:7068", "right": "pull_request:7277", "accept": false, - "reason": "Typos-only cleanup vs deprecated-parameter/test updates plus typos; overlapping theme, but not the same change." + "reason": "General typo fixes vs deprecated-parameter test updates plus typos; overlapping wording only, not the same underlying fix." }, { "left": "pull_request:2959", "right": "pull_request:3037", "accept": false, - "reason": "UNet3D test adjustment vs skipping an unstable MPS test; different test maintenance issues." + "reason": "UNet3D norm-group test adjustment vs skipping an unstable MPS test; separate test issues." } ] }, @@ -147668,7 +147751,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 137, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -147755,10 +147838,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -147773,150 +147856,150 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: most pairs are only loosely related by subsystem, docs, or adjacent feature work, but they do not fix the same concrete bug/change. No single PR stands out as a canonical representative.", - "confidence": 0.96, + "summary": "These PRs are mostly unrelated feature additions, docs updates, small fixes, and test/util changes across different subsystems; none of the soft edges look like true duplicates.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items span unrelated features, docs, tests, workflow, and bug-fix changes across different code paths.", + "canonical_pr_reason": "No single canonical PR for the cluster: the items span unrelated changes, so there isn't one duplicate source of truth.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "If one representative must be chosen, PR 8652 is the most substantial self-contained core feature addition with broad docs/tests and a clear scope, but it still does not duplicate the others.", "soft_edge_verdicts": [ { "left": "pull_request:3499", "right": "pull_request:3632", "accept": false, - "reason": "Both touch panorama, but one fixes a dpmsolver bug while the other adds view batching; different problems and merge scopes." + "reason": "Same panorama file, but 3499 fixes a dpmsolver-related bug while 3632 adds view batching/docs; different changes." }, { "left": "pull_request:10021", "right": "pull_request:10968", "accept": false, - "reason": "Both are LTX-related, but one adds the initial LTX Video support and the other upgrades to LTX 0.9.5 with a new condition pipeline; not the same change." + "reason": "Both are LTX-related, but 10021 is the initial LTX Video integration and 10968 is the later LTX 0.9.5 update with different components and behavior." }, { "left": "pull_request:10537", "right": "pull_request:11142", "accept": false, - "reason": "Both are Sana-related, but one is a Sana-4K pipeline update and the other is a Sana Sprint converter fix; separate code paths and goals." + "reason": "Both touch Sana assets, but one is a Sana-4K pipeline PR and the other is a Sana Sprint converter fix; not the same bug/change." }, { "left": "pull_request:2595", "right": "pull_request:3928", "accept": false, - "reason": "They touch schedulers/controlnet broadly, but the actual bugs are different: scheduler switching behavior vs controlnet timestep accounting." + "reason": "Shared scheduler/controlnet files are broad overlap, but 2595 is about scheduler config switching and 3928 fixes controlnet guidance range handling; not duplicates." }, { "left": "pull_request:6311", "right": "pull_request:6405", "accept": false, - "reason": "One adds a tutorial, the other tweaks a doc workflow; unrelated changes." + "reason": "One is a docs tutorial addition and the other is a doc-builder workflow tweak; unrelated." }, { "left": "pull_request:4408", "right": "pull_request:4819", "accept": false, - "reason": "Both relate to inpainting, but one is from_single_file SDXL support and the other is latent-input handling across inpainting pipelines; not the same bug." + "reason": "Different code paths: SDXL from-single-file support versus inpainting masked latent handling." }, { "left": "pull_request:12490", "right": "pull_request:13022", "accept": false, - "reason": "Completely unrelated: Audioldm2 test config vs Dockerfile dependency fixes." + "reason": "A test fixture tweak versus Dockerfile/runtime environment fixes; unrelated." }, { "left": "pull_request:10943", "right": "pull_request:10981", "accept": false, - "reason": "Both add LoRA support, but for different model families and separate implementations." + "reason": "Both add LoRA support, but for different model families (Wan vs CogView4) and different code paths." }, { "left": "pull_request:10326", "right": "pull_request:11252", "accept": false, - "reason": "Different maintenance fixes in unrelated files: workflow syntax vs timeout constant usage." + "reason": "Workflow quoting fix versus timeout constant change; unrelated." }, { "left": "pull_request:10478", "right": "pull_request:11434", "accept": false, - "reason": "Style cleanup in code vs a test import fix; not the same underlying issue." + "reason": "Style-only cleanup in pipelines versus a test import fix; not the same bug." }, { "left": "pull_request:2928", "right": "pull_request:3708", "accept": false, - "reason": "One allows custom local datasets for ControlNet training; the other fixes release-version sorting logic." + "reason": "ControlNet local dataset loading and custom releases version sorting are unrelated changes." }, { "left": "pull_request:382", "right": "pull_request:401", "accept": false, - "reason": "Both are docs, but for different pipelines and different content changes." + "reason": "Both are docs PRs, but for different pipelines and different content." }, { "left": "pull_request:4112", "right": "pull_request:5747", "accept": false, - "reason": "Dreambooth LoRA rank support and dataset data_dir handling are unrelated example improvements." + "reason": "Dreambooth LoRA rank support versus dataset loading parameter propagation; different features." }, { "left": "pull_request:5393", "right": "pull_request:5407", "accept": false, - "reason": "Docs/example modernization vs mixed-precision training fix; different concerns." + "reason": "Examples/docs HFApi update versus mixed-precision training fix in a different example; not duplicates." }, { "left": "pull_request:5952", "right": "pull_request:6065", "accept": false, - "reason": "Both update pipeline docs, but they cover different pipelines and different documentation additions." + "reason": "Pipeline overview doc update versus SDXL Turbo docs; related area but different content and scope." }, { "left": "pull_request:8618", "right": "pull_request:8722", "accept": false, - "reason": "A tiny autoencoder shift-factor change is unrelated to the embeddings refactor." + "reason": "SD3 tiny autoencoder shift_factor addition versus a large embeddings refactor; not the same issue." }, { "left": "pull_request:10140", "right": "pull_request:10430", "accept": false, - "reason": "ConsisID implementation and TeaCache are separate features with no shared bug/change." + "reason": "ConsisID model/pipeline integration versus TeaCache hook infrastructure; unrelated." }, { "left": "pull_request:6308", "right": "pull_request:7461", "accept": false, - "reason": "A LoRA test fixture fix and a testing-utils boolean parser fix are unrelated." + "reason": "LoRA dummy component test changes versus a utility bool parser fix; unrelated." }, { "left": "pull_request:7247", "right": "pull_request:7771", "accept": false, - "reason": "Config-checking test expectations vs flaky test marking; different test issues." + "reason": "Config checking test expectations versus marking a different test flaky; not duplicates." }, { "left": "pull_request:10454", "right": "pull_request:10837", "accept": false, - "reason": "One fixes a Hunyuan Video attention mask dimension bug; the other adds SkyReels Hunyuan models and docs." + "reason": "Hunyuan attention mask dimension fix versus SkyReels Hunyuan model/pipeline addition; different change intent." }, { "left": "pull_request:10579", "right": "pull_request:9918", "accept": false, - "reason": "Separate examples: attentive eraser community pipeline vs a server example." + "reason": "A community example pipeline versus a server example; different features." }, { "left": "pull_request:11052", "right": "pull_request:11089", "accept": false, - "reason": "A CUDA test compatibility decorator and Wan docs memory-optimization notes are unrelated." + "reason": "Testing utility decorator addition versus Wan documentation updates; unrelated." }, { "left": "pull_request:11091", "right": "pull_request:12086", "accept": false, - "reason": "Both mention Flux, but one is remote encode support and the other is modular I2I support; different code paths and features." + "reason": "Flux remote encode support and modular Flux I2I are both Flux-related but different implementations and problems." } ] }, @@ -147932,7 +148015,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 138, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -148021,10 +148104,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, @@ -148039,168 +148122,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a set of false-positive similarity matches: most items are unrelated PRs spanning docs, tests, workflow, dependency pins, and distinct bug fixes. None of the soft-edge pairs look like the same underlying change or bug, so I would not merge any pair.", - "confidence": 0.97, + "summary": "These are mostly unrelated pull requests that happen to cluster around tests, docs, loaders, and scheduler fixes; there is no clear duplicate issue/PR group here. I would treat them as separate artifacts rather than one underlying change set.", + "confidence": 0.19, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "If one PR must represent the cluster, PR #12005 is the cleanest standalone bugfix: it fixes sharded checkpoint loading to respect `local_files_only`, with a narrow and concrete code path.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR #12005 is the best single PR candidate because it addresses a specific user-facing bug, has a tight scope, and is clearly distinct from the mostly test/doc-only or unrelated maintenance changes in the cluster.", "soft_edge_verdicts": [ { "left": "pull_request:12005", "right": "pull_request:12420", "accept": false, - "reason": "Different problems: sharded-checkpoint offline behavior vs conditional torch.distributed import." + "reason": "Different bugs: sharded checkpoint handling vs conditional import of `torch.distributed`; same broad area only." }, { "left": "pull_request:12137", "right": "pull_request:13081", "accept": false, - "reason": "Unrelated changes: parallel-loading flag refactor vs Qwen hidden-state contiguity for torchao." + "reason": "Unrelated fixes: constants/parallel loading flag vs Qwen hidden-state contiguity for torchao." }, { "left": "pull_request:13120", "right": "pull_request:13150", "accept": false, - "reason": "Both touch tests, but one adds AutoModel load-id coverage and the other changes recompile_limit handling." + "reason": "Both are test changes, but they target different behaviors and code paths." }, { "left": "pull_request:13120", "right": "pull_request:13472", "accept": false, - "reason": "Tests for modular loading are unrelated to adding float64 gotchas to docs." + "reason": "A test helper tweak and a docs-only gotcha addition are not the same underlying change." }, { "left": "pull_request:1468", "right": "pull_request:2892", "accept": false, - "reason": "Saving trained betas in schedulers/configs is unrelated to upscaler input/embed validation." + "reason": "Separate functional areas: saving trained betas vs upscaler input validation/embeds." }, { "left": "pull_request:152", "right": "pull_request:788", "accept": false, - "reason": "Local-folder dataset support is unrelated to PIL resampling warning fixes in image pipelines." + "reason": "Local image-folder support is unrelated to PIL resampling warning fixes." }, { "left": "pull_request:1525", "right": "pull_request:51", "accept": false, - "reason": "Different eras and issues: upscaler batch handling vs setup.py install_requires typo." + "reason": "Different scope and era: upscaler batch-size bugfix vs packaging metadata typo." }, { "left": "pull_request:1554", "right": "pull_request:2655", "accept": false, - "reason": "Docs corrections are not the same as LoRA save/load behavior changes." + "reason": "Docs correction vs LoRA loading/saving behavior are not the same change." }, { "left": "pull_request:1554", "right": "pull_request:402", "accept": false, - "reason": "Both are docs PRs, but they update different documentation sections with no shared bug or feature." + "reason": "Both touch docs, but they address different documentation reorganizations." }, { "left": "pull_request:1920", "right": "pull_request:402", "accept": false, - "reason": "Both are docs, but one adds TRANSLATING.md and the other reorganizes the intro docs tree." + "reason": "Both are docs PRs, but one adds translating guidance and the other reorganizes the intro docs tree." }, { "left": "pull_request:1937", "right": "pull_request:3536", "accept": false, - "reason": "Stable diffusion EMA extraction logic is unrelated to DeepFloyd IF PIL normalization." + "reason": "EMA-weight extraction logic is unrelated to IF super-resolution image normalization." }, { "left": "pull_request:2693", "right": "pull_request:3708", "accept": false, - "reason": "A doc image-link fix is unrelated to packaging/version sorting for custom releases." + "reason": "A doc image link fix is unrelated to version sorting for custom releases." }, { "left": "pull_request:2928", "right": "pull_request:3657", "accept": false, - "reason": "Both mention controlnet, but one adds local dataset loading and the other adds checkpoint-argument parsing." + "reason": "Both mention ControlNet, but one adds local dataset loading and the other tweaks checkpoint parser arguments." }, { "left": "pull_request:2970", "right": "pull_request:4396", "accept": false, - "reason": "Variable rename in a DreamBooth script is unrelated to allowing encode_prompt without a text-encoder instance variable." + "reason": "Variable rename in a DreamBooth script vs prompt-encoding behavior across pipelines are different bugs." }, { "left": "pull_request:2983", "right": "pull_request:4963", "accept": false, - "reason": "Test-dependency installation docs/workflow updates are different from installing accelerate main and related loader/workflow fixes." + "reason": "Installing test dependencies is separate from pinning accelerate/main in workflows." }, { "left": "pull_request:3054", "right": "pull_request:3321", "accept": false, - "reason": "Different documentation topics; no shared underlying change." + "reason": "Different docs pages and different content updates; not a duplicate change." }, { "left": "pull_request:3292", "right": "pull_request:4887", "accept": false, - "reason": "Input perturbation in text-to-image training is unrelated to removing latents from Kandinsky prior emb2emb." + "reason": "Input perturbation training option vs removing latent input from a different pipeline are unrelated." }, { "left": "pull_request:3321", "right": "pull_request:5407", "accept": false, - "reason": "Controlling-generation docs updates are unrelated to mixed-precision fixes in unconditional training." + "reason": "Docs update vs mixed-precision training fix; same repo area only by coincidence." }, { "left": "pull_request:3397", "right": "pull_request:3746", "accept": false, - "reason": "A documentation note on local paths is not the same as a code fix for os.path.join handling." + "reason": "Documentation note on local paths vs code fix for `from_ckpt` path joining are distinct." }, { "left": "pull_request:3561", "right": "pull_request:4112", "accept": false, - "reason": "Both touch DreamBooth code, but they solve different issues: IF stage-II upscaling vs adding LoRA rank plumbing." + "reason": "Both touch DreamBooth LoRA files, but one is IF stage-II upscaling and the other adds rank handling." }, { "left": "pull_request:4787", "right": "pull_request:5946", "accept": false, - "reason": "LoRA loading relaxation is unrelated to refactoring example tests." + "reason": "LoRA checkpoint filename filtering is unrelated to refactoring example tests." }, { "left": "pull_request:5298", "right": "pull_request:5331", "accept": false, - "reason": "Different LoRA code paths: compile support for conv layers vs replacing deprecated attention processors." + "reason": "Torch.compile fix for LoRA conv is different from switching DreamBooth LoRA training to LoRALinear." }, { "left": "pull_request:8563", "right": "pull_request:9721", "accept": false, - "reason": "These are unrelated dependency pins: accelerate/transformers version bumps vs torch version pinning in Dockerfiles." + "reason": "Version pins in Python deps are unrelated to Docker torch version pins." }, { "left": "pull_request:1215", "right": "pull_request:530", "accept": false, - "reason": "DDIM generator/device fix is unrelated to flaky img2img test assertions." + "reason": "Different pipelines and different failure modes: DDIM CPU generator handling vs img2img test flakiness." }, { "left": "pull_request:3001", "right": "pull_request:3476", "accept": false, - "reason": "Both add Karras-related scheduler support, but they target different scheduler classes and code paths." + "reason": "Related scheduler family, but not the same concrete change or bug." }, { "left": "pull_request:11979", "right": "pull_request:12736", "accept": false, - "reason": "Compilation test version guards are unrelated to enabling regional compilation on Z-Image." + "reason": "Torch version gating in tests is unrelated to enabling regional compilation on a model." } ] }, @@ -148216,7 +148299,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 139, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -148303,10 +148386,10 @@ "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, @@ -148321,162 +148404,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a heterogeneous mix of unrelated PRs spanning docs, examples, loaders, schedulers, Flux/Wan/ControlNet, and attention backends. The soft pairs only share broad subsystem labels or filenames; none appear to be the same underlying bug or change, so no merge should be accepted.", - "confidence": 0.99, + "summary": "This cluster is a loose similarity bucket, not a true duplicate set. The PRs span unrelated docs, training scripts, loaders, attention backends, Flux/Wan work, and workflow chores; none of the soft edges look like the same underlying change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a true duplicate canonical here: the items cover distinct fixes/features across unrelated code paths and documentation changes.", + "canonical_pr_reason": "No single PR is a safe canonical choice because the grouped PRs do not share one concrete bugfix or feature; they only share broad subsystem keywords or filename overlap.", "best_issue_reason": null, - "best_pr_reason": "No PR cleanly represents the cluster as a whole; the pairwise similarities are superficial and the changes do not collapse into one concrete fix.", + "best_pr_reason": "No PR is a good global representative for duplicate triage: the cluster is heterogeneous and the soft matches are mostly false positives.", "soft_edge_verdicts": [ { "left": "pull_request:5559", "right": "pull_request:6114", "accept": false, - "reason": "Both are img2img-related, but one adds a new LCM img2img pipeline while the other adds img2img/inpaint support to lpw-xl community code; different artifacts and code paths." + "reason": "Both are pipeline-related, but one adds Latent Consistency img2img and the other adds lpw-xl img2img/inpaint support; different changes and code paths." }, { "left": "pull_request:12812", "right": "pull_request:13280", "accept": false, - "reason": "Both touch attention backend code, but 12812 adds CP support for kernels-based backends while 13280 adds FA4 backend support/docs; separate backend features." + "reason": "They both touch attention dispatch, but one adds CP wrapping for kernels-based backends while the other adds FA4 support/docs; not the same fix." }, { "left": "pull_request:12419", "right": "pull_request:13169", "accept": false, - "reason": "Flux modular-pipeline metadata changes versus Flux Klein PEFT LoRA format support; unrelated functionality." + "reason": "Flux modular-pipeline mellon prep and Flux Klein PEFT LoRA format support are separate concerns with different code paths." }, { "left": "pull_request:9175", "right": "pull_request:9638", "accept": false, - "reason": "Both are Flux ControlNet work, but one adds Union ControlNet and the other adds Xlabs ControlNet support; different model variants." + "reason": "Both are Flux ControlNet related, but one introduces Flux Union ControlNet support and the other adds Xlabs ControlNets; not mergeable as one duplicate change." }, { "left": "pull_request:10200", "right": "pull_request:9307", "accept": false, - "reason": "StableDiffusionXLControlNetUnion refactor versus rotary embedding CPU-placement fix; no shared bug or change." + "reason": "ControlNet Union refactor and rotary embedding CPU fix are unrelated changes in different parts of the library." }, { "left": "pull_request:10704", "right": "pull_request:11455", "accept": false, - "reason": "Docs fix for img2img versus SDXL training script interpolation-mode change; unrelated scopes." + "reason": "Docs for img2img/upscaler usage versus a training-script interpolation default change; same theme, not the same underlying artifact." }, { "left": "pull_request:10704", "right": "pull_request:11465", "accept": false, - "reason": "Docs fix for img2img versus LoRA SDXL training interpolation change; same broad topic, but not the same concrete change." + "reason": "The docs PR and the LoRA training-script PR address different files and behaviors, so they are not duplicates." }, { "left": "pull_request:10850", "right": "pull_request:12092", "accept": false, - "reason": "Hook helper utilities versus GGUF checkpoint docs; only superficial similarity." + "reason": "Hook utility methods and GGUF docs are unrelated; no shared bug or code path." }, { "left": "pull_request:11254", "right": "pull_request:9696", "accept": false, - "reason": "ConsisID import/dependency wiring versus CI workflow input validation; unrelated changes." + "reason": "ConsisID import fixes and the test-workflow input rename are unrelated maintenance changes." }, { "left": "pull_request:11916", "right": "pull_request:12236", "accept": false, - "reason": "Both involve attention_dispatch.py, but 11916 is a broad Flux/Chroma refactor and 12236 is specifically hub-kernels FA3 support." + "reason": "Shared attention_dispatch.py overlap, but one is a large refactor and the other adds kernels-backed FLASH_3 hub support; different concrete goals." }, { "left": "pull_request:130", "right": "pull_request:1513", "accept": false, - "reason": "Both mention accelerate config in docs, but they update different example READMEs years apart and are not the same change." + "reason": "Both are documentation-only, but one adds accelerate config to examples README and the other adds notebook config guidance for dreambooth; not the same document change." }, { "left": "pull_request:1525", "right": "pull_request:3561", "accept": false, - "reason": "Upscaling batch-size bugfix versus DreamBooth IF stage II training logic; different pipelines and fixes." + "reason": "Batch-size fix in upscaling versus Dreambooth IF stage-II upscaling training changes; different pipelines and bugs." }, { "left": "pull_request:2314", "right": "pull_request:689", "accept": false, - "reason": "One corrects fast-test expectations, the other swaps an attention implementation for performance; not the same underlying issue." + "reason": "One adjusts fast-test expectations, the other changes attention math for performance; not the same issue." }, { "left": "pull_request:2968", "right": "pull_request:748", "accept": false, - "reason": "Post-processing bugfix in img2img pipelines versus push_to_hub call fix in training scripts; unrelated." + "reason": "Post-processing bugfixes and push_to_hub call fixes are unrelated." }, { "left": "pull_request:388", "right": "pull_request:390", "accept": false, - "reason": "Two separate docs additions for different topics: unconditional latent diffusion and Apple Silicon MPS usage." + "reason": "These are separate docs over different topics (latent diffusion vs MPS optimization), with no shared underlying change." }, { "left": "pull_request:4262", "right": "pull_request:6450", "accept": false, - "reason": "SDXL DreamBooth LoRA training fixes versus LoRA deprecation test cleanup; different code paths and intent." + "reason": "SDXL DreamBooth LoRA fixes are unrelated to LoRA deprecation test cleanup." }, { "left": "pull_request:4347", "right": "pull_request:6305", "accept": false, - "reason": "Both scheduler-related, but one adds a step-index counter while the other adds zero-terminal-SNR beta rescaling; distinct changes." + "reason": "A scheduler step-index counter and zero-terminal-SNR beta rescaling are distinct scheduler features." }, { "left": "pull_request:6140", "right": "pull_request:7319", "accept": false, - "reason": "Both are docs-only, but one adds IP-Adapter docs and the other adds EDM scheduler docs; unrelated content." + "reason": "Adding IP-Adapter docs and adding EDM scheduler docs are separate documentation additions." }, { "left": "pull_request:6450", "right": "pull_request:7007", "accept": false, - "reason": "LoRA deprecation test adjustments versus the actual LoRAAttnProcessor deprecation change; related area, but not the same fix." + "reason": "Test renaming/cleanup versus LoRA attention-processor deprecation are different changes." }, { "left": "pull_request:6886", "right": "pull_request:6994", "accept": false, - "reason": "Both touch dtype handling, but one removes `torch_dtype` from `.to()` calls while the other passes `torch_dtype` into single-file loading; opposite directions and separate bugs." + "reason": "One removes torch_dtype from to(), the other passes torch_dtype into single-file loaders; related API area, but not the same fix." }, { "left": "pull_request:7115", "right": "pull_request:7652", "accept": false, - "reason": "Dockerfile package-install change versus GitHub Actions buildx workflow change; unrelated maintenance." + "reason": "Dockerfile package-install changes and workflow buildx enforcement are unrelated infrastructure updates." }, { "left": "pull_request:8600", "right": "pull_request:8860", "accept": false, - "reason": "DreamBooth SD3 script updates versus a README-only update; different artifact and no shared bug." + "reason": "The SD3 DreamBooth update and the README_sd3.md-only PR do not fix the same concrete bug or change." }, { "left": "pull_request:12004", "right": "pull_request:12074", "accept": false, - "reason": "Both are Wan 2.2 related, but one adds core model/pipeline support and the other adds second-transformer LoRA loading; separate capabilities." + "reason": "Wan 2.2 core model support and Wan 2.2 LoRA loading support are related project work, but they are separate features and not one duplicate PR." }, { "left": "pull_request:11995", "right": "pull_request:12419", "accept": false, - "reason": "Flux modular pipeline implementation versus Flux MellON readiness metadata; not the same underlying change." + "reason": "Both are Flux modular-pipeline work, but one adds modular Flux text-to-image and the other prepares Flux for mellon; different concrete objectives." }, { "left": "pull_request:10354", "right": "pull_request:9676", "accept": false, - "reason": "Flux IP-Adapter tester fix versus an import-path cleanup in benchmarks; unrelated." + "reason": "Flux IP-adapter tester fix and a benchmark import cleanup are unrelated." } ] }, @@ -148492,7 +148575,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 140, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -148500,6 +148583,7 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10086", "pull_request:10127", "pull_request:10342", "pull_request:10351", @@ -148525,10 +148609,8 @@ "pull_request:2286", "pull_request:2452", "pull_request:3313", - "pull_request:3691", "pull_request:3797", "pull_request:4242", - "pull_request:4862", "pull_request:5213", "pull_request:5361", "pull_request:5653", @@ -148542,6 +148624,8 @@ "pull_request:7771", "pull_request:7825", "pull_request:788", + "pull_request:8452", + "pull_request:8510", "pull_request:8696", "pull_request:8797", "pull_request:9060", @@ -148559,7 +148643,6 @@ "pull_request:11252|pull_request:9676", "pull_request:1527|pull_request:788", "pull_request:1968|pull_request:3797", - "pull_request:3691|pull_request:4862", "pull_request:4242|pull_request:5213", "pull_request:5361|pull_request:5852", "pull_request:6330|pull_request:7714", @@ -148574,192 +148657,200 @@ "pull_request:5653|pull_request:5915", "pull_request:7246|pull_request:8696", "pull_request:1355|pull_request:1571", - "pull_request:10351|pull_request:11716" + "pull_request:10351|pull_request:11716", + "pull_request:10086|pull_request:8452", + "pull_request:10086|pull_request:8510" ], - "prepared_review_unit_hash": "b1ca12cea11cf0cbc5baaa7b0c8680ed97e43db2d6a96e212a826edc346f8eba", + "prepared_review_unit_hash": "9ee042164eb2a121d77a596c836eaa614ecd2a020079119b5a7ca356a10f7e4c", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 26, - "serialized_chars": 101456, - "estimated_input_tokens": 25364, - "estimated_eval_tokens": 50984 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 103654, + "estimated_input_tokens": 25914, + "estimated_eval_tokens": 52084 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous collection of unrelated pull requests across tests, docs, loaders, models, Docker, and pipelines. The soft-similarity links mostly reflect broad subsystem overlap or similar maintenance patterns, not the same underlying bug or change, so no duplicate cluster emerges.", - "confidence": 0.93, + "summary": "Mostly false positives: the cluster mixes unrelated bug fixes, feature additions, docs, and test-only tweaks that merely share broad subsystems or filenames. None of the soft pairs look like the same concrete change.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR; the items span unrelated fixes across loaders, pipelines, tests, docs, and infra.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 11039 is the strongest representative because it is a substantive core loader fix touching multiple code paths, unlike most others here which are test/doc/utility-only or clearly different bugs.", "soft_edge_verdicts": [ { "left": "pull_request:7771", "right": "pull_request:7825", "accept": false, - "reason": "Both are test-only fixes in different model areas; no shared bug or code path." + "reason": "Different test issues: one marks a flaky UNetControlNetXS test, the other skips offload/device-map tests for models without split modules." }, { "left": "pull_request:1128", "right": "pull_request:3313", "accept": false, - "reason": "Both touch inpaint-related code, but one adds tensor inputs and the other fixes torch compile issues; different problems." + "reason": "One adds tensor input support for inpaint; the other fixes torch.compile breaks across ControlNet/UNet code paths." }, { "left": "pull_request:2129", "right": "pull_request:2286", "accept": false, - "reason": "Both modify cross-attention/LoRA code, but they target different compatibility issues and code paths." + "reason": "Related LoRA/xformers area, but different changes: LoRA pipeline support vs making LoRA work with xformers attention processors." }, { "left": "pull_request:12806", "right": "pull_request:12812", "accept": false, - "reason": "Both are distributed/attention infrastructure changes, but they solve different backend concerns." + "reason": "Different problems: distributed progress/logging handling vs enabling CP for kernel-based attention backends." }, { "left": "pull_request:11639", "right": "pull_request:11726", "accept": false, - "reason": "Both relate to Wan LoRA/video work, but one standardizes VAE sampling and the other fixes LoRA conversion metadata." + "reason": "Different Wan fixes: VAE sampling-mode standardization vs LoRA conversion/loader support." }, { "left": "pull_request:10342", "right": "pull_request:9060", "accept": false, - "reason": "Different fixes in unrelated modules; no common bug or mergeable combined PR." + "reason": "Unrelated: dtype inference logic vs removing a stray debug print in transformer_flux." }, { "left": "pull_request:10351", "right": "pull_request:11252", "accept": false, - "reason": "A LoRA test tweak and a timeout-constant change are unrelated." + "reason": "Different scope entirely: a LoRA test helper tweak vs fixing a timeout constant in a release-branch utility script." }, { "left": "pull_request:11252", "right": "pull_request:9676", "accept": false, - "reason": "Both are small utility fixes, but they address different scripts and failure modes." + "reason": "Both are small maintenance fixes, but they address unrelated code paths: request timeout vs benchmark import path." }, { "left": "pull_request:1527", "right": "pull_request:788", "accept": false, - "reason": "Same inpaint area, but one is an implementation bugfix and the other is a docs/warnings-related change." + "reason": "Inpaint tensor-input support vs PIL sampler warning cleanup in img2img/textual-inversion; not the same bug." }, { "left": "pull_request:1968", "right": "pull_request:3797", "accept": false, - "reason": "Both affect core model behavior, but one is CPU offload docs/behavior and the other is attention-naming cleanup." - }, - { - "left": "pull_request:3691", - "right": "pull_request:4862", - "accept": false, - "reason": "Different model-loader/model-pipeline issues; no shared code path." + "reason": "Docs/CPU-offload corrections versus a core attention naming refactor; different underlying issues." }, { "left": "pull_request:4242", "right": "pull_request:5213", "accept": false, - "reason": "Both mention VAE loading/scaling, but they fix different loader and latent-scaling behaviors." + "reason": "Different VAE-related bugs: single-file VAE loading support vs scaling-factor handling in latent diffusion." }, { "left": "pull_request:5361", "right": "pull_request:5852", "accept": false, - "reason": "Both are PEFT-related workflow/test changes, not the same bug or feature." + "reason": "PR test workflow fix is unrelated to the PEFT integration test harness changes." }, { "left": "pull_request:6330", "right": "pull_request:7714", "accept": false, - "reason": "Both are documentation PRs with unrelated content." + "reason": "One fixes SVD docs rendering; the other is an AutoPipeline documentation update." }, { "left": "pull_request:7632", "right": "pull_request:9506", "accept": false, - "reason": "Both touch docs/examples, but they address different path/import issues." + "reason": "DreamBooth VAE loading path fix is unrelated to docs/link cleanup and imageio import changes." }, { "left": "pull_request:11039", "right": "pull_request:11716", "accept": false, - "reason": "Both touch LoRA loading, but one changes torch_dtype handling and the other fixes metadata return behavior." + "reason": "Both involve loaders, but one changes single-file torch_dtype/quantization behavior while the other fixes LoRA metadata preparation." }, { "left": "pull_request:2452", "right": "pull_request:750", "accept": false, - "reason": "A draft docs page and an old scheduler compatibility fix are unrelated." + "reason": "Draft docs page addition and LMS timestep back-compat are unrelated." }, { "left": "pull_request:12041", "right": "pull_request:12521", "accept": false, - "reason": "Both modify Wan autoencoder code, but one fixes patchify behavior and the other fixes tiling crash logic." + "reason": "Same model family, but different concrete Wan autoencoder bugs: patchify logic vs tiled decode crash handling." }, { "left": "pull_request:10127", "right": "pull_request:8696", "accept": false, - "reason": "Test xfails for SD configs and Marigold flaky markers are unrelated test maintenance." + "reason": "One xfails incompatible SD config tests; the other marks Marigold depth tests flaky. Different test failures." }, { "left": "pull_request:10127", "right": "pull_request:8797", "accept": false, - "reason": "Different test suites and different instability fixes; not the same defect." + "reason": "Unrelated test changes: incompatible SD config xfails vs sharding test fixes." }, { "left": "pull_request:12424", "right": "pull_request:13022", "accept": false, - "reason": "Both are Dockerfile hygiene changes, but they target different image definitions and environments." + "reason": "Both touch Dockerfiles, but they are different infra cleanups with different images and fixes." }, { "left": "pull_request:1283", "right": "pull_request:1299", "accept": false, - "reason": "Both add new pipelines/docs, but they introduce different models and codebases." + "reason": "Distinct features: adding Versatile Diffusion versus adding AltDiffusion." }, { "left": "pull_request:5653", "right": "pull_request:5915", "accept": false, - "reason": "Both involve PEFT/Lora-related infra, but one changes model unwrapping and the other adds IP-Adapter Plus support." + "reason": "PEFT-aware maybe_raise_or_warn support is unrelated to IP-Adapter Plus model support." }, { "left": "pull_request:7246", "right": "pull_request:8696", "accept": false, - "reason": "Unrelated test fixes in different areas." + "reason": "VAE tiling test device fix and Marigold flaky-test annotations are unrelated." }, { "left": "pull_request:1355", "right": "pull_request:1571", "accept": false, - "reason": "Different pipeline additions/fixes; one adds Flax img2img, the other changes image resizing rules." + "reason": "Flax img2img pipeline addition versus SD Img2Img resize-to-multiple-of-8 bug fix are different changes." }, { "left": "pull_request:10351", "right": "pull_request:11716", "accept": false, - "reason": "A test-input cleanup and a LoRA metadata fix are unrelated." + "reason": "Test cleanup in LoRA utils is unrelated to LoRA loader metadata handling." + }, + { + "left": "pull_request:10086", + "right": "pull_request:8452", + "accept": false, + "reason": "StableDiffusion3 guidance-layer fix and single-file loading strictness are unrelated bugs." + }, + { + "left": "pull_request:10086", + "right": "pull_request:8510", + "accept": false, + "reason": "Different fixes: SD3 skip-guidance-layer handling versus DreamBooth LR scheduler logic." } ] }, @@ -148775,7 +148866,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 141, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -148783,7 +148874,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10086", "pull_request:10271", "pull_request:10360", "pull_request:10547", @@ -148826,15 +148916,13 @@ "pull_request:5053", "pull_request:5111", "pull_request:5665", + "pull_request:6261", "pull_request:676", + "pull_request:7044", "pull_request:833", - "pull_request:8452", - "pull_request:8510", "pull_request:9918" ], "soft_pairs": [ - "pull_request:10086|pull_request:8452", - "pull_request:10086|pull_request:8510", "pull_request:10271|pull_request:10547", "pull_request:10360|pull_request:10547", "pull_request:11269|pull_request:9918", @@ -148858,192 +148946,187 @@ "pull_request:3657|pull_request:5665", "pull_request:3806|pull_request:5111", "pull_request:4040|pull_request:4248", - "pull_request:4248|pull_request:5665" + "pull_request:4248|pull_request:5665", + "pull_request:6261|pull_request:7044" ], - "prepared_review_unit_hash": "9e01118c249ad86c8f927056b594b8ecd1f9a4796f9f4d2bf8469d34cb199ce4", + "prepared_review_unit_hash": "82926a2cf352975562b6a2383dc66e6f7a80edb53d7de14bdb0b3b67938f7ca9", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 26, - "serialized_chars": 97846, - "estimated_input_tokens": 24462, - "estimated_eval_tokens": 49180 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 96561, + "estimated_input_tokens": 24141, + "estimated_eval_tokens": 48538 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a true duplicate cluster: the items span unrelated docs, CI, test refactors, dependency bumps, and distinct code fixes/features. No single canonical issue/PR emerges.", - "confidence": 0.99, + "summary": "These PRs are mostly unrelated, with only superficial overlap from shared subsystems like docs, CI, LoRA, or schedulers. I don\u2019t see any pair that clearly fixes the same concrete bug/change.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the PRs address different subsystems and change types, with no shared concrete bug or mergeable change target.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No global best PR for deduplication; the set is heterogeneous and the soft edges are only superficial similarity.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:10086", - "right": "pull_request:8452", - "accept": false, - "reason": "Different problems: SD3 skip-guidance/num_images_per_prompt vs single-file loading strictness." - }, - { - "left": "pull_request:10086", - "right": "pull_request:8510", - "accept": false, - "reason": "Unrelated: SD3 pipeline guidance-layer batching bug vs dreambooth LR scheduler math." - }, { "left": "pull_request:10271", "right": "pull_request:10547", "accept": false, - "reason": "Different test refactors for unrelated models and offload behavior." + "reason": "Both touch tests, but they exercise different features: TorchAO serialization refactoring vs offload test parameters for CogVideoX/CogView3." }, { "left": "pull_request:10360", "right": "pull_request:10547", "accept": false, - "reason": "Release workflow TestPyPI URL fix vs offload-test adjustments; no shared bug." + "reason": "Release workflow TestPyPI install fix is unrelated to model offload test tweaks." }, { "left": "pull_request:11269", "right": "pull_request:9918", "accept": false, - "reason": "ONNXRuntime backend availability handling vs adding a server example." + "reason": "ONNX Runtime package detection changes and the server example are unrelated changes." }, { "left": "pull_request:11292", "right": "pull_request:11999", "accept": false, - "reason": "LoRA state-dict conversion cleanup vs PEFT config injection API adaptation." + "reason": "Both are LoRA-related, but one expands SDXL LoRA mapping support while the other adapts to a new PEFT injection API; different code paths and fixes." }, { "left": "pull_request:12062", "right": "pull_request:12491", "accept": false, - "reason": "CI GPU allocation in workflows vs adding doc-link checks; unrelated changes." + "reason": "CI GPU container settings and documentation link checking are separate workflow changes." }, { "left": "pull_request:12092", "right": "pull_request:12331", "accept": false, - "reason": "GGUF docs/example content vs parallel docs restructuring; not the same change." + "reason": "GGUF documentation additions and parallel-inference docs updates are different documentation topics." }, { "left": "pull_request:1229", "right": "pull_request:130", "accept": false, - "reason": "Both edit docs, but one updates DreamBooth install guidance and the other adds accelerate config instructions." + "reason": "DreamBooth README wording and the accelerate config note are separate example-doc edits." }, { "left": "pull_request:12425", "right": "pull_request:13211", "accept": false, - "reason": "Different offload systems: torch.distributed import guard vs group-offloading runtime fallback." + "reason": "torch.distributed import guarding is unrelated to the group-offloading runtime bug fix." }, { "left": "pull_request:1508", "right": "pull_request:402", "accept": false, - "reason": "Both docs-related, but one clarifies DreamBooth class-image counts and the other reorganizes the intro docs tree." + "reason": "DreamBooth class-image clarification and the intro docs rewrite are unrelated documentation changes." }, { "left": "pull_request:152", "right": "pull_request:1527", "accept": false, - "reason": "Local-image-folder training support vs inpaint tensor input handling; different code paths." + "reason": "Local-folder training support and tensor image/mask input support solve different example/pipeline behaviors." }, { "left": "pull_request:2037", "right": "pull_request:3586", "accept": false, - "reason": "Both touch community docs, but they add different community pipelines." + "reason": "Both add community examples, but they are different pipelines/features and not the same underlying change." }, { "left": "pull_request:2062", "right": "pull_request:3708", "accept": false, - "reason": "Paint-by-example CPU offload fix vs custom release/version parsing fix; unrelated." + "reason": "Paint-by-example CPU offload and custom release sorting are unrelated fixes." }, { "left": "pull_request:2279", "right": "pull_request:249", "accept": false, - "reason": "Both update tooling pins, but one swaps flake8/ruff and black preview behavior while the other only pins black." + "reason": "Both involve tooling, but one replaces flake8 with ruff while the other pins black; not the same concrete change." }, { "left": "pull_request:2341", "right": "pull_request:2655", "accept": false, - "reason": "EMAModel checkpoint saving vs LoRA loading/saving logic; different functionality." + "reason": "EMA checkpoint saving and LoRA load/save behavior are different training vs loader issues." }, { "left": "pull_request:2345", "right": "pull_request:833", "accept": false, - "reason": "xFormers training warning across example scripts vs package release version bump." + "reason": "xFormers training warning and the 0.5.1 release bump are unrelated." }, { "left": "pull_request:2415", "right": "pull_request:676", "accept": false, - "reason": "Pipeline utility import/version guard vs a simple resnet variable rename." + "reason": "Pipeline version-check logic and a resnet variable rename are not the same bug/change." }, { "left": "pull_request:2467", "right": "pull_request:4187", "accept": false, - "reason": "Latent output-type plumbing across pipelines vs scheduler timestep/index-counter changes; not the same bug." + "reason": "Latent output support and DPMSolver index handling are unrelated scheduler/pipeline changes." }, { "left": "pull_request:2595", "right": "pull_request:3536", "accept": false, - "reason": "Scheduler switching/config normalization vs IF super-resolution PIL normalization." + "reason": "Scheduler config switching and IF super-resolution normalization are different code paths." }, { "left": "pull_request:2968", "right": "pull_request:5053", "accept": false, - "reason": "Post-processing fix in img2img pipelines vs validation_image argument parsing." + "reason": "Post-processing control flow and the validation_image CLI argument are unrelated." }, { "left": "pull_request:3042", "right": "pull_request:5035", "accept": false, - "reason": "Accelerate minimum version bump for training examples vs import-availability handling for transformers." + "reason": "They both adjust dependency/import handling, but for different packages and problems; not the same underlying fix." }, { "left": "pull_request:3657", "right": "pull_request:5665", "accept": false, - "reason": "ControlNet checkpoint parser tweaks vs passing height/width to image preprocessing." + "reason": "ControlNet checkpoint parsing tweaks and image preprocessing size propagation are distinct issues." }, { "left": "pull_request:3806", "right": "pull_request:5111", "accept": false, - "reason": "Checkpoint-loading docs note vs Karras sigma support in samplers; unrelated." + "reason": "Checkpoint loading docs and Karras sigma scheduler support are unrelated." }, { "left": "pull_request:4040", "right": "pull_request:4248", "accept": false, - "reason": "Different config/list handling issues: _use_default_values type vs tracker config sanitization." + "reason": "A config default-type fix and removing a list from tracker config are different configuration-related bugs." }, { "left": "pull_request:4248", "right": "pull_request:5665", "accept": false, - "reason": "Tracker config cleanup vs ControlNet preprocessing dimensions; unrelated." + "reason": "Accelerator tracker config cleanup and ControlNet image preprocessing are unrelated changes." + }, + { + "left": "pull_request:6261", + "right": "pull_request:7044", + "accept": false, + "reason": "Moving Attend-and-Excite out of stable_diffusion and re-adding a UNet refactor are different refactorings, not one merged change." } ] }, @@ -149059,7 +149142,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 142, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -149094,16 +149177,16 @@ "pull_request:12586", "pull_request:12797", "pull_request:12960", + "pull_request:13120", + "pull_request:13313", "pull_request:13365", "pull_request:1982", "pull_request:4276", "pull_request:4987", "pull_request:5010", "pull_request:6241", - "pull_request:6261", "pull_request:6426", "pull_request:6886", - "pull_request:7044", "pull_request:7116", "pull_request:7216", "pull_request:7455", @@ -149117,7 +149200,6 @@ "pull_request:9940" ], "soft_pairs": [ - "pull_request:6261|pull_request:7044", "pull_request:6426|pull_request:8719", "pull_request:6886|pull_request:7455", "pull_request:7116|pull_request:7679", @@ -149141,186 +149223,187 @@ "pull_request:12450|pull_request:13365", "pull_request:12451|pull_request:12586", "pull_request:12487|pull_request:12797", - "pull_request:12552|pull_request:12797" + "pull_request:12552|pull_request:12797", + "pull_request:13120|pull_request:13313" ], - "prepared_review_unit_hash": "4727b82f293c7f4581804e81c7fd79b6c329ad9b6366a96465d563d80c28b0fb", + "prepared_review_unit_hash": "1c0a1a62951b34119f1ce68cd4fb864ca13989d968126f128f76a883bb4b3025", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 95828, - "estimated_input_tokens": 23957, - "estimated_eval_tokens": 48170 + "serialized_chars": 95462, + "estimated_input_tokens": 23866, + "estimated_eval_tokens": 47988 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated docs, CI, dependency, and model/code fixes. The soft links are mostly superficial similarity on broad maintenance themes, not the same underlying bug or change.", - "confidence": 0.95, + "summary": "This cluster is heterogeneous: the suggested soft matches are mostly superficial title or subsystem similarities, not the same underlying bug/change. Reject all soft edges.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items span unrelated docs, CI, dependency, and model fixes, so there is no single underlying change to represent the cluster.", + "canonical_pr_reason": "No single PR cleanly represents the cluster; the artifacts span unrelated docs, CI, loader, model, and example fixes.", "best_issue_reason": null, - "best_pr_reason": "No clear best PR: even the closest matches only share broad subsystem themes, not one concrete code-path fix that could serve as the cluster representative.", + "best_pr_reason": "No PR is a good global representative because this is not a true duplicate set.", "soft_edge_verdicts": [ - { - "left": "pull_request:6261", - "right": "pull_request:7044", - "accept": false, - "reason": "Different changes: attend-and-excite package refactor vs a UNet mid-block refactor. Same area, not the same bug or patch." - }, { "left": "pull_request:6426", "right": "pull_request:8719", "accept": false, - "reason": "Both touch deprecation-related code, but one is broad LoRA behavior and the other is VQ output docs/typing. Not the same change." + "reason": "Both mention deprecation cleanup, but one is broad LoRA/API deprecation handling and the other is a VQ model output doc/type cleanup; different changes." }, { "left": "pull_request:6886", "right": "pull_request:7455", "accept": false, - "reason": "Unrelated fixes: deprecation cleanup for .to() usage vs removal of distutils. No shared bug or code path." + "reason": "One removes deprecated `torch_dtype` usage in scripts; the other removes `distutils` usage. Same style of cleanup, different code paths." }, { "left": "pull_request:7116", "right": "pull_request:7679", "accept": false, - "reason": "Both are workflow/install tweaks, but they address different installation commands and packages. Not one mergeable fix." + "reason": "Both touch CI installs, but one is a workflow-wide uv migration and the other is a specific temporary pip fallback for peft; not the same fix." }, { "left": "pull_request:7216", "right": "pull_request:8631", "accept": false, - "reason": "Different single-file loader issues: signature/docs update vs T5 checkpoint return handling. Not the same defect." + "reason": "Both relate to single-file loading, but one updates the public signature while the other fixes a T5 checkpoint return bug; different underlying changes." }, { "left": "pull_request:8522", "right": "pull_request:8563", "accept": false, - "reason": "Both edit dependency pins in setup.py, but they bump different libraries for different reasons. Too broad to merge as duplicates." + "reason": "Both update dependency pins, but they are separate version bumps with different scope and rationale." }, { "left": "pull_request:8563", "right": "pull_request:9601", "accept": false, - "reason": "Dependency version pinning vs VAE dtype logic in a training script. Separate changes entirely." + "reason": "Dependency version pinning and SD3 VAE dtype handling are unrelated problems." }, { "left": "pull_request:10122", "right": "pull_request:9473", "accept": false, - "reason": "Flux attention-mask fix and Flux training LR fix are unrelated bugs in different code paths." + "reason": "Flux attention-mask wiring and Flux DreamBooth learning-rate logic are different code-path fixes." }, { "left": "pull_request:4276", "right": "pull_request:4987", "accept": false, - "reason": "Pipeline import test coverage vs removing a duplicate GLIGEN import. Different goals and code paths." + "reason": "One adds import tests; the other removes a duplicate GLIGEN export. No shared bug." }, { "left": "pull_request:11290", "right": "pull_request:12424", "accept": false, - "reason": "Both affect Docker/CI packaging, but one unpins torch versions and the other fixes Dockerfile definitions. Not the same change." + "reason": "A Dockerfile torch-pin change and a broad Dockerfile definition cleanup are not the same change." }, { "left": "pull_request:11994", "right": "pull_request:12960", "accept": false, - "reason": "Hub auth/login docs and FSDP detection in training scripts are unrelated." + "reason": "Hub login/docs updates and `is_fsdp` detection in training scripts are unrelated." }, { "left": "pull_request:10105", "right": "pull_request:9940", "accept": false, - "reason": "Both are small cleanup PRs, but one fixes a broken link and the other fixes textual inversion wording/API usage. Not the same underlying issue." + "reason": "Both are docs-oriented, but one fixes a broken link and the other updates textual inversion docs/example wording; not the same issue." }, { "left": "pull_request:5010", "right": "pull_request:6241", "accept": false, - "reason": "GLIGEN text-image import fix and a runtime dtype error report are unrelated." + "reason": "These are unrelated: one is a GLIGEN import/export fix, the other is a runtime dtype mismatch report with no matching patch shown." }, { "left": "pull_request:11571", "right": "pull_request:12021", "accept": false, - "reason": "LTX docs update versus LTX attention-backend support. Same product family, different concrete change." + "reason": "LTX docs improvements and LTX attention backend support are distinct changes." }, { "left": "pull_request:10454", "right": "pull_request:10983", "accept": false, - "reason": "A narrow Hunyuan attention-mask bug fix is not the same as the broader Hunyuan I2V feature/release PR." + "reason": "Both touch Hunyuan video, but 10454 fixes attention-mask dimensions while 10983 is a much larger feature/enablement PR; not mergeable as one duplicate fix." }, { "left": "pull_request:10318", "right": "pull_request:11269", "accept": false, - "reason": "Flux LoRA integration tests and ONNXRuntime package support are unrelated changes." + "reason": "Flux LoRA integration tests and ONNX Runtime backend support are unrelated." }, { "left": "pull_request:10342", "right": "pull_request:11366", "accept": false, - "reason": "Parameter dtype helper logic and Hidream output API cleanup do not target the same bug." + "reason": "`get_parameter_dtype` behavior and Hidream transformer outputs are unrelated code paths." }, { "left": "pull_request:11044", "right": "pull_request:11535", "accept": false, - "reason": "LoRA no-op warning behavior and Framepack layer initialization order are different fixes." + "reason": "LoRA no-op warning changes and Framepack transformer initialization order are different changes." }, { "left": "pull_request:113", "right": "pull_request:1982", "accept": false, - "reason": "Scheduler initialization bug and example test path update are unrelated." + "reason": "An SDE-VE scheduler init fix and an example test path update are unrelated." }, { "left": "pull_request:12005", "right": "pull_request:12092", "accept": false, - "reason": "Sharded checkpoint local_files_only handling and GGUF docs are unrelated." + "reason": "Sharded-checkpoint `local_files_only` behavior and GGUF documentation are unrelated." }, { "left": "pull_request:12360", "right": "pull_request:12467", "accept": false, - "reason": "Chroma docs wording and QwenImage batch handling are different issues." + "reason": "Chroma docs wording and QwenImage batch handling are unrelated." }, { "left": "pull_request:12450", "right": "pull_request:13365", "accept": false, - "reason": "CI installation revamp and Claude review workflow fork support are separate infrastructure changes." + "reason": "CI installation changes and Claude review fork support are different workflow problems." }, { "left": "pull_request:12451", "right": "pull_request:12586", "accept": false, - "reason": "Doc-builder Dockerfile deps vs doc tree sorting are unrelated maintenance changes." + "reason": "One adds doc-builder dependencies; the other just reorders a toctree entry." }, { "left": "pull_request:12487", "right": "pull_request:12797", "accept": false, - "reason": "Both touch documentation, but one fixes broken links broadly and the other makes minor Kandinsky doc edits. Not the same change." + "reason": "Both are doc fixes, but they target different models and different link/wording issues." }, { "left": "pull_request:12552", "right": "pull_request:12797", "accept": false, - "reason": "Kandinsky typo fixes and Kandinsky minor doc updates overlap in topic only; they are not the same concrete documentation bug." + "reason": "Kandinsky5 typo fixes and broader Kandinsky doc tweaks are not the same underlying change." + }, + { + "left": "pull_request:13120", + "right": "pull_request:13313", + "accept": false, + "reason": "Modular pipeline tests and LoRA conversion utilities are unrelated." } ] }, @@ -149336,16 +149419,16 @@ "component_index": 1, "component_count": 266, "review_unit_index": 143, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:13120", - "pull_request:13313", + "pull_request:12714", + "pull_request:13321", "pull_request:1513", "pull_request:1913", "pull_request:1982", @@ -149394,7 +149477,6 @@ "pull_request:980" ], "soft_pairs": [ - "pull_request:13120|pull_request:13313", "pull_request:1513|pull_request:402", "pull_request:1913|pull_request:326", "pull_request:1982|pull_request:2209", @@ -149418,186 +149500,187 @@ "pull_request:5724|pull_request:6163", "pull_request:6215|pull_request:8510", "pull_request:6340|pull_request:7024", - "pull_request:8438|pull_request:8571" + "pull_request:8438|pull_request:8571", + "pull_request:12714|pull_request:13321" ], - "prepared_review_unit_hash": "dcaa43cf5ab1a046c9327a0fc929cfc83c1b158230587de5a647fd8f18a821be", + "prepared_review_unit_hash": "7a77e4ba17a0c90cba22e634e2754aacc1d740e4f5b3c347b83d87c90b3f5c0a", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 102847, - "estimated_input_tokens": 25712, - "estimated_eval_tokens": 51680 + "serialized_chars": 102788, + "estimated_input_tokens": 25697, + "estimated_eval_tokens": 51650 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated pull requests across docs, tests, dependency bumps, loaders, schedulers, and pipelines. The soft pairs mostly share only a broad subsystem or formatting/theme, not the same concrete bug or change, so none should be merged as duplicates.", + "summary": "These PRs are only superficially similar; the soft edges mostly connect unrelated docs, tests, dependency, and pipeline changes, so none should be deduplicated.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical duplicate target; the candidate pairs do not share the same underlying bug or change, only broad keyword overlap.", "best_issue_reason": null, - "best_pr_reason": "No single PR is a good canonical representative here; the items are heterogeneous and do not describe one underlying fix or feature.", + "best_pr_reason": "No PR is a strong global best candidate because the set is heterogeneous and the pairings are not true duplicates.", "soft_edge_verdicts": [ - { - "left": "pull_request:13120", - "right": "pull_request:13313", - "accept": false, - "reason": "One adds AutoModel load-id tagging tests; the other changes LoRA conversion/loading logic for Flux2. Same loading area, but different code paths and problems." - }, { "left": "pull_request:1513", "right": "pull_request:402", "accept": false, - "reason": "Both are documentation changes, but one adds an accelerate notebook config tip and the other reorganizes the docs toctree. Not the same change." + "reason": "Both are docs-related, but one adds notebook setup guidance while the other restructures the intro docs tree; different content and purpose." }, { "left": "pull_request:1913", "right": "pull_request:326", "accept": false, - "reason": "1913 adds runtime input-type validation for img2img pipelines; 326 is broader UNet/type-hint cleanup and input-length checks. Different fixes." + "reason": "One validates img2img input types in specific pipelines; the other is a broad typing/UNet API PR across many files, not the same fix." }, { "left": "pull_request:1982", "right": "pull_request:2209", "accept": false, - "reason": "One updates example test paths; the other corrects an argument name in docs. Different files, different issue." + "reason": "Example test path correction vs. an image-to-image docs argument-name fix; unrelated changes." }, { "left": "pull_request:2279", "right": "pull_request:538", "accept": false, - "reason": "Ruff/black workflow migration is unrelated to a hub version bump. Same repo maintenance category only." + "reason": "CI/lint tool migration is unrelated to a package version bump." }, { "left": "pull_request:230", "right": "pull_request:980", "accept": false, - "reason": "Both touch fp16/loading or dtype handling, but for different components and code paths: model loading vs Dance Diffusion pipeline audio dtype." + "reason": "Both touch fp16/dtype themes, but one changes loading APIs and the other adjusts Dance Diffusion pipeline/runtime dtype handling." }, { "left": "pull_request:2327", "right": "pull_request:3681", "accept": false, - "reason": "UNet input-length validation is unrelated to manual checkpoint-limit handling in training scripts." + "reason": "UNet constructor validation is unrelated to checkpoint limit handling in training scripts." }, { "left": "pull_request:2365", "right": "pull_request:3471", "accept": false, - "reason": "A scheduler enum note and a DreamBooth docs torch.compile note are unrelated documentation edits." + "reason": "A scheduler-type note in code is not the same as a DreamBooth docs note about torch.compile." }, { "left": "pull_request:2365", "right": "pull_request:3806", "accept": false, - "reason": "No shared bug/change: scheduler typing note vs checkpoint-loading docs update." + "reason": "Different docs topics: scheduler typing vs. checkpoint loading guidance." }, { "left": "pull_request:2415", "right": "pull_request:3054", "accept": false, - "reason": "Pipeline utils version guard and a documentation wording tweak are not the same underlying issue." + "reason": "Pipeline loader compatibility logic and a docs wording cleanup are unrelated." }, { "left": "pull_request:2470", "right": "pull_request:4262", "accept": false, - "reason": "MPS test skips/fixes are unrelated to SDXL DreamBooth LoRA training fixes." + "reason": "Both are test-oriented, but one is an MPS skip fix and the other is SDXL DreamBooth LoRA training changes." }, { "left": "pull_request:2487", "right": "pull_request:2553", "accept": false, - "reason": "One adds a JAX/Flax guide; the other changes classifier-free guidance logic in Flax pipelines. Different scope and intent." + "reason": "A JAX/Flax docs addition is unrelated to adding a CFG flag in Flax pipelines." }, { "left": "pull_request:2894", "right": "pull_request:2958", "accept": false, - "reason": "Adding a flax requirement for ControlNet training is unrelated to adding ControlNet inpaint support." + "reason": "A requirements file addition and a new ControlNet Inpaint feature are different changes." }, { "left": "pull_request:2970", "right": "pull_request:3957", "accept": false, - "reason": "Variable rename in a DreamBooth inpaint script is not the same as the WIP attention-name correction work." + "reason": "Variable rename in one DreamBooth script does not match the attn-name correction WIP." }, { "left": "pull_request:3656", "right": "pull_request:5220", "accept": false, - "reason": "Activation/normalization refactor is unrelated to moving tests from slow to nightly." + "reason": "Core activation refactor and moving tests to nightly are unrelated." }, { "left": "pull_request:3681", "right": "pull_request:4078", "accept": false, - "reason": "Manual checkpoint-limit checking in training scripts is unrelated to requiring accelerate for from_single_file conversion." + "reason": "Training checkpoint-limit logic and from_single_file accelerate gating are separate code paths." }, { "left": "pull_request:4669", "right": "pull_request:6448", "accept": false, - "reason": "One fixes LoRA rank handling for text encoders; the other adjusts LoRA tests and monkey-patch cleanup. Related area, but not the same bug or change." + "reason": "LoRA rank handling in loaders is not the same as LoRA test assertion cleanup." }, { "left": "pull_request:5057", "right": "pull_request:6447", "accept": false, - "reason": "SDXL clip-skip support and adding K-Diffusion pipelines are distinct features in different parts of the codebase." + "reason": "SDXL clip-skip support and a K-Diffusion pipeline addition are distinct features." }, { "left": "pull_request:5478", "right": "pull_request:6643", "accept": false, - "reason": "Japanese docs localization and a new UViT2D API doc page are different documentation additions." + "reason": "Japanese docs rollout and UViT2D docs are separate documentation additions." }, { "left": "pull_request:5493", "right": "pull_request:5946", "accept": false, - "reason": "Core activation/normalization refactor is unrelated to refactoring example tests." + "reason": "A core activation/normalization refactor is not the same as refactoring example tests." }, { "left": "pull_request:5698", "right": "pull_request:5702", "accept": false, - "reason": "LCM img2img fixes and AnimateDiff prompt handling are separate pipeline bugs." + "reason": "LCM img2img bugfix and AnimateDiff prompt bugfix affect different pipelines and failure modes." }, { "left": "pull_request:5724", "right": "pull_request:6163", "accept": false, - "reason": "Both are LoRA-test related, but one adds a decorator and extra coverage while the other changes different test setup/assertions. Not the same concrete fix." + "reason": "Both are LoRA test-related, but one adds backend gating and the other tweaks test coverage/behavior; not the same underlying fix." }, { "left": "pull_request:6215", "right": "pull_request:8510", "accept": false, - "reason": "PixArt VAE initialization test stabilization is unrelated to DreamBooth LoRA learning-rate scheduler fixes." + "reason": "PixArt test initialization fix and DreamBooth LoRA scheduler fixes are unrelated." }, { "left": "pull_request:6340", "right": "pull_request:7024", "accept": false, - "reason": "AnimateDiff conversion scripts and an AnimateLCM docs section are different artifacts with different purposes." + "reason": "AnimateDiff conversion scripts and AnimateLCM docs are different artifacts and changes." }, { "left": "pull_request:8438", "right": "pull_request:8571", "accept": false, - "reason": "Both are general errata-style PRs, but they touch different files and fixes; they are not the same underlying change." + "reason": "Both are broad errata/style PRs, but they touch different files and distinct content edits; not one underlying change." + }, + { + "left": "pull_request:12714", + "right": "pull_request:13321", + "accept": false, + "reason": "Flux2 LoRA NaN test fixes and an unguarded torchvision import fix in Cosmos are unrelated." } ] }, @@ -149613,11 +149696,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 144, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -149639,11 +149722,9 @@ "pull_request:12438", "pull_request:12490", "pull_request:12491", - "pull_request:12714", "pull_request:12960", "pull_request:13081", "pull_request:13134", - "pull_request:13321", "pull_request:13330", "pull_request:1338", "pull_request:13410", @@ -149653,11 +149734,13 @@ "pull_request:1525", "pull_request:1527", "pull_request:1706", + "pull_request:1788", "pull_request:2415", "pull_request:2902", "pull_request:3292", "pull_request:3473", "pull_request:3536", + "pull_request:3708", "pull_request:4245", "pull_request:4278", "pull_request:453", @@ -149671,7 +149754,6 @@ "pull_request:9676" ], "soft_pairs": [ - "pull_request:12714|pull_request:13321", "pull_request:2902|pull_request:4278", "pull_request:11522|pull_request:12438", "pull_request:12424|pull_request:13134", @@ -149695,186 +149777,187 @@ "pull_request:1525|pull_request:453", "pull_request:1527|pull_request:3292", "pull_request:1527|pull_request:3536", - "pull_request:1706|pull_request:2415" + "pull_request:1706|pull_request:2415", + "pull_request:1788|pull_request:3708" ], - "prepared_review_unit_hash": "e0222d07be5fed3a1ad671fc306916424a164a2925e4760b04c6286590e7f0a6", + "prepared_review_unit_hash": "89443ecaf70c628ef89400b1ee95ee86f57ef633c2a1e463cda14eb545f8a9d4", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 92744, - "estimated_input_tokens": 23186, - "estimated_eval_tokens": 46628 + "serialized_chars": 91971, + "estimated_input_tokens": 22993, + "estimated_eval_tokens": 46242 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a mixed bag of largely unrelated PRs: formatting/docs updates, CI/workflow tweaks, model/pipeline bug fixes, and training/test adjustments. The soft pairs are only superficially similar (often same subsystem or similar wording) but do not fix the same concrete bug or change.", - "confidence": 0.97, + "summary": "This cluster does not contain a real duplicate set; the soft pairs are mostly unrelated formatting, docs, CI, or small fixes in different code paths. None of the candidate pairs looks mergeable as one PR or the same underlying bug/change.", + "confidence": 0.93, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR #12438 is the strongest representative because it fixes a concrete runtime loading bug in core pipeline-loading code and has clear downstream impact; the rest are mostly doc, test, or unrelated subsystem changes.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:12714", - "right": "pull_request:13321", - "accept": false, - "reason": "Different problems: Flux2 LoRA NaN test vs unguarded torchvision import in Cosmos." - }, { "left": "pull_request:2902", "right": "pull_request:4278", "accept": false, - "reason": "Both involve img2img/upscale handling, but one is cross_attention_kwargs plumbing and the other is a prompt=None batch-size bug; not the same fix." + "reason": "Both touch image pipelines, but they fix different bugs in different classes and inputs; not the same concrete code-path issue." }, { "left": "pull_request:11522", "right": "pull_request:12438", "accept": false, - "reason": "Both touch loading utilities, but one is an AudioLDM2 transformers-version workaround and the other handles offload_state_dict for transformers model init; different code paths." + "reason": "Both are version/compatibility fixes around loading, but one is AudioLDM2 transformers handling and the other is generic offload_state_dict cleanup; different failure modes." }, { "left": "pull_request:12424", "right": "pull_request:13134", "accept": false, - "reason": "Unrelated: Dockerfile/workflow cleanup versus Cosmos transformer typing annotations." + "reason": "Unrelated scopes: Dockerfile/workflow cleanup versus Cosmos transformer typing; no shared bug or change." }, { "left": "pull_request:12490", "right": "pull_request:12960", "accept": false, - "reason": "Different issues in different areas: a test config change for AudioLDM2 vs FSDP detection in Flux2 training scripts." + "reason": "Audioldm2 test fixture change versus Flux FSDP check; both are tests/examples but not the same issue." }, { "left": "pull_request:10114", "right": "pull_request:8425", "accept": false, - "reason": "Both are docs/CI-ish formatting changes, but one is Stable Audio formatting and the other just renames a workflow; no shared bug." + "reason": "Stable Audio formatting fix and workflow name rename are unrelated cosmetic edits." }, { "left": "pull_request:13410", "right": "pull_request:13472", "accept": false, - "reason": "Different documentation/agent-guide updates: modular pipeline docs vs float64 gotcha guidance." + "reason": "Different documentation updates: modular-pipeline guidance versus float64 gotcha; not duplicate content." }, { "left": "pull_request:3473", "right": "pull_request:4245", "accept": false, - "reason": "Textual inversion docs vs adding safetensors flags across many docs; same broad docs area, but not the same change." + "reason": "Both are docs, but one adds textual inversion inference and the other adds safetensors examples across many docs; different changes." }, { "left": "pull_request:10188", "right": "pull_request:8798", "accept": false, - "reason": "One fixes LoRA debug/error messaging in Flux, the other fixes sharded checkpoint loading from subfolder; distinct bugs." + "reason": "Both concern loading behavior, but one is LoRA expansion error handling and the other is sharded checkpoint subfolder loading; distinct bugs." }, { "left": "pull_request:10200", "right": "pull_request:8591", "accept": false, - "reason": "ControlNet Union refactor versus SD3 training density/weighting utilities refactor; different features and code paths." + "reason": "ControlNet Union refactor versus SD3 training density/weighting utilities; different subsystems and purposes." }, { "left": "pull_request:10245", "right": "pull_request:9592", "accept": false, - "reason": "PEFT inference docs change is unrelated to Flax embeddings docstring refactoring." + "reason": "PEFT docs tweak versus Flax embedding docstring refactor; not the same underlying change." }, { "left": "pull_request:10245", "right": "pull_request:9593", "accept": false, - "reason": "PEFT inference docs change is unrelated to community example docstring refactoring." + "reason": "PEFT docs tweak versus community docstring cleanup; unrelated documentation work." }, { "left": "pull_request:10431", "right": "pull_request:10538", "accept": false, - "reason": "Sana 1k PE bug fix in the transformer is not the same as adding Sana to auto-pipeline registration." + "reason": "Sana transformer bug fix versus auto-pipeline registration for Sana; one fixes model internals, the other pipeline discovery." }, { "left": "pull_request:10539", "right": "pull_request:11814", "accept": false, - "reason": "Flux true-CFG condition fix versus gradient-checkpointing kwargs plumbing; same model family, different concrete issues." + "reason": "Flux true-CFG condition fix versus gradient-checkpointing kwargs propagation; same model family but different code paths and bugs." }, { "left": "pull_request:11238", "right": "pull_request:9676", "accept": false, - "reason": "SD3 validation/A100 execution fix is unrelated to an import cleanup in benchmarks." + "reason": "SD3 ControlNet validation/runtime fix versus benchmark import cleanup; unrelated." }, { "left": "pull_request:1137", "right": "pull_request:51", "accept": false, - "reason": "Different eras and bugs: inpaint safety-checker dtype fix vs install_requires typo fix." + "reason": "Different eras and bugs: dtype cast in safety checker versus install_requires typo." }, { "left": "pull_request:11479", "right": "pull_request:12363", "accept": false, - "reason": "Intel Gaudi documentation update versus QwenImageEdit Plus docs/tests; both docs-heavy but unrelated." + "reason": "Intel Gaudi doc rename versus QwenImage Edit Plus docs/tests; no common issue." }, { "left": "pull_request:11614", "right": "pull_request:12491", "accept": false, - "reason": "Docker CI security hardening versus documentation link-check CI; both workflow-related but different concerns." + "reason": "Docker build script security cleanup versus documentation link-check CI; both CI-related but not same change." }, { "left": "pull_request:13081", "right": "pull_request:13330", "accept": false, - "reason": "Qwen hidden-state contiguity fix is unrelated to torchao test adjustments." + "reason": "Qwen hidden-state contiguity fix versus torchao test backend updates; one is a model bug, the other is test infrastructure." }, { "left": "pull_request:1338", "right": "pull_request:748", "accept": false, - "reason": "ONNX inpainting cast order fix is unrelated to push_to_hub argument cleanup in training scripts." + "reason": "ONNX inpainting cast order fix versus push_to_hub API cleanup; unrelated pipeline bug and training utility fix." }, { "left": "pull_request:1508", "right": "pull_request:152", "accept": false, - "reason": "Dreambooth num_class_images clarification is unrelated to local image-folder dataset support in unconditional training." + "reason": "DreamBooth prior-preservation docs clarification versus local image-folder training support; docs vs feature change." }, { "left": "pull_request:1525", "right": "pull_request:453", "accept": false, - "reason": "Upscaling batch-size fix and installation typo are unrelated." + "reason": "Upscaling batch-size fix versus installation typo; completely unrelated." }, { "left": "pull_request:1527", "right": "pull_request:3292", "accept": false, - "reason": "Stable Diffusion inpaint tensor-input support is unrelated to adding input perturbation in text-to-image training." + "reason": "Stable Diffusion inpaint tensor-input support versus input perturbation option in text-to-image training; different pipelines and behavior." }, { "left": "pull_request:1527", "right": "pull_request:3536", "accept": false, - "reason": "Both involve image preprocessing, but one is Stable Diffusion inpaint input handling and the other is IF super-resolution PIL normalization; different pipelines and bugs." + "reason": "Stable Diffusion inpaint tensor-input support versus DeepFloyd IF PIL normalization; different pipeline families and bugs." }, { "left": "pull_request:1706", "right": "pull_request:2415", "accept": false, - "reason": "Main docs update versus pipeline_utils version handling; not the same artifact or bug." + "reason": "Main docs update versus pipeline_utils version handling; unrelated documentation and code maintenance." + }, + { + "left": "pull_request:1788", + "right": "pull_request:3708", + "accept": false, + "reason": "DDIM prediction_type logic bug versus custom release sorting/version parsing; different modules and failure modes." } ] }, @@ -149890,11 +149973,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 145, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -149902,14 +149985,14 @@ "pull_request:10359", "pull_request:10567", "pull_request:10810", + "pull_request:10985", "pull_request:11469", + "pull_request:11999", "pull_request:12151", - "pull_request:1788", "pull_request:2655", "pull_request:3473", "pull_request:3589", "pull_request:3656", - "pull_request:3708", "pull_request:3723", "pull_request:3797", "pull_request:3850", @@ -149947,7 +150030,6 @@ "pull_request:9307" ], "soft_pairs": [ - "pull_request:1788|pull_request:3708", "pull_request:2655|pull_request:3797", "pull_request:3473|pull_request:4070", "pull_request:3589|pull_request:3723", @@ -149970,180 +150052,181 @@ "pull_request:10359|pull_request:11469", "pull_request:449|pull_request:676", "pull_request:10136|pull_request:10567", - "pull_request:10810|pull_request:12151" + "pull_request:10810|pull_request:12151", + "pull_request:10985|pull_request:11999" ], - "prepared_review_unit_hash": "32ddc21d51c1c4e7d2e189524ece6835c327dae2360892b5e92549ce9c86669f", + "prepared_review_unit_hash": "4f180f56ec9342484c11e6598904ad2c7f38f7363988362e9a84b26d9a1d3b99", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 98770, - "estimated_input_tokens": 24693, - "estimated_eval_tokens": 49642 + "serialized_chars": 100013, + "estimated_input_tokens": 25004, + "estimated_eval_tokens": 50264 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of mostly unrelated PRs linked only by loose lexical similarity. No soft edge looks like the same underlying bug or change, so all should be rejected.", - "confidence": 0.84, + "summary": "This is a broad false-positive cluster: the PRs span unrelated docs, tests, workflows, LoRA, Hunyuan Video, IP-Adapter, and single-file loader work. None of the soft edges looks like the same underlying change or bug fix.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: the set is heterogeneous, and none of the PRs subsume the others or represent the whole cluster.", + "canonical_pr_reason": "No canonical PR: there isn\u2019t a single duplicate thread here, just many unrelated PRs that happen to share broad subsystem keywords.", "best_issue_reason": null, - "best_pr_reason": "PR 10136 is the strongest representative artifact here: it is the broad, foundational Hunyuan Video implementation with the largest scope and most central merged code footprint, even though it does not duplicate the others.", + "best_pr_reason": "No PR is a good global representative because the cluster is not coherent; PR 10136 is large, but it is still unrelated to most others.", "soft_edge_verdicts": [ - { - "left": "pull_request:1788", - "right": "pull_request:3708", - "accept": false, - "reason": "DDIM scheduler bug fix vs custom release version sorting; unrelated code paths." - }, { "left": "pull_request:2655", "right": "pull_request:3797", "accept": false, - "reason": "LoRA save/load behavior vs attention naming cleanup; different problems." + "reason": "LoRA save/load fix vs attention-head naming correction across model configs; different bugs and code paths." }, { "left": "pull_request:3473", "right": "pull_request:4070", "accept": false, - "reason": "Both are docs PRs, but one adds textual inversion docs and the other edits SDXL docs; not the same change." + "reason": "Textual inversion docs vs SDXL docs wording cleanup; both are documentation, but not the same change." }, { "left": "pull_request:3589", "right": "pull_request:3723", "accept": false, - "reason": "Both touch reference-related examples, but one updates README/community prose while the other changes pipeline input handling." + "reason": "Community README notes vs adding np.ndarray support to reference pipelines; same example area, different concrete changes." }, { "left": "pull_request:3656", "right": "pull_request:4451", "accept": false, - "reason": "Activation helper refactor vs moving tests to nightly; unrelated." + "reason": "Activation helper refactor vs moving tests to nightly; unrelated refactor and test-scheduling change." }, { "left": "pull_request:3850", "right": "pull_request:5948", "accept": false, - "reason": "AudioDiffusion test fixture adjustment vs variant-loading warning fix; unrelated." + "reason": "AudioDiffusion test fixture fix vs variant-loading warning tweak; no shared bug or code path." }, { "left": "pull_request:3957", "right": "pull_request:6195", "accept": false, - "reason": "WIP attention-head naming fix vs AnimateDiff SDXL implementation; different code paths and goals." + "reason": "WIP attention-head correction vs AnimateDiff SDXL feature work; shared file path is not enough to merge them." }, { "left": "pull_request:4749", "right": "pull_request:6972", "accept": false, - "reason": "SDXL LoRA checkpointing vs IP-Adapter unload test; not the same bug or change." + "reason": "SDXL LoRA checkpointing vs IP-Adapter unload test; different components and failure modes." }, { "left": "pull_request:5506", "right": "pull_request:7458", "accept": false, - "reason": "PEFT/xformers LoRA fix vs Python 3.12 test skip; unrelated." + "reason": "PEFT/xformers LoRA bugfix vs skipping dynamo tests on Python 3.12; unrelated." }, { "left": "pull_request:5747", "right": "pull_request:7634", "accept": false, - "reason": "Dataset `data_dir` plumbing vs high-res UNet/pipeline fix; unrelated." + "reason": "Dataset data_dir plumbing vs high-res generation model addition; different features." }, { "left": "pull_request:6150", "right": "pull_request:6564", "accept": false, - "reason": "Both are IP-Adapter related, but one adds a converter and the other fixes subfolder path generation; distinct changes." + "reason": "IP-Adapter converter method vs path handling in image-encoder loading; same loader family, but different concrete fixes." }, { "left": "pull_request:6215", "right": "pull_request:8425", "accept": false, - "reason": "PixArt VAE test initialization vs workflow name rename; unrelated." + "reason": "PixArt test determinism fix vs workflow name rename; unrelated." }, { "left": "pull_request:6373", "right": "pull_request:7202", "accept": false, - "reason": "Deleting doc-comment workflows vs removing a requirements dependency; unrelated." + "reason": "Deleting doc-comment workflows vs removing huggingface-cli from example requirements; no common bug." }, { "left": "pull_request:6426", "right": "pull_request:6886", "accept": false, - "reason": "LoRA deprecation handling vs removing `torch_dtype` deprecation warnings; different issues." + "reason": "LoRA deprecation cleanup vs removing torch_dtype deprecation in to(); different deprecation paths." }, { "left": "pull_request:6961", "right": "pull_request:8627", "accept": false, - "reason": "Single-file loader tests vs SD3 control test-size changes; unrelated." + "reason": "Single-file loader test additions vs SD3 ControlNet test-size reduction; both tests, but different targets and issues." }, { "left": "pull_request:6994", "right": "pull_request:7216", "accept": false, - "reason": "Both involve single-file loading, but one fixes `torch_dtype` propagation and the other updates the API signature/docs; not the same concrete fix." + "reason": "Passing torch_dtype through single-file loaders vs updating from_single_file signature/docs; related area, but not the same concrete fix." }, { "left": "pull_request:7410", "right": "pull_request:8761", "accept": false, - "reason": "FreeInit behavior fix vs Hunyuan embedding refactor; unrelated." + "reason": "FreeInit behavior fix vs Hunyuan embedding refactor; unrelated pipeline internals." }, { "left": "pull_request:7458", "right": "pull_request:9307", "accept": false, - "reason": "Skipping dynamo tests on Python 3.12 vs rotary embedding CPU/device refactor; unrelated." + "reason": "Skipping dynamo tests on Python 3.12 vs rotary embedding CPU/device refactor; different problems." }, { "left": "pull_request:7662", "right": "pull_request:8512", "accept": false, - "reason": "Flax PR test cleanup vs removing a deprecated inpaint helper; unrelated." + "reason": "Flax CI cleanup vs removing deprecated prepare_mask_and_masked_image; unrelated workflows/code paths." }, { "left": "pull_request:8618", "right": "pull_request:8755", "accept": false, - "reason": "SD3 tiny autoencoder `shift_factor` addition vs SD3 Dreambooth LoRA training fix; different code paths." + "reason": "SD3 tiny autoencoder shift_factor addition vs SD3 DreamBooth LoRA training fix; same product family, different components." }, { "left": "pull_request:10359", "right": "pull_request:11469", "accept": false, - "reason": "Both touch tests, but they target different pipelines and different xfail issues." + "reason": "General test fixes vs targeted xfails for specific pipeline methods; not the same test failure." }, { "left": "pull_request:449", "right": "pull_request:676", "accept": false, - "reason": "Similar variable-renaming style only; they are different refactors in different modules." + "reason": "Variable renaming in attention vs variable renaming in resnet; stylistically similar, but separate refactors." }, { "left": "pull_request:10136", "right": "pull_request:10567", "accept": false, - "reason": "Same Hunyuan Video area, but one is the base implementation and the other adds true CFG support; follow-up feature, not the same change." + "reason": "Hunyuan Video core model/pipeline addition vs adding true CFG support to that pipeline; same module, different feature." }, { "left": "pull_request:10810", "right": "pull_request:12151", "accept": false, - "reason": "LoRA Flux support vs Flax deprecation cleanup; unrelated." + "reason": "Flux LoRA conversion support vs deprecating Flax support; completely different subsystems." + }, + { + "left": "pull_request:10985", + "right": "pull_request:11999", + "accept": false, + "reason": "Flux ComfyUI LoRA conversion support vs new LoRA config injection method; related theme, but not the same concrete change." } ] }, @@ -150159,11 +150242,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 146, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -150171,12 +150254,12 @@ "pull_request:10246", "pull_request:10418", "pull_request:10808", - "pull_request:10985", "pull_request:11089", "pull_request:11250", + "pull_request:11366", "pull_request:11421", + "pull_request:11535", "pull_request:11896", - "pull_request:11999", "pull_request:12142", "pull_request:12143", "pull_request:1513", @@ -150216,7 +150299,6 @@ "pull_request:9909" ], "soft_pairs": [ - "pull_request:10985|pull_request:11999", "pull_request:3410|pull_request:3846", "pull_request:4622|pull_request:4635", "pull_request:5053|pull_request:5439", @@ -150240,192 +150322,187 @@ "pull_request:5713|pull_request:6828", "pull_request:10119|pull_request:11089", "pull_request:10246|pull_request:9909", - "pull_request:10418|pull_request:11250" + "pull_request:10418|pull_request:11250", + "pull_request:11366|pull_request:11535" ], - "prepared_review_unit_hash": "64a5eb0c6e129cfa9a2794514a065de32263829bdb54035d41587004bafb74b7", + "prepared_review_unit_hash": "ca41964ca9c5b43d080c7adb9d20aad55e7f21e3a713f34495a948748acbb584", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 100136, - "estimated_input_tokens": 25034, - "estimated_eval_tokens": 50324 + "serialized_chars": 99143, + "estimated_input_tokens": 24786, + "estimated_eval_tokens": 49828 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous and not a real duplicate set: docs-only PRs, unrelated training-script fixes, and separate LoRA/Wan/Pipeline changes. The only strong duplicate-like pair is the two Wan VACE fp16/bf16 mask-cast fixes, which appear to address the same bug with one adding tests.", - "confidence": 0.88, + "summary": "This cluster is mostly a false-positive similarity group: the PRs span unrelated docs, example scripts, workflows, and separate model/pipeline fixes. None of the suggested soft edges looks like the same underlying change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR fits the whole cluster because most items are unrelated changes; if forced to pick within the only near-duplicate subpair, PR #12143 is the more complete merged fix (same source change plus tests).", + "canonical_pr_reason": "No canonical PR: the items are too heterogeneous and do not converge on one concrete bug or feature change.", "best_issue_reason": null, - "best_pr_reason": "PR #12143 is the strongest representative of an actual bug fix in the cluster: it makes the Wan VACE mask dtype correction and includes a test, whereas the paired PR #12142 is an incomplete/unmerged variant.", + "best_pr_reason": "No single PR is a good global representative; the cluster is a mix of unrelated changes, so any one choice would be arbitrary.", "soft_edge_verdicts": [ - { - "left": "pull_request:10985", - "right": "pull_request:11999", - "accept": false, - "reason": "Both are LoRA-related, but they fix different problems: ComfyUI/Flux LoRA conversion support vs. adapting to a new PEFT LoRA injection API. Same subsystem, not the same change." - }, { "left": "pull_request:3410", "right": "pull_request:3846", "accept": false, - "reason": "Both touch training utilities, but one fixes DeepSpeed plugin retrieval and the other fixes random seeding import usage. No shared bug." + "reason": "Both touch training utilities, but one fixes DeepSpeed plugin retrieval and the other changes random-module seeding; different bugs and code paths." }, { "left": "pull_request:4622", "right": "pull_request:4635", "accept": false, - "reason": "Both are docs/controlnet-adjacent, but one adds PushToHub docs and the other adds MultiControlNet docs/examples. Different features." + "reason": "Docs-only similarity is too broad; PushToHub docs and MultiControlNet docs/example are unrelated changes." }, { "left": "pull_request:5053", "right": "pull_request:5439", "accept": false, - "reason": "Validation-image argument handling in a training script is unrelated to the from_single_file tokenizer/config loading regression." + "reason": "One fixes an argument parsing interface in a training script, the other fixes tokenizer/config loading in from_single_file; unrelated." }, { "left": "pull_request:5823", "right": "pull_request:7058", "accept": false, - "reason": "Both are LoRA loader work, but one adds adapter_names to fuse_lora and the other copies an input state dict before load_lora_weights. Different code paths and fixes." + "reason": "Both are LoRA-related, but they fix different mechanisms: adapter fusion by names vs copying an input state dict before loading." }, { "left": "pull_request:5928", "right": "pull_request:6328", "accept": false, - "reason": "Both add community pipelines, but they are different pipelines (AnimateDiff+ControlNet vs Latent Consistency interpolation). Not duplicates." + "reason": "One adds a community README entry, the other adds a new AnimateDiff video-to-video pipeline; not the same concrete change." }, { "left": "pull_request:6289", "right": "pull_request:6698", "accept": false, - "reason": "Both introduce new video-related models/pipelines, but one is SparseCntrl for AnimateDiff and the other is PIA. Distinct features." + "reason": "AnimateDiff SparseCntrl and PIA are distinct model/pipeline additions with different architectures and goals." }, { "left": "pull_request:9212", "right": "pull_request:9234", "accept": false, - "reason": "One fixes the auto-pipeline mapping for Kolors img2img, the other fixes safetensors compatibility regression. Shared utility area only." + "reason": "Kolors img2img auto-pipeline mapping and safetensors compatibility regression are unrelated code paths." }, { "left": "pull_request:11421", "right": "pull_request:12142", "accept": false, - "reason": "One changes DreamBooth image interpolation defaults; the other fixes Wan VACE fp16/bf16 mask handling. No overlap in bug or code path." + "reason": "Dreambooth interpolation-mode handling and Wan VACE bf16/fp16 mask casting are unrelated fixes." }, { "left": "pull_request:11421", "right": "pull_request:12143", "accept": false, - "reason": "Same as above: unrelated training-image resizing default vs. Wan VACE dtype fix." + "reason": "These are separate concerns: an interpolation CLI option vs a Wan VACE dtype fix; not the same bug." }, { "left": "pull_request:1513", "right": "pull_request:152", "accept": false, - "reason": "Both are docs/examples, but one adds notebook bootstrap config and the other adds local image-folder training support. Different changes." + "reason": "Dreambooth notebook setup docs and local-image-folder training support are different documentation/features." }, { "left": "pull_request:1913", "right": "pull_request:327", "accept": false, - "reason": "Input-type validation for img2img pipelines is unrelated to type-hint cleanup in the PNDM pipeline." + "reason": "Input-type validation across image pipelines is unrelated to type-hint cleanup for the PNDM pipeline." }, { "left": "pull_request:2398", "right": "pull_request:2641", "accept": false, - "reason": "Pix2Pix Zero docs expansion and documentation-build workflow changes are unrelated." + "reason": "Pix2Pix Zero docs and documentation-build workflow changes are unrelated." }, { "left": "pull_request:2641", "right": "pull_request:3707", "accept": false, - "reason": "Both touch documentation workflows, but one adds notebook build config and the other changes doc-comment secret handling. Different issues." + "reason": "Both affect docs workflows, but one adds notebook build plumbing and the other changes secret handling/deletion workflow; not the same change." }, { "left": "pull_request:3042", "right": "pull_request:3681", "accept": false, - "reason": "Accelerate minimum-version bumps are separate from checkpoint-limit handling changes in training scripts." + "reason": "Accelerate version pinning and manual checkpoint-limit handling solve different training-script issues." }, { "left": "pull_request:6225", "right": "pull_request:6743", "accept": false, - "reason": "LoRA alpha parsing fixes and hflip-argument correction are unrelated training-script fixes." + "reason": "LoRA alpha parsing and hflip argument correction are separate training-script bugs." }, { "left": "pull_request:6743", "right": "pull_request:6918", "accept": false, - "reason": "Different problems: one corrects a flip-argument name, the other blocks hub-token exposure with wandb. No duplicate evidence." + "reason": "A flip-flag fix and hub-token exposure guard are unrelated." }, { "left": "pull_request:5438", "right": "pull_request:5477", "accept": false, - "reason": "Both concern latent-consistency community examples, but they add different pipelines (txt2img vs img2img). Not the same change." + "reason": "These are related only at the topic level; one adds a community txt2img entry, the other adds an img2img pipeline implementation." }, { "left": "pull_request:7246", "right": "pull_request:7771", "accept": false, - "reason": "One fixes VAE tiling device selection in tests; the other marks a ControlNetXS test as flaky. Different test concerns." + "reason": "One fixes VAE tiling test device usage, the other marks a ControlNetXS test flaky; different test issues." }, { "left": "pull_request:10808", "right": "pull_request:11896", "accept": false, - "reason": "Both deal with LoRA/offload behavior, but one restricts PEFT config update keys and the other fixes hook disabling when loading LoRAs. Not the same bug." + "reason": "Both concern LoRA/offloading, but one restricts PEFT config key updates while the other fixes hook removal during loading; distinct bugs." }, { "left": "pull_request:4695", "right": "pull_request:5057", "accept": false, - "reason": "One adds TextToVideoZeroSDXLPipeline; the other adds clip-skip support to SDXL pipelines. Different features." + "reason": "Text-to-Video Zero SDXL support and SDXL clip-skip support are separate features." }, { "left": "pull_request:5713", "right": "pull_request:6828", "accept": false, - "reason": "Both involve IP-adapter-related code, but one is the full IP-Adapter feature set and the other adjusts UNetMotionModel IP-adapter handling. Not duplicate enough to merge." + "reason": "IP-Adapter feature work and a UNetMotionModel IP-adapter code fix are not the same concrete change." }, { "left": "pull_request:10119", "right": "pull_request:11089", "accept": false, - "reason": "Docs for load_lora_adapter and Wan docs updates are unrelated." + "reason": "Docs for load_lora_adapter and Wan memory-optimization docs are unrelated." }, { "left": "pull_request:10246", "right": "pull_request:9909", "accept": false, - "reason": "Both concern attention processors, but one adds/updates docs and processor exports while the other expands the AttentionProcessor union type. Related area, different concrete change." + "reason": "One updates docs for attention processors, the other changes the type union in code; related topic, but not the same PR-level change." }, { "left": "pull_request:10418", "right": "pull_request:11250", "accept": false, - "reason": "Internal link fixes in a tutorial are unrelated to adding AutoModel docs and nav entries." + "reason": "Internal-link docs cleanup and AutoModel docs addition are unrelated documentation changes." }, { - "left": "pull_request:12142", - "right": "pull_request:12143", - "accept": true, - "reason": "These appear to fix the same Wan VACE fp16/bf16 mask-cast bug in the same file; 12143 is the more complete version because it includes the source fix plus tests." + "left": "pull_request:11366", + "right": "pull_request:11535", + "accept": false, + "reason": "Hidream transformer output shape change and Framepack initialization-order change affect different models and code paths." } ] }, @@ -150441,16 +150518,16 @@ "component_index": 1, "component_count": 266, "review_unit_index": 147, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:11366", - "pull_request:11535", + "pull_request:11588", + "pull_request:11913", "pull_request:12425", "pull_request:13022", "pull_request:1494", @@ -150499,7 +150576,6 @@ "pull_request:9735" ], "soft_pairs": [ - "pull_request:11366|pull_request:11535", "pull_request:12425|pull_request:13022", "pull_request:1494|pull_request:2655", "pull_request:1564|pull_request:3561", @@ -150524,192 +150600,193 @@ "pull_request:6961|pull_request:8386", "pull_request:7627|pull_request:9447", "pull_request:8445|pull_request:9447", - "pull_request:8591|pull_request:9735" + "pull_request:8591|pull_request:9735", + "pull_request:11588|pull_request:11913" ], - "prepared_review_unit_hash": "99f0ac1a43732b432d79b7526f36db6be17900ffa6e4d0417b8f27d16a687ff5", + "prepared_review_unit_hash": "dc745efae11436d7b0047c986fd26e6ade5de343a2378559da107f250b07389b", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 94205, - "estimated_input_tokens": 23552, - "estimated_eval_tokens": 47360 + "serialized_chars": 95419, + "estimated_input_tokens": 23855, + "estimated_eval_tokens": 47966 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The items are a heterogeneous mix of unrelated docs, tests, training-script tweaks, and core bugfixes; no pair looks like a true duplicate PR fix, so I would not merge any of the soft-edge candidates.", + "summary": "This cluster is mostly a grab-bag of unrelated PRs across docs, tests, packaging, and a few feature/fix areas. The soft edges share broad keywords at best, but they do not line up on the same concrete bug or merged change path.", "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR stands out: the set spans unrelated subsystems and change types, so there isn\u2019t one PR that clearly represents the whole cluster.", + "canonical_pr_reason": "No single PR is a good canonical duplicate target here; the items span many unrelated changes, so any one PR would be misleading as a representative.", "best_issue_reason": null, - "best_pr_reason": "PR 8452 is the strongest representative if one must be chosen: it targets an explicit issue and makes a concrete loader fix in a core code path, unlike the more docs/test-oriented or broad refactors.", + "best_pr_reason": "PR 11913 is the broadest, most self-contained feature PR in the set (Wan modular diffusers), so it is the least arbitrary single PR to use as a representative if one must be chosen.", "soft_edge_verdicts": [ - { - "left": "pull_request:11366", - "right": "pull_request:11535", - "accept": false, - "reason": "Different features and code paths: Hidream transformer output shape vs Framepack initialization order." - }, { "left": "pull_request:12425", "right": "pull_request:13022", "accept": false, - "reason": "Both touch distributed/cuda-related plumbing, but one is a Python import guard and the other is Dockerfile dependency/version setup; not the same bug or change." + "reason": "Both touch import/platform issues, but one is a torch.distributed guard and the other is Dockerfile dependency pinning; unrelated changes." }, { "left": "pull_request:1494", "right": "pull_request:2655", "accept": false, - "reason": "Both involve save/load semantics, but one adds safetensors save support while the other adjusts LoRA loading behavior; different concrete problems." + "reason": "Safe serialization support and LoRA loading fixes are different code paths and different problems." }, { "left": "pull_request:1564", "right": "pull_request:3561", "accept": false, - "reason": "Docs-only accelerate config guidance vs a Dreambooth IF training feature addition; unrelated changes." + "reason": "Docs for accelerate defaults versus Dreambooth IF upscaling logic; not the same bug or change." }, { "left": "pull_request:1564", "right": "pull_request:748", "accept": false, - "reason": "README docs for accelerate config vs a push_to_hub bugfix in training scripts; not the same underlying issue." + "reason": "One is a docs tweak, the other fixes push_to_hub behavior in training scripts; different scope." }, { "left": "pull_request:1788", "right": "pull_request:711", "accept": false, - "reason": "DDIM prediction-type bugfix vs a scheduler design proposal; different scope and code paths." + "reason": "A concrete DDIM sampling bug fix does not match a broad scheduler design proposal." }, { "left": "pull_request:1850", "right": "pull_request:2470", "accept": false, - "reason": "Test additions for ResnetBlock2D vs MPS-specific test skips; unrelated test maintenance." + "reason": "Both are tests, but they cover different models and different MPS-related adjustments." }, { "left": "pull_request:2430", "right": "pull_request:715", "accept": false, - "reason": "One changes version-reporting text in pipeline utils; the other makes utils imports conditional on PyTorch availability; different issues." + "reason": "One changes transformers version labeling, the other guards PyTorch imports; unrelated." }, { "left": "pull_request:2464", "right": "pull_request:3054", "accept": false, - "reason": "Dreambooth training-script cleanup vs documentation punctuation/formatting; no shared bug or implementation change." + "reason": "Training-script xformers handling is unrelated to a docs punctuation/contribution edit." }, { "left": "pull_request:2557", "right": "pull_request:748", "accept": false, - "reason": "Both are Dreambooth-related, but one adds validation logging and the other fixes hub push invocation; not the same code-path problem." + "reason": "Both involve Dreambooth scripts, but one adds validation logging and the other fixes hub pushing; separate changes." }, { "left": "pull_request:2593", "right": "pull_request:833", "accept": false, - "reason": "ControlNet conversion tooling vs a package version bump; unrelated." + "reason": "ControlNet conversion tooling and a version-bump release PR are not duplicates." }, { "left": "pull_request:3241", "right": "pull_request:5289", "accept": false, - "reason": "Both are test adjustments, but one changes an expected diff threshold and the other changes a save/load tolerance; different failures." + "reason": "Both are test adjustments, but for different pipelines and different failure modes." }, { "left": "pull_request:3852", "right": "pull_request:4215", "accept": false, - "reason": "Separate documentation updates for different topics; no duplicate change." + "reason": "Different documentation edits in different files with no shared underlying issue." }, { "left": "pull_request:4498", "right": "pull_request:6114", "accept": false, - "reason": "Both are community/example related, but one adds SDXL reference support while the other extends lpw-xl img2img/inpaint; distinct features." + "reason": "Both mention SDXL community examples, but one adds a reference demo while the other adds img2img/inpaint support; different features." }, { "left": "pull_request:4787", "right": "pull_request:6762", "accept": false, - "reason": "LoRA loading relaxation and a flaky Kandinsky test fix are unrelated." + "reason": "LoRA loading filename filtering and a flaky Kandinsky test fix are unrelated." }, { "left": "pull_request:5220", "right": "pull_request:6261", "accept": false, - "reason": "Test-suite scheduling changes vs moving Attend and Excite out of stable_diffusion; unrelated." + "reason": "One retags slow tests, the other refactors pipeline exports; no shared concrete change." }, { "left": "pull_request:5639", "right": "pull_request:6340", "accept": false, - "reason": "AnimateDiff docs update vs conversion scripts; same area broadly, but not the same underlying fix." + "reason": "Motion LoRA docs and AnimateDiff conversion scripts are different artifacts and different tasks." }, { "left": "pull_request:5946", "right": "pull_request:6762", "accept": false, - "reason": "Refactoring example tests vs marking one test flaky; separate test-maintenance tasks." + "reason": "A broad example-test refactor is unrelated to a single flaky Kandinsky test adjustment." }, { "left": "pull_request:6065", "right": "pull_request:6643", "accept": false, - "reason": "Docs for SDXL Turbo vs docs for UViT2D; unrelated additions." + "reason": "SDXL Turbo docs and UViT2D docs are separate documentation additions." }, { "left": "pull_request:6147", "right": "pull_request:8452", "accept": false, - "reason": "Both concern single-file loading, but one adds refiner-model support and the other loosens strict loading for errors; different concrete code-path issues." + "reason": "Both are about single-file loading, but they address different failures and different code paths; too broad to merge as duplicates." }, { "left": "pull_request:6195", "right": "pull_request:6405", "accept": false, - "reason": "AnimateDiff SDXL WIP vs doc-builder workflow testing; unrelated." + "reason": "A WIP AnimateDiff SDXL feature PR and a doc-builder workflow test PR are unrelated." }, { "left": "pull_request:6775", "right": "pull_request:7779", "accept": false, - "reason": "Two docstring fixes in different APIs; not the same change." + "reason": "Both are docstring/docs fixes, but for different APIs and different issues." }, { "left": "pull_request:6961", "right": "pull_request:8386", "accept": false, - "reason": "Single-file loader tests vs model test input-shape refactors; different test concerns." + "reason": "Single-file checkpoint loader tests and output_shape-based model test cleanup are unrelated test maintenance." }, { "left": "pull_request:7627", "right": "pull_request:9447", "accept": false, - "reason": "Playground img2img latent handling vs pipeline save_pretrained shard-size defaults; unrelated." + "reason": "An img2img pipeline fix for Playground and a save_pretrained shard-size default change do not share the same bug." }, { "left": "pull_request:8445", "right": "pull_request:9447", "accept": false, - "reason": "Both touch checkpoint sharding defaults, but one changes model save_pretrained and the other changes pipeline save_pretrained; similar theme, not the same concrete fix." + "reason": "Both alter max_shard_size defaults, but one changes model saving shards and the other changes pipeline save semantics; not the same concrete fix." }, { "left": "pull_request:8591", "right": "pull_request:9735", "accept": false, - "reason": "SD3 training utility refactor vs an SD3 ControlNet example update; related model family, but different artifacts and problems." + "reason": "Both are SD3 training-related, but one refactors weighting utilities while the other updates a ControlNet example/training script; different changes." + }, + { + "left": "pull_request:11588", + "right": "pull_request:11913", + "accept": false, + "reason": "Both are Wan-related, but one adds LoRA support for I2V while the other introduces the modular Wan pipeline; separate code paths and features." } ] }, @@ -150725,11 +150802,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 148, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -150752,14 +150829,14 @@ "pull_request:11273", "pull_request:11290", "pull_request:1147", - "pull_request:11588", "pull_request:11614", "pull_request:11728", "pull_request:11823", - "pull_request:11913", "pull_request:12062", + "pull_request:1215", "pull_request:12419", "pull_request:12425", + "pull_request:13364", "pull_request:1496", "pull_request:2915", "pull_request:2956", @@ -150775,6 +150852,7 @@ "pull_request:7714", "pull_request:7771", "pull_request:7941", + "pull_request:833", "pull_request:8412", "pull_request:8709", "pull_request:8755", @@ -150782,7 +150860,6 @@ "pull_request:9721" ], "soft_pairs": [ - "pull_request:11588|pull_request:11913", "pull_request:8709|pull_request:9099", "pull_request:10376|pull_request:11066", "pull_request:10117|pull_request:10704", @@ -150806,186 +150883,194 @@ "pull_request:11290|pull_request:12425", "pull_request:1147|pull_request:2915", "pull_request:11614|pull_request:11728", - "pull_request:11614|pull_request:12062" + "pull_request:11614|pull_request:12062", + "pull_request:1215|pull_request:833", + "pull_request:12425|pull_request:13364" ], - "prepared_review_unit_hash": "8171a21a6c6223b0d7c6475da6d14f934a01e5b9f0603ca3df624a0f5137de56", + "prepared_review_unit_hash": "1883a9ff8f6b0880473f00a22e5dce8359289c72ac35067fc18cbe70338bee08", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 25, - "serialized_chars": 96108, - "estimated_input_tokens": 24027, - "estimated_eval_tokens": 48310 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 97146, + "estimated_input_tokens": 24287, + "estimated_eval_tokens": 48830 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated PRs that were grouped by superficial similarity rather than shared bug/change. The cluster spans Wan/Hunyuan/SD3 feature work, docs, CI, and test-only tweaks; none of the soft edges look like true duplicates.", + "summary": "This cluster is not a duplicate set; it\u2019s a grab bag of unrelated PRs spanning docs, CI/workflows, model code, LoRA conversions, and tests. All soft edges look like superficial similarity only, not the same underlying bug/change.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "If a representative is needed, PR 10922 is the broadest substantive feature change in the main Wan theme; however, it is not a duplicate of the others.", + "canonical_pr_reason": "No canonical PR: the items are heterogeneous and do not resolve to one underlying fix or feature.", "best_issue_reason": null, - "best_pr_reason": "PR 10922 is the strongest standalone representative because it is a large, central feature integration with real code and API surface changes. The rest are either smaller follow-ups or unrelated maintenance/docs/test updates.", + "best_pr_reason": "No single PR is a good global representative for deduping this cluster; the PRs cover unrelated areas and changes.", "soft_edge_verdicts": [ - { - "left": "pull_request:11588", - "right": "pull_request:11913", - "accept": false, - "reason": "Both are Wan-related, but one is LoRA I2V support and the other is the modular Wan pipeline; different code paths and different change intents." - }, { "left": "pull_request:8709", "right": "pull_request:9099", "accept": false, - "reason": "Both are SD3 inpainting-related, but one adds the SD3 inpaint pipeline and the other adds SD3 controlnet inpainting; distinct features, not one bug fix." + "reason": "Both are SD3-related pipeline additions, but they target different features: SD3 inpaint vs SD3 ControlNet inpainting." }, { "left": "pull_request:10376", "right": "pull_request:11066", "accept": false, - "reason": "HunyuanVideo LoRA conversion support and the new HunyuanVideo-I2V model pipeline are separate feature areas." + "reason": "Both touch HunyuanVideo, but one is LoRA format conversion support and the other is a new HunyuanVideo-I2V model/pipeline." }, { "left": "pull_request:10117", "right": "pull_request:10704", "accept": false, - "reason": "Both are docs PRs, but one fixes README links and the other corrects img2img usage examples; unrelated documentation changes." + "reason": "Both are docs fixes, but one repairs broken README links while the other updates image-to-image usage examples." }, { "left": "pull_request:10326", "right": "pull_request:10354", "accept": false, - "reason": "One fixes MPS workflow quoting; the other patches a Flux IP-Adapter tester by re-applying model dtype/device. Different problems." + "reason": "Unrelated CI/test changes: one fixes an MPS workflow install command, the other fixes a Flux IP-Adapter test helper." }, { "left": "pull_request:2959", "right": "pull_request:3410", "accept": false, - "reason": "UNet3D norm-group test adjustment is unrelated to the DeepSpeed plugin retrieval fix in training code." + "reason": "Different test fixes in different code paths: UNet3D norm-group test vs DeepSpeed plugin retrieval in training examples." }, { "left": "pull_request:4927", "right": "pull_request:4963", "accept": false, - "reason": "These are a revert and re-revert around accelerate/offloading behavior, but they are not the same concrete change; 4963 also adds accelerate main to CI." + "reason": "Both involve offloading/loaders, but one reverts accelerate hook handling while the other reinstalls accelerate main in workflows; not the same change." }, { "left": "pull_request:6645", "right": "pull_request:7086", "accept": false, - "reason": "Both touch community docs, but one is a README cleanup and the other adds an AnimateDiff community pipeline and controlnet updates." + "reason": "Both update community docs, but they cover different pipelines and different functional content." }, { "left": "pull_request:7714", "right": "pull_request:8412", "accept": false, - "reason": "AutoPipeline docs overhaul and single-file loader docs are different documentation topics with different APIs." + "reason": "Both are docs work, but AutoPipeline documentation and single-file documentation are distinct topics." }, { "left": "pull_request:7771", "right": "pull_request:7941", "accept": false, - "reason": "One marks a UNetControlNetXS test flaky; the other marks a StableDiffusion single-file slow test. Only a generic test-maintenance similarity." + "reason": "Both are test annotations, but one marks a ControlNetXS test flaky while the other marks a StableDiffusion21 single-file class as slow." }, { "left": "pull_request:11823", "right": "pull_request:12419", "accept": false, - "reason": "Both are LoRA-related, but one supports Fal Kontext LoRA conversion and the other adapts Flux modular pipeline metadata for Mellon; unrelated implementations." + "reason": "Both are LoRA-related, but they support different model families and conversion paths: Flux Kontext vs modular Flux readiness." }, { "left": "pull_request:6490", "right": "pull_request:6752", "accept": false, - "reason": "Different community projects: RAVE video editing versus PromptDiffusion pipeline docs/code." + "reason": "Both add community/research project content, but they introduce different projects and codepaths." }, { "left": "pull_request:10376", "right": "pull_request:10482", "accept": false, - "reason": "HunyuanVideo LoRA conversion support and a PyTorch<2.5 NaN fix in the transformer are unrelated issues." + "reason": "Both are HunyuanVideo changes, but one adds LoRA conversion support and the other fixes a NaN/attention mask bug." }, { "left": "pull_request:2956", "right": "pull_request:3020", "accept": false, - "reason": "Euler scheduler Karras-sigmas addition and DPM-SDE stochastic sampler introduction are different scheduler features." + "reason": "Both relate to schedulers, but they implement different scheduler features in different algorithms." }, { "left": "pull_request:10069", "right": "pull_request:11273", "accept": false, - "reason": "BnB quantizer device handling and Slack release bot timeout changes are unrelated code paths." + "reason": "Completely different concerns: BnB quantizer device handling vs notification bot timeout behavior." }, { "left": "pull_request:10126", "right": "pull_request:8755", "accept": false, - "reason": "LoRA save deprecation and SD3 DreamBooth LoRA training fixes are different changes in different subsystems." + "reason": "Different domains and fixes: LoRA API deprecation vs SD3 DreamBooth LoRA training logic." }, { "left": "pull_request:10271", "right": "pull_request:11052", "accept": false, - "reason": "TorchAO serialization test refactor and quanto test capability gating are separate test infrastructure changes." + "reason": "TorchAO test refactor vs a new compute-capability test decorator; related to tests but not the same bug/change." }, { "left": "pull_request:1089", "right": "pull_request:1496", "accept": false, - "reason": "Old scheduler docs addition versus Habana Gaudi docs addition; both docs-only but unrelated topics." + "reason": "Both are docs, but one expands scheduler usage docs and the other adds Habana Gaudi documentation." }, { "left": "pull_request:10922", "right": "pull_request:11059", "accept": false, - "reason": "Wan2.1 core integration versus Wan LoRA conversion support for non-diffusers checkpoints; different layers of the stack." + "reason": "Same broad Wan area, but one adds Wan model/pipeline support and the other converts non-diffusers Wan LoRAs; different changes." }, { "left": "pull_request:11067", "right": "pull_request:9721", "accept": false, - "reason": "Benchmark transformers pinning and Docker torch pinning are both environment changes, but they target different setups and failure modes." + "reason": "Both are Docker/benchmark config changes, but one pins transformers for benchmarking and the other pins torch versions across Dockerfiles." }, { "left": "pull_request:11196", "right": "pull_request:11273", "accept": false, - "reason": "Model save_pretrained safetensors behavior and Slack bot request timeout are unrelated maintenance fixes." + "reason": "Model save_pretrained behavior vs Slack release-notification timeout; unrelated code paths." }, { "left": "pull_request:11290", "right": "pull_request:12425", "accept": false, - "reason": "Onnx Docker torch version unpinning and distributed import guarding are unrelated infrastructure fixes." + "reason": "Dockerfile dependency pinning vs torch.distributed import guarding; no shared underlying bug." }, { "left": "pull_request:1147", "right": "pull_request:2915", "accept": false, - "reason": "Flax example text-encoder loading fix and a Stable Diffusion test threshold adjustment are not the same change." + "reason": "Flax example loading fix vs a tolerance adjustment in a slow test; unrelated." }, { "left": "pull_request:11614", "right": "pull_request:11728", "accept": false, - "reason": "Docker workflow bugfix versus adding quant libs to the doc-builder Docker image; different CI/container concerns." + "reason": "CI workflow hardening vs doc-builder Dockerfile adding quant libs; different files and goals." }, { "left": "pull_request:11614", "right": "pull_request:12062", "accept": false, - "reason": "Build-docker workflow cleanup and enabling all GPUs in CI are unrelated workflow changes." + "reason": "Both touch CI, but one is a workflow loop/quoting fix and the other enables all GPUs across multiple workflows; not the same issue." + }, + { + "left": "pull_request:1215", + "right": "pull_request:833", + "accept": false, + "reason": "DDIM generator/device bugfix vs a version bump release PR; unrelated." + }, + { + "left": "pull_request:12425", + "right": "pull_request:13364", + "accept": false, + "reason": "Torch distributed import guarding vs MotionConv2d dtype casting; completely different code paths and fixes." } ] }, @@ -151001,18 +151086,18 @@ "component_index": 1, "component_count": 266, "review_unit_index": 149, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:1215", - "pull_request:12425", + "pull_request:10132", + "pull_request:11044", + "pull_request:11767", "pull_request:1331", - "pull_request:13364", "pull_request:1706", "pull_request:2464", "pull_request:2892", @@ -151047,20 +151132,18 @@ "pull_request:7941", "pull_request:816", "pull_request:8267", - "pull_request:833", "pull_request:8370", "pull_request:8452", "pull_request:8496", "pull_request:8591", "pull_request:8652", + "pull_request:8660", "pull_request:8719", "pull_request:8761", "pull_request:8937", "pull_request:8973" ], "soft_pairs": [ - "pull_request:1215|pull_request:833", - "pull_request:12425|pull_request:13364", "pull_request:1331|pull_request:3337", "pull_request:1706|pull_request:2464", "pull_request:2892|pull_request:5053", @@ -151084,192 +151167,188 @@ "pull_request:8267|pull_request:8719", "pull_request:8591|pull_request:8761", "pull_request:8652|pull_request:8973", - "pull_request:6752|pull_request:7442" + "pull_request:6752|pull_request:7442", + "pull_request:10132|pull_request:8660", + "pull_request:11044|pull_request:11767" ], - "prepared_review_unit_hash": "cfc4f303a5ac6b0fb9fdd2e2e16f41c8a702c9db0357427040edcc5684085b05", + "prepared_review_unit_hash": "13d06b5c8209c5ad521343c130fb53d8997226349f53ac9dcaccd93fbdb3151a", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 99618, - "estimated_input_tokens": 24905, - "estimated_eval_tokens": 50066 + "serialized_chars": 100496, + "estimated_input_tokens": 25124, + "estimated_eval_tokens": 50504 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose grab bag of largely unrelated Diffusers PRs. A few pairs share broad areas (inpainting, LoRA, docs, Lumina), but the concrete bugs/changes differ, so none of the soft edges should be merged as duplicates.", - "confidence": 0.92, + "summary": "No duplicate cluster emerges here: the soft edges mostly connect unrelated PRs that happen to touch similar subsystems, docs, or tests. I would not merge any of these pairs as duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No issue artifacts are present, so there is no canonical issue to pick. As a cluster representative, PR 6119 is the clearest substantive bug-fix PR with a focused change and strong problem statement (fp16 LoRA training instability), making it a reasonable canonical PR.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 6119 is the best overall representative because it is a concrete, well-scoped code fix rather than a doc/test-only tweak or broad refactor, and its title/body clearly describe the underlying bug and fix.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:1215", - "right": "pull_request:833", - "accept": false, - "reason": "Different changes: DDIM eta>0 CPU generator handling vs release version bump. No shared bug or code path." - }, - { - "left": "pull_request:12425", - "right": "pull_request:13364", - "accept": false, - "reason": "Unrelated import guard for torch.distributed vs MotionConv2d blur-kernel dtype casting." - }, { "left": "pull_request:1331", "right": "pull_request:3337", "accept": false, - "reason": "Both touch inpainting, but one fixes ONNX inpaint scheduler input ordering and the other is a broad postprocessing refactor across many pipelines; not the same concrete bug." + "reason": "Both touch inpaint/postprocessing, but 1331 fixes ONNX inpainting scheduler input order while 3337 is a broader postprocessing refactor across multiple pipelines." }, { "left": "pull_request:1706", "right": "pull_request:2464", "accept": false, - "reason": "Docs update vs training-script xformers cleanup; different artifacts and no common fix." + "reason": "One is main docs cleanup; the other removes xformers training code from a LoRA example. Same repo area, different changes." }, { "left": "pull_request:2892", "right": "pull_request:5053", "accept": false, - "reason": "Upscaler input validation/embed support vs t2i-adapter validation_image nargs fix; unrelated pipelines and bugs." + "reason": "Upscaler input validation for embeds is unrelated to the t2i-adapter validation_image CLI argument change." }, { "left": "pull_request:2968", "right": "pull_request:816", "accept": false, - "reason": "Img2img post-processing bug fix vs deepspeed fp16 loss casting in training; different code paths." + "reason": "Both are pipeline/training fixes, but one is img2img post-processing and the other is deepspeed float casting in text-to-image training." }, { "left": "pull_request:3337", "right": "pull_request:4272", "accept": false, - "reason": "Broad postprocessing refactor vs optional import cleanup; same repo area but not the same change or bug." + "reason": "Postprocessing refactor is unrelated to optional import cleanup; different goals and code paths." }, { "left": "pull_request:3397", "right": "pull_request:5111", "accept": false, - "reason": "Docs note on local directory path vs k-diffusion scheduler support; unrelated." + "reason": "Docs note for local directory paths has nothing to do with K-Diffusion generator passthrough or scheduler changes." }, { "left": "pull_request:3589", "right": "pull_request:3590", "accept": false, - "reason": "Both mention community README content, but one adds reference-trick docs/examples and the other adds a new canvas pipeline; not duplicates." + "reason": "Both update community docs, but they add different community pipelines/features and are not the same change." }, { "left": "pull_request:3590", "right": "pull_request:5090", "accept": false, - "reason": "Stable Diffusion mixture/canvas feature addition vs t2i-adapter community fix; different features and code paths." + "reason": "Stable Diffusion Mixture Canvas docs/code are unrelated to T2I-Adapter bugfixes." }, { "left": "pull_request:4272", "right": "pull_request:6081", "accept": false, - "reason": "Optional import statements cleanup vs SDXL sigma conversion for torch.compile; unrelated." + "reason": "Optional import statement fixes are unrelated to the SDXL sigma/torch.compile pipeline tweak." }, { "left": "pull_request:4669", "right": "pull_request:4787", "accept": false, - "reason": "Both involve LoRA loading, but one fixes per-text-encoder rank handling and the other tightens file selection heuristics; different bugs." + "reason": "Both are LoRA-related loader changes, but one handles mixed LoRA ranks and the other tightens checkpoint file filtering." }, { "left": "pull_request:4783", "right": "pull_request:6007", "accept": false, - "reason": "SDXL prompt-embedding docs change vs docs about the copied-from mechanism; not the same issue." + "reason": "SDXL embedding docs and the copied-from docs refactor are different documentation changes with no shared bug or behavior." }, { "left": "pull_request:6119", "right": "pull_request:7932", "accept": false, - "reason": "Training-time fp16 gradient fix vs LoRA test expectation updates; one is production code, the other is test maintenance." + "reason": "Both mention LoRA tests, but they address different issues: fp16 gradient scaling vs RGB ordering side effects." }, { "left": "pull_request:6262", "right": "pull_request:7410", "accept": false, - "reason": "Pipeline relocation/refactor of Panorama vs FreeInit behavior fix across animation/video pipelines; different changes." + "reason": "Moving panorama out of stable_diffusion is unrelated to fixing FreeInit behavior in animation/PIA pipelines." }, { "left": "pull_request:6263", "right": "pull_request:7410", "accept": false, - "reason": "LDM3D refactor/move vs FreeInit behavior fix; unrelated code paths." + "reason": "LDM3D refactor is a pipeline move; FreeInit fix is a separate runtime behavior change." }, { "left": "pull_request:6426", "right": "pull_request:8370", "accept": false, - "reason": "LoRA deprecation cleanup across many modules vs safe handling of norm_type in model remapping; distinct problems." + "reason": "LoRA deprecation cleanup is unrelated to safely handling norm_type remapping in model loading utilities." }, { "left": "pull_request:6448", "right": "pull_request:8496", "accept": false, - "reason": "LoRA test assertion adjustments vs a warning-log guard in Transformer SD3; different scope and intent." + "reason": "LoRA test assertion updates do not match the Transformer SD3 warning-log guard change." }, { "left": "pull_request:6743", "right": "pull_request:7650", "accept": false, - "reason": "Different training arguments in different examples: hflip flag correction vs SDXL instruct-pix2pix is_cosxl_edit support." + "reason": "These are different training arguments in different scripts: hflip flag correction vs SDXL ip2p cosxl-edit support." }, { "left": "pull_request:6787", "right": "pull_request:8937", "accept": false, - "reason": "Marigold depth input normalization fix vs AuraFlow long-prompt handling; unrelated pipelines." + "reason": "Marigold depth input scaling fix is unrelated to AuraFlow long-prompt handling." }, { "left": "pull_request:6961", "right": "pull_request:7941", "accept": false, - "reason": "Single-file loader fast tests vs marking slow tests; pure test maintenance, not the same fix." + "reason": "Single-file loader fast-test additions are not the same as marking a separate single-file test class as slow." }, { "left": "pull_request:7295", "right": "pull_request:8452", "accept": false, - "reason": "Both are single-file loader related, but one is doc/copy cleanup for UNet loader and the other is a strictness fix for single-file loading errors; different concrete issues." + "reason": "UNet single-file loader docs cleanup is unrelated to the single-file loading strictness fix." }, { "left": "pull_request:8267", "right": "pull_request:8719", "accept": false, - "reason": "Single-file resume_download deprecation handling vs VQModelOutput deprecation/doc update; unrelated loader/model changes." + "reason": "Both involve single-file loading, but one changes resume_download handling and the other updates VQModelOutput deprecation wording." }, { "left": "pull_request:8591", "right": "pull_request:8761", "accept": false, - "reason": "SD3 density/weighting utility refactor vs Hunyuan embedding refactor; different models and functionality." + "reason": "SD3 training weighting refactor is unrelated to Hunyuan embedding refactoring." }, { "left": "pull_request:8652", "right": "pull_request:8973", "accept": false, - "reason": "Lumina feature addition vs a test-size reduction for Lumina; one adds the pipeline, the other only shrinks test configs." + "reason": "Lumina-T2X feature addition and the Lumina test-size reduction are not the same underlying change." }, { - "left": "pull_request:6752", - "right": "pull_request:7442", + "left": "pull_request:10132", + "right": "pull_request:8660", + "accept": false, + "reason": "Workflow tokenizers pinning is unrelated to SD3 ControlNet input validation." + }, + { + "left": "pull_request:11044", + "right": "pull_request:11767", "accept": false, - "reason": "PromptDiffusion pipeline addition vs Differential Diffusion community feature; both are research-project additions but they implement different methods and pipelines." + "reason": "LoRA no-op logging change and group-offloading unique-memory-address fix address completely different code paths." } ] }, @@ -151285,19 +151364,18 @@ "component_index": 1, "component_count": 266, "review_unit_index": 150, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10132", + "pull_request:10838", "pull_request:10970", - "pull_request:11044", + "pull_request:11261", "pull_request:1171", - "pull_request:11767", "pull_request:11975", "pull_request:1206", "pull_request:1229", @@ -151328,6 +151406,7 @@ "pull_request:4819", "pull_request:5024", "pull_request:51", + "pull_request:5111", "pull_request:5396", "pull_request:5456", "pull_request:5837", @@ -151338,13 +151417,10 @@ "pull_request:711", "pull_request:8422", "pull_request:8425", - "pull_request:8660", "pull_request:886", "pull_request:934" ], "soft_pairs": [ - "pull_request:10132|pull_request:8660", - "pull_request:11044|pull_request:11767", "pull_request:1171|pull_request:1206", "pull_request:1229|pull_request:2590", "pull_request:1229|pull_request:3275", @@ -151370,204 +151446,206 @@ "pull_request:2334|pull_request:2385", "pull_request:10970|pull_request:11975", "pull_request:12451|pull_request:13352", - "pull_request:340|pull_request:711" + "pull_request:340|pull_request:711", + "pull_request:5024|pull_request:5111", + "pull_request:10838|pull_request:11261" ], - "prepared_review_unit_hash": "b01c6f623814d0a14d067d500f26f46b44fb28962a47476dec6b31eecd49a1fc", + "prepared_review_unit_hash": "1486f83954114840f14958292a59eb9a2ee167314bd8572f55cc2cf20fbe1c40", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 96222, - "estimated_input_tokens": 24056, - "estimated_eval_tokens": 48368 + "serialized_chars": 95845, + "estimated_input_tokens": 23962, + "estimated_eval_tokens": 48180 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a grab bag of unrelated PRs across docs, CI, schedulers, pipelines, and tests; I don\u2019t see a single duplicate issue/change cluster here.", + "summary": "This cluster is mostly a bag of unrelated PRs across docs, CI, tests, schedulers, and new model features. I don\u2019t see a single underlying bug/change tying them together, so none of the soft pairs should be merged as duplicates.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a good canonical representative because the artifacts do not describe the same underlying bug or change.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is globally best-suited as a canonical duplicate target; the collection is too heterogeneous and lacks one shared concrete change.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:10132", - "right": "pull_request:8660", - "accept": false, - "reason": "Workflow pin/update for tokenizers in CI vs a ControlNet-SD3 input validation change; unrelated code paths." - }, - { - "left": "pull_request:11044", - "right": "pull_request:11767", - "accept": false, - "reason": "LoRA no-op warning/message changes vs group-offloading memory-address logic; different subsystems and bugs." - }, { "left": "pull_request:1171", "right": "pull_request:1206", "accept": false, - "reason": "Typo fix in docs/examples vs LDM checkpoint layer-name conversion logic; not the same change." + "reason": "Docs typo fix vs LDM layer-name fix; different files and changes." }, { "left": "pull_request:1229", "right": "pull_request:2590", "accept": false, - "reason": "Dreambooth README installation text vs CI deps-table check; documentation vs workflow enforcement." + "reason": "Dreambooth README wording vs CI deps-table check; unrelated." }, { "left": "pull_request:1229", "right": "pull_request:3275", "accept": false, - "reason": "Dreambooth docs edit vs AttnBlock upsample_size API change; no shared underlying bug." + "reason": "README/install docs vs UNet upsample_size API change; not the same change." }, { "left": "pull_request:12578", "right": "pull_request:13120", "accept": false, - "reason": "Early error handling in modular offload code vs tests for load-id tagging; one is implementation, the other test coverage." + "reason": "CPU offload error handling vs model-loading tests; different code paths." }, { "left": "pull_request:12739", "right": "pull_request:12921", "accept": false, - "reason": "Dummy object/type-check support for ZImageTransformer2DModel vs new GLM-Image model support; same area only broadly, not same fix." + "reason": "Dummy export for ZImage vs full GLM-Image model support; same area but different feature work." }, { "left": "pull_request:1554", "right": "pull_request:453", "accept": false, - "reason": "Docs correction in UNet text vs installation typo in docs; different documentation issues." + "reason": "Docs correction in pipeline docs vs installation typo fix; unrelated." }, { "left": "pull_request:1945", "right": "pull_request:3275", "accept": false, - "reason": "Tests for 2D UNet blocks vs adding upsample_size parameters to attn blocks; related subsystem but not the same concrete change." + "reason": "Block tests vs adding upsample_size to blocks; related subsystem but not the same underlying change." }, { "left": "pull_request:2327", "right": "pull_request:2590", "accept": false, - "reason": "UNet length validation vs CI dependency-table check; unrelated." + "reason": "UNet input validation vs CI workflow update; unrelated." }, { "left": "pull_request:3201", "right": "pull_request:3711", "accept": false, - "reason": "Community pipeline revert vs documentation workflow revert; different files and problems." + "reason": "Community pipeline revert vs doc-build workflow revert; no shared bug." }, { "left": "pull_request:3245", "right": "pull_request:3825", "accept": false, - "reason": "Chinese docs translation update vs a single English doc typo fix; unrelated docs changes." + "reason": "Chinese docs translation update vs a single English docs typo fix; unrelated." }, { "left": "pull_request:3265", "right": "pull_request:3711", "accept": false, - "reason": "lpw_stable_diffusion revert vs doc-build workflow revert; not the same code path." + "reason": "lpw community pipeline revert vs doc-build workflow revert; unrelated." }, { "left": "pull_request:3398", "right": "pull_request:3825", "accept": false, - "reason": "Two separate documentation typo fixes in different pages; not duplicates." + "reason": "Stable diffusion docs typo vs control_brightness docs update; different doc edits." }, { "left": "pull_request:3614", "right": "pull_request:5396", "accept": false, - "reason": "Test cleanup in stable diffusion vs timezone-aware stale-bot script fix; unrelated." + "reason": "Test cleanup in stable diffusion vs stale-bot timezone fix; unrelated." }, { "left": "pull_request:3711", "right": "pull_request:5858", "accept": false, - "reason": "Doc-build workflow revert vs tutorial text update; unrelated." + "reason": "Workflow revert vs training-doc timestamp dtype fix; unrelated." }, { "left": "pull_request:4187", "right": "pull_request:5024", "accept": false, - "reason": "DPMSolverMultistepScheduler timestep handling vs DPMSolverSDE step-index initialization; scheduler-related but distinct concrete problems." + "reason": "Different scheduler bugs in different scheduler classes; not mergeable as one PR." }, { "left": "pull_request:4187", "right": "pull_request:5456", "accept": false, - "reason": "Scheduler timestep logic vs RealFill feature addition; no overlap in underlying change." + "reason": "DPMSolverMultistep scheduler change vs RealFill feature addition; unrelated." }, { "left": "pull_request:453", "right": "pull_request:51", "accept": false, - "reason": "Docs installation typo vs setup.py dependency spec fix; different artifacts and scopes." + "reason": "Install docs typo vs setup.py dependency pin fix; similar theme but different artifacts and fixes." }, { "left": "pull_request:5837", "right": "pull_request:6915", "accept": false, - "reason": "Textual inversion multi-token handling vs attention-processor signature warnings; unrelated code paths." + "reason": "Textual inversion multi-token support vs attention processor signature handling; unrelated." }, { "left": "pull_request:6217", "right": "pull_request:8425", "accept": false, - "reason": "Torch-compile fix for SVD VAE vs workflow name tweak; completely unrelated." + "reason": "Torch-compile SVD VAE fix vs workflow rename; unrelated." }, { "left": "pull_request:6972", "right": "pull_request:8422", "accept": false, - "reason": "IPAdapter unload test assertion vs subfolder-loading test suite; different testing goals and bugs." + "reason": "IPAdapter unload test expectation vs subfolder-loading test suite; different test goals." }, { "left": "pull_request:886", "right": "pull_request:934", "accept": false, - "reason": "Both touch VAE tests, but they address different failures (generator seeding vs MPS-specific expected outputs), so not the same bug." + "reason": "VAE test seeding fix vs broader MPS test adjustments; same family of tests but not the same bug/change." }, { "left": "pull_request:4098", "right": "pull_request:4819", "accept": false, - "reason": "SD-XL inpainting feature addition vs masked-latent input handling fix across inpainting pipelines; related area but not the same concrete change." + "reason": "SDXL inpainting feature add vs masked-latent input handling across inpainting pipelines; overlapping area but distinct changes." }, { "left": "pull_request:2334", "right": "pull_request:2385", "accept": false, - "reason": "New Pix2Pix Zero pipeline vs adding a utility helper inside that pipeline; incremental follow-up, not the same underlying bug/change." + "reason": "Pix2Pix Zero pipeline introduction vs a utility helper added later in the same file; not the same underlying change." }, { "left": "pull_request:10970", "right": "pull_request:11975", "accept": false, - "reason": "Workflow action swap for style bot vs a style-only code fix in an example script; unrelated." + "reason": "Workflow reuse change vs a style fix in a training script; unrelated." }, { "left": "pull_request:12451", "right": "pull_request:13352", "accept": false, - "reason": "Doc-builder Dockerfile dependencies vs Claude review workflow checkout; separate CI/container maintenance." + "reason": "Doc-builder Dockerfile packages vs Claude review checkout step; unrelated CI infra changes." }, { "left": "pull_request:340", "right": "pull_request:711", "accept": false, - "reason": "Img2Img scheduler support vs an experimental scheduler design refactor; not the same fix or code path." + "reason": "Img2Img scheduler support fix vs a broad scheduler design proposal; not the same code-path bug." + }, + { + "left": "pull_request:5024", + "right": "pull_request:5111", + "accept": false, + "reason": "DPM-Solver-SDE step index/noise sampler vs KDPM2 Karras-sigmas support; different scheduler fixes." + }, + { + "left": "pull_request:10838", + "right": "pull_request:11261", + "accept": false, + "reason": "FlashVideo text-to-video pipeline vs VidTok autoencoder support; both video-related but unrelated changes." } ] }, @@ -151583,7 +151661,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 151, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -151594,12 +151672,10 @@ "pull_request:10326", "pull_request:10354", "pull_request:10810", - "pull_request:10838", "pull_request:10941", "pull_request:11231", "pull_request:11243", "pull_request:11254", - "pull_request:11261", "pull_request:11532", "pull_request:11814", "pull_request:11892", @@ -151620,9 +151696,12 @@ "pull_request:2365", "pull_request:2489", "pull_request:2706", + "pull_request:2959", "pull_request:3001", "pull_request:3351", "pull_request:3397", + "pull_request:3471", + "pull_request:3561", "pull_request:3766", "pull_request:3968", "pull_request:3989", @@ -151630,9 +151709,8 @@ "pull_request:4040", "pull_request:4115", "pull_request:4259", - "pull_request:5024", + "pull_request:4412", "pull_request:504", - "pull_request:5111", "pull_request:7328", "pull_request:8797", "pull_request:8798", @@ -151640,8 +151718,6 @@ "pull_request:9676" ], "soft_pairs": [ - "pull_request:5024|pull_request:5111", - "pull_request:10838|pull_request:11261", "pull_request:12117|pull_request:12301", "pull_request:12496|pull_request:12611", "pull_request:11243|pull_request:12151", @@ -151664,186 +151740,188 @@ "pull_request:2365|pull_request:3397", "pull_request:2365|pull_request:4040", "pull_request:2489|pull_request:504", - "pull_request:2706|pull_request:3766" + "pull_request:2706|pull_request:3766", + "pull_request:2959|pull_request:4412", + "pull_request:3471|pull_request:3561" ], - "prepared_review_unit_hash": "172e9a0b7fd782de8cb5ebf5d00dfb425b60f85b291e0dce24bf3c6ff0ae7316", + "prepared_review_unit_hash": "2f2b90b10a089b94eb43e4611b29867b246bd9d2a439891156bcf57e114614d5", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 100873, - "estimated_input_tokens": 25219, - "estimated_eval_tokens": 50694 + "serialized_chars": 98224, + "estimated_input_tokens": 24556, + "estimated_eval_tokens": 49368 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated CI, docs, model, loader, scheduler, and pipeline changes. None of the soft edges looks like the same underlying bug/change, so I would not merge any of these pairs as duplicates.", - "confidence": 0.95, + "summary": "This cluster is a broad grab-bag of unrelated PRs across CI, docs, LoRA conversion, Qwen/Flux/Wan pipelines, Flax deprecation, and assorted test fixes. The soft edges look like topic overlap only, not the same underlying bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR cleanly represents the cluster; the items span many unrelated subsystems and change types, so there is no coherent canonical PR to anchor duplicate triage.", + "canonical_pr_reason": "No canonical PR: the items are heterogeneous and cover distinct subsystems and fixes, so any single representative would be arbitrary.", "best_issue_reason": null, - "best_pr_reason": "No PR is a good global representative because the cluster is not centered on one bug or feature; the largest PRs are broad but still unrelated to the other candidates.", + "best_pr_reason": "No single PR is a strong global duplicate representative here; the cluster is too heterogeneous for a meaningful best-PR choice.", "soft_edge_verdicts": [ - { - "left": "pull_request:5024", - "right": "pull_request:5111", - "accept": false, - "reason": "Both touch Karras-related schedulers, but one fixes step_index/noise_sampler state while the other adds use_karras_sigmas to different schedulers; not the same concrete change." - }, - { - "left": "pull_request:10838", - "right": "pull_request:11261", - "accept": false, - "reason": "FlashVideo pipeline work and VidTok autoencoder work are different features in different code paths; they are not duplicates." - }, { "left": "pull_request:12117", "right": "pull_request:12301", "accept": false, - "reason": "Both are Qwen-Image additions, but one adds img2img/inpaint while the other adds ControlNet-inpaint; separate pipeline variants and not one mergeable fix." + "reason": "Both are Qwen-Image feature additions, but one adds img2img/inpaint pipelines while the other adds ControlNet inpainting; different changes." }, { "left": "pull_request:12496", "right": "pull_request:12611", "accept": false, - "reason": "One is a specific Wan i2v shape bug fix, the other is a large modular Wan feature PR; they are not the same underlying issue." + "reason": "Wan i2v shape-mismatch fix vs. broad Wan modular refactor; unrelated concrete problems." }, { "left": "pull_request:11243", "right": "pull_request:12151", "accept": false, - "reason": "Musubi Wan LoRA support and Flax deprecation are unrelated changes with no shared bug or code path." + "reason": "LoRA support for Musubi Wan vs. Flax deprecation; no shared bug/change." }, { "left": "pull_request:12530", "right": "pull_request:12618", "accept": false, - "reason": "A flaky test xfail and a CI uv-install typo are unrelated; one is test suppression, the other is workflow syntax." + "reason": "Both touch CI, but one xfails a flaky test and the other fixes a uv install typo; not the same issue." }, { "left": "pull_request:12530", "right": "pull_request:13296", "accept": false, - "reason": "The xfail for a stable-cascade test and the release-test transformer update address different problems in different workflows/tests." + "reason": "CI-related only; xfail for a test vs. release-test dependency update are different fixes." }, { "left": "pull_request:2489", "right": "pull_request:402", "accept": false, - "reason": "CoreML docs addition and intro-section docs cleanup are different documentation changes with no shared underlying bug." + "reason": "Docs coreml page addition vs. intro-section docs restructure; unrelated." }, { "left": "pull_request:3001", "right": "pull_request:3351", "accept": false, - "reason": "Both concern Karras sigmas, but they modify different schedulers and implement separate support paths; not one duplicate fix." + "reason": "Both involve Karras-related scheduler work, but they change different schedulers and different APIs." }, { "left": "pull_request:3989", "right": "pull_request:4259", "accept": false, - "reason": "Stable Diffusion XL test/offload fixes and a from_single_file circular import fix are unrelated code paths." + "reason": "Slow-test fixes in SDXL vs. circular-import fix in loaders; unrelated." }, { "left": "pull_request:7328", "right": "pull_request:8981", "accept": false, - "reason": "Adding torch to a PyPI publish step is unrelated to LoraLoaderMixin docs/import wiring." + "reason": "PyPI workflow dependency tweak vs. LoRA loader init/docs changes; no shared code-path bug." }, { "left": "pull_request:3968", "right": "pull_request:4115", "accept": false, - "reason": "Both affect SDXL, but one is a broad improvement PR and the other fixes discrete denoising timestep handling; not the same bug." + "reason": "Both touch SDXL, but one is a workflow/docs/model cleanup and the other fixes denoising_start/end semantics; not the same change." }, { "left": "pull_request:10354", "right": "pull_request:8797", "accept": false, - "reason": "Flux IP-Adapter tester behavior and sharded-checkpoint test determinism are unrelated." + "reason": "Flux tester mixin fix vs. sharded checkpoint test fix; different test issues." }, { "left": "pull_request:10941", "right": "pull_request:11892", "accept": false, - "reason": "Flux pipeline copy fixes and removal of k-diffusion dependency are different changes with no common code defect." + "reason": "Flux copy fixes vs. removing k-diffusion dependency; unrelated." }, { "left": "pull_request:10326", "right": "pull_request:8798", "accept": false, - "reason": "A workflow quoting fix and a subfolder checkpoint-loading fix are unrelated." + "reason": "MPS workflow package quoting fix vs. checkpoint subfolder loading fix; unrelated." }, { "left": "pull_request:10810", "right": "pull_request:11814", "accept": false, - "reason": "Flux LoRA conversion support and gradient-checkpointing kwarg plumbing are separate fixes, not the same bug." + "reason": "Flux LoRA conversion support vs. gradient-checkpointing kwargs passthrough; different code paths." }, { "left": "pull_request:11231", "right": "pull_request:11532", "accept": false, - "reason": "HiDream image pipeline/model support and HiDream LoRA conversion support are distinct features." + "reason": "HiDream model/pipeline addition vs. HiDream LoRA conversion support; same ecosystem but different features." }, { "left": "pull_request:11254", "right": "pull_request:9676", "accept": false, - "reason": "ConsisID import fixes and an EntryNotFoundError import fix are unrelated dependency/import changes." + "reason": "ConsisID import/dependency fix vs. EntryNotFoundError import fix; unrelated dependency cleanups." }, { "left": "pull_request:12099", "right": "pull_request:13081", "accept": false, - "reason": "Qwen hidden-state device placement and contiguity fixes target different tensor issues in different code locations." + "reason": "Qwen device placement typo vs. Qwen contiguous-tensor fix; separate bugs." }, { "left": "pull_request:12747", "right": "pull_request:13211", "accept": false, - "reason": "Group-offloading pinning is a feature addition, while the later PR is a bug fix for conditional module prefetching; related area but not the same change." + "reason": "Both concern group offloading, but one adds pinning features and the other patches a fallback bug; not one mergeable change." }, { "left": "pull_request:1494", "right": "pull_request:152", "accept": false, - "reason": "Safetensors save support and local-folder training support are unrelated proposals." + "reason": "Safetensors saving support vs. local image-folder training example; unrelated." }, { "left": "pull_request:2365", "right": "pull_request:3397", "accept": false, - "reason": "A scheduler type note and a LoRA docs note are not the same issue or code path." + "reason": "Scheduler type-note cleanup vs. LoRA docs note; no shared bug/change." }, { "left": "pull_request:2365", "right": "pull_request:4040", "accept": false, - "reason": "A scheduler typing note and keeping _use_default_values as a list are unrelated internal changes." + "reason": "Scheduler typing note vs. config _use_default_values list fix; unrelated." }, { "left": "pull_request:2489", "right": "pull_request:504", "accept": false, - "reason": "CoreML docs addition and relative-link doc fixes are separate documentation edits." + "reason": "CoreML docs addition vs. docs link-path fix; unrelated." }, { "left": "pull_request:2706", "right": "pull_request:3766", "accept": false, - "reason": "EMA save/restore in a training example and a Kandinsky conversion-script update are unrelated changes." + "reason": "Dreambooth EMA save/restore fix vs. Kandinsky conversion script update; different code paths." + }, + { + "left": "pull_request:2959", + "right": "pull_request:4412", + "accept": false, + "reason": "UNet3D norm-group test adjustment vs. float16 inference test fix; different test failures." + }, + { + "left": "pull_request:3471", + "right": "pull_request:3561", + "accept": false, + "reason": "Dreambooth docs note vs. Dreambooth/IF stage-II upscaling implementation; not the same underlying change." } ] }, @@ -151859,7 +151937,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 152, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -151877,24 +151955,24 @@ "pull_request:10358", "pull_request:11588", "pull_request:11728", + "pull_request:11975", "pull_request:12363", + "pull_request:12397", + "pull_request:12450", "pull_request:12611", "pull_request:12642", "pull_request:12919", "pull_request:12982", + "pull_request:12998", "pull_request:13017", "pull_request:13169", "pull_request:2902", - "pull_request:2959", "pull_request:3189", "pull_request:3313", - "pull_request:3471", - "pull_request:3561", "pull_request:3590", "pull_request:4112", "pull_request:4214", "pull_request:4262", - "pull_request:4412", "pull_request:4572", "pull_request:5186", "pull_request:5384", @@ -151916,8 +151994,6 @@ "pull_request:9593" ], "soft_pairs": [ - "pull_request:2959|pull_request:4412", - "pull_request:3471|pull_request:3561", "pull_request:3590|pull_request:5384", "pull_request:4112|pull_request:4262", "pull_request:4214|pull_request:5792", @@ -151939,180 +152015,182 @@ "pull_request:1030|pull_request:3189", "pull_request:10358|pull_request:9593", "pull_request:11728|pull_request:12363", - "pull_request:11728|pull_request:12642" + "pull_request:11728|pull_request:12642", + "pull_request:11975|pull_request:12998", + "pull_request:12397|pull_request:12450" ], - "prepared_review_unit_hash": "792a66421b0a6f992c6c30ac898db62f031ad94be0b1b33b0a8f6bfa95941a0a", + "prepared_review_unit_hash": "78266a825859431563a0846a8da7fe39e8408cf94ffd0db3aa8db2da8c1363fd", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 95973, - "estimated_input_tokens": 23994, - "estimated_eval_tokens": 48244 + "serialized_chars": 99026, + "estimated_input_tokens": 24757, + "estimated_eval_tokens": 49770 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a false positive: the PRs span unrelated docs, tests, CI, pipelines, quantization, and attention changes. The soft edges are driven by superficial keyword overlap, not the same underlying bug or change.", - "confidence": 0.98, + "summary": "This cluster is not a true duplicate set; it mixes many unrelated PRs across pipelines, docs, CI, LoRA, quantization, and NPU work. The soft-similarity links are mostly superficial (same broad area or same file) rather than the same concrete bug/change.", + "confidence": 0.76, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR really fits because the cluster is heterogeneous; if one representative is needed, PR 12611 is the broadest and most substantive merged change in the set.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 12611 is the most substantial standalone PR here, with the widest scope and the most core code touched, so it is the best representative artifact despite the cluster not being a true duplicate group.", "soft_edge_verdicts": [ - { - "left": "pull_request:2959", - "right": "pull_request:4412", - "accept": false, - "reason": "Both are small test fixes, but they target different models and failures (UNet3D norm groups vs generic float16 inference)." - }, - { - "left": "pull_request:3471", - "right": "pull_request:3561", - "accept": false, - "reason": "Both mention DreamBooth, but one is a README note and the other is a training-flow code change for IF stage II upscaling." - }, { "left": "pull_request:3590", "right": "pull_request:5384", "accept": false, - "reason": "Completely different changes: community canvas docs/code versus hypernetwork loading support." + "reason": "Unrelated changes: Stable Diffusion Mixture Canvas/community example work vs hypernetwork loader support." }, { "left": "pull_request:4112", "right": "pull_request:4262", "accept": false, - "reason": "Both touch DreamBooth/LoRA, but they fix different things: rank wiring vs unrelated SDXL training corrections." + "reason": "Both are DreamBooth-related, but they fix different things: rank plumbing vs SDXL DreamBooth batch/prior-preservation fixes." }, { "left": "pull_request:4214", "right": "pull_request:5792", "accept": false, - "reason": "Unrelated fixes; one addresses bf16 custom diffusion behavior, the other just updates import paths across examples." + "reason": "Custom Diffusion bf16 error handling and import-path cleanup are unrelated changes." }, { "left": "pull_request:4572", "right": "pull_request:5420", "accept": false, - "reason": "Example variable rename fix vs tutorial doc correction; no shared bug or code path." + "reason": "One fixes LoRA loading argument names; the other is a tutorial/doc snippet correction for create_repo return value." }, { "left": "pull_request:5186", "right": "pull_request:5552", "accept": false, - "reason": "Both mention FreeU, but one is a proof-of-concept feature and the other fixes the disable method; not the same change." + "reason": "Related FreeU area, but not the same underlying fix: a FreeU PoC vs correcting disable_freeu behavior." }, { "left": "pull_request:6759", "right": "pull_request:8660", "accept": false, - "reason": "Different pipeline bugs in different areas: InstantID image embedding repetition vs SD3 ControlNet input validation." + "reason": "InstantID image embedding repetition and SD3 ControlNet max_sequence_length validation are different code paths and bugs." }, { "left": "pull_request:7044", "right": "pull_request:9065", "accept": false, - "reason": "Unet refactor restoration is unrelated to adding sentencepiece as an optional dependency." + "reason": "UNet refactor restoration and sentencepiece soft-dependency support are unrelated." }, { "left": "pull_request:7679", "right": "pull_request:8285", "accept": false, - "reason": "Both are CI/build maintenance, but one changes how peft is installed and the other just adds wget to the Docker image." + "reason": "CI peft-install tweak and installing wget in the doc-builder Dockerfile are unrelated maintenance changes." }, { "left": "pull_request:9060", "right": "pull_request:9426", "accept": false, - "reason": "A debug-print removal in Flux code and a workflow artifact-uploader bump are unrelated." + "reason": "Removing a debug print in transformer_flux is unrelated to updating artifact uploader versions in CI." }, { "left": "pull_request:2902", "right": "pull_request:3313", "accept": false, - "reason": "Both are attention-related pipeline edits, but they fix different concrete issues and would not merge into one PR." + "reason": "Img2Img cross_attention_kwargs bugfix and broad torch.compile break fixes are different concrete problems." }, { "left": "pull_request:12919", "right": "pull_request:13017", "accept": false, - "reason": "Same NPU backend area, but different bugs: enabling Ulysses attention vs fixing QwenImageEditPlus attention-mask handling." + "reason": "Both touch NPU attention, but one adds a backend op for Ulysses while the other fixes QwenImageEditPlus-specific NPU masking/shape behavior." }, { "left": "pull_request:6489", "right": "pull_request:7442", "accept": false, - "reason": "StyleAligned pipeline/docs versus Differential Diffusion research project; unrelated features." + "reason": "StyleAligned pipeline and Differential Diffusion are separate community projects with different functionality." }, { "left": "pull_request:12982", "right": "pull_request:13169", "accept": false, - "reason": "Both are Flux2/Klein related, but one adds the Klein pipeline and the other only translates a PEFT LoRA format." + "reason": "Adding Flux2 klein support and converting Flux Klein PEFT LoRA format are related by model name but are distinct features/fixes." }, { "left": "pull_request:11588", "right": "pull_request:12611", "accept": false, - "reason": "Different LoRA features: I2V expansion support versus the broad Wan modular pipeline overhaul." + "reason": "LoRA unload support for Flux Control and the Wan modular pipeline are different features in different subsystems." }, { "left": "pull_request:10069", "right": "pull_request:9400", "accept": false, - "reason": "One fixes a BnB quantizer device-type bug; the other makes tests device-agnostic. Not the same issue." + "reason": "BnB quantizer device-type validation and device-agnostic test updates are not the same bug/change." }, { "left": "pull_request:10114", "right": "pull_request:8467", "accept": false, - "reason": "Pure formatting cleanup versus a semantic SEGA pipeline device fix." + "reason": "Formatting cleanup in Stable Audio is unrelated to the SEGA pipeline execution-device fix." }, { "left": "pull_request:10122", "right": "pull_request:10206", "accept": false, - "reason": "Different Flux/attention and LoRA problems: passing attn_mask is unrelated to unloading Flux Control LoRA weights." + "reason": "Passing attn_mask into SDPA for Flux and unloading LoRA weights for Flux Control are different code-path fixes." }, { "left": "pull_request:10271", "right": "pull_request:10293", "accept": false, - "reason": "Both touch TorchAO tests/versioning, but one is a test refactor and the other enforces a minimum torchao version." + "reason": "TorchAO test refactoring and minimum-version bump are related to the same package but are different changes, not one concrete fix." }, { "left": "pull_request:1030", "right": "pull_request:3189", "accept": false, - "reason": "DreamBooth padding and AudioLDM waveform dtype are unrelated bugs." + "reason": "DreamBooth padding behavior and AudioLDM waveform dtype are unrelated." }, { "left": "pull_request:10358", "right": "pull_request:9593", "accept": false, - "reason": "Community-projects docs entry vs docstring refactoring in an example script; no shared underlying change." + "reason": "Community projects table update and docstring refactoring are unrelated documentation changes." }, { "left": "pull_request:11728", "right": "pull_request:12363", "accept": false, - "reason": "Dockerfile quant-lib installation is unrelated to QwenImageEdit Plus docs/tests." + "reason": "Doc-builder quant-lib installation and QwenImageEdit Plus docs/tests are unrelated." }, { "left": "pull_request:11728", "right": "pull_request:12642", "accept": false, - "reason": "Dockerfile dependency additions and docs TOC updates are unrelated maintenance changes." + "reason": "Doc-builder Dockerfile changes and doc TOC updates do not fix the same issue." + }, + { + "left": "pull_request:11975", + "right": "pull_request:12998", + "accept": false, + "reason": "These are separate style-only cleanups in different training/pipeline files." + }, + { + "left": "pull_request:12397", + "right": "pull_request:12450", + "accept": false, + "reason": "Temporarily disabling main-branch installs in specific CI jobs and broadly revisiting CI installation strategy are related at a high level but not the same concrete change." } ] }, @@ -152128,7 +152206,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 153, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -152136,14 +152214,11 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10387", "pull_request:11534", "pull_request:11797", "pull_request:11950", - "pull_request:11975", - "pull_request:12397", - "pull_request:12450", "pull_request:12591", - "pull_request:12998", "pull_request:1468", "pull_request:1527", "pull_request:1564", @@ -152151,8 +152226,10 @@ "pull_request:2655", "pull_request:3673", "pull_request:3675", + "pull_request:3691", "pull_request:3797", "pull_request:3798", + "pull_request:3806", "pull_request:3850", "pull_request:453", "pull_request:4862", @@ -152182,16 +152259,16 @@ "pull_request:8719", "pull_request:9060", "pull_request:9323", + "pull_request:9489", "pull_request:9696", "pull_request:9918" ], "soft_pairs": [ - "pull_request:11975|pull_request:12998", - "pull_request:12397|pull_request:12450", "pull_request:1468|pull_request:1527", "pull_request:1564|pull_request:453", "pull_request:1937|pull_request:3675", "pull_request:2655|pull_request:3673", + "pull_request:3691|pull_request:3806", "pull_request:3797|pull_request:3798", "pull_request:3850|pull_request:6078", "pull_request:4862|pull_request:6772", @@ -152210,186 +152287,187 @@ "pull_request:8652|pull_request:9918", "pull_request:9060|pull_request:9323", "pull_request:11797|pull_request:11950", - "pull_request:11534|pull_request:12591" + "pull_request:11534|pull_request:12591", + "pull_request:10387|pull_request:9489" ], - "prepared_review_unit_hash": "336e4f643b2921f3c96206aaf8045b41bb6851faa5924a1730bd1998541ea303", + "prepared_review_unit_hash": "4c150f5fd59423a7dbd10d27ffcc58ac06d3d8b9d9515b1a45c5f48c0e85dd53", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 100157, - "estimated_input_tokens": 25040, - "estimated_eval_tokens": 50336 + "serialized_chars": 97892, + "estimated_input_tokens": 24473, + "estimated_eval_tokens": 49202 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated docs, CI, training, quantization, and model changes. The soft links are based on broad topical similarity rather than the same concrete bug or patch, so I reject all of them.", - "confidence": 0.98, + "summary": "These PRs are not true duplicates; they only share broad topical similarity (docs, workflows, training scripts, quantization, pipelines) and do not target the same concrete bug or change. No single canonical PR or issue stands out for the cluster.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items do not form a duplicate set; they span distinct fixes/features across workflows, pipelines, loaders, and quantization.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is a good duplicate representative here because none of the soft pairs are the same underlying change.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:11975", - "right": "pull_request:12998", - "accept": false, - "reason": "Both are style-only changes, but they touch different files and different cleanup issues; not the same patch." - }, - { - "left": "pull_request:12397", - "right": "pull_request:12450", - "accept": false, - "reason": "Both modify CI workflows, but one is a temporary pinning workaround for specific tests while the other is a broad CI installation rework." - }, { "left": "pull_request:1468", "right": "pull_request:1527", "accept": false, - "reason": "Completely different problems: scheduler beta serialization vs stable diffusion inpaint tensor inputs." + "reason": "Unrelated fixes: saving trained betas vs. inpaint tensor input handling." }, { "left": "pull_request:1564", "right": "pull_request:453", "accept": false, - "reason": "Both are documentation fixes, but they address unrelated typos/setup guidance in different docs." + "reason": "Different docs typos/notes in separate example READMEs." }, { "left": "pull_request:1937", "right": "pull_request:3675", "accept": false, - "reason": "Conversion-script EMA handling and a UniDiffuser test fix are unrelated." + "reason": "Conversion EMA extraction logic vs. a UniDiffuser test fix; no shared code path." }, { "left": "pull_request:2655", "right": "pull_request:3673", "accept": false, - "reason": "LoRA save/load behavior and DreamBooth multi-GPU correctness are different code-path fixes." + "reason": "LoRA save/load logic vs. DreamBooth multi-GPU fix; different subsystems and bugs." + }, + { + "left": "pull_request:3691", + "right": "pull_request:3806", + "accept": false, + "reason": "Text-encoder LoRA processor selection vs. checkpoint loading note; unrelated changes." }, { "left": "pull_request:3797", "right": "pull_request:3798", "accept": false, - "reason": "Attn naming correction across core models is unrelated to comment cleanup in training scripts." + "reason": "Different model naming/constructor fixes vs. comment cleanup; not the same bug." }, { "left": "pull_request:3850", "right": "pull_request:6078", "accept": false, - "reason": "AudioDiffusion test setup and Euler sigma handling are unrelated." + "reason": "AudioDiffusion test setup fix vs. Euler sigma scheduler fix; different targets." }, { "left": "pull_request:4862", "right": "pull_request:6772", "accept": false, - "reason": "A MultiControlNetModel bugfix is not the same as the large ControlNet-XS feature implementation." + "reason": "SDXL MultiControlNet height/width bug vs. ControlNet-XS implementation; distinct code paths." }, { "left": "pull_request:4963", "right": "pull_request:5754", "accept": false, - "reason": "Shared CI/training context only; one is an accelerate-main workaround and the other is formatting cleanup." + "reason": "Workflow/installing accelerate main vs. unrelated formatting cleanup in training scripts." }, { "left": "pull_request:5697", "right": "pull_request:7679", "accept": false, - "reason": "PEFT/xformers attention bugfix and a CI package-install change are different issues." + "reason": "PEFT/xformers attention scaling bug vs. workflow package-install change; not mergeable as one fix." }, { "left": "pull_request:6119", "right": "pull_request:8303", "accept": false, - "reason": "LoRA fp16 gradient scaling and a documentation wording change are unrelated." + "reason": "LoRA fp16 gradient handling vs. a documentation wording change." }, { "left": "pull_request:6261", "right": "pull_request:7410", "accept": false, - "reason": "Attend-and-Excite refactor and FreeInit behavior fix do not target the same code path." + "reason": "Attend-and-Excite refactor vs. FreeInit behavior fix; different pipelines and problems." }, { "left": "pull_request:6564", "right": "pull_request:8425", "accept": false, - "reason": "IP Adapter path construction and a workflow name rename are unrelated." + "reason": "IP Adapter path bug fix vs. workflow name rename; unrelated." }, { "left": "pull_request:6886", "right": "pull_request:7662", "accept": false, - "reason": "Removing deprecated torch_dtype args and trimming redundant flax test installs are different maintenance changes." + "reason": "Deprecated `to(torch_dtype=...)` cleanup vs. removing redundant apt install in Flax tests." }, { "left": "pull_request:6886", "right": "pull_request:8267", "accept": false, - "reason": "Both mention deprecations, but one is about to() kwargs and the other is resume_download handling." + "reason": "Deprecated `to(torch_dtype=...)` cleanup vs. `resume_download` deprecation handling." }, { "left": "pull_request:689", "right": "pull_request:816", "accept": false, - "reason": "A slow-test performance tweak and a deepspeed compatibility fix are not the same bug." + "reason": "Attention implementation performance tweak vs. Deepspeed mixed-precision training fix." }, { "left": "pull_request:7116", "right": "pull_request:7662", "accept": false, - "reason": "Both touch workflows, but one switches installers to uv while the other removes a redundant package install." + "reason": "Workflow dependency installation overhaul vs. removing redundant package install in one job." }, { "left": "pull_request:7295", "right": "pull_request:8719", "accept": false, - "reason": "UNet single-file loader cleanup and VQModel deprecation docs are unrelated." + "reason": "Single-file UNet loader docs cleanup vs. VQModelOutput deprecation messaging." }, { "left": "pull_request:7778", "right": "pull_request:9696", "accept": false, - "reason": "SSH debugging workflow and PR-test workflow input changes are different CI features." + "reason": "Debugging SSH workflow vs. PR test workflow input changes; different workflows and purposes." }, { "left": "pull_request:8618", "right": "pull_request:8627", "accept": false, - "reason": "SD3 tiny autoencoder shift_factor and SD3 ControlNet docs/feature scaffolding are not the same change." + "reason": "SD3 tiny autoencoder shift_factor change vs. SD3 ControlNet example/pipeline addition; different features." }, { "left": "pull_request:8652", "right": "pull_request:9918", "accept": false, - "reason": "Lumina-T2X feature integration and a server example are separate additions." + "reason": "Lumina feature addition vs. server example docs/code; only broad release-related similarity." }, { "left": "pull_request:9060", "right": "pull_request:9323", "accept": false, - "reason": "Removing a debug print in Flux and changing single-file defaults are unrelated." + "reason": "Remove debug print in Flux transformer vs. changing single-file default checkpoints; unrelated." }, { "left": "pull_request:11797", "right": "pull_request:11950", "accept": false, - "reason": "Higgs quantization and SVDQuant are different quantization methods with different code paths." + "reason": "Higgs quantization support vs. SVDQuant support; both quantization-related but distinct methods and implementations." }, { "left": "pull_request:11534", "right": "pull_request:12591", "accept": false, - "reason": "Framepack docs/pipeline work and Hunyuan Video attention dispatch are separate Hunyuan Video changes, not one concrete bug fix." + "reason": "Framepack docs/pipeline addition vs. attention dispatch backend support for Hunyuan Video; not the same change." + }, + { + "left": "pull_request:10387", + "right": "pull_request:9489", + "accept": false, + "reason": "Different SD3 controlnet training fixes: custom data loading vs. DDP/find_unused_parameters." } ] }, @@ -152405,7 +152483,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 154, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -152413,9 +152491,9 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10387", "pull_request:10576", "pull_request:11039", + "pull_request:1128", "pull_request:11290", "pull_request:11583", "pull_request:11595", @@ -152434,6 +152512,7 @@ "pull_request:13280", "pull_request:13454", "pull_request:13562", + "pull_request:2040", "pull_request:2795", "pull_request:3852", "pull_request:3912", @@ -152459,11 +152538,9 @@ "pull_request:8290", "pull_request:8684", "pull_request:8685", - "pull_request:8761", - "pull_request:9489" + "pull_request:8761" ], "soft_pairs": [ - "pull_request:10387|pull_request:9489", "pull_request:11614|pull_request:12397", "pull_request:11655|pull_request:12419", "pull_request:11699|pull_request:12331", @@ -152487,193 +152564,197 @@ "pull_request:11039|pull_request:12277", "pull_request:11583|pull_request:11618", "pull_request:6003|pull_request:6484", - "pull_request:12806|pull_request:13280" + "pull_request:12806|pull_request:13280", + "pull_request:1128|pull_request:2040" ], - "prepared_review_unit_hash": "d35c9c8a872535ff148d64197519865e82b845fec59ea3c88ca1003362df20fa", + "prepared_review_unit_hash": "cfb6711ab8555e1103a31881fcce4a1f02e68caa315a4a19500755523b9c8374", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 97633, - "estimated_input_tokens": 24409, - "estimated_eval_tokens": 49074 + "serialized_chars": 98200, + "estimated_input_tokens": 24550, + "estimated_eval_tokens": 49356 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a single duplicate set; it mixes unrelated PRs across docs, CI, loaders, tests, and separate model/pipeline fixes. The soft-similarity edges are mostly superficial topic matches rather than the same concrete change.", - "confidence": 0.11, + "summary": "The cluster is mostly a grab bag of unrelated PRs. I only found one true duplicate-like pair: two CONTRIBUTING.md link fixes for the supported Python versions reference. Everything else is either a different feature, different bug, or just the same broad subsystem without the same concrete change.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR fits this cluster: the items are heterogeneous and do not resolve to one underlying bug or change.", + "canonical_pr_reason": "PR 7638 is the cleanest canonical candidate: it directly fixes the broken supported-Python link in CONTRIBUTING.md with no extra unrelated changes, making it the most precise representative of that duplicate pair.", "best_issue_reason": null, - "best_pr_reason": "No global best PR for dedupe purposes; none of these PRs is a good representative for the whole cluster.", + "best_pr_reason": "PR 7638 is the best global PR pick because it is a focused, self-contained fix for the specific bad-link problem, whereas the nearby PR 6014 also bundles an unrelated Makefile formatting change.", "soft_edge_verdicts": [ - { - "left": "pull_request:10387", - "right": "pull_request:9489", - "accept": false, - "reason": "Both touch SD3 ControlNet training, but one adds multi-process data loading support while the other fixes DDP/unused-parameter setup. Different bugs, different code paths." - }, { "left": "pull_request:11614", "right": "pull_request:12397", "accept": false, - "reason": "Both are CI workflow edits, but one is a build-docker security tweak and the other stops installing transformers/accelerate from main for PR tests. Not the same fix." + "reason": "Both touch CI workflows, but one is about build-docker security and the other disables installing main-branch transformers/accelerate. Different concrete fixes." }, { "left": "pull_request:11655", "right": "pull_request:12419", "accept": false, - "reason": "Both mention Flux/LoRA, but one adds bnb 8-bit LoRA support and the other adjusts modular pipeline metadata for Mellon. Different changes." + "reason": "One adds bnb 8-bit LoRA support and tests; the other is a Flux/Mellon modular-pipeline change. Same broad area, not the same bug or change." }, { "left": "pull_request:11699", "right": "pull_request:12331", "accept": false, - "reason": "One is a torchao docs tip about FP8 hardware, the other is a broad docs update for PyTorch 2.0/parallel inference. Same docs area, not the same issue." + "reason": "Docs-only PRs in different areas: fp8 hardware tips vs parallel inference docs restructuring. No shared concrete issue." }, { "left": "pull_request:12331", "right": "pull_request:13165", "accept": false, - "reason": "Parallel inference docs versus FA3 wrapper implementation. Same general optimization/backend ecosystem, but unrelated concrete changes." + "reason": "Parallel inference docs vs FA3 hub wrapper implementation. Not the same artifact or fix." }, { "left": "pull_request:13164", "right": "pull_request:13454", "accept": false, - "reason": "One removes test deps from CI, the other xfails CLIP-related tests. Both are test infra changes, but not the same bug or patch." + "reason": "Different CI/test changes: dropping test deps vs xfail-ing CLIP-related tests. Not a duplicate." }, { "left": "pull_request:13164", "right": "pull_request:13562", "accept": false, - "reason": "Both alter CI installs, but one removes test dependencies while the other makes tokenizers installation stable. Different objectives and edits." + "reason": "Both are CI workflow edits, but one removes test deps and the other pins tokenizers. Different underlying change." }, { "left": "pull_request:2795", "right": "pull_request:3912", "accept": false, - "reason": "Both involve LoRA text-encoder training, but one is a small script behavior tweak and the other adds the finetune-text-encoder feature. Not duplicate changes." + "reason": "Both relate to LoRA training, but they target different scripts and different stages of the feature. Not the same concrete code-path fix." }, { "left": "pull_request:3928", "right": "pull_request:4248", "accept": false, - "reason": "ControlNet timestep handling versus tracker config cleanup in dreambooth. No shared code path." + "reason": "ControlNet timestep guidance bug fix vs accelerator tracker input cleanup. Unrelated issues." }, { "left": "pull_request:5299", "right": "pull_request:7134", "accept": false, - "reason": "Both are typing/docs cleanups, but they change different model files and different type annotations. Superficially similar only." + "reason": "Both are type-hint/doc cleanup in model code, but they touch different model files and different changes. Too broad to treat as one fix." }, { "left": "pull_request:5771", "right": "pull_request:6961", "accept": false, - "reason": "Loader refactor versus single-file loader tests. Related subsystem, but one is broad cleanup and the other is test coverage." + "reason": "A broad loaders refactor is not the same as single-file loader tests and conversion-script tweaks. Different underlying work." }, { "left": "pull_request:5943", "right": "pull_request:7650", "accept": false, - "reason": "Transformer2D gradient checkpointing versus SDXL InstructPix2Pix constructor arg. Different model features, not the same fix." + "reason": "Gradient-checkpointing support in Transformer2DModel and an SDXL InstructPix2Pix arg addition are unrelated." }, { "left": "pull_request:8290", "right": "pull_request:8761", "accept": false, - "reason": "Both are Hunyuan-related, but one is a large refactor PR and the other changes a specific embedding implementation. Not the same underlying change." + "reason": "Both are Hunyuan/Hunyuan-DIT changes, but one is a large refactor and the other is a specific embedding refactor. Not the same bug or patch." }, { "left": "pull_request:11290", "right": "pull_request:12451", "accept": false, - "reason": "One unpins torch versions in the onnxruntime Dockerfile; the other adds zip/wget to the doc-builder Dockerfile. Different Dockerfiles and problems." + "reason": "Dockerfile changes for different images and different dependencies; no shared concrete fix." }, { "left": "pull_request:3852", "right": "pull_request:4744", "accept": false, - "reason": "A docs contributor note versus an AudioLDM docstring update. No duplicate bug/change." + "reason": "Contributor-note docs vs AudioLDM docstring correction. Different documentation targets and problems." }, { "left": "pull_request:4301", "right": "pull_request:4352", "accept": false, - "reason": "Kandinsky docs update versus Kandinsky combined pipeline code fixes. Same project area, but doc-only and code changes are not the same fix." + "reason": "Kandinsky docs update vs combined-pipeline feature fixes. Same domain, but not the same change." }, { "left": "pull_request:5157", "right": "pull_request:7158", "accept": false, - "reason": "LoRA conversion logic versus benchmark cleanup. Unrelated changes." + "reason": "LoRA conversion bug fix vs benchmark cleanup. Completely different purpose." }, { "left": "pull_request:6014", "right": "pull_request:7638", - "accept": false, - "reason": "Both edit CONTRIBUTING.md links, but 6014 also touches the Makefile and the link targets differ. Not a clear duplicate of the same change." + "accept": true, + "reason": "Both fix the same broken CONTRIBUTING.md link to supported Python versions. 6014 also tweaks Makefile formatting, but the core artifact is the same link correction." }, { "left": "pull_request:8684", "right": "pull_request:8685", "accept": false, - "reason": "One fixes a Pillow import in a pipeline example; the other adds a float8 note to SD3 docs. Different files and purposes." + "reason": "Both are docs-level patches around different topics (Pillow import example vs float8 note). No duplicate bug/change." }, { "left": "pull_request:6241", "right": "pull_request:6787", "accept": false, - "reason": "One reports a runtime dtype mismatch; the other fixes Marigold input normalization range. Different bugs entirely." + "reason": "One addresses a float/half runtime error, the other fixes Marigold input normalization. Different code paths." }, { "left": "pull_request:10576", "right": "pull_request:11595", "accept": false, - "reason": "Both mention LoRA loading, but one tests 8-bit bnb quantized model loading and the other fixes a Flux LoRA loading edge case. Different concrete issues." + "reason": "Both mention LoRA loading with quantization, but one is a test addition for 8-bit bnb and the other is a Flux-specific loader fix. Not the same concrete fix." }, { "left": "pull_request:11039", "right": "pull_request:12277", "accept": false, - "reason": "Torch dtype/quantization loading behavior versus disabling xformers tests for some pipelines. No shared bug." + "reason": "Single-file torch_dtype handling vs disabling xformers tests for some pipelines. Different concerns entirely." }, { "left": "pull_request:11583", "right": "pull_request:11618", "accept": false, - "reason": "Both are docs changes, but one removes the fast diffusion tutorial and the other reorganizes PyTorch 2.0 docs links/content. Not the same page or fix." + "reason": "Both are docs/navigation updates, but one removes a tutorial and the other rewires PyTorch 2.0 doc links. Not a duplicate." }, { "left": "pull_request:6003", "right": "pull_request:6484", "accept": false, - "reason": "Both are Stable Video Diffusion, but one adds SVD inpaint and the other supports GPU-created generators. Distinct features/bugs." + "reason": "Same SVD pipeline file, but one adds SVD inpaint and the other fixes GPU generator handling. Different concrete changes." }, { "left": "pull_request:12806", "right": "pull_request:13280", "accept": false, - "reason": "Both touch attention backend internals, but one handles distributed logging/progress bars and the other adds FlashAttention-4 support. Different code paths and goals." + "reason": "Distributed logging/progress handling vs FlashAttention-4 support. Different subsystems and fixes." + }, + { + "left": "pull_request:1128", + "right": "pull_request:2040", + "accept": false, + "reason": "Inpaint tensor-input support vs adding InstructPix2Pix pipeline. Unrelated features." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The only accepted soft-edge pair, 6014/7638, is plausibly mergeable as a single CONTRIBUTING.md link fix despite 6014 also including a minor Makefile whitespace cleanup. The rest of the pairwise rejections are appropriately cautious and supported by the listed diffs." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -152682,7 +152763,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 155, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -152704,7 +152785,6 @@ "pull_request:10728", "pull_request:10808", "pull_request:11081", - "pull_request:1128", "pull_request:11479", "pull_request:11639", "pull_request:11904", @@ -152715,7 +152795,6 @@ "pull_request:12523", "pull_request:12909", "pull_request:1971", - "pull_request:2040", "pull_request:4062", "pull_request:415", "pull_request:433", @@ -152730,17 +152809,18 @@ "pull_request:7476", "pull_request:8631", "pull_request:8941", + "pull_request:9077", "pull_request:9175", "pull_request:9234", "pull_request:9542", "pull_request:9721", + "pull_request:9739", "pull_request:9840", "pull_request:9843", "pull_request:9896", "pull_request:9934" ], "soft_pairs": [ - "pull_request:1128|pull_request:2040", "pull_request:11904|pull_request:12028", "pull_request:7476|pull_request:9234", "pull_request:11639|pull_request:11913", @@ -152765,186 +152845,200 @@ "pull_request:102|pull_request:1971", "pull_request:102|pull_request:415", "pull_request:102|pull_request:433", - "pull_request:10218|pull_request:10554" + "pull_request:10218|pull_request:10554", + "pull_request:10218|pull_request:9077", + "pull_request:10218|pull_request:9739" ], - "prepared_review_unit_hash": "80f74980609a18f57d0fd6b1af5dde80088d4aeef0783edc753756a8f587b19a", + "prepared_review_unit_hash": "ed7f81e62309902f57d6b74f50524e97d2a35d0bb9db458b8b468c9cb5e746e0", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 26, - "serialized_chars": 104284, - "estimated_input_tokens": 26071, - "estimated_eval_tokens": 52398 + "soft_pair_count": 27, + "serialized_chars": 104203, + "estimated_input_tokens": 26051, + "estimated_eval_tokens": 52358 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a bag of unrelated pull requests with only superficial overlap in subsystem or file paths. None of the soft-edge pairs look like the same underlying bug or change, so I would not merge any of them as duplicates.", - "confidence": 0.95, + "summary": "This cluster is a noisy mix of unrelated PRs that only look similar at a broad subsystem level (model loading, Flux/Wan, ControlNet, docs, tests). None of the soft-edge pairs look like the same underlying change, so I would not merge any of them into a duplicate cluster.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:1128", - "right": "pull_request:2040", - "accept": false, - "reason": "Different features: inpaint tensor-input support vs adding a new InstructPix2Pix pipeline." - }, { "left": "pull_request:11904", "right": "pull_request:12028", "accept": false, - "reason": "Both touch loading internals, but one is a model-loading validation fix and the other is parallel shard loading; not the same change." + "reason": "Both touch loading utilities, but 11904 is a non-blocking single-file load speedup while 12028 is parallel shard loading; different mechanisms and bug targets." }, { "left": "pull_request:7476", "right": "pull_request:9234", "accept": false, - "reason": "Unrelated fixes in ControlNet inpaint vs safetensors compatibility regression." + "reason": "7476 fixes ControlNetUnion callback behavior in SDXL inpaint; 9234 fixes a safetensors compatibility regression. Same area, different problem." }, { "left": "pull_request:11639", "right": "pull_request:11913", "accept": false, - "reason": "Wan video2video sampling behavior vs modular Wan pipeline work; different code paths and goals." + "reason": "11639 changes Wan video-to-video latent sampling; 11913 adds the Wan modular pipeline. Not the same code-path or change." }, { "left": "pull_request:10192", "right": "pull_request:10728", "accept": false, - "reason": "Flux ControlNet/inpaint pipeline addition vs quantized Flux IP-Adapter handling; distinct features." + "reason": "10192 adds Flux Control/Inpaint support; 10728 changes Flux IP-Adapter handling for quantized models. Different features and fixes." }, { "left": "pull_request:5448", "right": "pull_request:6752", "accept": false, - "reason": "LCM pipeline addition and PromptDiffusion are separate pipelines with no shared bug/change." + "reason": "Both add new research pipelines, but they are entirely different models and implementations, not the same underlying change." }, { "left": "pull_request:12249", "right": "pull_request:12416", "accept": false, - "reason": "Different model families and pipelines: Step1X-Edit vs QwenImage Edit Plus modular support." + "reason": "12249 adds Step1X-Edit; 12416 adds QwenImage Edit Plus modular support. Distinct pipelines/models." }, { "left": "pull_request:5887", "right": "pull_request:6868", "accept": false, - "reason": "Related to IP-Adapter, but one adds support to a specific inpaint pipeline and the other introduces a broader input path across many pipelines." + "reason": "5887 adds IP-Adapter to ControlNet inpaint, while 6868 propagates ip_adapter_image_embeds through SD pipelines generally. Related topic, but not the same fix." }, { "left": "pull_request:11081", "right": "pull_request:9840", "accept": false, - "reason": "Group offload record_stream support and bitsandbytes CUDA placement are separate runtime features." + "reason": "11081 implements record_stream for group offloading; 9840 relaxes CUDA placement restrictions for bnb-loaded pipelines. Different behavior and code paths." }, { "left": "pull_request:4062", "right": "pull_request:4408", "accept": false, - "reason": "Shap-E mesh output work is unrelated to SDXL single-file inpainting support." + "reason": "4062 adds Shap-E mesh output; 4408 supports SDXL inpainting from single-file checkpoints. Unrelated changes." }, { "left": "pull_request:12523", "right": "pull_request:12909", "accept": false, - "reason": "AutoencoderKLWan skip-keys fix vs Wan context-parallel attention fix; different bug classes." + "reason": "12523 fixes AutoencoderKLWan skip keys; 12909 fixes Wan 2.1 context parallel attention. Separate bugs in different layers." }, { "left": "pull_request:10808", "right": "pull_request:9542", "accept": false, - "reason": "LoRA config key ambiguity fix vs Flux LoRA text-encoder compatibility; not the same underlying issue." + "reason": "10808 tightens PEFT config key handling; 9542 improves conversion of Kohya Flux LoRAs with text encoder keys. Not the same bug." }, { "left": "pull_request:4408", "right": "pull_request:5653", "accept": false, - "reason": "SDXL single-file inpainting support and peft-related warning handling are unrelated." + "reason": "4408 is SDXL single-file inpainting support; 5653 is a PEFT-related maybe_raise_or_warn enhancement. No shared underlying change." }, { "left": "pull_request:6484", "right": "pull_request:6646", "accept": false, - "reason": "Stable Video Diffusion GPU-generator support vs resolution multiple-of-8 support; distinct pipeline issues." + "reason": "6484 fixes GPU-created generators in SVD; 6646 changes SVD resolution constraints. Same model family, different issues." }, { "left": "pull_request:10000", "right": "pull_request:9934", "accept": false, - "reason": "Mochi video-to-video pipeline addition vs Mochi docs update; docs and code are not the same concrete change." + "reason": "10000 adds Mochi video-to-video support; 9934 updates Mochi docs. One is code, the other documentation." }, { "left": "pull_request:10008", "right": "pull_request:8941", "accept": false, - "reason": "CogVideoX table formatting docs vs Stable Diffusion 3 guidance-scale docstring fix; unrelated." + "reason": "10008 fixes a CogVideoX docs table; 8941 fixes an SD3 guidance_scale docstring mismatch. Both docs-only and unrelated." }, { "left": "pull_request:10069", "right": "pull_request:9721", "accept": false, - "reason": "BnB quantizer device validation fix vs docker torch-version pinning; no shared code path." + "reason": "10069 corrects a BnB device type check; 9721 pins torch versions in Dockerfiles. Different concerns." }, { "left": "pull_request:10101", "right": "pull_request:9843", "accept": false, - "reason": "PipelineTesterMixin sigma coverage vs SD3 LoRA test data fix; both tests, but different behaviors." + "reason": "10101 adds a generic sigmas test; 9843 adjusts an SD3 LoRA test fixture. Testing-related, but not the same concrete issue." }, { "left": "pull_request:10114", "right": "pull_request:8631", "accept": false, - "reason": "Stable Audio formatting cleanup vs single-file T5 loading fix; unrelated." + "reason": "10114 is a formatting-only stable audio cleanup; 8631 fixes from_single_file for T5 checkpoints. Completely different." }, { "left": "pull_request:10122", "right": "pull_request:9896", "accept": false, - "reason": "Flux attention-mask support vs Flux device/typecasting revert; different fixes in the same subsystem." + "reason": "10122 passes attn_mask into Flux attention; 9896 reverts Flux device/typecasting changes. Different code paths and fixes." }, { "left": "pull_request:10132", "right": "pull_request:11479", "accept": false, - "reason": "Tokenizers test-environment update vs Intel Gaudi documentation rename; unrelated." + "reason": "10132 updates tokenizers in CI; 11479 renames Habana Gaudi docs to Intel Gaudi. Unrelated." + }, + { + "left": "pull_request:10192", + "right": "pull_request:9175", + "accept": false, + "reason": "Both are Flux ControlNet-related, but 10192 adds Flux Control/Inpaint docs and pipeline support while 9175 adds Union ControlNet support. Different features, not duplicates." }, { "left": "pull_request:102", "right": "pull_request:1971", "accept": false, - "reason": "Training examples update vs a scheduler docstring edit; not the same change." + "reason": "102 updates training example commands; 1971 only edits a scheduler docstring. Not the same change." }, { "left": "pull_request:102", "right": "pull_request:415", "accept": false, - "reason": "Both are docs-oriented, but one updates examples and the other updates training docs structure/content." + "reason": "102 updates examples/README and training scripts, while 415 restructures training docs. Same broad docs area, but different artifacts and scope." }, { "left": "pull_request:102", "right": "pull_request:433", "accept": false, - "reason": "Training examples changes are unrelated to the black-version dependency bump." + "reason": "102 is training example maintenance; 433 updates black and unrelated code/style files. Not the same underlying bug or change." }, { - "left": "pull_request:10192", - "right": "pull_request:9175", + "left": "pull_request:10218", + "right": "pull_request:10554", + "accept": false, + "reason": "10218 fixes ControlNetUnion callback tensor inputs; 10554 fixes an SD3 LoRA training script key filter. Different pipelines and bugs." + }, + { + "left": "pull_request:10218", + "right": "pull_request:9077", "accept": false, - "reason": "Both involve Flux/ControlNet, but one adds a Flux control inpaint pipeline and the other adds Union ControlNet support; not the same concrete feature." + "reason": "10218 is a functional ControlNetUnion fix; 9077 is a typo-only cleanup. No overlap in the actual change." + }, + { + "left": "pull_request:10218", + "right": "pull_request:9739", + "accept": false, + "reason": "10218 changes ControlNetUnion callback inputs; 9739 adds SD3 single-file docs. Different purpose and files." } ] }, @@ -152960,7 +153054,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 156, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -152968,7 +153062,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10218", "pull_request:10230", "pull_request:10316", "pull_request:10322", @@ -152994,11 +153087,14 @@ "pull_request:10990", "pull_request:11087", "pull_request:11113", + "pull_request:11194", + "pull_request:11212", "pull_request:11238", "pull_request:11451", "pull_request:11556", "pull_request:11812", "pull_request:11813", + "pull_request:11838", "pull_request:11975", "pull_request:11999", "pull_request:2759", @@ -153008,18 +153104,14 @@ "pull_request:8845", "pull_request:8920", "pull_request:9036", - "pull_request:9077", "pull_request:9116", "pull_request:9687", "pull_request:9696", "pull_request:9736", - "pull_request:9739", "pull_request:9918", "pull_request:9934" ], "soft_pairs": [ - "pull_request:10218|pull_request:9077", - "pull_request:10218|pull_request:9739", "pull_request:10230|pull_request:11556", "pull_request:10230|pull_request:8755", "pull_request:10316|pull_request:9934", @@ -153045,204 +153137,206 @@ "pull_request:10922|pull_request:11087", "pull_request:10941|pull_request:11975", "pull_request:10990|pull_request:11812", - "pull_request:11113|pull_request:11975" + "pull_request:11113|pull_request:11975", + "pull_request:11194|pull_request:11838", + "pull_request:11212|pull_request:9934" ], - "prepared_review_unit_hash": "a4993713af04e323fa9e6de66733f943cc48d0e8855288f8666cee44192b5632", + "prepared_review_unit_hash": "79307d1870402690e1183df572775b9daca14c8fa811912d33cf9bb4ae2e6cca", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 99992, - "estimated_input_tokens": 24998, - "estimated_eval_tokens": 50252 + "serialized_chars": 101040, + "estimated_input_tokens": 25260, + "estimated_eval_tokens": 50776 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are a heterogeneous mix of unrelated docs, tests, CI, model additions, and bugfix PRs. The soft edges are mostly title/subsystem similarity rather than the same underlying change, so I would not merge any of them as duplicates.", - "confidence": 0.97, + "summary": "This cluster is a loose similarity group, not a true duplicate set: it mixes unrelated docs updates, training-script fixes, model additions, and pipeline features. None of the soft-edge pairs look like the same underlying bug or change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 10922 is the strongest representative artifact: it is a broad, merged core feature PR with many touched files and clear centrality to the Wan release, unlike the more narrow docs/tests fixes around it.", + "canonical_pr_reason": "No single PR is a good canonical for the cluster; the items span unrelated codepaths and content areas (docs, CI, training scripts, model/pipeline feature work).", "best_issue_reason": null, - "best_pr_reason": "PR 10922 is the best overall PR candidate because it is substantive, widely scoped, and the most representative merged change in this set.", + "best_pr_reason": "No PR stands out as a global representative duplicate target because the cluster is not centered on one concrete fix or feature.", "soft_edge_verdicts": [ - { - "left": "pull_request:10218", - "right": "pull_request:9077", - "accept": false, - "reason": "Unrelated changes: ControlNetUnion callback tensor cleanup vs assorted typo/doc/test fixes." - }, - { - "left": "pull_request:10218", - "right": "pull_request:9739", - "accept": false, - "reason": "Different code paths and purposes: ControlNetUnion callback inputs vs docs wording/checkpoint text." - }, { "left": "pull_request:10230", "right": "pull_request:11556", "accept": false, - "reason": "Both are docs edits, but they address different pages and topics; not the same change." + "reason": "Both are docs-only, but one adds LoRA loader mixins and the other adds regional compilation docs; unrelated content and codepaths." }, { "left": "pull_request:10230", "right": "pull_request:8755", "accept": false, - "reason": "LoRA docs additions are unrelated to SD3 dreambooth training logic." + "reason": "LoRA docs update vs SD3 DreamBooth training logic fix; same broad area only by keyword, not the same change." }, { "left": "pull_request:10316", "right": "pull_request:9934", "accept": false, - "reason": "Both are docs-only, but they cover different models and different documentation fixes." + "reason": "Different documentation fixes for different pipelines; no shared bug or underlying change." }, { "left": "pull_request:10322", "right": "pull_request:10431", "accept": false, - "reason": "Same model family, but one adds 2K conversion support while the other fixes a positional embedding bug; different concrete fixes." + "reason": "Sana converter/model support vs Sana transformer PE fix; both Sana-related but distinct problems and files." }, { "left": "pull_request:10376", "right": "pull_request:10567", "accept": false, - "reason": "HunyuanVideo LoRA loader support vs HunyuanVideo true CFG pipeline work are separate changes." + "reason": "HunyuanVideo LoRA conversion support vs HunyuanVideo true-CFG pipeline feature; unrelated code changes." }, { "left": "pull_request:10378", "right": "pull_request:11238", "accept": false, - "reason": "Different training scripts and bugs: SD3 LoRA distributed-saving changes vs SD3 ControlNet validation memory fix." + "reason": "Both touch training scripts, but one is SD3 LoRA saving and the other is SD3 ControlNet validation on A100; different fixes." }, { "left": "pull_request:1043", "right": "pull_request:2759", "accept": false, - "reason": "Both touch tests, but one adds a helper/print improvement and the other is a large unrelated training/runtime regression test change." + "reason": "The shared tests are superficial; one adds torch_all_close helper, the other is a broad attention/model refactor." }, { "left": "pull_request:1043", "right": "pull_request:774", "accept": false, - "reason": "Test utility/print improvements do not match the callback tolerance and chaining fixes in the older pipeline test PR." + "reason": "Test-helper/print improvements vs unrelated historical pipeline test tweaks; not the same defect." }, { "left": "pull_request:10430", "right": "pull_request:9736", "accept": false, - "reason": "TeaCache infrastructure is unrelated to Allegro model/pipeline introduction." + "reason": "TeaCache core hooks work vs Allegro model/pipeline addition; completely different changes." }, { "left": "pull_request:10434", "right": "pull_request:8920", "accept": false, - "reason": "Both are docs-link related, but one fixes many broken cross-links while the other tweaks PAG guide wording." + "reason": "Broken docs links across many pages vs a small PAG guide wording fix; same docs area only, not same issue." }, { "left": "pull_request:10473", "right": "pull_request:11813", "accept": false, - "reason": "Different docs scopes: advanced training README updates vs Flux Kontext training docs/code." + "reason": "General advanced training README text vs Flux Kontext training code and docs; unrelated feature/documentation work." }, { "left": "pull_request:10491", "right": "pull_request:8845", "accept": false, - "reason": "CI path typo fix in workflows vs PAG test typo fixes are separate maintenance changes." + "reason": "CI filename fix vs PAG test typo fix; both tests/CI, but unrelated concrete changes." }, { "left": "pull_request:10524", "right": "pull_request:9934", "accept": false, - "reason": "HunyuanVideo checkpoint-id docs correction is unrelated to Mochi docs expansion." + "reason": "HunyuanVideo checkpoint-id correction vs Mochi docs overhaul; different pipelines and no common bug." }, { "left": "pull_request:10554", "right": "pull_request:9036", "accept": false, - "reason": "SD3 miniature loader bugfix is unrelated to nightly dependency/install workflow changes." + "reason": "SD3 miniature training loader bugfix vs nightly dependency install fix; unrelated scopes." }, { "left": "pull_request:10554", "right": "pull_request:9116", "accept": false, - "reason": "One is an SD3 training-script bugfix; the other is a dead-link docs fix." + "reason": "SD3 training-script fix vs a dead-link docs fix; no shared codepath." }, { "left": "pull_request:10579", "right": "pull_request:9687", "accept": false, - "reason": "Community example addition vs Xlabs ControlNet support are different features." + "reason": "A new community pipeline README/code addition vs Flux ControlNet support; different features." }, { "left": "pull_request:10810", "right": "pull_request:11999", "accept": false, - "reason": "Flux LoRA conversion logic for supported key patterns is not the same change as adapting PEFT config injection/versioning." + "reason": "Flux LoRA conversion logic vs PEFT version/config-injection adaptation; the latter is dependency compatibility, not the same bug." }, { "left": "pull_request:10834", "right": "pull_request:11451", "accept": false, - "reason": "GGUF arg-check bugfix vs WanTransformer torch.compile test are unrelated." + "reason": "GGUF isinstance bugfix vs Wan transformer compile test; one fixes runtime logic, the other adds test coverage." }, { "left": "pull_request:10850", "right": "pull_request:9687", "accept": false, - "reason": "Hook utility refactor work is unrelated to Flux ControlNet support." + "reason": "Hooks utility refactor vs Xlabs ControlNet support; not the same concrete code-path problem." }, { "left": "pull_request:10850", "right": "pull_request:9918", "accept": false, - "reason": "Hook utilities are not the same as server example/docs work." + "reason": "Hooks utility refactor vs server example/docs addition; unrelated." }, { "left": "pull_request:1087", "right": "pull_request:3108", "accept": false, - "reason": "Early slow-test fixes and an empty/unknown checkpoint-fix PR are not the same underlying issue." + "reason": "Slow-test adjustments vs an empty/ckpt-fix PR with no comparable change; not duplicates." }, { "left": "pull_request:10894", "right": "pull_request:9696", "accept": false, - "reason": "Docs toctree sorting and PR-workflow input renaming are separate maintenance tasks." + "reason": "Docs toctree sorting vs PR workflow test runner changes; unrelated documentation vs CI automation edits." }, { "left": "pull_request:10922", "right": "pull_request:11087", "accept": false, - "reason": "Same Wan file family, but 10922 is the main Wan2.1 feature add while 11087 is a later I2V quality tweak to a specific latent-retrieval path; not the same change." + "reason": "Wan model/pipeline addition vs Wan I2V quality fix; same product line, but different code paths and merge scope." }, { "left": "pull_request:10941", "right": "pull_request:11975", "accept": false, - "reason": "Flux pipeline copy-fix work is unrelated to the Flux advanced training style cleanup." + "reason": "Flux pipeline copy-fixes vs a style fix in a Flux advanced training script; unrelated changes." }, { "left": "pull_request:10990", "right": "pull_request:11812", "accept": false, - "reason": "Flux LoRA quantization support and Flux Kontext pipeline introduction are different concrete features." + "reason": "Flux LoRA quantization support vs Flux Kontext pipeline feature; different underlying functionality." }, { "left": "pull_request:11113", "right": "pull_request:11975", "accept": false, - "reason": "AutoencoderKL training bugfix and Flux training script style fix are unrelated." + "reason": "AutoencoderKL training bugfix vs Flux training script style fix; unrelated models and issues." + }, + { + "left": "pull_request:11194", + "right": "pull_request:11838", + "accept": false, + "reason": "Loading guide docs for dtype maps vs deprecated-pipelines warning docs; both docs-only but unrelated subjects." + }, + { + "left": "pull_request:11212", + "right": "pull_request:9934", + "accept": false, + "reason": "MPS badge/docs update across many pipeline pages vs Mochi docs; same docs theme only, not the same change." } ] }, @@ -153258,7 +153352,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 157, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "cache", @@ -153266,8 +153360,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:11194", - "pull_request:11212", "pull_request:11237", "pull_request:11254", "pull_request:1133", @@ -153290,6 +153382,8 @@ "pull_request:11904", "pull_request:11916", "pull_request:1194", + "pull_request:11975", + "pull_request:120", "pull_request:12115", "pull_request:12231", "pull_request:12277", @@ -153300,7 +153394,9 @@ "pull_request:12519", "pull_request:12644", "pull_request:12747", + "pull_request:12960", "pull_request:1765", + "pull_request:1971", "pull_request:2101", "pull_request:2322", "pull_request:2691", @@ -153311,12 +153407,9 @@ "pull_request:840", "pull_request:9688", "pull_request:9827", - "pull_request:9878", - "pull_request:9934" + "pull_request:9878" ], "soft_pairs": [ - "pull_request:11194|pull_request:11838", - "pull_request:11212|pull_request:9934", "pull_request:11237|pull_request:9688", "pull_request:11254|pull_request:9827", "pull_request:1133|pull_request:2808", @@ -153341,198 +153434,200 @@ "pull_request:11894|pull_request:12519", "pull_request:11916|pull_request:12419", "pull_request:1194|pull_request:2101", - "pull_request:1194|pull_request:2322" + "pull_request:1194|pull_request:2322", + "pull_request:11975|pull_request:12960", + "pull_request:120|pull_request:1971" ], - "prepared_review_unit_hash": "ba04d9c7f4ef2e71032fc89bd59d97785589acae1da65fe5866146ee5aafbff7", + "prepared_review_unit_hash": "4b154bba76fc9fe27acf2a566862a7b7526f56d37fdeff2ef0a2ea6de10f00ed", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 105468, - "estimated_input_tokens": 26367, - "estimated_eval_tokens": 52990 + "serialized_chars": 106819, + "estimated_input_tokens": 26705, + "estimated_eval_tokens": 53666 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very heterogeneous: docs edits, CI/workflow changes, model/pipeline additions, and unrelated loader/refactor fixes. I found no pair that clearly represents the same underlying bug or change, so I reject all soft edges.", - "confidence": 0.95, + "summary": "This is a heterogeneous grab-bag of unrelated PRs across docs, examples, dependency tweaks, pipeline additions, and separate bug fixes. None of the suggested soft edges look like the same underlying change.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 11916 is the most central merged PR in this set: it has the broadest core-code impact, the most inbound references/review activity, and is the least arbitrary representative of the mixed cluster.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 11916 is the strongest representative if one must be chosen, because it is a substantial merged refactor with broad impact and high discussion/review signal, unlike the mostly isolated docs/typo/CI items.", + "best_pr_reason": "PR 11904 is the strongest standalone representative: it is a broad, self-contained code change that improves model loading across multiple loaders and utility paths, unlike the mostly narrow docs/example tweaks elsewhere in the set.", "soft_edge_verdicts": [ - { - "left": "pull_request:11194", - "right": "pull_request:11838", - "accept": false, - "reason": "Both are docs PRs, but they cover different topics and files; no shared underlying bug or change." - }, - { - "left": "pull_request:11212", - "right": "pull_request:9934", - "accept": false, - "reason": "Unrelated docs updates: MPS badges vs Mochi docs." - }, { "left": "pull_request:11237", "right": "pull_request:9688", "accept": false, - "reason": "Different doc fixes in different areas; only superficial docs similarity." + "reason": "One is a tiny docs dtype-map example update; the other is a Flux ControlNet docs change. Different topics and no shared code path." }, { "left": "pull_request:11254", "right": "pull_request:9827", "accept": false, - "reason": "ConsisID import fix vs LoRA device_map behavior are unrelated changes." + "reason": "ConsisID import fixes and LoRA device_map loading fixes are separate dependency/runtime issues in different areas." }, { "left": "pull_request:1133", "right": "pull_request:2808", "accept": false, - "reason": "Windows CI workflow vs test relaxation in model tests; not the same change." + "reason": "A Windows CI workflow and a test relaxation for DiT are unrelated changes." }, { "left": "pull_request:1133", "right": "pull_request:705", "accept": false, - "reason": "Same broad CI/workflow theme, but different workflows and goals." + "reason": "Both are CI-related, but one adds a Windows workflow while the other changes push-test GPU setup; not the same bug or change." }, { "left": "pull_request:1133", "right": "pull_request:824", "accept": false, - "reason": "Different CI/test setup changes; not the same underlying issue." + "reason": "CI workflow setup and a Flax test addition are unrelated." }, { "left": "pull_request:11416", "right": "pull_request:12231", "accept": false, - "reason": "Both are typo/doc cleanup, but they fix different typos in different files." + "reason": "Both are docs typo cleanups, but they touch different files and different text; not one underlying bug." }, { "left": "pull_request:11472", "right": "pull_request:9878", "accept": false, - "reason": "Interpolation-mode docs vs gradient-checkpointing behavior; unrelated." + "reason": "DreamBooth interpolation defaults and gradient-checkpointing behavior in model forwards are unrelated code paths." }, { "left": "pull_request:115", "right": "pull_request:498", "accept": false, - "reason": "Pipeline rename work vs attention slicing bugfix; different code-path problems." + "reason": "Pipeline renames and attention-slicing bug fixes are different changes." }, { "left": "pull_request:11556", "right": "pull_request:12331", "accept": false, - "reason": "Separate docs additions about different optimization topics." + "reason": "One adds regional compilation docs; the other reorganizes parallel inference docs. Same broad docs area, but not the same change." }, { "left": "pull_request:1162", "right": "pull_request:840", "accept": false, - "reason": "Community links cleanup vs adding a dummy one-step pipeline; unrelated." + "reason": "A links/doc update and a new community example are unrelated." }, { "left": "pull_request:11621", "right": "pull_request:12340", "accept": false, - "reason": "Two different pipeline additions, not the same feature change." + "reason": "Two different pipeline additions for different models; not the same feature or code path." }, { "left": "pull_request:11621", "right": "pull_request:12398", "accept": false, - "reason": "PhotoDoodle pipeline vs Mirage pipeline are distinct features." + "reason": "Two separate new pipelines for different architectures/models, not a duplicate change." }, { "left": "pull_request:11635", "right": "pull_request:12115", "accept": false, - "reason": "Flux-Kontext feature/docs vs an unrelated test PR." + "reason": "Flux Kontext pipeline work and an AnimateDiff test are entirely different artifacts." }, { "left": "pull_request:11684", "right": "pull_request:11904", "accept": false, - "reason": "Both touch single-file loading, but one adds GGUF support and the other speeds loading; not the same fix." + "reason": "Both touch single-file loading, but one adds GGUF checkpoint support while the other speeds up loading and sync behavior; different concrete behavior." }, { "left": "pull_request:11767", "right": "pull_request:11814", "accept": false, - "reason": "Group-offloading disk-memory fix vs Flux gradient-checkpointing kwargs forwarding; unrelated." + "reason": "Group-offloading memory-address fixes and Flux gradient-checkpointing kwargs plumbing are unrelated." }, { "left": "pull_request:11767", "right": "pull_request:12747", "accept": false, - "reason": "Same file area, but one fixes disk offloading behavior while the other adds pinning features." + "reason": "Same subsystem, but one fixes disk-offload address collisions and the other adds pin-group feature semantics; not the same underlying bug." }, { "left": "pull_request:118", "right": "pull_request:1765", "accept": false, - "reason": "Old pipeline final-fix rename vs Dreambooth flax fixes; unrelated." + "reason": "A ScoreSdeVe pipeline rename and Flax dreambooth fixes are unrelated." }, { "left": "pull_request:11838", "right": "pull_request:12644", "accept": false, - "reason": "Deprecated-pipelines docs vs AutoModel docs; different content and intent." + "reason": "Deprecated-pipelines docs and AutoModel docs are separate documentation changes." }, { "left": "pull_request:1184", "right": "pull_request:2691", "accept": false, - "reason": "Flash-attention addition vs attention rename refactor; not the same underlying change." + "reason": "Adding flash attention and renaming attention APIs are different code changes, even though both touch attention-related files." }, { "left": "pull_request:11848", "right": "pull_request:12644", "accept": false, - "reason": "LoRA metadata docs vs AutoModel docs; unrelated documentation changes." + "reason": "LoRA metadata docs and AutoModel docs are unrelated docs updates." }, { "left": "pull_request:11892", "right": "pull_request:12277", "accept": false, - "reason": "Removing k-diffusion from core vs disabling xformers tests; different maintenance tasks." + "reason": "Removing k-diffusion from core dependencies and disabling xformers tests are unrelated maintenance changes." }, { "left": "pull_request:11894", "right": "pull_request:12519", "accept": false, - "reason": "k-diffusion pinning for CI vs toctree sorting fix; unrelated CI/docs changes." + "reason": "Pinning k-diffusion for CI and fixing toctree sorting are unrelated." }, { "left": "pull_request:11916", "right": "pull_request:12419", "accept": false, - "reason": "Flux/Chroma attention-dispatch refactor vs Flux modular-pipeline readiness; related subsystem, but distinct changes and code paths." + "reason": "A large Flux/Chroma single-file refactor and a modular-pipeline metadata tweak for Flux are not the same concrete change." }, { "left": "pull_request:1194", "right": "pull_request:2101", "accept": false, - "reason": "Docs workflow updates vs pix2pix example/docs update; not the same change." + "reason": "Documentation workflow updates and a pix2pix example doc/code update are unrelated." }, { "left": "pull_request:1194", "right": "pull_request:2322", "accept": false, - "reason": "Docs workflow updates vs example/inpaint update; unrelated." + "reason": "PR docs action workflow changes and a broad example/inpainting update are unrelated." + }, + { + "left": "pull_request:11975", + "right": "pull_request:12960", + "accept": false, + "reason": "A style-only DreamBooth Flux change and an FSDP detection fix in Flux2 scripts are different issues." + }, + { + "left": "pull_request:120", + "right": "pull_request:1971", + "accept": false, + "reason": "A README expansion and a scheduler docstring cleanup are not the same change." } ] }, @@ -153548,7 +153643,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 158, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -153556,7 +153651,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:11975", "pull_request:120", "pull_request:12099", "pull_request:12109", @@ -153585,18 +153679,19 @@ "pull_request:12782", "pull_request:1283", "pull_request:12914", - "pull_request:12960", "pull_request:12961", "pull_request:13029", "pull_request:13039", "pull_request:13081", + "pull_request:1311", "pull_request:13146", + "pull_request:13164", "pull_request:13294", "pull_request:13321", "pull_request:13338", "pull_request:13472", "pull_request:13562", - "pull_request:1971", + "pull_request:1386", "pull_request:2892", "pull_request:2926", "pull_request:3337", @@ -153606,8 +153701,6 @@ "pull_request:79" ], "soft_pairs": [ - "pull_request:11975|pull_request:12960", - "pull_request:120|pull_request:1971", "pull_request:120|pull_request:433", "pull_request:12099|pull_request:12914", "pull_request:12109|pull_request:12387", @@ -153631,192 +153724,194 @@ "pull_request:1269|pull_request:3337", "pull_request:12782|pull_request:13321", "pull_request:1283|pull_request:483", - "pull_request:13029|pull_request:13472" + "pull_request:13029|pull_request:13472", + "pull_request:1311|pull_request:1386", + "pull_request:13164|pull_request:13294" ], - "prepared_review_unit_hash": "5203548f6b25420aa74f4f03c719db77f783274b5a00281b4c1aaf39f010a80f", + "prepared_review_unit_hash": "6995bec11d9ac229ca4bf5ef087ea5a70c1726adf2967415c988c10284741dc3", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 101665, - "estimated_input_tokens": 25417, - "estimated_eval_tokens": 51090 + "serialized_chars": 102119, + "estimated_input_tokens": 25530, + "estimated_eval_tokens": 51316 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate group: it mixes unrelated docs cleanup, pipeline additions, CI fixes, scheduler tweaks, and bugfixes across different subsystems. No soft pair looks like the same underlying change with enough specificity to merge.", - "confidence": 0.98, + "summary": "This is not a true duplicate cluster: the items span unrelated docs, CI, model, scheduler, and pipeline changes. The soft edges are mostly superficial similarity (same subsystem, shared filenames, or both being typo/docs PRs) rather than the same underlying bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a safe canonical here; the items are heterogeneous and do not share one concrete bug or change.", + "canonical_pr_reason": "No canonical PR: the items are a heterogeneous mix of unrelated docs, CI, model, scheduler, and pipeline changes.", "best_issue_reason": null, - "best_pr_reason": "No global best PR: none of these is representative of the whole set, and the cluster is too mixed to collapse around one artifact.", + "best_pr_reason": "No single PR is a good global representative; none of the items consolidates the cluster into one concrete fix.", "soft_edge_verdicts": [ - { - "left": "pull_request:11975", - "right": "pull_request:12960", - "accept": false, - "reason": "Different fixes in different files/subsystems: a style typo in a Flux example vs an FSDP detection bug in DreamBooth scripts." - }, - { - "left": "pull_request:120", - "right": "pull_request:1971", - "accept": false, - "reason": "Both are README/doc edits, but they address different pages and topics; not the same underlying change." - }, { "left": "pull_request:120", "right": "pull_request:433", "accept": false, - "reason": "One is README content expansion, the other is a broad dependency/code update; no shared bug or concrete change." + "reason": "Both touch README/docs, but one is a README example update and the other is a broad black-version/tooling refresh; not the same change." }, { "left": "pull_request:12099", "right": "pull_request:12914", "accept": false, - "reason": "Both fix typos, but in unrelated pipeline code paths (Qwen image device handling vs Cosmos revision string)." + "reason": "Both are tiny typo/device-revision fixes, but they affect different pipelines and different code paths." }, { "left": "pull_request:12109", "right": "pull_request:12387", "accept": false, - "reason": "Scheduler feature work vs attention-backend support; unrelated code paths and changes." + "reason": "Scheduler feature work vs attention-backend support; same broad area only, not the same bug or fix." }, { "left": "pull_request:1219", "right": "pull_request:2926", "accept": false, - "reason": "Both touch DDIM docs, but on different doc pages with different edits; not one concrete duplicate change." + "reason": "Both concern DDIM docs, but one is pipeline docs and the other is scheduler docs with unrelated wording cleanup." }, { "left": "pull_request:12244", "right": "pull_request:12604", "accept": false, - "reason": "Both are docs cleanups, but they target different documentation areas and unrelated content." + "reason": "Both are doc-only, but one removes Flax from installation/docs while the other cleans up modular docs content." }, { "left": "pull_request:12265", "right": "pull_request:12637", "accept": false, - "reason": "Typos/docs cleanup vs removing a copied-code comment in a transformer; not the same underlying fix." + "reason": "Both are typo-style cleanups, but they touch different files and unrelated content." }, { "left": "pull_request:12265", "right": "pull_request:13146", "accept": false, - "reason": "Different issues entirely: generic typo cleanup vs fixing a tokenizer type/import in PAG HunyuanDiT." + "reason": "A generic typo sweep is not the same as fixing an MT5Tokenizer type annotation in PAG HunYuanDiT." }, { "left": "pull_request:1229", "right": "pull_request:748", "accept": false, - "reason": "DreamBooth README install-note fix is unrelated to push_to_hub argument fixes in training scripts." + "reason": "Dreambooth README install guidance and push_to_hub call fixes are unrelated changes." }, { "left": "pull_request:12340", "right": "pull_request:12961", "accept": false, - "reason": "Both add pipelines, but they introduce different features/models and are not mergeable as one change." + "reason": "Adding LucyEditPipeline and adding PhotoDoodle are different pipeline additions with no shared code-path bug." }, { "left": "pull_request:12398", "right": "pull_request:12961", "accept": false, - "reason": "Different new pipeline families (Mirage vs PhotoDoodle); no shared bug or implementation path." + "reason": "Mirage pipeline implementation and PhotoDoodle community README content are unrelated." }, { "left": "pull_request:1240", "right": "pull_request:79", "accept": false, - "reason": "Both are large feature additions touching init files, but they add entirely different model/pipeline stacks." + "reason": "Both are early top-level model/pipeline additions, but they add different functionality and different code paths." }, { "left": "pull_request:12420", "right": "pull_request:13081", "accept": false, - "reason": "One guards a distributed import, the other makes Qwen tensors contiguous for torchao; unrelated problems." + "reason": "Conditional torch-distributed import and contiguous Qwen hidden states fix separate runtime issues." }, { "left": "pull_request:12451", "right": "pull_request:13338", "accept": false, - "reason": "Doc-builder Dockerfile package additions vs a Claude workflow permission change; not the same fix." + "reason": "Doc-builder Dockerfile packages vs GitHub workflow permissions are unrelated CI maintenance changes." }, { "left": "pull_request:12467", "right": "pull_request:12487", "accept": false, - "reason": "QwenImage batch handling bug vs broken-link doc fixes; completely different artifact types and code paths." + "reason": "QwenImage batch handling bug fix vs docs broken-link cleanup; not the same artifact or bug." }, { "left": "pull_request:12467", "right": "pull_request:12552", "accept": false, - "reason": "QwenImage batching issue vs Kandinsky5 doc typo cleanup; unrelated." + "reason": "A QwenImage runtime fix and Kandinsky5 doc typo fixes do not share the same underlying problem." }, { "left": "pull_request:1250", "right": "pull_request:2892", "accept": false, - "reason": "Community example README addition vs a pipeline input-validation bug; no common underlying change." + "reason": "Community pipeline addition and upscaler input-validation fix are different changes." }, { "left": "pull_request:12530", "right": "pull_request:13294", "accept": false, - "reason": "Both are test-stability related, but they target different models/tests and different failure modes." + "reason": "Both address test stability, but they target different tests and different failure modes." }, { "left": "pull_request:12541", "right": "pull_request:13039", "accept": false, - "reason": "Wan LoRA conversion compatibility vs WanVACE CPU upcasting; different code paths and bug classes." + "reason": "Wan LoRA conversion logic and WanVACE CPU upcasting are separate Wan-related fixes, not one code-path issue." }, { "left": "pull_request:12618", "right": "pull_request:13562", "accept": false, - "reason": "Both affect CI install commands, but one is a typo fix and the other changes tokenizers installation across multiple workflows." + "reason": "Both adjust CI install commands, but they fix different dependency/version problems." }, { "left": "pull_request:12686", "right": "pull_request:12775", "accept": false, - "reason": "Kandinsky5 docs/image pipelines vs Z-Image docs; separate features and documentation sets." + "reason": "Both are docs-related and share toctree changes, but one adds Kandinsky5 pipelines while the other adds Z-Image docs." }, { "left": "pull_request:1269", "right": "pull_request:3337", "accept": false, - "reason": "Both are refactor-like pipeline changes, but they are distinct refactors affecting different parts of stable-diffusion postprocessing and refactor scope." + "reason": "Both involve stable-diffusion-era refactors, but they refactor different code paths and different pipeline sets." }, { "left": "pull_request:12782", "right": "pull_request:13321", "accept": false, - "reason": "Offloading test tweak vs unguarded torchvision import in Cosmos; unrelated." + "reason": "Test offloading tweaks for HunyuanVideo and a guarded torchvision import for Cosmos are unrelated." }, { "left": "pull_request:1283", "right": "pull_request:483", "accept": false, - "reason": "Versatile Diffusion model addition vs typo/typo-check cleanup; not the same change." + "reason": "Versatile Diffusion model addition vs typo/typo-check docs work are completely different changes." }, { "left": "pull_request:13029", "right": "pull_request:13472", "accept": false, - "reason": "Modular quickstart docs overhaul vs an agents-docs float64 gotcha note; different documentation goals and content." + "reason": "A modular quickstart rewrite and an agent-docs float64 gotcha note are both docs, but not the same documentation fix." + }, + { + "left": "pull_request:1311", + "right": "pull_request:1386", + "accept": false, + "reason": "AltDiffusion tests and scheduler v-prediction implementation are unrelated." + }, + { + "left": "pull_request:13164", + "right": "pull_request:13294", + "accept": false, + "reason": "CI dependency cleanup and a flaky WanVACE test fix are both test-related, but they address different failures and are not mergeable as one PR." } ] }, @@ -153832,7 +153927,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 159, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -153840,15 +153935,11 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:1311", - "pull_request:13164", "pull_request:1325", - "pull_request:13294", "pull_request:13296", "pull_request:1341", "pull_request:13536", "pull_request:13562", - "pull_request:1386", "pull_request:1393", "pull_request:1402", "pull_request:1406", @@ -153868,11 +153959,14 @@ "pull_request:1765", "pull_request:1766", "pull_request:1850", + "pull_request:1868", + "pull_request:1939", "pull_request:1965", "pull_request:2009", "pull_request:2314", "pull_request:2322", "pull_request:2341", + "pull_request:286", "pull_request:2919", "pull_request:3230", "pull_request:3321", @@ -153880,6 +153974,8 @@ "pull_request:34", "pull_request:3402", "pull_request:3433", + "pull_request:3653", + "pull_request:3712", "pull_request:397", "pull_request:41", "pull_request:415", @@ -153889,8 +153985,6 @@ "pull_request:768" ], "soft_pairs": [ - "pull_request:1311|pull_request:1386", - "pull_request:13164|pull_request:13294", "pull_request:1325|pull_request:3343", "pull_request:1325|pull_request:3402", "pull_request:1325|pull_request:3433", @@ -153916,204 +154010,234 @@ "pull_request:1761|pull_request:3230", "pull_request:1765|pull_request:768", "pull_request:1766|pull_request:538", - "pull_request:1850|pull_request:2919" + "pull_request:1850|pull_request:2919", + "pull_request:1868|pull_request:286", + "pull_request:1868|pull_request:3653", + "pull_request:1868|pull_request:3712", + "pull_request:1939|pull_request:286", + "pull_request:1939|pull_request:3653", + "pull_request:1939|pull_request:3712" ], - "prepared_review_unit_hash": "7a18e13f5e30a402824c77ebe678ffe73cfed214c2a464e2f65a571c62ae9207", + "prepared_review_unit_hash": "8fb2b8940c1026a8f9ad939f0810f80b434238eb61775aacddde7658fe643b32", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 28, - "serialized_chars": 105004, - "estimated_input_tokens": 26251, - "estimated_eval_tokens": 52758 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 106188, + "estimated_input_tokens": 26547, + "estimated_eval_tokens": 53350 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: mostly unrelated PRs spanning CI, docs, tests, schedulers, and separate pipeline/features. No soft-edge pair looks like the same underlying change.", - "confidence": 0.97, + "summary": "This cluster is not a true duplicate set: it mixes unrelated feature additions, CI/workflow tweaks, docs updates, and test fixes. None of the soft edges should be accepted.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR fits this cluster well; the PRs are not duplicates and do not share one concrete bug/change.", + "canonical_pr_reason": "No single canonical PR fits: the artifacts span unrelated changes across pipelines, schedulers, CI, docs, and tests.", "best_issue_reason": null, - "best_pr_reason": "No single PR is a good global representative for dedupe; the set is too mixed. If forced, the largest feature PRs are still unrelated to the rest.", + "best_pr_reason": "PR 3230 is the broadest and most substantial artifact here, so it is the closest thing to a representative PR, but the cluster is too heterogeneous to treat as one change.", "soft_edge_verdicts": [ - { - "left": "pull_request:1311", - "right": "pull_request:1386", - "accept": false, - "reason": "AltDiffusion test additions vs scheduler v-prediction behavior; unrelated changes." - }, - { - "left": "pull_request:13164", - "right": "pull_request:13294", - "accept": false, - "reason": "CI dependency cleanup vs WanVACE flaky-test stabilization; different problems and code paths." - }, { "left": "pull_request:1325", "right": "pull_request:3343", "accept": false, - "reason": "Image2Image/OneFlow feature work vs DPM solver config fix; not the same bug/change." + "reason": "Different changes in different areas: OneFlow image2image support vs a DPM-Solver config fix." }, { "left": "pull_request:1325", "right": "pull_request:3402", "accept": false, - "reason": "Large feature PR vs Dockerfile typo/packaging fix; unrelated." + "reason": "Unrelated codepaths: OneFlow pipeline work vs a Dockerfile cleanup." }, { "left": "pull_request:1325", "right": "pull_request:3433", "accept": false, - "reason": "Feature implementation vs documentation/style-only update; not duplicate changes." + "reason": "Different concerns entirely: OneFlow image2image support vs documentation formatting/style fixes." }, { "left": "pull_request:1325", "right": "pull_request:41", "accept": false, - "reason": "Feature PR vs comment cleanup in resnet code; unrelated." + "reason": "Completely unrelated: OneFlow image2image work vs comment-only resnet cleanup." }, { "left": "pull_request:13296", "right": "pull_request:13562", "accept": false, - "reason": "Both touch CI, but one updates transformer version in release tests while the other changes tokenizers installs across workflows; not one concrete issue." + "reason": "Both touch CI, but they fix different package/version issues in the same workflow, not the same bug or change." }, { "left": "pull_request:1341", "right": "pull_request:3230", "accept": false, - "reason": "Different pipeline features (Stable Diffusion Repaint vs IF); not mergeable as one fix." + "reason": "Both add pipeline exports, but they implement different features: Stable Diffusion repaint vs IF support." }, { "left": "pull_request:13536", "right": "pull_request:13562", "accept": false, - "reason": "Workflow permissions/instructions for Claude vs CI package install changes; unrelated." + "reason": "Workflow permission/behavior changes are unrelated to the tokenizers/transformers CI install update." }, { "left": "pull_request:1393", "right": "pull_request:1406", "accept": false, - "reason": "Both scheduler/config deprecations, but they address different parameters and behaviors." + "reason": "Both are deprecation-related, but they target different config behaviors and different codepaths." }, { "left": "pull_request:1402", "right": "pull_request:34", "accept": false, - "reason": "Pipeline upscaling fix vs typo/comment cleanup; no shared underlying bug." + "reason": "A pipeline-loading bug fix is unrelated to a typo/comment fix in a scheduler file." }, { "left": "pull_request:1415", "right": "pull_request:1751", "accept": false, - "reason": "Stable diffusion broad fixes vs UnCLIP generator threading; different subsystems and bugs." + "reason": "Both affect Stable Diffusion-family pipelines, but one fixes all Stable Diffusion code paths while the other fixes UnCLIP generator handling." }, { "left": "pull_request:1458", "right": "pull_request:2009", "accept": false, - "reason": "ONNX runtime detection tweak vs textual inversion loading; unrelated code paths." + "reason": "ONNX runtime candidate detection is unrelated to textual inversion loader support." }, { "left": "pull_request:1468", "right": "pull_request:2341", "accept": false, - "reason": "Scheduler betas serialization vs DreamBooth EMAModel saving; different changes." + "reason": "Saving trained betas in schedulers is unrelated to EMA saving in a training example." }, { "left": "pull_request:1487", "right": "pull_request:3230", "accept": false, - "reason": "xFormers docs vs IF pipeline implementation; documentation does not match the feature PR." + "reason": "Docs for xFormers are unrelated to adding the IF pipeline." }, { "left": "pull_request:1548", "right": "pull_request:2322", "accept": false, - "reason": "Issue-template wording vs example/inpaint script update; unrelated." + "reason": "Bug-report template wording is unrelated to the example/pipeline code added in the other PR." }, { "left": "pull_request:1588", "right": "pull_request:2314", "accept": false, - "reason": "FP16 common-test fix vs fast-test threshold adjustment; both tests, but different concrete failures." + "reason": "Fast-test FP16 fixes are unrelated to image-slice tolerance updates." }, { "left": "pull_request:1667", "right": "pull_request:1965", "accept": false, - "reason": "Stable Diffusion conversion scheduler fix vs checkpoint merger import cleanup; unrelated." + "reason": "Scheduler config handling for SD conversion is unrelated to the checkpoint merger import fix." }, { "left": "pull_request:1706", "right": "pull_request:3321", "accept": false, - "reason": "General docs update vs controlling-generation docs update; same broad area, but not the same change or bug." + "reason": "General docs refresh is unrelated to the controlling-generation documentation update." }, { "left": "pull_request:1713", "right": "pull_request:423", "accept": false, - "reason": "Conversion-script refactor vs docs/output syntax improvements; unrelated." + "reason": "Different changes: Stable Diffusion conversion cleanup vs output/documentation syntax improvements." }, { "left": "pull_request:1724", "right": "pull_request:397", "accept": false, - "reason": "xFormers recommendation docs vs quicktour docs; both docs, but not the same underlying issue." + "reason": "Both are docs, but one is xFormers guidance and the other is the quicktour rewrite; not the same change." }, { "left": "pull_request:1724", "right": "pull_request:415", "accept": false, - "reason": "xFormers docs vs training docs rewrite; separate documentation changes." + "reason": "xFormers recommendations are unrelated to training overview docs refresh." }, { "left": "pull_request:1724", "right": "pull_request:418", "accept": false, - "reason": "xFormers docs vs DiffusionPipeline docs rework; not the same topic or fix." + "reason": "xFormers docs and DiffusionPipeline docs are separate documentation efforts." }, { "left": "pull_request:1757", "right": "pull_request:3230", "accept": false, - "reason": "Versatile Diffusion layer parameter propagation vs IF feature addition; unrelated." + "reason": "Adding resnet_time_scale_shift in versatile diffusion is unrelated to the IF pipeline feature." }, { "left": "pull_request:1761", "right": "pull_request:3230", "accept": false, - "reason": "UnCLIP CPU offload vs IF pipeline addition; different pipeline areas and goals." + "reason": "UnCLIP CPU offloading and IF pipeline addition are different features and codepaths." }, { "left": "pull_request:1765", "right": "pull_request:768", "accept": false, - "reason": "DreamBooth flax fixes vs DreamBooth enhancement script scaffolding; not the same concrete bug/change." + "reason": "Dreambooth Flax fixes are unrelated to the Dreambooth enhancement example." }, { "left": "pull_request:1766", "right": "pull_request:538", "accept": false, - "reason": "UnCLIP transformers version gating vs hub dependency version bump; unrelated dependency issues." + "reason": "UnCLIP version gating is unrelated to upgrading the huggingface-hub dependency." }, { "left": "pull_request:1850", "right": "pull_request:2919", "accept": false, - "reason": "ResnetBlock2D unit tests vs speeding up a 3D UNet test; both tests only, no shared bug/change." + "reason": "ResnetBlock2D tests and a 3D UNet test speedup are different test changes." + }, + { + "left": "pull_request:1868", + "right": "pull_request:286", + "accept": false, + "reason": "EMA decay logic in a training script is unrelated to a README link fix." + }, + { + "left": "pull_request:1868", + "right": "pull_request:3653", + "accept": false, + "reason": "EMA decay changes are unrelated to a linting cleanup in a community example." + }, + { + "left": "pull_request:1868", + "right": "pull_request:3712", + "accept": false, + "reason": "Training-script EMA logic is unrelated to documentation workflow secret renames." + }, + { + "left": "pull_request:1939", + "right": "pull_request:286", + "accept": false, + "reason": "README image path correction is unrelated to the stable-diffusion README link fix." + }, + { + "left": "pull_request:1939", + "right": "pull_request:3653", + "accept": false, + "reason": "A README asset-path fix is unrelated to a community-example lint fix." + }, + { + "left": "pull_request:1939", + "right": "pull_request:3712", + "accept": false, + "reason": "A README path correction is unrelated to documentation workflow secret changes." } ] }, @@ -154129,7 +154253,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 160, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -154137,8 +154261,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:1868", - "pull_request:1939", "pull_request:1971", "pull_request:2001", "pull_request:2006", @@ -154161,6 +154283,7 @@ "pull_request:2436", "pull_request:2475", "pull_request:252", + "pull_request:2526", "pull_request:2542", "pull_request:2655", "pull_request:2706", @@ -154170,11 +154293,12 @@ "pull_request:3069", "pull_request:3172", "pull_request:3262", + "pull_request:3400", "pull_request:3402", "pull_request:3581", - "pull_request:3653", "pull_request:3708", "pull_request:3712", + "pull_request:3757", "pull_request:3786", "pull_request:3916", "pull_request:397", @@ -154187,12 +154311,6 @@ "pull_request:979" ], "soft_pairs": [ - "pull_request:1868|pull_request:286", - "pull_request:1868|pull_request:3653", - "pull_request:1868|pull_request:3712", - "pull_request:1939|pull_request:286", - "pull_request:1939|pull_request:3653", - "pull_request:1939|pull_request:3712", "pull_request:1971|pull_request:2007", "pull_request:1971|pull_request:3262", "pull_request:2001|pull_request:3708", @@ -154220,240 +154338,218 @@ "pull_request:2436|pull_request:979", "pull_request:2475|pull_request:2542", "pull_request:2475|pull_request:2912", - "pull_request:2475|pull_request:2958" + "pull_request:2475|pull_request:2958", + "pull_request:2475|pull_request:3757", + "pull_request:2526|pull_request:3400" ], - "prepared_review_unit_hash": "b6d10d3df0a90966d5cb9fe90034c9dbe01e5d7ab1f6e55fc74dd2fbab6a0095", + "prepared_review_unit_hash": "3f7317e577ce08128800498ea668fc7b8baaff6f0db81d7390b94ade6be28cdf", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 34, - "serialized_chars": 96508, - "estimated_input_tokens": 24127, - "estimated_eval_tokens": 48510 + "soft_pair_count": 30, + "serialized_chars": 96641, + "estimated_input_tokens": 24161, + "estimated_eval_tokens": 48578 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose collection of unrelated diffusers PRs spanning docs, tests, examples, workflow, and code fixes. I found no pair that looks like the same underlying bug/change, so none of the soft edges should be merged.", + "summary": "This cluster is highly heterogeneous: mostly unrelated docs/typo/test/example maintenance PRs, plus a few ControlNet-related feature PRs and other isolated changes. I don\u2019t see a single underlying bug or change that ties them together, so no soft edges are accepted.", "confidence": 0.98, - "canonical_issue_reason": "No issue artifacts are present in this cluster, so there is no issue candidate to canonicalize.", - "canonical_pr_reason": "There is no single canonical PR here: the items cover many unrelated one-off changes rather than one duplicate fix or feature.", + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR fits the whole cluster; the items span unrelated docs, tests, examples, workflows, and feature work.", "best_issue_reason": null, - "best_pr_reason": "No clear best PR exists for the cluster as a whole. The PRs are not duplicates of one another, so picking any one as a representative would be arbitrary.", + "best_pr_reason": "No PR is a strong global representative. PR 2475 is the largest substantive item, but it only matches a small subset of the cluster and not the rest.", "soft_edge_verdicts": [ - { - "left": "pull_request:1868", - "right": "pull_request:286", - "accept": false, - "reason": "EMA decay fix in training code vs README link fix; unrelated files and problems." - }, - { - "left": "pull_request:1868", - "right": "pull_request:3653", - "accept": false, - "reason": "EMA decay logic vs a linting cleanup in mixture_tiling.py; different code paths." - }, - { - "left": "pull_request:1868", - "right": "pull_request:3712", - "accept": false, - "reason": "Training EMA bug fix vs workflow secret rename; no shared underlying change." - }, - { - "left": "pull_request:1939", - "right": "pull_request:286", - "accept": false, - "reason": "README logo path vs stable diffusion README link; both docs, but different targets and fixes." - }, - { - "left": "pull_request:1939", - "right": "pull_request:3653", - "accept": false, - "reason": "Logo path correction vs code lint cleanup; unrelated." - }, - { - "left": "pull_request:1939", - "right": "pull_request:3712", - "accept": false, - "reason": "README image path vs GitHub Actions secrets; unrelated changes." - }, { "left": "pull_request:1971", "right": "pull_request:2007", "accept": false, - "reason": "PNDM docstring wording vs black formatting sweep; not the same artifact or bug." + "reason": "Docstring cleanup vs broad black-formatting sweep; different changes." }, { "left": "pull_request:1971", "right": "pull_request:3262", "accept": false, - "reason": "PNDM docstring cleanup vs IF model name/docs update; different documentation changes." + "reason": "Docstring wording fix vs IF model-name update; unrelated." }, { "left": "pull_request:2001", "right": "pull_request:3708", "accept": false, - "reason": "ColossalAI example seed handling/typos vs dynamic module version sorting fix; unrelated code paths." + "reason": "ColossalAI seed launch fix vs version-sorting utility change; different code paths." }, { "left": "pull_request:2006", "right": "pull_request:3712", "accept": false, - "reason": "Checkpoint merger fixes vs docs workflow secret rename; different subsystems." + "reason": "Checkpoint merger fix vs workflow secret rename; unrelated." }, { "left": "pull_request:2007", "right": "pull_request:3581", "accept": false, - "reason": "Repository-wide black formatting update vs removal of an unused import in bit_diffusion.py; not duplicate fixes." + "reason": "Black formatting PR vs bit_diffusion import cleanup; not the same change." }, { "left": "pull_request:21", "right": "pull_request:7", "accept": false, - "reason": "Early model test additions vs initial VQ-VAE implementation; different work entirely." + "reason": "Model test additions vs initial VQ-VAE implementation; different artifacts." }, { "left": "pull_request:21", "right": "pull_request:909", "accept": false, - "reason": "Model tests vs inpaint pipeline test updates; both tests, but not the same concrete change." + "reason": "Different test updates on different model paths; not a duplicate change." }, { "left": "pull_request:2138", "right": "pull_request:2227", "accept": false, - "reason": "Single docs typo fix vs broad negative_prompt typo sweep across pipelines; related wording, but not one duplicate change." + "reason": "Docs typo fix vs pipeline API typos across many files; different scope and content." }, { "left": "pull_request:2138", "right": "pull_request:3402", "accept": false, - "reason": "Docs typo vs Dockerfile dependency fix; unrelated." + "reason": "Docs typo fix vs Dockerfile fix; unrelated." }, { "left": "pull_request:217", "right": "pull_request:286", "accept": false, - "reason": "Safety checker logic fix vs README link fix; different code and intent." + "reason": "Safety-checker logic bug vs README link correction; different concerns." }, { "left": "pull_request:2190", "right": "pull_request:3916", "accept": false, - "reason": "Two separate branch-sync PRs with different large content sets; not the same change." + "reason": "Unrelated branch syncs touching different example sets." }, { "left": "pull_request:2197", "right": "pull_request:4239", "accept": false, - "reason": "StableDiffusionInpaintPipeline typo/fix vs SDXL LoRA text_encoder fix; different pipelines and bugs." + "reason": "Inpaint pipeline typo/input fix vs LoRA text-encoder support fix; different code paths." }, { "left": "pull_request:22", "right": "pull_request:7", "accept": false, - "reason": "Model-loading warning fix/tests vs VQ-VAE introduction; unrelated." + "reason": "Modeling-utils warning typo vs VQ-VAE addition; unrelated." }, { "left": "pull_request:22", "right": "pull_request:909", "accept": false, - "reason": "Model-loading warning fix/tests vs inpaint test changes; different areas." + "reason": "Warning-logging typo fix vs inpaint test/model-id update; different changes." }, { "left": "pull_request:2208", "right": "pull_request:3172", "accept": false, - "reason": "xFormers docs update vs deterministic algorithms docs; separate documentation topics." + "reason": "Separate documentation updates on different topics." }, { "left": "pull_request:2295", "right": "pull_request:2706", "accept": false, - "reason": "Transformer layer configurability vs EMA save/restore in training; unrelated." + "reason": "Configurable transformer-depth feature vs EMA save/restore training fix; not the same bug." }, { "left": "pull_request:2314", "right": "pull_request:2655", "accept": false, - "reason": "Fast-test expectation updates vs LoRA save/load logic; different concerns." + "reason": "Slow-test threshold tweaks vs LoRA loading/saving logic; unrelated." }, { "left": "pull_request:2322", "right": "pull_request:252", "accept": false, - "reason": "New layer.ai inpaint example vs README examples expansion; both examples-related, but not the same underlying artifact." + "reason": "New inpaint example vs README examples update; different deliverables." }, { "left": "pull_request:2322", "right": "pull_request:3069", "accept": false, - "reason": "New example script vs Flax TPU test expectation changes; unrelated." + "reason": "Example script addition vs Flax TPU test updates; unrelated." }, { "left": "pull_request:2329", "right": "pull_request:774", "accept": false, - "reason": "Workflow refactor vs test assertion adjustments; same broad test area, but not the same fix." + "reason": "Workflow refactor vs callback test adjustments; different code paths." }, { "left": "pull_request:2362", "right": "pull_request:3786", "accept": false, - "reason": "MPS skip decorator vs PR template addition; unrelated repository plumbing." + "reason": "MPS skip decorator vs PR template addition; unrelated." }, { "left": "pull_request:2388", "right": "pull_request:397", "accept": false, - "reason": "Controlling generation docs vs quicktour docs; different docs pages and content." + "reason": "Different documentation pages and scopes, not the same change." }, { "left": "pull_request:2388", "right": "pull_request:415", "accept": false, - "reason": "Controlling generation docs vs training overview docs; separate documentation additions." + "reason": "Control docs vs training docs; unrelated documentation work." }, { "left": "pull_request:2401", "right": "pull_request:4081", "accept": false, - "reason": "Index page addition vs README model table update; unrelated docs changes." + "reason": "Docs index overhaul vs README table entry; not a duplicate." }, { "left": "pull_request:2436", "right": "pull_request:979", "accept": false, - "reason": "Attend-and-Excite demo link vs missing import in README code sample; unrelated." + "reason": "Demo-link docs change vs missing import in README example; different artifacts." }, { "left": "pull_request:2475", "right": "pull_request:2542", "accept": false, - "reason": "Large ControlNet implementation vs small ControlNet docs tweaks; docs and code are not the same duplicate change." + "reason": "ControlNet feature work vs small docs fixes; unrelated." }, { "left": "pull_request:2475", "right": "pull_request:2912", "accept": false, - "reason": "ControlNet implementation vs a punctuation fix in controlnet.mdx; not the same underlying bug/change." + "reason": "Major ControlNet addition vs punctuation/doc tweak; not the same change." }, { "left": "pull_request:2475", "right": "pull_request:2958", "accept": false, - "reason": "ControlNet implementation vs ControlNet inpaint addition; related area, but distinct pipelines and code paths." + "reason": "Base ControlNet feature vs ControlNet inpaint variant; related subsystem but different concrete PRs." + }, + { + "left": "pull_request:2475", + "right": "pull_request:3757", + "accept": false, + "reason": "Base ControlNet feature vs conditioning-channels config change; different functionality." + }, + { + "left": "pull_request:2526", + "right": "pull_request:3400", + "accept": false, + "reason": "Test-threshold adjustments vs dependency/Dockerfile changes; unrelated." } ] }, @@ -154469,7 +154565,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 161, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -154477,8 +154573,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2475", - "pull_request:2526", "pull_request:2533", "pull_request:2590", "pull_request:2693", @@ -154499,25 +154593,28 @@ "pull_request:3318", "pull_request:3343", "pull_request:34", - "pull_request:3400", "pull_request:3402", "pull_request:3433", "pull_request:3474", + "pull_request:3524", + "pull_request:3525", "pull_request:3608", "pull_request:3707", "pull_request:3709", - "pull_request:3757", "pull_request:3931", "pull_request:3955", "pull_request:4081", "pull_request:41", "pull_request:4495", "pull_request:4547", + "pull_request:4748", "pull_request:4963", "pull_request:4973", + "pull_request:5267", "pull_request:5286", "pull_request:5478", "pull_request:5622", + "pull_request:5855", "pull_request:693", "pull_request:705", "pull_request:828", @@ -154526,8 +154623,6 @@ "pull_request:975" ], "soft_pairs": [ - "pull_request:2475|pull_request:3757", - "pull_request:2526|pull_request:3400", "pull_request:2533|pull_request:4547", "pull_request:2590|pull_request:2970", "pull_request:2693|pull_request:3707", @@ -154554,210 +154649,233 @@ "pull_request:34|pull_request:948", "pull_request:3402|pull_request:5622", "pull_request:3433|pull_request:5622", - "pull_request:3474|pull_request:5478" + "pull_request:3474|pull_request:5478", + "pull_request:3524|pull_request:3525", + "pull_request:3524|pull_request:3931", + "pull_request:3524|pull_request:4748", + "pull_request:3524|pull_request:5267", + "pull_request:3524|pull_request:5855" ], - "prepared_review_unit_hash": "cefdfd67ecfc97b1046043da5a80a12149f9a86b5b2f59f63e437d22d198afdf", + "prepared_review_unit_hash": "bfb48f8c2be7228a509717dad28da84e4ad5a30eb0c68b2c13c55fe37e205b34", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 29, - "serialized_chars": 92850, - "estimated_input_tokens": 23213, - "estimated_eval_tokens": 46682 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 93442, + "estimated_input_tokens": 23361, + "estimated_eval_tokens": 46978 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These artifacts are mostly unrelated housekeeping or feature PRs (docs, tests, CI, small bugfixes). None of the soft pairs look like the same underlying change or bug, so no duplicates should be merged.", - "confidence": 0.98, - "canonical_issue_reason": null, - "canonical_pr_reason": null, - "best_issue_reason": null, - "best_pr_reason": "PR 3230 is the most substantial and central artifact in this set: it introduces the IF pipeline end-to-end with docs, code, and tests, and it is the one most frequently linked by the soft-similarity heuristic.", + "summary": "No duplicate cluster here: the PRs span unrelated docs, CI, example scripts, scheduler fixes, and feature work. The soft links are mostly superficial similarity on filenames or titles, not the same concrete change.", + "confidence": 0.97, + "canonical_issue_reason": "No issue artifacts are present in this cluster.", + "canonical_pr_reason": "No single PR is a credible canonical duplicate target; the items are heterogeneous and do not share one underlying bug or change.", + "best_issue_reason": "No issue artifacts are present in this cluster.", + "best_pr_reason": "No PR stands out as a global representative duplicate; the cluster is too mixed across unrelated changes.", "soft_edge_verdicts": [ - { - "left": "pull_request:2475", - "right": "pull_request:3757", - "accept": false, - "reason": "Both are ControlNet-related, but one adds the initial pipeline and the other adds configurable conditioning channels; different concrete changes." - }, - { - "left": "pull_request:2526", - "right": "pull_request:3400", - "accept": false, - "reason": "Unrelated: one relaxes slow test thresholds, the other adds omegaconf to packaging/build config." - }, { "left": "pull_request:2533", "right": "pull_request:4547", "accept": false, - "reason": "Different areas and goals: safetensors RC testing vs a Flax scheduler implementation." + "reason": "WIP testing rc1 vs Flax scheduler implementation; unrelated changes and code paths." }, { "left": "pull_request:2590", "right": "pull_request:2970", "accept": false, - "reason": "One updates CI to check deps tables; the other renames a local variable in a training script." + "reason": "CI dependency-table check vs a variable rename in a dreambooth example; not the same change." }, { "left": "pull_request:2693", "right": "pull_request:3707", "accept": false, - "reason": "Both are doc fixes, but for different docs and different problems; not the same underlying change." + "reason": "Doc image-link fix vs doc-build secrets/workflow changes; different files and problems." }, { "left": "pull_request:2701", "right": "pull_request:3230", "accept": false, - "reason": "Different scope: ONNX Runtime import support vs adding the IF pipeline." + "reason": "ONNX Runtime support list update vs the IF pipeline landing; shared utility area but unrelated functionality." }, { "left": "pull_request:2759", "right": "pull_request:828", "accept": false, - "reason": "Unrelated test/CI changes in different stacks; no shared bug or feature." + "reason": "Large test-run/workflow debug PR vs Flax test completion; different scopes and code paths." }, { "left": "pull_request:2762", "right": "pull_request:3230", "accept": false, - "reason": "Multiple ControlNets docs section vs the IF pipeline implementation/docs; not the same change." + "reason": "ControlNet docs for multi-conditioning vs IF feature implementation; both docs/feature work but not the same change." }, { "left": "pull_request:2811", "right": "pull_request:705", "accept": false, - "reason": "Both touch tests/CI, but one updates ONNX img2img expected values and the other changes CUDA CI setup." + "reason": "onnx PNDM test slice update vs CUDA CI workflow/dependency changes; unrelated." }, { "left": "pull_request:2814", "right": "pull_request:909", "accept": false, - "reason": "Stable UnCLIP offload support is unrelated to the stable diffusion inpaint test update." + "reason": "Stable UnCLIP completion vs Stable Diffusion inpaint test/model-path updates; different pipeline work." }, { "left": "pull_request:286", "right": "pull_request:41", "accept": false, - "reason": "Both are old doc/comment cleanups, but they affect different files and unrelated wording fixes." + "reason": "Stable Diffusion README link fix vs ancient resnet comment cleanup; no shared bug." }, { "left": "pull_request:286", "right": "pull_request:693", "accept": false, - "reason": "Different README citation/link tweaks; not the same underlying fix." + "reason": "README link fix vs citation-format typo; only superficial README overlap." }, { "left": "pull_request:3042", "right": "pull_request:4963", "accept": false, - "reason": "Different dependency/compatibility issues: accelerate minimum version vs installing accelerate main for tests." + "reason": "Accelerate minimum-version updates in example requirements vs installing accelerate main in CI; related dependency area but different concrete fixes." }, { "left": "pull_request:3063", "right": "pull_request:3955", "accept": false, - "reason": "Unrelated code paths: pipeline __setattr__ bug vs Kandinsky prior parameter docs." + "reason": "Pipeline __setattr__ None handling vs Kandinsky parameter docs; code fix vs documentation." }, { "left": "pull_request:3108", "right": "pull_request:3931", "accept": false, - "reason": "No diff context for 3108, but it does not match the ckpt-loading changes in 3931; insufficient basis for duplication." + "reason": "Empty/unknown checkpoint fix vs ckpt loading overhaul; no evidence of same underlying bug." }, { "left": "pull_request:3169", "right": "pull_request:975", "accept": false, - "reason": "Habana docs update vs a README title typo; different documentation changes." + "reason": "Habana docs update vs OpenVINO title typo; different docs and unrelated content." }, { "left": "pull_request:3217", "right": "pull_request:3230", "accept": false, - "reason": "Both touch pipeline_utils/docs utilities, but 3217 adds a PIL utility refactor while 3230 adds IF; distinct changes." + "reason": "Add pt_to_pil utility refactor vs IF feature addition; same module area but distinct changes." }, { "left": "pull_request:3230", "right": "pull_request:3709", "accept": false, - "reason": "Shared docs tree file, but 3230 adds IF docs and code whereas 3709 only adds an image processor doc page." + "reason": "IF pipeline addition vs adding image-processor docs; unrelated docs/feature work." }, { "left": "pull_request:3230", "right": "pull_request:4081", "accept": false, - "reason": "IF pipeline addition is unrelated to adding Kandinsky to the README table." + "reason": "IF pipeline landing vs Kandinsky README table row; different features and files." }, { "left": "pull_request:3230", "right": "pull_request:4495", "accept": false, - "reason": "Completely different feature areas: IF vs SDXL reference/community example." + "reason": "IF pipeline addition vs SDXL reference community example; unrelated pipeline work." }, { "left": "pull_request:3230", "right": "pull_request:4973", "accept": false, - "reason": "IF pipeline addition and README install badge are unrelated." + "reason": "IF pipeline feature vs monthly install badge in README; no shared underlying change." }, { "left": "pull_request:3230", "right": "pull_request:5286", "accept": false, - "reason": "IF pipeline addition is unrelated to adding Japanese documentation." + "reason": "IF pipeline addition vs Japanese docs skeleton; unrelated." }, { "left": "pull_request:3295", "right": "pull_request:3608", "accept": false, - "reason": "Both are tiny doc/example typo fixes, but in different tutorials and with different text." + "reason": "Tutorial typo in PIL conversion vs train_text_to_image arg typo; both typo fixes but different code locations and behavior." }, { "left": "pull_request:3318", "right": "pull_request:4547", "accept": false, - "reason": "IF textual inversion work is unrelated to the Flax Euler scheduler implementation." + "reason": "WIP IF textual inversion vs Flax scheduler implementation; different model families and changes." }, { "left": "pull_request:3343", "right": "pull_request:5622", "accept": false, - "reason": "Scheduler config bugfix vs a large examples/README overhaul; not the same change." + "reason": "DPMSolver config bugfix vs example-training optimization cleanup; unrelated." }, { "left": "pull_request:34", "right": "pull_request:948", "accept": false, - "reason": "Both fix comments, but they are separate typo corrections in different pipeline files." + "reason": "Comment typo cleanup in DDPM vs another DDPM comment typo fix years later; not the same concrete patch." }, { "left": "pull_request:3402", "right": "pull_request:5622", "accept": false, - "reason": "Dockerfile fix unrelated to example-training script/content changes." + "reason": "Dockerfile dependency install fix vs large examples/optimization refactor; unrelated." }, { "left": "pull_request:3433", "right": "pull_request:5622", "accept": false, - "reason": "Documentation style rendering tweaks are unrelated to the examples training overhaul." + "reason": "Docs style rendering cleanup vs examples/optimization changes; not the same issue." }, { "left": "pull_request:3474", "right": "pull_request:5478", "accept": false, - "reason": "Docs for attention processors vs adding Japanese docs; different documentation efforts." + "reason": "Attention processor docs vs Japanese docs rollout; both documentation but completely different artifacts." + }, + { + "left": "pull_request:3524", + "right": "pull_request:3525", + "accept": false, + "reason": "Two README edits, but one rewrites documentation sections and the other tweaks a usage snippet and badges; not one merged change." + }, + { + "left": "pull_request:3524", + "right": "pull_request:3931", + "accept": false, + "reason": "README rewrite vs ckpt loading logic; unrelated." + }, + { + "left": "pull_request:3524", + "right": "pull_request:4748", + "accept": false, + "reason": "README rewrite vs setup.py release-publishing instructions; different documentation targets and intent." + }, + { + "left": "pull_request:3524", + "right": "pull_request:5267", + "accept": false, + "reason": "README rewrite vs docs/README guidance updates; both docs but not the same concrete change." + }, + { + "left": "pull_request:3524", + "right": "pull_request:5855", + "accept": false, + "reason": "README rewrite vs community README typo fix; same file type but distinct edits and scopes." } ] }, @@ -154773,7 +154891,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 162, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -154781,7 +154899,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:3524", "pull_request:3525", "pull_request:3563", "pull_request:3641", @@ -154795,7 +154912,6 @@ "pull_request:390", "pull_request:391", "pull_request:392", - "pull_request:3931", "pull_request:3955", "pull_request:396", "pull_request:397", @@ -154812,30 +154928,26 @@ "pull_request:422", "pull_request:4239", "pull_request:4273", + "pull_request:4275", + "pull_request:4276", "pull_request:428", "pull_request:4284", "pull_request:4498", - "pull_request:4748", "pull_request:4807", "pull_request:4822", - "pull_request:5267", "pull_request:5729", - "pull_request:5855", + "pull_request:5792", "pull_request:5948", "pull_request:6078", "pull_request:6150", "pull_request:6337", "pull_request:6410", "pull_request:6491", + "pull_request:6499", "pull_request:6591", "pull_request:662" ], "soft_pairs": [ - "pull_request:3524|pull_request:3525", - "pull_request:3524|pull_request:3931", - "pull_request:3524|pull_request:4748", - "pull_request:3524|pull_request:5267", - "pull_request:3524|pull_request:5855", "pull_request:3525|pull_request:5729", "pull_request:3563|pull_request:3989", "pull_request:3641|pull_request:5948", @@ -154863,234 +154975,225 @@ "pull_request:418|pull_request:428", "pull_request:4203|pull_request:6491", "pull_request:4239|pull_request:6337", - "pull_request:4239|pull_request:6591" + "pull_request:4239|pull_request:6591", + "pull_request:4273|pull_request:6499", + "pull_request:4275|pull_request:6410", + "pull_request:4276|pull_request:5792" ], - "prepared_review_unit_hash": "47a461f1bc93d64ac52ae7257ecfea7651d7321c3d095b8352e49bfb7e28dcd6", + "prepared_review_unit_hash": "f923be4b9582f58e86d3d7a682c1f596ab4291910c51dbb12ec811919f9b2bcb", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 33, - "serialized_chars": 98991, - "estimated_input_tokens": 24748, - "estimated_eval_tokens": 49752 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 97774, + "estimated_input_tokens": 24444, + "estimated_eval_tokens": 49144 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab-bag of unrelated docs/typo cleanups and a few distinct code fixes; none of the soft-edge pairs look like the same underlying change.", - "confidence": 0.96, + "summary": "This cluster is mostly a grab bag of unrelated docs, typo, example, test, and small bugfix PRs. The soft edges are driven by superficial similarity (same repo area, docs, or tiny edits) rather than the same underlying change.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 6591 is the broadest concrete code-fix PR in the set, but it is still just a typo cleanup across many example pipelines and not a good cluster-wide duplicate target.", + "best_pr_reason": "PR 3835 is the broadest and most central documentation/API sweep in the set, so it is the closest thing to a representative anchor, though the cluster is too heterogeneous to treat as a true duplicate group.", "soft_edge_verdicts": [ - { - "left": "pull_request:3524", - "right": "pull_request:3525", - "accept": false, - "reason": "Both touch README.md, but one is a broad README cleanup and the other is a different README example/update; not the same change." - }, - { - "left": "pull_request:3524", - "right": "pull_request:3931", - "accept": false, - "reason": "README cleanup vs ckpt loading code changes in loaders.py; unrelated." - }, - { - "left": "pull_request:3524", - "right": "pull_request:4748", - "accept": false, - "reason": "README cleanup vs setup.py packaging instructions; unrelated." - }, - { - "left": "pull_request:3524", - "right": "pull_request:5267", - "accept": false, - "reason": "README formatting/docs style fix vs docs/README contribution syntax changes; different scope and content." - }, - { - "left": "pull_request:3524", - "right": "pull_request:5855", - "accept": false, - "reason": "Both edit README.md, but they change different examples/content and are not the same underlying edit." - }, { "left": "pull_request:3525", "right": "pull_request:5729", "accept": false, - "reason": "README.md tweak vs release/version check updates across training scripts; unrelated." + "reason": "README wording/usage examples vs a large training-script version check update; different change intents and code paths." }, { "left": "pull_request:3563", "right": "pull_request:3989", "accept": false, - "reason": "Different problems: vq pipeline norm handling vs SDXL offload/test fixes." + "reason": "A UNet block bugfix for vq pipeline vs SDXL test/offload changes; not the same underlying bug." }, { "left": "pull_request:3641", "right": "pull_request:5948", "accept": false, - "reason": "Inpainting latents bug vs deprecated variant warning logic; different code paths." + "reason": "Inpainting latent handling fix vs a from_pretrained deprecation warning tweak; unrelated functionality." }, { "left": "pull_request:3687", "right": "pull_request:4807", "accept": false, - "reason": "Doc-builder test page vs Chinese Stable Diffusion documentation; unrelated." + "reason": "Doc-builder test edit vs Chinese documentation additions; unrelated artifacts." }, { "left": "pull_request:3723", "right": "pull_request:4498", "accept": false, - "reason": "Reference pipeline input handling vs adding SDXL reference/adain support; different feature work." + "reason": "Reference-image example updates vs a new SDXL reference pipeline feature; not the same change." }, { "left": "pull_request:3798", "right": "pull_request:6078", "accept": false, - "reason": "Comment cleanup in training scripts vs Euler scheduler sigma logic; unrelated." + "reason": "Comment cleanup in training scripts vs Euler scheduler sigma logic; different issues." }, { "left": "pull_request:3806", "right": "pull_request:6150", "accept": false, - "reason": "Checkpoint-loading README note vs IP-Adapter conversion code in loaders; unrelated." + "reason": "Checkpoint-loading README note vs IP-Adapter conversion code; unrelated." }, { "left": "pull_request:3820", "right": "pull_request:3955", "accept": false, - "reason": "LDM3D typo/docs cleanup vs Kandinsky prior parameter description fixes; different components." + "reason": "LDM3D typo/docs cleanup vs Kandinsky prior parameter/default changes; different pipelines and bugs." }, { "left": "pull_request:3820", "right": "pull_request:4284", "accept": false, - "reason": "Both touch LDM3D, but they fix different typos/docs snippets and are not the same concrete change." + "reason": "Both touch LDM3D, but one is a small typo/example cleanup and the other renames the output class/docs; not one concrete fix." }, { "left": "pull_request:3835", "right": "pull_request:4273", "accept": false, - "reason": "Broad API docs cleanup vs AutoPipeline tutorial/doc additions; different work items." + "reason": "Broad API docs refactor vs AutoPipeline tutorial/docs; same docs area but not the same change." }, { "left": "pull_request:386", "right": "pull_request:396", "accept": false, - "reason": "Stable Diffusion pipeline docs vs DDIM docs; unrelated." + "reason": "Stable Diffusion pipeline docs vs DDIM docs; different models and pages." }, { "left": "pull_request:386", "right": "pull_request:416", "accept": false, - "reason": "Pipeline docs vs model API docs; different documentation areas." + "reason": "Stable Diffusion pipeline docs vs models API docs; unrelated documentation targets." }, { "left": "pull_request:390", "right": "pull_request:416", "accept": false, - "reason": "MPS optimization docs vs model API docs; unrelated." + "reason": "MPS optimization docs vs models API docs; unrelated topics." }, { "left": "pull_request:391", "right": "pull_request:392", "accept": false, - "reason": "PNDM docs vs conceptual contribution docs; unrelated." + "reason": "PNDM docs vs contribution docs; no shared underlying bug or feature." }, { "left": "pull_request:391", "right": "pull_request:404", "accept": false, - "reason": "PNDM docs vs fp16 optimization docs; different guides." + "reason": "PNDM docs vs fp16 optimization docs; different documentation subjects." }, { "left": "pull_request:391", "right": "pull_request:417", "accept": false, - "reason": "PNDM docs vs inference pipelines docs; not the same change." + "reason": "PNDM docs vs inference pipeline tutorials; unrelated docs changes." }, { "left": "pull_request:397", "right": "pull_request:422", "accept": false, - "reason": "Quicktour docs vs outputs API docs; unrelated." + "reason": "Quicktour rewrite vs outputs API docs; different documentation areas." }, { "left": "pull_request:397", "right": "pull_request:428", "accept": false, - "reason": "Quicktour docs vs using-diffusers guide rewrite; different documentation set and intent." + "reason": "Quicktour overhaul vs Using Diffusers docs consolidation; both docs, but not the same concrete change." }, { "left": "pull_request:4082", "right": "pull_request:6410", "accept": false, - "reason": "SDXL requirement pin vs RL module registration fix; unrelated." + "reason": "SDXL dependency pin vs RL module registration fix; unrelated code paths." }, { "left": "pull_request:41", "right": "pull_request:662", "accept": false, - "reason": "Comment cleanup in core models vs textual inversion example/package changes; unrelated." + "reason": "Resnet comment cleanup vs textual inversion example/assets update; clearly different." }, { "left": "pull_request:4113", "right": "pull_request:4822", "accept": false, - "reason": "SDXL inpaint dtype fix vs LoRA warning typo fix; different bug classes." + "reason": "SDXL inpaint dtype fix vs LoRA warning typo; unrelated bug fixes." }, { "left": "pull_request:415", "right": "pull_request:422", "accept": false, - "reason": "Training overview docs vs outputs API docs; unrelated." + "reason": "Training docs overhaul vs outputs API docs; different subjects." }, { "left": "pull_request:415", "right": "pull_request:428", "accept": false, - "reason": "Training docs vs using-diffusers docs; different guides." + "reason": "Training docs vs Using Diffusers guides; both docs, but not the same change or page." }, { "left": "pull_request:418", "right": "pull_request:422", "accept": false, - "reason": "DiffusionPipeline docs vs outputs docs; unrelated." + "reason": "DiffusionPipeline docs vs outputs API docs; separate foundational docs pages." }, { "left": "pull_request:418", "right": "pull_request:428", "accept": false, - "reason": "DiffusionPipeline docs vs using-diffusers docs; different content." + "reason": "DiffusionPipeline docs vs Using Diffusers docs; related only at a high level, not the same change." }, { "left": "pull_request:4203", "right": "pull_request:6491", "accept": false, - "reason": "Dreambooth README typo vs empty/no-diff typo PR; not a concrete same change." + "reason": "Dreambooth README typo fix vs a zero-diff typo PR; not a duplicate fix." }, { "left": "pull_request:4239", "right": "pull_request:6337", "accept": false, - "reason": "SDXL LoRA text encoder fix vs TensorRT img2img latent retrieval fix; unrelated." + "reason": "Text-encoder loading fix vs TensorRT latent retrieval change; different code paths and problems." }, { "left": "pull_request:4239", "right": "pull_request:6591", "accept": false, - "reason": "SDXL LoRA fix vs broad typo cleanup across many examples; different underlying changes." + "reason": "SDXL LoRA/text_encoder fix vs broad typo cleanup across examples and pipelines; unrelated." + }, + { + "left": "pull_request:4273", + "right": "pull_request:6499", + "accept": false, + "reason": "AutoPipeline tutorial/docs vs PEFT adapter API docs; separate documentation features." + }, + { + "left": "pull_request:4275", + "right": "pull_request:6410", + "accept": false, + "reason": "AutoPipeline docs correction vs RL register_modules refactor; unrelated." + }, + { + "left": "pull_request:4276", + "right": "pull_request:5792", + "accept": false, + "reason": "Pipeline import test vs import-path fixes in examples; test coverage change is not the same as the example import cleanup." } ] }, @@ -155106,7 +155209,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 163, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -155114,9 +155217,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:4273", - "pull_request:4275", - "pull_request:4276", "pull_request:4301", "pull_request:4320", "pull_request:433", @@ -155142,10 +155242,14 @@ "pull_request:5622", "pull_request:5624", "pull_request:5682", + "pull_request:5697", + "pull_request:5698", + "pull_request:5699", + "pull_request:5707", "pull_request:5729", - "pull_request:5792", "pull_request:5916", "pull_request:5982", + "pull_request:6053", "pull_request:6065", "pull_request:6237", "pull_request:6262", @@ -155153,19 +155257,16 @@ "pull_request:6410", "pull_request:6432", "pull_request:6437", - "pull_request:6499", "pull_request:6775", "pull_request:6886", "pull_request:70", "pull_request:7134", + "pull_request:7200", "pull_request:7510", "pull_request:7650", "pull_request:788" ], "soft_pairs": [ - "pull_request:4273|pull_request:6499", - "pull_request:4275|pull_request:6410", - "pull_request:4276|pull_request:5792", "pull_request:4301|pull_request:4887", "pull_request:4320|pull_request:6410", "pull_request:433|pull_request:70", @@ -155188,192 +155289,195 @@ "pull_request:5552|pull_request:7650", "pull_request:5602|pull_request:6065", "pull_request:5622|pull_request:7510", - "pull_request:5624|pull_request:5916" + "pull_request:5624|pull_request:5916", + "pull_request:5697|pull_request:7200", + "pull_request:5698|pull_request:5699", + "pull_request:5707|pull_request:6053" ], - "prepared_review_unit_hash": "d1ba7774d1af1d3396958592a6444e768522461fb0b22d2b62737190c80b9a08", + "prepared_review_unit_hash": "dd73365133cd72f28b61b9038aabf5f1d224e37637aa6ecc3de39a02bde7c78c", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 101337, - "estimated_input_tokens": 25335, - "estimated_eval_tokens": 50926 + "serialized_chars": 104051, + "estimated_input_tokens": 26013, + "estimated_eval_tokens": 52282 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous: documentation updates, small maintenance fixes, and a few unrelated feature/code changes that only share superficial wording or subsystem tags. I do not see any true duplicate PR pairs here.", - "confidence": 0.19, + "summary": "This cluster is a grab bag of mostly unrelated PRs that were linked only by weak textual similarity. The soft edges mostly connect docs, training-script, scheduler, and model changes that do not fix the same underlying bug or implement the same change, so none should be merged as duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 5164 is the most substantial and heavily reviewed change in the set, so it is the best fallback representative, but it is not actually a duplicate of the others.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 5164 is the strongest standalone PR in the cluster: broad impact, large diff, and high review activity. The rest are mostly smaller, unrelated maintenance or doc changes.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:4273", - "right": "pull_request:6499", - "accept": false, - "reason": "AutoPipeline tutorial docs vs PEFT loader API docs/code; different scope and change." - }, - { - "left": "pull_request:4275", - "right": "pull_request:6410", - "accept": false, - "reason": "Autopipeline docs tweak vs RL value-guided sampling module registration fix; unrelated." - }, - { - "left": "pull_request:4276", - "right": "pull_request:5792", - "accept": false, - "reason": "Dependency test addition vs import-path cleanup in examples; not the same bug." - }, { "left": "pull_request:4301", "right": "pull_request:4887", "accept": false, - "reason": "Kandinsky documentation update vs removal of a latents argument from a pipeline API; different changes." + "reason": "Kandinsky docs cleanup vs Kandinsky prior_emb2emb API removal; one is documentation, the other changes pipeline behavior." }, { "left": "pull_request:4320", "right": "pull_request:6410", "accept": false, - "reason": "Typos in ONNX docs vs a runtime module refactor; no shared code path." + "reason": "Typo/doc wording fix vs RL value-guided sampling refactor; unrelated code paths and goals." }, { "left": "pull_request:433", "right": "pull_request:70", "accept": false, - "reason": "Black version bump/formatting cleanup vs an early attention-layer refactor; unrelated maintenance." + "reason": "Black version bump and formatting cleanup vs an early attention/unet refactor; no shared bug or feature." }, { "left": "pull_request:4396", "right": "pull_request:6262", "accept": false, - "reason": "Prompt-encoding fallback fix across pipelines vs panorama refactor; different problems." + "reason": "Prompt-encoding fallback fix vs moving panorama out of stable_diffusion; unrelated pipeline changes." }, { "left": "pull_request:4472", "right": "pull_request:5729", "accept": false, - "reason": "DreamBooth README edit vs version-gating update across many training scripts; not the same change." + "reason": "DreamBooth README edit vs broad setup/version updates in many scripts; not the same concrete change." }, { "left": "pull_request:4547", "right": "pull_request:4746", "accept": false, - "reason": "Flax scheduler export work vs SDXL inpainting WIP; unrelated features." + "reason": "Flax scheduler WIP vs SDXL inpainting work; different features and code paths." }, { "left": "pull_request:4547", "right": "pull_request:5682", "accept": false, - "reason": "Flax scheduler addition vs AnimateDiff checks; different code paths and goals." + "reason": "Flax scheduler WIP vs AnimateDiff checks; unrelated implementations." }, { "left": "pull_request:4748", "right": "pull_request:5729", "accept": false, - "reason": "setup.py/upload instructions cleanup vs min-version bumps in training scripts; not duplicate." + "reason": "Packaging/release instruction cleanup vs example script minimum-version updates; only loosely related to setup/versioning." }, { "left": "pull_request:4783", "right": "pull_request:6237", "accept": false, - "reason": "SDXL ControlNet doc args vs batched-seed warning docs; unrelated documentation topics." + "reason": "SDXL embedding-force docs vs batched seed documentation; different user-facing topics." }, { "left": "pull_request:4787", "right": "pull_request:6437", "accept": false, - "reason": "LoRA loading filename filter logic vs CI test exclusion tweak; different fixes." + "reason": "LoRA weight-file selection logic vs CI test exclusion for PEFT env; same broad area, but not the same bug or mergeable fix." }, { "left": "pull_request:4887", "right": "pull_request:6886", "accept": false, - "reason": "Kandinsky pipeline API removal vs deprecation cleanup for to(); no common bug." + "reason": "Kandinsky pipeline API change vs removing torch_dtype deprecation usage; unrelated changes." }, { "left": "pull_request:4937", "right": "pull_request:6305", "accept": false, - "reason": "T2I-Adapter conditioning factor vs DDPMScheduler zero-SNR support; different features." + "reason": "Adapter conditioning-factor addition vs DDPM beta rescaling; different scheduler/model functionality." }, { "left": "pull_request:51", "right": "pull_request:788", "accept": false, - "reason": "Old packaging fix vs PIL sampler warning fix in examples; unrelated." + "reason": "install_requires pin fix vs PIL sampler warning fix; no common underlying issue." }, { "left": "pull_request:5122", "right": "pull_request:5164", "accept": false, - "reason": "Workflow/python-version and setup updates vs FreeU algorithm feature; not the same issue." + "reason": "CI Python version bump vs adding FreeU mechanism; unrelated maintenance vs new model feature." }, { "left": "pull_request:5267", "right": "pull_request:5729", "accept": false, - "reason": "README markdown formatting vs version check bumps; only superficial repository overlap." + "reason": "Markdown code-fence cleanup in docs vs broad example/setup version changes; not the same artifact or bug." }, { "left": "pull_request:5312", "right": "pull_request:7134", "accept": false, - "reason": "General model docs/typehints cleanup vs specific UNet type-hint docs; related area but not the same change." + "reason": "Broad model docs/type-hint cleanup vs specific UNet type-hint adjustments; same general theme, but not the same concrete change." }, { "left": "pull_request:5369", "right": "pull_request:6775", "accept": false, - "reason": "Minor docs fixes vs missing-parameter docs for DDPMScheduler; different documentation targets." + "reason": "Minor docs fixes in img2img/inpaint vs adding a missing DDPM parameter doc; different documentation edits." }, { "left": "pull_request:5393", "right": "pull_request:6432", "accept": false, - "reason": "Training tutorial API modernization vs SDXL DreamBooth learning-rate tweak; unrelated docs/examples." + "reason": "HFApi tutorial migration vs DreamBooth learning-rate tweak; unrelated example updates." }, { "left": "pull_request:5478", "right": "pull_request:5982", "accept": false, - "reason": "Japanese docs rollout vs SDXL Turbo docs feature addition; different deliverables." + "reason": "Japanese docs rollout vs SDXL Turbo docs; different doc scopes and features." }, { "left": "pull_request:5552", "right": "pull_request:7650", "accept": false, - "reason": "FreeU disable-method bugfix vs new `is_cosxl_edit` argument; not the same underlying bug." + "reason": "FreeU disable-method bugfix vs SDXL InstructPix2Pix arg addition; unrelated pipeline/model changes." }, { "left": "pull_request:5602", "right": "pull_request:6065", "accept": false, - "reason": "Whitespace/docs cleanup vs SDXL Turbo documentation; unrelated." + "reason": "Whitespace cleanup in docs vs SDXL Turbo docs; not the same change." }, { "left": "pull_request:5622", "right": "pull_request:7510", "accept": false, - "reason": "Large examples/README and script cleanup vs VAE tiling OOM test fix; different issues." + "reason": "Example training script/README and ignore-file cleanup vs VAE tiling OOM test fix; unrelated." }, { "left": "pull_request:5624", "right": "pull_request:5916", "accept": false, - "reason": "Large examples workflow/README changes vs DreamBooth output_dir rename; not duplicate." + "reason": "Both touch DreamBooth, but one is a broad argument/README update and the other only renames the output_dir default; not the same underlying fix." + }, + { + "left": "pull_request:5697", + "right": "pull_request:7200", + "accept": false, + "reason": "PEFT LoRA+xformers bugfix vs Transformer2D validation/type behavior; different code paths." + }, + { + "left": "pull_request:5698", + "right": "pull_request:5699", + "accept": false, + "reason": "LCM img2img functional fix vs stable unCLIP styling cleanup; unrelated despite adjacency." + }, + { + "left": "pull_request:5707", + "right": "pull_request:6053", + "accept": false, + "reason": "LCM tests vs logging step_loss in text-to-image training; different purposes and files." } ] }, @@ -155389,7 +155493,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 164, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -155397,10 +155501,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:5697", - "pull_request:5698", - "pull_request:5699", - "pull_request:5707", "pull_request:5727", "pull_request:5729", "pull_request:5833", @@ -155409,7 +155509,6 @@ "pull_request:5937", "pull_request:594", "pull_request:6051", - "pull_request:6053", "pull_request:6114", "pull_request:6192", "pull_request:6205", @@ -155433,23 +155532,25 @@ "pull_request:7077", "pull_request:7115", "pull_request:7118", + "pull_request:7153", "pull_request:7181", - "pull_request:7200", "pull_request:7277", "pull_request:7339", "pull_request:7455", "pull_request:7461", + "pull_request:748", "pull_request:7627", + "pull_request:7650", + "pull_request:816", "pull_request:825", "pull_request:8290", "pull_request:8378", "pull_request:8412", - "pull_request:8652" + "pull_request:8652", + "pull_request:8722", + "pull_request:9247" ], "soft_pairs": [ - "pull_request:5697|pull_request:7200", - "pull_request:5698|pull_request:5699", - "pull_request:5707|pull_request:6053", "pull_request:5727|pull_request:6051", "pull_request:5729|pull_request:6861", "pull_request:5833|pull_request:7461", @@ -155474,211 +155575,254 @@ "pull_request:6759|pull_request:8378", "pull_request:7050|pull_request:7277", "pull_request:7115|pull_request:7455", - "pull_request:7118|pull_request:7277" + "pull_request:7118|pull_request:7277", + "pull_request:7153|pull_request:9247", + "pull_request:748|pull_request:816", + "pull_request:7627|pull_request:7650", + "pull_request:7627|pull_request:8722" ], - "prepared_review_unit_hash": "d5b0f37d005b6591eb60d25cd17ae390d042457257becb0dc9870a20f93731d8", + "prepared_review_unit_hash": "5de4d9545da3b0072a5c79ea547f3cad42e6f7b127d0a0da4b7225cd41d85d01", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 28, - "serialized_chars": 105867, - "estimated_input_tokens": 26467, - "estimated_eval_tokens": 53190 + "soft_pair_count": 29, + "serialized_chars": 104789, + "estimated_input_tokens": 26198, + "estimated_eval_tokens": 52652 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster does not contain a true duplicate set; the soft pairs mostly link unrelated docs, refactors, or sequential follow-up PRs that happen to touch similar files or topics. I would reject all soft edges and treat this as a noisy similarity cluster rather than a mergeable duplicate group.", - "confidence": 0.93, + "summary": "Mostly a heterogeneous mix of unrelated PRs (docs, CI, examples, and distinct model/pipeline fixes). The only true duplicate-looking subcluster is the repeated text-to-image training-script scheduler tweak around 6299/6300/6301/6322; everything else should stay separate.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR for the whole cluster; the only genuine duplicate subcluster is the repeated text-to-image training fix, where PR 6322 is the clearest representative.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 6322 is the best representative of the only real duplicate subcluster because it matches the same training-script code path and appears to be the most consolidated version of that change.", "soft_edge_verdicts": [ - { - "left": "pull_request:5697", - "right": "pull_request:7200", - "accept": false, - "reason": "Both touch attention-related code, but one is a PEFT/xFormers LoRA bugfix and the other is a Transformer2D validation/error-check change. Different concrete problems and code paths." - }, - { - "left": "pull_request:5698", - "right": "pull_request:5699", - "accept": false, - "reason": "LCM img2img bugfix versus unrelated stable unclip styling cleanup. Shared file overlap is incidental, not the same fix." - }, - { - "left": "pull_request:5707", - "right": "pull_request:6053", - "accept": false, - "reason": "5707 adds LCM tests across many pipelines; 6053 only adds step_loss logging in the text-to-image trainer. Different change intent and scope." - }, { "left": "pull_request:5727", "right": "pull_request:6051", "accept": false, - "reason": "LCM script addition versus Flax LCM scheduler integration. Related ecosystem, but not the same concrete change or mergeable into one PR." + "reason": "Both are LCM-related, but one adds example scripts while the other adds Flax scheduler support; different changes and code paths." }, { "left": "pull_request:5729", "right": "pull_request:6861", "accept": false, - "reason": "One updates example training setup/version pins; the other changes SDXL ControlNet cfg_end handling across multiple pipelines. No duplicate bug/change." + "reason": "Both touch training/pipeline setup, but one is a version-check update across examples and the other adds cfg_end support in SDXL pipelines." }, { "left": "pull_request:5833", "right": "pull_request:7461", "accept": false, - "reason": "Docs for PixArt low-VRAM usage and a testing-utils boolean helper relocation are unrelated." + "reason": "Docs for PixArt low-VRAM usage vs a testing-utils helper fix; unrelated artifacts." }, { "left": "pull_request:5834", "right": "pull_request:7461", "accept": false, - "reason": "Single-file ControlNet import fix versus testing-utils boolean helper relocation. No shared bug or code path." + "reason": "Single-file loading import path fix vs textual inversion boolean helper; not the same bug or change." }, { "left": "pull_request:5899", "right": "pull_request:5937", "accept": false, - "reason": "Kandinsky3 model/pipeline addition versus docs for Kandinsky3. The latter documents the former but is not the same underlying change." + "reason": "One adds Kandinsky3 code, the other only documents it; docs follow-up, not a duplicate code change." }, { "left": "pull_request:594", "right": "pull_request:825", "accept": false, - "reason": "Both involve Flax utilities, but 594 fixes Flax UNet/DDIM scheduler behavior while 825 removes the safety checker dummy. Different bugs and code paths." + "reason": "Both are Flax maintenance, but one fixes UNet/DDIM embedding math and the other removes Flax safety-checker dummy handling." }, { "left": "pull_request:6114", "right": "pull_request:7627", "accept": false, - "reason": "LPW XL img2img/inpaint support versus Playground img2img latent handling. Both are pipeline feature/bug work but not the same concrete issue." + "reason": "Both involve img2img/inpaint, but they target different pipelines and distinct bugs." }, { "left": "pull_request:6192", "right": "pull_request:6800", "accept": false, - "reason": "IP-Adapter unload support versus adding is_torchvision_available utility. No overlap beyond general loader/utilities area." + "reason": "IP-Adapter unload support vs a torchvision availability helper; unrelated changes." }, { "left": "pull_request:6205", "right": "pull_request:7077", "accept": false, - "reason": "Doc URL typo fix versus a code fix for head_to_batch_dim in IPAdapterAttnProcessor. Not the same bug/change." + "reason": "Docs URL typo vs attention tensor reshaping bug; unrelated." }, { "left": "pull_request:6205", "right": "pull_request:7181", "accept": false, - "reason": "Doc URL typo fix versus broad typo/fix-up PR touching many unrelated files. Similarity is only superficial." + "reason": "Both are small fixes, but one is a blog-link correction and the other is a broad typo sweep across multiple files." }, { "left": "pull_request:6262", "right": "pull_request:8290", "accept": false, - "reason": "6262 refactors panorama out of stable_diffusion; 8290 is a large HunyuanDiT refactor. Shared package structure is not enough." + "reason": "Panorama refactor vs HunyuanDIT refactor; different modules and changes." }, { "left": "pull_request:6263", "right": "pull_request:8290", "accept": false, - "reason": "6263 refactors LDM3D out of stable_diffusion; 8290 is unrelated HunyuanDIT refactoring and docs/build changes." + "reason": "LDM3D refactor vs HunyuanDIT refactor; different code paths and goals." }, { "left": "pull_request:6296", "right": "pull_request:6343", "accept": false, - "reason": "Both modify amused docs, but one adds the base pipeline docs and the other adds other pipeline docs. Could be related documentation follow-ups, but not the same concrete change." + "reason": "Initial aMUSEd docs/implementation PR vs later docs expansion for extra pipelines; follow-up documentation, not duplicate change." }, { "left": "pull_request:6296", "right": "pull_request:6453", "accept": false, - "reason": "Base amused docs versus a paper-link edit. The latter is a small doc tweak, not the same underlying addition." + "reason": "Initial aMUSEd docs/implementation PR vs paper-link doc tweak; not the same underlying change." }, { "left": "pull_request:6299", "right": "pull_request:6322", - "accept": false, - "reason": "Both change text-to-image training scripts, but 6299 alters validation scheduler choice while 6322 is about attn-train behavior. Different fixes." + "accept": true, + "reason": "Same text-to-image training-script scheduler adjustment in the same files; plausibly one merged PR." }, { "left": "pull_request:6300", "right": "pull_request:6322", - "accept": false, - "reason": "Same as above: shared file paths, but different functional changes and failure modes." + "accept": true, + "reason": "Same training-script change and file set as the other text-to-image validation scheduler fix." }, { "left": "pull_request:6301", "right": "pull_request:6322", - "accept": false, - "reason": "These are separate training-script adjustments, not one duplicate bug fix." + "accept": true, + "reason": "Same code-path and diff theme as the other text-to-image training updates; looks duplicate." }, { "left": "pull_request:6308", "right": "pull_request:7339", "accept": false, - "reason": "LoRA PEFT dummy components test fix versus release workflow dependency fix. No shared code path." + "reason": "LoRA test fixture fix vs release-workflow dependency changes; unrelated." }, { "left": "pull_request:6311", "right": "pull_request:8652", "accept": false, - "reason": "6311 adds fast diffusion tutorial docs; 8652 adds Lumina-T2X model/pipeline/scheduler support. Different artifacts and purpose." + "reason": "A docs/tutorial addition vs a major model/pipeline addition; only share generic documentation tree edits." }, { "left": "pull_request:6330", "right": "pull_request:8412", "accept": false, - "reason": "SVD video rendering docs fix versus single-file loading docs rewrite. Both are documentation, but unrelated topics." + "reason": "SVD docs rendering fix vs single-file usage docs rewrite; different documentation topics." }, { "left": "pull_request:6656", "right": "pull_request:6800", "accept": false, - "reason": "Textual inversion unload method versus is_torchvision_available utility exposure. Unrelated changes." + "reason": "Textual inversion unload API vs torchvision availability helper; unrelated." }, { "left": "pull_request:6759", "right": "pull_request:8378", "accept": false, - "reason": "InstantID prompt-image batch handling versus AsymmetricAutoencoderKL forward bugfix. Distinct code paths and symptoms." + "reason": "InstantID example fix vs asymmetric autoencoder forward fix; different areas and bugs." }, { "left": "pull_request:7050", "right": "pull_request:7277", "accept": false, - "reason": "Typo fixes in text-to-image examples versus broad deprecated-parameter/test cleanup. Not the same fix." + "reason": "Both mention typos, but one is text_to_image examples and the other is a broader tests/docs cleanup with a deprecated-parameter update." }, { "left": "pull_request:7115", "right": "pull_request:7455", "accept": false, - "reason": "Dockerfile dependency change versus removing distutils from setup/testing utils. Different maintenance tasks." + "reason": "Dockerfile packaging change vs distutils removal across setup/testing utilities; unrelated." }, { "left": "pull_request:7118", "right": "pull_request:7277", "accept": false, - "reason": "Docs typo cleanup versus test/deprecated-parameter cleanup. Similar hygiene theme, but not duplicate changes." + "reason": "Docs typo cleanup vs tests/deprecated-parameter update; not the same change." + }, + { + "left": "pull_request:7153", + "right": "pull_request:9247", + "accept": false, + "reason": "Documentation workflow path filters vs pipeline-loading variant detection fix; unrelated." + }, + { + "left": "pull_request:748", + "right": "pull_request:816", + "accept": false, + "reason": "Push-to-hub API cleanup vs mixed-precision/deepspeed training loss fix; different problems." + }, + { + "left": "pull_request:7627", + "right": "pull_request:7650", + "accept": false, + "reason": "Both are SDXL-related, but one fixes img2img latents handling and the other adds an is_cosxl_edit argument for instruct-pix2pix." + }, + { + "left": "pull_request:825", + "right": "pull_request:8722", + "accept": false, + "reason": "Flax safety-checker cleanup vs a large embeddings refactor; different code paths." + }, + { + "left": "pull_request:8290", + "right": "pull_request:8722", + "accept": false, + "reason": "HunyuanDIT refactor vs embeddings refactor; both broad refactors but not the same underlying change." + }, + { + "left": "pull_request:8378", + "right": "pull_request:8722", + "accept": false, + "reason": "AsymmetricAutoencoderKL forward bugfix vs embeddings refactor; unrelated." + }, + { + "left": "pull_request:8412", + "right": "pull_request:8652", + "accept": false, + "reason": "Single-file docs vs Lumina model/pipeline addition; different artifacts and scope." + }, + { + "left": "pull_request:8652", + "right": "pull_request:8722", + "accept": false, + "reason": "Lumina-T2X feature addition vs embeddings refactor; not the same code-path problem." + }, + { + "left": "pull_request:8722", + "right": "pull_request:9247", + "accept": false, + "reason": "Large embeddings refactor vs a small model-variant discovery bugfix; unrelated." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: most pairs are clearly unrelated, and the only plausible duplicate subcluster is the repeated text-to-image training-script scheduler fix around 6299/6300/6301/6322. The conservative rejections look appropriate, and the single accepted subcluster is supported by the identical file set and diff theme." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -155687,7 +155831,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 165, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -155701,19 +155845,20 @@ "pull_request:12419", "pull_request:12439", "pull_request:12549", + "pull_request:2421", + "pull_request:3034", "pull_request:4597", + "pull_request:4799", + "pull_request:5127", + "pull_request:5653", "pull_request:5887", "pull_request:6003", + "pull_request:6975", "pull_request:6999", - "pull_request:7153", - "pull_request:748", - "pull_request:7627", "pull_request:7637", - "pull_request:7650", "pull_request:7717", "pull_request:7893", "pull_request:7977", - "pull_request:816", "pull_request:8308", "pull_request:8310", "pull_request:840", @@ -155722,7 +155867,6 @@ "pull_request:8481", "pull_request:8660", "pull_request:8719", - "pull_request:8722", "pull_request:8761", "pull_request:8978", "pull_request:8984", @@ -155730,7 +155874,6 @@ "pull_request:9060", "pull_request:909", "pull_request:91", - "pull_request:9247", "pull_request:9269", "pull_request:9289", "pull_request:9307", @@ -155745,10 +155888,6 @@ "pull_request:9994" ], "soft_pairs": [ - "pull_request:7153|pull_request:9247", - "pull_request:748|pull_request:816", - "pull_request:7627|pull_request:7650", - "pull_request:7627|pull_request:8722", "pull_request:7637|pull_request:7717", "pull_request:7893|pull_request:9307", "pull_request:7977|pull_request:9739", @@ -155771,198 +155910,195 @@ "pull_request:10396|pull_request:10578", "pull_request:12439|pull_request:12549", "pull_request:12086|pull_request:12419", - "pull_request:6003|pull_request:6999" + "pull_request:6003|pull_request:6999", + "pull_request:4799|pull_request:6975", + "pull_request:5127|pull_request:5653", + "pull_request:2421|pull_request:3034" ], - "prepared_review_unit_hash": "5aa6417bcf941a36844a023662daa83d5d75940233b7733b30549dc0628f8fd5", + "prepared_review_unit_hash": "91ce99af5b432fff383e7ff40e70e7417aba370a68bc4ec3c77a1814b1f8858f", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 27, - "serialized_chars": 103921, - "estimated_input_tokens": 25981, - "estimated_eval_tokens": 52218 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 101541, + "estimated_input_tokens": 25386, + "estimated_eval_tokens": 51028 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated LoRA, ControlNet, Flux, Stable Diffusion, Hunyuan, CI, docs, and workflow changes. I do not see a single underlying bug/change that unifies the set, and none of the soft pairs look mergeable as duplicates.", + "summary": "This cluster is highly heterogeneous: LoRA, ControlNet, Flux, Hunyuan, SD3, CI, docs, and unrelated maintenance changes. The soft edges are mostly superficial overlaps in filenames or broad subsystem names, not true duplicate bugs or mergeable duplicate PRs.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR spans the cluster\u2019s themes; the changes are too unrelated to justify one canonical PR.", + "canonical_pr_reason": "No single canonical PR stands out; the items are unrelated changes spanning many subsystems, so there is no meaningful duplicate anchor.", "best_issue_reason": null, - "best_pr_reason": "If forced to pick one, pull_request:10578 is the most substantive core code fix, but it is not representative of the cluster as a whole.", + "best_pr_reason": "No globally best PR for deduplication: the set is too mixed, so any single PR would be arbitrary rather than representative of the cluster.", "soft_edge_verdicts": [ - { - "left": "pull_request:7153", - "right": "pull_request:9247", - "accept": false, - "reason": "Docs workflow path filtering and model-variant detection are unrelated changes." - }, - { - "left": "pull_request:748", - "right": "pull_request:816", - "accept": false, - "reason": "Different example-script fixes with different failure modes; not the same bug." - }, - { - "left": "pull_request:7627", - "right": "pull_request:7650", - "accept": false, - "reason": "Both touch SDXL img2img, but one fixes Playground latent scaling and the other adds a new `is_cosxl_edit` arg." - }, - { - "left": "pull_request:7627", - "right": "pull_request:8722", - "accept": false, - "reason": "Pipeline latent handling vs embeddings refactor; no shared underlying bug." - }, { "left": "pull_request:7637", "right": "pull_request:7717", "accept": false, - "reason": "Ruff config cleanup is unrelated to docstring style fixes." + "reason": "Different maintenance areas: Ruff config deprecation cleanup vs docstring style fixes." }, { "left": "pull_request:7893", "right": "pull_request:9307", "accept": false, - "reason": "Dockerfile Python upgrade and rotary embedding refactor are separate changes." + "reason": "Unrelated changes: Python 3.10 Dockerfile upgrade vs rotary embedding CPU/device refactor." }, { "left": "pull_request:7977", "right": "pull_request:9739", "accept": false, - "reason": "A typo fix in docs and SD3 doc/loader updates are not the same issue." + "reason": "One fixes a typo in docs; the other adds an SD3 single-file loading example. Not the same change." }, { "left": "pull_request:8308", "right": "pull_request:9289", "accept": false, - "reason": "Hunyuan model refactor and Animatediff model addition are different features." + "reason": "Different models/features: Hunyuan refactor vs AnimateDiff model work." }, { "left": "pull_request:8310", "right": "pull_request:9289", "accept": false, - "reason": "Same as above; unrelated model refactors." + "reason": "HunyuanDiT refactor and AnimateDiff model work are unrelated code paths." }, { "left": "pull_request:840", "right": "pull_request:9", "accept": false, - "reason": "Community example addition versus Grad TTS model/pipeline implementation." + "reason": "One-step U-Net community example vs Grad TTS model/pipeline addition are unrelated." }, { "left": "pull_request:8403", "right": "pull_request:8978", "accept": false, - "reason": "Tailscale action update and DreamBooth/Imagic content update are unrelated." + "reason": "Updating a Tailscale action is unrelated to DreamBooth/imagic content changes." }, { "left": "pull_request:8479", "right": "pull_request:9620", "accept": false, - "reason": "Hunyuan ControlNet support and ControlNet interrupt support fix different code paths." + "reason": "Both involve ControlNet, but one adds Hunyuan ControlNet support and the other adds interrupt handling across ControlNet pipelines." }, { "left": "pull_request:8481", "right": "pull_request:9739", "accept": false, - "reason": "ImageProcessor error-message tweak and SD3 doc/loader work are unrelated." + "reason": "Image input error-message tweak vs SD3 docs/typo fixes are unrelated." }, { "left": "pull_request:8660", "right": "pull_request:9426", "accept": false, - "reason": "SD3 ControlNet input validation and CI artifact uploader version are unrelated." + "reason": "SD3 ControlNet validation change vs CI artifact uploader upgrade are different concerns." }, { "left": "pull_request:8719", "right": "pull_request:9676", "accept": false, - "reason": "VQModel deprecation cleanup and benchmark import fix are unrelated maintenance changes." + "reason": "VQModelOutput deprecation/docs cleanup vs benchmark import fix; no shared bug or change." }, { "left": "pull_request:8761", "right": "pull_request:9269", "accept": false, - "reason": "Both edit embeddings, but they address different embedding classes and different bugs." + "reason": "Shared embeddings.py file, but these patch different functions for different models (Hunyuan embedding refactor vs CogVideoX 3D rope refactor)." }, { "left": "pull_request:8978", "right": "pull_request:9734", "accept": false, - "reason": "Notebook/content update versus SD3 transformer/doc changes are not the same change." + "reason": "DreamBooth/imagic update and SD3 doc/test update are unrelated documentation/example changes." }, { "left": "pull_request:8984", "right": "pull_request:9620", "accept": false, - "reason": "SD3 ONNX support and ControlNet interrupt are separate features." + "reason": "ONNX conversion/pipeline support for SD3 is unrelated to ControlNet interrupt handling." }, { "left": "pull_request:9060", "right": "pull_request:9938", "accept": false, - "reason": "Removing a debug print is unrelated to updating the SDXL reference pipeline." + "reason": "Removing a debug print in Flux is unrelated to SDXL reference pipeline updates." }, { "left": "pull_request:909", "right": "pull_request:91", "accept": false, - "reason": "Test fixture updates and scheduler API refactor are different changes." + "reason": "Different scopes and eras: test model ID cleanup vs a broad scheduler API finish." }, { "left": "pull_request:9483", "right": "pull_request:9699", "accept": false, - "reason": "Nightly test expectation updates and runner selection changes do not fix the same bug." + "reason": "Nightly test numeric expectation updates vs adding a new SSH runner are unrelated CI changes." }, { "left": "pull_request:9620", "right": "pull_request:9994", "accept": false, - "reason": "ControlNet interrupt support and Omnigen model/pipeline support are unrelated." + "reason": "ControlNet interrupt support and Omnigen model/pipeline support are unrelated features." }, { "left": "pull_request:4597", "right": "pull_request:5887", "accept": false, - "reason": "Both are ControlNet-related, but one adds MultiControlNet SDXL support and the other adds IP-Adapter support to inpaint; distinct features." + "reason": "Both touch ControlNet, but one adds MultiControlNet SDXL support and the other adds IP-Adapter support for inpaint; distinct PRs." }, { "left": "pull_request:10396", "right": "pull_request:10578", "accept": false, - "reason": "Both touch LoRA loading, but one fixes missing-state indexing and the other handles 4bit quantized Flux shape calculation; different bugs." + "reason": "Both are LoRA-related, but one handles missing base layers in LoRA expansion while the other supports 4-bit quantized Flux LoRA loading." }, { "left": "pull_request:12439", "right": "pull_request:12549", "accept": false, - "reason": "Both concern attention backends, but they add different backends and code paths." + "reason": "Both concern attention backends, but one adds kernels-based Sage/FA2 support while the other adds AITER backend support." }, { "left": "pull_request:12086", "right": "pull_request:12419", "accept": false, - "reason": "Both target Flux modular pipelines, but one adds I2I support while the other tweaks Mellon-compatible input typing." + "reason": "Both touch Flux modular pipelines, but one adds Flux I2I support and the other adds Mellon-specific kwargs typing." }, { "left": "pull_request:6003", "right": "pull_request:6999", "accept": false, - "reason": "Same Stable Video Diffusion pipeline, but one adds inpaint support and the other fixes tensor-image encoding behavior." + "reason": "Both are Stable Video Diffusion, but one adds inpaint support and the other fixes tensor image input handling." + }, + { + "left": "pull_request:4799", + "right": "pull_request:6975", + "accept": false, + "reason": "SDXL ControlNet guess_mode bugfix vs ControlNet inpaint single-file conversion fix are separate issues." + }, + { + "left": "pull_request:5127", + "right": "pull_request:5653", + "accept": false, + "reason": "Multi-adapter SDXL support and PEFT model unwrapping are unrelated changes." + }, + { + "left": "pull_request:2421", + "right": "pull_request:3034", + "accept": false, + "reason": "Deprecation call cleanup in cross_attention is unrelated to fixing an accelerate parameter name." } ] }, @@ -155978,7 +156114,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 166, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -156013,31 +156149,29 @@ "pull_request:2129", "pull_request:2200", "pull_request:2393", - "pull_request:2421", + "pull_request:2584", "pull_request:2721", "pull_request:2928", - "pull_request:3034", "pull_request:3042", "pull_request:3128", + "pull_request:3590", + "pull_request:3746", "pull_request:4062", + "pull_request:4214", + "pull_request:4659", + "pull_request:4661", "pull_request:4715", - "pull_request:4799", "pull_request:5117", - "pull_request:5127", - "pull_request:5653", "pull_request:5833", "pull_request:5834", + "pull_request:6009", "pull_request:6133", "pull_request:6447", "pull_request:6650", "pull_request:689", - "pull_request:6975", "pull_request:9918" ], "soft_pairs": [ - "pull_request:4799|pull_request:6975", - "pull_request:5127|pull_request:5653", - "pull_request:2421|pull_request:3034", "pull_request:12074|pull_request:12679", "pull_request:4062|pull_request:6447", "pull_request:10578|pull_request:11130", @@ -156058,180 +156192,190 @@ "pull_request:11814|pull_request:11985", "pull_request:13029|pull_request:13100", "pull_request:1513|pull_request:3042", - "pull_request:1554|pull_request:3128" + "pull_request:1554|pull_request:3128", + "pull_request:2584|pull_request:3590", + "pull_request:3746|pull_request:4659", + "pull_request:3746|pull_request:4661", + "pull_request:4214|pull_request:6009" ], - "prepared_review_unit_hash": "99f63f205c82362dd8885cd7ae8a3fd7cdbce3b8355809b2175bd2370afeed7e", + "prepared_review_unit_hash": "a5abc4e5a115cb3bc398c447a78bbf91538a9f4499995bbf4f7205f60f31aa50", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 102638, - "estimated_input_tokens": 25660, - "estimated_eval_tokens": 51576 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 106313, + "estimated_input_tokens": 26579, + "estimated_eval_tokens": 53414 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose similarity set of unrelated docs, feature, and bugfix PRs. I don\u2019t see any pair that is clearly the same underlying change, so all soft edges should be rejected.", - "confidence": 0.84, + "summary": "This cluster is heterogeneous: docs tweaks, example additions, model/pipeline features, and unrelated bug fixes. None of the suggested soft pairs look like the same underlying change or bug.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a clear duplicate anchor for the whole cluster; if forced to pick a representative, PR #10136 is the strongest central core change because it is a substantial Hunyuan Video implementation PR with broad downstream references and related follow-up work.", + "canonical_pr_reason": "No canonical PR fits well; the cluster spans unrelated features, docs, and fixes rather than one duplicated change.", "best_issue_reason": null, - "best_pr_reason": "PR #10136 is the best overall representative PR in this set: it is a substantial, well-scoped core feature PR with the most evidence of being a foundational change rather than a minor doc tweak.", + "best_pr_reason": "No single PR is a strong representative for duplicate triage because there is no shared concrete code-path or bug across the items.", "soft_edge_verdicts": [ - { - "left": "pull_request:4799", - "right": "pull_request:6975", - "accept": false, - "reason": "Both touch ControlNet-related code, but one fixes SDXL guess_mode behavior while the other fixes single-file ControlNet inpaint loading; different bugs and code paths." - }, - { - "left": "pull_request:5127", - "right": "pull_request:5653", - "accept": false, - "reason": "Multi-adapter support in SDXL Adapter and PEFT-aware model unwrapping are separate features affecting different subsystems." - }, - { - "left": "pull_request:2421", - "right": "pull_request:3034", - "accept": false, - "reason": "One is a small deprecate call cleanup, the other renames/uses an accelerate CLI flag correctly; unrelated changes." - }, { "left": "pull_request:12074", "right": "pull_request:12679", "accept": false, - "reason": "Both mention multiple LoRAs, but they target different models and workflows (Wan 2.2 transformer loading vs ChronoEdit LoRA support)." + "reason": "Both involve LoRA support, but one is Wan 2.2 second-transformer loading and the other is ChronoEdit multi-LoRA docs/loaders; different models and changes." }, { "left": "pull_request:4062", "right": "pull_request:6447", "accept": false, - "reason": "Shap-E mesh output and Stable Diffusion K-Diffusion are unrelated pipeline additions." + "reason": "Shap-E mesh output support is unrelated to adding K-Diffusion pipelines." }, { "left": "pull_request:10578", "right": "pull_request:11130", "accept": false, - "reason": "Both are quantization-related, but one fixes LoRA loading into 4-bit Flux models while the other introduces pipeline-level quantization config; not the same concrete change." + "reason": "4bit Flux LoRA loading and pipeline-level quantization config are different feature sets, not the same fix." }, { "left": "pull_request:4715", "right": "pull_request:5833", "accept": false, - "reason": "Renaming community test files is unrelated to PixArt docs/import guidance." + "reason": "File renames for community tests vs fixing imports in single-file loading are unrelated changes." }, { "left": "pull_request:4715", "right": "pull_request:5834", "accept": false, - "reason": "Renaming community test files is unrelated to single-file ControlNet import fixes." + "reason": "File renames for community tests vs fixing imports in single-file loading are unrelated changes." }, { "left": "pull_request:10482", "right": "pull_request:10567", "accept": false, - "reason": "The NaN fix and true-CFG enablement both touch HunyuanVideo, but they solve different problems in different code paths." + "reason": "NaN fix in HunyuanVideo attention masking is unrelated to the true-cfg feature work." }, { "left": "pull_request:2129", "right": "pull_request:2721", "accept": false, - "reason": "Pipeline LoRA support and the ModelEditing pipeline are distinct features." + "reason": "Allowing LoRA from pipeline and adding ModelEditing pipeline are different capabilities." }, { "left": "pull_request:2200", "right": "pull_request:2393", "accept": false, - "reason": "These are related panorama-style pipelines, but they implement different pipeline variants and should not be merged as duplicates." + "reason": "Both are panorama-related, but one is a community multi-diffusion pipeline and the other is an official stable-diffusion panorama pipeline; not the same concrete change." }, { "left": "pull_request:12519", "right": "pull_request:12530", "accept": false, - "reason": "One fixes a toctree sorting/CI issue; the other xfails a flaky test. Different maintenance tasks." + "reason": "Toctree sorting/quality fix and xfail-ing a flaky test solve different CI issues." }, { "left": "pull_request:1579", "right": "pull_request:2002", "accept": false, - "reason": "Intel-optimized examples and LoRA fine-tuning support are different example/documentation additions." + "reason": "Intel-optimized examples are unrelated to LoRA fine-tuning support." }, { "left": "pull_request:10136", "right": "pull_request:10837", "accept": false, - "reason": "SkyReels Hunyuan work builds on the HunyuanVideo stack, but it is a separate follow-on pipeline/example set, not the same PR change." + "reason": "Base Hunyuan Video support and SkyReels Hunyuan T2V/I2V are distinct model integrations." }, { "left": "pull_request:1942", "right": "pull_request:689", "accept": false, - "reason": "Upcast-attention flag handling in a conversion script is unrelated to an attention backend performance tweak." + "reason": "Fixing an upcast_attention flag in a conversion script is unrelated to the attention performance change." }, { "left": "pull_request:2928", "right": "pull_request:5117", "accept": false, - "reason": "Custom dataset loading for ControlNet training and training resume log behavior are unrelated fixes." + "reason": "Custom local dataset loading for ControlNet training and training resume log handling are unrelated." }, { "left": "pull_request:12416", "right": "pull_request:12445", "accept": false, - "reason": "QwenImage Edit Plus modular support and Flux modular alignment are separate modular-pipeline changes." + "reason": "QwenImage Edit Plus modular support and Flux modular alignment are different model families and codepaths." }, { "left": "pull_request:6133", "right": "pull_request:6650", "accept": false, - "reason": "The core UFOGen scheduler/pipeline addition and a community example for it are related, but not the same code change." + "reason": "UFOGen pipeline/scheduler implementation and a community example for the scheduler are not the same change." }, { "left": "pull_request:10008", "right": "pull_request:10435", "accept": false, - "reason": "Both are docs-only, but they fix different documentation issues in different files and models." + "reason": "Table formatting docs for CogVideoX and doc variable-name fixes are unrelated documentation edits." }, { "left": "pull_request:10719", "right": "pull_request:9918", "accept": false, - "reason": "Adding provider_options to from_pretrained is unrelated to adding a server example." + "reason": "Adding provider_options to pipeline loading and adding a server example are unrelated." }, { "left": "pull_request:11814", "right": "pull_request:11985", "accept": false, - "reason": "Gradient-checkpointing keyword passthrough in Flux is distinct from adding Flux Kontext support." + "reason": "Passing joint_attention_kwargs in gradient checkpointing and adding Flux Kontext are different features." }, { "left": "pull_request:13029", "right": "pull_request:13100", "accept": false, - "reason": "A new modular quickstart guide and a ModularPipeline docs update are different documentation efforts." + "reason": "Both are modular docs, but one is a quickstart guide and the other is a broader ModularPipeline doc update; not the same change." }, { "left": "pull_request:1513", "right": "pull_request:3042", "accept": false, - "reason": "write_basic_config docs for DreamBooth and accelerate minimum-version pinning solve different accelerate-related issues." + "reason": "Dreambooth notebook config guidance and accelerate minimum-version updates are different docs/dependency fixes." }, { "left": "pull_request:1554", "right": "pull_request:3128", "accept": false, - "reason": "Doc corrections for model shapes are unrelated to the img2img timestep bug fix." + "reason": "Docs corrections for UNet text shapes are unrelated to the img2img timestep bug fix." + }, + { + "left": "pull_request:2584", + "right": "pull_request:3590", + "accept": false, + "reason": "Community ControlNet img2img and Stable Diffusion Mixture Canvas are separate community pipelines." + }, + { + "left": "pull_request:3746", + "right": "pull_request:4659", + "accept": false, + "reason": "Path-join fix in from_ckpt is unrelated to the SDXL long weighted prompt pipeline/docs changes." + }, + { + "left": "pull_request:3746", + "right": "pull_request:4661", + "accept": false, + "reason": "Path-join fix in from_ckpt is unrelated to the SDXL long weighted prompt pipeline/docs changes." + }, + { + "left": "pull_request:4214", + "right": "pull_request:6009", + "accept": false, + "reason": "bf16 error fix in Custom Diffusion is unrelated to the TensorRT super-args error fix." } ] }, @@ -156247,7 +156391,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 167, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -156270,23 +156414,23 @@ "pull_request:1331", "pull_request:13352", "pull_request:2111", + "pull_request:2345", "pull_request:2467", - "pull_request:2584", + "pull_request:2693", + "pull_request:3278", "pull_request:3313", - "pull_request:3590", "pull_request:3607", - "pull_request:3746", - "pull_request:4214", + "pull_request:3691", + "pull_request:4272", + "pull_request:4301", "pull_request:4412", - "pull_request:4659", - "pull_request:4661", "pull_request:4695", + "pull_request:4715", "pull_request:4887", "pull_request:4987", "pull_request:5033", "pull_request:5299", "pull_request:5979", - "pull_request:6009", "pull_request:6085", "pull_request:6197", "pull_request:6261", @@ -156304,10 +156448,6 @@ "pull_request:9601" ], "soft_pairs": [ - "pull_request:2584|pull_request:3590", - "pull_request:3746|pull_request:4659", - "pull_request:3746|pull_request:4661", - "pull_request:4214|pull_request:6009", "pull_request:4412|pull_request:6085", "pull_request:4887|pull_request:4987", "pull_request:5299|pull_request:6658", @@ -156328,186 +156468,190 @@ "pull_request:12541|pull_request:13352", "pull_request:13100|pull_request:13120", "pull_request:1331|pull_request:2467", - "pull_request:2111|pull_request:3607" + "pull_request:2111|pull_request:3607", + "pull_request:2345|pull_request:3691", + "pull_request:2467|pull_request:4272", + "pull_request:2693|pull_request:4301", + "pull_request:3278|pull_request:4715" ], - "prepared_review_unit_hash": "b8738ce4bee3cc38547317258df66ab66b294d9c066a9b49c1437ce61f4b5d94", + "prepared_review_unit_hash": "629bb44bdb346522d2f2fa78877f410d82168d28f20620217fab3172dbdfbdf9", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 102259, - "estimated_input_tokens": 25565, - "estimated_eval_tokens": 51386 + "serialized_chars": 101102, + "estimated_input_tokens": 25276, + "estimated_eval_tokens": 50808 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose similarity group of mostly unrelated PRs across docs, LoRA, attention, pipelines, schedulers, and CI. The paired items only share broad subsystem names or one common file in a larger refactor, but they do not describe the same concrete bug or change.", - "confidence": 0.97, + "summary": "This set is not a duplicate cluster; it\u2019s a grab bag of mostly unrelated Diffusers PRs spanning docs, small bug fixes, loader/quantization work, and pipeline refactors. The soft pairs are mostly same-subsystem but different concrete changes, so they should be rejected.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:2584", - "right": "pull_request:3590", - "accept": false, - "reason": "Different community pipelines: one adds ControlNet img2img, the other adds Mixture Canvas. Shared theme is only community examples." - }, - { - "left": "pull_request:3746", - "right": "pull_request:4659", - "accept": false, - "reason": "Both touch path/doc loading areas, but one fixes ckpt path joining and the other removes a docs/tutorial page. No same bug or code path." - }, - { - "left": "pull_request:3746", - "right": "pull_request:4661", - "accept": false, - "reason": "Path-handling fix vs. SDXL long prompt community pipeline. Unrelated changes." - }, - { - "left": "pull_request:4214", - "right": "pull_request:6009", - "accept": false, - "reason": "Different fixes in different areas: bf16/custom diffusion attention vs. TensorRT constructor args. No shared defect." - }, { "left": "pull_request:4412", "right": "pull_request:6085", "accept": false, - "reason": "Both adjust testing/scheduler behavior, but they are not the same issue or implementation path." + "reason": "Both touch test/common utility code, but one is a float16 inference test fix and the other is a scheduler add_noise bug fix. Different code paths and different failures." }, { "left": "pull_request:4887", "right": "pull_request:4987", "accept": false, - "reason": "One removes a latent input from a pipeline; the other removes a duplicate import. Unrelated." + "reason": "These are unrelated model API cleanup changes: one removes a latent input from Kandinsky prior_emb2emb, the other removes a duplicate GLIGEN export." }, { "left": "pull_request:5299", "right": "pull_request:6658", "accept": false, - "reason": "Typehint/docs cleanup in models vs. adding a memory-footprint utility. Distinct changes." + "reason": "Both add utility methods, but one is about typehints/docs in models and the other introduces a memory-footprint helper for models/pipelines. Not the same underlying change." }, { "left": "pull_request:6261", "right": "pull_request:8290", "accept": false, - "reason": "Both affect docs/import structure, but one refactors Attend-and-Excite and the other is a large HunyuanDiT docs/workflow refactor. Not the same change." + "reason": "Both affect pipeline import/docs organization, but one refactors Attend-and-Excite out of stable_diffusion while the other is a broad HunyuanDiT/docs workflow refactor. No shared bug or change." }, { "left": "pull_request:6975", "right": "pull_request:8816", "accept": false, - "reason": "Single-file controlnet inpaint loader fix vs. PAG support for SD1.5 ControlNet. Different features and code paths." + "reason": "One fixes single-file controlnet inpaint loading; the other adds PAG support for SD1.5 ControlNet. Different features and code paths." }, { "left": "pull_request:6975", "right": "pull_request:8820", "accept": false, - "reason": "Single-file loader fix vs. PAG support docs/code for ControlNet. Not the same bug." + "reason": "Single-file controlnet inpaint loading fix versus PAG support docs/code for SD1.5 ControlNet. Same broad area, different concrete work." }, { "left": "pull_request:3313", "right": "pull_request:5033", "accept": false, - "reason": "Both touch inpainting, but one is torch.compile cleanup across many pipelines while the other fixes regular UNet inpaint logic. Too broad to merge as duplicates." + "reason": "Both involve inpainting, but one is a torch.compile compatibility sweep and the other fixes regular text2image UNet handling in inpaint. Not duplicative." }, { "left": "pull_request:11950", "right": "pull_request:12207", "accept": false, - "reason": "Both concern quantization/nunchaku, but one is initial SVDQuant support and the other is nunchaku integration. Different features." + "reason": "Both are quantization-related and touch loader/import plumbing, but one introduces SVDQuant support and the other incorporates Nunchaku. Different backends and functionality." }, { "left": "pull_request:4695", "right": "pull_request:6447", "accept": false, - "reason": "Both add pipeline variants, but one is TextToVideoZero SDXL and the other is K-Diffusion pipelines. Different pipeline families." + "reason": "Both add pipeline documentation and exports, but they introduce entirely different pipelines: TextToVideoZero SDXL vs StableDiffusionXL K-Diffusion." }, { "left": "pull_request:5979", "right": "pull_request:6650", "accept": false, - "reason": "DDGAN pipeline/scheduler addition vs. UFOGen community scheduler. Separate additions." + "reason": "Both add new pipeline/scheduler code, but DDGAN and UFOGen are distinct models and implementations. Not the same bug or change." }, { "left": "pull_request:6197", "right": "pull_request:7371", "accept": false, - "reason": "DeepCache community support vs. DoRA LoRA support. No shared concrete bug/change." + "reason": "DeepCache support for Stable Diffusion versus DoRA LoRA support. Different features, different subsystems." }, { "left": "pull_request:10058", "right": "pull_request:11583", "accept": false, - "reason": "Docs link fix vs. removal of fast diffusion tutorial. Both documentation-only, but not the same issue." + "reason": "Both are docs-only edits, but one fixes a training dataset link while the other removes a fast diffusion tutorial. No shared underlying issue." }, { "left": "pull_request:10114", "right": "pull_request:8452", "accept": false, - "reason": "Formatting-only pipeline cleanup vs. single-file loading strictness fix. Different scopes and problems." + "reason": "A formatting cleanup in stable audio versus a single-file loading error fix. Completely different problem statements." }, { "left": "pull_request:10114", "right": "pull_request:8510", "accept": false, - "reason": "Formatting cleanup vs. LR scheduler math fix for dreambooth LoRA. Unrelated." + "reason": "Formatting cleanup in a pipeline versus learning-rate scheduler math fixes in dreambooth LoRA training. Not the same code path." }, { "left": "pull_request:10122", "right": "pull_request:10941", "accept": false, - "reason": "Attention mask propagation in Flux attention vs. copy-fix in Flux pipelines. Similar subsystem, different change." + "reason": "Flux attention-mask plumbing in the attention processor is unrelated to flux pipeline copy/refactor cleanup." }, { "left": "pull_request:10126", "right": "pull_request:8691", "accept": false, - "reason": "LoRA deprecation/API cleanup vs. fixing vanilla fine-tuned LoRA loading. Distinct LoRA issues." + "reason": "One deprecates save_attn_procs and updates tests; the other fixes vanilla fine-tuned LoRA loading behavior. Different loader paths and goals." }, { "left": "pull_request:10544", "right": "pull_request:9601", "accept": false, - "reason": "Docs for ParaAttention optimization vs. SD3 training VAE dtype flag. Entirely different areas." + "reason": "Docs for ParaAttention optimization versus an SD3 training flag for VAE dtype. No duplicate bug or change." }, { "left": "pull_request:12541", "right": "pull_request:13352", "accept": false, - "reason": "LoRA conversion logic for Wan models vs. CI workflow checkout step. No common underlying bug." + "reason": "Wan LoRA conversion support and a CI workflow checkout step are unrelated changes." }, { "left": "pull_request:13100", "right": "pull_request:13120", "accept": false, - "reason": "ModularPipeline documentation update vs. tests for AutoModel load-id tagging. Related to modular pipeline work, but not the same change." + "reason": "ModularPipeline docs update versus a test for AutoModel load-id tagging. Related project area, but not the same issue or change." }, { "left": "pull_request:1331", "right": "pull_request:2467", "accept": false, - "reason": "Both mention inpainting, but one fixes scheduler support for inpaint pipelines and the other adds latent output type across many pipelines. Not the same concrete bug." + "reason": "Both mention output-type/inpainting-related code, but one is an inpainting scheduler/path fix and the other adds latent output type across many pipelines. Different feature scope." }, { "left": "pull_request:2111", "right": "pull_request:3607", "accept": false, - "reason": "Both involve attention internals, but one fuses attention masks and the other fixes temb attention/setup in a different code path." + "reason": "Both affect attention-related code, but one fuses attention masking in cross_attention while the other changes temb attention and adds docker/testing updates. Not the same concrete fix." + }, + { + "left": "pull_request:2345", + "right": "pull_request:3691", + "accept": false, + "reason": "Both are xFormers/PT2-era loader changes, but one is a warning about xformers 0.0.16 training issues and the other chooses LoRAAttnProcessor2_0 for text encoder LoRA. Different behavior." + }, + { + "left": "pull_request:2467", + "right": "pull_request:4272", + "accept": false, + "reason": "Latent output-type plumbing across pipelines versus missing optional imports in pipeline folders. Same repository area, not the same bug or change." + }, + { + "left": "pull_request:2693", + "right": "pull_request:4301", + "accept": false, + "reason": "Both are docs fixes, but one is an image link fix in inpaint docs and the other updates Kandinsky docs and pipeline links. No shared underlying defect." + }, + { + "left": "pull_request:3278", + "right": "pull_request:4715", + "accept": false, + "reason": "One temporarily disables spectrogram diffusion tests and dependency entries; the other renames community example test files to run scripts. Different maintenance tasks." } ] }, @@ -156523,7 +156667,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 168, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -156532,7 +156676,14 @@ "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10148", + "pull_request:10230", + "pull_request:10326", + "pull_request:10539", + "pull_request:10547", "pull_request:10850", + "pull_request:10966", + "pull_request:11269", + "pull_request:11479", "pull_request:11602", "pull_request:11916", "pull_request:12301", @@ -156540,19 +156691,11 @@ "pull_request:12534", "pull_request:12919", "pull_request:2334", - "pull_request:2345", "pull_request:2397", - "pull_request:2467", - "pull_request:2693", - "pull_request:3278", "pull_request:3673", - "pull_request:3691", "pull_request:3850", "pull_request:4099", - "pull_request:4272", - "pull_request:4301", "pull_request:4621", - "pull_request:4715", "pull_request:4947", "pull_request:5186", "pull_request:5526", @@ -156574,16 +156717,14 @@ "pull_request:8650", "pull_request:8684", "pull_request:8698", + "pull_request:8755", "pull_request:8932", "pull_request:9247", "pull_request:9257", + "pull_request:9524", "pull_request:9896" ], "soft_pairs": [ - "pull_request:2345|pull_request:3691", - "pull_request:2467|pull_request:4272", - "pull_request:2693|pull_request:4301", - "pull_request:3278|pull_request:4715", "pull_request:3673|pull_request:4621", "pull_request:3850|pull_request:5526", "pull_request:4099|pull_request:6195", @@ -156604,186 +156745,209 @@ "pull_request:2334|pull_request:2397", "pull_request:11916|pull_request:12439", "pull_request:12534|pull_request:12919", - "pull_request:10148|pull_request:10850" + "pull_request:10148|pull_request:10850", + "pull_request:10230|pull_request:11269", + "pull_request:10326|pull_request:8755", + "pull_request:10539|pull_request:9257", + "pull_request:10547|pull_request:9524", + "pull_request:10966|pull_request:11479" ], - "prepared_review_unit_hash": "0eb9bdfad3e6f0aa5d4c2e182f69b936b3a783b0718b7ce333d6e67fb1a391b6", + "prepared_review_unit_hash": "594bc55285b00f03361b8c5563511461a4e33aff3dbdbddb793fc3f148056a64", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 25, - "serialized_chars": 102003, - "estimated_input_tokens": 25501, - "estimated_eval_tokens": 51258 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 102578, + "estimated_input_tokens": 25645, + "estimated_eval_tokens": 51546 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a loose similarity bucket, not a duplicate cluster: the PRs span unrelated features, docs, tests, refactors, and platform-specific fixes. None of the soft edges looks like the same underlying change.", - "confidence": 0.98, + "summary": "The cluster is heterogeneous: the soft links mostly connect unrelated PRs that happen to share a subsystem, doc area, or test target. None look like true duplicates or a single mergeable code change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: the items are heterogeneous and do not represent one underlying bug or feature change.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No good global PR representative; the set mixes unrelated changes across different subsystems and release eras.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:2345", - "right": "pull_request:3691", - "accept": false, - "reason": "xformers 0.0.16 warning vs text-encoder LoRA processor selection; unrelated fixes." - }, - { - "left": "pull_request:2467", - "right": "pull_request:4272", - "accept": false, - "reason": "latent output type/docs cleanup vs missing optional import statements; different concerns." - }, - { - "left": "pull_request:2693", - "right": "pull_request:4301", - "accept": false, - "reason": "inpaint doc link fix vs Kandinsky doc update; not the same change." - }, - { - "left": "pull_request:3278", - "right": "pull_request:4715", - "accept": false, - "reason": "disable spectrogram diffusion tests vs renaming example test files; unrelated maintenance." - }, { "left": "pull_request:3673", "right": "pull_request:4621", "accept": false, - "reason": "multi-GPU DreamBooth fix vs T2IAdapter validation/downscale logic; different code paths." + "reason": "Different fixes: DreamBooth multi-GPU logic vs T2I-Adapter downscale-factor validation." }, { "left": "pull_request:3850", "right": "pull_request:5526", "accept": false, - "reason": "audio diffusion test adjustments vs SDXL scheduler timestep logic; not duplicative." + "reason": "Unrelated test cleanup vs scheduler/inpaint timestep behavior fix." }, { "left": "pull_request:4099", "right": "pull_request:6195", "accept": false, - "reason": "SDXL prompt-embed bug fix vs AnimateDiff SDXL implementation work; distinct changes." + "reason": "Both touch UNet/SDXL-related code, but one fixes prompt-embed handling and the other adds AnimateDiff SDXL support." }, { "left": "pull_request:4947", "right": "pull_request:5853", "accept": false, - "reason": "T2I adapter training test addition vs doctest/workflow infrastructure; unrelated." + "reason": "Training-script tests vs doctest workflow infrastructure; no shared bug/change." }, { "left": "pull_request:5186", "right": "pull_request:7328", "accept": false, - "reason": "FreeU proof-of-concept vs PyPI workflow dependency fix; no shared code-path problem." + "reason": "FreeU proof-of-concept vs PyPI workflow dependency fix are unrelated." }, { "left": "pull_request:6007", "right": "pull_request:6237", "accept": false, - "reason": "docs for #Copied from mechanism vs batched seeds guidance; separate documentation topics." + "reason": "Docs wording change vs batched seed guidance; different documentation updates." }, { "left": "pull_request:6007", "right": "pull_request:7134", "accept": false, - "reason": "docs mechanism update vs UNet type hints; unrelated doc edits." + "reason": "Contribution docs update vs UNet type-hint cleanup; unrelated." }, { "left": "pull_request:6864", "right": "pull_request:8684", "accept": false, - "reason": "PEFT/torch.compile note vs Pillow import fix in controlnet docs; different documentation changes." + "reason": "PEFT inference docs vs Pillow import fix in a controlnet example; unrelated." }, { "left": "pull_request:6961", "right": "pull_request:7295", "accept": false, - "reason": "single-file checkpoint loader tests vs UNet single-file loader doc cleanup; different objectives." + "reason": "Single-file loader test utilities vs UNet single-file loader copy-paste cleanup; different targets." }, { "left": "pull_request:7044", "right": "pull_request:7134", "accept": false, - "reason": "UNet refactor restoration vs type-hint documentation cleanup; not the same underlying change." + "reason": "Same file family, but one re-adds UNet refactor code and the other is type-hint cleanup." }, { "left": "pull_request:7202", "right": "pull_request:8698", "accept": false, - "reason": "remove huggingface-cli dependency vs transformer2d output deprecation removal; unrelated chores." + "reason": "Requirements-file cleanup vs transformer output-class deprecation removal; unrelated." }, { "left": "pull_request:7206", "right": "pull_request:8650", "accept": false, - "reason": "training script tags update vs SD3 DreamBooth test suite; different scopes entirely." + "reason": "Training-script tag metadata vs SD3 DreamBooth test suite; different concerns." }, { "left": "pull_request:7410", "right": "pull_request:9247", "accept": false, - "reason": "FreeInit behavior refactor vs model-variant identification fix; different subsystems." + "reason": "FreeInit behavior fix vs pipeline variant-identification bug; unrelated code paths." }, { "left": "pull_request:8404", "right": "pull_request:8932", "accept": false, - "reason": "Latte video model addition vs temporal transformer test coverage; not the same feature." + "reason": "Latte model/pipeline addition vs temporal transformer test coverage; not the same change." }, { "left": "pull_request:8522", "right": "pull_request:8932", "accept": false, - "reason": "transformers version pin vs temporal transformer tests; unrelated release/test work." + "reason": "Transformers version pinning vs temporal transformer tests; unrelated." }, { "left": "pull_request:9257", "right": "pull_request:9896", "accept": false, - "reason": "DreamBooth Flux bug fix vs revert of Flux device/typecasting changes; different code paths and intents." + "reason": "Flux DreamBooth training-script bug fix vs reverting explicit device transfers in Flux pipelines; different issues." }, { "left": "pull_request:11602", "right": "pull_request:12301", "accept": false, - "reason": "Sana Sprint img2img pipeline addition vs Qwen controlnet inpaint support; separate pipelines." + "reason": "Sana Sprint img2img pipeline addition vs Qwen-Image ControlNet inpaint pipeline addition; different features." }, { "left": "pull_request:2334", "right": "pull_request:2397", "accept": false, - "reason": "Pix2Pix Zero initial pipeline addition vs later ddim inversion support; related family, but not the same concrete change." + "reason": "Both are Pix2Pix Zero-related, but one adds the pipeline and the other adds DDIM inversion support; not the same concrete change." }, { "left": "pull_request:11916", "right": "pull_request:12439", "accept": false, - "reason": "Flux/Chroma attention dispatcher refactor vs adding kernels-backed attention backends; distinct fixes." + "reason": "Attention dispatcher refactor vs adding SageAttention/FA2 kernels support; different backend changes." }, { "left": "pull_request:12534", "right": "pull_request:12919", "accept": false, - "reason": "Flux pos-embed NPU inference tweak vs NPU attention backend support; separate NPU work." + "reason": "Ascend NPU positional-embedding handling vs NPU attention backend implementation; different problems." }, { "left": "pull_request:10148", "right": "pull_request:10850", "accept": false, - "reason": "OmniGen feature addition vs hook helper utilities; unrelated PRs despite touching hook-related files." + "reason": "OmniGen feature addition vs hook utility-method refactor; unrelated." + }, + { + "left": "pull_request:10230", + "right": "pull_request:11269", + "accept": false, + "reason": "Docs for LoRA loader mixins vs ONNX Runtime package detection updates; unrelated." + }, + { + "left": "pull_request:10326", + "right": "pull_request:8755", + "accept": false, + "reason": "CI workflow quoting fix vs SD3 DreamBooth LoRA training bug fix; unrelated." + }, + { + "left": "pull_request:10539", + "right": "pull_request:9257", + "accept": false, + "reason": "Flux true-CFG condition fix vs Flux DreamBooth script bug fix; both Flux but different code paths and failures." + }, + { + "left": "pull_request:10547", + "right": "pull_request:9524", + "accept": false, + "reason": "Offload-test parameterization vs CogVideoX VAE conv_cache refactor; different areas." + }, + { + "left": "pull_request:10966", + "right": "pull_request:11479", + "accept": false, + "reason": "CogView4 input validation fix vs Intel Gaudi docs rename; unrelated." + }, + { + "left": "pull_request:110?", + "right": "pull_request:110?", + "accept": false, + "reason": "No matching pair provided." + }, + { + "left": "pull_request:7206", + "right": "pull_request:7206", + "accept": false, + "reason": "No duplicate edge here; listed for completeness in the candidate set." } ] }, @@ -156799,7 +156963,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 169, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -156807,20 +156971,16 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10230", - "pull_request:10326", - "pull_request:10539", - "pull_request:10547", - "pull_request:10966", "pull_request:11269", "pull_request:11449", - "pull_request:11479", "pull_request:11495", "pull_request:12092", "pull_request:12586", "pull_request:12911", "pull_request:13022", "pull_request:13211", + "pull_request:1324", + "pull_request:1579", "pull_request:1623", "pull_request:1850", "pull_request:2119", @@ -156846,22 +157006,20 @@ "pull_request:6864", "pull_request:689", "pull_request:7050", + "pull_request:7652", "pull_request:7712", "pull_request:7778", "pull_request:8404", - "pull_request:8755", "pull_request:9065", + "pull_request:9074", "pull_request:9207", - "pull_request:9257", - "pull_request:9524", - "pull_request:9687" + "pull_request:9400", + "pull_request:9489", + "pull_request:9579", + "pull_request:9687", + "pull_request:9988" ], "soft_pairs": [ - "pull_request:10230|pull_request:11269", - "pull_request:10326|pull_request:8755", - "pull_request:10539|pull_request:9257", - "pull_request:10547|pull_request:9524", - "pull_request:10966|pull_request:11479", "pull_request:11269|pull_request:9687", "pull_request:11449|pull_request:11495", "pull_request:12092|pull_request:12586", @@ -156883,198 +157041,196 @@ "pull_request:6311|pull_request:7778", "pull_request:6864|pull_request:9065", "pull_request:7050|pull_request:7712", - "pull_request:7050|pull_request:9207" + "pull_request:7050|pull_request:9207", + "pull_request:7652|pull_request:9400", + "pull_request:9489|pull_request:9579", + "pull_request:9074|pull_request:9988", + "pull_request:1324|pull_request:1579" ], - "prepared_review_unit_hash": "805f70aaafa75bd3c0ed557e6fd91b4116d1fafd6e7e86d3343d79d54f56926f", + "prepared_review_unit_hash": "dbcd6e8cd8f9f14d22ad7756dc0d038aa61b4be44b573e836c3aff04e7409fd6", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 27, - "serialized_chars": 96464, - "estimated_input_tokens": 24116, - "estimated_eval_tokens": 48488 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 96688, + "estimated_input_tokens": 24172, + "estimated_eval_tokens": 48600 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft links are overwhelmingly false positives: the cluster mixes unrelated docs, test-only changes, workflow/config fixes, dependency bumps, and distinct feature/bug fixes. Even the few same-subsystem pairs target different scripts or code paths, so there is no duplicate PR cluster here.", - "confidence": 0.96, + "summary": "This cluster is not a duplicate set: it mixes unrelated docs, example-script edits, build/test fixes, and Flux/ControlNet/FreeU feature work. All soft edges are rejected because they do not match the same concrete bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No clear canonical PR: the items are heterogeneous and do not share one underlying bug or change.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:10230", - "right": "pull_request:11269", - "accept": false, - "reason": "Docs for LoRA loader mixins vs ONNX runtime backend additions; unrelated changes." - }, - { - "left": "pull_request:10326", - "right": "pull_request:8755", - "accept": false, - "reason": "Workflow packaging fix vs SD3 Dreambooth LoRA training bugfix; different code paths." - }, - { - "left": "pull_request:10539", - "right": "pull_request:9257", - "accept": false, - "reason": "Flux true-CFG logic vs Dreambooth Flux prompt/tokenization changes; not the same bug." - }, - { - "left": "pull_request:10547", - "right": "pull_request:9524", - "accept": false, - "reason": "Offload test tweaks vs CogVideoX VAE refactor; unrelated." - }, - { - "left": "pull_request:10966", - "right": "pull_request:11479", - "accept": false, - "reason": "CogView4 input validation vs Gaudi docs rename; unrelated." - }, { "left": "pull_request:11269", "right": "pull_request:9687", "accept": false, - "reason": "ONNX runtime availability detection vs Xlabs ControlNet support; unrelated." + "reason": "Different scopes: ONNXRuntime package detection vs Flux Redux model/pipeline conversion changes." }, { "left": "pull_request:11449", "right": "pull_request:11495", "accept": false, - "reason": "Both touch image interpolation, but in different training scripts and different fixes; not one concrete change." + "reason": "Both touch image interpolation defaults, but for different training scripts and separate code paths." }, { "left": "pull_request:12092", "right": "pull_request:12586", "accept": false, - "reason": "GGUF docs content vs doc-toctree sort/reorder; unrelated." + "reason": "GGUF docs content vs a docs tree sorting/move change; not the same change." }, { "left": "pull_request:12092", "right": "pull_request:12911", "accept": false, - "reason": "GGUF docs vs discrete diffusion feature addition; unrelated." + "reason": "GGUF conversion docs are unrelated to discrete diffusion docs/examples." }, { "left": "pull_request:13022", "right": "pull_request:13211", "accept": false, - "reason": "Dockerfile compatibility fixes vs group offloading runtime bug; unrelated." + "reason": "Dockerfile dependency/runtime fixes vs group-offloading runtime bug fix are unrelated." }, { "left": "pull_request:1623", "right": "pull_request:538", "accept": false, - "reason": "Different dependency updates in different files and times; not the same underlying issue." + "reason": "Requirements bumps for example scripts vs core hub dependency version update; different artifacts and purpose." }, { "left": "pull_request:1850", "right": "pull_request:2983", "accept": false, - "reason": "ResnetBlock2D tests vs contributor docs update; unrelated." + "reason": "A model-layer test addition is unrelated to documentation about installing test dependencies." }, { "left": "pull_request:2119", "right": "pull_request:538", "accept": false, - "reason": "Local-files-only Hub access fix vs hub version bump; unrelated." + "reason": "Local-files-only Hub access guard vs dependency version bump; different problems." }, { "left": "pull_request:2470", "right": "pull_request:2983", "accept": false, - "reason": "MPS test skips vs CONTRIBUTING.md docs; unrelated." + "reason": "MPS-specific test skips vs contributing-doc test setup guidance are unrelated." }, { "left": "pull_request:3077", "right": "pull_request:3675", "accept": false, - "reason": "Flax text-to-image sample logic vs a one-off UniDiffuser test fix; unrelated." + "reason": "Flax dataset preprocessing change vs a V100 test fix; no shared underlying bug." }, { "left": "pull_request:3471", "right": "pull_request:4112", "accept": false, - "reason": "Dreambooth torch.compile docs note vs Dreambooth rank change; different artifacts and issues." + "reason": "DreamBooth README note vs LoRA rank support in training code are distinct changes." }, { "left": "pull_request:3723", "right": "pull_request:4112", "accept": false, - "reason": "Reference-image community example update vs Dreambooth rank support; unrelated." + "reason": "Reference image input enhancement vs DreamBooth LoRA rank support; different features." }, { "left": "pull_request:3957", "right": "pull_request:4595", "accept": false, - "reason": "Attention-name correction WIP vs SDXL-to-SD conversion WIP; unrelated." + "reason": "WIP attention-name correction vs XL-to-SD conversion work are unrelated." }, { "left": "pull_request:4248", "right": "pull_request:6081", "accept": false, - "reason": "Tracker config cleanup vs scheduler sigmas handling; unrelated." + "reason": "Trackers config fix in DreamBooth vs SDXL scheduler-sigmas handling; different code paths." }, { "left": "pull_request:4396", "right": "pull_request:6261", "accept": false, - "reason": "Encode-prompt dtype fallback vs moving Attend-and-Excite into its own module; unrelated." + "reason": "Prompt dtype fallback fix vs pipeline package refactor; not the same bug." }, { "left": "pull_request:51", "right": "pull_request:689", "accept": false, - "reason": "Install-requires pin fix vs attention performance rewrite; unrelated." + "reason": "Install requirements fix vs attention implementation performance tweak are unrelated." }, { "left": "pull_request:5164", "right": "pull_request:6007", "accept": false, - "reason": "FreeU core algorithm change vs docs about #Copied from; unrelated." + "reason": "FreeU core mechanism vs docs for the copied-from convention; completely different changes." }, { "left": "pull_request:6265", "right": "pull_request:8404", "accept": false, - "reason": "GLIGEN refactor vs Latte video model introduction; unrelated." + "reason": "Both touch import exposure, but one splits out GLIGEN while the other adds Latte; different features." }, { "left": "pull_request:6311", "right": "pull_request:7778", "accept": false, - "reason": "Tutorial doc addition vs debugging workflow; unrelated." + "reason": "Fast diffusion tutorial docs vs debugging workflow YAML are unrelated." }, { "left": "pull_request:6864", "right": "pull_request:9065", "accept": false, - "reason": "PEFT compile docs note vs adding sentencepiece as a soft dependency; unrelated." + "reason": "PEFT torch.compile docs vs sentencepiece soft-dependency support; not the same issue." }, { "left": "pull_request:7050", "right": "pull_request:7712", "accept": false, - "reason": "Typos in text-to-image examples vs import fixes; unrelated." + "reason": "Typo cleanup in examples vs import-path fixes; unrelated." }, { "left": "pull_request:7050", "right": "pull_request:9207", "accept": false, - "reason": "Typos in text-to-image examples vs CPU offloading docs/usages; unrelated." + "reason": "Text-to-image typo fixes vs CPU offloading docs fix; different changes." + }, + { + "left": "pull_request:7652", + "right": "pull_request:9400", + "accept": false, + "reason": "Docker buildx workflow setup vs device-agnostic test adjustments; unrelated." + }, + { + "left": "pull_request:9489", + "right": "pull_request:9579", + "accept": false, + "reason": "SD3 ControlNet training bug fix vs embeddings doc expansion; not the same underlying change." + }, + { + "left": "pull_request:9074", + "right": "pull_request:9988", + "accept": false, + "reason": "Both are Flux-related, but one is a core follow-up bug/processor fix while the other is a broader Redux/conversion feature PR; too distinct to merge as one fix." + }, + { + "left": "pull_request:1324", + "right": "pull_request:1579", + "accept": false, + "reason": "INT8 Stable Diffusion quantization docs/feature vs Intel-optimized example set; related theme but not the same concrete change." } ] }, @@ -157090,7 +157246,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 170, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -157100,26 +157256,29 @@ "nodes": [ "pull_request:10259", "pull_request:10326", + "pull_request:10354", "pull_request:10557", "pull_request:10808", "pull_request:10894", "pull_request:1129", "pull_request:11655", "pull_request:11663", + "pull_request:11726", "pull_request:11738", "pull_request:11869", + "pull_request:11913", "pull_request:1206", "pull_request:12119", "pull_request:12324", "pull_request:12387", + "pull_request:12418", "pull_request:12534", "pull_request:13134", "pull_request:13208", - "pull_request:1324", "pull_request:13294", "pull_request:13296", + "pull_request:13365", "pull_request:13562", - "pull_request:1579", "pull_request:2279", "pull_request:3746", "pull_request:3968", @@ -157131,26 +157290,20 @@ "pull_request:5887", "pull_request:6757", "pull_request:6868", + "pull_request:6929", + "pull_request:7062", "pull_request:7068", - "pull_request:7652", "pull_request:7712", "pull_request:8419", "pull_request:8452", "pull_request:8564", "pull_request:8652", + "pull_request:8684", "pull_request:8836", - "pull_request:9074", - "pull_request:9400", - "pull_request:9473", - "pull_request:9489", - "pull_request:9579", - "pull_request:9988" + "pull_request:9206", + "pull_request:9473" ], "soft_pairs": [ - "pull_request:7652|pull_request:9400", - "pull_request:9489|pull_request:9579", - "pull_request:9074|pull_request:9988", - "pull_request:1324|pull_request:1579", "pull_request:11663|pull_request:11738", "pull_request:1129|pull_request:1206", "pull_request:13134|pull_request:13296", @@ -157171,193 +157324,214 @@ "pull_request:11869|pull_request:12387", "pull_request:12324|pull_request:13208", "pull_request:10326|pull_request:10557", - "pull_request:10326|pull_request:10894" + "pull_request:10326|pull_request:10894", + "pull_request:10354|pull_request:9206", + "pull_request:10557|pull_request:9206", + "pull_request:12418|pull_request:13365", + "pull_request:6929|pull_request:7062", + "pull_request:6929|pull_request:8684", + "pull_request:11726|pull_request:11913" ], - "prepared_review_unit_hash": "2a6cbf2f2a399ac72e5828d5ea5418e4a9af970abae2e6c9db94d0d83f581c6c", + "prepared_review_unit_hash": "0298a94aa8925f18a35a7faf9a0012366843c688f5053eec85e5b1068c38bb58", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 25, - "serialized_chars": 104855, - "estimated_input_tokens": 26214, - "estimated_eval_tokens": 52684 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 104893, + "estimated_input_tokens": 26224, + "estimated_eval_tokens": 52704 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated PRs that only look similar at a very broad level (docs, CI, loaders, LoRA, Flux, IP-Adapter, quantization). The soft edges are all false positives; none appear to be the same underlying bug or change.", - "confidence": 0.95, + "summary": "Mostly a heterogeneous set of unrelated PRs; only two plausible duplicate threads stand out: bnb deterministic test stabilization (11663/11738) and SDXL watermarker optionality (3968/4346). The rest are different fixes/features despite superficial subsystem overlap.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR fits the whole set cleanly; the cluster is mostly false positives with only small duplicate subthreads.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "pull_request:3968 is the broadest representative of the only clear duplicate thread, covering the SDXL watermarker optionality change that 4346 narrows further.", "soft_edge_verdicts": [ - { - "left": "pull_request:7652", - "right": "pull_request:9400", - "accept": false, - "reason": "Docker Buildx workflow setup is unrelated to making single-file/pipeline tests device-agnostic." - }, - { - "left": "pull_request:9489", - "right": "pull_request:9579", - "accept": false, - "reason": "SD3 ControlNet training DDP kwargs and embeddings doc updates are different fixes in different areas." - }, - { - "left": "pull_request:9074", - "right": "pull_request:9988", - "accept": false, - "reason": "Both are Flux-related, but one is Flux processor/model cleanup and the other is Flux conversion/pipeline work; not the same concrete change." - }, - { - "left": "pull_request:1324", - "right": "pull_request:1579", - "accept": false, - "reason": "README quantization docs and Intel optimization example additions are unrelated scope changes." - }, { "left": "pull_request:11663", "right": "pull_request:11738", - "accept": false, - "reason": "One adds deterministic checks in test bodies; the other moves determinism to class setup. Same area, but not the same change." + "accept": true, + "reason": "Same bnb test determinism fix on the same files; 11738 is a refactor/consolidation of the change introduced in 11663." }, { "left": "pull_request:1129", "right": "pull_request:1206", "accept": false, - "reason": "DreamBooth Flax sample batch sizing and LDM layer-name conversion are unrelated code paths." + "reason": "Unrelated: DreamBooth Flax sample batch sizing vs. LDM checkpoint layer-name conversion." }, { "left": "pull_request:13134", "right": "pull_request:13296", "accept": false, - "reason": "Cosmos typing cleanup and CI transformer version updates are unrelated." + "reason": "Unrelated: Cosmos transformer typing vs. CI release-test transformer version pinning." }, { "left": "pull_request:13294", "right": "pull_request:13562", "accept": false, - "reason": "WanVACE flaky-test cleanup and CI tokenizers install changes do not fix the same issue." + "reason": "Unrelated: WanVACE flaky test cleanup vs. CI tokenizers install policy." }, { "left": "pull_request:2279", "right": "pull_request:3746", "accept": false, - "reason": "Ruff/black migration and from_ckpt path-join fix are unrelated maintenance changes." + "reason": "Unrelated: lint-tool migration vs. from_ckpt path joining fix." }, { "left": "pull_request:4944", "right": "pull_request:6868", "accept": false, - "reason": "Both concern IP-Adapter, but 4944 is the core support implementation while 6868 is a later pipeline-call extension; they are not one mergeable bug fix." + "reason": "Related subsystem, but different changes: initial IP-Adapter support vs. adding ip_adapter_image_embeds to pipeline calls." }, { "left": "pull_request:6757", "right": "pull_request:8419", "accept": false, - "reason": "Moving sigma tensors to device and removing cached_download are unrelated changes." + "reason": "Unrelated: SDXL k-diffusion sigma device placement vs. dynamic module download backend change." }, { "left": "pull_request:7068", "right": "pull_request:7712", "accept": false, - "reason": "Typo fixes and import fixes are different cleanup changes." + "reason": "Unrelated: docstring/typo cleanup vs. import path fixes." }, { "left": "pull_request:7712", "right": "pull_request:8836", "accept": false, - "reason": "Import path corrections and sliced-attention loop bugfix are unrelated." + "reason": "Unrelated: import fixes vs. sliced attention loop bug fix." }, { "left": "pull_request:8452", "right": "pull_request:8652", "accept": false, - "reason": "Single-file loading strictness fix and Lumina model addition are unrelated." + "reason": "Unrelated: single-file loading error fix vs. adding Lumina-T2X." }, { "left": "pull_request:8564", "right": "pull_request:9473", "accept": false, - "reason": "SD3 training token-limit docs/code and Flux DreamBooth LoRA learning-rate fix are unrelated training changes." + "reason": "Both mention training/T5, but they fix different scripts and different bugs." }, { "left": "pull_request:10808", "right": "pull_request:12119", "accept": false, - "reason": "PEFT config key handling and Qwen-Image-LoRA support are both loader-related but fix different model paths and problems." + "reason": "Unrelated: peft config key handling vs. Qwen-Image-Lightning LoRA loading support." }, { "left": "pull_request:11655", "right": "pull_request:12534", "accept": false, - "reason": "bitsandbytes 8-bit LoRA support and Flux NPU pos-embed handling are unrelated subsystems." + "reason": "Same broad Flux area, but different concrete problems: 8-bit LoRA loading vs. NPU positional embedding inference." }, { "left": "pull_request:3968", "right": "pull_request:4346", - "accept": false, - "reason": "Both touch SDXL, but one is a broader pipeline improvement and the other is optional watermark behavior; not the same concrete change." + "accept": true, + "reason": "Same SDXL watermarker optionality change on overlapping files; 3968 is the broader umbrella and 4346 is a focused follow-up on the same code path." }, { "left": "pull_request:5713", "right": "pull_request:5887", "accept": false, - "reason": "General IP-Adapter support and ControlNet-inpaint IP-Adapter support are related, but the latter is a separate pipeline integration, not the same change." + "reason": "5887 extends IP-Adapter to one pipeline, but it is not the same change as the core IP-Adapter feature in 5713." }, { "left": "pull_request:5298", "right": "pull_request:5439", "accept": false, - "reason": "LoRA conv torch.compile fix and from_single_file config/tokenizer loading fix are unrelated." + "reason": "Unrelated: LoRA conv torch.compile fix vs. single-file tokenizer/config loading." }, { "left": "pull_request:10259", "right": "pull_request:10808", "accept": false, - "reason": "Flux Turbo LoRA loading support and PEFT config update restrictions solve different issues." + "reason": "Unrelated: Flux LoRA loading support vs. PEFT config-update logic." }, { "left": "pull_request:11869", "right": "pull_request:12387", "accept": false, - "reason": "GGUF CUDA-kernel support and flash-attention through kernels are distinct backend features, not one fix." + "reason": "Both involve kernels, but one adds GGUF CUDA kernels and the other adds flash-attention backend support; different code paths." }, { "left": "pull_request:12324", "right": "pull_request:13208", "accept": false, - "reason": "Wan2.2 VACE conversion/pipeline work and Helios video pipelines are unrelated model families." + "reason": "Unrelated model additions: Wan2.2 VACE-Fun vs. Helios video pipelines." }, { "left": "pull_request:10326", "right": "pull_request:10557", "accept": false, - "reason": "MPS workflow quoting fix and InstructPix2Pix single-file slow-test fix are unrelated." + "reason": "Unrelated CI workflow fix vs. single-file slow-test fix." }, { "left": "pull_request:10326", "right": "pull_request:10894", "accept": false, - "reason": "MPS workflow quoting fix and docs toctree sorting are unrelated maintenance changes." + "reason": "Unrelated CI workflow fix vs. docs toctree sorting." + }, + { + "left": "pull_request:10354", + "right": "pull_request:9206", + "accept": false, + "reason": "Unrelated: Flux IP-Adapter tester dtype/device fix vs. xformers test skips." + }, + { + "left": "pull_request:10557", + "right": "pull_request:9206", + "accept": false, + "reason": "Unrelated test areas and failure modes." + }, + { + "left": "pull_request:12418", + "right": "pull_request:13365", + "accept": false, + "reason": "Unrelated: xfail CI tests vs. Claude review workflow support for forks." + }, + { + "left": "pull_request:6929", + "right": "pull_request:7062", + "accept": false, + "reason": "Unrelated docs changes: community pipelines loading vs. imagic docstring correction." + }, + { + "left": "pull_request:6929", + "right": "pull_request:8684", + "accept": false, + "reason": "Unrelated docs/example fixes in different pipelines." + }, + { + "left": "pull_request:11726", + "right": "pull_request:11913", + "accept": false, + "reason": "Both Wan-related, but one is LoRA conversion support and the other is the modular Wan pipeline; not the same change." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: most pairs are clearly unrelated, and the two accepted threads have plausible same-code-path overlap with the later PRs looking like narrow follow-ups. The verdicts are generally conservative, and I don\u2019t see a clear overstatement of evidence." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -157366,18 +157540,16 @@ "component_index": 1, "component_count": 266, "review_unit_index": 171, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10114", - "pull_request:10354", "pull_request:10419", - "pull_request:10557", "pull_request:10723", "pull_request:10758", "pull_request:10922", @@ -157386,18 +157558,15 @@ "pull_request:11252", "pull_request:1129", "pull_request:1171", - "pull_request:11726", "pull_request:11770", "pull_request:11913", "pull_request:11995", "pull_request:12062", "pull_request:12395", - "pull_request:12418", "pull_request:1250", "pull_request:12934", "pull_request:1309", "pull_request:13229", - "pull_request:13365", "pull_request:1338", "pull_request:1588", "pull_request:1968", @@ -157408,28 +157577,26 @@ "pull_request:2489", "pull_request:2529", "pull_request:2565", + "pull_request:2590", + "pull_request:2593", "pull_request:2861", "pull_request:3121", "pull_request:326", + "pull_request:3299", + "pull_request:3506", + "pull_request:3563", + "pull_request:3589", + "pull_request:3681", "pull_request:3711", "pull_request:4112", "pull_request:4395", "pull_request:4465", "pull_request:530", - "pull_request:6929", - "pull_request:7062", + "pull_request:5499", "pull_request:788", - "pull_request:8684", - "pull_request:9206", "pull_request:9968" ], "soft_pairs": [ - "pull_request:10354|pull_request:9206", - "pull_request:10557|pull_request:9206", - "pull_request:12418|pull_request:13365", - "pull_request:6929|pull_request:7062", - "pull_request:6929|pull_request:8684", - "pull_request:11726|pull_request:11913", "pull_request:10723|pull_request:10758", "pull_request:11913|pull_request:11995", "pull_request:10114|pull_request:11252", @@ -157449,186 +157616,184 @@ "pull_request:2347|pull_request:4112", "pull_request:2384|pull_request:3711", "pull_request:2489|pull_request:3121", - "pull_request:2529|pull_request:4465" + "pull_request:2529|pull_request:4465", + "pull_request:2590|pull_request:3681", + "pull_request:2593|pull_request:3506", + "pull_request:3299|pull_request:3589", + "pull_request:3563|pull_request:5499" ], - "prepared_review_unit_hash": "95b8e5eaa358bb0faf9c9a27c8d62dcf688a1b56808eed4255b1db15578f07e6", + "prepared_review_unit_hash": "844999e387b261f711415d318517fa6ff5c9a305c627a8bd1d7f18322bb346f2", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 26, - "serialized_chars": 106383, - "estimated_input_tokens": 26596, - "estimated_eval_tokens": 53448 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 104852, + "estimated_input_tokens": 26213, + "estimated_eval_tokens": 52682 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the items span unrelated formatting fixes, docs tweaks, model/runtime features, CI changes, and test updates. None of the soft-edge pairs look like the same underlying bug or change.", - "confidence": 0.93, + "summary": "The soft-edge pairs are largely false positives: they share broad keywords or subsystems (Wan, ControlNet, docs, CI, DreamBooth) but do not appear to fix the same concrete bug or implement the same change. I would not merge any of these pairs as duplicates.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR stands out because the cluster is heterogeneous; if forced, PR 11995 is the broadest self-contained feature, but it is not a true duplicate anchor.", "best_issue_reason": null, - "best_pr_reason": "PR 12934 is the strongest standalone artifact here: it is a large, coherent feature PR with substantial review activity and a clearly defined scope. Even so, it is not a duplicate anchor for the rest of the mixed cluster.", + "best_pr_reason": "PR 11995 is the closest thing to a broad representative, but this set is too mixed for any PR to serve as a strong canonical duplicate target.", "soft_edge_verdicts": [ - { - "left": "pull_request:10354", - "right": "pull_request:9206", - "accept": false, - "reason": "Different fixes: Flux IP-Adapter tester/device handling vs broken xformers attention tests." - }, - { - "left": "pull_request:10557", - "right": "pull_request:9206", - "accept": false, - "reason": "Different test areas: single-file Stable Diffusion slow tests vs xformers-related animation tests." - }, - { - "left": "pull_request:12418", - "right": "pull_request:13365", - "accept": false, - "reason": "One is CI test xfail stabilization; the other updates Claude review workflow for forks." - }, - { - "left": "pull_request:6929", - "right": "pull_request:7062", - "accept": false, - "reason": "Both are docs-related, but one adds community pipeline overview content while the other fixes an Imagic docstring." - }, - { - "left": "pull_request:6929", - "right": "pull_request:8684", - "accept": false, - "reason": "Unrelated docs/import tweaks in different areas; not the same bug or change." - }, - { - "left": "pull_request:11726", - "right": "pull_request:11913", - "accept": false, - "reason": "Wan LoRA conversion support and Wan modular pipeline support are separate features, not one fix." - }, { "left": "pull_request:10723", "right": "pull_request:10758", "accept": false, - "reason": "ControlNet Union scaling and from_pretrained kwarg typing are unrelated changes." + "reason": "Different changes: ControlNet Union scaling vs from_pretrained type-checking docs/code." }, { "left": "pull_request:11913", "right": "pull_request:11995", "accept": false, - "reason": "Both touch modular pipelines, but they add different model families and different codepaths." + "reason": "Both are modular-pipeline work, but they target different models and features (Wan vs Flux)." }, { "left": "pull_request:10114", "right": "pull_request:11252", "accept": false, - "reason": "Stable Audio formatting cleanup vs timeout-constant change in a utility script." + "reason": "Unrelated edits: Stable Audio formatting vs a timeout constant fix in a utility script." }, { "left": "pull_request:10419", "right": "pull_request:11770", "accept": false, - "reason": "Both mention docs, but they fix different docstring/doc layout issues in different modules." + "reason": "Both touch docs, but one fixes scheduler docstrings and the other cleans up LoRA docs/base helpers." }, { "left": "pull_request:10922", "right": "pull_request:10943", "accept": false, - "reason": "Wan model/pipeline support and Wan LoRA support are related ecosystem work, but not the same concrete change." + "reason": "Same Wan ecosystem, but one adds Wan pipeline/model support and the other adds LoRA support; distinct changes." }, { "left": "pull_request:11095", "right": "pull_request:9968", "accept": false, - "reason": "A docs/example fix in Flux ControlNet is unrelated to the ControlNet module refactor." + "reason": "A small Flux ControlNet doc fix is not the same as a broad ControlNet module refactor." }, { "left": "pull_request:1129", "right": "pull_request:1171", "accept": false, - "reason": "DreamBooth Flax batch sizing and a Stable Diffusion typo fix are unrelated." + "reason": "DreamBooth batch-size bug fix vs a typography/doc cleanup; unrelated underlying changes." }, { "left": "pull_request:12062", "right": "pull_request:12395", "accept": false, - "reason": "Both are CI workflow edits, but they address different runner/setup problems." + "reason": "Different CI concerns: GPU selection in workflows vs prerelease dependency installation." }, { "left": "pull_request:1250", "right": "pull_request:326", "accept": false, - "reason": "A community pipeline addition and a PNDM type-hint cleanup are unrelated." + "reason": "Community pipeline addition is unrelated to a PNDM type-hint/doc change." }, { "left": "pull_request:12934", "right": "pull_request:13229", "accept": false, - "reason": "Both are video-model support changes, but they target different pipelines/checkpoints and different implementation paths." + "reason": "Both are video-model releases, but they introduce different model support and code paths." }, { "left": "pull_request:1309", "right": "pull_request:788", "accept": false, - "reason": "Both concern PIL handling, but one is broad compatibility handling while the other is a specific resampling warning fix." + "reason": "Both mention PIL, but one improves PIL handling broadly while the other fixes a specific warning issue." + }, + { + "left": "pull_request:1338", + "right": "pull_request:530", + "accept": false, + "reason": "ONNX inpainting cast order and img2img test flakiness are unrelated." }, { "left": "pull_request:1588", "right": "pull_request:2861", "accept": false, - "reason": "FP16 test fixes and StableUnCLIP image_embeds=None handling are unrelated." + "reason": "FP16 test adjustments and StableUnCLIP image_embeds handling are different bugs." }, { "left": "pull_request:1968", "right": "pull_request:2565", "accept": false, - "reason": "CPU offload behavior and UNet1D doc correction are not the same issue." + "reason": "CPU offload docs/code and a UNet1D doc correction are unrelated." }, { "left": "pull_request:2037", "right": "pull_request:326", "accept": false, - "reason": "A Stable UnCLIP community pipeline and a scheduler type-hint cleanup are unrelated." + "reason": "A Stable UnCLIP community pipeline and PNDM type hints are not the same change." }, { "left": "pull_request:2339", "right": "pull_request:4395", "accept": false, - "reason": "Both touch DreamBooth, but one is checkpointing docs/tests and the other fixes class-prompt pre-encoding behavior." + "reason": "Both are DreamBooth-related, but one is checkpointing/docs and the other is class-prompt pre-encoding logic." }, { "left": "pull_request:2347", "right": "pull_request:4112", "accept": false, - "reason": "Scheduler noising schedule work and DreamBooth LoRA rank support are unrelated." + "reason": "Scheduler noising-schedule work is unrelated to adding rank to DreamBooth LoRA." }, { "left": "pull_request:2384", "right": "pull_request:3711", "accept": false, - "reason": "Pix2Pix0 utility revert and doc-build workflow revert are unrelated." + "reason": "A Pix2Pix0 utility revert and a doc-build workflow revert are unrelated." }, { "left": "pull_request:2489", "right": "pull_request:3121", "accept": false, - "reason": "CoreML docs addition and ControlNet global pooling are unrelated." + "reason": "Core ML docs and ControlNet global pooling fix different concerns entirely." }, { "left": "pull_request:2529", "right": "pull_request:4465", "accept": false, - "reason": "SD checkpoint conversion fix and ControlNet-to-TensorRT conversion are different codepaths and changes." + "reason": "Stable Diffusion checkpoint conversion and ControlNet-to-TensorRT conversion are different conversion pipelines." + }, + { + "left": "pull_request:2590", + "right": "pull_request:3681", + "accept": false, + "reason": "CI deps-table checking and checkpoint-limit handling are unrelated." + }, + { + "left": "pull_request:2593", + "right": "pull_request:3506", + "accept": false, + "reason": "ControlNet checkpoint conversion and DreamBooth checkpoint resume logic are unrelated." + }, + { + "left": "pull_request:3299", + "right": "pull_request:3589", + "accept": false, + "reason": "DreamBooth LoRA doc/example updates and README reference tricks are separate doc changes." + }, + { + "left": "pull_request:3563", + "right": "pull_request:5499", + "accept": false, + "reason": "A VQ pipeline bug fix and general docs link fixes do not share the same underlying issue." } ] }, @@ -157644,27 +157809,27 @@ "component_index": 1, "component_count": 266, "review_unit_index": 172, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10324", + "pull_request:10630", + "pull_request:11124", + "pull_request:11469", "pull_request:11995", "pull_request:12041", + "pull_request:12277", "pull_request:12416", "pull_request:12523", + "pull_request:1277", + "pull_request:1579", "pull_request:1966", - "pull_request:2590", - "pull_request:2593", "pull_request:2727", - "pull_request:3299", - "pull_request:3506", - "pull_request:3563", - "pull_request:3589", - "pull_request:3681", "pull_request:3850", "pull_request:3928", "pull_request:4040", @@ -157674,7 +157839,6 @@ "pull_request:5122", "pull_request:5312", "pull_request:5440", - "pull_request:5499", "pull_request:5695", "pull_request:5698", "pull_request:6007", @@ -157697,14 +157861,11 @@ "pull_request:8282", "pull_request:8531", "pull_request:8761", + "pull_request:9206", "pull_request:9473", "pull_request:9751" ], "soft_pairs": [ - "pull_request:2590|pull_request:3681", - "pull_request:2593|pull_request:3506", - "pull_request:3299|pull_request:3589", - "pull_request:3563|pull_request:5499", "pull_request:3850|pull_request:5695", "pull_request:3850|pull_request:6099", "pull_request:3928|pull_request:4040", @@ -157724,180 +157885,184 @@ "pull_request:1966|pull_request:2727", "pull_request:12041|pull_request:12523", "pull_request:11995|pull_request:12416", - "pull_request:5053|pull_request:5440" + "pull_request:5053|pull_request:5440", + "pull_request:10324|pull_request:9206", + "pull_request:10630|pull_request:11124", + "pull_request:11469|pull_request:12277", + "pull_request:1277|pull_request:1579" ], - "prepared_review_unit_hash": "aaf54eba61651f05c2cf85477c23dd168762aaf67d8965a8ce7dd59a1fae27ba", + "prepared_review_unit_hash": "dca7a9469f7b4427eb618e5e1af86ee4217f477149b7058c930c6f0ee6e41955", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 100792, - "estimated_input_tokens": 25198, - "estimated_eval_tokens": 50652 + "serialized_chars": 101413, + "estimated_input_tokens": 25354, + "estimated_eval_tokens": 50964 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated PRs across modular pipelines, training scripts, schedulers, docs, tests, and model utilities. The suggested soft links are all superficial (shared subsystem or file, not the same concrete bug/change), so none should be merged as duplicates.", + "summary": "This cluster is highly heterogeneous: it mixes unrelated test tweaks, docs edits, scheduler/model fixes, and feature PRs across many subsystems. I don\u2019t see any true duplicate pair here, so there is no sensible canonical artifact for the cluster.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR fits: the items span many independent features/fixes with no shared underlying change or bug.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is a good global representative because the cluster is heterogeneous and not a duplicate set.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:2590", - "right": "pull_request:3681", - "accept": false, - "reason": "CI workflow tweak vs checkpoint resume behavior in controlnet training; different subsystems and bugs." - }, - { - "left": "pull_request:2593", - "right": "pull_request:3506", - "accept": false, - "reason": "ControlNet conversion script addition vs training resume-step fix; unrelated changes." - }, - { - "left": "pull_request:3299", - "right": "pull_request:3589", - "accept": false, - "reason": "Different docs updates for different topics; no shared underlying issue." - }, - { - "left": "pull_request:3563", - "right": "pull_request:5499", - "accept": false, - "reason": "VQ pipeline bug fix vs docs link cleanup; code-path problem is unrelated to documentation edits." - }, { "left": "pull_request:3850", "right": "pull_request:5695", "accept": false, - "reason": "Audio diffusion test adjustment vs PixArt masking bug; unrelated model/pipeline areas." + "reason": "Different pipelines and fixes: AudioDiffusion test setup vs PixArt-Alpha mask-feature logic." }, { "left": "pull_request:3850", "right": "pull_request:6099", "accept": false, - "reason": "Audio diffusion test fix vs TextToVideoZero constructor fix; different code paths and failures." + "reason": "AudioDiffusion test fix is unrelated to TextToVideoZero constructor/super-class argument handling." }, { "left": "pull_request:3928", "right": "pull_request:4040", "accept": false, - "reason": "ControlNet timestep logic fix vs config default-value type change; unrelated bugs." + "reason": "ControlNet guidance-timestep bug and `_use_default_values` type change are different code paths and bug classes." }, { "left": "pull_request:4078", "right": "pull_request:5122", "accept": false, - "reason": "Single-file loading dependency check vs workflow Python version update; not the same bug/change." + "reason": "Single-file accelerate requirement is unrelated to workflow Python-version bumps." }, { "left": "pull_request:4690", "right": "pull_request:6081", "accept": false, - "reason": "DPMSolver scheduler refactor vs SDXL pipeline sigmas edit; different concrete code-path problems." + "reason": "DPMSolver scheduler refactor is not the same issue as the Stable Diffusion XL torch.compile sigma workaround." }, { "left": "pull_request:5312", "right": "pull_request:6658", "accept": false, - "reason": "Model docs/typehint cleanup vs memory-footprint utility addition; unrelated changes." + "reason": "Docs/typehint cleanup in models is unrelated to adding memory-footprint utilities." }, { "left": "pull_request:5698", "right": "pull_request:6145", "accept": false, - "reason": "LCM img2img bug fix vs comment cleanup in distillation scripts; different code and purpose." + "reason": "LCM img2img bug fix is unrelated to comment cleanup in distillation scripts." }, { "left": "pull_request:6007", "right": "pull_request:8282", "accept": false, - "reason": "Docs mechanism note vs I2VGenXL docs clarification; not the same underlying change." + "reason": "Both are docs, but they cover different topics and changes, not the same underlying change." }, { "left": "pull_request:6217", "right": "pull_request:8531", "accept": false, - "reason": "Torch-compile VAE fix vs sharding when no device_map is passed; different bugs." + "reason": "Torch-compile VAE fix and sharding/device_map loading fix affect different execution paths." }, { "left": "pull_request:652", "right": "pull_request:754", "accept": false, - "reason": "Gradient-checkpointing test tolerance tweak vs dynamic module caching fix; unrelated." + "reason": "Gradient-checkpointing test tolerance change is unrelated to community-pipeline dynamic module handling." }, { "left": "pull_request:6893", "right": "pull_request:8761", "accept": false, - "reason": "3D UNet attention refactor vs Hunyuan embedding refactor; same broad area but different changes." + "reason": "3D UNet attention refactor and Hunyuan embedding refactor touch different models and problems." }, { "left": "pull_request:6918", "right": "pull_request:7919", "accept": false, - "reason": "Flux/Dreambooth hub-token exposure fix vs offload behavior with device_map; different issues." + "reason": "Hub-token security fix in training scripts is unrelated to offload behavior under `device_map`." }, { "left": "pull_request:6965", "right": "pull_request:7158", "accept": false, - "reason": "PEFT workflow test reshuffle vs LCM benchmark fix; unrelated test/benchmark changes." + "reason": "A PEFT workflow test move and an LCM benchmark fix are unrelated test changes." }, { "left": "pull_request:7335", "right": "pull_request:7930", "accept": false, - "reason": "Shared a test filename, but one is a ControlNet SDXL config test and the other is callback framework tests; not the same bug." + "reason": "ControlNet SDXL config/test cleanup and official callback tests do not fix the same bug." }, { "left": "pull_request:7455", "right": "pull_request:9473", "accept": false, - "reason": "Remove distutils vs FLUX Dreambooth LR fix; unrelated core utility and training script changes." + "reason": "Removing distutils and Flux Dreambooth LR logic are unrelated changes." }, { "left": "pull_request:7916", "right": "pull_request:9751", "accept": false, - "reason": "Both touch NPU support, but one is SDXL training and the other is FLUX Dreambooth; different pipelines and fixes." + "reason": "Both add NPU support, but for different models/pipelines and different code paths." }, { "left": "pull_request:1966", "right": "pull_request:2727", "accept": false, - "reason": "Flax inpainting implementation vs Flax ControlNet implementation; separate features, not the same change." + "reason": "Flax inpainting implementation and Flax ControlNet are separate feature additions." }, { "left": "pull_request:12041", "right": "pull_request:12523", "accept": false, - "reason": "Both touch AutoencoderKLWan, but one fixes patchify and the other skip-keys/load behavior; different concrete bugs." + "reason": "Both touch AutoencoderKLWan, but one fixes patchify logic while the other changes skip-key handling; not the same concrete bug." }, { "left": "pull_request:11995", "right": "pull_request:12416", "accept": false, - "reason": "Flux modular pipeline support vs QwenImage Edit Plus modular support; same framework area but different model families and changes." + "reason": "Different modular-pipeline additions: Flux text-to-image vs QwenImage Edit Plus." }, { "left": "pull_request:5053", "right": "pull_request:5440", "accept": false, - "reason": "Validation-image argparse fix vs local single-file loading fix; unrelated." + "reason": "Validation-image arg parsing and single-file loading are unrelated changes." + }, + { + "left": "pull_request:10324", + "right": "pull_request:9206", + "accept": false, + "reason": "Kandinsky CPU offload API fix is unrelated to broken xformers test suppression." + }, + { + "left": "pull_request:10630", + "right": "pull_request:11124", + "accept": false, + "reason": "Flax test slice updates and CogView4 tokenizer test fix are unrelated test-only changes." + }, + { + "left": "pull_request:11469", + "right": "pull_request:12277", + "accept": false, + "reason": "Both disable tests, but for different pipelines and reasons; they do not share one concrete bug." + }, + { + "left": "pull_request:1277", + "right": "pull_request:1579", + "accept": false, + "reason": "INT8 quantization docs and Intel optimization example additions are different documentation/features." } ] }, @@ -157913,67 +158078,63 @@ "component_index": 1, "component_count": 266, "review_unit_index": 173, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10188", "pull_request:10318", - "pull_request:10324", "pull_request:10360", "pull_request:10378", - "pull_request:10630", "pull_request:10719", "pull_request:10921", "pull_request:11087", - "pull_request:11124", "pull_request:11196", "pull_request:11319", "pull_request:11421", - "pull_request:11469", "pull_request:11479", "pull_request:11504", "pull_request:12099", - "pull_request:12277", "pull_request:12363", "pull_request:12397", "pull_request:12491", "pull_request:12493", "pull_request:12687", - "pull_request:1277", "pull_request:13029", "pull_request:13119", "pull_request:13204", "pull_request:1331", - "pull_request:1579", "pull_request:1968", "pull_request:2467", "pull_request:3023", "pull_request:3297", "pull_request:3446", "pull_request:349", + "pull_request:3607", "pull_request:3673", + "pull_request:3708", + "pull_request:3804", + "pull_request:3847", + "pull_request:3912", "pull_request:3957", "pull_request:4040", "pull_request:4596", "pull_request:5057", "pull_request:5653", "pull_request:5754", + "pull_request:5948", + "pull_request:6144", + "pull_request:6150", "pull_request:833", "pull_request:8691", "pull_request:8964", - "pull_request:9206", "pull_request:9878" ], "soft_pairs": [ - "pull_request:10324|pull_request:9206", - "pull_request:10630|pull_request:11124", - "pull_request:11469|pull_request:12277", - "pull_request:1277|pull_request:1579", "pull_request:5057|pull_request:5653", "pull_request:10188|pull_request:8691", "pull_request:10318|pull_request:10719", @@ -157993,180 +158154,184 @@ "pull_request:3023|pull_request:3957", "pull_request:3023|pull_request:833", "pull_request:3297|pull_request:4596", - "pull_request:3446|pull_request:5754" + "pull_request:3446|pull_request:5754", + "pull_request:3607|pull_request:3804", + "pull_request:3708|pull_request:5948", + "pull_request:3847|pull_request:6150", + "pull_request:3912|pull_request:6144" ], - "prepared_review_unit_hash": "c6658df85cfa84fcf5529381730a3712bb9023a1c2625da29153d1ba17498d9c", + "prepared_review_unit_hash": "297f21d7ebf9c63c4c14a0f01d93deefdb658c26653a223fc7f4971e7f6bbfbd", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 98030, - "estimated_input_tokens": 24508, - "estimated_eval_tokens": 49272 + "serialized_chars": 97525, + "estimated_input_tokens": 24382, + "estimated_eval_tokens": 49020 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Broad false-positive cluster: the items span unrelated docs, CI, tests, training, LoRA, modular, and model-feature PRs. The suggested soft edges are superficial area overlaps, not the same underlying bug or change.", + "summary": "No soft duplicate pair looks like the same underlying bug/change. Most pairs only share a broad subsystem or theme (LoRA, Wan, docs, CI, attention), but the concrete edits and intent differ, so this cluster should not be merged into a single canonical PR.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: this is not one change set, but a heterogeneous mix of unrelated PRs across different subsystems and time periods.", + "canonical_pr_reason": "No single PR is a safe canonical representative here: the items are unrelated fixes/features across different code paths and docs, not one duplicate change.", "best_issue_reason": null, - "best_pr_reason": "No strong global best PR exists because the cluster is not a duplicate group. If forced, 10921 is the largest standalone feature PR, but it is not representative of the rest.", + "best_pr_reason": "No PR stands out as a global best duplicate target because the cluster is not one coherent topic.", "soft_edge_verdicts": [ - { - "left": "pull_request:10324", - "right": "pull_request:9206", - "accept": false, - "reason": "Both touch tests around pipeline behavior, but one is Kandinsky CPU offload and the other is xformers test handling across other pipelines; different bugs and code paths." - }, - { - "left": "pull_request:10630", - "right": "pull_request:11124", - "accept": false, - "reason": "Unrelated test updates: one adjusts Flax scheduler slice expectations, the other changes a CogView4 tokenizer fixture." - }, - { - "left": "pull_request:11469", - "right": "pull_request:12277", - "accept": false, - "reason": "Both are test skips/xfails, but they target different pipelines and failure modes; not the same underlying issue." - }, - { - "left": "pull_request:1277", - "right": "pull_request:1579", - "accept": false, - "reason": "Both mention documentation/examples, but one is a publish-doc workflow and the other adds research-project examples; no shared bug or change." - }, { "left": "pull_request:5057", "right": "pull_request:5653", "accept": false, - "reason": "SDXL clip-skip support and PEFT-aware warning/unwrapping are unrelated feature fixes in different code paths." + "reason": "Different changes: SDXL clip-skip support vs PEFT-aware model unwrapping in pipeline loading." }, { "left": "pull_request:10188", "right": "pull_request:8691", "accept": false, - "reason": "Both concern LoRA loading, but one fixes Flux shape expansion and the other changes vanilla fine-tuned LoRA loading logic; not mergeable into one PR." + "reason": "Both touch LoRA loading, but one fixes mixed LoRA expansion handling in Flux and the other changes vanilla fine-tuned LoRA loading logic." }, { "left": "pull_request:10318", "right": "pull_request:10719", "accept": false, - "reason": "A Flux LoRA integration test and ONNX provider_options plumbing are unrelated." + "reason": "Integration tests for Flux LoRA expansion are unrelated to adding provider_options to from_pretrained." }, { "left": "pull_request:10360", "right": "pull_request:8964", "accept": false, - "reason": "CI release-workflow install URL fix and AudioLDM2 generation cache-position fix are unrelated." + "reason": "Release workflow TestPyPI URL fix is unrelated to AudioLDM2 cache-position generation fix." }, { "left": "pull_request:10378", "right": "pull_request:11319", "accept": false, - "reason": "DreamBooth LoRA training support and a docs/community entry are unrelated changes." + "reason": "Dreambooth LoRA training support is unrelated to adding a community-project docs entry." }, { "left": "pull_request:10921", "right": "pull_request:11087", "accept": false, - "reason": "Both are Wan-related, but one adds Wan 2.1 support and the other fixes Wan I2V latent handling; different scope and code paths." + "reason": "Wan 2.1 support is a large feature PR; the other is a small Wan I2V quality fix in one function." }, { "left": "pull_request:11196", "right": "pull_request:12363", "accept": false, - "reason": "Safe-serialization save_pretrained behavior and QwenImageEdit Plus docs/tests are unrelated." + "reason": "Safe-serialization behavior in ModelMixin is unrelated to QwenImageEdit Plus docs/tests." }, { "left": "pull_request:11421", "right": "pull_request:9878", "accept": false, - "reason": "Dreambooth interpolation default and gradient-checkpointing eval-mode changes are unrelated training/model behaviors." + "reason": "Interpolation default in dreambooth training is unrelated to gradient-checkpointing behavior across eval mode." }, { "left": "pull_request:11479", "right": "pull_request:11504", "accept": false, - "reason": "Docs rename for Intel Gaudi and dependency/workflow updates are not the same change." + "reason": "Intel Gaudi doc rename is unrelated to dependency-table and CI trigger updates." }, { "left": "pull_request:12099", "right": "pull_request:12687", "accept": false, - "reason": "Two typo fixes in different subsystems (Qwen image pipeline vs autoencoder_dc); no shared bug." + "reason": "Qwen device placement typo fix is unrelated to an autoencoder_dc typo rename." }, { "left": "pull_request:12397", "right": "pull_request:12491", "accept": false, - "reason": "Both are CI/workflow changes, but one disables transformers-from-main installs and the other adds doc link checking; distinct purposes." + "reason": "CI disabling main-branch installs is unrelated to adding a documentation link-check job." }, { "left": "pull_request:12493", "right": "pull_request:13119", "accept": false, - "reason": "LTX-Video docs/pipeline work and Flux2 LoRA key conversion are unrelated model-family changes." + "reason": "LTX-Video 0.9.8 docs/code updates are unrelated to Flux2 LoRA key conversion handling." }, { "left": "pull_request:13029", "right": "pull_request:13204", "accept": false, - "reason": "Modular quickstart docs and modular trust_remote_code/component-loading logic are different changes, not duplicates." + "reason": "Modular quickstart docs are unrelated to modular pipeline trust_remote_code handling." }, { "left": "pull_request:1331", "right": "pull_request:349", "accept": false, - "reason": "An inpainting scheduler-order bugfix and DDPMScheduler type hints are unrelated." + "reason": "Inpainting scheduler-order fix is unrelated to DDPM type-hint cleanup." }, { "left": "pull_request:1968", "right": "pull_request:3673", "accept": false, - "reason": "CPU offload behavior/docs and multi-GPU DreamBooth fixes are different issues." + "reason": "CPU offload doc/correction work is unrelated to Dreambooth multi-GPU config handling." }, { "left": "pull_request:2467", "right": "pull_request:4040", "accept": false, - "reason": "Latent output-type support and _use_default_values list typing are unrelated." + "reason": "Latent output type feature PR is unrelated to keeping _use_default_values as a list." }, { "left": "pull_request:3023", "right": "pull_request:3957", "accept": false, - "reason": "Both involve attention code, but one changes head/batch reshaping behavior and the other is an unfinished attn-name correction; not the same concrete fix." + "reason": "Both are attention-related, but one adds KV attention processor behavior and the other is an attn-name correction WIP; different code paths and goals." }, { "left": "pull_request:3023", "right": "pull_request:833", "accept": false, - "reason": "Attention processor changes and a version bump release PR are unrelated." + "reason": "Attention/KV processor work is unrelated to a release-version bump." }, { "left": "pull_request:3297", "right": "pull_request:4596", "accept": false, - "reason": "Retrieval-augmented diffusion examples and SDXL conversion scripts are unrelated project additions." + "reason": "Retrieval-augmented diffusion models project code is unrelated to an SDXL checkpoint conversion script." }, { "left": "pull_request:3446", "right": "pull_request:5754", "accept": false, - "reason": "Pipeline loading warning text fix and general style cleanup are unrelated." + "reason": "Pipeline loading warning-message cleanup is unrelated to styling fixes in example training scripts." + }, + { + "left": "pull_request:3607", + "right": "pull_request:3804", + "accept": false, + "reason": "Both mention attention, but one fixes temb attention and the other changes legacy attention block weight conversion for device maps." + }, + { + "left": "pull_request:3708", + "right": "pull_request:5948", + "accept": false, + "reason": "Custom release-version sorting is unrelated to from_pretrained variant warning logic." + }, + { + "left": "pull_request:3847", + "right": "pull_request:6150", + "accept": false, + "reason": "Soft-dependency tests/CI setup are unrelated to IP-Adapter converter implementation." + }, + { + "left": "pull_request:3912", + "right": "pull_request:6144", + "accept": false, + "reason": "Dreambooth LoRA text-encoder finetuning feature is unrelated to unwrapped UNet LoRA state saving." } ] }, @@ -158182,17 +158347,23 @@ "component_index": 1, "component_count": 266, "review_unit_index": 174, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10086", "pull_request:10096", + "pull_request:10119", + "pull_request:10132", + "pull_request:10188", + "pull_request:10342", + "pull_request:10544", "pull_request:10922", + "pull_request:11243", "pull_request:12109", "pull_request:12215", "pull_request:12416", @@ -158200,11 +158371,6 @@ "pull_request:13352", "pull_request:3001", "pull_request:3128", - "pull_request:3607", - "pull_request:3708", - "pull_request:3804", - "pull_request:3847", - "pull_request:3912", "pull_request:4352", "pull_request:4417", "pull_request:4855", @@ -158214,9 +158380,6 @@ "pull_request:5419", "pull_request:5771", "pull_request:5938", - "pull_request:5948", - "pull_request:6144", - "pull_request:6150", "pull_request:6262", "pull_request:6263", "pull_request:6264", @@ -158234,16 +158397,14 @@ "pull_request:8496", "pull_request:8618", "pull_request:8627", + "pull_request:9060", "pull_request:9175", "pull_request:9569", "pull_request:9918", + "pull_request:9938", "pull_request:9985" ], "soft_pairs": [ - "pull_request:3607|pull_request:3804", - "pull_request:3708|pull_request:5948", - "pull_request:3847|pull_request:6150", - "pull_request:3912|pull_request:6144", "pull_request:4855|pull_request:6904", "pull_request:5771|pull_request:6262", "pull_request:5771|pull_request:6263", @@ -158265,192 +158426,196 @@ "pull_request:12109|pull_request:12416", "pull_request:12215|pull_request:12416", "pull_request:10086|pull_request:10922", - "pull_request:10096|pull_request:8627" + "pull_request:10096|pull_request:8627", + "pull_request:10119|pull_request:10544", + "pull_request:10132|pull_request:9060", + "pull_request:10188|pull_request:11243", + "pull_request:10342|pull_request:9938" ], - "prepared_review_unit_hash": "fe253d67134473eed10f665714f23bf98fedfbcd9af309bb49060634088b80e1", + "prepared_review_unit_hash": "29fdd1a6f55ef8dc7d4a6370a5e3db034d1b8993cf14f78a2aca82e6fbe333c0", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 105313, - "estimated_input_tokens": 26329, - "estimated_eval_tokens": 52914 + "serialized_chars": 104279, + "estimated_input_tokens": 26070, + "estimated_eval_tokens": 52396 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: the soft pairs mostly share broad subsystems or filenames, but they fix different bugs, add different features, or touch unrelated docs/tests/workflows. No duplicate groups are supported by the content.", - "confidence": 0.98, + "summary": "The soft-edge candidates are mostly false positives: they share broad subsystems or adjacent themes, but they do not fix the same concrete bug or implement the same change. No issue is present, and there is no strong canonical PR anchor; the Flux/ControlNet pair is the closest on surface area but still too broad and different.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items span many unrelated changes across pipelines, schedulers, docs, workflows, and examples, so there is no single representative duplicate target.", + "canonical_pr_reason": "No single PR cleanly represents a duplicate cluster here; the items span unrelated fixes/features, so there is no strong canonical PR anchor.", "best_issue_reason": null, - "best_pr_reason": "No global best PR stands out; the cluster is too mixed for a meaningful representative artifact.", + "best_pr_reason": "PR 9985 is the broadest and most central feature PR in the set, but it is not a good duplicate representative because it bundles multiple Flux features rather than one specific fix.", "soft_edge_verdicts": [ - { - "left": "pull_request:3607", - "right": "pull_request:3804", - "accept": false, - "reason": "Same file, but different fixes: temb attention handling vs legacy attention block weight conversion. Not the same bug." - }, - { - "left": "pull_request:3708", - "right": "pull_request:5948", - "accept": false, - "reason": "Version sorting for custom releases vs a variant-loading warning fix. Unrelated behavior." - }, - { - "left": "pull_request:3847", - "right": "pull_request:6150", - "accept": false, - "reason": "Test/workflow addition for soft dependencies vs IP-Adapter conversion helper. Different changes." - }, - { - "left": "pull_request:3912", - "right": "pull_request:6144", - "accept": false, - "reason": "Text-encoder finetuning feature vs checkpoint-save unwrapping in the LoRA script. Same script family, not the same change." - }, { "left": "pull_request:4855", "right": "pull_request:6904", "accept": false, - "reason": "Offload-hook handling vs making load_image accept PIL images. Different code paths and bugs." + "reason": "Unrelated fixes: CPU offload/state-hook handling vs PIL image loading support. Same repo area, different code paths." }, { "left": "pull_request:5771", "right": "pull_request:6262", "accept": false, - "reason": "Loader refactor vs moving Panorama out of stable_diffusion. Not the same underlying change." + "reason": "Refactor of loaders vs moving Panorama out of stable_diffusion; different subsystems and different changes." }, { "left": "pull_request:5771", "right": "pull_request:6263", "accept": false, - "reason": "Loader refactor vs moving LDM3D out of stable_diffusion. Unrelated." + "reason": "Refactor of loaders vs moving LDM3D out of stable_diffusion; not the same bug or change." }, { "left": "pull_request:5771", "right": "pull_request:6264", "accept": false, - "reason": "Loader refactor vs moving SAG out of stable_diffusion. Unrelated." + "reason": "Refactor of loaders vs moving SAG out of stable_diffusion; unrelated code paths." }, { "left": "pull_request:6432", "right": "pull_request:7216", "accept": false, - "reason": "README learning-rate tweak vs from_single_file signature update. Different scope." + "reason": "README learning-rate update vs from_single_file signature/documentation change. No shared concrete fix." }, { "left": "pull_request:7044", "right": "pull_request:7662", "accept": false, - "reason": "UNet refactor/re-addition vs removal of a redundant flax-test install command. Different changes." + "reason": "Unet refactor restoration vs workflow cleanup in flax tests; unrelated changes." }, { "left": "pull_request:7455", "right": "pull_request:7662", "accept": false, - "reason": "Distutils removal vs workflow cleanup. No shared bug or change." + "reason": "Distutils removal vs removing redundant packages in CI. Same general maintenance theme, but not the same change." }, { "left": "pull_request:7738", "right": "pull_request:8496", "accept": false, - "reason": "LoRA device-placement fix vs SD3 warning-log gating. Unrelated." + "reason": "LoRA device-placement fix vs SD3 warning-log guard. Different bugs and code paths." }, { "left": "pull_request:8467", "right": "pull_request:9569", "accept": false, - "reason": "SEGA pipeline device fix vs CogVideoX autoencoder decode fix. Different pipelines and failures." + "reason": "SEGA pipeline device fix vs CogVideoX autoencoder decode fix; unrelated models and failures." }, { "left": "pull_request:8618", "right": "pull_request:9918", "accept": false, - "reason": "SD3 tiny autoencoder shift_factor change vs adding a server example. Different artifacts and purpose." + "reason": "SD3 tiny autoencoder shift_factor change vs adding a server example; totally different deliverables." }, { "left": "pull_request:9175", "right": "pull_request:9985", "accept": false, - "reason": "Both touch Flux, but one adds union ControlNet support while the other adds Fill/Canny/Depth/Redux variants. Not the same concrete feature." + "reason": "Both touch Flux, but one adds Union ControlNet support while the other adds Fill/Canny/Depth/Redux variants. Broadly related, but not the same concrete feature or mergeable single PR." }, { "left": "pull_request:5938", "right": "pull_request:6975", "accept": false, - "reason": "ControlNet LoRA support draft vs single-file ControlNet inpaint fix. Different functionality." + "reason": "ControlNet LoRA support vs single-file ControlNet inpaint loading fix. Different functionality and code paths." }, { "left": "pull_request:12519", "right": "pull_request:13352", "accept": false, - "reason": "Toctree sorting fix vs Claude review workflow checkout step. Unrelated CI/docs changes." + "reason": "Toctree sorting fix vs adding a checkout step to Claude review workflow; unrelated CI/docs changes." }, { "left": "pull_request:4352", "right": "pull_request:4417", "accept": false, - "reason": "Kandinsky combined pipeline fixes vs Zero123 community pipeline support. Different models and goals." + "reason": "Kandinsky combined pipeline fixes vs Zero123 community pipeline support; unrelated features." }, { "left": "pull_request:5361", "right": "pull_request:5419", "accept": false, - "reason": "Temporary PEFT integration testing branch vs the actual push-test addition. Not the same mergeable change." + "reason": "Temporary PEFT integration testing PR vs adding PEFT CUDA slow tests on push. Same testing domain, but not the same change." }, { "left": "pull_request:6303", "right": "pull_request:7359", "accept": false, - "reason": "ADD training scripts vs ResAdapter training script. Different research projects." + "reason": "ADD training script vs ResAdapter training script; both research examples, but different projects and code." }, { "left": "pull_request:3001", "right": "pull_request:4986", "accept": false, - "reason": "Karras support for DPMSolverMultistepScheduler vs a broader sigma-based refactor across schedulers. Related area, but not the same change." + "reason": "Karras support in one scheduler vs a broad sigma-based refactor across multiple schedulers. Overlapping area, but not the same underlying change." }, { "left": "pull_request:3128", "right": "pull_request:5261", "accept": false, - "reason": "Correct img2img timesteps vs fixing callback step indexing. Same subsystem, different bug." + "reason": "Img2img timestep selection fix vs callback step indexing fix. Both scheduler-adjacent, but distinct bugs." }, { "left": "pull_request:12109", "right": "pull_request:12416", "accept": false, - "reason": "UniPC scheduler custom sigmas/timesteps vs QwenImage Edit Plus modular pipeline support. Unrelated." + "reason": "UniPC custom sigmas/timesteps support vs QwenImage Edit Plus modular pipeline support; unrelated features." }, { "left": "pull_request:12215", "right": "pull_request:12416", "accept": false, - "reason": "Qwen-Image ControlNet support vs QwenImage Edit Plus modular support. Same product family, but different features and code paths." + "reason": "Qwen-Image ControlNet support vs QwenImage Edit Plus modular support. Same product family, but different concrete capabilities." }, { "left": "pull_request:10086", "right": "pull_request:10922", "accept": false, - "reason": "StableDiffusion3 skip-guidance bugfix vs Wan2.1 continuation work. Unrelated." + "reason": "SD3 skip-guidance bug fix vs Wan2.1 continuation PR; unrelated." }, { "left": "pull_request:10096", "right": "pull_request:8627", "accept": false, - "reason": "Stale-bot exemption change vs SD3 ControlNet fast-test PR. Different maintenance areas." + "reason": "Stale-bot label exemption vs SD3 ControlNet test trimming. No shared change." + }, + { + "left": "pull_request:10119", + "right": "pull_request:10544", + "accept": false, + "reason": "Docs for load_lora_adapter vs ParaAttention docs. Both documentation-only, but unrelated topics." + }, + { + "left": "pull_request:10132", + "right": "pull_request:9060", + "accept": false, + "reason": "Tokenizers CI update vs removing a debug print in transformer_flux. Different concerns and code paths." + }, + { + "left": "pull_request:10188", + "right": "pull_request:11243", + "accept": false, + "reason": "Flux LoRA error handling/test vs Wan LoRA conversion support. Both LoRA-related, but they fix different model families and behaviors." + }, + { + "left": "pull_request:10342", + "right": "pull_request:9938", + "accept": false, + "reason": "get_parameter_dtype behavior update vs SDXL reference pipeline docs/example update. Unrelated." } ] }, @@ -158466,7 +158631,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 175, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -158474,18 +158639,12 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10119", - "pull_request:10132", - "pull_request:10188", - "pull_request:10342", "pull_request:10351", - "pull_request:10544", "pull_request:10547", "pull_request:10557", "pull_request:10810", "pull_request:10894", "pull_request:11059", - "pull_request:11243", "pull_request:11252", "pull_request:11743", "pull_request:11813", @@ -158510,24 +158669,26 @@ "pull_request:2101", "pull_request:225", "pull_request:2314", + "pull_request:2470", "pull_request:2529", + "pull_request:2553", + "pull_request:2637", + "pull_request:2730", "pull_request:2734", + "pull_request:3245", + "pull_request:3398", "pull_request:3614", "pull_request:3766", + "pull_request:4188", + "pull_request:889", "pull_request:8964", - "pull_request:9060", "pull_request:908", "pull_request:957", "pull_request:9676", "pull_request:9684", - "pull_request:9687", - "pull_request:9938" + "pull_request:9687" ], "soft_pairs": [ - "pull_request:10119|pull_request:10544", - "pull_request:10132|pull_request:9060", - "pull_request:10188|pull_request:11243", - "pull_request:10342|pull_request:9938", "pull_request:10351|pull_request:10557", "pull_request:10547|pull_request:8964", "pull_request:10557|pull_request:9676", @@ -158549,192 +158710,210 @@ "pull_request:1951|pull_request:225", "pull_request:1965|pull_request:908", "pull_request:2101|pull_request:2734", - "pull_request:2314|pull_request:3614" + "pull_request:2314|pull_request:3614", + "pull_request:2470|pull_request:4188", + "pull_request:2553|pull_request:889", + "pull_request:2637|pull_request:3245", + "pull_request:2637|pull_request:3398", + "pull_request:2730|pull_request:3245", + "pull_request:2730|pull_request:3398" ], - "prepared_review_unit_hash": "2bc27d37be836f237183bdfa2acb4b97910c3d4bad7aa95e9de094c3966dea59", + "prepared_review_unit_hash": "32ce370ca94c5130e0be4829bf0abdc24c22302abe7b3e5d77042fdba5aa6ae7", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 26, - "serialized_chars": 96031, - "estimated_input_tokens": 24008, - "estimated_eval_tokens": 48272 + "soft_pair_count": 28, + "serialized_chars": 97523, + "estimated_input_tokens": 24381, + "estimated_eval_tokens": 49018 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a duplicate set; it mixes unrelated docs, CI, LoRA conversion, modular pipeline, and test-only PRs across many months/years.", - "confidence": 0.18, + "summary": "This set is a grab bag of unrelated PRs across tests, docs, CI, training scripts, and loader/model fixes. None of the soft pairs look like the same underlying change, so all should be rejected as duplicates.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: there is no single underlying bug or change tying these PRs together.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is a good cluster representative because the items are heterogeneous and touch different subsystems and code paths.", + "best_pr_reason": "pull_request:10810 is the strongest standalone code fix in the set: it updates LoRA conversion logic in src/diffusers/loaders/lora_conversion_utils.py for Flux support, making it the most substantive and representative PR here.", "soft_edge_verdicts": [ - { - "left": "pull_request:10119", - "right": "pull_request:10544", - "accept": false, - "reason": "Different docs topics: LoRA adapter loading vs ParaAttention optimization docs." - }, - { - "left": "pull_request:10132", - "right": "pull_request:9060", - "accept": false, - "reason": "Unrelated changes: CI dependency install vs removing a debug print in FLUX transformer code." - }, - { - "left": "pull_request:10188", - "right": "pull_request:11243", - "accept": false, - "reason": "Both LoRA-related, but one is Flux loader error handling and the other adds Musubi Wan LoRA conversion support." - }, - { - "left": "pull_request:10342", - "right": "pull_request:9938", - "accept": false, - "reason": "Unrelated: parameter dtype utility vs SDXL reference pipeline docs/example updates." - }, { "left": "pull_request:10351", "right": "pull_request:10557", "accept": false, - "reason": "Different test fixes for different areas; not the same underlying bug." + "reason": "Different test targets: LoRA NaN-fusing test cleanup vs CogVideoX/CogView3 offload test tuning." }, { "left": "pull_request:10547", "right": "pull_request:8964", "accept": false, - "reason": "Offload test tuning for CogVideoX/CogView3 is unrelated to AudioLDM2 cache-position generation." + "reason": "Unrelated subsystems: transformer offload test settings vs AudioLDM2 generation cache-position fix." }, { "left": "pull_request:10557", "right": "pull_request:9676", "accept": false, - "reason": "Single-file test fixture change vs import-path cleanup; no shared bug." + "reason": "One is a LoRA test helper tweak; the other is a CI import-path cleanup for a benchmark script." }, { "left": "pull_request:10810", "right": "pull_request:9684", "accept": false, - "reason": "Different LoRA/Wan changes vs VAE framewise batching in CogVideoX." + "reason": "Different fixes: Flux LoRA conversion support vs CogVideoX VAE batching behavior." }, { "left": "pull_request:10810", "right": "pull_request:9687", "accept": false, - "reason": "Flux LoRA conversion support is unrelated to adding Xlabs ControlNet support." + "reason": "Both are model-related but not the same change: Flux LoRA conversion vs Xlabs ControlNet support." }, { "left": "pull_request:10894", "right": "pull_request:9676", "accept": false, - "reason": "Docs toctree sorting vs benchmark import cleanup are unrelated." + "reason": "Docs toctree sorting is unrelated to the benchmark import fix." }, { "left": "pull_request:11059", "right": "pull_request:12315", "accept": false, - "reason": "Wan LoRA conversion support and scheduler-test simplification are different changes." + "reason": "Non-diffusers WAN LoRA conversion support is unrelated to pruning scheduler setup in LoRA tests." }, { "left": "pull_request:11252", "right": "pull_request:12424", "accept": false, - "reason": "Timeout constant fix in a utility script vs broad Dockerfile/workflow definitions cleanup." + "reason": "Timeout constant fix in a utility script is unrelated to Dockerfile definition cleanup." }, { "left": "pull_request:11743", "right": "pull_request:11813", "accept": false, - "reason": "Both Flux training-related, but one is a shortened research script and the other is Kontext training support; not the same change." + "reason": "Different training artifacts: a shortened Flux research script vs Kontext training implementation and docs." }, { "left": "pull_request:11894", "right": "pull_request:12450", "accept": false, - "reason": "Both touch dependency installation, but one pins k-diffusion for CI and the other revisits many CI/Docker installs broadly." + "reason": "Both touch CI/dependencies, but one pins k-diffusion and the other broadly revises CI installation flow." }, { "left": "pull_request:12018", "right": "pull_request:12231", "accept": false, - "reason": "Docs link text fix vs typo cleanup across examples are unrelated." + "reason": "A docs link change is not the same as a typo fix across community example docs." }, { "left": "pull_request:12315", "right": "pull_request:13119", "accept": false, - "reason": "Scheduler-test consolidation is unrelated to Flux2 LoRA key conversion handling." + "reason": "LoRA test simplification and Flux2 LoRA key conversion address different code paths and failure modes." }, { "left": "pull_request:13120", "right": "pull_request:13204", "accept": false, - "reason": "Both modular-pipeline related, but one adds tests and the other changes update_components behavior." + "reason": "Shared modular-pipeline test file aside, one adds AutoModel load-id tests while the other changes trust_remote_code handling in pipeline code." }, { "left": "pull_request:13338", "right": "pull_request:13536", "accept": false, - "reason": "Same workflow file, but different permission changes and different workflow behavior; not one concrete fix." + "reason": "Same workflow file, but distinct permission changes and workflow behavior; not one mergeable bug fix." }, { "left": "pull_request:1414", "right": "pull_request:957", "accept": false, - "reason": "Clip-guided cutout sizing fix is unrelated to the text-to-image scheduler argument change." + "reason": "Very old clip-guided pipeline fix vs a training-scheduler bug fix in text-to-image; unrelated." }, { "left": "pull_request:1415", "right": "pull_request:2529", "accept": false, - "reason": "Broad Stable Diffusion fixes vs open-clip checkpoint conversion; different code paths and bugs." + "reason": "Stable Diffusion docs/code cleanup across many files is unrelated to the specific SD checkpoint conversion bug." }, { "left": "pull_request:1468", "right": "pull_request:1494", "accept": false, - "reason": "Saving trained betas is unrelated to adding safetensors serialization support." + "reason": "Saving trained betas in schedulers is unrelated to adding safetensor serialization support." }, { "left": "pull_request:1945", "right": "pull_request:3766", "accept": false, - "reason": "UNet block tests are unrelated to Kandinsky UNet conversion script updates." + "reason": "UNet block tests do not match a Kandinsky UNet conversion-script update." }, { "left": "pull_request:1951", "right": "pull_request:225", "accept": false, - "reason": "ColossalAI integration update and README release notes are unrelated." + "reason": "ColossalAI dependency update is unrelated to README release-note edits." }, { "left": "pull_request:1965", "right": "pull_request:908", "accept": false, - "reason": "Checkpoint merger utility vs mega inpaint pipeline routing are different changes." + "reason": "Checkpoint merger cleanup is unrelated to swapping the inpaint pipeline in the community mega script." }, { "left": "pull_request:2101", "right": "pull_request:2734", "accept": false, - "reason": "InstructPix2Pix docs/example update is unrelated to Habana documentation numbers." + "reason": "Pix2Pix docs/example update is unrelated to Habana documentation number updates." }, { "left": "pull_request:2314", "right": "pull_request:3614", "accept": false, - "reason": "Both are test tweaks, but they address different pipelines and different expected-slice issues." + "reason": "Both are test-related, but one adjusts numerical expectations and the other touches a renamed community script and determinism in a different test." + }, + { + "left": "pull_request:2470", + "right": "pull_request:4188", + "accept": false, + "reason": "MPS test skips and SDXL ControlNet dataset fingerprinting are different fixes in different areas." + }, + { + "left": "pull_request:2553", + "right": "pull_request:889", + "accept": false, + "reason": "CFG flag addition in Flax pipelines is unrelated to adding a components accessor for Stable Diffusion." + }, + { + "left": "pull_request:2637", + "right": "pull_request:3245", + "accept": false, + "reason": "Quicktour typo cleanup is unrelated to Chinese docs navigation/tree updates." + }, + { + "left": "pull_request:2637", + "right": "pull_request:3398", + "accept": false, + "reason": "Both are doc typos, but they fix different files and different wording; not the same underlying change." + }, + { + "left": "pull_request:2730", + "right": "pull_request:3245", + "accept": false, + "reason": "English training-doc newline/grammar cleanup is unrelated to the zh translated docs overhaul." + }, + { + "left": "pull_request:2730", + "right": "pull_request:3398", + "accept": false, + "reason": "These are separate documentation typos in different pages with no shared bug or change." } ] }, @@ -158750,7 +158929,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 176, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -158758,25 +158937,19 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2470", - "pull_request:2553", - "pull_request:2637", - "pull_request:2730", "pull_request:2915", "pull_request:2919", "pull_request:3003", "pull_request:3037", "pull_request:3047", "pull_request:3241", - "pull_request:3245", "pull_request:3370", - "pull_request:3398", "pull_request:3536", "pull_request:3568", "pull_request:3675", + "pull_request:3691", "pull_request:3850", "pull_request:4058", - "pull_request:4188", "pull_request:4211", "pull_request:4319", "pull_request:4564", @@ -158798,22 +158971,22 @@ "pull_request:564", "pull_request:5672", "pull_request:5699", + "pull_request:5707", + "pull_request:6049", + "pull_request:6065", "pull_request:6075", "pull_request:6163", "pull_request:6215", "pull_request:6305", "pull_request:6450", + "pull_request:6588", + "pull_request:6997", + "pull_request:7464", "pull_request:748", - "pull_request:7679", - "pull_request:889" + "pull_request:7679" ], "soft_pairs": [ - "pull_request:2470|pull_request:4188", - "pull_request:2553|pull_request:889", - "pull_request:2637|pull_request:3245", - "pull_request:2637|pull_request:3398", - "pull_request:2730|pull_request:3245", - "pull_request:2730|pull_request:3398", + "pull_request:2915|pull_request:3691", "pull_request:2915|pull_request:5053", "pull_request:2919|pull_request:3037", "pull_request:3003|pull_request:4058", @@ -158833,192 +159006,183 @@ "pull_request:5275|pull_request:6305", "pull_request:5506|pull_request:7679", "pull_request:5626|pull_request:6163", - "pull_request:5672|pull_request:6215" + "pull_request:5672|pull_request:6215", + "pull_request:5707|pull_request:6997", + "pull_request:6049|pull_request:7464", + "pull_request:6065|pull_request:6588" ], - "prepared_review_unit_hash": "be0e104aea1b83f8bba80d22731d4719768a4876414610cb77e92e078f250b75", + "prepared_review_unit_hash": "8bc81a82cfc011c3d9b1aa3989de392b627ba97d59cfc8380cb2b778de9337d2", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 26, - "serialized_chars": 95513, - "estimated_input_tokens": 23879, - "estimated_eval_tokens": 48014 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 93137, + "estimated_input_tokens": 23285, + "estimated_eval_tokens": 46826 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated small PRs (docs, tests, workflow tweaks, and a few distinct code fixes). I don\u2019t see one underlying bug/change that ties them together, so I would not merge any of the soft-edge pairs.", - "confidence": 0.93, + "summary": "This cluster is a grab bag of mostly unrelated PRs that share only superficial signals like \u201cfix\u201d, tests, docs, or common subsystems such as SDXL/LoRA. There is no single underlying bug or change tying them together, so the soft duplicate edges should be rejected.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR cleanly represents a duplicate cluster here; the items span many unrelated fixes. If one must be chosen as the most substantive core-code change, PR 3003 is the best candidate because it fixes a concrete scheduler bug across multiple implementations and includes tests/docs, but it is not a duplicate anchor for the rest of the set.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 3003 is the strongest standalone representative of a real code-path fix: it corrects dynamic thresholding behavior in several schedulers and adds coverage. It is broader and more canonical than the mostly test/doc/metadata-only PRs, though it still does not unify the cluster.", "soft_edge_verdicts": [ { - "left": "pull_request:2470", - "right": "pull_request:4188", - "accept": false, - "reason": "Both are SD-related maintenance, but one skips MPS tests and the other fixes SDXL ControlNet embedding caching; different code paths and different problems." - }, - { - "left": "pull_request:2553", - "right": "pull_request:889", - "accept": false, - "reason": "Both touch pipeline utilities/config behavior, but 2553 adds classifier-free guidance logic in Flax Stable Diffusion while 889 adds a components API; not the same fix." - }, - { - "left": "pull_request:2637", - "right": "pull_request:3245", - "accept": false, - "reason": "Both are docs edits, but one is an English quicktour wording fix and the other is a large Chinese docs translation/update; unrelated changes." - }, - { - "left": "pull_request:2637", - "right": "pull_request:3398", - "accept": false, - "reason": "Both are minor docs fixes, but they correct different files and different typos; no shared bug or change." - }, - { - "left": "pull_request:2730", - "right": "pull_request:3245", - "accept": false, - "reason": "One is an English training-doc grammar fix; the other is a broader translated docs update. Same area, not same underlying issue." - }, - { - "left": "pull_request:2730", - "right": "pull_request:3398", + "left": "pull_request:2915", + "right": "pull_request:3691", "accept": false, - "reason": "Both are docs typos, but they fix separate pages and separate wording mistakes." + "reason": "One is a test threshold relaxation; the other changes LoRA text-encoder processor selection. Different code paths and different problems." }, { "left": "pull_request:2915", "right": "pull_request:5053", "accept": false, - "reason": "One relaxes a test tolerance for textual inversion; the other changes a CLI argument shape for SDXL validation images. Different subsystems and failures." + "reason": "A slow-test tweak and a validation_image argument fix are unrelated changes." }, { "left": "pull_request:2919", "right": "pull_request:3037", "accept": false, - "reason": "Both are test-related, but 2919 speeds up a 3D UNet test while 3037 skips an unstable MPS test in Stable UnCLIP img2img; not the same concrete issue." + "reason": "Both touch tests, but one speeds up a UNet3D test and the other skips an unstable MPS test in Stable UnCLIP. Not the same bug." }, { "left": "pull_request:3003", "right": "pull_request:4058", "accept": false, - "reason": "3003 fixes dynamic thresholding logic across schedulers; 4058 only changes SDXL documentation rendering. No overlap beyond broad SDXL/Diffusers context." + "reason": "Scheduler dynamic-thresholding fixes are unrelated to SDXL documentation updates." }, { "left": "pull_request:3047", "right": "pull_request:3241", "accept": false, - "reason": "3047 fixes workflow report naming, while 3241 loosens an UnCLIP attention-slicing test. Different artifact types and problems." + "reason": "Fixing report names in CI workflows is unrelated to loosening an UnCLIP test assertion." }, { "left": "pull_request:3370", "right": "pull_request:3568", "accept": false, - "reason": "Both are docs updates, but 3370 revises PyTorch 2.0 optimization docs and 3568 fixes broken links in basic training docs; unrelated content." + "reason": "Both are docs PRs, but they fix different pages and different link/text issues." }, { "left": "pull_request:3536", "right": "pull_request:3675", "accept": false, - "reason": "3536 fixes image normalization in DeepFloyd IF super-resolution pipelines; 3675 is a test fix for UniDiffuser on V100. Different code paths and goals." + "reason": "DeepFloyd IF PIL normalization and a UniDiffuser test skip are unrelated." }, { "left": "pull_request:3850", "right": "pull_request:5396", "accept": false, - "reason": "3850 fixes audio diffusion test setup; 5396 fixes timezone handling in the stale bot. Completely different areas." + "reason": "Audio diffusion test adjustments and a timezone-aware stale-bot fix do not share an underlying bug." }, { "left": "pull_request:4211", "right": "pull_request:5699", "accept": false, - "reason": "Both touch SDXL-style pipeline docs/code, but 4211 is a docstring fix for inpainting examples and 5699 is a styling refactor across Stable UnCLIP/UnCLIP. Not the same change." + "reason": "Docstring cleanup for SDXL inpainting is unrelated to style-only refactors in UnCLIP/Versatile Diffusion." }, { "left": "pull_request:4319", "right": "pull_request:4564", "accept": false, - "reason": "4319 updates SDXL training-script licensing text; 4564 renames the ControlNet SDXL model-card repo reference. Same ecosystem, different fixes." + "reason": "Both mention SDXL, but one updates training-script licensing and the other fixes a ControlNet repo name in docs. Different changes." }, { "left": "pull_request:4666", "right": "pull_request:6450", "accept": false, - "reason": "4666 adds LoRA support for SDXL ControlNet; 6450 changes LoRA test naming/coverage around different scales. One is feature code, the other is test cleanup." + "reason": "Enabling LoRA for SDXL ControlNets is a feature change; the other is LoRA test/deprecation cleanup. Not the same fix." }, { "left": "pull_request:4749", "right": "pull_request:5393", "accept": false, - "reason": "4749 fixes SDXL LoRA checkpointing for text encoders; 5393 updates examples/docs to use HFApi. No shared bug." + "reason": "SDXL DreamBooth LoRA checkpointing and HFApi updates are separate example/script changes." }, { "left": "pull_request:4787", "right": "pull_request:5157", "accept": false, - "reason": "Both edit loaders.py, but 4787 tightens LoRA checkpoint file selection while 5157 corrects kohya/controlnet LoRA name conversion rules. Different conversion/loading bugs." + "reason": "Both affect loaders.py and LoRA handling, but one tightens weight-file selection while the other fixes Kohya/ControlNet name conversion. Different concrete bugs." }, { "left": "pull_request:5030", "right": "pull_request:6075", "accept": false, - "reason": "5030 fixes model CPU offload when a component key is missing; 6075 fixes backend cache clearing in tests. Different problem domains." + "reason": "Model CPU offload key handling and backend cache clearing are unrelated runtime/test issues." }, { "left": "pull_request:508", "right": "pull_request:564", "accept": false, - "reason": "Both are Flax scheduler-related, but 508 adds/ports schedulers and 564 updates scheduler docs/behavior wording. Not the same concrete bug or code-path fix." + "reason": "Both involve schedulers, but one adds Flax scheduler implementations and the other updates Flax scheduler docs. Not the same change." }, { "left": "pull_request:51", "right": "pull_request:748", "accept": false, - "reason": "51 is a packaging dependency pin fix; 748 fixes push_to_hub calls in example training scripts. Unrelated." + "reason": "Packaging dependency correction and push_to_hub example fixes are unrelated." }, { "left": "pull_request:5122", "right": "pull_request:5579", "accept": false, - "reason": "5122 updates workflow/setup Python versions; 5579 bumps the minimum transformers version for PEFT integration. Both dependency-related, but not the same issue." + "reason": "Workflow Python-version changes are unrelated to bumping the transformers minimum version for PEFT integration." }, { "left": "pull_request:5275", "right": "pull_request:6305", "accept": false, - "reason": "5275 adds an SDXL JAX/TPU example project; 6305 adds zero-terminal-SNR support to DDPMScheduler. Feature areas do not match." + "reason": "Adding an SDXL Flax example and adding zero-terminal-SNR support to DDPM are different tasks." }, { "left": "pull_request:5506", "right": "pull_request:7679", "accept": false, - "reason": "5506 fixes PEFT LoRA/xformers attention scaling in model code; 7679 changes the CI install command for PEFT. One runtime bug, one workflow tweak." + "reason": "PEFT/xformers attention handling and CI installation method changes are unrelated." }, { "left": "pull_request:5626", "right": "pull_request:6163", "accept": false, - "reason": "5626 fixes a CPU-offload test assertion; 6163 adds LoRA/xformers tests. Same test suite area, but different objectives and code paths." + "reason": "CPU offload test fixes and LoRA test additions are not the same bug or change." }, { "left": "pull_request:5672", "right": "pull_request:6215", "accept": false, - "reason": "Both concern PixArt tests, but 5672 supports non-square images in the model code while 6215 changes PixArt VAE test initialization and expected outputs. Different underlying fixes." + "reason": "Both touch PixArt tests, but one implements non-square image support and the other changes VAE test initialization/expected outputs. Not a duplicate pair." + }, + { + "left": "pull_request:5707", + "right": "pull_request:6997", + "accept": false, + "reason": "LCM test additions and UNetSpatioTemporalCondition LoRA support are unrelated." + }, + { + "left": "pull_request:6049", + "right": "pull_request:7464", + "accept": false, + "reason": "DemoFusion documentation/code adjustments and an LDM3D import fix are distinct changes." + }, + { + "left": "pull_request:6065", + "right": "pull_request:6588", + "accept": false, + "reason": "SDXL Turbo docs and SDXL k-diffusion test updates are unrelated." } ] }, @@ -159034,7 +159198,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 177, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -159042,11 +159206,17 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:11067", + "pull_request:11522", + "pull_request:11809", "pull_request:12119", + "pull_request:1229", "pull_request:12416", "pull_request:12445", + "pull_request:12530", "pull_request:13169", "pull_request:2034", + "pull_request:2339", "pull_request:2959", "pull_request:3313", "pull_request:3859", @@ -159054,22 +159224,16 @@ "pull_request:5057", "pull_request:5261", "pull_request:5579", - "pull_request:5707", "pull_request:5807", "pull_request:5823", - "pull_request:6049", - "pull_request:6065", "pull_request:6229", "pull_request:6389", - "pull_request:6588", "pull_request:6764", "pull_request:6772", - "pull_request:6997", "pull_request:7189", "pull_request:7328", "pull_request:7359", "pull_request:7442", - "pull_request:7464", "pull_request:7653", "pull_request:7771", "pull_request:7778", @@ -159091,9 +159255,6 @@ "pull_request:9718" ], "soft_pairs": [ - "pull_request:5707|pull_request:6997", - "pull_request:6049|pull_request:7464", - "pull_request:6065|pull_request:6588", "pull_request:6229|pull_request:7771", "pull_request:6229|pull_request:8386", "pull_request:6389|pull_request:7653", @@ -159115,186 +159276,189 @@ "pull_request:8856|pull_request:9311", "pull_request:2034|pull_request:2959", "pull_request:3859|pull_request:5057", - "pull_request:5579|pull_request:5823" + "pull_request:5579|pull_request:5823", + "pull_request:11067|pull_request:11522", + "pull_request:11809|pull_request:12530", + "pull_request:1229|pull_request:2339" ], - "prepared_review_unit_hash": "5ea72f00ce84105433cd80ae3a50ddb5b0e1bdcc104495795e73df1ccb0430dc", + "prepared_review_unit_hash": "0489b7eca5d77e6e79750c316b68ba476101daf1748845e9cda1b1cebaad1bef", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 25, - "serialized_chars": 104319, - "estimated_input_tokens": 26080, - "estimated_eval_tokens": 52416 + "serialized_chars": 102612, + "estimated_input_tokens": 25653, + "estimated_eval_tokens": 51562 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly superficial title/subsystem similarity across unrelated pull requests. None of the soft-edge pairs look like the same underlying bug or mergeable code change, so I would not collapse any of them as duplicates.", - "confidence": 0.96, + "summary": "This cluster is highly heterogeneous: most pairs are only superficially related by broad subsystem, shared filenames, or both being tests/docs/CI. I don\u2019t see a single underlying bug or change that ties the set together, so no canonical duplicate anchor stands out.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "If a single PR must anchor the cluster, 3859 is the broadest/highest-activity item, but it is not actually a duplicate of the others.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "3859 is the most substantial and broadly scoped PR here, so it is the least bad representative, though the set is not a true duplicate cluster.", + "best_pr_reason": "No single PR is a strong canonical representative; the items span unrelated fixes across CI, docs, tests, and pipeline features rather than one mergeable change.", "soft_edge_verdicts": [ - { - "left": "pull_request:5707", - "right": "pull_request:6997", - "accept": false, - "reason": "Different changes: 5707 updates LoRA weight-name handling and tests across pipelines; 6997 adds LoRA support to a specific UNet model. Shared LoRA theme only." - }, - { - "left": "pull_request:6049", - "right": "pull_request:7464", - "accept": false, - "reason": "Unrelated fixes: DemoFusion docs/pipeline cleanup vs an LDM3D community script import path fix." - }, - { - "left": "pull_request:6065", - "right": "pull_request:6588", - "accept": false, - "reason": "Docs for SDXL Turbo vs a slow-test expectation update for the SDXL k-diffusion pipeline; no shared bug." - }, { "left": "pull_request:6229", "right": "pull_request:7771", "accept": false, - "reason": "CI dependency installation vs marking a flaky ControlNet XS test. Different problems and code paths." + "reason": "Unrelated: CI dependency install vs a flaky UNetControlNetXS test." }, { "left": "pull_request:6229", "right": "pull_request:8386", "accept": false, - "reason": "Fast-test dependency setup is unrelated to changing model test dummy inputs/output shapes." + "reason": "Unrelated: CI dependency install vs broad test-shape refactors." }, { "left": "pull_request:6389", "right": "pull_request:7653", "accept": false, - "reason": "Stable Video Diffusion type/doc fix vs a configuration-utils decorator/refactor fix; not the same issue." + "reason": "Different problems: SVD API typing vs a configuration-utils decorator fix." }, { "left": "pull_request:6764", "right": "pull_request:7778", "accept": false, - "reason": "Docs note about saving fused LoRAs vs a new debugging workflow. No common bug/change." + "reason": "Docs note about serialization is unrelated to a debugging workflow workflow file." }, { "left": "pull_request:6772", "right": "pull_request:8825", "accept": false, - "reason": "ControlNet-XS implementation work is unrelated to adding ControlNet/IP-Adapter support to the SDXL k-diffusion pipeline." + "reason": "Both mention ControlNet/IP-Adapter, but they fix different code paths and features, not the same bug." }, { "left": "pull_request:7328", "right": "pull_request:9065", "accept": false, - "reason": "PyPI publish workflow dependency tweak vs adding sentencepiece as a soft dependency. Different concerns." + "reason": "Unrelated dependency/workflow change vs adding sentencepiece as an optional dependency." }, { "left": "pull_request:7359", "right": "pull_request:7442", "accept": false, - "reason": "Two different research-project additions (ResAdapter training vs Differential Diffusion) with no shared code-path bug." + "reason": "Different community projects: ResAdapter training vs Differential Diffusion." }, { "left": "pull_request:7825", "right": "pull_request:9442", "accept": false, - "reason": "Both mention device_map tests, but 7825 fixes model offload tests while 9442 changes CI pytest invocation for deprecated attention blocks. Not the same defect." + "reason": "Different test/CI issues; no shared failure mode." }, { "left": "pull_request:8627", "right": "pull_request:9307", "accept": false, - "reason": "SD3 ControlNet test-size reduction vs rotary embedding refactor in embeddings.py; unrelated." + "reason": "SD3 control test sizing vs rotary embedding refactor are unrelated." }, { "left": "pull_request:8660", "right": "pull_request:9592", "accept": false, - "reason": "SD3 ControlNet input validation vs Flax embedding docstring refactoring. No shared underlying bug." + "reason": "Pipeline input validation vs docstring cleanup; not the same change." }, { "left": "pull_request:8660", "right": "pull_request:9593", "accept": false, - "reason": "SD3 ControlNet input validation vs unrelated docstring cleanup in hd_painter.py." + "reason": "Pipeline validation bug vs unrelated docs refactor." }, { "left": "pull_request:9242", "right": "pull_request:9718", "accept": false, - "reason": "XPU test-backend support vs a community prompt-scheduling callback entry. Different areas entirely." + "reason": "XPU test backend support is unrelated to a prompt-scheduling community-script update." }, { "left": "pull_request:9446", "right": "pull_request:9705", "accept": false, - "reason": "TPU/XLA training-script updates vs Flux LoRA README wording updates. Not the same change." + "reason": "TPU training script updates vs a README tweak for a Flux LoRA script." }, { "left": "pull_request:3313", "right": "pull_request:4278", "accept": false, - "reason": "Both touch StableDiffusionUpscalePipeline, but 3313 fixes torch.compile breakages broadly while 4278 fixes prompt=None handling in input checks. Different bugs." + "reason": "Same file, but different bugs: torch.compile break fixes vs prompt=None handling in upscale." }, { "left": "pull_request:12445", "right": "pull_request:13169", "accept": false, - "reason": "Flux modular alignment refactor vs PEFT LoRA format support in the loader. Same ecosystem, different problem." + "reason": "Different Flux-related changes: modular alignment vs LoRA format conversion." }, { "left": "pull_request:7189", "right": "pull_request:8856", "accept": false, - "reason": "Both are img2img latent prep fixes, but 7189 fixes IP-Adapter embed repetition while 8856 fixes batch duplication/error handling for generator lists. Not mergeable as one bug fix." + "reason": "Both touch img2img helpers, but one is IP-Adapter embed repetition and the other is batch/image duplication handling." }, { "left": "pull_request:12119", "right": "pull_request:12416", "accept": false, - "reason": "Qwen-Image-Lightning LoRA loading support vs QwenImage Edit Plus modular support. Separate model features." + "reason": "QwenImage ecosystem overlap only; one adds Lightning LoRA support, the other adds QwenImage Edit Plus modular support." }, { "left": "pull_request:5261", "right": "pull_request:5807", "accept": false, - "reason": "Callback step-index fix in schedulers vs enabling num_images_per_prompt in a community SDXL LPW script. Different code paths." + "reason": "Callback-step fix vs prompt-embedding duplication for lpw SDXL; different issues." }, { "left": "pull_request:8856", "right": "pull_request:9311", "accept": false, - "reason": "Img2img batch-size error handling vs SDXL ControlNet CFG cutoff callbacks. No overlap beyond broad pipeline area." + "reason": "Image batch duplication bug vs SDXL ControlNet CFG cutoff callback; unrelated." }, { "left": "pull_request:2034", "right": "pull_request:2959", "accept": false, - "reason": "Unrelated test fixes for DiT vs UNet3D norm-group behavior." + "reason": "Different model test fixes with no shared underlying defect." }, { "left": "pull_request:3859", "right": "pull_request:5057", "accept": false, - "reason": "SDXL pipeline additions vs adding clip-skip support to SDXL. Related subsystem, but distinct features and change sets." + "reason": "Adding SDXL pipelines is not the same as adding clip-skip support; too broad to merge as duplicates." }, { "left": "pull_request:5579", "right": "pull_request:5823", "accept": false, - "reason": "Transformers min-version bump for PEFT integration vs adding adapter_names to fuse_lora. Different fixes." + "reason": "Transformers minimum-version bump vs LoRA adapter fusion API; unrelated changes." + }, + { + "left": "pull_request:11067", + "right": "pull_request:11522", + "accept": false, + "reason": "Benchmark version pinning vs Audioldm2 transformers compatibility fix are separate CI/runtime issues." + }, + { + "left": "pull_request:11809", + "right": "pull_request:12530", + "accept": false, + "reason": "Different test adjustments for different failures; no common bug." + }, + { + "left": "pull_request:1229", + "right": "pull_request:2339", + "accept": false, + "reason": "Both DreamBooth docs, but one is install guidance and the other is checkpointing behavior/tests." } ] }, @@ -159310,7 +159474,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 178, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -159321,14 +159485,9 @@ "pull_request:10245", "pull_request:10482", "pull_request:11066", - "pull_request:11067", - "pull_request:11522", "pull_request:11556", - "pull_request:11809", "pull_request:11823", "pull_request:11896", - "pull_request:1229", - "pull_request:12530", "pull_request:12805", "pull_request:12960", "pull_request:13211", @@ -159340,24 +159499,29 @@ "pull_request:2037", "pull_request:2295", "pull_request:2327", - "pull_request:2339", "pull_request:243", "pull_request:274", "pull_request:3189", "pull_request:3321", + "pull_request:3632", "pull_request:3707", + "pull_request:3795", "pull_request:3852", "pull_request:3859", "pull_request:4397", + "pull_request:4595", "pull_request:4695", "pull_request:4715", "pull_request:4749", "pull_request:4761", + "pull_request:4869", "pull_request:4887", "pull_request:5003", + "pull_request:5057", "pull_request:6099", "pull_request:6208", "pull_request:6225", + "pull_request:6405", "pull_request:6772", "pull_request:6965", "pull_request:6972", @@ -159368,9 +159532,6 @@ "pull_request:8282" ], "soft_pairs": [ - "pull_request:11067|pull_request:11522", - "pull_request:11809|pull_request:12530", - "pull_request:1229|pull_request:2339", "pull_request:12960|pull_request:13211", "pull_request:13330|pull_request:13355", "pull_request:153|pull_request:2037", @@ -159391,180 +159552,183 @@ "pull_request:1901|pull_request:3852", "pull_request:243|pull_request:274", "pull_request:3189|pull_request:4397", - "pull_request:3321|pull_request:3707" + "pull_request:3321|pull_request:3707", + "pull_request:3632|pull_request:5057", + "pull_request:3795|pull_request:4869", + "pull_request:4595|pull_request:6405" ], - "prepared_review_unit_hash": "b769017dc59989d34a5c83fa0d9ee702e408df5575c7acce19b17d5d1feb5793", + "prepared_review_unit_hash": "755d6b604cf26cdee5c5be12695a035e44a4b315f4224bd9e04d77ab21d32a86", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 96439, - "estimated_input_tokens": 24110, - "estimated_eval_tokens": 48476 + "serialized_chars": 96622, + "estimated_input_tokens": 24156, + "estimated_eval_tokens": 48568 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All suggested soft edges are false positives: the paired PRs are mostly unrelated docs, CI, tests, or feature changes that merely share broad keywords or occasional filenames.", - "confidence": 0.99, + "summary": "The cluster is highly heterogeneous: most pairs are superficial similarity matches across unrelated docs, CI, example, and model changes. None of the proposed soft edges look like true duplicates or the same concrete change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the cluster does not represent one underlying change, but many unrelated pull requests.", + "canonical_pr_reason": "No single canonical PR fits this cluster; the items span unrelated subsystems and change types, so the similarity links are spurious rather than duplicate PRs.", "best_issue_reason": null, - "best_pr_reason": "No single PR is a good global representative because the set spans distinct features and maintenance fixes rather than duplicates.", + "best_pr_reason": "PR 13211 is the strongest standalone bug fix in the set because it targets a concrete runtime offloading issue and includes a focused test, but it still does not serve as a duplicate anchor for the rest of the cluster.", "soft_edge_verdicts": [ - { - "left": "pull_request:11067", - "right": "pull_request:11522", - "accept": false, - "reason": "Benchmark transformers pinning vs an AudioLDM2 transformers-version compatibility fix; different subsystems and code paths." - }, - { - "left": "pull_request:11809", - "right": "pull_request:12530", - "accept": false, - "reason": "A tolerance-criteria test adjustment vs an xfail added for a different stable-cascade prior test failure." - }, - { - "left": "pull_request:1229", - "right": "pull_request:2339", - "accept": false, - "reason": "Dreambooth README wording change vs checkpointing docs/tests; same area but not the same bug or change." - }, { "left": "pull_request:12960", "right": "pull_request:13211", "accept": false, - "reason": "Flux2 FSDP detection tweak vs group-offloading bug fix in hooks; unrelated execution paths." + "reason": "Unrelated fixes: one determines FSDP handling in DreamBooth examples, the other patches group offloading prefetch behavior. Different code paths and failure modes." }, { "left": "pull_request:13330", "right": "pull_request:13355", "accept": false, - "reason": "TorchAO benchmark/test plumbing vs TorchAoConfig version-guard change; related ecosystem, different fixes." + "reason": "Both touch torchao, but one is a test/workflow tweak and the other is a version guard bump. Not the same bug or mergeable change." }, { "left": "pull_request:153", "right": "pull_request:2037", "accept": false, - "reason": "Issue templates vs adding a community Stable UnCLIP pipeline; completely different content." + "reason": "Issue templates vs a community pipeline addition; no shared underlying change." }, { "left": "pull_request:2295", "right": "pull_request:2327", "accept": false, - "reason": "Both touch UNet internals, but one makes transformer layers configurable while the other adds input-length validation." + "reason": "Both concern UNet internals, but one adds configurability and the other adds input validation. Different changes, not duplicates." }, { "left": "pull_request:4749", "right": "pull_request:6099", "accept": false, - "reason": "SDXL dreambooth LoRA checkpointing vs TextToVideoZero superclass-argument fix; different pipelines and failures." + "reason": "DreamBooth SDXL checkpointing and TextToVideoZero superclass argument fixes are unrelated code paths." }, { "left": "pull_request:4761", "right": "pull_request:6972", "accept": false, - "reason": "Paint-by-example encoder fix vs IP-Adapter unload test correction; different components and symptoms." + "reason": "Paint-by-example encoder initialization and IPAdapter unload test behavior are different issues." }, { "left": "pull_request:4887", "right": "pull_request:6208", "accept": false, - "reason": "Kandinsky prior_emb2emb API removal vs Stable Diffusion dtype mismatch on decode; unrelated code paths." + "reason": "Kandinsky prior_emb2emb API cleanup vs Stable Diffusion half/bias dtype runtime fix; unrelated." }, { "left": "pull_request:5003", "right": "pull_request:7158", "accept": false, - "reason": "LoRA fuse/unfuse logic fix vs LCM benchmark setup; one is model code, the other is benchmarking infra." + "reason": "LoRA fuse/unfuse logic and an LCM benchmark cleanup are not the same bug or change." }, { "left": "pull_request:6225", "right": "pull_request:7932", "accept": false, - "reason": "LoRA alpha parsing fix vs LoRA test expectation updates; tests may relate to the same area but not the same underlying bug." + "reason": "Both are LoRA-related, but one fixes alpha parsing and the other adjusts test expectations/side effects. Not mergeable as one PR." }, { "left": "pull_request:6772", "right": "pull_request:8282", "accept": false, - "reason": "ControlNet-XS implementation vs I2VGen-XL docs update; feature implementation and unrelated documentation." + "reason": "ControlNet-XS implementation work and I2VGenXL docs are completely different artifacts." }, { "left": "pull_request:7712", "right": "pull_request:7720", "accept": false, - "reason": "Import cleanup vs dead-code removal/f-string cleanup; small maintenance changes, not the same change." + "reason": "Import fixes and dead-code removal are separate maintenance changes." }, { "left": "pull_request:11823", "right": "pull_request:11896", "accept": false, - "reason": "Fal Kontext LoRA conversion support vs hook disabling bug fix; both LoRA-related but different mechanisms and fixes." + "reason": "Both involve LoRA infrastructure, but one adds Fal Kontext conversion support while the other fixes hook disabling. Different concrete problems." }, { "left": "pull_request:4715", "right": "pull_request:6965", "accept": false, - "reason": "Renaming community scripts to run vs moving SDXL T2I Adapter LoRA tests into PEFT workflow; distinct test/packaging changes." + "reason": "File renames for example test discovery and moving an SDXL LoRA test into PEFT workflow are unrelated." }, { "left": "pull_request:3859", "right": "pull_request:4695", "accept": false, - "reason": "SDXL pipeline additions vs TextToVideoZeroSDXLPipeline; both add pipelines but for different model families and functionality." + "reason": "Both are pipeline additions, but they implement different pipelines and code paths; too broad to treat as duplicates." }, { "left": "pull_request:10245", "right": "pull_request:11556", "accept": false, - "reason": "PEFT adapter docs update vs regional compilation docs; both docs-only, but unrelated topics." + "reason": "PEFT inference docs and regional compilation docs are unrelated documentation updates." }, { "left": "pull_request:10482", "right": "pull_request:11066", "accept": false, - "reason": "NaN bug fix in HunyuanVideo transformer vs new HunyuanVideo-I2V model support; different change types and goals." + "reason": "Same HunyuanVideo file, but one fixes NaNs in attention masking while the other adds a new I2V model variant. Different underlying changes." }, { "left": "pull_request:12805", "right": "pull_request:13352", "accept": false, - "reason": "PyPI publish workflow tweaks vs Claude review workflow checkout step; unrelated CI workflows." + "reason": "PyPI publish workflow changes and Claude review workflow checkout are unrelated CI edits." }, { "left": "pull_request:1901", "right": "pull_request:3852", "accept": false, - "reason": "Removing duplicated API autodoc entries vs adding a contributor note in docs; unrelated documentation updates." + "reason": "Both are docs-only, but they address different documentation pages and issues." }, { "left": "pull_request:243", "right": "pull_request:274", "accept": false, - "reason": "RePaint pipeline addition vs Stable Diffusion README addition; unrelated feature vs docs." + "reason": "RePaint pipeline WIP vs Stable Diffusion README addition; no shared bug/change." }, { "left": "pull_request:3189", "right": "pull_request:4397", "accept": false, - "reason": "AudioLDM waveform dtype fix vs SDXL docstring default guidance change; different modules and problems." + "reason": "AudioLDM waveform dtype fix and SDXL docstring correction are unrelated." }, { "left": "pull_request:3321", "right": "pull_request:3707", "accept": false, - "reason": "Controlling-generation docs expansion vs documentation-build workflow secrets changes; unrelated docs vs CI." + "reason": "Generation docs update and doc-build secrets/workflow changes are separate concerns." + }, + { + "left": "pull_request:3632", + "right": "pull_request:5057", + "accept": false, + "reason": "Panorama view batching and SDXL clip-skip support are different features." + }, + { + "left": "pull_request:3795", + "right": "pull_request:4869", + "accept": false, + "reason": "Overview-page docs updates and combined-pipeline tests are not the same change." + }, + { + "left": "pull_request:4595", + "right": "pull_request:6405", + "accept": false, + "reason": "Different doc-builder/testing work; no concrete shared bug or feature." } ] }, @@ -159580,7 +159744,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 179, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -159593,7 +159757,12 @@ "pull_request:10148", "pull_request:10187", "pull_request:10188", + "pull_request:10230", + "pull_request:10358", + "pull_request:10370", + "pull_request:10544", "pull_request:10630", + "pull_request:10970", "pull_request:11081", "pull_request:11130", "pull_request:11196", @@ -159606,19 +159775,13 @@ "pull_request:12534", "pull_request:1355", "pull_request:2727", - "pull_request:3632", - "pull_request:3795", - "pull_request:4595", - "pull_request:4869", "pull_request:4899", - "pull_request:5057", "pull_request:5842", "pull_request:5880", "pull_request:5887", "pull_request:5922", "pull_request:6217", "pull_request:628", - "pull_request:6405", "pull_request:6578", "pull_request:6646", "pull_request:670", @@ -159629,6 +159792,7 @@ "pull_request:8229", "pull_request:8480", "pull_request:8496", + "pull_request:8691", "pull_request:8696", "pull_request:8702", "pull_request:8765", @@ -159638,9 +159802,6 @@ "pull_request:9862" ], "soft_pairs": [ - "pull_request:3632|pull_request:5057", - "pull_request:3795|pull_request:4869", - "pull_request:4595|pull_request:6405", "pull_request:5842|pull_request:7924", "pull_request:5880|pull_request:6217", "pull_request:5922|pull_request:7924", @@ -159662,186 +159823,189 @@ "pull_request:10069|pull_request:11196", "pull_request:10119|pull_request:9105", "pull_request:10148|pull_request:8696", - "pull_request:10188|pull_request:11196" + "pull_request:10188|pull_request:11196", + "pull_request:10230|pull_request:8691", + "pull_request:10358|pull_request:10544", + "pull_request:10370|pull_request:10970" ], - "prepared_review_unit_hash": "532df7e39268aedf2bb5ab048fa2b7d6e0525e8205e43b5d4fad80a2eaf9956f", + "prepared_review_unit_hash": "33ea2c9a8ff00389d839eaf0cad48ef67f20535bb78b54038a8d027f288f3752", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 103486, - "estimated_input_tokens": 25872, - "estimated_eval_tokens": 52000 + "serialized_chars": 103578, + "estimated_input_tokens": 25895, + "estimated_eval_tokens": 52046 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly unrelated PRs spanning docs, new model/pipeline features, test updates, and isolated bug fixes. The apparent similarity is mostly superficial (shared subsystems or filenames), not a single underlying duplicate change.", - "confidence": 0.97, + "summary": "This is a heterogeneous set of mostly unrelated PRs (docs updates, new model/pipeline additions, LoRA/quantization changes, and isolated bug fixes). The soft edges are mostly surface-level similarity and do not indicate duplicate underlying changes.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a good canonical representative because the set is heterogeneous and does not center on one bug or feature.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No global best PR; the items split across unrelated features and fixes, so any single PR would be arbitrary.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:3632", - "right": "pull_request:5057", - "accept": false, - "reason": "Different problems: panorama view batching vs SDXL clip-skip support." - }, - { - "left": "pull_request:3795", - "right": "pull_request:4869", - "accept": false, - "reason": "Docs overview cleanup is unrelated to combined-pipeline test coverage." - }, - { - "left": "pull_request:4595", - "right": "pull_request:6405", - "accept": false, - "reason": "A WIP conversion PR is not the same change as a doc-builder workflow tweak." - }, { "left": "pull_request:5842", "right": "pull_request:7924", "accept": false, - "reason": "PixArt 512px resolution fix vs ControlNet IP-Adapter condition handling are unrelated code paths." + "reason": "Both touch image-generation pipeline behavior, but one fixes PixArt 512px resolution mapping and the other fixes IP-Adapter handling in SDXL inpaint; different models, code paths, and bugs." }, { "left": "pull_request:5880", "right": "pull_request:6217", "accept": false, - "reason": "VAE config registration across many pipelines vs Stable Video Diffusion torch-compile fix." + "reason": "Both involve Stable Video Diffusion/VAE-related code, but 5880 is a broad VAE compatibility/config change while 6217 is a torch.compile-specific VAE forward introspection fix; not the same bug." }, { "left": "pull_request:5922", "right": "pull_request:7924", "accept": false, - "reason": "MultiAdapter inpainting example fix is different from SDXL ControlNet IP-Adapter kwargs handling." + "reason": "Both mention adapter/inpaint flows, but 5922 fixes MultiAdapter image preprocessing in a community example and 7924 fixes added_cond_kwargs for IP-Adapter in the SDXL ControlNet pipeline; distinct changes." }, { "left": "pull_request:628", "right": "pull_request:670", "accept": false, - "reason": "Both touch docs index, but one adds a link and the other removes one; not the same change." + "reason": "These are unrelated documentation index tweaks from early repo history; they do not fix the same underlying issue or change." }, { "left": "pull_request:6578", "right": "pull_request:8480", "accept": false, - "reason": "Docs heading fix vs HunyuanDIT ControlNet model support are unrelated." + "reason": "One is a docs heading/indent fix for ControlNet; the other adds HunyuanDiT ControlNet support. Same broad area, but not the same change." }, { "left": "pull_request:7034", "right": "pull_request:8765", "accept": false, - "reason": "ControlNet single-file loading fix vs a docs typo in the single-file page." + "reason": "7034 fixes `load_model_dict_into_meta` usage in single-file ControlNet loading, while 8765 corrects a docs API argument typo; unrelated." }, { "left": "pull_request:7247", "right": "pull_request:8229", "accept": false, - "reason": "Config tuple/list normalization is unrelated to the Stable Video Diffusion resize fix." + "reason": "7247 fixes config/type checking in single-file loading tests, while 8229 changes Stable Video Diffusion resizing behavior. Different defects and code paths." }, { "left": "pull_request:7919", "right": "pull_request:9736", "accept": false, - "reason": "Offload behavior with device_map vs adding Allegro T2V support are different changes." + "reason": "One fixes offload behavior with device maps; the other adds the Allegro text-to-video model/pipeline. No shared underlying bug or change." }, { "left": "pull_request:8496", "right": "pull_request:8702", "accept": false, - "reason": "SD3 warning-log adjustment vs an SD3 conversion script; different goals and code paths." + "reason": "Both relate to SD3, but 8496 only narrows a warning log condition in the transformer, while 8702 adds a conversion script. Not mergeable as one PR." }, { "left": "pull_request:11081", "right": "pull_request:11130", "accept": false, - "reason": "Group-offloading record_stream support and pipeline-level quantization config are separate features." + "reason": "11081 implements `record_stream` for group offloading; 11130 introduces pipeline-level quantization config and tests. Different features." }, { "left": "pull_request:11231", "right": "pull_request:11995", "accept": false, - "reason": "HiDream Image support vs modular Flux pipeline support are distinct model families and implementations." + "reason": "Both add large model families, but HiDream Image and modular Flux are separate model integrations with unrelated implementation work." }, { "left": "pull_request:11869", "right": "pull_request:12534", "accept": false, - "reason": "GGUF CUDA kernel support vs Flux NPU positional-embedding handling are unrelated fixes." + "reason": "11869 adds CUDA kernel support for GGUF inference, while 12534 adjusts Flux pos-embedding behavior on Ascend NPU. Different hardware targets and fixes." }, { "left": "pull_request:6646", "right": "pull_request:8856", "accept": false, - "reason": "SVD resolution handling vs image duplication for img2img batch errors are different bugs." + "reason": "Both concern image/video resolution or img2img handling, but 6646 changes SVD resolution constraints and 8856 fixes batch duplication error handling across img2img pipelines; not the same bug." }, { "left": "pull_request:10630", "right": "pull_request:11469", "accept": false, - "reason": "Flax test-slice adjustments vs xfail markers for PyTorch-version-specific failures." + "reason": "Both are test-only changes, but they address different failing tests on different pipelines and PyTorch versions." }, { "left": "pull_request:12226", "right": "pull_request:12330", "accept": false, - "reason": "A new Qwen-Image ControlNet test is unrelated to cleaning pipeline group-offloading tests." + "reason": "One adds a fast Qwen-Image ControlNet test; the other cleans up unrelated group offloading tests. No duplicate change." }, { "left": "pull_request:1355", "right": "pull_request:2727", "accept": false, - "reason": "Flax img2img pipeline addition vs Flax ControlNet addition are separate features, not one duplicate change." + "reason": "Both are Flax pipeline additions, but one adds Stable Diffusion img2img and the other adds Flax ControlNet; separate features, not duplicates." }, { "left": "pull_request:4899", "right": "pull_request:5887", "accept": false, - "reason": "ControlNet LoRA support and IP-Adapter support solve different adapter integration problems." + "reason": "4899 is a work-in-progress ControlNet LoRA support change, while 5887 adds IP-Adapter support to ControlNet inpaint. Different loader paths and goals." }, { "left": "pull_request:10187", "right": "pull_request:9862", "accept": false, - "reason": "LoRA no-op warning cleanup vs save_lora_adapter implementation are different LoRA code paths." + "reason": "Both are LoRA-related, but 10187 improves no-op warnings during loading, whereas 9862 adds `save_lora_adapter()`. Different functionality." }, { "left": "pull_request:10069", "right": "pull_request:11196", "accept": false, - "reason": "Device-type validation in load_model_dict_into_meta vs removing save_model fallback in save_pretrained are unrelated." + "reason": "10069 adds device-type validation in model loading; 11196 reverts `save_model` usage in `save_pretrained`. Distinct model serialization/loading issues." }, { "left": "pull_request:10119", "right": "pull_request:9105", "accept": false, - "reason": "Docs for loading LoRA adapters vs a deprecation-related docs change are not the same change." + "reason": "10119 is docs for `load_lora_adapter`; 9105 is a docs deprecation cleanup. No shared underlying change." }, { "left": "pull_request:10148", "right": "pull_request:8696", "accept": false, - "reason": "Adding OmniGen support vs marking Marigold tests flaky are unrelated." + "reason": "An OmniGen feature addition and a Marigold test-flakiness annotation are unrelated." }, { "left": "pull_request:10188", "right": "pull_request:11196", "accept": false, - "reason": "Flux LoRA load-message test fix vs save_pretrained serialization behavior are different issues." + "reason": "10188 fixes Flux LoRA expansion/no-op loading behavior, while 11196 changes safe serialization behavior in model saving. Different code paths." + }, + { + "left": "pull_request:10230", + "right": "pull_request:8691", + "accept": false, + "reason": "10230 updates LoRA loader docs; 8691 fixes vanilla fine-tuned LoRA loading. One is documentation only, the other is runtime behavior." + }, + { + "left": "pull_request:10358", + "right": "pull_request:10544", + "accept": false, + "reason": "Both are documentation additions, but one adds a community project entry and the other adds ParaAttention optimization docs; not the same artifact." + }, + { + "left": "pull_request:10370", + "right": "pull_request:10970", + "accept": false, + "reason": "10370 is a formatting/style change in torchao quantizer code; 10970 swaps in a reusable style-bot GitHub Action workflow. Different changes." } ] }, @@ -159857,21 +160021,16 @@ "component_index": 1, "component_count": 266, "review_unit_index": 180, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10230", - "pull_request:10358", - "pull_request:10370", "pull_request:10506", - "pull_request:10544", "pull_request:10579", - "pull_request:10970", "pull_request:11001", "pull_request:11113", "pull_request:11186", @@ -159903,20 +160062,22 @@ "pull_request:3297", "pull_request:335", "pull_request:3934", + "pull_request:4276", "pull_request:4595", + "pull_request:4596", + "pull_request:4655", "pull_request:4887", "pull_request:4963", "pull_request:5053", "pull_request:5111", "pull_request:5615", - "pull_request:8691", + "pull_request:5697", + "pull_request:5853", + "pull_request:6114", "pull_request:8981", "pull_request:9307" ], "soft_pairs": [ - "pull_request:10230|pull_request:8691", - "pull_request:10358|pull_request:10544", - "pull_request:10370|pull_request:10970", "pull_request:10506|pull_request:9307", "pull_request:10579|pull_request:8981", "pull_request:11001|pull_request:11186", @@ -159939,192 +160100,195 @@ "pull_request:3275|pull_request:5111", "pull_request:3297|pull_request:4595", "pull_request:3934|pull_request:4963", - "pull_request:3934|pull_request:5615" + "pull_request:3934|pull_request:5615", + "pull_request:4276|pull_request:5853", + "pull_request:4596|pull_request:6114", + "pull_request:4655|pull_request:5697" ], - "prepared_review_unit_hash": "2129973bc55fb104339daf4930f98c1db2dc3299223ea2246b84a55a898ad1da", + "prepared_review_unit_hash": "52a9285fd84d27b333321d2e5dd8ef98f16f58df1a8edf80c3f52fdf8a41680d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 26, - "serialized_chars": 99624, - "estimated_input_tokens": 24906, - "estimated_eval_tokens": 50068 + "serialized_chars": 100689, + "estimated_input_tokens": 25173, + "estimated_eval_tokens": 50602 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No strong duplicate clusters here: the soft edges are mostly unrelated docs, tests, style, or feature PRs that only share a broad subsystem. I\u2019d keep them separate and use a core bug-fix PR as the representative item.", - "confidence": 0.91, + "summary": "This cluster is not a true duplicate set: the soft edges mostly connect unrelated PRs that only share broad themes like tests, CI, docs, or pipeline work. I rejected all pairings.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 8691 is a clear, self-contained core library bug fix in LoRA loading, with a specific code-path change and a focused title, making it a good representative PR.", + "canonical_pr_reason": "No single PR is a valid canonical duplicate anchor here; the items span unrelated changes across tests, docs, CI, training scripts, and core pipeline/model code.", "best_issue_reason": null, - "best_pr_reason": "PR 8691 is the strongest canonical PR candidate among these items because it fixes a concrete runtime behavior in core code rather than only docs, tests, or tooling.", + "best_pr_reason": "No global best PR for deduplication: none of these PRs subsumes the others or represents the same underlying change.", "soft_edge_verdicts": [ - { - "left": "pull_request:10230", - "right": "pull_request:8691", - "accept": false, - "reason": "Docs-only LoRA mixin additions versus a core LoRA loading bug fix; related theme, but not the same change or bug." - }, - { - "left": "pull_request:10358", - "right": "pull_request:10544", - "accept": false, - "reason": "Both are documentation PRs, but they cover unrelated topics: community projects listing vs ParaAttention docs." - }, - { - "left": "pull_request:10370", - "right": "pull_request:10970", - "accept": false, - "reason": "A style tweak for a prior PR and a workflow change to use a style bot action are not the same concrete bug/change." - }, { "left": "pull_request:10506", "right": "pull_request:9307", "accept": false, - "reason": "A quantization offload test and a rotary-embedding refactor touch different code paths and solve different problems." + "reason": "Unrelated changes: a torchao offload test vs a rotary-embedding refactor." }, { "left": "pull_request:10579", "right": "pull_request:8981", "accept": false, - "reason": "Different features entirely: a Wan VACE LoRA test versus adding LoraLoaderMixin to init exports/docs." + "reason": "Different topics: an XL attentive eraser community pipeline vs LoRA loader mixin exports/docs." }, { "left": "pull_request:11001", "right": "pull_request:11186", "accept": false, - "reason": "Both mention device portability, but one enables XPU CI support across code/tests while the other only removes a hard-coded cuda call in one test." + "reason": "Both are device-portability related, but one enables bnb tests on xpu while the other only removes a hard-coded cuda call in one test." }, { "left": "pull_request:11113", "right": "pull_request:11743", "accept": false, - "reason": "AutoencoderKL training-script bug fix versus adding a new Flux LoRA quantization training script; unrelated changes." + "reason": "Different training scripts and goals; one fixes AutoencoderKL training, the other adds a quantized Flux DreamBooth script." }, { "left": "pull_request:11843", "right": "pull_request:12493", "accept": false, - "reason": "Wan VACE LoRA exclude_modules tests and LTX 0.9.8 pipeline/docs are unrelated." + "reason": "A Wan VACE LoRA test is unrelated to LTX-Video 0.9.8 pipeline/docs changes." }, { "left": "pull_request:12062", "right": "pull_request:12397", "accept": false, - "reason": "Both are CI workflow edits, but they address different pipeline behavior and cannot plausibly be merged as one fix." + "reason": "Both touch CI workflows, but one changes GPU allocation and the other stops installing transformers/accelerate from main; not the same fix." }, { "left": "pull_request:12363", "right": "pull_request:13100", "accept": false, - "reason": "QwenImageEdit Plus docs/tests versus ModularPipeline docs and runtime-state changes are unrelated." + "reason": "QwenImage Edit Plus docs/tests are unrelated to modular pipeline state documentation." }, { "left": "pull_request:12642", "right": "pull_request:13100", "accept": false, - "reason": "Doc TOC validation updates and ModularPipeline doc/runtime updates are different concerns." + "reason": "Docs/toctree maintenance is not the same change as modular pipeline state docs and code updates." }, { "left": "pull_request:1331", "right": "pull_request:335", "accept": false, - "reason": "Inpainting scheduler handling versus PNDM type hints; not the same bug or code-path problem." + "reason": "Different code paths: inpainting scheduler input ordering vs PNDM type hints." }, { "left": "pull_request:1494", "right": "pull_request:2122", "accept": false, - "reason": "Safetensors save support and safetensors documentation are related by topic only, not the same underlying change." + "reason": "Safetensors saving support and safetensors docs are related only by topic, not the same underlying change." }, { "left": "pull_request:153", "right": "pull_request:1945", "accept": false, - "reason": "Issue templates and UNet block tests are unrelated." + "reason": "Issue templates are unrelated to 2D UNet block tests." }, { "left": "pull_request:2009", "right": "pull_request:274", "accept": false, - "reason": "Textual inversion loading and a Stable Diffusion README are unrelated." + "reason": "Textual inversion loader integration is unrelated to a Stable Diffusion README addition." }, { "left": "pull_request:2009", "right": "pull_request:3275", "accept": false, - "reason": "Textual inversion loading versus adding upsample_size to attention blocks are different code changes." + "reason": "Textual inversion loader support and upsample_size support in UNet blocks are different features." }, { "left": "pull_request:2279", "right": "pull_request:2827", "accept": false, - "reason": "General formatting/tooling modernization and applying Ruff rules are not the same concrete fix." + "reason": "Both concern lint/style tooling, but these are different repo-wide hygiene changes, not one concrete bug or change." }, { "left": "pull_request:2793", "right": "pull_request:2928", "accept": false, - "reason": "A docs removal about cuDNN benchmark and controlnet local dataset loading are unrelated." + "reason": "A docs removal about cuDNN benchmark is unrelated to allowing custom local datasets in ControlNet training." }, { "left": "pull_request:2793", "right": "pull_request:4887", "accept": false, - "reason": "cuDNN benchmark docs and removing a latent input from Kandinsky prior_emb2emb are unrelated." + "reason": "No shared code-path problem: docs cleanup vs removing latents from Kandinsky prior_emb2emb." }, { "left": "pull_request:2861", "right": "pull_request:3208", "accept": false, - "reason": "A Stable UnCLIP test for image_embeds=None and Token Merging docs are unrelated." + "reason": "A Stable UnCLIP test for image_embeds=None is unrelated to Token Merging docs." }, { "left": "pull_request:2959", "right": "pull_request:5053", "accept": false, - "reason": "UNet3D norm-group test correction versus a t2i-adapter validation_image arg fix are unrelated." + "reason": "A UNet3D norm-group test fix is unrelated to the validation_image CLI argument change." }, { "left": "pull_request:3023", "right": "pull_request:3127", "accept": false, - "reason": "Attention processor changes for torch 2.0 versus a safety-checker img2img fix are different bugs and code paths." + "reason": "Different subsystems: attention processor/KV handling vs safety checker image masking." }, { "left": "pull_request:3275", "right": "pull_request:5111", "accept": false, - "reason": "Adding upsample_size to attention blocks and adding use_karras_sigmas to schedulers are unrelated." + "reason": "UNet block upsample_size plumbing and Karras sigma scheduler support are unrelated." }, { "left": "pull_request:3297", "right": "pull_request:4595", "accept": false, - "reason": "A new retrieval-augmented diffusion project and a WIP converter pipeline are unrelated." + "reason": "A retrieval-augmented diffusion model project is not the same as an SDXL-to-original conversion script." }, { "left": "pull_request:3934", "right": "pull_request:4963", "accept": false, - "reason": "Both are revert-related CI/pipeline changes, but they revert different behaviors in different areas." + "reason": "These are unrelated reverts: automatic chunking vs accelerate-main install changes." }, { "left": "pull_request:3934", "right": "pull_request:5615", "accept": false, - "reason": "Automatic chunking revert versus SDXL adapter LoRA revert are separate changes and not the same underlying issue." + "reason": "Different revert scopes: text-to-video chunking behavior vs SDXL adapter LoRA loading." + }, + { + "left": "pull_request:4276", + "right": "pull_request:5853", + "accept": false, + "reason": "A pipeline import smoke test and doctests workflow are different test infrastructure changes." + }, + { + "left": "pull_request:4596", + "right": "pull_request:6114", + "accept": false, + "reason": "A conversion script for SDXL-to-original checkpoints is unrelated to lpw-xl img2img/inpaint support." + }, + { + "left": "pull_request:4655", + "right": "pull_request:5697", + "accept": false, + "reason": "Adding data_dir to load_dataset is unrelated to the PEFT LoRA xformers bugfix." } ] }, @@ -160140,11 +160304,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 181, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -160154,10 +160318,12 @@ "pull_request:13365", "pull_request:1470", "pull_request:1966", - "pull_request:4276", - "pull_request:4596", - "pull_request:4655", + "pull_request:4365", + "pull_request:4749", + "pull_request:4793", + "pull_request:4862", "pull_request:5038", + "pull_request:5085", "pull_request:5204", "pull_request:530", "pull_request:5393", @@ -160165,15 +160331,14 @@ "pull_request:5771", "pull_request:5833", "pull_request:5834", - "pull_request:5853", "pull_request:594", - "pull_request:6114", "pull_request:6237", "pull_request:6261", "pull_request:6426", "pull_request:6633", "pull_request:6752", "pull_request:6864", + "pull_request:6929", "pull_request:7044", "pull_request:7116", "pull_request:7153", @@ -160197,9 +160362,6 @@ "pull_request:9735" ], "soft_pairs": [ - "pull_request:4276|pull_request:5853", - "pull_request:4596|pull_request:6114", - "pull_request:4655|pull_request:5697", "pull_request:530|pull_request:748", "pull_request:5393|pull_request:7216", "pull_request:5771|pull_request:6261", @@ -160222,192 +160384,202 @@ "pull_request:6752|pull_request:8739", "pull_request:10105|pull_request:9103", "pull_request:1269|pull_request:1470", - "pull_request:13294|pull_request:13365" + "pull_request:13294|pull_request:13365", + "pull_request:4365|pull_request:4749", + "pull_request:4793|pull_request:4862", + "pull_request:5085|pull_request:5697", + "pull_request:6237|pull_request:6929" ], - "prepared_review_unit_hash": "13279d9eb754b2c27ffe37cf5a243f93d096cce25b2fdf52fa533a8735c4b73f", + "prepared_review_unit_hash": "a4735a384a5cd3ecf21fdf35f8b6bb4eccba665f8cd1fa48a98ee15976f77182", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 26, - "serialized_chars": 100378, - "estimated_input_tokens": 25095, - "estimated_eval_tokens": 50446 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 102667, + "estimated_input_tokens": 25667, + "estimated_eval_tokens": 51590 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a broad false-positive similarity set: the paired PRs share generic themes (docs, workflows, LoRA, SD3, single-file loading) but do not fix the same concrete bug or implement the same change. I would not consolidate any pair.", + "summary": "The cluster is a grab bag of unrelated diffusers PRs spanning docs, CI, LoRA, single-file loading, SD3, and examples. None of the soft-edge pairs look like the same underlying bug or change, so I would not merge any of them as duplicates.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR is a true duplicate of the others. If a representative PR is needed, #5771 is the broadest core-code change in the set (large loaders/LoRA refactor across many files), so it is the best stand-in for the cluster.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "#5771 is the strongest representative because it is a substantial, cross-cutting code change in the core loader/LoRA path and is more central than the many narrow docs/tests/CI fixes in this group.", "soft_edge_verdicts": [ - { - "left": "pull_request:4276", - "right": "pull_request:5853", - "accept": false, - "reason": "`4276` adds a pipeline import test; `5853` introduces doctest/link-check workflows. Different maintenance areas and no shared bug/change." - }, - { - "left": "pull_request:4596", - "right": "pull_request:6114", - "accept": false, - "reason": "`4596` adds an SDXL conversion script; `6114` updates a community pipeline/README for img2img and inpaint. Not the same code path." - }, - { - "left": "pull_request:4655", - "right": "pull_request:5697", - "accept": false, - "reason": "`4655` passes `data_dir` into dataset loading in training scripts; `5697` fixes PEFT LoRA/xformers attention scaling. Unrelated fixes." - }, { "left": "pull_request:530", "right": "pull_request:748", "accept": false, - "reason": "`530` changes img2img test assertions; `748` fixes `push_to_hub` calls in training examples. Test cleanup vs training API bug." + "reason": "Different fixes in different areas: img2img test adjustment vs push_to_hub call cleanup for training scripts." }, { "left": "pull_request:5393", "right": "pull_request:7216", "accept": false, - "reason": "`5393` migrates examples/docs to HFApi; `7216` changes `from_single_file` signature/docs. Same ecosystem, but different changes." + "reason": "Both touch docs/examples around hub/file handling, but one updates API usage while the other changes the from_single_file signature and loader behavior." }, { "left": "pull_request:5771", "right": "pull_request:6261", "accept": false, - "reason": "`5771` refactors LoRA loaders; `6261` moves Attend-and-Excite out of `stable_diffusion`. Different refactors and code paths." + "reason": "Unrelated changes: a broad loaders/LoRA refactor vs moving Attend-and-Excite into its own module." }, { "left": "pull_request:5833", "right": "pull_request:7339", "accept": false, - "reason": "`5833` is PixArt documentation; `7339` fixes release workflow dependencies. No concrete overlap." + "reason": "Single-file ControlNet import fix and release-workflow dependency changes are unrelated." }, { "left": "pull_request:5834", "right": "pull_request:7339", "accept": false, - "reason": "`5834` fixes single-file ControlNet imports; `7339` updates release deps. Completely different subsystems." + "reason": "Single-file import path correction is unrelated to release workflow dependency setup." }, { "left": "pull_request:6237", "right": "pull_request:6864", "accept": false, - "reason": "Both are docs, but one explains batched seeds and the other `torch.compile` with LoRA. Distinct guidance, not a duplicate change." + "reason": "Docs-only changes about seeds vs LoRA/torch.compile guidance; different concerns and code paths." }, { "left": "pull_request:6426", "right": "pull_request:8267", "accept": false, - "reason": "Both mention deprecations, but `6426` removes deprecated LoRA backend handling while `8267` adjusts `resume_download` behavior. Different deprecation paths." + "reason": "Both mention deprecation, but one is a broad LoRA deprecation cleanup and the other is specific resume_download handling in single-file loaders." }, { "left": "pull_request:6633", "right": "pull_request:7247", "accept": false, - "reason": "`6633` lets `convert_from_ckpt.py` read a local YAML config; `7247` changes single-file config checking tests/config shapes. Related area, but not the same fix." + "reason": "One fixes checkpoint config file reading; the other changes single-file config-shape tests. Not the same bug." }, { "left": "pull_request:7044", "right": "pull_request:7295", "accept": false, - "reason": "`7044` re-adds UNet refactor helpers; `7295` cleans a single-file UNet loader docstring. Not the same code-path problem." + "reason": "Different loader/refactor changes: restoring a UNet refactor vs cleaning copy-paste residue in the UNet single-file loader." }, { "left": "pull_request:7116", "right": "pull_request:7153", "accept": false, - "reason": "`7116` switches workflows to `uv`; `7153` corrects documentation workflow path filters. Separate CI changes." + "reason": "CI workflow install tooling vs documentation path filtering; not the same underlying change." }, { "left": "pull_request:7913", "right": "pull_request:9447", "accept": false, - "reason": "`7913` adds custom timesteps/sigmas to SDXL ControlNet; `9447` changes pipeline save sharding defaults. Different features and failure modes." + "reason": "ControlNet SDXL timestep/sigma support is unrelated to pipeline save_pretrained sharding defaults." }, { "left": "pull_request:8564", "right": "pull_request:8631", "accept": false, - "reason": "`8564` adds SD3 T5 token-limit training support; `8631` fixes `from_single_file` for T5 checkpoints. Same model family, but not the same underlying issue." + "reason": "SD3 training docs/code for T5 sequence length vs a single-file loader fix for T5 checkpoints; same model family, different problems." }, { "left": "pull_request:8591", "right": "pull_request:8618", "accept": false, - "reason": "`8591` refactors SD3 density/weighting utilities in training; `8618` adds `shift_factor` to the tiny autoencoder. Different code paths." + "reason": "SD3 training utility refactor vs adding shift_factor to the tiny autoencoder; different code paths and purposes." }, { "left": "pull_request:8591", "right": "pull_request:8627", "accept": false, - "reason": "`8591` is SD3 training-loss math; `8627` adds SD3 ControlNet model/pipeline/docs. Broadly related to SD3, but not one change." + "reason": "Training-utility refactor is unrelated to adding SD3 ControlNet docs/models/tests." }, { "left": "pull_request:8618", "right": "pull_request:9735", "accept": false, - "reason": "`8618` changes the tiny autoencoder's `shift_factor`; `9735` updates an SD3 ControlNet example and training script imports. Unrelated." + "reason": "Tiny autoencoder shift_factor addition and SD3 ControlNet example update are separate features." }, { "left": "pull_request:8627", "right": "pull_request:9735", "accept": false, - "reason": "`8627` introduces SD3 ControlNet model/pipeline files; `9735` updates the SD3 ControlNet example. Example update is not the same as the implementation PR." + "reason": "SD3 ControlNet model/docs addition is broader and different from updating a training example." }, { "left": "pull_request:9400", "right": "pull_request:9536", "accept": false, - "reason": "`9400` makes single-file pipeline tests device-agnostic; `9536` fixes a tokenizer padding-side argument. Different bug fixes." + "reason": "Device-agnostic test updates and ChatGLMTokenizer padding signature fixes are unrelated." }, { "left": "pull_request:1966", "right": "pull_request:594", "accept": false, - "reason": "`1966` adds Flax inpainting; `594` fixes Flax UNet/DDIM embeddings/scheduler behavior. Both Flax, but not the same concrete change." + "reason": "Flax inpainting implementation vs Flax UNet/DDIM embedding/scheduler fix are different feature areas." }, { "left": "pull_request:5038", "right": "pull_request:5204", "accept": false, - "reason": "Both touch LoRA-related logic, but `5038` makes optional args explicit in `transformer_2d.py` while `5204` fixes text-encoder LoRA scaling across pipelines. Not mergeable as one PR." + "reason": "LoRA-related, but one makes optional args explicit in Transformer2D while the other fixes text encoder scaling across many pipelines." }, { "left": "pull_request:6752", "right": "pull_request:8739", "accept": false, - "reason": "`6752` adds PromptDiffusion community support; `8739` adds ConsistencyTTA community support. Different community pipelines." + "reason": "Distinct community pipeline additions with different pipelines and goals." }, { "left": "pull_request:10105", "right": "pull_request:9103", "accept": false, - "reason": "`10105` fixes a broken README link; `9103` fixes distributed LR scheduler math. Documentation vs training logic." + "reason": "Broken doc link fix vs distributed LR scheduler logic correction; unrelated." }, { "left": "pull_request:1269", "right": "pull_request:1470", "accept": false, - "reason": "`1269` is a stable diffusion refactor; `1470` makes DreamBooth compatible with Alt Diffusion. Different feature work." + "reason": "Stable-diffusion refactor finalization vs DreamBooth AltDiffusion compatibility are separate changes." }, { "left": "pull_request:13294", "right": "pull_request:13365", "accept": false, - "reason": "`13294` addresses flaky WanVACE LoRA tests; `13365` updates the Claude review workflow for forks. Test stabilization vs CI permissions/workflow." + "reason": "Flaky WanVACE test cleanup vs fork-support CI workflow changes; no shared bug/change." + }, + { + "left": "pull_request:4365", + "right": "pull_request:4749", + "accept": false, + "reason": "Both mention SDXL LoRA, but one adds a new training example while the other fixes checkpoint saving/loading for an existing script." + }, + { + "left": "pull_request:4793", + "right": "pull_request:4862", + "accept": false, + "reason": "Auto-pipeline kwargs forwarding and ControlNet SDXL MultiControlNet height/width handling are unrelated bugs." + }, + { + "left": "pull_request:5085", + "right": "pull_request:5697", + "accept": false, + "reason": "Different LoRA issues: offloading incompatible checkpoints vs xformers-enabled PEFT scaling." + }, + { + "left": "pull_request:6237", + "right": "pull_request:6929", + "accept": false, + "reason": "Seed batching docs and community pipeline docs are different documentation topics with no shared code change." } ] }, @@ -160423,11 +160595,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 182, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -160444,6 +160616,10 @@ "pull_request:10326", "pull_request:10420", "pull_request:10430", + "pull_request:10585", + "pull_request:10648", + "pull_request:10784", + "pull_request:10810", "pull_request:10837", "pull_request:10894", "pull_request:11080", @@ -160451,23 +160627,17 @@ "pull_request:11366", "pull_request:11428", "pull_request:11639", + "pull_request:11647", + "pull_request:11778", "pull_request:11997", "pull_request:12024", "pull_request:12909", "pull_request:2954", "pull_request:3125", - "pull_request:4365", - "pull_request:4749", "pull_request:4780", - "pull_request:4793", - "pull_request:4862", "pull_request:5038", - "pull_request:5085", "pull_request:5202", - "pull_request:5697", "pull_request:5823", - "pull_request:6237", - "pull_request:6929", "pull_request:7096", "pull_request:7906", "pull_request:8618", @@ -160476,15 +160646,13 @@ "pull_request:8985", "pull_request:9015", "pull_request:9107", + "pull_request:9765", "pull_request:9769", + "pull_request:9807", "pull_request:9832", "pull_request:9934" ], "soft_pairs": [ - "pull_request:4365|pull_request:4749", - "pull_request:4793|pull_request:4862", - "pull_request:5085|pull_request:5697", - "pull_request:6237|pull_request:6929", "pull_request:2954|pull_request:3125", "pull_request:5038|pull_request:5823", "pull_request:7096|pull_request:7906", @@ -160505,186 +160673,197 @@ "pull_request:10322|pull_request:8985", "pull_request:10326|pull_request:8684", "pull_request:10420|pull_request:9015", - "pull_request:10430|pull_request:9769" + "pull_request:10430|pull_request:9769", + "pull_request:10585|pull_request:10784", + "pull_request:10648|pull_request:11647", + "pull_request:10784|pull_request:11778", + "pull_request:10784|pull_request:9765", + "pull_request:10810|pull_request:9807" ], - "prepared_review_unit_hash": "71e87d45913213156fb444684bb2e070ca57a4067f7c587bcc9ca0f578c0c40f", + "prepared_review_unit_hash": "ccc0b56992256f6a0d5018cd8b449985cfaaa5089c9c8ccefc008eaa2378418d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 97120, - "estimated_input_tokens": 24280, - "estimated_eval_tokens": 48816 + "soft_pair_count": 26, + "serialized_chars": 97099, + "estimated_input_tokens": 24275, + "estimated_eval_tokens": 48806 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is an overbroad mix of unrelated PRs: CI/workflow cleanups, docs-only updates, feature additions (Mochi, OmniGen, Hunyuan, Wan, etc.), and a few small bugfixes. The soft-edge candidates mostly share broad subsystems or filenames but not the same concrete change, so none should be merged as duplicates.", - "confidence": 0.93, + "summary": "All soft edges look like false positives: these PRs mostly cover unrelated docs, CI, tests, or distinct feature/fix work within broad subsystems. No single canonical duplicate pair emerges.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR cleanly represents the whole cluster; the items span many unrelated changes and the apparent similarities are mostly superficial subsystem overlap.", + "canonical_pr_reason": "No canonical PR: the cluster is heterogeneous, and the apparent similarities are mostly broad subsystem overlap rather than the same underlying change.", "best_issue_reason": null, - "best_pr_reason": "If a representative PR must be chosen, pull_request:9769 (Mochi T2V) is the strongest candidate because it is a substantial, self-contained feature PR with clear scope and broad content, unlike the many small docs/CI edits.", + "best_pr_reason": "PR 10148 is the strongest standalone representative by scope and review activity, but it is still not a duplicate anchor for the other items.", "soft_edge_verdicts": [ - { - "left": "pull_request:4365", - "right": "pull_request:4749", - "accept": false, - "reason": "Both are SDXL/PEFT-related docs or example updates, but they fix different workflows: one adds SDXL training support, the other fixes LoRA checkpoint loading. Shared test file alone is not enough." - }, - { - "left": "pull_request:4793", - "right": "pull_request:4862", - "accept": false, - "reason": "Both touch pipeline internals, but one fixes auto_pipeline config forwarding while the other fixes ControlNet img2img dimension handling. Different code paths and bugs." - }, - { - "left": "pull_request:5085", - "right": "pull_request:5697", - "accept": false, - "reason": "Both involve LoRA/offloading behavior, but the concrete fixes differ: incompatible LoRA checkpoint offloading vs xformers-enabled LoRA scaling in attention. Not the same bug." - }, - { - "left": "pull_request:6237", - "right": "pull_request:6929", - "accept": false, - "reason": "Both are docs-focused, but one is batched seed guidance and the other is community pipeline loading docs. No shared underlying change." - }, { "left": "pull_request:2954", "right": "pull_request:3125", "accept": false, - "reason": "Both are multi-control/video-generation related examples, but one adds TextToVideoZero and the other extends StableDiffusionControlNetInpaint. Different pipelines and functionality." + "reason": "Different changes: TextToVideoZero pipeline addition vs multi-control-image handling in ControlNet inpaint docs/example code." }, { "left": "pull_request:5038", "right": "pull_request:5823", "accept": false, - "reason": "Both concern LoRA API ergonomics, but one makes optional args explicit in Transformer2DModel while the other adds adapter_names to fuse_lora. Different change requests." + "reason": "Both relate to LoRA, but one is an API signature cleanup and the other adds adapter_names support; different code paths and goals." }, { "left": "pull_request:7096", "right": "pull_request:7906", "accept": false, - "reason": "Both are training-script/example additions, but ControlNet training enhancements and GLIGEN training are unrelated features." + "reason": "Both are training/examples work, but they target different projects and scripts; not the same bug or change." }, { "left": "pull_request:4780", "right": "pull_request:5202", "accept": false, - "reason": "Both relate to PEFT/LoRA support, but 4780 is broad PEFT integration work while 5202 only adds Kohya state-dict key mapping. Not mergeable as one PR." + "reason": "PEFT/LoRA-related, but one is broad PEFT integration and the other adds Kohya checkpoint key mapping; not mergeable as one fix." }, { "left": "pull_request:10837", "right": "pull_request:11428", "accept": false, - "reason": "Both are HunyuanVideo-related, but one adds SkyReels Hunyuan T2V/I2V while the other adds Hunyuan Video Framepack. They target different model variants and code paths." + "reason": "Both touch HunyuanVideo docs/pipeline surface area, but one adds SkyReels T2V/I2V and the other adds Framepack; distinct features." }, { "left": "pull_request:11639", "right": "pull_request:11997", "accept": false, - "reason": "Both touch Wan video pipelines, but one standardizes VAE sampling in a specific video-to-video pipeline while the other adds modular Wan I2V/FLF2V blocks. Different scope and implementation." + "reason": "Wan-related, but one standardizes v2v VAE sampling while the other adds modular I2V/FLF2V blocks; separate changes." }, { "left": "pull_request:12024", "right": "pull_request:12909", "accept": false, - "reason": "Both edit transformer_wan.py, but one is a wan2.2 follow-up across pipeline/tests and the other fixes context parallel behavior in Wan 2.1/animate attention. Not the same concrete bug." + "reason": "Both modify Wan transformer code, but they fix different issues: timestep reshaping vs context-parallel handling." }, { "left": "pull_request:10042", "right": "pull_request:9832", "accept": false, - "reason": "One merges a PEFT PR workflow into the main PR workflow, the other adds nightly quantization CI. Both are CI-related only at a very broad level." + "reason": "Both CI/workflow-related, but one removes a workflow and the other adds quantization nightly tests; unrelated." }, { "left": "pull_request:10114", "right": "pull_request:10894", "accept": false, - "reason": "One is a stable_audio formatting fix, the other is a docs toctree sorting change. No shared underlying issue." + "reason": "One is a formatting fix in Stable Audio; the other is docs toctree sorting. No shared bug or change." }, { "left": "pull_request:10115", "right": "pull_request:9107", "accept": false, - "reason": "Flux rotary embedding fix vs SD3 LoRA loading-hook fix: different models, different bugs, no common code path." + "reason": "Flux rotary embedding bugfix vs SD3 LoRA loading hook fix; different models and code paths." }, { "left": "pull_request:10132", "right": "pull_request:11366", "accept": false, - "reason": "Tokenizers backend update for PEFT tests vs Hidream transformer output-shape fix. Unrelated changes." + "reason": "Tokenizers update in PEFT backend workflow vs Hidream transformer output shape/API change; unrelated." }, { "left": "pull_request:10148", "right": "pull_request:11319", "accept": false, - "reason": "OmniGen feature addition and community-project docs entry are unrelated; one adds code and tests, the other is just documentation." + "reason": "OmniGen feature addition vs a community-projects docs entry; not the same underlying change." }, { "left": "pull_request:10148", "right": "pull_request:8618", "accept": false, - "reason": "OmniGen addition vs SD3 tiny autoencoder shift_factor change are distinct features in different subsystems." + "reason": "OmniGen is a large new pipeline/model feature; the other changes SD3 tiny autoencoder config. Different work." }, { "left": "pull_request:10245", "right": "pull_request:8685", "accept": false, - "reason": "PEFT docs wording update and float8 note in SD3 docs are unrelated documentation fixes." + "reason": "PEFT docs text change vs SD3 float8 note; both documentation-only but about different topics." }, { "left": "pull_request:10299", "right": "pull_request:9934", "accept": false, - "reason": "Both are docs changes mentioning model usage, but one updates a GGUF example and the other expands Mochi documentation. Different topics." + "reason": "GGUF docs generator placement fix vs Mochi docs expansion; unrelated documentation updates." }, { "left": "pull_request:10312", "right": "pull_request:11080", "accept": false, - "reason": "Attention.head_to_batch_dim bugfix and Kandinsky image resizing bugfix are separate code issues in different modules." + "reason": "Attention head_to_batch_dim bugfix vs Kandinsky image resize/divisibility fix; different subsystems and bugs." }, { "left": "pull_request:10318", "right": "pull_request:11319", "accept": false, - "reason": "Flux LoRA integration tests and community-project docs have no shared underlying bug or feature." + "reason": "Flux LoRA integration tests vs community projects docs entry; no shared implementation or bug." }, { "left": "pull_request:10322", "right": "pull_request:8985", "accept": false, - "reason": "Sana 2K converter support and Lumina test-size reduction are unrelated changes; one is a converter update, the other is test simplification." + "reason": "Sana 2K model conversion support vs reducing Lumina test size; unrelated." }, { "left": "pull_request:10326", "right": "pull_request:8684", "accept": false, - "reason": "MPS workflow quoting fix and Pillow import docs fix are unrelated housekeeping edits." + "reason": "MPS workflow quoting fix vs Pillow import docs fix; no common code-path problem." }, { "left": "pull_request:10420", "right": "pull_request:9015", "accept": false, - "reason": "LTX checkpoint-id docs fix and PIA example fix are different pipeline docs updates, not the same change." + "reason": "LTX checkpoint ID docs fix vs PIA example cleanup; unrelated doc updates." }, { "left": "pull_request:10430", "right": "pull_request:9769", "accept": false, - "reason": "TeaCache core infrastructure work and Mochi feature integration are different large changes; sharing broad modeling infrastructure does not make them duplicates." + "reason": "TeaCache core implementation vs Mochi T2V model/pipeline addition; not the same change." + }, + { + "left": "pull_request:10585", + "right": "pull_request:10784", + "accept": false, + "reason": "Mochi VAE annotation cleanup vs Lumina2 input validation fix; different models and bugs." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11647", + "accept": false, + "reason": "DDPM docs adding sigmoid schedule vs adding Finegrained FP8 quantization; unrelated." + }, + { + "left": "pull_request:10784", + "right": "pull_request:11778", + "accept": false, + "reason": "Lumina2 input validation vs Wan VACE mask padding; separate bugs in different pipelines." + }, + { + "left": "pull_request:10784", + "right": "pull_request:9765", + "accept": false, + "reason": "Lumina2 validation bugfix vs SD3 docstring typo fix; different scope and no shared code-path." + }, + { + "left": "pull_request:10810", + "right": "pull_request:9807", + "accept": false, + "reason": "Flux LoRA conversion improvements vs AutoencoderKL LoRA adapter support; both LoRA-related but distinct fixes." } ] }, @@ -160700,7 +160879,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 183, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -160708,10 +160887,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10585", - "pull_request:10648", - "pull_request:10784", - "pull_request:10810", "pull_request:10819", "pull_request:10834", "pull_request:10837", @@ -160726,7 +160901,6 @@ "pull_request:11583", "pull_request:11647", "pull_request:11722", - "pull_request:11778", "pull_request:1194", "pull_request:12018", "pull_request:12062", @@ -160743,26 +160917,26 @@ "pull_request:12434", "pull_request:12450", "pull_request:12551", + "pull_request:12631", + "pull_request:12644", + "pull_request:12675", + "pull_request:12686", "pull_request:12705", "pull_request:12797", + "pull_request:12856", "pull_request:13022", + "pull_request:13076", "pull_request:13164", "pull_request:13218", "pull_request:13454", + "pull_request:13562", "pull_request:2983", "pull_request:3172", "pull_request:45", "pull_request:9687", - "pull_request:9735", - "pull_request:9765", - "pull_request:9807" + "pull_request:9735" ], "soft_pairs": [ - "pull_request:10585|pull_request:10784", - "pull_request:10648|pull_request:11647", - "pull_request:10784|pull_request:11778", - "pull_request:10784|pull_request:9765", - "pull_request:10810|pull_request:9807", "pull_request:10819|pull_request:11564", "pull_request:10834|pull_request:11537", "pull_request:10837|pull_request:11066", @@ -160784,198 +160958,196 @@ "pull_request:12330|pull_request:13218", "pull_request:12360|pull_request:12434", "pull_request:12450|pull_request:13164", - "pull_request:12551|pull_request:13454" + "pull_request:12551|pull_request:13454", + "pull_request:12551|pull_request:13562", + "pull_request:12631|pull_request:13076", + "pull_request:12644|pull_request:12675", + "pull_request:12686|pull_request:12856" ], - "prepared_review_unit_hash": "5d480f210f2d16841c23e5bcdadfb097a305bfb125af8745708da49a30256b2b", + "prepared_review_unit_hash": "5c3af2a27dd4e7c14b52d9a7b5cbdd816132453e0582b76a0eebc19c30f5983b", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 27, - "serialized_chars": 95884, - "estimated_input_tokens": 23971, - "estimated_eval_tokens": 48198 + "soft_pair_count": 26, + "serialized_chars": 97146, + "estimated_input_tokens": 24287, + "estimated_eval_tokens": 48830 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous PRs across docs, CI, model support, and small bugfixes. I only found one true duplicate-style pair: the two Wan VACE bf16/fp16 fixes (12142/12143), which target the same mask casting bug in the same code path.", - "confidence": 0.89, + "summary": "This cluster is a set of largely unrelated PRs that only look superficially similar by subsystem or docs/CI overlap. I do not see any true duplicate/mergeable PR pairs here, so all soft edges should be rejected.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR for the whole cluster; most items are unrelated. If one representative is needed, 12143 is the strongest because it fixes the same Wan VACE dtype bug as 12142 and also adds test coverage.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "12143 is the best representative of the only real duplicate pair: it fixes the Wan VACE mask-casting bug in `pipeline_wan_vace.py` and includes the test adjustment, making it the more complete change.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:10585", - "right": "pull_request:10784", - "accept": false, - "reason": "Different components and bugs: Mochi VAE annotation vs Lumina2 input validation." - }, - { - "left": "pull_request:10648", - "right": "pull_request:11647", - "accept": false, - "reason": "Unrelated changes: DDPM docs for sigmoid beta schedule vs adding Finegrained FP8 quantization." - }, - { - "left": "pull_request:10784", - "right": "pull_request:11778", - "accept": false, - "reason": "Both touch pipeline validation/padding, but they fix different pipelines and different bugs." - }, - { - "left": "pull_request:10784", - "right": "pull_request:9765", - "accept": false, - "reason": "Lumina2 input validation vs SD3 docstring typo; not the same underlying change." - }, - { - "left": "pull_request:10810", - "right": "pull_request:9807", - "accept": false, - "reason": "Both involve LoRA support, but one is Flux conversion logic and the other is AutoencoderKL adapter support." - }, { "left": "pull_request:10819", "right": "pull_request:11564", "accept": false, - "reason": "Docs entry addition vs adding prodigy deps; no shared bug or change." + "reason": "Both touch docs/example-related content, but one adds LoRA docs entries while the other adds a dreambooth dependency; no shared bug or change." }, { "left": "pull_request:10834", "right": "pull_request:11537", "accept": false, - "reason": "GGUF parameter check bugfix vs torch.compile test addition; different code paths." + "reason": "GGUF isinstance fix vs adding a torch.compile test for UNet are unrelated code/test changes." }, { "left": "pull_request:10837", "right": "pull_request:11066", "accept": false, - "reason": "Both are Hunyuan-related, but they introduce different model/pipeline updates and are not the same fix." + "reason": "Both are HunyuanVideo-related, but one adds SkyReels Hunyuan T2V/I2V support and the other introduces a new HunyuanVideo-I2V model; different functionality." }, { "left": "pull_request:10885", "right": "pull_request:11647", "accept": false, - "reason": "CogVideoX scheduler docs vs Finegrained FP8 quantization; unrelated." + "reason": "Both are docs/feature additions, but CogVideoX schedulers and Finegrained FP8 quantization are separate subsystems and changes." }, { "left": "pull_request:11067", "right": "pull_request:12062", "accept": false, - "reason": "Both are CI-related, but one pins a benchmark dependency and the other enables all GPUs; different changes." + "reason": "Both modify CI/benchmarking, but one pins transformers for benchmarks while the other enables all GPUs in CI containers; not the same issue." }, { "left": "pull_request:11243", "right": "pull_request:9687", "accept": false, - "reason": "Musubi Wan LoRA support vs Xlabs ControlNet support; different features and code paths." + "reason": "Musubi Wan LoRA conversion support and Xlabs ControlNet support are different model-path features with different code paths." }, { "left": "pull_request:11290", "right": "pull_request:12420", "accept": false, - "reason": "Docker version unpinning vs conditional torch.distributed import; unrelated." + "reason": "Dockerfile torch pinning and conditional import of torch.distributed are unrelated fixes." }, { "left": "pull_request:11366", "right": "pull_request:9735", "accept": false, - "reason": "Hidream transformer output shape change vs SD3 ControlNet example update; different areas." + "reason": "Hidream transformer output-shape change and SD3 ControlNet example updates are unrelated." }, { "left": "pull_request:11583", "right": "pull_request:12134", "accept": false, - "reason": "Both are docs cleanup, but one removes a tutorial and the other refreshes a different stable diffusion doc." + "reason": "Both are docs removals/refreshes, but one removes the fast diffusion tutorial and the other rewrites the Stable Diffusion intro; different content." }, { "left": "pull_request:11722", "right": "pull_request:12233", "accept": false, - "reason": "Different documentation fixes: a misleading comment in a script vs a title/link text update." + "reason": "Both are documentation edits, but a misleading code comment in a script is not the same as renaming a quantization docs title." }, { "left": "pull_request:1194", "right": "pull_request:3172", "accept": false, - "reason": "Doc-builder workflow setup vs reproducibility docs; unrelated." + "reason": "Different doc-builder workflow updates years apart; one updates PR docs actions, the other documents deterministic algorithms." }, { "left": "pull_request:1194", "right": "pull_request:45", "accept": false, - "reason": "Both touch documentation workflows, but they address different workflow revisions and broad docs plumbing changes." + "reason": "Both touch doc workflows, but 45 is the initial docs workflow setup while 1194 changes action references/secrets; not the same change." }, { "left": "pull_request:12018", "right": "pull_request:12391", "accept": false, - "reason": "Minor link/text fix in quantization overview vs a large quantization docs reorganization; not the same change." + "reason": "A link-text fix in quantization overview is not the same as a broad quantization docs reorganization/rewrite." }, { "left": "pull_request:12018", "right": "pull_request:12797", "accept": false, - "reason": "Quantization link fix vs Kandinsky docs edits; unrelated." + "reason": "Quantization link text fix vs Kandinsky 5 docs fixes are unrelated." }, { "left": "pull_request:12142", "right": "pull_request:13022", "accept": false, - "reason": "Wan VACE dtype bugfix vs Dockerfile compatibility changes; no shared issue." + "reason": "WAN VACE bf16/fp16 mask handling and CUDA/xformers Dockerfile fixes are unrelated." }, { "left": "pull_request:12143", "right": "pull_request:13022", "accept": false, - "reason": "Wan VACE dtype bugfix vs Dockerfile compatibility changes; unrelated." + "reason": "The WAN VACE test/mask dtype fix is not the same as Dockerfile CUDA/xformers adjustments." }, { "left": "pull_request:12231", "right": "pull_request:12705", "accept": false, - "reason": "General typo fix in example docs vs a broader typo cleanup across many files; similar category but not the same underlying change." + "reason": "A broad typo cleanup across many example scripts is not the same change as a smaller typo cleanup in assorted docs/examples." }, { "left": "pull_request:1229", "right": "pull_request:2983", "accept": false, - "reason": "Dreambooth README wording change vs CONTRIBUTING guidance on test deps; unrelated docs." + "reason": "Dreambooth README dependency guidance and CONTRIBUTING test-dependency instructions are different documentation changes." }, { "left": "pull_request:12330", "right": "pull_request:13218", "accept": false, - "reason": "Pipeline test cleanup vs Helios-specific skipped test; different test issues." + "reason": "Both are test-related, but one cleans group offloading tests while the other skips an invalid Helios save/load case; different failures." }, { "left": "pull_request:12360", "right": "pull_request:12434", "accept": false, - "reason": "Chroma doc rewrite vs text-to-image PEFT version bump; unrelated." + "reason": "Chroma doc wording fixes and text-to-image training PEFT version changes are unrelated." }, { "left": "pull_request:12450", "right": "pull_request:13164", "accept": false, - "reason": "Both revise CI installs, but they target different workflows and different dependency cleanup goals." + "reason": "Both alter CI installs, but 12450 revisits overall CI installation strategy whereas 13164 removes test deps from CI jobs; not the same concrete change." }, { "left": "pull_request:12551", "right": "pull_request:13454", "accept": false, - "reason": "Different flaky-test mitigations for different models/issues: Sana layerwise casting vs CLIP-related xfails." + "reason": "Skipping Sana layerwise-casting tests in CI and xfailing CLIP-related pipeline tests address different failures." + }, + { + "left": "pull_request:12551", + "right": "pull_request:13562", + "accept": false, + "reason": "Sana-specific CI skips and tokenizers installation stability in CI are unrelated." + }, + { + "left": "pull_request:12631", + "right": "pull_request:13076", + "accept": false, + "reason": "Distributed inference docs and a quantization syntax fix are unrelated documentation changes." + }, + { + "left": "pull_request:12644", + "right": "pull_request:12675", + "accept": false, + "reason": "AutoModel docs and Sana video typo fixes are unrelated docs edits." + }, + { + "left": "pull_request:12686", + "right": "pull_request:12856", + "accept": false, + "reason": "Kandinsky5 pipeline implementation and z-image-omni-base implementation are completely different model introductions." } ] }, @@ -160991,7 +161163,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 184, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -160999,15 +161171,8 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:12551", - "pull_request:12631", - "pull_request:12644", - "pull_request:12675", "pull_request:12686", - "pull_request:12856", "pull_request:1292", - "pull_request:13076", - "pull_request:13562", "pull_request:13567", "pull_request:1397", "pull_request:1423", @@ -161029,29 +161194,33 @@ "pull_request:2365", "pull_request:2384", "pull_request:2405", + "pull_request:2490", + "pull_request:2598", + "pull_request:2811", + "pull_request:2817", "pull_request:2959", "pull_request:3006", + "pull_request:3172", "pull_request:3201", "pull_request:3226", "pull_request:326", "pull_request:3265", "pull_request:3331", "pull_request:3561", + "pull_request:3709", "pull_request:3716", "pull_request:396", "pull_request:416", "pull_request:533", "pull_request:564", "pull_request:570", + "pull_request:703", "pull_request:748", "pull_request:806", - "pull_request:840" + "pull_request:840", + "pull_request:975" ], "soft_pairs": [ - "pull_request:12551|pull_request:13562", - "pull_request:12631|pull_request:13076", - "pull_request:12644|pull_request:12675", - "pull_request:12686|pull_request:12856", "pull_request:12686|pull_request:13567", "pull_request:1292|pull_request:2405", "pull_request:1397|pull_request:1423", @@ -161073,192 +161242,196 @@ "pull_request:2365|pull_request:326", "pull_request:2384|pull_request:3201", "pull_request:2384|pull_request:3265", - "pull_request:2384|pull_request:533" + "pull_request:2384|pull_request:533", + "pull_request:2490|pull_request:3709", + "pull_request:2598|pull_request:3172", + "pull_request:2811|pull_request:703", + "pull_request:2817|pull_request:975" ], - "prepared_review_unit_hash": "055827450ef5dbf10a71773d02ad555da96e2962704cf869572adf60e91b2a53", + "prepared_review_unit_hash": "1022873c5465a39f544ced1f83a934a4116852901b1472880fc8bb92cee34ba2", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 101815, - "estimated_input_tokens": 25454, - "estimated_eval_tokens": 51164 + "serialized_chars": 101889, + "estimated_input_tokens": 25473, + "estimated_eval_tokens": 51202 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: it mixes unrelated docs-only edits, CI tweaks, model/pipeline feature work, and scheduler fixes. The soft edges are mostly superficial similarity (shared words or filenames) rather than the same underlying bug/change.", - "confidence": 0.98, + "summary": "All soft edges are false positives: the paired PRs are mostly unrelated docs edits, feature additions, or separate reverts in different code paths. No duplicate cluster emerges from this set.", + "confidence": 0.99, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No issue artifacts are present. If a PR must be chosen as the closest representative of the set, #2009 is the strongest core-library change with broad downstream impact and high discussion activity, but it is not a duplicate of the others.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "#2009 is the best representative PR because it is a substantial, central library change affecting loaders and many pipelines; the rest are mostly isolated docs/typo/test/revert changes.", "soft_edge_verdicts": [ - { - "left": "pull_request:12551", - "right": "pull_request:13562", - "accept": false, - "reason": "12551 skips SANA layerwise-casting tests in CI; 13562 changes CI dependency installation for tokenizers. Different CI concerns." - }, - { - "left": "pull_request:12631", - "right": "pull_request:13076", - "accept": false, - "reason": "12631 is a distributed inference doc update; 13076 fixes a TorchAo quantization config syntax example. Different docs and different technical issues." - }, - { - "left": "pull_request:12644", - "right": "pull_request:12675", - "accept": false, - "reason": "12644 adds AutoModel docs; 12675 fixes typos in the Sana video docs. Same broad area (docs) but unrelated content." - }, - { - "left": "pull_request:12686", - "right": "pull_request:12856", - "accept": false, - "reason": "12686 introduces Kandinsky5 pipelines; 12856 adds z-image-omni-base implementation. Different model families and codepaths." - }, { "left": "pull_request:12686", "right": "pull_request:13567", "accept": false, - "reason": "12686 is a pipeline implementation PR; 13567 only updates agent guidance docs. Unrelated." + "reason": "Kandinsky5 pipeline docs/code vs agents documentation update; unrelated scopes." }, { "left": "pull_request:1292", "right": "pull_request:2405", "accept": false, - "reason": "1292 adds OneFlow support; 2405 reverts a release version bump. No shared underlying bug/change." + "reason": "Both are reverts, but they revert different changes in different areas and do not fix the same bug." }, { "left": "pull_request:1397", "right": "pull_request:1423", "accept": false, - "reason": "1397 adds SD2 attention slicing support; 1423 adds Flax support for SD2. Same product area, but distinct features and codepaths." + "reason": "SD2 attention slicing vs Flax SD2 support; different concrete changes." }, { "left": "pull_request:1437", "right": "pull_request:3331", "accept": false, - "reason": "Both are typo/doc cleanups, but in different files and with different text. Not the same change." + "reason": "Separate typo fixes in different pipeline/docs files; not the same issue." }, { "left": "pull_request:1508", "right": "pull_request:3561", "accept": false, - "reason": "1508 clarifies DreamBooth prior-preservation docs; 3561 changes DreamBooth IF stage II upscaling training code. Different problems." + "reason": "Dreambooth docs clarification vs Dreambooth IF stage II upscaling; unrelated changes." }, { "left": "pull_request:1508", "right": "pull_request:748", "accept": false, - "reason": "1508 is a docs clarification; 748 fixes push_to_hub calls in training scripts. Different issues." + "reason": "Docs clarification vs push_to_hub fix; different bugs." }, { "left": "pull_request:1697", "right": "pull_request:1951", "accept": false, - "reason": "1697 updates a community composable diffusion example; 1951 updates ColossalAI training integration. Unrelated." + "reason": "Community pipeline update vs ColossalAI update; no shared concrete code-path problem." }, { "left": "pull_request:1765", "right": "pull_request:564", "accept": false, - "reason": "1765 is a DreamBooth Flax hotfix; 564 is a Flax scheduler fix. Different codepaths and bugs." + "reason": "Dreambooth Flax fixes and Flax scheduler fixes are different subsystems and bugs." }, { "left": "pull_request:1788", "right": "pull_request:570", "accept": false, - "reason": "1788 fixes DDIM behavior when prediction_type is sample; 570 fixes BaseOutput initialization from dict. Completely different subsystems." + "reason": "DDIM sampler bug fix vs BaseOutput dict initialization; unrelated code paths." }, { "left": "pull_request:1850", "right": "pull_request:2959", "accept": false, - "reason": "1850 adds ResnetBlock2D tests; 2959 adjusts a UNet3D condition test. Similar test area, but not the same behavior." + "reason": "Test additions for different models; not the same underlying issue." }, { "left": "pull_request:1903", "right": "pull_request:3006", "accept": false, - "reason": "1903 only renames comments in depth2img; 3006 fixes DDPM variance math. Different scope and code." + "reason": "Comment cleanup in depth2img vs DDPM variance logic fix; unrelated." }, { "left": "pull_request:2009", "right": "pull_request:2183", "accept": false, - "reason": "2009 adds textual inversion loading support; 2183 is a logging hotfix in textual inversion training. Different changes." + "reason": "Textual inversion embedding loading vs logging hotfix; different functionality." }, { "left": "pull_request:2103", "right": "pull_request:806", "accept": false, - "reason": "2103 adds lora tags to model cards; 806 is an early WIP adding EMA training options. Not the same underlying change." + "reason": "LoRA model tag docs vs Dreambooth EMA training work; no duplicate bug/change." }, { "left": "pull_request:2129", "right": "pull_request:3716", "accept": false, - "reason": "2129 adds LoRA plumbing in pipelines; 3716 adds ParaDiGMS parallel sampling. Different features and mergeable units." + "reason": "LoRA pipeline support vs ParaDiGMS pipeline addition; distinct features." }, { "left": "pull_request:2257", "right": "pull_request:840", "accept": false, - "reason": "2257 adds an UnCLIP text interpolation community pipeline; 840 adds a dummy one-step UNet example. Unrelated examples." + "reason": "Both are community examples, but they implement different pipelines and changes." }, { "left": "pull_request:2330", "right": "pull_request:3226", "accept": false, - "reason": "2330 adds ethical guidelines docs; 3226 adds ControlNet v1.1 docs. Different documentation topics." + "reason": "Ethical guidelines docs vs ControlNet v1.1 docs; unrelated documentation updates." }, { "left": "pull_request:2349", "right": "pull_request:396", "accept": false, - "reason": "2349 documents KarrasDiffusionSchedulers; 396 adds DDIM docs. Distinct docs pages and subjects." + "reason": "KarrasDiffusionSchedulers overview docs vs DDIM docs; different documentation targets." }, { "left": "pull_request:2349", "right": "pull_request:416", "accept": false, - "reason": "2349 is scheduler documentation; 416 is broader models documentation. No shared underlying change." + "reason": "KarrasDiffusionSchedulers docs note vs general models docs; unrelated." }, { "left": "pull_request:2365", "right": "pull_request:326", "accept": false, - "reason": "2365 adds a note about KarrasDiffusionSchedulers; 326 is a type-hint PR for PNDM pipeline. Not the same bug/change." + "reason": "Type-note change for Karras schedulers vs PNDM pipeline typing; not the same change." }, { "left": "pull_request:2384", "right": "pull_request:3201", "accept": false, - "reason": "2384 reverts a Pix2Pix Zero helper; 3201 reverts a different community pipeline update. Different codepaths." + "reason": "Revert of a Pix2Pix Zero utility vs revert of lpw pipeline update; different underlying PRs." }, { "left": "pull_request:2384", "right": "pull_request:3265", "accept": false, - "reason": "2384 reverts Pix2Pix Zero code; 3265 reverts a community lpw_stable_diffusion change. Different artifacts." + "reason": "Different revert targets and code paths; not duplicates." }, { "left": "pull_request:2384", "right": "pull_request:533", "accept": false, - "reason": "2384 is a Pix2Pix Zero revert; 533 reverts DDIM typehint changes. Unrelated." + "reason": "Pix2Pix Zero revert vs DDIM scheduler typehint revert; unrelated." + }, + { + "left": "pull_request:2490", + "right": "pull_request:3709", + "accept": false, + "reason": "MultiDiffusion docs addition vs image processor docs; different documentation areas." + }, + { + "left": "pull_request:2598", + "right": "pull_request:3172", + "accept": false, + "reason": "Paint By Example docs vs deterministic algorithms docs; no shared bug or change." + }, + { + "left": "pull_request:2811", + "right": "pull_request:703", + "accept": false, + "reason": "ONNX test expectation update vs CI workflow change; unrelated." + }, + { + "left": "pull_request:2817", + "right": "pull_request:975", + "accept": false, + "reason": "UNet1D documentation vs OpenVINO title typo; different docs changes." } ] }, @@ -161274,7 +161447,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 185, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -161282,13 +161455,8 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2490", - "pull_request:2598", - "pull_request:2811", - "pull_request:2817", "pull_request:2894", "pull_request:3042", - "pull_request:3172", "pull_request:3183", "pull_request:3226", "pull_request:3294", @@ -161301,7 +161469,6 @@ "pull_request:3608", "pull_request:3673", "pull_request:3681", - "pull_request:3709", "pull_request:3786", "pull_request:3806", "pull_request:382", @@ -161319,22 +161486,25 @@ "pull_request:4155", "pull_request:416", "pull_request:417", + "pull_request:4196", "pull_request:4211", + "pull_request:4227", + "pull_request:4259", + "pull_request:4284", "pull_request:4347", + "pull_request:4498", + "pull_request:4652", "pull_request:5024", "pull_request:5201", + "pull_request:5384", "pull_request:5505", + "pull_request:5698", "pull_request:5747", "pull_request:5768", - "pull_request:6485", - "pull_request:703", - "pull_request:975" + "pull_request:6163", + "pull_request:6485" ], "soft_pairs": [ - "pull_request:2490|pull_request:3709", - "pull_request:2598|pull_request:3172", - "pull_request:2811|pull_request:703", - "pull_request:2817|pull_request:975", "pull_request:2894|pull_request:3226", "pull_request:3042|pull_request:4078", "pull_request:3183|pull_request:3541", @@ -161357,198 +161527,209 @@ "pull_request:388|pull_request:416", "pull_request:390|pull_request:391", "pull_request:4058|pull_request:4155", - "pull_request:4058|pull_request:6485" + "pull_request:4058|pull_request:6485", + "pull_request:4196|pull_request:4227", + "pull_request:4259|pull_request:5698", + "pull_request:4284|pull_request:5768", + "pull_request:4498|pull_request:5384", + "pull_request:4652|pull_request:6163" ], - "prepared_review_unit_hash": "0aeb93e405673d94d77f0b6645ce30f82b2439f7d55464242911eca3588c2abf", + "prepared_review_unit_hash": "66a990464013a2f10242ab1f63b8b62b3d50712f28b1e93f7664883328c924d6", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 27, - "serialized_chars": 92529, - "estimated_input_tokens": 23133, - "estimated_eval_tokens": 46522 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 95607, + "estimated_input_tokens": 23902, + "estimated_eval_tokens": 48060 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a heterogeneous mix of unrelated docs, example, CI, and code-fix PRs. None of the proposed soft pairs look like the same underlying change, so all are rejected.", - "confidence": 0.14, + "summary": "This cluster is a false positive: the PRs span unrelated docs edits, example-script fixes, dependency bumps, loader changes, and scheduler/controlnet logic changes. There is no single underlying bug or change family to deduplicate.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR stands out; the items are mostly unrelated standalone changes rather than duplicate implementations of one fix.", + "canonical_pr_reason": "No canonical PR fits the cluster; the items are too heterogeneous across unrelated files and code paths, so any representative would be arbitrary.", "best_issue_reason": null, - "best_pr_reason": "PR 3681 is the broadest substantive fix in the set and the most plausible representative if one PR must be chosen, but it is not a duplicate of the others.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:2490", - "right": "pull_request:3709", - "accept": false, - "reason": "Both add docs, but for different topics: MultiDiffusion panorama vs image processor reference docs." - }, - { - "left": "pull_request:2598", - "right": "pull_request:3172", - "accept": false, - "reason": "One updates PaintByExample docs; the other adds reproducibility guidance about deterministic algorithms." - }, - { - "left": "pull_request:2811", - "right": "pull_request:703", - "accept": false, - "reason": "An ONNX test expectation update is unrelated to CI workflow changes for torch/cuda cache." - }, - { - "left": "pull_request:2817", - "right": "pull_request:975", - "accept": false, - "reason": "UNet1D documentation for extra channels is unrelated to a doc title typo in the toctree." - }, { "left": "pull_request:2894", "right": "pull_request:3226", "accept": false, - "reason": "Adding a Flax requirements file is not the same change as adding ControlNet v1.1 docs." + "reason": "Both are docs/example-related, but one adds a flax requirements file while the other edits ControlNet docs; different changes and no shared bug." }, { "left": "pull_request:3042", "right": "pull_request:4078", "accept": false, - "reason": "Both mention accelerate, but one bumps example requirements while the other enforces accelerate in a single-file converter path." + "reason": "Both involve accelerate, but 3042 raises a minimum version across many examples while 4078 only adds an install guard for from_single_file; different issues." }, { "left": "pull_request:3183", "right": "pull_request:3541", "accept": false, - "reason": "Both are DreamBooth fixes, but one corrects LoRA checkpoint registration and the other fixes attention-mask collation." + "reason": "Different DreamBooth problems: one fixes LoRA checkpointing, the other fixes attention mask collation in the base training script." }, { "left": "pull_request:3294", "right": "pull_request:5201", "accept": false, - "reason": "LoRA weight loading and xformers/attention-processor LoRA handling are different code paths and bug fixes." + "reason": "Both touch LoRA internals, but 3294 adds lora prefix loading support and 5201 fixes xformers/processor offload behavior; not the same concrete change." }, { "left": "pull_request:3295", "right": "pull_request:5505", "accept": false, - "reason": "These are separate documentation typos in different tutorials and files." + "reason": "These are unrelated typo/docs fixes in different files and subsystems." }, { "left": "pull_request:3331", "right": "pull_request:4211", "accept": false, - "reason": "Both touch inpainting docs, but one is a simple typo and the other is an SDXL docstring/example fix." + "reason": "Both are doc typos, but they target different SDXL/inpainting docs and different example text; no same underlying bug." }, { "left": "pull_request:3410", "right": "pull_request:4078", "accept": false, - "reason": "Both relate to accelerate availability, but one fixes deepspeed_plugin retrieval in training and the other guards from_single_file conversion." + "reason": "One fixes deepspeed plugin retrieval in a training script; the other makes accelerate mandatory for single-file loading. Different code paths." }, { "left": "pull_request:3462", "right": "pull_request:3673", "accept": false, - "reason": "One makes DreamBooth LoRA more robust to the original UNet; the other fixes multi-GPU handling in DreamBooth." + "reason": "Same DreamBooth script area, but one changes robustness around orig UNet handling and the other fixes multi-GPU behavior via unwrap_model; distinct bugs." }, { "left": "pull_request:3589", "right": "pull_request:5024", "accept": false, - "reason": "Community reference-tricks docs and a scheduler step-index bug are unrelated changes." + "reason": "Unrelated topics: reference-trick docs/example updates versus scheduler step-index bookkeeping." }, { "left": "pull_request:3608", "right": "pull_request:5505", "accept": false, - "reason": "These are distinct typo fixes in different files and contexts." + "reason": "Both are typo-related, but they affect different examples/docs and different text; not a duplicate change." }, { "left": "pull_request:3681", "right": "pull_request:4132", "accept": false, - "reason": "Checkpoint-limit handling in training scripts is unrelated to making accelerate optional in checkpoint conversion." + "reason": "Both concern accelerate usage, but 3681 replaces project configuration handling in training scripts while 4132 makes accelerate optional in checkpoint conversion; different problems." }, { "left": "pull_request:3786", "right": "pull_request:4347", "accept": false, - "reason": "PR template content and scheduler step-index logic have no shared bug or change." + "reason": "PR template changes are unrelated to scheduler step-index logic." }, { "left": "pull_request:3806", "right": "pull_request:5747", "accept": false, - "reason": "One adds a checkpoint-loading note to README; the other adds a dataset data_dir parameter in training scripts." + "reason": "One is a README note about loading from checkpoints; the other adds a data_dir parameter to dataset loading. Different concerns." }, { "left": "pull_request:382", "right": "pull_request:392", "accept": false, - "reason": "DDPM pipeline docs are unrelated to the conceptual contribution page." + "reason": "Both are early docs expansions, but they cover different pages and unrelated content areas." }, { "left": "pull_request:382", "right": "pull_request:404", "accept": false, - "reason": "Stable Diffusion pipeline docs and fp16 optimization docs are different documentation pages with different scope." + "reason": "Stable Diffusion pipeline docs vs fp16 optimization docs are separate documentation changes, not the same issue." }, { "left": "pull_request:382", "right": "pull_request:417", "accept": false, - "reason": "DDPM docs and general inference-pipeline docs are separate documentation efforts." + "reason": "These are different documentation pages for different user guides; no shared underlying change." }, { "left": "pull_request:3820", "right": "pull_request:5768", "accept": false, - "reason": "SDXL doc typo fixes and consistency-decoder documentation are unrelated changes." + "reason": "LDM3D typo/docs cleanup is unrelated to consistency decoder documentation fixes." }, { "left": "pull_request:386", "right": "pull_request:391", "accept": false, - "reason": "Stable Diffusion pipeline docs are not the same as PNDM pipeline docs." + "reason": "Two different pipeline docs (Stable Diffusion vs PNDM); same general docs theme but not the same change." }, { "left": "pull_request:388", "right": "pull_request:396", "accept": false, - "reason": "Unconditional latent diffusion docs and DDIM docs describe different pipelines and papers." + "reason": "Different pipeline documentation pages and code references; no shared bug or feature." }, { "left": "pull_request:388", "right": "pull_request:416", "accept": false, - "reason": "A latent-diffusion docs page is unrelated to the broad models API documentation update." + "reason": "Unconditional latent diffusion docs versus model API docs are unrelated." }, { "left": "pull_request:390", "right": "pull_request:391", "accept": false, - "reason": "Apple Silicon MPS optimization docs are unrelated to PNDM pipeline docs." + "reason": "Different optimization/docs pages (MPS hardware guidance vs PNDM pipeline docs)." }, { "left": "pull_request:4058", "right": "pull_request:4155", "accept": false, - "reason": "One is a doc-only SDXL correction; the other is an unfinished ControlNet SDXL pipeline fix affecting code and tests." + "reason": "SDXL docs cleanup and SDXL ControlNet pipeline follow-up fixes are not the same change; one is documentation-only, the other changes code and tests." }, { "left": "pull_request:4058", "right": "pull_request:6485", "accept": false, - "reason": "SDXL docs and PR-template reviewer metadata are unrelated." + "reason": "SDXL docs correction and PR template reviewer update are unrelated." + }, + { + "left": "pull_request:4196", + "right": "pull_request:4227", + "accept": false, + "reason": "SDXL docs improvements versus a missing import in textual inversion inference are different files and bugs." + }, + { + "left": "pull_request:4259", + "right": "pull_request:5698", + "accept": false, + "reason": "Circular-import fix for from_single_file loaders is unrelated to latent-consistency img2img behavior." + }, + { + "left": "pull_request:4284", + "right": "pull_request:5768", + "accept": false, + "reason": "Both are LDM3D-related docs fixes, but they touch different files and separate typos; not the same concrete issue." + }, + { + "left": "pull_request:4498", + "right": "pull_request:5384", + "accept": false, + "reason": "Different feature work: SDXL reference control support versus basic hypernetwork support for StableDiffusionPipeline." + }, + { + "left": "pull_request:4652", + "right": "pull_request:6163", + "accept": false, + "reason": "Both are LoRA tests, but they cover different test suites and behaviors; not a single fix that could be merged into one PR." } ] }, @@ -161564,7 +161745,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 186, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -161572,22 +161753,14 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:4196", - "pull_request:4227", - "pull_request:4259", - "pull_request:4284", - "pull_request:4498", - "pull_request:4652", "pull_request:4669", "pull_request:5188", "pull_request:5239", "pull_request:5242", - "pull_request:5384", "pull_request:5456", "pull_request:5552", "pull_request:5698", "pull_request:5707", - "pull_request:5768", "pull_request:5771", "pull_request:5822", "pull_request:5895", @@ -161597,7 +161770,6 @@ "pull_request:6053", "pull_request:6085", "pull_request:6119", - "pull_request:6163", "pull_request:6192", "pull_request:6205", "pull_request:6215", @@ -161608,10 +161780,18 @@ "pull_request:6643", "pull_request:6656", "pull_request:6775", + "pull_request:6997", + "pull_request:7077", + "pull_request:7103", + "pull_request:7124", + "pull_request:7153", + "pull_request:7166", + "pull_request:7206", "pull_request:7319", "pull_request:7328", "pull_request:7410", "pull_request:7634", + "pull_request:7662", "pull_request:7827", "pull_request:7941", "pull_request:8229", @@ -161619,14 +161799,10 @@ "pull_request:8303", "pull_request:8425", "pull_request:8510", - "pull_request:8521" + "pull_request:8521", + "pull_request:9247" ], "soft_pairs": [ - "pull_request:4196|pull_request:4227", - "pull_request:4259|pull_request:5698", - "pull_request:4284|pull_request:5768", - "pull_request:4498|pull_request:5384", - "pull_request:4652|pull_request:6163", "pull_request:4669|pull_request:6437", "pull_request:5188|pull_request:7328", "pull_request:5239|pull_request:6775", @@ -161650,156 +161826,131 @@ "pull_request:6215|pull_request:8303", "pull_request:6217|pull_request:8510", "pull_request:6308|pull_request:8229", - "pull_request:6342|pull_request:8521" + "pull_request:6342|pull_request:8521", + "pull_request:6997|pull_request:7206", + "pull_request:7077|pull_request:7124", + "pull_request:7103|pull_request:9247", + "pull_request:7153|pull_request:7662", + "pull_request:7166|pull_request:9247" ], - "prepared_review_unit_hash": "dc2aacb0b69ae09fd33cc7cc942f8d9c72dd972520522a2190470a5a15c81585", + "prepared_review_unit_hash": "6fa1018a61a6f3798a85e5d204a369fdaa700fea11f9559aef4134e39f32ef9a", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 97613, - "estimated_input_tokens": 24404, - "estimated_eval_tokens": 49064 + "serialized_chars": 96959, + "estimated_input_tokens": 24240, + "estimated_eval_tokens": 48736 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated docs-only edits, test tweaks, training-script changes, scheduler fixes, and pipeline/model features. None of the proposed soft pairs look like the same underlying bug or change, so all soft edges should be rejected.", + "summary": "This cluster is heterogeneous: it mixes unrelated docs, CI, training-script, scheduler, pipeline, and loader changes. None of the soft pairs look like the same underlying bug or mergeable change, so there is no duplicate group to collapse.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No clear canonical PR: the cluster is a grab bag of mostly unrelated changes. If forced to pick one representative, PR 5771 is the broadest foundational code change, but it is not a duplicate umbrella for the rest.", + "canonical_pr_reason": "No canonical PR: the items span many unrelated concerns, and no single PR meaningfully represents the rest as a duplicate fix/change.", "best_issue_reason": null, - "best_pr_reason": "PR 5771 is the best single representative by breadth and centrality because it refactors loaders and touches many downstream files, but it still doesn't match the other items as a duplicate target.", + "best_pr_reason": "No PR is a strong global representative; the largest/broadest PRs are still about different subsystems and goals, not the same concrete code-path problem.", "soft_edge_verdicts": [ - { - "left": "pull_request:4196", - "right": "pull_request:4227", - "accept": false, - "reason": "SDXL docs cleanup vs a missing import in a textual inversion README; unrelated artifacts and code paths." - }, - { - "left": "pull_request:4259", - "right": "pull_request:5698", - "accept": false, - "reason": "Circular import fix in loaders vs an LCM img2img pipeline fix; different bugs and different subsystems." - }, - { - "left": "pull_request:4284", - "right": "pull_request:5768", - "accept": false, - "reason": "LDM3D documentation typo cleanup vs consistency decoder docs cleanup; both docs-only, but not the same change." - }, - { - "left": "pull_request:4498", - "right": "pull_request:5384", - "accept": false, - "reason": "SDXL reference/adain feature addition vs hypernetwork support loader work; different code paths and goals." - }, - { - "left": "pull_request:4652", - "right": "pull_request:6163", - "accept": false, - "reason": "Both are LoRA tests, but one is an SDXL test fix and the other adds a separate xformers test; not the same concrete bug." - }, { "left": "pull_request:4669", "right": "pull_request:6437", "accept": false, - "reason": "LoRA rank-handling bugfix vs disabling a test in PEFT CI; one fixes functionality, the other adjusts test selection." + "reason": "Different problems: LoRA rank handling in text-encoder loading vs excluding a PEFT-specific test from CI." }, { "left": "pull_request:5188", "right": "pull_request:7328", "accept": false, - "reason": "Training utility refactor vs adding torch to a publish workflow; completely unrelated changes." + "reason": "Unrelated scopes: adding compute_snr to training utils vs adding torch to a PyPI workflow step." }, { "left": "pull_request:5239", "right": "pull_request:6775", "accept": false, - "reason": "AutoencoderKL encode docstring vs DDPM scheduler parameter docs; both docs, but unrelated components." + "reason": "Different artifacts: AutoencoderKL encode docstring vs DDPMScheduler parameter docs." }, { "left": "pull_request:5242", "right": "pull_request:6775", "accept": false, - "reason": "AutoencoderKL decode docstring vs DDPM scheduler parameter docs; unrelated documentation updates." + "reason": "Different artifacts: AutoencoderKL decode docstring vs DDPMScheduler parameter docs." }, { "left": "pull_request:5456", "right": "pull_request:6192", "accept": false, - "reason": "New RealFill example project vs adding unload_ip_adapter; separate features in different areas." + "reason": "RealFill example/package addition vs IP-Adapter unload behavior in loaders are unrelated changes." }, { "left": "pull_request:5456", "right": "pull_request:6656", "accept": false, - "reason": "RealFill example project vs textual inversion unload API; unrelated functionality." + "reason": "Different features: RealFill example vs textual inversion unload API." }, { "left": "pull_request:5456", "right": "pull_request:7634", "accept": false, - "reason": "RealFill example project vs high-res-fix UNet/pipeline addition; different features." + "reason": "Different domains: RealFill example vs high-res UNet fix implementation." }, { "left": "pull_request:5552", "right": "pull_request:6085", "accept": false, - "reason": "FreeU disable-method fix vs scheduler add_noise bugfix; different code paths." + "reason": "FreeU disable-method bugfix vs scheduler add_noise indexing bugfix; different code paths." }, { "left": "pull_request:5698", "right": "pull_request:5707", "accept": false, - "reason": "The former fixes an LCM img2img pipeline bug; the latter only adds tests across pipelines." + "reason": "LCM img2img pipeline fix vs separate LCM test additions; not the same change or bug." }, { "left": "pull_request:5707", "right": "pull_request:5822", "accept": false, - "reason": "LCM test expansion vs adding LCM support in ControlNet/Adapter pipelines; tests are not the same as the implementation bugfix." + "reason": "Test coverage for LCM vs ControlNet/Adapter support for LCM; different concrete implementations." }, { "left": "pull_request:5707", "right": "pull_request:5895", "accept": false, - "reason": "LCM tests vs the large Stable Video Diffusion feature addition; unrelated." + "reason": "LCM tests vs SVD feature PR; no shared underlying fix." }, { "left": "pull_request:5771", "right": "pull_request:7410", "accept": false, - "reason": "Loader refactor vs FreeInit behavior fix; unrelated areas and problems." + "reason": "LoRA loaders refactor vs FreeInit behavior fix; unrelated code paths and goals." }, { "left": "pull_request:5895", "right": "pull_request:6053", "accept": false, - "reason": "SVD feature addition vs logging step loss in a training script; unrelated." + "reason": "Stable Video Diffusion feature work vs step_loss logging in a training example; not the same bug/change." }, { "left": "pull_request:5982", "right": "pull_request:6053", "accept": false, - "reason": "SDXL Turbo docs vs training-loss logging; different artifacts and purposes." + "reason": "SDXL Turbo docs vs training loss logging; unrelated documentation and example code." }, { "left": "pull_request:5982", "right": "pull_request:6643", "accept": false, - "reason": "Both touch docs/toctree, but one is SDXL Turbo docs and the other is UViT2D docs; not the same change." + "reason": "Both are docs, but for different models/pages (SDXL Turbo vs UViT2D), so not duplicates." }, { "left": "pull_request:5988", @@ -161811,49 +161962,79 @@ "left": "pull_request:6002", "right": "pull_request:8230", "accept": false, - "reason": "Stable Video Diffusion guidance-scale logic fix vs CPU offloading usage docs; different bug domains." + "reason": "SVD guidance-scale fix vs CPU offloading usage/docs cleanup; different problems." }, { "left": "pull_request:6119", "right": "pull_request:8425", "accept": false, - "reason": "LoRA fp16 gradient fix vs a workflow-name rename; unrelated." + "reason": "LoRA training fp16 fix vs workflow rename; no shared code path." }, { "left": "pull_request:6205", "right": "pull_request:7827", "accept": false, - "reason": "T2I-Adapter blog URL correction vs a ControlNet docstring CPU-offload tweak; unrelated docs." + "reason": "T2I-Adapter blog URL docs fix vs ControlNet docstring CPU-offload example fix; unrelated documentation edits." }, { "left": "pull_request:6215", "right": "pull_request:7941", "accept": false, - "reason": "PixArt test initialization fix vs marking a single-file test class slow; not the same bug." + "reason": "PixArt test seeding/expected values vs marking a single-file test class slow; different test concerns." }, { "left": "pull_request:6215", "right": "pull_request:8303", "accept": false, - "reason": "PixArt test changes vs a philosophy.md typo fix; unrelated." + "reason": "PixArt test determinism fix vs philosophy typo edit; unrelated." }, { "left": "pull_request:6217", "right": "pull_request:8510", "accept": false, - "reason": "SVD torch-compile VAE fix vs dreambooth LR scheduler logic change; different code paths." + "reason": "SVD torch.compile VAE fix vs DreamBooth LR scheduler fix; different model/pipeline issues." }, { "left": "pull_request:6308", "right": "pull_request:8229", "accept": false, - "reason": "LoRA PEFT dummy-component test setup vs SVD resize behavior fix; unrelated." + "reason": "LoRA PEFT dummy-component test fixes vs SVD VideoProcessor resize fix; unrelated test/setup vs runtime behavior." }, { "left": "pull_request:6342", "right": "pull_request:8521", "accept": false, - "reason": "Installation docs update vs requirements file formatting fix; both maintenance but not the same change." + "reason": "Installation docs note vs requirements file formatting; different files and concerns." + }, + { + "left": "pull_request:6997", + "right": "pull_request:7206", + "accept": false, + "reason": "Draft LoRA support for UNetSpatioTemporalConditionModel vs training-script tagging metadata; unrelated." + }, + { + "left": "pull_request:7077", + "right": "pull_request:7124", + "accept": false, + "reason": "IP-Adapter head_to_batch_dim fix vs DreamBooth LoRA kwarg rename; different code paths." + }, + { + "left": "pull_request:7103", + "right": "pull_request:9247", + "accept": false, + "reason": "ControlNet check_image bugfix vs pipeline_loading_utils variant detection bugfix; different subsystems and failures." + }, + { + "left": "pull_request:7153", + "right": "pull_request:7662", + "accept": false, + "reason": "Documentation workflow path filter fix vs removing redundant packages from flax PR tests; unrelated CI changes." + }, + { + "left": "pull_request:7166", + "right": "pull_request:9247", + "accept": false, + "reason": "SDXL LoRA loading fix vs variant-folder identification fix; different loader logic and symptoms." } ] }, @@ -161869,7 +162050,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 187, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -161890,25 +162071,27 @@ "pull_request:13055", "pull_request:13365", "pull_request:3770", + "pull_request:3797", + "pull_request:4262", + "pull_request:4370", "pull_request:4498", + "pull_request:5331", + "pull_request:570", "pull_request:5807", + "pull_request:5813", + "pull_request:6237", "pull_request:6316", - "pull_request:6997", - "pull_request:7077", - "pull_request:7103", - "pull_request:7124", - "pull_request:7153", - "pull_request:7166", + "pull_request:6893", "pull_request:7181", - "pull_request:7206", + "pull_request:7410", "pull_request:7638", "pull_request:7652", "pull_request:7653", - "pull_request:7662", "pull_request:7893", "pull_request:7913", "pull_request:7919", "pull_request:7938", + "pull_request:826", "pull_request:8272", "pull_request:8322", "pull_request:8386", @@ -161919,18 +162102,12 @@ "pull_request:8702", "pull_request:8722", "pull_request:9100", - "pull_request:9247", "pull_request:9408", "pull_request:9620", "pull_request:9721", "pull_request:9735" ], "soft_pairs": [ - "pull_request:6997|pull_request:7206", - "pull_request:7077|pull_request:7124", - "pull_request:7103|pull_request:9247", - "pull_request:7153|pull_request:7662", - "pull_request:7166|pull_request:9247", "pull_request:7181|pull_request:7653", "pull_request:7638|pull_request:7893", "pull_request:7652|pull_request:8627", @@ -161950,186 +162127,191 @@ "pull_request:10837|pull_request:11534", "pull_request:10324|pull_request:10325", "pull_request:11475|pull_request:11975", - "pull_request:12519|pull_request:13365" + "pull_request:12519|pull_request:13365", + "pull_request:3797|pull_request:4370", + "pull_request:4262|pull_request:5331", + "pull_request:570|pull_request:826", + "pull_request:5813|pull_request:6237", + "pull_request:6893|pull_request:7410" ], - "prepared_review_unit_hash": "0bd7ebe974b9c49d8b6a778adc703211816bb15664779cd80641f5fc2c31e0e4", + "prepared_review_unit_hash": "2c3ea0fb16be38b251a7b07c28327f2d7a563678fdabc1c2221bb7388d43d016", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 100848, - "estimated_input_tokens": 25212, - "estimated_eval_tokens": 50680 + "serialized_chars": 104543, + "estimated_input_tokens": 26136, + "estimated_eval_tokens": 52528 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: the paired PRs are only superficially similar (same subsystem, same file, or same kind of maintenance change) but address different bugs, features, or test/docs updates. All soft edges should be rejected.", - "confidence": 0.96, + "summary": "This is not a true duplicate cluster: the items span unrelated fixes across docs, CI, Docker, WAN, Hunyuan, SD3, ControlNet, and test-only changes. The soft edges are mostly superficial theme matches and should be rejected.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No single PR cleanly represents a duplicate cluster here; the items span unrelated changes across many areas, so there is no safe canonical PR.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:6997", - "right": "pull_request:7206", - "accept": false, - "reason": "Different work: UNetSpatioTemporalConditionModel LoRA support versus adding training tags across many example scripts." - }, - { - "left": "pull_request:7077", - "right": "pull_request:7124", - "accept": false, - "reason": "Unrelated fixes: attention head reshaping for IPAdapterAttnProcessor versus an SDXL DreamBooth kwarg rename." - }, - { - "left": "pull_request:7103", - "right": "pull_request:9247", - "accept": false, - "reason": "Both touch control/model-loading code, but one fixes ControlNet image validation while the other fixes variant folder detection in pipeline loading." - }, - { - "left": "pull_request:7153", - "right": "pull_request:7662", - "accept": false, - "reason": "Both are CI-related workflow edits, but one fixes documentation path filters and the other removes redundant flax test package installs." - }, - { - "left": "pull_request:7166", - "right": "pull_request:9247", - "accept": false, - "reason": "Different problems: SDXL LoRA loading bug versus model variant identification in pipeline loading utils." - }, { "left": "pull_request:7181", "right": "pull_request:7653", "accept": false, - "reason": "Both are typo/cleanup style changes, but they affect unrelated docs and config logic; not the same underlying bug." + "reason": "Both touch config-related code, but 7181 is docs/example cleanup while 7653 fixes a classmethod bug in configuration_utils; different problems." }, { "left": "pull_request:7638", "right": "pull_request:7893", "accept": false, - "reason": "Both touch project setup/packaging, but one fixes a CONTRIBUTING link while the other upgrades Dockerfiles to Python 3.10." + "reason": "Both are docs/infra maintenance, but one fixes a broken CONTRIBUTING link and the other upgrades Python in Dockerfiles; unrelated changes." }, { "left": "pull_request:7652", "right": "pull_request:8627", "accept": false, - "reason": "Docker buildx setup versus SD3 controlnet fast-test reduction; entirely different concerns." + "reason": "Docker buildx workflow setup is unrelated to the SD3 ControlNet feature work." }, { "left": "pull_request:7913", "right": "pull_request:8467", "accept": false, - "reason": "One adds custom sigmas/timesteps to ControlNet SDXL; the other fixes SEGA pipeline device usage. Different pipelines and code paths." + "reason": "ControlNet SDXL timestep support and SEGA device handling are different code paths and bugs." }, { "left": "pull_request:7919", "right": "pull_request:8722", "accept": false, - "reason": "Offload-behavior fix versus a large embeddings refactor; the latter is a broad structural change, not the same bug." + "reason": "Offload/device_map behavior fixes are unrelated to the embeddings refactor." }, { "left": "pull_request:7938", "right": "pull_request:8386", "accept": false, - "reason": "Docker status reporting workflow versus test-shape-property refactors; unrelated maintenance changes." + "reason": "Docker workflow status reporting and test shape-property refactors are unrelated." }, { "left": "pull_request:8272", "right": "pull_request:8322", "accept": false, - "reason": "Both are typo/errata style edits, but they target different files and broadly different documentation/scheduler text." + "reason": "Both are typo/errata cleanup, but they fix different text across different files and are not the same patch." }, { "left": "pull_request:8322", "right": "pull_request:9100", "accept": false, - "reason": "General errata across docs/examples versus a single typo fix in an SDXL LoRA script; not the same change." + "reason": "Both are typo fixes, but they are separate wording corrections in unrelated docs/examples." }, { "left": "pull_request:8522", "right": "pull_request:9721", "accept": false, - "reason": "One pins transformers in setup.py; the other pins torch versions in Dockerfiles. Similar packaging theme, different dependency and execution contexts." + "reason": "Transformers version pin and torch version pin are separate dependency changes." }, { "left": "pull_request:8591", "right": "pull_request:8702", "accept": false, - "reason": "SD3 training density/weighting refactor versus adding an SD3 conversion script; related ecosystem, but not the same bug or change." + "reason": "SD3 training weighting refactor and the SD3 conversion script are different deliverables with different purposes." }, { "left": "pull_request:8702", "right": "pull_request:9735", "accept": false, - "reason": "Conversion script addition versus updating SD3 controlnet example/training script; different artifacts and purposes." + "reason": "A conversion script and an example update for SD3 ControlNet do not fix the same code-path problem." }, { "left": "pull_request:9408", "right": "pull_request:9620", "accept": false, - "reason": "Both are ControlNet feature work, but one adds Flux ControlNet img2img/inpaint pipelines while the other adds interrupt support across ControlNet pipelines." + "reason": "Flux ControlNet img2img/inpaint additions and interrupt support are related to ControlNet but solve different issues." }, { "left": "pull_request:11639", "right": "pull_request:12611", "accept": false, - "reason": "WanVideoToVideo latent sampling fix versus a large modular WAN/guiders integration PR; same broad domain but not the same concrete bug." + "reason": "Wan video-to-video sampling-mode cleanup and the modular WAN refactor are not the same concrete bug/change." }, { "left": "pull_request:5807", "right": "pull_request:6316", "accept": false, - "reason": "LPW SDXL prompt embedding duplication fix versus moving ControlNetXS into the community folder; unrelated changes." + "reason": "LPW SDXL prompt embedding batching and moving ControlNetXS to community are unrelated." }, { "left": "pull_request:12551", "right": "pull_request:13055", "accept": false, - "reason": "Both involve WAN/CI/testing, but one skips Sana layerwise casting tests in CI while the other fixes WAN 2.2 patch-size handling when a transformer is missing." + "reason": "Skipping Sana layerwise-casting tests in CI is unrelated to WAN 2.2 transformer fallback handling." }, { "left": "pull_request:3770", "right": "pull_request:4498", "accept": false, - "reason": "Control guidance start/stop support versus an SDXL reference-adain community example; different features and code paths." + "reason": "Control guidance start/stop and SDXL reference-attention/adain are different feature areas and code paths." }, { "left": "pull_request:10837", "right": "pull_request:11534", "accept": false, - "reason": "Both touch Hunyuan Video docs, but one adds SkyReels Hunyuan T2V/I2V support and the other adds Framepack documentation and pipeline support." + "reason": "Both are HunyuanVideo-related, but one adds SkyReels T2V/I2V support and the other adds Framepack F1; not the same change." }, { "left": "pull_request:10324", "right": "pull_request:10325", "accept": false, - "reason": "A Kandinsky pipeline API fix and an EMA test fix are completely unrelated despite both being small bugfix PRs." + "reason": "Pipeline offload API fix and EMA test adjustment are unrelated." }, { "left": "pull_request:11475", "right": "pull_request:11975", "accept": false, - "reason": "WAN recompilation fix versus a style-only training-script correction; different intent and code paths." + "reason": "WAN recompilation fix and a Flux training style correction are unrelated." }, { "left": "pull_request:12519", "right": "pull_request:13365", "accept": false, - "reason": "Both modify GitHub workflows, but one fixes toctree sorting while the other changes Claude review fork support; not the same underlying issue." + "reason": "Toctree sorting/CI formatting and Claude review workflow fork support are unrelated CI changes." + }, + { + "left": "pull_request:3797", + "right": "pull_request:4370", + "accept": false, + "reason": "Bad attention naming in model init and a dummy import path fix are not the same underlying bug." + }, + { + "left": "pull_request:4262", + "right": "pull_request:5331", + "accept": false, + "reason": "Both touch DreamBooth LoRA, but they address different issues: multiple training fixes vs replacing deprecated attention LoRA procs." + }, + { + "left": "pull_request:570", + "right": "pull_request:826", + "accept": false, + "reason": "BaseOutput dict initialization and DreamBooth fp16 loss casting are unrelated." + }, + { + "left": "pull_request:5813", + "right": "pull_request:6237", + "accept": false, + "reason": "Loader API docs split and batched seed documentation are separate documentation updates." + }, + { + "left": "pull_request:6893", + "right": "pull_request:7410", + "accept": false, + "reason": "3D UNet attention refactor and FreeInit behavior fix target different subsystems and bugs." } ] }, @@ -162145,7 +162327,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 188, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -162177,36 +162359,32 @@ "pull_request:1910", "pull_request:1926", "pull_request:2345", + "pull_request:2347", + "pull_request:2398", "pull_request:241", + "pull_request:2470", + "pull_request:2565", + "pull_request:2655", + "pull_request:2919", "pull_request:3275", "pull_request:3297", - "pull_request:3797", + "pull_request:3657", + "pull_request:3928", "pull_request:402", - "pull_request:4262", - "pull_request:4370", + "pull_request:4188", + "pull_request:4214", + "pull_request:4223", "pull_request:453", "pull_request:4819", "pull_request:4944", "pull_request:5033", "pull_request:51", - "pull_request:5331", "pull_request:538", - "pull_request:570", "pull_request:5713", - "pull_request:5813", - "pull_request:6237", "pull_request:630", - "pull_request:6893", - "pull_request:7410", - "pull_request:826", "pull_request:9269" ], "soft_pairs": [ - "pull_request:3797|pull_request:4370", - "pull_request:4262|pull_request:5331", - "pull_request:570|pull_request:826", - "pull_request:5813|pull_request:6237", - "pull_request:6893|pull_request:7410", "pull_request:4819|pull_request:5033", "pull_request:4944|pull_request:5713", "pull_request:10132|pull_request:10494", @@ -162225,190 +162403,200 @@ "pull_request:1583|pull_request:538", "pull_request:1910|pull_request:402", "pull_request:1926|pull_request:241", - "pull_request:2345|pull_request:3275" + "pull_request:2345|pull_request:3275", + "pull_request:2347|pull_request:2398", + "pull_request:2470|pull_request:4223", + "pull_request:2565|pull_request:2655", + "pull_request:2565|pull_request:402", + "pull_request:2919|pull_request:4188", + "pull_request:3657|pull_request:3928", + "pull_request:3928|pull_request:4214" ], - "prepared_review_unit_hash": "735d78f4c16f6a4362db43eb853fa56a8e42e99cae65b0c043b074e938bed7e9", + "prepared_review_unit_hash": "9c30df62caa737b77a1493ae3778d3d7d6cc479e1186d297fc9a18351e1dc402", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 98189, - "estimated_input_tokens": 24548, - "estimated_eval_tokens": 49352 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 102064, + "estimated_input_tokens": 25516, + "estimated_eval_tokens": 51288 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated PRs with a few broad-theme overlaps. The only strong duplicate-like pair is the IP-Adapter work (4944/5713); the rest overlap only at subsystem or filename level and should stay separate.", - "confidence": 0.95, + "summary": "All soft-edge candidates are false positives: they may touch related areas, but each pair targets a different bug, feature, or documentation/test update and should not be deduplicated.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 5713 is the best canonical PR candidate: it is the merged, comprehensive IP-Adapter implementation and subsumes the earlier WIP support in 4944.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 5713 is the best overall representative of the only true duplicate-like theme in this set because it is the final, merged, feature-complete IP-Adapter PR.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:3797", - "right": "pull_request:4370", - "accept": false, - "reason": "Both mention naming/refactor changes in model code, but they fix different issues: bad attention-head naming vs a dummy import path typo." - }, - { - "left": "pull_request:4262", - "right": "pull_request:5331", - "accept": false, - "reason": "Both touch SDXL DreamBooth LoRA, but they address different changes: license/prior-preservation and training logic vs replacing deprecated attention processors." - }, - { - "left": "pull_request:570", - "right": "pull_request:826", - "accept": false, - "reason": "Unrelated fixes: BaseOutput dict initialization vs DreamBooth fp16 loss casting." - }, - { - "left": "pull_request:5813", - "right": "pull_request:6237", - "accept": false, - "reason": "Both are docs PRs, but one splits loader APIs and the other adds batched-seed guidance; different content and goals." - }, - { - "left": "pull_request:6893", - "right": "pull_request:7410", - "accept": false, - "reason": "Different code paths and problems: 3D UNet attention setup refactor vs FreeInit progress-bar behavior." - }, { "left": "pull_request:4819", "right": "pull_request:5033", "accept": false, - "reason": "Both are inpainting-related, but they fix different latent-handling bugs; they are not the same concrete change." + "reason": "Both are inpainting-related, but one adds support for 4-channel image inputs while the other fixes batch replication/mask handling. Different code-path issues." }, { "left": "pull_request:4944", "right": "pull_request:5713", - "accept": true, - "reason": "Both implement IP-Adapter support in the same loader/attention/pipeline area; 5713 is the broader merged follow-up to the earlier WIP 4944 and could plausibly have been one PR." + "accept": false, + "reason": "Same IP-Adapter theme, but 4944 is core loader/model support and 5713 is broader feature/docs integration across pipelines. Not the same concrete change." }, { "left": "pull_request:10132", "right": "pull_request:10494", "accept": false, - "reason": "Both edit CI tokenizers installs, but for different test jobs and different purposes; not the same bug or change." + "reason": "Both edit the CI workflow, but they fix different tokenizers install commands for different test contexts." }, { "left": "pull_request:10200", "right": "pull_request:9269", "accept": false, - "reason": "Completely different features: ControlNet-Union refactor vs CogVideoX 3D RoPE refactor." + "reason": "Unrelated refactors in different subsystems: ControlNet Union vs CogVideoX 3D RoPE." }, { "left": "pull_request:1038", "right": "pull_request:51", "accept": false, - "reason": "Unrelated: Flax dtype fix vs setup.py install_requires typo fix." + "reason": "Flax dtype bugfix vs setup.py dependency typo; completely different changes." }, { "left": "pull_request:10597", "right": "pull_request:10633", "accept": false, - "reason": "Both are docs updates, but one changes an SD3 IP-Adapter model ID and the other adds a Flux IP-Adapter example." + "reason": "Both are docs, but for different pipelines and different features." }, { "left": "pull_request:1147", "right": "pull_request:630", "accept": false, - "reason": "Different areas entirely: Flax example loading vs Flax from_pretrained mismatch handling." + "reason": "One fixes Flax example loading from subfolders; the other cleans up Flax from_pretrained mismatch handling. Different problems." }, { "left": "pull_request:11504", "right": "pull_request:11728", "accept": false, - "reason": "CI/dependency table update vs Docker doc-builder quant-library install; not the same change." + "reason": "Dependency/workflow update vs Dockerfile quant-lib additions. Same broad tooling area, but not the same bug or change." }, { "left": "pull_request:11613", "right": "pull_request:12425", "accept": false, - "reason": "One reverts torch.compile CI/tests, the other fixes torch.distributed imports; unrelated." + "reason": "A revert of torch.compile CI/tests is unrelated to guarding distributed imports." }, { "left": "pull_request:12363", "right": "pull_request:13029", "accept": false, - "reason": "Different docs projects: QwenImageEdit Plus documentation/test additions vs modular quickstart rewrite." + "reason": "QwenImageEdit Plus docs/tests vs Modular Diffusers quickstart rewrite; unrelated content." }, { "left": "pull_request:1269", "right": "pull_request:3297", "accept": false, - "reason": "Stable Diffusion refactor vs retrieval-augmented diffusion model example; unrelated." + "reason": "Stable Diffusion refactor follow-up vs retrieval-augmented diffusion model feature; different artifacts entirely." }, { "left": "pull_request:12973", "right": "pull_request:13365", "accept": false, - "reason": "GLM-Image feature support vs Claude review workflow changes; no code-path overlap." + "reason": "GLM-Image model support vs Claude review CI fork support. No overlap in underlying change." }, { "left": "pull_request:13150", "right": "pull_request:13332", "accept": false, - "reason": "Test helper recompile-limit behavior vs dependency-testing workflow tightening; different concerns." + "reason": "Test helper recompile-limit behavior vs dependency-testing workflow tightening; separate concerns." }, { "left": "pull_request:1508", "right": "pull_request:453", "accept": false, - "reason": "Dreambooth class-image docs clarification vs installation typo fix; unrelated." + "reason": "DreamBooth docs clarification vs installation typo fix. Different documentation issues." }, { "left": "pull_request:1582", "right": "pull_request:538", "accept": false, - "reason": "RePaint scheduler bug fix vs hub version bump; no shared bug/change." + "reason": "RePaint scheduler bugfix vs hub version bump. Unrelated." }, { "left": "pull_request:1583", "right": "pull_request:538", "accept": false, - "reason": "Legacy inpaint pipeline bug fix vs hub version bump; unrelated." + "reason": "Legacy inpaint pipeline fix vs hub version bump. Unrelated changes." }, { "left": "pull_request:1910", "right": "pull_request:402", "accept": false, - "reason": "Both are docs reorganizations, but one localizes docs while the other finishes the intro section; different scopes." + "reason": "Korean docs initialization and toctree changes vs intro-section docs cleanup. Same docs area, but not the same change." }, { "left": "pull_request:1926", "right": "pull_request:241", "accept": false, - "reason": "A large inpainting support branch vs a standalone inpainting example script; same broad topic, not the same change." + "reason": "One is a large inpainting support/support-support PR, the other is an example script addition. Not mergeable as one duplicate change." }, { "left": "pull_request:2345", "right": "pull_request:3275", "accept": false, - "reason": "xFormers warning message vs upsample_size argument plumbing; unrelated fixes." + "reason": "xFormers warning in training scripts vs adding upsample_size to attention blocks. Different code paths." + }, + { + "left": "pull_request:2347", + "right": "pull_request:2398", + "accept": false, + "reason": "Scheduler noising-schedule changes vs Pix2Pix Zero docs updates; unrelated." + }, + { + "left": "pull_request:2470", + "right": "pull_request:4223", + "accept": false, + "reason": "MPS test skips vs ControlNet SDXL training fixes. Different problems." + }, + { + "left": "pull_request:2565", + "right": "pull_request:2655", + "accept": false, + "reason": "UNet1D docs correction vs LoRA save/load logic change. Different scope and behavior." + }, + { + "left": "pull_request:2919", + "right": "pull_request:4188", + "accept": false, + "reason": "Test speed-up changes vs SDXL ControlNet embedding-cache follow-up fixes. Not the same issue." + }, + { + "left": "pull_request:3657", + "right": "pull_request:3928", + "accept": false, + "reason": "ControlNet checkpoint parsing tweaks vs out-of-list runtime fix in ControlNet pipelines/schedulers. Different underlying bugs." + }, + { + "left": "pull_request:3928", + "right": "pull_request:4214", + "accept": false, + "reason": "ControlNet runtime fix vs bf16 error resolution in Custom Diffusion. Unrelated." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet: most pairs are indeed only loose subsystem/doc/workflow overlaps, and the IP-Adapter pair (4944/5713) is the only clearly related theme. The soft-edge rejections are conservative and supported by the diffs. The single accepted pair is plausible as a merged follow-up rather than an overclaimed exact duplicate." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -162417,7 +162605,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 189, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -162425,27 +162613,22 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2347", - "pull_request:2398", - "pull_request:2470", - "pull_request:2565", - "pull_request:2655", - "pull_request:2919", - "pull_request:3657", + "pull_request:1006", + "pull_request:12086", + "pull_request:12117", + "pull_request:12418", + "pull_request:12519", + "pull_request:2476", + "pull_request:3652", "pull_request:3681", - "pull_request:3691", - "pull_request:3928", - "pull_request:402", "pull_request:4070", "pull_request:4099", - "pull_request:4188", - "pull_request:4214", - "pull_request:4223", "pull_request:4245", "pull_request:4595", "pull_request:4887", "pull_request:4987", "pull_request:51", + "pull_request:5301", "pull_request:5385", "pull_request:5615", "pull_request:5788", @@ -162454,6 +162637,7 @@ "pull_request:6119", "pull_request:6208", "pull_request:6217", + "pull_request:6225", "pull_request:6229", "pull_request:6265", "pull_request:6759", @@ -162471,18 +162655,14 @@ "pull_request:8389", "pull_request:8452", "pull_request:8510", + "pull_request:8564", + "pull_request:8702", + "pull_request:8973", "pull_request:9323", - "pull_request:9569" + "pull_request:9569", + "pull_request:9896" ], "soft_pairs": [ - "pull_request:2347|pull_request:2398", - "pull_request:2470|pull_request:4223", - "pull_request:2565|pull_request:2655", - "pull_request:2565|pull_request:402", - "pull_request:2919|pull_request:4188", - "pull_request:3657|pull_request:3928", - "pull_request:3681|pull_request:3691", - "pull_request:3928|pull_request:4214", "pull_request:4070|pull_request:4245", "pull_request:4099|pull_request:6078", "pull_request:4595|pull_request:6114", @@ -162501,198 +162681,199 @@ "pull_request:7158|pull_request:8389", "pull_request:7353|pull_request:9323", "pull_request:8452|pull_request:9569", - "pull_request:8510|pull_request:9569" + "pull_request:8510|pull_request:9569", + "pull_request:8564|pull_request:8702", + "pull_request:8973|pull_request:9896", + "pull_request:12086|pull_request:12117", + "pull_request:1006|pull_request:2476", + "pull_request:12418|pull_request:12519", + "pull_request:3652|pull_request:3681", + "pull_request:5301|pull_request:6225" ], - "prepared_review_unit_hash": "edece5c1c49c51deebb2550c40164905618e2ad17e789a422ffcb96e3bd493fc", + "prepared_review_unit_hash": "7e9e4c8be89a19c42195b3c5efff1aa687743de3304105ee9028106f0b98cd7a", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 27, - "serialized_chars": 93978, - "estimated_input_tokens": 23495, - "estimated_eval_tokens": 47246 + "soft_pair_count": 26, + "serialized_chars": 94167, + "estimated_input_tokens": 23542, + "estimated_eval_tokens": 47340 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose bucket of unrelated PRs that happen to share generic maintenance/fix wording. The soft pairs do not look like the same underlying bug or change, so none should be merged as duplicates.", + "summary": "This set is a heterogeneous mix of unrelated PRs across docs, pipelines, training scripts, benchmarks, and utilities. The soft pairs mostly share broad subsystems or filenames, but they do not fix the same concrete bug/change, so none should be merged as duplicates.", "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:2347", - "right": "pull_request:2398", - "accept": false, - "reason": "Different changes: scheduler noising schedule vs Pix2Pix Zero docs update." - }, - { - "left": "pull_request:2470", - "right": "pull_request:4223", - "accept": false, - "reason": "Different test-only MPS skip fix vs ControlNet SDXL training fixes." - }, - { - "left": "pull_request:2565", - "right": "pull_request:2655", - "accept": false, - "reason": "Unet1d docs correction is unrelated to LoRA loading/saving logic." - }, - { - "left": "pull_request:2565", - "right": "pull_request:402", - "accept": false, - "reason": "Docs-only Unet1d correction does not match intro docs restructuring." - }, - { - "left": "pull_request:2919", - "right": "pull_request:4188", - "accept": false, - "reason": "Test speed-up changes are unrelated to SDXL ControlNet dataset fingerprinting." - }, - { - "left": "pull_request:3657", - "right": "pull_request:3928", - "accept": false, - "reason": "Checkpoint parsing tweaks are unrelated to ControlNet timesteps bug fix." - }, - { - "left": "pull_request:3681", - "right": "pull_request:3691", - "accept": false, - "reason": "Training-script checkpoint handling and LoRA processor class selection are different fixes." - }, - { - "left": "pull_request:3928", - "right": "pull_request:4214", - "accept": false, - "reason": "ControlNet timestep bounds bug and bf16 custom diffusion error are distinct code paths." - }, { "left": "pull_request:4070", "right": "pull_request:4245", "accept": false, - "reason": "SDXL docs link cleanup is unrelated to adding safetensors examples." + "reason": "Both are docs-only PRs, but one fixes SDXL doc links while the other adds safetensors examples across many docs; not the same change." }, { "left": "pull_request:4099", "right": "pull_request:6078", "accept": false, - "reason": "SDXL prompt-embedding bug fix is unrelated to Euler sigma/import cleanup." + "reason": "Both touch model/pipeline internals, but one fixes SDXL prompt embedding handling and the other changes Euler scheduler sigma imports/ordering; different bugs." }, { "left": "pull_request:4595", "right": "pull_request:6114", "accept": false, - "reason": "WIP converter work does not match LPW XL img2img/inpaint support." + "reason": "One is an unfinished XL converter scaffold; the other adds img2img/inpaint support to a community LPW script. No shared concrete fix." }, { "left": "pull_request:4887", "right": "pull_request:7115", "accept": false, - "reason": "Kandinsky pipeline API removal is unrelated to Dockerfile dependency cleanup." + "reason": "Kandinsky prior emb2emb latent removal and Dockerfile cleanup are unrelated code paths and goals." }, { "left": "pull_request:4987", "right": "pull_request:6265", "accept": false, - "reason": "Import list cleanup is unrelated to moving GLIGEN into its own module." + "reason": "Both are import-structure refactors, but one removes a duplicate GLIGEN export and the other moves GLIGEN into its own package. Different scope." }, { "left": "pull_request:4987", "right": "pull_request:6886", "accept": false, - "reason": "Import cleanup is unrelated to dtype/to() deprecation fixes." + "reason": "Duplicate GLIGEN import cleanup versus deprecation cleanup for `to(dtype=...)`; unrelated changes." }, { "left": "pull_request:51", "right": "pull_request:816", "accept": false, - "reason": "Packaging typo fix is unrelated to deepspeed fine-tuning loss casting." + "reason": "A packaging pin fix and a training-loss dtype cast are not the same underlying issue." }, { "left": "pull_request:5385", "right": "pull_request:7202", "accept": false, - "reason": "CI copy-check changes are unrelated to removing huggingface-cli from requirements." + "reason": "Both are CI/example maintenance, but one updates fix-copies logic for method signatures while the other removes a dependency from example requirements." }, { "left": "pull_request:5615", "right": "pull_request:7790", "accept": false, - "reason": "Reverting LoRA loading in T2I adapter is unrelated to SDXL masked img2img adaptation." + "reason": "Both concern SDXL-adjacent community code, but one reverts `load_lora_weights` behavior and the other adapts a masked img2img pipeline for SDXL." }, { "left": "pull_request:5788", "right": "pull_request:7158", "accept": false, - "reason": "LCM timestep validation is unrelated to benchmark workflow changes." + "reason": "LCM scheduler error handling and benchmark workflow changes are unrelated; shared benchmark naming is incidental." }, { "left": "pull_request:6119", "right": "pull_request:8378", "accept": false, - "reason": "LoRA fp16 training fix is unrelated to asymmetric autoencoder forward arguments." + "reason": "One fixes LoRA alpha parsing in training scripts; the other fixes AsymmetricAutoencoderKL forward decode arguments. Different code paths and bugs." }, { "left": "pull_request:6208", "right": "pull_request:8510", "accept": false, - "reason": "Stable Diffusion decode dtype mismatch is unrelated to DreamBooth LR scheduler logic." + "reason": "One fixes Stable Diffusion VAE dtype mismatch at decode time; the other changes LR scheduler math in DreamBooth LoRA training." }, { "left": "pull_request:6217", "right": "pull_request:8303", "accept": false, - "reason": "Stable Video Diffusion torch-compile fix is unrelated to a philosophy doc typo." + "reason": "Torch-compile VAE signature handling and a documentation typo fix are unrelated." }, { "left": "pull_request:6229", "right": "pull_request:7778", "accept": false, - "reason": "CI dependency install in fast tests is unrelated to adding a debugging workflow." + "reason": "CI dependency installation and a debugging workflow are both workflow-related, but they address different purposes and code paths." }, { "left": "pull_request:6759", "right": "pull_request:6952", "accept": false, - "reason": "InstantID prompt-image embedding repeat fix is unrelated to textual inversion model-card updates." + "reason": "Both touch example/community code, but one fixes image embedding repetition in an InstantID example and the other rewrites textual inversion model card handling." }, { "left": "pull_request:7158", "right": "pull_request:8389", "accept": false, - "reason": "Benchmark test fix and benchmarking workflow overhaul are related to benchmarks but not the same concrete change." + "reason": "Both involve benchmarking, but one fixes an LCM benchmark test and the other rewrites the benchmarking workflow/container/reporting. Not the same bug." }, { "left": "pull_request:7353", "right": "pull_request:9323", "accept": false, - "reason": "ControlNet SDXL img2img mixin addition is unrelated to single-file default model updates." + "reason": "One adds FromSingleFileMixin to SDXL ControlNet img2img; the other changes single-file default checkpoint repos. Different functionality." }, { "left": "pull_request:8452", "right": "pull_request:9569", "accept": false, - "reason": "Single-file loading strictness fix is unrelated to CogVideoX autoencoder decode batching." + "reason": "Single-file loading strictness and CogVideoX autoencoder decode batching are unrelated." }, { "left": "pull_request:8510", "right": "pull_request:9569", "accept": false, - "reason": "DreamBooth LR scheduler fix is unrelated to CogVideoX autoencoder decode batching." + "reason": "DreamBooth LR scheduler fixes and CogVideoX decode fixes are different subsystems with no shared concrete bug." + }, + { + "left": "pull_request:8564", + "right": "pull_request:8702", + "accept": false, + "reason": "Both are SD3-related, but one adds T5 token-limit documentation/training support while the other adds an SD3 conversion script. Different changes." + }, + { + "left": "pull_request:8973", + "right": "pull_request:9896", + "accept": false, + "reason": "A test-size reduction and a Flux revert are unrelated; neither fixes the same code-path problem." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12117", + "accept": false, + "reason": "Both add Flux/QwenImage modular pipeline support, but they target different features and filesets: Flux I2I modularization versus QwenImage img2img/inpaint pipelines." + }, + { + "left": "pull_request:1006", + "right": "pull_request:2476", + "accept": false, + "reason": "Both mention interpolation/image-size issues, but one fixes large-batch `F.interpolate()` behavior in resnet, while the other adds an `attn_res` parameter for Attend-and-Excite. Not the same bug." + }, + { + "left": "pull_request:12418", + "right": "pull_request:12519", + "accept": false, + "reason": "One xfails a failing test in CI; the other fixes a docs toctree sorting issue. Shared CI context only." + }, + { + "left": "pull_request:3652", + "right": "pull_request:3681", + "accept": false, + "reason": "Both involve checkpoint naming in training/example code, but one refactors checkpoint directory layout in tests/examples, while the other manually enforces checkpoint limits instead of relying on Accelerate." + }, + { + "left": "pull_request:5301", + "right": "pull_request:6225", + "accept": false, + "reason": "One fixes SDXL pipelines without text encoders; the other fixes LoRA alpha parsing across training scripts. Related area, but not the same underlying bug." } ] }, @@ -162708,7 +162889,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 190, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -162716,7 +162897,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:1006", "pull_request:10322", "pull_request:10731", "pull_request:1089", @@ -162725,12 +162905,8 @@ "pull_request:11522", "pull_request:11595", "pull_request:11999", - "pull_request:12086", - "pull_request:12117", "pull_request:12188", - "pull_request:12418", "pull_request:12424", - "pull_request:12519", "pull_request:12581", "pull_request:12805", "pull_request:12960", @@ -162743,35 +162919,33 @@ "pull_request:1977", "pull_request:2279", "pull_request:2314", - "pull_request:2476", - "pull_request:3652", - "pull_request:3681", "pull_request:3723", "pull_request:5038", - "pull_request:5301", + "pull_request:5111", "pull_request:5697", "pull_request:5842", "pull_request:5974", - "pull_request:6225", "pull_request:6241", + "pull_request:6311", + "pull_request:6332", + "pull_request:6365", + "pull_request:6673", "pull_request:6787", + "pull_request:7206", "pull_request:7507", + "pull_request:788", "pull_request:7924", "pull_request:7934", - "pull_request:8564", - "pull_request:8702", - "pull_request:8973", + "pull_request:8467", + "pull_request:8618", + "pull_request:8627", + "pull_request:8722", + "pull_request:9160", "pull_request:9442", - "pull_request:9896" + "pull_request:9736", + "pull_request:978" ], "soft_pairs": [ - "pull_request:8564|pull_request:8702", - "pull_request:8973|pull_request:9896", - "pull_request:12086|pull_request:12117", - "pull_request:1006|pull_request:2476", - "pull_request:12418|pull_request:12519", - "pull_request:3652|pull_request:3681", - "pull_request:5301|pull_request:6225", "pull_request:5842|pull_request:5974", "pull_request:6241|pull_request:7924", "pull_request:6787|pull_request:7507", @@ -162789,186 +162963,200 @@ "pull_request:1968|pull_request:2314", "pull_request:1977|pull_request:2279", "pull_request:2279|pull_request:3723", - "pull_request:5038|pull_request:5697" + "pull_request:5038|pull_request:5697", + "pull_request:5111|pull_request:6365", + "pull_request:6311|pull_request:8627", + "pull_request:6332|pull_request:8467", + "pull_request:6673|pull_request:6787", + "pull_request:7206|pull_request:9160", + "pull_request:788|pull_request:978", + "pull_request:8618|pull_request:9736", + "pull_request:8722|pull_request:9736" ], - "prepared_review_unit_hash": "bb4b5f5b1542cb8bfd1df6cf373660af3482e20171d2e8e0f83efea5f8657851", + "prepared_review_unit_hash": "a584c62610311dfce2bf3579bd0eee266069c4986d715d3099ab7fce85e37c97", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 25, - "serialized_chars": 100168, - "estimated_input_tokens": 25042, - "estimated_eval_tokens": 50340 + "soft_pair_count": 26, + "serialized_chars": 101768, + "estimated_input_tokens": 25442, + "estimated_eval_tokens": 51140 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated docs, CI, example, pipeline, scheduler, and LoRA changes. None of the soft pairs look like the same underlying bug or mergeable change, so they should all be rejected.", - "confidence": 0.99, + "summary": "This cluster is mostly false-positive similarity matches: many pairs share a broad subsystem or even a filename, but they target different bugs, features, or cleanup changes. I don\u2019t see a single coherent duplicate PR/issue to canonicalize around.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR stands out; the items span unrelated features and fixes rather than one duplicate change set.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is a good global representative because the cluster is not a coherent duplicate group.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:8564", - "right": "pull_request:8702", - "accept": false, - "reason": "Different SD3-related changes: token-limit training docs/code vs a new conversion script." - }, - { - "left": "pull_request:8973", - "right": "pull_request:9896", - "accept": false, - "reason": "Unrelated areas: lumina test-size reduction vs Flux device/typecasting revert." - }, - { - "left": "pull_request:12086", - "right": "pull_request:12117", - "accept": false, - "reason": "Different features in different subsystems: Flux modular I2I support vs QwenImage pipelines." - }, - { - "left": "pull_request:1006", - "right": "pull_request:2476", - "accept": false, - "reason": "Both mention image sizing, but one fixes large-batch interpolate behavior and the other adds an Attend-and-Excite size parameter." - }, - { - "left": "pull_request:12418", - "right": "pull_request:12519", - "accept": false, - "reason": "CI test xfail and docs toctree sorting are unrelated changes." - }, - { - "left": "pull_request:3652", - "right": "pull_request:3681", - "accept": false, - "reason": "Both touch checkpointing, but one refactors checkpoint naming while the other changes checkpoint-limit handling." - }, - { - "left": "pull_request:5301", - "right": "pull_request:6225", - "accept": false, - "reason": "SDXL no-text-encoder support and LoRA alpha parsing are different concrete problems." - }, { "left": "pull_request:5842", "right": "pull_request:5974", "accept": false, - "reason": "Same file, but different bugs: PixArt resolution binning vs timestep retrieval for LCM." + "reason": "Both touch PixArtAlpha, but one fixes 512px resolution binning while the other adapts the pipeline for pixart-lcm; different changes, not one bug." }, { "left": "pull_request:6241", "right": "pull_request:7924", "accept": false, - "reason": "Runtime dtype mismatch and IP-Adapter added_cond_kwargs are unrelated code paths." + "reason": "Unrelated failures: generic dtype mismatch runtime error vs IP-Adapter `added_cond_kwargs` handling in ControlNet inpaint." }, { "left": "pull_request:6787", "right": "pull_request:7507", "accept": false, - "reason": "Marigold input normalization fix vs adding an InstantID SDXL img2img pipeline." + "reason": "Different pipelines and problems: Marigold input normalization vs adding an InstantID SDXL img2img pipeline." }, { "left": "pull_request:7934", "right": "pull_request:9442", "accept": false, - "reason": "Single-file dict config loading and CI behavior for deprecated attention blocks are unrelated." + "reason": "Single-file config loading with dicts is unrelated to CI changes for deprecated attention block tests." }, { "left": "pull_request:10322", "right": "pull_request:11142", "accept": false, - "reason": "Both are Sana converter edits, but they address different model variants and script issues, not one shared bug." + "reason": "Both modify the Sana converter, but they add support for different model families/variants and different scheduler setup changes." }, { "left": "pull_request:10731", "right": "pull_request:10975", "accept": false, - "reason": "Video pipeline output typing cleanup vs Wan pipeline bug fixes are unrelated." + "reason": "Type-hint/output cleanup across video pipelines is not the same as fixing Wan I2V pipeline bugs." }, { "left": "pull_request:1089", "right": "pull_request:1386", "accept": false, - "reason": "Docs on scheduler usage and the scheduler v-prediction implementation are not the same change." + "reason": "Docs for swapping schedulers are unrelated to implementing v-prediction in schedulers." }, { "left": "pull_request:11142", "right": "pull_request:12188", "accept": false, - "reason": "Sana Sprint converter fixes and Qwen-Image-Edit width/height logic are unrelated." + "reason": "Sana Sprint converter updates do not match the Qwen-Image-Edit dimension bug fix." }, { "left": "pull_request:11522", "right": "pull_request:12424", "accept": false, - "reason": "AudioLDM2 transformers compatibility and Docker/CI cleanup are unrelated." + "reason": "AudioLDM2 transformers-version compatibility is unrelated to Dockerfile/workflow definition cleanup." }, { "left": "pull_request:11595", "right": "pull_request:11999", "accept": false, - "reason": "Both involve LoRA, but one fixes Flux load_lora_weights expansion and the other updates PEFT config injection; different code paths." + "reason": "Both are LoRA-related, but one fixes Flux weight loading logic and the other adapts PEFT injection APIs across loaders and tests." }, { "left": "pull_request:12424", "right": "pull_request:12960", "accept": false, - "reason": "Docker/workflow fixes and the FSDP detection change are unrelated." + "reason": "Docker/workflow cleanup is unrelated to the `is_fsdp` detection fix in Flux2 training scripts." }, { "left": "pull_request:12581", "right": "pull_request:13056", "accept": false, - "reason": "Support for more Qwen LoRAs and Z-Image LoRA training are different features." + "reason": "Both mention LoRA support, but they address different models and workflows: Qwen LoRA conversion vs Z-Image DreamBooth training." }, { "left": "pull_request:12805", "right": "pull_request:12998", "accept": false, - "reason": "PyPI publish workflow updates and style-only doc/pipeline cleanup are unrelated." + "reason": "PyPI publish workflow changes are unrelated to docstring/style cleanup in several pipelines." }, { "left": "pull_request:1299", "right": "pull_request:130", "accept": false, - "reason": "AltDiffusion feature addition vs a README note about accelerate config are unrelated." + "reason": "AltDiffusion feature addition and an examples README note about accelerate config are unrelated." }, { "left": "pull_request:1968", "right": "pull_request:2314", "accept": false, - "reason": "CPU offload behavior fix and fast-test expectation tweaks do not address the same bug." + "reason": "CPU offload behavior fixes are unrelated to adjusting fast-test expected values." }, { "left": "pull_request:1977", "right": "pull_request:2279", "accept": false, - "reason": "Example/script behavior changes and lint tooling migration are unrelated." + "reason": "Training script seed/launch behavior and tooling migration from flake8 to ruff are different changes." }, { "left": "pull_request:2279", "right": "pull_request:3723", "accept": false, - "reason": "Style/ruff migration vs reference pipeline enhancement are unrelated." + "reason": "Repository lint tooling changes are unrelated to the reference pipeline example update." }, { "left": "pull_request:5038", "right": "pull_request:5697", "accept": false, - "reason": "Both mention LoRA, but one makes optional args explicit and the other fixes PEFT/xformers attention behavior; not the same code-path problem." + "reason": "Both involve LoRA scale handling, but they patch different code paths: transformer projection args vs attention processor/xformers backend behavior." + }, + { + "left": "pull_request:5111", + "right": "pull_request:6365", + "accept": false, + "reason": "Karras-sigma support and MPS dtype handling in KDPM2AncestralDiscreteScheduler are separate issues." + }, + { + "left": "pull_request:6311", + "right": "pull_request:8627", + "accept": false, + "reason": "These are different documentation additions for different feature areas; shared toctree edits are incidental." + }, + { + "left": "pull_request:6332", + "right": "pull_request:8467", + "accept": false, + "reason": "Rerender-A-Video community pipeline and SEGA pipeline fixes are unrelated community examples." + }, + { + "left": "pull_request:6673", + "right": "pull_request:6787", + "accept": false, + "reason": "InstantID pipeline addition and Marigold input-range correction are unrelated." + }, + { + "left": "pull_request:7206", + "right": "pull_request:9160", + "accept": false, + "reason": "Training script tags/metadata and Flux DreamBooth latent caching are not the same change." + }, + { + "left": "pull_request:788", + "right": "pull_request:978", + "accept": false, + "reason": "PIL resampling warning fixes are unrelated to small typo/documentation cleanup." + }, + { + "left": "pull_request:8618", + "right": "pull_request:9736", + "accept": false, + "reason": "SD3 tiny autoencoder shift_factor and Allegro T2V are different features with no shared bug." + }, + { + "left": "pull_request:8722", + "right": "pull_request:9736", + "accept": false, + "reason": "Large embeddings refactor and Allegro model addition only overlap on touched file names, not on the underlying change." } ] }, @@ -162984,7 +163172,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 191, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -163016,39 +163204,32 @@ "pull_request:1913", "pull_request:1942", "pull_request:2001", + "pull_request:2655", + "pull_request:3034", + "pull_request:3128", + "pull_request:3294", + "pull_request:3446", + "pull_request:3590", + "pull_request:3673", + "pull_request:3681", "pull_request:4062", - "pull_request:5111", + "pull_request:4262", + "pull_request:4690", + "pull_request:4706", + "pull_request:4744", + "pull_request:4793", + "pull_request:5393", "pull_request:5438", - "pull_request:6311", - "pull_request:6332", - "pull_request:6365", - "pull_request:6673", + "pull_request:5807", "pull_request:676", - "pull_request:6787", - "pull_request:7206", - "pull_request:788", + "pull_request:6994", "pull_request:823", - "pull_request:8467", - "pull_request:8618", - "pull_request:8627", - "pull_request:8722", - "pull_request:9160", "pull_request:9175", "pull_request:9696", - "pull_request:9736", - "pull_request:978", "pull_request:9807", "pull_request:9816" ], "soft_pairs": [ - "pull_request:5111|pull_request:6365", - "pull_request:6311|pull_request:8627", - "pull_request:6332|pull_request:8467", - "pull_request:6673|pull_request:6787", - "pull_request:7206|pull_request:9160", - "pull_request:788|pull_request:978", - "pull_request:8618|pull_request:9736", - "pull_request:8722|pull_request:9736", "pull_request:9175|pull_request:9816", "pull_request:12416|pull_request:13126", "pull_request:12519|pull_request:13321", @@ -163065,186 +163246,194 @@ "pull_request:1582|pull_request:676", "pull_request:1583|pull_request:676", "pull_request:1913|pull_request:823", - "pull_request:1942|pull_request:2001" + "pull_request:1942|pull_request:2001", + "pull_request:2655|pull_request:3294", + "pull_request:3034|pull_request:3446", + "pull_request:3128|pull_request:3673", + "pull_request:3590|pull_request:5807", + "pull_request:3681|pull_request:4690", + "pull_request:4262|pull_request:4706", + "pull_request:4744|pull_request:5393", + "pull_request:4793|pull_request:6994" ], - "prepared_review_unit_hash": "392ccab25ccb1537513a3a0d31d40d578dbb25a2ec1afd8ebd4a8dab2899421a", + "prepared_review_unit_hash": "58dbe707b5866d8fe3958db77bd0474d134995b28b287975a224293a3481027d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 97245, - "estimated_input_tokens": 24312, - "estimated_eval_tokens": 48880 + "serialized_chars": 96357, + "estimated_input_tokens": 24090, + "estimated_eval_tokens": 48436 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set; it groups many unrelated pull requests that only share broad file areas or superficial syntax changes. None of the soft edges look like the same underlying bug or change.", - "confidence": 0.95, + "summary": "This cluster is a loose collection of unrelated PRs across docs, CI, loaders, models, and pipelines. The soft-edge pairs mostly share only a broad subsystem or keyword (e.g. Flux, LoRA, LTX) but not the same concrete bug or change.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR serves as a coherent canonical representative because the cluster is heterogeneous; the items span unrelated docs, CI, model, scheduler, and example changes.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is a good global canonical choice for duplicate triage here. The set lacks a shared concrete code-path problem, so picking any one PR would be misleading.", + "best_pr_reason": "PR 9816 is the strongest standalone representative because it is a large, self-contained feature addition with clear scope and substantial review activity; the rest are mostly small, unrelated fixes or infra/docs changes.", "soft_edge_verdicts": [ - { - "left": "pull_request:5111", - "right": "pull_request:6365", - "accept": false, - "reason": "Same scheduler file, but different fixes: one adds Karras sigma support, the other adjusts MPS timestep casting." - }, - { - "left": "pull_request:6311", - "right": "pull_request:8627", - "accept": false, - "reason": "Both touch toctree/docs, but one adds Fast Diffusion docs while the other adds SD3 ControlNet docs; unrelated content." - }, - { - "left": "pull_request:6332", - "right": "pull_request:8467", - "accept": false, - "reason": "Different community pipeline changes with no shared bug or code path." - }, - { - "left": "pull_request:6673", - "right": "pull_request:6787", - "accept": false, - "reason": "Both are community pipeline work, but one adds InstantID while the other fixes Marigold depth image scaling; not duplicates." - }, - { - "left": "pull_request:7206", - "right": "pull_request:9160", - "accept": false, - "reason": "Unrelated training-script documentation/tag updates for different purposes." - }, - { - "left": "pull_request:788", - "right": "pull_request:978", - "accept": false, - "reason": "One fixes PIL resampling warnings in img2img/inpaint; the other is a generic typo/doc cleanup. Different changes." - }, - { - "left": "pull_request:8618", - "right": "pull_request:9736", - "accept": false, - "reason": "SD3 tiny autoencoder shift-factor work is unrelated to Allegro T2V model/pipeline addition." - }, - { - "left": "pull_request:8722", - "right": "pull_request:9736", - "accept": false, - "reason": "Refactor of embeddings internals vs adding Allegro T2V; not the same issue or change." - }, { "left": "pull_request:9175", "right": "pull_request:9816", "accept": false, - "reason": "Flux Union ControlNet support is unrelated to RF-inversion Flux pipeline addition." + "reason": "Both are Flux-related, but one adds Union ControlNet support and the other adds an RF-inversion Flux pipeline; different code paths and user-facing features." }, { "left": "pull_request:12416", "right": "pull_request:13126", "accept": false, - "reason": "Both are modular pipeline/lora-adjacent but for different QwenImage/Z-Image features; not the same concrete bug or feature." + "reason": "Both involve model-adjacent modular/LoRA work, but they target different projects and problems: QwenImage Edit Plus vs Z-Image distilled LoRAs." }, { "left": "pull_request:12519", "right": "pull_request:13321", "accept": false, - "reason": "Toctree sorting cleanup vs guarding torchvision import in Cosmos are unrelated." + "reason": "One is a docs toctree sorting cleanup, the other fixes an unguarded torchvision import in Cosmos; unrelated changes." }, { "left": "pull_request:4062", "right": "pull_request:5438", "accept": false, - "reason": "Shap-E mesh output support and latent consistency pipeline addition are different features." + "reason": "Shap-E mesh output support and Latent Consistency support are different features with no shared bug or mergeable unified patch." }, { "left": "pull_request:12021", "right": "pull_request:12095", "accept": false, - "reason": "LTX attention-backend support vs LTX 0.9.8 model release; different code paths and goals." + "reason": "Both touch LTX, but one adds attention backend support in core code while the other introduces a new LTX Video model release and pipeline/docs updates." }, { "left": "pull_request:10354", "right": "pull_request:9696", "accept": false, - "reason": "Flux IP-Adapter tester fix and PR test workflow input changes are unrelated." + "reason": "A Flux IP-Adapter tester fix and a workflow input-name update are unrelated." }, { "left": "pull_request:10497", "right": "pull_request:11067", "accept": false, - "reason": "Tokenizers install path fix in CI is unrelated to pinning transformers for benchmarks." + "reason": "Both are CI-related, but one adjusts tokenizers installation and the other pins transformers for benchmarking; different workflows and problems." }, { "left": "pull_request:10557", "right": "pull_request:9696", "accept": false, - "reason": "Slow test fixes for InstructPix2Pix are unrelated to PR workflow changes." + "reason": "One fixes StableDiffusion single-file slow tests; the other changes a PR test workflow. No shared underlying bug." }, { "left": "pull_request:10894", "right": "pull_request:11975", "accept": false, - "reason": "Docs toctree sorting and Flux Dreambooth style fix are unrelated." + "reason": "Docs toctree sorting and a style-only training script edit are not the same change." }, { "left": "pull_request:11243", "right": "pull_request:9807", "accept": false, - "reason": "LoRA conversion support for Musubi Wan and AutoencoderKL LoRA adapter support are different features/code paths." + "reason": "Both concern LoRA, but they fix different conversion/loading paths for different model families; not the same concrete bug." }, { "left": "pull_request:1229", "right": "pull_request:1458", "accept": false, - "reason": "Dreambooth README bitsandbytes guidance and onnxruntime candidate detection are unrelated." + "reason": "A README install note and an ONNX runtime candidate list update are unrelated." }, { "left": "pull_request:12424", "right": "pull_request:13338", "accept": false, - "reason": "Dockerfile definition cleanup and Claude workflow permissions are unrelated CI/workflow changes." + "reason": "Dockerfile cleanup and adding id-token permissions to a workflow are separate infra fixes." }, { "left": "pull_request:12960", "right": "pull_request:13364", "accept": false, - "reason": "FSDP detection in Flux Dreambooth scripts and MotionConv2d dtype casting are different problems." + "reason": "DreamBooth FSDP detection and MotionConv2d dtype casting are different code paths and bugs." }, { "left": "pull_request:1582", "right": "pull_request:676", "accept": false, - "reason": "RePaint scheduler math fix is unrelated to a variable rename in ResNet upsampling." + "reason": "RePaint scheduler logic and a variable rename in resnet are unrelated." }, { "left": "pull_request:1583", "right": "pull_request:676", "accept": false, - "reason": "Legacy inpaint scheduler logic change and ResNet variable rename are unrelated." + "reason": "A Stable Diffusion inpaint timestep fix and a resnet variable rename are not the same issue." }, { "left": "pull_request:1913", "right": "pull_request:823", "accept": false, - "reason": "Input-type validation for img2img pipelines and negative-prompt type mismatch handling are distinct validation fixes." + "reason": "Input-type validation in img2img pipelines and negative-prompt type checking are distinct validation fixes." }, { "left": "pull_request:1942", "right": "pull_request:2001", "accept": false, - "reason": "Unused upcast flag fix in a conversion script is unrelated to ColossalAI seed/launch typo fixes." + "reason": "A conversion-script flag rename and a ColossalAI seed-launch fix are unrelated." + }, + { + "left": "pull_request:2655", + "right": "pull_request:3294", + "accept": false, + "reason": "Both are LoRA loader changes, but one corrects loading/saving behavior while the other adds support for a new 'lora_' prefix; they are related only at a broad subsystem level." + }, + { + "left": "pull_request:3034", + "right": "pull_request:3446", + "accept": false, + "reason": "Wrong accelerate parameter name vs a warning-message wording fix; different concrete bugs." + }, + { + "left": "pull_request:3128", + "right": "pull_request:3673", + "accept": false, + "reason": "Img2img timestep selection and multi-GPU DreamBooth behavior are unrelated." + }, + { + "left": "pull_request:3590", + "right": "pull_request:5807", + "accept": false, + "reason": "StableDiffusionCanvasPipeline support and lpw SDXL prompt-embedding duplication are separate community-pipeline features." + }, + { + "left": "pull_request:3681", + "right": "pull_request:4690", + "accept": false, + "reason": "Checkpoint limit handling in training scripts and a DPMSolver scheduler refactor are unrelated." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4706", + "accept": false, + "reason": "An SDXL DreamBooth LoRA training patch set and a missing-default fix for fc alphas are different LoRA issues." + }, + { + "left": "pull_request:4744", + "right": "pull_request:5393", + "accept": false, + "reason": "A docstring correction in AudioLDM and a training/docs API update are not the same change." + }, + { + "left": "pull_request:4793", + "right": "pull_request:6994", + "accept": false, + "reason": "AutoPipeline kwargs passthrough and passing torch_dtype in from_single_file are different loader fixes with no shared failing code path." } ] }, @@ -163260,7 +163449,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 192, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -163268,10 +163457,20 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10122", + "pull_request:10126", + "pull_request:10140", + "pull_request:10188", + "pull_request:10326", + "pull_request:10358", + "pull_request:10431", + "pull_request:10595", "pull_request:10723", "pull_request:11916", "pull_request:11967", + "pull_request:12109", "pull_request:12226", + "pull_request:12301", "pull_request:12490", "pull_request:12534", "pull_request:12549", @@ -163281,51 +163480,33 @@ "pull_request:13490", "pull_request:2193", "pull_request:2452", - "pull_request:2655", - "pull_request:3034", - "pull_request:3128", - "pull_request:3294", - "pull_request:3446", + "pull_request:2902", "pull_request:3587", - "pull_request:3590", - "pull_request:3673", - "pull_request:3681", "pull_request:3770", - "pull_request:4262", "pull_request:4499", "pull_request:4500", "pull_request:4501", "pull_request:4502", - "pull_request:4690", - "pull_request:4706", - "pull_request:4744", - "pull_request:4793", + "pull_request:5033", "pull_request:5301", - "pull_request:5393", "pull_request:5559", "pull_request:5697", - "pull_request:5807", "pull_request:5898", "pull_request:6217", - "pull_request:6994", "pull_request:7062", "pull_request:7410", "pull_request:7461", "pull_request:7714", "pull_request:7720", + "pull_request:8691", + "pull_request:8897", "pull_request:9247", + "pull_request:9257", "pull_request:9311", + "pull_request:9592", "pull_request:9723" ], "soft_pairs": [ - "pull_request:2655|pull_request:3294", - "pull_request:3034|pull_request:3446", - "pull_request:3128|pull_request:3673", - "pull_request:3590|pull_request:5807", - "pull_request:3681|pull_request:4690", - "pull_request:4262|pull_request:4706", - "pull_request:4744|pull_request:5393", - "pull_request:4793|pull_request:6994", "pull_request:5697|pull_request:7461", "pull_request:5898|pull_request:7714", "pull_request:7062|pull_request:7461", @@ -163343,192 +163524,182 @@ "pull_request:3770|pull_request:4500", "pull_request:3770|pull_request:4501", "pull_request:3770|pull_request:4502", - "pull_request:5301|pull_request:6217" + "pull_request:5301|pull_request:6217", + "pull_request:10140|pull_request:8897", + "pull_request:12109|pull_request:12301", + "pull_request:2902|pull_request:5033", + "pull_request:10122|pull_request:10188", + "pull_request:10126|pull_request:9257", + "pull_request:10326|pull_request:8691", + "pull_request:10358|pull_request:9592", + "pull_request:10431|pull_request:10595" ], - "prepared_review_unit_hash": "cefce8640c987051096fb79a4e80f92158f9585a0dd9ca4f68c3080e1bf9f026", + "prepared_review_unit_hash": "cce8983c8de946dac4aa5c05c9a52dde0e1b01458b4049019bd799221d2af53a", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 102385, - "estimated_input_tokens": 25597, - "estimated_eval_tokens": 51450 + "serialized_chars": 101778, + "estimated_input_tokens": 25445, + "estimated_eval_tokens": 51146 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set does not form a real duplicate cluster. Most pairs only share a broad subsystem, file, or superficial wording; they fix different bugs or add different features, so all soft edges should be rejected.", - "confidence": 0.91, + "summary": "This cluster is a set of mostly unrelated pull requests; the soft-similarity signals are broad topical overlaps, not true duplicates or mergeable variants of the same fix.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 11916 is the broadest and most central artifact here: it makes the largest attention-dispatch refactor and other attention-backend PRs are closest in scope to it, but it is still not a duplicate of them.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 11916 is the best single representative of the attention-backend subset because it has the broadest code impact and anchors the later attention-related follow-up PRs in this list.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:2655", - "right": "pull_request:3294", - "accept": false, - "reason": "Both touch loaders.py and LoRA loading, but one fixes safetensors fallback while the other adds support for lora_ prefixes; different bugs and changes." - }, - { - "left": "pull_request:3034", - "right": "pull_request:3446", - "accept": false, - "reason": "Both are small parameter/doc fixes, but they address unrelated call sites and issues; not the same bug or change." - }, - { - "left": "pull_request:3128", - "right": "pull_request:3673", - "accept": false, - "reason": "Both involve timestep handling / training logic, but one fixes img2img timestep slicing and the other fixes multi-GPU DreamBooth; unrelated." - }, - { - "left": "pull_request:3590", - "right": "pull_request:5807", - "accept": false, - "reason": "Both are community pipeline additions/fixes, but they target different pipelines and features; no shared underlying change." - }, - { - "left": "pull_request:3681", - "right": "pull_request:4690", - "accept": false, - "reason": "Checkpoints_total_limit handling in training scripts is unrelated to the DPMSolverMultistepScheduler sigma refactor." - }, - { - "left": "pull_request:4262", - "right": "pull_request:4706", - "accept": false, - "reason": "One is SDXL DreamBooth/UNet training fixes; the other is LoRA alpha fallback in loaders. Different code paths and bugs." - }, - { - "left": "pull_request:4744", - "right": "pull_request:5393", - "accept": false, - "reason": "AudioLDM docstring cleanup is unrelated to the accelerate API update in training examples." - }, - { - "left": "pull_request:4793", - "right": "pull_request:6994", - "accept": false, - "reason": "AutoPipeline kwargs propagation and single-file torch_dtype propagation solve different loading issues." - }, { "left": "pull_request:5697", "right": "pull_request:7461", "accept": false, - "reason": "A PEFT/xFormers attention bug fix is unrelated to moving str_to_bool in testing utils." + "reason": "Completely different areas: LoRA/xformers attention bug vs a testing-utils helper definition fix." }, { "left": "pull_request:5898", "right": "pull_request:7714", "accept": false, - "reason": "A docs spacing tweak is unrelated to the AutoPipeline documentation overhaul." + "reason": "Both are docs-only, but they document different features and would not be merged as the same change." }, { "left": "pull_request:7062", "right": "pull_request:7461", "accept": false, - "reason": "Imagic docstring correction and testing-utils helper relocation are unrelated." + "reason": "Imagic docstring cleanup is unrelated to testing-utils boolean parsing." }, { "left": "pull_request:7410", "right": "pull_request:7720", "accept": false, - "reason": "FreeInit behavior in video/animatediff is unrelated to dead-code cleanup in assorted scripts." + "reason": "FreeInit behavior fix vs dead-code/f-string cleanup; different code paths and bugs." }, { "left": "pull_request:9247", "right": "pull_request:9723", "accept": false, - "reason": "Variant folder identification in pipeline loading is unrelated to Matryoshka schedule_shifted_power image examples." + "reason": "Different fixes in different modules: model variant detection vs Matryoshka schedule_shifted_power usage." }, { "left": "pull_request:12534", "right": "pull_request:13490", "accept": false, - "reason": "Both mention Flux/NPU, but one fixes positional embeddings for Ascend inference and the other changes NPU attention mask shape handling; different code paths." + "reason": "Both touch NPU-related code, but one fixes Flux rotary embeddings on Ascend and the other adjusts NPU attention mask shapes; separate problems." }, { "left": "pull_request:11916", "right": "pull_request:11967", "accept": false, - "reason": "Both touch attention_dispatch, but 11916 is a Flux/Chroma refactor while 11967 changes backend error handling; not the same fix." + "reason": "Shared attention-dispatch area, but one is a large refactor and the other changes runtime backend warning/error behavior." }, { "left": "pull_request:12549", "right": "pull_request:12919", "accept": false, - "reason": "Both are attention-backend related, but AITER backend support and NPU ulysses support are separate backend implementations and can\u2019t be merged as one PR." + "reason": "Different backends/platforms: AITER backend support vs NPU attention Ulysses enablement." }, { "left": "pull_request:10723", "right": "pull_request:9311", "accept": false, - "reason": "ControlNet Union per-control scaling and SDXL ControlNet CFG cutoff are different ControlNet features with different behaviors." + "reason": "Both are ControlNet-related, but one adds per-control scaling and the other adds CFG cutoff; distinct features." }, { "left": "pull_request:12226", "right": "pull_request:12736", "accept": false, - "reason": "A Qwen-Image ControlNet test and z-image regional compilation are unrelated." + "reason": "A fast pipeline test and a regional compilation flag for another model are unrelated." }, { "left": "pull_request:12490", "right": "pull_request:13321", "accept": false, - "reason": "Audioldm2 ClapConfig test fix and Cosmos torchvision import guard are unrelated." + "reason": "Test fixture correction for AudioLDM2 is unrelated to a guarded torchvision import in Cosmos." }, { "left": "pull_request:2193", "right": "pull_request:2452", "accept": false, - "reason": "Adding a SAG docs entry and adding core-api docs are separate documentation changes." + "reason": "Both are docs additions, but they cover entirely different documentation sections and topics." }, { "left": "pull_request:3587", "right": "pull_request:5559", "accept": false, - "reason": "Community CLIP-guided mixing docs/code and latent-consistency img2img pipeline work are unrelated." + "reason": "Community README update vs adding a latent-consistency img2img pipeline; not the same artifact." }, { "left": "pull_request:3770", "right": "pull_request:4499", "accept": false, - "reason": "ControlNet guidance start/stop and SDXL reference-attn/adain community pipeline are unrelated features." + "reason": "ControlNet guidance start/stop and SDXL reference-attn are different pipeline features." }, { "left": "pull_request:3770", "right": "pull_request:4500", "accept": false, - "reason": "ControlNet guidance scheduling and SDXL reference pipeline are unrelated." + "reason": "ControlNet guidance start/stop and SDXL reference-attn are different pipeline features." }, { "left": "pull_request:3770", "right": "pull_request:4501", "accept": false, - "reason": "ControlNet guidance scheduling and SDXL reference pipeline are unrelated." + "reason": "ControlNet guidance start/stop and SDXL reference-attn are different pipeline features." }, { "left": "pull_request:3770", "right": "pull_request:4502", "accept": false, - "reason": "ControlNet guidance scheduling and SDXL reference pipeline are unrelated." + "reason": "ControlNet guidance start/stop and SDXL reference-attn are different pipeline features." }, { "left": "pull_request:5301", "right": "pull_request:6217", "accept": false, - "reason": "SDXL pipeline optional-component fixes are unrelated to Stable Video Diffusion torch-compile VAE inspection changes." + "reason": "SDXL text-encoder handling and Stable Video Diffusion torch-compile support are unrelated." + }, + { + "left": "pull_request:10122", + "right": "pull_request:10188", + "accept": false, + "reason": "Flux attention-mask plumbing vs Flux LoRA-loading error handling are separate concerns." + }, + { + "left": "pull_request:10126", + "right": "pull_request:9257", + "accept": false, + "reason": "LoRA save_attn_procs deprecation is unrelated to a DreamBooth Flux training-script fix." + }, + { + "left": "pull_request:10326", + "right": "pull_request:8691", + "accept": false, + "reason": "MPS workflow quoting fix vs LoRA loading behavior; no common bug." + }, + { + "left": "pull_request:10358", + "right": "pull_request:9592", + "accept": false, + "reason": "Community projects docs update and Flax docstring refactor are unrelated docs changes." + }, + { + "left": "pull_request:10431", + "right": "pull_request:10595", + "accept": false, + "reason": "Different model-specific fixes: Sana positional embedding initialization vs bf16 casting in quadratic attention." } ] }, @@ -163544,23 +163715,15 @@ "component_index": 1, "component_count": 266, "review_unit_index": 193, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10122", - "pull_request:10126", - "pull_request:10140", - "pull_request:10188", - "pull_request:10326", - "pull_request:10358", - "pull_request:10431", "pull_request:10539", - "pull_request:10595", "pull_request:10704", "pull_request:11025", "pull_request:11059", @@ -163577,39 +163740,39 @@ "pull_request:11892", "pull_request:11894", "pull_request:12005", - "pull_request:12109", "pull_request:12137", "pull_request:1215", "pull_request:12231", "pull_request:1230", - "pull_request:12301", "pull_request:12363", + "pull_request:12425", "pull_request:12527", "pull_request:12530", + "pull_request:12586", "pull_request:12805", "pull_request:12998", + "pull_request:1311", + "pull_request:13134", + "pull_request:13472", + "pull_request:1360", + "pull_request:1425", + "pull_request:1625", + "pull_request:171", + "pull_request:1780", + "pull_request:1849", "pull_request:1968", - "pull_request:2902", - "pull_request:5033", + "pull_request:1978", + "pull_request:227", + "pull_request:2386", + "pull_request:2640", + "pull_request:2715", "pull_request:796", - "pull_request:8691", - "pull_request:8897", "pull_request:8932", "pull_request:9103", - "pull_request:9257", - "pull_request:9592", "pull_request:9688", "pull_request:9769" ], "soft_pairs": [ - "pull_request:10140|pull_request:8897", - "pull_request:12109|pull_request:12301", - "pull_request:2902|pull_request:5033", - "pull_request:10122|pull_request:10188", - "pull_request:10126|pull_request:9257", - "pull_request:10326|pull_request:8691", - "pull_request:10358|pull_request:9592", - "pull_request:10431|pull_request:10595", "pull_request:10539|pull_request:8932", "pull_request:10704|pull_request:9103", "pull_request:11025|pull_request:11066", @@ -163625,180 +163788,188 @@ "pull_request:1215|pull_request:1968", "pull_request:12231|pull_request:12527", "pull_request:1230|pull_request:796", - "pull_request:12363|pull_request:12530" + "pull_request:12363|pull_request:12530", + "pull_request:12425|pull_request:13134", + "pull_request:12586|pull_request:13472", + "pull_request:1311|pull_request:227", + "pull_request:1360|pull_request:1978", + "pull_request:1425|pull_request:2386", + "pull_request:1625|pull_request:2640", + "pull_request:171|pull_request:1849", + "pull_request:1780|pull_request:2715" ], - "prepared_review_unit_hash": "3de8fc26c8f92d2eca3fd0b7d4495036654daa52e652af549d5295ac5f1c202f", + "prepared_review_unit_hash": "350fe8614020bb454791f03aeb7d1c89332d4429fe311faddf50b60f7176aaf6", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 99539, - "estimated_input_tokens": 24885, - "estimated_eval_tokens": 50026 + "serialized_chars": 97581, + "estimated_input_tokens": 24396, + "estimated_eval_tokens": 49048 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab-bag of unrelated PRs spanning docs, CI/workflows, loaders, schedulers, pipelines, and model additions. A few pairs share broad subsystems like Flux, LoRA, or testing, but none appear to fix the same concrete bug or implement the same change.", - "confidence": 0.97, + "summary": "The soft-similarity pairs are mostly superficial matches on keywords or broad subsystems (Flux, docs, CI, tests), but they do not describe the same underlying bug or change. No duplicate pair is convincing here.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:10140", - "right": "pull_request:8897", - "accept": false, - "reason": "Both are large feature PRs, but they add different model families (ConsisID vs AnimateDiff SparseCtrl) and touch different codepaths." - }, - { - "left": "pull_request:12109", - "right": "pull_request:12301", - "accept": false, - "reason": "UniPC scheduler sigma handling and Qwen-Image ControlNet inpainting are unrelated changes with no shared bug." - }, - { - "left": "pull_request:2902", - "right": "pull_request:5033", - "accept": false, - "reason": "One fixes cross_attention_kwargs in img2img pipelines; the other fixes inpaint latent/mask handling. Different bugs." - }, - { - "left": "pull_request:10122", - "right": "pull_request:10188", - "accept": false, - "reason": "Flux attention-mask wiring is unrelated to LoRA loading/shape-expansion logging in loaders." - }, - { - "left": "pull_request:10126", - "right": "pull_request:9257", - "accept": false, - "reason": "Deprecating save_attn_procs and fixing DreamBooth Flux prompt tokenization are separate issues." - }, - { - "left": "pull_request:10326", - "right": "pull_request:8691", - "accept": false, - "reason": "CI workflow quoting fix versus LoRA vanilla loading logic; no common underlying defect." - }, - { - "left": "pull_request:10358", - "right": "pull_request:9592", - "accept": false, - "reason": "Community projects docs update and Flax embedding docstring refactor are unrelated documentation edits." - }, - { - "left": "pull_request:10431", - "right": "pull_request:10595", - "accept": false, - "reason": "Both mention Sana, but they fix different attention/positional-encoding problems in different codepaths." - }, { "left": "pull_request:10539", "right": "pull_request:8932", "accept": false, - "reason": "Flux true-CFG condition fix is unrelated to adding Temporal Transformer tests." + "reason": "Flux CFG logic fix vs. Temporal Transformer test coverage; unrelated changes." }, { "left": "pull_request:10704", "right": "pull_request:9103", "accept": false, - "reason": "Image-to-image docs cleanup and distributed LR scheduler logic are not the same change." + "reason": "Documentation edit for img2img vs. distributed LR scheduler fix; different code paths and goals." }, { "left": "pull_request:11025", "right": "pull_request:11066", "accept": false, - "reason": "Wan LoRA conversion support and HunyuanVideo I2V model support are distinct feature additions." + "reason": "Wan LoRA conversion support vs. new HunyuanVideo I2V model support; separate features." }, { "left": "pull_request:11059", "right": "pull_request:9769", "accept": false, - "reason": "Wan t2v LoRA conversion fix and Mochi T2V core model addition are unrelated." + "reason": "Wan LoRA conversion tweak vs. Mochi model/pipeline addition; not the same change." }, { "left": "pull_request:11067", "right": "pull_request:11614", "accept": false, - "reason": "Benchmark dependency pinning and docker build script changes address different CI concerns." + "reason": "Benchmark transformers pinning vs. docker CI security/workflow fix; unrelated CI changes." }, { "left": "pull_request:11236", "right": "pull_request:9688", "accept": false, - "reason": "A Sana Sprint docs title tweak and Flux ControlNet docs update are unrelated docs changes." + "reason": "Two different documentation updates for different pipelines; no shared underlying bug or feature." }, { "left": "pull_request:11325", "right": "pull_request:11770", "accept": false, - "reason": "HiDream docstring replacement and LoRA docs cleanup do not share the same underlying fix." + "reason": "HiDream docstring decorator fix vs. LoRA docs cleanup; both docs-oriented but not the same issue." }, { "left": "pull_request:11743", "right": "pull_request:11814", "accept": false, - "reason": "Flux LoRA quantization example script and Flux gradient-checkpointing kwargs forwarding are different problems." + "reason": "Research training script addition vs. Flux gradient-checkpointing bug fix; unrelated." }, { "left": "pull_request:11767", "right": "pull_request:12005", "accept": false, - "reason": "Group-offloading disk-memory-address fixes and sharded checkpoint local_files_only behavior are separate core utility changes." + "reason": "Shared modeling tests file, but one fixes group-offloading disk addressing and the other sharded checkpoint local-only behavior." }, { "left": "pull_request:11786", "right": "pull_request:11894", "accept": false, - "reason": "Big GPU marker fix and k-diffusion pinning are unrelated CI/dependency changes." + "reason": "Big-GPU test marker correction vs. k-diffusion dependency pin; different subsystems." }, { "left": "pull_request:11892", "right": "pull_request:12805", "accept": false, - "reason": "Removing k-diffusion from core dependencies and changing the PyPI workflow are not the same change." + "reason": "Removing k-diffusion from core deps vs. PyPI publish workflow edits; not the same change." }, { "left": "pull_request:12137", "right": "pull_request:12998", "accept": false, - "reason": "Parallel-loading constant refactor and style-only doc/pipeline whitespace cleanup are unrelated." + "reason": "Parallel-loading constant refactor vs. broad style/docstring cleanup; unrelated." }, { "left": "pull_request:1215", "right": "pull_request:1968", "accept": false, - "reason": "DDIM eta/cpu-generator behavior fix is unrelated to CPU offload documentation corrections." + "reason": "DDIM eta/cpu-generator bug fix vs. CPU offload doc corrections; different artifact types and issues." }, { "left": "pull_request:12231", "right": "pull_request:12527", "accept": false, - "reason": "A typo/docs cleanup across examples is unrelated to the Kandinsky5 no-CFG logic fix." + "reason": "Typo fix in example docs vs. Kandinsky5 CFG bug fix; no overlap beyond generic cleanup." }, { "left": "pull_request:1230", "right": "pull_request:796", "accept": false, - "reason": "Both involve MPS/testing, but one fixes DDIM/fast test behavior while the other adds Apple M1 test infrastructure; not the same bug." + "reason": "MPS fast-test fix vs. adding Apple M1 tests; related platform, but not the same concrete bug/change." }, { "left": "pull_request:12363", "right": "pull_request:12530", "accept": false, - "reason": "QwenImageEdit Plus docs/tests and an xfailed Stable Cascade test are unrelated." + "reason": "QwenImageEdit Plus docs/test addition vs. Stable Cascade test xfail; unrelated." + }, + { + "left": "pull_request:12425", + "right": "pull_request:13134", + "accept": false, + "reason": "torch.distributed import guard vs. Cosmos transformer typing fix; different code areas." + }, + { + "left": "pull_request:12586", + "right": "pull_request:13472", + "accept": false, + "reason": "Documentation ordering change vs. agents docs float64 guidance; unrelated docs changes." + }, + { + "left": "pull_request:1311", + "right": "pull_request:227", + "accept": false, + "reason": "AltDiffusion test additions vs. scheduler add_noise implementation; unrelated." + }, + { + "left": "pull_request:1360", + "right": "pull_request:1978", + "accept": false, + "reason": "k-diffusion community pipeline docs/code vs. push_to_hub implementation work; different features." + }, + { + "left": "pull_request:1425", + "right": "pull_request:2386", + "accept": false, + "reason": "Simple docs typo fixes vs. UniPC docs rewrite; not the same underlying issue." + }, + { + "left": "pull_request:1625", + "right": "pull_request:2640", + "accept": false, + "reason": "k-diffusion pipeline sizing fix vs. telemetry removal in pipeline loading; unrelated." + }, + { + "left": "pull_request:171", + "right": "pull_request:1849", + "accept": false, + "reason": "LDM eta-condition fix vs. cross_attention_kwargs support in a different attention module; not duplicates." + }, + { + "left": "pull_request:1780", + "right": "pull_request:2715", + "accept": false, + "reason": "Upscale preprocessing comment fix vs. broad typo cleanup; superficial text similarity only." } ] }, @@ -163814,40 +163985,24 @@ "component_index": 1, "component_count": 266, "review_unit_index": 194, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:12425", - "pull_request:12586", - "pull_request:1311", - "pull_request:13134", - "pull_request:13472", - "pull_request:1360", - "pull_request:1425", - "pull_request:1625", - "pull_request:171", - "pull_request:1780", - "pull_request:1849", "pull_request:1850", - "pull_request:1978", "pull_request:2122", "pull_request:2126", - "pull_request:227", "pull_request:2365", - "pull_request:2386", "pull_request:2529", "pull_request:2530", "pull_request:2553", "pull_request:2598", "pull_request:2604", - "pull_request:2640", "pull_request:2693", - "pull_request:2715", "pull_request:2757", "pull_request:2861", "pull_request:2968", @@ -163859,26 +164014,34 @@ "pull_request:3642", "pull_request:3644", "pull_request:3797", + "pull_request:386", + "pull_request:3942", "pull_request:3957", + "pull_request:3989", + "pull_request:3995", + "pull_request:400", + "pull_request:401", "pull_request:402", + "pull_request:4058", + "pull_request:4106", + "pull_request:4203", + "pull_request:4285", "pull_request:4442", "pull_request:5036", + "pull_request:5098", + "pull_request:5276", + "pull_request:5369", "pull_request:5396", "pull_request:5526", + "pull_request:5699", + "pull_request:5724", + "pull_request:5738", "pull_request:5844", "pull_request:628", "pull_request:796", "pull_request:819" ], "soft_pairs": [ - "pull_request:12425|pull_request:13134", - "pull_request:12586|pull_request:13472", - "pull_request:1311|pull_request:227", - "pull_request:1360|pull_request:1978", - "pull_request:1425|pull_request:2386", - "pull_request:1625|pull_request:2640", - "pull_request:171|pull_request:1849", - "pull_request:1780|pull_request:2715", "pull_request:1850|pull_request:2861", "pull_request:2122|pull_request:402", "pull_request:2126|pull_request:2693", @@ -163894,187 +164057,212 @@ "pull_request:3572|pull_request:5036", "pull_request:3607|pull_request:5526", "pull_request:3644|pull_request:5844", - "pull_request:3797|pull_request:3957" + "pull_request:3797|pull_request:3957", + "pull_request:386|pull_request:401", + "pull_request:3942|pull_request:5098", + "pull_request:3942|pull_request:5276", + "pull_request:3942|pull_request:5724", + "pull_request:3989|pull_request:5699", + "pull_request:3995|pull_request:4058", + "pull_request:3995|pull_request:4285", + "pull_request:400|pull_request:401", + "pull_request:4106|pull_request:5738", + "pull_request:4203|pull_request:5369" ], - "prepared_review_unit_hash": "4e06fdcc98cc3484ba134390b82660a702337240bd9828f804b5ec5e517c7de0", + "prepared_review_unit_hash": "b6d2e0fb0a29c9420f49ecfb1b0f9d99754908a0d4af32894eb6836314e300c4", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 94295, - "estimated_input_tokens": 23574, - "estimated_eval_tokens": 47404 + "soft_pair_count": 26, + "serialized_chars": 98789, + "estimated_input_tokens": 24698, + "estimated_eval_tokens": 49652 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft links are mostly false positives: this cluster mixes unrelated docs, tests, typing, and bug-fix PRs across different subsystems. I would not deduplicate any of the proposed pairs.", - "confidence": 0.96, + "summary": "Almost all soft-similarity pairs are false positives across unrelated docs/tests/maintenance changes. The only plausible mergeable duplicate is the attention-naming pair `3797`/`3957`, which both appear to address the same core model attention-head naming correction.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR fits this cluster; the items are heterogeneous and do not resolve to one underlying bug or change.", + "canonical_pr_reason": "PR 3797 is the strongest representative PR: it is merged and makes a broad attention-head naming correction across core model files. It should not be described as explicitly issue-linked, because the packet shows `explicit_issue_links: 0`.", "best_issue_reason": null, - "best_pr_reason": "PR #3797 is the strongest standalone representative by substance and scope, but it still only covers its own attention-naming fix and should not serve as a duplicate umbrella for the rest.", + "best_pr_reason": "PR 3797 is the best standalone PR in this set because it is a concrete core-model bug fix affecting many attention-related files, whereas the other PRs are mostly docs/tests or narrower maintenance changes. Its standalone strength does not rely on any explicit issue link.", "soft_edge_verdicts": [ - { - "left": "pull_request:12425", - "right": "pull_request:13134", - "accept": false, - "reason": "Different problems: a torch.distributed import guard vs a typing annotation change in Cosmos transformer." - }, - { - "left": "pull_request:12586", - "right": "pull_request:13472", - "accept": false, - "reason": "Unrelated docs sorting change vs a model-gotcha docs update about float64." - }, - { - "left": "pull_request:1311", - "right": "pull_request:227", - "accept": false, - "reason": "AltDiffusion test/dependency changes and LMS/PNDM scheduler add_noise implementation are not the same fix." - }, - { - "left": "pull_request:1360", - "right": "pull_request:1978", - "accept": false, - "reason": "Community pipeline addition vs hub push-to-hub infrastructure; different code paths and goals." - }, - { - "left": "pull_request:1425", - "right": "pull_request:2386", - "accept": false, - "reason": "Both are docs-only, but they address different documentation pages and issues." - }, - { - "left": "pull_request:1625", - "right": "pull_request:2640", - "accept": false, - "reason": "k-diffusion height/width defaults and telemetry removal are unrelated pipeline changes." - }, - { - "left": "pull_request:171", - "right": "pull_request:1849", - "accept": false, - "reason": "Scheduler eta-condition bugfix vs cross_attention_kwargs plumbing in a different model path." - }, - { - "left": "pull_request:1780", - "right": "pull_request:2715", - "accept": false, - "reason": "A comment typo in upscale preprocessing is not the same as a broad multi-file typo cleanup." - }, { "left": "pull_request:1850", "right": "pull_request:2861", "accept": false, - "reason": "Different tests for different components: ResnetBlock2D vs StableUnCLIP image_embeds handling." + "reason": "Different test additions for different features: ResnetBlock2D coverage vs StableUnCLIP `image_embeds=None` handling." }, { "left": "pull_request:2122", "right": "pull_request:402", "accept": false, - "reason": "Separate docs efforts: safetensors onboarding vs intro-section restructuring." + "reason": "Both are docs-related, but they cover unrelated topics: safetensors usage vs a broad docs intro/toctree overhaul." }, { "left": "pull_request:2126", "right": "pull_request:2693", "accept": false, - "reason": "Docstring cleanup in pipeline loading is unrelated to fixing an inpaint image link." + "reason": "Unrelated doc fixes: duplicate `torch_dtype` docstring cleanup vs an inpaint image-link correction." }, { "left": "pull_request:2365", "right": "pull_request:327", "accept": false, - "reason": "Scheduler enum note and PNDM pipeline typing are not the same underlying change." + "reason": "A scheduler enum note is not the same change as adding type hints to the PNDM pipeline." }, { "left": "pull_request:2529", "right": "pull_request:2757", "accept": false, - "reason": "Checkpoint conversion fallback and depth-map batching are unrelated bug fixes." + "reason": "Different code paths: OpenCLIP checkpoint conversion fallback vs depth-map batching in depth2img." }, { "left": "pull_request:2530", "right": "pull_request:796", "accept": false, - "reason": "EMA tests and Apple M1 test infra additions are separate test coverage changes." + "reason": "Both add tests, but they target different features and failure modes; not the same bug fix." }, { "left": "pull_request:2553", "right": "pull_request:3642", "accept": false, - "reason": "Flax CFG flag implementation and TensorRT inpainting pipeline addition are different features." + "reason": "CFG flag handling in Flax Stable Diffusion is unrelated to adding TensorRT inpainting pipeline support." }, { "left": "pull_request:2598", "right": "pull_request:628", "accept": false, - "reason": "Paint-by-Example docs update is not the same as the DDPM/colab table edit." + "reason": "Separate documentation edits for different pages/features; no shared underlying change." }, { "left": "pull_request:2604", "right": "pull_request:628", "accept": false, - "reason": "Both touch docs tables, but one fixes a link and the other reorders pipeline listings; not the same change." + "reason": "A colab-link fix in the index is unrelated to the DDPM colab placement change." }, { "left": "pull_request:2968", "right": "pull_request:819", "accept": false, - "reason": "Post-processing logic fix and DDIM timestep dtype fix address different bugs." + "reason": "Different bugs in different code paths: img2img post-processing vs DDIM timestep dtype on Windows." }, { "left": "pull_request:3127", "right": "pull_request:5396", "accept": false, - "reason": "Safety checker image handling and stale-bot timezone handling are unrelated." + "reason": "Safety checker tensor handling is unrelated to the stale bot timezone datetime fix." }, { "left": "pull_request:3473", "right": "pull_request:4442", "accept": false, - "reason": "Different docs additions: textual inversion guidance vs distilled SD inference." + "reason": "Both add docs, but for different model features and tutorial pages; not one underlying change." }, { "left": "pull_request:3572", "right": "pull_request:5036", "accept": false, - "reason": "ControlNet inpaint docs and import/dummy-object fixes are not the same artifact." + "reason": "ControlNet inpaint docs and import/dummy-object plumbing are unrelated maintenance changes." }, { "left": "pull_request:3607", "right": "pull_request:5526", "accept": false, - "reason": "Attention-processor temb fix and scheduler order handling are separate code paths." + "reason": "Core attention processor changes are unrelated to the second-order scheduler logic fix." }, { "left": "pull_request:3644", "right": "pull_request:5844", "accept": false, - "reason": "Two unrelated docs updates, one for InstructPix2Pix and one for a Japanese SDXL reference." + "reason": "Both are docs updates, but they document different examples and features." }, { "left": "pull_request:3797", "right": "pull_request:3957", + "accept": true, + "reason": "Both concern the same attention-head naming/correction work in the same UNet model code paths, and `3957` looks like a plausible follow-up piece of the same fix." + }, + { + "left": "pull_request:386", + "right": "pull_request:401", + "accept": false, + "reason": "Different pipeline docs and code: Stable Diffusion pipeline overview vs Karras VE docs." + }, + { + "left": "pull_request:3942", + "right": "pull_request:5098", + "accept": false, + "reason": "A test harness fix for np outputs is unrelated to Kandinsky precision test coverage." + }, + { + "left": "pull_request:3942", + "right": "pull_request:5276", + "accept": false, + "reason": "Test-output handling and scheduler 1D support are unrelated changes." + }, + { + "left": "pull_request:3942", + "right": "pull_request:5724", + "accept": false, + "reason": "A common test helper change does not match a LoRA backend test/decorator change." + }, + { + "left": "pull_request:3989", + "right": "pull_request:5699", + "accept": false, + "reason": "Slow-test fixes in SDXL do not share the same underlying change as a styling-only stable unclip/text-unet cleanup." + }, + { + "left": "pull_request:3995", + "right": "pull_request:4058", + "accept": false, + "reason": "SDXL prompt-embedding plumbing is unrelated to a docs image/table update." + }, + { + "left": "pull_request:3995", + "right": "pull_request:4285", + "accept": false, + "reason": "Core SDXL pipeline changes are unrelated to ONNX/OpenVINO documentation work." + }, + { + "left": "pull_request:400", + "right": "pull_request:401", + "accept": false, + "reason": "They are different docs pages for different pipelines; no shared concrete change." + }, + { + "left": "pull_request:4106", + "right": "pull_request:5738", "accept": false, - "reason": "Both are about attention-head naming, but 3957 is a separate follow-up/WIP patch, not the same concrete PR change." + "reason": "A new Tortoise TTS pipeline and LoRA adapter deletion are entirely separate features." + }, + { + "left": "pull_request:4203", + "right": "pull_request:5369", + "accept": false, + "reason": "DreamBooth README typo cleanup and general docs minor fixes are separate doc edits, not one fix." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, + "evaluator_result": { + "accept": false, + "feedback": "The evaluation is mostly conservative, but it overstates the evidence for PR 3797. In the packet, PR 3797 is merged and clearly about the attention naming fix, but it does **not** have an explicit issue link (`explicit_issue_links: 0`), so calling it \u201cexplicitly issue-linked\u201d and citing \u201cclearest issue context\u201d is unsupported. The borderline 3797\u21943957 pair is plausibly the same attention-name correction, but the standalone PR reasoning needs to stay grounded in the packet." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -164083,7 +164271,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 195, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -164091,17 +164279,7 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:386", - "pull_request:3942", - "pull_request:3989", - "pull_request:3995", - "pull_request:400", - "pull_request:401", - "pull_request:4058", - "pull_request:4106", - "pull_request:4203", "pull_request:4211", - "pull_request:4285", "pull_request:4347", "pull_request:4397", "pull_request:4499", @@ -164109,10 +164287,8 @@ "pull_request:4811", "pull_request:4856", "pull_request:5097", - "pull_request:5098", "pull_request:5164", "pull_request:5275", - "pull_request:5276", "pull_request:5369", "pull_request:5384", "pull_request:5391", @@ -164121,9 +164297,6 @@ "pull_request:5602", "pull_request:5653", "pull_request:5698", - "pull_request:5699", - "pull_request:5724", - "pull_request:5738", "pull_request:5741", "pull_request:5831", "pull_request:5982", @@ -164131,25 +164304,30 @@ "pull_request:6198", "pull_request:6221", "pull_request:6305", + "pull_request:6308", "pull_request:6360", "pull_request:6365", + "pull_request:6369", + "pull_request:6498", + "pull_request:6564", "pull_request:6997", + "pull_request:6998", + "pull_request:7247", + "pull_request:7260", "pull_request:7328", + "pull_request:7353", "pull_request:7371", "pull_request:7473", - "pull_request:7679" + "pull_request:7857", + "pull_request:8298", + "pull_request:8404", + "pull_request:8452", + "pull_request:9242", + "pull_request:9247", + "pull_request:9426", + "pull_request:9718" ], "soft_pairs": [ - "pull_request:386|pull_request:401", - "pull_request:3942|pull_request:5098", - "pull_request:3942|pull_request:5276", - "pull_request:3942|pull_request:5724", - "pull_request:3989|pull_request:5699", - "pull_request:3995|pull_request:4058", - "pull_request:3995|pull_request:4285", - "pull_request:400|pull_request:401", - "pull_request:4106|pull_request:5738", - "pull_request:4203|pull_request:5369", "pull_request:4211|pull_request:6198", "pull_request:4347|pull_request:5275", "pull_request:4397|pull_request:5391", @@ -164165,193 +164343,195 @@ "pull_request:5602|pull_request:5982", "pull_request:5653|pull_request:7371", "pull_request:5698|pull_request:7473", - "pull_request:5741|pull_request:7679", - "pull_request:6305|pull_request:6997" + "pull_request:6305|pull_request:6997", + "pull_request:6308|pull_request:6369", + "pull_request:6498|pull_request:7857", + "pull_request:6564|pull_request:8404", + "pull_request:6998|pull_request:7260", + "pull_request:7247|pull_request:9247", + "pull_request:7353|pull_request:9426", + "pull_request:7857|pull_request:9242", + "pull_request:7857|pull_request:9718", + "pull_request:8298|pull_request:8452" ], - "prepared_review_unit_hash": "8a79f5b6633b15264bba71a0cead8b8cf29c7340690f0e2f9d863f8853ccb003", + "prepared_review_unit_hash": "fdb82ad4f1c62775f8b14ec3f684621b99fe52398142a7344e614e2a1ed0bd71", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 27, - "serialized_chars": 101363, - "estimated_input_tokens": 25341, - "estimated_eval_tokens": 50938 + "soft_pair_count": 25, + "serialized_chars": 98415, + "estimated_input_tokens": 24604, + "estimated_eval_tokens": 49464 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous docs/test/feature PRs with one true duplicate pair: the two SDXL reference-attn/adain PRs. Everything else is a broad-similarity false positive.", - "confidence": 0.97, + "summary": "This cluster is mostly a bundle of unrelated PRs spanning docs, schedulers, pipelines, tests, and CI. The only clear duplicate pair is the SDXL reference-attention PR submission. Everything else should stay separate.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 4499 is the cleanest representative of the only real duplicate pair (4499/4500): same SDXL reference_attn/reference_adain feature, same files, same change; 4500 looks like a near-identical repost.", + "canonical_pr_reason": "No single PR is a good canonical for the whole cluster because the items are unrelated; if forced to pick one representative, the near-identical SDXL reference-attention PR (#4499) is the clearest duplicate anchor.", "best_issue_reason": null, - "best_pr_reason": "4499 is the best representative PR in this cluster because it matches its twin exactly and has the cleaner title; the rest are unrelated one-off fixes/docs changes.", + "best_pr_reason": "PR #4499 is the strongest canonical candidate only because it has an exact twin (#4500) with the same title, files, and diff intent. Outside that pair, the cluster is too heterogeneous for a meaningful global canonical PR.", "soft_edge_verdicts": [ { - "left": "pull_request:386", - "right": "pull_request:401", + "left": "pull_request:4211", + "right": "pull_request:6198", "accept": false, - "reason": "Both are docs-only PRs, but for different pipelines and different topics (Stable Diffusion vs Karras VE). No shared underlying bug/change." + "reason": "Different fixes: docstring wording for SDXL inpainting vs benchmarking/reporting code. No shared bug or change path." }, { - "left": "pull_request:3942", - "right": "pull_request:5098", + "left": "pull_request:4347", + "right": "pull_request:5275", "accept": false, - "reason": "Both touch tests around numerical behavior, but they address different pipelines and different failure modes. Not the same bug." + "reason": "A scheduler step-index change and adding an SDXL research example are unrelated." }, { - "left": "pull_request:3942", - "right": "pull_request:5276", + "left": "pull_request:4397", + "right": "pull_request:5391", "accept": false, - "reason": "One fixes numpy test behavior; the other fixes a 1D scheduler issue. Different code paths and problems." + "reason": "Both touch docs, but one is an SDXL docstring tweak and the other is broad typehint/docs cleanup across models; not the same change." }, { - "left": "pull_request:3942", - "right": "pull_request:5724", + "left": "pull_request:4499", + "right": "pull_request:5384", "accept": false, - "reason": "Test cleanup versus LoRA test decorator support. Same broad testing area, but not the same underlying change." + "reason": "SDXL reference-attn/adain support is a feature PR; hypernetwork support is a different loader/feature area." }, { - "left": "pull_request:3989", - "right": "pull_request:5699", + "left": "pull_request:4500", + "right": "pull_request:5384", "accept": false, - "reason": "Slow-test fixes for SDXL offload versus unrelated styling cleanup. No shared bug." + "reason": "Same as above: unrelated features in different code paths." }, { - "left": "pull_request:3995", - "right": "pull_request:4058", + "left": "pull_request:4811", + "right": "pull_request:6360", "accept": false, - "reason": "SDXL prompt embeddings implementation versus doc link correction. Different artifacts and concerns." + "reason": "ControlNet inpaint pipeline addition vs adding padding_mask_crop to inpaint pipelines are distinct changes." }, { - "left": "pull_request:3995", - "right": "pull_request:4285", + "left": "pull_request:4856", + "right": "pull_request:6365", "accept": false, - "reason": "Feature addition in SDXL versus ONNX/OpenVINO docs updates. Not the same change." + "reason": "Documentation link fix is unrelated to an MPS scheduler bugfix." }, { - "left": "pull_request:400", - "right": "pull_request:401", + "left": "pull_request:5097", + "right": "pull_request:6221", "accept": false, - "reason": "Score SDE VE docs and Karras VE docs are separate model docs, not one duplicate fix." + "reason": "Float16 test adjustments and slow-test cleanup do not fix the same underlying issue." }, { - "left": "pull_request:4106", - "right": "pull_request:5738", + "left": "pull_request:5164", + "right": "pull_request:7328", "accept": false, - "reason": "Tortoise TTS implementation versus LoRA adapter deletion feature. Completely different code paths." + "reason": "FreeU core mechanism vs CI workflow dependency update are unrelated." }, { - "left": "pull_request:4203", - "right": "pull_request:5369", + "left": "pull_request:5369", + "right": "pull_request:5831", "accept": false, - "reason": "Both are minor docs fixes, but for different files and unrelated typos/examples." + "reason": "Minor docs examples vs a docs title fix are different documentation edits, not one bug." }, { - "left": "pull_request:4211", - "right": "pull_request:6198", + "left": "pull_request:5420", + "right": "pull_request:5741", "accept": false, - "reason": "SDXL inpainting docstring correction versus benchmark reporting fix. Different areas and changes." + "reason": "A follow-up training-doc correction and PEFT unload speedup are not the same code-path problem." }, { - "left": "pull_request:4347", - "right": "pull_request:5275", + "left": "pull_request:5526", + "right": "pull_request:6179", "accept": false, - "reason": "Scheduler step-index work versus adding an SDXL example project. Unrelated." + "reason": "Both affect diffusion pipelines, but one fixes 2nd-order scheduler handling and the other adds fused attention projections; not mergeable as one bug fix." }, { - "left": "pull_request:4397", - "right": "pull_request:5391", + "left": "pull_request:5602", + "right": "pull_request:5982", "accept": false, - "reason": "SDXL docstring default tweak versus broad typehint/doc cleanup in models. Not the same concrete change." + "reason": "Different docs additions for different topics/models." }, { - "left": "pull_request:4499", - "right": "pull_request:4500", - "accept": true, - "reason": "Near-identical PRs with the same SDXL reference_attn/reference_adain support, same files, and same feature change; they look like duplicate submissions." - }, - { - "left": "pull_request:4811", - "right": "pull_request:6360", + "left": "pull_request:5653", + "right": "pull_request:7371", "accept": false, - "reason": "ControlNet inpaint pipeline introduction versus padding_mask_crop support across inpaint pipelines. Different features." + "reason": "Both mention PEFT/LoRA, but one changes maybe_raise_or_warn and the other adds DoRA LoRA support; different underlying changes." }, { - "left": "pull_request:4856", - "right": "pull_request:6365", + "left": "pull_request:5698", + "right": "pull_request:7473", "accept": false, - "reason": "API link fix versus MPS scheduler compatibility fix. No shared bug." + "reason": "LCM img2img fix and missing raise statements in check_inputs are unrelated." }, { - "left": "pull_request:5097", - "right": "pull_request:6221", + "left": "pull_request:6305", + "right": "pull_request:6997", "accept": false, - "reason": "Float16 test adjustments versus assorted slow-test minor fixes. Similar testing theme only, not same issue." + "reason": "DDPM zero-SNR scheduler support and LoRA support for a spatio-temporal UNet are unrelated." }, { - "left": "pull_request:5164", - "right": "pull_request:7328", + "left": "pull_request:6308", + "right": "pull_request:6369", "accept": false, - "reason": "FreeU core mechanism versus PyPI workflow dependency tweak. Completely unrelated." + "reason": "Both are LoRA-related tests/backend work, but one fixes dummy components and the other removes deprecated LoRA APIs; different changes." }, { - "left": "pull_request:5369", - "right": "pull_request:5831", + "left": "pull_request:6498", + "right": "pull_request:7857", "accept": false, - "reason": "Minor docs/example fixes versus a toctree title fix. Different docs changes." + "reason": "Inpaint docs for mask blur/padding mask crop and PixArtSigma docs are unrelated." }, { - "left": "pull_request:5420", - "right": "pull_request:5741", + "left": "pull_request:6564", + "right": "pull_request:8404", "accept": false, - "reason": "Training docs follow-up versus LoRA unload performance work. Different artifacts and goals." + "reason": "IP-Adapter path fix and Latte video model addition are unrelated." }, { - "left": "pull_request:5526", - "right": "pull_request:6179", + "left": "pull_request:6998", + "right": "pull_request:7260", "accept": false, - "reason": "Scheduler-order fix in ControlNet inpaint versus fused attention projection support. Different subsystems and bugs." + "reason": "A docs callout fix is unrelated to SAG IP-Adapter support." }, { - "left": "pull_request:5602", - "right": "pull_request:5982", + "left": "pull_request:7247", + "right": "pull_request:9247", "accept": false, - "reason": "Docs cleanup versus SDXL Turbo docs addition. Not the same underlying change." + "reason": "Single-file loading config tests and model variant identification fixes are different issues." }, { - "left": "pull_request:5653", - "right": "pull_request:7371", + "left": "pull_request:7353", + "right": "pull_request:9426", "accept": false, - "reason": "PEFT helper/wrapper support versus DoRA LoRA loading support. Related ecosystem, but different concrete changes." + "reason": "ControlNet SDXL img2img pipeline update and artifact-uploader CI changes are unrelated." }, { - "left": "pull_request:5698", - "right": "pull_request:7473", + "left": "pull_request:7857", + "right": "pull_request:9242", "accept": false, - "reason": "LCM img2img fix versus missing-raise validation fix. Different bugs." + "reason": "PixArtSigma documentation and xpu testing support do not address the same underlying bug/change." }, { - "left": "pull_request:5741", - "right": "pull_request:7679", + "left": "pull_request:7857", + "right": "pull_request:9718", "accept": false, - "reason": "LoRA unload speedup versus CI workflow install change. Not the same issue." + "reason": "Different docs/community-script changes with no shared concrete bug." }, { - "left": "pull_request:6305", - "right": "pull_request:6997", + "left": "pull_request:8298", + "right": "pull_request:8452", "accept": false, - "reason": "DDPM zero-SNR scheduler argument versus LoRA support for a spatio-temporal UNet. Unrelated changes." + "reason": "CLI env simplification and single-file loading error fix are unrelated." } ] }, @@ -164367,23 +164547,31 @@ "component_index": 1, "component_count": 266, "review_unit_index": 196, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10148", "pull_request:10351", + "pull_request:10360", "pull_request:10567", + "pull_request:10622", "pull_request:10630", + "pull_request:10985", "pull_request:11130", "pull_request:11290", + "pull_request:11582", "pull_request:11738", "pull_request:11979", + "pull_request:11997", "pull_request:12275", "pull_request:12383", + "pull_request:13063", + "pull_request:13216", "pull_request:1327", "pull_request:1513", "pull_request:3252", @@ -164392,48 +164580,31 @@ "pull_request:5057", "pull_request:5073", "pull_request:5470", - "pull_request:6308", - "pull_request:6369", "pull_request:6447", - "pull_request:6498", - "pull_request:6564", "pull_request:6789", "pull_request:6816", "pull_request:6929", - "pull_request:6998", "pull_request:7058", - "pull_request:7247", - "pull_request:7260", - "pull_request:7353", - "pull_request:7857", - "pull_request:8298", - "pull_request:8404", - "pull_request:8452", + "pull_request:7213", + "pull_request:7990", "pull_request:8496", "pull_request:8616", "pull_request:8685", + "pull_request:8777", "pull_request:8964", "pull_request:9060", + "pull_request:9159", "pull_request:9242", - "pull_request:9247", "pull_request:9256", - "pull_request:9426", "pull_request:9649", "pull_request:9696", "pull_request:9709", "pull_request:9718", - "pull_request:9968" + "pull_request:9862", + "pull_request:9968", + "pull_request:9988" ], "soft_pairs": [ - "pull_request:6308|pull_request:6369", - "pull_request:6498|pull_request:7857", - "pull_request:6564|pull_request:8404", - "pull_request:6998|pull_request:7260", - "pull_request:7247|pull_request:9247", - "pull_request:7353|pull_request:9426", - "pull_request:7857|pull_request:9242", - "pull_request:7857|pull_request:9718", - "pull_request:8298|pull_request:8452", "pull_request:8496|pull_request:8964", "pull_request:9060|pull_request:9256", "pull_request:9242|pull_request:9649", @@ -164451,198 +164622,193 @@ "pull_request:1327|pull_request:399", "pull_request:1513|pull_request:3252", "pull_request:4744|pull_request:6929", - "pull_request:6929|pull_request:8685" + "pull_request:6929|pull_request:8685", + "pull_request:7213|pull_request:7990", + "pull_request:11582|pull_request:11997", + "pull_request:9159|pull_request:9988", + "pull_request:13063|pull_request:13216", + "pull_request:10985|pull_request:9862", + "pull_request:10148|pull_request:8777", + "pull_request:10360|pull_request:10622" ], - "prepared_review_unit_hash": "2a7b73fabe2fe0abab2d03612b646646ba9d2257ab27efa2b19a80fb4b798a76", + "prepared_review_unit_hash": "0d9d290ebf2ae127b19fda6ec8468c0d4772611fa8a0cf47cffabdcb5ade77c2", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 27, - "serialized_chars": 98356, - "estimated_input_tokens": 24589, - "estimated_eval_tokens": 49434 + "soft_pair_count": 25, + "serialized_chars": 102955, + "estimated_input_tokens": 25739, + "estimated_eval_tokens": 51734 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is highly heterogeneous: mostly unrelated test, docs, CI, quantization, LoRA, and pipeline changes. I don\u2019t see a single duplicate cluster or a clear canonical artifact shared across the items.", - "confidence": 0.94, + "summary": "This cluster is mostly a set of unrelated Diffusers PRs spanning different pipelines, docs, CI, tests, and utilities. I do not see any true duplicate PR pairs; the soft matches are mostly superficial topic overlap or shared subsystems, not the same concrete bug/change.", + "confidence": 0.9, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: the items span unrelated bugs/features in different subsystems, so any one PR would be arbitrary as a cluster representative.", + "canonical_pr_reason": "No PR is a clear duplicate anchor because the cluster does not center on one underlying change; the items are heterogeneous and the soft edges are not strong enough to merge them.", "best_issue_reason": null, - "best_pr_reason": "PR 11130 is the strongest standalone PR in the set because it has a substantial code change plus tests/docs and a clear feature scope; the other PRs are mostly narrower fixes, docs, or CI updates.", + "best_pr_reason": "PR 11130 is the strongest representative only because it is a substantial, well-scoped merged change with broad impact and review activity. It is not a duplicate canonical in the strict sense, just the best single representative among unrelated PRs.", "soft_edge_verdicts": [ - { - "left": "pull_request:6308", - "right": "pull_request:6369", - "accept": false, - "reason": "Both are LoRA-related tests, but one fixes PEFT dummy components and the other removes deprecated old-backend APIs; different bugs/changes." - }, - { - "left": "pull_request:6498", - "right": "pull_request:7857", - "accept": false, - "reason": "Both are docs around pipelines, but one is about inpaint mask blur/padding docs and the other adds PixArtSigma docs; unrelated changes." - }, - { - "left": "pull_request:6564", - "right": "pull_request:8404", - "accept": false, - "reason": "IP-Adapter path handling vs. a new Latte video model/pipeline; completely different code paths and goals." - }, - { - "left": "pull_request:6998", - "right": "pull_request:7260", - "accept": false, - "reason": "Both touch adapter-related docs/logic, but one fixes a callout and the other adds IP-Adapter support for SAG; not the same bug." - }, - { - "left": "pull_request:7247", - "right": "pull_request:9247", - "accept": false, - "reason": "One fixes config tuple/list handling in single-file utils, the other fixes variant discovery in pipeline loading; different issues." - }, - { - "left": "pull_request:7353", - "right": "pull_request:9426", - "accept": false, - "reason": "One adds FromSingleFileMixin to an SDXL ControlNet pipeline, the other updates artifact uploader versions in CI; unrelated." - }, - { - "left": "pull_request:7857", - "right": "pull_request:9242", - "accept": false, - "reason": "PixArtSigma docs/model addition vs XPU test backend support; no shared underlying bug or change." - }, - { - "left": "pull_request:7857", - "right": "pull_request:9718", - "accept": false, - "reason": "PixArtSigma docs/model addition vs prompt scheduling callback docs; both are docs/features, but not the same change." - }, - { - "left": "pull_request:8298", - "right": "pull_request:8452", - "accept": false, - "reason": "Different targets: CLI env output simplification vs single-file loading error fix." - }, { "left": "pull_request:8496", "right": "pull_request:8964", "accept": false, - "reason": "Transformer SD3 warning-log gating vs AudioLDM2 cache-position fix; unrelated code paths." + "reason": "Different fixes in different code paths: SD3 warning-log behavior vs AudioLDM2 cache-position handling." }, { "left": "pull_request:9060", "right": "pull_request:9256", "accept": false, - "reason": "One removes a debug print in FLUX, the other fixes an image type check in Kandinsky2.2; unrelated bugs." + "reason": "Unrelated changes: removing a debug print in Flux vs fixing an image type check in Kandinsky2.2." }, { "left": "pull_request:9242", "right": "pull_request:9649", "accept": false, - "reason": "XPU test support vs scheduler return-tuple changes; unrelated areas and purposes." + "reason": "Different concerns: xpu test backend support vs adding pred_original_sample to scheduler tuple returns." }, { "left": "pull_request:9649", "right": "pull_request:9718", "accept": false, - "reason": "Scheduler API return-shape fix vs community README docs update; not the same underlying change." + "reason": "One is scheduler return behavior; the other is a docs update for a community-script callback. Not the same change." }, { "left": "pull_request:9696", "right": "pull_request:9968", "accept": false, - "reason": "Run-tests workflow inputs vs controlnet module refactor; unrelated CI/workflow and model API changes." + "reason": "CI workflow input validation vs ControlNet module refactor; no shared underlying bug." }, { "left": "pull_request:5057", "right": "pull_request:5073", "accept": false, - "reason": "Both are SDXL-related, but one adds clip-skip support and the other fixes multi-batch prompt embed handling; distinct functional changes." + "reason": "Both touch SDXL, but one adds clip skip support and the other fixes multi-batch prompt embeds. Different concrete problems." }, { "left": "pull_request:10567", "right": "pull_request:11130", "accept": false, - "reason": "Hunyuan Video true-CFG feature work vs pipeline-level quantization config; separate subsystems and changes." + "reason": "HunyuanVideo true-CFG feature work vs pipeline-level quantization config. Different subsystems and goals." }, { "left": "pull_request:11738", "right": "pull_request:11979", "accept": false, - "reason": "Deterministic bnb test setup vs torch version gating for compile tests; both test-related but different problems." + "reason": "Test determinism for bnb quantization vs torch-version gating for compile tests. Not the same bug or change." }, { "left": "pull_request:5470", "right": "pull_request:6789", "accept": false, - "reason": "ControlNet inpaint pipeline fixes/docs vs PixArt 256px inference fixes; unrelated pipelines and bugs." + "reason": "ControlNet inpaint doc/fix work and PixArt 256px inference conversion are unrelated." }, { "left": "pull_request:6447", "right": "pull_request:8616", "accept": false, - "reason": "K-Diffusion pipeline addition vs LoRA fusion support in SD3; different features and code paths." + "reason": "K-Diffusion pipeline addition vs LoRA fusion support in SD3. Different features and code paths." }, { "left": "pull_request:6816", "right": "pull_request:7058", "accept": false, - "reason": "SDXL training script DDP/PEFT fix vs LoRA state-dict copy fix; both LoRA-adjacent, but different concrete bugs." + "reason": "SDXL LoRA training/DDP fix vs copying LoRA state dicts before mutation. Related area, but not the same concrete bug." }, { "left": "pull_request:12275", "right": "pull_request:12383", "accept": false, - "reason": "TorchAO config class support vs modular single-file support; quantization vs modular pipeline features." + "reason": "TorchAO config support vs modular single-file support. Different feature work." }, { "left": "pull_request:10351", "right": "pull_request:10630", "accept": false, - "reason": "Two separate test fixes: LoRA NaN test input handling vs flax expected output slices." + "reason": "LoRA test adjustment vs flax test slice update. Both are test-only changes, but not duplicates." }, { "left": "pull_request:11290", "right": "pull_request:9709", "accept": false, - "reason": "Dockerfile torch version unpinning vs setup.py torch max-version pin; opposite dependency-management changes in different files." + "reason": "One unpins torch in an ONNX Dockerfile, the other caps torch for CI. Opposite policy in different contexts." }, { "left": "pull_request:1327", "right": "pull_request:399", "accept": false, - "reason": "Both touch pipeline loading, but one enables local custom pipeline files and the other is an ONNX exporter/pipeline implementation; not the same bug." + "reason": "Custom local pipeline loading vs ONNX Stable Diffusion exporter/pipeline. No shared underlying change." }, { "left": "pull_request:1513", "right": "pull_request:3252", "accept": false, - "reason": "Dreambooth docs addition vs fp16 docs about stateful offload hooks; unrelated documentation updates." + "reason": "Accelerate config docs vs stateful offload notes. Different documentation topics." }, { "left": "pull_request:4744", "right": "pull_request:6929", "accept": false, - "reason": "AudioLDM docstring correction vs community pipeline loading docs; different documentation targets and issues." + "reason": "AudioLDM docstring fix vs community pipeline docs. Both docs, but not the same issue." }, { "left": "pull_request:6929", "right": "pull_request:8685", "accept": false, - "reason": "Community pipeline docs/loading changes vs SD3 float8 note; unrelated docs for different features." + "reason": "Community pipeline loading docs vs SD3 float8 note. Different documentation updates." + }, + { + "left": "pull_request:7213", + "right": "pull_request:7990", + "accept": false, + "reason": "Merge-LoRAs docs vs scheduler-features docs. Same broad docs area, but unrelated content." + }, + { + "left": "pull_request:11582", + "right": "pull_request:11997", + "accept": false, + "reason": "Both are Wan-related, but one adds Wan VACE support and the other refactors Wan modular I2V/FLF2V support. Not mergeable as one duplicate PR." + }, + { + "left": "pull_request:9159", + "right": "pull_request:9988", + "accept": false, + "reason": "Flux sharding support vs Flux Redux conversion/pipeline work. Different code paths and goals." + }, + { + "left": "pull_request:13063", + "right": "pull_request:13216", + "accept": false, + "reason": "Different modular pipeline families: Wan refactor vs Helios addition." + }, + { + "left": "pull_request:10985", + "right": "pull_request:9862", + "accept": false, + "reason": "ComfyUI LoRA compatibility improvements vs saving LoRA adapters. Same area, but distinct functionality and bug scope." + }, + { + "left": "pull_request:10148", + "right": "pull_request:8777", + "accept": false, + "reason": "OmniGen feature addition vs dummy attention processor placeholders. Unrelated changes." + }, + { + "left": "pull_request:10360", + "right": "pull_request:10622", + "accept": false, + "reason": "Release TestPyPI install fix vs general uv installation docs. Different workflows and fixes." } ] }, @@ -164658,7 +164824,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 197, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -164666,16 +164832,10 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10148", - "pull_request:10360", "pull_request:1038", - "pull_request:10622", "pull_request:10838", - "pull_request:10985", "pull_request:11254", - "pull_request:11582", "pull_request:11894", - "pull_request:11997", "pull_request:12005", "pull_request:12425", "pull_request:12438", @@ -164687,11 +164847,9 @@ "pull_request:12960", "pull_request:12973", "pull_request:130", - "pull_request:13063", "pull_request:1309", "pull_request:13100", "pull_request:13120", - "pull_request:13216", "pull_request:13364", "pull_request:1525", "pull_request:1554", @@ -164699,29 +164857,31 @@ "pull_request:1920", "pull_request:2001", "pull_request:2274", + "pull_request:2325", "pull_request:2481", + "pull_request:2584", "pull_request:2793", + "pull_request:2861", + "pull_request:3127", "pull_request:3275", "pull_request:3410", "pull_request:3506", + "pull_request:3607", + "pull_request:3681", + "pull_request:3691", + "pull_request:4034", + "pull_request:4262", + "pull_request:4276", + "pull_request:4315", + "pull_request:4498", + "pull_request:4655", + "pull_request:5695", + "pull_request:5747", "pull_request:630", - "pull_request:7213", - "pull_request:7990", - "pull_request:8777", - "pull_request:9159", "pull_request:9838", - "pull_request:9862", - "pull_request:9875", - "pull_request:9988" + "pull_request:9875" ], "soft_pairs": [ - "pull_request:7213|pull_request:7990", - "pull_request:11582|pull_request:11997", - "pull_request:9159|pull_request:9988", - "pull_request:13063|pull_request:13216", - "pull_request:10985|pull_request:9862", - "pull_request:10148|pull_request:8777", - "pull_request:10360|pull_request:10622", "pull_request:1038|pull_request:630", "pull_request:10838|pull_request:9875", "pull_request:11254|pull_request:9838", @@ -164739,186 +164899,200 @@ "pull_request:1920|pull_request:3275", "pull_request:2001|pull_request:2274", "pull_request:2001|pull_request:2481", - "pull_request:2001|pull_request:3410" + "pull_request:2001|pull_request:3410", + "pull_request:2325|pull_request:4498", + "pull_request:2584|pull_request:3127", + "pull_request:2584|pull_request:4276", + "pull_request:2861|pull_request:3681", + "pull_request:3607|pull_request:5695", + "pull_request:3691|pull_request:4655", + "pull_request:4034|pull_request:5747", + "pull_request:4262|pull_request:4315" ], - "prepared_review_unit_hash": "c2b6fc5f7b3af08c2eaea03593210dd6ac7bca7fc9ebcb50a6f21dbf2ebc3d09", + "prepared_review_unit_hash": "3b3063609fee916416360b91f1980be35d8deda5e5e93b095a8ac460b5dd7304", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 25, - "serialized_chars": 99457, - "estimated_input_tokens": 24865, - "estimated_eval_tokens": 49986 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 96696, + "estimated_input_tokens": 24174, + "estimated_eval_tokens": 48604 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: it mixes unrelated PRs spanning docs, model support, converters, training examples, and small bug fixes. The soft links appear to be driven by superficial filename/subsystem similarity rather than the same underlying change.", - "confidence": 0.98, + "summary": "The candidate pairs are largely unrelated changes across different subsystems, versions, and goals; I don\u2019t see any pair that clearly fixes the same concrete bug or implements the same change.", + "confidence": 0.96, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:7213", - "right": "pull_request:7990", - "accept": false, - "reason": "Both are docs PRs, but they cover different topics: LoRA inference docs vs scheduler features. No shared bug or change." - }, - { - "left": "pull_request:11582", - "right": "pull_request:11997", - "accept": false, - "reason": "Both touch Wan-related code, but one adds Wan VACE while the other adds Wan I2V/FLF2V modular support. Different features and code paths." - }, - { - "left": "pull_request:9159", - "right": "pull_request:9988", - "accept": false, - "reason": "Both involve Flux transformer code, but one is sharding support and the other is Flux Redux conversion/model wiring. Not the same fix." - }, - { - "left": "pull_request:13063", - "right": "pull_request:13216", - "accept": false, - "reason": "Both are modular-pipeline refactors, but they introduce different model families/tasks (Wan refactor vs Helios support). Separate changes." - }, - { - "left": "pull_request:10985", - "right": "pull_request:9862", - "accept": false, - "reason": "Both concern LoRA/PEFT internals, but one expands ComfyUI LoRA compatibility and the other adds save_lora_adapter. Different functionality." - }, - { - "left": "pull_request:10148", - "right": "pull_request:8777", - "accept": false, - "reason": "OmniGen support vs dummy LoRA attention processor shims are unrelated." - }, - { - "left": "pull_request:10360", - "right": "pull_request:10622", - "accept": false, - "reason": "One fixes release workflow TestPyPI installation; the other updates installation docs. No common bug." - }, { "left": "pull_request:1038", "right": "pull_request:630", "accept": false, - "reason": "Flax dtype fix and Flax from_pretrained key cleanup are unrelated changes." + "reason": "One fixes a Flax training dtype mismatch; the other cleans up Flax checkpoint key handling. Different bugs, different code paths." }, { "left": "pull_request:10838", "right": "pull_request:9875", "accept": false, - "reason": "Both add new model/pipeline support, but for different models (FlashVideo vs Meissonic). Not mergeable into one PR." + "reason": "FlashVideo pipeline addition and Meissonic support are separate model/pipeline features, not the same change." }, { "left": "pull_request:11254", "right": "pull_request:9838", "accept": false, - "reason": "ConsisID import fixes and Flux training memory/bug fixes are unrelated." + "reason": "ConsisID import dependency fix is unrelated to Flux LoRA training behavior." }, { "left": "pull_request:11894", "right": "pull_request:12911", "accept": false, - "reason": "k-diffusion pinning for CI and adding discrete diffusion docs/examples are completely different changes." + "reason": "Pinning k-diffusion for CI is unrelated to adding discrete diffusion pipelines and docs." }, { "left": "pull_request:12005", "right": "pull_request:12438", "accept": false, - "reason": "Both alter loading behavior, but one fixes sharded checkpoint local_files_only handling and the other handles transformers offload_state_dict. Different bugs." + "reason": "Both touch loading utilities, but one handles local_files_only for sharded checkpoints while the other drops offload_state_dict for newer transformers models." }, { "left": "pull_request:12425", "right": "pull_request:12960", "accept": false, - "reason": "Torch.distributed import guarding and Flux2 FSDP detection are unrelated small fixes." + "reason": "Torch distributed import guarding is unrelated to how is_fsdp is computed in Flux2 training scripts." }, { "left": "pull_request:12586", "right": "pull_request:13100", "accept": false, - "reason": "Docs tree sorting vs ModularPipeline docs updates are both documentation-only but not the same change." + "reason": "Doc ordering change vs modular pipeline documentation updates; not the same underlying issue." }, { "left": "pull_request:12714", "right": "pull_request:13364", "accept": false, - "reason": "Flux2 LoRA NaN test changes and Wan MotionConv2d dtype casting fix address different components and failures." + "reason": "LoRA NaN test adjustments are unrelated to Wan Animate blur-kernel dtype casting." }, { "left": "pull_request:12863", "right": "pull_request:12914", "accept": false, - "reason": "Cosmos conversion script/model support vs a typo fix in a pipeline docstring are not the same underlying issue." + "reason": "Cosmos conversion support is a separate feature from correcting a revision typo in pipeline docs." }, { "left": "pull_request:12973", "right": "pull_request:13120", "accept": false, - "reason": "GLM-Image support and modular loading robustness tests are unrelated." + "reason": "GLM-Image model support and modular loading tests are unrelated changes." }, { "left": "pull_request:130", "right": "pull_request:1309", "accept": false, - "reason": "Accelerate config example docs and PIL handling/docs cleanup are unrelated docs PRs." + "reason": "Example docs onboarding step and PIL handling improvements have no shared bug or change target." }, { "left": "pull_request:1525", "right": "pull_request:3410", "accept": false, - "reason": "Upscaling batch-size fix and DeepSpeed plugin retrieval fix are different training bugs." + "reason": "Upscaling batch-size fix and DeepSpeed state access fix are different training bugs." }, { "left": "pull_request:1554", "right": "pull_request:2793", "accept": false, - "reason": "Docs correctness for UNet shapes and removing cuDNN benchmark advice are unrelated documentation edits." + "reason": "One corrects API docs, the other removes a cuDNN benchmark suggestion from optimization docs." }, { "left": "pull_request:1878", "right": "pull_request:3506", "accept": false, - "reason": "One is a checkpoint parsing guard; the other adjusts checkpoint resume step scaling. Different code paths." + "reason": "Checkpoint format handling and global-step restoration are separate training/checkpointing issues." }, { "left": "pull_request:1920", "right": "pull_request:3275", "accept": false, - "reason": "TRANSLATING docs addition and adding upsample_size to Attn blocks are unrelated." + "reason": "Documentation translation file addition and UNet upsample_size support are unrelated." }, { "left": "pull_request:2001", "right": "pull_request:2274", "accept": false, - "reason": "ColossalAI typo fix and transform naming clash in examples are unrelated example fixes." + "reason": "ColossalAI typo fixes and a transforms naming clash are different example changes." }, { "left": "pull_request:2001", "right": "pull_request:2481", "accept": false, - "reason": "ColossalAI typo fix and unconditional training logger fix are unrelated." + "reason": "ColossalAI typo fixes are unrelated to the unconditional training test fix." }, { "left": "pull_request:2001", "right": "pull_request:3410", "accept": false, - "reason": "ColossalAI typo fix and DeepSpeed plugin retrieval fix are unrelated." + "reason": "No shared code path or bug: ColossalAI typo fix vs accelerator state retrieval." + }, + { + "left": "pull_request:2325", + "right": "pull_request:4498", + "accept": false, + "reason": "ONNX VAE conversion support and SDXL reference pipeline support are separate features." + }, + { + "left": "pull_request:2584", + "right": "pull_request:3127", + "accept": false, + "reason": "Community ControlNet pipeline addition is unrelated to the img2img safety-checker fix." + }, + { + "left": "pull_request:2584", + "right": "pull_request:4276", + "accept": false, + "reason": "Pipeline addition and pipeline import tests are not the same concrete change." + }, + { + "left": "pull_request:2861", + "right": "pull_request:3681", + "accept": false, + "reason": "StableUnCLIP None-case test and checkpoint-limit handling are unrelated." + }, + { + "left": "pull_request:3607", + "right": "pull_request:5695", + "accept": false, + "reason": "Temb attention fix and PixArt masking condition fix address different model behaviors." + }, + { + "left": "pull_request:3691", + "right": "pull_request:4655", + "accept": false, + "reason": "LoRA processor selection for PyTorch 2.0 is unrelated to adding data_dir to dataset loading." + }, + { + "left": "pull_request:4034", + "right": "pull_request:5747", + "accept": false, + "reason": "LoRA unload utility and dataset data_dir plumbing are unrelated." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4315", + "accept": false, + "reason": "SDXL DreamBooth LoRA training fixes do not match the torch.compile graph-break fix in LoRA layers; different concrete problems." } ] }, @@ -164934,38 +165108,37 @@ "component_index": 1, "component_count": 266, "review_unit_index": 198, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:11261", + "pull_request:11268", + "pull_request:11582", + "pull_request:11894", + "pull_request:11979", "pull_request:12215", "pull_request:12220", - "pull_request:2325", - "pull_request:2584", - "pull_request:2861", - "pull_request:3127", - "pull_request:3607", - "pull_request:3681", - "pull_request:3691", - "pull_request:4034", - "pull_request:4262", - "pull_request:4276", - "pull_request:4315", + "pull_request:12324", + "pull_request:12830", + "pull_request:1369", + "pull_request:3846", "pull_request:4343", "pull_request:4395", - "pull_request:4498", "pull_request:4629", "pull_request:4655", + "pull_request:4714", "pull_request:4947", "pull_request:5024", "pull_request:5508", - "pull_request:5695", - "pull_request:5747", + "pull_request:5668", "pull_request:5788", + "pull_request:5833", + "pull_request:5834", "pull_request:6213", "pull_request:6254", "pull_request:6346", @@ -164982,23 +165155,17 @@ "pull_request:7650", "pull_request:7797", "pull_request:7803", + "pull_request:819", "pull_request:8378", "pull_request:8422", "pull_request:8702", "pull_request:8981", "pull_request:9569", "pull_request:9662", + "pull_request:9875", "pull_request:9918" ], "soft_pairs": [ - "pull_request:2325|pull_request:4498", - "pull_request:2584|pull_request:3127", - "pull_request:2584|pull_request:4276", - "pull_request:2861|pull_request:3681", - "pull_request:3607|pull_request:5695", - "pull_request:3691|pull_request:4655", - "pull_request:4034|pull_request:5747", - "pull_request:4262|pull_request:4315", "pull_request:4343|pull_request:5508", "pull_request:4395|pull_request:4629", "pull_request:4655|pull_request:6511", @@ -165016,192 +165183,200 @@ "pull_request:7803|pull_request:9662", "pull_request:8422|pull_request:9918", "pull_request:8702|pull_request:8981", - "pull_request:12215|pull_request:12220" + "pull_request:12215|pull_request:12220", + "pull_request:11582|pull_request:12324", + "pull_request:11261|pull_request:9875", + "pull_request:11268|pull_request:11894", + "pull_request:11979|pull_request:12830", + "pull_request:1369|pull_request:819", + "pull_request:3846|pull_request:5833", + "pull_request:3846|pull_request:5834", + "pull_request:4714|pull_request:5668" ], - "prepared_review_unit_hash": "328a27c5d4a3761cf00f36bfd4d4819c9f6d00ce024528049a52914317bc526a", + "prepared_review_unit_hash": "416940ba83e124cee87f341a41c862cce9ff53b596ae61c3b2f2c712153b5070", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 97749, - "estimated_input_tokens": 24438, - "estimated_eval_tokens": 49132 + "serialized_chars": 98032, + "estimated_input_tokens": 24508, + "estimated_eval_tokens": 49272 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: most PRs are unrelated training, scheduler, docs, or test tweaks, and the suggested soft edges do not look like the same underlying change. The only loosely related pair is the Qwen-Image work, but even there the PRs target different code paths (ControlNet support vs modular pipeline plumbing).", - "confidence": 0.21, + "summary": "This cluster is heterogeneous: most pairs are clearly unrelated PRs that just share a subsystem or file pattern. I found no true duplicate issue/PR pair; the strongest representative PR is the broad Qwen modular work (#12220), but it is not a duplicate of the other Qwen/Wan items.", + "confidence": 0.31, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR fits the cluster well; most items are unrelated, so any choice would be arbitrary. If forced, the broadest and most central-looking artifact is the Qwen modular pipeline PR because it introduces the core Qwen pipeline scaffolding and related exports/docs.", + "canonical_pr_reason": "No single underlying change dominates the cluster; if one PR must stand in as the representative artifact, #12220 is the most substantial and central merged PR in the set.", "best_issue_reason": null, - "best_pr_reason": "PR 12220 is the best representative only in a very loose sense: it is the most substantial and central Qwen-related change in the set, touching the modular pipeline implementation and integration points rather than a narrow follow-up.", + "best_pr_reason": "#12220 has the broadest scope and highest review signal, so it is the best standalone representative of this mixed PR cluster, though it should not be treated as a duplicate target.", "soft_edge_verdicts": [ - { - "left": "pull_request:2325", - "right": "pull_request:4498", - "accept": false, - "reason": "Different features in different eras: ONNX VAE converter vs SDXL reference-attention community pipeline." - }, - { - "left": "pull_request:2584", - "right": "pull_request:3127", - "accept": false, - "reason": "Both are ControlNet-related, but one adds a community img2img pipeline while the other fixes safety-checker image handling in core stable diffusion img2img." - }, - { - "left": "pull_request:2584", - "right": "pull_request:4276", - "accept": false, - "reason": "A new community ControlNet pipeline and a dependency-import test are unrelated changes." - }, - { - "left": "pull_request:2861", - "right": "pull_request:3681", - "accept": false, - "reason": "StableUnCLIP test for `image_embeds=None` is unrelated to checkpoint-limit handling in training scripts." - }, - { - "left": "pull_request:3607", - "right": "pull_request:5695", - "accept": false, - "reason": "Different problems: temb attention graph behavior vs PixArt mask-feature condition logic." - }, - { - "left": "pull_request:3691", - "right": "pull_request:4655", - "accept": false, - "reason": "LoRA processor selection for text-encoder loading is unrelated to passing `data_dir` into dataset loading." - }, - { - "left": "pull_request:4034", - "right": "pull_request:5747", - "accept": false, - "reason": "LoRA unload utility docs/code vs dataset `data_dir` plumbing in training scripts." - }, - { - "left": "pull_request:4262", - "right": "pull_request:4315", - "accept": false, - "reason": "DreamBooth LoRA training fixes are not the same code-path problem as the torch.compile graph-break fix in LoRA conv forward." - }, { "left": "pull_request:4343", "right": "pull_request:5508", "accept": false, - "reason": "Both touch `train_dreambooth_lora_sdxl.py`, but one adds a rank arg while the other is a broader feature/fix set; not the same concrete change." + "reason": "Both touch DreamBooth LoRA SDXL, but one adds a rank argument while the other adds several training features; different changes, not the same bug/fix." }, { "left": "pull_request:4395", "right": "pull_request:4629", "accept": false, - "reason": "DreamBooth class-prompt preprocessing fix vs SDXL long weighted prompt pipeline docs/example additions." + "reason": "4395 fixes pre-encoding class prompts in DreamBooth; 4629 is an unrelated SDXL long weighted prompt/push-to-hub cleanup effort." }, { "left": "pull_request:4655", "right": "pull_request:6511", "accept": false, - "reason": "Dataset loading `data_dir` fix and torch.compile checkpointing compatibility are different issues." + "reason": "One passes data_dir to load_dataset; the other makes checkpointing compatible with torch.compile. Different code paths and goals." }, { "left": "pull_request:4947", "right": "pull_request:5024", "accept": false, - "reason": "T2I-Adapter training arg/model init changes are unrelated to the DPMSolver-SDE step-index fix." + "reason": "Training-script tests for T2I-Adapter versus scheduler step-index logic; no shared underlying bug." }, { "left": "pull_request:5788", "right": "pull_request:6254", "accept": false, - "reason": "LCM schedule validation error vs torch.fx-friendly randn implementation are unrelated scheduler/utils changes." + "reason": "LCM error-message guard and randn_like/tracing behavior are unrelated changes." }, { "left": "pull_request:6213", "right": "pull_request:6346", "accept": false, - "reason": "Text-to-video pipeline cleanup is unrelated to LoRA state-dict conversion in training examples." + "reason": "Text-to-video pipeline cleanup and LoRA state-dict conversion follow-up are different changes." }, { "left": "pull_request:6759", "right": "pull_request:7096", "accept": false, - "reason": "InstantID prompt-image embedding repeat fix vs ControlNet training-script enhancements are different code paths." + "reason": "InstantID batch embedding fix is unrelated to ControlNet training-script enhancements." }, { "left": "pull_request:6764", "right": "pull_request:8981", "accept": false, - "reason": "A PEFT inference docs note is not the same change as adding `LoraLoaderMixin` exports and docs across the library." + "reason": "Documentation about saving after fusing adapters is not the same change as adding LoraLoaderMixin to inits." }, { "left": "pull_request:7062", "right": "pull_request:8378", "accept": false, - "reason": "Imagic docstring cleanup and AsymmetricAutoencoderKL forward fix address unrelated parts of the library." + "reason": "A community pipeline docstring fix and an asymmetric autoencoder forward fix do not share the same underlying bug." }, { "left": "pull_request:7103", "right": "pull_request:8378", "accept": false, - "reason": "ControlNet image validation fix is unrelated to the AsymmetricAutoencoderKL forward bug." + "reason": "ControlNet input validation fix and autoencoder forward fix are separate code-path issues." }, { "left": "pull_request:7134", "right": "pull_request:7295", "accept": false, - "reason": "UNet type-hint cleanup and single-file loader residue cleanup are different chores in different modules." + "reason": "UNet type-hint cleanup is unrelated to single-file loader residue cleanup." }, { "left": "pull_request:7200", "right": "pull_request:7650", "accept": false, - "reason": "Transformer2D norm/patch-input validation is unrelated to adding the `is_cosxl_edit` argument in SDXL InstructPix2Pix." + "reason": "Transformer norm-type validation and SDXL InstructPix2Pix `is_cosxl_edit` are distinct changes." }, { "left": "pull_request:7627", "right": "pull_request:9569", "accept": false, - "reason": "Playground img2img latent handling and CogVideoX autoencoder decode batching are different model paths." + "reason": "Playground img2img latent scaling for SDXL is unrelated to CogVideoX autoencoder decode batching." }, { "left": "pull_request:7797", "right": "pull_request:9662", "accept": false, - "reason": "DDPM test size reduction and LoRA/transformers version gate update are unrelated test maintenance changes." + "reason": "DDPM fast-test size reduction and a transformers-version test bump are separate CI/test adjustments." }, { "left": "pull_request:7803", "right": "pull_request:9662", "accept": false, - "reason": "DDIM test size reduction and LoRA dependency version bump are unrelated." + "reason": "DDIM fast-test size reduction is unrelated to the transformers version gating change." }, { "left": "pull_request:8422", "right": "pull_request:9918", "accept": false, - "reason": "Subfolder-loading test suite and server example/docs are not the same underlying bug or feature." + "reason": "A subfolder-loading test suite and a server example are different additions, not duplicate fixes." }, { "left": "pull_request:8702", "right": "pull_request:8981", "accept": false, - "reason": "SD3 conversion script and LoraLoaderMixin export/docs are separate functionality additions." + "reason": "SD3 conversion script creation is unrelated to the LoRA loader init/docs refactor." }, { "left": "pull_request:12215", "right": "pull_request:12220", "accept": false, - "reason": "Both are Qwen-Image related, but one adds ControlNet support while the other adds modular pipeline infrastructure; different code paths and merge surfaces." + "reason": "Both are Qwen-Image related, but one adds ControlNet support while the other builds a modular Qwen pipeline; different concrete features." + }, + { + "left": "pull_request:11582", + "right": "pull_request:12324", + "accept": false, + "reason": "Wan VACE initial support and Wan2.2 VACE-Fun conversion updates are related follow-ups, but not the same underlying PR/change." + }, + { + "left": "pull_request:11261", + "right": "pull_request:9875", + "accept": false, + "reason": "VidTok autoencoders and Meissonic support are unrelated feature PRs." + }, + { + "left": "pull_request:11268", + "right": "pull_request:11894", + "accept": false, + "reason": "Relaxing UnCLIP test tolerance and pinning k-diffusion for CI are unrelated test/packaging fixes." + }, + { + "left": "pull_request:11979", + "right": "pull_request:12830", + "accept": false, + "reason": "Torch version gating in compilation tests is unrelated to attention-backend test enhancements." + }, + { + "left": "pull_request:1369", + "right": "pull_request:819", + "accept": false, + "reason": "Flax sinusoidal embedding behavior and DDIM Windows int64 timesteps are unrelated bugs." + }, + { + "left": "pull_request:3846", + "right": "pull_request:5833", + "accept": false, + "reason": "Random module seeding and a docs-only PixArt change do not address the same code-path." + }, + { + "left": "pull_request:3846", + "right": "pull_request:5834", + "accept": false, + "reason": "Random module seeding and a ControlNet single-file import fix are unrelated." + }, + { + "left": "pull_request:4714", + "right": "pull_request:5668", + "accept": false, + "reason": "Both concern missing-Transformers behavior, but one fixes a dummy-object error path while the other broadly enables diffusers/DDPM without Transformers; not the same concrete fix." } ] }, @@ -165217,11 +165392,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 199, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -165234,54 +165409,47 @@ "pull_request:10497", "pull_request:10633", "pull_request:10730", + "pull_request:10894", "pull_request:10898", "pull_request:11004", "pull_request:11095", + "pull_request:11243", "pull_request:11261", - "pull_request:11268", "pull_request:1129", "pull_request:11556", - "pull_request:11582", - "pull_request:11894", "pull_request:11979", + "pull_request:11994", + "pull_request:11999", "pull_request:12117", "pull_request:12119", + "pull_request:12288", "pull_request:12301", - "pull_request:12324", "pull_request:12416", "pull_request:12439", "pull_request:12445", "pull_request:12611", - "pull_request:12830", "pull_request:13216", - "pull_request:1369", + "pull_request:1496", + "pull_request:1920", + "pull_request:1942", "pull_request:1966", "pull_request:2040", - "pull_request:3846", + "pull_request:2347", + "pull_request:2467", + "pull_request:2516", + "pull_request:3850", "pull_request:4098", - "pull_request:4714", - "pull_request:5668", - "pull_request:5833", - "pull_request:5834", "pull_request:6003", "pull_request:6841", "pull_request:7158", - "pull_request:819", + "pull_request:750", "pull_request:8229", "pull_request:8981", "pull_request:9175", "pull_request:9257", - "pull_request:9875" + "pull_request:9709" ], "soft_pairs": [ - "pull_request:11582|pull_request:12324", - "pull_request:11261|pull_request:9875", - "pull_request:11268|pull_request:11894", - "pull_request:11979|pull_request:12830", - "pull_request:1369|pull_request:819", - "pull_request:3846|pull_request:5833", - "pull_request:3846|pull_request:5834", - "pull_request:4714|pull_request:5668", "pull_request:6841|pull_request:8981", "pull_request:7158|pull_request:8229", "pull_request:4098|pull_request:6003", @@ -165297,180 +165465,182 @@ "pull_request:10230|pull_request:11004", "pull_request:10494|pull_request:10497", "pull_request:10633|pull_request:10898", - "pull_request:10730|pull_request:9257" + "pull_request:10730|pull_request:9257", + "pull_request:10894|pull_request:9709", + "pull_request:11243|pull_request:11999", + "pull_request:11261|pull_request:12288", + "pull_request:11979|pull_request:11994", + "pull_request:1496|pull_request:2516", + "pull_request:1920|pull_request:750", + "pull_request:1942|pull_request:3850", + "pull_request:2347|pull_request:2467" ], - "prepared_review_unit_hash": "3d90d9b3c856bf02c2c8b0a3456ddd4a0ea3b3ae6cd1effc544e93c7ba678c01", + "prepared_review_unit_hash": "ba86b03b4d550b24a09574bf8dfb868e6e97caf6eabed86b3ef636261cbe1715", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 24, - "serialized_chars": 97839, - "estimated_input_tokens": 24460, - "estimated_eval_tokens": 49176 + "serialized_chars": 103827, + "estimated_input_tokens": 25957, + "estimated_eval_tokens": 52170 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: docs tweaks, CI/test adjustments, and unrelated feature additions. None of the soft pairs are true duplicates; most only share a broad subsystem or filename pattern.", - "confidence": 0.98, + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs, CI, scheduler, LoRA, and pipeline PRs across many different code paths. There is no single canonical artifact that cleanly represents the cluster.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR cleanly represents the cluster because the items do not share one underlying bug or change; they span unrelated docs, CI, model, and pipeline work.", + "canonical_pr_reason": "No canonical PR: the items are heterogeneous and do not share one underlying bug or change.", "best_issue_reason": null, - "best_pr_reason": "No PR is a strong global representative here; the cluster lacks a shared concrete code-path problem, so any choice would be arbitrary.", + "best_pr_reason": "No best PR: none of the PRs is representative of the whole cluster, and selecting one would be arbitrary.", "soft_edge_verdicts": [ - { - "left": "pull_request:11582", - "right": "pull_request:12324", - "accept": false, - "reason": "Both touch Wan VACE conversion, but one adds Wan VACE and the other adds Wan2.2 VACE-Fun; different models and configs, not the same change." - }, - { - "left": "pull_request:11261", - "right": "pull_request:9875", - "accept": false, - "reason": "Both add model support, but VidTok autoencoders and Meissonic transformer/pipeline support are unrelated feature work." - }, - { - "left": "pull_request:11268", - "right": "pull_request:11894", - "accept": false, - "reason": "One relaxes an unclip test tolerance; the other pins k-diffusion. Different issues and code paths." - }, - { - "left": "pull_request:11979", - "right": "pull_request:12830", - "accept": false, - "reason": "One adds version guards to compile tests; the other changes attention-backend test coverage. Shared test area only, not the same bug." - }, - { - "left": "pull_request:1369", - "right": "pull_request:819", - "accept": false, - "reason": "A Flax sinusoidal embedding fix and DDIM timestep int64 fix are unrelated numerical issues." - }, - { - "left": "pull_request:3846", - "right": "pull_request:5833", - "accept": false, - "reason": "Random module seeding and single-file ControlNet import fixes address different problems." - }, - { - "left": "pull_request:3846", - "right": "pull_request:5834", - "accept": false, - "reason": "Random seeding and ControlNet single-file import path fixes are unrelated." - }, - { - "left": "pull_request:4714", - "right": "pull_request:5668", - "accept": false, - "reason": "One fixes a dummy-object error path; the other makes LoRA text-encoder typing/install behavior work without transformers. Not the same change." - }, { "left": "pull_request:6841", "right": "pull_request:8981", "accept": false, - "reason": "Both touch example/docs-style files, but one restores BasicTransformerBlock args and the other broadens LoRA docs. Different concerns." + "reason": "Different changes: one restores BasicTransformerBlock params; the other is LoRA docs cleanup. Same broad area, not same bug." }, { "left": "pull_request:7158", "right": "pull_request:8229", "accept": false, - "reason": "LCM benchmark cleanup and SVD resize behavior are unrelated." + "reason": "Unrelated fixes: benchmark cleanup vs Stable Video Diffusion resize behavior." }, { "left": "pull_request:4098", "right": "pull_request:6003", "accept": false, - "reason": "Both involve video diffusion, but SDXL inpainting and Stable Video Diffusion inpaint work are different pipelines and code paths." + "reason": "Both are inpainting-related, but for different pipelines and different code paths; not the same concrete change." }, { "left": "pull_request:1966", "right": "pull_request:2040", "accept": false, - "reason": "Flax inpaint and InstructPix2Pix are separate pipeline additions with different behavior and targets." + "reason": "Different pipeline additions: Flax inpainting vs InstructPix2Pix. Not the same underlying change." }, { "left": "pull_request:12301", "right": "pull_request:12445", "accept": false, - "reason": "Qwen image ControlNet inpainting support and Flux modular alignment are unrelated features." + "reason": "Qwen ControlNet inpaint support and Flux modular alignment are separate features." }, { "left": "pull_request:12117", "right": "pull_request:12416", "accept": false, - "reason": "QwenImage inpaint/img2img pipelines and QwenImage Edit Plus modular support are different feature sets." + "reason": "QwenImage pipelines vs QwenImage Edit Plus modular support are distinct changes." }, { "left": "pull_request:10020", "right": "pull_request:9175", "accept": false, - "reason": "SD3 ControlNet conversion/support and Flux Union ControlNet are different model families and implementation paths." + "reason": "Different features: SD3.5 ControlNet conversion tooling vs Flux ControlNet support." }, { "left": "pull_request:12119", "right": "pull_request:12439", "accept": false, - "reason": "QwenImage Lightning LoRA support and attention-backend kernel support are unrelated." + "reason": "LoRA loading support and attention backend support are unrelated code paths." }, { "left": "pull_request:12611", "right": "pull_request:13216", "accept": false, - "reason": "Both are modular-pipeline expansions, but Wan modular work and Helios modular work are distinct model families." + "reason": "Separate modular pipeline families (Wan vs Helios), not duplicates." }, { "left": "pull_request:10008", "right": "pull_request:11095", "accept": false, - "reason": "CogVideoX docs formatting and Flux controlnet example fixes are unrelated." + "reason": "Docs table fix for CogVideoX vs Flux ControlNet example fix are unrelated." }, { "left": "pull_request:10119", "right": "pull_request:11556", "accept": false, - "reason": "LoRA docs wording and regional compilation docs cover different topics." + "reason": "Docs for load_lora_adapter vs regional compilation docs; no shared bug/change." }, { "left": "pull_request:1012", "right": "pull_request:1129", "accept": false, - "reason": "DreamBooth training mode for text encoder and Flax sample batch sizing are unrelated training fixes." + "reason": "Both touch DreamBooth, but one is PyTorch text-encoder train mode and the other is Flax sample batch sizing." }, { "left": "pull_request:10230", "right": "pull_request:11004", "accept": false, - "reason": "Docs for LoRA loader mixins and PEFT full-key-prefix handling are not the same bug or change." + "reason": "Docs update for LoRA mixins and PEFT key-prefix handling are different issues." }, { "left": "pull_request:10494", "right": "pull_request:10497", "accept": false, - "reason": "Both modify tokenizers installation, but they propose different, mutually exclusive fixes; not a single merged change." + "reason": "Both address tokenizers install in CI, but they propose different fixes to the same line; related, but not the same concrete change." }, { "left": "pull_request:10633", "right": "pull_request:10898", "accept": false, - "reason": "Flux IP-Adapter docs and remote_decode implementation are unrelated." + "reason": "Flux IP-Adapter docs and remote_decode infrastructure are unrelated." }, { "left": "pull_request:10730", "right": "pull_request:9257", "accept": false, - "reason": "Hook unwrapping in pipeline utilities and DreamBooth Flux prompt-tokenization fixes are different code paths." + "reason": "Hook-unwrapping fix and DreamBooth Flux script bug fix are separate problems." + }, + { + "left": "pull_request:10894", + "right": "pull_request:9709", + "accept": false, + "reason": "Toctree sorting/docs ordering is unrelated to the torch version pin." + }, + { + "left": "pull_request:11243", + "right": "pull_request:11999", + "accept": false, + "reason": "Musubi Wan LoRA conversion support and PEFT LoRA injection changes are different LoRA issues." + }, + { + "left": "pull_request:11261", + "right": "pull_request:12288", + "accept": false, + "reason": "VidTok autoencoder addition and InfiniteTalk pipeline addition are unrelated model/pipeline features." + }, + { + "left": "pull_request:1496", + "right": "pull_request:2516", + "accept": false, + "reason": "Habana Gaudi docs and evaluation docs are unrelated documentation additions." + }, + { + "left": "pull_request:1920", + "right": "pull_request:750", + "accept": false, + "reason": "Translating docs and LMS scheduler back-compat are completely unrelated." + }, + { + "left": "pull_request:1942", + "right": "pull_request:3850", + "accept": false, + "reason": "Script flag rename and audio_diffusion test fixes do not share an underlying bug." + }, + { + "left": "pull_request:2347", + "right": "pull_request:2467", + "accept": false, + "reason": "Scheduler schedule updates and latent output-type plumbing are unrelated changes." } ] }, @@ -165486,36 +165656,22 @@ "component_index": 1, "component_count": 266, "review_unit_index": 200, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10894", - "pull_request:11243", - "pull_request:11261", - "pull_request:11979", - "pull_request:11994", - "pull_request:11999", - "pull_request:12288", - "pull_request:1496", - "pull_request:1920", - "pull_request:1942", - "pull_request:2347", "pull_request:237", - "pull_request:2467", "pull_request:2487", - "pull_request:2516", "pull_request:3260", "pull_request:3370", "pull_request:3552", "pull_request:3561", "pull_request:3711", "pull_request:3806", - "pull_request:3850", "pull_request:3877", "pull_request:4242", "pull_request:4276", @@ -165531,27 +165687,32 @@ "pull_request:5196", "pull_request:5204", "pull_request:5391", + "pull_request:5420", + "pull_request:5421", "pull_request:5456", "pull_request:5602", "pull_request:5614", + "pull_request:5626", + "pull_request:5724", "pull_request:5754", "pull_request:5880", + "pull_request:5974", + "pull_request:5988", + "pull_request:6081", + "pull_request:6105", + "pull_request:6140", "pull_request:6163", + "pull_request:6174", + "pull_request:6213", "pull_request:6346", + "pull_request:6908", + "pull_request:6909", "pull_request:7134", - "pull_request:750", - "pull_request:934", - "pull_request:9709" + "pull_request:7359", + "pull_request:7780", + "pull_request:934" ], "soft_pairs": [ - "pull_request:10894|pull_request:9709", - "pull_request:11243|pull_request:11999", - "pull_request:11261|pull_request:12288", - "pull_request:11979|pull_request:11994", - "pull_request:1496|pull_request:2516", - "pull_request:1920|pull_request:750", - "pull_request:1942|pull_request:3850", - "pull_request:2347|pull_request:2467", "pull_request:237|pull_request:934", "pull_request:2487|pull_request:492", "pull_request:3260|pull_request:4887", @@ -165568,186 +165729,201 @@ "pull_request:4635|pull_request:5602", "pull_request:5097|pull_request:6163", "pull_request:5204|pull_request:5754", - "pull_request:5391|pull_request:7134" + "pull_request:5391|pull_request:7134", + "pull_request:5420|pull_request:6213", + "pull_request:5421|pull_request:5626", + "pull_request:5724|pull_request:6174", + "pull_request:5974|pull_request:6908", + "pull_request:5974|pull_request:6909", + "pull_request:5988|pull_request:6140", + "pull_request:6081|pull_request:6213", + "pull_request:6105|pull_request:7359", + "pull_request:6140|pull_request:7780" ], - "prepared_review_unit_hash": "148a35fb7c8fb551995262726c128f11b97fd7637eb2d7f18a5f7a16cbf169a7", + "prepared_review_unit_hash": "a989fb38b8c981937581eb7aaab3acffa991d08e4c83331d7ae2e39a5aeb1ab8", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 105964, - "estimated_input_tokens": 26491, - "estimated_eval_tokens": 53238 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 101304, + "estimated_input_tokens": 25326, + "estimated_eval_tokens": 50908 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a loose set of mostly unrelated docs, tests, and feature PRs. The soft links are superficial (shared subsystem, file overlap, or broad maintenance theme) rather than the same concrete bug/change, so none should be merged as duplicates.", - "confidence": 0.98, + "summary": "This cluster is not a duplicate set: it contains many unrelated pull requests spanning docs, tests, example scripts, and feature work, with only superficial thematic similarity in a few pairs.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR represents a shared underlying change; the paired PRs span unrelated docs, tests, and feature work.", + "canonical_pr_reason": "No single PR is a good canonical representative; the items are heterogeneous and do not share one concrete bug or change.", "best_issue_reason": null, - "best_pr_reason": "No PR is a good global canonical candidate because the cluster does not show a single duplicate target.", + "best_pr_reason": "No PR here is a strong global representative because the cluster is broad and the soft-linked pairs still address different code paths or documentation updates.", "soft_edge_verdicts": [ - { - "left": "pull_request:10894", - "right": "pull_request:9709", - "accept": false, - "reason": "Docs toctree sorting vs a torch version pin in setup.py; unrelated changes." - }, - { - "left": "pull_request:11243", - "right": "pull_request:11999", - "accept": false, - "reason": "Musubi WAN LoRA conversion logic vs PEFT LoRA config injection; different code paths and problems." - }, - { - "left": "pull_request:11261", - "right": "pull_request:12288", - "accept": false, - "reason": "VidTok autoencoder addition vs InfiniteTalk pipeline work; different models/features." - }, - { - "left": "pull_request:11979", - "right": "pull_request:11994", - "accept": false, - "reason": "Test decorators for torch version gating vs huggingface-hub auth/docs updates; not the same issue." - }, - { - "left": "pull_request:1496", - "right": "pull_request:2516", - "accept": false, - "reason": "Habana Gaudi docs vs diffusion evaluation docs; both docs-only but unrelated topics." - }, - { - "left": "pull_request:1920", - "right": "pull_request:750", - "accept": false, - "reason": "Translation docs vs LMS timestep back-compat in a scheduler; no shared underlying bug." - }, - { - "left": "pull_request:1942", - "right": "pull_request:3850", - "accept": false, - "reason": "Converter CLI flag fix vs audio diffusion test adjustments; different areas and intents." - }, - { - "left": "pull_request:2347", - "right": "pull_request:2467", - "accept": false, - "reason": "Scheduler noising schedule changes vs latent output type support; different pipeline/scheduler changes." - }, { "left": "pull_request:237", "right": "pull_request:934", "accept": false, - "reason": "Test workflow setup vs MPS test fixes; both touch tests/CI but not the same defect." + "reason": "One adds CI/fixes a test; the other fixes MPS test expectations. Different maintenance changes, not the same change or bug." }, { "left": "pull_request:2487", "right": "pull_request:492", "accept": false, - "reason": "A Flax/JAX docs guide vs initial Flax PNDM scheduler implementation; related ecosystem, but not one change." + "reason": "Docs for JAX/Flax stable diffusion vs initial Flax PNDM scheduler support; different features and code paths." }, { "left": "pull_request:3260", "right": "pull_request:4887", "accept": false, - "reason": "ControlNet tracker arg cleanup vs Kandinsky latent input removal; unrelated bug fixes." + "reason": "Removing a required CLI arg for ControlNet training is unrelated to removing latents from Kandinsky prior_emb2emb." }, { "left": "pull_request:3370", "right": "pull_request:5196", "accept": false, - "reason": "PyTorch 2.0 doc refresh vs xformers docs cleanup; both documentation maintenance, not duplicate work." + "reason": "Both touch docs about optimization, but one updates PyTorch 2.0 docs and the other edits xformers/memory docs; not the same change." }, { "left": "pull_request:3552", "right": "pull_request:5180", "accept": false, - "reason": "General docs maintenance vs a Wuerstchen doc-builder fix; different pages and edits." + "reason": "General docs maintenance versus a small Wuerstchen citation fix; unrelated documentation edits." }, { "left": "pull_request:3561", "right": "pull_request:3877", "accept": false, - "reason": "Dreambooth IF upscaling changes vs SDXL DreamBooth LoRA work; separate training pipelines." + "reason": "DreamBooth IF stage II upscaling changes are unrelated to the SDXL DreamBooth LoRA WIP implementation." }, { "left": "pull_request:3711", "right": "pull_request:5614", "accept": false, - "reason": "Different revert PRs touching unrelated doc/script changes; no shared underlying change." + "reason": "Reverting a docs-build secrets workflow change is unrelated to reverting SDXL training width/height ordering." }, { "left": "pull_request:3806", "right": "pull_request:5456", "accept": false, - "reason": "A checkpoint loading note in a README vs RealFill feature addition; not the same artifact or bug." + "reason": "A README note about checkpoint loading is unrelated to adding the RealFill training/inference project." }, { "left": "pull_request:4242", "right": "pull_request:5880", "accept": false, - "reason": "From-single-file VAE loading support vs VAE config registration for latent diffusion compatibility; related subject, different fixes." + "reason": "Allowing VAE loading from single-file checkpoints is a loader change; the VAE PR fixes config support across latent-diffusion models, not the same bug." }, { "left": "pull_request:4276", "right": "pull_request:4508", "accept": false, - "reason": "Pipeline import test coverage vs use_onnx parameter plumbing; unrelated test vs loader bug." + "reason": "A pipeline import test and a fix for use_onnx in from_pretrained are different concerns." }, { "left": "pull_request:4276", "right": "pull_request:5456", "accept": false, - "reason": "Pipeline import test coverage vs RealFill project addition; unrelated." + "reason": "Pipeline import coverage is unrelated to the RealFill training project." }, { "left": "pull_request:4285", "right": "pull_request:4422", "accept": false, - "reason": "Separate ONNX/OpenVINO docs updates with different content and scope; not the same documentation change." + "reason": "Both edit ONNX/OpenVINO docs, but they are separate documentation updates for different content and examples." }, { "left": "pull_request:4365", "right": "pull_request:6346", "accept": false, - "reason": "SDXL LoRA example support vs a follow-up state-dict conversion update; related script, but not the same concrete change." + "reason": "Both touch SDXL LoRA-related files, but one adds the example/docs and the other fixes state-dict conversion; not the same concrete change." }, { "left": "pull_request:4635", "right": "pull_request:5602", "accept": false, - "reason": "MultiControlNet docs vs table whitespace cleanup in pipeline overview; clearly different doc work." + "reason": "MultiControlNet docs vs table-format cleanup in stable diffusion overview; unrelated documentation work." }, { "left": "pull_request:5097", "right": "pull_request:6163", "accept": false, - "reason": "Fast CUDA float16 test tuning vs LoRA test fixes; both are tests, but on different features and failures." + "reason": "Both are test fixes for float16/LoRA-related paths, but they target different test cases and failure modes." }, { "left": "pull_request:5204", "right": "pull_request:5754", "accept": false, - "reason": "LoRA text-encoder scaling logic vs style-only formatting cleanup; not duplicate functionality." + "reason": "LoRA text-encoder scaling fix versus minor styling cleanup in training scripts; not the same issue." }, { "left": "pull_request:5391", "right": "pull_request:7134", "accept": false, - "reason": "Broad model typehint/docs cleanup vs a specific UNet typehint fix; the overlap is stylistic, not the same change." + "reason": "General typehint/docs cleanup in model classes versus UNet type-hint adjustments; separate documentation/style changes." + }, + { + "left": "pull_request:5420", + "right": "pull_request:6213", + "accept": false, + "reason": "A docs example fix for create_repo/repo_id is unrelated to a text-to-video pipeline cleanup." + }, + { + "left": "pull_request:5421", + "right": "pull_request:5626", + "accept": false, + "reason": "Import/deprecation fix in versatile diffusion is unrelated to the CPU offload test correction." + }, + { + "left": "pull_request:5724", + "right": "pull_request:6174", + "accept": false, + "reason": "LoRA test decorator changes are unrelated to adding a missing backslash in a docs command." + }, + { + "left": "pull_request:5974", + "right": "pull_request:6908", + "accept": false, + "reason": "PixArtAlpha pipeline adaptation is unrelated to SDXL ControlNet model-card standardization." + }, + { + "left": "pull_request:5974", + "right": "pull_request:6909", + "accept": false, + "reason": "PixArtAlpha pipeline adaptation is unrelated to ControlNet Flax model-card standardization." + }, + { + "left": "pull_request:5988", + "right": "pull_request:6140", + "accept": false, + "reason": "Kandinsky 3 docs/pipeline work is unrelated to the IP-Adapter API doc addition." + }, + { + "left": "pull_request:6081", + "right": "pull_request:6213", + "accept": false, + "reason": "Adjusting SDXL scheduler sigmas for torch.compile is unrelated to text-to-video pipeline cleanup." + }, + { + "left": "pull_request:6105", + "right": "pull_request:7359", + "accept": false, + "reason": "SDE Drag community pipeline addition is unrelated to the ResAdapter training script." + }, + { + "left": "pull_request:6140", + "right": "pull_request:7780", + "accept": false, + "reason": "IP-Adapter API doc introduction versus an IP-Adapter masking wording fix; same area but not the same concrete change." } ] }, @@ -165763,11 +165939,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 201, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -165785,32 +165961,30 @@ "pull_request:2347", "pull_request:2617", "pull_request:2994", + "pull_request:3900", "pull_request:3924", + "pull_request:4215", + "pull_request:4508", + "pull_request:4744", + "pull_request:4859", + "pull_request:5035", "pull_request:5057", - "pull_request:5420", - "pull_request:5421", - "pull_request:5626", - "pull_request:5724", - "pull_request:5974", - "pull_request:5988", - "pull_request:6081", - "pull_request:6105", - "pull_request:6140", - "pull_request:6174", + "pull_request:5164", + "pull_request:5579", + "pull_request:6003", "pull_request:6209", - "pull_request:6213", + "pull_request:6369", "pull_request:6757", - "pull_request:6908", - "pull_request:6909", "pull_request:6927", "pull_request:6998", "pull_request:7230", - "pull_request:7359", + "pull_request:7662", "pull_request:7780", "pull_request:7913", "pull_request:8322", "pull_request:8531", "pull_request:8571", + "pull_request:8755", "pull_request:8825", "pull_request:8937", "pull_request:9040", @@ -165818,18 +165992,10 @@ "pull_request:9213", "pull_request:9234", "pull_request:9707", - "pull_request:9807" + "pull_request:9807", + "pull_request:9968" ], "soft_pairs": [ - "pull_request:5420|pull_request:6213", - "pull_request:5421|pull_request:5626", - "pull_request:5724|pull_request:6174", - "pull_request:5974|pull_request:6908", - "pull_request:5974|pull_request:6909", - "pull_request:5988|pull_request:6140", - "pull_request:6081|pull_request:6213", - "pull_request:6105|pull_request:7359", - "pull_request:6140|pull_request:7780", "pull_request:6209|pull_request:6927", "pull_request:6757|pull_request:7913", "pull_request:6998|pull_request:7780", @@ -165846,192 +166012,187 @@ "pull_request:13294|pull_request:13352", "pull_request:1945|pull_request:2617", "pull_request:2002|pull_request:2347", - "pull_request:2994|pull_request:3924" + "pull_request:2994|pull_request:3924", + "pull_request:4215|pull_request:4744", + "pull_request:4508|pull_request:4859", + "pull_request:5035|pull_request:5579", + "pull_request:5164|pull_request:5579", + "pull_request:6369|pull_request:7662", + "pull_request:8755|pull_request:9968", + "pull_request:3900|pull_request:6003" ], - "prepared_review_unit_hash": "2a201de16693128320314f9bd479dae13eb576b821133ab291f642adea58ed49", + "prepared_review_unit_hash": "bf1832a5dbb444a623fc4fb318b883753536856f88723c0c6e94dad831704ccc", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 26, - "serialized_chars": 101104, - "estimated_input_tokens": 25276, - "estimated_eval_tokens": 50808 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 100160, + "estimated_input_tokens": 25040, + "estimated_eval_tokens": 50336 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a real duplicate group: it mixes unrelated PRs across LoRA, quantization, docs, CI, training scripts, and pipeline features. The soft edges are mostly superficial file/subsystem similarities rather than the same underlying bug or change.", - "confidence": 0.97, + "summary": "The proposed soft-edge pairs are not true duplicates: they mostly share a subsystem or release theme, but each PR targets a different concrete change. I rejected all soft edges and would treat PR #9213 as the best representative only because it is the broadest, most central PR in the set.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a valid canonical duplicate target here; the items do not converge on one shared bug or feature.", + "canonical_pr_reason": "No single PR is a true canonical duplicate for this set; if a representative is needed, #9213 is the strongest anchor because it is the broad quantization feature PR with the most discussion and inbound references.", "best_issue_reason": null, - "best_pr_reason": "No PR is suitable as a global best representative because the cluster spans many distinct changes with only loose topical overlap.", + "best_pr_reason": "#9213 is the best global representative because it is the largest, most impactful, and most referenced PR in the cluster, whereas the others are smaller point fixes or unrelated docs/CI changes.", "soft_edge_verdicts": [ - { - "left": "pull_request:5420", - "right": "pull_request:6213", - "accept": false, - "reason": "Different changes: a docs/example parameter fix vs a text-to-video pipeline cleanup." - }, - { - "left": "pull_request:5421", - "right": "pull_request:5626", - "accept": false, - "reason": "Unrelated: an import deprecation cleanup in a pipeline vs a CPU offload test fix." - }, - { - "left": "pull_request:5724", - "right": "pull_request:6174", - "accept": false, - "reason": "Different scopes: LoRA test decorator changes vs a docs formatting edit." - }, - { - "left": "pull_request:5974", - "right": "pull_request:6908", - "accept": false, - "reason": "PixArt pipeline logic vs ControlNet SDXL model-card standardization; not the same bug or change." - }, - { - "left": "pull_request:5974", - "right": "pull_request:6909", - "accept": false, - "reason": "PixArt pipeline logic vs a different ControlNet flax model-card refactor." - }, - { - "left": "pull_request:5988", - "right": "pull_request:6140", - "accept": false, - "reason": "Kandinsky 3 docs/pipeline work and IP-Adapter API docs are separate features." - }, - { - "left": "pull_request:6081", - "right": "pull_request:6213", - "accept": false, - "reason": "SDXL scheduler sigma handling vs text-to-video pipeline cleanup; different code paths." - }, - { - "left": "pull_request:6105", - "right": "pull_request:7359", - "accept": false, - "reason": "Two separate community/research pipeline additions, not one shared fix." - }, - { - "left": "pull_request:6140", - "right": "pull_request:7780", - "accept": false, - "reason": "Same docs file family, but one adds API docs and the other fixes masking wording; not one underlying change." - }, { "left": "pull_request:6209", "right": "pull_request:6927", "accept": false, - "reason": "README test-script fix vs prompt2prompt import change; unrelated." + "reason": "Different changes: README example fix vs community prompt2prompt import refactor. Same docs area, not the same underlying bug or change." }, { "left": "pull_request:6757", "right": "pull_request:7913", "accept": false, - "reason": "Both touch sigma/timestep handling, but in different pipelines and for different bugs/features." + "reason": "Both touch timestep/sigma handling, but one is a device-placement fix and the other adds custom timesteps/sigmas for ControlNet. Not mergeable as one duplicate change." }, { "left": "pull_request:6998", "right": "pull_request:7780", "accept": false, - "reason": "Both edit IP-Adapter docs, but they fix different text issues and do not form one change." + "reason": "Both edit the IP-Adapter docs, but they fix different doc sections and issues: callout markup vs masking/face-id wording." }, { "left": "pull_request:8322", "right": "pull_request:8571", "accept": false, - "reason": "Broad errata/docs cleanup vs philosophy/style edits; not the same change set." + "reason": "Both are typo/style errata, but they are broad, unrelated documentation cleanup PRs rather than the same concrete change." }, { "left": "pull_request:8531", "right": "pull_request:9234", "accept": false, - "reason": "Both are loading-related, but one fixes sharding with no device_map and the other fixes safetensors compatibility regression." + "reason": "Completely different fixes: sharding/device_map behavior versus safetensors compatibility regression." }, { "left": "pull_request:8825", "right": "pull_request:9807", "accept": false, - "reason": "k-diffusion ControlNet/IP-Adapter support vs AutoencoderKL LoRA adapter support; unrelated subsystems." + "reason": "Unrelated code paths: k-diffusion pipeline feature work versus AutoencoderKL LoRA adapter support." }, { "left": "pull_request:8937", "right": "pull_request:9707", "accept": false, - "reason": "Runtime long-prompt handling bug vs docstring refactoring; not duplicates." + "reason": "One fixes long-prompt handling in AuraFlow code; the other is a docstring refactor in a different model file." }, { "left": "pull_request:9040", "right": "pull_request:9041", "accept": false, - "reason": "Related HunyuanDiT area, but one adds differential diffusion and the other adds an img2img variant; different features." + "reason": "Same project area but different additions: Differential Diffusion vs an Img2Img variant for HunyuanDiT." }, { "left": "pull_request:5057", "right": "pull_request:7230", "accept": false, - "reason": "SDXL clip-skip support vs single-file Playground v2.5 loading support; different functionality." + "reason": "Different features: SDXL clip-skip support versus single-file checkpoint loading support." }, { "left": "pull_request:10578", "right": "pull_request:9213", "accept": false, - "reason": "Narrow 4bit Flux LoRA loading fix vs broad bitsandbytes quantization support; related topic but not the same concrete change." + "reason": "#10578 is a specific 4-bit Flux LoRA loading fix; #9213 is a broad bitsandbytes quantization framework addition. Related topic, not the same bug/change." }, { "left": "pull_request:11004", "right": "pull_request:11802", "accept": false, - "reason": "LoRA prefix handling cleanup vs deprecation removal; unrelated maintenance changes." + "reason": "Both involve LoRA cleanup, but one removes a PEFT prefix import path and the other removes deprecated lora_base methods. Not the same underlying change." }, { "left": "pull_request:11595", "right": "pull_request:12226", "accept": false, - "reason": "Flux LoRA loader fix vs a Qwen-Image ControlNet test addition; completely different targets." + "reason": "Different domains entirely: Flux LoRA loading fix versus a Qwen-Image ControlNet test addition." }, { "left": "pull_request:11894", "right": "pull_request:12618", "accept": false, - "reason": "CI dependency pinning vs a workflow uninstall typo fix; different CI issues." + "reason": "Both are CI-related, but one pins k-diffusion and the other fixes a uv install typo. Different failure modes." }, { "left": "pull_request:13294", "right": "pull_request:13352", "accept": false, - "reason": "WanVACE test flakiness cleanup vs Claude review workflow checkout; unrelated." + "reason": "Unrelated CI/workflow fixes: WanVACE test flakiness versus adding a checkout step to Claude review workflow." }, { "left": "pull_request:1945", "right": "pull_request:2617", "accept": false, - "reason": "UNet block tests vs introduction of VAE image processor; different code and purpose." + "reason": "Tests for UNet blocks versus introducing the image processor are unrelated changes." }, { "left": "pull_request:2002", "right": "pull_request:2347", "accept": false, - "reason": "LoRA example/training support vs scheduler noising schedule work; unrelated features." + "reason": "One adds a LoRA training example; the other changes scheduler noise scheduling. No shared concrete bug." }, { "left": "pull_request:2994", "right": "pull_request:3924", "accept": false, - "reason": "Runtime dtype-casting fix in ControlNet training vs a docs code-sample parameter rename." + "reason": "ControlNet training-script casting fix versus a doc example parameter update. Same example family, different change type and problem." + }, + { + "left": "pull_request:4215", + "right": "pull_request:4744", + "accept": false, + "reason": "Different documentation updates in different pipelines; not the same underlying issue." + }, + { + "left": "pull_request:4508", + "right": "pull_request:4859", + "accept": false, + "reason": "`use_onnx` loading behavior and an SDXL inpaint fast test tweak are unrelated." + }, + { + "left": "pull_request:5035", + "right": "pull_request:5579", + "accept": false, + "reason": "One avoids forcing transformers installation; the other bumps the minimum transformers version for PEFT. Different dependency policy changes." + }, + { + "left": "pull_request:5164", + "right": "pull_request:5579", + "accept": false, + "reason": "FreeU architecture changes are unrelated to a PEFT/transformers minimum-version bump." + }, + { + "left": "pull_request:6369", + "right": "pull_request:7662", + "accept": false, + "reason": "LoRA API deprecation cleanup versus removing redundant test dependencies from Flax workflows. Different code paths." + }, + { + "left": "pull_request:8755", + "right": "pull_request:9968", + "accept": false, + "reason": "Different model areas and fixes: SD3 DreamBooth LoRA training versus ControlNet module refactor." + }, + { + "left": "pull_request:3900", + "right": "pull_request:6003", + "accept": false, + "reason": "Video-to-video pipeline addition versus Stable Video Diffusion inpaint support. Similar high-level area, but not the same change." } ] }, @@ -166047,11 +166208,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 202, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -166080,38 +166241,30 @@ "pull_request:1945", "pull_request:1977", "pull_request:2048", + "pull_request:2279", "pull_request:2314", "pull_request:2325", - "pull_request:3900", + "pull_request:2398", + "pull_request:2489", + "pull_request:3054", + "pull_request:3334", + "pull_request:3546", + "pull_request:4090", "pull_request:4132", - "pull_request:4215", - "pull_request:4508", "pull_request:4597", - "pull_request:4744", - "pull_request:4859", - "pull_request:5035", + "pull_request:4712", "pull_request:5053", + "pull_request:5057", "pull_request:5127", - "pull_request:5164", "pull_request:538", "pull_request:5388", - "pull_request:5579", + "pull_request:5736", + "pull_request:5754", + "pull_request:5853", "pull_request:5991", - "pull_request:6003", - "pull_request:6369", - "pull_request:6816", - "pull_request:7662", - "pull_request:8755", - "pull_request:9968" + "pull_request:6816" ], "soft_pairs": [ - "pull_request:4215|pull_request:4744", - "pull_request:4508|pull_request:4859", - "pull_request:5035|pull_request:5579", - "pull_request:5164|pull_request:5579", - "pull_request:6369|pull_request:7662", - "pull_request:8755|pull_request:9968", - "pull_request:3900|pull_request:6003", "pull_request:1128|pull_request:1355", "pull_request:4597|pull_request:5127", "pull_request:5053|pull_request:5991", @@ -166129,162 +166282,126 @@ "pull_request:1496|pull_request:164", "pull_request:1945|pull_request:2314", "pull_request:1977|pull_request:538", - "pull_request:2048|pull_request:4132" + "pull_request:2048|pull_request:4132", + "pull_request:2279|pull_request:3054", + "pull_request:2398|pull_request:4090", + "pull_request:2489|pull_request:3334", + "pull_request:3546|pull_request:5057", + "pull_request:4712|pull_request:5853", + "pull_request:5736|pull_request:5754" ], - "prepared_review_unit_hash": "f2e6ec8d14739935cff9e1111b6591d1661d41835bb9147be9b6f5529a06634e", + "prepared_review_unit_hash": "18a36dc7a4aaa4a62b93adb56f3987184041f4a83ab74dccda739db35b436b06", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 101796, - "estimated_input_tokens": 25449, - "estimated_eval_tokens": 51154 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 100518, + "estimated_input_tokens": 25130, + "estimated_eval_tokens": 50516 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is noisy: the items span unrelated docs, pipeline features, quantization, LoRA, Flax deprecation, and model additions. The soft links look like superficial title/subsystem similarity rather than the same underlying bug or change, so none should be merged as duplicates.", - "confidence": 0.97, + "summary": "This cluster is very noisy: the soft edges mostly link unrelated PRs that only share broad keywords or a subsystem name. None of the candidate pairs looks like the same concrete fix/change, so all soft edges should be rejected.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No canonical PR: the cluster is heterogeneous, and the candidate pairings do not describe the same underlying change.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "No single PR is a safe global representative; the items span unrelated docs, tests, LoRA, Wan, SDXL, and scheduler changes.", "soft_edge_verdicts": [ - { - "left": "pull_request:4215", - "right": "pull_request:4744", - "accept": false, - "reason": "Different doc updates in different areas (ControlNet guidance vs AudioLDM docstring); no shared codepath or concrete change." - }, - { - "left": "pull_request:4508", - "right": "pull_request:4859", - "accept": false, - "reason": "One fixes `from_pretrained` ONNX argument handling, the other adjusts an SDXL inpaint fast test; unrelated." - }, - { - "left": "pull_request:5035", - "right": "pull_request:5579", - "accept": false, - "reason": "Both touch dependency/version logic, but one removes a hard transformers dependency and the other bumps the minimum version for PEFT; different problems." - }, - { - "left": "pull_request:5164", - "right": "pull_request:5579", - "accept": false, - "reason": "FreeU mechanism implementation is unrelated to the transformers minimum-version bump." - }, - { - "left": "pull_request:6369", - "right": "pull_request:7662", - "accept": false, - "reason": "LoRA API cleanup versus Flax CI workflow cleanup; no shared bug or change." - }, - { - "left": "pull_request:8755", - "right": "pull_request:9968", - "accept": false, - "reason": "SD3 DreamBooth LoRA training fix and ControlNet module refactor are different codepaths and fixes." - }, - { - "left": "pull_request:3900", - "right": "pull_request:6003", - "accept": false, - "reason": "Both video-related, but one adds video2video and the other adds SVD inpaint; distinct features." - }, { "left": "pull_request:1128", "right": "pull_request:1355", "accept": false, - "reason": "Inpaint tensor-input support is unrelated to adding a Flax img2img pipeline." + "reason": "Inpaint tensor-input support and a Flax img2img pipeline are unrelated features with different pipelines and codepaths." }, { "left": "pull_request:4597", "right": "pull_request:5127", "accept": false, - "reason": "Both add multi-input support, but for different abstractions (MultiControlNet vs MultiAdapter) and different pipelines." + "reason": "MultiControlNet SDXL support and multi-adapter SDXL adapter handling are different integration changes, not the same bug fix." }, { "left": "pull_request:5053", "right": "pull_request:5991", "accept": false, - "reason": "Validation-image CLI parsing fix is unrelated to LoRA serialization prefix handling." + "reason": "`validation_image` CLI handling and LoRA prefix serialization are unrelated fixes in different areas." }, { "left": "pull_request:5388", "right": "pull_request:6816", "accept": false, - "reason": "Both mention PEFT/LoRA, but one broadly adapts example scripts while the other fixes one SDXL training script under DDP; not the same concrete change." + "reason": "Both touch DreamBooth LoRA, but one adds PEFT support and the other fixes DDP embedding grad handling; different code paths and goals." }, { "left": "pull_request:11582", "right": "pull_request:12611", "accept": false, - "reason": "Wan VACE feature addition and Wan modular-pipeline refactor are different changes, not mergeable as one fix." + "reason": "Both are Wan-related, but Wan VACE and modular Wan are separate feature sets and not the same change." }, { "left": "pull_request:10983", "right": "pull_request:11025", "accept": false, - "reason": "Both are Hunyuan-related, but one adds Hunyuan I2V support and the other adds Wan I2V LoRA support; different models and codepaths." + "reason": "Hunyuan I2V model support and Wan LoRA support are unrelated model integrations." }, { "left": "pull_request:10322", "right": "pull_request:10537", "accept": false, - "reason": "Sana 2K model support and Sana-4K support are separate model additions with different targets." + "reason": "Sana 2K model conversion and Sana 4K pipeline changes are different model-size additions, not one underlying fix." }, { "left": "pull_request:10358", "right": "pull_request:11556", "accept": false, - "reason": "Community-projects docs entry and regional compilation docs are unrelated documentation changes." + "reason": "A community-projects docs entry and regional compilation docs are unrelated documentation updates." }, { "left": "pull_request:10370", "right": "pull_request:11059", "accept": false, - "reason": "A style-only follow-up PR is unrelated to Wan LoRA conversion logic." + "reason": "TorchAO quantizer style/import cleanup and Wan LoRA conversion logic are unrelated changes." }, { "left": "pull_request:10854", "right": "pull_request:11894", "accept": false, - "reason": "Hunyuan DiT test-module rename and k-diffusion pinning for CI are unrelated maintenance changes." + "reason": "Hunyuan DiT test-module renaming and k-diffusion pinning for CI are unrelated CI/test maintenance tasks." }, { "left": "pull_request:11524", "right": "pull_request:11535", "accept": false, - "reason": "Conditional torchvision import in Cosmos and Framepack transformer init-order changes affect different models and bugs." + "reason": "Cosmos torchvision import gating and Framepack transformer initialization order are different model implementation fixes." }, { "left": "pull_request:11556", "right": "pull_request:11715", "accept": false, - "reason": "Docs for regional compilation are unrelated to adding bnb version guards in quantization tests." + "reason": "Regional compilation docs and bnb test annotations do not fix the same issue." }, { "left": "pull_request:12151", "right": "pull_request:12541", "accept": false, - "reason": "Flax deprecation work is unrelated to Wan LoRA conversion support." + "reason": "Flax deprecation and late-step Wan LoRA conversion are unrelated changes." }, { "left": "pull_request:1283", "right": "pull_request:2325", "accept": false, - "reason": "Versatile Diffusion model integration and ONNX VAE converter support are unrelated feature additions." + "reason": "Versatile Diffusion model support and a VAE ONNX converter are separate features." }, { "left": "pull_request:1496", @@ -166296,19 +166413,55 @@ "left": "pull_request:1945", "right": "pull_request:2314", "accept": false, - "reason": "UNet block tests and fast-test expectation tuning are different test changes, not the same bug." + "reason": "UNet block tests and fast-test tolerance updates are different test additions." }, { "left": "pull_request:1977", "right": "pull_request:538", "accept": false, - "reason": "DreamBooth/ColossalAI script updates and hub-version bump are unrelated." + "reason": "DreamBooth ColossalAI script adjustments and hub-version pinning are unrelated updates." }, { "left": "pull_request:2048", "right": "pull_request:4132", "accept": false, - "reason": "Both involve accelerate, but one changes checkpoint save/load hooks while the other makes accelerate optional in a CKPT conversion script; different codepaths and goals." + "reason": "Accelerate checkpoint-hook loading and making accelerate optional in a converter solve different problems." + }, + { + "left": "pull_request:2279", + "right": "pull_request:3054", + "accept": false, + "reason": "Ruff/black tooling changes and a contribution-doc wording tweak are unrelated maintenance edits." + }, + { + "left": "pull_request:2398", + "right": "pull_request:4090", + "accept": false, + "reason": "Pix2Pix Zero docs and BentoML integration docs are unrelated documentation additions." + }, + { + "left": "pull_request:2489", + "right": "pull_request:3334", + "accept": false, + "reason": "Core ML docs and a docstring formatting fix are not the same underlying change." + }, + { + "left": "pull_request:3546", + "right": "pull_request:5057", + "accept": false, + "reason": "Panorama scheduler compatibility and SDXL clip-skip support are distinct pipeline features." + }, + { + "left": "pull_request:4712", + "right": "pull_request:5853", + "accept": false, + "reason": "Tests fetcher infrastructure and doctest workflows are separate test-system changes." + }, + { + "left": "pull_request:5736", + "right": "pull_request:5754", + "accept": false, + "reason": "RealFill torchvision compatibility and style cleanup in training scripts are unrelated." } ] }, @@ -166324,11 +166477,11 @@ "component_index": 1, "component_count": 266, "review_unit_index": 203, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -166336,30 +166489,29 @@ "pull_request:10188", "pull_request:10324", "pull_request:10376", + "pull_request:10378", + "pull_request:1038", + "pull_request:10473", + "pull_request:10579", "pull_request:10622", "pull_request:10747", "pull_request:10808", + "pull_request:10966", + "pull_request:11039", + "pull_request:11124", + "pull_request:11261", + "pull_request:11273", "pull_request:11684", "pull_request:11823", "pull_request:11869", + "pull_request:12040", "pull_request:12439", - "pull_request:2279", - "pull_request:2398", "pull_request:2421", - "pull_request:2489", "pull_request:2561", - "pull_request:3054", - "pull_request:3334", + "pull_request:2593", "pull_request:3386", - "pull_request:3546", - "pull_request:4090", "pull_request:4690", - "pull_request:4712", - "pull_request:5057", - "pull_request:5736", - "pull_request:5754", "pull_request:5807", - "pull_request:5853", "pull_request:6006", "pull_request:6213", "pull_request:6355", @@ -166375,18 +166527,13 @@ "pull_request:8797", "pull_request:9257", "pull_request:9311", + "pull_request:9426", "pull_request:9507", "pull_request:9735", "pull_request:9968", "pull_request:9985" ], "soft_pairs": [ - "pull_request:2279|pull_request:3054", - "pull_request:2398|pull_request:4090", - "pull_request:2489|pull_request:3334", - "pull_request:3546|pull_request:5057", - "pull_request:4712|pull_request:5853", - "pull_request:5736|pull_request:5754", "pull_request:5807|pull_request:6213", "pull_request:6355|pull_request:6787", "pull_request:7117|pull_request:7541", @@ -166404,180 +166551,193 @@ "pull_request:6929|pull_request:7086", "pull_request:10119|pull_request:10622", "pull_request:10188|pull_request:9257", - "pull_request:10376|pull_request:10808" + "pull_request:10376|pull_request:10808", + "pull_request:10378|pull_request:10473", + "pull_request:1038|pull_request:2593", + "pull_request:10579|pull_request:11261", + "pull_request:10808|pull_request:12040", + "pull_request:10966|pull_request:11124", + "pull_request:10966|pull_request:9426", + "pull_request:11039|pull_request:11273" ], - "prepared_review_unit_hash": "78adfff618358ee861568ae85c7c43c75cf6467cd6428a2984220a1b46d0cab7", + "prepared_review_unit_hash": "de0c771330a4c7f40573d1f16eb2d55495e9230cfe6814247db2c4d753697aba", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 100465, - "estimated_input_tokens": 25117, - "estimated_eval_tokens": 50490 + "soft_pair_count": 25, + "serialized_chars": 99311, + "estimated_input_tokens": 24828, + "estimated_eval_tokens": 49912 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly unrelated docs, infra, scheduler, and feature PRs that only look superficially similar. No soft edge looks like the same underlying bug/change, so all candidate duplicate links should be rejected.", - "confidence": 0.86, + "summary": "This cluster is highly heterogeneous: the soft pairs mostly connect unrelated docs, tests, scheduler, LoRA, and feature PRs rather than true duplicates.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a true canonical duplicate target for this cluster; the items span different subsystems and changes. If forced to pick one representative, PR 9985 is the broadest and most discussed feature PR in the set.", + "canonical_pr_reason": "PR 9985 is the broadest and most central change in the set by scope and discussion volume, though the cluster itself is not a true duplicate group.", "best_issue_reason": null, - "best_pr_reason": "PR 9985 is the strongest cluster representative by scope, discussion, and impact, but it should be treated as a representative feature PR rather than a duplicate anchor.", + "best_pr_reason": "PR 9985 is the strongest representative by breadth and activity, but it should not be treated as a duplicate canonical for the other items.", "soft_edge_verdicts": [ - { - "left": "pull_request:2279", - "right": "pull_request:3054", - "accept": false, - "reason": "Both are docs edits, but they address different documentation sections and changes; not the same underlying change." - }, - { - "left": "pull_request:2398", - "right": "pull_request:4090", - "accept": false, - "reason": "Different docs topics and files; no shared bug or concrete change." - }, - { - "left": "pull_request:2489", - "right": "pull_request:3334", - "accept": false, - "reason": "Unrelated docs additions; one adds Core ML docs, the other is a docstring cleanup." - }, - { - "left": "pull_request:3546", - "right": "pull_request:5057", - "accept": false, - "reason": "Different pipeline features: panorama scheduler support vs SDXL clip skip support." - }, - { - "left": "pull_request:4712", - "right": "pull_request:5853", - "accept": false, - "reason": "Both are test/CI infrastructure, but they solve different problems and touch different mechanisms." - }, - { - "left": "pull_request:5736", - "right": "pull_request:5754", - "accept": false, - "reason": "Both are style/compatibility cleanups, but they affect different examples and different issues." - }, { "left": "pull_request:5807", "right": "pull_request:6213", "accept": false, - "reason": "Different community pipelines and different fixes; not the same code-path problem." + "reason": "Both are community pipeline work, but they fix different areas: LPW SDXL embeddings vs. text-to-video pipeline cleanup." }, { "left": "pull_request:6355", "right": "pull_request:6787", "accept": false, - "reason": "Unrelated example/doc tweak vs image value-range bug fix." + "reason": "Unrelated fixes: one adjusts widget metadata, the other corrects Marigold input normalization." }, { "left": "pull_request:7117", "right": "pull_request:7541", "accept": false, - "reason": "Both touch nightly workflows, but one adds notifications and the other makes the workflow dispatchable." + "reason": "Both touch nightly workflows, but one adds Slack/reporting support while the other makes the workflow dispatchable for release tests." }, { "left": "pull_request:7156", "right": "pull_request:8229", "accept": false, - "reason": "Different bug fixes in unrelated areas: stalebot label handling vs SVD resize behavior." + "reason": "Completely different code paths: stale-bot label handling vs. Stable Video Diffusion resize behavior." }, { "left": "pull_request:7246", "right": "pull_request:8229", "accept": false, - "reason": "A test-device fix is not the same as the SVD pipeline resize bug." + "reason": "These are unrelated tests/pipeline fixes; no shared underlying bug or change." }, { "left": "pull_request:9735", "right": "pull_request:9968", "accept": false, - "reason": "Both involve ControlNet/SD3, but one updates an example while the other is a broad module refactor." + "reason": "One updates an SD3 ControlNet example, the other is a broad ControlNet module refactor; not the same concrete change." }, { "left": "pull_request:11684", "right": "pull_request:11823", "accept": false, - "reason": "Both are loader/conversion related, but they target different model families and distinct formats." + "reason": "Different LoRA features and model families: GGUF single-file loading vs. Fal-Kontext LoRA conversion." }, { "left": "pull_request:2561", "right": "pull_request:3386", "accept": false, - "reason": "Community inpainting pipelines and ControlNet refactor are different changes with different scopes." + "reason": "Both mention ControlNet inpainting, but one adds community inpaint scripts and the other refactors core ControlNet pipelines; not mergeable as one fix." }, { "left": "pull_request:9507", "right": "pull_request:9985", "accept": false, - "reason": "Both are Flux-related, but one is a specific ControlNet batch bug fix and the other is a broad Flux feature expansion." + "reason": "Both are Flux-related, but 9507 fixes a specific controlnet batch-addition bug while 9985 adds multiple Flux feature pipelines." }, { "left": "pull_request:10747", "right": "pull_request:9311", "accept": false, - "reason": "Different ControlNet additions: union model support vs CFG cutoff callback behavior." + "reason": "Different ControlNet changes: union model support vs. SDXL CFG cutoff callback." }, { "left": "pull_request:10324", "right": "pull_request:8797", "accept": false, - "reason": "Unrelated kandinsky offload fix vs model sharding test adjustments." + "reason": "Unrelated areas: Kandinsky CPU offload API fix vs. test sharding behavior." }, { "left": "pull_request:11869", "right": "pull_request:12439", "accept": false, - "reason": "Both mention kernels, but one adds GGUF CUDA kernel support and the other adds attention backend support." + "reason": "Both mention kernels, but GGUF CUDA inference and attention backend support are separate features with different code paths." }, { "left": "pull_request:2421", "right": "pull_request:819", "accept": false, - "reason": "A deprecation-message cleanup and a DDIM timestep dtype fix are unrelated." + "reason": "Unrelated fixes in different subsystems: deprecate-message cleanup vs. DDIM timestep dtype on Windows." }, { "left": "pull_request:4690", "right": "pull_request:6006", "accept": false, - "reason": "Related scheduler refactors, but not the same concrete change; one is a draft refactor and the other adds sigmas broadly." + "reason": "Related scheduler area, but not the same change: a draft DPMSolver refactor vs. a broad addition of `self.sigmas` across schedulers." }, { "left": "pull_request:6929", "right": "pull_request:7086", "accept": false, - "reason": "Community pipeline docs/update work vs AnimateDiff pipeline bug fixes are unrelated." + "reason": "Different community-pipeline changes: custom-pipeline loading docs vs. AnimateDiff img2vid/controlnet updates." }, { "left": "pull_request:10119", "right": "pull_request:10622", "accept": false, - "reason": "Different docs changes with no shared bug or feature." + "reason": "Both are docs-only, but they cover different topics: LoRA adapter loading vs. uv installation." }, { "left": "pull_request:10188", "right": "pull_request:9257", "accept": false, - "reason": "Both mention Flux, but they fix different code paths and different scripts/loaders." + "reason": "Different problems: Flux LoRA loading/expansion test behavior vs. DreamBooth Flux training script bugs." }, { "left": "pull_request:10376", "right": "pull_request:10808", "accept": false, - "reason": "One adds HunyuanVideo LoRA conversion support; the other changes PEFT config key handling." + "reason": "Separate LoRA work: HunyuanVideo format conversion vs. PEFT config key filtering." + }, + { + "left": "pull_request:10378", + "right": "pull_request:10473", + "accept": false, + "reason": "One adds distributed training support for SD3 LoRA; the other is just README wording updates." + }, + { + "left": "pull_request:1038", + "right": "pull_request:2593", + "accept": false, + "reason": "Different tasks entirely: Flax dtype fix vs. ControlNet checkpoint conversion tooling." + }, + { + "left": "pull_request:10579", + "right": "pull_request:11261", + "accept": false, + "reason": "Unrelated additions: an attentive eraser community pipeline vs. new VidTok autoencoder models." + }, + { + "left": "pull_request:10808", + "right": "pull_request:12040", + "accept": false, + "reason": "Different LoRA concerns: PEFT config adjustment vs. Wan Lightx2v LoRA/model support." + }, + { + "left": "pull_request:10966", + "right": "pull_request:11124", + "accept": false, + "reason": "The test PR is only a harness update; the bug-fix PR changes CogView4 input validation, so they are not the same change." + }, + { + "left": "pull_request:10966", + "right": "pull_request:9426", + "accept": false, + "reason": "Unrelated: CogView4 validation logic vs. CI artifact uploader version bump." + }, + { + "left": "pull_request:11039", + "right": "pull_request:11273", + "accept": false, + "reason": "Completely unrelated: single-file torch_dtype handling vs. Slack notification bot timeout handling." } ] }, @@ -166593,7 +166753,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 204, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -166601,16 +166761,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10378", - "pull_request:1038", - "pull_request:10473", - "pull_request:10579", - "pull_request:10808", - "pull_request:10966", - "pull_request:11039", - "pull_request:11124", - "pull_request:11261", - "pull_request:11273", "pull_request:11290", "pull_request:11366", "pull_request:1147", @@ -166618,7 +166768,6 @@ "pull_request:11728", "pull_request:11894", "pull_request:1192", - "pull_request:12040", "pull_request:12151", "pull_request:12491", "pull_request:12493", @@ -166637,27 +166786,30 @@ "pull_request:2048", "pull_request:2341", "pull_request:2584", - "pull_request:2593", "pull_request:2595", "pull_request:26", "pull_request:3334", "pull_request:3471", + "pull_request:3476", "pull_request:3586", + "pull_request:3589", "pull_request:3614", "pull_request:3675", + "pull_request:3716", + "pull_request:4034", + "pull_request:4044", + "pull_request:4177", + "pull_request:4188", "pull_request:5299", "pull_request:538", - "pull_request:788", - "pull_request:9426" + "pull_request:5702", + "pull_request:5741", + "pull_request:5747", + "pull_request:6450", + "pull_request:6511", + "pull_request:788" ], "soft_pairs": [ - "pull_request:10378|pull_request:10473", - "pull_request:1038|pull_request:2593", - "pull_request:10579|pull_request:11261", - "pull_request:10808|pull_request:12040", - "pull_request:10966|pull_request:11124", - "pull_request:10966|pull_request:9426", - "pull_request:11039|pull_request:11273", "pull_request:11290|pull_request:11728", "pull_request:11366|pull_request:11504", "pull_request:1147|pull_request:130", @@ -166675,186 +166827,186 @@ "pull_request:1956|pull_request:3334", "pull_request:2584|pull_request:3586", "pull_request:2595|pull_request:3675", - "pull_request:3471|pull_request:5299" + "pull_request:3471|pull_request:5299", + "pull_request:3476|pull_request:3716", + "pull_request:3589|pull_request:5747", + "pull_request:4034|pull_request:6450", + "pull_request:4044|pull_request:5702", + "pull_request:4177|pull_request:6511", + "pull_request:4188|pull_request:5741" ], - "prepared_review_unit_hash": "f7f4dcd2947b3385b1eec95103027a691293830d252da52ac552e0613063ad63", + "prepared_review_unit_hash": "418c6ad1ce79cf07ed308a1592ce9f36a7f6b9b3af16b1d387b88f7cbb268217", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 94387, - "estimated_input_tokens": 23597, - "estimated_eval_tokens": 47450 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 91919, + "estimated_input_tokens": 22980, + "estimated_eval_tokens": 46216 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a heterogeneous grab-bag of unrelated PRs spanning docs, CI, model additions, loaders, schedulers, and tests. The soft edges are all false positives; none of the paired PRs appear to fix the same concrete bug or implement the same change.", + "summary": "The soft pairs are only superficially similar; they span unrelated PRs across Dockerfiles, docs, training scripts, schedulers, loaders, and tests. None look like the same concrete bug or change, so all soft edges should be rejected.", "confidence": 0.98, - "canonical_issue_reason": "No issue artifact here; the cluster is PR-only.", - "canonical_pr_reason": "No single canonical PR: the items cover unrelated maintenance, docs, CI, model, loader, scheduler, and test changes across different code paths.", - "best_issue_reason": "No issue candidate is suitable as a representative because there are no issues in the cluster.", - "best_pr_reason": "No PR is a good global representative since the cluster is not one duplicate change set; it is a collection of unrelated PRs.", + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:10378", - "right": "pull_request:10473", - "accept": false, - "reason": "Different training-example changes: DS/LoRA SD3 code vs advanced-training README edits." - }, - { - "left": "pull_request:1038", - "right": "pull_request:2593", - "accept": false, - "reason": "Unrelated fixes: Flax dtype handling vs ControlNet checkpoint conversion tooling." - }, - { - "left": "pull_request:10579", - "right": "pull_request:11261", - "accept": false, - "reason": "Different features: a community pipeline addition vs VidTok autoencoder support." - }, - { - "left": "pull_request:10808", - "right": "pull_request:12040", - "accept": false, - "reason": "Both touch LoRA code, but one is PEFT config adjustment and the other is Wan LightX2V LoRA support; not the same bug/change." - }, - { - "left": "pull_request:10966", - "right": "pull_request:11124", - "accept": false, - "reason": "Same model family only; one fixes CogView4 input validation, the other changes a test tokenizer fixture." - }, - { - "left": "pull_request:10966", - "right": "pull_request:9426", - "accept": false, - "reason": "CogView4 input validation is unrelated to CI artifact uploader updates." - }, - { - "left": "pull_request:11039", - "right": "pull_request:11273", - "accept": false, - "reason": "Single-file loader dtype logic is unrelated to a Slack notification bot timeout change." - }, { "left": "pull_request:11290", "right": "pull_request:11728", "accept": false, - "reason": "Both are Dockerfile dependency edits, but for different images and dependency sets." + "reason": "Both touch Dockerfiles, but one unpins torch in an onnxruntime CPU image while the other adds quantization libraries to the doc-builder image; different purpose and code path." }, { "left": "pull_request:11366", "right": "pull_request:11504", "accept": false, - "reason": "HiDream transformer output shape change is unrelated to dependency/version table updates." + "reason": "Hidream transformer output-shape fix vs dependency-table/CI update; no shared underlying bug or change." }, { "left": "pull_request:1147", "right": "pull_request:130", "accept": false, - "reason": "Separate Flax example docs/code loading fix vs an early examples README setup note." + "reason": "Flax text encoder loading fix vs adding accelerate config docs; unrelated feature areas and no duplicate behavior." }, { "left": "pull_request:11894", "right": "pull_request:12491", "accept": false, - "reason": "CI pinning of k-diffusion is unrelated to documentation link checking." + "reason": "k-diffusion pinning for CI is unrelated to the documentation link-check workflow change." }, { "left": "pull_request:1192", "right": "pull_request:2048", "accept": false, - "reason": "Loading edge cases work vs accelerate checkpoint hooks are different broad loading/checkpoint topics and not the same code-path fix." + "reason": "General loading edge cases in pipeline utils vs accelerate checkpoint hooks and Dreambooth training flow; different problem statements." }, { "left": "pull_request:12151", "right": "pull_request:12973", "accept": false, - "reason": "Flax deprecation sweep is unrelated to adding GLM-Image support." + "reason": "Deprecating Flax support is the opposite of adding GLM-Image support; clearly different changes." }, { "left": "pull_request:12493", "right": "pull_request:12541", "accept": false, - "reason": "LTX-Video documentation/model support is unrelated to Wan LoRA conversion logic." + "reason": "LTX-Video pipeline/docs addition vs Wan LoRA conversion fix; different models and code paths." }, { "left": "pull_request:12493", "right": "pull_request:13255", "accept": false, - "reason": "LTX-Video pipeline/docs changes are unrelated to LoRA alpha-scale loading fixes." + "reason": "LTX-Video support and Z-Image LoRA-loading fix are unrelated model/pipeline changes." }, { "left": "pull_request:13042", "right": "pull_request:13150", "accept": false, - "reason": "QwenImage prompt-isolation fixes are unrelated to test-only recompile_limit plumbing." + "reason": "QwenImage prompt-isolation fix vs test helper change for recompile limits; not the same bug or feature." }, { "left": "pull_request:1338", "right": "pull_request:788", "accept": false, - "reason": "ONNX inpaint cast-order fix is unrelated to a PIL resampling warning fix in textual inversion and img2img." + "reason": "ONNX inpainting cast-order fix vs PIL sampler warning fix in different pipelines; not duplicative." }, { "left": "pull_request:1494", "right": "pull_request:2341", "accept": false, - "reason": "Safetensors save support is unrelated to EMAModel checkpoint saving test changes." + "reason": "Safetensors save support vs EMA checkpoint-saving test adjustments; different functionality and no shared code defect." }, { "left": "pull_request:1525", "right": "pull_request:26", "accept": false, - "reason": "Upscaling batch-size fix is unrelated to DDPM alphas_cumprod math correction." + "reason": "Upscaling batch-size fix vs DDPM alpha_cumprod fix; both numerical but entirely different algorithms and code paths." }, { "left": "pull_request:1706", "right": "pull_request:538", "accept": false, - "reason": "Main docs refresh is unrelated to hub version bump." + "reason": "Main docs cleanup vs hub dependency version bump; documentation and packaging changes are unrelated." }, { "left": "pull_request:1945", "right": "pull_request:3614", "accept": false, - "reason": "2D UNet block tests are unrelated to a community StableDiffusionTilingPipeline plus test tweak." + "reason": "Adding 2D UNet block tests vs tweaking an existing test file/comment; not the same underlying bug/change." }, { "left": "pull_request:1956", "right": "pull_request:3334", "accept": false, - "reason": "Typo fix in a community script is unrelated to a pipeline docstring fix." + "reason": "Imagic typo fix in example code vs pipeline docstring cleanup; different files and no common bug." }, { "left": "pull_request:2584", "right": "pull_request:3586", "accept": false, - "reason": "Different community pipelines: ControlNet img2img vs StableDiffusionTilingPipeline." + "reason": "Community ControlNet img2img pipeline vs StableDiffusionTilingPipeline community example; separate additions." }, { "left": "pull_request:2595", "right": "pull_request:3675", "accept": false, - "reason": "Scheduler config-switching logic is unrelated to a V100-specific test fix." + "reason": "Scheduler switching/config normalization PR vs a test fix for V100; unrelated." }, { "left": "pull_request:3471", "right": "pull_request:5299", "accept": false, - "reason": "Dreambooth torch.compile docs note is unrelated to model typehint/doc cleanup." + "reason": "Dreambooth docs note about torch.compile vs typehint/docs improvements in models; not the same change." + }, + { + "left": "pull_request:3476", + "right": "pull_request:3716", + "accept": false, + "reason": "Adding use_karras_sigmas to a scheduler vs introducing ParaDiGMS pipeline and parallel schedulers; different feature areas." + }, + { + "left": "pull_request:3589", + "right": "pull_request:5747", + "accept": false, + "reason": "README/reference-tricks docs update vs dataset data_dir parameter in training scripts; unrelated docs vs training input handling." + }, + { + "left": "pull_request:4034", + "right": "pull_request:6450", + "accept": false, + "reason": "LoRA unload utility implementation vs LoRA deprecation test refactor; same broad subsystem but not the same concrete bug/change." + }, + { + "left": "pull_request:4044", + "right": "pull_request:5702", + "accept": false, + "reason": "AudioLDM example docstring typo fix vs AnimateDiff prompt bug fix; different pipelines and issues." + }, + { + "left": "pull_request:4177", + "right": "pull_request:6511", + "accept": false, + "reason": "Making safetensors required dependency vs torch.compile-compatible checkpointing for SDXL LoRA training; unrelated." + }, + { + "left": "pull_request:4188", + "right": "pull_request:5741", + "accept": false, + "reason": "SDXL ControlNet training cache/fingerprint fix vs speeding up PEFT LoRA unload; different code paths and goals." } ] }, @@ -166870,7 +167022,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 205, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -166878,14 +167030,10 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10126", + "pull_request:10127", "pull_request:12933", "pull_request:12934", - "pull_request:3476", - "pull_request:3589", - "pull_request:3716", - "pull_request:4034", - "pull_request:4044", - "pull_request:4177", "pull_request:4214", "pull_request:4595", "pull_request:4944", @@ -166895,8 +167043,7 @@ "pull_request:5424", "pull_request:5653", "pull_request:5665", - "pull_request:5702", - "pull_request:5747", + "pull_request:5713", "pull_request:5754", "pull_request:5788", "pull_request:5823", @@ -166906,7 +167053,7 @@ "pull_request:6332", "pull_request:6432", "pull_request:6448", - "pull_request:6450", + "pull_request:6490", "pull_request:6511", "pull_request:6772", "pull_request:6884", @@ -166914,6 +167061,7 @@ "pull_request:7016", "pull_request:7115", "pull_request:7200", + "pull_request:7359", "pull_request:7458", "pull_request:7638", "pull_request:7652", @@ -166923,16 +167071,14 @@ "pull_request:8512", "pull_request:8566", "pull_request:8722", + "pull_request:8932", + "pull_request:8973", "pull_request:9400", + "pull_request:9827", "pull_request:9896", "pull_request:9987" ], "soft_pairs": [ - "pull_request:3476|pull_request:3716", - "pull_request:3589|pull_request:5747", - "pull_request:4034|pull_request:6450", - "pull_request:4044|pull_request:5702", - "pull_request:4177|pull_request:6511", "pull_request:4214|pull_request:5754", "pull_request:4595|pull_request:6195", "pull_request:5030|pull_request:5424", @@ -166954,198 +167100,203 @@ "pull_request:4944|pull_request:5653", "pull_request:6884|pull_request:7016", "pull_request:12933|pull_request:12934", - "pull_request:5388|pull_request:5823" + "pull_request:5388|pull_request:5823", + "pull_request:6490|pull_request:7359", + "pull_request:5713|pull_request:7016", + "pull_request:10126|pull_request:9827", + "pull_request:10127|pull_request:8932", + "pull_request:10127|pull_request:8973" ], - "prepared_review_unit_hash": "6b0f7307c75e7b46981887f9229e975b9bb860815939b5fead66fee77d318fe8", + "prepared_review_unit_hash": "0c920336df852c3d431874248e774115307633986f3114f5037bbfa56fbaa7b8", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 102037, - "estimated_input_tokens": 25510, - "estimated_eval_tokens": 51276 + "serialized_chars": 103034, + "estimated_input_tokens": 25759, + "estimated_eval_tokens": 51774 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set; it mixes unrelated PRs across LoRA, schedulers, ControlNet, IP-Adapter, docs, tests, and various small fixes. The soft edges are mostly title/scope coincidences or shared subsystem files, not the same underlying change.", - "confidence": 0.95, + "summary": "This cluster is mostly noisy soft-similarity matches across unrelated diffusers PRs. None of the candidate pairs look like true duplicates or the same concrete code change, so all soft edges should be rejected.", + "confidence": 0.69, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR #5713 is the broadest foundational change in the cluster: the main IP-Adapter feature rollout with wide pipeline/loaders impact and multiple later adapter-related follow-ups nearby. It is the best representative PR among an otherwise noisy set.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR #5713 is the strongest standalone representative of the cluster because it is substantive, merged, broadly referenced, and spans the widest relevant adapter-related surface area.", "soft_edge_verdicts": [ - { - "left": "pull_request:3476", - "right": "pull_request:3716", - "accept": false, - "reason": "Different changes: DPMSolver scheduler option vs a new ParaDiGMS pipeline; only both are scheduler/pipeline work." - }, - { - "left": "pull_request:3589", - "right": "pull_request:5747", - "accept": false, - "reason": "Unrelated: community README/reference tricks vs a dataset `data_dir` argument fix in training scripts." - }, - { - "left": "pull_request:4034", - "right": "pull_request:6450", - "accept": false, - "reason": "Both LoRA-related, but one adds unload utility while the other changes LoRA test assertions/deprecation coverage; not the same bug." - }, - { - "left": "pull_request:4044", - "right": "pull_request:5702", - "accept": false, - "reason": "Different pipelines and different fixes: AudioLDM docstring typo vs AnimateDiff prompt default/embedding handling." - }, - { - "left": "pull_request:4177", - "right": "pull_request:6511", - "accept": false, - "reason": "Dependency pinning vs training checkpointing under torch.compile; no shared code path beyond general infra." - }, { "left": "pull_request:4214", "right": "pull_request:5754", "accept": false, - "reason": "Different scope: bf16 error fix in custom diffusion vs formatting-only dataset loading cleanup." + "reason": "Custom Diffusion bf16 fix vs docs/style cleanup; completely different changes and code paths." }, { "left": "pull_request:4595", "right": "pull_request:6195", "accept": false, - "reason": "WIP unrelated to the other SDXL/AnimateDiff model changes; no common concrete fix." + "reason": "A WIP SDXL conversion effort vs AnimateDiff SDXL; unrelated features." }, { "left": "pull_request:5030", "right": "pull_request:5424", "accept": false, - "reason": "Both touch offload/test utilities, but one fixes a missing key during model offload and the other changes pipeline test matrix filtering." + "reason": "Model offload key handling vs slow-test pipeline fetcher logic; no shared bug/change." }, { "left": "pull_request:5157", "right": "pull_request:6448", "accept": false, - "reason": "Different LoRA bugs: kohya/controlnet conversion mapping vs LoRA test refactor/assertion cleanup." + "reason": "ControlNet LoRA conversion fix vs LoRA test assertion cleanup; different objectives and diffs." }, { "left": "pull_request:5157", "right": "pull_request:6772", "accept": false, - "reason": "Converter bug fix vs large ControlNet-XS implementation; only superficial overlap in model code." + "reason": "ControlNet LoRA conversion fix vs ControlNet-XS implementation; same broad area but not the same concrete change." }, { "left": "pull_request:5665", "right": "pull_request:6772", "accept": false, - "reason": "ControlNet img2img preprocess arguments vs a new ControlNet-XS feature implementation; not mergeable as one fix." + "reason": "ControlNet img2img height/width preprocessing vs ControlNet-XS feature implementation; unrelated." }, { "left": "pull_request:5788", "right": "pull_request:6231", "accept": false, - "reason": "Both about training stability, but one improves an LCM scheduler error message and the other upcasts fp16 LoRA params in SDXL training." + "reason": "LCM timestep error handling vs SDXL fp16 training fix; different subsystems and problems." }, { "left": "pull_request:6081", "right": "pull_request:6332", "accept": false, - "reason": "Different targets: SDXL scheduler sigma handling vs a community pipeline for rerender-a-video." + "reason": "SDXL sigma handling vs community Rerender-A-Video pipeline; unrelated." }, { "left": "pull_request:6332", "right": "pull_request:8404", "accept": false, - "reason": "Both community/video-oriented, but one is Rerender-A-Video and the other is Latte video-generation support; distinct pipelines." + "reason": "Rerender-A-Video community pipeline vs Latte video model/pipeline; different features." }, { "left": "pull_request:6432", "right": "pull_request:7638", "accept": false, - "reason": "Different content: README learning-rate tweak vs fixing a broken contributing docs link." + "reason": "README learning-rate tweak vs CONTRIBUTING link fix; unrelated docs changes." }, { "left": "pull_request:6432", "right": "pull_request:8303", "accept": false, - "reason": "One is a training README value change; the other is a philosophy wording typo." + "reason": "DreamBooth README update vs philosophy typo fix; unrelated." }, { "left": "pull_request:6511", "right": "pull_request:7652", "accept": false, - "reason": "Checkpointing with torch.compile vs Docker buildx workflow setup; unrelated maintenance areas." + "reason": "torch.compile checkpointing support vs Docker buildx workflow change; no shared code path." }, { "left": "pull_request:689", "right": "pull_request:819", "accept": false, - "reason": "Both schedulers, but one changes attention implementation for speed and the other fixes DDIM timestep dtype on Windows." + "reason": "Attention slow-test optimization vs DDIM int64 timestep fix; different bugs and code paths." }, { "left": "pull_request:7115", "right": "pull_request:8512", "accept": false, - "reason": "Dockerfile dependency install change vs removing deprecated mask helper from inpaint pipelines; unrelated." + "reason": "Dockerfile package-install tweak vs deprecated mask helper removal; unrelated." }, { "left": "pull_request:7200", "right": "pull_request:8722", "accept": false, - "reason": "Transformers2D validation/error behavior vs a broad embeddings refactor; not the same bug or change." + "reason": "Transformer2D norm/patch-input error handling vs embeddings refactor; not the same bug or change." }, { "left": "pull_request:7458", "right": "pull_request:7638", "accept": false, - "reason": "Python-version test skipping vs CONTRIBUTING link fix; no shared underlying issue." + "reason": "Skip dynamo tests on Python 3.12 vs docs link correction; unrelated." }, { "left": "pull_request:8566", "right": "pull_request:9987", "accept": false, - "reason": "Shared SD3-related files, but 8566 adds SD3 ControlNet/Multi-ControlNet while 9987 adds SD3.5 IP-Adapter integration; different features and code paths." + "reason": "SD3 ControlNet/Multi-ControlNet vs SD3.5 IP-Adapter integration; both SD3-related but distinct features and code paths." }, { "left": "pull_request:9400", "right": "pull_request:9896", "accept": false, - "reason": "Both tweak testing/pipeline execution behavior, but one is device-agnostic single-file tests and the other reverts Flux device/typecasting changes." + "reason": "Device-agnostic single-file pipeline tests vs Flux device/typecast revert; unrelated." }, { "left": "pull_request:4944", "right": "pull_request:5653", "accept": false, - "reason": "IP-Adapter support vs a PEFT-aware `maybe_raise_or_warn` helper; related area, but different concrete problems." + "reason": "IP-Adapter support feature vs PEFT-aware maybe_raise_or_warn helper; same area only loosely, not the same change." }, { "left": "pull_request:6884", "right": "pull_request:7016", "accept": false, - "reason": "Both in IP-Adapter, but 6884 validates scale length while 7016 refactors image-embed preparation and skips loading the image encoder; separate changes." + "reason": "Scale-length validation in set_ip_adapter_scale vs image-embedding refactor/load skipping; separate IP-Adapter follow-ups, not one mergeable change." }, { "left": "pull_request:12933", "right": "pull_request:12934", "accept": false, - "reason": "Same LTX2 subsystem, but one adds LoRA support and the other adds distilled checkpoint/two-stage generation support; not the same PR-worthy fix." + "reason": "LTX-2 LoRA support vs LTX2 distilled checkpoint/two-stage generation support; related project, but distinct feature work." }, { "left": "pull_request:5388", "right": "pull_request:5823", "accept": false, - "reason": "Both involve PEFT/LoRA examples, but one migrates training scripts to PEFT and the other adds `adapter_names` to `fuse_lora`; different functionality." + "reason": "PEFT example-script adaptation vs adding adapter_names to fuse_lora; different LoRA/PEFT changes." + }, + { + "left": "pull_request:6490", + "right": "pull_request:7359", + "accept": false, + "reason": "RAVE video editing community pipeline vs ResAdapter training script; unrelated community additions." + }, + { + "left": "pull_request:5713", + "right": "pull_request:7016", + "accept": false, + "reason": "Initial IP-Adapter feature rollout vs later image-embed refactor/optimization; same subsystem, but not the same concrete code-path problem." + }, + { + "left": "pull_request:10126", + "right": "pull_request:9827", + "accept": false, + "reason": "save_attn_procs deprecation vs LoRA device-map loading fix; both LoRA-related but separate issues." + }, + { + "left": "pull_request:10127", + "right": "pull_request:8932", + "accept": false, + "reason": "xfail for incompatible SD configs vs temporal transformer test coverage; unrelated test changes." + }, + { + "left": "pull_request:10127", + "right": "pull_request:8973", + "accept": false, + "reason": "xfail for incompatible SD configs vs lumina model-size test reduction; unrelated." } ] }, @@ -167161,7 +167312,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 206, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -167169,8 +167320,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10126", - "pull_request:10127", "pull_request:10230", "pull_request:10342", "pull_request:10506", @@ -167203,27 +167352,23 @@ "pull_request:2062", "pull_request:2122", "pull_request:2274", + "pull_request:2325", + "pull_request:2347", + "pull_request:2467", + "pull_request:2481", "pull_request:2793", "pull_request:2795", + "pull_request:326", "pull_request:3410", "pull_request:3850", - "pull_request:5713", - "pull_request:6490", + "pull_request:4276", + "pull_request:4499", + "pull_request:4500", "pull_request:689", - "pull_request:7016", - "pull_request:7359", - "pull_request:8932", - "pull_request:8973", "pull_request:9244", - "pull_request:9687", - "pull_request:9827" + "pull_request:9687" ], "soft_pairs": [ - "pull_request:6490|pull_request:7359", - "pull_request:5713|pull_request:7016", - "pull_request:10126|pull_request:9827", - "pull_request:10127|pull_request:8932", - "pull_request:10127|pull_request:8973", "pull_request:10230|pull_request:10719", "pull_request:10342|pull_request:10966", "pull_request:10506|pull_request:11124", @@ -167245,198 +167390,210 @@ "pull_request:1706|pull_request:2795", "pull_request:2062|pull_request:3410", "pull_request:2274|pull_request:3850", - "pull_request:2274|pull_request:689" + "pull_request:2274|pull_request:689", + "pull_request:2325|pull_request:4499", + "pull_request:2325|pull_request:4500", + "pull_request:2347|pull_request:4276", + "pull_request:2467|pull_request:326", + "pull_request:2481|pull_request:3850", + "pull_request:2481|pull_request:689" ], - "prepared_review_unit_hash": "fc54e041a2873f02dafd14b4eeacac96757e3b806bc930fc8bd13d3e42439a6a", + "prepared_review_unit_hash": "572b2b0a712e7229fb01a4381ccf23762513c3ce8bbb7309f19051028c66f131", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 27, - "serialized_chars": 103770, - "estimated_input_tokens": 25943, - "estimated_eval_tokens": 52142 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 100738, + "estimated_input_tokens": 25185, + "estimated_eval_tokens": 50626 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: the PRs span unrelated docs, tests, CI, and feature fixes across many subsystems. The soft edges are mostly superficial topic overlap (for example LoRA, Flux, or IP-Adapter) rather than the same underlying change or bug.", - "confidence": 0.97, + "summary": "These PRs are a heterogeneous mix of unrelated docs, tests, workflow, and code changes; the soft-similarity edges look like false positives rather than duplicate fixes.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:6490", - "right": "pull_request:7359", - "accept": false, - "reason": "Different community projects and issues; they only share the broad \u201cresearch project\u201d category, not the same feature or bug." - }, - { - "left": "pull_request:5713", - "right": "pull_request:7016", - "accept": false, - "reason": "Both are IP-Adapter related, but 5713 adds the feature while 7016 refactors embedding preparation and loading behavior; not the same concrete change." - }, - { - "left": "pull_request:10126", - "right": "pull_request:9827", - "accept": false, - "reason": "One deprecates `save_attn_procs()`, the other fixes LoRA device-map loading; same LoRA area but different code paths and problems." - }, - { - "left": "pull_request:10127", - "right": "pull_request:8932", - "accept": false, - "reason": "An xfail tweak for SD config tests is unrelated to adding temporal transformer test coverage." - }, - { - "left": "pull_request:10127", - "right": "pull_request:8973", - "accept": false, - "reason": "Both are test-only changes, but for different models and failure modes; not mergeable as one fix." - }, { "left": "pull_request:10230", "right": "pull_request:10719", "accept": false, - "reason": "Docs for LoRA loader mixins versus adding `provider_options` to ONNX model loading; unrelated changes." + "reason": "Unrelated docs-only LoRA mixin documentation vs ONNX provider_options plumbing; different change goals and code paths." }, { "left": "pull_request:10342", "right": "pull_request:10966", "accept": false, - "reason": "`get_parameter_dtype` behavior update is unrelated to CogView4 input validation." + "reason": "Generic utility dtype helper update vs CogView4 input validation fix; no shared bug or mergeable common change." }, { "left": "pull_request:10506", "right": "pull_request:11124", "accept": false, - "reason": "A TorchAO CPU offload test and a CogView4 tokenizer/test fix are different test issues." + "reason": "TorchAO sequential offload test vs CogView4 tokenizer test update; both are tests only and address different features." }, { "left": "pull_request:10539", "right": "pull_request:9244", "accept": false, - "reason": "Both touch Flux, but one changes CFG logic in the main pipeline while the other adds single-file support; distinct code paths." + "reason": "Both touch Flux, but one fixes true CFG gating in the pipeline while the other adds single-file loading support across Flux pipelines; separate changes." }, { "left": "pull_request:10719", "right": "pull_request:9687", "accept": false, - "reason": "ONNX `provider_options` loading support is unrelated to Xlabs ControlNet support." + "reason": "ONNX provider_options loading path vs Flux ControlNet model support; unrelated subsystems and functionality." }, { "left": "pull_request:11292", "right": "pull_request:12315", "accept": false, - "reason": "LoRA conversion support and LoRA test simplification are not the same underlying change." + "reason": "LoRA SDXL conversion logic vs a sweep of scheduler selection changes in tests; not the same concrete bug or change." }, { "left": "pull_request:1136", "right": "pull_request:2002", "accept": false, - "reason": "A README link fix is unrelated to adding LoRA training support and API changes." + "reason": "Old README link fix vs LoRA training documentation/code support; only broad docs overlap." }, { "left": "pull_request:11613", "right": "pull_request:12530", "accept": false, - "reason": "A large CI/test revert is unrelated to xfails for a specific Stable Cascade test." + "reason": "CI/docker/test revert for torch.compile vs xfail of one stable-cascade test; completely different maintenance actions." }, { "left": "pull_request:11823", "right": "pull_request:12275", "accept": false, - "reason": "Flux Kontext LoRA support and TorchAO quantization config support are different features." + "reason": "Fal Kontext LoRA conversion/support implementation vs TorchAO quantization feature support; unrelated loader/quantization areas." }, { "left": "pull_request:11877", "right": "pull_request:11892", "accept": false, - "reason": "Deprecating a pipeline and removing `k-diffusion` are separate chores with no shared bug fix." + "reason": "Deprecating a pipeline class vs removing a dependency from setup; both chores, but not the same fix." }, { "left": "pull_request:1229", "right": "pull_request:1309", "accept": false, - "reason": "DreamBooth docs and PIL handling changes are unrelated." + "reason": "DreamBooth install-doc wording vs broad PIL handling overhaul; same repository area, different underlying changes." }, { "left": "pull_request:12782", "right": "pull_request:13364", "accept": false, - "reason": "HunyuanVideo offloading test adjustments and MotionConv2d dtype casting fix different failures." + "reason": "HunyuanVideo offloading test tweak vs MotionConv2d blur-kernel dtype fix; test adjustment is not the same code-path bug." }, { "left": "pull_request:1309", "right": "pull_request:1945", "accept": false, - "reason": "PIL handling/doc changes and UNet block tests are unrelated." + "reason": "PIL handling overhaul across pipelines vs adding 2D UNet block tests; unrelated scope." }, { "left": "pull_request:13120", "right": "pull_request:13332", "accept": false, - "reason": "Modular AutoModel tagging tests and dependency-testing workflow tightening are unrelated." + "reason": "Modular pipeline AutoModel tagging tests vs dependency-testing workflow hardening; both test-related but different purposes and files." }, { "left": "pull_request:13330", "right": "pull_request:13332", "accept": false, - "reason": "TorchAO test fixes and dependency workflow changes do not target the same code path." + "reason": "TorchAO tests/workflow changes vs dependency testing workflow hardening; no shared concrete bug fix." }, { "left": "pull_request:1508", "right": "pull_request:2122", "accept": false, - "reason": "DreamBooth clarification and safetensors docs are separate documentation changes." + "reason": "DreamBooth prior-preservation clarification vs safetensors docs; both documentation, but unrelated content." }, { "left": "pull_request:1508", "right": "pull_request:2793", "accept": false, - "reason": "DreamBooth prior-preservation docs and cuDNN benchmark doc removal are unrelated docs edits." + "reason": "DreamBooth docs clarification vs cuDNN benchmark docs removal; different docs topics and goals." }, { "left": "pull_request:1564", "right": "pull_request:1910", "accept": false, - "reason": "Default accelerate config guidance and Korean docs initialization are unrelated." + "reason": "Default accelerate config guidance vs Korean docs init workflow; unrelated documentation changes." }, { "left": "pull_request:1706", "right": "pull_request:2795", "accept": false, - "reason": "Main docs update versus a text-to-image LoRA example code change; not the same artifact." + "reason": "Main README/docs update vs text-to-image LoRA training code fix; not the same bug or feature." }, { "left": "pull_request:2062", "right": "pull_request:3410", "accept": false, - "reason": "Paint-by-example CPU offload fix is unrelated to the Accelerate DeepSpeed plugin retrieval change." + "reason": "Paint-by-example CPU offload fix vs DeepSpeed plugin retrieval fix in a training script; different code paths and problems." }, { "left": "pull_request:2274", "right": "pull_request:3850", "accept": false, - "reason": "Both are test-related, but they address different example pipelines and different bugs." + "reason": "Function naming clash fix in unconditional training vs audio diffusion test setup fix; unrelated examples and failures." }, { "left": "pull_request:2274", "right": "pull_request:689", "accept": false, - "reason": "Function naming clash in training scripts is unrelated to an attention implementation performance change." + "reason": "Unconditional training script naming fix vs attention performance change; completely different area and intent." + }, + { + "left": "pull_request:2325", + "right": "pull_request:4499", + "accept": false, + "reason": "ONNX VAE converter addition vs SDXL reference-attention community example; no shared underlying change." + }, + { + "left": "pull_request:2325", + "right": "pull_request:4500", + "accept": false, + "reason": "ONNX VAE converter addition vs duplicate SDXL reference example; unrelated artifacts." + }, + { + "left": "pull_request:2347", + "right": "pull_request:4276", + "accept": false, + "reason": "Scheduler noising schedule updates vs pipeline import test; different functionality and files." + }, + { + "left": "pull_request:2467", + "right": "pull_request:326", + "accept": false, + "reason": "Both are broad pipeline/model refactors, but one adds latent output handling and the other is type-hint cleanup; not the same concrete bug/change." + }, + { + "left": "pull_request:2481", + "right": "pull_request:3850", + "accept": false, + "reason": "Training script logger fix vs audio diffusion test adjustment; different failure modes and locations." + }, + { + "left": "pull_request:2481", + "right": "pull_request:689", + "accept": false, + "reason": "Training script logger version guard vs attention implementation change; no common code-path problem." } ] }, @@ -167452,7 +167609,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 207, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -167460,10 +167617,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2325", - "pull_request:2347", - "pull_request:2467", - "pull_request:2481", "pull_request:2641", "pull_request:2793", "pull_request:2994", @@ -167477,14 +167630,10 @@ "pull_request:3675", "pull_request:3691", "pull_request:3806", - "pull_request:3850", "pull_request:3852", "pull_request:4099", "pull_request:4272", - "pull_request:4276", "pull_request:4301", - "pull_request:4499", - "pull_request:4500", "pull_request:453", "pull_request:4572", "pull_request:4690", @@ -167500,22 +167649,24 @@ "pull_request:6330", "pull_request:6432", "pull_request:6834", - "pull_request:689", "pull_request:7096", "pull_request:7309", "pull_request:7738", + "pull_request:7790", + "pull_request:7938", "pull_request:816", "pull_request:826", + "pull_request:8660", + "pull_request:8875", "pull_request:8964", - "pull_request:9060" + "pull_request:9060", + "pull_request:9175", + "pull_request:9447", + "pull_request:9735", + "pull_request:9918", + "pull_request:9938" ], "soft_pairs": [ - "pull_request:2325|pull_request:4499", - "pull_request:2325|pull_request:4500", - "pull_request:2347|pull_request:4276", - "pull_request:2467|pull_request:326", - "pull_request:2481|pull_request:3850", - "pull_request:2481|pull_request:689", "pull_request:2641|pull_request:3397", "pull_request:2793|pull_request:3691", "pull_request:2994|pull_request:3657", @@ -167535,192 +167686,191 @@ "pull_request:5837|pull_request:7096", "pull_request:6330|pull_request:6834", "pull_request:7309|pull_request:9060", - "pull_request:7738|pull_request:8964" + "pull_request:7738|pull_request:8964", + "pull_request:7790|pull_request:9447", + "pull_request:7938|pull_request:9735", + "pull_request:8660|pull_request:9938", + "pull_request:9735|pull_request:9918", + "pull_request:8875|pull_request:9175" ], - "prepared_review_unit_hash": "6dbf9d95402c8bd98fb3a7157a53dce395f066a85fc740b26493699253cc7844", + "prepared_review_unit_hash": "36fcc3a8268a85c169a3529def5e33314ae358618cf09eff323813aff1df37f2", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 26, - "serialized_chars": 93133, - "estimated_input_tokens": 23284, - "estimated_eval_tokens": 46824 + "soft_pair_count": 25, + "serialized_chars": 91985, + "estimated_input_tokens": 22997, + "estimated_eval_tokens": 46250 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated PRs spanning docs, examples, schedulers, loaders, and model fixes. I found one true duplicate-looking pair (4499/4500) for the SDXL reference_attn/reference_adain addition; the rest are not the same underlying change.", - "confidence": 0.92, + "summary": "This cluster is a heterogeneous grab-bag of unrelated docs, examples, loader, scheduler, and pipeline changes. None of the soft pairs look like true duplicates; they mostly share only superficial wording patterns.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 4499 is the best canonical for the only true duplicate pair here: it matches 4500\u2019s SDXL reference_attn/reference_adain change almost exactly, with the earlier/lower-number submission.", + "canonical_pr_reason": "No single canonical PR: the items span unrelated docs, examples, schedulers, loaders, and model/pipeline changes.", "best_issue_reason": null, - "best_pr_reason": "PR 4499 is the strongest representative of the only near-identical pair in the cluster; the remaining PRs are unrelated enough that no broader cluster-wide canonical PR emerges.", + "best_pr_reason": "PR 9175 is the most substantial and highest-discussion item, so it is the best cluster representative even though it is not a duplicate hub.", "soft_edge_verdicts": [ - { - "left": "pull_request:2325", - "right": "pull_request:4499", - "accept": false, - "reason": "Different changes: ONNX VAE converter vs SDXL reference pipeline/docs." - }, - { - "left": "pull_request:2325", - "right": "pull_request:4500", - "accept": false, - "reason": "Different changes: ONNX VAE converter vs SDXL reference pipeline/docs." - }, - { - "left": "pull_request:2347", - "right": "pull_request:4276", - "accept": false, - "reason": "Both touch utilities/tests around pipelines, but they fix different problems and code paths." - }, - { - "left": "pull_request:2467", - "right": "pull_request:326", - "accept": false, - "reason": "Different features: latent output support vs type-hint cleanup for PNDM and scheduler/model files." - }, - { - "left": "pull_request:2481", - "right": "pull_request:3850", - "accept": false, - "reason": "One is an example test fix; the other is audio_diffusion test setup changes." - }, - { - "left": "pull_request:2481", - "right": "pull_request:689", - "accept": false, - "reason": "Completely different area: training-script test fix vs attention performance change." - }, { "left": "pull_request:2641", "right": "pull_request:3397", "accept": false, - "reason": "Docs-only changes in different topics; not the same underlying bug or change." + "reason": "Different doc tasks: JAX notebook build setup vs a LoRA training doc note." }, { "left": "pull_request:2793", "right": "pull_request:3691", "accept": false, - "reason": "Different scopes: docs suggestion removal vs LoRA loader behavior for PT 2.0." + "reason": "Unrelated docs removal vs a code change in LoRA text-encoder loading." }, { "left": "pull_request:2994", "right": "pull_request:3657", "accept": false, - "reason": "Different fixes in separate scripts: ControlNet type casting vs checkpoint parsing arguments." + "reason": "Different bugs in different code paths: ControlNet type casting vs checkpoint parser tweaks." }, { "left": "pull_request:3034", "right": "pull_request:4572", "accept": false, - "reason": "Unrelated parameter-name fixes in different training scripts." + "reason": "Same theme of parameter-name fixes, but different scripts and APIs; not one concrete fix." }, { "left": "pull_request:3080", "right": "pull_request:3675", "accept": false, - "reason": "Pipeline save logic vs a V100 test workaround; not the same change." + "reason": "Pipeline save logic vs a test-only UniDiffuser fix; unrelated changes." }, { "left": "pull_request:326", "right": "pull_request:826", "accept": false, - "reason": "Type-hint/scheduler cleanup vs mixed-precision loss casting in training." + "reason": "Old type-hint refactor across many files vs DreamBooth fp16 loss casting; not the same bug." }, { "left": "pull_request:3321", "right": "pull_request:3339", "accept": false, - "reason": "Docs update vs dependency pinning; no shared bug/change." + "reason": "Docs update vs dependency pinning; no shared underlying change." }, { "left": "pull_request:3675", "right": "pull_request:5526", "accept": false, - "reason": "Test-environment workaround vs scheduler-order logic fix." + "reason": "A V100 test fix is unrelated to scheduler order handling." }, { "left": "pull_request:3806", "right": "pull_request:5056", "accept": false, - "reason": "Both mention checkpoints/LoRAs, but one is a README note and the other is multi-LoRA fusing docs." + "reason": "Checkpoint-loading note vs LoRA fusion documentation; different documentation topics." }, { "left": "pull_request:3852", "right": "pull_request:5898", "accept": false, - "reason": "Different documentation updates in unrelated pipeline/tutorial pages." + "reason": "Contributor credit note vs a prose cleanup in unrelated docs." }, { "left": "pull_request:4099", "right": "pull_request:6231", "accept": false, - "reason": "Both SDXL-related, but one fixes prompt embeds and the other fixes fp16 LoRA training stability." + "reason": "SDXL prompt-embedding bug fix vs LoRA fp16 training stability; different code paths." }, { "left": "pull_request:4272", "right": "pull_request:4987", "accept": false, - "reason": "Import-path maintenance vs removing a duplicated GLIGEN export; not the same bug." + "reason": "Missing optional imports cleanup vs removing a duplicate GLIGEN export; unrelated maintenance." }, { "left": "pull_request:4301", "right": "pull_request:6432", "accept": false, - "reason": "Different docs fixes: Kandinsky navigation/content vs DreamBooth LR example." + "reason": "Kandinsky docs update vs an SDXL dreambooth learning-rate edit; no overlap." }, { "left": "pull_request:453", "right": "pull_request:816", "accept": false, - "reason": "Installation typo in docs vs deepspeed compatibility in training loss; unrelated." + "reason": "Installation typo in docs vs deepspeed loss casting in training code." }, { "left": "pull_request:4690", "right": "pull_request:4862", "accept": false, - "reason": "Scheduler refactor and test work vs an SDXL ControlNet img2img bug fix; not the same issue." + "reason": "Different scheduler redesign/refactor vs an SDXL ControlNet img2img height/width fix." }, { "left": "pull_request:5672", "right": "pull_request:5974", "accept": false, - "reason": "PixArt non-square support vs PixArtAlpha LCM adaptation; different code paths." + "reason": "PixArt non-square image support vs adapting PixArtAlpha for a different model variant." }, { "left": "pull_request:5837", "right": "pull_request:7096", "accept": false, - "reason": "Textual inversion token handling vs ControlNet training script/docs enhancements." + "reason": "Textual inversion multi-token support vs ControlNet example enhancements; unrelated functionality." }, { "left": "pull_request:6330", "right": "pull_request:6834", "accept": false, - "reason": "SVD docs rendering tweak vs callback property typo; unrelated." + "reason": "SVD docs rendering fix vs callback API typo fix; different docs-only changes." }, { "left": "pull_request:7309", "right": "pull_request:9060", "accept": false, - "reason": "Different fixes in different modules: scheduler order validation vs removing a debug print." + "reason": "Scheduler order validation vs removing a stray debug print in Flux transformer." }, { "left": "pull_request:7738", "right": "pull_request:8964", "accept": false, - "reason": "LoRA device-placement fix vs AudioLDM2 cache-position fix; different components and bugs." + "reason": "LoRA device placement fix vs AudioLDM2 cache-position fix; different subsystems." + }, + { + "left": "pull_request:7790", + "right": "pull_request:9447", + "accept": false, + "reason": "Community SDXL pipeline adaptation vs pipeline save shard-size behavior; not the same issue." + }, + { + "left": "pull_request:7938", + "right": "pull_request:9735", + "accept": false, + "reason": "Docker Slack reporting workflow vs SD3 ControlNet example updates; unrelated CI vs example code." + }, + { + "left": "pull_request:8660", + "right": "pull_request:9938", + "accept": false, + "reason": "SD3 ControlNet input validation vs SDXL reference pipeline update; different pipelines and bugs." + }, + { + "left": "pull_request:9735", + "right": "pull_request:9918", + "accept": false, + "reason": "SD3 ControlNet example update vs adding a server example; no common code path." + }, + { + "left": "pull_request:8875", + "right": "pull_request:9175", + "accept": false, + "reason": "Different feature work: PAG inpaint support vs Flux Union ControlNet support; not mergeable into one PR." } ] }, @@ -167736,7 +167886,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 208, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -167770,34 +167920,30 @@ "pull_request:2389", "pull_request:2452", "pull_request:2565", + "pull_request:26", "pull_request:274", "pull_request:3252", "pull_request:3632", + "pull_request:3831", + "pull_request:4078", + "pull_request:4099", + "pull_request:4262", + "pull_request:4404", "pull_request:4408", "pull_request:4461", "pull_request:453", + "pull_request:4847", + "pull_request:4910", "pull_request:5127", + "pull_request:530", + "pull_request:5695", "pull_request:689", "pull_request:6946", - "pull_request:7790", - "pull_request:7938", - "pull_request:8660", - "pull_request:8875", - "pull_request:9175", "pull_request:9212", "pull_request:9244", - "pull_request:9447", - "pull_request:9507", - "pull_request:9735", - "pull_request:9918", - "pull_request:9938" + "pull_request:9507" ], "soft_pairs": [ - "pull_request:7790|pull_request:9447", - "pull_request:7938|pull_request:9735", - "pull_request:8660|pull_request:9938", - "pull_request:9735|pull_request:9918", - "pull_request:8875|pull_request:9175", "pull_request:10747|pull_request:10951", "pull_request:13217|pull_request:13229", "pull_request:5127|pull_request:6946", @@ -167817,186 +167963,191 @@ "pull_request:1920|pull_request:2452", "pull_request:2345|pull_request:4461", "pull_request:2367|pull_request:3252", - "pull_request:2565|pull_request:453" + "pull_request:2565|pull_request:453", + "pull_request:26|pull_request:530", + "pull_request:3831|pull_request:4847", + "pull_request:4078|pull_request:4910", + "pull_request:4099|pull_request:5695", + "pull_request:4262|pull_request:4404" ], - "prepared_review_unit_hash": "c4dd1b071fca8f3566f7ba746cc5b364485f4c0d3b2efe00214dfa933acfa8e6", + "prepared_review_unit_hash": "c1bbf9cd84a6a07b525305cac40679291cefcde71631d2b359ba294dac688ab0", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 97478, - "estimated_input_tokens": 24370, - "estimated_eval_tokens": 48996 + "serialized_chars": 97771, + "estimated_input_tokens": 24443, + "estimated_eval_tokens": 49142 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a set of mostly unrelated merged PRs that only look superficially similar by subsystem or wording. None of the soft-edge pairs appear to be the same underlying bug or change, so I would not merge any of them as duplicates.", + "summary": "This is a loose similarity cluster of mostly unrelated PRs across docs, pipelines, loaders, CI, and model support. The shared-file or shared-subsystem overlaps are coincidental rather than the same concrete bug/change, so all soft edges should be rejected.", "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR: the items are heterogeneous and do not share one underlying fix or feature.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 13217 is the most substantial and coherent standalone change in the set, with the strongest review/discussion footprint, so it is the best representative if one must pick a single PR.", "soft_edge_verdicts": [ - { - "left": "pull_request:7790", - "right": "pull_request:9447", - "accept": false, - "reason": "Different changes: SDXL masked img2img pipeline adaptation vs pipeline save_pretrained shard-size behavior. No shared code path." - }, - { - "left": "pull_request:7938", - "right": "pull_request:9735", - "accept": false, - "reason": "Unrelated: Docker build status reporting workflow vs SD3 controlnet example update. Same repo area only at a very broad level." - }, - { - "left": "pull_request:8660", - "right": "pull_request:9938", - "accept": false, - "reason": "Different features in different pipelines: SD3 controlnet input validation vs SDXL reference pipeline refresh." - }, - { - "left": "pull_request:9735", - "right": "pull_request:9918", - "accept": false, - "reason": "Different domains: SD3 controlnet example docs/training tweaks vs adding a server example and docs." - }, - { - "left": "pull_request:8875", - "right": "pull_request:9175", - "accept": false, - "reason": "Both touch PAG/ControlNet, but they add different pipeline classes and docs for different models (SD inpaint vs Flux union controlnet)." - }, { "left": "pull_request:10747", "right": "pull_request:10951", "accept": false, - "reason": "Different implementations and targets: MultiControlNetUnionModel on SDXL vs a community ControlNet tile upscaler pipeline." + "reason": "Both are SDXL ControlNet-related, but one adds a union model and the other adds a tile upscaler pipeline; different features and codepaths." }, { "left": "pull_request:13217", "right": "pull_request:13229", "accept": false, - "reason": "Both are LTX/Sana-related, but one adds LTX-2.3 model support and the other swaps in an LTX2 VAE for Sana-video. Not the same change." + "reason": "LTX-2.3 model support vs Sana-video VAE integration; different model families and conversion paths." }, { "left": "pull_request:5127", "right": "pull_request:6946", "accept": false, - "reason": "Different adapter features: SDXL multi-adapter support vs IP-Adapter low_cpu_mem_usage loading support." + "reason": "MultiAdapter support and IP-Adapter low_cpu_mem_usage loading are unrelated loader/pipeline changes." }, { "left": "pull_request:10552", "right": "pull_request:11823", "accept": false, - "reason": "Different loader paths: from_single_file support for StableDiffusionAdapterPipeline vs Flux Kontext LoRA conversion/loading support." + "reason": "From_single_file support for StableDiffusionAdapterPipeline is unrelated to Flux Kontext LoRA conversion support." }, { "left": "pull_request:3632", "right": "pull_request:4408", "accept": false, - "reason": "Same broad concept of pipeline docs/behavior, but one adds panorama view batching and the other fixes from_single_file for SDXL inpainting." + "reason": "Panorama view batching and SDXL from_single_file inpainting support are separate changes." }, { "left": "pull_request:10072", "right": "pull_request:9507", "accept": false, - "reason": "Different Flux issues: offloading SD3.5 controlnets vs Flux controlnet batch/control_mode fixes." + "reason": "SD3.5 controlnet offloading and Flux controlnet batch handling are different bugs in different codepaths." }, { "left": "pull_request:10292", "right": "pull_request:9212", "accept": false, - "reason": "AutoPipeline additions for Flux Control vs Kolors img2img routing; unrelated pipeline families." + "reason": "Same file, but distinct registrations for different pipeline families; not the same change." }, { "left": "pull_request:10719", "right": "pull_request:9244", "accept": false, - "reason": "Different loading features: provider_options passthrough for ONNX in from_pretrained vs adding single-file support for Flux pipelines." + "reason": "ONNX provider_options plumbing and Flux single-file support are unrelated loader features." }, { "left": "pull_request:10911", "right": "pull_request:11975", "accept": false, - "reason": "Test-only fix for Lumina2 LoRA nan fusion vs an unrelated style fix in a training example." + "reason": "A Lumina2 LoRA test fix and a training-script style fix are not the same underlying bug." }, { "left": "pull_request:11004", "right": "pull_request:11818", "accept": false, - "reason": "LoRA/PEFT cleanup vs Flux denoising sync removal; separate code paths and motivations." + "reason": "PEFT key-prefix cleanup and Flux denoising sync removal affect different subsystems and problems." }, { "left": "pull_request:11614", "right": "pull_request:12424", "accept": false, - "reason": "Both involve Docker workflows, but one changes a shell loop/env handling and the other rewires Dockerfile definitions. Not the same bug." + "reason": "Both touch docker/CI, but one fixes workflow file handling and the other changes Dockerfile definitions; not one concrete change." }, { "left": "pull_request:1282", "right": "pull_request:2389", "accept": false, - "reason": "Two unrelated documentation typo/link fixes in different docs pages." + "reason": "Different documentation typos in different pages; too granular and unrelated to merge." }, { "left": "pull_request:1309", "right": "pull_request:274", "accept": false, - "reason": "Large cross-cutting PIL handling update vs adding a Stable Diffusion README; no concrete overlap." + "reason": "PIL handling improvements across many components and adding an SD README are unrelated." }, { "left": "pull_request:1564", "right": "pull_request:3252", "accept": false, - "reason": "Dreambooth accelerate config docs vs notes about stateful model offloading; distinct documentation changes." + "reason": "Default accelerate config docs and stateful offloading notes are different documentation topics." }, { "left": "pull_request:1579", "right": "pull_request:274", "accept": false, - "reason": "Research-project examples addition vs Stable Diffusion README addition; different content and scope." + "reason": "Intel-optimized research examples and the Stable Diffusion README are unrelated additions." }, { "left": "pull_request:1787", "right": "pull_request:689", "accept": false, - "reason": "UnCLIP image count fix in training/test workflows vs an attention implementation change for performance." + "reason": "UnCLIP num_images_per_prompt behavior and attention speed changes are different bugs/optimizations." }, { "left": "pull_request:1920", "right": "pull_request:2452", "accept": false, - "reason": "Docs file addition vs draft core-api doc page; different documentation tasks." + "reason": "A translation guide and a core API docs scaffold are separate doc additions." }, { "left": "pull_request:2345", "right": "pull_request:4461", "accept": false, - "reason": "xformers training warning across scripts vs adding a pipeline_class_name argument to a conversion script." + "reason": "xFormers training warning and pipeline_class_name support in conversion scripts are unrelated fixes." }, { "left": "pull_request:2367", "right": "pull_request:3252", "accept": false, - "reason": "Checkpoint limit feature in training scripts vs offloading docs notes; unrelated." + "reason": "Checkpoint limit plumbing in training scripts and offloading documentation are not the same change." }, { "left": "pull_request:2565", "right": "pull_request:453", "accept": false, - "reason": "UNet1D doc correction vs installation typo fix; both are docs, but not the same change." + "reason": "UNet1D doc correction and installation typo fix are unrelated documentation edits." + }, + { + "left": "pull_request:26", + "right": "pull_request:530", + "accept": false, + "reason": "A scheduler math bugfix and a flaky-test assertion change are different issues." + }, + { + "left": "pull_request:3831", + "right": "pull_request:4847", + "accept": false, + "reason": "New DDPM inversion pipeline and prompt-free diffusion community example are distinct artifacts." + }, + { + "left": "pull_request:4078", + "right": "pull_request:4910", + "accept": false, + "reason": "Accelerate enforcement for from_single_file and an inpaint docs typo are unrelated." + }, + { + "left": "pull_request:4099", + "right": "pull_request:5695", + "accept": false, + "reason": "SDXL prompt-embedding bugfix and PixArt masking-condition fix are different model-specific bugs." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4404", + "accept": false, + "reason": "SDXL DreamBooth LoRA fixes and removing textual inversion mentions from SDXL docs are not the same code-path problem." } ] }, @@ -168012,7 +168163,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 209, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -168024,29 +168175,28 @@ "pull_request:11017", "pull_request:11421", "pull_request:11645", + "pull_request:1256", + "pull_request:1258", + "pull_request:12618", + "pull_request:1282", + "pull_request:13164", "pull_request:2040", "pull_request:241", - "pull_request:26", - "pull_request:3831", - "pull_request:4078", - "pull_request:4099", - "pull_request:4262", - "pull_request:4404", + "pull_request:2994", + "pull_request:3137", + "pull_request:3189", "pull_request:4617", "pull_request:4632", "pull_request:4783", "pull_request:4833", - "pull_request:4847", "pull_request:4910", "pull_request:4927", "pull_request:5057", "pull_request:5085", - "pull_request:530", "pull_request:5393", "pull_request:5407", "pull_request:5506", "pull_request:5615", - "pull_request:5695", "pull_request:5813", "pull_request:6099", "pull_request:6208", @@ -168070,11 +168220,6 @@ "pull_request:9988" ], "soft_pairs": [ - "pull_request:26|pull_request:530", - "pull_request:3831|pull_request:4847", - "pull_request:4078|pull_request:4910", - "pull_request:4099|pull_request:5695", - "pull_request:4262|pull_request:4404", "pull_request:4632|pull_request:5393", "pull_request:4783|pull_request:5813", "pull_request:4833|pull_request:6099", @@ -168094,193 +168239,194 @@ "pull_request:8616|pull_request:9423", "pull_request:9175|pull_request:9311", "pull_request:10633|pull_request:11017", - "pull_request:11421|pull_request:11645" + "pull_request:11421|pull_request:11645", + "pull_request:1256|pull_request:1258", + "pull_request:12618|pull_request:13164", + "pull_request:1282|pull_request:2994", + "pull_request:3137|pull_request:3189" ], - "prepared_review_unit_hash": "e2b656faccf41333a14dc24b3731a67f84130af08bf8ca0e1303e59eb607787c", + "prepared_review_unit_hash": "dbc07c4dcb7d7fa3b8d918109d04c0a0972211aabec166f6991c1762fe0c6eb9", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 100512, - "estimated_input_tokens": 25128, - "estimated_eval_tokens": 50512 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 97276, + "estimated_input_tokens": 24319, + "estimated_eval_tokens": 48894 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No duplicate cluster here: the soft pairs are mostly superficial matches across unrelated docs, examples, and bug fixes. None of the paired PRs appear to fix the same concrete code path or represent mergeable duplicates.", - "confidence": 0.98, + "summary": "The cluster is mostly heterogeneous: docs, CI, examples, and unrelated model/pipeline fixes. None of the soft edges except one look like true duplicates; they mostly share broad keywords only. The only likely duplicate pair is the two tiny pipeline-test PRs about adding negative_prompt handling.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR cleanly represents the cluster because the items span unrelated changes. If a representative is needed, pull_request:1258 is the clearest true-duplicate candidate in the only exact same-change pair.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "pull_request:1258 is the best overall PR candidate because it is the cleaner, explicit version of the only near-identical change (negative_prompt handling in the pipeline test).", "soft_edge_verdicts": [ - { - "left": "pull_request:26", - "right": "pull_request:530", - "accept": false, - "reason": "Different bugs in different areas: DDPM scheduler math vs img2img test flakiness." - }, - { - "left": "pull_request:3831", - "right": "pull_request:4847", - "accept": false, - "reason": "Unrelated pipelines: DDPM inversion pipeline addition vs a WIP community prompt-free diffusion example." - }, - { - "left": "pull_request:4078", - "right": "pull_request:4910", - "accept": false, - "reason": "Different fixes in different docs/code paths: single-file accelerate dependency vs an inpainting docs typo." - }, - { - "left": "pull_request:4099", - "right": "pull_request:5695", - "accept": false, - "reason": "Both are SDXL-related, but one fixes prompt embedding handling and the other fixes PixArt mask-feature logic." - }, - { - "left": "pull_request:4262", - "right": "pull_request:4404", - "accept": false, - "reason": "Not the same change: DreamBooth LoRA training fixes vs removing textual-inversion mentions from SDXL docs." - }, { "left": "pull_request:4632", "right": "pull_request:5393", "accept": false, - "reason": "Both touch examples/docs, but the changes are unrelated checkpointing fixes vs HFApi tutorial updates." + "reason": "Different changes: SDXL LoRA checkpointing/casting bugfix vs unrelated HFApi docs/example updates." }, { "left": "pull_request:4783", "right": "pull_request:5813", "accept": false, - "reason": "Different scopes: SDXL CFG-cutoff docs vs a docs reorganization of loader APIs." + "reason": "One is SDXL ControlNet CFG-cutoff docs/code, the other is a loaders docs reorganization; same subsystem, not same change." }, { "left": "pull_request:4833", "right": "pull_request:6099", "accept": false, - "reason": "Unrelated LoRA-offload fix vs TextToVideoZero constructor/superclass argument fix." + "reason": "Unfuse LoRA bugfix vs TextToVideoZero constructor args; unrelated code paths." }, { "left": "pull_request:4910", "right": "pull_request:5407", "accept": false, - "reason": "Different problems: inpainting docs typo vs mixed-precision training fix for unconditional generation." + "reason": "Inpainting docs typo fix vs mixed-precision training example fix; not the same underlying bug." }, { "left": "pull_request:4927", "right": "pull_request:5615", "accept": false, - "reason": "Revert of offloading behavior is not the same as reverting load_lora logic in the SDXL adapter." + "reason": "Both are reverts, but they revert different features in different pipelines and address different problems." }, { "left": "pull_request:5085", "right": "pull_request:5506", "accept": false, - "reason": "Same broad LoRA/offload area, but different concrete bugs: incompatible LoRA offloading vs xformers/PEFT attention call error." + "reason": "LoRA offloading incompatibility fix vs xformers/PEFT attention scale argument fix; distinct issues." }, { "left": "pull_request:6208", "right": "pull_request:6787", "accept": false, - "reason": "One is a Stable Diffusion dtype crash fix; the other is a Marigold depth input-range fix." + "reason": "Stable Diffusion latent dtype mismatch vs Marigold input range normalization; unrelated." }, { "left": "pull_request:6841", "right": "pull_request:8702", "accept": false, - "reason": "Both are unrelated refactors: ADA layer norm params vs an SD3 conversion script." + "reason": "BasicTransformerBlock parameter restore vs SD3 conversion script; no shared code-path bug." }, { "left": "pull_request:6841", "right": "pull_request:9065", "accept": false, - "reason": "Different changes: AdaLayerNorm parameter restoration vs adding sentencepiece as a soft dependency." + "reason": "Ada-layer-norm parameter fix vs sentencepiece soft-dependency wiring; unrelated changes." }, { "left": "pull_request:6893", "right": "pull_request:8722", "accept": false, - "reason": "3D UNet attention setup refactor is unrelated to the embeddings module refactor." + "reason": "3D UNet attention refactor vs embeddings module refactor; broad architectural overlap only." }, { "left": "pull_request:7134", "right": "pull_request:7213", "accept": false, - "reason": "Type-hint cleanup for UNet is not the same as the LoRA merge docs update." + "reason": "UNet type-hint cleanup vs LoRA docs update; not the same artifact or bug." }, { "left": "pull_request:7632", "right": "pull_request:9569", "accept": false, - "reason": "Different training/runtime bugs: DreamBooth VAE loading vs CogVideoX autoencoder decode batching." + "reason": "DreamBooth VAE loading fix vs CogVideoX decode batch-size fix; different model paths." }, { "left": "pull_request:8685", "right": "pull_request:9918", "accept": false, - "reason": "Different docs additions: SD3 float8 note vs a server example guide." + "reason": "A documentation note about float8 vs a new server example; unrelated." }, { "left": "pull_request:9049", "right": "pull_request:9988", "accept": false, - "reason": "Both Flux-related, but one is a narrow encode_prompt dtype fix and the other is a large Flux Redux feature/conversion PR." + "reason": "Flux prompt-encoding fix when called separately vs broader Flux Redux conversion/architecture work; not mergeable as one bugfix." }, { "left": "pull_request:2040", "right": "pull_request:241", "accept": false, - "reason": "InstructPix2Pix pipeline addition and inpainting example script are separate features, not duplicates." + "reason": "InstructPix2Pix pipeline addition vs inpainting example script; different features." }, { "left": "pull_request:4617", "right": "pull_request:5057", "accept": false, - "reason": "Core encode_prompt refactor is broad and separate from SDXL clip-skip support." + "reason": "Encode-prompt refactor vs adding clip-skip support; same area but different change and not a single fix." }, { "left": "pull_request:8616", "right": "pull_request:9423", "accept": false, - "reason": "LoRA fusion for SD3 and Differential Diffusion for Kolors are unrelated model features." + "reason": "SD3 LoRA fusion support vs Kolors differential diffusion example; unrelated." }, { "left": "pull_request:9175", "right": "pull_request:9311", "accept": false, - "reason": "Flux ControlNet support and SDXL ControlNet CFG cutoff are different features/fixes." + "reason": "Flux ControlNet support vs SDXL ControlNet CFG cutoff callback; different pipelines and behaviors." }, { "left": "pull_request:10633", "right": "pull_request:11017", "accept": false, - "reason": "Flux docs example vs hybrid inference VAE encode feature; same docs area but not the same change." + "reason": "Flux docs IP-Adapter example vs VAE encode hybrid-inference feature; unrelated docs/features." }, { "left": "pull_request:11421", "right": "pull_request:11645", "accept": false, - "reason": "Different fixes: DreamBooth interpolation default vs setting torch version to N/A when torch is disabled." + "reason": "DreamBooth interpolation default change vs torch-version sentinel when torch is disabled; unrelated." + }, + { + "left": "pull_request:1256", + "right": "pull_request:1258", + "accept": true, + "reason": "Same tiny test change in the same file/test: both add negative_prompt handling to the pipeline test; 1258 is the clearer keyworded version." + }, + { + "left": "pull_request:12618", + "right": "pull_request:13164", + "accept": false, + "reason": "One is a uv install typo fix; the other is broader CI dependency pruning. Same workflow area, different change." + }, + { + "left": "pull_request:1282", + "right": "pull_request:2994", + "accept": false, + "reason": "Docs link fix vs ControlNet training type-casting fix; unrelated." + }, + { + "left": "pull_request:3137", + "right": "pull_request:3189", + "accept": false, + "reason": "UNet/class-label dtype cast vs AudioLDM waveform dtype fix; different code paths and symptoms." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative on the soft edges. The only accepted pair, 1256/1258, is plausibly the same bugfix in the same test file, while the other pairwise rejections match the shown diffs and titles. No obvious overstatement beyond minor wording around \u2018same-change,\u2019 which is acceptable here." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -168289,7 +168435,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 210, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -168299,43 +168445,43 @@ "nodes": [ "pull_request:10056", "pull_request:10630", + "pull_request:10983", "pull_request:11475", "pull_request:11485", + "pull_request:11588", "pull_request:11786", "pull_request:11979", "pull_request:11994", + "pull_request:11995", "pull_request:12041", + "pull_request:12249", "pull_request:12330", - "pull_request:1256", - "pull_request:1258", - "pull_request:12618", - "pull_request:1282", - "pull_request:13164", "pull_request:1327", "pull_request:1355", "pull_request:1458", "pull_request:1470", "pull_request:241", "pull_request:2557", - "pull_request:2994", - "pull_request:3137", - "pull_request:3189", "pull_request:4352", "pull_request:4461", "pull_request:4712", + "pull_request:4714", "pull_request:4793", "pull_request:5021", "pull_request:5085", "pull_request:5115", + "pull_request:5261", "pull_request:5299", "pull_request:5823", "pull_request:5880", "pull_request:6236", + "pull_request:6424", "pull_request:6673", "pull_request:7062", "pull_request:7158", "pull_request:7216", "pull_request:7632", + "pull_request:7924", "pull_request:821", "pull_request:823", "pull_request:8688", @@ -168347,10 +168493,6 @@ "pull_request:9838" ], "soft_pairs": [ - "pull_request:1256|pull_request:1258", - "pull_request:12618|pull_request:13164", - "pull_request:1282|pull_request:2994", - "pull_request:3137|pull_request:3189", "pull_request:4352|pull_request:4793", "pull_request:5021|pull_request:7216", "pull_request:5085|pull_request:5823", @@ -168371,166 +168513,197 @@ "pull_request:4461|pull_request:5115", "pull_request:7158|pull_request:8688", "pull_request:10630|pull_request:11786", - "pull_request:4712|pull_request:6673" + "pull_request:4712|pull_request:6673", + "pull_request:4714|pull_request:5261", + "pull_request:10983|pull_request:11588", + "pull_request:11995|pull_request:12249", + "pull_request:6424|pull_request:7924" ], - "prepared_review_unit_hash": "4cecf9b90b91fb6b7dd7aef66f3725d6eea3e0412d28c8f7e9720d8880c7c118", + "prepared_review_unit_hash": "4340fa2873ff93cc33cab18ec7ee44e1c1e7f7787b8d8e79b71565dcdafa7229", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 98828, - "estimated_input_tokens": 24707, - "estimated_eval_tokens": 49670 + "serialized_chars": 102064, + "estimated_input_tokens": 25516, + "estimated_eval_tokens": 51288 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated Diffusers PRs. The only close duplicate-like pair is the two negative-prompt test edits in `tests/test_pipelines.py` (1256/1258); the rest target different bugs, features, docs, CI, or subsystems.", + "summary": "This cluster is a heterogeneous mix of unrelated PRs across pipelines, LoRA/loading, schedulers, docs, CI, and tests. The soft pairs are mostly broad-subsystem similarities or shared filenames, not the same underlying bug or change, so none should be merged as duplicates.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 1258 is the cleaner version of the only genuine near-duplicate pair; it makes the same negative-prompt test change in a clearer, explicit form.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 1258 is the best representative of the cluster\u2019s only near-duplicate change, but the overall set is too heterogeneous for a strong canonical PR.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:1256", - "right": "pull_request:1258", - "accept": true, - "reason": "Same underlying negative-prompt test update in the same file; the second is just the clearer keyword-arg version." - }, - { - "left": "pull_request:12618", - "right": "pull_request:13164", - "accept": false, - "reason": "Different CI problems: uv install typo vs removing test deps from workflows." - }, - { - "left": "pull_request:1282", - "right": "pull_request:2994", - "accept": false, - "reason": "Unrelated changes: a docs link fix vs a ControlNet training type-casting bug." - }, - { - "left": "pull_request:3137", - "right": "pull_request:3189", - "accept": false, - "reason": "Different code paths and bugs: timestep/class-label dtype casting vs AudioLDM waveform dtype." - }, { "left": "pull_request:4352", "right": "pull_request:4793", "accept": false, - "reason": "Different functionality: Kandinsky combined offload/xformers helpers vs auto-pipeline kwargs propagation." + "reason": "Different problems: one adds combined Kandinsky pipeline helpers, the other fixes kwargs propagation in auto_pipeline. Same general area only." }, { "left": "pull_request:5021", "right": "pull_request:7216", "accept": false, - "reason": "Different changes: scheduler mapping docs vs from_single_file signature/docs expansion." + "reason": "Both touch conversion/mapping-related code, but one updates scheduler docs mapping and the other changes from_single_file signature/documentation. No shared bug." }, { "left": "pull_request:5085", "right": "pull_request:5823", "accept": false, - "reason": "Different LoRA issues: incompatible checkpoint offloading vs adding `adapter_names` to fuse_lora." + "reason": "Both involve LoRA loaders, but one fixes incompatible LoRA offloading behavior and the other adds adapter_names support for fuse_lora. Distinct changes." }, { "left": "pull_request:5299", "right": "pull_request:6236", "accept": false, - "reason": "Typehints/docs cleanup is unrelated to the ControlNet inpainting doc example fix." + "reason": "Both are small fixes, but one is typehints/docs in model utilities and the other corrects ControlNet inpainting example values. Not the same issue." }, { "left": "pull_request:5880", "right": "pull_request:7632", "accept": false, - "reason": "Different VAE-related fixes in different code paths: broad latent-diffusion compatibility vs DreamBooth VAE loading." + "reason": "Both concern VAE-related loading, but one is a broad latent-diffusion compatibility fix and the other is a specific DreamBooth VAE path issue. Different code paths." }, { "left": "pull_request:6673", "right": "pull_request:7062", "accept": false, - "reason": "Different artifacts entirely: adding InstantID pipeline vs fixing imagic docstrings." + "reason": "Both are community example/doc updates, but they address different pipeline docstrings and different functionality. No duplicate bug/change." }, { "left": "pull_request:8722", "right": "pull_request:8952", "accept": false, - "reason": "Core embeddings refactor is not the same as adding QKV fusion to specific transformer models." + "reason": "One is a large embeddings refactor; the other adds QKV fusion support. Same model area, but not the same change or bug." }, { "left": "pull_request:9099", "right": "pull_request:9638", "accept": false, - "reason": "Both involve ControlNet, but they address different models/features and are not the same concrete change." + "reason": "Both relate to ControlNet/SD3, but one adds SD3 controlnet inpainting while the other adds Xlabs controlnet support. Different features and code paths." }, { "left": "pull_request:9473", "right": "pull_request:9838", "accept": false, - "reason": "Different DreamBooth Flux training bugs in different scripts; they are not the same fix." + "reason": "Both touch Flux DreamBooth LoRA training, but one fixes LR handling for T5 in the base script and the other adjusts repetition/memory behavior in advanced training. Not the same bug." }, { "left": "pull_request:10056", "right": "pull_request:11485", "accept": false, - "reason": "Unrelated: Flux Redux prompt/multi-image support vs a new Flex.2 pipeline feature." + "reason": "One adds Flux Redux prompt/multi-image support; the other adds Flex.2 pipeline files. Different pipelines and functionality." }, { "left": "pull_request:11475", "right": "pull_request:11994", "accept": false, - "reason": "WAN recompilation fix is unrelated to hub-auth/docs CI changes." + "reason": "One fixes WAN recompilation behavior; the other fixes HF Hub login/test failures in docs and CI. No overlap in underlying issue." }, { "left": "pull_request:11979", "right": "pull_request:12330", "accept": false, - "reason": "Different goals: enforcing torch version in compile tests vs cleaning group-offloading tests." + "reason": "One enforces torch version for compile tests; the other cleans group offloading tests. Both are test maintenance only and unrelated." }, { "left": "pull_request:1327", "right": "pull_request:821", "accept": false, - "reason": "Custom local pipeline loading and an all-in-one Stable Diffusion pipeline are unrelated features." + "reason": "Both are early pipeline additions, but one improves loading local custom pipelines and the other adds a mega Stable Diffusion example. Different artifacts and goals." }, { "left": "pull_request:1458", "right": "pull_request:823", "accept": false, - "reason": "ONNX runtime candidate list change is unrelated to the negative-prompt type-mismatch fix." + "reason": "One adds an onnxruntime candidate package; the other fixes negative-prompt type mismatch across SD pipelines. Different subsystems." }, { "left": "pull_request:1470", "right": "pull_request:2557", "accept": false, - "reason": "DreamBooth AltDiffusion compatibility and intermediate logging are different changes." + "reason": "Both touch DreamBooth, but one adds AltDiffusion compatibility and the other adds intermediate logging. Different changes, not duplicate fixes." + }, + { + "left": "pull_request:11485", + "right": "pull_request:12041", + "accept": false, + "reason": "Different features: Flex.2 pipeline addition versus WAN VAE patch fix. Same broad release period only." + }, + { + "left": "pull_request:1355", + "right": "pull_request:241", + "accept": false, + "reason": "Both are pipeline additions, but one adds Flax img2img and the other adds an inpainting example script. Not the same underlying capability." + }, + { + "left": "pull_request:4461", + "right": "pull_request:5115", + "accept": false, + "reason": "Both modify conversion scripts, but one adds a pipeline_class_name argument and the other adds config_files parsing. Related tooling, different fixes." + }, + { + "left": "pull_request:7158", + "right": "pull_request:8688", + "accept": false, + "reason": "One fixes benchmark LCM test cleanup; the other fixes DoRA LoRA conversion. No shared bug or code path." + }, + { + "left": "pull_request:10630", + "right": "pull_request:11786", + "accept": false, + "reason": "Both are test/CI adjustments, but one updates Flax scheduler expected slices and the other changes the big accelerator marker. Different failures." }, { "left": "pull_request:4712", "right": "pull_request:6673", "accept": false, - "reason": "Tests fetcher infrastructure is unrelated to the InstantID pipeline addition." + "reason": "One adds a tests fetcher utility; the other adds InstantID pipeline docs/example. No common change." + }, + { + "left": "pull_request:4714", + "right": "pull_request:5261", + "accept": false, + "reason": "One changes DummyObject error handling for missing transformers; the other fixes callback step indexing in community schedulers. Unrelated." + }, + { + "left": "pull_request:10983", + "right": "pull_request:11588", + "accept": false, + "reason": "Both touch Hunyuan/Wan-style model support, but one adds Hunyuan I2V and the other adds WAN LoRA support for I2V. Different models and fix scopes." + }, + { + "left": "pull_request:11995", + "right": "pull_request:12249", + "accept": false, + "reason": "One adds modular Flux text-to-image support; the other adds Step1X-Edit pipeline and converter. Different pipeline families." + }, + { + "left": "pull_request:6424", + "right": "pull_request:7924", + "accept": false, + "reason": "Both are model/pipeline fixes, but one is a PixArt weight conversion bug and the other is an IP-Adapter conditional kwargs fix in SDXL ControlNet inpaint. Not the same issue." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, 1256/1258, is a plausible near-duplicate pair in the same test file with the same negative-prompt update, and the rest of the pairwise rejections are consistent with the differing files/themes shown in the packet. The canonicalization language is appropriately cautious given the heterogeneous cluster." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -168539,7 +168712,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 211, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -168549,20 +168722,21 @@ "nodes": [ "pull_request:1", "pull_request:10005", + "pull_request:10058", + "pull_request:10069", + "pull_request:10086", + "pull_request:101", "pull_request:10867", - "pull_request:10983", + "pull_request:10970", "pull_request:11130", "pull_request:11231", "pull_request:11564", - "pull_request:11588", "pull_request:11660", "pull_request:11797", - "pull_request:11995", "pull_request:12024", "pull_request:12074", "pull_request:12109", "pull_request:12119", - "pull_request:12249", "pull_request:12387", "pull_request:12584", "pull_request:12723", @@ -168578,15 +168752,14 @@ "pull_request:3386", "pull_request:3476", "pull_request:4098", - "pull_request:4714", - "pull_request:5261", + "pull_request:428", "pull_request:5298", "pull_request:5440", - "pull_request:6424", "pull_request:7", "pull_request:7359", "pull_request:7541", - "pull_request:7924", + "pull_request:8412", + "pull_request:8425", "pull_request:8640", "pull_request:8739", "pull_request:9099", @@ -168596,10 +168769,6 @@ "pull_request:9988" ], "soft_pairs": [ - "pull_request:4714|pull_request:5261", - "pull_request:10983|pull_request:11588", - "pull_request:11995|pull_request:12249", - "pull_request:6424|pull_request:7924", "pull_request:2385|pull_request:3351", "pull_request:5298|pull_request:5440", "pull_request:7359|pull_request:8739", @@ -168619,180 +168788,184 @@ "pull_request:9507|pull_request:9988", "pull_request:12897|pull_request:12919", "pull_request:1|pull_request:7", - "pull_request:10005|pull_request:11564" + "pull_request:10005|pull_request:11564", + "pull_request:10058|pull_request:8412", + "pull_request:10069|pull_request:10970", + "pull_request:10086|pull_request:8425", + "pull_request:101|pull_request:428" ], - "prepared_review_unit_hash": "a35c8ee960f72c64416cd22276d71be42619f98907cf72a8cbb50a3786b5d082", + "prepared_review_unit_hash": "dfff5b20b2c5b127c59986faed864f6d3c191804eca746b1044fe2e84f4c7be4", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 24, - "serialized_chars": 106971, - "estimated_input_tokens": 26743, - "estimated_eval_tokens": 53742 + "serialized_chars": 105234, + "estimated_input_tokens": 26309, + "estimated_eval_tokens": 52874 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: the PRs span unrelated features, docs, schedulers, loaders, quantization, and model additions across many different code paths. The soft-similarity links are mostly superficial (shared subsystem names or adjacent filenames) rather than true duplicate fixes.", + "summary": "These PRs are only superficially similar by subsystem or shared filenames. They cover distinct features, refactors, docs updates, and separate bug fixes, so none of the suggested soft edges should be merged as duplicates.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR fits this cluster; the artifacts are not the same change or bug, so any canonical PR choice would be arbitrary.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No clear global best PR: nothing here represents the whole set, and the PRs target unrelated models/features rather than one duplicated underlying change.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:4714", - "right": "pull_request:5261", - "accept": false, - "reason": "Different bugs: import error handling vs scheduler callback indexing." - }, - { - "left": "pull_request:10983", - "right": "pull_request:11588", - "accept": false, - "reason": "Different changes: Hunyuan I2V model integration vs LoRA loading for I2V." - }, - { - "left": "pull_request:11995", - "right": "pull_request:12249", - "accept": false, - "reason": "Different models/features: modular Flux pipeline refactor vs Step1X-Edit pipeline addition." - }, - { - "left": "pull_request:6424", - "right": "pull_request:7924", - "accept": false, - "reason": "Unrelated fixes: PixArt checkpoint conversion vs IP-Adapter conditioning in SDXL ControlNet inpaint." - }, { "left": "pull_request:2385", "right": "pull_request:3351", "accept": false, - "reason": "Utility method for Pix2Pix Zero is unrelated to adding Karras sigmas in LMSDiscreteScheduler." + "reason": "One adds a Pix2Pix Zero utility method; the other adds Karras sigmas to LMSDiscreteScheduler. Different components and unrelated changes." }, { "left": "pull_request:5298", "right": "pull_request:5440", "accept": false, - "reason": "LoRA conv torch.compile fix is unrelated to local single-file loading." + "reason": "These fix different LoRA/compile and single-file loading issues in different code paths; same broad area, but not the same bug." }, { "left": "pull_request:7359", "right": "pull_request:8739", "accept": false, - "reason": "Different community examples/scripts; not the same underlying bug or change." + "reason": "ResAdapter training script versus ConsistencyTTA community example. Both are community content, but the models and goals are unrelated." }, { "left": "pull_request:7541", "right": "pull_request:8640", "accept": false, - "reason": "Workflow dispatch changes are unrelated to relaxing a flax test tolerance." + "reason": "Workflow dispatch/release test changes versus a Flax scheduler test tolerance tweak. No shared underlying defect." }, { "left": "pull_request:2129", "right": "pull_request:2954", "accept": false, - "reason": "LoRA-from-pipeline support is unrelated to adding Text-to-Video Zero." + "reason": "One enables LoRA from pipeline; the other adds TextToVideoZeroPipeline. Both are pipeline additions, but not the same change or bug." }, { "left": "pull_request:9099", "right": "pull_request:9573", "accept": false, - "reason": "Both touch SD3, but one adds ControlNet inpainting while the other fixes dimension divisibility; different issues." + "reason": "Both touch SD3 controlnet/inpainting, but one adds SD3 controlnet inpainting support and the other fixes height/width divisibility. Different concrete issues." }, { "left": "pull_request:13063", "right": "pull_request:13208", "accept": false, - "reason": "Wan modular refactor is unrelated to adding Helios video generation pipelines." + "reason": "Wan modular pipeline refactor versus Helios video generation pipelines. Shared infra files do not indicate the same change." }, { "left": "pull_request:11130", "right": "pull_request:11797", "accept": false, - "reason": "Quantization config plumbing is not the same as implementing Higgs quantization." + "reason": "Pipeline-level quantization config is a general feature; Higgs quantization is a separate quantizer implementation and loading adjustment." }, { "left": "pull_request:10867", "right": "pull_request:9542", "accept": false, - "reason": "Both are Flux-adjacent LoRA work, but they solve different loading/conversion problems and are not mergeable into one fix." + "reason": "Multi IP-Adapter for Flux and Kohya Flux LoRA text-encoder support are distinct Flux loader changes, not one bug fix." }, { "left": "pull_request:2721", "right": "pull_request:3476", "accept": false, - "reason": "ModelEditing pipeline addition is unrelated to scheduler Karras-sigmas support." + "reason": "Model editing pipeline addition versus Karras sigmas for DPMSolverSinglestepScheduler. Different functionality entirely." }, { "left": "pull_request:3386", "right": "pull_request:4098", "accept": false, - "reason": "ControlNet refactor/additions are not the same concrete change as SDXL inpainting." + "reason": "ControlNet refactor/additions versus SDXL inpainting. They both expand pipelines, but they do not fix the same concrete problem." }, { "left": "pull_request:12109", "right": "pull_request:12119", "accept": false, - "reason": "Scheduler custom sigma support is unrelated to Qwen-Image Lightning LoRA loading." + "reason": "Custom sigma/timestep support in UniPCMultistepScheduler versus Qwen-Image-Lightning LoRA loading. Separate scheduler and loader work." }, { "left": "pull_request:12584", "right": "pull_request:12723", "accept": false, - "reason": "SANA-Video base inference and LongSANA minute-length video are different model/pipeline changes." + "reason": "Both are SANA-Video related, but one adds pretrained 5s inference and the other adds LongSANA minute-length generation with causal transformer changes. Not mergeable as one PR." }, { "left": "pull_request:11231", "right": "pull_request:12119", "accept": false, - "reason": "HiDream Image pipeline work is unrelated to Qwen-Image Lightning LoRA support." + "reason": "HiDream Image pipeline/model support versus Qwen-Image Lightning LoRA loading. Different models and code paths." }, { "left": "pull_request:11660", "right": "pull_request:12387", "accept": false, - "reason": "HiDream LoRA support and flash-attention backend support are unrelated." + "reason": "HiDream LoRA conversion support and FlashAttention-through-kernels support are unrelated features in different subsystems." }, { "left": "pull_request:12024", "right": "pull_request:12074", "accept": false, - "reason": "Wan follow-up fixes and Wan 2.2 LoRA loading are different code paths and not the same bug." + "reason": "Wan follow-up fixes versus Wan 2.2 second-transformer LoRA loading. Same model family, but different concrete behavior." }, { "left": "pull_request:9507", "right": "pull_request:9988", "accept": false, - "reason": "Flux ControlNet batching fix is unrelated to Flux Redux pipeline/model additions." + "reason": "Flux ControlNet control-mode batching fix versus Flux Redux conversion/runtime changes. Shared Flux branding only; not the same bug." }, { "left": "pull_request:12897", "right": "pull_request:12919", "accept": false, - "reason": "Both mention NPU, but one changes Flux pos-embed placement and the other adds NPU attention backend support." + "reason": "NPU pos-embed optimization for Flux versus NPU attention backend support. Different NPU code paths and different objectives." }, { "left": "pull_request:1", "right": "pull_request:7", "accept": false, - "reason": "Early GLIDE model files and VQ-VAE addition are separate foundational model imports." + "reason": "These are foundational early PRs adding GLIDE modeling and VQ-VAE support. Both are initial model bring-up, but not the same artifact." }, { "left": "pull_request:10005", "right": "pull_request:11564", "accept": false, - "reason": "Docs-only pipeline additions are unrelated to adding prodigy dependencies." + "reason": "Docs-only Flux pipeline additions versus a requirements update for DreamBooth. No shared underlying bug or feature." + }, + { + "left": "pull_request:10058", + "right": "pull_request:8412", + "accept": false, + "reason": "Both are docs fixes about single-file/loading or links, but they address different documentation pages and issues." + }, + { + "left": "pull_request:10069", + "right": "pull_request:10970", + "accept": false, + "reason": "BnB quantizer device handling versus PR style bot workflow swap. Completely unrelated." + }, + { + "left": "pull_request:10086", + "right": "pull_request:8425", + "accept": false, + "reason": "Stable Diffusion 3 Skip Guidance Layers bugfix versus a workflow name change. No overlap beyond being repository maintenance." + }, + { + "left": "pull_request:101", + "right": "pull_request:428", + "accept": false, + "reason": "Early core model/pipeline bootstrap PR versus general docs cleanup for using diffusers. Not the same change set." } ] }, @@ -168808,7 +168981,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 212, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -168816,10 +168989,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10058", - "pull_request:10069", - "pull_request:10086", - "pull_request:101", "pull_request:10109", "pull_request:10114", "pull_request:10115", @@ -168836,40 +169005,40 @@ "pull_request:10218", "pull_request:10220", "pull_request:10245", + "pull_request:10246", + "pull_request:10251", + "pull_request:1027", + "pull_request:10293", "pull_request:10358", "pull_request:10506", "pull_request:10730", - "pull_request:10970", "pull_request:11096", "pull_request:11129", "pull_request:11475", "pull_request:11531", "pull_request:11635", + "pull_request:11647", "pull_request:1885", "pull_request:2007", "pull_request:2322", "pull_request:3172", - "pull_request:428", + "pull_request:3230", "pull_request:628", "pull_request:70", - "pull_request:8412", - "pull_request:8425", "pull_request:8497", "pull_request:8600", "pull_request:8627", "pull_request:8774", + "pull_request:9021", "pull_request:9036", "pull_request:9116", "pull_request:9266", "pull_request:9286", + "pull_request:9709", "pull_request:9769", "pull_request:9893" ], "soft_pairs": [ - "pull_request:10058|pull_request:8412", - "pull_request:10069|pull_request:10970", - "pull_request:10086|pull_request:8425", - "pull_request:101|pull_request:428", "pull_request:10109|pull_request:9286", "pull_request:10114|pull_request:10730", "pull_request:10114|pull_request:11475", @@ -168894,210 +169063,221 @@ "pull_request:10218|pull_request:9036", "pull_request:10218|pull_request:9116", "pull_request:10220|pull_request:11635", - "pull_request:10245|pull_request:10358" + "pull_request:10245|pull_request:10358", + "pull_request:10246|pull_request:11647", + "pull_request:10251|pull_request:9021", + "pull_request:1027|pull_request:3230", + "pull_request:1027|pull_request:628", + "pull_request:10293|pull_request:9709" ], - "prepared_review_unit_hash": "b93e7046600f598d2a41ddc887edda5e2df7be2fe61dd6987308e4985412b780", + "prepared_review_unit_hash": "6dfc1f355b5426a481ac9cb65fa7fa4b730897116d54e3017bd83e8a327b651d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 29, - "serialized_chars": 98585, - "estimated_input_tokens": 24647, - "estimated_eval_tokens": 49550 + "soft_pair_count": 30, + "serialized_chars": 98746, + "estimated_input_tokens": 24687, + "estimated_eval_tokens": 49630 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All soft matches are superficial title/scope overlaps; the pairs span unrelated docs, CI, formatting, feature additions, and bug fixes. I would not collapse any of them as duplicates.", - "confidence": 0.98, + "summary": "This cluster is a set of mostly unrelated PRs that only share broad topical similarity (docs, CI, LoRA, new-model additions, small bugfixes). I don\u2019t see any pair that looks like the same underlying bug or change.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a duplicate canonical here: the candidates cover unrelated changes across docs, workflows, model/features, and tests, so there is no shared underlying fix/change to anchor on.", + "canonical_pr_reason": "No single PR is a credible duplicate anchor here; the cluster is heterogeneous and the similarities are too coarse to treat any PR as canonical.", "best_issue_reason": null, - "best_pr_reason": "PR 9769 ([core] Mochi T2V) is the clearest standalone, well-scoped feature PR with an explicit tracking issue and high review activity, so it is the strongest representative if one must be chosen.", + "best_pr_reason": "If forced to pick a representative, PR #10140 is the broadest substantial code addition, but it is still not a duplicate target for the rest of the cluster.", "soft_edge_verdicts": [ - { - "left": "pull_request:10058", - "right": "pull_request:8412", - "accept": false, - "reason": "Both are docs PRs, but one fixes a broken link while the other rewrites single-file loading docs; different underlying changes." - }, - { - "left": "pull_request:10069", - "right": "pull_request:10970", - "accept": false, - "reason": "Model-loading device validation and a PR-style-bot workflow update are unrelated code paths." - }, - { - "left": "pull_request:10086", - "right": "pull_request:8425", - "accept": false, - "reason": "StableDiffusion3 guidance-layer behavior and a workflow rename have no shared bug/change." - }, - { - "left": "pull_request:101", - "right": "pull_request:428", - "accept": false, - "reason": "A large early core-code PR and a broad docs refactor are not the same change." - }, { "left": "pull_request:10109", "right": "pull_request:9286", "accept": false, - "reason": "Stable Audio XLA support and exporting a new control-net class are distinct features." + "reason": "Different subsystems and goals: Stable Audio XLA support vs __init__ export cleanup." }, { "left": "pull_request:10114", "right": "pull_request:10730", "accept": false, - "reason": "Formatting cleanup in Stable Audio is unrelated to hook unwrapping behavior in pipeline_utils." + "reason": "Formatting-only Stable Audio edit vs hook-unwrapping bugfix in pipeline utils." }, { "left": "pull_request:10114", "right": "pull_request:11475", "accept": false, - "reason": "Stable Audio formatting changes do not overlap with WAN recompilation fixes." + "reason": "Stable Audio formatting cleanup vs WAN recompilation fix; unrelated code paths." }, { "left": "pull_request:10115", "right": "pull_request:8497", "accept": false, - "reason": "Flux positional-embedding bugfix and a spelling correction in a scheduler docstring are unrelated." + "reason": "Flux positional embedding bugfix vs spelling/doc cleanup in a scheduler; not the same change." }, { "left": "pull_request:10119", "right": "pull_request:10245", "accept": false, - "reason": "Both touch LoRA docs, but one documents load_lora_adapter while the other updates a delete_adapters tutorial; different API changes." + "reason": "Both are docs about LoRA APIs, but they cover different methods and different behavior." }, { "left": "pull_request:10126", "right": "pull_request:8774", "accept": false, - "reason": "Deprecating save_attn_procs and introducing LoraBaseMixin are different LoRA refactors with different code paths." + "reason": "Deprecating save_attn_procs is separate from the LoraBaseMixin refactor." }, { "left": "pull_request:10140", "right": "pull_request:9769", "accept": false, - "reason": "ConsisID and Mochi are separate model/pipeline feature additions, not the same underlying change." + "reason": "Different model launches (ConsisID vs Mochi); both are new feature additions, not duplicates." }, { "left": "pull_request:10148", "right": "pull_request:9893", "accept": false, - "reason": "OmniGen implementation and an SDXL ControlNet reference community pipeline are unrelated features." + "reason": "OmniGen implementation vs SDXL ControlNet reference community pipeline; distinct features." }, { "left": "pull_request:1017", "right": "pull_request:1885", "accept": false, - "reason": "Dreambooth argument typing and a typo fix in dreambooth inpaint use different files and problems." + "reason": "Unrelated DreamBooth script type-hint tweak vs a typo fix in DreamBooth inpaint." }, { "left": "pull_request:10188", "right": "pull_request:10506", "accept": false, - "reason": "Flux LoRA expansion error handling and a TorchAo sequential offload test are not the same bug or change." + "reason": "LoRA expansion logging/test fix vs TorchAO sequential offload test; different concerns." }, { "left": "pull_request:10197", "right": "pull_request:11096", "accept": false, - "reason": "CI TPU configuration changes and Flux TPU README updates are separate maintenance tasks." + "reason": "CI TPU workflow update vs Flux TPU README docs; different artifact types and changes." }, { "left": "pull_request:10197", "right": "pull_request:11129", "accept": false, - "reason": "Workflow updates and modeling import cleanup are unrelated." + "reason": "CI workflow changes vs modeling import cleanup; unrelated." }, { "left": "pull_request:10197", "right": "pull_request:11531", "accept": false, - "reason": "CI workflow edits and an inpaint doc variant example are different concerns." + "reason": "CI workflow changes vs inpaint documentation tweak; unrelated." }, { "left": "pull_request:10197", "right": "pull_request:8600", "accept": false, - "reason": "TPU CI changes and SD3 DreamBooth cleanup are unrelated." + "reason": "CI workflow changes vs DreamBooth script formatting; no shared bug/change." }, { "left": "pull_request:10197", "right": "pull_request:9266", "accept": false, - "reason": "CI workflow edits and adding a community project entry are unrelated." + "reason": "CI workflow changes vs community-projects docs entry; unrelated." }, { "left": "pull_request:102", "right": "pull_request:2007", "accept": false, - "reason": "Training example updates and a Black formatting sweep are not the same underlying change." + "reason": "Training README/example updates vs broad Black formatting sweep; not the same underlying change." }, { "left": "pull_request:102", "right": "pull_request:70", "accept": false, - "reason": "Example docs/script updates and a core attention/unet refactor are unrelated." + "reason": "Training example docs vs an old UNet/attention refactor; unrelated." }, { "left": "pull_request:1020", "right": "pull_request:2322", "accept": false, - "reason": "Training docs updates and an inpaint example PR address different documentation surfaces and topics." + "reason": "Training/fine-tuning docs update vs a separate inpaint example update." }, { "left": "pull_request:1020", "right": "pull_request:3172", "accept": false, - "reason": "Training docs updates and reproducibility docs are distinct documentation changes." + "reason": "Both are docs, but one is training docs and the other is reproducibility guidance; not the same change." }, { "left": "pull_request:1020", "right": "pull_request:628", "accept": false, - "reason": "Training/fine-tuning docs and a Colab table tweak do not reflect the same change." + "reason": "Different documentation targets and content; no shared bug or patch." }, { "left": "pull_request:10206", "right": "pull_request:8627", "accept": false, - "reason": "Flux Control LoRA unload support and SD3 ControlNet fast-test trimming are unrelated." + "reason": "Flux Control LoRA unload support vs SD3 ControlNet docs/model work; unrelated." }, { "left": "pull_request:10218", "right": "pull_request:9036", "accept": false, - "reason": "ControlNetUnion callback-input fixes and nightly dependency workflow changes are unrelated." + "reason": "ControlNetUnion callback tensor fix vs nightly dependency pinning; unrelated." }, { "left": "pull_request:10218", "right": "pull_request:9116", "accept": false, - "reason": "ControlNetUnion callback-input fixes and a dead-link docs patch are unrelated." + "reason": "ControlNetUnion callback inputs fix vs a dead-link doc fix; unrelated." }, { "left": "pull_request:10220", "right": "pull_request:11635", "accept": false, - "reason": "Renaming a Mochi integration test and adding Flux Kontext are separate changes." + "reason": "Mochi test rename vs Flux Kontext feature addition; unrelated." }, { "left": "pull_request:10245", "right": "pull_request:10358", "accept": false, - "reason": "LoRA adapter API docs and a community-project listing are unrelated documentation changes." + "reason": "PEFT docs cleanup vs community projects listing update; different docs areas and goals." + }, + { + "left": "pull_request:10246", + "right": "pull_request:11647", + "accept": false, + "reason": "Missing AttnProcessors docs/code exposure vs adding Finegrained FP8 quantization; unrelated." + }, + { + "left": "pull_request:10251", + "right": "pull_request:9021", + "accept": false, + "reason": "Regional prompting typo fix vs SD3 ControlNet transform-layer bugfix; unrelated." + }, + { + "left": "pull_request:1027", + "right": "pull_request:3230", + "accept": false, + "reason": "DDPM training example option addition vs large IF pipeline implementation; not the same change." + }, + { + "left": "pull_request:1027", + "right": "pull_request:628", + "accept": false, + "reason": "Both touch training-example adjacent content, but one adds an argument and the other updates README launch commands." + }, + { + "left": "pull_request:10293", + "right": "pull_request:9709", + "accept": false, + "reason": "TorchAO version floor bump vs torch upper-bound pinning for CI; different dependency issues." } ] }, @@ -169113,7 +169293,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 213, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -169121,10 +169301,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10246", - "pull_request:10251", - "pull_request:1027", - "pull_request:10293", "pull_request:10312", "pull_request:10318", "pull_request:10324", @@ -169142,40 +169318,38 @@ "pull_request:10544", "pull_request:10548", "pull_request:10554", + "pull_request:10576", "pull_request:10579", "pull_request:10651", "pull_request:10852", + "pull_request:10889", + "pull_request:10898", "pull_request:11004", "pull_request:11124", "pull_request:11196", "pull_request:11249", + "pull_request:11262", "pull_request:11319", "pull_request:11564", - "pull_request:11647", "pull_request:21", "pull_request:22", "pull_request:268", "pull_request:2759", "pull_request:3108", - "pull_request:3230", "pull_request:545", - "pull_request:628", "pull_request:828", "pull_request:886", "pull_request:8932", "pull_request:8964", "pull_request:8973", - "pull_request:9021", "pull_request:9077", + "pull_request:9244", "pull_request:9709", - "pull_request:9739" + "pull_request:9739", + "pull_request:9821", + "pull_request:9827" ], "soft_pairs": [ - "pull_request:10246|pull_request:11647", - "pull_request:10251|pull_request:9021", - "pull_request:1027|pull_request:3230", - "pull_request:1027|pull_request:628", - "pull_request:10293|pull_request:9709", "pull_request:10312|pull_request:10651", "pull_request:10318|pull_request:11004", "pull_request:10324|pull_request:11124", @@ -169202,90 +169376,66 @@ "pull_request:10548|pull_request:10554", "pull_request:10548|pull_request:9077", "pull_request:10548|pull_request:9739", - "pull_request:10554|pull_request:10852" + "pull_request:10554|pull_request:10852", + "pull_request:10554|pull_request:10889", + "pull_request:10554|pull_request:11262", + "pull_request:10576|pull_request:9821", + "pull_request:10576|pull_request:9827", + "pull_request:10579|pull_request:10898", + "pull_request:10579|pull_request:9244" ], - "prepared_review_unit_hash": "c2aec6a76d854a94923b9be276ef1eb3f2a911e924b4fe4099005bec7f6cae10", + "prepared_review_unit_hash": "4d9607cc9131acaaf5b1e5df2e3d90e8671b22c3635d6cf590ec22de29a0ccbc", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 32, - "serialized_chars": 91544, - "estimated_input_tokens": 22886, - "estimated_eval_tokens": 46028 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 33, + "serialized_chars": 90710, + "estimated_input_tokens": 22678, + "estimated_eval_tokens": 45612 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated PRs across docs, tests, CI, model/pipeline fixes, and feature additions. The soft edges mostly reflect superficial overlap in subsystem or filenames, not the same underlying bug/change.", - "confidence": 0.97, + "summary": "This cluster is mostly a grab bag of unrelated PRs (tests, docs, small fixes, and feature work). The only vaguely related area is LoRA/device-map handling, but the paired items are still fix-vs-test or different subproblems, so I would not merge any soft edges.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 11647 is the most substantial self-contained change in the set: a broad feature addition with code, docs, config, and tests, unlike the many tiny typo/test/maintenance patches.", + "canonical_pr_reason": "PR 9827 is the strongest canonical anchor if one must pick a representative: it is a substantive code fix around LoRA/device-map behavior, with broad downstream impact and an explicit problem statement.", "best_issue_reason": null, - "best_pr_reason": "PR 11647 is the strongest representative PR here because it is coherent and feature-complete, whereas the others are mostly narrow fixes or unrelated maintenance changes.", + "best_pr_reason": "PR 9827 is the best single PR candidate because it addresses a concrete runtime bug in shared loading logic, unlike the mostly test/doc-only items around it.", "soft_edge_verdicts": [ - { - "left": "pull_request:10246", - "right": "pull_request:11647", - "accept": false, - "reason": "Docs for missing attention processors vs a new Finegrained FP8 feature; unrelated changes." - }, - { - "left": "pull_request:10251", - "right": "pull_request:9021", - "accept": false, - "reason": "Regional prompting typo fix vs SD3 ControlNet layer-loading bug; different code paths and problems." - }, - { - "left": "pull_request:1027", - "right": "pull_request:3230", - "accept": false, - "reason": "DDPM training example CLI fix vs adding the IF pipeline; not the same change." - }, - { - "left": "pull_request:1027", - "right": "pull_request:628", - "accept": false, - "reason": "DDPM example CLI fix vs notebook link/docs update; only broadly training-example related." - }, - { - "left": "pull_request:10293", - "right": "pull_request:9709", - "accept": false, - "reason": "TorchAO minimum-version bump vs torch upper-bound pin; both dependency-related but different packages and fixes." - }, { "left": "pull_request:10312", "right": "pull_request:10651", "accept": false, - "reason": "Attention reshaping bug fix vs Lumina input validation change; unrelated issues." + "reason": "Different bugs in different pipelines: Attention reshape docs vs Lumina input validation." }, { "left": "pull_request:10318", "right": "pull_request:11004", "accept": false, - "reason": "Flux LoRA integration tests vs PEFT key-prefix handling; different LoRA concerns and code paths." + "reason": "One is a Flux LoRA integration test, the other changes PEFT prefix handling; related topic, not the same fix." }, { "left": "pull_request:10324", "right": "pull_request:11124", "accept": false, - "reason": "Kandinsky offload API fix vs CogView4 test tokenizer update; unrelated." + "reason": "Kandinsky offload test helper vs a CogView4 tokenizer test; unrelated changes." }, { "left": "pull_request:10332", "right": "pull_request:11564", "accept": false, - "reason": "Single-file shape error fix vs adding prodigy dependencies; completely different targets." + "reason": "Shape error message fix vs adding a dependency; no shared underlying change." }, { "left": "pull_request:10351", @@ -169297,133 +169447,169 @@ "left": "pull_request:10351", "right": "pull_request:8964", "accept": false, - "reason": "LoRA test cleanup vs AudioLDM2 generation cache fix; unrelated." + "reason": "Different code paths: LoRA fuse test vs AudioLDM2 cache-position fix." }, { "left": "pull_request:10364", "right": "pull_request:11249", "accept": false, - "reason": "Torch XLA support for Sana vs Flux Xlab ControlNet support; different models and features." + "reason": "Sana XLA support and Flux Xlab ControlNet support are different features in different pipelines." }, { "left": "pull_request:1041", "right": "pull_request:268", "accept": false, - "reason": "Both touch tests, but one is a failing-tests cleanup and the other changes expected outputs for slow tests." + "reason": "Both touch tests, but they fix different historical failures and different pipeline/model issues." }, { "left": "pull_request:1041", "right": "pull_request:2759", "accept": false, - "reason": "General test fixes vs a large run-tests refactor across many files; not the same bug." + "reason": "General test cleanup vs a large test-run plumbing change; not the same concrete bug." }, { "left": "pull_request:1041", "right": "pull_request:3108", "accept": false, - "reason": "Test-fix PR vs empty checkpoint-fix placeholder; no evidence of the same change." + "reason": "No evidence they address the same issue; one is a broad test fix, the other is an opaque checkpoint fix." }, { "left": "pull_request:1041", "right": "pull_request:886", "accept": false, - "reason": "General test cleanup vs autoencoder test fix; different artifacts." + "reason": "Different test fixes for different components; no common underlying bug." }, { "left": "pull_request:10420", "right": "pull_request:9077", "accept": false, - "reason": "LTX checkpoint-id docs fix vs typo cleanup across docs and code; unrelated." + "reason": "LTX checkpoint ID doc fix vs typo cleanup across unrelated files." }, { "left": "pull_request:1043", "right": "pull_request:21", "accept": false, - "reason": "Improved test assertions/logging vs adding model tests; different purposes." + "reason": "Better print helper vs a model test addition; not duplicate changes." }, { "left": "pull_request:1043", "right": "pull_request:22", "accept": false, - "reason": "Improved test assertions/logging vs GLIDESuperResUNetModel tests; unrelated." + "reason": "Better print helper vs GLIDE test coverage; unrelated." }, { "left": "pull_request:1043", "right": "pull_request:545", "accept": false, - "reason": "Test helper printing changes vs configuration logging changes; only superficial logging overlap." + "reason": "Test utility print formatting vs configuration logging behavior; different concerns." }, { "left": "pull_request:1043", "right": "pull_request:828", "accept": false, - "reason": "Better test print helpers vs Flax test suite completion; different changes." + "reason": "Test utility change vs Flax test completion; not the same bug or change." }, { "left": "pull_request:10481", "right": "pull_request:10537", "accept": false, - "reason": "Both touch Sana, but 10481 is only the clean_caption default fix while 10537 is a broader Sana-4K/pag update." + "reason": "Both involve Sana, but 10481 only flips a default while 10537 adds a broader Sana-4K update and extra pipeline changes." }, { "left": "pull_request:10497", "right": "pull_request:9709", "accept": false, - "reason": "Tokenizers install path fix in CI vs torch dependency pin; unrelated CI/dependency changes." + "reason": "Tokenizers install path fix vs torch version pin; unrelated CI adjustments." }, { "left": "pull_request:10502", "right": "pull_request:10579", "accept": false, - "reason": "Model output serialization metadata vs adding a community pipeline; unrelated." + "reason": "Serialized type name support in model outputs vs a community-project docs entry; unrelated." }, { "left": "pull_request:10506", "right": "pull_request:11196", "accept": false, - "reason": "TorchAO offload test vs save_pretrained serialization revert; unrelated." + "reason": "Sequential CPU offload regression test vs a save_pretrained serialization revert; not the same problem." }, { "left": "pull_request:10506", "right": "pull_request:8932", "accept": false, - "reason": "TorchAO offload test vs temporal transformer tests; no shared concrete bug." + "reason": "TorchAO offload test vs temporal transformer test coverage; different subsystems." }, { "left": "pull_request:10506", "right": "pull_request:8973", "accept": false, - "reason": "TorchAO offload test vs lumina test-size reduction; unrelated testing changes." + "reason": "TorchAO offload test vs Lumina test size reduction; unrelated." }, { "left": "pull_request:10544", "right": "pull_request:11319", "accept": false, - "reason": "ParaAttention docs vs community-project catalog entry; different documentation topics." + "reason": "ParaAttention docs vs community-project listing; no shared underlying change." }, { "left": "pull_request:10548", "right": "pull_request:10554", "accept": false, - "reason": "HunyuanVideo batch>1 attention-mask fix vs SD3 LoRA script fix; different models and bugs." + "reason": "HunyuanVideo batch-mask fix vs SD3 LoRA training script fix; different code paths and bugs." }, { "left": "pull_request:10548", "right": "pull_request:9077", "accept": false, - "reason": "HunyuanVideo transformer fix vs generic typo cleanup; unrelated." + "reason": "Transformer batch masking bug vs typo/doc cleanup; unrelated." }, { "left": "pull_request:10548", "right": "pull_request:9739", "accept": false, - "reason": "HunyuanVideo batch handling fix vs SD3.5 single-file docs example; unrelated." + "reason": "HunyuanVideo runtime fix vs documentation typo/update; unrelated." }, { "left": "pull_request:10554", "right": "pull_request:10852", "accept": false, - "reason": "SD3 LoRA loading bug fix vs issue-template label edit; unrelated." + "reason": "SD3 LoRA state-dict key fix vs issue-template wording; unrelated." + }, + { + "left": "pull_request:10554", + "right": "pull_request:10889", + "accept": false, + "reason": "SD3 LoRA loading bug vs fp16 unscaling in a different training script; not duplicate." + }, + { + "left": "pull_request:10554", + "right": "pull_request:11262", + "accept": false, + "reason": "SD3 LoRA loader fix vs Wan ftfy import fix; different models and failures." + }, + { + "left": "pull_request:10576", + "right": "pull_request:9821", + "accept": false, + "reason": "Regression test for 8-bit LoRA loading vs a broader device_map behavior fix; test-only versus implementation and not the same change." + }, + { + "left": "pull_request:10576", + "right": "pull_request:9827", + "accept": false, + "reason": "This is a test for LoRA loading in quantized models, while 9827 is the actual device_map fix; they are related but not the same PR-worthy change." + }, + { + "left": "pull_request:10579", + "right": "pull_request:10898", + "accept": false, + "reason": "Community-project docs entry vs remote_decode feature/docs/implementation; unrelated." + }, + { + "left": "pull_request:10579", + "right": "pull_request:9244", + "accept": false, + "reason": "Attentive eraser pipeline addition vs Flux single-file support; different features in different areas." } ] }, @@ -169439,7 +169625,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 214, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -169447,9 +169633,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10554", - "pull_request:10576", - "pull_request:10579", "pull_request:10622", "pull_request:10648", "pull_request:10676", @@ -169463,14 +169646,18 @@ "pull_request:10885", "pull_request:10889", "pull_request:1089", - "pull_request:10898", "pull_request:109", "pull_request:11039", + "pull_request:1104", + "pull_request:11044", + "pull_request:11087", "pull_request:11194", - "pull_request:11262", "pull_request:11681", "pull_request:11838", "pull_request:11841", + "pull_request:12005", + "pull_request:12142", + "pull_request:12143", "pull_request:1352", "pull_request:1489", "pull_request:164", @@ -169480,6 +169667,7 @@ "pull_request:2337", "pull_request:2428", "pull_request:289", + "pull_request:3230", "pull_request:34", "pull_request:774", "pull_request:839", @@ -169487,22 +169675,13 @@ "pull_request:87", "pull_request:9060", "pull_request:909", - "pull_request:9244", "pull_request:9442", "pull_request:9447", "pull_request:9449", "pull_request:9601", - "pull_request:9739", - "pull_request:9821", - "pull_request:9827" + "pull_request:9739" ], "soft_pairs": [ - "pull_request:10554|pull_request:10889", - "pull_request:10554|pull_request:11262", - "pull_request:10576|pull_request:9821", - "pull_request:10576|pull_request:9827", - "pull_request:10579|pull_request:10898", - "pull_request:10579|pull_request:9244", "pull_request:10622|pull_request:11841", "pull_request:10648|pull_request:11194", "pull_request:10648|pull_request:11838", @@ -169530,234 +169709,232 @@ "pull_request:109|pull_request:21", "pull_request:109|pull_request:87", "pull_request:11039|pull_request:9442", - "pull_request:11039|pull_request:9447" + "pull_request:11039|pull_request:9447", + "pull_request:1104|pull_request:3230", + "pull_request:11044|pull_request:12005", + "pull_request:11087|pull_request:12142", + "pull_request:11087|pull_request:12143" ], - "prepared_review_unit_hash": "6414ba4c119d93c1938566ac4ef7ade464f86713098a1b1ae333e9ed0003de24", + "prepared_review_unit_hash": "b9b87e4608558a0f953bc335b5760a56cc53c926b41a0265cfb414070d142a5d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 34, - "serialized_chars": 102042, - "estimated_input_tokens": 25511, - "estimated_eval_tokens": 51278 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 98575, + "estimated_input_tokens": 24644, + "estimated_eval_tokens": 49544 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated PRs grouped by superficial similarity. The only real substantive theme is LoRA/device_map handling, but even there the candidate pairs are either test-only, docs-only, or adjacent but distinct fixes rather than the same change.", - "confidence": 0.86, + "summary": "This cluster is very noisy and mostly unrelated: it mixes docs-only edits, test cleanups, scheduler/docs updates, LoRA/device-map fixes, and separate pipeline feature changes. I do not see any true duplicate pairs among the soft edges.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 9449 is the clearest representative of the substantive device_map/LoRA-loading subtheme: it contains a concrete code-path fix plus broad regression coverage, unlike the many docs/test/typo-only PRs.", + "canonical_pr_reason": "PR 9449 is the strongest representative of the cluster because it fixes a concrete runtime bug in LoRA loading with device-mapped models and spans the most substantive code paths among the candidates.", "best_issue_reason": null, - "best_pr_reason": "PR 9449 is the strongest overall representative because it fixes a real runtime bug, has broad test coverage, and is more central than the surrounding docs or cosmetic changes.", + "best_pr_reason": "PR 9449 is the best global PR representative: it addresses a real functional issue with clear user impact, whereas most other items are docs, tests, or unrelated feature/refactor changes.", "soft_edge_verdicts": [ - { - "left": "pull_request:10554", - "right": "pull_request:10889", - "accept": false, - "reason": "Different bugs in different training scripts: SD3 LoRA state loading vs fp16 unscaling in SDXL validation." - }, - { - "left": "pull_request:10554", - "right": "pull_request:11262", - "accept": false, - "reason": "Unrelated fixes: SD3 miniature LoRA loading path vs Wan ftfy import handling." - }, - { - "left": "pull_request:10576", - "right": "pull_request:9821", - "accept": false, - "reason": "One is a quantization test addition; the other is a device_map behavior fix. Same broad area, not the same change." - }, - { - "left": "pull_request:10576", - "right": "pull_request:9827", - "accept": false, - "reason": "Test-only LoRA loading coverage vs the actual device_map/LoRA loading fix; related theme, but not the same concrete patch." - }, - { - "left": "pull_request:10579", - "right": "pull_request:10898", - "accept": false, - "reason": "Completely different features: community pipeline addition vs remote_decode documentation/API." - }, - { - "left": "pull_request:10579", - "right": "pull_request:9244", - "accept": false, - "reason": "Different pipelines and problems: attentive eraser community script vs Flux single-file support." - }, { "left": "pull_request:10622", "right": "pull_request:11841", "accept": false, - "reason": "Both are docs, but about unrelated topics: virtualenv/uv installation vs batch generation documentation." + "reason": "Both are docs PRs, but one updates installation instructions while the other adds batch generation docs; unrelated changes." }, { "left": "pull_request:10648", "right": "pull_request:11194", "accept": false, - "reason": "Scheduler docs update vs component-specific torch_dtype docs; same docs domain, different content." + "reason": "Scheduler docs vs a torch_dtype mapping docs update; different topics and no shared bug/change." }, { "left": "pull_request:10648", "right": "pull_request:11838", "accept": false, - "reason": "Unrelated documentation changes: DDPM sigmoid scheduler docs vs deprecated pipelines notices." + "reason": "One is scheduler documentation, the other marks pipelines deprecated; not the same change." }, { "left": "pull_request:10676", "right": "pull_request:9060", "accept": false, - "reason": "README update for XLA/SPMD vs removing a debug print in FLUX transformer code; not the same issue." + "reason": "Different pipelines and different purposes: SPMD inference example vs a transformer debug-print cleanup." }, { "left": "pull_request:10698", "right": "pull_request:9601", "accept": false, - "reason": "Different bugs: inconsistent augmentations in instruct-pix2pix vs VAE dtype handling under mixed precision." + "reason": "Instruct Pix2Pix transform bugfix and SD3 VAE dtype handling are unrelated code-paths." }, { "left": "pull_request:1073", "right": "pull_request:34", "accept": false, - "reason": "Both are small text cleanups, but they touch unrelated files and fixes (README typo vs comment typo)." + "reason": "Dreambooth README typo fix vs a scheduler comment typo fix; purely separate documentation edits." }, { "left": "pull_request:10730", "right": "pull_request:11681", "accept": false, - "reason": "Hook unwrapping fix vs device_map documentation/tests clarification; related subsystem, but not the same code-path bug." + "reason": "Both touch pipeline_utils, but one fixes hook unwrapping and the other clarifies device_map docs/tests; not the same bug." }, { "left": "pull_request:1077", "right": "pull_request:774", "accept": false, - "reason": "Scheduler test refactor vs pipeline test expectation changes; different test changes, not one underlying bug." + "reason": "Scheduler test refactor vs pipeline test assertions; different test areas and changes." }, { "left": "pull_request:1077", "right": "pull_request:909", "accept": false, - "reason": "Both are test-related, but one refactors scheduler tests while the other changes pipeline assertions/models." + "reason": "These are unrelated test changes in different eras and different behaviors." }, { "left": "pull_request:10808", "right": "pull_request:9449", "accept": false, - "reason": "PEFT config rank-pattern adjustment is a different LoRA issue from device_mapped LoRA loading." + "reason": "PEFT config key restriction and LoRA loading with device-mapped models are distinct issues." }, { "left": "pull_request:10852", "right": "pull_request:9739", "accept": false, - "reason": "Issue template label fix vs docs for Stable Diffusion 3; unrelated maintenance changes." + "reason": "Issue template label change vs Stable Diffusion 3.5 single-file docs; unrelated." }, { "left": "pull_request:1086", "right": "pull_request:1489", "accept": false, - "reason": "Attention-module refactor vs removing a reminder comment in tests; unrelated changes." + "reason": "Attention module refactor vs a test comment removal; no shared underlying bug or change." }, { "left": "pull_request:1086", "right": "pull_request:2130", "accept": false, - "reason": "Model attention refactor vs pipeline allow_patterns cleanup; not the same underlying bug or feature." + "reason": "Attention refactor and allow_patterns cleanup are unrelated implementation changes." }, { "left": "pull_request:1086", "right": "pull_request:2188", "accept": false, - "reason": "Core attention implementation change vs docs license text removal." + "reason": "Model attention refactor vs inpaint docs license removal; unrelated." }, { "left": "pull_request:1086", "right": "pull_request:2337", "accept": false, - "reason": "Attention refactor vs removing unnecessary tests; different scope and purpose." + "reason": "Attention implementation change vs test removals; not the same fix." }, { "left": "pull_request:1086", "right": "pull_request:2428", "accept": false, - "reason": "Attention code refactor vs docs tweak for panorama examples; unrelated." + "reason": "Attention refactor and panorama docs cleanup are unrelated." }, { "left": "pull_request:1086", "right": "pull_request:289", "accept": false, - "reason": "Attention implementation refactor vs seed-handling refactor across pipelines; not the same concrete bug." + "reason": "Attention module refactor and seed-related pipeline refactor are different code changes." }, { "left": "pull_request:1086", "right": "pull_request:839", "accept": false, - "reason": "Model refactor vs README auth-token cleanup; unrelated." + "reason": "Model refactor vs DreamBooth README edit; no duplicate relation." }, { "left": "pull_request:1086", "right": "pull_request:842", "accept": false, - "reason": "Attention refactor vs test output-access cleanup; different changes." + "reason": "Attention refactor and pipeline test cleanup are unrelated." }, { "left": "pull_request:10885", "right": "pull_request:11194", "accept": false, - "reason": "CogVideoX scheduler docs vs torch_dtype mapping docs; unrelated documentation topics." + "reason": "CogVideoX scheduler docs and torch_dtype mapping docs are separate documentation additions." }, { "left": "pull_request:10885", "right": "pull_request:11838", "accept": false, - "reason": "Scheduler docs addition vs deprecated pipeline notices; not the same change." + "reason": "Scheduler docs addition vs deprecated-pipelines docs; different documentation goals." }, { "left": "pull_request:10889", "right": "pull_request:9739", "accept": false, - "reason": "fp16 unscaling fix in training vs Stable Diffusion 3 docs/example update; different artifacts and problems." + "reason": "Dreambooth fp16 validation fix and SD3.5 single-file docs are not the same change." + }, + { + "left": "pull_request:1089", + "right": "pull_request:164", + "accept": false, + "reason": "Stable Diffusion docs about scheduler usage vs an old scheduler schedule implementation change; different levels and concerns." }, { "left": "pull_request:109", "right": "pull_request:1352", "accept": false, - "reason": "Large model renaming refactor vs a docs sample-model swap; not the same underlying change." + "reason": "Major model renaming and AltDiffusion sample-model docs are unrelated." }, { "left": "pull_request:109", "right": "pull_request:21", "accept": false, - "reason": "Model renaming/refactor vs adding model tests; different work items." + "reason": "Big model renaming vs adding model tests; not a duplicate bug/change." }, { "left": "pull_request:109", "right": "pull_request:87", "accept": false, - "reason": "Both are early large refactors, but they target different model abstractions and files rather than one duplicate patch." + "reason": "Both are large refactors, but they address different model architecture migrations and cannot be merged as one concrete fix." }, { "left": "pull_request:11039", "right": "pull_request:9442", "accept": false, - "reason": "Single-file torch_dtype handling vs CI test/workflow changes; unrelated." + "reason": "Single-file dtype handling and CI behavior for deprecated attention tests are unrelated." }, { "left": "pull_request:11039", "right": "pull_request:9447", "accept": false, - "reason": "Loader dtype behavior vs pipeline save_pretrained shard-size default; different code paths." + "reason": "Single-file loader dtype behavior vs pipeline save_pretrained sharding defaults; different code paths." + }, + { + "left": "pull_request:1104", + "right": "pull_request:3230", + "accept": false, + "reason": "CI test additions for MPS and the IF model addition are entirely different changes." + }, + { + "left": "pull_request:11044", + "right": "pull_request:12005", + "accept": false, + "reason": "LoRA no-op warning text and sharded-checkpoint local_files_only handling are unrelated." + }, + { + "left": "pull_request:11087", + "right": "pull_request:12142", + "accept": false, + "reason": "Wan I2V quality fix vs Wan VACE bf16/fp16 mask casting; different pipeline variants and fixes." + }, + { + "left": "pull_request:11087", + "right": "pull_request:12143", + "accept": false, + "reason": "Wan I2V quality work and Wan VACE dtype handling are separate changes, not one bug fix." } ] }, @@ -169773,7 +169950,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 215, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -169781,9 +169958,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:1104", - "pull_request:11044", - "pull_request:11087", "pull_request:11095", "pull_request:11096", "pull_request:11113", @@ -169807,19 +169981,22 @@ "pull_request:11522", "pull_request:11529", "pull_request:11531", + "pull_request:11537", "pull_request:11546", + "pull_request:11556", + "pull_request:11564", "pull_request:11635", + "pull_request:11736", + "pull_request:11747", + "pull_request:11841", "pull_request:11897", "pull_request:11983", - "pull_request:12005", - "pull_request:12142", - "pull_request:12143", "pull_request:12244", "pull_request:12391", + "pull_request:12586", "pull_request:1278", "pull_request:1751", "pull_request:2009", - "pull_request:3230", "pull_request:42", "pull_request:580", "pull_request:703", @@ -169831,10 +170008,6 @@ "pull_request:9938" ], "soft_pairs": [ - "pull_request:1104|pull_request:3230", - "pull_request:11044|pull_request:12005", - "pull_request:11087|pull_request:12142", - "pull_request:11087|pull_request:12143", "pull_request:11095|pull_request:11350", "pull_request:11096|pull_request:11129", "pull_request:11096|pull_request:11531", @@ -169860,216 +170033,220 @@ "pull_request:11353|pull_request:11546", "pull_request:11366|pull_request:9938", "pull_request:1147|pull_request:2009", - "pull_request:11529|pull_request:11531" + "pull_request:11529|pull_request:11531", + "pull_request:11537|pull_request:11736", + "pull_request:11556|pull_request:11841", + "pull_request:11556|pull_request:12586", + "pull_request:11564|pull_request:11747" ], - "prepared_review_unit_hash": "a6cfb908e443fa609efb9249a64d2de6eb753a20baa65822bbc3fd69798eaec1", + "prepared_review_unit_hash": "c3c2569f39c679b11a7c561d1d0777c822f97fc74191e6afd0dd3e0ffeb18a8d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 97487, - "estimated_input_tokens": 24372, - "estimated_eval_tokens": 49000 + "serialized_chars": 95325, + "estimated_input_tokens": 23832, + "estimated_eval_tokens": 47920 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly a grab bag of unrelated PRs. The only strong duplicate-like match is the Wan VACE bf16/fp16 mask-dtype fix pair, but nearly all other soft edges are false positives across different features, docs, or pipelines.", - "confidence": 0.24, + "summary": "This cluster is a loose collection of unrelated pull requests, mostly small docs, CI, and model/pipeline fixes. I do not see a single underlying bug or change theme that would justify deduping any pair.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 12143 is the cleanest self-contained code fix in the set: it addresses the Wan VACE mask dtype bug and includes the related test adjustment, making it the best single PR anchor here.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 12143 is the strongest representative of a concrete bug fix among otherwise unrelated changes, and it is the only clearly duplicate-adjacent Wan VACE fix candidate in this cluster.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:1104", - "right": "pull_request:3230", - "accept": false, - "reason": "Unrelated changes: CI slow MPS tests vs. a new IF pipeline/docs feature." - }, - { - "left": "pull_request:11044", - "right": "pull_request:12005", - "accept": false, - "reason": "Both touch loaders/utilities, but one is a LoRA log-level tweak and the other is sharded-checkpoint offline behavior; different bugs." - }, - { - "left": "pull_request:11087", - "right": "pull_request:12142", - "accept": false, - "reason": "Same broad Wan area, but these are different pipeline fixes and not the same concrete bug." - }, - { - "left": "pull_request:11087", - "right": "pull_request:12143", - "accept": false, - "reason": "Same broad Wan area, but the changes target different code paths and are not clearly the same fix." - }, { "left": "pull_request:11095", "right": "pull_request:11350", "accept": false, - "reason": "Both edit controlnet_flux.py, but one is an example/doc correction while the other changes forward aggregation logic." + "reason": "Same file, but different problems: one fixes a pipeline example variable in docs, the other changes ControlNet union forward logic." }, { "left": "pull_request:11096", "right": "pull_request:11129", "accept": false, - "reason": "README wording update vs. modeling import cleanup; unrelated changes." + "reason": "README/docs wording for Flux TPU vs internal modeling import cleanup; unrelated changes." }, { "left": "pull_request:11096", "right": "pull_request:11531", "accept": false, - "reason": "Flux README updates and an inpaint doc variant fix are separate documentation changes." + "reason": "Both are docs, but one is Flux TPU README content and the other is an inpaint usage example; different topics." }, { "left": "pull_request:11113", "right": "pull_request:9506", "accept": false, - "reason": "Autoencoder training script fix and a doc/path tweak are unrelated." + "reason": "AutoencoderKL training bugfix vs unrelated docs/path correction; no shared code-path problem." }, { "left": "pull_request:11115", "right": "pull_request:11250", "accept": false, - "reason": "Implementation of AutoModel vs. AutoModel docs; related feature area, but not the same concrete change." + "reason": "Code export/import work for AutoModel vs adding AutoModel docs; the latter is documentation only." }, { "left": "pull_request:11129", "right": "pull_request:11531", "accept": false, - "reason": "Model import cleanup and inpaint docs are unrelated." + "reason": "Model import refactor vs inpaint documentation tweak; not the same underlying change." }, { "left": "pull_request:11129", "right": "pull_request:11983", "accept": false, - "reason": "Model import cleanup vs. installation docs update; different topics." + "reason": "Modeling import cleanup and installation docs edit are unrelated." }, { "left": "pull_request:11129", "right": "pull_request:9760", "accept": false, - "reason": "Model import cleanup vs. README wording cleanup; no shared bug or change." + "reason": "Model import changes vs README wording cleanup; no shared bug or fix." }, { "left": "pull_request:11237", "right": "pull_request:12391", "accept": false, - "reason": "Minor dtype-map docs edit and quantization docs restructure are different documentation changes." + "reason": "Minor dtype-map docs update vs broad quantization docs restructuring; unrelated documentation changes." }, { "left": "pull_request:11244", "right": "pull_request:11522", "accept": false, - "reason": "Both touch AudioLDM2, but they fix different issues: model type/offload handling vs. transformers main cache-position compatibility." + "reason": "Both touch AudioLDM2, but they fix different concrete issues: model typing/offload handling versus transformer-version cache-position compatibility." }, { "left": "pull_request:11250", "right": "pull_request:12244", "accept": false, - "reason": "AutoModel docs vs. Flax removal across the docs tree; unrelated." + "reason": "AutoModel docs addition vs removing Flax from many docs; different documentation scopes and goals." }, { "left": "pull_request:11257", "right": "pull_request:11635", "accept": false, - "reason": "A workflow token test and a Flux Kontext feature/docs PR are unrelated." + "reason": "Workflow token test change vs Flux Kontext feature implementation; completely different concerns." }, { "left": "pull_request:11260", "right": "pull_request:11897", "accept": false, - "reason": "Dummy modeling_utils edit vs. large modular-diffusers merge test; not the same change." + "reason": "Temporary modeling_utils dummy edit vs large modular-diffusers CI/docs merge test; unrelated." }, { "left": "pull_request:11261", "right": "pull_request:11269", "accept": false, - "reason": "VidTok autoencoder feature and onnxruntime package detection are unrelated." + "reason": "VidTok autoencoder support vs ONNX Runtime package detection updates; different subsystems and bugs." }, { "left": "pull_request:11262", "right": "pull_request:9739", "accept": false, - "reason": "Wan ftfy import fix and Stable Diffusion 3 typo/doc changes are different problems." + "reason": "Wan ftfy import guard vs Stable Diffusion 3 single-file doc/example fix; unrelated." }, { "left": "pull_request:113", "right": "pull_request:1751", "accept": false, - "reason": "Scheduler init fix vs. UnCLIP generator plumbing; different code paths." + "reason": "SDE-VE scheduler init bug vs UnCLIP generator plumbing; different schedulers and code paths." }, { "left": "pull_request:1130", "right": "pull_request:840", "accept": false, - "reason": "Community pipeline attention/offload enhancement vs. a one-step demo pipeline; unrelated." + "reason": "lpw_stable_diffusion offload/weighting fixes vs a dummy one-step U-Net example; unrelated." }, { "left": "pull_request:1133", "right": "pull_request:1278", "accept": false, - "reason": "Windows test workflow and HF hub RC testing are separate CI changes." + "reason": "Windows test workflow addition vs CI cache test workflow changes; different CI goals." }, { "left": "pull_request:1133", "right": "pull_request:42", "accept": false, - "reason": "Windows CI workflow vs. a modeling test cleanup; unrelated." + "reason": "Windows test workflow vs early modeling-utils test cleanup; no common bug." }, { "left": "pull_request:1133", "right": "pull_request:580", "accept": false, - "reason": "Windows CI workflow vs. a Flax scheduler test/implementation change; unrelated." + "reason": "Windows CI workflow vs Flax scheduler implementation/test work; unrelated." }, { "left": "pull_request:1133", "right": "pull_request:703", "accept": false, - "reason": "Windows CI workflow vs. CUDA cache CI; different test infrastructure changes." + "reason": "Windows test workflow vs CUDA cache CI changes; different test infra problems." }, { "left": "pull_request:1133", "right": "pull_request:885", "accept": false, - "reason": "Windows CI workflow vs. a DPM encoder WIP scheduler implementation; unrelated." + "reason": "Windows CI workflow vs new DPM encoder scheduler; unrelated." }, { "left": "pull_request:11353", "right": "pull_request:11546", "accept": false, - "reason": "Wan FLF2V support and Hunyuan LoRA loading are different features in different pipelines." + "reason": "Wan FLF2V feature support vs Hunyuan LoRA-loading support; different model families and changes." }, { "left": "pull_request:11366", "right": "pull_request:9938", "accept": false, - "reason": "Hidream transformer output shape fix vs. SDXL reference pipeline update; unrelated." + "reason": "HiDream transformer output-shape fix vs SDXL reference pipeline docs/example update; unrelated." }, { "left": "pull_request:1147", "right": "pull_request:2009", "accept": false, - "reason": "Flax example path changes vs. textual inversion loader support; related to loading, but not the same fix." + "reason": "Flax example loading-from-subfolder fix vs broad textual inversion loader integration; different feature scopes." }, { "left": "pull_request:11529", "right": "pull_request:11531", "accept": false, - "reason": "No-diff/placeholder PR vs. an actual inpaint doc update; not enough evidence of the same change." + "reason": "No meaningful diff is shown for 11529, and 11531 is a specific inpaint docs edit; insufficient evidence of sameness." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11736", + "accept": false, + "reason": "Torch.compile test coverage for UNet2DConditionModel vs quantizer compileability API; different layers and problems." + }, + { + "left": "pull_request:11556", + "right": "pull_request:11841", + "accept": false, + "reason": "Regional compilation docs vs batch inference docs; both documentation, but not the same underlying change." + }, + { + "left": "pull_request:11556", + "right": "pull_request:12586", + "accept": false, + "reason": "Regional compilation docs vs doc-toctree sorting; unrelated docs maintenance." + }, + { + "left": "pull_request:11564", + "right": "pull_request:11747", + "accept": false, + "reason": "Dreambooth FLUX dependency addition vs HiDream license header addition; unrelated PRs." } ] }, @@ -170085,7 +170262,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 216, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -170093,9 +170270,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:11537", - "pull_request:11556", - "pull_request:11564", "pull_request:11566", "pull_request:11635", "pull_request:11636", @@ -170107,23 +170281,23 @@ "pull_request:11703", "pull_request:11716", "pull_request:11728", - "pull_request:11736", "pull_request:11744", - "pull_request:11747", "pull_request:118", "pull_request:11812", "pull_request:1182", "pull_request:1184", - "pull_request:11841", "pull_request:11847", "pull_request:11869", "pull_request:11875", "pull_request:11891", "pull_request:11894", + "pull_request:11897", + "pull_request:119", "pull_request:11920", "pull_request:11955", "pull_request:12062", "pull_request:12109", + "pull_request:12115", "pull_request:12236", "pull_request:12244", "pull_request:12315", @@ -170137,16 +170311,15 @@ "pull_request:12705", "pull_request:12848", "pull_request:1713", + "pull_request:1868", "pull_request:1903", + "pull_request:1939", + "pull_request:217", "pull_request:38", "pull_request:59", "pull_request:75" ], "soft_pairs": [ - "pull_request:11537|pull_request:11736", - "pull_request:11556|pull_request:11841", - "pull_request:11556|pull_request:12586", - "pull_request:11564|pull_request:11747", "pull_request:11566|pull_request:11812", "pull_request:11635|pull_request:11920", "pull_request:11635|pull_request:11955", @@ -170172,216 +170345,220 @@ "pull_request:11869|pull_request:12236", "pull_request:11875|pull_request:12398", "pull_request:11891|pull_request:12360", - "pull_request:11894|pull_request:12062" + "pull_request:11894|pull_request:12062", + "pull_request:11897|pull_request:12115", + "pull_request:119|pull_request:1868", + "pull_request:119|pull_request:1939", + "pull_request:119|pull_request:217" ], - "prepared_review_unit_hash": "1ecbdde51400e447f47183143f9b5cbc79c3ce9d32d7b9a77aefcfedd52f950f", + "prepared_review_unit_hash": "c29606e53dabb03a84ee073c8338247d50a03008b5a7eb45ac90fb99dbb96368", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 99117, - "estimated_input_tokens": 24780, - "estimated_eval_tokens": 49816 + "serialized_chars": 98982, + "estimated_input_tokens": 24746, + "estimated_eval_tokens": 49748 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose similarity bucket of unrelated docs, tests, feature PRs, and maintenance changes. I found no pair that clearly reflects the same underlying bug or change, so all soft edges should be rejected.", - "confidence": 0.98, + "summary": "Mostly a noisy mixed cluster of unrelated Diffusers PRs. The only coherent theme is model/pipeline work around Chroma/Flux/Kontext, but the proposed soft pairs are generally adjacent feature, docs, or test changes rather than true duplicates.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 11698 is the broadest and most central substantive feature change in the cluster: it adds the Chroma pipeline/model stack, has the strongest integration footprint, and explicitly targets issue 11010.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 11698 is the best representative artifact because it is the largest real product change here and anchors the Chroma/Flux family work better than the smaller docs/test/follow-up PRs.", "soft_edge_verdicts": [ - { - "left": "pull_request:11537", - "right": "pull_request:11736", - "accept": false, - "reason": "UNet torch.compile test addition vs quantizer `is_compileable` property; different codepaths and purposes." - }, - { - "left": "pull_request:11556", - "right": "pull_request:11841", - "accept": false, - "reason": "Regional compilation docs vs batch inference docs; both docs, but unrelated topics." - }, - { - "left": "pull_request:11556", - "right": "pull_request:12586", - "accept": false, - "reason": "Regional compilation docs vs doc tree reordering; not the same change." - }, - { - "left": "pull_request:11564", - "right": "pull_request:11747", - "accept": false, - "reason": "Add prodigy dependency vs HiDream license header; unrelated maintenance." - }, { "left": "pull_request:11566", "right": "pull_request:11812", "accept": false, - "reason": "Chroma-as-FLUX variant conversion vs Flux Kontext pipeline feature; different models and logic." + "reason": "Both are FLUX-family changes, but one adds Chroma variant loading and the other adds the Flux Kontext pipeline; different artifacts and code paths." }, { "left": "pull_request:11635", "right": "pull_request:11920", "accept": false, - "reason": "Chroma pipeline docs/implementation vs Wan test slices; unrelated model families." + "reason": "Both are test-only PRs, but they cover different pipelines (Flux Kontext vs Wan) and unrelated behaviors." }, { "left": "pull_request:11635", "right": "pull_request:11955", "accept": false, - "reason": "Chroma pipeline work vs Cosmos test slices; same broad domain only." + "reason": "Different test additions for different pipelines (Flux Kontext vs Cosmos); same broad pattern, not the same change." }, { "left": "pull_request:11636", "right": "pull_request:11875", "accept": false, - "reason": "Community pipeline docs vs auto-pipeline hookup for Kontext; not the same concrete change." + "reason": "One is general docs for custom community pipelines; the other wires Kontext into auto-pipeline. Related topic, but not a duplicate change." }, { "left": "pull_request:11655", "right": "pull_request:11684", "accept": false, - "reason": "LoRA 8-bit bitsandbytes support vs GGUF single-file loading support; different loader paths." + "reason": "Both touch loader/quantization areas, but one adds bnb 8-bit LoRA support and the other adds GGUF single-file loading; distinct fixes." }, { "left": "pull_request:11660", "right": "pull_request:12109", "accept": false, - "reason": "HiDream LoRA conversion/support vs UniPC scheduler sigma support; unrelated subsystems." + "reason": "HiDream LoRA conversion work and UniPC scheduler custom sigma support are unrelated code changes." }, { "left": "pull_request:11681", "right": "pull_request:12315", "accept": false, - "reason": "device_map doc clarifications vs scheduler simplification in LoRA tests; not duplicates." + "reason": "Model/pipeline device_map docs versus LoRA test cleanup; no shared bug or implementation path." }, { "left": "pull_request:11698", "right": "pull_request:12398", "accept": false, - "reason": "Chroma pipeline feature vs Mirage pipeline feature; both add new models, but they are distinct pipelines and codepaths." + "reason": "Chroma pipeline integration and Mirage pipeline implementation are different feature areas despite overlapping __init__ export churn." }, { "left": "pull_request:11703", "right": "pull_request:12244", "accept": false, - "reason": "Quantization/compile docs vs removing Flax docs; unrelated documentation edits." + "reason": "Docs PR about quantization/compile guidance vs docs PR removing Flax; unrelated documentation changes." }, { "left": "pull_request:11716", "right": "pull_request:12424", "accept": false, - "reason": "Flux LoRA metadata fix vs Dockerfile definitions cleanup; unrelated changes." + "reason": "Flux LoRA metadata fix and Dockerfile cleanup are unrelated." }, { "left": "pull_request:11716", "right": "pull_request:12438", "accept": false, - "reason": "Flux LoRA metadata fix vs offload_state_dict handling in pipeline loading; separate bug fixes." + "reason": "LoRA metadata handling and offload_state_dict loading logic address different bugs." }, { "left": "pull_request:11728", "right": "pull_request:12586", "accept": false, - "reason": "Doc-builder quant libs Dockerfile vs doc sort in toctree; unrelated maintenance." + "reason": "Doc-builder Docker image dependencies and doc tree sorting are unrelated." }, { "left": "pull_request:11744", "right": "pull_request:12848", "accept": false, - "reason": "Dreambooth LoRA metadata/training changes vs adding ChromaInpaintPipeline; different feature areas." + "reason": "DreamBooth LoRA metadata handling and adding ChromaInpaintPipeline are separate features." }, { "left": "pull_request:118", "right": "pull_request:1903", "accept": false, - "reason": "Early ScoreSdeVe pipeline rename vs SD2 depth2img comment fixes; unrelated and far apart." + "reason": "Old pipeline rename/fix and SD2 depth2img comment reordering are not the same underlying change." }, { "left": "pull_request:118", "right": "pull_request:59", "accept": false, - "reason": "ScoreSdeVe pipeline rename vs ResNet merge; different historical changes." + "reason": "Different old-core refactors in unrelated pipelines/models." }, { "left": "pull_request:11812", "right": "pull_request:11875", "accept": false, - "reason": "Flux Kontext implementation/docs vs auto-pipeline entry addition; related project, but not the same concrete fix." + "reason": "Adding the Flux Kontext pipeline and registering it in auto-pipeline are related follow-ups, but not the same concrete change." }, { "left": "pull_request:11812", "right": "pull_request:12337", "accept": false, - "reason": "Flux Kontext feature vs xformers memory-efficient attention fix for Flux; different issues." + "reason": "Flux Kontext pipeline creation and xformers attention fix for Flux are different code-path fixes." }, { "left": "pull_request:1182", "right": "pull_request:1713", "accept": false, - "reason": "Long-prompt-weighting docs vs stable diffusion conversion cleanup; unrelated docs/code." + "reason": "Both are documentation-oriented, but one is LPW docs and the other is conversion script cleanup; not duplicates." }, { "left": "pull_request:1184", "right": "pull_request:38", "accept": false, - "reason": "Flash-attention integration vs unifying attention modules; same file family, but different change goals and eras." + "reason": "Both involve attention code historically, but one is flash-attention integration and the other is early attention module unification; different changes separated by scope and time." }, { "left": "pull_request:1184", "right": "pull_request:75", "accept": false, - "reason": "Flash-attention integration vs Glide attention fix; different attention bugs/features." + "reason": "Flash-attention work and Glide attention fix are distinct bug fixes." }, { "left": "pull_request:11847", "right": "pull_request:12637", "accept": false, - "reason": "Wan single-file example fix vs Sana video copyright/comment cleanup; unrelated." + "reason": "Wan single-file example docs and a copy-comment cleanup in Sana video are unrelated." }, { "left": "pull_request:11847", "right": "pull_request:12705", "accept": false, - "reason": "Wan single-file docs fix vs typo cleanup across assorted files; not the same issue." + "reason": "Wan single-file docs and typo fixes across many files are not the same change." }, { "left": "pull_request:11869", "right": "pull_request:12236", "accept": false, - "reason": "GGUF CUDA kernel support vs hub _flash_3 attention backend; both kernel-related, but distinct implementations." + "reason": "GGUF CUDA kernel support and hub flash-attn kernel support are both performance/backend work, but they target different mechanisms and code paths." }, { "left": "pull_request:11875", "right": "pull_request:12398", "accept": false, - "reason": "Kontext auto-pipeline wiring vs Mirage pipeline implementation; unrelated pipeline features." + "reason": "Kontext auto-pipeline registration and Mirage pipeline implementation are unrelated." }, { "left": "pull_request:11891", "right": "pull_request:12360", "accept": false, - "reason": "Chroma.md typo fix vs Chroma docs guidance rewrite; same docs area but different edits." + "reason": "Both are Chroma docs, but one fixes a usage typo and the other rewrites guidance text for guidance_scale; not a duplicate docs fix." }, { "left": "pull_request:11894", "right": "pull_request:12062", "accept": false, - "reason": "Pin k-diffusion for CI vs enable all GPUs in CI; unrelated CI maintenance." + "reason": "k-diffusion pinning for CI and enabling all GPUs in CI are unrelated infrastructure changes." + }, + { + "left": "pull_request:11897", + "right": "pull_request:12115", + "accept": false, + "reason": "A large test merge and an AnimateDiff sparse control test addition are different scopes and different failures." + }, + { + "left": "pull_request:119", + "right": "pull_request:1868", + "accept": false, + "reason": "Very old unrelated var-name fix vs EMA decay fix; not the same bug." + }, + { + "left": "pull_request:119", + "right": "pull_request:1939", + "accept": false, + "reason": "Very old unrelated fixes in hub utils vs README image path." + }, + { + "left": "pull_request:119", + "right": "pull_request:217", + "accept": false, + "reason": "Hub util variable-name fix and NSFW checker logic fix are unrelated." } ] }, @@ -170397,7 +170574,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 217, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -170405,7 +170582,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:11897", "pull_request:119", "pull_request:11953", "pull_request:11968", @@ -170419,8 +170595,9 @@ "pull_request:121", "pull_request:1210", "pull_request:1211", - "pull_request:12115", "pull_request:1215", + "pull_request:1218", + "pull_request:1219", "pull_request:12226", "pull_request:12467", "pull_request:12560", @@ -170434,30 +170611,27 @@ "pull_request:12984", "pull_request:1326", "pull_request:1415", + "pull_request:1610", "pull_request:1759", - "pull_request:1868", "pull_request:19", - "pull_request:1939", "pull_request:1971", "pull_request:2007", - "pull_request:217", "pull_request:2345", "pull_request:2701", "pull_request:3023", "pull_request:312", "pull_request:3343", + "pull_request:408", "pull_request:41", + "pull_request:423", "pull_request:433", "pull_request:498", "pull_request:693", "pull_request:70", - "pull_request:75" + "pull_request:75", + "pull_request:921" ], "soft_pairs": [ - "pull_request:11897|pull_request:12115", - "pull_request:119|pull_request:1868", - "pull_request:119|pull_request:1939", - "pull_request:119|pull_request:217", "pull_request:119|pull_request:312", "pull_request:119|pull_request:41", "pull_request:119|pull_request:498", @@ -170486,234 +170660,238 @@ "pull_request:1210|pull_request:3343", "pull_request:1211|pull_request:1326", "pull_request:1215|pull_request:2345", - "pull_request:1215|pull_request:3023" + "pull_request:1215|pull_request:3023", + "pull_request:1218|pull_request:1610", + "pull_request:1218|pull_request:423", + "pull_request:1219|pull_request:408", + "pull_request:1219|pull_request:921" ], - "prepared_review_unit_hash": "56bb10d2e460a823bf12b06b4379d2e3d3b6fcd6a1a35a256b8947b4110064a0", + "prepared_review_unit_hash": "20bd9f73142c4a89d5349012c60cf583546c716632fc30a70e72b8006cca41be", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 100000, - "estimated_input_tokens": 25000, - "estimated_eval_tokens": 50256 + "serialized_chars": 104498, + "estimated_input_tokens": 26125, + "estimated_eval_tokens": 52506 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated PRs were grouped by weak title/file similarity. I don\u2019t see a single coherent duplicate cluster here; almost all soft edges should be rejected.", + "summary": "This cluster is a grab bag of unrelated PRs (docs typos, small bug fixes, tests, modular pipeline work). None of the soft pairs appear to address the same concrete bug or change, so all should be rejected and there is no coherent canonical issue/PR.", "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No PR here is a true duplicate anchor; the set is heterogeneous. If forced, 11897 is only the broadest umbrella change, not a real canonical duplicate.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "11897 is the broadest artifact in the set, but it mixes test/docs/code updates and is not a strong representative for any duplicate family.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:11897", - "right": "pull_request:12115", - "accept": false, - "reason": "Test/docs mega-merge vs an AnimateDiff control-image test; different change and bug." - }, - { - "left": "pull_request:119", - "right": "pull_request:1868", - "accept": false, - "reason": "Tiny hub_utils variable-name fix vs EMA decay logic in training; unrelated code paths." - }, - { - "left": "pull_request:119", - "right": "pull_request:1939", - "accept": false, - "reason": "Hub utils fix vs README logo path fix; unrelated changes." - }, - { - "left": "pull_request:119", - "right": "pull_request:217", - "accept": false, - "reason": "Different files and bugs: hub_utils vs stable-diffusion safety checker." - }, { "left": "pull_request:119", "right": "pull_request:312", "accept": false, - "reason": "Different documentation/link cleanup topics; not the same bug/change." + "reason": "Different changes: a variable-name bug fix in hub_utils.py vs a README link/docs edit." }, { "left": "pull_request:119", "right": "pull_request:41", "accept": false, - "reason": "Variable-name typo fix vs README/comment formatting cleanup; unrelated." + "reason": "Unrelated fixes: model-card arg typo vs resnet comment cleanup." }, { "left": "pull_request:119", "right": "pull_request:498", "accept": false, - "reason": "Hub utils fix vs attention-slicing API bug; different code paths." + "reason": "Different code paths and problems; one is a model-card typo, the other is an attention-slicing API fix." }, { "left": "pull_request:119", "right": "pull_request:693", "accept": false, - "reason": "Hub utils fix vs BibTeX citation formatting; unrelated." + "reason": "README BibTeX citation formatting is unrelated to the hub_utils variable-name fix." }, { "left": "pull_request:119", "right": "pull_request:75", "accept": false, - "reason": "Hub utils bug vs attention module fix for Glide; unrelated." + "reason": "Different model code; attention implementation changes are not the same as the hub_utils typo fix." }, { "left": "pull_request:11953", "right": "pull_request:12686", "accept": false, - "reason": "Fast test slices for HiDream vs new Kandinsky5 pipeline feature; unrelated." + "reason": "Test-slice update for HiDream is unrelated to Kandinsky5 pipeline implementation." }, { "left": "pull_request:11968", "right": "pull_request:12644", "accept": false, - "reason": "Hook refactor/docs vs AutoModel docs; different scope and code paths." + "reason": "Hook refactor/shared constants and AutoModel docs are different artifacts with no shared bug/change." }, { "left": "pull_request:1198", "right": "pull_request:1415", "accept": false, - "reason": "CycleDiffusion docs/calling fix vs broad Stable Diffusion cleanup; not the same concrete bug." + "reason": "Both touch diffusion pipelines, but one fixes CycleDiffusion invocation while the other is a broad stable-diffusion cleanup; not the same concrete issue." }, { "left": "pull_request:11986", "right": "pull_request:12560", "accept": false, - "reason": "Different modular pipeline features for different models; not the same underlying change." + "reason": "Both are modular-pipeline work, but they target different features (LTX vs general modular updates) and are not mergeable as one fix." }, { "left": "pull_request:11986", "right": "pull_request:12628", "accept": false, - "reason": "LTX modular pipeline implementation vs a generic modular consistency check; unrelated." + "reason": "LTX modular pipeline implementation is unrelated to the modular block-name consistency check." }, { "left": "pull_request:11986", "right": "pull_request:12932", "accept": false, - "reason": "LTX pipeline implementation vs docstring cleanup for QwenImage modular blocks." + "reason": "Different Qwen/LTX modular-pipeline changes; one adds LTX pipeline code, the other only improves docstrings/types." }, { "left": "pull_request:11986", "right": "pull_request:12959", "accept": false, - "reason": "LTX pipeline work vs WAN I2V modular parameter additions; different features." + "reason": "LTX modular pipeline work is unrelated to WAN I2V parameter additions." }, { "left": "pull_request:120", "right": "pull_request:19", "accept": false, - "reason": "Both touch README, but one adds examples while the other renders latex; different changes." + "reason": "Both touch README, but one adds examples and the other renders math/formatting; not the same underlying change." }, { "left": "pull_request:120", "right": "pull_request:2007", "accept": false, - "reason": "README examples vs broad black-formatting sweep; unrelated." + "reason": "README examples and black formatting updates are unrelated." }, { "left": "pull_request:120", "right": "pull_request:70", "accept": false, - "reason": "README examples vs attention/mid-block model refactor; unrelated." + "reason": "Main README examples vs attention-model refactor are completely different changes." }, { "left": "pull_request:12018", "right": "pull_request:12467", "accept": false, - "reason": "Docs link wording fix vs QwenImage batch-image pipeline bug; unrelated." + "reason": "A docs link fix is unrelated to a QwenImage batch-handling code fix." }, { "left": "pull_request:12030", "right": "pull_request:12628", "accept": false, - "reason": "Add IP-Adapter support vs modular block-name validation; unrelated." + "reason": "Flux IP-Adapter support and modular pipeline validation are different features/problems." }, { "left": "pull_request:12030", "right": "pull_request:12846", "accept": false, - "reason": "IP-Adapter feature in Flux fill vs AutoPipeline text2video support; different features." + "reason": "Flux IP-Adapter support is unrelated to adding AutoPipelineForText2Video." }, { "left": "pull_request:12030", "right": "pull_request:12848", "accept": false, - "reason": "IP-Adapter support vs Chroma inpaint pipeline addition; unrelated features." + "reason": "Flux IP-Adapter support is unrelated to adding ChromaInpaintPipeline." }, { "left": "pull_request:12030", "right": "pull_request:12984", "accept": false, - "reason": "Flux fill IP-Adapter work vs Flux2 Klein docs/pipeline; unrelated." + "reason": "Flux IP-Adapter support and Flux2 Klein docs/code are different changes." }, { "left": "pull_request:1208", "right": "pull_request:1759", "accept": false, - "reason": "OneFlow/DPMsolver support vs xformers attn2 None fix; unrelated." + "reason": "One adds OneFlow/DPMsolver support; the other only guards attn2=None in xformers handling." }, { "left": "pull_request:1208", "right": "pull_request:2701", "accept": false, - "reason": "OneFlow/DPMsolver support vs onnxruntime-rocm import detection; unrelated." + "reason": "OneFlow/DPMsolver support is unrelated to ONNX Runtime ROCm support." }, { "left": "pull_request:12099", "right": "pull_request:12226", "accept": false, - "reason": "QwenImage device typo fix vs adding a controlnet test; different issues." + "reason": "A device-variable typo fix in QwenImage is unrelated to a new QwenImage ControlNet test." }, { "left": "pull_request:121", "right": "pull_request:1971", "accept": false, - "reason": "README examples vs PNDM docstring cleanup; unrelated." + "reason": "README examples and a scheduler docstring cleanup are not the same underlying change." }, { "left": "pull_request:121", "right": "pull_request:433", "accept": false, - "reason": "README examples vs black-version update and code formatting sweep; unrelated." + "reason": "README examples and broad black-style formatting changes are unrelated." }, { "left": "pull_request:1210", "right": "pull_request:3343", "accept": false, - "reason": "CycleDiffusion slow-test fix vs DPM solver config fix; different bugs." + "reason": "Slow-test fixes for cycle diffusion/schedulers are unrelated to a DPMSolver config-doc fix." }, { "left": "pull_request:1211", "right": "pull_request:1326", "accept": false, - "reason": "Flax usage comment cleanup vs gpu_id offload fix; unrelated." + "reason": "Flax usage-comment cleanup is unrelated to gpu_id handling in sequential CPU offload." }, { "left": "pull_request:1215", "right": "pull_request:2345", "accept": false, - "reason": "DDIM eta/cpu-generator bug fix vs xformers warning message; unrelated." + "reason": "DDIM eta/cpu-generator logic fix is unrelated to xformers warning messaging in training scripts." }, { "left": "pull_request:1215", "right": "pull_request:3023", "accept": false, - "reason": "DDIM generator bug vs attention processor API change; unrelated code paths." + "reason": "DDIM variance-noise logic and attention head/batch-dim changes are different code paths and problems." + }, + { + "left": "pull_request:1218", + "right": "pull_request:1610", + "accept": false, + "reason": "Conversion-script scheduler support is unrelated to attention logging/test changes." + }, + { + "left": "pull_request:1218", + "right": "pull_request:423", + "accept": false, + "reason": "A conversion script update is not the same as general output/docs syntax cleanup." + }, + { + "left": "pull_request:1219", + "right": "pull_request:408", + "accept": false, + "reason": "DDIM docs edits and general docs typo fixes are too broad and not the same specific change." + }, + { + "left": "pull_request:1219", + "right": "pull_request:921", + "accept": false, + "reason": "DDIM docs only vs a large v1-5 docs update; not a duplicate change." } ] }, @@ -170729,7 +170907,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 218, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -170737,8 +170915,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:1218", - "pull_request:1219", "pull_request:1221", "pull_request:12236", "pull_request:12275", @@ -170761,35 +170937,34 @@ "pull_request:12637", "pull_request:12675", "pull_request:12686", + "pull_request:12687", "pull_request:12705", + "pull_request:12782", "pull_request:12846", "pull_request:12848", + "pull_request:12868", "pull_request:12883", "pull_request:12919", "pull_request:12932", + "pull_request:12984", "pull_request:13038", "pull_request:13055", "pull_request:13100", "pull_request:13146", + "pull_request:13150", "pull_request:13158", "pull_request:13164", + "pull_request:13174", "pull_request:13259", + "pull_request:13275", "pull_request:13331", "pull_request:13344", "pull_request:13443", - "pull_request:1610", "pull_request:2631", "pull_request:393", - "pull_request:408", - "pull_request:423", - "pull_request:427", - "pull_request:921" + "pull_request:427" ], "soft_pairs": [ - "pull_request:1218|pull_request:1610", - "pull_request:1218|pull_request:423", - "pull_request:1219|pull_request:408", - "pull_request:1219|pull_request:921", "pull_request:1221|pull_request:393", "pull_request:12236|pull_request:12919", "pull_request:12275|pull_request:12305", @@ -170821,252 +170996,286 @@ "pull_request:12637|pull_request:13038", "pull_request:12637|pull_request:13443", "pull_request:12675|pull_request:13158", - "pull_request:12686|pull_request:13344" + "pull_request:12686|pull_request:13344", + "pull_request:12687|pull_request:13275", + "pull_request:12705|pull_request:13038", + "pull_request:12705|pull_request:13443", + "pull_request:12782|pull_request:13150", + "pull_request:12846|pull_request:12848", + "pull_request:12846|pull_request:12984", + "pull_request:12846|pull_request:13259", + "pull_request:12848|pull_request:12984", + "pull_request:12848|pull_request:13259", + "pull_request:12868|pull_request:13174" ], - "prepared_review_unit_hash": "d67bb36e664a98027a7b13bf70a4676a7fa3979ee00720e477a2577284c3c779", + "prepared_review_unit_hash": "ebd167387e5c1afdc49acfdc829c7b397cdd945ce99d2fcde236c3dc03178196", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 36, - "serialized_chars": 105702, - "estimated_input_tokens": 26426, - "estimated_eval_tokens": 53108 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 42, + "serialized_chars": 107761, + "estimated_input_tokens": 26941, + "estimated_eval_tokens": 54138 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a duplicate cluster; it\u2019s a grab bag of unrelated PRs spanning docs typos, CI, modular-pipeline plumbing, attention backends, and separate model/pipeline features. The soft links mostly reflect superficial topical overlap, not the same underlying bug or change.", - "confidence": 0.97, + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs, CI, modular-pipeline cleanup, model feature additions, and separate bug fixes. I would not merge any of the soft pairs; they only share broad subsystem labels or filenames.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a clear canonical representative: the set spans unrelated changes across docs, CI, modular pipelines, attention backends, and new model/pipeline features.", + "canonical_pr_reason": "No canonical PR: the items are only loosely related by topic and do not fix the same concrete change.", "best_issue_reason": null, - "best_pr_reason": "No PR is a strong global duplicate anchor because none of the items appear to address the same concrete code-path problem.", + "best_pr_reason": "No single PR is a strong global representative because the cluster spans unrelated docs, CI, feature work, and model fixes.", "soft_edge_verdicts": [ - { - "left": "pull_request:1218", - "right": "pull_request:1610", - "accept": false, - "reason": "Conversion-script work and pipeline test cleanup are unrelated." - }, - { - "left": "pull_request:1218", - "right": "pull_request:423", - "accept": false, - "reason": "A Stable Diffusion conversion script and general output/doc syntax cleanup are different changes." - }, - { - "left": "pull_request:1219", - "right": "pull_request:408", - "accept": false, - "reason": "DDIM docs and broad typo/docs cleanup are not the same documentation change." - }, - { - "left": "pull_request:1219", - "right": "pull_request:921", - "accept": false, - "reason": "Both touch docs, but they cover different generations of pipeline docs and different content." - }, { "left": "pull_request:1221", "right": "pull_request:393", "accept": false, - "reason": "Weight-reuse docs and text inversion training docs are unrelated topics." + "reason": "Weight reuse docs vs textual inversion docs: unrelated documentation changes." }, { "left": "pull_request:12236", "right": "pull_request:12919", "accept": false, - "reason": "Both touch attention_dispatch, but one adds Hub FA3 support and the other adds NPU Ulysses attention; different code paths." + "reason": "Both touch attention code, but they fix different backends and different code paths." }, { "left": "pull_request:12275", "right": "pull_request:12305", "accept": false, - "reason": "TorchAO config support and single-file allocator warmup solve different problems." + "reason": "TorchAO config support and single-file allocator warmup are unrelated features." }, { "left": "pull_request:12307", "right": "pull_request:12637", "accept": false, - "reason": "Modular-pipeline fixes and a Sana video copy cleanup are unrelated." + "reason": "Modular pipeline tweaks vs Sana video copy cleanup; different problems and files." }, { "left": "pull_request:12307", "right": "pull_request:12705", "accept": false, - "reason": "A modular-pipeline behavior change and a typo/docs sweep are not the same bug." + "reason": "One is modular-pipeline logic, the other is typo cleanup across docs/examples." }, { "left": "pull_request:12307", "right": "pull_request:13146", "accept": false, - "reason": "Modular pipeline updates and a tokenizer type fix are separate changes." + "reason": "Modular pipeline fixes vs a tokenizer-type cleanup in PAG hunyuan code." }, { "left": "pull_request:12337", "right": "pull_request:12398", "accept": false, - "reason": "Flux xformers attention fix and Mirage pipeline introduction are unrelated." + "reason": "xFormers attention enablement and Mirage pipeline work are unrelated." }, { "left": "pull_request:12337", "right": "pull_request:12637", "accept": false, - "reason": "Flux attention backend fix and Sana video copy cleanup do not address the same issue." + "reason": "Different pipelines and different fixes; only the broad theme is code cleanup." }, { "left": "pull_request:12337", "right": "pull_request:13146", "accept": false, - "reason": "Flux attention fix and MT5/T5 tokenizer typing are unrelated." + "reason": "Flux attention fix and Hunyuan tokenizer cleanup are distinct changes." }, { "left": "pull_request:12360", "right": "pull_request:12527", "accept": false, - "reason": "Chroma docs wording and Kandinsky5 CFG logic fix are different changes." + "reason": "Chroma docs edits and Kandinsky5 no-CFG fix are different artifacts." }, { "left": "pull_request:12360", "right": "pull_request:12637", "accept": false, - "reason": "Chroma docs updates and Sana video copy cleanup are unrelated." + "reason": "Docs wording changes do not match the Sana copy-fix PR." }, { "left": "pull_request:12360", "right": "pull_request:13146", "accept": false, - "reason": "Chroma docs and tokenizer type fixes are not the same underlying issue." + "reason": "Chroma docs and PAG type cleanup are unrelated." }, { "left": "pull_request:12398", "right": "pull_request:13331", "accept": false, - "reason": "Mirage pipeline addition and CogVideoX inpaint pipeline addition are different features." + "reason": "Different model families and different pipeline additions; not the same feature." }, { "left": "pull_request:1240", "right": "pull_request:2631", "accept": false, - "reason": "OneFlow pipeline implementation and training-loop process checks are unrelated." + "reason": "OneFlow image-to-image port and training main-process checks are unrelated." }, { "left": "pull_request:1240", "right": "pull_request:427", "accept": false, - "reason": "OneFlow image-to-image support and test cleanup are not the same change." + "reason": "Both involve tests/pipelines, but the fixes target different behaviors." }, { "left": "pull_request:12434", "right": "pull_request:12467", "accept": false, - "reason": "Dependency version bump and QwenImage batch handling fix are unrelated." + "reason": "PEFT version pinning and QwenImage batch handling are unrelated changes." }, { "left": "pull_request:12434", "right": "pull_request:12487", "accept": false, - "reason": "A requirements update and broken-link docs cleanup are different issues." + "reason": "Training requirement pin vs broken-link docs cleanup; no shared bug." }, { "left": "pull_request:12434", "right": "pull_request:12552", "accept": false, - "reason": "Requirements pin change and typo fixes in docs are unrelated." + "reason": "PEFT requirement bump and Kandinsky5 docs typo are unrelated." }, { "left": "pull_request:12491", "right": "pull_request:13164", "accept": false, - "reason": "Both are CI-related, but one adds link checking and the other removes test deps; not the same bug." + "reason": "Both touch CI, but one adds link checking while the other trims test deps." }, { "left": "pull_request:12527", "right": "pull_request:12883", "accept": false, - "reason": "Wan CFG fix and torchao docs typo fix are unrelated." + "reason": "Kandinsky5 no-CFG fix and torchao typo correction address different issues." }, { "left": "pull_request:12541", "right": "pull_request:13055", "accept": false, - "reason": "LoRA conversion compatibility and Wan patchification/device-selection fixes are distinct issues." + "reason": "LoRA conversion compatibility and Wan transformer fallback fix different code paths." }, { "left": "pull_request:12560", "right": "pull_request:12628", "accept": false, - "reason": "Both are modular-pipeline changes, but one is a broad update and the other adds a specific validation check; not the same bug." + "reason": "General modular updates vs an added consistency check; related area, different change." }, { "left": "pull_request:12560", "right": "pull_request:12932", "accept": false, - "reason": "Broad modular updates and docstring improvements are not the same change." + "reason": "Code-path modular updates and docstring cleanup are not the same fix." }, { "left": "pull_request:12578", "right": "pull_request:13100", "accept": false, - "reason": "Early erroring in auto CPU offload and modular pipeline documentation updates are unrelated." + "reason": "Auto CPU offload error handling and modular docs update are unrelated." }, { "left": "pull_request:12628", "right": "pull_request:12846", "accept": false, - "reason": "A modular block-name consistency check and AutoPipeline for text2video are unrelated." + "reason": "Modular pipeline validation and AutoPipelineForText2Video are different features." }, { "left": "pull_request:12628", "right": "pull_request:12848", "accept": false, - "reason": "Block-name validation in modular pipeline code and adding ChromaInpaintPipeline are different changes." + "reason": "A modular check and Chroma inpaint pipeline addition do not share the same bug." }, { "left": "pull_request:12628", "right": "pull_request:12932", "accept": false, - "reason": "A runtime consistency check and QwenImage docstrings do not fix the same problem." + "reason": "Structural modular validation vs docstring improvements only overlap broadly." }, { "left": "pull_request:12628", "right": "pull_request:13259", "accept": false, - "reason": "A modular pipeline validation bug and AGENTS.md/process/docs changes are unrelated." + "reason": "Pipeline validation code and repository AGENTS docs are unrelated." }, { "left": "pull_request:12637", "right": "pull_request:13038", "accept": false, - "reason": "Sana video copy cleanup and Wan patchification fixes are different issues." + "reason": "Sana copy cleanup and Wan patchification fix different pipeline behaviors." }, { "left": "pull_request:12637", "right": "pull_request:13443", "accept": false, - "reason": "Copy cleanup in Sana video and FA4 backend integration are unrelated." + "reason": "Different model families and unrelated attention/backend tweaks." }, { "left": "pull_request:12675", "right": "pull_request:13158", "accept": false, - "reason": "Kandinsky5 docs typo cleanup and QwenImage docs addition are not the same documentation change." + "reason": "Kandinsky5 docs typo and QwenImage layered docs addition are different doc changes." }, { "left": "pull_request:12686", "right": "pull_request:13344", "accept": false, - "reason": "Kandinsky5 image-pipeline feature work and GLM image model tests are unrelated." + "reason": "Kandinsky5 pipeline expansion and GLM Image transformer tests are unrelated." + }, + { + "left": "pull_request:12687", + "right": "13275", + "accept": false, + "reason": "Autoencoder typo fix and Hunyuan test kwargs cleanup are unrelated." + }, + { + "left": "pull_request:12705", + "right": "pull_request:13150", + "accept": false, + "reason": "Typo cleanup across docs/examples and a test helper API change are different issues." + }, + { + "left": "pull_request:12846", + "right": "pull_request:12848", + "accept": false, + "reason": "Both add pipelines, but for different models and different code paths." + }, + { + "left": "pull_request:12846", + "right": "pull_request:12984", + "accept": false, + "reason": "AutoPipeline support and Flux2 Klein docs are unrelated." + }, + { + "left": "pull_request:12846", + "right": "pull_request:13259", + "accept": false, + "reason": "Auto pipeline support and AGENTS.md are different changes entirely." + }, + { + "left": "pull_request:12848", + "right": "pull_request:12984", + "accept": false, + "reason": "Chroma inpaint pipeline and Flux2 Klein docs do not match." + }, + { + "left": "pull_request:12848", + "right": "pull_request:13259", + "accept": false, + "reason": "New pipeline implementation vs contributor guidance docs; unrelated." + }, + { + "left": "pull_request:12868", + "right": "pull_request:13174", + "accept": false, + "reason": "QwenImage attention splitting and attention-backend tests are not the same concrete fix." } ] }, @@ -171082,7 +171291,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 219, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -171090,15 +171299,8 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:12687", - "pull_request:12705", - "pull_request:12782", - "pull_request:12846", - "pull_request:12848", - "pull_request:12868", "pull_request:1291", "pull_request:1292", - "pull_request:12984", "pull_request:1299", "pull_request:12998", "pull_request:13038", @@ -171108,25 +171310,29 @@ "pull_request:1311", "pull_request:13146", "pull_request:13150", - "pull_request:13174", "pull_request:13204", "pull_request:1322", "pull_request:1325", "pull_request:13259", "pull_request:13269", "pull_request:1327", - "pull_request:13275", "pull_request:1330", "pull_request:13340", "pull_request:1338", + "pull_request:1341", "pull_request:13436", "pull_request:13443", + "pull_request:1352", + "pull_request:1360", "pull_request:1382", + "pull_request:1406", "pull_request:15", "pull_request:1624", "pull_request:17", "pull_request:1907", "pull_request:1939", + "pull_request:21", + "pull_request:2516", "pull_request:2529", "pull_request:2530", "pull_request:2540", @@ -171137,19 +171343,11 @@ "pull_request:432", "pull_request:483", "pull_request:51", - "pull_request:62" + "pull_request:62", + "pull_request:7", + "pull_request:842" ], "soft_pairs": [ - "pull_request:12687|pull_request:13275", - "pull_request:12705|pull_request:13038", - "pull_request:12705|pull_request:13443", - "pull_request:12782|pull_request:13150", - "pull_request:12846|pull_request:12848", - "pull_request:12846|pull_request:12984", - "pull_request:12846|pull_request:13259", - "pull_request:12848|pull_request:12984", - "pull_request:12848|pull_request:13259", - "pull_request:12868|pull_request:13174", "pull_request:1291|pull_request:2529", "pull_request:1292|pull_request:1907", "pull_request:1299|pull_request:483", @@ -171176,258 +171374,233 @@ "pull_request:1330|pull_request:2958", "pull_request:1330|pull_request:432", "pull_request:13340|pull_request:13436", - "pull_request:1338|pull_request:51" + "pull_request:1338|pull_request:51", + "pull_request:1341|pull_request:7", + "pull_request:1352|pull_request:1406", + "pull_request:1352|pull_request:21", + "pull_request:1352|pull_request:842", + "pull_request:1360|pull_request:2516" ], - "prepared_review_unit_hash": "91beb4015e70d16c608c31599ab2cc4b9f2f3663fb470e46dbde2eeac920ed0a", + "prepared_review_unit_hash": "4074d0a2107aba40f59084f8d17b2ec547f6d9b42e592e65fd8049db2aede7a7", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 37, - "serialized_chars": 102794, - "estimated_input_tokens": 25699, - "estimated_eval_tokens": 51654 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 99949, + "estimated_input_tokens": 24988, + "estimated_eval_tokens": 50232 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very noisy: it mixes unrelated typo/docs tweaks, test-only changes, model/pipeline feature work, and a few small core fixes. I reject every soft edge because none of the paired PRs look like the same underlying change.", - "confidence": 0.93, + "summary": "The cluster is heterogeneous: most PRs are unrelated docs, tests, or different feature/fix areas. None of the soft-similarity pairs looks like the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a convincing duplicate anchor here; the cluster is too heterogeneous to treat any one change as canonical.", + "canonical_pr_reason": "No single PR is a valid canonical for this cluster; the items cover unrelated changes across docs, pipelines, models, tests, and tooling.", "best_issue_reason": null, - "best_pr_reason": "PR 12705 ('Fix typos') is the broadest generic maintenance change in the set, so it is the least-bad representative, but it still does not unify the cluster.", + "best_pr_reason": "No PR is a strong global representative because the soft-similarity links are mostly superficial and do not share the same concrete code-path problem.", "soft_edge_verdicts": [ - { - "left": "pull_request:12687", - "right": "pull_request:13275", - "accept": false, - "reason": "Unrelated fixes: autoencoder typo vs. Hunyuan image-to-video test kwargs." - }, - { - "left": "pull_request:12705", - "right": "pull_request:13038", - "accept": false, - "reason": "Different topics: typo/docs cleanup vs. Wan patchification bug fix." - }, - { - "left": "pull_request:12705", - "right": "pull_request:13443", - "accept": false, - "reason": "Different topics: typo/docs cleanup vs. attention backend version check." - }, - { - "left": "pull_request:12782", - "right": "pull_request:13150", - "accept": false, - "reason": "Both are tests, but they cover different models and failures." - }, - { - "left": "pull_request:12846", - "right": "pull_request:12848", - "accept": false, - "reason": "Both add pipelines, but they are different concrete pipelines and features." - }, - { - "left": "pull_request:12846", - "right": "pull_request:12984", - "accept": false, - "reason": "Feature addition vs. docs-only klein update; not the same change." - }, - { - "left": "pull_request:12846", - "right": "pull_request:13259", - "accept": false, - "reason": "AutoPipeline feature work is unrelated to AGENTS/docs tooling changes." - }, - { - "left": "pull_request:12848", - "right": "pull_request:12984", - "accept": false, - "reason": "Different features: Chroma inpaint pipeline vs. Flux2 Klein docs/pipeline wording." - }, - { - "left": "pull_request:12848", - "right": "pull_request:13259", - "accept": false, - "reason": "Pipeline feature addition vs. AGENTS/docs work are unrelated." - }, - { - "left": "pull_request:12868", - "right": "pull_request:13174", - "accept": false, - "reason": "Attention backend split vs. attention backend tests; related subsystem, different change." - }, { "left": "pull_request:1291", "right": "pull_request:2529", "accept": false, - "reason": "Notebook link fix vs. checkpoint conversion logic fix; unrelated." + "reason": "Different fixes: README notebook link vs SD checkpoint conversion logic; no shared bug." }, { "left": "pull_request:1292", "right": "pull_request:1907", "accept": false, - "reason": "Revert of one change set vs. Dockerfile merge work; unrelated." + "reason": "One is OneFlow integration code, the other is a docker/build repo change; unrelated changes." }, { "left": "pull_request:1299", "right": "pull_request:483", "accept": false, - "reason": "AltDiffusion addition vs. typo/docs workflow changes; unrelated." + "reason": "AltDiffusion feature addition vs typo/doc cleanup; not the same underlying change." }, { "left": "pull_request:12998", "right": "pull_request:13081", "accept": false, - "reason": "Style-only whitespace cleanup vs. tensor contiguity fix; unrelated." + "reason": "Different style/runtime issues in different pipelines; not one code-path problem." }, { "left": "pull_request:13038", "right": "pull_request:13146", "accept": false, - "reason": "Wan patchification fix vs. tokenizer type fix; unrelated." + "reason": "Wan patchification fix vs tokenizer type fix; separate failures in different components." }, { "left": "pull_request:13039", "right": "pull_request:13150", "accept": false, - "reason": "Different test failures in different WAN areas; not the same bug." + "reason": "CPU upcasting bug in a model forward pass vs a test helper tweak; unrelated." }, { "left": "pull_request:13100", "right": "pull_request:13204", "accept": false, - "reason": "Same file, but different modular pipeline behaviors and fixes." + "reason": "Modular pipeline docs/state handling vs trust_remote_code behavior; different concerns." }, { "left": "pull_request:1311", "right": "pull_request:2530", "accept": false, - "reason": "AltDiffusion tests vs. EMAModel test suite; unrelated." + "reason": "AltDiffusion tests vs EMAModel tests; same area only at a very broad level." }, { "left": "pull_request:13146", "right": "pull_request:13443", "accept": false, - "reason": "Tokenizer type cleanup vs. attention backend version gating; unrelated." + "reason": "Tokenizer type fix vs attention backend version check; no common bug path." }, { "left": "pull_request:1322", "right": "pull_request:1330", "accept": false, - "reason": "Different inpainting-related fixes; one is scheduler path, the other is inpaint scaling." + "reason": "Both mention scheduler/path usage, but one fixes examples' pretrained path and the other inpaint scaling; not the same bug." }, { "left": "pull_request:1322", "right": "pull_request:1624", "accept": false, - "reason": "Example scheduler path fix vs. DPM timestep calculation; unrelated." + "reason": "Examples scheduler path fix vs DPM timestep computation; different code paths." }, { "left": "pull_request:1322", "right": "pull_request:1939", "accept": false, - "reason": "Scheduler path fix vs. README logo path fix; unrelated." + "reason": "Scheduler path fix vs README logo path; only superficial path wording overlap." }, { "left": "pull_request:1322", "right": "pull_request:2540", "accept": false, - "reason": "Example scheduler fix vs. section-doc cleanup; unrelated." + "reason": "Scheduler path fix vs docs section correction; unrelated." }, { "left": "pull_request:1322", "right": "pull_request:432", "accept": false, - "reason": "Different documentation/link fixes with no shared underlying bug." + "reason": "Different documentation/example link fixes; not the same concrete issue." }, { "left": "pull_request:1325", "right": "pull_request:1382", "accept": false, - "reason": "Image2image fix vs. dependency import gating; unrelated." + "reason": "OneFlow image2image fix vs dependency/import gating; different bugs and scopes." }, { "left": "pull_request:1325", "right": "pull_request:15", "accept": false, - "reason": "Different fixes: image2image behavior vs. README sample path." + "reason": "No overlap beyond being repo housekeeping; unrelated changes." }, { "left": "pull_request:1325", "right": "pull_request:17", "accept": false, - "reason": "Image2image bug fix vs. typo cleanup; unrelated." + "reason": "OneFlow code vs typo fix; not mergeable as one PR." }, { "left": "pull_request:1325", "right": "pull_request:27", "accept": false, - "reason": "Image2image fix vs. TODO checklist formatting; unrelated." + "reason": "OneFlow integration vs README checklist formatting; unrelated." }, { "left": "pull_request:1325", "right": "pull_request:62", "accept": false, - "reason": "Different pipeline behaviors: image2image fix vs. latent diffusion uncond fix." + "reason": "OneFlow image2image fix vs LDM unconditional pipeline fix; different pipelines and bugs." }, { "left": "pull_request:13259", "right": "pull_request:13269", "accept": false, - "reason": "Same docs/tooling area, but 13269 expands AGENTS with skills support rather than duplicating 13259." + "reason": "AGENTS.md introduction vs later skill/doc expansion; related docs, but not the same change." }, { "left": "pull_request:1327", "right": "pull_request:3397", "accept": false, - "reason": "Custom local pipeline loading vs. LoRA local-directory note; unrelated." + "reason": "Local custom pipeline loading behavior vs LoRA local directory docs; different functionality." }, { "left": "pull_request:1330", "right": "pull_request:1624", "accept": false, - "reason": "Inpaint scaling vs. DPM timestep scheduling; unrelated." + "reason": "Inpaint scaling fix vs DPM scheduler timesteps; unrelated." }, { "left": "pull_request:1330", "right": "pull_request:2667", "accept": false, - "reason": "Both touch inpainting, but one fixes scaling while the other implements strength behavior." + "reason": "Scheduler/path issue in training examples vs implementing inpaint strength behavior; not the same bug." }, { "left": "pull_request:1330", "right": "pull_request:2958", "accept": false, - "reason": "Inpaint scaling fix vs. ControlNet inpaint addition; different changes." + "reason": "Example scheduler path fix vs controlnet inpaint addition; different features." }, { "left": "pull_request:1330", "right": "pull_request:432", "accept": false, - "reason": "Inpaint scaling bug vs. link cleanup; unrelated." + "reason": "Inpaint scaling change vs docs link corrections; unrelated." }, { "left": "pull_request:13340", "right": "pull_request:13436", "accept": false, - "reason": "Transformer test refactor vs. trivial README edit; unrelated." + "reason": "Model test refactor vs a trivial README test edit; not the same underlying issue." }, { "left": "pull_request:1338", "right": "pull_request:51", "accept": false, - "reason": "ONNX inpaint cast-order fix vs. dependency pin correction; unrelated." + "reason": "ONNX inpaint cast-order fix vs setup dependency pin; completely unrelated." + }, + { + "left": "pull_request:1341", + "right": "pull_request:7", + "accept": false, + "reason": "Stable Diffusion repaint pipeline vs VQ-VAE model addition; different features entirely." + }, + { + "left": "pull_request:1352", + "right": "pull_request:1406", + "accept": false, + "reason": "AltDiffusion sample checkpoint/docs update vs deprecating sample size; different docs/product changes." + }, + { + "left": "pull_request:1352", + "right": "pull_request:21", + "accept": false, + "reason": "AltDiffusion docs vs early model tests; no shared bug or change." + }, + { + "left": "pull_request:1352", + "right": "pull_request:842", + "accept": false, + "reason": "AltDiffusion sample model docs vs removing ['sample'] from pipeline outputs; unrelated." + }, + { + "left": "pull_request:1360", + "right": "pull_request:2516", + "accept": false, + "reason": "Community pipeline code/README update vs evaluation docs page; no common code-path problem." } ] }, @@ -171443,17 +171616,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 220, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:1341", - "pull_request:1352", - "pull_request:1360", "pull_request:1373", "pull_request:1374", "pull_request:1382", @@ -171470,16 +171640,19 @@ "pull_request:1510", "pull_request:1520", "pull_request:1525", + "pull_request:153", + "pull_request:154", + "pull_request:1547", "pull_request:159", + "pull_request:1605", "pull_request:1610", "pull_request:1756", + "pull_request:1788", "pull_request:1903", - "pull_request:21", "pull_request:2130", "pull_request:2188", "pull_request:227", "pull_request:2496", - "pull_request:2516", "pull_request:286", "pull_request:289", "pull_request:299", @@ -171493,19 +171666,14 @@ "pull_request:43", "pull_request:570", "pull_request:662", - "pull_request:7", "pull_request:718", + "pull_request:768", "pull_request:774", "pull_request:800", "pull_request:839", "pull_request:842" ], "soft_pairs": [ - "pull_request:1341|pull_request:7", - "pull_request:1352|pull_request:1406", - "pull_request:1352|pull_request:21", - "pull_request:1352|pull_request:842", - "pull_request:1360|pull_request:2516", "pull_request:1373|pull_request:299", "pull_request:1373|pull_request:800", "pull_request:1374|pull_request:43", @@ -171537,258 +171705,249 @@ "pull_request:1510|pull_request:1610", "pull_request:1510|pull_request:423", "pull_request:1520|pull_request:3608", - "pull_request:1525|pull_request:570" + "pull_request:1525|pull_request:570", + "pull_request:153|pull_request:1788", + "pull_request:154|pull_request:1605", + "pull_request:1547|pull_request:768" ], - "prepared_review_unit_hash": "fc1de577dbc95b860ed5a5bc79fa0ee235be05daa80dbe60d3bc65ca826cda5d", + "prepared_review_unit_hash": "14c15f67111aa2dd1fb4533de8a8b2b4d228909348e927f2bb1bfb69e4381a3e", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 37, - "serialized_chars": 108860, - "estimated_input_tokens": 27215, - "estimated_eval_tokens": 54686 + "soft_pair_count": 35, + "serialized_chars": 107187, + "estimated_input_tokens": 26797, + "estimated_eval_tokens": 53850 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a heterogeneous mix of unrelated PRs: feature additions (new pipelines/schedulers), docs-only edits, typo fixes, and isolated bug fixes. The soft edges are mostly superficial file overlap or broad subsystem similarity, not the same underlying change.", + "summary": "The cluster is heterogeneous: the PRs cover unrelated docs, examples, schedulers, pipeline fixes, and feature additions. The soft links are mostly superficial overlaps in shared subsystem files, not true duplicate bugs/changes.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR is a canonical representative; the items are unrelated changes across docs, examples, schedulers, and different pipelines.", "best_issue_reason": null, - "best_pr_reason": "No PR is a strong canonical representative because the cluster does not form a single duplicate/change set; the items span many unrelated features and fixes.", + "best_pr_reason": "No single PR is a good global representative for duplicate triage because the cluster is not about one underlying change.", "soft_edge_verdicts": [ - { - "left": "pull_request:1341", - "right": "pull_request:7", - "accept": false, - "reason": "Both touch __init__.py, but one adds Stable Diffusion Repaint and the other adds VQ-VAE; different features and code paths." - }, - { - "left": "pull_request:1352", - "right": "pull_request:1406", - "accept": false, - "reason": "Sample model docs vs config/deprecation handling; unrelated changes despite the shared 'sample' wording." - }, - { - "left": "pull_request:1352", - "right": "pull_request:21", - "accept": false, - "reason": "A docs checkpoint update is unrelated to model tests and UNet refactoring." - }, - { - "left": "pull_request:1352", - "right": "pull_request:842", - "accept": false, - "reason": "Changing a doc sample output format is not the same as removing ['sample'] access from pipeline outputs." - }, - { - "left": "pull_request:1360", - "right": "pull_request:2516", - "accept": false, - "reason": "Community k-diffusion pipeline work is unrelated to a docs page about evaluating diffusion models." - }, { "left": "pull_request:1373", "right": "pull_request:299", "accept": false, - "reason": "Versatile Diffusion feature expansion and a PNDM hotfix are different changes." + "reason": "Different changes: Versatile Diffusion extension vs a PNDM scheduler hotfix." }, { "left": "pull_request:1373", "right": "pull_request:800", "accept": false, - "reason": "Versatile Diffusion extension and WIP VQ Diffusion are separate model families and implementations." + "reason": "Unrelated features: Versatile Diffusion docs/code vs VQ Diffusion work-in-progress." }, { "left": "pull_request:1374", "right": "pull_request:43", "accept": false, - "reason": "An AI template PR is unrelated to adding a model card template." + "reason": "Unrelated examples/ops changes vs model card/template and scheduler refactor work." }, { "left": "pull_request:1382", "right": "pull_request:286", "accept": false, - "reason": "Import fixes in pipeline init files are not the same as a README link fix." + "reason": "Import/dependency fix vs a docs link correction; not the same code path." }, { "left": "pull_request:1382", "right": "pull_request:662", "accept": false, - "reason": "Import/dependency cleanup is unrelated to a textual inversion 6GB example change." + "reason": "Bad-dependency import cleanup vs DreamBooth/textual inversion content; different concerns." }, { "left": "pull_request:1382", "right": "pull_request:718", "accept": false, - "reason": "Bad-dependency import fixes do not match typo fixes in docs/examples." + "reason": "Import fix vs typo fixes; no shared underlying bug." }, { "left": "pull_request:1382", "right": "pull_request:774", "accept": false, - "reason": "Import guards are unrelated to test expectation tweaks." + "reason": "Import fix vs pipeline test adjustments; not a duplicate change." }, { "left": "pull_request:1386", "right": "pull_request:227", "accept": false, - "reason": "Adding v-prediction support in schedulers is different from adding add_noise methods." + "reason": "v-prediction scheduler support vs adding add_noise methods; different scheduler behavior." }, { "left": "pull_request:1402", "right": "pull_request:1425", "accept": false, - "reason": "A pipeline loading fix is unrelated to a docs typo cleanup." + "reason": "Pipeline loading fix vs docs typo cleanup; only broad repo overlap." }, { "left": "pull_request:1406", "right": "pull_request:842", "accept": false, - "reason": "Deprecating sample size and removing ['sample'] output access are distinct API changes." + "reason": "Config/deprecation behavior change vs output-access test cleanup; unrelated." }, { "left": "pull_request:1412", "right": "pull_request:3230", "accept": false, - "reason": "SD2 inpainting tests and the IF pipeline implementation are different artifacts and code paths." + "reason": "SD2 inpainting tests vs IF pipeline feature addition; different pipelines and purposes." }, { "left": "pull_request:1424", "right": "pull_request:1903", "accept": false, - "reason": "Stable Diffusion 2 docs and SD2 depth2img comment fixes are not the same change." + "reason": "Stable Diffusion 2 docs vs depth2img comment cleanup; not the same change." }, { "left": "pull_request:1424", "right": "pull_request:396", "accept": false, - "reason": "SD2 docs and DDIM docs are unrelated documentation additions." + "reason": "SD2 docs vs DDIM docs/pipeline work; unrelated documentation sets." }, { "left": "pull_request:1424", "right": "pull_request:408", "accept": false, - "reason": "SD2 docs and generic docs typo cleanup are different documentation changes." + "reason": "SD2 docs vs general docs typo cleanup; no duplicate bug/change." }, { "left": "pull_request:1425", "right": "pull_request:159", "accept": false, - "reason": "Typo fixes in unrelated docs files are not the same underlying change." + "reason": "Docs typo fixes in different docs areas; not the same underlying issue." }, { "left": "pull_request:1425", "right": "pull_request:718", "accept": false, - "reason": "These are separate typo/doc cleanup PRs in different files and contexts." + "reason": "Docs typo cleanup vs broader typo/example fixes; too different to merge as one." }, { "left": "pull_request:1470", "right": "pull_request:3080", "accept": false, - "reason": "DreamBooth Alt Diffusion compatibility is unrelated to pipeline save_pretrained behavior." + "reason": "DreamBooth alt-diffusion compatibility vs pipeline save_pretrained behavior; different code paths." }, { "left": "pull_request:1487", "right": "pull_request:1756", "accept": false, - "reason": "xformers docs updates are not the same as adding attention masks to UCLIP attention." + "reason": "xformers docs update vs attention-mask code fix; one is docs only, the other runtime behavior." }, { "left": "pull_request:1489", "right": "pull_request:2130", "accept": false, - "reason": "Removing a reminder comment is unrelated to allow_patterns cleanup in pipeline loading." + "reason": "Removing a reminder comment vs pipeline allow_patterns cleanup; unrelated maintenance." }, { "left": "pull_request:1489", "right": "pull_request:2188", "accept": false, - "reason": "Comment removal in tests is unrelated to removing a license notice from docs." + "reason": "Removing a comment vs deleting a docs license note; different files and goals." }, { "left": "pull_request:1489", "right": "pull_request:2496", "accept": false, - "reason": "A test comment deletion is unrelated to the image_noiser/image_normalizer comment edit." + "reason": "Comment cleanup vs a code comment rename in UnCLIP; not the same issue." }, { "left": "pull_request:1489", "right": "pull_request:289", "accept": false, - "reason": "Test comment cleanup and seed-refactor pipeline changes are different scopes." + "reason": "Comment cleanup vs refactoring seed handling; unrelated changes." }, { "left": "pull_request:1489", "right": "pull_request:34", "accept": false, - "reason": "Removing a reminder comment is not the same as fixing scheduler comment typos." + "reason": "A single test comment removal vs a typo fix in DDPM scheduler comments; not duplicates." }, { "left": "pull_request:1489", "right": "pull_request:839", "accept": false, - "reason": "A scheduler test comment change is unrelated to DreamBooth README wording." + "reason": "Comment cleanup vs dreambooth README auth-token change; unrelated." }, { "left": "pull_request:1489", "right": "pull_request:842", "accept": false, - "reason": "Comment removal in tests is unrelated to output API changes removing ['sample']." + "reason": "Comment cleanup vs pipeline output API cleanup; different scope and code." }, { "left": "pull_request:15", "right": "pull_request:286", "accept": false, - "reason": "An example output path rename is unrelated to a broken README link." + "reason": "Example output-path rename vs stable diffusion docs link fix; no shared bug/change." }, { "left": "pull_request:15", "right": "pull_request:662", "accept": false, - "reason": "Example path naming has nothing to do with the 6GB textual inversion update." + "reason": "Example path rename vs textual inversion changes; unrelated example content." }, { "left": "pull_request:15", "right": "pull_request:718", "accept": false, - "reason": "Output path naming and typo fixes are separate trivial edits." + "reason": "Example path rename vs typo fixes; only superficial repo overlap." }, { "left": "pull_request:1510", "right": "pull_request:1610", "accept": false, - "reason": "Scheduler conversion-script improvements do not match attention-test adjustments." + "reason": "Scheduler/docs conversion-script work vs pipeline test improvements; not the same issue." }, { "left": "pull_request:1510", "right": "pull_request:423", "accept": false, - "reason": "A conversion script/scheduler PR is unrelated to output syntax docs and pipeline changes." + "reason": "Conversion script/scheduler additions vs output syntax updates; different feature sets." }, { "left": "pull_request:1520", "right": "pull_request:3608", "accept": false, - "reason": "Imagic example fixes and a text-to-image argument typo are different example-level changes." + "reason": "Imagic example fixes vs text-to-image training script typo fix; different examples and behavior." }, { "left": "pull_request:1525", "right": "pull_request:570", "accept": false, - "reason": "Batch-size logic in upscaling is unrelated to BaseOutput dict initialization." + "reason": "Upscaling batch-size bug vs BaseOutput dict initialization bug; unrelated code paths." + }, + { + "left": "pull_request:153", + "right": "pull_request:1788", + "accept": false, + "reason": "Issue templates vs DDIM prediction_type bug fix; no overlap beyond being repo maintenance." + }, + { + "left": "pull_request:154", + "right": "pull_request:1605", + "accept": false, + "reason": "Stable diffusion conversion script vs Versatile Diffusion upcast_attention; different models/features." + }, + { + "left": "pull_request:1547", + "right": "pull_request:768", + "accept": false, + "reason": "DreamBooth collate_fn bugfix vs broad dreambooth enhancement; not the same concrete fix." } ] }, @@ -171804,17 +171963,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 221, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:153", - "pull_request:154", - "pull_request:1547", "pull_request:1548", "pull_request:1558", "pull_request:1564", @@ -171823,7 +171979,6 @@ "pull_request:159", "pull_request:1591", "pull_request:1601", - "pull_request:1605", "pull_request:1610", "pull_request:1623", "pull_request:1624", @@ -171836,12 +171991,16 @@ "pull_request:1761", "pull_request:1763", "pull_request:1765", - "pull_request:1788", + "pull_request:1827", + "pull_request:1849", "pull_request:1971", "pull_request:2122", "pull_request:217", "pull_request:2279", "pull_request:2282", + "pull_request:2467", + "pull_request:2542", + "pull_request:258", "pull_request:2691", "pull_request:3230", "pull_request:3241", @@ -171859,12 +172018,9 @@ "pull_request:61", "pull_request:63", "pull_request:662", - "pull_request:768" + "pull_request:828" ], "soft_pairs": [ - "pull_request:153|pull_request:1788", - "pull_request:154|pull_request:1605", - "pull_request:1547|pull_request:768", "pull_request:1548|pull_request:1971", "pull_request:1548|pull_request:433", "pull_request:1558|pull_request:3653", @@ -171891,222 +172047,220 @@ "pull_request:1763|pull_request:2691", "pull_request:1763|pull_request:3653", "pull_request:1763|pull_request:3712", - "pull_request:1765|pull_request:2282" + "pull_request:1765|pull_request:2282", + "pull_request:1765|pull_request:2542", + "pull_request:1765|pull_request:828", + "pull_request:1827|pull_request:2467", + "pull_request:1849|pull_request:258" ], - "prepared_review_unit_hash": "a7df839a3d526d71392d8997a920ef0544a98a05ebcb55632e1c0168db28b4eb", + "prepared_review_unit_hash": "2d22dc04aba9ac67b221ec3f4a4d7a40a4b73543af5bdaf140b3cc6fb1f3181f", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 30, - "serialized_chars": 102075, - "estimated_input_tokens": 25519, - "estimated_eval_tokens": 51294 + "soft_pair_count": 31, + "serialized_chars": 103700, + "estimated_input_tokens": 25925, + "estimated_eval_tokens": 52106 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab-bag of small maintenance PRs: docs typos, README/template edits, scheduler bug fixes, attention/memory changes, and a few example-script tweaks. I don\u2019t see any true duplicate pairs; most soft links are only title-level similarity.", - "confidence": 0.16, + "summary": "Mostly unrelated PRs grouped by superficial similarity (typos, docs, scheduler/model fixes). No true duplicate pair stands out; all soft edges should be rejected.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 1558 is the closest representative of the cluster\u2019s most common theme: small, generic maintenance/typo cleanup. That said, the cluster is very heterogeneous, so it is only a weak canonical anchor.", + "canonical_pr_reason": "No clear canonical PR: the set mixes unrelated docs, lint, scheduler, model, and example changes rather than one duplicated change.", "best_issue_reason": null, - "best_pr_reason": "PR 1558 is the best global representative among these PRs because it is a concise, generic maintenance fix and sits near several other typo/docs cleanup PRs. It is not a strong duplicate target for the code-fix PRs, but it is the most canonical-looking maintenance entry overall.", + "best_pr_reason": "No global best PR: there is no single PR that meaningfully represents the rest of the cluster because they target different code paths and maintenance tasks.", "soft_edge_verdicts": [ - { - "left": "pull_request:153", - "right": "pull_request:1788", - "accept": false, - "reason": "Issue templates vs DDIM scheduler logic; unrelated change types and code paths." - }, - { - "left": "pull_request:154", - "right": "pull_request:1605", - "accept": false, - "reason": "Stable-diffusion conversion script vs versatile-diffusion upcast_attention; no shared bug or change." - }, - { - "left": "pull_request:1547", - "right": "pull_request:768", - "accept": false, - "reason": "Both are Dreambooth-related, but one fixes collate/tokenization and the other adds default-args/enhancement plumbing; not the same concrete change." - }, { "left": "pull_request:1548", "right": "pull_request:1971", "accept": false, - "reason": "Bug-report template update vs PNDM docstring cleanup; unrelated." + "reason": "Both are text-only edits, but one updates an issue template and the other a scheduler docstring; not the same change." }, { "left": "pull_request:1548", "right": "pull_request:433", "accept": false, - "reason": "Issue template markdown update vs black/versioning changes; unrelated." + "reason": "Issue-template wording and black/version updates are unrelated maintenance changes." }, { "left": "pull_request:1558", "right": "pull_request:3653", "accept": false, - "reason": "Both are minor cleanup, but one is typo fixes and the other is linting refactor in a community example; not the same underlying change." + "reason": "Both are small cleanup PRs, but one fixes typos in pipelines and the other is a lint refactor in an example; different scope." }, { "left": "pull_request:1564", "right": "pull_request:2122", "accept": false, - "reason": "Dreambooth README accelerate config docs vs safetensors docs; different documentation topics." + "reason": "Both touch docs, but one adds an accelerate config note and the other adds safetensors documentation; unrelated topics." }, { "left": "pull_request:1582", "right": "pull_request:1623", "accept": false, - "reason": "RePaint scheduler bug fix vs requirements version bump; no shared code-path problem." + "reason": "A RePaint scheduler bug fix and dependency version bumps are unrelated." }, { "left": "pull_request:1583", "right": "pull_request:1623", "accept": false, - "reason": "Legacy inpaint scheduler fix vs dependency update; unrelated." + "reason": "An inpaint-legacy masking fix and requirements updates do not address the same bug." }, { "left": "pull_request:159", "right": "pull_request:17", "accept": false, - "reason": "Both are typo/doc cleanup, but they touch different files and different typo sets; not one concrete duplicate change." + "reason": "Both are typo cleanup, but they update different documents in different areas and do not share one underlying defect." }, { "left": "pull_request:159", "right": "pull_request:408", "accept": false, - "reason": "General typo cleanup, but separate doc files and edits; too broad to treat as the same change." + "reason": "Two separate documentation typo PRs; no shared code-path or single fix." }, { "left": "pull_request:159", "right": "pull_request:568", "accept": false, - "reason": "Overlapping typo-cleanup theme only; the actual file sets and edits differ substantially." + "reason": "Scheduler README typos and broader typo cleanup are different maintenance changes." }, { "left": "pull_request:1591", "right": "pull_request:3241", "accept": false, - "reason": "Slice-attention upcast fix vs relaxed UnCLIP test threshold; code fix vs test-only change." + "reason": "One fixes slice-attention upcasting; the other loosens a fast test threshold. Test adjustment is not the same bug fix." }, { "left": "pull_request:1601", "right": "pull_request:535", "accept": false, - "reason": "Classifier-free guidance control-flow fix vs resnet upsampling implementation fix; unrelated subsystems." + "reason": "Classifier-free-guidance control flow and a resnet upsample implementation fix are unrelated code paths." }, { "left": "pull_request:1610", "right": "pull_request:423", "accept": false, - "reason": "Pipeline tests/attention logger work vs output syntax/docs cleanup; not the same bug or change." + "reason": "An attention/xformers-related change and docs/output syntax cleanup are unrelated." }, { "left": "pull_request:1623", "right": "pull_request:2279", "accept": false, - "reason": "Example requirement pins vs repo-wide linting/tooling migration; related only at a very broad maintenance level." + "reason": "Both are maintenance/tooling changes, but one bumps requirements while the other swaps linters and black versions; not the same change." }, { "left": "pull_request:1624", "right": "pull_request:3343", "accept": false, - "reason": "Different DPM scheduler variants and different bugs (timesteps computation vs config field reference)." + "reason": "Both are DPM scheduler-related, but one corrects timestep generation and the other fixes a config reference; different concrete bugs." }, { "left": "pull_request:1624", "right": "pull_request:35", "accept": false, - "reason": "Scheduler timestep correction vs adding an analytic DPM pipeline; not duplicates." + "reason": "Scheduler timestep logic and an analytic DPM pipeline addition are unrelated." }, { "left": "pull_request:1624", "right": "pull_request:432", "accept": false, - "reason": "Scheduler code fix vs documentation link updates; unrelated." + "reason": "A scheduler algorithm fix and a docs link correction are unrelated." }, { "left": "pull_request:1627", "right": "pull_request:3230", "accept": false, - "reason": "xformers enable plumbing for UNet vs adding IF pipelines/docs; different feature areas." + "reason": "Re-enabling xformers attention on UNet2DCondition and adding IF are unrelated feature/fix changes." }, { "left": "pull_request:17", "right": "pull_request:662", "accept": false, - "reason": "README typo cleanup vs textual-inversion example addition; no shared underlying bug." + "reason": "Both mention README/docs, but one is a typo fix and the other is a 6GB textual inversion tweak; different change sets." }, { "left": "pull_request:1706", "right": "pull_request:3723", "accept": false, - "reason": "Main docs update vs reference-image example enhancement; different scopes and artifacts." + "reason": "Main docs refresh and reference-example handling are separate documentation/example updates." }, { "left": "pull_request:171", "right": "pull_request:217", "accept": false, - "reason": "Eta-condition fix in latent diffusion vs safety checker bug fix; unrelated code paths." + "reason": "Eta-condition logic in latent diffusion and a safety checker bug are unrelated." }, { "left": "pull_request:1713", "right": "pull_request:61", "accept": false, - "reason": "Both are conversion-related, but one is a generic conversion script tweak and the other is an LDM-uncond converter; different converters and behavior." + "reason": "Different conversion scripts for different pipelines; not the same underlying bug or change." }, { "left": "pull_request:1713", "right": "pull_request:63", "accept": false, - "reason": "Different conversion scripts for different targets; not the same bug/change." + "reason": "Both are conversion helpers, but they target different source pipelines and conversion logic." }, { "left": "pull_request:1756", "right": "pull_request:2691", "accept": false, - "reason": "Attention-mask handling for UCLIP vs large attention-module rename/refactor; not one concrete fix." + "reason": "Attention-mask support and module renaming touch the same file area but solve different problems." }, { - "left": "pull_request:1756", + "left": "pull_request:1763", "right": "pull_request:3653", "accept": false, - "reason": "Attention-mask code change vs linting cleanup in an example; unrelated." + "reason": "Versatile-diffusion attention-mask handling and a linting cleanup are unrelated." }, { - "left": "pull_request:1756", + "left": "pull_request:1763", "right": "pull_request:3712", "accept": false, - "reason": "Model attention-mask fix vs workflow secret rename; unrelated." + "reason": "A model behavior fix and workflow secret renaming are unrelated." }, { - "left": "pull_request:1763", - "right": "pull_request:2691", + "left": "pull_request:1765", + "right": "pull_request:2282", "accept": false, - "reason": "Versatile attention-mask fix vs attention-module rename/refactor; different problem statements and code paths." + "reason": "DreamBooth Flax fixes and miscellaneous converter/docs fixes do not share the same code-path problem." }, { - "left": "pull_request:1763", - "right": "pull_request:3653", + "left": "pull_request:1765", + "right": "pull_request:2542", "accept": false, - "reason": "Attention-mask behavior fix vs example linting cleanup; unrelated." + "reason": "Flax DreamBooth training fixes and ControlNet docs fixes are unrelated." }, { "left": "pull_request:1765", - "right": "pull_request:2282", + "right": "pull_request:828", "accept": false, - "reason": "Dreambooth flax fixes vs broad misc fixes across converters/schedulers/tests; too broad and not the same concrete change." + "reason": "Both involve Flax, but one changes training logic and the other updates tests; not the same concrete bug." + }, + { + "left": "pull_request:1827", + "right": "pull_request:2467", + "accept": false, + "reason": "Adding add_noise to SDE-VE and adding latent output types are unrelated pipeline/scheduler changes." + }, + { + "left": "pull_request:1849", + "right": "pull_request:258", + "accept": false, + "reason": "Versatile Diffusion cross_attention_kwargs plumbing and an inpainting script cleanup are unrelated." } ] }, @@ -172122,17 +172276,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 222, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:1765", - "pull_request:1827", - "pull_request:1849", "pull_request:1861", "pull_request:1868", "pull_request:1907", @@ -172148,11 +172299,13 @@ "pull_request:2006", "pull_request:2007", "pull_request:2048", + "pull_request:2057", + "pull_request:2086", + "pull_request:21", "pull_request:217", "pull_request:2190", "pull_request:2208", - "pull_request:2467", - "pull_request:2542", + "pull_request:2489", "pull_request:258", "pull_request:2641", "pull_request:286", @@ -172169,6 +172322,7 @@ "pull_request:3681", "pull_request:3712", "pull_request:3804", + "pull_request:4177", "pull_request:433", "pull_request:662", "pull_request:718", @@ -172179,10 +172333,6 @@ "pull_request:828" ], "soft_pairs": [ - "pull_request:1765|pull_request:2542", - "pull_request:1765|pull_request:828", - "pull_request:1827|pull_request:2467", - "pull_request:1849|pull_request:258", "pull_request:1861|pull_request:433", "pull_request:1868|pull_request:217", "pull_request:1868|pull_request:312", @@ -172214,252 +172364,249 @@ "pull_request:2006|pull_request:3433", "pull_request:2007|pull_request:2208", "pull_request:2007|pull_request:3262", - "pull_request:2048|pull_request:3681" + "pull_request:2048|pull_request:3681", + "pull_request:2057|pull_request:4177", + "pull_request:2086|pull_request:2489", + "pull_request:21|pull_request:828" ], - "prepared_review_unit_hash": "79ca44fb7fbbecb235fd307b40b0a35208db58187901287c9576900d5bc0fbe2", + "prepared_review_unit_hash": "74f17c9999791889be88050c279595498fe2fbc5231dd9a0c50848d746f2615d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 36, - "serialized_chars": 99654, - "estimated_input_tokens": 24914, - "estimated_eval_tokens": 50084 + "soft_pair_count": 35, + "serialized_chars": 98594, + "estimated_input_tokens": 24649, + "estimated_eval_tokens": 49554 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a true duplicate cluster: the items span unrelated docs, example scripts, schedulers, tests, and infrastructure changes. None of the proposed soft pairs look like the same underlying bug or change.", - "confidence": 0.97, - "canonical_issue_reason": "No single issue unifies this set; the PRs target different subsystems and fixes.", - "canonical_pr_reason": "No canonical PR fits the cluster because the items are heterogeneous and do not represent one mergeable change.", - "best_issue_reason": "There is no good global issue representative here; the set is too mixed to map cleanly to one bug report.", - "best_pr_reason": "There is no good global PR representative; none of these PRs captures the cluster as one coherent change.", + "summary": "This cluster is a noisy mix of unrelated pull requests across docs, formatting, example scripts, conversion utilities, schedulers, and training helpers. The soft edges are false positives driven by shared files or broad subsystem overlap, not the same underlying bug/change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits: the items span unrelated fixes and features, and the apparent overlaps are mostly superficial (shared files, docs, or style cleanup).", + "best_issue_reason": null, + "best_pr_reason": "If a single anchor is needed, PR 2048 is the most substantial cross-cutting maintenance change in the set, but it still is not a duplicate target for the other PRs.", "soft_edge_verdicts": [ - { - "left": "pull_request:1765", - "right": "pull_request:2542", - "accept": false, - "reason": "Different changes: DreamBooth Flax fixes vs. ControlNet docs tweaks." - }, - { - "left": "pull_request:1765", - "right": "pull_request:828", - "accept": false, - "reason": "Different fixes in unrelated Flax/example code; not the same bug or change." - }, - { - "left": "pull_request:1827", - "right": "pull_request:2467", - "accept": false, - "reason": "Scheduler add_noise implementation vs. latent output type support in pipelines are unrelated." - }, - { - "left": "pull_request:1849", - "right": "pull_request:258", - "accept": false, - "reason": "Versatile Diffusion cross_attention_kwargs plumbing is unrelated to the inpainting script cleanup." - }, { "left": "pull_request:1861", "right": "pull_request:433", "accept": false, - "reason": "Same file appears, but one is loss computation and the other is a broad black-version update." + "reason": "Both touch training/example code, but one changes loss computation while the other is a black-version/lint update; different changes." }, { "left": "pull_request:1868", "right": "pull_request:217", "accept": false, - "reason": "EMA decay fix is unrelated to the stable diffusion safety checker bug." + "reason": "EMA decay logic and the safety checker bug are unrelated code paths." }, { "left": "pull_request:1868", "right": "pull_request:312", "accept": false, - "reason": "EMA decay logic and README link fixes are unrelated changes." + "reason": "Different fixes: EMA scheduling vs safety-checker image handling." }, { "left": "pull_request:1868", "right": "pull_request:3343", "accept": false, - "reason": "Different training utility fix vs. DPM solver config reference." + "reason": "Training EMA decay and a scheduler config typo are not the same underlying issue." }, { "left": "pull_request:1868", "right": "pull_request:3402", "accept": false, - "reason": "EMA decay in training code is unrelated to the Dockerfile fix." + "reason": "EMA decay logic and a Dockerfile repair are unrelated." }, { "left": "pull_request:1868", "right": "pull_request:3433", "accept": false, - "reason": "Training EMA logic and documentation/style rendering are not the same change." + "reason": "One is EMA behavior in training code; the other is doc formatting cleanup." }, { "left": "pull_request:1868", "right": "pull_request:75", "accept": false, - "reason": "Different code paths and eras: EMA decay vs. Glide attention constructor logic." + "reason": "Different model/code-path changes; no concrete bug overlap." }, { "left": "pull_request:1907", "right": "pull_request:2190", "accept": false, - "reason": "Docker build merge commit is unrelated to the DreamBooth/Imagic notebook bundle." + "reason": "A Dockerfile commit and a large notebook/example merge are unrelated." }, { "left": "pull_request:1910", "right": "pull_request:3471", "accept": false, - "reason": "Korean docs initialization and DreamBooth torch.compile note are different documentation changes." + "reason": "Korean docs initialization and a DreamBooth compile note are different documentation changes." }, { "left": "pull_request:1920", "right": "pull_request:2641", "accept": false, - "reason": "TRANSLATING.md addition is unrelated to Jax notebook build workflow changes." + "reason": "Docs translation guide vs notebook-build workflow/config change are different tasks." }, { "left": "pull_request:1937", "right": "pull_request:3804", "accept": false, - "reason": "Stable diffusion EMA extraction script fix is unrelated to attention processor device-map conversion." + "reason": "EMA-weight extraction in the conversion script is unrelated to legacy attention weight conversion." }, { "left": "pull_request:1939", "right": "pull_request:217", "accept": false, - "reason": "README logo path fix and safety-checker bug fix are unrelated." + "reason": "Logo path fix and safety-checker logic are unrelated." }, { "left": "pull_request:1939", "right": "pull_request:312", "accept": false, - "reason": "Logo path correction is unrelated to pipeline README link cleanup." + "reason": "A README asset path fix is not the same as the safety-checker/img2img bug fix." }, { "left": "pull_request:1939", "right": "pull_request:3402", "accept": false, - "reason": "README image path fix and Docker packaging fix are not the same change." + "reason": "README logo path and Docker build changes are unrelated." }, { "left": "pull_request:1939", "right": "pull_request:3433", "accept": false, - "reason": "Unrelated docs/path fix vs. rendering/style fix." + "reason": "A path correction in README is unrelated to style-rendering tweaks." }, { "left": "pull_request:1942", "right": "pull_request:3127", "accept": false, - "reason": "Converter CLI flag rename is unrelated to img2img safety-checker behavior." + "reason": "Converter CLI flag cleanup and img2img safety-checker behavior are distinct changes." }, { "left": "pull_request:1950", "right": "pull_request:286", "accept": false, - "reason": "OneFlow graph config support and Stable Diffusion README link fix are unrelated." + "reason": "OneFlow integration and a README link fix are unrelated." }, { "left": "pull_request:1950", "right": "pull_request:3653", "accept": false, - "reason": "OneFlow support and mixture tiling lint cleanup are different changes." + "reason": "OneFlow feature code and a linting-only change are not the same fix." }, { "left": "pull_request:1950", "right": "pull_request:3712", "accept": false, - "reason": "OneFlow integration and docs workflow secret rename do not fix the same code path." + "reason": "OneFlow integration and CI secret wiring are unrelated." }, { "left": "pull_request:1950", "right": "pull_request:662", "accept": false, - "reason": "OneFlow backend integration and 6GB textual inversion example changes are unrelated." + "reason": "They both touch example/pipeline plumbing, but one adds OneFlow support while the other is a 6GB textual-inversion workaround; different features." }, { "left": "pull_request:1950", "right": "pull_request:718", "accept": false, - "reason": "OneFlow support and typo fixes are not the same underlying change." + "reason": "OneFlow integration and typo cleanup are unrelated." }, { "left": "pull_request:1950", "right": "pull_request:774", "accept": false, - "reason": "OneFlow backend work and pipeline test adjustments are unrelated." + "reason": "A OneFlow port and a test-adjustment PR are different changes." }, { "left": "pull_request:1951", "right": "pull_request:822", "accept": false, - "reason": "ColossalAI upgrade and Flax scheduler API update are unrelated dependency/code changes." + "reason": "Latest ColossalAI integration and Flax scheduler API updates are unrelated." }, { "left": "pull_request:1971", "right": "pull_request:3069", "accept": false, - "reason": "PNDM docstring cleanup is unrelated to Flax TPU test updates." + "reason": "PNDM docstring wording and Flax TPU test updates are unrelated." }, { "left": "pull_request:1971", "right": "pull_request:3074", "accept": false, - "reason": "Docstring wording fix and Transformer2DModel docstring correction are separate documentation edits." + "reason": "A scheduler docstring edit and a Transformer2DModel docstring correction are separate documentation fixes." }, { "left": "pull_request:1971", "right": "pull_request:798", "accept": false, - "reason": "Both touch docs, but they are different docstrings for different models." + "reason": "Different docstring cleanup in different modules; not the same issue." }, { "left": "pull_request:2004", "right": "pull_request:258", "accept": false, - "reason": "DreamBooth README typo fix is unrelated to inpainting script refactoring." + "reason": "DreamBooth README typos and an inpainting script fix are unrelated." }, { "left": "pull_request:2006", "right": "pull_request:3402", "accept": false, - "reason": "Checkpoint merger fixes and Dockerfile cleanup are unrelated." + "reason": "Checkpoint-merger logic and Dockerfile changes are unrelated." }, { "left": "pull_request:2006", "right": "pull_request:3433", "accept": false, - "reason": "Checkpoint merger code changes and style-rendering docs tweaks are not the same bug or change." + "reason": "Checkpoint-merger fixes are not the same as docs style rendering changes." }, { "left": "pull_request:2007", "right": "pull_request:2208", "accept": false, - "reason": "Black-formatting sweep and xFormers docs update are unrelated." + "reason": "Black-version update and xFormers docs updates are unrelated maintenance changes." }, { "left": "pull_request:2007", "right": "pull_request:3262", "accept": false, - "reason": "Black formatting changes and IF model naming/docs updates are unrelated." + "reason": "A formatting bump and IF model-name renaming are unrelated." }, { "left": "pull_request:2048", "right": "pull_request:3681", "accept": false, - "reason": "Accelerate checkpoint hooks overhaul and manual checkpoints_total_limit handling are different training-system changes." + "reason": "Both concern training/checkpointing, but 2048 changes accelerate save/load hooks while 3681 enforces manual checkpoint limits across scripts; different concrete fixes." + }, + { + "left": "pull_request:2057", + "right": "pull_request:4177", + "accept": false, + "reason": "Safetensors download gating and making safetensors a required dependency are distinct concerns." + }, + { + "left": "pull_request:2086", + "right": "pull_request:2489", + "accept": false, + "reason": "LoRA docs and Core ML docs are separate documentation additions." + }, + { + "left": "pull_request:21", + "right": "pull_request:828", + "accept": false, + "reason": "These are different test suites and different model families; no shared bug/change." } ] }, @@ -172475,17 +172622,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 223, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2057", - "pull_request:2086", - "pull_request:21", "pull_request:2118", "pull_request:2122", "pull_request:2126", @@ -172505,8 +172649,11 @@ "pull_request:2293", "pull_request:2322", "pull_request:2327", - "pull_request:2489", + "pull_request:2329", + "pull_request:2335", + "pull_request:2337", "pull_request:2542", + "pull_request:286", "pull_request:2861", "pull_request:289", "pull_request:3012", @@ -172519,7 +172666,6 @@ "pull_request:3972", "pull_request:3976", "pull_request:41", - "pull_request:4177", "pull_request:4211", "pull_request:4214", "pull_request:4239", @@ -172533,9 +172679,6 @@ "pull_request:842" ], "soft_pairs": [ - "pull_request:2057|pull_request:4177", - "pull_request:2086|pull_request:2489", - "pull_request:21|pull_request:828", "pull_request:2118|pull_request:4239", "pull_request:2122|pull_request:3798", "pull_request:2126|pull_request:3321", @@ -172566,240 +172709,271 @@ "pull_request:2282|pull_request:4380", "pull_request:2293|pull_request:4211", "pull_request:2322|pull_request:3663", - "pull_request:2327|pull_request:2861" + "pull_request:2327|pull_request:2861", + "pull_request:2329|pull_request:289", + "pull_request:2329|pull_request:828", + "pull_request:2335|pull_request:286", + "pull_request:2337|pull_request:289", + "pull_request:2337|pull_request:3972", + "pull_request:2337|pull_request:839", + "pull_request:2337|pull_request:842" ], - "prepared_review_unit_hash": "7959e9c9091d544d2dd01d4c974e7b74f7f6dca77199ef5d57051e8db11c4a0c", + "prepared_review_unit_hash": "9be03f22a11b43905feda38e467580634a57ccf33261c0c93ae85b05f40ec7d7", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 34, - "serialized_chars": 96919, - "estimated_input_tokens": 24230, - "estimated_eval_tokens": 48716 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 38, + "serialized_chars": 95329, + "estimated_input_tokens": 23833, + "estimated_eval_tokens": 47922 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated Diffusers PRs (docs updates, typo fixes, small refactors, and a few code changes). The soft edges are mostly surface-level similarity and do not look like the same underlying bug or change.", + "summary": "This cluster is mostly a grab bag of unrelated tiny PRs: typo fixes, docs edits, test cleanup, and a few small code changes. The surface similarity is high, but the underlying changes do not match closely enough to treat them as duplicates.", "confidence": 0.96, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:2057", - "right": "pull_request:4177", - "accept": false, - "reason": "Both mention safetensors, but one changes pipeline download ignore patterns while the other makes safetensors a required dependency; different problems and code paths." - }, - { - "left": "pull_request:2086", - "right": "pull_request:2489", - "accept": false, - "reason": "Both are docs additions, but one adds LoRA docs and the other adds Core ML docs; unrelated content." - }, - { - "left": "pull_request:21", - "right": "pull_request:828", - "accept": false, - "reason": "Early model-test PR vs Flax test completion; different models, different goals, not the same fix." - }, { "left": "pull_request:2118", "right": "pull_request:4239", "accept": false, - "reason": "One fixes an import-utils spelling bug; the other is an SDXL LoRA script change. No shared code-path issue." + "reason": "Different naming fixes in different files; one corrects an import-utils typo, the other changes SDXL training script imports/logic." }, { "left": "pull_request:2122", "right": "pull_request:3798", "accept": false, - "reason": "One is safetensors docs, the other is a comment cleanup in training scripts; unrelated." + "reason": "Both are docs-related, but one adds safetensors docs while the other fixes comments in training scripts; no same underlying change." }, { "left": "pull_request:2126", "right": "pull_request:3321", "accept": false, - "reason": "Both are docs typos, but one removes a duplicate torch_dtype docstring entry and the other expands controlling-generation docs; not the same change." + "reason": "Unrelated docstring cleanup in pipeline_utils vs a large docs expansion for controlling generation." }, { "left": "pull_request:2130", "right": "pull_request:289", "accept": false, - "reason": "One trims redundant allow_patterns in from_pretrained; the other is a scheduler seed refactor across pipelines. Different functionality." + "reason": "Different fixes in pipeline_utils allow_patterns vs scheduler/generator refactor across multiple pipelines." }, { "left": "pull_request:2130", "right": "pull_request:3972", "accept": false, - "reason": "One is pipeline allow-pattern cleanup; the other is documentation workflow YAML restructuring. Unrelated." + "reason": "Different code paths: allow_patterns cleanup versus GitHub Actions workflow refactor." }, { "left": "pull_request:2130", "right": "pull_request:839", "accept": false, - "reason": "Pipeline download pattern cleanup vs README auth-token removal; different issues." + "reason": "Different files and issues; pipeline loading patterns vs example README token usage." }, { "left": "pull_request:2130", "right": "pull_request:842", "accept": false, - "reason": "Pipeline download pattern cleanup vs test API return-shape cleanup; unrelated." + "reason": "Different scope; pipeline_utils loader cleanup does not match removing ['sample'] indexing in tests." }, { "left": "pull_request:2134", "right": "pull_request:4239", "accept": false, - "reason": "Dreambooth README typo fix vs SDXL LoRA code changes; not the same underlying bug." + "reason": "Both are typo-oriented, but one is a README spelling fix and the other is a code-level SDXL training script change." }, { "left": "pull_request:2137", "right": "pull_request:4239", "accept": false, - "reason": "Typo/docs fixes in loaders.py versus substantive SDXL LoRA training script changes; different scope and code paths." + "reason": "Loaders.py typo/docstring cleanup is unrelated to the SDXL text_encoder code changes." }, { "left": "pull_request:2138", "right": "pull_request:3012", "accept": false, - "reason": "Doc typo in loaders overview vs comment typo in textual inversion training script; unrelated." + "reason": "Docs typo fix in loaders.mdx versus a comment typo in a training script; unrelated content." }, { "left": "pull_request:2138", "right": "pull_request:3331", "accept": false, - "reason": "Loader docs typo vs inpainting docs typo; both docs-only but different pages and changes." + "reason": "One is loaders API docs, the other is an inpainting docs variable-name typo; different documents and fixes." }, { "left": "pull_request:2138", "right": "pull_request:3976", "accept": false, - "reason": "Loader docs typo vs safetensors error-message fix in code; different artifacts and purposes." + "reason": "Different typo fix areas: loaders docs vs safetensors error-message typos in code." }, { "left": "pull_request:217", "right": "pull_request:41", "accept": false, - "reason": "Very old safety-check bug fix vs comment cleanup in resnets; no shared bug." + "reason": "Both touch comments/cleanup, but one fixes a safety checker bug and the other only comment formatting in resnet files." }, { "left": "pull_request:217", "right": "pull_request:693", "accept": false, - "reason": "Safety-check logic fix vs BibTeX citation formatting; unrelated." + "reason": "Safety-check logic bug fix is unrelated to a BibTeX citation formatting change." }, { "left": "pull_request:217", "right": "pull_request:75", "accept": false, - "reason": "Safety-check fix vs attention class constructor refactor for Glide; different code paths." + "reason": "Code bug in safety checker vs attention initialization/refactor; different concrete problems." }, { "left": "pull_request:2188", "right": "pull_request:289", "accept": false, - "reason": "Docs removing license warning vs scheduler seed refactor; not the same change." + "reason": "Docs removal of license text is unrelated to the generator/scheduler refactor in sampling pipelines." }, { "left": "pull_request:2188", "right": "pull_request:3972", "accept": false, - "reason": "Docs license removal vs documentation workflow YAML change; unrelated." + "reason": "Docs-only license removal vs workflow refactor; no shared bug/change." }, { "left": "pull_request:2188", "right": "pull_request:839", "accept": false, - "reason": "Docs license removal vs README auth-token cleanup; different docs issues." + "reason": "Different docs: inpaint license warning removal vs token usage cleanup in DreamBooth README." }, { "left": "pull_request:2188", "right": "pull_request:842", "accept": false, - "reason": "Docs license removal vs test API cleanup; unrelated." + "reason": "Inpaint docs license section removal is unrelated to test output-access cleanup." }, { "left": "pull_request:2197", "right": "pull_request:2227", "accept": false, - "reason": "Both touch Stable Diffusion inpaint-related code, but one fixes a prompt_embeds typo and input handling, while the other corrects negative_prompt spelling across many pipelines; not the same concrete bug." + "reason": "Both touch prompt argument docs, but one is a typo fix in a single inpaint pipeline; the other updates negative_prompt docs across many pipelines." }, { "left": "pull_request:2198", "right": "pull_request:4214", "accept": false, - "reason": "Missing-keys load error handling vs custom diffusion bf16 fix; different functionality and failure modes." + "reason": "Missing-key load error handling is unrelated to the custom diffusion bf16 error fix." }, { "left": "pull_request:22", "right": "pull_request:828", "accept": false, - "reason": "Model-loading warning typo fix vs Flax test expansion; unrelated." + "reason": "Very different changes: model loading warnings/error messages vs Flax test completion and tolerance updates." }, { "left": "pull_request:2208", "right": "pull_request:3262", "accept": false, - "reason": "xFormers docs update vs IF model naming update; both docs, but different topics and targets." + "reason": "xFormers docs update for installation is unrelated to IF model naming changes." }, { "left": "pull_request:2208", "right": "pull_request:3370", "accept": false, - "reason": "Both docs-related, but one updates xFormers installation guidance and the other updates PyTorch 2.0 optimization docs; not duplicates." + "reason": "Both are optimization docs, but one is a targeted xFormers install update and the other is a broad PyTorch 2.0 doc rewrite." }, { "left": "pull_request:2227", "right": "pull_request:4320", "accept": false, - "reason": "Negative_prompt typo across pipelines vs ONNX docs typo fix; unrelated." + "reason": "Negative_prompt typo docs across pipelines is unrelated to a small ONNX Runtime docs wording fix." }, { "left": "pull_request:2267", "right": "pull_request:737", "accept": false, - "reason": "Both remove unwanted datasets usage in example scripts/docs, but one changes logging imports and the other removes use_auth_token instructions. Similar theme, not same fix." + "reason": "Both edit training examples, but one removes datasets logging/imports and the other removes use_auth_token instructions." }, { "left": "pull_request:2282", "right": "pull_request:2542", "accept": false, - "reason": "Misc fixes across scripts/schedulers/tests vs controlnet docs fixes; no common bug." + "reason": "Mixed script/doc typo fixes in conversion tooling vs ControlNet docs link fixes; not the same change." }, { "left": "pull_request:2282", "right": "pull_request:4380", "accept": false, - "reason": "Misc script/scheduler fixes vs mixed docs additions for SDXL training. Different changes." + "reason": "Conversion-script typo fixes are unrelated to SDXL training docs additions." }, { "left": "pull_request:2293", "right": "pull_request:4211", "accept": false, - "reason": "Docstring fixes for ckpt conversion vs SDXL inpainting docstring/example fix; unrelated pipeline areas." + "reason": "Docstring fixes in convert_from_ckpt.py are unrelated to SDXL inpainting docstring corrections." }, { "left": "pull_request:2322", "right": "pull_request:3663", "accept": false, - "reason": "Layer.ai example/update PR vs Compel textual-inversion docs update; different artifacts and intent." + "reason": "These are not the same underlying update; one is a new example app and the other is a docs addition for textual inversion utilities." }, { "left": "pull_request:2327", "right": "pull_request:2861", "accept": false, - "reason": "UNet input-length validation fix vs StableUnCLIP img2img test for image_embeds=None; different bugs and code paths." + "reason": "UNet input validation changes are unrelated to adding a Stable UnCLIP test for image_embeds=None." + }, + { + "left": "pull_request:2329", + "right": "pull_request:289", + "accept": false, + "reason": "Workflow refactor and seed-handling refactor are different kinds of changes with no shared bug." + }, + { + "left": "pull_request:2329", + "right": "pull_request:828", + "accept": false, + "reason": "CI workflow refactor is unrelated to Flax test changes." + }, + { + "left": "pull_request:2335", + "right": "pull_request:286", + "accept": false, + "reason": "Issue template/forum link updates are unrelated to a broken README link fix." + }, + { + "left": "pull_request:2337", + "right": "pull_request:289", + "accept": false, + "reason": "Test pruning in common pipeline tests is unrelated to generator/seeding refactors in pipelines." + }, + { + "left": "pull_request:2337", + "right": "pull_request:3972", + "accept": false, + "reason": "Removing unnecessary tests is unrelated to a documentation workflow restructure." + }, + { + "left": "pull_request:2337", + "right": "pull_request:839", + "accept": false, + "reason": "Different scope: test cleanup versus README auth-token cleanup." + }, + { + "left": "pull_request:2337", + "right": "pull_request:842", + "accept": false, + "reason": "Test removal in one area is not the same as changing pipeline return handling in another test file." } ] }, @@ -172815,17 +172989,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 224, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2329", - "pull_request:2335", - "pull_request:2337", "pull_request:2349", "pull_request:2362", "pull_request:2367", @@ -172842,10 +173013,11 @@ "pull_request:2533", "pull_request:2539", "pull_request:2542", + "pull_request:2557", + "pull_request:2569", "pull_request:2641", "pull_request:2715", "pull_request:274", - "pull_request:286", "pull_request:2873", "pull_request:289", "pull_request:3022", @@ -172853,12 +173025,14 @@ "pull_request:3271", "pull_request:3389", "pull_request:340", + "pull_request:3561", "pull_request:3687", "pull_request:3757", "pull_request:391", "pull_request:3972", "pull_request:403", "pull_request:4050", + "pull_request:4320", "pull_request:4352", "pull_request:4380", "pull_request:4442", @@ -172868,18 +173042,10 @@ "pull_request:4807", "pull_request:710", "pull_request:800", - "pull_request:828", "pull_request:839", "pull_request:842" ], "soft_pairs": [ - "pull_request:2329|pull_request:289", - "pull_request:2329|pull_request:828", - "pull_request:2335|pull_request:286", - "pull_request:2337|pull_request:289", - "pull_request:2337|pull_request:3972", - "pull_request:2337|pull_request:839", - "pull_request:2337|pull_request:842", "pull_request:2349|pull_request:2873", "pull_request:2349|pull_request:391", "pull_request:2349|pull_request:710", @@ -172906,252 +173072,212 @@ "pull_request:2539|pull_request:4807", "pull_request:2542|pull_request:3757", "pull_request:2542|pull_request:403", - "pull_request:2542|pull_request:4380" + "pull_request:2542|pull_request:4380", + "pull_request:2557|pull_request:3561", + "pull_request:2569|pull_request:4320" ], - "prepared_review_unit_hash": "b64f6b322ca4ae65977730695a3c397cad43f49b843ec137caf176b90b393566", + "prepared_review_unit_hash": "b7d0770b35d0ff45e981578bf530826b69e997f02e957e7ac3c41fa1ec8eb002", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 34, - "serialized_chars": 99611, - "estimated_input_tokens": 24903, - "estimated_eval_tokens": 50062 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 97436, + "estimated_input_tokens": 24359, + "estimated_eval_tokens": 48974 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a noisy mix of unrelated docs, test, workflow, and small refactor PRs; the soft links are driven by superficial similarity (same subsystem, same file family, or both being cleanup/docs work) rather than the same concrete change.", - "confidence": 0.25, + "summary": "This cluster is a set of mostly unrelated docs, test, and small code PRs that were linked only by weak textual similarity; I do not see any true duplicate pair among the soft edges.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 2337 is the clearest concrete cleanup in the set and has the strongest local similarity to other test-cleanup items, so it is the most plausible representative if one PR must anchor the cluster.", + "canonical_pr_reason": "No issue exists in this cluster, so there is no canonical issue. As a representative PR, #4508 is the clearest self-contained code fix with a concrete runtime behavior change and a matching test update.", "best_issue_reason": null, - "best_pr_reason": "PR 2337 is a self-contained, well-scoped test cleanup with multiple related test-oriented soft links; it is the best representative of the cluster, though not a duplicate of the others.", + "best_pr_reason": "#4508 is the strongest standalone PR in the set: it fixes a specific `from_pretrained` argument-forwarding bug (`use_onnx`) and includes a targeted test update, unlike the many docs-only or unrelated maintenance PRs.", "soft_edge_verdicts": [ - { - "left": "pull_request:2329", - "right": "pull_request:289", - "accept": false, - "reason": "Different changes: workflow refactor for push tests vs removing seed-related code in pipeline tests. Shared testing context is too broad." - }, - { - "left": "pull_request:2329", - "right": "pull_request:828", - "accept": false, - "reason": "Unrelated test areas: CI workflow cleanup vs Flax pipeline test updates." - }, - { - "left": "pull_request:2335", - "right": "pull_request:286", - "accept": false, - "reason": "Both touch docs/links, but one adds a forum contact link and the other fixes a README link target. Not the same bug/change." - }, - { - "left": "pull_request:2337", - "right": "pull_request:289", - "accept": false, - "reason": "Both are cleanup-oriented, but one removes unnecessary tests while the other removes set_seed plumbing across pipelines. Different code paths and goals." - }, - { - "left": "pull_request:2337", - "right": "pull_request:3972", - "accept": false, - "reason": "Test cleanup vs documentation workflow restructuring; no shared fix." - }, - { - "left": "pull_request:2337", - "right": "pull_request:839", - "accept": false, - "reason": "Different scope: removing tests versus removing an auth token flag from a dreambooth README." - }, - { - "left": "pull_request:2337", - "right": "pull_request:842", - "accept": false, - "reason": "Both involve tests, but one removes obsolete tests while the other changes pipeline output accessors. Not the same underlying issue." - }, { "left": "pull_request:2349", "right": "pull_request:2873", "accept": false, - "reason": "Different documentation edits in different files; one adds scheduler overview content, the other just rephrases contribution text." + "reason": "Both are docs PRs, but one adds scheduler docs while the other edits contribution wording; different topics and no shared bug/change." }, { "left": "pull_request:2349", "right": "pull_request:391", "accept": false, - "reason": "Both are scheduler docs, but they cover different schedulers and different pages. Not one shared doc fix." + "reason": "One documents Karras schedulers, the other adds PNDM docs; separate documentation work, not the same change." }, { "left": "pull_request:2349", "right": "pull_request:710", "accept": false, - "reason": "Different doc targets: scheduler overview prose vs a docstring formatting fix in model classes." + "reason": "Karras scheduler overview docs vs a docstring fix for UNet models; unrelated content." }, { "left": "pull_request:2362", "right": "pull_request:3389", "accept": false, - "reason": "One adds an MPS skip decorator for tests; the other is an attention refactor integration-test branch. Different problems." + "reason": "MPS test skip helper vs attention refactor integration testing; different purposes and code paths." }, { "left": "pull_request:2367", "right": "pull_request:274", "accept": false, - "reason": "Training-script checkpoint limit support is unrelated to adding a Stable Diffusion README." + "reason": "Checkpoint-limit training script updates vs adding a Stable Diffusion README; unrelated." }, { "left": "pull_request:2386", "right": "pull_request:2715", "accept": false, - "reason": "UniPC doc cleanup is unrelated to broad typo fixes across model/pipeline docs and code comments." + "reason": "UniPC docs cleanup vs broad typo fixes across models/pipelines; not the same concrete issue." }, { "left": "pull_request:2398", "right": "pull_request:340", "accept": false, - "reason": "Pix2Pix Zero docs update versus img2img scheduler support. Different artifacts and no shared code path." + "reason": "Pix2Pix Zero docs expansion vs re-adding LMS scheduler support to img2img; different artifacts and behavior." }, { "left": "pull_request:2400", "right": "pull_request:4457", "accept": false, - "reason": "Both touch community pipeline docs, but one adds a new UnCLIP community pipeline and the other makes TensorRT inpaint example fixes. Not the same change." + "reason": "Both touch community docs, but they introduce different pipelines and examples; not the same underlying change." }, { "left": "pull_request:2415", "right": "pull_request:4050", "accept": false, - "reason": "Pipeline loading bug fix versus README demo table edits; unrelated files and outcomes." + "reason": "Pipeline utility bug handling vs README demo table updates; unrelated." }, { "left": "pull_request:2415", "right": "pull_request:4508", "accept": false, - "reason": "Both touch pipeline_utils.py, but they fix different behaviors: handling missing transformers vs use_onnx parameter plumbing. Not mergeable as one duplicate fix." + "reason": "Same file, but different bugs: one guards missing transformers in version checks, the other forwards `use_onnx`; they are distinct fixes." }, { "left": "pull_request:2428", "right": "pull_request:289", "accept": false, - "reason": "Docs author cleanup versus removing set_seed; different domains." + "reason": "Docs-only cleanup for panorama/pix2pix pages vs a refactor removing `set_seed`; no shared code-path problem." }, { "left": "pull_request:2428", "right": "pull_request:3972", "accept": false, - "reason": "Documentation wording/link cleanup is unrelated to build workflow restructuring." + "reason": "Docs page cleanup vs workflow restructuring; unrelated." }, { "left": "pull_request:2428", "right": "pull_request:839", "accept": false, - "reason": "Different docs edits in different files; no shared underlying bug." + "reason": "Docs cleanup vs removing `use_auth_token` from Dreambooth README; unrelated." }, { "left": "pull_request:2428", "right": "pull_request:842", "accept": false, - "reason": "README/docs cleanup vs test API output cleanup; not the same change." + "reason": "Docs cleanup vs test assertion change in pipelines; unrelated." }, { "left": "pull_request:2436", "right": "pull_request:3022", "accept": false, - "reason": "Docs demo link addition is unrelated to an attention-processor type union fix." + "reason": "Adding a demo link in docs vs adding `AttnProcessor2_0` to a type union; different scope and target." }, { "left": "pull_request:2436", "right": "pull_request:3271", "accept": false, - "reason": "Demo/docs addition and training-link docs are both documentation, but they address different pages and content." + "reason": "Attend-and-Excite docs update vs adding training-script links across docs; unrelated documentation changes." }, { "left": "pull_request:2467", "right": "pull_request:327", "accept": false, - "reason": "Broad latent-output support across many pipelines is not the same as typing the PNDM pipeline." - }, - { - "left": "pull_request:2481", - "right": "pull_request:3687", - "accept": false, - "reason": "Training-script bug fix versus doc-builder testing placeholder text; unrelated." + "reason": "Both involve pipeline output/API shape, but one adds latent output support broadly while the other is type hints for PNDM; not the same concrete fix." }, { "left": "pull_request:2481", - "right": "pull_request:4746", - "accept": false, - "reason": "A concrete training-script logging fix is unrelated to the SDXL inpainting draft work." - }, - { - "left": "pull_request:2481", - "right": "pull_request:800", + "right": "pull_request:4352", "accept": false, - "reason": "Training-script fix versus WIP VQ Diffusion implementation; no shared code path." + "reason": "Unconditional training-script test fix vs Kandinsky combined pipeline feature additions; unrelated." }, { "left": "pull_request:2487", "right": "pull_request:4442", "accept": false, - "reason": "Both add guides, but one is JAX/Flax Stable Diffusion and the other is distilled SD. Different content and target use cases." + "reason": "JAX/Flax Stable Diffusion docs vs distilled SD docs; both docs, but different guides and topics." }, { "left": "pull_request:2489", "right": "pull_request:2641", "accept": false, - "reason": "Core ML docs addition versus doc-builder notebook wiring; unrelated docs infrastructure work." + "reason": "CoreML docs addition vs doc-builder workflow/notebook config changes; unrelated." }, { "left": "pull_request:2533", "right": "pull_request:3687", "accept": false, - "reason": "CI/testing WIP changes versus a doc-builder test string edit; not the same change." + "reason": "RC1 testing workflow/setup pin vs a doc-builder test PR; different changes and intent." }, { "left": "pull_request:2533", "right": "pull_request:4746", "accept": false, - "reason": "Testing RC1 setup changes are unrelated to SDXL inpainting work." + "reason": "Dependency/workflow test PR vs SDXL inpainting WIP code changes; unrelated." }, { "left": "pull_request:2533", "right": "pull_request:800", "accept": false, - "reason": "Temporary testing/setup changes versus the VQ Diffusion implementation draft; unrelated." + "reason": "RC1 setup/workflow test PR vs VQ-Diffusion implementation work; unrelated." }, { "left": "pull_request:2539", "right": "pull_request:4807", "accept": false, - "reason": "Doc default-value fix in a pipeline docstring versus adding a Chinese Stable Diffusion guide. Different docs and issues." + "reason": "A small doc default-value fix in code vs adding Chinese Stable Diffusion docs; unrelated." }, { "left": "pull_request:2542", "right": "pull_request:3757", "accept": false, - "reason": "Both are ControlNet-related, but one is docs/navigation cleanup and the other adds a model constructor parameter. Different layers and fixes." + "reason": "ControlNet docs/overview updates vs a ControlNet model parameter change; same subsystem but not the same bug/change." }, { "left": "pull_request:2542", "right": "pull_request:403", "accept": false, - "reason": "ControlNet docs fixes are unrelated to philosophy text edits." + "reason": "ControlNet docs fixes vs philosophy wording cleanup; unrelated." }, { "left": "pull_request:2542", "right": "pull_request:4380", "accept": false, - "reason": "Both are minor docs fixes, but they affect different pages and content. No shared bug." + "reason": "ControlNet docs changes vs broader training-doc edits; not the same underlying change." + }, + { + "left": "pull_request:2557", + "right": "pull_request:3561", + "accept": false, + "reason": "Both touch Dreambooth, but one adds intermediate logging and the other changes Dreambooth IF stage II upscaling behavior; distinct PRs." + }, + { + "left": "pull_request:2569", + "right": "pull_request:4320", + "accept": false, + "reason": "OpenVINO docs addition vs ONNX docs typo cleanup; different optimization docs and different edits." } ] }, @@ -173167,16 +173293,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 225, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2557", - "pull_request:2569", "pull_request:26", "pull_request:2608", "pull_request:2667", @@ -173196,12 +173320,13 @@ "pull_request:2897", "pull_request:2912", "pull_request:2958", + "pull_request:2959", + "pull_request:2968", "pull_request:3054", "pull_request:3172", "pull_request:3449", - "pull_request:3561", "pull_request:3757", - "pull_request:4320", + "pull_request:3846", "pull_request:4556", "pull_request:4614", "pull_request:4741", @@ -173217,6 +173342,7 @@ "pull_request:774", "pull_request:78", "pull_request:802", + "pull_request:823", "pull_request:824", "pull_request:839", "pull_request:842", @@ -173224,8 +173350,6 @@ "pull_request:948" ], "soft_pairs": [ - "pull_request:2557|pull_request:3561", - "pull_request:2569|pull_request:4320", "pull_request:26|pull_request:689", "pull_request:2608|pull_request:4741", "pull_request:2667|pull_request:2958", @@ -173252,210 +173376,212 @@ "pull_request:2897|pull_request:4614", "pull_request:2912|pull_request:3757", "pull_request:2958|pull_request:3757", - "pull_request:2958|pull_request:4811" + "pull_request:2958|pull_request:4811", + "pull_request:2959|pull_request:3846", + "pull_request:2968|pull_request:823" ], - "prepared_review_unit_hash": "d0c765a55dd66b19b86f9b7279e6e8ae97304e29aa84fb506315ad3cadd1bc08", + "prepared_review_unit_hash": "5fa264394462f3b72f36e882c2a70cd1408f7b4feacc6595ad50e931b005049a", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 95932, - "estimated_input_tokens": 23983, - "estimated_eval_tokens": 48222 + "serialized_chars": 94862, + "estimated_input_tokens": 23716, + "estimated_eval_tokens": 47688 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is not a true duplicate set: it mixes unrelated docs, test, CI, refactor, and feature PRs. The soft overlaps are mostly superficial (same broad subsystem or shared filenames) rather than the same concrete bug/change, so all soft edges should be rejected.", - "confidence": 0.97, + "summary": "Mostly unrelated PRs with a few superficial title/file overlaps; I don\u2019t see any true duplicate pairs here. The strongest single change is the post-processing fix in PR 2968, but it is not a canonical duplicate target for the rest of the set.", + "confidence": 0.86, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR is a clear duplicate center for this mixed set; the items span docs typos, example scripts, tests, and unrelated API/features. If a representative PR must be chosen, 2968 is the cleanest concrete bug fix.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 2968 is the strongest standalone fix: it corrects a concrete post-processing bug in two image-to-image pipelines and is a plausible mergeable unit.", "soft_edge_verdicts": [ - { - "left": "pull_request:2557", - "right": "pull_request:3561", - "accept": false, - "reason": "Both touch DreamBooth training, but one adds intermediate logging and the other adds IF stage II upscaling; different change sets and code paths." - }, - { - "left": "pull_request:2569", - "right": "pull_request:4320", - "accept": false, - "reason": "Both are docs, but OpenVINO vs ONNX Runtime are separate optimization guides with unrelated content." - }, { "left": "pull_request:26", "right": "pull_request:689", "accept": false, - "reason": "Different subsystems: scheduler math fix vs attention implementation optimization." + "reason": "Both touch diffusion internals, but one fixes a scheduler state bug and the other is a performance refactor in attention; different underlying changes." }, { "left": "pull_request:2608", "right": "pull_request:4741", "accept": false, - "reason": "One fixes README typos; the other fixes dummy imports for AudioLDM2. No shared bug." + "reason": "Both are small docs/import cleanup style PRs, but they affect unrelated areas and do not fix the same bug or change." }, { "left": "pull_request:2667", "right": "pull_request:2958", "accept": false, - "reason": "Both involve inpainting/controlnet, but one adds an inpaint example/strength handling and the other adds a new ControlNet inpaint pipeline." + "reason": "Both mention inpainting/controlnet-related pipelines, but they implement different features in different code paths and are not the same change." }, { "left": "pull_request:2677", "right": "pull_request:4556", "accept": false, - "reason": "Both update docs navigation, but one adds training ControlNet docs and the other adds SDXL ControlNet API docs; separate pages and purposes." + "reason": "Both edit docs navigation, but one adds training docs and the other adds API docs for a different ControlNet variant." }, { "left": "pull_request:2693", "right": "pull_request:570", "accept": false, - "reason": "A doc image-link fix and BaseOutput dict initialization are unrelated." + "reason": "An image link typo fix and a BaseOutput dict-init fix are unrelated." }, { "left": "pull_request:27", "right": "pull_request:286", "accept": false, - "reason": "Both are small README/doc fixes, but they address different text and different links/checklist content." + "reason": "Both are minor README edits, but one fixes a checklist format and the other fixes a pipeline doc link." }, { "left": "pull_request:27", "right": "pull_request:662", "accept": false, - "reason": "Checklist formatting in the README is unrelated to the textual inversion 6GB example changes." + "reason": "Checklist formatting in the top-level README is unrelated to the textual inversion example overhaul." }, { "left": "pull_request:27", "right": "pull_request:718", "accept": false, - "reason": "Both are typo/README cleanup PRs, but they fix different documents and issues." + "reason": "Both are typo/cleanup PRs, but they change different files and different content; not the same bug." }, { "left": "pull_request:27", "right": "pull_request:774", "accept": false, - "reason": "Checklist formatting and test expectation adjustments are unrelated changes." + "reason": "A checklist formatting tweak and a test adjustment are unrelated changes." }, { "left": "pull_request:2759", "right": "pull_request:3449", "accept": false, - "reason": "Both affect test infrastructure, but one is a broad test-run PR and the other specifically reworks MPS/ONNX test jobs." + "reason": "Both are test/CI related, but one changes model training/test coverage and the other reworks workflow jobs; not the same fix." }, { "left": "pull_request:2764", "right": "pull_request:3172", "accept": false, - "reason": "Both mention optimization/docs, but Torch 2.0 benchmark guidance and deterministic algorithms are different topics." + "reason": "Both touch docs about modern PyTorch features, but one is a benchmark/docs update and the other is a reproducibility guide update tied to deterministic algorithms." }, { "left": "pull_request:2767", "right": "pull_request:948", "accept": false, - "reason": "LoRA script argument handling and a DDPM comment typo are unrelated." + "reason": "A LoRA script argument rename is unrelated to a DDPM doc comment typo." }, { "left": "pull_request:2794", "right": "pull_request:768", "accept": false, - "reason": "Both edit DreamBooth code, but one adds validation prompt support while the other is a generic enhancement scaffold; not the same concrete fix." + "reason": "Both modify dreambooth training, but one adds validation support while the other is a separate default-args enhancement." }, { "left": "pull_request:2795", "right": "pull_request:3054", "accept": false, - "reason": "LoRA training code change vs contribution docs update; no shared bug or change." + "reason": "One is a LoRA training-script bugfix; the other is a docs wording cleanup. Different artifacts and change types." }, { "left": "pull_request:2808", "right": "pull_request:705", "accept": false, - "reason": "Relaxing model tests is unrelated to CI workflow and dependency changes." + "reason": "Both involve tests/CI, but one relaxes model tests while the other changes CUDA workflow infrastructure; not the same issue." }, { "left": "pull_request:2808", "right": "pull_request:824", "accept": false, - "reason": "Both are test-related, but DiT test relaxation and Flax test additions are distinct." + "reason": "Test relaxations in different pipelines are not the same concrete bug." }, { "left": "pull_request:2823", "right": "pull_request:4807", "accept": false, - "reason": "Both are docs, but one updates English Stable UnCLIP docs and the other adds Chinese Stable Diffusion docs." + "reason": "Both are Stable UnCLIP docs, but one fixes usage examples and the other adds Chinese docs; different changes." }, { "left": "pull_request:2846", "right": "pull_request:5063", "accept": false, - "reason": "One fixes a slow test invocation; the other updates a docstring across several pipelines." + "reason": "Both touch Stable UnCLIP docs/tests, but one fixes test call order while the other updates guidance_rescale documentation." }, { "left": "pull_request:286", "right": "pull_request:78", "accept": false, - "reason": "A README link fix and a UNet mid-block refactor are unrelated." + "reason": "A broken README link fix is unrelated to a mid-block model refactor." }, { "left": "pull_request:286", "right": "pull_request:802", "accept": false, - "reason": "Both touch documentation/examples, but a link fix is not the same as an indentation fix in a code sample." + "reason": "Both are documentation edits, but one fixes a link and the other fixes code-example indentation in a different guide." }, { "left": "pull_request:289", "right": "pull_request:839", "accept": false, - "reason": "A refactor removing set_seed and a DreamBooth README token cleanup are unrelated." + "reason": "One removes set_seed from pipeline code; the other removes use_auth_token from dreambooth docs. No shared bug." }, { "left": "pull_request:289", "right": "pull_request:842", "accept": false, - "reason": "Refactoring pipeline seeding behavior and changing output access from [\"sample\"] to .images are different fixes." + "reason": "A refactor removing set_seed is unrelated to a test API cleanup." }, { "left": "pull_request:289", "right": "pull_request:87", "accept": false, - "reason": "One is a targeted refactor of sampling code; the other is a broad unconditional-model refactor. Same area, but not one mergeable change." + "reason": "Both are broad refactors, but they target entirely different subsystems and eras; not the same underlying change." }, { "left": "pull_request:2897", "right": "pull_request:4614", "accept": false, - "reason": "Stable UnCLIP docs example vs from_single_file config-file support are unrelated." + "reason": "Both concern docs/loaders around Stable UnCLIP or single-file loading, but one adds an example and the other adds a config-file parameter; different behavior." }, { "left": "pull_request:2912", "right": "pull_request:3757", "accept": false, - "reason": "A punctuation/doc wording fix and a ControlNet model API change do not address the same issue." + "reason": "A punctuation doc fix and a ControlNet constructor API change are unrelated." }, { "left": "pull_request:2958", "right": "pull_request:3757", "accept": false, - "reason": "New ControlNet inpaint pipeline and configurable conditioning channels are different features in the same broad area." + "reason": "ControlNet inpaint pipeline addition and configurable conditioning channels are separate features, not one bug fix." }, { "left": "pull_request:2958", "right": "pull_request:4811", "accept": false, - "reason": "Both add ControlNet inpaint variants, but SD ControlNet inpaint and SDXL ControlNet inpaint are separate pipeline implementations and cannot be merged as one PR." + "reason": "Both add inpaint/controlnet pipelines, but one is for ControlNet in general and the other is SDXL-specific; they are different concrete implementations." + }, + { + "left": "pull_request:2959", + "right": "pull_request:3846", + "accept": false, + "reason": "A UNet3D test correction and a random-module seeding fix are unrelated." + }, + { + "left": "pull_request:2968", + "right": "pull_request:823", + "accept": false, + "reason": "Both adjust pipeline behavior, but one fixes post-processing placement and the other fixes negative-prompt type handling/tests; different code-path problems." } ] }, @@ -173471,16 +173597,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 226, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:2959", - "pull_request:2968", "pull_request:2983", "pull_request:299", "pull_request:3022", @@ -173501,11 +173625,13 @@ "pull_request:3326", "pull_request:3331", "pull_request:3334", + "pull_request:3337", + "pull_request:3341", + "pull_request:3433", "pull_request:344", "pull_request:3471", "pull_request:3474", "pull_request:3806", - "pull_request:3846", "pull_request:3894", "pull_request:3924", "pull_request:4034", @@ -173514,6 +173640,7 @@ "pull_request:4215", "pull_request:432", "pull_request:4508", + "pull_request:4690", "pull_request:4741", "pull_request:4746", "pull_request:5239", @@ -173521,15 +173648,12 @@ "pull_request:5506", "pull_request:75", "pull_request:800", - "pull_request:823", "pull_request:826", "pull_request:909", "pull_request:972", "pull_request:975" ], "soft_pairs": [ - "pull_request:2959|pull_request:3846", - "pull_request:2968|pull_request:823", "pull_request:2983|pull_request:4508", "pull_request:299|pull_request:800", "pull_request:3022|pull_request:3474", @@ -173556,156 +173680,146 @@ "pull_request:3294|pull_request:5506", "pull_request:3318|pull_request:4746", "pull_request:3326|pull_request:3331", - "pull_request:3334|pull_request:4215" + "pull_request:3334|pull_request:4215", + "pull_request:3337|pull_request:4690", + "pull_request:3341|pull_request:3433" ], - "prepared_review_unit_hash": "7bbb8eb2b34bf1022cae48c45c12705f7401cb02c5ae25cbfb1e4c5338b07dd8", + "prepared_review_unit_hash": "343c1f1476866205fddce8f967c0c9d2ad03f6f06b2cdbafd5d5f82b143a1849", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 91284, - "estimated_input_tokens": 22821, - "estimated_eval_tokens": 45898 + "serialized_chars": 94087, + "estimated_input_tokens": 23522, + "estimated_eval_tokens": 47300 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a duplicate cluster: it mixes unrelated PRs across docs, tests, pipeline fixes, model changes, and feature additions. No single underlying bug or change ties the set together.", - "confidence": 0.99, + "summary": "This cluster is a grab bag of unrelated diffusers PRs: docs edits, type-hint cleanups, scheduler tweaks, example updates, and several distinct feature additions. The soft links are mostly superficial filename/subsystem overlap rather than the same underlying fix or change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the artifacts are heterogeneous and do not represent one shared bug or change.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No best PR for deduplication: there is no single concrete fix that the rest of the items can be compared against.", + "best_pr_reason": "No single PR is a good canonical representative because the cluster spans many unrelated changes with no common bug or shared code-path fix.", "soft_edge_verdicts": [ - { - "left": "pull_request:2959", - "right": "pull_request:3846", - "accept": false, - "reason": "UNet3D norm-group test adjustment vs random seeding fix; unrelated codepaths and intents." - }, - { - "left": "pull_request:2968", - "right": "pull_request:823", - "accept": false, - "reason": "Img2img post-processing/safety-checker flow vs negative-prompt type-mismatch handling; different bugs." - }, { "left": "pull_request:2983", "right": "pull_request:4508", "accept": false, - "reason": "Contributing docs update vs from_pretrained use_onnx parameter fix; no shared underlying change." + "reason": "Docs for installing test deps vs ONNX parameter handling in from_pretrained; unrelated changes." }, { "left": "pull_request:299", "right": "pull_request:800", "accept": false, - "reason": "PNDM timestep computation hotfix vs VQ Diffusion work; completely different areas." + "reason": "PNDM timestep fix vs VQ Diffusion model import/codebase addition; different problems and code paths." }, { "left": "pull_request:3022", "right": "pull_request:3474", "accept": false, - "reason": "Code fix adding AttnProcessor2_0 to a union vs docs page for attention processors; docs-only versus API change." + "reason": "Both touch attention_processor.py, but one fixes a type-export union and the other adds docs pages and exports; not the same change." }, { "left": "pull_request:3042", "right": "pull_request:3924", "accept": false, - "reason": "Example dependency pinning vs tutorial parameter rename; unrelated documentation/example maintenance." + "reason": "Accelerate minimum version bump for example requirements vs a docs parameter rename; unrelated." }, { "left": "pull_request:3068", "right": "pull_request:909", "accept": false, - "reason": "Textual inversion docs/load example fix vs inpainting test model update; different features and failure modes." + "reason": "Textual inversion docs update vs inpaint test model-id update; different artifacts and purpose." }, { "left": "pull_request:3069", "right": "pull_request:909", "accept": false, - "reason": "Flax TPU test expectations vs inpainting test model update; not the same bug." + "reason": "Flax TPU test expectation changes vs pipeline inpaint test updates; unrelated test maintenance." }, { "left": "pull_request:3108", "right": "pull_request:4741", "accept": false, - "reason": "Empty/unknown checkpoint fix vs AudioLDM2 dummy import correction; no concrete overlap." + "reason": "Empty/unknown checkpoint-fix PR vs AudioLDM2 dummy-import fix; no shared bug." }, { "left": "pull_request:3108", "right": "pull_request:972", "accept": false, - "reason": "Empty/unknown checkpoint fix vs README typo correction; unrelated." + "reason": "Empty/unknown checkpoint-fix PR vs torch_dtype typo fix in README; unrelated." }, { "left": "pull_request:3108", "right": "pull_request:975", "accept": false, - "reason": "Empty/unknown checkpoint fix vs documentation title capitalization; unrelated." + "reason": "Empty/unknown checkpoint-fix PR vs documentation title typo; unrelated." }, { "left": "pull_request:312", "right": "pull_request:41", "accept": false, - "reason": "README link correction vs resnet comment cleanup; superficial similarity only." + "reason": "README link typo fix vs comment cleanup in resnet/unet files; different maintenance areas." }, { "left": "pull_request:312", "right": "pull_request:432", "accept": false, - "reason": "Both are link/docs edits, but they target different broken links in different docs pages." + "reason": "Both are docs/link fixes, but one updates README community link and the other rewrites notebook links across many docs; not the same concrete change." }, { "left": "pull_request:312", "right": "pull_request:75", "accept": false, - "reason": "README link fix vs Glide attention implementation fix; different code paths." + "reason": "README link fix vs attention implementation behavior change for Glide; unrelated." }, { "left": "pull_request:3126", "right": "pull_request:4034", "accept": false, - "reason": "Multi-GPU example docs vs LoRA unload implementation; not the same underlying issue." + "reason": "Multi-GPU example support docs/scripts vs LoRA unload utility and tests; distinct features." }, { "left": "pull_request:3166", "right": "pull_request:3326", "accept": false, - "reason": "Textual inversion loader docs/example fix vs new 'Adapt a model' training doc; unrelated documentation work." + "reason": "Textual inversion docs fix vs generic model-adaptation docs; unrelated documentation topics." }, { "left": "pull_request:3187", "right": "pull_request:3230", "accept": false, - "reason": "x4 upscaler offload support vs adding the IF pipeline; different features entirely." + "reason": "Upcaler CPU offload change vs large IF pipeline addition; different pipeline and bug scope." }, { "left": "pull_request:3230", "right": "pull_request:3894", "accept": false, - "reason": "IF pipeline addition vs Flax UNet dropout change; unrelated components and bugs." + "reason": "IF pipeline addition vs Flax UNet dropout feature; unrelated subsystems." }, { "left": "pull_request:3230", "right": "pull_request:4172", "accept": false, - "reason": "IF pipeline addition vs Flax model mixin casting; different subsystems and purposes." + "reason": "IF pipeline addition vs Flax casting utility refactor; unrelated code paths." }, { "left": "pull_request:3230", "right": "pull_request:5239", "accept": false, - "reason": "IF pipeline addition vs AutoencoderKL encode docstring; no shared code issue." + "reason": "IF pipeline addition vs AutoencoderKL encode docstring; unrelated." }, { "left": "pull_request:3230", @@ -173717,49 +173831,61 @@ "left": "pull_request:325", "right": "pull_request:344", "accept": false, - "reason": "Score SDE VE type hints vs VAE model type hints; both typing-only, but not the same artifact or bug." + "reason": "Score SDE VE type-hint cleanup vs VAE type-hint cleanup; same style category, but different files and no shared bug." }, { "left": "pull_request:3252", "right": "pull_request:3471", "accept": false, - "reason": "Stateful offload docs vs DreamBooth torch.compile note; different documentation topics." + "reason": "Docs for stateful offload methods vs DreamBooth torch.compile note; unrelated docs." }, { "left": "pull_request:3252", "right": "pull_request:3806", "accept": false, - "reason": "Stateful offload docs vs checkpoint-loading note; unrelated docs fixes." + "reason": "Offload-state docs vs checkpoint-loading note for text-to-image training; different topics." }, { "left": "pull_request:327", "right": "pull_request:826", "accept": false, - "reason": "PNDM pipeline typing cleanup vs DreamBooth fp16 loss fix; different codepaths." + "reason": "PNDM pipeline type hints vs DreamBooth fp16/prior loss fix; unrelated code paths." }, { "left": "pull_request:3294", "right": "pull_request:5506", "accept": false, - "reason": "LoRA weight-loading implementation vs PEFT/xformers attention scaling fix; both mention LoRA, but not the same concrete bug." + "reason": "LoRA prefix implementation work vs xformers/PEFT scale handling fix; not the same concrete fix." }, { "left": "pull_request:3318", "right": "pull_request:4746", "accept": false, - "reason": "IF textual inversion work vs SDXL inpainting WIP; different features and files." + "reason": "WIP IF textual inversion vs SDXL inpainting WIP; different model families and goals." }, { "left": "pull_request:3326", "right": "pull_request:3331", "accept": false, - "reason": "New training guide vs a single inpainting docs typo; not the same change." + "reason": "New training guide for adapting a model vs an inpainting doc typo; unrelated docs." }, { "left": "pull_request:3334", "right": "pull_request:4215", "accept": false, - "reason": "Pipeline docstring cleanup vs control generation docs spacing/wording; unrelated documentation edits." + "reason": "Docstring punctuation/links fix vs controlled-generation docs cleanup; different documentation changes." + }, + { + "left": "pull_request:3337", + "right": "pull_request:4690", + "accept": false, + "reason": "Pipeline postprocessing refactor vs DPMSolverMultistepScheduler refactor; different components and code paths." + }, + { + "left": "pull_request:3341", + "right": "pull_request:3433", + "accept": false, + "reason": "Dependency/style fix PR vs doc rendering spacing fixes; unrelated maintenance." } ] }, @@ -173775,20 +173901,17 @@ "component_index": 1, "component_count": 266, "review_unit_index": 227, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:3337", - "pull_request:3341", "pull_request:3345", "pull_request:3359", "pull_request:3388", - "pull_request:3433", "pull_request:345", "pull_request:3471", "pull_request:3472", @@ -173809,7 +173932,10 @@ "pull_request:3727", "pull_request:3766", "pull_request:38", + "pull_request:3806", + "pull_request:382", "pull_request:3866", + "pull_request:390", "pull_request:3924", "pull_request:3979", "pull_request:4081", @@ -173818,8 +173944,8 @@ "pull_request:4352", "pull_request:4547", "pull_request:4607", - "pull_request:4690", "pull_request:4807", + "pull_request:5024", "pull_request:5312", "pull_request:5396", "pull_request:5470", @@ -173833,8 +173959,6 @@ "pull_request:819" ], "soft_pairs": [ - "pull_request:3337|pull_request:4690", - "pull_request:3341|pull_request:3433", "pull_request:3345|pull_request:3524", "pull_request:3345|pull_request:5622", "pull_request:3359|pull_request:3866", @@ -173861,210 +173985,218 @@ "pull_request:3709|pull_request:4807", "pull_request:3727|pull_request:4547", "pull_request:3766|pull_request:4352", - "pull_request:38|pull_request:75" + "pull_request:38|pull_request:75", + "pull_request:3806|pull_request:5024", + "pull_request:382|pull_request:390" ], - "prepared_review_unit_hash": "d85802f2e875e3f8b104faa3515f02282675130949a8cbd2de6bbfcd6c54aeba", + "prepared_review_unit_hash": "d8daef4c57f67c0df18ef02843857cccdd0bb7d67c9229d8d3a9a672225e8b34", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 99879, - "estimated_input_tokens": 24970, - "estimated_eval_tokens": 50196 + "serialized_chars": 97637, + "estimated_input_tokens": 24410, + "estimated_eval_tokens": 49076 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: most pairs are superficial matches across docs, tests, or unrelated fixes, not true duplicates. There is no clear canonical issue, and only a broad representative PR can be chosen at all.", - "confidence": 0.63, + "summary": "All candidate pairs look like unrelated PRs rather than duplicates. They span separate docs edits, example-script changes, and distinct bug fixes/features, often only sharing a broad subsystem or filename.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR cleanly subsumes the cluster; if a representative is needed, PR 3337 is the broadest substantive code change, but the set is too mixed to treat it as a true duplicate target.", + "canonical_pr_reason": "No canonical PR: the cluster is heterogeneous and does not represent one duplicate change set.", "best_issue_reason": null, - "best_pr_reason": "PR 3337 is the most substantial and broadly scoped item here, touching many pipelines with a shared postprocessing refactor, so it is the best representative PR despite the cluster mismatch.", + "best_pr_reason": "No single PR is a good representative duplicate target because the items are unrelated and should stay separate.", "soft_edge_verdicts": [ - { - "left": "pull_request:3337", - "right": "pull_request:4690", - "accept": false, - "reason": "Both are refactors, but 3337 is a VAE postprocessing change across pipelines while 4690 is a DPMSolverMultistep scheduler sigma refactor; different bugs/paths." - }, - { - "left": "pull_request:3341", - "right": "pull_request:3433", - "accept": false, - "reason": "3341 pins urllib3 and touches style tooling; 3433 only fixes docs formatting/rendering. No shared code-path problem." - }, { "left": "pull_request:3345", "right": "pull_request:3524", "accept": false, - "reason": "3345 is a README pipeline-class fix for RePaint; 3524 is a broader README overhaul. Same file, different change." + "reason": "Both touch README docs, but one is a specific pipeline-class fix in community examples and the other is a broad README rewrite; different changes." }, { "left": "pull_request:3345", "right": "pull_request:5622", "accept": false, - "reason": "A README class-name fix is unrelated to a large examples/training optimization patch." + "reason": "Completely unrelated: a small README pipeline-class fix vs a large examples/training-ops overhaul." }, { "left": "pull_request:3359", "right": "pull_request:3866", "accept": false, - "reason": "Same training script, but 3359 disables the safety checker and 3866 adds LoRA rank support; separate features, not one bug." + "reason": "Same example file, but different change: disabling safety_checker vs adding LoRA rank args; not the same bug or fix." }, { "left": "pull_request:3388", "right": "pull_request:5478", "accept": false, - "reason": "3388 updates installation docs for transformers; 5478 adds Japanese docs and workflow language support. Different docs work." + "reason": "Different documentation areas and purposes: install docs update vs full Japanese docs addition." }, { "left": "pull_request:345", "right": "pull_request:819", "accept": false, - "reason": "345 adds type hints to DDIM/DDPM pipelines; 819 fixes DDIM timestep dtype on Windows. Different issues in the scheduler area." + "reason": "Different pipelines and issues: type hints cleanup vs Windows int64 timestep bug fix." }, { "left": "pull_request:3471", "right": "pull_request:5312", "accept": false, - "reason": "3471 is a Dreambooth docs note; 5312 is model docs/typehint cleanup. Same theme of docs, not same change." + "reason": "Dreambooth docs note vs model-library typehints/docs refactor; unrelated changes." }, { "left": "pull_request:3472", "right": "pull_request:4343", "accept": false, - "reason": "3472 adds IF Dreambooth docs; 4343 adds a rank arg to SDXL LoRA training. Different training flows and fixes." + "reason": "One adds IF training docs, the other adds a rank CLI arg for SDXL LoRA training; different changes." }, { "left": "pull_request:352", "right": "pull_request:7", "accept": false, - "reason": "352 adds the diffusers CLI env command; 7 adds VQ-VAE model support. Completely unrelated." + "reason": "Commands/env CLI addition vs initial VQ-VAE model implementation; unrelated." }, { "left": "pull_request:3524", "right": "pull_request:4081", "accept": false, - "reason": "Both touch README, but 3524 is a broad README rewrite while 4081 just adds Kandinsky to a table." + "reason": "Both edit README.md, but they add different content sections/features and are not the same change." }, { "left": "pull_request:3541", "right": "pull_request:3979", "accept": false, - "reason": "3541 fixes DreamBooth attention mask collation; 3979 changes SDXL upscale/xformers detection logic. Different code paths." + "reason": "Different bug fixes in different code paths: Dreambooth attention-mask collation vs SDXL attention-processor check." }, { "left": "pull_request:3547", "right": "pull_request:4607", "accept": false, - "reason": "3547 adds bibliography references; 4607 documents PushToHubMixin. Docs additions, but unrelated content." + "reason": "Bibtex references file vs docs overview for PushToHubMixin; unrelated documentation work." }, { "left": "pull_request:3563", "right": "pull_request:5698", "accept": false, - "reason": "3563 fixes a UNet block norm condition; 5698 fixes LCM img2img behavior and LoRA handling. Different model/pipeline bugs." + "reason": "Different model/pipeline fixes in unrelated code paths; no shared underlying bug." }, { "left": "pull_request:3572", "right": "pull_request:4196", "accept": false, - "reason": "3572 is a ControlNet inpainting docs/example tweak; 4196 improves SDXL ControlNet docs and related pipeline docs. Not the same underlying change." + "reason": "Both are docs-related, but one corrects ControlNet inpaint docs and the other improves SDXL API docs; different artifacts." }, { "left": "pull_request:3577", "right": "pull_request:5729", "accept": false, - "reason": "3577 fixes broken README links; 5729 updates version checks across training scripts. Different scope and purpose." + "reason": "README link fixes vs updating minimum version checks across training scripts; not the same issue." }, { "left": "pull_request:3587", "right": "pull_request:4352", "accept": false, - "reason": "3587 adds a community pipeline and README entry; 4352 adds CPU offload/xformers support to Kandinsky combined pipelines. Different feature sets." + "reason": "Different Kandinsky-related changes: a new community pipeline vs fixes to combined pipeline helper methods." }, { "left": "pull_request:3607", "right": "pull_request:3924", "accept": false, - "reason": "3607 changes attention code and tests; 3924 updates a docs code sample to newer accelerator parameters. Not the same bug." + "reason": "Different subsystems: temb attention fix vs a docs code-snippet parameter rename." }, { "left": "pull_request:3607", "right": "pull_request:5396", "accept": false, - "reason": "3607 is an attention/model change, while 5396 fixes timezone handling in a stale-bot utility. No overlap beyond being code changes." + "reason": "Attention/LoRA runtime fix vs stale-bot timezone handling; unrelated." }, { "left": "pull_request:3614", "right": "pull_request:5526", "accept": false, - "reason": "3614 only tweaks tests/rename behavior; 5526 fixes second-order scheduler stepping logic. Different concrete problems." + "reason": "Test cleanup vs scheduler step-count logic fix; not the same bug or change." }, { "left": "pull_request:3625", "right": "pull_request:4547", "accept": false, - "reason": "3625 is a doc-builder workflow WIP; 4547 is a Flax scheduler implementation. Unrelated." + "reason": "Doc-builder workflow test PR vs a Flax scheduler implementation PR; unrelated." }, { "left": "pull_request:3625", "right": "pull_request:4807", "accept": false, - "reason": "3625 is workflow plumbing for docs testing; 4807 adds Chinese docs. Same docs project, different artifact." + "reason": "Doc-builder workflow test PR vs adding Chinese docs; unrelated." }, { "left": "pull_request:3657", "right": "pull_request:5470", "accept": false, - "reason": "3657 updates ControlNet conversion-script argument parsing; 5470 fixes ControlNet inpaint pipeline docs and API pages. Different areas." + "reason": "ControlNet checkpoint parsing tweaks vs ControlNet inpaint docs; different code paths and goals." }, { "left": "pull_request:3676", "right": "pull_request:4547", "accept": false, - "reason": "3676 is a doc-builder test WIP; 4547 is a Flax Euler scheduler PR. No shared bug/change." + "reason": "Testing doc-builder change vs Flax scheduler implementation; unrelated." }, { "left": "pull_request:3676", "right": "pull_request:4807", "accept": false, - "reason": "Test docs-builder workflow changes are unrelated to adding Japanese/Chinese docs." + "reason": "Testing doc-builder change vs adding Japanese docs; unrelated." }, { "left": "pull_request:3709", "right": "pull_request:4807", "accept": false, - "reason": "3709 adds the image processor docs page; 4807 adds Chinese docs. Different documentation additions." + "reason": "New image-processor doc page vs a full Japanese docs translation; different changes." }, { "left": "pull_request:3727", "right": "pull_request:4547", "accept": false, - "reason": "3727 is a draft Dreambooth T5 finetuning change; 4547 is a Flax scheduler WIP. Not the same code path." + "reason": "IF Dreambooth fine-tuning work vs scheduler implementation; unrelated." }, { "left": "pull_request:3766", "right": "pull_request:4352", "accept": false, - "reason": "3766 updates a Kandinsky conversion script; 4352 fixes combined pipeline runtime methods. Related project, but different change." + "reason": "Both Kandinsky-related, but one updates conversion script and the other adds combined-pipeline methods/tests; not the same concrete fix." }, { "left": "pull_request:38", "right": "pull_request:75", "accept": false, - "reason": "Both touch attention code, but 38 is a broad unification refactor while 75 is a specific Glide attention fix. Too different to merge as one PR." + "reason": "PR 38 is a broad attention-module unification; PR 75 is a later specific Glide attention fix. Related area, but not the same underlying change." + }, + { + "left": "pull_request:3806", + "right": "pull_request:5024", + "accept": false, + "reason": "Checkpoint-loading note vs scheduler step-index/noise-sampler fix; unrelated." + }, + { + "left": "pull_request:382", + "right": "pull_request:390", + "accept": false, + "reason": "DDPM pipeline docs vs Apple Silicon/MPS optimization docs; different documentation topics." + }, + { + "left": "pull_request:3866", + "right": "pull_request:3924", + "accept": false, + "reason": "LoRA rank argument enhancement vs a docs parameter-name update; unrelated changes." } ] }, @@ -174080,7 +174212,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 228, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -174088,11 +174220,8 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:3806", - "pull_request:382", "pull_request:3850", "pull_request:388", - "pull_request:390", "pull_request:391", "pull_request:3928", "pull_request:3931", @@ -174110,24 +174239,28 @@ "pull_request:408", "pull_request:4082", "pull_request:41", + "pull_request:4106", + "pull_request:4112", "pull_request:415", "pull_request:416", "pull_request:418", "pull_request:4196", "pull_request:4237", "pull_request:4238", + "pull_request:4276", "pull_request:4396", "pull_request:4748", "pull_request:4807", "pull_request:4850", "pull_request:498", "pull_request:4984", - "pull_request:5024", "pull_request:508", "pull_request:5097", + "pull_request:5456", "pull_request:5602", "pull_request:5853", "pull_request:6075", + "pull_request:6150", "pull_request:6198", "pull_request:6322", "pull_request:6337", @@ -174137,8 +174270,6 @@ "pull_request:921" ], "soft_pairs": [ - "pull_request:3806|pull_request:5024", - "pull_request:382|pull_request:390", "pull_request:3850|pull_request:5853", "pull_request:388|pull_request:391", "pull_request:388|pull_request:401", @@ -174170,240 +174301,249 @@ "pull_request:4082|pull_request:6337", "pull_request:41|pull_request:498", "pull_request:41|pull_request:693", - "pull_request:41|pull_request:75" + "pull_request:41|pull_request:75", + "pull_request:4106|pull_request:4276", + "pull_request:4106|pull_request:6150", + "pull_request:4112|pull_request:5456" ], - "prepared_review_unit_hash": "aa9e447efa86338c4af9a00951eafefd41e6614c9f7a4af81c231ac7849b3525", + "prepared_review_unit_hash": "bd72a1ce5e3ec82db4ab37a234220b206cf0458bc61ba4dc87112d8e5e693b82", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 34, - "serialized_chars": 104414, - "estimated_input_tokens": 26104, - "estimated_eval_tokens": 52464 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 35, + "serialized_chars": 107300, + "estimated_input_tokens": 26825, + "estimated_eval_tokens": 53906 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated docs updates, test fixes, and code bugfixes across different subsystems; the soft suggestions are similarity-based but do not represent the same underlying change.", - "confidence": 0.97, + "summary": "This cluster is mostly heterogeneous: docs refreshes, unrelated bug fixes, and test-only tweaks across different subsystems. I found no soft-edge pair that clearly describes the same underlying change, so I rejected all candidates.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a valid canonical representative because the items span many unrelated changes (docs, workflows, tests, schedulers, SDXL/controlnet, and miscellaneous fixes) rather than one duplicate fix.", + "canonical_pr_reason": "PR 4237 is the strongest representative if one must be chosen: it fixes a concrete, shared SDXL code-path bug, has an explicit issue target, and includes regression tests.", "best_issue_reason": null, - "best_pr_reason": "No global best PR: the cluster is not a coherent duplicate set, so picking one PR would be arbitrary and misleading.", + "best_pr_reason": "PR 4237 is the best standalone triage candidate: it is a real bug fix with tests and a clear problem statement, unlike the many docs/test-only PRs in the cluster.", "soft_edge_verdicts": [ - { - "left": "pull_request:3806", - "right": "pull_request:5024", - "accept": false, - "reason": "Unrelated changes: a checkpoint-loading note in training docs vs scheduler step-index/noise-sampler logic." - }, - { - "left": "pull_request:382", - "right": "pull_request:390", - "accept": false, - "reason": "Both are docs PRs, but for different pipelines/topics (DDPM vs Apple Silicon/MPS) with no shared bug or change." - }, { "left": "pull_request:3850", "right": "pull_request:5853", "accept": false, - "reason": "Test-only PRs, but they address different problems in different areas (audio_diffusion test setup vs doctest infrastructure)." + "reason": "Audio Diffusion test fix vs doctest workflow/CI additions; unrelated areas and different underlying problems." }, { "left": "pull_request:388", "right": "pull_request:391", "accept": false, - "reason": "Separate pipeline docs for different algorithms (latent diffusion vs PNDM); no common underlying fix." + "reason": "Both are docs PRs, but they document different pipelines (latent diffusion vs PNDM), not the same change." }, { "left": "pull_request:388", "right": "pull_request:401", "accept": false, - "reason": "Both are docs/pipeline additions, but for different pipelines and releases; not the same change." + "reason": "Different pipeline docs and code updates: latent diffusion unconditional vs Karras-VE, no shared bug or change." }, { "left": "pull_request:3928", "right": "pull_request:4396", "accept": false, - "reason": "Different bug fixes: ControlNet timestep gating vs prompt encoding when text_encoder may be absent." + "reason": "Both touch SD/ControlNet code, but one fixes timestep gating and the other fixes prompt embedding dtype; distinct bugs." }, { "left": "pull_request:3931", "right": "pull_request:4748", "accept": false, - "reason": "Different codepaths and purposes: checkpoint loading logic vs packaging/setup instructions." + "reason": "Checkpoint loading logic vs packaging/release instructions; no overlap in underlying issue." }, { "left": "pull_request:396", "right": "pull_request:397", "accept": false, - "reason": "Different docs additions: DDIM pipeline docs vs quicktour/general getting-started docs." + "reason": "DDIM docs vs Quicktour docs; both documentation-only and not the same artifact or problem." }, { "left": "pull_request:396", "right": "pull_request:415", "accept": false, - "reason": "Unrelated docs work: DDIM pipeline docs vs training overview/docs." + "reason": "Different docs areas: DDIM pipeline docs versus training docs." }, { "left": "pull_request:396", "right": "pull_request:418", "accept": false, - "reason": "DDIM pipeline docs and DiffusionPipeline docs are distinct documentation changes, not one bug or feature." + "reason": "DDIM pipeline docs vs DiffusionPipeline docs; separate documentation changes." }, { "left": "pull_request:396", "right": "pull_request:508", "accept": false, - "reason": "Different scope entirely: DDIM docs vs scheduler/library-wide flax scheduler work." + "reason": "Docs PR versus scheduler implementation/Flax scheduler updates; not the same underlying change." }, { "left": "pull_request:397", "right": "pull_request:408", "accept": false, - "reason": "Quicktour rewrite vs typos cleanup across docs are separate documentation tasks." + "reason": "Quicktour rewrite vs typo cleanup across docs; related only at a very broad docs level." }, { "left": "pull_request:397", "right": "pull_request:416", "accept": false, - "reason": "Quicktour docs vs model API docs; different content and target surface." + "reason": "Quicktour docs vs models API docs; different content and purpose." }, { "left": "pull_request:397", "right": "pull_request:921", "accept": false, - "reason": "Different docs updates and broad release notes; not the same underlying change." + "reason": "Both are docs, but one is quicktour/training content and the other is v1.5 release docs updates; not duplicate work." }, { "left": "pull_request:3971", "right": "pull_request:4010", "accept": false, - "reason": "Workflow/docs maintenance vs SDXL sequential CPU offload bugfix; unrelated code paths." + "reason": "SDXL docs cleanup versus sequential offload code fix; different artifacts and different issues." }, { "left": "pull_request:3971", "right": "pull_request:4196", "accept": false, - "reason": "Different changes: docs build workflow/SDXL docs plumbing vs SDXL parameter documentation improvements." + "reason": "Both are SDXL docs, but they cover different sections and edits; not the same concrete change." }, { "left": "pull_request:3971", "right": "pull_request:4984", "accept": false, - "reason": "Workflow/docs setup vs PR template typo fix; no shared underlying bug." + "reason": "Docs workflow/content fix vs PR template edit; unrelated." }, { "left": "pull_request:3972", "right": "pull_request:4850", "accept": false, - "reason": "Build-doc workflow refactor vs deprecation-warning API cleanup; unrelated changes." + "reason": "Documentation workflow refactor vs deprecation-warning code change; unrelated." }, { "left": "pull_request:3979", "right": "pull_request:6410", "accept": false, - "reason": "Different bugs in different systems: SDXL attention/type handling vs RL value-guided sampling module registration." + "reason": "SDXL pipeline bug fix vs RL module registration refactor; no shared code path." }, { "left": "pull_request:3985", "right": "pull_request:4807", "accept": false, - "reason": "Both are docs, but one is SDXL English docs and the other is Chinese docs; separate localization/documentation work." + "reason": "Both are docs, but one is SDXL docs and the other is Chinese stable diffusion docs; different documentation targets." }, { "left": "pull_request:3989", "right": "pull_request:5097", "accept": false, - "reason": "Both touch tests, but one fixes slow-test offload behavior and the other adjusts float16 inference tolerances." + "reason": "Sequential offload fix vs float16 test adjustments; test changes do not show the same bug being fixed." }, { "left": "pull_request:3989", "right": "pull_request:6075", "accept": false, - "reason": "Different test maintenance issues: SDXL/text-to-video slow tests vs backend cache clearing." + "reason": "SDXL CPU offload fix vs backend cache-clearing test cleanup; unrelated problems." }, { "left": "pull_request:3989", "right": "pull_request:6198", "accept": false, - "reason": "Test-related, but benchmark reporting and slow-test fixes are unrelated." + "reason": "Pipeline offload bug fix vs benchmark reporting helper; different code paths and intent." }, { "left": "pull_request:4", "right": "pull_request:75", "accept": false, - "reason": "Both involve Glide-era code, but one converts weights and the other fixes attention module logic; different underlying changes." + "reason": "Glide weight conversion utility vs attention-layer constructor fix; same broad model family, but not the same change." }, { "left": "pull_request:4058", "right": "pull_request:4237", "accept": false, - "reason": "SDXL docs cleanup vs SDXL prompt-encoding bugfix; documentation and runtime behavior are distinct." + "reason": "SDXL docs correction vs SDXL prompt-encoding bug fix; different type of change and no shared underlying defect." }, { "left": "pull_request:4058", "right": "pull_request:4238", "accept": false, - "reason": "SDXL docs changes vs ControlNet SDXL inference fix; not the same code-path problem." + "reason": "Docs-only PR vs SDXL ControlNet inference fix; not the same artifact or issue." }, { "left": "pull_request:4058", "right": "pull_request:5602", "accept": false, - "reason": "Both are SDXL docs-related, but one is a docs correction and the other is whitespace/overview-table cleanup." + "reason": "Both touch docs, but one is SDXL prose/image formatting and the other is stable-diffusion overview layout; not duplicate." }, { "left": "pull_request:408", "right": "pull_request:415", "accept": false, - "reason": "General docs typo fixes vs training docs overhaul; different scope and intent." + "reason": "General doc typo cleanup versus training docs rewrite; too broad to be the same change." }, { "left": "pull_request:408", "right": "pull_request:418", "accept": false, - "reason": "Typos in multiple docs vs DiffusionPipeline API docs rewrite; not the same change." + "reason": "Typo fixes across docs versus DiffusionPipeline API docs rewrite; not the same underlying work." }, { "left": "pull_request:4082", "right": "pull_request:6322", "accept": false, - "reason": "Requirement pinning in SDXL example vs training script scheduler/attention changes; unrelated." + "reason": "Requirement pin fix vs training-script scheduler change; unrelated." }, { "left": "pull_request:4082", "right": "pull_request:6337", "accept": false, - "reason": "Dependency/version fix vs community example latents retrieval change; different problems." + "reason": "SDXL requirement adjustment vs retrieve_latents usage fix; different issues." }, { "left": "pull_request:41", "right": "pull_request:498", "accept": false, - "reason": "Comment cleanup in model code vs disabling attention slicing bugfix in pipelines; unrelated." + "reason": "Comment cleanup in model code versus attention-slicing API fix; unrelated." }, { "left": "pull_request:41", "right": "pull_request:693", "accept": false, - "reason": "Comment cleanup vs BibTeX citation formatting; no shared underlying issue." + "reason": "Comment formatting cleanup versus BibTeX citation correction; unrelated." }, { "left": "pull_request:41", "right": "pull_request:75", "accept": false, - "reason": "Shared old Glide context, but one is comment cleanup and the other changes attention implementation; not the same bug or mergeable change." + "reason": "Both involve Glide/attention code, but one changes comments and the other changes runtime constructor logic; not the same fix." + }, + { + "left": "pull_request:4106", + "right": "pull_request:4276", + "accept": false, + "reason": "New Tortoise TTS feature work vs import test coverage; different goals and code paths." + }, + { + "left": "pull_request:4106", + "right": "pull_request:6150", + "accept": false, + "reason": "Tortoise TTS implementation vs IP-Adapter image projection conversion; unrelated subsystems." + }, + { + "left": "pull_request:4112", + "right": "pull_request:5456", + "accept": false, + "reason": "DreamBooth LoRA rank enhancement vs RealFill feature addition; separate projects and changes." } ] }, @@ -174419,16 +174559,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 229, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:4106", - "pull_request:4112", "pull_request:4113", "pull_request:415", "pull_request:416", @@ -174454,16 +174592,19 @@ "pull_request:4437", "pull_request:4495", "pull_request:4563", + "pull_request:4581", + "pull_request:4596", + "pull_request:46", "pull_request:4635", "pull_request:4690", "pull_request:5456", + "pull_request:5552", "pull_request:5707", "pull_request:5768", "pull_request:5937", "pull_request:6002", "pull_request:6053", "pull_request:6078", - "pull_request:6150", "pull_request:6326", "pull_request:6337", "pull_request:6360", @@ -174476,9 +174617,6 @@ "pull_request:96" ], "soft_pairs": [ - "pull_request:4106|pull_request:4276", - "pull_request:4106|pull_request:6150", - "pull_request:4112|pull_request:5456", "pull_request:4113|pull_request:6002", "pull_request:415|pull_request:921", "pull_request:416|pull_request:418", @@ -174506,222 +174644,225 @@ "pull_request:4495|pull_request:6360", "pull_request:4495|pull_request:6800", "pull_request:4563|pull_request:5707", - "pull_request:4563|pull_request:6053" + "pull_request:4563|pull_request:6053", + "pull_request:4581|pull_request:5552", + "pull_request:4596|pull_request:5456", + "pull_request:46|pull_request:96" ], - "prepared_review_unit_hash": "3b93862b139e4cf1589ccb47bbdabe1b08ba2f7a757e7c214b9b578f919ec6bf", + "prepared_review_unit_hash": "5b7be3e4d407a036485d5742252cb2ee61b22fde953bd371cc15985df8dca220", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 102983, - "estimated_input_tokens": 25746, - "estimated_eval_tokens": 51748 + "serialized_chars": 104577, + "estimated_input_tokens": 26145, + "estimated_eval_tokens": 52546 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a mixed bag of unrelated docs updates, example scripts, and scheduler/pipeline fixes. None of the soft-edge pairs look like the same underlying bug/change closely enough to merge, so all soft edges are rejected and there is no clear canonical issue or PR for the cluster.", - "confidence": 0.94, + "summary": "This cluster is mostly a grab bag of unrelated docs, examples, tests, and small bugfix PRs. The only near-match is the DPMSolverMultistepScheduler pair, but even that looks like two separate scheduler changes rather than one duplicate fix.", + "confidence": 0.36, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 4187 is the most concrete bug-fix style change in the only plausible overlap area (DPMSolverMultistepScheduler), while the rest are mostly unrelated docs/tests or separate feature work.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 4187 is the best representative because it targets a specific code-path issue and is closer to a canonical bugfix than the surrounding documentation-heavy PRs.", "soft_edge_verdicts": [ - { - "left": "pull_request:4106", - "right": "pull_request:4276", - "accept": false, - "reason": "Both are import-related test/helper changes, but one adds Tortoise TTS exports while the other adds a pipeline import test; different problems and different code paths." - }, - { - "left": "pull_request:4106", - "right": "pull_request:6150", - "accept": false, - "reason": "Add-Tortoise-TTS vs IP-Adapter conversion logic are unrelated features/fixes in different subsystems." - }, - { - "left": "pull_request:4112", - "right": "pull_request:5456", - "accept": false, - "reason": "Dreambooth LoRA rank support and RealFill are different training/example changes with no shared bug." - }, { "left": "pull_request:4113", "right": "pull_request:6002", "accept": false, - "reason": "One fixes SDXL inpaint dtype handling; the other fixes Stable Video Diffusion guidance scale behavior. Different pipelines and failures." + "reason": "Different pipelines and different problems: SDXL inpaint dtype handling vs Stable Video Diffusion guidance-scale logic." }, { "left": "pull_request:415", "right": "pull_request:921", "accept": false, - "reason": "Both are documentation PRs, but they cover different docs areas and content; not the same change." + "reason": "Both are docs, but they cover different sections and scope; not the same underlying change." }, { "left": "pull_request:416", "right": "pull_request:418", "accept": false, - "reason": "Models docs/API cleanup versus DiffusionPipeline docs/API cleanup are related only at a broad documentation level, not the same artifact or bug." + "reason": "Separate documentation updates for different API areas; no shared bug or change." }, { "left": "pull_request:418", "right": "pull_request:921", "accept": false, - "reason": "Both touch docs, but one is API reference cleanup and the other is a broad release docs update; too different to be duplicates." + "reason": "Broad docs work in both cases, but the targets are different and not a single mergeable fix." }, { "left": "pull_request:4187", "right": "pull_request:4690", "accept": false, - "reason": "Both are about DPMSolverMultistepScheduler internals, but one fixes duplicate timesteps/indexing while the other refactors the scheduler around sigmas; different concrete issues." + "reason": "Same scheduler file, but one is about duplicate timestep/index handling and the other is a sigma-based refactor; related, not the same concrete fix." }, { "left": "pull_request:4188", "right": "pull_request:4370", "accept": false, - "reason": "ControlNet SDXL training cache fingerprinting and SDXL dummy import naming are unrelated fixes." + "reason": "ControlNet training cache fingerprinting vs SDXL dummy import cleanup are unrelated changes." }, { "left": "pull_request:4196", "right": "pull_request:4635", "accept": false, - "reason": "SDXL docs improvement versus MultiControlNet docs addition; same broad area, but not the same underlying change." + "reason": "SDXL docs cleanup vs MultiControlNet documentation; different documentation topics." }, { "left": "pull_request:42", "right": "pull_request:96", "accept": false, - "reason": "Early test updates for different model additions; they share a test file but not the same bug or change." + "reason": "Both touch early latent-diffusion modeling code, but they are different pieces of model/test work rather than one duplicate bug." }, { "left": "pull_request:421", "right": "pull_request:422", "accept": false, - "reason": "Scheduler docs and outputs docs are separate documentation updates with no shared underlying issue." + "reason": "Scheduler docs and BaseOutput docs are separate documentation changes." }, { "left": "pull_request:421", "right": "pull_request:91", "accept": false, - "reason": "Docs fix versus scheduler API/model forward renames and compatibility work; not the same change." + "reason": "One is a docs fix, the other is a broad scheduler/model API implementation change." }, { "left": "pull_request:4211", "right": "pull_request:6002", "accept": false, - "reason": "SDXL inpainting docstring correction and Stable Video Diffusion guidance logic are unrelated." + "reason": "Different SDXL docs vs a Stable Video Diffusion runtime fix; no common underlying issue." }, { "left": "pull_request:422", "right": "pull_request:423", "accept": false, - "reason": "Outputs docs and syntax improvement across many pipeline docs are different documentation tasks." + "reason": "Outputs docs versus a wider syntax/docs cleanup across many pipeline pages; not the same concrete change." }, { "left": "pull_request:4239", "right": "pull_request:6326", "accept": false, - "reason": "Dreambooth text_encoder fix and TensorRT guidance_scale fix are unrelated example-specific bugs." + "reason": "DreamBooth text-encoder naming fix and TensorRT guidance-scale assignment fix are unrelated." }, { "left": "pull_request:4275", "right": "pull_request:6337", "accept": false, - "reason": "AutoPipeline docs correction and TensorRT latent retrieval fix have no shared code path." + "reason": "Autopipeline docs fix vs a TensorRT latent-retrieval code fix; different concerns." }, { "left": "pull_request:4275", "right": "pull_request:6591", "accept": false, - "reason": "Docs for AutoPipeline versus broad typo fixes across community/controlnet examples are not the same change." + "reason": "Autopipeline docs and the text typo sweep across many examples are separate doc/code cleanup efforts." }, { "left": "pull_request:4276", "right": "pull_request:4417", "accept": false, - "reason": "Pipeline import test and Zero123 support are different objectives and code paths." + "reason": "Pipeline import test coverage vs Zero123 feature support are unrelated." }, { "left": "pull_request:4276", "right": "pull_request:6437", "accept": false, - "reason": "Pipeline import test versus CI exclusion change for PEFT LoRA tests are unrelated." + "reason": "Pipeline import test vs CI exclusion change; not the same bug/change." }, { "left": "pull_request:43", "right": "pull_request:7", "accept": false, - "reason": "Model card template/workflow updates versus adding VQ-VAE are unrelated foundational changes." + "reason": "Old model-card/template work and VQ-VAE addition are different features." }, { "left": "pull_request:4320", "right": "pull_request:6337", "accept": false, - "reason": "A documentation typo fix and a runtime latent retrieval fix do not address the same bug." + "reason": "ONNX docs typo fix vs latent retrieval code fix are unrelated." }, { "left": "pull_request:4320", "right": "pull_request:6591", "accept": false, - "reason": "Docs typo fix versus broad spelling fixes across many examples are only text-level similarity, not the same change." + "reason": "Different doc cleanup efforts with no shared code-path." }, { "left": "pull_request:4339", "right": "pull_request:5937", "accept": false, - "reason": "A guidance_scale docstring default correction and Kandinsky3 docs are unrelated documentation updates." + "reason": "A guidance-scale default typo in one example is unrelated to Kandinsky 3 docs." }, { "left": "pull_request:4347", "right": "pull_request:6053", "accept": false, - "reason": "Step-index scheduler fix versus logging an extra step_loss metric are unrelated and touch different code paths." + "reason": "Scheduler step-index work versus training loss logging are unrelated changes." }, { "left": "pull_request:4395", "right": "pull_request:6078", "accept": false, - "reason": "Dreambooth class prompt pre-encoding fix and Euler discrete sigma cleanup are unrelated." + "reason": "DreamBooth class-prompt encoding fix and Euler sigma fix are separate code paths." }, { "left": "pull_request:4437", "right": "pull_request:5768", "accept": false, - "reason": "Consistency-model sampling bug fixes versus consistency decoder doc cleanup are not the same underlying issue." + "reason": "Consistency-model sampling bugs vs consistency decoder documentation; not the same change." }, { "left": "pull_request:4495", "right": "pull_request:6360", "accept": false, - "reason": "A new SDXL reference community pipeline and adding padding_mask_crop to inpaint pipelines are different features/fixes." + "reason": "A community SDXL reference example and inpaint padding-mask support are different features." }, { "left": "pull_request:4495", "right": "pull_request:6800", "accept": false, - "reason": "New community SDXL reference pipeline versus adding torchvision availability utility are unrelated." + "reason": "Example pipeline work versus import-utility availability changes; unrelated." }, { "left": "pull_request:4563", "right": "pull_request:5707", "accept": false, - "reason": "Prompt2Prompt community pipeline and LCM test additions are different work in different areas." + "reason": "Prompt2Prompt feature addition and LCM test additions do not fix the same issue." }, { "left": "pull_request:4563", "right": "pull_request:6053", "accept": false, - "reason": "Prompt-to-prompt pipeline addition and training loss logging are unrelated changes." + "reason": "Prompt2Prompt pipeline work versus training loss logging are unrelated." + }, + { + "left": "pull_request:4581", + "right": "pull_request:5552", + "accept": false, + "reason": "UniPC sigma handling fix and FreeU disable-method fix are unrelated code paths." + }, + { + "left": "pull_request:4596", + "right": "pull_request:5456", + "accept": false, + "reason": "SDXL checkpoint conversion script and RealFill training docs/code are different features." + }, + { + "left": "pull_request:46", + "right": "pull_request:96", + "accept": false, + "reason": "ResNet refactor and latent-diffusion finalization are separate architectural changes." } ] }, @@ -174737,17 +174878,14 @@ "component_index": 1, "component_count": 266, "review_unit_index": 230, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:4581", - "pull_request:4596", - "pull_request:46", "pull_request:4632", "pull_request:4635", "pull_request:4655", @@ -174769,12 +174907,14 @@ "pull_request:4934", "pull_request:4955", "pull_request:4973", + "pull_request:498", + "pull_request:4984", + "pull_request:5", + "pull_request:5011", "pull_request:5098", "pull_request:5307", - "pull_request:5456", "pull_request:5499", "pull_request:5506", - "pull_request:5552", "pull_request:5682", "pull_request:5698", "pull_request:5699", @@ -174790,13 +174930,11 @@ "pull_request:6490", "pull_request:6591", "pull_request:6800", + "pull_request:7", "pull_request:7153", - "pull_request:96" + "pull_request:75" ], "soft_pairs": [ - "pull_request:4581|pull_request:5552", - "pull_request:4596|pull_request:5456", - "pull_request:46|pull_request:96", "pull_request:4632|pull_request:5922", "pull_request:4635|pull_request:5982", "pull_request:4655|pull_request:5506", @@ -174820,198 +174958,208 @@ "pull_request:4881|pull_request:4934", "pull_request:4886|pull_request:7153", "pull_request:4955|pull_request:5698", - "pull_request:4973|pull_request:6800" + "pull_request:4973|pull_request:6800", + "pull_request:498|pull_request:75", + "pull_request:4984|pull_request:5698", + "pull_request:5|pull_request:7", + "pull_request:5011|pull_request:5098" ], - "prepared_review_unit_hash": "d29b1a677e65338e8cd4ce8f9cda636079d756e66204c4495b5276349091aeee", + "prepared_review_unit_hash": "c7e6407593ac1aa585f33053100983e559c04b27cb36da4562e3ac4d57f33b3a", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 27, - "serialized_chars": 98213, - "estimated_input_tokens": 24554, - "estimated_eval_tokens": 49364 + "soft_pair_count": 28, + "serialized_chars": 97249, + "estimated_input_tokens": 24313, + "estimated_eval_tokens": 48882 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated PRs; the soft edges are mostly superficial title/subsystem similarity rather than the same underlying bug or change. No pair should be merged.", + "summary": "This cluster is a grab-bag of unrelated PRs across docs, examples, workflows, and distinct pipeline/model bugs. The soft similarities are mostly lexical (e.g. SDXL, ControlNet, LoRA) rather than the same concrete fix, so none of the candidate edges should be merged.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR stands out: the items span unrelated docs edits, training scripts, pipeline fixes, CI changes, and utility tweaks, so any canonical choice would be arbitrary.", + "canonical_pr_reason": "No single PR represents a common underlying change; the set is heterogeneous and spans unrelated fixes, docs, and tests.", "best_issue_reason": null, - "best_pr_reason": "No global best PR is a good representative because the cluster does not correspond to one duplicate family; the strongest choice would still be misleading.", + "best_pr_reason": "No PR is a good global anchor because there is no shared bug/change theme across the cluster.", "soft_edge_verdicts": [ - { - "left": "pull_request:4581", - "right": "pull_request:5552", - "accept": false, - "reason": "Different bugs in different code paths: UniPC Karras sigma conversion vs FreeU disable logic." - }, - { - "left": "pull_request:4596", - "right": "pull_request:5456", - "accept": false, - "reason": "An SDXL conversion script is unrelated to adding the RealFill training/inference example." - }, - { - "left": "pull_request:46", - "right": "pull_request:96", - "accept": false, - "reason": "Early ResNet/UNet refactor work vs later LDM finalization; not the same concrete change." - }, { "left": "pull_request:4632", "right": "pull_request:5922", "accept": false, - "reason": "Checkpointing/casting fixes in an SDXL LoRA trainer are unrelated to MultiAdapter inpainting handling." + "reason": "Different problems: SDXL LoRA checkpointing/casting vs MultiAdapter inpainting handling." }, { "left": "pull_request:4635", "right": "pull_request:5982", "accept": false, - "reason": "Both are docs-related, but one is MultiControlNet guidance and the other is SDXL Turbo docs; different changes." + "reason": "Unrelated docs additions: MultiControlNet guide vs SDXL Turbo docs." }, { "left": "pull_request:4655", "right": "pull_request:5506", "accept": false, - "reason": "Loading dataset with data_dir is unrelated to the PEFT/xformers attention processor fix." + "reason": "Different code paths: load_dataset data_dir plumbing vs PEFT/xformers attention scaling." }, { "left": "pull_request:4666", "right": "pull_request:5853", "accept": false, - "reason": "Enabling LoRA for SDXL ControlNet is unrelated to adding doctest workflows." + "reason": "SDXL ControlNet LoRA support vs adding doctest CI workflows; not the same change." }, { "left": "pull_request:4700", "right": "pull_request:5499", "accept": false, - "reason": "Shap-E guide documentation and broad docs link-fixing are not the same change." + "reason": "Different documentation topics: Shap-E guide vs general link fixes across many docs." }, { "left": "pull_request:4706", "right": "pull_request:4862", "accept": false, - "reason": "LoRA alpha default handling and MultiControlNet img2img size handling affect different code paths." + "reason": "Different bugs: missing LoRA alpha defaults vs MultiControlNet img2img size handling." }, { "left": "pull_request:4721", "right": "pull_request:6591", "accept": false, - "reason": "Korean docs cleanup is unrelated to a broad typo sweep across examples and pipelines." + "reason": "A Korean docs typo fix is unrelated to a broad typo cleanup across examples and pipelines." }, { "left": "pull_request:4741", "right": "pull_request:6049", "accept": false, - "reason": "AudioLDM2 dummy import fixes are unrelated to the DemoFusion pipeline fix." + "reason": "AudioLDM2 dummy import fix vs DemoFusion pipeline/README changes; unrelated." }, { "left": "pull_request:4746", "right": "pull_request:5682", "accept": false, - "reason": "A WIP SDXL inpainting PR is unrelated to AnimateDiff model/pipeline additions and checks." + "reason": "WIP SDXL inpainting work vs AnimateDiff checks; different features entirely." }, { "left": "pull_request:4748", "right": "pull_request:5307", "accept": false, - "reason": "Setup.py packaging documentation and creating shell.py are unrelated." + "reason": "Packaging/setup instructions vs an unrelated new shell.py script." }, { "left": "pull_request:4761", "right": "pull_request:5098", "accept": false, - "reason": "Paint-by-example proj_size fix and Kandinsky precision test adjustments are unrelated." + "reason": "Paint-by-example proj_size fix vs Kandinsky precision test updates; different code paths." }, { "left": "pull_request:4793", "right": "pull_request:5506", "accept": false, - "reason": "AutoPipeline kwargs forwarding is unrelated to the attention processor LoRA/xformers bug fix." + "reason": "Auto-pipeline config kwargs vs attention processor LoRA/xformers behavior." }, { "left": "pull_request:4793", "right": "pull_request:5754", "accept": false, - "reason": "AutoPipeline load_config kwargs and dataset call styling changes do not fix the same bug." + "reason": "Auto-pipeline loading fix vs formatting/data_dir changes in training scripts." }, { "left": "pull_request:4807", "right": "pull_request:5937", "accept": false, - "reason": "Chinese Stable Diffusion docs and Kandinsky3 docs are separate documentation additions." + "reason": "Chinese docs additions vs Kandinsky3 docs; unrelated documentation work." }, { "left": "pull_request:4807", "right": "pull_request:6017", "accept": false, - "reason": "A Chinese docs page and a doc-builder test PR are unrelated." + "reason": "Chinese docs vs a doc-builder test WIP; not the same underlying change." }, { "left": "pull_request:4807", "right": "pull_request:6466", "accept": false, - "reason": "Chinese Stable Diffusion docs and SVD doc edits are different pages and different updates." + "reason": "Chinese docs vs SVD doc cleanup; different pages and purposes." }, { "left": "pull_request:4822", "right": "pull_request:5699", "accept": false, - "reason": "A LoRA warning typo fix is unrelated to a styling cleanup in Stable UnCLIP." + "reason": "LoRA warning text typo vs stable_unclip styling refactors; unrelated." }, { "left": "pull_request:4822", "right": "pull_request:6389", "accept": false, - "reason": "LoRA warning text and SVD noise_aug_strength typing address different issues." + "reason": "LoRA warning text typo vs SVD parameter type fix; different issue." }, { "left": "pull_request:4822", "right": "pull_request:6591", "accept": false, - "reason": "A single warning-string typo is not the same as a broad typo cleanup across many files." + "reason": "Single warning-string typo vs broad typo cleanup across many files." }, { "left": "pull_request:4847", "right": "pull_request:6490", "accept": false, - "reason": "Prompt Free Diffusion community example and RAVE video editing are unrelated community additions." + "reason": "Prompt Free Diffusion community example vs RAVE video editing community example; different additions." }, { "left": "pull_request:4881", "right": "pull_request:4934", "accept": false, - "reason": "SDXL pix2pix vae_precision support and T2I-Adapter training script/docs are different features." + "reason": "SDXL pix2pix VAE precision option vs T2I-Adapter training script; not the same feature." }, { "left": "pull_request:4886", "right": "pull_request:7153", "accept": false, - "reason": "A pipeline safety-checker bug fix is unrelated to CI path-filter changes." + "reason": "Upscale safety checker bug vs documentation workflow path filtering." }, { "left": "pull_request:4955", "right": "pull_request:5698", "accept": false, - "reason": "Parser type cleanup in pix2pix is unrelated to the LCM img2img fix." + "reason": "Argument type fix in pix2pix script vs LCM img2img fix; unrelated." }, { "left": "pull_request:4973", "right": "pull_request:6800", "accept": false, - "reason": "Adding a README badge is unrelated to exposing is_torchvision_available." + "reason": "README badge update vs adding is_torchvision_available utility; different concerns." + }, + { + "left": "pull_request:498", + "right": "pull_request:75", + "accept": false, + "reason": "Attention-slicing typo fix vs a much older Glide attention implementation change." + }, + { + "left": "pull_request:4984", + "right": "pull_request:5698", + "accept": false, + "reason": "PR template editor names vs LCM img2img bug fix; no shared code path." + }, + { + "left": "pull_request:5", + "right": "pull_request:7", + "accept": false, + "reason": "Foundational latent UNet addition vs VQVAE addition; separate model components, not duplicates." + }, + { + "left": "pull_request:5011", + "right": "pull_request:5098", + "accept": false, + "reason": "Flax/PT class-name handling vs Kandinsky precision tests; unrelated changes." } ] }, @@ -175027,25 +175175,20 @@ "component_index": 1, "component_count": 266, "review_unit_index": 231, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:498", - "pull_request:4984", - "pull_request:5", - "pull_request:5011", "pull_request:5021", "pull_request:5030", "pull_request:5038", "pull_request:5056", "pull_request:5058", "pull_request:5063", - "pull_request:5098", "pull_request:51", "pull_request:5111", "pull_request:5122", @@ -175058,9 +175201,12 @@ "pull_request:5289", "pull_request:5304", "pull_request:5307", + "pull_request:5312", "pull_request:5331", + "pull_request:5391", + "pull_request:5424", + "pull_request:5454", "pull_request:5506", - "pull_request:5698", "pull_request:570", "pull_request:5729", "pull_request:5780", @@ -175069,26 +175215,24 @@ "pull_request:5834", "pull_request:5910", "pull_request:6085", + "pull_request:6236", "pull_request:6364", + "pull_request:6658", "pull_request:6754", "pull_request:6765", "pull_request:6766", "pull_request:6800", "pull_request:6861", "pull_request:6875", - "pull_request:7", + "pull_request:7062", "pull_request:7161", + "pull_request:7166", "pull_request:7272", "pull_request:7307", "pull_request:7328", - "pull_request:7339", - "pull_request:75" + "pull_request:7339" ], "soft_pairs": [ - "pull_request:498|pull_request:75", - "pull_request:4984|pull_request:5698", - "pull_request:5|pull_request:7", - "pull_request:5011|pull_request:5098", "pull_request:5021|pull_request:6364", "pull_request:5030|pull_request:7272", "pull_request:5038|pull_request:5506", @@ -175113,210 +175257,227 @@ "pull_request:5307|pull_request:6754", "pull_request:5307|pull_request:6765", "pull_request:5307|pull_request:6766", - "pull_request:5307|pull_request:7307" + "pull_request:5307|pull_request:7307", + "pull_request:5312|pull_request:6236", + "pull_request:5331|pull_request:7166", + "pull_request:5391|pull_request:6658", + "pull_request:5424|pull_request:7062", + "pull_request:5454|pull_request:6861" ], - "prepared_review_unit_hash": "6a1a6ace8b718ccb04054c4d0bc85a90a15a8a7230f9f315b2df4aa5f21080c8", + "prepared_review_unit_hash": "c613bef807317c672b6b4907f8489843eda73907c1cf0f09b57600897207d53b", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 29, - "serialized_chars": 97397, - "estimated_input_tokens": 24350, - "estimated_eval_tokens": 48956 + "soft_pair_count": 30, + "serialized_chars": 97239, + "estimated_input_tokens": 24310, + "estimated_eval_tokens": 48876 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly a grab bag of unrelated PRs; the soft pairs only share broad subsystem keywords (LoRA, docs, schedulers, workflows) but not the same concrete bug or change.", - "confidence": 0.97, + "summary": "Most pairs are false positives spanning unrelated docs, workflow, test, and LoRA/IP-Adapter changes. The only clear duplicate-like pair is 6765/6766, which both patch HF cache path construction for the same underlying path-handling issue.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 5201 is the strongest representative if one must pick a canonical PR: it is a substantive, concrete fix in attention_processor affecting LoRA/xformers behavior across multiple model classes, unlike the many doc-only or cosmetic PRs.", + "canonical_pr_reason": "Among the accepted duplicate-like pair, 6765 is the more central change: it fixes the main cache-path logic in hub_utils, while 6766 is the companion constants adjustment.", "best_issue_reason": null, - "best_pr_reason": "PR 5201 is the best global representative because it addresses a real runtime bug in a shared code path, with clear implementation changes and broad applicability.", + "best_pr_reason": "6765 is the best representative PR for this cluster because it contains the core cache-path fix and matches the only strong duplicate pair.", "soft_edge_verdicts": [ - { - "left": "pull_request:498", - "right": "pull_request:75", - "accept": false, - "reason": "Both touch attention-related code, but one fixes stable-diffusion pipeline slicing and the other changes Glide/GradTTS attention module initialization; different bugs and different code paths." - }, - { - "left": "pull_request:4984", - "right": "pull_request:5698", - "accept": false, - "reason": "PR template typo fix vs LCM img2img bug fix; completely unrelated artifacts." - }, - { - "left": "pull_request:5", - "right": "pull_request:7", - "accept": false, - "reason": "Both are early model additions, but one adds UNet-LDM and the other adds VQVAE; separate components, not the same change." - }, - { - "left": "pull_request:5011", - "right": "pull_request:5098", - "accept": false, - "reason": "Flax-to-PT loading fix vs Kandinsky precision test adjustments; unrelated fixes." - }, { "left": "pull_request:5021", "right": "pull_request:6364", "accept": false, - "reason": "Both are docs, but one updates scheduler mapping while the other revises inpaint guidance; different content and purpose." + "reason": "Both are docs, but they cover different topics: scheduler mapping vs ControlNet inpaint overlay guidance." }, { "left": "pull_request:5030", "right": "pull_request:7272", "accept": false, - "reason": "Both involve loader code, but one fixes missing-key offload behavior and the other targets custom diffusion/PEFT loader compatibility; not the same bug." + "reason": "Different bugs in different loader paths; not the same concrete fix." }, { "left": "pull_request:5038", "right": "pull_request:5506", "accept": false, - "reason": "Both are LoRA-related, but one makes optional args explicit in transformer_2d while the other fixes xformers-enabled LoRA attention calls; different concrete issues." + "reason": "Both touch PEFT/LoRA code, but one makes arguments explicit and the other fixes xformers-enabled LoRA calls." }, { "left": "pull_request:5056", "right": "pull_request:5111", "accept": false, - "reason": "LoRA fusion docs vs scheduler feature support; unrelated." + "reason": "Unrelated changes: LoRA docs vs Karras sigma scheduler support." }, { "left": "pull_request:5058", "right": "pull_request:7161", "accept": false, - "reason": "PEFT/LoRA integration code vs documentation for torch.compile usage; not the same underlying change." + "reason": "Both are PEFT docs, but they describe different workflows and fixes." }, { "left": "pull_request:5063", "right": "pull_request:5190", "accept": false, - "reason": "One fixes a guidance_rescale docstring, the other wraps a different docstring line; cosmetic docs only, no shared bug/change." + "reason": "Both edit docstrings, but they target different functions and files with unrelated content." }, { "left": "pull_request:51", "right": "pull_request:570", "accept": false, - "reason": "Dependency pin fix vs BaseOutput dict initialization behavior; unrelated." + "reason": "Completely unrelated: packaging metadata vs BaseOutput initialization behavior." }, { "left": "pull_request:5111", "right": "pull_request:7328", "accept": false, - "reason": "Scheduler changes for karras sigmas vs a PyPI workflow dependency tweak; unrelated." + "reason": "Scheduler feature work vs release workflow dependency changes; different problems." }, { "left": "pull_request:5122", "right": "pull_request:5833", "accept": false, - "reason": "Workflow/python-version update vs a ControlNet single-file import fix; different scopes and bugs." + "reason": "Workflow/Python version changes are unrelated to the PixArt docs change shown here." }, { "left": "pull_request:5122", "right": "pull_request:5834", "accept": false, - "reason": "Workflow/python-version update vs ControlNet single-file import fix; unrelated." + "reason": "Workflow setup updates vs a single-file import fix; no shared code-path bug." }, { "left": "pull_request:5122", "right": "pull_request:7339", "accept": false, - "reason": "Workflow/python-version update vs release workflow dependency installation; not the same change." + "reason": "Different release/workflow dependency changes with distinct goals." }, { "left": "pull_request:5185", "right": "pull_request:5823", "accept": false, - "reason": "Both are LoRA/PEFT-adjacent, but one fixes active_adapter handling in CI and the other adds adapter_names support to fuse_lora; different functionality and bug." + "reason": "Both are PEFT-related, but one is a CI fix and the other adds adapter_names to fuse_lora." }, { "left": "pull_request:5201", "right": "pull_request:5331", "accept": false, - "reason": "Both relate to LoRA, but one fixes xformers/attention processor behavior while the other rewrites examples to use LoRALinear; not the same code-path problem." + "reason": "Shared LoRA vocabulary, but one fixes xformers processor removal and the other updates example training code." }, { "left": "pull_request:5201", "right": "pull_request:6085", "accept": false, - "reason": "LoRA/xformers attention fix vs scheduler add_noise bug; unrelated." + "reason": "Different code paths: LoRA/xformers processor handling vs scheduler add_noise logic." }, { "left": "pull_request:5250", "right": "pull_request:5910", "accept": false, - "reason": "DDPO docs entry vs IPAdapterFull implementation; unrelated." + "reason": "Docs for DDPO vs IPAdapterFull loader support; unrelated changes." }, { "left": "pull_request:5250", "right": "pull_request:6800", "accept": false, - "reason": "DDPO docs vs adding a utility import/export; unrelated." + "reason": "DDPO docs and adding torchvision availability are unrelated." }, { "left": "pull_request:5267", "right": "pull_request:6861", "accept": false, - "reason": "Docs README formatting cleanup vs SDXL pipeline cfg_end-related code changes; unrelated." + "reason": "README formatting cleanup vs a large SDXL pipeline update; no duplicate bug/change." }, { "left": "pull_request:5286", "right": "pull_request:6800", "accept": false, - "reason": "Japanese docs addition vs torchvision availability utility; unrelated." + "reason": "Japanese documentation addition vs import-utility availability helper; unrelated." }, { "left": "pull_request:5289", "right": "pull_request:6875", "accept": false, - "reason": "Different test tolerance updates for different pipelines; not the same bug." + "reason": "Both adjust test expectations, but they concern different pipelines and different failures." }, { "left": "pull_request:5304", "right": "pull_request:5729", "accept": false, - "reason": "DeepFloyd IF docstring typing cleanup vs broad example setup-version updates; unrelated." + "reason": "DeepFloyd IF docstrings vs example min-version updates; no same underlying change." }, { "left": "pull_request:5307", "right": "pull_request:5780", "accept": false, - "reason": "A stray shell.py addition vs a checkpoint_merger docstring cleanup; completely unrelated." + "reason": "Completely unrelated new file vs checkpoint_merger doc cleanup." }, { "left": "pull_request:5307", "right": "pull_request:6754", "accept": false, - "reason": "Shell script addition vs diffusion DPO loss correction; unrelated." + "reason": "Unrelated: shell sort demo file vs DPO training loss fix." }, { "left": "pull_request:5307", "right": "pull_request:6765", "accept": false, - "reason": "Shell.py is unrelated to hub cache-path handling." + "reason": "No relation: new shell.py demo vs cache path handling." }, { "left": "pull_request:5307", "right": "pull_request:6766", "accept": false, - "reason": "Shell.py is unrelated to constants.py cache-path handling." + "reason": "No relation: new shell.py demo vs constants cache path handling." }, { "left": "pull_request:5307", "right": "pull_request:7307", "accept": false, - "reason": "Shell.py addition vs TensorRT example update; unrelated." + "reason": "Unrelated demo script vs TensorRT/example environment update." + }, + { + "left": "pull_request:5312", + "right": "pull_request:6236", + "accept": false, + "reason": "Both are docs-ish, but one improves model docs/typehints and the other fixes a ControlNet example." + }, + { + "left": "pull_request:5331", + "right": "pull_request:7166", + "accept": false, + "reason": "Both mention LoRA/dreambooth, but they fix different issues in different scripts." + }, + { + "left": "pull_request:5391", + "right": "pull_request:6658", + "accept": false, + "reason": "General model docs cleanup vs introducing a memory-footprint utility; unrelated." + }, + { + "left": "pull_request:5424", + "right": "pull_request:7062", + "accept": false, + "reason": "Slow-test pipeline fetcher vs a community pipeline docstring fix." + }, + { + "left": "pull_request:5454", + "right": "pull_request:6861", + "accept": false, + "reason": "Scheduler docs link tweak vs SDXL pipeline feature update; unrelated." + }, + { + "left": "pull_request:6765", + "right": "pull_request:6766", + "accept": true, + "reason": "Both address the same cache-path construction issue around HF cache locations and are natural companions under issue 6585." } ] }, @@ -175332,7 +175493,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 232, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -175340,11 +175501,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:5312", - "pull_request:5331", - "pull_request:5391", - "pull_request:5424", - "pull_request:5454", "pull_request:5478", "pull_request:5499", "pull_request:5533", @@ -175357,32 +175513,37 @@ "pull_request:5707", "pull_request:5723", "pull_request:5727", - "pull_request:5741", "pull_request:5750", "pull_request:5780", "pull_request:5788", "pull_request:5818", "pull_request:5831", + "pull_request:5842", + "pull_request:5844", + "pull_request:5855", "pull_request:5858", "pull_request:5895", + "pull_request:5910", "pull_request:5911", + "pull_request:5920", "pull_request:5961", "pull_request:5974", + "pull_request:5982", "pull_request:6051", "pull_request:6065", "pull_request:6192", - "pull_request:6236", + "pull_request:6241", + "pull_request:6276", "pull_request:6326", "pull_request:6339", "pull_request:6498", + "pull_request:6643", "pull_request:6656", - "pull_request:6658", "pull_request:6800", "pull_request:6861", - "pull_request:6864", + "pull_request:6941", "pull_request:7062", "pull_request:7077", - "pull_request:7166", "pull_request:7307", "pull_request:7324", "pull_request:774", @@ -175390,11 +175551,6 @@ "pull_request:909" ], "soft_pairs": [ - "pull_request:5312|pull_request:6236", - "pull_request:5331|pull_request:7166", - "pull_request:5391|pull_request:6658", - "pull_request:5424|pull_request:7062", - "pull_request:5454|pull_request:6861", "pull_request:5478|pull_request:5831", "pull_request:5499|pull_request:6065", "pull_request:5533|pull_request:5707", @@ -175414,214 +175570,240 @@ "pull_request:5727|pull_request:5911", "pull_request:5727|pull_request:6192", "pull_request:5727|pull_request:6656", - "pull_request:5741|pull_request:6864", "pull_request:5750|pull_request:5895", "pull_request:5780|pull_request:6861", "pull_request:5780|pull_request:7307", - "pull_request:5788|pull_request:5974" + "pull_request:5788|pull_request:5974", + "pull_request:5842|pull_request:6241", + "pull_request:5842|pull_request:7062", + "pull_request:5844|pull_request:6643", + "pull_request:5855|pull_request:6861", + "pull_request:5895|pull_request:5982", + "pull_request:5910|pull_request:6941", + "pull_request:5911|pull_request:6192", + "pull_request:5911|pull_request:6656", + "pull_request:5920|pull_request:6276" ], - "prepared_review_unit_hash": "5af11928b23d10b1a085d967d4bf6937553666a70538b25124b0663bacdf80a9", + "prepared_review_unit_hash": "16178bd8f469dbfb1fe5d22d95a63315a65030ed0a0748525251bc56515d8538", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 29, - "serialized_chars": 101167, - "estimated_input_tokens": 25292, - "estimated_eval_tokens": 50840 + "soft_pair_count": 32, + "serialized_chars": 100495, + "estimated_input_tokens": 25124, + "estimated_eval_tokens": 50504 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: docs-only edits, test tweaks, minor bug fixes, and unrelated feature additions across different code paths. None of the soft pairs looks like the same underlying change, so they should stay separate.", - "confidence": 0.98, + "summary": "This cluster is heterogeneous: docs cleanups, test adjustments, new model/pipeline features, and unrelated loader/scheduler fixes. None of the soft edges look like true duplicates or mergeable into one PR, so I reject them all.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR stands out because these items do not share one concrete bug/change; they span unrelated docs, tests, and feature work, so any single PR would be arbitrary.", + "canonical_pr_reason": "No single PR cleanly represents a duplicate cluster here; the items span unrelated docs, tests, and feature work rather than one underlying change.", "best_issue_reason": null, - "best_pr_reason": "No PR is a good global representative for this cluster. The overlap is only superficial (same repo or broad subsystem), not a shared fixable code-path problem.", + "best_pr_reason": "PR 5911 is the most substantive, self-contained implementation change with a clear issue target, but even it does not unify the rest of the set into one duplicate family.", "soft_edge_verdicts": [ - { - "left": "pull_request:5312", - "right": "pull_request:6236", - "accept": false, - "reason": "Different docs/fix scopes: model typehints/docs vs a ControlNet inpainting doc correction." - }, - { - "left": "pull_request:5331", - "right": "pull_request:7166", - "accept": false, - "reason": "Both touch DreamBooth/LoRA, but one updates example LoRA code and the other fixes an SDXL loading bug; not the same change." - }, - { - "left": "pull_request:5391", - "right": "pull_request:6658", - "accept": false, - "reason": "Broadly model docs versus a new memory-footprint utility; unrelated changes." - }, - { - "left": "pull_request:5424", - "right": "pull_request:7062", - "accept": false, - "reason": "Slow-test pipeline fetcher fix and Imagic docstring cleanup are unrelated." - }, - { - "left": "pull_request:5454", - "right": "pull_request:6861", - "accept": false, - "reason": "One is a docs link tweak; the other adds cfg_end plumbing across SDXL/controlnet pipelines." - }, { "left": "pull_request:5478", "right": "pull_request:5831", "accept": false, - "reason": "Japanese docs addition and an English toctree title fix are separate documentation changes." + "reason": "Japanese docs versus a small title-only docs fix; different changes and no shared bug." }, { "left": "pull_request:5499", "right": "pull_request:6065", "accept": false, - "reason": "Link-fix docs PR versus adding SDXL Turbo docs and registry entries; not the same underlying change." + "reason": "Broad docs link cleanup versus SDXL Turbo docs addition; same area, different change." }, { "left": "pull_request:5533", "right": "pull_request:5707", "accept": false, - "reason": "Different test work: mixture-test speedups vs LCM test additions." + "reason": "Both are test-related, but one speeds up SDXL expert tests and the other adds LCM tests; unrelated code paths." }, { "left": "pull_request:5533", "right": "pull_request:5818", "accept": false, - "reason": "One speeds up SDXL tests; the other rewrites PR test-matrix setup." + "reason": "Speeding up expert tests and restructuring PR test workflow are different testing infrastructure changes." }, { "left": "pull_request:5568", "right": "pull_request:5961", "accept": false, - "reason": "Custom pipeline loading bugfix versus a PixArt docstring/default typo; unrelated." + "reason": "Custom pipeline loading fix versus a PixArt docstring typo; not the same bug or change." }, { "left": "pull_request:56", "right": "pull_request:774", "accept": false, - "reason": "Both are test adjustments, but they fix different tests and behaviors." + "reason": "Both touch tests, but they fix different assertions in different eras and files; not the same issue." }, { "left": "pull_request:56", "right": "pull_request:909", "accept": false, - "reason": "Old test tolerance changes and unrelated inpaint test/model-id updates do not share one bug." + "reason": "General test tolerance tweaks versus inpaint test model/path updates; unrelated fixes." }, { "left": "pull_request:5614", "right": "pull_request:5858", "accept": false, - "reason": "Reverting SDXL width/height order is unrelated to the basic_training dtype doc example." + "reason": "Both are revert-style docs/code edits, but they revert different changes in different places." }, { "left": "pull_request:5622", "right": "pull_request:6326", "accept": false, - "reason": "A broad, unrelated examples/optim PR versus a specific TRT guidance_scale bugfix." + "reason": "Example script optimization/config work versus a guidance_scale setter fix in a community TensorRT pipeline." }, { "left": "pull_request:5622", "right": "pull_request:7077", "accept": false, - "reason": "No shared code-path problem: unrelated examples/optim work and an IP-Adapter attention reshape fix." + "reason": "Different test/optimization work versus a specific attention processor fix; no shared underlying bug." }, { "left": "pull_request:5626", "right": "pull_request:7767", "accept": false, - "reason": "CPU offload test fix versus PixArt cleanup; different targets and intent." + "reason": "CPU offload test fix versus PixArt Sigma nits; unrelated test and pipeline changes." }, { "left": "pull_request:5682", "right": "pull_request:6339", "accept": false, - "reason": "Both mention AnimateDiff, but one is a WIP feature/checks PR and the other is only a docs correction." + "reason": "AnimateDiff checks implementation versus AnimateDiff docs example correction; one is feature/checks, the other documentation." }, { "left": "pull_request:5707", "right": "pull_request:7324", "accept": false, - "reason": "LCM add-tests work and Stable Cascade test updates are unrelated." + "reason": "LCM test additions versus Stable Cascade test updates; different models and failures." }, { "left": "pull_request:5723", "right": "pull_request:6051", "accept": false, - "reason": "LCM docs versus Flax LCM implementation work; different deliverables." + "reason": "LCM documentation and Flax LCM implementation are related broadly, but they are not the same concrete change." }, { "left": "pull_request:5723", "right": "pull_request:6498", "accept": false, - "reason": "LCM docs addition and inpaint docs update are separate documentation changes." + "reason": "LCM docs versus inpaint mask blur docs; only both are docs, not the same change." }, { "left": "pull_request:5723", "right": "pull_request:6800", "accept": false, - "reason": "LCM docs versus adding is_torchvision_available; unrelated." + "reason": "LCM docs versus utility availability export; completely different concerns." }, { "left": "pull_request:5727", "right": "pull_request:5911", "accept": false, - "reason": "LCM scripts and IPAdapterFull support are different features." + "reason": "LCM scripts versus IP-Adapter Full support; different features and code paths." }, { "left": "pull_request:5727", "right": "pull_request:6192", "accept": false, - "reason": "LCM distillation scripts versus unload_ip_adapter; no shared fix." + "reason": "LCM training scripts versus unloading IP-Adapter support; unrelated model/tooling changes." }, { "left": "pull_request:5727", "right": "pull_request:6656", "accept": false, - "reason": "LCM distillation scripts versus unload_textual_inversion; separate loader features." - }, - { - "left": "pull_request:5741", - "right": "pull_request:6864", - "accept": false, - "reason": "PEFT unload-speedup implementation and a docs note about torch.compile are not the same change." + "reason": "LCM scripts versus unload_textual_inversion; different loader features." }, { "left": "pull_request:5750", "right": "pull_request:5895", "accept": false, - "reason": "These are completely different: a placeholder/reference docs PR versus Stable Video Diffusion feature work." + "reason": "Reference-only docs introduction versus a full Stable Video Diffusion feature PR; not the same underlying change." }, { "left": "pull_request:5780", "right": "pull_request:6861", "accept": false, - "reason": "Checkpoint merger doc cleanup versus SDXL/controlnet pipeline feature plumbing are unrelated." + "reason": "Checkpoint merger doc cleanup versus a ControlNet/SDXL pipeline update; unrelated." }, { "left": "pull_request:5780", "right": "pull_request:7307", "accept": false, - "reason": "Checkpoint merger doc update does not match the tensorrt txt2img update; no shared bug or change." + "reason": "Checkpoint merger comment cleanup versus Tensorrt example update; different example codebases." }, { "left": "pull_request:5788", "right": "pull_request:5974", "accept": false, - "reason": "LCM scheduler error handling and PixArtAlpha LCM adaptation are different code paths and goals." + "reason": "LCM scheduler error handling versus PixArt-LCM pipeline adaptation; same model family only in name, but different concrete fixes." + }, + { + "left": "pull_request:5842", + "right": "pull_request:6241", + "accept": false, + "reason": "PixArt resolution bug fix versus a runtime dtype mismatch report with no diff shown; no evidence they are the same issue." + }, + { + "left": "pull_request:5842", + "right": "pull_request:7062", + "accept": false, + "reason": "PixArt resolution logic versus a community Imagic docstring fix; unrelated." + }, + { + "left": "pull_request:5844", + "right": "pull_request:6643", + "accept": false, + "reason": "Docs reference to Japanese SDXL versus UViT2D docs; different documentation additions." + }, + { + "left": "pull_request:5855", + "right": "pull_request:6861", + "accept": false, + "reason": "Community README typo fix versus SDXL ControlNet update; unrelated edits." + }, + { + "left": "pull_request:5895", + "right": "pull_request:5982", + "accept": false, + "reason": "Stable Video Diffusion feature PR versus SDXL Turbo docs; different products and deliverables." + }, + { + "left": "pull_request:5910", + "right": "pull_request:6941", + "accept": false, + "reason": "IPAdapterFull support and SDXL ControlNet IP-Adapter support share a subsystem, but they fix different pipelines and are not one mergeable change." + }, + { + "left": "pull_request:5911", + "right": "pull_request:6192", + "accept": false, + "reason": "Adding IPAdapterFull support versus unloading IP-Adapter support; opposite lifecycle operations, not the same bug." + }, + { + "left": "pull_request:5911", + "right": "pull_request:6656", + "accept": false, + "reason": "IP-Adapter support versus textual inversion unload support; different loader mechanisms." + }, + { + "left": "pull_request:5920", + "right": "pull_request:6276", + "accept": false, + "reason": "IPEX LoRA support versus IPAdapter FaceID community pipeline; different features and code paths." } ] }, @@ -175637,7 +175819,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 233, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -175645,25 +175827,14 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:5842", - "pull_request:5844", - "pull_request:5855", - "pull_request:5895", - "pull_request:5910", - "pull_request:5911", - "pull_request:5920", - "pull_request:5982", "pull_request:5983", "pull_request:6", "pull_request:6002", - "pull_request:6192", "pull_request:62", "pull_request:6236", - "pull_request:6241", "pull_request:6262", "pull_request:6263", "pull_request:6264", - "pull_request:6276", "pull_request:6296", "pull_request:6308", "pull_request:6322", @@ -175673,36 +175844,39 @@ "pull_request:6339", "pull_request:6344", "pull_request:6373", + "pull_request:6389", "pull_request:6410", + "pull_request:6433", + "pull_request:6453", "pull_request:6462", + "pull_request:6487", + "pull_request:6498", + "pull_request:6564", "pull_request:662", - "pull_request:6643", - "pull_request:6656", "pull_request:6759", - "pull_request:6861", + "pull_request:6800", "pull_request:6910", - "pull_request:6941", "pull_request:7", - "pull_request:7062", + "pull_request:7077", + "pull_request:7087", "pull_request:7510", "pull_request:7653", + "pull_request:7720", + "pull_request:78", + "pull_request:8268", "pull_request:8272", "pull_request:8419", "pull_request:8425", + "pull_request:8426", "pull_request:8481", + "pull_request:8483", + "pull_request:8491", + "pull_request:8654", + "pull_request:8656", "pull_request:8698", "pull_request:96" ], "soft_pairs": [ - "pull_request:5842|pull_request:6241", - "pull_request:5842|pull_request:7062", - "pull_request:5844|pull_request:6643", - "pull_request:5855|pull_request:6861", - "pull_request:5895|pull_request:5982", - "pull_request:5910|pull_request:6941", - "pull_request:5911|pull_request:6192", - "pull_request:5911|pull_request:6656", - "pull_request:5920|pull_request:6276", "pull_request:5983|pull_request:7510", "pull_request:6|pull_request:7", "pull_request:6|pull_request:96", @@ -175725,228 +175899,265 @@ "pull_request:6337|pull_request:8481", "pull_request:6339|pull_request:7510", "pull_request:6339|pull_request:8481", - "pull_request:6373|pull_request:8698" + "pull_request:6373|pull_request:8698", + "pull_request:6389|pull_request:7087", + "pull_request:6410|pull_request:7077", + "pull_request:6410|pull_request:8426", + "pull_request:6410|pull_request:8491", + "pull_request:6410|pull_request:8656", + "pull_request:6433|pull_request:6487", + "pull_request:6453|pull_request:6800", + "pull_request:6462|pull_request:7653", + "pull_request:6487|pull_request:8483", + "pull_request:6498|pull_request:8268", + "pull_request:6498|pull_request:8654", + "pull_request:6564|pull_request:7720", + "pull_request:662|pull_request:78" ], - "prepared_review_unit_hash": "a8c3f1e8fa4aa2f058a5cf38c2c3f50136aa2d551f0387d8f01cf4a14a4a31a7", + "prepared_review_unit_hash": "98e23c0f4cda5170230f61e898d3743245ed11250dcde2528ff98815c269a47e", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 32, - "serialized_chars": 97441, - "estimated_input_tokens": 24361, - "estimated_eval_tokens": 48978 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 36, + "serialized_chars": 102149, + "estimated_input_tokens": 25538, + "estimated_eval_tokens": 51332 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of unrelated PRs across docs, pipeline additions/refactors, loader features, and test fixes; the soft pairs do not look like duplicate bug fixes or mergeable single changes.", - "confidence": 0.96, + "summary": "This cluster is a grab bag of mostly unrelated PRs across docs, workflow cleanup, model imports, and feature work. There are a few same-subsystem follow-ups, but none look like true duplicates of the same concrete change.", + "confidence": 0.77, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a good canonical representative: the items span many unrelated subsystems and change types, and even the superficially similar PRs target different concrete features or bugs.", + "canonical_pr_reason": "PR 8483 is the broadest and most central artifact here: it adds Stable Diffusion 3 and drives several downstream follow-up fixes/docs around the same model family.", "best_issue_reason": null, - "best_pr_reason": "No clear best PR for duplicate triage. The cluster is too heterogeneous, so picking one would be arbitrary rather than representative of a shared underlying change.", + "best_pr_reason": "PR 8483 is the strongest representative PR in the set because it is a major feature addition with substantial code and docs impact, unlike the smaller one-off fixes and docs tweaks.", "soft_edge_verdicts": [ - { - "left": "pull_request:5842", - "right": "pull_request:6241", - "accept": false, - "reason": "PixArt 512px resolution fix vs a runtime float/half type error; unrelated bugs." - }, - { - "left": "pull_request:5842", - "right": "pull_request:7062", - "accept": false, - "reason": "PixArt resolution bug vs imagic docstring cleanup; different change types and code paths." - }, - { - "left": "pull_request:5844", - "right": "pull_request:6643", - "accept": false, - "reason": "Both are docs PRs, but one adds a Japanese SDXL reference and the other adds UViT2D docs; unrelated content." - }, - { - "left": "pull_request:5855", - "right": "pull_request:6861", - "accept": false, - "reason": "README typo fix vs a large SDXL ControlNet BD change; no shared underlying bug." - }, - { - "left": "pull_request:5895", - "right": "pull_request:5982", - "accept": false, - "reason": "SVD feature addition vs SDXL Turbo docs; different artifacts and purposes." - }, - { - "left": "pull_request:5910", - "right": "pull_request:6941", - "accept": false, - "reason": "IPAdapterFull support vs IP-Adapter support for SDXL ControlNet inpaint; different code paths and problems." - }, - { - "left": "pull_request:5911", - "right": "pull_request:6192", - "accept": false, - "reason": "IPAdapterFull support vs unload_ip_adapter; one adds loading support, the other adds removal support." - }, - { - "left": "pull_request:5911", - "right": "pull_request:6656", - "accept": false, - "reason": "IPAdapterFull support vs unload_textual_inversion; different loaders and unrelated functionality." - }, - { - "left": "pull_request:5920", - "right": "pull_request:6276", - "accept": false, - "reason": "IPEX LoRA support vs community IPAdapter FaceID pipeline; separate features." - }, { "left": "pull_request:5983", "right": "pull_request:7510", "accept": false, - "reason": "SVD documentation tweak vs a VAE tiling OOM test fix; unrelated." + "reason": "Different bugs: SVD docs cleanup vs a VAE tiling OOM test fix." }, { "left": "pull_request:6", "right": "pull_request:7", "accept": false, - "reason": "Adding UNetLDMModel init export vs adding VQModel export; same era, but different model additions." + "reason": "Both are import/init updates, but they expose different models and solve different changes." }, { "left": "pull_request:6", "right": "pull_request:96", "accept": false, - "reason": "UNetLDM init export vs broader LDM finalization and conditional UNet work; not the same change." + "reason": "Different model-setup changes in the early LDM stack; not the same code change." }, { "left": "pull_request:6002", "right": "pull_request:7653", "accept": false, - "reason": "SVD guidance-scale fix vs configuration decorator refactor; unrelated code paths." + "reason": "SVD guidance-scale logic vs a configuration decorator fix; unrelated code paths." }, { "left": "pull_request:6002", "right": "pull_request:8272", "accept": false, - "reason": "SVD guidance-scale logic vs grammatical error fixes in scheduler messages; unrelated." + "reason": "Different fixes: guidance-scale handling versus grammar cleanup in error strings." }, { "left": "pull_request:62", "right": "pull_request:662", "accept": false, - "reason": "Latent diffusion pipeline decode fix vs early 6GB textual inversion setup; different problems." + "reason": "Early latent-diffusion pipeline fix vs a separate 6GB textual-inversion/docs change." }, { "left": "pull_request:62", "right": "pull_request:96", "accept": false, - "reason": "Bug fix in unconditional LDM pipeline vs model init/finalization work; not the same underlying issue." + "reason": "Different latent-diffusion refactors; one fixes unconditional decode behavior, the other finalizes model wiring." }, { "left": "pull_request:6236", "right": "pull_request:6910", "accept": false, - "reason": "ControlNet inpaint docs example fix vs ControlNet model-card standardization; both touch ControlNet but not the same bug/change." + "reason": "Both mention ControlNet, but one is an inpainting doc example and the other is model-card generation in training code." }, { "left": "pull_request:6262", "right": "pull_request:8419", "accept": false, - "reason": "Panorama refactor vs cached_download removal in dynamic modules; unrelated." + "reason": "Panorama refactor versus dynamic module download logic; unrelated changes." }, { "left": "pull_request:6263", "right": "pull_request:8419", "accept": false, - "reason": "LDM3D refactor vs cached_download removal; no overlap in underlying issue." + "reason": "LDM3D refactor versus dynamic module download logic; unrelated changes." }, { "left": "pull_request:6264", "right": "pull_request:8419", "accept": false, - "reason": "SAG refactor vs cached_download removal; unrelated code and intent." + "reason": "SAG refactor versus dynamic module download logic; unrelated changes." }, { "left": "pull_request:6296", "right": "pull_request:6344", "accept": false, - "reason": "Initial aMUSEd pipeline addition vs later repo-link updates for the same feature; related topic but not the same concrete change." + "reason": "Same feature area (aMUSEd), but 6344 is only a later link-update follow-up, not the same change." }, { "left": "pull_request:6296", "right": "pull_request:6462", "accept": false, - "reason": "aMUSEd pipeline addition vs paper-link correction in docs; docs-only follow-up, not duplicate fix." + "reason": "Both touch aMUSEd docs, but one adds the pipeline and the other only updates the paper link." }, { "left": "pull_request:6308", "right": "pull_request:8425", "accept": false, - "reason": "LoRA PEFT dummy component test fix vs workflow rename; unrelated." + "reason": "LoRA dummy-component test fix versus a workflow filename/name tweak; unrelated." }, { "left": "pull_request:6322", "right": "pull_request:7510", "accept": false, - "reason": "Training script attention fix vs VAE tiling OOM test fix; different code paths." + "reason": "Training-script scheduler change versus VAE tiling OOM test fix; different issues." }, { "left": "pull_request:6322", "right": "pull_request:8481", "accept": false, - "reason": "Training script fix vs image processor error-message bug; unrelated." + "reason": "Training-script fix versus image-processor error-message fix; unrelated." }, { "left": "pull_request:6326", "right": "pull_request:6410", "accept": false, - "reason": "Guidance-scale handling in TensorRT img2img vs value-guided sampling module registration; different bugs." + "reason": "TensorRT guidance_scale bug fix versus module-registration refactor in experimental RL code." }, { "left": "pull_request:6332", "right": "pull_request:6759", "accept": false, - "reason": "Rerender-A-Video community pipeline addition vs InstantID bug fix; unrelated." + "reason": "Community video-rerender pipeline versus an InstantID batching fix; different features." }, { "left": "pull_request:6337", "right": "pull_request:7510", "accept": false, - "reason": "retrieve_latents fix in TensorRT img2img vs VAE tiling OOM test fix; unrelated." + "reason": "retrieve_latents bug fix versus a VAE tiling OOM test fix; unrelated." }, { "left": "pull_request:6337", "right": "pull_request:8481", "accept": false, - "reason": "retrieve_latents change vs image processor format error fix; different code paths." + "reason": "Different code paths: latent retrieval in TensorRT img2img versus image input validation messaging." }, { "left": "pull_request:6339", "right": "pull_request:7510", "accept": false, - "reason": "AnimateDiff docs tweak vs VAE tiling OOM test fix; unrelated." + "reason": "AnimateDiff docs update versus VAE tiling OOM test fix; unrelated." }, { "left": "pull_request:6339", "right": "pull_request:8481", "accept": false, - "reason": "AnimateDiff docs tweak vs image processor error-message fix; unrelated." + "reason": "AnimateDiff docs versus image-processor validation messaging; unrelated." }, { "left": "pull_request:6373", "right": "pull_request:8698", "accept": false, - "reason": "Workflow deletion vs transformer2d deprecation cleanup; unrelated maintenance changes." + "reason": "Workflow deletion versus transformer output deprecation cleanup; unrelated." + }, + { + "left": "pull_request:6389", + "right": "pull_request:7087", + "accept": false, + "reason": "Both are SVD-related, but one changes a parameter type and the other adds docs/example content." + }, + { + "left": "pull_request:6410", + "right": "pull_request:7077", + "accept": false, + "reason": "Experimental RL module registration versus IP-Adapter attention reshaping bug; different problems." + }, + { + "left": "pull_request:6410", + "right": "pull_request:8426", + "accept": false, + "reason": "Experimental RL cleanup versus workflow dependency changes; unrelated." + }, + { + "left": "pull_request:6410", + "right": "pull_request:8491", + "accept": false, + "reason": "Experimental RL module registration versus DualTransformer2D import-path fix; unrelated." + }, + { + "left": "pull_request:6410", + "right": "pull_request:8656", + "accept": false, + "reason": "Experimental RL module registration versus shard-checkpoint loading fixes; unrelated." + }, + { + "left": "pull_request:6433", + "right": "pull_request:6487", + "accept": false, + "reason": "Community gluegen pipeline versus Stable Cascade/Wuerstchen release work; unrelated." + }, + { + "left": "pull_request:6453", + "right": "pull_request:6800", + "accept": false, + "reason": "aMUSEd paper-link docs update versus adding a torchvision availability helper; unrelated." + }, + { + "left": "pull_request:6462", + "right": "pull_request:7653", + "accept": false, + "reason": "aMUSEd docs link correction versus config initializer decorator fix; unrelated." + }, + { + "left": "pull_request:6487", + "right": "pull_request:8483", + "accept": false, + "reason": "Both are major model-release PRs, but they implement different model families and code paths." + }, + { + "left": "pull_request:6498", + "right": "pull_request:8268", + "accept": false, + "reason": "Inpainting docs about mask blur/padding versus from_single_file documentation cleanup; unrelated docs changes." + }, + { + "left": "pull_request:6498", + "right": "pull_request:8654", + "accept": false, + "reason": "Different docs topics: inpainting mask handling versus SD3 token-limit guidance." + }, + { + "left": "pull_request:6564", + "right": "pull_request:7720", + "accept": false, + "reason": "IP-Adapter path handling versus dead-code/f-string cleanup; unrelated." + }, + { + "left": "pull_request:662", + "right": "pull_request:78", + "accept": false, + "reason": "Textual inversion 6GB fix versus a mid-block architecture fix; different areas and code paths." } ] }, @@ -175962,7 +176173,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 234, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -175970,14 +176181,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:6389", - "pull_request:6410", - "pull_request:6433", - "pull_request:6453", - "pull_request:6462", - "pull_request:6487", - "pull_request:6498", - "pull_request:6564", "pull_request:662", "pull_request:6645", "pull_request:6754", @@ -175993,46 +176196,40 @@ "pull_request:693", "pull_request:6989", "pull_request:6998", + "pull_request:7", + "pull_request:7024", + "pull_request:7034", + "pull_request:705", "pull_request:7077", - "pull_request:7087", + "pull_request:7113", + "pull_request:7134", + "pull_request:7156", "pull_request:7259", "pull_request:7307", + "pull_request:74", "pull_request:75", - "pull_request:7653", + "pull_request:7618", "pull_request:7709", - "pull_request:7720", - "pull_request:78", + "pull_request:79", "pull_request:802", "pull_request:822", - "pull_request:8268", + "pull_request:824", "pull_request:8378", - "pull_request:8426", + "pull_request:8452", "pull_request:8483", - "pull_request:8491", "pull_request:8607", - "pull_request:8654", - "pull_request:8656", + "pull_request:8684", "pull_request:8688", "pull_request:8705", + "pull_request:8836", "pull_request:8920", "pull_request:8990", + "pull_request:9065", "pull_request:921", - "pull_request:978" + "pull_request:978", + "pull_request:979" ], "soft_pairs": [ - "pull_request:6389|pull_request:7087", - "pull_request:6410|pull_request:7077", - "pull_request:6410|pull_request:8426", - "pull_request:6410|pull_request:8491", - "pull_request:6410|pull_request:8656", - "pull_request:6433|pull_request:6487", - "pull_request:6453|pull_request:6800", - "pull_request:6462|pull_request:7653", - "pull_request:6487|pull_request:8483", - "pull_request:6498|pull_request:8268", - "pull_request:6498|pull_request:8654", - "pull_request:6564|pull_request:7720", - "pull_request:662|pull_request:78", "pull_request:662|pull_request:802", "pull_request:6645|pull_request:8378", "pull_request:6754|pull_request:7307", @@ -176050,222 +176247,221 @@ "pull_request:6904|pull_request:8688", "pull_request:693|pull_request:75", "pull_request:6989|pull_request:8990", - "pull_request:6998|pull_request:8920" + "pull_request:6998|pull_request:8920", + "pull_request:7|pull_request:74", + "pull_request:7|pull_request:79", + "pull_request:7|pull_request:979", + "pull_request:7024|pull_request:8607", + "pull_request:7034|pull_request:8452", + "pull_request:705|pull_request:824", + "pull_request:7077|pull_request:8836", + "pull_request:7113|pull_request:9065", + "pull_request:7134|pull_request:8684", + "pull_request:7156|pull_request:7618", + "pull_request:7156|pull_request:8688" ], - "prepared_review_unit_hash": "40174d632267727b8a4f15e820192f03abb4efc2423eb1618c32548e7fe03e42", + "prepared_review_unit_hash": "02cb5c579305005ee78e4fa98942a7624ce1448b99d13cab918d6baef49410a5", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 31, - "serialized_chars": 98973, - "estimated_input_tokens": 24744, - "estimated_eval_tokens": 49744 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 96408, + "estimated_input_tokens": 24102, + "estimated_eval_tokens": 48460 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly unrelated PRs spanning docs, CI, feature additions, and small bug fixes. The soft links are all false positives; none of the paired PRs look like the same underlying change or could be merged into one PR.", - "confidence": 0.98, + "summary": "This cluster is a false positive: the PRs cover unrelated docs, CI, dependency, and code-fix changes. None of the soft-edge pairs look like the same underlying bug or change, so no merge candidate is safe.", + "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR #8483 is the most substantial and central artifact in the set: it introduces Stable Diffusion 3 with broad code, docs, loader, and model changes. It is the best representative PR if one must be chosen, though it is not a duplicate anchor for the rest of the cluster.", + "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "pull_request:6389", - "right": "pull_request:7087", - "accept": false, - "reason": "Both touch stable video diffusion, but #6389 only fixes the `noise_aug_strength` type while #7087 is a docs/API cleanup for the pipeline. Different change types and no shared bug." - }, - { - "left": "pull_request:6410", - "right": "pull_request:7077", - "accept": false, - "reason": "#6410 fixes `register_modules` usage in an RL helper; #7077 changes `head_to_batch_dim` to support IPAdapter attention shapes. Unrelated code paths." - }, - { - "left": "pull_request:6410", - "right": "pull_request:8426", - "accept": false, - "reason": "#6410 is an RL module registration fix, while #8426 updates a CI mirror workflow. No overlap in bug or component." - }, - { - "left": "pull_request:6410", - "right": "pull_request:8491", - "accept": false, - "reason": "#6410 concerns RL sampling internals; #8491 fixes a DualTransformer2D import path. Different subsystems and problems." - }, - { - "left": "pull_request:6410", - "right": "pull_request:8656", - "accept": false, - "reason": "#6410 is about module registration in experimental RL; #8656 fixes shard checkpoint loading in modeling utils. Not the same issue." - }, - { - "left": "pull_request:6433", - "right": "pull_request:6487", - "accept": false, - "reason": "#6433 adds the GlueGen community pipeline; #6487 adds Stable Cascade pipeline/docs/model files. Both are feature additions, but completely different pipelines." - }, - { - "left": "pull_request:6453", - "right": "pull_request:6800", - "accept": false, - "reason": "#6453 only adds a paper link in aMUSEd docs; #6800 adds `is_torchvision_available` plumbing across utils/training. Different scope and intent." - }, - { - "left": "pull_request:6462", - "right": "pull_request:7653", - "accept": false, - "reason": "#6462 is a docs link correction for aMUSEd; #7653 fixes a wrong decorator/helper signature in configuration utils. No shared code path." - }, - { - "left": "pull_request:6487", - "right": "pull_request:8483", - "accept": false, - "reason": "Both are large model additions, but Stable Cascade and Stable Diffusion 3 are separate feature sets with different models, docs, loaders, and tests." - }, - { - "left": "pull_request:6498", - "right": "pull_request:8268", - "accept": false, - "reason": "#6498 adds inpaint docs for mask blur/padding; #8268 cleans up from_single_file docs. Both are docs-only, but unrelated topics." - }, - { - "left": "pull_request:6498", - "right": "pull_request:8654", - "accept": false, - "reason": "#6498 is inpainting documentation; #8654 adds SD3 T5 token-limit docs. Different documentation subjects and no shared code change." - }, - { - "left": "pull_request:6564", - "right": "pull_request:7720", - "accept": false, - "reason": "#6564 fixes IP Adapter path generation; #7720 removes dead code and fixes an f-string issue across several files. Only superficial utility overlap." - }, - { - "left": "pull_request:662", - "right": "pull_request:78", - "accept": false, - "reason": "#662 is a large 6GB memory/textual inversion notebook and example update; #78 is an early attention/mid-block implementation change. Not the same bug or feature." - }, { "left": "pull_request:662", "right": "pull_request:802", "accept": false, - "reason": "#662 is a broad example/model adaptation for low-memory textual inversion; #802 only fixes indentation in a custom-pipeline code example." + "reason": "Different docs/code changes in different areas; not the same bug or fix." }, { "left": "pull_request:6645", "right": "pull_request:8378", "accept": false, - "reason": "#6645 updates the community README; #8378 fixes AsymmetricAutoencoderKL forward behavior. Different file types and different problems." + "reason": "Community README edit vs asymmetric autoencoder code fix; unrelated." }, { "left": "pull_request:6754", "right": "pull_request:7307", "accept": false, - "reason": "#6754 fixes DPO loss reduction in training scripts; #7307 adds a TensorRT community pipeline and environment spec. Unrelated training bug vs feature work." + "reason": "DPO loss correction vs TensorRT example update; different problems." }, { "left": "pull_request:6759", "right": "pull_request:6929", "accept": false, - "reason": "#6759 fixes batching in an InstantID community pipeline; #6929 adds docs and local-loading support for community pipelines. Different artifacts and code paths." + "reason": "InstantID prompt image embedding repetition vs docs for community pipelines; unrelated." }, { "left": "pull_request:6765", "right": "pull_request:7307", "accept": false, - "reason": "#6765 changes cache path handling in hub_utils; #7307 is a TensorRT pipeline update with a conda env. No same underlying issue." + "reason": "Cache-path utility fix vs TensorRT example/diffusers.yml update; not same change." }, { "left": "pull_request:678", "right": "pull_request:921", "accept": false, - "reason": "#678 is an early DreamBooth/Stable Diffusion update with model and notebook changes; #921 is v1-5 docs and API updates. Broadly related to Stable Diffusion, but not the same change." + "reason": "DreamBooth notebook/model updates vs v1-5 docs refresh; far broader and different." }, { "left": "pull_request:6800", "right": "pull_request:6989", "accept": false, - "reason": "#6800 adds a torchvision availability helper; #6989 only adds `peft` to a benchmark workflow. Different scope and no shared bug." + "reason": "Torchvision availability helper vs benchmark workflow dependency; unrelated." }, { "left": "pull_request:6800", "right": "pull_request:7259", "accept": false, - "reason": "#6800 is a utility/import availability fix; #7259 introduces TCD docs and navigation. No common code-path problem." + "reason": "Torchvision helper addition vs TCD docs; no shared code-path bug." }, { "left": "pull_request:6800", "right": "pull_request:7709", "accept": false, - "reason": "#6800 affects import utilities; #7709 changes a push-tests workflow and container privileges. Unrelated." + "reason": "Import helper addition vs CI/Tailscale workflow; different subsystems." }, { "left": "pull_request:6800", "right": "pull_request:8483", "accept": false, - "reason": "#6800 is a small utility addition; #8483 is a major SD3 feature landing. Not the same fix or change." + "reason": "Torchvision helper addition vs SD3 feature work; not the same bug or mergeable fix." }, { "left": "pull_request:6800", "right": "pull_request:8705", "accept": false, - "reason": "#6800 is about torchvision detection; #8705 adds a benchmarking Docker container. Different infrastructure concerns." + "reason": "Import helper addition vs new benchmarking Docker container; unrelated." }, { "left": "pull_request:6837", "right": "pull_request:8607", "accept": false, - "reason": "#6837 only tweaks example import order in i2vgen-xl docs; #8607 adds TAESD3 docs. Both docs-related, but unrelated models/topics." + "reason": "I2VGen docs import order vs SD3 TAESD docs; only both are docs." }, { "left": "pull_request:688", "right": "pull_request:822", "accept": false, - "reason": "#688 fixes Flax/PyTorch import handling; #822 updates the Flax scheduler API and latents scaling. Different functional issues." + "reason": "Flax/PyTorch import handling vs Flax scheduler API update; different underlying issues." }, { "left": "pull_request:689", "right": "pull_request:978", "accept": false, - "reason": "#689 changes attention internals for performance; #978 is typo/doc cleanup across examples and pipeline docs. No shared bug." + "reason": "Attention performance tweak vs typo/docs cleanup; unrelated." }, { "left": "pull_request:6904", "right": "pull_request:8688", "accept": false, - "reason": "#6904 adds PIL image support in `load_image`; #8688 fixes DoRA detection in LoRA conversion. Both are loader utilities, but they address different input paths and bugs." + "reason": "PIL-image support in load_image vs LoRA DoRA conversion logic; different code paths." }, { "left": "pull_request:693", "right": "pull_request:75", "accept": false, - "reason": "#693 only corrects a BibTeX citation; #75 changes attention/unet implementation for Glide. Completely different artifact types and intent." + "reason": "BibTeX citation fix vs Glide/attention model plumbing; unrelated." }, { "left": "pull_request:6989", "right": "pull_request:8990", "accept": false, - "reason": "#6989 updates a benchmark workflow dependency list; #8990 adds the StableDiffusionXLControlNetPAGImg2Img pipeline and tests. Infrastructure vs model feature." + "reason": "Benchmark workflow dependency install vs new PAG pipeline implementation; different scope." }, { "left": "pull_request:6998", "right": "pull_request:8920", "accept": false, - "reason": "Both are docs fixes, but #6998 only changes a callout marker in the IP-Adapter guide while #8920 makes small PAG guide edits. Different guide pages and no shared bug." + "reason": "Callout markdown fix vs PAG guide text edits; both docs but not the same change." + }, + { + "left": "pull_request:7", + "right": "pull_request:74", + "accept": false, + "reason": "Initial VQ-VAE addition vs Grad-TTS mid-block change; distinct model features." + }, + { + "left": "pull_request:7", + "right": "pull_request:79", + "accept": false, + "reason": "VQ-VAE addition vs unconditional image generation; only shares early repo history." + }, + { + "left": "pull_request:7", + "right": "pull_request:979", + "accept": false, + "reason": "VQ-VAE model addition vs missing import in README; unrelated." + }, + { + "left": "pull_request:7024", + "right": "pull_request:8607", + "accept": false, + "reason": "AnimateLCM docs vs SD3 TAESD docs; different topics." + }, + { + "left": "pull_request:7034", + "right": "pull_request:8452", + "accept": false, + "reason": "Both touch single_file_utils, but one fixes ControlNet dtype handling and the other CLIP load strictness; not same bug." + }, + { + "left": "pull_request:705", + "right": "pull_request:824", + "accept": false, + "reason": "CI CUDA workflow changes vs Flax scheduler/test updates; unrelated." + }, + { + "left": "pull_request:7077", + "right": "pull_request:8836", + "accept": false, + "reason": "IP-Adapter head_to_batch_dim fix vs SlicedAttnProcessor loop fix; different attention bugs." + }, + { + "left": "pull_request:7113", + "right": "pull_request:9065", + "accept": false, + "reason": "Docker Slack notification workflow vs sentencepiece optional dependency support; unrelated." + }, + { + "left": "pull_request:7134", + "right": "pull_request:8684", + "accept": false, + "reason": "UNet type hints vs Pillow import docs fix; no shared underlying bug." + }, + { + "left": "pull_request:7156", + "right": "pull_request:7618", + "accept": false, + "reason": "Stalebot label handling vs CPU offload hook cleanup; unrelated automation vs pipeline behavior." + }, + { + "left": "pull_request:7156", + "right": "pull_request:8688", + "accept": false, + "reason": "Stalebot logic vs LoRA DoRA conversion; unrelated." } ] }, @@ -176281,7 +176477,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 235, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -176289,20 +176485,11 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:7", - "pull_request:7024", - "pull_request:7034", - "pull_request:705", - "pull_request:7077", - "pull_request:7113", - "pull_request:7134", - "pull_request:7156", "pull_request:718", "pull_request:7212", "pull_request:7213", "pull_request:7259", "pull_request:7339", - "pull_request:74", "pull_request:7458", "pull_request:7464", "pull_request:7472", @@ -176311,45 +176498,42 @@ "pull_request:7520", "pull_request:7564", "pull_request:7594", - "pull_request:7618", "pull_request:7627", "pull_request:7709", "pull_request:774", "pull_request:7778", + "pull_request:7790", + "pull_request:7791", "pull_request:78", - "pull_request:79", + "pull_request:7857", "pull_request:7913", + "pull_request:7919", + "pull_request:7938", + "pull_request:7977", + "pull_request:800", "pull_request:819", "pull_request:8224", + "pull_request:8225", "pull_request:824", + "pull_request:828", "pull_request:8412", "pull_request:8426", - "pull_request:8452", + "pull_request:8467", "pull_request:8491", - "pull_request:8607", - "pull_request:8684", - "pull_request:8688", "pull_request:8836", "pull_request:8990", + "pull_request:9015", "pull_request:9036", - "pull_request:9065", + "pull_request:904", "pull_request:9116", "pull_request:9251", "pull_request:9463", - "pull_request:979" + "pull_request:9649", + "pull_request:9684", + "pull_request:9721", + "pull_request:9769" ], "soft_pairs": [ - "pull_request:7|pull_request:74", - "pull_request:7|pull_request:79", - "pull_request:7|pull_request:979", - "pull_request:7024|pull_request:8607", - "pull_request:7034|pull_request:8452", - "pull_request:705|pull_request:824", - "pull_request:7077|pull_request:8836", - "pull_request:7113|pull_request:9065", - "pull_request:7134|pull_request:8684", - "pull_request:7156|pull_request:7618", - "pull_request:7156|pull_request:8688", "pull_request:718|pull_request:78", "pull_request:7212|pull_request:8836", "pull_request:7213|pull_request:8412", @@ -176368,216 +176552,227 @@ "pull_request:7594|pull_request:8224", "pull_request:7627|pull_request:7913", "pull_request:7709|pull_request:8990", - "pull_request:774|pull_request:78" + "pull_request:774|pull_request:78", + "pull_request:7790|pull_request:8467", + "pull_request:7791|pull_request:8836", + "pull_request:7857|pull_request:9649", + "pull_request:7919|pull_request:9684", + "pull_request:7919|pull_request:9769", + "pull_request:7938|pull_request:9721", + "pull_request:7977|pull_request:9036", + "pull_request:7977|pull_request:9116", + "pull_request:800|pull_request:904", + "pull_request:8225|pull_request:9015", + "pull_request:824|pull_request:828" ], - "prepared_review_unit_hash": "eaf255858bfd31b8bbe7d39ec1e08b2525568c429f32a17c9af246a1ae60eaeb", + "prepared_review_unit_hash": "78cb2a596b55b568694a77965810cc92e784316e0579f5df9f1e7ac009c6ce76", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 94884, - "estimated_input_tokens": 23721, - "estimated_eval_tokens": 47698 + "serialized_chars": 96503, + "estimated_input_tokens": 24126, + "estimated_eval_tokens": 48508 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a false-positive mix of unrelated PRs spanning model code, docs, CI/workflows, tests, and small utility fixes. None of the proposed soft edges appear to be the same underlying bug or change.", - "confidence": 0.99, + "summary": "This cluster is highly heterogeneous: it mixes typo/docs fixes, workflow changes, test-only adjustments, and unrelated core model/pipeline feature work. I don\u2019t see a single underlying bug/change tying the items together, so there is no true duplicate canonical artifact here.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: there is no single underlying fix here, just a heterogeneous set of unrelated changes.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No best PR candidate for duplicate triage; the cluster is not centered on one concrete code-path or feature change.", + "best_pr_reason": "If forced to pick one representative PR, 9769 is the most substantial and self-contained change set, with an explicit issue target and broad, coherent feature work.", "soft_edge_verdicts": [ - { - "left": "pull_request:7", - "right": "pull_request:74", - "accept": false, - "reason": "VQModel addition vs Grad-TTS mid-block change; unrelated model features." - }, - { - "left": "pull_request:7", - "right": "pull_request:79", - "accept": false, - "reason": "Shared init exports only; VQModel addition and unconditional image generation are different changes." - }, - { - "left": "pull_request:7", - "right": "pull_request:979", - "accept": false, - "reason": "Model addition vs README import fix; not the same bug/change." - }, - { - "left": "pull_request:7024", - "right": "pull_request:8607", - "accept": false, - "reason": "Separate docs for AnimateLCM and TAESD3; different pipelines and topics." - }, - { - "left": "pull_request:7034", - "right": "pull_request:8452", - "accept": false, - "reason": "Both touch single_file_utils, but one fixes ControlNet dtype loading and the other relaxes CLIP load strictness; different code paths." - }, - { - "left": "pull_request:705", - "right": "pull_request:824", - "accept": false, - "reason": "CI CUDA workflow changes vs Flax test setup; unrelated." - }, - { - "left": "pull_request:7077", - "right": "pull_request:8836", - "accept": false, - "reason": "Different attention bugs: IPAdapter head reshaping vs sliced attention loop bounds." - }, - { - "left": "pull_request:7113", - "right": "pull_request:9065", - "accept": false, - "reason": "Docker CI Slack notification workflow vs sentencepiece optional dependency wiring; unrelated." - }, - { - "left": "pull_request:7134", - "right": "pull_request:8684", - "accept": false, - "reason": "Type-hint/doc cleanup in UNet vs Pillow import fix in a pipeline docstring; unrelated." - }, - { - "left": "pull_request:7156", - "right": "pull_request:7618", - "accept": false, - "reason": "Stalebot label handling vs CPU offload slow-test fix; different subsystems." - }, - { - "left": "pull_request:7156", - "right": "pull_request:8688", - "accept": false, - "reason": "Stalebot utility change vs LoRA DoRA conversion fix; unrelated." - }, { "left": "pull_request:718", "right": "pull_request:78", "accept": false, - "reason": "Typos/docs cleanup vs mid-block implementation fix; not the same change." + "reason": "Typos/docs cleanup versus a TemporalBasicTransformerBlock code fix; unrelated changes." }, { "left": "pull_request:7212", "right": "pull_request:8836", "accept": false, - "reason": "TemporalBasicTransformerBlock cleanup vs sliced attention processor bug; different components." + "reason": "One removes duplicated block code; the other fixes a sliced-attention loop boundary bug. Different concrete problems." }, { "left": "pull_request:7213", "right": "pull_request:8412", "accept": false, - "reason": "Docs for merging LoRAs vs single-file loading docs; related theme but not the same change." + "reason": "Merge-LoRAs docs versus single-file loading docs; both docs, but different features and intent." }, { "left": "pull_request:7259", "right": "pull_request:8990", "accept": false, - "reason": "TCD docs addition vs PAG img2img pipeline implementation; different features." + "reason": "TCD intro docs versus PAG img2img feature work; different pipelines and deliverables." }, { "left": "pull_request:7339", "right": "pull_request:7778", "accept": false, - "reason": "Release workflow dependency fix vs debugging SSH workflow; unrelated CI changes." + "reason": "Release dependency workflow fix versus a debugging SSH workflow; unrelated CI changes." }, { "left": "pull_request:7458", "right": "pull_request:9251", "accept": false, - "reason": "Python 3.12 test skipping vs patch-branch test triggering; different test-infra concerns." + "reason": "Skipping dynamo tests on Python 3.12 is unrelated to triggering fast tests on patch branches." }, { "left": "pull_request:7464", "right": "pull_request:8426", "accept": false, - "reason": "LDM3D example import fix vs mirror community pipeline workflow; unrelated." + "reason": "LDM3D import fix versus mirror-community workflow dependency fix; no shared bug." }, { "left": "pull_request:7464", "right": "pull_request:8491", "accept": false, - "reason": "LDM3D example fix vs DualTransformer2D import path fix; unrelated." + "reason": "LDM3D fix and dual-transformer import move touch different modules and issues." }, { "left": "pull_request:7472", "right": "pull_request:9036", "accept": false, - "reason": "Broken link in examples README vs nightly dependency install fix; unrelated." + "reason": "Broken link in examples README versus nightly deps workflow fix; different scope." }, { "left": "pull_request:748", "right": "pull_request:819", "accept": false, - "reason": "Training push_to_hub API fix vs DDIM timestep int64 fix; different bugs." + "reason": "push_to_hub call-site fix in examples versus DDIM timestep int64 fix; unrelated." }, { "left": "pull_request:7510", "right": "pull_request:9036", "accept": false, - "reason": "VAE tiling OOM fix vs nightly deps workflow fix; unrelated." + "reason": "OOM-related test tweak versus nightly dependency workflow update; no common bug." }, { "left": "pull_request:7510", "right": "pull_request:9116", "accept": false, - "reason": "VAE tiling OOM fix vs dead-link docs fix; unrelated." + "reason": "Test VAE tiling memory fix versus dead-link docs fix; unrelated." }, { "left": "pull_request:7520", "right": "pull_request:9463", "accept": false, - "reason": "HD-Painter pipeline addition vs PAG img2img support; different features." + "reason": "HD-Painter community pipeline addition versus PAG SD img2img support; different features." }, { "left": "pull_request:7564", "right": "pull_request:9036", "accept": false, - "reason": "Test fixes/cpu offload changes vs nightly dependency workflow fix; unrelated." + "reason": "Test adjustments for Python 3.12/dynamo versus a workflow dependency change." }, { "left": "pull_request:7564", "right": "pull_request:9116", "accept": false, - "reason": "Test fixes vs dead-link docs fix; unrelated." + "reason": "Test/runtime compatibility fixes versus a documentation dead-link fix." }, { "left": "pull_request:7594", "right": "pull_request:8224", "accept": false, - "reason": "Callback docs bugfix vs DeepFloyd training docs; unrelated documentation changes." + "reason": "Callback docs bugfix versus DeepFloyd training docs; different documentation topics." }, { "left": "pull_request:7627", "right": "pull_request:7913", "accept": false, - "reason": "Playground img2img pipeline fix vs custom timesteps/sigmas support; different pipeline behaviors." + "reason": "Playground img2img latent handling fix versus custom sigmas/timesteps support; separate pipeline behaviors." }, { "left": "pull_request:7709", "right": "pull_request:8990", "accept": false, - "reason": "Tailscale CI workflow addition vs PAG pipeline implementation; unrelated." + "reason": "Workflow tailscale setup versus a large Mochi model/pipeline feature addition." }, { "left": "pull_request:774", "right": "pull_request:78", "accept": false, - "reason": "Test tolerance tweak vs mid-block code fix; not the same bug." + "reason": "Tiny test assertion tweaks versus a mid-block model refactor; unrelated." + }, + { + "left": "pull_request:7790", + "right": "pull_request:8467", + "accept": false, + "reason": "Masked SDXL community pipeline adaptation versus SEGA pipeline device placement fix; different code paths." + }, + { + "left": "pull_request:7791", + "right": "pull_request:8836", + "accept": false, + "reason": "AnimateDiff/IP-Adapter compatibility fix versus sliced-attention loop bug; separate bugs." + }, + { + "left": "pull_request:7857", + "right": "pull_request:9649", + "accept": false, + "reason": "PixArtSigma docs/exports versus scheduler return-value API change; unrelated." + }, + { + "left": "pull_request:7919", + "right": "pull_request:9684", + "accept": false, + "reason": "Device-map offload behavior fix versus CogVideoX frame batching improvements; not the same issue." + }, + { + "left": "pull_request:7919", + "right": "pull_request:9769", + "accept": false, + "reason": "Offload behavior fix versus Mochi model/pipeline feature work; completely different changes." + }, + { + "left": "pull_request:7938", + "right": "pull_request:9721", + "accept": false, + "reason": "Slack reporting workflow refactor versus pinning torch versions in Dockerfiles; unrelated infra updates." + }, + { + "left": "pull_request:7977", + "right": "pull_request:9036", + "accept": false, + "reason": "Attention typo doc fix versus nightly deps workflow fix; no overlap." + }, + { + "left": "pull_request:7977", + "right": "pull_request:9116", + "accept": false, + "reason": "Attention typo docs fix versus dead-link docs fix; both docs, but different problems." + }, + { + "left": "pull_request:800", + "right": "pull_request:904", + "accept": false, + "reason": "WIP VQ Diffusion implementation versus stable diffusion inpainting; entirely different model families." + }, + { + "left": "pull_request:8225", + "right": "pull_request:9015", + "accept": false, + "reason": "Ignoring checkpoint uploads during push versus fixing a PIA example snippet; unrelated." + }, + { + "left": "pull_request:824", + "right": "pull_request:828", + "accept": false, + "reason": "Both are Flax-related test work, but 824 fixes pipeline scheduler formatting while 828 expands/adjusts tests and imports; not the same concrete bug/change." } ] }, @@ -176593,7 +176788,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 236, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -176601,23 +176796,18 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:7790", - "pull_request:7791", - "pull_request:7857", - "pull_request:7919", - "pull_request:7938", - "pull_request:7977", - "pull_request:800", - "pull_request:8225", - "pull_request:824", - "pull_request:828", + "pull_request:6532", + "pull_request:6534", + "pull_request:7030", + "pull_request:7031", + "pull_request:7613", + "pull_request:7615", "pull_request:8298", "pull_request:8310", "pull_request:8386", "pull_request:839", "pull_request:842", "pull_request:8426", - "pull_request:8467", "pull_request:8478", "pull_request:8481", "pull_request:8483", @@ -176631,36 +176821,30 @@ "pull_request:8656", "pull_request:8685", "pull_request:8705", - "pull_request:8836", + "pull_request:8737", + "pull_request:8779", + "pull_request:8851", "pull_request:8978", "pull_request:8990", - "pull_request:9015", "pull_request:9036", - "pull_request:904", + "pull_request:9070", "pull_request:9077", - "pull_request:9116", + "pull_request:9083", + "pull_request:9130", + "pull_request:9196", "pull_request:9249", + "pull_request:9426", "pull_request:9536", + "pull_request:9569", "pull_request:9592", - "pull_request:9649", + "pull_request:9658", "pull_request:9662", - "pull_request:9684", - "pull_request:9721", "pull_request:9739", - "pull_request:9769" + "pull_request:9934", + "pull_request:9938", + "pull_request:9994" ], "soft_pairs": [ - "pull_request:7790|pull_request:8467", - "pull_request:7791|pull_request:8836", - "pull_request:7857|pull_request:9649", - "pull_request:7919|pull_request:9684", - "pull_request:7919|pull_request:9769", - "pull_request:7938|pull_request:9721", - "pull_request:7977|pull_request:9036", - "pull_request:7977|pull_request:9116", - "pull_request:800|pull_request:904", - "pull_request:8225|pull_request:9015", - "pull_request:824|pull_request:828", "pull_request:8298|pull_request:8652", "pull_request:8310|pull_request:8478", "pull_request:8386|pull_request:9536", @@ -176678,220 +176862,228 @@ "pull_request:8656|pull_request:9077", "pull_request:8656|pull_request:9739", "pull_request:8685|pull_request:9592", - "pull_request:8705|pull_request:8990" + "pull_request:8705|pull_request:8990", + "pull_request:8737|pull_request:9070", + "pull_request:8779|pull_request:9569", + "pull_request:8851|pull_request:9934", + "pull_request:8990|pull_request:9130", + "pull_request:8990|pull_request:9196", + "pull_request:9083|pull_request:9658", + "pull_request:9083|pull_request:9994", + "pull_request:9426|pull_request:9938", + "pull_request:7030|pull_request:7031", + "pull_request:6532|pull_request:6534", + "pull_request:7613|pull_request:7615" ], - "prepared_review_unit_hash": "de9dc122a20b999e4033753bb0c59b040014d3913c17a85aa344ff46acaa0bd3", + "prepared_review_unit_hash": "92489cec02a03a3beec180a37ce2851ae7bfda5fbcfb4a04706d1acf6d210f60", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 98406, - "estimated_input_tokens": 24602, - "estimated_eval_tokens": 49460 + "serialized_chars": 96590, + "estimated_input_tokens": 24148, + "estimated_eval_tokens": 48552 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous PRs with only one clearly related pair; there is no meaningful duplicate cluster here beyond a few weak thematic overlaps.", - "confidence": 0.95, + "summary": "The soft-edge suggestions are mostly false positives across unrelated feature, docs, CI, and test changes. Only three pairs are true duplicates of the same concrete change: the DreamBooth torch.compile compatibility fix, the SDXL ControlNet Inpaint IP-Adapter support, and the advanced SDXL LoRA model-card standardization.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR stands out: the set mixes unrelated docs, tests, workflows, and feature additions, so there is no strong duplicate anchor.", + "canonical_pr_reason": "No single canonical PR cleanly spans this mixed cluster; if a fallback is needed, PR 8483 is the strongest general representative because it is a large merged feature PR with substantial review and downstream references.", "best_issue_reason": null, - "best_pr_reason": "No strong global best PR for duplicate triage; if forced, the broadest feature PRs are the most representative, but none is a good canonical duplicate target.", + "best_pr_reason": "PR 8483 is the best overall anchor among these items for the same reason: it is the most substantial merged artifact in the set, whereas most other PRs are narrow and unrelated to one another.", "soft_edge_verdicts": [ - { - "left": "pull_request:7790", - "right": "pull_request:8467", - "accept": false, - "reason": "Different pipelines and different bugs; only a loose SDXL/SEGA theme overlap." - }, - { - "left": "pull_request:7791", - "right": "pull_request:8836", - "accept": false, - "reason": "IP-Adapter AnimateDiff creation fix vs sliced attention loop bug are unrelated." - }, - { - "left": "pull_request:7857", - "right": "pull_request:9649", - "accept": false, - "reason": "Docs for PixArtSigma vs scheduler return behavior in DDIM are unrelated." - }, - { - "left": "pull_request:7919", - "right": "pull_request:9684", - "accept": false, - "reason": "Offload/device_map behavior and CogVideoX VAE batching address different code paths." - }, - { - "left": "pull_request:7919", - "right": "pull_request:9769", - "accept": false, - "reason": "Offload behavior fix vs Mochi model/pipeline addition are unrelated." - }, - { - "left": "pull_request:7938", - "right": "pull_request:9721", - "accept": false, - "reason": "Both touch Docker/workflows, but one changes Slack reporting and the other pins torch versions." - }, - { - "left": "pull_request:7977", - "right": "pull_request:9036", - "accept": false, - "reason": "A typo fix in docs is unrelated to nightly dependency installation." - }, - { - "left": "pull_request:7977", - "right": "pull_request:9116", - "accept": false, - "reason": "Both are docs-only, but they fix different issues (typo vs dead link)." - }, - { - "left": "pull_request:800", - "right": "pull_request:904", - "accept": false, - "reason": "Very old VQ Diffusion work vs stable diffusion inpainting are unrelated." - }, - { - "left": "pull_request:8225", - "right": "pull_request:9015", - "accept": false, - "reason": "Checkpoint upload ignore patterns vs PIA example docs are unrelated." - }, - { - "left": "pull_request:824", - "right": "pull_request:828", - "accept": false, - "reason": "Both are Flax test-related, but one modifies pipeline/test infra and the other only adjusts tests; not the same bug or change." - }, { "left": "pull_request:8298", "right": "pull_request:8652", "accept": false, - "reason": "CLI env platform-info cleanup vs Lumina feature addition are unrelated." + "reason": "Unrelated changes: CLI env platform info cleanup vs a new Lumina-T2X feature PR." }, { "left": "pull_request:8310", "right": "pull_request:8478", "accept": false, - "reason": "HunyuanDiT refactor and HunyuanDiT ControlNet/pipeline addition are separate changes, not one fix." + "reason": "Both are Hunyuan-related, but one is a model refactor and the other adds controlnet/pipeline support; not the same concrete change." }, { "left": "pull_request:8386", "right": "pull_request:9536", "accept": false, - "reason": "Test input-shape cleanup vs ChatGLM tokenizer padding fix are unrelated." + "reason": "Test helper shape cleanup vs ChatGLM tokenizer padding fix; no shared bug or code path." }, { "left": "pull_request:839", "right": "pull_request:842", "accept": false, - "reason": "Different cleanup patches in different areas; not the same underlying change." + "reason": "README auth-token cleanup and pipeline test sample-output cleanup are unrelated." }, { "left": "pull_request:8426", "right": "pull_request:9739", "accept": false, - "reason": "Mirror-pipeline workflow change vs SD3 single-file docs/code are unrelated." + "reason": "Mirror-community CI fix vs SD3 typo/single-file docs change; different concerns." }, { "left": "pull_request:8481", "right": "pull_request:9036", "accept": false, - "reason": "Image processor error message fix and nightly deps update are unrelated." + "reason": "Image processor error-message tweak vs nightly dependency workflow fix; unrelated." }, { "left": "pull_request:8483", "right": "pull_request:8990", "accept": false, - "reason": "SD3 core feature vs SDXL ControlNet PAG Img2Img are different model features and code paths." + "reason": "Major SD3 feature addition vs SDXL ControlNet PAG img2img pipeline addition; same broad area, different feature and code paths." }, { "left": "pull_request:8491", "right": "pull_request:9077", "accept": false, - "reason": "Import-path fix and typo fixes are not the same concrete issue." + "reason": "Dual-transformer import fix vs typo cleanup and controlnet_xs changes; unrelated." }, { "left": "pull_request:8491", "right": "pull_request:9739", "accept": false, - "reason": "Both mention fixes, but they touch different files and problems; no shared bug." + "reason": "Import-path fix vs SD3 typo/single-file docs change; unrelated." }, { "left": "pull_request:8521", "right": "pull_request:8978", "accept": false, - "reason": "Requirement file formatting vs a large example update are unrelated." + "reason": "Requirements pin cleanup vs broad notebook/example update; not the same underlying issue." }, { "left": "pull_request:8522", "right": "pull_request:9662", "accept": false, - "reason": "Both touch transformers versions, but one changes install requirements and the other a test version gate; not a duplicate fix." + "reason": "Both touch transformers versions, but one changes package pins and the other adjusts a test requirement; not the same concrete change." }, { "left": "pull_request:8583", "right": "pull_request:8607", "accept": false, - "reason": "XFormers support in SD3 and TAESD3 docs are separate feature areas." + "reason": "XFormers SD3 support vs TAESD3 docs; same ecosystem, different changes." }, { "left": "pull_request:8607", "right": "pull_request:8618", - "accept": true, - "reason": "Same SD3 TAESD3 shift_factor change: the docs example and the code support line up on the same feature." + "accept": false, + "reason": "Docs for TAESD3 usage vs code support for shift_factor; closely related, but not the same artifact or bug fix." }, { "left": "pull_request:8607", "right": "pull_request:9249", "accept": false, - "reason": "TAESD3 docs vs SD3 ControlNet example are unrelated." + "reason": "TAESD3 docs vs SD3 ControlNet training example; unrelated feature docs." }, { "left": "pull_request:8656", "right": "pull_request:9077", "accept": false, - "reason": "Shard-checkpoint fix vs typo cleanup are unrelated." + "reason": "Shard-checkpoint loading fix vs typo cleanup/controlnet_xs edits; unrelated." }, { "left": "pull_request:8656", "right": "pull_request:9739", "accept": false, - "reason": "Shard-checkpoint fix and single-file SD3 docs/code are unrelated." + "reason": "Sharded checkpoint fix vs SD3 typo/single-file docs update; unrelated." }, { "left": "pull_request:8685", "right": "pull_request:9592", "accept": false, - "reason": "Float8 note in SD3 docs vs Flax docstring refactor are unrelated." + "reason": "SD3 float8 note vs Flax docstring refactor; both docs-oriented but not the same change." }, { "left": "pull_request:8705", "right": "pull_request:8990", "accept": false, - "reason": "Benchmarking docker container addition is unrelated to the SDXL PAG pipeline." + "reason": "Docker benchmarking container addition vs SDXL ControlNet PAG pipeline feature; unrelated." + }, + { + "left": "pull_request:8737", + "right": "pull_request:9070", + "accept": false, + "reason": "Large CI/benchmarking overhaul vs FLUX img2img pipeline draft; different scopes and code paths." + }, + { + "left": "pull_request:8779", + "right": "pull_request:9569", + "accept": false, + "reason": "SD3 img2img test tweak vs CogVideoX autoencoder decode fix; unrelated." + }, + { + "left": "pull_request:8851", + "right": "pull_request:9934", + "accept": false, + "reason": "AuraFlow docs vs Mochi docs; same doc-only category, but different models/features." + }, + { + "left": "pull_request:8990", + "right": "pull_request:9130", + "accept": false, + "reason": "SDXL ControlNet PAG img2img pipeline vs textual inversion README update; unrelated." + }, + { + "left": "pull_request:8990", + "right": "pull_request:9196", + "accept": false, + "reason": "Pipeline feature addition vs PixArt attention-processor chore; not the same underlying change." + }, + { + "left": "pull_request:9083", + "right": "pull_request:9658", + "accept": false, + "reason": "Flux single-file transformer support vs a Flux ControlNet community example; related model family, but different code paths and goals." + }, + { + "left": "pull_request:9083", + "right": "pull_request:9994", + "accept": false, + "reason": "Flux single-file loader support vs Omnigen model/pipeline addition; unrelated." + }, + { + "left": "pull_request:9426", + "right": "pull_request:9938", + "accept": false, + "reason": "CI artifact uploader version bump vs SDXL reference pipeline update; unrelated." + }, + { + "left": "pull_request:7030", + "right": "pull_request:7031", + "accept": true, + "reason": "Same SDXL ControlNet Inpaint pipeline change adding IPAdapterMixin/imports, and both target the same issue." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6534", + "accept": true, + "reason": "Same DreamBooth torch.compile compatibility fix across the SD and LoRA scripts, with the same explicit issue target." + }, + { + "left": "pull_request:7613", + "right": "pull_request:7615", + "accept": true, + "reason": "Same advanced diffusion training model-card standardization work, split only by SD15 vs SDXL script variants." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The analysis is grounded in the packet and stays conservative. The one accepted pair (8607, 8618) is a plausible same-feature docs/code match around TAESD3 shift_factor, and the other rejections do not overclaim duplicate evidence. The summary is also appropriately cautious about the rest of the set being heterogeneous." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -176900,7 +177092,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 237, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -176911,64 +177103,53 @@ "pull_request:10364", "pull_request:10365", "pull_request:10783", + "pull_request:10830", + "pull_request:10888", "pull_request:10889", "pull_request:11395", "pull_request:11426", "pull_request:11427", + "pull_request:11463", "pull_request:11491", "pull_request:11492", "pull_request:11744", "pull_request:11783", + "pull_request:4659", + "pull_request:4661", "pull_request:5065", "pull_request:5066", "pull_request:5299", + "pull_request:5303", + "pull_request:5304", "pull_request:5312", "pull_request:5910", "pull_request:5911", "pull_request:6195", + "pull_request:6328", + "pull_request:6509", "pull_request:6525", "pull_request:6532", "pull_request:6534", "pull_request:6555", "pull_request:6556", "pull_request:6721", - "pull_request:7030", - "pull_request:7031", + "pull_request:6765", + "pull_request:6766", + "pull_request:6767", + "pull_request:6768", + "pull_request:6769", "pull_request:7052", "pull_request:7307", - "pull_request:7613", - "pull_request:7615", "pull_request:7865", "pull_request:7898", - "pull_request:8737", - "pull_request:8779", - "pull_request:8851", - "pull_request:8990", - "pull_request:9070", - "pull_request:9083", - "pull_request:9130", - "pull_request:9196", - "pull_request:9426", - "pull_request:9569", + "pull_request:9579", + "pull_request:9582", "pull_request:9583", - "pull_request:9658", + "pull_request:9606", "pull_request:9724", - "pull_request:9934", - "pull_request:9938", - "pull_request:9994" + "pull_request:9815" ], "soft_pairs": [ - "pull_request:8737|pull_request:9070", - "pull_request:8779|pull_request:9569", - "pull_request:8851|pull_request:9934", - "pull_request:8990|pull_request:9130", - "pull_request:8990|pull_request:9196", - "pull_request:9083|pull_request:9658", - "pull_request:9083|pull_request:9994", - "pull_request:9426|pull_request:9938", - "pull_request:7030|pull_request:7031", - "pull_request:6532|pull_request:6534", - "pull_request:7613|pull_request:7615", "pull_request:5910|pull_request:5911", "pull_request:6195|pull_request:6721", "pull_request:5299|pull_request:5312", @@ -176989,662 +177170,271 @@ "pull_request:7052|pull_request:7307", "pull_request:10783|pull_request:10889", "pull_request:11491|pull_request:11492", - "pull_request:11744|pull_request:11783" + "pull_request:11744|pull_request:11783", + "pull_request:6765|pull_request:6766", + "pull_request:6765|pull_request:6767", + "pull_request:6765|pull_request:6768", + "pull_request:6765|pull_request:6769", + "pull_request:6766|pull_request:6767", + "pull_request:6766|pull_request:6768", + "pull_request:6767|pull_request:6768", + "pull_request:11463|pull_request:11491", + "pull_request:9724|pull_request:9815", + "pull_request:6328|pull_request:6509", + "pull_request:5303|pull_request:5304", + "pull_request:9579|pull_request:9606", + "pull_request:11395|pull_request:11492", + "pull_request:10830|pull_request:10888", + "pull_request:9582|pull_request:9583", + "pull_request:9582|pull_request:9815", + "pull_request:6766|pull_request:6769", + "pull_request:6767|pull_request:6769", + "pull_request:4659|pull_request:4661" ], - "prepared_review_unit_hash": "9fdbe5fed187c1c14097a2587c8430e7d64991e252f7206004242c208cc7fb8f", + "prepared_review_unit_hash": "aa28d71954a969def8cf352fcdac7324709d4d4cb9885b20ffbfbc611e3e08eb", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 32, - "serialized_chars": 104603, - "estimated_input_tokens": 26151, - "estimated_eval_tokens": 52558 + "soft_pair_count": 40, + "serialized_chars": 108355, + "estimated_input_tokens": 27089, + "estimated_eval_tokens": 54434 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is not one monolithic duplicate cluster; it breaks into several independent PR families (compile-compat training scripts, interpolation defaults, IP-Adapter/AnimateDiff pipeline features, docs cleanup, etc.). I accepted only pairs that clearly share the same concrete change or the same code-path fix, and rejected cross-topic or merely same-subsystem pairs.", - "confidence": 0.89, + "summary": "This cluster is heterogeneous: it mixes several unrelated PR families (XLA support, torch.compile compatibility, interpolation defaults, metadata handling, docstring cleanup, and test tweaks). Only a subset are true duplicates or direct follow-ups; many same-issue or same-subsystem PRs are too different in concrete code path to merge safely.", + "confidence": 0.54, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR spans the whole set: the items are split across multiple unrelated feature/fix families, so any one PR would only be a local representative.", + "canonical_pr_reason": "No single canonical PR cleanly covers the whole cluster because the items split into multiple unrelated change families. If forced to pick one representative, PR 5911 is the strongest merged exemplar for the IPAdapterFull feature because it is the completed, user-facing implementation with docs/tests.", "best_issue_reason": null, - "best_pr_reason": "No global best PR exists across the full set. The best representative is only meaningful within each subcluster (for example, the merged, code-complete PRs like 5911, 6721, or 10889 within their own families).", + "best_pr_reason": "PR 5911 is the most complete representative artifact in the set: merged, clearly scoped, and backed by docs/tests. It best fits a canonical representative for one of the true duplicate subclusters.", "soft_edge_verdicts": [ - { - "left": "pull_request:8737", - "right": "pull_request:9070", - "accept": false, - "reason": "Unrelated changes: CI/workflow and docs noise versus a new FLUX img2img pipeline." - }, - { - "left": "pull_request:8779", - "right": "pull_request:9569", - "accept": false, - "reason": "Different code paths and problems: SD3 img2img test tweak versus CogVideoX autoencoder decode fix." - }, - { - "left": "pull_request:8851", - "right": "pull_request:9934", - "accept": false, - "reason": "Different documentation additions for different models/pipelines; not the same underlying change." - }, - { - "left": "pull_request:8990", - "right": "pull_request:9130", - "accept": false, - "reason": "PAG pipeline feature versus textual inversion README example; unrelated artifacts." - }, - { - "left": "pull_request:8990", - "right": "pull_request:9196", - "accept": false, - "reason": "Different changes: pipeline addition versus PixArt attention-processor utility." - }, - { - "left": "pull_request:9083", - "right": "pull_request:9658", - "accept": false, - "reason": "Both are Flux-related, but one is single-file transformer support and the other is a community controlnet script; not the same fix." - }, - { - "left": "pull_request:9083", - "right": "pull_request:9994", - "accept": false, - "reason": "Unrelated model/pipeline work: Flux transformer support versus Omnigen model/pipeline implementation." - }, - { - "left": "pull_request:9426", - "right": "pull_request:9938", - "accept": false, - "reason": "CI artifact uploader update versus SDXL reference-pipeline docs; no shared bug/change." - }, - { - "left": "pull_request:7030", - "right": "pull_request:7031", - "accept": true, - "reason": "Same StableDiffusionXLControlNetInpaint IP-Adapter support change; one is the draft/alternative form of the same code-path fix." - }, - { - "left": "pull_request:6532", - "right": "pull_request:6534", - "accept": true, - "reason": "Same torch.compile compatibility workaround applied to sibling DreamBooth training scripts; plausibly one combined PR." - }, - { - "left": "pull_request:7613", - "right": "pull_request:7615", - "accept": true, - "reason": "Same model-card standardization refactor for advanced diffusion training scripts; same underlying maintenance change." - }, { "left": "pull_request:5910", "right": "pull_request:5911", "accept": true, - "reason": "Same IPAdapterFull support feature; docs/tests and loader changes line up on the same code path." + "reason": "Same IPAdapterFull feature update in the same loader/model paths; 5911 looks like the completed merged version." }, { "left": "pull_request:6195", "right": "pull_request:6721", "accept": true, - "reason": "Same AnimateDiff SDXL feature branch; one is WIP groundwork, the other is the finalized pipeline/docs implementation." + "reason": "Same AnimateDiff SDXL feature work; 6721 is the finished PR for the same pipeline change." }, { "left": "pull_request:5299", "right": "pull_request:5312", "accept": true, - "reason": "Same docs/typehint cleanup in diffusers/models; 5312 is the broader follow-up on the same maintenance change." + "reason": "Both are broad model-doc/typehint cleanups in the same `src/diffusers/models` area and could be combined." }, { "left": "pull_request:11426", "right": "pull_request:11427", "accept": true, - "reason": "Same image-interpolation feature added to paired text-to-image training scripts." + "reason": "Same interpolation-mode feature for closely related text-to-image training scripts; one PR could cover both files." }, { "left": "pull_request:5065", "right": "pull_request:5066", "accept": true, - "reason": "Same attention_processor typo fix across the same concrete code path; one PR is the broader version of the same correction." + "reason": "Same attention-processor typo fix; same underlying correction with examples and core code." }, { "left": "pull_request:7865", "right": "pull_request:7898", "accept": false, - "reason": "Both reduce test model size, but they are separate pipeline tests and not the same concrete fix." + "reason": "Different test targets (Dance Diffusion vs DiT) and no shared concrete bug beyond reducing test size." }, { - "left": "pull_request:6534", - "right": "pull_request:6555", + "left": "pull_request:6555", + "right": "pull_request:6556", "accept": true, - "reason": "Same compiled-module compatibility workaround across training scripts; could be merged into one broader PR." + "reason": "Same torch.compile compatibility fix applied to sibling text-to-image LoRA scripts." }, { - "left": "pull_request:6532", + "left": "pull_request:6534", "right": "pull_request:6555", - "accept": true, - "reason": "Same torch.compile compatibility fix pattern applied to another training script." + "accept": false, + "reason": "Different training scripts and code paths; same issue tag isn't enough." }, { "left": "pull_request:6525", "right": "pull_request:6555", - "accept": true, - "reason": "Same compile-compatibility pattern in training scripts; different file, same underlying fix family." + "accept": false, + "reason": "ControlNet vs DreamBooth are separate scripts and fixes." }, { "left": "pull_request:6532", - "right": "pull_request:6556", - "accept": true, - "reason": "Same compile-compatibility workaround family across sibling training scripts." + "right": "pull_request:6555", + "accept": false, + "reason": "Different script families; too broad to be a single duplicate change." }, { - "left": "pull_request:6525", - "right": "pull_request:6532", - "accept": true, - "reason": "Same torch.compile compatibility issue family; both add the same kind of unwrap/return_dict adjustments." + "left": "pull_request:6532", + "right": "pull_request:6534", + "accept": false, + "reason": "DreamBooth SD and LoRA are distinct implementations, not the same PR." }, { "left": "pull_request:6525", "right": "pull_request:6534", - "accept": true, - "reason": "Same compile-compatibility fix family, just on another training script." + "accept": false, + "reason": "Different training scripts and code paths." }, { "left": "pull_request:11395", "right": "pull_request:11491", "accept": true, - "reason": "Same LANCZOS interpolation-default change applied to different DreamBooth training scripts." + "reason": "Same interpolation default change across DreamBooth-style scripts; same feature and same kind of edit." }, { "left": "pull_request:10364", "right": "pull_request:10365", - "accept": true, - "reason": "Same torch_xla support pattern: add XLA detection plus xm.mark_step() in similar pipeline loops." + "accept": false, + "reason": "Separate XLA support patches for different pipelines, not one concrete code path." }, { "left": "pull_request:9583", "right": "pull_request:9724", "accept": false, - "reason": "Both are docstring refactors, but they touch different utility modules and different fixes." + "reason": "Unrelated docstring refactors in different modules." + }, + { + "left": "pull_request:6534", + "right": "pull_request:6556", + "accept": false, + "reason": "Different script families; same compile-compat theme only." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6556", + "accept": false, + "reason": "Different script families and files." }, { "left": "pull_request:7052", "right": "pull_request:7307", "accept": true, - "reason": "Same stable_diffusion_tensorrt_txt2img update; 7307 looks like the expanded follow-up of the same script change." + "reason": "Same stable_diffusion_tensorrt_txt2img update; 7307 looks like the follow-up/expanded version of 7052." }, { "left": "pull_request:10783", "right": "pull_request:10889", "accept": true, - "reason": "Same fp16 unscaling fix in train_dreambooth_lora_sdxl; 10889 is the cleaned-up final version of the same change." + "reason": "Same fp16 unscaling fix in the same dreambooth_lora_sdxl script; 10889 is the merged final PR." }, { "left": "pull_request:11491", "right": "pull_request:11492", - "accept": true, - "reason": "Same LANCZOS default interpolation feature applied to two DreamBooth variants." + "accept": false, + "reason": "Different training-script families; same interpolation theme but not the same concrete change." }, { "left": "pull_request:11744", "right": "pull_request:11783", - "accept": true, - "reason": "Same lora_alpha/metadata handling feature across Sana and SD3 DreamBooth scripts." + "accept": false, + "reason": "Sana vs SD3 are different scripts/models; metadata support is analogous but not a duplicate." }, - { - "left": "pull_request:6555", - "right": "pull_request:6556", - "accept": true, - "reason": "Same torch.compile compatibility fix family across text-to-image SD and SDXL LoRA scripts." - } - ] - }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, - "error_kind": null, - "error_message": null - }, - { - "label": "pull_request", - "component_index": 1, - "component_count": 266, - "review_unit_index": 238, - "review_unit_count": 273, - "status": "reviewed", - "reason": null, - "source": "llm", - "cache_hit": false, - "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-1-3019", - "nodes": [ - "pull_request:10830", - "pull_request:10888", - "pull_request:11395", - "pull_request:11455", - "pull_request:11463", - "pull_request:11465", - "pull_request:11491", - "pull_request:11492", - "pull_request:4595", - "pull_request:4596", - "pull_request:4659", - "pull_request:4661", - "pull_request:5303", - "pull_request:5304", - "pull_request:5382", - "pull_request:5614", - "pull_request:5725", - "pull_request:5744", - "pull_request:5748", - "pull_request:5906", - "pull_request:5914", - "pull_request:6208", - "pull_request:6241", - "pull_request:6259", - "pull_request:6328", - "pull_request:6509", - "pull_request:6558", - "pull_request:6576", - "pull_request:6654", - "pull_request:6655", - "pull_request:6765", - "pull_request:6766", - "pull_request:6767", - "pull_request:6768", - "pull_request:6769", - "pull_request:7233", - "pull_request:7235", - "pull_request:9579", - "pull_request:9582", - "pull_request:9583", - "pull_request:9591", - "pull_request:9606", - "pull_request:9706", - "pull_request:9707", - "pull_request:9714", - "pull_request:9724", - "pull_request:9815" - ], - "soft_pairs": [ - "pull_request:6765|pull_request:6766", - "pull_request:6765|pull_request:6767", - "pull_request:6765|pull_request:6768", - "pull_request:6765|pull_request:6769", - "pull_request:6766|pull_request:6767", - "pull_request:6766|pull_request:6768", - "pull_request:6767|pull_request:6768", - "pull_request:11463|pull_request:11491", - "pull_request:9724|pull_request:9815", - "pull_request:6328|pull_request:6509", - "pull_request:5303|pull_request:5304", - "pull_request:9579|pull_request:9606", - "pull_request:11395|pull_request:11492", - "pull_request:10830|pull_request:10888", - "pull_request:9582|pull_request:9583", - "pull_request:9582|pull_request:9815", - "pull_request:6766|pull_request:6769", - "pull_request:6767|pull_request:6769", - "pull_request:4659|pull_request:4661", - "pull_request:7233|pull_request:7235", - "pull_request:4595|pull_request:4596", - "pull_request:9707|pull_request:9714", - "pull_request:5725|pull_request:5748", - "pull_request:5906|pull_request:5914", - "pull_request:6208|pull_request:6259", - "pull_request:6241|pull_request:6259", - "pull_request:5725|pull_request:5744", - "pull_request:6208|pull_request:6241", - "pull_request:5382|pull_request:5614", - "pull_request:11455|pull_request:11465", - "pull_request:9591|pull_request:9706", - "pull_request:6558|pull_request:6576", - "pull_request:6654|pull_request:6655" - ], - "prepared_review_unit_hash": "03fe67c2489e009af682a4d71cb107320ec878df6e59f934ceee48ca8278c77e", - "component_budget": { - "node_count": 3019, - "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 - }, - "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 33, - "serialized_chars": 102023, - "estimated_input_tokens": 25506, - "estimated_eval_tokens": 51268 - }, - "overflow_policy": "truncate_then_skip", - "trimmed": false, - "aggressively_trimmed": false, - "split": true, - "analyst_result": { - "summary": "This cluster actually contains several unrelated duplicate families, not one coherent issue: SD3 ControlNet AutoPipeline wiring, interpolation-mode updates across training scripts, docs/template cleanups, and a few isolated runtime/docstring fixes. I only accepted pairs that look like the same concrete change or a direct revision of it.", - "confidence": 0.78, - "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR fits the whole set: the items split into multiple unrelated duplicate families, so one PR would be misleading as a global representative.", - "best_issue_reason": null, - "best_pr_reason": "No single global best PR for the same reason; the strongest matches are per-subcluster, not across the entire item set.", - "soft_edge_verdicts": [ { "left": "pull_request:6765", "right": "pull_request:6766", "accept": false, - "reason": "Same tracking issue, but different utility files/code paths; too broad to be a duplicate PR pair." + "reason": "Repo-wide path cleanup in different modules, not one shared code path." }, { "left": "pull_request:6765", "right": "pull_request:6767", "accept": false, - "reason": "Related cache/path cleanup, but different files and functions; not the same concrete change." + "reason": "Different utility modules; same path-handling style change only." }, { "left": "pull_request:6765", "right": "pull_request:6768", "accept": false, - "reason": "Different code path in a different module; same issue label is not enough." + "reason": "Different utility modules and code paths." }, { "left": "pull_request:6765", "right": "pull_request:6769", "accept": false, - "reason": "Different utility module and implementation path; these are related fixes, not duplicates." + "reason": "Different utility modules." }, { "left": "pull_request:6766", "right": "pull_request:6767", "accept": false, - "reason": "Same broad path-handling theme, but separate files/functions." + "reason": "Different utility modules." }, { "left": "pull_request:6766", "right": "pull_request:6768", "accept": false, - "reason": "Broad same-issue overlap only; the concrete changes are in different modules." + "reason": "Different utility modules." }, { "left": "pull_request:6767", "right": "pull_request:6768", "accept": false, - "reason": "Different files and code paths; too loosely related to merge as duplicates." + "reason": "Different utility modules." }, { "left": "pull_request:11463", "right": "pull_request:11491", "accept": false, - "reason": "Same interpolation feature idea, but different training scripts/code paths; not the same artifact." + "reason": "ControlNet SDXL and DreamBooth Lora Lumina2 are different training scripts." }, { "left": "pull_request:9724", "right": "pull_request:9815", "accept": true, - "reason": "Same training_utils docstring refactor in the same file; 9815 looks like the follow-on/replacement of 9724." + "reason": "Same docstring refactor/fix series in `training_utils.py`; 9815 is the more complete follow-up." }, { "left": "pull_request:6328", "right": "pull_request:6509", "accept": false, - "reason": "Both are AnimateDiff extensions, but video-to-video and image-to-video are different pipelines/features." + "reason": "AnimateDiff video-to-video vs img-to-video are related but separate features." }, { "left": "pull_request:5303", "right": "pull_request:5304", "accept": true, - "reason": "Same DeepFloyd-IF docstring/type cleanup in the same code area; 5304 is a direct extension of the earlier fix." + "reason": "Same DeepFloyd IF docstring/typehint cleanup in the same pipeline file set." }, { "left": "pull_request:9579", "right": "pull_request:9606", "accept": false, - "reason": "Both are docstring cleanup, but in different files and different functions; not the same concrete change." - }, - { - "left": "pull_request:11395", - "right": "pull_request:11492", - "accept": false, - "reason": "Same issue theme, but different scripts; this is a feature propagation, not a duplicate PR." - }, - { - "left": "pull_request:10830", - "right": "pull_request:10888", - "accept": true, - "reason": "Same SD3 ControlNet AutoPipeline export/wiring fix; 10888 is the merged concrete version of the same change." + "reason": "Different modules and unrelated doc fixes." }, { "left": "pull_request:9582", "right": "pull_request:9583", "accept": false, - "reason": "Both are documentation fixes, but they touch different files and unrelated docstrings." + "reason": "Different modules; same docstring cleanup style only." }, { "left": "pull_request:9582", "right": "pull_request:9815", "accept": false, - "reason": "Different files and different docstring targets; same issue link is too broad." - }, - { - "left": "pull_request:6766", - "right": "pull_request:6769", - "accept": false, - "reason": "Different utility modules; same issue target does not make them duplicates." - }, - { - "left": "pull_request:6767", - "right": "pull_request:6769", - "accept": false, - "reason": "Separate file-level path fixes, not one concrete bug/change." + "reason": "Different modules and issue scope." }, { "left": "pull_request:4659", "right": "pull_request:4661", "accept": true, - "reason": "Same SDXL long-weighted-prompt pipeline change; 4661 is the merged replacement for the same feature." - }, - { - "left": "pull_request:7233", - "right": "pull_request:7235", - "accept": true, - "reason": "Same DoRA support addition for the SDXL DreamBooth LoRA script; 7235 adds the matching docs and aligns with the same change." - }, - { - "left": "pull_request:4595", - "right": "pull_request:4596", - "accept": true, - "reason": "Same SDXL converter feature; 4596 is the concrete implementation of the earlier WIP idea." - }, - { - "left": "pull_request:9707", - "right": "pull_request:9714", - "accept": true, - "reason": "Same AuraFlow transformer docstring refactor in the same file; these look like duplicate/revised versions of the same cleanup." - }, - { - "left": "pull_request:5725", - "right": "pull_request:5748", - "accept": true, - "reason": "Both are the same API pipelines docs cleanup series; 5748 appears to be a split follow-up of the broader 5725 change." - }, - { - "left": "pull_request:5906", - "right": "pull_request:5914", - "accept": true, - "reason": "Same small README/model-card template fix in the same training script; 5914 is a direct refinement of 5906." - }, - { - "left": "pull_request:6208", - "right": "pull_request:6241", - "accept": true, - "reason": "Same RuntimeError fix for the Stable Diffusion pipeline dtype mismatch; title and patch intent match closely." - }, - { - "left": "pull_request:5744", - "right": "pull_request:5748", - "accept": false, - "reason": "Related docs cleanup, but these are split into different page batches and do not look like the same exact change." - }, - { - "left": "pull_request:5382", - "right": "pull_request:5614", - "accept": false, - "reason": "The second PR is a revert of the first width/height fix, so they are opposite changes, not duplicates." - }, - { - "left": "pull_request:11455", - "right": "pull_request:11465", - "accept": false, - "reason": "Same interpolation-mode idea, but for different SDXL training scripts; not the same concrete change." - }, - { - "left": "pull_request:11463", - "right": "pull_request:11491", - "accept": false, - "reason": "Same feature family, but different training scripts and code paths." - }, - { - "left": "pull_request:6558", - "right": "pull_request:6576", - "accept": false, - "reason": "Both add torch.compile compatibility, but they target different training scripts (standard vs SDXL)." - }, - { - "left": "pull_request:6654", - "right": "pull_request:6655", - "accept": false, - "reason": "Closely related DreamBooth LoRA inference updates, but for different script variants; not one duplicate bug fix." - }, - { - "left": "pull_request:9591", - "right": "pull_request:9706", - "accept": true, - "reason": "Same docstring refactor in the same AuraFlow transformer file; these are effectively the same cleanup revision." - }, - { - "left": "pull_request:6208", - "right": "pull_request:6259", - "accept": false, - "reason": "Different problems: one fixes a VAE dtype mismatch in the pipeline, the other adds autocast in a training validation loop." - }, - { - "left": "pull_request:6241", - "right": "pull_request:6259", - "accept": false, - "reason": "Unrelated fixes in different code paths; not the same concrete bug." - }, - { - "left": "pull_request:5725", - "right": "pull_request:5744", - "accept": true, - "reason": "Same docs cleanup family; 5744 looks like a split/updated slice of the broader 5725 edit set." - }, - { - "left": "pull_request:6765", - "right": "pull_request:6769", - "accept": false, - "reason": "Different modules and path-handling fixes; same issue target is too broad." - }, - { - "left": "pull_request:9579", - "right": "pull_request:9606", - "accept": false, - "reason": "Different files and different docstring targets; related housekeeping only." - }, - { - "left": "pull_request:9582", - "right": "pull_request:9583", - "accept": false, - "reason": "Different files, different helper docs; not a duplicate pair." - }, - { - "left": "pull_request:6766", - "right": "pull_request:6769", - "accept": false, - "reason": "Separate module-level path fixes, not the same change." - }, - { - "left": "pull_request:6767", - "right": "pull_request:6769", - "accept": false, - "reason": "Same issue target but distinct code paths and files." - }, - { - "left": "pull_request:11455", - "right": "pull_request:11465", - "accept": false, - "reason": "Same interpolation-mode feature, but distinct training scripts and edits." - }, - { - "left": "pull_request:11395", - "right": "pull_request:11492", - "accept": false, - "reason": "Different scripts; these are related ports of the same idea, not duplicates." - }, - { - "left": "pull_request:11463", - "right": "pull_request:11491", - "accept": false, - "reason": "Same feature family but different concrete training scripts." - }, - { - "left": "pull_request:6558", - "right": "pull_request:6576", - "accept": false, - "reason": "Separate script variants, so this is not one concrete fix." - }, - { - "left": "pull_request:6654", - "right": "pull_request:6655", - "accept": false, - "reason": "Same general refactor, but separate standard/SDXL scripts." - }, - { - "left": "pull_request:9591", - "right": "pull_request:9706", - "accept": true, - "reason": "Same file and same docstring cleanup; clearly the same underlying change." - }, - { - "left": "pull_request:6208", - "right": "pull_request:6259", - "accept": false, - "reason": "Different code paths and different failures being addressed." - }, - { - "left": "pull_request:6241", - "right": "pull_request:6259", - "accept": false, - "reason": "Not the same bug or patch scope." - }, - { - "left": "pull_request:5725", - "right": "pull_request:5744", - "accept": true, - "reason": "Same docs cleanup, with 5744 acting as a split/revised subset of the broader PR." - }, - { - "left": "pull_request:5725", - "right": "pull_request:5748", - "accept": true, - "reason": "Same API docs cleanup family; 5748 is another split/revised subset of the broader PR." - }, - { - "left": "pull_request:5744", - "right": "pull_request:5748", - "accept": false, - "reason": "They cover different split batches of the docs cleanup, so they are related but not the same artifact." + "reason": "Same SDXL long weighted prompt pipeline change; 4661 is clearly the replacement/follow-up to 4659." } ] }, @@ -177659,8 +177449,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 239, - "review_unit_count": 273, + "review_unit_index": 238, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -177669,55 +177459,68 @@ "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:11431", + "pull_request:11455", + "pull_request:11465", "pull_request:11471", "pull_request:11472", "pull_request:11491", "pull_request:11511", "pull_request:11512", - "pull_request:11782", - "pull_request:11783", "pull_request:2046", "pull_request:2383", "pull_request:2384", - "pull_request:3201", - "pull_request:3265", "pull_request:3633", "pull_request:3952", - "pull_request:4629", - "pull_request:4659", - "pull_request:4661", + "pull_request:4595", + "pull_request:4596", + "pull_request:5382", "pull_request:5506", + "pull_request:5614", "pull_request:5697", + "pull_request:5725", + "pull_request:5744", + "pull_request:5748", + "pull_request:5906", + "pull_request:5914", + "pull_request:6208", + "pull_request:6241", + "pull_request:6259", "pull_request:6525", "pull_request:6526", "pull_request:6556", + "pull_request:6558", "pull_request:6576", "pull_request:6577", + "pull_request:6654", + "pull_request:6655", "pull_request:7072", "pull_request:7233", "pull_request:7235", - "pull_request:7797", - "pull_request:7803", - "pull_request:7804", - "pull_request:7833", - "pull_request:7846", - "pull_request:7849", - "pull_request:8283", - "pull_request:8335", - "pull_request:8773", - "pull_request:8810", - "pull_request:8864", "pull_request:8916", "pull_request:8919", "pull_request:8932", - "pull_request:8976", "pull_request:8987", + "pull_request:9591", "pull_request:9706", "pull_request:9707", "pull_request:9714", "pull_request:9715" ], "soft_pairs": [ + "pull_request:7233|pull_request:7235", + "pull_request:4595|pull_request:4596", + "pull_request:9707|pull_request:9714", + "pull_request:5725|pull_request:5748", + "pull_request:5906|pull_request:5914", + "pull_request:6208|pull_request:6259", + "pull_request:6241|pull_request:6259", + "pull_request:5725|pull_request:5744", + "pull_request:6208|pull_request:6241", + "pull_request:5382|pull_request:5614", + "pull_request:11455|pull_request:11465", + "pull_request:9591|pull_request:9706", + "pull_request:6558|pull_request:6576", + "pull_request:6654|pull_request:6655", "pull_request:8916|pull_request:8919", "pull_request:11431|pull_request:11512", "pull_request:6526|pull_request:6576", @@ -177741,715 +177544,667 @@ "pull_request:7072|pull_request:7233", "pull_request:11472|pull_request:11491", "pull_request:7072|pull_request:7235", - "pull_request:8932|pull_request:8987", - "pull_request:11782|pull_request:11783", - "pull_request:8810|pull_request:8864", - "pull_request:7797|pull_request:7803", - "pull_request:7797|pull_request:7804", - "pull_request:7797|pull_request:7833", - "pull_request:7797|pull_request:7846", - "pull_request:7803|pull_request:7804", - "pull_request:7803|pull_request:7833", - "pull_request:7803|pull_request:7846", - "pull_request:7804|pull_request:7833", - "pull_request:7804|pull_request:7846", - "pull_request:7833|pull_request:7846", - "pull_request:4629|pull_request:4659", - "pull_request:4629|pull_request:4661", - "pull_request:7797|pull_request:7849", - "pull_request:7803|pull_request:7849", - "pull_request:7804|pull_request:7849", - "pull_request:7833|pull_request:7849", - "pull_request:7846|pull_request:7849", - "pull_request:9706|pull_request:9707", - "pull_request:3201|pull_request:3265", - "pull_request:8773|pull_request:8976", - "pull_request:8283|pull_request:8335" + "pull_request:8932|pull_request:8987" ], - "prepared_review_unit_hash": "0ff9e0fec33fa4bfa837c286313d8313af18a2a76c6b553aa5086ad2772fe1fa", + "prepared_review_unit_hash": "acfce77d04156eab0f7b24f8eeac1e293cceb25ceaa4980e011d821038d3ad5a", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 47, - "serialized_chars": 109967, - "estimated_input_tokens": 27492, - "estimated_eval_tokens": 55240 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 38, + "serialized_chars": 99899, + "estimated_input_tokens": 24975, + "estimated_eval_tokens": 50206 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a mixed cluster of several small PR revision/revert threads, not one single bug or feature. The true duplicates are mostly exact replacements or reversions in the same file/change set; the cross-model/script torch.compile, DoRA, test-size, and interpolation PRs are related only at a broad theme level and should not be merged.", - "confidence": 0.78, + "summary": "This is not one coherent duplicate cluster; it splits into several small near-duplicate groups around the same feature/fix family (DoRA, torch.compile compatibility, interpolation-mode flags, docs sweeps), plus many unrelated PRs. I accepted only pairs that look like the same concrete change or a direct revert; I rejected pairs that only share a broad theme like docs, tests, or training-scripts in the same area.", + "confidence": 0.71, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR cleanly represents the whole heterogeneous cluster; 11783 is the closest canonical representative because it is the latest open revision of an exact duplicate pair and is explicitly issue-linked.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "11783 is the strongest representative PR here: it matches 11782 exactly, is the current revision, and has an explicit issue target. It is still only representative of one subthread, not the entire mixed cluster.", + "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "pull_request:7233", + "right": "pull_request:7235", + "accept": true, + "reason": "Same DreamBooth LoRA SDXL DoRA feature; one is the code change, the other adds the matching docs/script follow-up in the same file family." + }, + { + "left": "pull_request:4595", + "right": "pull_request:4596", + "accept": true, + "reason": "Both are about adding the SDXL conversion script; 4595 looks like the WIP precursor and 4596 is the realized implementation." + }, + { + "left": "pull_request:9707", + "right": "pull_request:9714", + "accept": true, + "reason": "Same docstring refactor in the same AuraFlow transformer file with essentially the same patch." + }, + { + "left": "pull_request:5725", + "right": "pull_request:5748", + "accept": false, + "reason": "Both are docs cleanup sweeps, but they are split into different page/file sets; too broad to treat as the same concrete change." + }, + { + "left": "pull_request:5906", + "right": "pull_request:5914", + "accept": true, + "reason": "Same SDXL LoRA README/template bug in the same script, with 5914 extending/finalizing the same fix." + }, + { + "left": "pull_request:6208", + "right": "pull_request:6259", + "accept": false, + "reason": "Same error text, but different code paths and different files; these are not the same bug fix." + }, + { + "left": "pull_request:6241", + "right": "pull_request:6259", + "accept": false, + "reason": "No concrete diff/file evidence for 6241, and nothing shows it is the same fix as 6259." + }, + { + "left": "pull_request:5725", + "right": "pull_request:5744", + "accept": false, + "reason": "This is still a docs sweep split into chunks; similar topic, but not one identical change set." + }, + { + "left": "pull_request:6208", + "right": "pull_request:6241", + "accept": false, + "reason": "No shared concrete implementation or file-level evidence tying them to the same fix." + }, + { + "left": "pull_request:5382", + "right": "pull_request:5614", + "accept": true, + "reason": "Direct revert/fix of the same SDXL training original-size ordering bug in the same pair of scripts." + }, + { + "left": "pull_request:11455", + "right": "pull_request:11465", + "accept": true, + "reason": "Same interpolation-mode feature in closely related SDXL training scripts; effectively the same change pattern." + }, + { + "left": "pull_request:9591", + "right": "pull_request:9706", + "accept": false, + "reason": "Both are docs-oriented, but they touch different files and different model docs; too general to be a duplicate change." + }, + { + "left": "pull_request:6558", + "right": "pull_request:6576", + "accept": false, + "reason": "Both add torch.compile compatibility, but to different training scripts and code paths; same theme, not the same change." + }, + { + "left": "pull_request:6654", + "right": "pull_request:6655", + "accept": false, + "reason": "Related DreamBooth modularization work, but for different script variants and not one single concrete fix." + }, { "left": "pull_request:8916", "right": "pull_request:8919", "accept": false, - "reason": "Same general torch.compile test pattern, but different model files and different target models (Hunyuan DiT vs Latte), so not the same change." + "reason": "Both add model test coverage, but they are for different models and different files; too broad to merge as duplicates." }, { "left": "pull_request:11431", "right": "pull_request:11512", "accept": false, - "reason": "Both add torch.compile tests, but they cover different transformer models and different test files, so they are not the same concrete fix." + "reason": "Both are torch.compile test additions, but for different transformer models and different test files." }, { "left": "pull_request:6526", "right": "pull_request:6576", "accept": false, - "reason": "Both are torch.compile compatibility updates, but for different training scripts and code paths (ControlNet SDXL vs InstructPix2Pix SDXL)." + "reason": "Same broad compile-compatibility idea, but different pipelines and different code paths." }, { "left": "pull_request:9714", "right": "pull_request:9715", "accept": true, - "reason": "Same source file and same docstring-refactor patch; this looks like a direct replacement/revision of the same documentation change." + "reason": "Same AuraFlow docstring refactor in the same file family; these look like the same change iteration." }, { "left": "pull_request:6526", "right": "pull_request:6577", "accept": false, - "reason": "Different scripts and model-training paths, so this is only a broad torch.compile theme match, not the same underlying change." + "reason": "Both are compile-compat fixes, but they target different training scripts and are not the same code-path problem." }, { "left": "pull_request:6576", "right": "pull_request:6577", "accept": false, - "reason": "Different example scripts; they share a compile-compatibility goal but not the same concrete code-path fix." + "reason": "Different training scripts with a similar unwrap-model pattern; related, but not the same concrete fix." }, { "left": "pull_request:9707", "right": "pull_request:9715", "accept": true, - "reason": "Same file and effectively the same docstring cleanup; 9715 appears to supersede the earlier revision." + "reason": "Same AuraFlow transformer docstring refactor in the same file, with highly overlapping edits." }, { "left": "pull_request:11431", "right": "pull_request:11511", "accept": false, - "reason": "Both are torch.compile test additions, but for different model classes and different test modules." + "reason": "Same testing theme, but different transformer models and different files." }, { "left": "pull_request:5506", "right": "pull_request:5697", "accept": true, - "reason": "Both patch the same attention_processor scale/PEFT/xformers issue in the same file and are effectively overlapping fixes for the same code path." + "reason": "Same attention_processor PEFT/xformers bug fix; 5697 is the cleaner follow-up of the same code-path issue." }, { "left": "pull_request:11471", "right": "pull_request:11491", "accept": false, - "reason": "Same interpolation feature idea, but applied to different training scripts and model families; not the same patch." + "reason": "Same interpolation-mode idea, but applied to different DreamBooth variants and different training scripts." }, { "left": "pull_request:11511", "right": "pull_request:11512", "accept": false, - "reason": "Different model tests in different files; same coverage theme, not the same change." + "reason": "Both are compile-test additions, but they cover different models and are separate test PRs." }, { "left": "pull_request:2383", "right": "pull_request:2384", "accept": true, - "reason": "This is an exact revert pair in the same file, so they are the same underlying change in opposite directions." + "reason": "Exact same Pix2Pix Zero utility-function addition and its revert; clearly the same change." }, { "left": "pull_request:3633", "right": "pull_request:3952", "accept": false, - "reason": "Both pin setup.py to a release candidate, but they target different hub versions and are separate test PRs." + "reason": "Both pin huggingface-hub RCs in setup.py, but they are different version experiments and not the same change." }, { "left": "pull_request:8919", "right": "pull_request:8932", "accept": false, - "reason": "Different transformer model test files; only the test-coverage style is shared." + "reason": "Model test coverage additions for different models; same category, not the same artifact." }, { "left": "pull_request:6526", "right": "pull_request:6556", "accept": false, - "reason": "Different scripts and different model/training code paths, despite both being compile-compatibility work." + "reason": "Both address compile compatibility, but on different scripts and different training flows." }, { "left": "pull_request:6556", "right": "pull_request:6576", "accept": false, - "reason": "Different example scripts and different concrete training loops; not the same fix." + "reason": "Similar compile-compat pattern, but different scripts and not one shared fix." }, { "left": "pull_request:2046", "right": "pull_request:3633", "accept": false, - "reason": "Both touch setup.py, but they are unrelated hub-version test pins from different dates and versions." + "reason": "Different dependency pin experiments in setup.py, separated by time and target version." }, { "left": "pull_request:11471", "right": "pull_request:11472", - "accept": false, - "reason": "Same interpolation option idea, but in different example scripts; too broad to treat as one change." + "accept": true, + "reason": "Same new interpolation-mode option added to closely related advanced DreamBooth training variants." }, { "left": "pull_request:2046", "right": "pull_request:3952", "accept": false, - "reason": "Same broad dependency-pin test theme, but different versions and different PRs." + "reason": "Both are temporary dependency pins, but they target different RC versions and different experiments." }, { "left": "pull_request:6525", "right": "pull_request:6526", "accept": false, - "reason": "Both are torch.compile compatibility updates for ControlNet scripts, but they are separate SD and SDXL code paths, not one patch." + "reason": "Same general compile-compat theme, but separate SD and SDXL training scripts; not the same concrete change." }, { "left": "pull_request:7072", "right": "pull_request:7233", "accept": false, - "reason": "Both add DoRA support, but they modify different scripts and different training flows; not the same concrete change." + "reason": "Both mention DoRA for DreamBooth LoRA, but they are different script variants and not one code patch." }, { "left": "pull_request:11472", "right": "pull_request:11491", "accept": false, - "reason": "Different DreamBooth scripts; same feature theme but not the same patch or code path." - }, - { - "left": "pull_request:7072", - "right": "pull_request:7235", - "accept": false, - "reason": "Related DoRA work, but one is the advanced-script feature and the other is the standard SDXL script/docs update; separate changes." + "reason": "Similar interpolation default change, but across different model-specific training scripts; too broad to merge as a duplicate." }, { - "left": "pull_request:8932", - "right": "pull_request:8987", - "accept": false, - "reason": "Both are new transformer-model tests, but for different models and files, so not duplicates." - }, + "left": "pull_request:9706", + "right": "pull_request:9715", + "accept": true, + "reason": "Same docstring refactor in the same AuraFlow transformer file, with strongly overlapping edits." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 239, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10299", + "pull_request:10316", + "pull_request:11449", + "pull_request:11492", + "pull_request:11782", + "pull_request:11783", + "pull_request:2825", + "pull_request:2826", + "pull_request:3201", + "pull_request:3237", + "pull_request:3265", + "pull_request:3285", + "pull_request:4629", + "pull_request:4659", + "pull_request:4661", + "pull_request:4899", + "pull_request:5938", + "pull_request:6526", + "pull_request:6532", + "pull_request:6534", + "pull_request:6556", + "pull_request:6558", + "pull_request:6559", + "pull_request:6569", + "pull_request:6577", + "pull_request:6654", + "pull_request:6655", + "pull_request:7603", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7833", + "pull_request:7846", + "pull_request:7849", + "pull_request:8283", + "pull_request:8335", + "pull_request:8773", + "pull_request:8810", + "pull_request:8864", + "pull_request:8916", + "pull_request:8932", + "pull_request:8973", + "pull_request:8976", + "pull_request:8985", + "pull_request:9706", + "pull_request:9707", + "pull_request:9714" + ], + "soft_pairs": [ + "pull_request:11782|pull_request:11783", + "pull_request:8810|pull_request:8864", + "pull_request:7797|pull_request:7803", + "pull_request:7797|pull_request:7804", + "pull_request:7797|pull_request:7833", + "pull_request:7797|pull_request:7846", + "pull_request:7803|pull_request:7804", + "pull_request:7803|pull_request:7833", + "pull_request:7803|pull_request:7846", + "pull_request:7804|pull_request:7833", + "pull_request:7804|pull_request:7846", + "pull_request:7833|pull_request:7846", + "pull_request:4629|pull_request:4659", + "pull_request:4629|pull_request:4661", + "pull_request:7797|pull_request:7849", + "pull_request:7803|pull_request:7849", + "pull_request:7804|pull_request:7849", + "pull_request:7833|pull_request:7849", + "pull_request:7846|pull_request:7849", + "pull_request:9706|pull_request:9707", + "pull_request:3201|pull_request:3265", + "pull_request:8773|pull_request:8976", + "pull_request:8283|pull_request:8335", + "pull_request:10299|pull_request:10316", + "pull_request:11449|pull_request:11492", + "pull_request:3237|pull_request:3285", + "pull_request:6559|pull_request:6654", + "pull_request:6559|pull_request:7603", + "pull_request:6559|pull_request:8283", + "pull_request:6654|pull_request:7603", + "pull_request:6654|pull_request:8283", + "pull_request:7603|pull_request:8283", + "pull_request:6526|pull_request:6558", + "pull_request:4899|pull_request:5938", + "pull_request:9706|pull_request:9714", + "pull_request:2825|pull_request:2826", + "pull_request:8916|pull_request:8932", + "pull_request:6526|pull_request:6532", + "pull_request:8973|pull_request:8985", + "pull_request:6556|pull_request:6577", + "pull_request:6569|pull_request:6655", + "pull_request:6569|pull_request:7603", + "pull_request:6526|pull_request:6534", + "pull_request:6532|pull_request:6558" + ], + "prepared_review_unit_hash": "717e10bb0c087ffc4e4b17d1484cb487922992d9ec1665025d9e0b5d374453b2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 44, + "serialized_chars": 108616, + "estimated_input_tokens": 27154, + "estimated_eval_tokens": 54564 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set mostly splits into a few clear duplicate chains: exact/near-exact reruns of the same PR, replacement PRs for the same feature, and a number of similar-but-distinct example/test updates that should stay separate. The strongest duplicate clusters are the SDXL long weighted prompt pipeline, ControlNet LoRA support, disk offload support, and a few same-file doc/test follow-ups.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4661 is the best canonical PR: it is the merged final replacement for the SDXL long weighted prompt pipeline work and most directly represents the finished change in that cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 4661 is the strongest overall representative among these PRs because it is merged, final, and corresponds to a coherent feature rather than a partial draft or a narrow follow-up.", + "soft_edge_verdicts": [ { "left": "pull_request:11782", "right": "pull_request:11783", "accept": true, - "reason": "Same files and same SD3 LoRA metadata change; 11783 is a direct continuation/revision of 11782." + "reason": "Same DreamBooth LoRA SD3 feature and same files; this looks like the same PR/change in duplicate form." }, { "left": "pull_request:8810", "right": "pull_request:8864", "accept": true, - "reason": "Same PAG ControlNet Img2Img feature in the same pipeline area; 8864 reads like a revised/superseding version of 8810 rather than a distinct feature." + "reason": "Same PAG support for SD ControlNet Img2Img, same implementation area and same concrete change." }, { "left": "pull_request:7797", "right": "pull_request:7803", "accept": false, - "reason": "Both shrink fast-test models, but for different pipelines and different test modules; not the same change." + "reason": "Both shrink fast tests, but they target different pipelines (DDPM vs DDIM) and are separate changes." }, { "left": "pull_request:7797", "right": "pull_request:7804", "accept": false, - "reason": "Different pipeline fast tests with separate model-size reductions." + "reason": "Different test suites and different model types; only the general goal is shared." }, { "left": "pull_request:7797", "right": "pull_request:7833", "accept": false, - "reason": "Same general goal of faster tests, but different pipeline test files and model definitions." + "reason": "Same optimization theme, but this is a different pipeline test with a distinct code path." }, { "left": "pull_request:7797", "right": "pull_request:7846", "accept": false, - "reason": "Different audio/image test suites and different dummy-model changes." + "reason": "Different audio pipeline test; not the same underlying bug or change." }, { "left": "pull_request:7803", "right": "pull_request:7804", "accept": false, - "reason": "Different fast tests in different modules; only the size-reduction theme is shared." + "reason": "Different test files/pipelines, so this is not one mergeable change." }, { "left": "pull_request:7803", "right": "pull_request:7833", "accept": false, - "reason": "Different pipeline tests and different model configs." + "reason": "Separate pipeline test reductions, not the same concrete update." }, { "left": "pull_request:7803", "right": "pull_request:7846", "accept": false, - "reason": "Different test files and model shapes; not one change." + "reason": "Different fast test and different model setup." }, { "left": "pull_request:7804", "right": "pull_request:7833", "accept": false, - "reason": "Different fast-test reductions across unrelated pipelines." + "reason": "Only broadly related test-size reductions; they are distinct PRs." }, { "left": "pull_request:7804", "right": "pull_request:7846", "accept": false, - "reason": "Different fast tests, same broad optimization theme only." + "reason": "Different pipelines and test files; not a duplicate change." }, { "left": "pull_request:7833", "right": "pull_request:7846", "accept": false, - "reason": "Separate pipeline test reductions in different files." + "reason": "Separate audio pipeline test reductions, not one concrete fix." }, { "left": "pull_request:4629", "right": "pull_request:4659", "accept": true, - "reason": "4659 is a replacement/rework of the same SDXL long-weighted-prompt pipeline change originally introduced in 4629." + "reason": "Same SDXL long weighted prompt pipeline work; 4659 is a replacement of 4629." }, { "left": "pull_request:4629", "right": "pull_request:4661", "accept": true, - "reason": "4661 is the later replacement PR for the same SDXL long-weighted-prompt pipeline work." + "reason": "Same feature and same affected area; 4661 is the final merged version of the same work." }, { "left": "pull_request:7797", "right": "pull_request:7849", "accept": false, - "reason": "Same fast-test-size reduction theme, but different pipeline test files and dummy models." + "reason": "Different pipelines and test files; only the broad goal of shrinking tests matches." }, { "left": "pull_request:7803", "right": "pull_request:7849", "accept": false, - "reason": "Different test modules; same broad goal only." + "reason": "Different fast test and different model architecture." }, { "left": "pull_request:7804", "right": "pull_request:7849", "accept": false, - "reason": "Different pipeline fast tests, not the same code change." + "reason": "Unrelated test suites, so not the same underlying change." }, { "left": "pull_request:7833", "right": "pull_request:7849", "accept": false, - "reason": "Separate test files and model configs." + "reason": "Different pipeline tests with different model reductions." }, { "left": "pull_request:7846", "right": "pull_request:7849", "accept": false, - "reason": "Different fast-test model reductions; not duplicates." + "reason": "Different test suite and different pipeline." }, { "left": "pull_request:9706", "right": "pull_request:9707", "accept": true, - "reason": "Same file and the same docstring refactor patch, with 9707 looking like the updated revision." + "reason": "Same file and same docstring refactor in AuraFlow transformer docs." }, { "left": "pull_request:3201", "right": "pull_request:3265", "accept": true, - "reason": "These are a direct revert/re-revert pair on the same community pipeline file, i.e. the same underlying change in opposite directions." + "reason": "Direct revert / re-revert of the same community pipeline change in the same file." }, { "left": "pull_request:8773", "right": "pull_request:8976", "accept": true, - "reason": "8976 is the broad revert of the same LoraBaseMixin introduction that 8773 added; same underlying change, reversed." + "reason": "Both are the same LoRA base mixin revert work and touch the same core files and functionality." }, { - "left": "pull_request:9707", - "right": "pull_request:9715", - "accept": true, - "reason": "Same auraflow docstring-refactor change in the same file; 9715 is just another revision of that patch." - } - ] - }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, - "error_kind": null, - "error_message": null - }, - { - "label": "pull_request", - "component_index": 1, - "component_count": 266, - "review_unit_index": 240, - "review_unit_count": 273, - "status": "reviewed", - "reason": null, - "source": "llm", - "cache_hit": false, - "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-1-3019", - "nodes": [ - "pull_request:10299", - "pull_request:10316", - "pull_request:10386", - "pull_request:10387", - "pull_request:11449", - "pull_request:11492", - "pull_request:1346", - "pull_request:1347", - "pull_request:1788", - "pull_request:2094", - "pull_request:2825", - "pull_request:2826", - "pull_request:3237", - "pull_request:3285", - "pull_request:3458", - "pull_request:3460", - "pull_request:4499", - "pull_request:4500", - "pull_request:4501", - "pull_request:4502", - "pull_request:4899", - "pull_request:5833", - "pull_request:5834", - "pull_request:5938", - "pull_request:6207", - "pull_request:6266", - "pull_request:6526", - "pull_request:6532", - "pull_request:6534", - "pull_request:6556", - "pull_request:6558", - "pull_request:6559", - "pull_request:6569", - "pull_request:6577", - "pull_request:6654", - "pull_request:6655", - "pull_request:6940", - "pull_request:6942", - "pull_request:6944", - "pull_request:6947", - "pull_request:7603", - "pull_request:8283", - "pull_request:8916", - "pull_request:8932", - "pull_request:8973", - "pull_request:8985", - "pull_request:9706", - "pull_request:9714" - ], - "soft_pairs": [ - "pull_request:10299|pull_request:10316", - "pull_request:11449|pull_request:11492", - "pull_request:3237|pull_request:3285", - "pull_request:6559|pull_request:6654", - "pull_request:6559|pull_request:7603", - "pull_request:6559|pull_request:8283", - "pull_request:6654|pull_request:7603", - "pull_request:6654|pull_request:8283", - "pull_request:7603|pull_request:8283", - "pull_request:6526|pull_request:6558", - "pull_request:4899|pull_request:5938", - "pull_request:9706|pull_request:9714", - "pull_request:2825|pull_request:2826", - "pull_request:8916|pull_request:8932", - "pull_request:6526|pull_request:6532", - "pull_request:8973|pull_request:8985", - "pull_request:6556|pull_request:6577", - "pull_request:6569|pull_request:6655", - "pull_request:6569|pull_request:7603", - "pull_request:6526|pull_request:6534", - "pull_request:6532|pull_request:6558", - "pull_request:1346|pull_request:1347", - "pull_request:6207|pull_request:6266", - "pull_request:6534|pull_request:6558", - "pull_request:1788|pull_request:2094", - "pull_request:10386|pull_request:10387", - "pull_request:3458|pull_request:3460", - "pull_request:4499|pull_request:4500", - "pull_request:4499|pull_request:4501", - "pull_request:4499|pull_request:4502", - "pull_request:4500|pull_request:4501", - "pull_request:4500|pull_request:4502", - "pull_request:4501|pull_request:4502", - "pull_request:6942|pull_request:6944", - "pull_request:5833|pull_request:5834", - "pull_request:6940|pull_request:6944", - "pull_request:6942|pull_request:6947", - "pull_request:6944|pull_request:6947" - ], - "prepared_review_unit_hash": "1f7f4f3f25d1e35345632d636c83da99793d812a85606e16965174550ae07b5a", - "component_budget": { - "node_count": 3019, - "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 - }, - "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 38, - "serialized_chars": 101473, - "estimated_input_tokens": 25369, - "estimated_eval_tokens": 50994 - }, - "overflow_policy": "truncate_then_skip", - "trimmed": false, - "aggressively_trimmed": false, - "split": true, - "analyst_result": { - "summary": "This set is not one homogeneous duplicate cluster; it contains several exact or near-exact duplicate subgroups plus many same-theme but script-specific PRs that should stay separate. I accepted only the pairs that clearly target the same concrete change in the same file/path (or an obvious one-PR consolidation of identical docs/example fixes).", - "confidence": 0.61, - "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a canonical representative for the whole set because the items split into unrelated subclusters. If forced to anchor one subgroup, PR 4502 is the strongest representative of the repeated SDXL reference-example change because it is merged and the most complete instance.", - "best_issue_reason": null, - "best_pr_reason": "PR 4502 is the best single representative among the clear duplicate subgroups: merged, complete, and central to the exact-duplicate SDXL reference example lineage.", - "soft_edge_verdicts": [ + "left": "pull_request:8283", + "right": "pull_request:8335", + "accept": false, + "reason": "Both modularize validation/inference, but they target different training scripts and code paths." + }, { "left": "pull_request:10299", "right": "pull_request:10316", - "accept": true, - "reason": "Same docs bug pattern: remove generator from from_pretrained usage in pipeline examples; plausibly one docs PR." + "accept": false, + "reason": "Same documentation pattern, but different pipeline docs and separate edits." }, { "left": "pull_request:11449", "right": "pull_request:11492", "accept": false, - "reason": "Same theme, but different training scripts and code paths; not one concrete change." + "reason": "Both set LANCZOS defaults, but for different training scripts; too broad to treat as one duplicate." }, { "left": "pull_request:3237", "right": "pull_request:3285", "accept": true, - "reason": "Same offload feature/fix across the same core files; clearly the same underlying change." + "reason": "Same disk-offload feature in the same core offload codepath; 3285 looks like the focused implementation behind the broader PR." }, { "left": "pull_request:6559", "right": "pull_request:6654", "accept": false, - "reason": "Both are training-example inference refactors, but for different scripts and not the same code path." + "reason": "Different training scripts and separate modularization changes." }, { "left": "pull_request:6559", "right": "pull_request:7603", "accept": false, - "reason": "Different example pipelines/scripts; similar refactor theme only." + "reason": "Different example scripts; same general refactor pattern is not enough." }, { "left": "pull_request:6559", "right": "pull_request:8283", "accept": false, - "reason": "Different training scripts; shared validation/inference theme is too broad." + "reason": "Different pipelines and different training examples." }, { "left": "pull_request:6654", "right": "pull_request:7603", "accept": false, - "reason": "Different scripts and model families; not the same concrete fix." + "reason": "Different examples and code paths despite similar validation refactors." }, { "left": "pull_request:6654", "right": "pull_request:8283", "accept": false, - "reason": "Same broad refactor pattern, but separate code paths." + "reason": "Separate training scripts, not one mergeable fix." }, { "left": "pull_request:7603", "right": "pull_request:8283", "accept": false, - "reason": "Distinct training scripts; not one mergeable bug fix." + "reason": "Different scripts and separate inference workflows." }, { "left": "pull_request:6526", "right": "pull_request:6558", "accept": false, - "reason": "Both add torch.compile compatibility, but for different training scripts and code paths." + "reason": "Different training scripts; compile-compatibility is the theme, but not the same code change." }, { "left": "pull_request:4899", "right": "pull_request:5938", "accept": true, - "reason": "Same ControlNet LoRA feature/fix family with overlapping files and targets; same underlying change." + "reason": "Same ControlNet LoRA support work touching the same LoRA/controlnet internals." }, { "left": "pull_request:9706", "right": "pull_request:9714", "accept": true, - "reason": "Same docstring refactor in the same transformer file." + "reason": "Same docstring refactor in the same file; 9714 is another version of the same change." }, { "left": "pull_request:2825", "right": "pull_request:2826", "accept": true, - "reason": "Same Dreambooth offset-noise training change in the same script." + "reason": "Same DreamBooth offset-noise feature in the same script; 2826 is the completed implementation." }, { "left": "pull_request:8916", "right": "pull_request:8932", "accept": false, - "reason": "Different model test additions; same test-coverage theme only." + "reason": "Different test modules for different model families; only both are coverage additions." }, { "left": "pull_request:6526", "right": "pull_request:6532", "accept": false, - "reason": "Both are torch.compile compatibility patches, but for different training scripts." - }, - { - "left": "pull_request:1346", - "right": "pull_request:1347", - "accept": false, - "reason": "Same README file, but different IMAGIC issues and different fixes." - }, - { - "left": "pull_request:6207", - "right": "pull_request:6266", - "accept": true, - "reason": "Same deprecated-helper removal across closely related LoRA training scripts; same cleanup change." - }, - { - "left": "pull_request:6534", - "right": "pull_request:6558", - "accept": false, - "reason": "Different training scripts; similar compile-compat work, not one concrete fix." - }, - { - "left": "pull_request:1788", - "right": "pull_request:2094", - "accept": true, - "reason": "Same DDIM scheduler bug around prediction_type='sample'; 2094 subsumes the earlier fix." - }, - { - "left": "pull_request:10386", - "right": "pull_request:10387", - "accept": true, - "reason": "Exact same new example file and title; duplicate submissions of the same loading fix." - }, - { - "left": "pull_request:3458", - "right": "pull_request:3460", - "accept": true, - "reason": "Same inpaint LoRA-scale support change; 3460 is the fuller version with tests." - }, - { - "left": "pull_request:4499", - "right": "pull_request:4500", - "accept": true, - "reason": "Same SDXL reference-example addition; duplicate submissions of the same content." - }, - { - "left": "pull_request:4499", - "right": "pull_request:4501", - "accept": true, - "reason": "Same SDXL reference-example addition; identical underlying change." - }, - { - "left": "pull_request:4499", - "right": "pull_request:4502", - "accept": true, - "reason": "Same SDXL reference-example addition; 4502 is the merged final form." - }, - { - "left": "pull_request:4500", - "right": "pull_request:4501", - "accept": true, - "reason": "Exact same example addition and file set." - }, - { - "left": "pull_request:4500", - "right": "pull_request:4502", - "accept": true, - "reason": "Exact same example addition; 4502 is the merged final version." - }, - { - "left": "pull_request:4501", - "right": "pull_request:4502", - "accept": true, - "reason": "Exact same example addition; 4502 is the merged final version." - }, - { - "left": "pull_request:6940", - "right": "pull_request:6942", - "accept": false, - "reason": "Both standardize model cards, but for different training scripts and artifacts." + "reason": "Both are torch.compile compatibility fixes, but for different training scripts and model paths." }, { - "left": "pull_request:5833", - "right": "pull_request:5834", + "left": "pull_request:8973", + "right": "pull_request:8985", "accept": true, - "reason": "Same single-file ControlNet import path bug; 5834 is the direct fix." - }, - { - "left": "pull_request:6940", - "right": "pull_request:6944", - "accept": false, - "reason": "Different model-card scripts; same broad standardization theme only." - }, - { - "left": "pull_request:6942", - "right": "pull_request:6944", - "accept": false, - "reason": "Different scripts and outputs; not the same PR." - }, - { - "left": "pull_request:6942", - "right": "pull_request:6947", - "accept": false, - "reason": "Different scripts; same model-card standardization theme is too broad." - }, - { - "left": "pull_request:6944", - "right": "pull_request:6947", - "accept": false, - "reason": "Different scripts and model cards; not one concrete change." + "reason": "Same Lumina test-size reduction in the same test file; these are effectively the same fast-test change." }, { "left": "pull_request:6556", "right": "pull_request:6577", "accept": false, - "reason": "Both are compile-compat patches, but for different training scripts." + "reason": "Different example scripts and different model families; not one concrete change." }, { "left": "pull_request:6569", "right": "pull_request:6655", "accept": false, - "reason": "Different training scripts; same inferencing-refactor pattern only." + "reason": "Different training scripts and separate inference/validation refactors." }, { "left": "pull_request:6569", "right": "pull_request:7603", "accept": false, - "reason": "Separate example scripts and code paths." + "reason": "Different example pipelines; same refactor pattern is too broad." }, { - "left": "pull_request:6532", - "right": "pull_request:6558", + "left": "pull_request:6526", + "right": "pull_request:6534", "accept": false, - "reason": "Different example scripts; both touch compile-compat, but not the same bug." - }, - { - "left": "pull_request:8973", - "right": "pull_request:8985", - "accept": true, - "reason": "Same Lumina test file and same model-size reduction goal; 8985 is the narrower follow-up of the same change." + "reason": "Different scripts (ControlNet SDXL vs DreamBooth LoRA SD), so not the same underlying fix." }, { - "left": "pull_request:6940", - "right": "pull_request:6944", + "left": "pull_request:6532", + "right": "pull_request:6558", "accept": false, - "reason": "Different training scripts; same standardization theme only." + "reason": "Separate training scripts and separate compile-compatibility changes." } ] }, @@ -178464,50 +178219,53 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 241, - "review_unit_count": 273, + "review_unit_index": 240, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10268", - "pull_request:10288", + "pull_request:10386", + "pull_request:10387", "pull_request:10781", "pull_request:11421", "pull_request:11451", "pull_request:11472", "pull_request:11537", "pull_request:11552", - "pull_request:11744", - "pull_request:11782", "pull_request:12269", - "pull_request:12387", - "pull_request:12439", "pull_request:12454", - "pull_request:2602", - "pull_request:3197", - "pull_request:3201", - "pull_request:3265", - "pull_request:3952", + "pull_request:1346", + "pull_request:1347", + "pull_request:1788", + "pull_request:2094", + "pull_request:3458", + "pull_request:3460", + "pull_request:4499", + "pull_request:4500", + "pull_request:4501", + "pull_request:4502", "pull_request:5299", "pull_request:5312", "pull_request:5391", - "pull_request:6119", - "pull_request:6231", - "pull_request:6526", - "pull_request:6532", - "pull_request:6555", + "pull_request:5833", + "pull_request:5834", + "pull_request:6207", + "pull_request:6266", + "pull_request:6534", + "pull_request:6558", "pull_request:6559", - "pull_request:6569", - "pull_request:6577", "pull_request:6654", "pull_request:6655", + "pull_request:6940", + "pull_request:6942", "pull_request:6944", + "pull_request:6947", "pull_request:7603", - "pull_request:7615", "pull_request:7797", "pull_request:7803", "pull_request:7804", @@ -178516,12 +178274,27 @@ "pull_request:8335", "pull_request:8973", "pull_request:9083", - "pull_request:9257", - "pull_request:9473", "pull_request:9706", "pull_request:9715" ], "soft_pairs": [ + "pull_request:1346|pull_request:1347", + "pull_request:6207|pull_request:6266", + "pull_request:6534|pull_request:6558", + "pull_request:1788|pull_request:2094", + "pull_request:10386|pull_request:10387", + "pull_request:3458|pull_request:3460", + "pull_request:4499|pull_request:4500", + "pull_request:4499|pull_request:4501", + "pull_request:4499|pull_request:4502", + "pull_request:4500|pull_request:4501", + "pull_request:4500|pull_request:4502", + "pull_request:4501|pull_request:4502", + "pull_request:6942|pull_request:6944", + "pull_request:5833|pull_request:5834", + "pull_request:6940|pull_request:6944", + "pull_request:6942|pull_request:6947", + "pull_request:6944|pull_request:6947", "pull_request:7797|pull_request:8973", "pull_request:7803|pull_request:8973", "pull_request:7804|pull_request:8973", @@ -178540,241 +178313,252 @@ "pull_request:6559|pull_request:6655", "pull_request:6655|pull_request:7603", "pull_request:6655|pull_request:8283", - "pull_request:10268|pull_request:10781", - "pull_request:11744|pull_request:11782", - "pull_request:6944|pull_request:7615", - "pull_request:2602|pull_request:3952", - "pull_request:11537|pull_request:11552", - "pull_request:10288|pull_request:9083", - "pull_request:6526|pull_request:6555", - "pull_request:6532|pull_request:6577", - "pull_request:9257|pull_request:9473", - "pull_request:12387|pull_request:12439", - "pull_request:3197|pull_request:3201", - "pull_request:3197|pull_request:3265", - "pull_request:6569|pull_request:8335", - "pull_request:6119|pull_request:6231" + "pull_request:10268|pull_request:10781" ], - "prepared_review_unit_hash": "4a7bc80b07c8bf7d381d6e358d033d1005e480868549938667e30dda89f5f7c3", + "prepared_review_unit_hash": "244b6941e6f3e5e5734fad2abed774de1323ba02d7c2efa21534421b6291be97", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 32, - "serialized_chars": 101571, - "estimated_input_tokens": 25393, - "estimated_eval_tokens": 51042 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 36, + "serialized_chars": 103523, + "estimated_input_tokens": 25881, + "estimated_eval_tokens": 52018 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are superficial matches on a shared theme (tests, docs, training-script cleanup) but not the same underlying change. The only clear duplicate-like/subsumable pairs are the Flux Kontext modular support PRs, the DreamBooth interpolation default updates, the AuraFlow docstring cleanup, and the fp16 LoRA upcast fix.", - "confidence": 0.71, + "summary": "This cluster is really several smaller duplicate subgroups rather than one single bug. I accepted only the pairs that are the same concrete fix/feature on the same code path; most other pairs are just similar maintenance work in different scripts or subsystems.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 12269 is the broadest merged feature PR here and the best representative of the few truly same-change pairs in this set (Flux Kontext modular support).", + "canonical_pr_reason": "PR 4502 is the best canonical representative for the large SDXL Reference subcluster: it is the merged, most complete version of the same exact feature added by the other identical PRs.", "best_issue_reason": null, - "best_pr_reason": "PR 12269 is the strongest standalone representative: merged, broad, and central to one of the accepted duplicate families.", + "best_pr_reason": "PR 4502 is the strongest overall representative here because it is merged, has the most discussion/inbound references, and sits at the center of the largest exact-duplicate group in this cluster.", "soft_edge_verdicts": [ + { + "left": "pull_request:1346", + "right": "pull_request:1347", + "accept": false, + "reason": "Both touch the same README, but they fix different example snippets and different issues; not the same underlying change." + }, + { + "left": "pull_request:6207", + "right": "pull_request:6266", + "accept": false, + "reason": "Both are LoRA-script cleanup, but they remove different deprecated helpers from different training scripts; too broad to merge as one duplicate." + }, + { + "left": "pull_request:6534", + "right": "pull_request:6558", + "accept": false, + "reason": "Same torch.compile theme, but they target different training scripts and code paths, so they are not the same fix." + }, + { + "left": "pull_request:1788", + "right": "pull_request:2094", + "accept": true, + "reason": "Both fix the DDIM scheduler handling for prediction_type='sample' in the same code path." + }, + { + "left": "pull_request:10386", + "right": "pull_request:10387", + "accept": true, + "reason": "Same title, same file, same new example; these are effectively duplicate submissions of the same loading-support change." + }, + { + "left": "pull_request:3458", + "right": "pull_request:3460", + "accept": true, + "reason": "Both implement the same inpaint LoRA scale/cross-attention support in the same pipeline, with 3460 adding the supporting tests/import cleanup." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4500", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4501", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4502", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set; 4502 is the merged final version." + }, + { + "left": "pull_request:4500", + "right": "pull_request:4501", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set." + }, + { + "left": "pull_request:4500", + "right": "pull_request:4502", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set; 4502 is the merged final version." + }, + { + "left": "pull_request:4501", + "right": "pull_request:4502", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set; 4502 is the merged final version." + }, + { + "left": "pull_request:6942", + "right": "pull_request:6944", + "accept": false, + "reason": "Both are model-card standardization work, but they target different training scripts and different model families." + }, + { + "left": "pull_request:5833", + "right": "pull_request:5834", + "accept": false, + "reason": "These are both single-file/import related, but they fix different paths and different problems; not the same concrete bug." + }, + { + "left": "pull_request:6940", + "right": "pull_request:6944", + "accept": false, + "reason": "Both standardize model cards, but one is for plain T2I LoRA and the other is for SDXL LoRA; separate changes." + }, + { + "left": "pull_request:6942", + "right": "pull_request:6947", + "accept": false, + "reason": "Same general model-card cleanup theme, but they apply to different scripts and adapter types, not one shared change." + }, + { + "left": "pull_request:6944", + "right": "pull_request:6947", + "accept": false, + "reason": "Different training scripts and model-card targets; only the broad maintenance area overlaps." + }, { "left": "pull_request:7797", "right": "pull_request:8973", "accept": false, - "reason": "Same class of test-size reduction, but different pipelines and no shared concrete change path." + "reason": "Both reduce fast-test model sizes, but they are for different pipeline tests and unrelated code paths." }, { "left": "pull_request:7803", "right": "pull_request:8973", "accept": false, - "reason": "Different pipeline/test target; this is only a similar optimization pattern." + "reason": "Different test files and models; same broad test-speed theme only." }, { "left": "pull_request:7804", "right": "pull_request:8973", "accept": false, - "reason": "Different test files and models; not the same bug/change." + "reason": "Different test files and models; same broad test-speed theme only." }, { "left": "pull_request:7833", "right": "pull_request:8973", "accept": false, - "reason": "Both shrink fast tests, but they touch unrelated pipelines and are not duplicates." + "reason": "Different test files and models; same broad test-speed theme only." }, { "left": "pull_request:11421", "right": "pull_request:11472", - "accept": true, - "reason": "Both make the same interpolation-mode/default-lanczos change in parallel training scripts; same concrete fix pattern." + "accept": false, + "reason": "Same interpolation default idea, but applied to different training scripts and pipelines; not one concrete change." }, { "left": "pull_request:11451", "right": "pull_request:11552", "accept": false, - "reason": "Both add torch.compile tests, but for different models and code paths." + "reason": "Both add torch.compile tests, but for different model classes and test files." }, { "left": "pull_request:10268", "right": "pull_request:9083", "accept": false, - "reason": "Both add single-file support, but for different model families (Mochi vs Flux) and distinct implementations." + "reason": "Both add single-file support, but for different models (Mochi vs Flux), so these are distinct features." }, { "left": "pull_request:5299", "right": "pull_request:5391", "accept": false, - "reason": "Both are docs/typehint cleanups, but the diffs are unrelated and not the same change." + "reason": "Both improve docs/type hints, but they touch different files and a broad documentation sweep rather than one shared bug." }, { "left": "pull_request:5312", "right": "pull_request:5391", "accept": false, - "reason": "Same broad docs/typehint theme, but not the same concrete edit set." + "reason": "Same broad docs/type-hint cleanup, but not the same concrete change set or code path." }, { "left": "pull_request:12269", "right": "pull_request:12454", "accept": true, - "reason": "12454 looks like follow-up export/plumbing for the same Flux Kontext modular T2I/I2I feature introduced in 12269." + "reason": "Both add the same Flux Kontext modular pipeline support across the same modular-pipeline files." }, { "left": "pull_request:11451", "right": "pull_request:11537", "accept": false, - "reason": "Both concern torch.compile testing, but they target different model classes and different test logic." + "reason": "One adds a Wan compile test and the other targets UNet2DConditionModel; same testing category only." }, { "left": "pull_request:6559", "right": "pull_request:8335", "accept": false, - "reason": "Both modularize training-time inference, but for different training scripts and not the same code path." + "reason": "Both modularize training inference, but for different scripts and generations; not the same underlying fix." }, { "left": "pull_request:6654", "right": "pull_request:8335", "accept": false, - "reason": "Related training-script refactors, but they are different example scripts with different inference flows." + "reason": "Different training script and pipeline family; related refactor pattern only." }, { "left": "pull_request:6655", "right": "pull_request:8335", "accept": false, - "reason": "Same refactor style, but different script families and not one mergeable concrete fix." + "reason": "Different training script and pipeline family; related refactor pattern only." }, { "left": "pull_request:9706", "right": "pull_request:9715", "accept": true, - "reason": "Both touch the same AuraFlow transformer docstrings in the same file and appear to be the same documentation refactor." + "reason": "Same AuraFlow docstring refactor in the same file; this is the same underlying documentation change." }, { "left": "pull_request:6559", "right": "pull_request:6655", "accept": false, - "reason": "Different example scripts; only the high-level refactor theme matches." + "reason": "Both are training-script inference refactors, but they affect different model families and code paths." }, { "left": "pull_request:6655", "right": "pull_request:7603", "accept": false, - "reason": "Both modularize validation/inference, but for different pipelines and no shared concrete bug." + "reason": "Same broad modularization pattern, but different scripts and different inference flows." }, { "left": "pull_request:6655", "right": "pull_request:8283", "accept": false, - "reason": "Same general training-script cleanup theme, but distinct scripts and implementation details." + "reason": "Same broad modularization pattern, but different scripts and different inference flows." }, { "left": "pull_request:10268", "right": "pull_request:10781", "accept": false, - "reason": "Single-file support for different transformer models; not the same change." - }, - { - "left": "pull_request:11744", - "right": "pull_request:11782", - "accept": false, - "reason": "Same metadata-handling idea, but applied to different training scripts and model families." - }, - { - "left": "pull_request:6944", - "right": "pull_request:7615", - "accept": false, - "reason": "Both standardize model-card code, but they are separate example scripts and not one underlying fix." - }, - { - "left": "pull_request:2602", - "right": "pull_request:3952", - "accept": false, - "reason": "Both are dependency pin tests, but they pin different versions and are unrelated temporary changes." - }, - { - "left": "pull_request:11537", - "right": "pull_request:11552", - "accept": false, - "reason": "Both add torch.compile tests, but for different models and no shared failure mode is shown." - }, - { - "left": "pull_request:10288", - "right": "pull_request:9083", - "accept": false, - "reason": "Same single-file loader area, but these add support for different model types and different checkpoints." - }, - { - "left": "pull_request:6526", - "right": "pull_request:6555", - "accept": false, - "reason": "Both address torch.compile compatibility, but in different training scripts and code paths." - }, - { - "left": "pull_request:6532", - "right": "pull_request:6577", - "accept": false, - "reason": "Same compile-compatibility theme, but different example scripts and unwrapping logic targets." - }, - { - "left": "pull_request:9257", - "right": "pull_request:9473", - "accept": false, - "reason": "Both involve Flux DreamBooth scripts, but the fixes target different bugs and different parameter-handling paths." - }, - { - "left": "pull_request:12387", - "right": "pull_request:12439", - "accept": false, - "reason": "Both extend attention-backend support, but they add different backends and are not the same concrete change." - }, - { - "left": "pull_request:3197", - "right": "pull_request:3201", - "accept": false, - "reason": "Revert vs original update; they are opposites, not duplicates." - }, - { - "left": "pull_request:3197", - "right": "pull_request:3265", - "accept": false, - "reason": "This is a revert-of-revert sequence, not the same underlying change as the original update." - }, - { - "left": "pull_request:6569", - "right": "pull_request:8335", - "accept": false, - "reason": "Both modularize validation/inference, but they touch different scripts and training setups." - }, - { - "left": "pull_request:6119", - "right": "pull_request:6231", - "accept": true, - "reason": "Same fp16 LoRA gradient/upcast workaround across closely related training scripts; same concrete bug fix pattern." + "reason": "Both are single-file support work, but for different transformer models and different code changes." } ] }, @@ -178789,15 +178573,16 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 242, - "review_unit_count": 273, + "review_unit_index": 241, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10288", "pull_request:10504", "pull_request:10505", "pull_request:11407", @@ -178805,49 +178590,60 @@ "pull_request:11471", "pull_request:11476", "pull_request:11492", + "pull_request:11537", + "pull_request:11552", + "pull_request:11744", + "pull_request:11782", "pull_request:121", - "pull_request:12142", - "pull_request:12143", - "pull_request:12307", - "pull_request:13115", + "pull_request:12387", + "pull_request:12439", "pull_request:2046", "pull_request:252", "pull_request:2602", - "pull_request:2812", - "pull_request:2813", - "pull_request:326", - "pull_request:327", - "pull_request:3554", - "pull_request:3555", - "pull_request:4022", - "pull_request:4684", - "pull_request:6262", - "pull_request:6263", - "pull_request:6264", + "pull_request:3197", + "pull_request:3201", + "pull_request:3265", + "pull_request:3952", + "pull_request:6119", + "pull_request:6231", + "pull_request:6526", + "pull_request:6532", "pull_request:6550", + "pull_request:6555", "pull_request:6556", "pull_request:6559", "pull_request:6569", + "pull_request:6577", "pull_request:6654", - "pull_request:6908", - "pull_request:6909", - "pull_request:6910", - "pull_request:6939", - "pull_request:6940", - "pull_request:6942", "pull_request:6944", + "pull_request:7615", "pull_request:7797", "pull_request:7803", "pull_request:7804", "pull_request:7833", "pull_request:8283", - "pull_request:8816", - "pull_request:8820", + "pull_request:8335", "pull_request:8985", + "pull_request:9083", + "pull_request:9257", + "pull_request:9473", "pull_request:9592", "pull_request:9707" ], "soft_pairs": [ + "pull_request:11744|pull_request:11782", + "pull_request:6944|pull_request:7615", + "pull_request:2602|pull_request:3952", + "pull_request:11537|pull_request:11552", + "pull_request:10288|pull_request:9083", + "pull_request:6526|pull_request:6555", + "pull_request:6532|pull_request:6577", + "pull_request:9257|pull_request:9473", + "pull_request:12387|pull_request:12439", + "pull_request:3197|pull_request:3201", + "pull_request:3197|pull_request:3265", + "pull_request:6569|pull_request:8335", + "pull_request:6119|pull_request:6231", "pull_request:11471|pull_request:11492", "pull_request:7797|pull_request:8985", "pull_request:7803|pull_request:8985", @@ -178862,238 +178658,198 @@ "pull_request:2046|pull_request:2602", "pull_request:10504|pull_request:10505", "pull_request:11407|pull_request:11476", - "pull_request:121|pull_request:252", - "pull_request:12307|pull_request:13115", - "pull_request:2812|pull_request:2813", - "pull_request:3554|pull_request:3555", - "pull_request:8816|pull_request:8820", - "pull_request:12142|pull_request:12143", - "pull_request:326|pull_request:327", - "pull_request:6908|pull_request:6910", - "pull_request:6909|pull_request:6910", - "pull_request:6939|pull_request:6940", - "pull_request:6262|pull_request:6263", - "pull_request:6262|pull_request:6264", - "pull_request:6263|pull_request:6264", - "pull_request:6908|pull_request:6909", - "pull_request:6939|pull_request:6942", - "pull_request:6939|pull_request:6944", - "pull_request:4022|pull_request:4684" + "pull_request:121|pull_request:252" ], - "prepared_review_unit_hash": "b9150f1f4f979768c882056e02483e18196d7f41cbc4f34a18efdb01dad389ab", + "prepared_review_unit_hash": "fb6af76d9774c6a6c613965551bc7d0d9ffdc0cb7175bc8eb26242dfa5fe7f96", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 31, - "serialized_chars": 112126, - "estimated_input_tokens": 28032, - "estimated_eval_tokens": 56320 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 101050, + "estimated_input_tokens": 25263, + "estimated_eval_tokens": 50782 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a mixed bag of several unrelated duplicate families: exact doc/test cleanups, a few very similar typo/model-card sweeps, and multiple broader refactors that should stay separate. I only accepted pairs that look like the same concrete change or bug fix, not just the same subsystem or style of cleanup.", - "confidence": 0.68, + "summary": "This cluster is mostly a loose collection of similarly-shaped PRs (test cleanups, training-script refactors, docs tweaks) rather than true duplicates. I only accepted the exact or near-exact same-change pairs: the repeated lpw_stable_diffusion revert/update cycle, the identical print-removal test PRs, and the shared fp16 LoRA upcasting fix.", + "confidence": 0.79, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR spans the whole set because it splits into multiple unrelated subclusters. If a representative is required, 10505 is the cleanest exact-duplicate exemplar: same print removals as 10504, merged, and low-risk.", + "canonical_pr_reason": "PR 10505 is the cleanest canonical representative: it is a merged copy of the exact same test print-removal change as 10504, with identical touched files and diff intent.", "best_issue_reason": null, - "best_pr_reason": "10505 is the strongest overall representative only within the exact-duplicate cleanup subcluster: it is merged, minimal, and matches 10504 exactly. For the full cluster, there is no single best PR because the items are not one duplicate.", + "best_pr_reason": "PR 10505 is the strongest overall representative because it is an exact duplicate of 10504, already merged, and the change is unambiguous and self-contained.", "soft_edge_verdicts": [ { - "left": "pull_request:11471", - "right": "pull_request:11492", - "accept": true, - "reason": "Same concrete change: add image interpolation mode support and switch the default to lanczos in closely related DreamBooth LoRA scripts." - }, - { - "left": "pull_request:7797", - "right": "pull_request:8985", + "left": "pull_request:11744", + "right": "pull_request:11782", "accept": false, - "reason": "Both reduce fast-test model sizes, but they are different pipeline tests and not the same underlying bug or change." + "reason": "Both add LoRA metadata handling to DreamBooth-related scripts, but Sana and SD3 are separate model-specific code paths; this is similar work, not the same change." }, { - "left": "pull_request:7803", - "right": "pull_request:8985", + "left": "pull_request:6944", + "right": "pull_request:7615", "accept": false, - "reason": "Same broad test-speedup theme, but this is a different DDIM test and not a duplicate of the Lumina test change." + "reason": "Both standardize model-card handling, but they target different training scripts and different implementations; too broad to treat as duplicates." }, { - "left": "pull_request:7804", - "right": "pull_request:8985", + "left": "pull_request:2602", + "right": "pull_request:3952", "accept": false, - "reason": "Different fast test and different model/code path; size reductions alone are too broad to be duplicates." + "reason": "These are unrelated temporary dependency-pin test PRs for different huggingface-hub RC versions, not the same bug or change." }, { - "left": "pull_request:7833", - "right": "pull_request:8985", + "left": "pull_request:11537", + "right": "pull_request:11552", "accept": false, - "reason": "Another separate fast test size reduction; same motivation, but not the same concrete PR." - }, - { - "left": "pull_request:11421", - "right": "pull_request:11471", - "accept": true, - "reason": "Both apply the same lanczos-default interpolation fix to related DreamBooth training scripts, so they could plausibly be merged together." - }, - { - "left": "pull_request:6550", - "right": "pull_request:6556", - "accept": true, - "reason": "Both address the same torch.compile compatibility issue in sibling SDXL training scripts with very similar code-path changes." + "reason": "Both add torch.compile tests, but for different models and test files; same chore pattern, not the same underlying artifact." }, { - "left": "pull_request:6559", - "right": "pull_request:6569", + "left": "pull_request:10288", + "right": "pull_request:9083", "accept": false, - "reason": "Both are training/inference refactors, but they touch different scripts and different concrete flows." + "reason": "Both add Flux single-file support, but for different model variants and code paths (Canny/Depth/Fill vs Transformer)." }, { - "left": "pull_request:6569", - "right": "pull_request:6654", + "left": "pull_request:6526", + "right": "pull_request:6555", "accept": false, - "reason": "These are separate modularization refactors in different training scripts, not one shared fix." + "reason": "Both make training scripts torch.compile-compatible, but they patch different scripts and different unwrap/save paths; not the same PR." }, { - "left": "pull_request:6569", - "right": "pull_request:8283", + "left": "pull_request:6532", + "right": "pull_request:6577", "accept": false, - "reason": "Same general pattern in different scripts, but not the same underlying code-path problem." + "reason": "Different training scripts and different compile-compatibility fixes; similar subsystem, but not the same change." }, { - "left": "pull_request:9592", - "right": "pull_request:9707", + "left": "pull_request:9257", + "right": "pull_request:9473", "accept": false, - "reason": "Both are docstring cleanup, but they target different transformer files and are not the same change." + "reason": "Both touch Flux DreamBooth code, but one changes prompt/tokenization flow and the other fixes Prodigy learning-rate handling; distinct bugs." }, { - "left": "pull_request:2046", - "right": "pull_request:2602", + "left": "pull_request:12387", + "right": "pull_request:12439", "accept": false, - "reason": "Same dependency pin area, but they pin different release candidates at different times, so they are not duplicates." - }, - { - "left": "pull_request:10504", - "right": "pull_request:10505", - "accept": true, - "reason": "Exact duplicate cleanup: both remove the same stray print statements from the same tests." + "reason": "Both concern attention backends, but they add different backend support and different core changes; too broad to merge as duplicates." }, { - "left": "pull_request:11407", - "right": "pull_request:11476", + "left": "pull_request:3197", + "right": "pull_request:3201", "accept": false, - "reason": "Both are broad typo sweeps, but the touched file sets and edits differ enough that they are not the same concrete change." + "reason": "3201 is a revert of 3197, so the changes are opposite, not duplicates." }, { - "left": "pull_request:121", - "right": "pull_request:252", - "accept": false, - "reason": "Both update README examples, but they are different documentation revisions from different points in time." + "left": "pull_request:3197", + "right": "pull_request:3265", + "accept": true, + "reason": "3265 restores the exact same lpw_stable_diffusion update as 3197; same file and same change set." }, { - "left": "pull_request:12307", - "right": "pull_request:13115", + "left": "pull_request:6569", + "right": "pull_request:8335", "accept": false, - "reason": "These are different modular-pipeline fixes; one adds broader module support while the other is a small Flux-specific tweak." + "reason": "Both add validation/inference modularization, but in different example scripts with different logic; not the same code-path fix." }, { - "left": "pull_request:2812", - "right": "pull_request:2813", + "left": "pull_request:6119", + "right": "pull_request:6231", "accept": true, - "reason": "Identical Stable unCLIP docs update with the same edited content and targets." + "reason": "Both apply the same fp16 LoRA fix pattern: upcast trainable parameters to fp32 to avoid gradient-scaling issues in LoRA training scripts." }, { - "left": "pull_request:3554", - "right": "pull_request:3555", - "accept": true, - "reason": "These are duplicate Moreh/textual-inversion example updates with the same files and content." + "left": "pull_request:11471", + "right": "pull_request:11492", + "accept": false, + "reason": "Same interpolation-default idea, but implemented in different DreamBooth training scripts; related refactor, not a duplicate change." }, { - "left": "pull_request:8816", - "right": "pull_request:8820", - "accept": true, - "reason": "Same PAG controlnet feature work and same code path; the later PR mainly adds the docs hookup on top." + "left": "pull_request:7797", + "right": "pull_request:8985", + "accept": false, + "reason": "These are fast-test size reductions for different pipelines; same optimization theme, but not the same test or bug." }, { - "left": "pull_request:12142", - "right": "pull_request:12143", - "accept": true, - "reason": "Same Wan VACE bf16/fp16 fix in the same pipeline; the second PR just adds related test cleanup." + "left": "pull_request:7803", + "right": "pull_request:8985", + "accept": false, + "reason": "Different pipeline tests and different model-size reductions; not the same concrete change." }, { - "left": "pull_request:326", - "right": "pull_request:327", + "left": "pull_request:7804", + "right": "pull_request:8985", "accept": false, - "reason": "Related type-hint work, but not the same concrete change: one is a broad sweep and the other is a specific PNDM pipeline update." + "reason": "Different amused vs lumina test simplifications; broad test-speedup theme only." }, { - "left": "pull_request:6908", - "right": "pull_request:6910", + "left": "pull_request:7833", + "right": "pull_request:8985", "accept": false, - "reason": "Same model-card standardization theme, but different training scripts and not one shared PR-worthy code-path change." + "reason": "Different audio vs lumina test reductions; not the same underlying artifact." }, { - "left": "pull_request:6909", - "right": "pull_request:6910", + "left": "pull_request:11421", + "right": "pull_request:11471", "accept": false, - "reason": "Similar controlnet model-card cleanup, but this is a separate flax implementation, not a duplicate of the base controlnet PR." + "reason": "Both set LANCZOS defaults, but in different training scripts and contexts; similar refactor, not a duplicate." }, { - "left": "pull_request:6939", - "right": "pull_request:6940", + "left": "pull_request:6550", + "right": "pull_request:6556", "accept": false, - "reason": "Both standardize T2I model cards, but they affect different scripts and are separate refactors." + "reason": "Both address torch.compile compatibility, but for different SD/SDXL training scripts and different code paths." }, { - "left": "pull_request:6262", - "right": "pull_request:6263", + "left": "pull_request:6559", + "right": "pull_request:6569", "accept": false, - "reason": "Different pipeline refactors (panorama vs LDM3D); same general move-out-of-stable_diffusion theme is not enough." + "reason": "Both are training/inference workflow refactors, but they touch different scripts and solve different behavior." }, { - "left": "pull_request:6262", - "right": "pull_request:6264", + "left": "pull_request:6569", + "right": "pull_request:6654", "accept": false, - "reason": "These are distinct pipeline moves for different features, not the same underlying change." + "reason": "Different example scripts and different inference modularization changes; same family, not same PR." }, { - "left": "pull_request:6263", - "right": "pull_request:6264", + "left": "pull_request:9592", + "right": "pull_request:9707", "accept": false, - "reason": "Separate refactors for different pipelines; they should not be merged as duplicates." + "reason": "Both are docstring cleanup PRs, but for unrelated modules and separate changes." }, { - "left": "pull_request:6908", - "right": "pull_request:6909", + "left": "pull_request:2046", + "right": "pull_request:2602", "accept": false, - "reason": "Both standardize controlnet model cards, but they target different backends and are not the same concrete fix." + "reason": "These are different temporary hub version test bumps at different versions, not duplicates." }, { - "left": "pull_request:6939", - "right": "pull_request:6942", - "accept": false, - "reason": "Same model-card standardization family, but different training scripts and different artifacts." + "left": "pull_request:10504", + "right": "pull_request:10505", + "accept": true, + "reason": "These are effectively identical print-removal test PRs across the same files, with no meaningful change differences." }, { - "left": "pull_request:6939", - "right": "pull_request:6944", + "left": "pull_request:11407", + "right": "pull_request:11476", "accept": false, - "reason": "Related documentation cleanup, but this is a separate SDXL LoRA model-card update, not a duplicate." + "reason": "Both are typo-only sweeps, but across different sets of files and unrelated strings/comments." }, { - "left": "pull_request:4022", - "right": "pull_request:4684", + "left": "pull_request:121", + "right": "pull_request:252", "accept": false, - "reason": "Both are Korean translation updates, but they cover different docs sets and are separate localization passes." + "reason": "Both update the README, but the content and scope are different example additions rather than the same change." } ] }, @@ -179108,36 +178864,34 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 243, - "review_unit_count": 273, + "review_unit_index": 242, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10109", - "pull_request:10183", - "pull_request:10268", - "pull_request:10365", "pull_request:1157", "pull_request:1186", "pull_request:11920", - "pull_request:11954", "pull_request:11955", - "pull_request:13069", - "pull_request:13071", - "pull_request:13253", + "pull_request:12142", + "pull_request:12143", + "pull_request:12307", + "pull_request:13115", "pull_request:13254", "pull_request:13315", "pull_request:13335", - "pull_request:13341", - "pull_request:13342", - "pull_request:5104", - "pull_request:5105", - "pull_request:5239", - "pull_request:5242", + "pull_request:2812", + "pull_request:2813", + "pull_request:326", + "pull_request:327", + "pull_request:3554", + "pull_request:3555", + "pull_request:4022", + "pull_request:4684", "pull_request:5828", "pull_request:6261", "pull_request:6262", @@ -179145,13 +178899,13 @@ "pull_request:6264", "pull_request:6265", "pull_request:6267", - "pull_request:6907", "pull_request:6908", "pull_request:6909", "pull_request:6910", "pull_request:6939", "pull_request:6940", "pull_request:6942", + "pull_request:6944", "pull_request:6947", "pull_request:7692", "pull_request:7746", @@ -179160,6 +178914,8 @@ "pull_request:8064", "pull_request:8426", "pull_request:8427", + "pull_request:8816", + "pull_request:8820", "pull_request:8946", "pull_request:9591", "pull_request:9592", @@ -179167,6 +178923,22 @@ "pull_request:9707" ], "soft_pairs": [ + "pull_request:12307|pull_request:13115", + "pull_request:2812|pull_request:2813", + "pull_request:3554|pull_request:3555", + "pull_request:8816|pull_request:8820", + "pull_request:12142|pull_request:12143", + "pull_request:326|pull_request:327", + "pull_request:6908|pull_request:6910", + "pull_request:6909|pull_request:6910", + "pull_request:6939|pull_request:6940", + "pull_request:6262|pull_request:6263", + "pull_request:6262|pull_request:6264", + "pull_request:6263|pull_request:6264", + "pull_request:6908|pull_request:6909", + "pull_request:6939|pull_request:6942", + "pull_request:6939|pull_request:6944", + "pull_request:4022|pull_request:4684", "pull_request:5828|pull_request:7746", "pull_request:6261|pull_request:6262", "pull_request:6261|pull_request:6263", @@ -179188,336 +178960,264 @@ "pull_request:1157|pull_request:1186", "pull_request:11920|pull_request:11955", "pull_request:13254|pull_request:13315", - "pull_request:13254|pull_request:13335", - "pull_request:13254|pull_request:13341", - "pull_request:13254|pull_request:13342", - "pull_request:13315|pull_request:13335", - "pull_request:13315|pull_request:13341", - "pull_request:13315|pull_request:13342", - "pull_request:13335|pull_request:13341", - "pull_request:13335|pull_request:13342", - "pull_request:13341|pull_request:13342", - "pull_request:5104|pull_request:5105", - "pull_request:5239|pull_request:5242", - "pull_request:7903|pull_request:8064", - "pull_request:13069|pull_request:13071", - "pull_request:6907|pull_request:6908", - "pull_request:6907|pull_request:6909", - "pull_request:6907|pull_request:6910", - "pull_request:10109|pull_request:10365", - "pull_request:10183|pull_request:10268", - "pull_request:11920|pull_request:11954", - "pull_request:11954|pull_request:11955", - "pull_request:13253|pull_request:13254", - "pull_request:13253|pull_request:13315", - "pull_request:13253|pull_request:13335", - "pull_request:13253|pull_request:13341", - "pull_request:13253|pull_request:13342" + "pull_request:13254|pull_request:13335" ], - "prepared_review_unit_hash": "b6f1de791b8d8647cb77f10535f9ac104315ee955436edcf1a8937e4c8eb0675", + "prepared_review_unit_hash": "155add5443c79604821105006f29fa128234062fd220ba1907f7aa310ba0b4aa", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 46, - "serialized_chars": 106285, - "estimated_input_tokens": 26572, - "estimated_eval_tokens": 53400 + "soft_pair_count": 38, + "serialized_chars": 110612, + "estimated_input_tokens": 27653, + "estimated_eval_tokens": 55562 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous: it mixes unrelated PRs across test fixes, pipeline refactors, docs updates, and workflow tweaks. Only two soft-edge pairs look like true duplicates/near-duplicates: the ConsistencyDecoderVAE test fixes and the PAG community pipeline rework.", - "confidence": 0.94, + "summary": "This cluster actually contains several unrelated subtopics (modular pipeline refactors, model-card standardization, translation/docs updates, test slices, etc.). I only accepted soft edges where the diffs look like the same concrete change or a clear rework of the same file path; most lookalike PRs are distinct follow-up changes and should stay separate.", + "confidence": 0.84, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a good canonical for the whole cluster because the items split into several unrelated subclusters rather than one duplicate change.", + "canonical_pr_reason": "No single canonical PR spans the whole set; the items split into multiple unrelated duplicate subclusters. If forced to anchor on one, the PAG controlnet PR pair is the clearest true duplicate cluster.", "best_issue_reason": null, - "best_pr_reason": "No single PR cleanly represents the cluster; the only strong duplicate signals are isolated to small subclusters, so a global representative would be misleading.", + "best_pr_reason": "8820 is the strongest single PR anchor among the true duplicate pairs: it is the more complete, merged version of the PAG controlnet change, including code, docs, and tests.", "soft_edge_verdicts": [ { - "left": "pull_request:5828", - "right": "pull_request:7746", - "accept": true, - "reason": "Both are fixes for the ConsistencyDecoderVAE test path, so they look like the same underlying test issue." - }, - { - "left": "pull_request:6261", - "right": "pull_request:6262", - "accept": false, - "reason": "Same broad stable_diffusion refactor theme, but they move different pipelines/components and are not the same change." - }, - { - "left": "pull_request:6261", - "right": "pull_request:6263", - "accept": false, - "reason": "Different pipeline move: Attend-and-Excite vs LDM3D, so not the same concrete change." - }, - { - "left": "pull_request:6261", - "right": "pull_request:6264", - "accept": false, - "reason": "Different pipelines (Attend-and-Excite vs SAG); related refactor style but not duplicate work." - }, - { - "left": "pull_request:6262", - "right": "pull_request:6267", - "accept": false, - "reason": "Panorama and K-Diffusion are separate pipeline moves, not one underlying change." - }, - { - "left": "pull_request:6263", - "right": "pull_request:6267", - "accept": false, - "reason": "LDM3D and K-Diffusion are distinct pipeline refactors." - }, - { - "left": "pull_request:6264", - "right": "pull_request:6267", - "accept": false, - "reason": "SAG and K-Diffusion are separate pipeline moves, so reject." - }, - { - "left": "pull_request:7903", - "right": "pull_request:8946", + "left": "pull_request:12307", + "right": "pull_request:13115", "accept": false, - "reason": "Unrelated test changes: video/image processor validation vs a request-count assertion in model loading tests." + "reason": "Related modular-pipeline cleanup, but not the same concrete change; 13115 is a follow-up export tweak, not a duplicate of the larger 12307 fix." }, { - "left": "pull_request:8064", - "right": "pull_request:8946", - "accept": false, - "reason": "Marigold test work and model-loading request-count test tweaks are unrelated." + "left": "pull_request:2812", + "right": "pull_request:2813", + "accept": true, + "reason": "Same doc update to the same files; the diffs are effectively identical." }, { - "left": "pull_request:8426", - "right": "pull_request:8427", - "accept": false, - "reason": "Same workflow file, but one changes dependency install and the other adds env printing; different fixes." + "left": "pull_request:3554", + "right": "pull_request:3555", + "accept": true, + "reason": "Same generated example/script bundle with the same files and content; clearly duplicated." }, { - "left": "pull_request:6940", - "right": "pull_request:6942", - "accept": false, - "reason": "Both standardize model cards, but for different example scripts (LoRA vs SDXL), so not the same change." + "left": "pull_request:8816", + "right": "pull_request:8820", + "accept": true, + "reason": "Same PAG controlnet change on the same code path; 8820 is the fuller follow-up with docs/tests." }, { - "left": "pull_request:7692", - "right": "pull_request:7876", + "left": "pull_request:12142", + "right": "pull_request:12143", "accept": true, - "reason": "These are the same PAG community pipeline modification/rework, with matching file and diff intent." + "reason": "Same bf16/fp16 fix in `pipeline_wan_vace.py` on the same line; 12143 just packages it with extra test changes." }, { - "left": "pull_request:9592", - "right": "pull_request:9593", - "accept": false, - "reason": "Docstring refactors in different files and different components; only same broad docs style." + "left": "pull_request:326", + "right": "pull_request:327", + "accept": true, + "reason": "Same PNDM typing change in the same pipeline file; 327 is a narrower rework of the same edit." }, { - "left": "pull_request:6261", - "right": "pull_request:6267", + "left": "pull_request:6908", + "right": "pull_request:6910", "accept": false, - "reason": "Different pipeline migrations out of stable_diffusion; same area but not the same fix." + "reason": "Both are model-card standardizations, but they touch different training scripts and are not the same underlying change." }, { - "left": "pull_request:6265", - "right": "pull_request:6267", + "left": "pull_request:6909", + "right": "pull_request:6910", "accept": false, - "reason": "GLIGEN and K-Diffusion are separate pipeline extractions." + "reason": "Different ControlNet training scripts; similar cleanup pattern, but not a duplicate change." }, { "left": "pull_request:6939", - "right": "pull_request:6947", - "accept": false, - "reason": "Both are model-card standardizations, but for different training scripts and model families." - }, - { - "left": "pull_request:6940", - "right": "pull_request:6947", + "right": "pull_request:6940", "accept": false, - "reason": "LoRA model-card updates and T2I Adapter SDXL model-card updates are different concrete changes." + "reason": "Same style of model-card refactor, but applied to different scripts and different outputs." }, { - "left": "pull_request:9591", - "right": "pull_request:9707", + "left": "pull_request:6262", + "right": "pull_request:6263", "accept": false, - "reason": "Unrelated docstring refactors in different modules." + "reason": "Both move different Stable Diffusion variants out into separate packages; same refactor pattern, but distinct pipelines." }, { - "left": "pull_request:1157", - "right": "pull_request:1186", + "left": "pull_request:6262", + "right": "pull_request:6264", "accept": false, - "reason": "Different Flax training scripts with separate argument/reporting fixes; not the same bug." + "reason": "Different pipeline split-outs (Panorama vs SAG), so not the same change." }, { - "left": "pull_request:11920", - "right": "pull_request:11955", + "left": "pull_request:6263", + "right": "pull_request:6264", "accept": false, - "reason": "Different video test suites (Wan vs Cosmos), so not duplicate work." + "reason": "Different pipeline split-outs with separate targets; related refactor family, not duplicates." }, { - "left": "pull_request:13254", - "right": "pull_request:13315", + "left": "pull_request:6908", + "right": "pull_request:6909", "accept": false, - "reason": "Different model test refactors for unrelated transformer families." + "reason": "Same standardization theme, but different training scripts and file targets." }, { - "left": "pull_request:13254", - "right": "pull_request:13335", + "left": "pull_request:6939", + "right": "pull_request:6942", "accept": false, - "reason": "LTX and Cosmos test refactors are unrelated." + "reason": "Both standardize model cards, but for different scripts (T2I vs SDXL) and different metadata." }, { - "left": "pull_request:13254", - "right": "pull_request:13341", + "left": "pull_request:6939", + "right": "pull_request:6944", "accept": false, - "reason": "LTX and Bria test refactors are separate changes." + "reason": "Different LoRA vs base training scripts; similar cleanup, not the same PR." }, { - "left": "pull_request:13254", - "right": "pull_request:13342", + "left": "pull_request:4022", + "right": "pull_request:4684", "accept": false, - "reason": "LTX and Hunyuan transformer test refactors are unrelated." + "reason": "Both are Korean doc updates, but they are separate translation passes with different content and pages." }, { - "left": "pull_request:13315", - "right": "pull_request:13335", + "left": "pull_request:5828", + "right": "pull_request:7746", "accept": false, - "reason": "Sana and Cosmos test refactors are different model-specific changes." + "reason": "Different tests and different bugs; one is a no-grad annotation, the other fixes VAE tiling dtype." }, { - "left": "pull_request:13315", - "right": "pull_request:13341", + "left": "pull_request:6261", + "right": "pull_request:6262", "accept": false, - "reason": "Sana and Bria refactors are unrelated." + "reason": "Same refactor family, but different pipelines and different code paths." }, { - "left": "pull_request:13315", - "right": "pull_request:13342", + "left": "pull_request:6261", + "right": "pull_request:6263", "accept": false, - "reason": "Sana and Hunyuan refactors are separate model test changes." + "reason": "Different moved pipeline modules; not duplicate changes." }, { - "left": "pull_request:13335", - "right": "pull_request:13341", + "left": "pull_request:6261", + "right": "pull_request:6264", "accept": false, - "reason": "Cosmos and Bria test refactors are different model families." + "reason": "Different moved pipeline modules; same broad refactor, not the same PR." }, { - "left": "pull_request:13335", - "right": "pull_request:13342", + "left": "pull_request:6262", + "right": "pull_request:6267", "accept": false, - "reason": "Cosmos and Hunyuan test refactors are unrelated." + "reason": "Both relocate pipelines, but they target different variants and files." }, { - "left": "pull_request:13341", - "right": "pull_request:13342", + "left": "pull_request:6263", + "right": "pull_request:6267", "accept": false, - "reason": "Bria and Hunyuan test refactors are separate model-specific changes." + "reason": "Different pipeline refactors; similar packaging work, not duplicates." }, { - "left": "pull_request:5104", - "right": "pull_request:5105", + "left": "pull_request:6264", + "right": "pull_request:6267", "accept": false, - "reason": "Same timestep-bias idea, but applied to different training scripts (SDXL LoRA vs Dreambooth) and not one mergeable PR." + "reason": "Different moved pipeline targets; not the same underlying change." }, { - "left": "pull_request:5239", - "right": "pull_request:5242", + "left": "pull_request:7903", + "right": "pull_request:8946", "accept": false, - "reason": "Different AutoencoderKL methods: encode docstring vs decode docstring; related but distinct." + "reason": "Unrelated test changes in different areas; no shared concrete bug/change." }, { - "left": "pull_request:7903", - "right": "pull_request:8064", + "left": "pull_request:8064", + "right": "pull_request:8946", "accept": false, - "reason": "Video processor/input validation work and Marigold tests are unrelated." + "reason": "Different model/test areas and different intent; not duplicates." }, { - "left": "pull_request:11920", - "right": "pull_request:11954", + "left": "pull_request:8426", + "right": "pull_request:8427", "accept": false, - "reason": "Wan and Hunyuan video test slices are different pipeline test suites." + "reason": "Both touch the same workflow file, but one changes install steps while the other adds debug output; separate fixes." }, { - "left": "pull_request:11954", - "right": "pull_request:11955", + "left": "pull_request:6940", + "right": "pull_request:6942", "accept": false, - "reason": "Hunyuan video and Cosmos test slices are separate model families." + "reason": "Similar model-card standardization, but different scripts and different training contexts." }, { - "left": "pull_request:13253", - "right": "pull_request:13254", - "accept": false, - "reason": "Z Image and LTX transformer test refactors are unrelated." + "left": "pull_request:7692", + "right": "pull_request:7876", + "accept": true, + "reason": "Same community PAG pipeline behavior change; 7876 is a reworked version of the same fix." }, { - "left": "pull_request:13253", - "right": "pull_request:13315", + "left": "pull_request:9592", + "right": "pull_request:9593", "accept": false, - "reason": "Z Image and Sana test refactors are different model-specific changes." + "reason": "Docstring refactors in different files; same style, different code." }, { - "left": "pull_request:13253", - "right": "pull_request:13335", + "left": "pull_request:6261", + "right": "pull_request:6267", "accept": false, - "reason": "Z Image and Cosmos test refactors are unrelated." + "reason": "Both are modular pipeline refactors, but they move different pipelines and add different exports." }, { - "left": "pull_request:13253", - "right": "pull_request:13341", + "left": "pull_request:6265", + "right": "pull_request:6267", "accept": false, - "reason": "Z Image and Bria test refactors are different model families." + "reason": "Different stable-diffusion subpipeline moves; related but not duplicate." }, { - "left": "pull_request:13253", - "right": "pull_request:13342", + "left": "pull_request:6939", + "right": "pull_request:6947", "accept": false, - "reason": "Z Image and Hunyuan test refactors are unrelated." + "reason": "Both standardize model cards, but for different pipelines and different code paths." }, { - "left": "pull_request:6907", - "right": "pull_request:6908", + "left": "pull_request:6940", + "right": "pull_request:6947", "accept": false, - "reason": "ControlNet model-card standardization for different variants/scripts; same pattern, not the same change." + "reason": "Different training scripts and different model-card metadata; not the same change." }, { - "left": "pull_request:6907", - "right": "pull_request:6909", + "left": "pull_request:9591", + "right": "pull_request:9707", "accept": false, - "reason": "ControlNet and ControlNet Flax model-card updates are separate scripts and PRs." + "reason": "Different docstring cleanups in different models; no duplicate bug/change." }, { - "left": "pull_request:6907", - "right": "pull_request:6910", + "left": "pull_request:1157", + "right": "pull_request:1186", "accept": false, - "reason": "ControlNet SDXL model-card update is distinct from the base ControlNet change." + "reason": "Similar Flax training-script fixes, but they target different examples and separate code paths." }, { - "left": "pull_request:10109", - "right": "pull_request:10365", + "left": "pull_request:11920", + "right": "pull_request:11955", "accept": false, - "reason": "Both add torch_xla support, but to different pipelines and code paths." + "reason": "Different model test-slice additions for different pipelines; same pattern, not duplicates." }, { - "left": "pull_request:10183", - "right": "pull_request:10268", + "left": "pull_request:13254", + "right": "pull_request:13315", "accept": false, - "reason": "Single-file support for AutoencoderDC and Mochi Transformer are different features." + "reason": "Different transformer test refactors for different models; not the same underlying change." }, { - "left": "pull_request:6940", - "right": "pull_request:6947", + "left": "pull_request:13254", + "right": "pull_request:13335", "accept": false, - "reason": "These target different training scripts and model-card variants, so they are not the same concrete change." + "reason": "Different transformer test refactors for different models; related CI cleanup, not duplicates." } ] }, @@ -179532,12 +179232,12 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 244, - "review_unit_count": 273, + "review_unit_index": 243, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -179545,39 +179245,39 @@ "pull_request:10183", "pull_request:10268", "pull_request:10364", + "pull_request:10365", "pull_request:10402", "pull_request:10434", - "pull_request:10947", "pull_request:10991", "pull_request:11920", - "pull_request:11953", + "pull_request:11954", "pull_request:11955", - "pull_request:1291", + "pull_request:13069", + "pull_request:13071", "pull_request:13253", "pull_request:13254", "pull_request:13315", "pull_request:13335", "pull_request:13341", - "pull_request:13345", + "pull_request:13342", "pull_request:13346", "pull_request:1513", "pull_request:1564", - "pull_request:2486", - "pull_request:2487", - "pull_request:2488", "pull_request:2812", "pull_request:2813", "pull_request:2815", - "pull_request:309", "pull_request:3568", "pull_request:3577", - "pull_request:555", + "pull_request:5104", + "pull_request:5105", + "pull_request:5239", + "pull_request:5242", "pull_request:5833", "pull_request:5834", - "pull_request:5871", - "pull_request:5979", - "pull_request:6133", + "pull_request:6907", "pull_request:6908", + "pull_request:6909", + "pull_request:6910", "pull_request:6942", "pull_request:6944", "pull_request:6947", @@ -179585,11 +179285,36 @@ "pull_request:7615", "pull_request:7903", "pull_request:7934", + "pull_request:8064", "pull_request:9083", "pull_request:9591", "pull_request:9715" ], "soft_pairs": [ + "pull_request:13254|pull_request:13341", + "pull_request:13254|pull_request:13342", + "pull_request:13315|pull_request:13335", + "pull_request:13315|pull_request:13341", + "pull_request:13315|pull_request:13342", + "pull_request:13335|pull_request:13341", + "pull_request:13335|pull_request:13342", + "pull_request:13341|pull_request:13342", + "pull_request:5104|pull_request:5105", + "pull_request:5239|pull_request:5242", + "pull_request:7903|pull_request:8064", + "pull_request:13069|pull_request:13071", + "pull_request:6907|pull_request:6908", + "pull_request:6907|pull_request:6909", + "pull_request:6907|pull_request:6910", + "pull_request:10109|pull_request:10365", + "pull_request:10183|pull_request:10268", + "pull_request:11920|pull_request:11954", + "pull_request:11954|pull_request:11955", + "pull_request:13253|pull_request:13254", + "pull_request:13253|pull_request:13315", + "pull_request:13253|pull_request:13335", + "pull_request:13253|pull_request:13341", + "pull_request:13253|pull_request:13342", "pull_request:13254|pull_request:13346", "pull_request:13315|pull_request:13346", "pull_request:13335|pull_request:13346", @@ -179612,265 +179337,276 @@ "pull_request:5834|pull_request:7934", "pull_request:10109|pull_request:10364", "pull_request:10183|pull_request:10991", - "pull_request:10183|pull_request:9083", - "pull_request:10268|pull_request:10947", - "pull_request:11920|pull_request:11953", - "pull_request:11953|pull_request:11955", - "pull_request:1291|pull_request:309", - "pull_request:13253|pull_request:13346", - "pull_request:13254|pull_request:13345", - "pull_request:13315|pull_request:13345", - "pull_request:13335|pull_request:13345", - "pull_request:13341|pull_request:13345", - "pull_request:2486|pull_request:2488", - "pull_request:2487|pull_request:555", - "pull_request:5871|pull_request:7903", - "pull_request:5979|pull_request:6133" + "pull_request:10183|pull_request:9083" ], - "prepared_review_unit_hash": "9ae59dc50828abfd20880a56eeb6c7bfa432a59f3c6866629f83583b832b2a15", + "prepared_review_unit_hash": "dcabbfaa1e574170cc1957a5995e1b933a2d4c6974ca4fd3a1d715113e90da69", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 36, - "serialized_chars": 103656, - "estimated_input_tokens": 25914, - "estimated_eval_tokens": 52084 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 47, + "serialized_chars": 107733, + "estimated_input_tokens": 26934, + "estimated_eval_tokens": 54124 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: most items are unrelated docs, model-card, single-file, or test-refactor PRs that only look similar at a high level. I only accepted a few true near-duplicates where the concrete doc fix is the same or clearly subsumed by the later PR.", + "summary": "Mostly a heterogeneous maintenance cluster: many PRs are similar in style but apply to different models, pipelines, or example scripts. The only clear duplicate-looking subgroup is the docs link cleanup, especially the Stable UnCLIP and Hunyuan docs fixes.", "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "10434 is the strongest cluster representative: it is issue-linked, broad, and covers the same docs-link cleanup theme across many pages, making it the best canonical PR in this mixed set.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "10434 has the clearest scope and broadest coverage among the merged PRs here, so it is the best overall representative even though the cluster contains several unrelated subfamilies.", "soft_edge_verdicts": [ { "left": "pull_request:13254", - "right": "pull_request:13346", + "right": "pull_request:13341", "accept": false, - "reason": "Different transformer test refactors for different model families; same broad test-cleanup pattern, not the same change." + "reason": "Same refactor style, but different model test files (LTX vs Bria); not the same change." }, { - "left": "pull_request:13315", - "right": "pull_request:13346", + "left": "pull_request:13254", + "right": "pull_request:13342", "accept": false, - "reason": "Different Sana vs CogVideoX/CogView transformer test refactors; not the same underlying bug/change." + "reason": "Different model families and files; both are test refactors, not duplicates." }, { - "left": "pull_request:13335", - "right": "pull_request:13346", + "left": "pull_request:13315", + "right": "pull_request:13335", "accept": false, - "reason": "Different Cosmos vs CogVideoX/CogView test refactors; only similar CI cleanup theme." + "reason": "Different models (Sana vs Cosmos); similar refactor pattern only." }, { - "left": "pull_request:13341", - "right": "pull_request:13346", + "left": "pull_request:13315", + "right": "pull_request:13341", "accept": false, - "reason": "Different Bria/Chroma vs CogVideoX/CogView test refactors; not mergeable as one concrete change." + "reason": "Different test targets and code paths; not the same underlying change." }, { - "left": "pull_request:2812", - "right": "pull_request:2815", - "accept": true, - "reason": "Same Stable unCLIP docs update around broken links/updated checkpoints; later PR is a clean variant of the same doc fix." + "left": "pull_request:13315", + "right": "pull_request:13342", + "accept": false, + "reason": "Different models/pipelines; only the CI refactor pattern matches." }, { - "left": "pull_request:2813", - "right": "pull_request:2815", - "accept": true, - "reason": "Same Stable unCLIP docs fix in the same file and same topic; clearly the same underlying documentation change." + "left": "pull_request:13335", + "right": "pull_request:13341", + "accept": false, + "reason": "Different model test files; same refactor family, not a duplicate change." }, { - "left": "pull_request:3568", - "right": "pull_request:3577", + "left": "pull_request:13335", + "right": "pull_request:13342", "accept": false, - "reason": "Both fix broken links, but in different docs pages with different exact edits; too generic to be duplicates." + "reason": "Different model targets and test code; not the same PR." }, { - "left": "pull_request:6908", - "right": "pull_request:6942", + "left": "pull_request:13341", + "right": "pull_request:13342", "accept": false, - "reason": "Both standardize model cards, but for different training scripts and model types; same pattern, different change." + "reason": "Different model families and files; too broad to treat as the same change." }, { - "left": "pull_request:10402", - "right": "pull_request:10434", - "accept": true, - "reason": "Same broken-link documentation issue, including the same hunyuan docs files; 10434 is a broader superset of the same fix." + "left": "pull_request:5104", + "right": "pull_request:5105", + "accept": false, + "reason": "Same idea, but separate training scripts (SDXL LoRA vs base DreamBooth); not one concrete code-path change." }, { - "left": "pull_request:6908", - "right": "pull_request:6944", + "left": "pull_request:5239", + "right": "pull_request:5242", "accept": false, - "reason": "Different training example model-card standardizations; not the same code path or artifact." + "reason": "Same class, but different methods (encode vs decode) and separate docstring fixes." }, { - "left": "pull_request:6908", - "right": "pull_request:6947", + "left": "pull_request:7903", + "right": "pull_request:8064", "accept": false, - "reason": "Different example script/model-card cleanup; only shares the general standardization theme." + "reason": "Unrelated work: one is a video processor test PR, the other is a Marigold pipeline addition/test PR." }, { - "left": "pull_request:6908", - "right": "pull_request:6963", + "left": "pull_request:13069", + "right": "pull_request:13071", "accept": false, - "reason": "Different training examples and docs paths; not the same concrete change." + "reason": "Different transformer models and different test refactors; same CI pattern only." }, { - "left": "pull_request:6908", - "right": "pull_request:7615", + "left": "pull_request:6907", + "right": "pull_request:6908", "accept": false, - "reason": "Both are model-card cleanup PRs, but for different scripts and workflows; not duplicates." + "reason": "Both are ControlNet model-card cleanups, but for different example scripts and scopes." }, { - "left": "pull_request:9591", - "right": "pull_request:9715", + "left": "pull_request:6907", + "right": "pull_request:6909", "accept": false, - "reason": "Different docstring refactors for different transformer classes; same formatting theme only." + "reason": "Different training entrypoints; same documentation standardization theme only." }, { - "left": "pull_request:6942", - "right": "pull_request:6963", + "left": "pull_request:6907", + "right": "pull_request:6910", "accept": false, - "reason": "Separate SDXL training-script model-card updates; different examples and content." + "reason": "Same broad docs work, but different script and change set; not duplicates." }, { - "left": "pull_request:6944", - "right": "pull_request:6963", + "left": "pull_request:10109", + "right": "pull_request:10365", "accept": false, - "reason": "Different example scripts and model-card text; not the same fix." + "reason": "Both add torch_xla support, but to different pipelines and code paths." }, { - "left": "pull_request:6947", - "right": "pull_request:6963", + "left": "pull_request:10183", + "right": "pull_request:10268", "accept": false, - "reason": "Different training examples; same model-card cleanup style, not the same change." + "reason": "Different single-file loaders for different models; same loader subsystem only." }, { - "left": "pull_request:1513", - "right": "pull_request:1564", - "accept": true, - "reason": "Both edit the DreamBooth README to improve Accelerate setup guidance; same doc section and same underlying user-facing issue." + "left": "pull_request:11920", + "right": "pull_request:11954", + "accept": false, + "reason": "Both add test slices, but for different pipelines and expected outputs." }, { - "left": "pull_request:5833", - "right": "pull_request:7934", + "left": "pull_request:11954", + "right": "pull_request:11955", "accept": false, - "reason": "Both involve single-file loading, but one is an import-path fix and the other is config-dict handling; different bugs." + "reason": "Different pipeline test updates; same testing style, not the same bug/change." }, { - "left": "pull_request:5834", - "right": "pull_request:7934", + "left": "pull_request:13253", + "right": "pull_request:13254", "accept": false, - "reason": "Import fix vs config-loading fix in single-file code; related area, not the same concrete problem." + "reason": "Different model test refactors (Z-Image vs LTX); not the same concrete change." }, { - "left": "pull_request:10109", - "right": "pull_request:10364", + "left": "pull_request:13253", + "right": "pull_request:13315", "accept": false, - "reason": "Both add torch_xla support, but to different pipelines and code paths; not mergeable as one PR." + "reason": "Different models and files; same refactor pattern only." }, { - "left": "pull_request:10183", - "right": "pull_request:10991", + "left": "pull_request:13253", + "right": "pull_request:13335", "accept": false, - "reason": "Different single-file support additions for different models; same framework, unrelated model targets." + "reason": "Different model test targets; not a duplicate." }, { - "left": "pull_request:10183", - "right": "pull_request:9083", + "left": "pull_request:13253", + "right": "pull_request:13341", "accept": false, - "reason": "AutoencoderDC vs Flux single-file support are separate model integrations, not the same change." + "reason": "Different model families and tests; no shared concrete change." }, { - "left": "pull_request:10268", - "right": "pull_request:10947", + "left": "pull_request:13253", + "right": "pull_request:13342", "accept": false, - "reason": "Different model support additions in single-file loader; distinct targets and converters." + "reason": "Different model test refactors, not the same PR." }, { - "left": "pull_request:11920", - "right": "pull_request:11953", + "left": "pull_request:13254", + "right": "pull_request:13346", "accept": false, - "reason": "Different test-slice updates for Wan vs HiDream; same test style only." + "reason": "Different model test files and model families; only the refactor pattern overlaps." }, { - "left": "pull_request:11953", - "right": "pull_request:11955", + "left": "pull_request:13315", + "right": "pull_request:13346", "accept": false, - "reason": "Different pipelines and expected slices; both test updates, not duplicates." + "reason": "Different models and test code paths; not duplicates." }, { - "left": "pull_request:1291", - "right": "pull_request:309", + "left": "pull_request:13335", + "right": "pull_request:13346", "accept": false, - "reason": "Both touch links, but one fixes a README notebook link and the other updates a different reference; not the same artifact." + "reason": "Different model test refactors; same theme but not the same change." }, { - "left": "pull_request:13253", + "left": "pull_request:13341", "right": "pull_request:13346", "accept": false, - "reason": "Z-Image test refactor vs Cog transformer test refactor; different models and files." + "reason": "Different model families and files; not the same underlying fix." }, { - "left": "pull_request:13254", - "right": "pull_request:13345", + "left": "pull_request:2812", + "right": "pull_request:2815", + "accept": true, + "reason": "Same Stable UnCLIP docs cleanup theme, same docs file family, and 2815 is the broader final docs update." + }, + { + "left": "pull_request:2813", + "right": "pull_request:2815", + "accept": true, + "reason": "Same Stable UnCLIP docs fix set; 2815 looks like the consolidated/broader version." + }, + { + "left": "pull_request:3568", + "right": "pull_request:3577", "accept": false, - "reason": "LTX test refactor vs Chroma/LongCat/HiDream refactor; same refactor category, different models." + "reason": "Both fix broken links, but in different documentation pages and contexts." }, { - "left": "pull_request:13315", - "right": "pull_request:13345", + "left": "pull_request:6908", + "right": "pull_request:6942", "accept": false, - "reason": "Sana test refactor vs Chroma/LongCat/HiDream refactor; different model families." + "reason": "Different example scripts and model-card templates; same standardization theme only." }, { - "left": "pull_request:13335", - "right": "pull_request:13345", + "left": "pull_request:6908", + "right": "pull_request:6944", "accept": false, - "reason": "Cosmos test refactor vs Chroma/LongCat/HiDream refactor; unrelated concrete edits." + "reason": "Different training scripts and docs content; not one concrete change." }, { - "left": "pull_request:13341", - "right": "pull_request:13345", + "left": "pull_request:6908", + "right": "pull_request:6947", "accept": false, - "reason": "Bria test refactor vs Chroma/LongCat/HiDream refactor; not the same change." + "reason": "Same broad model-card cleanup category, but different pipelines and code paths." }, { - "left": "pull_request:2486", - "right": "pull_request:2488", + "left": "pull_request:6908", + "right": "pull_request:6963", "accept": false, - "reason": "Both move test logic into mixins, but they adjust different behaviors in different test files; not duplicates." + "reason": "Different example scripts; similar docs standardization, not duplicates." }, { - "left": "pull_request:2487", - "right": "pull_request:555", + "left": "pull_request:6908", + "right": "pull_request:7615", "accept": false, - "reason": "Stable Diffusion JAX docs addition vs Flax VAE implementation; entirely different artifacts." + "reason": "Both standardize model cards, but for different training recipes and files." }, { - "left": "pull_request:5871", - "right": "pull_request:7903", + "left": "pull_request:10402", + "right": "pull_request:10434", + "accept": true, + "reason": "Same broken-links/docs-fix family; 10434 includes the same hunyuan pages and broadens the cleanup across more docs." + }, + { + "left": "pull_request:6942", + "right": "pull_request:6963", "accept": false, - "reason": "Deprecated scheduler rebase/test PR vs video processor validation PR; unrelated code paths." + "reason": "Different training examples and model-card text; not the same change." }, { - "left": "pull_request:5979", - "right": "pull_request:6133", + "left": "pull_request:6944", + "right": "pull_request:6963", + "accept": false, + "reason": "Different scripts and card generation logic; only the docs standardization theme matches." + }, + { + "left": "pull_request:6947", + "right": "pull_request:6963", "accept": false, - "reason": "DDGAN pipeline/scheduler work vs UFOGen pipeline/scheduler work; same architectural shape, different features." + "reason": "Different training entrypoints and model-card content; not duplicates." } ] }, @@ -179885,26 +179621,26 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 245, - "review_unit_count": 273, + "review_unit_index": 244, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10259", "pull_request:10268", "pull_request:1073", + "pull_request:10947", "pull_request:10991", - "pull_request:11059", - "pull_request:11243", - "pull_request:11516", - "pull_request:11571", "pull_request:11655", + "pull_request:11920", "pull_request:11953", "pull_request:11954", + "pull_request:11955", + "pull_request:1291", "pull_request:13253", "pull_request:13254", "pull_request:13315", @@ -179918,15 +179654,16 @@ "pull_request:1420", "pull_request:178", "pull_request:1998", + "pull_request:2486", + "pull_request:2487", + "pull_request:2488", "pull_request:260", "pull_request:2839", - "pull_request:3168", - "pull_request:3676", - "pull_request:4513", - "pull_request:5626", - "pull_request:5671", + "pull_request:309", + "pull_request:555", "pull_request:5871", - "pull_request:6532", + "pull_request:5979", + "pull_request:6133", "pull_request:6534", "pull_request:6589", "pull_request:6940", @@ -179934,16 +179671,27 @@ "pull_request:6944", "pull_request:6963", "pull_request:7425", - "pull_request:7544", "pull_request:7613", "pull_request:7615", + "pull_request:7903", "pull_request:7910", - "pull_request:7984", - "pull_request:7988", "pull_request:9591", "pull_request:9714" ], "soft_pairs": [ + "pull_request:10268|pull_request:10947", + "pull_request:11920|pull_request:11953", + "pull_request:11953|pull_request:11955", + "pull_request:1291|pull_request:309", + "pull_request:13253|pull_request:13346", + "pull_request:13254|pull_request:13345", + "pull_request:13315|pull_request:13345", + "pull_request:13335|pull_request:13345", + "pull_request:13341|pull_request:13345", + "pull_request:2486|pull_request:2488", + "pull_request:2487|pull_request:555", + "pull_request:5871|pull_request:7903", + "pull_request:5979|pull_request:6133", "pull_request:6534|pull_request:6589", "pull_request:7425|pull_request:7910", "pull_request:9591|pull_request:9714", @@ -179968,242 +179716,264 @@ "pull_request:13341|pull_request:13348", "pull_request:13344|pull_request:13346", "pull_request:1420|pull_request:178", - "pull_request:1998|pull_request:2839", - "pull_request:3168|pull_request:3676", - "pull_request:4513|pull_request:5671", - "pull_request:5626|pull_request:7544", - "pull_request:5671|pull_request:5871", - "pull_request:6532|pull_request:7984", - "pull_request:6532|pull_request:7988", - "pull_request:11516|pull_request:11571", - "pull_request:11059|pull_request:11243" + "pull_request:1998|pull_request:2839" ], - "prepared_review_unit_hash": "dd8ddf1103c031813d89b5bfec957019fe97ba9d7fb409d1b1c888ec38d86488", + "prepared_review_unit_hash": "266da5cbaf8d179a337081a621b7ddd812a56d06020bb7e0b8583861896abe6c", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 33, - "serialized_chars": 98596, - "estimated_input_tokens": 24649, - "estimated_eval_tokens": 49554 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 38, + "serialized_chars": 102933, + "estimated_input_tokens": 25734, + "estimated_eval_tokens": 51724 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not one coherent duplicate cluster; it splits into several unrelated mini-groups (docs edits, training-script compile fixes, single-file loaders, LoRA support, and test refactors). Only a couple of soft edges look like true duplicates: the doc-builder header tweak pair and the CPU offload test assertion fixes.", - "confidence": 0.88, + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs typos, pipeline/model feature additions, and model-test refactors across many different components. No soft edge looks like the same underlying bug or change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: the items are too heterogeneous to treat as one duplicate set.", + "canonical_pr_reason": "No single canonical PR fits well: the items span many unrelated PRs and codepaths, so there is no coherent duplicate anchor for the cluster.", "best_issue_reason": null, - "best_pr_reason": "If forced to pick a representative, 7544 is the strongest concrete follow-up in the only clearly duplicate subgroup (CPU offload test fixes), but the overall cluster is too mixed for a true global best.", + "best_pr_reason": "No best PR: the cluster is heterogeneous, and the PRs address different models, scripts, docs, and loader behaviors rather than one shared change.", "soft_edge_verdicts": [ + { + "left": "pull_request:10268", + "right": "pull_request:10947", + "accept": false, + "reason": "Different single-file targets: Mochi Transformer vs SANA Transformer." + }, + { + "left": "pull_request:11920", + "right": "pull_request:11953", + "accept": false, + "reason": "Different model test slices: Wan vs HiDream-Image." + }, + { + "left": "pull_request:11953", + "right": "pull_request:11955", + "accept": false, + "reason": "Different model families: HiDream-Image vs Cosmos." + }, + { + "left": "pull_request:1291", + "right": "pull_request:309", + "accept": false, + "reason": "Both are doc link fixes, but they touch different notebooks/URLs and are unrelated changes." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13346", + "accept": false, + "reason": "Different transformer models and test suites: Z Image vs CogVideoX/CogView." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test refactors: LTX vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test refactors: Sana vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test refactors: Cosmos vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:13341", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test refactors: Bria vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:2486", + "right": "pull_request:2488", + "accept": false, + "reason": "Both change test mixin behavior, but they address different helpers and different concrete behaviors." + }, + { + "left": "pull_request:2487", + "right": "pull_request:555", + "accept": false, + "reason": "Unrelated: a Flax/JAX guide versus a VAE implementation." + }, + { + "left": "pull_request:5871", + "right": "pull_request:7903", + "accept": false, + "reason": "Unrelated: deprecated scheduler docs/rebase work versus video processor tests." + }, + { + "left": "pull_request:5979", + "right": "pull_request:6133", + "accept": false, + "reason": "Different new features: DDGAN pipeline/scheduler vs UFOGen pipeline/scheduler." + }, { "left": "pull_request:6534", "right": "pull_request:6589", "accept": false, - "reason": "Same broad torch.compile theme, but different training scripts/model families and different fixes; not the same concrete change." + "reason": "Similar torch.compile theme, but different training scripts and code paths." }, { "left": "pull_request:7425", "right": "pull_request:7910", "accept": false, - "reason": "Both are doc-only changes, but they fix unrelated docs in different files and are not the same underlying change." + "reason": "Different documentation areas: PEFT tutorial snippet vs image processor API docs." }, { "left": "pull_request:9591", "right": "pull_request:9714", "accept": false, - "reason": "Unrelated docstring refactors for different models/files; too broad to be duplicates." + "reason": "Unrelated docstring edits in different model classes." }, { "left": "pull_request:10259", "right": "pull_request:11655", "accept": false, - "reason": "Both touch LoRA loading, but one adds regular Flux LoRA support and the other adds bnb 8-bit handling; different fixes." + "reason": "Same loader file, but different problems: Flux LoRA loading vs bnb 8-bit dequantization support." }, { "left": "pull_request:6942", "right": "pull_request:7615", "accept": false, - "reason": "Both standardize model cards, but they target different scripts and templates." + "reason": "Both standardize model cards, but for different training scripts and model variants." }, { "left": "pull_request:6940", "right": "pull_request:7613", "accept": false, - "reason": "Different training scripts and model-card content; same style, not the same change." + "reason": "Different model-card updates for different training scripts and outputs." }, { "left": "pull_request:6940", "right": "pull_request:7615", "accept": false, - "reason": "Same as above: related docs cleanup, but not one concrete duplicate fix." + "reason": "Different model-card updates for different training scripts and outputs." }, { "left": "pull_request:6944", "right": "pull_request:7613", "accept": false, - "reason": "Different scripts and outputs; too loosely related to merge as duplicates." + "reason": "Different training scripts: T2I LoRA SDXL vs advanced SD15 LoRA." }, { "left": "pull_request:6963", "right": "pull_request:7615", "accept": false, - "reason": "Both are model-card standardization PRs, but for different training recipes and files." + "reason": "Different training scripts: textual inversion SDXL vs advanced SDXL LoRA." }, { "left": "pull_request:10268", "right": "pull_request:10991", "accept": false, - "reason": "Both add single-file support, but for different model families and checkpoint converters." + "reason": "Different single-file targets: Mochi Transformer vs Wan models." }, { "left": "pull_request:1073", "right": "pull_request:260", "accept": false, - "reason": "Both are typo/doc fixes, but in unrelated files with no shared underlying bug." + "reason": "Different README typo fixes in unrelated example folders." }, { "left": "pull_request:11953", "right": "pull_request:11954", "accept": false, - "reason": "Both add test slices, but for different pipelines and different test coverage." + "reason": "Different fast test slices for different pipelines: HiDream-Image vs Hunyuan Video." }, { "left": "pull_request:13253", "right": "pull_request:13344", "accept": false, - "reason": "Different transformer models and different goals; not the same change." + "reason": "Different model test suites: Z Image vs GLM Image." }, { "left": "pull_request:13253", "right": "pull_request:13345", "accept": false, - "reason": "Both are test refactors, but for different model families and files." + "reason": "Different model test suites: Z Image vs Chroma/LongCat/HiDream." }, { "left": "pull_request:13254", "right": "pull_request:13347", "accept": false, - "reason": "Different transformer test suites and different model groups." + "reason": "Different transformer test groups and different models." }, { "left": "pull_request:13254", "right": "pull_request:13348", "accept": false, - "reason": "Same test-refactor theme, but not the same concrete code path or model set." + "reason": "Different transformer test groups and different models." }, { "left": "pull_request:13315", "right": "pull_request:13347", "accept": false, - "reason": "Different transformer families and test suites; too disjoint." + "reason": "Different transformer test groups and different models." }, { "left": "pull_request:13315", "right": "pull_request:13348", "accept": false, - "reason": "Not the same bug/change, just a similar refactor pattern." + "reason": "Different transformer test groups and different models." }, { "left": "pull_request:13335", "right": "pull_request:13347", "accept": false, - "reason": "Different models and separate test refactors." + "reason": "Different transformer test groups and different models." }, { "left": "pull_request:13335", "right": "pull_request:13348", "accept": false, - "reason": "Different transformer test suites; not duplicates." + "reason": "Different transformer test groups and different models." }, { "left": "pull_request:13341", "right": "pull_request:13347", "accept": false, - "reason": "Different model families; same refactor style is not enough." + "reason": "Different transformer test groups and different models." }, { "left": "pull_request:13341", "right": "pull_request:13348", "accept": false, - "reason": "Unrelated transformer test refactors." + "reason": "Different transformer test groups and different models." }, { "left": "pull_request:13344", "right": "pull_request:13346", "accept": false, - "reason": "Different models and different test additions; same general area only." + "reason": "Different model tests: GLM Image vs CogVideoX/CogView." }, { "left": "pull_request:1420", "right": "pull_request:178", "accept": false, - "reason": "Both are pipeline tests, but they cover different Stable Diffusion versions and different behaviors." + "reason": "Both are Stable Diffusion tests, but they target different variants and different test goals." }, { "left": "pull_request:1998", "right": "pull_request:2839", "accept": false, - "reason": "Both touch README files, but they are unrelated documentation updates." - }, - { - "left": "pull_request:3168", - "right": "pull_request:3676", - "accept": true, - "reason": "Both are the same doc-builder test tweak in the same file and same line, just with a different temporary title string." - }, - { - "left": "pull_request:4513", - "right": "pull_request:5671", - "accept": false, - "reason": "Both are CI/test related, but they address unrelated workflows and checks." - }, - { - "left": "pull_request:5626", - "right": "pull_request:7544", - "accept": true, - "reason": "Same test file and same CPU offload assertion path; 7544 is a follow-up cleanup of the same underlying offload-test bug." - }, - { - "left": "pull_request:5671", - "right": "pull_request:5871", - "accept": false, - "reason": "Different CI/test maintenance tasks with no shared underlying code change." - }, - { - "left": "pull_request:6532", - "right": "pull_request:7984", - "accept": false, - "reason": "Different torch.compile compatibility fixes in different components (training script vs VAE)." - }, - { - "left": "pull_request:6532", - "right": "pull_request:7988", - "accept": false, - "reason": "Same broad theme, but the IPAdapter compile fix is a different code path from Dreambooth training." - }, - { - "left": "pull_request:11516", - "right": "pull_request:11571", - "accept": false, - "reason": "Both are LTX Video updates, but they add different variants/features and touch overlapping docs/pipeline areas only broadly." - }, - { - "left": "pull_request:11059", - "right": "pull_request:11243", - "accept": false, - "reason": "Same LoRA conversion module, but one is for non-diffusers Wan T2V LoRAs and the other for musubi Wan LoRAs; different source formats." + "reason": "Unrelated README updates in different top-level/example docs." } ] }, @@ -180218,49 +179988,50 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 246, - "review_unit_count": 273, + "review_unit_index": 245, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10183", - "pull_request:10268", "pull_request:10279", "pull_request:10288", "pull_request:10323", - "pull_request:10437", - "pull_request:10508", "pull_request:10568", "pull_request:10781", - "pull_request:10947", - "pull_request:11058", + "pull_request:11059", + "pull_request:11243", + "pull_request:11516", + "pull_request:11571", "pull_request:12215", "pull_request:12301", - "pull_request:12933", "pull_request:13082", "pull_request:13119", - "pull_request:13217", "pull_request:13253", "pull_request:13255", "pull_request:13346", "pull_request:13347", "pull_request:13348", - "pull_request:2257", - "pull_request:2400", + "pull_request:3168", + "pull_request:3676", "pull_request:3923", "pull_request:3941", + "pull_request:4513", "pull_request:5093", "pull_request:5220", "pull_request:5626", + "pull_request:5671", + "pull_request:5871", "pull_request:6532", "pull_request:6534", "pull_request:6555", "pull_request:6589", "pull_request:7542", + "pull_request:7544", "pull_request:7984", "pull_request:7988", "pull_request:8571", @@ -180270,12 +180041,19 @@ "pull_request:9152", "pull_request:9257", "pull_request:9318", - "pull_request:9401", "pull_request:9473", "pull_request:9553", "pull_request:9993" ], "soft_pairs": [ + "pull_request:3168|pull_request:3676", + "pull_request:4513|pull_request:5671", + "pull_request:5626|pull_request:7544", + "pull_request:5671|pull_request:5871", + "pull_request:6532|pull_request:7984", + "pull_request:6532|pull_request:7988", + "pull_request:11516|pull_request:11571", + "pull_request:11059|pull_request:11243", "pull_request:8755|pull_request:9993", "pull_request:6532|pull_request:6589", "pull_request:8571|pull_request:8572", @@ -180296,204 +180074,210 @@ "pull_request:8755|pull_request:9473", "pull_request:10568|pull_request:9318", "pull_request:6555|pull_request:6589", - "pull_request:8926|pull_request:9257", - "pull_request:12933|pull_request:13217", - "pull_request:10437|pull_request:11058", - "pull_request:10268|pull_request:10288", - "pull_request:10508|pull_request:9401", - "pull_request:10781|pull_request:10947", - "pull_request:2257|pull_request:2400" + "pull_request:8926|pull_request:9257" ], - "prepared_review_unit_hash": "7bebdefe5c8ab6379fe2603ba09b45c468995c911b302c05d0b30f76890a4994", + "prepared_review_unit_hash": "fffb1cf07a5458f235ee7eaa913fb301f7b52231e85791b776d6154e93a03eb1", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 27, - "serialized_chars": 105735, - "estimated_input_tokens": 26434, - "estimated_eval_tokens": 53124 + "soft_pair_count": 29, + "serialized_chars": 99334, + "estimated_input_tokens": 24834, + "estimated_eval_tokens": 49924 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a grab bag of mostly unrelated PRs: single-file model support, docs/link fixes, test refactors, CI token updates, LoRA conversion tweaks, and training-script bug fixes. The soft edges are driven by broad topical similarity or shared files, but they are not the same underlying change.", - "confidence": 0.96, + "summary": "This cluster is mostly a mix of unrelated docs, tests, training-script, and model-support PRs. I only found a few true overlap pairs: doc-builder header tweaks, CPU offload test assertions, and quantization link fixes.", + "confidence": 0.72, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single canonical PR for the whole set: the items split into several unrelated subclusters. Among the overlapping docs-link fixes, PR 10279 is the cleanest representative.", "best_issue_reason": null, - "best_pr_reason": "PR 13217 is the most substantial, self-contained feature change in the set, with broad code and test updates plus many references; if one PR must anchor the cluster, this is the strongest representative.", + "best_pr_reason": "PR 10279 is the strongest representative of the small overlap subcluster because it fixes the same quantization overview links as 10323 and also includes the related GGUF doc correction.", "soft_edge_verdicts": [ + { + "left": "pull_request:3168", + "right": "pull_request:3676", + "accept": true, + "reason": "Same doc-builder WIP on the same docs index header change; same underlying docs-only tweak." + }, + { + "left": "pull_request:4513", + "right": "pull_request:5671", + "accept": false, + "reason": "Unrelated changes: safetensors/version pinning in setup/deps vs dependency-test workflow debugging." + }, + { + "left": "pull_request:5626", + "right": "pull_request:7544", + "accept": true, + "reason": "Both fix the CPU offload test assertions in test_pipelines_common.py; same underlying offload-test bug." + }, + { + "left": "pull_request:5671", + "right": "pull_request:5871", + "accept": false, + "reason": "Different targets: dependency-test workflow tweaks vs deprecated-scheduler doc/test rebasing." + }, + { + "left": "pull_request:6532", + "right": "pull_request:7984", + "accept": false, + "reason": "Different code paths: Dreambooth compile-compatibility vs VAE compile-compatibility." + }, + { + "left": "pull_request:6532", + "right": "pull_request:7988", + "accept": false, + "reason": "Different code paths and model surfaces: Dreambooth training script vs IPAdapter compile compatibility." + }, + { + "left": "pull_request:11516", + "right": "pull_request:11571", + "accept": false, + "reason": "Related project, but not the same change: initial LTX 0.9.7 support vs later distilled-model addition and docs." + }, + { + "left": "pull_request:11059", + "right": "pull_request:11243", + "accept": false, + "reason": "Same converter file, but different WAN LoRA formats (t2v vs musubi) and different concrete key layouts." + }, { "left": "pull_request:8755", "right": "pull_request:9993", "accept": false, - "reason": "Same DreamBooth SD3 script, but they fix different bugs: text-encoder training vs checkpoint saving." + "reason": "Separate SD3 Dreambooth LoRA fixes: text-encoder training bug vs checkpoint-saving bug." }, { "left": "pull_request:6532", "right": "pull_request:6589", "accept": false, - "reason": "Both address torch.compile compatibility, but in different training scripts and code paths; not one duplicate fix." + "reason": "Both are compile-related, but they fix different training scripts and different unwrap/save behavior." }, { "left": "pull_request:8571", "right": "pull_request:8572", "accept": false, - "reason": "Both are typo/style cleanup PRs, but they touch different docs areas and are separate edits." + "reason": "Both are broad typo/style errata, but they touch different docs and code areas; not the same underlying change." }, { "left": "pull_request:13119", "right": "pull_request:13255", "accept": false, - "reason": "Both modify LoRA conversion utilities, but for different model families and different loading bugs." + "reason": "Different model-specific LoRA loading fixes: Flux2 key handling vs Z-Image loading." }, { "left": "pull_request:10183", "right": "pull_request:10288", "accept": false, - "reason": "Single-file support, but for different models (AutoencoderDC vs Flux Canny/Depth/Fill)." + "reason": "Different single-file support additions for different models (AutoencoderDC vs Flux variants)." }, { "left": "pull_request:10183", "right": "pull_request:10781", "accept": false, - "reason": "Different single-file feature work for different models; not the same concrete change." + "reason": "Different single-file support additions for different models (AutoencoderDC vs Lumina2 transformer)." }, { "left": "pull_request:13082", "right": "pull_request:13346", "accept": false, - "reason": "Both are test refactors, but they target different transformer families and separate files." + "reason": "Unrelated test-refactor PRs for different transformer families; same broad theme, not same bug/change." }, { "left": "pull_request:13253", "right": "pull_request:13347", "accept": false, - "reason": "Both refactor transformer tests, but for different models and unrelated test suites." + "reason": "Different test refactors for different transformer families; not the same underlying change." }, { "left": "pull_request:13253", "right": "pull_request:13348", "accept": false, - "reason": "Different model test refactors; same theme, not the same change." + "reason": "Different test refactors for different transformer families; not the same underlying change." }, { "left": "pull_request:3923", "right": "pull_request:3941", "accept": false, - "reason": "Both are docs fixes, but they address different documentation issues." + "reason": "Different documentation fixes on different pages; both are torch-import/doc-link cleanups, but not the same change." }, { "left": "pull_request:6534", "right": "pull_request:7984", "accept": false, - "reason": "Both mention torch.compile, but one is a training-script compatibility fix and the other is a VAE implementation change." + "reason": "Dreambooth LoRA compile compatibility vs VAE compile compatibility; separate fixes." }, { "left": "pull_request:6534", "right": "pull_request:7988", "accept": false, - "reason": "Different components and different fixes; only the general compile-compat theme overlaps." + "reason": "Dreambooth LoRA compile compatibility vs IPAdapter compile compatibility; separate fixes." }, { "left": "pull_request:9152", "right": "pull_request:9553", "accept": false, - "reason": "Same docs page, but one adds a fuller example while the other adds a device-map tip; separate changes." + "reason": "Same docs file and topic, but different distributed-inference enhancements rather than one shared fix." }, { "left": "pull_request:12215", "right": "pull_request:12301", "accept": false, - "reason": "Related Qwen-Image features, but one adds ControlNet and the other adds ControlNet-Inpainting; not the same PR." + "reason": "Qwen-Image ControlNet vs ControlNet-inpaint are different features and code paths." }, { "left": "pull_request:10279", "right": "pull_request:10323", - "accept": false, - "reason": "Both are quantization docs fixes, but they change different pages and different links." + "accept": true, + "reason": "Both make the same quantization overview link corrections; 10279 just includes an extra related GGUF docs change." }, { "left": "pull_request:5626", "right": "pull_request:7542", "accept": false, - "reason": "Both touch the same CPU-offload test file, but they fix different issues: assertion logic vs a typo." + "reason": "Same offload test file, but different assertions: CPU-offload device check vs a sequential-offload typo fix." }, { "left": "pull_request:5093", "right": "pull_request:5220", "accept": false, - "reason": "Both move slow tests to nightly, but they cover different test groups and separate PRs." + "reason": "Both move slow tests to nightly, but they cover different test batches and are not the same change." }, { "left": "pull_request:8755", "right": "pull_request:9473", "accept": false, - "reason": "Different DreamBooth scripts and different bugs; same broad training area only." + "reason": "Different SD3 Dreambooth LoRA fixes in the same script family; not the same bug." }, { "left": "pull_request:10568", "right": "pull_request:9318", "accept": false, - "reason": "Both update HF tokens in CI, but they target different workflows and different job scopes." + "reason": "Both update HF token usage in CI, but one is fast GPU tests and the other is nightly tests; separate workflows." }, { "left": "pull_request:6555", "right": "pull_request:6589", "accept": false, - "reason": "Both are compile-compat training fixes, but for different scripts and model variants." + "reason": "Different scripts and different fixes; same compile-compatibility theme only at a very high level." }, { "left": "pull_request:8926", "right": "pull_request:9257", "accept": false, - "reason": "Different training scripts and different bugs; not a duplicate fix." - }, - { - "left": "pull_request:12933", - "right": "pull_request:13217", - "accept": false, - "reason": "Both concern LTX-2, but one adds LoRA support while the other adds LTX-2.3 model support." - }, - { - "left": "pull_request:10437", - "right": "pull_request:11058", - "accept": false, - "reason": "Both make tests more device-agnostic, but they modify different test infrastructure and different coverage areas." - }, - { - "left": "pull_request:10268", - "right": "pull_request:10288", - "accept": false, - "reason": "Single-file support for different models; same subsystem, different concrete feature." - }, - { - "left": "pull_request:10508", - "right": "pull_request:9401", - "accept": false, - "reason": "Both are test device-agnostic efforts, but they target different suites and different accelerator assumptions." - }, - { - "left": "pull_request:10781", - "right": "pull_request:10947", - "accept": false, - "reason": "Different model single-file support work: Lumina2 transformer vs SANA transformer." - }, - { - "left": "pull_request:2257", - "right": "pull_request:2400", - "accept": false, - "reason": "Different community pipelines and different interpolation modes; not the same artifact." + "reason": "Different Dreambooth/Flux fixes touching different scripts and failure modes." } ] }, @@ -180508,28 +180292,36 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 247, - "review_unit_count": 273, + "review_unit_index": 246, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10268", + "pull_request:10288", + "pull_request:10437", + "pull_request:10508", + "pull_request:10781", "pull_request:1087", + "pull_request:10947", + "pull_request:11058", "pull_request:11386", "pull_request:11391", - "pull_request:11407", - "pull_request:11416", - "pull_request:11476", "pull_request:1210", "pull_request:1218", "pull_request:12236", "pull_request:12387", + "pull_request:12933", + "pull_request:13217", "pull_request:1510", "pull_request:2193", + "pull_request:2257", "pull_request:2271", + "pull_request:2400", "pull_request:2740", "pull_request:2741", "pull_request:2751", @@ -180538,8 +180330,6 @@ "pull_request:2939", "pull_request:3472", "pull_request:3525", - "pull_request:3676", - "pull_request:3687", "pull_request:4498", "pull_request:4499", "pull_request:4500", @@ -180547,12 +180337,8 @@ "pull_request:4502", "pull_request:4946", "pull_request:5267", - "pull_request:5439", - "pull_request:5440", "pull_request:5855", "pull_request:6526", - "pull_request:703", - "pull_request:705", "pull_request:7934", "pull_request:7984", "pull_request:7988", @@ -180560,12 +180346,17 @@ "pull_request:8490", "pull_request:8860", "pull_request:888", - "pull_request:8992", - "pull_request:9040", "pull_request:9255", - "pull_request:9355" + "pull_request:9355", + "pull_request:9401" ], "soft_pairs": [ + "pull_request:12933|pull_request:13217", + "pull_request:10437|pull_request:11058", + "pull_request:10268|pull_request:10288", + "pull_request:10508|pull_request:9401", + "pull_request:10781|pull_request:10947", + "pull_request:2257|pull_request:2400", "pull_request:3472|pull_request:4946", "pull_request:6526|pull_request:7984", "pull_request:6526|pull_request:7988", @@ -180592,240 +180383,234 @@ "pull_request:4498|pull_request:4499", "pull_request:4498|pull_request:4500", "pull_request:4498|pull_request:4501", - "pull_request:4498|pull_request:4502", - "pull_request:8992|pull_request:9040", - "pull_request:3676|pull_request:3687", - "pull_request:703|pull_request:705", - "pull_request:5439|pull_request:5440", - "pull_request:11407|pull_request:11416", - "pull_request:11416|pull_request:11476" + "pull_request:4498|pull_request:4502" ], - "prepared_review_unit_hash": "33d929f691059cb3ddf2143a9fbdad83525408ecd75a69aa969b69c2d246ec7a", + "prepared_review_unit_hash": "67c067db2c909601ae2bb2fd223c017f0db4506f29c5208ce80caaeb13f62ced", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 96679, - "estimated_input_tokens": 24170, - "estimated_eval_tokens": 48596 + "serialized_chars": 101307, + "estimated_input_tokens": 25327, + "estimated_eval_tokens": 50910 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is really several separate duplicate families: SDXL reference example PRs, typo-sweep docs PRs, CI workflow tweaks, and unrelated bug fixes. Only the clearly same-file/same-change pairs should be treated as duplicates; most soft edges are false positives.", - "confidence": 0.73, + "summary": "This cluster is mostly a mix of unrelated PRs that happen to share broad subsystems or doc/test themes. The only clear duplicate set is the SDXL reference-pipeline PRs, plus one near-duplicate text-inversion docs edit; most other soft edges are false positives because they touch different concrete features or code paths.", + "confidence": 0.71, "canonical_issue_reason": null, - "canonical_pr_reason": "No single global canonical PR fits the whole cluster because it contains multiple unrelated change families. If one representative is needed, PR 4502 is the strongest merged exemplar of the largest coherent family (the SDXL reference example updates).", + "canonical_pr_reason": "4502 is the clearest merged final version of the repeated SDXL reference-pipeline PRs and the strongest representative of the only obvious duplicate group in this set.", "best_issue_reason": null, - "best_pr_reason": "PR 4502 is the best overall representative: it is merged, concrete, and appears to be the final complete version of the SDXL reference example family, which is the cleanest duplicate group in this cluster.", + "best_pr_reason": "4502 has the most complete implementation and discussion among the true duplicate candidates, so it is the best PR anchor here.", "soft_edge_verdicts": [ + { + "left": "pull_request:12933", + "right": "pull_request:13217", + "accept": false, + "reason": "Different features: LTX-2 LoRA support vs LTX-2.3 model support. Same area, but not the same underlying change." + }, + { + "left": "pull_request:10437", + "right": "pull_request:11058", + "accept": false, + "reason": "Both are test-infra refactors, but one adds xpu/device-agnostic support while the other adds multi-accelerator decorators and offload tweaks. Not the same concrete change." + }, + { + "left": "pull_request:10268", + "right": "pull_request:10288", + "accept": false, + "reason": "Both extend single-file support, but for different models (Mochi vs Flux Canny/Depth/Fill)." + }, + { + "left": "pull_request:10508", + "right": "pull_request:9401", + "accept": false, + "reason": "Different test suites and different device-agnostic changes; shared testing utilities are not enough to make this a duplicate." + }, + { + "left": "pull_request:10781", + "right": "pull_request:10947", + "accept": false, + "reason": "Different model support additions: Lumina Image 2.0 vs SANA Transformer." + }, + { + "left": "pull_request:2257", + "right": "pull_request:2400", + "accept": false, + "reason": "Related community pipelines, but one is text interpolation and the other is image interpolation. Different pipeline changes." + }, { "left": "pull_request:3472", "right": "pull_request:4946", "accept": false, - "reason": "Different docs/features: IF Dreambooth instructions vs SDXL support note in training overview." + "reason": "Different docs updates for different training topics; not the same change." }, { "left": "pull_request:6526", "right": "pull_request:7984", "accept": false, - "reason": "Both mention torch.compile, but they fix different code paths in different files (ControlNet training vs VAE decode)." + "reason": "Different torch.compile fixes in different code paths: ControlNet SDXL training vs VAE decode." }, { "left": "pull_request:6526", "right": "pull_request:7988", "accept": false, - "reason": "Different compatibility fixes in different modules; not the same underlying bug." + "reason": "Different compile-compatibility work in different models; same theme, not the same fix." }, { "left": "pull_request:9255", "right": "pull_request:9355", "accept": false, - "reason": "Same workflow file, but different CI changes (release-branch test matrix vs manual dispatch)." + "reason": "Same workflow file, but one changes branch triggering and naming while the other adds workflow_dispatch. Separate changes." }, { "left": "pull_request:7934", "right": "pull_request:8452", "accept": false, - "reason": "Both single-file loading related, but one fixes dict config handling and the other relaxes state-dict loading; distinct bugs." + "reason": "Both are single-file loading fixes, but one handles dict config loading and the other relaxes state_dict strictness. Different bugs." }, { "left": "pull_request:2193", "right": "pull_request:888", "accept": false, - "reason": "Different feature additions: SAG pipeline vs Cycle Diffusion pipeline." + "reason": "Different pipelines entirely: SAG Stable Diffusion vs Cycle Diffusion." }, { "left": "pull_request:12236", "right": "pull_request:12387", "accept": false, - "reason": "Both are kernels/attention work, but they target different backend support and are not the same concrete change." + "reason": "Both involve kernels/flash attention, but they target different backend capabilities and code paths." }, { "left": "pull_request:1087", "right": "pull_request:1210", "accept": false, - "reason": "Different fixes in stable diffusion test/code paths; not the same bug." + "reason": "Both are slow-test fixes, but they touch different tests and expected outputs." }, { "left": "pull_request:1087", "right": "pull_request:2271", "accept": false, - "reason": "Different test expectations/fix scopes; no shared concrete change." + "reason": "Different slow-test adjustments in different files and pipelines." }, { "left": "pull_request:1087", "right": "pull_request:2846", "accept": false, - "reason": "Unrelated stable-diffusion test fix vs stable-unclip input-order fix." + "reason": "Different test fixes with only superficial overlap in the 'Fix slow tests' theme." }, { "left": "pull_request:11386", "right": "pull_request:11391", "accept": false, - "reason": "Same title, but no diff/files shown for 11391; insufficient evidence to confirm the same concrete change." + "reason": "Same filename/title, but 11391 has no diff/files shown here, so there isn't enough evidence it is the same concrete change." }, { "left": "pull_request:1210", "right": "pull_request:2271", "accept": false, - "reason": "Both are test fixes, but they touch different pipelines and different assertions." + "reason": "Both are test-output updates, but they modify different pipelines/schedulers and are not the same change." }, { "left": "pull_request:1210", "right": "pull_request:2846", "accept": false, - "reason": "Cycle Diffusion test changes vs Stable Unclip img2img test changes are unrelated." + "reason": "Different slow-test fixes in different test modules." }, { "left": "pull_request:1218", "right": "pull_request:1510", "accept": false, - "reason": "Both improve the conversion script, but they add different scheduler support/features and are not the same PR." + "reason": "Both touch conversion tooling, but one adds scheduler options while the other improves scheduler/documentation coverage. Different PRs." }, { "left": "pull_request:2271", "right": "pull_request:2846", "accept": false, - "reason": "Different test files and different behaviors; not duplicates." + "reason": "Different slow-test fixes for different test files." }, { "left": "pull_request:2740", "right": "pull_request:2741", "accept": false, - "reason": "2740 changes Dreambooth and Text Inversion docs; 2741 is only Text Inversion, so they are not the same change." + "reason": "2740 includes an additional DreamBooth docs edit, so it is not the same change as the text-inversion-only PR." }, { "left": "pull_request:2740", "right": "pull_request:2751", "accept": false, - "reason": "2740 is broader and includes Dreambooth; 2751 is just the Text Inversion typo fix." + "reason": "2740 spans DreamBooth plus text inversion; 2751 is only the text-inversion docs edit." }, { "left": "pull_request:2741", "right": "pull_request:2751", "accept": true, - "reason": "Same one-line typo fix in the same file (text_inversion.mdx), with 2751 as the merged version." + "reason": "Same text_inversion.md wording fix on the same sentence; 2751 looks like the later merged revision of the same docs change." }, { "left": "pull_request:2924", "right": "pull_request:2939", "accept": false, - "reason": "Different docs edits in different files (pipeline overview vs scheduler docs), not the same change." + "reason": "Both are tiny docs tweaks for VQ Diffusion, but they modify different docs pages and are not the same edit." }, { "left": "pull_request:3525", "right": "pull_request:5267", "accept": false, - "reason": "Both touch README content, but they are different files and different edits." + "reason": "Both are README edits, but they address different content and different sections." }, { "left": "pull_request:3525", "right": "pull_request:5855", "accept": false, - "reason": "Root README update vs community README typo fix; unrelated changes." + "reason": "Unrelated README edits; same file alone is not enough." }, { "left": "pull_request:5267", "right": "pull_request:5855", "accept": false, - "reason": "Docs README formatting changes vs community README import typo; not the same bug." + "reason": "Different documentation fixes in the same README, not the same change." }, { "left": "pull_request:8490", "right": "pull_request:8860", "accept": false, - "reason": "Same README filename, but 8860 has no visible diff/files here, so the concrete change cannot be verified as identical." + "reason": "Same README filename, but the previews indicate different SD3 README edits and not a single duplicate change." }, { "left": "pull_request:4498", "right": "pull_request:4499", "accept": true, - "reason": "Same SDXL reference example family: same files and same feature area, just title wording differs." + "reason": "Same SDXL reference-pipeline feature on the same files; effectively the same change." }, { "left": "pull_request:4498", "right": "pull_request:4500", "accept": true, - "reason": "Same SDXL reference example addition in the same files; clearly the same underlying change family." + "reason": "Same SDXL reference-pipeline implementation and docs, just a parallel PR variant." }, { "left": "pull_request:4498", "right": "pull_request:4501", "accept": true, - "reason": "Same SDXL reference example change set; only the PR snapshot/title differs." + "reason": "Same underlying SDXL reference-pipeline addition with the same touched files." }, { "left": "pull_request:4498", "right": "pull_request:4502", "accept": true, - "reason": "Same SDXL reference example, with 4502 appearing to be the completed merged version." - }, - { - "left": "pull_request:8992", - "right": "pull_request:9040", - "accept": true, - "reason": "Same HunyuanDiT differential diffusion addition; 9040 is the broader/merged continuation including the README update." - }, - { - "left": "pull_request:3676", - "right": "pull_request:3687", - "accept": true, - "reason": "Identical doc-builder edit to docs/source/en/index.mdx; same file and same change." - }, - { - "left": "pull_request:703", - "right": "pull_request:705", - "accept": false, - "reason": "Both CI-related, but one is a CUDA cache test setup and the other is a different workflow/dependency update." - }, - { - "left": "pull_request:5439", - "right": "pull_request:5440", - "accept": false, - "reason": "Both are single-file loading fixes, but they address different bugs (config/tokenizer loading vs local-only CLIP config loading)." - }, - { - "left": "pull_request:11407", - "right": "pull_request:11416", - "accept": false, - "reason": "Both are typo sweeps, but they target different file sets and different concrete edits." - }, - { - "left": "pull_request:11416", - "right": "pull_request:11476", - "accept": false, - "reason": "Both are broad typo/comment sweeps, but they are distinct PRs with different file sets and fixes." + "reason": "Same SDXL reference-pipeline feature; 4502 is the merged final variant of the same change." } ] }, @@ -180840,12 +180625,12 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 248, - "review_unit_count": 273, + "review_unit_index": 247, + "review_unit_count": 272, "status": "reviewed", "reason": null, - "source": "llm", - "cache_hit": false, + "source": "cache", + "cache_hit": true, "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ @@ -180857,6 +180642,9 @@ "pull_request:10852", "pull_request:1087", "pull_request:1131", + "pull_request:11407", + "pull_request:11416", + "pull_request:11476", "pull_request:1210", "pull_request:13071", "pull_request:13082", @@ -180864,41 +180652,43 @@ "pull_request:17", "pull_request:2271", "pull_request:2526", - "pull_request:2864", - "pull_request:2925", "pull_request:3169", "pull_request:344", - "pull_request:3625", "pull_request:365", "pull_request:3676", + "pull_request:3687", "pull_request:386", "pull_request:3863", "pull_request:388", "pull_request:408", - "pull_request:4330", - "pull_request:4472", - "pull_request:4652", "pull_request:5201", + "pull_request:5439", + "pull_request:5440", "pull_request:5506", - "pull_request:6017", - "pull_request:6018", - "pull_request:61", "pull_request:6261", "pull_request:6262", "pull_request:6263", "pull_request:6264", "pull_request:6265", - "pull_request:63", - "pull_request:6647", "pull_request:6962", + "pull_request:703", + "pull_request:705", "pull_request:7131", "pull_request:7451", "pull_request:831", "pull_request:8720", "pull_request:8946", + "pull_request:8992", + "pull_request:9040", "pull_request:9131" ], "soft_pairs": [ + "pull_request:8992|pull_request:9040", + "pull_request:3676|pull_request:3687", + "pull_request:703|pull_request:705", + "pull_request:5439|pull_request:5440", + "pull_request:11407|pull_request:11416", + "pull_request:11416|pull_request:11476", "pull_request:3169|pull_request:3863", "pull_request:6962|pull_request:8720", "pull_request:344|pull_request:365", @@ -180919,218 +180709,192 @@ "pull_request:13071|pull_request:13082", "pull_request:1425|pull_request:408", "pull_request:17|pull_request:408", - "pull_request:2271|pull_request:2526", - "pull_request:2864|pull_request:2925", - "pull_request:3625|pull_request:3676", - "pull_request:3676|pull_request:6017", - "pull_request:3676|pull_request:6018", - "pull_request:4330|pull_request:4472", - "pull_request:4652|pull_request:6647", - "pull_request:6017|pull_request:6018", - "pull_request:61|pull_request:63" + "pull_request:2271|pull_request:2526" ], - "prepared_review_unit_hash": "336c0096697e902963db5bef94726b9b65f1105b9ac767ed8340c02005c3af92", + "prepared_review_unit_hash": "945003d9d172f514de6ded6d9d806a3c733d576dab2427ae3bc563d6a60e8758", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 29, - "serialized_chars": 98047, - "estimated_input_tokens": 24512, - "estimated_eval_tokens": 49280 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 103573, + "estimated_input_tokens": 25894, + "estimated_eval_tokens": 52044 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a mixed bag of unrelated docs/test/refactor PRs. The only convincing duplicate-style pairs are the two FLUX NPU adaptation PRs and the two Remote VAE template PRs; most other pairs are same-area but different concrete changes, so I rejected them.", - "confidence": 0.89, + "summary": "A few pairs are clear duplicates or follow-up fixes in the same file/change, but most are merely similar PRs in the same subsystem and should stay separate.", + "confidence": 0.71, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 10464 is the broader, more complete FLUX NPU adaptation change and best represents the shared code-path fix among the accepted duplicates.", + "canonical_pr_reason": "PR 10784 is the cleanest canonical representative: it is merged, issue-linked, and contains a tight self-contained fix to the same validation bug.", "best_issue_reason": null, - "best_pr_reason": "PR 10464 is the strongest representative PR here: it contains the main NPU flash-attention removal/adaptation plus the related example/model updates, making it the best canonical PR for the cluster.", + "best_pr_reason": "PR 10784 is the best overall candidate because it is a focused merged fix with an explicit issue target and minimal scope.", "soft_edge_verdicts": [ { - "left": "pull_request:3169", - "right": "pull_request:3863", + "left": "pull_request:8992", + "right": "pull_request:9040", "accept": false, - "reason": "Same Habana docs area, but these are different documentation updates for different versions/configs, not the same underlying change." + "reason": "Same feature area, but one is the implementation PR and the other is a follow-up/community README update; not the same concrete change." }, { - "left": "pull_request:6962", - "right": "pull_request:8720", - "accept": false, - "reason": "Both adjust slow-test expectations, but for different pipelines and different precision issues; not the same bug fix." + "left": "pull_request:3676", + "right": "pull_request:3687", + "accept": true, + "reason": "Same file and identical doc-builder test edit; these look like duplicate/overlapping PRs for the same docs title change." }, { - "left": "pull_request:344", - "right": "pull_request:365", + "left": "pull_request:703", + "right": "pull_request:705", "accept": false, - "reason": "Both touch VAE typing, but they change different methods/signatures and are separate cleanup edits." + "reason": "Both touch CI CUDA testing, but they change different workflows and different failure modes." }, { - "left": "pull_request:6262", - "right": "pull_request:6265", + "left": "pull_request:5439", + "right": "pull_request:5440", "accept": false, - "reason": "Both refactor stable_diffusion imports, but they move different pipelines; broad subsystem overlap is not enough." + "reason": "Related single-file loading fixes, but one is config/tokenizer loading while the other is local_files_only propagation." }, { - "left": "pull_request:6263", - "right": "pull_request:6265", + "left": "pull_request:11407", + "right": "pull_request:11416", "accept": false, - "reason": "Different pipeline moved out of stable_diffusion; related refactor, but not the same concrete change." + "reason": "Both are typo cleanup PRs, but they cover different text sets and docs sections rather than one shared fix." }, { - "left": "pull_request:6264", - "right": "pull_request:6265", + "left": "pull_request:11416", + "right": "pull_request:11476", "accept": false, - "reason": "Different pipeline/refactor target; same namespace cleanup, but not one shared fix." + "reason": "Different typo batches across docs versus code/tests; same cleanup style, not the same change." }, { - "left": "pull_request:10678", - "right": "pull_request:10784", + "left": "pull_request:3169", + "right": "pull_request:3863", "accept": false, - "reason": "Same validation pattern, but different pipeline files and independent fixes." + "reason": "Same Habana doc file, but the versions/config updates are different follow-on documentation revisions." }, { - "left": "pull_request:8946", - "right": "pull_request:9131", + "left": "pull_request:6962", + "right": "pull_request:8720", "accept": false, - "reason": "Completely unrelated: one changes test expectations, the other edits setup dependencies." + "reason": "Both target flaky tests, but they fix different pipelines with different assertions and root causes." }, { - "left": "pull_request:6261", + "left": "pull_request:344", + "right": "pull_request:365", + "accept": true, + "reason": "Same VAE model file and same type-hinting effort; 365 looks like a narrow follow-up to the broader PR." + }, + { + "left": "pull_request:6262", "right": "pull_request:6265", "accept": false, - "reason": "Both are stable_diffusion refactors, but they move different pipelines and do not fix one shared issue." + "reason": "Same refactor theme, but these move different pipelines out of stable_diffusion, so they are not the same change." }, { - "left": "pull_request:386", - "right": "pull_request:388", + "left": "pull_request:6263", + "right": "pull_request:6265", "accept": false, - "reason": "Different pipeline docs/code additions; not the same underlying feature or bug fix." + "reason": "Same refactor theme, but different pipeline extraction work in different modules." }, { - "left": "pull_request:5201", - "right": "pull_request:5506", + "left": "pull_request:6264", + "right": "pull_request:6265", "accept": false, - "reason": "Same file, but one fixes LoRA offload behavior while the other fixes xformers/PEFT scaling; different code paths." + "reason": "Same refactor theme, but this is a separate pipeline move, not the same concrete change." }, { - "left": "pull_request:10464", - "right": "pull_request:10465", + "left": "pull_request:10678", + "right": "pull_request:10784", "accept": true, - "reason": "These are clearly overlapping FLUX NPU-adaptation PRs with the same removed flash-attention path and matching core diff." + "reason": "Both patch the same height/width divisibility validation bug pattern with nearly identical code in different pipelines." }, { - "left": "pull_request:7131", - "right": "pull_request:7451", + "left": "pull_request:11407", + "right": "pull_request:11416", + "accept": false, + "reason": "Docs typo batches overlap in style only; they are not the same text fix." + }, + { + "left": "pull_request:11416", + "right": "pull_request:11476", "accept": false, - "reason": "Both are typo/documentation cleanups, but they touch different docs and unrelated content." + "reason": "One is docs/comments typos, the other is example/code string typos; too different to merge as duplicates." }, { "left": "pull_request:10849", "right": "pull_request:10852", "accept": true, - "reason": "Second PR is a direct follow-up tweak to the same remote VAE issue template file, so they are the same underlying template change." + "reason": "Same remote VAE issue template file, with 10852 clearly a follow-up fix to the template introduced in 10849." }, { "left": "pull_request:1087", "right": "pull_request:2526", "accept": false, - "reason": "Both are slow-test fixes, but for different pipelines and different expected outputs." + "reason": "Both are slow-test fixes, but they touch different tests and different nondeterminism sources." }, { "left": "pull_request:1131", "right": "pull_request:831", "accept": false, - "reason": "Both bump versions, but they are different release bumps with different associated scheduler changes." + "reason": "Both bump versions, but they are distinct release bumps to different versions." }, { "left": "pull_request:1210", "right": "pull_request:2526", "accept": false, - "reason": "Separate slow-test adjustments for different pipelines; not the same bug." + "reason": "Both adjust test expectations, but they are separate flake fixes in unrelated pipelines." }, { "left": "pull_request:13071", "right": "pull_request:13082", "accept": false, - "reason": "Different model test refactors for different architectures; same test area, but not one duplicate fix." + "reason": "Both are model test refactors, but for different model families and code paths." }, { "left": "pull_request:1425", "right": "pull_request:408", "accept": false, - "reason": "Both fix docs typos, but in different pages and different wording; not one shared change." + "reason": "Both fix typos in docs, but they are separate content edits across different pages." }, { "left": "pull_request:17", "right": "pull_request:408", "accept": false, - "reason": "Different typo cleanup scopes and files; too broad to treat as the same change." + "reason": "General typo cleanup, but on different docs with no shared concrete bug." }, { "left": "pull_request:2271", "right": "pull_request:2526", "accept": false, - "reason": "Different test baseline updates for different pipelines." + "reason": "Both are test tolerance fixes, but they address different pipelines and numerical expectations." }, { - "left": "pull_request:2864", - "right": "pull_request:2925", - "accept": false, - "reason": "Both are docs overview edits, but they affect different documents and unrelated text changes." - }, - { - "left": "pull_request:3625", - "right": "pull_request:3676", - "accept": false, - "reason": "Both are doc-builder test/WIP PRs, but they are separate experimental edits to the same page rather than one concrete fix." - }, - { - "left": "pull_request:3676", - "right": "pull_request:6017", - "accept": false, - "reason": "Same test-doc-builder theme, but these are distinct WIP PRs with different edits." - }, - { - "left": "pull_request:3676", - "right": "pull_request:6018", - "accept": false, - "reason": "Same reason as above: separate WIP doc-builder test branches, not the same change." - }, - { - "left": "pull_request:4330", - "right": "pull_request:4472", - "accept": false, - "reason": "Different README files and different content changes; not the same underlying issue." - }, - { - "left": "pull_request:4652", - "right": "pull_request:6647", + "left": "pull_request:5201", + "right": "pull_request:5506", "accept": false, - "reason": "One fixes LoRA save/load tests, the other adjusts SDXL k-diffusion test expectations; unrelated." + "reason": "Same file, but one fixes LoRA offload/set_processor behavior while the other fixes xformers+PEFT scaling; different bugs." }, { - "left": "pull_request:6017", - "right": "pull_request:6018", - "accept": false, - "reason": "Separate WIP doc-builder test PRs making different title/content tweaks." + "left": "pull_request:10464", + "right": "pull_request:10465", + "accept": true, + "reason": "Same FLUX NPU adaptation theme and overlapping transformer/controlnet changes; 10465 reads like a narrowed duplicate/follow-up of 10464." }, { - "left": "pull_request:61", - "right": "pull_request:63", + "left": "pull_request:7131", + "right": "pull_request:7451", "accept": false, - "reason": "Different conversion scripts for different pipelines; same repo area, but not the same change." + "reason": "Both are docs typo fixes, but they update different documentation areas and are not the same change." } ] }, @@ -181145,8 +180909,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 249, - "review_unit_count": 273, + "review_unit_index": 248, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -181168,28 +180932,27 @@ "pull_request:2290", "pull_request:2388", "pull_request:2446", - "pull_request:2539", "pull_request:2757", - "pull_request:2839", - "pull_request:3099", + "pull_request:2864", + "pull_request:2925", "pull_request:325", "pull_request:327", "pull_request:333", "pull_request:335", "pull_request:349", - "pull_request:3525", "pull_request:359", "pull_request:360", - "pull_request:3709", - "pull_request:3710", - "pull_request:3819", - "pull_request:4451", - "pull_request:4612", - "pull_request:5220", - "pull_request:5267", - "pull_request:5280", - "pull_request:5292", - "pull_request:5855", + "pull_request:3625", + "pull_request:3676", + "pull_request:4330", + "pull_request:4472", + "pull_request:4612", + "pull_request:4652", + "pull_request:6017", + "pull_request:6018", + "pull_request:61", + "pull_request:63", + "pull_request:6647", "pull_request:6909", "pull_request:6910", "pull_request:6939", @@ -181204,6 +180967,14 @@ "pull_request:9760" ], "soft_pairs": [ + "pull_request:2864|pull_request:2925", + "pull_request:3625|pull_request:3676", + "pull_request:3676|pull_request:6017", + "pull_request:3676|pull_request:6018", + "pull_request:4330|pull_request:4472", + "pull_request:4652|pull_request:6647", + "pull_request:6017|pull_request:6018", + "pull_request:61|pull_request:63", "pull_request:8470|pull_request:9760", "pull_request:8478|pull_request:8479", "pull_request:12117|pull_request:12225", @@ -181223,212 +180994,168 @@ "pull_request:1415|pull_request:2757", "pull_request:1490|pull_request:831", "pull_request:2290|pull_request:2446", - "pull_request:2388|pull_request:4612", - "pull_request:2539|pull_request:3099", - "pull_request:2839|pull_request:3525", - "pull_request:3525|pull_request:3819", - "pull_request:3709|pull_request:3710", - "pull_request:3819|pull_request:5267", - "pull_request:3819|pull_request:5855", - "pull_request:4451|pull_request:5220", - "pull_request:5280|pull_request:5292" + "pull_request:2388|pull_request:4612" ], - "prepared_review_unit_hash": "7ebaf6c352bfe8bbbb7581f71304aa12635c548c9df9dcaabb3b4f30de86e658", + "prepared_review_unit_hash": "10279bfafd289cc48f1c2e1dcc71988410ceac53c96736ae248e8e7ed8c63a1f", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 102323, - "estimated_input_tokens": 25581, - "estimated_eval_tokens": 51418 + "serialized_chars": 96890, + "estimated_input_tokens": 24223, + "estimated_eval_tokens": 48702 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous, with only a couple of true duplicate/near-duplicate PR pairs. Most pairs are broad topical similarities (docs, tests, version bumps, scheduler/model maintenance) rather than the same concrete change.", - "confidence": 0.93, + "summary": "Mostly unrelated PRs spanning docs tweaks, type-hint refactors, release bumps, and separate feature additions; only the HunyuanDiT controlnet/pipeline pair looks like an exact duplicate.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 8478 is the cleanest canonical choice because PR 8479 appears to be the same HunyuanDiT controlnet/pipeline patch duplicated; 8478 is the earlier, representative copy.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 8478 best represents the only true duplicate-style pair in the set, and its content matches 8479 closely enough to serve as the canonical PR.", "soft_edge_verdicts": [ + { + "left": "pull_request:2864", + "right": "pull_request:2925", + "accept": false, + "reason": "Both are tiny docs punctuation tweaks, but they hit different overview pages and are not the same change." + }, + { + "left": "pull_request:3625", + "right": "pull_request:3676", + "accept": false, + "reason": "Both are doc-builder-related, but one changes workflow config while the other changes the docs index title; different fixes." + }, + { + "left": "pull_request:3676", + "right": "pull_request:6017", + "accept": false, + "reason": "Same broad docs test theme, but the edits are different title strings in separate PRs, not one underlying change." + }, + { + "left": "pull_request:3676", + "right": "pull_request:6018", + "accept": false, + "reason": "These are separate docs-index title tweaks with different test strings, not a single duplicated patch." + }, + { + "left": "pull_request:4330", + "right": "pull_request:4472", + "accept": false, + "reason": "Different README files and different content; they are not the same bug or change." + }, + { + "left": "pull_request:4652", + "right": "pull_request:6647", + "accept": false, + "reason": "Unrelated test fixes in different areas; sharing the word \"fix\" is not enough to make them duplicates." + }, + { + "left": "pull_request:6017", + "right": "pull_request:6018", + "accept": false, + "reason": "Both are doc-builder test edits to the same file family, but they are distinct title changes and not the same patch." + }, + { + "left": "pull_request:61", + "right": "pull_request:63", + "accept": false, + "reason": "Different conversion scripts for different pipelines and checkpoints; not the same underlying work." + }, { "left": "pull_request:8470", "right": "pull_request:9760", "accept": false, - "reason": "Both are README updates, but they touch different content and different release eras; not the same change." + "reason": "Both touch README text, but they are separate documentation updates for different content." }, { "left": "pull_request:8478", "right": "pull_request:8479", "accept": true, - "reason": "Same HunyuanDiT controlnet/pipeline feature set, same files, and effectively the same code addition." + "reason": "These look like the same HunyuanDiT controlnet/pipeline addition duplicated almost verbatim." }, { "left": "pull_request:12117", "right": "pull_request:12225", "accept": false, - "reason": "Both are QwenImage additions, but one adds img2img/inpaint pipelines while the other adds a distinct edit-inpaint pipeline." + "reason": "Related QwenImage work, but one adds img2img/inpaint while the other adds edit-inpaint; different concrete code paths." }, { "left": "pull_request:327", "right": "pull_request:335", "accept": false, - "reason": "Type-hint cleanup, but one is a pipeline PR and the other is a scheduler PR; not the same concrete change." + "reason": "One is a pipeline type-hint refactor and the other is a scheduler type-hint refactor; not the same change." }, { "left": "pull_request:7903", "right": "pull_request:9131", "accept": false, - "reason": "Unrelated changes: video processor tests versus a tokenizers pin in setup.py." + "reason": "A video/image processor test refactor versus a dependency pin are unrelated changes." }, { "left": "pull_request:8064", "right": "pull_request:9131", "accept": false, - "reason": "Unrelated changes: Marigold tests versus a tokenizers pin." + "reason": "Marigold test changes and a setup dependency change are unrelated." }, { "left": "pull_request:6910", "right": "pull_request:6939", "accept": false, - "reason": "Both standardize model cards, but they target different training scripts and different models." + "reason": "Both standardize model-card code, but for different training scripts and different artifacts." }, { "left": "pull_request:325", "right": "pull_request:333", "accept": false, - "reason": "Both are type-hint PRs, but they cover different pipelines and code paths." + "reason": "Different pipelines, different files, and different type-hint refactors." }, { "left": "pull_request:359", "right": "pull_request:360", "accept": false, - "reason": "Related scheduler typing work, but 360 adds an extra scheduler and is not the same PR change." + "reason": "Both are scheduler annotations, but they target different schedulers and are separate changes." }, { "left": "pull_request:335", "right": "pull_request:349", "accept": false, - "reason": "Different scheduler type-hint updates for different classes; same style, not same bug/change." - }, - { - "left": "pull_request:6909", - "right": "pull_request:6939", - "accept": false, - "reason": "Same model-card standardization theme, but different training scripts and outputs." + "reason": "Type-hint updates for different schedulers; not the same concrete bug or patch." }, { "left": "pull_request:12521", "right": "pull_request:12562", "accept": false, - "reason": "Both are Wan-related fixes, but one is autoencoder tiling and the other is context-parallel crash handling." + "reason": "Different Wan crash fixes: one is tiling decode, the other is context-parallel handling." }, { "left": "pull_request:1131", "right": "pull_request:1490", "accept": false, - "reason": "Different release bumps and different follow-up changes; not the same version-bump PR." - }, - { - "left": "pull_request:1131", - "right": "pull_request:912", - "accept": false, - "reason": "Different version bumps to different dev releases." - }, - { - "left": "pull_request:13069", - "right": "pull_request:13082", - "accept": false, - "reason": "Different model test refactors for different architectures; not the same change." - }, - { - "left": "pull_request:13259", - "right": "pull_request:13567", - "accept": false, - "reason": "Both touch AGENTS docs, but they are separate documentation edits with different scope." - }, - { - "left": "pull_request:1415", - "right": "pull_request:2757", - "accept": false, - "reason": "Broad stable-diffusion cleanup versus a specific depth batching fix; not the same underlying issue." - }, - { - "left": "pull_request:1490", - "right": "pull_request:831", - "accept": false, - "reason": "Different version bumps across different releases, even though both touch scheduler defaults." + "reason": "Both are version/release-related, but they are different bumps with different accompanying code changes." }, { "left": "pull_request:2290", "right": "pull_request:2446", "accept": false, - "reason": "Both are pix2pix docs fixes, but they address different doc snippets and different mistakes." + "reason": "Separate docs snippet fixes for different examples; not the same issue." }, { "left": "pull_request:2388", "right": "pull_request:4612", "accept": false, - "reason": "One adds the controlling-generation doc; the other only fixes links inside it." - }, - { - "left": "pull_request:2539", - "right": "pull_request:3099", - "accept": false, - "reason": "Same kind of doc/default-value cleanup, but for different pipelines and different defaults." - }, - { - "left": "pull_request:2839", - "right": "pull_request:3525", - "accept": false, - "reason": "Both update README files, but the content and subject matter are unrelated." - }, - { - "left": "pull_request:3525", - "right": "pull_request:3819", - "accept": false, - "reason": "Both are README edits, but they are unrelated documents and different changes." - }, - { - "left": "pull_request:3709", - "right": "pull_request:3710", - "accept": true, - "reason": "Same image-processor documentation addition, with matching files and nearly identical content." - }, - { - "left": "pull_request:3819", - "right": "pull_request:5267", - "accept": false, - "reason": "Unrelated README/doc formatting changes; not the same underlying update." - }, - { - "left": "pull_request:3819", - "right": "pull_request:5855", - "accept": false, - "reason": "Unrelated README edits in different areas of the repo." - }, - { - "left": "pull_request:4451", - "right": "pull_request:5220", - "accept": false, - "reason": "Both move tests to nightly, but they are different batches of tests and not the same concrete change." - }, - { - "left": "pull_request:5280", - "right": "pull_request:5292", - "accept": false, - "reason": "Both speed up ControlNet tests, but they target different test files and different dummy setups." + "reason": "One adds the controlling-generation doc page, the other fixes links within it; related but not duplicate." } ] }, @@ -181443,8 +181170,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 250, - "review_unit_count": 273, + "review_unit_index": 249, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -181453,32 +181180,32 @@ "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10206", - "pull_request:10323", - "pull_request:10434", "pull_request:10810", - "pull_request:11879", - "pull_request:11895", - "pull_request:12490", - "pull_request:13255", - "pull_request:13293", - "pull_request:13313", - "pull_request:13338", - "pull_request:13352", - "pull_request:1508", "pull_request:1525", - "pull_request:1564", + "pull_request:2539", + "pull_request:2839", + "pull_request:3099", "pull_request:3214", "pull_request:326", "pull_request:327", "pull_request:333", "pull_request:335", "pull_request:343", + "pull_request:3525", "pull_request:359", "pull_request:360", + "pull_request:3709", + "pull_request:3710", + "pull_request:3819", "pull_request:4078", "pull_request:4132", - "pull_request:4188", + "pull_request:4451", + "pull_request:5220", + "pull_request:5267", + "pull_request:5280", + "pull_request:5292", "pull_request:5420", + "pull_request:5855", "pull_request:6327", "pull_request:6342", "pull_request:6346", @@ -181502,6 +181229,14 @@ "pull_request:9760" ], "soft_pairs": [ + "pull_request:2539|pull_request:3099", + "pull_request:2839|pull_request:3525", + "pull_request:3525|pull_request:3819", + "pull_request:3709|pull_request:3710", + "pull_request:3819|pull_request:5267", + "pull_request:3819|pull_request:5855", + "pull_request:4451|pull_request:5220", + "pull_request:5280|pull_request:5292", "pull_request:6342|pull_request:6438", "pull_request:6453|pull_request:6462", "pull_request:6617|pull_request:6620", @@ -181519,193 +181254,192 @@ "pull_request:1525|pull_request:3214", "pull_request:5420|pull_request:6346", "pull_request:7246|pull_request:7825", - "pull_request:327|pull_request:333", - "pull_request:13255|pull_request:13313", - "pull_request:4188|pull_request:6346", - "pull_request:10323|pull_request:10434", - "pull_request:11879|pull_request:11895", - "pull_request:12490|pull_request:13293", - "pull_request:13338|pull_request:13352", - "pull_request:1508|pull_request:1564" + "pull_request:327|pull_request:333" ], - "prepared_review_unit_hash": "33125b1f12908b22cbbd4030a09d2cf386eece47d8d2f6fbb0fb714bd919e988", + "prepared_review_unit_hash": "1863e2ec5701a907aaa9ce1654ae50d4fc5af4fc64e710efcb9b112bbad2709d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 94963, - "estimated_input_tokens": 23741, - "estimated_eval_tokens": 47738 + "soft_pair_count": 26, + "serialized_chars": 96056, + "estimated_input_tokens": 24014, + "estimated_eval_tokens": 48284 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a grab bag of small docs/tests/LoRA follow-ups. The only clear duplicate-style pairs are the installation.md Windows note, the aMUSEd paper-link correction, and the README Colab-link follow-up; most other pairs are related but distinct cleanup or feature changes.", - "confidence": 0.64, + "summary": "Mostly a heterogeneous mix of unrelated docs, tests, version bumps, and small code fixes. Only a few pairs are true duplicates around the exact same documentation change; most other soft edges should be rejected under strict PR triage.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "6438 is the best representative: merged, explicitly issue-linked, and the clearest complete fix for the repeated installation-doc Windows/Python note.", + "canonical_pr_reason": "No single canonical PR fits the whole cluster: the items span unrelated edits across docs, tests, schedulers, LoRA utilities, and examples. The cluster is too mixed for one representative PR to safely stand in for the rest.", "best_issue_reason": null, - "best_pr_reason": "6438 is the strongest cluster anchor overall because it is the merged, issue-backed doc fix with the most complete scope among the accepted duplicates.", + "best_pr_reason": "No strong global best PR; if forced, the cleanest representative is one of the accepted doc-fix PRs, but the overall cluster is not cohesive enough to justify a single canonical PR.", "soft_edge_verdicts": [ + { + "left": "pull_request:2539", + "right": "pull_request:3099", + "accept": false, + "reason": "Same broad theme of default-value docs, but they touch different pipeline files and different parameters; not the same underlying fix." + }, + { + "left": "pull_request:2839", + "right": "pull_request:3525", + "accept": false, + "reason": "Both update README content, but they are different files and different changes (examples table vs general README copy)." + }, + { + "left": "pull_request:3525", + "right": "pull_request:3819", + "accept": false, + "reason": "Unrelated README edits in different locations; one is a general README rewrite, the other adds an FAQ section in examples." + }, + { + "left": "pull_request:3709", + "right": "pull_request:3710", + "accept": true, + "reason": "Same new documentation page and same toctree wiring; this is the same underlying doc addition with minor follow-up wording changes." + }, + { + "left": "pull_request:3819", + "right": "pull_request:5267", + "accept": false, + "reason": "Different documentation areas and different intent; FAQ expansion in examples is not the same change as README formatting guidance." + }, + { + "left": "pull_request:3819", + "right": "pull_request:5855", + "accept": false, + "reason": "Different README and different content; one is a FAQ addition, the other is a typo fix in a community example." + }, + { + "left": "pull_request:4451", + "right": "pull_request:5220", + "accept": false, + "reason": "Both retag tests, but they affect different test suites and different subsets; not the same concrete change." + }, + { + "left": "pull_request:5280", + "right": "pull_request:5292", + "accept": false, + "reason": "Both speed up ControlNet-related tests, but they target different test files and different setups, so they are not the same fix." + }, { "left": "pull_request:6342", "right": "pull_request:6438", "accept": true, - "reason": "Same installation.md Windows/Python support note; 6438 is the cleaner merged correction of the same doc issue." + "reason": "Same installation doc and same Windows/PyTorch version-compatibility note; this is effectively the same doc fix with refined wording." }, { "left": "pull_request:6453", "right": "pull_request:6462", "accept": true, - "reason": "Same aMUSEd paper-link update in the same doc page; 6462 corrects the same citation target." + "reason": "Same file and same aMUSEd paper-link correction; the second PR updates the exact same documentation bug." }, { "left": "pull_request:6617", "right": "pull_request:6620", "accept": false, - "reason": "Insufficient evidence of the same change; these look like separate docs wording fixes rather than one duplicate patch." + "reason": "Insufficient evidence on the left artifact and the visible right diff is a docs-arguments fix; not enough to conclude the same change." }, { "left": "pull_request:7118", "right": "pull_request:7451", "accept": false, - "reason": "Both are docs cleanup sweeps, but they cover different files and different typo sets, so not the same underlying fix." + "reason": "Both are typo/doc cleanup PRs, but they span different files and broad documentation sets, so they are not one concrete change." }, { "left": "pull_request:7544", "right": "pull_request:7827", "accept": false, - "reason": "Different code paths: CPU offload test assertions vs a docstring example fix." + "reason": "Different bug classes: CPU offload test assertions versus a docstring example fix." }, { "left": "pull_request:7767", "right": "pull_request:7966", "accept": false, - "reason": "One changes PixArt pipeline code, the other updates PixArt Sigma test expectations; not the same bug or change." + "reason": "One changes PixArt pipeline code, the other updates PixArt Sigma tests; related area but not the same underlying bug or patch." }, { "left": "pull_request:831", "right": "pull_request:912", "accept": false, - "reason": "These are version/type-hint housekeeping PRs, not one concrete duplicate change." + "reason": "Both are versioning/type-hint-era maintenance, but they correspond to different release bumps and different scheduler edits; not the same PR-worthy change." }, { "left": "pull_request:8775", "right": "pull_request:9760", - "accept": true, - "reason": "Same README Colab-link fix in the same file; 9760 is a follow-up correction of the same broken link." + "accept": false, + "reason": "Same top-level README only in the broadest sense; the content changes are unrelated." }, { "left": "pull_request:6327", "right": "pull_request:6356", "accept": false, - "reason": "Same script, but the scopes differ: 6327 includes broader LoRA-script edits while 6356 targets SNR gamma/prior-preservation behavior." + "reason": "Overlapping advanced SDXL LoRA script area, but they fix different issues and include different code paths; too broad to merge as duplicates." }, { "left": "pull_request:326", "right": "pull_request:335", "accept": false, - "reason": "Related type-hint work, but 326 is a broad sweep and 335 is a separate scheduler-specific PR." + "reason": "Both add typing to PNDM-related code, but one is pipeline typing and the other is scheduler typing; not the same change." }, { - "left": "pull_request:343", - "right": "pull_request:359", + "left": "pull_request:4078", + "right": "pull_request:4132", "accept": false, - "reason": "Both are type-hint PRs, but they touch different scheduler files and are not the same concrete change." + "reason": "Opposite behaviors in the same file: one forces accelerate to be present, the other makes accelerate optional." }, { "left": "pull_request:343", - "right": "pull_request:360", + "right": "pull_request:359", "accept": false, - "reason": "Different files and different scheduler-type-hint scope; not mergeable as one duplicate fix." + "reason": "Both are scheduler typing cleanups, but they affect different schedulers and code paths." }, { - "left": "pull_request:4078", - "right": "pull_request:4132", + "left": "pull_request:343", + "right": "pull_request:360", "accept": false, - "reason": "Same file, but opposite behavior: one forces accelerate to be installed, the other makes accelerate optional." + "reason": "Same as above; different schedulers, so not the same underlying fix." }, { "left": "pull_request:10206", "right": "pull_request:10810", "accept": false, - "reason": "Both touch LoRA utilities, but they address different Flux/Kohya conversion problems and different code paths." + "reason": "Both are LoRA/Flux-related, but one is unload logic and the other is conversion support; related subsystem, different concrete changes." }, { "left": "pull_request:1525", "right": "pull_request:3214", "accept": false, - "reason": "Both mention batch sizing, but one is the upscaling pipeline and the other is attention head reshaping; different bugs." + "reason": "Different bugs in different components: batch sizing in the upscaler versus attention head reshaping." }, { "left": "pull_request:5420", "right": "pull_request:6346", "accept": false, - "reason": "Different training-doc/example updates; they do not fix the same underlying issue." + "reason": "Both touch training examples, but they address different example scripts and different behaviors; not a single mergeable fix." }, { "left": "pull_request:7246", "right": "pull_request:7825", "accept": false, - "reason": "Test-only changes in different areas: VAE tiling vs model device-map/offload behavior." + "reason": "Different test scopes and different failure modes; one is VAE tiling, the other is device-map/offload coverage." }, { "left": "pull_request:327", "right": "pull_request:333", "accept": false, - "reason": "Separate type-hint PRs for different pipelines; not the same change." - }, - { - "left": "pull_request:13255", - "right": "pull_request:13313", - "accept": false, - "reason": "Same LoRA area, but one fixes missing alpha handling and the other adds Flux2/Kohya conversion support; distinct code paths." - }, - { - "left": "pull_request:4188", - "right": "pull_request:6346", - "accept": false, - "reason": "Different training-example fixes and different files; these are not the same underlying bug." - }, - { - "left": "pull_request:10323", - "right": "pull_request:10434", - "accept": false, - "reason": "Both fix broken docs links, but 10434 is a broader multi-file sweep rather than the same single-link change." - }, - { - "left": "pull_request:11879", - "right": "pull_request:11895", - "accept": false, - "reason": "Different deprecated test removals in different pipeline sets; not one duplicate change." - }, - { - "left": "pull_request:12490", - "right": "pull_request:13293", - "accept": false, - "reason": "One is a test-only ClapConfig adjustment, the other changes Audioldm2 pipeline cache-position logic; different problems." - }, - { - "left": "pull_request:13338", - "right": "pull_request:13352", - "accept": false, - "reason": "Workflow permission change vs adding a checkout step; related file, but different CI fixes." - }, - { - "left": "pull_request:1508", - "right": "pull_request:1564", - "accept": false, - "reason": "Both are DreamBooth docs, but they address different topics: prior-preservation clarification vs default accelerate config." + "reason": "Both are typing updates for different pipelines, but they are separate code paths and not one duplicate change." } ] }, @@ -181720,8 +181454,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 251, - "review_unit_count": 273, + "review_unit_index": 250, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -181730,19 +181464,31 @@ "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10319", + "pull_request:10323", + "pull_request:10434", "pull_request:11257", "pull_request:11701", + "pull_request:11879", + "pull_request:11895", + "pull_request:12490", "pull_request:12675", "pull_request:12883", "pull_request:1291", "pull_request:13254", + "pull_request:13255", + "pull_request:13293", + "pull_request:13313", "pull_request:13315", "pull_request:13335", + "pull_request:13338", "pull_request:13340", "pull_request:13341", + "pull_request:13352", "pull_request:1415", "pull_request:1423", "pull_request:1490", + "pull_request:1508", + "pull_request:1564", "pull_request:178", "pull_request:1998", "pull_request:2406", @@ -181753,31 +181499,27 @@ "pull_request:3470", "pull_request:3472", "pull_request:3525", + "pull_request:4188", "pull_request:4285", - "pull_request:4351", - "pull_request:4472", - "pull_request:4574", "pull_request:4612", - "pull_request:5267", "pull_request:555", - "pull_request:5855", "pull_request:6119", - "pull_request:6314", - "pull_request:6462", - "pull_request:6647", - "pull_request:6837", + "pull_request:6346", "pull_request:7166", - "pull_request:7335", - "pull_request:7425", "pull_request:889", "pull_request:9040", "pull_request:912", - "pull_request:9249", "pull_request:9423", - "pull_request:9765", - "pull_request:9860" + "pull_request:9765" ], "soft_pairs": [ + "pull_request:13255|pull_request:13313", + "pull_request:4188|pull_request:6346", + "pull_request:10323|pull_request:10434", + "pull_request:11879|pull_request:11895", + "pull_request:12490|pull_request:13293", + "pull_request:13338|pull_request:13352", + "pull_request:1508|pull_request:1564", "pull_request:6119|pull_request:7166", "pull_request:9040|pull_request:9423", "pull_request:10319|pull_request:9765", @@ -181797,219 +181539,210 @@ "pull_request:2406|pull_request:4612", "pull_request:2569|pull_request:4285", "pull_request:325|pull_request:327", - "pull_request:3470|pull_request:3472", - "pull_request:3525|pull_request:4351", - "pull_request:4351|pull_request:5267", - "pull_request:4351|pull_request:5855", - "pull_request:4472|pull_request:4574", - "pull_request:555|pull_request:889", - "pull_request:6314|pull_request:6462", - "pull_request:6647|pull_request:7335", - "pull_request:6837|pull_request:7425", - "pull_request:9249|pull_request:9860" + "pull_request:3470|pull_request:3472" ], - "prepared_review_unit_hash": "276691bf4bfa79f87c752183de50c7939388a643369f99db716f9544db68eba2", + "prepared_review_unit_hash": "05f016a4814fbb95dac2d7bf1cbffa3e1331a7ae3328279c034292343d9f2843", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 29, - "serialized_chars": 92971, - "estimated_input_tokens": 23243, - "estimated_eval_tokens": 46742 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 96652, + "estimated_input_tokens": 24163, + "estimated_eval_tokens": 48582 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated docs/test/maintenance PRs; only the style-bot token workflow fix and the paired DreamBooth IF doc additions look like true near-duplicates.", - "confidence": 0.67, + "summary": "The set is highly heterogeneous: docs nits, workflow tweaks, test refactors, and unrelated bug fixes across different code paths. I don\u2019t see any soft pair that is truly the same underlying change, so I reject all duplicate links. If a single canonical PR is needed, #13255 is the cleanest concrete bug fix in a core utility.", + "confidence": 0.61, "canonical_issue_reason": null, - "canonical_pr_reason": "11701 is the cleanest final form of the shared style-bot token workflow fix; 11257 is an earlier variant of the same one-line change.", + "canonical_pr_reason": "#13255 is a small, focused code fix in a central loader utility, with a clear bug description and the strongest signal of a real standalone change among this mixed set.", "best_issue_reason": null, - "best_pr_reason": "11701 is merged, concrete, and represents the clearest canonical fix among the accepted duplicate-like pairs.", + "best_pr_reason": "#13255 is the best representative PR here because it is a precise, mergeable bug fix with minimal scope and clearer code-path impact than the docs/test/workflow-only PRs.", "soft_edge_verdicts": [ + { + "left": "pull_request:13255", + "right": "pull_request:13313", + "accept": false, + "reason": "Same file, but different LoRA-loading fixes and different bug shapes; not the same concrete change." + }, + { + "left": "pull_request:4188", + "right": "pull_request:6346", + "accept": false, + "reason": "Both touch training examples, but they fix different issues in different scripts and are not one mergeable bug fix." + }, + { + "left": "pull_request:10323", + "right": "pull_request:10434", + "accept": false, + "reason": "Both are docs-link cleanups, but they target different pages and different broken-link sets." + }, + { + "left": "pull_request:11879", + "right": "pull_request:11895", + "accept": false, + "reason": "Both remove deprecated tests, but they are separate cleanup batches for different pipelines." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13293", + "accept": false, + "reason": "These affect different files and different bugs; one is a test fixture change, the other is pipeline cache-position logic." + }, + { + "left": "pull_request:13338", + "right": "pull_request:13352", + "accept": false, + "reason": "Same workflow file, but one changes permissions and the other adds a checkout step; distinct fixes." + }, + { + "left": "pull_request:1508", + "right": "pull_request:1564", + "accept": false, + "reason": "Both are DreamBooth docs edits, but they cover unrelated documentation topics." + }, { "left": "pull_request:6119", "right": "pull_request:7166", "accept": false, - "reason": "Same example file, but different bugs: fp16 gradient scaling vs a loading/state-dict issue. Not the same underlying change." + "reason": "Both are SDXL LoRA-related, but they fix different code paths: gradient scaling vs loading state dict into the correct text encoder." }, { "left": "pull_request:9040", "right": "pull_request:9423", "accept": false, - "reason": "Both are differential-diffusion community additions, but they target different models/pipelines (HunyuanDiT vs Kolors) and are separate features." + "reason": "Different community pipelines and different features; not the same underlying change." }, { "left": "pull_request:10319", "right": "pull_request:9765", "accept": false, - "reason": "Both are docstring fixes, but for different pipelines and different text; not the same bug or edit." + "reason": "Different pipeline docs and different docstring mistakes; no shared bug/change." }, { "left": "pull_request:11257", "right": "pull_request:11701", - "accept": true, - "reason": "Same workflow file and same bot_token line; both are variants of the style-bot token swap, with 11701 looking like the corrected final secret name." + "accept": false, + "reason": "Both edit the style-bot workflow token, but they point to different secrets and are not the same exact fix." }, { "left": "pull_request:12675", "right": "pull_request:12883", "accept": false, - "reason": "Both are docs typo fixes, but for unrelated pages and different API names; not a duplicate." + "reason": "Both are typo fixes in docs, but for unrelated quantization pages and different strings." }, { "left": "pull_request:1291", "right": "pull_request:1415", "accept": false, - "reason": "One fixes a notebook link; the other is a broad stable-diffusion cleanup. Same area, but not the same change." + "reason": "One fixes a single README link; the other is a broad stable-diffusion cleanup PR." }, { "left": "pull_request:13254", "right": "pull_request:13340", "accept": false, - "reason": "Different transformer test refactors for different models; they do not fix the same concrete problem." + "reason": "Different transformer models and different test refactors; not the same bug or change." }, { "left": "pull_request:13315", "right": "pull_request:13340", "accept": false, - "reason": "Different model test refactors; same subsystem but distinct changes." + "reason": "Different model test refactors with no shared concrete code-path problem." }, { "left": "pull_request:13335", "right": "pull_request:13340", "accept": false, - "reason": "Cosmos vs SD3 transformer test refactors are separate changes, not duplicates." + "reason": "Different transformer test suites; these are separate refactors." }, { "left": "pull_request:13340", "right": "pull_request:13341", "accept": false, - "reason": "SD3 vs Bria transformer test refactors are different model-specific changes." + "reason": "Different models and test code; not mergeable as one duplicate fix." }, { "left": "pull_request:1415", "right": "pull_request:309", "accept": false, - "reason": "Both touch stable-diffusion docs, but one is a broad cleanup and the other is a specific notebook link fix." + "reason": "Both mention Stable Diffusion docs, but one is a broad cleanup and the other is a specific notebook link fix." }, { "left": "pull_request:1423", "right": "pull_request:555", "accept": false, - "reason": "Stable Diffusion 2 Flax support and Flax VAE addition are related feature work, not the same fix." + "reason": "One adds SD2 Flax support, the other adds a Flax VAE; separate features, not duplicates." }, { "left": "pull_request:1490", "right": "pull_request:912", "accept": false, - "reason": "Both bump versions, but one also adds deprecations and wider scheduler changes; not the same PR-worthy change." + "reason": "Both are version/deprecation releases, but for different versions and unrelated release work." }, { "left": "pull_request:178", "right": "pull_request:555", "accept": false, - "reason": "Stable diffusion pipeline tests versus Flax VAE implementation are unrelated." + "reason": "Different artifacts and different scope: a pipeline test vs a Flax model implementation." }, { "left": "pull_request:178", "right": "pull_request:889", "accept": false, - "reason": "One adds tests, the other adds a components API; different changes." + "reason": "One is a pipeline test, the other adds components support; unrelated changes." }, { "left": "pull_request:1998", "right": "pull_request:3525", "accept": false, - "reason": "DreamBooth blog/docs content and a generic README rewrite are not the same change." + "reason": "Both touch README.md, but they are unrelated content changes." }, { "left": "pull_request:2406", "right": "pull_request:4612", "accept": false, - "reason": "Both edit the controlling-generation docs, but they address different issues: guidance wording vs broken links." + "reason": "Both are docs in the controlling-generation area, but they fix different text and link issues." }, { "left": "pull_request:2569", "right": "pull_request:4285", "accept": false, - "reason": "OpenVINO docs and ONNX/OpenVINO SDXL docs are separate documentation additions." + "reason": "Different optimization docs and different runtime backends; not the same change." }, { "left": "pull_request:325", "right": "pull_request:327", "accept": false, - "reason": "Type-hint updates for different pipelines; not the same concrete fix." + "reason": "Different pipelines and different type-hint updates; not the same bug fix." }, { "left": "pull_request:3470", "right": "pull_request:3472", - "accept": true, - "reason": "Both add the same IF DreamBooth documentation section to closely related docs; they look like two halves of the same doc update." - }, - { - "left": "pull_request:3525", - "right": "pull_request:4351", - "accept": false, - "reason": "Both touch README.md, but one is a generic rewrite and the other is a path fix; unrelated changes." - }, - { - "left": "pull_request:4351", - "right": "pull_request:5267", - "accept": false, - "reason": "Both are documentation edits, but for different files and different purposes." - }, - { - "left": "pull_request:4351", - "right": "pull_request:5855", - "accept": false, - "reason": "README path fix versus a community README import typo; different content." - }, - { - "left": "pull_request:4472", - "right": "pull_request:4574", - "accept": false, - "reason": "Both update README_sdxl.md, but one removes auth instructions and the other adds push_to_hub mentions; not the same issue." - }, - { - "left": "pull_request:555", - "right": "pull_request:889", - "accept": false, - "reason": "Flax VAE implementation and components API exposure are distinct changes." - }, - { - "left": "pull_request:6314", - "right": "pull_request:6462", - "accept": false, - "reason": "Different paper-link typos in different docs; no shared underlying change." - }, - { - "left": "pull_request:6647", - "right": "pull_request:7335", "accept": false, - "reason": "Different test files and different failures; both are test fixes, but not the same bug." + "reason": "Both add IF DreamBooth material, but one is the training docs and the other is the README; separate documentation updates." }, { - "left": "pull_request:6837", - "right": "pull_request:7425", + "left": "pull_request:13254", + "right": "pull_request:13340", "accept": false, - "reason": "I2VGenXL doc dependency ordering and a PEFT tutorial image addition are unrelated." + "reason": "Different model test refactors with no shared concrete code-path problem." }, { - "left": "pull_request:9249", - "right": "pull_request:9860", + "left": "pull_request:13335", + "right": "pull_request:13341", "accept": false, - "reason": "9860 extends the controlnet example to SD3.5, but 9249 is the initial SD3 example; related follow-up work, not the same change." + "reason": "Different transformer test suites; these are separate refactors." } ] }, @@ -182024,8 +181757,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 252, - "review_unit_count": 273, + "review_unit_index": 251, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -182033,37 +181766,27 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10005", "pull_request:10021", - "pull_request:10109", - "pull_request:10183", - "pull_request:10222", "pull_request:10330", "pull_request:10378", - "pull_request:10819", - "pull_request:10947", - "pull_request:1137", "pull_request:11428", - "pull_request:11431", - "pull_request:11522", "pull_request:11534", - "pull_request:11537", "pull_request:11913", "pull_request:11995", "pull_request:11997", "pull_request:12086", - "pull_request:12442", - "pull_request:12490", - "pull_request:12526", "pull_request:12660", "pull_request:12909", "pull_request:2464", "pull_request:2795", - "pull_request:3127", "pull_request:325", + "pull_request:3525", "pull_request:359", "pull_request:4262", + "pull_request:4351", "pull_request:4451", + "pull_request:4472", + "pull_request:4574", "pull_request:4597", "pull_request:4694", "pull_request:4749", @@ -182071,18 +181794,37 @@ "pull_request:5003", "pull_request:5093", "pull_request:5201", - "pull_request:5674", + "pull_request:5267", + "pull_request:555", "pull_request:5697", - "pull_request:6122", + "pull_request:5855", + "pull_request:6314", + "pull_request:6462", + "pull_request:6647", + "pull_request:6837", "pull_request:6907", "pull_request:6939", + "pull_request:7335", + "pull_request:7425", "pull_request:8755", "pull_request:8798", + "pull_request:889", "pull_request:8913", + "pull_request:9249", + "pull_request:9860", "pull_request:9893", "pull_request:9938" ], "soft_pairs": [ + "pull_request:3525|pull_request:4351", + "pull_request:4351|pull_request:5267", + "pull_request:4351|pull_request:5855", + "pull_request:4472|pull_request:4574", + "pull_request:555|pull_request:889", + "pull_request:6314|pull_request:6462", + "pull_request:6647|pull_request:7335", + "pull_request:6837|pull_request:7425", + "pull_request:9249|pull_request:9860", "pull_request:11913|pull_request:11997", "pull_request:4262|pull_request:4749", "pull_request:8798|pull_request:8913", @@ -182098,188 +181840,180 @@ "pull_request:5201|pull_request:5697", "pull_request:9893|pull_request:9938", "pull_request:10021|pull_request:10330", - "pull_request:2464|pull_request:2795", - "pull_request:11522|pull_request:12490", - "pull_request:12442|pull_request:12526", - "pull_request:11431|pull_request:11537", - "pull_request:5674|pull_request:6122", - "pull_request:1137|pull_request:3127", - "pull_request:10005|pull_request:10819", - "pull_request:10109|pull_request:10222", - "pull_request:10183|pull_request:10947" + "pull_request:2464|pull_request:2795" ], - "prepared_review_unit_hash": "ef0ae85d13198cdc6fd7808dc0810d700d92cd0a6543d4f157cacc16c5cd322f", + "prepared_review_unit_hash": "47f6ab87fce74ed74f7e0bdd38c0204b8ce774197182beec70c426c7803868b3", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 24, - "serialized_chars": 108616, - "estimated_input_tokens": 27154, - "estimated_eval_tokens": 54564 + "soft_pair_count": 25, + "serialized_chars": 102375, + "estimated_input_tokens": 25594, + "estimated_eval_tokens": 51444 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most links are broad topical overlaps rather than true duplicates. I only accepted pairs that share the same concrete bugfix/change: sharded checkpoint subfolder loading, LoRA fuse/unfuse behavior, and the Wan Animate continuation. The rest are feature follow-ups, docs/test housekeeping, or related-but-distinct pipeline/model work.", - "confidence": 0.69, + "summary": "This cluster is mostly heterogeneous feature/docs PRs, not a single duplicate set. Only two soft-edge pairs look like true same-change follow-ups: LoRA fuse/unfuse and sharded-checkpoint subfolder loading.", + "confidence": 0.79, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 12526 is the most complete Wan Animate artifact here: it explicitly continues #12442 and adds the final pipeline/docs/tests wiring for the same feature.", + "canonical_pr_reason": "No single canonical PR: the items span unrelated features and docs, so there isn\u2019t one PR that cleanly represents the whole cluster.", "best_issue_reason": null, - "best_pr_reason": "PR 12526 is the strongest canonical PR candidate overall because it is merged, self-contained, and the clearest complete implementation among the related items in this set.", + "best_pr_reason": "PR 8798 is the best anchor among the few real near-duplicates: it\u2019s a focused fix in the shared checkpoint-loading utility, has a direct follow-up pair, and includes test coverage.", "soft_edge_verdicts": [ + { + "left": "pull_request:3525", + "right": "pull_request:4351", + "accept": false, + "reason": "Both touch README.md, but they are unrelated docs edits with different goals and no shared bug/change." + }, + { + "left": "pull_request:4351", + "right": "pull_request:5267", + "accept": false, + "reason": "Different documentation files and different edits; not the same underlying change." + }, + { + "left": "pull_request:4351", + "right": "pull_request:5855", + "accept": false, + "reason": "Generic README cleanup versus a community example typo fix; unrelated." + }, + { + "left": "pull_request:4472", + "right": "pull_request:4574", + "accept": false, + "reason": "Different example READMEs with different content changes; not the same change." + }, + { + "left": "pull_request:555", + "right": "pull_request:889", + "accept": false, + "reason": "Flax VAE addition versus DiffusionPipeline components support are distinct features." + }, + { + "left": "pull_request:6314", + "right": "pull_request:6462", + "accept": false, + "reason": "Both are paper-link doc fixes, but for different docs and different papers; not duplicates." + }, + { + "left": "pull_request:6647", + "right": "pull_request:7335", + "accept": false, + "reason": "One updates SDXL k-diffusion test outputs; the other adjusts ControlNet SDXL config/docs. Different problems." + }, + { + "left": "pull_request:6837", + "right": "pull_request:7425", + "accept": false, + "reason": "Different docs fixes in unrelated pipelines/tutorials." + }, + { + "left": "pull_request:9249", + "right": "pull_request:9860", + "accept": false, + "reason": "Initial SD3 ControlNet example versus later SD3.5 support extension; related area, but not the same concrete change." + }, { "left": "pull_request:11913", "right": "pull_request:11997", "accept": false, - "reason": "Related Wan modular work, but this looks like a follow-up feature extension (I2V/FLF2V) rather than the same concrete change." + "reason": "Both are Wan modular-pipeline work, but one introduces the base pipeline and the other adds I2V/FLF2V support; separate feature changes." }, { "left": "pull_request:4262", "right": "pull_request:4749", "accept": false, - "reason": "Both are SDXL DreamBooth LoRA fixes, but they address different behaviors (general fixes vs checkpointing/state-dict handling) and are not the same PR-worthy change." + "reason": "Both touch the SDXL DreamBooth LoRA script, but they address different issues and code paths." }, { "left": "pull_request:8798", "right": "pull_request:8913", "accept": true, - "reason": "Same underlying bug: loading sharded checkpoints from a subfolder. The second PR refines the same `_get_checkpoint_shard_files` path for local/subfolder handling." - }, - { - "left": "pull_request:11995", - "right": "pull_request:12086", - "accept": false, - "reason": "Same modular Flux area, but one is base text-to-image modular support and the other adds img2img support; distinct feature work." + "reason": "Same checkpoint-loading bug family: both fix sharded checkpoint loading from subfolders in the shared hub/model-loading path." }, { "left": "pull_request:4451", "right": "pull_request:5093", "accept": false, - "reason": "Both move tests to nightly, but this is broad test-maintenance rather than the same concrete bug/change." + "reason": "Both move tests to nightly, but they affect different test suites and are not the same underlying issue." }, { "left": "pull_request:4833", "right": "pull_request:5003", "accept": true, - "reason": "Both fix the same LoRA fuse/unfuse behavior in the same code path; 5003 is effectively a follow-up/cleanup on the same bug." + "reason": "Same LoRA fuse/unfuse bug in the same code path and files; these are plausibly one merged fix." }, { "left": "pull_request:4597", "right": "pull_request:4694", "accept": false, - "reason": "Both are ControlNet SDXL related, but one is MultiControlNet support and the other is SDXL inpainting support; different features." + "reason": "Same broad ControlNet area, but one is MultiControlNet SDXL support and the other is SDXL inpainting support; distinct features." }, { "left": "pull_request:12660", "right": "pull_request:12909", "accept": false, - "reason": "Both touch context-parallel attention behavior, but they target different models/code paths (chronoedit vs WAN) and are not the same change." + "reason": "Both tweak parallel_config handling, but they fix different model processors and different code paths." }, { "left": "pull_request:6907", "right": "pull_request:6939", "accept": false, - "reason": "Both standardize model cards, but they affect different training scripts and are just parallel docs/metadata cleanup." + "reason": "Both standardize model cards, but they are separate scripts and separate doc/model-card updates." }, { "left": "pull_request:11428", "right": "pull_request:11534", "accept": false, - "reason": "Both are Hunyuan Video Framepack-related, but 11534 is a distinct F1 variant/follow-up rather than the same concrete feature change." + "reason": "Same family of Hunyuan/Framepack work, but the PRs target different variants and documentation additions rather than the same exact change." }, { "left": "pull_request:10378", "right": "pull_request:8755", "accept": false, - "reason": "Same SD3 DreamBooth LoRA script, but they fix different issues in different parts of the training flow." + "reason": "Both modify the SD3 DreamBooth LoRA script, but they solve different training/saving issues." }, { "left": "pull_request:325", "right": "pull_request:359", "accept": false, - "reason": "Both are type-hint cleanups, but for different files and unrelated code paths." + "reason": "Both are type-hint cleanups, but on unrelated files and APIs." }, { "left": "pull_request:5201", "right": "pull_request:5697", "accept": false, - "reason": "Both involve xformers/LoRA attention behavior, but they fix different concrete bugs in different attention-path logic." + "reason": "Both involve LoRA/xformers behavior, but they fix different mechanisms in different code paths." }, { "left": "pull_request:9893", "right": "pull_request:9938", "accept": false, - "reason": "Both update community SDXL reference docs/examples, but they are separate example pipelines and not the same change." + "reason": "Both update SDXL reference community docs, but they describe different pipelines and changes." }, { "left": "pull_request:10021", "right": "pull_request:10330", "accept": false, - "reason": "Same LTX Video family, but 10330 is a later version/update rather than the same implementation change." + "reason": "Same LTX video model family, but one introduces the feature and the other updates it to 0.9.1; related follow-up, not a duplicate." }, { "left": "pull_request:2464", "right": "pull_request:2795", "accept": false, - "reason": "Both touch the text-to-image LoRA training script, but they address different behaviors and are not the same bugfix." - }, - { - "left": "pull_request:11522", - "right": "pull_request:12490", - "accept": false, - "reason": "Different Audioldm2-related fixes: one updates pipeline code for transformers compatibility, the other adjusts a test config." - }, - { - "left": "pull_request:12442", - "right": "pull_request:12526", - "accept": true, - "reason": "12526 is explicitly a continuation of 12442 and covers the same Wan Animate feature, so they are the same underlying change." - }, - { - "left": "pull_request:11431", - "right": "pull_request:11537", - "accept": false, - "reason": "Both add torch.compile tests, but for different model families and test targets." - }, - { - "left": "pull_request:5674", - "right": "pull_request:6122", - "accept": false, - "reason": "Both are docs typos/cleanup, but they are unrelated documentation sweeps rather than one shared fix." - }, - { - "left": "pull_request:1137", - "right": "pull_request:3127", - "accept": false, - "reason": "Both mention safety-checker/img2img behavior, but one is a narrow dtype cast in a legacy pipeline and the other is a broader safety-checker tensor-handling fix." - }, - { - "left": "pull_request:10005", - "right": "pull_request:10819", - "accept": false, - "reason": "Both are docs updates, but they target different model families and unrelated missing-entry cleanup." - }, - { - "left": "pull_request:10109", - "right": "pull_request:10222", - "accept": false, - "reason": "Both add torch_xla support, but to different pipelines and code paths; not the same concrete fix." - }, - { - "left": "pull_request:10183", - "right": "pull_request:10947", - "accept": false, - "reason": "Both add single-file support, but for different model classes and converter mappings, so they are distinct features." + "reason": "Different training scripts and different example fixes; not the same change." } ] }, @@ -182294,8 +182028,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 253, - "review_unit_count": 273, + "review_unit_index": 252, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -182303,22 +182037,35 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10005", + "pull_request:10109", + "pull_request:10183", + "pull_request:10222", "pull_request:10278", "pull_request:10394", "pull_request:1041", "pull_request:10434", + "pull_request:10819", "pull_request:1087", + "pull_request:10947", "pull_request:1103", "pull_request:1104", + "pull_request:1137", + "pull_request:11431", "pull_request:11473", "pull_request:11505", + "pull_request:11522", + "pull_request:11537", "pull_request:11634", "pull_request:11857", "pull_request:121", "pull_request:1210", "pull_request:12272", "pull_request:1230", + "pull_request:12442", "pull_request:12487", + "pull_request:12490", + "pull_request:12526", "pull_request:12775", "pull_request:12856", "pull_request:1291", @@ -182326,33 +182073,28 @@ "pull_request:13340", "pull_request:1437", "pull_request:1467", - "pull_request:168", - "pull_request:178", - "pull_request:1937", "pull_request:1998", "pull_request:2271", "pull_request:2362", - "pull_request:2383", "pull_request:2396", - "pull_request:2446", - "pull_request:252", - "pull_request:2595", "pull_request:2757", "pull_request:2766", - "pull_request:2792", - "pull_request:2846", - "pull_request:2935", - "pull_request:3119", + "pull_request:3127", "pull_request:3166", - "pull_request:3217", "pull_request:353", - "pull_request:386", - "pull_request:3987", - "pull_request:3989", - "pull_request:9015", - "pull_request:948" + "pull_request:5674", + "pull_request:6122", + "pull_request:9015" ], "soft_pairs": [ + "pull_request:11522|pull_request:12490", + "pull_request:12442|pull_request:12526", + "pull_request:11431|pull_request:11537", + "pull_request:5674|pull_request:6122", + "pull_request:1137|pull_request:3127", + "pull_request:10005|pull_request:10819", + "pull_request:10109|pull_request:10222", + "pull_request:10183|pull_request:10947", "pull_request:10278|pull_request:11473", "pull_request:10278|pull_request:9015", "pull_request:10394|pull_request:11634", @@ -182370,228 +182112,192 @@ "pull_request:1291|pull_request:2757", "pull_request:13253|pull_request:13340", "pull_request:1437|pull_request:353", - "pull_request:1467|pull_request:2271", - "pull_request:1467|pull_request:2846", - "pull_request:168|pull_request:178", - "pull_request:178|pull_request:386", - "pull_request:1937|pull_request:2595", - "pull_request:1998|pull_request:252", - "pull_request:2271|pull_request:3989", - "pull_request:2362|pull_request:2792", - "pull_request:2383|pull_request:3217", - "pull_request:2446|pull_request:3987", - "pull_request:2846|pull_request:3989", - "pull_request:2935|pull_request:3119", - "pull_request:353|pull_request:948" + "pull_request:1467|pull_request:2271" ], - "prepared_review_unit_hash": "94c8362f5f3bab54d835a6d88b71b6add29af7ccafd1283ed4fe83bf2c29c715", + "prepared_review_unit_hash": "c852f9c8cd6002b9969422e5b06a763f3f5d941fcbea33359bdc0c045884d3c4", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 30, - "serialized_chars": 98104, - "estimated_input_tokens": 24526, - "estimated_eval_tokens": 49308 + "soft_pair_count": 26, + "serialized_chars": 102079, + "estimated_input_tokens": 25520, + "estimated_eval_tokens": 51296 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly superficial similarity among unrelated maintenance PRs: docs typos/link cleanups, test-flakiness fixes, and a few MPS-related test changes. The pairs generally share broad themes or filenames, but not the same concrete bug or change.", + "summary": "Mostly unrelated docs/test/infra fixes with one true continuation pair for Wan Animate. Accept only the Wan Animate continuation; the rest are too broad or cover different concrete changes.", "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 10434 is the broadest and clearest docs-maintenance representative here: it fixes broken links across many files and has an explicit issue target, unlike the narrower typo/test PRs.", + "canonical_pr_reason": "PR 12526 is the clearest representative: it explicitly continues #12442, targets the same Wan Animate implementation/docs/tests, and is the more complete follow-up.", "best_issue_reason": null, - "best_pr_reason": "PR 10434 is the best overall anchor for the cluster because it most clearly reflects a real, multi-file docs cleanup and is more representative than the many one-off typo or test adjustments.", + "best_pr_reason": "PR 12526 is the best overall anchor for this cluster because it is the largest substantive change here and the only pair with explicit continuation/overlap strong enough to justify duplicate-style grouping.", "soft_edge_verdicts": [ + { + "left": "pull_request:11522", + "right": "pull_request:12490", + "accept": false, + "reason": "Both are small docs edits, but one is an AudioLDM2 transformers-version compatibility fix and the other is a ClapConfig test cleanup; different problems and code paths." + }, + { + "left": "pull_request:12442", + "right": "pull_request:12526", + "accept": true, + "reason": "Same Wan Animate feature work: 12526 is explicitly a continuation of 12442 and overlaps heavily in files, tests, and implementation scope." + }, + { + "left": "pull_request:11431", + "right": "pull_request:11537", + "accept": false, + "reason": "Both add torch.compile-related tests, but they target different models and different test refactors; not the same concrete fix." + }, + { + "left": "pull_request:5674", + "right": "pull_request:6122", + "accept": false, + "reason": "Both are broad documentation typo passes, but they touch different docs sets and do not represent one shared change." + }, + { + "left": "pull_request:1137", + "right": "pull_request:3127", + "accept": false, + "reason": "Both involve safety checker behavior, but one fixes inpaint legacy dtype handling and the other fixes img2img safety-checker processing; distinct bugs." + }, + { + "left": "pull_request:10005", + "right": "pull_request:10819", + "accept": false, + "reason": "Both are docs updates, but one adds missing flux pipelines and the other adds LoRA loader entries; separate content." + }, + { + "left": "pull_request:10109", + "right": "pull_request:10222", + "accept": false, + "reason": "Both add torch_xla support, but in different pipelines and code paths; not mergeable as one PR." + }, + { + "left": "pull_request:10183", + "right": "pull_request:10947", + "accept": false, + "reason": "Both are single-file loading work, but they implement support for different models with different converters and tests." + }, { "left": "pull_request:10278", "right": "pull_request:11473", "accept": false, - "reason": "Same docs file, but one changes the TorchAo example and the other fixes an fp8 table typo; different fixes." + "reason": "Same docs file, but different torchao doc fixes: one updates the example, the other fixes fp8 granular quantization naming." }, { "left": "pull_request:10278", "right": "pull_request:9015", "accept": false, - "reason": "Both are docs PRs, but they target unrelated examples in different pipelines and fix different issues." + "reason": "Both are docs/example updates, but for unrelated components and unrelated issues." }, { "left": "pull_request:10394", "right": "pull_request:11634", "accept": false, - "reason": "Unrelated docs edits: peft-loading link text vs a TorchAO documentation link update." + "reason": "Both are docs link fixes, but they affect different guides and different link targets; no shared underlying change." }, { "left": "pull_request:1041", "right": "pull_request:2396", "accept": false, - "reason": "One fixes failing tests in pipelines; the other removes scheduler warnings and adjusts UniPC tests. Different underlying problems." + "reason": "Both touch tests, but one is general failing-test cleanup and the other is UniPC-specific warning/trained_beta test fixes." }, { "left": "pull_request:10434", "right": "pull_request:11505", "accept": false, - "reason": "Both touch docs links, but they clean up different broken-link sets and are not the same change." + "reason": "Both fix docs links, but they are separate link sets across different doc areas and not one code/doc change." }, { "left": "pull_request:1087", "right": "pull_request:1467", "accept": false, - "reason": "Both are test-fix PRs, but they address different failures in different pipeline areas." + "reason": "Both are test-stability PRs, but they patch different test suites and different failure modes; too broad to merge." }, { "left": "pull_request:1103", "right": "pull_request:3166", "accept": false, - "reason": "Both are docs-related, but one updates a model-card URL and the other fixes textual inversion docs usage." + "reason": "Both are textual-inversion-related docs/code updates, but one fixes a README URL and the other fixes loader usage docs; different scope." }, { "left": "pull_request:1104", "right": "pull_request:2362", "accept": false, - "reason": "One adds slow MPS test infrastructure; the other adds a skip decorator. Related theme, but not the same concrete change." + "reason": "One adds slow MPS tests, the other adds an MPS skip decorator; related area, but not the same fix." }, { "left": "pull_request:11505", "right": "pull_request:12487", "accept": false, - "reason": "Both are link-fix docs PRs, but they target different files and different broken links." + "reason": "Both are docs link-fix PRs, but they cover different documentation families and different broken links." }, { "left": "pull_request:11857", "right": "pull_request:12272", "accept": false, - "reason": "Flux docs reference edits are unrelated to the modular pipeline fix for default height/width handling." + "reason": "Both touch Flux docs/pipelines, but one fixes pipeline references while the other fixes modular pipeline height/width handling; different changes." }, { "left": "pull_request:121", "right": "pull_request:1998", "accept": false, - "reason": "Both touch README-style docs, but they add different examples and content with no shared bug/change." + "reason": "Both are README/example updates, but one adds usage examples to the main README and the other links a DreamBooth blog post; unrelated." }, { "left": "pull_request:1210", "right": "pull_request:1467", "accept": false, - "reason": "Cycle-diffusion scheduler/test warnings are unrelated to generic slow-test adjustments." + "reason": "Both are test fixes, but one is cycle-diffusion/scheduler reproducibility work and the other is stable-diffusion slow-test tolerance changes." }, { "left": "pull_request:1230", "right": "pull_request:2766", "accept": false, - "reason": "Both mention MPS tests, but one fixes generator handling and the other changes expected outputs for torch 2.0. Different issues." + "reason": "Both address MPS-related test behavior, but one fixes generator handling in DDIM/DDPM tests and the other changes UNet block expectations for torch 2.0." }, { "left": "pull_request:12775", "right": "pull_request:12856", "accept": false, - "reason": "Docs addition for Z-Image versus the actual Z-Image implementation PR; one documents, the other adds the feature." + "reason": "The docs PR and the implementation PR are about Z-Image, but they are not the same artifact or concrete change; one is documentation, the other is model implementation." }, { "left": "pull_request:1291", "right": "pull_request:2757", "accept": false, - "reason": "Stable Diffusion notebook link fix is unrelated to the depth batching bug." + "reason": "Both are Stable Diffusion-related, but one fixes a README link and the other fixes depth batching behavior; unrelated problems." }, { "left": "pull_request:13253", "right": "pull_request:13340", "accept": false, - "reason": "Both are test refactors, but for different models and code paths." + "reason": "Both refactor transformer tests, but for different models (Z-Image vs SD3) and different test mixins; not the same change." }, { "left": "pull_request:1437", "right": "pull_request:353", "accept": false, - "reason": "These are unrelated typo fixes in different source files." + "reason": "Both are typo fixes, but in different files and on different messages; no shared bug/change." }, { "left": "pull_request:1467", "right": "pull_request:2271", "accept": false, - "reason": "Both are slow-test adjustments, but they cover different pipelines and different flaky expectations." - }, - { - "left": "pull_request:1467", - "right": "pull_request:2846", - "accept": false, - "reason": "Both are slow-test fixes, but they address different test cases in different pipelines." - }, - { - "left": "pull_request:168", - "right": "pull_request:178", - "accept": false, - "reason": "Feature addition for Stable Diffusion is not the same change as adding tests for it." - }, - { - "left": "pull_request:178", - "right": "pull_request:386", - "accept": false, - "reason": "Stable Diffusion tests and Stable Diffusion docs are separate artifacts with different purposes." - }, - { - "left": "pull_request:1937", - "right": "pull_request:2595", - "accept": false, - "reason": "EMA extraction in conversion code is unrelated to scheduler switchability fixes." - }, - { - "left": "pull_request:1998", - "right": "pull_request:252", - "accept": false, - "reason": "Both are README docs, but they add different example content and context." - }, - { - "left": "pull_request:2271", - "right": "pull_request:3989", - "accept": false, - "reason": "Both are slow-test fixes, but for different model families and different failure modes." - }, - { - "left": "pull_request:2362", - "right": "pull_request:2792", - "accept": false, - "reason": "Both relate to MPS skipping, but one introduces the decorator and the other applies it in specific tests." - }, - { - "left": "pull_request:2383", - "right": "pull_request:3217", - "accept": false, - "reason": "Pix2Pix utility code and a generic pt-to-pil utility are unrelated changes." - }, - { - "left": "pull_request:2446", - "right": "pull_request:3987", - "accept": false, - "reason": "These are docs snippets for different pipelines with no shared bug or code path." - }, - { - "left": "pull_request:2846", - "right": "pull_request:3989", - "accept": false, - "reason": "Both are slow-test PRs, but they fix different pipelines and different expectations." - }, - { - "left": "pull_request:2935", - "right": "pull_request:3119", - "accept": false, - "reason": "RePaint scheduler docs and RePaint pipeline docs are separate documentation updates." - }, - { - "left": "pull_request:353", - "right": "pull_request:948", - "accept": false, - "reason": "Two unrelated typo fixes in different pipeline files." + "reason": "Both are slow-test maintenance, but they change different expected outputs in different test suites and were not one merged fix." } ] }, @@ -182606,8 +182312,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 254, - "review_unit_count": 273, + "review_unit_index": 253, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -182615,20 +182321,27 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10323", - "pull_request:11438", - "pull_request:11473", - "pull_request:11512", - "pull_request:11537", - "pull_request:11737", - "pull_request:12915", - "pull_request:13217", - "pull_request:326", - "pull_request:3286", - "pull_request:3313", - "pull_request:333", + "pull_request:1467", + "pull_request:168", + "pull_request:178", + "pull_request:1937", + "pull_request:1998", + "pull_request:2271", + "pull_request:2362", + "pull_request:2383", + "pull_request:2446", + "pull_request:252", + "pull_request:2595", + "pull_request:2792", + "pull_request:2846", + "pull_request:2935", + "pull_request:3119", + "pull_request:3217", + "pull_request:353", + "pull_request:386", "pull_request:395", - "pull_request:4099", + "pull_request:3987", + "pull_request:3989", "pull_request:4145", "pull_request:4175", "pull_request:4472", @@ -182636,35 +182349,39 @@ "pull_request:4540", "pull_request:4652", "pull_request:4797", - "pull_request:4862", "pull_request:5051", "pull_request:5065", "pull_request:5066", "pull_request:5332", "pull_request:5411", - "pull_request:5506", "pull_request:5844", - "pull_request:5982", - "pull_request:6065", "pull_request:6122", - "pull_request:6308", "pull_request:6775", "pull_request:679", "pull_request:6837", "pull_request:688", - "pull_request:7118", - "pull_request:7131", "pull_request:715", - "pull_request:7301", "pull_request:7425", "pull_request:7472", - "pull_request:7747", "pull_request:8470", "pull_request:8775", "pull_request:8923", - "pull_request:8946" + "pull_request:8946", + "pull_request:948" ], "soft_pairs": [ + "pull_request:1467|pull_request:2846", + "pull_request:168|pull_request:178", + "pull_request:178|pull_request:386", + "pull_request:1937|pull_request:2595", + "pull_request:1998|pull_request:252", + "pull_request:2271|pull_request:3989", + "pull_request:2362|pull_request:2792", + "pull_request:2383|pull_request:3217", + "pull_request:2446|pull_request:3987", + "pull_request:2846|pull_request:3989", + "pull_request:2935|pull_request:3119", + "pull_request:353|pull_request:948", "pull_request:395|pull_request:679", "pull_request:4145|pull_request:4175", "pull_request:4472|pull_request:4540", @@ -182678,197 +182395,192 @@ "pull_request:6775|pull_request:7425", "pull_request:688|pull_request:715", "pull_request:8470|pull_request:8775", - "pull_request:8923|pull_request:8946", - "pull_request:326|pull_request:333", - "pull_request:7118|pull_request:7131", - "pull_request:3286|pull_request:3313", - "pull_request:10323|pull_request:11473", - "pull_request:11438|pull_request:11737", - "pull_request:5982|pull_request:6065", - "pull_request:7301|pull_request:7747", - "pull_request:4099|pull_request:4862", - "pull_request:5506|pull_request:6308", - "pull_request:12915|pull_request:13217", - "pull_request:11512|pull_request:11537" + "pull_request:8923|pull_request:8946" ], - "prepared_review_unit_hash": "fc8ac2b2d3375ac4f2ab481b0c652e4997e2087200cc909a6961c78a656ab5ee", + "prepared_review_unit_hash": "5d3cff56a9eef1f5a3e26fea380874c399eb5abe209d6cefe26c933f84a54220", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 102197, - "estimated_input_tokens": 25550, - "estimated_eval_tokens": 51356 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 92869, + "estimated_input_tokens": 23218, + "estimated_eval_tokens": 46692 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: mostly unrelated docs, tests, feature additions, and compile-fix PRs. I found no true duplicate pair; the soft matches are mostly same-file or same-subsystem overlaps rather than the same concrete bug/change.", - "confidence": 0.97, + "summary": "Mostly unrelated maintenance/doc/test PRs. The only credible duplicate pair is the two resnet variable-renaming cleanups; everything else changes different files or fixes different problems.", + "confidence": 0.78, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 395 is the clearest representative of the only real duplicate pair: a self-contained refactor in src/diffusers/models/resnet.py that renames internal variables for readability.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "PR 395 is the best cluster representative because it is the broadest, cleanest code-only cleanup among the plausible duplicate pair and has the most complete local context.", "soft_edge_verdicts": [ { - "left": "pull_request:395", - "right": "pull_request:679", + "left": "pull_request:1467", + "right": "pull_request:2846", "accept": false, - "reason": "Both touch resnet.py, but they fix different cleanup spots: one renames _upsample_2d args, the other renames forward inputs. Same file, not the same change." + "reason": "Both are flaky-test fixes, but they target different pipelines/files and different failure modes; not the same concrete bug." }, { - "left": "pull_request:4145", - "right": "pull_request:4175", + "left": "pull_request:168", + "right": "pull_request:178", "accept": false, - "reason": "Unrelated PRs: a demo UnCLIP draft vs a ControlNet webdataset training example. No shared bug or code path." + "reason": "One adds the Stable Diffusion pipeline implementation, the other adds tests for it; related work, but not the same change or bug." }, { - "left": "pull_request:4472", - "right": "pull_request:4540", + "left": "pull_request:178", + "right": "pull_request:386", "accept": false, - "reason": "Both edit README_sdxl.md, but one removes a login note and the other adds a Colab notebook link. Different doc changes." + "reason": "Tests for Stable Diffusion vs docs for Stable Diffusion; different artifacts and different changes." }, { - "left": "pull_request:4495", - "right": "pull_request:5844", + "left": "pull_request:1937", + "right": "pull_request:2595", "accept": false, - "reason": "Different examples and goals: SDXL reference community pipeline vs Japanese SDXL reference docs. Not the same artifact or fix." + "reason": "EMA extraction in the conversion script vs scheduler config mapping are unrelated code paths." }, { - "left": "pull_request:4652", - "right": "pull_request:4797", + "left": "pull_request:1998", + "right": "pull_request:252", "accept": false, - "reason": "Both are LoRA-related, but one changes test serialization behavior and the other changes loader mapping logic. Different bugs." + "reason": "Both are README/docs edits, but they update different pages for different purposes and are not the same underlying change." }, { - "left": "pull_request:5051", - "right": "pull_request:6122", + "left": "pull_request:2271", + "right": "pull_request:3989", "accept": false, - "reason": "Both are docs typo PRs, but they address different documents and unrelated wording changes." + "reason": "Both fix slow tests, but in different pipelines and with different adjustments; not the same bug." }, { - "left": "pull_request:5065", - "right": "pull_request:5332", + "left": "pull_request:2362", + "right": "pull_request:2792", "accept": false, - "reason": "Same broad theme of typo cleanup, but 5065 targets attention/pipeline code paths while 5332 only fixes a comment in train_dreambooth_lora.py. Not the same change." + "reason": "One adds the skip_mps helper and applies it to some tests; the other just applies it to text-to-video tests. Related, but not a duplicate change." }, { - "left": "pull_request:5066", - "right": "pull_request:5332", + "left": "pull_request:2383", + "right": "pull_request:3217", "accept": false, - "reason": "Prompt2Prompt type-check cleanup vs a DreamBooth LoRA comment fix. Different files and unrelated code paths." + "reason": "Different utility additions in different files and for different pipelines; not the same fix." }, { - "left": "pull_request:5411", - "right": "pull_request:7472", + "left": "pull_request:2446", + "right": "pull_request:3987", "accept": false, - "reason": "A broken training-doc link fix is unrelated to an examples README link fix. No duplicate bug." + "reason": "Two unrelated docs snippet fixes for different pipelines." }, { - "left": "pull_request:6775", - "right": "pull_request:6837", + "left": "pull_request:2846", + "right": "pull_request:3989", "accept": false, - "reason": "DDPM scheduler doc parameter addition vs i2vgen docs dependency ordering. Different docs issues." + "reason": "Different slow-test fixes in different pipelines; same broad theme only." }, { - "left": "pull_request:6775", - "right": "pull_request:7425", + "left": "pull_request:2935", + "right": "pull_request:3119", "accept": false, - "reason": "DDPM scheduler docs vs PEFT inference docs; unrelated documentation fixes." + "reason": "Both touch RePaint docs, but they fix different doc snippets and are not the same concrete issue." }, { - "left": "pull_request:688", - "right": "pull_request:715", + "left": "pull_request:353", + "right": "pull_request:948", "accept": false, - "reason": "Both are import-guard fixes for no-torch environments, but they hit different modules and different concrete failures." + "reason": "Two unrelated typo fixes in different pipeline/model files." }, { - "left": "pull_request:8470", - "right": "pull_request:8775", + "left": "pull_request:395", + "right": "pull_request:679", + "accept": true, + "reason": "Both are small resnet.py refactors renaming single-letter variables in the same code path; they could plausibly be merged into one cleanup PR." + }, + { + "left": "pull_request:4145", + "right": "pull_request:4175", "accept": false, - "reason": "Unrelated README updates: MaPO project content vs SD3 Colab link." + "reason": "A draft UnCLIP pipeline toy change vs a ControlNet webdataset training example; unrelated." }, { - "left": "pull_request:8923", - "right": "pull_request:8946", + "left": "pull_request:4472", + "right": "pull_request:4540", "accept": false, - "reason": "CI/debug workflow changes vs a test expectation adjustment. Not the same issue." + "reason": "Both edit the same README_sdxl.md, but one removes a login note and the other adds a Colab link; separate documentation updates." }, { - "left": "pull_request:326", - "right": "pull_request:333", + "left": "pull_request:4495", + "right": "pull_request:5844", "accept": false, - "reason": "Both are type-hint PRs, but they affect different pipelines and different files. No shared bug/change." + "reason": "Different docs/examples additions for different SDXL references; not the same change." }, { - "left": "pull_request:7118", - "right": "pull_request:7131", + "left": "pull_request:4652", + "right": "pull_request:4797", "accept": false, - "reason": "Both are doc typo fixes, but on different pages and different content areas. No duplicate." + "reason": "One is a LoRA test tweak, the other changes the LoRA loader mapping logic; not the same bug or fix." }, { - "left": "pull_request:3286", - "right": "pull_request:3313", + "left": "pull_request:5051", + "right": "pull_request:6122", "accept": false, - "reason": "Both target torch.compile stability, but they fix different breakages in different code paths. Too broad to merge as one PR." + "reason": "Different documentation typo sweeps across different pages/files." }, { - "left": "pull_request:10323", - "right": "pull_request:11473", + "left": "pull_request:5065", + "right": "pull_request:6837", "accept": false, - "reason": "Quantization overview link fix vs a TorchAO FP8 docs typo. Different documentation problems." + "reason": "LoRA/attention-processor cleanup vs i2vgen docs dependency tweak; unrelated." }, { - "left": "pull_request:11438", - "right": "pull_request:11737", + "left": "pull_request:5066", + "right": "pull_request:5332", "accept": false, - "reason": "Both modify the hidream training script, but one adds aspect-ratio bucketing and the other adds DeepSpeed support. Distinct feature changes." + "reason": "One is a broader code cleanup in attention-related examples/files; the other is a DreamBooth LoRA comment typo. Different changes." }, { - "left": "pull_request:5982", - "right": "pull_request:6065", + "left": "pull_request:5411", + "right": "pull_request:7472", "accept": false, - "reason": "Both concern SDXL Turbo docs, but one adds docs and the other expands overview/module exposure. Related feature work, not the same concrete change." + "reason": "Both are broken-link fixes in docs, but they target different pages and links." }, { - "left": "pull_request:7301", - "right": "pull_request:7747", + "left": "pull_request:6775", + "right": "pull_request:6837", "accept": false, - "reason": "VAE scaling constant test fix vs VAE tiling test fix. Different test failures." + "reason": "Missing parameter docs in DDPM vs missing deps in I2VGen docs; unrelated." }, { - "left": "pull_request:4099", - "right": "pull_request:4862", + "left": "pull_request:6775", + "right": "pull_request:7425", "accept": false, - "reason": "SDXL prompt-embed bug vs ControlNet img2img height/width handling. Different runtime issues." + "reason": "Different docs fixes in different parts of the docs tree." }, { - "left": "pull_request:5506", - "right": "pull_request:6308", + "left": "pull_request:688", + "right": "pull_request:715", "accept": false, - "reason": "PEFT/xformers attention fix vs LoRA dummy-component test setup. Different problems, not mergeable as one fix." + "reason": "Both address importability without PyTorch, but in different modules and different failure points; too broad to treat as the same bug." }, { - "left": "pull_request:12915", - "right": "pull_request:13217", + "left": "pull_request:8470", + "right": "pull_request:8775", "accept": false, - "reason": "13217 builds on the LTX-2 pipeline work, but it adds LTX-2.3 support and new conversion/model changes. Follow-up feature, not duplicate." + "reason": "Different README updates for different research projects; not a duplicate change." }, { - "left": "pull_request:11512", - "right": "pull_request:11537", + "left": "pull_request:8923", + "right": "pull_request:8946", "accept": false, - "reason": "Both add torch.compile tests, but for different model classes and files. Same testing theme, different artifacts." + "reason": "CI/debug workflow changes vs a test expectation change; unrelated." } ] }, @@ -182883,8 +182595,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 255, - "review_unit_count": 273, + "review_unit_index": 254, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -182892,7 +182604,13 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10323", "pull_request:1087", + "pull_request:11438", + "pull_request:11473", + "pull_request:11512", + "pull_request:11537", + "pull_request:11737", "pull_request:1210", "pull_request:1218", "pull_request:12265", @@ -182900,48 +182618,52 @@ "pull_request:12526", "pull_request:12705", "pull_request:1277", + "pull_request:12915", + "pull_request:13217", "pull_request:1324", "pull_request:1510", "pull_request:1582", "pull_request:1583", "pull_request:1713", - "pull_request:2007", "pull_request:2608", - "pull_request:2637", - "pull_request:2642", - "pull_request:2647", - "pull_request:2684", - "pull_request:2685", - "pull_request:2686", - "pull_request:2687", - "pull_request:2688", - "pull_request:2690", - "pull_request:2730", - "pull_request:2740", - "pull_request:2741", - "pull_request:2742", - "pull_request:2746", - "pull_request:2749", - "pull_request:2751", - "pull_request:2752", - "pull_request:2862", - "pull_request:2863", - "pull_request:2864", - "pull_request:2865", - "pull_request:433", + "pull_request:326", + "pull_request:3286", + "pull_request:3313", + "pull_request:333", + "pull_request:4099", + "pull_request:4862", + "pull_request:5506", + "pull_request:5982", + "pull_request:6065", "pull_request:6216", "pull_request:6291", "pull_request:6299", "pull_request:6300", "pull_request:6301", + "pull_request:6308", "pull_request:6907", + "pull_request:7118", + "pull_request:7131", "pull_request:718", + "pull_request:7301", "pull_request:7613", "pull_request:774", + "pull_request:7747", "pull_request:9077", "pull_request:9739" ], "soft_pairs": [ + "pull_request:326|pull_request:333", + "pull_request:7118|pull_request:7131", + "pull_request:3286|pull_request:3313", + "pull_request:10323|pull_request:11473", + "pull_request:11438|pull_request:11737", + "pull_request:5982|pull_request:6065", + "pull_request:7301|pull_request:7747", + "pull_request:4099|pull_request:4862", + "pull_request:5506|pull_request:6308", + "pull_request:12915|pull_request:13217", + "pull_request:11512|pull_request:11537", "pull_request:6216|pull_request:6291", "pull_request:6907|pull_request:7613", "pull_request:12288|pull_request:12526", @@ -182956,266 +182678,192 @@ "pull_request:1210|pull_request:774", "pull_request:1218|pull_request:1713", "pull_request:12265|pull_request:12705", - "pull_request:1510|pull_request:1713", - "pull_request:2007|pull_request:433", - "pull_request:2637|pull_request:2730", - "pull_request:2642|pull_request:2647", - "pull_request:2642|pull_request:2684", - "pull_request:2642|pull_request:2685", - "pull_request:2642|pull_request:2686", - "pull_request:2642|pull_request:2687", - "pull_request:2642|pull_request:2688", - "pull_request:2642|pull_request:2690", - "pull_request:2642|pull_request:2740", - "pull_request:2642|pull_request:2741", - "pull_request:2642|pull_request:2742", - "pull_request:2642|pull_request:2746", - "pull_request:2642|pull_request:2749", - "pull_request:2642|pull_request:2751", - "pull_request:2642|pull_request:2752", - "pull_request:2642|pull_request:2862", - "pull_request:2642|pull_request:2863", - "pull_request:2642|pull_request:2864", - "pull_request:2642|pull_request:2865" + "pull_request:1510|pull_request:1713" ], - "prepared_review_unit_hash": "461b53f62bb711e12ed7415da1fa938d40702e8e94d365268b3aa95212e23875", + "prepared_review_unit_hash": "78724315c2b28bc4e6a022f134f920f94fab64bce028a5ef3e7ea010dfc1d95d", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 35, - "serialized_chars": 103145, - "estimated_input_tokens": 25787, - "estimated_eval_tokens": 51830 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 106361, + "estimated_input_tokens": 26591, + "estimated_eval_tokens": 53438 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of small docs/test/maintenance PRs. The only clear duplicate set is the identical \"Better train\" trio; the rest are similar-by-theme but not the same underlying change.", + "summary": "This set is mostly unrelated PRs across different docs, tests, and feature work. The only clear duplicate group is the identical text-to-image training PR trio (6299/6300/6301); SDXL Turbo docs (5982/6065) look like a plausible continuation of the same docs rollout, but most other soft edges are false positives.", "confidence": 0.9, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 6299 is the best canonical PR because it is one of the exact duplicate \"Better train\" submissions and cleanly represents the only true duplicate set in the cluster.", + "canonical_pr_reason": "PR 6299 is the best canonical PR candidate because it is part of the only exact-duplicate trio in the set (6299/6300/6301) with identical filenames and diff shape.", "best_issue_reason": null, - "best_pr_reason": "PR 6299 is the earliest exact match in the duplicate trio and has the same touched files/diff as 6300 and 6301, so it is the strongest representative PR here.", + "best_pr_reason": "PR 6299 is the strongest overall PR candidate here: it has an exact duplicate match and is the clearest representative of a duplicate cluster, unlike the other mostly unrelated PRs.", "soft_edge_verdicts": [ + { + "left": "pull_request:326", + "right": "pull_request:333", + "accept": false, + "reason": "Different pipelines and different type-hint cleanups; same theme, but not the same underlying change." + }, + { + "left": "pull_request:7118", + "right": "pull_request:7131", + "accept": false, + "reason": "Both are docs typo fixes, but they touch different documentation areas and correct different wording issues." + }, + { + "left": "pull_request:3286", + "right": "pull_request:3313", + "accept": false, + "reason": "Both target torch.compile compatibility, but they fix different code paths and different breakages." + }, + { + "left": "pull_request:10323", + "right": "pull_request:11473", + "accept": false, + "reason": "Different quantization docs changes: link formatting vs a torchao fp8 terminology correction." + }, + { + "left": "pull_request:11438", + "right": "pull_request:11737", + "accept": false, + "reason": "One adds aspect-ratio bucketing, the other changes DeepSpeed support behavior; not the same bug or feature." + }, + { + "left": "pull_request:5982", + "right": "pull_request:6065", + "accept": true, + "reason": "Both are part of the SDXL Turbo docs rollout and could plausibly be merged into one documentation PR." + }, + { + "left": "pull_request:7301", + "right": "pull_request:7747", + "accept": false, + "reason": "Unrelated test fixes: a VAE scaling constant vs a VAE tiling assertion." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4862", + "accept": false, + "reason": "Different SDXL bugs in different code paths: prompt embedding handling vs MultiControlNet img2img sizing." + }, + { + "left": "pull_request:5506", + "right": "pull_request:6308", + "accept": false, + "reason": "Different LoRA-related fixes; one is xformers/PEFT interaction, the other is dummy component test setup." + }, + { + "left": "pull_request:12915", + "right": "pull_request:13217", + "accept": false, + "reason": "Same family of LTX2 work, but these are different model-version additions and not the same concrete change." + }, + { + "left": "pull_request:11512", + "right": "pull_request:11537", + "accept": false, + "reason": "Different torch.compile test additions for different model classes." + }, { "left": "pull_request:6216", "right": "pull_request:6291", "accept": false, - "reason": "Same LoRA test file, but one fixes PEFT 0.7 key remapping while the other changes fusion/unfusion tolerance; different bugs." + "reason": "Different test fixes in the same file; one handles PEFT key remapping, the other changes a LoRA unfusion assertion." }, { "left": "pull_request:6907", "right": "pull_request:7613", "accept": false, - "reason": "Both are model-card related, but they target different training scripts and different card standardization work." + "reason": "Different model-card standardization work for different training scripts." }, { "left": "pull_request:12288", "right": "pull_request:12526", "accept": false, - "reason": "Both add video-generation pieces, but InfiniteTalk and Wan Animate are different pipelines/code paths, not one shared change." + "reason": "Both add new video-generation pipelines, but they are different models/features and not one mergeable fix." }, { "left": "pull_request:2608", "right": "pull_request:718", "accept": false, - "reason": "Both are typo/docs cleanups, but they touch different files and unrelated wording fixes." + "reason": "Both are typo fixes, but they address different text and different locations; too broad to treat as the same change." }, { "left": "pull_request:6299", "right": "pull_request:6300", "accept": true, - "reason": "Identical title, files, and diff; this is the same change duplicated across two PRs." + "reason": "Identical diff preview and filenames; these are effectively the same PR change." }, { "left": "pull_request:6299", "right": "pull_request:6301", "accept": true, - "reason": "Same exact patch and touched files as 6299; clearly the same PR content." + "reason": "Identical diff preview and filenames; same underlying change." }, { "left": "pull_request:6300", "right": "pull_request:6301", "accept": true, - "reason": "Exact same diff and file set; these are duplicate submissions of the same change." + "reason": "Identical diff preview and filenames; same underlying change." }, { "left": "pull_request:9077", "right": "pull_request:9739", "accept": false, - "reason": "Both are typo-fix PRs, but they touch different docs/code paths and are not the same underlying edit." + "reason": "Both are typo/docs PRs, but they touch unrelated docs and different content changes." }, { "left": "pull_request:1277", "right": "pull_request:1324", "accept": false, - "reason": "Both mention quantization, but one adds docs/workflow changes while the other updates the README for Optimum integration." + "reason": "One adds INT8 quantization documentation, the other adds an INT8 Stable Diffusion feature path; related topic but not the same change." }, { "left": "pull_request:1582", "right": "pull_request:1583", "accept": false, - "reason": "Both concern inpainting timestep handling, but one fixes RePaintScheduler math and the other fixes legacy inpaint masking; different code paths." + "reason": "Both touch repaint/inpaint logic, but they fix different scheduler/pipeline steps and are not the same bug." }, { "left": "pull_request:1087", "right": "pull_request:774", "accept": false, - "reason": "One is a slow-tests fix with pipeline/config changes; the other only adjusts test assertions/setup." + "reason": "Different slow-test fixes affecting different tests and code paths." }, { "left": "pull_request:1210", "right": "pull_request:774", "accept": false, - "reason": "CycleDiffusion scheduler/generator plumbing is a different fix from a generic pipeline test adjustment." + "reason": "Same as above: unrelated test/scheduler fixes." }, { "left": "pull_request:1218", "right": "pull_request:1713", "accept": false, - "reason": "Same converter script, but one adds scheduler support and the other is a small helper refactor; not duplicate." + "reason": "Both improve the conversion script, but they change different conversion behaviors and are not one concrete fix." }, { "left": "pull_request:12265", "right": "pull_request:12705", "accept": false, - "reason": "Different typos/docs PRs across unrelated files; not the same change." + "reason": "Both are typo PRs, but they are broad unrelated cleanup passes over different files." }, { "left": "pull_request:1510", "right": "pull_request:1713", "accept": false, - "reason": "Scheduler/conversion feature work versus conversion-script cleanup; not the same bug or patch." - }, - { - "left": "pull_request:2007", - "right": "pull_request:433", - "accept": false, - "reason": "Both are black-related maintenance, but they are separate formatting/dependency updates affecting different areas." - }, - { - "left": "pull_request:2637", - "right": "pull_request:2730", - "accept": false, - "reason": "Different documentation pages with unrelated wording fixes." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2647", - "accept": false, - "reason": "Separate docs pages and separate wording edits; not a duplicate." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2684", - "accept": false, - "reason": "Different docs pages; same broad typo theme is too weak to merge as one change." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2685", - "accept": false, - "reason": "Different docs pages and different sentence-level edits." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2686", - "accept": false, - "reason": "Different docs pages; these are independent prose fixes." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2687", - "accept": false, - "reason": "Different docs pages and unrelated wording changes." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2688", - "accept": false, - "reason": "Different docs pages; not the same underlying issue." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2690", - "accept": false, - "reason": "Different docs pages and distinct prose cleanup." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2740", - "accept": false, - "reason": "Different documentation areas; one is loading docs, the other is training docs." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2741", - "accept": false, - "reason": "Unrelated docs files and separate wording fixes." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2742", - "accept": false, - "reason": "Different docs pages; no shared underlying change." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2746", - "accept": false, - "reason": "Different docs pages and different typo fixes." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2749", - "accept": false, - "reason": "Different docs pages; only the general typo theme overlaps." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2751", - "accept": false, - "reason": "Separate docs pages with unrelated text edits." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2752", - "accept": false, - "reason": "Different conceptual docs and different sentence fixes." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2862", - "accept": false, - "reason": "Different docs pages and different local prose corrections." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2863", - "accept": false, - "reason": "Different docs pages; not one shared bug or change." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2864", - "accept": false, - "reason": "Different docs pages and different wording fixes." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2865", - "accept": false, - "reason": "Different docs pages; the overlap is only that both are small editorial edits." + "reason": "Different conversion-script changes with different scheduler/model support; not the same underlying change." } ] }, @@ -183230,8 +182878,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 256, - "review_unit_count": 273, + "review_unit_index": 255, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -183247,11 +182895,10 @@ "pull_request:10774", "pull_request:12004", "pull_request:12006", - "pull_request:12024", "pull_request:12915", - "pull_request:13039", - "pull_request:13042", "pull_request:13121", + "pull_request:2007", + "pull_request:2637", "pull_request:2642", "pull_request:2647", "pull_request:2684", @@ -183260,9 +182907,14 @@ "pull_request:2687", "pull_request:2688", "pull_request:2690", + "pull_request:2730", + "pull_request:2740", + "pull_request:2741", "pull_request:2742", + "pull_request:2746", "pull_request:2748", "pull_request:2749", + "pull_request:2751", "pull_request:2752", "pull_request:2862", "pull_request:2863", @@ -183271,24 +182923,40 @@ "pull_request:2870", "pull_request:2923", "pull_request:2954", - "pull_request:327", - "pull_request:365", "pull_request:3690", "pull_request:3695", + "pull_request:433", "pull_request:4695", "pull_request:5058", "pull_request:5151", - "pull_request:6952", - "pull_request:6978", "pull_request:7068", "pull_request:7246", "pull_request:7247", "pull_request:7411", - "pull_request:8572", "pull_request:9423", "pull_request:9597" ], "soft_pairs": [ + "pull_request:2007|pull_request:433", + "pull_request:2637|pull_request:2730", + "pull_request:2642|pull_request:2647", + "pull_request:2642|pull_request:2684", + "pull_request:2642|pull_request:2685", + "pull_request:2642|pull_request:2686", + "pull_request:2642|pull_request:2687", + "pull_request:2642|pull_request:2688", + "pull_request:2642|pull_request:2690", + "pull_request:2642|pull_request:2740", + "pull_request:2642|pull_request:2741", + "pull_request:2642|pull_request:2742", + "pull_request:2642|pull_request:2746", + "pull_request:2642|pull_request:2749", + "pull_request:2642|pull_request:2751", + "pull_request:2642|pull_request:2752", + "pull_request:2642|pull_request:2862", + "pull_request:2642|pull_request:2863", + "pull_request:2642|pull_request:2864", + "pull_request:2642|pull_request:2865", "pull_request:2642|pull_request:2870", "pull_request:2642|pull_request:2923", "pull_request:2647|pull_request:2684", @@ -183332,311 +183000,408 @@ "pull_request:2684|pull_request:2748", "pull_request:9423|pull_request:9597", "pull_request:10581|pull_request:10589", - "pull_request:3690|pull_request:3695", - "pull_request:327|pull_request:365", - "pull_request:12006|pull_request:12024", - "pull_request:7068|pull_request:8572", - "pull_request:13039|pull_request:13042", - "pull_request:6952|pull_request:6978" + "pull_request:3690|pull_request:3695" ], - "prepared_review_unit_hash": "2583f23b2791d4979c0cbc08d5804b7d138d2d58eddd48610d321a4c2f9050a6", + "prepared_review_unit_hash": "ac3cc3bd4dc987a1fab6b5688e0c7ee9516906704ed50f152f9cf60c853e894e", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 49, - "serialized_chars": 109663, - "estimated_input_tokens": 27416, - "estimated_eval_tokens": 55088 + "soft_pair_count": 64, + "serialized_chars": 114078, + "estimated_input_tokens": 28520, + "estimated_eval_tokens": 57296 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous cluster of unrelated docs cleanups, feature additions, and fixes. I only accept a couple of soft edges where the PRs appear to target the same concrete bug/change; the rest are too broad or clearly distinct.", - "confidence": 0.34, - "canonical_issue_reason": "No issue artifacts are present in this cluster.", - "canonical_pr_reason": "10537 is the strongest duplicate-style anchor: it contains the concrete Sana 4K PAG/pipeline fix, unlike the many unrelated docs or feature PRs.", - "best_issue_reason": "No issue artifacts are present in this cluster.", - "best_pr_reason": "10537 is the best representative of the only substantive duplicate-like subcluster; the remaining items are too mixed to support a single canonical PR for the whole set.", + "summary": "Mostly a noisy mixed cluster: many items are unrelated docs/typo or feature PRs that only share broad subsystem/file overlap. Only a few pairs look like true duplicates or near-duplicates: the black update, the Sana 4K pipeline fix, and the Kandinsky docstring example fix.", + "confidence": 0.74, + "canonical_issue_reason": null, + "canonical_pr_reason": "10537 is the clearest representative: it is a merged, concrete Sana 4K pipeline fix with an earlier matching PR (10517) that appears to be the same change.", + "best_issue_reason": null, + "best_pr_reason": "10537 is the best overall PR candidate because it fixes an actual code path in the Sana 4K pipeline, unlike the many docs-only or broad-sweep PRs in this cluster.", "soft_edge_verdicts": [ + { + "left": "pull_request:2007", + "right": "pull_request:433", + "accept": true, + "reason": "Same black-version bump / formatting sweep; 433 and 2007 are effectively the same maintenance change." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2730", + "accept": false, + "reason": "Both are doc edits, but they touch different pages and unrelated wording fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2647", + "accept": false, + "reason": "Different docs typo fixes in unrelated files; same style, not the same change." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2684", + "accept": false, + "reason": "Unrelated docs edits in different pages; no shared concrete bug or change." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2685", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2686", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2687", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2688", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2690", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2740", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2741", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2742", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2746", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2748", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2749", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2751", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2752", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2862", + "accept": false, + "reason": "Different docs edits in unrelated conceptual pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2863", + "accept": false, + "reason": "Different docs edits in unrelated API pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2864", + "accept": false, + "reason": "Different docs edits in unrelated API pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2865", + "accept": false, + "reason": "Different docs edits in unrelated API pages." + }, { "left": "pull_request:2642", "right": "pull_request:2870", "accept": false, - "reason": "Both are small docs edits; they are not the same concrete change or bug fix." + "reason": "Different docs edits in unrelated API pages." }, { "left": "pull_request:2642", "right": "pull_request:2923", "accept": false, - "reason": "Punctuation/wording cleanup in docs only; no shared underlying bug or change." + "reason": "Different docs edits in unrelated API pages." }, { "left": "pull_request:2647", "right": "pull_request:2684", "accept": false, - "reason": "Separate documentation typo fixes, not one underlying change." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2685", "accept": false, - "reason": "Both are minor prose edits; they do not fix the same issue." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2686", "accept": false, - "reason": "These are unrelated docs wording updates despite similar style." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2687", "accept": false, - "reason": "Different documentation sentence fixes; not duplicates." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2688", "accept": false, - "reason": "Separate docs wording changes, not the same bug/change." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2690", "accept": false, - "reason": "Different docs fixes in the same handbook area, but not the same concrete edit." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2742", "accept": false, - "reason": "General doc cleanup PRs; the text edits are distinct." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2749", "accept": false, - "reason": "Both are typo/wording fixes, but they are different sentences and files." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2752", "accept": false, - "reason": "Not the same underlying documentation issue." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2862", "accept": false, - "reason": "Different docs corrections; no single shared change." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2863", "accept": false, - "reason": "Separate wording fixes in logging docs, not duplicates." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2864", "accept": false, - "reason": "A different docs sentence/punctuation fix, not the same patch." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2865", "accept": false, - "reason": "Minor prose cleanup only; not the same concrete change." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2870", "accept": false, - "reason": "Documentation cleanup in different files; not duplicate work." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2647", "right": "pull_request:2923", "accept": false, - "reason": "Same broad docs area, but different edits and no shared bug." + "reason": "Different docs pages and separate typo fixes." }, { "left": "pull_request:2684", "right": "pull_request:2685", "accept": false, - "reason": "Independent docs typo fixes; not the same change." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2686", "accept": false, - "reason": "Different sentence-level cleanup, not a duplicate." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2687", "accept": false, - "reason": "Both are docs edits, but they fix different text." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2688", "accept": false, - "reason": "Separate prose corrections; no shared code-path or issue." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2690", "accept": false, - "reason": "Not the same underlying documentation change." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2742", "accept": false, - "reason": "These are unrelated docs typos in different sections." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2749", "accept": false, - "reason": "Different wording fixes; not one duplicate patch." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2752", "accept": false, - "reason": "Both touch docs, but the actual edits differ." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2862", "accept": false, - "reason": "No shared concrete bug or edit." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2863", "accept": false, - "reason": "Different small docs corrections, not duplicates." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2865", "accept": false, - "reason": "Separate typo cleanup in another page." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2870", "accept": false, - "reason": "Not the same docs sentence or file." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:2684", "right": "pull_request:2923", "accept": false, - "reason": "Different docs wording fixes; too broad to merge as duplicates." + "reason": "Separate docs typo fixes, not the same underlying change." }, { "left": "pull_request:7068", "right": "pull_request:7411", "accept": false, - "reason": "Both are typo-focused PRs, but they touch different files and edits." + "reason": "Both are typo sweeps, but they touch different file sets and different documentation/code areas." }, { "left": "pull_request:10759", "right": "pull_request:10774", "accept": false, - "reason": "One adds a community pipeline, the other fixes a rendering bug in that pipeline; related, but not the same concrete change." + "reason": "One adds the community pipeline, the other fixes flattened image elements; related files but distinct changes." }, { "left": "pull_request:12915", "right": "pull_request:13121", "accept": false, - "reason": "A large LTX-2 feature add versus a specific inference bug fix; not the same change." + "reason": "New LTX2 video pipelines versus a later inference bug fix in the same family; not the same change." }, { "left": "pull_request:10517", "right": "pull_request:10537", "accept": true, - "reason": "Both are about the Sana 4K PAG/pipeline bug; 10537 contains the concrete fix in the same area, so these look like duplicate attempts at the same change." + "reason": "Same Sana 4K PAG pipeline bug fix; 10517 and 10537 target the same underlying issue." }, { "left": "pull_request:12004", "right": "pull_request:12006", "accept": false, - "reason": "Umbrella Wan2.2 work versus a specific i2v addition; related feature area, but not the same concrete patch." - }, - { - "left": "pull_request:12006", - "right": "pull_request:12024", - "accept": false, - "reason": "Same broader Wan2.2 effort, but 12024 is a follow-up fix rather than the same exact code-path change." + "reason": "Broad Wan2.2 rollout versus a specific 5B I2V addition; different scope and code paths." }, { "left": "pull_request:2954", "right": "pull_request:4695", "accept": false, - "reason": "TextToVideoZero versus the SDXL variant are different pipelines with different implementations." + "reason": "TextToVideoZero and TextToVideoZeroSDXLPipeline are related but separate features, not one duplicate fix." }, { "left": "pull_request:7246", "right": "pull_request:7247", "accept": false, - "reason": "One is a VAE tiling test fix, the other a config-checking test fix; not the same bug." + "reason": "Different test fixes: one is device handling for VAE tiling, the other config checking for single-file conversion." }, { "left": "pull_request:5058", "right": "pull_request:5151", "accept": false, - "reason": "Both concern PEFT/LoRA integration, but one targets the text encoder and the other the UNet; too broad to treat as duplicates." + "reason": "Both are PEFT-related, but one integrates the text encoder and the other integrates the UNet; distinct code paths." }, { "left": "pull_request:9423", "right": "pull_request:9597", "accept": false, - "reason": "Different community pipelines and different diffusion variants; not the same underlying change." + "reason": "Different community pipelines for different backends; not the same underlying bug or change." }, { "left": "pull_request:10581", "right": "pull_request:10589", "accept": false, - "reason": "Both add SD3 IP-Adapter support, but to different pipeline classes and code paths; related, not duplicate." + "reason": "Separate Stable Diffusion 3 pipeline variants with different IP-Adapter plumbing." }, { "left": "pull_request:3690", "right": "pull_request:3695", "accept": true, - "reason": "Both fix the same stale Kandinsky example code in the pipeline docs; 3695 broadens the same correction to the prior example, so they are mergeable as one change." - }, - { - "left": "pull_request:327", - "right": "pull_request:365", - "accept": false, - "reason": "Type-hint cleanup for different modules; not the same bug or patch." - }, - { - "left": "pull_request:13039", - "right": "pull_request:13042", - "accept": false, - "reason": "Unrelated fixes in different subsystems: layerwise upcasting CPU behavior versus prompt-isolation test handling." - }, - { - "left": "pull_request:6952", - "right": "pull_request:6978", - "accept": false, - "reason": "Different training scripts and different model-card handling changes; not duplicate work." + "reason": "Both correct Kandinsky docstring examples and usage snippets; they look like the same documentation fix family and could be merged." } ] }, @@ -183651,8 +183416,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 257, - "review_unit_count": 273, + "review_unit_index": 256, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -183664,9 +183429,12 @@ "pull_request:1087", "pull_request:11096", "pull_request:12004", + "pull_request:12006", "pull_request:12024", "pull_request:1210", "pull_request:1219", + "pull_request:13039", + "pull_request:13042", "pull_request:13275", "pull_request:13293", "pull_request:13442", @@ -183679,37 +183447,39 @@ "pull_request:2197", "pull_request:22", "pull_request:2208", - "pull_request:2271", - "pull_request:2290", "pull_request:2415", "pull_request:2464", - "pull_request:2540", - "pull_request:2587", "pull_request:260", "pull_request:2608", "pull_request:2624", "pull_request:3066", + "pull_request:327", "pull_request:3470", "pull_request:353", - "pull_request:3942", + "pull_request:365", "pull_request:396", "pull_request:407", - "pull_request:421", - "pull_request:4275", "pull_request:429", - "pull_request:432", "pull_request:4320", "pull_request:500", "pull_request:5552", - "pull_request:568", + "pull_request:6952", + "pull_request:6978", + "pull_request:7068", "pull_request:718", "pull_request:7627", "pull_request:774", "pull_request:798", "pull_request:8467", + "pull_request:8572", "pull_request:9760" ], "soft_pairs": [ + "pull_request:327|pull_request:365", + "pull_request:12006|pull_request:12024", + "pull_request:7068|pull_request:8572", + "pull_request:13039|pull_request:13042", + "pull_request:6952|pull_request:6978", "pull_request:2415|pull_request:2464", "pull_request:7627|pull_request:8467", "pull_request:5552|pull_request:7627", @@ -183734,249 +183504,216 @@ "pull_request:2175|pull_request:3470", "pull_request:2175|pull_request:429", "pull_request:2175|pull_request:500", - "pull_request:2197|pull_request:4320", - "pull_request:2271|pull_request:3066", - "pull_request:2271|pull_request:3942", - "pull_request:2290|pull_request:421", - "pull_request:2290|pull_request:4275", - "pull_request:2540|pull_request:432", - "pull_request:2587|pull_request:429", - "pull_request:2587|pull_request:500", - "pull_request:2608|pull_request:568", - "pull_request:2624|pull_request:500" + "pull_request:2197|pull_request:4320" ], - "prepared_review_unit_hash": "dba89d3716ff97da3c1b3bd70e7f724651b0c05ee934321b76ed16dcf385586b", + "prepared_review_unit_hash": "ff3743af661516ddb86abe46c8526694d3d9f29841a01b65059ff5b990503f4b", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 34, - "serialized_chars": 95972, - "estimated_input_tokens": 23993, - "estimated_eval_tokens": 48242 + "soft_pair_count": 30, + "serialized_chars": 95218, + "estimated_input_tokens": 23805, + "estimated_eval_tokens": 47866 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a mixed bag of unrelated docs/test typo fixes, with one genuine duplicate-like follow-up pair in Wan2.2. There isn\u2019t a single coherent cluster-wide bug; the only strong merge candidate is the Wan 2.2 follow-up pair.", - "confidence": 0.44, + "summary": "This cluster is mostly a grab bag of small maintenance PRs: typo/doc cleanup, test-only fixes, and a few unrelated model/pipeline patches. Only the Wan2.2 follow-up PRs look like they belong to the same concrete change set; the rest are too different to merge as duplicates.", + "confidence": 0.86, "canonical_issue_reason": null, - "canonical_pr_reason": "No issue is present. As a PR canonical, pull_request:12004 is the best anchor because it is the broadest substantive change in the set and the only one with a clear follow-up relationship to another PR in the cluster.", + "canonical_pr_reason": "PR 8572 is the broadest maintenance cleanup in the set, touching many docs/examples/code/test files and best matches the cluster\u2019s overall typo/maintenance character.", "best_issue_reason": null, - "best_pr_reason": "pull_request:12004 is the most representative substantive PR here: it introduces the Wan2.2 model/pipeline support and has the clearest concrete code-path relationship to its follow-up PR.", + "best_pr_reason": "PR 12004 is the most substantive standalone change here: a large Wan2.2 support PR spanning converter, core model/pipelines, and tests, so it is the strongest representative code change.", "soft_edge_verdicts": [ + { + "left": "pull_request:327", + "right": "pull_request:365", + "accept": false, + "reason": "Both are type-hint PRs, but they fix different pipelines/models and are unrelated changes." + }, + { + "left": "pull_request:12006", + "right": "pull_request:12024", + "accept": true, + "reason": "Same Wan2.2 image-to-video feature series; 12024 is a follow-up fix on the same pipeline/code path and could plausibly have been one PR." + }, + { + "left": "pull_request:7068", + "right": "pull_request:8572", + "accept": false, + "reason": "Both are typo/errata cleanups, but they target different files and different scopes of maintenance work." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13042", + "accept": false, + "reason": "Different bugs in different areas: Wan Animate CPU upcasting vs QwenImage prompt isolation tests." + }, + { + "left": "pull_request:6952", + "right": "pull_request:6978", + "accept": false, + "reason": "Both touch training/model-card code, but they are separate scripts and not the same underlying bug/change." + }, { "left": "pull_request:2415", "right": "pull_request:2464", "accept": false, - "reason": "Different problems: one adjusts pipeline loading/version handling, the other removes xFormers setup from a training example. Same broad tooling area, not the same change." + "reason": "One changes pipeline loading/version checks; the other removes xFormers enablement from a LoRA example. Not the same change." }, { "left": "pull_request:7627", "right": "pull_request:8467", "accept": false, - "reason": "Unrelated fixes in different pipelines: Playground img2img latent handling vs SEGA device selection/text encoding." + "reason": "Both are img2img pipeline fixes, but they address different concrete issues in different pipelines." }, { "left": "pull_request:5552", "right": "pull_request:7627", "accept": false, - "reason": "Both touch pipeline internals, but one is FreeU disable behavior and the other is Playground img2img latent scaling. Not the same bug." + "reason": "FreeU disable logic vs Playground img2img latent handling are unrelated code paths." }, { "left": "pull_request:12004", "right": "pull_request:12024", "accept": true, - "reason": "Clear follow-up on the same Wan2.2 code path: both touch Wan transformer/pipeline/test files and the later PR fixes issues in the feature introduced by the first. These could plausibly have been merged together." + "reason": "Both are Wan2.2 support changes; 12024 is a follow-up patch to the same transformer/pipeline/test work introduced by 12004." }, { "left": "pull_request:1041", "right": "pull_request:774", "accept": false, - "reason": "Both are test-focused, but they fix different failures: one adds model split metadata for LDM, the other relaxes callback tolerances and adjusts test plumbing." + "reason": "Both tweak tests, but they cover different failures and different pipelines." }, { "left": "pull_request:1087", "right": "pull_request:3066", "accept": false, - "reason": "Different test regressions in different areas: config extraction/inpaint handling versus layer utility expected values." + "reason": "Different test fixes in different areas; no shared concrete bug." }, { "left": "pull_request:11096", "right": "pull_request:9760", "accept": false, - "reason": "Both are README edits, but they are unrelated documentation wording changes with different scopes and targets." + "reason": "Both are README edits, but one is Flux TPU instructions and the other is a generic README wording update." }, { "left": "pull_request:1210", "right": "pull_request:3066", "accept": false, - "reason": "Cycle diffusion/scheduler noise handling is unrelated to layer utility test expectations." + "reason": "Both are test-adjustment PRs, but they fix different failures and unrelated code paths." }, { "left": "pull_request:1219", "right": "pull_request:396", "accept": false, - "reason": "Same topic name, but not the same concrete change: 396 adds DDIM docs and pipeline support, while 1219 is a later docs-only update." + "reason": "Same DDIM docs file, but these are not clearly the same underlying change; one is a later doc tweak while the other adds the docs page/code reference." }, { "left": "pull_request:13275", "right": "pull_request:13293", "accept": false, - "reason": "Different test files and different subsystems: Hunyuan video tests versus AudioLDM2 pipeline tests." + "reason": "Hunyuan test kwargs vs AudioLDM2 cache-position handling are unrelated." }, { "left": "pull_request:13275", "right": "pull_request:13442", "accept": false, - "reason": "Both are test fixes, but they target unrelated training/test setups and different model families." + "reason": "Both are test-related, but they fix different models/scripts and different issues." }, { "left": "pull_request:17", "right": "pull_request:718", "accept": false, - "reason": "Generic typo fixes in unrelated files; not the same underlying change." + "reason": "General typo fixes in separate documents; not the same artifact or bug." }, { "left": "pull_request:1724", "right": "pull_request:2208", "accept": false, - "reason": "Both concern xFormers docs, but they are separate documentation updates rather than one shared bug/change." + "reason": "Both concern xFormers docs, but one adds/reworks the section while the other updates installation details; too broad to treat as duplicates." }, { "left": "pull_request:21", "right": "pull_request:22", "accept": false, - "reason": "Both add tests, but for different models and different fixes. Same testing area only at a very broad level." + "reason": "Different model/test additions: UNet tests vs GLIDESuperResUNetModel tests." }, { "left": "pull_request:2137", "right": "pull_request:2608", "accept": false, - "reason": "Independent typo/docs fixes in different files with no shared code path." + "reason": "Unrelated typo fixes in loaders.py vs README.md." }, { "left": "pull_request:2137", "right": "pull_request:718", "accept": false, - "reason": "Different typo fixes in different docs/code files; no common underlying bug." + "reason": "Both are typo cleanups, but in different files and contexts." }, { "left": "pull_request:2138", "right": "pull_request:260", "accept": false, - "reason": "Both are small docs/wording edits, but they affect unrelated files and content." + "reason": "A docs typo change and a README typo change are not the same underlying change." }, { "left": "pull_request:2138", "right": "pull_request:353", "accept": false, - "reason": "A docs typo fix is unrelated to a UNet block error-message typo." + "reason": "Docs wording vs unet_blocks error-message typo are unrelated." }, { "left": "pull_request:2138", "right": "pull_request:407", "accept": false, - "reason": "No shared concrete fix: docs wording versus attention-slicing API typo." + "reason": "Docs wording vs attention-slicing typo are unrelated changes." }, { "left": "pull_request:2138", "right": "pull_request:798", "accept": false, - "reason": "Docs typo in loaders is unrelated to a UNet docstring correction." + "reason": "Docs wording vs unet docstring typo are different fixes." }, { "left": "pull_request:2175", "right": "pull_request:2624", "accept": false, - "reason": "Both are docs changes, but one adds components docs and the other adds cache_dir docs. Different API surface." + "reason": "Both are docs, but one adds a pipeline API entry while the other documents cache_dir; not the same change." }, { "left": "pull_request:2175", "right": "pull_request:3470", "accept": false, - "reason": "Unrelated documentation topics: diffusion pipeline API vs IF dreambooth usage docs." + "reason": "Different docs/features: pipeline API docs vs IF DreamBooth README additions." }, { "left": "pull_request:2175", "right": "pull_request:429", "accept": false, - "reason": "Different docs additions with no shared bug or change set." + "reason": "Docs API entry vs configuration docs are unrelated." }, { "left": "pull_request:2175", "right": "pull_request:500", "accept": false, - "reason": "Both are docs, but they document different configuration features and cannot be treated as one fix." + "reason": "Both are docs-related, but they document different configuration concepts and file sets." }, { "left": "pull_request:2197", "right": "pull_request:4320", "accept": false, - "reason": "Stable Diffusion inpaint typo fix is unrelated to ONNX docs wording." - }, - { - "left": "pull_request:2271", - "right": "pull_request:3066", - "accept": false, - "reason": "Different test failures in different modules; no shared code path." - }, - { - "left": "pull_request:2271", - "right": "pull_request:3942", - "accept": false, - "reason": "Both are test adjustments, but one updates legacy inpaint expected values while the other changes pipeline common xFormers test behavior." - }, - { - "left": "pull_request:2290", - "right": "pull_request:421", - "accept": false, - "reason": "Pix2Pix docs wording is unrelated to scheduler documentation cleanup." - }, - { - "left": "pull_request:2290", - "right": "pull_request:4275", - "accept": false, - "reason": "Both are docs, but for unrelated pipelines and different documentation scopes." - }, - { - "left": "pull_request:2540", - "right": "pull_request:432", - "accept": false, - "reason": "Different documentation edits in different pages: ControlNet section labels versus README link corrections." - }, - { - "left": "pull_request:2587", - "right": "pull_request:429", - "accept": false, - "reason": "Translation docs setup is unrelated to configuration API docs." - }, - { - "left": "pull_request:2587", - "right": "pull_request:500", - "accept": false, - "reason": "Different docs trees and different API subjects; not the same change." - }, - { - "left": "pull_request:2608", - "right": "pull_request:568", - "accept": false, - "reason": "Both are typo-related, but 568 is a broad multi-file typo pass and 2608 is a separate README punctuation fix." - }, - { - "left": "pull_request:2624", - "right": "pull_request:500", - "accept": false, - "reason": "Different configuration docs additions: cache_dir versus subfolder." + "reason": "StableDiffusionInpaint typo fix and ONNX docs typo fix are unrelated." } ] }, @@ -183991,8 +183728,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 258, - "review_unit_count": 273, + "review_unit_index": 257, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -184000,17 +183737,21 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:11892", - "pull_request:11894", - "pull_request:11995", - "pull_request:12269", - "pull_request:12611", - "pull_request:13063", + "pull_request:2271", + "pull_request:2290", + "pull_request:2540", + "pull_request:2587", + "pull_request:2608", + "pull_request:2624", "pull_request:299", + "pull_request:3066", "pull_request:3525", + "pull_request:3942", "pull_request:408", + "pull_request:421", "pull_request:4275", "pull_request:429", + "pull_request:432", "pull_request:4320", "pull_request:4472", "pull_request:4572", @@ -184039,16 +183780,21 @@ "pull_request:7181", "pull_request:7236", "pull_request:7472", - "pull_request:7977", - "pull_request:8308", - "pull_request:8310", - "pull_request:8498", "pull_request:8600", "pull_request:904", "pull_request:9077", "pull_request:9116" ], "soft_pairs": [ + "pull_request:2271|pull_request:3066", + "pull_request:2271|pull_request:3942", + "pull_request:2290|pull_request:421", + "pull_request:2290|pull_request:4275", + "pull_request:2540|pull_request:432", + "pull_request:2587|pull_request:429", + "pull_request:2587|pull_request:500", + "pull_request:2608|pull_request:568", + "pull_request:2624|pull_request:500", "pull_request:299|pull_request:904", "pull_request:3525|pull_request:5454", "pull_request:408|pull_request:718", @@ -184086,299 +183832,318 @@ "pull_request:7068|pull_request:7131", "pull_request:7068|pull_request:7181", "pull_request:7181|pull_request:9077", - "pull_request:7472|pull_request:9116", - "pull_request:7977|pull_request:8498", - "pull_request:8308|pull_request:8310", - "pull_request:11995|pull_request:12269", - "pull_request:11892|pull_request:11894", - "pull_request:12611|pull_request:13063" + "pull_request:7472|pull_request:9116" ], - "prepared_review_unit_hash": "be75d0fd6dcab5971706ac26865aeb1919291c7227e0591a3d73b83644b3e0d3", + "prepared_review_unit_hash": "27bbbe38d8e1901c4283c705c0ecdd15fd76fb046fe36bc27b035505f0abd649", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 47, "item_count": 47, - "soft_pair_count": 43, - "serialized_chars": 104024, - "estimated_input_tokens": 26006, - "estimated_eval_tokens": 52268 + "soft_pair_count": 47, + "serialized_chars": 97914, + "estimated_input_tokens": 24479, + "estimated_eval_tokens": 49214 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab-bag of unrelated small PRs, with many false-positive soft links caused by generic similarity (typos, docs, LoRA, modular pipelines, path handling). Only the HunyuanDiT refactor pair looks like the same concrete change; the rest are related-by-theme at best, not duplicates.", - "confidence": 0.9, + "summary": "This is a loose similarity cluster, not a true duplicate set: it mixes typo/docs edits, test expectation refreshes, and unrelated code-path fixes. I would reject all soft edges and avoid forcing a single canonical duplicate target.", + "confidence": 0.21, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR fits the whole cluster; if one representative is needed, pull_request:8308 is the best choice because pull_request:8310 appears to be the closest true duplicate with the same HunyuanDiT refactor and file set.", + "canonical_pr_reason": "No exact canonical duplicate stands out. If a representative is needed, PR 7181 is the broadest generic cleanup, but it is still not a true duplicate target for the rest of the cluster.", "best_issue_reason": null, - "best_pr_reason": "pull_request:8308 is the strongest duplicate candidate overall: it has the clearest near-identical sibling (8310) and a coherent substantive refactor, unlike the many theme-only matches elsewhere in the cluster.", + "best_pr_reason": "PR 7181 is the best overall representative of the cluster\u2019s dominant low-risk docs/typo-cleanup theme, but the cluster is too heterogeneous for it to serve as a genuine duplicate target.", "soft_edge_verdicts": [ + { + "left": "pull_request:2271", + "right": "pull_request:3066", + "accept": false, + "reason": "Both touch tests, but they fix different assertions in different files and different pipelines." + }, + { + "left": "pull_request:2271", + "right": "pull_request:3942", + "accept": false, + "reason": "Separate test failures in unrelated code paths; not the same bug." + }, + { + "left": "pull_request:2290", + "right": "pull_request:421", + "accept": false, + "reason": "Both are docs fixes, but they target different pages and different typo sets." + }, + { + "left": "pull_request:2290", + "right": "pull_request:4275", + "accept": false, + "reason": "Different documentation areas and unrelated wording changes." + }, + { + "left": "pull_request:2540", + "right": "pull_request:432", + "accept": false, + "reason": "Both are docs-only, but they fix different pages and different link/text issues." + }, + { + "left": "pull_request:2587", + "right": "pull_request:429", + "accept": false, + "reason": "A translated docs add-on and a configuration docs rewrite are not the same change." + }, + { + "left": "pull_request:2587", + "right": "pull_request:500", + "accept": false, + "reason": "Different documentation changes with no shared concrete bug." + }, + { + "left": "pull_request:2608", + "right": "pull_request:568", + "accept": false, + "reason": "Both are typo cleanups, but they affect unrelated files and unrelated wording." + }, + { + "left": "pull_request:2624", + "right": "pull_request:500", + "accept": false, + "reason": "Different docs topics: cache_dir API docs versus subfolder docs." + }, { "left": "pull_request:299", "right": "pull_request:904", "accept": false, - "reason": "Different changes: scheduler hotfix vs a full inpainting pipeline rewrite." + "reason": "A scheduler implementation hotfix and an inpainting pipeline rewrite are unrelated code paths." }, { "left": "pull_request:3525", "right": "pull_request:5454", "accept": false, - "reason": "Both touch docs, but they fix different pages and different issues." + "reason": "A README cleanup and a scheduler docs edit are not the same underlying change." }, { "left": "pull_request:408", "right": "pull_request:718", "accept": false, - "reason": "Both are typo/docs cleanups, but not the same concrete edit or code path." + "reason": "Both are broad docs typo fixes, but they patch different files and different mistakes." }, { "left": "pull_request:4275", "right": "pull_request:6339", "accept": false, - "reason": "Different documentation fixes for different pipelines; same theme, not same change." + "reason": "Both touch docs, but for different pipelines and different example text." }, { "left": "pull_request:429", "right": "pull_request:500", "accept": false, - "reason": "Different docs/features: config docs addition vs subfolder docs update." + "reason": "Both are configuration docs, but they address different API surfaces and different additions." }, { "left": "pull_request:4320", "right": "pull_request:4822", "accept": false, - "reason": "Unrelated typo fixes in different subsystems." + "reason": "A docs typo fix and a loader warning typo fix are different changes in different files." }, { "left": "pull_request:4472", "right": "pull_request:6438", "accept": false, - "reason": "Different README/install changes for different user flows." + "reason": "One removes SDXL login instructions; the other adds installation platform guidance." }, { "left": "pull_request:4572", "right": "pull_request:6759", "accept": false, - "reason": "Different bugs: LoRA network_alpha naming vs InstantID prompt-image embedding repetition." + "reason": "Different training/example bugs in different scripts and code paths." }, { "left": "pull_request:5003", "right": "pull_request:5201", "accept": false, - "reason": "Both involve LoRA, but they fix different behaviors and code paths." + "reason": "Both involve LoRA, but one fixes fuse/unfuse internals while the other changes processor removal behavior." }, { "left": "pull_request:568", "right": "pull_request:718", "accept": false, - "reason": "Generic typo cleanup overlap only; not the same underlying change." + "reason": "Generic typo fixes, but they touch different subsystems and unrelated wording." }, { "left": "pull_request:5780", "right": "pull_request:6754", "accept": false, - "reason": "Different fixes: checkpoint_merger wording vs diffusion DPO loss math." + "reason": "A community example typo and a diffusion DPO loss fix are unrelated." }, { "left": "pull_request:5780", "right": "pull_request:6765", "accept": false, - "reason": "Different files and different problems; only a broad docs/code cleanup similarity." + "reason": "Example doc wording versus cache-path utility logic; not the same bug." }, { "left": "pull_request:5780", "right": "pull_request:6766", "accept": false, - "reason": "Unrelated edits: checkpoint_merger typo vs path-handling constant change." + "reason": "Example doc wording versus constants path handling; unrelated changes." }, { "left": "pull_request:5780", "right": "pull_request:6767", "accept": false, - "reason": "Different code paths and no shared concrete bug." + "reason": "Example docs and Flax model path handling are different code paths." }, { "left": "pull_request:5780", "right": "pull_request:6768", "accept": false, - "reason": "Different fixes; only same broad maintenance theme." + "reason": "Example typo fix and config path handling are unrelated." }, { "left": "pull_request:5780", "right": "pull_request:6769", "accept": false, - "reason": "Different utilities and different path fixes; not a duplicate." + "reason": "Example typo fix and dynamic module path handling are unrelated." }, { "left": "pull_request:5780", "right": "pull_request:7052", "accept": false, - "reason": "Completely different features: typo fix vs TensorRT example update." + "reason": "A small example typo and a large TensorRT example rewrite are not the same change." }, { "left": "pull_request:5780", "right": "pull_request:7236", "accept": false, - "reason": "Different scope: typo fix vs DreamBooth training script bugfix." + "reason": "Different community/example edits with no shared concrete bug." }, { "left": "pull_request:6192", "right": "pull_request:6656", "accept": false, - "reason": "Both add unload-style loader methods, but for different loaders and different functionality." + "reason": "Both add unload-related APIs, but for different loaders and different model components." }, { "left": "pull_request:6754", "right": "pull_request:6765", "accept": false, - "reason": "Different bugs and unrelated files; DPO loss fix vs cache-path handling." + "reason": "Different fixes: DPO loss math versus cache-path construction." }, { "left": "pull_request:6754", "right": "pull_request:6766", "accept": false, - "reason": "Different changes entirely: DPO math vs constants path handling." + "reason": "Different code paths and unrelated bugs." }, { "left": "pull_request:6754", "right": "pull_request:6767", "accept": false, - "reason": "Different files and different path/file resolution issue, not the same bug." + "reason": "A training-loss fix and a Flax path fix are unrelated." }, { "left": "pull_request:6754", "right": "pull_request:6768", "accept": false, - "reason": "No shared concrete bug; one is DPO loss, the other config path handling." + "reason": "Different functionality: DPO loss calculation versus config path joining." }, { "left": "pull_request:6754", "right": "pull_request:6769", "accept": false, - "reason": "Different code paths and unrelated fixes." + "reason": "Different functionality and different subsystems." }, { "left": "pull_request:6754", "right": "pull_request:7052", "accept": false, - "reason": "Different domains: training loss bug vs community example update." + "reason": "A loss-function fix and a TensorRT example update are unrelated." }, { "left": "pull_request:6754", "right": "pull_request:7236", "accept": false, - "reason": "Different training scripts and different bugs." + "reason": "Different DreamBooth/DPO training scripts and different bugs." }, { "left": "pull_request:6754", "right": "pull_request:8600", "accept": false, - "reason": "Different model families and different training-script fixes." + "reason": "Different training scripts and unrelated fixes." }, { "left": "pull_request:6765", "right": "pull_request:7052", "accept": false, - "reason": "Path/cache utility fix vs community example rewrite; not the same change." + "reason": "Cache-path utility logic and a TensorRT example are unrelated." }, { "left": "pull_request:6765", "right": "pull_request:7236", "accept": false, - "reason": "Different code paths and no shared concrete bug." + "reason": "Different code paths and different bug classes." }, { "left": "pull_request:6765", "right": "pull_request:8600", "accept": false, - "reason": "Different scopes: cache-path fix vs DreamBooth script cleanup." + "reason": "Cache-path handling and DreamBooth text-encoder cleanup are unrelated." }, { "left": "pull_request:6766", "right": "pull_request:7052", "accept": false, - "reason": "Different files and unrelated bug fixes." + "reason": "Constants path construction and a TensorRT example rewrite are not the same change." }, { "left": "pull_request:6766", "right": "pull_request:7236", "accept": false, - "reason": "Different code paths; same issue label is not enough." + "reason": "Different files, different behavior, no shared concrete bug." }, { "left": "pull_request:6766", "right": "pull_request:8600", "accept": false, - "reason": "Different fixes in different files." + "reason": "Cross-platform constants fix versus DreamBooth script cleanup are unrelated." }, { "left": "pull_request:6767", "right": "pull_request:8600", "accept": false, - "reason": "Different bug classes and unrelated file sets." + "reason": "Flax model path handling and DreamBooth cleanup are unrelated code paths." }, { "left": "pull_request:7068", "right": "pull_request:7131", "accept": false, - "reason": "Both are typo docs PRs, but they are separate edits in different files." + "reason": "Both are typo/doc fixes, but they update different files and different content." }, { "left": "pull_request:7068", "right": "pull_request:7181", "accept": false, - "reason": "Generic typo cleanup only; not the same concrete change." + "reason": "Overlapping style, but not the same underlying change or files." }, { "left": "pull_request:7181", "right": "pull_request:9077", "accept": false, - "reason": "Both are typo sweeps, but across different files and topics." + "reason": "Both are typo sweeps, but they touch different areas and different exact edits." }, { "left": "pull_request:7472", "right": "pull_request:9116", "accept": false, - "reason": "Both fix links, but they are different broken links in different docs." - }, - { - "left": "pull_request:7977", - "right": "pull_request:8498", - "accept": false, - "reason": "Different typo fixes in unrelated docs, not the same underlying issue." - }, - { - "left": "pull_request:8308", - "right": "pull_request:8310", - "accept": true, - "reason": "Same HunyuanDiT model refactor with the same file set and the same new exports; this looks like the same concrete change." - }, - { - "left": "pull_request:11995", - "right": "pull_request:12269", - "accept": false, - "reason": "Same broad modular-pipeline area, but 12269 adds Kontext/I2I and other broader changes; not the same change." - }, - { - "left": "pull_request:11892", - "right": "pull_request:11894", - "accept": false, - "reason": "Opposite dependency changes: remove k-diffusion vs pin k-diffusion." - }, - { - "left": "pull_request:12611", - "right": "pull_request:13063", - "accept": false, - "reason": "Related Wan work, but 13063 is a broader task-based refactor with additional Wan22/I2V changes." + "reason": "A broken link in examples README and a dead link in stable audio docs are separate fixes." } ] }, @@ -184393,8 +184158,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 259, - "review_unit_count": 273, + "review_unit_index": 258, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -184403,38 +184168,34 @@ "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10136", - "pull_request:10259", "pull_request:10351", "pull_request:10454", "pull_request:10810", "pull_request:11124", "pull_request:11254", "pull_request:1137", + "pull_request:11892", + "pull_request:11894", + "pull_request:11995", + "pull_request:12269", "pull_request:12425", + "pull_request:12611", "pull_request:12782", "pull_request:13039", "pull_request:13042", - "pull_request:1312", + "pull_request:13063", "pull_request:13275", - "pull_request:13293", "pull_request:13424", - "pull_request:13442", - "pull_request:2602", - "pull_request:2637", "pull_request:2693", - "pull_request:2915", "pull_request:325", "pull_request:326", "pull_request:343", "pull_request:349", "pull_request:365", - "pull_request:3825", - "pull_request:3850", "pull_request:4572", "pull_request:4780", "pull_request:4783", "pull_request:483", - "pull_request:4947", "pull_request:5058", "pull_request:5188", "pull_request:5407", @@ -184445,13 +184206,21 @@ "pull_request:7627", "pull_request:7825", "pull_request:7919", + "pull_request:7977", + "pull_request:8308", + "pull_request:8310", + "pull_request:8498", "pull_request:9257", "pull_request:9736", "pull_request:9769", - "pull_request:978", - "pull_request:9985" + "pull_request:978" ], "soft_pairs": [ + "pull_request:7977|pull_request:8498", + "pull_request:8308|pull_request:8310", + "pull_request:11995|pull_request:12269", + "pull_request:11892|pull_request:11894", + "pull_request:12611|pull_request:13063", "pull_request:7247|pull_request:7825", "pull_request:10810|pull_request:9257", "pull_request:4572|pull_request:5407", @@ -184471,187 +184240,168 @@ "pull_request:326|pull_request:349", "pull_request:11254|pull_request:12425", "pull_request:1137|pull_request:2693", - "pull_request:13042|pull_request:13275", - "pull_request:2637|pull_request:3825", - "pull_request:2915|pull_request:3850", - "pull_request:4947|pull_request:5188", - "pull_request:13042|pull_request:13293", - "pull_request:13042|pull_request:13442", - "pull_request:10259|pull_request:9985", - "pull_request:1312|pull_request:2602" + "pull_request:13042|pull_request:13275" ], - "prepared_review_unit_hash": "6ce9e8c8a8fd4eb4f2fe2abeb46131ea073e1d870445ddd579fae459b05d0c3b", + "prepared_review_unit_hash": "5f298f24a0fec59103b992769911a2f954707d83af3de6bcac196883d103f486", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 27, - "serialized_chars": 102381, - "estimated_input_tokens": 25596, - "estimated_eval_tokens": 51448 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 106152, + "estimated_input_tokens": 26538, + "estimated_eval_tokens": 53332 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly a heterogeneous mix of unrelated feature, docs, test, and dependency PRs. The soft edges are driven by superficial similarity (same subsystem, same type of change, or same broad feature family) rather than the same underlying bug or mergeable change.", - "confidence": 0.95, + "summary": "This cluster is mostly a grab bag of unrelated PRs. The only plausible duplicate-like pair is the two Hunyuan model refactor PRs; the rest are separate fixes, docs tweaks, or feature additions that should stay distinct.", + "confidence": 0.9, "canonical_issue_reason": null, - "canonical_pr_reason": "No issue artifacts are present. Among the PRs, 9985 is the best single anchor only because it is the most substantial, broad feature PR in the set; it is not a true duplicate target for the others.", + "canonical_pr_reason": "PR 8310 is the clearest representative of the only plausible duplicate pair: it is the more specific Hunyuan DiT refactor and overlaps the same file set as 8308.", "best_issue_reason": null, - "best_pr_reason": "PR 9985 is the strongest representative node: large scope, high review activity, and a central Flux feature release. Even so, the cluster is not a real duplicate set.", + "best_pr_reason": "PR 8310 is the best single PR to anchor the only real duplicate candidate in this cluster; the other items are mostly unrelated changes.", "soft_edge_verdicts": [ + { + "left": "pull_request:7977", + "right": "pull_request:8498", + "accept": false, + "reason": "Both are typo/doc fixes, but they fix different typos in different files and are unrelated changes." + }, + { + "left": "pull_request:8308", + "right": "pull_request:8310", + "accept": true, + "reason": "Same Hunyuan/Hunyuan DiT model refactor over the same file set; these look like the same underlying change and could plausibly be one PR." + }, + { + "left": "pull_request:11995", + "right": "pull_request:12269", + "accept": false, + "reason": "Both involve modular Flux work, but one adds Flux modular T2I and the other adds Flux Kontext T2I/I2I support; different changes." + }, + { + "left": "pull_request:11892", + "right": "pull_request:11894", + "accept": false, + "reason": "One removes k-diffusion, the other pins it; these are opposite dependency changes, not duplicates." + }, + { + "left": "pull_request:12611", + "right": "pull_request:13063", + "accept": false, + "reason": "Both concern Wan modular pipelines, but the second is a later task-based refactor with new Wan22/Wan i2v blocks, not the same change." + }, { "left": "pull_request:7247", "right": "pull_request:7825", "accept": false, - "reason": "Both are test-related, but one fixes config-checking tests and the other fixes device/offload tests; different components and failures." + "reason": "Config-checking tests and device/offload tests are different test fixes for different behavior." }, { "left": "pull_request:10810", "right": "pull_request:9257", "accept": false, - "reason": "Flux LoRA conversion support vs. a DreamBooth Flux training-script bug fix; related area, but not the same code-path problem." + "reason": "Flux LoRA conversion support and a DreamBooth Flux prompt/tokenization fix are separate code paths." }, { "left": "pull_request:4572", "right": "pull_request:5407", "accept": false, - "reason": "Separate training-example fixes in different scripts for different bugs; only a broad theme overlap." + "reason": "Different training/example fixes: network_alpha naming versus mixed-precision unconditional training." }, { "left": "pull_request:4783", "right": "pull_request:6764", "accept": false, - "reason": "Both are docs updates, but one documents SDXL embeddings behavior and the other adds PEFT serialization guidance; unrelated changes." + "reason": "One adds SDXL embedding docs, the other adds a serialization note for fused adapters; unrelated documentation changes." }, { "left": "pull_request:9736", "right": "pull_request:9769", "accept": false, - "reason": "Different model releases (Allegro vs Mochi) with distinct codebases and pipelines." - }, - { - "left": "pull_request:12782", - "right": "pull_request:13039", - "accept": false, - "reason": "Hunyuan Video offloading-test tuning vs Wan animate CPU upcasting fix; unrelated bugs." - }, - { - "left": "pull_request:483", - "right": "pull_request:978", - "accept": false, - "reason": "Both are typo cleanups, but they address different files and different typo sets; not the same concrete change." + "reason": "Different model releases entirely: Allegro T2V versus Mochi T2V." }, { "left": "pull_request:326", "right": "pull_request:365", "accept": false, - "reason": "Type-hint cleanups in unrelated modules (PNDM pipeline vs VAE model); no shared bug or change." + "reason": "Type-hint cleanup in different modules; no shared bug or change." }, { "left": "pull_request:13275", "right": "pull_request:13424", "accept": false, - "reason": "Hunyuan test kwargs fix vs AutoencoderDC test adjustments; separate test failures." + "reason": "Separate test fixes in different areas: Hunyuan video tests versus AutoencoderDC tests." }, { "left": "pull_request:10351", "right": "pull_request:11124", "accept": false, - "reason": "A LoRA test fix and a CogView4 tokenizer test fix are unrelated test issues." + "reason": "A LoRA test fix and a CogView4 tokenizer test fix are unrelated." }, { "left": "pull_request:325", "right": "pull_request:343", "accept": false, - "reason": "Different API/type-hint updates in different scheduler/pipeline classes; not mergeable as one fix." + "reason": "Different scheduler/pipeline type-hint updates with no shared underlying issue." }, { "left": "pull_request:10136", "right": "pull_request:10454", "accept": false, - "reason": "Initial Hunyuan Video feature addition vs a later attention-mask dim bugfix; same family, different underlying changes." + "reason": "The first adds Hunyuan Video support; the second fixes an attention-mask bug later. Same subsystem, but not the same change." }, { "left": "pull_request:5993", "right": "pull_request:6009", "accept": false, - "reason": "Both fix constructor-argument mismatches, but in different community pipelines and different code paths." + "reason": "These fix different constructor/super-call issues in different community examples." }, { "left": "pull_request:5188", "right": "pull_request:5407", "accept": false, - "reason": "Training-utils helper addition vs mixed-precision training fix; unrelated changes." + "reason": "Compute-SNR utility work and unconditional training mixed-precision behavior are distinct changes." }, { "left": "pull_request:7627", "right": "pull_request:7919", "accept": false, - "reason": "One fixes Playground img2img latent handling, the other fixes offload behavior with device_map; different bugs." + "reason": "Playground img2img latent handling and device_map/offload behavior are separate fixes." }, { "left": "pull_request:4780", "right": "pull_request:5058", "accept": false, - "reason": "General PEFT integration draft vs a follow-up text-encoder backend PR; too broad and not the same concrete change." - }, - { - "left": "pull_request:326", - "right": "pull_request:349", - "accept": false, - "reason": "Separate type-hint updates in different scheduler/pipeline files; no duplicate relationship." + "reason": "PEFT integration draft work and later LoRA backend support are related, but not the same concrete change." }, { "left": "pull_request:11254", "right": "pull_request:12425", "accept": false, - "reason": "Different import-guard fixes in unrelated optional-dependency/torch.distributed areas." + "reason": "ConsisID import fixes and torch.distributed import guarding are unrelated dependency/import changes." }, { "left": "pull_request:1137", "right": "pull_request:2693", "accept": false, - "reason": "Code fix vs docs image-link typo; not the same artifact or bug." + "reason": "A pipeline dtype fix and a docs image-link typo are unrelated." }, { "left": "pull_request:13042", "right": "pull_request:13275", "accept": false, - "reason": "Different test fixes in different model families (QwenImage vs Hunyuan)." - }, - { - "left": "pull_request:13042", - "right": "pull_request:13293", - "accept": false, - "reason": "Different test fixes in different pipelines (QwenImage vs AudioLDM2)." - }, - { - "left": "pull_request:13042", - "right": "pull_request:13442", - "accept": false, - "reason": "Different test fixes in unrelated training/pipeline areas; only shared 'tests' label-level similarity." - }, - { - "left": "pull_request:10259", - "right": "pull_request:9985", - "accept": false, - "reason": "Both are Flux-related, but one adds LoRA-loading support and the other adds Flux Fill/Canny/Depth/Redux features; distinct changes." - }, - { - "left": "pull_request:1312", - "right": "pull_request:2602", - "accept": false, - "reason": "Both pin huggingface-hub for prerelease testing, but they target different versions and separate test intents." + "reason": "Prompt-isolation tests for QwenImage and Llava-kwargs fixes for Hunyuan tests are different test changes." } ] }, @@ -184666,8 +184416,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 260, - "review_unit_count": 273, + "review_unit_index": 259, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -184675,16 +184425,20 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10259", "pull_request:11534", "pull_request:11546", "pull_request:12782", "pull_request:13039", "pull_request:13042", + "pull_request:1312", "pull_request:13275", "pull_request:13293", "pull_request:13352", "pull_request:13365", + "pull_request:13442", "pull_request:13469", + "pull_request:2602", "pull_request:2637", "pull_request:2642", "pull_request:2647", @@ -184694,7 +184448,6 @@ "pull_request:2687", "pull_request:2688", "pull_request:2690", - "pull_request:2730", "pull_request:2740", "pull_request:2741", "pull_request:2742", @@ -184707,24 +184460,28 @@ "pull_request:2864", "pull_request:2865", "pull_request:2870", - "pull_request:2923", - "pull_request:2928", - "pull_request:3214", - "pull_request:3607", - "pull_request:3798", + "pull_request:2915", + "pull_request:3825", + "pull_request:3850", "pull_request:4343", - "pull_request:4546", - "pull_request:4822", "pull_request:4862", - "pull_request:4984", + "pull_request:4947", + "pull_request:5188", "pull_request:6021", - "pull_request:6485", "pull_request:7050", "pull_request:7103", "pull_request:7118", - "pull_request:7411" + "pull_request:7411", + "pull_request:9985" ], "soft_pairs": [ + "pull_request:2637|pull_request:3825", + "pull_request:2915|pull_request:3850", + "pull_request:4947|pull_request:5188", + "pull_request:13042|pull_request:13293", + "pull_request:13042|pull_request:13442", + "pull_request:10259|pull_request:9985", + "pull_request:1312|pull_request:2602", "pull_request:4343|pull_request:6021", "pull_request:7050|pull_request:7411", "pull_request:7118|pull_request:7411", @@ -184754,272 +184511,258 @@ "pull_request:2637|pull_request:2863", "pull_request:2637|pull_request:2864", "pull_request:2637|pull_request:2865", - "pull_request:2637|pull_request:2870", - "pull_request:2637|pull_request:2923", - "pull_request:2642|pull_request:2730", - "pull_request:2647|pull_request:2730", - "pull_request:2684|pull_request:2730", - "pull_request:2928|pull_request:3798", - "pull_request:3214|pull_request:3607", - "pull_request:4546|pull_request:4822", - "pull_request:4984|pull_request:6485" + "pull_request:2637|pull_request:2870" ], - "prepared_review_unit_hash": "98c11ccf1b4654c590416ab7bcbd247f3026726ce0b7e9535529ca8114a06129", + "prepared_review_unit_hash": "0145d76f7c5ea9a693cdfb068a0491ea0730cbd89ea00ffadd564667f53cfc94", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 38, - "serialized_chars": 95321, - "estimated_input_tokens": 23831, - "estimated_eval_tokens": 47918 + "soft_pair_count": 37, + "serialized_chars": 98009, + "estimated_input_tokens": 24503, + "estimated_eval_tokens": 49262 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a noisy, heterogeneous set: most pairs only share a broad area (docs, tests, workflow config, or a subsystem) but not the same concrete bug/change. I found no true duplicate PR pairs here.", - "confidence": 0.97, + "summary": "This cluster is mostly a pile of superficially similar but unrelated PRs: docs typo cleanups, test adjustments, CI workflow edits, and distinct model/loader feature changes. The soft edges are driven by overlapping keywords or shared files, not the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical representative: the set spans unrelated docs, tests, CI, and feature work, and the apparent overlaps are too shallow to indicate one shared change.", "best_issue_reason": null, - "best_pr_reason": "PR 7411 is the closest representative only because it\u2019s the broadest multi-file docs-typo sweep among the candidates, but it still isn\u2019t a duplicate target for the others.", + "best_pr_reason": "No PR is a strong global duplicate anchor. If forced, PR 9985 is the broadest and most central feature change, but it is still not equivalent to the others in the cluster.", "soft_edge_verdicts": [ + { + "left": "pull_request:2637", + "right": "pull_request:3825", + "accept": false, + "reason": "Both are doc updates, but they edit different pages with unrelated wording changes; this is not the same underlying change." + }, + { + "left": "pull_request:2915", + "right": "pull_request:3850", + "accept": false, + "reason": "Different test suites for different pipelines; they address separate test fixes, not one shared bug." + }, + { + "left": "pull_request:4947", + "right": "pull_request:5188", + "accept": false, + "reason": "Both touch training examples, but they fix different scripts and different behaviors." + }, + { + "left": "pull_request:13042", + "right": "pull_request:13293", + "accept": false, + "reason": "One fixes QwenImage prompt-isolation tests; the other changes AudioLDM2 cache-position handling. Different code paths." + }, + { + "left": "pull_request:13042", + "right": "pull_request:13442", + "accept": false, + "reason": "Different training/test fixes in different subsystems; no shared concrete bug or merged change." + }, + { + "left": "pull_request:10259", + "right": "pull_request:9985", + "accept": false, + "reason": "Both are Flux-related, but one is LoRA-loading support for control/fill and the other is a much broader Flux feature expansion. Not the same change." + }, + { + "left": "pull_request:1312", + "right": "pull_request:2602", + "accept": false, + "reason": "Both pin huggingface-hub prereleases, but they target different versions and different release-testing contexts." + }, { "left": "pull_request:4343", "right": "pull_request:6021", "accept": false, - "reason": "Different SDXL DreamBooth LoRA changes: rank argument vs variant argument; related script, not same fix." + "reason": "Both add a rank argument to LoRA training scripts, but they modify different scripts with different surrounding logic." }, { "left": "pull_request:7050", "right": "pull_request:7411", "accept": false, - "reason": "Both are typo/doc cleanup, but they touch different docs sets and are not the same change." + "reason": "Both are typo/doc cleanups, but they span different doc files and unrelated edits." }, { "left": "pull_request:7118", "right": "pull_request:7411", "accept": false, - "reason": "Both are documentation typo fixes, but for different pages/files and different edits." + "reason": "Different documentation pages and different typo fixes; not one underlying change." }, { "left": "pull_request:13293", "right": "pull_request:13469", "accept": false, - "reason": "Unrelated test fixes in different areas: AudioLDM2 tests vs attention-processor deprecation handling." + "reason": "Unrelated fixes: one is AudioLDM2 pipeline behavior, the other is attention-processor test deprecation handling." }, { "left": "pull_request:13352", "right": "pull_request:13365", "accept": false, - "reason": "Same workflow file, but different concrete changes; 13365 is a broader fork-support update, not a duplicate of the checkout-only PR." + "reason": "Same workflow file, but the changes are different: one adds checkout, the other rewrites fork support and sanitization. Too distinct to treat as one fix." }, { "left": "pull_request:4862", "right": "pull_request:7103", "accept": false, - "reason": "Both involve ControlNet, but they fix different bugs in different code paths (SDXL img2img MultiControlNet vs check_image validation)." + "reason": "Both involve ControlNet, but they fix different bugs in different functions and should not be merged as duplicates." }, { "left": "pull_request:11534", "right": "pull_request:11546", "accept": false, - "reason": "Framepack docs/pipeline addition vs a PEFT loader mapping tweak; not the same underlying change." + "reason": "Framepack feature/docs vs Hunyuan LoRA-loading support are separate changes, not one code-path problem." }, { "left": "pull_request:12782", "right": "pull_request:13042", "accept": false, - "reason": "Different test fixes in different models/projects; no shared bug path." + "reason": "Different test areas for different models; no shared underlying bug." }, { "left": "pull_request:12782", "right": "pull_request:13275", "accept": false, - "reason": "Both are Hunyuan test adjustments, but one is offloading split settings and the other is Llava kwargs; separate issues." + "reason": "Both are Hunyuan-related tests, but they address different test failures and different setups." }, { "left": "pull_request:13039", "right": "pull_request:13275", "accept": false, - "reason": "Different Hunyuan-related fixes: CPU upcasting in WAN animate vs Llava config kwargs in tests." + "reason": "Different models and different issues: WAN CPU upcasting vs Hunyuan test kwargs." }, { "left": "pull_request:2637", "right": "pull_request:2642", "accept": false, - "reason": "Both are docs edits, but they modify different pages and address unrelated wording fixes." + "reason": "Separate documentation pages with unrelated prose edits." }, { "left": "pull_request:2637", "right": "pull_request:2647", "accept": false, - "reason": "Same broad docs cleanup theme, but different document and wording change." + "reason": "Different docs pages and different wording fixes; not the same change." }, { "left": "pull_request:2637", "right": "pull_request:2684", "accept": false, - "reason": "Different docs page and different sentence-level correction; not a duplicate." + "reason": "Unrelated documentation edits on different pages." }, { "left": "pull_request:2637", "right": "pull_request:2685", "accept": false, - "reason": "Unrelated docs typo fixes on different pages." + "reason": "Different doc page and content; only the general docs-maintenance theme overlaps." }, { "left": "pull_request:2637", "right": "pull_request:2686", "accept": false, - "reason": "Different docs page and wording fix; not the same change." + "reason": "Different docs page and unrelated wording correction." }, { "left": "pull_request:2637", "right": "pull_request:2687", "accept": false, - "reason": "Different docs page and wording fix; not the same change." + "reason": "Separate documentation cleanup on another page." }, { "left": "pull_request:2637", "right": "pull_request:2688", "accept": false, - "reason": "Different docs page and wording fix; not a duplicate." + "reason": "Different docs page and edits; not a duplicate." }, { "left": "pull_request:2637", "right": "pull_request:2690", "accept": false, - "reason": "Different docs page and wording fix; not the same underlying issue." + "reason": "Different documentation content and target page." }, { "left": "pull_request:2637", "right": "pull_request:2740", "accept": false, - "reason": "Different documentation updates; unrelated content and files." + "reason": "Different training docs with distinct edits; no shared underlying fix." }, { "left": "pull_request:2637", "right": "pull_request:2741", "accept": false, - "reason": "Different docs page and wording fix; not the same change." + "reason": "Different docs file and separate wording change." }, { "left": "pull_request:2637", "right": "pull_request:2742", "accept": false, - "reason": "Different docs page and wording fix; no shared concrete bug." + "reason": "Dreambooth docs vs quicktour docs; unrelated changes." }, { "left": "pull_request:2637", "right": "pull_request:2746", "accept": false, - "reason": "Different docs page and wording fix; unrelated." + "reason": "Different docs page and typo fix only." }, { "left": "pull_request:2637", "right": "pull_request:2749", "accept": false, - "reason": "Different docs page and wording fix; not a duplicate." + "reason": "Different optimization docs with unrelated wording edits." }, { "left": "pull_request:2637", "right": "pull_request:2751", "accept": false, - "reason": "Different docs page and wording fix; unrelated." + "reason": "Different training documentation page and different change." }, { "left": "pull_request:2637", "right": "pull_request:2752", "accept": false, - "reason": "Different conceptual docs edit; not the same change." + "reason": "Philosophy docs vs quicktour docs; separate edits." }, { "left": "pull_request:2637", "right": "pull_request:2862", "accept": false, - "reason": "Different docs page and wording fix; unrelated." + "reason": "Different conceptual docs page and different wording cleanup." }, { "left": "pull_request:2637", "right": "pull_request:2863", "accept": false, - "reason": "Different docs page and wording fix; not the same issue." + "reason": "Different docs page and standalone typo fix." }, { "left": "pull_request:2637", "right": "pull_request:2864", "accept": false, - "reason": "Different docs page and wording fix; unrelated." + "reason": "Different docs page; only topical similarity exists." }, { "left": "pull_request:2637", "right": "pull_request:2865", "accept": false, - "reason": "Different docs page and wording fix; not the same change." + "reason": "Different pipeline docs pages and separate typo fixes." }, { "left": "pull_request:2637", "right": "pull_request:2870", "accept": false, - "reason": "Different docs page and wording fix; not a duplicate." - }, - { - "left": "pull_request:2637", - "right": "pull_request:2923", - "accept": false, - "reason": "Different docs page and wording fix; unrelated." - }, - { - "left": "pull_request:2642", - "right": "pull_request:2730", - "accept": false, - "reason": "Different docs changes on different pages; only superficially similar typo cleanup." - }, - { - "left": "pull_request:2647", - "right": "pull_request:2730", - "accept": false, - "reason": "Different docs changes on different pages; not the same underlying change." - }, - { - "left": "pull_request:2684", - "right": "pull_request:2730", - "accept": false, - "reason": "Different docs pages and unrelated wording fixes." - }, - { - "left": "pull_request:2928", - "right": "pull_request:3798", - "accept": false, - "reason": "ControlNet dataset-loading fix vs unrelated comment cleanup in training scripts." - }, - { - "left": "pull_request:3214", - "right": "pull_request:3607", - "accept": false, - "reason": "Different bugs: attention reshape mismatch vs temb attention behavior." - }, - { - "left": "pull_request:4546", - "right": "pull_request:4822", - "accept": false, - "reason": "Different fixes in different files: controlnet conversion typo vs LoRA warning string." - }, - { - "left": "pull_request:4984", - "right": "pull_request:6485", - "accept": false, - "reason": "Both touch the PR template reviewer list, but they correct different entries and are not the same change." + "reason": "Different docs page and unrelated formatting/title cleanup." } ] }, @@ -185034,8 +184777,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 261, - "review_unit_count": 273, + "review_unit_index": 260, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -185043,18 +184786,10 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:102", - "pull_request:1020", - "pull_request:1077", - "pull_request:11588", - "pull_request:11726", "pull_request:12069", - "pull_request:121", "pull_request:12416", "pull_request:12454", "pull_request:12490", - "pull_request:12584", - "pull_request:12634", "pull_request:12782", "pull_request:13039", "pull_request:13042", @@ -185062,14 +184797,23 @@ "pull_request:13424", "pull_request:13442", "pull_request:154", - "pull_request:1861", - "pull_request:2329", - "pull_request:252", + "pull_request:2637", + "pull_request:2642", + "pull_request:2647", + "pull_request:2684", + "pull_request:2730", + "pull_request:2923", + "pull_request:2928", + "pull_request:3214", "pull_request:356", - "pull_request:415", + "pull_request:3607", + "pull_request:3798", + "pull_request:4546", "pull_request:4621", "pull_request:4749", + "pull_request:4822", "pull_request:4947", + "pull_request:4984", "pull_request:5164", "pull_request:5188", "pull_request:5552", @@ -185078,6 +184822,7 @@ "pull_request:6276", "pull_request:6346", "pull_request:6433", + "pull_request:6485", "pull_request:7062", "pull_request:7247", "pull_request:7353", @@ -185085,14 +184830,20 @@ "pull_request:7825", "pull_request:7934", "pull_request:8467", - "pull_request:8631", "pull_request:8797", - "pull_request:8798", "pull_request:9060", "pull_request:9206", "pull_request:9696" ], "soft_pairs": [ + "pull_request:2637|pull_request:2923", + "pull_request:2642|pull_request:2730", + "pull_request:2647|pull_request:2730", + "pull_request:2684|pull_request:2730", + "pull_request:2928|pull_request:3798", + "pull_request:3214|pull_request:3607", + "pull_request:4546|pull_request:4822", + "pull_request:4984|pull_request:6485", "pull_request:5768|pull_request:7746", "pull_request:6276|pull_request:6433", "pull_request:7247|pull_request:7934", @@ -185113,225 +184864,204 @@ "pull_request:12416|pull_request:12454", "pull_request:13042|pull_request:13424", "pull_request:7825|pull_request:8797", - "pull_request:5188|pull_request:6346", - "pull_request:8631|pull_request:8798", - "pull_request:12584|pull_request:12634", - "pull_request:12490|pull_request:13042", - "pull_request:11588|pull_request:11726", - "pull_request:102|pull_request:121", - "pull_request:102|pull_request:1861", - "pull_request:102|pull_request:252", - "pull_request:1020|pull_request:415", - "pull_request:1077|pull_request:2329" + "pull_request:5188|pull_request:6346" ], - "prepared_review_unit_hash": "223f9dc3004449fe398494ce9de14baeac8cc0d05777ee0d89c5df31dc394d85", + "prepared_review_unit_hash": "4a5fc1219889358300bbeb15d5d09c6853a46930942188be71ed731e578aea0e", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 30, - "serialized_chars": 103021, - "estimated_input_tokens": 25756, - "estimated_eval_tokens": 51768 + "soft_pair_count": 29, + "serialized_chars": 98002, + "estimated_input_tokens": 24501, + "estimated_eval_tokens": 49258 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of superficially similar but substantively different PRs: many are test fixes, docs updates, or feature work in different subsystems. I do not see a true duplicate pair among the soft candidates; the only near-matches are feature-adjacent follow-ups, not the same underlying change.", - "confidence": 0.88, + "summary": "This cluster is not a true duplicate set; it mixes unrelated docs tweaks, test fixes, small bug fixes, and feature work across different subsystems. Most soft-edge pairs should stay separate.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "No issue artifacts are present. Among the PRs, #12584 is the strongest representative anchor: it is the most substantial multi-file feature addition with heavy review activity, so it best stands in for the cluster despite the cluster\u2019s heterogeneity.", + "canonical_pr_reason": "If a single anchor is needed, PR 5164 is the strongest representative: it is the broadest substantive code change, spans many files, and had the most discussion/review activity. But the set is heterogeneous, so no real canonical PR exists.", "best_issue_reason": null, - "best_pr_reason": "#12584 is the most complete and central PR in the set, with the broadest scope and the most discussion; if one PR must represent this mixed group, it is the clearest anchor.", + "best_pr_reason": "PR 5164 is the best overall representative because it is the largest, most impactful core-library change in the set and has the clearest product significance.", "soft_edge_verdicts": [ + { + "left": "pull_request:2637", + "right": "pull_request:2923", + "accept": false, + "reason": "Both are docs edits, but they change different pages and different wording; not the same underlying change." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2730", + "accept": false, + "reason": "Different documentation pages with unrelated edits; no shared bug or change request." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2730", + "accept": false, + "reason": "Unrelated docs wording fix versus training docs newline/wording cleanup." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2730", + "accept": false, + "reason": "Different docs files and different edits; not a duplicate." + }, + { + "left": "pull_request:2928", + "right": "pull_request:3798", + "accept": false, + "reason": "ControlNet dataset-loading change is unrelated to the comment-only training-script cleanup." + }, + { + "left": "pull_request:3214", + "right": "pull_request:3607", + "accept": false, + "reason": "Attention reshape bug fix is a concrete code-path issue; the other PR is a separate temb-attention change with different scope." + }, + { + "left": "pull_request:4546", + "right": "pull_request:4822", + "accept": false, + "reason": "Both are typo/warning text fixes, but in different scripts and different messages; not the same change." + }, + { + "left": "pull_request:4984", + "right": "pull_request:6485", + "accept": false, + "reason": "Same template file, but the reviewer-list update is a later separate maintenance edit, not the same underlying fix." + }, { "left": "pull_request:5768", "right": "pull_request:7746", "accept": false, - "reason": "Both are small test-related PRs, but they fix different model test failures in different codepaths and are not the same bug." + "reason": "One adjusts docs/config handling for consistency decoder; the other changes a VAE test dtype. Different problems." }, { "left": "pull_request:6276", "right": "pull_request:6433", "accept": false, - "reason": "Both touch community examples, but they add different pipelines/features and do not target the same change." + "reason": "Community README entry versus a separate community pipeline addition; same area, but not the same artifact or bug." }, { "left": "pull_request:7247", "right": "pull_request:7934", "accept": false, - "reason": "One adjusts config checking tests, the other fixes single-file config loading from dicts; related area, different defect." + "reason": "Both involve single-file/loading utilities, but they fix different code paths and different failure modes." }, { "left": "pull_request:7247", "right": "pull_request:8797", "accept": false, - "reason": "These are unrelated fixes: config-checking tests versus sharded-checkpoint loading from a subfolder." + "reason": "Single-file config normalization and sharding tests are unrelated." }, { "left": "pull_request:9206", "right": "pull_request:9696", "accept": false, - "reason": "A test fix for xformers issues is unrelated to a workflow change for running tests from a PR." + "reason": "Broken xformers tests versus workflow input validation; different concerns and not mergeable as one fix." }, { "left": "pull_request:12782", "right": "pull_request:13293", "accept": false, - "reason": "Different AudioLDM2 issues: offloading test parameters versus cache-position handling in the pipeline." + "reason": "Hunyuan offloading test tweak is unrelated to AudioLDM2 cache-position logic." }, { "left": "pull_request:12782", "right": "pull_request:13442", "accept": false, - "reason": "These are unrelated test/training fixes in different areas and do not share the same underlying bug." + "reason": "Different test suites and different model families; no shared underlying bug." }, { "left": "pull_request:13039", "right": "pull_request:13293", "accept": false, - "reason": "Wan animate CPU upcasting and AudioLDM2 cache-position logic are separate bugs in different models." + "reason": "WAN layerwise upcasting on CPU is unrelated to AudioLDM2 cache-position handling." }, { "left": "pull_request:13039", "right": "pull_request:13442", "accept": false, - "reason": "No overlap beyond being test/bug-fix PRs; they address different subsystems and failures." + "reason": "Different model-specific test fixes; not the same issue." }, { "left": "pull_request:4621", "right": "pull_request:4947", "accept": false, - "reason": "Multi-adapter downscale validation and SDXL adapter training arguments are unrelated changes." + "reason": "Multi-adapter downscale validation and T2I adapter training support are separate changes." }, { "left": "pull_request:4749", "right": "pull_request:5615", "accept": false, - "reason": "Both mention SDXL LoRA behavior, but one fixes checkpointing while the other reverts load-lora handling in a different pipeline." + "reason": "SDXL LoRA checkpointing and SDXL Adapter load-lora revert address different code paths." }, { "left": "pull_request:5164", "right": "pull_request:5552", "accept": false, - "reason": "FreeU feature implementation and a later disable-method fix are related follow-ups, not the same concrete change." + "reason": "FreeU mechanism introduction and fixing disable_freeu are related, but the PRs are different changes with different concrete fixes." }, { "left": "pull_request:7062", "right": "pull_request:8467", "accept": false, - "reason": "Docstring cleanup for Imagick is unrelated to the SEGA pipeline runtime fix." + "reason": "Community imagic docstring cleanup versus SEGA pipeline fix are unrelated." }, { "left": "pull_request:7353", "right": "pull_request:9060", "accept": false, - "reason": "Adding FromSingleFileMixin to SDXL ControlNet img2img is unrelated to removing a debug print in Flux." + "reason": "Adding FromSingleFileMixin to a ControlNet XL pipeline is unrelated to removing a debug print in Flux." }, { "left": "pull_request:12069", "right": "pull_request:12490", "accept": false, - "reason": "Both are AudioLDM2 test-related, but they fix different issues: transformers cache handling versus ClapConfig construction." + "reason": "Both are AudioLDM2-related, but one fixes pipeline compatibility logic while the other only adjusts test component construction." }, { "left": "pull_request:154", "right": "pull_request:356", "accept": false, - "reason": "A conversion script for Stable Diffusion is not the same change as the later fine-tuning example addition." + "reason": "Stable Diffusion conversion script creation and training README/implementation work are not the same change." }, { "left": "pull_request:7247", "right": "pull_request:9206", "accept": false, - "reason": "Single-file config tuple/list handling and xformers test fixes are unrelated." + "reason": "Single-file config conversion and xformers test failures are unrelated." }, { "left": "pull_request:12416", "right": "pull_request:12454", "accept": false, - "reason": "These add support for different modular pipelines (QwenImage Edit Plus vs Flux Kontext)." - }, - { - "left": "pull_request:13042", - "right": "pull_request:13424", - "accept": false, - "reason": "Prompt-isolation fixes for QwenImage are unrelated to AutoencoderDC test adjustments." + "reason": "Both add modular pipeline support, but for different models/features; not the same underlying bug or change." }, { "left": "pull_request:7825", "right": "pull_request:8797", "accept": false, - "reason": "Both modify test utilities, but one handles device-map/offload tests while the other fixes sharded checkpoint loading." + "reason": "Model offload guards and sharded-checkpoint test fixes are separate test issues." }, { "left": "pull_request:5188", "right": "pull_request:6346", "accept": false, - "reason": "Adding compute_snr to training utils and a later LCM LoRA follow-up are not the same bug or feature." - }, - { - "left": "pull_request:8631", - "right": "pull_request:8798", - "accept": false, - "reason": "A T5 single-file fix and sharded-checkpoint subfolder loading address different loader paths." - }, - { - "left": "pull_request:12584", - "right": "pull_request:12634", - "accept": false, - "reason": "Same project family, but one adds SANA-Video inference assets and the other adds image-to-video pipeline support; distinct capabilities." - }, - { - "left": "pull_request:12490", - "right": "pull_request:13042", - "accept": false, - "reason": "AudioLDM2 test cleanup and QwenImage prompt isolation fixes are entirely unrelated." - }, - { - "left": "pull_request:11588", - "right": "pull_request:11726", - "accept": false, - "reason": "Both concern Wan LoRA support, but one fixes I2V expansion in the pipeline while the other changes LoRA conversion logic; different codepaths and bugs." - }, - { - "left": "pull_request:102", - "right": "pull_request:121", - "accept": false, - "reason": "These are separate documentation/example updates with different scope and content." - }, - { - "left": "pull_request:102", - "right": "pull_request:1861", - "accept": false, - "reason": "A training example update and a loss-computation change are not the same change." - }, - { - "left": "pull_request:102", - "right": "pull_request:252", - "accept": false, - "reason": "Both touch README examples, but they are different documentation revisions and not duplicates." - }, - { - "left": "pull_request:1020", - "right": "pull_request:415", - "accept": false, - "reason": "Training docs updates overlap in area, but they are separate doc sets and not the same underlying change." - }, - { - "left": "pull_request:1077", - "right": "pull_request:2329", - "accept": false, - "reason": "Scheduler integration test refactoring and push-test workflow refactoring are unrelated test-infrastructure changes." + "reason": "compute_snr utility addition and LoRA state-dict conversion are unrelated training-example fixes." } ] }, @@ -185346,8 +185076,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 262, - "review_unit_count": 273, + "review_unit_index": 261, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -185355,7 +185085,12 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:102", + "pull_request:1020", + "pull_request:1077", + "pull_request:11588", "pull_request:1162", + "pull_request:11726", "pull_request:1182", "pull_request:119", "pull_request:1194", @@ -185364,47 +185099,51 @@ "pull_request:1218", "pull_request:12265", "pull_request:12307", + "pull_request:12490", "pull_request:12552", "pull_request:12573", + "pull_request:12584", "pull_request:12628", + "pull_request:12634", "pull_request:12705", "pull_request:12763", "pull_request:12782", "pull_request:12932", "pull_request:13039", + "pull_request:13042", "pull_request:13115", "pull_request:13424", "pull_request:1360", "pull_request:1487", "pull_request:15", - "pull_request:1510", "pull_request:1610", "pull_request:168", "pull_request:1724", - "pull_request:1763", - "pull_request:1950", - "pull_request:1968", + "pull_request:1861", "pull_request:2006", - "pull_request:2062", - "pull_request:21", - "pull_request:2197", "pull_request:2208", + "pull_request:2329", "pull_request:252", - "pull_request:260", "pull_request:2624", "pull_request:312", - "pull_request:3343", "pull_request:3470", - "pull_request:353", - "pull_request:356", + "pull_request:415", "pull_request:429", - "pull_request:43", - "pull_request:500", "pull_request:61", "pull_request:63", - "pull_request:75" + "pull_request:8631", + "pull_request:8798" ], "soft_pairs": [ + "pull_request:8631|pull_request:8798", + "pull_request:12584|pull_request:12634", + "pull_request:12490|pull_request:13042", + "pull_request:11588|pull_request:11726", + "pull_request:102|pull_request:121", + "pull_request:102|pull_request:1861", + "pull_request:102|pull_request:252", + "pull_request:1020|pull_request:415", + "pull_request:1077|pull_request:2329", "pull_request:1162|pull_request:2006", "pull_request:1162|pull_request:312", "pull_request:1182|pull_request:1610", @@ -185426,238 +185165,222 @@ "pull_request:1487|pull_request:2208", "pull_request:1487|pull_request:2624", "pull_request:1487|pull_request:3470", - "pull_request:1487|pull_request:429", - "pull_request:1487|pull_request:500", - "pull_request:1510|pull_request:61", - "pull_request:1510|pull_request:63", - "pull_request:168|pull_request:356", - "pull_request:1763|pull_request:75", - "pull_request:1950|pull_request:3343", - "pull_request:1968|pull_request:2062", - "pull_request:21|pull_request:43", - "pull_request:2197|pull_request:260", - "pull_request:2197|pull_request:353" + "pull_request:1487|pull_request:429" ], - "prepared_review_unit_hash": "93ba4d4ae0ae40d3ea6511430866e5469d8dd9cb95cad2231b52556921f14353", + "prepared_review_unit_hash": "3dfa5666dfab9ad7f3b3df8607435b3c02977a16441cd7d0e16a01a73d707759", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 32, - "serialized_chars": 102081, - "estimated_input_tokens": 25521, - "estimated_eval_tokens": 51298 + "soft_pair_count": 31, + "serialized_chars": 102733, + "estimated_input_tokens": 25684, + "estimated_eval_tokens": 51624 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is mostly a grab bag of small, unrelated maintenance PRs: docs/link fixes, typo cleanups, test adjustments, and a few narrow code-path fixes. The similarity edges are mostly superficial (same subsystem or same file family) rather than true duplicates, so almost all should be rejected.", - "confidence": 0.91, + "summary": "This cluster is mostly heterogeneous, with a few superficially similar docs/tests/feature PRs but no true duplicate pairs. The only close-looking pair is the Sana-Video feature work, but the PRs target different concrete capabilities (initial inference vs image-to-video support).", + "confidence": 0.19, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #12265 is the broadest generic maintenance cleanup in the set, spanning multiple docs/examples and code files with a clear typo-fix theme, so it is the best umbrella representative of this cluster.", + "canonical_pr_reason": "No issue exists and the PRs do not form a real duplicate set; if one representative must be chosen, PR 12584 is the most substantial self-contained feature landing in the cluster, with broad SANA-Video docs/code/tests and high review activity.", "best_issue_reason": null, - "best_pr_reason": "PR #12265 is the best single representative because it covers the widest small-fix surface area and matches the dominant low-level maintenance/doc-cleanup character of the cluster.", + "best_pr_reason": "PR 12584 is the strongest single representative because it is a complete, well-reviewed feature addition with broad scope; the rest of the items are mostly unrelated docs, tests, or small fixes.", "soft_edge_verdicts": [ + { + "left": "pull_request:8631", + "right": "pull_request:8798", + "accept": false, + "reason": "Both are loading-related bugfixes, but they touch different code paths: single-file T5 checkpoint creation vs sharded checkpoint loading from subfolders." + }, + { + "left": "pull_request:12584", + "right": "pull_request:12634", + "accept": false, + "reason": "Same SANA-Video family, but different concrete changes: 12584 adds 5s 480p inference, while 12634 adds image-to-video pipeline support and new I2V code paths." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13042", + "accept": false, + "reason": "Unrelated test fixes for different models/pipelines: AudioLDM2 ClapConfig setup vs QwenImage prompt isolation." + }, + { + "left": "pull_request:11588", + "right": "pull_request:11726", + "accept": false, + "reason": "Both concern Wan LoRA handling, but one fixes I2V LoRA support and the other expands conversion support for VACE loras; not the same change." + }, + { + "left": "pull_request:102", + "right": "pull_request:121", + "accept": false, + "reason": "Both edit examples docs, but one updates training examples while the other updates the main README examples with code blocks; separate docs changes." + }, + { + "left": "pull_request:102", + "right": "pull_request:1861", + "accept": false, + "reason": "Training examples README updates vs loss-computation changes in training scripts are different artifacts and fixes." + }, + { + "left": "pull_request:102", + "right": "pull_request:252", + "accept": false, + "reason": "Both touch README/example content, but they are different documentation updates with different scope and content." + }, + { + "left": "pull_request:1020", + "right": "pull_request:415", + "accept": false, + "reason": "Both are training-docs PRs, but 1020 is a broad docs refresh across training/fine-tuning pages while 415 is an earlier restructuring of training docs." + }, + { + "left": "pull_request:1077", + "right": "pull_request:2329", + "accept": false, + "reason": "Different test infrastructure changes: scheduler integration test refactor vs push-tests workflow refactor." + }, { "left": "pull_request:1162", "right": "pull_request:2006", "accept": false, - "reason": "Both are maintenance PRs, but one fixes community pipeline links while the other fixes checkpoint merger behavior; different problems and code paths." + "reason": "Different community maintenance fixes: link updates vs checkpoint-merger behavior and loading fixes." }, { "left": "pull_request:1162", "right": "pull_request:312", "accept": false, - "reason": "Both touch links/docs, but they fix different broken links in different docs locations; too broad to treat as the same change." + "reason": "Both fix links, but they target different pages and link targets; not the same underlying issue." }, { "left": "pull_request:1182", "right": "pull_request:1610", "accept": false, - "reason": "LPW docs improvements and xformers-related test/logging changes are unrelated changes despite both being minor maintenance." + "reason": "Docs for the LPW pipeline vs attention/xFormers pipeline test changes are unrelated." }, { "left": "pull_request:119", "right": "pull_request:15", "accept": false, - "reason": "A variable-name bugfix in hub_utils and a README output-path rename are unrelated." + "reason": "A variable-name bugfix in hub_utils is unrelated to the examples README output-path rename." }, { "left": "pull_request:1194", "right": "pull_request:2208", "accept": false, - "reason": "Doc-builder workflow updates and xFormers docs updates are different maintenance areas with no shared concrete bug." + "reason": "Different documentation/workflow edits: PR docs action wiring vs xFormers docs updates." }, { "left": "pull_request:120", "right": "pull_request:121", "accept": false, - "reason": "Both edit the README examples section, but one adds example code while the other expands and refactors examples; not the same underlying change." + "reason": "Both update README examples, but 120 adds the examples section and 121 adds code samples; they are complementary, not duplicates." }, { "left": "pull_request:120", "right": "pull_request:252", "accept": false, - "reason": "Both are README/example updates, but they are separate documentation edits made at different times and not a single duplicate fix." + "reason": "Both are README example updates, but they add different example content in different edits." }, { "left": "pull_request:1218", "right": "pull_request:61", "accept": false, - "reason": "The conversion script improvement and the LatentDiffusionUncond conversion script are distinct scripts and changes." + "reason": "Both are conversion-script work, but one extends the Stable Diffusion conversion script and the other adds an LDM unconditional conversion script." }, { "left": "pull_request:1218", "right": "pull_request:63", "accept": false, - "reason": "Improving the stable diffusion conversion script and adding a BDDM conversion script are different features." + "reason": "Different conversion scripts for different pipelines (Stable Diffusion vs BDDM), so not the same fix." }, { "left": "pull_request:12265", "right": "pull_request:12307", "accept": false, - "reason": "Both are broad maintenance PRs, but one is a typo sweep and the other adds/modifies modular-pipeline infrastructure; not the same fix." + "reason": "One is typo cleanup across assorted files; the other is a large modular-pipeline feature/fix set." }, { "left": "pull_request:12265", "right": "pull_request:13115", "accept": false, - "reason": "A typo cleanup sweep is unrelated to the modular-pipeline import/export fix." + "reason": "Typos/cleanup PR vs modular-pipeline import/export additions for Flux2 are not the same change." }, { "left": "pull_request:12552", "right": "pull_request:12705", "accept": false, - "reason": "Kandinsky5 doc typos and a broad typo-fix sweep across many files are different scopes and not the same bug." + "reason": "Both are typo fixes, but they modify different docs/code locations and address different wording issues." }, { "left": "pull_request:12573", "right": "pull_request:12932", "accept": false, - "reason": "A better warning message for modular loading is not the same as improving QwenImage docstrings." + "reason": "Different small maintenance changes in modular pipelines: warning text vs docstring/output metadata updates." }, { "left": "pull_request:12628", "right": "pull_request:12763", "accept": false, - "reason": "One adds a modular-pipeline validation check; the other adds the Flux2 modular pipeline feature set. Different changes." + "reason": "A validation check in modular pipeline internals is not the same as adding the Flux2 modular pipeline feature." }, { "left": "pull_request:12782", "right": "pull_request:13424", "accept": false, - "reason": "These fix different test failures in different models; same general area, but not the same underlying bug." + "reason": "Unrelated test fixes for HunyuanVideo offloading vs AutoencoderDC dtype inference." }, { "left": "pull_request:13039", "right": "pull_request:13424", "accept": false, - "reason": "WAN layerwise upcasting CPU tests and AutoencoderDC dtype inference tests are unrelated test fixes." + "reason": "Different test fixes for Wan layerwise upcasting vs AutoencoderDC tests; no shared code path." }, { "left": "pull_request:1360", "right": "pull_request:168", "accept": false, - "reason": "A community K-diffusion pipeline addition and the original Stable Diffusion pipeline introduction are separate features." + "reason": "Community k-diffusion pipeline addition is a separate feature from the original stable diffusion pipeline landing." }, { "left": "pull_request:1487", "right": "pull_request:1724", "accept": false, - "reason": "Both concern xformers docs, but one is API docs and the other is a broader recommendation/addition of an optimization guide; not the same change." + "reason": "Both mention xFormers docs, but 1487 adds xFormers API docs while 1724 recommends xFormers broadly; not duplicate work." }, { "left": "pull_request:1487", "right": "pull_request:2208", "accept": false, - "reason": "They both mention xFormers, but one updates pipeline docs and the other updates the dedicated xFormers guide; different doc changes." + "reason": "Different xFormers documentation edits at different times and scopes." }, { "left": "pull_request:1487", "right": "pull_request:2624", "accept": false, - "reason": "xFormers docs and cache_dir docs are unrelated documentation updates." + "reason": "xFormers docs are unrelated to adding cache_dir documentation to pipeline_utils." }, { "left": "pull_request:1487", "right": "pull_request:3470", "accept": false, - "reason": "xFormers documentation and DreamBooth IF docs are completely different subjects." + "reason": "xFormers documentation and IF DreamBooth documentation are unrelated changes." }, { "left": "pull_request:1487", "right": "pull_request:429", "accept": false, - "reason": "xFormers docs and configuration docs are unrelated documentation areas." - }, - { - "left": "pull_request:1487", - "right": "pull_request:500", - "accept": false, - "reason": "xFormers documentation and subfolder configuration docs are different topics." - }, - { - "left": "pull_request:1510", - "right": "pull_request:61", - "accept": false, - "reason": "A scheduler/conversion script improvement is not the same as adding the uncond conversion script." - }, - { - "left": "pull_request:1510", - "right": "pull_request:63", - "accept": false, - "reason": "A scheduler/conversion improvement and a BDDM converter are different code additions." - }, - { - "left": "pull_request:168", - "right": "pull_request:356", - "accept": false, - "reason": "The Stable Diffusion pipeline introduction and the fine-tuning README/script are separate features." - }, - { - "left": "pull_request:1763", - "right": "pull_request:75", - "accept": false, - "reason": "Both touch attention code, but they fix different concrete bugs in different code paths and should not be merged as one PR." - }, - { - "left": "pull_request:1950", - "right": "pull_request:3343", - "accept": false, - "reason": "One fixes a OneFlow graph/config issue; the other is a DPM-Solver config docs/code correction. Unrelated." - }, - { - "left": "pull_request:1968", - "right": "pull_request:2062", - "accept": false, - "reason": "Both relate to CPU offload, but one corrects docs/behavior broadly while the other fixes Paint by Example-specific offload mechanics." - }, - { - "left": "pull_request:21", - "right": "pull_request:43", - "accept": false, - "reason": "Model tests and the model card template are unrelated changes." - }, - { - "left": "pull_request:2197", - "right": "pull_request:260", - "accept": false, - "reason": "A StableDiffusionInpaintPipeline typo fix and a README spelling fix are different small edits." - }, - { - "left": "pull_request:2197", - "right": "pull_request:353", - "accept": false, - "reason": "One fixes an inpaint pipeline typo; the other fixes a UNet block error message. Different concrete issues." + "reason": "xFormers docs vs configuration docs touch different parts of the documentation and codebase." } ] }, @@ -185672,8 +185395,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 263, - "review_unit_count": 273, + "review_unit_index": 262, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -185681,6 +185404,14 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:1487", + "pull_request:1510", + "pull_request:168", + "pull_request:1763", + "pull_request:1950", + "pull_request:1968", + "pull_request:2062", + "pull_request:21", "pull_request:2197", "pull_request:2290", "pull_request:2333", @@ -185692,45 +185423,47 @@ "pull_request:2540", "pull_request:2569", "pull_request:2587", + "pull_request:260", "pull_request:2604", "pull_request:2608", "pull_request:2715", "pull_request:2788", "pull_request:2812", - "pull_request:2813", "pull_request:2814", - "pull_request:2823", - "pull_request:2861", - "pull_request:2922", - "pull_request:3", - "pull_request:3068", - "pull_request:3069", "pull_request:3166", - "pull_request:3183", - "pull_request:3241", - "pull_request:3318", - "pull_request:3554", + "pull_request:3343", + "pull_request:353", + "pull_request:356", "pull_request:3676", "pull_request:3744", "pull_request:3798", "pull_request:3821", - "pull_request:3979", "pull_request:407", "pull_request:408", "pull_request:421", "pull_request:4275", - "pull_request:4276", "pull_request:429", + "pull_request:43", "pull_request:4479", "pull_request:4721", "pull_request:500", - "pull_request:5280", + "pull_request:61", + "pull_request:63", "pull_request:718", - "pull_request:798", - "pull_request:828", - "pull_request:88" + "pull_request:75", + "pull_request:798" ], "soft_pairs": [ + "pull_request:1487|pull_request:500", + "pull_request:1510|pull_request:61", + "pull_request:1510|pull_request:63", + "pull_request:168|pull_request:356", + "pull_request:1763|pull_request:75", + "pull_request:1950|pull_request:3343", + "pull_request:1968|pull_request:2062", + "pull_request:21|pull_request:43", + "pull_request:2197|pull_request:260", + "pull_request:2197|pull_request:353", "pull_request:2197|pull_request:407", "pull_request:2197|pull_request:798", "pull_request:2290|pull_request:3166", @@ -185751,252 +185484,222 @@ "pull_request:2608|pull_request:2715", "pull_request:2608|pull_request:4479", "pull_request:2715|pull_request:718", - "pull_request:2812|pull_request:2814", - "pull_request:2812|pull_request:2922", - "pull_request:2813|pull_request:2814", - "pull_request:2813|pull_request:2922", - "pull_request:2814|pull_request:2823", - "pull_request:2861|pull_request:4276", - "pull_request:3|pull_request:88", - "pull_request:3068|pull_request:3318", - "pull_request:3069|pull_request:828", - "pull_request:3166|pull_request:4721", - "pull_request:3183|pull_request:3979", - "pull_request:3241|pull_request:5280", - "pull_request:3318|pull_request:3554" + "pull_request:2812|pull_request:2814" ], - "prepared_review_unit_hash": "0203f4c3ecc8c24623f5ac1ba0d77f341a7bf1d46c366cb590dea497bc4224af", + "prepared_review_unit_hash": "1f350d5380b8f4e4ec5e9a615a1d9934b2b5ee0244d0c7425d41c42c222237e5", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 33, - "serialized_chars": 97936, - "estimated_input_tokens": 24484, - "estimated_eval_tokens": 49224 + "soft_pair_count": 31, + "serialized_chars": 100221, + "estimated_input_tokens": 25056, + "estimated_eval_tokens": 50368 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly small, unrelated docs/typo/test cleanup PRs. There are two clear duplicate pairs: the identical Stable UnCLIP docs PRs, and the controlling_generation doc nits PR. Everything else is too different in file scope or bug/change target to merge as duplicates.", - "confidence": 0.67, + "summary": "Mostly a false-positive cluster: the items span unrelated docs nits, code fixes, example changes, and translation updates. The soft edges are driven by shared filenames or broad subsystem names, not the same underlying bug/change.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 2715 is the broadest representative cleanup: it spans several core model files plus a test, matching the cluster\u2019s general typo/fix pattern better than the narrow one-file doc PRs.", + "canonical_pr_reason": "No single PR covers the cluster; the items are unrelated docs and code changes with only superficial topic overlap.", "best_issue_reason": null, - "best_pr_reason": "PR 2715 is the best overall representative of this cluster because it is a generic multi-file typo cleanup in core code, whereas most other items are narrower or unrelated docs/test tweaks.", + "best_pr_reason": "No clear best PR. The largest PRs are still one-off fixes or docs updates, not a representative duplicate target for the whole cluster.", "soft_edge_verdicts": [ + { + "left": "pull_request:1487", + "right": "pull_request:500", + "accept": false, + "reason": "Unrelated docs changes: xformers API docs vs configuration subfolder docs." + }, + { + "left": "pull_request:1510", + "right": "pull_request:61", + "accept": false, + "reason": "Both are conversion-related, but one improves a broad converter while the other adds an unconditional conversion script." + }, + { + "left": "pull_request:1510", + "right": "pull_request:63", + "accept": false, + "reason": "Different conversion targets and code paths: scheduler/pipeline conversion work vs BDDM conversion script." + }, + { + "left": "pull_request:168", + "right": "pull_request:356", + "accept": false, + "reason": "Stable diffusion pipeline addition is unrelated to text-to-image fine-tuning example code." + }, + { + "left": "pull_request:1763", + "right": "pull_request:75", + "accept": false, + "reason": "Both touch attention code, but one fixes mask handling and the other changes Glide attention initialization." + }, + { + "left": "pull_request:1950", + "right": "pull_request:3343", + "accept": false, + "reason": "Different bugs: one is UNet graph config / OneFlow support, the other is a DPM-Solver config reference fix." + }, + { + "left": "pull_request:1968", + "right": "pull_request:2062", + "accept": false, + "reason": "Both are CPU offload fixes, but for different pipelines and different offload code paths." + }, + { + "left": "pull_request:21", + "right": "pull_request:43", + "accept": false, + "reason": "Different models/pipelines and different work: model tests vs model card/template and pipeline migration." + }, + { + "left": "pull_request:2197", + "right": "pull_request:260", + "accept": false, + "reason": "Inpaint pipeline typo fix vs README typo fix; unrelated artifacts and scopes." + }, + { + "left": "pull_request:2197", + "right": "pull_request:353", + "accept": false, + "reason": "Different typo fixes in unrelated pipeline code; same broad repo area is not enough." + }, { "left": "pull_request:2197", "right": "pull_request:407", "accept": false, - "reason": "Both are typo-oriented, but they fix different code paths in different files: inpaint pipeline input handling vs attention-slicing method naming." + "reason": "Different stable-diffusion pipeline typos; one is an inpaint signature fix, the other attention slicing API cleanup." }, { "left": "pull_request:2197", "right": "pull_request:798", "accept": false, - "reason": "Different artifacts and targets: StableDiffusionInpaintPipeline input typo vs a UNet2DConditionModel docstring fix." + "reason": "Different docstring/signature typo fixes in separate pipeline/model files." }, { "left": "pull_request:2290", "right": "pull_request:3166", "accept": false, - "reason": "Pix2Pix docs formatting is unrelated to the textual inversion loader example change." + "reason": "Pix2Pix docs typo vs textual inversion loader doc/example fix." }, { "left": "pull_request:2290", "right": "pull_request:408", "accept": false, - "reason": "Both are docs typos, but they touch different pages and different content; not the same underlying change." + "reason": "Pix2Pix docs formatting fix is unrelated to general docs typo cleanup." }, { "left": "pull_request:2333", "right": "pull_request:421", "accept": false, - "reason": "Ethical guidelines wording vs scheduler documentation; separate doc surfaces and fixes." + "reason": "Different documentation pages and different fixes: ethical guidelines formatting vs scheduler docs typo cleanup." }, { "left": "pull_request:2333", "right": "pull_request:4275", "accept": false, - "reason": "Different docs pages with no shared concrete change beyond being documentation." + "reason": "Ethical guidelines docs fix is unrelated to AutoPipeline docs." }, { "left": "pull_request:2367", "right": "pull_request:3798", "accept": false, - "reason": "PR 2367 adds checkpoint-limit plumbing; PR 3798 only fixes comments. Same training area, but not the same bug/change." + "reason": "Training-script feature addition vs comment cleanup; same files but not the same underlying change." }, { "left": "pull_request:2388", "right": "pull_request:2406", - "accept": true, - "reason": "Both update the same controlling_generation docs page with adjacent nits; this is effectively the same documentation change and could be merged together." + "accept": false, + "reason": "Same docs page, but 2388 adds the whole page and 2406 adds follow-up nits; not the same underlying change." }, { "left": "pull_request:2433", "right": "pull_request:429", "accept": false, - "reason": "SDEdit note in img2img docs is unrelated to the configuration docs update." + "reason": "SDEdit note in img2img docs vs configuration docs; unrelated documentation topics." }, { "left": "pull_request:2433", "right": "pull_request:500", "accept": false, - "reason": "Different docs topics and files: img2img SDEdit note vs subfolder/config docs." + "reason": "Img2Img docs note vs configuration subfolder docs; no shared bug or change." }, { "left": "pull_request:2533", "right": "pull_request:3676", "accept": false, - "reason": "The workflow/setup dependency tweak is unrelated to the doc-builder test PR." + "reason": "Slow-test workflow/rc1 setup is unrelated to a doc-builder test PR." }, { "left": "pull_request:2540", "right": "pull_request:2788", "accept": false, - "reason": "ControlNet docs correction and training-section toctree deduplication are different doc changes." + "reason": "ControlNet section docs vs training section deduplication; different pages and objectives." }, { "left": "pull_request:2540", "right": "pull_request:3744", "accept": false, - "reason": "ControlNet docs and workflow token rename are unrelated." + "reason": "ControlNet docs correction vs workflow secret rename; unrelated changes." }, { "left": "pull_request:2569", "right": "pull_request:3821", "accept": false, - "reason": "OpenVINO docs vs translated Korean docs; different pages and languages, not the same change." + "reason": "OpenVINO docs vs translated Korean docs; different content and purpose." }, { "left": "pull_request:2587", "right": "pull_request:3821", "accept": false, - "reason": "Both are translation PRs, but Chinese vs Korean doc additions are distinct content changes." + "reason": "Chinese translated docs set vs Korean translated docs page; separate translation additions." }, { "left": "pull_request:2604", "right": "pull_request:4275", "accept": false, - "reason": "A colab link fix in the homepage is unrelated to autopipeline docs." + "reason": "Colab link fix in index vs AutoPipeline docs correction; unrelated docs changes." }, { "left": "pull_request:2604", "right": "pull_request:4721", "accept": false, - "reason": "Homepage colab link fix vs Korean DreamBooth translation; different artifacts entirely." + "reason": "Colab link fix vs Korean DreamBooth translation fix; different documents and edits." }, { "left": "pull_request:2608", "right": "pull_request:2715", "accept": false, - "reason": "README punctuation fixes are not the same change as the multi-file code/test typo cleanup." + "reason": "README punctuation fixes vs broad code/docs typo cleanup across models and pipelines." }, { "left": "pull_request:2608", "right": "pull_request:4479", "accept": false, - "reason": "README typos vs assorted pipeline doc typos; same broad theme, but not one concrete fix." + "reason": "README punctuation fixes are unrelated to miscellaneous pipeline typo fixes." }, { "left": "pull_request:2715", "right": "pull_request:718", "accept": false, - "reason": "Multi-file code typos/test fix is not the same underlying change as older docs and README typo fixes." + "reason": "Broad typo cleanup across model files is not the same change as the earlier mixed docs/example typo PR." }, { "left": "pull_request:2812", "right": "pull_request:2814", "accept": false, - "reason": "Docs-only Stable UnCLIP update vs implementation work in Stable UnCLIP code; not the same change." - }, - { - "left": "pull_request:2812", - "right": "pull_request:2922", - "accept": false, - "reason": "Both are unCLIP docs-related, but one is a broader docs update while the other is a small punctuation fix in a different page." - }, - { - "left": "pull_request:2813", - "right": "pull_request:2814", - "accept": false, - "reason": "Same as above: docs update vs pipeline implementation change." - }, - { - "left": "pull_request:2813", - "right": "pull_request:2922", - "accept": false, - "reason": "Different docs PRs on different pages; not a duplicate concrete change." - }, - { - "left": "pull_request:2814", - "right": "pull_request:2823", - "accept": false, - "reason": "Implementation of Stable UnCLIP CPU offload and a docs improvement are different changes." - }, - { - "left": "pull_request:2861", - "right": "pull_request:4276", - "accept": false, - "reason": "A pipeline test for image_embeds=None is unrelated to a dependency/import smoke test." - }, - { - "left": "pull_request:3", - "right": "pull_request:88", - "accept": false, - "reason": "Foundational config/import plumbing and DDPM model work are not the same bug/change." - }, - { - "left": "pull_request:3068", - "right": "pull_request:3318", - "accept": false, - "reason": "Textual inversion docs and IF textual inversion code are related only by topic, not by the same concrete fix." - }, - { - "left": "pull_request:3069", - "right": "pull_request:828", - "accept": false, - "reason": "Both touch Flax tests, but they adjust different test expectations in different contexts; not one duplicate change." - }, - { - "left": "pull_request:3166", - "right": "pull_request:4721", - "accept": false, - "reason": "Textual inversion loader docs/example and Korean DreamBooth translation are unrelated." - }, - { - "left": "pull_request:3183", - "right": "pull_request:3979", - "accept": false, - "reason": "DreamBooth LoRA bugfix and SDXL bugfix address different code paths." - }, - { - "left": "pull_request:3241", - "right": "pull_request:5280", - "accept": false, - "reason": "Different fast-test adjustments for different pipelines; same testing theme but not the same change." - }, - { - "left": "pull_request:3318", - "right": "pull_request:3554", - "accept": false, - "reason": "IF textual inversion work and the Moreh textual inversion example PR are different changes with different scopes." - }, - { - "left": "pull_request:2812", - "right": "pull_request:2813", - "accept": true, - "reason": "These are effectively identical Stable UnCLIP docs PRs with the same title, files, and diff preview, so they are duplicate attempts at the same change." + "reason": "Stable unCLIP docs update vs Stable unCLIP implementation/offload changes; related feature area but different artifacts and scope." } ] }, @@ -186011,8 +185714,8 @@ "label": "pull_request", "component_index": 1, "component_count": 266, - "review_unit_index": 264, - "review_unit_count": 273, + "review_unit_index": 263, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -186020,9 +185723,21 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:11522", - "pull_request:12069", + "pull_request:2812", + "pull_request:2813", + "pull_request:2814", + "pull_request:2823", + "pull_request:2861", + "pull_request:2922", + "pull_request:3", + "pull_request:3068", + "pull_request:3069", + "pull_request:3166", + "pull_request:3183", + "pull_request:3241", + "pull_request:3318", "pull_request:3470", + "pull_request:3554", "pull_request:3625", "pull_request:3687", "pull_request:37", @@ -186032,9 +185747,12 @@ "pull_request:407", "pull_request:4082", "pull_request:42", + "pull_request:4276", "pull_request:4380", "pull_request:4652", + "pull_request:4721", "pull_request:5250", + "pull_request:5280", "pull_request:531", "pull_request:5407", "pull_request:5723", @@ -186047,29 +185765,26 @@ "pull_request:6215", "pull_request:6466", "pull_request:6475", - "pull_request:6647", "pull_request:6759", - "pull_request:70", - "pull_request:7015", - "pull_request:7050", - "pull_request:7068", - "pull_request:7118", - "pull_request:7451", "pull_request:75", - "pull_request:7564", "pull_request:774", - "pull_request:78", - "pull_request:7825", - "pull_request:7977", - "pull_request:800", "pull_request:824", - "pull_request:8498", - "pull_request:8656", - "pull_request:8845", - "pull_request:9206", - "pull_request:971" + "pull_request:828", + "pull_request:88" ], "soft_pairs": [ + "pull_request:2812|pull_request:2922", + "pull_request:2813|pull_request:2814", + "pull_request:2813|pull_request:2922", + "pull_request:2814|pull_request:2823", + "pull_request:2861|pull_request:4276", + "pull_request:3|pull_request:88", + "pull_request:3068|pull_request:3318", + "pull_request:3069|pull_request:828", + "pull_request:3166|pull_request:4721", + "pull_request:3183|pull_request:3979", + "pull_request:3241|pull_request:5280", + "pull_request:3318|pull_request:3554", "pull_request:3470|pull_request:5723", "pull_request:3625|pull_request:3687", "pull_request:3687|pull_request:6017", @@ -186088,234 +185803,505 @@ "pull_request:5723|pull_request:5727", "pull_request:580|pull_request:824", "pull_request:5983|pull_request:6466", - "pull_request:5983|pull_request:6475", - "pull_request:5983|pull_request:7015", - "pull_request:6647|pull_request:7564", - "pull_request:70|pull_request:78", - "pull_request:7050|pull_request:7068", - "pull_request:7050|pull_request:7451", - "pull_request:7068|pull_request:7118", - "pull_request:7564|pull_request:8656", - "pull_request:7825|pull_request:9206", - "pull_request:7977|pull_request:8845", - "pull_request:800|pull_request:971", - "pull_request:8498|pull_request:8845", - "pull_request:11522|pull_request:12069" + "pull_request:5983|pull_request:6475" ], - "prepared_review_unit_hash": "4b899d5d302562500ea4d51f99bff9856ac4df074803766995153211bf713ed7", + "prepared_review_unit_hash": "e4aa36be9bfda169cfd363e31446f77c02b94c6f282de70453b191518271d24b", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 98705, - "estimated_input_tokens": 24677, - "estimated_eval_tokens": 49610 + "serialized_chars": 94631, + "estimated_input_tokens": 23658, + "estimated_eval_tokens": 47572 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly noisy: it contains several unrelated docs/tests/typo PRs. Only a few soft edges look like true same-change duplicates, notably AudioLDM2, the early attention refactor, and the SVD doc edit.", - "confidence": 0.93, + "summary": "This cluster is mostly a grab bag of unrelated docs, tests, and small feature/fix PRs. The only clear duplicate-like item is the near-identical Stable unCLIP docs patch pair (2812/2813, not one of the provided soft edges); most other soft pairs are same subsystem at best, but not the same concrete bug/change.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR cleanly covers the whole cluster; it fragments into separate mini-pairs. If forced, 12069 is the strongest representative code-fix PR because it directly supersedes the same AudioLDM2 bug path as 11522.", + "canonical_pr_reason": "PR 2812 is the cleanest duplicate representative in the only obvious exact-match pair (2812/2813): same Stable unCLIP docs patch, same hunk and filenames, created minutes apart.", "best_issue_reason": null, - "best_pr_reason": "12069 is the best overall representative among the true duplicates: it fixes the AudioLDM2 transformer-version branch in the same pipeline code path as 11522 and is the more complete follow-up.", + "best_pr_reason": "PR 2814 is the most substantive representative here: it actually completes the Stable UnCLIP pipeline code path across multiple source files, whereas most other items are docs/test tweaks.", "soft_edge_verdicts": [ + { + "left": "pull_request:2812", + "right": "pull_request:2922", + "accept": false, + "reason": "Both are Stable unCLIP-related docs, but they change different pages and different content; not the same change." + }, + { + "left": "pull_request:2813", + "right": "pull_request:2814", + "accept": false, + "reason": "Docs-only Stable unCLIP update vs core pipeline implementation; related area, not the same bug or change." + }, + { + "left": "pull_request:2813", + "right": "pull_request:2922", + "accept": false, + "reason": "Different docs edits on different files; same topic, but not duplicate work." + }, + { + "left": "pull_request:2814", + "right": "pull_request:2823", + "accept": false, + "reason": "Code completion PR vs follow-up docs improvement; not the same underlying change." + }, + { + "left": "pull_request:2861", + "right": "pull_request:4276", + "accept": false, + "reason": "StableUnCLIPImg2Img test fix vs generic pipeline import test; unrelated." + }, + { + "left": "pull_request:3", + "right": "pull_request:88", + "accept": false, + "reason": "Foundational hub-loading work vs DDPM implementation work; different code paths and goals." + }, + { + "left": "pull_request:3068", + "right": "pull_request:3318", + "accept": false, + "reason": "Textual inversion docs cleanup vs IF textual inversion feature work; not the same artifact." + }, + { + "left": "pull_request:3069", + "right": "pull_request:828", + "accept": false, + "reason": "Both are Flax test updates, but they target different test suites and fixes." + }, + { + "left": "pull_request:3166", + "right": "pull_request:4721", + "accept": false, + "reason": "Text inversion loader doc example vs unrelated Korean DreamBooth doc edits." + }, + { + "left": "pull_request:3183", + "right": "pull_request:3979", + "accept": false, + "reason": "DreamBooth LoRA checkpoint bug vs SDXL attention bug; different subsystems." + }, + { + "left": "pull_request:3241", + "right": "pull_request:5280", + "accept": false, + "reason": "UnCLIP test tolerance change vs ControlNet fast-test resizing; unrelated tests." + }, + { + "left": "pull_request:3318", + "right": "pull_request:3554", + "accept": false, + "reason": "IF textual inversion WIP vs separate textual inversion/example branch; not the same change." + }, { "left": "pull_request:3470", "right": "pull_request:5723", "accept": false, - "reason": "Different feature areas: DreamBooth IF docs vs LCM docs; same broad docs category, not the same change." + "reason": "IF DreamBooth docs vs LCM docs; different model families and docs." }, { "left": "pull_request:3625", "right": "pull_request:3687", "accept": false, - "reason": "Both are doc-builder test/workflow tweaks, but they touch different files and are not the same concrete bug or change." + "reason": "Both are doc-builder test PRs, but they modify different files with different test intents." }, { "left": "pull_request:3687", "right": "pull_request:6017", "accept": false, - "reason": "Separate test-only doc-builder/title edits; not the same underlying fix." + "reason": "Same broad doc-builder testing theme, but different edits and different PRs." }, { "left": "pull_request:3687", "right": "pull_request:6018", "accept": false, - "reason": "Same general doc-builder testing area, but different edit and no shared concrete bug." + "reason": "Same broad doc-builder testing theme, but different edits and different PRs." }, { "left": "pull_request:37", "right": "pull_request:75", "accept": false, - "reason": "Related attention code, but PR 37 is a larger merge/refactor while 75 is a specific Glide attention bugfix; not the same concrete change." + "reason": "Attention refactor/merge vs Glide-specific attention fix; same subsystem, not same change." }, { "left": "pull_request:3850", "right": "pull_request:6215", "accept": false, - "reason": "AudioDiffusion test adjustments vs PixArt test initialization; unrelated test fixes." + "reason": "Audio diffusion test fix vs PixArt test initialization fix; unrelated test suites." }, { "left": "pull_request:3979", "right": "pull_request:4082", "accept": false, - "reason": "SDXL pipeline bugfix vs SDXL requirement/version doc change; different artifacts and problems." + "reason": "SDXL bug fix vs requirement pin fix; different problem classes." }, { "left": "pull_request:3985", "right": "pull_request:4380", "accept": false, - "reason": "Both are docs, but they address different sections and unrelated content changes." + "reason": "SDXL doc polish vs broader docs cleanup in other training pages; not duplicate content." }, { "left": "pull_request:407", "right": "pull_request:75", "accept": false, - "reason": "Attention slicing typo fix vs Glide attention implementation fix; not the same bug." + "reason": "Attention slicing typo fix vs Glide attention implementation fix; different changes." }, { "left": "pull_request:4082", "right": "pull_request:4652", "accept": false, - "reason": "Requirement pin change vs LoRA test adjustment; unrelated changes." + "reason": "Package requirement tweak vs LoRA SDXL test adjustment; unrelated." }, { "left": "pull_request:42", "right": "pull_request:824", "accept": false, - "reason": "Unrelated test additions in different subsystems (modeling utils vs Flax pipeline/testing utilities)." + "reason": "Old LDM unconditional test addition vs Flax pipeline test work; different code paths." }, { "left": "pull_request:4380", "right": "pull_request:6466", "accept": false, - "reason": "Different docs areas and different edits; both are minor doc cleanup, not one shared issue." + "reason": "Different docs pages and edits; both minor docs fixes, but not the same change." }, { "left": "pull_request:5250", "right": "pull_request:5911", "accept": false, - "reason": "DDPO docs entry vs IP-Adapter full support implementation/docs; different features." + "reason": "DDPO docs entry vs IP-Adapter Full support implementation; unrelated." }, { "left": "pull_request:531", "right": "pull_request:774", "accept": false, - "reason": "Both are tests, but they target different regressions in different pipelines." + "reason": "Two separate test-fix PRs in different eras and files; not the same bug." }, { "left": "pull_request:5407", "right": "pull_request:6759", "accept": false, - "reason": "Mixed-precision training example fix vs InstantID prompt embedding repetition; unrelated pipeline fixes." + "reason": "Mixed-precision training fix vs InstantID prompt-embedding shape fix; different pipelines and bugs." }, { "left": "pull_request:5723", "right": "pull_request:5727", "accept": false, - "reason": "LCM docs vs LCM script addition: same feature family, but not the same concrete bug/change." + "reason": "LCM docs vs LCM scripts; same feature area, but docs and implementation are separate changes." }, { "left": "pull_request:580", "right": "pull_request:824", "accept": false, - "reason": "Flax scheduler test work vs general testing utility changes; different issues." + "reason": "Early Flax scheduler test work vs later Flax pipeline test work; different changes." }, { "left": "pull_request:5983", "right": "pull_request:6466", - "accept": true, - "reason": "Both edit the same SVD doc file and are minor doc cleanups for the same page, so they look like the same underlying change." + "accept": false, + "reason": "Both touch SVD docs, but they make different edits to different sections; not a duplicate change." }, { "left": "pull_request:5983", "right": "pull_request:6475", "accept": false, - "reason": "SVD docs vs SVD pipeline type annotation; different artifacts." - }, + "reason": "SVD doc cleanup vs Stable Video Diffusion VAE type fix; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 264, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11428", + "pull_request:11522", + "pull_request:11546", + "pull_request:12069", + "pull_request:12490", + "pull_request:130", + "pull_request:13275", + "pull_request:13442", + "pull_request:13469", + "pull_request:1564", + "pull_request:2001", + "pull_request:2062", + "pull_request:327", + "pull_request:3397", + "pull_request:343", + "pull_request:3471", + "pull_request:349", + "pull_request:4262", + "pull_request:4666", + "pull_request:4749", + "pull_request:5201", + "pull_request:5639", + "pull_request:5983", + "pull_request:6645", + "pull_request:6647", + "pull_request:6759", + "pull_request:70", + "pull_request:7015", + "pull_request:7024", + "pull_request:7050", + "pull_request:7068", + "pull_request:7118", + "pull_request:7206", + "pull_request:7451", + "pull_request:7564", + "pull_request:78", + "pull_request:7825", + "pull_request:7977", + "pull_request:800", + "pull_request:8498", + "pull_request:8656", + "pull_request:8845", + "pull_request:9040", + "pull_request:9086", + "pull_request:9206", + "pull_request:9597", + "pull_request:971" + ], + "soft_pairs": [ + "pull_request:5983|pull_request:7015", + "pull_request:6647|pull_request:7564", + "pull_request:70|pull_request:78", + "pull_request:7050|pull_request:7068", + "pull_request:7050|pull_request:7451", + "pull_request:7068|pull_request:7118", + "pull_request:7564|pull_request:8656", + "pull_request:7825|pull_request:9206", + "pull_request:7977|pull_request:8845", + "pull_request:800|pull_request:971", + "pull_request:8498|pull_request:8845", + "pull_request:11522|pull_request:12069", + "pull_request:12490|pull_request:13442", + "pull_request:327|pull_request:343", + "pull_request:3397|pull_request:3471", + "pull_request:4262|pull_request:4666", + "pull_request:11428|pull_request:11546", + "pull_request:2001|pull_request:2062", + "pull_request:4749|pull_request:5201", + "pull_request:6645|pull_request:6759", + "pull_request:327|pull_request:349", + "pull_request:13275|pull_request:13469", + "pull_request:5639|pull_request:7024", + "pull_request:7206|pull_request:9086", + "pull_request:9040|pull_request:9597", + "pull_request:130|pull_request:1564" + ], + "prepared_review_unit_hash": "a91719b292ba1a940b22d79d10ff9f62e9c78c0d80feea855a7ecd34706f6fe4", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 97871, + "estimated_input_tokens": 24468, + "estimated_eval_tokens": 49192 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated docs/test/feature PRs that only look similar at the subsystem or wording level. The only strong duplicate-like pair is the two AudioLDM2 fixes, which both patch the same `generate_language_model` compatibility path for newer Transformers.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:12069 is the best canonical PR because it is the more complete AudioLDM2 compatibility fix: it keeps the version guard, moves `model_kwargs` handling into the shared path, and includes the test update, making it a better representative than the earlier partial fix in 11522.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:12069 is the strongest representative of the only true duplicate pair in this cluster; it fixes the concrete failing AudioLDM2 code path rather than a broader or unrelated docs/test change.", + "soft_edge_verdicts": [ { "left": "pull_request:5983", "right": "pull_request:7015", "accept": false, - "reason": "SVD docs vs single-file loader docs/code; unrelated changes." + "reason": "Both are docs edits, but they target different pages and unrelated fixes (SVD docs vs single-file SDXL docs)." }, { "left": "pull_request:6647", "right": "pull_request:7564", "accept": false, - "reason": "Both are test-related, but they address different pipelines and failures." + "reason": "Both touch tests, but they cover different pipelines and different failure modes; not the same bug." }, { "left": "pull_request:70", "right": "pull_request:78", - "accept": true, - "reason": "Same attention/unet refactor line: PR 78 continues and supersedes the mid-block attention work started in 70." + "accept": false, + "reason": "Both are broad architectural mid-block changes, but not the same concrete change or failure; too coarse to merge as duplicates." }, { "left": "pull_request:7050", "right": "pull_request:7068", "accept": false, - "reason": "Both are typo cleanups, but in different docs/examples and not the same concrete typo fix." + "reason": "Both are typo/docs PRs, but they edit different files and unrelated content." }, { "left": "pull_request:7050", "right": "pull_request:7451", "accept": false, - "reason": "Broad typo-fix PRs touching different files; too unrelated to treat as duplicates." + "reason": "Same general docs-typo theme, but different documentation areas and unrelated edits." }, { "left": "pull_request:7068", "right": "pull_request:7118", "accept": false, - "reason": "Both are doc typo fixes, but they cover different doc sections and different exact edits." + "reason": "Both are docs typo cleanups, but they fix different docs pages and examples." }, { "left": "pull_request:7564", "right": "pull_request:8656", "accept": false, - "reason": "Different test failures and code paths; not the same underlying bug." + "reason": "Both are test maintenance PRs, but they address different model behaviors and checkpoint handling." }, { "left": "pull_request:7825", "right": "pull_request:9206", "accept": false, - "reason": "Model device-map tests vs xformers pipeline tests; unrelated regressions." + "reason": "Different test suites and different issues: device-map/offload tests vs xformers attention tests." }, { "left": "pull_request:7977", "right": "pull_request:8845", "accept": false, - "reason": "Docs typo in optimization guide vs typo fixes in PAG tests; different scopes." + "reason": "Both are typo fixes, but they are in different docs/test contexts and not the same change." }, { "left": "pull_request:800", "right": "pull_request:971", "accept": false, - "reason": "Different feature additions (VQ Diffusion vs bit diffusion), not duplicate work." + "reason": "Both relate to diffusion model additions, but they are different models/features (VQ Diffusion vs bit diffusion)." }, { "left": "pull_request:8498", "right": "pull_request:8845", "accept": false, - "reason": "SD3 doc typo vs PAG test typo; unrelated." + "reason": "Unrelated fixes: a single doc typo versus typo corrections in PAG tests." }, { "left": "pull_request:11522", "right": "pull_request:12069", "accept": true, - "reason": "Same AudioLDM2 cache_position/version-guard bug in the same file; 12069 is the follow-up fix to 11522." + "reason": "Same AudioLDM2 `generate_language_model` compatibility bug for Transformers main/newer versions; 12069 is a direct follow-up/complete fix to 11522." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13442", + "accept": false, + "reason": "Different topics: a CLAP config test fix vs training-script test fixes." + }, + { + "left": "pull_request:327", + "right": "pull_request:343", + "accept": false, + "reason": "Both are type-hint cleanup PRs, but they touch different schedulers/pipelines and are not the same underlying change." + }, + { + "left": "pull_request:3397", + "right": "pull_request:3471", + "accept": false, + "reason": "Both are docs, but one is a local-path note and the other is a torch.compile note for DreamBooth." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4666", + "accept": false, + "reason": "Both involve SDXL/LoRA, but one is DreamBooth training/model fixes and the other is ControlNet SDXL LoRA support; different code paths." + }, + { + "left": "pull_request:11428", + "right": "pull_request:11546", + "accept": false, + "reason": "FramePack pipeline addition vs a LoRA-loader mapping for that new model; related feature area, but not the same change." + }, + { + "left": "pull_request:2001", + "right": "pull_request:2062", + "accept": false, + "reason": "Completely different scopes: a ColossalAI example typo fix vs Paint by Example CPU offload behavior." + }, + { + "left": "pull_request:4749", + "right": "pull_request:5201", + "accept": false, + "reason": "Both are LoRA-related, but one is SDXL DreamBooth checkpointing and the other is xformers/LoRA processor behavior." + }, + { + "left": "pull_request:6645", + "right": "pull_request:6759", + "accept": false, + "reason": "Both are community example edits, but they implement different community pipelines with no shared bug/change." + }, + { + "left": "pull_request:327", + "right": "pull_request:349", + "accept": false, + "reason": "Both are type-hint PRs, but they affect different scheduler classes and are unrelated." + }, + { + "left": "pull_request:13275", + "right": "pull_request:13469", + "accept": false, + "reason": "Different test fixes in different areas: Hunyuan/LLaVA kwargs vs deprecated attention-processor testing." + }, + { + "left": "pull_request:5639", + "right": "pull_request:7024", + "accept": false, + "reason": "Both are Animatediff docs, but they add different feature sections (Motion LoRAs vs AnimateLCM)." + }, + { + "left": "pull_request:7206", + "right": "pull_request:9086", + "accept": false, + "reason": "Both touch training scripts, but one is tag metadata and the other is actual Flux DreamBooth/LoRA training code." + }, + { + "left": "pull_request:9040", + "right": "pull_request:9597", + "accept": false, + "reason": "Same general 'differential diffusion' concept, but for different community pipelines/models; not the same patch." + }, + { + "left": "pull_request:130", + "right": "pull_request:1564", + "accept": false, + "reason": "Both mention accelerate config, but one adds a setup step and the other adds a default config option for DreamBooth docs." } ] }, @@ -186331,7 +186317,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 265, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -186340,17 +186326,9 @@ "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10858", - "pull_request:11428", - "pull_request:11546", + "pull_request:12006", "pull_request:12018", - "pull_request:12490", - "pull_request:130", - "pull_request:13275", - "pull_request:13442", - "pull_request:13469", - "pull_request:1564", - "pull_request:2001", - "pull_request:2062", + "pull_request:12041", "pull_request:2172", "pull_request:2406", "pull_request:2490", @@ -186359,50 +186337,43 @@ "pull_request:2637", "pull_request:2693", "pull_request:2748", - "pull_request:327", - "pull_request:3397", - "pull_request:343", - "pull_request:3471", - "pull_request:349", + "pull_request:3351", + "pull_request:3476", + "pull_request:3586", + "pull_request:3589", + "pull_request:3590", "pull_request:3942", "pull_request:3945", "pull_request:4262", "pull_request:4343", + "pull_request:4498", + "pull_request:4572", "pull_request:4626", - "pull_request:4666", "pull_request:4749", "pull_request:4833", "pull_request:4862", + "pull_request:5003", + "pull_request:5157", "pull_request:5201", - "pull_request:5639", + "pull_request:5615", "pull_request:5665", + "pull_request:5768", + "pull_request:5828", "pull_request:6342", - "pull_request:6645", "pull_request:6759", - "pull_request:7024", - "pull_request:7206", + "pull_request:7050", + "pull_request:7131", + "pull_request:7246", + "pull_request:7825", + "pull_request:7827", "pull_request:788", + "pull_request:8230", + "pull_request:8797", "pull_request:8913", - "pull_request:9040", - "pull_request:9086", "pull_request:9506", - "pull_request:9597" + "pull_request:9696" ], "soft_pairs": [ - "pull_request:12490|pull_request:13442", - "pull_request:327|pull_request:343", - "pull_request:3397|pull_request:3471", - "pull_request:4262|pull_request:4666", - "pull_request:11428|pull_request:11546", - "pull_request:2001|pull_request:2062", - "pull_request:4749|pull_request:5201", - "pull_request:6645|pull_request:6759", - "pull_request:327|pull_request:349", - "pull_request:13275|pull_request:13469", - "pull_request:5639|pull_request:7024", - "pull_request:7206|pull_request:9086", - "pull_request:9040|pull_request:9597", - "pull_request:130|pull_request:1564", "pull_request:2693|pull_request:788", "pull_request:4833|pull_request:5201", "pull_request:8913|pull_request:9506", @@ -186413,186 +186384,193 @@ "pull_request:2637|pull_request:2748", "pull_request:3945|pull_request:4343", "pull_request:4626|pull_request:6342", - "pull_request:4862|pull_request:5665" + "pull_request:4862|pull_request:5665", + "pull_request:5003|pull_request:5157", + "pull_request:5768|pull_request:5828", + "pull_request:7246|pull_request:8797", + "pull_request:7825|pull_request:9696", + "pull_request:7827|pull_request:8230", + "pull_request:12006|pull_request:12041", + "pull_request:3351|pull_request:3476", + "pull_request:3589|pull_request:4498", + "pull_request:3586|pull_request:3590", + "pull_request:4262|pull_request:5615", + "pull_request:4572|pull_request:4749", + "pull_request:4749|pull_request:6759", + "pull_request:7050|pull_request:7131" ], - "prepared_review_unit_hash": "1e8e37d214f59d0a7e4da24af9218a6a5171cdd976bd642308c9062896c15c09", + "prepared_review_unit_hash": "1765ba9f485742a1de799413f2913d2088f3e3129277a29a2a4c874d5e335af5", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 94538, - "estimated_input_tokens": 23635, - "estimated_eval_tokens": 47526 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 94485, + "estimated_input_tokens": 23622, + "estimated_eval_tokens": 47500 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly superficial matches (same broad area, similar wording, or same docs/tests folder) rather than true duplicate bugs/changes. I don\u2019t see a single underlying issue or PR that unifies the cluster.", - "confidence": 0.95, + "summary": "This is a heterogeneous cluster, not a duplicate set: it mixes unrelated docs edits, example updates, scheduler/LoRA fixes, ControlNet bugs, tests, and workflow changes. There is no single underlying bug or change tying these PRs together.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: the items span unrelated docs, tests, and feature work, with similarity mostly coming from shared keywords or neighboring subsystems.", + "canonical_pr_reason": "No single canonical PR fits the cluster. The items span many unrelated code paths; the broadest docs cleanup (PR 7050) is only a loose representative, not a true duplicate anchor.", "best_issue_reason": null, - "best_pr_reason": "No best PR candidate stands out because there is no coherent duplicate topic to canonicalize across the set.", + "best_pr_reason": "PR 7050 is the closest representative only because it is a broad cross-file docs typo cleanup, but it still does not subsume the rest of the unrelated fixes in this cluster.", "soft_edge_verdicts": [ { - "left": "pull_request:12490", - "right": "pull_request:13442", - "accept": false, - "reason": "Both touch tests, but for different models and failures; not the same bug or fix." - }, - { - "left": "pull_request:327", - "right": "pull_request:343", + "left": "pull_request:2693", + "right": "pull_request:788", "accept": false, - "reason": "Different scheduler/pipeline components and unrelated type-hint cleanup; not one concrete change." + "reason": "One is an inpaint doc image-link fix; the other is a PIL sampler warning/code compatibility change. Different artifacts and no shared bug." }, { - "left": "pull_request:3397", - "right": "pull_request:3471", + "left": "pull_request:4833", + "right": "pull_request:5201", "accept": false, - "reason": "Both are docs edits, but they address different pages and different guidance." + "reason": "Both involve LoRA, but PR 4833 fixes unfuse/save behavior while PR 5201 fixes xformers/attention processor LoRA handling across models. Different concrete code paths." }, { - "left": "pull_request:4262", - "right": "pull_request:4666", + "left": "pull_request:8913", + "right": "pull_request:9506", "accept": false, - "reason": "One is DreamBooth LoRA training fixes; the other is ControlNet SDXL LoRA support. Same theme, different code paths." + "reason": "PR 8913 fixes sharded checkpoint loading from a subfolder; PR 9506 is a docs change adding an import and path fix. Unrelated." }, { - "left": "pull_request:11428", - "right": "pull_request:11546", + "left": "pull_request:10858", + "right": "pull_request:12018", "accept": false, - "reason": "FramePack pipeline implementation vs PEFT loader mapping support; related feature area, but not the same fix/change." + "reason": "Both are docs edits, but one changes a pipeline overview table entry while the other fixes a quantization docs link label. Not the same issue." }, { - "left": "pull_request:2001", - "right": "pull_request:2062", + "left": "pull_request:2172", + "right": "pull_request:2595", "accept": false, - "reason": "Completely unrelated examples/tests changes in different subsystems." + "reason": "Multi-GPU validation in training scripts and scheduler switching for DEIS/DPM/UniPC are unrelated problems in different subsystems." }, { - "left": "pull_request:4749", - "right": "pull_request:5201", + "left": "pull_request:2406", + "right": "pull_request:2490", "accept": false, - "reason": "One is SDXL DreamBooth checkpointing; the other is xformers/LoRA processor behavior. Not the same issue." + "reason": "These are separate documentation updates in the same file, but one is a generic nits edit and the other adds a MultiDiffusion section. Not duplicates." }, { - "left": "pull_request:6645", - "right": "pull_request:6759", + "left": "pull_request:2526", + "right": "pull_request:3942", "accept": false, - "reason": "Different community README/pipeline changes; no shared underlying bug or feature." + "reason": "PR 2526 adjusts slow-test tolerances; PR 3942 fixes xformers-related NumPy test behavior. Different test failures and causes." }, { - "left": "pull_request:327", - "right": "pull_request:349", + "left": "pull_request:2637", + "right": "pull_request:2748", "accept": false, - "reason": "Separate scheduler classes with independent type-hint updates, not one duplicate fix." + "reason": "These are unrelated typo fixes in different docs pages; they do not reflect the same underlying change." }, { - "left": "pull_request:13275", - "right": "pull_request:13469", + "left": "pull_request:3945", + "right": "pull_request:4343", "accept": false, - "reason": "Hunyuan test kwargs vs attention-processor test deprecation handling; unrelated test fixes." + "reason": "Both add a LoRA rank argument, but to different DreamBooth training scripts (base vs SDXL). Similar feature family, not the same concrete change." }, { - "left": "pull_request:5639", - "right": "pull_request:7024", + "left": "pull_request:4626", + "right": "pull_request:6342", "accept": false, - "reason": "Both are AnimateDiff docs, but they document different capabilities (Motion LoRAs vs AnimateLCM)." + "reason": "One fixes text2image docs links; the other adds an installation note about Python 3.12 on Windows. Unrelated." }, { - "left": "pull_request:7206", - "right": "pull_request:9086", + "left": "pull_request:4862", + "right": "pull_request:5665", "accept": false, - "reason": "Training-script tags vs FLUX DreamBooth/LoRA scripts; different changes and scopes." + "reason": "Different ControlNet img2img bugs: one is an SDXL MultiControlNetModel height/width issue, the other passes dimensions into image preprocessing. Not the same bug." }, { - "left": "pull_request:9040", - "right": "pull_request:9597", + "left": "pull_request:5003", + "right": "pull_request:5157", "accept": false, - "reason": "Both update the community README, but they add different pipeline entries and are not duplicates." + "reason": "PR 5003 fixes LoRA fuse/unfuse logic; PR 5157 fixes LoRA conversion naming for ControlNet/Kohya formats. Different code paths." }, { - "left": "pull_request:130", - "right": "pull_request:1564", + "left": "pull_request:5768", + "right": "pull_request:5828", "accept": false, - "reason": "Both mention accelerate config, but one adds setup instructions broadly and the other adds a default-config option in DreamBooth docs." + "reason": "Docs cleanup versus a memory test annotation for the consistency decoder. No shared bug or change." }, { - "left": "pull_request:2693", - "right": "pull_request:788", + "left": "pull_request:7246", + "right": "pull_request:8797", "accept": false, - "reason": "A doc image-link fix and a PIL sampler warning fix are unrelated." + "reason": "VAE tiling device handling and sharded checkpoint tests are unrelated test fixes." }, { - "left": "pull_request:4833", - "right": "pull_request:5201", + "left": "pull_request:7825", + "right": "pull_request:9696", "accept": false, - "reason": "LoRA state/unfuse logic vs xformers LoRA processor behavior; different code paths and failure modes." + "reason": "Model offload test guards and a GitHub Actions workflow update are unrelated." }, { - "left": "pull_request:8913", - "right": "pull_request:9506", + "left": "pull_request:7827", + "right": "pull_request:8230", "accept": false, - "reason": "Sharded checkpoint loading fix vs unrelated docs path/imageio update." + "reason": "Both mention CPU offloading, but one is a docstring example fix and the other is a broad usage/docs update. Not the same concrete issue." }, { - "left": "pull_request:10858", - "right": "pull_request:12018", + "left": "pull_request:12006", + "right": "pull_request:12041", "accept": false, - "reason": "One is pipeline overview wording; the other is quantization docs terminology/link cleanup. Not the same change." + "reason": "One adds Wan 5b i2v support; the other fixes VAE patchify behavior. Different components and bugs." }, { - "left": "pull_request:2172", - "right": "pull_request:2595", + "left": "pull_request:3351", + "right": "pull_request:3476", "accept": false, - "reason": "Multi-GPU validation in training scripts vs scheduler config switching; no duplicate underlying bug." + "reason": "Both add Karras sigmas, but to different schedulers with separate implementations. Similar enhancement theme, not one duplicate change." }, { - "left": "pull_request:2406", - "right": "pull_request:2490", + "left": "pull_request:3589", + "right": "pull_request:4498", "accept": false, - "reason": "Intro typo/wording cleanup vs adding a new MultiDiffusion section; different doc changes." + "reason": "PR 3589 updates README examples/tricks; PR 4498 adds SDXL reference pipeline support. Docs-only vs implementation work, not duplicates." }, { - "left": "pull_request:2526", - "right": "pull_request:3942", + "left": "pull_request:3586", + "right": "pull_request:3590", "accept": false, - "reason": "Both adjust tests, but they address different failures and test suites." + "reason": "These are separate community pipelines/example additions (tiling vs canvas), not the same underlying feature." }, { - "left": "pull_request:2637", - "right": "pull_request:2748", + "left": "pull_request:4262", + "right": "pull_request:5615", "accept": false, - "reason": "Two separate documentation wording fixes in the same file family, not the same underlying issue." + "reason": "Different SDXL-related changes in different pipelines: DreamBooth LoRA fixes versus reverting load_lora behavior in the T2I adapter." }, { - "left": "pull_request:3945", - "right": "pull_request:4343", + "left": "pull_request:4572", + "right": "pull_request:4749", "accept": false, - "reason": "Same idea of adding a rank arg, but to different training scripts (SD1.5 vs SDXL); separate PRs, not one duplicate change." + "reason": "One renames network_alpha handling in the text-to-image LoRA script; the other fixes SDXL DreamBooth LoRA checkpoint loading. Different problems." }, { - "left": "pull_request:4626", - "right": "pull_request:6342", + "left": "pull_request:4749", + "right": "pull_request:6759", "accept": false, - "reason": "A docs link fix and a rejected installation note are unrelated." + "reason": "Checkpoint loading for SDXL DreamBooth LoRA and prompt embedding repetition in InstantID are unrelated." }, { - "left": "pull_request:4862", - "right": "pull_request:5665", + "left": "pull_request:7050", + "right": "pull_request:7131", "accept": false, - "reason": "Both are ControlNet img2img fixes, but one handles MultiControlNet height/width calculation and the other passes size into preprocessing; different bugs." + "reason": "Both are typo/docs cleanup PRs, but they touch different documentation areas and do not address the same underlying issue." } ] }, @@ -186608,7 +186586,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 266, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -186616,13 +186594,15 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ + "pull_request:10192", "pull_request:11353", "pull_request:11479", + "pull_request:11655", "pull_request:11726", "pull_request:11728", - "pull_request:12006", - "pull_request:12041", "pull_request:1229", + "pull_request:1269", + "pull_request:1508", "pull_request:1527", "pull_request:154", "pull_request:1554", @@ -186630,55 +186610,39 @@ "pull_request:168", "pull_request:1788", "pull_request:1951", + "pull_request:2009", "pull_request:2367", "pull_request:2565", "pull_request:2693", "pull_request:2892", "pull_request:2928", - "pull_request:3351", + "pull_request:2976", "pull_request:3370", "pull_request:3446", - "pull_request:3476", - "pull_request:3586", - "pull_request:3589", - "pull_request:3590", + "pull_request:3471", + "pull_request:3568", "pull_request:3641", - "pull_request:4262", - "pull_request:4498", - "pull_request:4572", - "pull_request:4749", - "pull_request:5003", - "pull_request:5157", - "pull_request:5615", - "pull_request:5768", - "pull_request:5828", - "pull_request:6759", - "pull_request:7050", - "pull_request:7131", - "pull_request:7246", + "pull_request:4272", + "pull_request:4621", + "pull_request:4856", + "pull_request:5090", + "pull_request:5299", + "pull_request:5391", + "pull_request:5411", + "pull_request:5759", + "pull_request:6119", + "pull_request:6645", + "pull_request:7200", + "pull_request:7319", "pull_request:7451", - "pull_request:7825", - "pull_request:7827", - "pull_request:8230", - "pull_request:8797", + "pull_request:7857", + "pull_request:7919", + "pull_request:822", + "pull_request:827", "pull_request:9207", - "pull_request:9696", "pull_request:98" ], "soft_pairs": [ - "pull_request:5003|pull_request:5157", - "pull_request:5768|pull_request:5828", - "pull_request:7246|pull_request:8797", - "pull_request:7825|pull_request:9696", - "pull_request:7827|pull_request:8230", - "pull_request:12006|pull_request:12041", - "pull_request:3351|pull_request:3476", - "pull_request:3589|pull_request:4498", - "pull_request:3586|pull_request:3590", - "pull_request:4262|pull_request:5615", - "pull_request:4572|pull_request:4749", - "pull_request:4749|pull_request:6759", - "pull_request:7050|pull_request:7131", "pull_request:7451|pull_request:9207", "pull_request:11353|pull_request:11726", "pull_request:11479|pull_request:11728", @@ -186692,198 +186656,211 @@ "pull_request:1951|pull_request:3370", "pull_request:2367|pull_request:2928", "pull_request:2892|pull_request:3446", - "pull_request:2892|pull_request:3641" + "pull_request:2892|pull_request:3641", + "pull_request:2892|pull_request:4272", + "pull_request:2976|pull_request:4856", + "pull_request:3568|pull_request:5411", + "pull_request:4621|pull_request:6119", + "pull_request:5090|pull_request:6645", + "pull_request:5391|pull_request:5759", + "pull_request:7319|pull_request:7857", + "pull_request:822|pull_request:827", + "pull_request:10192|pull_request:11655", + "pull_request:1269|pull_request:2009", + "pull_request:1508|pull_request:3471", + "pull_request:5299|pull_request:5759", + "pull_request:7200|pull_request:7919" ], - "prepared_review_unit_hash": "08df8cf5947af0ffa307df7f79b414e7bfc34fb327958e5264f0d22f3f73352c", + "prepared_review_unit_hash": "45a9ba5aed339a066c7cdd04aa5be3f3d8699ce6b40bc8595835d15bd7884339", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 98082, - "estimated_input_tokens": 24521, - "estimated_eval_tokens": 49298 + "serialized_chars": 101394, + "estimated_input_tokens": 25349, + "estimated_eval_tokens": 50954 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous: docs typos, training-script tweaks, scheduler changes, Wan feature work, and a few unrelated tests/utilities. I don\u2019t see any true duplicate PR pairs among the soft edges.", - "confidence": 0.9, + "summary": "This cluster is mostly a false-positive grouping of unrelated PRs: docs tweaks, feature additions, and bug fixes across different subsystems. None of the soft edges look like the same underlying change, so I would not deduplicate any pair.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No clear canonical PR for deduping; the set is too mixed and none of the soft-edge pairs are the same concrete change.", + "canonical_pr_reason": "No single canonical PR: the set is heterogeneous and does not share one concrete bug/change. If forced to pick a representative, PR 5759 is the broadest docs/type-hint cleanup, but it is not a true duplicate target.", "best_issue_reason": null, - "best_pr_reason": "If a single PR must represent the cluster, #12006 is the best anchor because it is the broadest substantive code change here and sits in the only loosely related Wan subgroup.", + "best_pr_reason": "PR 5759 is the closest to an umbrella representative because it spans the broadest docs/type-hint maintenance pass, but the cluster still lacks a true common underlying fix.", "soft_edge_verdicts": [ { - "left": "pull_request:5003", - "right": "pull_request:5157", + "left": "pull_request:7451", + "right": "pull_request:9207", "accept": false, - "reason": "Both are LoRA-related, but they fix different problems: fuse/unfuse logic vs. Kohya/controlnet LoRA name conversion." + "reason": "Both are docs-only, but one fixes typos across training docs while the other updates CPU offloading usage guidance; different change targets and intent." }, { - "left": "pull_request:5768", - "right": "pull_request:5828", + "left": "pull_request:11353", + "right": "pull_request:11726", "accept": false, - "reason": "Unrelated changes: documentation cleanup for consistency decoder vs. a memory-related test tweak." + "reason": "Different Wan work: FLF2V support versus Wan LoRA conversion fixes. They touch the same project area but not the same bug or feature." }, { - "left": "pull_request:7246", - "right": "pull_request:8797", + "left": "pull_request:11479", + "right": "pull_request:11728", "accept": false, - "reason": "Different test fixes in the same file, but they target separate behaviors: VAE tiling vs. sharded checkpoint handling." + "reason": "One is a doc title rename for Intel Gaudi; the other adds quantization libraries to the doc-builder Docker image. Unrelated changes." }, { - "left": "pull_request:7825", - "right": "pull_request:9696", + "left": "pull_request:1229", + "right": "pull_request:1564", "accept": false, - "reason": "One is a model-device-map test guard; the other changes a GitHub Actions workflow. Not the same bug/change." + "reason": "Both touch DreamBooth docs, but one is about bitsandbytes installation and the other about default accelerate config. Different user problems." }, { - "left": "pull_request:7827", - "right": "pull_request:8230", + "left": "pull_request:1229", + "right": "pull_request:2367", "accept": false, - "reason": "Both mention CPU offloading, but one fixes a docstring example and the other updates broader docs/examples. Not one concrete change." + "reason": "Install guidance for bitsandbytes is unrelated to adding checkpoint-retention controls in training scripts." }, { - "left": "pull_request:12006", - "right": "pull_request:12041", + "left": "pull_request:1527", + "right": "pull_request:2693", "accept": false, - "reason": "Both are Wan-related, but one adds 5b i2v pipeline support while the other fixes VAE patchify behavior. Different code paths." + "reason": "Stable Diffusion inpaint tensor-input support is a code fix; the other is only a broken image link in docs." }, { - "left": "pull_request:3351", - "right": "pull_request:3476", + "left": "pull_request:154", + "right": "pull_request:168", "accept": false, - "reason": "Same feature idea (`use_karras_sigmas`) but applied to different schedulers and code paths; not a single mergeable fix." + "reason": "A conversion script for original SD checkpoints is not the same as the initial Stable Diffusion pipeline integration." }, { - "left": "pull_request:3589", - "right": "pull_request:4498", + "left": "pull_request:154", + "right": "pull_request:98", "accept": false, - "reason": "One adds README reference tricks; the other implements SDXL reference support. Documentation vs. feature implementation." + "reason": "These are different conversion scripts for different model families (LDM/Stable Diffusion vs NCSN++)." }, { - "left": "pull_request:3586", - "right": "pull_request:3590", + "left": "pull_request:1554", + "right": "pull_request:2565", "accept": false, - "reason": "Both are Mixture-themed community examples, but they add different pipelines (tiling vs. canvas) and are not the same change." + "reason": "Both are docs corrections, but they target different model APIs and different dimensionality docs; no shared underlying bug." }, { - "left": "pull_request:4262", - "right": "pull_request:5615", + "left": "pull_request:1788", + "right": "pull_request:3641", "accept": false, - "reason": "Different SDXL LoRA-related fixes in different components; not the same underlying bug." + "reason": "DDIM sample-prediction handling and inpainting initial-latent handling are separate scheduler/pipeline bugs." }, { - "left": "pull_request:4572", - "right": "pull_request:4749", + "left": "pull_request:1951", + "right": "pull_request:3370", "accept": false, - "reason": "One fixes a parameter name in an example; the other fixes SDXL DreamBooth LoRA checkpoint loading/splitting." + "reason": "ColossalAI upgrade work is unrelated to PyTorch 2.0 documentation updates." }, { - "left": "pull_request:4749", - "right": "pull_request:6759", + "left": "pull_request:2367", + "right": "pull_request:2928", "accept": false, - "reason": "Checkpoint-loading fixes for DreamBooth vs. InstantID prompt-image embedding repetition. Unrelated code paths." + "reason": "Checkpoint-limit plumbing in training scripts and ControlNet local dataset loading are independent features." }, { - "left": "pull_request:7050", - "right": "pull_request:7131", + "left": "pull_request:2892", + "right": "pull_request:3446", "accept": false, - "reason": "Both are typo/docs cleanup, but they touch different docs and different issues; not a single duplicated fix." + "reason": "Allowing embed inputs in the upscaler pipeline is unrelated to a warning-message wording fix." }, { - "left": "pull_request:7451", - "right": "pull_request:9207", + "left": "pull_request:2892", + "right": "pull_request:3641", "accept": false, - "reason": "Different docs fixes: broad typo cleanup vs. CPU offloading usage updates. Too broad to be the same change." + "reason": "These are different inpainting/upscaler behaviors with distinct code paths and failure modes." }, { - "left": "pull_request:11353", - "right": "pull_request:11726", + "left": "pull_request:2892", + "right": "pull_request:4272", "accept": false, - "reason": "Wan feature support vs. Wan LoRA conversion support; same ecosystem, but different functionality and code paths." + "reason": "Upgrader input validation is unrelated to fixing missing optional imports in pipeline folders." }, { - "left": "pull_request:11479", - "right": "pull_request:11728", + "left": "pull_request:2976", + "right": "pull_request:4856", "accept": false, - "reason": "Doc title rename for Gaudi vs. Dockerfile quant-library additions. Completely different changes." + "reason": "Both are docs link fixes, but they address different pages and different broken links, not the same change." }, { - "left": "pull_request:1229", - "right": "pull_request:1564", + "left": "pull_request:3568", + "right": "pull_request:5411", "accept": false, - "reason": "Both touch DreamBooth docs, but one is about bitsandbytes installation and the other about default accelerate config." + "reason": "Both involve documentation links, but one fixes general tutorial links and the other is specifically about textual inversion training links." }, { - "left": "pull_request:1229", - "right": "pull_request:2367", + "left": "pull_request:4621", + "right": "pull_request:6119", "accept": false, - "reason": "DreamBooth docs install guidance vs. training-script checkpoint limits. Not the same bug or change." + "reason": "Multi-adapter downscale-factor validation and fp16 LoRA gradient scaling are unrelated training/runtime issues." }, { - "left": "pull_request:1527", - "right": "pull_request:2693", + "left": "pull_request:5090", + "right": "pull_request:6645", "accept": false, - "reason": "Inpaint tensor-input support vs. a broken image-link doc fix. No overlap in underlying issue." + "reason": "A T2I Adapter bugfix is not the same as a community README cleanup." }, { - "left": "pull_request:154", - "right": "pull_request:168", + "left": "pull_request:5391", + "right": "pull_request:5759", "accept": false, - "reason": "A conversion script addition and the stable diffusion pipeline introduction are unrelated changes." + "reason": "Both are broad docs/type-hint cleanups, but they are different edit sets with different scopes and not one concrete fix." }, { - "left": "pull_request:154", - "right": "pull_request:98", + "left": "pull_request:7319", + "right": "pull_request:7857", "accept": false, - "reason": "Different conversion scripts for different model families; not the same underlying bug/change." + "reason": "Adding EDM scheduler docs and adding PixArt-Sigma docs are separate documentation additions." }, { - "left": "pull_request:1554", - "right": "pull_request:2565", + "left": "pull_request:822", + "right": "pull_request:827", "accept": false, - "reason": "Both are doc corrections, but they refer to different model APIs and different doc pages." + "reason": "Flax scheduler API changes and PyTorch/Flax checkpoint-loading alignment are different compatibility fixes." }, { - "left": "pull_request:1788", - "right": "pull_request:3641", + "left": "pull_request:10192", + "right": "pull_request:11655", "accept": false, - "reason": "DDIM scheduler bug fix vs. inpainting latent handling. Different concrete problems." + "reason": "Flux Control Inpaint pipeline support and bnb 8-bit LoRA loading support are different code paths and different features." }, { - "left": "pull_request:1951", - "right": "pull_request:3370", + "left": "pull_request:1269", + "right": "pull_request:2009", "accept": false, - "reason": "ColossalAI version update vs. PyTorch 2.0 docs update. Unrelated." + "reason": "Stable diffusion refactor work and textual inversion loading support are unrelated." }, { - "left": "pull_request:2367", - "right": "pull_request:2928", + "left": "pull_request:1508", + "right": "pull_request:3471", "accept": false, - "reason": "Checkpoint retention in training scripts vs. controlnet local dataset loading. Different features." + "reason": "Both are DreamBooth docs changes, but they cover different topics: prior-preservation vs torch.compile guidance." }, { - "left": "pull_request:2892", - "right": "pull_request:3446", + "left": "pull_request:5299", + "right": "pull_request:5759", "accept": false, - "reason": "Upcaster/embed input support vs. a pipeline warning-message wording fix. Not the same bug." + "reason": "These overlap only at a high level (docs/type hints). The concrete files and edits do not indicate the same underlying bug/change." }, { - "left": "pull_request:2892", - "right": "pull_request:3641", + "left": "pull_request:7200", + "right": "pull_request:7919", "accept": false, - "reason": "Both involve inpainting/upscaling, but they fix different latent-handling behavior in different pipelines." + "reason": "Transformers2D norm-type validation and offload behavior with device_map are separate core changes with different failure modes." } ] }, @@ -186899,7 +186876,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 267, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -186907,13 +186884,21 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10192", + "pull_request:10539", + "pull_request:10810", "pull_request:10941", - "pull_request:11655", + "pull_request:10983", + "pull_request:11066", + "pull_request:1136", + "pull_request:11704", "pull_request:11739", "pull_request:11744", + "pull_request:11856", + "pull_request:11997", + "pull_request:12006", "pull_request:12490", - "pull_request:1269", + "pull_request:12523", + "pull_request:12611", "pull_request:12782", "pull_request:13039", "pull_request:13055", @@ -186921,35 +186906,27 @@ "pull_request:13217", "pull_request:13275", "pull_request:1508", - "pull_request:2009", - "pull_request:2892", - "pull_request:2976", + "pull_request:1525", + "pull_request:1982", + "pull_request:2001", + "pull_request:2339", + "pull_request:2693", "pull_request:326", - "pull_request:3471", "pull_request:3546", - "pull_request:3568", "pull_request:359", "pull_request:360", + "pull_request:3607", "pull_request:3632", - "pull_request:4272", - "pull_request:4621", - "pull_request:4856", - "pull_request:5090", - "pull_request:5299", - "pull_request:5391", - "pull_request:5411", - "pull_request:5759", + "pull_request:5407", "pull_request:5898", - "pull_request:6119", "pull_request:6330", - "pull_request:6645", - "pull_request:7200", + "pull_request:6346", + "pull_request:6994", + "pull_request:7068", "pull_request:7246", - "pull_request:7319", - "pull_request:7857", - "pull_request:7919", - "pull_request:822", - "pull_request:827", + "pull_request:774", + "pull_request:8571", + "pull_request:8631", "pull_request:8797", "pull_request:8992", "pull_request:9206", @@ -186957,19 +186934,6 @@ "pull_request:9676" ], "soft_pairs": [ - "pull_request:2892|pull_request:4272", - "pull_request:2976|pull_request:4856", - "pull_request:3568|pull_request:5411", - "pull_request:4621|pull_request:6119", - "pull_request:5090|pull_request:6645", - "pull_request:5391|pull_request:5759", - "pull_request:7319|pull_request:7857", - "pull_request:822|pull_request:827", - "pull_request:10192|pull_request:11655", - "pull_request:1269|pull_request:2009", - "pull_request:1508|pull_request:3471", - "pull_request:5299|pull_request:5759", - "pull_request:7200|pull_request:7919", "pull_request:10941|pull_request:9676", "pull_request:326|pull_request:359", "pull_request:326|pull_request:360", @@ -186983,198 +186947,211 @@ "pull_request:11739|pull_request:11744", "pull_request:12490|pull_request:12782", "pull_request:12490|pull_request:13039", - "pull_request:13039|pull_request:13055" + "pull_request:13039|pull_request:13055", + "pull_request:1508|pull_request:2339", + "pull_request:11997|pull_request:12611", + "pull_request:10539|pull_request:10810", + "pull_request:1136|pull_request:2001", + "pull_request:1136|pull_request:2693", + "pull_request:11704|pull_request:11856", + "pull_request:5407|pull_request:6346", + "pull_request:7068|pull_request:8571", + "pull_request:1982|pull_request:774", + "pull_request:10983|pull_request:11066", + "pull_request:12006|pull_request:12523", + "pull_request:1525|pull_request:3607", + "pull_request:6994|pull_request:8631" ], - "prepared_review_unit_hash": "58e5f7ecd7906ae7c67f535f165d983a87b6fd44106e7730a5d7269b22a7422e", + "prepared_review_unit_hash": "624ece7fc0b1f4e388f48fe4efc3c6828816a17ee1a12cdaf3663ddd3348956f", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 103834, - "estimated_input_tokens": 25959, - "estimated_eval_tokens": 52174 + "serialized_chars": 100657, + "estimated_input_tokens": 25165, + "estimated_eval_tokens": 50586 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: mostly unrelated docs, test, chore, and feature PRs that only look similar at a high level. I found no convincing duplicate pairings and no clear canonical artifact.", + "summary": "The cluster is mostly heterogeneous: the proposed soft edges pair up superficially similar PRs (same subsystem, docs/tests, or related model families) but not the same underlying bug or change. I rejected all candidate duplicate links.", "confidence": 0.96, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No single PR is a good canonical representative because the items do not share one underlying bug or change.", + "best_pr_reason": "PR 13217 is the most substantial and activity-heavy item, so it is the best representative anchor if one must be chosen, but this set is not truly a duplicate cluster.", "soft_edge_verdicts": [ { - "left": "pull_request:2892", - "right": "pull_request:4272", + "left": "pull_request:10941", + "right": "pull_request:9676", "accept": false, - "reason": "Different problems: upscaler input/embed handling vs missing optional imports. Only broad code-maintenance similarity." + "reason": "Flux true-cfg logic fix vs an import cleanup; unrelated changes." }, { - "left": "pull_request:2976", - "right": "pull_request:4856", + "left": "pull_request:326", + "right": "pull_request:359", "accept": false, - "reason": "Both are docs link fixes, but they target different pages and broken links, not the same underlying change." + "reason": "Type-hint cleanup in a pipeline vs scheduler typing; same style, different code paths." }, { - "left": "pull_request:3568", - "right": "pull_request:5411", + "left": "pull_request:326", + "right": "pull_request:360", "accept": false, - "reason": "Both fix broken docs links, but for different guides and different link targets." + "reason": "Different scheduler typing refactors, not the same change." }, { - "left": "pull_request:4621", - "right": "pull_request:6119", + "left": "pull_request:5898", + "right": "pull_request:6330", "accept": false, - "reason": "Unrelated: multi-adapter downscale validation vs LoRA fp16 gradient scaling in training." + "reason": "Both are docs fixes, but on different pages and different issues." }, { - "left": "pull_request:5090", - "right": "pull_request:6645", + "left": "pull_request:7246", + "right": "pull_request:9206", "accept": false, - "reason": "Unrelated t2i adapter logic vs community README cleanup." + "reason": "Different test fixes in different pipelines; only broad test hygiene overlaps." }, { - "left": "pull_request:5391", - "right": "pull_request:5759", + "left": "pull_request:8797", + "right": "pull_request:9206", "accept": false, - "reason": "Both are broad docs/type-hint cleanups, but they are not the same concrete fix or code path." + "reason": "Unrelated model/modeling test adjustments, not one bug." }, { - "left": "pull_request:7319", - "right": "pull_request:7857", + "left": "pull_request:13121", + "right": "pull_request:13217", "accept": false, - "reason": "Different docs additions for different features/schedulers; not a duplicate change." + "reason": "LTX-2 CFG fix vs broader LTX-2.3 model support; related area, but not the same concrete change." }, { - "left": "pull_request:822", - "right": "pull_request:827", + "left": "pull_request:8992", + "right": "pull_request:9423", "accept": false, - "reason": "One updates Flax scheduler API; the other aligns PT/Flax loading behavior. Same area, different change." + "reason": "Different community differential-diffusion pipelines for different models." }, { - "left": "pull_request:10192", - "right": "pull_request:11655", + "left": "pull_request:12490", + "right": "pull_request:13275", "accept": false, - "reason": "Different Flux issues: pipeline inpaint/control support vs 8-bit LoRA loading support." + "reason": "Audioldm2 test config fix vs Hunyuan test kwargs fix; unrelated." }, { - "left": "pull_request:1269", - "right": "pull_request:2009", + "left": "pull_request:3546", + "right": "pull_request:3632", "accept": false, - "reason": "Refactor/finalization PR vs textual inversion loading feature; not the same bug or change." + "reason": "Panorama scheduler compatibility vs view-batch feature/docs; same subsystem, different functionality." }, { - "left": "pull_request:1508", - "right": "pull_request:3471", + "left": "pull_request:11739", + "right": "pull_request:11744", "accept": false, - "reason": "Both are DreamBooth docs updates, but they address different documentation topics." + "reason": "Both touch LoRA metadata/alpha, but on different training scripts and example flows." }, { - "left": "pull_request:5299", - "right": "pull_request:5759", + "left": "pull_request:12490", + "right": "pull_request:12782", "accept": false, - "reason": "Overlap on type hints/docs, but these are separate broad cleanup PRs touching different sets of files." + "reason": "Different test-only fixes for different models." }, { - "left": "pull_request:7200", - "right": "pull_request:7919", + "left": "pull_request:12490", + "right": "pull_request:13039", "accept": false, - "reason": "Different code paths: Transformer2D input validation vs offload behavior when device_map is enabled." + "reason": "Audioldm2 test config vs WAN CPU upcasting test behavior; unrelated." }, { - "left": "pull_request:10941", - "right": "pull_request:9676", + "left": "pull_request:13039", + "right": "pull_request:13055", "accept": false, - "reason": "Different fixes in different areas: Flux copy cleanup vs benchmark import path." + "reason": "One fixes layerwise upcasting on CPU; the other fixes WAN pipeline fallback when a transformer is missing." }, { - "left": "pull_request:326", - "right": "pull_request:359", + "left": "pull_request:1508", + "right": "pull_request:2339", "accept": false, - "reason": "Both are type-hint PRs, but they modify different scheduler files and are not the same change." + "reason": "Dreambooth docs clarification vs checkpointing docs/tests; related topic, not same issue." }, { - "left": "pull_request:326", - "right": "pull_request:360", + "left": "pull_request:11997", + "right": "pull_request:12611", "accept": false, - "reason": "Same as above: separate scheduler type-hint cleanup, not one underlying fix." + "reason": "Both are Wan-related, but one is a modular pipeline feature PR and the other is a broad guider/modular update; not the same concrete change." }, { - "left": "pull_request:5898", - "right": "pull_request:6330", + "left": "pull_request:10539", + "right": "pull_request:10810", "accept": false, - "reason": "Different docs edits for unrelated topics: unconditional generation vs SVD rendering." + "reason": "True-cfg fix in Flux vs LoRA support conversion changes; unrelated." }, { - "left": "pull_request:7246", - "right": "pull_request:9206", + "left": "pull_request:1136", + "right": "pull_request:2001", "accept": false, - "reason": "Unrelated test fixes: VAE tiling device handling vs xformers test coverage." + "reason": "404 link fix in examples README vs ColossalAI typo/seed fix; unrelated." }, { - "left": "pull_request:8797", - "right": "pull_request:9206", + "left": "pull_request:1136", + "right": "pull_request:2693", "accept": false, - "reason": "Different test failures in different subsystems; not the same bug." + "reason": "Different documentation pages and different broken links." }, { - "left": "pull_request:13121", - "right": "pull_request:13217", + "left": "pull_request:11704", + "right": "pull_request:11856", "accept": false, - "reason": "LTX-2 inference edge case vs new LTX-2.3 model support; distinct code paths and goals." + "reason": "Both touch Wan LoRA conversion, but they fix different conversion paths and failure modes." }, { - "left": "pull_request:8992", - "right": "pull_request:9423", + "left": "pull_request:5407", + "right": "pull_request:6346", "accept": false, - "reason": "Both add differential diffusion, but to different model families and implementations." + "reason": "Mixed-precision training example fix vs LoRA state-dict conversion follow-up; not the same bug." }, { - "left": "pull_request:12490", - "right": "pull_request:13275", + "left": "pull_request:7068", + "right": "pull_request:8571", "accept": false, - "reason": "Separate test fixes for different models/configurations; no shared underlying bug." + "reason": "Generic typo/style cleanup PRs; too broad and unrelated." }, { - "left": "pull_request:3546", - "right": "pull_request:3632", + "left": "pull_request:1982", + "right": "pull_request:774", "accept": false, - "reason": "Related panorama pipeline area, but the changes are different: scheduler support vs view-batch feature." + "reason": "Example test path tweak vs pipeline callback/test tolerance changes; unrelated." }, { - "left": "pull_request:11739", - "right": "pull_request:11744", + "left": "pull_request:10983", + "right": "pull_request:11066", "accept": false, - "reason": "One is an open training-script alpha/metadata fix; the other is a completed Sana training metadata PR. Similar theme, but not the same artifact." + "reason": "Both are HunyuanVideo I2V changes, but they target different model variants/releases and different code paths." }, { - "left": "pull_request:12490", - "right": "pull_request:12782", + "left": "pull_request:12006", + "right": "pull_request:12523", "accept": false, - "reason": "Different test adjustments for different model families." + "reason": "Wan model addition vs AutoencoderKLWan skip-key fix; different problems." }, { - "left": "pull_request:12490", - "right": "pull_request:13039", + "left": "pull_request:1525", + "right": "pull_request:3607", "accept": false, - "reason": "Unrelated: Audioldm2 test config vs Wan upcasting fix." + "reason": "Upscaling batch-size fix vs temb attention fix; unrelated." }, { - "left": "pull_request:13039", - "right": "pull_request:13055", + "left": "pull_request:6994", + "right": "pull_request:8631", "accept": false, - "reason": "Different Wan issues: CPU upcasting test behavior vs missing-transformer fallback handling." + "reason": "Two separate from_single_file loader fixes touching different conversion utilities." } ] }, @@ -187190,7 +187167,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 268, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -187199,68 +187176,54 @@ "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10105", - "pull_request:10539", + "pull_request:10200", + "pull_request:10579", "pull_request:10704", - "pull_request:10810", - "pull_request:10983", - "pull_request:11066", "pull_request:1128", - "pull_request:1136", "pull_request:11366", "pull_request:11384", "pull_request:11571", "pull_request:11572", "pull_request:11635", - "pull_request:11704", - "pull_request:11856", + "pull_request:11726", "pull_request:11897", - "pull_request:11997", - "pull_request:12006", + "pull_request:12315", + "pull_request:12324", "pull_request:12496", + "pull_request:12522", "pull_request:12523", - "pull_request:12611", "pull_request:12982", "pull_request:13002", + "pull_request:13150", "pull_request:13208", "pull_request:13223", "pull_request:1338", - "pull_request:1508", - "pull_request:1525", + "pull_request:13454", "pull_request:1926", - "pull_request:1982", - "pull_request:2001", "pull_request:2322", - "pull_request:2339", "pull_request:2691", "pull_request:2693", - "pull_request:3607", + "pull_request:2968", + "pull_request:3298", "pull_request:391", "pull_request:416", + "pull_request:4242", "pull_request:4294", "pull_request:433", - "pull_request:5407", - "pull_request:6346", + "pull_request:4412", + "pull_request:5098", + "pull_request:6489", + "pull_request:6490", "pull_request:689", - "pull_request:6994", - "pull_request:7068", - "pull_request:774", - "pull_request:8571", - "pull_request:8631" + "pull_request:6962", + "pull_request:7247", + "pull_request:7941", + "pull_request:8322", + "pull_request:8761", + "pull_request:9008", + "pull_request:9065" ], "soft_pairs": [ - "pull_request:1508|pull_request:2339", - "pull_request:11997|pull_request:12611", - "pull_request:10539|pull_request:10810", - "pull_request:1136|pull_request:2001", - "pull_request:1136|pull_request:2693", - "pull_request:11704|pull_request:11856", - "pull_request:5407|pull_request:6346", - "pull_request:7068|pull_request:8571", - "pull_request:1982|pull_request:774", - "pull_request:10983|pull_request:11066", - "pull_request:12006|pull_request:12523", - "pull_request:1525|pull_request:3607", - "pull_request:6994|pull_request:8631", "pull_request:13208|pull_request:13223", "pull_request:11635|pull_request:11897", "pull_request:2322|pull_request:433", @@ -187272,186 +187235,186 @@ "pull_request:10105|pull_request:10704", "pull_request:1338|pull_request:689", "pull_request:1128|pull_request:1926", - "pull_request:11366|pull_request:11384" + "pull_request:11366|pull_request:11384", + "pull_request:10579|pull_request:9065", + "pull_request:2693|pull_request:4412", + "pull_request:2968|pull_request:4412", + "pull_request:5098|pull_request:6962", + "pull_request:12522|pull_request:13454", + "pull_request:11726|pull_request:12324", + "pull_request:12315|pull_request:13150", + "pull_request:3298|pull_request:4242", + "pull_request:6489|pull_request:6490", + "pull_request:8322|pull_request:9008", + "pull_request:7247|pull_request:7941", + "pull_request:10200|pull_request:8761" ], - "prepared_review_unit_hash": "cbfaf21418d4bd4879dcccc9ecbafdc9ca1cf7445eb89238cad2c00f48b1173f", + "prepared_review_unit_hash": "8865b013118fcaab7d0b9f3ad3f0fef2e0121c5173dfe86abb2c14aaaa7b1719", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 25, - "serialized_chars": 106946, - "estimated_input_tokens": 26737, - "estimated_eval_tokens": 53730 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 105037, + "estimated_input_tokens": 26260, + "estimated_eval_tokens": 52776 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated PRs with a few false-positive soft links from shared filenames or broad subsystem overlap; I found no true duplicate pairs under a strict same-bug/same-change standard.", - "confidence": 0.92, + "summary": "This is a heterogeneous cluster of unrelated pull requests across different features, docs, tests, and refactors. The soft edges are mostly superficial filename/subsystem matches rather than the same underlying change.", + "confidence": 0.99, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 13208 is the best representative if one must be chosen: it is a coherent, self-contained feature addition with docs, pipeline/model code, loaders, and test integration. The rest are mostly smaller docs/typo edits or narrower follow-up fixes.", + "canonical_pr_reason": "No single PR is a safe canonical representative: the cluster spans unrelated PRs from different subsystems and release trains, with no shared concrete bug/change theme.", "best_issue_reason": null, - "best_pr_reason": "PR 13208 is the strongest overall artifact because it introduces a full Helios video-generation feature rather than a narrow patch, and it has the broadest implementation surface and integration coverage.", + "best_pr_reason": "No PR is globally best-suited for duplicate collapsing here; the items do not form one coherent fix or feature.", "soft_edge_verdicts": [ { - "left": "pull_request:1508", - "right": "pull_request:2339", - "accept": false, - "reason": "Both touch DreamBooth docs/train script, but one is about `num_class_images` clarification and the other is checkpointing docs/tests; different user problems and different changes." - }, - { - "left": "pull_request:11997", - "right": "pull_request:12611", - "accept": false, - "reason": "Both are Wan modular-pipeline work, but 11997 adds Wan I2V/FLF2V blocks while 12611 is a broader modular refactor with guider and pipeline plumbing; not the same concrete change." - }, - { - "left": "pull_request:10539", - "right": "pull_request:10810", + "left": "pull_request:13208", + "right": "pull_request:13223", "accept": false, - "reason": "One fixes Flux true-CFG gating; the other improves LoRA conversion support for Flux. Same subsystem, different bug." + "reason": "Same Helios area, but one adds the pipelines and docs while the other changes context-parallel behavior in the transformer; different changes, not one duplicate fix." }, { - "left": "pull_request:1136", - "right": "pull_request:2001", + "left": "pull_request:11635", + "right": "pull_request:11897", "accept": false, - "reason": "Both are small docs/example edits, but one fixes README links and the other fixes a ColossalAI example typo; unrelated." + "reason": "Both touch docs/pipelines, but 11635 adds Flux Kontext while 11897 is a broad modular-diffusers/test merge; unrelated work." }, { - "left": "pull_request:1136", - "right": "pull_request:2693", + "left": "pull_request:2322", + "right": "pull_request:433", "accept": false, - "reason": "Both are link/doc cleanups, but they target different files and different broken links; not a duplicate change." + "reason": "Different eras and completely different concerns: inpaint example/update vs black/versioning cleanup." }, { - "left": "pull_request:11704", - "right": "pull_request:11856", + "left": "pull_request:2691", + "right": "pull_request:4294", "accept": false, - "reason": "Both concern Wan LoRA handling, but 11704 adjusts conversion thresholds for AccVideo support while 11856 changes fuse_lora logic for AccVideo/CausVid; related area, not the same fix." + "reason": "One renames attention code; the other updates SDXL checkpoint/docs naming. No shared code-path bug." }, { - "left": "pull_request:5407", - "right": "pull_request:6346", + "left": "pull_request:391", + "right": "pull_request:416", "accept": false, - "reason": "One fixes mixed-precision behavior in an unconditional training example; the other updates LoRA state-dict conversion in training examples. Different code paths and symptoms." + "reason": "PNDM docs versus core model docs restructuring; unrelated documentation changes." }, { - "left": "pull_request:7068", - "right": "pull_request:8571", + "left": "pull_request:12982", + "right": "pull_request:13002", "accept": false, - "reason": "Both are style/typo cleanup PRs, but they touch different files and different text; no shared bug or change." + "reason": "Both are about FLUX.2 klein, but one is training/docs/pipeline support and the other is modular-pipeline implementation; not the same concrete change." }, { - "left": "pull_request:1982", - "right": "pull_request:774", + "left": "pull_request:11571", + "right": "pull_request:11572", "accept": false, - "reason": "Both are test-related, but one fixes an example test path and the other adjusts pipeline test tolerances/usage; not the same underlying issue." + "reason": "Same LTX-Video theme, but one adds distilled model/docs and the other adds LoRA conversion support; different code paths." }, { - "left": "pull_request:10983", - "right": "pull_request:11066", + "left": "pull_request:10105", + "right": "pull_request:10704", "accept": false, - "reason": "Both are Hunyuan I2V docs/pipeline updates, but they introduce different model variants and distinct implementation changes." + "reason": "Broken README link fix versus img2img docs cleanup; unrelated doc edits." }, { - "left": "pull_request:12006", - "right": "pull_request:12523", + "left": "pull_request:1338", + "right": "pull_request:689", "accept": false, - "reason": "Wan i2v shape mismatch vs AutoencoderKLWan skip-key handling are separate bugs in different components." + "reason": "ONNX inpaint cast-order fix is unrelated to an attention performance tweak." }, { - "left": "pull_request:1525", - "right": "pull_request:3607", + "left": "pull_request:1128", + "right": "pull_request:1926", "accept": false, - "reason": "One fixes upscale batch sizing; the other fixes attention behavior and slow tests. Different problems." + "reason": "Both mention inpainting, but one adds tensor input support and the other is OneFlow dev support; different implementations." }, { - "left": "pull_request:6994", - "right": "pull_request:8631", + "left": "pull_request:11366", + "right": "pull_request:11384", "accept": false, - "reason": "Both affect single-file loaders, but one passes `torch_dtype` through loaders while the other returns a missing T5 model object; separate loader bugs." + "reason": "Same Hidream file, but one changes model outputs and the other only shifts deprecation versions; not the same bug/change." }, { - "left": "pull_request:13208", - "right": "pull_request:13223", + "left": "pull_request:10579", + "right": "pull_request:9065", "accept": false, - "reason": "13208 adds Helios pipelines and docs; 13223 changes Helios context-parallel execution. Same model family, but not the same change." + "reason": "Community pipeline addition versus sentencepiece optional dependency plumbing; unrelated." }, { - "left": "pull_request:11635", - "right": "pull_request:11897", + "left": "pull_request:2693", + "right": "pull_request:4412", "accept": false, - "reason": "11635 adds Flux Kontext functionality; 11897 is a workflow/docs merge. No overlap in concrete bug or feature." + "reason": "Doc link fix versus test fixture cleanup; no shared underlying issue." }, { - "left": "pull_request:2322", - "right": "pull_request:433", + "left": "pull_request:2968", + "right": "pull_request:4412", "accept": false, - "reason": "One is an inpaint example/update bundle; the other is a black version bump and formatting cleanup. Unrelated." + "reason": "Post-processing bug fix versus a test setup change; unrelated." }, { - "left": "pull_request:2691", - "right": "pull_request:4294", + "left": "pull_request:5098", + "right": "pull_request:6962", "accept": false, - "reason": "Both rename/adjust docs and APIs, but they target different model families and different rename tasks." + "reason": "Different pipelines and different goals: Kandinsky precision tests versus PixArt slow-test stabilization." }, { - "left": "pull_request:391", - "right": "pull_request:416", + "left": "pull_request:12522", + "right": "pull_request:13454", "accept": false, - "reason": "PNDM docs and model API docs are separate documentation updates with no shared underlying fix." + "reason": "Toctree styling cleanup versus xfail markers for CLIP/transformers compatibility; unrelated." }, { - "left": "pull_request:12982", - "right": "pull_request:13002", + "left": "pull_request:11726", + "right": "pull_request:12324", "accept": false, - "reason": "Both are Flux2/Klein support, but 12982 adds training/pipeline support and 13002 adds modular-pipeline support; related feature area, not one duplicate PR." + "reason": "Both involve Wan, but one extends LoRA conversion support and the other adds a new VACE-Fun model/config path; distinct changes." }, { - "left": "pull_request:11571", - "right": "pull_request:11572", + "left": "pull_request:12315", + "right": "pull_request:13150", "accept": false, - "reason": "One adds LTX 0.9.7-distilled model/pipeline docs; the other adds non-diffusers LTX LoRA conversion support. Same product area, different changes." + "reason": "One normalizes scheduler usage in LoRA tests; the other changes a test helper API. Not the same bug or feature." }, { - "left": "pull_request:12496", - "right": "pull_request:12523", + "left": "pull_request:3298", + "right": "pull_request:4242", "accept": false, - "reason": "Wan I2V condition shape fix vs AutoencoderKLWan skip-key handling; unrelated fixes." + "reason": "Image aspect-ratio support in IF super-resolution is unrelated to allowing VAE loading in from_single_file." }, { - "left": "pull_request:10105", - "right": "pull_request:10704", + "left": "pull_request:6489", + "right": "pull_request:6490", "accept": false, - "reason": "Broken README link cleanup and image-to-image docs correction are different documentation edits." + "reason": "Two different community pipelines; no shared code-path or duplicate fix." }, { - "left": "pull_request:1338", - "right": "pull_request:689", + "left": "pull_request:8322", + "right": "pull_request:9008", "accept": false, - "reason": "Onnx inpainting cast-order fix and attention performance change are unrelated." + "reason": "Generic errata/typo cleanup versus broader typo and formatting fixes; separate maintenance work." }, { - "left": "pull_request:1128", - "right": "pull_request:1926", + "left": "pull_request:7247", + "right": "pull_request:7941", "accept": false, - "reason": "Tensor-input inpaint support and a separate Stable Diffusion OneFlow support PR are different features." + "reason": "Config-checking test fix versus adding a slow-test decorator; unrelated test maintenance." }, { - "left": "pull_request:11366", - "right": "pull_request:11384", + "left": "pull_request:10200", + "right": "pull_request:8761", "accept": false, - "reason": "Both touch HiDream code, but one changes transformer outputs and the other moves deprecation warnings; different concerns." + "reason": "ControlNet Union refactor and Hunyuan embedding refactor are separate model changes with no common bug." } ] }, @@ -187467,7 +187430,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 269, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -187475,68 +187438,56 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10200", - "pull_request:10579", - "pull_request:11726", + "pull_request:10330", + "pull_request:10431", + "pull_request:10537", + "pull_request:10815", + "pull_request:11252", + "pull_request:11556", "pull_request:11916", + "pull_request:11986", "pull_request:12006", - "pull_request:12315", - "pull_request:12324", + "pull_request:12092", + "pull_request:12387", "pull_request:12496", - "pull_request:12522", "pull_request:12530", "pull_request:12549", - "pull_request:13150", + "pull_request:12647", + "pull_request:12763", + "pull_request:12961", + "pull_request:13039", + "pull_request:13331", "pull_request:13454", "pull_request:1926", "pull_request:1966", + "pull_request:1982", + "pull_request:2367", "pull_request:2464", "pull_request:2470", - "pull_request:2693", + "pull_request:2593", "pull_request:2892", "pull_request:2959", - "pull_request:2968", "pull_request:327", - "pull_request:3298", "pull_request:344", "pull_request:3586", + "pull_request:3657", "pull_request:382", "pull_request:3912", "pull_request:396", "pull_request:4098", "pull_request:416", - "pull_request:4242", "pull_request:4408", "pull_request:4412", "pull_request:4780", - "pull_request:5098", "pull_request:5151", "pull_request:5384", - "pull_request:6489", - "pull_request:6490", + "pull_request:6633", "pull_request:6961", - "pull_request:6962", - "pull_request:7247", - "pull_request:7941", - "pull_request:8322", + "pull_request:802", "pull_request:8452", - "pull_request:8761", - "pull_request:9008", - "pull_request:9065" + "pull_request:9696" ], "soft_pairs": [ - "pull_request:10579|pull_request:9065", - "pull_request:2693|pull_request:4412", - "pull_request:2968|pull_request:4412", - "pull_request:5098|pull_request:6962", - "pull_request:12522|pull_request:13454", - "pull_request:11726|pull_request:12324", - "pull_request:12315|pull_request:13150", - "pull_request:3298|pull_request:4242", - "pull_request:6489|pull_request:6490", - "pull_request:8322|pull_request:9008", - "pull_request:7247|pull_request:7941", - "pull_request:10200|pull_request:8761", "pull_request:327|pull_request:344", "pull_request:3586|pull_request:5384", "pull_request:2470|pull_request:2959", @@ -187550,192 +187501,204 @@ "pull_request:2892|pull_request:4412", "pull_request:4780|pull_request:5151", "pull_request:2464|pull_request:3912", - "pull_request:12006|pull_request:12496" + "pull_request:12006|pull_request:12496", + "pull_request:10330|pull_request:10815", + "pull_request:10431|pull_request:10537", + "pull_request:11556|pull_request:12092", + "pull_request:11986|pull_request:12763", + "pull_request:12647|pull_request:13039", + "pull_request:12961|pull_request:13331", + "pull_request:1982|pull_request:802", + "pull_request:2367|pull_request:3657", + "pull_request:2593|pull_request:3657", + "pull_request:6633|pull_request:6961", + "pull_request:11252|pull_request:9696", + "pull_request:11916|pull_request:12387" ], - "prepared_review_unit_hash": "9240c5295d807147d3c7d65dd3ac1a1a5c987af8f9cb6c551e243258168acf15", + "prepared_review_unit_hash": "3d5b69faf66793c7a1dfe4d5f9580a25f8266af242ef97202d33163aec38588f", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 105343, - "estimated_input_tokens": 26336, - "estimated_eval_tokens": 52928 + "serialized_chars": 102040, + "estimated_input_tokens": 25510, + "estimated_eval_tokens": 51276 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: it mixes unrelated docs, tests, model refactors, and feature PRs. None of the soft edges look like true duplicates of the same concrete change, so no canonical artifact stands out.", - "confidence": 0.94, + "summary": "The cluster is mostly heterogeneous: many unrelated docs, test, and feature PRs. The only strong duplicate-like pair is the PEFT integration work (4780 and 5151).", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: the apparent similarities are only broad subsystem/topic overlap (Wan/VACE, PEFT, attention backends, docs/tests) rather than the same underlying fix or feature.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No global best PR for deduplication: the set is too mixed, and the strongest overlaps still target different code paths or distinct feature additions.", + "best_pr_reason": "PR 5151 is the best representative because it is the clearest mature continuation of the only truly overlapping feature thread in the set (PEFT integration with 4780).", "soft_edge_verdicts": [ { - "left": "pull_request:10579", - "right": "pull_request:9065", + "left": "pull_request:327", + "right": "pull_request:344", "accept": false, - "reason": "Community pipeline addition vs soft-dependency packaging change; unrelated changes." + "reason": "Different type-hint cleanup PRs for unrelated codepaths (PNDM vs VAE)." }, { - "left": "pull_request:2693", - "right": "pull_request:4412", + "left": "pull_request:3586", + "right": "pull_request:5384", "accept": false, - "reason": "Doc image-link fix vs float16 test harness fix; different artifacts and code paths." + "reason": "Both are broad feature PRs, but they target different subsystems and problems (tiling pipeline vs hypernetwork support)." }, { - "left": "pull_request:2968", - "right": "pull_request:4412", + "left": "pull_request:2470", + "right": "pull_request:2959", "accept": false, - "reason": "Img2img post-processing bugfix vs test-only adjustment; not the same change." + "reason": "Both are test fixes, but for different models and failure modes." }, { - "left": "pull_request:5098", - "right": "pull_request:6962", + "left": "pull_request:382", + "right": "pull_request:396", "accept": false, - "reason": "Different pipelines/models and different test issues; only both are test-related." + "reason": "Docs for different pipelines (DDPM vs DDIM), not the same change." }, { - "left": "pull_request:12522", - "right": "pull_request:13454", + "left": "pull_request:382", + "right": "pull_request:416", "accept": false, - "reason": "Docs TOC styling cleanup vs xfail for CLIP-related tests; unrelated." + "reason": "DDPM/DDIM docs versus general model docs; unrelated documentation updates." }, { - "left": "pull_request:11726", - "right": "pull_request:12324", + "left": "pull_request:1926", + "right": "pull_request:1966", "accept": false, - "reason": "Wan VACE LoRA conversion support vs new Wan2.2 VACE feature/pipeline; related theme, but different concrete work." + "reason": "Both involve inpainting, but they are separate framework/backend implementations and not one concrete fix." }, { - "left": "pull_request:12315", - "right": "pull_request:13150", + "left": "pull_request:6961", + "right": "pull_request:8452", "accept": false, - "reason": "Scheduler test simplification vs compile-test API change; separate test maintenance." + "reason": "Both touch single-file loading, but one adds tests and the other changes loader behavior for a different bug." }, { - "left": "pull_request:3298", - "right": "pull_request:4242", + "left": "pull_request:4098", + "right": "pull_request:4408", "accept": false, - "reason": "Pipeline aspect-ratio support vs from_single_file VAE loading support; not the same bug/change." + "reason": "One adds SDXL inpainting; the other adds from_single_file support for that feature. Related area, but distinct PRs." }, { - "left": "pull_request:6489", - "right": "pull_request:6490", + "left": "pull_request:11916", + "right": "pull_request:12549", "accept": false, - "reason": "Different community pipelines, only sharing README placement." + "reason": "Attention-dispatch refactor versus adding a new AITER backend; same subsystem, different concrete changes." }, { - "left": "pull_request:8322", - "right": "pull_request:9008", + "left": "pull_request:12530", + "right": "pull_request:13454", "accept": false, - "reason": "Broad errata sweep vs typo/whitespace cleanup; not one underlying issue." + "reason": "Separate xfail annotations for unrelated failing tests." }, { - "left": "pull_request:7247", - "right": "pull_request:7941", + "left": "pull_request:2892", + "right": "pull_request:4412", "accept": false, - "reason": "Single-file config test fixes vs marking a different slow test; unrelated test changes." + "reason": "Unrelated changes: upscaler input validation versus a float16 test fix." }, { - "left": "pull_request:10200", - "right": "pull_request:8761", - "accept": false, - "reason": "Different model refactors in unrelated components." + "left": "pull_request:4780", + "right": "pull_request:5151", + "accept": true, + "reason": "Both are PEFT integration work for UNet; 5151 looks like the completed follow-through of the same feature thread." }, { - "left": "pull_request:327", - "right": "pull_request:344", + "left": "pull_request:2464", + "right": "pull_request:3912", "accept": false, - "reason": "Separate type-hint PRs for different modules; no shared bug or feature." + "reason": "Same training script file, but different changes: xformers cleanup versus adding text-encoder finetuning support." }, { - "left": "pull_request:3586", - "right": "pull_request:5384", + "left": "pull_request:12006", + "right": "pull_request:12496", "accept": false, - "reason": "Community tiling pipeline vs hypernetwork loader support; completely different changes." + "reason": "Different Wan i2v changes: feature expansion for 5B support versus a shape-mismatch bug fix." }, { - "left": "pull_request:2470", - "right": "pull_request:2959", + "left": "pull_request:10330", + "right": "pull_request:10815", "accept": false, - "reason": "Different test fixes for different models/pipelines." + "reason": "LTX Video release/docs work versus Enhance-A-Video hook implementation; not the same bug or change." }, { - "left": "pull_request:382", - "right": "pull_request:396", + "left": "pull_request:10431", + "right": "pull_request:10537", "accept": false, - "reason": "DDPM docs vs DDIM docs; adjacent topics but distinct additions." + "reason": "Both are Sana-related, but one fixes positional embeddings and the other adds 4K pipeline adjustments." }, { - "left": "pull_request:382", - "right": "pull_request:416", + "left": "pull_request:11556", + "right": "pull_request:12092", "accept": false, - "reason": "Pipeline docs vs model docs; no shared underlying change." + "reason": "Different docs topics: regional compilation versus GGUF checkpoint conversion." }, { - "left": "pull_request:1926", - "right": "pull_request:1966", + "left": "pull_request:11986", + "right": "pull_request:12763", "accept": false, - "reason": "Both add inpainting support, but for different backends/implementations; not mergeable as one change." + "reason": "Different modular pipelines for different model families (LTX vs Flux2)." }, { - "left": "pull_request:6961", - "right": "pull_request:8452", + "left": "pull_request:12647", + "right": "pull_request:13039", "accept": false, - "reason": "Test additions for single-file loaders vs loader error-handling fix; different scopes." + "reason": "Both concern layerwise-casting tests, but they are model-specific and fix different failures." }, { - "left": "pull_request:4098", - "right": "pull_request:4408", + "left": "pull_request:12961", + "right": "pull_request:13331", "accept": false, - "reason": "Adding SDXL inpainting vs enabling single-file loading for SDXL inpainting; related feature area, but different code paths." + "reason": "Different pipeline additions altogether (PhotoDoodle vs CogVideoX inpaint)." }, { - "left": "pull_request:11916", - "right": "pull_request:12549", + "left": "pull_request:1982", + "right": "pull_request:802", "accept": false, - "reason": "Attention-dispatch refactor for Flux/Chroma vs adding the AITER backend; different changes despite shared files." + "reason": "Example test path fix versus an indentation fix in a custom pipeline example." }, { - "left": "pull_request:12530", - "right": "pull_request:13454", + "left": "pull_request:2367", + "right": "pull_request:3657", "accept": false, - "reason": "Different xfail markers for different failing tests; no same defect." + "reason": "Both touch training/conversion scripts, but they address unrelated features and scripts." }, { - "left": "pull_request:2892", - "right": "pull_request:4412", + "left": "pull_request:2593", + "right": "pull_request:3657", "accept": false, - "reason": "Upscaler input-validation change vs float16 test setup fix; unrelated." + "reason": "Same ControlNet converter script, but different supported checkpoint behaviors; not the same concrete change." }, { - "left": "pull_request:4780", - "right": "pull_request:5151", + "left": "pull_request:6633", + "right": "pull_request:6961", "accept": false, - "reason": "Broad draft PEFT integration work vs a later Unet-specific PEFT integration PR; overlapping area, but not the same concrete change." + "reason": "Different single-file loader fixes: YAML config reading versus test scaffolding changes." }, { - "left": "pull_request:2464", - "right": "pull_request:3912", + "left": "pull_request:11252", + "right": "pull_request:9696", "accept": false, - "reason": "Generic script cleanup vs a feature to finetune the text encoder; different intents and diffs." + "reason": "Unrelated utility/workflow updates: timeout constant cleanup versus PR test workflow input changes." }, { - "left": "pull_request:12006", - "right": "pull_request:12496", + "left": "pull_request:11916", + "right": "pull_request:12387", "accept": false, - "reason": "Wan2.2 i2v feature addition vs a separate shape-mismatch bugfix in the Wan i2v pipeline." + "reason": "Attention dispatcher refactor versus kernels-backed flash-attention backend support; distinct features." } ] }, @@ -187751,7 +187714,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 270, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -187760,27 +187723,23 @@ "cluster_id": "cluster-1-3019", "nodes": [ "pull_request:10330", - "pull_request:10431", "pull_request:10537", - "pull_request:10815", - "pull_request:11252", - "pull_request:11556", - "pull_request:11916", - "pull_request:11986", + "pull_request:10538", + "pull_request:10547", + "pull_request:11174", "pull_request:11995", - "pull_request:12092", - "pull_request:12387", "pull_request:12454", "pull_request:12593", - "pull_request:12647", + "pull_request:12644", "pull_request:12679", - "pull_request:12763", - "pull_request:12961", - "pull_request:13039", - "pull_request:13331", + "pull_request:12934", + "pull_request:12983", + "pull_request:13158", + "pull_request:13208", + "pull_request:13216", + "pull_request:13223", "pull_request:1788", "pull_request:1913", - "pull_request:1982", "pull_request:2327", "pull_request:2367", "pull_request:2593", @@ -187789,37 +187748,30 @@ "pull_request:2959", "pull_request:3292", "pull_request:3607", - "pull_request:3657", "pull_request:3806", + "pull_request:382", + "pull_request:386", + "pull_request:390", + "pull_request:391", + "pull_request:392", + "pull_request:404", + "pull_request:417", "pull_request:4187", + "pull_request:5506", "pull_request:5665", "pull_request:6426", - "pull_request:6633", "pull_request:6759", - "pull_request:6961", "pull_request:6994", "pull_request:7116", + "pull_request:7200", "pull_request:7246", "pull_request:7825", "pull_request:7941", - "pull_request:802", "pull_request:8937", - "pull_request:9676", - "pull_request:9696" + "pull_request:9251", + "pull_request:9676" ], "soft_pairs": [ - "pull_request:10330|pull_request:10815", - "pull_request:10431|pull_request:10537", - "pull_request:11556|pull_request:12092", - "pull_request:11986|pull_request:12763", - "pull_request:12647|pull_request:13039", - "pull_request:12961|pull_request:13331", - "pull_request:1982|pull_request:802", - "pull_request:2367|pull_request:3657", - "pull_request:2593|pull_request:3657", - "pull_request:6633|pull_request:6961", - "pull_request:11252|pull_request:9696", - "pull_request:11916|pull_request:12387", "pull_request:6426|pull_request:7116", "pull_request:1788|pull_request:3607", "pull_request:2367|pull_request:4187", @@ -187832,491 +187784,489 @@ "pull_request:5665|pull_request:6994", "pull_request:2919|pull_request:2959", "pull_request:7246|pull_request:7941", - "pull_request:11995|pull_request:12454" + "pull_request:11995|pull_request:12454", + "pull_request:12934|pull_request:12983", + "pull_request:13208|pull_request:13216", + "pull_request:10330|pull_request:11174", + "pull_request:12644|pull_request:13158", + "pull_request:382|pull_request:391", + "pull_request:386|pull_request:392", + "pull_request:386|pull_request:404", + "pull_request:386|pull_request:417", + "pull_request:390|pull_request:392", + "pull_request:390|pull_request:404", + "pull_request:390|pull_request:417", + "pull_request:5506|pull_request:7200", + "pull_request:13216|pull_request:13223", + "pull_request:10537|pull_request:10538", + "pull_request:10547|pull_request:9251" ], - "prepared_review_unit_hash": "d7f93deca492c4953298f64e9346c3c69d4d265bf6aaf0f607b8c3cc00153af2", + "prepared_review_unit_hash": "402ba18f5bc06a7cbb0b725be80c6ea23f56a5e559c3acac6bb813e862a0ad5b", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 25, - "serialized_chars": 98298, - "estimated_input_tokens": 24575, - "estimated_eval_tokens": 49406 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 103629, + "estimated_input_tokens": 25908, + "estimated_eval_tokens": 52072 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated PRs across different models, pipelines, docs, workflows, and tests. None of the soft-edge pairs looks like the same concrete bug or change under a strict duplicate standard.", - "confidence": 0.97, + "summary": "This cluster is a set of mostly unrelated PRs that only look similar at a very broad level (docs, tests, model/pipeline additions). None of the soft-pair candidates appear to fix the same concrete bug or implement the same change, so they should not be merged as duplicates.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a good canonical for the cluster: the items span unrelated subsystems and model families, so collapsing them would be misleading.", + "canonical_pr_reason": "No single canonical PR stands out as a duplicate target; the items are not a coherent duplicate cluster.", "best_issue_reason": null, - "best_pr_reason": "No PR stands out as a global representative duplicate target; the overlap is mostly superficial (shared docs/files or similar model areas) rather than the same fix.", + "best_pr_reason": "PR 13216 is the strongest single representative if one must be chosen: it is a broad, self-contained Helios modular pipeline feature with supporting code and tests.", "soft_edge_verdicts": [ - { - "left": "pull_request:10330", - "right": "pull_request:10815", - "accept": false, - "reason": "Different video features for different projects; same broad domain but not the same bug or change." - }, - { - "left": "pull_request:10431", - "right": "pull_request:10537", - "accept": false, - "reason": "Both are Sana-related, but one fixes positional encoding while the other changes 4K pipeline behavior; not the same underlying change." - }, - { - "left": "pull_request:11556", - "right": "pull_request:12092", - "accept": false, - "reason": "Both touch docs only, but they document unrelated topics and features." - }, - { - "left": "pull_request:11986", - "right": "pull_request:12763", - "accept": false, - "reason": "Different modular pipelines for different model families; not a duplicate change." - }, - { - "left": "pull_request:12647", - "right": "pull_request:13039", - "accept": false, - "reason": "Separate test/model fixes for different architectures and CPU issues." - }, - { - "left": "pull_request:12961", - "right": "pull_request:13331", - "accept": false, - "reason": "Unrelated additions: a community PhotoDoodle README entry versus a CogVideoX inpaint pipeline." - }, - { - "left": "pull_request:1982", - "right": "pull_request:802", - "accept": false, - "reason": "One updates example tests, the other fixes code indentation in a tutorial example; not the same change." - }, - { - "left": "pull_request:2367", - "right": "pull_request:3657", - "accept": false, - "reason": "Training-script checkpoint limit support is unrelated to controlnet conversion-script argument parsing." - }, - { - "left": "pull_request:2593", - "right": "pull_request:3657", - "accept": false, - "reason": "Same script area, but one adds the controlnet conversion script and the other tweaks parsing options; related, not duplicate." - }, - { - "left": "pull_request:6633", - "right": "pull_request:6961", - "accept": false, - "reason": "Single-file YAML loading fix versus fast tests for loaders; different objectives and code paths." - }, - { - "left": "pull_request:11252", - "right": "pull_request:9696", - "accept": false, - "reason": "Unrelated workflow changes: timeout constant cleanup versus PR test workflow inputs/validation." - }, - { - "left": "pull_request:11916", - "right": "pull_request:12387", - "accept": false, - "reason": "A broad attention refactor versus adding a flash-attention backend through kernels; same subsystem, different change." - }, { "left": "pull_request:6426", "right": "pull_request:7116", "accept": false, - "reason": "LoRA deprecation cleanup and CI workflow packaging changes are unrelated." + "reason": "LoRA deprecation cleanup vs workflow install changes; unrelated changes and not the same code-path problem." }, { "left": "pull_request:1788", "right": "pull_request:3607", "accept": false, - "reason": "Different bug fixes in unrelated code paths: DDIM scheduler logic versus attention temb handling." + "reason": "DDIM scheduler bug fix vs attention/PEFT/xformers fix; different components and failure modes." }, { "left": "pull_request:2367", "right": "pull_request:4187", "accept": false, - "reason": "Checkpoint retention in training scripts is unrelated to DPMSolver timestep indexing." + "reason": "Training checkpoint retention option vs DPM-Solver timestep handling; unrelated functionality." }, { "left": "pull_request:6759", "right": "pull_request:8937", "accept": false, - "reason": "Separate example/pipeline fixes with no shared concrete bug." + "reason": "InstantID example fix vs long-prompt handling in AuraFlow; different pipelines and issues." }, { "left": "pull_request:12593", "right": "pull_request:12679", "accept": false, - "reason": "ChronoEdit model/pipeline addition versus multi-LoRA support; distinct features." + "reason": "ChronoEdit feature addition vs ChronoEdit LoRA support; related project, but not the same concrete change." }, { "left": "pull_request:7825", "right": "pull_request:9676", "accept": false, - "reason": "Device-map test guards and an import-path fix are unrelated." + "reason": "Model offload test guard vs benchmark import cleanup; no shared bug or implementation target." }, { "left": "pull_request:1913", "right": "pull_request:3292", "accept": false, - "reason": "Input-type validation and input perturbation are different training/pipeline behaviors." + "reason": "Input type validation for img2img pipelines vs training input perturbation; separate concerns." }, { "left": "pull_request:2327", "right": "pull_request:2706", "accept": false, - "reason": "UNet shape validation versus EMA save/restore in training; not the same fix." + "reason": "UNet config length validation vs EMA save/restore in training; different bugs and files." }, { "left": "pull_request:2593", "right": "pull_request:3806", "accept": false, - "reason": "ControlNet conversion tooling and a README note about checkpoint loading are not the same change." + "reason": "ControlNet checkpoint conversion support vs README inference note; docs/support update, not same fix." }, { "left": "pull_request:5665", "right": "pull_request:6994", "accept": false, - "reason": "Image preprocessing sizing and torch_dtype propagation in from_single_file are separate bugs." + "reason": "ControlNet img2img preprocessing fix vs single-file dtype propagation; distinct code paths." }, { "left": "pull_request:2919", "right": "pull_request:2959", "accept": false, - "reason": "Both modify the same test file, but one speeds it up and the other fixes a norm-group-specific test; not duplicates." + "reason": "Two different test adjustments for UNet3D; both are test-only and not the same underlying change." }, { "left": "pull_request:7246", "right": "pull_request:7941", "accept": false, - "reason": "VAE tiling device handling and marking a slow single-file test are unrelated." + "reason": "VAE tiling device fix vs slow-test decorator; unrelated test maintenance." }, { "left": "pull_request:11995", "right": "pull_request:12454", "accept": false, - "reason": "Both are modular-pipeline work, but for different model families and distinct feature sets." - } - ] - }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, - "error_kind": null, - "error_message": null - }, - { - "label": "pull_request", - "component_index": 1, - "component_count": 266, - "review_unit_index": 271, - "review_unit_count": 273, - "status": "reviewed", - "reason": null, - "source": "llm", - "cache_hit": false, - "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-1-3019", - "nodes": [ - "pull_request:10021", - "pull_request:10330", - "pull_request:10537", - "pull_request:10538", - "pull_request:10547", - "pull_request:10815", - "pull_request:11174", - "pull_request:12387", - "pull_request:12549", - "pull_request:12644", - "pull_request:12934", - "pull_request:12983", - "pull_request:13158", - "pull_request:13208", - "pull_request:13216", - "pull_request:13223", - "pull_request:13338", - "pull_request:13365", - "pull_request:1878", - "pull_request:2593", - "pull_request:326", - "pull_request:344", - "pull_request:3471", - "pull_request:3806", - "pull_request:382", - "pull_request:386", - "pull_request:390", - "pull_request:391", - "pull_request:392", - "pull_request:404", - "pull_request:417", - "pull_request:4242", - "pull_request:4262", - "pull_request:4395", - "pull_request:4572", - "pull_request:4963", - "pull_request:5053", - "pull_request:5393", - "pull_request:5506", - "pull_request:5615", - "pull_request:5837", - "pull_request:6743", - "pull_request:6764", - "pull_request:6886", - "pull_request:7200", - "pull_request:8512", - "pull_request:8652", - "pull_request:9251" - ], - "soft_pairs": [ - "pull_request:12934|pull_request:12983", - "pull_request:13208|pull_request:13216", - "pull_request:10330|pull_request:11174", - "pull_request:12644|pull_request:13158", - "pull_request:382|pull_request:391", - "pull_request:386|pull_request:392", - "pull_request:386|pull_request:404", - "pull_request:386|pull_request:417", - "pull_request:390|pull_request:392", - "pull_request:390|pull_request:404", - "pull_request:390|pull_request:417", - "pull_request:5506|pull_request:7200", - "pull_request:13216|pull_request:13223", - "pull_request:10537|pull_request:10538", - "pull_request:10547|pull_request:9251", - "pull_request:1878|pull_request:2593", - "pull_request:4242|pull_request:5837", - "pull_request:4262|pull_request:5837", - "pull_request:4963|pull_request:5615", - "pull_request:6764|pull_request:8652", - "pull_request:6886|pull_request:8512", - "pull_request:12387|pull_request:12549", - "pull_request:10021|pull_request:10815", - "pull_request:13338|pull_request:13365", - "pull_request:326|pull_request:344", - "pull_request:3471|pull_request:3806", - "pull_request:3806|pull_request:4395", - "pull_request:4572|pull_request:5393", - "pull_request:5053|pull_request:6743" - ], - "prepared_review_unit_hash": "61d183a351ac6b9c93bb452ea9090b8cee5bc376893cee138f0d70b5e92346be", - "component_budget": { - "node_count": 3019, - "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 - }, - "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 29, - "serialized_chars": 102719, - "estimated_input_tokens": 25680, - "estimated_eval_tokens": 51616 - }, - "overflow_policy": "truncate_then_skip", - "trimmed": false, - "aggressively_trimmed": false, - "split": true, - "analyst_result": { - "summary": "This cluster is heterogeneous: LTX/LTX2, Helios, Sana, CI workflow, docs, and assorted older utility PRs. The soft links are mostly filename or theme similarity, not the same underlying change, so they should be rejected as duplicates.", - "confidence": 0.98, - "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the items are unrelated across multiple subsystems and change types, so there isn\u2019t one underlying fix or feature to represent the cluster.", - "best_issue_reason": null, - "best_pr_reason": "No global best PR; none of these artifacts is a clear representative of the whole set because the set does not share one concrete bug or change.", - "soft_edge_verdicts": [ + "reason": "Flux modular pipeline addition vs Kontext modular support; same family, but distinct features and pipelines." + }, { "left": "pull_request:12934", "right": "pull_request:12983", "accept": false, - "reason": "Both are LTX2-related, but one adds distilled checkpoint support and the other adds single-file conversion support; different layers of the stack." + "reason": "LTX2 distilled checkpoint feature support vs single-file loader support; related release but not the same concrete change." }, { "left": "pull_request:13208", "right": "pull_request:13216", "accept": false, - "reason": "Helios-related, but one adds the video generation pipelines/docs and the other adds the modular helios implementation; not the same change." + "reason": "Helios docs/pipeline registration vs Helios modular pipeline implementation; different product surfaces." + }, + { + "left": "pull_request:13216", + "right": "pull_request:13223", + "accept": false, + "reason": "Helios modular pipeline addition vs context-parallelism fix in Helios transformer; separate implementation and bug fix." }, { "left": "pull_request:10330", "right": "pull_request:11174", "accept": false, - "reason": "Both touch LTX docs, but 10330 is a 0.9.1 update while 11174 refactors the text-only conditioning pipeline; different underlying work." + "reason": "LTX Video 0.9.1 update vs LTX0.9.5 conditioning refactor; different versions and different changes." }, { "left": "pull_request:12644", "right": "pull_request:13158", "accept": false, - "reason": "Unrelated docs PRs for AutoModel and QwenImage layered pipeline; same broad docs area only." + "reason": "AutoModel docs vs QwenImageLayered docs; unrelated documentation changes." }, { "left": "pull_request:382", "right": "pull_request:391", "accept": false, - "reason": "Both are early docs PRs, but DDPM and PNDM are different pipelines and different content." + "reason": "DDPM docs vs PNDM docs; separate pipeline documentation pages." }, { "left": "pull_request:386", "right": "pull_request:392", "accept": false, - "reason": "Stable Diffusion docs versus contribution docs; no shared bug or feature." + "reason": "Stable Diffusion docs vs conceptual contribution docs; unrelated documentation scope." }, { "left": "pull_request:386", "right": "pull_request:404", "accept": false, - "reason": "Stable Diffusion pipeline docs are unrelated to fp16 optimization docs and pipeline tweaks." - }, - { - "left": "pull_request:386", - "right": "pull_request:417", - "accept": false, - "reason": "Stable Diffusion docs versus conditional-image-generation docs; different pages and purposes." + "reason": "Stable Diffusion docs vs fp16/memory-speed docs; both docs, but different topics and targets." }, { "left": "pull_request:390", "right": "pull_request:392", "accept": false, - "reason": "MPS optimization docs and contribution docs are unrelated." + "reason": "Apple Silicon MPS guide vs contribution docs; unrelated documentation areas." }, { "left": "pull_request:390", "right": "pull_request:404", "accept": false, - "reason": "MPS hardware docs and fp16 optimization docs are adjacent topics but not the same change." + "reason": "MPS optimization guide vs fp16/memory-speed guide; different optimization topics." }, { "left": "pull_request:390", "right": "pull_request:417", "accept": false, - "reason": "Different documentation guides; no common underlying bug or feature." + "reason": "MPS optimization guide vs inference pipeline docs; not the same change." }, { "left": "pull_request:5506", "right": "pull_request:7200", "accept": false, - "reason": "Both are model/attention fixes, but one is a PEFT+xFormers attention call fix and the other is a transformer norm-type validation change." - }, - { - "left": "pull_request:13216", - "right": "pull_request:13223", - "accept": false, - "reason": "Helios modular pipeline addition versus Helios context-parallelism fix; separate code paths and problems." + "reason": "PEFT/xformers LoRA fix vs transformer norm-type validation; different attention/model issues." }, { "left": "pull_request:10537", "right": "pull_request:10538", "accept": false, - "reason": "Sana-4K pipeline tweaks and adding Sana to auto-pipeline registration are related, but not the same concrete fix." + "reason": "Sana PAG pipeline behavior tweak vs auto-pipeline registration; separate feature plumbing." }, { "left": "pull_request:10547", "right": "pull_request:9251", "accept": false, - "reason": "Test offload adjustments for model tests versus CI trigger changes on patch branches; unrelated." - }, + "reason": "Offload test data for CogVideoX/CogView3 vs CI trigger branch change; unrelated test vs workflow update." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 271, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10021", + "pull_request:10063", + "pull_request:10105", + "pull_request:10136", + "pull_request:10245", + "pull_request:10326", + "pull_request:10815", + "pull_request:10983", + "pull_request:11613", + "pull_request:11614", + "pull_request:12387", + "pull_request:12490", + "pull_request:12549", + "pull_request:13338", + "pull_request:13365", + "pull_request:13454", + "pull_request:1878", + "pull_request:2593", + "pull_request:3160", + "pull_request:326", + "pull_request:3351", + "pull_request:344", + "pull_request:3471", + "pull_request:3806", + "pull_request:4242", + "pull_request:4262", + "pull_request:4395", + "pull_request:4572", + "pull_request:4963", + "pull_request:5053", + "pull_request:5393", + "pull_request:5615", + "pull_request:5665", + "pull_request:5788", + "pull_request:5837", + "pull_request:6426", + "pull_request:6743", + "pull_request:6764", + "pull_request:6886", + "pull_request:7113", + "pull_request:7200", + "pull_request:7627", + "pull_request:7938", + "pull_request:8512", + "pull_request:8652", + "pull_request:8684", + "pull_request:8797" + ], + "soft_pairs": [ + "pull_request:1878|pull_request:2593", + "pull_request:4242|pull_request:5837", + "pull_request:4262|pull_request:5837", + "pull_request:4963|pull_request:5615", + "pull_request:6764|pull_request:8652", + "pull_request:6886|pull_request:8512", + "pull_request:12387|pull_request:12549", + "pull_request:10021|pull_request:10815", + "pull_request:13338|pull_request:13365", + "pull_request:326|pull_request:344", + "pull_request:3471|pull_request:3806", + "pull_request:3806|pull_request:4395", + "pull_request:4572|pull_request:5393", + "pull_request:5053|pull_request:6743", + "pull_request:5665|pull_request:7627", + "pull_request:5788|pull_request:7200", + "pull_request:6426|pull_request:8512", + "pull_request:10063|pull_request:10105", + "pull_request:10245|pull_request:8684", + "pull_request:3160|pull_request:3351", + "pull_request:10326|pull_request:8797", + "pull_request:7113|pull_request:7938", + "pull_request:11613|pull_request:11614", + "pull_request:12490|pull_request:13454", + "pull_request:10136|pull_request:10983" + ], + "prepared_review_unit_hash": "c6810c295ae11c0bf0dcda8c47966b19d67584d5a6683651493994d9bbc92d96", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 98849, + "estimated_input_tokens": 24713, + "estimated_eval_tokens": 49682 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly false-positive similarity links across unrelated PRs; no duplicate pair is strong enough to accept under strict PR duplicate rules.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents this cluster: the linked PRs span unrelated docs, CI, scheduler, model, and example changes rather than one shared bug or feature.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR is suitable because the cluster is heterogeneous. If forced, #10136 is the broadest substantive feature PR, but it still does not match the rest of the items.", + "soft_edge_verdicts": [ { "left": "pull_request:1878", "right": "pull_request:2593", "accept": false, - "reason": "Stable-diffusion checkpoint loader guard versus ControlNet checkpoint conversion script; different conversion paths." + "reason": "Different changes: checkpoint loading guard in a conversion script vs a separate ControlNet conversion script. Same general area only." }, { "left": "pull_request:4242", "right": "pull_request:5837", "accept": false, - "reason": "Allowing VAE loading from single-file checkpoints is unrelated to textual inversion multi-token support." + "reason": "Unrelated loader docs/code: allowing VAE loading vs allowing multiple textual inversion tokens." }, { "left": "pull_request:4262", "right": "pull_request:5837", "accept": false, - "reason": "SDXL DreamBooth LoRA fixes and textual inversion multi-token support address different features." + "reason": "Different example fixes in different training paths; not the same underlying bug." }, { "left": "pull_request:4963", "right": "pull_request:5615", "accept": false, - "reason": "Installing accelerate main in CI and reverting SDXL adapter LoRA loading are unrelated changes." + "reason": "One PR installs accelerate main; the other reverts LoRA loading in an SDXL adapter pipeline. No shared bug." }, { "left": "pull_request:6764", "right": "pull_request:8652", "accept": false, - "reason": "PEFT serialization docs versus Lumina-T2X model/pipeline support; different domains." + "reason": "A serialization note for PEFT adapters is unrelated to adding the Lumina-T2X model/pipeline." }, { "left": "pull_request:6886", "right": "pull_request:8512", "accept": false, - "reason": "Both are cleanup/revert-style changes, but one removes torch_dtype deprecation usage and the other removes a deprecated helper function." + "reason": "Deprecation cleanup for `to(dtype=...)` is unrelated to removing a deprecated mask helper." }, { "left": "pull_request:12387", "right": "pull_request:12549", "accept": false, - "reason": "Both concern attention backends, but one adds flash attention via kernels and the other adds AITER support; distinct backends and implementations." + "reason": "Both touch attention backends, but they add different backends (`kernels` flash vs AITER). Not the same concrete change." }, { "left": "pull_request:10021", "right": "pull_request:10815", "accept": false, - "reason": "LTX Video core support and Enhance-A-Video hooks are unrelated features." + "reason": "LTX Video core support and Enhance-A-Video hooks are different feature areas, not one bug fix." }, { "left": "pull_request:13338", "right": "pull_request:13365", "accept": false, - "reason": "Same workflow file, but one adds id-token permission and the other changes fork-review execution; different workflow fixes." + "reason": "Same workflow file, but one adds `id-token` permission while the other changes fork-review behavior. Different changes." }, { "left": "pull_request:326", "right": "pull_request:344", "accept": false, - "reason": "PNDM pipeline type-hint work and VAE model type-hint work are separate refactors." + "reason": "Both are type-hint cleanup PRs, but they target different modules and problems." }, { "left": "pull_request:3471", "right": "pull_request:3806", "accept": false, - "reason": "Dreambooth torch.compile docs note versus checkpoint-loading guidance for text-to-image training; different documentation changes." + "reason": "Different docs examples: one adds a torch.compile note, the other explains loading from a checkpoint." }, { "left": "pull_request:3806", "right": "pull_request:4395", "accept": false, - "reason": "Checkpoint-loading note and DreamBooth class-prompt pre-encoding fix do not address the same issue." + "reason": "Checkpoint-loading guidance is unrelated to the DreamBooth class-prompt pre-encoding fix." }, { "left": "pull_request:4572", "right": "pull_request:5393", "accept": false, - "reason": "LoRA network_alpha naming fix in training code versus HFApi-based docs/example updates; not the same bug." + "reason": "`network_alpha` naming fix in an example is unrelated to the HFApi docs update." }, { "left": "pull_request:5053", "right": "pull_request:6743", "accept": false, - "reason": "Validation image argument handling in T2I Adapter training versus hflip flag correction in Diffusion DPO training; unrelated example fixes." + "reason": "Different training script fixes: validation image argument vs hflip argument naming." + }, + { + "left": "pull_request:5665", + "right": "pull_request:7627", + "accept": false, + "reason": "Both are img2img-related, but one fixes control image preprocessing dimensions and the other handles latents stats for Playground." + }, + { + "left": "pull_request:5788", + "right": "pull_request:7200", + "accept": false, + "reason": "LCM scheduler error handling and Transformer2D layernorm/patch-input validation are unrelated changes." + }, + { + "left": "pull_request:6426", + "right": "pull_request:8512", + "accept": false, + "reason": "LoRA deprecation handling is unrelated to removing the deprecated mask helper function." + }, + { + "left": "pull_request:10063", + "right": "pull_request:10105", + "accept": false, + "reason": "Both are docs nits, but they fix different pages and different broken text/links." + }, + { + "left": "pull_request:10245", + "right": "pull_request:8684", + "accept": false, + "reason": "PEFT adapter docs cleanup is unrelated to the Pillow import fix." + }, + { + "left": "pull_request:3160", + "right": "pull_request:3351", + "accept": false, + "reason": "Same feature family (`use_karras_sigmas`) but different scheduler implementations; not one concrete bug/change." + }, + { + "left": "pull_request:10326", + "right": "pull_request:8797", + "accept": false, + "reason": "Workflow quoting fix and sharded-checkpoint test fixes are unrelated." + }, + { + "left": "pull_request:7113", + "right": "pull_request:7938", + "accept": false, + "reason": "Both touch Docker CI notifications, but one adds Slack reporting and the other swaps the notification implementation. Not the same change." + }, + { + "left": "pull_request:11613", + "right": "pull_request:11614", + "accept": false, + "reason": "A revert of torch.compile CI changes is unrelated to a Docker CI security fix." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13454", + "accept": false, + "reason": "A CLAP config test fix is unrelated to xfail markers for CLIP/transformers compatibility issues." + }, + { + "left": "pull_request:10136", + "right": "pull_request:10983", + "accept": false, + "reason": "Related Hunyuan family features, but one adds core Hunyuan Video while the other adds the I2V pipeline/model extension; separate feature work." } ] }, @@ -188332,7 +188282,7 @@ "component_index": 1, "component_count": 266, "review_unit_index": 272, - "review_unit_count": 273, + "review_unit_count": 272, "status": "reviewed", "reason": null, "source": "llm", @@ -188340,66 +188290,37 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-1-3019", "nodes": [ - "pull_request:10063", - "pull_request:10105", - "pull_request:10136", "pull_request:10148", - "pull_request:10245", - "pull_request:10326", "pull_request:10431", - "pull_request:10983", "pull_request:11001", "pull_request:11058", "pull_request:11142", "pull_request:11231", "pull_request:11269", - "pull_request:11613", - "pull_request:11614", "pull_request:11660", "pull_request:12450", - "pull_request:12490", "pull_request:12491", - "pull_request:13454", "pull_request:1878", "pull_request:2892", "pull_request:2915", - "pull_request:3160", "pull_request:3298", - "pull_request:3351", "pull_request:3586", "pull_request:3589", "pull_request:3806", "pull_request:4412", "pull_request:4749", - "pull_request:5665", "pull_request:5672", - "pull_request:5788", - "pull_request:6426", "pull_request:6450", "pull_request:6752", - "pull_request:7113", - "pull_request:7200", + "pull_request:6886", + "pull_request:7116", "pull_request:7359", - "pull_request:7627", - "pull_request:7938", - "pull_request:8512", - "pull_request:8684", - "pull_request:8797", + "pull_request:8386", + "pull_request:8719", "pull_request:9049", "pull_request:9159" ], "soft_pairs": [ - "pull_request:5665|pull_request:7627", - "pull_request:5788|pull_request:7200", - "pull_request:6426|pull_request:8512", - "pull_request:10063|pull_request:10105", - "pull_request:10245|pull_request:8684", - "pull_request:3160|pull_request:3351", - "pull_request:10326|pull_request:8797", - "pull_request:7113|pull_request:7938", - "pull_request:11613|pull_request:11614", - "pull_request:12490|pull_request:13454", - "pull_request:10136|pull_request:10983", "pull_request:11001|pull_request:11058", "pull_request:2915|pull_request:4412", "pull_request:11231|pull_request:11660", @@ -188412,246 +188333,128 @@ "pull_request:2892|pull_request:3298", "pull_request:3586|pull_request:3589", "pull_request:3586|pull_request:5672", - "pull_request:4749|pull_request:6450" + "pull_request:4749|pull_request:6450", + "pull_request:6886|pull_request:8719", + "pull_request:7116|pull_request:8386" ], - "prepared_review_unit_hash": "255b1d8cc2effbc5d37136201df4e63697aa16220655e6b9ff4bde4026c64292", + "prepared_review_unit_hash": "4fa0c962b7c274ffa1ae75b0c0ac23b28ba3a05c808a954ddd68e2a28116ebb2", "component_budget": { "node_count": 3019, "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 24, - "serialized_chars": 99222, - "estimated_input_tokens": 24806, - "estimated_eval_tokens": 49868 + "node_count": 29, + "item_count": 29, + "soft_pair_count": 15, + "serialized_chars": 62349, + "estimated_input_tokens": 15588, + "estimated_eval_tokens": 31432 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated docs cleanups, CI tweaks, feature additions, and test-only changes. I don\u2019t see any soft-edge pair that represents the same underlying bug or change, so I reject all pairings and treat one broad feature PR as the best representative only by coverage/centrality, not as a duplicate match.", - "confidence": 0.16, + "summary": "This cluster is not a true duplicate set. The PRs span unrelated features, CI/doc changes, and isolated test fixes across different subsystems; the soft edges are false positives from generic similarity (tests, workflows, model family names, or docs).", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No pair looks like a true duplicate. If one PR must represent the cluster, #10136 is the broadest and most central feature PR (largest scope, highest discussion/review activity) among the candidates.", + "canonical_pr_reason": "No single PR is a valid canonical here: the items are not the same underlying bug or change, so picking one would misrepresent the cluster.", "best_issue_reason": null, - "best_pr_reason": "#10136 is the best global representative because it is the most substantial core feature PR in the set and has the strongest activity signal, but it is not a duplicate of the others.", + "best_pr_reason": "No best PR candidate exists for deduplication because there is no coherent duplicate theme across the set; the cluster is heterogeneous.", "soft_edge_verdicts": [ - { - "left": "pull_request:5665", - "right": "pull_request:7627", - "accept": false, - "reason": "Both touch img2img-related code, but they fix different problems in different pipelines: ControlNet height/width preprocessing vs Playground latents handling." - }, - { - "left": "pull_request:5788", - "right": "pull_request:7200", - "accept": false, - "reason": "Both add validation/error handling, but they target different subsystems and failure modes: scheduler step sizing vs Transformer2D patch-input/layernorm validation." - }, - { - "left": "pull_request:6426", - "right": "pull_request:8512", - "accept": false, - "reason": "These are unrelated refactors in different areas: LoRA deprecation cleanup vs removing a deprecated mask helper." - }, - { - "left": "pull_request:10063", - "right": "pull_request:10105", - "accept": false, - "reason": "Both are docs-only fixes, but they address different files and different issues: evaluation doc nits vs a broken README link." - }, - { - "left": "pull_request:10245", - "right": "pull_request:8684", - "accept": false, - "reason": "Unrelated documentation changes: PEFT adapter method naming vs fixing a Pillow import in a ControlNet XL docs example." - }, - { - "left": "pull_request:3160", - "right": "pull_request:3351", - "accept": false, - "reason": "Both add Karras sigmas, but they are separate scheduler implementations and not the same concrete bug/change." - }, - { - "left": "pull_request:10326", - "right": "pull_request:8797", - "accept": false, - "reason": "Different scopes entirely: MPS workflow quoting fix vs test sharding determinism fixes." - }, - { - "left": "pull_request:7113", - "right": "pull_request:7938", - "accept": false, - "reason": "Both modify Docker CI notifications, but one adds Slack reporting while the other swaps the notification implementation; not the same underlying change." - }, - { - "left": "pull_request:11613", - "right": "pull_request:11614", - "accept": false, - "reason": "A CI/test revert and a Docker CI security fix are unrelated." - }, - { - "left": "pull_request:12490", - "right": "pull_request:13454", - "accept": false, - "reason": "Audioldm2 ClapConfig test setup and CLIP-related xfails are different test issues." - }, - { - "left": "pull_request:10136", - "right": "pull_request:10983", - "accept": false, - "reason": "Both are Hunyuan-related, but one adds the core Hunyuan Video pipeline while the other adds the I2V variant and related docs/converters; distinct changes." - }, { "left": "pull_request:11001", "right": "pull_request:11058", "accept": false, - "reason": "Both concern hardware/testing portability, but one enables bnb tests on XPU and the other makes CUDA-only tests device-agnostic; different code paths." + "reason": "Both touch test/device handling, but one enables bnb tests on xpu while the other makes CUDA-only tests device-agnostic; different fixes and code paths." }, { "left": "pull_request:2915", "right": "pull_request:4412", "accept": false, - "reason": "A relaxed golden-image tolerance in one test is unrelated to fixing float16 inference test setup." + "reason": "Both are test-only adjustments, but they fix different failures in different tests; not the same bug or change." }, { "left": "pull_request:11231", "right": "pull_request:11660", "accept": false, - "reason": "Core HiDream model/pipeline addition vs HiDream LoRA support; related ecosystem, but not the same concrete fix." + "reason": "Same model family, but one adds the HiDream Image pipeline and the other adds HiDream LoRA conversion support; separate features." }, { "left": "pull_request:9049", "right": "pull_request:9159", "accept": false, - "reason": "Different Flux issues: prompt-encoding dtype handling vs making the transformer shardable during inference." + "reason": "Flux encode_prompt dtype handling and Flux transformer sharding are unrelated code-path changes." }, { "left": "pull_request:6752", "right": "pull_request:7359", "accept": false, - "reason": "PromptDiffusion pipeline addition and ResAdapter training-script work are separate research-project integrations." + "reason": "Different research-project additions: PromptDiffusion pipeline vs ResAdapter training script." }, { "left": "pull_request:10148", "right": "pull_request:11269", "accept": false, - "reason": "OmniGen implementation vs ONNX runtime backend availability detection are unrelated." + "reason": "OmniGen integration and onnxruntime package-detection changes are unrelated." }, { "left": "pull_request:10431", "right": "pull_request:11142", "accept": false, - "reason": "Sana transformer positional-embedding fix and Sana-Sprint converter/script changes do not address the same bug." + "reason": "Both are Sana-related, but one fixes positional embedding behavior in the transformer and the other adjusts a converter script for Sana Sprint checkpoints; not the same issue." }, { "left": "pull_request:12450", "right": "pull_request:12491", "accept": false, - "reason": "CI dependency-installation cleanup and documentation link checking are separate workflow changes." + "reason": "Both are CI workflow edits, but they address different pipeline/install concerns; not mergeable as one bug fix." }, { "left": "pull_request:1878", "right": "pull_request:3806", "accept": false, - "reason": "Checkpoint-loading robustness in a conversion script is unrelated to documentation guidance for loading from checkpoints." + "reason": "Stable diffusion checkpoint parsing fix vs documentation note about loading from checkpoints; different artifacts and purposes." }, { "left": "pull_request:2892", "right": "pull_request:3298", "accept": false, - "reason": "Different upscaler issues: allowing prompt/image embeds in inputs vs supporting arbitrary aspect ratios." + "reason": "One fixes upscaler input validation for embeds, the other adds arbitrary aspect ratio support to a different pipeline." }, { "left": "pull_request:3586", "right": "pull_request:3589", "accept": false, - "reason": "One adds a community pipeline; the other only updates README references to reference-trick examples." + "reason": "A new community pipeline and README reference-trick updates are not the same underlying change." + }, + { + "left": "pull_request:3586", + "right": "pull_request:5672", + "accept": false, + "reason": "Completely different features: a community Stable Diffusion tiling pipeline vs PixArt non-square image support." }, { "left": "pull_request:4749", "right": "pull_request:6450", "accept": false, - "reason": "Both touch LoRA training/testing, but one fixes SDXL checkpointing while the other is broader LoRA deprecation cleanup." - } - ] - }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, - "error_kind": null, - "error_message": null - }, - { - "label": "pull_request", - "component_index": 1, - "component_count": 266, - "review_unit_index": 273, - "review_unit_count": 273, - "status": "reviewed", - "reason": null, - "source": "llm", - "cache_hit": false, - "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-1-3019", - "nodes": [ - "pull_request:6886", - "pull_request:7116", - "pull_request:8386", - "pull_request:8719" - ], - "soft_pairs": [ - "pull_request:6886|pull_request:8719", - "pull_request:7116|pull_request:8386" - ], - "prepared_review_unit_hash": "db95eb152695d733ac7e91e284ab34e0608ceba583bedac41c298b35fbf444c7", - "component_budget": { - "node_count": 3019, - "item_count": 3019, - "soft_pair_count": 7654, - "serialized_chars": 8728470, - "estimated_input_tokens": 2182118, - "estimated_eval_tokens": 4364492 - }, - "budget": { - "node_count": 4, - "item_count": 4, - "soft_pair_count": 2, - "serialized_chars": 9952, - "estimated_input_tokens": 2488, - "estimated_eval_tokens": 5232 - }, - "overflow_policy": "truncate_then_skip", - "trimmed": false, - "aggressively_trimmed": false, - "split": true, - "analyst_result": { - "summary": "This cluster is not a single duplicate group: it contains two unrelated PR pairs. One pair is about deprecation cleanup in `to()`/VQ docs, the other is about workflow/install changes versus test helper refactors. The soft links are too loose to merge.", - "confidence": 0.96, - "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR cleanly represents the whole cluster because the items split into two unrelated change topics; if one representative must be chosen, PR #6886 is the strongest candidate since it is a merged code-path deprecation fix with accompanying test updates.", - "best_issue_reason": null, - "best_pr_reason": "PR #6886 is the best overall representative: it fixes a concrete API deprecation across real conversion/pipeline code and is more substantive than the doc-only or CI/test-only changes in the other PRs.", - "soft_edge_verdicts": [ + "reason": "DreamBooth SDXL LoRA checkpointing and LoRA deprecation test cleanup are unrelated changes." + }, { "left": "pull_request:6886", "right": "pull_request:8719", "accept": false, - "reason": "Both relate to deprecation wording, but they touch different classes and different code paths (`pipe.to(dtype=...)` vs VQ output docs). They are not the same underlying bug/change and would not be merged as one PR." + "reason": "Both are maintenance-oriented, but one removes deprecated torch_dtype usage across converters/tests while the other updates VQ model deprecation docs; different changes." }, { "left": "pull_request:7116", "right": "pull_request:8386", "accept": false, - "reason": "These are unrelated: one changes workflow dependency installation to use `uv`, while the other refactors test dummy-input shape helpers. Same broad repo area, but not the same fix." + "reason": "CI installation workflow changes and test input-shape refactors are unrelated." } ] }, @@ -189151,30 +188954,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is centered on Flux/SD3 attention-mask support. PR 10044 is the broadest implementation, while 10051 and 10053 are narrower variants focused on wiring an attention mask into Flux attention.", - "confidence": 0.79, + "summary": "All three PRs are variations of the same Flux/SD3 attention-mask work in `attention_processor.py`. PR 10044 is the broadest implementation; 10051 and 10053 are narrower follow-ups around wiring `attn_mask` into the Flux attention call and related conversion/pipeline plumbing.", + "confidence": 0.87, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 10044 is the most comprehensive and representative fix: it adds joint attention-mask preparation in the core attention processor and also touches Flux/SD3-related pipeline paths. The other two PRs are narrower variants of Flux attention-mask wiring.", + "canonical_pr_reason": "PR 10044 is the best canonical PR: it has the broadest scope, explicit Flux & SD3 coverage, the most discussion/review activity, and looks like the main implementation rather than a narrow follow-up.", "best_issue_reason": null, - "best_pr_reason": "PR 10044 best captures the full underlying change set for this cluster, rather than only the narrower SDPA attn_mask hookup seen in 10051/10053.", + "best_pr_reason": "PR 10044 is the strongest representative of the cluster because it addresses the feature at the source with the most complete scope and evidence of review.", "soft_edge_verdicts": [ { "left": "pull_request:10044", "right": "pull_request:10051", - "accept": false, - "reason": "Related but not the same concrete fix: 10044 introduces a new joint-mask preparation path for Flux/SD3, while 10051 mainly threads an existing mask into scaled_dot_product_attention and adjusts Flux pipeline/scheduler code." + "accept": true, + "reason": "Both target the same underlying attention-mask support in `attention_processor.py` for Flux, and both are tied to the same issue. 10051 looks like a narrower implementation slice of the broader 10044 fix." }, { "left": "pull_request:10051", "right": "pull_request:10053", "accept": true, - "reason": "These are near-duplicate implementations of the same Flux attention-mask wiring change, sharing the same core attention_processor edit and issue target, with 10053 looking like a reduced subset of 10051." + "reason": "These are effectively the same Flux attention-mask change: both modify the same attention call to pass `attn_mask`, with only minor surrounding plumbing differences." }, { "left": "pull_request:10044", "right": "pull_request:10053", "accept": false, - "reason": "They overlap in the same file and issue target, but the fixes are materially different: 10044 builds/normalizes a joint attention mask, while 10053 only passes attn_mask through to SDPA." + "reason": "They overlap on `attention_processor.py`, but 10044 is a broader Flux/SD3 mask-preparation change while 10053 is a narrower Flux-only wiring change; not enough evidence that they are the same concrete change." } ] }, @@ -193085,144 +192888,144 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Mostly a loose XPU-related test-fix set rather than true duplicates: many PRs update different failing tests with backend-specific expectations or device helpers, while 11404 and 13535 are separate code-path fixes in quantization/LoRA loading.", - "confidence": 0.94, + "summary": "This cluster is not a true duplicate set: it mixes many unrelated XPU/A100 test-porting PRs across different pipelines plus a few real code fixes. They share a broad accelerator-testing theme, but not the same underlying bug or change.", + "confidence": 0.22, "canonical_issue_reason": null, - "canonical_pr_reason": "11264 is the cleanest representative of the dominant pattern in this cluster: a focused XPU test-failure adjustment using backend-specific expected values.", + "canonical_pr_reason": "No single PR cleanly represents the whole cluster because the items are mostly unrelated. If one must be chosen, PR #11458 is the broadest accelerator-portability test fix, but it still only covers one slice of the cluster.", "best_issue_reason": null, - "best_pr_reason": "11264 best captures the cluster\u2019s main theme without being overly broad: one concrete pipeline test updated for XPU-specific numeric drift.", + "best_pr_reason": "PR #11458 is the best broad representative of the shared theme because it updates test setup/helpers for accelerator-agnostic execution and spans multiple test cases, but it is still not a duplicate of the other PRs.", "soft_edge_verdicts": [ { "left": "pull_request:11219", "right": "pull_request:12350", "accept": false, - "reason": "Different failing tests in unrelated areas (bnb mixed-int8 training vs. Marigold depth); same backend target is not enough." + "reason": "Both are XPU-related test fixes, but they target different components and failures: mixed int8 autocast vs Marigold depth tests." }, { "left": "pull_request:11219", "right": "pull_request:11458", "accept": false, - "reason": "Separate test suites and different failure modes; one is quantization training, the other is Stable Diffusion Safe test setup." + "reason": "Different test failures and code paths; one is a bnb autocast test tweak, the other is Safe Diffusion test setup for XPU/A100." }, { "left": "pull_request:11245", "right": "pull_request:11263", "accept": false, - "reason": "Both are XPU-related, but they fix different pipeline tests in different subsystems (Flux Redux vs. LoRA SD)." + "reason": "Both adjust test expectations for accelerator runs, but Flux Redux inference and LoRA finetuning are unrelated pipelines with different failures." }, { "left": "pull_request:11245", "right": "pull_request:11264", "accept": false, - "reason": "Different pipeline tests and different expected outputs; not the same underlying bug." + "reason": "Different pipeline tests and outputs; no shared bug or merged change beyond broad XPU test flakiness." }, { "left": "pull_request:11245", "right": "pull_request:11266", "accept": false, - "reason": "Different test cases in different pipelines; same backend adaptation theme only." + "reason": "Flux Redux inference and IP-Adapter masking failures are separate pipeline behaviors and should not be merged as one fix." }, { "left": "pull_request:11245", "right": "pull_request:11310", "accept": false, - "reason": "Distinct Stable Diffusion/Flux test failures; no shared concrete code-path bug." + "reason": "These are distinct pipeline test failures in different models; only the accelerator target is shared." }, { "left": "pull_request:11264", "right": "pull_request:11266", "accept": false, - "reason": "Both adjust XPU expectations, but they are separate tests with separate outputs and can be merged independently." + "reason": "Stable Diffusion inpaint vs IP-Adapter multiple masks are different failing tests and different code paths." }, { "left": "pull_request:11264", "right": "pull_request:11310", "accept": false, - "reason": "Same general test-maintenance pattern, but different Stable Diffusion test cases and failure points." + "reason": "Both touch image-generation tests, but the failing assertions and pipeline implementations are unrelated." + }, + { + "left": "pull_request:11264", + "right": "pull_request:11404", + "accept": false, + "reason": "One is a test expectation update; the other includes GGUF loader/quantizer code changes plus tests, so they are not the same fix." }, { "left": "pull_request:11264", "right": "pull_request:11458", "accept": false, - "reason": "Different test modules and distinct failures; not a single duplicate fix." + "reason": "Different Stable Diffusion variants and different failures; both are XPU test adjustments, not one underlying bug." }, { "left": "pull_request:11264", - "right": "pull_request:11404", + "right": "pull_request:12350", "accept": false, - "reason": "One is a test expectation tweak; the other changes LoRA/GGUF device handling in runtime code." + "reason": "Separate pipelines and assertions; no evidence of the same concrete bug." }, { "left": "pull_request:11265", "right": "pull_request:11459", "accept": false, - "reason": "Different tests and different pipelines; both are XPU portability fixes, not duplicates." + "reason": "i2vgen-xl tuple output tolerance and semantic/panorama accelerator setup are unrelated test fixes." }, { "left": "pull_request:11265", "right": "pull_request:12350", "accept": false, - "reason": "Unrelated failures: i2vgen_xl output tolerance vs. Marigold depth test naming/expectations." + "reason": "Different models, different test names, different behavior changes." }, { "left": "pull_request:11266", "right": "pull_request:11310", "accept": false, - "reason": "Both use XPU-specific expectations, but they target separate Stable Diffusion tests with different code paths." + "reason": "IP-Adapter mask outputs and SDXL K-Diffusion sigmas are unrelated test failures." }, { "left": "pull_request:11266", "right": "pull_request:11458", "accept": false, - "reason": "Different test suites and different failure semantics; not mergeable as one fix." + "reason": "Both are XPU test fixes, but they target different pipelines and output slices." }, { "left": "pull_request:11266", "right": "pull_request:11444", "accept": false, - "reason": "Different pipelines and different test infrastructure changes; no single underlying bug." + "reason": "IP-Adapter masking vs Unidiffuser device/offload tests are separate issues." }, { - "left": "pull_request:11444", - "right": "pull_request:11458", - "accept": false, - "reason": "Both are XPU-enablement test PRs, but they cover different pipelines and helper behavior." - }, - { - "left": "pull_request:11458", + "left": "pull_request:11266", "right": "pull_request:12350", "accept": false, - "reason": "Unrelated test failures in different domains; no common concrete bug." + "reason": "Different model families and different failing expectations." }, { - "left": "pull_request:11264", - "right": "pull_request:12350", + "left": "pull_request:11444", + "right": "pull_request:11458", "accept": false, - "reason": "Separate pipeline tests with different expected-output mismatches; not duplicates." + "reason": "Unidiffuser test portability and Safe Diffusion test portability are different test suites with no shared concrete bug." }, { - "left": "pull_request:11266", + "left": "pull_request:11458", "right": "pull_request:12350", "accept": false, - "reason": "Different pipelines and different failure points; only the accelerator theme overlaps." + "reason": "Safe Diffusion XPU/A100 test changes do not overlap with Marigold depth test changes." }, { "left": "pull_request:12350", "right": "pull_request:12842", "accept": false, - "reason": "Marigold depth test fix and PixArt test fix are unrelated changes." + "reason": "Marigold depth and PixArt test expectations are unrelated pipeline tests." }, { "left": "pull_request:12842", "right": "pull_request:13270", "accept": false, - "reason": "One adjusts image expectation values; the other changes distributed parallelism test backend/device handling." + "reason": "PixArt test expectation updates and parallelism infrastructure changes are not the same underlying issue." }, { "left": "pull_request:13270", "right": "pull_request:13535", "accept": false, - "reason": "Different layers entirely: model parallelism test support vs. PEFT hotswap runtime behavior." + "reason": "Parallelism backend selection and PEFT hotswap compilation behavior are distinct code paths and fixes." } ] }, @@ -195539,36 +195342,36 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs cluster loosely around group offloading and test maintenance, but they do not share the same concrete bug or change. The only close pair is the group-offloading test work, yet the edits target different code paths and test files.", - "confidence": 0.56, + "summary": "Mostly loose topical overlap around group offloading and test maintenance, but these PRs fix different problems: a Qwen-specific test skip removal, a PEFT delete_adapters regression, pytest migration work, and a separate group-offloading-with-disk test adjustment. I don\u2019t see true duplicate PR pairs here.", + "confidence": 0.66, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 13491 is the most concrete fix in the cluster: it adjusts the group-offloading disk test path to pass the model\u2019s block modules, which directly addresses a failing behavior rather than just refactoring tests or removing a skip.", + "canonical_pr_reason": "PR 13491 is the closest representative of the shared theme because it directly fixes a concrete group-offloading-with-disk test path, but the cluster is not actually one duplicate change set.", "best_issue_reason": null, - "best_pr_reason": "PR 13491 is the strongest overall candidate because it makes the specific group-offloading-with-disk test logic work, whereas the others are either test migrations, a skip removal, or a different adapter/offloading bug fix.", + "best_pr_reason": "PR 13491 is the strongest standalone candidate since it targets a specific failing test path in group offloading with disk and makes a concrete behavioral adjustment, unlike the broader test-migration PRs.", "soft_edge_verdicts": [ { "left": "pull_request:12382", "right": "pull_request:13234", "accept": false, - "reason": "One removes a Qwen-specific skip; the other is only a pytest migration for group offloading tests. Same broad area, but not the same underlying change or bug." + "reason": "Different scope: 12382 only removes a QwenImageTransformer2DModel test skip, while 13234 is a broad pytest migration for group offloading tests." }, { "left": "pull_request:12639", "right": "pull_request:13491", "accept": false, - "reason": "12639 fixes adapter deletion reapplying group offloading; 13491 fixes a test helper for disk offloading. Different code paths and different failure modes." + "reason": "Different underlying bug: 12639 fixes delete_adapters breaking leaf-level offloading, while 13491 adjusts test setup for group offloading with disk." }, { "left": "pull_request:13234", "right": "pull_request:13242", "accept": false, - "reason": "Both are pytest migrations, but for different test modules and hooks vs group-offloading tests. Refactor similarity is not enough to treat them as duplicates." + "reason": "Both are pytest migration/refactor PRs, but they touch different test files and different subsystems; not the same concrete change." }, { "left": "pull_request:13234", "right": "pull_request:13491", "accept": false, - "reason": "13234 is a test-framework migration, while 13491 is a targeted functional fix for group-offloading disk tests. They are not the same concrete change." + "reason": "13234 is a test-framework migration, whereas 13491 is a specific group-offloading test fix; similar area, but not the same bug or change." } ] }, @@ -195945,30 +195748,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two PRs are closely related CI environment fixes for installing prerelease transformers in workflow tests; the third is a separate model code fix. The CI PRs are similar enough to merge conceptually, but the model fix is unrelated.", - "confidence": 0.87, + "summary": "Two PRs are closely related CI fixes for the same prerelease dependency issue in workflow setup; the BnB quantization PR is unrelated and should not be clustered with them.", + "confidence": 0.84, "canonical_issue_reason": null, - "canonical_pr_reason": "pull_request:12671 is the best representative of the shared CI dependency-install workaround because it targets the broader nightly/release workflow coverage and shows the same fix pattern across multiple jobs.", + "canonical_pr_reason": "pull_request:12671 is the better cluster representative because it targets the same CI dependency regression more broadly across nightly/release workflows, whereas pull_request:12617 is a narrower push-tests-only patch.", "best_issue_reason": null, - "best_pr_reason": "pull_request:12671 is the strongest canonical PR candidate for the CI install issue; pull_request:12617 is narrower and already merged, while pull_request:13481 addresses a different runtime bug in model attention code.", + "best_pr_reason": "pull_request:12671 is the strongest PR candidate for the shared CI regression: it applies the same transformers/huggingface_hub prerelease install workaround in the more comprehensive workflow context.", "soft_edge_verdicts": [ { "left": "pull_request:12617", "right": "pull_request:12671", "accept": true, - "reason": "Both PRs address the same underlying CI failure mode: test workflows needing a prerelease transformers install to work with the updated dependency stack. They modify the same kind of workflow setup and could plausibly be combined into one PR." + "reason": "Both PRs apply the same underlying CI fix: reinstalling prerelease transformers and removing conflicting huggingface_hub/accelerate state in workflow setup to unblock tests. They could plausibly be merged into one PR touching multiple workflow files." }, { "left": "pull_request:12617", "right": "pull_request:13481", "accept": false, - "reason": "These fix different problems in different code paths: one is a GitHub Actions workflow dependency install workaround, the other is a BnB attention/model tensor-shape fix in source code and tests." + "reason": "These address different problems: 12617 is a CI workflow dependency workaround, while 13481 changes Wan attention code and related tests to fix BnB quantization behavior. They are not the same underlying bug or change." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: 12617 and 12671 both patch CI workflow dependency setup to install prerelease transformers, and 13481 is a separate model-code fix. The soft-edge accept between the two CI PRs is reasonable because they could be combined into one workflow-fix PR, and the non-accept for 12617 vs 13481 is appropriately conservative." + "feedback": "The cluster call is grounded in the packet: 12617 and 12671 are both CI workflow patches that reinstall prerelease transformers and remove conflicting huggingface_hub/accelerate state, so treating them as mergeable into one broader CI-fix PR is reasonable. The rejection of 12617 with 13481 is also conservative and supported by the very different model/quantization scope. Minor caution: the wording about \"nightly/release workflows\" leans on the PR title more than the diff snippet, but it does not overstate the evidence enough to invalidate the result." }, "evaluator_used": true, "retried": false, @@ -198286,18 +198089,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are not duplicates: one PR fixes attention-backend LSE broadcasting for flash/flash-3 ring CP, while the other fixes Flux PyTorch/XLA TPU inference timesteps and step marking. Different code paths, backends, and failure modes.", - "confidence": 0.96, + "summary": "These two pull requests are unrelated: one fixes attention backend behavior in core model code, the other patches a specific PyTorch/XLA Flux inference example for TPU execution.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 13182 is the stronger canonical PR because it is merged and fixes a concrete backend bug in `src/diffusers/models/attention_dispatch.py` with a targeted broadcast-handling change.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 13182 is the best representative of the cluster: it is merged, more directly about a specific bug fix, and its diff is tightly scoped to the failing attention backend behavior.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:13182", "right": "pull_request:13457", "accept": false, - "reason": "Different underlying problems: 13182 adjusts LSE dimensionality in attention dispatch for flash/flash-3 ring CP, while 13457 changes Flux XLA inference behavior to keep timesteps on XLA and restore `xm.mark_step()`. They do not appear mergeable into one fix." + "reason": "Different bugs and code paths: PR 13182 changes core attention dispatch logic to handle LSE shape across flash/cudnn backends, while PR 13457 only adjusts a TPU inference example and XLA step handling for Flux. They do not look like the same underlying change and would not plausibly be merged into one PR." } ] }, @@ -198745,18 +198548,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are Neuron-related infrastructure changes with heavy file overlap, but they target different goals: general AWS Neuron device support vs. torch.compile compatibility on Neuron. Related, but not the same duplicate change.", - "confidence": 0.88, + "summary": "Two highly overlapping Neuron PRs touch mostly the same files and some near-identical setup code, but the titles suggest different goals: broad Trainium/Inferentia device support vs torch.compile compatibility. They look related, not the same concrete change.", + "confidence": 0.77, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 13289 is the broader umbrella change for official Neuron device support and is the better representative of the shared cluster topic.", + "canonical_pr_reason": "PR 13289 is the broader umbrella change for official Neuron device support and is the better representative of this cluster; 13485 reads like a narrower follow-up around compile compatibility.", "best_issue_reason": null, - "best_pr_reason": "PR 13289 is the best representative because it covers the wider Neuron enablement effort, while 13485 reads like a narrower follow-up focused on torch.compile compatibility.", + "best_pr_reason": "PR 13289 is the most canonical-looking artifact because it frames the overall device support effort and appears to subsume the shared infrastructure changes.", "soft_edge_verdicts": [ { "left": "pull_request:13289", "right": "pull_request:13485", "accept": false, - "reason": "Do not merge: the titles indicate different concrete objectives, and the shared files suggest related Neuron plumbing rather than the same exact bug or change." + "reason": "High file overlap, but the PRs target different outcomes and are not clearly the same concrete fix; they could be related follow-ups rather than duplicates." } ] }, @@ -199635,18 +199438,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs target the same hook path in `src/diffusers/hooks/pyramid_attention_broadcast.py` and implement the same concrete fix: drop the redundant `iteration == 0` guard and invalidate cached attention outside the timestep window. They are near-duplicates of the same bugfix.", - "confidence": 0.95, + "summary": "Both PRs target the same cache/iteration logic in `pyramid_attention_broadcast.py` and are likely alternate fixes for the same underlying hook bug. PR 13497 is the cleaner representative because it explicitly removes the redundant `iteration == 0` guard and clears stale cache outside the timestep range.", + "confidence": 0.82, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #13467 is the better canonical PR because it most explicitly covers the full bugfix in its title and diff intent: removing the redundant iteration-0 recompute path and clearing stale cache outside the timestep range.", + "canonical_pr_reason": "PR 13497 is the best canonical PR: it is open, narrowly focused, and directly addresses the same cache invalidation / redundant-guard logic in the hook.", "best_issue_reason": null, - "best_pr_reason": "PR #13467 is the strongest representative of the cluster: it addresses the same code path and the broader cache-staleness behavior, making it the clearest standalone fix for this issue.", + "best_pr_reason": "PR 13497 is the strongest cluster representative because it presents the same fix more cleanly and explicitly than the older closed PR 13467.", "soft_edge_verdicts": [ { "left": "pull_request:13467", "right": "pull_request:13497", "accept": true, - "reason": "Same file, same function, same underlying bugfix. Both remove the iteration-0 special case and address stale cache invalidation outside the timestep range, so they are effectively the same change with minor wording/diff differences." + "reason": "Accept: both PRs modify the same file and appear to fix the same concrete code path in the PAB hook\u2014redundant iteration-0 recomputation plus stale-cache handling outside the timestep range." } ] }, @@ -199698,18 +199501,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs target the same profiling decorator bug in `examples/profiling/profiling_utils.py` and the same tracking issue, but they use different fixes: #13468 applies a broad class-level rebinding, while #13471 is a later merged LTX2-specific workaround. They are related, but not the same concrete change.", - "confidence": 0.86, + "summary": "Two related profiling PRs address method-decoration isolation in `examples/profiling/profiling_utils.py`, but they are not the same change: one proposes a broad class-level patching strategy, while the other applies a narrower LTX2-only fix and was merged.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "#13471 is the merged fix and the narrower, final implementation for the profiling instance-isolation bug.", + "canonical_pr_reason": "PR #13471 is the better canonical PR because it was merged and contains the final, narrower fix for the concrete isolation bug in the profiling helper.", "best_issue_reason": null, - "best_pr_reason": "#13471 is the strongest representative because it is merged and directly addresses the bug in the same code path.", + "best_pr_reason": "PR #13471 is the best representative artifact: it resolves the bug in the shipped code path, is merged, and its diff reflects the implemented solution rather than an exploratory/generalized approach.", "soft_edge_verdicts": [ { "left": "pull_request:13468", "right": "pull_request:13471", "accept": false, - "reason": "Same general bug area and tracking issue, but the diffs implement different fixes: a broad class-level patch versus an LTX2-only conditional workaround. They are competing alternatives, not one combined duplicate change." + "reason": "They touch the same file and issue target, but the implementations differ materially: #13468 patches sub-component methods at the class level for all pipelines, while #13471 applies an LTX2-specific workaround and preserves original behavior elsewhere. That is related, not the same concrete change." } ] }, @@ -199764,30 +199567,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two PRs are effectively the same maintenance change to the main documentation build workflow; the upload-doc workflow bump is related but not the same concrete change.", - "confidence": 0.91, + "summary": "Two PRs are direct duplicates of the same documentation workflow maintenance change; the upload-workflow bump is separate. The merged main-doc workflow PR is the best representative.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 13555 is the strongest representative: it is merged, directly updates the main documentation build workflow, and matches the most clearly duplicated change among the candidates.", + "canonical_pr_reason": "PR #13555 is the strongest canonical PR: it is merged, touches the main documentation build workflow, and is the cleanest representative of the repeated doc-builder SHA bump change.", "best_issue_reason": null, - "best_pr_reason": "PR 13555 is the best overall choice because it is merged, concrete, and the cleanest example of the shared doc-builder SHA bump pattern in the cluster.", + "best_pr_reason": "PR #13555 is the best overall PR because it is merged and implements the same workflow SHA bump in the main doc build path, making it the most complete and authoritative instance.", "soft_edge_verdicts": [ { "left": "pull_request:13476", "right": "pull_request:13493", "accept": false, - "reason": "Both bump doc-builder SHAs, but they touch different workflow paths (`upload_pr_documentation.yml` vs `build_documentation.yml`) and are not the same concrete change." + "reason": "Different workflows and different files: one updates the PR documentation upload workflow, the other the main documentation build workflow. Same kind of dependency bump, but not the same underlying change." }, { "left": "pull_request:13476", "right": "pull_request:13555", "accept": false, - "reason": "These update different documentation workflows with different behavior, so they are related maintenance changes rather than the same underlying fix/change." + "reason": "These target different doc-builder workflows and different YAML files, so they are not the same concrete bug/change despite both being SHA bumps." }, { "left": "pull_request:13493", "right": "pull_request:13555", "accept": true, - "reason": "Both PRs make the same concrete change in the same file and workflow path: bumping the doc-builder SHA for `build_documentation.yml`." + "reason": "Both PRs update the same `.github/workflows/build_documentation.yml` doc-builder reference for the main doc build workflow, with nearly identical intent and only the SHA differing. They are the same underlying maintenance change." } ] }, @@ -199902,18 +199705,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are not duplicates: one adds TorchAO dequantization/XPU-related quantization support and broader test coverage, while the other only shrinks WanAnimate test inputs to avoid OOM. They touch one overlapping test file but fix different problems.", - "confidence": 0.96, + "summary": "Two open PRs share one test file but address different problems: TorchAO quantization/XPU support vs. shrinking WanAnimate test tensors to avoid OOM. They should not be deduplicated.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #13530 is the more substantive and central change: it modifies the TorchAO quantizer itself and expands coverage/support, whereas #13541 is a narrow test-only size reduction.", + "canonical_pr_reason": "PR #13530 is the broader, more substantive TorchAO quantization change and best represents the cluster\u2019s main code-path work; #13541 is a narrow test-only size reduction.", "best_issue_reason": null, - "best_pr_reason": "#13530 best represents the cluster because it addresses the underlying TorchAO quantization behavior rather than only adjusting a test fixture.", + "best_pr_reason": "#13530 is the best PR to keep as the representative because it changes the quantizer implementation and associated tests, whereas #13541 only adjusts test inputs to reduce memory use.", "soft_edge_verdicts": [ { "left": "pull_request:13530", "right": "pull_request:13541", "accept": false, - "reason": "Do not merge: they address different concerns. #13530 adds dequantization/XPU quantization support, while #13541 only reduces WanAnimate test tensor sizes to prevent OOM. Shared test-file overlap is incidental." + "reason": "Shared test-file overlap is incidental; one PR adds TorchAO dequantization/XPU-related behavior, the other only reduces WanAnimate test tensor sizes to prevent OOM. They are not the same underlying fix." } ] }, @@ -200811,7 +200614,7 @@ "pull_request:2067|pull_request:2147", "pull_request:2096|pull_request:2147" ], - "prepared_review_unit_hash": "4b269ce556d1ad28cdb8a80db9d2f03d99dad3485ae49d8cead2a87b65c0d9e5", + "prepared_review_unit_hash": "82935e60182d57dabd2f2815ebf7f72eade0be9ac1efd565f421e03d0be4bafa", "component_budget": { "node_count": 3, "item_count": 3, @@ -200833,36 +200636,36 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All three PRs target the same legacy Stable Diffusion inpaint bug: mask tensors were being handled at the wrong shape/scale and the final latents were incorrectly noised. PR #2147 is the merged, most complete fix with tests, so it is the best canonical PR.", - "confidence": 0.93, + "summary": "All three pull requests target the same legacy Stable Diffusion inpaint bug/fix set: resizing the mask tensor to latent size and preventing extra noise from being added to the final latents. PR 2147 appears to be the finalized, merged version with tests; 2067 and 2096 are earlier overlapping attempts.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #2147 is the merged, most complete implementation of the legacy inpaint fix and includes tests; it covers the same concrete mask-resize/noise bug as the other two PRs.", + "canonical_pr_reason": "PR 2147 is the canonical PR: it is merged, covers the same concrete legacy inpaint code path, and looks like the most complete/finalized fix with accompanying tests.", "best_issue_reason": null, - "best_pr_reason": "#2147 is the strongest representative because it is merged, addresses the full legacy inpaint bug path, and adds regression tests. The other PRs are earlier partial/alternative implementations of the same fix.", + "best_pr_reason": "PR 2147 is the best representative overall because it is merged and most fully addresses the underlying bug, while the other two are earlier overlapping iterations.", "soft_edge_verdicts": [ { "left": "pull_request:2067", "right": "pull_request:2096", "accept": true, - "reason": "Same concrete bug in the same pipeline: resizing/formatting the mask tensor for legacy inpainting and preventing noise from being added to final latents. The changes are overlapping enough to plausibly merge into one PR." + "reason": "Same legacy inpaint bug/fix set in the same file: both address scaling the mask tensor to latent size and avoiding noise on final latents, so they look like overlapping iterations of one fix." }, { "left": "pull_request:2067", "right": "pull_request:2147", "accept": true, - "reason": "Both PRs fix the same legacy inpaint code path: mask tensor resizing/handling and final-latent noise behavior. #2147 is a cleaner final form, but the underlying change is the same." + "reason": "Same concrete bug in the same code path; PR 2147 is a later, more complete version of the same legacy inpaint mask-resize/noise fix." }, { "left": "pull_request:2096", "right": "pull_request:2147", "accept": true, - "reason": "These are clearly the same bug fix in the same file and code path, with essentially the same goal of correcting mask tensor scaling and latent noise handling." + "reason": "Both target the same underlying legacy inpaint correction, and PR 2147 clearly refines the same fix into the merged final form." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays reasonably conservative. All three PRs clearly target the same legacy inpaint code path, and the canonical choice of #2147 is supported by it being merged and including tests. The soft-edge accepts are also plausible because the diffs overlap on the same file and same bugfix theme, so they look mergeable into a single PR." + "feedback": "The summary is well grounded in the packet: all three PRs target the same legacy inpaint code path and the same mask-resize / final-latents noise fix, and PR 2147 is the only merged one with a test file. The pairwise overlap verdicts are conservative enough based on the shared file and closely related diffs." }, "evaluator_used": true, "retried": false, @@ -207419,24 +207222,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are not duplicates: 6933 is a conversion-script cleanup, 8183 fixes a logger typo in single_file_utils.py, and 9288 adapts FreeNoise for AnimateDiff V3. The shared file between 6933 and 8183 is incidental; 8183 and 9288 do not share a concrete code path.", - "confidence": 0.98, + "summary": "These three PRs are not duplicates: they address different bugs in different code paths (SD conversion, logger typo handling, and FreeNoise/AnimateDiff V3 compatibility).", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR for this cluster; the three PRs fix unrelated bugs, so none should be treated as the duplicate representative.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 8183 is the most self-contained concrete bug fix with an explicit linked issue, but it is still not a duplicate of the others.", + "best_pr_reason": "No single PR cleanly represents the cluster because the changes are unrelated; if forced, PR #8183 is the most self-contained and explicitly issue-linked, but it is not a good cluster canonical.", "soft_edge_verdicts": [ { "left": "pull_request:6933", "right": "pull_request:8183", "accept": false, - "reason": "Shared filename only; 6933 changes conversion logic across scripts, while 8183 is a narrow logger.warning typo fix in a different bug path." + "reason": "Shared file alone is not enough; one PR fixes SD conversion reshaping/text encoder conversion, the other fixes a logger.warning typo in checkpoint loading." }, { "left": "pull_request:8183", "right": "pull_request:9288", "accept": false, - "reason": "Different features and code paths: 8183 fixes a logging attribute error, while 9288 changes FreeNoise block handling for AnimateDiff V3." + "reason": "Different bugs and different behavior: logger warning attribute fix versus FreeNoise mid_block handling for AnimateDiff V3." } ] }, diff --git a/analysis/current/manifest.json b/analysis/current/manifest.json index 68cad1859aba187e98970b46b99d3671dc7e2d67..a081afe0e8356f66e66f2a5e6ec646e00aea0573 100644 --- a/analysis/current/manifest.json +++ b/analysis/current/manifest.json @@ -1,8 +1,8 @@ { - "analysis_id": "hybrid-model-20260427t112541z", + "analysis_id": "hybrid-model-20260427t113512z", "archived_artifacts": { - "hybrid": "snapshots/20260427T112541Z/analysis-runs/hybrid-model-20260427t112541z/analysis-report-hybrid.json", - "hybrid_reviews": "snapshots/20260427T112541Z/analysis-runs/hybrid-model-20260427t112541z/analysis-report-hybrid.llm-reviews.json" + "hybrid": "snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.llm-reviews.json" }, "artifacts": { "hybrid": "analysis/current/analysis-report-hybrid.json", @@ -10,9 +10,9 @@ }, "channel": "canonical", "model": null, - "published_at": "2026-04-27T11:44:48Z", + "published_at": "2026-04-27T12:02:06Z", "repo": "huggingface/diffusers", "schema_version": 1, - "snapshot_id": "20260427T112541Z", + "snapshot_id": "20260427T113512Z", "variant": "hybrid" } diff --git a/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.json b/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.json new file mode 100644 index 0000000000000000000000000000000000000000..f3ca9db7dfe531e32f2973dc803cd1d8a8efd030 --- /dev/null +++ b/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.json @@ -0,0 +1,17836 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/diffusers", + "snapshot_id": "20260427T113512Z", + "generated_at": "2026-04-27T12:02:03Z", + "evidence_quality": "full", + "llm_enrichment": true, + "meta_bugs": [ + { + "cluster_id": "cluster-1246-28", + "summary": "Cluster of 23 issues and 5 PRs centered on issue #13124.", + "status": "open", + "confidence": 0.99, + "canonical_issue_number": 13124, + "canonical_pr_number": 6566, + "issue_numbers": [ + 1246, + 2780, + 4566, + 5412, + 5757, + 5840, + 6086, + 6098, + 6124, + 6277, + 6363, + 6366, + 6442, + 6454, + 6482, + 7330, + 8659, + 8871, + 9628, + 9973, + 10752, + 10910, + 13124 + ], + "pr_numbers": [ + 6566, + 10783, + 10889, + 13273, + 13510 + ], + "evidence_types": [ + "closing_reference", + "duplicate_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6566, + "right_pr_number": 10783, + "code_similarity": 0.052, + "size_similarity": 0.235, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.031, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6566, + "right_pr_number": 10889, + "code_similarity": 0.046, + "size_similarity": 0.206, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.031, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6566, + "right_pr_number": 13273, + "code_similarity": 0.039, + "size_similarity": 0.178, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.025, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6566, + "right_pr_number": 13510, + "code_similarity": 0.081, + "size_similarity": 0.294, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.15, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10783, + "right_pr_number": 10889, + "code_similarity": 0.893, + "size_similarity": 0.875, + "file_overlap": 1.0, + "area_overlap": 0.765, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 203, + 210 + ], + [ + 214, + 221 + ] + ], + "right_ranges": [ + [ + 203, + 209 + ], + [ + 213, + 219 + ] + ] + } + ] + }, + { + "left_pr_number": 10783, + "right_pr_number": 13273, + "code_similarity": 0.177, + "size_similarity": 0.042, + "file_overlap": 0.059, + "area_overlap": 0.353, + "patch_similarity": 0.186, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 203, + 210 + ], + [ + 214, + 221 + ] + ], + "right_ranges": [ + [ + 216, + 222 + ] + ] + } + ] + }, + { + "left_pr_number": 10783, + "right_pr_number": 13510, + "code_similarity": 0.17, + "size_similarity": 0.8, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.065, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10889, + "right_pr_number": 13273, + "code_similarity": 0.135, + "size_similarity": 0.037, + "file_overlap": 0.059, + "area_overlap": 0.235, + "patch_similarity": 0.186, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 203, + 209 + ], + [ + 213, + 219 + ] + ], + "right_ranges": [ + [ + 216, + 222 + ] + ] + } + ] + }, + { + "left_pr_number": 10889, + "right_pr_number": 13510, + "code_similarity": 0.15, + "size_similarity": 0.7, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.065, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 13273, + "right_pr_number": 13510, + "code_similarity": 0.037, + "size_similarity": 0.052, + "file_overlap": 0.059, + "area_overlap": 0.0, + "patch_similarity": 0.057, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora.py", + "left_ranges": [ + [ + 155, + 173 + ] + ], + "right_ranges": [ + [ + 1398, + 1413 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6585-11", + "summary": "Cluster of 1 issues and 10 PRs centered on issue #6585.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 6585, + "canonical_pr_number": 6737, + "issue_numbers": [ + 6585 + ], + "pr_numbers": [ + 6672, + 6719, + 6737, + 6740, + 6765, + 6766, + 6767, + 6768, + 6769, + 6781 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6672, + "right_pr_number": 6719, + "code_similarity": 0.917, + "size_similarity": 0.938, + "file_overlap": 1.0, + "area_overlap": 0.824, + "patch_similarity": 0.941, + "shared_filenames": [ + "src/diffusers/loaders/unet.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders/unet.py", + "left_ranges": [ + [ + 17, + 23 + ], + [ + 504, + 511 + ] + ], + "right_ranges": [ + [ + 16, + 22 + ], + [ + 504, + 512 + ] + ] + } + ] + }, + { + "left_pr_number": 6672, + "right_pr_number": 6737, + "code_similarity": 0.19, + "size_similarity": 0.8, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.2, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6672, + "right_pr_number": 6740, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6672, + "right_pr_number": 6765, + "code_similarity": 0.131, + "size_similarity": 0.577, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.105, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6672, + "right_pr_number": 6766, + "code_similarity": 0.218, + "size_similarity": 0.867, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.3, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6672, + "right_pr_number": 6767, + "code_similarity": 0.159, + "size_similarity": 0.6, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.259, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6672, + "right_pr_number": 6768, + "code_similarity": 0.178, + "size_similarity": 0.682, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.28, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6672, + "right_pr_number": 6769, + "code_similarity": 0.188, + "size_similarity": 0.833, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.143, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6672, + "right_pr_number": 6781, + "code_similarity": 0.242, + "size_similarity": 0.789, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.56, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6719, + "right_pr_number": 6737, + "code_similarity": 0.179, + "size_similarity": 0.75, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.19, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6719, + "right_pr_number": 6740, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6719, + "right_pr_number": 6765, + "code_similarity": 0.138, + "size_similarity": 0.615, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.103, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6719, + "right_pr_number": 6766, + "code_similarity": 0.205, + "size_similarity": 0.812, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.286, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6719, + "right_pr_number": 6767, + "code_similarity": 0.166, + "size_similarity": 0.64, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.25, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6719, + "right_pr_number": 6768, + "code_similarity": 0.186, + "size_similarity": 0.727, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.269, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6719, + "right_pr_number": 6769, + "code_similarity": 0.198, + "size_similarity": 0.889, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.138, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6719, + "right_pr_number": 6781, + "code_similarity": 0.249, + "size_similarity": 0.842, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.538, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6737, + "right_pr_number": 6740, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6737, + "right_pr_number": 6765, + "code_similarity": 0.112, + "size_similarity": 0.462, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.133, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6737, + "right_pr_number": 6766, + "code_similarity": 0.227, + "size_similarity": 0.923, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.286, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6737, + "right_pr_number": 6767, + "code_similarity": 0.123, + "size_similarity": 0.48, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.182, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6737, + "right_pr_number": 6768, + "code_similarity": 0.139, + "size_similarity": 0.545, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.2, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6737, + "right_pr_number": 6769, + "code_similarity": 0.163, + "size_similarity": 0.667, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.2, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6737, + "right_pr_number": 6781, + "code_similarity": 0.149, + "size_similarity": 0.632, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.148, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6740, + "right_pr_number": 6765, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6740, + "right_pr_number": 6766, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6740, + "right_pr_number": 6767, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6740, + "right_pr_number": 6768, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6740, + "right_pr_number": 6769, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6740, + "right_pr_number": 6781, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6765, + "right_pr_number": 6766, + "code_similarity": 0.13, + "size_similarity": 0.5, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.2, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6765, + "right_pr_number": 6767, + "code_similarity": 0.231, + "size_similarity": 0.962, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.257, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6765, + "right_pr_number": 6768, + "code_similarity": 0.199, + "size_similarity": 0.846, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.2, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6765, + "right_pr_number": 6769, + "code_similarity": 0.159, + "size_similarity": 0.692, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.135, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6765, + "right_pr_number": 6781, + "code_similarity": 0.18, + "size_similarity": 0.731, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.225, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6766, + "right_pr_number": 6767, + "code_similarity": 0.145, + "size_similarity": 0.52, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.273, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6766, + "right_pr_number": 6768, + "code_similarity": 0.163, + "size_similarity": 0.591, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.3, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6766, + "right_pr_number": 6769, + "code_similarity": 0.172, + "size_similarity": 0.722, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.182, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6766, + "right_pr_number": 6781, + "code_similarity": 0.17, + "size_similarity": 0.684, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.222, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6767, + "right_pr_number": 6768, + "code_similarity": 0.239, + "size_similarity": 0.88, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.417, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6767, + "right_pr_number": 6769, + "code_similarity": 0.17, + "size_similarity": 0.72, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.172, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6767, + "right_pr_number": 6781, + "code_similarity": 0.188, + "size_similarity": 0.76, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.242, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6768, + "right_pr_number": 6769, + "code_similarity": 0.191, + "size_similarity": 0.818, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.185, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6768, + "right_pr_number": 6781, + "code_similarity": 0.206, + "size_similarity": 0.864, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.219, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6769, + "right_pr_number": 6781, + "code_similarity": 0.207, + "size_similarity": 0.947, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.114, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-4478-9", + "summary": "Cluster of 6 issues and 3 PRs centered on issue #4619.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 4619, + "canonical_pr_number": 4796, + "issue_numbers": [ + 4478, + 4619, + 4752, + 5510, + 7572, + 7854 + ], + "pr_numbers": [ + 4632, + 4796, + 7858 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 4632, + "right_pr_number": 4796, + "code_similarity": 0.14, + "size_similarity": 0.682, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.026, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 4632, + "right_pr_number": 7858, + "code_similarity": 0.034, + "size_similarity": 0.16, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.015, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 4796, + "right_pr_number": 7858, + "code_similarity": 0.093, + "size_similarity": 0.109, + "file_overlap": 0.111, + "area_overlap": 0.0, + "patch_similarity": 0.25, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py", + "left_ranges": [ + [ + 908, + 926 + ] + ], + "right_ranges": [ + [ + 525, + 532 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9567-8", + "summary": "Cluster of 1 issues and 7 PRs centered on issue #9567.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 9567, + "canonical_pr_number": 9608, + "issue_numbers": [ + 9567 + ], + "pr_numbers": [ + 9579, + 9582, + 9583, + 9606, + 9608, + 9724, + 9815 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9579, + "right_pr_number": 9582, + "code_similarity": 0.043, + "size_similarity": 0.175, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.053, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9579, + "right_pr_number": 9583, + "code_similarity": 0.061, + "size_similarity": 0.292, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.015, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9579, + "right_pr_number": 9606, + "code_similarity": 0.088, + "size_similarity": 0.417, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.032, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9579, + "right_pr_number": 9608, + "code_similarity": 0.104, + "size_similarity": 0.376, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.194, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9579, + "right_pr_number": 9724, + "code_similarity": 0.054, + "size_similarity": 0.225, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.062, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9579, + "right_pr_number": 9815, + "code_similarity": 0.05, + "size_similarity": 0.2, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.068, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9582, + "right_pr_number": 9583, + "code_similarity": 0.128, + "size_similarity": 0.6, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.053, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9582, + "right_pr_number": 9606, + "code_similarity": 0.088, + "size_similarity": 0.42, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.027, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9582, + "right_pr_number": 9608, + "code_similarity": 0.022, + "size_similarity": 0.066, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.06, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9582, + "right_pr_number": 9724, + "code_similarity": 0.158, + "size_similarity": 0.778, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.019, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9582, + "right_pr_number": 9815, + "code_similarity": 0.179, + "size_similarity": 0.875, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.027, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9583, + "right_pr_number": 9606, + "code_similarity": 0.142, + "size_similarity": 0.7, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.014, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9583, + "right_pr_number": 9608, + "code_similarity": 0.024, + "size_similarity": 0.11, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.013, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9583, + "right_pr_number": 9724, + "code_similarity": 0.157, + "size_similarity": 0.771, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.02, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9583, + "right_pr_number": 9815, + "code_similarity": 0.141, + "size_similarity": 0.686, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.029, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9606, + "right_pr_number": 9608, + "code_similarity": 0.038, + "size_similarity": 0.157, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.044, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9606, + "right_pr_number": 9724, + "code_similarity": 0.426, + "size_similarity": 0.54, + "file_overlap": 1.0, + "area_overlap": 0.017, + "patch_similarity": 0.081, + "shared_filenames": [ + "src/diffusers/training_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/training_utils.py", + "left_ranges": [ + [ + 36, + 44 + ], + [ + 195, + 207 + ], + [ + 233, + 240 + ], + [ + 253, + 260 + ], + [ + 271, + 279 + ], + [ + 506, + 513 + ], + [ + 541, + 565 + ], + [ + 575, + 584 + ] + ], + "right_ranges": [ + [ + 43, + 51 + ], + [ + 59, + 76 + ] + ] + } + ] + }, + { + "left_pr_number": 9606, + "right_pr_number": 9815, + "code_similarity": 0.413, + "size_similarity": 0.48, + "file_overlap": 1.0, + "area_overlap": 0.018, + "patch_similarity": 0.071, + "shared_filenames": [ + "src/diffusers/training_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/training_utils.py", + "left_ranges": [ + [ + 36, + 44 + ], + [ + 195, + 207 + ], + [ + 233, + 240 + ], + [ + 253, + 260 + ], + [ + 271, + 279 + ], + [ + 506, + 513 + ], + [ + 541, + 565 + ], + [ + 575, + 584 + ] + ], + "right_ranges": [ + [ + 43, + 51 + ], + [ + 61, + 77 + ] + ] + } + ] + }, + { + "left_pr_number": 9608, + "right_pr_number": 9724, + "code_similarity": 0.026, + "size_similarity": 0.085, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.061, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9608, + "right_pr_number": 9815, + "code_similarity": 0.025, + "size_similarity": 0.075, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.067, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9724, + "right_pr_number": 9815, + "code_similarity": 0.884, + "size_similarity": 0.889, + "file_overlap": 1.0, + "area_overlap": 0.893, + "patch_similarity": 0.628, + "shared_filenames": [ + "src/diffusers/training_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/training_utils.py", + "left_ranges": [ + [ + 43, + 51 + ], + [ + 59, + 76 + ] + ], + "right_ranges": [ + [ + 43, + 51 + ], + [ + 61, + 77 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6984-7", + "summary": "Cluster of 1 issues and 6 PRs centered on issue #6984.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 6984, + "canonical_pr_number": 7071, + "issue_numbers": [ + 6984 + ], + "pr_numbers": [ + 7052, + 7071, + 7211, + 7307, + 8376, + 8387 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 7052, + "right_pr_number": 7071, + "code_similarity": 0.063, + "size_similarity": 0.139, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.233, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7052, + "right_pr_number": 7211, + "code_similarity": 0.069, + "size_similarity": 0.185, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.216, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7052, + "right_pr_number": 7307, + "code_similarity": 0.403, + "size_similarity": 0.269, + "file_overlap": 0.5, + "area_overlap": 0.453, + "patch_similarity": 0.274, + "shared_filenames": [ + "examples/community/stable_diffusion_tensorrt_txt2img.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/stable_diffusion_tensorrt_txt2img.py", + "left_ranges": [ + [ + 19, + 28 + ], + [ + 41, + 58 + ], + [ + 592, + 602 + ], + [ + 678, + 685 + ], + [ + 697, + 771 + ], + [ + 1008, + 1013 + ] + ], + "right_ranges": [ + [ + 19, + 36 + ], + [ + 45, + 62 + ], + [ + 594, + 606 + ], + [ + 683, + 780 + ], + [ + 791, + 865 + ], + [ + 1102, + 1107 + ] + ] + } + ] + }, + { + "left_pr_number": 7052, + "right_pr_number": 8376, + "code_similarity": 0.088, + "size_similarity": 0.299, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.187, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7052, + "right_pr_number": 8387, + "code_similarity": 0.088, + "size_similarity": 0.299, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.187, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7071, + "right_pr_number": 7211, + "code_similarity": 0.274, + "size_similarity": 0.754, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.818, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7071, + "right_pr_number": 7307, + "code_similarity": 0.156, + "size_similarity": 0.517, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.352, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7071, + "right_pr_number": 8376, + "code_similarity": 0.186, + "size_similarity": 0.466, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.619, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7071, + "right_pr_number": 8387, + "code_similarity": 0.186, + "size_similarity": 0.466, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.619, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7211, + "right_pr_number": 7307, + "code_similarity": 0.186, + "size_similarity": 0.685, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.326, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7211, + "right_pr_number": 8376, + "code_similarity": 0.223, + "size_similarity": 0.618, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.665, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7211, + "right_pr_number": 8387, + "code_similarity": 0.223, + "size_similarity": 0.618, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.665, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7307, + "right_pr_number": 8376, + "code_similarity": 0.231, + "size_similarity": 0.901, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.341, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7307, + "right_pr_number": 8387, + "code_similarity": 0.231, + "size_similarity": 0.901, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.341, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8376, + "right_pr_number": 8387, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/regional_prompting_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/regional_prompting_stable_diffusion.py", + "left_ranges": [ + [ + 1, + 28 + ], + [ + 33, + 41 + ], + [ + 83, + 182 + ], + [ + 186, + 376 + ], + [ + 399, + 405 + ] + ], + "right_ranges": [ + [ + 1, + 28 + ], + [ + 33, + 41 + ], + [ + 83, + 182 + ], + [ + 186, + 376 + ], + [ + 399, + 405 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-11390-7", + "summary": "Cluster of 1 issues and 6 PRs centered on issue #11390.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 11390, + "canonical_pr_number": 11463, + "issue_numbers": [ + 11390 + ], + "pr_numbers": [ + 11395, + 11426, + 11427, + 11463, + 11491, + 11492 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 11395, + "right_pr_number": 11426, + "code_similarity": 0.271, + "size_similarity": 0.8, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.739, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11395, + "right_pr_number": 11427, + "code_similarity": 0.271, + "size_similarity": 0.8, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.739, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11395, + "right_pr_number": 11463, + "code_similarity": 0.179, + "size_similarity": 0.444, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.604, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11395, + "right_pr_number": 11491, + "code_similarity": 0.328, + "size_similarity": 0.889, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11395, + "right_pr_number": 11492, + "code_similarity": 0.338, + "size_similarity": 0.96, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.972, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11426, + "right_pr_number": 11427, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11426, + "right_pr_number": 11463, + "code_similarity": 0.194, + "size_similarity": 0.556, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.55, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11426, + "right_pr_number": 11491, + "code_similarity": 0.291, + "size_similarity": 0.9, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.739, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11426, + "right_pr_number": 11492, + "code_similarity": 0.28, + "size_similarity": 0.833, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.756, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11427, + "right_pr_number": 11463, + "code_similarity": 0.194, + "size_similarity": 0.556, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.55, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11427, + "right_pr_number": 11491, + "code_similarity": 0.291, + "size_similarity": 0.9, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.739, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11427, + "right_pr_number": 11492, + "code_similarity": 0.28, + "size_similarity": 0.833, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.756, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11463, + "right_pr_number": 11491, + "code_similarity": 0.191, + "size_similarity": 0.5, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.604, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11463, + "right_pr_number": 11492, + "code_similarity": 0.185, + "size_similarity": 0.463, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.615, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11491, + "right_pr_number": 11492, + "code_similarity": 0.331, + "size_similarity": 0.926, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.972, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-6503-6", + "summary": "Cluster of 1 issues and 5 PRs centered on issue #6503.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 6503, + "canonical_pr_number": 6532, + "issue_numbers": [ + 6503 + ], + "pr_numbers": [ + 6525, + 6532, + 6534, + 6555, + 6556 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6525, + "right_pr_number": 6532, + "code_similarity": 0.172, + "size_similarity": 0.549, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.417, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6525, + "right_pr_number": 6534, + "code_similarity": 0.18, + "size_similarity": 0.651, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.333, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6525, + "right_pr_number": 6555, + "code_similarity": 0.252, + "size_similarity": 0.929, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.444, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6525, + "right_pr_number": 6556, + "code_similarity": 0.149, + "size_similarity": 0.528, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.288, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6532, + "right_pr_number": 6534, + "code_similarity": 0.264, + "size_similarity": 0.843, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.638, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6532, + "right_pr_number": 6555, + "code_similarity": 0.162, + "size_similarity": 0.51, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.4, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6532, + "right_pr_number": 6556, + "code_similarity": 0.252, + "size_similarity": 0.962, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.397, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6534, + "right_pr_number": 6555, + "code_similarity": 0.202, + "size_similarity": 0.605, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.541, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6534, + "right_pr_number": 6556, + "code_similarity": 0.244, + "size_similarity": 0.811, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.545, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6555, + "right_pr_number": 6556, + "code_similarity": 0.166, + "size_similarity": 0.491, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.452, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-2098-6", + "summary": "Cluster of 3 issues and 3 PRs centered on issue #3008.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 3008, + "canonical_pr_number": 3197, + "issue_numbers": [ + 2098, + 3008, + 3184 + ], + "pr_numbers": [ + 3197, + 3201, + 3265 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 3197, + "right_pr_number": 3201, + "code_similarity": 0.79, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 0.401, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/lpw_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/lpw_stable_diffusion.py", + "left_ranges": [ + [ + 1, + 30 + ], + [ + 135, + 141 + ], + [ + 196, + 202 + ], + [ + 236, + 242 + ], + [ + 252, + 258 + ], + [ + 340, + 354 + ], + [ + 368, + 417 + ], + [ + 440, + 635 + ], + [ + 646, + 655 + ], + [ + 669, + 739 + ], + [ + 742, + 783 + ], + [ + 791, + 797 + ], + [ + 815, + 865 + ], + [ + 875, + 893 + ], + [ + 927, + 952 + ], + [ + 964, + 973 + ], + [ + 982, + 1028 + ], + [ + 1035, + 1043 + ], + [ + 1050, + 1119 + ], + [ + 1130, + 1146 + ], + [ + 1168, + 1187 + ], + [ + 1199, + 1211 + ], + [ + 1223, + 1237 + ], + [ + 1244, + 1259 + ], + [ + 1286, + 1301 + ], + [ + 1313, + 1325 + ], + [ + 1335, + 1349 + ], + [ + 1356, + 1373 + ], + [ + 1401, + 1422 + ], + [ + 1434, + 1446 + ], + [ + 1455, + 1470 + ] + ], + "right_ranges": [ + [ + 1, + 39 + ], + [ + 144, + 150 + ], + [ + 205, + 211 + ], + [ + 245, + 251 + ], + [ + 261, + 267 + ], + [ + 349, + 363 + ], + [ + 377, + 406 + ], + [ + 429, + 494 + ], + [ + 505, + 512 + ], + [ + 526, + 572 + ], + [ + 575, + 591 + ], + [ + 599, + 605 + ], + [ + 623, + 665 + ], + [ + 675, + 689 + ], + [ + 723, + 738 + ], + [ + 750, + 755 + ], + [ + 764, + 800 + ], + [ + 807, + 813 + ], + [ + 820, + 862 + ], + [ + 873, + 886 + ], + [ + 908, + 920 + ], + [ + 932, + 938 + ], + [ + 950, + 961 + ], + [ + 968, + 980 + ], + [ + 1007, + 1015 + ], + [ + 1027, + 1034 + ], + [ + 1044, + 1055 + ], + [ + 1062, + 1075 + ], + [ + 1103, + 1114 + ], + [ + 1126, + 1133 + ], + [ + 1142, + 1153 + ] + ] + } + ] + }, + { + "left_pr_number": 3197, + "right_pr_number": 3265, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/lpw_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/lpw_stable_diffusion.py", + "left_ranges": [ + [ + 1, + 30 + ], + [ + 135, + 141 + ], + [ + 196, + 202 + ], + [ + 236, + 242 + ], + [ + 252, + 258 + ], + [ + 340, + 354 + ], + [ + 368, + 417 + ], + [ + 440, + 635 + ], + [ + 646, + 655 + ], + [ + 669, + 739 + ], + [ + 742, + 783 + ], + [ + 791, + 797 + ], + [ + 815, + 865 + ], + [ + 875, + 893 + ], + [ + 927, + 952 + ], + [ + 964, + 973 + ], + [ + 982, + 1028 + ], + [ + 1035, + 1043 + ], + [ + 1050, + 1119 + ], + [ + 1130, + 1146 + ], + [ + 1168, + 1187 + ], + [ + 1199, + 1211 + ], + [ + 1223, + 1237 + ], + [ + 1244, + 1259 + ], + [ + 1286, + 1301 + ], + [ + 1313, + 1325 + ], + [ + 1335, + 1349 + ], + [ + 1356, + 1373 + ], + [ + 1401, + 1422 + ], + [ + 1434, + 1446 + ], + [ + 1455, + 1470 + ] + ], + "right_ranges": [ + [ + 1, + 30 + ], + [ + 135, + 141 + ], + [ + 196, + 202 + ], + [ + 236, + 242 + ], + [ + 252, + 258 + ], + [ + 340, + 354 + ], + [ + 368, + 417 + ], + [ + 440, + 635 + ], + [ + 646, + 655 + ], + [ + 669, + 739 + ], + [ + 742, + 783 + ], + [ + 791, + 797 + ], + [ + 815, + 865 + ], + [ + 875, + 893 + ], + [ + 927, + 952 + ], + [ + 964, + 973 + ], + [ + 982, + 1028 + ], + [ + 1035, + 1043 + ], + [ + 1050, + 1119 + ], + [ + 1130, + 1146 + ], + [ + 1168, + 1187 + ], + [ + 1199, + 1211 + ], + [ + 1223, + 1237 + ], + [ + 1244, + 1259 + ], + [ + 1286, + 1301 + ], + [ + 1313, + 1325 + ], + [ + 1335, + 1349 + ], + [ + 1356, + 1373 + ], + [ + 1401, + 1422 + ], + [ + 1434, + 1446 + ], + [ + 1455, + 1470 + ] + ] + } + ] + }, + { + "left_pr_number": 3201, + "right_pr_number": 3265, + "code_similarity": 0.79, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 0.401, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/lpw_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/lpw_stable_diffusion.py", + "left_ranges": [ + [ + 1, + 39 + ], + [ + 144, + 150 + ], + [ + 205, + 211 + ], + [ + 245, + 251 + ], + [ + 261, + 267 + ], + [ + 349, + 363 + ], + [ + 377, + 406 + ], + [ + 429, + 494 + ], + [ + 505, + 512 + ], + [ + 526, + 572 + ], + [ + 575, + 591 + ], + [ + 599, + 605 + ], + [ + 623, + 665 + ], + [ + 675, + 689 + ], + [ + 723, + 738 + ], + [ + 750, + 755 + ], + [ + 764, + 800 + ], + [ + 807, + 813 + ], + [ + 820, + 862 + ], + [ + 873, + 886 + ], + [ + 908, + 920 + ], + [ + 932, + 938 + ], + [ + 950, + 961 + ], + [ + 968, + 980 + ], + [ + 1007, + 1015 + ], + [ + 1027, + 1034 + ], + [ + 1044, + 1055 + ], + [ + 1062, + 1075 + ], + [ + 1103, + 1114 + ], + [ + 1126, + 1133 + ], + [ + 1142, + 1153 + ] + ], + "right_ranges": [ + [ + 1, + 30 + ], + [ + 135, + 141 + ], + [ + 196, + 202 + ], + [ + 236, + 242 + ], + [ + 252, + 258 + ], + [ + 340, + 354 + ], + [ + 368, + 417 + ], + [ + 440, + 635 + ], + [ + 646, + 655 + ], + [ + 669, + 739 + ], + [ + 742, + 783 + ], + [ + 791, + 797 + ], + [ + 815, + 865 + ], + [ + 875, + 893 + ], + [ + 927, + 952 + ], + [ + 964, + 973 + ], + [ + 982, + 1028 + ], + [ + 1035, + 1043 + ], + [ + 1050, + 1119 + ], + [ + 1130, + 1146 + ], + [ + 1168, + 1187 + ], + [ + 1199, + 1211 + ], + [ + 1223, + 1237 + ], + [ + 1244, + 1259 + ], + [ + 1286, + 1301 + ], + [ + 1313, + 1325 + ], + [ + 1335, + 1349 + ], + [ + 1356, + 1373 + ], + [ + 1401, + 1422 + ], + [ + 1434, + 1446 + ], + [ + 1455, + 1470 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9514-5", + "summary": "Cluster of 3 issues and 2 PRs centered on issue #10989.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 10989, + "canonical_pr_number": 10990, + "issue_numbers": [ + 9514, + 10337, + 10989 + ], + "pr_numbers": [ + 10338, + 10990 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10338, + "right_pr_number": 10990, + "code_similarity": 0.016, + "size_similarity": 0.076, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.005, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-9237-5", + "summary": "Cluster of 3 issues and 2 PRs centered on issue #9237.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 9237, + "canonical_pr_number": 9419, + "issue_numbers": [ + 9237, + 9497, + 9548 + ], + "pr_numbers": [ + 9419, + 9549 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9419, + "right_pr_number": 9549, + "code_similarity": 0.143, + "size_similarity": 0.144, + "file_overlap": 0.167, + "area_overlap": 0.029, + "patch_similarity": 0.364, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_flux.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_flux.py", + "left_ranges": [ + [ + 170, + 183 + ], + [ + 1786, + 1792 + ], + [ + 1834, + 1840 + ] + ], + "right_ranges": [ + [ + 183, + 190 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9591-5", + "summary": "Cluster of 5 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 9591, + "issue_numbers": [], + "pr_numbers": [ + 9591, + 9706, + 9707, + 9714, + 9715 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9591, + "right_pr_number": 9706, + "code_similarity": 0.099, + "size_similarity": 0.422, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.097, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9591, + "right_pr_number": 9707, + "code_similarity": 0.099, + "size_similarity": 0.422, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.097, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9591, + "right_pr_number": 9714, + "code_similarity": 0.099, + "size_similarity": 0.422, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.097, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9591, + "right_pr_number": 9715, + "code_similarity": 0.099, + "size_similarity": 0.422, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.097, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9706, + "right_pr_number": 9707, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", + "left_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ], + "right_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ] + } + ] + }, + { + "left_pr_number": 9706, + "right_pr_number": 9714, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", + "left_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ], + "right_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ] + } + ] + }, + { + "left_pr_number": 9706, + "right_pr_number": 9715, + "code_similarity": 0.996, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 0.975, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", + "left_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ], + "right_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ] + } + ] + }, + { + "left_pr_number": 9707, + "right_pr_number": 9714, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", + "left_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ], + "right_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ] + } + ] + }, + { + "left_pr_number": 9707, + "right_pr_number": 9715, + "code_similarity": 0.996, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 0.975, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", + "left_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ], + "right_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ] + } + ] + }, + { + "left_pr_number": 9714, + "right_pr_number": 9715, + "code_similarity": 0.996, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 0.975, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", + "left_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ], + "right_ranges": [ + [ + 187, + 193 + ], + [ + 257, + 277 + ], + [ + 347, + 355 + ], + [ + 373, + 385 + ], + [ + 406, + 415 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-10109-5", + "summary": "Cluster of 1 issues and 4 PRs centered on issue #10357.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 10357, + "canonical_pr_number": 10498, + "issue_numbers": [ + 10357 + ], + "pr_numbers": [ + 10109, + 10364, + 10365, + 10498 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10109, + "right_pr_number": 10364, + "code_similarity": 0.34, + "size_similarity": 0.952, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10109, + "right_pr_number": 10365, + "code_similarity": 0.3, + "size_similarity": 0.909, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.786, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10109, + "right_pr_number": 10498, + "code_similarity": 0.08, + "size_similarity": 0.008, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.524, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10364, + "right_pr_number": 10365, + "code_similarity": 0.309, + "size_similarity": 0.955, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.786, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10364, + "right_pr_number": 10498, + "code_similarity": 0.08, + "size_similarity": 0.008, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.524, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10365, + "right_pr_number": 10498, + "code_similarity": 0.102, + "size_similarity": 0.009, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.667, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-8446-5", + "summary": "Cluster of 5 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 8461, + "issue_numbers": [], + "pr_numbers": [ + 8446, + 8461, + 8476, + 9308, + 9316 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 8446, + "right_pr_number": 8461, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8446, + "right_pr_number": 8476, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8446, + "right_pr_number": 9308, + "code_similarity": 0.321, + "size_similarity": 0.921, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.914, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8446, + "right_pr_number": 9316, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8461, + "right_pr_number": 8476, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8461, + "right_pr_number": 9308, + "code_similarity": 0.321, + "size_similarity": 0.921, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.914, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8461, + "right_pr_number": 9316, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8476, + "right_pr_number": 9308, + "code_similarity": 0.321, + "size_similarity": 0.921, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.914, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 8476, + "right_pr_number": 9316, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 9308, + "right_pr_number": 9316, + "code_similarity": 0.321, + "size_similarity": 0.921, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.914, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-4498-5", + "summary": "Cluster of 5 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 4502, + "issue_numbers": [], + "pr_numbers": [ + 4498, + 4499, + 4500, + 4501, + 4502 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 4498, + "right_pr_number": 4499, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1850, + 1887 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 788 + ] + ], + "right_ranges": [ + [ + 1, + 788 + ] + ] + } + ] + }, + { + "left_pr_number": 4498, + "right_pr_number": 4500, + "code_similarity": 0.999, + "size_similarity": 0.999, + "file_overlap": 1.0, + "area_overlap": 0.999, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1850, + 1887 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 788 + ] + ], + "right_ranges": [ + [ + 1, + 789 + ] + ] + } + ] + }, + { + "left_pr_number": 4498, + "right_pr_number": 4501, + "code_similarity": 0.997, + "size_similarity": 0.994, + "file_overlap": 1.0, + "area_overlap": 0.994, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1850, + 1887 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 788 + ] + ], + "right_ranges": [ + [ + 1, + 793 + ] + ] + } + ] + }, + { + "left_pr_number": 4498, + "right_pr_number": 4502, + "code_similarity": 0.937, + "size_similarity": 0.953, + "file_overlap": 1.0, + "area_overlap": 0.87, + "patch_similarity": 0.948, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1889, + 1957 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 788 + ] + ], + "right_ranges": [ + [ + 1, + 799 + ] + ] + } + ] + }, + { + "left_pr_number": 4499, + "right_pr_number": 4500, + "code_similarity": 0.999, + "size_similarity": 0.999, + "file_overlap": 1.0, + "area_overlap": 0.999, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1850, + 1887 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 788 + ] + ], + "right_ranges": [ + [ + 1, + 789 + ] + ] + } + ] + }, + { + "left_pr_number": 4499, + "right_pr_number": 4501, + "code_similarity": 0.997, + "size_similarity": 0.994, + "file_overlap": 1.0, + "area_overlap": 0.994, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1850, + 1887 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 788 + ] + ], + "right_ranges": [ + [ + 1, + 793 + ] + ] + } + ] + }, + { + "left_pr_number": 4499, + "right_pr_number": 4502, + "code_similarity": 0.937, + "size_similarity": 0.953, + "file_overlap": 1.0, + "area_overlap": 0.87, + "patch_similarity": 0.948, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1889, + 1957 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 788 + ] + ], + "right_ranges": [ + [ + 1, + 799 + ] + ] + } + ] + }, + { + "left_pr_number": 4500, + "right_pr_number": 4501, + "code_similarity": 0.997, + "size_similarity": 0.995, + "file_overlap": 1.0, + "area_overlap": 0.995, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1850, + 1887 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 789 + ] + ], + "right_ranges": [ + [ + 1, + 793 + ] + ] + } + ] + }, + { + "left_pr_number": 4500, + "right_pr_number": 4502, + "code_similarity": 0.938, + "size_similarity": 0.954, + "file_overlap": 1.0, + "area_overlap": 0.871, + "patch_similarity": 0.948, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1889, + 1957 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 789 + ] + ], + "right_ranges": [ + [ + 1, + 799 + ] + ] + } + ] + }, + { + "left_pr_number": 4501, + "right_pr_number": 4502, + "code_similarity": 0.94, + "size_similarity": 0.958, + "file_overlap": 1.0, + "area_overlap": 0.875, + "patch_similarity": 0.948, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/stable_diffusion_xl_reference.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 1850, + 1887 + ] + ], + "right_ranges": [ + [ + 1889, + 1957 + ] + ] + }, + { + "filename": "examples/community/stable_diffusion_xl_reference.py", + "left_ranges": [ + [ + 1, + 793 + ] + ], + "right_ranges": [ + [ + 1, + 799 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-4780-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 5151, + "issue_numbers": [], + "pr_numbers": [ + 4780, + 5151 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 4780, + "right_pr_number": 5151, + "code_similarity": 0.148, + "size_similarity": 0.208, + "file_overlap": 0.148, + "area_overlap": 0.102, + "patch_similarity": 0.175, + "shared_filenames": [ + "src/diffusers/loaders.py", + "src/diffusers/models/attention.py", + "src/diffusers/models/attention_processor.py", + "src/diffusers/models/resnet.py", + "src/diffusers/models/transformer_2d.py", + "src/diffusers/models/unet_2d_condition.py", + "src/diffusers/utils/__init__.py", + "src/diffusers/utils/peft_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders.py", + "left_ranges": [ + [ + 31, + 44 + ], + [ + 52, + 60 + ], + [ + 1392, + 1421 + ], + [ + 1444, + 1449 + ], + [ + 1462, + 1494 + ], + [ + 1497, + 1512 + ], + [ + 1533, + 1538 + ] + ], + "right_ranges": [ + [ + 11, + 16 + ], + [ + 31, + 46 + ], + [ + 72, + 77 + ], + [ + 400, + 406 + ], + [ + 514, + 523 + ], + [ + 528, + 563 + ], + [ + 680, + 756 + ], + [ + 1169, + 1174 + ], + [ + 1210, + 1216 + ], + [ + 1520, + 1559 + ], + [ + 1571, + 1579 + ], + [ + 1600, + 1655 + ], + [ + 1712, + 1718 + ], + [ + 1725, + 1731 + ], + [ + 1749, + 1779 + ], + [ + 1845, + 1865 + ], + [ + 2235, + 2254 + ], + [ + 2289, + 2295 + ], + [ + 2342, + 2357 + ], + [ + 2384, + 2390 + ], + [ + 2402, + 2408 + ], + [ + 2435, + 2441 + ], + [ + 2452, + 2597 + ], + [ + 3176, + 3187 + ], + [ + 3199, + 3207 + ], + [ + 3219, + 3227 + ], + [ + 3230, + 3236 + ], + [ + 3242, + 3248 + ], + [ + 3309, + 3325 + ] + ] + }, + { + "filename": "src/diffusers/models/attention.py", + "left_ranges": [ + [ + 21, + 26 + ], + [ + 295, + 308 + ], + [ + 339, + 354 + ] + ], + "right_ranges": [ + [ + 17, + 23 + ], + [ + 301, + 307 + ], + [ + 318, + 332 + ], + [ + 371, + 379 + ], + [ + 382, + 389 + ] + ] + }, + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 21, + 27 + ], + [ + 137, + 158 + ], + [ + 413, + 456 + ], + [ + 585, + 599 + ], + [ + 604, + 610 + ], + [ + 737, + 753 + ], + [ + 761, + 767 + ], + [ + 797, + 803 + ], + [ + 806, + 813 + ], + [ + 824, + 830 + ], + [ + 952, + 966 + ], + [ + 973, + 979 + ], + [ + 1030, + 1044 + ], + [ + 1058, + 1064 + ] + ], + "right_ranges": [ + [ + 18, + 24 + ], + [ + 137, + 163 + ], + [ + 550, + 557 + ], + [ + 569, + 583 + ], + [ + 588, + 594 + ], + [ + 1014, + 1033 + ], + [ + 1047, + 1055 + ] + ] + }, + { + "filename": "src/diffusers/models/resnet.py", + "left_ranges": [ + [ + 23, + 28 + ], + [ + 126, + 132 + ], + [ + 165, + 173 + ], + [ + 196, + 202 + ], + [ + 212, + 224 + ], + [ + 534, + 546 + ], + [ + 557, + 563 + ], + [ + 583, + 589 + ], + [ + 624, + 635 + ], + [ + 646, + 655 + ] + ], + "right_ranges": [ + [ + 20, + 26 + ], + [ + 150, + 162 + ], + [ + 195, + 206 + ], + [ + 239, + 248 + ], + [ + 258, + 275 + ], + [ + 616, + 624 + ], + [ + 629, + 641 + ], + [ + 652, + 658 + ], + [ + 678, + 684 + ], + [ + 719, + 734 + ], + [ + 745, + 756 + ] + ] + }, + { + "filename": "src/diffusers/models/transformer_2d.py", + "left_ranges": [ + [ + 23, + 28 + ], + [ + 138, + 146 + ], + [ + 196, + 204 + ], + [ + 274, + 280 + ], + [ + 283, + 295 + ], + [ + 325, + 333 + ] + ], + "right_ranges": [ + [ + 20, + 26 + ], + [ + 100, + 108 + ], + [ + 142, + 150 + ], + [ + 200, + 208 + ], + [ + 295, + 315 + ], + [ + 345, + 361 + ] + ] + }, + { + "filename": "src/diffusers/models/unet_2d_condition.py", + "left_ranges": [ + [ + 731, + 737 + ] + ], + "right_ranges": [ + [ + 20, + 26 + ], + [ + 995, + 1003 + ], + [ + 1097, + 1106 + ] + ] + }, + { + "filename": "src/diffusers/utils/__init__.py", + "left_ranges": [ + [ + 67, + 73 + ], + [ + 83, + 94 + ] + ], + "right_ranges": [ + [ + 26, + 36 + ], + [ + 88, + 94 + ], + [ + 98, + 108 + ] + ] + }, + { + "filename": "src/diffusers/utils/peft_utils.py", + "left_ranges": [ + [ + 1, + 153 + ] + ], + "right_ranges": [ + [ + 15, + 25 + ], + [ + 56, + 61 + ], + [ + 108, + 118 + ], + [ + 121, + 142 + ], + [ + 167, + 175 + ], + [ + 194, + 216 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-10025-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #10025.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 10025, + "canonical_pr_number": 10044, + "issue_numbers": [ + 10025 + ], + "pr_numbers": [ + 10044, + 10051, + 10053 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10044, + "right_pr_number": 10051, + "code_similarity": 0.133, + "size_similarity": 0.25, + "file_overlap": 0.2, + "area_overlap": 0.0, + "patch_similarity": 0.152, + "shared_filenames": [ + "src/diffusers/models/attention_processor.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 731, + 773 + ], + [ + 1495, + 1508 + ], + [ + 2356, + 2366 + ] + ], + "right_ranges": [ + [ + 1762, + 1770 + ], + [ + 1858, + 1866 + ] + ] + } + ] + }, + { + "left_pr_number": 10044, + "right_pr_number": 10053, + "code_similarity": 0.27, + "size_similarity": 0.508, + "file_overlap": 0.5, + "area_overlap": 0.0, + "patch_similarity": 0.121, + "shared_filenames": [ + "src/diffusers/models/attention_processor.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 731, + 773 + ], + [ + 1495, + 1508 + ], + [ + 2356, + 2366 + ] + ], + "right_ranges": [ + [ + 1907, + 1915 + ], + [ + 2014, + 2025 + ] + ] + } + ] + }, + { + "left_pr_number": 10051, + "right_pr_number": 10053, + "code_similarity": 0.213, + "size_similarity": 0.127, + "file_overlap": 0.4, + "area_overlap": 0.152, + "patch_similarity": 0.094, + "shared_filenames": [ + "scripts/convert_flux_to_diffusers.py", + "src/diffusers/models/attention_processor.py" + ], + "shared_file_areas": [ + { + "filename": "scripts/convert_flux_to_diffusers.py", + "left_ranges": [ + [ + 31, + 37 + ] + ], + "right_ranges": [ + [ + 31, + 37 + ] + ] + }, + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 1762, + 1770 + ], + [ + 1858, + 1866 + ] + ], + "right_ranges": [ + [ + 1907, + 1915 + ], + [ + 2014, + 2025 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-8810-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #9886.", + "status": "open", + "confidence": 0.6, + "canonical_issue_number": 9886, + "canonical_pr_number": 8864, + "issue_numbers": [ + 9886 + ], + "pr_numbers": [ + 8810, + 8864, + 9917 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 8810, + "right_pr_number": 8864, + "code_similarity": 0.849, + "size_similarity": 0.755, + "file_overlap": 0.875, + "area_overlap": 0.886, + "patch_similarity": 0.835, + "shared_filenames": [ + "docs/source/en/api/pipelines/pag.md", + "src/diffusers/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/pipelines/pag/__init__.py", + "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/pag.md", + "left_ranges": [ + [ + 44, + 51 + ] + ], + "right_ranges": [ + [ + 71, + 81 + ] + ] + }, + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ + [ + 290, + 296 + ], + [ + 690, + 696 + ] + ], + "right_ranges": [ + [ + 332, + 338 + ], + [ + 790, + 796 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/__init__.py", + "left_ranges": [ + [ + 145, + 151 + ], + [ + 493, + 499 + ] + ], + "right_ranges": [ + [ + 168, + 174 + ], + [ + 572, + 578 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/auto_pipeline.py", + "left_ranges": [ + [ + 47, + 53 + ], + [ + 93, + 99 + ] + ], + "right_ranges": [ + [ + 59, + 65 + ], + [ + 135, + 141 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/__init__.py", + "left_ranges": [ + [ + 22, + 28 + ], + [ + 37, + 43 + ] + ], + "right_ranges": [ + [ + 23, + 29 + ], + [ + 47, + 53 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "left_ranges": [ + [ + 1, + 1297 + ] + ], + "right_ranges": [ + [ + 1, + 1291 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "left_ranges": [ + [ + 1247, + 1267 + ] + ], + "right_ranges": [ + [ + 1832, + 1852 + ] + ] + } + ] + }, + { + "left_pr_number": 8810, + "right_pr_number": 9917, + "code_similarity": 0.808, + "size_similarity": 0.773, + "file_overlap": 0.75, + "area_overlap": 0.877, + "patch_similarity": 0.81, + "shared_filenames": [ + "src/diffusers/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/pipelines/pag/__init__.py", + "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ + [ + 290, + 296 + ], + [ + 690, + 696 + ] + ], + "right_ranges": [ + [ + 340, + 346 + ], + [ + 806, + 812 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/__init__.py", + "left_ranges": [ + [ + 145, + 151 + ], + [ + 493, + 499 + ] + ], + "right_ranges": [ + [ + 169, + 175 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/auto_pipeline.py", + "left_ranges": [ + [ + 47, + 53 + ], + [ + 93, + 99 + ] + ], + "right_ranges": [ + [ + 64, + 70 + ], + [ + 136, + 142 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/__init__.py", + "left_ranges": [ + [ + 22, + 28 + ], + [ + 37, + 43 + ] + ], + "right_ranges": [ + [ + 25, + 31 + ], + [ + 48, + 54 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "left_ranges": [ + [ + 1, + 1297 + ] + ], + "right_ranges": [ + [ + 1, + 1342 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "left_ranges": [ + [ + 1247, + 1267 + ] + ], + "right_ranges": [ + [ + 1786, + 1804 + ] + ] + } + ] + }, + { + "left_pr_number": 8864, + "right_pr_number": 9917, + "code_similarity": 0.912, + "size_similarity": 0.977, + "file_overlap": 0.857, + "area_overlap": 0.9, + "patch_similarity": 0.961, + "shared_filenames": [ + "src/diffusers/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/pipelines/pag/__init__.py", + "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ + [ + 332, + 338 + ], + [ + 790, + 796 + ] + ], + "right_ranges": [ + [ + 340, + 346 + ], + [ + 806, + 812 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/__init__.py", + "left_ranges": [ + [ + 168, + 174 + ], + [ + 572, + 578 + ] + ], + "right_ranges": [ + [ + 169, + 175 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/auto_pipeline.py", + "left_ranges": [ + [ + 59, + 65 + ], + [ + 135, + 141 + ] + ], + "right_ranges": [ + [ + 64, + 70 + ], + [ + 136, + 142 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/__init__.py", + "left_ranges": [ + [ + 23, + 29 + ], + [ + 47, + 53 + ] + ], + "right_ranges": [ + [ + 25, + 31 + ], + [ + 48, + 54 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd_img2img.py", + "left_ranges": [ + [ + 1, + 1291 + ] + ], + "right_ranges": [ + [ + 1, + 1342 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "left_ranges": [ + [ + 1832, + 1852 + ] + ], + "right_ranges": [ + [ + 1786, + 1804 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-281-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #281.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 281, + "canonical_pr_number": 361, + "issue_numbers": [ + 281, + 725 + ], + "pr_numbers": [ + 361, + 727 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 361, + "right_pr_number": 727, + "code_similarity": 0.0, + "size_similarity": 0.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-6137-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #8236.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 8236, + "canonical_pr_number": 8312, + "issue_numbers": [ + 6137, + 8236 + ], + "pr_numbers": [ + 6143, + 8312 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6143, + "right_pr_number": 8312, + "code_similarity": 0.239, + "size_similarity": 0.493, + "file_overlap": 0.333, + "area_overlap": 0.0, + "patch_similarity": 0.269, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image_lora.py" + ], + "shared_file_areas": [ + { + "filename": "examples/text_to_image/train_text_to_image_lora.py", + "left_ranges": [ + [ + 630, + 638 + ], + [ + 646, + 651 + ] + ], + "right_ranges": [ + [ + 697, + 718 + ], + [ + 722, + 735 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5249-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #5249.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 5249, + "canonical_pr_number": 5293, + "issue_numbers": [ + 5249 + ], + "pr_numbers": [ + 5293, + 5299, + 5312 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5293, + "right_pr_number": 5299, + "code_similarity": 0.044, + "size_similarity": 0.098, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.164, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5293, + "right_pr_number": 5312, + "code_similarity": 0.028, + "size_similarity": 0.078, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.08, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5299, + "right_pr_number": 5312, + "code_similarity": 0.188, + "size_similarity": 0.795, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.192, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-11730-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #11730.", + "status": "open", + "confidence": 0.6, + "canonical_issue_number": 11730, + "canonical_pr_number": 11744, + "issue_numbers": [ + 11730 + ], + "pr_numbers": [ + 11744, + 11782, + 11783 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 11744, + "right_pr_number": 11782, + "code_similarity": 0.339, + "size_similarity": 0.963, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.975, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11744, + "right_pr_number": 11783, + "code_similarity": 0.339, + "size_similarity": 0.963, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.975, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11782, + "right_pr_number": 11783, + "code_similarity": 0.964, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 0.896, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/dreambooth/test_dreambooth_lora_sd3.py", + "examples/dreambooth/train_dreambooth_lora_sd3.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/test_dreambooth_lora_sd3.py", + "left_ranges": [ + [ + 13, + 27 + ], + [ + 209, + 254 + ] + ], + "right_ranges": [ + [ + 13, + 27 + ], + [ + 209, + 254 + ] + ] + }, + { + "filename": "examples/dreambooth/train_dreambooth_lora_sd3.py", + "left_ranges": [ + [ + 53, + 59 + ], + [ + 322, + 333 + ], + [ + 1273, + 1279 + ], + [ + 1302, + 1316 + ], + [ + 1333, + 1339 + ], + [ + 1935, + 1946 + ], + [ + 1957, + 1963 + ] + ], + "right_ranges": [ + [ + 54, + 60 + ], + [ + 323, + 334 + ], + [ + 1274, + 1280 + ], + [ + 1303, + 1317 + ], + [ + 1334, + 1340 + ], + [ + 1936, + 1947 + ], + [ + 1958, + 1964 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-8384-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #8384.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 8384, + "canonical_pr_number": 11239, + "issue_numbers": [ + 8384 + ], + "pr_numbers": [ + 10973, + 11239, + 11240 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10973, + "right_pr_number": 11239, + "code_similarity": 0.344, + "size_similarity": 0.972, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 10973, + "right_pr_number": 11240, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11239, + "right_pr_number": 11240, + "code_similarity": 0.344, + "size_similarity": 0.972, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-13074-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #13074.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 13074, + "canonical_pr_number": 13075, + "issue_numbers": [ + 13074 + ], + "pr_numbers": [ + 13075, + 13093, + 13514 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13075, + "right_pr_number": 13093, + "code_similarity": 0.695, + "size_similarity": 0.85, + "file_overlap": 1.0, + "area_overlap": 0.5, + "patch_similarity": 0.333, + "shared_filenames": [ + "src/diffusers/models/_modeling_parallel.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/_modeling_parallel.py", + "left_ranges": [ + [ + 22, + 28 + ], + [ + 291, + 304 + ] + ], + "right_ranges": [ + [ + 290, + 301 + ] + ] + } + ] + }, + { + "left_pr_number": 13075, + "right_pr_number": 13514, + "code_similarity": 0.593, + "size_similarity": 0.95, + "file_overlap": 1.0, + "area_overlap": 0.207, + "patch_similarity": 0.205, + "shared_filenames": [ + "src/diffusers/models/_modeling_parallel.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/_modeling_parallel.py", + "left_ranges": [ + [ + 22, + 28 + ], + [ + 291, + 304 + ] + ], + "right_ranges": [ + [ + 299, + 312 + ] + ] + } + ] + }, + { + "left_pr_number": 13093, + "right_pr_number": 13514, + "code_similarity": 0.576, + "size_similarity": 0.895, + "file_overlap": 1.0, + "area_overlap": 0.13, + "patch_similarity": 0.345, + "shared_filenames": [ + "src/diffusers/models/_modeling_parallel.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/_modeling_parallel.py", + "left_ranges": [ + [ + 290, + 301 + ] + ], + "right_ranges": [ + [ + 299, + 312 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6514-4", + "summary": "Cluster of 4 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 6514, + "issue_numbers": [], + "pr_numbers": [ + 6514, + 6554, + 6840, + 7242 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6514, + "right_pr_number": 6554, + "code_similarity": 0.182, + "size_similarity": 0.444, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.623, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6514, + "right_pr_number": 6840, + "code_similarity": 0.161, + "size_similarity": 0.378, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.569, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6514, + "right_pr_number": 7242, + "code_similarity": 0.118, + "size_similarity": 0.089, + "file_overlap": 0.333, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 34, + 40 + ], + [ + 53, + 65 + ], + [ + 1002, + 1007 + ], + [ + 1058, + 1096 + ], + [ + 1105, + 1121 + ], + [ + 1533, + 1539 + ] + ], + "right_ranges": [ + [ + 877, + 884 + ] + ] + } + ] + }, + { + "left_pr_number": 6554, + "right_pr_number": 6840, + "code_similarity": 0.297, + "size_similarity": 0.85, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.847, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6554, + "right_pr_number": 7242, + "code_similarity": 0.04, + "size_similarity": 0.2, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 6840, + "right_pr_number": 7242, + "code_similarity": 0.047, + "size_similarity": 0.235, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-12771-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #12771.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 12771, + "canonical_pr_number": 12788, + "issue_numbers": [ + 12771 + ], + "pr_numbers": [ + 12788, + 13239, + 13240 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 12788, + "right_pr_number": 13239, + "code_similarity": 0.238, + "size_similarity": 0.24, + "file_overlap": 0.5, + "area_overlap": 0.038, + "patch_similarity": 0.178, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", + "left_ranges": [ + [ + 24, + 31 + ], + [ + 412, + 445 + ], + [ + 550, + 587 + ] + ], + "right_ranges": [ + [ + 442, + 452 + ], + [ + 472, + 488 + ] + ] + } + ] + }, + { + "left_pr_number": 12788, + "right_pr_number": 13240, + "code_similarity": 0.236, + "size_similarity": 0.25, + "file_overlap": 0.5, + "area_overlap": 0.038, + "patch_similarity": 0.15, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", + "left_ranges": [ + [ + 24, + 31 + ], + [ + 412, + 445 + ], + [ + 550, + 587 + ] + ], + "right_ranges": [ + [ + 442, + 454 + ], + [ + 465, + 480 + ] + ] + } + ] + }, + { + "left_pr_number": 13239, + "right_pr_number": 13240, + "code_similarity": 0.756, + "size_similarity": 0.962, + "file_overlap": 1.0, + "area_overlap": 0.541, + "patch_similarity": 0.5, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/schedulers/scheduling_dpmsolver_multistep.py", + "left_ranges": [ + [ + 442, + 452 + ], + [ + 472, + 488 + ] + ], + "right_ranges": [ + [ + 442, + 454 + ], + [ + 465, + 480 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-13077-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #13077.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 13077, + "canonical_pr_number": 13136, + "issue_numbers": [ + 13077 + ], + "pr_numbers": [ + 13102, + 13128, + 13136 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 13102, + "right_pr_number": 13128, + "code_similarity": 0.27, + "size_similarity": 0.102, + "file_overlap": 0.5, + "area_overlap": 0.253, + "patch_similarity": 0.074, + "shared_filenames": [ + "src/diffusers/models/controlnets/controlnet_z_image.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", + "left_ranges": [ + [ + 517, + 535 + ] + ], + "right_ranges": [ + [ + 517, + 591 + ] + ] + } + ] + }, + { + "left_pr_number": 13102, + "right_pr_number": 13136, + "code_similarity": 0.693, + "size_similarity": 0.9, + "file_overlap": 1.0, + "area_overlap": 0.531, + "patch_similarity": 0.182, + "shared_filenames": [ + "src/diffusers/models/controlnets/controlnet_z_image.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", + "left_ranges": [ + [ + 517, + 535 + ] + ], + "right_ranges": [ + [ + 517, + 528 + ], + [ + 531, + 548 + ] + ] + } + ] + }, + { + "left_pr_number": 13128, + "right_pr_number": 13136, + "code_similarity": 0.321, + "size_similarity": 0.092, + "file_overlap": 0.5, + "area_overlap": 0.4, + "patch_similarity": 0.083, + "shared_filenames": [ + "src/diffusers/models/controlnets/controlnet_z_image.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/controlnets/controlnet_z_image.py", + "left_ranges": [ + [ + 517, + 591 + ] + ], + "right_ranges": [ + [ + 517, + 528 + ], + [ + 531, + 548 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-1257-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #1257.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 1257, + "canonical_pr_number": 1361, + "issue_numbers": [ + 1257, + 2950 + ], + "pr_numbers": [ + 1361, + 3000 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 1361, + "right_pr_number": 3000, + "code_similarity": 0.018, + "size_similarity": 0.041, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.063, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-1407-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #1792.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 1792, + "canonical_pr_number": 1416, + "issue_numbers": [ + 1407, + 1792 + ], + "pr_numbers": [ + 1416, + 1796 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 1416, + "right_pr_number": 1796, + "code_similarity": 0.149, + "size_similarity": 0.691, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.071, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-11315-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #11315.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 11315, + "canonical_pr_number": 12665, + "issue_numbers": [ + 11315, + 12653 + ], + "pr_numbers": [ + 11316, + 12665 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 11316, + "right_pr_number": 12665, + "code_similarity": 0.35, + "size_similarity": 1.0, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-1466-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #1466.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 1466, + "canonical_pr_number": 2070, + "issue_numbers": [ + 1466, + 2041 + ], + "pr_numbers": [ + 1547, + 2070 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 1547, + "right_pr_number": 2070, + "code_similarity": 0.211, + "size_similarity": 0.636, + "file_overlap": 0.2, + "area_overlap": 0.0, + "patch_similarity": 0.155, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth.py", + "left_ranges": [ + [ + 300, + 309 + ], + [ + 313, + 349 + ], + [ + 534, + 545 + ] + ], + "right_ranges": [ + [ + 240, + 253 + ], + [ + 660, + 666 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-11702-4", + "summary": "Cluster of 2 issues and 2 PRs centered on issue #11709.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 11709, + "canonical_pr_number": 11704, + "issue_numbers": [ + 11702, + 11709 + ], + "pr_numbers": [ + 11704, + 11726 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 11704, + "right_pr_number": 11726, + "code_similarity": 0.545, + "size_similarity": 0.711, + "file_overlap": 1.0, + "area_overlap": 0.215, + "patch_similarity": 0.183, + "shared_filenames": [ + "src/diffusers/loaders/lora_conversion_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders/lora_conversion_utils.py", + "left_ranges": [ + [ + 1605, + 1622 + ], + [ + 1664, + 1679 + ], + [ + 1682, + 1697 + ], + [ + 1726, + 1745 + ] + ], + "right_ranges": [ + [ + 1596, + 1605 + ], + [ + 1625, + 1681 + ], + [ + 1689, + 1698 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6866-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #6866.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 6866, + "canonical_pr_number": 6941, + "issue_numbers": [ + 6866 + ], + "pr_numbers": [ + 6941, + 7030, + 7031 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6941, + "right_pr_number": 7030, + "code_similarity": 0.345, + "size_similarity": 0.139, + "file_overlap": 1.0, + "area_overlap": 0.039, + "patch_similarity": 0.027, + "shared_filenames": [ + "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", + "left_ranges": [ + [ + 19, + 35 + ], + [ + 201, + 208 + ], + [ + 218, + 225 + ], + [ + 507, + 572 + ], + [ + 636, + 643 + ], + [ + 824, + 834 + ], + [ + 1177, + 1184 + ], + [ + 1273, + 1282 + ], + [ + 1409, + 1416 + ], + [ + 1463, + 1474 + ], + [ + 1740, + 1748 + ] + ], + "right_ranges": [ + [ + 28, + 34 + ], + [ + 146, + 152 + ], + [ + 158, + 164 + ] + ] + } + ] + }, + { + "left_pr_number": 6941, + "right_pr_number": 7031, + "code_similarity": 0.356, + "size_similarity": 0.185, + "file_overlap": 1.0, + "area_overlap": 0.044, + "patch_similarity": 0.027, + "shared_filenames": [ + "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", + "left_ranges": [ + [ + 19, + 35 + ], + [ + 201, + 208 + ], + [ + 218, + 225 + ], + [ + 507, + 572 + ], + [ + 636, + 643 + ], + [ + 824, + 834 + ], + [ + 1177, + 1184 + ], + [ + 1273, + 1282 + ], + [ + 1409, + 1416 + ], + [ + 1463, + 1474 + ], + [ + 1740, + 1748 + ] + ], + "right_ranges": [ + [ + 28, + 39 + ], + [ + 151, + 157 + ], + [ + 163, + 169 + ] + ] + } + ] + }, + { + "left_pr_number": 7030, + "right_pr_number": 7031, + "code_similarity": 0.707, + "size_similarity": 0.75, + "file_overlap": 1.0, + "area_overlap": 0.306, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py", + "left_ranges": [ + [ + 28, + 34 + ], + [ + 146, + 152 + ], + [ + 158, + 164 + ] + ], + "right_ranges": [ + [ + 28, + 39 + ], + [ + 151, + 157 + ], + [ + 163, + 169 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5262-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #5262.", + "status": "merged", + "confidence": 0.75, + "canonical_issue_number": 5262, + "canonical_pr_number": 5370, + "issue_numbers": [ + 5262 + ], + "pr_numbers": [ + 5288, + 5366, + 5370 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 5288, + "right_pr_number": 5366, + "code_similarity": 0.344, + "size_similarity": 0.222, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [ + "tests/pipelines/stable_diffusion/test_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", + "left_ranges": [ + [ + 90, + 97 + ] + ], + "right_ranges": [ + [ + 103, + 129 + ], + [ + 134, + 170 + ] + ] + } + ] + }, + { + "left_pr_number": 5288, + "right_pr_number": 5370, + "code_similarity": 0.356, + "size_similarity": 0.279, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [ + "tests/pipelines/stable_diffusion/test_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", + "left_ranges": [ + [ + 90, + 97 + ] + ], + "right_ranges": [ + [ + 82, + 88 + ], + [ + 104, + 118 + ], + [ + 123, + 145 + ], + [ + 186, + 192 + ], + [ + 320, + 326 + ], + [ + 338, + 344 + ], + [ + 378, + 384 + ], + [ + 397, + 403 + ], + [ + 416, + 422 + ], + [ + 488, + 494 + ], + [ + 641, + 647 + ], + [ + 668, + 674 + ] + ] + } + ] + }, + { + "left_pr_number": 5366, + "right_pr_number": 5370, + "code_similarity": 0.559, + "size_similarity": 0.796, + "file_overlap": 1.0, + "area_overlap": 0.246, + "patch_similarity": 0.088, + "shared_filenames": [ + "tests/pipelines/stable_diffusion/test_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", + "left_ranges": [ + [ + 103, + 129 + ], + [ + 134, + 170 + ] + ], + "right_ranges": [ + [ + 82, + 88 + ], + [ + 104, + 118 + ], + [ + 123, + 145 + ], + [ + 186, + 192 + ], + [ + 320, + 326 + ], + [ + 338, + 344 + ], + [ + 378, + 384 + ], + [ + 397, + 403 + ], + [ + 416, + 422 + ], + [ + 488, + 494 + ], + [ + 641, + 647 + ], + [ + 668, + 674 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-916-4", + "summary": "Cluster of 4 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 1110, + "issue_numbers": [], + "pr_numbers": [ + 916, + 1007, + 1110, + 1225 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 916, + "right_pr_number": 1007, + "code_similarity": 0.334, + "size_similarity": 0.238, + "file_overlap": 0.833, + "area_overlap": 0.065, + "patch_similarity": 0.091, + "shared_filenames": [ + "examples/unconditional_image_generation/train_unconditional.py", + "src/diffusers/models/unet_2d.py", + "src/diffusers/pipeline_utils.py", + "src/diffusers/pipelines/ddpm/pipeline_ddpm.py", + "tests/test_models_unet.py" + ], + "shared_file_areas": [ + { + "filename": "examples/unconditional_image_generation/train_unconditional.py", + "left_ranges": [ + [ + 138, + 144 + ] + ], + "right_ranges": [ + [ + 59, + 69 + ], + [ + 144, + 150 + ], + [ + 231, + 237 + ] + ] + }, + { + "filename": "src/diffusers/models/unet_2d.py", + "left_ranges": [ + [ + 258, + 261 + ] + ], + "right_ranges": [ + [ + 245, + 248 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 72, + 80 + ] + ], + "right_ranges": [ + [ + 53, + 61 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/ddpm/pipeline_ddpm.py", + "left_ranges": [ + [ + 77, + 83 + ] + ], + "right_ranges": [ + [ + 90, + 96 + ] + ] + }, + { + "filename": "tests/test_models_unet.py", + "left_ranges": [ + [ + 131, + 137 + ], + [ + 141, + 147 + ], + [ + 219, + 225 + ], + [ + 294, + 300 + ], + [ + 412, + 418 + ], + [ + 437, + 443 + ] + ], + "right_ranges": [ + [ + 77, + 83 + ], + [ + 129, + 135 + ], + [ + 149, + 155 + ], + [ + 202, + 208 + ], + [ + 216, + 222 + ], + [ + 338, + 344 + ], + [ + 363, + 369 + ] + ] + } + ] + }, + { + "left_pr_number": 916, + "right_pr_number": 1110, + "code_similarity": 0.734, + "size_similarity": 0.919, + "file_overlap": 0.286, + "area_overlap": 0.955, + "patch_similarity": 0.872, + "shared_filenames": [ + "examples/unconditional_image_generation/train_unconditional_ort.py", + "src/diffusers/pipeline_utils.py" + ], + "shared_file_areas": [ + { + "filename": "examples/unconditional_image_generation/train_unconditional_ort.py", + "left_ranges": [ + [ + 1, + 251 + ] + ], + "right_ranges": [ + [ + 1, + 251 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 72, + 80 + ] + ], + "right_ranges": [ + [ + 78, + 86 + ] + ] + } + ] + }, + { + "left_pr_number": 916, + "right_pr_number": 1225, + "code_similarity": 0.137, + "size_similarity": 0.102, + "file_overlap": 0.333, + "area_overlap": 0.019, + "patch_similarity": 0.063, + "shared_filenames": [ + "examples/unconditional_image_generation/train_unconditional.py", + "src/diffusers/pipeline_utils.py" + ], + "shared_file_areas": [ + { + "filename": "examples/unconditional_image_generation/train_unconditional.py", + "left_ranges": [ + [ + 138, + 144 + ] + ], + "right_ranges": [ + [ + 59, + 69 + ], + [ + 144, + 153 + ], + [ + 245, + 251 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 72, + 80 + ] + ], + "right_ranges": [ + [ + 53, + 61 + ] + ] + } + ] + }, + { + "left_pr_number": 1007, + "right_pr_number": 1110, + "code_similarity": 0.103, + "size_similarity": 0.259, + "file_overlap": 0.143, + "area_overlap": 0.0, + "patch_similarity": 0.057, + "shared_filenames": [ + "src/diffusers/pipeline_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 53, + 61 + ] + ], + "right_ranges": [ + [ + 78, + 86 + ] + ] + } + ] + }, + { + "left_pr_number": 1007, + "right_pr_number": 1225, + "code_similarity": 0.494, + "size_similarity": 0.43, + "file_overlap": 0.4, + "area_overlap": 0.614, + "patch_similarity": 0.486, + "shared_filenames": [ + "examples/unconditional_image_generation/train_unconditional.py", + "src/diffusers/pipeline_utils.py" + ], + "shared_file_areas": [ + { + "filename": "examples/unconditional_image_generation/train_unconditional.py", + "left_ranges": [ + [ + 59, + 69 + ], + [ + 144, + 150 + ], + [ + 231, + 237 + ] + ], + "right_ranges": [ + [ + 59, + 69 + ], + [ + 144, + 153 + ], + [ + 245, + 251 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 53, + 61 + ] + ], + "right_ranges": [ + [ + 53, + 61 + ] + ] + } + ] + }, + { + "left_pr_number": 1110, + "right_pr_number": 1225, + "code_similarity": 0.106, + "size_similarity": 0.111, + "file_overlap": 0.25, + "area_overlap": 0.0, + "patch_similarity": 0.059, + "shared_filenames": [ + "src/diffusers/pipeline_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 78, + 86 + ] + ], + "right_ranges": [ + [ + 53, + 61 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5062-4", + "summary": "Cluster of 1 issues and 3 PRs centered on issue #5062.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 5062, + "canonical_pr_number": 5066, + "issue_numbers": [ + 5062 + ], + "pr_numbers": [ + 5065, + 5066, + 5332 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5065, + "right_pr_number": 5066, + "code_similarity": 0.257, + "size_similarity": 0.089, + "file_overlap": 0.15, + "area_overlap": 0.543, + "patch_similarity": 0.028, + "shared_filenames": [ + "examples/community/run_onnx_controlnet.py", + "examples/community/run_tensorrt_controlnet.py", + "src/diffusers/models/attention_processor.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/run_onnx_controlnet.py", + "left_ranges": [ + [ + 8, + 14 + ] + ], + "right_ranges": [ + [ + 8, + 13 + ], + [ + 21, + 27 + ] + ] + }, + { + "filename": "examples/community/run_tensorrt_controlnet.py", + "left_ranges": [ + [ + 11, + 17 + ] + ], + "right_ranges": [ + [ + 11, + 16 + ], + [ + 25, + 31 + ] + ] + }, + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 382, + 388 + ] + ], + "right_ranges": [ + [ + 382, + 388 + ] + ] + } + ] + }, + { + "left_pr_number": 5065, + "right_pr_number": 5332, + "code_similarity": 0.003, + "size_similarity": 0.015, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.003, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5066, + "right_pr_number": 5332, + "code_similarity": 0.033, + "size_similarity": 0.167, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-6123-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6123.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 6123, + "canonical_pr_number": 6328, + "issue_numbers": [ + 6123 + ], + "pr_numbers": [ + 6328, + 6509 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 6328, + "right_pr_number": 6509, + "code_similarity": 0.241, + "size_similarity": 0.686, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.693, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-7736-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #7736.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 7736, + "canonical_pr_number": 7761, + "issue_numbers": [ + 7736 + ], + "pr_numbers": [ + 7739, + 7761 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 7739, + "right_pr_number": 7761, + "code_similarity": 0.155, + "size_similarity": 0.462, + "file_overlap": 0.05, + "area_overlap": 0.071, + "patch_similarity": 0.15, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", + "left_ranges": [ + [ + 19, + 25 + ], + [ + 122, + 128 + ], + [ + 254, + 261 + ], + [ + 969, + 984 + ], + [ + 1023, + 1032 + ] + ], + "right_ranges": [ + [ + 11, + 24 + ], + [ + 775, + 783 + ], + [ + 847, + 857 + ], + [ + 883, + 891 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6158-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6158.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 6158, + "canonical_pr_number": 6721, + "issue_numbers": [ + 6158 + ], + "pr_numbers": [ + 6195, + 6721 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6195, + "right_pr_number": 6721, + "code_similarity": 0.215, + "size_similarity": 0.615, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.61, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-6299-4", + "summary": "Cluster of 4 related pull requests linked by soft_similarity.", + "status": "closed", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 6322, + "issue_numbers": [], + "pr_numbers": [ + 6299, + 6300, + 6301, + 6322 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6299, + "right_pr_number": 6300, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/text_to_image/train_text_to_image.py", + "left_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ], + "right_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "left_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ], + "right_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ] + } + ] + }, + { + "left_pr_number": 6299, + "right_pr_number": 6301, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/text_to_image/train_text_to_image.py", + "left_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ], + "right_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "left_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ], + "right_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ] + } + ] + }, + { + "left_pr_number": 6299, + "right_pr_number": 6322, + "code_similarity": 0.716, + "size_similarity": 0.748, + "file_overlap": 1.0, + "area_overlap": 0.422, + "patch_similarity": 0.791, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/text_to_image/train_text_to_image.py", + "left_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ], + "right_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 520 + ], + [ + 580, + 591 + ], + [ + 597, + 613 + ], + [ + 634, + 658 + ], + [ + 735, + 741 + ], + [ + 1018, + 1024 + ], + [ + 1033, + 1040 + ], + [ + 1055, + 1090 + ], + [ + 1093, + 1100 + ], + [ + 1107, + 1120 + ], + [ + 1144, + 1155 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "left_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ], + "right_ranges": [ + [ + 34, + 41 + ], + [ + 54, + 60 + ], + [ + 172, + 178 + ], + [ + 181, + 192 + ], + [ + 478, + 485 + ], + [ + 586, + 599 + ], + [ + 801, + 812 + ], + [ + 882, + 940 + ] + ] + } + ] + }, + { + "left_pr_number": 6300, + "right_pr_number": 6301, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/text_to_image/train_text_to_image.py", + "left_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ], + "right_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "left_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ], + "right_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ] + } + ] + }, + { + "left_pr_number": 6300, + "right_pr_number": 6322, + "code_similarity": 0.716, + "size_similarity": 0.748, + "file_overlap": 1.0, + "area_overlap": 0.422, + "patch_similarity": 0.791, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/text_to_image/train_text_to_image.py", + "left_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ], + "right_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 520 + ], + [ + 580, + 591 + ], + [ + 597, + 613 + ], + [ + 634, + 658 + ], + [ + 735, + 741 + ], + [ + 1018, + 1024 + ], + [ + 1033, + 1040 + ], + [ + 1055, + 1090 + ], + [ + 1093, + 1100 + ], + [ + 1107, + 1120 + ], + [ + 1144, + 1155 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "left_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ], + "right_ranges": [ + [ + 34, + 41 + ], + [ + 54, + 60 + ], + [ + 172, + 178 + ], + [ + 181, + 192 + ], + [ + 478, + 485 + ], + [ + 586, + 599 + ], + [ + 801, + 812 + ], + [ + 882, + 940 + ] + ] + } + ] + }, + { + "left_pr_number": 6301, + "right_pr_number": 6322, + "code_similarity": 0.716, + "size_similarity": 0.748, + "file_overlap": 1.0, + "area_overlap": 0.422, + "patch_similarity": 0.791, + "shared_filenames": [ + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/text_to_image/train_text_to_image.py", + "left_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 515 + ], + [ + 575, + 586 + ], + [ + 592, + 608 + ], + [ + 1014, + 1021 + ], + [ + 1036, + 1071 + ], + [ + 1125, + 1136 + ] + ], + "right_ranges": [ + [ + 41, + 47 + ], + [ + 152, + 158 + ], + [ + 166, + 179 + ], + [ + 485, + 520 + ], + [ + 580, + 591 + ], + [ + 597, + 613 + ], + [ + 634, + 658 + ], + [ + 735, + 741 + ], + [ + 1018, + 1024 + ], + [ + 1033, + 1040 + ], + [ + 1055, + 1090 + ], + [ + 1093, + 1100 + ], + [ + 1107, + 1120 + ], + [ + 1144, + 1155 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "left_ranges": [ + [ + 34, + 40 + ], + [ + 54, + 60 + ], + [ + 580, + 593 + ], + [ + 874, + 920 + ] + ], + "right_ranges": [ + [ + 34, + 41 + ], + [ + 54, + 60 + ], + [ + 172, + 178 + ], + [ + 181, + 192 + ], + [ + 478, + 485 + ], + [ + 586, + 599 + ], + [ + 801, + 812 + ], + [ + 882, + 940 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-3896-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 4097, + "issue_numbers": [], + "pr_numbers": [ + 3896, + 4016, + 4097 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 3896, + "right_pr_number": 4016, + "code_similarity": 0.196, + "size_similarity": 0.42, + "file_overlap": 0.095, + "area_overlap": 0.008, + "patch_similarity": 0.535, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", + "left_ranges": [ + [ + 1, + 805 + ] + ], + "right_ranges": [ + [ + 73, + 79 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py", + "left_ranges": [ + [ + 1, + 869 + ] + ], + "right_ranges": [ + [ + 78, + 84 + ] + ] + } + ] + }, + { + "left_pr_number": 3896, + "right_pr_number": 4097, + "code_similarity": 0.096, + "size_similarity": 0.214, + "file_overlap": 0.05, + "area_overlap": 0.008, + "patch_similarity": 0.237, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", + "left_ranges": [ + [ + 1, + 805 + ] + ], + "right_ranges": [ + [ + 13, + 19 + ], + [ + 810, + 875 + ] + ] + } + ] + }, + { + "left_pr_number": 4016, + "right_pr_number": 4097, + "code_similarity": 0.366, + "size_similarity": 0.509, + "file_overlap": 0.4, + "area_overlap": 0.299, + "patch_similarity": 0.267, + "shared_filenames": [ + "examples/dreambooth/README_sdxl.md", + "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "examples/test_examples.py", + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/README_sdxl.md", + "left_ranges": [ + [ + 1, + 178 + ] + ], + "right_ranges": [ + [ + 164, + 180 + ] + ] + }, + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 1, + 1266 + ] + ], + "right_ranges": [ + [ + 16, + 22 + ], + [ + 46, + 56 + ], + [ + 64, + 70 + ], + [ + 92, + 99 + ], + [ + 128, + 139 + ], + [ + 424, + 455 + ], + [ + 461, + 466 + ], + [ + 485, + 513 + ], + [ + 528, + 556 + ], + [ + 602, + 607 + ], + [ + 635, + 641 + ], + [ + 700, + 713 + ], + [ + 729, + 738 + ], + [ + 772, + 837 + ], + [ + 861, + 871 + ], + [ + 874, + 954 + ], + [ + 969, + 989 + ], + [ + 1042, + 1050 + ], + [ + 1053, + 1068 + ], + [ + 1078, + 1107 + ], + [ + 1129, + 1139 + ], + [ + 1184, + 1205 + ], + [ + 1227, + 1237 + ], + [ + 1257, + 1288 + ], + [ + 1334, + 1340 + ] + ] + }, + { + "filename": "examples/test_examples.py", + "left_ranges": [ + [ + 353, + 390 + ] + ], + "right_ranges": [ + [ + 385, + 426 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", + "left_ranges": [ + [ + 73, + 79 + ] + ], + "right_ranges": [ + [ + 13, + 19 + ], + [ + 810, + 875 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9403-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #9403.", + "status": "merged", + "confidence": 0.75, + "canonical_issue_number": 9403, + "canonical_pr_number": 10261, + "issue_numbers": [ + 9403 + ], + "pr_numbers": [ + 9787, + 10261 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 9787, + "right_pr_number": 10261, + "code_similarity": 0.589, + "size_similarity": 0.715, + "file_overlap": 0.583, + "area_overlap": 0.468, + "patch_similarity": 0.715, + "shared_filenames": [ + "scripts/convert_flux_xlabs_ipadapter_to_diffusers.py", + "src/diffusers/loaders/__init__.py", + "src/diffusers/loaders/ip_adapter.py", + "src/diffusers/loaders/transformer_flux.py", + "src/diffusers/models/attention_processor.py", + "src/diffusers/models/transformers/transformer_flux.py", + "src/diffusers/pipelines/flux/pipeline_flux.py" + ], + "shared_file_areas": [ + { + "filename": "scripts/convert_flux_xlabs_ipadapter_to_diffusers.py", + "left_ranges": [ + [ + 1, + 97 + ] + ], + "right_ranges": [ + [ + 1, + 97 + ] + ] + }, + { + "filename": "src/diffusers/loaders/__init__.py", + "left_ranges": [ + [ + 55, + 61 + ], + [ + 70, + 89 + ] + ], + "right_ranges": [ + [ + 55, + 61 + ], + [ + 77, + 83 + ], + [ + 87, + 100 + ] + ] + }, + { + "filename": "src/diffusers/loaders/ip_adapter.py", + "left_ranges": [ + [ + 41, + 48 + ], + [ + 348, + 611 + ] + ], + "right_ranges": [ + [ + 38, + 45 + ], + [ + 355, + 644 + ] + ] + }, + { + "filename": "src/diffusers/loaders/transformer_flux.py", + "left_ranges": [ + [ + 1, + 174 + ] + ], + "right_ranges": [ + [ + 1, + 179 + ] + ] + }, + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 482, + 488 + ], + [ + 1987, + 2215 + ], + [ + 4614, + 4620 + ] + ], + "right_ranges": [ + [ + 575, + 581 + ], + [ + 2653, + 2801 + ], + [ + 6039, + 6045 + ] + ] + }, + { + "filename": "src/diffusers/models/transformers/transformer_flux.py", + "left_ranges": [ + [ + 21, + 27 + ], + [ + 207, + 215 + ] + ], + "right_ranges": [ + [ + 21, + 27 + ], + [ + 177, + 194 + ], + [ + 200, + 207 + ], + [ + 219, + 227 + ], + [ + 491, + 501 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/flux/pipeline_flux.py", + "left_ranges": [ + [ + 17, + 33 + ], + [ + 149, + 155 + ], + [ + 190, + 197 + ], + [ + 203, + 210 + ], + [ + 387, + 436 + ], + [ + 611, + 618 + ], + [ + 662, + 672 + ], + [ + 783, + 799 + ] + ], + "right_ranges": [ + [ + 17, + 33 + ], + [ + 149, + 155 + ], + [ + 177, + 184 + ], + [ + 190, + 197 + ], + [ + 203, + 210 + ], + [ + 389, + 448 + ], + [ + 477, + 509 + ], + [ + 632, + 640 + ], + [ + 645, + 656 + ], + [ + 700, + 716 + ], + [ + 748, + 759 + ], + [ + 775, + 781 + ], + [ + 790, + 810 + ], + [ + 846, + 888 + ], + [ + 898, + 919 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9772-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #9772.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 9772, + "canonical_pr_number": 10131, + "issue_numbers": [ + 9772 + ], + "pr_numbers": [ + 9776, + 10131 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9776, + "right_pr_number": 10131, + "code_similarity": 0.789, + "size_similarity": 0.969, + "file_overlap": 0.444, + "area_overlap": 0.947, + "patch_similarity": 0.87, + "shared_filenames": [ + "docs/source/en/_toctree.yml", + "src/diffusers/__init__.py", + "src/diffusers/models/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/controlnet/__init__.py", + "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py", + "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py", + "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/_toctree.yml", + "left_ranges": [ + [ + 217, + 224 + ] + ], + "right_ranges": [ + [ + 252, + 259 + ], + [ + 368, + 375 + ] + ] + }, + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ + [ + 88, + 94 + ], + [ + 364, + 372 + ], + [ + 571, + 577 + ], + [ + 825, + 833 + ] + ], + "right_ranges": [ + [ + 91, + 97 + ], + [ + 377, + 385 + ], + [ + 587, + 593 + ], + [ + 851, + 859 + ] + ] + }, + { + "filename": "src/diffusers/models/__init__.py", + "left_ranges": [ + [ + 39, + 45 + ], + [ + 94, + 100 + ] + ], + "right_ranges": [ + [ + 44, + 50 + ], + [ + 101, + 107 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/__init__.py", + "left_ranges": [ + [ + 156, + 164 + ], + [ + 489, + 497 + ] + ], + "right_ranges": [ + [ + 162, + 170 + ], + [ + 498, + 506 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/controlnet/__init__.py", + "left_ranges": [ + [ + 30, + 38 + ], + [ + 60, + 68 + ] + ], + "right_ranges": [ + [ + 1, + 86 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py", + "left_ranges": [ + [ + 1, + 1824 + ] + ], + "right_ranges": [ + [ + 1, + 1801 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py", + "left_ranges": [ + [ + 1, + 1500 + ] + ], + "right_ranges": [ + [ + 1, + 1531 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py", + "left_ranges": [ + [ + 1, + 1624 + ] + ], + "right_ranges": [ + [ + 1, + 1646 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6688-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6688.", + "status": "closed", + "confidence": 0.55, + "canonical_issue_number": 6688, + "canonical_pr_number": 7005, + "issue_numbers": [ + 6688 + ], + "pr_numbers": [ + 6844, + 7005 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 6844, + "right_pr_number": 7005, + "code_similarity": 0.471, + "size_similarity": 0.968, + "file_overlap": 0.182, + "area_overlap": 0.304, + "patch_similarity": 0.778, + "shared_filenames": [ + "src/diffusers/models/attention.py", + "src/diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/attention.py", + "left_ranges": [ + [ + 431, + 437 + ], + [ + 475, + 488 + ], + [ + 498, + 510 + ], + [ + 531, + 544 + ] + ], + "right_ranges": [ + [ + 425, + 431 + ] + ] + }, + { + "filename": "src/diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py", + "left_ranges": [ + [ + 205, + 211 + ], + [ + 227, + 233 + ], + [ + 331, + 341 + ] + ], + "right_ranges": [ + [ + 205, + 211 + ], + [ + 227, + 233 + ], + [ + 331, + 339 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-8710-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #8710.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 8710, + "canonical_pr_number": 9386, + "issue_numbers": [ + 8710 + ], + "pr_numbers": [ + 8725, + 9386 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 8725, + "right_pr_number": 9386, + "code_similarity": 0.463, + "size_similarity": 0.85, + "file_overlap": 0.545, + "area_overlap": 0.029, + "patch_similarity": 0.797, + "shared_filenames": [ + "docs/source/en/api/pipelines/pag.md", + "src/diffusers/__init__.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/pipelines/pag/__init__.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/pag.md", + "left_ranges": [ + [ + 20, + 30 + ] + ], + "right_ranges": [ + [ + 48, + 58 + ] + ] + }, + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ + [ + 304, + 310 + ], + [ + 703, + 709 + ] + ], + "right_ranges": [ + [ + 363, + 369 + ], + [ + 835, + 841 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/__init__.py", + "left_ranges": [ + [ + 141, + 147 + ], + [ + 492, + 498 + ] + ], + "right_ranges": [ + [ + 174, + 180 + ], + [ + 596, + 602 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/auto_pipeline.py", + "left_ranges": [ + [ + 47, + 53 + ], + [ + 89, + 95 + ] + ], + "right_ranges": [ + [ + 66, + 72 + ], + [ + 161, + 167 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/pag/__init__.py", + "left_ranges": [ + [ + 23, + 29 + ], + [ + 37, + 43 + ] + ], + "right_ranges": [ + [ + 34, + 41 + ], + [ + 60, + 66 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "left_ranges": [ + [ + 1232, + 1252 + ] + ], + "right_ranges": [ + [ + 1757, + 1777 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6891-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6891.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 6891, + "canonical_pr_number": 7615, + "issue_numbers": [ + 6891 + ], + "pr_numbers": [ + 7613, + 7615 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 7613, + "right_pr_number": 7615, + "code_similarity": 0.319, + "size_similarity": 0.882, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.948, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-7677-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #7677.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 7677, + "canonical_pr_number": 7865, + "issue_numbers": [ + 7677 + ], + "pr_numbers": [ + 7865, + 7898 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 7865, + "right_pr_number": 7898, + "code_similarity": 0.159, + "size_similarity": 0.682, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.15, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-12441-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12441.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 12441, + "canonical_pr_number": 12526, + "issue_numbers": [ + 12441 + ], + "pr_numbers": [ + 12442, + 12526 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 12442, + "right_pr_number": 12526, + "code_similarity": 0.687, + "size_similarity": 0.728, + "file_overlap": 0.684, + "area_overlap": 0.669, + "patch_similarity": 0.68, + "shared_filenames": [ + "docs/source/en/api/pipelines/wan.md", + "scripts/convert_wan_to_diffusers.py", + "src/diffusers/__init__.py", + "src/diffusers/models/__init__.py", + "src/diffusers/models/transformers/__init__.py", + "src/diffusers/models/transformers/transformer_wan_animate.py", + "src/diffusers/pipelines/__init__.py", + "src/diffusers/pipelines/wan/__init__.py", + "src/diffusers/pipelines/wan/pipeline_wan_animate.py", + "src/diffusers/utils/dummy_pt_objects.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "tests/pipelines/wan/test_wan_animate.py", + "tests/quantization/gguf/test_gguf.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/wan.md", + "left_ranges": [ + [ + 40, + 46 + ], + [ + 96, + 110 + ], + [ + 151, + 165 + ], + [ + 250, + 469 + ], + [ + 496, + 505 + ], + [ + 574, + 585 + ] + ], + "right_ranges": [ + [ + 40, + 46 + ], + [ + 96, + 110 + ], + [ + 151, + 165 + ], + [ + 250, + 469 + ], + [ + 496, + 505 + ], + [ + 574, + 585 + ] + ] + }, + { + "filename": "scripts/convert_wan_to_diffusers.py", + "left_ranges": [ + [ + 1, + 27 + ], + [ + 116, + 308 + ], + [ + 560, + 590 + ], + [ + 601, + 612 + ], + [ + 620, + 643 + ], + [ + 1148, + 1311 + ], + [ + 1324, + 1330 + ], + [ + 1340, + 1346 + ], + [ + 1352, + 1359 + ], + [ + 1416, + 1449 + ] + ], + "right_ranges": [ + [ + 6, + 25 + ], + [ + 114, + 316 + ], + [ + 568, + 604 + ], + [ + 615, + 626 + ], + [ + 634, + 645 + ], + [ + 1168, + 1174 + ], + [ + 1184, + 1190 + ], + [ + 1196, + 1203 + ], + [ + 1260, + 1280 + ] + ] + }, + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ + [ + 259, + 265 + ], + [ + 621, + 627 + ], + [ + 954, + 960 + ], + [ + 1286, + 1292 + ] + ], + "right_ranges": [ + [ + 268, + 274 + ], + [ + 637, + 643 + ], + [ + 979, + 985 + ], + [ + 1318, + 1324 + ] + ] + }, + { + "filename": "src/diffusers/models/__init__.py", + "left_ranges": [ + [ + 101, + 107 + ], + [ + 201, + 207 + ] + ], + "right_ranges": [ + [ + 108, + 114 + ], + [ + 215, + 221 + ] + ] + }, + { + "filename": "src/diffusers/models/transformers/__init__.py", + "left_ranges": [ + [ + 37, + 41 + ] + ], + "right_ranges": [ + [ + 42, + 46 + ] + ] + }, + { + "filename": "src/diffusers/models/transformers/transformer_wan_animate.py", + "left_ranges": [ + [ + 1, + 611 + ] + ], + "right_ranges": [ + [ + 1, + 1298 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/__init__.py", + "left_ranges": [ + [ + 381, + 393 + ], + [ + 794, + 806 + ] + ], + "right_ranges": [ + [ + 385, + 397 + ], + [ + 809, + 821 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/wan/__init__.py", + "left_ranges": [ + [ + 23, + 29 + ], + [ + 36, + 45 + ] + ], + "right_ranges": [ + [ + 23, + 29 + ], + [ + 36, + 45 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/wan/pipeline_wan_animate.py", + "left_ranges": [ + [ + 1, + 1072 + ] + ], + "right_ranges": [ + [ + 1, + 1204 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_pt_objects.py", + "left_ranges": [ + [ + 1488, + 1508 + ] + ], + "right_ranges": [ + [ + 1623, + 1643 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "left_ranges": [ + [ + 3407, + 3427 + ] + ], + "right_ranges": [ + [ + 3512, + 3532 + ] + ] + }, + { + "filename": "tests/pipelines/wan/test_wan_animate.py", + "left_ranges": [ + [ + 1, + 245 + ] + ], + "right_ranges": [ + [ + 1, + 239 + ] + ] + }, + { + "filename": "tests/quantization/gguf/test_gguf.py", + "left_ranges": [ + [ + 16, + 22 + ], + [ + 722, + 754 + ] + ], + "right_ranges": [ + [ + 16, + 22 + ], + [ + 722, + 754 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5886-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5886.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 5886, + "canonical_pr_number": 5911, + "issue_numbers": [ + 5886 + ], + "pr_numbers": [ + 5910, + 5911 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5910, + "right_pr_number": 5911, + "code_similarity": 0.294, + "size_similarity": 0.411, + "file_overlap": 0.333, + "area_overlap": 0.26, + "patch_similarity": 0.134, + "shared_filenames": [ + "src/diffusers/loaders/unet.py", + "src/diffusers/models/embeddings.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders/unet.py", + "left_ranges": [ + [ + 21, + 27 + ], + [ + 664, + 670 + ], + [ + 711, + 726 + ] + ], + "right_ranges": [ + [ + 21, + 27 + ], + [ + 675, + 683 + ], + [ + 747, + 778 + ] + ] + }, + { + "filename": "src/diffusers/models/embeddings.py", + "left_ranges": [ + [ + 458, + 482 + ] + ], + "right_ranges": [ + [ + 461, + 478 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-3880-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #3880.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 3880, + "canonical_pr_number": 4536, + "issue_numbers": [ + 3880 + ], + "pr_numbers": [ + 3921, + 4536 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 3921, + "right_pr_number": 4536, + "code_similarity": 0.061, + "size_similarity": 0.229, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.1, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-4899-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5800.", + "status": "closed", + "confidence": 0.6, + "canonical_issue_number": 5800, + "canonical_pr_number": 5938, + "issue_numbers": [ + 5800 + ], + "pr_numbers": [ + 4899, + 5938 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 4899, + "right_pr_number": 5938, + "code_similarity": 0.472, + "size_similarity": 0.741, + "file_overlap": 0.375, + "area_overlap": 0.354, + "patch_similarity": 0.586, + "shared_filenames": [ + "src/diffusers/models/controlnet.py", + "src/diffusers/models/embeddings.py", + "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/controlnet.py", + "left_ranges": [ + [ + 19, + 26 + ], + [ + 81, + 87 + ], + [ + 97, + 103 + ], + [ + 109, + 117 + ], + [ + 251, + 257 + ], + [ + 723, + 738 + ], + [ + 771, + 777 + ] + ], + "right_ranges": [ + [ + 20, + 27 + ], + [ + 82, + 88 + ], + [ + 108, + 116 + ], + [ + 253, + 259 + ] + ] + }, + { + "filename": "src/diffusers/models/embeddings.py", + "left_ranges": [ + [ + 18, + 24 + ], + [ + 167, + 176 + ], + [ + 180, + 186 + ] + ], + "right_ranges": [ + [ + 18, + 24 + ], + [ + 201, + 210 + ], + [ + 214, + 220 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", + "left_ranges": [ + [ + 377, + 397 + ], + [ + 429, + 450 + ], + [ + 458, + 483 + ], + [ + 623, + 631 + ], + [ + 635, + 647 + ], + [ + 652, + 672 + ] + ], + "right_ranges": [ + [ + 304, + 311 + ], + [ + 323, + 328 + ], + [ + 377, + 397 + ], + [ + 429, + 450 + ], + [ + 458, + 483 + ], + [ + 531, + 536 + ], + [ + 623, + 631 + ], + [ + 635, + 647 + ], + [ + 652, + 672 + ], + [ + 830, + 836 + ], + [ + 965, + 971 + ], + [ + 1178, + 1183 + ], + [ + 1262, + 1275 + ], + [ + 1308, + 1332 + ], + [ + 1346, + 1353 + ], + [ + 1431, + 1439 + ], + [ + 1482, + 1492 + ], + [ + 1498, + 1505 + ], + [ + 1509, + 1516 + ], + [ + 1541, + 1549 + ], + [ + 1576, + 1585 + ], + [ + 1592, + 1602 + ], + [ + 1632, + 1649 + ], + [ + 1663, + 1668 + ], + [ + 1677, + 1690 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12004-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 12006, + "issue_numbers": [], + "pr_numbers": [ + 12004, + 12006, + 12024 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 12004, + "right_pr_number": 12006, + "code_similarity": 0.151, + "size_similarity": 0.056, + "file_overlap": 0.222, + "area_overlap": 0.166, + "patch_similarity": 0.103, + "shared_filenames": [ + "src/diffusers/pipelines/skyreels_v2/pipeline_skyreels_v2_i2v.py", + "src/diffusers/pipelines/wan/pipeline_wan_i2v.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/skyreels_v2/pipeline_skyreels_v2_i2v.py", + "left_ranges": [ + [ + 316, + 321 + ] + ], + "right_ranges": [ + [ + 370, + 375 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", + "left_ranges": [ + [ + 149, + 180 + ], + [ + 186, + 194 + ], + [ + 339, + 345 + ], + [ + 383, + 394 + ], + [ + 504, + 510 + ], + [ + 549, + 558 + ], + [ + 615, + 621 + ], + [ + 625, + 635 + ], + [ + 662, + 675 + ], + [ + 700, + 730 + ], + [ + 734, + 748 + ] + ], + "right_ranges": [ + [ + 175, + 181 + ], + [ + 189, + 198 + ], + [ + 420, + 431 + ], + [ + 458, + 470 + ], + [ + 674, + 680 + ], + [ + 694, + 701 + ], + [ + 708, + 717 + ], + [ + 738, + 754 + ], + [ + 792, + 800 + ] + ] + } + ] + }, + { + "left_pr_number": 12004, + "right_pr_number": 12024, + "code_similarity": 0.387, + "size_similarity": 0.855, + "file_overlap": 0.455, + "area_overlap": 0.157, + "patch_similarity": 0.164, + "shared_filenames": [ + "src/diffusers/models/transformers/transformer_wan.py", + "src/diffusers/pipelines/wan/pipeline_wan.py", + "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", + "tests/pipelines/wan/test_wan.py", + "tests/pipelines/wan/test_wan_image_to_video.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/transformer_wan.py", + "left_ranges": [ + [ + 170, + 180 + ], + [ + 312, + 334 + ], + [ + 486, + 507 + ], + [ + 517, + 530 + ] + ], + "right_ranges": [ + [ + 324, + 330 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/wan/pipeline_wan.py", + "left_ranges": [ + [ + 112, + 131 + ], + [ + 134, + 142 + ], + [ + 146, + 157 + ], + [ + 285, + 291 + ], + [ + 318, + 326 + ], + [ + 388, + 394 + ], + [ + 427, + 436 + ], + [ + 485, + 491 + ], + [ + 495, + 505 + ], + [ + 549, + 592 + ], + [ + 595, + 609 + ] + ], + "right_ranges": [ + [ + 125, + 139 + ], + [ + 526, + 532 + ], + [ + 536, + 546 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", + "left_ranges": [ + [ + 149, + 180 + ], + [ + 186, + 194 + ], + [ + 339, + 345 + ], + [ + 383, + 394 + ], + [ + 504, + 510 + ], + [ + 549, + 558 + ], + [ + 615, + 621 + ], + [ + 625, + 635 + ], + [ + 662, + 675 + ], + [ + 700, + 730 + ], + [ + 734, + 748 + ] + ], + "right_ranges": [ + [ + 162, + 178 + ], + [ + 669, + 681 + ], + [ + 710, + 716 + ] + ] + }, + { + "filename": "tests/pipelines/wan/test_wan.py", + "left_ranges": [ + [ + 85, + 113 + ] + ], + "right_ranges": [ + [ + 13, + 22 + ], + [ + 87, + 99 + ], + [ + 141, + 185 + ] + ] + }, + { + "filename": "tests/pipelines/wan/test_wan_image_to_video.py", + "left_ranges": [ + [ + 86, + 108 + ], + [ + 126, + 132 + ], + [ + 182, + 193 + ], + [ + 242, + 265 + ], + [ + 283, + 289 + ], + [ + 340, + 348 + ] + ], + "right_ranges": [ + [ + 12, + 21 + ], + [ + 27, + 33 + ], + [ + 88, + 93 + ], + [ + 111, + 117 + ], + [ + 167, + 210 + ], + [ + 260, + 265 + ], + [ + 283, + 289 + ], + [ + 341, + 381 + ] + ] + } + ] + }, + { + "left_pr_number": 12006, + "right_pr_number": 12024, + "code_similarity": 0.121, + "size_similarity": 0.065, + "file_overlap": 0.125, + "area_overlap": 0.161, + "patch_similarity": 0.093, + "shared_filenames": [ + "src/diffusers/pipelines/wan/pipeline_wan_i2v.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/wan/pipeline_wan_i2v.py", + "left_ranges": [ + [ + 175, + 181 + ], + [ + 189, + 198 + ], + [ + 420, + 431 + ], + [ + 458, + 470 + ], + [ + 674, + 680 + ], + [ + 694, + 701 + ], + [ + 708, + 717 + ], + [ + 738, + 754 + ], + [ + 792, + 800 + ] + ], + "right_ranges": [ + [ + 162, + 178 + ], + [ + 669, + 681 + ], + [ + 710, + 716 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-7038-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #7038.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 7038, + "canonical_pr_number": 7550, + "issue_numbers": [ + 7038 + ], + "pr_numbers": [ + 7442, + 7550 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 7442, + "right_pr_number": 7550, + "code_similarity": 0.083, + "size_similarity": 0.271, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.192, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-13377-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13377.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 13377, + "canonical_pr_number": 13379, + "issue_numbers": [ + 13377 + ], + "pr_numbers": [ + 13379, + 13449 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 13379, + "right_pr_number": 13449, + "code_similarity": 0.212, + "size_similarity": 0.179, + "file_overlap": 0.562, + "area_overlap": 0.002, + "patch_similarity": 0.046, + "shared_filenames": [ + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet_inpaint.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_inpaint.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py", + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_layered.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage.py", + "left_ranges": [ + [ + 311, + 332 + ], + [ + 600, + 606 + ] + ], + "right_ranges": [ + [ + 585, + 591 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet.py", + "left_ranges": [ + [ + 101, + 107 + ], + [ + 239, + 245 + ], + [ + 248, + 254 + ], + [ + 287, + 293 + ], + [ + 318, + 330 + ], + [ + 376, + 397 + ], + [ + 718, + 724 + ] + ], + "right_ranges": [ + [ + 701, + 707 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_controlnet_inpaint.py", + "left_ranges": [ + [ + 74, + 80 + ], + [ + 221, + 227 + ], + [ + 230, + 236 + ], + [ + 247, + 253 + ], + [ + 269, + 275 + ], + [ + 280, + 286 + ], + [ + 300, + 317 + ], + [ + 359, + 377 + ], + [ + 751, + 757 + ] + ], + "right_ranges": [ + [ + 740, + 746 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit.py", + "left_ranges": [ + [ + 247, + 261 + ], + [ + 306, + 318 + ], + [ + 360, + 378 + ], + [ + 714, + 720 + ] + ], + "right_ranges": [ + [ + 706, + 712 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_inpaint.py", + "left_ranges": [ + [ + 258, + 272 + ], + [ + 318, + 330 + ], + [ + 392, + 412 + ], + [ + 895, + 901 + ] + ], + "right_ranges": [ + [ + 879, + 885 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", + "left_ranges": [ + [ + 260, + 274 + ], + [ + 320, + 332 + ], + [ + 375, + 393 + ], + [ + 702, + 708 + ] + ], + "right_ranges": [ + [ + 694, + 700 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py", + "left_ranges": [ + [ + 358, + 379 + ], + [ + 693, + 699 + ] + ], + "right_ranges": [ + [ + 678, + 684 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py", + "left_ranges": [ + [ + 385, + 405 + ], + [ + 837, + 843 + ] + ], + "right_ranges": [ + [ + 823, + 829 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_layered.py", + "left_ranges": [ + [ + 385, + 403 + ], + [ + 704, + 710 + ] + ], + "right_ranges": [ + [ + 698, + 704 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12458-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12458.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 12458, + "canonical_pr_number": 12698, + "issue_numbers": [ + 12458 + ], + "pr_numbers": [ + 12467, + 12698 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 12467, + "right_pr_number": 12698, + "code_similarity": 0.133, + "size_similarity": 0.125, + "file_overlap": 0.067, + "area_overlap": 0.208, + "patch_similarity": 0.105, + "shared_filenames": [ + "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/qwenimage/pipeline_qwenimage_edit_plus.py", + "left_ranges": [ + [ + 252, + 268 + ], + [ + 635, + 646 + ], + [ + 686, + 692 + ], + [ + 695, + 704 + ], + [ + 736, + 760 + ] + ], + "right_ranges": [ + [ + 226, + 250 + ], + [ + 255, + 286 + ], + [ + 312, + 317 + ], + [ + 322, + 380 + ], + [ + 680, + 706 + ], + [ + 739, + 810 + ], + [ + 853, + 868 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5524-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5524.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 5524, + "canonical_pr_number": 5704, + "issue_numbers": [ + 5524 + ], + "pr_numbers": [ + 5696, + 5704 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 5696, + "right_pr_number": 5704, + "code_similarity": 0.183, + "size_similarity": 0.165, + "file_overlap": 0.25, + "area_overlap": 0.194, + "patch_similarity": 0.048, + "shared_filenames": [ + "src/diffusers/pipelines/wuerstchen/modeling_wuerstchen_prior.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/wuerstchen/modeling_wuerstchen_prior.py", + "left_ranges": [ + [ + 59, + 64 + ] + ], + "right_ranges": [ + [ + 28, + 36 + ], + [ + 41, + 55 + ], + [ + 59, + 65 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-11285-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #11285.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 11285, + "canonical_pr_number": 12964, + "issue_numbers": [ + 11285 + ], + "pr_numbers": [ + 11306, + 12964 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 11306, + "right_pr_number": 12964, + "code_similarity": 0.057, + "size_similarity": 0.214, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.096, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-12754-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12754.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 12754, + "canonical_pr_number": 12762, + "issue_numbers": [ + 12754 + ], + "pr_numbers": [ + 12762, + 13103 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 12762, + "right_pr_number": 13103, + "code_similarity": 0.538, + "size_similarity": 0.923, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.357, + "shared_filenames": [ + "src/diffusers/pipelines/pipeline_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/pipeline_utils.py", + "left_ranges": [ + [ + 2115, + 2121 + ] + ], + "right_ranges": [ + [ + 2082, + 2089 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12905-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12905.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 12905, + "canonical_pr_number": 13183, + "issue_numbers": [ + 12905 + ], + "pr_numbers": [ + 13094, + 13183 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13094, + "right_pr_number": 13183, + "code_similarity": 0.597, + "size_similarity": 0.632, + "file_overlap": 1.0, + "area_overlap": 0.438, + "patch_similarity": 0.119, + "shared_filenames": [ + "src/diffusers/pipelines/z_image/pipeline_z_image.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/z_image/pipeline_z_image.py", + "left_ranges": [ + [ + 276, + 282 + ], + [ + 335, + 343 + ] + ], + "right_ranges": [ + [ + 276, + 282 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5712-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5712.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 5712, + "canonical_pr_number": 6438, + "issue_numbers": [ + 5712 + ], + "pr_numbers": [ + 6342, + 6438 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6342, + "right_pr_number": 6438, + "code_similarity": 0.54, + "size_similarity": 0.917, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.375, + "shared_filenames": [ + "docs/source/en/installation.md" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/installation.md", + "left_ranges": [ + [ + 48, + 54 + ] + ], + "right_ranges": [ + [ + 37, + 46 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9841-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #9841.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 9841, + "canonical_pr_number": 9879, + "issue_numbers": [ + 9841 + ], + "pr_numbers": [ + 9842, + 9879 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 9842, + "right_pr_number": 9879, + "code_similarity": 0.839, + "size_similarity": 0.891, + "file_overlap": 1.0, + "area_overlap": 0.767, + "patch_similarity": 0.614, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 1183, + 1218 + ] + ], + "right_ranges": [ + [ + 67, + 73 + ], + [ + 1184, + 1216 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-13292-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13292.", + "status": "open", + "confidence": 0.6, + "canonical_issue_number": 13292, + "canonical_pr_number": 13415, + "issue_numbers": [ + 13292 + ], + "pr_numbers": [ + 13307, + 13415 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13307, + "right_pr_number": 13415, + "code_similarity": 0.48, + "size_similarity": 0.978, + "file_overlap": 0.667, + "area_overlap": 0.138, + "patch_similarity": 0.238, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_flux2.py", + "examples/dreambooth/train_dreambooth_lora_flux2_klein.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_flux2.py", + "left_ranges": [ + [ + 1734, + 1743 + ] + ], + "right_ranges": [ + [ + 1740, + 1751 + ], + [ + 1812, + 1822 + ] + ] + }, + { + "filename": "examples/dreambooth/train_dreambooth_lora_flux2_klein.py", + "left_ranges": [ + [ + 1674, + 1683 + ] + ], + "right_ranges": [ + [ + 1680, + 1691 + ], + [ + 1755, + 1765 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12656-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12656.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 12656, + "canonical_pr_number": 13110, + "issue_numbers": [ + 12656 + ], + "pr_numbers": [ + 13110, + 13516 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13110, + "right_pr_number": 13516, + "code_similarity": 0.696, + "size_similarity": 0.696, + "file_overlap": 1.0, + "area_overlap": 0.632, + "patch_similarity": 0.237, + "shared_filenames": [ + "src/diffusers/models/transformers/auraflow_transformer_2d.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/auraflow_transformer_2d.py", + "left_ranges": [ + [ + 76, + 87 + ] + ], + "right_ranges": [ + [ + 76, + 94 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12555-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13062.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 13062, + "canonical_pr_number": 12555, + "issue_numbers": [ + 13062 + ], + "pr_numbers": [ + 12555, + 13518 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 12555, + "right_pr_number": 13518, + "code_similarity": 0.599, + "size_similarity": 0.379, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.154, + "shared_filenames": [ + "src/diffusers/utils/import_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/utils/import_utils.py", + "left_ranges": [ + [ + 140, + 146 + ] + ], + "right_ranges": [ + [ + 140, + 146 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-7730-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #7730.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 7730, + "canonical_pr_number": 11156, + "issue_numbers": [ + 7730 + ], + "pr_numbers": [ + 7742, + 11156 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 7742, + "right_pr_number": 11156, + "code_similarity": 0.633, + "size_similarity": 0.833, + "file_overlap": 0.917, + "area_overlap": 0.188, + "patch_similarity": 0.837, + "shared_filenames": [ + "examples/instruct_pix2pix/train_instruct_pix2pix.py", + "examples/research_projects/instructpix2pix_lora/train_instruct_pix2pix_lora.py", + "examples/research_projects/promptdiffusion/convert_original_promptdiffusion_to_diffusers.py", + "scripts/convert_dance_diffusion_to_diffusers.py", + "scripts/convert_vae_pt_to_diffusers.py", + "src/diffusers/loaders/single_file_utils.py", + "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", + "src/diffusers/utils/loading_utils.py", + "src/diffusers/utils/testing_utils.py", + "utils/fetch_latest_release_branch.py", + "utils/notify_slack_about_release.py" + ], + "shared_file_areas": [ + { + "filename": "examples/instruct_pix2pix/train_instruct_pix2pix.py", + "left_ranges": [ + [ + 418, + 424 + ] + ], + "right_ranges": [ + [ + 49, + 55 + ], + [ + 419, + 425 + ] + ] + }, + { + "filename": "examples/research_projects/instructpix2pix_lora/train_instruct_pix2pix_lora.py", + "left_ranges": [ + [ + 378, + 384 + ] + ], + "right_ranges": [ + [ + 54, + 60 + ], + [ + 476, + 482 + ] + ] + }, + { + "filename": "examples/research_projects/promptdiffusion/convert_original_promptdiffusion_to_diffusers.py", + "left_ranges": [ + [ + 1435, + 1441 + ] + ], + "right_ranges": [ + [ + 59, + 65 + ], + [ + 1436, + 1442 + ] + ] + }, + { + "filename": "scripts/convert_dance_diffusion_to_diffusers.py", + "left_ranges": [ + [ + 74, + 80 + ] + ], + "right_ranges": [ + [ + 11, + 17 + ], + [ + 75, + 81 + ] + ] + }, + { + "filename": "scripts/convert_vae_pt_to_diffusers.py", + "left_ranges": [ + [ + 122, + 128 + ] + ], + "right_ranges": [ + [ + 13, + 19 + ], + [ + 123, + 130 + ] + ] + }, + { + "filename": "src/diffusers/loaders/single_file_utils.py", + "left_ranges": [ + [ + 381, + 387 + ], + [ + 402, + 408 + ] + ], + "right_ranges": [ + [ + 44, + 50 + ], + [ + 444, + 450 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py", + "left_ranges": [ + [ + 1324, + 1330 + ] + ], + "right_ranges": [ + [ + 52, + 58 + ], + [ + 1325, + 1331 + ] + ] + }, + { + "filename": "src/diffusers/utils/loading_utils.py", + "left_ranges": [ + [ + 25, + 31 + ] + ], + "right_ranges": [ + [ + 7, + 13 + ], + [ + 30, + 36 + ] + ] + }, + { + "filename": "src/diffusers/utils/testing_utils.py", + "left_ranges": [ + [ + 396, + 402 + ], + [ + 417, + 423 + ], + [ + 436, + 442 + ] + ], + "right_ranges": [ + [ + 26, + 32 + ], + [ + 595, + 601 + ], + [ + 616, + 622 + ], + [ + 635, + 641 + ] + ] + }, + { + "filename": "utils/fetch_latest_release_branch.py", + "left_ranges": [ + [ + 27, + 33 + ] + ], + "right_ranges": [ + [ + 13, + 24 + ], + [ + 30, + 40 + ] + ] + }, + { + "filename": "utils/notify_slack_about_release.py", + "left_ranges": [ + [ + 26, + 32 + ], + [ + 38, + 44 + ] + ], + "right_ranges": [ + [ + 17, + 24 + ], + [ + 28, + 34 + ], + [ + 40, + 46 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12540-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12540.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 12540, + "canonical_pr_number": 13109, + "issue_numbers": [ + 12540 + ], + "pr_numbers": [ + 13109, + 13517 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 13109, + "right_pr_number": 13517, + "code_similarity": 0.441, + "size_similarity": 0.667, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.052, + "shared_filenames": [ + "src/diffusers/models/transformers/sana_transformer.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/sana_transformer.py", + "left_ranges": [ + [ + 108, + 121 + ] + ], + "right_ranges": [ + [ + 457, + 475 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-6545-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #6545.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 6545, + "canonical_pr_number": 6569, + "issue_numbers": [ + 6545 + ], + "pr_numbers": [ + 6569, + 8283 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 6569, + "right_pr_number": 8283, + "code_similarity": 0.258, + "size_similarity": 0.981, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.411, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-10076-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #10076.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 10076, + "canonical_pr_number": 10449, + "issue_numbers": [ + 10076 + ], + "pr_numbers": [ + 10449, + 13435 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 10449, + "right_pr_number": 13435, + "code_similarity": 0.247, + "size_similarity": 0.979, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.342, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-13411-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13411.", + "status": "open", + "confidence": 0.55, + "canonical_issue_number": 13411, + "canonical_pr_number": 13466, + "issue_numbers": [ + 13411 + ], + "pr_numbers": [ + 13412, + 13466 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 13412, + "right_pr_number": 13466, + "code_similarity": 0.138, + "size_similarity": 0.05, + "file_overlap": 0.25, + "area_overlap": 0.12, + "patch_similarity": 0.075, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_ltx_euler_ancestral_rf.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/schedulers/scheduling_ltx_euler_ancestral_rf.py", + "left_ranges": [ + [ + 1, + 4 + ], + [ + 65, + 73 + ], + [ + 83, + 97 + ], + [ + 237, + 259 + ], + [ + 271, + 278 + ], + [ + 358, + 377 + ] + ], + "right_ranges": [ + [ + 233, + 246 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-11208-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #11208.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 11208, + "canonical_pr_number": 11251, + "issue_numbers": [ + 11208 + ], + "pr_numbers": [ + 11251, + 12400 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 11251, + "right_pr_number": 12400, + "code_similarity": 0.336, + "size_similarity": 0.584, + "file_overlap": 0.5, + "area_overlap": 0.087, + "patch_similarity": 0.261, + "shared_filenames": [ + "src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py", + "left_ranges": [ + [ + 15, + 22 + ], + [ + 41, + 47 + ], + [ + 229, + 236 + ], + [ + 576, + 639 + ], + [ + 716, + 791 + ], + [ + 1157, + 1180 + ], + [ + 1242, + 1267 + ] + ], + "right_ranges": [ + [ + 1119, + 1171 + ], + [ + 1174, + 1183 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12913-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12913.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 12913, + "canonical_pr_number": 13513, + "issue_numbers": [ + 12913 + ], + "pr_numbers": [ + 13272, + 13513 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13272, + "right_pr_number": 13513, + "code_similarity": 0.45, + "size_similarity": 0.2, + "file_overlap": 0.5, + "area_overlap": 0.667, + "patch_similarity": 0.18, + "shared_filenames": [ + "src/diffusers/pipelines/lumina2/pipeline_lumina2.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/lumina2/pipeline_lumina2.py", + "left_ranges": [ + [ + 696, + 703 + ] + ], + "right_ranges": [ + [ + 696, + 707 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-4704-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #4704.", + "status": "closed", + "confidence": 0.55, + "canonical_issue_number": 4704, + "canonical_pr_number": 4926, + "issue_numbers": [ + 4704 + ], + "pr_numbers": [ + 4707, + 4926 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 4707, + "right_pr_number": 4926, + "code_similarity": 0.197, + "size_similarity": 0.962, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.029, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-9610-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #9610.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 9610, + "canonical_pr_number": 9659, + "issue_numbers": [ + 9610 + ], + "pr_numbers": [ + 9659, + 9843 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 9659, + "right_pr_number": 9843, + "code_similarity": 0.292, + "size_similarity": 0.331, + "file_overlap": 0.5, + "area_overlap": 0.11, + "patch_similarity": 0.246, + "shared_filenames": [ + "tests/lora/test_lora_layers_sd3.py" + ], + "shared_file_areas": [ + { + "filename": "tests/lora/test_lora_layers_sd3.py", + "left_ranges": [ + [ + 12, + 40 + ], + [ + 45, + 54 + ], + [ + 128, + 215 + ] + ], + "right_ranges": [ + [ + 166, + 179 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-13243-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13243.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 13243, + "canonical_pr_number": 13512, + "issue_numbers": [ + 13243 + ], + "pr_numbers": [ + 13246, + 13512 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 13246, + "right_pr_number": 13512, + "code_similarity": 0.475, + "size_similarity": 0.26, + "file_overlap": 0.5, + "area_overlap": 0.714, + "patch_similarity": 0.152, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_flow_match_euler_discrete.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/schedulers/scheduling_flow_match_euler_discrete.py", + "left_ranges": [ + [ + 127, + 137 + ], + [ + 144, + 149 + ] + ], + "right_ranges": [ + [ + 127, + 139 + ], + [ + 146, + 151 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-13445-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13445.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 13445, + "canonical_pr_number": 13455, + "issue_numbers": [ + 13445 + ], + "pr_numbers": [ + 13455, + 13509 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 13455, + "right_pr_number": 13509, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_flux2_klein.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_flux2_klein.py", + "left_ranges": [ + [ + 358, + 364 + ] + ], + "right_ranges": [ + [ + 358, + 364 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-4553-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #4553.", + "status": "closed", + "confidence": 0.6, + "canonical_issue_number": 4553, + "canonical_pr_number": 4558, + "issue_numbers": [ + 4553 + ], + "pr_numbers": [ + 4558, + 4603 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 4558, + "right_pr_number": 4603, + "code_similarity": 0.423, + "size_similarity": 0.12, + "file_overlap": 0.125, + "area_overlap": 1.0, + "patch_similarity": 0.078, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", + "left_ranges": [ + [ + 540, + 546 + ] + ], + "right_ranges": [ + [ + 540, + 546 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-12496-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #12499.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 12499, + "canonical_pr_number": 12523, + "issue_numbers": [ + 12499 + ], + "pr_numbers": [ + 12496, + 12523 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 12496, + "right_pr_number": 12523, + "code_similarity": 0.059, + "size_similarity": 0.281, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.02, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-5214-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5216.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 5216, + "canonical_pr_number": 5305, + "issue_numbers": [ + 5216 + ], + "pr_numbers": [ + 5214, + 5305 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5214, + "right_pr_number": 5305, + "code_similarity": 0.064, + "size_similarity": 0.022, + "file_overlap": 0.019, + "area_overlap": 0.144, + "patch_similarity": 0.026, + "shared_filenames": [ + "examples/custom_diffusion/train_custom_diffusion.py", + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen.py", + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen_text_image.py" + ], + "shared_file_areas": [ + { + "filename": "examples/custom_diffusion/train_custom_diffusion.py", + "left_ranges": [ + [ + 1075, + 1104 + ] + ], + "right_ranges": [ + [ + 207, + 213 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen.py", + "left_ranges": [ + [ + 26, + 32 + ], + [ + 274, + 283 + ], + [ + 402, + 411 + ], + [ + 803, + 811 + ], + [ + 840, + 847 + ] + ], + "right_ranges": [ + [ + 803, + 811 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen_text_image.py", + "left_ranges": [ + [ + 32, + 38 + ], + [ + 306, + 315 + ], + [ + 434, + 443 + ], + [ + 965, + 973 + ], + [ + 1014, + 1021 + ] + ], + "right_ranges": [ + [ + 965, + 973 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5302-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5302.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 5302, + "canonical_pr_number": 5304, + "issue_numbers": [ + 5302 + ], + "pr_numbers": [ + 5303, + 5304 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5303, + "right_pr_number": 5304, + "code_similarity": 0.093, + "size_similarity": 0.059, + "file_overlap": 0.167, + "area_overlap": 0.089, + "patch_similarity": 0.0, + "shared_filenames": [ + "src/diffusers/pipelines/deepfloyd_if/pipeline_if.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/deepfloyd_if/pipeline_if.py", + "left_ranges": [ + [ + 158, + 165 + ], + [ + 540, + 546 + ] + ], + "right_ranges": [ + [ + 161, + 171 + ], + [ + 174, + 187 + ], + [ + 193, + 200 + ], + [ + 570, + 582 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-4629-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 4661, + "issue_numbers": [], + "pr_numbers": [ + 4629, + 4659, + 4661 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 4629, + "right_pr_number": 4659, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "docs/source/en/_toctree.yml", + "docs/source/en/api/models/overview.md", + "docs/source/en/using-diffusers/push_to_hub.md", + "examples/community/README.md", + "examples/community/lpw_stable_diffusion_xl.py", + "examples/controlnet/train_controlnet.py", + "examples/controlnet/train_controlnet_flax.py", + "examples/controlnet/train_controlnet_sdxl.py", + "examples/custom_diffusion/train_custom_diffusion.py", + "examples/dreambooth/train_dreambooth.py", + "examples/dreambooth/train_dreambooth_flax.py", + "examples/dreambooth/train_dreambooth_lora.py", + "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "examples/instruct_pix2pix/train_instruct_pix2pix.py", + "examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py", + "examples/research_projects/controlnet/train_controlnet_webdataset.py", + "examples/text_to_image/train_text_to_image.py", + "examples/text_to_image/train_text_to_image_flax.py", + "examples/text_to_image/train_text_to_image_lora.py", + "examples/text_to_image/train_text_to_image_lora_sdxl.py", + "examples/text_to_image/train_text_to_image_sdxl.py", + "examples/textual_inversion/textual_inversion.py", + "examples/textual_inversion/textual_inversion_flax.py", + "examples/unconditional_image_generation/train_unconditional.py", + "setup.py", + "src/diffusers/__init__.py", + "src/diffusers/pipelines/controlnet/multicontrolnet.py", + "src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py", + "src/diffusers/utils/hub_utils.py", + "tests/models/test_lora_layers.py", + "tests/pipelines/controlnet/test_controlnet_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/_toctree.yml", + "left_ranges": [ + [ + 32, + 37 + ] + ], + "right_ranges": [ + [ + 32, + 37 + ] + ] + }, + { + "filename": "docs/source/en/api/models/overview.md", + "left_ranges": [ + [ + 11, + 16 + ] + ], + "right_ranges": [ + [ + 11, + 16 + ] + ] + }, + { + "filename": "docs/source/en/using-diffusers/push_to_hub.md", + "left_ranges": [ + [ + 0, + 0 + ] + ], + "right_ranges": [ + [ + 0, + 0 + ] + ] + }, + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 39, + 46 + ], + [ + 1530, + 1567 + ] + ], + "right_ranges": [ + [ + 39, + 46 + ], + [ + 1530, + 1567 + ] + ] + }, + { + "filename": "examples/community/lpw_stable_diffusion_xl.py", + "left_ranges": [ + [ + 1, + 1297 + ] + ], + "right_ranges": [ + [ + 1, + 1297 + ] + ] + }, + { + "filename": "examples/controlnet/train_controlnet.py", + "left_ranges": [ + [ + 56, + 62 + ] + ], + "right_ranges": [ + [ + 56, + 62 + ] + ] + }, + { + "filename": "examples/controlnet/train_controlnet_flax.py", + "left_ranges": [ + [ + 59, + 65 + ] + ], + "right_ranges": [ + [ + 59, + 65 + ] + ] + }, + { + "filename": "examples/controlnet/train_controlnet_sdxl.py", + "left_ranges": [ + [ + 58, + 64 + ] + ], + "right_ranges": [ + [ + 58, + 64 + ] + ] + }, + { + "filename": "examples/custom_diffusion/train_custom_diffusion.py", + "left_ranges": [ + [ + 58, + 64 + ] + ], + "right_ranges": [ + [ + 58, + 64 + ] + ] + }, + { + "filename": "examples/dreambooth/train_dreambooth.py", + "left_ranges": [ + [ + 60, + 66 + ] + ], + "right_ranges": [ + [ + 60, + 66 + ] + ] + }, + { + "filename": "examples/dreambooth/train_dreambooth_flax.py", + "left_ranges": [ + [ + 36, + 42 + ] + ], + "right_ranges": [ + [ + 36, + 42 + ] + ] + }, + { + "filename": "examples/dreambooth/train_dreambooth_lora.py", + "left_ranges": [ + [ + 70, + 76 + ] + ], + "right_ranges": [ + [ + 70, + 76 + ] + ] + }, + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 58, + 64 + ] + ], + "right_ranges": [ + [ + 58, + 64 + ] + ] + }, + { + "filename": "examples/instruct_pix2pix/train_instruct_pix2pix.py", + "left_ranges": [ + [ + 52, + 58 + ] + ], + "right_ranges": [ + [ + 52, + 58 + ] + ] + }, + { + "filename": "examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py", + "left_ranges": [ + [ + 55, + 61 + ] + ], + "right_ranges": [ + [ + 55, + 61 + ] + ] + }, + { + "filename": "examples/research_projects/controlnet/train_controlnet_webdataset.py", + "left_ranges": [ + [ + 0, + 0 + ] + ], + "right_ranges": [ + [ + 0, + 0 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image.py", + "left_ranges": [ + [ + 53, + 59 + ] + ], + "right_ranges": [ + [ + 53, + 59 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_flax.py", + "left_ranges": [ + [ + 33, + 39 + ] + ], + "right_ranges": [ + [ + 33, + 39 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_lora.py", + "left_ranges": [ + [ + 48, + 54 + ] + ], + "right_ranges": [ + [ + 48, + 54 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_lora_sdxl.py", + "left_ranges": [ + [ + 57, + 63 + ] + ], + "right_ranges": [ + [ + 57, + 63 + ] + ] + }, + { + "filename": "examples/text_to_image/train_text_to_image_sdxl.py", + "left_ranges": [ + [ + 57, + 63 + ] + ], + "right_ranges": [ + [ + 57, + 63 + ] + ] + }, + { + "filename": "examples/textual_inversion/textual_inversion.py", + "left_ranges": [ + [ + 79, + 85 + ] + ], + "right_ranges": [ + [ + 79, + 85 + ] + ] + }, + { + "filename": "examples/textual_inversion/textual_inversion_flax.py", + "left_ranges": [ + [ + 56, + 62 + ] + ], + "right_ranges": [ + [ + 56, + 62 + ] + ] + }, + { + "filename": "examples/unconditional_image_generation/train_unconditional.py", + "left_ranges": [ + [ + 30, + 36 + ] + ], + "right_ranges": [ + [ + 30, + 36 + ] + ] + }, + { + "filename": "setup.py", + "left_ranges": [ + [ + 233, + 239 + ] + ], + "right_ranges": [ + [ + 233, + 239 + ] + ] + }, + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ + [ + 1, + 4 + ] + ], + "right_ranges": [ + [ + 1, + 4 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/controlnet/multicontrolnet.py", + "left_ranges": [ + [ + 39, + 44 + ], + [ + 53, + 58 + ] + ], + "right_ranges": [ + [ + 39, + 44 + ], + [ + 53, + 58 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py", + "left_ranges": [ + [ + 149, + 163 + ], + [ + 530, + 535 + ], + [ + 540, + 545 + ], + [ + 551, + 564 + ], + [ + 569, + 574 + ], + [ + 606, + 611 + ], + [ + 888, + 893 + ], + [ + 933, + 938 + ], + [ + 963, + 974 + ] + ], + "right_ranges": [ + [ + 149, + 163 + ], + [ + 530, + 535 + ], + [ + 540, + 545 + ], + [ + 551, + 564 + ], + [ + 569, + 574 + ], + [ + 606, + 611 + ], + [ + 888, + 893 + ], + [ + 933, + 938 + ], + [ + 963, + 974 + ] + ] + }, + { + "filename": "src/diffusers/utils/hub_utils.py", + "left_ranges": [ + [ + 410, + 432 + ] + ], + "right_ranges": [ + [ + 410, + 432 + ] + ] + }, + { + "filename": "tests/models/test_lora_layers.py", + "left_ranges": [ + [ + 664, + 669 + ] + ], + "right_ranges": [ + [ + 664, + 669 + ] + ] + }, + { + "filename": "tests/pipelines/controlnet/test_controlnet_sdxl.py", + "left_ranges": [ + [ + 26, + 31 + ], + [ + 46, + 52 + ], + [ + 297, + 299 + ] + ], + "right_ranges": [ + [ + 26, + 31 + ], + [ + 46, + 52 + ], + [ + 297, + 299 + ] + ] + } + ] + }, + { + "left_pr_number": 4629, + "right_pr_number": 4661, + "code_similarity": 0.514, + "size_similarity": 0.357, + "file_overlap": 0.065, + "area_overlap": 0.996, + "patch_similarity": 0.5, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/lpw_stable_diffusion_xl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 39, + 46 + ], + [ + 1530, + 1567 + ] + ], + "right_ranges": [ + [ + 39, + 46 + ], + [ + 1530, + 1573 + ] + ] + }, + { + "filename": "examples/community/lpw_stable_diffusion_xl.py", + "left_ranges": [ + [ + 1, + 1297 + ] + ], + "right_ranges": [ + [ + 1, + 1297 + ] + ] + } + ] + }, + { + "left_pr_number": 4659, + "right_pr_number": 4661, + "code_similarity": 0.514, + "size_similarity": 0.357, + "file_overlap": 0.065, + "area_overlap": 0.996, + "patch_similarity": 0.5, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/lpw_stable_diffusion_xl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 39, + 46 + ], + [ + 1530, + 1567 + ] + ], + "right_ranges": [ + [ + 39, + 46 + ], + [ + 1530, + 1573 + ] + ] + }, + { + "filename": "examples/community/lpw_stable_diffusion_xl.py", + "left_ranges": [ + [ + 1, + 1297 + ] + ], + "right_ranges": [ + [ + 1, + 1297 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-13462-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #13462.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 13462, + "canonical_pr_number": 13471, + "issue_numbers": [ + 13462 + ], + "pr_numbers": [ + 13468, + 13471 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 13468, + "right_pr_number": 13471, + "code_similarity": 0.43, + "size_similarity": 0.344, + "file_overlap": 1.0, + "area_overlap": 0.132, + "patch_similarity": 0.102, + "shared_filenames": [ + "examples/profiling/profiling_utils.py" + ], + "shared_file_areas": [ + { + "filename": "examples/profiling/profiling_utils.py", + "left_ranges": [ + [ + 1, + 6 + ], + [ + 29, + 46 + ], + [ + 49, + 56 + ], + [ + 59, + 95 + ], + [ + 169, + 177 + ], + [ + 217, + 225 + ] + ], + "right_ranges": [ + [ + 45, + 60 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5285-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5285.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 5285, + "canonical_pr_number": 5478, + "issue_numbers": [ + 5285 + ], + "pr_numbers": [ + 5286, + 5478 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 5286, + "right_pr_number": 5478, + "code_similarity": 0.144, + "size_similarity": 0.035, + "file_overlap": 0.143, + "area_overlap": 0.224, + "patch_similarity": 0.104, + "shared_filenames": [ + "docs/source/ja/index.md" + ], + "shared_file_areas": [ + { + "filename": "docs/source/ja/index.md", + "left_ranges": [ + [ + 1, + 22 + ] + ], + "right_ranges": [ + [ + 1, + 98 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-10817-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #10817.", + "status": "merged", + "confidence": 0.8, + "canonical_issue_number": 10817, + "canonical_pr_number": 10888, + "issue_numbers": [ + 10817 + ], + "pr_numbers": [ + 10830, + 10888 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10830, + "right_pr_number": 10888, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/diffusers/__init__.py", + "src/diffusers/pipelines/auto_pipeline.py", + "src/diffusers/utils/dummy_torch_and_transformers_objects.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/__init__.py", + "left_ranges": [ + [ + 865, + 871 + ] + ], + "right_ranges": [ + [ + 865, + 871 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/auto_pipeline.py", + "left_ranges": [ + [ + 34, + 43 + ], + [ + 124, + 130 + ], + [ + 183, + 189 + ] + ], + "right_ranges": [ + [ + 34, + 43 + ], + [ + 124, + 130 + ], + [ + 183, + 189 + ] + ] + }, + { + "filename": "src/diffusers/utils/dummy_torch_and_transformers_objects.py", + "left_ranges": [ + [ + 1517, + 1537 + ] + ], + "right_ranges": [ + [ + 1517, + 1537 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5504-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5504.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 5504, + "canonical_pr_number": 5697, + "issue_numbers": [ + 5504 + ], + "pr_numbers": [ + 5506, + 5697 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5506, + "right_pr_number": 5697, + "code_similarity": 0.518, + "size_similarity": 0.4, + "file_overlap": 1.0, + "area_overlap": 0.073, + "patch_similarity": 0.75, + "shared_filenames": [ + "src/diffusers/models/attention_processor.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/attention_processor.py", + "left_ranges": [ + [ + 909, + 916 + ], + [ + 938, + 952 + ], + [ + 959, + 965 + ] + ], + "right_ranges": [ + [ + 879, + 887 + ], + [ + 894, + 910 + ], + [ + 918, + 924 + ], + [ + 949, + 957 + ], + [ + 964, + 970 + ], + [ + 973, + 980 + ], + [ + 991, + 997 + ], + [ + 1183, + 1190 + ], + [ + 1215, + 1222 + ], + [ + 1236, + 1242 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-3690-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #3693.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 3693, + "canonical_pr_number": 3695, + "issue_numbers": [ + 3693 + ], + "pr_numbers": [ + 3690, + 3695 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 3690, + "right_pr_number": 3695, + "code_similarity": 0.747, + "size_similarity": 0.529, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.938, + "shared_filenames": [ + "src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py", + "left_ranges": [ + [ + 41, + 61 + ] + ], + "right_ranges": [ + [ + 41, + 61 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-10453-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #10453.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 10453, + "canonical_pr_number": 10454, + "issue_numbers": [ + 10453 + ], + "pr_numbers": [ + 10454, + 10482 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 10454, + "right_pr_number": 10482, + "code_similarity": 0.657, + "size_similarity": 0.611, + "file_overlap": 1.0, + "area_overlap": 0.467, + "patch_similarity": 0.474, + "shared_filenames": [ + "src/diffusers/models/transformers/transformer_hunyuan_video.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/models/transformers/transformer_hunyuan_video.py", + "left_ranges": [ + [ + 721, + 727 + ] + ], + "right_ranges": [ + [ + 713, + 727 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-2067-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 2147, + "issue_numbers": [], + "pr_numbers": [ + 2067, + 2096, + 2147 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 2067, + "right_pr_number": 2096, + "code_similarity": 0.393, + "size_similarity": 0.3, + "file_overlap": 1.0, + "area_overlap": 0.0, + "patch_similarity": 0.221, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", + "left_ranges": [ + [ + 21, + 27 + ], + [ + 45, + 70 + ], + [ + 429, + 435 + ], + [ + 486, + 495 + ], + [ + 522, + 535 + ], + [ + 583, + 609 + ] + ], + "right_ranges": [ + [ + 509, + 519 + ], + [ + 567, + 575 + ] + ] + } + ] + }, + { + "left_pr_number": 2067, + "right_pr_number": 2147, + "code_similarity": 0.457, + "size_similarity": 0.811, + "file_overlap": 0.5, + "area_overlap": 0.296, + "patch_similarity": 0.273, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", + "left_ranges": [ + [ + 21, + 27 + ], + [ + 45, + 70 + ], + [ + 429, + 435 + ], + [ + 486, + 495 + ], + [ + 522, + 535 + ], + [ + 583, + 609 + ] + ], + "right_ranges": [ + [ + 45, + 78 + ], + [ + 515, + 522 + ], + [ + 603, + 609 + ], + [ + 657, + 665 + ] + ] + } + ] + }, + { + "left_pr_number": 2096, + "right_pr_number": 2147, + "code_similarity": 0.256, + "size_similarity": 0.243, + "file_overlap": 0.5, + "area_overlap": 0.068, + "patch_similarity": 0.226, + "shared_filenames": [ + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py", + "left_ranges": [ + [ + 509, + 519 + ], + [ + 567, + 575 + ] + ], + "right_ranges": [ + [ + 45, + 78 + ], + [ + 515, + 522 + ], + [ + 603, + 609 + ], + [ + 657, + 665 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-5264-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #5264.", + "status": "merged", + "confidence": 0.55, + "canonical_issue_number": 5264, + "canonical_pr_number": 5535, + "issue_numbers": [ + 5264 + ], + "pr_numbers": [ + 5280, + 5535 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 5280, + "right_pr_number": 5535, + "code_similarity": 0.553, + "size_similarity": 0.581, + "file_overlap": 1.0, + "area_overlap": 0.176, + "patch_similarity": 0.5, + "shared_filenames": [ + "tests/pipelines/controlnet/test_controlnet_img2img.py" + ], + "shared_file_areas": [ + { + "filename": "tests/pipelines/controlnet/test_controlnet_img2img.py", + "left_ranges": [ + [ + 72, + 79 + ], + [ + 83, + 90 + ] + ], + "right_ranges": [ + [ + 72, + 96 + ], + [ + 102, + 114 + ], + [ + 189, + 203 + ], + [ + 207, + 231 + ], + [ + 239, + 251 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-709-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #709.", + "status": "merged", + "confidence": 0.6, + "canonical_issue_number": 709, + "canonical_pr_number": 710, + "issue_numbers": [ + 709 + ], + "pr_numbers": [ + 710, + 798 + ], + "evidence_types": [ + "closing_reference", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 710, + "right_pr_number": 798, + "code_similarity": 0.25, + "size_similarity": 0.5, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 1.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-8960-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #8960.", + "status": "merged", + "confidence": 0.75, + "canonical_issue_number": 8960, + "canonical_pr_number": 8966, + "issue_numbers": [ + 8960 + ], + "pr_numbers": [ + 8961, + 8966 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 8961, + "right_pr_number": 8966, + "code_similarity": 0.838, + "size_similarity": 0.786, + "file_overlap": 1.0, + "area_overlap": 0.875, + "patch_similarity": 0.5, + "shared_filenames": [ + "docs/source/en/tutorials/basic_training.md" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/tutorials/basic_training.md", + "left_ranges": [ + [ + 340, + 346 + ] + ], + "right_ranges": [ + [ + 340, + 347 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-9083-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 9083, + "issue_numbers": [], + "pr_numbers": [ + 9083, + 10183, + 10991 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 9083, + "right_pr_number": 10183, + "code_similarity": 0.316, + "size_similarity": 0.937, + "file_overlap": 0.333, + "area_overlap": 0.022, + "patch_similarity": 0.14, + "shared_filenames": [ + "src/diffusers/loaders/single_file_model.py", + "src/diffusers/loaders/single_file_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders/single_file_model.py", + "left_ranges": [ + [ + 24, + 30 + ], + [ + 75, + 84 + ] + ], + "right_ranges": [ + [ + 23, + 29 + ], + [ + 83, + 89 + ] + ] + }, + { + "filename": "src/diffusers/loaders/single_file_utils.py", + "left_ranges": [ + [ + 77, + 83 + ], + [ + 111, + 118 + ], + [ + 506, + 516 + ], + [ + 1867, + 2061 + ] + ], + "right_ranges": [ + [ + 92, + 99 + ], + [ + 140, + 149 + ], + [ + 570, + 592 + ], + [ + 2221, + 2295 + ] + ] + } + ] + }, + { + "left_pr_number": 9083, + "right_pr_number": 10991, + "code_similarity": 0.186, + "size_similarity": 0.466, + "file_overlap": 0.2, + "area_overlap": 0.003, + "patch_similarity": 0.209, + "shared_filenames": [ + "src/diffusers/loaders/single_file_model.py", + "src/diffusers/loaders/single_file_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders/single_file_model.py", + "left_ranges": [ + [ + 24, + 30 + ], + [ + 75, + 84 + ] + ], + "right_ranges": [ + [ + 39, + 46 + ], + [ + 119, + 132 + ] + ] + }, + { + "filename": "src/diffusers/loaders/single_file_utils.py", + "left_ranges": [ + [ + 77, + 83 + ], + [ + 111, + 118 + ], + [ + 506, + 516 + ], + [ + 1867, + 2061 + ] + ], + "right_ranges": [ + [ + 117, + 124 + ], + [ + 178, + 186 + ], + [ + 669, + 689 + ], + [ + 2490, + 2496 + ], + [ + 2499, + 2520 + ], + [ + 2523, + 2606 + ], + [ + 2895, + 3146 + ] + ] + } + ] + }, + { + "left_pr_number": 10183, + "right_pr_number": 10991, + "code_similarity": 0.194, + "size_similarity": 0.437, + "file_overlap": 0.2, + "area_overlap": 0.0, + "patch_similarity": 0.31, + "shared_filenames": [ + "src/diffusers/loaders/single_file_model.py", + "src/diffusers/loaders/single_file_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/loaders/single_file_model.py", + "left_ranges": [ + [ + 23, + 29 + ], + [ + 83, + 89 + ] + ], + "right_ranges": [ + [ + 39, + 46 + ], + [ + 119, + 132 + ] + ] + }, + { + "filename": "src/diffusers/loaders/single_file_utils.py", + "left_ranges": [ + [ + 92, + 99 + ], + [ + 140, + 149 + ], + [ + 570, + 592 + ], + [ + 2221, + 2295 + ] + ], + "right_ranges": [ + [ + 117, + 124 + ], + [ + 178, + 186 + ], + [ + 669, + 689 + ], + [ + 2490, + 2496 + ], + [ + 2499, + 2520 + ], + [ + 2523, + 2606 + ], + [ + 2895, + 3146 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-4076-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 4079, + "issue_numbers": [], + "pr_numbers": [ + 4076, + 4079 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 4076, + "right_pr_number": 4079, + "code_similarity": 0.657, + "size_similarity": 0.79, + "file_overlap": 0.2, + "area_overlap": 0.909, + "patch_similarity": 0.809, + "shared_filenames": [ + "examples/instruct_pix2pix/train_instruct_pix2pix_xl.py", + "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py" + ], + "shared_file_areas": [ + { + "filename": "examples/instruct_pix2pix/train_instruct_pix2pix_xl.py", + "left_ranges": [ + [ + 1, + 1051 + ] + ], + "right_ranges": [ + [ + 1, + 1205 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py", + "left_ranges": [ + [ + 1, + 963 + ] + ], + "right_ranges": [ + [ + 1, + 919 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-1028-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 1142, + "issue_numbers": [], + "pr_numbers": [ + 1028, + 1071, + 1142 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 1028, + "right_pr_number": 1071, + "code_similarity": 0.317, + "size_similarity": 0.62, + "file_overlap": 0.2, + "area_overlap": 0.236, + "patch_similarity": 0.34, + "shared_filenames": [ + "README.md", + "docs/source/api/schedulers.mdx", + "docs/source/optimization/fp16.mdx", + "examples/community/README.md", + "examples/community/multilingual_stable_diffusion.py", + "setup.py", + "src/diffusers/dependency_versions_table.py", + "src/diffusers/pipeline_flax_utils.py", + "src/diffusers/pipeline_utils.py", + "src/diffusers/pipelines/README.md", + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", + "src/diffusers/utils/__init__.py", + "src/diffusers/utils/testing_utils.py", + "tests/models/test_models_unet_2d.py", + "tests/models/test_models_vae.py", + "tests/pipelines/stable_diffusion/test_stable_diffusion.py", + "tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py", + "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint.py", + "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint_legacy.py", + "tests/test_pipelines.py" + ], + "shared_file_areas": [ + { + "filename": "README.md", + "left_ranges": [ + [ + 182, + 190 + ], + [ + 331, + 355 + ] + ], + "right_ranges": [ + [ + 142, + 148 + ] + ] + }, + { + "filename": "docs/source/api/schedulers.mdx", + "left_ranges": [ + [ + 89, + 95 + ], + [ + 99, + 107 + ] + ], + "right_ranges": [ + [ + 112, + 130 + ] + ] + }, + { + "filename": "docs/source/optimization/fp16.mdx", + "left_ranges": [ + [ + 18, + 26 + ] + ], + "right_ranges": [ + [ + 22, + 28 + ], + [ + 291, + 331 + ] + ] + }, + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 16, + 24 + ], + [ + 374, + 514 + ] + ], + "right_ranges": [ + [ + 17, + 25 + ], + [ + 375, + 423 + ], + [ + 501, + 572 + ] + ] + }, + { + "filename": "examples/community/multilingual_stable_diffusion.py", + "left_ranges": [ + [ + 1, + 356 + ] + ], + "right_ranges": [ + [ + 1, + 436 + ] + ] + }, + { + "filename": "setup.py", + "left_ranges": [ + [ + 94, + 100 + ], + [ + 182, + 188 + ] + ], + "right_ranges": [ + [ + 89, + 98 + ], + [ + 180, + 185 + ] + ] + }, + { + "filename": "src/diffusers/dependency_versions_table.py", + "left_ranges": [ + [ + 18, + 24 + ] + ], + "right_ranges": [ + [ + 13, + 22 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_flax_utils.py", + "left_ranges": [ + [ + 444, + 454 + ] + ], + "right_ranges": [ + [ + 29, + 35 + ], + [ + 272, + 278 + ], + [ + 302, + 314 + ], + [ + 319, + 325 + ] + ] + }, + { + "filename": "src/diffusers/pipeline_utils.py", + "left_ranges": [ + [ + 625, + 635 + ] + ], + "right_ranges": [ + [ + 30, + 38 + ], + [ + 360, + 366 + ], + [ + 398, + 411 + ], + [ + 602, + 608 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/README.md", + "left_ranges": [ + [ + 30, + 49 + ] + ], + "right_ranges": [ + [ + 16, + 22 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", + "left_ranges": [ + [ + 208, + 213 + ] + ], + "right_ranges": [ + [ + 9, + 21 + ], + [ + 58, + 66 + ], + [ + 80, + 98 + ], + [ + 113, + 136 + ], + [ + 172, + 179 + ], + [ + 373, + 383 + ] + ] + }, + { + "filename": "src/diffusers/utils/__init__.py", + "left_ranges": [ + [ + 40, + 55 + ] + ], + "right_ranges": [ + [ + 42, + 48 + ] + ] + }, + { + "filename": "src/diffusers/utils/testing_utils.py", + "left_ranges": [ + [ + 4, + 17 + ], + [ + 37, + 50 + ], + [ + 168, + 186 + ] + ], + "right_ranges": [ + [ + 139, + 167 + ], + [ + 191, + 203 + ] + ] + }, + { + "filename": "tests/models/test_models_unet_2d.py", + "left_ranges": [ + [ + 21, + 33 + ], + [ + 68, + 73 + ], + [ + 150, + 163 + ], + [ + 208, + 214 + ], + [ + 288, + 294 + ], + [ + 340, + 348 + ], + [ + 356, + 362 + ], + [ + 378, + 384 + ], + [ + 403, + 596 + ] + ], + "right_ranges": [ + [ + 21, + 35 + ], + [ + 431, + 437 + ], + [ + 447, + 453 + ] + ] + }, + { + "filename": "tests/models/test_models_vae.py", + "left_ranges": [ + [ + 1, + 303 + ] + ], + "right_ranges": [ + [ + 20, + 26 + ], + [ + 147, + 153 + ] + ] + }, + { + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion.py", + "left_ranges": [ + [ + 528, + 534 + ], + [ + 548, + 554 + ], + [ + 576, + 582 + ], + [ + 595, + 604 + ], + [ + 634, + 643 + ], + [ + 672, + 678 + ], + [ + 714, + 720 + ], + [ + 740, + 746 + ] + ], + "right_ranges": [ + [ + 24, + 31 + ], + [ + 363, + 458 + ], + [ + 640, + 652 + ] + ] + }, + { + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py", + "left_ranges": [ + [ + 488, + 494 + ], + [ + 530, + 536 + ], + [ + 582, + 588 + ] + ], + "right_ranges": [ + [ + 523, + 530 + ] + ] + }, + { + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint.py", + "left_ranges": [ + [ + 288, + 294 + ], + [ + 330, + 336 + ], + [ + 368, + 376 + ] + ], + "right_ranges": [ + [ + 28, + 34 + ], + [ + 278, + 287 + ], + [ + 306, + 312 + ], + [ + 317, + 326 + ], + [ + 347, + 353 + ], + [ + 358, + 370 + ], + [ + 385, + 388 + ] + ] + }, + { + "filename": "tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint_legacy.py", + "left_ranges": [ + [ + 368, + 374 + ], + [ + 414, + 420 + ], + [ + 471, + 477 + ] + ], + "right_ranges": [ + [ + 31, + 37 + ], + [ + 358, + 367 + ], + [ + 388, + 394 + ], + [ + 400, + 412 + ], + [ + 432, + 438 + ] + ] + }, + { + "filename": "tests/test_pipelines.py", + "left_ranges": [ + [ + 77, + 83 + ], + [ + 86, + 92 + ], + [ + 98, + 104 + ], + [ + 111, + 117 + ], + [ + 121, + 127 + ], + [ + 316, + 324 + ], + [ + 341, + 349 + ], + [ + 366, + 372 + ], + [ + 382, + 393 + ], + [ + 404, + 418 + ], + [ + 426, + 432 + ], + [ + 448, + 454 + ], + [ + 472, + 478 + ] + ], + "right_ranges": [ + [ + 42, + 48 + ], + [ + 446, + 454 + ], + [ + 462, + 485 + ], + [ + 494, + 510 + ] + ] + } + ] + }, + { + "left_pr_number": 1028, + "right_pr_number": 1142, + "code_similarity": 0.304, + "size_similarity": 0.196, + "file_overlap": 0.043, + "area_overlap": 0.569, + "patch_similarity": 0.354, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/multilingual_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 16, + 24 + ], + [ + 374, + 514 + ] + ], + "right_ranges": [ + [ + 18, + 28 + ], + [ + 504, + 577 + ], + [ + 604, + 607 + ] + ] + }, + { + "filename": "examples/community/multilingual_stable_diffusion.py", + "left_ranges": [ + [ + 1, + 356 + ] + ], + "right_ranges": [ + [ + 1, + 436 + ] + ] + } + ] + }, + { + "left_pr_number": 1071, + "right_pr_number": 1142, + "code_similarity": 0.401, + "size_similarity": 0.122, + "file_overlap": 0.027, + "area_overlap": 0.888, + "patch_similarity": 0.386, + "shared_filenames": [ + "examples/community/README.md", + "examples/community/multilingual_stable_diffusion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/community/README.md", + "left_ranges": [ + [ + 17, + 25 + ], + [ + 375, + 423 + ], + [ + 501, + 572 + ] + ], + "right_ranges": [ + [ + 18, + 28 + ], + [ + 504, + 577 + ], + [ + 604, + 607 + ] + ] + }, + { + "filename": "examples/community/multilingual_stable_diffusion.py", + "left_ranges": [ + [ + 1, + 436 + ] + ], + "right_ranges": [ + [ + 1, + 436 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-2812-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 2815, + "issue_numbers": [], + "pr_numbers": [ + 2812, + 2813, + 2815 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 2812, + "right_pr_number": 2813, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "docs/source/en/api/pipelines/stable_unclip.mdx", + "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", + "left_ranges": [ + [ + 16, + 25 + ], + [ + 28, + 42 + ], + [ + 50, + 57 + ] + ], + "right_ranges": [ + [ + 16, + 25 + ], + [ + 28, + 42 + ], + [ + 50, + 57 + ] + ] + }, + { + "filename": "src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py", + "left_ranges": [ + [ + 537, + 563 + ], + [ + 568, + 580 + ] + ], + "right_ranges": [ + [ + 537, + 563 + ], + [ + 568, + 580 + ] + ] + } + ] + }, + { + "left_pr_number": 2812, + "right_pr_number": 2815, + "code_similarity": 0.601, + "size_similarity": 0.725, + "file_overlap": 0.5, + "area_overlap": 0.66, + "patch_similarity": 0.503, + "shared_filenames": [ + "docs/source/en/api/pipelines/stable_unclip.mdx" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", + "left_ranges": [ + [ + 16, + 25 + ], + [ + 28, + 42 + ], + [ + 50, + 57 + ] + ], + "right_ranges": [ + [ + 16, + 25 + ], + [ + 28, + 42 + ], + [ + 50, + 74 + ] + ] + } + ] + }, + { + "left_pr_number": 2813, + "right_pr_number": 2815, + "code_similarity": 0.601, + "size_similarity": 0.725, + "file_overlap": 0.5, + "area_overlap": 0.66, + "patch_similarity": 0.503, + "shared_filenames": [ + "docs/source/en/api/pipelines/stable_unclip.mdx" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/api/pipelines/stable_unclip.mdx", + "left_ranges": [ + [ + 16, + 25 + ], + [ + 28, + 42 + ], + [ + 50, + 57 + ] + ], + "right_ranges": [ + [ + 16, + 25 + ], + [ + 28, + 42 + ], + [ + 50, + 74 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-386-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 386, + "issue_numbers": [], + "pr_numbers": [ + 386, + 388, + 417 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 386, + "right_pr_number": 388, + "code_similarity": 0.089, + "size_similarity": 0.244, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.268, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 386, + "right_pr_number": 417, + "code_similarity": 0.068, + "size_similarity": 0.262, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.106, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 388, + "right_pr_number": 417, + "code_similarity": 0.204, + "size_similarity": 0.933, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.113, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-643-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 645, + "issue_numbers": [], + "pr_numbers": [ + 643, + 644, + 645 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 643, + "right_pr_number": 644, + "code_similarity": 0.505, + "size_similarity": 0.216, + "file_overlap": 1.0, + "area_overlap": 0.304, + "patch_similarity": 0.373, + "shared_filenames": [ + "examples/textual_inversion/textual_inversion.py", + "src/diffusers/dependency_versions_table.py", + "src/diffusers/models/resnet.py", + "src/diffusers/schedulers/scheduling_sde_ve.py", + "src/diffusers/schedulers/scheduling_sde_ve_flax.py" + ], + "shared_file_areas": [ + { + "filename": "examples/textual_inversion/textual_inversion.py", + "left_ranges": [ + [ + 29, + 49 + ], + [ + 552, + 559 + ], + [ + 579, + 585 + ] + ], + "right_ranges": [ + [ + 29, + 49 + ], + [ + 552, + 559 + ], + [ + 579, + 585 + ] + ] + }, + { + "filename": "src/diffusers/dependency_versions_table.py", + "left_ranges": [ + [ + 17, + 22 + ] + ], + "right_ranges": [ + [ + 17, + 22 + ] + ] + }, + { + "filename": "src/diffusers/models/resnet.py", + "left_ranges": [ + [ + 1, + 482 + ] + ], + "right_ranges": [ + [ + 109, + 132 + ], + [ + 200, + 222 + ], + [ + 378, + 400 + ], + [ + 413, + 434 + ] + ] + }, + { + "filename": "src/diffusers/schedulers/scheduling_sde_ve.py", + "left_ranges": [ + [ + 94, + 101 + ], + [ + 120, + 129 + ] + ], + "right_ranges": [ + [ + 94, + 101 + ], + [ + 120, + 129 + ] + ] + }, + { + "filename": "src/diffusers/schedulers/scheduling_sde_ve_flax.py", + "left_ranges": [ + [ + 104, + 111 + ], + [ + 132, + 141 + ] + ], + "right_ranges": [ + [ + 104, + 111 + ], + [ + 132, + 141 + ] + ] + } + ] + }, + { + "left_pr_number": 643, + "right_pr_number": 645, + "code_similarity": 0.433, + "size_similarity": 0.03, + "file_overlap": 0.2, + "area_overlap": 1.0, + "patch_similarity": 0.115, + "shared_filenames": [ + "examples/textual_inversion/textual_inversion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/textual_inversion/textual_inversion.py", + "left_ranges": [ + [ + 29, + 49 + ], + [ + 552, + 559 + ], + [ + 579, + 585 + ] + ], + "right_ranges": [ + [ + 29, + 49 + ], + [ + 552, + 559 + ], + [ + 579, + 585 + ] + ] + } + ] + }, + { + "left_pr_number": 644, + "right_pr_number": 645, + "code_similarity": 0.472, + "size_similarity": 0.138, + "file_overlap": 0.2, + "area_overlap": 1.0, + "patch_similarity": 0.228, + "shared_filenames": [ + "examples/textual_inversion/textual_inversion.py" + ], + "shared_file_areas": [ + { + "filename": "examples/textual_inversion/textual_inversion.py", + "left_ranges": [ + [ + 29, + 49 + ], + [ + 552, + 559 + ], + [ + 579, + 585 + ] + ], + "right_ranges": [ + [ + 29, + 49 + ], + [ + 552, + 559 + ], + [ + 579, + 585 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-7072-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 7072, + "issue_numbers": [], + "pr_numbers": [ + 7072, + 7233, + 7235 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 7072, + "right_pr_number": 7233, + "code_similarity": 0.087, + "size_similarity": 0.253, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.243, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7072, + "right_pr_number": 7235, + "code_similarity": 0.272, + "size_similarity": 0.687, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.899, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 7233, + "right_pr_number": 7235, + "code_similarity": 0.614, + "size_similarity": 0.368, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.271, + "shared_filenames": [ + "examples/dreambooth/train_dreambooth_lora_sdxl.py" + ], + "shared_file_areas": [ + { + "filename": "examples/dreambooth/train_dreambooth_lora_sdxl.py", + "left_ranges": [ + [ + 647, + 661 + ], + [ + 1156, + 1162 + ], + [ + 1168, + 1174 + ] + ], + "right_ranges": [ + [ + 647, + 661 + ], + [ + 1156, + 1162 + ], + [ + 1168, + 1174 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-11445-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 11503, + "issue_numbers": [], + "pr_numbers": [ + 11445, + 11503, + 11507 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 11445, + "right_pr_number": 11503, + "code_similarity": 0.056, + "size_similarity": 0.236, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.056, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11445, + "right_pr_number": 11507, + "code_similarity": 0.111, + "size_similarity": 0.509, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.06, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 11503, + "right_pr_number": 11507, + "code_similarity": 0.03, + "size_similarity": 0.12, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.042, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-326-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 327, + "issue_numbers": [], + "pr_numbers": [ + 326, + 327, + 349 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 326, + "right_pr_number": 327, + "code_similarity": 0.549, + "size_similarity": 0.144, + "file_overlap": 0.067, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "src/diffusers/pipelines/pndm/pipeline_pndm.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/pipelines/pndm/pipeline_pndm.py", + "left_ranges": [ + [ + 15, + 47 + ] + ], + "right_ranges": [ + [ + 15, + 47 + ] + ] + } + ] + }, + { + "left_pr_number": 326, + "right_pr_number": 349, + "code_similarity": 0.13, + "size_similarity": 0.202, + "file_overlap": 0.067, + "area_overlap": 0.119, + "patch_similarity": 0.189, + "shared_filenames": [ + "src/diffusers/schedulers/scheduling_ddpm.py" + ], + "shared_file_areas": [ + { + "filename": "src/diffusers/schedulers/scheduling_ddpm.py", + "left_ranges": [ + [ + 60, + 65 + ] + ], + "right_ranges": [ + [ + 15, + 21 + ], + [ + 51, + 64 + ], + [ + 87, + 93 + ], + [ + 179, + 191 + ] + ] + } + ] + }, + { + "left_pr_number": 327, + "right_pr_number": 349, + "code_similarity": 0.17, + "size_similarity": 0.711, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.189, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-5833-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "merged", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 5834, + "issue_numbers": [], + "pr_numbers": [ + 5833, + 5834, + 7934 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 5833, + "right_pr_number": 5834, + "code_similarity": 0.027, + "size_similarity": 0.128, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.006, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5833, + "right_pr_number": 7934, + "code_similarity": 0.047, + "size_similarity": 0.22, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.017, + "shared_filenames": [], + "shared_file_areas": [] + }, + { + "left_pr_number": 5834, + "right_pr_number": 7934, + "code_similarity": 0.117, + "size_similarity": 0.583, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.0, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + } + ], + "duplicate_issues": [ + { + "cluster_id": "cluster-11315-4", + "canonical_issue_number": 11315, + "duplicate_issue_numbers": [ + 12653 + ], + "reason": "Issues in cluster-11315-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-11702-4", + "canonical_issue_number": 11709, + "duplicate_issue_numbers": [ + 11702 + ], + "reason": "Issues in cluster-11702-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-1246-28", + "canonical_issue_number": 13124, + "duplicate_issue_numbers": [ + 1246, + 2780, + 4566, + 5412, + 5757, + 5840, + 6086, + 6098, + 6124, + 6277, + 6363, + 6366, + 6442, + 6454, + 6482, + 7330, + 8659, + 8871, + 9628, + 9973, + 10752, + 10910 + ], + "reason": "Issues in cluster-1246-28 are treated as duplicates because they share closing_reference, duplicate_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-1257-4", + "canonical_issue_number": 1257, + "duplicate_issue_numbers": [ + 2950 + ], + "reason": "Issues in cluster-1257-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-1407-4", + "canonical_issue_number": 1792, + "duplicate_issue_numbers": [ + 1407 + ], + "reason": "Issues in cluster-1407-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-1466-4", + "canonical_issue_number": 1466, + "duplicate_issue_numbers": [ + 2041 + ], + "reason": "Issues in cluster-1466-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-2098-6", + "canonical_issue_number": 3008, + "duplicate_issue_numbers": [ + 2098, + 3184 + ], + "reason": "Issues in cluster-2098-6 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-281-4", + "canonical_issue_number": 281, + "duplicate_issue_numbers": [ + 725 + ], + "reason": "Issues in cluster-281-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-4478-9", + "canonical_issue_number": 4619, + "duplicate_issue_numbers": [ + 4478, + 4752, + 5510, + 7572, + 7854 + ], + "reason": "Issues in cluster-4478-9 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-6137-4", + "canonical_issue_number": 8236, + "duplicate_issue_numbers": [ + 6137 + ], + "reason": "Issues in cluster-6137-4 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-9237-5", + "canonical_issue_number": 9237, + "duplicate_issue_numbers": [ + 9497, + 9548 + ], + "reason": "Issues in cluster-9237-5 are treated as duplicates because they share closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-9514-5", + "canonical_issue_number": 10989, + "duplicate_issue_numbers": [ + 9514, + 10337 + ], + "reason": "Issues in cluster-9514-5 are treated as duplicates because they share closing_reference, soft_similarity evidence." + } + ], + "duplicate_prs": [ + { + "cluster_id": "cluster-10025-4", + "canonical_pr_number": 10044, + "duplicate_pr_numbers": [ + 10051, + 10053 + ], + "target_issue_number": 10025, + "reason": "PRs in cluster-10025-4 are treated as duplicates because they converge on issue #10025 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-10076-3", + "canonical_pr_number": 10449, + "duplicate_pr_numbers": [ + 13435 + ], + "target_issue_number": 10076, + "reason": "PRs in cluster-10076-3 are treated as duplicates because they converge on issue #10076 with closing_reference evidence." + }, + { + "cluster_id": "cluster-10109-5", + "canonical_pr_number": 10498, + "duplicate_pr_numbers": [ + 10109, + 10364, + 10365 + ], + "target_issue_number": 10357, + "reason": "PRs in cluster-10109-5 are treated as duplicates because they converge on issue #10357 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-10132-2", + "canonical_pr_number": 10132, + "duplicate_pr_numbers": [ + 10497 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10132-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10279-2", + "canonical_pr_number": 10279, + "duplicate_pr_numbers": [ + 10323 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10279-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-1028-3", + "canonical_pr_number": 1142, + "duplicate_pr_numbers": [ + 1028, + 1071 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1028-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10386-2", + "canonical_pr_number": 10386, + "duplicate_pr_numbers": [ + 10387 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10386-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10402-2", + "canonical_pr_number": 10434, + "duplicate_pr_numbers": [ + 10402 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10402-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10453-3", + "canonical_pr_number": 10454, + "duplicate_pr_numbers": [ + 10482 + ], + "target_issue_number": 10453, + "reason": "PRs in cluster-10453-3 are treated as duplicates because they converge on issue #10453 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-10464-2", + "canonical_pr_number": 10465, + "duplicate_pr_numbers": [ + 10464 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10464-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10504-2", + "canonical_pr_number": 10505, + "duplicate_pr_numbers": [ + 10504 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10504-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10517-2", + "canonical_pr_number": 10537, + "duplicate_pr_numbers": [ + 10517 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10517-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10678-2", + "canonical_pr_number": 10784, + "duplicate_pr_numbers": [ + 10678 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10678-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10817-3", + "canonical_pr_number": 10888, + "duplicate_pr_numbers": [ + 10830 + ], + "target_issue_number": 10817, + "reason": "PRs in cluster-10817-3 are treated as duplicates because they converge on issue #10817 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-10849-2", + "canonical_pr_number": 10849, + "duplicate_pr_numbers": [ + 10852 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10849-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-10921-2", + "canonical_pr_number": 10922, + "duplicate_pr_numbers": [ + 10921 + ], + "target_issue_number": null, + "reason": "PRs in cluster-10921-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-111-2", + "canonical_pr_number": 111, + "duplicate_pr_numbers": [ + 112 + ], + "target_issue_number": null, + "reason": "PRs in cluster-111-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-11208-3", + "canonical_pr_number": 11251, + "duplicate_pr_numbers": [ + 12400 + ], + "target_issue_number": 11208, + "reason": "PRs in cluster-11208-3 are treated as duplicates because they converge on issue #11208 with closing_reference evidence." + }, + { + "cluster_id": "cluster-11285-3", + "canonical_pr_number": 12964, + "duplicate_pr_numbers": [ + 11306 + ], + "target_issue_number": 11285, + "reason": "PRs in cluster-11285-3 are treated as duplicates because they converge on issue #11285 with closing_reference evidence." + }, + { + "cluster_id": "cluster-11315-4", + "canonical_pr_number": 12665, + "duplicate_pr_numbers": [ + 11316 + ], + "target_issue_number": 11315, + "reason": "PRs in cluster-11315-4 are treated as duplicates because they converge on issue #11315 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-11358-2", + "canonical_pr_number": 11747, + "duplicate_pr_numbers": [ + 11358 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11358-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-11390-7", + "canonical_pr_number": 11463, + "duplicate_pr_numbers": [ + 11395, + 11426, + 11427, + 11491, + 11492 + ], + "target_issue_number": 11390, + "reason": "PRs in cluster-11390-7 are treated as duplicates because they converge on issue #11390 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-11445-3", + "canonical_pr_number": 11503, + "duplicate_pr_numbers": [ + 11445, + 11507 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11445-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-11455-2", + "canonical_pr_number": 11455, + "duplicate_pr_numbers": [ + 11465 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11455-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-11471-2", + "canonical_pr_number": 11471, + "duplicate_pr_numbers": [ + 11472 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11471-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-11522-2", + "canonical_pr_number": 12069, + "duplicate_pr_numbers": [ + 11522 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11522-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-11663-2", + "canonical_pr_number": 11738, + "duplicate_pr_numbers": [ + 11663 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11663-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-11702-4", + "canonical_pr_number": 11704, + "duplicate_pr_numbers": [ + 11726 + ], + "target_issue_number": 11709, + "reason": "PRs in cluster-11702-4 are treated as duplicates because they converge on issue #11709 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-11730-4", + "canonical_pr_number": 11744, + "duplicate_pr_numbers": [ + 11782, + 11783 + ], + "target_issue_number": 11730, + "reason": "PRs in cluster-11730-4 are treated as duplicates because they converge on issue #11730 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-11995-2", + "canonical_pr_number": 11995, + "duplicate_pr_numbers": [ + 12086 + ], + "target_issue_number": null, + "reason": "PRs in cluster-11995-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12004-3", + "canonical_pr_number": 12006, + "duplicate_pr_numbers": [ + 12004, + 12024 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12004-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12049-2", + "canonical_pr_number": 12076, + "duplicate_pr_numbers": [ + 12049 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12049-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12142-2", + "canonical_pr_number": 12143, + "duplicate_pr_numbers": [ + 12142 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12142-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12267-2", + "canonical_pr_number": 12289, + "duplicate_pr_numbers": [ + 12267 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12267-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12269-2", + "canonical_pr_number": 12454, + "duplicate_pr_numbers": [ + 12269 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12269-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12322-2", + "canonical_pr_number": 12322, + "duplicate_pr_numbers": [ + 12820 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12322-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12336-2", + "canonical_pr_number": 12342, + "duplicate_pr_numbers": [ + 12336 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12336-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12351-2", + "canonical_pr_number": 12366, + "duplicate_pr_numbers": [ + 12351 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12351-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12441-3", + "canonical_pr_number": 12526, + "duplicate_pr_numbers": [ + 12442 + ], + "target_issue_number": 12441, + "reason": "PRs in cluster-12441-3 are treated as duplicates because they converge on issue #12441 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-12458-3", + "canonical_pr_number": 12698, + "duplicate_pr_numbers": [ + 12467 + ], + "target_issue_number": 12458, + "reason": "PRs in cluster-12458-3 are treated as duplicates because they converge on issue #12458 with closing_reference evidence." + }, + { + "cluster_id": "cluster-1246-28", + "canonical_pr_number": 6566, + "duplicate_pr_numbers": [ + 10783, + 10889, + 13273, + 13510 + ], + "target_issue_number": 13124, + "reason": "PRs in cluster-1246-28 are treated as duplicates because they converge on issue #13124 with closing_reference, duplicate_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-12496-3", + "canonical_pr_number": 12523, + "duplicate_pr_numbers": [ + 12496 + ], + "target_issue_number": 12499, + "reason": "PRs in cluster-12496-3 are treated as duplicates because they converge on issue #12499 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-12540-3", + "canonical_pr_number": 13109, + "duplicate_pr_numbers": [ + 13517 + ], + "target_issue_number": 12540, + "reason": "PRs in cluster-12540-3 are treated as duplicates because they converge on issue #12540 with closing_reference evidence." + }, + { + "cluster_id": "cluster-12555-3", + "canonical_pr_number": 12555, + "duplicate_pr_numbers": [ + 13518 + ], + "target_issue_number": 13062, + "reason": "PRs in cluster-12555-3 are treated as duplicates because they converge on issue #13062 with closing_reference evidence." + }, + { + "cluster_id": "cluster-1256-2", + "canonical_pr_number": 1258, + "duplicate_pr_numbers": [ + 1256 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1256-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-1257-4", + "canonical_pr_number": 1361, + "duplicate_pr_numbers": [ + 3000 + ], + "target_issue_number": 1257, + "reason": "PRs in cluster-1257-4 are treated as duplicates because they converge on issue #1257 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-12617-2", + "canonical_pr_number": 12617, + "duplicate_pr_numbers": [ + 12671 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12617-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12656-3", + "canonical_pr_number": 13110, + "duplicate_pr_numbers": [ + 13516 + ], + "target_issue_number": 12656, + "reason": "PRs in cluster-12656-3 are treated as duplicates because they converge on issue #12656 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-12754-3", + "canonical_pr_number": 12762, + "duplicate_pr_numbers": [ + 13103 + ], + "target_issue_number": 12754, + "reason": "PRs in cluster-12754-3 are treated as duplicates because they converge on issue #12754 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-12771-4", + "canonical_pr_number": 12788, + "duplicate_pr_numbers": [ + 13239, + 13240 + ], + "target_issue_number": 12771, + "reason": "PRs in cluster-12771-4 are treated as duplicates because they converge on issue #12771 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-12782-2", + "canonical_pr_number": 13039, + "duplicate_pr_numbers": [ + 12782 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12782-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12869-2", + "canonical_pr_number": 12870, + "duplicate_pr_numbers": [ + 12869 + ], + "target_issue_number": null, + "reason": "PRs in cluster-12869-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-12905-3", + "canonical_pr_number": 13183, + "duplicate_pr_numbers": [ + 13094 + ], + "target_issue_number": 12905, + "reason": "PRs in cluster-12905-3 are treated as duplicates because they converge on issue #12905 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-12913-3", + "canonical_pr_number": 13513, + "duplicate_pr_numbers": [ + 13272 + ], + "target_issue_number": 12913, + "reason": "PRs in cluster-12913-3 are treated as duplicates because they converge on issue #12913 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-13042-2", + "canonical_pr_number": 13042, + "duplicate_pr_numbers": [ + 13424 + ], + "target_issue_number": null, + "reason": "PRs in cluster-13042-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-13069-2", + "canonical_pr_number": 13082, + "duplicate_pr_numbers": [ + 13069 + ], + "target_issue_number": null, + "reason": "PRs in cluster-13069-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-13074-4", + "canonical_pr_number": 13075, + "duplicate_pr_numbers": [ + 13093, + 13514 + ], + "target_issue_number": 13074, + "reason": "PRs in cluster-13074-4 are treated as duplicates because they converge on issue #13074 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-13077-4", + "canonical_pr_number": 13136, + "duplicate_pr_numbers": [ + 13102, + 13128 + ], + "target_issue_number": 13077, + "reason": "PRs in cluster-13077-4 are treated as duplicates because they converge on issue #13077 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-13118-2", + "canonical_pr_number": 13189, + "duplicate_pr_numbers": [ + 13118 + ], + "target_issue_number": null, + "reason": "PRs in cluster-13118-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-13243-3", + "canonical_pr_number": 13512, + "duplicate_pr_numbers": [ + 13246 + ], + "target_issue_number": 13243, + "reason": "PRs in cluster-13243-3 are treated as duplicates because they converge on issue #13243 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-13259-2", + "canonical_pr_number": 13259, + "duplicate_pr_numbers": [ + 13567 + ], + "target_issue_number": null, + "reason": "PRs in cluster-13259-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-13292-3", + "canonical_pr_number": 13415, + "duplicate_pr_numbers": [ + 13307 + ], + "target_issue_number": 13292, + "reason": "PRs in cluster-13292-3 are treated as duplicates because they converge on issue #13292 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-1334-2", + "canonical_pr_number": 1426, + "duplicate_pr_numbers": [ + 1334 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1334-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-13377-3", + "canonical_pr_number": 13379, + "duplicate_pr_numbers": [ + 13449 + ], + "target_issue_number": 13377, + "reason": "PRs in cluster-13377-3 are treated as duplicates because they converge on issue #13377 with closing_reference evidence." + }, + { + "cluster_id": "cluster-13411-3", + "canonical_pr_number": 13466, + "duplicate_pr_numbers": [ + 13412 + ], + "target_issue_number": 13411, + "reason": "PRs in cluster-13411-3 are treated as duplicates because they converge on issue #13411 with closing_reference evidence." + }, + { + "cluster_id": "cluster-13445-3", + "canonical_pr_number": 13455, + "duplicate_pr_numbers": [ + 13509 + ], + "target_issue_number": 13445, + "reason": "PRs in cluster-13445-3 are treated as duplicates because they converge on issue #13445 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-13462-3", + "canonical_pr_number": 13471, + "duplicate_pr_numbers": [ + 13468 + ], + "target_issue_number": 13462, + "reason": "PRs in cluster-13462-3 are treated as duplicates because they converge on issue #13462 with closing_reference evidence." + }, + { + "cluster_id": "cluster-13467-2", + "canonical_pr_number": 13497, + "duplicate_pr_numbers": [ + 13467 + ], + "target_issue_number": null, + "reason": "PRs in cluster-13467-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-13493-2", + "canonical_pr_number": 13555, + "duplicate_pr_numbers": [ + 13493 + ], + "target_issue_number": null, + "reason": "PRs in cluster-13493-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-1407-4", + "canonical_pr_number": 1416, + "duplicate_pr_numbers": [ + 1796 + ], + "target_issue_number": 1792, + "reason": "PRs in cluster-1407-4 are treated as duplicates because they converge on issue #1792 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-1415-2", + "canonical_pr_number": 2757, + "duplicate_pr_numbers": [ + 1415 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1415-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-1466-4", + "canonical_pr_number": 2070, + "duplicate_pr_numbers": [ + 1547 + ], + "target_issue_number": 1466, + "reason": "PRs in cluster-1466-4 are treated as duplicates because they converge on issue #1466 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-1513-2", + "canonical_pr_number": 1513, + "duplicate_pr_numbers": [ + 1564 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1513-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-1567-2", + "canonical_pr_number": 1618, + "duplicate_pr_numbers": [ + 1567 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1567-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-1711-2", + "canonical_pr_number": 1714, + "duplicate_pr_numbers": [ + 1711 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1711-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-1788-2", + "canonical_pr_number": 2094, + "duplicate_pr_numbers": [ + 1788 + ], + "target_issue_number": null, + "reason": "PRs in cluster-1788-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2067-3", + "canonical_pr_number": 2147, + "duplicate_pr_numbers": [ + 2067, + 2096 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2067-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2098-6", + "canonical_pr_number": 3197, + "duplicate_pr_numbers": [ + 3201, + 3265 + ], + "target_issue_number": 3008, + "reason": "PRs in cluster-2098-6 are treated as duplicates because they converge on issue #3008 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-2206-2", + "canonical_pr_number": 2206, + "duplicate_pr_numbers": [ + 2260 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2206-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2217-2", + "canonical_pr_number": 2259, + "duplicate_pr_numbers": [ + 2217 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2217-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2383-2", + "canonical_pr_number": 2383, + "duplicate_pr_numbers": [ + 2384 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2383-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2417-2", + "canonical_pr_number": 2418, + "duplicate_pr_numbers": [ + 2417 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2417-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2450-2", + "canonical_pr_number": 2451, + "duplicate_pr_numbers": [ + 2450 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2450-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2541-2", + "canonical_pr_number": 2566, + "duplicate_pr_numbers": [ + 2541 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2541-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2647-3", + "canonical_pr_number": 2684, + "duplicate_pr_numbers": [ + 2647, + 2748 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2647-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2741-2", + "canonical_pr_number": 2751, + "duplicate_pr_numbers": [ + 2741 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2741-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-281-4", + "canonical_pr_number": 361, + "duplicate_pr_numbers": [ + 727 + ], + "target_issue_number": 281, + "reason": "PRs in cluster-281-4 are treated as duplicates because they converge on issue #281 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-2812-3", + "canonical_pr_number": 2815, + "duplicate_pr_numbers": [ + 2812, + 2813 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2812-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2825-2", + "canonical_pr_number": 2826, + "duplicate_pr_numbers": [ + 2825 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2825-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-2842-2", + "canonical_pr_number": 2962, + "duplicate_pr_numbers": [ + 2842 + ], + "target_issue_number": null, + "reason": "PRs in cluster-2842-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3084-2", + "canonical_pr_number": 3459, + "duplicate_pr_numbers": [ + 3084 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3084-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3168-3", + "canonical_pr_number": 3168, + "duplicate_pr_numbers": [ + 3676, + 3687 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3168-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3237-2", + "canonical_pr_number": 3285, + "duplicate_pr_numbers": [ + 3237 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3237-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-326-3", + "canonical_pr_number": 327, + "duplicate_pr_numbers": [ + 326, + 349 + ], + "target_issue_number": null, + "reason": "PRs in cluster-326-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3414-2", + "canonical_pr_number": 3414, + "duplicate_pr_numbers": [ + 3415 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3414-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-344-2", + "canonical_pr_number": 344, + "duplicate_pr_numbers": [ + 365 + ], + "target_issue_number": null, + "reason": "PRs in cluster-344-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3458-2", + "canonical_pr_number": 3460, + "duplicate_pr_numbers": [ + 3458 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3458-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3554-2", + "canonical_pr_number": 3555, + "duplicate_pr_numbers": [ + 3554 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3554-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3690-3", + "canonical_pr_number": 3695, + "duplicate_pr_numbers": [ + 3690 + ], + "target_issue_number": 3693, + "reason": "PRs in cluster-3690-3 are treated as duplicates because they converge on issue #3693 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-3709-2", + "canonical_pr_number": 3710, + "duplicate_pr_numbers": [ + 3709 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3709-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3797-2", + "canonical_pr_number": 3797, + "duplicate_pr_numbers": [ + 3957 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3797-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-386-3", + "canonical_pr_number": 386, + "duplicate_pr_numbers": [ + 388, + 417 + ], + "target_issue_number": null, + "reason": "PRs in cluster-386-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3880-3", + "canonical_pr_number": 4536, + "duplicate_pr_numbers": [ + 3921 + ], + "target_issue_number": 3880, + "reason": "PRs in cluster-3880-3 are treated as duplicates because they converge on issue #3880 with closing_reference evidence." + }, + { + "cluster_id": "cluster-3896-3", + "canonical_pr_number": 4097, + "duplicate_pr_numbers": [ + 3896, + 4016 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3896-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-395-2", + "canonical_pr_number": 395, + "duplicate_pr_numbers": [ + 679 + ], + "target_issue_number": null, + "reason": "PRs in cluster-395-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-3968-2", + "canonical_pr_number": 3968, + "duplicate_pr_numbers": [ + 4346 + ], + "target_issue_number": null, + "reason": "PRs in cluster-3968-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4076-2", + "canonical_pr_number": 4079, + "duplicate_pr_numbers": [ + 4076 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4076-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4131-2", + "canonical_pr_number": 4138, + "duplicate_pr_numbers": [ + 4131 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4131-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4315-2", + "canonical_pr_number": 5298, + "duplicate_pr_numbers": [ + 4315 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4315-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-433-2", + "canonical_pr_number": 433, + "duplicate_pr_numbers": [ + 2007 + ], + "target_issue_number": null, + "reason": "PRs in cluster-433-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4475-2", + "canonical_pr_number": 4493, + "duplicate_pr_numbers": [ + 4475 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4475-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4478-9", + "canonical_pr_number": 4796, + "duplicate_pr_numbers": [ + 4632, + 7858 + ], + "target_issue_number": 4619, + "reason": "PRs in cluster-4478-9 are treated as duplicates because they converge on issue #4619 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-4485-2", + "canonical_pr_number": 4486, + "duplicate_pr_numbers": [ + 4485 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4485-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4498-5", + "canonical_pr_number": 4502, + "duplicate_pr_numbers": [ + 4498, + 4499, + 4500, + 4501 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4498-5 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4544-2", + "canonical_pr_number": 4590, + "duplicate_pr_numbers": [ + 4544 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4544-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4553-3", + "canonical_pr_number": 4558, + "duplicate_pr_numbers": [ + 4603 + ], + "target_issue_number": 4553, + "reason": "PRs in cluster-4553-3 are treated as duplicates because they converge on issue #4553 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-4595-2", + "canonical_pr_number": 4596, + "duplicate_pr_numbers": [ + 4595 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4595-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4629-3", + "canonical_pr_number": 4661, + "duplicate_pr_numbers": [ + 4629, + 4659 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4629-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4655-2", + "canonical_pr_number": 5747, + "duplicate_pr_numbers": [ + 4655 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4655-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4704-3", + "canonical_pr_number": 4926, + "duplicate_pr_numbers": [ + 4707 + ], + "target_issue_number": 4704, + "reason": "PRs in cluster-4704-3 are treated as duplicates because they converge on issue #4704 with closing_reference evidence." + }, + { + "cluster_id": "cluster-4780-2", + "canonical_pr_number": 5151, + "duplicate_pr_numbers": [ + 4780 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4780-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4810-2", + "canonical_pr_number": 4810, + "duplicate_pr_numbers": [ + 4891 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4810-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-483-2", + "canonical_pr_number": 483, + "duplicate_pr_numbers": [ + 978 + ], + "target_issue_number": null, + "reason": "PRs in cluster-483-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4833-2", + "canonical_pr_number": 4833, + "duplicate_pr_numbers": [ + 5003 + ], + "target_issue_number": null, + "reason": "PRs in cluster-4833-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-4899-3", + "canonical_pr_number": 5938, + "duplicate_pr_numbers": [ + 4899 + ], + "target_issue_number": 5800, + "reason": "PRs in cluster-4899-3 are treated as duplicates because they converge on issue #5800 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-5062-4", + "canonical_pr_number": 5066, + "duplicate_pr_numbers": [ + 5065, + 5332 + ], + "target_issue_number": 5062, + "reason": "PRs in cluster-5062-4 are treated as duplicates because they converge on issue #5062 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-5214-3", + "canonical_pr_number": 5305, + "duplicate_pr_numbers": [ + 5214 + ], + "target_issue_number": 5216, + "reason": "PRs in cluster-5214-3 are treated as duplicates because they converge on issue #5216 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-5243-2", + "canonical_pr_number": 5255, + "duplicate_pr_numbers": [ + 5243 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5243-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-5249-4", + "canonical_pr_number": 5293, + "duplicate_pr_numbers": [ + 5299, + 5312 + ], + "target_issue_number": 5249, + "reason": "PRs in cluster-5249-4 are treated as duplicates because they converge on issue #5249 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-5262-4", + "canonical_pr_number": 5370, + "duplicate_pr_numbers": [ + 5288, + 5366 + ], + "target_issue_number": 5262, + "reason": "PRs in cluster-5262-4 are treated as duplicates because they converge on issue #5262 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-5264-3", + "canonical_pr_number": 5535, + "duplicate_pr_numbers": [ + 5280 + ], + "target_issue_number": 5264, + "reason": "PRs in cluster-5264-3 are treated as duplicates because they converge on issue #5264 with closing_reference evidence." + }, + { + "cluster_id": "cluster-5285-3", + "canonical_pr_number": 5478, + "duplicate_pr_numbers": [ + 5286 + ], + "target_issue_number": 5285, + "reason": "PRs in cluster-5285-3 are treated as duplicates because they converge on issue #5285 with closing_reference evidence." + }, + { + "cluster_id": "cluster-5302-3", + "canonical_pr_number": 5304, + "duplicate_pr_numbers": [ + 5303 + ], + "target_issue_number": 5302, + "reason": "PRs in cluster-5302-3 are treated as duplicates because they converge on issue #5302 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-5347-2", + "canonical_pr_number": 5349, + "duplicate_pr_numbers": [ + 5347 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5347-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-5382-2", + "canonical_pr_number": 5382, + "duplicate_pr_numbers": [ + 5614 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5382-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-5504-3", + "canonical_pr_number": 5697, + "duplicate_pr_numbers": [ + 5506 + ], + "target_issue_number": 5504, + "reason": "PRs in cluster-5504-3 are treated as duplicates because they converge on issue #5504 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-5524-3", + "canonical_pr_number": 5704, + "duplicate_pr_numbers": [ + 5696 + ], + "target_issue_number": 5524, + "reason": "PRs in cluster-5524-3 are treated as duplicates because they converge on issue #5524 with closing_reference evidence." + }, + { + "cluster_id": "cluster-5626-3", + "canonical_pr_number": 5626, + "duplicate_pr_numbers": [ + 7544, + 7868 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5626-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-5712-3", + "canonical_pr_number": 6438, + "duplicate_pr_numbers": [ + 6342 + ], + "target_issue_number": 5712, + "reason": "PRs in cluster-5712-3 are treated as duplicates because they converge on issue #5712 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-5833-3", + "canonical_pr_number": 5834, + "duplicate_pr_numbers": [ + 5833, + 7934 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5833-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-5878-2", + "canonical_pr_number": 5878, + "duplicate_pr_numbers": [ + 5879 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5878-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-5886-3", + "canonical_pr_number": 5911, + "duplicate_pr_numbers": [ + 5910 + ], + "target_issue_number": 5886, + "reason": "PRs in cluster-5886-3 are treated as duplicates because they converge on issue #5886 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-5906-2", + "canonical_pr_number": 5914, + "duplicate_pr_numbers": [ + 5906 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5906-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-5970-2", + "canonical_pr_number": 5971, + "duplicate_pr_numbers": [ + 5970 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5970-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-5982-2", + "canonical_pr_number": 5982, + "duplicate_pr_numbers": [ + 6065 + ], + "target_issue_number": null, + "reason": "PRs in cluster-5982-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6014-2", + "canonical_pr_number": 7638, + "duplicate_pr_numbers": [ + 6014 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6014-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6119-2", + "canonical_pr_number": 6119, + "duplicate_pr_numbers": [ + 6231 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6119-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6123-3", + "canonical_pr_number": 6328, + "duplicate_pr_numbers": [ + 6509 + ], + "target_issue_number": 6123, + "reason": "PRs in cluster-6123-3 are treated as duplicates because they converge on issue #6123 with closing_reference evidence." + }, + { + "cluster_id": "cluster-6137-4", + "canonical_pr_number": 8312, + "duplicate_pr_numbers": [ + 6143 + ], + "target_issue_number": 8236, + "reason": "PRs in cluster-6137-4 are treated as duplicates because they converge on issue #8236 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-6158-3", + "canonical_pr_number": 6721, + "duplicate_pr_numbers": [ + 6195 + ], + "target_issue_number": 6158, + "reason": "PRs in cluster-6158-3 are treated as duplicates because they converge on issue #6158 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-6261-2", + "canonical_pr_number": 6261, + "duplicate_pr_numbers": [ + 6265 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6261-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6299-4", + "canonical_pr_number": 6322, + "duplicate_pr_numbers": [ + 6299, + 6300, + 6301 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6299-4 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-643-3", + "canonical_pr_number": 645, + "duplicate_pr_numbers": [ + 643, + 644 + ], + "target_issue_number": null, + "reason": "PRs in cluster-643-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6453-2", + "canonical_pr_number": 6453, + "duplicate_pr_numbers": [ + 6462 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6453-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6503-6", + "canonical_pr_number": 6532, + "duplicate_pr_numbers": [ + 6525, + 6534, + 6555, + 6556 + ], + "target_issue_number": 6503, + "reason": "PRs in cluster-6503-6 are treated as duplicates because they converge on issue #6503 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-6514-4", + "canonical_pr_number": 6514, + "duplicate_pr_numbers": [ + 6554, + 6840, + 7242 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6514-4 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6545-3", + "canonical_pr_number": 6569, + "duplicate_pr_numbers": [ + 8283 + ], + "target_issue_number": 6545, + "reason": "PRs in cluster-6545-3 are treated as duplicates because they converge on issue #6545 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-6585-11", + "canonical_pr_number": 6737, + "duplicate_pr_numbers": [ + 6672, + 6719, + 6740, + 6765, + 6766, + 6767, + 6768, + 6769, + 6781 + ], + "target_issue_number": 6585, + "reason": "PRs in cluster-6585-11 are treated as duplicates because they converge on issue #6585 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-6587-2", + "canonical_pr_number": 6587, + "duplicate_pr_numbers": [ + 6803 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6587-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6627-2", + "canonical_pr_number": 6849, + "duplicate_pr_numbers": [ + 6627 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6627-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6688-3", + "canonical_pr_number": 7005, + "duplicate_pr_numbers": [ + 6844 + ], + "target_issue_number": 6688, + "reason": "PRs in cluster-6688-3 are treated as duplicates because they converge on issue #6688 with closing_reference evidence." + }, + { + "cluster_id": "cluster-6690-2", + "canonical_pr_number": 6692, + "duplicate_pr_numbers": [ + 6690 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6690-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6866-4", + "canonical_pr_number": 6941, + "duplicate_pr_numbers": [ + 7030, + 7031 + ], + "target_issue_number": 6866, + "reason": "PRs in cluster-6866-4 are treated as duplicates because they converge on issue #6866 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-6891-3", + "canonical_pr_number": 7615, + "duplicate_pr_numbers": [ + 7613 + ], + "target_issue_number": 6891, + "reason": "PRs in cluster-6891-3 are treated as duplicates because they converge on issue #6891 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-6909-2", + "canonical_pr_number": 6909, + "duplicate_pr_numbers": [ + 6939 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6909-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6980-2", + "canonical_pr_number": 6983, + "duplicate_pr_numbers": [ + 6980 + ], + "target_issue_number": null, + "reason": "PRs in cluster-6980-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-6984-7", + "canonical_pr_number": 7071, + "duplicate_pr_numbers": [ + 7052, + 7211, + 7307, + 8376, + 8387 + ], + "target_issue_number": 6984, + "reason": "PRs in cluster-6984-7 are treated as duplicates because they converge on issue #6984 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-7038-3", + "canonical_pr_number": 7550, + "duplicate_pr_numbers": [ + 7442 + ], + "target_issue_number": 7038, + "reason": "PRs in cluster-7038-3 are treated as duplicates because they converge on issue #7038 with closing_reference evidence." + }, + { + "cluster_id": "cluster-7072-3", + "canonical_pr_number": 7072, + "duplicate_pr_numbers": [ + 7233, + 7235 + ], + "target_issue_number": null, + "reason": "PRs in cluster-7072-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-709-3", + "canonical_pr_number": 710, + "duplicate_pr_numbers": [ + 798 + ], + "target_issue_number": 709, + "reason": "PRs in cluster-709-3 are treated as duplicates because they converge on issue #709 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-7092-2", + "canonical_pr_number": 7092, + "duplicate_pr_numbers": [ + 8687 + ], + "target_issue_number": null, + "reason": "PRs in cluster-7092-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-7677-3", + "canonical_pr_number": 7865, + "duplicate_pr_numbers": [ + 7898 + ], + "target_issue_number": 7677, + "reason": "PRs in cluster-7677-3 are treated as duplicates because they converge on issue #7677 with closing_reference evidence." + }, + { + "cluster_id": "cluster-7692-2", + "canonical_pr_number": 7876, + "duplicate_pr_numbers": [ + 7692 + ], + "target_issue_number": null, + "reason": "PRs in cluster-7692-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-7709-2", + "canonical_pr_number": 7719, + "duplicate_pr_numbers": [ + 7709 + ], + "target_issue_number": null, + "reason": "PRs in cluster-7709-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-7730-3", + "canonical_pr_number": 11156, + "duplicate_pr_numbers": [ + 7742 + ], + "target_issue_number": 7730, + "reason": "PRs in cluster-7730-3 are treated as duplicates because they converge on issue #7730 with closing_reference evidence." + }, + { + "cluster_id": "cluster-7736-3", + "canonical_pr_number": 7761, + "duplicate_pr_numbers": [ + 7739 + ], + "target_issue_number": 7736, + "reason": "PRs in cluster-7736-3 are treated as duplicates because they converge on issue #7736 with closing_reference evidence." + }, + { + "cluster_id": "cluster-8308-2", + "canonical_pr_number": 8310, + "duplicate_pr_numbers": [ + 8308 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8308-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-831-2", + "canonical_pr_number": 831, + "duplicate_pr_numbers": [ + 1490 + ], + "target_issue_number": null, + "reason": "PRs in cluster-831-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8384-4", + "canonical_pr_number": 11239, + "duplicate_pr_numbers": [ + 10973, + 11240 + ], + "target_issue_number": 8384, + "reason": "PRs in cluster-8384-4 are treated as duplicates because they converge on issue #8384 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-8446-5", + "canonical_pr_number": 8461, + "duplicate_pr_numbers": [ + 8446, + 8476, + 9308, + 9316 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8446-5 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8478-2", + "canonical_pr_number": 8478, + "duplicate_pr_numbers": [ + 8479 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8478-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8710-3", + "canonical_pr_number": 9386, + "duplicate_pr_numbers": [ + 8725 + ], + "target_issue_number": 8710, + "reason": "PRs in cluster-8710-3 are treated as duplicates because they converge on issue #8710 with closing_reference evidence." + }, + { + "cluster_id": "cluster-8773-2", + "canonical_pr_number": 8773, + "duplicate_pr_numbers": [ + 8976 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8773-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8798-2", + "canonical_pr_number": 8798, + "duplicate_pr_numbers": [ + 8913 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8798-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8810-4", + "canonical_pr_number": 8864, + "duplicate_pr_numbers": [ + 8810, + 9917 + ], + "target_issue_number": 9886, + "reason": "PRs in cluster-8810-4 are treated as duplicates because they converge on issue #9886 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-8814-2", + "canonical_pr_number": 8815, + "duplicate_pr_numbers": [ + 8814 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8814-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8816-2", + "canonical_pr_number": 8820, + "duplicate_pr_numbers": [ + 8816 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8816-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8899-2", + "canonical_pr_number": 8899, + "duplicate_pr_numbers": [ + 9037 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8899-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8932-2", + "canonical_pr_number": 8987, + "duplicate_pr_numbers": [ + 8932 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8932-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8946-2", + "canonical_pr_number": 9131, + "duplicate_pr_numbers": [ + 8946 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8946-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8949-2", + "canonical_pr_number": 8949, + "duplicate_pr_numbers": [ + 8980 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8949-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-8960-3", + "canonical_pr_number": 8966, + "duplicate_pr_numbers": [ + 8961 + ], + "target_issue_number": 8960, + "reason": "PRs in cluster-8960-3 are treated as duplicates because they converge on issue #8960 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-8973-2", + "canonical_pr_number": 8985, + "duplicate_pr_numbers": [ + 8973 + ], + "target_issue_number": null, + "reason": "PRs in cluster-8973-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-9080-2", + "canonical_pr_number": 9082, + "duplicate_pr_numbers": [ + 9080 + ], + "target_issue_number": null, + "reason": "PRs in cluster-9080-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-9083-3", + "canonical_pr_number": 9083, + "duplicate_pr_numbers": [ + 10183, + 10991 + ], + "target_issue_number": null, + "reason": "PRs in cluster-9083-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-912-2", + "canonical_pr_number": 912, + "duplicate_pr_numbers": [ + 1131 + ], + "target_issue_number": null, + "reason": "PRs in cluster-912-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-916-4", + "canonical_pr_number": 1110, + "duplicate_pr_numbers": [ + 916, + 1007, + 1225 + ], + "target_issue_number": null, + "reason": "PRs in cluster-916-4 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-9237-5", + "canonical_pr_number": 9419, + "duplicate_pr_numbers": [ + 9549 + ], + "target_issue_number": 9237, + "reason": "PRs in cluster-9237-5 are treated as duplicates because they converge on issue #9237 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-9403-3", + "canonical_pr_number": 10261, + "duplicate_pr_numbers": [ + 9787 + ], + "target_issue_number": 9403, + "reason": "PRs in cluster-9403-3 are treated as duplicates because they converge on issue #9403 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-9514-5", + "canonical_pr_number": 10990, + "duplicate_pr_numbers": [ + 10338 + ], + "target_issue_number": 10989, + "reason": "PRs in cluster-9514-5 are treated as duplicates because they converge on issue #10989 with closing_reference, soft_similarity evidence." + }, + { + "cluster_id": "cluster-9564-3", + "canonical_pr_number": 9564, + "duplicate_pr_numbers": [ + 9854, + 9855 + ], + "target_issue_number": null, + "reason": "PRs in cluster-9564-3 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-9567-8", + "canonical_pr_number": 9608, + "duplicate_pr_numbers": [ + 9579, + 9582, + 9583, + 9606, + 9724, + 9815 + ], + "target_issue_number": 9567, + "reason": "PRs in cluster-9567-8 are treated as duplicates because they converge on issue #9567 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-9591-5", + "canonical_pr_number": 9591, + "duplicate_pr_numbers": [ + 9706, + 9707, + 9714, + 9715 + ], + "target_issue_number": null, + "reason": "PRs in cluster-9591-5 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-9610-3", + "canonical_pr_number": 9659, + "duplicate_pr_numbers": [ + 9843 + ], + "target_issue_number": 9610, + "reason": "PRs in cluster-9610-3 are treated as duplicates because they converge on issue #9610 with closing_reference evidence." + }, + { + "cluster_id": "cluster-9631-2", + "canonical_pr_number": 9642, + "duplicate_pr_numbers": [ + 9631 + ], + "target_issue_number": null, + "reason": "PRs in cluster-9631-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-9772-3", + "canonical_pr_number": 10131, + "duplicate_pr_numbers": [ + 9776 + ], + "target_issue_number": 9772, + "reason": "PRs in cluster-9772-3 are treated as duplicates because they converge on issue #9772 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-9841-3", + "canonical_pr_number": 9879, + "duplicate_pr_numbers": [ + 9842 + ], + "target_issue_number": 9841, + "reason": "PRs in cluster-9841-3 are treated as duplicates because they converge on issue #9841 with closing_reference, shared_issue_target evidence." + } + ], + "best_issue": { + "cluster_id": "cluster-1246-28", + "issue_number": 13124, + "reason": "Issue #13124 is the strongest global issue candidate because it is open, belongs to a cluster with 28 artifacts, and carries 2 discussion comments plus 6 inbound references.", + "score": 390.0 + }, + "best_pr": { + "cluster_id": "cluster-10025-4", + "pr_number": 10044, + "reason": "PR #10044 is the strongest global PR candidate because it is open, belongs to a cluster with 4 artifacts, links to 2 issues, and carries 32 review/discussion events.", + "score": 233.939 + } +} diff --git a/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.llm-reviews.json b/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.llm-reviews.json new file mode 100644 index 0000000000000000000000000000000000000000..86fac2f1a1889e2ddf4df96763c182ad0211d4fb --- /dev/null +++ b/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.llm-reviews.json @@ -0,0 +1,209386 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/diffusers", + "snapshot_id": "20260427T113512Z", + "generated_at": "2026-04-27T12:02:03Z", + "model": "gpt-5.4-mini?service_tier=flex", + "reviews": [ + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10035", + "issue:10091", + "issue:10343", + "issue:10804", + "issue:10954", + "issue:11083", + "issue:11169", + "issue:11225", + "issue:11417", + "issue:12044", + "issue:12436", + "issue:12480", + "issue:12481", + "issue:2044", + "issue:2780", + "issue:2819", + "issue:2995", + "issue:3809", + "issue:3954", + "issue:4104", + "issue:4325", + "issue:4557", + "issue:4649", + "issue:4754", + "issue:4868", + "issue:5082", + "issue:5259", + "issue:5295", + "issue:5475", + "issue:5510", + "issue:5805", + "issue:6324", + "issue:6363", + "issue:6517", + "issue:6594", + "issue:6727", + "issue:6812", + "issue:6815", + "issue:6996", + "issue:7330", + "issue:7854", + "issue:8236", + "issue:8759", + "issue:9293", + "issue:9298", + "issue:9546", + "issue:9726", + "issue:9727" + ], + "soft_pairs": [ + "issue:12480|issue:12481", + "issue:5295|issue:5475", + "issue:10035|issue:10091", + "issue:5510|issue:7854", + "issue:3954|issue:8236", + "issue:12436|issue:12480", + "issue:12436|issue:12481", + "issue:6815|issue:8759", + "issue:5805|issue:7854", + "issue:10804|issue:10954", + "issue:4325|issue:8759", + "issue:4754|issue:5259", + "issue:6727|issue:6812", + "issue:4649|issue:6996", + "issue:11083|issue:3809", + "issue:12044|issue:6324", + "issue:10343|issue:6815", + "issue:2995|issue:5082", + "issue:2819|issue:6594", + "issue:9293|issue:9298", + "issue:4557|issue:4868", + "issue:11225|issue:9546", + "issue:9726|issue:9727", + "issue:6363|issue:7330", + "issue:2780|issue:6363", + "issue:4104|issue:6815", + "issue:11169|issue:6517", + "issue:11417|issue:2044" + ], + "prepared_review_unit_hash": "0133da4827865cc18ff504a8d40a95b1bf43e4a1d70288f3555e8f0c674ee8d9", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36481, + "estimated_input_tokens": 9121, + "estimated_eval_tokens": 18498 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The strongest duplicate clusters are the exact-title or near-exact error repeats, especially the CLIPTextModel/offload_state_dict reports. I kept broader symptom matches conservative when the titles point to different model or pipeline contexts and the packet lacks body text or traces to prove a shared root cause.", + "confidence": 0.76, + "canonical_issue_reason": "issue:12480 is the cleanest anchor for the offload_state_dict/CLIPTextModel cluster: it has the most complete wording, and 12481 plus 12436 are very close variants of the same report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:12480 is the best overall issue anchor because it is the clearest, most detailed representative of the strongest duplicate cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12480", + "right": "issue:12481", + "accept": true, + "reason": "Near-identical titles and the same CLIPTextModel/offload_state_dict TypeError; same bug." + }, + { + "left": "issue:5295", + "right": "issue:5475", + "accept": true, + "reason": "Same LaVie/LaVIE project title with only capitalization variation; duplicate report." + }, + { + "left": "issue:10035", + "right": "issue:10091", + "accept": true, + "reason": "Exact same CogvideoX VAE decoder memory issue title; duplicate." + }, + { + "left": "issue:5510", + "right": "issue:7854", + "accept": false, + "reason": "Both mention dtype mismatches, but the pipeline/context differs and the packet does not show one concrete shared code path." + }, + { + "left": "issue:3954", + "right": "issue:8236", + "accept": false, + "reason": "Same broad scheduler step-count symptom, but not enough shared evidence here to confirm the same underlying bug." + }, + { + "left": "issue:12436", + "right": "issue:12480", + "accept": true, + "reason": "Same offload_state_dict / CLIPTextModel TypeError family; clearly overlapping reports." + }, + { + "left": "issue:12436", + "right": "issue:12481", + "accept": true, + "reason": "Same offload_state_dict / CLIPTextModel TypeError family; duplicate variant." + }, + { + "left": "issue:6815", + "right": "issue:8759", + "accept": false, + "reason": "Both are black-image / invalid-cast symptom reports, but the model context differs and the packet does not prove the same root cause." + }, + { + "left": "issue:5805", + "right": "issue:7854", + "accept": false, + "reason": "Related dtype mismatch wording, but the specific failing path and context differ too much to merge them confidently." + }, + { + "left": "issue:10804", + "right": "issue:10954", + "accept": true, + "reason": "Same FLUX1.dev LoRA load-failure report, just reworded." + }, + { + "left": "issue:4325", + "right": "issue:8759", + "accept": false, + "reason": "The error text is similar, but this is still only a broad black-image symptom match across different model contexts." + }, + { + "left": "issue:4754", + "right": "issue:5259", + "accept": true, + "reason": "Same missing model_index.json / textual inversion loading error; duplicate." + }, + { + "left": "issue:6727", + "right": "issue:6812", + "accept": true, + "reason": "Same Query/Key/Value dtype mismatch during LoRA training; same bug family." + }, + { + "left": "issue:4649", + "right": "issue:6996", + "accept": true, + "reason": "Same NoneType / added_cond_kwargs text_embeds error; duplicate." + }, + { + "left": "issue:11083", + "right": "issue:3809", + "accept": true, + "reason": "Same torch-DDP stride mismatch warning; near-exact duplicate wording." + }, + { + "left": "issue:12044", + "right": "issue:6324", + "accept": true, + "reason": "Exact same AttributeError about bool.__module__; duplicate." + }, + { + "left": "issue:10343", + "right": "issue:6815", + "accept": false, + "reason": "Both mention black output / invalid-cast symptoms, but they are model-specific reports and not clearly the same bug." + }, + { + "left": "issue:2995", + "right": "issue:5082", + "accept": true, + "reason": "Same mat1/mat2 shape-mismatch error; duplicate." + }, + { + "left": "issue:2819", + "right": "issue:6594", + "accept": false, + "reason": "Both are raw.githubusercontent.com download failures, but the target files and underlying cause are not shown to be the same." + }, + { + "left": "issue:9293", + "right": "issue:9298", + "accept": true, + "reason": "Exact same Float8_e4m3fnStorage lookup failure; duplicate." + }, + { + "left": "issue:4557", + "right": "issue:4868", + "accept": true, + "reason": "Same super object __getattr__ AttributeError; duplicate." + }, + { + "left": "issue:11225", + "right": "issue:9546", + "accept": false, + "reason": "Both are validation-step OOM reports, but they are model-family-specific and only symptom-level similar here." + }, + { + "left": "issue:9726", + "right": "issue:9727", + "accept": false, + "reason": "Related FLUX.1-dev multi-GPU reports, but one is a training problem and the other a save problem; not the same bug." + }, + { + "left": "issue:6363", + "right": "issue:7330", + "accept": true, + "reason": "Same Attempting to unscale FP16 gradients error in the LoRA training script family; duplicate." + }, + { + "left": "issue:2780", + "right": "issue:6363", + "accept": true, + "reason": "Same FP16 unscale failure in text-to-image LoRA training; duplicate." + }, + { + "left": "issue:4104", + "right": "issue:6815", + "accept": false, + "reason": "Same broad black-image/invalid-cast symptom, but not enough evidence of one shared root cause across the different reports." + }, + { + "left": "issue:11169", + "right": "issue:6517", + "accept": false, + "reason": "Both are import failures, but the reported errors differ and do not indicate the same bug." + }, + { + "left": "issue:11417", + "right": "issue:2044", + "accept": true, + "reason": "Exact same DistributedDataParallel has no dtype AttributeError; duplicate." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The canonical choice is reasonably grounded, and several exact-title or near-exact duplicate pairs are fine. However, the soft-edge section is too assertive in a few places given the packet only provides titles/body sizes, not body text or code traces. In particular, pairs like issue:11225 vs issue:9546, issue:6815 vs issue:8759, issue:10343 vs issue:6815, and issue:3954 vs issue:8236 read like symptom-level similarity rather than clearly the same bug. For this task, those should stay conservative unless the packet shows stronger shared evidence. The summary should also avoid implying direct proof of sameness for broad symptom matches when the model/pipeline context differs." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 2, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10520", + "issue:10575", + "issue:10650", + "issue:10673", + "issue:10958", + "issue:11413", + "issue:11436", + "issue:12009", + "issue:12104", + "issue:12559", + "issue:1433", + "issue:1875", + "issue:2500", + "issue:2780", + "issue:3489", + "issue:5110", + "issue:5510", + "issue:5607", + "issue:5749", + "issue:5805", + "issue:5897", + "issue:6086", + "issue:6363", + "issue:6368", + "issue:6445", + "issue:6517", + "issue:6570", + "issue:6821", + "issue:6832", + "issue:6879", + "issue:7261", + "issue:7286", + "issue:7330", + "issue:736", + "issue:7681", + "issue:7708", + "issue:7854", + "issue:7891", + "issue:7935", + "issue:7978", + "issue:8465", + "issue:8581", + "issue:8833", + "issue:8863", + "issue:9362", + "issue:9448", + "issue:9476" + ], + "soft_pairs": [ + "issue:6517|issue:7681", + "issue:5749|issue:9362", + "issue:12559|issue:7978", + "issue:5607|issue:6570", + "issue:7935|issue:8833", + "issue:12104|issue:9362", + "issue:1875|issue:5510", + "issue:10650|issue:10673", + "issue:7854|issue:9476", + "issue:11413|issue:6517", + "issue:7708|issue:7935", + "issue:5110|issue:736", + "issue:5897|issue:7891", + "issue:6086|issue:6363", + "issue:7261|issue:7286", + "issue:1875|issue:7854", + "issue:11436|issue:6879", + "issue:10958|issue:12009", + "issue:10958|issue:2500", + "issue:8863|issue:9448", + "issue:1875|issue:5805", + "issue:6445|issue:6821", + "issue:2780|issue:7330", + "issue:6832|issue:8863", + "issue:7935|issue:8465", + "issue:10520|issue:7891", + "issue:6368|issue:8581", + "issue:10575|issue:5749", + "issue:1433|issue:3489" + ], + "prepared_review_unit_hash": "85ad060b999b6dfe4d5e443055781096e05483f557cb1d6f1af4713b8d076769", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 36851, + "estimated_input_tokens": 9213, + "estimated_eval_tokens": 18682 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster still breaks into a few well-supported duplicate subgroups, especially the LoRA FP16 unscale reports and a handful of exact-message matches. However, several tuple.shape links are too weakly grounded to treat as confirmed duplicates, so I\u2019d keep the triage conservative and not collapse the broader cluster too aggressively.", + "confidence": 0.69, + "canonical_issue_reason": "Issue 6086 is the strongest canonical anchor for the accepted duplicate subgroup: it has the richest discussion, a clear repro, and matches the same FP16-gradient unscale failure seen in 6363 and 7330.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6086 is the best overall issue to anchor the confirmed subgroup because it is detailed, well-scoped, and explicitly captures the shared LoRA training failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6517", + "right": "issue:7681", + "accept": false, + "reason": "Both are import-time failures, but they involve different modules and different underlying causes." + }, + { + "left": "issue:5749", + "right": "issue:9362", + "accept": false, + "reason": "Both are IndexErrors, but the reported sizes differ and there is no clear shared failing code path." + }, + { + "left": "issue:12559", + "right": "issue:7978", + "accept": true, + "reason": "Same concrete torchvision::nms missing-operator failure; the later report is just a more specific instance of the same bug." + }, + { + "left": "issue:5607", + "right": "issue:6570", + "accept": true, + "reason": "Both report the same tensor-size mismatch in the StableDiffusionXLInstructPix2Pix pipeline family." + }, + { + "left": "issue:7935", + "right": "issue:8833", + "accept": false, + "reason": "Both mention missing config.json, but they point to different models and loader contexts." + }, + { + "left": "issue:12104", + "right": "issue:9362", + "accept": false, + "reason": "Same broad IndexError wording, but the tensor sizes and failure context are clearly different." + }, + { + "left": "issue:1875", + "right": "issue:5510", + "accept": false, + "reason": "Both are dtype mismatch errors, but the pipeline context and concrete failure mode are not the same." + }, + { + "left": "issue:10650", + "right": "issue:10673", + "accept": false, + "reason": "The shape text matches, but the pipelines/models differ enough that this is not clearly the same underlying bug." + }, + { + "left": "issue:7854", + "right": "issue:9476", + "accept": false, + "reason": "Both are dtype mismatch reports, but they arise in different training/validation contexts and are too generic to merge." + }, + { + "left": "issue:11413", + "right": "issue:6517", + "accept": false, + "reason": "Both are import failures, but the traceback targets different pipelines and does not indicate a shared root cause." + }, + { + "left": "issue:7708", + "right": "issue:7935", + "accept": false, + "reason": "Both involve missing config.json, but the model/path context differs and the evidence is too weak for duplication." + }, + { + "left": "issue:5110", + "right": "issue:736", + "accept": true, + "reason": "Same HFValidationError about invalid repo IDs; the second report is the same validation bug in another form." + }, + { + "left": "issue:5897", + "right": "issue:7891", + "accept": false, + "reason": "The device-mismatch wording is similar, but the reports are too context-poor to prove the same concrete bug." + }, + { + "left": "issue:6086", + "right": "issue:6363", + "accept": true, + "reason": "Same exact FP16 unscale failure in the LoRA text-to-image training script family." + }, + { + "left": "issue:7261", + "right": "issue:7286", + "accept": true, + "reason": "Both are unsuccessful cross-attention weight loading reports for Custom Diffusion, with the second adding a pipeline variant." + }, + { + "left": "issue:1875", + "right": "issue:7854", + "accept": false, + "reason": "Both are half/float mismatch reports, but the surrounding pipelines differ and the bug is not concretely the same." + }, + { + "left": "issue:11436", + "right": "issue:6879", + "accept": false, + "reason": "Both are mat1/mat2 dtype mismatch errors, but one is half vs bfloat16 and the other is half vs float." + }, + { + "left": "issue:10958", + "right": "issue:12009", + "accept": false, + "reason": "Both mention meta-tensor loading problems, but the model format and loader path differ enough to keep them separate." + }, + { + "left": "issue:10958", + "right": "issue:2500", + "accept": false, + "reason": "Same meta-tensor symptom, but one is a specific pipeline loader report and the other is test failures; too broad to merge." + }, + { + "left": "issue:8863", + "right": "issue:9448", + "accept": false, + "reason": "The tuple.shape symptom is similar, but the evidence is only a generic exception string plus a broad pipeline family, so this is not confirmed duplicate-level evidence." + }, + { + "left": "issue:1875", + "right": "issue:5805", + "accept": false, + "reason": "Both are dtype-related, but the titles do not show the same concrete failure path." + }, + { + "left": "issue:6445", + "right": "issue:6821", + "accept": false, + "reason": "Both are caption_column validation issues, but the accepted values and dataset-schema contexts differ." + }, + { + "left": "issue:2780", + "right": "issue:7330", + "accept": true, + "reason": "Same exact Attempting to unscale FP16 gradients failure in the LoRA text-to-image training script family." + }, + { + "left": "issue:6832", + "right": "issue:8863", + "accept": false, + "reason": "This is only supported by a generic tuple.shape exception and broad pipeline-family similarity, which is not enough to confirm a duplicate." + }, + { + "left": "issue:7935", + "right": "issue:8465", + "accept": false, + "reason": "Both mention missing config.json, but they involve different pipelines and model-loading paths." + }, + { + "left": "issue:10520", + "right": "issue:7891", + "accept": false, + "reason": "Same device-mismatch wording, but the reports are too vague to prove the same code path." + }, + { + "left": "issue:6368", + "right": "issue:8581", + "accept": false, + "reason": "Both are unsupported-lora-layer errors, but they reference different layers and are not the same concrete bug." + }, + { + "left": "issue:10575", + "right": "issue:5749", + "accept": false, + "reason": "Both are index-out-of-bounds errors, but the dimensions and sizes differ and there is no clear shared failure path." + }, + { + "left": "issue:1433", + "right": "issue:3489", + "accept": true, + "reason": "Same OnnxRuntimeModel has no attribute config error; the second is a pipeline-specific duplicate of the first." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary is mostly grounded, but a few soft-edge accepts are too strong for the evidence in the packet. In particular, the tuple.shape links (issue:6832 \u2194 issue:8863 and issue:8863 \u2194 issue:9448) are only supported by a generic exception string plus a broad pipeline family, so treating them as confirmed duplicates overstates the packet. The LoRA FP16-unscale subgroup and the exact-message pairs (e.g. HFValidationError, torchvision::nms, OnnxRuntimeModel.config) are better supported. Because the output accepts some weakly grounded merges, I\u2019d reject for precision." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 3, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10520", + "issue:10722", + "issue:10896", + "issue:10897", + "issue:10952", + "issue:11436", + "issue:12436", + "issue:13261", + "issue:13311", + "issue:3357", + "issue:3651", + "issue:4410", + "issue:4752", + "issue:4932", + "issue:4933", + "issue:5022", + "issue:5025", + "issue:5399", + "issue:5510", + "issue:5603", + "issue:5661", + "issue:6124", + "issue:6278", + "issue:6363", + "issue:6482", + "issue:7003", + "issue:7330", + "issue:7708", + "issue:7978", + "issue:8518", + "issue:859", + "issue:8833", + "issue:9063", + "issue:9109", + "issue:9291", + "issue:9292", + "issue:9443", + "issue:9448", + "issue:9602", + "issue:9628", + "issue:9726", + "issue:9727", + "issue:9790", + "issue:9799", + "issue:9895", + "issue:9901", + "issue:9973" + ], + "soft_pairs": [ + "issue:10896|issue:10897", + "issue:9063|issue:9443", + "issue:4932|issue:4933", + "issue:3651|issue:6278", + "issue:9291|issue:9292", + "issue:5022|issue:5661", + "issue:10722|issue:9799", + "issue:9727|issue:9790", + "issue:6363|issue:6482", + "issue:10952|issue:9602", + "issue:9726|issue:9790", + "issue:3357|issue:8518", + "issue:7330|issue:9628", + "issue:6124|issue:7330", + "issue:9628|issue:9973", + "issue:9448|issue:9901", + "issue:7708|issue:8833", + "issue:10520|issue:9895", + "issue:13261|issue:13311", + "issue:11436|issue:5399", + "issue:7978|issue:9109", + "issue:5025|issue:5603", + "issue:12436|issue:859", + "issue:4752|issue:5510", + "issue:4410|issue:7003" + ], + "prepared_review_unit_hash": "607f7cefb8a8b1a8a7a41bf1917a6d6b32b91415c039c9b64fcb91cb7e765f87", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 34728, + "estimated_input_tokens": 8682, + "estimated_eval_tokens": 17620 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is heterogeneous overall, but many pairs are clear duplicate issues around the same error or request. I accepted only pairs that appear to describe the same underlying bug/feature and rejected pairs that merely share a subsystem or broad error class.", + "confidence": 0.89, + "canonical_issue_reason": "No single canonical issue covers the full set because the items split into many unrelated subclusters. The cleanest duplicate anchor is issue 10897, which is an exact same-title match to 10896 and remains open.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10897 is the best issue to keep as the canonical representative for its subcluster: it is an exact duplicate of 10896, has the same request text, and is open.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10896", + "right": "issue:10897", + "accept": true, + "reason": "Exact same title and same feature request; these are duplicates." + }, + { + "left": "issue:9063", + "right": "issue:9443", + "accept": false, + "reason": "Same error string, but different pipelines/models; likely separate tokenizer-loading cases." + }, + { + "left": "issue:4932", + "right": "issue:4933", + "accept": true, + "reason": "Same DreamBooth output complaint with near-identical wording; duplicate reports." + }, + { + "left": "issue:3651", + "right": "issue:6278", + "accept": false, + "reason": "Both mention multi-GPU, but they affect different training scripts and likely different failure points." + }, + { + "left": "issue:9291", + "right": "issue:9292", + "accept": true, + "reason": "Both report FLUX LoRA key incompatibility with diffusers loaders; same underlying compatibility bug." + }, + { + "left": "issue:5022", + "right": "issue:5661", + "accept": true, + "reason": "Both are the same missing PreTrainedTokenizer import/name error in diffusers." + }, + { + "left": "issue:10722", + "right": "issue:9799", + "accept": false, + "reason": "Both are tensor-size mismatches, but the actual dimensions and training contexts differ." + }, + { + "left": "issue:9727", + "right": "issue:9790", + "accept": false, + "reason": "One is a save problem and the other is a training problem; not clearly the same bug." + }, + { + "left": "issue:6363", + "right": "issue:6482", + "accept": true, + "reason": "Same FP16 unscale error in closely related DreamBooth/LoRA training paths." + }, + { + "left": "issue:10952", + "right": "issue:9602", + "accept": true, + "reason": "Both are adapter-name lookup failures during LoRA loading; same adapter handling issue." + }, + { + "left": "issue:9726", + "right": "issue:9790", + "accept": true, + "reason": "Same FLUX.1-dev DreamBooth multigpu training failure theme and likely same code path." + }, + { + "left": "issue:3357", + "right": "issue:8518", + "accept": false, + "reason": "Both are torch.compile errors, but they are different unsupported operations and different stack paths." + }, + { + "left": "issue:7330", + "right": "issue:9628", + "accept": true, + "reason": "Same \"Attempting to unscale FP16 gradients\" training failure across related scripts." + }, + { + "left": "issue:6124", + "right": "issue:7330", + "accept": true, + "reason": "Same FP16 gradient unscale error in DreamBooth/LoRA training flows." + }, + { + "left": "issue:9628", + "right": "issue:9973", + "accept": true, + "reason": "Both report the same FP16 unscale failure in DreamBooth LoRA SD3 training." + }, + { + "left": "issue:9448", + "right": "issue:9901", + "accept": true, + "reason": "Both describe the same tuple-vs-shape attention/processor bug path." + }, + { + "left": "issue:7708", + "right": "issue:8833", + "accept": false, + "reason": "Both are missing-file model loading errors, but they concern different repos and different missing files." + }, + { + "left": "issue:10520", + "right": "issue:9895", + "accept": false, + "reason": "Same device-mismatch symptom, but different models and no evidence they share the same fix." + }, + { + "left": "issue:13261", + "right": "issue:13311", + "accept": true, + "reason": "Both are FLUX.2-klein-9B LoRA compatibility failures in the same transformer/LoRA loading path." + }, + { + "left": "issue:11436", + "right": "issue:5399", + "accept": false, + "reason": "Both are dtype mismatches, but they involve different pipelines and are too generic to merge confidently." + }, + { + "left": "issue:7978", + "right": "issue:9109", + "accept": true, + "reason": "Exact same torchvision::nms missing-operator error." + }, + { + "left": "issue:5025", + "right": "issue:5603", + "accept": false, + "reason": "Both are import errors from diffusers.utils, but they are different missing symbols and likely separate breakages." + }, + { + "left": "issue:12436", + "right": "issue:859", + "accept": false, + "reason": "Both are unexpected-argument constructor errors, but for different classes and different arguments." + }, + { + "left": "issue:4752", + "right": "issue:5510", + "accept": false, + "reason": "Both are dtype mismatches in SDXL-family pipelines, but the concrete pipelines and fixes are different." + }, + { + "left": "issue:4410", + "right": "issue:7003", + "accept": false, + "reason": "Both involve weight-shape mismatches, but they are different in/out channel expectations and different loading scenarios." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The canonical anchor choice is supported by the identical SkyReels-A1 titles and open state on 10897, and the accepted soft-edge pairs all have reasonably close title/body alignment rather than just broad subsystem similarity. The rejected pairs are also appropriately cautious. No clear overstatement of evidence stood out." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 4, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10154", + "issue:10343", + "issue:10467", + "issue:10928", + "issue:11218", + "issue:11410", + "issue:11417", + "issue:11702", + "issue:11709", + "issue:11849", + "issue:11865", + "issue:11901", + "issue:12104", + "issue:12174", + "issue:12778", + "issue:13261", + "issue:13350", + "issue:1925", + "issue:1931", + "issue:1946", + "issue:1976", + "issue:2100", + "issue:2947", + "issue:3325", + "issue:3481", + "issue:3829", + "issue:4265", + "issue:4752", + "issue:4868", + "issue:5259", + "issue:5510", + "issue:5805", + "issue:6139", + "issue:6363", + "issue:6821", + "issue:6914", + "issue:7049", + "issue:720", + "issue:7210", + "issue:7708", + "issue:7851", + "issue:7854", + "issue:8759", + "issue:8975", + "issue:9210", + "issue:9628", + "issue:9690", + "issue:9901" + ], + "soft_pairs": [ + "issue:10154|issue:7210", + "issue:1931|issue:720", + "issue:4752|issue:7854", + "issue:11417|issue:12778", + "issue:1946|issue:3325", + "issue:6363|issue:9628", + "issue:6914|issue:9901", + "issue:13261|issue:13350", + "issue:1925|issue:1931", + "issue:2947|issue:6139", + "issue:10928|issue:5259", + "issue:10343|issue:8759", + "issue:5510|issue:5805", + "issue:4752|issue:5805", + "issue:12104|issue:7851", + "issue:10467|issue:9690", + "issue:6821|issue:9210", + "issue:12174|issue:4868", + "issue:3481|issue:3829", + "issue:11417|issue:11865", + "issue:11218|issue:8975", + "issue:11702|issue:11709", + "issue:1976|issue:7708", + "issue:2100|issue:4265", + "issue:11410|issue:7049", + "issue:11849|issue:11901" + ], + "prepared_review_unit_hash": "73578b402a4ad4b2104ce6c628216c4a575d394ab066dcd47fccc3a7c973f6e2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35606, + "estimated_input_tokens": 8902, + "estimated_eval_tokens": 18060 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is not one tight duplicate cluster; it contains several separate bug families. Clear duplicate-like pairs mostly involve repeated dtype-mismatch errors, repeated missing-file/import lookups, repeated attribute errors, and repeated LoRA key-loading failures. Several other soft pairs are only broadly similar and should stay separate.", + "confidence": 0.84, + "canonical_issue_reason": "No single canonical issue covers the whole set; the items split into multiple unrelated bug families. If one representative must be chosen, issue 1931 best captures the recurring half-precision dtype-mismatch subgroup.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1931 is the strongest overall representative because it is a clear, recurring half-precision/float dtype-mismatch report and aligns with other near-duplicate reports in that subgroup.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10154", + "right": "issue:7210", + "accept": false, + "reason": "Both are AttributeError reports, but they hit different APIs and code paths: load_lora_weights vs accelerate.load_checkpoint_and_dispatch. Not the same bug." + }, + { + "left": "issue:1931", + "right": "issue:720", + "accept": true, + "reason": "Same underlying float16/Half vs Float mismatch during Stable Diffusion inference; the symptom and fix path are effectively the same." + }, + { + "left": "issue:4752", + "right": "issue:7854", + "accept": true, + "reason": "These are the same input/bias dtype mismatch error in SDXL code, with nearly identical wording and likely the same fix." + }, + { + "left": "issue:11417", + "right": "issue:12778", + "accept": false, + "reason": "Both are missing-attribute errors, but on different objects and different failures. Too little evidence of one concrete shared bug." + }, + { + "left": "issue:1946", + "right": "issue:3325", + "accept": false, + "reason": "Both mention CUDA invalid argument, but one is xformers-related and the other is train_dreambooth.py. Same broad symptom, not the same bug." + }, + { + "left": "issue:6363", + "right": "issue:9628", + "accept": false, + "reason": "Same error text, but different training scripts and model setups. Insufficient evidence that one code-path fix would address both." + }, + { + "left": "issue:6914", + "right": "issue:9901", + "accept": false, + "reason": "Both are tuple/shape-related AttributeErrors, but they occur in different contexts (IP-Adapter/cpu offload vs AttnProcessor) and do not clearly share one root cause." + }, + { + "left": "issue:13261", + "right": "issue:13350", + "accept": false, + "reason": "One is a LoRA/lokr loading failure; the other is a batch inference usage question. Not the same underlying bug." + }, + { + "left": "issue:1925", + "right": "issue:1931", + "accept": true, + "reason": "Both describe the same half-precision/float mismatch class of bug in Stable Diffusion inference." + }, + { + "left": "issue:2947", + "right": "issue:6139", + "accept": false, + "reason": "These are different missing attributes on UNet2DConditionModel, so they are not the same concrete defect." + }, + { + "left": "issue:10928", + "right": "issue:5259", + "accept": false, + "reason": "Different missing-file errors: fp16 weight file vs model_index.json. Separate problems." + }, + { + "left": "issue:10343", + "right": "issue:8759", + "accept": true, + "reason": "Same invalid-cast warning in image postprocessing leading to black images; the symptom and code area match closely." + }, + { + "left": "issue:5510", + "right": "issue:5805", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different pipelines/platforms and the exact failure context differs too much to merge confidently." + }, + { + "left": "issue:4752", + "right": "issue:5805", + "accept": false, + "reason": "Same broad dtype mismatch theme, but one is an SDXL pipeline bug and the other is an XLA-specific issue with opposite type ordering. Not a clean duplicate." + }, + { + "left": "issue:12104", + "right": "issue:7851", + "accept": false, + "reason": "Both are index-out-of-bounds errors, but one is size 0 and the other is a concurrency-related size 11 case. Different bugs." + }, + { + "left": "issue:10467", + "right": "issue:9690", + "accept": false, + "reason": "One is about a generated temp filename in FP8 example output; the other is a cleanup request. Not the same issue." + }, + { + "left": "issue:6821", + "right": "issue:9210", + "accept": false, + "reason": "These are opposite dataset-column validation errors (caption_column vs image_column) and do not indicate the same defect." + }, + { + "left": "issue:12174", + "right": "issue:4868", + "accept": true, + "reason": "Same AttributeError on __getattr__ with the same underlying failure signature." + }, + { + "left": "issue:3481", + "right": "issue:3829", + "accept": false, + "reason": "Both discuss image differences across inference settings, but they involve different schedulers/pipelines and are too broad to treat as one bug." + }, + { + "left": "issue:11417", + "right": "issue:11865", + "accept": false, + "reason": "Completely different attribute errors on different objects; no shared code-path is apparent." + }, + { + "left": "issue:11218", + "right": "issue:8975", + "accept": false, + "reason": "Both are import errors from diffusers.loaders, but they concern different symbols and likely different missing exports." + }, + { + "left": "issue:11702", + "right": "issue:11709", + "accept": true, + "reason": "Both are LoRA key-loading failures in diffusers for closely related video-model use cases; a single loader fix could plausibly address both." + }, + { + "left": "issue:1976", + "right": "issue:7708", + "accept": true, + "reason": "Same incorrect tokenizer/config.json file lookup against a model repo that does not provide it." + }, + { + "left": "issue:2100", + "right": "issue:4265", + "accept": false, + "reason": "Different missing imports for different pipelines/classes; same category, but not the same concrete bug." + }, + { + "left": "issue:11410", + "right": "issue:7049", + "accept": true, + "reason": "Exact same too-many-values-to-unpack error in the attention/IP-adapter code path." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The accepted soft-edge pairs are mostly supported by near-identical error signatures/titles, and the rejected pairs avoid overclaiming when only broad symptom overlap is present. No clear overstatement stands out." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 5, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10154", + "issue:10472", + "issue:11218", + "issue:12174", + "issue:12436", + "issue:12480", + "issue:12481", + "issue:12482", + "issue:12719", + "issue:12722", + "issue:12895", + "issue:1709", + "issue:2044", + "issue:2402", + "issue:246", + "issue:2696", + "issue:3816", + "issue:4104", + "issue:4198", + "issue:4325", + "issue:4456", + "issue:4478", + "issue:4557", + "issue:4619", + "issue:4752", + "issue:5171", + "issue:5399", + "issue:5510", + "issue:5518", + "issue:5596", + "issue:5805", + "issue:6444", + "issue:6734", + "issue:6832", + "issue:6858", + "issue:6879", + "issue:7506", + "issue:7516", + "issue:7572", + "issue:7854", + "issue:7922", + "issue:8463", + "issue:8556", + "issue:8578", + "issue:9112", + "issue:9448", + "issue:9645", + "issue:9799" + ], + "soft_pairs": [ + "issue:2402|issue:2696", + "issue:12174|issue:4557", + "issue:246|issue:3816", + "issue:4619|issue:7854", + "issue:1709|issue:2044", + "issue:4198|issue:5399", + "issue:5399|issue:6879", + "issue:4104|issue:4325", + "issue:9112|issue:9799", + "issue:4619|issue:5510", + "issue:6734|issue:8463", + "issue:7516|issue:7922", + "issue:11218|issue:9645", + "issue:8556|issue:8578", + "issue:7506|issue:8578", + "issue:4456|issue:5518", + "issue:5596|issue:6858", + "issue:4478|issue:5805", + "issue:4752|issue:7572", + "issue:12895|issue:8463", + "issue:10472|issue:12436", + "issue:10154|issue:6444", + "issue:12436|issue:5171", + "issue:6832|issue:9448", + "issue:12719|issue:12722", + "issue:12481|issue:12482", + "issue:12480|issue:12482" + ], + "prepared_review_unit_hash": "69f6c3b012e089969579c58875f2c9d8070b05eafba8e2ed3f84f646c5061c15", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36285, + "estimated_input_tokens": 9072, + "estimated_eval_tokens": 18400 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous, but there are a few clear duplicate mini-clusters: CUDA-not-enabled, super.__getattr__, black-image/cast warning, NoneType.tokenize, meta-tensor copy, dtype mismatch in SDXL, and the offload_state_dict CLIPTextModel reports. The strongest duplicate group is the offload_state_dict trio.", + "confidence": 0.82, + "canonical_issue_reason": "issue:12482 has the cleanest, most direct title for the CLIPTextModel offload_state_dict bug, while 12480/12481 are near-identical noisy variants of the same report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:12482 is the best representative issue in this set because it captures the core error succinctly and sits in the tightest duplicate cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2402", + "right": "issue:2696", + "accept": true, + "reason": "Same underlying runtime failure: Torch was built without CUDA support." + }, + { + "left": "issue:12174", + "right": "issue:4557", + "accept": true, + "reason": "Exact same AttributeError on super.__getattr__; clearly the same bug report." + }, + { + "left": "issue:246", + "right": "issue:3816", + "accept": false, + "reason": "Both are missing-transformers-symbol errors, but they involve different attributes and likely different breakages." + }, + { + "left": "issue:4619", + "right": "issue:7854", + "accept": true, + "reason": "Same concrete dtype mismatch error in the diffusion model path; likely duplicate reports of the same bug." + }, + { + "left": "issue:1709", + "right": "issue:2044", + "accept": false, + "reason": "Different AttributeErrors in different code paths; only superficially similar." + }, + { + "left": "issue:4198", + "right": "issue:5399", + "accept": false, + "reason": "Both mention mat1/mat2 dtype issues, but the described scenarios and code paths are too different to treat as one bug." + }, + { + "left": "issue:5399", + "right": "issue:6879", + "accept": false, + "reason": "Same broad dtype-mismatch theme, but different scenarios and no evidence of the same concrete defect." + }, + { + "left": "issue:4104", + "right": "issue:4325", + "accept": true, + "reason": "Black-image output with invalid-value-encountered-in-cast is the same failure mode." + }, + { + "left": "issue:9112", + "right": "issue:9799", + "accept": false, + "reason": "Both are tensor-size mismatch errors, but the shapes differ substantially and suggest different underlying bugs." + }, + { + "left": "issue:4619", + "right": "issue:5510", + "accept": false, + "reason": "Same class of dtype mismatch, but different pipelines/code paths and not enough evidence of one shared bug." + }, + { + "left": "issue:6734", + "right": "issue:8463", + "accept": true, + "reason": "Same NoneType.tokenize failure in the InstructP2P/SDXL path; this looks like the same bug." + }, + { + "left": "issue:7516", + "right": "issue:7922", + "accept": false, + "reason": "Both are missing diffusers attributes, but they concern different pipelines and likely different version-gated symbols." + }, + { + "left": "issue:11218", + "right": "issue:9645", + "accept": false, + "reason": "Both are import failures from diffusers.loaders, but the missing mixins are different and not clearly the same defect." + }, + { + "left": "issue:8556", + "right": "issue:8578", + "accept": true, + "reason": "Same meta-tensor copy failure during model loading; the model names differ, but the code-path problem is the same." + }, + { + "left": "issue:7506", + "right": "issue:8578", + "accept": false, + "reason": "Both mention meta tensors, but the reported loading contexts are different enough that this is not clearly the same bug." + }, + { + "left": "issue:4456", + "right": "issue:5518", + "accept": false, + "reason": "Both involve LoRA and ModuleList, but the failing attribute access differs and the underlying bugs are not clearly the same." + }, + { + "left": "issue:5596", + "right": "issue:6858", + "accept": false, + "reason": "Both are convert() type errors, but the wrong input types and training contexts differ." + }, + { + "left": "issue:4478", + "right": "issue:5805", + "accept": false, + "reason": "Same general dtype mismatch theme, but one is SDXL from_pretrained and the other is XLA LCM-SDXL; not the same concrete bug." + }, + { + "left": "issue:4752", + "right": "issue:7572", + "accept": true, + "reason": "Same exact input/bias dtype mismatch error in Stable Diffusion XL; these look like duplicate reports." + }, + { + "left": "issue:12895", + "right": "issue:8463", + "accept": false, + "reason": "Both have AttributeError wording, but the failure modes and contexts are unrelated." + }, + { + "left": "issue:12481", + "right": "issue:12482", + "accept": true, + "reason": "Near-exact duplicate offload_state_dict CLIPTextModel TypeError report." + }, + { + "left": "issue:12480", + "right": "issue:12482", + "accept": true, + "reason": "Same offload_state_dict CLIPTextModel TypeError; 12480 is just a noisier variant of 12482." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the canonical/best-issue choice for the offload_state_dict trio is well supported by the near-identical titles. The accepted soft-edge pairs are conservative enough based on the issue titles alone, and the rejected pairs avoid overclaiming where the evidence is only thematic." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 6, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10453", + "issue:10542", + "issue:10910", + "issue:11218", + "issue:11435", + "issue:12147", + "issue:12535", + "issue:12609", + "issue:12612", + "issue:12778", + "issue:12779", + "issue:13311", + "issue:13350", + "issue:1399", + "issue:2294", + "issue:2780", + "issue:3010", + "issue:3017", + "issue:3140", + "issue:4224", + "issue:4409", + "issue:4529", + "issue:4615", + "issue:4837", + "issue:4868", + "issue:5054", + "issue:539", + "issue:5737", + "issue:5992", + "issue:5998", + "issue:6079", + "issue:6082", + "issue:6098", + "issue:6124", + "issue:6324", + "issue:6363", + "issue:6442", + "issue:6454", + "issue:6482", + "issue:6832", + "issue:6914", + "issue:7210", + "issue:8182", + "issue:8975", + "issue:9237", + "issue:9548", + "issue:9970", + "issue:9973" + ], + "soft_pairs": [ + "issue:12609|issue:12612", + "issue:10453|issue:10542", + "issue:4224|issue:4409", + "issue:10910|issue:6454", + "issue:12778|issue:12779", + "issue:6363|issue:6454", + "issue:9237|issue:9548", + "issue:3010|issue:3017", + "issue:4529|issue:539", + "issue:9970|issue:9973", + "issue:11435|issue:4615", + "issue:4837|issue:5737", + "issue:10910|issue:6363", + "issue:11435|issue:8975", + "issue:2780|issue:6454", + "issue:6079|issue:6082", + "issue:2294|issue:4615", + "issue:13311|issue:13350", + "issue:6832|issue:6914", + "issue:5054|issue:7210", + "issue:3140|issue:4615", + "issue:11218|issue:4615", + "issue:6098|issue:6482", + "issue:6124|issue:6442", + "issue:1399|issue:5998", + "issue:12147|issue:12535", + "issue:4868|issue:6324", + "issue:4868|issue:8182", + "issue:5992|issue:6324" + ], + "prepared_review_unit_hash": "3956d44b6e51cf614bf31430d31769dbd7ca253143c224f3ff7755b21c5f9a25", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 35765, + "estimated_input_tokens": 8942, + "estimated_eval_tokens": 18140 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Use issue:6454 as the canonical anchor for the recurring FP16-unscale reports, but apply a strict title-only policy for soft edges. Most proposed duplicate links are too weakly evidenced and should be rejected; only the 9237/9548 pair has explicit duplicate-style evidence.", + "confidence": 0.79, + "canonical_issue_reason": "issue:6454 is the most generic FP16-unscale report in the set and is the best anchor for that recurring family.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6454 is the best single issue representative because it is the broadest, least script-specific FP16-unscale report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12609", + "right": "issue:12612", + "accept": false, + "reason": "Same broad WAN NaN wording, but the packet provides only title similarity and no concrete shared repro/root-cause evidence." + }, + { + "left": "issue:10453", + "right": "issue:10542", + "accept": false, + "reason": "Both are Hunyuan Video batch-size complaints, but the recurrence claim is only in the title and is not enough to mark a duplicate." + }, + { + "left": "issue:4224", + "right": "issue:4409", + "accept": false, + "reason": "Different missing attributes (`control_model` vs `alphas_cumprod`); related area, not the same bug." + }, + { + "left": "issue:10910", + "right": "issue:6454", + "accept": false, + "reason": "FP16-unscale title match alone is too broad; no body-level evidence that these are the same concrete failure." + }, + { + "left": "issue:12778", + "right": "issue:12779", + "accept": false, + "reason": "Different Flux2 failures in different code paths, not the same underlying issue." + }, + { + "left": "issue:6363", + "right": "issue:6454", + "accept": false, + "reason": "Both mention FP16-unscale, but the packet gives no direct evidence they are the same bug beyond the shared error text." + }, + { + "left": "issue:9237", + "right": "issue:9548", + "accept": true, + "reason": "Accept: #9548 explicitly says it is still the issue on #9237, which is direct duplicate-style evidence." + }, + { + "left": "issue:3010", + "right": "issue:3017", + "accept": false, + "reason": "Both are docs requests about multi-GPU inference, but there is no explicit link or enough evidence to treat them as the same request." + }, + { + "left": "issue:4529", + "right": "issue:539", + "accept": false, + "reason": "Unrelated keyword-argument failures in different APIs." + }, + { + "left": "issue:9970", + "right": "issue:9973", + "accept": false, + "reason": "Same general DreamBooth SDXL LoRA area, but only title similarity is available and that is not enough for a duplicate call." + }, + { + "left": "issue:11435", + "right": "issue:4615", + "accept": false, + "reason": "Different missing imports/symbols; not the same code-path failure." + }, + { + "left": "issue:4837", + "right": "issue:5737", + "accept": false, + "reason": "Both concern SDXL conversion, but the packet does not show enough shared evidence to confirm the same bug." + }, + { + "left": "issue:10910", + "right": "issue:6363", + "accept": false, + "reason": "FP16-unscale phrasing alone is insufficient to conclude the same underlying defect." + }, + { + "left": "issue:11435", + "right": "issue:8975", + "accept": false, + "reason": "Different missing import targets, so not duplicates." + }, + { + "left": "issue:2780", + "right": "issue:6454", + "accept": false, + "reason": "Specific training script report vs generic issue, but no concrete evidence they are the same failure." + }, + { + "left": "issue:6079", + "right": "issue:6082", + "accept": false, + "reason": "Very similar half/float error wording, but no body evidence that the two reports share the same root cause." + }, + { + "left": "issue:2294", + "right": "issue:4615", + "accept": false, + "reason": "Different import targets in different pipelines; too different to merge." + }, + { + "left": "issue:13311", + "right": "issue:13350", + "accept": false, + "reason": "One is about missing LoRA keys, the other about batch inference usage; not the same bug." + }, + { + "left": "issue:6832", + "right": "issue:6914", + "accept": false, + "reason": "Both are IP-Adapter-FaceID shape-related, but the packet does not provide enough shared reproduction detail to confirm a duplicate." + }, + { + "left": "issue:5054", + "right": "issue:7210", + "accept": false, + "reason": "Different attribute errors in different offload/load paths; related subsystem, not the same bug." + }, + { + "left": "issue:3140", + "right": "issue:4615", + "accept": false, + "reason": "Import-related, but the packet does not establish that these are the same concrete missing-symbol breakage." + }, + { + "left": "issue:11218", + "right": "issue:4615", + "accept": false, + "reason": "Different missing imports; not the same issue." + }, + { + "left": "issue:6098", + "right": "issue:6482", + "accept": false, + "reason": "Both mention FP16-unscale, but the resume-from-checkpoint trigger is not enough evidence to merge them." + }, + { + "left": "issue:6124", + "right": "issue:6442", + "accept": false, + "reason": "Same generic error text, but no direct evidence beyond title similarity." + }, + { + "left": "issue:1399", + "right": "issue:5998", + "accept": false, + "reason": "Both are VersatileDiffusion signature mismatches, but the differing kwargs mean the packet does not prove the same bug." + }, + { + "left": "issue:12147", + "right": "issue:12535", + "accept": false, + "reason": "Similar Wan2.2 Lightning LoRA loading complaints, but no explicit duplicate evidence and different contexts in the titles." + }, + { + "left": "issue:4868", + "right": "issue:6324", + "accept": false, + "reason": "Different attribute errors with different missing members." + }, + { + "left": "issue:4868", + "right": "issue:8182", + "accept": false, + "reason": "Completely different error messages; not duplicates." + }, + { + "left": "issue:5992", + "right": "issue:6324", + "accept": false, + "reason": "Different root causes and different failing symbols; not the same bug." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The canonical anchor around issue:6454 for the recurring FP16-unscale family is plausible, but the soft-edge handling is not conservative enough. Several verdicts are asserted as same-bug duplicates based only on title similarity, while the packet provides no direct evidence beyond that. In particular, claims like issue:10453/issue:10542 being \"clearly a recurrence,\" or multiple FP16-unscale pairs being the same bug, overstate certainty. Given the evaluator\u2019s conservative standard, this should be rejected." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 7, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10281", + "issue:10594", + "issue:10910", + "issue:11417", + "issue:11435", + "issue:11464", + "issue:1466", + "issue:1544", + "issue:1709", + "issue:1817", + "issue:1866", + "issue:1867", + "issue:2041", + "issue:2044", + "issue:2294", + "issue:2780", + "issue:384", + "issue:3973", + "issue:4186", + "issue:4478", + "issue:4529", + "issue:4615", + "issue:4724", + "issue:4868", + "issue:4969", + "issue:5025", + "issue:539", + "issue:5510", + "issue:5572", + "issue:5576", + "issue:5578", + "issue:6098", + "issue:6124", + "issue:6363", + "issue:6454", + "issue:6482", + "issue:6561", + "issue:6636", + "issue:6842", + "issue:7330", + "issue:8182", + "issue:8637", + "issue:8871", + "issue:8975", + "issue:9628", + "issue:9645", + "issue:9973" + ], + "soft_pairs": [ + "issue:3973|issue:4529", + "issue:4969|issue:8637", + "issue:10281|issue:5576", + "issue:10594|issue:11464", + "issue:1866|issue:1867", + "issue:6842|issue:8637", + "issue:4615|issue:8975", + "issue:8637|issue:9645", + "issue:10910|issue:6098", + "issue:1709|issue:4868", + "issue:11435|issue:2294", + "issue:4478|issue:5510", + "issue:1544|issue:2044", + "issue:6124|issue:9628", + "issue:6098|issue:6454", + "issue:384|issue:5572", + "issue:1466|issue:2041", + "issue:5025|issue:8637", + "issue:5578|issue:6561", + "issue:7330|issue:8871", + "issue:1817|issue:4529", + "issue:10910|issue:6482", + "issue:4969|issue:5025", + "issue:4186|issue:4724", + "issue:10910|issue:2780", + "issue:6098|issue:6363", + "issue:6124|issue:9973", + "issue:5025|issue:6636", + "issue:1817|issue:539", + "issue:11417|issue:8182", + "issue:6098|issue:6124" + ], + "prepared_review_unit_hash": "6c7c3989b5921fff2a044b0dc4e3d97b5c61abb4b24ed3fd34ff8c984ac20528", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36268, + "estimated_input_tokens": 9067, + "estimated_eval_tokens": 18390 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a mixed issue cluster with one dominant duplicate family around \"ValueError: Attempting to unscale FP16 gradients\" in LoRA/DreamBooth training, plus a smaller img2img low-strength pair. Several other pairs are only superficially similar (same subsystem/error class) and should not be merged.", + "confidence": 0.88, + "canonical_issue_reason": "issue:6363 is the most representative FP16-unscale report: exact error title, substantial discussion, inbound references, and an explicit duplicate reference to issue:8871; it best anchors the repeated LoRA/DreamBooth training bug reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6363 has the clearest, most central description of the recurring training failure and appears to be the main hub for related reports across scripts and variants.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3973", + "right": "issue:4529", + "accept": false, + "reason": "Both are import/signature errors, but they affect different functions ('download_from_original_stable_diffusion_ckpt' vs 'load_lora_into_unet') and are not the same bug." + }, + { + "left": "issue:4969", + "right": "issue:8637", + "accept": false, + "reason": "Different missing symbols from different areas ('CrossAttention' vs 'PixArtSigmaPipeline'); same error class, not the same underlying issue." + }, + { + "left": "issue:10281", + "right": "issue:5576", + "accept": false, + "reason": "Both are scheduler feature requests, but FreeScale and FreeNoise are distinct proposals and should not be merged." + }, + { + "left": "issue:10594", + "right": "issue:11464", + "accept": false, + "reason": "Both involve FLUX controls/LoRA, but one is about multiple controls on FLUX.1-dev and the other about loading a specific FLUX Canny LoRA; too different to treat as duplicates." + }, + { + "left": "issue:1866", + "right": "issue:1867", + "accept": true, + "reason": "Both describe img2img failing at very low strength values; one names a scheduler-specific manifestation, the other generalizes the same low-strength pipeline bug." + }, + { + "left": "issue:6842", + "right": "issue:8637", + "accept": false, + "reason": "Both are import errors, but for different pipelines ('Prompt2PromptPipeline' vs 'PixArtSigmaPipeline'); not the same change or bug." + }, + { + "left": "issue:4615", + "right": "issue:8975", + "accept": false, + "reason": "These are separate missing-import reports for different loader/pipeline classes and do not share the same code-path failure." + }, + { + "left": "issue:8637", + "right": "issue:9645", + "accept": false, + "reason": "Different missing imports from different modules; they are not the same underlying issue." + }, + { + "left": "issue:10910", + "right": "issue:6098", + "accept": true, + "reason": "Same exact FP16-unscale training error; these are duplicate reports of the same bug." + }, + { + "left": "issue:1709", + "right": "issue:4868", + "accept": false, + "reason": "Both are AttributeErrors, but in different pipelines and with unrelated missing attributes ('astype' vs '__getattr__')." + }, + { + "left": "issue:11435", + "right": "issue:2294", + "accept": false, + "reason": "Both are import errors, but for different pipeline names and unrelated missing symbols." + }, + { + "left": "issue:4478", + "right": "issue:5510", + "accept": true, + "reason": "Both report the same SDXL dtype mismatch error ('Input type and bias type should be the same') across closely related pipelines, so they likely share the same fix." + }, + { + "left": "issue:1544", + "right": "issue:2044", + "accept": false, + "reason": "Different training/runtime failures: missing 'get_velocity' on the scheduler vs DDP wrapper missing 'dtype'." + }, + { + "left": "issue:6124", + "right": "issue:9628", + "accept": true, + "reason": "Same FP16-unscale error in DreamBooth/LoRA training variants; these are the same underlying bug." + }, + { + "left": "issue:6098", + "right": "issue:6454", + "accept": true, + "reason": "Exact same error message and training failure; clear duplicate." + }, + { + "left": "issue:384", + "right": "issue:5572", + "accept": false, + "reason": "Both involve missing VAE-related attributes, but the missing fields and contexts differ enough that they are not the same bug." + }, + { + "left": "issue:1466", + "right": "issue:2041", + "accept": true, + "reason": "Both report the Dreambooth example being incompatible with Windows/Mac; the later issue is a narrower version of the same cross-platform problem." + }, + { + "left": "issue:5025", + "right": "issue:8637", + "accept": false, + "reason": "Different missing imports ('randn_tensor' vs 'PixArtSigmaPipeline'); same import-error family, not duplicate." + }, + { + "left": "issue:5578", + "right": "issue:6561", + "accept": false, + "reason": "Both are model-loading OSErrors, but they point to different missing files in different models, so they are not the same issue." + }, + { + "left": "issue:7330", + "right": "issue:8871", + "accept": true, + "reason": "Same FP16-unscale failure in LoRA training; one is a specific example script, the other a broader follow-up." + }, + { + "left": "issue:1817", + "right": "issue:4529", + "accept": false, + "reason": "Unrelated API-signature issues in different utilities/mixins; not the same underlying change." + }, + { + "left": "issue:10910", + "right": "issue:6482", + "accept": true, + "reason": "Same FP16-unscale bug, with 6482 being a DreamBooth LoRA SDXL variant of the same training failure." + }, + { + "left": "issue:4969", + "right": "issue:5025", + "accept": false, + "reason": "Both are import problems, but they concern different symbols and unrelated code paths." + }, + { + "left": "issue:4186", + "right": "issue:4724", + "accept": false, + "reason": "Both are SDXL pipeline behavior reports, but one is about inpaint+refiner and the other about img2img with a base checkpoint; too different to merge." + }, + { + "left": "issue:10910", + "right": "issue:2780", + "accept": true, + "reason": "Same FP16-unscale training bug; 2780 is just another script-specific report." + }, + { + "left": "issue:6098", + "right": "issue:6363", + "accept": true, + "reason": "Same recurring FP16-unscale error; 6363 is essentially the same bug in a more specific script context." + }, + { + "left": "issue:6124", + "right": "issue:9973", + "accept": true, + "reason": "Same FP16-unscale failure in SDXL DreamBooth/LoRA training scripts." + }, + { + "left": "issue:5025", + "right": "issue:6636", + "accept": false, + "reason": "Different missing imports from different modules; they are not the same bug." + }, + { + "left": "issue:1817", + "right": "issue:539", + "accept": false, + "reason": "Different API signature regressions in unrelated helper functions." + }, + { + "left": "issue:11417", + "right": "issue:8182", + "accept": false, + "reason": "Completely different AttributeErrors in unrelated objects ('DistributedDataParallel.dtype' vs logger typo), so not duplicates." + }, + { + "left": "issue:6098", + "right": "issue:6124", + "accept": true, + "reason": "Same FP16-unscale training failure, just reported through a different notebook/script." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The dominant FP16-unscale family is supported by the titles and the one explicit duplicate reference (6363 \u2194 8871), and the smaller img2img pair is also defensible. The rejected soft pairs are generally framed by clear title-level differences, so I don\u2019t see overstatement that would justify rejection." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 8, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10722", + "issue:12436", + "issue:12482", + "issue:13137", + "issue:13261", + "issue:2044", + "issue:2132", + "issue:2294", + "issue:2669", + "issue:2780", + "issue:2947", + "issue:3040", + "issue:3787", + "issue:4118", + "issue:4232", + "issue:4234", + "issue:4265", + "issue:4557", + "issue:4615", + "issue:4619", + "issue:4969", + "issue:5012", + "issue:5016", + "issue:5029", + "issue:5075", + "issue:5572", + "issue:5603", + "issue:5712", + "issue:5992", + "issue:6124", + "issue:6126", + "issue:6324", + "issue:6353", + "issue:6371", + "issue:6442", + "issue:6482", + "issue:6636", + "issue:7239", + "issue:7516", + "issue:7681", + "issue:8182", + "issue:8320", + "issue:858", + "issue:8871", + "issue:8915", + "issue:9112", + "issue:9248", + "issue:9476" + ], + "soft_pairs": [ + "issue:5075|issue:7239", + "issue:6442|issue:8871", + "issue:4557|issue:5029", + "issue:6353|issue:6371", + "issue:4232|issue:4234", + "issue:13137|issue:13261", + "issue:12436|issue:12482", + "issue:5712|issue:9248", + "issue:2294|issue:3787", + "issue:4615|issue:5603", + "issue:10722|issue:9112", + "issue:4619|issue:9476", + "issue:3040|issue:5992", + "issue:4557|issue:858", + "issue:2044|issue:2669", + "issue:2780|issue:6482", + "issue:5012|issue:5016", + "issue:7516|issue:7681", + "issue:6124|issue:8871", + "issue:2044|issue:5572", + "issue:6324|issue:8182", + "issue:2132|issue:2669", + "issue:4265|issue:8915", + "issue:4118|issue:8320", + "issue:5603|issue:6636", + "issue:4969|issue:5603", + "issue:2947|issue:6126" + ], + "prepared_review_unit_hash": "47e9b2c76d107d685e3e0503811621195cae0c6ba28adef8017000913a377e5c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35818, + "estimated_input_tokens": 8955, + "estimated_eval_tokens": 18166 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous cluster of unrelated diffusers issues, with only a few true duplicate families. The strongest duplicate thread is the repeated \"Attempting to unscale FP16 gradients\" LoRA training failures; a separate exact kwarg regression around `offload_state_dict` also matches cleanly.", + "confidence": 0.69, + "canonical_issue_reason": "Issue 6442 is the best anchor for the recurring FP16-unscale LoRA training bug: it has the exact error text, is broadly framed, and matches the same failure reported in 6124 and 8871.", + "canonical_pr_reason": null, + "best_issue_reason": "6442 is the cleanest representative of a real duplicate family in this cluster, with a precise title and enough context to subsume the same training error reports; it is better as the canonical issue than the more one-off, script-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5075", + "right": "issue:7239", + "accept": false, + "reason": "Both are SDXL shape-mismatch reports, but they involve different dimensions and workflows (prompt embeddings/refiner vs DreamBooth LoRA); not the same concrete bug." + }, + { + "left": "issue:6442", + "right": "issue:8871", + "accept": true, + "reason": "Same exact training error (`Attempting to unscale FP16 gradients`) in the same LoRA training family; these look like duplicate reports of the same bug." + }, + { + "left": "issue:4557", + "right": "issue:5029", + "accept": false, + "reason": "Different attribute errors on different objects; no shared underlying failure." + }, + { + "left": "issue:6353", + "right": "issue:6371", + "accept": false, + "reason": "Both mention controlnet examples, but the failures differ (training example issue vs controlnet_xs assert) and do not look mergeable into one fix." + }, + { + "left": "issue:4232", + "right": "issue:4234", + "accept": false, + "reason": "These are related feature requests/questions about instruct-pix2pix/controlnet, but they are not the same concrete bug or change request." + }, + { + "left": "issue:13137", + "right": "issue:13261", + "accept": true, + "reason": "Both describe the same FLUX.2-klein LoRA/LoKr loading problem; the titles differ slightly but point to the same underlying failure." + }, + { + "left": "issue:12436", + "right": "issue:12482", + "accept": true, + "reason": "Same `offload_state_dict` kwarg being passed into `CLIPTextModel`; this is the same bug stated with slightly different wording." + }, + { + "left": "issue:5712", + "right": "issue:9248", + "accept": false, + "reason": "Both are install/upgrade complaints, but the first is explicitly Windows/Python 3.12-specific while the second is generic; not enough to treat as the same bug." + }, + { + "left": "issue:2294", + "right": "issue:3787", + "accept": false, + "reason": "Different import errors for different symbols and different failure modes." + }, + { + "left": "issue:4615", + "right": "issue:5603", + "accept": false, + "reason": "Different missing imports (`StableDiffusionControlnetPipeline` vs `is_safetensors_available`); these are separate API breakages." + }, + { + "left": "issue:10722", + "right": "issue:9112", + "accept": false, + "reason": "Both are tensor-size mismatch errors, but they occur in different models/workflows with different shapes; too broad to merge." + }, + { + "left": "issue:4619", + "right": "issue:9476", + "accept": false, + "reason": "Both are dtype mismatch reports during LoRA training, but they involve different scripts, models, and dtype combinations." + }, + { + "left": "issue:3040", + "right": "issue:5992", + "accept": false, + "reason": "Both mention `__module__`-style attribute errors, but the actual failing objects and contexts differ; not the same bug." + }, + { + "left": "issue:4557", + "right": "issue:858", + "accept": false, + "reason": "Unrelated attribute errors on different objects (`super` getattr vs `Parameter.named_children`)." + }, + { + "left": "issue:2044", + "right": "issue:2669", + "accept": false, + "reason": "No shared concrete failure: DDP `dtype` access vs `NoneType.replace`." + }, + { + "left": "issue:2780", + "right": "issue:6482", + "accept": true, + "reason": "Same exact `Attempting to unscale FP16 gradients` failure in LoRA training, with the second being a variant of the same mixed-precision checkpoint/resume problem." + }, + { + "left": "issue:5012", + "right": "issue:5016", + "accept": false, + "reason": "Both concern text-encoder training in SDXL LoRA scripts, but they are separate script-specific reports and not clearly the same concrete bug/change." + }, + { + "left": "issue:7516", + "right": "issue:7681", + "accept": false, + "reason": "Completely unrelated failures: missing pipeline attribute vs OpenSSL import-chain error." + }, + { + "left": "issue:6124", + "right": "issue:8871", + "accept": true, + "reason": "Both report the exact same FP16 unscale error in LoRA/DreamBooth training; clear duplicate family." + }, + { + "left": "issue:2044", + "right": "issue:5572", + "accept": false, + "reason": "Different failures and different objects; no duplicate signal." + }, + { + "left": "issue:6324", + "right": "issue:8182", + "accept": false, + "reason": "`bool.__module__` attribute error and logger typo are unrelated." + }, + { + "left": "issue:2132", + "right": "issue:2669", + "accept": false, + "reason": "Different runtime errors in different code paths; no same underlying bug." + }, + { + "left": "issue:4265", + "right": "issue:8915", + "accept": false, + "reason": "Both are import errors, but for different pipeline symbols and unrelated API surfaces." + }, + { + "left": "issue:4118", + "right": "issue:8320", + "accept": false, + "reason": "Different attribute errors involving `AutoencoderKL`; not the same bug." + }, + { + "left": "issue:5603", + "right": "issue:6636", + "accept": false, + "reason": "Both are import-name failures from `diffusers.utils`, but they affect different exported symbols and are not clearly the same issue." + }, + { + "left": "issue:4969", + "right": "issue:5603", + "accept": false, + "reason": "Different missing imports with different symbols; not the same failure." + }, + { + "left": "issue:2947", + "right": "issue:6126", + "accept": false, + "reason": "Unrelated model attribute errors (`load_attn_procs` vs `tile_sample_min_size`)." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet. The chosen canonical issue 6442 is a conservative and supported anchor for the repeated FP16-unscale LoRA training failures, and the other accepted soft edges are limited to close textual matches. The rejected pairs are generally handled cautiously, without overstating duplication from similarity alone." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 9, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10062", + "issue:10752", + "issue:11149", + "issue:11218", + "issue:11417", + "issue:1249", + "issue:12779", + "issue:1709", + "issue:1988", + "issue:2089", + "issue:2132", + "issue:2947", + "issue:3094", + "issue:3211", + "issue:3382", + "issue:3787", + "issue:3816", + "issue:3973", + "issue:4086", + "issue:4186", + "issue:4265", + "issue:4478", + "issue:4529", + "issue:4557", + "issue:4615", + "issue:4868", + "issue:4969", + "issue:5025", + "issue:5029", + "issue:5054", + "issue:539", + "issue:5518", + "issue:5919", + "issue:6086", + "issue:6098", + "issue:6126", + "issue:6324", + "issue:6442", + "issue:6444", + "issue:6741", + "issue:6842", + "issue:7330", + "issue:7854", + "issue:8307", + "issue:858", + "issue:8871", + "issue:9026" + ], + "soft_pairs": [ + "issue:1709|issue:6324", + "issue:11149|issue:9026", + "issue:11218|issue:4265", + "issue:1249|issue:2947", + "issue:11417|issue:12779", + "issue:6098|issue:6442", + "issue:2089|issue:3382", + "issue:3973|issue:539", + "issue:4529|issue:9026", + "issue:6086|issue:7330", + "issue:4868|issue:5029", + "issue:4557|issue:6126", + "issue:2947|issue:3211", + "issue:3787|issue:4615", + "issue:4478|issue:7854", + "issue:2947|issue:5518", + "issue:10062|issue:6444", + "issue:1988|issue:4478", + "issue:5054|issue:858", + "issue:2132|issue:5919", + "issue:2947|issue:858", + "issue:12779|issue:6741", + "issue:2947|issue:5029", + "issue:3816|issue:8307", + "issue:5025|issue:6842", + "issue:3094|issue:3382", + "issue:10752|issue:8871", + "issue:4086|issue:4186", + "issue:4969|issue:6842", + "issue:5029|issue:5054" + ], + "prepared_review_unit_hash": "e6af63249f54c248fc5f968d8763bc15874c65dd9677237d3c48bedb53b861f1", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36069, + "estimated_input_tokens": 9018, + "estimated_eval_tokens": 18292 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous set of unrelated diffusers issues. The clear duplicate families are the FP16 gradient unscale bug, the half/float input-bias type mismatch runtime error, and the LoRA/PEFT peft_config AttributeError family only where the evidence is explicit enough.", + "confidence": 0.82, + "canonical_issue_reason": "issue:6442 is the most representative of the FP16 unscale failure cluster: it uses the exact shared error text and is the cleanest generic report among the variants.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6442 is the best global issue representative because it captures the recurring FP16 gradient unscale bug in the most generic form, without extra script-specific noise.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1709", + "right": "issue:6324", + "accept": false, + "reason": "Both are AttributeErrors, but they involve different missing attributes on different object types and do not look like the same bug." + }, + { + "left": "issue:11149", + "right": "issue:9026", + "accept": false, + "reason": "Both are unexpected-kwarg failures, but in different functions and with different signatures; same theme, not same defect." + }, + { + "left": "issue:11218", + "right": "issue:4265", + "accept": false, + "reason": "Both are import errors, but they are for different missing symbols in different modules." + }, + { + "left": "issue:1249", + "right": "issue:2947", + "accept": false, + "reason": "Different failures: ONNX numpy/device handling versus missing UNet load_attn_procs support." + }, + { + "left": "issue:11417", + "right": "issue:12779", + "accept": false, + "reason": "Different AttributeErrors in unrelated contexts: DDP dtype access versus a missing CLI argument." + }, + { + "left": "issue:6098", + "right": "issue:6442", + "accept": true, + "reason": "Same FP16 gradient unscale failure; the titles and reported error text align closely." + }, + { + "left": "issue:2089", + "right": "issue:3382", + "accept": false, + "reason": "Both mention OOM, but they are different training scripts and do not clearly share the same underlying bug." + }, + { + "left": "issue:3973", + "right": "issue:539", + "accept": false, + "reason": "Different unexpected keyword arguments on different APIs; not the same underlying change." + }, + { + "left": "issue:4529", + "right": "issue:9026", + "accept": false, + "reason": "Different kwarg regressions in different loader functions." + }, + { + "left": "issue:6086", + "right": "issue:7330", + "accept": true, + "reason": "Same FP16 unscale gradients failure, reported in another training script." + }, + { + "left": "issue:4868", + "right": "issue:5029", + "accept": false, + "reason": "Both are AttributeErrors, but on different objects with unrelated missing fields." + }, + { + "left": "issue:4557", + "right": "issue:6126", + "accept": false, + "reason": "Different missing-attribute problems in unrelated model code paths." + }, + { + "left": "issue:2947", + "right": "issue:3211", + "accept": false, + "reason": "Different missing attributes in different model classes; no shared bug is evident." + }, + { + "left": "issue:3787", + "right": "issue:4615", + "accept": false, + "reason": "Separate import issues for different pipeline symbols." + }, + { + "left": "issue:4478", + "right": "issue:7854", + "accept": true, + "reason": "Same half/float input-bias type mismatch runtime error; the failure text matches closely." + }, + { + "left": "issue:2947", + "right": "issue:5518", + "accept": false, + "reason": "Different missing attributes in different model components." + }, + { + "left": "issue:10062", + "right": "issue:6444", + "accept": false, + "reason": "The titles both mention peft_config, but the packet does not establish that they are the same DreamBooth SDXL/checkpointing code path, so this is too weak to merge." + }, + { + "left": "issue:12779", + "right": "issue:6741", + "accept": false, + "reason": "Different missing CLI arguments in different scripts." + }, + { + "left": "issue:2947", + "right": "issue:5029", + "accept": false, + "reason": "Different missing attributes on different model objects with no clear shared path." + }, + { + "left": "issue:3816", + "right": "issue:8307", + "accept": false, + "reason": "Different import errors for unrelated classes." + }, + { + "left": "issue:5025", + "right": "issue:6842", + "accept": false, + "reason": "Different missing imports from different modules." + }, + { + "left": "issue:3094", + "right": "issue:3382", + "accept": false, + "reason": "Both are OOM reports, but they describe different training setups and do not clearly share the same bug." + }, + { + "left": "issue:10752", + "right": "issue:8871", + "accept": true, + "reason": "Same FP16 gradient unscale failure, described as the same bug in another LoRA training report." + }, + { + "left": "issue:4086", + "right": "issue:4186", + "accept": false, + "reason": "Different SDXL/controlnet/inpaint training problems; not the same defect." + }, + { + "left": "issue:4969", + "right": "issue:6842", + "accept": false, + "reason": "Different import errors involving different missing symbols." + }, + { + "left": "issue:5029", + "right": "issue:5054", + "accept": false, + "reason": "Different missing attributes on different model objects." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Mostly grounded, but one soft-edge accept is too strong: issue:10062 vs issue:6444 only share the peft_config AttributeError in the titles. The packet does not establish that they are the same DreamBooth SDXL/checkpointing code path, so that verdict is more assertive than the evidence supports. The other pair calls look conservative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 10, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10062", + "issue:10337", + "issue:10472", + "issue:10989", + "issue:11149", + "issue:11435", + "issue:11973", + "issue:12492", + "issue:12547", + "issue:1544", + "issue:170", + "issue:1709", + "issue:1842", + "issue:2132", + "issue:2669", + "issue:3015", + "issue:3202", + "issue:3569", + "issue:3728", + "issue:3787", + "issue:3816", + "issue:384", + "issue:3973", + "issue:409", + "issue:4410", + "issue:4615", + "issue:4868", + "issue:4969", + "issue:5054", + "issue:5466", + "issue:6086", + "issue:6124", + "issue:6139", + "issue:6363", + "issue:6442", + "issue:6482", + "issue:672", + "issue:6858", + "issue:7516", + "issue:7593", + "issue:8182", + "issue:8463", + "issue:859", + "issue:8915", + "issue:9026", + "issue:9645", + "issue:9973" + ], + "soft_pairs": [ + "issue:4410|issue:5466", + "issue:1709|issue:5054", + "issue:6124|issue:6482", + "issue:10337|issue:10989", + "issue:6442|issue:9973", + "issue:6124|issue:6363", + "issue:3816|issue:7516", + "issue:10472|issue:11149", + "issue:1544|issue:2669", + "issue:4969|issue:9645", + "issue:10062|issue:6139", + "issue:3973|issue:9026", + "issue:859|issue:9026", + "issue:7593|issue:8463", + "issue:4615|issue:4969", + "issue:1544|issue:2132", + "issue:11973|issue:12547", + "issue:11435|issue:3787", + "issue:4615|issue:8915", + "issue:4868|issue:6139", + "issue:170|issue:1842", + "issue:3569|issue:3728", + "issue:1709|issue:8182", + "issue:384|issue:6858", + "issue:12492|issue:3816", + "issue:1709|issue:409", + "issue:3202|issue:5054", + "issue:6086|issue:6124", + "issue:3015|issue:672" + ], + "prepared_review_unit_hash": "4c67400215ebeffbcda3220ae12ccf18d91aaa5fd9ad3f21999b3782ccc5c424", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35857, + "estimated_input_tokens": 8965, + "estimated_eval_tokens": 18186 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a loose similarity cluster, not a single duplicate set. The only clear duplicate families are the repeated \"Attempting to unscale FP16 gradients\" reports, the repeated \"--instance_prompt\" Flux Kontext reports, and the FLUX quantized-transformers loading failures.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 6442 is the cleanest umbrella for the repeated \"Attempting to unscale FP16 gradients.\" failures and is the most generic, reusable representative of that bug family.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6442 best represents the most recurring concrete bug pattern in this cluster: the FP16 gradient unscale failure appearing across multiple training-script variants.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4410", + "right": "issue:5466", + "accept": false, + "reason": "Different failures: UNet shape mismatch vs LoRA weight loading shape mismatch; same broad area but not the same bug." + }, + { + "left": "issue:1709", + "right": "issue:5054", + "accept": false, + "reason": "Unrelated attribute errors on different objects/methods; no shared concrete code path." + }, + { + "left": "issue:6124", + "right": "issue:6482", + "accept": true, + "reason": "Same FP16 gradient-unscale error in closely related DreamBooth/LoRA SDXL training flows." + }, + { + "left": "issue:10337", + "right": "issue:10989", + "accept": true, + "reason": "Both describe FLUX adapter loading failing when transformers are quantized; same underlying quantized-load bug." + }, + { + "left": "issue:6442", + "right": "issue:9973", + "accept": true, + "reason": "Exact same \"Attempting to unscale FP16 gradients\" bug family in training scripts." + }, + { + "left": "issue:6124", + "right": "issue:6363", + "accept": true, + "reason": "Same exact FP16 gradient-unscale failure, just reported from a different LoRA training entry point." + }, + { + "left": "issue:3816", + "right": "issue:7516", + "accept": false, + "reason": "Different missing-transformers-attribute failures (CLIPImageProcessor vs DepthNormalEstimationPipeline)." + }, + { + "left": "issue:10472", + "right": "issue:11149", + "accept": false, + "reason": "Different bugs: unexpected lora_bias kwarg vs unexpected dtype kwarg in quantizer code." + }, + { + "left": "issue:1544", + "right": "issue:2669", + "accept": false, + "reason": "Both are generic attribute errors, but the missing attributes and code paths are unrelated." + }, + { + "left": "issue:4969", + "right": "issue:9645", + "accept": false, + "reason": "Both are import errors, but they involve different symbols and different failure causes." + }, + { + "left": "issue:10062", + "right": "issue:6139", + "accept": false, + "reason": "Different UNet attribute failures: missing peft_config vs missing attn_processors." + }, + { + "left": "issue:3973", + "right": "issue:9026", + "accept": false, + "reason": "Different unexpected-kwarg problems in different functions." + }, + { + "left": "issue:859", + "right": "issue:9026", + "accept": false, + "reason": "Different constructor/load kwargs; same error shape but not the same bug." + }, + { + "left": "issue:7593", + "right": "issue:8463", + "accept": false, + "reason": "Both are NoneType attribute errors, but on different objects and unrelated code paths." + }, + { + "left": "issue:4615", + "right": "issue:4969", + "accept": false, + "reason": "Different import targets from diffusers; not the same missing symbol." + }, + { + "left": "issue:1544", + "right": "issue:2132", + "accept": false, + "reason": "Completely different failures: scheduler API mismatch vs Accelerator trackers attribute." + }, + { + "left": "issue:11973", + "right": "issue:12547", + "accept": true, + "reason": "Same Flux Kontext issue: the script requires --instance_prompt in both reports." + }, + { + "left": "issue:11435", + "right": "issue:3787", + "accept": false, + "reason": "Both are import errors, but they reference different modules/symbols and different root causes." + }, + { + "left": "issue:4615", + "right": "issue:8915", + "accept": false, + "reason": "Both are import errors, but one is ControlNet pipeline and the other is AutoPipelineForText2Image; not the same bug." + }, + { + "left": "issue:4868", + "right": "issue:6139", + "accept": false, + "reason": "Different attribute errors on different objects; no evidence of a shared fix." + }, + { + "left": "issue:170", + "right": "issue:1842", + "accept": true, + "reason": "Same unexpected-eta error in the Stable Diffusion step call." + }, + { + "left": "issue:3569", + "right": "issue:3728", + "accept": false, + "reason": "Two separate new-pipeline requests, not duplicates." + }, + { + "left": "issue:1709", + "right": "issue:8182", + "accept": false, + "reason": "Unrelated attribute errors on Tensor vs Logger." + }, + { + "left": "issue:384", + "right": "issue:6858", + "accept": false, + "reason": "Both mention attribute errors, but the failing objects and contexts are unrelated." + }, + { + "left": "issue:12492", + "right": "issue:3816", + "accept": false, + "reason": "Both are transformers attribute issues, but they are different missing names and not clearly the same code-path problem." + }, + { + "left": "issue:1709", + "right": "issue:409", + "accept": false, + "reason": "Different missing-method failures in different APIs." + }, + { + "left": "issue:3202", + "right": "issue:5054", + "accept": false, + "reason": "Different pipeline attribute gaps; no shared underlying bug." + }, + { + "left": "issue:6086", + "right": "issue:6124", + "accept": true, + "reason": "Same FP16 gradient-unscale error across LoRA training examples; plausibly one fix covers both." + }, + { + "left": "issue:3015", + "right": "issue:672", + "accept": false, + "reason": "Related conversion topics, but one is ControlNet-specific and the other is a broader ckpt conversion request." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The rejected pairs are mostly justified by different concrete error signatures or unrelated code paths, and the accepted pairs are limited to clearly recurring families in the packet (FP16 unscale failures, --instance_prompt, eta keyword, and quantized-transformer load failures)." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 11, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10062", + "issue:10180", + "issue:10512", + "issue:11435", + "issue:11978", + "issue:1249", + "issue:13015", + "issue:1544", + "issue:1817", + "issue:1991", + "issue:2044", + "issue:2294", + "issue:2669", + "issue:2947", + "issue:3040", + "issue:3202", + "issue:3211", + "issue:3638", + "issue:3728", + "issue:3973", + "issue:409", + "issue:4265", + "issue:4557", + "issue:4615", + "issue:5025", + "issue:5246", + "issue:5518", + "issue:5572", + "issue:5578", + "issue:5603", + "issue:5897", + "issue:5919", + "issue:6086", + "issue:6098", + "issue:6139", + "issue:6842", + "issue:7891", + "issue:7931", + "issue:8307", + "issue:858", + "issue:8833", + "issue:8975", + "issue:9187", + "issue:9628", + "issue:9645", + "issue:9873", + "issue:9895" + ], + "soft_pairs": [ + "issue:2294|issue:8975", + "issue:1991|issue:5897", + "issue:5025|issue:9645", + "issue:6098|issue:9628", + "issue:2044|issue:5919", + "issue:5578|issue:8833", + "issue:6086|issue:9628", + "issue:7931|issue:9187", + "issue:11435|issue:5603", + "issue:13015|issue:9895", + "issue:1817|issue:3973", + "issue:5246|issue:8307", + "issue:1249|issue:4557", + "issue:3638|issue:3728", + "issue:6842|issue:9645", + "issue:10180|issue:10512", + "issue:1544|issue:5572", + "issue:10062|issue:2947", + "issue:1544|issue:2947", + "issue:2669|issue:2947", + "issue:4265|issue:4615", + "issue:13015|issue:7891", + "issue:3040|issue:3211", + "issue:2947|issue:409", + "issue:4557|issue:6139", + "issue:11978|issue:9873", + "issue:2947|issue:3202", + "issue:5518|issue:858" + ], + "prepared_review_unit_hash": "095f40d9ea6986d0cd036195504f536c04230a5e954a0907ebe4607637b15e79", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35372, + "estimated_input_tokens": 8843, + "estimated_eval_tokens": 17942 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose set of similar-looking diffusers issues rather than true duplicates. No soft-edge pair is supported strongly enough to accept; the IP-Adapter pair (#7931/#9187) in particular has different failure surfaces and lacks evidence of a shared root cause.", + "confidence": 0.84, + "canonical_issue_reason": "No single issue is a perfect canonical for the whole cluster; #2947 is the closest representative of the recurring UNet/LoRA attribute-error subgroup and is the earliest central report in that smaller theme.", + "canonical_pr_reason": null, + "best_issue_reason": "#2947 is the best global representative among these issues because it anchors the common UNet/LoRA API-compatibility pattern more clearly than the other items, which are mostly separate import/config/device problems.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2294", + "right": "issue:8975", + "accept": false, + "reason": "Both are import errors, but they fail on different missing symbols in different modules." + }, + { + "left": "issue:1991", + "right": "issue:5897", + "accept": false, + "reason": "Both mention device mismatches, but they involve different workflows and likely different root causes." + }, + { + "left": "issue:5025", + "right": "issue:9645", + "accept": false, + "reason": "Different missing exports from different namespaces; no shared concrete fix." + }, + { + "left": "issue:6098", + "right": "issue:9628", + "accept": false, + "reason": "Same broad FP16 unscale message, but different scripts/model setups and no evidence of one fix." + }, + { + "left": "issue:2044", + "right": "issue:5919", + "accept": false, + "reason": "Different missing attributes on different object types; not the same bug." + }, + { + "left": "issue:5578", + "right": "issue:8833", + "accept": false, + "reason": "Both are missing file errors, but for different files and different model/package contexts." + }, + { + "left": "issue:6086", + "right": "issue:9628", + "accept": false, + "reason": "Shared error text, but the training setups differ enough that the same fix is not established." + }, + { + "left": "issue:7931", + "right": "issue:9187", + "accept": false, + "reason": "Both involve IP-Adapter, but the titles point to different failure surfaces and there is no packet evidence they share the same root cause or fix." + }, + { + "left": "issue:11435", + "right": "issue:5603", + "accept": false, + "reason": "Different missing imports from different parts of diffusers; unrelated symbols." + }, + { + "left": "issue:13015", + "right": "issue:9895", + "accept": false, + "reason": "Both are device mismatch errors, but on different models/hardware paths and not clearly the same bug." + }, + { + "left": "issue:1817", + "right": "issue:3973", + "accept": false, + "reason": "Both are unexpected keyword errors, but in different helper functions and code paths." + }, + { + "left": "issue:5246", + "right": "issue:8307", + "accept": false, + "reason": "Different missing attributes on the diffusers module; not the same issue." + }, + { + "left": "issue:1249", + "right": "issue:4557", + "accept": false, + "reason": "Completely different failures: ndarray `.device` versus `super().__getattr__`." + }, + { + "left": "issue:3638", + "right": "issue:3728", + "accept": false, + "reason": "Both are new-pipeline proposals, but for different models and different changes." + }, + { + "left": "issue:6842", + "right": "issue:9645", + "accept": false, + "reason": "Different missing pipeline/import symbols; same package, not the same fix." + }, + { + "left": "issue:10180", + "right": "issue:10512", + "accept": false, + "reason": "Both concern Flux LoRA loading, but the reported scenarios are different enough to avoid merging." + }, + { + "left": "issue:1544", + "right": "issue:5572", + "accept": false, + "reason": "Different missing attributes on different components." + }, + { + "left": "issue:10062", + "right": "issue:2947", + "accept": false, + "reason": "Both are UNet/LoRA-related attribute errors, but they fail on different methods and likely need different fixes." + }, + { + "left": "issue:1544", + "right": "issue:2947", + "accept": false, + "reason": "Scheduler `get_velocity` and UNet `load_attn_procs` are different components and different failures." + }, + { + "left": "issue:2669", + "right": "issue:2947", + "accept": false, + "reason": "`NoneType.replace` is unrelated to UNet attention-processor loading." + }, + { + "left": "issue:4265", + "right": "issue:4615", + "accept": false, + "reason": "Both are ControlNet-related import errors, but they reference different pipeline symbols and exports." + }, + { + "left": "issue:4557", + "right": "issue:6139", + "accept": false, + "reason": "Different attribute errors on different object types; no shared code path." + }, + { + "left": "issue:11978", + "right": "issue:9873", + "accept": false, + "reason": "Different pipeline feature requests for different models." + }, + { + "left": "issue:2947", + "right": "issue:3202", + "accept": false, + "reason": "`load_attn_procs` and `enable_vae_tiling` are unrelated missing methods on different pipeline classes." + }, + { + "left": "issue:5518", + "right": "issue:858", + "accept": false, + "reason": "Different attribute errors in different module types; not the same bug." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The writeup is mostly conservative, but it overstates the only accepted soft-edge pair. From the packet, issue:7931 and issue:9187 both involve IP-Adapter, yet the titles point to different failure surfaces (`ip_adapter_image_embeds` in SDXL ControlNet InpaintPipeline vs `image_encoder_folder=None` after a Kolors update), and there is no explicit evidence in the packet that they share the same root cause or fix. That makes the accept verdict too strong. The rest of the pair judgments look appropriately cautious, but this one unsupported accept is enough to reject." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 12, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10472", + "issue:11214", + "issue:11417", + "issue:11865", + "issue:12044", + "issue:1249", + "issue:12630", + "issue:1315", + "issue:1433", + "issue:1817", + "issue:2044", + "issue:2098", + "issue:2669", + "issue:2904", + "issue:2947", + "issue:3008", + "issue:3211", + "issue:3325", + "issue:384", + "issue:409", + "issue:4557", + "issue:4615", + "issue:4868", + "issue:5025", + "issue:5054", + "issue:5125", + "issue:5510", + "issue:5522", + "issue:5572", + "issue:6086", + "issue:6098", + "issue:6176", + "issue:6351", + "issue:6363", + "issue:6382", + "issue:6442", + "issue:6636", + "issue:6811", + "issue:7228", + "issue:7593", + "issue:8463", + "issue:858", + "issue:9278", + "issue:9476", + "issue:9790", + "issue:9973" + ], + "soft_pairs": [ + "issue:3325|issue:5125", + "issue:3211|issue:4557", + "issue:11417|issue:4868", + "issue:6176|issue:6382", + "issue:12630|issue:2904", + "issue:9278|issue:9790", + "issue:1315|issue:6636", + "issue:12044|issue:1433", + "issue:409|issue:858", + "issue:6363|issue:6442", + "issue:1249|issue:5054", + "issue:10472|issue:1817", + "issue:2669|issue:5572", + "issue:10010|issue:6811", + "issue:2044|issue:6351", + "issue:2044|issue:384", + "issue:11214|issue:5522", + "issue:6098|issue:9973", + "issue:1249|issue:8463", + "issue:2098|issue:3008", + "issue:4615|issue:5025", + "issue:11865|issue:2947", + "issue:6086|issue:6098", + "issue:409|issue:5054", + "issue:7228|issue:7593", + "issue:5510|issue:9476" + ], + "prepared_review_unit_hash": "12a59477cb507f962dabf77c3435a9c97764dc5c891bd90cf48261590398c71b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 35150, + "estimated_input_tokens": 8788, + "estimated_eval_tokens": 17832 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous set of diffusers issues with a few clear duplicate mini-clusters. I accepted only the exact same-error pairs or very close same-script duplicates; the rest are too broad or point to different code paths.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 6363 is the strongest representative of the FP16-gradient duplicate set: it has the clearest problem statement and the richest discussion around the same exact failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6363 is the best overall issue to anchor this cluster because it matches multiple accepted duplicates and describes the core error cleanly.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3325", + "right": "issue:5125", + "accept": false, + "reason": "Both are CUDA failures, but they involve different scripts and different error conditions; not the same concrete bug." + }, + { + "left": "issue:3211", + "right": "issue:4557", + "accept": false, + "reason": "Different missing-attribute errors on different objects; only superficial similarity." + }, + { + "left": "issue:11417", + "right": "issue:4868", + "accept": false, + "reason": "Unrelated attribute errors in different code paths." + }, + { + "left": "issue:6176", + "right": "issue:6382", + "accept": false, + "reason": "One is LoRA weight loading after training, the other is an IP-adapter key error during SDXL/LCM use; not the same issue." + }, + { + "left": "issue:12630", + "right": "issue:2904", + "accept": false, + "reason": "Different AudioLDM-related failures with different missing attributes and root causes." + }, + { + "left": "issue:9278", + "right": "issue:9790", + "accept": true, + "reason": "Both describe the same FLUX.1-dev DreamBooth multigpu training problem and appear to be duplicate reports of the same workflow failure." + }, + { + "left": "issue:1315", + "right": "issue:6636", + "accept": false, + "reason": "Both are import errors from diffusers.utils, but they are different missing symbols and likely different breakages." + }, + { + "left": "issue:12044", + "right": "issue:1433", + "accept": false, + "reason": "Different attribute errors on different types; not the same bug." + }, + { + "left": "issue:409", + "right": "issue:858", + "accept": false, + "reason": "These failures are unrelated and arise from different objects and APIs." + }, + { + "left": "issue:6363", + "right": "issue:6442", + "accept": true, + "reason": "Same exact FP16 gradient unscale error; clearly duplicate reports of the same training problem." + }, + { + "left": "issue:1249", + "right": "issue:5054", + "accept": false, + "reason": "Different pipeline attribute failures with no shared concrete code path." + }, + { + "left": "issue:10472", + "right": "issue:1817", + "accept": false, + "reason": "Both are TypeErrors, but they concern different unexpected arguments and different workflows." + }, + { + "left": "issue:2669", + "right": "issue:5572", + "accept": false, + "reason": "Different AttributeErrors with different failing objects and causes." + }, + { + "left": "issue:10010", + "right": "issue:6811", + "accept": false, + "reason": "Both are 404s for model_index.json, but for different models/examples; same symptom, not the same underlying issue." + }, + { + "left": "issue:2044", + "right": "issue:6351", + "accept": false, + "reason": "Both involve DDP-wrapped models, but they fail on different missing attributes and code paths." + }, + { + "left": "issue:2044", + "right": "issue:384", + "accept": false, + "reason": "Completely different attribute errors; not duplicates." + }, + { + "left": "issue:11214", + "right": "issue:5522", + "accept": false, + "reason": "Different TypeErrors from different APIs and subsystems." + }, + { + "left": "issue:6098", + "right": "issue:9973", + "accept": true, + "reason": "Exact same FP16 gradient unscale failure; duplicate training error report." + }, + { + "left": "issue:1249", + "right": "issue:8463", + "accept": false, + "reason": "Different runtime failures in different pipelines; no shared concrete bug." + }, + { + "left": "issue:2098", + "right": "issue:3008", + "accept": true, + "reason": "Both report the same invalid torch.randn argument combination in the community pipeline path." + }, + { + "left": "issue:4615", + "right": "issue:5025", + "accept": false, + "reason": "Different import errors involving different symbols; not the same bug." + }, + { + "left": "issue:11865", + "right": "issue:2947", + "accept": false, + "reason": "Both are missing-method errors on model classes, but they are different methods and different feature paths." + }, + { + "left": "issue:6086", + "right": "issue:6098", + "accept": true, + "reason": "Same FP16-gradient unscale error, with one issue giving extra training-script context." + }, + { + "left": "issue:409", + "right": "issue:5054", + "accept": false, + "reason": "Different missing-attribute failures on different pipeline classes." + }, + { + "left": "issue:7228", + "right": "issue:7593", + "accept": false, + "reason": "Different model-loading/runtime errors on different pipelines; not the same underlying issue." + }, + { + "left": "issue:5510", + "right": "issue:9476", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different pipelines and likely different code paths." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet and the conservative stance is appropriate. The accepted duplicate mini-clusters are well supported by the titles and activity in the packet: the repeated FP16-gradient reports, the FLUX.1-dev multigpu DreamBooth pair, and the torch.randn invalid-arguments pair. The rejected pairs are phrased cautiously and do not overclaim sameness from superficial symptom overlap. No mergeability issue appears here since this is an issue cluster, not a PR pair." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 13, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11218", + "issue:11435", + "issue:1249", + "issue:12778", + "issue:13015", + "issue:1315", + "issue:1544", + "issue:1552", + "issue:1817", + "issue:1875", + "issue:1976", + "issue:1991", + "issue:2044", + "issue:2669", + "issue:2904", + "issue:2947", + "issue:3124", + "issue:3787", + "issue:384", + "issue:409", + "issue:4229", + "issue:4456", + "issue:4478", + "issue:4557", + "issue:4969", + "issue:5259", + "issue:5578", + "issue:5603", + "issue:5805", + "issue:5919", + "issue:6124", + "issue:6126", + "issue:6139", + "issue:6146", + "issue:6442", + "issue:6444", + "issue:6454", + "issue:6482", + "issue:6858", + "issue:7593", + "issue:7891", + "issue:8871", + "issue:8915", + "issue:8975", + "issue:9026", + "issue:9430", + "issue:9476" + ], + "soft_pairs": [ + "issue:6124|issue:6454", + "issue:1817|issue:9026", + "issue:4456|issue:6139", + "issue:5805|issue:9476", + "issue:4969|issue:8915", + "issue:1544|issue:5919", + "issue:2904|issue:7593", + "issue:5603|issue:8915", + "issue:6442|issue:6482", + "issue:3787|issue:8975", + "issue:1552|issue:3124", + "issue:1976|issue:5578", + "issue:4229|issue:6146", + "issue:2669|issue:6858", + "issue:1544|issue:4557", + "issue:409|issue:4557", + "issue:1991|issue:7891", + "issue:13015|issue:1991", + "issue:5259|issue:5578", + "issue:12778|issue:6444", + "issue:1875|issue:4478", + "issue:11435|issue:8915", + "issue:2947|issue:9430", + "issue:11218|issue:1315", + "issue:2044|issue:6126", + "issue:6454|issue:8871", + "issue:1249|issue:409", + "issue:2669|issue:384" + ], + "prepared_review_unit_hash": "fcb9b3e8dc5f6c5005491f093528b68cefd94f2670fdde432bd199b40a5813e2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35476, + "estimated_input_tokens": 8869, + "estimated_eval_tokens": 17994 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated import, attribute, or device/dtype errors. A small FP16-gradient subcluster appears to be genuine duplicates; the rest of the soft pairs are only title-similar and should stay separate.", + "confidence": 0.82, + "canonical_issue_reason": "issue:6442 is the clearest representative of the recurring \"Attempting to unscale FP16 gradients.\" bug; it matches multiple later reports with the same failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6442 is the strongest global representative because it is the most generic, high-signal report for the repeated FP16-unscale training failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6124", + "right": "issue:6454", + "accept": true, + "reason": "Same exact FP16 unscale-gradient error in the same SDXL/DreamBooth LoRA training area; looks like a duplicate report." + }, + { + "left": "issue:1817", + "right": "issue:9026", + "accept": false, + "reason": "Different failures: scheduler argument mismatch vs. checkpoint dispatch keyword mismatch." + }, + { + "left": "issue:4456", + "right": "issue:6139", + "accept": false, + "reason": "Both are LoRA-related attribute errors, but they hit different missing attributes in different code paths." + }, + { + "left": "issue:5805", + "right": "issue:9476", + "accept": false, + "reason": "Both are dtype mismatch reports, but they involve different pipelines/devices and are not clearly the same bug." + }, + { + "left": "issue:4969", + "right": "issue:8915", + "accept": false, + "reason": "Different missing imports from different diffusers symbols." + }, + { + "left": "issue:1544", + "right": "issue:5919", + "accept": false, + "reason": "Unrelated attribute errors: scheduler get_velocity vs. Linear set_lora_layer." + }, + { + "left": "issue:2904", + "right": "issue:7593", + "accept": false, + "reason": "Different NoneType failures in different pipelines and code paths." + }, + { + "left": "issue:5603", + "right": "issue:8915", + "accept": false, + "reason": "Different import symbols and likely different version-compatibility issues." + }, + { + "left": "issue:6442", + "right": "issue:6482", + "accept": true, + "reason": "Same exact FP16 unscale-gradient error; 6482 is just the resume-from-checkpoint variant of the same bug." + }, + { + "left": "issue:3787", + "right": "issue:8975", + "accept": false, + "reason": "Both are import errors, but they concern different classes and submodules." + }, + { + "left": "issue:1552", + "right": "issue:3124", + "accept": false, + "reason": "Attention slicing on SD2 is a different issue from model conversion failures." + }, + { + "left": "issue:1976", + "right": "issue:5578", + "accept": false, + "reason": "Both are OSError model_index.json complaints, but the reported contexts indicate different inputs and failure setups." + }, + { + "left": "issue:4229", + "right": "issue:6146", + "accept": false, + "reason": "Both mention multi-GPU training, but the titles do not identify the same concrete failure." + }, + { + "left": "issue:2669", + "right": "issue:6858", + "accept": false, + "reason": "Different attribute failures: NoneType.replace vs. string.convert." + }, + { + "left": "issue:1544", + "right": "issue:4557", + "accept": false, + "reason": "Completely different missing-attribute errors." + }, + { + "left": "issue:409", + "right": "issue:4557", + "accept": false, + "reason": "StableDiffusionPipeline attention slicing vs. a generic __getattr__ error are not the same bug." + }, + { + "left": "issue:1991", + "right": "issue:7891", + "accept": false, + "reason": "Both mention device mismatches, but one is checkpoint-merger specific and the other is a generic runtime exception." + }, + { + "left": "issue:13015", + "right": "issue:1991", + "accept": false, + "reason": "Different device-mismatch contexts and different pipelines." + }, + { + "left": "issue:5259", + "right": "issue:5578", + "accept": false, + "reason": "Closely similar OSError wording, but not enough to confirm the same underlying bug." + }, + { + "left": "issue:12778", + "right": "issue:6444", + "accept": false, + "reason": "Different missing attributes in unrelated model components." + }, + { + "left": "issue:1875", + "right": "issue:4478", + "accept": false, + "reason": "Both are type/device mismatch symptoms, but the pipelines and reported failure modes differ." + }, + { + "left": "issue:11435", + "right": "issue:8915", + "accept": false, + "reason": "Different missing imports from diffusers; no concrete shared code path." + }, + { + "left": "issue:2947", + "right": "issue:9430", + "accept": false, + "reason": "Different attribute errors with no evidence of the same underlying defect." + }, + { + "left": "issue:11218", + "right": "issue:1315", + "accept": false, + "reason": "Both are import errors after API changes, but they involve different symbols and modules." + }, + { + "left": "issue:2044", + "right": "issue:6126", + "accept": false, + "reason": "Different missing attributes on unrelated objects." + }, + { + "left": "issue:6454", + "right": "issue:8871", + "accept": true, + "reason": "Same exact FP16 unscale-gradient failure repeated in later LoRA training; clearly the same bug family." + }, + { + "left": "issue:1249", + "right": "issue:409", + "accept": false, + "reason": "Different subsystems: ONNX pipeline ndarray device handling vs. attention slicing." + }, + { + "left": "issue:2669", + "right": "issue:384", + "accept": false, + "reason": "Different NoneType/latent attribute failures with no shared concrete code path." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the accepted soft pairs are conservative because they share the same exact failure text or an obviously narrow variant. The rejected pairs stay appropriately cautious about title similarity without stronger evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 14, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11865", + "issue:12126", + "issue:12195", + "issue:12494", + "issue:12779", + "issue:13015", + "issue:1709", + "issue:1991", + "issue:2132", + "issue:2669", + "issue:2904", + "issue:2995", + "issue:3202", + "issue:3211", + "issue:3290", + "issue:3382", + "issue:4086", + "issue:409", + "issue:4265", + "issue:4551", + "issue:4557", + "issue:5025", + "issue:5054", + "issue:5081", + "issue:539", + "issue:5518", + "issue:5572", + "issue:5603", + "issue:5919", + "issue:6363", + "issue:6444", + "issue:6697", + "issue:6858", + "issue:7239", + "issue:7593", + "issue:8463", + "issue:8537", + "issue:858", + "issue:8637", + "issue:8863", + "issue:8915", + "issue:9026", + "issue:9055", + "issue:9278", + "issue:9450", + "issue:9726", + "issue:9727", + "issue:9973" + ], + "soft_pairs": [ + "issue:12195|issue:8537", + "issue:539|issue:9026", + "issue:5603|issue:8637", + "issue:6363|issue:9973", + "issue:12126|issue:8463", + "issue:3382|issue:6697", + "issue:12494|issue:7239", + "issue:5025|issue:8915", + "issue:409|issue:7593", + "issue:3290|issue:5054", + "issue:4265|issue:9055", + "issue:4086|issue:4551", + "issue:1709|issue:858", + "issue:5572|issue:6858", + "issue:1991|issue:9450", + "issue:11865|issue:4557", + "issue:2669|issue:3211", + "issue:1709|issue:3211", + "issue:9278|issue:9727", + "issue:2995|issue:5081", + "issue:12779|issue:6444", + "issue:3202|issue:5518", + "issue:2132|issue:5572", + "issue:3290|issue:409", + "issue:4265|issue:5603", + "issue:8637|issue:8915", + "issue:8463|issue:8863", + "issue:9278|issue:9726", + "issue:2904|issue:5919", + "issue:13015|issue:9450" + ], + "prepared_review_unit_hash": "5f1b445255407c2ce8e2a3cc917ae04c41b863a9a9faa09a8a0519d3ca21a06e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36608, + "estimated_input_tokens": 9152, + "estimated_eval_tokens": 18560 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated diffusers bug reports and API breakages. None of the proposed soft pairs look like the same underlying issue closely enough to merge; they share only broad subsystem keywords or generic error shapes.", + "confidence": 0.96, + "canonical_issue_reason": "Issue 9278 is the best anchor only by breadth and discussion volume: it is the most prominent of the FLUX multi-GPU dreambooth reports and has the strongest cluster visibility, though the cluster overall is not truly cohesive.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9278 is the strongest single representative for the largest visible subtheme in this set, but it still only covers one FLUX dreambooth/multi-GPU problem and does not generalize to the many unrelated API/import/shape/device errors in the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12195", + "right": "issue:8537", + "accept": false, + "reason": "Both are version/compatibility errors, but one is torch.library.custom_op in AutoencoderKL and the other is torch.float8_e4m3fn in StableDiffusion3 from_single_file; different missing APIs and code paths." + }, + { + "left": "issue:539", + "right": "issue:9026", + "accept": false, + "reason": "Both mention unexpected keyword arguments in loading utilities, but snapshot_download(allow_patterns) and load_checkpoint_and_dispatch(force_hook) are unrelated functions and failures." + }, + { + "left": "issue:5603", + "right": "issue:8637", + "accept": false, + "reason": "Both are import errors, but they concern different symbols in different modules: is_safetensors_available vs PixArtSigmaPipeline." + }, + { + "left": "issue:6363", + "right": "issue:9973", + "accept": false, + "reason": "Same general FP16/unscale wording, but one is a training-script gradient scaling failure and the other is a dreambooth LoRA script report; no evidence of the same bug." + }, + { + "left": "issue:12126", + "right": "issue:8463", + "accept": false, + "reason": "Both are NoneType attribute errors, but cache_context while switching WAN transformers and tokenize in InstructP2P-SDXL are different pipelines and failure points." + }, + { + "left": "issue:3382", + "right": "issue:6697", + "accept": false, + "reason": "Both are OOM/training complaints, but one is multi-GPU text-to-image training and the other is SDXL LoRA OOM; too broad to be the same defect." + }, + { + "left": "issue:12494", + "right": "issue:7239", + "accept": false, + "reason": "Both are matmul shape mismatch reports, but the tensor dimensions and training contexts differ; no clear shared code-path." + }, + { + "left": "issue:5025", + "right": "issue:8915", + "accept": false, + "reason": "Both are import errors from diffusers.utils/diffusers, but they are for different helpers: randn_tensor vs AutoPipelineForText2Image." + }, + { + "left": "issue:409", + "right": "issue:7593", + "accept": false, + "reason": "Both are pipeline attribute errors, but enable_attention_slicing in StableDiffusionPipeline and image_projection_layers in prepare_ip_adapter_image_embeds are unrelated features." + }, + { + "left": "issue:3290", + "right": "issue:5054", + "accept": false, + "reason": "Both involve missing attributes, but NoneType.flush and StableDiffusionXLPipeline.final_offload_hook are different objects and code paths." + }, + { + "left": "issue:4265", + "right": "issue:9055", + "accept": false, + "reason": "Both are import errors for pipeline/loaders symbols, but StableDiffusionXLControlNetPipeline and StableDiffusionLoraLoaderMixin are unrelated APIs." + }, + { + "left": "issue:4086", + "right": "issue:4551", + "accept": false, + "reason": "Both concern training with SDXL-ish models, but one is ControlNet training on SDXL base-0.9 and the other is a DreamBooth crash on SDXL base-1.0; not the same bug." + }, + { + "left": "issue:1709", + "right": "issue:858", + "accept": false, + "reason": "Both are AttributeError reports, but astype on Tensor in ONNX img2img and named_children on Parameter are different mistakes in different layers." + }, + { + "left": "issue:5572", + "right": "issue:6858", + "accept": false, + "reason": "Both are attribute failures, but scaling_factor on FrozenDict and convert on str arise in different model/pipeline setup paths." + }, + { + "left": "issue:1991", + "right": "issue:9450", + "accept": false, + "reason": "Both mention tensors on different devices, but checkpoint-merger and FluxPipeline transformer-device placement are separate subsystems and problems." + }, + { + "left": "issue:11865", + "right": "issue:4557", + "accept": false, + "reason": "Both are AttributeError on model objects, but from_single_file on CosmosTransformer3DModel and super().__getattr__ are unrelated failures." + }, + { + "left": "issue:2669", + "right": "issue:3211", + "accept": false, + "reason": "Both are NoneType/object attribute issues, but replace on NoneType and requires_safety_checker on StableDiffusionMegaPipeline are different pipeline behaviors." + }, + { + "left": "issue:1709", + "right": "issue:3211", + "accept": false, + "reason": "Different attribute errors in different pipelines: Tensor.astype in ONNX img2img vs StableDiffusionMegaPipeline.requires_safety_checker." + }, + { + "left": "issue:9278", + "right": "issue:9727", + "accept": false, + "reason": "Both are FLUX.1-dev dreambooth multigpu reports, but one is training on multigpu and the other is a save problem after multigpu training; not the same concrete bug." + }, + { + "left": "issue:2995", + "right": "issue:5081", + "accept": false, + "reason": "Both are shape mismatch errors, but the dimensions and model contexts differ; no evidence they share the same faulty interface." + }, + { + "left": "issue:12779", + "right": "issue:6444", + "accept": false, + "reason": "Both are training-script argument/attribute issues in dreambooth LoRA, but validation_image_path missing and peft_config missing under checkpointing are distinct bugs." + }, + { + "left": "issue:3202", + "right": "issue:5518", + "accept": false, + "reason": "Both are missing-attribute reports, but enable_vae_tiling on StableDiffusionControlNetPipeline and time on ModuleList are different APIs." + }, + { + "left": "issue:2132", + "right": "issue:5572", + "accept": false, + "reason": "Different missing attributes in different contexts: Accelerator.trackers versus FrozenDict.scaling_factor." + }, + { + "left": "issue:3290", + "right": "issue:409", + "accept": false, + "reason": "Both are attribute-related pipeline errors, but NoneType.flush and enable_attention_slicing are not the same bug or code path." + }, + { + "left": "issue:4265", + "right": "issue:5603", + "accept": false, + "reason": "Both are import errors, but the missing symbols are different utility/pipeline APIs and do not indicate one shared regression." + }, + { + "left": "issue:8637", + "right": "issue:8915", + "accept": false, + "reason": "Both are import errors, but PixArtSigmaPipeline and AutoPipelineForText2Image are separate exports with no shared failure evidence." + }, + { + "left": "issue:8463", + "right": "issue:8863", + "accept": false, + "reason": "Both are runtime errors in model usage, but tokenize on NoneType and tuple.shape are unrelated failure modes." + }, + { + "left": "issue:9278", + "right": "issue:9726", + "accept": false, + "reason": "Same broad FLUX dreambooth/multigpu area, but one is the general training issue and the other is a later problem report; titles do not point to the same concrete failure." + }, + { + "left": "issue:2904", + "right": "issue:5919", + "accept": false, + "reason": "Both are attribute errors, but AudioLDM scheduler access and LoRA set_lora_layer on Linear are unrelated pipeline/model components." + }, + { + "left": "issue:13015", + "right": "issue:9450", + "accept": false, + "reason": "Both mention device placement, but the NPU tensor-device mismatch in QwenImageEditPlus and FluxPipeline transformer device mismatch are different models and setups." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 15, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10154", + "issue:10180", + "issue:10472", + "issue:11651", + "issue:11732", + "issue:11865", + "issue:1249", + "issue:13015", + "issue:13300", + "issue:1433", + "issue:170", + "issue:2044", + "issue:2669", + "issue:2856", + "issue:2904", + "issue:3040", + "issue:3202", + "issue:3290", + "issue:3382", + "issue:3973", + "issue:409", + "issue:4456", + "issue:4529", + "issue:4557", + "issue:4868", + "issue:5029", + "issue:5054", + "issue:5081", + "issue:5082", + "issue:5171", + "issue:5518", + "issue:5897", + "issue:5919", + "issue:6086", + "issue:6126", + "issue:6351", + "issue:6454", + "issue:6482", + "issue:6869", + "issue:7114", + "issue:7228", + "issue:858", + "issue:859", + "issue:861", + "issue:8863", + "issue:8871", + "issue:9227", + "issue:9430" + ], + "soft_pairs": [ + "issue:10472|issue:4529", + "issue:13300|issue:9227", + "issue:2856|issue:861", + "issue:11865|issue:2904", + "issue:3290|issue:4557", + "issue:5081|issue:5082", + "issue:11732|issue:6869", + "issue:10154|issue:7228", + "issue:11865|issue:4868", + "issue:3040|issue:5518", + "issue:2044|issue:858", + "issue:3202|issue:409", + "issue:6086|issue:6454", + "issue:13015|issue:5897", + "issue:1433|issue:4557", + "issue:3290|issue:4868", + "issue:5054|issue:9430", + "issue:3382|issue:7114", + "issue:2044|issue:5029", + "issue:1249|issue:6126", + "issue:10154|issue:6351", + "issue:170|issue:859", + "issue:1249|issue:858", + "issue:5029|issue:8863", + "issue:6482|issue:8871", + "issue:10180|issue:11651", + "issue:3973|issue:5171", + "issue:4456|issue:6126", + "issue:2669|issue:5919" + ], + "prepared_review_unit_hash": "f60bf3fd5d4786bdce866babc7115723d1e67cdcdf738f20cc85ace54a5ba534", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36354, + "estimated_input_tokens": 9089, + "estimated_eval_tokens": 18434 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of superficially similar error-report issues, with only a few true duplicates among the soft pairs. The strongest duplication signals are the ckpt-to-diffusers conversion reports and the FP16-gradient unscale LoRA-training reports; most other pairs share only a broad error pattern or subsystem.", + "confidence": 0.72, + "canonical_issue_reason": "Issue 10154 is the most generic, cluster-central LoRA-loading report and best represents the recurring `load_lora_weights` attribute-error symptom in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "10154 is the clearest representative for triage because its title is broad, directly reflects the shared LoRA-loading failure mode, and can anchor related reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10472", + "right": "issue:4529", + "accept": false, + "reason": "Different missing kwargs in different LoRA APIs (`lora_bias` vs `network_alpha`); same general area, but not the same bug." + }, + { + "left": "issue:13300", + "right": "issue:9227", + "accept": false, + "reason": "Both mention FLUX, but one is CPU inference and the other is a Flux LoRA training/loading problem; different concrete code paths." + }, + { + "left": "issue:2856", + "right": "issue:861", + "accept": true, + "reason": "Both are about converting Stable Diffusion checkpoints/models to Diffusers with the same converter path; this looks like the same underlying conversion bug." + }, + { + "left": "issue:11865", + "right": "issue:2904", + "accept": false, + "reason": "Completely different failures (`from_single_file` on CosmosTransformer3DModel vs AudioLDM scheduler NoneType)." + }, + { + "left": "issue:3290", + "right": "issue:4557", + "accept": false, + "reason": "Different errors on unrelated objects (`flush` vs `__getattr__`); no shared code-path evidence." + }, + { + "left": "issue:5081", + "right": "issue:5082", + "accept": false, + "reason": "Both are matmul shape mismatches, but the dimensions and reported context differ; not enough to call them the same bug." + }, + { + "left": "issue:11732", + "right": "issue:6869", + "accept": false, + "reason": "Same general unpacking error class, but different functions and stack contexts; too broad to merge." + }, + { + "left": "issue:10154", + "right": "issue:7228", + "accept": false, + "reason": "Both are attribute errors, but one is about `load_lora_weights` and the other about traced model `dtype`; unrelated failures." + }, + { + "left": "issue:11865", + "right": "issue:4868", + "accept": false, + "reason": "Different missing attributes in different objects; no indication of a shared underlying defect." + }, + { + "left": "issue:3040", + "right": "issue:5518", + "accept": false, + "reason": "Different attribute failures in different loader/model paths (`__module__` vs `time`)." + }, + { + "left": "issue:2044", + "right": "issue:858", + "accept": false, + "reason": "Both are PyTorch object attribute errors, but the objects and failure modes are unrelated (`dtype` vs `named_children`)." + }, + { + "left": "issue:3202", + "right": "issue:409", + "accept": false, + "reason": "Different missing pipeline methods (`enable_vae_tiling` vs `enable_attention_slicing`); version-skew style issue, not the same bug." + }, + { + "left": "issue:6086", + "right": "issue:6454", + "accept": true, + "reason": "Same LoRA-training FP16 gradient unscale failure; the titles point to the same concrete bug." + }, + { + "left": "issue:13015", + "right": "issue:5897", + "accept": false, + "reason": "Both are device-mismatch errors, but on different models/hardware and different code paths; too broad to merge." + }, + { + "left": "issue:1433", + "right": "issue:4557", + "accept": false, + "reason": "Different objects and different missing attribute symptoms (`config` vs `__getattr__`)." + }, + { + "left": "issue:3290", + "right": "issue:4868", + "accept": false, + "reason": "Different NoneType/super attribute errors with no shared implementation path." + }, + { + "left": "issue:5054", + "right": "issue:9430", + "accept": false, + "reason": "Unrelated attribute errors in different components (`final_offload_hook` vs `NoneType.get`)." + }, + { + "left": "issue:3382", + "right": "issue:7114", + "accept": false, + "reason": "Both are OOM reports, but they involve different training scripts and settings; not the same concrete bug." + }, + { + "left": "issue:2044", + "right": "issue:5029", + "accept": false, + "reason": "Different missing attributes on different object types; not a duplicate." + }, + { + "left": "issue:1249", + "right": "issue:6126", + "accept": false, + "reason": "Different runtime failures in different subsystems (`numpy.ndarray.device` vs VAE tiling attribute)." + }, + { + "left": "issue:10154", + "right": "issue:6351", + "accept": false, + "reason": "Both mention missing attributes, but one is LoRA loading and the other is DDP adapter disabling; unrelated code paths." + }, + { + "left": "issue:170", + "right": "issue:859", + "accept": false, + "reason": "Different API signature mismatches (`eta` vs `device_map`) in different functions." + }, + { + "left": "issue:1249", + "right": "issue:858", + "accept": false, + "reason": "Different object types and missing attributes; only the broad error pattern matches." + }, + { + "left": "issue:5029", + "right": "issue:8863", + "accept": false, + "reason": "Different failures (`latent_dist` missing vs tuple shape access); no shared bug evidence." + }, + { + "left": "issue:6482", + "right": "issue:8871", + "accept": true, + "reason": "Same `Attempting to unscale FP16 gradients` LoRA-training failure, with the second issue clearly describing the same bug." + }, + { + "left": "issue:10180", + "right": "issue:11651", + "accept": false, + "reason": "Both are FluxControl LoRA load problems, but one is about multiple LoRAs and the other a specific flux-fill LoRA case; not enough to assert one bug." + }, + { + "left": "issue:3973", + "right": "issue:5171", + "accept": false, + "reason": "Different unexpected-argument errors in different functions (`vae_path` vs `safety_checker`)." + }, + { + "left": "issue:4456", + "right": "issue:6126", + "accept": false, + "reason": "Different attribute errors in different parts of the LoRA/VAE stack (`ModuleList[4]` vs `tile_sample_min_size`)." + }, + { + "left": "issue:2669", + "right": "issue:5919", + "accept": false, + "reason": "Both are `AttributeError`s, but the missing attributes and object types differ completely (`replace` vs `set_lora_layer`)." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The issue summary is grounded in the packet, and the soft-edge calls are conservative overall. The three accepted pairs have strong title-level duplication signals, while the rejected pairs mostly differ by concrete API, object, or code path rather than just sharing a broad error class." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 16, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10722", + "issue:11117", + "issue:11435", + "issue:11436", + "issue:11464", + "issue:1255", + "issue:12609", + "issue:12612", + "issue:12613", + "issue:1263", + "issue:12666", + "issue:13162", + "issue:13203", + "issue:13221", + "issue:13311", + "issue:1377", + "issue:1444", + "issue:1500", + "issue:1616", + "issue:1991", + "issue:2661", + "issue:2662", + "issue:3576", + "issue:3580", + "issue:3627", + "issue:4557", + "issue:4868", + "issue:5022", + "issue:5023", + "issue:5075", + "issue:5399", + "issue:6086", + "issue:6442", + "issue:6454", + "issue:6549", + "issue:6880", + "issue:7033", + "issue:7185", + "issue:7799", + "issue:7922", + "issue:7931", + "issue:8589", + "issue:9062", + "issue:9378", + "issue:9403", + "issue:9430", + "issue:9514", + "issue:9895" + ], + "soft_pairs": [ + "issue:1991|issue:9895", + "issue:6086|issue:6442", + "issue:11464|issue:9514", + "issue:4557|issue:9430", + "issue:4868|issue:7922", + "issue:6442|issue:6454", + "issue:2661|issue:2662", + "issue:12609|issue:12613", + "issue:12612|issue:12613", + "issue:3576|issue:3580", + "issue:13203|issue:13221", + "issue:13162|issue:13311", + "issue:5022|issue:5023", + "issue:11117|issue:11436", + "issue:1500|issue:1616", + "issue:7799|issue:7931", + "issue:1377|issue:1444", + "issue:3627|issue:6880", + "issue:12666|issue:7185", + "issue:9378|issue:9403", + "issue:10722|issue:8589", + "issue:11435|issue:9062", + "issue:6549|issue:7033", + "issue:5075|issue:5399", + "issue:1255|issue:1263" + ], + "prepared_review_unit_hash": "74f3f67064eed847ecf9def9f7fda1519d271344509796da2856c29f1e12883e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34491, + "estimated_input_tokens": 8623, + "estimated_eval_tokens": 17502 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mix of superficially similar issue reports rather than proven duplicates. The clearest anchor is the active Wan 2.2 i2v NaN-latents report, but the soft-edge pairs mostly do not have enough evidence beyond title overlap to justify collapse/merge conclusions. No PRs were present.", + "confidence": 0.66, + "canonical_issue_reason": "issue:12613 is the strongest cluster anchor because it is the most active open Wan 2.2/i2v NaN-latents report and the most actionable focal point for related discussion, while still leaving room for nearby reports to remain separate without stronger evidence.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:12613 is the best overall issue candidate for cluster navigation: it is open, specific, and has the most discussion/inbound attention among the Wan latent-NaN reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1991", + "right": "issue:9895", + "accept": false, + "reason": "Both mention device-mismatch style failures, but they occur in different pipelines and contexts (checkpoint merger vs Flux inference), so this is too broad to treat as the same bug." + }, + { + "left": "issue:6086", + "right": "issue:6442", + "accept": false, + "reason": "The titles are similar, but the available evidence here is only symptom-level AMP/GradScaler overlap; that is not enough to confirm a single underlying bug." + }, + { + "left": "issue:11464", + "right": "issue:9514", + "accept": false, + "reason": "Both are Flux/LoRA-related, but one is a pipeline loading problem and the other is an fp8 quantization weight-loading issue; different code paths." + }, + { + "left": "issue:4557", + "right": "issue:9430", + "accept": false, + "reason": "Different exceptions and different failure points; the shared AttributeError wording is not enough to call them duplicates." + }, + { + "left": "issue:4868", + "right": "issue:7922", + "accept": false, + "reason": "Both are generic attribute/import errors, but they refer to different missing symbols in different subsystems, so the evidence is too weak for a merge." + }, + { + "left": "issue:6442", + "right": "issue:6454", + "accept": false, + "reason": "The titles match exactly, but the packet provides no concrete code-path evidence; treat as related reports rather than a confirmed duplicate pair." + }, + { + "left": "issue:2661", + "right": "issue:2662", + "accept": false, + "reason": "The write-error wording is close, but one report is step-specific and the other is broader; the shown evidence does not prove the same bug." + }, + { + "left": "issue:12609", + "right": "issue:12613", + "accept": false, + "reason": "Both mention Wan latent NaNs, but the model/version and i2v context differ; title similarity alone is not enough to merge them." + }, + { + "left": "issue:12612", + "right": "issue:12613", + "accept": false, + "reason": "These are related Wan NaN reports, but they target different model/version contexts (2.1 vs 2.2 i2v), so they should stay separate without stronger evidence." + }, + { + "left": "issue:3576", + "right": "issue:3580", + "accept": false, + "reason": "Both are XFormers TypeError reports, but one is tied to Textual Inversion and the other is a generic XFormers failure; not clearly the same defect." + }, + { + "left": "issue:13203", + "right": "issue:13221", + "accept": false, + "reason": "The titles are very close, but the packet only shows another similarly named ZImage LoRA support issue; that is not enough to confidently treat them as one change/bug." + }, + { + "left": "issue:13162", + "right": "issue:13311", + "accept": false, + "reason": "Both concern Flux.2-klein LoRA support, but one is a general support request and the other is a specific loading error for a different model size; not the same underlying issue." + }, + { + "left": "issue:5022", + "right": "issue:5023", + "accept": false, + "reason": "Both are import errors in the same release, but they are for different symbols and pipelines, so they do not look like one concrete defect." + }, + { + "left": "issue:11117", + "right": "issue:11436", + "accept": false, + "reason": "Both involve dtype mismatches, but the failing operations and pipelines differ, so this is not strong enough evidence for duplication." + }, + { + "left": "issue:1500", + "right": "issue:1616", + "accept": false, + "reason": "These are both LPW community pipeline compatibility reports, but they target different Diffusers versions and different descriptions of the breakage." + }, + { + "left": "issue:7799", + "right": "issue:7931", + "accept": false, + "reason": "Both involve IP-Adapter, but one is a matmul size mismatch and the other is an inpaint/embed-parameter issue; different concrete failures." + }, + { + "left": "issue:1377", + "right": "issue:1444", + "accept": false, + "reason": "Two different missing-class import errors; the common import-error pattern is too generic to merge them." + }, + { + "left": "issue:3627", + "right": "issue:6880", + "accept": false, + "reason": "Both reference Custom Diffusion retrieve.py, but the packet does not provide enough concrete shared failure detail to safely call them the same bug." + }, + { + "left": "issue:12666", + "right": "issue:7185", + "accept": false, + "reason": "Completely different topics: a missing Kandinsky pipeline versus a general PyTorch support question." + }, + { + "left": "issue:9378", + "right": "issue:9403", + "accept": false, + "reason": "These are separate Xlabs support requests for different components (ControlNet vs IP-Adapter), so they should not be merged as one change." + }, + { + "left": "issue:10722", + "right": "issue:8589", + "accept": false, + "reason": "The reports are in the same DreamBooth/prior-preservation area, but one is a concrete tensor-shape mismatch and the other is a broader training inability report; not enough to confirm identity." + }, + { + "left": "issue:11435", + "right": "issue:9062", + "accept": false, + "reason": "Different missing pipeline imports with different symbols and paths; no evidence of the same underlying issue." + }, + { + "left": "issue:6549", + "right": "issue:7033", + "accept": false, + "reason": "Both are IP-Adapter warnings, but the observed behaviors are different (LoRA weight logging vs attention masking), so they are not clearly duplicates." + }, + { + "left": "issue:5075", + "right": "issue:5399", + "accept": false, + "reason": "A shape-mismatch in SDXL Refiner prompt embeddings is not the same as a dtype mismatch in custom inpainting inference." + }, + { + "left": "issue:1255", + "right": "issue:1263", + "accept": false, + "reason": "Both are import errors, but for different classes/modules; the shared symptom is too generic to justify a duplicate call." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The issue-level summary is mostly grounded, but the soft-edge handling is not conservative enough. Several pairs are marked accept even though the packet only gives title-level similarity and the packet\u2019s deterministic label is false for those pairs. The strongest overstatements are 6086/6442, 6442/6454, 2661/2662, 10722/8589, 12609/12613, 13203/13221, and 3627/6880. These may be related, but the evidence shown here does not justify confidently treating them as the same bug. The writeup should stay softer on duplicate/merge conclusions." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 17, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10180", + "issue:10752", + "issue:10848", + "issue:10963", + "issue:10989", + "issue:11199", + "issue:11283", + "issue:11581", + "issue:11827", + "issue:1255", + "issue:12826", + "issue:13124", + "issue:1444", + "issue:1538", + "issue:1611", + "issue:2294", + "issue:2780", + "issue:2908", + "issue:2947", + "issue:2984", + "issue:2991", + "issue:3140", + "issue:3279", + "issue:3569", + "issue:3638", + "issue:4052", + "issue:4063", + "issue:4069", + "issue:4118", + "issue:4754", + "issue:4969", + "issue:5025", + "issue:5757", + "issue:579", + "issue:6044", + "issue:6098", + "issue:6112", + "issue:6734", + "issue:7221", + "issue:7223", + "issue:7681", + "issue:8320", + "issue:8482", + "issue:8615", + "issue:9527", + "issue:9630", + "issue:9911" + ], + "soft_pairs": [ + "issue:10963|issue:1255", + "issue:10963|issue:4969", + "issue:2908|issue:2991", + "issue:13124|issue:2780", + "issue:10026|issue:9630", + "issue:6112|issue:7221", + "issue:4754|issue:579", + "issue:8482|issue:8615", + "issue:2294|issue:3140", + "issue:12826|issue:4969", + "issue:11199|issue:7681", + "issue:11283|issue:11827", + "issue:12826|issue:5025", + "issue:3569|issue:3638", + "issue:10848|issue:11581", + "issue:10180|issue:10989", + "issue:1444|issue:4969", + "issue:1538|issue:1611", + "issue:2984|issue:7223", + "issue:4063|issue:4069", + "issue:4052|issue:4118", + "issue:9527|issue:9911", + "issue:10752|issue:6098", + "issue:4052|issue:8320", + "issue:6044|issue:6734", + "issue:3279|issue:5757", + "issue:2947|issue:6044" + ], + "prepared_review_unit_hash": "7f40a94b37c2715112e66bf1b0d0d27fffcf6a9b315a9e0575c4bc93b2609656", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35239, + "estimated_input_tokens": 8810, + "estimated_eval_tokens": 17876 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose mix of diffusers issues around imports, training, and Flux/IP-Adapter loading; only a few pairs are clear duplicates, chiefly the ControlNet mixed-precision and textual-inversion/from_single_file reports.", + "confidence": 0.68, + "canonical_issue_reason": "Issue 2908 is the best canonical anchor because it is a concrete, script-specific bug report and has a near-identical duplicate in 2991.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 2908 is the most actionable issue in the cluster: it names the exact example script and failure mode, making it the strongest duplicate target among the accepted pairs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10963", + "right": "issue:1255", + "accept": false, + "reason": "Both are import errors, but they concern different missing symbols and different code paths." + }, + { + "left": "issue:10963", + "right": "issue:4969", + "accept": false, + "reason": "Different missing imports (`AutoencoderKLWan` vs `CrossAttention`), so not the same bug." + }, + { + "left": "issue:2908", + "right": "issue:2991", + "accept": true, + "reason": "Both report ControlNet PyTorch training/example mixed-precision failures in the same training script family." + }, + { + "left": "issue:13124", + "right": "issue:2780", + "accept": false, + "reason": "Same error phrase, but different training scripts and no clear evidence of the same underlying fix." + }, + { + "left": "issue:10026", + "right": "issue:9630", + "accept": false, + "reason": "Both involve Flux ControlNet multi-GPU/single-GPU setups, but one is about memory scaling and the other a specific configuration error." + }, + { + "left": "issue:6112", + "right": "issue:7221", + "accept": false, + "reason": "Both describe noisy outputs, but they are different models and likely different inference/setup issues." + }, + { + "left": "issue:4754", + "right": "issue:579", + "accept": false, + "reason": "Both are textual inversion loading problems, but the missing file and failure mode differ." + }, + { + "left": "issue:8482", + "right": "issue:8615", + "accept": false, + "reason": "These are different feature requests for Stable Diffusion 3, not the same change." + }, + { + "left": "issue:2294", + "right": "issue:3140", + "accept": false, + "reason": "Different missing pipeline imports (`VideoInpaint` vs `ControlNet`) with no shared concrete bug." + }, + { + "left": "issue:12826", + "right": "issue:4969", + "accept": false, + "reason": "Different missing imports (`export_to_video` vs `CrossAttention`)." + }, + { + "left": "issue:11199", + "right": "issue:7681", + "accept": false, + "reason": "Both are import/runtime failures, but one is on `onnx_utils` and the other is an OpenSSL/`lib` dependency error." + }, + { + "left": "issue:11283", + "right": "issue:11827", + "accept": false, + "reason": "Both mention safetensors header deserialization, but they are on different model families and likely different file/data issues." + }, + { + "left": "issue:12826", + "right": "issue:5025", + "accept": false, + "reason": "Different import symbols and unrelated code paths." + }, + { + "left": "issue:3569", + "right": "issue:3638", + "accept": false, + "reason": "These are separate new-pipeline feature requests for different models." + }, + { + "left": "issue:10848", + "right": "issue:11581", + "accept": false, + "reason": "Both concern IP-Adapter loading in FLUX, but the reported constructor kwarg errors differ (`device` vs `dtype`)." + }, + { + "left": "issue:10180", + "right": "issue:10989", + "accept": false, + "reason": "Both are Flux Control LoRA issues, but one is about multiple LoRAs and the other about quantized transformers." + }, + { + "left": "issue:1444", + "right": "issue:4969", + "accept": false, + "reason": "Different import failures for different classes." + }, + { + "left": "issue:1538", + "right": "issue:1611", + "accept": false, + "reason": "Same pipeline name, but different base models and likely different compatibility regressions." + }, + { + "left": "issue:2984", + "right": "issue:7223", + "accept": false, + "reason": "One is a research paper/model request and the other is a new pipeline request; not duplicates." + }, + { + "left": "issue:4063", + "right": "issue:4069", + "accept": true, + "reason": "Near-identical textual inversion unload issues; same underlying pipeline cleanup behavior." + }, + { + "left": "issue:4052", + "right": "issue:4118", + "accept": true, + "reason": "Both are the same from_single_file regression around `AutoencoderKL`, with one being the class-specific variant of the other." + }, + { + "left": "issue:9527", + "right": "issue:9911", + "accept": false, + "reason": "Both are dtype/controlnet-related errors, but the scenarios are different: mixed precision mismatch vs multiple ControlNet layer lengths." + }, + { + "left": "issue:10752", + "right": "issue:6098", + "accept": false, + "reason": "The error text is similar, but there is no evidence they arise from the same training code path or fix." + }, + { + "left": "issue:4052", + "right": "issue:8320", + "accept": false, + "reason": "Same feature area, but `from_single_file` with `__name__` is a distinct error manifestation and likely a different regression." + }, + { + "left": "issue:6044", + "right": "issue:6734", + "accept": false, + "reason": "Different failures: missing `do_classifier_free_guidance` vs tokenizer being `None`." + }, + { + "left": "issue:3279", + "right": "issue:5757", + "accept": false, + "reason": "Both are resume-from-checkpoint problems, but they affect different scripts/models and are not clearly the same bug." + }, + { + "left": "issue:2947", + "right": "issue:6044", + "accept": false, + "reason": "Different missing attributes on different pipeline classes; not the same underlying issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and the soft-edge decisions are appropriately conservative. The accepted duplicate pairs are all reasonably supported by the titles/body descriptions, and the rejected pairs avoid overstating similarity where the bug paths differ. Canonicalizing around issue 2908 is defensible given the script-specific ControlNet mixed-precision report and its close match with 2991." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 18, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10066", + "issue:10591", + "issue:10752", + "issue:1082", + "issue:1100", + "issue:11014", + "issue:11410", + "issue:12047", + "issue:12097", + "issue:12147", + "issue:1255", + "issue:1260", + "issue:1263", + "issue:13087", + "issue:13124", + "issue:13258", + "issue:1444", + "issue:1478", + "issue:1572", + "issue:1842", + "issue:1946", + "issue:2044", + "issue:2132", + "issue:2294", + "issue:2780", + "issue:3420", + "issue:3728", + "issue:3884", + "issue:4483", + "issue:4542", + "issue:4619", + "issue:4724", + "issue:4872", + "issue:5897", + "issue:6363", + "issue:6454", + "issue:6697", + "issue:6879", + "issue:7133", + "issue:7210", + "issue:8414", + "issue:852", + "issue:8534", + "issue:859", + "issue:8975", + "issue:9062", + "issue:9841" + ], + "soft_pairs": [ + "issue:4619|issue:6879", + "issue:12047|issue:12147", + "issue:12097|issue:13258", + "issue:1478|issue:4542", + "issue:8975|issue:9062", + "issue:11014|issue:4872", + "issue:3884|issue:5897", + "issue:10752|issue:2780", + "issue:13124|issue:6454", + "issue:1842|issue:859", + "issue:1100|issue:7210", + "issue:852|issue:859", + "issue:1263|issue:1444", + "issue:7133|issue:9841", + "issue:4483|issue:4724", + "issue:1255|issue:13087", + "issue:10066|issue:11410", + "issue:1082|issue:1946", + "issue:1260|issue:9062", + "issue:10591|issue:8534", + "issue:2044|issue:2132", + "issue:13124|issue:6363", + "issue:4872|issue:6697", + "issue:3728|issue:8414", + "issue:1260|issue:2294", + "issue:1572|issue:3420" + ], + "prepared_review_unit_hash": "cfc781f3b34e6c28466d63d9db766ad180309eca3ec31a5f8083fe671b4ffb53", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 34477, + "estimated_input_tokens": 8620, + "estimated_eval_tokens": 17496 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "A small set of true duplicates is present around FP16-unscale and import-signature regressions; most other pairs are only loosely similar symptoms across different models, scripts, or pipelines and should remain separate.", + "confidence": 0.74, + "canonical_issue_reason": "Issue #6454 is the broadest, most generic report of the recurring \"Attempting to unscale FP16 gradients\" failure, so it best represents the duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #6454 is the best representative because it states the core error cleanly without tying it to one narrow script-specific trigger.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4619", + "right": "issue:6879", + "accept": false, + "reason": "Both are dtype-mismatch runtime errors, but they affect different training paths (SDXL LoRA vs Custom Diffusion) and are not clearly the same bug." + }, + { + "left": "issue:12047", + "right": "issue:12147", + "accept": false, + "reason": "Both mention Wan2.2 LoRA problems, but one is fusion and the other loading; different code paths and not clearly one concrete defect." + }, + { + "left": "issue:12097", + "right": "issue:13258", + "accept": false, + "reason": "One is VRAM OOM, the other is a missing Image-to-Video model; these are different problems." + }, + { + "left": "issue:1478", + "right": "issue:4542", + "accept": false, + "reason": "Both are multi-image output limitations, but they hit different pipeline classes and are not clearly the same underlying defect." + }, + { + "left": "issue:8975", + "right": "issue:9062", + "accept": false, + "reason": "Both are import errors, but they involve different symbols and unrelated modules." + }, + { + "left": "issue:11014", + "right": "issue:4872", + "accept": false, + "reason": "Both are OOM reports, but one is generic high-RAM A100 training and the other is SDXL T4 finetuning; too broad to treat as one bug." + }, + { + "left": "issue:3884", + "right": "issue:5897", + "accept": false, + "reason": "Same script family, but the titles do not identify the same concrete failure; #5897 is a device mismatch while #3884 is generic." + }, + { + "left": "issue:10752", + "right": "issue:2780", + "accept": true, + "reason": "Same FP16-gradient unscale ValueError in LoRA training; #2780 is the exact script-specific report of the same bug." + }, + { + "left": "issue:13124", + "right": "issue:6454", + "accept": false, + "reason": "Both mention the FP16-unscale error, but one is DreamBooth LoRA with a validation_prompt trigger and the other is a broader/general report; not clearly one concrete code path." + }, + { + "left": "issue:1842", + "right": "issue:859", + "accept": false, + "reason": "Both are unexpected-kwarg TypeErrors, but the kwargs differ (eta vs device_map) and point to different incompatibilities." + }, + { + "left": "issue:1100", + "right": "issue:7210", + "accept": false, + "reason": "Different attribute errors in different wrappers/paths; not the same bug." + }, + { + "left": "issue:852", + "right": "issue:859", + "accept": true, + "reason": "Both report the same device_map unexpected-keyword regression in initialization; same underlying API mismatch." + }, + { + "left": "issue:1263", + "right": "issue:1444", + "accept": true, + "reason": "These are the same missing StableDiffusionSafetyChecker import error, phrased slightly differently." + }, + { + "left": "issue:7133", + "right": "issue:9841", + "accept": false, + "reason": "Different DreamBooth SDXL LoRA problems: checkpoint resume size mismatch vs use_dora TypeError." + }, + { + "left": "issue:4483", + "right": "issue:4724", + "accept": false, + "reason": "Both involve SDXL pipelines, but the reported bugs are different (determinism vs checkpoint/pipeline behavior)." + }, + { + "left": "issue:1255", + "right": "issue:13087", + "accept": false, + "reason": "Both are missing-import errors, but they are different symbols and unrelated fixes." + }, + { + "left": "issue:10066", + "right": "issue:11410", + "accept": false, + "reason": "Both are attention-related unpacking errors, but the expected arity and failing call sites differ, so they are not clearly the same bug." + }, + { + "left": "issue:1082", + "right": "issue:1946", + "accept": false, + "reason": "Different low-level CUDA failures in different contexts; too generic to merge." + }, + { + "left": "issue:1260", + "right": "issue:9062", + "accept": false, + "reason": "Different missing import symbols; not the same issue." + }, + { + "left": "issue:10591", + "right": "issue:8534", + "accept": false, + "reason": "Both are SD3 training-script complaints, but they describe different problems in different scripts." + }, + { + "left": "issue:2044", + "right": "issue:2132", + "accept": false, + "reason": "Different attribute errors on different objects; unrelated failures." + }, + { + "left": "issue:13124", + "right": "issue:6363", + "accept": false, + "reason": "Same broad error family, but the scripts and trigger differ; not enough evidence of one identical underlying bug." + }, + { + "left": "issue:4872", + "right": "issue:6697", + "accept": false, + "reason": "Both are OOMs in SDXL finetuning, but the reports are too general and context-specific to be the same concrete bug." + }, + { + "left": "issue:3728", + "right": "issue:8414", + "accept": false, + "reason": "Completely different model domains and feature requests; not duplicates." + }, + { + "left": "issue:1260", + "right": "issue:2294", + "accept": false, + "reason": "Both are import errors, but they concern different pipeline classes and different missing names." + }, + { + "left": "issue:1572", + "right": "issue:3420", + "accept": false, + "reason": "Both mention pipeline incompatibilities, but the affected pipelines and failure modes differ." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet, and the soft-edge decisions are conservative overall. The accepted pairs are well-supported duplicates (same FP16-unscale error, same device_map init regression, same SafetyChecker import error), while the rejections stay cautious where the packet only shows shared symptoms or related but distinct code paths. The chosen canonical issue #6454 is also a reasonable generic representative of the FP16-unscale cluster." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 19, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10066", + "issue:10453", + "issue:10860", + "issue:10963", + "issue:11790", + "issue:12096", + "issue:12141", + "issue:12492", + "issue:1255", + "issue:1260", + "issue:1263", + "issue:13124", + "issue:1614", + "issue:1734", + "issue:1946", + "issue:1976", + "issue:2296", + "issue:246", + "issue:2947", + "issue:2984", + "issue:3382", + "issue:3812", + "issue:409", + "issue:4241", + "issue:4446", + "issue:4615", + "issue:4911", + "issue:5016", + "issue:5197", + "issue:5594", + "issue:6435", + "issue:6446", + "issue:6482", + "issue:6636", + "issue:6894", + "issue:7041", + "issue:7049", + "issue:7333", + "issue:7516", + "issue:7766", + "issue:7940", + "issue:8817", + "issue:9051", + "issue:9062", + "issue:9396", + "issue:9534", + "issue:9889" + ], + "soft_pairs": [ + "issue:10963|issue:4615", + "issue:246|issue:409", + "issue:10963|issue:1263", + "issue:10066|issue:7049", + "issue:1255|issue:6636", + "issue:7766|issue:8817", + "issue:3812|issue:6435", + "issue:12096|issue:12141", + "issue:10453|issue:9051", + "issue:1260|issue:4615", + "issue:4446|issue:4911", + "issue:7940|issue:9396", + "issue:12492|issue:7516", + "issue:1976|issue:2296", + "issue:5016|issue:6894", + "issue:2984|issue:7333", + "issue:13124|issue:6482", + "issue:4241|issue:5594", + "issue:7041|issue:9889", + "issue:10860|issue:11790", + "issue:2947|issue:9534", + "issue:1734|issue:5197", + "issue:1614|issue:6446", + "issue:4615|issue:9062", + "issue:1946|issue:3382" + ], + "prepared_review_unit_hash": "e303f34fa6daec21415633e668d593403af0ddc737a420d74d37847688d7dec0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33895, + "estimated_input_tokens": 8474, + "estimated_eval_tokens": 17204 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a loose lexical cluster, not one coherent duplicate group. I found only two true duplicate pairs; the rest are related only at a broad subsystem or symptom level and should stay separate.", + "confidence": 0.91, + "canonical_issue_reason": "No single canonical issue fits the whole cluster: most items are unrelated import errors, training bugs, pipeline limitations, or feature requests. If forced to anchor the only exact duplicate pair, issue 1255 is the cleanest representative of that narrow import-error duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1255 is the strongest issue candidate for a canonical anchor because it is a clear, specific import-error report with a near-verbatim duplicate (1260). It is still not a good representative of the full set, which is too heterogeneous.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10963", + "right": "issue:4615", + "accept": false, + "reason": "Both are import errors from diffusers, but they concern different missing symbols ('AutoencoderKLWan' vs 'StableDiffusionControlnetPipeline') and likely different code paths." + }, + { + "left": "issue:246", + "right": "issue:409", + "accept": false, + "reason": "Both mention missing transformers-era APIs, but the actual failures are unrelated attributes ('FeatureExtractionMixin' vs 'enable_attention_slicing')." + }, + { + "left": "issue:10963", + "right": "issue:1263", + "accept": false, + "reason": "Same broad theme of missing imports, but different classes/modules and no evidence of the same underlying bug." + }, + { + "left": "issue:10066", + "right": "issue:7049", + "accept": false, + "reason": "Both are unpacking errors, but the expected tuple arity differs and the contexts are different enough to suggest separate bugs." + }, + { + "left": "issue:1255", + "right": "issue:6636", + "accept": false, + "reason": "Both are import-related, but one is a scheduler symbol and the other is a utils constant; not the same underlying issue." + }, + { + "left": "issue:7766", + "right": "issue:8817", + "accept": false, + "reason": "Both concern IP-Adapter FaceID usage, but one is a how-to/question thread and the other is a concrete failure report; too little evidence they share one bug." + }, + { + "left": "issue:3812", + "right": "issue:6435", + "accept": false, + "reason": "Both are default/config initialization problems, but they refer to different fields and different models." + }, + { + "left": "issue:12096", + "right": "issue:12141", + "accept": false, + "reason": "Both are WanVACEPipeline complaints, but one is about group offloading and the other is about bf16/fp16 VAE handling; different failures." + }, + { + "left": "issue:10453", + "right": "issue:9051", + "accept": false, + "reason": "Both are batch-size limitations, but they are for different pipelines/models and do not look like one concrete bug." + }, + { + "left": "issue:1260", + "right": "issue:4615", + "accept": false, + "reason": "Both are import errors, but for different public classes; not the same missing-export bug." + }, + { + "left": "issue:4446", + "right": "issue:4911", + "accept": true, + "reason": "Same StableDiffusionPipeline.from_single_file custom text_encoder loading failure, described with nearly identical wording." + }, + { + "left": "issue:7940", + "right": "issue:9396", + "accept": false, + "reason": "Both are Korean translation tracking issues, not the same bug or change request." + }, + { + "left": "issue:12492", + "right": "issue:7516", + "accept": false, + "reason": "Different missing attributes in different objects/modules; same general API-version theme, but not the same bug." + }, + { + "left": "issue:1976", + "right": "issue:2296", + "accept": false, + "reason": "One is a bad file lookup during loading, the other is xformers behavior on T4; unrelated root causes." + }, + { + "left": "issue:5016", + "right": "issue:6894", + "accept": false, + "reason": "Both involve LoRA/dreambooth scripts, but one is about SDXL text-encoder training and the other about LoRA compatibility in A1111; not the same fix." + }, + { + "left": "issue:2984", + "right": "issue:7333", + "accept": false, + "reason": "Different video-generation research/feature threads with no concrete shared bug or change." + }, + { + "left": "issue:13124", + "right": "issue:6482", + "accept": false, + "reason": "Both mention FP16 unscale errors, but the triggering conditions differ and one is specifically tied to validation prompt handling while the other is resume-from-checkpoint." + }, + { + "left": "issue:4241", + "right": "issue:5594", + "accept": false, + "reason": "Both mention LoRA, but one asks about unidiffusers fine-tuning support and the other is about SDXL memory usage; unrelated." + }, + { + "left": "issue:7041", + "right": "issue:9889", + "accept": false, + "reason": "Both are segfaults, but on different inference setups and with no shared code-path evidence." + }, + { + "left": "issue:10860", + "right": "issue:11790", + "accept": false, + "reason": "Both are LoRA weight-loading problems, but one is specific to ComfyUI incompatibility and the other is a generic load failure in a different training script." + }, + { + "left": "issue:2947", + "right": "issue:9534", + "accept": false, + "reason": "Different missing attributes on different model classes; not the same underlying API regression." + }, + { + "left": "issue:1734", + "right": "issue:5197", + "accept": false, + "reason": "Both concern multi-GPU performance, but they describe different training scripts and one is slowdown while the other is OOM; too broad to merge." + }, + { + "left": "issue:1614", + "right": "issue:6446", + "accept": false, + "reason": "Different symptoms in stable diffusion loading/inference; one is blank output under autocast, the other is model loading on Colab." + }, + { + "left": "issue:4615", + "right": "issue:9062", + "accept": false, + "reason": "Both are missing imports from diffusers, but they are for different pipelines/classes and are not the same export issue." + }, + { + "left": "issue:1946", + "right": "issue:3382", + "accept": false, + "reason": "One is an xformers CUDA invalid-argument crash, the other is multi-GPU OOM; different failure modes and fixes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The cluster does look heterogeneous, and the two near-duplicate issue pairs you singled out are well supported by the packet: 1255/1260 and 4446/4911. The remaining soft-edge rejections are conservative and do not overclaim equivalence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 20, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1013", + "issue:10436", + "issue:10752", + "issue:1100", + "issue:11033", + "issue:11853", + "issue:12195", + "issue:1246", + "issue:12529", + "issue:1255", + "issue:1315", + "issue:13258", + "issue:1377", + "issue:1611", + "issue:1925", + "issue:1988", + "issue:2947", + "issue:3094", + "issue:3140", + "issue:3382", + "issue:4186", + "issue:4213", + "issue:4615", + "issue:4724", + "issue:4736", + "issue:4969", + "issue:5246", + "issue:5510", + "issue:5563", + "issue:5873", + "issue:6124", + "issue:6146", + "issue:6298", + "issue:6454", + "issue:6697", + "issue:7114", + "issue:7146", + "issue:7185", + "issue:7210", + "issue:7228", + "issue:7854", + "issue:858", + "issue:9421", + "issue:9497", + "issue:9534", + "issue:9790", + "issue:9841", + "issue:9970" + ], + "soft_pairs": [ + "issue:5563|issue:9841", + "issue:10436|issue:9421", + "issue:1925|issue:4186", + "issue:3382|issue:6146", + "issue:1013|issue:1611", + "issue:9497|issue:9790", + "issue:5563|issue:9970", + "issue:1255|issue:4615", + "issue:1988|issue:7854", + "issue:5873|issue:7185", + "issue:3094|issue:4736", + "issue:11033|issue:4213", + "issue:2947|issue:7228", + "issue:1377|issue:4969", + "issue:6298|issue:7210", + "issue:1100|issue:9534", + "issue:12195|issue:5246", + "issue:12529|issue:13258", + "issue:3094|issue:6697", + "issue:11853|issue:4724", + "issue:1315|issue:3140", + "issue:7210|issue:858", + "issue:1246|issue:6454", + "issue:10752|issue:6124", + "issue:7114|issue:7146", + "issue:1988|issue:5510" + ], + "prepared_review_unit_hash": "15d5746fc52be98ddaaa0894bd21db42991798336a148bf0d50ed56056135fed", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35100, + "estimated_input_tokens": 8775, + "estimated_eval_tokens": 17806 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a loose similarity cluster spanning several unrelated Diffusers issues, with one clear subcluster around the DreamBooth/LoRA FP16 unscale error and a smaller pair around StableDiffusionMegaPipeline / stable-diffusion-inpainting loading.", + "confidence": 0.72, + "canonical_issue_reason": "issue:6454 is the cleanest representative of the recurring FP16-unscale bug family: it has the generic error title, aligns with the accepted DreamBooth/LoRA duplicates, and is less script-specific than the older example reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6454 is the best cluster representative because it names the core failure directly and captures the broadest version of the shared training bug without tying it to one notebook or example.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5563", + "right": "issue:9841", + "accept": false, + "reason": "Same broad DreamBooth/SDXL LoRA area, but 9841 is a specific use_dora TypeError while 5563 is only a generic umbrella report; not clearly the same bug." + }, + { + "left": "issue:10436", + "right": "issue:9421", + "accept": true, + "reason": "Both report the same model repo, runwayml/stable-diffusion-inpainting, failing to load/access; this looks like the same underlying availability/loading problem." + }, + { + "left": "issue:1925", + "right": "issue:4186", + "accept": false, + "reason": "Both mention inpainting, but one is SD 2 inpainting half-precision and the other is SDXL inpaint with refiner; different model paths and failure modes." + }, + { + "left": "issue:3382", + "right": "issue:6146", + "accept": false, + "reason": "Both involve multi-GPU training, but one is an OOM in train_text_to_image and the other is a general accelerate-based multi-GPU training question; not the same concrete bug." + }, + { + "left": "issue:1013", + "right": "issue:1611", + "accept": true, + "reason": "These are essentially the same StableDiffusionMegaPipeline regression with runwayml/stable-diffusion-v1-5 after v0.6.0; same pipeline and same breakage." + }, + { + "left": "issue:9497", + "right": "issue:9790", + "accept": false, + "reason": "Both are Flux DreamBooth training reports, but the concrete failures differ (mat2 1-D tensor vs general multi-GPU training problem)." + }, + { + "left": "issue:5563", + "right": "issue:9970", + "accept": false, + "reason": "Both are generic SDXL DreamBooth LoRA training failures, but the reports are too broad to show the same underlying code-path bug." + }, + { + "left": "issue:1255", + "right": "issue:4615", + "accept": false, + "reason": "Both are import errors, but they concern different missing symbols (DPMSolverMultistepScheduler vs StableDiffusionControlnetPipeline) and are not the same issue." + }, + { + "left": "issue:1988", + "right": "issue:7854", + "accept": false, + "reason": "Same high-level dtype mismatch symptom, but they occur in different contexts/pipelines; not enough evidence of one shared bug." + }, + { + "left": "issue:5873", + "right": "issue:7185", + "accept": false, + "reason": "ONNX export of scaled_dot_product_attention and general PyTorch 2.2 compatibility are different problems." + }, + { + "left": "issue:3094", + "right": "issue:4736", + "accept": false, + "reason": "Both mention memory/instability during training, but they are different scripts and one includes an additional invalid-cast symptom; too broad to merge." + }, + { + "left": "issue:11033", + "right": "issue:4213", + "accept": false, + "reason": "Both are from_single_file loading problems, but one is SD1.5 UNet loading and the other is SDXL without accelerate; different code paths." + }, + { + "left": "issue:2947", + "right": "issue:7228", + "accept": false, + "reason": "Different attribute errors on different objects ('load_attn_procs' vs 'dtype'); not the same underlying bug." + }, + { + "left": "issue:1377", + "right": "issue:4969", + "accept": false, + "reason": "Different missing imports from different packages/classes; only similar by being ImportErrors." + }, + { + "left": "issue:6298", + "right": "issue:7210", + "accept": false, + "reason": "CLIPImageProjection missing in GLIGEN and named_parameters missing with accelerate dispatch are unrelated failures." + }, + { + "left": "issue:1100", + "right": "issue:9534", + "accept": false, + "reason": "A CLIPTextTransformer device attribute issue and a FluxMultiControlNetModel config issue are unrelated." + }, + { + "left": "issue:12195", + "right": "issue:5246", + "accept": false, + "reason": "One is a torch.library.custom_op compatibility failure in AutoencoderKL; the other is a missing CLIPImageProjection attribute. Different bugs." + }, + { + "left": "issue:12529", + "right": "issue:13258", + "accept": false, + "reason": "Same model family, but one is a tiled VAE tensor size mismatch and the other is a missing image-to-video model; not the same change or bug." + }, + { + "left": "issue:3094", + "right": "issue:6697", + "accept": false, + "reason": "Both are OOM reports during training, but on different training scripts and settings; memory pressure alone is too broad." + }, + { + "left": "issue:11853", + "right": "issue:4724", + "accept": false, + "reason": "Different SDXL composition problems: controlnet+IP-Adapter size conflict versus img2img with the base checkpoint." + }, + { + "left": "issue:1315", + "right": "issue:3140", + "accept": false, + "reason": "Different missing imports affecting different symbols and pipelines; not a duplicate pair." + }, + { + "left": "issue:7210", + "right": "issue:858", + "accept": false, + "reason": "AttributeError on named_parameters during accelerate dispatch and named_children on Parameter are unrelated failures." + }, + { + "left": "issue:1246", + "right": "issue:6454", + "accept": true, + "reason": "Both are the same FP16 unscale error in Dreambooth training/example code; this is the same underlying bug resurfacing in a newer report." + }, + { + "left": "issue:10752", + "right": "issue:6124", + "accept": true, + "reason": "Exact same FP16 gradient unscale error in DreamBooth/LoRA training, just reported in different wording and notebook/script contexts." + }, + { + "left": "issue:7114", + "right": "issue:7146", + "accept": false, + "reason": "Both are SDXL training OOM reports, but they are separate scripts and likely separate tuning situations rather than one concrete bug." + }, + { + "left": "issue:1988", + "right": "issue:5510", + "accept": false, + "reason": "Same runtime error string, but the reports are from different pipelines and contexts; not enough to treat as one fixable bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The representative choice and canonicalization are grounded in the packet, and the accepted soft-edge pairs are conservative matches. The only notable cluster is the FP16 unscale DreamBooth/LoRA family, with a couple of separate loading/regression pairs elsewhere; no verdict looks overstated." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 21, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10436", + "issue:10722", + "issue:10969", + "issue:1100", + "issue:11033", + "issue:11104", + "issue:11143", + "issue:12009", + "issue:12195", + "issue:1255", + "issue:12826", + "issue:13087", + "issue:1444", + "issue:3816", + "issue:4052", + "issue:4086", + "issue:4118", + "issue:4448", + "issue:4615", + "issue:4724", + "issue:4736", + "issue:4848", + "issue:5029", + "issue:5565", + "issue:5603", + "issue:5607", + "issue:5960", + "issue:6044", + "issue:6101", + "issue:6131", + "issue:6415", + "issue:6446", + "issue:6570", + "issue:6598", + "issue:6734", + "issue:7432", + "issue:7506", + "issue:7666", + "issue:7681", + "issue:7708", + "issue:8537", + "issue:8826", + "issue:9111", + "issue:9171", + "issue:9841", + "issue:989", + "issue:993", + "issue:9970" + ], + "soft_pairs": [ + "issue:1100|issue:9111", + "issue:11104|issue:11143", + "issue:10722|issue:6570", + "issue:9841|issue:9970", + "issue:1444|issue:4615", + "issue:3816|issue:7681", + "issue:12009|issue:7506", + "issue:6598|issue:7432", + "issue:4052|issue:8537", + "issue:10969|issue:989", + "issue:12195|issue:4118", + "issue:4086|issue:7708", + "issue:5607|issue:8826", + "issue:5565|issue:9970", + "issue:6131|issue:6415", + "issue:11033|issue:9171", + "issue:4448|issue:6101", + "issue:1255|issue:12826", + "issue:13087|issue:5603", + "issue:10436|issue:6446", + "issue:1100|issue:6044", + "issue:4736|issue:5960", + "issue:4848|issue:7666", + "issue:4724|issue:993", + "issue:5029|issue:6734" + ], + "prepared_review_unit_hash": "35e5abdeab95fc3796c97765b0a133ae52f3c35982bcc3f92e856e6a792e06fb", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 35003, + "estimated_input_tokens": 8751, + "estimated_eval_tokens": 17758 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated diffusers loading/training/import issues. None of the soft pairs look like true duplicates or the same fixable code-path bug, so I would not merge any of them into a single canonical artifact. There are no PRs in the set.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1100", + "right": "issue:9111", + "accept": false, + "reason": "Both are runtime complaints, but one is a CLIPTextTransformer device attribute error and the other is BitDiffusion missing a scheduler. Different components and failures." + }, + { + "left": "issue:11104", + "right": "issue:11143", + "accept": false, + "reason": "Offload-method breakage and poor generated video quality are different problems; same model family, not the same bug." + }, + { + "left": "issue:10722", + "right": "issue:6570", + "accept": false, + "reason": "Both are tensor-size mismatch errors, but they occur in different pipelines/workflows (DreamBooth prior preservation vs SDXL InstructPix2Pix). Too broad to be the same bug." + }, + { + "left": "issue:9841", + "right": "issue:9970", + "accept": false, + "reason": "Both involve SDXL DreamBooth/LoRA training, but the reported failures differ (use_dora TypeError vs generic script failure). Not enough evidence of one underlying defect." + }, + { + "left": "issue:1444", + "right": "issue:4615", + "accept": false, + "reason": "Both are import errors, but for different symbols and different code paths. Not the same missing-export issue." + }, + { + "left": "issue:3816", + "right": "issue:7681", + "accept": false, + "reason": "CLIPImageProcessor missing from transformers and an OpenSSL/lib import failure are unrelated." + }, + { + "left": "issue:12009", + "right": "issue:7506", + "accept": false, + "reason": "Both mention meta tensors, but they affect different models and loaders (Wan GGUF vs CLIPTextModelWithProjection). Not the same bug." + }, + { + "left": "issue:6598", + "right": "issue:7432", + "accept": false, + "reason": "Both discuss image quality differences versus WebUI, but they are subjective output comparisons rather than one concrete defect." + }, + { + "left": "issue:4052", + "right": "issue:8537", + "accept": false, + "reason": "Missing from_single_file and missing torch.float8_e4m3fn are unrelated compatibility problems." + }, + { + "left": "issue:10969", + "right": "issue:989", + "accept": false, + "reason": "Both say 'weight must be 2-D', but one is FLUX-controlnet zero3 training and the other is a different generic runtime error. Too little evidence they are the same code-path failure." + }, + { + "left": "issue:12195", + "right": "issue:4118", + "accept": false, + "reason": "Torch custom_op missing and AutoencoderKL lacking from_single_file are different API/version issues." + }, + { + "left": "issue:4086", + "right": "issue:7708", + "accept": false, + "reason": "ControlNet training failure and missing tokenizer/config.json for an SDXL checkpoint are different problems." + }, + { + "left": "issue:5607", + "right": "issue:8826", + "accept": false, + "reason": "Tensor-size mismatch and torch.cat on an empty list are different runtime failures." + }, + { + "left": "issue:5565", + "right": "issue:9970", + "accept": false, + "reason": "Both are DreamBooth SDXL LoRA training related, but one is an example script usability issue and the other is a training failure. Not the same bug." + }, + { + "left": "issue:6131", + "right": "issue:6415", + "accept": false, + "reason": "IP-Adapter Plus support issues are related, but one is a general 'not working' report and the other is a FaceID/FaceID Plus support request. Different concrete artifacts." + }, + { + "left": "issue:11033", + "right": "issue:9171", + "accept": false, + "reason": "Both touch from_single_file loading for SD1.5, but 9171 is a specific v_prediction regression while 11033 is a broader UNet loading failure. Not clearly the same fix." + }, + { + "left": "issue:4448", + "right": "issue:6101", + "accept": false, + "reason": "Both compare diffusers output quality to A1111/controlnet, but one is outpainting quality and the other is inpaint pipeline performance. Different issues." + }, + { + "left": "issue:1255", + "right": "issue:12826", + "accept": false, + "reason": "Missing DPMSolverMultistepScheduler and missing export_to_video are unrelated import/export problems." + }, + { + "left": "issue:13087", + "right": "issue:5603", + "accept": false, + "reason": "Missing Flux2KleinPipeline and missing is_safetensors_available are unrelated API surface issues." + }, + { + "left": "issue:10436", + "right": "issue:6446", + "accept": false, + "reason": "Both are model loading failures, but for different checkpoints and symptoms. Not the same underlying load bug." + }, + { + "left": "issue:1100", + "right": "issue:6044", + "accept": false, + "reason": "A device attribute error on CLIPTextTransformer is unrelated to StableDiffusionImg2ImgPipeline missing do_classifier_free_guidance." + }, + { + "left": "issue:4736", + "right": "issue:5960", + "accept": false, + "reason": "Both mention invalid casts during training, but one also includes OOM and the other is specifically ControlNet SDXL image casting. Not the same concrete defect." + }, + { + "left": "issue:4848", + "right": "issue:7666", + "accept": false, + "reason": "A custom pipeline not working and that same pipeline not working with from_single_file are related, but the reports are too broad to confirm one identical bug." + }, + { + "left": "issue:4724", + "right": "issue:993", + "accept": false, + "reason": "SDXL img2img behavior and fp16 dtype mismatch are different issues." + }, + { + "left": "issue:5029", + "right": "issue:6734", + "accept": false, + "reason": "AutoencoderTinyOutput missing latent_dist and a NoneType tokenize error are unrelated failures." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 22, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10472", + "issue:10839", + "issue:1260", + "issue:13087", + "issue:1315", + "issue:1444", + "issue:1500", + "issue:1604", + "issue:1709", + "issue:2044", + "issue:2294", + "issue:2548", + "issue:2780", + "issue:3140", + "issue:3325", + "issue:3565", + "issue:3639", + "issue:3728", + "issue:3787", + "issue:3788", + "issue:3973", + "issue:4186", + "issue:4206", + "issue:4529", + "issue:4615", + "issue:5374", + "issue:5563", + "issue:5661", + "issue:5894", + "issue:5897", + "issue:5972", + "issue:6044", + "issue:6230", + "issue:6277", + "issue:6967", + "issue:7002", + "issue:7146", + "issue:7483", + "issue:7708", + "issue:861", + "issue:9062", + "issue:9111", + "issue:9237", + "issue:945", + "issue:9962", + "issue:9970", + "issue:9973" + ], + "soft_pairs": [ + "issue:5563|issue:9973", + "issue:1709|issue:6044", + "issue:3325|issue:7146", + "issue:2294|issue:9062", + "issue:3639|issue:3728", + "issue:1260|issue:3787", + "issue:5894|issue:7002", + "issue:6967|issue:9970", + "issue:1315|issue:1444", + "issue:1500|issue:1604", + "issue:10472|issue:2548", + "issue:2780|issue:6277", + "issue:4529|issue:9962", + "issue:13087|issue:4615", + "issue:2044|issue:9111", + "issue:861|issue:945", + "issue:5661|issue:7483", + "issue:3973|issue:5374", + "issue:4206|issue:5972", + "issue:3639|issue:3788", + "issue:3325|issue:5897", + "issue:10839|issue:9237", + "issue:4186|issue:7708", + "issue:3565|issue:6230", + "issue:3140|issue:3787" + ], + "prepared_review_unit_hash": "a989c457dab79a77c996acfa09dff927fb61d3be5cb97ab082c4df17536be373", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 34078, + "estimated_input_tokens": 8520, + "estimated_eval_tokens": 17296 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: import errors, OOMs, pipeline requests, and several unrelated training bugs. Only one soft pair looks like a strong duplicate candidate (the LoRA gradient issue around train_text_to_image_lora.py).", + "confidence": 0.91, + "canonical_issue_reason": "If a canonical issue must be chosen, issue 2780 is the most concrete and actionable bug report: it names the exact script and exact failure (`Attempting to unscale FP16 gradients`), and it has the clearest duplicate-looking counterpart (6277).", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 2780 is the best representative because it is specific, reproducible, and tightly scoped to a single code path; most other issues here are either generic, version-specific import errors, or separate feature requests.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5563", + "right": "issue:9973", + "accept": false, + "reason": "Both are SDXL/DreamBooth/LoRA training reports, but they describe different failures: generic training errors vs an FP16 unscale error at the last step." + }, + { + "left": "issue:1709", + "right": "issue:6044", + "accept": false, + "reason": "Different pipelines and different exceptions (`astype` on Tensor vs missing `do_classifier_free_guidance`)." + }, + { + "left": "issue:3325", + "right": "issue:7146", + "accept": false, + "reason": "Same broad training area, but one is CUDA invalid-argument in DreamBooth and the other is a clear OOM in SDXL text-to-image." + }, + { + "left": "issue:2294", + "right": "issue:9062", + "accept": false, + "reason": "Both are missing-import issues, but they are for different symbols and different pipelines (`StableDiffusionVideoInpaintPipeline` vs `StableAudioPipeline`)." + }, + { + "left": "issue:3639", + "right": "issue:3728", + "accept": false, + "reason": "Different new pipelines/model proposals; no shared bug or code path." + }, + { + "left": "issue:1260", + "right": "issue:3787", + "accept": false, + "reason": "Both are import failures, but one is a missing top-level symbol and the other is a partially initialized-module import cycle." + }, + { + "left": "issue:5894", + "right": "issue:7002", + "accept": false, + "reason": "Similar error wording, but 7002 is specifically an img2img pipeline buffer-size failure while 5894 is a generic invalid-buffer-size report without enough evidence of the same root cause." + }, + { + "left": "issue:6967", + "right": "issue:9970", + "accept": false, + "reason": "Both concern SDXL DreamBooth LoRA training, but one is a prior-preservation shape mismatch and the other is a vague training failure; not enough to treat as the same bug." + }, + { + "left": "issue:1315", + "right": "issue:1444", + "accept": false, + "reason": "Different missing imports (`PIL_INTERPOLATION` vs `StableDiffusionSafetyChecker`)." + }, + { + "left": "issue:1500", + "right": "issue:1604", + "accept": false, + "reason": "Same file, but the reports are tied to different Diffusers versions and do not clearly describe the same concrete regression." + }, + { + "left": "issue:10472", + "right": "issue:2548", + "accept": false, + "reason": "Both are LoRA-related, but one is an inference-time unexpected kwarg (`lora_bias`) and the other is a saving-time finetuning error." + }, + { + "left": "issue:2780", + "right": "issue:6277", + "accept": true, + "reason": "Same example script and same gradient-related failure domain; the titles strongly indicate the same underlying LoRA training bug." + }, + { + "left": "issue:4529", + "right": "issue:9962", + "accept": false, + "reason": "Different unexpected keyword arguments in different code paths (`network_alpha` vs `use_cuda_graph`)." + }, + { + "left": "issue:13087", + "right": "issue:4615", + "accept": false, + "reason": "Different missing imports for different pipeline classes." + }, + { + "left": "issue:2044", + "right": "issue:9111", + "accept": false, + "reason": "Unrelated failures: DDP has no `dtype` vs BitDiffusion missing `scheduler`." + }, + { + "left": "issue:861", + "right": "issue:945", + "accept": false, + "reason": "Both are model-conversion related, but one is ckpt-to-diffusers and the other is ckpt-to-ONNX; different tools and failure modes." + }, + { + "left": "issue:5661", + "right": "issue:7483", + "accept": false, + "reason": "Different problems: a missing tokenizer type in 0.22.1 vs SDXL LoRA-to-Diffusers conversion failure." + }, + { + "left": "issue:3973", + "right": "issue:5374", + "accept": false, + "reason": "Both are unexpected-kwarg errors in conversion utilities, but they involve different functions and different parameters." + }, + { + "left": "issue:4206", + "right": "issue:5972", + "accept": false, + "reason": "One is a broken SDXL controlnet training path; the other is documentation for Min-SNR weighting." + }, + { + "left": "issue:3639", + "right": "issue:3788", + "accept": false, + "reason": "Both are new pipeline proposals, but they are for different models and not the same change." + }, + { + "left": "issue:3325", + "right": "issue:5897", + "accept": false, + "reason": "Different training failures: CUDA invalid argument vs device mismatch between CPU and CUDA." + }, + { + "left": "issue:10839", + "right": "issue:9237", + "accept": false, + "reason": "Both mention Flux DreamBooth LoRA training, but the reports are too generic to confirm the same concrete bug." + }, + { + "left": "issue:4186", + "right": "issue:7708", + "accept": false, + "reason": "Completely different issues: SDXL inpaint/refiner behavior vs missing tokenizer config for a base model." + }, + { + "left": "issue:3565", + "right": "issue:6230", + "accept": false, + "reason": "Both are OOM reports, but they concern different training scripts/models and do not look like one mergeable bug." + }, + { + "left": "issue:3140", + "right": "issue:3787", + "accept": false, + "reason": "Different import failures with different root causes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative on the soft-edge pairs. The one accepted pair (issue:2780 / issue:6277) is well supported by the shared example script and gradient-failure wording, and the rejected pairs are generally justified by clear differences in symbols, pipelines, error types, or issue scope. No obvious overstatement of evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 23, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10436", + "issue:10453", + "issue:10612", + "issue:10752", + "issue:10928", + "issue:10963", + "issue:11214", + "issue:12195", + "issue:1246", + "issue:1249", + "issue:12778", + "issue:13073", + "issue:1315", + "issue:1444", + "issue:1544", + "issue:1946", + "issue:1976", + "issue:246", + "issue:255", + "issue:3140", + "issue:4224", + "issue:4356", + "issue:4409", + "issue:4488", + "issue:4566", + "issue:4840", + "issue:5374", + "issue:539", + "issue:5595", + "issue:5840", + "issue:6126", + "issue:6363", + "issue:6741", + "issue:7146", + "issue:7210", + "issue:7228", + "issue:7644", + "issue:7716", + "issue:7784", + "issue:8307", + "issue:8534", + "issue:8788", + "issue:8887", + "issue:9067", + "issue:9450", + "issue:9895", + "issue:9914", + "issue:9962" + ], + "soft_pairs": [ + "issue:8534|issue:8887", + "issue:10612|issue:9914", + "issue:12195|issue:8307", + "issue:4356|issue:9067", + "issue:4488|issue:4840", + "issue:1946|issue:7146", + "issue:255|issue:5595", + "issue:1249|issue:246", + "issue:10928|issue:7716", + "issue:10453|issue:13073", + "issue:1544|issue:7210", + "issue:10436|issue:1976", + "issue:7228|issue:7784", + "issue:7644|issue:8788", + "issue:4566|issue:5840", + "issue:10963|issue:1315", + "issue:12778|issue:6741", + "issue:1249|issue:7210", + "issue:1444|issue:3140", + "issue:4409|issue:6126", + "issue:539|issue:9962", + "issue:4224|issue:6126", + "issue:1246|issue:6363", + "issue:10752|issue:1246", + "issue:11214|issue:5374", + "issue:9450|issue:9895" + ], + "prepared_review_unit_hash": "1e8114d3be6be3050f2e26cea6d9b8ca93875b84ce65341a25b6a5a1ab0ed90d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35105, + "estimated_input_tokens": 8777, + "estimated_eval_tokens": 17810 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are only loosely related and should stay separate. I accepted only clear same-bug duplicates around FP16 unscale failures, SDXL checkpoint-resume, and Flux device-placement regression.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 6363 is the strongest duplicate anchor: it has the exact FP16 unscale error and specific training-script context, making it the most representative of the accepted duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Best overall issue because it is precise, reproducible, and clearly shared by multiple near-duplicate reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8534", + "right": "issue:8887", + "accept": false, + "reason": "Both are SD3 training-script problems, but they describe different bugs: logit-normal weighting vs. VAE encode." + }, + { + "left": "issue:10612", + "right": "issue:9914", + "accept": false, + "reason": "Both involve Flux LoRA loading, but one is NF4 fill-pipeline loading and the other is Xlabs-trained LoRA loading; not the same code path." + }, + { + "left": "issue:12195", + "right": "issue:8307", + "accept": false, + "reason": "Unrelated failures: a PyTorch custom_op compatibility error vs. a missing StableCascadeUNet export." + }, + { + "left": "issue:4356", + "right": "issue:9067", + "accept": false, + "reason": "Both are import/module errors, but they refer to different missing modules and likely different breakages." + }, + { + "left": "issue:4488", + "right": "issue:4840", + "accept": false, + "reason": "Different loading bugs in different pipelines/models; not a single concrete duplicate." + }, + { + "left": "issue:1946", + "right": "issue:7146", + "accept": false, + "reason": "xformers CUDA invalid-argument and SDXL fine-tuning OOM are different failure modes." + }, + { + "left": "issue:255", + "right": "issue:5595", + "accept": false, + "reason": "These are about image size validation, but one reports an error on valid divisible-by-8 sizes while the other reports an expected error for non-divisible dimensions." + }, + { + "left": "issue:1249", + "right": "issue:246", + "accept": false, + "reason": "Completely different problems: an ONNX pipeline ndarray/device issue vs. a missing transformers mixin import." + }, + { + "left": "issue:10928", + "right": "issue:7716", + "accept": false, + "reason": "Both are missing-file OSErrors, but the missing files and model directories are different." + }, + { + "left": "issue:10453", + "right": "issue:13073", + "accept": false, + "reason": "Both are unsupported-parameter complaints, but one is batch size and the other is guidance_scale; unrelated bugs." + }, + { + "left": "issue:1544", + "right": "issue:7210", + "accept": false, + "reason": "Different components and different missing attributes: scheduler get_velocity vs. pipeline named_parameters." + }, + { + "left": "issue:10436", + "right": "issue:1976", + "accept": false, + "reason": "Both involve loading models from Hugging Face, but the failing models and incorrect paths/files are different." + }, + { + "left": "issue:7228", + "right": "issue:7784", + "accept": false, + "reason": "Different missing attributes on different pipeline classes; not the same underlying bug." + }, + { + "left": "issue:7644", + "right": "issue:8788", + "accept": false, + "reason": "Both mention non-power-of-2 images, but they affect different pipelines and appear to be different limitations." + }, + { + "left": "issue:4566", + "right": "issue:5840", + "accept": true, + "reason": "Same underlying SDXL LoRA checkpoint-resume bug, both centered on training resuming from checkpoint." + }, + { + "left": "issue:10963", + "right": "issue:1315", + "accept": false, + "reason": "Both are import errors, but they are for unrelated symbols and modules." + }, + { + "left": "issue:12778", + "right": "issue:6741", + "accept": false, + "reason": "FluxImageProcessor attribute error and Diffusion DPO argparse failure are unrelated." + }, + { + "left": "issue:1249", + "right": "issue:7210", + "accept": false, + "reason": "Unrelated errors in different areas: ONNX pipeline ndarray handling vs. pipeline checkpoint dispatch." + }, + { + "left": "issue:1444", + "right": "issue:3140", + "accept": false, + "reason": "Both are missing imports, but they are different pipeline classes with no clear shared bug." + }, + { + "left": "issue:4409", + "right": "issue:6126", + "accept": false, + "reason": "Different model classes and different missing attributes; not a single duplicate." + }, + { + "left": "issue:539", + "right": "issue:9962", + "accept": false, + "reason": "Both are unexpected keyword-argument errors, but the APIs and arguments are unrelated." + }, + { + "left": "issue:4224", + "right": "issue:6126", + "accept": false, + "reason": "Different attribute errors on different model types; no shared concrete failure." + }, + { + "left": "issue:1246", + "right": "issue:6363", + "accept": true, + "reason": "Same FP16 unscale gradient error, with the same underlying training bug." + }, + { + "left": "issue:10752", + "right": "issue:1246", + "accept": true, + "reason": "Same exact FP16 gradient unscale failure, just described at different levels of specificity." + }, + { + "left": "issue:11214", + "right": "issue:5374", + "accept": false, + "reason": "Different signatures and different errors: attention forward kwarg handling vs. checkpoint_path argument parsing." + }, + { + "left": "issue:9450", + "right": "issue:9895", + "accept": true, + "reason": "Both describe the same FluxPipeline device-placement regression leading to tensors on CPU and CUDA at the same time." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analyst output is conservative and grounded in the packet. The accepted pairs are plausible same-bug duplicates (FP16 unscale, SDXL checkpoint-resume, Flux device mismatch), and the rejected soft-edge pairs are generally separated by different error modes or different code paths. I don\u2019t see an overstatement of evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 24, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10928", + "issue:11014", + "issue:11362", + "issue:12157", + "issue:12174", + "issue:1246", + "issue:12492", + "issue:12529", + "issue:1263", + "issue:12826", + "issue:13087", + "issue:13124", + "issue:1552", + "issue:1709", + "issue:1925", + "issue:2089", + "issue:2132", + "issue:2947", + "issue:3040", + "issue:3279", + "issue:3420", + "issue:3794", + "issue:4073", + "issue:4086", + "issue:4186", + "issue:4304", + "issue:4409", + "issue:4483", + "issue:4733", + "issue:5054", + "issue:5374", + "issue:5518", + "issue:5857", + "issue:6028", + "issue:6353", + "issue:6382", + "issue:6427", + "issue:6482", + "issue:6636", + "issue:6697", + "issue:6955", + "issue:7178", + "issue:7330", + "issue:7784", + "issue:8182", + "issue:8667", + "issue:9044", + "issue:9901" + ], + "soft_pairs": [ + "issue:4409|issue:5518", + "issue:11014|issue:6028", + "issue:13124|issue:7330", + "issue:5857|issue:6382", + "issue:12157|issue:6353", + "issue:3040|issue:9901", + "issue:3794|issue:4304", + "issue:12529|issue:8667", + "issue:6427|issue:7178", + "issue:10928|issue:9044", + "issue:11362|issue:5374", + "issue:1709|issue:4073", + "issue:2089|issue:6697", + "issue:1246|issue:6482", + "issue:1925|issue:4086", + "issue:12174|issue:8182", + "issue:12492|issue:6955", + "issue:3279|issue:4733", + "issue:12826|issue:6636", + "issue:1552|issue:3420", + "issue:4186|issue:4483", + "issue:5054|issue:7784", + "issue:2132|issue:2947", + "issue:1263|issue:13087" + ], + "prepared_review_unit_hash": "5f5a308e8a393ad072ba352ce20eca79efc5a411d02f796f68a4c07d5e1c157f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34684, + "estimated_input_tokens": 8671, + "estimated_eval_tokens": 17598 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy and unrelated. The only strong duplicate subgroup is the recurring \"Attempting to unscale FP16 gradients\" training-script failures; the rest of the soft pairs are not the same underlying bug/change.", + "confidence": 0.64, + "canonical_issue_reason": "issue:7330 is the clearest representative of the recurring FP16 unscale-gradient bug: the title states the exact error and it matches multiple near-duplicate training reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:7330 is the best issue to keep as the representative report for the FP16 gradient-unscale problem because it is specific, explicit, and directly matches the repeated failure signature across scripts.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4409", + "right": "issue:5518", + "accept": false, + "reason": "Both are AttributeErrors, but they involve different missing attributes in different code paths ('alphas_cumprod' vs 'time')." + }, + { + "left": "issue:11014", + "right": "issue:6028", + "accept": false, + "reason": "Both mention CUDA OOM, but they are different models/workflows and do not indicate the same bug." + }, + { + "left": "issue:13124", + "right": "issue:7330", + "accept": true, + "reason": "Same concrete failure signature: \"Attempting to unscale FP16 gradients\" in DreamBooth/text-to-image training scripts; the validation_prompt mention looks like a trigger of the same bug." + }, + { + "left": "issue:5857", + "right": "issue:6382", + "accept": false, + "reason": "Different problems: LCM LoRA/AnimateDiff behavior versus an IP-adapter key error." + }, + { + "left": "issue:12157", + "right": "issue:6353", + "accept": false, + "reason": "Different failures in different training setups: bnb quantization config iteration error vs SDXL controlnet example failure." + }, + { + "left": "issue:3040", + "right": "issue:9901", + "accept": false, + "reason": "Unrelated AttributeErrors with different objects and code paths." + }, + { + "left": "issue:3794", + "right": "issue:4304", + "accept": false, + "reason": "These are feature requests about LoRA/SDXL support, not the same bug." + }, + { + "left": "issue:12529", + "right": "issue:8667", + "accept": false, + "reason": "Both are tensor-shape mismatch reports, but they affect different models/pipelines and likely different shape-handling code." + }, + { + "left": "issue:6427", + "right": "issue:7178", + "accept": false, + "reason": "Different functionality entirely: model conversion to .ckpt versus export_to_video." + }, + { + "left": "issue:10928", + "right": "issue:9044", + "accept": false, + "reason": "Both are missing-file OSErrors, but they refer to different filenames and model directories, so they are not clearly the same bug." + }, + { + "left": "issue:11362", + "right": "issue:5374", + "accept": false, + "reason": "Different unexpected-keyword-argument errors in different classes; no shared concrete code path is evident." + }, + { + "left": "issue:1709", + "right": "issue:4073", + "accept": false, + "reason": "Same symptom class ('astype' on Tensor) but in different pipelines/code paths; not enough to treat as the same bug." + }, + { + "left": "issue:2089", + "right": "issue:6697", + "accept": false, + "reason": "Both are CUDA OOM reports, but they concern different training setups and models." + }, + { + "left": "issue:1246", + "right": "issue:6482", + "accept": true, + "reason": "Same exact FP16 unscale-gradient error, with the SDXL advanced report describing the same training-time failure under resume_from_checkpoint." + }, + { + "left": "issue:1925", + "right": "issue:4086", + "accept": false, + "reason": "Different model/pipeline failures: half-precision in SD2 inpainting versus SDXL ControlNet training behavior." + }, + { + "left": "issue:12174", + "right": "issue:8182", + "accept": false, + "reason": "Unrelated AttributeErrors on different classes ('super' __getattr__ vs Logger.warninging)." + }, + { + "left": "issue:12492", + "right": "issue:6955", + "accept": false, + "reason": "Different import/version compatibility issues involving different symbols and modules." + }, + { + "left": "issue:3279", + "right": "issue:4733", + "accept": false, + "reason": "Both mention checkpoint resume problems, but the reported failures are not the same concrete bug or error signature." + }, + { + "left": "issue:12826", + "right": "issue:6636", + "accept": false, + "reason": "Different import errors for different symbols; not the same issue." + }, + { + "left": "issue:1552", + "right": "issue:3420", + "accept": false, + "reason": "Different pipeline/model behavior complaints; no shared bug." + }, + { + "left": "issue:4186", + "right": "issue:4483", + "accept": false, + "reason": "Different topics entirely: SDXL inpaint/refiner behavior versus generator determinism." + }, + { + "left": "issue:5054", + "right": "issue:7784", + "accept": false, + "reason": "Both are missing-attribute pipeline errors, but they affect different pipeline classes and methods." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The two accepted soft pairs are well-supported by the shared exact FP16 unscale-gradient error signature, while the rejections generally avoid over-calling similarity on only generic symptom overlap. No PR-pair issue is present." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 25, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10226", + "issue:10839", + "issue:11005", + "issue:11045", + "issue:11581", + "issue:12436", + "issue:1246", + "issue:12492", + "issue:1709", + "issue:1976", + "issue:1988", + "issue:2044", + "issue:246", + "issue:2571", + "issue:2616", + "issue:3382", + "issue:3565", + "issue:4137", + "issue:4308", + "issue:4324", + "issue:4492", + "issue:4649", + "issue:4736", + "issue:4868", + "issue:4925", + "issue:5016", + "issue:5259", + "issue:5805", + "issue:6044", + "issue:6098", + "issue:6142", + "issue:6281", + "issue:646", + "issue:6697", + "issue:7114", + "issue:7133", + "issue:7146", + "issue:7228", + "issue:7432", + "issue:7516", + "issue:7681", + "issue:8307", + "issue:858", + "issue:8613", + "issue:9227", + "issue:9393", + "issue:9548", + "issue:9901" + ], + "soft_pairs": [ + "issue:7228|issue:9901", + "issue:10226|issue:7133", + "issue:2616|issue:646", + "issue:4925|issue:6697", + "issue:11005|issue:4308", + "issue:246|issue:858", + "issue:2571|issue:6281", + "issue:1246|issue:6098", + "issue:4324|issue:4649", + "issue:7146|issue:8613", + "issue:1709|issue:8307", + "issue:3382|issue:3565", + "issue:4736|issue:7114", + "issue:10839|issue:9548", + "issue:11045|issue:9227", + "issue:4868|issue:9901", + "issue:5016|issue:9393", + "issue:4137|issue:7432", + "issue:7516|issue:858", + "issue:1976|issue:5259", + "issue:4492|issue:6142", + "issue:12492|issue:7681", + "issue:11581|issue:12436", + "issue:1988|issue:5805", + "issue:2044|issue:6044" + ], + "prepared_review_unit_hash": "8cd08451bcbb6ff7c9c0318d6ffcf2e4b5906fc1f322346ebac87f23e6ffd576", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 35073, + "estimated_input_tokens": 8769, + "estimated_eval_tokens": 17794 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a mix of unrelated issues. Two soft-edge pairs look like true duplicates: the FP16 unscale-gradients reports, and the NoneType/added_cond_kwargs error in UNet forward. The Flux/Dreambooth items are related in topic but not clearly the same underlying bug.", + "confidence": 0.88, + "canonical_issue_reason": "Issue #1246 is the strongest canonical representative: it is older, has much more discussion/inbound references, and matches #6098\u2019s exact FP16 unscale-gradients error.", + "canonical_pr_reason": null, + "best_issue_reason": "#1246 is the best overall issue candidate because it is the clearest established duplicate target among the provided items and has the most evidence/history.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7228", + "right": "issue:9901", + "accept": false, + "reason": "Different failure modes: traced-model dtype access vs AttnProcessor returning a tuple/shape error." + }, + { + "left": "issue:10226", + "right": "issue:7133", + "accept": false, + "reason": "Both mention size mismatch, but one is Flux Dreambooth LoRA and the other is SDXL resuming from checkpoint; not the same bug." + }, + { + "left": "issue:2616", + "right": "issue:646", + "accept": false, + "reason": "Unrelated errors: save_attn_procs NoneType.replace vs PNDMScheduler set_format missing." + }, + { + "left": "issue:4925", + "right": "issue:6697", + "accept": false, + "reason": "Both are OOM reports, but the training setups and symptoms are not the same concrete bug." + }, + { + "left": "issue:11005", + "right": "issue:4308", + "accept": false, + "reason": "Docstring/default mismatch in different pipeline code; not a duplicate bug report." + }, + { + "left": "issue:246", + "right": "issue:858", + "accept": false, + "reason": "Different missing-attribute problems in transformers vs a Parameter.named_children error." + }, + { + "left": "issue:2571", + "right": "issue:6281", + "accept": false, + "reason": "Different import errors for different classes; no shared underlying issue." + }, + { + "left": "issue:1246", + "right": "issue:6098", + "accept": true, + "reason": "Same FP16 gradient unscale error; #6098 is essentially the same bug wording as the older Dreambooth report." + }, + { + "left": "issue:4324", + "right": "issue:4649", + "accept": true, + "reason": "Same underlying NoneType/added_cond_kwargs problem in UNet forward; the error wording and code path align closely." + }, + { + "left": "issue:7146", + "right": "issue:8613", + "accept": false, + "reason": "Both are memory-related, but one is SDXL fine-tuning OOM and the other is caption_column-triggered OOM." + }, + { + "left": "issue:1709", + "right": "issue:8307", + "accept": false, + "reason": "Completely different errors: ONNX Tensor.astype vs missing StableCascadeUNet attribute." + }, + { + "left": "issue:3382", + "right": "issue:3565", + "accept": false, + "reason": "Both mention CUDA OOM in text-to-image training, but they are not the same concrete failure path." + }, + { + "left": "issue:4736", + "right": "issue:7114", + "accept": false, + "reason": "Related SDXL OOM theme, but one also has invalid cast issues and they are separate reports." + }, + { + "left": "issue:10839", + "right": "issue:9548", + "accept": false, + "reason": "Both are Flux Dreambooth LoRA complaints, but the titles indicate different symptoms and no clear identical bug." + }, + { + "left": "issue:11045", + "right": "issue:9227", + "accept": false, + "reason": "Different Flux Dreambooth LoRA reports with different models/symptoms; not enough to treat as the same bug." + }, + { + "left": "issue:4868", + "right": "issue:9901", + "accept": false, + "reason": "Different attribute/type errors with no shared code-path evidence." + }, + { + "left": "issue:5016", + "right": "issue:9393", + "accept": false, + "reason": "Different training problems: text_encoder training failure vs model save failure in deepspeed Flux training." + }, + { + "left": "issue:4137", + "right": "issue:7432", + "accept": false, + "reason": "Both compare diffusers to WebUI, but they are broad result-difference complaints, not the same bug." + }, + { + "left": "issue:7516", + "right": "issue:858", + "accept": false, + "reason": "Different missing attribute errors in different subsystems." + }, + { + "left": "issue:1976", + "right": "issue:5259", + "accept": false, + "reason": "Different file/path resolution problems: tokenizer/config.json vs model_index.json." + }, + { + "left": "issue:4492", + "right": "issue:6142", + "accept": false, + "reason": "Different NameError causes; one is lowercase false, the other is undefined start_training." + }, + { + "left": "issue:12492", + "right": "issue:7681", + "accept": false, + "reason": "Different failures: missing CLIPFeatureExtractor vs import failure due to an OpenSSL/lib issue." + }, + { + "left": "issue:11581", + "right": "issue:12436", + "accept": false, + "reason": "Both are constructor-argument incompatibilities, but for different classes and different kwargs." + }, + { + "left": "issue:1988", + "right": "issue:5805", + "accept": false, + "reason": "Same broad dtype mismatch theme, but not the same model/training path or concrete bug." + }, + { + "left": "issue:2044", + "right": "issue:6044", + "accept": false, + "reason": "Different attribute errors in different parts of the pipeline/runtime." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and the soft-edge judgments are conservative. The two accepted pairs are plausible duplicates from the titles, while the rejected pairs are appropriately cautious rather than overstated. Canonical choice #1246 is also well supported by age/activity and the exact FP16 wording match with #6098." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 26, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10062", + "issue:10154", + "issue:10500", + "issue:10591", + "issue:1249", + "issue:1255", + "issue:1260", + "issue:13087", + "issue:1377", + "issue:1544", + "issue:1709", + "issue:1925", + "issue:234", + "issue:2556", + "issue:2947", + "issue:3026", + "issue:3140", + "issue:3202", + "issue:3489", + "issue:4086", + "issue:4104", + "issue:4224", + "issue:4265", + "issue:4282", + "issue:4724", + "issue:5518", + "issue:6028", + "issue:6324", + "issue:6371", + "issue:646", + "issue:7114", + "issue:7276", + "issue:7483", + "issue:7516", + "issue:7574", + "issue:7681", + "issue:7728", + "issue:8320", + "issue:8537", + "issue:8759", + "issue:8887", + "issue:8975", + "issue:9111", + "issue:9450", + "issue:9505", + "issue:9527", + "issue:9630" + ], + "soft_pairs": [ + "issue:3026|issue:7728", + "issue:4104|issue:8759", + "issue:10062|issue:1544", + "issue:646|issue:8320", + "issue:4086|issue:4724", + "issue:3489|issue:6324", + "issue:1925|issue:234", + "issue:1260|issue:8975", + "issue:10500|issue:9450", + "issue:2556|issue:4282", + "issue:4224|issue:5518", + "issue:9527|issue:9630", + "issue:1255|issue:4265", + "issue:1377|issue:3140", + "issue:3202|issue:7516", + "issue:10591|issue:8887", + "issue:13087|issue:4265", + "issue:2947|issue:3489", + "issue:10154|issue:1709", + "issue:6371|issue:7276", + "issue:7681|issue:8537", + "issue:7574|issue:9505", + "issue:1249|issue:9111", + "issue:4724|issue:7483", + "issue:6028|issue:7114" + ], + "prepared_review_unit_hash": "b36d41ed75d9e31e7385b0c50d20c313cf8f5c1b4395c34826aff3f12a5e64e1", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 34464, + "estimated_input_tokens": 8616, + "estimated_eval_tokens": 17488 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of diffusers bug reports: many missing-attribute/import failures, plus a few training/runtime regression reports. Only a few pairs are true duplicates; most are merely similar symptoms in different code paths.", + "confidence": 0.78, + "canonical_issue_reason": "issue:10062 is the most representative open bug: a concrete core-model attribute error tied to LoRA/PEFT, which matches the cluster\u2019s recurring missing-attribute theme better than the more niche or version-specific reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10062 is the best single issue to keep as the cluster representative because it is specific enough to be actionable yet broad enough to stand in for the repeated attribute-missing failures in this set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3026", + "right": "issue:7728", + "accept": false, + "reason": "Both are training-related errors, but they affect different examples and failure modes; not the same underlying bug." + }, + { + "left": "issue:4104", + "right": "issue:8759", + "accept": true, + "reason": "Same SD3 image-postprocessing symptom: black/invalid cast output and the same runtime warning in the image processor code path." + }, + { + "left": "issue:10062", + "right": "issue:1544", + "accept": false, + "reason": "Both are AttributeErrors, but on different objects and methods; no clear shared code-path bug." + }, + { + "left": "issue:646", + "right": "issue:8320", + "accept": false, + "reason": "These are unrelated attribute errors on different components; same broad pattern only." + }, + { + "left": "issue:4086", + "right": "issue:4724", + "accept": false, + "reason": "Both mention SDXL/control-related problems, but they are different pipelines and different failure descriptions." + }, + { + "left": "issue:3489", + "right": "issue:6324", + "accept": false, + "reason": "Different errors in different components; only generic runtime failure similarity." + }, + { + "left": "issue:1925", + "right": "issue:234", + "accept": false, + "reason": "Both mention half precision, but they are different reports and not clearly the same concrete bug." + }, + { + "left": "issue:1260", + "right": "issue:8975", + "accept": false, + "reason": "Different import failures for different symbols; same general import-error class only." + }, + { + "left": "issue:10500", + "right": "issue:9450", + "accept": false, + "reason": "Both are device-mismatch errors, but they occur in different pipelines and setups." + }, + { + "left": "issue:2556", + "right": "issue:4282", + "accept": true, + "reason": "Identical feature request title and same underlying change request: adding Multi-ControlNet support." + }, + { + "left": "issue:4224", + "right": "issue:5518", + "accept": false, + "reason": "Both are attribute errors involving models, but they concern different subsystems and different missing members." + }, + { + "left": "issue:9527", + "right": "issue:9630", + "accept": false, + "reason": "Both are Flux/controlnet training issues, but one is a dtype mismatch and the other is a single-layer/GPU configuration problem." + }, + { + "left": "issue:1255", + "right": "issue:4265", + "accept": false, + "reason": "Both are import errors, but for different pipeline classes and different missing names." + }, + { + "left": "issue:1377", + "right": "issue:3140", + "accept": false, + "reason": "Both are ImportErrors, but one is from transformers and the other is from diffusers; different missing symbols." + }, + { + "left": "issue:3202", + "right": "issue:7516", + "accept": false, + "reason": "Different attribute/module errors in different code paths; not the same bug." + }, + { + "left": "issue:10591", + "right": "issue:8887", + "accept": false, + "reason": "Both concern SD3 training scripts, but the titles point to different concrete problems; too little evidence they are the same bug." + }, + { + "left": "issue:13087", + "right": "issue:4265", + "accept": false, + "reason": "Different missing pipeline names and different models; same import-error theme only." + }, + { + "left": "issue:2947", + "right": "issue:3489", + "accept": false, + "reason": "Different missing-attribute failures in different components (UNet vs ONNX runtime model)." + }, + { + "left": "issue:10154", + "right": "issue:1709", + "accept": false, + "reason": "Both mention attribute-style errors during loading/conversion, but the objects and failing methods differ." + }, + { + "left": "issue:6371", + "right": "issue:7276", + "accept": false, + "reason": "An assertion failure in a ControlNet-XS example is not the same as a DreamBooth network-fetch issue." + }, + { + "left": "issue:7681", + "right": "issue:8537", + "accept": false, + "reason": "These are environment/import failures with different root causes and different missing capabilities." + }, + { + "left": "issue:7574", + "right": "issue:9505", + "accept": true, + "reason": "Same KeyError on 'middle_block_out.0.weight' with the same upscaling-related failure mode." + }, + { + "left": "issue:1249", + "right": "issue:9111", + "accept": false, + "reason": "Different failures on different objects; one is a NumPy device attribute error, the other a missing scheduler attribute." + }, + { + "left": "issue:4724", + "right": "issue:7483", + "accept": false, + "reason": "Both relate to SDXL, but one is an img2img pipeline behavior issue and the other is a LoRA conversion script problem." + }, + { + "left": "issue:6028", + "right": "issue:7114", + "accept": false, + "reason": "Both are OOM reports, but on different models/scripts and not the same concrete issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays appropriately conservative. The representative issue choice is defensible given the cluster\u2019s broad mix of missing-attribute/import/runtime reports, and the accepted soft edges are well supported by the titles (notably the duplicate Multi-ControlNet requests, the shared SD3 image_processor warning/black-output pair, and the identical middle_block_out.0.weight KeyError pair). The rejected pairs are also phrased cautiously without overstating equivalence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 27, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10472", + "issue:10594", + "issue:11199", + "issue:11436", + "issue:11581", + "issue:12459", + "issue:1263", + "issue:1264", + "issue:12689", + "issue:12826", + "issue:1315", + "issue:13300", + "issue:1377", + "issue:1544", + "issue:1875", + "issue:1925", + "issue:1988", + "issue:3211", + "issue:3489", + "issue:3648", + "issue:4118", + "issue:4409", + "issue:4557", + "issue:4724", + "issue:5246", + "issue:5603", + "issue:5757", + "issue:5840", + "issue:6044", + "issue:6252", + "issue:6298", + "issue:6415", + "issue:6517", + "issue:7228", + "issue:7311", + "issue:7708", + "issue:7766", + "issue:851", + "issue:8537", + "issue:858", + "issue:8659", + "issue:8863", + "issue:9111", + "issue:9630", + "issue:9780", + "issue:9911", + "issue:993" + ], + "soft_pairs": [ + "issue:7228|issue:858", + "issue:7311|issue:8659", + "issue:6415|issue:7766", + "issue:3211|issue:4409", + "issue:5757|issue:5840", + "issue:1263|issue:1315", + "issue:4118|issue:8537", + "issue:3489|issue:4557", + "issue:10594|issue:12459", + "issue:12826|issue:5603", + "issue:1875|issue:1988", + "issue:13300|issue:993", + "issue:1264|issue:3648", + "issue:6298|issue:9780", + "issue:12689|issue:6252", + "issue:5246|issue:8537", + "issue:1544|issue:9111", + "issue:11199|issue:6517", + "issue:9630|issue:9911", + "issue:1925|issue:4724", + "issue:1315|issue:1377", + "issue:10472|issue:11581", + "issue:11436|issue:851", + "issue:4724|issue:7708", + "issue:6044|issue:8863" + ], + "prepared_review_unit_hash": "18e437a9e1d73e80ccbe37cc0c48d6b38b49cadebe656b60ee4ed49ef4213a22", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 34548, + "estimated_input_tokens": 8637, + "estimated_eval_tokens": 17530 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly heterogeneous, but there are two clear duplicate pairs: the LoRA checkpoint-loading reports and the DreamBooth LoRA SDXL checkpoint-resume reports. The rest only share broad themes like import errors, dtype mismatches, or checkpoint handling and should stay separate.", + "confidence": 0.74, + "canonical_issue_reason": "issue:12689 is the clearest representative for the LoRA checkpoint-loading bug: it is concise, specific, and has an exact duplicate in issue:6252.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:12689 is the strongest global issue candidate because it cleanly captures a concrete, recurring bug with an exact duplicate and a clear symptom statement.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7228", + "right": "issue:858", + "accept": false, + "reason": "Both are generic attribute errors, but they affect different objects and code paths; not the same underlying bug." + }, + { + "left": "issue:7311", + "right": "issue:8659", + "accept": false, + "reason": "Both mention checkpoint/resume, but one is about saving at checkpoint and the other about resuming/loading for inference; too broad to merge as the same bug without stronger evidence." + }, + { + "left": "issue:6415", + "right": "issue:7766", + "accept": false, + "reason": "Both are IP-Adapter FaceID-related, but one is a support/availability complaint and the other is a usage question; not duplicate bug reports." + }, + { + "left": "issue:3211", + "right": "issue:4409", + "accept": false, + "reason": "Different missing attributes on different pipeline/model objects; same symptom class, but not the same concrete defect." + }, + { + "left": "issue:5757", + "right": "issue:5840", + "accept": true, + "reason": "Same script, same resume-from-checkpoint problem, and near-identical titles; these look like the same underlying bug." + }, + { + "left": "issue:1263", + "right": "issue:1315", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols from different modules; likely separate compatibility issues." + }, + { + "left": "issue:4118", + "right": "issue:8537", + "accept": false, + "reason": "Both are from single-file/model-loading workflows, but the missing attributes differ completely; not the same bug." + }, + { + "left": "issue:3489", + "right": "issue:4557", + "accept": false, + "reason": "Different failure modes in ONNX/pipeline code; the shared shape is only a generic AttributeError." + }, + { + "left": "issue:10594", + "right": "issue:12459", + "accept": false, + "reason": "One is a multiple-controls-with-LoRA issue, the other is batch inference throughput; not the same bug." + }, + { + "left": "issue:12826", + "right": "issue:5603", + "accept": false, + "reason": "Both are import-name errors, but they refer to different exports and likely different version skew cases." + }, + { + "left": "issue:1875", + "right": "issue:1988", + "accept": false, + "reason": "Both are fp16/bf16 dtype problems, but they occur in different training examples and error contexts." + }, + { + "left": "issue:13300", + "right": "issue:993", + "accept": false, + "reason": "Both involve precision/device mismatch, but one is CPU generation on modern models and the other is an fp16 stable-diffusion mismatch; not the same bug." + }, + { + "left": "issue:1264", + "right": "issue:3648", + "accept": false, + "reason": "No real overlap beyond stable-diffusion training references; one is about a hard-coded scheduler checkpoint, the other about VRAM requirements." + }, + { + "left": "issue:6298", + "right": "issue:9780", + "accept": false, + "reason": "Different missing attributes in different pipelines/models; not the same concrete code path." + }, + { + "left": "issue:12689", + "right": "issue:6252", + "accept": true, + "reason": "These are essentially the same bug: LoRA weights/adapters not being loaded from checkpoints, with near-identical titles." + }, + { + "left": "issue:5246", + "right": "issue:8537", + "accept": false, + "reason": "Different missing symbol vs torch-version error; both import/model-compatibility issues but not the same defect." + }, + { + "left": "issue:1544", + "right": "issue:9111", + "accept": false, + "reason": "Different missing attributes in unrelated model classes; not a duplicate." + }, + { + "left": "issue:11199", + "right": "issue:6517", + "accept": false, + "reason": "Both are import failures, but one is about onnx_utils specifically and the other about a ControlNet SDXL import traceback; too different." + }, + { + "left": "issue:9630", + "right": "issue:9911", + "accept": false, + "reason": "Both are FLUX multi-GPU/controlnet-related, but one is a num_single_layers setting error and the other is a two-ControlNet layer-length mismatch; separate bugs." + }, + { + "left": "issue:1925", + "right": "issue:4724", + "accept": false, + "reason": "Different SDXL/inpainting pipeline issues; the shared theme is only that a pipeline behaves incorrectly." + }, + { + "left": "issue:1315", + "right": "issue:1377", + "accept": false, + "reason": "Both are import errors, but they concern different missing names from different packages and are not the same fix." + }, + { + "left": "issue:10472", + "right": "issue:11581", + "accept": false, + "reason": "Different adapter/component errors: one is LoRA bias arg handling, the other is IP Adapter dtype handling; not the same bug." + }, + { + "left": "issue:11436", + "right": "issue:851", + "accept": false, + "reason": "Both are dtype mismatches, but the models, symptoms, and likely fixes differ." + }, + { + "left": "issue:4724", + "right": "issue:7708", + "accept": false, + "reason": "Both concern SDXL pipeline setup, but one is about expected behavior and the other about a missing tokenizer file; not the same issue." + }, + { + "left": "issue:6044", + "right": "issue:8863", + "accept": false, + "reason": "Different missing-attribute failures in different pipelines; no evidence they share a single root cause." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The two accepted soft pairs are well-supported by near-identical titles and the same checkpoint/resume or LoRA-loading failure mode. The rejected pairs are also phrased cautiously, avoiding overclaiming duplicates from only broad symptom similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 28, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10062", + "issue:1100", + "issue:11447", + "issue:11978", + "issue:1249", + "issue:1255", + "issue:1572", + "issue:1709", + "issue:1717", + "issue:1842", + "issue:2044", + "issue:2100", + "issue:2234", + "issue:2984", + "issue:3094", + "issue:3211", + "issue:3382", + "issue:3728", + "issue:4194", + "issue:4224", + "issue:4316", + "issue:4331", + "issue:4409", + "issue:4667", + "issue:4969", + "issue:499", + "issue:5016", + "issue:5509", + "issue:5565", + "issue:5619", + "issue:5760", + "issue:5992", + "issue:6028", + "issue:6044", + "issue:6154", + "issue:6362", + "issue:6382", + "issue:6536", + "issue:7114", + "issue:7214", + "issue:7228", + "issue:7660", + "issue:8307", + "issue:8939", + "issue:9026", + "issue:9115", + "issue:9534", + "issue:9973" + ], + "soft_pairs": [ + "issue:1842|issue:9026", + "issue:2984|issue:5760", + "issue:10062|issue:1100", + "issue:5509|issue:7214", + "issue:6362|issue:6536", + "issue:2044|issue:8307", + "issue:3211|issue:4224", + "issue:1249|issue:9534", + "issue:1709|issue:9534", + "issue:1255|issue:2100", + "issue:1249|issue:7228", + "issue:3382|issue:6028", + "issue:11978|issue:3728", + "issue:1572|issue:5619", + "issue:1255|issue:4969", + "issue:2234|issue:8939", + "issue:4409|issue:5992", + "issue:499|issue:7660", + "issue:5016|issue:6154", + "issue:1249|issue:6044", + "issue:3094|issue:7114", + "issue:4316|issue:4331", + "issue:11447|issue:1717", + "issue:5565|issue:9973", + "issue:6382|issue:9115", + "issue:4194|issue:4667" + ], + "prepared_review_unit_hash": "3733e3f243bce438bed9dc5e7a31adc697f994c9840ae64ce19a86038711f6a9", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35383, + "estimated_input_tokens": 8846, + "estimated_eval_tokens": 17948 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is over-broad: it mixes many unrelated diffusers issues (missing attrs, import errors, OOMs, example regressions, and new-pipeline requests). I only see two plausible duplicate pairs: the HF_HUB_OFFLINE loading failures and the StableDiffusionPipeline.from_single_file regression.", + "confidence": 0.9, + "canonical_issue_reason": "No single canonical issue cleanly represents the whole cluster; the strongest duplicate-like subgroup is the offline-cache bug, where issue 1717 is the broader, more canonical wording compared with 11447.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1717 is the best issue representative because it states the broad offline-cache loading failure and also matches 11447\u2019s shard-loading symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1842", + "right": "issue:9026", + "accept": false, + "reason": "Both are TypeErrors, but they come from different code paths and different kwargs; not the same underlying bug." + }, + { + "left": "issue:2984", + "right": "issue:5760", + "accept": false, + "reason": "Completely unrelated research/model announcement issues, not duplicates." + }, + { + "left": "issue:10062", + "right": "issue:1100", + "accept": false, + "reason": "Both mention missing attributes, but the models, attributes, and failure contexts differ." + }, + { + "left": "issue:5509", + "right": "issue:7214", + "accept": false, + "reason": "Different failures: image batch-size validation vs scheduler karras_sigma behavior." + }, + { + "left": "issue:6362", + "right": "issue:6536", + "accept": false, + "reason": "Both are OOM reports, but for different models and training paths; too broad to merge." + }, + { + "left": "issue:2044", + "right": "issue:8307", + "accept": false, + "reason": "Missing dtype on DDP vs missing StableCascadeUNet export are unrelated attribute errors." + }, + { + "left": "issue:3211", + "right": "issue:4224", + "accept": false, + "reason": "Both are missing pipeline attrs, but they affect different attributes and different pipeline features." + }, + { + "left": "issue:1249", + "right": "issue:9534", + "accept": false, + "reason": "Different pipeline/model attribute errors with no clear shared code path." + }, + { + "left": "issue:1709", + "right": "issue:9534", + "accept": false, + "reason": "ONNX Tensor astype error and FluxMultiControlNetModel config error are unrelated." + }, + { + "left": "issue:1255", + "right": "issue:2100", + "accept": false, + "reason": "Both are import errors, but for different symbols and likely different export gaps." + }, + { + "left": "issue:1249", + "right": "issue:7228", + "accept": false, + "reason": "Different missing-attribute failures in different loading/tracing paths." + }, + { + "left": "issue:3382", + "right": "issue:6028", + "accept": false, + "reason": "Both involve OOM, but in different scripts and models; not the same bug." + }, + { + "left": "issue:11978", + "right": "issue:3728", + "accept": false, + "reason": "A pipeline feature request and a different new-pipeline request are not duplicates." + }, + { + "left": "issue:1572", + "right": "issue:5619", + "accept": false, + "reason": "Both concern img2img-like behavior, but they target different pipelines and symptoms." + }, + { + "left": "issue:1255", + "right": "issue:4969", + "accept": false, + "reason": "Separate import failures for different classes; no shared concrete bug." + }, + { + "left": "issue:2234", + "right": "issue:8939", + "accept": false, + "reason": "Training/xformers issue vs tokenizer loading issue for a different model release." + }, + { + "left": "issue:4409", + "right": "issue:5992", + "accept": false, + "reason": "Different attribute/error classes in unrelated code paths." + }, + { + "left": "issue:499", + "right": "issue:7660", + "accept": false, + "reason": "Both mention examples, but they are different examples with different failures." + }, + { + "left": "issue:5016", + "right": "issue:6154", + "accept": false, + "reason": "Both are SDXL LoRA training reports, but one is about text encoder training and the other about M2 Mac compatibility." + }, + { + "left": "issue:1249", + "right": "issue:6044", + "accept": false, + "reason": "Different missing-attribute errors in different pipeline behaviors." + }, + { + "left": "issue:3094", + "right": "issue:7114", + "accept": false, + "reason": "Both are training OOMs, but in different scripts and setups; too broad to be the same bug." + }, + { + "left": "issue:4316", + "right": "issue:4331", + "accept": false, + "reason": "Validation-option error and differing DreamBooth outputs are not the same underlying issue." + }, + { + "left": "issue:11447", + "right": "issue:1717", + "accept": true, + "reason": "Both describe the same HF_HUB_OFFLINE loading failure: cached/sharded checkpoint files are not loaded correctly offline." + }, + { + "left": "issue:5565", + "right": "issue:9973", + "accept": false, + "reason": "Different DreamBooth LoRA training failures with no shared concrete code path." + }, + { + "left": "issue:6382", + "right": "issue:9115", + "accept": false, + "reason": "Different adapter/model combination errors; not the same bug." + }, + { + "left": "issue:4194", + "right": "issue:4667", + "accept": true, + "reason": "Both point to the same from_single_file regression in StableDiffusionPipeline; 4194 is a specific VAE symptom of the broader API break." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative about the cluster being over-broad. The two accepted soft pairs, 11447\u21941717 and 4194\u21944667, are plausibly the same bugfix class; the rest are appropriately rejected as too broad or clearly unrelated." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 29, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10029", + "issue:10169", + "issue:12174", + "issue:1263", + "issue:1444", + "issue:1565", + "issue:1709", + "issue:1866", + "issue:2044", + "issue:246", + "issue:3290", + "issue:3489", + "issue:3886", + "issue:4086", + "issue:4137", + "issue:4255", + "issue:4324", + "issue:4483", + "issue:4615", + "issue:4648", + "issue:4868", + "issue:490", + "issue:5374", + "issue:5565", + "issue:5919", + "issue:6218", + "issue:6324", + "issue:646", + "issue:6536", + "issue:6570", + "issue:6842", + "issue:6967", + "issue:6996", + "issue:7210", + "issue:7708", + "issue:7784", + "issue:8307", + "issue:8320", + "issue:852", + "issue:8552", + "issue:867", + "issue:9044", + "issue:9111", + "issue:9534", + "issue:9799", + "issue:9889", + "issue:9936", + "issue:9962" + ], + "soft_pairs": [ + "issue:5374|issue:852", + "issue:8552|issue:9889", + "issue:12174|issue:7210", + "issue:5374|issue:9962", + "issue:7784|issue:8320", + "issue:4324|issue:6996", + "issue:4137|issue:4255", + "issue:6570|issue:9799", + "issue:1263|issue:4615", + "issue:1444|issue:6842", + "issue:12174|issue:6324", + "issue:490|issue:867", + "issue:3489|issue:4868", + "issue:3290|issue:8307", + "issue:5565|issue:6967", + "issue:5919|issue:646", + "issue:1709|issue:9111", + "issue:10169|issue:4648", + "issue:1709|issue:246", + "issue:4483|issue:9044", + "issue:4086|issue:4483", + "issue:2044|issue:9534", + "issue:1565|issue:6536", + "issue:3886|issue:9936", + "issue:4483|issue:7708", + "issue:10029|issue:12174", + "issue:1866|issue:6218" + ], + "prepared_review_unit_hash": "7ff83e33041923527703caf93b96ae2e1b5e100dadf3c46504b58f4e607b5f04", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35394, + "estimated_input_tokens": 8849, + "estimated_eval_tokens": 17954 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated attribute/import/runtime errors that only match at a very broad textual level. The only clear duplicate is the missing `config.json` report between issues 10169 and 4648.", + "confidence": 0.9, + "canonical_issue_reason": "issue:4648 is the most concrete version of the `config.json`-missing problem, with the explicit failing path and error text; issue:10169 is the same bug but stated more generically.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4648 is the best representative of the true duplicate pair because it names the exact missing file and directory, making the underlying issue clearest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5374", + "right": "issue:852", + "accept": false, + "reason": "Both mention unexpected keyword arguments, but they refer to different arguments and likely different call sites." + }, + { + "left": "issue:8552", + "right": "issue:9889", + "accept": false, + "reason": "Both are segfault reports, but they involve different pipelines and different runtime setups; not the same concrete bug." + }, + { + "left": "issue:12174", + "right": "issue:7210", + "accept": false, + "reason": "Both are AttributeErrors, but on different missing methods in different code paths." + }, + { + "left": "issue:5374", + "right": "issue:9962", + "accept": false, + "reason": "Different unexpected keyword arguments; same error shape but not the same underlying defect." + }, + { + "left": "issue:7784", + "right": "issue:8320", + "accept": false, + "reason": "Different missing attributes in unrelated code paths." + }, + { + "left": "issue:4324", + "right": "issue:6996", + "accept": false, + "reason": "Both are NoneType-related errors, but the failing expressions and contexts are different." + }, + { + "left": "issue:4137", + "right": "issue:4255", + "accept": false, + "reason": "Both complain about output quality versus another implementation, but they are not the same bug or fix." + }, + { + "left": "issue:6570", + "right": "issue:9799", + "accept": false, + "reason": "Both are tensor size mismatches, but the shapes and pipelines differ; no clear shared root cause." + }, + { + "left": "issue:1263", + "right": "issue:4615", + "accept": false, + "reason": "Different missing import symbols; likely separate compatibility issues." + }, + { + "left": "issue:1444", + "right": "issue:6842", + "accept": false, + "reason": "Different missing imports for different pipeline classes." + }, + { + "left": "issue:12174", + "right": "issue:6324", + "accept": false, + "reason": "Different AttributeErrors with different objects and failure modes." + }, + { + "left": "issue:490", + "right": "issue:867", + "accept": false, + "reason": "Both concern memory, but one is general GPU memory use and the other is memory not being released in FastAPI; not the same bug." + }, + { + "left": "issue:3489", + "right": "issue:4868", + "accept": false, + "reason": "Different missing attributes on different objects and in different pipelines." + }, + { + "left": "issue:3290", + "right": "issue:8307", + "accept": false, + "reason": "A file flush NoneType error is unrelated to a missing exported class attribute." + }, + { + "left": "issue:5565", + "right": "issue:6967", + "accept": false, + "reason": "Both are dreambooth/LoRA script issues, but one is script startup and the other is a prior-preservation shape mismatch." + }, + { + "left": "issue:5919", + "right": "issue:646", + "accept": false, + "reason": "Both are missing-attribute errors, but on unrelated classes and code paths." + }, + { + "left": "issue:1709", + "right": "issue:9111", + "accept": false, + "reason": "ONNX tensor dtype conversion error versus a missing scheduler attribute; unrelated." + }, + { + "left": "issue:10169", + "right": "issue:4648", + "accept": true, + "reason": "Same underlying missing `config.json` problem; 4648 is the concrete report and 10169 is the generic duplicate." + }, + { + "left": "issue:1709", + "right": "issue:246", + "accept": false, + "reason": "Different errors in different components: ONNX tensor conversion versus missing transformer mixin import." + }, + { + "left": "issue:4483", + "right": "issue:9044", + "accept": false, + "reason": "Different SDXL-related failures: nondeterministic generation versus missing model file." + }, + { + "left": "issue:4086", + "right": "issue:4483", + "accept": false, + "reason": "Both are SDXL training/inference complaints, but one is controlnet training not working and the other is generator nondeterminism." + }, + { + "left": "issue:2044", + "right": "issue:9534", + "accept": false, + "reason": "Different missing attributes on different wrapper/model classes." + }, + { + "left": "issue:1565", + "right": "issue:6536", + "accept": false, + "reason": "Both mention OOM, but one is GPU selection behavior and the other is UNet3D training memory exhaustion." + }, + { + "left": "issue:3886", + "right": "issue:9936", + "accept": false, + "reason": "Both involve distributed training timeouts, but the scripts, models, and likely root causes differ." + }, + { + "left": "issue:4483", + "right": "issue:7708", + "accept": false, + "reason": "Nondeterminism versus missing tokenizer config file; unrelated." + }, + { + "left": "issue:10029", + "right": "issue:12174", + "accept": false, + "reason": "Both are attribute errors, but they concern different missing attributes and are not clearly the same bug." + }, + { + "left": "issue:1866", + "right": "issue:6218", + "accept": false, + "reason": "Both are pipeline behavior issues, but one is scheduler strength behavior and the other is batched seed handling." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: issue:10169 and issue:4648 both describe the same missing `config.json` problem, and the other soft-edge rejections stay conservative by relying on clearly different error contexts. No overstatement of evidence is apparent." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 30, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10436", + "issue:10612", + "issue:10933", + "issue:10989", + "issue:11062", + "issue:11178", + "issue:11209", + "issue:11224", + "issue:1260", + "issue:1263", + "issue:13429", + "issue:13430", + "issue:1495", + "issue:1925", + "issue:1962", + "issue:2390", + "issue:2437", + "issue:2635", + "issue:2676", + "issue:2830", + "issue:2838", + "issue:2904", + "issue:2984", + "issue:3432", + "issue:3453", + "issue:3728", + "issue:3788", + "issue:4094", + "issue:4241", + "issue:4728", + "issue:4779", + "issue:4945", + "issue:5455", + "issue:5630", + "issue:5689", + "issue:6149", + "issue:6636", + "issue:7114", + "issue:7228", + "issue:7513", + "issue:7593", + "issue:858", + "issue:9161", + "issue:9163", + "issue:9208", + "issue:9497", + "issue:9500", + "issue:9501" + ], + "soft_pairs": [ + "issue:10436|issue:1925", + "issue:10933|issue:6149", + "issue:1260|issue:6636", + "issue:1263|issue:1495", + "issue:2904|issue:7228", + "issue:4094|issue:4945", + "issue:4728|issue:7114", + "issue:7593|issue:858", + "issue:11209|issue:11224", + "issue:13429|issue:13430", + "issue:11062|issue:11209", + "issue:11062|issue:11224", + "issue:9500|issue:9501", + "issue:5455|issue:7513", + "issue:3728|issue:3788", + "issue:9161|issue:9163", + "issue:2830|issue:4779", + "issue:1962|issue:2984", + "issue:10612|issue:10989", + "issue:5630|issue:5689", + "issue:2838|issue:4241", + "issue:2390|issue:2437", + "issue:9497|issue:9500", + "issue:2635|issue:2676", + "issue:11178|issue:9208", + "issue:3432|issue:3453" + ], + "prepared_review_unit_hash": "df5c3c4fb92ea3d48294abc11942a661c60b835b7b2bc73b89ba09ab59c520ab", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 34391, + "estimated_input_tokens": 8598, + "estimated_eval_tokens": 17452 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of model-loading/support issues, with a small number of clear duplicates inside narrower subtopics (LoRA loading, scheduler-name questions, T2I-Adapter support, and ControlNet from_single_file loading).", + "confidence": 0.86, + "canonical_issue_reason": "issue:11209 is the best canonical issue for the LoRA-loading subgroup: it is generic, open, and has a near-identical twin in issue:11224.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11209 is the strongest global issue representative among these items, but only for the LoRA-loading slice of the cluster; the overall set is too mixed to support one broad canonical issue perfectly.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10436", + "right": "issue:1925", + "accept": false, + "reason": "Both mention inpainting, but one is a model-loading failure and the other is a half-precision runtime issue on a different model/version." + }, + { + "left": "issue:10933", + "right": "issue:6149", + "accept": false, + "reason": "Both are IP-Adapter load errors, but they target different adapter/model variants and do not look like the same concrete bug." + }, + { + "left": "issue:1260", + "right": "issue:6636", + "accept": false, + "reason": "Different import symbols and different API breakages; same general category, not the same bug." + }, + { + "left": "issue:1263", + "right": "issue:1495", + "accept": false, + "reason": "Both are import errors, but for unrelated symbols and modules." + }, + { + "left": "issue:2904", + "right": "issue:7228", + "accept": false, + "reason": "Unrelated failures: AudioLDM scheduler initialization vs traced StableDiffusionPipeline missing dtype." + }, + { + "left": "issue:4094", + "right": "issue:4945", + "accept": false, + "reason": "One asks for a training script, the other asks for tests for training; related area but not the same change." + }, + { + "left": "issue:4728", + "right": "issue:7114", + "accept": false, + "reason": "Both involve OOM during training, but on different scripts and contexts, so not the same underlying issue." + }, + { + "left": "issue:7593", + "right": "issue:858", + "accept": false, + "reason": "Different attribute errors in different code paths; only superficial 'bug' similarity." + }, + { + "left": "issue:11209", + "right": "issue:11224", + "accept": true, + "reason": "Near-identical titles and very high textual overlap; both are generic 'Error in loading Lora' reports." + }, + { + "left": "issue:13429", + "right": "issue:13430", + "accept": true, + "reason": "Same JoyAI-Image-Edit support request, phrased as 'add' vs 'support'." + }, + { + "left": "issue:11062", + "right": "issue:11209", + "accept": false, + "reason": "Both are LoRA loading problems, but 11062 is a specific CivitAI LoRA load error and not clearly the same bug as the generic loading failure." + }, + { + "left": "issue:11062", + "right": "issue:11224", + "accept": false, + "reason": "Same reason as 11062 vs 11209: similar area, but not enough evidence of the same concrete failure." + }, + { + "left": "issue:9500", + "right": "issue:9501", + "accept": false, + "reason": "Both concern Flux DreamBooth saving, but one is a checkpoint failure and the other is a long save delay; likely distinct symptoms." + }, + { + "left": "issue:5455", + "right": "issue:7513", + "accept": false, + "reason": "Both mention DreamBooth/LoRA training problems, but the titles point to different workflows and failure modes." + }, + { + "left": "issue:3728", + "right": "issue:3788", + "accept": false, + "reason": "Separate new-pipeline requests for different models; same broad theme, not the same change." + }, + { + "left": "issue:9161", + "right": "issue:9163", + "accept": false, + "reason": "Same script area, but one is distributed bugs and the other is tensor bugs; too vague to merge as one issue." + }, + { + "left": "issue:2830", + "right": "issue:4779", + "accept": false, + "reason": "Both are about video generation on ModelScope, but one is a question and the other a pipeline request; not clearly the same request." + }, + { + "left": "issue:1962", + "right": "issue:2984", + "accept": false, + "reason": "Different video-generation topics and different model families." + }, + { + "left": "issue:10612", + "right": "issue:10989", + "accept": false, + "reason": "Both involve LoRA loading with quantization, but one is Flux Fill and the other is Flux ControlNet; related bug class, not the same concrete code path." + }, + { + "left": "issue:5630", + "right": "issue:5689", + "accept": false, + "reason": "Both report residual noise in SDXL, but they refer to different schedulers and scenarios." + }, + { + "left": "issue:2838", + "right": "issue:4241", + "accept": false, + "reason": "Both are about control/T2I-adapter style support, but one is a ControlNet pipeline question and the other is fine-tuning unidiffusers with LoRA." + }, + { + "left": "issue:2390", + "right": "issue:2437", + "accept": true, + "reason": "Both are the same T2I-Adapter feature request, one phrased as support and the other as adding the pipeline." + }, + { + "left": "issue:9497", + "right": "issue:9500", + "accept": false, + "reason": "Both are Flux DreamBooth training issues, but one is a matrix-shape runtime error and the other is checkpoint saving failure." + }, + { + "left": "issue:2635", + "right": "issue:2676", + "accept": true, + "reason": "Same underlying scheduler-name mapping question for DPM++ Karras variants." + }, + { + "left": "issue:11178", + "right": "issue:9208", + "accept": true, + "reason": "Same from_single_file ControlNet loading problem; 11178 is a concrete instance of the broader 9208 issue." + }, + { + "left": "issue:3432", + "right": "issue:3453", + "accept": false, + "reason": "Both are precision/runtime errors in examples, but on different ops, dtypes, and scripts." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: it correctly treats the cluster as mixed and only elevates clearly duplicated subtopics where the textual overlap is strong. The accepted soft-edge pairs are conservative enough, and the rejected pairs avoid overclaiming sameness from broad topical similarity alone. No PR-pair mergeability issue applies here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 31, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10144", + "issue:10343", + "issue:11062", + "issue:11209", + "issue:11224", + "issue:11286", + "issue:11374", + "issue:11569", + "issue:11657", + "issue:12047", + "issue:12052", + "issue:12458", + "issue:12459", + "issue:12499", + "issue:12896", + "issue:13311", + "issue:1962", + "issue:197", + "issue:199", + "issue:2462", + "issue:2838", + "issue:2941", + "issue:3061", + "issue:3094", + "issue:3284", + "issue:3363", + "issue:3451", + "issue:3884", + "issue:4680", + "issue:4708", + "issue:5118", + "issue:5594", + "issue:5603", + "issue:5633", + "issue:5889", + "issue:6418", + "issue:6584", + "issue:6636", + "issue:6689", + "issue:7330", + "issue:7418", + "issue:9042", + "issue:9148", + "issue:9237", + "issue:9497", + "issue:9551", + "issue:9692" + ], + "soft_pairs": [ + "issue:12499|issue:12896", + "issue:12458|issue:12459", + "issue:3284|issue:3363", + "issue:11209|issue:11286", + "issue:11224|issue:11286", + "issue:3094|issue:3451", + "issue:1962|issue:5889", + "issue:4680|issue:5594", + "issue:6584|issue:6689", + "issue:9042|issue:9148", + "issue:12047|issue:12052", + "issue:3884|issue:7330", + "issue:4708|issue:5603", + "issue:4708|issue:6636", + "issue:2462|issue:2838", + "issue:2941|issue:3061", + "issue:5118|issue:9551", + "issue:11374|issue:11569", + "issue:10144|issue:6418", + "issue:11062|issue:11286", + "issue:10343|issue:9692", + "issue:197|issue:199", + "issue:9237|issue:9497", + "issue:5633|issue:7418", + "issue:11657|issue:13311" + ], + "prepared_review_unit_hash": "89cc845ffdb129dfe49ad785f6410e104a612028193b9f84bdca2ef65539bb8c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33415, + "estimated_input_tokens": 8354, + "estimated_eval_tokens": 16964 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly heterogeneous: many feature requests and unrelated model-specific bugs. The clearest duplicate pocket is the LoRA-loading reports (11209/11224/11286), with a likely duplicate pair for Flux Dreambooth training (9237/9497).", + "confidence": 0.74, + "canonical_issue_reason": "issue:11224 is the strongest representative of the LoRA-loading duplicate set: it has the same generic failure title as 11209/11286 and the most discussion activity among them.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11224 is the best consolidation target for the only clearly duplicated issue family in this batch.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12499", + "right": "issue:12896", + "accept": false, + "reason": "Same Wan 2.2 I2V area, but one is a condition-shape mismatch and the other is an MPS-specific device/backend failure; not the same concrete bug." + }, + { + "left": "issue:12458", + "right": "issue:12459", + "accept": false, + "reason": "Both mention batch inference, but they are for different models/pipelines and one is a throughput request while the other is an issue report." + }, + { + "left": "issue:3284", + "right": "issue:3363", + "accept": true, + "reason": "Both report train_dreambooth_lora.py failing, with very similar wording and no sign of different code paths; plausible duplicate." + }, + { + "left": "issue:11209", + "right": "issue:11286", + "accept": true, + "reason": "Near-identical LoRA loading failure wording in the same subsystem; likely the same underlying bug." + }, + { + "left": "issue:11224", + "right": "issue:11286", + "accept": true, + "reason": "Near-identical LoRA loading failure wording in the same subsystem; likely the same underlying bug." + }, + { + "left": "issue:3094", + "right": "issue:3451", + "accept": false, + "reason": "Both are training-memory related, but one is an OOM during finetuning and the other is an fp16-vs-fp32 memory regression; different bugs." + }, + { + "left": "issue:1962", + "right": "issue:5889", + "accept": false, + "reason": "One is a generic request for video generation with Stable Diffusion, the other is about Stable Video Diffusion specifically; not the same issue." + }, + { + "left": "issue:4680", + "right": "issue:5594", + "accept": false, + "reason": "Both touch SDXL fine-tuning, but one asks about inpainting support and the other about LoRA memory usage; different requests." + }, + { + "left": "issue:6584", + "right": "issue:6689", + "accept": false, + "reason": "Both are conversion-to-Diffusers problems, but they involve different source models and no shared concrete failure is evident." + }, + { + "left": "issue:9042", + "right": "issue:9148", + "accept": false, + "reason": "Related Flux feature requests, but one is the base pipeline and the other is inpainting support; not the same change." + }, + { + "left": "issue:12047", + "right": "issue:12052", + "accept": false, + "reason": "Both involve Wan 2.2 and LightX2V, but one is LoRA fusion on GGUF and the other is an offloading device-mismatch bug; different code paths." + }, + { + "left": "issue:3884", + "right": "issue:7330", + "accept": false, + "reason": "Same training script, but the failure descriptions are too generic vs. a specific FP16 unscale error; not enough evidence of the same bug." + }, + { + "left": "issue:4708", + "right": "issue:5603", + "accept": false, + "reason": "Both are import errors from diffusers.utils, but the concrete missing symbol/module failure is different and not clearly the same root cause." + }, + { + "left": "issue:4708", + "right": "issue:6636", + "accept": false, + "reason": "Both are import errors from diffusers.utils, but they point to different missing names and could be separate regressions." + }, + { + "left": "issue:2462", + "right": "issue:2838", + "accept": false, + "reason": "Different feature requests in different areas; only loosely related by training/pipeline support." + }, + { + "left": "issue:2941", + "right": "issue:3061", + "accept": false, + "reason": "Both concern Stable Diffusion outputs, but one is scheduler-specific undesired results and the other is general reproducibility; not the same bug." + }, + { + "left": "issue:5118", + "right": "issue:9551", + "accept": false, + "reason": "Both mention ControlNet, but one is about reference_only usage and the other about Flux ControlNet models; different tasks." + }, + { + "left": "issue:11374", + "right": "issue:11569", + "accept": false, + "reason": "Both are model-support feature requests, but for different models and different integration work." + }, + { + "left": "issue:10144", + "right": "issue:6418", + "accept": false, + "reason": "Both are video-quality complaints, but they involve different models and different symptoms; not the same underlying issue." + }, + { + "left": "issue:11062", + "right": "issue:11286", + "accept": false, + "reason": "Both are LoRA-loading problems, but 11062 is specific to a particular CivitAI LoRA while 11286 is a generic loading error; not enough to treat as the same bug." + }, + { + "left": "issue:10343", + "right": "issue:9692", + "accept": false, + "reason": "Same CogView3-Plus family, but one is a black-image/cast warning and the other is a startup error; different failures." + }, + { + "left": "issue:197", + "right": "issue:199", + "accept": false, + "reason": "Both are dead-code cleanup items, but they target different modules and are not the same change." + }, + { + "left": "issue:9237", + "right": "issue:9497", + "accept": true, + "reason": "Both are Flux Dreambooth/LoRA training problems, and 9497 looks like a concrete runtime manifestation of the broader training issue in 9237." + }, + { + "left": "issue:5633", + "right": "issue:7418", + "accept": false, + "reason": "Different files and different bug reports; no evidence they are the same defect." + }, + { + "left": "issue:11657", + "right": "issue:13311", + "accept": false, + "reason": "Both mention missing LoRA keys with prefix='transformer', but they affect different model families and likely different loaders." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: the LoRA-loading trio is the strongest duplicate pocket, and the 9237/9497 Flux Dreambooth pair is plausibly related. The non-duplicate calls are generally conservative and supported by the differing titles/symptoms." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 32, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10226", + "issue:10282", + "issue:10749", + "issue:10848", + "issue:10869", + "issue:10872", + "issue:10958", + "issue:11315", + "issue:11824", + "issue:11839", + "issue:11865", + "issue:1214", + "issue:12216", + "issue:12645", + "issue:12653", + "issue:1925", + "issue:2830", + "issue:2966", + "issue:3019", + "issue:3026", + "issue:3420", + "issue:3774", + "issue:3885", + "issue:4520", + "issue:4708", + "issue:4733", + "issue:4736", + "issue:4872", + "issue:5226", + "issue:5227", + "issue:5757", + "issue:6794", + "issue:7168", + "issue:7263", + "issue:7306", + "issue:7492", + "issue:7513", + "issue:7621", + "issue:8359", + "issue:8939", + "issue:9026", + "issue:9141", + "issue:9237", + "issue:9448", + "issue:9497", + "issue:9548", + "issue:9857" + ], + "soft_pairs": [ + "issue:1214|issue:1925", + "issue:11315|issue:12653", + "issue:4708|issue:6794", + "issue:7168|issue:7306", + "issue:2966|issue:3026", + "issue:7621|issue:8359", + "issue:4736|issue:4872", + "issue:10749|issue:10872", + "issue:7263|issue:9448", + "issue:5226|issue:5227", + "issue:3420|issue:4520", + "issue:7513|issue:9237", + "issue:10226|issue:9857", + "issue:11824|issue:11839", + "issue:10848|issue:9026", + "issue:4733|issue:5757", + "issue:10869|issue:10958", + "issue:7492|issue:8939", + "issue:12216|issue:12645", + "issue:3774|issue:3885", + "issue:11865|issue:9141", + "issue:2830|issue:3019", + "issue:10282|issue:11839", + "issue:9497|issue:9548" + ], + "prepared_review_unit_hash": "c8d315b9364c00ee5b1089325b80a4be00eeeeb25158216210e1892d3cc59ae0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33858, + "estimated_input_tokens": 8465, + "estimated_eval_tokens": 17186 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated issues across different pipelines/models. The only clear duplicate-looking pair is the rope float64 MPS regression. No PRs are present.", + "confidence": 0.71, + "canonical_issue_reason": "issue:9237 is the best representative anchor because it is the broadest FLUX dreambooth/LoRA training thread and has the strongest follow-up activity among the Flux-training issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9237 is the most useful global issue to keep as the canonical reference for the recurring Flux dreambooth/LoRA reports; it is broad, central, and has the most discussion/inbound references in that subtopic.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1214", + "right": "issue:1925", + "accept": false, + "reason": "Both mention half-precision, but one is CLIPTextModel-specific and the other is SD2 inpainting; too broad to treat as the same bug." + }, + { + "left": "issue:11315", + "right": "issue:12653", + "accept": true, + "reason": "Both describe the same rope implementation float64 regression breaking MPS/NPU; same underlying bug and likely one fix." + }, + { + "left": "issue:4708", + "right": "issue:6794", + "accept": false, + "reason": "Both are import errors, but they point to different symbols/modules and do not clearly share the same root cause." + }, + { + "left": "issue:7168", + "right": "issue:7306", + "accept": false, + "reason": "Different problems: IP-Adapter image-embed shape handling versus float16 support in LEditsPPPipelineStableDiffusionXL." + }, + { + "left": "issue:2966", + "right": "issue:3026", + "accept": false, + "reason": "Same script family, but one is multi-GPU training and the other is a single-GPU accelerate launch failure; not enough evidence of the same concrete bug." + }, + { + "left": "issue:7621", + "right": "issue:8359", + "accept": false, + "reason": "Different features: cosxl_edit pipeline compatibility versus from_single_file with custom checkpoints." + }, + { + "left": "issue:4736", + "right": "issue:4872", + "accept": false, + "reason": "Both mention SDXL finetuning OOM, but one also has an invalid-cast error and they do not look like the same specific failure." + }, + { + "left": "issue:10749", + "right": "issue:10872", + "accept": false, + "reason": "Both are feature requests for new loading support, but they target different models and formats." + }, + { + "left": "issue:7263", + "right": "issue:9448", + "accept": false, + "reason": "Both involve IP-Adapter, but the failures are different: attention slicing versus a tuple/shape error in ControlNet inpaint." + }, + { + "left": "issue:5226", + "right": "issue:5227", + "accept": false, + "reason": "Both concern LoRA loading, but one is AdapterPipeline compatibility and the other is torch.compile interaction; different code paths." + }, + { + "left": "issue:3420", + "right": "issue:4520", + "accept": false, + "reason": "Different subsystems and symptoms: ControlNet/Diffusion pipeline mismatch versus LoRA scale with attention slicing." + }, + { + "left": "issue:7513", + "right": "issue:9237", + "accept": false, + "reason": "Both are Dreambooth/LoRA training reports, but one is SDXL on a 3060 and the other is FLUX training; not the same bug." + }, + { + "left": "issue:10226", + "right": "issue:9857", + "accept": false, + "reason": "Both are Flux tensor-shape failures, but one is Dreambooth LoRA training and the other is ControlNet embedding mismatch; too different to merge." + }, + { + "left": "issue:11824", + "right": "issue:11839", + "accept": false, + "reason": "Same pipeline family, but one is multi-image input handling and the other is GGUF loading with invalid input size; not clearly the same defect." + }, + { + "left": "issue:10848", + "right": "issue:9026", + "accept": false, + "reason": "Both are unexpected-kwarg errors, but they occur in different loader functions with different arguments and likely different fixes." + }, + { + "left": "issue:4733", + "right": "issue:5757", + "accept": false, + "reason": "Both are checkpoint-resume issues, but they are on different training scripts and the reports do not show the same underlying failure." + }, + { + "left": "issue:10869", + "right": "issue:10958", + "accept": false, + "reason": "Both are meta-tensor errors, but one is Tensor.item() and the other is copy-out; similar class of symptom, not the same concrete bug." + }, + { + "left": "issue:7492", + "right": "issue:8939", + "accept": false, + "reason": "Both are loading failures, but one is a missing repository and the other is a tokenizer load error for a different model." + }, + { + "left": "issue:12216", + "right": "issue:12645", + "accept": false, + "reason": "Same model family, but one is quality comparison and the other is speed comparison; not the same issue." + }, + { + "left": "issue:3774", + "right": "issue:3885", + "accept": false, + "reason": "Both concern VRAM/memory in train_text_to_image.py, but one is zero3 not saving memory and the other is a general VRAM reduction question." + }, + { + "left": "issue:11865", + "right": "issue:9141", + "accept": false, + "reason": "Both mention missing from_single_file support, but they target different model classes and different user requests." + }, + { + "left": "issue:2830", + "right": "issue:3019", + "accept": false, + "reason": "Image-to-video/modelscope question versus segment-anything for text-guided editing; unrelated." + }, + { + "left": "issue:10282", + "right": "issue:11839", + "accept": false, + "reason": "Both involve shape/input-size mismatch, but one is a Flux x_embedder weight shape issue and the other is GGUF load input-size validation; different failures." + }, + { + "left": "issue:9497", + "right": "issue:9548", + "accept": false, + "reason": "Both are Flux Dreambooth/LoRA training reports, but one is a mat2 dimensionality error and the other is a follow-up on the earlier Flux issue; not the same concrete bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: issue:9237 is a reasonable canonical anchor for the Flux dreambooth/LoRA cluster, the rope float64 MPS pair is the only clearly duplicate-looking one, and the note that no PRs are present is supported. The soft-edge verdicts are generally conservative and do not overstate duplicate evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 33, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10937", + "issue:11464", + "issue:11630", + "issue:11674", + "issue:11864", + "issue:12009", + "issue:12047", + "issue:12459", + "issue:2052", + "issue:2089", + "issue:2244", + "issue:2250", + "issue:2636", + "issue:3289", + "issue:3391", + "issue:3728", + "issue:3788", + "issue:3826", + "issue:4185", + "issue:4213", + "issue:4398", + "issue:4733", + "issue:4872", + "issue:5005", + "issue:5226", + "issue:5412", + "issue:5840", + "issue:6201", + "issue:6274", + "issue:6275", + "issue:6277", + "issue:6363", + "issue:7057", + "issue:7142", + "issue:7355", + "issue:7432", + "issue:7644", + "issue:771", + "issue:8645", + "issue:8723", + "issue:9012", + "issue:9115", + "issue:9187", + "issue:9343", + "issue:9667", + "issue:9873", + "issue:9914" + ], + "soft_pairs": [ + "issue:4185|issue:4398", + "issue:2636|issue:5005", + "issue:8723|issue:9012", + "issue:6274|issue:6275", + "issue:11464|issue:12459", + "issue:10937|issue:11864", + "issue:6277|issue:6363", + "issue:3728|issue:9873", + "issue:9343|issue:9667", + "issue:7142|issue:7432", + "issue:7355|issue:7644", + "issue:3788|issue:9873", + "issue:9115|issue:9187", + "issue:2052|issue:4733", + "issue:3289|issue:3391", + "issue:11630|issue:11674", + "issue:2244|issue:2250", + "issue:9343|issue:9914", + "issue:7057|issue:771", + "issue:6201|issue:8645", + "issue:2089|issue:4872", + "issue:3788|issue:3826", + "issue:5412|issue:5840", + "issue:4213|issue:5226", + "issue:12009|issue:12047" + ], + "prepared_review_unit_hash": "4a9321ff7cf7fd748e76f181fc590f2669c771739f70523e759d5abf66c8e845", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33845, + "estimated_input_tokens": 8462, + "estimated_eval_tokens": 17180 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a mix of unrelated issues, with a few clear duplicate-style pairs around the same concrete bug (resume-from-checkpoint progress handling, Stable Cascade resolution failures, and some model-loading errors). I did not find a single global canonical issue/PR for the whole cluster.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4185", + "right": "issue:4398", + "accept": false, + "reason": "Both mention black images, but they involve different pipelines and likely different root causes." + }, + { + "left": "issue:2636", + "right": "issue:5005", + "accept": true, + "reason": "Same resume-from-checkpoint progress-bar/skip-step bug in training scripts." + }, + { + "left": "issue:8723", + "right": "issue:9012", + "accept": true, + "reason": "Both target SD3ControlNetModel.from_transformer loading/initialization failures on the same code path." + }, + { + "left": "issue:6274", + "right": "issue:6275", + "accept": false, + "reason": "Both are T2I-Adapter issues, but one is about multiple adapters and the other about model coverage." + }, + { + "left": "issue:11464", + "right": "issue:12459", + "accept": false, + "reason": "Different FLUX problems: model loading vs batch inference throughput." + }, + { + "left": "issue:10937", + "right": "issue:11864", + "accept": false, + "reason": "Both are torch.compile encode failures, but they affect different models and error modes." + }, + { + "left": "issue:6277", + "right": "issue:6363", + "accept": true, + "reason": "Same train_text_to_image_lora.py gradient/unscale failure reported at different specificity levels." + }, + { + "left": "issue:3728", + "right": "issue:9873", + "accept": false, + "reason": "Both are new pipeline requests, but for completely different models." + }, + { + "left": "issue:9343", + "right": "issue:9667", + "accept": false, + "reason": "Different FLUX load-time issues with different flags and failure contexts." + }, + { + "left": "issue:7142", + "right": "issue:7432", + "accept": false, + "reason": "Both compare diffusers to WebUI, but the model/context differ and the reports are too broad to treat as the same bug." + }, + { + "left": "issue:7355", + "right": "issue:7644", + "accept": true, + "reason": "Same Stable Cascade resolution-handling failure with arbitrary / non-power-of-two sizes." + }, + { + "left": "issue:3788", + "right": "issue:9873", + "accept": false, + "reason": "Different new pipeline proposals with no shared concrete bug or feature." + }, + { + "left": "issue:9115", + "right": "issue:9187", + "accept": false, + "reason": "Different adapter integrations and different failure modes." + }, + { + "left": "issue:2052", + "right": "issue:4733", + "accept": false, + "reason": "Both involve checkpoint resume, but they are different training scripts and errors." + }, + { + "left": "issue:3289", + "right": "issue:3391", + "accept": false, + "reason": "Same DeepFloyd area, but the bugs are different symptoms and likely different causes." + }, + { + "left": "issue:11630", + "right": "issue:11674", + "accept": false, + "reason": "Same Wan VACE family, but one is single-file support and the other is ref-image/multi-control-video support." + }, + { + "left": "issue:2244", + "right": "issue:2250", + "accept": false, + "reason": "Both are DreamBooth inference issues, but the failures are unrelated." + }, + { + "left": "issue:9343", + "right": "issue:9914", + "accept": false, + "reason": "Both concern FLUX loading, but one is a mismatched-size/load flag issue and the other is LoRA loading with Xlabs." + }, + { + "left": "issue:7057", + "right": "issue:771", + "accept": false, + "reason": "Both are checkpoint or precision-related, but the actual models and failure paths differ." + }, + { + "left": "issue:6201", + "right": "issue:8645", + "accept": false, + "reason": "Different code paths entirely: prompt embedding handling vs missing single_file_utils helper." + }, + { + "left": "issue:2089", + "right": "issue:4872", + "accept": false, + "reason": "Both are OOM training reports, but on different models and scenarios." + }, + { + "left": "issue:3788", + "right": "issue:3826", + "accept": false, + "reason": "Both are new pipeline requests, but they are unrelated models/features." + }, + { + "left": "issue:5412", + "right": "issue:5840", + "accept": true, + "reason": "Same train_dreambooth_lora_sdxl.py checkpoint-resume failure." + }, + { + "left": "issue:4213", + "right": "issue:5226", + "accept": false, + "reason": "Different SDXL loading features and different failures." + }, + { + "left": "issue:12009", + "right": "issue:12047", + "accept": false, + "reason": "Same model family, but one is single-file loading and the other is LoRA fusion on GGUF; not clearly the same bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet, and the soft-edge decisions are conservative overall. The accepted pairs are plausible duplicates or near-duplicates based on the titles, while the rejected pairs are appropriately cautious and do not overstate identity across different models, scripts, or failure modes. No canonical issue/PR is asserted without support, and the summary does not overclaim a single global bug." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 34, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10350", + "issue:10472", + "issue:10653", + "issue:10848", + "issue:11948", + "issue:12082", + "issue:12195", + "issue:12458", + "issue:12645", + "issue:1495", + "issue:2089", + "issue:2294", + "issue:2392", + "issue:2459", + "issue:2560", + "issue:2772", + "issue:2774", + "issue:2883", + "issue:2891", + "issue:3008", + "issue:3094", + "issue:3184", + "issue:3569", + "issue:3631", + "issue:3639", + "issue:4000", + "issue:4580", + "issue:4779", + "issue:5433", + "issue:5465", + "issue:5577", + "issue:5661", + "issue:5846", + "issue:5917", + "issue:6073", + "issue:6183", + "issue:6295", + "issue:7147", + "issue:7185", + "issue:7209", + "issue:7210", + "issue:7267", + "issue:771", + "issue:8494", + "issue:852", + "issue:9501", + "issue:9727", + "issue:9953" + ], + "soft_pairs": [ + "issue:2883|issue:4779", + "issue:10653|issue:9953", + "issue:10350|issue:12082", + "issue:3569|issue:3639", + "issue:5433|issue:6295", + "issue:2772|issue:7267", + "issue:5577|issue:6183", + "issue:9501|issue:9727", + "issue:2089|issue:3094", + "issue:1495|issue:7147", + "issue:3008|issue:3184", + "issue:12195|issue:7185", + "issue:5661|issue:5917", + "issue:2774|issue:4000", + "issue:11948|issue:7209", + "issue:12458|issue:12645", + "issue:3631|issue:4580", + "issue:5846|issue:6073", + "issue:2392|issue:2560", + "issue:2459|issue:771", + "issue:2891|issue:5465", + "issue:10472|issue:10848", + "issue:8494|issue:852", + "issue:2294|issue:7147", + "issue:6183|issue:7210" + ], + "prepared_review_unit_hash": "3869d19c7f88ed381f1cd72f0597f44f79c60aa0528ca70d94baf2dcb24dacca", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34770, + "estimated_input_tokens": 8693, + "estimated_eval_tokens": 17642 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous. Two pairs look like true duplicates or near-duplicates around the same code path; the rest are only loosely similar by topic and should stay separate.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2883", + "right": "issue:4779", + "accept": false, + "reason": "Both mention image-to-video, but one is a generic pipeline request and the other is a ModelScope-specific pipeline item; not the same bug/change." + }, + { + "left": "issue:10653", + "right": "issue:9953", + "accept": false, + "reason": "Both involve quantized modules and device moves, but one is 8-bit to CPU and the other is moving a quantized component to CUDA; different failure paths." + }, + { + "left": "issue:10350", + "right": "issue:12082", + "accept": false, + "reason": "Both mention VAE memory, but they target different models and triggers (SD2.1 decoder vs Wan2.1 after compile)." + }, + { + "left": "issue:3569", + "right": "issue:3639", + "accept": false, + "reason": "These are separate new-pipeline requests for different features (Uni-ControlNet vs GlyphControl)." + }, + { + "left": "issue:5433", + "right": "issue:6295", + "accept": false, + "reason": "Both are about DPM++ artifacts, but the reported conditions differ enough (missing refiner vs use_lu_lambdas) that they do not clearly describe one concrete bug." + }, + { + "left": "issue:2772", + "right": "issue:7267", + "accept": true, + "reason": "Both describe the same train_controlnet.py local-dataset/local-loading problem; the titles point to the same underlying script bug." + }, + { + "left": "issue:5577", + "right": "issue:6183", + "accept": false, + "reason": "Both involve from_single_file, but they hit different model types and different errors (ControlNet key error vs SDXL shape issue)." + }, + { + "left": "issue:9501", + "right": "issue:9727", + "accept": false, + "reason": "Both are DreamBooth/FLUX save problems, but one is delayed saving and the other is a multigpu save failure; not clearly the same bug." + }, + { + "left": "issue:2089", + "right": "issue:3094", + "accept": false, + "reason": "Both are OOM reports during training, but they involve different examples and contexts; too broad to be one bug." + }, + { + "left": "issue:1495", + "right": "issue:7147", + "accept": false, + "reason": "Both are import errors, but for different symbols and different releases; not the same issue." + }, + { + "left": "issue:3008", + "right": "issue:3184", + "accept": true, + "reason": "Both are about lpw_stable_diffusion handling generator lists/invalid torch.randn arguments, which looks like the same concrete pipeline bug." + }, + { + "left": "issue:12195", + "right": "issue:7185", + "accept": false, + "reason": "One is a specific AutoencoderKL failure on PyTorch 2.3.1; the other is a general question about PyTorch 2.2.0 support." + }, + { + "left": "issue:5661", + "right": "issue:5917", + "accept": false, + "reason": "Both are upgrade-related errors, but they concern unrelated missing names in different code paths." + }, + { + "left": "issue:2774", + "right": "issue:4000", + "accept": false, + "reason": "Both are conversion problems, but one is ControlNet safetensors conversion and the other is SD 2.0 ckpt conversion; different tools and likely different bugs." + }, + { + "left": "issue:11948", + "right": "issue:7209", + "accept": false, + "reason": "Both concern offline loading, but one is from_pretrained for WanTransformer3DModel and the other is from_single_file with a local safetensor; different loading paths." + }, + { + "left": "issue:12458", + "right": "issue:12645", + "accept": false, + "reason": "One is a batch inference issue/feature request and the other is a speed comparison complaint; not the same bug." + }, + { + "left": "issue:3631", + "right": "issue:4580", + "accept": false, + "reason": "Different pipelines and errors: inpainting latents exception vs UniPC schedule/use_karras_sigmas exception." + }, + { + "left": "issue:5846", + "right": "issue:6073", + "accept": false, + "reason": "Both are beginner DreamBooth support reports, but they do not identify the same concrete failure." + }, + { + "left": "issue:2392", + "right": "issue:2560", + "accept": false, + "reason": "Both touch StableDiffusionImg2ImgPipeline, but one is a prompt_embeds validation error and the other is differing outputs with a generator list." + }, + { + "left": "issue:2459", + "right": "issue:771", + "accept": false, + "reason": "Unrelated training/setup failures: xformers enablement vs BF16 support." + }, + { + "left": "issue:2891", + "right": "issue:5465", + "accept": false, + "reason": "Both mention unsupported options/docs mismatches, but they are different features (embeds vs inpaint_strength) and not one bug." + }, + { + "left": "issue:10472", + "right": "issue:10848", + "accept": false, + "reason": "Both are unexpected-keyword errors, but for different arguments and loading functions; too generic to merge." + }, + { + "left": "issue:8494", + "right": "issue:852", + "accept": false, + "reason": "Different unexpected keyword arguments in different versions/code paths; not the same defect." + }, + { + "left": "issue:2294", + "right": "issue:7147", + "accept": false, + "reason": "Both are import errors, but they reference different missing pipeline symbols and different eras." + }, + { + "left": "issue:6183", + "right": "issue:7210", + "accept": false, + "reason": "One is a from_single_file shape issue and the other is an accelerate dispatch attribute error; unrelated failures." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The evaluation is grounded in the packet and stays conservative. The two accepted pairs are reasonably supported as likely duplicates/near-duplicates, and the remaining pairs are appropriately rejected for being only loosely related or clearly different failure modes. The cluster summary also matches the evidence: mostly heterogeneous with only two plausible duplicate pairs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 35, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1013", + "issue:10994", + "issue:11045", + "issue:11127", + "issue:1181", + "issue:1214", + "issue:12195", + "issue:12666", + "issue:13277", + "issue:13319", + "issue:1538", + "issue:1614", + "issue:1962", + "issue:2163", + "issue:2371", + "issue:2480", + "issue:3112", + "issue:3221", + "issue:3450", + "issue:3694", + "issue:4210", + "issue:4480", + "issue:4566", + "issue:4616", + "issue:4708", + "issue:4724", + "issue:5025", + "issue:5128", + "issue:5226", + "issue:5314", + "issue:5442", + "issue:5633", + "issue:579", + "issue:5931", + "issue:6574", + "issue:692", + "issue:7223", + "issue:7239", + "issue:7243", + "issue:7483", + "issue:7708", + "issue:7871", + "issue:9343", + "issue:9525", + "issue:9803", + "issue:9914", + "issue:993", + "issue:9971" + ], + "soft_pairs": [ + "issue:3112|issue:5314", + "issue:12195|issue:12666", + "issue:1013|issue:1538", + "issue:2371|issue:2480", + "issue:4708|issue:5025", + "issue:9803|issue:9914", + "issue:1181|issue:692", + "issue:4480|issue:5931", + "issue:10994|issue:11045", + "issue:4566|issue:7239", + "issue:11127|issue:9971", + "issue:5442|issue:7243", + "issue:4724|issue:6574", + "issue:3221|issue:3450", + "issue:579|issue:7708", + "issue:4616|issue:7483", + "issue:5128|issue:7871", + "issue:1214|issue:5226", + "issue:1962|issue:7223", + "issue:4210|issue:5633", + "issue:9343|issue:9525", + "issue:1614|issue:993", + "issue:13277|issue:13319", + "issue:2163|issue:3694" + ], + "prepared_review_unit_hash": "964a7151bb94f18a804812e8aaa39c3bb98644817efdda03b912b21b9665ed7b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33985, + "estimated_input_tokens": 8497, + "estimated_eval_tokens": 17250 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated diffusers bugs/feature requests; only one soft pair looks like a true duplicate around the SDXL LoRA-to-Diffusers conversion script. The rest are too broad, different code paths, or only share a subsystem name.", + "confidence": 0.84, + "canonical_issue_reason": "issue:7483 is the better canonical issue for the accepted duplicate pair because it is more specific about the failing SDXL 1.0 base-model conversion path, while issue:4616 is broader and less explicit.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:7483 is the strongest representative issue in this set: it clearly states the broken script and the SDXL 1.0 base-model condition, making it easier to anchor discussion and deduping.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3112", + "right": "issue:5314", + "accept": false, + "reason": "Both concern ControlNet/inpainting, but one is a multi-controlnet request and the other is reference inpainting; different change requests, not the same bug." + }, + { + "left": "issue:12195", + "right": "issue:12666", + "accept": false, + "reason": "Completely different problems: an AutoencoderKL custom_op compatibility error versus a missing Kandinsky video pipeline release feature." + }, + { + "left": "issue:1013", + "right": "issue:1538", + "accept": false, + "reason": "Same pipeline family, but the failures are different: one is a version-compatibility issue and the other is model-specific compatibility with stable-diffusion-2-base." + }, + { + "left": "issue:2371", + "right": "issue:2480", + "accept": false, + "reason": "Different Dreambooth output/checkpoint problems; one is about generated objects not appearing, the other about checkpoint directory layout." + }, + { + "left": "issue:4708", + "right": "issue:5025", + "accept": false, + "reason": "Both are import-related, but one is a generic diffusers.utils import error and the other is specifically missing randn_tensor; not enough evidence they share the same root cause." + }, + { + "left": "issue:9803", + "right": "issue:9914", + "accept": false, + "reason": "Both mention Flux LoRA loading, but the titles point to different failure modes and Xlabs-specific loading behavior; too broad to treat as the same concrete bug." + }, + { + "left": "issue:1181", + "right": "issue:692", + "accept": false, + "reason": "A training-from-scratch issue and a request for a fine-tuning example are not the same underlying bug or change." + }, + { + "left": "issue:4480", + "right": "issue:5931", + "accept": false, + "reason": "Different symptoms and likely different root causes: MPS kernel crash with fp16 versus nondeterministic generation." + }, + { + "left": "issue:10994", + "right": "issue:11045", + "accept": false, + "reason": "Both involve train_dreambooth_lora_flux, but one crashes with batch size > 1 during training while the other crashes when loading Flux Schnell 1; not the same code-path problem." + }, + { + "left": "issue:4566", + "right": "issue:7239", + "accept": false, + "reason": "Checkpoint resume failure and SDXL LoRA shape-mismatch error are distinct issues." + }, + { + "left": "issue:11127", + "right": "issue:9971", + "accept": false, + "reason": "Both are Flux-related, but one is a QwenImage single-file loading issue and the other is a scheduler issue; different concrete bugs." + }, + { + "left": "issue:5442", + "right": "issue:7243", + "accept": false, + "reason": "These are separate model/pipeline feature requests (ScaleCrafter vs ResAdapter), not duplicates." + }, + { + "left": "issue:4724", + "right": "issue:6574", + "accept": false, + "reason": "An SDXL img2img correctness issue and an SVD pipeline warning are unrelated." + }, + { + "left": "issue:3221", + "right": "issue:3450", + "accept": false, + "reason": "Different loading APIs and different failure contexts; only the general theme of pipeline loading overlaps." + }, + { + "left": "issue:579", + "right": "issue:7708", + "accept": false, + "reason": "Both mention missing config.json, but they target different loading workflows and model types; not clearly the same bug." + }, + { + "left": "issue:4616", + "right": "issue:7483", + "accept": true, + "reason": "Both describe the SDXL LoRA-to-Diffusers conversion script failing for SDXL-based models; same concrete bug/change with closely matching scope." + }, + { + "left": "issue:5128", + "right": "issue:7871", + "accept": false, + "reason": "A single-file loading error and a blockwise-scale LoRA issue are different problems." + }, + { + "left": "issue:1214", + "right": "issue:5226", + "accept": false, + "reason": "Half-precision CLIPTextModel behavior and SDXL adapter LoRA loading are unrelated." + }, + { + "left": "issue:1962", + "right": "issue:7223", + "accept": false, + "reason": "A generic video generation question and a request to add Latte are not duplicates." + }, + { + "left": "issue:4210", + "right": "issue:5633", + "accept": false, + "reason": "UNet attention_mask usage and a transformer_temporal bug are different code paths and different bugs." + }, + { + "left": "issue:9343", + "right": "issue:9525", + "accept": false, + "reason": "Both are Flux-related, but one is a loading/size mismatch error and the other is lora_scale not taking effect; different defects." + }, + { + "left": "issue:1614", + "right": "issue:993", + "accept": false, + "reason": "Both are precision-related, but black-image output on autocast and generic fp16 dtype mismatch are not the same concrete failure." + }, + { + "left": "issue:13277", + "right": "issue:13319", + "accept": false, + "reason": "Same Ulysses SP topic, but one fails with batch inputs in the pipeline while the other fails in backward pass; different bugs." + }, + { + "left": "issue:2163", + "right": "issue:3694", + "accept": false, + "reason": "UNet dtype mismatch during training and missing model/config files in the example are unrelated failures." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analyst output is grounded in the packet and stays conservative on the soft-edge pairs. The one accepted duplicate pair (issue:4616 / issue:7483) is plausibly the same SDXL LoRA conversion-script bug, and the rest are reasonably rejected as different failure modes, feature requests, or only loosely related subsystem overlaps. The summary does not overstate evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 36, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10169", + "issue:10591", + "issue:10671", + "issue:10722", + "issue:11314", + "issue:11417", + "issue:1214", + "issue:1976", + "issue:2132", + "issue:2459", + "issue:2941", + "issue:3094", + "issue:3325", + "issue:3706", + "issue:3884", + "issue:4389", + "issue:4492", + "issue:4593", + "issue:4697", + "issue:4872", + "issue:5963", + "issue:6028", + "issue:6137", + "issue:6176", + "issue:6383", + "issue:6415", + "issue:6548", + "issue:6741", + "issue:6791", + "issue:6914", + "issue:7057", + "issue:7185", + "issue:7405", + "issue:7513", + "issue:7589", + "issue:8236", + "issue:8307", + "issue:861", + "issue:8613", + "issue:8659", + "issue:8839", + "issue:8909", + "issue:9141", + "issue:9343", + "issue:9450", + "issue:9548", + "issue:9799", + "issue:9857" + ], + "soft_pairs": [ + "issue:6791|issue:8659", + "issue:5963|issue:6028", + "issue:4593|issue:4697", + "issue:6383|issue:7185", + "issue:6176|issue:8839", + "issue:7513|issue:9548", + "issue:3094|issue:4872", + "issue:1214|issue:7589", + "issue:9799|issue:9857", + "issue:10671|issue:9450", + "issue:7057|issue:861", + "issue:2132|issue:6741", + "issue:3706|issue:7405", + "issue:4389|issue:9343", + "issue:4697|issue:6176", + "issue:6137|issue:8236", + "issue:10722|issue:9857", + "issue:6415|issue:6914", + "issue:8613|issue:8909", + "issue:2941|issue:6548", + "issue:10591|issue:11314", + "issue:10169|issue:1976", + "issue:3325|issue:3884", + "issue:11417|issue:4492", + "issue:2459|issue:7057", + "issue:8307|issue:9141" + ], + "prepared_review_unit_hash": "935735cde9dff9cc7f164243105317ba0189630ce8f92a638882fd9f74a87d28", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 34775, + "estimated_input_tokens": 8694, + "estimated_eval_tokens": 17644 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous: many unrelated bug reports across different models, training scripts, and loading paths. Only one clear duplicate pair stands out: 6137 and 8236, both about incorrect LR scheduler step-count calculation in multi-GPU training.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 6137 is the best canonical issue because 8236 is a near-duplicate of the same multi-GPU LR scheduler step-count bug.", + "canonical_pr_reason": null, + "best_issue_reason": "6137 is the strongest representative of the only real duplicate thread in this set; the rest of the items are too diverse to share one underlying fix.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6791", + "right": "issue:8659", + "accept": false, + "reason": "Both mention SDXL training, but one is a generic can't-run issue and the other is checkpoint resume/load behavior; different failures and code paths." + }, + { + "left": "issue:5963", + "right": "issue:6028", + "accept": false, + "reason": "Same model family, but one is failing when passing embeds instead of prompts and the other is a CUDA OOM; not the same bug." + }, + { + "left": "issue:4593", + "right": "issue:4697", + "accept": false, + "reason": "Both are LoRA-loading problems, but they affect different scenarios and error conditions; not clearly the same underlying defect." + }, + { + "left": "issue:6383", + "right": "issue:7185", + "accept": false, + "reason": "A scheduler feature question and a PyTorch version-compatibility question are not duplicate bug reports." + }, + { + "left": "issue:6176", + "right": "issue:8839", + "accept": false, + "reason": "Different LoRA issues on different models/scripts; one is loading weights after DreamBooth training, the other is an SD3 LoRA weight error." + }, + { + "left": "issue:7513", + "right": "issue:9548", + "accept": false, + "reason": "Both involve FLUX/LoRA training, but the reported problems and scripts differ too much to be the same bug." + }, + { + "left": "issue:3094", + "right": "issue:4872", + "accept": false, + "reason": "Both are OOM reports, but they concern different examples and model setups; likely separate memory issues." + }, + { + "left": "issue:1214", + "right": "issue:7589", + "accept": false, + "reason": "Half-precision CLIPTextModel failure is unrelated to GLIGEN layout-to-image behavior." + }, + { + "left": "issue:9799", + "right": "issue:9857", + "accept": false, + "reason": "Both are tensor-size mismatch errors, but the titles point to different FLUX training paths and likely different root causes." + }, + { + "left": "issue:10671", + "right": "issue:9450", + "accept": false, + "reason": "Both mention device/dtype issues, but one is DiffusionPipeline dtype/device access and the other is a multi-GPU tensor device mismatch; not the same bug." + }, + { + "left": "issue:7057", + "right": "issue:861", + "accept": false, + "reason": "A broken checkpoint load is not the same as a conversion-from-ckpt problem." + }, + { + "left": "issue:2132", + "right": "issue:6741", + "accept": false, + "reason": "Different missing-attribute errors in different scripts: Accelerator.trackers vs Namespace.no_flip." + }, + { + "left": "issue:3706", + "right": "issue:7405", + "accept": false, + "reason": "These are model announcement/new-model issues, not duplicates of the same bug or change." + }, + { + "left": "issue:4389", + "right": "issue:9343", + "accept": false, + "reason": "Both are model-loading reports, but ControlNet loading and FLUX low_cpu_mem_usage/ignore_mismatched_sizes failures are distinct." + }, + { + "left": "issue:4697", + "right": "issue:6176", + "accept": false, + "reason": "Both mention loading LoRA weights, but they are different scenarios with different symptoms and no clear shared code-path bug." + }, + { + "left": "issue:6137", + "right": "issue:8236", + "accept": true, + "reason": "These describe the same bug: LR scheduler step-count is computed incorrectly in multi-GPU training, with 8236 giving a specific num_train_epochs case of 6137." + }, + { + "left": "issue:10722", + "right": "issue:9857", + "accept": false, + "reason": "Both are shape-mismatch reports, but they affect different training setups and do not clearly share one root cause." + }, + { + "left": "issue:6415", + "right": "issue:6914", + "accept": false, + "reason": "Both involve IP-Adapter-FaceID, but one is about unsupported features and the other is an offload-related AttributeError; not the same defect." + }, + { + "left": "issue:8613", + "right": "issue:8909", + "accept": false, + "reason": "OOM while using caption_column is unrelated to problems saving text_encoder_ti in a DreamBooth LoRA script." + }, + { + "left": "issue:2941", + "right": "issue:6548", + "accept": false, + "reason": "Scheduler output quality and RGBA latent handling are different pipeline bugs." + }, + { + "left": "issue:10591", + "right": "issue:11314", + "accept": false, + "reason": "Different training scripts, different models, and different reported mistakes; not a duplicate bug." + }, + { + "left": "issue:10169", + "right": "issue:1976", + "accept": false, + "reason": "Both mention config.json, but 1976 is specifically about tokenizer/config.json lookup; 10169 is too generic to confirm the same bug." + }, + { + "left": "issue:3325", + "right": "issue:3884", + "accept": false, + "reason": "Different scripts and errors: DreamBooth CUDA invalid argument vs text-to-image LoRA failure." + }, + { + "left": "issue:11417", + "right": "issue:4492", + "accept": false, + "reason": "An AttributeError on DistributedDataParallel.dtype is unrelated to a lowercase boolean NameError." + }, + { + "left": "issue:2459", + "right": "issue:7057", + "accept": false, + "reason": "xFormers attention not working in a training script is unrelated to a checkpoint load failure." + }, + { + "left": "issue:8307", + "right": "issue:9141", + "accept": false, + "reason": "Missing StableCascadeUNet API and missing FluxPipeline.from_single_file are different missing-symbol issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and suitably conservative. The summary stays within the packet evidence, and the only affirmative soft-edge pair (6137/8236) is well-supported by the closely matching LR-scheduler/multi-GPU titles. The other pairs are rejected cautiously rather than overstated." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 37, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10128", + "issue:10381", + "issue:10798", + "issue:10928", + "issue:11117", + "issue:11396", + "issue:1166", + "issue:12657", + "issue:13249", + "issue:1398", + "issue:1475", + "issue:1552", + "issue:1614", + "issue:1976", + "issue:2089", + "issue:2299", + "issue:2857", + "issue:3325", + "issue:3706", + "issue:3764", + "issue:4000", + "issue:4229", + "issue:4453", + "issue:4566", + "issue:4648", + "issue:4701", + "issue:4708", + "issue:4736", + "issue:4969", + "issue:5578", + "issue:5934", + "issue:6444", + "issue:6794", + "issue:7133", + "issue:7681", + "issue:7728", + "issue:7757", + "issue:7854", + "issue:8414", + "issue:8500", + "issue:8534", + "issue:8557", + "issue:8614", + "issue:8708", + "issue:8833", + "issue:8939", + "issue:9421" + ], + "soft_pairs": [ + "issue:8534|issue:8708", + "issue:6444|issue:7728", + "issue:8500|issue:8557", + "issue:1398|issue:1552", + "issue:3706|issue:8414", + "issue:10381|issue:10798", + "issue:1166|issue:1475", + "issue:3325|issue:4701", + "issue:10128|issue:2857", + "issue:11396|issue:4453", + "issue:4648|issue:5578", + "issue:4708|issue:4969", + "issue:3764|issue:5934", + "issue:10928|issue:8939", + "issue:6794|issue:7681", + "issue:8614|issue:8833", + "issue:11117|issue:7854", + "issue:2299|issue:4000", + "issue:1976|issue:9421", + "issue:2089|issue:4736", + "issue:1398|issue:1614", + "issue:4566|issue:7133", + "issue:12657|issue:13249", + "issue:4229|issue:7757" + ], + "prepared_review_unit_hash": "df4e2576b5d9e3761534940afdb53d42c41937e94e7a389001aa04b7a4f13a9d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 34171, + "estimated_input_tokens": 8543, + "estimated_eval_tokens": 17342 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a heterogeneous mix of unrelated issues (feature requests, model-support questions, file-loading/import errors, and distinct training/runtime bugs). I found no convincing duplicate pairs, and there are no PRs in the cluster.", + "confidence": 0.95, + "canonical_issue_reason": "No canonical issue: the items span multiple unrelated problem types and code paths, so collapsing them into one issue would be incorrect.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10381 is the most detailed, diagnosis-rich report, so it is the best representative anchor if one must pick an issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8534", + "right": "issue:8708", + "accept": false, + "reason": "Both mention SD3 training, but one is about wrong logit-normal weighting and the other about a missing `shift_factor`; different bugs." + }, + { + "left": "issue:6444", + "right": "issue:7728", + "accept": false, + "reason": "Both involve SDXL LoRA training, but one is a `peft_config` checkpointing issue and the other is an Accelerate runtime error; not the same concrete failure." + }, + { + "left": "issue:8500", + "right": "issue:8557", + "accept": false, + "reason": "Both mention token limits, but one is a general question and the other is a training-specific SD3 report; not enough to treat as one bug." + }, + { + "left": "issue:1398", + "right": "issue:1552", + "accept": false, + "reason": "Different problems: Apple Silicon/MPS support for SD2 versus `enable_attention_slicing()` incompatibility." + }, + { + "left": "issue:3706", + "right": "issue:8414", + "accept": false, + "reason": "Both are model-related announcements, but Wuerstchen and ConsistencyTTA are different models/changes." + }, + { + "left": "issue:10381", + "right": "issue:10798", + "accept": false, + "reason": "Both are Flux quantization/LoRA reports, but they involve different backends and symptoms (`Quanto`/Zero GPU spaces vs `bitsandbytes` device-combination bug)." + }, + { + "left": "issue:1166", + "right": "issue:1475", + "accept": false, + "reason": "Textual inversion appears in both, but one is a TPU save hang and the other is a multi-class DreamBooth/textual inversion training question." + }, + { + "left": "issue:3325", + "right": "issue:4701", + "accept": false, + "reason": "Different failures: DreamBooth CUDA invalid argument versus AutoencoderKL differentiation CUDA error." + }, + { + "left": "issue:10128", + "right": "issue:2857", + "accept": false, + "reason": "These are unrelated feature requests for different models/capabilities (`fastercache` vs UniDiffuser)." + }, + { + "left": "issue:11396", + "right": "issue:4453", + "accept": false, + "reason": "Both ask about LoRA conversion, but they target different source models and output ecosystems (hidream/comfyui vs SDXL/AUTO1111)." + }, + { + "left": "issue:4648", + "right": "issue:5578", + "accept": false, + "reason": "Both are missing-file load errors, but they refer to different missing files and different loading contexts." + }, + { + "left": "issue:4708", + "right": "issue:4969", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols/modules and likely different root causes." + }, + { + "left": "issue:3764", + "right": "issue:5934", + "accept": false, + "reason": "Both mention `torch.compile`, but the reported failures are different and not clearly the same code-path bug." + }, + { + "left": "issue:10928", + "right": "issue:8939", + "accept": false, + "reason": "Different loading issues: missing `diffusion_pytorch_model.fp16.bin` versus tokenizer loading failure for SD3 medium." + }, + { + "left": "issue:6794", + "right": "issue:7681", + "accept": false, + "reason": "Both are import-time errors, but the missing symbol and dependency failure are unrelated." + }, + { + "left": "issue:8614", + "right": "issue:8833", + "accept": false, + "reason": "Both mention `config.json`, but they concern different artifacts/models and are not the same loading bug." + }, + { + "left": "issue:11117", + "right": "issue:7854", + "accept": false, + "reason": "Both are dtype mismatch runtime errors, but they affect different code paths and error conditions." + }, + { + "left": "issue:2299", + "right": "issue:4000", + "accept": false, + "reason": "One asks about fine-tuned checkpoints in Diffusers; the other is about converting an SD2 checkpoint to Diffusers format. Related topic, not the same issue." + }, + { + "left": "issue:1976", + "right": "issue:9421", + "accept": false, + "reason": "Both are model access/load errors, but they refer to different models and different missing-path situations." + }, + { + "left": "issue:2089", + "right": "issue:4736", + "accept": false, + "reason": "Both mention training and OOM, but one is DreamBooth on AltDiffusion and the other is SDXL LoRA with an additional invalid-value symptom." + }, + { + "left": "issue:1398", + "right": "issue:1614", + "accept": false, + "reason": "Different SD2/MPS versus autocast black-image behavior; not the same bug." + }, + { + "left": "issue:4566", + "right": "issue:7133", + "accept": false, + "reason": "Both are checkpoint-resume problems for SDXL LoRA, but one is a generic resume failure and the other is a text-encoder size mismatch; too different to merge." + }, + { + "left": "issue:12657", + "right": "issue:13249", + "accept": false, + "reason": "Different models and failure reports: Qwen Image Edit quantization versus Z-Image LoRA support." + }, + { + "left": "issue:4229", + "right": "issue:7757", + "accept": false, + "reason": "Both are multi-GPU training issues, but they involve different scripts and different error classes (accelerated training error vs OOM)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 38, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10453", + "issue:10650", + "issue:10760", + "issue:11417", + "issue:11839", + "issue:1214", + "issue:1925", + "issue:2091", + "issue:3032", + "issue:3061", + "issue:3325", + "issue:3382", + "issue:3454", + "issue:3497", + "issue:3774", + "issue:4213", + "issue:4325", + "issue:4375", + "issue:5314", + "issue:5455", + "issue:5465", + "issue:5474", + "issue:5502", + "issue:5811", + "issue:5884", + "issue:5897", + "issue:5919", + "issue:6096", + "issue:6131", + "issue:6159", + "issue:6226", + "issue:6351", + "issue:6446", + "issue:6492", + "issue:6548", + "issue:6561", + "issue:6821", + "issue:7418", + "issue:7619", + "issue:7681", + "issue:7735", + "issue:7848", + "issue:9497", + "issue:964", + "issue:9643", + "issue:9727", + "issue:9799" + ], + "soft_pairs": [ + "issue:3032|issue:3497", + "issue:3497|issue:5919", + "issue:4325|issue:6492", + "issue:5811|issue:5884", + "issue:3325|issue:7619", + "issue:1925|issue:964", + "issue:6226|issue:6446", + "issue:11417|issue:6351", + "issue:5314|issue:5474", + "issue:3454|issue:5502", + "issue:5465|issue:6159", + "issue:1214|issue:4213", + "issue:6131|issue:6561", + "issue:2091|issue:3382", + "issue:10453|issue:9643", + "issue:4375|issue:6096", + "issue:10650|issue:11839", + "issue:2091|issue:3774", + "issue:9497|issue:9727", + "issue:3061|issue:6548", + "issue:6821|issue:7735", + "issue:9497|issue:9799", + "issue:5455|issue:5897", + "issue:10760|issue:7681", + "issue:7418|issue:7848" + ], + "prepared_review_unit_hash": "766a327c71b41c64442ec4085daf13dcfeb6314509edb5a2fd6149b59050b7b6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 34384, + "estimated_input_tokens": 8596, + "estimated_eval_tokens": 17448 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: many unrelated training/inference/import/runtime bug reports with only one clear duplicate-like pair around `caption_column` validation. Most soft matches are superficial (same pipeline, same framework, or same broad error class) rather than the same underlying bug.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 6821 is the clearest, older report of the `caption_column` validation failure; 7735 appears to be the same bug phrased differently.", + "canonical_pr_reason": null, + "best_issue_reason": "6821 is the strongest representative issue in the only near-duplicate pair and has the most explicit, concise failure description.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3032", + "right": "issue:3497", + "accept": false, + "reason": "Both are attribute errors on different pipeline APIs (`from_pretrained` vs `from_ckpt`); not the same bug." + }, + { + "left": "issue:3497", + "right": "issue:5919", + "accept": false, + "reason": "Different missing attributes (`from_ckpt` vs `set_lora_layer`) and different code paths." + }, + { + "left": "issue:4325", + "right": "issue:6492", + "accept": false, + "reason": "Both mention black images, but one is an SDXL runtime warning/cast issue and the other is a broad symptom report; too vague to treat as the same bug." + }, + { + "left": "issue:5811", + "right": "issue:5884", + "accept": false, + "reason": "One is about callback arguments, the other about IP-Adapter support; distinct feature changes." + }, + { + "left": "issue:3325", + "right": "issue:7619", + "accept": false, + "reason": "Both involve `train_dreambooth.py`, but the reported failures are different (`CUDA invalid argument` vs channel mismatch)." + }, + { + "left": "issue:1925", + "right": "issue:964", + "accept": false, + "reason": "Half-precision inpainting vs generic inpainting failure; not clearly the same issue." + }, + { + "left": "issue:6226", + "right": "issue:6446", + "accept": false, + "reason": "Both are loading problems, but the targets and failure modes differ substantially." + }, + { + "left": "issue:11417", + "right": "issue:6351", + "accept": false, + "reason": "Both are DDP attribute errors, but they involve different missing attributes and likely different fixes." + }, + { + "left": "issue:5314", + "right": "issue:5474", + "accept": false, + "reason": "Related ControlNet/reference themes, but one is a pipeline request and the other is a different reference-only concept." + }, + { + "left": "issue:3454", + "right": "issue:5502", + "accept": false, + "reason": "Different training scripts and different compatibility problems; not the same bug." + }, + { + "left": "issue:5465", + "right": "issue:6159", + "accept": false, + "reason": "Different pipelines and different unsupported parameters (`inpaint_strength` vs `denoising_end`)." + }, + { + "left": "issue:1214", + "right": "issue:4213", + "accept": false, + "reason": "Half-precision CLIP bug vs single-file SDXL loading without accelerate; unrelated." + }, + { + "left": "issue:6131", + "right": "issue:6561", + "accept": false, + "reason": "One is an IP-Adapter support failure, the other is a missing model file/config error." + }, + { + "left": "issue:2091", + "right": "issue:3382", + "accept": false, + "reason": "Both concern multi-GPU training, but one is checkpoint resume and the other is OOM behavior." + }, + { + "left": "issue:10453", + "right": "issue:9643", + "accept": false, + "reason": "Different model families and different limitations (`batch size > 1` vs multiple ControlNets)." + }, + { + "left": "issue:4375", + "right": "issue:6096", + "accept": false, + "reason": "Both involve `torch.compile`, but the model-specific failure modes differ; not enough to call them the same bug." + }, + { + "left": "issue:10650", + "right": "issue:11839", + "accept": false, + "reason": "Both are shape/input-size errors, but on different models and different loading paths." + }, + { + "left": "issue:2091", + "right": "issue:3774", + "accept": false, + "reason": "Different multi-GPU training issues: checkpoint resume vs zero3 memory savings." + }, + { + "left": "issue:9497", + "right": "issue:9727", + "accept": false, + "reason": "Both are Flux Dreambooth reports, but the errors are different (`mat2` shape vs save problem on multigpu)." + }, + { + "left": "issue:3061", + "right": "issue:6548", + "accept": false, + "reason": "Reproducibility issue vs RGBA/image-as-latents bug; not the same underlying defect." + }, + { + "left": "issue:6821", + "right": "issue:7735", + "accept": true, + "reason": "Near-identical `caption_column` validation failure in the training CLI; same underlying bug and likely duplicate report." + }, + { + "left": "issue:9497", + "right": "issue:9799", + "accept": false, + "reason": "Both are Flux training errors, but the tensor-shape failures are different." + }, + { + "left": "issue:5455", + "right": "issue:5897", + "accept": false, + "reason": "Both are LoRA training failures, but they are different scripts and different error types." + }, + { + "left": "issue:10760", + "right": "issue:7681", + "accept": false, + "reason": "An import/export attribute bug vs an environment/OpenSSL import failure; unrelated." + }, + { + "left": "issue:7418", + "right": "issue:7848", + "accept": false, + "reason": "Same community example file family, but one is a code bug and the other is a decoding issue; not clearly the same bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: it correctly identifies the only clear duplicate-like pair (6821/7735) and treats the rest conservatively as superficially similar rather than the same bug. The soft-edge rejections are cautious and supported by the titles/body metadata. No overstatement noted." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 39, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1082", + "issue:11247", + "issue:11410", + "issue:12009", + "issue:12587", + "issue:1260", + "issue:12699", + "issue:12720", + "issue:13073", + "issue:1672", + "issue:1976", + "issue:222", + "issue:2299", + "issue:2331", + "issue:2437", + "issue:2578", + "issue:281", + "issue:3152", + "issue:4213", + "issue:4566", + "issue:4911", + "issue:4969", + "issue:5128", + "issue:579", + "issue:5897", + "issue:6146", + "issue:6154", + "issue:6203", + "issue:6218", + "issue:6353", + "issue:6415", + "issue:6561", + "issue:684", + "issue:6964", + "issue:6967", + "issue:7146", + "issue:7147", + "issue:725", + "issue:7757", + "issue:8871", + "issue:8939", + "issue:8956", + "issue:9051", + "issue:9208", + "issue:9527", + "issue:9548", + "issue:9667", + "issue:9841" + ], + "soft_pairs": [ + "issue:8871|issue:9548", + "issue:12009|issue:9208", + "issue:1672|issue:2299", + "issue:3152|issue:4969", + "issue:1260|issue:7147", + "issue:12699|issue:12720", + "issue:4911|issue:5128", + "issue:2578|issue:4213", + "issue:6353|issue:7146", + "issue:6154|issue:6967", + "issue:6415|issue:6561", + "issue:11247|issue:9527", + "issue:2331|issue:2437", + "issue:222|issue:684", + "issue:1082|issue:5897", + "issue:12587|issue:6203", + "issue:4566|issue:9841", + "issue:1976|issue:579", + "issue:6146|issue:7757", + "issue:6218|issue:6964", + "issue:11410|issue:9667", + "issue:281|issue:725", + "issue:8939|issue:8956", + "issue:13073|issue:9051" + ], + "prepared_review_unit_hash": "13c5d6ef4dbf97229d641e2f2f6b6d9660dc32031f0ca176be8da301e28ffce2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34565, + "estimated_input_tokens": 8642, + "estimated_eval_tokens": 17540 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose set of unrelated training/loading bugs. Only the Accelerate/model-loading feature requests (281 and 725) look like the same underlying change; the rest are too contextually different to merge as duplicates.", + "confidence": 0.63, + "canonical_issue_reason": "Issue 725 is the clearest concrete version of the shared model-loading optimization request and overlaps directly with issue 281\u2019s ask to natively use Accelerate to reduce memory during model loading.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 725 is the strongest canonical candidate because it is specific, actionable, and the only near-duplicate pair in the set clearly points to the same feature area.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8871", + "right": "issue:9548", + "accept": false, + "reason": "Both mention LoRA/Flux training, but the reported failures and scopes differ; this is not the same concrete bug." + }, + { + "left": "issue:12009", + "right": "issue:9208", + "accept": false, + "reason": "Both are from_single_file loading failures, but they affect different model families and fail for different reasons." + }, + { + "left": "issue:1672", + "right": "issue:2299", + "accept": false, + "reason": "Both relate to checkpoint conversion/loading, but one is about wrong output image size and the other is about fast-stable-diffusion checkpoint support." + }, + { + "left": "issue:3152", + "right": "issue:4969", + "accept": false, + "reason": "Generic import errors, but they refer to different missing symbols and unrelated code paths." + }, + { + "left": "issue:1260", + "right": "issue:7147", + "accept": false, + "reason": "Both are import-name failures, but they concern different pipeline classes and do not indicate the same bug." + }, + { + "left": "issue:12699", + "right": "issue:12720", + "accept": false, + "reason": "One is a video quality regression over repeated prompts, the other is a quantization-after-LoRA question; different problems." + }, + { + "left": "issue:4911", + "right": "issue:5128", + "accept": false, + "reason": "Both involve from_single_file, but they concern different model types and different missing/invalid loading data." + }, + { + "left": "issue:2578", + "right": "issue:4213", + "accept": false, + "reason": "Editable install trouble is unrelated to SDXL from_single_file needing accelerate." + }, + { + "left": "issue:6353", + "right": "issue:7146", + "accept": false, + "reason": "Training-example error vs CUDA OOM; same broad area, but not the same underlying bug." + }, + { + "left": "issue:6154", + "right": "issue:6967", + "accept": false, + "reason": "Different training scripts and different failures; not mergeable as one defect." + }, + { + "left": "issue:6415", + "right": "issue:6561", + "accept": false, + "reason": "One is about feature/model support, the other about a missing repository file during loading." + }, + { + "left": "issue:11247", + "right": "issue:9527", + "accept": false, + "reason": "Multi-GPU distribution problem vs dtype mismatch; related to ControlNet/FLUX but not the same code-path bug." + }, + { + "left": "issue:2331", + "right": "issue:2437", + "accept": false, + "reason": "Separate feature requests for different pipelines; same subsystem, different changes." + }, + { + "left": "issue:222", + "right": "issue:684", + "accept": false, + "reason": "Broad training runtime failures, but one is general and the other is AMD-specific memory optimization behavior." + }, + { + "left": "issue:1082", + "right": "issue:5897", + "accept": false, + "reason": "Different LoRA/DreamBooth runtime errors with distinct failure modes and code paths." + }, + { + "left": "issue:12587", + "right": "issue:6203", + "accept": false, + "reason": "These are different questions about different ControlNet conditions." + }, + { + "left": "issue:4566", + "right": "issue:9841", + "accept": false, + "reason": "Checkpoint resume issue vs use_dora TypeError; both in SDXL LoRA training, but not the same bug." + }, + { + "left": "issue:1976", + "right": "issue:579", + "accept": false, + "reason": "Both are config-file lookup errors, but they are for different missing files and different loading flows." + }, + { + "left": "issue:6146", + "right": "issue:7757", + "accept": false, + "reason": "Both concern multi-GPU training, but one is DreamBooth LoRA and the other is ControlNet SDXL OOM; not the same defect." + }, + { + "left": "issue:6218", + "right": "issue:6964", + "accept": false, + "reason": "Both involve ControlNet pipelines, but the failures are different and not the same underlying bug." + }, + { + "left": "issue:11410", + "right": "issue:9667", + "accept": false, + "reason": "Completely different failures: attention tensor unpacking vs loading a flux fp8 model." + }, + { + "left": "issue:281", + "right": "issue:725", + "accept": true, + "reason": "Same underlying feature request: integrate Accelerate/device_map into DiffusionPipeline loading to reduce memory use; 725 is a more concrete restatement of 281." + }, + { + "left": "issue:8939", + "right": "issue:8956", + "accept": false, + "reason": "Both are model-loading issues, but they involve different pipelines/models and different missing tokenizer/model assets." + }, + { + "left": "issue:13073", + "right": "issue:9051", + "accept": false, + "reason": "Different pipeline behavior problems: guidance-scale support in ZImageControlNetPipeline vs batch inference in FluxPipeline." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded in the packet and appropriately conservative: the 281/725 pair is the only clearly same-bug/feature match, and the other soft-edge rejections do not overclaim equivalence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 40, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10226", + "issue:10350", + "issue:10569", + "issue:10920", + "issue:11023", + "issue:1305", + "issue:13079", + "issue:1486", + "issue:2210", + "issue:2774", + "issue:2904", + "issue:2941", + "issue:2947", + "issue:3019", + "issue:3061", + "issue:3451", + "issue:3565", + "issue:4137", + "issue:4213", + "issue:4483", + "issue:5510", + "issue:5846", + "issue:6082", + "issue:6224", + "issue:6280", + "issue:6481", + "issue:6964", + "issue:7142", + "issue:7146", + "issue:7178", + "issue:7185", + "issue:7221", + "issue:730", + "issue:7330", + "issue:761", + "issue:7629", + "issue:7782", + "issue:8504", + "issue:852", + "issue:8686", + "issue:9026", + "issue:9141", + "issue:9270", + "issue:9497", + "issue:9514", + "issue:9810", + "issue:9911" + ], + "soft_pairs": [ + "issue:3061|issue:4483", + "issue:6964|issue:8686", + "issue:2904|issue:2947", + "issue:10920|issue:1486", + "issue:5510|issue:6082", + "issue:2941|issue:4483", + "issue:6224|issue:7146", + "issue:10350|issue:10569", + "issue:6481|issue:761", + "issue:7185|issue:9810", + "issue:4213|issue:7178", + "issue:9270|issue:9514", + "issue:1305|issue:3019", + "issue:13079|issue:7629", + "issue:5846|issue:7330", + "issue:2210|issue:8504", + "issue:2774|issue:7782", + "issue:11023|issue:7221", + "issue:4137|issue:7142", + "issue:852|issue:9026", + "issue:10226|issue:9497", + "issue:6280|issue:9141", + "issue:10026|issue:9911", + "issue:6481|issue:730", + "issue:3451|issue:3565" + ], + "prepared_review_unit_hash": "634f678bdf46a0b81863211b16965048c72ad2a00f506b35a3584cd8041a6230", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34469, + "estimated_input_tokens": 8618, + "estimated_eval_tokens": 17492 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: the soft pairs share only broad topical similarity (e.g., memory, LoRA, scheduler, Flux) but not the same concrete bug or fix path. No duplicate pair is strong enough to merge, so there is no clear canonical issue in this set.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3061", + "right": "issue:4483", + "accept": false, + "reason": "Both mention nondeterminism, but one is a generic Stable Diffusion reproducibility complaint and the other is an SDXL generator determinism issue; different code paths and failure modes." + }, + { + "left": "issue:6964", + "right": "issue:8686", + "accept": false, + "reason": "One is about 4-channel UNet support in inpaint pipelines; the other is a callback incompatibility with ControlNet pipelines. Related subsystem, different bug." + }, + { + "left": "issue:2904", + "right": "issue:2947", + "accept": false, + "reason": "AudioLDM scheduler AttributeError and missing load_attn_procs on UNet are unrelated API/runtime failures." + }, + { + "left": "issue:10920", + "right": "issue:1486", + "accept": false, + "reason": "Both involve DDIM-like schedulers, but one is DDIMInverseScheduler inf on first iteration and the other is DDIMScheduler with prediction_type='sample'; different bug locations." + }, + { + "left": "issue:5510", + "right": "issue:6082", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different pipelines with different concrete errors and likely different fixes." + }, + { + "left": "issue:2941", + "right": "issue:4483", + "accept": false, + "reason": "EulerAncestralDiscreteScheduler incorrect results is not the same as generator non-determinism; different symptoms and likely different causes." + }, + { + "left": "issue:6224", + "right": "issue:7146", + "accept": false, + "reason": "TorchDynamo fine-tuning failure and SDXL training OOM are separate training issues." + }, + { + "left": "issue:10350", + "right": "issue:10569", + "accept": false, + "reason": "Both are memory complaints, but one is VAE decoder memory in SD2.1 and the other is CPU memory usage in HunyuanVideo; different components." + }, + { + "left": "issue:6481", + "right": "issue:761", + "accept": false, + "reason": "DreamBooth batch-size failure and 8-bit Adam failure are distinct training/optimizer problems." + }, + { + "left": "issue:7185", + "right": "issue:9810", + "accept": false, + "reason": "PyTorch version support question and LoRA module_to_save() support are unrelated issues." + }, + { + "left": "issue:4213", + "right": "issue:7178", + "accept": false, + "reason": "from_single_file() dependency issue and export_to_video malfunction are unrelated pipeline problems." + }, + { + "left": "issue:9270", + "right": "issue:9514", + "accept": false, + "reason": "Both are Flux/LoRA-related, but one is incomplete support in SimpleTuner and the other is loading LoRA after fp8 quantization; not the same bug." + }, + { + "left": "issue:1305", + "right": "issue:3019", + "accept": false, + "reason": "Direct Inversion and SAM-based text-guided editing are different feature requests/proposals." + }, + { + "left": "issue:13079", + "right": "issue:7629", + "accept": false, + "reason": "Both mention OOM, but one is FLUX.2-klein in a loop and the other is SVD on PyTorch 1.13.1; different model/runtime contexts." + }, + { + "left": "issue:5846", + "right": "issue:7330", + "accept": false, + "reason": "Both are training-script errors, but resume_from_checkpoint and FP16 unscale are different bugs." + }, + { + "left": "issue:2210", + "right": "issue:8504", + "accept": false, + "reason": "General request to train text encoder with LoRA and a specific SD3-Medium text-encoder selection feature are overlapping themes, but not the same change." + }, + { + "left": "issue:2774", + "right": "issue:7782", + "accept": false, + "reason": "Both are conversion errors, but one is ControlNet safetensors-to-diffusers and the other is invalid safetensors loading; different failure modes." + }, + { + "left": "issue:11023", + "right": "issue:7221", + "accept": false, + "reason": "Both concern bad image outputs, but one is blur with a custom pipeline and the other is noisy output with a model file; not enough to unify." + }, + { + "left": "issue:4137", + "right": "issue:7142", + "accept": false, + "reason": "Both compare diffusers to WebUI, but one is LoRA result mismatch and the other is SDXL-lightning inference mismatch; different pipelines and change targets." + }, + { + "left": "issue:852", + "right": "issue:9026", + "accept": false, + "reason": "Both are unexpected keyword-argument errors, but they affect different APIs and versions; not the same underlying bug." + }, + { + "left": "issue:10226", + "right": "issue:9497", + "accept": false, + "reason": "Both are Flux DreamBooth/LoRA training errors, but the reported failures are different shape/type mismatches and may stem from different code paths." + }, + { + "left": "issue:6280", + "right": "issue:9141", + "accept": false, + "reason": "Both are about missing from_single_file support, but one is a specific ControlNet img2img pipeline and the other is the FluxPipeline API." + }, + { + "left": "issue:10026", + "right": "issue:9911", + "accept": false, + "reason": "DeepSpeed memory scaling and multi-ControlNet layer-length mismatch are unrelated Flux issues." + }, + { + "left": "issue:6481", + "right": "issue:730", + "accept": false, + "reason": "DreamBooth batch-size >1 failure and local-vs-Colab training differences are not the same defect." + }, + { + "left": "issue:3451", + "right": "issue:3565", + "accept": false, + "reason": "Both involve memory during training, but one is fp16 using more memory than fp32 and the other is generic OOM in a text-to-image example; different root causes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 41, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10226", + "issue:10392", + "issue:10848", + "issue:11117", + "issue:11853", + "issue:11922", + "issue:12047", + "issue:12459", + "issue:12535", + "issue:1540", + "issue:1541", + "issue:1672", + "issue:2185", + "issue:2875", + "issue:3284", + "issue:3454", + "issue:4086", + "issue:4094", + "issue:4125", + "issue:4368", + "issue:4398", + "issue:4405", + "issue:4875", + "issue:5016", + "issue:5178", + "issue:5510", + "issue:5565", + "issue:5791", + "issue:5798", + "issue:5956", + "issue:6117", + "issue:6154", + "issue:6232", + "issue:6297", + "issue:6362", + "issue:6382", + "issue:6967", + "issue:7589", + "issue:8589", + "issue:9278", + "issue:9420", + "issue:9497", + "issue:9546", + "issue:9667", + "issue:9841", + "issue:9857", + "issue:9914", + "issue:993" + ], + "soft_pairs": [ + "issue:2185|issue:2875", + "issue:4125|issue:5956", + "issue:11853|issue:6382", + "issue:11117|issue:5510", + "issue:1672|issue:993", + "issue:10226|issue:6967", + "issue:5791|issue:8589", + "issue:4398|issue:9841", + "issue:3284|issue:3454", + "issue:9667|issue:9914", + "issue:4086|issue:7589", + "issue:9278|issue:9497", + "issue:9546|issue:9857", + "issue:4875|issue:5178", + "issue:5565|issue:6154", + "issue:11922|issue:12459", + "issue:4094|issue:4405", + "issue:10848|issue:5798", + "issue:12047|issue:12535", + "issue:10392|issue:9420", + "issue:5016|issue:6297", + "issue:1540|issue:1541", + "issue:4368|issue:6362", + "issue:6117|issue:6232" + ], + "prepared_review_unit_hash": "26b2ca33dc142aeb2bbbf022113fff08b71276a52ee6d194dbb861ad20722b53", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34163, + "estimated_input_tokens": 8541, + "estimated_eval_tokens": 17338 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mixed set of training/loading/performance reports across DreamBooth, LoRA, Flux, SDXL, IP-Adapter, and schedulers. Only the two DreamBooth MODEL_TYPE issues look like a true duplicate pair; the rest are similar in wording but describe different bugs or change requests.", + "confidence": 0.79, + "canonical_issue_reason": "issue:1541 is the clearest representative of the duplicate DreamBooth MODEL_TYPE / CKPT loading failure and states the root symptom more directly than 1540.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:1541 is the best single issue to anchor this cluster because it cleanly captures the DreamBooth checkpoint-loading / MODEL_TYPE failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2185", + "right": "issue:2875", + "accept": false, + "reason": "One is missing training config files; the other is a training-stability/nan-loss problem." + }, + { + "left": "issue:4125", + "right": "issue:5956", + "accept": false, + "reason": "Both concern training behavior, but they are different models and different symptoms (expected loss curve vs degraded image quality)." + }, + { + "left": "issue:11853", + "right": "issue:6382", + "accept": false, + "reason": "Both mention IP-Adapter, but one is a ControlNet/IP-Adapter loading size conflict and the other is an LCM LoRA key error after training." + }, + { + "left": "issue:11117", + "right": "issue:5510", + "accept": false, + "reason": "Different pipelines and different dtype mismatch sites; not the same bug." + }, + { + "left": "issue:1672", + "right": "issue:993", + "accept": false, + "reason": "Wrong image size after conversion is unrelated to an fp16 data-type mismatch." + }, + { + "left": "issue:10226", + "right": "issue:6967", + "accept": false, + "reason": "Both are shape-mismatch reports, but they involve different models and different training contexts (Flux LoRA vs SDXL prior preservation)." + }, + { + "left": "issue:5791", + "right": "issue:8589", + "accept": false, + "reason": "Both mention prior preservation, but one is a parameter-combination question and the other is an SD3-specific prior-preservation training failure." + }, + { + "left": "issue:4398", + "right": "issue:9841", + "accept": false, + "reason": "Black outputs and a use_dora TypeError are different failure modes." + }, + { + "left": "issue:3284", + "right": "issue:3454", + "accept": false, + "reason": "Same script family, but the reported failures are not specific enough to conclude the same underlying bug." + }, + { + "left": "issue:9667", + "right": "issue:9914", + "accept": false, + "reason": "One is loading a Flux fp8 model through a local transformer file; the other is loading a trained LoRA with Xlabs on Diffusers." + }, + { + "left": "issue:4086", + "right": "issue:7589", + "accept": false, + "reason": "Different features entirely: SDXL ControlNet training vs GLIGEN layout-to-image synthesis." + }, + { + "left": "issue:9278", + "right": "issue:9497", + "accept": false, + "reason": "Both are Flux DreamBooth issues, but one is multigpu training and the other is a matmul shape error." + }, + { + "left": "issue:9546", + "right": "issue:9857", + "accept": false, + "reason": "Validation OOM and embedding tensor size mismatch are different problems." + }, + { + "left": "issue:4875", + "right": "issue:5178", + "accept": false, + "reason": "Both involve DPM SDE state across runs, but the triggering condition and scheduler variant differ enough that they do not look like the same bug." + }, + { + "left": "issue:5565", + "right": "issue:6154", + "accept": false, + "reason": "Generic script failure versus an M2 Mac-specific LoRA training problem." + }, + { + "left": "issue:11922", + "right": "issue:12459", + "accept": false, + "reason": "Different model families and symptoms; one is FLUX.1-Fill-dev running problem, the other is a Kontext-dev batch-throughput issue." + }, + { + "left": "issue:4094", + "right": "issue:4405", + "accept": false, + "reason": "These are separate feature requests for different training scripts." + }, + { + "left": "issue:10848", + "right": "issue:5798", + "accept": false, + "reason": "Both are unexpected-argument errors, but they hit different functions/pipelines and do not share the same code path." + }, + { + "left": "issue:12047", + "right": "issue:12535", + "accept": false, + "reason": "Both are Wan2.2 LoRA-related, but one is fusing a Lightx2v LoRA on GGUF and the other is failing to load a Lightning LoRA." + }, + { + "left": "issue:10392", + "right": "issue:9420", + "accept": false, + "reason": "Different Flux+LoRA loading problems with different model variants and failure points." + }, + { + "left": "issue:5016", + "right": "issue:6297", + "accept": false, + "reason": "Text-encoder training is a different issue from a general complaint that the current script cannot train on photos." + }, + { + "left": "issue:1540", + "right": "issue:1541", + "accept": true, + "reason": "Both report the same DreamBooth MODEL_TYPE / CKPT loading failure; 1541 is the clearer root-cause description." + }, + { + "left": "issue:4368", + "right": "issue:6362", + "accept": false, + "reason": "One is training-time OOM for SDXL LoRA; the other is dataset mapping OOM." + }, + { + "left": "issue:6117", + "right": "issue:6232", + "accept": false, + "reason": "Both are scheduler add_noise bugs, but they involve different scheduler classes and different failing settings." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only clearly supported duplicate is the DreamBooth MODEL_TYPE / CKPT-loading pair (1540/1541), and the other soft-edge verdicts appropriately avoid overclaiming equivalence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 42, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10833", + "issue:11023", + "issue:11117", + "issue:11127", + "issue:11133", + "issue:11352", + "issue:11432", + "issue:12216", + "issue:12458", + "issue:13043", + "issue:13105", + "issue:13258", + "issue:1552", + "issue:175", + "issue:2549", + "issue:2891", + "issue:3382", + "issue:3451", + "issue:3803", + "issue:4206", + "issue:4520", + "issue:4724", + "issue:4872", + "issue:4925", + "issue:5075", + "issue:5254", + "issue:5455", + "issue:5769", + "issue:5805", + "issue:5966", + "issue:6159", + "issue:6278", + "issue:6695", + "issue:6832", + "issue:7466", + "issue:7658", + "issue:7666", + "issue:771", + "issue:7799", + "issue:8405", + "issue:8414", + "issue:8667", + "issue:9112", + "issue:9237", + "issue:938", + "issue:9500", + "issue:9667", + "issue:9953" + ], + "soft_pairs": [ + "issue:5966|issue:7658", + "issue:11352|issue:9953", + "issue:6832|issue:7799", + "issue:3803|issue:6278", + "issue:4206|issue:4925", + "issue:11127|issue:9667", + "issue:12216|issue:12458", + "issue:3382|issue:4925", + "issue:8405|issue:9112", + "issue:8414|issue:938", + "issue:8667|issue:9112", + "issue:4872|issue:4925", + "issue:175|issue:5769", + "issue:13105|issue:13258", + "issue:1552|issue:771", + "issue:9237|issue:9500", + "issue:2891|issue:6159", + "issue:2549|issue:4520", + "issue:3382|issue:3451", + "issue:11117|issue:5805", + "issue:5254|issue:7466", + "issue:11432|issue:4724", + "issue:10833|issue:5075", + "issue:5455|issue:6695", + "issue:11023|issue:7666", + "issue:11133|issue:13043" + ], + "prepared_review_unit_hash": "7264d9dac5b6d610f15a81d5433473f2f4196c1fd2d5ad568e1e98e8ac904f67", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 34814, + "estimated_input_tokens": 8704, + "estimated_eval_tokens": 17664 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a broad grab-bag of diffusion/pipeline bug reports, mostly about model compatibility, dtype/device mismatches, quantization, and training/inference failures. The soft matches are mostly superficial (same subsystem or similar error wording) rather than true duplicates.", + "confidence": 0.42, + "canonical_issue_reason": "issue:9953 is the best representative of the cluster\u2019s most recurring concrete theme: pipeline failures when moving quantized components across devices. It is specific enough to anchor the cluster, while still broader than the more model-specific reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9953 is the most generally useful issue here because it describes a common runtime/device-transfer bug pattern that plausibly spans multiple similar reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5966", + "right": "issue:7658", + "accept": false, + "reason": "Related ControlNet topic, but one is SD 2.1 support and the other is SDXL multiple-conditioning; not the same bug." + }, + { + "left": "issue:11352", + "right": "issue:9953", + "accept": false, + "reason": "Both involve quantized pipelines and device moves, but one is missing a CPU warning and the other is a CUDA error; different failures." + }, + { + "left": "issue:6832", + "right": "issue:7799", + "accept": false, + "reason": "Both are IP-Adapter matmul/shape problems, but they hit different pipelines and likely different code paths." + }, + { + "left": "issue:3803", + "right": "issue:6278", + "accept": false, + "reason": "Both mention multi-GPU training, but the scripts and failure modes are different; not a duplicate bug." + }, + { + "left": "issue:4206", + "right": "issue:4925", + "accept": false, + "reason": "Both concern SDXL ControlNet training, but one is a general main-branch breakage and the other is a distributed OOM issue." + }, + { + "left": "issue:11127", + "right": "issue:9667", + "accept": false, + "reason": "Both are Flux loading issues, but one is a single-file compatibility problem and the other is fp8/local transformer loading." + }, + { + "left": "issue:12216", + "right": "issue:12458", + "accept": false, + "reason": "One reports inferior Qwen-Image-Edit results, the other is a batch inference request/issue; different concerns." + }, + { + "left": "issue:3382", + "right": "issue:4925", + "accept": false, + "reason": "Both are training OOM reports, but they are from different training setups and not the same underlying defect." + }, + { + "left": "issue:8405", + "right": "issue:9112", + "accept": false, + "reason": "Both mention tensor-size errors, but on different models and different inference paths." + }, + { + "left": "issue:8414", + "right": "issue:938", + "accept": false, + "reason": "Completely different models and domains: text-to-audio vs text-to-motion." + }, + { + "left": "issue:8667", + "right": "issue:9112", + "accept": false, + "reason": "Similar size-mismatch wording, but the model families and likely root causes differ." + }, + { + "left": "issue:4872", + "right": "issue:4925", + "accept": false, + "reason": "Both are OOM complaints, but one is SDXL finetuning on T4 and the other is distributed ControlNet training." + }, + { + "left": "issue:175", + "right": "issue:5769", + "accept": false, + "reason": "Both ask about training correctness, but one is about seeding and the other about mixed precision; not the same issue." + }, + { + "left": "issue:13105", + "right": "issue:13258", + "accept": false, + "reason": "Different Wan model versions and different symptoms: noisy output vs missing image-to-video behavior." + }, + { + "left": "issue:1552", + "right": "issue:771", + "accept": false, + "reason": "One is attention slicing on SD2, the other is BF16 in dreambooth; unrelated precision/memory issues." + }, + { + "left": "issue:9237", + "right": "issue:9500", + "accept": false, + "reason": "Both are Flux dreambooth-related, but one is a training failure and the other is a checkpoint-saving failure." + }, + { + "left": "issue:2891", + "right": "issue:6159", + "accept": false, + "reason": "Different pipelines and different API gaps; one is embeds support, the other is denoising_end support." + }, + { + "left": "issue:2549", + "right": "issue:4520", + "accept": false, + "reason": "Different bugs in different features: SD 2.0 UNet loading vs LoRA scale with attention slicing." + }, + { + "left": "issue:3382", + "right": "issue:3451", + "accept": false, + "reason": "Both are training-memory reports, but one is multi-GPU OOM and the other is fp16 memory regression." + }, + { + "left": "issue:11117", + "right": "issue:5805", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different pipelines and environments." + }, + { + "left": "issue:5254", + "right": "issue:7466", + "accept": false, + "reason": "One is a feature request for multi-ControlNet support, the other is a torch.compile failure; not the same change or bug." + }, + { + "left": "issue:11432", + "right": "issue:4724", + "accept": false, + "reason": "A from_pretrained dtype-auto issue and an SDXL img2img behavior problem are not the same underlying defect." + }, + { + "left": "issue:10833", + "right": "issue:5075", + "accept": false, + "reason": "Both are matmul shape errors, but they happen in different pipelines with different inputs and likely different root causes." + }, + { + "left": "issue:5455", + "right": "issue:6695", + "accept": false, + "reason": "Same generic subprocess error string, but the surrounding workflows are too different to call them duplicates." + }, + { + "left": "issue:11023", + "right": "issue:7666", + "accept": false, + "reason": "Both mention lpw_stable_diffusion_xl, but one reports blurry output and the other single-file loading failure." + }, + { + "left": "issue:11133", + "right": "issue:13043", + "accept": false, + "reason": "Unrelated models and failures: CogVideoX pipeline bug vs Blackwell kernel-image compatibility issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 43, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10040", + "issue:1082", + "issue:10937", + "issue:11133", + "issue:11285", + "issue:12108", + "issue:12195", + "issue:12235", + "issue:12666", + "issue:13351", + "issue:1875", + "issue:2234", + "issue:2296", + "issue:2578", + "issue:2904", + "issue:3382", + "issue:3481", + "issue:3497", + "issue:4052", + "issue:4316", + "issue:4356", + "issue:4450", + "issue:4542", + "issue:4872", + "issue:5006", + "issue:539", + "issue:5455", + "issue:5607", + "issue:6239", + "issue:6280", + "issue:6761", + "issue:7133", + "issue:7146", + "issue:7147", + "issue:7178", + "issue:720", + "issue:7263", + "issue:7378", + "issue:758", + "issue:7644", + "issue:7735", + "issue:8613", + "issue:8667", + "issue:8817", + "issue:8939", + "issue:9112", + "issue:9187", + "issue:9202" + ], + "soft_pairs": [ + "issue:11285|issue:8939", + "issue:5455|issue:7133", + "issue:1082|issue:7146", + "issue:5006|issue:7146", + "issue:3382|issue:4872", + "issue:2578|issue:7178", + "issue:12235|issue:13351", + "issue:4450|issue:7735", + "issue:4542|issue:7378", + "issue:539|issue:6761", + "issue:7644|issue:8667", + "issue:8817|issue:9202", + "issue:10040|issue:3481", + "issue:1875|issue:720", + "issue:11133|issue:12108", + "issue:4052|issue:6280", + "issue:12666|issue:4356", + "issue:5607|issue:9112", + "issue:10937|issue:6239", + "issue:2904|issue:3497", + "issue:12195|issue:4356", + "issue:2234|issue:2296", + "issue:4316|issue:8613", + "issue:7263|issue:9187", + "issue:7147|issue:758" + ], + "prepared_review_unit_hash": "ec26074868e2c6559578d3129c346308007255b7cf092be68bf3b5a42282c67d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34966, + "estimated_input_tokens": 8742, + "estimated_eval_tokens": 17740 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are unrelated API, training, memory, or model-specific bugs that only share superficial wording. There is no single coherent duplicate target across the whole set.", + "confidence": 0.94, + "canonical_issue_reason": "No clear canonical issue: the items span many unrelated failure modes and subsystems, so no single issue cleanly represents the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "No strong best issue candidate for the whole cluster; the only near-duplicate-looking pair is a narrow API-missing report, but it does not unify the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11285", + "right": "issue:8939", + "accept": false, + "reason": "Different problems: conversion/value errors vs tokenizer loading failure." + }, + { + "left": "issue:5455", + "right": "issue:7133", + "accept": false, + "reason": "Dreambooth notebook subprocess failure is not the same as SDXL checkpoint-resume size mismatch." + }, + { + "left": "issue:1082", + "right": "issue:7146", + "accept": false, + "reason": "CUDA cublas execution failure differs from a generic SDXL OOM during finetuning." + }, + { + "left": "issue:5006", + "right": "issue:7146", + "accept": false, + "reason": "Both mention OOM, but one is a BF16 end-of-training spike and the other is a broader training OOM; not the same concrete bug." + }, + { + "left": "issue:3382", + "right": "issue:4872", + "accept": false, + "reason": "Different training contexts and symptoms: multi-GPU OOM vs T4 Colab SDXL finetuning OOM." + }, + { + "left": "issue:2578", + "right": "issue:7178", + "accept": false, + "reason": "Editable install failure is unrelated to export_to_video malfunction." + }, + { + "left": "issue:12235", + "right": "issue:13351", + "accept": false, + "reason": "Related feature area, but different request scopes and model variants; not one concrete change." + }, + { + "left": "issue:4450", + "right": "issue:7735", + "accept": false, + "reason": "Textual inversion multi-text support is unrelated to a missing caption_column field." + }, + { + "left": "issue:4542", + "right": "issue:7378", + "accept": false, + "reason": "One is a multi-image output limit, the other is memory use in latent decoding." + }, + { + "left": "issue:539", + "right": "issue:6761", + "accept": false, + "reason": "Different unexpected-argument errors on unrelated APIs." + }, + { + "left": "issue:7644", + "right": "issue:8667", + "accept": false, + "reason": "Both involve non-power-of-two images, but different model/pipeline code paths and failure modes." + }, + { + "left": "issue:8817", + "right": "issue:9202", + "accept": false, + "reason": "Different IP-Adapter variants and pipeline behavior; not the same bug." + }, + { + "left": "issue:10040", + "right": "issue:3481", + "accept": false, + "reason": "Grey-image generation is unrelated to DDIM determinism across inference steps." + }, + { + "left": "issue:1875", + "right": "issue:720", + "accept": false, + "reason": "Both are dtype mismatch errors, but in different pipelines and code paths." + }, + { + "left": "issue:11133", + "right": "issue:12108", + "accept": false, + "reason": "CogVideoX image-to-video bug is unrelated to Qwen/Chroma flow-matching scheduler breakage." + }, + { + "left": "issue:4052", + "right": "issue:6280", + "accept": true, + "reason": "Both report the same missing `from_single_file` pipeline API; the titles differ only in how specifically the pipeline is named." + }, + { + "left": "issue:12666", + "right": "issue:4356", + "accept": false, + "reason": "Kandinsky pipeline availability is unrelated to a missing dummy module import." + }, + { + "left": "issue:5607", + "right": "issue:9112", + "accept": false, + "reason": "Tensor size mismatch here is a generic shape issue, but the pipelines and reported failure contexts differ." + }, + { + "left": "issue:10937", + "right": "issue:6239", + "accept": false, + "reason": "Both touch VAE/torch FX, but one is a compile error on vae.encode and the other is a torch FX support request for latent sampling." + }, + { + "left": "issue:2904", + "right": "issue:3497", + "accept": false, + "reason": "AudioLDM scheduler NoneType error is unrelated to a missing from_ckpt attribute." + }, + { + "left": "issue:12195", + "right": "issue:4356", + "accept": false, + "reason": "A PyTorch version incompatibility around `torch.library.custom_op` is unrelated to the missing dummy module import." + }, + { + "left": "issue:2234", + "right": "issue:2296", + "accept": false, + "reason": "Both mention xformers attention, but one is training failure and the other is T4 inference incompatibility." + }, + { + "left": "issue:4316", + "right": "issue:8613", + "accept": false, + "reason": "Validation-option failure during training is unrelated to an OOM triggered by caption_column usage." + }, + { + "left": "issue:7263", + "right": "issue:9187", + "accept": false, + "reason": "Different IP-Adapter failure modes: attention slicing vs SDXL/image_encoder_folder=None after Kolors update." + }, + { + "left": "issue:7147", + "right": "issue:758", + "accept": false, + "reason": "StableCascade import failure is unrelated to the urllib3 PROTOCOL_TLS import error." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall: the cluster is clearly heterogeneous, the no-canonical-issue conclusion is conservative, and the one accepted soft edge (issue:4052 \u2194 issue:6280) is a reasonable same-API-missing match. The false soft-edge rejections are also cautious and do not overstate similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 44, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10381", + "issue:10428", + "issue:10848", + "issue:10937", + "issue:1100", + "issue:11033", + "issue:11149", + "issue:11464", + "issue:11651", + "issue:1214", + "issue:1249", + "issue:13137", + "issue:13446", + "issue:13484", + "issue:1434", + "issue:1614", + "issue:2210", + "issue:2780", + "issue:2856", + "issue:3045", + "issue:3480", + "issue:4194", + "issue:4368", + "issue:4453", + "issue:4534", + "issue:4724", + "issue:4835", + "issue:4874", + "issue:4969", + "issue:5006", + "issue:5374", + "issue:5846", + "issue:6278", + "issue:6524", + "issue:6584", + "issue:6761", + "issue:6794", + "issue:6879", + "issue:7033", + "issue:7238", + "issue:7660", + "issue:8236", + "issue:8708", + "issue:8887", + "issue:9091", + "issue:9497", + "issue:9660", + "issue:9841" + ], + "soft_pairs": [ + "issue:1100|issue:1249", + "issue:10848|issue:11149", + "issue:10937|issue:9091", + "issue:6879|issue:9497", + "issue:10381|issue:10428", + "issue:1614|issue:4724", + "issue:11033|issue:1214", + "issue:4368|issue:5006", + "issue:1214|issue:7660", + "issue:13137|issue:13484", + "issue:13446|issue:9660", + "issue:2210|issue:3480", + "issue:2856|issue:6584", + "issue:3045|issue:4453", + "issue:4194|issue:6524", + "issue:4969|issue:6794", + "issue:5374|issue:6761", + "issue:7033|issue:7238", + "issue:8708|issue:8887", + "issue:11464|issue:11651", + "issue:4835|issue:4874", + "issue:2780|issue:5846", + "issue:1434|issue:4874", + "issue:6278|issue:8236", + "issue:4534|issue:9841" + ], + "prepared_review_unit_hash": "84b65760e8007d4542dda8ad5c8523265c7b1113e30aa07424cefa405b7b8337", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34268, + "estimated_input_tokens": 8567, + "estimated_eval_tokens": 17390 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is heterogeneous; most soft pairs share only broad subsystem vocabulary and do not look like the same underlying bug. The closest mini-cluster is Flux/LoRA loading, but even there the reports appear to be different variants rather than duplicates.", + "confidence": 0.68, + "canonical_issue_reason": "No single issue cleanly represents the whole set. If a representative is needed, issue 11464 is the best fit for the Flux/LoRA-loading subset because it is the broadest FluxControlPipeline report and is nearest to 11651.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11464 is the most suitable canonical issue among the provided items: it is the clearest, broadest Flux/LoRA loading failure and the only one with an obvious nearby sibling in the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1100", + "right": "issue:1249", + "accept": false, + "reason": "Both mention device-related attribute errors, but one is a CLIPTextTransformer attribute lookup and the other is a numpy.ndarray/device issue in OnnxPipeline. Different code paths and root causes." + }, + { + "left": "issue:10848", + "right": "issue:11149", + "accept": false, + "reason": "Both are unexpected keyword-argument failures, but they occur in different loaders: load_model_dict_into_meta(device) vs BnB4BitDiffusersQuantizer.create_quantized_param(dtype). Not the same bug." + }, + { + "left": "issue:10937", + "right": "issue:9091", + "accept": false, + "reason": "Both involve torch.compile and VAE components, but one is an encode-time error while the other is decoder support for a specific VAE. Different failure modes." + }, + { + "left": "issue:6879", + "right": "issue:9497", + "accept": false, + "reason": "Both are dtype/matmul runtime errors, but one is Custom Diffusion training and the other is Flux Dreambooth training. Different models and likely different code paths." + }, + { + "left": "issue:10381", + "right": "issue:10428", + "accept": false, + "reason": "Similar Flux inference wording, but one is quantization + LoRA on zero-GPU spaces and the other is Ascend NPU inference. Different backend/hardware-specific problems." + }, + { + "left": "issue:1614", + "right": "issue:4724", + "accept": false, + "reason": "Both are Stable Diffusion pipeline failures, but black output with autocast is unrelated to an img2img checkpoint mismatch. Not the same underlying bug." + }, + { + "left": "issue:11033", + "right": "issue:1214", + "accept": false, + "reason": "One is SD1.5 from_single_file loading for UNet, the other is CLIPTextModel half-precision behavior. Different components and symptoms." + }, + { + "left": "issue:4368", + "right": "issue:5006", + "accept": false, + "reason": "Both are training memory issues, but one is SDXL LoRA OOM and the other is end-of-training BF16 OOM. Too broad to treat as the same bug." + }, + { + "left": "issue:1214", + "right": "issue:7660", + "accept": false, + "reason": "Half-precision CLIPTextModel failure and a trace example failure are unrelated." + }, + { + "left": "issue:13137", + "right": "issue:13484", + "accept": false, + "reason": "Both are Flux 2 LoRA loading problems, but one is LoKr weight loading and the other is FAL LoRA loading. Close topic, but not clearly the same code-path bug." + }, + { + "left": "issue:13446", + "right": "issue:9660", + "accept": false, + "reason": "Custom pipeline trust checking and from_pretrained eval-mode handling are unrelated issues in different parts of the API." + }, + { + "left": "issue:2210", + "right": "issue:3480", + "accept": false, + "reason": "Both concern text-encoder LoRA, but one is a training-capability request and the other is a scale-parameter feature request. Same area, not the same change." + }, + { + "left": "issue:2856", + "right": "issue:6584", + "accept": false, + "reason": "Generic SD-to-Diffusers conversion versus a specific DreamShaper v8 inpaint conversion failure. Not the same underlying conversion bug." + }, + { + "left": "issue:3045", + "right": "issue:4453", + "accept": false, + "reason": "These are opposite directions and different ecosystems: converting LoRA safetensors to Diffusers vs converting Diffusers SDXL LoRA to AUTO1111-compatible safetensors." + }, + { + "left": "issue:4194", + "right": "issue:6524", + "accept": false, + "reason": "VAE not working in from_single_file and LoRA failing on second load are different subsystems." + }, + { + "left": "issue:4969", + "right": "issue:6794", + "accept": false, + "reason": "Importing CrossAttention and importing cast_training_params are unrelated import errors." + }, + { + "left": "issue:5374", + "right": "issue:6761", + "accept": false, + "reason": "Both are unexpected keyword-argument errors, but they affect different methods with different parameters and components." + }, + { + "left": "issue:7033", + "right": "issue:7238", + "accept": false, + "reason": "Both mention IP-Adapter masking, but one is warning-log noise and the other is a masking behavior issue. Not enough evidence of the same bug." + }, + { + "left": "issue:8708", + "right": "issue:8887", + "accept": false, + "reason": "Both are SD3 training/VAE-related, but one is a missing config field in dreambooth examples and the other is a wrong VAE encode path. Related, but not the same defect." + }, + { + "left": "issue:11464", + "right": "issue:11651", + "accept": false, + "reason": "Both are FluxControl LoRA loading failures, but they refer to different LoRA variants (Canny-dev vs flux-fill) and likely separate loading cases. Too distinct to merge." + }, + { + "left": "issue:4835", + "right": "issue:4874", + "accept": false, + "reason": "Different pipeline bugs: ControlNet img2img with MultiControlNetModel versus UpscalePipeline low_res_scheduler when loading locally." + }, + { + "left": "issue:2780", + "right": "issue:5846", + "accept": false, + "reason": "Both are LoRA training issues, but one is FP16 gradient unscale handling and the other is resume_from_checkpoint behavior in a different script/version." + }, + { + "left": "issue:1434", + "right": "issue:4874", + "accept": false, + "reason": "Memory blowout during upscaling is unrelated to the low_res_scheduler local-loading bug." + }, + { + "left": "issue:6278", + "right": "issue:8236", + "accept": false, + "reason": "Both involve multi-GPU examples, but one cannot run and the other has an incorrect scheduler step count. Different training bugs." + }, + { + "left": "issue:4534", + "right": "issue:9841", + "accept": false, + "reason": "SDXL dreambooth validation crash on mps and SDXL LoRA type error with use_dora=False are different failures in different execution paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 45, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10055", + "issue:10266", + "issue:10421", + "issue:10738", + "issue:10823", + "issue:10952", + "issue:11209", + "issue:11224", + "issue:11285", + "issue:11702", + "issue:1214", + "issue:13162", + "issue:13484", + "issue:1742", + "issue:1931", + "issue:1979", + "issue:234", + "issue:2856", + "issue:3382", + "issue:3481", + "issue:3582", + "issue:3619", + "issue:3774", + "issue:3910", + "issue:4073", + "issue:4086", + "issue:4340", + "issue:4468", + "issue:4476", + "issue:4736", + "issue:5259", + "issue:5619", + "issue:6130", + "issue:6176", + "issue:6226", + "issue:6536", + "issue:6761", + "issue:7056", + "issue:7178", + "issue:7310", + "issue:7483", + "issue:7708", + "issue:771", + "issue:8514", + "issue:9179", + "issue:9180", + "issue:9962" + ], + "soft_pairs": [ + "issue:3582|issue:5619", + "issue:3619|issue:771", + "issue:1979|issue:2856", + "issue:10055|issue:7310", + "issue:4340|issue:6176", + "issue:11285|issue:6226", + "issue:4073|issue:9180", + "issue:13162|issue:13484", + "issue:5259|issue:7708", + "issue:1214|issue:7178", + "issue:10421|issue:4468", + "issue:10952|issue:3910", + "issue:1742|issue:3481", + "issue:4086|issue:7483", + "issue:4476|issue:9179", + "issue:7056|issue:8514", + "issue:3382|issue:3774", + "issue:10823|issue:6536", + "issue:6761|issue:9962", + "issue:11209|issue:11702", + "issue:11209|issue:6130", + "issue:11224|issue:11702", + "issue:11224|issue:6130", + "issue:10266|issue:10738", + "issue:3382|issue:4736", + "issue:1931|issue:234" + ], + "prepared_review_unit_hash": "eb8fbc1e0e9e88ab8ccfb0df70cb0dd9818349af7b968c3838bb7833831fe5fe", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 33814, + "estimated_input_tokens": 8454, + "estimated_eval_tokens": 17164 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of Diffusers issues around training, LoRA loading, conversion, and a few scheduler/runtime bugs. The only clear duplicate pair is the repeated generic \"Error in loading Lora\" reports; the rest are thematically similar but not the same underlying defect.", + "confidence": 0.66, + "canonical_issue_reason": "issue:11224 is the clearest representative of the repeated generic LoRA-loading failure reports, with the same broad title as issue:11209 and slightly richer discussion activity.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11224 is the best cluster anchor because it is a broad, user-facing LoRA loading failure that matches multiple nearby reports, while still being specific enough to represent the duplicate family.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3582", + "right": "issue:5619", + "accept": false, + "reason": "Different problems: ControlNet inpaint strength handling vs LCM img2img AutoencodeTiny encode behavior." + }, + { + "left": "issue:3619", + "right": "issue:771", + "accept": false, + "reason": "Unrelated runtime areas: Text2Video-Zero CrossFrameAttention vs BF16 dreambooth precision support." + }, + { + "left": "issue:1979", + "right": "issue:2856", + "accept": false, + "reason": "Both are conversion-related, but one is a general SD-to-diffusers conversion failure and the other is a specific SD conversion script issue; not clearly the same bug." + }, + { + "left": "issue:10055", + "right": "issue:7310", + "accept": false, + "reason": "Both mention training scripts, but one is SD3 ControlNet training and the other is SDXL LoRA defaults; different code paths and failure modes." + }, + { + "left": "issue:4340", + "right": "issue:6176", + "accept": false, + "reason": "Both concern LoRA training/loading, but one is SDXL LoRA training with text encoder and the other is loading LoRA weights after DreamBooth LoRA training; not the same defect." + }, + { + "left": "issue:11285", + "right": "issue:6226", + "accept": false, + "reason": "One is about conversion value errors between original Stable Diffusion and diffusers, the other about loading errors caused by a conversion script; related area but not the same underlying issue." + }, + { + "left": "issue:4073", + "right": "issue:9180", + "accept": false, + "reason": "Different failure classes: DeepFloyd IF safety checker attribute error vs pipeline missing _execution_device." + }, + { + "left": "issue:13162", + "right": "issue:13484", + "accept": false, + "reason": "Both are Flux LoRA support/load issues, but they appear to target different models and loading scenarios; too broad to treat as the same concrete bug." + }, + { + "left": "issue:5259", + "right": "issue:7708", + "accept": false, + "reason": "Both are missing-file load errors, but one is generic model_index.json missing and the other is a tokenizer/config.json path issue for SDXL." + }, + { + "left": "issue:1214", + "right": "issue:7178", + "accept": false, + "reason": "Half-precision CLIPTextModel failure is unrelated to export_to_video behavior." + }, + { + "left": "issue:10421", + "right": "issue:4468", + "accept": false, + "reason": "Both involve multi-GPU behavior, but one is a CPU memory leak during pipeline migration and the other is differing inference performance on multi-GPU." + }, + { + "left": "issue:10952", + "right": "issue:3910", + "accept": false, + "reason": "Both are LoRA-related, but one is a specific adapter-name mismatch in SDXL long-prompt weighting and the other is a generic \"certain LoRA will not load\" report." + }, + { + "left": "issue:1742", + "right": "issue:3481", + "accept": false, + "reason": "Performance variability vs output-dissimilarity across inference steps are different issues despite both mentioning prompt/inference behavior." + }, + { + "left": "issue:4086", + "right": "issue:7483", + "accept": false, + "reason": "Both are about ControlNet/SDXL training or conversion, but one is SDXL ControlNet training not working and the other is LoRA-to-Diffusers conversion for SDXL 1.0 base models." + }, + { + "left": "issue:4476", + "right": "issue:9179", + "accept": false, + "reason": "Both mention SDXL+ControlNet, but one is fill50k results not matching expectations and the other is training results generally not making sense; too broad to merge." + }, + { + "left": "issue:7056", + "right": "issue:8514", + "accept": false, + "reason": "Both are training errors, but one is SDXL text-to-image LoRA with train_text_encoder and the other is EMAModel causing InstructP2P parallel fine-tuning failure." + }, + { + "left": "issue:3382", + "right": "issue:3774", + "accept": false, + "reason": "Multi-GPU OOM in text-to-image training and ZeRO3 memory not saving are related only at a high level; different training setups and symptoms." + }, + { + "left": "issue:10823", + "right": "issue:6536", + "accept": false, + "reason": "Different problems: missing inference files from fine-tuning script vs UNet3DConditionModel fine-tuning out-of-memory." + }, + { + "left": "issue:6761", + "right": "issue:9962", + "accept": false, + "reason": "Different API breakages: unexpected keyword on VaeImageProcessor.preprocess vs unexpected keyword use_cuda_graph." + }, + { + "left": "issue:11209", + "right": "issue:11702", + "accept": false, + "reason": "Both are LoRA loading issues, but 11702 is an AccVid-specific key error with diffusers; not the same as the generic loading failure." + }, + { + "left": "issue:11209", + "right": "issue:6130", + "accept": false, + "reason": "Same broad theme, but 6130 is an older generic loading issue with no evidence it is the exact same adapter/key failure as 11209." + }, + { + "left": "issue:11224", + "right": "issue:11702", + "accept": false, + "reason": "Generic LoRA loading failure vs AccVid LoRA key error; different concrete failure modes." + }, + { + "left": "issue:11224", + "right": "issue:6130", + "accept": false, + "reason": "Both are generic LoRA-loading complaints, but the evidence is too weak to say they are the same underlying bug." + }, + { + "left": "issue:10266", + "right": "issue:10738", + "accept": false, + "reason": "Both are out-of-bounds scheduler issues, but one is UniPC with FlowMatch and the other is a scheduler sigma index error; not enough to identify a single bug." + }, + { + "left": "issue:3382", + "right": "issue:4736", + "accept": false, + "reason": "Different training failures: multi-GPU OOM in text-to-image vs SDXL LoRA OOM plus invalid cast." + }, + { + "left": "issue:1931", + "right": "issue:234", + "accept": false, + "reason": "Both concern half precision, but one is a Stable Diffusion 2 dtype mismatch and the other is a generic pipeline half-precision failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 46, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10040", + "issue:10169", + "issue:10436", + "issue:1070", + "issue:10860", + "issue:11160", + "issue:11247", + "issue:11396", + "issue:12108", + "issue:12141", + "issue:12359", + "issue:1251", + "issue:12771", + "issue:12896", + "issue:13300", + "issue:1565", + "issue:1988", + "issue:2495", + "issue:2578", + "issue:2991", + "issue:3191", + "issue:3235", + "issue:3812", + "issue:4551", + "issue:4619", + "issue:4768", + "issue:4885", + "issue:4911", + "issue:4936", + "issue:5226", + "issue:579", + "issue:5884", + "issue:5886", + "issue:6082", + "issue:7057", + "issue:7228", + "issue:7666", + "issue:7854", + "issue:7974", + "issue:817", + "issue:8668", + "issue:8833", + "issue:8922", + "issue:9343", + "issue:9500", + "issue:9548", + "issue:9911", + "issue:9970" + ], + "soft_pairs": [ + "issue:11247|issue:9911", + "issue:11160|issue:12896", + "issue:4911|issue:7228", + "issue:1988|issue:4619", + "issue:2578|issue:5226", + "issue:10169|issue:3812", + "issue:4768|issue:5884", + "issue:12141|issue:7057", + "issue:7057|issue:7666", + "issue:13300|issue:8668", + "issue:9500|issue:9970", + "issue:2991|issue:817", + "issue:12771|issue:9343", + "issue:3235|issue:4885", + "issue:6082|issue:7854", + "issue:12108|issue:12359", + "issue:10040|issue:8922", + "issue:3191|issue:4936", + "issue:1565|issue:4551", + "issue:9500|issue:9548", + "issue:579|issue:8833", + "issue:10436|issue:1070", + "issue:1251|issue:2495", + "issue:10860|issue:11396", + "issue:5886|issue:7974" + ], + "prepared_review_unit_hash": "b7e2aed33fba7a1bfe6a8fadaef0b31b21ee0ed2659b2814cf8e43f56685ac5f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34435, + "estimated_input_tokens": 8609, + "estimated_eval_tokens": 17474 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very loose: most items are distinct model-loading, training, or GPU/device-specific bugs rather than true duplicates. No soft pair looks safe to merge as the same underlying issue.", + "confidence": 0.37, + "canonical_issue_reason": "issue:9500 is the most substantial and actionable report in the set, with the richest reproduction details and highest discussion activity, so it is the best representative anchor for this heterogeneous cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9500 is the strongest standalone issue for triage: detailed, active, and clearly described, even though the surrounding items are not true duplicates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11247", + "right": "issue:9911", + "accept": false, + "reason": "Both concern Flux/controlnet multi-device or multi-controlnet behavior, but they describe different failure modes and likely different code paths." + }, + { + "left": "issue:11160", + "right": "issue:12896", + "accept": false, + "reason": "Same WAN i2v family, but one is inconsistent inference output and the other is a CUDA-vs-MPS execution failure; too different to treat as one bug." + }, + { + "left": "issue:4911", + "right": "issue:7228", + "accept": false, + "reason": "Both are loading-related, but one is about missing custom text_encoder in from_single_file and the other is a traced-model dtype attribute error." + }, + { + "left": "issue:1988", + "right": "issue:4619", + "accept": false, + "reason": "Both mention fp16/bf16 type mismatch during training, but they are different training scripts and not clearly the same concrete bug." + }, + { + "left": "issue:2578", + "right": "issue:5226", + "accept": false, + "reason": "Editable install failure and LoRA loading failure are unrelated problems." + }, + { + "left": "issue:10169", + "right": "issue:3812", + "accept": false, + "reason": "Missing config.json and missing controlnet key in config are different config-loading issues." + }, + { + "left": "issue:4768", + "right": "issue:5884", + "accept": false, + "reason": "One is about checkpointing multiple T2I adapters; the other is a feature request for broader IP-Adapter support." + }, + { + "left": "issue:12141", + "right": "issue:7057", + "accept": false, + "reason": "VAE dtype compatibility and checkpoint/model loading failure are not the same underlying bug." + }, + { + "left": "issue:7057", + "right": "issue:7666", + "accept": false, + "reason": "Different loading paths and different errors; no evidence they share the same root cause." + }, + { + "left": "issue:13300", + "right": "issue:8668", + "accept": false, + "reason": "CPU generation failure for SD3.5/FLUX is unrelated to the image-size alignment bug in SD3 img2img." + }, + { + "left": "issue:9500", + "right": "issue:9970", + "accept": false, + "reason": "Both are DreamBooth training failures, but one is Flux checkpoint saving and the other is SDXL LoRA script failure; too broad to merge." + }, + { + "left": "issue:2991", + "right": "issue:817", + "accept": false, + "reason": "Both are mixed-precision example failures, but they affect different examples and are not clearly the same code-path defect." + }, + { + "left": "issue:12771", + "right": "issue:9343", + "accept": false, + "reason": "Scheduler out-of-bounds with Karras sigmas is unrelated to FLUX loading with mismatched sizes." + }, + { + "left": "issue:3235", + "right": "issue:4885", + "accept": false, + "reason": "Custom Diffusion weights with biases failing to load is not the same as SDXL ControlNet Img2Img LoRA loading." + }, + { + "left": "issue:6082", + "right": "issue:7854", + "accept": false, + "reason": "Both are half/float mismatch reports, but they are separate reports without enough evidence of the same concrete failure." + }, + { + "left": "issue:12108", + "right": "issue:12359", + "accept": false, + "reason": "Pipeline breakage with flow-matching schedulers is not the same as the Chroma documentation issue about guidance_scale." + }, + { + "left": "issue:10040", + "right": "issue:8922", + "accept": false, + "reason": "Grey images and black images are similar symptoms, but they are not enough to conclude the same root cause." + }, + { + "left": "issue:3191", + "right": "issue:4936", + "accept": false, + "reason": "Dreambooth+LoRA being broken and resume in Dreambooth LoRA training are different training problems." + }, + { + "left": "issue:1565", + "right": "issue:4551", + "accept": false, + "reason": "GPU selection/OOM on SD2 is unrelated to SDXL Dreambooth crashing." + }, + { + "left": "issue:9500", + "right": "issue:9548", + "accept": false, + "reason": "Both involve Flux DreamBooth/LoRA, but one is checkpoint saving failure and the other is a broader 'still issue' follow-up; not the same concrete bug." + }, + { + "left": "issue:579", + "right": "issue:8833", + "accept": false, + "reason": "Both mention missing config.json, but one is textual inversion and the other is a model repo/config packaging issue." + }, + { + "left": "issue:10436", + "right": "issue:1070", + "accept": false, + "reason": "Inpainting model not loading and inpainting faces bug are different issues." + }, + { + "left": "issue:1251", + "right": "issue:2495", + "accept": false, + "reason": "Black output on MPS for RePaint and black output on MPS for upscale pipeline may look similar, but they affect different pipelines and code paths." + }, + { + "left": "issue:10860", + "right": "issue:11396", + "accept": false, + "reason": "Both involve LoRA interoperability with ComfyUI, but one is about using a trained LoRA and the other is about converting a HiDream LoRA format." + }, + { + "left": "issue:5886", + "right": "issue:7974", + "accept": false, + "reason": "Face model support request and an inpaint bug with IP-Adapter face+style are related only broadly; they are not the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 47, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10180", + "issue:10223", + "issue:10436", + "issue:11062", + "issue:11214", + "issue:11362", + "issue:11374", + "issue:11422", + "issue:11702", + "issue:11777", + "issue:11989", + "issue:12126", + "issue:12329", + "issue:1444", + "issue:1552", + "issue:1708", + "issue:1944", + "issue:2296", + "issue:2947", + "issue:3152", + "issue:3619", + "issue:4213", + "issue:4450", + "issue:5246", + "issue:5465", + "issue:5607", + "issue:5661", + "issue:6159", + "issue:6226", + "issue:646", + "issue:6574", + "issue:6584", + "issue:6594", + "issue:668", + "issue:6758", + "issue:696", + "issue:6964", + "issue:7057", + "issue:7168", + "issue:7293", + "issue:7432", + "issue:7660", + "issue:7708", + "issue:817", + "issue:861", + "issue:8939", + "issue:9180", + "issue:9643" + ], + "soft_pairs": [ + "issue:5607|issue:6574", + "issue:668|issue:696", + "issue:4213|issue:6964", + "issue:4213|issue:7660", + "issue:10436|issue:6226", + "issue:10180|issue:8939", + "issue:12126|issue:12329", + "issue:7168|issue:7432", + "issue:1708|issue:1944", + "issue:11422|issue:11777", + "issue:6584|issue:861", + "issue:11214|issue:11362", + "issue:11374|issue:7293", + "issue:11062|issue:11702", + "issue:5465|issue:9643", + "issue:2947|issue:9180", + "issue:5246|issue:646", + "issue:1552|issue:7057", + "issue:4450|issue:817", + "issue:5661|issue:7708", + "issue:10223|issue:11989", + "issue:6159|issue:6758", + "issue:1552|issue:3619", + "issue:2296|issue:6594", + "issue:1444|issue:3152" + ], + "prepared_review_unit_hash": "70d23bc9f223654ea9fb32ed0897a20237bf54b30a635dc9102994638c8d5ff4", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34969, + "estimated_input_tokens": 8743, + "estimated_eval_tokens": 17742 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly a loose mix of unrelated model-loading, LoRA, pipeline-argument, and training questions/bugs. None of the soft pairs look like true duplicates of the same underlying defect, so I would not merge any of them.", + "confidence": 0.88, + "canonical_issue_reason": "issue:10180 is the clearest concrete bug report in the set: a reproducible Flux Control LoRA loading failure with real discussion and references, making it the strongest single issue to anchor triage.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10180 is the best representative issue for the dominant theme here (LoRA/model-loading failures), but the overall set is too heterogeneous to imply one duplicate family.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5607", + "right": "issue:6574", + "accept": false, + "reason": "One is a tensor shape mismatch runtime error; the other is a warning about SVD pipeline input range. Different failure modes." + }, + { + "left": "issue:668", + "right": "issue:696", + "accept": false, + "reason": "Both mention DreamBooth training, but one is a T4 runtime failure and the other is plain CUDA OOM. Not the same bug." + }, + { + "left": "issue:4213", + "right": "issue:6964", + "accept": false, + "reason": "from_single_file/accelerate loading issue vs 4-channel UNet incompatibility. Different code paths and root causes." + }, + { + "left": "issue:4213", + "right": "issue:7660", + "accept": false, + "reason": "Both are example/loading complaints, but one is about from_single_file() and the other about a trace example. Too generic to be duplicates." + }, + { + "left": "issue:10436", + "right": "issue:6226", + "accept": false, + "reason": "One is a model-specific loading failure; the other is a conversion script causing webui loading errors. Related area, but not the same defect." + }, + { + "left": "issue:10180", + "right": "issue:8939", + "accept": false, + "reason": "Flux Control LoRA multi-load bug vs SD3 tokenizer load error. Different models and different failure layers." + }, + { + "left": "issue:12126", + "right": "issue:12329", + "accept": false, + "reason": "Wan2.2 cache_context inference error vs pipeline misclassifying a single-file transformer as Wan2.1. Related subsystem, distinct bugs." + }, + { + "left": "issue:7168", + "right": "issue:7432", + "accept": false, + "reason": "IP-Adapter embedding tensor-shape problem vs general output quality mismatch with WebUI. Not the same issue." + }, + { + "left": "issue:1708", + "right": "issue:1944", + "accept": false, + "reason": "One asks how to pass a YAML config; the other reports random dreambooth results. Question vs training quality bug." + }, + { + "left": "issue:11422", + "right": "issue:11777", + "accept": false, + "reason": "A concrete HiDream LoRA validation-prompt crash vs a general question about training HiDream-I1 LoRAs. Not duplicates." + }, + { + "left": "issue:6584", + "right": "issue:861", + "accept": false, + "reason": "Specific DreamShaper inpaint conversion failure vs generic ckpt-to-diffusers conversion complaint. Same broad area, different bug." + }, + { + "left": "issue:11214", + "right": "issue:11362", + "accept": false, + "reason": "Similar 'unexpected keyword argument' pattern, but different model classes and arguments; not clearly the same code path." + }, + { + "left": "issue:11374", + "right": "issue:7293", + "accept": false, + "reason": "SkyReels-V2 integration request and AudioLDM2 integration request are unrelated feature requests." + }, + { + "left": "issue:11062", + "right": "issue:11702", + "accept": false, + "reason": "Both are LoRA loading problems, but they involve different models and different errors (CivitAI load error vs key error)." + }, + { + "left": "issue:5465", + "right": "issue:9643", + "accept": false, + "reason": "DiffEdit inpaint_strength support vs Flux multiple ControlNets. Different pipelines and unrelated bugs." + }, + { + "left": "issue:2947", + "right": "issue:9180", + "accept": false, + "reason": "Missing load_attn_procs attribute vs missing _execution_device attribute. Both attribute errors, but on different objects and code paths." + }, + { + "left": "issue:5246", + "right": "issue:646", + "accept": false, + "reason": "CLIPImageProjection import issue vs PNDMScheduler set_format issue. Completely different components." + }, + { + "left": "issue:1552", + "right": "issue:7057", + "accept": false, + "reason": "Attention slicing regression vs an unspecified model checkpoint failure. No clear shared defect." + }, + { + "left": "issue:4450", + "right": "issue:817", + "accept": false, + "reason": "Multi-textual-inversion support bug vs mixed-precision DreamBooth issue. Same training umbrella only, not same bug." + }, + { + "left": "issue:5661", + "right": "issue:7708", + "accept": false, + "reason": "Tokenizer symbol missing vs tokenizer/config.json missing. Both tokenizer-related, but one is an import issue and the other is a model repo layout issue." + }, + { + "left": "issue:10223", + "right": "issue:11989", + "accept": false, + "reason": "A question about where to obtain a LoRA ID vs a black-image inference bug. Not duplicates." + }, + { + "left": "issue:6159", + "right": "issue:6758", + "accept": false, + "reason": "Both are unsupported pipeline kwargs, but for different arguments and different pipelines. Too broad to merge." + }, + { + "left": "issue:1552", + "right": "issue:3619", + "accept": false, + "reason": "Attention slicing on Stable Diffusion 2 vs CrossFrameAttention in Text2Video-Zero. Different features and failures." + }, + { + "left": "issue:2296", + "right": "issue:6594", + "accept": false, + "reason": "xformers behavior on T4 vs a network error fetching a YAML config. Entirely unrelated." + }, + { + "left": "issue:1444", + "right": "issue:3152", + "accept": false, + "reason": "StableDiffusionSafetyChecker import error vs TensorFlow Lite analyzer wrapper import error. Different ecosystems and bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 48, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10169", + "issue:11045", + "issue:11702", + "issue:13035", + "issue:1552", + "issue:1572", + "issue:1709", + "issue:1909", + "issue:1974", + "issue:2089", + "issue:2161", + "issue:2185", + "issue:2941", + "issue:3451", + "issue:3487", + "issue:3651", + "issue:3761", + "issue:3774", + "issue:3803", + "issue:3811", + "issue:4213", + "issue:4265", + "issue:4925", + "issue:5014", + "issue:5197", + "issue:5226", + "issue:5259", + "issue:5605", + "issue:5619", + "issue:5860", + "issue:5923", + "issue:5992", + "issue:6130", + "issue:6158", + "issue:6218", + "issue:6224", + "issue:6284", + "issue:6427", + "issue:6967", + "issue:7133", + "issue:746", + "issue:761", + "issue:8236", + "issue:9500", + "issue:9525", + "issue:9790", + "issue:9841" + ], + "soft_pairs": [ + "issue:3761|issue:746", + "issue:2185|issue:5197", + "issue:5014|issue:6284", + "issue:5619|issue:5860", + "issue:3651|issue:8236", + "issue:5226|issue:6218", + "issue:761|issue:9841", + "issue:9500|issue:9790", + "issue:5923|issue:6224", + "issue:1552|issue:1572", + "issue:3451|issue:3774", + "issue:4213|issue:6427", + "issue:10169|issue:5259", + "issue:2089|issue:4925", + "issue:11045|issue:9525", + "issue:2161|issue:3803", + "issue:1709|issue:5992", + "issue:2941|issue:3487", + "issue:3811|issue:9525", + "issue:11702|issue:6130", + "issue:13035|issue:4265", + "issue:1909|issue:1974", + "issue:5605|issue:6158", + "issue:6967|issue:7133" + ], + "prepared_review_unit_hash": "82752a503da4f0484f877432c820aecbc9ee4d7d9c6ba873d1288c470a2c2afe", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33490, + "estimated_input_tokens": 8373, + "estimated_eval_tokens": 17002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a broad false-positive grouping: the items span unrelated loading, training, and pipeline bugs rather than one underlying defect. None of the soft-edge pairs look like true duplicates or mergeable PR-equivalents.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 6130 is the most general and representative user-facing problem in the set ('Error while loading Lora'), but the cluster is too heterogeneous to have a true single canonical duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6130 is the best representative issue only in the sense that it is a broad, concrete loading problem with substantial discussion; it is not a strong canonical for the whole cluster because the rest of the items are mostly different bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3761", + "right": "issue:746", + "accept": false, + "reason": "Different failures: community pipeline process-loading vs ONNX revision import in img2img." + }, + { + "left": "issue:2185", + "right": "issue:5197", + "accept": false, + "reason": "Training config-file omission is unrelated to multi-GPU slowdown/performance behavior." + }, + { + "left": "issue:5014", + "right": "issue:6284", + "accept": false, + "reason": "Dreambooth LoRA output quality issue is distinct from SDXL LoRA validation/test result mismatch." + }, + { + "left": "issue:5619", + "right": "issue:5860", + "accept": false, + "reason": "AutoencodeTiny img2img encode bug is a different code path from Tiny Autoencoder + ControlNet pipeline failure." + }, + { + "left": "issue:3651", + "right": "issue:8236", + "accept": false, + "reason": "Multi-GPU DreamBooth execution failure is not the same as incorrect LR scheduler step counting." + }, + { + "left": "issue:5226", + "right": "issue:6218", + "accept": false, + "reason": "Both involve batching/LoRA-like generation behavior, but one is AdapterPipeline load_lora_weights and the other is batched seeds in ControlNet/img2img; different bugs." + }, + { + "left": "issue:761", + "right": "issue:9841", + "accept": false, + "reason": "Old DreamBooth 8-bit Adam training failure is unrelated to SDXL DreamBooth LoRA TypeError with use_dora=False." + }, + { + "left": "issue:9500", + "right": "issue:9790", + "accept": false, + "reason": "Checkpoint save failure during Flux Dreambooth training is distinct from multi-GPU training problems." + }, + { + "left": "issue:5923", + "right": "issue:6224", + "accept": false, + "reason": "Both concern training instability, but one is stuck execution in multi-GPU fine-tuning and the other is a TorchDynamo error in text-to-image training." + }, + { + "left": "issue:1552", + "right": "issue:1572", + "accept": false, + "reason": "Attention slicing on SD2 and stable_diffusion_mega img2img/inpainting are separate pipeline issues." + }, + { + "left": "issue:3451", + "right": "issue:3774", + "accept": false, + "reason": "fp16 memory regression in a training script is not the same as zero3 not saving memory in another script." + }, + { + "left": "issue:4213", + "right": "issue:6427", + "accept": false, + "reason": "SDXL from_single_file accelerate dependency issue is unrelated to broken diffusers-to-ckpt conversion." + }, + { + "left": "issue:10169", + "right": "issue:5259", + "accept": false, + "reason": "Both are missing-file load errors, but config.json missing and model_index.json missing are different concrete problems." + }, + { + "left": "issue:2089", + "right": "issue:4925", + "accept": false, + "reason": "AltDiffusion DreamBooth OOM and SDXL ControlNet distributed OOM are different models/training setups, not the same bug." + }, + { + "left": "issue:11045", + "right": "issue:9525", + "accept": false, + "reason": "Flux Schnell loading crash during Dreambooth LoRA training is distinct from lora_scale having no effect in Flux loading." + }, + { + "left": "issue:2161", + "right": "issue:3803", + "accept": false, + "reason": "LoRA training on an inpainting model and multi-GPU LoRA model sharing are different training issues." + }, + { + "left": "issue:1709", + "right": "issue:5992", + "accept": false, + "reason": "ONNX img2img Tensor.astype AttributeError is unrelated to llm_guided_diffusion bool/__module__ error." + }, + { + "left": "issue:2941", + "right": "issue:3487", + "accept": false, + "reason": "EulerAncestralDiscreteScheduler result quality in standard SD vs inpaint-with-strength poor results are related scheduler names but different code paths and symptoms." + }, + { + "left": "issue:3811", + "right": "issue:9525", + "accept": false, + "reason": "Both involve LoRA loading in Flux-like contexts, but 'no effect on generation' and 'lora_scale has no effect' are not clearly the same defect." + }, + { + "left": "issue:11702", + "right": "issue:6130", + "accept": false, + "reason": "AccVid LoRA key error when loading with diffusers is a different loading failure than the generic 'Error while loading Lora' report." + }, + { + "left": "issue:13035", + "right": "issue:4265", + "accept": false, + "reason": "Missing MT5Tokenizer import from transformers is unrelated to missing StableDiffusionXLControlNetPipeline import from diffusers." + }, + { + "left": "issue:1909", + "right": "issue:1974", + "accept": false, + "reason": "Generic scheduler-generator API mismatch and Euler-specific reproducibility failure are related only by generators, not the same bug." + }, + { + "left": "issue:5605", + "right": "issue:6158", + "accept": false, + "reason": "Per-image caption support request for SDXL dreambooth is a feature request, not the same as adding SDXL support to animatediff." + }, + { + "left": "issue:6967", + "right": "issue:7133", + "accept": false, + "reason": "Both are SDXL Dreambooth LoRA training problems, but one is prior-preservation shape mismatch and the other is checkpoint resume size mismatch; different failure points." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 49, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10382", + "issue:10472", + "issue:10512", + "issue:10653", + "issue:11169", + "issue:11410", + "issue:11464", + "issue:11907", + "issue:11922", + "issue:12436", + "issue:12728", + "issue:13035", + "issue:13286", + "issue:1377", + "issue:1439", + "issue:1538", + "issue:1717", + "issue:1985", + "issue:2324", + "issue:2371", + "issue:2480", + "issue:3152", + "issue:3223", + "issue:3296", + "issue:3357", + "issue:4139", + "issue:4256", + "issue:4534", + "issue:4885", + "issue:490", + "issue:499", + "issue:5178", + "issue:5226", + "issue:5633", + "issue:6069", + "issue:6218", + "issue:6295", + "issue:6382", + "issue:646", + "issue:6536", + "issue:6964", + "issue:7082", + "issue:7178", + "issue:7286", + "issue:7882", + "issue:852", + "issue:9200", + "issue:9556" + ], + "soft_pairs": [ + "issue:11169|issue:9200", + "issue:499|issue:5226", + "issue:11410|issue:5633", + "issue:2324|issue:2480", + "issue:1717|issue:3296", + "issue:5226|issue:6964", + "issue:6069|issue:6295", + "issue:1439|issue:646", + "issue:4256|issue:4534", + "issue:2324|issue:2371", + "issue:5178|issue:6069", + "issue:6382|issue:9556", + "issue:499|issue:7178", + "issue:10472|issue:6536", + "issue:1985|issue:7882", + "issue:10382|issue:5633", + "issue:1538|issue:3223", + "issue:12436|issue:852", + "issue:11907|issue:6218", + "issue:12728|issue:3357", + "issue:10512|issue:4885", + "issue:11464|issue:11922", + "issue:490|issue:6536", + "issue:10653|issue:13286", + "issue:13035|issue:7082", + "issue:1377|issue:3152", + "issue:4139|issue:7286" + ], + "prepared_review_unit_hash": "333f4665592bed45b7c019df73c043c69b5e533e7fc642bfdfcbf0550fbc9c33", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35688, + "estimated_input_tokens": 8922, + "estimated_eval_tokens": 18100 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very heterogeneous: it mixes LoRA-loading bugs, scheduler/runtime errors, import failures, training-script issues, and torch.compile problems. I found no soft pair that looks like the same underlying bug/change, so all soft edges are rejected.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 10512 is the best anchor only in a loose sense: it is a concrete, user-facing Flux/LoRA loading failure and has the most discussion among the similar-looking subset. It is still not a true duplicate of the rest of the cluster, which spans many unrelated subsystems.", + "canonical_pr_reason": null, + "best_issue_reason": "10512 is the most central representative of the nearest theme in this noisy cluster, but it is not a good global canonical for all items because the cluster is too broad and mixed.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11169", + "right": "issue:9200", + "accept": false, + "reason": "Different failure classes: numpy ABI/binary incompatibility during import vs a dtype error in StableDiffusionXLControlNetPipeline." + }, + { + "left": "issue:499", + "right": "issue:5226", + "accept": false, + "reason": "Both are pipeline-related, but one is an example unconditional training script failure and the other is LoRA loading for SDXL Adapter; different code paths." + }, + { + "left": "issue:11410", + "right": "issue:5633", + "accept": false, + "reason": "Attention unpacking ValueError in a processor call is unrelated to a bug in transformer_temporal.py." + }, + { + "left": "issue:2324", + "right": "issue:2480", + "accept": false, + "reason": "Both concern Dreambooth training, but one is about instance data reading and the other about checkpoint directory structure; not the same bug." + }, + { + "left": "issue:1717", + "right": "issue:3296", + "accept": false, + "reason": "Offline model-cache loading and Dreambooth checkpoint saving/loading are different mechanisms and failure modes." + }, + { + "left": "issue:5226", + "right": "issue:6964", + "accept": false, + "reason": "Both are load-related pipeline issues, but one is about LoRA weights on an SDXL adapter pipeline and the other is 4-channel UNet support in ControlNet inpaint." + }, + { + "left": "issue:6069", + "right": "issue:6295", + "accept": false, + "reason": "Different scheduler bugs: add_noise at t=0 vs artifacts with DPM++ 2M SDE Karras/use_lu_lambdas." + }, + { + "left": "issue:1439", + "right": "issue:646", + "accept": false, + "reason": "Different import/API regressions: FeatureExtractionMixin missing vs PNDMScheduler lacking set_format." + }, + { + "left": "issue:4256", + "right": "issue:4534", + "accept": false, + "reason": "Both are Dreambooth-related, but one is a generic training error and the other is an MPS validation crash." + }, + { + "left": "issue:2324", + "right": "issue:2371", + "accept": false, + "reason": "Same training area, but one is about reading instance data and the other is about generated objects not appearing in output; not a duplicate." + }, + { + "left": "issue:5178", + "right": "issue:6069", + "accept": false, + "reason": "Different scheduler issues: resolution changes after first run vs add_noise failing at t=0." + }, + { + "left": "issue:6382", + "right": "issue:9556", + "accept": false, + "reason": "Both mention LoRA-related workflows, but one is an IP-adapter key error during SDXL training/use and the other is problems with saved LoRA models under rslora." + }, + { + "left": "issue:499", + "right": "issue:7178", + "accept": false, + "reason": "Training script failure vs export_to_video behavior; unrelated code paths." + }, + { + "left": "issue:10472", + "right": "issue:6536", + "accept": false, + "reason": "Different issues entirely: unexpected lora_bias argument in inference vs out-of-memory during UNet3DConditionModel fine-tuning." + }, + { + "left": "issue:1985", + "right": "issue:7882", + "accept": false, + "reason": "Both relate to adding/loading capabilities, but embeddings on Stable Diffusion vs missing from_single_file on a T2I adapter pipeline are not the same bug." + }, + { + "left": "issue:10382", + "right": "issue:5633", + "accept": false, + "reason": "VAE forward return-type mismatch causing AttributeError is unrelated to a bug in transformer_temporal.py." + }, + { + "left": "issue:1538", + "right": "issue:3223", + "accept": false, + "reason": "Different runtime problems: StableDiffusionMegaPipeline model compatibility vs memory not clearing after moving pipeline to cuda." + }, + { + "left": "issue:12436", + "right": "issue:852", + "accept": false, + "reason": "Both are constructor keyword regressions, but offload_state_dict in CLIPTextModel is different from device_map on MPS in diffusers 0.5.1." + }, + { + "left": "issue:11907", + "right": "issue:6218", + "accept": false, + "reason": "Scheduler compatibility bug in FLUX vs batched seeds behavior in ControlNet/img2img; not the same underlying issue." + }, + { + "left": "issue:12728", + "right": "issue:3357", + "accept": false, + "reason": "Both involve torch.compile/Dynamo, but one is a channels_last + Wan 2.2 runtime/unsupported behavior problem and the other is a hasattr TensorVariable compile error." + }, + { + "left": "issue:10512", + "right": "issue:4885", + "accept": false, + "reason": "Both are LoRA-related, but one is unable to load Quanto Flux LoRA and the other is SDXL ControlNet Img2Img not loading LoRA weights; different pipelines and failure modes." + }, + { + "left": "issue:11464", + "right": "issue:11922", + "accept": false, + "reason": "Both are FLUX model issues, but one is loading a Canny-dev LoRA into FluxControlPipeline and the other is a generic FLUX.1-Fill-dev runtime problem." + }, + { + "left": "issue:490", + "right": "issue:6536", + "accept": false, + "reason": "Memory pressure in diffusion pipelines vs out-of-memory while fine-tuning UNet3DConditionModel; too broad and different contexts." + }, + { + "left": "issue:10653", + "right": "issue:13286", + "accept": false, + "reason": "8-bit .to(cpu) limitation for HunyuanVideoTransformer3DModel is unrelated to torchao quantization support in general." + }, + { + "left": "issue:13035", + "right": "issue:7082", + "accept": false, + "reason": "Import errors from different dependency packages: transformers MT5Tokenizer vs accelerate.hooks CpuOffload." + }, + { + "left": "issue:1377", + "right": "issue:3152", + "accept": false, + "reason": "Different external import failures: CLIPVisionModelWithProjection from transformers vs a TensorFlow Lite analyzer wrapper symbol." + }, + { + "left": "issue:4139", + "right": "issue:7286", + "accept": false, + "reason": "Both involve attention/custom diffusion, but one is precision casting for a custom attention processor and the other is cross-attention weight loading in Attend-and-Excite; not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 50, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10180", + "issue:10450", + "issue:10612", + "issue:10653", + "issue:10989", + "issue:11225", + "issue:11241", + "issue:11352", + "issue:11408", + "issue:11464", + "issue:11585", + "issue:1643", + "issue:1842", + "issue:2482", + "issue:2560", + "issue:2819", + "issue:2875", + "issue:3094", + "issue:3450", + "issue:3565", + "issue:3595", + "issue:3604", + "issue:3884", + "issue:4052", + "issue:4213", + "issue:4483", + "issue:4885", + "issue:499", + "issue:5014", + "issue:5016", + "issue:5197", + "issue:5621", + "issue:5628", + "issue:5798", + "issue:6069", + "issue:6362", + "issue:646", + "issue:7276", + "issue:7378", + "issue:7418", + "issue:9170", + "issue:9171", + "issue:9343", + "issue:9370", + "issue:9546", + "issue:9841", + "issue:9911" + ], + "soft_pairs": [ + "issue:5014|issue:9841", + "issue:3450|issue:3595", + "issue:5014|issue:5016", + "issue:10989|issue:11464", + "issue:3884|issue:7276", + "issue:4213|issue:499", + "issue:10180|issue:4885", + "issue:4213|issue:9171", + "issue:3565|issue:6362", + "issue:4052|issue:646", + "issue:10612|issue:9343", + "issue:1842|issue:5798", + "issue:10653|issue:11352", + "issue:1643|issue:9370", + "issue:11585|issue:4483", + "issue:2482|issue:3604", + "issue:11225|issue:3094", + "issue:2875|issue:5197", + "issue:7378|issue:9170", + "issue:10010|issue:10450", + "issue:2819|issue:7418", + "issue:5628|issue:6069", + "issue:11241|issue:9546", + "issue:11408|issue:5621", + "issue:2560|issue:9911" + ], + "prepared_review_unit_hash": "3c124435ad4ec2cbb77478f8f1864611b0eb78ac9e9008ed81275472e309148f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34781, + "estimated_input_tokens": 8696, + "estimated_eval_tokens": 17648 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is noisy and mostly groups Flux/LoRA/ControlNet/model-loading issues by lexical similarity. Only one pair looks like a true duplicate; the rest are related only at a broad subsystem level or describe different failure modes.", + "confidence": 0.73, + "canonical_issue_reason": "issue:10989 is the clearest representative of the main Flux LoRA loading problem, with a concrete failure mode (quantized transformers) and the strongest overlap with other Flux/LoRA loader reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10989 is the best issue to anchor this cluster because it states the core Flux ControlNet LoRA loading bug most concretely and broadly.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5014", + "right": "issue:9841", + "accept": false, + "reason": "Both are DreamBooth/LoRA-related, but one is a vague poor-results report and the other is a specific TypeError when use_dora=False; different bugs." + }, + { + "left": "issue:3450", + "right": "issue:3595", + "accept": true, + "reason": "Both report the same from_ckpt loading failure, with nearly identical wording and same code path." + }, + { + "left": "issue:5014", + "right": "issue:5016", + "accept": false, + "reason": "General bad LoRA training results vs a specific SDXL text-encoder training limitation; not the same concrete defect." + }, + { + "left": "issue:10989", + "right": "issue:11464", + "accept": false, + "reason": "Both mention Flux/LoRA loading, but one is quantization-specific and the other is a model/pipeline loading failure for a Canny LoRA; too different to merge." + }, + { + "left": "issue:3884", + "right": "issue:7276", + "accept": false, + "reason": "Different examples and different failure classes: LoRA script error vs dreambooth pretrained_model network errors." + }, + { + "left": "issue:4213", + "right": "issue:499", + "accept": false, + "reason": "from_single_file accelerate dependency issue vs an unrelated unconditional training script problem." + }, + { + "left": "issue:10180", + "right": "issue:4885", + "accept": false, + "reason": "Both involve LoRA loading, but they affect different pipelines and describe different adapter-loading behavior." + }, + { + "left": "issue:4213", + "right": "issue:9171", + "accept": false, + "reason": "from_single_file without accelerate is unrelated to the SD1.5 v_prediction regression." + }, + { + "left": "issue:3565", + "right": "issue:6362", + "accept": false, + "reason": "Both are OOM reports, but they occur in different training stages and scripts; not the same bug." + }, + { + "left": "issue:4052", + "right": "issue:646", + "accept": false, + "reason": "Missing from_single_file method vs scheduler.set_format attribute error; unrelated API issues." + }, + { + "left": "issue:10612", + "right": "issue:9343", + "accept": false, + "reason": "Both are FLUX load issues, but one is NF4/LoRA loading and the other is low_cpu_mem_usage/ignore_mismatched_sizes; different code paths." + }, + { + "left": "issue:1842", + "right": "issue:5798", + "accept": false, + "reason": "Different missing-argument API errors (`eta` vs `callback_on_step_end`) in different pipelines." + }, + { + "left": "issue:10653", + "right": "issue:11352", + "accept": false, + "reason": "Both concern bitsandbytes 8bit/CPU movement, but one is a hard `.to()` error and the other is about missing warning behavior; not same concrete bug." + }, + { + "left": "issue:1643", + "right": "issue:9370", + "accept": false, + "reason": "Completely different model API errors: xformers attention signature vs missing pooled_projection." + }, + { + "left": "issue:11585", + "right": "issue:4483", + "accept": false, + "reason": "A subjective output-quality complaint is not the same as a determinism bug." + }, + { + "left": "issue:2482", + "right": "issue:3604", + "accept": false, + "reason": "Both discuss weak LoRA training, but one is non-convergence and the other is loss not decreasing; too vague to treat as duplicates." + }, + { + "left": "issue:11225", + "right": "issue:3094", + "accept": false, + "reason": "Both are OOM during training, but on different models/examples and different phases." + }, + { + "left": "issue:2875", + "right": "issue:5197", + "accept": false, + "reason": "Training loss NaN and multi-GPU performance regression are unrelated issues." + }, + { + "left": "issue:7378", + "right": "issue:9170", + "accept": false, + "reason": "Different memory problems in different pipelines; one is latent decoding, the other SDXL/ControlNet training memory size." + }, + { + "left": "issue:10010", + "right": "issue:10450", + "accept": false, + "reason": "Missing model_index.json for FLUX Fill example is unrelated to an architectural difference between transformer blocks." + }, + { + "left": "issue:2819", + "right": "issue:7418", + "accept": false, + "reason": "A broken raw.githubusercontent example URL and a bug in a specific community pipeline are unrelated." + }, + { + "left": "issue:5628", + "right": "issue:6069", + "accept": false, + "reason": "Both are scheduler bugs, but they affect different schedulers and different numerical edge cases." + }, + { + "left": "issue:11241", + "right": "issue:9546", + "accept": false, + "reason": "Floating point exception during Flux ControlNet training vs validation-step OOM; different failure modes." + }, + { + "left": "issue:11408", + "right": "issue:5621", + "accept": false, + "reason": "Compiled-model adapter support and SDXL ControlNet encoder_hid_proj support are unrelated feature gaps." + }, + { + "left": "issue:2560", + "right": "issue:9911", + "accept": false, + "reason": "Determinism with a list of generators vs multi-ControlNet layer-length incompatibility are different issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The one accepted pair, issue:3450 vs issue:3595, is well supported by the nearly identical from_ckpt failure descriptions. The other pair judgments are appropriately cautious given the titles/bodies available. The only slightly broad phrasing is the claim that issue:10989 is the cluster anchor, since it is quantization-specific, but it still works as a representative Flux LoRA loading issue." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 51, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1070", + "issue:10804", + "issue:1082", + "issue:11062", + "issue:11464", + "issue:11585", + "issue:11700", + "issue:1188", + "issue:1614", + "issue:1617", + "issue:1672", + "issue:2288", + "issue:2296", + "issue:3788", + "issue:4006", + "issue:4061", + "issue:4094", + "issue:4256", + "issue:4529", + "issue:4530", + "issue:460", + "issue:490", + "issue:4916", + "issue:4925", + "issue:5006", + "issue:5012", + "issue:5225", + "issue:5412", + "issue:5433", + "issue:5442", + "issue:5474", + "issue:6130", + "issue:6159", + "issue:6176", + "issue:6353", + "issue:6385", + "issue:6761", + "issue:6894", + "issue:7146", + "issue:725", + "issue:7322", + "issue:7619", + "issue:8514", + "issue:8939", + "issue:9421", + "issue:9643", + "issue:9732", + "issue:9804" + ], + "soft_pairs": [ + "issue:4006|issue:4916", + "issue:1082|issue:4256", + "issue:3788|issue:5442", + "issue:4529|issue:6761", + "issue:4925|issue:6353", + "issue:1614|issue:1672", + "issue:1070|issue:9421", + "issue:4530|issue:5433", + "issue:2288|issue:4094", + "issue:11585|issue:2296", + "issue:10804|issue:11464", + "issue:1617|issue:7619", + "issue:5012|issue:6894", + "issue:11062|issue:6130", + "issue:1188|issue:5474", + "issue:1672|issue:490", + "issue:11700|issue:725", + "issue:7146|issue:8514", + "issue:5412|issue:7322", + "issue:8939|issue:9804", + "issue:6159|issue:9643", + "issue:1070|issue:460", + "issue:4061|issue:6176", + "issue:5225|issue:6385", + "issue:5006|issue:9732" + ], + "prepared_review_unit_hash": "cbd696c0f32312decf3ce21663d7e0e80826c294848e77b60919c9041f529da6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34720, + "estimated_input_tokens": 8680, + "estimated_eval_tokens": 17616 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes many unrelated training, OOM, pipeline, and LoRA-loading reports. The soft-similarity pairs mostly reflect broad thematic overlap rather than the same underlying bug, so none should be merged as duplicates.", + "confidence": 0.9, + "canonical_issue_reason": "issue:6130 is the broadest and most central report in the LoRA-loading subset ('Error while loading Lora'); it can serve as an umbrella reference, but it is not a true duplicate of the other issue-specific failures.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6130 is the best representative issue because it is the most generic and closest to a catch-all for the repeated LoRA-loading complaints in this set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4006", + "right": "issue:4916", + "accept": false, + "reason": "Both are training-related and mention gradient checkpointing/DeepSpeed, but they describe different failure modes and contexts; not the same concrete bug." + }, + { + "left": "issue:1082", + "right": "issue:4256", + "accept": false, + "reason": "Both are training errors, but one is a CUDA cublas failure in Dreambooth and the other is a generic SDXL-Dreambooth training error; too little evidence they share one root cause." + }, + { + "left": "issue:3788", + "right": "issue:5442", + "accept": false, + "reason": "Different new-pipeline feature requests (UniControl vs ScaleCrafter); no indication of the same change or bug." + }, + { + "left": "issue:4529", + "right": "issue:6761", + "accept": false, + "reason": "Both are API/compatibility errors, but one is a LoRA loader keyword mismatch and the other is a VaeImageProcessor.preprocess argument issue; different code paths." + }, + { + "left": "issue:4925", + "right": "issue:6353", + "accept": false, + "reason": "Both concern SDXL training examples, but one is distributed ControlNet OOM and the other is an example-command failure; not the same underlying issue." + }, + { + "left": "issue:1614", + "right": "issue:1672", + "accept": false, + "reason": "Both relate to incorrect image outputs after conversion/autocast, but the symptoms and workflows differ enough that they do not look like one bug." + }, + { + "left": "issue:1070", + "right": "issue:9421", + "accept": false, + "reason": "Both mention inpainting models, but one is a bug in a new inpainting version and the other is inability to access a specific model repository; different problems." + }, + { + "left": "issue:4530", + "right": "issue:5433", + "accept": false, + "reason": "Both involve SDXL artifacts, but one is a refiner artifact issue and the other is DPM++ scheduler artifacts without refiner; not the same concrete defect." + }, + { + "left": "issue:2288", + "right": "issue:4094", + "accept": false, + "reason": "These are separate feature requests for different training scripts (InstructPix2Pix vs T2I adapters)." + }, + { + "left": "issue:11585", + "right": "issue:2296", + "accept": false, + "reason": "A generic poor-result report is not the same as xformers failing on T4; too broad to be duplicates." + }, + { + "left": "issue:10804", + "right": "issue:11464", + "accept": false, + "reason": "Both are LoRA loading problems, but one is FLUX1.dev LoRA loading and the other is FLUX.1-Canny-dev-lora into FluxControlPipeline; similar area, different concrete failure." + }, + { + "left": "issue:1617", + "right": "issue:7619", + "accept": false, + "reason": "Both are training-script errors, but one is missing required args for Dreambooth and the other is a channel-count mismatch; distinct bugs." + }, + { + "left": "issue:5012", + "right": "issue:6894", + "accept": false, + "reason": "Both concern SDXL LoRA text-encoder behavior, but one is inability to train the text encoder and the other is a LoRA not working in A1111; different manifestations and code paths." + }, + { + "left": "issue:11062", + "right": "issue:6130", + "accept": false, + "reason": "Both are LoRA loading complaints, but one is a specific CivitAI LoRA loading error and the other is a generic 'Error while loading Lora'; likely related only at a broad level, not clearly the same bug." + }, + { + "left": "issue:1188", + "right": "issue:5474", + "accept": false, + "reason": "Different feature requests for different pipelines (CLIPSeg+SD vs Stable Diffusion Reference Only)." + }, + { + "left": "issue:1672", + "right": "issue:490", + "accept": false, + "reason": "Both are model conversion/memory-related issues, but one is wrong output size after ckpt-to-diffusers conversion and the other is general memory problems; not the same defect." + }, + { + "left": "issue:11700", + "right": "issue:725", + "accept": false, + "reason": "Both concern loading pipelines/models with lower memory use, but one requests pruna integration and the other asks for device map/accelerate support; different changes." + }, + { + "left": "issue:7146", + "right": "issue:8514", + "accept": false, + "reason": "Both are training failures, but one is SDXL fine-tuning CUDA OOM and the other is an EMAModel parallel fine-tuning error; different root causes." + }, + { + "left": "issue:5412", + "right": "issue:7322", + "accept": false, + "reason": "Both mention SDXL Dreambooth LoRA training failures, but one is checkpoint resume failure and the other is SIGKILL/OOM with a newer diffusers version; not the same concrete bug." + }, + { + "left": "issue:8939", + "right": "issue:9804", + "accept": false, + "reason": "Both are model-loading errors, but one is a tokenizer load failure for SD3 and the other is local SDXL ControlNet model loading; unrelated." + }, + { + "left": "issue:6159", + "right": "issue:9643", + "accept": false, + "reason": "Both involve unsupported ControlNet/Flux-related API behavior, but one is a missing denoising_end argument on SDXL ControlNetXSPipeline and the other is multiple ControlNets on Flux; different features." + }, + { + "left": "issue:1070", + "right": "issue:460", + "accept": false, + "reason": "Inpainting model issue vs ONNX/CUDA runtime dependency issue; unrelated subsystems." + }, + { + "left": "issue:4061", + "right": "issue:6176", + "accept": false, + "reason": "Both are LoRA-loading errors, but one is an SDXL load-lora issue and the other is loading LoRA weights after Dreambooth training; similar area, not the same bug." + }, + { + "left": "issue:5225", + "right": "issue:6385", + "accept": false, + "reason": "Both are VRAM-heavy SDXL-related problems, but one is a training forward-pass OOM and the other is unexpectedly high VRAM in StableVideoDiffusionPipeline; different code paths." + }, + { + "left": "issue:5006", + "right": "issue:9732", + "accept": false, + "reason": "Both are SDXL training OOMs, but one is BF16-specific end-of-training CUDA OOM and the other is a Flux dreambooth guide OOM on 3090 Ti; not clearly the same issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 52, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10282", + "issue:10650", + "issue:11045", + "issue:1552", + "issue:1565", + "issue:1974", + "issue:2185", + "issue:2459", + "issue:3124", + "issue:3177", + "issue:3382", + "issue:3420", + "issue:3582", + "issue:3619", + "issue:3802", + "issue:3803", + "issue:3884", + "issue:4468", + "issue:5028", + "issue:5319", + "issue:5466", + "issue:5515", + "issue:5619", + "issue:5884", + "issue:6069", + "issue:6176", + "issue:6362", + "issue:6446", + "issue:6610", + "issue:6866", + "issue:7002", + "issue:7064", + "issue:7146", + "issue:7157", + "issue:7197", + "issue:7210", + "issue:7263", + "issue:771", + "issue:7728", + "issue:7799", + "issue:8578", + "issue:8909", + "issue:9026", + "issue:9141", + "issue:9343", + "issue:9525", + "issue:9556" + ], + "soft_pairs": [ + "issue:2459|issue:3802", + "issue:3803|issue:4468", + "issue:6610|issue:9141", + "issue:10282|issue:8578", + "issue:1552|issue:6446", + "issue:11045|issue:9343", + "issue:3420|issue:771", + "issue:5319|issue:7157", + "issue:5466|issue:6176", + "issue:2185|issue:7197", + "issue:2459|issue:3124", + "issue:5028|issue:9026", + "issue:5884|issue:6866", + "issue:8909|issue:9556", + "issue:3582|issue:3619", + "issue:3884|issue:7728", + "issue:10650|issue:7002", + "issue:6362|issue:7146", + "issue:1974|issue:5619", + "issue:2459|issue:3619", + "issue:10026|issue:3177", + "issue:7210|issue:9525", + "issue:6866|issue:7064", + "issue:5515|issue:6069", + "issue:1565|issue:3382", + "issue:7263|issue:7799" + ], + "prepared_review_unit_hash": "7699020fdb07eede8b06a675c0209b0d6b9281b74099c55ad05b38d47584a339", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 34844, + "estimated_input_tokens": 8711, + "estimated_eval_tokens": 17678 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose lexical grouping of many unrelated diffusers training/loading issues. None of the soft pairs look like the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2459", + "right": "issue:3802", + "accept": false, + "reason": "Different bugs: xformers attention in one training script vs checkpoint retention limit not working in training examples." + }, + { + "left": "issue:3803", + "right": "issue:4468", + "accept": false, + "reason": "Both mention multi-GPU behavior, but one is about LoRA model sharing/state across GPUs and the other is about performance differences; not the same concrete failure." + }, + { + "left": "issue:6610", + "right": "issue:9141", + "accept": false, + "reason": "Different APIs and failure modes: SDXL inpaint `num_in_channels` ignored vs FluxPipeline missing `from_single_file()`." + }, + { + "left": "issue:10282", + "right": "issue:8578", + "accept": false, + "reason": "Both are loading/meta-tensor errors, but the model shapes and root causes differ; not the same bug." + }, + { + "left": "issue:1552", + "right": "issue:6446", + "accept": false, + "reason": "Stable Diffusion 2 attention slicing issue is unrelated to minimal-loading/Colab model loading failure." + }, + { + "left": "issue:11045", + "right": "issue:9343", + "accept": false, + "reason": "Both involve Flux loading/training, but one is a DreamBooth LoRA crash and the other is a parameter-loading mismatch with low_cpu_mem_usage/ignore_mismatched_sizes." + }, + { + "left": "issue:3420", + "right": "issue:771", + "accept": false, + "reason": "ControlNet inpaint pipeline integration issue vs BF16 dreambooth support; different code paths and symptoms." + }, + { + "left": "issue:5319", + "right": "issue:7157", + "accept": false, + "reason": "Both touch PEFT/LoRA, but one is about loading speed and the other about unexpected keys when loading an LCM LoRA; not the same defect." + }, + { + "left": "issue:5466", + "right": "issue:6176", + "accept": false, + "reason": "Different LoRA loading problems: SDXL shape mismatch on image-to-image load vs failure after training dreambooth_lora." + }, + { + "left": "issue:2185", + "right": "issue:7197", + "accept": false, + "reason": "Missing config files for naive text-to-image training is unrelated to the KeyError while resuming training from local SD 1.5." + }, + { + "left": "issue:2459", + "right": "issue:3124", + "accept": false, + "reason": "One is a training-script xformers issue; the other is a model conversion failure. No shared concrete bug." + }, + { + "left": "issue:5028", + "right": "issue:9026", + "accept": false, + "reason": "Unexpected keyword/forward signature problems in different components: reference pipeline forward hook vs accelerate dispatch argument mismatch." + }, + { + "left": "issue:5884", + "right": "issue:6866", + "accept": false, + "reason": "Both are IP-Adapter feature requests, but one is broad support across ControlNet/T2I pipelines and the other is specific to SDXL ControlNet Inpaint." + }, + { + "left": "issue:8909", + "right": "issue:9556", + "accept": false, + "reason": "Different LoRA-save problems: text_encoder_ti saving in an advanced DreamBooth script vs rslora save-format problems." + }, + { + "left": "issue:3582", + "right": "issue:3619", + "accept": false, + "reason": "Different pipelines and bugs: ControlNet inpaint strength handling vs Text2Video-Zero CrossFrameAttention." + }, + { + "left": "issue:3884", + "right": "issue:7728", + "accept": false, + "reason": "Both are training-script errors, but they involve different scripts and different failure conditions; not the same issue." + }, + { + "left": "issue:10650", + "right": "issue:7002", + "accept": false, + "reason": "Different shape/buffer failures in different pipelines and models; too generic to treat as one bug." + }, + { + "left": "issue:6362", + "right": "issue:7146", + "accept": false, + "reason": "Both mention SDXL OOM, but one is dataset mapping memory usage and the other is fine-tuning OOM; related theme, not same root cause." + }, + { + "left": "issue:1974", + "right": "issue:5619", + "accept": false, + "reason": "Euler scheduler generator-list reproducibility bug is unrelated to AutoencodeTiny/LCM image-encoding failure." + }, + { + "left": "issue:2459", + "right": "issue:3619", + "accept": false, + "reason": "xformers memory-efficient attention in a training script is unrelated to CrossFrameAttention not working in Text2Video-Zero." + }, + { + "left": "issue:10026", + "right": "issue:3177", + "accept": false, + "reason": "Both concern DeepSpeed stage 3 memory/offload behavior, but one is Flux ControlNet multi-GPU training and the other is DreamBooth stage-3 offloading; different code paths." + }, + { + "left": "issue:7210", + "right": "issue:9525", + "accept": false, + "reason": "Different Flux-related problems: missing `named_parameters` under checkpoint dispatch vs `lora_scale` having no effect when loading." + }, + { + "left": "issue:6866", + "right": "issue:7064", + "accept": false, + "reason": "Both are IP-Adapter support issues, but they target different pipelines and capabilities, so they are not the same bug/change." + }, + { + "left": "issue:5515", + "right": "issue:6069", + "accept": false, + "reason": "LCM pipeline initialization without a scheduler is unrelated to DPM++ Karras `add_noise` failing at `t=0`." + }, + { + "left": "issue:1565", + "right": "issue:3382", + "accept": false, + "reason": "GPU selection/placement complaint vs multi-GPU OOM in text-to-image training; similar theme, different underlying issue." + }, + { + "left": "issue:7263", + "right": "issue:7799", + "accept": false, + "reason": "IP-Adapter plus attention slicing vs IP-Adapter matmul size mismatch in a different pipeline; not the same concrete failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 53, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10154", + "issue:10337", + "issue:10542", + "issue:10952", + "issue:11006", + "issue:11127", + "issue:11145", + "issue:11436", + "issue:12104", + "issue:12141", + "issue:1261", + "issue:12656", + "issue:12689", + "issue:13300", + "issue:1478", + "issue:1552", + "issue:1614", + "issue:1672", + "issue:222", + "issue:2296", + "issue:2549", + "issue:2838", + "issue:2904", + "issue:3420", + "issue:3884", + "issue:4086", + "issue:4104", + "issue:4256", + "issue:4316", + "issue:4557", + "issue:5037", + "issue:5607", + "issue:5757", + "issue:6073", + "issue:6082", + "issue:6086", + "issue:6415", + "issue:6741", + "issue:684", + "issue:8482", + "issue:8556", + "issue:8659", + "issue:8667", + "issue:8817", + "issue:9443", + "issue:9525", + "issue:9599", + "issue:9861" + ], + "soft_pairs": [ + "issue:11145|issue:9599", + "issue:4557|issue:6741", + "issue:4256|issue:6073", + "issue:12104|issue:12656", + "issue:10154|issue:9525", + "issue:12141|issue:1552", + "issue:10952|issue:12689", + "issue:8556|issue:9443", + "issue:11436|issue:6082", + "issue:6415|issue:8817", + "issue:10337|issue:13300", + "issue:4086|issue:684", + "issue:1261|issue:222", + "issue:10542|issue:11006", + "issue:2838|issue:8482", + "issue:1614|issue:4104", + "issue:11127|issue:9861", + "issue:4316|issue:5037", + "issue:1478|issue:1614", + "issue:2549|issue:3420", + "issue:1672|issue:2296", + "issue:2904|issue:4557", + "issue:5607|issue:8667", + "issue:5757|issue:8659", + "issue:3884|issue:6086" + ], + "prepared_review_unit_hash": "794f3cca398046a0d53b2534f632a4a71917c9e0bc1dd2fd0660ee2ed54334e6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34676, + "estimated_input_tokens": 8669, + "estimated_eval_tokens": 17594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of unrelated Diffusers issues. Only a few pairs look like true duplicates: the train_dreambooth validation-image bug, the SD3ControlNetModel non-power-of-two size mismatch, and the IP-Adapter FaceID support issue.", + "confidence": 0.71, + "canonical_issue_reason": "issue:6415 is the clearest representative of one of the real duplicate-like subclusters: it is explicit about IP-Adapter FaceID/FaceID-plus support, broad enough to anchor the related reports, and has the most discussion context among that group.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6415 is the best overall issue in this set to keep as the canonical anchor because it is concrete, clearly scoped, and the most discussion-rich of the near-duplicate FaceID/IP-Adapter reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11145", + "right": "issue:9599", + "accept": false, + "reason": "Both mention FLUX LoRA, but one is a support request for Flux Fill LoRA/dreambooth finetuning and the other asks for a LoRA-only example; not the same bug or change." + }, + { + "left": "issue:4557", + "right": "issue:6741", + "accept": false, + "reason": "Both are generic AttributeError reports, but they are about different missing attributes in different code paths and do not look like one underlying defect." + }, + { + "left": "issue:4256", + "right": "issue:6073", + "accept": false, + "reason": "Both involve dreambooth training, but one is a specific SDXL training error and the other is a beginner\u2019s general error while following README instructions; too broad to merge." + }, + { + "left": "issue:12104", + "right": "issue:12656", + "accept": false, + "reason": "Both mention out-of-bounds indexing, but one is a generic IndexError and the other is a specific AuraFlow positional-embedding bug; likely different defects." + }, + { + "left": "issue:10154", + "right": "issue:9525", + "accept": false, + "reason": "Both concern Flux LoRA loading, but one is an attribute-error failure on load_lora_weights and the other is lora_scale having no effect; different symptoms and likely different fixes." + }, + { + "left": "issue:12141", + "right": "issue:1552", + "accept": false, + "reason": "These are unrelated dtype/precision and attention-slicing problems in different pipelines; not the same underlying bug." + }, + { + "left": "issue:10952", + "right": "issue:12689", + "accept": false, + "reason": "Both are LoRA-loading related, but one is an adapter-name mismatch in SDXL long-prompt weighting and the other is weights not being loaded from checkpoints; not the same concrete issue." + }, + { + "left": "issue:8556", + "right": "issue:9443", + "accept": false, + "reason": "One is a meta-tensor copy failure when loading SD3 from single file, the other is a tokenizer instantiation problem; different code paths." + }, + { + "left": "issue:11436", + "right": "issue:6082", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different models/pipelines and are not clearly the same bug." + }, + { + "left": "issue:6415", + "right": "issue:8817", + "accept": true, + "reason": "Both are about IP-Adapter FaceID not working / not being supported, so they appear to describe the same missing-feature bug." + }, + { + "left": "issue:10337", + "right": "issue:13300", + "accept": false, + "reason": "These are different model/runtime problems: quantized-transformer FLUX IPAdapter failure versus CPU generation failures for SD3/FLUX." + }, + { + "left": "issue:4086", + "right": "issue:684", + "accept": false, + "reason": "Both touch training or hardware compatibility, but one is SDXL ControlNet training and the other is AMD memory optimizations; unrelated." + }, + { + "left": "issue:1261", + "right": "issue:222", + "accept": false, + "reason": "The xformers backward assert is a specific attention-kernel bug, while the training runtime error report is too generic and not the same issue." + }, + { + "left": "issue:10542", + "right": "issue:11006", + "accept": false, + "reason": "Both are video-model issues, but they involve different pipelines and failure modes (batching vs quantized-transformer output corruption)." + }, + { + "left": "issue:2838", + "right": "issue:8482", + "accept": false, + "reason": "One requests ControlNet pipeline behavior, the other asks for SD3-medium model support; different feature requests." + }, + { + "left": "issue:1614", + "right": "issue:4104", + "accept": false, + "reason": "Both mention black output / casting, but the reported contexts and code paths differ enough that they do not clearly share one root cause." + }, + { + "left": "issue:11127", + "right": "issue:9861", + "accept": false, + "reason": "One is about single-file FLUX loading failure, the other about Flux training not updating the transformer; different stages and likely different fixes." + }, + { + "left": "issue:4316", + "right": "issue:5037", + "accept": true, + "reason": "Both are train_dreambooth.py validation-image failures, with the second validation image option triggering the same kind of error." + }, + { + "left": "issue:1478", + "right": "issue:1614", + "accept": false, + "reason": "One is an upscale pipeline multi-output limitation and the other is a black-image/autocast issue; not the same bug." + }, + { + "left": "issue:2549", + "right": "issue:3420", + "accept": false, + "reason": "These involve different pipelines and failures; the titles do not point to one shared underlying code-path problem." + }, + { + "left": "issue:1672", + "right": "issue:2296", + "accept": false, + "reason": "Image-size conversion errors and xformers memory-efficient attention failures are unrelated." + }, + { + "left": "issue:2904", + "right": "issue:4557", + "accept": false, + "reason": "Different AttributeError contexts in different pipelines; no sign of a shared underlying defect." + }, + { + "left": "issue:5607", + "right": "issue:8667", + "accept": true, + "reason": "The titles describe the same tensor-size-mismatch bug for non-power-of-two images in SD3ControlNetModel." + }, + { + "left": "issue:5757", + "right": "issue:8659", + "accept": false, + "reason": "Both mention checkpoint resume issues, but they are different scripts and 8659 also includes inference-loading problems; not clearly the same bug." + }, + { + "left": "issue:3884", + "right": "issue:6086", + "accept": false, + "reason": "Both concern train_text_to_image_lora.py, but one is a generic error report and the other is a specific FP16 gradient unscale issue; not enough to treat as duplicates." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The three accepted duplicate-like pairs are well supported by the issue titles, and the rejected soft pairs are mostly framed cautiously with reasonable distinctions. Canonical anchoring on issue:6415 is also defensible given the discussion activity and the FaceID/IP-Adapter subcluster." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 54, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10055", + "issue:10640", + "issue:11314", + "issue:11581", + "issue:11585", + "issue:1214", + "issue:1249", + "issue:1552", + "issue:1572", + "issue:1709", + "issue:1867", + "issue:1974", + "issue:1976", + "issue:2044", + "issue:2052", + "issue:2089", + "issue:2296", + "issue:2299", + "issue:2560", + "issue:259", + "issue:3032", + "issue:4125", + "issue:490", + "issue:499", + "issue:5577", + "issue:5603", + "issue:6444", + "issue:6610", + "issue:6741", + "issue:6865", + "issue:7079", + "issue:7197", + "issue:7431", + "issue:746", + "issue:7483", + "issue:7574", + "issue:7708", + "issue:776", + "issue:8317", + "issue:8494", + "issue:8511", + "issue:8534", + "issue:8589", + "issue:859", + "issue:8939", + "issue:9026", + "issue:9398" + ], + "soft_pairs": [ + "issue:10640|issue:7079", + "issue:2044|issue:6741", + "issue:7483|issue:7708", + "issue:8317|issue:9398", + "issue:8494|issue:859", + "issue:1572|issue:1867", + "issue:11581|issue:9026", + "issue:2299|issue:490", + "issue:11585|issue:1976", + "issue:6444|issue:6610", + "issue:259|issue:776", + "issue:1974|issue:2560", + "issue:10055|issue:2052", + "issue:1249|issue:3032", + "issue:5577|issue:7574", + "issue:1214|issue:499", + "issue:4125|issue:7197", + "issue:5603|issue:6865", + "issue:746|issue:8939", + "issue:8511|issue:8589", + "issue:1552|issue:2296", + "issue:1709|issue:3032", + "issue:2089|issue:7431", + "issue:11314|issue:8534" + ], + "prepared_review_unit_hash": "b64a33e49e6f3b3620aebd705cf0f3cde94908131be539fad2acb16d89023b11", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 34234, + "estimated_input_tokens": 8559, + "estimated_eval_tokens": 17374 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose set of training/inference issues around diffusion models; only one pair looks plausibly like the same underlying generator-list reproducibility bug. The rest are distinct failures in different pipelines, models, or error paths.", + "confidence": 0.77, + "canonical_issue_reason": "Issue #10055 is the most direct representative of the cluster\u2019s main theme: an SD3 ControlNet training script not working. It is specific enough to anchor the cluster, while several neighboring issues are narrower variants or unrelated training-script bugs.", + "canonical_pr_reason": null, + "best_issue_reason": "#10055 is the strongest global issue choice because it is explicit about the broken training script and the SD3 ControlNet context, which matches the dominant topic better than the more model-specific or secondary training complaints.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10640", + "right": "issue:7079", + "accept": false, + "reason": "Different bugs: a FluxInpaint/DataParallel multi-GPU error versus a Transformer2DModel patch/norm_type failure." + }, + { + "left": "issue:2044", + "right": "issue:6741", + "accept": false, + "reason": "Unrelated attribute errors in different code paths: DDP dtype access vs missing no_flip argument in a diffusion DPO script." + }, + { + "left": "issue:7483", + "right": "issue:7708", + "accept": false, + "reason": "Both involve SDXL-related loading/conversion, but one is LoRA-to-Diffusers conversion and the other is a tokenizer file lookup failure; not the same concrete bug." + }, + { + "left": "issue:8317", + "right": "issue:9398", + "accept": false, + "reason": "Both are missing-argument runtime errors, but they occur in different classes and call sites, so they do not appear to be the same underlying defect." + }, + { + "left": "issue:8494", + "right": "issue:859", + "accept": false, + "reason": "Different API breakages: SD3 force_hooks/force_hook mismatch versus a generic unexpected device_map constructor argument." + }, + { + "left": "issue:1572", + "right": "issue:1867", + "accept": false, + "reason": "Both are img2img-related, but one is a broad pipeline incompatibility while the other is a low-strength scheduler behavior issue." + }, + { + "left": "issue:11581", + "right": "issue:9026", + "accept": false, + "reason": "Different errors from different layers: IP-Adapter loading in FLUX with a dtype issue versus load_checkpoint_and_dispatch force_hook mismatch." + }, + { + "left": "issue:2299", + "right": "issue:490", + "accept": false, + "reason": "One is about loading fine-tuned checkpoints into Diffusers, the other is a memory-usage comparison issue." + }, + { + "left": "issue:11585", + "right": "issue:1976", + "accept": false, + "reason": "No shared underlying bug: Stable Diffusion Web UI output quality vs an incorrect tokenizer/config path error." + }, + { + "left": "issue:6444", + "right": "issue:6610", + "accept": false, + "reason": "Different training/inference problems: missing peft_config during checkpointing versus num_in_channels being ignored in from_single_file." + }, + { + "left": "issue:259", + "right": "issue:776", + "accept": false, + "reason": "Different issues in different areas: progress bar display versus img2img batch-size behavior." + }, + { + "left": "issue:1974", + "right": "issue:2560", + "accept": true, + "reason": "Both describe the same generator-list reproducibility problem in img2img/scheduler behavior: passing a list of generators yields incorrect or non-deterministic batch outputs." + }, + { + "left": "issue:5603", + "right": "issue:6865", + "accept": false, + "reason": "Different import/utility issues: one is a missing diffusers utility import, the other is a load_image Image-instance validation problem." + }, + { + "left": "issue:746", + "right": "issue:8939", + "accept": false, + "reason": "Both are loading-related, but one is an ONNX revision problem for img2img and the other is a tokenizer file resolution error for SD3." + }, + { + "left": "issue:8511", + "right": "issue:8589", + "accept": false, + "reason": "Both concern SD3 DreamBooth training, but one is about unloading text encoders and the other is prior-preservation support." + }, + { + "left": "issue:1552", + "right": "issue:2296", + "accept": false, + "reason": "Different attention/memory issues in different setups: attention slicing on SD2 versus xformers on T4." + }, + { + "left": "issue:1709", + "right": "issue:3032", + "accept": false, + "reason": "Different attribute errors in different pipelines: ONNX img2img astype misuse versus ControlNet img2img missing from_pretrained." + }, + { + "left": "issue:2089", + "right": "issue:7431", + "accept": false, + "reason": "Both are DreamBooth/training related, but one is OOM on AltDiffusion and the other is a DoRA/text-encoder training problem." + }, + { + "left": "issue:11314", + "right": "issue:8534", + "accept": false, + "reason": "Both are SD/FLUX/SD3 training-script complaints, but they concern different models and different defects: a ControlNet script mistake versus wrong logit-normal weighting." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative on the soft-edge pairs. The only accepted pair, 1974/2560, is plausibly the same generator-list reproducibility bug, and the other negative verdicts do not overstate the evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 55, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10673", + "issue:1082", + "issue:11006", + "issue:11160", + "issue:11839", + "issue:1214", + "issue:12157", + "issue:12433", + "issue:12559", + "issue:12771", + "issue:1407", + "issue:1552", + "issue:1919", + "issue:1949", + "issue:2578", + "issue:2645", + "issue:337", + "issue:3382", + "issue:3432", + "issue:3582", + "issue:4213", + "issue:4386", + "issue:4453", + "issue:4480", + "issue:4580", + "issue:4677", + "issue:487", + "issue:4935", + "issue:5012", + "issue:5227", + "issue:5619", + "issue:5798", + "issue:6280", + "issue:6371", + "issue:6382", + "issue:6964", + "issue:7049", + "issue:7660", + "issue:771", + "issue:7751", + "issue:8320", + "issue:8494", + "issue:8551", + "issue:8668", + "issue:9187", + "issue:9497", + "issue:9726" + ], + "soft_pairs": [ + "issue:1949|issue:7751", + "issue:1552|issue:3582", + "issue:6382|issue:9187", + "issue:2578|issue:6964", + "issue:2578|issue:7660", + "issue:4386|issue:4453", + "issue:10673|issue:11839", + "issue:12433|issue:1919", + "issue:11006|issue:11160", + "issue:12559|issue:3432", + "issue:9497|issue:9726", + "issue:2645|issue:4480", + "issue:4213|issue:5227", + "issue:5798|issue:8494", + "issue:12771|issue:4580", + "issue:6280|issue:8320", + "issue:1407|issue:7049", + "issue:5619|issue:771", + "issue:1214|issue:487", + "issue:1082|issue:3382", + "issue:12157|issue:6371", + "issue:1214|issue:5227", + "issue:337|issue:4677", + "issue:4935|issue:5012", + "issue:8551|issue:8668" + ], + "prepared_review_unit_hash": "6262e8414168cba15c27fda93922fdc4e7b05d132f1652b9aefcf182e67f758d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 34484, + "estimated_input_tokens": 8621, + "estimated_eval_tokens": 17498 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the items cover unrelated bugs in training, pipelines, schedulers, docs, and platform-specific failures. None of the soft pairs look like the same underlying issue/change, so I would not merge any of them as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical issue stands out because the set does not form one duplicate family; the issues span different models, code paths, and failure modes.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a strong global representative for the cluster for the same reason: the topics are too unrelated to anchor on one issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1949", + "right": "issue:7751", + "accept": false, + "reason": "Different problems: Dreambooth tutorial/requirements setup vs an AnimateDiff conversion error." + }, + { + "left": "issue:1552", + "right": "issue:3582", + "accept": false, + "reason": "Both are pipeline behavior issues, but attention slicing and ControlNet inpaint strength are unrelated code paths." + }, + { + "left": "issue:6382", + "right": "issue:9187", + "accept": false, + "reason": "Both mention IP Adapter, but one is an LCM training/use key error and the other is an SDXL config regression after Kolors update." + }, + { + "left": "issue:2578", + "right": "issue:6964", + "accept": false, + "reason": "Editable install problems are unrelated to a ControlNet inpaint bug on 4-channel UNets." + }, + { + "left": "issue:2578", + "right": "issue:7660", + "accept": false, + "reason": "Editable install failure and a trace example failure are distinct issues with no shared code-path evidence." + }, + { + "left": "issue:4386", + "right": "issue:4453", + "accept": false, + "reason": "Both concern LoRA conversion, but one is a general question and the other is a specific A1111 compatibility question." + }, + { + "left": "issue:10673", + "right": "issue:11839", + "accept": false, + "reason": "Both report shape/input-size errors, but they are different pipelines/models (AuraFlow vs FluxKontext gguf)." + }, + { + "left": "issue:12433", + "right": "issue:1919", + "accept": false, + "reason": "Documentation version incompatibility is unrelated to missing datasets in a Dreambooth requirements file." + }, + { + "left": "issue:11006", + "right": "issue:11160", + "accept": false, + "reason": "Same Wan 2.1 family, but quantized-transformer video corruption and inconsistent inference are not the same concrete bug." + }, + { + "left": "issue:12559", + "right": "issue:3432", + "accept": false, + "reason": "Different backend errors: missing torchvision::nms on Windows vs half-precision CPU conv failure." + }, + { + "left": "issue:9497", + "right": "issue:9726", + "accept": false, + "reason": "Both are FLUX Dreambooth reports, but one is a mat2 shape error and the other is a multigpu training problem." + }, + { + "left": "issue:2645", + "right": "issue:4480", + "accept": false, + "reason": "Apple-silicon backend failures, but one is a CUDA-on-M1 assertion and the other is an MPS float16 crash." + }, + { + "left": "issue:4213", + "right": "issue:5227", + "accept": false, + "reason": "From_single_file loading without accelerate and load_lora_weight with torch.compile are unrelated issues." + }, + { + "left": "issue:5798", + "right": "issue:8494", + "accept": false, + "reason": "Both are unexpected-kwarg errors, but they affect different pipelines and different arguments." + }, + { + "left": "issue:12771", + "right": "issue:4580", + "accept": false, + "reason": "Both involve scheduler failures with karras sigmas, but they are different schedulers and failure points." + }, + { + "left": "issue:6280", + "right": "issue:8320", + "accept": false, + "reason": "Both touch from_single_file, but one lacks a pipeline method while the other is an AutoencoderKL type-handling bug." + }, + { + "left": "issue:1407", + "right": "issue:7049", + "accept": false, + "reason": "An ONNX feature_extractor mismatch and an IP Adapter unpacking error are unrelated." + }, + { + "left": "issue:5619", + "right": "issue:771", + "accept": false, + "reason": "LCM img2img autoencoder behavior and BF16 Dreambooth are separate backend/model issues." + }, + { + "left": "issue:1214", + "right": "issue:487", + "accept": false, + "reason": "Half-precision CLIPTextModel and Stable Diffusion on M1 CPU are different precision/backend problems." + }, + { + "left": "issue:1082", + "right": "issue:3382", + "accept": false, + "reason": "CUBLAS matmul execution failure and multi-GPU OOM are distinct Dreambooth training failures." + }, + { + "left": "issue:12157", + "right": "issue:6371", + "accept": false, + "reason": "A bnb quantization config iterable bug in Qwen-Image training is unrelated to a ControlNet XS assert error." + }, + { + "left": "issue:1214", + "right": "issue:5227", + "accept": false, + "reason": "Half-precision CLIPTextModel and torch.compile LoRA loading are unrelated code paths." + }, + { + "left": "issue:337", + "right": "issue:4677", + "accept": false, + "reason": "Both are inpainting-related, but one is a general feature/problem report and the other is a documentation example failure." + }, + { + "left": "issue:4935", + "right": "issue:5012", + "accept": false, + "reason": "Both mention the same LoRA SDXL training script, but one is a usage question and the other is a text-encoder training limitation." + }, + { + "left": "issue:8551", + "right": "issue:8668", + "accept": false, + "reason": "Both involve image latent handling, but one is a device-map transfer bug and the other is a size-alignment failure in SD3 img2img." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 56, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10266", + "issue:10453", + "issue:10591", + "issue:11006", + "issue:12034", + "issue:12547", + "issue:1552", + "issue:1976", + "issue:2091", + "issue:2459", + "issue:2668", + "issue:3354", + "issue:3454", + "issue:3582", + "issue:3613", + "issue:3774", + "issue:4134", + "issue:4137", + "issue:4728", + "issue:4733", + "issue:4880", + "issue:5054", + "issue:5259", + "issue:5395", + "issue:5465", + "issue:5524", + "issue:5619", + "issue:5749", + "issue:5860", + "issue:6492", + "issue:6536", + "issue:6570", + "issue:6598", + "issue:6661", + "issue:6809", + "issue:6815", + "issue:6914", + "issue:746", + "issue:7766", + "issue:8514", + "issue:8833", + "issue:8956", + "issue:9044", + "issue:9112", + "issue:9200", + "issue:945", + "issue:9613" + ], + "soft_pairs": [ + "issue:10266|issue:5749", + "issue:4134|issue:6661", + "issue:3582|issue:5860", + "issue:4137|issue:6598", + "issue:5395|issue:6809", + "issue:6570|issue:9112", + "issue:10453|issue:5465", + "issue:12547|issue:8514", + "issue:2091|issue:4728", + "issue:11006|issue:12034", + "issue:10591|issue:3354", + "issue:1552|issue:5619", + "issue:1976|issue:9044", + "issue:3613|issue:3774", + "issue:5259|issue:8833", + "issue:5054|issue:5524", + "issue:1552|issue:2459", + "issue:2668|issue:4880", + "issue:3454|issue:4733", + "issue:746|issue:8956", + "issue:6492|issue:6815", + "issue:746|issue:945", + "issue:4728|issue:6536", + "issue:6914|issue:7766", + "issue:9200|issue:9613" + ], + "prepared_review_unit_hash": "c697d7a1a594387a6129bbeca017eb9128f0a3e909d97a708883d62bb61bb1d1", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 34341, + "estimated_input_tokens": 8586, + "estimated_eval_tokens": 17428 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly false-positive similarity matches: the set mixes unrelated bugs across model loading, training scripts, ControlNet, video pipelines, and OOM/runtime issues. Only the black-image pair looks like the same underlying defect; the rest should stay separate.", + "confidence": 0.86, + "canonical_issue_reason": "issue:6815 is the most concrete/root-cause report in the only clear duplicate pair (black images caused by invalid value casting). The broader cluster is otherwise too heterogeneous to have a strong single canonical issue.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6815 is the best representative for the one real overlap here because it states the warning and symptom together; it is still only a local best fit, not a true cluster-wide anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10266", + "right": "issue:5749", + "accept": false, + "reason": "Both mention index errors, but UniPC+FlowMatch out-of-bounds is a specific scheduler bug, while 5749 is a generic tensor index OOB with no shared code path." + }, + { + "left": "issue:4134", + "right": "issue:6661", + "accept": false, + "reason": "ControlNet MPS tensor-shape mismatch is a model compatibility bug; 6661 is an OOM on certain image sizes, which is a different failure mode." + }, + { + "left": "issue:3582", + "right": "issue:5860", + "accept": false, + "reason": "These affect different pipelines and symptoms: inpaint strength handling vs Tiny Autoencoder + ControlNet pipeline failure." + }, + { + "left": "issue:4137", + "right": "issue:6598", + "accept": false, + "reason": "Both compare outputs, but one is LoRA-vs-WebUI divergence and the other is img2img mismatch; they do not point to the same concrete bug." + }, + { + "left": "issue:5395", + "right": "issue:6809", + "accept": false, + "reason": "PEFT LoRA merging problems and PEFT adapter loading failures are related, but these describe different operations and likely different bugs." + }, + { + "left": "issue:6570", + "right": "issue:9112", + "accept": false, + "reason": "Different model families and code paths: SDXL InstructPix2Pix tensor mismatch versus Auraflow inference tensor mismatch." + }, + { + "left": "issue:10453", + "right": "issue:5465", + "accept": false, + "reason": "Batch-size limitation in Hunyuan Video and DiffEdit inpaint_strength restriction are unrelated pipeline behaviors." + }, + { + "left": "issue:12547", + "right": "issue:8514", + "accept": false, + "reason": "Missing required CLI argument for Flux Dreambooth fine-tuning is not the same as EMAModel causing parallel fine-tuning errors." + }, + { + "left": "issue:2091", + "right": "issue:4728", + "accept": false, + "reason": "Both involve checkpoint resume, but one is a multi-GPU resume bug and the other is an OOM during resume; not the same underlying issue." + }, + { + "left": "issue:11006", + "right": "issue:12034", + "accept": false, + "reason": "Both are Wan I2V reports, but one is broken output with a quantized transformer and the other is poor quality versus an official space." + }, + { + "left": "issue:10591", + "right": "issue:3354", + "accept": false, + "reason": "These are different LoRA training-script bugs: a general problem in the SD3 script versus an fp16 casting issue in another script." + }, + { + "left": "issue:1552", + "right": "issue:5619", + "accept": false, + "reason": "Attention slicing on SD2 and AutoencodeTiny LCM img2img encode are separate features with unrelated failure points." + }, + { + "left": "issue:1976", + "right": "issue:9044", + "accept": false, + "reason": "Both are model/file lookup errors, but they reference different missing files, different models, and likely different packaging issues." + }, + { + "left": "issue:3613", + "right": "issue:3774", + "accept": false, + "reason": "Both mention training memory, but one is Dreambooth OOM on torch 2.0 and the other is zero3 not saving memory in text-to-image training." + }, + { + "left": "issue:5259", + "right": "issue:8833", + "accept": false, + "reason": "Missing model_index.json versus missing config.json are similar loading symptoms, but they are not the same artifact or code path." + }, + { + "left": "issue:5054", + "right": "issue:5524", + "accept": false, + "reason": "Missing final_offload_hook and missing lora_layer are distinct attribute errors in different subsystems." + }, + { + "left": "issue:1552", + "right": "issue:2459", + "accept": false, + "reason": "Both involve attention optimizations, but attention slicing and xformers memory-efficient attention failing are not the same bug." + }, + { + "left": "issue:2668", + "right": "issue:4880", + "accept": false, + "reason": "Both are ControlNet-related, but one misroutes ControlNet image input and the other raises a validation TypeError for missing/invalid image input." + }, + { + "left": "issue:3454", + "right": "issue:4733", + "accept": false, + "reason": "These are different training-script failures: one says the script does not run on latest, the other is a checkpoint-resume optimizer state error." + }, + { + "left": "issue:746", + "right": "issue:8956", + "accept": false, + "reason": "These are different load failures in different backends/models: ONNX img2img revision loading versus OpenVINO StableDiffusionXL pipeline model loading." + }, + { + "left": "issue:6492", + "right": "issue:6815", + "accept": true, + "reason": "Both describe black-image output, and 6815 adds the likely root cause: invalid values during image cast to uint8. This looks like the same bug." + }, + { + "left": "issue:746", + "right": "issue:945", + "accept": false, + "reason": "Both concern loading/ONNX-related behavior, but they target different model-conversion scenarios and do not clearly share one bug." + }, + { + "left": "issue:4728", + "right": "issue:6536", + "accept": false, + "reason": "Both are training/OOM reports, but one is a checkpoint-resume OOM and the other is generic fine-tuning out-of-memory for UNet3DConditionModel." + }, + { + "left": "issue:6914", + "right": "issue:7766", + "accept": false, + "reason": "AttributeError from enable_model_cpu_offload breaking IP-Adapter-FaceID is a concrete bug; 7766 is a usage/question thread, not the same issue." + }, + { + "left": "issue:9200", + "right": "issue:9613", + "accept": false, + "reason": "These are both 0.30.x compatibility complaints, but one is a ControlNet dtype error and the other is a Timesteps.scale attribute error; different failures." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted overlap, issue:6492 with issue:6815, is well supported by the titles, and the other pair rejections do not overclaim shared bugs beyond vague surface similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 57, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10150", + "issue:10180", + "issue:10421", + "issue:10712", + "issue:1100", + "issue:11148", + "issue:11865", + "issue:12535", + "issue:1310", + "issue:1709", + "issue:2132", + "issue:2459", + "issue:2578", + "issue:2904", + "issue:2947", + "issue:3032", + "issue:3565", + "issue:3582", + "issue:3661", + "issue:4086", + "issue:4103", + "issue:4255", + "issue:4368", + "issue:4648", + "issue:4701", + "issue:4923", + "issue:5125", + "issue:5474", + "issue:5524", + "issue:5794", + "issue:5884", + "issue:5897", + "issue:6028", + "issue:6218", + "issue:6610", + "issue:696", + "issue:7053", + "issue:7064", + "issue:720", + "issue:7293", + "issue:7357", + "issue:7432", + "issue:746", + "issue:7799", + "issue:7854", + "issue:9051", + "issue:9643", + "issue:9673" + ], + "soft_pairs": [ + "issue:1709|issue:2904", + "issue:4648|issue:5794", + "issue:11148|issue:12535", + "issue:5125|issue:5897", + "issue:4923|issue:7293", + "issue:3565|issue:4368", + "issue:4255|issue:7432", + "issue:5884|issue:7064", + "issue:3661|issue:4086", + "issue:10150|issue:7357", + "issue:10712|issue:7799", + "issue:10421|issue:9673", + "issue:1310|issue:5474", + "issue:720|issue:7854", + "issue:2578|issue:6218", + "issue:9051|issue:9643", + "issue:11865|issue:6610", + "issue:2947|issue:5524", + "issue:6028|issue:696", + "issue:2459|issue:3582", + "issue:4701|issue:5897", + "issue:10180|issue:746", + "issue:1100|issue:3032", + "issue:2132|issue:3032", + "issue:4103|issue:7053" + ], + "prepared_review_unit_hash": "0e30def0ff333ba406688f1a9a7a8b778d1c095d07c490610778bbf8b72d2dac", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34809, + "estimated_input_tokens": 8703, + "estimated_eval_tokens": 17662 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of false-positive soft similarities across unrelated issues. The candidates share broad themes like pipelines, LoRAs, OOMs, or missing attributes, but not the same concrete bug or fix path. No PRs are present.", + "confidence": 0.93, + "canonical_issue_reason": "issue:10150 is the broadest umbrella-style report in the set: SDXL pipeline loading/inference slowness. It is the best representative issue if this cluster needs a single canonical issue, even though it is not a true duplicate of the others.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10150 is the most general and central issue among the listed items, covering the broad performance theme better than the more specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1709", + "right": "issue:2904", + "accept": false, + "reason": "Both are pipeline errors, but one is an ONNX Tensor.astype mismatch and the other is an AudioLDM scheduler NoneType error; different code paths." + }, + { + "left": "issue:4648", + "right": "issue:5794", + "accept": false, + "reason": "Both mention missing config files during loading, but they affect different models/artifacts and are not the same load failure." + }, + { + "left": "issue:11148", + "right": "issue:12535", + "accept": false, + "reason": "Both involve LoRA/adapter loading, but one is a PEFT state-flag bug and the other is a Wan2.2 Lightning LoRA loading issue." + }, + { + "left": "issue:5125", + "right": "issue:5897", + "accept": false, + "reason": "Both are runtime/CUDA-related, but one is an xformers kernel failure and the other is a device-placement mismatch during LoRA training." + }, + { + "left": "issue:4923", + "right": "issue:7293", + "accept": false, + "reason": "Both are AudioLDM feature requests, but they ask for different extensions and do not describe the same change." + }, + { + "left": "issue:3565", + "right": "issue:4368", + "accept": false, + "reason": "Both concern OOM during training, but they are different training setups and not the same underlying memory bug." + }, + { + "left": "issue:4255", + "right": "issue:7432", + "accept": false, + "reason": "Both compare Diffusers to WebUI quality, but one is a broad quality complaint and the other is about a specific image similarity workflow." + }, + { + "left": "issue:5884", + "right": "issue:7064", + "accept": false, + "reason": "Both are IP-Adapter support items, but one is a broad add-support request and the other is a narrow fix for SAG/panorama pipelines." + }, + { + "left": "issue:3661", + "right": "issue:4086", + "accept": false, + "reason": "Both involve model loading/training, but one is about from_ckpt on SD2.x weights and the other is about ControlNet training on SDXL 0.9." + }, + { + "left": "issue:10150", + "right": "issue:7357", + "accept": false, + "reason": "Both are SDXL performance complaints, but one is general slow loading/inference and the other is specifically slow inference despite compile/fuse_loras; not the same bug." + }, + { + "left": "issue:10712", + "right": "issue:7799", + "accept": false, + "reason": "Both report runtime errors, but one is about prompt_embeds in SD3 and the other is a matmul size mismatch with IP-Adapter in InstructPix2Pix." + }, + { + "left": "issue:10421", + "right": "issue:9673", + "accept": false, + "reason": "Both mention multi-GPU/multi-LoRA overhead, but one is a CPU memory leak when moving pipelines and the other is high CPU usage while loading multiple LoRAs." + }, + { + "left": "issue:1310", + "right": "issue:5474", + "accept": false, + "reason": "These are different pipeline ideas: one is K-Diffusion + Stable Diffusion, the other is a new reference-only pipeline feature request." + }, + { + "left": "issue:720", + "right": "issue:7854", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different contexts and describe different failing operations." + }, + { + "left": "issue:2578", + "right": "issue:6218", + "accept": false, + "reason": "One is editable-install breakage and the other is a batch-seed pipeline bug; unrelated failures." + }, + { + "left": "issue:9051", + "right": "issue:9643", + "accept": false, + "reason": "Both concern Flux limitations, but one is batch inference and the other is multiple ControlNets; separate feature gaps." + }, + { + "left": "issue:11865", + "right": "issue:6610", + "accept": false, + "reason": "Both involve missing functionality around single-file loading, but one is a missing from_single_file attribute and the other is num_in_channels being ignored." + }, + { + "left": "issue:2947", + "right": "issue:5524", + "accept": false, + "reason": "Both relate to attention/LoRA loading, but one is a missing load_attn_procs attribute and the other is a missing lora_layer attribute on Linear." + }, + { + "left": "issue:6028", + "right": "issue:696", + "accept": false, + "reason": "Both are OOM reports, but they are separate models/workflows and not the same memory issue." + }, + { + "left": "issue:2459", + "right": "issue:3582", + "accept": false, + "reason": "Both say something doesn't work in training/inpaint workflows, but they are unrelated features with different failure modes." + }, + { + "left": "issue:4701", + "right": "issue:5897", + "accept": false, + "reason": "Both are CUDA/device-type problems, but one is a decoder backprop error at float32 and the other is a cross-device tensor placement issue in LoRA training." + }, + { + "left": "issue:10180", + "right": "issue:746", + "accept": false, + "reason": "Both mention loading behavior, but one is about multiple LoRAs with Flux Control LoRA and the other is an ONNX revision loading problem." + }, + { + "left": "issue:1100", + "right": "issue:3032", + "accept": false, + "reason": "Both are attribute errors in pipeline components, but they involve different missing attributes and different code paths." + }, + { + "left": "issue:2132", + "right": "issue:3032", + "accept": false, + "reason": "Both are attribute errors in examples/pipelines, but one is Accelerator.trackers and the other is a pipeline class missing from_pretrained." + }, + { + "left": "issue:4103", + "right": "issue:7053", + "accept": false, + "reason": "Both are model-behavior issues, but one is AutoencoderKL scaling_factor handling and the other is UNet3D class conditioning; unrelated fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 58, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10180", + "issue:10518", + "issue:10612", + "issue:1125", + "issue:12141", + "issue:1263", + "issue:1363", + "issue:1896", + "issue:2363", + "issue:2851", + "issue:2941", + "issue:2991", + "issue:3221", + "issue:3223", + "issue:3700", + "issue:3802", + "issue:4213", + "issue:4307", + "issue:4368", + "issue:4566", + "issue:4728", + "issue:499", + "issue:539", + "issue:5798", + "issue:5897", + "issue:5917", + "issue:6028", + "issue:6094", + "issue:6176", + "issue:6218", + "issue:6689", + "issue:6956", + "issue:7057", + "issue:7146", + "issue:7147", + "issue:7263", + "issue:7418", + "issue:7483", + "issue:7589", + "issue:7591", + "issue:7666", + "issue:7728", + "issue:7931", + "issue:8228", + "issue:8359", + "issue:8956", + "issue:9527", + "issue:9953" + ], + "soft_pairs": [ + "issue:3802|issue:7057", + "issue:4368|issue:4566", + "issue:4728|issue:7146", + "issue:10518|issue:7418", + "issue:1363|issue:6176", + "issue:2941|issue:3223", + "issue:1263|issue:7147", + "issue:10180|issue:8956", + "issue:10612|issue:9953", + "issue:5897|issue:7728", + "issue:12141|issue:7666", + "issue:7263|issue:7931", + "issue:4307|issue:7591", + "issue:6028|issue:6956", + "issue:1125|issue:1896", + "issue:499|issue:7589", + "issue:4213|issue:6218", + "issue:539|issue:5798", + "issue:2363|issue:6689", + "issue:8228|issue:8359", + "issue:2851|issue:3700", + "issue:3221|issue:7483", + "issue:5917|issue:6094", + "issue:2991|issue:9527" + ], + "prepared_review_unit_hash": "7e0ffae3caefff71bff55d6a6a26c70661f3cbf75cb76248f4a87075435de566", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34291, + "estimated_input_tokens": 8573, + "estimated_eval_tokens": 17402 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All proposed soft edges look like false positives. The items span different pipelines, scripts, versions, and failure modes, so there is no clear duplicate cluster or mergeable pair.", + "confidence": 0.96, + "canonical_issue_reason": "No clear canonical issue: this set is heterogeneous and does not center on one underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10180 is the closest thing to an anchor because it is a concrete, well-scoped Flux LoRA loading bug, but it still does not represent the rest of the set as duplicates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3802", + "right": "issue:7057", + "accept": false, + "reason": "Both mention checkpoints, but one is about a training argument not working and the other is a vague model checkpoint failure; different bugs." + }, + { + "left": "issue:4368", + "right": "issue:4566", + "accept": false, + "reason": "Both are SDXL LoRA training issues, but one is CUDA OOM and the other is checkpoint resume failure; not the same defect." + }, + { + "left": "issue:4728", + "right": "issue:7146", + "accept": false, + "reason": "Both involve OOM during SDXL training, but the contexts differ: resume_from_checkpoint on UNet fine-tuning vs a general SDXL fine-tuning OOM." + }, + { + "left": "issue:10518", + "right": "issue:7418", + "accept": false, + "reason": "Unrelated: one is a script problem in sd3_lora_colab, the other is a bug in a community upscale pipeline line." + }, + { + "left": "issue:1363", + "right": "issue:6176", + "accept": false, + "reason": "Different domains: ONNX scheduler error versus LoRA-weight loading after dreambooth training." + }, + { + "left": "issue:2941", + "right": "issue:3223", + "accept": false, + "reason": "Different failure modes: scheduler output quality issue versus RAM not being cleared after moving a pipeline to CUDA." + }, + { + "left": "issue:1263", + "right": "issue:7147", + "accept": false, + "reason": "Both are import errors, but they refer to different pipeline classes and different missing symbols." + }, + { + "left": "issue:10180", + "right": "issue:8956", + "accept": false, + "reason": "Different models and symptoms: Flux Control LoRA loading versus OVStableDiffusionXLPipeline model loading." + }, + { + "left": "issue:10612", + "right": "issue:9953", + "accept": false, + "reason": "Both involve quantization, but one is loading a LoRA into a quantized Flux Fill pipeline and the other is moving a pipeline with a quantized component to CUDA." + }, + { + "left": "issue:5897", + "right": "issue:7728", + "accept": false, + "reason": "Both are training-time errors, but the scripts and concrete device mismatch contexts differ; not the same bug." + }, + { + "left": "issue:12141", + "right": "issue:7666", + "accept": false, + "reason": "Unrelated: WanVACE VAE dtype incompatibility versus from_single_file custom pipeline loading." + }, + { + "left": "issue:7263", + "right": "issue:7931", + "accept": false, + "reason": "Both involve IP-Adapter, but they hit different pipelines and different parameter paths, so they are not the same issue." + }, + { + "left": "issue:4307", + "right": "issue:7591", + "accept": false, + "reason": "Both are loading-related, but one is from_pretrained(local_files_only=True) and the other is from_single_file(local_files_only=True); different code paths and errors." + }, + { + "left": "issue:6028", + "right": "issue:6956", + "accept": false, + "reason": "Both are CUDA OOM reports, but for different pipelines and environments; too broad to be duplicates." + }, + { + "left": "issue:1125", + "right": "issue:1896", + "accept": false, + "reason": "Different pipelines and different runtime errors; no shared underlying bug." + }, + { + "left": "issue:499", + "right": "issue:7589", + "accept": false, + "reason": "Different examples and different bug classes; not the same underlying change or defect." + }, + { + "left": "issue:4213", + "right": "issue:6218", + "accept": false, + "reason": "Different problems: from_single_file dependency behavior versus batched seeds handling in ControlNet/Img2Img." + }, + { + "left": "issue:539", + "right": "issue:5798", + "accept": false, + "reason": "Both are unexpected keyword-argument errors, but they concern different APIs and call sites." + }, + { + "left": "issue:2363", + "right": "issue:6689", + "accept": false, + "reason": "One is a general question about converting LoRA files; the other is a specific SSD-1B conversion failure." + }, + { + "left": "issue:8228", + "right": "issue:8359", + "accept": false, + "reason": "Both are from_single_file issues, but one is a specific CLIPTextModel load failure while the other is a broader custom-checkpoint incompatibility." + }, + { + "left": "issue:2851", + "right": "issue:3700", + "accept": false, + "reason": "Completely different training features: push_to_hub on TPU versus min_snr argument support." + }, + { + "left": "issue:3221", + "right": "issue:7483", + "accept": false, + "reason": "Different workflows: load_attn_procs regression versus LoRA-to-Diffusers conversion for SDXL-based models." + }, + { + "left": "issue:5917", + "right": "issue:6094", + "accept": false, + "reason": "Both are version regressions, but they affect different conversion/pipeline code paths and are not the same bug." + }, + { + "left": "issue:2991", + "right": "issue:9527", + "accept": false, + "reason": "Related to ControlNet precision, but one is a mixed-precision example failure and the other is an explicit dtype mismatch between controlnet fp32 and main pipe bf16; not identical." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 59, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10266", + "issue:10290", + "issue:10453", + "issue:10512", + "issue:10575", + "issue:1398", + "issue:1544", + "issue:1611", + "issue:170", + "issue:1717", + "issue:1925", + "issue:1931", + "issue:2163", + "issue:3015", + "issue:3537", + "issue:3585", + "issue:3802", + "issue:4213", + "issue:4368", + "issue:5128", + "issue:5465", + "issue:5798", + "issue:5846", + "issue:5857", + "issue:6092", + "issue:6159", + "issue:6218", + "issue:6230", + "issue:6382", + "issue:6602", + "issue:6741", + "issue:6858", + "issue:7229", + "issue:7263", + "issue:7483", + "issue:776", + "issue:7773", + "issue:7985", + "issue:8686", + "issue:9051", + "issue:9115", + "issue:9393", + "issue:9514", + "issue:9667", + "issue:9907", + "issue:993" + ], + "soft_pairs": [ + "issue:4213|issue:5857", + "issue:5465|issue:9051", + "issue:6382|issue:6602", + "issue:5128|issue:9667", + "issue:10512|issue:9514", + "issue:1931|issue:993", + "issue:776|issue:7773", + "issue:7263|issue:9115", + "issue:170|issue:5798", + "issue:6218|issue:8686", + "issue:10026|issue:9907", + "issue:1717|issue:7483", + "issue:3015|issue:3537", + "issue:4368|issue:6230", + "issue:6858|issue:7229", + "issue:10026|issue:3802", + "issue:1611|issue:3585", + "issue:6092|issue:9393", + "issue:2163|issue:5846", + "issue:1398|issue:1925", + "issue:1544|issue:6741", + "issue:10453|issue:6159", + "issue:10290|issue:7985", + "issue:10266|issue:10575" + ], + "prepared_review_unit_hash": "1111e93d49dff33abe159346e6c22eb2a47ea48c69f62b3de36c33462537253e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33671, + "estimated_input_tokens": 8418, + "estimated_eval_tokens": 17092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous set of Diffusers issues; the proposed soft pairs are mostly superficial title matches rather than true duplicates. I do not see a single underlying bug/change that unifies the cluster.", + "confidence": 0.96, + "canonical_issue_reason": "No canonical issue: the items span distinct bugs and features across Flux, ControlNet, schedulers, LoRA loading, batching, dtype, and conversion scripts. The soft pairs only share broad subsystem words, not the same concrete failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a good global representative because there is no single root cause or code path shared by the cluster. If forced, any choice would misrepresent several unrelated reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4213", + "right": "issue:5857", + "accept": false, + "reason": "Both mention SDXL/LoRA-ish workflows, but one is about from_single_file() needing accelerate and the other about LCM LoRA behavior in AnimateDiff SDXL. Different failures and code paths." + }, + { + "left": "issue:5465", + "right": "issue:9051", + "accept": false, + "reason": "Both are batching-related pipeline bugs, but DiffEdit inpaint_strength handling and FluxPipeline batch inference are different concrete issues." + }, + { + "left": "issue:6382", + "right": "issue:6602", + "accept": false, + "reason": "Both involve LoRA/IP-adapter-style model loading errors, but one is an IP adapter key mismatch after training and the other is loading a saved model after fuse_lora(). Not the same bug." + }, + { + "left": "issue:5128", + "right": "issue:9667", + "accept": false, + "reason": "Both concern loading Flux/XLPipeline local weights, but one is a safetensors load error and the other is a Flux fp8 transformer file issue. Different symptom and likely fix." + }, + { + "left": "issue:10512", + "right": "issue:9514", + "accept": false, + "reason": "LoRA loading in Quanto/Flux vs loading LoRA after fp8 quantization are related only at a high level; the quantization stack and failure modes differ." + }, + { + "left": "issue:1931", + "right": "issue:993", + "accept": false, + "reason": "Both are dtype/float16 mismatch reports, but one is Stable Diffusion 2 half precision and the other is a generic fp16 mismatch. Too broad to merge as one bug." + }, + { + "left": "issue:776", + "right": "issue:7773", + "accept": false, + "reason": "Both are batch-size>1 regressions, but one is image-to-image and the other is Stable Video Diffusion generation. Different pipelines and likely causes." + }, + { + "left": "issue:7263", + "right": "issue:9115", + "accept": false, + "reason": "Both mention IP Adapter/AnimateDiff-ish features, but one is an attention_slicing incompatibility and the other is SDXL motion adapter usage. Not the same concrete issue." + }, + { + "left": "issue:170", + "right": "issue:5798", + "accept": false, + "reason": "Both are callback/step-argument compatibility problems, but one is legacy eta support and the other is callback_on_step_end on ControlNet. Different API breakages." + }, + { + "left": "issue:6218", + "right": "issue:8686", + "accept": false, + "reason": "Both involve SDXL ControlNet pipeline behavior, but one is batched seeds and the other is SDXLCFGCutoffCallback. Different feature bugs, not one root cause." + }, + { + "left": "issue:10026", + "right": "issue:9907", + "accept": false, + "reason": "Both mention Flux ControlNet training, but one is DeepSpeed memory not reducing and the other is maximum sequence length warnings. Different runtime problems." + }, + { + "left": "issue:1717", + "right": "issue:7483", + "accept": false, + "reason": "Offline model caching vs LoRA-to-Diffusers conversion on SDXL 1.0 are unrelated." + }, + { + "left": "issue:3015", + "right": "issue:3537", + "accept": false, + "reason": "These are both ControlNet conversion requests, but one is a diffusers-to-CompVis/WebUI conversion and the other asks for a generic conversion script. Overlapping topic, not the same change." + }, + { + "left": "issue:4368", + "right": "issue:6230", + "accept": false, + "reason": "Both are OOM reports during training, but one is SDXL LoRA training and the other is train_text_to_image_sdxl.py. Generic memory pressure, not a shared bug." + }, + { + "left": "issue:6858", + "right": "issue:7229", + "accept": false, + "reason": "ControlNet training attribute error versus checkpoint-saving issue in advanced training. Different code paths and failures." + }, + { + "left": "issue:10026", + "right": "issue:3802", + "accept": false, + "reason": "Flux ControlNet DeepSpeed memory issue and --checkpoints_total_limit not working are unrelated training problems." + }, + { + "left": "issue:1611", + "right": "issue:3585", + "accept": false, + "reason": "Both are model-loading/training failures in SD pipelines, but one is a pipeline regression with v1-5 and the other is a LoRA synthesis/use failure. No common root cause shown." + }, + { + "left": "issue:6092", + "right": "issue:9393", + "accept": false, + "reason": "Both are LoRA training problems, but one says the LoRA example does not train anything while the other says a Flux dreambooth LoRA cannot save. Different failures." + }, + { + "left": "issue:2163", + "right": "issue:5846", + "accept": false, + "reason": "Both involve training scripts and errors, but one is UNet dtype mismatch during text-to-image training and the other is a resume_from_checkpoint issue in SDXL LoRA. Not the same bug." + }, + { + "left": "issue:1398", + "right": "issue:1925", + "accept": false, + "reason": "Both relate to Stable Diffusion 2 on Apple Silicon/half precision, but one is MPS support and the other is inpainting fp16 behavior. Too broad to treat as duplicates." + }, + { + "left": "issue:1544", + "right": "issue:6741", + "accept": false, + "reason": "Scheduler API attribute error vs Diffusion DPO script missing no_flip argument. Different script/version issues." + }, + { + "left": "issue:10453", + "right": "issue:6159", + "accept": false, + "reason": "Both are pipeline support gaps, but one is Hunyuan Video batch size >1 and the other is SDXL ControlNet XSPipeline denoising_end support. Different pipelines and fixes." + }, + { + "left": "issue:10290", + "right": "issue:7985", + "accept": false, + "reason": "Both mention torch.compile/Flux-related performance, but one is Flux ControlNet compilation failure and the other is IPAdapter compilation incompatibility. Not the same code path." + }, + { + "left": "issue:10266", + "right": "issue:10575", + "accept": false, + "reason": "Both are index-out-of-bounds style errors, but one is UniPC with FlowMatch and the other is DPMSolverMultistepScheduler. Different scheduler bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 60, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10470", + "issue:10612", + "issue:10743", + "issue:10989", + "issue:11127", + "issue:1214", + "issue:1925", + "issue:1946", + "issue:2288", + "issue:2363", + "issue:2459", + "issue:2578", + "issue:2806", + "issue:2856", + "issue:2904", + "issue:3124", + "issue:3224", + "issue:3420", + "issue:3631", + "issue:3651", + "issue:3803", + "issue:4101", + "issue:4134", + "issue:4405", + "issue:4880", + "issue:5226", + "issue:5621", + "issue:5963", + "issue:6069", + "issue:6093", + "issue:6548", + "issue:7056", + "issue:7057", + "issue:7082", + "issue:7146", + "issue:7589", + "issue:7891", + "issue:817", + "issue:8613", + "issue:8686", + "issue:8826", + "issue:9055", + "issue:9089", + "issue:9092", + "issue:9450", + "issue:9514", + "issue:9803" + ], + "soft_pairs": [ + "issue:3124|issue:7057", + "issue:3631|issue:6548", + "issue:7056|issue:7146", + "issue:2856|issue:4101", + "issue:3651|issue:3803", + "issue:1214|issue:2578", + "issue:7082|issue:9055", + "issue:1946|issue:8613", + "issue:10026|issue:10743", + "issue:5226|issue:8686", + "issue:2363|issue:2856", + "issue:2578|issue:7589", + "issue:10470|issue:11127", + "issue:10612|issue:9803", + "issue:2806|issue:2904", + "issue:4134|issue:5621", + "issue:10989|issue:9514", + "issue:2288|issue:4405", + "issue:3224|issue:6093", + "issue:9089|issue:9092", + "issue:7891|issue:9450", + "issue:5963|issue:6069", + "issue:4880|issue:8826", + "issue:1925|issue:817", + "issue:2459|issue:3420" + ], + "prepared_review_unit_hash": "0112d456267b49e7634e42d2fdc593744abf10f804715932271750f4e5faa6b7", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34340, + "estimated_input_tokens": 8585, + "estimated_eval_tokens": 17426 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly a loose Flux/diffusers similarity cluster rather than true duplicates. The best representative is the broad Flux LoRA loading failure report; most other issues are about different pipelines, precision modes, or unrelated training/conversion errors.", + "confidence": 0.72, + "canonical_issue_reason": "issue:9803 is the broadest and most representative Flux LoRA-loading failure report, and it aligns with the main recurring theme across the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9803 is the best global issue representative because it is open, recent, and phrased broadly enough to cover the quantized/Flux LoRA-loading variants without overfitting to one subcase.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3124", + "right": "issue:7057", + "accept": false, + "reason": "Both mention model conversion/loading failures, but one is a generic Stable Diffusion model conversion problem and the other is a checkpoint-specific issue; not the same bug." + }, + { + "left": "issue:3631", + "right": "issue:6548", + "accept": false, + "reason": "Both involve latents, but one is a generic inpainting exception and the other is a specific RGBA image-handling bug in prepare_latents; different failure paths." + }, + { + "left": "issue:7056", + "right": "issue:7146", + "accept": false, + "reason": "One is a LoRA fine-tuning/train_text_encoder error and the other is an SDXL OOM report; same area, different underlying problem." + }, + { + "left": "issue:2856", + "right": "issue:4101", + "accept": false, + "reason": "Both are conversion-related, but they target different model types and conversion flows (SD vs ControlNet)." + }, + { + "left": "issue:3651", + "right": "issue:3803", + "accept": false, + "reason": "Different multi-GPU training scripts and different reported symptoms; not the same concrete bug." + }, + { + "left": "issue:1214", + "right": "issue:2578", + "accept": false, + "reason": "Half-precision CLIP behavior and editable-install failure are unrelated." + }, + { + "left": "issue:7082", + "right": "issue:9055", + "accept": false, + "reason": "Both are import errors, but for different symbols and different modules; unrelated failures." + }, + { + "left": "issue:1946", + "right": "issue:8613", + "accept": false, + "reason": "CUDA/xformers runtime error versus data-column-driven OOM; no shared bug." + }, + { + "left": "issue:10026", + "right": "issue:10743", + "accept": false, + "reason": "Stage-3 memory reduction bug versus a request to support zero-3; related topic, but not the same change or defect." + }, + { + "left": "issue:5226", + "right": "issue:8686", + "accept": false, + "reason": "Different pipelines and different feature interactions; not the same underlying issue." + }, + { + "left": "issue:2363", + "right": "issue:2856", + "accept": false, + "reason": "Both concern conversion, but LoRA conversion and Stable Diffusion model conversion are distinct tasks." + }, + { + "left": "issue:2578", + "right": "issue:7589", + "accept": false, + "reason": "Editable install failure and GLIGEN pipeline failure are unrelated." + }, + { + "left": "issue:10470", + "right": "issue:11127", + "accept": false, + "reason": "TorchAO inference and a single-file Flux model loading problem are different bugs." + }, + { + "left": "issue:10612", + "right": "issue:9803", + "accept": false, + "reason": "Both are Flux LoRA-loading related, but 10612 is specifically NF4 quantized Flux Fill pipeline loading; too specific to safely treat as the same bug as the broad general failure report." + }, + { + "left": "issue:2806", + "right": "issue:2904", + "accept": false, + "reason": "Pipeline-printing customization and an AudioLDM scheduler NoneType error are unrelated." + }, + { + "left": "issue:4134", + "right": "issue:5621", + "accept": false, + "reason": "Both involve ControlNet, but one is an MPS tensor mismatch and the other is encoder_hid_proj support; different issues." + }, + { + "left": "issue:10989", + "right": "issue:9514", + "accept": true, + "reason": "Both describe Flux LoRA loading failing under quantization, and they plausibly share the same loader/code-path defect." + }, + { + "left": "issue:2288", + "right": "issue:4405", + "accept": false, + "reason": "These are both example-script requests, but they are for different training scripts and different model families." + }, + { + "left": "issue:3224", + "right": "issue:6093", + "accept": false, + "reason": "Learning-rate scaling duplication and push_to_hub repo-creation behavior are unrelated training-script issues." + }, + { + "left": "issue:9089", + "right": "issue:9092", + "accept": false, + "reason": "Both mention FluxPipeline, but one is missing the class while the other lacks model_index.json; different setup/load failures." + }, + { + "left": "issue:7891", + "right": "issue:9450", + "accept": false, + "reason": "Both are device-mismatch errors, but the specific contexts differ enough that they are not clearly the same bug." + }, + { + "left": "issue:5963", + "right": "issue:6069", + "accept": false, + "reason": "Kandinsky embed/prompt handling and a scheduler add_noise edge case are unrelated." + }, + { + "left": "issue:4880", + "right": "issue:8826", + "accept": false, + "reason": "ControlNet image input type error versus empty torch.cat tensor list are different runtime failures." + }, + { + "left": "issue:1925", + "right": "issue:817", + "accept": false, + "reason": "Both mention half precision, but they are different pipelines/scripts and not the same defect." + }, + { + "left": "issue:2459", + "right": "issue:3420", + "accept": false, + "reason": "xformers attention setup in a LoRA training script and a ControlNet inpaint pipeline compatibility issue are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The chosen representative issue is broad enough for the Flux LoRA-loading subset, and the single accepted soft pair is phrased cautiously as a plausible shared loader/code-path defect rather than a firm duplicate claim. No overstatement is evident." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 61, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10226", + "issue:10472", + "issue:10591", + "issue:10612", + "issue:11014", + "issue:11046", + "issue:11374", + "issue:11581", + "issue:12529", + "issue:12666", + "issue:13258", + "issue:2044", + "issue:2518", + "issue:2669", + "issue:2891", + "issue:3224", + "issue:3325", + "issue:3645", + "issue:4061", + "issue:4368", + "issue:4419", + "issue:4756", + "issue:4868", + "issue:4932", + "issue:5059", + "issue:5224", + "issue:5226", + "issue:5466", + "issue:5524", + "issue:5577", + "issue:5894", + "issue:5992", + "issue:6082", + "issue:6112", + "issue:6158", + "issue:6427", + "issue:6758", + "issue:7293", + "issue:7690", + "issue:771", + "issue:7891", + "issue:8465", + "issue:8614", + "issue:8708", + "issue:8726", + "issue:8956", + "issue:9505", + "issue:9667" + ], + "soft_pairs": [ + "issue:4419|issue:6112", + "issue:2044|issue:5992", + "issue:2669|issue:5992", + "issue:10472|issue:5224", + "issue:11374|issue:6158", + "issue:12666|issue:13258", + "issue:4061|issue:5466", + "issue:4932|issue:771", + "issue:5226|issue:6427", + "issue:7690|issue:8956", + "issue:8708|issue:8726", + "issue:10226|issue:12529", + "issue:11046|issue:11581", + "issue:5577|issue:9505", + "issue:2891|issue:6758", + "issue:8465|issue:8614", + "issue:6158|issue:7293", + "issue:3325|issue:5894", + "issue:6082|issue:7891", + "issue:4868|issue:5524", + "issue:11014|issue:4368", + "issue:10591|issue:3224", + "issue:2518|issue:4756", + "issue:10612|issue:9667", + "issue:3645|issue:5059" + ], + "prepared_review_unit_hash": "eebd36e65174c35c6fed23c8e7d4c7f74ef3f99b72702aa8c49eb176a47a79b3", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34423, + "estimated_input_tokens": 8606, + "estimated_eval_tokens": 17468 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft matches are mostly false positives: they share broad LoRA/pipeline/error vocabulary, but the concrete bugs, code paths, and model types differ. No pair looks safe to collapse as a duplicate or mergeable PR analogue.", + "confidence": 0.9, + "canonical_issue_reason": "No true canonical duplicate stands out; the set is too heterogeneous. If one issue must anchor the cluster, issue 11581 is the clearest self-contained bug report with a specific traceback and actionable scope.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11581 is the strongest representative because it has a precise failure mode, explicit exception text, and a focused Flux/IP-Adapter code path, making it the most useful single issue in this set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4419", + "right": "issue:6112", + "accept": false, + "reason": "Both mention SDXL output problems, but one is about offset noise not being used at inference and the other is generic turbo noise; different bug mechanisms." + }, + { + "left": "issue:2044", + "right": "issue:5992", + "accept": false, + "reason": "Both are AttributeError-type failures, but one is DDP lacking dtype and the other is a bool object lacking __module__; unrelated code paths." + }, + { + "left": "issue:2669", + "right": "issue:5992", + "accept": false, + "reason": "Different exception targets and symptoms; NoneType.replace versus bool.__module__ indicate separate defects." + }, + { + "left": "issue:10472", + "right": "issue:5224", + "accept": false, + "reason": "Both involve LoRA loading, but one is lora_bias kwarg handling and the other is StableDiffusionXLAdapterPipeline load_lora_weights shape behavior; not the same bug." + }, + { + "left": "issue:11374", + "right": "issue:6158", + "accept": false, + "reason": "Feature requests for different models/pipelines (SkyReels-V2 vs animatediff SDXL); not the same change." + }, + { + "left": "issue:12666", + "right": "issue:13258", + "accept": false, + "reason": "Both are about Wan2.2 video support, but one is a missing pipeline in a release and the other is a missing image-to-video model; different requests." + }, + { + "left": "issue:4061", + "right": "issue:5466", + "accept": false, + "reason": "Both are LoRA load errors, but SDXL load_lora error and image-to-image expected shape mismatch are distinct failure modes." + }, + { + "left": "issue:4932", + "right": "issue:771", + "accept": false, + "reason": "Dreambooth output file omission is unrelated to BF16 training support." + }, + { + "left": "issue:5226", + "right": "issue:6427", + "accept": false, + "reason": "AdapterPipeline load_lora_weights incompatibility is unrelated to diffusers-to-ckpt conversion." + }, + { + "left": "issue:7690", + "right": "issue:8956", + "accept": false, + "reason": "Tokenizer loading failure and OVStableDiffusionXLPipeline model loading failure are different loader issues." + }, + { + "left": "issue:8708", + "right": "issue:8726", + "accept": false, + "reason": "Both are SD3 dreambooth-related, but one is missing vae.shift_factor and the other is text encoder errors; separate bugs." + }, + { + "left": "issue:10226", + "right": "issue:12529", + "accept": false, + "reason": "Both report tensor size mismatches, but one is Dreambooth Flux with LoRA and the other is Wan2.2 tiled VAE; different models and code paths." + }, + { + "left": "issue:11046", + "right": "issue:11581", + "accept": false, + "reason": "Different pipelines and errors: complex Flux inpainting/controlnet/IP-adapter inference versus Flux IP-Adapter dtype constructor error." + }, + { + "left": "issue:5577", + "right": "issue:9505", + "accept": false, + "reason": "Both are key mismatches, but one is ControlNet single-file time_embed and the other is upscaler middle_block_out; unrelated checkpoints." + }, + { + "left": "issue:2891", + "right": "issue:6758", + "accept": false, + "reason": "One is docs/functionality mismatch for StableDiffusionUpscalePipeline embeds, the other is num_images_per_prompt unsupported in a different pipeline." + }, + { + "left": "issue:8465", + "right": "issue:8614", + "accept": false, + "reason": "Both mention missing config.json, but they target different loaders/models and are not the same artifact." + }, + { + "left": "issue:6158", + "right": "issue:7293", + "accept": false, + "reason": "Different feature requests for different pipelines/models; no shared concrete change." + }, + { + "left": "issue:3325", + "right": "issue:5894", + "accept": false, + "reason": "CUDA invalid argument during Dreambooth training is not the same as invalid buffer size runtime errors." + }, + { + "left": "issue:6082", + "right": "issue:7891", + "accept": false, + "reason": "Half/Float dtype mismatch versus CPU/CUDA device mismatch are distinct runtime errors." + }, + { + "left": "issue:4868", + "right": "issue:5524", + "accept": false, + "reason": "Different attribute errors on different objects: super.__getattr__ versus Linear.lora_layer." + }, + { + "left": "issue:11014", + "right": "issue:4368", + "accept": false, + "reason": "Both involve memory pressure, but one is generic CUDA OOM and the other is SDXL LoRA training OOM; not the same specific bug." + }, + { + "left": "issue:10591", + "right": "issue:3224", + "accept": false, + "reason": "SD3 training script issues are unrelated to duplicated learning-rate scaling in a Dreambooth LoRA script." + }, + { + "left": "issue:2518", + "right": "issue:4756", + "accept": false, + "reason": "xformers support in unconditional image generation is unrelated to adding style-training support in SDXL Dreambooth examples." + }, + { + "left": "issue:10612", + "right": "issue:9667", + "accept": false, + "reason": "Flux Fill NF4 quantized LoRA loading error and local transformer_flux.py fp8 loading error are different Flux loader problems." + }, + { + "left": "issue:3645", + "right": "issue:5059", + "accept": false, + "reason": "ControlNet inpaint training help and SDXL multi-concept finetuning are different workflows, not one bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 62, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10055", + "issue:1033", + "issue:10573", + "issue:10705", + "issue:10866", + "issue:11088", + "issue:11160", + "issue:11169", + "issue:1125", + "issue:11436", + "issue:12085", + "issue:12499", + "issue:13249", + "issue:1572", + "issue:1604", + "issue:234", + "issue:2459", + "issue:2819", + "issue:304", + "issue:3582", + "issue:4000", + "issue:4104", + "issue:4118", + "issue:4198", + "issue:4229", + "issue:4557", + "issue:5016", + "issue:5447", + "issue:5524", + "issue:5563", + "issue:5619", + "issue:5749", + "issue:6107", + "issue:6280", + "issue:6353", + "issue:6446", + "issue:6697", + "issue:7681", + "issue:7728", + "issue:7960", + "issue:8405", + "issue:9044", + "issue:9156", + "issue:9237", + "issue:9239", + "issue:9501", + "issue:9630", + "issue:9895" + ], + "soft_pairs": [ + "issue:11088|issue:12085", + "issue:4000|issue:9044", + "issue:11436|issue:4198", + "issue:6353|issue:6697", + "issue:4104|issue:9239", + "issue:11169|issue:7681", + "issue:1125|issue:234", + "issue:10705|issue:9895", + "issue:10573|issue:8405", + "issue:10866|issue:13249", + "issue:11160|issue:12499", + "issue:1572|issue:234", + "issue:6107|issue:6446", + "issue:3582|issue:5447", + "issue:10055|issue:9501", + "issue:7960|issue:9630", + "issue:304|issue:7681", + "issue:5016|issue:5563", + "issue:4118|issue:6280", + "issue:1033|issue:5749", + "issue:4557|issue:5524", + "issue:9156|issue:9237", + "issue:4229|issue:7728", + "issue:1604|issue:2819", + "issue:2459|issue:5619" + ], + "prepared_review_unit_hash": "d01afacd7494aa78fd21b5da7ccaa9713c7e743075ebf4336b6a702f8eccfc1c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34933, + "estimated_input_tokens": 8734, + "estimated_eval_tokens": 17724 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a noisy mix of unrelated diffusion/training/inference/import bugs. None of the soft pairs look like true duplicates of the same underlying issue, so I rejected all of them. The best representative issue is the broad Flux training report, but the cluster does not really form a tight duplicate set.", + "confidence": 0.39, + "canonical_issue_reason": "issue:9237 is the broadest and most central report here, with strong discussion activity and inbound references, and its title covers a recurring Flux training area better than the narrower reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9237 is the most suitable umbrella issue for this cluster because it is generic, well-discussed, and closest to the recurring training-related subset of items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11088", + "right": "issue:12085", + "accept": false, + "reason": "Both are gguf load errors, but they target different WAN model classes and lack evidence of the same failing code path." + }, + { + "left": "issue:4000", + "right": "issue:9044", + "accept": false, + "reason": "One is about converting an SD2 ckpt; the other is a missing file in the SDXL repo. Different problems entirely." + }, + { + "left": "issue:11436", + "right": "issue:4198", + "accept": false, + "reason": "Both mention dtype mismatch, but the model/pipeline context is different and there is no sign of the same bug." + }, + { + "left": "issue:6353", + "right": "issue:6697", + "accept": false, + "reason": "These are both SDXL-related, but one is a training example error and the other is CUDA OOM; not the same issue." + }, + { + "left": "issue:4104", + "right": "issue:9239", + "accept": false, + "reason": "Both mention black images, but the reports are too broad and do not show the same concrete code-path defect." + }, + { + "left": "issue:11169", + "right": "issue:7681", + "accept": false, + "reason": "These are distinct import failures with different underlying dependency problems." + }, + { + "left": "issue:1125", + "right": "issue:234", + "accept": false, + "reason": "Both are Stable Diffusion Mega-related, but one is an inpaint runtime error and the other is a half-precision failure." + }, + { + "left": "issue:10705", + "right": "issue:9895", + "accept": false, + "reason": "Inference slowdown and device-mismatch error are different symptoms with different root causes." + }, + { + "left": "issue:10573", + "right": "issue:8405", + "accept": false, + "reason": "Both involve multiple controlnets, but one is a padding_mask_crop limitation and the other is a tensor-size error." + }, + { + "left": "issue:10866", + "right": "issue:13249", + "accept": false, + "reason": "Different model families and different LoRA compatibility problems." + }, + { + "left": "issue:11160", + "right": "issue:12499", + "accept": false, + "reason": "Both are WAN I2V issues, but the reports point to different failures and versions, not a clearly shared bug." + }, + { + "left": "issue:1572", + "right": "issue:234", + "accept": false, + "reason": "Same general pipeline family, but one is img2img/inpainting support and the other is a half-precision runtime issue." + }, + { + "left": "issue:6107", + "right": "issue:6446", + "accept": false, + "reason": "An IP-Adapter EC2 deployment issue and an SD2-1 loading issue are unrelated." + }, + { + "left": "issue:3582", + "right": "issue:5447", + "accept": false, + "reason": "Both concern image-to-image/inpaint behavior, but the specific scheduler/strength failures differ." + }, + { + "left": "issue:10055", + "right": "issue:9501", + "accept": false, + "reason": "ControlNet SD3 training failure and Flux DreamBooth save delay are different training paths and failures." + }, + { + "left": "issue:7960", + "right": "issue:9630", + "accept": false, + "reason": "Both are multi-GPU training issues, but one is generic SDXL training and the other is a specific flux-controlnet single-GPU-on-multi-GPU-server error." + }, + { + "left": "issue:304", + "right": "issue:7681", + "accept": false, + "reason": "A transformers import error and an OpenSSL-related import error are not the same bug." + }, + { + "left": "issue:5016", + "right": "issue:5563", + "accept": false, + "reason": "Related SDXL DreamBooth/LoRA area, but one is specifically about text encoder training and the other is a broader error report." + }, + { + "left": "issue:4118", + "right": "issue:6280", + "accept": false, + "reason": "Both mention missing from_single_file support, but on different classes and code paths." + }, + { + "left": "issue:1033", + "right": "issue:5749", + "accept": false, + "reason": "Both are index errors, but the affected pipelines and failure details are unrelated." + }, + { + "left": "issue:4557", + "right": "issue:5524", + "accept": false, + "reason": "Different attribute errors in different components; not the same underlying defect." + }, + { + "left": "issue:9156", + "right": "issue:9237", + "accept": false, + "reason": "Both are Flux LoRA training reports, but one is CUDA OOM and the other is a broader training issue with different symptoms." + }, + { + "left": "issue:4229", + "right": "issue:7728", + "accept": false, + "reason": "Both involve training/accelerate, but the affected scripts and error contexts differ too much to merge as duplicates." + }, + { + "left": "issue:1604", + "right": "issue:2819", + "accept": false, + "reason": "A broken example file and a network download failure are unrelated." + }, + { + "left": "issue:2459", + "right": "issue:5619", + "accept": false, + "reason": "Different scripts and different failure modes; only a broad training/inference theme is shared." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 63, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11117", + "issue:12157", + "issue:12195", + "issue:13053", + "issue:1363", + "issue:1434", + "issue:1572", + "issue:1604", + "issue:1875", + "issue:1931", + "issue:2185", + "issue:2616", + "issue:2806", + "issue:3032", + "issue:3174", + "issue:3411", + "issue:3420", + "issue:3450", + "issue:3574", + "issue:3660", + "issue:3816", + "issue:4256", + "issue:4398", + "issue:487", + "issue:5014", + "issue:5447", + "issue:5510", + "issue:5577", + "issue:5678", + "issue:5860", + "issue:6158", + "issue:6176", + "issue:6218", + "issue:6298", + "issue:639", + "issue:6879", + "issue:6967", + "issue:720", + "issue:7347", + "issue:7589", + "issue:7985", + "issue:8613", + "issue:8729", + "issue:9179", + "issue:9180", + "issue:9516", + "issue:9527", + "issue:9902" + ], + "soft_pairs": [ + "issue:2806|issue:3420", + "issue:5510|issue:720", + "issue:3174|issue:4398", + "issue:1572|issue:6218", + "issue:11117|issue:6879", + "issue:1363|issue:9527", + "issue:1604|issue:5577", + "issue:3411|issue:5014", + "issue:4256|issue:6967", + "issue:12195|issue:9902", + "issue:5447|issue:5860", + "issue:3816|issue:9180", + "issue:6176|issue:7347", + "issue:487|issue:7589", + "issue:5678|issue:6298", + "issue:3450|issue:3574", + "issue:3660|issue:7985", + "issue:12157|issue:2185", + "issue:1875|issue:1931", + "issue:13053|issue:6158", + "issue:1434|issue:8613", + "issue:2616|issue:3032", + "issue:639|issue:9516", + "issue:8729|issue:9179" + ], + "prepared_review_unit_hash": "715c63ec5d135ac9e40bcd8d4f9ec24b53ed36656a622e6bc0b9b279222b7583", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34240, + "estimated_input_tokens": 8560, + "estimated_eval_tokens": 17376 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft pairs are false positives: they share generic error/pipeline keywords, but each issue points to a different model, feature, or failure mode. No duplicate cluster emerges, so there is no clear canonical issue or PR.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good representative; the set is too heterogeneous and the titles mostly overlap only on broad terms like dtype, training, or pipeline.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2806", + "right": "issue:3420", + "accept": false, + "reason": "Both involve pipelines, but one is a printing/customization bug and the other is a ControlNet diffusion-pipeline incompatibility." + }, + { + "left": "issue:5510", + "right": "issue:720", + "accept": false, + "reason": "Both are dtype mismatch errors, but they affect different pipelines and have different input/bias mismatch directions." + }, + { + "left": "issue:3174", + "right": "issue:4398", + "accept": false, + "reason": "Both mention LoRA/DreamBooth, but one is a model-loading regression and the other is black-image output from a different SDXL workflow." + }, + { + "left": "issue:1572", + "right": "issue:6218", + "accept": false, + "reason": "These are separate pipeline bugs: one is img2img/inpainting support in stable_diffusion_mega, the other is batched seeds in ControlNet/img2img." + }, + { + "left": "issue:11117", + "right": "issue:6879", + "accept": false, + "reason": "Both are dtype errors, but they occur in different code paths and models: StableDiffusion3Pipeline vs Custom Diffusion." + }, + { + "left": "issue:1363", + "right": "issue:9527", + "accept": false, + "reason": "One is an ONNX long-prompt/LMS scheduler issue; the other is a ControlNet fp32/bf16 dtype mismatch." + }, + { + "left": "issue:1604", + "right": "issue:5577", + "accept": false, + "reason": "One breaks a long-prompt SD script on a specific diffusers version; the other is a ControlNet single-file conversion KeyError." + }, + { + "left": "issue:3411", + "right": "issue:5014", + "accept": false, + "reason": "One is a test for DreamBooth+LoRA training with text encoder enabled; the other is a user report about poor training results." + }, + { + "left": "issue:4256", + "right": "issue:6967", + "accept": false, + "reason": "Both are SDXL DreamBooth training reports, but one is a general training failure and the other is a prior-preservation shape mismatch." + }, + { + "left": "issue:12195", + "right": "issue:9902", + "accept": false, + "reason": "These are unrelated: a PyTorch custom_op compatibility failure versus a conda version pinning issue." + }, + { + "left": "issue:5447", + "right": "issue:5860", + "accept": false, + "reason": "One is a scheduler denoising_start bug in SDXL img2img; the other is a broader Tiny Autoencoder + ControlNet pipeline failure." + }, + { + "left": "issue:3816", + "right": "issue:9180", + "accept": false, + "reason": "Different missing-attribute problems: transformers CLIPImageProcessor vs pipeline _execution_device." + }, + { + "left": "issue:6176", + "right": "issue:7347", + "accept": false, + "reason": "Both concern LoRA training/loading, but one is loading weights after DreamBooth-LoRA training and the other is a regression after an update." + }, + { + "left": "issue:487", + "right": "issue:7589", + "accept": false, + "reason": "A StableDiffusionPipeline M1 CPU problem is unrelated to a GLIGEN layout-to-image synthesis failure." + }, + { + "left": "issue:5678", + "right": "issue:6298", + "accept": false, + "reason": "One concerns higher-order autograd with scaled_dot_product_attention; the other is a CLIPImageProjection attribute error in GLIGEN." + }, + { + "left": "issue:3450", + "right": "issue:3574", + "accept": false, + "reason": "Both involve from_ckpt, but one is a dev-version regression and the other is a Windows-specific failure." + }, + { + "left": "issue:3660", + "right": "issue:7985", + "accept": false, + "reason": "Both mention torch.compile, but they are different adapters/code paths: LoRA/xformers versus IPAdapter." + }, + { + "left": "issue:12157", + "right": "issue:2185", + "accept": false, + "reason": "A bnb quantization config parsing error is unrelated to missing config files for naive text-to-image training." + }, + { + "left": "issue:1875", + "right": "issue:1931", + "accept": false, + "reason": "Both are float16/float dtype mismatches, but they are distinct pipelines and error contexts, not one concrete bug." + }, + { + "left": "issue:13053", + "right": "issue:6158", + "accept": false, + "reason": "These are separate feature requests for different pipelines: Unipic 3.0 support vs SDXL support in animatediff." + }, + { + "left": "issue:1434", + "right": "issue:8613", + "accept": false, + "reason": "One is an upscale-pipeline OOM at larger sizes; the other is an OOM triggered by caption_column usage in training." + }, + { + "left": "issue:2616", + "right": "issue:3032", + "accept": false, + "reason": "One is a NoneType replace failure during LoRA attention saving; the other is a missing from_pretrained attribute on a ControlNet pipeline." + }, + { + "left": "issue:639", + "right": "issue:9516", + "accept": false, + "reason": "These are unrelated model-initialization and FLUX transformer-argument issues." + }, + { + "left": "issue:8729", + "right": "issue:9179", + "accept": false, + "reason": "Both are SDXL ControlNet training complaints, but one is about validation-image selection and the other about bad training results." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 64, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10026", + "issue:10198", + "issue:10712", + "issue:11041", + "issue:11088", + "issue:11907", + "issue:12078", + "issue:12141", + "issue:1407", + "issue:1572", + "issue:1946", + "issue:1988", + "issue:2361", + "issue:3348", + "issue:3619", + "issue:4256", + "issue:4724", + "issue:4728", + "issue:4752", + "issue:5012", + "issue:5075", + "issue:5246", + "issue:5374", + "issue:5433", + "issue:5475", + "issue:5504", + "issue:5577", + "issue:5593", + "issue:5619", + "issue:5760", + "issue:579", + "issue:6351", + "issue:6610", + "issue:6869", + "issue:6894", + "issue:7133", + "issue:7735", + "issue:7935", + "issue:7960", + "issue:8494", + "issue:9047", + "issue:9067", + "issue:9112", + "issue:9546", + "issue:9556", + "issue:9576", + "issue:9857" + ], + "soft_pairs": [ + "issue:3348|issue:9556", + "issue:5593|issue:9576", + "issue:5504|issue:6894", + "issue:9112|issue:9857", + "issue:10026|issue:7960", + "issue:10010|issue:10198", + "issue:12078|issue:9546", + "issue:5075|issue:5433", + "issue:12141|issue:5619", + "issue:4724|issue:5577", + "issue:5012|issue:7735", + "issue:5246|issue:9067", + "issue:579|issue:7935", + "issue:5475|issue:5760", + "issue:6351|issue:6610", + "issue:5374|issue:8494", + "issue:11041|issue:11088", + "issue:1572|issue:3619", + "issue:4256|issue:7133", + "issue:2361|issue:4728", + "issue:11907|issue:9047", + "issue:1988|issue:4752", + "issue:10712|issue:1946", + "issue:1407|issue:6869" + ], + "prepared_review_unit_hash": "d959af8e30b50a4dbbf93c45a51626de6f2c6598b95b8ecb490b4a449e268186", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34392, + "estimated_input_tokens": 8598, + "estimated_eval_tokens": 17452 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy and not a true duplicate set: the issues span unrelated diffusion pipelines, training scripts, model-loading errors, scheduler bugs, and device-specific failures. None of the soft edges look like the same underlying bug or change.", + "confidence": 0.91, + "canonical_issue_reason": "issue:11041 is the most concrete, well-scoped bug report in the set, with substantial discussion and an explicit failure mode, so it is the best anchor if one issue must represent this noisy cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11041 has the clearest reproduction and impact signal among the issues here; the others are either different subsystems or more generic follow-ups/examples rather than the same bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3348", + "right": "issue:9556", + "accept": false, + "reason": "Both concern LoRA-related behavior, but one is about xformers scale handling and the other about saved rslora models; different bugs and code paths." + }, + { + "left": "issue:5593", + "right": "issue:9576", + "accept": false, + "reason": "Both involve pipeline composition/loading, but one is AutoPipeline with ControlNet and the other is use_safetensors propagation; not the same defect." + }, + { + "left": "issue:5504", + "right": "issue:6894", + "accept": false, + "reason": "Both mention LoRA and xformers/A1111 compatibility, but they report different symptoms in different integrations and are not the same change request." + }, + { + "left": "issue:9112", + "right": "issue:9857", + "accept": false, + "reason": "Both are tensor-shape mismatch reports, but they affect different models/pipelines (Auraflow vs FLUX ControlNet) and different failing dimensions." + }, + { + "left": "issue:10026", + "right": "issue:7960", + "accept": false, + "reason": "Both mention multi-GPU training memory behavior, but one is Flux ControlNet DeepSpeed Stage-3 and the other is generic SDXL training; not one bug." + }, + { + "left": "issue:10010", + "right": "issue:10198", + "accept": false, + "reason": "One is a missing model_index.json for FLUX Fill example; the other is a Flux.1-dev weight mismatch. Different load/fetch failures." + }, + { + "left": "issue:12078", + "right": "issue:9546", + "accept": false, + "reason": "Both are Flux Controlnet training example issues, but one is bad validation input and the other is validation OOM; different underlying problems." + }, + { + "left": "issue:5075", + "right": "issue:5433", + "accept": false, + "reason": "Both involve prompt/scheduler-related SDXL behavior, but one is a shape mismatch with prompt embeddings and the other is visual artifacts with DPM++ schedulers; not the same bug." + }, + { + "left": "issue:12141", + "right": "issue:5619", + "accept": false, + "reason": "Both are VAE/encoding-related runtime failures, but one is WanVACE bfloat16/float16 incompatibility and the other is AutoencodeTiny image encoding in LCM img2img." + }, + { + "left": "issue:4724", + "right": "issue:5577", + "accept": false, + "reason": "One is SDXL img2img checkpoint behavior; the other is ControlNetModel.from_single_file key lookup. Different pipeline components and failure modes." + }, + { + "left": "issue:5012", + "right": "issue:7735", + "accept": false, + "reason": "Both are training-script usability issues, but one is LoRA text encoder training support and the other is an invalid dataset caption column name." + }, + { + "left": "issue:5246", + "right": "issue:9067", + "accept": false, + "reason": "Both are missing-module/attribute style import problems, but one is CLIPImageProjection availability and the other is a missing diffusers.hub_utils module." + }, + { + "left": "issue:579", + "right": "issue:7935", + "accept": false, + "reason": "Both are config/file-not-found loader errors, but they target different models and missing files (textual inversion config.json vs stable-diffusion-2-1 config.json)." + }, + { + "left": "issue:5475", + "right": "issue:5760", + "accept": false, + "reason": "These are entirely different projects/models: LaVie video generation vs DIFF-FOLEY video-to-audio synthesis." + }, + { + "left": "issue:6351", + "right": "issue:6610", + "accept": false, + "reason": "Both are SDXL training/loading issues, but one is DDP missing disable_adapters during LCM distillation and the other is from_single_file ignoring num_in_channels." + }, + { + "left": "issue:5374", + "right": "issue:8494", + "accept": false, + "reason": "One is an unexpected checkpoint_path argument error in a training script; the other is SD3 force_hooks support. Different APIs and code paths." + }, + { + "left": "issue:11041", + "right": "issue:11088", + "accept": false, + "reason": "Both are WAN2.1-related, but one is apply_group_offloading returning an error and the other is a transformer gguf load failure; distinct defects." + }, + { + "left": "issue:1572", + "right": "issue:3619", + "accept": false, + "reason": "Both are legacy pipeline feature bugs, but one concerns stable_diffusion_mega img2img/inpainting and the other CrossFrameAttention in Text2Video-Zero." + }, + { + "left": "issue:4256", + "right": "issue:7133", + "accept": false, + "reason": "Both are training-time issues, but one is SDXL Dreambooth and the other is resuming SDXL Dreambooth LoRA with text-encoder size mismatch; not the same bug." + }, + { + "left": "issue:2361", + "right": "issue:4728", + "accept": false, + "reason": "Both mention training/resume and memory, but one is checkpoint saving crash and the other is OOM when resuming UNet finetuning. Different failure points." + }, + { + "left": "issue:11907", + "right": "issue:9047", + "accept": false, + "reason": "Both are Flux inference compatibility problems, but one is scheduler support in the pipeline and the other is MPS device support; separate issues." + }, + { + "left": "issue:1988", + "right": "issue:4752", + "accept": false, + "reason": "Both are fp16/bf16 type-mismatch errors, but one occurs in the text-to-image trainer example and the other in pipeline_stable_diffusion_xl. Different call sites." + }, + { + "left": "issue:10712", + "right": "issue:1946", + "accept": false, + "reason": "Both are runtime errors during generation, but one is prompt_embeds in StableDiffusion3 and the other is xformers CUDA invalid argument; unrelated root causes." + }, + { + "left": "issue:1407", + "right": "issue:6869", + "accept": false, + "reason": "Both are ValueError-style pipeline issues, but one is an ONNX pipeline feature_extractor expectation and the other is export_to_video unpacking. Not the same problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 65, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10282", + "issue:10689", + "issue:1070", + "issue:10712", + "issue:10738", + "issue:11285", + "issue:11474", + "issue:11585", + "issue:11864", + "issue:13043", + "issue:1363", + "issue:1709", + "issue:1925", + "issue:1944", + "issue:2036", + "issue:2250", + "issue:228", + "issue:2776", + "issue:3152", + "issue:3497", + "issue:4000", + "issue:4183", + "issue:490", + "issue:4925", + "issue:5054", + "issue:5504", + "issue:5522", + "issue:5563", + "issue:5603", + "issue:5661", + "issue:5884", + "issue:6761", + "issue:6914", + "issue:7003", + "issue:7056", + "issue:7333", + "issue:7483", + "issue:776", + "issue:7864", + "issue:8589", + "issue:8817", + "issue:8939", + "issue:8950", + "issue:9180", + "issue:9484", + "issue:9841", + "issue:9914", + "issue:9959" + ], + "soft_pairs": [ + "issue:7333|issue:7864", + "issue:6914|issue:8817", + "issue:1709|issue:9180", + "issue:1070|issue:1925", + "issue:10738|issue:11474", + "issue:10282|issue:7003", + "issue:4925|issue:5563", + "issue:10689|issue:5884", + "issue:7056|issue:9841", + "issue:8589|issue:8950", + "issue:2776|issue:9484", + "issue:3497|issue:5054", + "issue:11864|issue:5661", + "issue:10712|issue:1363", + "issue:1944|issue:2250", + "issue:13043|issue:4183", + "issue:5522|issue:6761", + "issue:5504|issue:7483", + "issue:2036|issue:8939", + "issue:11285|issue:4000", + "issue:9914|issue:9959", + "issue:11585|issue:490", + "issue:228|issue:776", + "issue:3152|issue:5603" + ], + "prepared_review_unit_hash": "d69af19c9591d8a80c79a684948dbfc109d07d2e24bebef5ac374c53ce58244b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34488, + "estimated_input_tokens": 8622, + "estimated_eval_tokens": 17500 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated or only share a broad subsystem. The only clear duplicate-like pair is the Flux Xlabs LoRA loading error pair (#9914/#9959); the rest are distinct bugs or feature requests.", + "confidence": 0.58, + "canonical_issue_reason": "Issue #9914 is the best representative of the only strong duplicate pair: it is broader and more descriptive, while #9959 appears to be a narrower manifestation of the same Flux Xlabs LoRA-loading problem.", + "canonical_pr_reason": null, + "best_issue_reason": "#9914 is the clearest global issue to anchor the cluster because it names the concrete failure and includes a fix proposal, making it the most useful canonical issue among the provided items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7333", + "right": "issue:7864", + "accept": false, + "reason": "Both are video-generation paper/feature topics, but they describe different methods and problems, not the same bug or change." + }, + { + "left": "issue:6914", + "right": "issue:8817", + "accept": false, + "reason": "Both mention IP-Adapter FaceID, but #6914 is specifically an offload-related AttributeError while #8817 is a generic usage failure; not enough evidence of the same concrete code path." + }, + { + "left": "issue:1709", + "right": "issue:9180", + "accept": false, + "reason": "Different failures in different paths: ONNX img2img astype conversion vs pipeline missing _execution_device." + }, + { + "left": "issue:1070", + "right": "issue:1925", + "accept": false, + "reason": "Both concern inpainting, but one is a generic inpainting bug and the other is a half-precision issue; too different to merge as duplicates." + }, + { + "left": "issue:10738", + "right": "issue:11474", + "accept": false, + "reason": "Both are out-of-bounds errors, but the titles indicate different indices and likely different root causes." + }, + { + "left": "issue:10282", + "right": "issue:7003", + "accept": false, + "reason": "Both are tensor shape mismatches, but they affect different layers and dimensions, so they are not the same bug." + }, + { + "left": "issue:4925", + "right": "issue:5563", + "accept": false, + "reason": "Both involve training instability, but one is an OOM in distributed ControlNet SDXL training and the other is an SDXL DreamBooth LoRA error." + }, + { + "left": "issue:10689", + "right": "issue:5884", + "accept": false, + "reason": "Both are IP-Adapter support requests, but one is Flux-specific and the other is about ControlNet/T2I pipelines; they are different feature scopes." + }, + { + "left": "issue:7056", + "right": "issue:9841", + "accept": false, + "reason": "Different DreamBooth/LoRA training errors with different parameters and code paths." + }, + { + "left": "issue:8589", + "right": "issue:8950", + "accept": false, + "reason": "Same model family, but one is prior-preservation handling and the other is a LoRA training bug; not the same underlying issue." + }, + { + "left": "issue:2776", + "right": "issue:9484", + "accept": false, + "reason": "Different training setups and failures: 8GB DreamBooth/DeepSpeed vs Flux multigpu DreamBooth." + }, + { + "left": "issue:3497", + "right": "issue:5054", + "accept": false, + "reason": "Unrelated API/attribute issues in different pipelines." + }, + { + "left": "issue:11864", + "right": "issue:5661", + "accept": false, + "reason": "Different import/compile failures in different modules; no shared code path indicated." + }, + { + "left": "issue:10712", + "right": "issue:1363", + "accept": false, + "reason": "Different pipeline families and errors: SD3 prompt_embeds runtime error vs ONNX long-prompt/LMS scheduler issue." + }, + { + "left": "issue:1944", + "right": "issue:2250", + "accept": false, + "reason": "Both mention DreamBooth inference, but one is random outputs and the other is failure to generate; not clearly the same bug." + }, + { + "left": "issue:13043", + "right": "issue:4183", + "accept": false, + "reason": "Different problems: GPU kernel incompatibility vs guidance-scale behavior in ControlNet." + }, + { + "left": "issue:5522", + "right": "issue:6761", + "accept": false, + "reason": "Different unexpected-argument errors in different APIs." + }, + { + "left": "issue:5504", + "right": "issue:7483", + "accept": false, + "reason": "Different LoRA-related problems: xformers interaction vs SDXL base conversion script failure." + }, + { + "left": "issue:2036", + "right": "issue:8939", + "accept": false, + "reason": "Different model-loading failures for different model families and components." + }, + { + "left": "issue:11285", + "right": "issue:4000", + "accept": false, + "reason": "Both concern conversion, but they target different model versions and error descriptions; not clearly the same defect." + }, + { + "left": "issue:9914", + "right": "issue:9959", + "accept": true, + "reason": "Both describe the same Flux Xlabs LoRA loading failure; #9959 is a narrower manifestation in single_blocks of the broader error in #9914." + }, + { + "left": "issue:11585", + "right": "issue:490", + "accept": false, + "reason": "Generic poor results vs a memory-usage comparison; not the same bug." + }, + { + "left": "issue:228", + "right": "issue:776", + "accept": false, + "reason": "Broken release link and batch-size image-to-image failure are unrelated." + }, + { + "left": "issue:3152", + "right": "issue:5603", + "accept": false, + "reason": "Completely different import errors in different libraries." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative. The only accepted soft pair (#9914/#9959) is reasonably supported by the titles/body descriptions as the same Flux Xlabs LoRA-loading problem, and the rest of the pairs are appropriately rejected without overclaiming." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 66, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11033", + "issue:11149", + "issue:11464", + "issue:11585", + "issue:13274", + "issue:1335", + "issue:1552", + "issue:1883", + "issue:1946", + "issue:1976", + "issue:234", + "issue:2471", + "issue:3235", + "issue:3296", + "issue:3401", + "issue:3573", + "issue:3761", + "issue:4000", + "issue:4030", + "issue:4186", + "issue:4304", + "issue:4697", + "issue:4701", + "issue:490", + "issue:5014", + "issue:5488", + "issue:5798", + "issue:6388", + "issue:7229", + "issue:7267", + "issue:7589", + "issue:7666", + "issue:7708", + "issue:817", + "issue:8482", + "issue:859", + "issue:8686", + "issue:8805", + "issue:8885", + "issue:8887", + "issue:8956", + "issue:8989", + "issue:9101", + "issue:9179", + "issue:964", + "issue:9667", + "issue:9913", + "issue:9941" + ], + "soft_pairs": [ + "issue:3761|issue:8956", + "issue:1335|issue:859", + "issue:1883|issue:8482", + "issue:8887|issue:9101", + "issue:234|issue:2471", + "issue:11585|issue:4186", + "issue:1552|issue:7666", + "issue:11033|issue:1976", + "issue:1976|issue:490", + "issue:4000|issue:7708", + "issue:4030|issue:9941", + "issue:3401|issue:3573", + "issue:7589|issue:8686", + "issue:13274|issue:8989", + "issue:11464|issue:4697", + "issue:1946|issue:4701", + "issue:3235|issue:8885", + "issue:4304|issue:6388", + "issue:5014|issue:9179", + "issue:8805|issue:9913", + "issue:5798|issue:859", + "issue:817|issue:964", + "issue:11149|issue:5798", + "issue:5488|issue:7589", + "issue:7267|issue:9667", + "issue:3296|issue:7229" + ], + "prepared_review_unit_hash": "9cac7a00a66d80b3ca4a5c02bf77250fc6bdcb5b2419c50b5c3b2499183a6283", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 34292, + "estimated_input_tokens": 8573, + "estimated_eval_tokens": 17402 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issues with one strong duplicate pair around the `class_labels` / `num_class_embeds` error. The other soft matches are broad title similarities, not the same concrete bug.", + "confidence": 0.76, + "canonical_issue_reason": "Issue 3401 is the best canonical issue because it states the core bug generally (`class_labels should be provided when num_class_embeds > 0`), while 3573 is a narrower pipeline-specific report of the same error.", + "canonical_pr_reason": null, + "best_issue_reason": "3401 is the most suitable global issue to anchor the duplicate pair: it names the root condition directly and is broader than the local upscaler manifestation.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3761", + "right": "issue:8956", + "accept": false, + "reason": "Different failures: multi-community pipeline loading in one process vs a model load failure in OVStableDiffusionXLPipeline." + }, + { + "left": "issue:1335", + "right": "issue:859", + "accept": false, + "reason": "Both are init/signature errors, but they involve different missing/unsupported arguments and different pipeline classes." + }, + { + "left": "issue:1883", + "right": "issue:8482", + "accept": false, + "reason": "VAE gradient-checkpointing support vs a request for SD3 model support; not the same bug." + }, + { + "left": "issue:8887", + "right": "issue:9101", + "accept": false, + "reason": "One is a wrong VAE encode bug in SD3 training; the other is a feature/support request for a different SD3 training script." + }, + { + "left": "issue:234", + "right": "issue:2471", + "accept": false, + "reason": "Half-precision runtime problems and a size-specific Attend-and-Excite issue are unrelated." + }, + { + "left": "issue:11585", + "right": "issue:4186", + "accept": false, + "reason": "Generic Web UI dissatisfaction vs a specific SDXL inpaint/refiner failure." + }, + { + "left": "issue:1552", + "right": "issue:7666", + "accept": false, + "reason": "Attention slicing on SD2 and `from_single_file` custom pipeline loading are different code paths and symptoms." + }, + { + "left": "issue:11033", + "right": "issue:1976", + "accept": false, + "reason": "Both mention loading, but one is SD1.5 `from_single_file` UNet loading and the other is a tokenizer path lookup error." + }, + { + "left": "issue:1976", + "right": "issue:490", + "accept": false, + "reason": "Tokenizer config lookup error vs a memory-usage complaint; no shared underlying bug." + }, + { + "left": "issue:4000", + "right": "issue:7708", + "accept": false, + "reason": "Checkpoint conversion to diffusers format is not the same as an SDXL tokenizer config missing-file error." + }, + { + "left": "issue:4030", + "right": "issue:9941", + "accept": false, + "reason": "A v2.1 embedding runtime error and a generic Colab failure are too different to be duplicates." + }, + { + "left": "issue:3401", + "right": "issue:3573", + "accept": true, + "reason": "Same underlying bug: both report the `class_labels should be provided when num_class_embeds > 0` error; 3573 is just the local upscaler manifestation." + }, + { + "left": "issue:7589", + "right": "issue:8686", + "accept": false, + "reason": "GLIGEN layout-to-image failure and SDXL ControlNet callback issues are separate problems." + }, + { + "left": "issue:13274", + "right": "issue:8989", + "accept": false, + "reason": "Both involve Stable Audio, but one is an AMD recursion error and the other is a sequential offloading bug; different failure modes." + }, + { + "left": "issue:11464", + "right": "issue:4697", + "accept": false, + "reason": "Both are LoRA loading complaints, but they concern different models/pipelines and likely different fixes." + }, + { + "left": "issue:1946", + "right": "issue:4701", + "accept": false, + "reason": "Different CUDA issues: xformers invalid-argument vs AutoencoderKL float32 differentiation error." + }, + { + "left": "issue:3235", + "right": "issue:8885", + "accept": false, + "reason": "Custom Diffusion weight-loading failure vs a Colab dependency/modifier-token setup problem." + }, + { + "left": "issue:4304", + "right": "issue:6388", + "accept": false, + "reason": "SDXL LoRA training support request vs gradient-checkpointing being ignored; not the same bug." + }, + { + "left": "issue:5014", + "right": "issue:9179", + "accept": false, + "reason": "Generic bad training results in Web UI vs sdxl+controlnet training correctness; too broad and different." + }, + { + "left": "issue:8805", + "right": "issue:9913", + "accept": false, + "reason": "A conversion script regression and a `set_lora_device` behavior report are unrelated." + }, + { + "left": "issue:5798", + "right": "issue:859", + "accept": false, + "reason": "Unexpected callback kwarg vs unexpected device_map kwarg are different API incompatibilities." + }, + { + "left": "issue:817", + "right": "issue:964", + "accept": false, + "reason": "Mixed-precision dreambooth vs inpainting example failure; no shared code-path evidence." + }, + { + "left": "issue:11149", + "right": "issue:5798", + "accept": false, + "reason": "Different unexpected-keyword errors in unrelated APIs." + }, + { + "left": "issue:5488", + "right": "issue:7589", + "accept": false, + "reason": "Demo image generation failure and GLIGEN layout-to-image failure are not the same defect." + }, + { + "left": "issue:7267", + "right": "issue:9667", + "accept": false, + "reason": "Local data loading in ControlNet training vs Flux fp8 loading failure are distinct issues." + }, + { + "left": "issue:3296", + "right": "issue:7229", + "accept": false, + "reason": "Both mention checkpoints, but one is DreamBooth text-encoder save/load and the other is a broader advanced-training checkpoint issue; not clearly the same bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative on the soft-edge pairs. The one accepted duplicate, issue:3401 vs issue:3573, is well supported by the shared `class_labels should be provided when num_class_embeds > 0` error. No overstatement stands out." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 67, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1070", + "issue:10712", + "issue:11285", + "issue:11853", + "issue:12003", + "issue:12009", + "issue:1249", + "issue:13043", + "issue:1305", + "issue:13298", + "issue:1556", + "issue:1604", + "issue:1616", + "issue:1672", + "issue:2420", + "issue:2459", + "issue:2775", + "issue:2871", + "issue:2904", + "issue:3124", + "issue:4137", + "issue:4183", + "issue:4213", + "issue:4389", + "issue:4460", + "issue:4542", + "issue:490", + "issue:5465", + "issue:5524", + "issue:5606", + "issue:6280", + "issue:6298", + "issue:6427", + "issue:6594", + "issue:6610", + "issue:6879", + "issue:6964", + "issue:7228", + "issue:7483", + "issue:7509", + "issue:771", + "issue:858", + "issue:8686", + "issue:8826", + "issue:9476", + "issue:9525", + "issue:969" + ], + "soft_pairs": [ + "issue:1604|issue:7483", + "issue:13043|issue:5606", + "issue:5606|issue:9525", + "issue:12003|issue:12009", + "issue:1305|issue:7509", + "issue:13298|issue:1556", + "issue:490|issue:6594", + "issue:4542|issue:969", + "issue:2775|issue:8826", + "issue:5524|issue:858", + "issue:4389|issue:7228", + "issue:1070|issue:2871", + "issue:6280|issue:6610", + "issue:6427|issue:6964", + "issue:4137|issue:4460", + "issue:4213|issue:8686", + "issue:3124|issue:771", + "issue:11853|issue:4183", + "issue:11285|issue:1672", + "issue:1249|issue:2904", + "issue:2420|issue:2459", + "issue:6879|issue:9476", + "issue:10712|issue:6298", + "issue:1616|issue:5465" + ], + "prepared_review_unit_hash": "44e609b81163eb9707edf748386e8995dfed5f70daa29ab1fa616d6fe8fae3e6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 34016, + "estimated_input_tokens": 8504, + "estimated_eval_tokens": 17264 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy: several unrelated issue pairs share broad topic words, but only two pairs look like true duplicates or near-duplicates with the same concrete failure mode.", + "confidence": 0.91, + "canonical_issue_reason": "issue:12003 is the best representative because it states the exact meta-tensor copy failure in a broad form, and issue:12009 reads like a specific instance of that same loader error.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:12003 is the strongest cluster anchor among the clearly matching bug reports; it is concrete, reusable as the umbrella report, and maps cleanly onto its specific duplicate-like counterpart.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1604", + "right": "issue:7483", + "accept": false, + "reason": "Both involve conversion-related scripts, but they describe different breakages, different models, and different code paths." + }, + { + "left": "issue:13043", + "right": "issue:5606", + "accept": false, + "reason": "One is a CUDA kernel-image compatibility error on RTX 5090; the other is a LoRA alpha-loading bug. Not the same bug." + }, + { + "left": "issue:5606", + "right": "issue:9525", + "accept": false, + "reason": "Both mention LoRA, but one is about missing alpha handling and the other about lora_scale behavior in Flux. Different defects." + }, + { + "left": "issue:12003", + "right": "issue:12009", + "accept": true, + "reason": "Both report the same meta-tensor copy failure during model loading; 12009 is a concrete case of the exact error in 12003." + }, + { + "left": "issue:1305", + "right": "issue:7509", + "accept": false, + "reason": "These are different pipeline features with different goals; no shared bug or code-path failure is indicated." + }, + { + "left": "issue:13298", + "right": "issue:1556", + "accept": false, + "reason": "Both are CUDA/device-related, but one says the generator device being cuda breaks generation and the other says images are black with an NSFW warning. Too different." + }, + { + "left": "issue:490", + "right": "issue:6594", + "accept": false, + "reason": "Memory usage and a network download failure are unrelated failure modes." + }, + { + "left": "issue:4542", + "right": "issue:969", + "accept": false, + "reason": "Image-count behavior in SDXL img2img is unrelated to multi-GPU test failures." + }, + { + "left": "issue:2775", + "right": "issue:8826", + "accept": false, + "reason": "TorchDynamo on a non-function object and torch.cat on an empty list are unrelated runtime errors." + }, + { + "left": "issue:5524", + "right": "issue:858", + "accept": false, + "reason": "A missing lora_layer attribute in Wurstchen loading is unrelated to a named_children AttributeError on Parameter." + }, + { + "left": "issue:4389", + "right": "issue:7228", + "accept": false, + "reason": "ControlNet loading failure and traced-model dtype attribute failure are different bugs." + }, + { + "left": "issue:1070", + "right": "issue:2871", + "accept": false, + "reason": "Both concern inpainting, but one is a vague version-specific bug report and the other is a specific latent-output handling bug." + }, + { + "left": "issue:6280", + "right": "issue:6610", + "accept": false, + "reason": "Both involve from_single_file on SDXL variants, but one is a missing method and the other is an in_channels conversion issue. Different concrete problems." + }, + { + "left": "issue:6427", + "right": "issue:6964", + "accept": false, + "reason": "Model-format conversion and a 4-channel UNet ControlNet inpaint issue are not the same defect." + }, + { + "left": "issue:4137", + "right": "issue:4460", + "accept": false, + "reason": "Both compare diffusers to other Stable Diffusion implementations, but they concern different result-mismatch scenarios with no shared concrete bug." + }, + { + "left": "issue:4213", + "right": "issue:8686", + "accept": false, + "reason": "One is about from_single_file requiring accelerate; the other is a callback incompatibility with ControlNet. No duplicate evidence." + }, + { + "left": "issue:3124", + "right": "issue:771", + "accept": false, + "reason": "Model conversion failure and BF16 dreambooth support are unrelated." + }, + { + "left": "issue:11853", + "right": "issue:4183", + "accept": false, + "reason": "ControlNet/IP-Adapter loading conflict and guidance_scale<=1 error are different ControlNet problems." + }, + { + "left": "issue:11285", + "right": "issue:1672", + "accept": false, + "reason": "Both touch conversion utilities, but one is a value-error during conversion and the other is an output-size bug. Not the same issue." + }, + { + "left": "issue:1249", + "right": "issue:2904", + "accept": false, + "reason": "An OnnxPipeline device attribute error and an AudioLDM missing scheduler error are unrelated." + }, + { + "left": "issue:2420", + "right": "issue:2459", + "accept": true, + "reason": "These are effectively the same training-script xformers bug: both say enable_xformers_memory_efficient_attention does not work in train_text_to_image_lora.py." + }, + { + "left": "issue:6879", + "right": "issue:9476", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different workflows with different triggers and code paths." + }, + { + "left": "issue:10712", + "right": "issue:6298", + "accept": false, + "reason": "StableDiffusion3 prompt_embeds runtime failure and GLIGEN CLIPImageProjection missing attribute are unrelated." + }, + { + "left": "issue:1616", + "right": "issue:5465", + "accept": false, + "reason": "LPW pipeline version compatibility and DiffEdit inpaint_strength handling are different pipeline bugs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: the two accepted pairs are clearly supported by the issue titles, and the rejected pairs are not overstated beyond the evidence available. The canonical issue choice for 12003 over 12009 is also reasonable and well-justified." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 68, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1013", + "issue:1084", + "issue:11033", + "issue:11396", + "issue:1214", + "issue:1398", + "issue:1495", + "issue:1572", + "issue:1931", + "issue:2645", + "issue:2769", + "issue:2806", + "issue:2875", + "issue:3045", + "issue:3124", + "issue:3140", + "issue:3348", + "issue:3379", + "issue:3619", + "issue:3803", + "issue:4302", + "issue:4340", + "issue:4460", + "issue:4724", + "issue:4726", + "issue:4885", + "issue:4925", + "issue:499", + "issue:5016", + "issue:5219", + "issue:5578", + "issue:639", + "issue:6427", + "issue:6584", + "issue:6699", + "issue:7057", + "issue:7197", + "issue:7283", + "issue:7293", + "issue:7432", + "issue:7466", + "issue:7681", + "issue:9067", + "issue:9092", + "issue:9179", + "issue:9841", + "issue:9861" + ], + "soft_pairs": [ + "issue:4340|issue:9841", + "issue:1398|issue:1931", + "issue:11396|issue:7283", + "issue:1214|issue:6427", + "issue:1495|issue:3140", + "issue:1572|issue:2806", + "issue:4925|issue:9179", + "issue:639|issue:7057", + "issue:5219|issue:6584", + "issue:1084|issue:7293", + "issue:2769|issue:3045", + "issue:3124|issue:3619", + "issue:4460|issue:7432", + "issue:7681|issue:9067", + "issue:2875|issue:7197", + "issue:11033|issue:499", + "issue:4726|issue:6699", + "issue:2645|issue:7466", + "issue:3348|issue:9841", + "issue:3379|issue:5016", + "issue:3803|issue:9861", + "issue:5578|issue:9092", + "issue:4302|issue:4885", + "issue:1013|issue:4724" + ], + "prepared_review_unit_hash": "ee680fe862e8fc4464e11401e71b80eb17cdb9280957c1c1121cfddadd90a5a9", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33569, + "estimated_input_tokens": 8393, + "estimated_eval_tokens": 17042 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy and spans many unrelated Diffusers bugs/features. I don\u2019t see any true duplicate pairs among the soft edges; most are only loosely similar by subsystem (conversion, LoRA, training, pipeline typing).", + "confidence": 0.9, + "canonical_issue_reason": "#6427 is the broadest, most general issue in the set around model conversion failures, so it is the best anchor if this cluster must be represented by one issue.", + "canonical_pr_reason": null, + "best_issue_reason": "#6427 is the strongest global issue candidate because it is concrete, widely applicable to nearby conversion/loading reports, and better suited as a triage anchor than the more niche or symptom-specific issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4340", + "right": "issue:9841", + "accept": false, + "reason": "Both are SDXL LoRA training problems, but they describe different failure modes and different code paths (`train_text_encoder` vs `use_dora=False`)." + }, + { + "left": "issue:1398", + "right": "issue:1931", + "accept": false, + "reason": "One is an MPS/Apple Silicon compatibility issue; the other is a float16 dtype mismatch. Same broad area, not the same bug." + }, + { + "left": "issue:11396", + "right": "issue:7283", + "accept": false, + "reason": "Both mention LoRA/model handling, but one is about converting a HiDream LoRA for ComfyUI and the other is about loading a Stable Cascade LoRA. Different tasks and model families." + }, + { + "left": "issue:1214", + "right": "issue:6427", + "accept": false, + "reason": "Half-precision CLIPTextModel failure is unrelated to diffusers-to-ckpt conversion failures." + }, + { + "left": "issue:1495", + "right": "issue:3140", + "accept": false, + "reason": "Different import errors affecting different symbols (`init_git_repo` vs `StableDiffusionControlNetPipeline`)." + }, + { + "left": "issue:1572", + "right": "issue:2806", + "accept": false, + "reason": "`stable_diffusion_mega` img2img/inpainting failure is unrelated to pipeline printing not honoring customized pipelines." + }, + { + "left": "issue:4925", + "right": "issue:9179", + "accept": false, + "reason": "Both involve SDXL ControlNet training, but one is an OOM/distributed-memory problem and the other is a bad-results/training-quality complaint." + }, + { + "left": "issue:639", + "right": "issue:7057", + "accept": false, + "reason": "Parameter initialization mismatch is not the same as a checkpoint loading failure." + }, + { + "left": "issue:5219", + "right": "issue:6584", + "accept": false, + "reason": "These are opposite-direction conversion/load problems on different models; not the same concrete bug." + }, + { + "left": "issue:1084", + "right": "issue:7293", + "accept": false, + "reason": "A docs request about multiple pipelines reusing weights is unrelated to an AudioLDM2 feature request." + }, + { + "left": "issue:2769", + "right": "issue:3045", + "accept": false, + "reason": "Both are conversion-related, but one is about round-tripping ckpt/safetensors through diffusers and the other is LoRA .safetensors to diffusers. Different workflows and likely different fixes." + }, + { + "left": "issue:3124", + "right": "issue:3619", + "accept": false, + "reason": "Model conversion failure and Text2Video-Zero CrossFrameAttention failure are unrelated." + }, + { + "left": "issue:4460", + "right": "issue:7432", + "accept": false, + "reason": "Both compare Diffusers outputs to another implementation, but they are not the same specific discrepancy or code path." + }, + { + "left": "issue:7681", + "right": "issue:9067", + "accept": false, + "reason": "One is an OpenSSL/lib import failure; the other is a missing `diffusers.hub_utils` module. Different causes and fixes." + }, + { + "left": "issue:2875", + "right": "issue:7197", + "accept": false, + "reason": "Training loss NaN and resume-training KeyError are different training failures." + }, + { + "left": "issue:11033", + "right": "issue:499", + "accept": false, + "reason": "A broken `from_single_file` load for SD1.5 UNet is unrelated to the example unconditional training script." + }, + { + "left": "issue:4726", + "right": "issue:6699", + "accept": false, + "reason": "Both are type-hint complaints, but they target different callback parameters (`callback` vs `callback_on_step_end`) and are not the same bug." + }, + { + "left": "issue:2645", + "right": "issue:7466", + "accept": false, + "reason": "M1/CUDA assertion in ControlNet and torch.compile failure for SDXL multi-controlnet inpaint are different issues." + }, + { + "left": "issue:3348", + "right": "issue:9841", + "accept": false, + "reason": "LoRA scale behavior with xformers is unrelated to a DoRA-related TypeError in SDXL DreamBooth training." + }, + { + "left": "issue:3379", + "right": "issue:5016", + "accept": false, + "reason": "Both are DreamBooth-related, but one is about loading an intermediate checkpoint and the other is about training text encoders." + }, + { + "left": "issue:3803", + "right": "issue:9861", + "accept": false, + "reason": "Multi-GPU LoRA model sharing and Flux transformer not updating are different training bugs." + }, + { + "left": "issue:5578", + "right": "issue:9092", + "accept": false, + "reason": "Both mention `model_index.json`, but the contexts differ enough that they are not clearly the same underlying load bug." + }, + { + "left": "issue:4302", + "right": "issue:4885", + "accept": false, + "reason": "Both involve LoRA loading in SDXL, but one is a specific third-party SDXL LoRA weight-loading failure and the other is a ControlNet img2img pipeline not loading LoRA weights. Different concrete paths." + }, + { + "left": "issue:1013", + "right": "issue:4724", + "accept": false, + "reason": "StableDiffusionMegaPipeline breakage in Diffusers 6.0 is unrelated to an SDXL img2img mismatch with the base checkpoint." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 69, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1037", + "issue:10472", + "issue:10866", + "issue:11978", + "issue:1214", + "issue:12141", + "issue:122", + "issue:12535", + "issue:13035", + "issue:13043", + "issue:1463", + "issue:1817", + "issue:2774", + "issue:2830", + "issue:2973", + "issue:3045", + "issue:3224", + "issue:3481", + "issue:3487", + "issue:3497", + "issue:3816", + "issue:4080", + "issue:4086", + "issue:4183", + "issue:4279", + "issue:4399", + "issue:4701", + "issue:4835", + "issue:4852", + "issue:487", + "issue:5125", + "issue:5227", + "issue:5798", + "issue:6362", + "issue:6964", + "issue:7483", + "issue:7660", + "issue:771", + "issue:7962", + "issue:8534", + "issue:8556", + "issue:8649", + "issue:9044", + "issue:9055", + "issue:955", + "issue:9576", + "issue:9911", + "issue:9914" + ], + "soft_pairs": [ + "issue:12141|issue:771", + "issue:3497|issue:3816", + "issue:4183|issue:9911", + "issue:4701|issue:5125", + "issue:8556|issue:8649", + "issue:1037|issue:4080", + "issue:10866|issue:12535", + "issue:487|issue:7660", + "issue:4086|issue:4279", + "issue:122|issue:1463", + "issue:7962|issue:955", + "issue:13043|issue:4835", + "issue:1214|issue:6964", + "issue:3224|issue:8534", + "issue:4852|issue:9914", + "issue:3487|issue:9576", + "issue:7483|issue:9044", + "issue:1817|issue:5798", + "issue:2774|issue:3045", + "issue:11978|issue:2830", + "issue:2973|issue:3481", + "issue:10472|issue:6362", + "issue:4399|issue:5227", + "issue:13035|issue:9055" + ], + "prepared_review_unit_hash": "0f88e6547e1a861291b8f4fcfd82a9a5636a1d87c069f071e392de3c74c051ca", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34296, + "estimated_input_tokens": 8574, + "estimated_eval_tokens": 17404 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are not true duplicates; they span many unrelated bugs, feature requests, and model/pipeline-specific errors. The soft pairs mostly share only broad vocabulary like LoRA, ControlNet, precision, or import/runtime errors, not the same concrete code path.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical issue stands out: the set is heterogeneous and the overlaps are only superficial.", + "canonical_pr_reason": null, + "best_issue_reason": "No strong global best issue for deduping; the list does not form one coherent bug cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12141", + "right": "issue:771", + "accept": false, + "reason": "Both mention BF16/bfloat16, but one is a WanVACE VAE precision bug and the other is Dreambooth training; different code paths." + }, + { + "left": "issue:3497", + "right": "issue:3816", + "accept": false, + "reason": "An attribute error in an inpaint pipeline is unrelated to a missing transformers class import." + }, + { + "left": "issue:4183", + "right": "issue:9911", + "accept": false, + "reason": "Both involve ControlNet, but one is a guidance_scale edge case and the other is Flux multi-ControlNet layer-length mismatch." + }, + { + "left": "issue:4701", + "right": "issue:5125", + "accept": false, + "reason": "Both are CUDA errors, but they occur in different components: AutoencoderKL decoding vs XFormers attention processing." + }, + { + "left": "issue:8556", + "right": "issue:8649", + "accept": false, + "reason": "Both are SD3-related, but one is a meta-tensor copy failure and the other is a num_images_per_prompt regression." + }, + { + "left": "issue:1037", + "right": "issue:4080", + "accept": false, + "reason": "A community pipeline request and an SDXL inpainting pipeline request are different feature additions." + }, + { + "left": "issue:10866", + "right": "issue:12535", + "accept": false, + "reason": "Both are LoRA-loading complaints, but they target different models and failure modes." + }, + { + "left": "issue:487", + "right": "issue:7660", + "accept": false, + "reason": "M1 CPU support for StableDiffusionPipeline is unrelated to a trace example not working." + }, + { + "left": "issue:4086", + "right": "issue:4279", + "accept": false, + "reason": "SDXL controlnet training and SDXL LoRA loading are distinct problems, not one concrete bug." + }, + { + "left": "issue:122", + "right": "issue:1463", + "accept": false, + "reason": "Missing 'exceptions' and missing 'datasets' are unrelated dependency/import errors." + }, + { + "left": "issue:7962", + "right": "issue:955", + "accept": false, + "reason": "Both ask for extra call arguments, but they request different parameters and different pipeline behavior." + }, + { + "left": "issue:13043", + "right": "issue:4835", + "accept": false, + "reason": "An attention-backend kernel compatibility issue is not the same as a MultiControlNetModel bug." + }, + { + "left": "issue:1214", + "right": "issue:6964", + "accept": false, + "reason": "Half-precision CLIPTextModel and 4-channel U-Net/VAE support are different precision/model-shape issues." + }, + { + "left": "issue:3224", + "right": "issue:8534", + "accept": false, + "reason": "Training-script weighting bugs, but one is duplicated LR scaling and the other is logit-normal weighting." + }, + { + "left": "issue:4852", + "right": "issue:9914", + "accept": false, + "reason": "Both involve LoRA loading, but one is generic loading and the other is a Flux/Xlabs-specific fix proposal." + }, + { + "left": "issue:3487", + "right": "issue:9576", + "accept": false, + "reason": "Different inference behavior with steps is unrelated to safetensors being passed to submodules." + }, + { + "left": "issue:7483", + "right": "issue:9044", + "accept": false, + "reason": "LoRA conversion for SDXL 1.0 base and a missing diffusion_pytorch_model.bin are separate loading/conversion issues." + }, + { + "left": "issue:1817", + "right": "issue:5798", + "accept": false, + "reason": "Unexpected scheduler arguments in Dreambooth and unexpected callback arguments in ControlNet are different API mismatches." + }, + { + "left": "issue:2774", + "right": "issue:3045", + "accept": false, + "reason": "Both are conversion scripts, but ControlNet-to-diffusers and LoRA-to-diffusers are different conversion paths." + }, + { + "left": "issue:11978", + "right": "issue:2830", + "accept": false, + "reason": "A new Bria pipeline request is unrelated to a question about using a ModelScope video model." + }, + { + "left": "issue:2973", + "right": "issue:3481", + "accept": false, + "reason": "Non-deterministic outputs from offload vs scheduler-dependent differences are not the same bug." + }, + { + "left": "issue:10472", + "right": "issue:6362", + "accept": false, + "reason": "A lora_bias constructor error during inference is unrelated to OOM while mapping a fine-tuning dataset." + }, + { + "left": "issue:4399", + "right": "issue:5227", + "accept": false, + "reason": "Both mention torch.compile, but one is compile plus CPU offload and the other is LoRA loading under compile." + }, + { + "left": "issue:13035", + "right": "issue:9055", + "accept": false, + "reason": "Different import errors involving different symbols from different packages; not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 70, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10169", + "issue:10382", + "issue:10472", + "issue:10722", + "issue:10848", + "issue:10869", + "issue:11864", + "issue:12009", + "issue:12492", + "issue:1255", + "issue:1500", + "issue:1534", + "issue:1536", + "issue:1572", + "issue:1842", + "issue:1905", + "issue:2276", + "issue:2578", + "issue:2984", + "issue:3152", + "issue:4210", + "issue:4324", + "issue:4368", + "issue:4492", + "issue:4520", + "issue:4932", + "issue:499", + "issue:5022", + "issue:5028", + "issue:5125", + "issue:5274", + "issue:5678", + "issue:6700", + "issue:7114", + "issue:730", + "issue:7708", + "issue:771", + "issue:7731", + "issue:7801", + "issue:7985", + "issue:8494", + "issue:858", + "issue:867", + "issue:9267", + "issue:9497", + "issue:9576", + "issue:9913" + ], + "soft_pairs": [ + "issue:1572|issue:4520", + "issue:2578|issue:499", + "issue:1255|issue:3152", + "issue:10382|issue:9267", + "issue:6700|issue:9576", + "issue:10472|issue:5028", + "issue:10848|issue:5028", + "issue:2984|issue:7731", + "issue:10169|issue:7708", + "issue:1905|issue:867", + "issue:4932|issue:730", + "issue:5274|issue:7801", + "issue:1500|issue:1536", + "issue:4368|issue:7114", + "issue:5125|issue:5678", + "issue:1842|issue:8494", + "issue:1534|issue:2276", + "issue:771|issue:9913", + "issue:10722|issue:9497", + "issue:11864|issue:7985", + "issue:10869|issue:12009", + "issue:4210|issue:4324", + "issue:4492|issue:5022", + "issue:12492|issue:858" + ], + "prepared_review_unit_hash": "966697e66ab25c93edb4216f3c3855918a5a79019e0376f2031819bc87762688", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33913, + "estimated_input_tokens": 8479, + "estimated_eval_tokens": 17214 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous bug reports with only a couple of clear duplicate-like pairs. Most soft edges are false positives because they share a broad subsystem or error family, not the same concrete bug.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 5274 is the cleanest duplicate anchor: it describes a specific fp16/mixed_precision training mismatch, and 7801 is nearly the same report.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 5274 is the best representative issue overall because it is precise, reproducible, and has the strongest near-exact duplicate match in the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1572", + "right": "issue:4520", + "accept": false, + "reason": "Same broad area (pipeline/training) but different bugs: img2img/inpainting support vs LoRA scale with attention slicing." + }, + { + "left": "issue:2578", + "right": "issue:499", + "accept": false, + "reason": "Unrelated failure modes: editable install problem vs unconditional training script issue." + }, + { + "left": "issue:1255", + "right": "issue:3152", + "accept": false, + "reason": "Different libraries and errors: diffusers scheduler import vs TensorFlow Lite import failure." + }, + { + "left": "issue:10382", + "right": "issue:9267", + "accept": false, + "reason": "Both are AttributeErrors, but on different objects/code paths: VAE forward return type vs missing rope attribute in Flux transformer." + }, + { + "left": "issue:6700", + "right": "issue:9576", + "accept": false, + "reason": "Both involve argument passing, but one is attention processor temb handling and the other is pipeline loading with use_safetensors." + }, + { + "left": "issue:10472", + "right": "issue:5028", + "accept": false, + "reason": "Both are unexpected-kwarg TypeErrors, but the kwargs and code paths differ (lora_bias vs scale)." + }, + { + "left": "issue:10848", + "right": "issue:5028", + "accept": false, + "reason": "Different loaders and parameters: load_model_dict_into_meta(device) vs a custom forward hook receiving scale." + }, + { + "left": "issue:2984", + "right": "issue:7731", + "accept": false, + "reason": "Both mention video generation, but they are different model/paper topics rather than the same bug." + }, + { + "left": "issue:10169", + "right": "issue:7708", + "accept": false, + "reason": "Both mention missing config.json, but the reports are too generic/specific mismatch to confirm the same underlying bug." + }, + { + "left": "issue:1905", + "right": "issue:867", + "accept": false, + "reason": "Both are memory-retention complaints, but one is a pipeline leak and the other is a FastAPI deployment context issue." + }, + { + "left": "issue:4932", + "right": "issue:730", + "accept": false, + "reason": "Different DreamBooth problems: missing output artifact vs training behavior differing by environment." + }, + { + "left": "issue:5274", + "right": "issue:7801", + "accept": true, + "reason": "Near-verbatim duplicate: both report mixed_precision/fp16 not matching the model dtype during training." + }, + { + "left": "issue:1500", + "right": "issue:1536", + "accept": true, + "reason": "Same lpw_stable_diffusion compatibility problem with newer diffusers versions." + }, + { + "left": "issue:4368", + "right": "issue:7114", + "accept": false, + "reason": "Both are OOM-related training reports, but they are different scripts and different failure contexts." + }, + { + "left": "issue:5125", + "right": "issue:5678", + "accept": false, + "reason": "Different failures in attention code: CUDA invalid configuration vs higher-order derivative issues with scaled_dot_product_attention." + }, + { + "left": "issue:1842", + "right": "issue:8494", + "accept": false, + "reason": "Both are unexpected keyword argument errors, but they concern different kwargs and likely different call sites." + }, + { + "left": "issue:1534", + "right": "issue:2276", + "accept": false, + "reason": "DreamBooth notebook issue vs xformers install problem; not the same underlying bug." + }, + { + "left": "issue:771", + "right": "issue:9913", + "accept": false, + "reason": "Different topics: BF16 support in DreamBooth vs set_lora_device behavior." + }, + { + "left": "issue:10722", + "right": "issue:9497", + "accept": false, + "reason": "Both are DreamBooth training errors, but the tensor shape mismatch and mat2 1-D tensor error are distinct." + }, + { + "left": "issue:11864", + "right": "issue:7985", + "accept": false, + "reason": "Both mention torch.compile, but the concrete failures differ: undefined torch in compiled encode vs IPAdapter incompatibility." + }, + { + "left": "issue:10869", + "right": "issue:12009", + "accept": false, + "reason": "Both mention meta tensors, but one is Tensor.item() on meta tensors and the other is copy-out during GGUF loading." + }, + { + "left": "issue:4210", + "right": "issue:4324", + "accept": false, + "reason": "Same UNet file, but different bugs: attention_mask usage question vs NoneType iteration error." + }, + { + "left": "issue:4492", + "right": "issue:5022", + "accept": false, + "reason": "Completely different import/name errors in unrelated packages and code paths." + }, + { + "left": "issue:12492", + "right": "issue:858", + "accept": false, + "reason": "Different AttributeErrors in different subsystems: transformers CLIPFeatureExtractor removal vs Parameter.named_children." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the two accepted soft-edge pairs are conservative and well-supported by the titles. The only clear duplicate-like pairs are indeed 5274\u21947801 and 1500\u21941536; the rest are appropriately rejected as broad similarity rather than the same concrete bug." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 71, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10154", + "issue:10594", + "issue:10650", + "issue:10940", + "issue:10969", + "issue:11464", + "issue:11581", + "issue:11922", + "issue:122", + "issue:12656", + "issue:1552", + "issue:1604", + "issue:1720", + "issue:1985", + "issue:2091", + "issue:2480", + "issue:2548", + "issue:259", + "issue:2871", + "issue:2891", + "issue:3382", + "issue:3565", + "issue:3774", + "issue:3811", + "issue:4080", + "issue:4593", + "issue:5006", + "issue:5227", + "issue:5565", + "issue:5621", + "issue:5846", + "issue:5860", + "issue:5894", + "issue:6146", + "issue:6366", + "issue:6894", + "issue:746", + "issue:7483", + "issue:7690", + "issue:7931", + "issue:7985", + "issue:861", + "issue:8939", + "issue:9362", + "issue:9448", + "issue:9643", + "issue:9911", + "issue:9973" + ], + "soft_pairs": [ + "issue:7931|issue:9448", + "issue:5621|issue:7483", + "issue:6366|issue:6894", + "issue:11464|issue:11581", + "issue:10650|issue:5894", + "issue:861|issue:8939", + "issue:10594|issue:11922", + "issue:10940|issue:122", + "issue:10969|issue:9911", + "issue:1985|issue:4080", + "issue:259|issue:2871", + "issue:2891|issue:9643", + "issue:11464|issue:4593", + "issue:3565|issue:5846", + "issue:5006|issue:9973", + "issue:10154|issue:3811", + "issue:1552|issue:5860", + "issue:746|issue:7690", + "issue:2480|issue:5565", + "issue:1720|issue:2548", + "issue:5227|issue:7985", + "issue:3382|issue:5006", + "issue:12656|issue:9362", + "issue:1604|issue:3774", + "issue:2091|issue:6146" + ], + "prepared_review_unit_hash": "15d6b9c8705174b6ab25fc6931135cd714fc0bebec47f8ce364de246ed5378d0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34868, + "estimated_input_tokens": 8717, + "estimated_eval_tokens": 17690 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated issues about LoRA loading, FLUX/ControlNet, training/checkpointing, tokenizer/import failures, and model-specific runtime errors. None of the soft pairs look like the same concrete bug or change, so all are rejected as duplicates.", + "confidence": 0.84, + "canonical_issue_reason": "No clear canonical issue: the items span several unrelated bug families rather than one duplicate thread.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:3811 is the strongest representative of the only recurring theme here (`load_lora_weights`/LoRA loading) and has a clear, concrete symptom compared with the vaguer nearby reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7931", + "right": "issue:9448", + "accept": false, + "reason": "Both involve IP-Adapter/controlnet inpainting, but the failures differ: one is about `ip_adapter_image_embeds` compatibility, the other is a tuple/shape error. Not the same bug." + }, + { + "left": "issue:5621", + "right": "issue:7483", + "accept": false, + "reason": "Different subjects and code paths: SDXL ControlNet `encoder_hid_proj` support vs LoRA-to-Diffusers conversion for SDXL-base models." + }, + { + "left": "issue:6366", + "right": "issue:6894", + "accept": false, + "reason": "Both concern LoRA checkpoints, but one is about resuming/loading saved checkpoints in training scripts and the other is about LoRA output not working in A1111. Not the same underlying issue." + }, + { + "left": "issue:11464", + "right": "issue:11581", + "accept": false, + "reason": "Both mention FLUX, but one is a LoRA load failure in `FluxControlPipeline` and the other is an IP-Adapter constructor dtype error. Different problems." + }, + { + "left": "issue:10650", + "right": "issue:5894", + "accept": false, + "reason": "Unrelated runtime errors in different contexts: a Lumina shape reshape failure vs an invalid buffer size error." + }, + { + "left": "issue:861", + "right": "issue:8939", + "accept": false, + "reason": "One is about converting ckpt to diffusers; the other is about loading a tokenizer for SD3. Different subsystems and failure modes." + }, + { + "left": "issue:10594", + "right": "issue:11922", + "accept": false, + "reason": "Both are FLUX-related, but one is about multiple controls with LoRA and the other is a FLUX Fill pipeline problem. Not the same concrete bug." + }, + { + "left": "issue:10940", + "right": "issue:122", + "accept": false, + "reason": "Completely different import failures: missing `triton.ops` vs missing `exceptions`." + }, + { + "left": "issue:10969", + "right": "issue:9911", + "accept": false, + "reason": "Both touch FLUX/controlnet multi-control scenarios, but one is a training-time `'weight' must be 2-D` failure and the other is an inference/setup error with different layer lengths. Not mergeable as one bug." + }, + { + "left": "issue:1985", + "right": "issue:4080", + "accept": false, + "reason": "Loading embeddings on Stable Diffusion vs adding an SDXL inpainting pipeline; unrelated." + }, + { + "left": "issue:259", + "right": "issue:2871", + "accept": false, + "reason": "Progress bar behavior is unrelated to latent output handling in inpainting." + }, + { + "left": "issue:2891", + "right": "issue:9643", + "accept": false, + "reason": "A docs/feature mismatch for StableDiffusionUpscalePipeline is unrelated to Flux multi-ControlNet support." + }, + { + "left": "issue:11464", + "right": "issue:4593", + "accept": false, + "reason": "Different LoRA-related failures: FluxControlPipeline load failure vs Automatic1111 LoRA rank constraint." + }, + { + "left": "issue:3565", + "right": "issue:5846", + "accept": false, + "reason": "Training OOM vs checkpoint resume error; different failure classes and code paths." + }, + { + "left": "issue:5006", + "right": "issue:9973", + "accept": false, + "reason": "Both are training issues, but one is end-of-training OOM and the other is FP16 gradient unscale failure. Not the same bug." + }, + { + "left": "issue:10154", + "right": "issue:3811", + "accept": false, + "reason": "Both mention `load_lora_weights`, but one reports an attribute error during loading while the other says the LoRA loads but has no effect. Same area, different concrete failure." + }, + { + "left": "issue:1552", + "right": "issue:5860", + "accept": false, + "reason": "Stable Diffusion 2 attention slicing is unrelated to tiny autoencoder + ControlNet pipeline failures." + }, + { + "left": "issue:746", + "right": "issue:7690", + "accept": false, + "reason": "ONNX revision loading vs tokenizer path resolution; different issues." + }, + { + "left": "issue:2480", + "right": "issue:5565", + "accept": false, + "reason": "Checkpoint folder creation in DreamBooth training is unrelated to the LoRA SDXL example script failing to run." + }, + { + "left": "issue:1720", + "right": "issue:2548", + "accept": false, + "reason": "Both are training-related, but one is M1 finetuning errors and the other is LoRA save errors. Not the same underlying defect." + }, + { + "left": "issue:5227", + "right": "issue:7985", + "accept": false, + "reason": "Both involve `torch.compile`, but one is LoRA loading and the other is IP-Adapter compatibility. Too broad to merge." + }, + { + "left": "issue:1604", + "right": "issue:3774", + "accept": false, + "reason": "Version-specific LPW script breakage is unrelated to zero3 training memory behavior." + }, + { + "left": "issue:2091", + "right": "issue:6146", + "accept": false, + "reason": "Both are checkpoint/multi-GPU training issues, but they affect different scripts and report different failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 72, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10266", + "issue:10512", + "issue:10575", + "issue:10738", + "issue:10833", + "issue:1100", + "issue:11088", + "issue:11218", + "issue:11298", + "issue:11370", + "issue:11474", + "issue:12009", + "issue:12414", + "issue:1264", + "issue:13035", + "issue:13067", + "issue:1974", + "issue:2674", + "issue:2871", + "issue:2880", + "issue:2995", + "issue:3689", + "issue:4073", + "issue:4286", + "issue:4304", + "issue:4630", + "issue:5227", + "issue:5455", + "issue:5606", + "issue:5617", + "issue:5862", + "issue:5873", + "issue:6158", + "issue:6298", + "issue:6382", + "issue:639", + "issue:6607", + "issue:6996", + "issue:7185", + "issue:7228", + "issue:7276", + "issue:7418", + "issue:7483", + "issue:7485", + "issue:776", + "issue:7782", + "issue:9180", + "issue:9341" + ], + "soft_pairs": [ + "issue:5606|issue:7228", + "issue:5455|issue:7276", + "issue:1974|issue:2674", + "issue:4286|issue:7483", + "issue:6996|issue:7485", + "issue:11298|issue:6382", + "issue:5227|issue:5617", + "issue:11088|issue:12009", + "issue:10266|issue:11474", + "issue:10512|issue:9341", + "issue:10575|issue:10738", + "issue:10833|issue:2995", + "issue:1100|issue:9180", + "issue:11218|issue:13035", + "issue:11370|issue:7782", + "issue:12414|issue:4304", + "issue:1264|issue:6607", + "issue:13067|issue:6158", + "issue:2871|issue:776", + "issue:2880|issue:639", + "issue:3689|issue:4630", + "issue:4073|issue:6298", + "issue:5862|issue:7185", + "issue:5873|issue:7418" + ], + "prepared_review_unit_hash": "f756dd60a782333752f7a167b2d502b339ebe086912678784d0671cd4c011d16", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34253, + "estimated_input_tokens": 8564, + "estimated_eval_tokens": 17384 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly unrelated issues. The only plausible duplicate pair in the provided soft edges is the scheduler index-out-of-bounds case: issue:10266 and issue:11474. No PRs are present.", + "confidence": 0.68, + "canonical_issue_reason": "issue:10738 is the best canonical issue because it states the broadest recurring failure mode: a scheduler sigma/index out-of-bounds bug, which best covers the duplicate-looking scheduler reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10738 is the strongest global representative for the cluster\u2019s only clear bug family, since its title is the most general version of the out-of-bounds scheduler problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5606", + "right": "issue:7228", + "accept": false, + "reason": "Different bugs: LoRA alpha handling vs a traced pipeline missing dtype." + }, + { + "left": "issue:5455", + "right": "issue:7276", + "accept": false, + "reason": "Both are notebook failures, but the concrete problems differ: CalledProcessError vs network errors." + }, + { + "left": "issue:1974", + "right": "issue:2674", + "accept": false, + "reason": "Unrelated scheduler-generator reproducibility bug vs missing safetensors dependency." + }, + { + "left": "issue:4286", + "right": "issue:7483", + "accept": false, + "reason": "Both involve LoRA/SDXL, but one is loading LoRA weights and the other is a conversion script issue." + }, + { + "left": "issue:6996", + "right": "issue:7485", + "accept": false, + "reason": "Different code paths and symptoms: NoneType in added_cond_kwargs vs safety_checker callable handling." + }, + { + "left": "issue:11298", + "right": "issue:6382", + "accept": false, + "reason": "Both mention key errors, but one is LoRA hotswapping/PEFT and the other is IP-Adapter during SDXL training." + }, + { + "left": "issue:5227", + "right": "issue:5617", + "accept": false, + "reason": "Shared torch.compile context, but the failures are different enough to not be the same concrete bug." + }, + { + "left": "issue:11088", + "right": "issue:12009", + "accept": false, + "reason": "Both are Wan GGUF load issues, but they describe different load failures and likely different root causes." + }, + { + "left": "issue:10266", + "right": "issue:11474", + "accept": true, + "reason": "Both report scheduler-related index out-of-bounds errors and look like the same off-by-one/sigma indexing bug." + }, + { + "left": "issue:10512", + "right": "issue:9341", + "accept": false, + "reason": "Both are Flux LoRA load issues, but one is quantization-related and the other is a text_model LoRA compatibility problem." + }, + { + "left": "issue:10833", + "right": "issue:2995", + "accept": false, + "reason": "Both are matmul shape errors, but they occur in different pipelines and do not clearly share one root cause." + }, + { + "left": "issue:1100", + "right": "issue:9180", + "accept": false, + "reason": "Different attribute errors on different objects; no clear shared bug." + }, + { + "left": "issue:11218", + "right": "issue:13035", + "accept": false, + "reason": "Both are import errors, but for different missing symbols in unrelated modules." + }, + { + "left": "issue:11370", + "right": "issue:7782", + "accept": false, + "reason": "Different invalid-load-key failures on different conversion paths." + }, + { + "left": "issue:12414", + "right": "issue:4304", + "accept": false, + "reason": "A metrics-validation request is not the same as a feature request for SDXL LoRA support." + }, + { + "left": "issue:1264", + "right": "issue:6607", + "accept": false, + "reason": "Both mention scheduler settings, but they refer to different example scripts and different questions." + }, + { + "left": "issue:13067", + "right": "issue:6158", + "accept": false, + "reason": "Both are feature requests, but for different model/pipeline support and not the same change." + }, + { + "left": "issue:2871", + "right": "issue:776", + "accept": false, + "reason": "Different pipeline bugs: latent output handling vs batch-size behavior in img2img." + }, + { + "left": "issue:2880", + "right": "issue:639", + "accept": false, + "reason": "Completely different subjects: JAX Colab runtime failure vs latent diffusion parameter initialization." + }, + { + "left": "issue:3689", + "right": "issue:4630", + "accept": false, + "reason": "Both involve LoRA after loading, but unloading weights and modifying weights are distinct requests." + }, + { + "left": "issue:4073", + "right": "issue:6298", + "accept": false, + "reason": "Different component attribute errors in different pipelines." + }, + { + "left": "issue:5862", + "right": "issue:7185", + "accept": false, + "reason": "Gradient checkpointing support and PyTorch version compatibility are not the same issue." + }, + { + "left": "issue:5873", + "right": "issue:7418", + "accept": false, + "reason": "ONNX export limitation vs a community pipeline bug; unrelated failures." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Rejecting because one soft-edge verdict is not grounded in the packet: it discusses issue:10575 vs issue:10738, but that pair does not appear in the provided soft_pairs list. The rest of the judgments are mostly conservative, and the 10266\u201311474 acceptance is plausible, but the inclusion of a nonexistent pair makes the output imprecise." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 73, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10049", + "issue:1962", + "issue:200", + "issue:201", + "issue:2015", + "issue:202", + "issue:2272", + "issue:2838", + "issue:338", + "issue:339", + "issue:3580", + "issue:3593", + "issue:3728", + "issue:3970", + "issue:4109", + "issue:4241", + "issue:4312", + "issue:4375", + "issue:4399", + "issue:5227", + "issue:6096", + "issue:6331", + "issue:6382", + "issue:6423", + "issue:6549", + "issue:6811", + "issue:6818", + "issue:692", + "issue:7146", + "issue:7147", + "issue:7168", + "issue:7303", + "issue:7766", + "issue:7799", + "issue:8307", + "issue:8577", + "issue:8655", + "issue:8667", + "issue:8700", + "issue:8710", + "issue:8788", + "issue:8813", + "issue:9067", + "issue:9115", + "issue:9448", + "issue:945", + "issue:9497", + "issue:9501" + ], + "soft_pairs": [ + "issue:6382|issue:7799", + "issue:7146|issue:9115", + "issue:7168|issue:7766", + "issue:7799|issue:9448", + "issue:8307|issue:9067", + "issue:4312|issue:6331", + "issue:4241|issue:692", + "issue:338|issue:339", + "issue:200|issue:201", + "issue:4241|issue:7303", + "issue:4375|issue:4399", + "issue:3580|issue:3593", + "issue:201|issue:202", + "issue:2272|issue:945", + "issue:3970|issue:4109", + "issue:8577|issue:8813", + "issue:8700|issue:8710", + "issue:9497|issue:9501", + "issue:2015|issue:3728", + "issue:6549|issue:6818", + "issue:8667|issue:8788", + "issue:2838|issue:6423", + "issue:2838|issue:7303", + "issue:10049|issue:7147", + "issue:1962|issue:3728", + "issue:5227|issue:6096", + "issue:6811|issue:8655" + ], + "prepared_review_unit_hash": "5ab23896ffc83d5d08a8279a5d946454a69978a86e672698a5825fcaed7c6777", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 34182, + "estimated_input_tokens": 8546, + "estimated_eval_tokens": 17348 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a loose thematic collection rather than one duplicate cluster. The only strong near-duplicate is the SAG inpainting request pair (4312/6331); the rest differ by concrete model, pipeline, or failure mode.", + "confidence": 0.41, + "canonical_issue_reason": "issue:4312 is the best representative because it is the broader, clearer feature request and subsumes issue:6331's narrower SAG-inpainting ask.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4312 is the most suitable global issue representative among the near-related items; it is the only one that cleanly overlaps a soft duplicate candidate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6382", + "right": "issue:7799", + "accept": false, + "reason": "Both are IP-Adapter-related, but one is a training key error and the other is an inference matmul shape mismatch in a different pipeline." + }, + { + "left": "issue:7146", + "right": "issue:9115", + "accept": false, + "reason": "Different problems: SDXL fine-tuning OOM versus AnimateDiff-Lightning / motion-adapter usage error." + }, + { + "left": "issue:7168", + "right": "issue:7766", + "accept": false, + "reason": "One asks how to handle IP-Adapter image embeddings, the other is about FaceID Plus usage; same area, not the same issue." + }, + { + "left": "issue:7799", + "right": "issue:9448", + "accept": false, + "reason": "Both mention IP-Adapter, but the failing pipelines and concrete errors are different." + }, + { + "left": "issue:8307", + "right": "issue:9067", + "accept": false, + "reason": "Unrelated missing-symbol problems: StableCascadeUNet versus diffusers.hub_utils." + }, + { + "left": "issue:4312", + "right": "issue:6331", + "accept": true, + "reason": "Same underlying feature request: support SAG for inpainting; 4312 is just broader by also mentioning img2img." + }, + { + "left": "issue:4241", + "right": "issue:692", + "accept": false, + "reason": "Both are about fine-tuning help, but one is a specific LoRA/unidiffusers request while the other is a general example request." + }, + { + "left": "issue:338", + "right": "issue:339", + "accept": false, + "reason": "Different scheduler test requests (LMS Discrete vs Karras VE), so not the same change." + }, + { + "left": "issue:200", + "right": "issue:201", + "accept": false, + "reason": "Different test targets: resnet class versus attention class." + }, + { + "left": "issue:4241", + "right": "issue:7303", + "accept": false, + "reason": "Fine-tuning support request versus scheduler zero-SNR question; not the same underlying change." + }, + { + "left": "issue:4375", + "right": "issue:4399", + "accept": false, + "reason": "Both involve torch 2 compile, but one is a crash during inference and the other is CPU offload compatibility; distinct bugs." + }, + { + "left": "issue:3580", + "right": "issue:3593", + "accept": false, + "reason": "Both involve xformers, but the reported failures are different and not clearly the same code path." + }, + { + "left": "issue:201", + "right": "issue:202", + "accept": false, + "reason": "Different test additions for different components (attention vs unet_blocks)." + }, + { + "left": "issue:2272", + "right": "issue:945", + "accept": false, + "reason": "Both are ONNX conversion-related, but one is a conversion script failing and the other is custom checkpoint conversion support." + }, + { + "left": "issue:3970", + "right": "issue:4109", + "accept": false, + "reason": "Different SDXL/IP-Adapter issues: prompt embeddings versus inpaint pipeline failure." + }, + { + "left": "issue:8577", + "right": "issue:8813", + "accept": false, + "reason": "Same feature family (differential diffusion), but they target different models and are not the same concrete request." + }, + { + "left": "issue:8700", + "right": "issue:8710", + "accept": false, + "reason": "PAG support requests, but one is a specific SDXL ControlNet img2img pipeline and the other is generic SD1.5 support." + }, + { + "left": "issue:9497", + "right": "issue:9501", + "accept": false, + "reason": "Both are Dreambooth Flux training topics, but the errors are different: save latency versus mat2 shape runtime error." + }, + { + "left": "issue:2015", + "right": "issue:3728", + "accept": false, + "reason": "Different pipeline requests: categorical diffusion versus shifted diffusion text-to-image." + }, + { + "left": "issue:6549", + "right": "issue:6818", + "accept": false, + "reason": "Both involve IP-Adapter and LoRA, but one is a warning-log issue and the other is an unload/constructor error." + }, + { + "left": "issue:8667", + "right": "issue:8788", + "accept": false, + "reason": "Both mention non-power-of-two image sizes, but one is SD3ControlNetModel shape mismatch and the other is VAE tiling support." + }, + { + "left": "issue:2838", + "right": "issue:6423", + "accept": false, + "reason": "One asks about ControlNet pipeline behavior, the other is about reproducing fill50k results; unrelated." + }, + { + "left": "issue:2838", + "right": "issue:7303", + "accept": false, + "reason": "Different requests: ControlNet pipeline behavior versus zero-SNR scheduler creation." + }, + { + "left": "issue:10049", + "right": "issue:7147", + "accept": false, + "reason": "Both are import errors, but for different symbols and different code paths." + }, + { + "left": "issue:1962", + "right": "issue:3728", + "accept": false, + "reason": "Video generation with stable diffusion is unrelated to shifted diffusion text-to-image generation." + }, + { + "left": "issue:5227", + "right": "issue:6096", + "accept": false, + "reason": "Both are torch.compile issues, but they fail in different scenarios and are not the same bug." + }, + { + "left": "issue:6811", + "right": "issue:8655", + "accept": false, + "reason": "Same model family (i2vgen-xl), but one is a missing model_index URL and the other is black GIF output." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge decisions stay conservative. The only accepted pair, issue:4312 / issue:6331, is strongly supported by near-identical SAG inpainting requests, with 4312 clearly broader. The rejections generally cite concrete differences in model, pipeline, or failure mode rather than overstating duplicate evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 74, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10049", + "issue:10940", + "issue:12412", + "issue:13232", + "issue:1572", + "issue:1677", + "issue:202", + "issue:2984", + "issue:3094", + "issue:337", + "issue:338", + "issue:339", + "issue:3432", + "issue:3453", + "issue:3706", + "issue:3728", + "issue:4042", + "issue:4185", + "issue:4206", + "issue:4305", + "issue:4356", + "issue:4731", + "issue:4779", + "issue:4828", + "issue:5227", + "issue:5314", + "issue:5433", + "issue:5689", + "issue:5889", + "issue:6069", + "issue:6232", + "issue:6914", + "issue:7209", + "issue:7243", + "issue:7330", + "issue:7681", + "issue:7690", + "issue:7716", + "issue:7931", + "issue:8493", + "issue:8589", + "issue:8723", + "issue:8827", + "issue:8828", + "issue:8975", + "issue:9148", + "issue:9202", + "issue:993" + ], + "soft_pairs": [ + "issue:10049|issue:8975", + "issue:7931|issue:9202", + "issue:5433|issue:5689", + "issue:4731|issue:5227", + "issue:5314|issue:9148", + "issue:8493|issue:8723", + "issue:4185|issue:4206", + "issue:7690|issue:7716", + "issue:1572|issue:337", + "issue:3706|issue:3728", + "issue:3728|issue:4779", + "issue:6914|issue:8827", + "issue:3094|issue:3453", + "issue:10940|issue:7681", + "issue:2984|issue:7243", + "issue:3432|issue:7330", + "issue:12412|issue:13232", + "issue:4305|issue:4356", + "issue:1677|issue:993", + "issue:4828|issue:7209", + "issue:4042|issue:5314", + "issue:202|issue:338", + "issue:202|issue:339", + "issue:8589|issue:8828", + "issue:2984|issue:5889", + "issue:6069|issue:6232" + ], + "prepared_review_unit_hash": "820b1e33f0b4fe99ce29adce1131eb7ac0bd74eccfabe9480c4b8f105facde47", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 34439, + "estimated_input_tokens": 8610, + "estimated_eval_tokens": 17476 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly heterogeneous, but a few issue pairs look like real duplicates: from_single_file offline-loading errors, img2img/inpainting pipeline failure, and a DPM++/scheduler add_noise bug. Most other pairs are only broadly similar and should stay separate.", + "confidence": 0.79, + "canonical_issue_reason": "No single issue cleanly represents the whole set because it spans unrelated topics; within the strongest duplicate-like scheduler pair, issue 6232 is the best umbrella since it states the broader add_noise failure while 6069 is a specific t=0 case.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6232 is the best global issue candidate among the duplicate-like pairs because it is the broadest, most reusable description of the scheduler bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10049", + "right": "issue:8975", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols/modules and no shared code path is evident." + }, + { + "left": "issue:7931", + "right": "issue:9202", + "accept": false, + "reason": "Both mention IP-Adapter, but they target different pipelines and different failure modes." + }, + { + "left": "issue:5433", + "right": "issue:5689", + "accept": false, + "reason": "Very similar DPM++/SDXL symptom, but one is artifacting without the refiner and the other is residual noise; not enough to confirm the same bug." + }, + { + "left": "issue:4731", + "right": "issue:5227", + "accept": false, + "reason": "Both involve torch.compile, but one is multi-ControlNet and the other is load_lora_weight, so the concrete failures differ." + }, + { + "left": "issue:5314", + "right": "issue:9148", + "accept": false, + "reason": "These are both inpainting-related feature requests, but for different models/pipelines and not the same change." + }, + { + "left": "issue:8493", + "right": "issue:8723", + "accept": false, + "reason": "Both are SD3 model issues, but one is state_dict loading for SD3Transformer2DModel and the other is SD3ControlNetModel initialization." + }, + { + "left": "issue:4185", + "right": "issue:4206", + "accept": false, + "reason": "One is SDXL ControlNet inference producing black images; the other is SDXL ControlNet training broken on main, so the code paths differ." + }, + { + "left": "issue:7690", + "right": "issue:7716", + "accept": false, + "reason": "Both are loading failures, but one is tokenizer lookup and the other is a missing config.json in a local model directory." + }, + { + "left": "issue:1572", + "right": "issue:337", + "accept": true, + "reason": "Both report img2img/inpainting failing in Stable Diffusion-style pipelines, with 1572 being the broader pipeline-specific version." + }, + { + "left": "issue:2984", + "right": "issue:7243", + "accept": false, + "reason": "Completely different model/paper requests with no shared bug or implementation target." + }, + { + "left": "issue:3432", + "right": "issue:7330", + "accept": false, + "reason": "Both are training/runtime failures, but one is CPU Half convolution support and the other is FP16 gradient unscaling in LoRA training." + }, + { + "left": "issue:3706", + "right": "issue:3728", + "accept": false, + "reason": "Different new-model/pipeline requests; same general area but not the same feature." + }, + { + "left": "issue:3728", + "right": "issue:4779", + "accept": false, + "reason": "Text-to-image shifted diffusion and image-to-video ModelScope are unrelated pipeline requests." + }, + { + "left": "issue:6914", + "right": "issue:8827", + "accept": false, + "reason": "The FaceID/IP-Adapter context overlaps, but the specific error and triggering condition are not clearly the same." + }, + { + "left": "issue:3094", + "right": "issue:3453", + "accept": false, + "reason": "Different training problems with different root causes: OOM vs unsupported bfloat16 CUDA op." + }, + { + "left": "issue:10940", + "right": "issue:7681", + "accept": false, + "reason": "Both are import-time failures, but they point to unrelated missing dependencies/environment problems." + }, + { + "left": "issue:2984", + "right": "issue:5889", + "accept": false, + "reason": "A research/model request versus a stable video diffusion feature request; not duplicates." + }, + { + "left": "issue:6069", + "right": "issue:6232", + "accept": true, + "reason": "These are the same scheduler.add_noise failure family; 6232 is the broader description and 6069 is a specific t=0 manifestation." + }, + { + "left": "issue:4042", + "right": "issue:5314", + "accept": false, + "reason": "Both mention reference/inpainting, but one asks how to combine features while the other requests a dedicated pipeline." + }, + { + "left": "issue:202", + "right": "issue:338", + "accept": false, + "reason": "Both are test-related, but they target different test additions and do not describe the same bug or change." + }, + { + "left": "issue:202", + "right": "issue:339", + "accept": false, + "reason": "Same as above: different test targets, not a duplicate." + }, + { + "left": "issue:8589", + "right": "issue:8828", + "accept": false, + "reason": "Both are SD3 training issues, but one is prior-preservation behavior and the other is a general 16GB training error." + }, + { + "left": "issue:4828", + "right": "issue:7209", + "accept": true, + "reason": "Both describe from_single_file offline/local loading hitting an unwanted network/connection path, so they look like the same underlying bug." + }, + { + "left": "issue:2984", + "right": "issue:7243", + "accept": false, + "reason": "Unrelated feature requests; no shared concrete problem." + }, + { + "left": "issue:4305", + "right": "issue:4356", + "accept": false, + "reason": "A torch.compile graph-break report and a missing dummy module import error are different issues." + }, + { + "left": "issue:1677", + "right": "issue:993", + "accept": false, + "reason": "Both are dtype-related, but one is a wrong-call-data-type problem and the other is a stable diffusion fp16 mismatch." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The accepted soft-edge pairs are well supported by the issue titles, especially 6069/6232, 4828/7209, and 1572/337. The rejected pairs are also reasonably distinguished by different pipelines, failure modes, or feature requests. No overstatement stood out." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 75, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10321", + "issue:1062", + "issue:1076", + "issue:10798", + "issue:10833", + "issue:11258", + "issue:1185", + "issue:12348", + "issue:12369", + "issue:12529", + "issue:13351", + "issue:1962", + "issue:2674", + "issue:3124", + "issue:3357", + "issue:3638", + "issue:3639", + "issue:3698", + "issue:4631", + "issue:4731", + "issue:4912", + "issue:5075", + "issue:5227", + "issue:5628", + "issue:5687", + "issue:6149", + "issue:6415", + "issue:6561", + "issue:6794", + "issue:7021", + "issue:7219", + "issue:7330", + "issue:7417", + "issue:7569", + "issue:7646", + "issue:7799", + "issue:8503", + "issue:8827", + "issue:8839", + "issue:9025", + "issue:9038", + "issue:9148", + "issue:9171", + "issue:9202", + "issue:9264", + "issue:9303", + "issue:9332" + ], + "soft_pairs": [ + "issue:10798|issue:10833", + "issue:3638|issue:3639", + "issue:8503|issue:8839", + "issue:6149|issue:6415", + "issue:7330|issue:7646", + "issue:3357|issue:5227", + "issue:1076|issue:1962", + "issue:7799|issue:9202", + "issue:7219|issue:9303", + "issue:3357|issue:4731", + "issue:9171|issue:9332", + "issue:12369|issue:13351", + "issue:9148|issue:9264", + "issue:5628|issue:5687", + "issue:4631|issue:5075", + "issue:6561|issue:8827", + "issue:12348|issue:12529", + "issue:3698|issue:4912", + "issue:9025|issue:9038", + "issue:2674|issue:3124", + "issue:10321|issue:11258", + "issue:6794|issue:7021", + "issue:7417|issue:7569", + "issue:1062|issue:1185" + ], + "prepared_review_unit_hash": "7e7ab5c52a588ac02993f646d7cfc8e313cbce6983f86eb4b5ef135207c68c98", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33158, + "estimated_input_tokens": 8290, + "estimated_eval_tokens": 16836 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Only one soft edge looks like a true duplicate: the LTX single-file loading/support reports. The rest are title-level similarities across different models, pipelines, or unrelated bugs/features and should stay separate.", + "confidence": 0.62, + "canonical_issue_reason": "issue:11258 is the broader LTX single-file support request and best captures the underlying problem behind the missing-file error in issue:10321.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11258 is the clearest representative of the LTX single-file issue family; issue:10321 reads as a specific failure symptom of that same gap.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10798", + "right": "issue:10833", + "accept": false, + "reason": "Both mention Flux + bitsandbytes quantization, but they describe different bugs: one is a device-combination issue, the other a FluxFillPipeline matmul shape failure." + }, + { + "left": "issue:3638", + "right": "issue:3639", + "accept": false, + "reason": "These are separate new-pipeline proposals (Prompt-Free Diffusion vs GlyphControl), not the same feature or bug." + }, + { + "left": "issue:8503", + "right": "issue:8839", + "accept": false, + "reason": "Both are SD3-related, but one is a gradient-checkpointing crash and the other is an SD3 LoRA weight error; different code paths and symptoms." + }, + { + "left": "issue:6149", + "right": "issue:6415", + "accept": false, + "reason": "Different IP-Adapter topics: one is a specific full-face_sd15 load error, the other is support for FaceID/FaceID Plus." + }, + { + "left": "issue:7330", + "right": "issue:7646", + "accept": false, + "reason": "Same training script, but unrelated problems: FP16 gradient unscale error versus dataloader worker configuration." + }, + { + "left": "issue:3357", + "right": "issue:5227", + "accept": false, + "reason": "Both involve torch.compile, but the underlying failures differ: a Dynamo hasattr limitation versus load_lora_weight incompatibility." + }, + { + "left": "issue:1076", + "right": "issue:1962", + "accept": false, + "reason": "These are unrelated questions about stable diffusion and video generation, not the same underlying bug or change." + }, + { + "left": "issue:7799", + "right": "issue:9202", + "accept": false, + "reason": "Both are IP-Adapter shape mismatches, but they affect different pipelines/models and appear to be different failures." + }, + { + "left": "issue:7219", + "right": "issue:9303", + "accept": false, + "reason": "Both are upscaler requests, but SUPIR and VEnhancer for CogVideoX-5b are distinct feature requests." + }, + { + "left": "issue:3357", + "right": "issue:4731", + "accept": false, + "reason": "Both mention torch.compile, but one is a TensorVariable hasattr error and the other is a multi-controlnet SDXL compile failure." + }, + { + "left": "issue:9171", + "right": "issue:9332", + "accept": false, + "reason": "Different loading bugs: from_single_file regression for SD 1.5 v_prediction versus from_pretrained loading the wrong identical weights." + }, + { + "left": "issue:12369", + "right": "issue:13351", + "accept": false, + "reason": "These are unrelated feature requests around different Wan/Fun ControlNet topics, not a duplicate bug/change." + }, + { + "left": "issue:9148", + "right": "issue:9264", + "accept": false, + "reason": "Both concern Flux inpainting, but one requests adding the pipeline and the other asks for an inpainting model for flux; not the same artifact." + }, + { + "left": "issue:5628", + "right": "issue:5687", + "accept": false, + "reason": "Both involve Karras sigmas/timesteps, but the concrete bugs differ: denoising offset in LMS scheduler versus duplicate timesteps breaking Img2Img." + }, + { + "left": "issue:4631", + "right": "issue:5075", + "accept": false, + "reason": "Both are matmul/runtime shape errors, but they occur in different scheduler/pipeline contexts and are not clearly the same code-path bug." + }, + { + "left": "issue:6561", + "right": "issue:8827", + "accept": false, + "reason": "Both are IP-Adapter loading/usage errors, but one is a missing config file and the other is an error using faceid_sdxl; different failures." + }, + { + "left": "issue:12348", + "right": "issue:12529", + "accept": false, + "reason": "Both are tensor-size mismatches in Wan models, but one is resolution-specific in WanImageToVideoPipeline and the other is a tiled-VAE mismatch in Wan2.2 TI2V-5B." + }, + { + "left": "issue:3698", + "right": "issue:4912", + "accept": false, + "reason": "Both are about conversion to diffusers, but one targets inpainting models and the other custom CivitAI SDXL checkpoints; not the same change." + }, + { + "left": "issue:9025", + "right": "issue:9038", + "accept": false, + "reason": "These are unrelated FlaxStableDiffusionPipeline questions: single-file loading on Kaggle TPU versus prompt weighting." + }, + { + "left": "issue:2674", + "right": "issue:3124", + "accept": false, + "reason": "Different model-conversion/LoRA dependency issues; they do not describe the same bug." + }, + { + "left": "issue:10321", + "right": "issue:11258", + "accept": true, + "reason": "Both point to the same LTX single-file loading gap: issue:10321 is the missing-model-file error, while issue:11258 is the broader single-file support request." + }, + { + "left": "issue:6794", + "right": "issue:7021", + "accept": false, + "reason": "These are unrelated import/environment errors: one is a training-utils import issue, the other is HF_HOME import configuration." + }, + { + "left": "issue:7417", + "right": "issue:7569", + "accept": false, + "reason": "Both are Mac test failures, but they involve different tests and likely different regressions." + }, + { + "left": "issue:1062", + "right": "issue:1185", + "accept": false, + "reason": "Both mention Dreambooth reproducibility, but they are separate reports with no evidence of the same underlying bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The single accepted edge (issue:10321 \u2194 issue:11258) is plausibly the same LTX single-file support gap, while the rejected pairs are explained in a cautious way without overstating duplication. No mergeability issues apply here since these are issue-to-issue pairs, not PRs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 76, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10518", + "issue:10591", + "issue:10893", + "issue:1101", + "issue:11104", + "issue:11258", + "issue:11374", + "issue:2163", + "issue:2471", + "issue:2531", + "issue:2603", + "issue:2780", + "issue:280", + "issue:2830", + "issue:2838", + "issue:2883", + "issue:3184", + "issue:3279", + "issue:3454", + "issue:3565", + "issue:3700", + "issue:3728", + "issue:3774", + "issue:4178", + "issue:4263", + "issue:4376", + "issue:4631", + "issue:4704", + "issue:5017", + "issue:5399", + "issue:5761", + "issue:5960", + "issue:5997", + "issue:6091", + "issue:6183", + "issue:6224", + "issue:6879", + "issue:7147", + "issue:7330", + "issue:8649", + "issue:9042", + "issue:938", + "issue:9410", + "issue:9443", + "issue:9551", + "issue:9665" + ], + "soft_pairs": [ + "issue:2163|issue:3565", + "issue:10893|issue:11374", + "issue:3728|issue:938", + "issue:5960|issue:5997", + "issue:6224|issue:7330", + "issue:4178|issue:4263", + "issue:10518|issue:10591", + "issue:3565|issue:3774", + "issue:4704|issue:6879", + "issue:8649|issue:9443", + "issue:7147|issue:9410", + "issue:4376|issue:6183", + "issue:11104|issue:11258", + "issue:3279|issue:3454", + "issue:10026|issue:9665", + "issue:2838|issue:5017", + "issue:2780|issue:3700", + "issue:5761|issue:6091", + "issue:2471|issue:2603", + "issue:1101|issue:280", + "issue:2830|issue:2883", + "issue:2531|issue:3184", + "issue:4631|issue:5399", + "issue:9042|issue:9551" + ], + "prepared_review_unit_hash": "bc8938b55d28108414e056db5f9bb4d923114ed42b1f40d6f9b114d37cedd421", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33395, + "estimated_input_tokens": 8349, + "estimated_eval_tokens": 16954 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a loose similarity bucket of unrelated issues. Only a few pairs look like true duplicates or near-duplicates around the same concrete bug/feature request; the rest are broader topic overlaps only.", + "confidence": 0.41, + "canonical_issue_reason": "issue:4263 is the clearest representative of a real duplicate set: it is the broader, more canonical description of the LoRA unloading bug, and issue:4178 looks like a specific xformers-related manifestation of the same failure.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4263 is the best overall issue to anchor the cluster because it states the underlying bug most generically and has the stronger discussion signal among the clearly related items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2163", + "right": "issue:3565", + "accept": false, + "reason": "Both are training-example problems, but one is a dtype mismatch and the other is OOM; different failure modes." + }, + { + "left": "issue:10893", + "right": "issue:11374", + "accept": false, + "reason": "Both are feature requests for model support, but for different models and different integrations." + }, + { + "left": "issue:3728", + "right": "issue:938", + "accept": false, + "reason": "Different requested pipelines: text-to-image shifted diffusion vs text-to-motion human motion diffusion." + }, + { + "left": "issue:5960", + "right": "issue:5997", + "accept": false, + "reason": "Different ControlNet SDXL problems: a runtime warning in training vs a tokenizer/model type mismatch in the tutorial." + }, + { + "left": "issue:6224", + "right": "issue:7330", + "accept": false, + "reason": "Both are training issues, but one is TorchDynamo-related and the other is FP16 gradient unscale handling." + }, + { + "left": "issue:4178", + "right": "issue:4263", + "accept": true, + "reason": "Same underlying LoRA unloading bug; 4178 is a specific xformers symptom of the broader broken-unloading issue." + }, + { + "left": "issue:10518", + "right": "issue:10591", + "accept": true, + "reason": "Near-identical SD3 LoRA training-script bug reports with the same vague failure description." + }, + { + "left": "issue:3565", + "right": "issue:3774", + "accept": false, + "reason": "Both mention memory, but one is plain OOM in text-to-image and the other is a zero3/DeepSpeed memory-sharding complaint." + }, + { + "left": "issue:4704", + "right": "issue:6879", + "accept": false, + "reason": "Both concern Custom Diffusion, but the titles point to different bugs: generic failure vs a specific dtype mismatch." + }, + { + "left": "issue:8649", + "right": "issue:9443", + "accept": false, + "reason": "SD3 issues, but one is about num_images_per_prompt behavior and the other about tokenizer instantiation from slow version." + }, + { + "left": "issue:7147", + "right": "issue:9410", + "accept": false, + "reason": "Both are import errors, but for different pipeline classes and different code paths." + }, + { + "left": "issue:4376", + "right": "issue:6183", + "accept": false, + "reason": "Different Stable Diffusion XL problems: textual inversion crash vs single-file shape mismatch." + }, + { + "left": "issue:11104", + "right": "issue:11258", + "accept": false, + "reason": "Both are LTX-Video related, but one is offload-method breakage and the other is a single-file loading topic." + }, + { + "left": "issue:3279", + "right": "issue:3454", + "accept": false, + "reason": "Both concern the same script family, but one is checkpoint resume and the other is a general latest-version regression." + }, + { + "left": "issue:10026", + "right": "issue:9665", + "accept": false, + "reason": "Both mention DeepSpeed/training, but they describe different problems and code paths." + }, + { + "left": "issue:2838", + "right": "issue:5017", + "accept": false, + "reason": "Both are ControlNet feature requests, but the requested change is not concrete enough to treat as the same bug/change." + }, + { + "left": "issue:2780", + "right": "issue:3700", + "accept": false, + "reason": "Different LoRA training-script issues: FP16 gradient scaling vs min_snr argument support." + }, + { + "left": "issue:5761", + "right": "issue:6091", + "accept": false, + "reason": "Both touch consistency decoder, but one is a cost/efficiency concern and the other is a size-argument memory issue." + }, + { + "left": "issue:2471", + "right": "issue:2603", + "accept": true, + "reason": "Same Attend-and-Excite size/shape limitation: both report failure away from the default square-image case." + }, + { + "left": "issue:1101", + "right": "issue:280", + "accept": false, + "reason": "Different scheduler feature requests: DPM-Solver vs DEIS." + }, + { + "left": "issue:2830", + "right": "issue:2883", + "accept": false, + "reason": "Both relate to video generation, but one asks about a specific model and the other requests a generic pipeline." + }, + { + "left": "issue:2531", + "right": "issue:3184", + "accept": false, + "reason": "Same community pipeline, but the bugs are different: meta-tensor offload failure vs generator-list handling." + }, + { + "left": "issue:4631", + "right": "issue:5399", + "accept": false, + "reason": "Unrelated concrete failures: scheduler scalar conversion error vs custom inpainting dtype mismatch." + }, + { + "left": "issue:9042", + "right": "issue:9551", + "accept": false, + "reason": "One requests a Flux pipeline; the other asks how to use Flux ControlNet models. Related topic, not the same change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative overall. The three accepted soft edges are the only ones that read like strong duplicate/near-duplicate matches from titles alone, and the rejected pairs are framed cautiously with appropriate distinctions." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 77, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10150", + "issue:10302", + "issue:10334", + "issue:10462", + "issue:10674", + "issue:10797", + "issue:10914", + "issue:11229", + "issue:12037", + "issue:12047", + "issue:12358", + "issue:13311", + "issue:13547", + "issue:1594", + "issue:1962", + "issue:201", + "issue:228", + "issue:2689", + "issue:2722", + "issue:3382", + "issue:3406", + "issue:3453", + "issue:3468", + "issue:3694", + "issue:3728", + "issue:3826", + "issue:4305", + "issue:4686", + "issue:4731", + "issue:4874", + "issue:5061", + "issue:5219", + "issue:5374", + "issue:5886", + "issue:6008", + "issue:6096", + "issue:6427", + "issue:6561", + "issue:671", + "issue:7223", + "issue:8503", + "issue:8649", + "issue:8787", + "issue:8839", + "issue:9303", + "issue:9850", + "issue:9873" + ], + "soft_pairs": [ + "issue:228|issue:2689", + "issue:12037|issue:12047", + "issue:201|issue:3406", + "issue:5886|issue:6561", + "issue:10674|issue:10797", + "issue:3382|issue:3453", + "issue:4305|issue:6008", + "issue:5061|issue:5374", + "issue:1594|issue:1962", + "issue:5219|issue:6427", + "issue:8649|issue:8839", + "issue:3728|issue:7223", + "issue:8503|issue:9850", + "issue:12358|issue:9873", + "issue:10302|issue:10914", + "issue:3468|issue:3694", + "issue:10334|issue:11229", + "issue:10462|issue:9303", + "issue:3728|issue:3826", + "issue:4686|issue:4874", + "issue:10150|issue:8787", + "issue:4731|issue:6096", + "issue:13311|issue:13547", + "issue:2722|issue:671" + ], + "prepared_review_unit_hash": "c50ec72c72c5e5263446112c7beb882c98e5d13a3d4fb5cd258f942c08494221", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33142, + "estimated_input_tokens": 8286, + "estimated_eval_tokens": 16828 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issue pairs; only the model-loading slowdown reports look plausibly like the same bug. The rest share subsystem keywords but describe different failures, features, or model-specific regressions.", + "confidence": 0.46, + "canonical_issue_reason": "issue:8787 is the broadest representative for the slowdown regression and can cover the more specific SDXL loading/inference complaint in issue:10150.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8787 is the best global anchor because it names the generic slowdown symptom, making it the most reusable duplicate target among the candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:228", + "right": "issue:2689", + "accept": false, + "reason": "Both mention broken links, but one is a specific release asset and the other is a generic image-link report; not enough evidence of the same underlying defect." + }, + { + "left": "issue:12037", + "right": "issue:12047", + "accept": false, + "reason": "Same model family and LoRA terminology, but one is compatibility/self-forcing guidance and the other is GGUF fusion failure; likely distinct code paths." + }, + { + "left": "issue:201", + "right": "issue:3406", + "accept": false, + "reason": "Both are about attention classes, but the concrete failure differs: a test request versus a specific only_cross_attention bug." + }, + { + "left": "issue:5886", + "right": "issue:6561", + "accept": false, + "reason": "IP-Adapter-related, but one asks for face-model support while the other is a missing config.json load error; not the same bug." + }, + { + "left": "issue:10674", + "right": "issue:10797", + "accept": false, + "reason": "Different pipelines and different offloading/GGUF behavior; too little to treat as one underlying issue." + }, + { + "left": "issue:3382", + "right": "issue:3453", + "accept": false, + "reason": "Both are training OOM/runtime problems, but one is multi-GPU memory pressure and the other is a BFloat16 kernel limitation." + }, + { + "left": "issue:4305", + "right": "issue:6008", + "accept": false, + "reason": "Shared release-regression framing only; one is a torch.compile graph break and the other is IP-Adapter face + LCM LoRA failure." + }, + { + "left": "issue:5061", + "right": "issue:5374", + "accept": false, + "reason": "Both are converter argument errors, but they reference different CLI parameters and likely different call sites." + }, + { + "left": "issue:1594", + "right": "issue:1962", + "accept": false, + "reason": "Related to video generation, but one is an img2img feature request and the other is general video generation support." + }, + { + "left": "issue:5219", + "right": "issue:6427", + "accept": false, + "reason": "Both concern conversion/loading, but one is loading an SDXL safetensors export and the other is converting diffusers to .ckpt; opposite directions and different symptoms." + }, + { + "left": "issue:8649", + "right": "issue:8839", + "accept": false, + "reason": "Both involve SD3, but one is a num_images_per_prompt error and the other is a LoRA weight issue." + }, + { + "left": "issue:3728", + "right": "issue:7223", + "accept": false, + "reason": "Both are new pipeline proposals, but they are different projects and not the same pipeline change." + }, + { + "left": "issue:8503", + "right": "issue:9850", + "accept": false, + "reason": "Gradient checkpointing appears in both, but one is an SD3 crash and the other is a feature request for frozen models; not the same bug." + }, + { + "left": "issue:12358", + "right": "issue:9873", + "accept": false, + "reason": "Both are pipeline-addition requests, but for different models and different feature scopes." + }, + { + "left": "issue:10302", + "right": "issue:10914", + "accept": false, + "reason": "Both mention offloading, but one reports FP8 noise without CPU offloading while the other reports unintended CPU offloading; different problems." + }, + { + "left": "issue:3468", + "right": "issue:3694", + "accept": false, + "reason": "Both involve missing model_index.json, but the contexts differ enough\u2014Dreambooth output generation versus a finetuning example load path\u2014to avoid merging." + }, + { + "left": "issue:10334", + "right": "issue:11229", + "accept": false, + "reason": "Both mention MPS/NaNs, but the first is a Sana MacOS regression and the second is attention slicing on SDXL; different triggers and models." + }, + { + "left": "issue:10462", + "right": "issue:9303", + "accept": false, + "reason": "One is a customization issue for CogVideoX-5B and the other is a request to add an upscaler/interpolator pipeline; not the same change." + }, + { + "left": "issue:3728", + "right": "issue:3826", + "accept": false, + "reason": "Both are new pipeline requests, but for different modalities and models; only a broad category match." + }, + { + "left": "issue:4686", + "right": "issue:4874", + "accept": false, + "reason": "Both touch StableDiffusionUpscalePipeline local loading, but one is a from_pretrained load failure and the other is a low_res_scheduler-specific bug." + }, + { + "left": "issue:10150", + "right": "issue:8787", + "accept": true, + "reason": "Both report slowdown in model loading/inference, and issue:8787 is the broader generic version of the same performance regression." + }, + { + "left": "issue:4731", + "right": "issue:6096", + "accept": false, + "reason": "Both mention torch.compile, but one is a multi-ControlNet SDXL failure and the other is an SVD dynamic output shape exception." + }, + { + "left": "issue:13311", + "right": "issue:13547", + "accept": false, + "reason": "Both are FLUX.2-klein LoRA loading problems, but one is a missing key-prefix association and the other is a generic adapter-loading error; likely distinct." + }, + { + "left": "issue:2722", + "right": "issue:671", + "accept": false, + "reason": "Both concern converted models not matching, but one is ControlNet UNet conversion and the other is original Stable Diffusion conversion identity; different artifacts and failure modes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:10150 with issue:8787, is plausibly the same slowdown regression based on the titles, and the rejected pairs are generally justified as related only by broad subsystem keywords rather than the same defect. No obvious overstatement." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 78, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10202", + "issue:10594", + "issue:10639", + "issue:1082", + "issue:11010", + "issue:11182", + "issue:11408", + "issue:12037", + "issue:12052", + "issue:12367", + "issue:12412", + "issue:1280", + "issue:1305", + "issue:13065", + "issue:13162", + "issue:1350", + "issue:13547", + "issue:1594", + "issue:2132", + "issue:2390", + "issue:2412", + "issue:2596", + "issue:2984", + "issue:3422", + "issue:3564", + "issue:3728", + "issue:3884", + "issue:4427", + "issue:4584", + "issue:4708", + "issue:4753", + "issue:4754", + "issue:4849", + "issue:488", + "issue:5108", + "issue:5168", + "issue:5284", + "issue:5576", + "issue:5623", + "issue:5849", + "issue:7021", + "issue:7646", + "issue:8511", + "issue:8594", + "issue:9576", + "issue:9628", + "issue:9876" + ], + "soft_pairs": [ + "issue:4427|issue:5108", + "issue:4708|issue:7021", + "issue:1280|issue:488", + "issue:12367|issue:13065", + "issue:1594|issue:2984", + "issue:1305|issue:1350", + "issue:4584|issue:5849", + "issue:5168|issue:5284", + "issue:3728|issue:5576", + "issue:8511|issue:9876", + "issue:11182|issue:9576", + "issue:4754|issue:4849", + "issue:12037|issue:12052", + "issue:10639|issue:12412", + "issue:2596|issue:4708", + "issue:3422|issue:5623", + "issue:8594|issue:9628", + "issue:11010|issue:11408", + "issue:3884|issue:7646", + "issue:2132|issue:4753", + "issue:13162|issue:13547", + "issue:2390|issue:3564", + "issue:10202|issue:10594", + "issue:1082|issue:2412" + ], + "prepared_review_unit_hash": "2a1cb77d7e3e97cd12f27db54f57810a035d2c17885f7a13d7afc9c9c79e08d4", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32912, + "estimated_input_tokens": 8228, + "estimated_eval_tokens": 16712 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a single coherent duplicate cluster: most items are unrelated model-support or training-error reports. I only see one likely true duplicate pair around grayscale/single-channel image handling.", + "confidence": 0.78, + "canonical_issue_reason": "No single canonical issue cleanly represents the whole set; the items span many unrelated topics, so a global duplicate target would be unsafe.", + "canonical_pr_reason": null, + "best_issue_reason": "488 is the clearest representative only for the grayscale/single-channel image bug, but it is not a good global canonical for the full mixed cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4427", + "right": "issue:5108", + "accept": false, + "reason": "Same broad T2I-MultiAdapter area, but different example variants and failure modes; not the same concrete bug." + }, + { + "left": "issue:4708", + "right": "issue:7021", + "accept": false, + "reason": "Both mention import problems, but one is about diffusers.utils and the other HF_HOME; too different to be duplicates." + }, + { + "left": "issue:1280", + "right": "issue:488", + "accept": true, + "reason": "Both describe grayscale/single-channel image handling failures in pipelines; same underlying bug." + }, + { + "left": "issue:12367", + "right": "issue:13065", + "accept": false, + "reason": "Different Qwen-Image feature requests: LoRA support vs controlnet inpaint/canny combination." + }, + { + "left": "issue:1594", + "right": "issue:2984", + "accept": false, + "reason": "Both are video-related, but they ask for different features and are not the same change." + }, + { + "left": "issue:1305", + "right": "issue:1350", + "accept": false, + "reason": "Completely unrelated topics: direct inversion editing versus object detection." + }, + { + "left": "issue:4584", + "right": "issue:5849", + "accept": false, + "reason": "Same training script family, but one is a checkpoint-resume error and the other is an accelerate entrypoint TypeError." + }, + { + "left": "issue:5168", + "right": "issue:5284", + "accept": false, + "reason": "Both are ControlNet variants, but they are distinct model-support requests." + }, + { + "left": "issue:3728", + "right": "issue:5576", + "accept": false, + "reason": "Different scheduler proposals; not the same underlying feature." + }, + { + "left": "issue:8511", + "right": "issue:9876", + "accept": false, + "reason": "Both involve LoRA VRAM usage, but one is specifically about SD3 text encoders not unloading while the other is a broad memory-release complaint." + }, + { + "left": "issue:11182", + "right": "issue:9576", + "accept": false, + "reason": "Different pipeline-loading bugs: unnecessary BF16 downloads versus use_safetensors propagation." + }, + { + "left": "issue:4754", + "right": "issue:4849", + "accept": false, + "reason": "One is a load-time missing model_index.json error; the other is a multi-GPU training question." + }, + { + "left": "issue:12037", + "right": "issue:12052", + "accept": false, + "reason": "Both involve Wan 2.2 and LightX2V, but one is a LoRA compatibility issue and the other is a device-mismatch offloading bug." + }, + { + "left": "issue:10639", + "right": "issue:12412", + "accept": false, + "reason": "Separate model support requests for Lumina-Image 2.0 and HunyuanImage-3.0." + }, + { + "left": "issue:2596", + "right": "issue:4708", + "accept": false, + "reason": "Both mention diffusers.utils, but one is a docs typo and the other is a generic import failure; not enough evidence of the same bug." + }, + { + "left": "issue:3422", + "right": "issue:5623", + "accept": false, + "reason": "Both touch safety checker behavior, but one is a disable request and the other is an enhancement request." + }, + { + "left": "issue:8594", + "right": "issue:9628", + "accept": false, + "reason": "Same SD3 dreambooth/LoRA script, but different concrete failures and code paths." + }, + { + "left": "issue:11010", + "right": "issue:11408", + "accept": false, + "reason": "Flux model support versus compiled-model adapter support; unrelated changes." + }, + { + "left": "issue:3884", + "right": "issue:7646", + "accept": false, + "reason": "Same script name, but one is an execution error and the other is a dataloader_num_workers question." + }, + { + "left": "issue:2132", + "right": "issue:4753", + "accept": false, + "reason": "Both are dreambooth_lora.py reports, but they describe different errors and no shared root cause is clear." + }, + { + "left": "issue:13162", + "right": "issue:13547", + "accept": false, + "reason": "Same Flux2-klein LoRA area, but one is support and the other is an adapter-loading error; not the same bug." + }, + { + "left": "issue:2390", + "right": "issue:3564", + "accept": false, + "reason": "Different control-method support requests: T2I-Adapter versus Uni-ControlNet." + }, + { + "left": "issue:10202", + "right": "issue:10594", + "accept": false, + "reason": "Both mention FLUX and LoRA, but one is about unloading a LoRA and the other about combining multiple controls with LoRA." + }, + { + "left": "issue:1082", + "right": "issue:2412", + "accept": false, + "reason": "Two different Dreambooth failures with no clear shared code-path." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: the cluster is clearly mixed, and the only confidently duplicate-looking pair is 1280 \u2194 488 on grayscale/single-channel image handling. The other pair verdicts stay conservative and are supported by the issue titles/bodies; no PR-merge concerns apply here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 79, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10321", + "issue:10460", + "issue:10760", + "issue:11010", + "issue:11024", + "issue:11104", + "issue:11363", + "issue:11432", + "issue:12221", + "issue:1574", + "issue:2177", + "issue:2321", + "issue:2412", + "issue:2466", + "issue:2728", + "issue:3613", + "issue:378", + "issue:4704", + "issue:4837", + "issue:4852", + "issue:4911", + "issue:4993", + "issue:5227", + "issue:579", + "issue:6069", + "issue:6073", + "issue:6162", + "issue:6503", + "issue:6520", + "issue:6527", + "issue:6545", + "issue:6565", + "issue:6709", + "issue:6741", + "issue:7159", + "issue:7210", + "issue:7355", + "issue:7538", + "issue:8307", + "issue:861", + "issue:8939", + "issue:9209", + "issue:9223", + "issue:9370", + "issue:9546", + "issue:9551", + "issue:9643", + "issue:9971" + ], + "soft_pairs": [ + "issue:11363|issue:9546", + "issue:2412|issue:6073", + "issue:11010|issue:9643", + "issue:6565|issue:6741", + "issue:10321|issue:11104", + "issue:10460|issue:9370", + "issue:12221|issue:9551", + "issue:2321|issue:2466", + "issue:10760|issue:7210", + "issue:11024|issue:8307", + "issue:11432|issue:9971", + "issue:6069|issue:6527", + "issue:5227|issue:7159", + "issue:4993|issue:9223", + "issue:378|issue:579", + "issue:4704|issue:6709", + "issue:6520|issue:7538", + "issue:8939|issue:9209", + "issue:1574|issue:3613", + "issue:4852|issue:4911", + "issue:2177|issue:2728", + "issue:6162|issue:7355", + "issue:4837|issue:861", + "issue:6503|issue:6545" + ], + "prepared_review_unit_hash": "a4953576e08a2b1abfe34f8dca38e095eaab44a933212b5c9f3028879b8f7bfd", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34355, + "estimated_input_tokens": 8589, + "estimated_eval_tokens": 17434 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Highly heterogeneous issue cluster; the soft pairs share only loose topical similarity, not the same concrete bug/change. No PRs present.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 6503 is the closest canonical representative because it is an umbrella/tracker-style item about training-script compatibility and has the strongest centrality among the set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6503 is the best issue anchor only by breadth and reference count; the rest are mostly unrelated narrow bugs, API mismatches, or feature requests, so this is not a true duplicate hub.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11363", + "right": "issue:9546", + "accept": false, + "reason": "Both mention OOM, but one is a ControlNet inference example and the other is FLUX ControlNet validation during training; different code paths and likely different fixes." + }, + { + "left": "issue:2412", + "right": "issue:6073", + "accept": false, + "reason": "Both are Dreambooth-related errors, but the reports describe different workflows and failure contexts; no evidence of the same underlying bug." + }, + { + "left": "issue:11010", + "right": "issue:9643", + "accept": false, + "reason": "One is a request for Chroma/Flux architecture changes, the other asks about multiple ControlNets in Flux; these are different feature/support questions." + }, + { + "left": "issue:6565", + "right": "issue:6741", + "accept": false, + "reason": "Both are Diffusion DPO training issues, but one is about loss spikes/noised samples and the other is a missing CLI argument; different defects." + }, + { + "left": "issue:10321", + "right": "issue:11104", + "accept": false, + "reason": "Both involve LTX-Video, but one is a missing-model-file load error while the other says both offload methods are broken; different failure modes." + }, + { + "left": "issue:10460", + "right": "issue:9370", + "accept": false, + "reason": "These are separate API signature bugs on different classes with different missing arguments (`mid_block_type` vs `pooled_projection`)." + }, + { + "left": "issue:12221", + "right": "issue:9551", + "accept": false, + "reason": "Both are Flux ControlNet support topics, but they ask for different integrations and usage guidance rather than the same bug/change." + }, + { + "left": "issue:2321", + "right": "issue:2466", + "accept": false, + "reason": "One concerns EMA/model copying in inference, the other checkpoint retention behavior during training; not the same underlying issue." + }, + { + "left": "issue:10760", + "right": "issue:7210", + "accept": false, + "reason": "A missing import/export attribute for a pipeline class is unrelated to a `named_parameters` attribute error on a pipeline object." + }, + { + "left": "issue:11024", + "right": "issue:8307", + "accept": false, + "reason": "These affect different classes/models (`SiglipImageProcessor` vs `StableCascadeUNet`) and different error conditions." + }, + { + "left": "issue:11432", + "right": "issue:9971", + "accept": false, + "reason": "`from_pretrained(torch_dtype=\"auto\")` behavior is unrelated to DEISMultistepScheduler not working on FLUX." + }, + { + "left": "issue:6069", + "right": "issue:6527", + "accept": false, + "reason": "Both are `add_noise` complaints, but they involve different schedulers and different edge conditions, so they are not the same bug." + }, + { + "left": "issue:5227", + "right": "issue:7159", + "accept": false, + "reason": "Both mention `torch.compile`, but one is about `load_lora_weight` and the other about fused LoRA still failing under compile; distinct code paths." + }, + { + "left": "issue:4993", + "right": "issue:9223", + "accept": false, + "reason": "A missing export for `CLIPImageProjection` is unrelated to a dependency version conflict around `pyarrow`." + }, + { + "left": "issue:378", + "right": "issue:579", + "accept": false, + "reason": "Both are textual inversion issues, but one is about a distribution class and the other about a missing config file; different root causes." + }, + { + "left": "issue:4704", + "right": "issue:6709", + "accept": false, + "reason": "Both are Custom Diffusion training reports, but one is a general training failure and the other is a concept-list-specific bug." + }, + { + "left": "issue:6520", + "right": "issue:7538", + "accept": false, + "reason": "A blank validation output in T2I adapter training is unrelated to a Custom Fill50k style dataset training issue." + }, + { + "left": "issue:8939", + "right": "issue:9209", + "accept": false, + "reason": "One is a tokenizer/model loading path problem; the other is a mirror connectivity failure. Different causes and fixes." + }, + { + "left": "issue:1574", + "right": "issue:3613", + "accept": false, + "reason": "Both are Dreambooth memory problems, but the reports are about different setups and symptoms; not enough to treat as the same concrete bug." + }, + { + "left": "issue:4852", + "right": "issue:4911", + "accept": false, + "reason": "Loading LoRA weights and loading a custom text encoder in `from_single_file` are different loading pathways and errors." + }, + { + "left": "issue:2177", + "right": "issue:2728", + "accept": false, + "reason": "These are different model classes and different loading failures (`AutoencoderKL` vs `UNet2DConditionModel` with safetensors)." + }, + { + "left": "issue:6162", + "right": "issue:7355", + "accept": false, + "reason": "An IP-Adapter SDXL shape mismatch is unrelated to Stable Cascade arbitrary-resolution/bfloat16 generation failures." + }, + { + "left": "issue:4837", + "right": "issue:861", + "accept": false, + "reason": "Both touch the converter script, but one is an SDXL conversion failure and the other is a general ckpt-to-diffusers conversion problem; not clearly the same bug." + }, + { + "left": "issue:6503", + "right": "issue:6545", + "accept": false, + "reason": "Both are tracker-style training issues, but they cover different initiatives: torch.compile compatibility versus modularizing inference during/after training." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 80, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10150", + "issue:10202", + "issue:10594", + "issue:10925", + "issue:11024", + "issue:11109", + "issue:11363", + "issue:11464", + "issue:11642", + "issue:12082", + "issue:12565", + "issue:12599", + "issue:12613", + "issue:12699", + "issue:1350", + "issue:1823", + "issue:228", + "issue:2984", + "issue:3019", + "issue:3728", + "issue:4061", + "issue:4194", + "issue:4303", + "issue:4305", + "issue:4354", + "issue:4376", + "issue:4492", + "issue:4885", + "issue:5412", + "issue:5558", + "issue:6418", + "issue:6482", + "issue:6598", + "issue:7049", + "issue:7142", + "issue:7348", + "issue:7773", + "issue:7985", + "issue:8182", + "issue:8626", + "issue:8787", + "issue:8828", + "issue:8839", + "issue:8939", + "issue:9237", + "issue:9448", + "issue:9525", + "issue:9732" + ], + "soft_pairs": [ + "issue:7049|issue:9448", + "issue:10925|issue:9237", + "issue:5558|issue:7348", + "issue:12613|issue:12699", + "issue:4305|issue:7985", + "issue:7049|issue:8626", + "issue:12082|issue:12565", + "issue:8828|issue:8839", + "issue:10150|issue:12599", + "issue:11642|issue:9525", + "issue:4303|issue:4376", + "issue:11024|issue:8182", + "issue:10202|issue:11464", + "issue:1823|issue:228", + "issue:3019|issue:3728", + "issue:8787|issue:9525", + "issue:11363|issue:9732", + "issue:4492|issue:8939", + "issue:1350|issue:2984", + "issue:6418|issue:7773", + "issue:6598|issue:7142", + "issue:4194|issue:4354", + "issue:10594|issue:11109", + "issue:4061|issue:4885", + "issue:5412|issue:6482" + ], + "prepared_review_unit_hash": "cc872a3ba161db0a707726691d41ccf1b05bb5f05d619896be02271f22d35b5f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34377, + "estimated_input_tokens": 8595, + "estimated_eval_tokens": 17446 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous. Only one soft pair looks like a genuine duplicate: the SDXL DreamBooth LoRA resume_from_checkpoint failures in the basic and advanced training scripts. The rest are related at a broad subsystem level but describe different bugs or features.", + "confidence": 0.77, + "canonical_issue_reason": "issue:6482 is the clearest representative of the likely duplicate training-resume bug because it names the exact failing script variant and the concrete FP16 unscale exception.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6482 is the best issue to anchor the cluster because it is the most specific and actionable of the likely duplicate pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7049", + "right": "issue:9448", + "accept": false, + "reason": "Both are IP-Adapter-related, but they report different failures in different pipelines/errors; not the same concrete bug." + }, + { + "left": "issue:10925", + "right": "issue:9237", + "accept": false, + "reason": "Both mention FLUX DreamBooth/LoRA, but one is a CLIPTextModel finetune issue and the other is a LoRA training issue; too broad." + }, + { + "left": "issue:5558", + "right": "issue:7348", + "accept": false, + "reason": "ControlNet inpaint artifacts vs scheduler weird results are different symptoms and likely different code paths." + }, + { + "left": "issue:12613", + "right": "issue:12699", + "accept": false, + "reason": "Wan 2.2 latent NaNs in i2v and video quality degrading after many prompts are not the same bug." + }, + { + "left": "issue:4305", + "right": "issue:7985", + "accept": false, + "reason": "Both involve torch.compile, but one is a graph break regression and the other is IP-Adapter incompatibility; distinct issues." + }, + { + "left": "issue:7049", + "right": "issue:8626", + "accept": false, + "reason": "One is a runtime error; the other is a guidance/docs request about multiple IP-Adapter images/masks." + }, + { + "left": "issue:12082", + "right": "issue:12565", + "accept": false, + "reason": "Both are Wan2.1 issues, but one is VAE memory after compile and the other is a general i2v running problem." + }, + { + "left": "issue:8828", + "right": "issue:8839", + "accept": false, + "reason": "SD3 training error vs SD3 LoRA weight error are different failures." + }, + { + "left": "issue:10150", + "right": "issue:12599", + "accept": false, + "reason": "Both mention slowness, but one is SDXL pipeline loading/inference generally and the other is moving to CUDA after loading; not clearly the same bug." + }, + { + "left": "issue:11642", + "right": "issue:9525", + "accept": false, + "reason": "Both say a scale parameter has no effect in Flux, but they concern different parameters and different pipelines." + }, + { + "left": "issue:4303", + "right": "issue:4376", + "accept": false, + "reason": "Both are SDXL/Flax load crashes, but one is loading SDXL in FlaxStableDiffusionPipeline and the other is load_textual_inversion crashing." + }, + { + "left": "issue:11024", + "right": "issue:8182", + "accept": false, + "reason": "Completely unrelated errors: SiglipImageProcessor .to() missing vs a Logger typo." + }, + { + "left": "issue:10202", + "right": "issue:11464", + "accept": false, + "reason": "Both involve FluxControlPipeline LoRA handling, but one is unloading correctly and the other is inability to load a specific LoRA; not enough to merge as one bug." + }, + { + "left": "issue:1823", + "right": "issue:228", + "accept": false, + "reason": "Swapped comments in depth2img and a broken model link are unrelated." + }, + { + "left": "issue:3019", + "right": "issue:3728", + "accept": false, + "reason": "Different model/pipeline feature requests with no common failure mode." + }, + { + "left": "issue:8787", + "right": "issue:9525", + "accept": false, + "reason": "Generic loading slowdown and a Flux LoRA scale bug are unrelated." + }, + { + "left": "issue:11363", + "right": "issue:9732", + "accept": false, + "reason": "ControlNet example CUDA OOM and FLUX DreamBooth training OOM are different workloads and likely different fixes." + }, + { + "left": "issue:4492", + "right": "issue:8939", + "accept": false, + "reason": "A Python boolean typo error and an SD3 tokenizer loading OSError are unrelated." + }, + { + "left": "issue:1350", + "right": "issue:2984", + "accept": false, + "reason": "DiffusionDet object detection and latent video diffusion are different projects/features." + }, + { + "left": "issue:6418", + "right": "issue:7773", + "accept": false, + "reason": "Both are Stable Video Diffusion issues, but one is identical first-frame output after finetuning and the other is a batch-size>1 generation bug." + }, + { + "left": "issue:6598", + "right": "issue:7142", + "accept": false, + "reason": "Both compare outputs, but one is generic img2img result mismatch and the other is SDXL-lightning vs webui discrepancy." + }, + { + "left": "issue:4194", + "right": "issue:4354", + "accept": false, + "reason": "Both are single-file loading problems, but one is VAE not working and the other is wrong pipeline type returned." + }, + { + "left": "issue:10594", + "right": "issue:11109", + "accept": false, + "reason": "Both mention FLUX and LoRA/control, but one is multiple controls with LoRA and the other is a depth LoRA transformer-layer mismatch." + }, + { + "left": "issue:4061", + "right": "issue:4885", + "accept": false, + "reason": "Both are SDXL LoRA-loading problems, but they target different pipelines and likely different code paths." + }, + { + "left": "issue:5412", + "right": "issue:6482", + "accept": true, + "reason": "Same SDXL DreamBooth LoRA checkpoint-resume failure across two script variants; likely one underlying bug in resume handling." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge calls stay conservative overall. The lone accepted pair, issue:5412 and issue:6482, is sufficiently supported by their shared resume_from_checkpoint SDXL DreamBooth LoRA training-failure context. The rest are reasonably rejected as broad subsystem similarities rather than the same bug." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 81, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1005", + "issue:10596", + "issue:10673", + "issue:10674", + "issue:10795", + "issue:10833", + "issue:10914", + "issue:11023", + "issue:11143", + "issue:11258", + "issue:13043", + "issue:13298", + "issue:13484", + "issue:13487", + "issue:1367", + "issue:2560", + "issue:3382", + "issue:3428", + "issue:3717", + "issue:3774", + "issue:4141", + "issue:4354", + "issue:4418", + "issue:462", + "issue:4911", + "issue:5623", + "issue:5857", + "issue:5877", + "issue:5889", + "issue:5927", + "issue:6001", + "issue:6008", + "issue:6243", + "issue:6252", + "issue:6446", + "issue:6574", + "issue:7229", + "issue:7363", + "issue:7405", + "issue:7629", + "issue:7728", + "issue:7757", + "issue:8435", + "issue:8518", + "issue:8787", + "issue:941", + "issue:9664", + "issue:9971" + ], + "soft_pairs": [ + "issue:4141|issue:4418", + "issue:462|issue:941", + "issue:11143|issue:11258", + "issue:4354|issue:6001", + "issue:4911|issue:8435", + "issue:2560|issue:3428", + "issue:1005|issue:1367", + "issue:3382|issue:7757", + "issue:5623|issue:7363", + "issue:5927|issue:6243", + "issue:6574|issue:7629", + "issue:13043|issue:13298", + "issue:3717|issue:3774", + "issue:7229|issue:7728", + "issue:6252|issue:6446", + "issue:5857|issue:6008", + "issue:5889|issue:7405", + "issue:13484|issue:13487", + "issue:10674|issue:9971", + "issue:10914|issue:11023", + "issue:10596|issue:10673", + "issue:10833|issue:8787", + "issue:10795|issue:8518", + "issue:5877|issue:9664" + ], + "prepared_review_unit_hash": "33eeeb8a033e802ad9d67b7ea4d949e77e5a6869c98a5bb350526f455931a931", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33921, + "estimated_input_tokens": 8481, + "estimated_eval_tokens": 17218 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issues were grouped by loose title similarity. Only the `from_single_file` SDXL pipeline-type mismatch pair looks like a true duplicate; the rest are similar only at a broad subsystem level.", + "confidence": 0.71, + "canonical_issue_reason": "issue:6001 is the clearest concrete report of the shared `from_single_file` SDXL loader bug: the refiner is returned as `Img2ImgPipeline` instead of the expected pipeline type.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6001 is the best canonical anchor because it states the wrong-return-type failure most specifically and is directly actionable.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4141", + "right": "issue:4418", + "accept": false, + "reason": "Both are inpaint/control-related, but one is about second ControlNet handling and the other about `denoising_start`; different bugs." + }, + { + "left": "issue:462", + "right": "issue:941", + "accept": false, + "reason": "Both involve MPS img2img behavior, but the symptoms and triggers differ (`0.3.0` regression vs `num_images_per_prompt` noise)." + }, + { + "left": "issue:11143", + "right": "issue:11258", + "accept": false, + "reason": "Different LTX-Video topics: one is quality feedback, the other is a single-file loading request." + }, + { + "left": "issue:4354", + "right": "issue:6001", + "accept": true, + "reason": "Both report `from_single_file` returning the wrong SDXL pipeline class for inpaint/refiner paths, which looks like the same underlying loader bug." + }, + { + "left": "issue:4911", + "right": "issue:8435", + "accept": false, + "reason": "Both mention `from_single_file`, but one is about custom text encoders and the other is a generic runtime error; not the same failure." + }, + { + "left": "issue:2560", + "right": "issue:3428", + "accept": false, + "reason": "Both concern StableDiffusionImg2ImgPipeline behavior, but they describe different output anomalies and likely different code paths." + }, + { + "left": "issue:1005", + "right": "issue:1367", + "accept": false, + "reason": "Scheduler timestep limits and custom `trained_betas` config saving are separate scheduler issues." + }, + { + "left": "issue:3382", + "right": "issue:7757", + "accept": false, + "reason": "Both are training OOMs, but they target different scripts and settings; too broad to be the same bug." + }, + { + "left": "issue:5623", + "right": "issue:7363", + "accept": false, + "reason": "One is a safety-checker enhancement request, the other a request for SDXL safety checker support; not the same change." + }, + { + "left": "issue:5927", + "right": "issue:6243", + "accept": false, + "reason": "Both mention IP-Adapter, but the configurations and failures are different." + }, + { + "left": "issue:6574", + "right": "issue:7629", + "accept": false, + "reason": "Both mention SVD, but one is a warning/shape issue and the other is a version-specific OOM; unrelated." + }, + { + "left": "issue:13043", + "right": "issue:13298", + "accept": false, + "reason": "Different CUDA/device failures: a specific Blackwell kernel-image issue vs a general generator-device breakage." + }, + { + "left": "issue:3717", + "right": "issue:3774", + "accept": false, + "reason": "Both are memory-related training issues, but one is multi-GPU OOM and the other is ZeRO-3 not saving memory." + }, + { + "left": "issue:7229", + "right": "issue:7728", + "accept": false, + "reason": "Checkpoint saving problems and an Accelerate error in an advanced DreamBooth script are not the same bug." + }, + { + "left": "issue:6252", + "right": "issue:6446", + "accept": false, + "reason": "Both are loading-related, but one is LoRA weights from a checkpoint and the other is a minimal-loading model init failure." + }, + { + "left": "issue:5857", + "right": "issue:6008", + "accept": false, + "reason": "Both involve LCM LoRA, but the affected pipelines and failure modes differ." + }, + { + "left": "issue:5889", + "right": "issue:7405", + "accept": false, + "reason": "Completely different models/projects: Stable Video Diffusion vs DynamiCrafter." + }, + { + "left": "issue:13484", + "right": "issue:13487", + "accept": false, + "reason": "Both are LoRA-loading issues in modular/Flux contexts, but the titles point to different components and failure scenarios." + }, + { + "left": "issue:10674", + "right": "issue:9971", + "accept": false, + "reason": "Both are FLUX-related, but one is a GGUF pipeline issue and the other is a scheduler compatibility issue." + }, + { + "left": "issue:10914", + "right": "issue:11023", + "accept": false, + "reason": "A device offloading problem and a blurry-image custom-pipeline report are unrelated." + }, + { + "left": "issue:10596", + "right": "issue:10673", + "accept": false, + "reason": "Both are invalid-shape errors, but they occur in different pipelines and likely stem from different shape assumptions." + }, + { + "left": "issue:10833", + "right": "issue:8787", + "accept": false, + "reason": "A bitsandbytes matrix-shape failure and a model-loading slowdown are not the same issue." + }, + { + "left": "issue:10795", + "right": "issue:8518", + "accept": false, + "reason": "Both are torch.compile failures, but one is GGUF/from_single_file related while the other is an SD3-specific dynamo unsupported-op error." + }, + { + "left": "issue:5877", + "right": "issue:9664", + "accept": false, + "reason": "Both are subprocess training failures, but they involve different scripts and no clear shared root cause." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet metadata, and the soft-edge calls are conservative. The one accepted pair (4354/6001) is well supported by the titles as the same `from_single_file` SDXL pipeline-type mismatch bug, while the rest are reasonably rejected as broader subsystem similarities rather than clear duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 82, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10180", + "issue:10321", + "issue:10334", + "issue:10350", + "issue:1061", + "issue:10925", + "issue:11010", + "issue:11143", + "issue:11298", + "issue:11342", + "issue:11907", + "issue:12501", + "issue:1367", + "issue:1734", + "issue:2156", + "issue:228", + "issue:2299", + "issue:2674", + "issue:3026", + "issue:3250", + "issue:3357", + "issue:3717", + "issue:4134", + "issue:4236", + "issue:4480", + "issue:4483", + "issue:4993", + "issue:5075", + "issue:5604", + "issue:5753", + "issue:5917", + "issue:5931", + "issue:6096", + "issue:6124", + "issue:6709", + "issue:6818", + "issue:6879", + "issue:7033", + "issue:7377", + "issue:7513", + "issue:7889", + "issue:8645", + "issue:8649", + "issue:9227", + "issue:9643", + "issue:9667", + "issue:993" + ], + "soft_pairs": [ + "issue:4236|issue:5075", + "issue:1061|issue:228", + "issue:3026|issue:3717", + "issue:4483|issue:5931", + "issue:6818|issue:7033", + "issue:7377|issue:8649", + "issue:11907|issue:9643", + "issue:4134|issue:4480", + "issue:4993|issue:5917", + "issue:6124|issue:7513", + "issue:10334|issue:11342", + "issue:11010|issue:9667", + "issue:10350|issue:12501", + "issue:10321|issue:11143", + "issue:10180|issue:11298", + "issue:3357|issue:6096", + "issue:2674|issue:5753", + "issue:1734|issue:2156", + "issue:2299|issue:993", + "issue:6709|issue:6879", + "issue:2674|issue:5604", + "issue:7889|issue:8645", + "issue:1367|issue:3250", + "issue:10925|issue:9227" + ], + "prepared_review_unit_hash": "1e8ef735dcd1e3bad03d97632103d9826e19280b6ce9978fb885f9f4da1f5ed8", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33409, + "estimated_input_tokens": 8353, + "estimated_eval_tokens": 16962 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issue set; the only strong duplicate-like pair is the multiple-LoRA PEFT error around Flux. Everything else is either a different model/pipeline, a different failure mode, or just a broad subsystem overlap.", + "confidence": 0.39, + "canonical_issue_reason": "issue:11298 is the best canonical issue because it is the clearest, still-open report for the concrete multiple-LoRA/PEFT key-error path and is the most likely umbrella for the closely related Flux LoRA-loading report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11298 is the strongest global issue candidate: specific, actionable, and aligned with the only plausible duplicate pair in the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4236", + "right": "issue:5075", + "accept": false, + "reason": "Both involve SDXL refiner shape issues, but one is text-encoder registration and the other is prompt-embedding matmul mismatch; not the same bug." + }, + { + "left": "issue:1061", + "right": "issue:228", + "accept": false, + "reason": "One is a docs/example update request; the other is a broken release link. Different artifact types and different problems." + }, + { + "left": "issue:3026", + "right": "issue:3717", + "accept": false, + "reason": "Different training/runtime failures: one is an instruct-pix2pix accelerate launch error, the other is multi-GPU OOM." + }, + { + "left": "issue:4483", + "right": "issue:5931", + "accept": false, + "reason": "Both mention determinism, but they target different pipelines and likely different causes; not enough to treat as one bug." + }, + { + "left": "issue:6818", + "right": "issue:7033", + "accept": false, + "reason": "Unloading an IP-Adapter and attention-masking warning logs are unrelated failure modes." + }, + { + "left": "issue:7377", + "right": "issue:8649", + "accept": false, + "reason": "Both mention num_images_per_prompt, but they are different pipelines/model families with different reported breakages." + }, + { + "left": "issue:11907", + "right": "issue:9643", + "accept": false, + "reason": "Both are Flux-related, but one is scheduler incompatibility and the other is multiple-ControlNet support; different concrete issues." + }, + { + "left": "issue:4134", + "right": "issue:4480", + "accept": false, + "reason": "Both are MPS/Mac issues, but one is a ControlNet tensor-shape mismatch and the other is a kernel crash with fp16 generation." + }, + { + "left": "issue:4993", + "right": "issue:5917", + "accept": false, + "reason": "Import-structure exposure vs checkpoint conversion regression; different code paths and symptoms." + }, + { + "left": "issue:6124", + "right": "issue:7513", + "accept": false, + "reason": "Both concern DreamBooth/LoRA training, but one is an FP16 gradient unscale error and the other is a generic GPU memory issue." + }, + { + "left": "issue:10334", + "right": "issue:11342", + "accept": false, + "reason": "Same model family and platform, but the reported failures differ too much (grey images/NaNs vs pipeline failing) to confidently merge." + }, + { + "left": "issue:11010", + "right": "issue:9667", + "accept": false, + "reason": "Both involve Flux-style architecture/model loading, but one is feature/support discussion and the other is an fp8 loading error." + }, + { + "left": "issue:10350", + "right": "issue:12501", + "accept": false, + "reason": "Both mention the VAE decoder, but one is memory consumption and the other is a torch graph-break issue; different bugs." + }, + { + "left": "issue:10321", + "right": "issue:11143", + "accept": false, + "reason": "Both are about LTX-Video, but one is a missing model file error and the other is video quality complaints." + }, + { + "left": "issue:10180", + "right": "issue:11298", + "accept": true, + "reason": "Both describe multiple-LoRA handling failing in the Flux/PEFT path, and the titles point to the same underlying key-error/load-management bug." + }, + { + "left": "issue:3357", + "right": "issue:6096", + "accept": false, + "reason": "Both are torch.compile issues, but they are different compiler failures with different reported exceptions and code paths." + }, + { + "left": "issue:2674", + "right": "issue:5753", + "accept": false, + "reason": "LoRA-related, but one is a missing dependency/safetensors packaging problem while the other is a pipeline/load_lora_weights incompatibility." + }, + { + "left": "issue:1734", + "right": "issue:2156", + "accept": false, + "reason": "Different training performance problems: multi-GPU slowness versus running on CPU instead of GPU." + }, + { + "left": "issue:2299", + "right": "issue:993", + "accept": false, + "reason": "Different issues: checkpoint compatibility/support versus a datatype mismatch in fp16." + }, + { + "left": "issue:6709", + "right": "issue:6879", + "accept": false, + "reason": "Both are Custom Diffusion, but one is concept-list behavior and the other is a dtype mismatch; not the same bug." + }, + { + "left": "issue:2674", + "right": "issue:5604", + "accept": false, + "reason": "Both concern LoRA handling, but one is safetensors packaging and the other is inability to unfuse multiple LoRAs; distinct problems." + }, + { + "left": "issue:7889", + "right": "issue:8645", + "accept": false, + "reason": "Same file area, but one is a general single_file_utils bug report and the other is a missing helper function request." + }, + { + "left": "issue:1367", + "right": "issue:3250", + "accept": false, + "reason": "Completely different scheduler config save bug versus pipeline execution failure." + }, + { + "left": "issue:10925", + "right": "issue:9227", + "accept": false, + "reason": "Both are FLUX DreamBooth training reports, but they point to different scripts/models and different failure modes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The only accepted soft edge, issue:10180 \u2194 issue:11298, is supported by the titles and clearly centers on the same Flux/multiple-LoRA loading failure mode. The rest of the pairwise rejections are appropriately cautious." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 83, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10186", + "issue:10639", + "issue:10833", + "issue:10893", + "issue:10937", + "issue:11014", + "issue:11023", + "issue:11033", + "issue:11225", + "issue:11408", + "issue:11569", + "issue:11907", + "issue:12082", + "issue:12120", + "issue:12358", + "issue:1367", + "issue:1565", + "issue:1617", + "issue:1823", + "issue:2089", + "issue:2371", + "issue:2444", + "issue:259", + "issue:3565", + "issue:3694", + "issue:4206", + "issue:4566", + "issue:4664", + "issue:4731", + "issue:5313", + "issue:5432", + "issue:5516", + "issue:5846", + "issue:6252", + "issue:6368", + "issue:6561", + "issue:6809", + "issue:6823", + "issue:6839", + "issue:7159", + "issue:761", + "issue:7982", + "issue:8469", + "issue:861", + "issue:8909", + "issue:9599", + "issue:9643", + "issue:9953" + ], + "soft_pairs": [ + "issue:1617|issue:2444", + "issue:10186|issue:10833", + "issue:4731|issue:7159", + "issue:11033|issue:11907", + "issue:11408|issue:9643", + "issue:3565|issue:3694", + "issue:1823|issue:259", + "issue:2371|issue:761", + "issue:5432|issue:5846", + "issue:7982|issue:9953", + "issue:10893|issue:11569", + "issue:6823|issue:6839", + "issue:11023|issue:8909", + "issue:1565|issue:2089", + "issue:5313|issue:6809", + "issue:6561|issue:8469", + "issue:10639|issue:12358", + "issue:11014|issue:11225", + "issue:10937|issue:12082", + "issue:4206|issue:4566", + "issue:12120|issue:9599", + "issue:4664|issue:5516", + "issue:1367|issue:861", + "issue:6252|issue:6368" + ], + "prepared_review_unit_hash": "68a061821407b0018af6439160d6ae7537a73e0930e0d90647ed14109368629b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34014, + "estimated_input_tokens": 8504, + "estimated_eval_tokens": 17264 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated issues and feature requests across different pipelines, training scripts, and runtime failures. The soft pairs share broad keywords but not the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue is a strong canonical representative because the cluster is heterogeneous; the soft pairs do not point to one shared defect or feature.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "No best issue: any one choice would overfit to a narrow subtopic and ignore the rest of the mixed set.", + "best_pr_reason": "No pull requests are present, so there is no PR candidate to select.", + "soft_edge_verdicts": [ + { + "left": "issue:1617", + "right": "issue:2444", + "accept": false, + "reason": "Different scripts and failures: missing CLI args vs inpaint training subprocess error." + }, + { + "left": "issue:10186", + "right": "issue:10833", + "accept": false, + "reason": "Same pipeline name, but one is an import error and the other is a quantization matrix-shape bug." + }, + { + "left": "issue:4731", + "right": "issue:7159", + "accept": false, + "reason": "Both involve torch.compile, but they hit different code paths and symptoms." + }, + { + "left": "issue:11033", + "right": "issue:11907", + "accept": false, + "reason": "Unet loading from_single_file vs FLUX scheduler compatibility are unrelated bugs." + }, + { + "left": "issue:11408", + "right": "issue:9643", + "accept": false, + "reason": "Compiled-model adapter setting and multiple-ControlNet support are different problems." + }, + { + "left": "issue:3565", + "right": "issue:3694", + "accept": false, + "reason": "OOM during training vs missing model/config files while running an example." + }, + { + "left": "issue:1823", + "right": "issue:259", + "accept": false, + "reason": "A comment swap in depth2img is unrelated to a progress bar fix." + }, + { + "left": "issue:2371", + "right": "issue:761", + "accept": false, + "reason": "DreamBooth output missing objects is not the same as 8-bit Adam optimizer failure." + }, + { + "left": "issue:5432", + "right": "issue:5846", + "accept": false, + "reason": "Blank validation images and resume-from-checkpoint errors are different LoRA training issues." + }, + { + "left": "issue:7982", + "right": "issue:9953", + "accept": false, + "reason": "from_pipe with MultiControlNetModel and moving a quantized pipeline to CUDA are distinct failures." + }, + { + "left": "issue:10893", + "right": "issue:11569", + "accept": false, + "reason": "Different feature requests for different models and pipelines." + }, + { + "left": "issue:6823", + "right": "issue:6839", + "accept": false, + "reason": "General controlnet usage problem vs StableVideoDiffusion from_single_file support." + }, + { + "left": "issue:11023", + "right": "issue:8909", + "accept": false, + "reason": "Blurred inference output and saving text_encoder_ti in training are unrelated." + }, + { + "left": "issue:1565", + "right": "issue:2089", + "accept": false, + "reason": "Both mention OOM, but one is GPU selection behavior and the other is DreamBooth optimizer memory use." + }, + { + "left": "issue:5313", + "right": "issue:6809", + "accept": false, + "reason": "Non-revertible LoRA fuse and PEFT load regressions are different adapter-management bugs." + }, + { + "left": "issue:6561", + "right": "issue:8469", + "accept": false, + "reason": "Missing IP-Adapter config file and corrupted faces are not the same underlying issue." + }, + { + "left": "issue:10639", + "right": "issue:12358", + "accept": false, + "reason": "Separate feature requests for different Lumina models, not one duplicated request." + }, + { + "left": "issue:11014", + "right": "issue:11225", + "accept": false, + "reason": "General CUDA OOM on Colab and SD3 controlnet validation OOM are different scenarios." + }, + { + "left": "issue:10937", + "right": "issue:12082", + "accept": false, + "reason": "torch.compile error on vae.encode is different from increased VAE memory after compile." + }, + { + "left": "issue:4206", + "right": "issue:4566", + "accept": false, + "reason": "Controlnet training on main branch and LoRA checkpoint resume are unrelated." + }, + { + "left": "issue:12120", + "right": "issue:9599", + "accept": false, + "reason": "Training guidance for distilled FLUX and a request for a LoRA-only FLUX example are not the same bug." + }, + { + "left": "issue:4664", + "right": "issue:5516", + "accept": false, + "reason": "SDXL LoRA loading mismatch and LoRA with t2i adapters are different integration bugs." + }, + { + "left": "issue:1367", + "right": "issue:861", + "accept": false, + "reason": "Scheduler save_config with trained_betas is unrelated to ckpt-to-diffusers conversion." + }, + { + "left": "issue:6252", + "right": "issue:6368", + "accept": false, + "reason": "Checkpoint not loading and unsupported LoRA layer alpha are different checkpoint-format problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 84, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10016", + "issue:10315", + "issue:10428", + "issue:10863", + "issue:11225", + "issue:11584", + "issue:11791", + "issue:12175", + "issue:12216", + "issue:12359", + "issue:1305", + "issue:1500", + "issue:1611", + "issue:1889", + "issue:2676", + "issue:2772", + "issue:2878", + "issue:2883", + "issue:3061", + "issue:3706", + "issue:3728", + "issue:460", + "issue:4615", + "issue:4724", + "issue:4925", + "issue:4935", + "issue:5432", + "issue:5466", + "issue:5575", + "issue:5576", + "issue:5873", + "issue:6224", + "issue:6520", + "issue:6536", + "issue:6549", + "issue:6586", + "issue:6790", + "issue:6794", + "issue:6891", + "issue:7180", + "issue:7466", + "issue:7766", + "issue:7871", + "issue:9112", + "issue:9343", + "issue:938", + "issue:9497", + "issue:9576" + ], + "soft_pairs": [ + "issue:6586|issue:6891", + "issue:10016|issue:5466", + "issue:2772|issue:4935", + "issue:7180|issue:7466", + "issue:4724|issue:7871", + "issue:5575|issue:6790", + "issue:11225|issue:4925", + "issue:1889|issue:2676", + "issue:12175|issue:12216", + "issue:11584|issue:11791", + "issue:2878|issue:5873", + "issue:12359|issue:9576", + "issue:3061|issue:460", + "issue:10315|issue:9497", + "issue:4615|issue:6794", + "issue:5432|issue:6520", + "issue:1305|issue:938", + "issue:2883|issue:3728", + "issue:6224|issue:6536", + "issue:10428|issue:9112", + "issue:3706|issue:5576", + "issue:6549|issue:7766", + "issue:1500|issue:1611", + "issue:10863|issue:9343" + ], + "prepared_review_unit_hash": "6bebef1785c0d0cb6a87882f2200b0c560f054d0313fa2678b2fd5dcfacc06fa", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34299, + "estimated_input_tokens": 8575, + "estimated_eval_tokens": 17406 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issues with a few loose topical similarities. None of the pairs are strong enough to call duplicates; the SDXL blank-validation pair is related, but not confirmed as the same bug.", + "confidence": 0.77, + "canonical_issue_reason": "Issue 6520 is the best representative of the SDXL blank-validation symptom because it explicitly names `log_validation` and the broader training-adapter context, but the evidence is still too weak to treat it as a confirmed duplicate of 5432.", + "canonical_pr_reason": null, + "best_issue_reason": "6520 is the most diagnostic issue for the shared SDXL validation-output theme, even though the pair does not clear duplicate threshold.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6586", + "right": "issue:6891", + "accept": false, + "reason": "Both are trainer-related tracker issues, but one is about micro-conditioning and the other about model-card saving utilities; different changes." + }, + { + "left": "issue:10016", + "right": "issue:5466", + "accept": false, + "reason": "Both involve load-time shape mismatches, but they affect different models and different parameters (`pos_embed` vs LoRA `down.weight`)." + }, + { + "left": "issue:2772", + "right": "issue:4935", + "accept": false, + "reason": "One is about local dataset handling in `train_controlnet.py`; the other is an SDXL distributed-training OOM. Same area, different bug." + }, + { + "left": "issue:7180", + "right": "issue:7466", + "accept": false, + "reason": "Both touch SDXL ControlNet, but one is CPU offload failure and the other is `torch.compile` with multi-ControlNet inpaint; not the same defect." + }, + { + "left": "issue:4724", + "right": "issue:7871", + "accept": false, + "reason": "Different problems: SDXL img2img checkpoint behavior versus blockwise scales with a specific LoRA file." + }, + { + "left": "issue:5575", + "right": "issue:6790", + "accept": false, + "reason": "Unrelated: a LoRA hidden-size question versus an LCM training state-dict mismatch." + }, + { + "left": "issue:11225", + "right": "issue:4925", + "accept": false, + "reason": "Both are OOM reports, but they involve different trainers and different validation/training contexts." + }, + { + "left": "issue:1889", + "right": "issue:2676", + "accept": false, + "reason": "Scheduler-related, but one is a feature request and the other is a naming question; not duplicates." + }, + { + "left": "issue:12175", + "right": "issue:12216", + "accept": false, + "reason": "Both concern Qwen image quality, but one is generic Qwen-Image quality and the other is Qwen-Image-Edit versus ComfyUI; too different." + }, + { + "left": "issue:11584", + "right": "issue:11791", + "accept": false, + "reason": "Both mention group offloading, but one is a `torch.compile` incompatibility and the other is LoRA weights failing to load after offloading." + }, + { + "left": "issue:2878", + "right": "issue:5873", + "accept": false, + "reason": "Both are ONNX conversion issues, but one is broad PyTorch 2.0 converter discussion and the other is a specific unsupported operator export." + }, + { + "left": "issue:12359", + "right": "issue:9576", + "accept": false, + "reason": "Completely different topics: Chroma docs for `guidance_scale` versus `use_safetensors` not being passed to submodules." + }, + { + "left": "issue:3061", + "right": "issue:460", + "accept": false, + "reason": "Both are loading/reproducibility concerns, but the underlying failures are unrelated." + }, + { + "left": "issue:10315", + "right": "issue:9497", + "accept": false, + "reason": "Different training errors for different models: CogVideo training versus Dreambooth Flux matmul shape failure." + }, + { + "left": "issue:4615", + "right": "issue:6794", + "accept": false, + "reason": "Both are import/runtime errors, but they involve unrelated symbols and modules." + }, + { + "left": "issue:5432", + "right": "issue:6520", + "accept": false, + "reason": "Both are related SDXL blank-validation reports, but the packet does not establish a shared `log_validation` code path or a true duplicate relationship." + }, + { + "left": "issue:1305", + "right": "issue:938", + "accept": false, + "reason": "Different papers/features entirely: Direct Inversion versus Human Motion Diffusion Model." + }, + { + "left": "issue:2883", + "right": "issue:3728", + "accept": false, + "reason": "Different pipeline requests: image-to-video versus shifted diffusion text-to-image." + }, + { + "left": "issue:6224", + "right": "issue:6536", + "accept": false, + "reason": "Different training/runtime failures: TorchDynamo issue in text-to-image fine-tuning versus UNet3DConditionModel OOM." + }, + { + "left": "issue:10428", + "right": "issue:9112", + "accept": false, + "reason": "Both are inference tensor-shape errors, but they affect different models and likely different assumptions." + }, + { + "left": "issue:3706", + "right": "issue:5576", + "accept": false, + "reason": "Different feature requests/models: Wuerstchen pipeline versus FreeNoise scheduler request." + }, + { + "left": "issue:6549", + "right": "issue:7766", + "accept": false, + "reason": "Both mention IP-Adapter, but one is a LoRA warning log and the other is a usage question." + }, + { + "left": "issue:1500", + "right": "issue:1611", + "accept": false, + "reason": "Both are old pipeline regressions, but the affected pipelines and breakage descriptions differ." + }, + { + "left": "issue:10863", + "right": "issue:9343", + "accept": false, + "reason": "Both are load/safetensors-related, but one is a local safetensors file error and the other is a FLUX loading issue with different flags." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is mostly cautious, but the one accepted pair is too strong for the packet: 5432 and 6520 both involve SDXL validation behavior, yet the packet only gives titles and does not explicitly establish a shared `log_validation` code path or a true duplicate relationship. The packet\u2019s deterministic signal also marks that pair as false. A safer phrasing would be that they are related reports, not an accepted duplicate pair." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 85, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10671", + "issue:11023", + "issue:11674", + "issue:11779", + "issue:12175", + "issue:1556", + "issue:1614", + "issue:3198", + "issue:3279", + "issue:3284", + "issue:3357", + "issue:4305", + "issue:4375", + "issue:4724", + "issue:4935", + "issue:4936", + "issue:4954", + "issue:4985", + "issue:5227", + "issue:5432", + "issue:5563", + "issue:5889", + "issue:5931", + "issue:5934", + "issue:6095", + "issue:6124", + "issue:6232", + "issue:6415", + "issue:6527", + "issue:6574", + "issue:7041", + "issue:7159", + "issue:7185", + "issue:7223", + "issue:7347", + "issue:7538", + "issue:8507", + "issue:8511", + "issue:8552", + "issue:8649", + "issue:8732", + "issue:8759", + "issue:8827", + "issue:8828", + "issue:8839", + "issue:9515", + "issue:9904" + ], + "soft_pairs": [ + "issue:4375|issue:5934", + "issue:12175|issue:9904", + "issue:3357|issue:7159", + "issue:3198|issue:4954", + "issue:4724|issue:5931", + "issue:3279|issue:3284", + "issue:7041|issue:8552", + "issue:8649|issue:8828", + "issue:10671|issue:11023", + "issue:4935|issue:5432", + "issue:6232|issue:6527", + "issue:8507|issue:8759", + "issue:6095|issue:6574", + "issue:8511|issue:8839", + "issue:4375|issue:5227", + "issue:6124|issue:8732", + "issue:6415|issue:8827", + "issue:5563|issue:7347", + "issue:7185|issue:9515", + "issue:4985|issue:7538", + "issue:11674|issue:11779", + "issue:5889|issue:7223", + "issue:4305|issue:4936", + "issue:1556|issue:1614" + ], + "prepared_review_unit_hash": "581c21b32515a2e90025b1dc6213f5d02a291556aaf555879b60691cd9860aa9", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33055, + "estimated_input_tokens": 8264, + "estimated_eval_tokens": 16784 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: several pairs are only loosely related by subsystem (torch.compile, SDXL/SD3 training, schedulers, black-image symptoms) but not the same underlying bug. I would reject all soft edges and treat the issue set as separate reports, with issue 5934 as the broadest representative of the compile-error subset.", + "confidence": 0.89, + "canonical_issue_reason": "issue:5934 is the broadest and most generic report in the set (\u201cError when using torch.compile\u201d), so it best represents the compile-related subset even though it is not a duplicate of the other issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:5934 is the most central, high-level issue title in the cluster and the best anchor for triage because it cleanly captures the recurring torch.compile failures without overfitting to one narrow stack trace.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4375", + "right": "issue:5934", + "accept": false, + "reason": "Both involve torch.compile, but one is a crash during inference and the other is a generic compile error; different concrete failures." + }, + { + "left": "issue:12175", + "right": "issue:9904", + "accept": false, + "reason": "Both mention image quality/shape issues, but they affect different models and different code paths." + }, + { + "left": "issue:3357", + "right": "issue:7159", + "accept": false, + "reason": "Both are torch.compile problems, but the reported errors differ (hasattr/TensorVariable vs fused LoRA compile failure)." + }, + { + "left": "issue:3198", + "right": "issue:4954", + "accept": false, + "reason": "Both are textual inversion loading failures, but one is safetensors loading and the other is an unpickling error." + }, + { + "left": "issue:4724", + "right": "issue:5931", + "accept": false, + "reason": "One is an SDXL img2img correctness issue; the other is nondeterministic generation. Same area, not the same bug." + }, + { + "left": "issue:3279", + "right": "issue:3284", + "accept": false, + "reason": "Both touch dreambooth LoRA training, but one is checkpoint resume and the other is accelerate launch/runtime setup." + }, + { + "left": "issue:7041", + "right": "issue:8552", + "accept": false, + "reason": "Both are crashes, but the pipelines and triggering conditions are different." + }, + { + "left": "issue:8649", + "right": "issue:8828", + "accept": false, + "reason": "Both are SD3 training-related, but one is a parameter-handling regression and the other is a resource/training error." + }, + { + "left": "issue:10671", + "right": "issue:11023", + "accept": false, + "reason": "Different symptoms and likely different causes: dtype/device determinism vs blurry custom-pipeline output." + }, + { + "left": "issue:4935", + "right": "issue:5432", + "accept": false, + "reason": "Both concern SDXL LoRA training, but one is about a helper class usage question and the other about blank validation images." + }, + { + "left": "issue:6232", + "right": "issue:6527", + "accept": false, + "reason": "Same general scheduler area, but different schedulers and likely different add_noise bugs." + }, + { + "left": "issue:8507", + "right": "issue:8759", + "accept": false, + "reason": "One is SD3 train_dreambooth text-encoder failure; the other is an image processor warning/cast issue." + }, + { + "left": "issue:6095", + "right": "issue:6574", + "accept": false, + "reason": "Both are SVD pipeline issues, but one is an IndexError in scheduling and the other is a dtype/range warning." + }, + { + "left": "issue:8511", + "right": "issue:8839", + "accept": false, + "reason": "Both are SD3 training issues, but the concrete problems differ: text encoders not unloading vs LoRA weight error." + }, + { + "left": "issue:4375", + "right": "issue:5227", + "accept": false, + "reason": "Both mention torch.compile, but one is a generic inference crash and the other is a specific LoRA-loading compile incompatibility." + }, + { + "left": "issue:6124", + "right": "issue:8732", + "accept": false, + "reason": "Both are training-related, but one is FP16 unscale failure and the other is checkpoint-saving behavior." + }, + { + "left": "issue:6415", + "right": "issue:8827", + "accept": false, + "reason": "One is a feature-support request for FaceID plus; the other is an actual runtime error with ip-adapter-faceid_sdxl." + }, + { + "left": "issue:5563", + "right": "issue:7347", + "accept": false, + "reason": "Both are training errors, but the titles are too broad and do not indicate the same concrete failure." + }, + { + "left": "issue:7185", + "right": "issue:9515", + "accept": false, + "reason": "One is a pytorch version support question; the other is a missing RMSNorm implementation detail in SD3 attention." + }, + { + "left": "issue:4985", + "right": "issue:7538", + "accept": false, + "reason": "Both are training issues, but one is an OOM and the other is a custom dataset/style training problem." + }, + { + "left": "issue:11674", + "right": "issue:11779", + "accept": false, + "reason": "Feature request vs bug report; they concern different changes in Wan VACE." + }, + { + "left": "issue:5889", + "right": "issue:7223", + "accept": false, + "reason": "Different model requests: stable video diffusion discussion vs Latte addition request." + }, + { + "left": "issue:4305", + "right": "issue:4936", + "accept": false, + "reason": "Both are regressions around 0.19.0, but one is a torch.compile graph break and the other is DreamBooth LoRA resume failure." + }, + { + "left": "issue:1556", + "right": "issue:1614", + "accept": false, + "reason": "Both produce black images, but they appear to stem from different runtime conditions and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 86, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10421", + "issue:10569", + "issue:11088", + "issue:11182", + "issue:11222", + "issue:11223", + "issue:12575", + "issue:13105", + "issue:13176", + "issue:1704", + "issue:1823", + "issue:2015", + "issue:2364", + "issue:3250", + "issue:3422", + "issue:3674", + "issue:3700", + "issue:3816", + "issue:3826", + "issue:4000", + "issue:4213", + "issue:4356", + "issue:4488", + "issue:4566", + "issue:4708", + "issue:4985", + "issue:5466", + "issue:5540", + "issue:5565", + "issue:5892", + "issue:6028", + "issue:6112", + "issue:6252", + "issue:6749", + "issue:7003", + "issue:7159", + "issue:7306", + "issue:7363", + "issue:7392", + "issue:776", + "issue:7782", + "issue:7908", + "issue:8518", + "issue:8885", + "issue:929", + "issue:9525", + "issue:9889" + ], + "soft_pairs": [ + "issue:7392|issue:7908", + "issue:2364|issue:5565", + "issue:3250|issue:7306", + "issue:7159|issue:8518", + "issue:11088|issue:13105", + "issue:4213|issue:4488", + "issue:4566|issue:6252", + "issue:6112|issue:9889", + "issue:5466|issue:7003", + "issue:2015|issue:3826", + "issue:3816|issue:4356", + "issue:5540|issue:6028", + "issue:3674|issue:3700", + "issue:4566|issue:4985", + "issue:4000|issue:7782", + "issue:10421|issue:10569", + "issue:3422|issue:7363", + "issue:6749|issue:8885", + "issue:11182|issue:9525", + "issue:11222|issue:11223", + "issue:12575|issue:13176", + "issue:1704|issue:929", + "issue:1823|issue:776", + "issue:4708|issue:5892" + ], + "prepared_review_unit_hash": "1b69802de30fb65f455ff642d2accdea112f2476239b5eaaf0c62c14475209e3", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32947, + "estimated_input_tokens": 8237, + "estimated_eval_tokens": 16730 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: the suggested pairs mostly share only broad subsystem/theme similarity (training, loading, attention, memory, imports) rather than the same underlying bug or change. No pull requests are present. I\u2019d treat issue 11223 as the best representative of the closest concrete subcluster, but with low duplicate confidence overall.", + "confidence": 0.69, + "canonical_issue_reason": "issue:11223 is the most concrete and actionable report in the closest apparent subcluster (WAN + LoRA + OOM), with a specific failure mode and enough detail to anchor triage better than the more generic or feature-oriented items.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11223 is the strongest representative artifact here: it describes a specific reproducible bug, is more detailed than neighboring WAN/LoRA items, and is the clearest candidate for a canonical issue among the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7392", + "right": "issue:7908", + "accept": false, + "reason": "Both concern aspect ratios in training, but one is a how-to/problem report and the other is a feature request for bucketing; not the same bug/change." + }, + { + "left": "issue:2364", + "right": "issue:5565", + "accept": false, + "reason": "Different training scripts and different failure reports; only a broad \u201ccan\u2019t run training\u201d similarity." + }, + { + "left": "issue:3250", + "right": "issue:7306", + "accept": false, + "reason": "Different pipelines and different defects: one can\u2019t run a pipeline at all, the other is a float16-specific issue." + }, + { + "left": "issue:7159", + "right": "issue:8518", + "accept": false, + "reason": "Both involve torch.compile, but the model paths and error conditions differ materially; not one fixable code-path bug." + }, + { + "left": "issue:11088", + "right": "issue:13105", + "accept": false, + "reason": "One is a gguf load error and the other is noisy output from a WAN model; different symptoms and likely different root causes." + }, + { + "left": "issue:4213", + "right": "issue:4488", + "accept": false, + "reason": "Both touch SDXL loading, but they target different APIs and bug conditions; too broad to be the same underlying issue." + }, + { + "left": "issue:4566", + "right": "issue:6252", + "accept": false, + "reason": "Checkpoint resume and LoRA weight loading are related but distinct behaviors with different failure modes." + }, + { + "left": "issue:6112", + "right": "issue:9889", + "accept": false, + "reason": "Both mention sdxl-turbo problems, but one is output corruption/noise while the other is a segfault during inference." + }, + { + "left": "issue:5466", + "right": "issue:7003", + "accept": false, + "reason": "Different pipelines and different shape-mismatch contexts; not the same concrete bug." + }, + { + "left": "issue:2015", + "right": "issue:3826", + "accept": false, + "reason": "Separate pipeline ideas in different domains; no shared underlying bug or implementation change." + }, + { + "left": "issue:3816", + "right": "issue:4356", + "accept": false, + "reason": "Both are import-related errors, but they involve different missing symbols/modules and different fixes." + }, + { + "left": "issue:5540", + "right": "issue:6028", + "accept": false, + "reason": "Only the generic symptom overlaps (OOM); the models and circumstances are different." + }, + { + "left": "issue:3674", + "right": "issue:3700", + "accept": false, + "reason": "Same script, but one is about metadata parsing and the other about an argument not working; distinct issues." + }, + { + "left": "issue:4566", + "right": "issue:4985", + "accept": false, + "reason": "Both are training/LoRA-related, but one is checkpoint resume and the other is an OOM during a different training flow." + }, + { + "left": "issue:4000", + "right": "issue:7782", + "accept": false, + "reason": "Both involve conversion/loading problems, but they concern different file formats and error causes." + }, + { + "left": "issue:10421", + "right": "issue:10569", + "accept": false, + "reason": "Both are memory-related, but one is a multi-GPU pipeline leak and the other is high CPU memory use for a specific model." + }, + { + "left": "issue:3422", + "right": "issue:7363", + "accept": false, + "reason": "Both mention the safety checker, but one is a general question and the other is a specific SDXL request; not the same bug/change." + }, + { + "left": "issue:6749", + "right": "issue:8885", + "accept": false, + "reason": "Different community examples and distinct setup/runtime failures; no shared underlying defect." + }, + { + "left": "issue:11182", + "right": "issue:9525", + "accept": false, + "reason": "Unnecessary download for BF16 loading and Flux lora_scale behavior are unrelated issues." + }, + { + "left": "issue:11222", + "right": "issue:11223", + "accept": false, + "reason": "Related WAN/LoRA context, but one is a feature request for more LoRAs and the other is an OOM bug under layer-wise upcasting." + }, + { + "left": "issue:12575", + "right": "issue:13176", + "accept": false, + "reason": "Both are attention-related errors, but they appear to hit different components and failure surfaces." + }, + { + "left": "issue:1704", + "right": "issue:929", + "accept": false, + "reason": "A general question about an attention block is not the same as a specific implementation/performance concern." + }, + { + "left": "issue:1823", + "right": "issue:776", + "accept": false, + "reason": "Different Stable Diffusion bugs: swapped comments versus batch-size behavior in image-to-image." + }, + { + "left": "issue:4708", + "right": "issue:5892", + "accept": false, + "reason": "Both are generic import errors, but they are not the same missing module/symbol or fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 87, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:1013", + "issue:10139", + "issue:10180", + "issue:10421", + "issue:11033", + "issue:11214", + "issue:11653", + "issue:1195", + "issue:12609", + "issue:12699", + "issue:1976", + "issue:2091", + "issue:2321", + "issue:2459", + "issue:3651", + "issue:3901", + "issue:4708", + "issue:4830", + "issue:4954", + "issue:5023", + "issue:5398", + "issue:5429", + "issue:5442", + "issue:5529", + "issue:5628", + "issue:6149", + "issue:6295", + "issue:6561", + "issue:6761", + "issue:6911", + "issue:6967", + "issue:7133", + "issue:7168", + "issue:7223", + "issue:7239", + "issue:7782", + "issue:7982", + "issue:8405", + "issue:8464", + "issue:8515", + "issue:8626", + "issue:8659", + "issue:9063", + "issue:9141", + "issue:9392", + "issue:9484", + "issue:9971" + ], + "soft_pairs": [ + "issue:7168|issue:8626", + "issue:5628|issue:6295", + "issue:4954|issue:7782", + "issue:6149|issue:6561", + "issue:5398|issue:5429", + "issue:8515|issue:9392", + "issue:9063|issue:9141", + "issue:2459|issue:6911", + "issue:11214|issue:6761", + "issue:6967|issue:7239", + "issue:10180|issue:11653", + "issue:7982|issue:9971", + "issue:5442|issue:7223", + "issue:10421|issue:8464", + "issue:5529|issue:8659", + "issue:10026|issue:9484", + "issue:4708|issue:5023", + "issue:12609|issue:12699", + "issue:4830|issue:8405", + "issue:2091|issue:7133", + "issue:1976|issue:2321", + "issue:3651|issue:3901", + "issue:1013|issue:1195", + "issue:10139|issue:11033" + ], + "prepared_review_unit_hash": "63434410e73bf3ae0d706226e6d592ba4bb3be2b30a42507bae4c80e9702b37d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34051, + "estimated_input_tokens": 8513, + "estimated_eval_tokens": 17282 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly topical overlap rather than true duplication. The packet provides only soft-similarity signals, so none of the pairs should be promoted to accepted duplicate status; even the closest-looking MultiControlNet pair remains unconfirmed.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 8405 is the clearest anchor for the narrow MultiControlNet/multiple-ControlNet tensor-size symptom, but it should be treated only as the most representative soft match, not a confirmed duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 8405 is the best issue representative because its title states the shared failure mode most directly and broadly, even though the packet does not support treating it as a true duplicate cluster center.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7168", + "right": "issue:8626", + "accept": false, + "reason": "Both mention IP-Adapter usage, but the concrete problems differ: image-embed shape handling vs guidance for multiple images/masks." + }, + { + "left": "issue:5628", + "right": "issue:6295", + "accept": false, + "reason": "Scheduler/artifact wording overlaps, but the samplers and failure modes are different." + }, + { + "left": "issue:4954", + "right": "issue:7782", + "accept": false, + "reason": "Both involve unpickling errors, but the contexts are different: textual inversion loading vs safetensors conversion." + }, + { + "left": "issue:6149", + "right": "issue:6561", + "accept": false, + "reason": "Both are IP-Adapter load errors, but the titles do not indicate the same missing file or same root cause." + }, + { + "left": "issue:5398", + "right": "issue:5429", + "accept": false, + "reason": "Different hacktoberfest tasks with different code paths: DreamBooth/DeepSpeed vs model support without VAE." + }, + { + "left": "issue:8515", + "right": "issue:9392", + "accept": false, + "reason": "Duplicate-code cleanup is unrelated to a scheduler feature request." + }, + { + "left": "issue:9063", + "right": "issue:9141", + "accept": false, + "reason": "Both are FLUXPipeline-related, but one is tokenizer instantiation and the other is missing from_single_file support." + }, + { + "left": "issue:2459", + "right": "issue:6911", + "accept": false, + "reason": "Different APIs and bug classes: xformers attention in a training script vs class-label handling in a 3D UNet." + }, + { + "left": "issue:11214", + "right": "issue:6761", + "accept": false, + "reason": "Unexpected keyword errors, but on different functions and different components." + }, + { + "left": "issue:6967", + "right": "issue:7239", + "accept": false, + "reason": "Both are SDXL training shape mismatches, but the scripts and mismatched tensors appear different." + }, + { + "left": "issue:10180", + "right": "issue:11653", + "accept": false, + "reason": "Different model families and fixes: Flux Control LoRA multi-lora loading vs HiDream .alpha key support." + }, + { + "left": "issue:7982", + "right": "issue:9971", + "accept": false, + "reason": "MultiControlNet pipeline conversion and a FLUX scheduler issue are different code paths." + }, + { + "left": "issue:5442", + "right": "issue:7223", + "accept": false, + "reason": "They are both video-generation topics, but not the same bug or feature request." + }, + { + "left": "issue:10421", + "right": "issue:8464", + "accept": false, + "reason": "A CPU memory leak during multi-GPU pipeline moves is not the same as SEGA offload failing when a seed is set." + }, + { + "left": "issue:5529", + "right": "issue:8659", + "accept": false, + "reason": "Both involve training scripts, but the failure modes are different: SDXL script bug vs checkpoint resume/inference loading." + }, + { + "left": "issue:10026", + "right": "issue:9484", + "accept": false, + "reason": "Both are FLUX multi-GPU/DeepSpeed-related, but one is ControlNet memory behavior and the other is DreamBooth training." + }, + { + "left": "issue:4708", + "right": "issue:5023", + "accept": false, + "reason": "Import errors in different modules/pipelines; no evidence of a shared underlying cause." + }, + { + "left": "issue:12609", + "right": "issue:12699", + "accept": false, + "reason": "Both mention wan 2.2, but one is NaNs in latents and the other is quality degradation over repeated prompts." + }, + { + "left": "issue:4830", + "right": "issue:8405", + "accept": false, + "reason": "Very similar MultiControlNet tensor-size wording, but the packet only marks soft similarity and does not support calling this an accepted duplicate." + }, + { + "left": "issue:2091", + "right": "issue:7133", + "accept": false, + "reason": "Both are resume-from-checkpoint issues, but they affect different scripts and different failure modes." + }, + { + "left": "issue:1976", + "right": "issue:2321", + "accept": false, + "reason": "A missing tokenizer/config file lookup is different from a model-copying/ema_weights logic issue." + }, + { + "left": "issue:3651", + "right": "issue:3901", + "accept": false, + "reason": "A multi-GPU DreamBooth runtime issue is unrelated to a missing requirements.txt file." + }, + { + "left": "issue:1013", + "right": "issue:1195", + "accept": false, + "reason": "Pipeline import/versioning issues and fp16 memory-efficient attention issues are different bugs." + }, + { + "left": "issue:10139", + "right": "issue:11033", + "accept": false, + "reason": "Both concern loading/init, but sample_size handling and from_single_file loading are distinct problems." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary is not conservative enough: the packet marks every soft pair, including issue:4830 vs issue:8405, with `deterministic_accept: false`. The analyst overstates that this is an accepted duplicate pair and that it is the only accepted soft duplicate. The rest of the writeup is broadly reasonable, but it should not elevate any pair to accepted/true-duplicate status without packet support." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 88, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10139", + "issue:10334", + "issue:10542", + "issue:10722", + "issue:10798", + "issue:10969", + "issue:10994", + "issue:11024", + "issue:11050", + "issue:11104", + "issue:11315", + "issue:11581", + "issue:11651", + "issue:12104", + "issue:12533", + "issue:12661", + "issue:12771", + "issue:13043", + "issue:2132", + "issue:2156", + "issue:2203", + "issue:2361", + "issue:2780", + "issue:3101", + "issue:3279", + "issue:3537", + "issue:3612", + "issue:3694", + "issue:6117", + "issue:6382", + "issue:6444", + "issue:6527", + "issue:713", + "issue:7229", + "issue:7593", + "issue:7629", + "issue:7722", + "issue:7979", + "issue:7982", + "issue:8469", + "issue:8503", + "issue:8614", + "issue:8828", + "issue:9047", + "issue:9227", + "issue:9546", + "issue:9971" + ], + "soft_pairs": [ + "issue:7629|issue:9546", + "issue:6444|issue:7229", + "issue:10798|issue:13043", + "issue:6382|issue:8469", + "issue:6117|issue:6527", + "issue:3101|issue:713", + "issue:11024|issue:7593", + "issue:11581|issue:11651", + "issue:10139|issue:7982", + "issue:2132|issue:3279", + "issue:6382|issue:7722", + "issue:10542|issue:11104", + "issue:10994|issue:9227", + "issue:11050|issue:2780", + "issue:3537|issue:3612", + "issue:7979|issue:8614", + "issue:10722|issue:10969", + "issue:2361|issue:3694", + "issue:12104|issue:12771", + "issue:2156|issue:2203", + "issue:8503|issue:8828", + "issue:12533|issue:12661", + "issue:10334|issue:11315", + "issue:9047|issue:9971" + ], + "prepared_review_unit_hash": "de0ee9d1a613d4f2fc71f31fe8a6649cc384e34f4fdfa4b8863e07f3e83cd3de", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33944, + "estimated_input_tokens": 8486, + "estimated_eval_tokens": 17228 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly superficially similar bug reports across different models, pipelines, schedulers, and training scripts, not true duplicates. All soft-edge pairs should be rejected.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a strong global canonical here; the set is too heterogeneous and the pairwise similarity is mostly generic wording rather than the same bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7629", + "right": "issue:9546", + "accept": false, + "reason": "Both mention OOM, but one is SVD on PyTorch 1.13.1 and the other is Flux ControlNet validation OOM; different model and code path." + }, + { + "left": "issue:6444", + "right": "issue:7229", + "accept": false, + "reason": "Both involve training/checkpointing, but the failures are different: peft_config attribute error vs checkpoint save issue in advanced training." + }, + { + "left": "issue:10798", + "right": "issue:13043", + "accept": false, + "reason": "Different problems: BitsAndBytes device-combination bug for Flux quantization vs missing kernel image on RTX 5090 with sage_hub attention." + }, + { + "left": "issue:6382", + "right": "issue:8469", + "accept": false, + "reason": "Both involve IP-Adapter, but one is a key error after LCM LoRA training and the other is broken face generation with a different adapter/model." + }, + { + "left": "issue:6117", + "right": "issue:6527", + "accept": false, + "reason": "Scheduler bugs, but they are distinct classes: CMStochasticIterativeScheduler add_noise vs EulerDiscreteScheduler add_noise." + }, + { + "left": "issue:3101", + "right": "issue:713", + "accept": false, + "reason": "The titles are too generic and the reports are not clearly the same concrete failure or code path." + }, + { + "left": "issue:11024", + "right": "issue:7593", + "accept": false, + "reason": "Different issues: missing .to() on SiglipImageProcessor vs NoneType/image_projection_layers error in IP-Adapter image embedding prep." + }, + { + "left": "issue:11581", + "right": "issue:11651", + "accept": false, + "reason": "Both are FLUX/IP-related load failures, but one is an unexpected dtype argument and the other is a FluxControl loader problem." + }, + { + "left": "issue:10139", + "right": "issue:7982", + "accept": false, + "reason": "Unet sample_size pipeline-init bug is unrelated to MultiControlNetModel.from_pipe behavior." + }, + { + "left": "issue:2132", + "right": "issue:3279", + "accept": false, + "reason": "Different failures: Accelerator.trackers missing vs resume-from-checkpoint errors in dreambooth training." + }, + { + "left": "issue:6382", + "right": "issue:7722", + "accept": false, + "reason": "Both mention IP-Adapter, but the affected pipelines and symptom are different enough that they are not the same bug." + }, + { + "left": "issue:10542", + "right": "issue:11104", + "accept": false, + "reason": "Different products and failure modes: Hunyuan Video batch-size regression vs LTX-Video offload-method breakage." + }, + { + "left": "issue:10994", + "right": "issue:9227", + "accept": false, + "reason": "Both are FLUX training reports, but one is a batch-size crash and the other is a broader model/script problem." + }, + { + "left": "issue:11050", + "right": "issue:2780", + "accept": false, + "reason": "Different dtype-related bugs in different scripts: SD3 controlnet embed casting vs FP16 gradient unscale error." + }, + { + "left": "issue:3537", + "right": "issue:3612", + "accept": false, + "reason": "One is a conversion script request, the other is a question about creating ControlNet without a model; not the same issue." + }, + { + "left": "issue:7979", + "right": "issue:8614", + "accept": false, + "reason": "Both mention config.json, but one is invalid JSON while the other is a missing config file for single-file ControlNet." + }, + { + "left": "issue:10722", + "right": "issue:10969", + "accept": false, + "reason": "Different training failures: tensor size mismatch in DreamBooth prior preservation vs a 2-D weight error in FLUX-controlnet zero3." + }, + { + "left": "issue:2361", + "right": "issue:3694", + "accept": false, + "reason": "Checkpoint saving failure and missing model_index/config.json are separate example-script problems." + }, + { + "left": "issue:12104", + "right": "issue:12771", + "accept": false, + "reason": "Both are out-of-bounds errors, but they arise in different contexts and are not clearly the same scheduler bug." + }, + { + "left": "issue:2156", + "right": "issue:2203", + "accept": false, + "reason": "Training uses CPU instead of GPU is unrelated to model download location." + }, + { + "left": "issue:8503", + "right": "issue:8828", + "accept": false, + "reason": "Both are SD3 training issues, but one is specifically gradient checkpointing and the other is a general 16GB training error." + }, + { + "left": "issue:12533", + "right": "issue:12661", + "accept": false, + "reason": "Context parallelism appears in both, but one is a hook conflict and the other is a ChronoEdit crash; not the same concrete bug." + }, + { + "left": "issue:10334", + "right": "issue:11315", + "accept": false, + "reason": "Both hit MPS, but they are different models and failures: Sana NaNs/gray images vs HiDream rope float64 issue." + }, + { + "left": "issue:9047", + "right": "issue:9971", + "accept": false, + "reason": "Flux on MPS device failure is unrelated to DEISMultistepScheduler behavior on FLUX." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 89, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11432", + "issue:11922", + "issue:12107", + "issue:2091", + "issue:2161", + "issue:2299", + "issue:3279", + "issue:3346", + "issue:3453", + "issue:4286", + "issue:4436", + "issue:4453", + "issue:4566", + "issue:460", + "issue:4790", + "issue:4974", + "issue:5016", + "issue:5226", + "issue:5317", + "issue:5341", + "issue:5687", + "issue:5757", + "issue:5886", + "issue:5997", + "issue:6008", + "issue:6069", + "issue:6117", + "issue:6226", + "issue:6295", + "issue:6584", + "issue:6732", + "issue:7053", + "issue:7138", + "issue:7286", + "issue:7440", + "issue:7879", + "issue:7982", + "issue:8359", + "issue:8469", + "issue:8827", + "issue:8885", + "issue:8900", + "issue:9142", + "issue:9209", + "issue:9309", + "issue:9421", + "issue:9643", + "issue:9804" + ], + "soft_pairs": [ + "issue:5016|issue:5757", + "issue:5687|issue:6295", + "issue:2091|issue:3453", + "issue:12107|issue:9142", + "issue:4790|issue:6226", + "issue:7286|issue:8885", + "issue:6732|issue:7053", + "issue:6226|issue:9421", + "issue:3279|issue:3346", + "issue:2299|issue:460", + "issue:4436|issue:4453", + "issue:7440|issue:7879", + "issue:9209|issue:9804", + "issue:6069|issue:6117", + "issue:4286|issue:5226", + "issue:11922|issue:9309", + "issue:2161|issue:4566", + "issue:5317|issue:5341", + "issue:6008|issue:8469", + "issue:9142|issue:9643", + "issue:5886|issue:8827", + "issue:11432|issue:7982", + "issue:8359|issue:8900", + "issue:4974|issue:6584", + "issue:5997|issue:7138" + ], + "prepared_review_unit_hash": "943e67a7463cb594eff764c1130da97d681319bb36a4503c1e22e2a954828f61", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34203, + "estimated_input_tokens": 8551, + "estimated_eval_tokens": 17358 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly lexical similarity, not true duplication: it mixes resume-from-checkpoint bugs, model-loading issues, conversion failures, scheduler problems, and adapter/ControlNet reports. Only one pair looks like the same concrete change request/bug in the same script.", + "confidence": 0.88, + "canonical_issue_reason": "issue:3346 is the best anchor because it explicitly asks for resume_from_checkpoint support in `examples/dreambooth/train_dreambooth_lora.py`, and issue:3279 reports the same checkpoint-resume failure in that same script.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:3346 is the clearest representative of the only real duplicate-like pair in the set; the rest of the items are unrelated bugs across different features.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5016", + "right": "issue:5757", + "accept": false, + "reason": "Both mention DreamBooth LoRA resume behavior, but one is about SDXL text-encoder training and the other is a checkpoint resume error in a different script/version." + }, + { + "left": "issue:5687", + "right": "issue:6295", + "accept": false, + "reason": "Both are scheduler/artifact complaints, but they describe different schedulers and different failure modes." + }, + { + "left": "issue:2091", + "right": "issue:3453", + "accept": false, + "reason": "One is a multi-GPU checkpoint resume issue; the other is a BF16 CUDA kernel unsupported error. Different bugs." + }, + { + "left": "issue:12107", + "right": "issue:9142", + "accept": false, + "reason": "Both involve lists/custom objects, but one is `accelerator.init_trackers` and the other is multiple ControlNet loading in a different pipeline." + }, + { + "left": "issue:4790", + "right": "issue:6226", + "accept": false, + "reason": "Same converter script, but the reports point to different downstream failures; not enough evidence they are the same underlying bug." + }, + { + "left": "issue:7286", + "right": "issue:8885", + "accept": false, + "reason": "Both mention loading-related problems, but they are for different features: Custom Diffusion cross-attention weights vs clip-retrieval/modifier tokens." + }, + { + "left": "issue:6732", + "right": "issue:7053", + "accept": false, + "reason": "Documentation for `trained_betas` is unrelated to `UNet3DConditionModel` class conditioning." + }, + { + "left": "issue:6226", + "right": "issue:9421", + "accept": false, + "reason": "Converter output compatibility is unrelated to simply failing to access a model on the hub." + }, + { + "left": "issue:3279", + "right": "issue:3346", + "accept": true, + "reason": "Same training example and same resume-from-checkpoint code path; one reports the error, the other requests the missing support." + }, + { + "left": "issue:2299", + "right": "issue:460", + "accept": false, + "reason": "Checkpoint conversion support and ONNX runtime dependency issues are different problems." + }, + { + "left": "issue:4436", + "right": "issue:4453", + "accept": false, + "reason": "Both are conversion topics, but one is safetensors-with-VAE to Diffusers and the other is Diffusers LoRA to A1111 safetensors." + }, + { + "left": "issue:7440", + "right": "issue:7879", + "accept": false, + "reason": "One says the pokemon dataset is unavailable; the other asks for a pokemon-blip-captions dataset. Related topic, not the same bug." + }, + { + "left": "issue:9209", + "right": "issue:9804", + "accept": false, + "reason": "A mirror/network access failure is unrelated to a local `ControlNetModel.from_pretrained` loading bug." + }, + { + "left": "issue:6069", + "right": "issue:6117", + "accept": false, + "reason": "Different schedulers and different `add_noise` failures." + }, + { + "left": "issue:4286", + "right": "issue:5226", + "accept": false, + "reason": "Both involve `load_lora_weights`, but they are different pipelines and not clearly the same concrete defect." + }, + { + "left": "issue:11922", + "right": "issue:9309", + "accept": false, + "reason": "Different Flux model/pipeline problems; same family name is not enough to merge them." + }, + { + "left": "issue:2161", + "right": "issue:4566", + "accept": false, + "reason": "LoRA training on inpainting models is a different issue from SDXL checkpoint resume errors." + }, + { + "left": "issue:5317", + "right": "issue:5341", + "accept": false, + "reason": "Documentation of PEFT API vs updating example scripts are related but distinct tasks." + }, + { + "left": "issue:6008", + "right": "issue:8469", + "accept": false, + "reason": "Both are IP-Adapter face issues, but they involve different models and symptoms." + }, + { + "left": "issue:9142", + "right": "issue:9643", + "accept": false, + "reason": "Both mention multiple ControlNets, but one is a Hunyuan DiT `list` error and the other is a Flux feature/support question." + }, + { + "left": "issue:5886", + "right": "issue:8827", + "accept": false, + "reason": "Different IP-Adapter face variants and different failure reports." + }, + { + "left": "issue:11432", + "right": "issue:7982", + "accept": false, + "reason": "`torch_dtype=\"auto\"` in `from_pretrained` is unrelated to `from_pipe` with `MultiControlNetModel`." + }, + { + "left": "issue:8359", + "right": "issue:8900", + "accept": false, + "reason": "Both use `from_single_file`, but one is about custom checkpoints broadly and the other is a specific SDXL refiner safetensors load path." + }, + { + "left": "issue:4974", + "right": "issue:6584", + "accept": false, + "reason": "General safetensors-to-Diffusers conversion and a specific DreamShaper inpaint conversion failure are not clearly the same bug." + }, + { + "left": "issue:5997", + "right": "issue:7138", + "accept": false, + "reason": "A tutorial/model-type mismatch is a different problem from SDXL UNet structure mismatch." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: it correctly identifies issue:3279 and issue:3346 as the only clearly duplicate-like pair, and the remaining pairwise judgments stay conservative rather than overclaiming duplication. No soft-edge verdict appears to overstate the evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 90, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10180", + "issue:10382", + "issue:10650", + "issue:10772", + "issue:10823", + "issue:10866", + "issue:10937", + "issue:11169", + "issue:11560", + "issue:12396", + "issue:12547", + "issue:12656", + "issue:12771", + "issue:1304", + "issue:1534", + "issue:1556", + "issue:2107", + "issue:2830", + "issue:2988", + "issue:3728", + "issue:3813", + "issue:434", + "issue:4580", + "issue:4631", + "issue:4724", + "issue:4925", + "issue:5012", + "issue:5026", + "issue:5152", + "issue:5197", + "issue:5412", + "issue:5432", + "issue:5846", + "issue:5997", + "issue:6407", + "issue:7041", + "issue:7168", + "issue:7238", + "issue:7420", + "issue:9237", + "issue:9331", + "issue:938", + "issue:9501", + "issue:9619", + "issue:9907", + "issue:9936" + ], + "soft_pairs": [ + "issue:10026|issue:5197", + "issue:10382|issue:9331", + "issue:2830|issue:938", + "issue:4580|issue:4631", + "issue:7168|issue:7238", + "issue:2988|issue:7041", + "issue:12656|issue:12771", + "issue:3728|issue:6407", + "issue:1534|issue:434", + "issue:10180|issue:12396", + "issue:4925|issue:5997", + "issue:5012|issue:5432", + "issue:9237|issue:9501", + "issue:9907|issue:9936", + "issue:10772|issue:9619", + "issue:10650|issue:10866", + "issue:10823|issue:12547", + "issue:10937|issue:7420", + "issue:1304|issue:3813", + "issue:4724|issue:7041", + "issue:5026|issue:5152", + "issue:5412|issue:5846", + "issue:11169|issue:11560", + "issue:1556|issue:2107" + ], + "prepared_review_unit_hash": "768e45bb287aed2b64b673c971313e626efc2b6169d0f37005a25727716342b2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33513, + "estimated_input_tokens": 8379, + "estimated_eval_tokens": 17014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster does not contain any conservatively grounded duplicate pair. Most pairs are only superficially related by subsystem or shared keywords, and the 5412/5846 resume_from_checkpoint pair is not strong enough to accept because the packet shows different training scripts with no body-level evidence of the same underlying bug.", + "confidence": 0.91, + "canonical_issue_reason": "No issue in this packet is strong enough to serve as a safe duplicate canonical for the cluster as a whole. If a best-in-group issue must be named, 5412 is the closest anchor for the resume_from_checkpoint topic, but not enough to justify merging 5846 into it.", + "canonical_pr_reason": null, + "best_issue_reason": "5412 is the strongest issue anchor only in the narrow sense that it names the script and failure mode most directly; however, it should not be treated as a confirmed duplicate canonical because 5846 points to a different script and the packet lacks shared-bug evidence.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10026", + "right": "issue:5197", + "accept": false, + "reason": "Both are multi-GPU training reports, but the symptoms differ (memory scaling vs slower throughput) and they do not indicate the same bug." + }, + { + "left": "issue:10382", + "right": "issue:9331", + "accept": false, + "reason": "VAE encoder return-type mismatch and FlowMatchEuler index errors are different components and failure modes." + }, + { + "left": "issue:2830", + "right": "issue:938", + "accept": false, + "reason": "These are different tasks altogether: image-to-video versus text-to-motion." + }, + { + "left": "issue:4580", + "right": "issue:4631", + "accept": false, + "reason": "Shared use_karras_sigmas alone is not enough; the schedulers and exceptions differ." + }, + { + "left": "issue:7168", + "right": "issue:7238", + "accept": false, + "reason": "One is about embedding tensor shape, the other about masking support; different issues." + }, + { + "left": "issue:2988", + "right": "issue:7041", + "accept": false, + "reason": "Different crash conditions around step counts; not the same defect." + }, + { + "left": "issue:12656", + "right": "issue:12771", + "accept": false, + "reason": "Both are out-of-bounds reports, but in different subsystems with different root causes." + }, + { + "left": "issue:3728", + "right": "issue:6407", + "accept": false, + "reason": "These are separate feature requests, not duplicate bugs or changes." + }, + { + "left": "issue:1534", + "right": "issue:434", + "accept": false, + "reason": "Both are training notebook complaints, but the reports are too generic and lack a shared concrete failure." + }, + { + "left": "issue:10180", + "right": "issue:12396", + "accept": false, + "reason": "Both mention adapters/LoRA, but the actual behaviors differ: loading multiple LoRAs versus delete_adapters/leaf_level errors." + }, + { + "left": "issue:4925", + "right": "issue:5997", + "accept": false, + "reason": "Distributed OOM in training and a tutorial type-mismatch import error are unrelated." + }, + { + "left": "issue:5012", + "right": "issue:5432", + "accept": false, + "reason": "Same script family, but one is about training the text encoder and the other is blank validation images." + }, + { + "left": "issue:9237", + "right": "issue:9501", + "accept": false, + "reason": "Both are Flux Dreambooth reports, but the actual problems differ and are not clearly the same code-path bug." + }, + { + "left": "issue:9907", + "right": "issue:9936", + "accept": false, + "reason": "A sequence-length warning and an NCCL timeout are distinct symptoms with different likely causes." + }, + { + "left": "issue:10772", + "right": "issue:9619", + "accept": false, + "reason": "A Sana ControlNet support request and a ControlNet interrupt feature request are unrelated." + }, + { + "left": "issue:10650", + "right": "issue:10866", + "accept": false, + "reason": "One is a Lumina shape/runtime error, the other a LoRA compatibility issue." + }, + { + "left": "issue:10823", + "right": "issue:12547", + "accept": false, + "reason": "Missing inference outputs and a required --instance_prompt argument are unrelated training-script issues." + }, + { + "left": "issue:10937", + "right": "issue:7420", + "accept": false, + "reason": "One is a specific vae.encode compile error, the other a broad torch.compile complaint; not enough to treat as one bug." + }, + { + "left": "issue:1304", + "right": "issue:3813", + "accept": false, + "reason": "These are broad how-to questions about different workflows, not duplicate bugs." + }, + { + "left": "issue:4724", + "right": "issue:7041", + "accept": false, + "reason": "Img2Img checkpoint behavior and an inference-step segfault are unrelated." + }, + { + "left": "issue:5026", + "right": "issue:5152", + "accept": false, + "reason": "A text_encoder KeyError and a Wurstchen alpha bug are different issues." + }, + { + "left": "issue:5412", + "right": "issue:5846", + "accept": false, + "reason": "Both mention resume_from_checkpoint, but they are different scripts and the packet provides no body evidence that they share the same underlying bug." + }, + { + "left": "issue:11169", + "right": "issue:11560", + "accept": false, + "reason": "A specific NumPy binary incompatibility traceback and a generic import error are not enough to establish one underlying bug." + }, + { + "left": "issue:1556", + "right": "issue:2107", + "accept": false, + "reason": "Both mention black images, but the reported conditions and likely causes differ." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The only accepted soft-edge pair is not sufficiently grounded in the packet. 5412 and 5846 do both mention `resume_from_checkpoint`, but they are different training scripts (`train_dreambooth_lora_sdxl.py` vs `train_text_to_image_lora_sdxl.py`), and the packet provides no body evidence showing they share the same underlying bug. This is too strong for a conservative duplicate verdict, so the cluster summary overstates the evidence." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 91, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10804", + "issue:10883", + "issue:11432", + "issue:11627", + "issue:11651", + "issue:1197", + "issue:12752", + "issue:1919", + "issue:2107", + "issue:2132", + "issue:2156", + "issue:2218", + "issue:2596", + "issue:3124", + "issue:3250", + "issue:3284", + "issue:3454", + "issue:3651", + "issue:3799", + "issue:3954", + "issue:4080", + "issue:4356", + "issue:4390", + "issue:4806", + "issue:4936", + "issue:557", + "issue:5603", + "issue:5753", + "issue:5798", + "issue:5799", + "issue:6159", + "issue:6224", + "issue:6278", + "issue:7223", + "issue:7377", + "issue:746", + "issue:761", + "issue:861", + "issue:8732", + "issue:9112", + "issue:9187", + "issue:9403", + "issue:9448", + "issue:9576", + "issue:9744", + "issue:9825", + "issue:9841" + ], + "soft_pairs": [ + "issue:3124|issue:5753", + "issue:2596|issue:5603", + "issue:7377|issue:9112", + "issue:3250|issue:861", + "issue:3799|issue:4806", + "issue:4356|issue:4936", + "issue:7223|issue:9744", + "issue:2132|issue:2156", + "issue:2218|issue:761", + "issue:10804|issue:11651", + "issue:1197|issue:4080", + "issue:2132|issue:3454", + "issue:3250|issue:746", + "issue:1919|issue:4390", + "issue:3954|issue:6278", + "issue:2107|issue:557", + "issue:6224|issue:8732", + "issue:11627|issue:9448", + "issue:9403|issue:9825", + "issue:9576|issue:9841", + "issue:2132|issue:3284", + "issue:11432|issue:9187", + "issue:3651|issue:3954", + "issue:5799|issue:6159", + "issue:10883|issue:12752", + "issue:5798|issue:5799" + ], + "prepared_review_unit_hash": "a550ebde87d1a6081131034e6f80e850ef2a8b2445f3827294ac86888d0c565d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 33793, + "estimated_input_tokens": 8449, + "estimated_eval_tokens": 17154 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are unrelated bug reports, feature requests, or training/examples issues. The soft pairs only share broad terminology and should not be deduplicated; none look like the same underlying bug or change.", + "confidence": 0.88, + "canonical_issue_reason": "If one issue must be chosen, issue 11651 is the most concrete standalone bug report: it has a specific FLUXControl LoRA load failure, a clear affected model/pipeline, and the most actionable problem statement. The rest of the cluster is too mixed to form a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11651 is the best overall issue candidate because it is a focused, reproducible runtime failure rather than a broad feature request or a vague compatibility report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3124", + "right": "issue:5753", + "accept": false, + "reason": "Both are model-loading/runtime problems, but one is a Stable Diffusion conversion failure and the other is an SDXL ControlNet LoRA-loading issue; different code paths and symptoms." + }, + { + "left": "issue:2596", + "right": "issue:5603", + "accept": false, + "reason": "These concern different documentation/import problems: missing `load_image` in docs versus an `is_safetensors_available` import error." + }, + { + "left": "issue:7377", + "right": "issue:9112", + "accept": false, + "reason": "Different pipelines and failures: StableCascade decoder `num_images_per_prompt` handling versus Auraflow tensor shape mismatch." + }, + { + "left": "issue:3250", + "right": "issue:861", + "accept": false, + "reason": "One is a KDiffusion pipeline runtime issue; the other is a checkpoint-to-diffusers conversion problem." + }, + { + "left": "issue:3799", + "right": "issue:4806", + "accept": false, + "reason": "These are unrelated typing bugs in different objects: scheduler `from_pretrained` inference versus `UNet2DModel` parameter typing." + }, + { + "left": "issue:4356", + "right": "issue:4936", + "accept": false, + "reason": "Different breakages: missing dummy module import on install versus DreamBooth LoRA resume logic." + }, + { + "left": "issue:7223", + "right": "issue:9744", + "accept": false, + "reason": "Both are model-addition requests, but for different video models and different pipeline surfaces; not the same change." + }, + { + "left": "issue:2132", + "right": "issue:2156", + "accept": false, + "reason": "Both are DreamBooth training issues, but one is an `Accelerator.trackers` attribute error and the other is training falling back to CPU; different failures." + }, + { + "left": "issue:2218", + "right": "issue:761", + "accept": false, + "reason": "Different training problems: overtraining behavior versus failure with 8-bit Adam optimizer." + }, + { + "left": "issue:10804", + "right": "issue:11651", + "accept": false, + "reason": "Both mention FLUX LoRA loading, but one is a ComfyUI-compatible FLUX1.dev LoRA load failure and the other is a FluxControl flux-fill-lora load failure; no evidence of the same root cause." + }, + { + "left": "issue:1197", + "right": "issue:4080", + "accept": false, + "reason": "A community OpenAI diffusion pipeline request is unrelated to an SDXL inpainting pipeline request." + }, + { + "left": "issue:2132", + "right": "issue:3454", + "accept": false, + "reason": "Both are DreamBooth LoRA training failures, but the reported symptoms and regression points differ; no concrete shared bug path is shown." + }, + { + "left": "issue:3250", + "right": "issue:746", + "accept": false, + "reason": "Different loading/execution paths: StableDiffusionKDiffusionPipeline runtime versus ONNX revision loading." + }, + { + "left": "issue:1919", + "right": "issue:4390", + "accept": false, + "reason": "Both are packaging/dependency complaints, but one is missing datasets for a DreamBooth example and the other is a generic missing requirements report." + }, + { + "left": "issue:3954", + "right": "issue:6278", + "accept": false, + "reason": "Both touch multi-GPU training, but one is about LR scheduler step counts and the other is about an example that cannot run on multi-GPU." + }, + { + "left": "issue:2107", + "right": "issue:557", + "accept": false, + "reason": "One reports low-resolution black/glitchy images, while the other compares DDIM results quality; not the same defect." + }, + { + "left": "issue:6224", + "right": "issue:8732", + "accept": false, + "reason": "TorchDynamo fine-tuning failure and checkpoint-saving failure are different training-stage problems." + }, + { + "left": "issue:11627", + "right": "issue:9448", + "accept": false, + "reason": "Different IP-Adapter issues in different pipelines: SD3.5 InstantX support versus a tuple `.shape` error in ControlNet inpainting." + }, + { + "left": "issue:9403", + "right": "issue:9825", + "accept": false, + "reason": "Both are FLUX/IP-Adapter feature requests, but they target different adapter support and are not the same concrete change." + }, + { + "left": "issue:9576", + "right": "issue:9841", + "accept": false, + "reason": "A loading bug with `use_safetensors` propagation and a DreamBooth SDXL LoRA `use_dora=False` type error are unrelated." + }, + { + "left": "issue:2132", + "right": "issue:3284", + "accept": false, + "reason": "Both mention Accelerate in DreamBooth training, but the specific failures differ: missing `trackers` attribute versus a broader run failure." + }, + { + "left": "issue:11432", + "right": "issue:9187", + "accept": false, + "reason": "These are separate loading issues: `torch_dtype=\"auto\"` on `from_pretrained` versus SDXL IP-Adapter behavior after Kolors update." + }, + { + "left": "issue:3651", + "right": "issue:3954", + "accept": false, + "reason": "Related only by multi-GPU context; one is inability to run DreamBooth multi-GPU, the other is incorrect scheduler step counting." + }, + { + "left": "issue:5799", + "right": "issue:6159", + "accept": false, + "reason": "Both are unsupported keyword-argument issues in inpainting-related pipelines, but they concern different kwargs, pipelines, and code paths." + }, + { + "left": "issue:10883", + "right": "issue:12752", + "accept": false, + "reason": "A Marigold model/docs update and a Z-Image img2img/inpainting feature request are unrelated." + }, + { + "left": "issue:5798", + "right": "issue:5799", + "accept": false, + "reason": "Both are kwarg compatibility bugs, but `callback_on_step_end` in ControlNet and `strength` in inpainting are distinct API gaps." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 92, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10302", + "issue:10428", + "issue:10514", + "issue:10542", + "issue:10962", + "issue:11010", + "issue:11363", + "issue:11436", + "issue:11777", + "issue:12396", + "issue:12738", + "issue:13079", + "issue:2089", + "issue:2412", + "issue:2616", + "issue:2707", + "issue:2844", + "issue:3481", + "issue:4307", + "issue:4402", + "issue:455", + "issue:4566", + "issue:457", + "issue:4849", + "issue:5128", + "issue:5395", + "issue:5518", + "issue:579", + "issue:5919", + "issue:6131", + "issue:6149", + "issue:6226", + "issue:6460", + "issue:6839", + "issue:684", + "issue:7142", + "issue:8435", + "issue:8543", + "issue:8659", + "issue:8708", + "issue:9096", + "issue:9200", + "issue:9258", + "issue:9264", + "issue:9297", + "issue:9673", + "issue:9861" + ], + "soft_pairs": [ + "issue:5395|issue:5518", + "issue:2616|issue:5919", + "issue:6131|issue:6149", + "issue:10514|issue:10962", + "issue:9297|issue:9673", + "issue:11010|issue:9861", + "issue:2089|issue:684", + "issue:6460|issue:9264", + "issue:8543|issue:8708", + "issue:4402|issue:6839", + "issue:455|issue:457", + "issue:11777|issue:9258", + "issue:2844|issue:4307", + "issue:2707|issue:6226", + "issue:10542|issue:12738", + "issue:10302|issue:12396", + "issue:4566|issue:8659", + "issue:4849|issue:579", + "issue:11363|issue:13079", + "issue:3481|issue:7142", + "issue:5128|issue:8435", + "issue:10428|issue:9096", + "issue:11436|issue:9200", + "issue:2089|issue:2412" + ], + "prepared_review_unit_hash": "4de6fa052fdf05a2171ae7d1ee687e5e2837530280602b7ee88d4ba1d3a34294", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33259, + "estimated_input_tokens": 8315, + "estimated_eval_tokens": 16886 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous set of issue reports across different pipelines/features. The only strong duplicate-like pairs are SDXL checkpoint-resume failures and multi-LoRA loading performance complaints; the rest share broad subsystem wording but appear to be different bugs or models.", + "confidence": 0.69, + "canonical_issue_reason": "issue:8659 is the best canonical issue: it is open, detailed, and covers the same SDXL checkpoint-resume/load-failure path as issue:4566, with broader symptom coverage.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8659 is the strongest global issue candidate because it is the clearest representative of one of the true duplicate pairs and is still open for consolidation.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5395", + "right": "issue:5518", + "accept": false, + "reason": "Both involve LoRA/PEFT, but one is about combining civitai loras and the other about a missing set_lora_layer attribute; different failures." + }, + { + "left": "issue:2616", + "right": "issue:5919", + "accept": false, + "reason": "Different bugs in LoRA saving: one is a NoneType replace error in save_attn_procs, the other is a missing set_lora_layer method." + }, + { + "left": "issue:6131", + "right": "issue:6149", + "accept": false, + "reason": "Both are IP-Adapter related, but the titles point to different adapter variants and different failure modes." + }, + { + "left": "issue:10514", + "right": "issue:10962", + "accept": false, + "reason": "Different model-specific shape problems: Sana resolution binning/sample size vs CogView4 prompt-embed shape handling." + }, + { + "left": "issue:9297", + "right": "issue:9673", + "accept": true, + "reason": "Same underlying multi-LoRA loading performance problem; one reports linear load-time growth and the other high CPU usage during the same operation." + }, + { + "left": "issue:11010", + "right": "issue:9861", + "accept": false, + "reason": "Both mention Flux, but one is architecture support and the other is a training-update symptom; not the same bug." + }, + { + "left": "issue:2089", + "right": "issue:684", + "accept": false, + "reason": "Both mention memory, but one is DreamBooth CUDA OOM on AltDiffusion and the other is AMD memory-optimization support; unrelated root causes." + }, + { + "left": "issue:6460", + "right": "issue:9264", + "accept": false, + "reason": "One is an inpainting model load hang, the other is a request for a Flux inpainting model; not duplicates." + }, + { + "left": "issue:8543", + "right": "issue:8708", + "accept": false, + "reason": "Different SD3 training issues: gradient checkpointing failure vs missing vae.config.shift_factor in examples." + }, + { + "left": "issue:4402", + "right": "issue:6839", + "accept": false, + "reason": "Both are from_single_file issues, but for different pipelines and failure modes; too broad to merge." + }, + { + "left": "issue:455", + "right": "issue:457", + "accept": false, + "reason": "Both are MPS-related, but one is an aten::index.Tensor NotImplementedError and the other is blue-artifact rendering; different bugs." + }, + { + "left": "issue:11777", + "right": "issue:9258", + "accept": false, + "reason": "Training question vs performance/debugging question; not the same issue." + }, + { + "left": "issue:2844", + "right": "issue:4307", + "accept": false, + "reason": "Explicit image-embedding runtime error vs local_files_only from_pretrained error; unrelated." + }, + { + "left": "issue:2707", + "right": "issue:6226", + "accept": false, + "reason": "Same script file, but one is a syntax error and the other is a loading regression; not the same concrete bug." + }, + { + "left": "issue:10542", + "right": "issue:12738", + "accept": false, + "reason": "Both are batch-size >1 failures, but they affect different pipelines with different assertions." + }, + { + "left": "issue:10302", + "right": "issue:12396", + "accept": false, + "reason": "FP8 inference noise without CPU offloading is unrelated to leaf_level/delete_adapters errors." + }, + { + "left": "issue:4566", + "right": "issue:8659", + "accept": true, + "reason": "Same SDXL checkpoint-resume/load-failure path; 8659 appears to subsume the resume issue and is the better merged target." + }, + { + "left": "issue:4849", + "right": "issue:579", + "accept": false, + "reason": "Textual inversion multi-GPU usage vs missing config.json; unrelated." + }, + { + "left": "issue:11363", + "right": "issue:13079", + "accept": false, + "reason": "Both are CUDA OOM reports, but on different Flux inference scenarios; not the same bug." + }, + { + "left": "issue:3481", + "right": "issue:7142", + "accept": false, + "reason": "Different determinism/output-difference complaints in different pipelines; not the same code-path issue." + }, + { + "left": "issue:5128", + "right": "issue:8435", + "accept": false, + "reason": "Both are from_single_file errors, but the specific pipelines and symptoms differ." + }, + { + "left": "issue:10428", + "right": "issue:9096", + "accept": false, + "reason": "Both mention Flux inference errors, but one is Ascend NPU-specific and the other is torch.half NaNs; different root causes." + }, + { + "left": "issue:11436", + "right": "issue:9200", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different models/pipelines with different tensor types." + }, + { + "left": "issue:2089", + "right": "issue:2412", + "accept": false, + "reason": "General DreamBooth training problems, but one is AltDiffusion OOM and the other is a broader error using dreambooth; not a duplicate." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary stays mostly grounded in the packet and the false-pair verdicts are conservative. The two accepted soft pairs are plausible from the titles, and the canonical/open-issue choice is reasonable. No clear overstatement that would force rejection." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 93, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10392", + "issue:10421", + "issue:10447", + "issue:10512", + "issue:10672", + "issue:10757", + "issue:10963", + "issue:11199", + "issue:11225", + "issue:11241", + "issue:11352", + "issue:11374", + "issue:11561", + "issue:11710", + "issue:1185", + "issue:13067", + "issue:1565", + "issue:1919", + "issue:1944", + "issue:2091", + "issue:2695", + "issue:2904", + "issue:2991", + "issue:3094", + "issue:3191", + "issue:3250", + "issue:4398", + "issue:4733", + "issue:4790", + "issue:4837", + "issue:5226", + "issue:5412", + "issue:5892", + "issue:6176", + "issue:6524", + "issue:6561", + "issue:6790", + "issue:6794", + "issue:6894", + "issue:7197", + "issue:7310", + "issue:817", + "issue:8493", + "issue:8511", + "issue:8790", + "issue:9109", + "issue:9358", + "issue:9876" + ], + "soft_pairs": [ + "issue:1565|issue:3094", + "issue:2695|issue:2991", + "issue:10392|issue:10512", + "issue:10672|issue:11710", + "issue:10963|issue:11199", + "issue:2091|issue:7197", + "issue:3191|issue:4398", + "issue:1919|issue:817", + "issue:5226|issue:9109", + "issue:10421|issue:11352", + "issue:2904|issue:3250", + "issue:6790|issue:8493", + "issue:10447|issue:10757", + "issue:4790|issue:4837", + "issue:4733|issue:7310", + "issue:6176|issue:9876", + "issue:6524|issue:6561", + "issue:11225|issue:11241", + "issue:11374|issue:13067", + "issue:11561|issue:8790", + "issue:1185|issue:1944", + "issue:5412|issue:6894", + "issue:5892|issue:6794", + "issue:8511|issue:9358" + ], + "prepared_review_unit_hash": "97c77e14cd9e480e4053b909b464c201395780e4b549d5d339ba9f2d6c560119", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33982, + "estimated_input_tokens": 8496, + "estimated_eval_tokens": 17248 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items do not form a true duplicate cluster; they span distinct bugs, feature requests, and converter/import/training failures. The soft similarities are mostly topical (LoRA/Flux/training/imports) rather than the same underlying defect, so all soft edges should be rejected.", + "confidence": 0.44, + "canonical_issue_reason": "Issue 9358 is the clearest, most concrete open bug in the set and best represents the Flux/LoRA text-encoder/memory-management subset, even though the overall collection is not a single duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9358 is the most actionable and well-scoped issue among mostly unrelated entries.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1565", + "right": "issue:3094", + "accept": false, + "reason": "Both involve OOM, but one is GPU selection/multi-GPU usage and the other is training fine-tuning memory usage; different bugs." + }, + { + "left": "issue:2695", + "right": "issue:2991", + "accept": false, + "reason": "Both are training-example validation problems, but they affect different examples and failure modes." + }, + { + "left": "issue:10392", + "right": "issue:10512", + "accept": false, + "reason": "Both mention Flux/LoRA, but one is a specific FP8/Hyper-FLUX setup and the other is a Quanto loading failure; not the same code path." + }, + { + "left": "issue:10672", + "right": "issue:11710", + "accept": false, + "reason": "One is a callback API feature request and the other is a video-cache feature request; unrelated changes." + }, + { + "left": "issue:10963", + "right": "issue:11199", + "accept": false, + "reason": "Both are import errors, but for different modules and likely different root causes." + }, + { + "left": "issue:2091", + "right": "issue:7197", + "accept": false, + "reason": "Both concern resuming training, but the reported errors and contexts differ enough to be separate bugs." + }, + { + "left": "issue:3191", + "right": "issue:4398", + "accept": false, + "reason": "Both involve Dreambooth/LoRA, but the symptoms and pipelines are different." + }, + { + "left": "issue:1919", + "right": "issue:817", + "accept": false, + "reason": "Missing dependency vs mixed-precision malfunction; unrelated fixes." + }, + { + "left": "issue:5226", + "right": "issue:9109", + "accept": false, + "reason": "Different subsystems entirely: LoRA loading in a pipeline versus a torchvision operator import issue." + }, + { + "left": "issue:10421", + "right": "issue:11352", + "accept": false, + "reason": "Memory leak moving pipelines across GPUs is not the same as a missing warning when moving an 8-bit model to CPU." + }, + { + "left": "issue:2904", + "right": "issue:3250", + "accept": false, + "reason": "Different pipelines and different runtime errors." + }, + { + "left": "issue:6790", + "right": "issue:8493", + "accept": false, + "reason": "Both are state_dict loading issues, but they involve different models and mismatch causes." + }, + { + "left": "issue:10447", + "right": "issue:10757", + "accept": false, + "reason": "Same script family, but one is a file-splitting bug and the other is a variable-name cleanup; separate changes." + }, + { + "left": "issue:4790", + "right": "issue:4837", + "accept": false, + "reason": "Both touch conversion scripts, but one is round-trip incompatibility and the other is an SDXL conversion failure." + }, + { + "left": "issue:4733", + "right": "issue:7310", + "accept": false, + "reason": "Checkpoint-resume optimizer error vs default training-script behavior; not the same bug." + }, + { + "left": "issue:6176", + "right": "issue:9876", + "accept": false, + "reason": "LoRA loading failure after training versus VRAM not being released after training are different issues." + }, + { + "left": "issue:6524", + "right": "issue:6561", + "accept": false, + "reason": "A second LoRA load breaking behavior is unrelated to a missing IP-Adapter config file." + }, + { + "left": "issue:11225", + "right": "issue:11241", + "accept": false, + "reason": "Different training failures: validation OOM versus floating-point exception." + }, + { + "left": "issue:11374", + "right": "issue:13067", + "accept": false, + "reason": "Separate feature requests for different model families." + }, + { + "left": "issue:11561", + "right": "issue:8790", + "accept": false, + "reason": "Both mention IP Adapter, but one is FluxFill support and the other is SDXL TensorRT integration; different work." + }, + { + "left": "issue:1185", + "right": "issue:1944", + "accept": false, + "reason": "Reproducibility issue versus poor generation quality; not the same defect." + }, + { + "left": "issue:5412", + "right": "issue:6894", + "accept": false, + "reason": "Both reference train_dreambooth_lora_sdxl, but one is resume_checkpoint failure and the other is A1111 LoRA compatibility." + }, + { + "left": "issue:5892", + "right": "issue:6794", + "accept": false, + "reason": "Generic import error versus a specific failed import from training_utils; insufficiently aligned." + }, + { + "left": "issue:8511", + "right": "issue:9358", + "accept": false, + "reason": "Both concern text-encoder handling in Flux/SD3 training, but they describe different scripts and different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 94, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10470", + "issue:10674", + "issue:10989", + "issue:1181", + "issue:12011", + "issue:12147", + "issue:12321", + "issue:12412", + "issue:12459", + "issue:12565", + "issue:13079", + "issue:1742", + "issue:2439", + "issue:2560", + "issue:2806", + "issue:3363", + "issue:3631", + "issue:3706", + "issue:378", + "issue:3984", + "issue:4241", + "issue:4368", + "issue:4551", + "issue:4665", + "issue:4724", + "issue:4733", + "issue:4849", + "issue:4975", + "issue:4985", + "issue:5474", + "issue:5529", + "issue:5575", + "issue:5606", + "issue:5633", + "issue:5719", + "issue:6230", + "issue:6520", + "issue:6561", + "issue:6879", + "issue:7238", + "issue:7239", + "issue:771", + "issue:8472", + "issue:8939", + "issue:8950", + "issue:9025", + "issue:9032" + ], + "soft_pairs": [ + "issue:4368|issue:8950", + "issue:12011|issue:13079", + "issue:6561|issue:7238", + "issue:378|issue:4849", + "issue:2806|issue:771", + "issue:4975|issue:5606", + "issue:10470|issue:10674", + "issue:5575|issue:8472", + "issue:10989|issue:12147", + "issue:1181|issue:4241", + "issue:3706|issue:5474", + "issue:6879|issue:7239", + "issue:9025|issue:9032", + "issue:4665|issue:5719", + "issue:4985|issue:6520", + "issue:3363|issue:4733", + "issue:4551|issue:8939", + "issue:12321|issue:12412", + "issue:2439|issue:3631", + "issue:5529|issue:5633", + "issue:5529|issue:6230", + "issue:1742|issue:2560", + "issue:3984|issue:4724", + "issue:12459|issue:12565" + ], + "prepared_review_unit_hash": "477a32ced2e2675fe82f4b5a2bcc4b1e189f56af64c727189ec225be33601f9e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33489, + "estimated_input_tokens": 8373, + "estimated_eval_tokens": 17002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated issues across Flux inference, Wan OOM/loading, Hunyuan support, SDXL training, LoRA loading, and assorted pipeline bugs. The soft pairs are mostly lexical overlaps rather than the same underlying defect.", + "confidence": 0.93, + "canonical_issue_reason": "No single issue cleanly represents the whole cluster; if a representative is required, issue 12459 is the most concrete and actionable bug report, but it does not cover the rest.", + "canonical_pr_reason": null, + "best_issue_reason": "issue 12459 is the strongest standalone candidate because it is a specific, reproducible performance bug with clear scope and current relevance, even though it is not a duplicate of the other items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4368", + "right": "issue:8950", + "accept": false, + "reason": "Both mention LoRA/SDXL OOM, but one is a training-memory problem and the other is an SD3 LoRA training bug; different model families and failure modes." + }, + { + "left": "issue:12011", + "right": "issue:13079", + "accept": false, + "reason": "Both are OOM reports, but Wan 2.2 i2v and FLUX.2-klein are different models and the repro conditions differ." + }, + { + "left": "issue:6561", + "right": "issue:7238", + "accept": false, + "reason": "One is an IP-Adapter checkpoint/file-path loading error; the other is an IP-Adapter masking feature request/bug. Same subsystem, not the same bug." + }, + { + "left": "issue:378", + "right": "issue:4849", + "accept": false, + "reason": "Textual inversion DiagonalGaussianDistribution is a specific runtime issue; the other is a general question about multi-GPU training." + }, + { + "left": "issue:2806", + "right": "issue:771", + "accept": false, + "reason": "Printing customized pipelines and BF16 dreambooth are unrelated problems." + }, + { + "left": "issue:4975", + "right": "issue:5606", + "accept": false, + "reason": "Both are LoRA loading/performance related, but one is about repeated datatype queries and the other about missing alpha metadata." + }, + { + "left": "issue:10470", + "right": "issue:10674", + "accept": false, + "reason": "Both are Flux inference failures, but torchao inference and GGUF loading are different backends/code paths." + }, + { + "left": "issue:5575", + "right": "issue:8472", + "accept": false, + "reason": "One asks about LoRA hidden size configuration; the other asks about simultaneous pre-training with other modules. Not the same defect." + }, + { + "left": "issue:10989", + "right": "issue:12147", + "accept": false, + "reason": "Both mention LoRA loading failing, but they concern different model families and different causes (quantized Transformers vs Wan2.2 Lightning loading)." + }, + { + "left": "issue:1181", + "right": "issue:4241", + "accept": false, + "reason": "Training from scratch vs fine-tuning LoRA are different usage questions, not duplicates." + }, + { + "left": "issue:3706", + "right": "issue:5474", + "accept": false, + "reason": "Wuerstchen new-model support and Stable Diffusion Reference Only proposal are separate model/pipeline feature requests." + }, + { + "left": "issue:6879", + "right": "issue:7239", + "accept": false, + "reason": "Custom Diffusion dtype mismatch and Dreambooth SDXL shape mismatch are different tensor-shape/dtype bugs." + }, + { + "left": "issue:9025", + "right": "issue:9032", + "accept": false, + "reason": "Both are FlaxStableDiffusionPipeline on TPU, but one is from_single_file usage in Kaggle and the other is a minimal example in Colab; not the same bug." + }, + { + "left": "issue:4665", + "right": "issue:5719", + "accept": false, + "reason": "Both involve LoRA/safetensors file selection on Windows, but one is a Windows alt-bin-file loading failure and the other is a specific preference order for LoRA weights." + }, + { + "left": "issue:4985", + "right": "issue:6520", + "accept": false, + "reason": "Both are training OOM-ish reports, but one is InstructPix2Pix SDXL training and the other is log_validation producing blank results in adapter training." + }, + { + "left": "issue:3363", + "right": "issue:4733", + "accept": false, + "reason": "Dreambooth LoRA failing on two machines and resume-from-checkpoint optimizer inf-check errors are different training bugs." + }, + { + "left": "issue:4551", + "right": "issue:8939", + "accept": false, + "reason": "SDXL dreambooth crash and SD3 tokenizer loading failure are unrelated model/setup issues." + }, + { + "left": "issue:12321", + "right": "issue:12412", + "accept": false, + "reason": "These are support requests for different HunyuanImage model versions, not the same issue." + }, + { + "left": "issue:2439", + "right": "issue:3631", + "accept": false, + "reason": "Inpainting pipeline loading/running problems and providing latents to inpainting are related but distinct pipeline bugs." + }, + { + "left": "issue:5529", + "right": "issue:5633", + "accept": false, + "reason": "train_text_to_image_sdxl.py bug and a transformer_temporal.py bug are different code paths." + }, + { + "left": "issue:5529", + "right": "issue:6230", + "accept": false, + "reason": "Both mention train_text_to_image_sdxl.py, but one is a generic bug report and the other is specifically CUDA OOM; not enough to treat as the same defect." + }, + { + "left": "issue:1742", + "right": "issue:2560", + "accept": false, + "reason": "Different generators/benchmarking speed behavior versus img2img output differences with a list of generators." + }, + { + "left": "issue:3984", + "right": "issue:4724", + "accept": false, + "reason": "Class-label validation error and SDXL img2img checkpoint behavior are unrelated." + }, + { + "left": "issue:12459", + "right": "issue:12565", + "accept": false, + "reason": "Flux.1-Kontext-dev throughput issue and Wan2.1 i2v running issue are separate model/runtime problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 95, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:1005", + "issue:10954", + "issue:11033", + "issue:11436", + "issue:11448", + "issue:11648", + "issue:1604", + "issue:2068", + "issue:2210", + "issue:2392", + "issue:2548", + "issue:3177", + "issue:3296", + "issue:3418", + "issue:3564", + "issue:3569", + "issue:3661", + "issue:3984", + "issue:4183", + "issue:4213", + "issue:4250", + "issue:4303", + "issue:4480", + "issue:4488", + "issue:4835", + "issue:5028", + "issue:5432", + "issue:5510", + "issue:5798", + "issue:6130", + "issue:6136", + "issue:6243", + "issue:6506", + "issue:6666", + "issue:7146", + "issue:7223", + "issue:7311", + "issue:7682", + "issue:8503", + "issue:861", + "issue:8649", + "issue:8790", + "issue:9051", + "issue:9092", + "issue:9443", + "issue:9613", + "issue:9873" + ], + "soft_pairs": [ + "issue:4480|issue:6136", + "issue:9443|issue:9613", + "issue:2392|issue:6506", + "issue:11436|issue:5510", + "issue:5028|issue:5798", + "issue:11033|issue:7682", + "issue:3661|issue:4213", + "issue:10010|issue:11448", + "issue:10954|issue:11648", + "issue:8503|issue:8649", + "issue:7146|issue:7311", + "issue:1604|issue:3984", + "issue:4183|issue:4835", + "issue:3564|issue:3569", + "issue:7223|issue:9873", + "issue:3177|issue:3296", + "issue:4303|issue:4488", + "issue:9051|issue:9092", + "issue:1005|issue:861", + "issue:2210|issue:3418", + "issue:6243|issue:8790", + "issue:2548|issue:6130", + "issue:5432|issue:6666", + "issue:2068|issue:4250" + ], + "prepared_review_unit_hash": "f325c5bea88e8ec1703764c88c428fbe1d0cb90b625de9056a9c208af96a3931", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34068, + "estimated_input_tokens": 8517, + "estimated_eval_tokens": 17290 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly heterogeneous: loading, dtype, pipeline support, training, and validation-image bugs. I only see a few true duplicates; most soft pairs are related by subsystem only, not the same underlying issue.", + "confidence": 0.91, + "canonical_issue_reason": "Issue:3564 is the clearest canonical among the true duplicates: it is the broader Uni-ControlNet feature request, and 3569 is the same request in a new-pipeline framing.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue:3564 is the best issue to anchor the duplicate subset, though the overall collection is too mixed to have one cluster-wide canonical bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4480", + "right": "issue:6136", + "accept": false, + "reason": "Both are Mac/cpu generation failures, but one is an MPS fp16 kernel crash and the other is SD2.1 NaNs on CPU." + }, + { + "left": "issue:9443", + "right": "issue:9613", + "accept": false, + "reason": "Different failure points: tokenizer instantiation for SD3 vs missing Timesteps.scale in 0.30.x." + }, + { + "left": "issue:2392", + "right": "issue:6506", + "accept": false, + "reason": "One is an img2img prompt_embeds validation error; the other is StableVideoDiffusion output_type formatting." + }, + { + "left": "issue:11436", + "right": "issue:5510", + "accept": false, + "reason": "Both are dtype mismatches, but they occur in different models and code paths with different tensors involved." + }, + { + "left": "issue:5028", + "right": "issue:5798", + "accept": false, + "reason": "Different unexpected-argument regressions in different pipelines: scale vs callback_on_step_end." + }, + { + "left": "issue:11033", + "right": "issue:7682", + "accept": false, + "reason": "One is from_single_file SD1.5 loading; the other is a multi-GPU training/runtime problem." + }, + { + "left": "issue:3661", + "right": "issue:4213", + "accept": false, + "reason": "Both involve model conversion/loading, but one is SD2.x from_ckpt and the other is SDXL from_single_file without accelerate." + }, + { + "left": "issue:10010", + "right": "issue:11448", + "accept": false, + "reason": "Both are missing-model errors, but for different pipelines and different absent artifacts/URLs." + }, + { + "left": "issue:10954", + "right": "issue:11648", + "accept": false, + "reason": "Similar LoRA wording, but one is a loading failure and the other is a general fp8 transformer usage question." + }, + { + "left": "issue:8503", + "right": "issue:8649", + "accept": false, + "reason": "Same SD3 area, but one crashes with gradient checkpointing while the other mishandles num_images_per_prompt." + }, + { + "left": "issue:7146", + "right": "issue:7311", + "accept": false, + "reason": "Different training bugs: CUDA OOM during SDXL finetuning versus checkpoint save failure." + }, + { + "left": "issue:1604", + "right": "issue:3984", + "accept": false, + "reason": "A diffusers version regression in one case, versus a class_labels requirement tied to num_class_embeds in the other." + }, + { + "left": "issue:4183", + "right": "issue:4835", + "accept": false, + "reason": "Both ControlNet-related, but one is guidance_scale <= 1 and the other is MultiControlNetModel in SDXL img2img." + }, + { + "left": "issue:3564", + "right": "issue:3569", + "accept": true, + "reason": "Same underlying request: Uni-ControlNet support/new pipeline for the same model family." + }, + { + "left": "issue:7223", + "right": "issue:9873", + "accept": false, + "reason": "Both are new model pipeline requests, but for entirely different models: Latte video generation vs OmniGen image generation." + }, + { + "left": "issue:3177", + "right": "issue:3296", + "accept": false, + "reason": "Same script, but different bugs: DeepSpeed stage-3 offloading versus checkpoint save/load with text_encoder." + }, + { + "left": "issue:4303", + "right": "issue:4488", + "accept": false, + "reason": "Both SDXL loading issues, but one is a Flax pipeline crash and the other is a conversion utility needing pipeline_class." + }, + { + "left": "issue:9051", + "right": "issue:9092", + "accept": false, + "reason": "One is batch inference support in FluxPipeline; the other is a model_index.json missing-file error." + }, + { + "left": "issue:1005", + "right": "issue:861", + "accept": false, + "reason": "Completely different topics: timestep bounds versus checkpoint-to-diffusers conversion." + }, + { + "left": "issue:2210", + "right": "issue:3418", + "accept": true, + "reason": "Same feature request: let LoRA training include the text encoder in the training script." + }, + { + "left": "issue:6243", + "right": "issue:8790", + "accept": false, + "reason": "Both mention IP-Adapter, but one is Face ID support and the other is SDXL TensorRT integration." + }, + { + "left": "issue:2548", + "right": "issue:6130", + "accept": false, + "reason": "Training-time LoRA save failure versus inference-time LoRA loading error; same area, different bugs." + }, + { + "left": "issue:5432", + "right": "issue:6666", + "accept": true, + "reason": "Same validation-image bug: the generated validation images are blank/black during training." + }, + { + "left": "issue:2068", + "right": "issue:4250", + "accept": false, + "reason": "Both are model-loading complaints, but 2068 is a dtype-conversion issue while 4250 is an unspecified incorrect-load bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall and appropriately conservative. The summary reflects the packet\u2019s mix, and the accepted true-duplicate pairs look supportable from the issue titles/descriptions: 3564/3569, 2210/3418, and 5432/6666. The rejected soft pairs are generally scoped narrowly enough to avoid overstating duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 96, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10139", + "issue:10150", + "issue:10223", + "issue:10453", + "issue:10514", + "issue:11223", + "issue:11657", + "issue:12009", + "issue:12011", + "issue:12037", + "issue:12108", + "issue:12147", + "issue:12738", + "issue:1969", + "issue:2163", + "issue:222", + "issue:3235", + "issue:3264", + "issue:3591", + "issue:3661", + "issue:3706", + "issue:3826", + "issue:3885", + "issue:3933", + "issue:4125", + "issue:4305", + "issue:4399", + "issue:4418", + "issue:4724", + "issue:4954", + "issue:5128", + "issue:5544", + "issue:5617", + "issue:579", + "issue:6183", + "issue:6368", + "issue:6686", + "issue:672", + "issue:6967", + "issue:7138", + "issue:7421", + "issue:8609", + "issue:8649", + "issue:8817", + "issue:9200", + "issue:9527", + "issue:9576" + ], + "soft_pairs": [ + "issue:4399|issue:4418", + "issue:9200|issue:9527", + "issue:1969|issue:222", + "issue:6967|issue:7138", + "issue:4724|issue:7421", + "issue:10150|issue:10223", + "issue:3933|issue:4305", + "issue:11223|issue:12011", + "issue:4724|issue:6368", + "issue:5617|issue:6686", + "issue:4954|issue:579", + "issue:3235|issue:3661", + "issue:8609|issue:8817", + "issue:12009|issue:12147", + "issue:2163|issue:3885", + "issue:10453|issue:12738", + "issue:5544|issue:8649", + "issue:10139|issue:10514", + "issue:11657|issue:12037", + "issue:5128|issue:6183", + "issue:3264|issue:672", + "issue:3706|issue:3826", + "issue:3591|issue:4125", + "issue:12108|issue:9576" + ], + "prepared_review_unit_hash": "ef9347a19b1e0c4006da7f67b15f681e9323fbee0cf2d9a51e7f847da7abfa8d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33484, + "estimated_input_tokens": 8371, + "estimated_eval_tokens": 16998 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose set of issue titles that only share broad themes like pipeline initialization, batch-size handling, dtype mismatches, loading, and OOMs. The only strong soft match is the sample_size-related pair.", + "confidence": 0.66, + "canonical_issue_reason": "issue:10139 is the most root-cause-like report: it directly says sample_size is not fully considered in pipeline init, which is a broader upstream bug description than the more specific downstream symptom in 10514.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10139 is the best global issue candidate because it is the broadest and most general description of the underlying sample_size/init problem, making it the most reusable canonical reference.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4399", + "right": "issue:4418", + "accept": false, + "reason": "Different failures: torch.compile/cpu offload compatibility vs SDXL inpaint denoising_start behavior." + }, + { + "left": "issue:9200", + "right": "issue:9527", + "accept": false, + "reason": "Both mention dtype errors, but the titles point to different pipeline/version contexts and do not clearly describe the same concrete bug." + }, + { + "left": "issue:1969", + "right": "issue:222", + "accept": false, + "reason": "Too generic; one is a specific tensor view/stride error, the other is a vague training RuntimeError." + }, + { + "left": "issue:6967", + "right": "issue:7138", + "accept": false, + "reason": "Related to SDXL shape/model-structure issues, but one is a training script mismatch and the other is a model structure discrepancy; not enough evidence of one underlying bug." + }, + { + "left": "issue:4724", + "right": "issue:7421", + "accept": false, + "reason": "Different problems: SDXL img2img checkpoint behavior vs seed handling in a KDiffusion pipeline." + }, + { + "left": "issue:10150", + "right": "issue:10223", + "accept": false, + "reason": "Unrelated: slow loading/inference vs a question about obtaining a LoRA identifier." + }, + { + "left": "issue:3933", + "right": "issue:4305", + "accept": false, + "reason": "Both involve graph breaks, but one is tied to logging.info and the other to a version-level torch.compile regression; not the same concrete fix." + }, + { + "left": "issue:11223", + "right": "issue:12011", + "accept": false, + "reason": "Both are WAN OOM reports, but one is specifically layer-wise upcasting with LoRAs and the other is a general i2v OOM; likely different causes." + }, + { + "left": "issue:4724", + "right": "issue:6368", + "accept": false, + "reason": "Different issues: SDXL img2img expected behavior vs unsupported LoRA alpha checkpoint loading." + }, + { + "left": "issue:5617", + "right": "issue:6686", + "accept": false, + "reason": "Both concern compiled models and save/load, but one is a loading failure and the other is lack of post-save performance gains." + }, + { + "left": "issue:4954", + "right": "issue:579", + "accept": false, + "reason": "Both are Textual Inversion load errors, but one is an unpickling error and the other is a missing config.json; different failure modes." + }, + { + "left": "issue:3235", + "right": "issue:3661", + "accept": false, + "reason": "Both are checkpoint-loading problems, but they involve different model families and distinct incompatibilities." + }, + { + "left": "issue:8609", + "right": "issue:8817", + "accept": false, + "reason": "Different IP-Adapter/custom pipeline issues without clear evidence of the same code-path bug." + }, + { + "left": "issue:12009", + "right": "issue:12147", + "accept": false, + "reason": "Both are Wan-related loading issues, but one is GGUF single-file loading and the other is Lightning LoRA loading; not the same bug." + }, + { + "left": "issue:2163", + "right": "issue:3885", + "accept": false, + "reason": "Unrelated: dtype mismatch in training vs a general VRAM-usage question." + }, + { + "left": "issue:10453", + "right": "issue:12738", + "accept": false, + "reason": "Both fail above batch size 1, but they concern different pipelines/models and do not clearly share the same underlying implementation bug." + }, + { + "left": "issue:5544", + "right": "issue:8649", + "accept": false, + "reason": "Different features and failures: FreeU disable error vs num_images_per_prompt handling in SD3." + }, + { + "left": "issue:10139", + "right": "issue:10514", + "accept": true, + "reason": "Both center on sample_size being mishandled in pipeline setup, with 10514 reading like a concrete manifestation of the broader init bug described in 10139." + }, + { + "left": "issue:11657", + "right": "issue:12037", + "accept": false, + "reason": "Both are Wan LoRA-related, but one is a missing-prefix warning and the other is a compatibility issue with a distilled LoRA type." + }, + { + "left": "issue:5128", + "right": "issue:6183", + "accept": false, + "reason": "Both are single-file loading issues, but one is a local safetensors loading problem and the other is an incorrect-shape problem." + }, + { + "left": "issue:3264", + "right": "issue:672", + "accept": false, + "reason": "Both involve conversion/loading of model weights, but they address different directions and different conversion paths." + }, + { + "left": "issue:3706", + "right": "issue:3826", + "accept": false, + "reason": "Not duplicates: one is a model introduction request, the other is a new pipeline request." + }, + { + "left": "issue:3591", + "right": "issue:4125", + "accept": false, + "reason": "Unrelated training/diagnostics questions about loss curves; no same bug or change." + }, + { + "left": "issue:12108", + "right": "issue:9576", + "accept": false, + "reason": "Different problems: flow-matching scheduler breakage vs a pipeline loading parameter propagation bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analyst\u2019s summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:10139 with issue:10514, is plausibly the strongest match based on the shared sample_size/init theme, and the other pairwise rejections avoid overstating weak title-level similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 97, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10040", + "issue:10311", + "issue:10475", + "issue:11006", + "issue:11135", + "issue:1161", + "issue:11779", + "issue:12108", + "issue:12175", + "issue:1524", + "issue:1572", + "issue:1604", + "issue:1616", + "issue:1704", + "issue:1823", + "issue:1866", + "issue:2161", + "issue:2185", + "issue:2774", + "issue:3539", + "issue:3698", + "issue:3728", + "issue:3761", + "issue:3788", + "issue:4186", + "issue:5005", + "issue:5128", + "issue:5226", + "issue:5337", + "issue:5753", + "issue:6141", + "issue:6354", + "issue:6570", + "issue:6823", + "issue:6866", + "issue:6935", + "issue:7021", + "issue:7350", + "issue:7982", + "issue:8296", + "issue:8414", + "issue:8588", + "issue:8708", + "issue:8826", + "issue:8939", + "issue:9044", + "issue:9619" + ], + "soft_pairs": [ + "issue:11135|issue:3728", + "issue:1161|issue:3698", + "issue:1524|issue:1604", + "issue:5226|issue:5753", + "issue:6866|issue:6935", + "issue:8588|issue:8708", + "issue:8939|issue:9044", + "issue:5005|issue:8296", + "issue:2774|issue:5128", + "issue:1616|issue:3761", + "issue:6570|issue:8826", + "issue:6141|issue:6354", + "issue:6141|issue:9619", + "issue:1572|issue:1866", + "issue:4186|issue:7982", + "issue:6823|issue:7350", + "issue:10040|issue:10475", + "issue:12108|issue:12175", + "issue:10311|issue:7021", + "issue:1704|issue:3539", + "issue:1823|issue:5337", + "issue:11006|issue:11779", + "issue:2161|issue:2185", + "issue:3788|issue:8414" + ], + "prepared_review_unit_hash": "29b2261832d1ad5a0a9a3511acf25b6dfcf9c651cc5738f0e4dcea4c88972caa", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33477, + "estimated_input_tokens": 8370, + "estimated_eval_tokens": 16996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very loose and mostly contains unrelated issues that only share superficial vocabulary around diffusion pipelines, training, or image generation. None of the soft-edge pairs look like the same underlying bug/change, so I would not merge any of them as duplicates.", + "confidence": 0.24, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a strong canonical candidate: the set spans distinct bugs and feature requests across SD3, Qwen Image, ControlNet, loaders, training scripts, and unrelated pipeline behavior.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11135", + "right": "issue:3728", + "accept": false, + "reason": "Both are new-pipeline ideas, but they describe different models and proposals (SmoothCache vs Shifted Diffusion), not the same change." + }, + { + "left": "issue:1161", + "right": "issue:3698", + "accept": false, + "reason": "Both are conversion-related, but one is Flax-to-PyTorch and the other is inpainting-model conversion; different workflows and targets." + }, + { + "left": "issue:1524", + "right": "issue:1604", + "accept": false, + "reason": "One is poor training quality, the other is a version-specific script breakage; not the same bug." + }, + { + "left": "issue:5226", + "right": "issue:5753", + "accept": false, + "reason": "Both mention LoRA with SDXL variants, but one is AdapterPipeline-specific and the other is ControlNet-specific; different code paths." + }, + { + "left": "issue:6866", + "right": "issue:6935", + "accept": false, + "reason": "Both are IP-Adapter requests, but one is ControlNet inpaint support and the other is Face ID; distinct functionality." + }, + { + "left": "issue:8588", + "right": "issue:8708", + "accept": false, + "reason": "Both involve SD3 example/script gaps, but one is a missing conversion script and the other is a missing VAE config field in dreambooth examples." + }, + { + "left": "issue:8939", + "right": "issue:9044", + "accept": false, + "reason": "Both are model-loading OSErrors, but they reference different models and missing files, so they are not the same issue." + }, + { + "left": "issue:5005", + "right": "issue:8296", + "accept": false, + "reason": "Both concern training resume/global step behavior, but one is a progress-bar truncation and the other is inconsistent step accounting across scripts." + }, + { + "left": "issue:2774", + "right": "issue:5128", + "accept": false, + "reason": "Both are file-conversion/load errors, but they affect different converters and model formats." + }, + { + "left": "issue:1616", + "right": "issue:3761", + "accept": false, + "reason": "Both are community pipeline loading issues, but one is LPW pipeline version compatibility and the other is loading multiple pipelines in one process." + }, + { + "left": "issue:6570", + "right": "issue:8826", + "accept": false, + "reason": "Both are runtime tensor-shape errors, but they occur in different pipelines and have different failure modes." + }, + { + "left": "issue:6141", + "right": "issue:6354", + "accept": false, + "reason": "One is a request for ControlNet parameters, the other is LoRA support for ControlNet; related area but not the same change." + }, + { + "left": "issue:6141", + "right": "issue:9619", + "accept": false, + "reason": "Different ControlNet feature requests: parameter support versus interrupt support." + }, + { + "left": "issue:1572", + "right": "issue:1866", + "accept": false, + "reason": "Both involve Stable Diffusion img2img/inpainting, but one is a broken pipeline and the other is scheduler-strength incompatibility." + }, + { + "left": "issue:4186", + "right": "issue:7982", + "accept": false, + "reason": "Both touch SDXL/ControlNet-style pipelines, but one is an inpaint/refiner runtime issue and the other is a from_pipe/MultiControlNetModel construction issue." + }, + { + "left": "issue:6823", + "right": "issue:7350", + "accept": false, + "reason": "Both mention ControlNet, but one is a general inability to use it and the other is initializing ControlNet from a LoRA-modified UNet." + }, + { + "left": "issue:10040", + "right": "issue:10475", + "accept": false, + "reason": "Both report low-quality/gray outputs, but one is Stable Diffusion 3 gray images and the other is a validation-vs-inference quality mismatch during fine-tuning." + }, + { + "left": "issue:12108", + "right": "issue:12175", + "accept": false, + "reason": "Both are Qwen Image-related, but one is a scheduler/flow-matching breakage and the other is generic bad image quality." + }, + { + "left": "issue:10311", + "right": "issue:7021", + "accept": false, + "reason": "Both are import/load errors, but they refer to different environment variables and different failure points." + }, + { + "left": "issue:1704", + "right": "issue:3539", + "accept": false, + "reason": "One is a question about an attention block, the other about main_process_first; not duplicates." + }, + { + "left": "issue:1823", + "right": "issue:5337", + "accept": false, + "reason": "Different topics entirely: swapped comments in depth2img versus a general question about callback behavior." + }, + { + "left": "issue:11006", + "right": "issue:11779", + "accept": false, + "reason": "Both involve Wan pipelines, but one is broken video output with quantized transformer and the other is incorrect mask padding in VACE; distinct bugs." + }, + { + "left": "issue:2161", + "right": "issue:2185", + "accept": false, + "reason": "One is LoRA training on inpainting, the other missing config files for naive text-to-image training; different issues." + }, + { + "left": "issue:3788", + "right": "issue:8414", + "accept": false, + "reason": "Both are new model/pipeline requests, but they target different models and modalities (visual generation vs text-to-audio)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 98, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10512", + "issue:10722", + "issue:10782", + "issue:10863", + "issue:11136", + "issue:11182", + "issue:11221", + "issue:11374", + "issue:12126", + "issue:13053", + "issue:1475", + "issue:2250", + "issue:2437", + "issue:2856", + "issue:3250", + "issue:3454", + "issue:3672", + "issue:3891", + "issue:3910", + "issue:4061", + "issue:4185", + "issue:4250", + "issue:4324", + "issue:4446", + "issue:4808", + "issue:4888", + "issue:5455", + "issue:579", + "issue:5821", + "issue:6371", + "issue:6548", + "issue:6666", + "issue:6689", + "issue:7185", + "issue:7420", + "issue:7708", + "issue:8787", + "issue:8894", + "issue:8939", + "issue:9108", + "issue:9112", + "issue:9187", + "issue:9343", + "issue:9362", + "issue:9415", + "issue:9667", + "issue:9971" + ], + "soft_pairs": [ + "issue:3910|issue:4061", + "issue:9112|issue:9362", + "issue:9187|issue:9971", + "issue:4808|issue:6548", + "issue:8787|issue:9667", + "issue:3454|issue:4324", + "issue:1475|issue:579", + "issue:10782|issue:7185", + "issue:2437|issue:3891", + "issue:2250|issue:3672", + "issue:11136|issue:12126", + "issue:5455|issue:6371", + "issue:10722|issue:11221", + "issue:8894|issue:9415", + "issue:4446|issue:7420", + "issue:11374|issue:13053", + "issue:7708|issue:8939", + "issue:3250|issue:5821", + "issue:4250|issue:4888", + "issue:10863|issue:9667", + "issue:11182|issue:9343", + "issue:10512|issue:9108", + "issue:4185|issue:6666", + "issue:2856|issue:6689" + ], + "prepared_review_unit_hash": "3806a7e108c3b18ba066b289948c610da8c0a8566e05a1ebd745da3df21c62dd", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33445, + "estimated_input_tokens": 8362, + "estimated_eval_tokens": 16980 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicate cluster emerges here: the soft-linked pairs are mostly superficial title matches across different models, pipelines, training scripts, or error types.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3910", + "right": "issue:4061", + "accept": false, + "reason": "Both mention LoRA loading, but one is generic and the other is SDXL-specific; likely different model/setup failures." + }, + { + "left": "issue:9112", + "right": "issue:9362", + "accept": false, + "reason": "Different error types and model paths: tensor-size mismatch in Auraflow vs an index-out-of-bounds issue." + }, + { + "left": "issue:9187", + "right": "issue:9971", + "accept": false, + "reason": "Unrelated subsystems: IP-Adapter/SDXL image encoder handling vs FLUX scheduler behavior." + }, + { + "left": "issue:4808", + "right": "issue:6548", + "accept": false, + "reason": "Both involve img2img-like code, but one is a SHAP-E integration test and the other is an RGBA latent-preprocessing bug." + }, + { + "left": "issue:8787", + "right": "issue:9667", + "accept": false, + "reason": "One is a performance regression in loading speed; the other is a concrete FLUX fp8 loading error." + }, + { + "left": "issue:3454", + "right": "issue:4324", + "accept": false, + "reason": "Different failures: a training script regression vs a NoneType error in UNet forward." + }, + { + "left": "issue:1475", + "right": "issue:579", + "accept": false, + "reason": "DreamBooth/textual-inversion training setup issue vs missing config.json for textual inversion models." + }, + { + "left": "issue:10782", + "right": "issue:7185", + "accept": false, + "reason": "A Lumina Image 2.0 bug and a general PyTorch version compatibility question are not the same issue." + }, + { + "left": "issue:2437", + "right": "issue:3891", + "accept": false, + "reason": "Both are feature requests for pipelines, but they target completely different pipelines and changes." + }, + { + "left": "issue:2250", + "right": "issue:3672", + "accept": false, + "reason": "DreamBooth prompt-generation failure and StableDiffusionPipeline thread-safety failure are different code-path bugs." + }, + { + "left": "issue:11136", + "right": "issue:12126", + "accept": false, + "reason": "Different pipelines and root causes: SCM timestep restriction vs a cache_context attribute error in wan2.2." + }, + { + "left": "issue:5455", + "right": "issue:6371", + "accept": false, + "reason": "Separate example/notebook failures in different training examples with unrelated assertions/process errors." + }, + { + "left": "issue:10722", + "right": "issue:11221", + "accept": false, + "reason": "Tensor-shape mismatch during DreamBooth prior preservation is unrelated to autoencoderkl non-convergence." + }, + { + "left": "issue:8894", + "right": "issue:9415", + "accept": false, + "reason": "A training-script runtime error vs a request to add a new parameter are not the same bug." + }, + { + "left": "issue:4446", + "right": "issue:7420", + "accept": false, + "reason": "Custom text_encoder loading and torch.compile support are different feature areas with different failures." + }, + { + "left": "issue:11374", + "right": "issue:13053", + "accept": false, + "reason": "Both are feature requests, but for different model pipelines and no shared concrete change." + }, + { + "left": "issue:7708", + "right": "issue:8939", + "accept": false, + "reason": "Both are tokenizer-loading problems, but they involve different models and likely different missing-file causes." + }, + { + "left": "issue:3250", + "right": "issue:5821", + "accept": false, + "reason": "A pipeline runtime failure and a CUDA backward/training issue are not the same underlying bug." + }, + { + "left": "issue:4250", + "right": "issue:4888", + "accept": false, + "reason": "Model conversion failure and offline safetensors loading are distinct workflows." + }, + { + "left": "issue:10863", + "right": "issue:9667", + "accept": false, + "reason": "Both involve local model loading, but one is a safetensors-file load error and the other is FLUX fp8 transformer loading." + }, + { + "left": "issue:11182", + "right": "issue:9343", + "accept": false, + "reason": "One reports an unnecessary download for BF16 variants; the other is a FLUX mismatched-size loading error." + }, + { + "left": "issue:10512", + "right": "issue:9108", + "accept": false, + "reason": "Quantization is shared, but one is LoRA load failure and the other is resuming training with a quantized base model." + }, + { + "left": "issue:4185", + "right": "issue:6666", + "accept": false, + "reason": "Both mention black images, but one is SDXL ControlNet inference and the other is validation imagery during training." + }, + { + "left": "issue:2856", + "right": "issue:6689", + "accept": false, + "reason": "Both are model-to-diffusers conversion complaints, but the titles point to different models and no shared concrete bug is evident." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 99, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10150", + "issue:10636", + "issue:10760", + "issue:1083", + "issue:10937", + "issue:11046", + "issue:1109", + "issue:11418", + "issue:11584", + "issue:11661", + "issue:12078", + "issue:12459", + "issue:12505", + "issue:12728", + "issue:1909", + "issue:2068", + "issue:2091", + "issue:2163", + "issue:2321", + "issue:2560", + "issue:2984", + "issue:3348", + "issue:3451", + "issue:3944", + "issue:4307", + "issue:4321", + "issue:4727", + "issue:4974", + "issue:5160", + "issue:5442", + "issue:5466", + "issue:5504", + "issue:5693", + "issue:5969", + "issue:6136", + "issue:6176", + "issue:6518", + "issue:6832", + "issue:7049", + "issue:7418", + "issue:7516", + "issue:7770", + "issue:834", + "issue:8518", + "issue:861", + "issue:9546", + "issue:9927" + ], + "soft_pairs": [ + "issue:4307|issue:4727", + "issue:10760|issue:7516", + "issue:1109|issue:834", + "issue:10150|issue:12505", + "issue:2984|issue:5442", + "issue:3348|issue:5504", + "issue:10636|issue:7418", + "issue:5693|issue:6176", + "issue:6832|issue:7049", + "issue:1083|issue:4321", + "issue:5466|issue:7770", + "issue:12728|issue:8518", + "issue:2091|issue:3451", + "issue:9546|issue:9927", + "issue:10937|issue:11584", + "issue:11661|issue:12078", + "issue:11046|issue:12459", + "issue:10010|issue:11418", + "issue:3944|issue:5160", + "issue:5969|issue:6136", + "issue:2163|issue:6518", + "issue:4974|issue:861", + "issue:2068|issue:2321", + "issue:1909|issue:2560" + ], + "prepared_review_unit_hash": "b01a53c0c58d58991fa25015767a4927dcf4fe3238b659e11c88f3332bd6c5dd", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34101, + "estimated_input_tokens": 8526, + "estimated_eval_tokens": 17308 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a mixed bag of unrelated issue reports; only a few soft duplicate mini-clusters stand out: offline/local_files_only loading, CPU-core underutilization, and list-of-generators handling.", + "confidence": 0.64, + "canonical_issue_reason": "No single canonical issue fits the whole set because the reports span many different bugs; the closest umbrella duplicate cluster is the offline/local_files_only loading family, with the broader representative being the auto-pipeline breakage report.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 4727 is the strongest representative of the most general, user-facing duplicate family here: offline/local_files_only loading failures affecting pipeline construction, not just one narrow from_pretrained call.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4307", + "right": "issue:4727", + "accept": true, + "reason": "Both report the same underlying offline/local_files_only model-loading failure; 4727 is the broader auto-pipeline manifestation of the same bug." + }, + { + "left": "issue:10760", + "right": "issue:7516", + "accept": false, + "reason": "Both are import/attribute errors, but they concern different missing pipeline symbols and do not look like the same concrete bug." + }, + { + "left": "issue:1109", + "right": "issue:834", + "accept": true, + "reason": "Both describe the same CPU underutilization problem; the titles differ, but the underlying complaint is the same performance issue." + }, + { + "left": "issue:10150", + "right": "issue:12505", + "accept": false, + "reason": "Both mention slowness/throughput, but they target different pipelines and different failure modes." + }, + { + "left": "issue:2984", + "right": "issue:5442", + "accept": false, + "reason": "These are different video-generation pipeline topics; no evidence they are the same bug or change." + }, + { + "left": "issue:3348", + "right": "issue:5504", + "accept": false, + "reason": "Both involve LoRA/xformers, but one is about the scale parameter and the other about PEFT LoRA loading; too broad to treat as the same bug." + }, + { + "left": "issue:10636", + "right": "issue:7418", + "accept": false, + "reason": "Different files and different reported failures; only superficial similarity in being code bugs." + }, + { + "left": "issue:5693", + "right": "issue:6176", + "accept": false, + "reason": "Both are loading-related errors, but one is a VAE load failure and the other is a LoRA post-training load issue." + }, + { + "left": "issue:6832", + "right": "issue:7049", + "accept": false, + "reason": "Both are IP-Adapter issues, but they are distinct shape/unpack errors in different examples." + }, + { + "left": "issue:1083", + "right": "issue:4321", + "accept": false, + "reason": "Both concern FP16/ONNX output problems, but the reported symptoms and environments differ materially." + }, + { + "left": "issue:5466", + "right": "issue:7770", + "accept": false, + "reason": "Both mention LoRA shape mismatches, but one is SDXL img2img LoRA loading and the other is a broader model+LoRA load mismatch." + }, + { + "left": "issue:12728", + "right": "issue:8518", + "accept": false, + "reason": "Both involve torch.compile, but they are different models and different Dynamo failure modes." + }, + { + "left": "issue:2091", + "right": "issue:3451", + "accept": false, + "reason": "These are unrelated training issues: checkpoint resume/multi-GPU versus fp16 memory use." + }, + { + "left": "issue:9546", + "right": "issue:9927", + "accept": false, + "reason": "Both are training-time problems, but one is validation OOM and the other is a header-size error; not the same bug." + }, + { + "left": "issue:10937", + "right": "issue:11584", + "accept": false, + "reason": "Both mention torch.compile, but one is a VAE encode error and the other is a group-offloading compatibility problem." + }, + { + "left": "issue:11661", + "right": "issue:12078", + "accept": false, + "reason": "Both are Flux-control training example issues, but one is a None handling bug and the other is validation-input content." + }, + { + "left": "issue:11046", + "right": "issue:12459", + "accept": false, + "reason": "Both are Flux inference issues, but one is a functional controlnet/inpainting/IP-adapter setup and the other is a batch throughput complaint." + }, + { + "left": "issue:10010", + "right": "issue:11418", + "accept": false, + "reason": "Both are about FLUX Fill assets, but one is a missing model_index.json fetch and the other is adding an fp8 safetensor file." + }, + { + "left": "issue:3944", + "right": "issue:5160", + "accept": false, + "reason": "Different Dreambooth/LoRA feature requests; same broad area, but not the same bug." + }, + { + "left": "issue:5969", + "right": "issue:6136", + "accept": false, + "reason": "Both are NaN reports, but the patch-specific DDIM issue and the Mac CPU SD2.1 issue have different triggers and code paths." + }, + { + "left": "issue:2163", + "right": "issue:6518", + "accept": false, + "reason": "Both are training-script errors, but they involve different root causes: dtype mismatch versus NoneType floor-division." + }, + { + "left": "issue:4974", + "right": "issue:861", + "accept": false, + "reason": "Both are model conversion topics, but one is safetensor-to-diffusers and the other is ckpt-to-diffusers; not the same bug." + }, + { + "left": "issue:2068", + "right": "issue:2321", + "accept": false, + "reason": "These are both dtype/model-loading discussions, but they report different problems and code paths." + }, + { + "left": "issue:1909", + "right": "issue:2560", + "accept": true, + "reason": "Both point to the same generator-list handling bug in pipelines/schedulers, where passing a list of generators produces incorrect behavior." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The clustering is mostly well-grounded in the packet and the soft-edge accepts are conservative. The accepted pairs (4307/4727, 1109/834, 1909/2560) are credible duplicate-family matches, while the rejections avoid overclaiming on only superficially similar reports. The summary stays appropriately narrow and does not overstate a single umbrella bug across unrelated issues." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 100, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10057", + "issue:10191", + "issue:10782", + "issue:1083", + "issue:11024", + "issue:11247", + "issue:11657", + "issue:12459", + "issue:12666", + "issue:12738", + "issue:13547", + "issue:1363", + "issue:1524", + "issue:1539", + "issue:1543", + "issue:2081", + "issue:2083", + "issue:2321", + "issue:2371", + "issue:2991", + "issue:3101", + "issue:3184", + "issue:3235", + "issue:3564", + "issue:3799", + "issue:4402", + "issue:4724", + "issue:4790", + "issue:5168", + "issue:5224", + "issue:5529", + "issue:5621", + "issue:5737", + "issue:5917", + "issue:6258", + "issue:6283", + "issue:6791", + "issue:7168", + "issue:7286", + "issue:7400", + "issue:817", + "issue:8534", + "issue:8557", + "issue:9223", + "issue:9637", + "issue:9717", + "issue:9856", + "issue:9911" + ], + "soft_pairs": [ + "issue:3564|issue:5168", + "issue:5917|issue:9223", + "issue:8534|issue:8557", + "issue:2083|issue:3184", + "issue:3235|issue:7286", + "issue:4790|issue:5737", + "issue:10057|issue:11024", + "issue:12459|issue:12738", + "issue:2371|issue:817", + "issue:1524|issue:1543", + "issue:1539|issue:3799", + "issue:2321|issue:5621", + "issue:2081|issue:5224", + "issue:4402|issue:7168", + "issue:10191|issue:9856", + "issue:2991|issue:3101", + "issue:6258|issue:7400", + "issue:5529|issue:6791", + "issue:9717|issue:9911", + "issue:10782|issue:12666", + "issue:1083|issue:1363", + "issue:11247|issue:9637", + "issue:4724|issue:6283", + "issue:11657|issue:13547" + ], + "prepared_review_unit_hash": "8e5f58cb0ab920866997484e92a98e2ecffd3883d449382ce20f58839b8a7a93", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34245, + "estimated_input_tokens": 8562, + "estimated_eval_tokens": 17380 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is broadly heterogeneous; the soft pairs mostly share subsystem vocabulary but not the same concrete bug or change. I would reject all soft edges and avoid selecting a single canonical issue/PR for the whole set.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No clear best issue: the items span unrelated topics (Flux, SDXL, ControlNet, training scripts, conversion, LoRA, video pipelines), so there is no single representative bug report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3564", + "right": "issue:5168", + "accept": false, + "reason": "Both are ControlNet feature requests, but Uni-ControlNet and ControlNet-XS are different variants and likely separate changes." + }, + { + "left": "issue:5917", + "right": "issue:9223", + "accept": false, + "reason": "One is a checkpoint conversion failure; the other is a dependency/version conflict about pyarrow. Different problems." + }, + { + "left": "issue:8534", + "right": "issue:8557", + "accept": false, + "reason": "Both concern SD3 training, but one is logit-normal weighting and the other is token-limit behavior. Not the same bug." + }, + { + "left": "issue:2083", + "right": "issue:3184", + "accept": false, + "reason": "Both mention generators, but one is DDIMScheduler eta behavior and the other is a community pipeline generator-list issue. Different code paths." + }, + { + "left": "issue:3235", + "right": "issue:7286", + "accept": false, + "reason": "Both are about Custom Diffusion weight loading, but the reported failures are different enough to look like separate loader issues." + }, + { + "left": "issue:4790", + "right": "issue:5737", + "accept": false, + "reason": "Both involve conversion scripts, but they are opposite directions and different model contexts; not one underlying bug." + }, + { + "left": "issue:10057", + "right": "issue:11024", + "accept": false, + "reason": "Flux single-file token handling and SiglipImageProcessor .to() are unrelated issues." + }, + { + "left": "issue:12459", + "right": "issue:12738", + "accept": false, + "reason": "Both mention batching, but one is a throughput concern and the other is an assertion failure in ZImagePipeline." + }, + { + "left": "issue:2371", + "right": "issue:817", + "accept": false, + "reason": "Both are Dreambooth-related, but one is missing trained objects and the other is mixed-precision training. Different failures." + }, + { + "left": "issue:1524", + "right": "issue:1543", + "accept": false, + "reason": "Same script family, but poor training quality and fp16 revision handling are not the same defect." + }, + { + "left": "issue:1539", + "right": "issue:3799", + "accept": false, + "reason": "These are both type-related, but one is a conversion-script type bug and the other is an inference typing issue. Not duplicates." + }, + { + "left": "issue:2321", + "right": "issue:5621", + "accept": false, + "reason": "Model-copy/EMA behavior is unrelated to SDXL ControlNet encoder_hid_proj support." + }, + { + "left": "issue:2081", + "right": "issue:5224", + "accept": false, + "reason": "Both involve JAX/Flax diffusion, but one is conversion from PyTorch weights and the other is a Flax SD 2.1 runtime error." + }, + { + "left": "issue:4402", + "right": "issue:7168", + "accept": false, + "reason": "Both concern SDXL/IP-adapter style usage, but one is from_single_file inpaint and the other is image-embed tensor shape handling." + }, + { + "left": "issue:10191", + "right": "issue:9856", + "accept": false, + "reason": "GPU selection for training and distributed port conflicts are unrelated." + }, + { + "left": "issue:2991", + "right": "issue:3101", + "accept": false, + "reason": "Both are ControlNet training example errors, but one is explicitly mixed-precision and the other is a generic failure; insufficient evidence of the same bug." + }, + { + "left": "issue:6258", + "right": "issue:7400", + "accept": false, + "reason": "Both are Stable Video Diffusion issues, but forward chunking quality degradation and black-image inference are different symptoms and likely different causes." + }, + { + "left": "issue:5529", + "right": "issue:6791", + "accept": false, + "reason": "Same training script, but one report is a bug and the other is a general inability to run. Too vague to merge as one defect." + }, + { + "left": "issue:9717", + "right": "issue:9911", + "accept": false, + "reason": "Flux docker-image setup and multi-ControlNet layer-length mismatch are unrelated." + }, + { + "left": "issue:10782", + "right": "issue:12666", + "accept": false, + "reason": "Lumina Image 2.0 and Kandinsky 5.0 Video Pipeline are different model families and different issues." + }, + { + "left": "issue:1083", + "right": "issue:1363", + "accept": false, + "reason": "FP16 ONNX output correctness and long-prompt/LMS scheduler errors are unrelated." + }, + { + "left": "issue:11247", + "right": "issue:9637", + "accept": false, + "reason": "Multi-GPU Flux controlnet distribution and XFormers attention-mask/bfloat16 behavior are different code paths." + }, + { + "left": "issue:4724", + "right": "issue:6283", + "accept": false, + "reason": "Both are SDXL img2img-related, but one is checkpoint compatibility and the other is strength=0 behavior in ControlNet img2img." + }, + { + "left": "issue:11657", + "right": "issue:13547", + "accept": false, + "reason": "Both are LoRA loading problems, but they target different transformer families (Wan vs Flux2) and do not look like the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 101, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10055", + "issue:10382", + "issue:10412", + "issue:10932", + "issue:1099", + "issue:1100", + "issue:11222", + "issue:11584", + "issue:11674", + "issue:1475", + "issue:1543", + "issue:1695", + "issue:1734", + "issue:2053", + "issue:2988", + "issue:3279", + "issue:3363", + "issue:3405", + "issue:3480", + "issue:4073", + "issue:4086", + "issue:4217", + "issue:4303", + "issue:4427", + "issue:4686", + "issue:4697", + "issue:4876", + "issue:4993", + "issue:5128", + "issue:5687", + "issue:5972", + "issue:6096", + "issue:6283", + "issue:660", + "issue:6879", + "issue:6880", + "issue:7003", + "issue:7159", + "issue:7660", + "issue:7770", + "issue:7982", + "issue:8604", + "issue:9012", + "issue:9045", + "issue:9496", + "issue:9525", + "issue:9803", + "issue:9895" + ], + "soft_pairs": [ + "issue:1099|issue:1475", + "issue:7003|issue:7770", + "issue:3279|issue:3363", + "issue:3405|issue:5972", + "issue:10055|issue:10412", + "issue:10932|issue:11584", + "issue:11222|issue:11674", + "issue:4217|issue:5128", + "issue:4686|issue:4697", + "issue:9045|issue:9895", + "issue:2988|issue:4303", + "issue:3480|issue:4876", + "issue:6879|issue:6880", + "issue:4427|issue:7660", + "issue:1100|issue:4073", + "issue:1543|issue:660", + "issue:1695|issue:1734", + "issue:9525|issue:9803", + "issue:9012|issue:9496", + "issue:4086|issue:7982", + "issue:2053|issue:4993", + "issue:5687|issue:6283", + "issue:6096|issue:7159", + "issue:10382|issue:8604" + ], + "prepared_review_unit_hash": "24602987c9c7ff8308129da3c4834b9985780d68118ec2df7c786f56c1c03b0c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33838, + "estimated_input_tokens": 8460, + "estimated_eval_tokens": 17176 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly a loose thematic cluster around diffusion training, ControlNet, LoRA, and device/dtype issues, but the soft pairs are not the same underlying bugs. I would not merge any of the candidate duplicates.", + "confidence": 0.68, + "canonical_issue_reason": "Issue 10055 is the best representative of the cluster\u2019s core theme: ControlNet training failing in a specific SD3 training script. The other issues are mostly separate regressions, different model families, or feature requests.", + "canonical_pr_reason": null, + "best_issue_reason": "10055 is the clearest and most central issue to use as a reference point for the controlnet-training failure subset, especially compared with the more peripheral dtype/loading/compile/feature-request items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1099", + "right": "issue:1475", + "accept": false, + "reason": "Both mention DreamBooth, but one is about missing class names while the other is about using two class types; different failures." + }, + { + "left": "issue:7003", + "right": "issue:7770", + "accept": false, + "reason": "Both are shape-mismatch style errors, but they occur in different code paths and with different artifacts (inpaint pipeline vs model/LoRA loading)." + }, + { + "left": "issue:3279", + "right": "issue:3363", + "accept": false, + "reason": "Resume-from-checkpoint errors and machine-specific failures are not the same bug." + }, + { + "left": "issue:3405", + "right": "issue:5972", + "accept": false, + "reason": "One is a ControlNet training bug, the other is a documentation issue about Min-SNR weighting; not duplicates." + }, + { + "left": "issue:10055", + "right": "issue:10412", + "accept": false, + "reason": "Both are SD3 training-related, but one is a broken ControlNet training script and the other is excessive VRAM usage in DreamBooth; different problems." + }, + { + "left": "issue:10932", + "right": "issue:11584", + "accept": false, + "reason": "Both involve torch.compile on HunyuanVideo, but the reported failures are different (recompile limit vs group offloading incompatibility)." + }, + { + "left": "issue:11222", + "right": "issue:11674", + "accept": false, + "reason": "Both are WAN-related feature requests, but they ask for different capabilities and are not the same change." + }, + { + "left": "issue:4217", + "right": "issue:5128", + "accept": false, + "reason": "Both concern safetensors/loading, but one is slow loading and the other is a load failure in a different API." + }, + { + "left": "issue:4686", + "right": "issue:4697", + "accept": false, + "reason": "Both are loading issues, but one is a saved pipeline reload problem and the other is a specific LoRA update regression." + }, + { + "left": "issue:9045", + "right": "issue:9895", + "accept": false, + "reason": "Both are device-placement issues, but they affect different pipelines and arise from different causes." + }, + { + "left": "issue:2988", + "right": "issue:4303", + "accept": false, + "reason": "Unrelated bugs: scheduler edge case versus Flax SDXL load crash." + }, + { + "left": "issue:3480", + "right": "issue:4876", + "accept": false, + "reason": "Both are LoRA parameter feature requests, but they request different knobs and do not describe the same change." + }, + { + "left": "issue:6879", + "right": "issue:6880", + "accept": false, + "reason": "Same broad Custom Diffusion area, but one is a dtype mismatch and the other is about retrieve.py; not the same bug." + }, + { + "left": "issue:4427", + "right": "issue:7660", + "accept": false, + "reason": "Both mention examples not working, but they are different examples with different functionality." + }, + { + "left": "issue:1100", + "right": "issue:4073", + "accept": false, + "reason": "Different exceptions in different components: CLIPTextTransformer device access vs DeepFloyd IF safety checker astype." + }, + { + "left": "issue:1543", + "right": "issue:660", + "accept": false, + "reason": "Both mention fp16/revision, but one is textual inversion-specific and the other is an MPS float16 runtime issue." + }, + { + "left": "issue:1695", + "right": "issue:1734", + "accept": false, + "reason": "Both are multi-GPU DreamBooth issues, but one is deepspeed training setup and the other is accelerate performance regression." + }, + { + "left": "issue:9525", + "right": "issue:9803", + "accept": false, + "reason": "Both are Flux LoRA-loading related, but one is about lora_scale being ignored and the other is a loading failure." + }, + { + "left": "issue:9012", + "right": "issue:9496", + "accept": false, + "reason": "Same model family, but one fails when loading weights and the other fails in forward execution; different code paths." + }, + { + "left": "issue:4086", + "right": "issue:7982", + "accept": false, + "reason": "Both involve ControlNet, but one is a training failure and the other is a from_pipe / MultiControlNetModel integration issue." + }, + { + "left": "issue:2053", + "right": "issue:4993", + "accept": false, + "reason": "Different regressions entirely: VRAM increase versus missing import structure entry." + }, + { + "left": "issue:5687", + "right": "issue:6283", + "accept": false, + "reason": "Both affect img2img-style behavior, but one is Karras timestep duplication and the other is a strength=0 ControlNet pipeline edge case." + }, + { + "left": "issue:6096", + "right": "issue:7159", + "accept": false, + "reason": "Both are torch.compile failures, but they involve different models and different compiler errors." + }, + { + "left": "issue:10382", + "right": "issue:8604", + "accept": false, + "reason": "Different bugs in different components: VAE forward return type mismatch versus T5 float16 corruption." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 102, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10150", + "issue:10178", + "issue:10334", + "issue:10674", + "issue:10782", + "issue:10914", + "issue:11827", + "issue:12082", + "issue:12195", + "issue:12459", + "issue:12529", + "issue:2364", + "issue:2554", + "issue:2784", + "issue:3174", + "issue:3191", + "issue:3450", + "issue:3694", + "issue:3803", + "issue:3805", + "issue:4043", + "issue:4186", + "issue:4305", + "issue:4389", + "issue:4788", + "issue:5259", + "issue:5314", + "issue:5575", + "issue:5577", + "issue:5630", + "issue:5894", + "issue:5932", + "issue:6201", + "issue:6333", + "issue:6372", + "issue:7142", + "issue:7243", + "issue:7263", + "issue:7420", + "issue:7722", + "issue:7785", + "issue:7982", + "issue:8472", + "issue:8832", + "issue:9106", + "issue:9112", + "issue:9484", + "issue:9763" + ], + "soft_pairs": [ + "issue:8832|issue:9106", + "issue:10178|issue:9112", + "issue:10782|issue:12195", + "issue:5894|issue:5932", + "issue:4389|issue:5577", + "issue:2784|issue:7243", + "issue:4043|issue:6201", + "issue:3803|issue:5575", + "issue:10334|issue:10914", + "issue:10674|issue:7982", + "issue:2364|issue:3805", + "issue:4305|issue:7420", + "issue:4186|issue:4788", + "issue:9484|issue:9763", + "issue:11827|issue:12459", + "issue:12082|issue:12529", + "issue:3694|issue:5259", + "issue:2554|issue:3450", + "issue:7263|issue:7722", + "issue:6333|issue:8472", + "issue:10150|issue:7785", + "issue:3174|issue:3191", + "issue:5314|issue:6372", + "issue:5630|issue:7142" + ], + "prepared_review_unit_hash": "fc43593349fb563d7a058ad6e7b7a857bde316bf9d1eb79d33489253a3c376fb", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33824, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated bugs and feature requests. The only clear duplicate-looking pair is the LoRA/Dreambooth loading regression (3174/3191); the rest are different models, error modes, or feature requests.", + "confidence": 0.89, + "canonical_issue_reason": "issue:3174 is the broader, more descriptive report of the LoRA/Dreambooth loading regression and likely subsumes issue:3191.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:3174 is the best canonical issue here because it clearly identifies the regression cause and scope, making it the strongest anchor for the duplicate pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8832", + "right": "issue:9106", + "accept": false, + "reason": "Different models and failures: a generic pipeline crash versus a PixArt-Alpha attribute error at time_step=1." + }, + { + "left": "issue:10178", + "right": "issue:9112", + "accept": false, + "reason": "Both are tensor-size problems, but they affect different components/models and appear to be different code paths." + }, + { + "left": "issue:10782", + "right": "issue:12195", + "accept": false, + "reason": "Unrelated issues: Lumina minor issue versus an AutoencoderKL custom_op compatibility break with PyTorch 2.3.1." + }, + { + "left": "issue:5894", + "right": "issue:5932", + "accept": false, + "reason": "Different runtime failures: invalid buffer size versus channel-mismatch in the UNet input." + }, + { + "left": "issue:4389", + "right": "issue:5577", + "accept": false, + "reason": "Both involve ControlNet loading, but one is a generic load error and the other is a specific from_single_file KeyError; not the same bug." + }, + { + "left": "issue:2784", + "right": "issue:7243", + "accept": false, + "reason": "Feature requests for different capabilities: Dreambooth on text2video versus ResAdapter support." + }, + { + "left": "issue:4043", + "right": "issue:6201", + "accept": false, + "reason": "Both concern prompt-embedding handling, but the exact pipeline behavior and failure condition differ." + }, + { + "left": "issue:3803", + "right": "issue:5575", + "accept": false, + "reason": "Multi-GPU LoRA training sharing issue versus a question about hidden-size configuration; not duplicates." + }, + { + "left": "issue:10334", + "right": "issue:10914", + "accept": false, + "reason": "Different symptoms and scopes: MacOS/MPS gray images and CPU NaNs versus unexpected CPU offloading." + }, + { + "left": "issue:10674", + "right": "issue:7982", + "accept": false, + "reason": "Different problems in different contexts: Flux GGUF loading versus from_pipe with MultiControlNetModel." + }, + { + "left": "issue:2364", + "right": "issue:3805", + "accept": false, + "reason": "Both are generic run failures, but they are separate execution environments and error reports." + }, + { + "left": "issue:4305", + "right": "issue:7420", + "accept": false, + "reason": "Related to torch.compile, but one is a version-specific graph break regression and the other is a general inability to use compile." + }, + { + "left": "issue:4186", + "right": "issue:4788", + "accept": false, + "reason": "Different failure modes: SDXL inpaint/refiner behavior versus out-of-memory on low VRAM." + }, + { + "left": "issue:9484", + "right": "issue:9763", + "accept": false, + "reason": "Both are Flux training issues, but one is multi-GPU Dreambooth training and the other is token truncation in ControlNet training." + }, + { + "left": "issue:11827", + "right": "issue:12459", + "accept": false, + "reason": "Completely different issues: safetensors header corruption versus batch inference throughput." + }, + { + "left": "issue:12082", + "right": "issue:12529", + "accept": false, + "reason": "Different Wan2.1 problems: post-compile VAE memory growth versus tiled VAE tensor-size mismatch." + }, + { + "left": "issue:3694", + "right": "issue:5259", + "accept": false, + "reason": "Similar missing-model_index wording, but one is an example-training path problem and the other is a generic directory loading error." + }, + { + "left": "issue:2554", + "right": "issue:3450", + "accept": false, + "reason": "Different dependency/version issues: unreleased accelerate requirement versus from_ckpt failing on dev version." + }, + { + "left": "issue:7263", + "right": "issue:7722", + "accept": false, + "reason": "Both mention IP-Adapter, but one is attention-slicing incompatibility and the other is an Animatediff-ControlNet+IPAdapter error." + }, + { + "left": "issue:6333", + "right": "issue:8472", + "accept": false, + "reason": "One is a feature/question about PEFT LoRA fine-tuning; the other is a separate question about simultaneous modules." + }, + { + "left": "issue:10150", + "right": "issue:7785", + "accept": false, + "reason": "Both are performance complaints, but one is broad SDXL loading/inference slowness and the other is first-inference latency in a specific version." + }, + { + "left": "issue:3174", + "right": "issue:3191", + "accept": true, + "reason": "Same underlying LoRA/Dreambooth loading regression; 3174 is the broader report and 3191 is a shorter restatement." + }, + { + "left": "issue:5314", + "right": "issue:6372", + "accept": false, + "reason": "Different pipeline integration requests: ControlNet reference inpainting versus SVD + ControlNet." + }, + { + "left": "issue:5630", + "right": "issue:7142", + "accept": false, + "reason": "Both are SDXL-related quality/performance issues, but one is scheduler noise residue and the other is SDXL-lightning output mismatch with webui." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet, and the duplicate call on issue:3174/issue:3191 is suitably conservative. The other soft-edge verdicts avoid overclaiming and are supported by the issue titles alone." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 103, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10180", + "issue:10511", + "issue:10512", + "issue:10526", + "issue:10533", + "issue:1076", + "issue:11241", + "issue:11922", + "issue:1214", + "issue:13008", + "issue:1310", + "issue:13351", + "issue:1392", + "issue:1704", + "issue:1962", + "issue:2988", + "issue:3019", + "issue:3064", + "issue:3174", + "issue:3824", + "issue:3901", + "issue:4306", + "issue:4476", + "issue:4542", + "issue:4724", + "issue:4852", + "issue:4888", + "issue:4975", + "issue:5474", + "issue:5606", + "issue:5997", + "issue:6069", + "issue:6130", + "issue:6196", + "issue:6278", + "issue:7782", + "issue:817", + "issue:8233", + "issue:8649", + "issue:8874", + "issue:9012", + "issue:938", + "issue:9393", + "issue:9729", + "issue:9825", + "issue:9876", + "issue:9927" + ], + "soft_pairs": [ + "issue:4476|issue:5997", + "issue:4852|issue:4975", + "issue:3174|issue:5606", + "issue:1076|issue:5474", + "issue:3064|issue:6130", + "issue:4542|issue:4724", + "issue:11241|issue:9927", + "issue:1310|issue:1962", + "issue:1392|issue:1962", + "issue:10533|issue:11922", + "issue:1214|issue:817", + "issue:1704|issue:3824", + "issue:3019|issue:938", + "issue:4306|issue:8649", + "issue:4888|issue:7782", + "issue:8874|issue:9012", + "issue:10511|issue:9825", + "issue:2988|issue:6069", + "issue:9393|issue:9729", + "issue:10180|issue:9876", + "issue:10512|issue:10526", + "issue:13008|issue:13351", + "issue:3901|issue:6278", + "issue:6196|issue:8233" + ], + "prepared_review_unit_hash": "3fd1a601ba9cc05add0f9ebc88489ea8522e264061f9b5cef9f83a93bd32452e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33188, + "estimated_input_tokens": 8297, + "estimated_eval_tokens": 16850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a false positive: the items span many distinct FLUX/LoRA/ControlNet/scheduler issues rather than one duplicate bug. The only moderately close pairs are same-theme reports about LoRA loading or Flux training, but they still appear to be different concrete failures or feature requests.", + "confidence": 0.86, + "canonical_issue_reason": "If one issue must anchor the cluster, #10512 is the clearest concrete bug report with substantial discussion and the most directly stated FLUX LoRA loading failure. It is still not a true umbrella duplicate for the rest of the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "#10512 is the strongest representative of the largest apparent theme (Flux LoRA loading problems): it is specific, actively discussed, and describes an actionable runtime failure. The other items are either broader feature requests, different model families, or unrelated errors.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4476", + "right": "issue:5997", + "accept": false, + "reason": "Both mention ControlNet/SDXL-ish setup problems, but one is about poor results on fill50k and the other is a model-type mismatch in a tutorial; different failure modes and not the same bug." + }, + { + "left": "issue:4852", + "right": "issue:4975", + "accept": false, + "reason": "Both are LoRA-loading related, but #4852 is a generic loading error while #4975 is an efficiency/performance complaint about repeated datatype queries." + }, + { + "left": "issue:3174", + "right": "issue:5606", + "accept": false, + "reason": "Both involve LoRA loading, but #3174 is DreamBooth + LoRA model-loading breakage from new LoRA features, while #5606 is specifically missing alpha handling." + }, + { + "left": "issue:1076", + "right": "issue:5474", + "accept": false, + "reason": "These are unrelated across time and topic; #1076 is an early SD 1.5 question, while #5474 proposes a new reference-only pipeline." + }, + { + "left": "issue:3064", + "right": "issue:6130", + "accept": false, + "reason": "Both are LoRA loading errors, but the reports differ in specifics and likely cause; no evidence they are the same underlying bug." + }, + { + "left": "issue:4542", + "right": "issue:4724", + "accept": false, + "reason": "Both concern SDXL img2img output behavior, but one is about producing multiple images and the other about checkpoint compatibility; not the same defect." + }, + { + "left": "issue:11241", + "right": "issue:9927", + "accept": false, + "reason": "Both are training-time failures around FLUX/ControlNet-like workflows, but one is a floating-point exception and the other is a header-too-large error; distinct bugs." + }, + { + "left": "issue:1310", + "right": "issue:1962", + "accept": false, + "reason": "These are different high-level pipeline questions: K-Diffusion integration vs video generation. They do not look like duplicate bug reports." + }, + { + "left": "issue:1392", + "right": "issue:1962", + "accept": false, + "reason": "Stable Diffusion 2 adoption vs video generation; same broad ecosystem, but no concrete shared bug or change." + }, + { + "left": "issue:10533", + "right": "issue:11922", + "accept": false, + "reason": "Both reference FLUX and fill-related problems, but one is a generated black image in flux fill fp16 and the other is a general running problem for FLUX.1-Fill-dev; too vague to treat as the same bug." + }, + { + "left": "issue:1214", + "right": "issue:817", + "accept": false, + "reason": "Both are mixed-precision/half-precision complaints, but they target different components (CLIPTextModel vs dreambooth example) and different symptoms." + }, + { + "left": "issue:1704", + "right": "issue:3824", + "accept": false, + "reason": "General questions about attention/inpaint pipelines, but they are unrelated topics and not duplicate bug reports." + }, + { + "left": "issue:3019", + "right": "issue:938", + "accept": false, + "reason": "A SAM text-guided editing request and a motion diffusion model request are completely different features." + }, + { + "left": "issue:4306", + "right": "issue:8649", + "accept": false, + "reason": "Different issues in different pipelines: reference pipeline list-index error vs SD3 num_images_per_prompt regression." + }, + { + "left": "issue:4888", + "right": "issue:7782", + "accept": false, + "reason": "Both mention safetensors/offline conversion, but one is about loading model safetensors offline and the other is a pickle/unpickling conversion failure." + }, + { + "left": "issue:8874", + "right": "issue:9012", + "accept": false, + "reason": "Both are load-weight errors for SD/SD3 models, but one is a single-file CKPT unpickling failure and the other is an SD3ControlNetModel.from_transformer load issue; not the same code path." + }, + { + "left": "issue:10511", + "right": "issue:9825", + "accept": false, + "reason": "#10511 is a scheduler redesign request, while #9825 is an IPAdapter support request for FLUX pipelines; same ecosystem but different changes." + }, + { + "left": "issue:2988", + "right": "issue:6069", + "accept": false, + "reason": "Both concern schedulers, but one is a training/inference step crash and the other is a specific add_noise failure at t=0; not the same bug." + }, + { + "left": "issue:9393", + "right": "issue:9729", + "accept": false, + "reason": "Both are FLUX LoRA training/save issues, but one is about DeepSpeed failing to save a model and the other is an unexpected T5EncoderModel save-model type error; related area but not clearly duplicate." + }, + { + "left": "issue:10180", + "right": "issue:9876", + "accept": false, + "reason": "#10180 is a Flux Control LoRA/multiple LoRA loading bug, while #9876 is a VRAM reclamation complaint after LoRA training; different problem domain." + }, + { + "left": "issue:10512", + "right": "issue:10526", + "accept": false, + "reason": "Both mention Flux/Quanto, but #10512 is a LoRA loading failure and #10526 is an optimum.quanto WeightQBytesTensor constructor TypeError in FP8; different underlying failure." + }, + { + "left": "issue:13008", + "right": "issue:13351", + "accept": false, + "reason": "One is a future FLUX.2 multi-image LoRA training pipeline feature request, the other is a Wan/Flux controlnet request; not duplicates." + }, + { + "left": "issue:3901", + "right": "issue:6278", + "accept": false, + "reason": "Completely unrelated: missing requirements.txt vs multi-GPU training issue." + }, + { + "left": "issue:6196", + "right": "issue:8233", + "accept": false, + "reason": "Both are depth/fp16-related errors, but one is ControlNetXS Depth SDXL loading and the other is a depth pipeline fp16 error; not enough evidence of the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 104, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10969", + "issue:11148", + "issue:11853", + "issue:12011", + "issue:12012", + "issue:12689", + "issue:1643", + "issue:1672", + "issue:1734", + "issue:1774", + "issue:2844", + "issue:3250", + "issue:3379", + "issue:350", + "issue:3788", + "issue:4000", + "issue:4061", + "issue:4086", + "issue:4303", + "issue:4354", + "issue:4736", + "issue:5150", + "issue:5436", + "issue:5465", + "issue:5575", + "issue:5693", + "issue:5840", + "issue:6238", + "issue:6250", + "issue:6323", + "issue:6407", + "issue:6540", + "issue:6661", + "issue:7142", + "issue:7310", + "issue:746", + "issue:7629", + "issue:7785", + "issue:8511", + "issue:8588", + "issue:8726", + "issue:8939", + "issue:9187", + "issue:962", + "issue:9664", + "issue:9799" + ], + "soft_pairs": [ + "issue:12011|issue:12012", + "issue:8511|issue:8939", + "issue:7142|issue:7785", + "issue:5840|issue:9664", + "issue:3788|issue:6407", + "issue:5465|issue:962", + "issue:10026|issue:1734", + "issue:10969|issue:9799", + "issue:1672|issue:4000", + "issue:6238|issue:6250", + "issue:4736|issue:7310", + "issue:3250|issue:6323", + "issue:8588|issue:8726", + "issue:4086|issue:5465", + "issue:2844|issue:4303", + "issue:1774|issue:350", + "issue:3379|issue:746", + "issue:11148|issue:12689", + "issue:5436|issue:6540", + "issue:11853|issue:9187", + "issue:5575|issue:6661", + "issue:4354|issue:5150", + "issue:4061|issue:5693", + "issue:1643|issue:7629" + ], + "prepared_review_unit_hash": "c2772506b6bbae604f3686f90531cc73f09d30d61d4624a67f9d4a035175d9a9", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33894, + "estimated_input_tokens": 8474, + "estimated_eval_tokens": 17204 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly false-positive similarity: most pairs share a broad subsystem but not the same concrete bug. Per the feedback, the SDXL VAE pair should also stay conservative; `issue:6238` and `issue:6250` are related, but the evidence is only broad VAE/load-path overlap, not enough to treat them as the same underlying defect.", + "confidence": 0.84, + "canonical_issue_reason": "`issue:6250` is the best canonical issue because it is the more specific, actionable report: it names the exact failure (`KeyError`) and the concrete path (`loading SDXL vae from single file`).", + "canonical_pr_reason": null, + "best_issue_reason": "`issue:6250` is the strongest representative for the cluster because it is symptom-specific and tied to a reproducible loading path, while `issue:6238` is broader and less diagnostic.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12011", + "right": "issue:12012", + "accept": false, + "reason": "Different failures: WAN i2v OOM vs a context/state `ValueError` from `apply_first_block_cache`." + }, + { + "left": "issue:8511", + "right": "issue:8939", + "accept": false, + "reason": "Different bugs in different areas: SD3 training-script unload behavior vs tokenizer loading for a model repo." + }, + { + "left": "issue:7142", + "right": "issue:7785", + "accept": false, + "reason": "Both mention inference, but one is a result mismatch and the other is a first-inference slowdown regression." + }, + { + "left": "issue:5840", + "right": "issue:9664", + "accept": false, + "reason": "Unrelated symptoms: resume/checkpoint freezing issue vs a subprocess failure in a LoRA training script." + }, + { + "left": "issue:3788", + "right": "issue:6407", + "accept": false, + "reason": "Both are new pipeline requests, but they are different projects/models and not the same change." + }, + { + "left": "issue:5465", + "right": "issue:962", + "accept": false, + "reason": "Both are inpaint-related, but one is about `inpaint_strength` and the other is about supported image size." + }, + { + "left": "issue:10026", + "right": "issue:1734", + "accept": false, + "reason": "Same broad training area, but one is about memory reduction under ZeRO-3 and the other is about multi-GPU slowness." + }, + { + "left": "issue:10969", + "right": "issue:9799", + "accept": false, + "reason": "Different runtime errors with no clear shared code path: 2-D weight failure vs tensor-size mismatch." + }, + { + "left": "issue:1672", + "right": "issue:4000", + "accept": false, + "reason": "Conversion-related, but one reports wrong output size while the other is a general conversion question." + }, + { + "left": "issue:6238", + "right": "issue:6250", + "accept": false, + "reason": "Same SDXL VAE area, but the evidence is only broad loading similarity; `6250` is a specific single-file `KeyError`, which is not enough to assert the same underlying bug as `6238`." + }, + { + "left": "issue:4736", + "right": "issue:7310", + "accept": false, + "reason": "Same training-script family, but OOM/invalid-cast symptoms are not the same bug as defaults not working." + }, + { + "left": "issue:3250", + "right": "issue:6323", + "accept": false, + "reason": "Different pipeline areas and failures: K-diffusion pipeline startup vs TensorRT img2img attribute handling." + }, + { + "left": "issue:8588", + "right": "issue:8726", + "accept": false, + "reason": "Missing conversion script vs text-encoder errors in SD3 dreambooth scripts; not the same defect." + }, + { + "left": "issue:4086", + "right": "issue:5465", + "accept": false, + "reason": "Different control issues: SDXL ControlNet training failure vs a DiffEdit `inpaint_strength` limitation." + }, + { + "left": "issue:2844", + "right": "issue:4303", + "accept": false, + "reason": "Completely different pipeline areas and errors: explicit image embeddings vs SDXL load crash in Flax pipeline." + }, + { + "left": "issue:1774", + "right": "issue:350", + "accept": false, + "reason": "One is a specific GPU OOM report; the other is a generic memory-usage question about UNet." + }, + { + "left": "issue:3379", + "right": "issue:746", + "accept": false, + "reason": "Dreambooth checkpoint loading vs ONNX revision loading are unrelated." + }, + { + "left": "issue:11148", + "right": "issue:12689", + "accept": false, + "reason": "Both are PEFT/LoRA-adjacent, but one is a loaded-flag bug and the other is checkpoint weights not being loaded." + }, + { + "left": "issue:5436", + "right": "issue:6540", + "accept": false, + "reason": "Both are ControlNet training issues, but the concrete failures and likely code paths differ." + }, + { + "left": "issue:11853", + "right": "issue:9187", + "accept": false, + "reason": "Both involve IP-Adapter/SDXL, but one is a conflict-size issue and the other is an `image_encoder_folder=None` loading regression." + }, + { + "left": "issue:5575", + "right": "issue:6661", + "accept": false, + "reason": "Unrelated topics: LoRA layer sizing question vs an MPS OOM report." + }, + { + "left": "issue:4354", + "right": "issue:5150", + "accept": false, + "reason": "Both are `from_single_file`-related, but they affect different pipelines and different failure modes." + }, + { + "left": "issue:4061", + "right": "issue:5693", + "accept": false, + "reason": "Both are load errors, but one is LoRA loading in SDXL and the other is VAE loading." + }, + { + "left": "issue:1643", + "right": "issue:7629", + "accept": false, + "reason": "No shared underlying bug: xformers argument mismatch vs version-specific OOM in SVD." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The report is mostly grounded, but it overstates one soft edge: `issue:6238` vs `issue:6250` is marked accepted even though the packet\u2019s own `deterministic_accept` is false and the available evidence is only broad SDXL/VAE similarity, not enough to assert the same underlying bug. The summary should stay more conservative here." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 105, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1005", + "issue:10512", + "issue:1060", + "issue:11584", + "issue:11824", + "issue:11839", + "issue:11864", + "issue:12085", + "issue:12126", + "issue:12147", + "issue:12529", + "issue:12599", + "issue:13258", + "issue:1962", + "issue:2596", + "issue:3250", + "issue:378", + "issue:4410", + "issue:4754", + "issue:5256", + "issue:5605", + "issue:5927", + "issue:6107", + "issue:6283", + "issue:6460", + "issue:6513", + "issue:6636", + "issue:671", + "issue:6832", + "issue:7306", + "issue:7770", + "issue:7935", + "issue:7936", + "issue:8482", + "issue:8495", + "issue:8556", + "issue:8637", + "issue:8790", + "issue:8827", + "issue:8939", + "issue:9106", + "issue:9167", + "issue:9187", + "issue:9270", + "issue:9320", + "issue:9410", + "issue:9953" + ], + "soft_pairs": [ + "issue:1962|issue:6460", + "issue:6513|issue:8482", + "issue:4410|issue:7770", + "issue:10512|issue:9270", + "issue:11824|issue:9106", + "issue:1005|issue:3250", + "issue:5927|issue:8790", + "issue:8637|issue:9167", + "issue:2596|issue:6636", + "issue:12147|issue:12529", + "issue:8939|issue:9320", + "issue:8495|issue:8556", + "issue:12599|issue:9953", + "issue:378|issue:4754", + "issue:6283|issue:7306", + "issue:1060|issue:671", + "issue:11584|issue:11864", + "issue:7935|issue:7936", + "issue:11839|issue:12085", + "issue:5256|issue:5605", + "issue:12126|issue:13258", + "issue:6832|issue:8827", + "issue:6636|issue:9410", + "issue:6107|issue:9187" + ], + "prepared_review_unit_hash": "7e06b15c45f6ce241c9c2297a141a9f1acb5b9e1a1d0f5a81f8860a9fcf47e6e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33596, + "estimated_input_tokens": 8399, + "estimated_eval_tokens": 17054 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All proposed soft pairs are topical overlaps only; none appear to describe the same underlying bug or mergeable PR change. The set spans distinct models, pipelines, and failure modes.", + "confidence": 0.92, + "canonical_issue_reason": "No safe canonical issue stands out: the items are mostly unrelated bug reports/features across different pipelines and models, so collapsing them would be too aggressive.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10512 is the strongest standalone issue candidate: it has the most discussion, a concrete loading failure, and enough detail to be actionable even though it is not a duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1962", + "right": "issue:6460", + "accept": false, + "reason": "Both involve video/inpainting loading behavior, but one is about video generation support and the other about endless model loading for a specific checkpoint; different bugs." + }, + { + "left": "issue:6513", + "right": "issue:8482", + "accept": false, + "reason": "Different features and models: T2IAdapter support vs Stable Diffusion 3 model support." + }, + { + "left": "issue:4410", + "right": "issue:7770", + "accept": false, + "reason": "Both mention shape mismatches, but the affected code paths and contexts differ: direct UNet loading vs model+LoRA loading in Python." + }, + { + "left": "issue:10512", + "right": "issue:9270", + "accept": false, + "reason": "Both concern LoRA, but one is a load failure and the other is incomplete support for a specific Flux training setup; not the same bug." + }, + { + "left": "issue:11824", + "right": "issue:9106", + "accept": false, + "reason": "Different pipelines and failure modes: FluxKontext multi-image input error vs PixArt-Alpha time_step=1 attribute error." + }, + { + "left": "issue:1005", + "right": "issue:3250", + "accept": false, + "reason": "Unrelated pipeline failures; one is a timestep bound question, the other is a KDiffusion pipeline run failure." + }, + { + "left": "issue:5927", + "right": "issue:8790", + "accept": false, + "reason": "Both mention IP-Adapter, but one is a composite inpaint/reference setup and the other is SDXL TensorRT integration; not the same concrete problem." + }, + { + "left": "issue:8637", + "right": "issue:9167", + "accept": false, + "reason": "Different missing imports for different pipeline classes." + }, + { + "left": "issue:2596", + "right": "issue:6636", + "accept": false, + "reason": "Documentation typo/missing symbol vs runtime import error for a different utility constant." + }, + { + "left": "issue:12147", + "right": "issue:12529", + "accept": false, + "reason": "Both are Wan2.2-related, but one is LoRA loading and the other is tiled VAE tensor mismatch; distinct subsystems and bugs." + }, + { + "left": "issue:8939", + "right": "issue:9320", + "accept": false, + "reason": "Tokenizer/config loading for SD3 vs saving/loading finetuned custom diffusion models; different failure points." + }, + { + "left": "issue:8495", + "right": "issue:8556", + "accept": false, + "reason": "Both are SD3 loading issues, but one is a tokenizer enum parse error and the other is a meta-tensor copy failure from single-file loading." + }, + { + "left": "issue:12599", + "right": "issue:9953", + "accept": false, + "reason": "Both concern moving pipelines to CUDA, but one is slowness and the other is an error with quantized components." + }, + { + "left": "issue:378", + "right": "issue:4754", + "accept": false, + "reason": "Textual inversion appears in both, but one is a model/component class issue and the other is a missing model_index.json when loading a saved artifact." + }, + { + "left": "issue:6283", + "right": "issue:7306", + "accept": false, + "reason": "Different SDXL bugs: strength=0 handling in ControlNet img2img vs float16 incompatibility in LEdits++." + }, + { + "left": "issue:1060", + "right": "issue:671", + "accept": false, + "reason": "One is a converter safety-checker load issue, the other is model non-identity after conversion; not the same bug." + }, + { + "left": "issue:11584", + "right": "issue:11864", + "accept": false, + "reason": "Both involve torch.compile, but one is a group-offloading incompatibility and the other is a fullgraph compile NameError in a different module." + }, + { + "left": "issue:7935", + "right": "issue:7936", + "accept": false, + "reason": "One is missing config.json for a repo, the other is invalid JSON when loading a local safetensors file." + }, + { + "left": "issue:11839", + "right": "issue:12085", + "accept": false, + "reason": "Both are gguf loading problems, but for different models and likely different input-shape expectations." + }, + { + "left": "issue:5256", + "right": "issue:5605", + "accept": false, + "reason": "Completely different feature requests: Insta-Flow training support vs per-image caption support for SDXL DreamBooth." + }, + { + "left": "issue:12126", + "right": "issue:13258", + "accept": false, + "reason": "Different Wan2.2 reports: a cache_context NoneType inference error vs a missing image-to-video model endpoint." + }, + { + "left": "issue:6832", + "right": "issue:8827", + "accept": false, + "reason": "Both involve IP-Adapter FaceID, but the specific errors and pipelines differ; not enough evidence of the same bug." + }, + { + "left": "issue:6636", + "right": "issue:9410", + "accept": false, + "reason": "Different import failures for different pipeline symbols." + }, + { + "left": "issue:6107", + "right": "issue:9187", + "accept": false, + "reason": "One is environment-specific IP-Adapter behavior on AWS, the other is an SDXL regression after a Kolors update with image_encoder_folder=None." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 106, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10100", + "issue:1013", + "issue:10139", + "issue:10447", + "issue:10472", + "issue:1061", + "issue:10639", + "issue:10653", + "issue:10823", + "issue:11232", + "issue:12107", + "issue:12157", + "issue:12599", + "issue:1398", + "issue:1823", + "issue:2459", + "issue:2466", + "issue:2784", + "issue:3101", + "issue:3487", + "issue:3643", + "issue:3700", + "issue:3802", + "issue:4031", + "issue:4109", + "issue:4134", + "issue:4446", + "issue:4631", + "issue:4801", + "issue:5126", + "issue:5128", + "issue:5425", + "issue:5558", + "issue:5576", + "issue:5607", + "issue:5932", + "issue:6382", + "issue:6839", + "issue:6858", + "issue:7002", + "issue:7133", + "issue:7197", + "issue:7974", + "issue:9454", + "issue:9576", + "issue:9613", + "issue:9719" + ], + "soft_pairs": [ + "issue:10472|issue:9613", + "issue:2784|issue:5576", + "issue:4801|issue:5425", + "issue:2459|issue:3700", + "issue:4446|issue:6839", + "issue:1398|issue:4134", + "issue:4031|issue:5128", + "issue:2466|issue:3802", + "issue:10653|issue:12599", + "issue:6382|issue:7974", + "issue:5607|issue:7002", + "issue:3487|issue:5558", + "issue:3101|issue:6858", + "issue:12107|issue:12157", + "issue:3643|issue:4631", + "issue:7133|issue:7197", + "issue:10823|issue:11232", + "issue:5126|issue:5932", + "issue:10010|issue:9454", + "issue:10100|issue:10639", + "issue:1013|issue:4109", + "issue:10139|issue:9576", + "issue:10447|issue:9719", + "issue:1061|issue:1823" + ], + "prepared_review_unit_hash": "32f20781ffee8a2cba5b88a49b2de489b0a84bc82258c79811b12d701c669e0e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34734, + "estimated_input_tokens": 8684, + "estimated_eval_tokens": 17624 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly heterogeneous and not one duplicate cluster. Only one soft pair looks like a true duplicate: the training-example checkpoint-limit bug. All other pairs share broad subsystem or symptom wording but describe different concrete failures or requests.", + "confidence": 0.94, + "canonical_issue_reason": "issue:2466 is the best canonical issue because it states the checkpoint-limit training-example bug most directly and cleanly; issue:3802 is the same underlying problem with slightly different wording.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:2466 is the strongest representative in this set: it is a clear, specific bug report with the clearest duplicate relationship to another item, while most other issues are unrelated or only superficially similar.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10472", + "right": "issue:9613", + "accept": false, + "reason": "Different failures: unexpected lora_bias argument during LoRA inference vs missing Timesteps.scale in 0.30.x. Same broad area, not the same bug." + }, + { + "left": "issue:2784", + "right": "issue:5576", + "accept": false, + "reason": "Both are feature requests, but they target different changes: Dreambooth for text2video models vs a new FreeNoise scheduler." + }, + { + "left": "issue:4801", + "right": "issue:5425", + "accept": false, + "reason": "Both mention training examples and checkpoints, but one is mixed-precision behavior in train_unconditional.py and the other is accelerate/PT 2.1 checkpoint-saving compatibility." + }, + { + "left": "issue:2459", + "right": "issue:3700", + "accept": false, + "reason": "Different training-script issues: xformers attention not working in LoRA training vs min_snr argument not usable. Similar script family, not the same bug." + }, + { + "left": "issue:4446", + "right": "issue:6839", + "accept": false, + "reason": "Both involve from_single_file-style loading, but one is custom text_encoder support for StableDiffusionPipeline and the other is StableVideoDiffusionPipeline single-file loading." + }, + { + "left": "issue:1398", + "right": "issue:4134", + "accept": false, + "reason": "Both are Apple/MPS-related, but the concrete problems differ: SD2 on MPS vs ControlNet tensor-size mismatch on M1 Max." + }, + { + "left": "issue:4031", + "right": "issue:5128", + "accept": false, + "reason": "Both concern loading from local files, but one is standalone safety checker loading and the other is XLPipeline local safetensors loading; different code paths." + }, + { + "left": "issue:2466", + "right": "issue:3802", + "accept": true, + "reason": "Same underlying bug: `checkpoints_total_limit` is not being enforced in training examples. The titles and descriptions point to the same checkpoint-rotation problem." + }, + { + "left": "issue:10653", + "right": "issue:12599", + "accept": false, + "reason": "Different problems around device movement/performance: bitsandbytes 8-bit module `.to()` CPU restriction vs slow CUDA move after loading." + }, + { + "left": "issue:6382", + "right": "issue:7974", + "accept": false, + "reason": "Both mention IP Adapter, but one is an IP-adapter key error in LCM LoRA SDXL training/use and the other is inpaint with IP Adapter face+style." + }, + { + "left": "issue:5607", + "right": "issue:7002", + "accept": false, + "reason": "Both are runtime shape/buffer errors, but the reported failures are different: tensor concat mismatch vs invalid buffer size in img2img decoding." + }, + { + "left": "issue:3487", + "right": "issue:5558", + "accept": false, + "reason": "Both involve inpaint output quality, but one is a scheduler/strength quality issue and the other is artifacts with multiple ControlNets." + }, + { + "left": "issue:3101", + "right": "issue:6858", + "accept": false, + "reason": "Both are controlnet-training related, but the actual errors differ: a training example yield error vs `'str' object has no attribute 'convert'`." + }, + { + "left": "issue:12107", + "right": "issue:12157", + "accept": false, + "reason": "Different accelerator/training errors: custom object list passed to `init_trackers` vs `'torch.device' object is not iterable` with bnb quantization config." + }, + { + "left": "issue:3643", + "right": "issue:4631", + "accept": false, + "reason": "Both are runtime exceptions, but one is an INT_MAX element limit while decoding latents and the other is a scalar conversion failure in DPMSolverSinglestepScheduler." + }, + { + "left": "issue:7133", + "right": "issue:7197", + "accept": false, + "reason": "Both involve resuming training, but one is SDXL Dreambooth LoRA text-encoder size mismatch and the other is a KeyError for 'Shortest Edge' in a different script." + }, + { + "left": "issue:10823", + "right": "issue:11232", + "accept": false, + "reason": "Both relate to fine-tuning workflows, but one is missing inference artifacts after training and the other is noise residuals during text-to-image fine-tuning." + }, + { + "left": "issue:5126", + "right": "issue:5932", + "accept": false, + "reason": "Different shape-related errors: a Conv2d forward signature mismatch vs a 4-channel input expectation being violated." + }, + { + "left": "issue:10010", + "right": "issue:9454", + "accept": false, + "reason": "Both mention a missing model/file, but they are for different models and failure points: FLUX Fill example model_index.json vs CogVideoX Image2Video model not found." + }, + { + "left": "issue:10100", + "right": "issue:10639", + "accept": false, + "reason": "Both are feature requests for model support, but they concern different models: ConsisID vs Lumina-Image 2.0." + }, + { + "left": "issue:1013", + "right": "issue:4109", + "accept": false, + "reason": "Both are pipeline-not-working issues, but they are different pipelines and likely different root causes: StableDiffusionMegaPipeline vs StableDiffusionXLInpaintPipeline." + }, + { + "left": "issue:10139", + "right": "issue:9576", + "accept": false, + "reason": "Different pipeline-loading bugs: `sample_size` handling in pipeline init vs `use_safetensors` not being passed to submodules." + }, + { + "left": "issue:10447", + "right": "issue:9719", + "accept": false, + "reason": "Different topics entirely: converter output splitting for large UNet models vs `disable_progress_bar` being ignored during checkpoint loading." + }, + { + "left": "issue:1061", + "right": "issue:1823", + "accept": false, + "reason": "One is documentation/instructions for textual inversion, the other is swapped comments in depth2img. Same area, not the same bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: the titles support that most pairs are only loosely related, and the 2466/3802 pair is the one clear same-bug match among the listed soft edges. The verdicts are conservative overall, with no obvious overreach from the available evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 107, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10671", + "issue:11006", + "issue:11627", + "issue:1188", + "issue:12494", + "issue:12565", + "issue:13300", + "issue:1486", + "issue:1949", + "issue:2437", + "issue:2579", + "issue:2673", + "issue:2800", + "issue:3346", + "issue:3728", + "issue:3851", + "issue:4037", + "issue:4206", + "issue:4975", + "issue:5215", + "issue:5337", + "issue:5457", + "issue:5593", + "issue:6159", + "issue:6842", + "issue:6879", + "issue:7389", + "issue:7390", + "issue:7441", + "issue:7982", + "issue:8532", + "issue:8534", + "issue:8540", + "issue:8543", + "issue:8805", + "issue:8817", + "issue:8834", + "issue:9042", + "issue:9109", + "issue:9156", + "issue:9167", + "issue:9309", + "issue:9450", + "issue:9527", + "issue:9546", + "issue:9630", + "issue:9873" + ], + "soft_pairs": [ + "issue:1188|issue:2437", + "issue:1486|issue:2579", + "issue:1486|issue:3851", + "issue:1949|issue:3346", + "issue:2673|issue:5215", + "issue:2800|issue:3728", + "issue:4037|issue:4206", + "issue:4975|issue:8532", + "issue:5337|issue:5457", + "issue:6842|issue:9167", + "issue:7441|issue:8805", + "issue:8534|issue:8543", + "issue:8540|issue:8834", + "issue:9042|issue:9873", + "issue:9450|issue:9630", + "issue:10671|issue:9527", + "issue:9309|issue:9546", + "issue:11627|issue:8817", + "issue:10026|issue:9156", + "issue:6159|issue:9109", + "issue:11006|issue:12565", + "issue:11627|issue:13300", + "issue:12494|issue:6879", + "issue:7389|issue:7390", + "issue:5593|issue:7982" + ], + "prepared_review_unit_hash": "933f2dd3f84b34c2576cc9c81d81d8b5c0f8d03d51123824c45912111dc769f0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34093, + "estimated_input_tokens": 8524, + "estimated_eval_tokens": 17304 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a loose similarity cluster of unrelated diffusers issues across Flux, SDXL, Wan, LoRA, and pipeline-feature requests. Only one pair looks like the same concrete bug family: SDXL `from_single_file` loading errors.", + "confidence": 0.83, + "canonical_issue_reason": "issue:7389 is the clearest representative: it has a concrete SDXL `from_single_file` failure, and it is tightly paired with a near-identical sibling issue on a related SDXL pipeline.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:7389 is the strongest issue to anchor the cluster because it describes a specific runtime failure in a shared loading path, making it the most useful representative if these are being collapsed.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1188", + "right": "issue:2437", + "accept": false, + "reason": "Both are pipeline feature requests, but they ask for different pipelines and do not describe the same bug/change." + }, + { + "left": "issue:1486", + "right": "issue:2579", + "accept": false, + "reason": "Different scheduler bug vs possible xFormers misuse; no concrete shared failure mode." + }, + { + "left": "issue:1486", + "right": "issue:3851", + "accept": false, + "reason": "Both involve DDIMScheduler, but the prediction type and failure conditions differ." + }, + { + "left": "issue:1949", + "right": "issue:3346", + "accept": false, + "reason": "Both are DreamBooth training issues, but one is tutorial errors and the other is checkpoint resume support." + }, + { + "left": "issue:2673", + "right": "issue:5215", + "accept": false, + "reason": "Both mention StableDiffusionUpscalePipeline, but the reported problems are unrelated." + }, + { + "left": "issue:2800", + "right": "issue:3728", + "accept": false, + "reason": "Both are new pipeline requests, but for different methods/models." + }, + { + "left": "issue:4037", + "right": "issue:4206", + "accept": false, + "reason": "Both are ControlNet training issues, but one is FSDP validation errors and the other is an SDXL main-branch breakage." + }, + { + "left": "issue:4975", + "right": "issue:8532", + "accept": false, + "reason": "Related to LoRA loading, but one is a performance issue and the other is memory growth." + }, + { + "left": "issue:5337", + "right": "issue:5457", + "accept": false, + "reason": "Different API documentation questions with no shared bug." + }, + { + "left": "issue:6842", + "right": "issue:9167", + "accept": false, + "reason": "Both are import errors, but they concern different pipeline classes and likely different export gaps." + }, + { + "left": "issue:7441", + "right": "issue:8805", + "accept": false, + "reason": "Completely different problems: Python 3.12 pytest failure vs an SDXL conversion script regression." + }, + { + "left": "issue:8534", + "right": "issue:8543", + "accept": false, + "reason": "Both are SD3 dreambooth training issues, but logit weighting and gradient checkpointing are distinct failures." + }, + { + "left": "issue:8540", + "right": "issue:8834", + "accept": false, + "reason": "Caching/optimization request vs a question about releasing ControlNet training code." + }, + { + "left": "issue:9042", + "right": "issue:9873", + "accept": false, + "reason": "Both are feature requests to add pipelines, but for different models." + }, + { + "left": "issue:9450", + "right": "issue:9630", + "accept": false, + "reason": "Both are Flux multi-GPU issues, but one is a tensor device mismatch and the other is a num_single_layers configuration error." + }, + { + "left": "issue:10671", + "right": "issue:9527", + "accept": false, + "reason": "Both touch dtype/device handling, but one is deterministic accessor behavior and the other is a mixed-precision controlnet runtime error." + }, + { + "left": "issue:9309", + "right": "issue:9546", + "accept": false, + "reason": "Both are Flux ControlNet training/runtime issues, but one is a schnell pipeline bug and the other is a validation-step OOM." + }, + { + "left": "issue:11627", + "right": "issue:8817", + "accept": false, + "reason": "Different model/pipeline topics: SD3.5 IP-adapter usage vs IP-Adapter FaceID trouble." + }, + { + "left": "issue:10026", + "right": "issue:9156", + "accept": false, + "reason": "Both are memory-related Flux training issues, but one is DeepSpeed stage-3 memory scaling and the other is LoRA CUDA OOM." + }, + { + "left": "issue:6159", + "right": "issue:9109", + "accept": false, + "reason": "Unrelated failures: an unsupported pipeline kwarg vs a missing torchvision operator." + }, + { + "left": "issue:11006", + "right": "issue:12565", + "accept": false, + "reason": "Both are Wan2.1 I2V issues, but the concrete symptoms differ and no shared code path is identified." + }, + { + "left": "issue:11627", + "right": "issue:13300", + "accept": false, + "reason": "Different generation setups and failures; no evidence they are the same bug." + }, + { + "left": "issue:12494", + "right": "issue:6879", + "accept": false, + "reason": "Shape-mismatch matmul error vs dtype mismatch in custom diffusion; different underlying faults." + }, + { + "left": "issue:7389", + "right": "issue:7390", + "accept": true, + "reason": "Both report `from_single_file` errors on closely related SDXL pipelines, suggesting the same loader bug in a shared code path." + }, + { + "left": "issue:5593", + "right": "issue:7982", + "accept": false, + "reason": "Both involve `from_pipe` and ControlNet, but one is AutoPipeline conversion and the other is MultiControlNetModel support; too distinct to merge confidently." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays mostly conservative. The only strong match identified is the SDXL `from_single_file` pair, which is supported by the titles. The remaining pairs are rejected with cautious reasoning and no clear overclaiming. The summary accurately reflects that this is a loose similarity cluster rather than a coherent single-bug set." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 108, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10472", + "issue:10760", + "issue:11962", + "issue:13351", + "issue:1556", + "issue:1814", + "issue:2507", + "issue:2560", + "issue:259", + "issue:2614", + "issue:2695", + "issue:2784", + "issue:2880", + "issue:2984", + "issue:2988", + "issue:3584", + "issue:3812", + "issue:3886", + "issue:3910", + "issue:4186", + "issue:4376", + "issue:4591", + "issue:4840", + "issue:4885", + "issue:5128", + "issue:5399", + "issue:5425", + "issue:5592", + "issue:5770", + "issue:5777", + "issue:5829", + "issue:5845", + "issue:5972", + "issue:6073", + "issue:6252", + "issue:6492", + "issue:6823", + "issue:7082", + "issue:7239", + "issue:7421", + "issue:7708", + "issue:7793", + "issue:9012", + "issue:9208", + "issue:9239", + "issue:941", + "issue:964", + "issue:9731" + ], + "soft_pairs": [ + "issue:6252|issue:7793", + "issue:2507|issue:5128", + "issue:2988|issue:4376", + "issue:3910|issue:4885", + "issue:3886|issue:5425", + "issue:6492|issue:9239", + "issue:4591|issue:7421", + "issue:4186|issue:964", + "issue:3584|issue:6073", + "issue:10472|issue:9731", + "issue:2695|issue:5972", + "issue:11962|issue:13351", + "issue:1556|issue:2507", + "issue:4885|issue:9012", + "issue:5777|issue:5845", + "issue:10760|issue:7082", + "issue:1814|issue:2614", + "issue:2560|issue:941", + "issue:259|issue:2880", + "issue:4840|issue:5592", + "issue:6823|issue:9208", + "issue:3812|issue:7708", + "issue:2784|issue:2984", + "issue:5770|issue:5829", + "issue:5399|issue:7239" + ], + "prepared_review_unit_hash": "dc9f7a1d94a82958b2870cae50b852f95799bade8004855170c4e75bc82d23f4", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34159, + "estimated_input_tokens": 8540, + "estimated_eval_tokens": 17336 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous set of loosely keyword-matched issues rather than a true duplicate cluster. Most links are only superficial similarity around LoRA, loading, ControlNet, training, or black-image symptoms; none look like the same concrete bug.", + "confidence": 0.96, + "canonical_issue_reason": "issue:6252 is the clearest concrete bug report in the set: a specific LoRA checkpoint-loading failure with actionable symptoms, making it the best representative of the loosely related loading/LoRA subset.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6252 is the most usable issue to anchor triage because it is specific, user-facing, and reproducible, unlike the more generic feature requests or broad compatibility reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6252", + "right": "issue:7793", + "accept": false, + "reason": "Both mention LoRA, but one is checkpoint loading and the other is unloading DeepFloyd LoRA weights; different code paths and failure modes." + }, + { + "left": "issue:2507", + "right": "issue:5128", + "accept": false, + "reason": "GPU safetensors load slowness is a performance issue, while the other is a missing import when loading a local .safetensors file." + }, + { + "left": "issue:2988", + "right": "issue:4376", + "accept": false, + "reason": "Scheduler step-count crash and SDXL inpaint/refiner crash are unrelated bugs in different pipeline paths." + }, + { + "left": "issue:3910", + "right": "issue:4885", + "accept": false, + "reason": "Generic LoRA load failure versus SDXL ControlNet img2img not loading LoRA weights; same broad area, but not the same concrete bug." + }, + { + "left": "issue:3886", + "right": "issue:5425", + "accept": false, + "reason": "A distributed watchdog timeout during training is not the same as an accelerate/PT 2.1 checkpoint-saving compatibility issue." + }, + { + "left": "issue:6492", + "right": "issue:9239", + "accept": false, + "reason": "Both involve black images, but one is generic image generation and the other is specifically ControlNet-Canny; too broad to merge." + }, + { + "left": "issue:4591", + "right": "issue:7421", + "accept": false, + "reason": "Deterministic generation problems in Karlo/unCLIP and seed handling in StableDiffusionXLKDiffusionPipeline are different model/code paths." + }, + { + "left": "issue:4186", + "right": "issue:964", + "accept": false, + "reason": "Both are inpainting-related, but one is SDXL inpaint with refiner and the other is an inpainting example failure; not the same bug." + }, + { + "left": "issue:3584", + "right": "issue:6073", + "accept": false, + "reason": "These are different DreamBooth tutorial failures with different setups and error contexts, not one underlying defect." + }, + { + "left": "issue:10472", + "right": "issue:9731", + "accept": false, + "reason": "The first is a LoRA inference TypeError about lora_bias; the second is a training/how-to request for Playground2.5 style LoRA use." + }, + { + "left": "issue:2695", + "right": "issue:5972", + "accept": false, + "reason": "ControlNet training validation crash and a documentation note about Min-SNR weighting are not the same issue." + }, + { + "left": "issue:11962", + "right": "issue:13351", + "accept": false, + "reason": "GGUF support for FLUX.1-Kontext-dev and Wan/Flux.2 ControlNet support requests are different feature requests." + }, + { + "left": "issue:1556", + "right": "issue:2507", + "accept": false, + "reason": "Black images on CUDA and safetensors GPU load performance are unrelated symptoms." + }, + { + "left": "issue:4885", + "right": "issue:9012", + "accept": false, + "reason": "One is LoRA weights not loading in an SDXL ControlNet pipeline; the other is SD3ControlNetModel weight-loading failure; different models and failure points." + }, + { + "left": "issue:5777", + "right": "issue:5845", + "accept": false, + "reason": "These are both DreamBooth/PixArt training issues, but they concern different model variants and error conditions." + }, + { + "left": "issue:10760", + "right": "issue:7082", + "accept": false, + "reason": "An attribute missing on an ONNX inpaint pipeline and an import error for CpuOffload are distinct import/runtime problems." + }, + { + "left": "issue:1814", + "right": "issue:2614", + "accept": false, + "reason": "Changing the CLIP model and changing the pretrained-model download directory are unrelated configuration requests." + }, + { + "left": "issue:2560", + "right": "issue:941", + "accept": false, + "reason": "Both are determinism/reproducibility complaints, but they affect different pipelines and different trigger conditions." + }, + { + "left": "issue:259", + "right": "issue:2880", + "accept": false, + "reason": "A progress bar fix is unrelated to a JAX Colab runtime failure." + }, + { + "left": "issue:4840", + "right": "issue:5592", + "accept": false, + "reason": "Both are checkpoint-loading issues, but they involve different model classes and checkpoint formats, so they are not the same bug." + }, + { + "left": "issue:6823", + "right": "issue:9208", + "accept": false, + "reason": "General cannot-use-ControlNet versus from_single_file() failing on most ControlNet models are related only at subsystem level, not one concrete defect." + }, + { + "left": "issue:3812", + "right": "issue:7708", + "accept": false, + "reason": "Missing controlnet config values and missing tokenizer/config.json are different missing-file/config problems." + }, + { + "left": "issue:2784", + "right": "issue:2984", + "accept": false, + "reason": "A DreamBooth feature request for text-to-video and a latent video diffusion model announcement/problem are not duplicates." + }, + { + "left": "issue:5770", + "right": "issue:5829", + "accept": false, + "reason": "Dataset access blocking LCM distillation and a missing unet_time_cond_proj_dim argument are unrelated training issues." + }, + { + "left": "issue:5399", + "right": "issue:7239", + "accept": false, + "reason": "A dtype mismatch during custom inpainting inference is different from an SDXL LoRA shape-mismatch during DreamBooth training." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 109, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10066", + "issue:10194", + "issue:10382", + "issue:10453", + "issue:10989", + "issue:11408", + "issue:11432", + "issue:11436", + "issue:1231", + "issue:1363", + "issue:1614", + "issue:2447", + "issue:2695", + "issue:2822", + "issue:3094", + "issue:3221", + "issue:3296", + "issue:3405", + "issue:4037", + "issue:4086", + "issue:4210", + "issue:4389", + "issue:4724", + "issue:4852", + "issue:4874", + "issue:5128", + "issue:5254", + "issue:5465", + "issue:5829", + "issue:5881", + "issue:5886", + "issue:5897", + "issue:5997", + "issue:6250", + "issue:6371", + "issue:7033", + "issue:7168", + "issue:7238", + "issue:7877", + "issue:835", + "issue:8410", + "issue:8464", + "issue:8645", + "issue:9096", + "issue:9370", + "issue:9825", + "issue:9856", + "issue:987" + ], + "soft_pairs": [ + "issue:3405|issue:4037", + "issue:3221|issue:4724", + "issue:835|issue:987", + "issue:1614|issue:987", + "issue:8410|issue:9825", + "issue:4852|issue:5128", + "issue:6250|issue:9856", + "issue:10382|issue:4210", + "issue:11436|issue:9096", + "issue:2447|issue:5897", + "issue:5829|issue:9370", + "issue:10453|issue:11408", + "issue:3094|issue:6371", + "issue:2822|issue:5881", + "issue:5254|issue:5465", + "issue:10989|issue:8464", + "issue:4210|issue:8645", + "issue:4874|issue:5128", + "issue:2695|issue:5997", + "issue:3221|issue:4086", + "issue:3296|issue:4389", + "issue:1231|issue:1363", + "issue:5886|issue:7238", + "issue:10066|issue:11432", + "issue:10194|issue:7877", + "issue:7033|issue:7168" + ], + "prepared_review_unit_hash": "f87043c58560835578dfcffd463dba6805645a8a8a269364597ee8ab97817fda", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 34789, + "estimated_input_tokens": 8698, + "estimated_eval_tokens": 17652 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very loose and mostly heterogeneous. A few items orbit attention/masking and FLUX/IP-Adapter support, but most pairs are different bugs or feature requests in adjacent subsystems, so I would not merge any soft edges.", + "confidence": 0.41, + "canonical_issue_reason": "Issue 10194 is the best central representative because it most directly names the shared theme of attention masking in FLUX, which is the closest common thread among the few related items in this cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "10194 is the broadest, most on-theme issue for the small attention/masking subgroup; it is more representative than the more specific error reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3405", + "right": "issue:4037", + "accept": false, + "reason": "Both are ControlNet training problems, but one is generic and the other is specifically validation errors under FSDP; not the same bug." + }, + { + "left": "issue:3221", + "right": "issue:4724", + "accept": false, + "reason": "Both involve loading/pipeline behavior, but they refer to different APIs and failure modes (attn procs vs SDXL img2img checkpoint behavior)." + }, + { + "left": "issue:835", + "right": "issue:987", + "accept": false, + "reason": "Both mention output image defects, but one is a flax pipeline conversion issue and the other is a speech-to-image green output; not the same code path." + }, + { + "left": "issue:1614", + "right": "issue:987", + "accept": false, + "reason": "Both are image-output corruption reports, but they are different models and symptoms (blank black vs green image)." + }, + { + "left": "issue:8410", + "right": "issue:9825", + "accept": false, + "reason": "Both are FLUX-related feature requests, but one asks for native TextDiffuser pipeline support and the other for IP-Adapter support." + }, + { + "left": "issue:4852", + "right": "issue:5128", + "accept": false, + "reason": "Both are loading errors, but one is LoRA loading and the other is single-file XL pipeline loading; too different to be duplicates." + }, + { + "left": "issue:6250", + "right": "issue:9856", + "accept": false, + "reason": "KeyError loading an SDXL VAE has nothing to do with a distributed port collision." + }, + { + "left": "issue:10382", + "right": "issue:4210", + "accept": false, + "reason": "VAE encoder return-type mismatch is unrelated to UNet attention_mask usage." + }, + { + "left": "issue:11436", + "right": "issue:9096", + "accept": false, + "reason": "Both are precision/dtype related, but one is a matmul dtype mismatch in HiDream and the other is NaNs in Flux half precision; not the same defect." + }, + { + "left": "issue:2447", + "right": "issue:5897", + "accept": false, + "reason": "Both mention LoRA errors, but one is a broad report and the other is a specific CPU/GPU device mismatch during training." + }, + { + "left": "issue:5829", + "right": "issue:9370", + "accept": false, + "reason": "Both involve missing arguments in model/embedding code, but they are different missing parameters in different components." + }, + { + "left": "issue:10453", + "right": "issue:11408", + "accept": false, + "reason": "Hunyuan batch-size limitations and compiled-model set_adapters support are unrelated issues." + }, + { + "left": "issue:3094", + "right": "issue:6371", + "accept": false, + "reason": "OOM during text-to-image finetuning is unrelated to an assert error in a controlnet_xs example." + }, + { + "left": "issue:2822", + "right": "issue:5881", + "accept": false, + "reason": "Training convergence quality complaints for ControlNet and Animatediff are not the same underlying bug." + }, + { + "left": "issue:5254", + "right": "issue:5465", + "accept": false, + "reason": "Multi-ControlNet support and DiffEdit inpaint_strength limitations are different feature gaps." + }, + { + "left": "issue:10989", + "right": "issue:8464", + "accept": false, + "reason": "Flux ControlNet LoRA loading under quantization and SEGA cpu-offload seed failure are unrelated." + }, + { + "left": "issue:4210", + "right": "issue:8645", + "accept": false, + "reason": "A question about UNet attention_mask is not the same as a missing helper function in single_file_utils.py." + }, + { + "left": "issue:4874", + "right": "issue:5128", + "accept": false, + "reason": "Both are local loading/scheduler related, but one is an upscale pipeline scheduler bug and the other is single-file safetensors loading." + }, + { + "left": "issue:2695", + "right": "issue:5997", + "accept": false, + "reason": "ControlNet training validation/TensorBoard behavior is unrelated to the SDXL tutorial model-type mismatch." + }, + { + "left": "issue:3221", + "right": "issue:4086", + "accept": false, + "reason": "Attn-proc loading in one pipeline is not the same as ControlNet training failing with SDXL base 0.9." + }, + { + "left": "issue:3296", + "right": "issue:4389", + "accept": false, + "reason": "Checkpoint save/load with text_encoder and ControlNetModel loading errors are different problems." + }, + { + "left": "issue:1231", + "right": "issue:1363", + "accept": false, + "reason": "Scheduler support in Flax and an ONNX long-prompt/LMS error are related only by scheduler name, not the underlying bug." + }, + { + "left": "issue:5886", + "right": "issue:7238", + "accept": false, + "reason": "IP-Adapter face model support and IP-Adapter masking are distinct feature requests." + }, + { + "left": "issue:10066", + "right": "issue:11432", + "accept": false, + "reason": "Attention unpacking failure and from_pretrained torch_dtype='auto' issues are different initialization paths." + }, + { + "left": "issue:10194", + "right": "issue:7877", + "accept": false, + "reason": "Flux attention masking support and adding out_dim to Attention block are different API changes." + }, + { + "left": "issue:7033", + "right": "issue:7168", + "accept": false, + "reason": "Attention masking warning logs and incorrect IP-Adapter image embeds are separate issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 110, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10121", + "issue:10925", + "issue:11185", + "issue:11464", + "issue:11878", + "issue:13105", + "issue:1315", + "issue:1572", + "issue:1944", + "issue:2163", + "issue:2596", + "issue:2875", + "issue:2947", + "issue:3111", + "issue:3405", + "issue:3480", + "issue:3631", + "issue:4073", + "issue:4086", + "issue:4186", + "issue:4969", + "issue:5128", + "issue:5515", + "issue:5619", + "issue:5892", + "issue:6016", + "issue:6130", + "issue:6196", + "issue:6256", + "issue:6439", + "issue:6548", + "issue:6565", + "issue:6870", + "issue:7003", + "issue:7119", + "issue:7293", + "issue:730", + "issue:7538", + "issue:7735", + "issue:7757", + "issue:7960", + "issue:9208", + "issue:9278", + "issue:9501", + "issue:9558", + "issue:9731", + "issue:9804" + ], + "soft_pairs": [ + "issue:2875|issue:6565", + "issue:2947|issue:4073", + "issue:3111|issue:3480", + "issue:4969|issue:5892", + "issue:3405|issue:4086", + "issue:1572|issue:3631", + "issue:6548|issue:9558", + "issue:5515|issue:5619", + "issue:11878|issue:13105", + "issue:6256|issue:7293", + "issue:6439|issue:7119", + "issue:6870|issue:7735", + "issue:9208|issue:9804", + "issue:10121|issue:11185", + "issue:6016|issue:7003", + "issue:7538|issue:9731", + "issue:7757|issue:7960", + "issue:1315|issue:2596", + "issue:9278|issue:9501", + "issue:2163|issue:6130", + "issue:5128|issue:6196", + "issue:1944|issue:730", + "issue:10925|issue:11464", + "issue:4186|issue:6870" + ], + "prepared_review_unit_hash": "096b87ac506d9e16ac3c2d6d58c883000e95619811940065267bed3ebea637f5", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33337, + "estimated_input_tokens": 8335, + "estimated_eval_tokens": 16926 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is noisy and heterogeneous: it mixes unrelated training, loading, import, and docs issues across different models and pipelines. None of the soft pairs look like the same underlying bug/change, so all should be rejected.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a strong canonical representative because the cluster spans multiple unrelated problem types; the closest subtheme is model-loading failures, but even those split across different APIs and models.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2875", + "right": "issue:6565", + "accept": false, + "reason": "Both are training-loss problems, but one is generic text-to-image NaNs and the other is DPO diffusion loss spikes/noised samples; different training setups and failure modes." + }, + { + "left": "issue:2947", + "right": "issue:4073", + "accept": false, + "reason": "Different components and errors: missing `load_attn_procs` on UNet vs safety-checker `astype` on a Tensor." + }, + { + "left": "issue:3111", + "right": "issue:3480", + "accept": false, + "reason": "One is about using a custom conditioning encoder; the other is a feature request for a LoRA scale parameter. Related area, but not the same change." + }, + { + "left": "issue:4969", + "right": "issue:5892", + "accept": false, + "reason": "Both mention import errors, but the actual missing symbols are unspecified vs `CrossAttention`; too vague to be the same bug." + }, + { + "left": "issue:3405", + "right": "issue:4086", + "accept": false, + "reason": "Both involve ControlNet training, but one is a general training bug and the other is an SDXL-specific training failure; not the same concrete issue." + }, + { + "left": "issue:1572", + "right": "issue:3631", + "accept": false, + "reason": "Both touch inpainting, but the first is a pipeline incompatibility and the second is an exception when passing latents; different code paths." + }, + { + "left": "issue:6548", + "right": "issue:9558", + "accept": false, + "reason": "RGBA image misdetected as latents vs black validation image in TensorBoard are unrelated symptoms." + }, + { + "left": "issue:5515", + "right": "issue:5619", + "accept": false, + "reason": "Both are LCM-related, but one is scheduler initialization and the other is AutoencodeTiny image encoding in img2img; not the same bug." + }, + { + "left": "issue:11878", + "right": "issue:13105", + "accept": false, + "reason": "Different Wan model problems: GGUF loading failure for 1.3B vs noisy outputs in Wan2.1 text-to-video." + }, + { + "left": "issue:6256", + "right": "issue:7293", + "accept": false, + "reason": "Both are feature requests, but they ask for unrelated pipeline capabilities." + }, + { + "left": "issue:6439", + "right": "issue:7119", + "accept": false, + "reason": "Both are docs link issues, but they concern different links/pages and are not clearly the same broken reference." + }, + { + "left": "issue:6870", + "right": "issue:7735", + "accept": false, + "reason": "VAE tiling missing in img2img/inpaint vs missing `caption_column` in custom dataset training are unrelated." + }, + { + "left": "issue:9208", + "right": "issue:9804", + "accept": false, + "reason": "Both are ControlNet-loading complaints, but they target different APIs (`from_single_file` vs `from_pretrained`) and different model types; not one concrete bug." + }, + { + "left": "issue:10121", + "right": "issue:11185", + "accept": false, + "reason": "Different models and errors: FLUX high-resolution generation vs CogView4 generation error." + }, + { + "left": "issue:6016", + "right": "issue:7003", + "accept": false, + "reason": "Both involve tensor shape issues, but one is in the converter and the other is in inpainting model weights; different failure points." + }, + { + "left": "issue:7538", + "right": "issue:9731", + "accept": false, + "reason": "One is a training issue on a custom Fill50k-style dataset; the other is a how-to/feature question for Playground2.5 LoRA training." + }, + { + "left": "issue:7757", + "right": "issue:7960", + "accept": false, + "reason": "Both are multi-GPU SDXL training failures, but one is a specific ControlNet script OOM and the other is a broader training failure; not enough to merge." + }, + { + "left": "issue:1315", + "right": "issue:2596", + "accept": false, + "reason": "Import-time missing symbol vs documentation link to a non-existing helper; different problem classes." + }, + { + "left": "issue:9278", + "right": "issue:9501", + "accept": false, + "reason": "Both involve FLUX DreamBooth training, but one is a multi-GPU training bug and the other is delayed model saving; different concrete symptoms." + }, + { + "left": "issue:2163", + "right": "issue:6130", + "accept": false, + "reason": "Training dtype mismatch vs LoRA loading error are different workflows and bugs." + }, + { + "left": "issue:5128", + "right": "issue:6196", + "accept": false, + "reason": "Both are loading issues, but one is local `.safetensors` via XLPipeline and the other is ControlNetXS Depth SDXL loading; different models and code paths." + }, + { + "left": "issue:1944", + "right": "issue:730", + "accept": false, + "reason": "Both are DreamBooth reports, but one is poor prompt learning on SD1.5/2.1-768 and the other is local-vs-Colab training behavior; not the same bug." + }, + { + "left": "issue:10925", + "right": "issue:11464", + "accept": false, + "reason": "Both mention FLUX dev and LoRA/finetuning, but one is a DreamBooth finetune question and the other is a ControlPipeline loading failure." + }, + { + "left": "issue:4186", + "right": "issue:6870", + "accept": false, + "reason": "Both are inpainting-related, but one is SDXL refiner incompatibility and the other is missing VAE tiling support; different underlying issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 111, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10421", + "issue:1062", + "issue:10705", + "issue:10893", + "issue:11247", + "issue:11363", + "issue:1185", + "issue:12009", + "issue:1251", + "issue:12529", + "issue:1572", + "issue:1944", + "issue:2026", + "issue:2465", + "issue:2875", + "issue:3643", + "issue:3803", + "issue:3954", + "issue:4356", + "issue:4446", + "issue:4488", + "issue:4593", + "issue:4975", + "issue:5259", + "issue:5447", + "issue:5488", + "issue:5570", + "issue:5592", + "issue:5619", + "issue:5917", + "issue:6157", + "issue:6218", + "issue:6821", + "issue:6870", + "issue:7146", + "issue:7293", + "issue:7310", + "issue:7378", + "issue:7936", + "issue:8540", + "issue:8555", + "issue:8832", + "issue:8833", + "issue:8900", + "issue:8956", + "issue:9450", + "issue:9804", + "issue:9895" + ], + "soft_pairs": [ + "issue:4356|issue:5259", + "issue:8956|issue:9804", + "issue:11363|issue:9895", + "issue:3643|issue:7378", + "issue:12009|issue:12529", + "issue:3803|issue:3954", + "issue:4446|issue:8900", + "issue:7146|issue:7310", + "issue:5488|issue:6870", + "issue:7936|issue:8833", + "issue:5570|issue:5917", + "issue:10705|issue:9450", + "issue:10893|issue:7293", + "issue:1251|issue:8832", + "issue:2875|issue:6821", + "issue:6157|issue:6218", + "issue:10421|issue:11247", + "issue:1185|issue:2465", + "issue:1572|issue:2026", + "issue:8540|issue:8555", + "issue:5447|issue:5619", + "issue:4488|issue:5592", + "issue:4593|issue:4975", + "issue:1062|issue:1944" + ], + "prepared_review_unit_hash": "003ce2aaf05091de7e1fd3051d58dba2f425f73612d10247fc30f5c4ed869c62", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34296, + "estimated_input_tokens": 8574, + "estimated_eval_tokens": 17404 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are unrelated bugs or feature requests that only share broad keywords like multi-GPU, loading, or training. None of the proposed soft edges look like the same concrete underlying issue.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue fits the set well; the items span distinct subsystems, model families, and failure modes. If forced to pick a representative, issue 10421 is the broadest multi-GPU-related report, but it still does not cover the rest.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10421 is the closest broad representative of the multi-GPU/device-placement subgroup, but the overall set is too mixed for a strong dedup canonical.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4356", + "right": "issue:5259", + "accept": false, + "reason": "Different failure modes: a missing packaged module vs a missing model_index.json in a local model directory." + }, + { + "left": "issue:8956", + "right": "issue:9804", + "accept": false, + "reason": "Both are load failures, but one is an OpenVINO pipeline model load issue and the other is a local SDXL ControlNet loading bug." + }, + { + "left": "issue:11363", + "right": "issue:9895", + "accept": false, + "reason": "OOM in a controlnet inference example is not the same bug as Flux device-mismatch after a specific commit." + }, + { + "left": "issue:3643", + "right": "issue:7378", + "accept": false, + "reason": "Both mention decoding latents, but one is an INT_MAX kernel limit and the other is a memory-usage complaint about separate decoding." + }, + { + "left": "issue:12009", + "right": "issue:12529", + "accept": false, + "reason": "Both are Wan-related, but one is single-file GGUF loading and the other is a tiled VAE tensor-size mismatch in a different path." + }, + { + "left": "issue:3803", + "right": "issue:3954", + "accept": false, + "reason": "Multi-GPU context only; one is model sharing in LoRA training, the other is incorrect scheduler step counting." + }, + { + "left": "issue:4446", + "right": "issue:8900", + "accept": false, + "reason": "Both involve from_single_file, but one is custom text_encoder support and the other is loading an SDXL refiner checkpoint." + }, + { + "left": "issue:7146", + "right": "issue:7310", + "accept": false, + "reason": "Both are SDXL training issues, but one is CUDA OOM and the other is a defaults/configuration bug in the LoRA script." + }, + { + "left": "issue:5488", + "right": "issue:6870", + "accept": false, + "reason": "A demo code failure is unrelated to missing enable_vae_tiling support in img2img/inpaint pipelines." + }, + { + "left": "issue:7936", + "right": "issue:8833", + "accept": false, + "reason": "A malformed local safetensors load is not the same as a missing config.json for an SD3 model repo." + }, + { + "left": "issue:5570", + "right": "issue:5917", + "accept": false, + "reason": "Missing compute_snr in a package release is a different issue from convert_ldm_unet_checkpoint breaking after an upgrade." + }, + { + "left": "issue:10705", + "right": "issue:9450", + "accept": false, + "reason": "Both are Flux-related, but one is a reuse/performance slowdown and the other is a multi-GPU device-placement crash." + }, + { + "left": "issue:10893", + "right": "issue:7293", + "accept": false, + "reason": "These are separate feature requests for different pipelines and models, not the same change request." + }, + { + "left": "issue:1251", + "right": "issue:8832", + "accept": false, + "reason": "RePaint on MPS returning black images is a different bug from a generic pipe(prompt) crash." + }, + { + "left": "issue:2875", + "right": "issue:6821", + "accept": false, + "reason": "Training NaNs and a caption_column validation error are unrelated training-path problems." + }, + { + "left": "issue:6157", + "right": "issue:6218", + "accept": false, + "reason": "An ip_adapter integration problem is distinct from a batched-seeds bug in ControlNet/Img2Img." + }, + { + "left": "issue:10421", + "right": "issue:11247", + "accept": false, + "reason": "Both are multi-GPU/device-placement themed, but one reports a CPU memory leak while the other is a controlnet distribution error for FluxTransformer2DModel." + }, + { + "left": "issue:1185", + "right": "issue:2465", + "accept": false, + "reason": "Reproducibility complaints vs asking for a reproducible training recipe are related topic-wise, not the same underlying bug." + }, + { + "left": "issue:1572", + "right": "issue:2026", + "accept": false, + "reason": "A pipeline failing for img2img/inpainting is not the same as cpu_offload not working in PaintByExample." + }, + { + "left": "issue:8540", + "right": "issue:8555", + "accept": false, + "reason": "One asks for caching embeddings/VAE representations in SD3 training; the other requests FP8 and text-encoder training support." + }, + { + "left": "issue:5447", + "right": "issue:5619", + "accept": false, + "reason": "Both are img2img-related, but they concern different models and different scheduler/encoding bugs." + }, + { + "left": "issue:4488", + "right": "issue:5592", + "accept": false, + "reason": "SDXL checkpoint loading without pipeline_class and UniDiffuser checkpoint loading are separate model-loading failures." + }, + { + "left": "issue:4593", + "right": "issue:4975", + "accept": false, + "reason": "Automatic1111 LoRA rank incompatibility is not the same as repeated datatype queries causing slow loading." + }, + { + "left": "issue:1062", + "right": "issue:1944", + "accept": false, + "reason": "Both mention Dreambooth reproducibility, but one is a generic inability to repeat paper results and the other is a specific prompt/result mismatch." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 112, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10139", + "issue:10671", + "issue:11133", + "issue:11160", + "issue:12612", + "issue:12699", + "issue:13258", + "issue:1434", + "issue:1478", + "issue:1572", + "issue:1672", + "issue:1976", + "issue:2364", + "issue:2439", + "issue:2480", + "issue:2659", + "issue:3101", + "issue:3428", + "issue:3812", + "issue:4255", + "issue:5577", + "issue:5593", + "issue:5630", + "issue:5647", + "issue:5687", + "issue:5881", + "issue:6069", + "issue:6250", + "issue:6297", + "issue:6540", + "issue:6870", + "issue:7243", + "issue:7405", + "issue:8359", + "issue:8909", + "issue:8939", + "issue:9026", + "issue:9052", + "issue:9069", + "issue:9164", + "issue:9208", + "issue:9227", + "issue:9237", + "issue:9278", + "issue:9613", + "issue:9667", + "issue:9768", + "issue:9911" + ], + "soft_pairs": [ + "issue:9164|issue:9227", + "issue:12612|issue:12699", + "issue:2364|issue:2480", + "issue:8359|issue:9052", + "issue:9237|issue:9278", + "issue:11133|issue:9613", + "issue:5647|issue:5687", + "issue:9026|issue:9069", + "issue:11160|issue:13258", + "issue:6297|issue:8939", + "issue:5577|issue:9208", + "issue:3101|issue:6540", + "issue:1672|issue:2659", + "issue:5593|issue:6069", + "issue:4255|issue:5881", + "issue:10671|issue:8909", + "issue:1976|issue:3812", + "issue:9768|issue:9911", + "issue:1572|issue:2439", + "issue:6250|issue:9667", + "issue:7243|issue:7405", + "issue:10139|issue:6870", + "issue:1434|issue:1478", + "issue:3428|issue:5630" + ], + "prepared_review_unit_hash": "fd2cb2415df01223e1826cb72b70c6fd1105cdaf362149b6e6aeeeacc44b6a39", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34378, + "estimated_input_tokens": 8595, + "estimated_eval_tokens": 17446 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose similarity cluster of mostly unrelated diffusers issues. The only pair that looks like a likely true duplicate is the from_single_file checkpoint-loading failure pair; the rest share broad subsystem vocabulary but not the same concrete bug.", + "confidence": 0.33, + "canonical_issue_reason": "issue:8359 is the broadest and most representative of the only credible duplicate-like subtopic in this set: single-file/custom checkpoint loading failures.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8359 is the best overall anchor because it is specific, broadly phrased, and most likely to subsume the closely related from_single_file checkpoint-loading report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9164", + "right": "issue:9227", + "accept": false, + "reason": "Same general FLUX training area, but one is a convergence complaint and the other is a runtime/model-loading problem; not the same bug." + }, + { + "left": "issue:12612", + "right": "issue:12699", + "accept": false, + "reason": "Both mention WAN, but one is NaN latents and the other is quality degradation over many prompts; different failure modes." + }, + { + "left": "issue:2364", + "right": "issue:2480", + "accept": false, + "reason": "Both relate to training scripts, but one is a generic inability to run training and the other is a missing checkpoint folder output; not the same issue." + }, + { + "left": "issue:8359", + "right": "issue:9052", + "accept": true, + "reason": "Both are from_single_file loading failures on nonstandard/custom checkpoints, pointing to the same underlying loader bug." + }, + { + "left": "issue:9237", + "right": "issue:9278", + "accept": false, + "reason": "Both are FLUX dreambooth training reports, but one is a general issue and the other is specifically multigpu; too broad to treat as one bug." + }, + { + "left": "issue:11133", + "right": "issue:13258", + "accept": false, + "reason": "Different video pipelines/models (CogVideoX vs Wan 2.2 TI2V) with different symptoms; not a shared concrete bug." + }, + { + "left": "issue:5647", + "right": "issue:5687", + "accept": false, + "reason": "Both affect img2img/inpaint strength handling, but the specific failure conditions and code paths differ." + }, + { + "left": "issue:9026", + "right": "issue:9069", + "accept": false, + "reason": "Completely different errors: one is an unexpected keyword in dispatch, the other is an ndarray type mismatch." + }, + { + "left": "issue:11160", + "right": "issue:13258", + "accept": false, + "reason": "Both are I2V-related, but they concern different models and different reported problems." + }, + { + "left": "issue:6297", + "right": "issue:8939", + "accept": false, + "reason": "Both are loading/training-related, but one is SDXL LoRA training and the other is a tokenizer load error for SD3; not the same bug." + }, + { + "left": "issue:5577", + "right": "issue:9208", + "accept": false, + "reason": "Both involve from_single_file/controlnet loading, but one is a specific missing key and the other is a broad inability to load most ControlNet models." + }, + { + "left": "issue:3101", + "right": "issue:6540", + "accept": false, + "reason": "Different training-time failures: ControlNet example error vs FSDP matmul error." + }, + { + "left": "issue:1672", + "right": "issue:2659", + "accept": false, + "reason": "Wrong image size after conversion is unrelated to batch-size-dependent output variation." + }, + { + "left": "issue:5593", + "right": "issue:6069", + "accept": false, + "reason": "Different code paths entirely: AutoPipeline.from_pipe with ControlNet vs scheduler.add_noise at t=0." + }, + { + "left": "issue:4255", + "right": "issue:5881", + "accept": false, + "reason": "Both are subjective quality complaints, not the same bug or code path." + }, + { + "left": "issue:10671", + "right": "issue:8909", + "accept": false, + "reason": "One is a deterministic dtype/device issue in DiffusionPipeline; the other is a training save problem for text_encoder_ti." + }, + { + "left": "issue:1976", + "right": "issue:3812", + "accept": false, + "reason": "Tokenizer file lookup and ControlNet config parsing are unrelated failures." + }, + { + "left": "issue:9768", + "right": "issue:9911", + "accept": false, + "reason": "Different problems: SDPA/CuDNN NaNs vs multi-ControlNet layer-length mismatch in Flux." + }, + { + "left": "issue:1572", + "right": "issue:2439", + "accept": false, + "reason": "Both mention inpainting, but one is a mega pipeline capability issue and the other is a load/run failure; not the same underlying bug." + }, + { + "left": "issue:6250", + "right": "issue:9667", + "accept": false, + "reason": "Different models and loaders: SDXL VAE single-file keyerror vs Flux fp8 local transformer loading." + }, + { + "left": "issue:7243", + "right": "issue:7405", + "accept": false, + "reason": "Separate new-model requests, not duplicate bugs." + }, + { + "left": "issue:10139", + "right": "issue:6870", + "accept": false, + "reason": "Both touch pipeline initialization/loading, but one is UNet sample size handling and the other is missing enable_vae_tiling; different fixes." + }, + { + "left": "issue:1434", + "right": "issue:1478", + "accept": false, + "reason": "Both are upscale pipeline problems, but one is OOM and the other is multi-image output support." + }, + { + "left": "issue:3428", + "right": "issue:5630", + "accept": false, + "reason": "Different img2img issues: unexpected results after changing the pipeline vs noise leftover with LMS scheduler." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The only clearly supported duplicate-like pair is the from_single_file checkpoint-loading one, and the other pair calls are mostly defensible as non-duplicates given the titles alone." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 113, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10144", + "issue:10223", + "issue:10317", + "issue:10472", + "issue:11199", + "issue:11218", + "issue:11298", + "issue:11351", + "issue:2052", + "issue:255", + "issue:2844", + "issue:3377", + "issue:3382", + "issue:3461", + "issue:3539", + "issue:3584", + "issue:3694", + "issue:3884", + "issue:3885", + "issue:4241", + "issue:4256", + "issue:4306", + "issue:4402", + "issue:4450", + "issue:4638", + "issue:4639", + "issue:4885", + "issue:5544", + "issue:5603", + "issue:6112", + "issue:6154", + "issue:6157", + "issue:6517", + "issue:6524", + "issue:6610", + "issue:7568", + "issue:7619", + "issue:8485", + "issue:8645", + "issue:8649", + "issue:8879", + "issue:8885", + "issue:9141", + "issue:9320", + "issue:9410", + "issue:9556", + "issue:9856" + ], + "soft_pairs": [ + "issue:10223|issue:10472", + "issue:11199|issue:11218", + "issue:3377|issue:3539", + "issue:5544|issue:6517", + "issue:6157|issue:6524", + "issue:8645|issue:9141", + "issue:11298|issue:8649", + "issue:2052|issue:4450", + "issue:3884|issue:4639", + "issue:10317|issue:8879", + "issue:3694|issue:8885", + "issue:3382|issue:3885", + "issue:255|issue:2844", + "issue:6154|issue:9856", + "issue:4638|issue:4885", + "issue:9320|issue:9556", + "issue:4306|issue:5544", + "issue:10144|issue:11351", + "issue:4241|issue:8485", + "issue:4256|issue:7619", + "issue:5603|issue:9410", + "issue:6112|issue:7568", + "issue:4402|issue:6610", + "issue:3461|issue:3584" + ], + "prepared_review_unit_hash": "720623ece6ca7f45e63738ef655342fb0e49e4c51269d2721d35ffabeda620a2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33979, + "estimated_input_tokens": 8495, + "estimated_eval_tokens": 17246 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly unrelated support/issues; the soft pairs share only superficial wording and do not look like the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10223", + "right": "issue:10472", + "accept": false, + "reason": "Different problems: one asks where to get a LoRA ID, the other reports a lora_bias constructor error during inference." + }, + { + "left": "issue:11199", + "right": "issue:11218", + "accept": false, + "reason": "Both are import failures, but for different modules/classes and likely different breakages." + }, + { + "left": "issue:3377", + "right": "issue:3539", + "accept": false, + "reason": "One is about SNR weighting in training, the other about main_process_first semantics; unrelated topics." + }, + { + "left": "issue:5544", + "right": "issue:6517", + "accept": false, + "reason": "FreeU disable error vs ControlNet SDXL import traceback; different code paths and symptoms." + }, + { + "left": "issue:6157", + "right": "issue:6524", + "accept": false, + "reason": "IP-Adapter in ControlNet img2img versus loading LoRA a second time; not the same defect." + }, + { + "left": "issue:8645", + "right": "issue:9141", + "accept": false, + "reason": "Missing single_file_utils helper versus missing FluxPipeline.from_single_file; different APIs." + }, + { + "left": "issue:11298", + "right": "issue:8649", + "accept": false, + "reason": "LoRA hot-swapping key error is not the same as SD3 num_images_per_prompt handling." + }, + { + "left": "issue:2052", + "right": "issue:4450", + "accept": false, + "reason": "Textual inversion resume-from-checkpoint issue differs from multi-text inversion not working." + }, + { + "left": "issue:3884", + "right": "issue:4639", + "accept": false, + "reason": "train_text_to_image_lora.py error and multiprocessing dataloader error are separate training failures." + }, + { + "left": "issue:10317", + "right": "issue:8879", + "accept": false, + "reason": "Both involve torch.compile, but one is StableVideoDiffusionPipeline with changing images and the other is StableDiffusionXL with cudagraphs." + }, + { + "left": "issue:3694", + "right": "issue:8885", + "accept": false, + "reason": "Pokemon finetuning file/config issue vs clip-retrieval/modifier-token setup problem; unrelated." + }, + { + "left": "issue:3382", + "right": "issue:3885", + "accept": false, + "reason": "Multi-GPU OOM in text-to-image training is different from general VRAM reduction question." + }, + { + "left": "issue:255", + "right": "issue:2844", + "accept": false, + "reason": "Both runtime errors in pipelines, but one is image-size validation in StableDiffusionPipeline and the other is explicit image embeddings in StableUnCLIPImg2ImgPipeline." + }, + { + "left": "issue:6154", + "right": "issue:9856", + "accept": false, + "reason": "M2 Mac LoRA training issue is unrelated to distributed port conflict." + }, + { + "left": "issue:4638", + "right": "issue:4885", + "accept": false, + "reason": "SDXL textual inversion loading problem versus SDXL ControlNet img2img LoRA loading problem; different loaders." + }, + { + "left": "issue:9320", + "right": "issue:9556", + "accept": false, + "reason": "Saved/loaded custom diffusion models and rslora saved-model issues concern different training artifacts." + }, + { + "left": "issue:4306", + "right": "issue:5544", + "accept": false, + "reason": "ReferencePipeline list index error is unrelated to FreeU toggle behavior in SDXL." + }, + { + "left": "issue:10144", + "right": "issue:11351", + "accept": false, + "reason": "Mochi video quality complaint is unrelated to Wan i2v processor dtype behavior." + }, + { + "left": "issue:4241", + "right": "issue:8485", + "accept": false, + "reason": "General question about unidiffusers LoRA support versus SD3 LoRA support request." + }, + { + "left": "issue:4256", + "right": "issue:7619", + "accept": false, + "reason": "SDXL Dreambooth training error and Dreambooth number-of-channels error are both training-related but not the same concrete bug." + }, + { + "left": "issue:5603", + "right": "issue:9410", + "accept": false, + "reason": "Different missing imports: utility function vs ControlNet inpainting pipeline class." + }, + { + "left": "issue:6112", + "right": "issue:7568", + "accept": false, + "reason": "SDXL turbo noise generation issue is not the same as prompt-tokens interpolation discussion." + }, + { + "left": "issue:4402", + "right": "issue:6610", + "accept": false, + "reason": "Single-file SDXL inpaint usage issue differs from num_in_channels being ignored in the same pipeline family." + }, + { + "left": "issue:3461", + "right": "issue:3584", + "accept": false, + "reason": "DreamBooth performance comparison and tutorial failure are distinct problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 114, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10302", + "issue:10608", + "issue:10795", + "issue:10972", + "issue:1100", + "issue:11014", + "issue:11133", + "issue:11247", + "issue:11642", + "issue:11702", + "issue:11989", + "issue:1363", + "issue:1614", + "issue:1709", + "issue:2089", + "issue:2359", + "issue:2495", + "issue:2603", + "issue:3222", + "issue:3631", + "issue:4241", + "issue:4788", + "issue:5254", + "issue:5540", + "issue:5670", + "issue:6107", + "issue:6176", + "issue:7049", + "issue:7159", + "issue:7221", + "issue:7431", + "issue:7441", + "issue:7616", + "issue:7682", + "issue:8511", + "issue:8556", + "issue:8644", + "issue:8673", + "issue:8708", + "issue:8828", + "issue:9187", + "issue:9237", + "issue:9514", + "issue:962", + "issue:9637", + "issue:9667", + "issue:9861", + "issue:9895" + ], + "soft_pairs": [ + "issue:3222|issue:6176", + "issue:6107|issue:7049", + "issue:11014|issue:2089", + "issue:7431|issue:9237", + "issue:11247|issue:7682", + "issue:8556|issue:8644", + "issue:10795|issue:7159", + "issue:5254|issue:962", + "issue:10972|issue:11702", + "issue:4788|issue:5540", + "issue:8673|issue:8708", + "issue:2603|issue:5670", + "issue:1100|issue:1709", + "issue:10302|issue:9895", + "issue:1614|issue:2495", + "issue:2359|issue:4241", + "issue:11133|issue:9637", + "issue:10608|issue:7441", + "issue:1363|issue:3631", + "issue:7221|issue:7616", + "issue:8511|issue:8828", + "issue:9187|issue:9861", + "issue:9514|issue:9667", + "issue:11642|issue:11989" + ], + "prepared_review_unit_hash": "757feb029606b016f9fa44d0173e696bae5574164ed19d2c62984222e88b7bcf", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34079, + "estimated_input_tokens": 8520, + "estimated_eval_tokens": 17296 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most pairs only share superficial topic words (LoRA, FP8, OOM, multi-GPU, black image, etc.) and do not look like the same underlying bug or change.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 5540 is the broadest, most generic report in the set (\u201cInference out of memory\u201d), so it is the closest thing to a representative bucket, though the cluster is not a single duplicate family.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 5540 is the least specific and most reusable as a placeholder canonical issue, but it is still far too generic to truly cover the other reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3222", + "right": "issue:6176", + "accept": false, + "reason": "Both involve LoRA weight loading, but the contexts differ (custom_diffusion training output vs dreambooth_lora loading), so they are not clearly the same bug." + }, + { + "left": "issue:6107", + "right": "issue:7049", + "accept": false, + "reason": "Both mention IP Adapter/Stable Diffusion, but one is environment-specific EC2 behavior and the other is a concrete tensor-unpack error in the basic example." + }, + { + "left": "issue:11014", + "right": "issue:2089", + "accept": false, + "reason": "Both are OOM reports, but one is inference on Colab A100 and the other is training DreamBooth AltDiffusion; different code paths and causes." + }, + { + "left": "issue:7431", + "right": "issue:9237", + "accept": false, + "reason": "Both are training-related LoRA reports, but one is DoRA with text encoder and the other is Flux Dreambooth LoRA; not the same underlying failure." + }, + { + "left": "issue:11247", + "right": "issue:7682", + "accept": false, + "reason": "Both concern multiple GPUs, but the model/pipeline and failure mode differ substantially (Flux+ControlNet distribution vs SD1.5 on weak GPUs)." + }, + { + "left": "issue:8556", + "right": "issue:8644", + "accept": false, + "reason": "Both mention meta-tensor/offload style errors, but one is SD3 from_single_file loading and the other is separately loading T5 with sequential/model CPU offload; too context-specific to merge." + }, + { + "left": "issue:10795", + "right": "issue:7159", + "accept": false, + "reason": "Both involve torch.compile, but the failures are on different model-loading paths (GGUF via from_single_file vs fused LoRA); not the same bug." + }, + { + "left": "issue:5254", + "right": "issue:962", + "accept": false, + "reason": "Both are inpaint-related, but one is multi-ControlNet support and the other is size limitation in new inpaint pipelines; different issues." + }, + { + "left": "issue:10972", + "right": "issue:11702", + "accept": false, + "reason": "Both are LoRA loading problems, but they involve different formats and error surfaces (OneTrainer Flux LoRAs vs AccVid LoRA key error)." + }, + { + "left": "issue:4788", + "right": "issue:5540", + "accept": false, + "reason": "Both are OOM-related, but one is SDXL inpaint on 6 GB VRAM and the other is a generic inference OOM report; not the same concrete bug." + }, + { + "left": "issue:8673", + "right": "issue:8708", + "accept": false, + "reason": "Both are SD3 issues, but one is missing attention masks and the other is a missing shift_factor config in training examples; unrelated fixes." + }, + { + "left": "issue:2603", + "right": "issue:5670", + "accept": false, + "reason": "Both concern non-square image generation, but they hit different pipelines and likely different shape-handling code paths." + }, + { + "left": "issue:1100", + "right": "issue:1709", + "accept": false, + "reason": "The errors are completely different (`device` attribute vs `astype` on Tensor), so these are not duplicates." + }, + { + "left": "issue:10302", + "right": "issue:9895", + "accept": false, + "reason": "One is FP8 inference noise without CPU offloading; the other is a device-mismatch regression after a commit. Different bugs." + }, + { + "left": "issue:1614", + "right": "issue:2495", + "accept": false, + "reason": "Both produce black images, but one is Stable Diffusion 2.1 with autocast and the other is MPS upscale pipeline; too broad to treat as the same bug." + }, + { + "left": "issue:2359", + "right": "issue:4241", + "accept": false, + "reason": "These are unrelated feature requests/issues: latent output_type in img2img versus LoRA fine-tuning support for unidiffusers." + }, + { + "left": "issue:11133", + "right": "issue:9637", + "accept": false, + "reason": "Different subsystems and failures: CogVideoX image-to-video bug versus XFormers attention-mask/bfloat handling." + }, + { + "left": "issue:10608", + "right": "issue:7441", + "accept": false, + "reason": "Multi-GPU pipeline failure and pytest on Python 3.12 are unrelated." + }, + { + "left": "issue:1363", + "right": "issue:3631", + "accept": false, + "reason": "ONNX LMS scheduler error and inpainting-with-latents exception are different pipeline bugs." + }, + { + "left": "issue:7221", + "right": "issue:7616", + "accept": false, + "reason": "Both mention noisy/artefactual output, but one is generic noise with a specific model and the other is low-denoise artifacts in img2img; not enough to merge." + }, + { + "left": "issue:8511", + "right": "issue:8828", + "accept": false, + "reason": "One is about text encoders not being unloaded during SD3 training scripts; the other is a general SD3 training error due to limited memory." + }, + { + "left": "issue:9187", + "right": "issue:9861", + "accept": false, + "reason": "IP Adapter/SDXL image-encoder handling and Flux training not updating the transformer are different code paths and symptoms." + }, + { + "left": "issue:9514", + "right": "issue:9667", + "accept": false, + "reason": "Both are Flux fp8-related, but one is LoRA loading after quantization and the other is loading a local transformer file; too different to be the same underlying bug." + }, + { + "left": "issue:11642", + "right": "issue:11989", + "accept": false, + "reason": "Prompt scaling having no effect and black inference images are unrelated symptoms with no clear shared code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 115, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10025", + "issue:10712", + "issue:10797", + "issue:11208", + "issue:12535", + "issue:12565", + "issue:13281", + "issue:13488", + "issue:1500", + "issue:2161", + "issue:2554", + "issue:2704", + "issue:2780", + "issue:3065", + "issue:3097", + "issue:3238", + "issue:3468", + "issue:3761", + "issue:3764", + "issue:3915", + "issue:4094", + "issue:4282", + "issue:4520", + "issue:460", + "issue:5399", + "issue:5429", + "issue:5565", + "issue:5873", + "issue:5932", + "issue:6096", + "issue:6196", + "issue:6250", + "issue:6283", + "issue:6685", + "issue:6811", + "issue:7009", + "issue:7168", + "issue:7210", + "issue:7480", + "issue:7982", + "issue:8540", + "issue:8788", + "issue:8886", + "issue:9306", + "issue:9343", + "issue:9497", + "issue:9803", + "issue:9895" + ], + "soft_pairs": [ + "issue:13281|issue:13488", + "issue:3764|issue:6096", + "issue:3097|issue:7480", + "issue:6283|issue:7168", + "issue:2704|issue:4282", + "issue:2780|issue:3238", + "issue:1500|issue:3761", + "issue:3468|issue:5565", + "issue:10712|issue:7210", + "issue:3065|issue:4094", + "issue:6196|issue:6250", + "issue:10025|issue:8886", + "issue:6811|issue:7009", + "issue:9306|issue:9895", + "issue:2161|issue:5399", + "issue:3915|issue:4520", + "issue:11208|issue:8788", + "issue:2554|issue:460", + "issue:10797|issue:7982", + "issue:5429|issue:8540", + "issue:5873|issue:6685", + "issue:12535|issue:12565", + "issue:9343|issue:9803", + "issue:5932|issue:9497" + ], + "prepared_review_unit_hash": "dc0ff4830ce6cf3cd29f1ce5eec357ff101696c6a9bfdfe41f265458d29de016", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34282, + "estimated_input_tokens": 8571, + "estimated_eval_tokens": 17398 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is highly heterogeneous: most pairs share only a broad subsystem or keyword, not the same concrete bug/change. No PRs are present. The strongest single issue candidate is a broad Flux loading regression, but it is not a clear duplicate anchor for the whole set.", + "confidence": 0.59, + "canonical_issue_reason": "issue:9343 is the broadest open Flux loading failure report and is the closest fit among the Flux-related items.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9343 is the best standalone issue candidate, but the overall collection is too mixed to treat as one duplicate cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13281", + "right": "issue:13488", + "accept": false, + "reason": "Both concern offloading, but one is a quantized Qwen image/device-mismatch case and the other is a diffusers release regression; not the same bug." + }, + { + "left": "issue:3764", + "right": "issue:6096", + "accept": false, + "reason": "Both mention torch.compile, but they fail in different pipelines with different exceptions and likely different root causes." + }, + { + "left": "issue:3097", + "right": "issue:7480", + "accept": false, + "reason": "Textual inversion loading vs DeepFloyd OOM are unrelated failure modes." + }, + { + "left": "issue:6283", + "right": "issue:7168", + "accept": false, + "reason": "ControlNet img2img strength=0 behavior is unrelated to IP-Adapter image embed tensor shape." + }, + { + "left": "issue:2704", + "right": "issue:4282", + "accept": false, + "reason": "One is a request for batch processing, the other is adding Multi-ControlNet support; related area but not the same change." + }, + { + "left": "issue:2780", + "right": "issue:3238", + "accept": false, + "reason": "FP16 gradient unscale error in LoRA training is unrelated to LoRA safetensor conversion." + }, + { + "left": "issue:1500", + "right": "issue:3761", + "accept": false, + "reason": "Community pipeline API compatibility vs loading two community pipelines in one process are different issues." + }, + { + "left": "issue:3468", + "right": "issue:5565", + "accept": false, + "reason": "Missing model_index.json generation and failing to run an SDXL DreamBooth script are different DreamBooth problems." + }, + { + "left": "issue:10712", + "right": "issue:7210", + "accept": false, + "reason": "SD3 prompt_embeds runtime error and missing named_parameters on an SDXL pipeline under accelerate are not the same bug." + }, + { + "left": "issue:3065", + "right": "issue:4094", + "accept": false, + "reason": "DreamBooth LoRA text-encoder tuning request vs adding a T2I adapter training script are different feature requests." + }, + { + "left": "issue:6196", + "right": "issue:6250", + "accept": false, + "reason": "ControlNetXS Depth loading and SDXL VAE single-file KeyError are separate loading failures." + }, + { + "left": "issue:10025", + "right": "issue:8886", + "accept": false, + "reason": "Both mention attention masks, but they affect different models and code paths (Flux transformer vs Auraflow SDPA)." + }, + { + "left": "issue:6811", + "right": "issue:7009", + "accept": false, + "reason": "Missing model_index.json for i2vgen-xl vs poor results from a text-to-video pipeline are not the same issue." + }, + { + "left": "issue:9306", + "right": "issue:9895", + "accept": false, + "reason": "Different models and different device-selection failures; not the same regression." + }, + { + "left": "issue:2161", + "right": "issue:5399", + "accept": false, + "reason": "LoRA training on inpainting models and dtype mismatch during custom inpainting inference are distinct bugs." + }, + { + "left": "issue:3915", + "right": "issue:4520", + "accept": false, + "reason": "torch.compile on text-to-video pipelines is unrelated to LoRA scale behavior with attention slicing." + }, + { + "left": "issue:11208", + "right": "issue:8788", + "accept": false, + "reason": "Multi-ControlNet support in SD3 inpainting and SD3 VAE tiling on non-power-of-two images are unrelated." + }, + { + "left": "issue:2554", + "right": "issue:460", + "accept": false, + "reason": "Accelerate version requirement and ONNXRuntime installation mismatch are unrelated environment issues." + }, + { + "left": "issue:10797", + "right": "issue:7982", + "accept": false, + "reason": "Group offloading on HunyuanVideo with use_stream and from_pipe with MultiControlNetModel are different failures." + }, + { + "left": "issue:5429", + "right": "issue:8540", + "accept": false, + "reason": "Supporting models without a VAE and SD3 DreamBooth embedding/VAE caching are not the same bug or change." + }, + { + "left": "issue:5873", + "right": "issue:6685", + "accept": false, + "reason": "ONNX export of scaled_dot_product_attention is unrelated to DreamBooth output_dir path handling." + }, + { + "left": "issue:12535", + "right": "issue:12565", + "accept": false, + "reason": "Both are Wan i2v issues, but they concern different model versions and symptoms." + }, + { + "left": "issue:9343", + "right": "issue:9803", + "accept": false, + "reason": "Both are Flux-related loading problems, but one is a mismatched-size/low_cpu_mem_usage load error and the other is LoRA loading failure; not clearly the same bug." + }, + { + "left": "issue:5932", + "right": "issue:9497", + "accept": false, + "reason": "Channel-count mismatch in inference and a Flux DreamBooth mat2 shape error are different underlying errors." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 116, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10165", + "issue:10674", + "issue:10855", + "issue:10952", + "issue:10969", + "issue:11247", + "issue:11298", + "issue:11370", + "issue:1539", + "issue:2412", + "issue:2459", + "issue:2674", + "issue:3431", + "issue:3487", + "issue:3566", + "issue:3761", + "issue:4256", + "issue:4324", + "issue:4398", + "issue:4419", + "issue:4436", + "issue:4468", + "issue:4578", + "issue:4806", + "issue:5259", + "issue:5502", + "issue:5607", + "issue:5689", + "issue:5886", + "issue:6008", + "issue:6107", + "issue:6382", + "issue:6482", + "issue:6520", + "issue:6914", + "issue:7310", + "issue:7782", + "issue:7936", + "issue:8405", + "issue:8440", + "issue:8507", + "issue:861", + "issue:8626", + "issue:8667", + "issue:9023", + "issue:9187", + "issue:9362", + "issue:9546" + ], + "soft_pairs": [ + "issue:3761|issue:861", + "issue:4419|issue:4578", + "issue:4468|issue:9023", + "issue:5259|issue:7936", + "issue:6382|issue:8626", + "issue:1539|issue:4806", + "issue:3487|issue:3566", + "issue:4324|issue:5502", + "issue:5886|issue:6914", + "issue:2459|issue:2674", + "issue:2412|issue:4256", + "issue:6482|issue:8507", + "issue:5607|issue:9362", + "issue:10855|issue:8440", + "issue:10969|issue:9546", + "issue:4436|issue:7782", + "issue:10674|issue:9187", + "issue:4398|issue:5689", + "issue:10165|issue:10952", + "issue:6520|issue:7310", + "issue:8405|issue:8667", + "issue:3431|issue:3487", + "issue:6008|issue:6107", + "issue:11247|issue:11298", + "issue:10674|issue:11370" + ], + "prepared_review_unit_hash": "e148e943b8b0ffc4e4eb72ea39a3268a44ee1d6d52271503fb237a9b1c5d19b3", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34515, + "estimated_input_tokens": 8629, + "estimated_eval_tokens": 17514 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly unrelated issue reports; none of the soft pairs look like true duplicates of the same underlying bug/change. The strongest standalone issue in the set is the active, concrete LoRA hotswapping PEFT-key-error report.", + "confidence": 0.84, + "canonical_issue_reason": "issue:11298 is the best canonical candidate: it is open, detailed, and describes a specific reproducible LoRA hotswapping failure with active discussion/inbound attention.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11298 is the strongest issue to anchor triage because it has the clearest concrete bug path and the most active discussion among the listed issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3761", + "right": "issue:861", + "accept": false, + "reason": "Different subsystems and failure modes: community pipeline loading vs ckpt-to-diffusers conversion." + }, + { + "left": "issue:4419", + "right": "issue:4578", + "accept": false, + "reason": "One is an inference behavior bug, the other is a documentation complaint about VAE usage." + }, + { + "left": "issue:4468", + "right": "issue:9023", + "accept": false, + "reason": "Both are sample-code related, but one is multi-GPU performance disparity and the other is a runtime failure." + }, + { + "left": "issue:5259", + "right": "issue:7936", + "accept": false, + "reason": "Different load/convert errors: missing model_index.json versus invalid local safetensors JSON format." + }, + { + "left": "issue:6382", + "right": "issue:8626", + "accept": false, + "reason": "One is a training/inference key error with IP-Adapter + LoRA; the other is a guidance/docs request about multiple IP-Adapter images/masks." + }, + { + "left": "issue:1539", + "right": "issue:4806", + "accept": false, + "reason": "Both mention parameter typing, but they affect different APIs and code paths, so they are not the same bug." + }, + { + "left": "issue:3487", + "right": "issue:3566", + "accept": false, + "reason": "Both are poor-result reports, but they involve different pipelines and scheduler/controlnet contexts." + }, + { + "left": "issue:4324", + "right": "issue:5502", + "accept": false, + "reason": "A forward-pass tensor/NoneType error is unrelated to fp16 support in a custom diffusion training script." + }, + { + "left": "issue:5886", + "right": "issue:6914", + "accept": false, + "reason": "One is a feature request for face-model support; the other is an offload-related FaceID runtime bug." + }, + { + "left": "issue:2459", + "right": "issue:2674", + "accept": false, + "reason": "Different LoRA setup issues: xformers integration in training versus missing safetensors requirement." + }, + { + "left": "issue:2412", + "right": "issue:4256", + "accept": false, + "reason": "Generic dreambooth errors, but they are different training setups and not clearly the same failure." + }, + { + "left": "issue:6482", + "right": "issue:8507", + "accept": false, + "reason": "Checkpoint resume FP16 unscale failure is unrelated to SD3 text-encoder training failure." + }, + { + "left": "issue:5607", + "right": "issue:9362", + "accept": false, + "reason": "Different tensor/index errors with no clear shared code path or root cause." + }, + { + "left": "issue:10855", + "right": "issue:8440", + "accept": false, + "reason": "Completely different topics: optimizer stepping under gradient accumulation vs a mixin subclass ValueError." + }, + { + "left": "issue:10969", + "right": "issue:9546", + "accept": false, + "reason": "Both are Flux controlnet training issues, but one is a weight-shape failure and the other is validation OOM; not the same bug." + }, + { + "left": "issue:4436", + "right": "issue:7782", + "accept": false, + "reason": "Both concern conversion/loading, but one is CIVITAI-with-VAE conversion and the other is an invalid safetensors pickle/load error." + }, + { + "left": "issue:10674", + "right": "issue:9187", + "accept": false, + "reason": "GGUF/FluxPipeline incompatibility is unrelated to SDXL IP-Adapter breakage after Kolors update." + }, + { + "left": "issue:4398", + "right": "issue:5689", + "accept": false, + "reason": "Both affect SDXL image quality, but one is all-black outputs in LoRA Dreambooth and the other is residual noise with DPM++ sampling." + }, + { + "left": "issue:10165", + "right": "issue:10952", + "accept": false, + "reason": "LoRA seed preservation on scale change is a different problem from adapter-name lookup failure when loading LoRA." + }, + { + "left": "issue:6520", + "right": "issue:7310", + "accept": false, + "reason": "Different training script failures: blank validation output versus defaults not working at all." + }, + { + "left": "issue:8405", + "right": "issue:8667", + "accept": false, + "reason": "Both are controlnet tensor issues, but one is multi-controlnet sizing and the other is non-power-of-two image sizing in SD3." + }, + { + "left": "issue:3431", + "right": "issue:3487", + "accept": false, + "reason": "Both are inpainting/controlnet quality problems, but the described pipelines and schedulers differ." + }, + { + "left": "issue:6008", + "right": "issue:6107", + "accept": false, + "reason": "IP-Adapter face/LCM-LoRA incompatibility is unrelated to environment-specific Colab-vs-EC2 runtime behavior." + }, + { + "left": "issue:11247", + "right": "issue:11298", + "accept": false, + "reason": "Both mention Flux/controlnet/LoRA adjacent areas, but the failures are different: multi-GPU transformer distribution versus LoRA hotswapping PEFT keys." + }, + { + "left": "issue:10674", + "right": "issue:11370", + "accept": false, + "reason": "Flux GGUF support problem is unrelated to a video conversion script failing to unpickle a file." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 117, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11199", + "issue:1125", + "issue:11343", + "issue:12052", + "issue:1251", + "issue:1524", + "issue:1536", + "issue:1614", + "issue:1616", + "issue:1925", + "issue:2156", + "issue:2609", + "issue:3045", + "issue:307", + "issue:3154", + "issue:3223", + "issue:3346", + "issue:3540", + "issue:3613", + "issue:4080", + "issue:4109", + "issue:4185", + "issue:4306", + "issue:4460", + "issue:4488", + "issue:4504", + "issue:4641", + "issue:4667", + "issue:5152", + "issue:5524", + "issue:5621", + "issue:5630", + "issue:5703", + "issue:5886", + "issue:5889", + "issue:5997", + "issue:6382", + "issue:6460", + "issue:7053", + "issue:7421", + "issue:7483", + "issue:8440", + "issue:8469", + "issue:8546", + "issue:8579", + "issue:8873", + "issue:9515", + "issue:9895" + ], + "soft_pairs": [ + "issue:1251|issue:1614", + "issue:307|issue:4080", + "issue:4306|issue:6382", + "issue:4488|issue:5621", + "issue:4504|issue:4641", + "issue:5152|issue:5524", + "issue:5886|issue:8469", + "issue:5889|issue:6460", + "issue:8440|issue:8873", + "issue:4185|issue:5997", + "issue:3045|issue:7483", + "issue:12052|issue:9895", + "issue:4109|issue:7421", + "issue:8546|issue:8579", + "issue:11199|issue:11343", + "issue:2609|issue:4460", + "issue:3154|issue:3223", + "issue:3540|issue:4667", + "issue:2156|issue:3613", + "issue:1125|issue:1925", + "issue:5630|issue:5703", + "issue:7053|issue:9515", + "issue:1524|issue:3346", + "issue:1536|issue:1616" + ], + "prepared_review_unit_hash": "63aedb876a1a583182644ef194e99212b3a79c8cc1e31b93a6b77918d1de28ea", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34022, + "estimated_input_tokens": 8506, + "estimated_eval_tokens": 17268 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous. The only strong duplicate-looking pair is the LPW pipeline compatibility issues; the rest are different bugs, features, or model-specific failures.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 1536 is the best representative because it states the core LPW pipeline compatibility problem in a general way; issue 1616 reads like the version-specific sibling of the same bug.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1536 is the clearest, most general duplicate target in the set and best captures the shared LPW/diffusers incompatibility theme.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1251", + "right": "issue:1614", + "accept": false, + "reason": "Both mention black images, but one is RePaint on MPS and the other is SD 2.1 autocast; different pipelines and likely different root causes." + }, + { + "left": "issue:307", + "right": "issue:4080", + "accept": false, + "reason": "One is a broad request to merge pipelines; the other asks for an SDXL inpainting pipeline. Not the same bug/change." + }, + { + "left": "issue:4306", + "right": "issue:6382", + "accept": false, + "reason": "Different failures: reference pipeline list-index error vs IP-Adapter key error in an SDXL/LCM training workflow." + }, + { + "left": "issue:4488", + "right": "issue:5621", + "accept": false, + "reason": "Both involve SDXL support, but one is checkpoint loading via from_original_ckpt and the other is ControlNet encoder_hid_proj support; distinct issues." + }, + { + "left": "issue:4504", + "right": "issue:4641", + "accept": false, + "reason": "Both involve LoRA loading, but one is a safetensor state_dict assignment bug and the other is local-file load_lora_weights behavior." + }, + { + "left": "issue:5152", + "right": "issue:5524", + "accept": false, + "reason": "Same model family, but one is a general Wurstchen alpha bug and the other is a specific lora_layer attribute error." + }, + { + "left": "issue:5886", + "right": "issue:8469", + "accept": false, + "reason": "Feature request for face-model support vs a report of corrupt faces from an existing face model; not the same underlying change." + }, + { + "left": "issue:5889", + "right": "issue:6460", + "accept": false, + "reason": "Unrelated topics: stable video diffusion vs loading stable-diffusion-inpainting hanging." + }, + { + "left": "issue:8440", + "right": "issue:8873", + "accept": false, + "reason": "One is a mixin/subclass ValueError; the other is xformers performance regression. Different code paths." + }, + { + "left": "issue:4185", + "right": "issue:5997", + "accept": false, + "reason": "Black-image inference bug with SDXL ControlNet vs a tutorial/model-type mismatch error; not the same bug." + }, + { + "left": "issue:3045", + "right": "issue:7483", + "accept": false, + "reason": "Both are LoRA-to-Diffusers conversion related, but one is a general conversion request and the other is an SDXL-base-specific failure." + }, + { + "left": "issue:12052", + "right": "issue:9895", + "accept": false, + "reason": "Both are device-mismatch/offloading failures, but they affect different pipelines/models and likely different fixes." + }, + { + "left": "issue:4109", + "right": "issue:7421", + "accept": false, + "reason": "Both mention SDXL pipelines, but one is an inpaint pipeline failure and the other is a seed/reproducibility issue in K-Diffusion." + }, + { + "left": "issue:8546", + "right": "issue:8579", + "accept": false, + "reason": "One is missing from-single-file support for SD3; the other is broken SD3 LoRA support. Different features." + }, + { + "left": "issue:11199", + "right": "issue:11343", + "accept": false, + "reason": "Both are import problems, but one is a specific onnx_utils import and the other is a broad post-transformers-update import failure." + }, + { + "left": "issue:2609", + "right": "issue:4460", + "accept": false, + "reason": "Canny ControlNet checkpoint request vs a mismatch in outputs between stable diffusion and diffusers; unrelated." + }, + { + "left": "issue:3154", + "right": "issue:3223", + "accept": false, + "reason": "Both are memory/offloading complaints, but one is disk offloading when RAM is full and the other is RAM not clearing after moving to CUDA." + }, + { + "left": "issue:3540", + "right": "issue:4667", + "accept": false, + "reason": "Different APIs and failures: prompt_embeds support in TextToVideoZeroPipeline vs from_single_file loading regression." + }, + { + "left": "issue:2156", + "right": "issue:3613", + "accept": false, + "reason": "One reports CPU-only training; the other reports GPU OOM in Dreambooth. Opposite symptoms, different issues." + }, + { + "left": "issue:1125", + "right": "issue:1925", + "accept": false, + "reason": "Both are inpainting/half-precision related, but one is a runtime error in a specific mega pipeline and the other is SD2 inpainting half-precision not working." + }, + { + "left": "issue:5630", + "right": "issue:5703", + "accept": false, + "reason": "Both concern sampling/output quality, but one is LMS scheduler noise in SDXL img2img and the other is DDIM inverse noise reconstruction." + }, + { + "left": "issue:7053", + "right": "issue:9515", + "accept": false, + "reason": "Different model internals: class conditioning in UNet3DConditionModel vs missing RMSNorm in SD3Transformer2DModel attention." + }, + { + "left": "issue:1524", + "right": "issue:3346", + "accept": false, + "reason": "Textual inversion training quality vs Dreambooth resume_from_checkpoint support; unrelated training concerns." + }, + { + "left": "issue:1536", + "right": "issue:1616", + "accept": true, + "reason": "Both describe the LPW pipeline breaking against newer diffusers versions; same underlying compatibility bug and plausible one-PR fix." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative on the soft-edge pairs. The LPW pair (1536/1616) is the only clearly duplicate-looking match from the titles, and the other rejections do not overclaim sameness based on the available evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 118, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10392", + "issue:10421", + "issue:10653", + "issue:11135", + "issue:11182", + "issue:11418", + "issue:11702", + "issue:12011", + "issue:12052", + "issue:12499", + "issue:12529", + "issue:12657", + "issue:13001", + "issue:13140", + "issue:1717", + "issue:2185", + "issue:2324", + "issue:2659", + "issue:3296", + "issue:3454", + "issue:3651", + "issue:3826", + "issue:3871", + "issue:4194", + "issue:4303", + "issue:4306", + "issue:4340", + "issue:4389", + "issue:4664", + "issue:4697", + "issue:4852", + "issue:5016", + "issue:5150", + "issue:5153", + "issue:5442", + "issue:6196", + "issue:6821", + "issue:6949", + "issue:7133", + "issue:7431", + "issue:761", + "issue:7793", + "issue:7851", + "issue:8885", + "issue:9187", + "issue:9421", + "issue:9527" + ], + "soft_pairs": [ + "issue:2324|issue:761", + "issue:8885|issue:9421", + "issue:4389|issue:6196", + "issue:4697|issue:9187", + "issue:1717|issue:4303", + "issue:4340|issue:5016", + "issue:4852|issue:7793", + "issue:3296|issue:4664", + "issue:4306|issue:7851", + "issue:11702|issue:13001", + "issue:10421|issue:10653", + "issue:10392|issue:11418", + "issue:12657|issue:13140", + "issue:3454|issue:3651", + "issue:11182|issue:9527", + "issue:5153|issue:5442", + "issue:2185|issue:3871", + "issue:7133|issue:7431", + "issue:11135|issue:3826", + "issue:2659|issue:6949", + "issue:12499|issue:12529", + "issue:3871|issue:6821", + "issue:12011|issue:12052", + "issue:4194|issue:5150" + ], + "prepared_review_unit_hash": "b2904344c1c680583dfbde321da03bb727581f5232343c156e3eebf45e8200cf", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33233, + "estimated_input_tokens": 8309, + "estimated_eval_tokens": 16874 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous; the soft-similarity pairs are mostly keyword matches around training, loading, LoRA, or device movement, but they do not describe the same underlying bug/change. I would reject all candidate duplicate links.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:10421 is the broadest and clearest representative of a concrete runtime/device-management bug in the set, but the cluster is too mixed to treat it as a true duplicate anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2324", + "right": "issue:761", + "accept": false, + "reason": "Both are training-related, but one is about Dreambooth instance data loading and the other about 8-bit Adam optimizer failure; different bug surfaces." + }, + { + "left": "issue:8885", + "right": "issue:9421", + "accept": false, + "reason": "One is a Custom Diffusion/clip-retrieval token-loading setup issue, the other is model access permissions for an inpainting repo; unrelated failures." + }, + { + "left": "issue:4389", + "right": "issue:6196", + "accept": false, + "reason": "Both mention ControlNet loading, but they concern different model classes and error conditions (ControlNetModel vs ControlNetXS Depth SDXL)." + }, + { + "left": "issue:4697", + "right": "issue:9187", + "accept": false, + "reason": "Different components and regressions: specific LoRA loading after 0.20.0 vs IP-Adapter SDXL breakage after Kolors update." + }, + { + "left": "issue:1717", + "right": "issue:4303", + "accept": false, + "reason": "Offline cache loading failure and Flax SDXL crash are unrelated code paths and error modes." + }, + { + "left": "issue:4340", + "right": "issue:5016", + "accept": false, + "reason": "Both involve SDXL text encoders, but they are separate training scripts and different failure reports; not the same concrete bug." + }, + { + "left": "issue:4852", + "right": "issue:7793", + "accept": false, + "reason": "One is a LoRA loading error, the other is an unload/cleanup error for DeepFloyd LoRA weights; opposite lifecycle stages and different bugs." + }, + { + "left": "issue:3296", + "right": "issue:4664", + "accept": false, + "reason": "Checkpoint save/load with text_encoder vs SDXL LoRA loading on two pipeline class names; not the same issue." + }, + { + "left": "issue:4306", + "right": "issue:7851", + "accept": false, + "reason": "Reference pipeline list-index error and concurrent-call out-of-bounds error are distinct bugs." + }, + { + "left": "issue:11702", + "right": "issue:13001", + "accept": false, + "reason": "AccVid LoRA key mismatch and Flux2.Klein GGUF shape mismatch are different models and different loading failures." + }, + { + "left": "issue:10421", + "right": "issue:10653", + "accept": false, + "reason": "Both are device-movement issues, but one reports a CPU memory leak moving pipelines across GPUs and the other is a bitsandbytes .to(cpu) unsupported error; not the same bug." + }, + { + "left": "issue:10392", + "right": "issue:11418", + "accept": false, + "reason": "Flux fp8 model usage versus adding a Flux fill fp8 safetensors file; similar model family only, not the same failure." + }, + { + "left": "issue:12657", + "right": "issue:13140", + "accept": false, + "reason": "Qwen image-edit quantization regression and Qwen Edit pipeline NoneType error are different symptoms and likely different code paths." + }, + { + "left": "issue:3454", + "right": "issue:3651", + "accept": false, + "reason": "train_dreambooth_lora.py regression on latest and multi-GPU Dreambooth failure are separate problems." + }, + { + "left": "issue:11182", + "right": "issue:9527", + "accept": false, + "reason": "Unnecessary BF16 download and dtype mismatch between ControlNet fp32 and main pipeline bf16 are different issues." + }, + { + "left": "issue:5153", + "right": "issue:5442", + "accept": false, + "reason": "FreeU is an algorithm/pipeline feature request, while ScaleCrafter is a different pipeline proposal; no shared bug." + }, + { + "left": "issue:2185", + "right": "issue:3871", + "accept": false, + "reason": "Missing config files for training and checkpoint-resume training error are unrelated training setup issues." + }, + { + "left": "issue:7133", + "right": "issue:7431", + "accept": false, + "reason": "Both mention text encoders in training, but one is a resume size-mismatch bug and the other is a DoRA training issue; different concrete defects." + }, + { + "left": "issue:11135", + "right": "issue:3826", + "accept": false, + "reason": "Both are new pipeline proposals, but they are entirely different pipelines and changes." + }, + { + "left": "issue:2659", + "right": "issue:6949", + "accept": false, + "reason": "Batch-size variation and even-step blank images are different generation correctness bugs." + }, + { + "left": "issue:12499", + "right": "issue:12529", + "accept": false, + "reason": "Both are Wan 2.2 shape issues, but one is conditioning shape mismatch and the other is tiled VAE tensor-size mismatch; different components." + }, + { + "left": "issue:3871", + "right": "issue:6821", + "accept": false, + "reason": "Checkpoint resume error and caption_column validation error are unrelated training CLI issues." + }, + { + "left": "issue:12011", + "right": "issue:12052", + "accept": false, + "reason": "Wan 2.2 OOM from large I2V loading and LightX2V offloading device-mismatch failure are different runtime problems." + }, + { + "left": "issue:4194", + "right": "issue:5150", + "accept": false, + "reason": "One is a from_single_file loading bug, the other is a missing from_single_file method on a different pipeline class; not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 119, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1013", + "issue:1055", + "issue:11006", + "issue:11148", + "issue:12012", + "issue:1261", + "issue:1363", + "issue:1565", + "issue:2324", + "issue:2361", + "issue:2465", + "issue:2534", + "issue:2830", + "issue:3468", + "issue:3584", + "issue:3706", + "issue:3970", + "issue:3984", + "issue:4117", + "issue:4134", + "issue:4236", + "issue:4303", + "issue:4634", + "issue:4704", + "issue:4779", + "issue:4969", + "issue:50", + "issue:5012", + "issue:5219", + "issue:5259", + "issue:5395", + "issue:5433", + "issue:5436", + "issue:5607", + "issue:5628", + "issue:6092", + "issue:6224", + "issue:6297", + "issue:6460", + "issue:6934", + "issue:7333", + "issue:7682", + "issue:785", + "issue:8277", + "issue:8477", + "issue:8939", + "issue:9167", + "issue:9780" + ], + "soft_pairs": [ + "issue:2830|issue:7333", + "issue:3706|issue:6460", + "issue:4779|issue:7333", + "issue:11148|issue:12012", + "issue:1363|issue:2534", + "issue:1013|issue:3970", + "issue:2324|issue:2465", + "issue:4236|issue:5433", + "issue:5436|issue:6092", + "issue:11006|issue:12012", + "issue:1565|issue:8939", + "issue:7682|issue:9780", + "issue:8277|issue:8477", + "issue:4634|issue:4704", + "issue:5219|issue:6934", + "issue:5395|issue:5628", + "issue:5012|issue:6297", + "issue:1261|issue:3584", + "issue:2361|issue:6224", + "issue:1055|issue:50", + "issue:1261|issue:4117", + "issue:3468|issue:5259", + "issue:3984|issue:785", + "issue:4134|issue:5607", + "issue:4303|issue:8939", + "issue:4969|issue:9167" + ], + "prepared_review_unit_hash": "ffaa672948898f1b1fe9416dd04548a2b2ae8f345c02999a24236d2571aaf268", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35039, + "estimated_input_tokens": 8760, + "estimated_eval_tokens": 17776 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous. Two soft edges look like real duplicates: the missing `LatentDiffusionPipeline` symbol pair and the SDXL diffusers-to-original-format conversion/loading pair. The rest are only loosely related by broad subsystems like DreamBooth, SDXL, schedulers, or video pipelines and should stay separate.", + "confidence": 0.78, + "canonical_issue_reason": "issue:50 is the clearest canonical duplicate anchor: it states the missing `LatentDiffusionPipeline` symbol directly, and issue:1055 is the same bug phrased as \u201cNo LatentDiffusionPipeline on the latest branch.\u201d", + "canonical_pr_reason": null, + "best_issue_reason": "issue:50 is the best overall issue to anchor the cluster because it is the most explicit, self-contained bug report and has the strongest duplicate evidence among the issues shown.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2830", + "right": "issue:7333", + "accept": false, + "reason": "Both are about video generation, but one is a capability question for ModelScope while the other is a specific image-to-video method/paper; not the same bug or change." + }, + { + "left": "issue:3706", + "right": "issue:6460", + "accept": false, + "reason": "Unrelated topics: Wuerstchen model announcement vs. an endless-loading bug while loading a Stable Diffusion inpainting model." + }, + { + "left": "issue:4779", + "right": "issue:7333", + "accept": false, + "reason": "Both mention image/video generation, but one is a pipeline request and the other is a specific image-to-video research issue; no shared concrete failure." + }, + { + "left": "issue:11148", + "right": "issue:12012", + "accept": false, + "reason": "Different code paths: PEFT LoRA adapter state flag handling vs. Wan 2.2 first-block cache context errors." + }, + { + "left": "issue:1363", + "right": "issue:2534", + "accept": false, + "reason": "Both are scheduler-related errors, but the affected pipelines and failure modes differ; too broad to be the same underlying bug." + }, + { + "left": "issue:1013", + "right": "issue:3970", + "accept": false, + "reason": "Separate SD pipeline problems: one is StableDiffusionMegaPipeline compatibility, the other is SDXL prompt embeddings not working." + }, + { + "left": "issue:2324", + "right": "issue:2465", + "accept": false, + "reason": "Dreambooth is the shared theme, but one is a concrete data-loading bug and the other is a request for a reproducible training recipe." + }, + { + "left": "issue:4236", + "right": "issue:5433", + "accept": false, + "reason": "One is SDXL Refiner text-encoder registration, the other is visual artifacts with DPM++ schedulers; not the same defect." + }, + { + "left": "issue:5436", + "right": "issue:6092", + "accept": false, + "reason": "Different training failures in different scripts: ControlNet XL example error vs. LoRA example not training anything." + }, + { + "left": "issue:11006", + "right": "issue:12012", + "accept": false, + "reason": "Wan 2.x video output corruption with quantized transformer is not the same as the first-block-cache context ValueError." + }, + { + "left": "issue:1565", + "right": "issue:8939", + "accept": false, + "reason": "GPU selection/OOM handling is unrelated to a tokenizer-loading failure for Stable Diffusion 3." + }, + { + "left": "issue:7682", + "right": "issue:9780", + "accept": false, + "reason": "Multi-GPU SD1.5 runtime behavior is unrelated to an SD3.5 model attribute error during loading." + }, + { + "left": "issue:8277", + "right": "issue:8477", + "accept": false, + "reason": "Both mention consistency training, but one is convergence and the other is a DistributedDataParallel script error; different problems." + }, + { + "left": "issue:4634", + "right": "issue:4704", + "accept": false, + "reason": "Both are Custom Diffusion, but one is safe serialization and the other is the training script failing; not the same code-path bug." + }, + { + "left": "issue:5219", + "right": "issue:6934", + "accept": true, + "reason": "These describe the same SDXL diffusers conversion issue from opposite ends: conversion from diffusers to original SDXL format is broken, leading to inability to load the converted safetensors model." + }, + { + "left": "issue:5395", + "right": "issue:5628", + "accept": false, + "reason": "Both involve LoRA/DPM-style topics only loosely, but they concern different scheduler/combination behaviors and are not the same bug." + }, + { + "left": "issue:5012", + "right": "issue:6297", + "accept": false, + "reason": "Both are LoRA training issues, but one is about SDXL text-encoder training in a script and the other says a DreamBooth LoRA training version cannot learn from photos." + }, + { + "left": "issue:1261", + "right": "issue:3584", + "accept": false, + "reason": "Xformers backward-pass assertion is unrelated to a DreamBooth tutorial failure." + }, + { + "left": "issue:2361", + "right": "issue:6224", + "accept": false, + "reason": "Both are training/checkpointing issues, but one is a checkpoint save crash and the other is a TorchDynamo fine-tuning problem." + }, + { + "left": "issue:1055", + "right": "issue:50", + "accept": true, + "reason": "Clear duplicate: both report `LatentDiffusionPipeline` missing/not available, with only wording differences." + }, + { + "left": "issue:1261", + "right": "issue:4117", + "accept": false, + "reason": "Xformers attention backward assertion is not the same as SDXL Dreambooth not working." + }, + { + "left": "issue:3468", + "right": "issue:5259", + "accept": false, + "reason": "Dreambooth missing `model_index.json` vs. a generic missing `model_index.json` load error are related by file name but not clearly the same underlying bug from the titles alone." + }, + { + "left": "issue:3984", + "right": "issue:785", + "accept": false, + "reason": "Class-label requirement in a model config is unrelated to PNDMScheduler performance cost." + }, + { + "left": "issue:4134", + "right": "issue:5607", + "accept": false, + "reason": "MPS ControlNet tensor-shape mismatch is different from a general tensor-size mismatch involving another pipeline." + }, + { + "left": "issue:4303", + "right": "issue:8939", + "accept": false, + "reason": "Both are import/load failures, but one is Flax SDXL loading and the other is tokenizer loading for SD3; different code paths." + }, + { + "left": "issue:4969", + "right": "issue:9167", + "accept": false, + "reason": "ImportError for `CrossAttention` and inability to import `StableDiffusion3Pipeline` are distinct missing-symbol issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The canonical anchor on issue:50 is supported by the matching LatentDiffusionPipeline missing-symbol reports in issue:1055, and the SDXL diffusers-conversion pair (issue:5219/6934) is a plausible duplicate relation from the titles. The remaining soft-edge rejections are conservative and stay within what the packet supports." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 120, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10382", + "issue:10823", + "issue:10963", + "issue:10986", + "issue:11062", + "issue:11127", + "issue:11136", + "issue:11359", + "issue:11569", + "issue:11709", + "issue:12012", + "issue:1434", + "issue:1643", + "issue:2081", + "issue:2161", + "issue:2673", + "issue:3174", + "issue:3325", + "issue:4104", + "issue:4255", + "issue:4551", + "issue:4639", + "issue:4724", + "issue:4788", + "issue:5016", + "issue:5150", + "issue:5458", + "issue:5515", + "issue:5558", + "issue:5840", + "issue:6492", + "issue:6517", + "issue:6598", + "issue:6790", + "issue:7146", + "issue:7223", + "issue:730", + "issue:7333", + "issue:8432", + "issue:8787", + "issue:8788", + "issue:8960", + "issue:9055", + "issue:9096", + "issue:9613", + "issue:9768", + "issue:9856" + ], + "soft_pairs": [ + "issue:5150|issue:5458", + "issue:5515|issue:6790", + "issue:7223|issue:7333", + "issue:1643|issue:730", + "issue:8787|issue:9613", + "issue:9096|issue:9768", + "issue:11136|issue:8788", + "issue:4104|issue:6492", + "issue:3325|issue:4639", + "issue:12012|issue:9856", + "issue:10382|issue:8960", + "issue:10986|issue:11709", + "issue:1643|issue:4551", + "issue:11062|issue:11127", + "issue:2081|issue:2673", + "issue:10823|issue:7146", + "issue:4255|issue:6598", + "issue:2161|issue:3174", + "issue:10963|issue:8432", + "issue:11359|issue:11569", + "issue:1434|issue:4788", + "issue:4724|issue:5558", + "issue:6517|issue:9055", + "issue:5016|issue:5840" + ], + "prepared_review_unit_hash": "69bd1cf3c6f788cacc282234e26073c5968feb220430f529a315e9b5ac688f50", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 34099, + "estimated_input_tokens": 8525, + "estimated_eval_tokens": 17306 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated Diffusers issues/feature requests. The proposed soft pairs share broad wording or subsystem names, but not the same concrete bug or change, so none should be merged as duplicates.", + "confidence": 0.18, + "canonical_issue_reason": "Issue 10382 is the most concrete bug report in the set: it describes a specific runtime type mismatch in the VAE encoder forward path with an actionable failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10382 is the best overall representative because it is specific, reproducible, and clearly bug-oriented, whereas most other items are broader troubleshooting, feature requests, or different failures.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5150", + "right": "issue:5458", + "accept": false, + "reason": "Both concern loader APIs, but one is about missing from_single_file support and the other is about local_files_only behavior; different bugs." + }, + { + "left": "issue:5515", + "right": "issue:6790", + "accept": false, + "reason": "One is a pipeline init failure without a scheduler, the other is a training state_dict key mismatch; not the same issue." + }, + { + "left": "issue:7223", + "right": "issue:7333", + "accept": false, + "reason": "Both are video-model related, but one is a feature request to add Latte and the other is about a specific image-to-video method; different changes." + }, + { + "left": "issue:1643", + "right": "issue:730", + "accept": false, + "reason": "Both mention DreamBooth, but the failures are unrelated: an xformers method signature error versus training quality/behavior on local vs Colab." + }, + { + "left": "issue:8787", + "right": "issue:9613", + "accept": false, + "reason": "Slow model loading is unrelated to the Timesteps.scale attribute error; different code paths and symptoms." + }, + { + "left": "issue:9096", + "right": "issue:9768", + "accept": false, + "reason": "Both mention NaNs, but one is Flux half-precision inference and the other is torch 2.5 SDPA/CuDNN; not the same bug." + }, + { + "left": "issue:11136", + "right": "issue:8788", + "accept": false, + "reason": "These are different pipeline support issues: SanaSprint intermediate timesteps vs SD3 VAE tiling on non-power-of-two images." + }, + { + "left": "issue:4104", + "right": "issue:6492", + "accept": false, + "reason": "Both produce black images, but the reported causes and contexts differ and there is no evidence of the same underlying defect." + }, + { + "left": "issue:3325", + "right": "issue:4639", + "accept": false, + "reason": "One is a CUDA invalid-argument error in DreamBooth training; the other is a dataloader multiprocessing error in a different script." + }, + { + "left": "issue:12012", + "right": "issue:9856", + "accept": false, + "reason": "A missing context for block cache state is unrelated to a distributed port collision." + }, + { + "left": "issue:10382", + "right": "issue:8960", + "accept": false, + "reason": "A VAE forward return-type mismatch is unrelated to gradient accumulation not working." + }, + { + "left": "issue:10986", + "right": "issue:11709", + "accept": false, + "reason": "ComplexDouble tensors on MPS and a LoRA key error are different failures in different parts of the loading/inference stack." + }, + { + "left": "issue:1643", + "right": "issue:4551", + "accept": false, + "reason": "Both involve DreamBooth, but one is an xformers API mismatch and the other is a crash with SDXL; not the same bug." + }, + { + "left": "issue:11062", + "right": "issue:11127", + "accept": false, + "reason": "Both mention CivitAI/Flux loading, but one is a specific LoRA key error and the other is a single-file incompatibility for a different model." + }, + { + "left": "issue:2081", + "right": "issue:2673", + "accept": false, + "reason": "Converting PyTorch weights to JAX is unrelated to StableDiffusionUpscalePipeline config validation." + }, + { + "left": "issue:10823", + "right": "issue:7146", + "accept": false, + "reason": "Missing inference files from a fine-tuning script and SDXL training OOM are different problems." + }, + { + "left": "issue:4255", + "right": "issue:6598", + "accept": false, + "reason": "Different-results complaints are too broad here; one compares Diffusers to sd-webui generally, the other is a specific img2img parity report." + }, + { + "left": "issue:2161", + "right": "issue:3174", + "accept": false, + "reason": "One is about LoRA training on an inpainting model; the other is a loading regression caused by new LoRA features." + }, + { + "left": "issue:10963", + "right": "issue:8432", + "accept": false, + "reason": "Both are import errors around autoencoder-related symbols, but they target different modules and symbols; not clearly the same bug." + }, + { + "left": "issue:11359", + "right": "issue:11569", + "accept": false, + "reason": "Both are LTX-Video feature requests, but one asks for faster inference and the other asks for a diffusers repo release; different requests." + }, + { + "left": "issue:1434", + "right": "issue:4788", + "accept": false, + "reason": "Both are OOM reports, but one is StableDiffusionUpscalePipeline at larger sizes and the other is SDXL inpaint on 6 GB VRAM; not the same failure." + }, + { + "left": "issue:4724", + "right": "issue:5558", + "accept": false, + "reason": "Both touch SDXL/control-like workflows, but one is an img2img checkpoint behavior issue and the other is artifacts with multiple ControlNets." + }, + { + "left": "issue:6517", + "right": "issue:9055", + "accept": false, + "reason": "These are different import errors from different loaders and different symbols; no shared code-path is evident." + }, + { + "left": "issue:5016", + "right": "issue:5840", + "accept": false, + "reason": "Both are SDXL DreamBooth LoRA issues, but one is about training text encoders and the other is about checkpoint resume/freezing; not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 121, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10049", + "issue:10428", + "issue:10446", + "issue:10797", + "issue:1166", + "issue:1367", + "issue:1922", + "issue:2036", + "issue:2132", + "issue:2218", + "issue:222", + "issue:2375", + "issue:2560", + "issue:2728", + "issue:2844", + "issue:3363", + "issue:3450", + "issue:3453", + "issue:3871", + "issue:4030", + "issue:4241", + "issue:4250", + "issue:4389", + "issue:4566", + "issue:4892", + "issue:4911", + "issue:5558", + "issue:5605", + "issue:5760", + "issue:5931", + "issue:6354", + "issue:6636", + "issue:7000", + "issue:7056", + "issue:7347", + "issue:746", + "issue:8435", + "issue:9096", + "issue:9187", + "issue:9415", + "issue:9450", + "issue:9515", + "issue:9810", + "issue:9856", + "issue:9895", + "issue:9911", + "issue:993" + ], + "soft_pairs": [ + "issue:2728|issue:4250", + "issue:4892|issue:5760", + "issue:4389|issue:4911", + "issue:222|issue:3453", + "issue:7056|issue:9415", + "issue:2132|issue:3363", + "issue:9515|issue:9810", + "issue:3450|issue:5931", + "issue:5605|issue:6354", + "issue:7000|issue:9096", + "issue:4030|issue:8435", + "issue:10049|issue:6636", + "issue:10446|issue:9450", + "issue:9856|issue:9911", + "issue:1922|issue:2375", + "issue:10797|issue:9187", + "issue:4566|issue:7347", + "issue:4241|issue:6354", + "issue:2844|issue:3871", + "issue:2560|issue:5558", + "issue:10428|issue:9895", + "issue:1367|issue:746", + "issue:1166|issue:2218", + "issue:2036|issue:993" + ], + "prepared_review_unit_hash": "51769a1d0308b17445d9d21a44290475a10761ee0b51d7a34710a50abd2f08f9", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33656, + "estimated_input_tokens": 8414, + "estimated_eval_tokens": 17084 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly distinct reports across different pipelines, training scripts, and feature requests. The soft-similarity pairs are largely broad false positives; even the Flux-related pair is too platform-specific/context-dependent to treat as the same bug.", + "confidence": 0.89, + "canonical_issue_reason": "No canonical issue: the collection is heterogeneous, spanning unrelated bugs and feature requests rather than one underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9895 is the strongest standalone report in the set: it has a precise FluxPipeline device-mismatch regression, a concrete error message, and clear reproduction context.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2728", + "right": "issue:4250", + "accept": false, + "reason": "Both are model-loading complaints, but one is UNet safetensors loading and the other is a generic from_pretrained issue; too broad to be the same bug." + }, + { + "left": "issue:4892", + "right": "issue:5760", + "accept": false, + "reason": "Both are video-related diffusion projects, but they concern entirely different systems and goals." + }, + { + "left": "issue:4389", + "right": "issue:4911", + "accept": false, + "reason": "Loading failures, but different code paths: ControlNetModel loading vs custom text_encoder loading from_single_file." + }, + { + "left": "issue:222", + "right": "issue:3453", + "accept": false, + "reason": "Both are training runtime errors, but the reported failures and causes are unrelated." + }, + { + "left": "issue:7056", + "right": "issue:9415", + "accept": false, + "reason": "Both mention errors after adding parameters, but they target different models/APIs and do not look like one bug." + }, + { + "left": "issue:2132", + "right": "issue:3363", + "accept": false, + "reason": "Both involve DreamBooth/LoRA training, but one is an Accelerator attribute error and the other is a machine-specific failure." + }, + { + "left": "issue:9515", + "right": "issue:9810", + "accept": false, + "reason": "These are different feature/implementation issues in different subsystems; no shared failing code path." + }, + { + "left": "issue:3450", + "right": "issue:5931", + "accept": false, + "reason": "One is a from_ckpt loading problem, the other is nondeterministic generation; not the same underlying issue." + }, + { + "left": "issue:5605", + "right": "issue:6354", + "accept": false, + "reason": "Both are LoRA-related feature requests, but they target different capabilities (SDXL DreamBooth captions vs ControlNet support)." + }, + { + "left": "issue:7000", + "right": "issue:9096", + "accept": false, + "reason": "A deprecation warning and a NaN inference bug are different problems, even if both touch Flux/torch internals." + }, + { + "left": "issue:4030", + "right": "issue:8435", + "accept": false, + "reason": "Both are runtime errors in Stable Diffusion paths, but the entrypoints and failure modes differ." + }, + { + "left": "issue:10049", + "right": "issue:6636", + "accept": false, + "reason": "Both are import errors, but they are for different symbols and different modules." + }, + { + "left": "issue:10446", + "right": "issue:9450", + "accept": false, + "reason": "Both mention Flux and multi-device behavior, but one is FluxControlNet sharding and the other is a FluxPipeline device-placement mismatch." + }, + { + "left": "issue:9856", + "right": "issue:9911", + "accept": false, + "reason": "Distributed port conflict vs multi-ControlNet layer-length mismatch; no shared bug." + }, + { + "left": "issue:1922", + "right": "issue:2375", + "accept": false, + "reason": "Undefined notebook helper vs missing safe_open import; unrelated failures." + }, + { + "left": "issue:10797", + "right": "issue:9187", + "accept": false, + "reason": "Different pipelines and problems: HunyuanVideo group offloading vs SDXL IP-Adapter image encoder handling." + }, + { + "left": "issue:4566", + "right": "issue:7347", + "accept": false, + "reason": "Both are LoRA training/resume complaints, but they describe different regressions and likely different fixes." + }, + { + "left": "issue:4241", + "right": "issue:6354", + "accept": false, + "reason": "A general support question about LoRA on unidiffusers is not the same as adding LoRA support to ControlNet." + }, + { + "left": "issue:2844", + "right": "issue:3871", + "accept": false, + "reason": "Explicit image-embedding loading in StableUnCLIP and checkpoint-resume training are unrelated." + }, + { + "left": "issue:2560", + "right": "issue:5558", + "accept": false, + "reason": "Deterministic generator behavior and ControlNet inpaint artifacts are different issues." + }, + { + "left": "issue:10428", + "right": "issue:9895", + "accept": false, + "reason": "These are the closest pair, but one is an Ascend NPU inference error and the other is a CUDA/CPU regression after a specific commit; too platform-specific to merge as the same bug." + }, + { + "left": "issue:1367", + "right": "issue:746", + "accept": false, + "reason": "Scheduler config serialization and ONNX revision loading are unrelated." + }, + { + "left": "issue:1166", + "right": "issue:2218", + "accept": false, + "reason": "TPU save hang vs overtrained models; no shared failure mechanism." + }, + { + "left": "issue:2036", + "right": "issue:993", + "accept": false, + "reason": "A model load OSError and an fp16 dtype mismatch are distinct problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 122, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10016", + "issue:10472", + "issue:11046", + "issue:1125", + "issue:11627", + "issue:12108", + "issue:12458", + "issue:13300", + "issue:1633", + "issue:2444", + "issue:2707", + "issue:2905", + "issue:3728", + "issue:3948", + "issue:4014", + "issue:4790", + "issue:4808", + "issue:5510", + "issue:5628", + "issue:5647", + "issue:5703", + "issue:617", + "issue:6218", + "issue:6323", + "issue:6520", + "issue:6570", + "issue:6749", + "issue:6758", + "issue:7049", + "issue:7237", + "issue:7509", + "issue:7606", + "issue:7708", + "issue:7793", + "issue:7936", + "issue:7974", + "issue:8228", + "issue:8551", + "issue:8626", + "issue:9012", + "issue:9063", + "issue:9089", + "issue:9202", + "issue:940", + "issue:941", + "issue:9660", + "issue:967" + ], + "soft_pairs": [ + "issue:9063|issue:9089", + "issue:5510|issue:6570", + "issue:6323|issue:9660", + "issue:12108|issue:12458", + "issue:10010|issue:7237", + "issue:6520|issue:6749", + "issue:10472|issue:7606", + "issue:11627|issue:8626", + "issue:3728|issue:7509", + "issue:7049|issue:9202", + "issue:2707|issue:4790", + "issue:1125|issue:13300", + "issue:3948|issue:4014", + "issue:6218|issue:6758", + "issue:2444|issue:2707", + "issue:5647|issue:8551", + "issue:940|issue:941", + "issue:10016|issue:8228", + "issue:4808|issue:6218", + "issue:7708|issue:7936", + "issue:7793|issue:9012", + "issue:1633|issue:2905", + "issue:5628|issue:5703", + "issue:11046|issue:7974", + "issue:617|issue:967" + ], + "prepared_review_unit_hash": "ecaf134a83c7dd8a92c09011e19bd837c6b125c4fa4011dcb96391d4efd500ed", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34949, + "estimated_input_tokens": 8738, + "estimated_eval_tokens": 17732 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicates here: the soft pairs are only title-similar, but each points to a different bug, feature request, or code path across loaders, pipelines, schedulers, and adapters.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a valid canonical duplicate target; the set is too heterogeneous and the soft pairs do not share the same underlying failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9063 is the clearest standalone bug report in the set, but only as a representative example\u2014not as a real cluster canonical, since the rest are unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9063", + "right": "issue:9089", + "accept": false, + "reason": "Both mention FluxPipeline, but one is a tokenizer-loading error and the other is a missing class/import; different root causes." + }, + { + "left": "issue:5510", + "right": "issue:6570", + "accept": false, + "reason": "Both are SDXL inpaint/pix2pix runtime errors, but they involve different tensor-shape/type mismatches and different code paths." + }, + { + "left": "issue:6323", + "right": "issue:9660", + "accept": false, + "reason": "One is a TensorRT pipeline attribute issue; the other is from_pretrained not setting eval mode. Not the same bug." + }, + { + "left": "issue:12108", + "right": "issue:12458", + "accept": false, + "reason": "Both are Qwen-image related, but one is scheduler flow-matching behavior and the other is batch inference/feature request." + }, + { + "left": "issue:10010", + "right": "issue:7237", + "accept": false, + "reason": "FLUX example 404 vs EDMEulerScheduler availability; unrelated failures." + }, + { + "left": "issue:6520", + "right": "issue:6749", + "accept": false, + "reason": "Training validation blank output vs InstantID community example replication; different scenarios and symptoms." + }, + { + "left": "issue:10472", + "right": "issue:7606", + "accept": false, + "reason": "Both touch LoRA, but one is an unexpected lora_bias kwarg and the other is a deprecated processor warning." + }, + { + "left": "issue:11627", + "right": "issue:8626", + "accept": false, + "reason": "One is an SD3.5 IP-Adapter usage issue; the other is guidance documentation for multiple IP-adapter images/masks." + }, + { + "left": "issue:3728", + "right": "issue:7509", + "accept": false, + "reason": "A new pipeline request vs a real-image inversion pipeline; not the same feature." + }, + { + "left": "issue:7049", + "right": "issue:9202", + "accept": false, + "reason": "Both involve IP-Adapter, but one is a value unpacking bug and the other is a specific FaceID Plus V2 incompatibility." + }, + { + "left": "issue:2707", + "right": "issue:4790", + "accept": false, + "reason": "Converter syntax error vs converter output compatibility problem; different issues." + }, + { + "left": "issue:1125", + "right": "issue:13300", + "accept": false, + "reason": "Inpaint pipeline runtime error vs CPU generation failures for SD3.5/FLUX; unrelated code paths." + }, + { + "left": "issue:3948", + "right": "issue:4014", + "accept": false, + "reason": "Pale images vs watermark pixel artifacts; different output-quality problems." + }, + { + "left": "issue:6218", + "right": "issue:6758", + "accept": false, + "reason": "Batched seeds bug in ControlNet/Img2Img vs num_images_per_prompt support in InstantID; not the same defect." + }, + { + "left": "issue:2444", + "right": "issue:2707", + "accept": false, + "reason": "Dreambooth inpaint script failure vs converter syntax error; unrelated scripts." + }, + { + "left": "issue:5647", + "right": "issue:8551", + "accept": false, + "reason": "Inpaint strength==1 bug vs VAE device placement bug under balanced device_map." + }, + { + "left": "issue:940", + "right": "issue:941", + "accept": false, + "reason": "Both are MPS-related, but one is an LMSDiscreteScheduler crash and the other is a num_images_per_prompt noise issue." + }, + { + "left": "issue:10016", + "right": "issue:8228", + "accept": false, + "reason": "Both are from_single_file loading errors, but one is a shape mismatch for SD3.5M and the other is a CLIPTextModel load failure." + }, + { + "left": "issue:4808", + "right": "issue:6218", + "accept": false, + "reason": "SHAP-E integration test failure vs batched seeds bug; different pipelines and symptoms." + }, + { + "left": "issue:7708", + "right": "issue:7936", + "accept": false, + "reason": "Missing tokenizer/config.json vs invalid JSON in local safetensors; different file-validation failures." + }, + { + "left": "issue:7793", + "right": "issue:9012", + "accept": false, + "reason": "LoRA unloading error vs SD3ControlNetModel.from_transformer weight loading error; unrelated loaders." + }, + { + "left": "issue:1633", + "right": "issue:2905", + "accept": false, + "reason": "One is an optional karras-sigmas wrapper enhancement; the other is a broad scheduler feature request. Same area, not same change." + }, + { + "left": "issue:5628", + "right": "issue:5703", + "accept": false, + "reason": "Karras/LMS denoising offset vs ddim_inverse noise reconstruction; different scheduler behaviors." + }, + { + "left": "issue:11046", + "right": "issue:7974", + "accept": false, + "reason": "General flux controlnet/inpainting/IP-adapter inference request vs a specific inpaint-with-IP-Adapter bug." + }, + { + "left": "issue:617", + "right": "issue:967", + "accept": false, + "reason": "NNPACK hardware warning vs ONNX scheduler incompatibility; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 123, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10066", + "issue:10202", + "issue:10492", + "issue:10542", + "issue:11702", + "issue:12078", + "issue:12080", + "issue:12157", + "issue:12175", + "issue:12319", + "issue:12459", + "issue:13065", + "issue:1310", + "issue:13281", + "issue:1434", + "issue:1934", + "issue:2015", + "issue:2091", + "issue:2361", + "issue:2507", + "issue:2618", + "issue:3296", + "issue:3377", + "issue:3565", + "issue:3613", + "issue:3761", + "issue:3885", + "issue:3937", + "issue:4080", + "issue:4400", + "issue:5044", + "issue:5143", + "issue:5606", + "issue:5607", + "issue:5873", + "issue:5966", + "issue:6069", + "issue:6157", + "issue:6418", + "issue:6545", + "issue:6548", + "issue:6594", + "issue:6891", + "issue:7237", + "issue:7418", + "issue:8668", + "issue:9178", + "issue:9476" + ], + "soft_pairs": [ + "issue:12080|issue:13065", + "issue:4080|issue:5966", + "issue:5143|issue:6418", + "issue:6594|issue:7418", + "issue:12319|issue:13281", + "issue:1434|issue:1934", + "issue:12078|issue:9476", + "issue:12157|issue:12175", + "issue:3937|issue:6157", + "issue:10066|issue:5607", + "issue:6069|issue:8668", + "issue:6545|issue:6891", + "issue:2361|issue:3296", + "issue:10492|issue:11702", + "issue:2507|issue:5606", + "issue:2618|issue:6548", + "issue:3377|issue:4400", + "issue:3565|issue:3613", + "issue:5873|issue:7237", + "issue:3761|issue:5044", + "issue:2091|issue:3885", + "issue:10202|issue:9178", + "issue:10542|issue:12459", + "issue:1310|issue:2015" + ], + "prepared_review_unit_hash": "6e54873bb0683fbd0dedc967cd557a77f92952c2cd72493f6b40c7605bcff371", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34484, + "estimated_input_tokens": 8621, + "estimated_eval_tokens": 17498 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a noisy mix of unrelated issue reports across different models and workflows. The soft pairs only share broad vocabulary (e.g. loading, inference, offloading, batch size) and do not describe the same concrete bug or change, so none should be merged as duplicates.", + "confidence": 0.79, + "canonical_issue_reason": "issue:6545 is the best anchor only because it is an explicit tracker with the most inbound references and the broadest coordination role; the rest are mostly isolated model-specific bugs rather than clear duplicates of one another.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6545 is the most suitable global issue representative for the cluster due to its tracker status, high discussion activity, and many references; it is the closest thing to a hub, even though the cluster itself does not form a true duplicate set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12080", + "right": "issue:13065", + "accept": false, + "reason": "Different requests: Qwen Image editing inference vs Qwen-Image controlnet inpaint+canny. Same model family, but not the same bug or code path." + }, + { + "left": "issue:4080", + "right": "issue:5966", + "accept": false, + "reason": "SDXL inpainting pipeline request vs ControlNet for SD 2.1. These are different feature requests for different pipelines/models." + }, + { + "left": "issue:5143", + "right": "issue:6418", + "accept": false, + "reason": "Unet output nondeterminism/precision difference vs stable video diffusion frames repeating. Different symptoms and likely different components." + }, + { + "left": "issue:6594", + "right": "issue:7418", + "accept": false, + "reason": "Raw GitHub config download failure vs a community pipeline bug. No shared underlying defect beyond both being example-related." + }, + { + "left": "issue:12319", + "right": "issue:13281", + "accept": false, + "reason": "Both involve offloading, but one is block_level group offloading generally and the other is torchao-quantized Qwen image with use_stream=True device mismatch. Not the same bug." + }, + { + "left": "issue:1434", + "right": "issue:1934", + "accept": false, + "reason": "OOM during upscale pipeline vs CPU offload VRAM consumption. Related to memory, but different pipelines and failure modes." + }, + { + "left": "issue:12078", + "right": "issue:9476", + "accept": false, + "reason": "Flux Control finetuning validation input issue vs FLUX LoRA validation dtype mismatch. Different training setups and errors." + }, + { + "left": "issue:12157", + "right": "issue:12175", + "accept": false, + "reason": "Quantization-config training device iterable error vs bad generated image quality. One is a runtime exception, the other is a quality complaint." + }, + { + "left": "issue:3937", + "right": "issue:6157", + "accept": false, + "reason": "ZeroscopeXL loading issue vs ip_adapter not working in controlnet img2img. Different models and pipeline integration points." + }, + { + "left": "issue:10066", + "right": "issue:5607", + "accept": false, + "reason": "Self.attn unpacking ValueError vs tensor size mismatch in a different runtime path. Similar error class only, not the same bug." + }, + { + "left": "issue:6069", + "right": "issue:8668", + "accept": false, + "reason": "Scheduler add_noise failing at t=0 vs SD3 img2img requiring aligned image size. Different code paths and symptoms." + }, + { + "left": "issue:6545", + "right": "issue:6891", + "accept": false, + "reason": "Both are trackers, but they target different maintenance efforts: example inferencing modularization vs model card utilities. Not duplicate work." + }, + { + "left": "issue:2361", + "right": "issue:3296", + "accept": false, + "reason": "Checkpoint save crash in text-to-image finetuning vs checkpoint save/load problems in DreamBooth with text encoder. Different scripts and failure specifics." + }, + { + "left": "issue:10492", + "right": "issue:11702", + "accept": false, + "reason": "torchao quantize + LoRA fuse error vs AccVid LoRA key error with diffusers. Both LoRA-related, but not the same bug or model path." + }, + { + "left": "issue:2507", + "right": "issue:5606", + "accept": false, + "reason": "Safetensors load speed on GPU vs LoRA loading without alpha value. Performance issue vs model metadata handling bug." + }, + { + "left": "issue:2618", + "right": "issue:6548", + "accept": false, + "reason": "ControlNet image input type handling vs RGBA image misclassified as latents. Both input-type issues, but different pipelines and failure conditions." + }, + { + "left": "issue:3377", + "right": "issue:4400", + "accept": false, + "reason": "Training-weighting question vs question about text_encoder support. Neither describes the same bug or fix." + }, + { + "left": "issue:3565", + "right": "issue:3613", + "accept": false, + "reason": "Both are memory-related training complaints, but they concern different scripts and different torch/runtime contexts." + }, + { + "left": "issue:5873", + "right": "issue:7237", + "accept": false, + "reason": "ONNX export of scaled_dot_product_attention vs missing EDMEulerScheduler symbol. Completely different functionality." + }, + { + "left": "issue:3761", + "right": "issue:5044", + "accept": false, + "reason": "Loading multiple community pipelines in one process vs Kandinsky 2.2 loading failure. Different pipeline-loading problems." + }, + { + "left": "issue:2091", + "right": "issue:3885", + "accept": false, + "reason": "Resume-from-checkpoint multi-GPU issue vs VRAM reduction question for training. Similar training context, but not the same underlying bug." + }, + { + "left": "issue:10202", + "right": "issue:9178", + "accept": false, + "reason": "Flux Control LoRA unloading bug vs flux LoRA training target modules error. Different stages: unloading/runtime cleanup vs training setup." + }, + { + "left": "issue:10542", + "right": "issue:12459", + "accept": false, + "reason": "Hunyuan batch-size regression vs FLUX.1-Kontext batch inference throughput issue. Both concern batching, but not the same defect." + }, + { + "left": "issue:1310", + "right": "issue:2015", + "accept": false, + "reason": "K-Diffusion + Stable Diffusion pipeline request vs categorical diffusion pipeline question. Different feature/topic entirely." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 124, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10472", + "issue:10514", + "issue:10526", + "issue:10760", + "issue:10798", + "issue:11135", + "issue:11232", + "issue:11352", + "issue:11569", + "issue:11651", + "issue:11962", + "issue:13015", + "issue:13053", + "issue:1392", + "issue:1478", + "issue:1643", + "issue:1979", + "issue:255", + "issue:2673", + "issue:2851", + "issue:3238", + "issue:3593", + "issue:3613", + "issue:3672", + "issue:3700", + "issue:3803", + "issue:4213", + "issue:457", + "issue:5474", + "issue:5877", + "issue:5923", + "issue:5932", + "issue:6008", + "issue:6230", + "issue:6277", + "issue:6695", + "issue:7180", + "issue:7228", + "issue:7592", + "issue:785", + "issue:861", + "issue:8738", + "issue:9138", + "issue:9210", + "issue:9264", + "issue:9270", + "issue:9690" + ], + "soft_pairs": [ + "issue:5923|issue:9210", + "issue:11962|issue:9264", + "issue:2851|issue:3803", + "issue:3613|issue:6230", + "issue:10526|issue:9690", + "issue:1478|issue:2673", + "issue:1979|issue:861", + "issue:11651|issue:9270", + "issue:3238|issue:6008", + "issue:5877|issue:6695", + "issue:457|issue:785", + "issue:3593|issue:7180", + "issue:4213|issue:7592", + "issue:10514|issue:9138", + "issue:10472|issue:11232", + "issue:10760|issue:7228", + "issue:10798|issue:13015", + "issue:11135|issue:5474", + "issue:11352|issue:8738", + "issue:11569|issue:13053", + "issue:1392|issue:5474", + "issue:1643|issue:5932", + "issue:255|issue:3672", + "issue:3700|issue:6277" + ], + "prepared_review_unit_hash": "13fc6b4946631934a2bf2b35d6fd94f64fee64aadc52a3413f52adc5ff42e3f8", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33835, + "estimated_input_tokens": 8459, + "estimated_eval_tokens": 17174 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a heterogeneous mix of diffusion bugs, training errors, quantization issues, and feature requests. The soft pairs are mostly superficial title-similarity matches, not the same underlying bug/change, so none should be merged.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 11651 is the closest concrete runtime-failure report in the recurring Flux/LoRA-loading subset, but the overall cluster is too heterogeneous to have a single true canonical issue.", + "canonical_pr_reason": null, + "best_issue_reason": "11651 is the best representative issue here because it describes a specific, reproducible failure in the dominant Flux/LoRA theme, even though it does not subsume the other unrelated items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5923", + "right": "issue:9210", + "accept": false, + "reason": "Different bugs: multi-GPU training stalls vs a dataset column validation error." + }, + { + "left": "issue:11962", + "right": "issue:9264", + "accept": false, + "reason": "Different requests: GGUF quantized-model support vs asking for an inpainting model." + }, + { + "left": "issue:2851", + "right": "issue:3803", + "accept": false, + "reason": "TPU push_to_hub failure and multi-GPU LoRA sharing are unrelated issues." + }, + { + "left": "issue:3613", + "right": "issue:6230", + "accept": false, + "reason": "Both are OOM reports, but they are separate training scripts and not the same bug." + }, + { + "left": "issue:10526", + "right": "issue:9690", + "accept": false, + "reason": "A Flux FP8 constructor TypeError is unrelated to code cleanup for an example." + }, + { + "left": "issue:1478", + "right": "issue:2673", + "accept": false, + "reason": "One is about multiple image outputs; the other is an upscale pipeline config/runtime mismatch." + }, + { + "left": "issue:1979", + "right": "issue:861", + "accept": false, + "reason": "Both mention conversion to diffusers, but they target different conversion paths and failures." + }, + { + "left": "issue:11651", + "right": "issue:9270", + "accept": false, + "reason": "Related Flux/LoRA area, but one is a FluxControl load failure and the other is broad missing LoRA support." + }, + { + "left": "issue:3238", + "right": "issue:6008", + "accept": false, + "reason": "LoRA safetensor conversion is not the same as IP-adapter/LCM LoRA failing." + }, + { + "left": "issue:5877", + "right": "issue:6695", + "accept": false, + "reason": "Generic subprocess errors in different contexts; too vague to be the same bug." + }, + { + "left": "issue:457", + "right": "issue:785", + "accept": false, + "reason": "Different subsystems: MPS rendering artifacts vs scheduler performance." + }, + { + "left": "issue:3593", + "right": "issue:7180", + "accept": false, + "reason": "LoRA/xformers second-call failure is unrelated to SDXL controlnet CPU offload." + }, + { + "left": "issue:4213", + "right": "issue:7592", + "accept": false, + "reason": "Model loading dependency issue vs DoRA state_dict key loading bug." + }, + { + "left": "issue:10514", + "right": "issue:9138", + "accept": false, + "reason": "Sana resolution binning limit and an int/str TypeError are unrelated." + }, + { + "left": "issue:10472", + "right": "issue:11232", + "accept": false, + "reason": "LoRA constructor keyword mismatch is not the same as noise residuals during fine-tuning." + }, + { + "left": "issue:10760", + "right": "issue:7228", + "accept": false, + "reason": "Importing a missing pipeline class is unrelated to a traced model lacking dtype." + }, + { + "left": "issue:10798", + "right": "issue:13015", + "accept": false, + "reason": "Different device/quantization problems in different models and backends." + }, + { + "left": "issue:11135", + "right": "issue:5474", + "accept": false, + "reason": "Both are pipeline feature requests, but they are distinct techniques and not one change." + }, + { + "left": "issue:11352", + "right": "issue:8738", + "accept": false, + "reason": "CPU warning on 8bit model movement is unrelated to SD3Transformer configuration failure." + }, + { + "left": "issue:11569", + "right": "issue:13053", + "accept": false, + "reason": "A repo request for a distilled model is not the same as adding Unipic 3.0 pipeline support." + }, + { + "left": "issue:1392", + "right": "issue:5474", + "accept": false, + "reason": "Generic Stable Diffusion 2 discussion does not match a specific reference-only pipeline request." + }, + { + "left": "issue:1643", + "right": "issue:5932", + "accept": false, + "reason": "xformers signature mismatch during DreamBooth and a 4-channel input error are different bugs." + }, + { + "left": "issue:255", + "right": "issue:3672", + "accept": false, + "reason": "UNet shape mismatch and concurrent thread failure are unrelated." + }, + { + "left": "issue:3700", + "right": "issue:6277", + "accept": false, + "reason": "Unsupported min_snr argument and broken gradients are separate training-example issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 125, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10302", + "issue:10337", + "issue:10797", + "issue:10914", + "issue:11298", + "issue:11352", + "issue:11370", + "issue:11872", + "issue:12329", + "issue:12529", + "issue:3280", + "issue:3442", + "issue:3454", + "issue:3694", + "issue:3781", + "issue:3880", + "issue:4073", + "issue:4109", + "issue:4183", + "issue:4354", + "issue:4386", + "issue:4450", + "issue:4554", + "issue:4754", + "issue:4782", + "issue:487", + "issue:4874", + "issue:5334", + "issue:5436", + "issue:6388", + "issue:6441", + "issue:6636", + "issue:6801", + "issue:8463", + "issue:8543", + "issue:8588", + "issue:8673", + "issue:8744", + "issue:8958", + "issue:9167", + "issue:9200", + "issue:9209", + "issue:9343", + "issue:9484", + "issue:9546", + "issue:9637", + "issue:9674", + "issue:9913" + ], + "soft_pairs": [ + "issue:3781|issue:5334", + "issue:3880|issue:4782", + "issue:4109|issue:4354", + "issue:4386|issue:6388", + "issue:4554|issue:8543", + "issue:5436|issue:9674", + "issue:6801|issue:9200", + "issue:8588|issue:8673", + "issue:3454|issue:487", + "issue:9484|issue:9546", + "issue:10797|issue:11370", + "issue:3280|issue:4073", + "issue:3442|issue:4450", + "issue:10302|issue:11872", + "issue:4183|issue:4874", + "issue:8463|issue:8958", + "issue:10337|issue:9343", + "issue:8744|issue:9209", + "issue:11298|issue:9637", + "issue:6636|issue:9167", + "issue:12329|issue:12529", + "issue:10914|issue:11352", + "issue:6441|issue:9913", + "issue:3694|issue:4754" + ], + "prepared_review_unit_hash": "a24f7d084f8b076d3e1207476d529b059bae1b8958f45857019dec369c90df3e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33846, + "estimated_input_tokens": 8462, + "estimated_eval_tokens": 17180 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is thematically loose: several issues concern offloading/quantization/device placement, but the proposed soft pairs are mostly different concrete bugs and should not be merged.", + "confidence": 0.37, + "canonical_issue_reason": "issue:10914 is the broadest representative of the shared offloading/device-placement theme: models being moved to CPU unexpectedly. It subsumes the general category better than the narrower FP8, 8-bit warning, or inefficiency reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10914 is the best single issue anchor because it is the most general and user-facing statement of the underlying offloading problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3781", + "right": "issue:5334", + "accept": false, + "reason": "LoRA loading regression vs Accelerate training API expectations; different components and failure modes." + }, + { + "left": "issue:3880", + "right": "issue:4782", + "accept": false, + "reason": "Both are inpainting-related, but one is degraded unmasked areas and the other is mask==0 changes; not clearly the same bug." + }, + { + "left": "issue:4109", + "right": "issue:4354", + "accept": false, + "reason": "SDXL inpaint pipeline failure vs wrong pipeline type returned from from_single_file; distinct issues." + }, + { + "left": "issue:4386", + "right": "issue:6388", + "accept": false, + "reason": "LoRA conversion question vs gradient-checkpointing ignored in SDXL+LoRA training; unrelated bugs." + }, + { + "left": "issue:4554", + "right": "issue:8543", + "accept": false, + "reason": "OOM in Kandinsky example vs SD3 gradient-checkpointing failure; different pipelines and symptoms." + }, + { + "left": "issue:5436", + "right": "issue:9674", + "accept": false, + "reason": "Training example error vs bad training results; not the same concrete defect." + }, + { + "left": "issue:6801", + "right": "issue:9200", + "accept": false, + "reason": "Scheduler loading problem vs SDXL ControlNet dtype error; different code paths." + }, + { + "left": "issue:8588", + "right": "issue:8673", + "accept": false, + "reason": "Missing conversion script vs missing SD3 attention masks; unrelated SD3 issues." + }, + { + "left": "issue:3454", + "right": "issue:487", + "accept": false, + "reason": "Different eras and failures: train script compatibility vs StableDiffusionPipeline on M1 CPU." + }, + { + "left": "issue:9484", + "right": "issue:9546", + "accept": false, + "reason": "FLUX multi-GPU/deepspeed training vs ControlNet validation OOM; separate training problems." + }, + { + "left": "issue:10797", + "right": "issue:11370", + "accept": false, + "reason": "HunyuanVideo group offloading failure vs conversion script parsing error; unrelated." + }, + { + "left": "issue:3280", + "right": "issue:4073", + "accept": false, + "reason": "Scheduler incompatibility in DeepFloyd IF vs safety-checker tensor/astype error; distinct subsystems." + }, + { + "left": "issue:3442", + "right": "issue:4450", + "accept": false, + "reason": "Textual inversion embedding dimension mismatch vs multi text inversion not working; related area but not the same bug." + }, + { + "left": "issue:10302", + "right": "issue:11872", + "accept": false, + "reason": "FP8 inference noise without CPU offloading vs repeated CPU/GPU transfers during loops; one is numerical quality, the other performance/offloading behavior." + }, + { + "left": "issue:4183", + "right": "issue:4874", + "accept": false, + "reason": "ControlNet guidance_scale edge case vs low_res_scheduler local-load bug; different triggers and failures." + }, + { + "left": "issue:8463", + "right": "issue:8958", + "accept": false, + "reason": "InstructP2P-SDXL tokenization failure vs torch.library.register_fake compatibility error; unrelated import/runtime problems." + }, + { + "left": "issue:10337", + "right": "issue:9343", + "accept": false, + "reason": "Quantized FLUX IPAdapter failure vs FLUX loading with low_cpu_mem_usage/ignore_mismatched_sizes; both FLUX-related but different loading paths." + }, + { + "left": "issue:8744", + "right": "issue:9209", + "accept": false, + "reason": "Training convenience feature request vs model download connectivity to hf-mirror; not the same issue." + }, + { + "left": "issue:11298", + "right": "issue:9637", + "accept": false, + "reason": "LoRA hot-swapping PEFT key error vs xFormers attention-mask/bfloat16 sequence-length bug; different APIs and failure modes." + }, + { + "left": "issue:6636", + "right": "issue:9167", + "accept": false, + "reason": "Different import errors: DIFFUSERS_SLOW_IMPORT vs StableDiffusion3Pipeline import name issue." + }, + { + "left": "issue:12329", + "right": "issue:12529", + "accept": false, + "reason": "Wrong Wan transformer variant loaded vs tiled VAE tensor-size mismatch; unrelated model components." + }, + { + "left": "issue:10914", + "right": "issue:11352", + "accept": false, + "reason": "Unexpected CPU offloading in general vs missing warning when moving an 8-bit quantized pipeline to CPU; related theme but not the same bug." + }, + { + "left": "issue:6441", + "right": "issue:9913", + "accept": false, + "reason": "from_pretrained failing in Colab vs set_lora_device not working; different API paths and symptoms." + }, + { + "left": "issue:3694", + "right": "issue:4754", + "accept": false, + "reason": "Missing model/config files during example finetuning vs corrupted textual-inversion load error; separate file-handling issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 126, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10514", + "issue:10520", + "issue:1100", + "issue:11178", + "issue:11819", + "issue:12009", + "issue:12085", + "issue:12536", + "issue:1261", + "issue:12661", + "issue:1271", + "issue:1305", + "issue:1367", + "issue:1541", + "issue:1821", + "issue:1865", + "issue:1919", + "issue:2107", + "issue:2331", + "issue:2518", + "issue:2786", + "issue:3382", + "issue:3487", + "issue:3717", + "issue:4014", + "issue:4185", + "issue:4229", + "issue:4398", + "issue:4848", + "issue:5205", + "issue:5227", + "issue:5379", + "issue:5380", + "issue:5429", + "issue:5474", + "issue:5923", + "issue:5934", + "issue:6372", + "issue:684", + "issue:817", + "issue:8228", + "issue:8503", + "issue:8786", + "issue:898", + "issue:9101", + "issue:9443", + "issue:9763", + "issue:9927" + ], + "soft_pairs": [ + "issue:1100|issue:898", + "issue:1367|issue:2786", + "issue:2518|issue:5429", + "issue:9763|issue:9927", + "issue:10514|issue:10520", + "issue:2331|issue:5474", + "issue:1821|issue:3487", + "issue:12536|issue:12661", + "issue:1541|issue:817", + "issue:4848|issue:8786", + "issue:1261|issue:684", + "issue:5379|issue:5380", + "issue:12009|issue:12085", + "issue:4014|issue:4398", + "issue:5474|issue:6372", + "issue:11178|issue:8228", + "issue:1271|issue:1919", + "issue:2107|issue:4185", + "issue:3717|issue:5923", + "issue:1865|issue:3382", + "issue:4229|issue:5923", + "issue:5227|issue:5934", + "issue:11819|issue:9101", + "issue:1305|issue:5205", + "issue:8503|issue:9443" + ], + "prepared_review_unit_hash": "b63594dfc58b4507bd01d6eacd890f2514af59fd15b041c7054d87ae8444e0f0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34542, + "estimated_input_tokens": 8636, + "estimated_eval_tokens": 17528 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: the items mostly cover unrelated bugs, feature requests, or model-specific failures across different pipelines/subsystems. The soft pairs share broad keywords at best, but there is no convincing evidence that any pair is the same underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": "No canonical issue: the set does not cohere around one fixable bug, and the apparent overlaps are only broad subsystem similarity (e.g. Sana, Wan, ControlNet, from_single_file, multi-GPU).", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10520 is the clearest concrete bug report in the set, with a specific runtime traceback and user-reproducible failure, so it is the best representative issue even though it is not a duplicate target for the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1100", + "right": "issue:898", + "accept": false, + "reason": "Different errors in different environments: CLIPTextTransformer device attribute vs TPU bfloat16 scalar type; no shared code-path bug." + }, + { + "left": "issue:1367", + "right": "issue:2786", + "accept": false, + "reason": "Unrelated issues: scheduler config serialization vs numpy input handling in Depth2Img; only generic API bug similarity." + }, + { + "left": "issue:2518", + "right": "issue:5429", + "accept": false, + "reason": "Both are example/support requests, but one is xformers support and the other is model-without-VAE support; not the same change." + }, + { + "left": "issue:9763", + "right": "issue:9927", + "accept": false, + "reason": "Both involve training tokenization limits, but one is Flux ControlNet truncation and the other is SD3 header-size failure; different failure modes." + }, + { + "left": "issue:10514", + "right": "issue:10520", + "accept": false, + "reason": "Same model family (Sana 4K) but different bugs: resolution binning/sample_size constraint vs cross-device tensor placement error." + }, + { + "left": "issue:2331", + "right": "issue:5474", + "accept": false, + "reason": "A general ControlNet pipeline request vs a new reference-only pipeline proposal; not the same underlying feature." + }, + { + "left": "issue:1821", + "right": "issue:3487", + "accept": false, + "reason": "Both are inpaint-related, but one is negative_prompt effectiveness and the other is strength/scheduler quality; distinct issues." + }, + { + "left": "issue:12536", + "right": "issue:12661", + "accept": false, + "reason": "Both mention context parallelism, but one is a sharding-dimension assertion in Wan image-to-video inference and the other is a ChronoEdit crash; no concrete same bug." + }, + { + "left": "issue:1541", + "right": "issue:817", + "accept": false, + "reason": "Dreambooth loading/model-type detection vs mixed-precision failure; different training issues." + }, + { + "left": "issue:4848", + "right": "issue:8786", + "accept": false, + "reason": "Both touch token/length behavior, but one is LPW SDXL support and the other is SD3 max_sequence_length handling; not the same code path." + }, + { + "left": "issue:1261", + "right": "issue:684", + "accept": false, + "reason": "XFormers backward-pass assertion vs AMD memory optimization not working; different subsystems and symptoms." + }, + { + "left": "issue:5379", + "right": "issue:5380", + "accept": false, + "reason": "Distinct community pipeline proposals with no shared bug or implementation target." + }, + { + "left": "issue:12009", + "right": "issue:12085", + "accept": false, + "reason": "Both are GGUF loading failures in Wan models, but they affect different model classes and report different errors; too little evidence of one shared fix." + }, + { + "left": "issue:4014", + "right": "issue:4398", + "accept": false, + "reason": "SDXL watermark artifacts vs black images in SDXL LoRA dreambooth are different defects, not one underlying bug." + }, + { + "left": "issue:5474", + "right": "issue:6372", + "accept": false, + "reason": "Different pipeline proposals: Stable Diffusion Reference Only vs SVD + ControlNet." + }, + { + "left": "issue:11178", + "right": "issue:8228", + "accept": false, + "reason": "Both are from_single_file load failures, but they involve different model components and failure types; not the same bug." + }, + { + "left": "issue:1271", + "right": "issue:1919", + "accept": false, + "reason": "Both are Dreambooth example-related, but one is dependency installation and the other is missing datasets; unrelated fixes." + }, + { + "left": "issue:2107", + "right": "issue:4185", + "accept": false, + "reason": "Both mention black images, but one is low-resolution output and the other is SDXL ControlNet inference; symptom overlap only." + }, + { + "left": "issue:3717", + "right": "issue:5923", + "accept": false, + "reason": "Both concern multi-GPU training, but one is OOM and the other is stuck execution; different problems." + }, + { + "left": "issue:1865", + "right": "issue:3382", + "accept": false, + "reason": "Different training failures: ZeRO stage3 dimension error vs multi-GPU CUDA OOM." + }, + { + "left": "issue:4229", + "right": "issue:5923", + "accept": false, + "reason": "Both are multi-GPU training problems, but one is an accelerated-training error and the other is a hang; no shared concrete bug." + }, + { + "left": "issue:5227", + "right": "issue:5934", + "accept": false, + "reason": "One is load_lora_weight under torch.compile, the other is a general torch.compile error; overlapping topic but not the same failure." + }, + { + "left": "issue:11819", + "right": "issue:9101", + "accept": false, + "reason": "Both ask for training scripts, but for different models/tasks (FLUX Kontext image editing vs SD3 InstructPix2Pix)." + }, + { + "left": "issue:1305", + "right": "issue:5205", + "accept": false, + "reason": "A paper request about Direct Inversion vs a repaint/controlnet pipeline request; no duplicate bug or change." + }, + { + "left": "issue:8503", + "right": "issue:9443", + "accept": false, + "reason": "Both are SD3-related, but one is gradient checkpointing crashes and the other is tokenizer slow-version instantiation; distinct issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 127, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10147", + "issue:10798", + "issue:11215", + "issue:1166", + "issue:11853", + "issue:12022", + "issue:12037", + "issue:1214", + "issue:1434", + "issue:1500", + "issue:2331", + "issue:2440", + "issue:2695", + "issue:3221", + "issue:3405", + "issue:3564", + "issue:3661", + "issue:3748", + "issue:4014", + "issue:4031", + "issue:4073", + "issue:4206", + "issue:4213", + "issue:4279", + "issue:4305", + "issue:4471", + "issue:4530", + "issue:4686", + "issue:4880", + "issue:4885", + "issue:4985", + "issue:5016", + "issue:5037", + "issue:5399", + "issue:5488", + "issue:5607", + "issue:5737", + "issue:5757", + "issue:5966", + "issue:6008", + "issue:6130", + "issue:6146", + "issue:7168", + "issue:7199", + "issue:7357", + "issue:761", + "issue:8482", + "issue:8729" + ], + "soft_pairs": [ + "issue:4014|issue:4530", + "issue:4213|issue:4279", + "issue:5037|issue:8729", + "issue:3748|issue:6130", + "issue:10147|issue:7199", + "issue:2331|issue:3564", + "issue:1214|issue:3661", + "issue:4031|issue:4073", + "issue:10798|issue:11215", + "issue:1166|issue:761", + "issue:1434|issue:4686", + "issue:4471|issue:4885", + "issue:5016|issue:5737", + "issue:5607|issue:7168", + "issue:2440|issue:5399", + "issue:11853|issue:6008", + "issue:4305|issue:7357", + "issue:2695|issue:3405", + "issue:12022|issue:12037", + "issue:5966|issue:8482", + "issue:4880|issue:5488", + "issue:5757|issue:6146", + "issue:4206|issue:4985", + "issue:1500|issue:3221" + ], + "prepared_review_unit_hash": "d233c7a566f4633b7b0f194252a07a3dee84fad299ef04ae7f0803b9742fe028", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34093, + "estimated_input_tokens": 8524, + "estimated_eval_tokens": 17304 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose mix of unrelated feature requests and bug reports around model support, LoRA/ControlNet, quantization, and training/runtime compatibility. None of the soft pairs look like true duplicates; they share only broad subsystem similarity.", + "confidence": 0.82, + "canonical_issue_reason": "issue:8482 is the broadest representative item: it is a general model-support request for a major new family (Stable Diffusion 3), which fits the cluster\u2019s overall support/compatibility theme better than the narrower bug reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8482 is the best issue anchor because it is a clear, standalone support request with broad scope. It is not a duplicate of the others, but it best represents the cluster\u2019s high-level theme.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4014", + "right": "issue:4530", + "accept": false, + "reason": "Both mention image artifacts, but one is SDXL watermark artifacts and the other is SDXL refiner color artifacts; different symptoms and fixes." + }, + { + "left": "issue:4213", + "right": "issue:4279", + "accept": false, + "reason": "Both are LoRA-loading problems, but one is `from_single_file()` without accelerate and the other is SDXL 0.9 LoRAs from CivitAI; not the same bug." + }, + { + "left": "issue:5037", + "right": "issue:8729", + "accept": false, + "reason": "Both involve validation images in training scripts, but one is about the second `--validation_image` argument and the other about an unexpected validation image in webdataset SDXL ControlNet training; too different." + }, + { + "left": "issue:3748", + "right": "issue:6130", + "accept": false, + "reason": "Both are LoRA load failures, but one is a float16/float32 QueryKey mismatch in safetensors and the other is a generic loading error; not clearly the same code-path." + }, + { + "left": "issue:10147", + "right": "issue:7199", + "accept": false, + "reason": "A multi-view generation support request and a ViewFusion research/feature request are related at a high level, but not the same underlying bug or change." + }, + { + "left": "issue:2331", + "right": "issue:3564", + "accept": false, + "reason": "Both concern ControlNet support, but one is adding a ControlNet pipeline and the other is Uni-ControlNet support; different feature scopes." + }, + { + "left": "issue:1214", + "right": "issue:3661", + "accept": false, + "reason": "Half-precision CLIPTextModel failure and `from_ckpt` not working for SD2.x are different compatibility issues." + }, + { + "left": "issue:4031", + "right": "issue:4073", + "accept": false, + "reason": "Both touch safety checker loading, but one fails loading from a local file while the other is a DeepFloyd IF `astype` tensor error; different paths." + }, + { + "left": "issue:10798", + "right": "issue:11215", + "accept": false, + "reason": "Both are about Flux quantization, but one is a device-combination bitsandbytes bug and the other is slow quantization plus poor LoRA support; not the same concrete problem." + }, + { + "left": "issue:1166", + "right": "issue:761", + "accept": false, + "reason": "TPU textual inversion saving hangs and DreamBooth 8-bit Adam failures are separate training issues." + }, + { + "left": "issue:1434", + "right": "issue:4686", + "accept": false, + "reason": "One is an OOM while upscaling; the other is loading a saved upscale pipeline. Related subsystem, different failure." + }, + { + "left": "issue:4471", + "right": "issue:4885", + "accept": false, + "reason": "Both involve LoRA loading in SDXL pipelines, but the affected pipelines and error modes differ; not enough to treat as the same bug." + }, + { + "left": "issue:5016", + "right": "issue:5737", + "accept": false, + "reason": "SDXL DreamBooth LoRA text encoder training and SDXL conversion-script failure are distinct issues." + }, + { + "left": "issue:5607", + "right": "issue:7168", + "accept": false, + "reason": "A tensor-dimension mismatch and IP-Adapter embed shape confusion are not the same underlying bug." + }, + { + "left": "issue:2440", + "right": "issue:5399", + "accept": false, + "reason": "Both are inpainting dtype/runtime problems, but one is loading in float16 and the other is a mat1/mat2 dtype mismatch during inference; not the same fix." + }, + { + "left": "issue:11853", + "right": "issue:6008", + "accept": false, + "reason": "Both mention IP-Adapter/LoRA interactions, but they target different model combinations and incompatibilities; not a single duplicate." + }, + { + "left": "issue:4305", + "right": "issue:7357", + "accept": false, + "reason": "One is a specific torch.compile graph break from a diffusers version change; the other is general slow inference despite optimizations. Same area, different issue." + }, + { + "left": "issue:2695", + "right": "issue:3405", + "accept": false, + "reason": "Both are ControlNet training problems, but one is a validation/TensorBoard reporting failure and the other is a broader training-process bug; too vague to merge." + }, + { + "left": "issue:12022", + "right": "issue:12037", + "accept": false, + "reason": "Flash-attn interface incompatibility and Wan 2.2 Lightx2v LoRA incompatibility are unrelated concrete problems." + }, + { + "left": "issue:5966", + "right": "issue:8482", + "accept": false, + "reason": "Both are model/pipeline support requests, but one is ControlNet for SD 2.1 and the other is Stable Diffusion 3 Medium support; different targets." + }, + { + "left": "issue:4880", + "right": "issue:5488", + "accept": false, + "reason": "A ControlNet image-type error and a demo image-generation failure are not the same bug." + }, + { + "left": "issue:5757", + "right": "issue:6146", + "accept": false, + "reason": "Both are training-script issues, but one is checkpoint resume errors and the other is multi-GPU training support; different code paths." + }, + { + "left": "issue:4206", + "right": "issue:4985", + "accept": false, + "reason": "ControlNet training broken on main and SDXL InstructPix2Pix OOM are unrelated training failures." + }, + { + "left": "issue:1500", + "right": "issue:3221", + "accept": false, + "reason": "Both are version-compatibility problems, but they affect different APIs and releases; not the same defect." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 128, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10573", + "issue:10795", + "issue:11010", + "issue:11045", + "issue:1185", + "issue:1188", + "issue:12009", + "issue:12113", + "issue:12329", + "issue:12575", + "issue:13043", + "issue:1672", + "issue:1807", + "issue:2324", + "issue:2442", + "issue:2507", + "issue:3348", + "issue:3551", + "issue:3826", + "issue:3878", + "issue:3891", + "issue:4057", + "issue:4117", + "issue:4183", + "issue:4217", + "issue:4646", + "issue:5075", + "issue:5395", + "issue:5474", + "issue:5565", + "issue:5875", + "issue:5953", + "issue:6232", + "issue:6382", + "issue:6517", + "issue:6758", + "issue:6879", + "issue:6964", + "issue:7021", + "issue:7056", + "issue:7133", + "issue:7306", + "issue:7681", + "issue:8834", + "issue:9101", + "issue:9630", + "issue:9643", + "issue:9861" + ], + "soft_pairs": [ + "issue:1672|issue:1807", + "issue:4117|issue:5565", + "issue:1188|issue:3891", + "issue:2442|issue:4057", + "issue:3348|issue:3551", + "issue:8834|issue:9101", + "issue:9630|issue:9861", + "issue:1185|issue:2324", + "issue:5075|issue:6879", + "issue:10573|issue:9643", + "issue:3878|issue:5875", + "issue:5395|issue:6232", + "issue:6382|issue:7306", + "issue:7021|issue:7681", + "issue:4646|issue:6517", + "issue:4183|issue:5953", + "issue:10795|issue:12009", + "issue:11010|issue:11045", + "issue:2507|issue:4217", + "issue:12113|issue:12329", + "issue:3826|issue:5474", + "issue:12575|issue:13043", + "issue:7056|issue:7133", + "issue:6758|issue:6964" + ], + "prepared_review_unit_hash": "1e4373916925de83c904b5a75e144e52dcb46bff752bedcbd53b17c1c3f4c152", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34758, + "estimated_input_tokens": 8690, + "estimated_eval_tokens": 17636 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The items are a heterogeneous set of unrelated issues; the suggested soft pairs share superficial subsystem keywords but not the same underlying bug or change. No PRs are present.", + "confidence": 0.36, + "canonical_issue_reason": "issue:10573 is the clearest concrete bug report in the set, with a specific failing pipeline and an explicit repro context; it is more representative than the broader or more speculative issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10573 is the best overall anchor because it describes a precise, actionable runtime failure rather than a broad question, feature request, or vague symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1672", + "right": "issue:1807", + "accept": false, + "reason": "Wrong image size after conversion and generation slowdown are different defects with different code paths." + }, + { + "left": "issue:4117", + "right": "issue:5565", + "accept": false, + "reason": "SDXL Dreambooth failure and LoRA SDXL script execution issues are separate training problems." + }, + { + "left": "issue:1188", + "right": "issue:3891", + "accept": false, + "reason": "A community request for CLIPSeg+SD is unrelated to a Tortoise TTS pipeline request." + }, + { + "left": "issue:2442", + "right": "issue:4057", + "accept": false, + "reason": "Loss-graph expectations for inpainting training and blurred outputs from a finetuned model are not the same bug." + }, + { + "left": "issue:3348", + "right": "issue:3551", + "accept": false, + "reason": "One is about LoRA scale support under xFormers, the other about LoRA disabling xFormers entirely; related area, but different behavior." + }, + { + "left": "issue:8834", + "right": "issue:9101", + "accept": false, + "reason": "Training code availability for SD3 ControlNet and an InstructPix2Pix training script for SD3 are distinct feature requests." + }, + { + "left": "issue:9630", + "right": "issue:9861", + "accept": false, + "reason": "Multi-GPU/single-GPU training configuration error and transformer not updating during Flux training are different bugs." + }, + { + "left": "issue:1185", + "right": "issue:2324", + "accept": false, + "reason": "Dreambooth reproducibility and instance-data loading are separate issues in the training pipeline." + }, + { + "left": "issue:5075", + "right": "issue:6879", + "accept": false, + "reason": "SDXL refiner prompt-embedding shape mismatch and Custom Diffusion dtype mismatch are different failure modes." + }, + { + "left": "issue:10573", + "right": "issue:9643", + "accept": false, + "reason": "Padding_mask_crop with multiple ControlNets in SDXL inpaint is not the same as the broader question of Flux multiple-ControlNet support." + }, + { + "left": "issue:3878", + "right": "issue:5875", + "accept": false, + "reason": "Pix2Pix training error and ONNX conversion prompt error are unrelated." + }, + { + "left": "issue:5395", + "right": "issue:6232", + "accept": false, + "reason": "LoRA combination problems with PEFT and a scheduler add_noise bug are unrelated code paths." + }, + { + "left": "issue:6382", + "right": "issue:7306", + "accept": false, + "reason": "IP-Adapter key errors in SDXL LoRA training and float16 failure in LEditsPP are different issues." + }, + { + "left": "issue:7021", + "right": "issue:7681", + "accept": false, + "reason": "HF_HOME import error and an OpenSSL symbol/import failure are distinct environment/import problems." + }, + { + "left": "issue:4646", + "right": "issue:6517", + "accept": false, + "reason": "ONNX conversion failure for ControlNet and an import failure for SDXL ControlNet img2img are different bugs." + }, + { + "left": "issue:4183", + "right": "issue:5953", + "accept": false, + "reason": "guidance_scale <= 1 controlnet error and size inconsistency crash are separate ControlNet issues." + }, + { + "left": "issue:10795", + "right": "issue:12009", + "accept": false, + "reason": "torch.compile on GGUF-loaded transformer models and Wan2.2 GGUF loading failure are related only by GGUF/from_single_file, not the same defect." + }, + { + "left": "issue:11010", + "right": "issue:11045", + "accept": false, + "reason": "A Flux architecture-change support request and a crash loading Flux Schnell in Dreambooth are different problems." + }, + { + "left": "issue:2507", + "right": "issue:4217", + "accept": false, + "reason": "Both mention safetensors loading speed, but they do not establish the same specific bug or code path." + }, + { + "left": "issue:12113", + "right": "issue:12329", + "accept": false, + "reason": "Channel-mismatch runtime error in Wan I2V and incorrect Wan2.1/Wan2.2 model selection are different symptoms." + }, + { + "left": "issue:3826", + "right": "issue:5474", + "accept": false, + "reason": "Audio-Journey pipeline request and Stable Diffusion Reference Only pipeline request are separate feature proposals." + }, + { + "left": "issue:12575", + "right": "issue:13043", + "accept": false, + "reason": "Attention-backend checks causing an error and Blackwell kernel-image availability are different backend/device issues." + }, + { + "left": "issue:7056", + "right": "issue:7133", + "accept": false, + "reason": "train_text_encoder parameter error and checkpoint resume text-encoder size mismatch are different training issues." + }, + { + "left": "issue:6758", + "right": "issue:6964", + "accept": false, + "reason": "InstantID num_images_per_prompt support and 4-channel UNet incompatibility are unrelated pipeline bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 129, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10282", + "issue:10514", + "issue:10891", + "issue:11006", + "issue:11133", + "issue:11208", + "issue:11314", + "issue:11321", + "issue:12003", + "issue:12334", + "issue:1604", + "issue:1865", + "issue:1876", + "issue:1979", + "issue:2015", + "issue:2412", + "issue:3174", + "issue:3585", + "issue:3613", + "issue:4185", + "issue:4368", + "issue:4697", + "issue:4936", + "issue:5442", + "issue:5596", + "issue:5689", + "issue:5934", + "issue:6096", + "issue:6131", + "issue:6481", + "issue:6545", + "issue:6552", + "issue:6741", + "issue:6870", + "issue:7197", + "issue:7735", + "issue:7945", + "issue:7974", + "issue:8182", + "issue:858", + "issue:8732", + "issue:8885", + "issue:8955", + "issue:9045", + "issue:9343", + "issue:9476", + "issue:9927", + "issue:9971" + ], + "soft_pairs": [ + "issue:1604|issue:4936", + "issue:1876|issue:1979", + "issue:5934|issue:6096", + "issue:4185|issue:5689", + "issue:11314|issue:11321", + "issue:2015|issue:5442", + "issue:3174|issue:3585", + "issue:4697|issue:8885", + "issue:10891|issue:9476", + "issue:8182|issue:858", + "issue:5596|issue:8732", + "issue:10514|issue:11208", + "issue:6131|issue:7974", + "issue:10282|issue:9045", + "issue:1865|issue:2412", + "issue:6545|issue:6552", + "issue:7197|issue:9927", + "issue:7735|issue:9971", + "issue:11006|issue:11133", + "issue:3613|issue:4368", + "issue:6481|issue:6741", + "issue:6870|issue:7945", + "issue:8955|issue:9343", + "issue:12003|issue:12334" + ], + "prepared_review_unit_hash": "2741407aec9818fe50c8f3395c6836cdb68f4104a79abbf1174376e4687d179b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33977, + "estimated_input_tokens": 8495, + "estimated_eval_tokens": 17246 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a false-positive similarity bucket: the issues span unrelated bugs across loading, training, compilation, attention, and model-specific pipelines. I found no true duplicate pair among the soft edges.", + "confidence": 0.94, + "canonical_issue_reason": "No single canonical issue stands out; the set is too heterogeneous and the soft pairs mostly share only superficial wording like \u201cerror,\u201d \u201cbug,\u201d or a broad subsystem name.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 5934 is the closest representative only because it is a broad, generic torch.compile failure and is the nearest thing to an umbrella bug here, but it still does not cover the rest of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1604", + "right": "issue:4936", + "accept": false, + "reason": "Both are training-script regressions, but one is lpw_stable_diffusion broken on diffusers 0.9.0 and the other is DreamBooth LoRA resume breakage; different code paths and failure modes." + }, + { + "left": "issue:1876", + "right": "issue:1979", + "accept": false, + "reason": "Both concern conversion, but one is about checkpoints missing state_dict and the other is a generic SD-to-diffusers conversion failure; not the same underlying bug." + }, + { + "left": "issue:5934", + "right": "issue:6096", + "accept": false, + "reason": "Both mention torch.compile, but 6096 is SVD-specific DynamicOutputShapeException while 5934 is a broader compile error; not one concrete fix." + }, + { + "left": "issue:4185", + "right": "issue:5689", + "accept": false, + "reason": "Same SDXL/controlnet neighborhood, but one reports black images in ControlNet inference and the other residual noise with DPM++; distinct symptoms and likely distinct causes." + }, + { + "left": "issue:11314", + "right": "issue:11321", + "accept": false, + "reason": "One is a fatal mistake in the FLUX ControlNet training script; the other is a README bug. Different artifact and not the same bug." + }, + { + "left": "issue:2015", + "right": "issue:5442", + "accept": false, + "reason": "Categorical diffusion pipeline versus ScaleCrafter higher-resolution generation; unrelated features with no shared failing code path." + }, + { + "left": "issue:3174", + "right": "issue:3585", + "accept": false, + "reason": "Both involve LoRA/dreambooth, but one is model loading broken by new LoRA features and the other is synthesizing LoRA into SD1.5 failing; not the same issue." + }, + { + "left": "issue:4697", + "right": "issue:8885", + "accept": false, + "reason": "A specific LoRA loading regression after 0.20.0 versus a Custom Diffusion example install/load-token problem; different failure surfaces." + }, + { + "left": "issue:10891", + "right": "issue:9476", + "accept": false, + "reason": "Both are dtype-related, but DCAE internal conversion on bfloat16 and FLUX validation bias/input mismatch are separate bugs." + }, + { + "left": "issue:8182", + "right": "issue:858", + "accept": false, + "reason": "A logger typo and a Parameter.named_children attribute error are clearly unrelated." + }, + { + "left": "issue:5596", + "right": "issue:8732", + "accept": false, + "reason": "pokemon-blip-captions build failure from a dict.convert call versus checkpoint-saving issues in advanced training; not the same path." + }, + { + "left": "issue:10514", + "right": "issue:11208", + "accept": false, + "reason": "Sana 4k resolution-binning/sample_size limitation versus SD3ControlNetInpaintingPipeline MultiControlNet support; different model features and bugs." + }, + { + "left": "issue:6131", + "right": "issue:7974", + "accept": false, + "reason": "Both touch IP-Adapter, but one is IP Adapter Plus not working generally and the other is inpaint with IP Adapter (face+style); too different to merge." + }, + { + "left": "issue:10282", + "right": "issue:9045", + "accept": false, + "reason": "Meta-tensor weight shape mismatch on load versus StableDiffusion3Pipeline CUDA conversion with accelerate/peft; unrelated underlying problems." + }, + { + "left": "issue:1865", + "right": "issue:2412", + "accept": false, + "reason": "Zero Stage3 SD2 dimension error and a generic DreamBooth error are different training failures, not the same bug." + }, + { + "left": "issue:6545", + "right": "issue:6552", + "accept": false, + "reason": "One is about modularizing inference in example scripts, the other about FP16 resume training; they are separate tracker items." + }, + { + "left": "issue:7197", + "right": "issue:9927", + "accept": false, + "reason": "KeyError on resuming SD1.5 text-image training versus HeaderTooLarge in SDv3 ControlNet training; unrelated bugs." + }, + { + "left": "issue:7735", + "right": "issue:9971", + "accept": false, + "reason": "Caption column lookup failure in custom dataset training versus DEISMultistepScheduler not working on FLUX; distinct problems." + }, + { + "left": "issue:11006", + "right": "issue:11133", + "accept": false, + "reason": "Both are video-pipeline reports, but Wan 2.1 I2V with quantized transformer and CogVideoX I2V are different models and code paths." + }, + { + "left": "issue:3613", + "right": "issue:4368", + "accept": false, + "reason": "Both are OOM reports in training, but DreamBooth torch2.0 OOM and SDXL LoRA OOM are not the same concrete bug." + }, + { + "left": "issue:6481", + "right": "issue:6741", + "accept": false, + "reason": "DreamBooth batch-size > 1 failure and missing no_flip arg in Diffusion DPO are unrelated training script issues." + }, + { + "left": "issue:6870", + "right": "issue:7945", + "accept": false, + "reason": "Missing enable_vae_tiling in SD img2img/inpaint and a community-pipeline combination request are not the same bug." + }, + { + "left": "issue:8955", + "right": "issue:9343", + "accept": false, + "reason": "CLIPTextModelWithProjection state_dict loading error versus FLUX loading with low_cpu_mem_usage=False and ignore_mismatched_sizes=True; different load failures." + }, + { + "left": "issue:12003", + "right": "issue:12334", + "accept": false, + "reason": "Meta-tensor copy error versus enable_xformers_memory_efficient_attention bug; no shared concrete code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 130, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10281", + "issue:10282", + "issue:10928", + "issue:1127", + "issue:1434", + "issue:1445", + "issue:1538", + "issue:2052", + "issue:2163", + "issue:2783", + "issue:2838", + "issue:307", + "issue:3233", + "issue:3428", + "issue:3453", + "issue:3612", + "issue:4037", + "issue:4109", + "issue:4185", + "issue:4353", + "issue:4356", + "issue:4453", + "issue:4788", + "issue:4826", + "issue:5224", + "issue:5540", + "issue:5607", + "issue:5621", + "issue:5630", + "issue:5757", + "issue:6028", + "issue:6252", + "issue:6333", + "issue:6382", + "issue:7306", + "issue:7418", + "issue:7441", + "issue:7480", + "issue:7991", + "issue:8414", + "issue:8540", + "issue:8729", + "issue:9047", + "issue:9051", + "issue:9063", + "issue:9101", + "issue:9913", + "issue:9996" + ], + "soft_pairs": [ + "issue:4185|issue:5621", + "issue:4788|issue:6028", + "issue:7418|issue:7991", + "issue:7441|issue:9913", + "issue:8540|issue:9101", + "issue:1434|issue:5540", + "issue:3428|issue:6382", + "issue:1538|issue:4109", + "issue:2163|issue:3453", + "issue:4037|issue:4826", + "issue:2052|issue:6252", + "issue:10281|issue:8414", + "issue:3233|issue:4453", + "issue:4185|issue:8729", + "issue:5224|issue:5607", + "issue:1127|issue:1445", + "issue:10282|issue:9996", + "issue:10928|issue:4356", + "issue:2783|issue:3612", + "issue:2838|issue:307", + "issue:2052|issue:5757", + "issue:6333|issue:9063", + "issue:4353|issue:5630", + "issue:7306|issue:7480", + "issue:9047|issue:9051" + ], + "prepared_review_unit_hash": "e620a1d10000b72bb26fdcc00f4150b3fce4c7164a7248d54a92e5eb53b308de", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34505, + "estimated_input_tokens": 8627, + "estimated_eval_tokens": 17510 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: a mix of unrelated Diffusers feature requests and bug reports across SD, SDXL, Flux, ControlNet, training, and export/loading issues. No pull requests are present, and none of the soft-edge pairs look like true duplicates or mergeable same-fix candidates.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 307 is the broadest umbrella-style request in the set, so it is the least misleading single issue to use as a representative anchor for this mixed cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 307 is the most general, cluster-like issue; the rest are mostly specific runtime bugs or narrowly scoped feature requests, so it best serves as a representative issue despite the cluster's weak coherence.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4185", + "right": "issue:5621", + "accept": false, + "reason": "Both involve ControlNet/SDXL, but one is about black images during inference and the other about unsupported `encoder_hid_proj` models; different failures and likely different fixes." + }, + { + "left": "issue:4788", + "right": "issue:6028", + "accept": false, + "reason": "Both are OOM reports, but for different pipelines/models (SDXL inpaint vs Kandinsky 3.0); not the same underlying bug." + }, + { + "left": "issue:7418", + "right": "issue:7991", + "accept": false, + "reason": "Both are code/tutorial bugs, but they refer to different example files and likely different mistakes." + }, + { + "left": "issue:7441", + "right": "issue:9913", + "accept": false, + "reason": "Python 3.12 pytest failure is unrelated to `set_lora_device` not working; different subsystem and symptom." + }, + { + "left": "issue:8540", + "right": "issue:9101", + "accept": false, + "reason": "Both are training-script feature requests, but one asks for SD3 Dreambooth caching and the other for an SD3 InstructPix2Pix script; not the same change." + }, + { + "left": "issue:1434", + "right": "issue:5540", + "accept": false, + "reason": "Both mention memory, but one is a specific upscale-pipeline OOM and the other is a generic inference OOM report." + }, + { + "left": "issue:3428", + "right": "issue:6382", + "accept": false, + "reason": "These are different problems: an unexpected result when modifying Img2ImgPipeline versus an IP-Adapter key error after LCM LoRA training." + }, + { + "left": "issue:1538", + "right": "issue:4109", + "accept": false, + "reason": "Both are pipeline-specific breakages, but one is about StableDiffusionMegaPipeline with SD 2-base and the other is SDXL inpaint not working; separate code paths." + }, + { + "left": "issue:2163", + "right": "issue:3453", + "accept": false, + "reason": "Both are training-script runtime errors, but one is dtype mismatch in `train_text_to_image.py` and the other is BFloat16 CUDA `triu_tril` support in a different training script." + }, + { + "left": "issue:4037", + "right": "issue:4826", + "accept": false, + "reason": "Both involve ControlNet training/reference behavior, but one is FSDP validation errors and the other is MultiControlNet reference errors; different failure modes." + }, + { + "left": "issue:2052", + "right": "issue:6252", + "accept": false, + "reason": "Both involve checkpoint/weights loading, but one is Textual Inversion resume support and the other is LoRA weights not loading from checkpoint; not the same bug." + }, + { + "left": "issue:10281", + "right": "issue:8414", + "accept": false, + "reason": "A new scheduler request and a new text-to-audio model request are unrelated feature proposals." + }, + { + "left": "issue:3233", + "right": "issue:4453", + "accept": false, + "reason": "Both are about LoRA conversion/export, but one is generic trained LoRA to safetensor and the other is SDXL LoRA interoperability with AUTO1111; distinct requests." + }, + { + "left": "issue:4185", + "right": "issue:8729", + "accept": false, + "reason": "Both are SDXL ControlNet training/inference issues, but black inference images and unexpected validation image generation are different behaviors." + }, + { + "left": "issue:5224", + "right": "issue:5607", + "accept": false, + "reason": "Flax Stable Diffusion 2.1 error and tensor-size mismatch are unrelated errors in different components." + }, + { + "left": "issue:1127", + "right": "issue:1445", + "accept": false, + "reason": "Both are ONNX pipeline issues, but one is an import traceback with DML execution provider and the other is an AttributeError in legacy inpainting; different concrete bugs." + }, + { + "left": "issue:10282", + "right": "issue:9996", + "accept": false, + "reason": "The model-load shape mismatch and the NF4 transformer-loading issue are separate loading/weight-format problems." + }, + { + "left": "issue:10928", + "right": "issue:4356", + "accept": false, + "reason": "A missing FP16 model file in a local UNet directory is unrelated to a module import error in diffusers." + }, + { + "left": "issue:2783", + "right": "issue:3612", + "accept": false, + "reason": "Both mention ControlNet creation, but one asks for an img2img/inpainting ControlNet pipeline and the other asks how to create a model-free ControlNet with a specific preprocessor; not the same change." + }, + { + "left": "issue:2838", + "right": "issue:307", + "accept": false, + "reason": "One asks about ControlNet pipeline behavior while the other is a broad request to merge Stable Diffusion pipelines; different scope and intent." + }, + { + "left": "issue:2052", + "right": "issue:5757", + "accept": false, + "reason": "Both are checkpoint resume problems, but they affect different scripts and training flows." + }, + { + "left": "issue:6333", + "right": "issue:9063", + "accept": false, + "reason": "A PEFT LoRA finetuning request and a Flux tokenizer-instantiation error are unrelated." + }, + { + "left": "issue:4353", + "right": "issue:5630", + "accept": false, + "reason": "Both concern SDXL image quality, but one is a generic poor-quality report and the other is a specific LMS scheduler artifact in Img2Img; not the same bug." + }, + { + "left": "issue:7306", + "right": "issue:7480", + "accept": false, + "reason": "Both are memory/precision-related, but one is float16 support in LEditsPPPipelineStableDiffusionXL and the other is generic OOM in DeepFloyd IF; different issues." + }, + { + "left": "issue:9047", + "right": "issue:9051", + "accept": false, + "reason": "Both are Flux-related, but one is MPS device support and the other is batch inference support; separate feature/compatibility requests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 131, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10533", + "issue:1076", + "issue:11160", + "issue:11258", + "issue:11581", + "issue:1188", + "issue:11989", + "issue:12329", + "issue:13008", + "issue:1305", + "issue:13105", + "issue:13232", + "issue:1445", + "issue:1851", + "issue:2695", + "issue:2776", + "issue:3284", + "issue:3391", + "issue:3613", + "issue:3812", + "issue:3984", + "issue:4733", + "issue:4885", + "issue:5004", + "issue:5025", + "issue:5442", + "issue:5798", + "issue:5917", + "issue:5931", + "issue:6252", + "issue:6336", + "issue:7237", + "issue:7306", + "issue:7341", + "issue:7432", + "issue:7722", + "issue:8708", + "issue:8922", + "issue:8970", + "issue:9148", + "issue:9167", + "issue:9171", + "issue:9178", + "issue:9665", + "issue:9861", + "issue:987", + "issue:9873", + "issue:9911" + ], + "soft_pairs": [ + "issue:9178|issue:9911", + "issue:10533|issue:11989", + "issue:3284|issue:4733", + "issue:1305|issue:5442", + "issue:2776|issue:3984", + "issue:5004|issue:8708", + "issue:5025|issue:9167", + "issue:5931|issue:9171", + "issue:9148|issue:9873", + "issue:9665|issue:9861", + "issue:11160|issue:13105", + "issue:1445|issue:5917", + "issue:3391|issue:987", + "issue:4885|issue:6252", + "issue:6336|issue:7722", + "issue:8922|issue:8970", + "issue:2695|issue:3812", + "issue:1076|issue:1188", + "issue:13008|issue:13232", + "issue:7237|issue:7341", + "issue:11258|issue:12329", + "issue:1851|issue:3613", + "issue:11581|issue:5798", + "issue:7306|issue:7432" + ], + "prepared_review_unit_hash": "33a0f6a84a2c84a0ca0963359536c1d1c910dd945a6d5e699128fca353fce22c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34171, + "estimated_input_tokens": 8543, + "estimated_eval_tokens": 17342 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: the items share broad themes like generation failures, training bugs, or pipeline feature requests, but the soft pairs do not look like the same underlying defect/change. I would not consolidate any of the candidate pairs.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue: the set spans many unrelated models, pipelines, and bug classes, and the apparent overlaps are only superficial symptom-level similarities.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "No global best issue for deduplication purposes; the issues are too diverse to use one as a representative canonical target.", + "best_pr_reason": "No pull requests are present in this cluster.", + "soft_edge_verdicts": [ + { + "left": "issue:9178", + "right": "issue:9911", + "accept": false, + "reason": "Both mention FLUX training/controlnet, but they describe different failures in different workflows; not the same concrete bug." + }, + { + "left": "issue:10533", + "right": "issue:11989", + "accept": false, + "reason": "Both mention black images, but one is FLUX fill fp16 and the other is a generic inference black-image report tied to weights; too broad to be the same bug." + }, + { + "left": "issue:3284", + "right": "issue:4733", + "accept": false, + "reason": "Both involve Dreambooth/accelerate, but one is a general run failure and the other is a checkpoint-resume optimizer issue; different failure modes." + }, + { + "left": "issue:1305", + "right": "issue:5442", + "accept": false, + "reason": "These are different pipeline feature requests, not the same change." + }, + { + "left": "issue:2776", + "right": "issue:3984", + "accept": false, + "reason": "One is a DeepSpeed/version requirement issue for Dreambooth on 8GB, the other is a class_labels ValueError; unrelated bugs." + }, + { + "left": "issue:5004", + "right": "issue:8708", + "accept": false, + "reason": "Both are fp16/Dreambooth-related, but one is resume-from-checkpoint in SDXL and the other is a missing VAE shift_factor in SD3 examples; not the same root cause." + }, + { + "left": "issue:5025", + "right": "issue:9167", + "accept": false, + "reason": "An import error for randn_tensor and an import problem for StableDiffusion3Pipeline are distinct issues." + }, + { + "left": "issue:5931", + "right": "issue:9171", + "accept": false, + "reason": "Determinism in StableDiffusionPipeline is unrelated to from_single_file handling of v_prediction models." + }, + { + "left": "issue:9148", + "right": "issue:9873", + "accept": false, + "reason": "Both are feature requests for new pipelines, but they request different models and capabilities." + }, + { + "left": "issue:9665", + "right": "issue:9861", + "accept": false, + "reason": "Both are FLUX training issues, but one is about accelerator.prepare/DeepSpeed config and the other about the transformer not updating; different code-path problems." + }, + { + "left": "issue:11160", + "right": "issue:13105", + "accept": false, + "reason": "Both mention WAN 2.1, but one is inconsistent inference in I2V and the other is noisy output in a T2V model; not clearly the same defect." + }, + { + "left": "issue:1445", + "right": "issue:5917", + "accept": false, + "reason": "Different errors in different tooling paths: ONNX inpaint legacy attribute error vs checkpoint conversion failure after upgrade." + }, + { + "left": "issue:3391", + "right": "issue:987", + "accept": false, + "reason": "Both describe bad image output, but they involve different pipelines and different symptom types (overexposure vs green image)." + }, + { + "left": "issue:4885", + "right": "issue:6252", + "accept": false, + "reason": "Both are about LoRA loading, but one is SDXL ControlNet Img2Img and the other is checkpoint loading generally; not enough evidence of the same bug." + }, + { + "left": "issue:6336", + "right": "issue:7722", + "accept": false, + "reason": "These are unrelated pipeline errors affecting different components and use cases." + }, + { + "left": "issue:8922", + "right": "issue:8970", + "accept": false, + "reason": "Both produce black images, but they are different models and environments (SD3 single-file vs Kolors on MPS), so this is only symptom-level similarity." + }, + { + "left": "issue:2695", + "right": "issue:3812", + "accept": false, + "reason": "ControlNet training validation/reporting and missing controlnet config are different issues." + }, + { + "left": "issue:1076", + "right": "issue:1188", + "accept": false, + "reason": "These are unrelated community/help requests from different eras and topics." + }, + { + "left": "issue:13008", + "right": "issue:13232", + "accept": false, + "reason": "Different feature requests for different models (FLUX LoRA multi-image training vs LTX-2.3 support)." + }, + { + "left": "issue:7237", + "right": "issue:7341", + "accept": false, + "reason": "One is a missing scheduler class, the other is a TorchVersion/Version comparison TypeError; unrelated." + }, + { + "left": "issue:11258", + "right": "issue:12329", + "accept": false, + "reason": "Different single-file model-loading issues for different model families (LTX vs Wan)." + }, + { + "left": "issue:1851", + "right": "issue:3613", + "accept": false, + "reason": "Both concern Dreambooth performance/resource usage, but one is multi-GPU slowness and the other is OOM with torch 2.0; different problems." + }, + { + "left": "issue:11581", + "right": "issue:5798", + "accept": false, + "reason": "Both are TypeErrors, but they occur in different APIs and with different causes; not the same fix." + }, + { + "left": "issue:7306", + "right": "issue:7432", + "accept": false, + "reason": "I cannot use float16 in one pipeline is unrelated to generating images that differ from WebUI." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 132, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10147", + "issue:10337", + "issue:10540", + "issue:10612", + "issue:1076", + "issue:10866", + "issue:11088", + "issue:11092", + "issue:11418", + "issue:11561", + "issue:11878", + "issue:11922", + "issue:11978", + "issue:1261", + "issue:12666", + "issue:13043", + "issue:13142", + "issue:1392", + "issue:1574", + "issue:1734", + "issue:1934", + "issue:2053", + "issue:2107", + "issue:228", + "issue:2420", + "issue:2480", + "issue:2495", + "issue:259", + "issue:3325", + "issue:3451", + "issue:3468", + "issue:3885", + "issue:4080", + "issue:4427", + "issue:4542", + "issue:4830", + "issue:5769", + "issue:6928", + "issue:7185", + "issue:7405", + "issue:7931", + "issue:8414", + "issue:8626", + "issue:8667", + "issue:9366", + "issue:964", + "issue:9744" + ], + "soft_pairs": [ + "issue:3451|issue:3885", + "issue:10337|issue:10612", + "issue:11418|issue:11922", + "issue:11978|issue:8414", + "issue:12666|issue:13142", + "issue:1574|issue:1734", + "issue:2107|issue:2495", + "issue:228|issue:259", + "issue:4830|issue:8667", + "issue:5769|issue:7185", + "issue:7405|issue:9744", + "issue:1261|issue:964", + "issue:11088|issue:11878", + "issue:2420|issue:3325", + "issue:10147|issue:11561", + "issue:4080|issue:6928", + "issue:11092|issue:9366", + "issue:1934|issue:2053", + "issue:2480|issue:3468", + "issue:10866|issue:13043", + "issue:4427|issue:7931", + "issue:4542|issue:8626", + "issue:10010|issue:10540", + "issue:1076|issue:1392" + ], + "prepared_review_unit_hash": "955460e2fb562d583b760f3e39be0939e7778bbfa501b310c309e095eede4251", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33946, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous. The only strong duplicate-like pair is the SDXL img2img timestep/strength failure pair; the rest differ by model, pipeline, or bug class and should stay separate.", + "confidence": 0.74, + "canonical_issue_reason": "issue:9366 is the broader, earlier report for the img2img timestep/strength failure family, so it works best as the canonical issue if these are grouped.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11092 is the most detailed reproduction of the same img2img timestep/strength code-path problem, making it the best representative issue for triage.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3451", + "right": "issue:3885", + "accept": false, + "reason": "Both mention VRAM usage in training, but one is a specific fp16-vs-fp32 memory regression and the other is a general \u201chow do I reduce VRAM\u201d question, not the same bug." + }, + { + "left": "issue:10337", + "right": "issue:10612", + "accept": false, + "reason": "Both involve Flux and quantization, but one is IP-Adapter failure and the other is LoRA loading in NF4 quantized Flux Fill; different adapters and code paths." + }, + { + "left": "issue:11418", + "right": "issue:11922", + "accept": false, + "reason": "One is about adding a specific fp8 safetensors file, the other is a vague FLUX.1-Fill-dev runtime problem; too little overlap to treat as the same bug." + }, + { + "left": "issue:11978", + "right": "issue:8414", + "accept": false, + "reason": "Different models and different feature requests; no shared underlying bug or change." + }, + { + "left": "issue:12666", + "right": "issue:13142", + "accept": false, + "reason": "These are unrelated breakages in different pipelines/model stacks; one is a video pipeline missing in a release, the other is a transformers 5.1.0 regression." + }, + { + "left": "issue:1574", + "right": "issue:1734", + "accept": false, + "reason": "Both are Dreambooth training issues, but one is a crash on 8GB GPU and the other is a multi-GPU slowdown; different problems." + }, + { + "left": "issue:2107", + "right": "issue:2495", + "accept": false, + "reason": "Both mention black/glitchy output, but they affect different pipelines and different conditions, so they do not look like the same bug." + }, + { + "left": "issue:228", + "right": "issue:259", + "accept": false, + "reason": "A broken model link and a progress-bar fix are unrelated." + }, + { + "left": "issue:4830", + "right": "issue:8667", + "accept": false, + "reason": "Both are tensor-size issues in controlnet-adjacent code, but one is MultiControlNet and the other is non-power-of-two image sizing in SD3ControlNetModel; not the same defect." + }, + { + "left": "issue:5769", + "right": "issue:7185", + "accept": false, + "reason": "Mixed-precision training correctness and PyTorch version support are different concerns." + }, + { + "left": "issue:7405", + "right": "issue:9744", + "accept": false, + "reason": "These are separate new model additions, not a duplicate bug or change." + }, + { + "left": "issue:1261", + "right": "issue:964", + "accept": false, + "reason": "Xformers backward-pass failure and an inpainting example not working are unrelated." + }, + { + "left": "issue:11088", + "right": "issue:11878", + "accept": false, + "reason": "Both involve Wan GGUF loading, but they target different model classes and failures; not clearly the same underlying issue." + }, + { + "left": "issue:2420", + "right": "issue:3325", + "accept": false, + "reason": "Different training scripts, different failures, and no shared concrete code-path bug." + }, + { + "left": "issue:10147", + "right": "issue:11561", + "accept": false, + "reason": "Support for MV-Adapter and support for loading IP-Adapter in FluxFill are distinct feature requests." + }, + { + "left": "issue:4080", + "right": "issue:6928", + "accept": false, + "reason": "An SDXL inpainting pipeline request and a Marigold depth colorization option are unrelated changes." + }, + { + "left": "issue:11092", + "right": "issue:9366", + "accept": true, + "reason": "Both describe img2img failures around step/strength-derived timesteps; they point to the same scheduler/timestep code path and could plausibly be merged into one bug report." + }, + { + "left": "issue:1934", + "right": "issue:2053", + "accept": false, + "reason": "Both are memory-related, but one concerns cpu_offload VRAM usage and the other a version regression in overall VRAM consumption." + }, + { + "left": "issue:2480", + "right": "issue:3468", + "accept": false, + "reason": "Both are Dreambooth checkpoint/export issues, but one is missing per-checkpoint unet folders and the other is missing model_index.json; different outputs and failures." + }, + { + "left": "issue:10866", + "right": "issue:13043", + "accept": false, + "reason": "Completely different models and failure modes; no duplicate relationship." + }, + { + "left": "issue:4427", + "right": "issue:7931", + "accept": false, + "reason": "A multi-adapter example failure and an IP-Adapter embedding issue in SDXL ControlNet inpaint are different adapter paths." + }, + { + "left": "issue:4542", + "right": "issue:8626", + "accept": false, + "reason": "One is about generating more than one image, the other is documentation/guidance for multiple IP-Adapter images and masks; not the same bug." + }, + { + "left": "issue:10010", + "right": "issue:10540", + "accept": false, + "reason": "Both are Flux Fill-related, but one is a missing model_index.json URL and the other is a from_single_file regression after adding Flux Fill support; different problems." + }, + { + "left": "issue:1076", + "right": "issue:1392", + "accept": false, + "reason": "Unrelated issues with no meaningful overlap." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The heterogeneity summary is supported by the packet, and the only accepted soft pair (issue:11092 / issue:9366) is a conservative same-family match around img2img timestep/strength behavior. The other soft-edge rejections are appropriately cautious and do not overstate duplicate evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 133, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10591", + "issue:11581", + "issue:11864", + "issue:11978", + "issue:12107", + "issue:13107", + "issue:13249", + "issue:13261", + "issue:13484", + "issue:1566", + "issue:1657", + "issue:2375", + "issue:2606", + "issue:2856", + "issue:3065", + "issue:4109", + "issue:4265", + "issue:4269", + "issue:4286", + "issue:4398", + "issue:4530", + "issue:4551", + "issue:4598", + "issue:4880", + "issue:4925", + "issue:4974", + "issue:5023", + "issue:5433", + "issue:5488", + "issue:5509", + "issue:5515", + "issue:5544", + "issue:5743", + "issue:6481", + "issue:6794", + "issue:7021", + "issue:7431", + "issue:7681", + "issue:8557", + "issue:9042", + "issue:9069", + "issue:9089", + "issue:9109", + "issue:9393", + "issue:9454", + "issue:9497", + "issue:9546", + "issue:9613" + ], + "soft_pairs": [ + "issue:11581|issue:9069", + "issue:1657|issue:3065", + "issue:4269|issue:4880", + "issue:4109|issue:4286", + "issue:7431|issue:9497", + "issue:13249|issue:13261", + "issue:11864|issue:2375", + "issue:4598|issue:9109", + "issue:11978|issue:9042", + "issue:5544|issue:7681", + "issue:6481|issue:9393", + "issue:1566|issue:4551", + "issue:12107|issue:9613", + "issue:5488|issue:5509", + "issue:10591|issue:8557", + "issue:2856|issue:4974", + "issue:4265|issue:6794", + "issue:4398|issue:4530", + "issue:4925|issue:9546", + "issue:9089|issue:9454", + "issue:2606|issue:5743", + "issue:13107|issue:13484", + "issue:5023|issue:7021", + "issue:5433|issue:5515" + ], + "prepared_review_unit_hash": "351905f6af6fa2d66d5fc4597a6e78856ccfda38982fbe471a9c06c60fc60b60", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34168, + "estimated_input_tokens": 8542, + "estimated_eval_tokens": 17340 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issues; none of the soft pairs look like true duplicates, and there are no PRs in this set.", + "confidence": 0.88, + "canonical_issue_reason": "issue:13261 is the most detailed and actively discussed report, so it is the best single anchor if one issue must represent this mixed set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:13261 has the richest context and strongest ongoing discussion, making it the best issue for triage/monitoring among these items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11581", + "right": "issue:9069", + "accept": false, + "reason": "Different bugs: FLUX IP-Adapter init dtype argument error vs np.ndarray type mismatch. Same framework family, not the same failure." + }, + { + "left": "issue:1657", + "right": "issue:3065", + "accept": false, + "reason": "Kaggle notebook request is unrelated to adding text-encoder LoRA support in DreamBooth." + }, + { + "left": "issue:4269", + "right": "issue:4880", + "accept": false, + "reason": "Both are TypeErrors in ControlNet, but one is about check_inputs defaults and the other about missing/invalid image input handling." + }, + { + "left": "issue:4109", + "right": "issue:4286", + "accept": false, + "reason": "StableDiffusionXLInpaintPipeline failure is unrelated to SDXL load_lora_weights not working." + }, + { + "left": "issue:7431", + "right": "issue:9497", + "accept": false, + "reason": "Different training failures on different model stacks: DoRA/text-encoder issues vs Flux matmul shape error." + }, + { + "left": "issue:13249", + "right": "issue:13261", + "accept": false, + "reason": "Both mention LoRA on Flux/Z-Image, but they concern different models and different symptoms; not enough to treat as one bug." + }, + { + "left": "issue:11864", + "right": "issue:2375", + "accept": false, + "reason": "Both are NameError-style failures, but one is torch.compile/fullgraph scoping and the other is a missing safe_open import." + }, + { + "left": "issue:4598", + "right": "issue:9109", + "accept": false, + "reason": "One is about improving missing dependency errors; the other is a specific torchvision::nms runtime failure." + }, + { + "left": "issue:11978", + "right": "issue:9042", + "accept": false, + "reason": "Separate feature requests for different pipelines (Bria 3.2 vs Flux)." + }, + { + "left": "issue:5544", + "right": "issue:7681", + "accept": false, + "reason": "Unrelated issues: FreeU disable handling in SDXL vs a low-level import failure from lib/OpenSSL." + }, + { + "left": "issue:6481", + "right": "issue:9393", + "accept": false, + "reason": "Different DreamBooth training problems: batch-size >1 behavior vs DeepSpeed save-model failure." + }, + { + "left": "issue:1566", + "right": "issue:4551", + "accept": false, + "reason": "Both are DreamBooth crashes, but the repros and failure points are different (checkpoint saving with fp16 output vs SDXL base training crash)." + }, + { + "left": "issue:12107", + "right": "issue:9613", + "accept": false, + "reason": "Custom tracker object handling in init_trackers is unrelated to the Timesteps.scale attribute error." + }, + { + "left": "issue:5488", + "right": "issue:5509", + "accept": false, + "reason": "Different demo/input-shape problems: one image generation not working vs image/prompt batch-size mismatch." + }, + { + "left": "issue:10591", + "right": "issue:8557", + "accept": false, + "reason": "SD3 LoRA training script problem is too vague to equate with the specific SD3 token-limit training issue." + }, + { + "left": "issue:2856", + "right": "issue:4974", + "accept": false, + "reason": "Both involve conversion to diffusers, but one is a general model conversion problem and the other is about .safetensor conversion; not the same bug." + }, + { + "left": "issue:4265", + "right": "issue:6794", + "accept": false, + "reason": "Different import errors in different modules: ControlNet pipeline export vs training_utils helper import." + }, + { + "left": "issue:4398", + "right": "issue:4530", + "accept": false, + "reason": "Both are SDXL image-quality problems, but black outputs from LoRA-dreambooth are not the same as refiner color artifacts." + }, + { + "left": "issue:4925", + "right": "issue:9546", + "accept": false, + "reason": "Both are OOM reports in ControlNet training, but they hit different models and steps, so they are not clearly the same bug." + }, + { + "left": "issue:9089", + "right": "issue:9454", + "accept": false, + "reason": "Both are missing-model issues, but they refer to entirely different models/pipelines (Flux vs CogVideoX)." + }, + { + "left": "issue:2606", + "right": "issue:5743", + "accept": false, + "reason": "DeepSpeed save_state hang is unrelated to LCD training stuck at 0%." + }, + { + "left": "issue:13107", + "right": "issue:13484", + "accept": false, + "reason": "Both are Flux LoRA loading problems, but the models and failure modes differ enough that they do not look like one concrete bug." + }, + { + "left": "issue:5023", + "right": "issue:7021", + "accept": false, + "reason": "Wuerstchen pipeline import error and HF_HOME import error are unrelated." + }, + { + "left": "issue:5433", + "right": "issue:5515", + "accept": false, + "reason": "Scheduler-related, but one is SDXL visual artifacts with DPM++ and the other is an LCM pipeline initialization failure without a scheduler." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 134, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10239", + "issue:10453", + "issue:10674", + "issue:10863", + "issue:11436", + "issue:1475", + "issue:1931", + "issue:2577", + "issue:3111", + "issue:3174", + "issue:3418", + "issue:3468", + "issue:3595", + "issue:3623", + "issue:3661", + "issue:3700", + "issue:3876", + "issue:4236", + "issue:4321", + "issue:4427", + "issue:4593", + "issue:4911", + "issue:4975", + "issue:5044", + "issue:5509", + "issue:5894", + "issue:5963", + "issue:6176", + "issue:6506", + "issue:6560", + "issue:6815", + "issue:6974", + "issue:7168", + "issue:7347", + "issue:7348", + "issue:7483", + "issue:761", + "issue:8287", + "issue:8871", + "issue:8922", + "issue:9007", + "issue:9051", + "issue:9069", + "issue:9112", + "issue:9164", + "issue:9393", + "issue:9914" + ], + "soft_pairs": [ + "issue:5894|issue:9112", + "issue:9164|issue:9393", + "issue:7347|issue:8871", + "issue:3468|issue:3700", + "issue:10239|issue:10453", + "issue:1475|issue:761", + "issue:3111|issue:3418", + "issue:1931|issue:4427", + "issue:10674|issue:9051", + "issue:4236|issue:6974", + "issue:4911|issue:8287", + "issue:5044|issue:5963", + "issue:10674|issue:7483", + "issue:10863|issue:9914", + "issue:6176|issue:6560", + "issue:3174|issue:4975", + "issue:3623|issue:4593", + "issue:3595|issue:3661", + "issue:6815|issue:9007", + "issue:11436|issue:9069", + "issue:3876|issue:7348", + "issue:5509|issue:6506", + "issue:2577|issue:7168", + "issue:4321|issue:8922" + ], + "prepared_review_unit_hash": "21b108b5b758e36ae6342eb99070b1f9efd4ecf1dcb4967e706e8ee2349ae2d2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33430, + "estimated_input_tokens": 8358, + "estimated_eval_tokens": 16972 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issue pairs; only the from_ckpt reports appear to be the same underlying bug. No PRs are present in this cluster.", + "confidence": 0.72, + "canonical_issue_reason": "issue:3595 is the broad root report for from_ckpt loading failures; issue:3661 is the narrower Stable Diffusion 2.x instance of the same loader problem.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:3595 is the best representative because it captures the general from_ckpt failure, while issue:3661 is a specific subtype/duplicate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5894", + "right": "issue:9112", + "accept": false, + "reason": "Both are runtime inference errors, but the titles point to different failures: invalid buffer size vs tensor shape/dtype mismatch in Auraflow." + }, + { + "left": "issue:9164", + "right": "issue:9393", + "accept": false, + "reason": "Both involve Flux DreamBooth training, but one is a convergence problem and the other is a model-saving failure under deepspeed." + }, + { + "left": "issue:7347", + "right": "issue:8871", + "accept": false, + "reason": "Both are LoRA training complaints, but the concrete failures differ: a generic post-update training breakage vs FP16 gradient unscale errors." + }, + { + "left": "issue:3468", + "right": "issue:3700", + "accept": false, + "reason": "One is about missing model_index.json after DreamBooth training; the other is about the min_snr argument in a LoRA training script." + }, + { + "left": "issue:10239", + "right": "issue:10453", + "accept": false, + "reason": "Same model family, but different problems: sequential CPU offload support vs batch size > 1 support." + }, + { + "left": "issue:1475", + "right": "issue:761", + "accept": false, + "reason": "DreamBooth-related, but one is about mixing two image classes for training and the other is an 8-bit Adam optimizer failure." + }, + { + "left": "issue:3111", + "right": "issue:3418", + "accept": false, + "reason": "Both mention conditioning/text encoders, but one requests a custom encoder for conditioning and the other asks to add text-encoder handling to a LoRA training script." + }, + { + "left": "issue:1931", + "right": "issue:4427", + "accept": false, + "reason": "Float16 scalar-type mismatch during SD2 inference is unrelated to the T2I Multiadapter example not working." + }, + { + "left": "issue:10674", + "right": "issue:9051", + "accept": false, + "reason": "Flux GGUF loading and Flux batch inference are different code paths and different bugs." + }, + { + "left": "issue:4236", + "right": "issue:6974", + "accept": false, + "reason": "One is about SDXL refiner text-encoder registration; the other is about unloading tokenizer/text-encoder components for textual inversion." + }, + { + "left": "issue:4911", + "right": "issue:8287", + "accept": false, + "reason": "Both involve text_encoder handling, but one is custom text_encoder loading from_single_file and the other is the text_encoder=None path." + }, + { + "left": "issue:5044", + "right": "issue:5963", + "accept": false, + "reason": "Different Kandinsky versions and different failure modes: loading failure vs embeds-instead-of-prompts handling." + }, + { + "left": "issue:10674", + "right": "issue:7483", + "accept": false, + "reason": "Flux GGUF loading is unrelated to LoRA-to-Diffusers conversion for SDXL-base models." + }, + { + "left": "issue:10863", + "right": "issue:9914", + "accept": false, + "reason": "Loading local safetensors files and loading trained Flux LoRA weights are not the same underlying bug." + }, + { + "left": "issue:6176", + "right": "issue:6560", + "accept": false, + "reason": "Both are LoRA-loading related, but the titles suggest different contexts and failure modes; not enough evidence of the same concrete bug." + }, + { + "left": "issue:3174", + "right": "issue:4975", + "accept": false, + "reason": "One reports LoRA+DreamBooth loading breakage from new features; the other reports inefficient datatype queries during LoRA loading." + }, + { + "left": "issue:3623", + "right": "issue:4593", + "accept": false, + "reason": "Unable to load larger models is not the same as LoRA rank being too high for Automatic1111 conversion/loading." + }, + { + "left": "issue:3595", + "right": "issue:3661", + "accept": true, + "reason": "These are the same from_ckpt loading bug, with issue:3661 being a specific Stable Diffusion 2.x manifestation of issue:3595." + }, + { + "left": "issue:6815", + "right": "issue:9007", + "accept": false, + "reason": "Both end up with black output, but the specific errors differ: invalid value during cast vs scheduler order-related black images." + }, + { + "left": "issue:11436", + "right": "issue:9069", + "accept": false, + "reason": "One is a dtype mismatch in Hidream inference; the other is a numpy type mismatch error when expecting np.ndarray." + }, + { + "left": "issue:3876", + "right": "issue:7348", + "accept": false, + "reason": "Both mention StableDiffusionControlNetInpaintPipeline, but one is a demo error and the other is a scheduler-quality regression; not clearly the same bug." + }, + { + "left": "issue:5509", + "right": "issue:6506", + "accept": false, + "reason": "Image batch-size mismatch and StableVideoDiffusion returning a list instead of np.ndarray are different API/shape issues." + }, + { + "left": "issue:2577", + "right": "issue:7168", + "accept": false, + "reason": "DDIMInverseScheduler trouble and IP-Adapter image-embed shape confusion are unrelated." + }, + { + "left": "issue:4321", + "right": "issue:8922", + "accept": false, + "reason": "Both produce black outputs, but one is DirectML FP16 ONNX behavior and the other is SD3 single-file quantized-T5 zero-output behavior; different causes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: all items are issues, no PRs are present, and the only clearly mergeable/same-bug pair is the from_ckpt reports (3595/3661). The other soft-edge rejections stay conservative and do not overclaim equivalence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 135, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10025", + "issue:10797", + "issue:10954", + "issue:11006", + "issue:11050", + "issue:11418", + "issue:11464", + "issue:1185", + "issue:12037", + "issue:12071", + "issue:12107", + "issue:122", + "issue:1398", + "issue:1475", + "issue:1534", + "issue:1537", + "issue:1572", + "issue:1672", + "issue:1949", + "issue:2036", + "issue:2107", + "issue:2371", + "issue:2552", + "issue:3222", + "issue:3238", + "issue:4101", + "issue:4255", + "issue:435", + "issue:4398", + "issue:4476", + "issue:4932", + "issue:4954", + "issue:5510", + "issue:5570", + "issue:6351", + "issue:6520", + "issue:6686", + "issue:7033", + "issue:7142", + "issue:7801", + "issue:861", + "issue:8668", + "issue:8729", + "issue:9136", + "issue:9179", + "issue:9223", + "issue:9717", + "issue:9971" + ], + "soft_pairs": [ + "issue:7033|issue:9136", + "issue:10797|issue:9971", + "issue:1534|issue:435", + "issue:3238|issue:861", + "issue:11050|issue:7801", + "issue:2107|issue:4398", + "issue:3222|issue:6686", + "issue:11464|issue:9717", + "issue:4476|issue:8729", + "issue:6520|issue:9179", + "issue:4255|issue:7142", + "issue:5570|issue:9223", + "issue:5510|issue:8668", + "issue:11006|issue:12037", + "issue:10025|issue:12071", + "issue:10954|issue:11418", + "issue:1185|issue:2371", + "issue:12107|issue:6351", + "issue:122|issue:4101", + "issue:1398|issue:1537", + "issue:1475|issue:4954", + "issue:1572|issue:4932", + "issue:1672|issue:2036", + "issue:1949|issue:2552" + ], + "prepared_review_unit_hash": "6743d62153b62ea1135f64d678bfda7402ae266adbc821be44c054fecb256e67", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34070, + "estimated_input_tokens": 8518, + "estimated_eval_tokens": 17292 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster looks mostly like topic-similarity false positives rather than true duplicates. The only mild thematic overlap is around Flux/attention-mask or training/runtime issues, but the paired items still describe different pipelines, symptoms, or code paths. I would keep the Flux attention-mask issue as the best representative of the cluster theme.", + "confidence": 0.72, + "canonical_issue_reason": "issue:10025 is the clearest core Flux transformer attention/mask report and best represents the small thematic overlap in the cluster; the other items are mostly unrelated training, conversion, or pipeline failures.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10025 is the most central and specific issue among the set for the only recurring theme (Flux transformer behavior).", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7033", + "right": "issue:9136", + "accept": false, + "reason": "Same broad IP-Adapter area, but one is attention-mask warning logs and the other is output quality at some resolutions; different bug/symptom." + }, + { + "left": "issue:10797", + "right": "issue:9971", + "accept": false, + "reason": "Different pipelines and failures: HunyuanVideo group offloading vs FLUX scheduler behavior." + }, + { + "left": "issue:1534", + "right": "issue:435", + "accept": false, + "reason": "Both are training notebook problems, but they are distinct errors in different notebooks and code paths." + }, + { + "left": "issue:3238", + "right": "issue:861", + "accept": false, + "reason": "Both involve model conversion, but one is LoRA safetensor conversion and the other is checkpoint-to-diffusers conversion; not the same issue." + }, + { + "left": "issue:11050", + "right": "issue:7801", + "accept": false, + "reason": "Both touch mixed precision, but one is a specific tensor-casting bug in a training script and the other is a model dtype expectation mismatch." + }, + { + "left": "issue:2107", + "right": "issue:4398", + "accept": false, + "reason": "Both mention bad/black images, but the first is low-resolution image generation and the second is an SDXL LoRA dreambooth pipeline issue." + }, + { + "left": "issue:3222", + "right": "issue:6686", + "accept": false, + "reason": "Unrelated problems: missing custom diffusion output file vs Dynamo save/load performance behavior." + }, + { + "left": "issue:11464", + "right": "issue:9717", + "accept": false, + "reason": "Different scopes: a specific FluxControlPipeline LoRA load failure vs a Flux-dev docker image error." + }, + { + "left": "issue:4476", + "right": "issue:8729", + "accept": false, + "reason": "Both involve SDXL controlnet training, but the symptoms differ: poor results on fill50k vs unexpected validation images in a webdataset script." + }, + { + "left": "issue:6520", + "right": "issue:9179", + "accept": false, + "reason": "Different training scripts and outcomes: blank validation images vs nonsensical SDXL+ControlNet training results." + }, + { + "left": "issue:4255", + "right": "issue:7142", + "accept": false, + "reason": "Both compare diffusers to webui, but they are separate model/pipeline quality complaints." + }, + { + "left": "issue:5570", + "right": "issue:9223", + "accept": false, + "reason": "Completely unrelated: missing training utility function vs dependency version conflict." + }, + { + "left": "issue:5510", + "right": "issue:8668", + "accept": false, + "reason": "Different failure modes in different pipelines: dtype mismatch in SDXL InstructPix2Pix vs SD3 image-size alignment error." + }, + { + "left": "issue:11006", + "right": "issue:12037", + "accept": false, + "reason": "Both concern Wan video models, but one is broken quantized I2V output and the other is LoRA compatibility with a different Wan version/model setup." + }, + { + "left": "issue:10025", + "right": "issue:12071", + "accept": false, + "reason": "Both are Flux transformer issues, but one is attention-mask handling and the other is a changed single-block forward behavior; not clearly the same bug." + }, + { + "left": "issue:10954", + "right": "issue:11418", + "accept": false, + "reason": "Different requests: LoRA loading failure vs how to add a specific Flux fill model file." + }, + { + "left": "issue:1185", + "right": "issue:2371", + "accept": false, + "reason": "Both are Dreambooth-related, but reproducibility is not the same as missing training output objects." + }, + { + "left": "issue:12107", + "right": "issue:6351", + "accept": false, + "reason": "Unrelated runtime issues: tracker initialization with custom objects vs DDP missing disable_adapters." + }, + { + "left": "issue:122", + "right": "issue:4101", + "accept": false, + "reason": "Different errors in different training/conversion paths; no concrete duplicate evidence." + }, + { + "left": "issue:1398", + "right": "issue:1537", + "accept": false, + "reason": "Both mention Apple MPS, but the reported bugs are different: SD2 on MPS vs specific scheduler issues on MPS." + }, + { + "left": "issue:1475", + "right": "issue:4954", + "accept": false, + "reason": "Different Dreambooth/textual inversion issues; one is dataset/class setup, the other is an unpickling error." + }, + { + "left": "issue:1572", + "right": "issue:4932", + "accept": false, + "reason": "Different failures in different pipelines: img2img/inpainting support vs missing saved model artifact." + }, + { + "left": "issue:1672", + "right": "issue:2036", + "accept": false, + "reason": "Unrelated model-conversion size bug vs OSError loading a specific Stable Diffusion model." + }, + { + "left": "issue:1949", + "right": "issue:2552", + "accept": false, + "reason": "Both are tutorial-related, but they concern different tutorials and different workflows." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 136, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11241", + "issue:11592", + "issue:11791", + "issue:12319", + "issue:1541", + "issue:1672", + "issue:1931", + "issue:2218", + "issue:2363", + "issue:2549", + "issue:2659", + "issue:2838", + "issue:2858", + "issue:2991", + "issue:3065", + "issue:3112", + "issue:3296", + "issue:3418", + "issue:3480", + "issue:3926", + "issue:4213", + "issue:4304", + "issue:4709", + "issue:4852", + "issue:5178", + "issue:5285", + "issue:5633", + "issue:5892", + "issue:5917", + "issue:5963", + "issue:6222", + "issue:6372", + "issue:6382", + "issue:6540", + "issue:6584", + "issue:7021", + "issue:7133", + "issue:7138", + "issue:7168", + "issue:7237", + "issue:7690", + "issue:8359", + "issue:8493", + "issue:8649", + "issue:9136", + "issue:9424", + "issue:9484", + "issue:9613" + ], + "soft_pairs": [ + "issue:2218|issue:5633", + "issue:2838|issue:2858", + "issue:2991|issue:6540", + "issue:3065|issue:4304", + "issue:3112|issue:6372", + "issue:3418|issue:3480", + "issue:3926|issue:5285", + "issue:5178|issue:5917", + "issue:5892|issue:7021", + "issue:5917|issue:5963", + "issue:7168|issue:9136", + "issue:7237|issue:9424", + "issue:11241|issue:9484", + "issue:2549|issue:4213", + "issue:2659|issue:6222", + "issue:1541|issue:3296", + "issue:2363|issue:6584", + "issue:4852|issue:8493", + "issue:7690|issue:8359", + "issue:7133|issue:7138", + "issue:11791|issue:12319", + "issue:11592|issue:9613", + "issue:6382|issue:8649", + "issue:1672|issue:6382", + "issue:1931|issue:4709" + ], + "prepared_review_unit_hash": "535118d01502948832345b26342853cafce70787712909f4ada7af6e4c51d9b4", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34137, + "estimated_input_tokens": 8535, + "estimated_eval_tokens": 17326 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is loosely related by diffusion-model training/loading topics, but the pairwise overlaps are too broad to treat as duplicates. Most items are distinct bugs or feature requests across different pipelines, models, and failure modes.", + "confidence": 0.44, + "canonical_issue_reason": "Issue 9484 is the best single representative of the cluster\u2019s strongest recurring theme: a concrete FLUX training problem. It is also open and specific enough to anchor the small FLUX/LoRA/controlnet subset better than the older, more divergent items.", + "canonical_pr_reason": null, + "best_issue_reason": "9484 is the most useful cluster representative because it is current, concrete, and centered on FLUX training/deepspeed behavior rather than a generic docs/import/conversion request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2218", + "right": "issue:5633", + "accept": false, + "reason": "Both mention bugs, but one is a training overfitting report and the other is a transformer model bug; they are not the same underlying issue." + }, + { + "left": "issue:2838", + "right": "issue:2858", + "accept": false, + "reason": "One is about ControlNet pipeline behavior and the other about adding a seeding image to an UnCLIP img2img pipeline; different features and failures." + }, + { + "left": "issue:2991", + "right": "issue:6540", + "accept": false, + "reason": "Both are ControlNet training bugs, but mixed precision failure and FSDP matrix-shape failure are different code-path problems." + }, + { + "left": "issue:3065", + "right": "issue:4304", + "accept": false, + "reason": "Both concern LoRA training support, but one is a text-encoder option for DreamBooth and the other is SDXL text-to-image LoRA support; not the same change." + }, + { + "left": "issue:3112", + "right": "issue:6372", + "accept": false, + "reason": "Multi-ControlNet inpainting and SVD+ControlNet are different pipeline combinations; too broad to merge." + }, + { + "left": "issue:3418", + "right": "issue:3480", + "accept": false, + "reason": "One adds text-encoder handling to a training script; the other adds a scale parameter for text-encoder LoRA layers. Related area, but different changes." + }, + { + "left": "issue:3926", + "right": "issue:5285", + "accept": false, + "reason": "These are both docs tasks, but one is an example request and the other is a translation request; not the same issue." + }, + { + "left": "issue:5178", + "right": "issue:5917", + "accept": false, + "reason": "Different bugs in different components: scheduler resolution dependence versus checkpoint conversion breaking after an upgrade." + }, + { + "left": "issue:5892", + "right": "issue:7021", + "accept": false, + "reason": "Both are import errors, but the titles indicate different causes and environments; not enough to \u0441\u0447\u0438\u0442\u0430\u0442\u044c them duplicates." + }, + { + "left": "issue:5917", + "right": "issue:5963", + "accept": false, + "reason": "Checkpoint conversion failure and Kandinsky embed-vs-prompt handling are unrelated failures." + }, + { + "left": "issue:7168", + "right": "issue:9136", + "accept": false, + "reason": "Both mention IP-Adapter, but one is about malformed image embeddings and the other about output quality at certain resolutions; different problems." + }, + { + "left": "issue:7237", + "right": "issue:9424", + "accept": false, + "reason": "A missing scheduler class and a broken notebook are unrelated issues." + }, + { + "left": "issue:11241", + "right": "issue:9484", + "accept": false, + "reason": "Both involve FLUX training, but one is a ControlNet floating-point crash and the other is multigpu/deepspeed DreamBooth training; not the same bug." + }, + { + "left": "issue:2549", + "right": "issue:4213", + "accept": false, + "reason": "Different model-loading failures: SD2 UNet incompatibility versus SDXL from_single_file needing accelerate." + }, + { + "left": "issue:2659", + "right": "issue:6222", + "accept": false, + "reason": "Both mention batch size, but one is image variation/nondeterminism and the other is DDPM scheduler state dependence; too different." + }, + { + "left": "issue:1541", + "right": "issue:3296", + "accept": false, + "reason": "Both are DreamBooth checkpoint/loading issues, but model-type detection and save/load with text_encoder are separate bugs." + }, + { + "left": "issue:2363", + "right": "issue:6584", + "accept": false, + "reason": "Both are conversion questions, but LoRA conversion and DreamShaper inpaint conversion are not the same underlying task." + }, + { + "left": "issue:4852", + "right": "issue:8493", + "accept": false, + "reason": "Both are load failures, but a generic LoRA load error and SD3Transformer2DModel state_dict mismatch are different problems." + }, + { + "left": "issue:7690", + "right": "issue:8359", + "accept": false, + "reason": "Tokenizer loading and custom checkpoint from_single_file failures are unrelated." + }, + { + "left": "issue:7133", + "right": "issue:7138", + "accept": false, + "reason": "Both concern SDXL structure mismatches, but one is text-encoder resume size mismatch and the other is UNet structure mismatch; different components." + }, + { + "left": "issue:11791", + "right": "issue:12319", + "accept": false, + "reason": "Both involve group offloading/LoRA loading, but one is inability to load LoRA after offloading and the other is block_level offloading being broken; not the same failure mode." + }, + { + "left": "issue:11592", + "right": "issue:9613", + "accept": false, + "reason": "A FluxLoraLoaderMixin regression and a Timesteps.scale attribute error are different code-path bugs." + }, + { + "left": "issue:6382", + "right": "issue:8649", + "accept": false, + "reason": "IP-Adapter key errors during training/inference and SD3 num_images_per_prompt handling are unrelated." + }, + { + "left": "issue:1672", + "right": "issue:6382", + "accept": false, + "reason": "Wrong image size after ckpt conversion and IP-Adapter key errors after LCM LoRA training are different issues." + }, + { + "left": "issue:1931", + "right": "issue:4709", + "accept": false, + "reason": "Float16 type mismatch in SD2 and SDXL ControlNet guess-mode support are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 137, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11149", + "issue:11436", + "issue:11978", + "issue:12358", + "issue:1451", + "issue:1478", + "issue:2495", + "issue:2663", + "issue:3209", + "issue:3347", + "issue:3363", + "issue:3401", + "issue:3432", + "issue:3454", + "issue:3540", + "issue:3564", + "issue:3565", + "issue:3761", + "issue:3937", + "issue:3980", + "issue:4198", + "issue:4206", + "issue:4250", + "issue:4460", + "issue:4773", + "issue:5075", + "issue:5224", + "issue:5281", + "issue:5284", + "issue:5319", + "issue:540", + "issue:5897", + "issue:5997", + "issue:6159", + "issue:6238", + "issue:6323", + "issue:720", + "issue:725", + "issue:7357", + "issue:7658", + "issue:7785", + "issue:7931", + "issue:7978", + "issue:8233", + "issue:9495", + "issue:9844", + "issue:9953" + ], + "soft_pairs": [ + "issue:3401|issue:5897", + "issue:4198|issue:5075", + "issue:11978|issue:12358", + "issue:1478|issue:2495", + "issue:2663|issue:4250", + "issue:4206|issue:5997", + "issue:11436|issue:720", + "issue:4773|issue:5319", + "issue:7931|issue:9495", + "issue:8233|issue:9953", + "issue:7357|issue:7785", + "issue:3209|issue:3347", + "issue:3540|issue:3937", + "issue:540|issue:725", + "issue:6238|issue:9844", + "issue:3432|issue:3565", + "issue:5281|issue:7658", + "issue:3564|issue:5284", + "issue:3761|issue:6323", + "issue:11149|issue:11436", + "issue:1451|issue:5224", + "issue:6159|issue:7978", + "issue:3980|issue:4460", + "issue:3363|issue:3454" + ], + "prepared_review_unit_hash": "3117c44dd3e310811656942b6c2acfb0a267261d51760c3a07e62a8cb38b8f46", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33518, + "estimated_input_tokens": 8380, + "estimated_eval_tokens": 17016 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated diffusers issues: dtype/device mismatches, pipeline feature requests, loading failures, performance complaints, and example-script bugs. The soft-similarity pairs only share broad vocabulary (e.g. \u201csame dtype\u201d, \u201cSDXL\u201d, \u201cpipeline\u201d) and do not look like the same underlying bug or change.", + "confidence": 0.95, + "canonical_issue_reason": "No single issue clearly unifies the cluster. If one must be picked, #11149 is the most concrete, self-contained bug report, but it is not a good duplicate umbrella for the others.", + "canonical_pr_reason": null, + "best_issue_reason": "#11149 is the strongest representative as a precise runtime error with a narrow code path and clear symptom; most other items are either different error classes or feature requests.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3401", + "right": "issue:5897", + "accept": false, + "reason": "Both mention tensor/device-related training errors, but one is about missing class labels for class embeddings and the other is a LoRA device placement bug." + }, + { + "left": "issue:4198", + "right": "issue:5075", + "accept": false, + "reason": "Both are dtype/matmul failures, but they occur in different contexts: a notebook example versus SDXL refiner prompt-embed shape mismatch." + }, + { + "left": "issue:11978", + "right": "issue:12358", + "accept": false, + "reason": "These are separate model pipeline requests for different models, not the same bug or change." + }, + { + "left": "issue:1478", + "right": "issue:2495", + "accept": false, + "reason": "Both involve StableDiffusionUpscalePipeline, but one is about multi-image output support and the other is an MPS black-image rendering bug." + }, + { + "left": "issue:2663", + "right": "issue:4250", + "accept": false, + "reason": "One is a safetensors loading failure for a safety checker; the other is a generic from_pretrained loading issue." + }, + { + "left": "issue:4206", + "right": "issue:5997", + "accept": false, + "reason": "Both are SDXL/controlnet-related, but one is a training breakage and the other is a tutorial/model-type mismatch error." + }, + { + "left": "issue:11436", + "right": "issue:720", + "accept": false, + "reason": "Both are dtype mismatch errors, but the underlying paths differ: Hidream inference matmul dtype mismatch versus traced UNet half/float tracing issue." + }, + { + "left": "issue:4773", + "right": "issue:5319", + "accept": false, + "reason": "These are unrelated LoRA-loading concerns: one is CivitAI LoRA loading, the other is a proposal to speed up LoRA loading with PEFT." + }, + { + "left": "issue:7931", + "right": "issue:9495", + "accept": false, + "reason": "Both involve SDXL plus IP-Adapter-style conditioning, but one is a ControlNet inpaint parameter bug and the other is a PAG + IPAdapter failure." + }, + { + "left": "issue:8233", + "right": "issue:9953", + "accept": false, + "reason": "Both mention fp16/quantization-adjacent behavior, but one is a depth pipeline fp16 error and the other is moving a quantized component to CUDA." + }, + { + "left": "issue:7357", + "right": "issue:7785", + "accept": false, + "reason": "Both are performance complaints, but one is about torch.compile/fuse_loras on SDXL and the other is slow first inference in diffusers 0.27.x." + }, + { + "left": "issue:3209", + "right": "issue:3347", + "accept": false, + "reason": "Both mention NaNs in diffusion outputs, but one is specifically AutoencoderKL encoder NaNs for large images and the other is a general pipeline producing NaN image arrays." + }, + { + "left": "issue:3540", + "right": "issue:3937", + "accept": false, + "reason": "These are unrelated pipeline issues: prompt embeddings on TextToVideoZeroPipeline versus ZeroscopeXL not working as a normal model." + }, + { + "left": "issue:540", + "right": "issue:725", + "accept": false, + "reason": "Both concern memory/offload behavior, but one proposes keeping only UNet on GPU while the other asks for device_map/accelerate support in DiffusionPipeline." + }, + { + "left": "issue:6238", + "right": "issue:9844", + "accept": false, + "reason": "Both are VAE encoder NaN reports, but they target different model families and are not clearly the same failure mode." + }, + { + "left": "issue:3432", + "right": "issue:3565", + "accept": false, + "reason": "Both are CUDA/half precision training/inference problems, but one is a CPU half-kernel runtime error and the other is a text-to-image training OOM." + }, + { + "left": "issue:5281", + "right": "issue:7658", + "accept": false, + "reason": "Both are multi-component SDXL issues, but one is a device placement error with CPU offload and the other is a question about multiple conditioning with ControlNet." + }, + { + "left": "issue:3564", + "right": "issue:5284", + "accept": false, + "reason": "One is a request for Uni-ControlNet support and the other is a request for ControlNet-LLLite support; they are different features." + }, + { + "left": "issue:3761", + "right": "issue:6323", + "accept": false, + "reason": "One is about loading two community pipelines in the same process; the other is a TensorRT img2img attribute-setting bug." + }, + { + "left": "issue:11149", + "right": "issue:11436", + "accept": false, + "reason": "Both are dtype-related runtime errors, but one is an unexpected keyword argument in quantizer code and the other is a mat1/mat2 dtype mismatch in inference." + }, + { + "left": "issue:1451", + "right": "issue:5224", + "accept": false, + "reason": "These are unrelated Flax/JAX Stable Diffusion errors with different model/version contexts." + }, + { + "left": "issue:6159", + "right": "issue:7978", + "accept": false, + "reason": "One is an unsupported denoising_end argument in an SDXL ControlNet pipeline; the other is a torchvision operator registration failure." + }, + { + "left": "issue:3980", + "right": "issue:4460", + "accept": false, + "reason": "Both mention different outputs versus other implementations, but one is a long-prompt tokenization issue and the other is a general stable-diffusion versus diffusers result difference." + }, + { + "left": "issue:3363", + "right": "issue:3454", + "accept": false, + "reason": "Both are train_dreambooth_lora.py failures, but they are described as different machine/environment breakages and do not indicate the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 138, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11010", + "issue:12078", + "issue:1404", + "issue:1445", + "issue:1500", + "issue:1672", + "issue:1807", + "issue:1823", + "issue:1866", + "issue:1896", + "issue:2089", + "issue:2420", + "issue:2459", + "issue:2880", + "issue:3026", + "issue:3124", + "issue:3565", + "issue:3647", + "issue:3694", + "issue:3975", + "issue:4566", + "issue:4733", + "issue:4932", + "issue:4993", + "issue:5128", + "issue:5465", + "issue:5578", + "issue:5678", + "issue:5719", + "issue:6183", + "issue:6230", + "issue:6277", + "issue:6283", + "issue:6561", + "issue:6832", + "issue:7333", + "issue:7480", + "issue:7682", + "issue:7722", + "issue:852", + "issue:867", + "issue:8898", + "issue:9208", + "issue:9227", + "issue:9270", + "issue:9793", + "issue:9873" + ], + "soft_pairs": [ + "issue:1445|issue:1500", + "issue:1866|issue:5465", + "issue:4993|issue:6283", + "issue:1672|issue:867", + "issue:4566|issue:4733", + "issue:5678|issue:7722", + "issue:1404|issue:1896", + "issue:1823|issue:2880", + "issue:2420|issue:6277", + "issue:3026|issue:6230", + "issue:3694|issue:7682", + "issue:5578|issue:9793", + "issue:12078|issue:9227", + "issue:11010|issue:9270", + "issue:1445|issue:852", + "issue:6183|issue:9208", + "issue:6561|issue:6832", + "issue:7333|issue:9873", + "issue:1807|issue:3124", + "issue:5719|issue:8898", + "issue:2459|issue:4932", + "issue:3565|issue:7480", + "issue:3975|issue:5128", + "issue:2089|issue:3647" + ], + "prepared_review_unit_hash": "79558fa6a97bdf2f0eef4fb51e11547023f4cec341209f6bcaf6197dde7f3b25", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33837, + "estimated_input_tokens": 8460, + "estimated_eval_tokens": 17176 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated Diffusers issues that only share broad model/training/loading vocabulary; none of the soft pairs look like the same underlying bug or change.", + "confidence": 0.94, + "canonical_issue_reason": "No single canonical issue stands out: the set spans distinct bugs across Flux training, loading, checkpointing, memory, and UI/docs behavior rather than one duplicate report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9227 is the most concrete and actionable report in the set, with a specific Flux training script, model, and failure context; it is the strongest standalone issue even though it does not unify the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1445", + "right": "issue:1500", + "accept": false, + "reason": "Both are pipeline compatibility problems, but one is an inpaint legacy AttributeError and the other is a community pipeline/version incompatibility; different code paths and fixes." + }, + { + "left": "issue:1866", + "right": "issue:5465", + "accept": false, + "reason": "Both mention strength-related behavior, but one is a scheduler/img2img failure at certain strengths and the other is a DiffEdit-only inpaint_strength limitation; not the same bug." + }, + { + "left": "issue:4993", + "right": "issue:6283", + "accept": false, + "reason": "CLIPImageProjection import/export structure vs SDXL ControlNet strength=0 behavior are unrelated failure modes." + }, + { + "left": "issue:1672", + "right": "issue:867", + "accept": false, + "reason": "Wrong output size after conversion and GPU memory not being released are different bugs with different symptoms and fixes." + }, + { + "left": "issue:4566", + "right": "issue:4733", + "accept": false, + "reason": "Both involve checkpoint resume, but one is SDXL LoRA resume support and the other is DreamBooth LoRA optimizer state validation; not the same concrete defect." + }, + { + "left": "issue:5678", + "right": "issue:7722", + "accept": false, + "reason": "Higher-order derivative errors in attention math and an Animatediff+IPAdapter pipeline error are unrelated." + }, + { + "left": "issue:1404", + "right": "issue:1896", + "accept": false, + "reason": "Both are CLIP-guided SD issues, but one is about a pipeline/model support question and the other is a size/type runtime error; too different to merge." + }, + { + "left": "issue:1823", + "right": "issue:2880", + "accept": false, + "reason": "A swapped comment typo and a JAX Colab runtime failure are not the same bug." + }, + { + "left": "issue:2420", + "right": "issue:6277", + "accept": false, + "reason": "xFormers attention enabling in LoRA training and broken gradients are different training-path failures." + }, + { + "left": "issue:3026", + "right": "issue:6230", + "accept": false, + "reason": "Instruct-pix2pix accelerate launch error on one GPU vs SDXL text-to-image OOM are different issues." + }, + { + "left": "issue:3694", + "right": "issue:7682", + "accept": false, + "reason": "Missing model/config files in the training example versus multi-GPU SD1.5 execution issues are not the same underlying bug." + }, + { + "left": "issue:5578", + "right": "issue:9793", + "accept": false, + "reason": "Missing model_index.json loading error and a progress-bar update bug are unrelated." + }, + { + "left": "issue:12078", + "right": "issue:9227", + "accept": false, + "reason": "Both are Flux-related, but one is an example validation-input problem and the other is a DreamBooth LoRA training failure; different scripts and code paths." + }, + { + "left": "issue:11010", + "right": "issue:9270", + "accept": false, + "reason": "Both mention Flux support, but one is about architecture changes for Chroma and the other about incomplete LoRA support in SimpleTuner; not a single concrete bug." + }, + { + "left": "issue:1445", + "right": "issue:852", + "accept": false, + "reason": "Both are compatibility crashes, but they affect different pipelines and different API surfaces (legacy inpaint vs MPS device_map handling)." + }, + { + "left": "issue:6183", + "right": "issue:9208", + "accept": false, + "reason": "Both are from_single_file loading issues, but one is incorrect tensor shape and the other is inability to load ControlNet models; distinct failure modes." + }, + { + "left": "issue:6561", + "right": "issue:6832", + "accept": false, + "reason": "An OSError about missing image_encoder config and an AttributeError in IP-Adapter Face ID are different bugs in the same broad area, but not the same fix." + }, + { + "left": "issue:7333", + "right": "issue:9873", + "accept": false, + "reason": "Image-to-video generation support and OmniGen pipeline addition are different feature requests/issues." + }, + { + "left": "issue:1807", + "right": "issue:3124", + "accept": false, + "reason": "A performance degradation over successive generations and a model conversion failure are unrelated." + }, + { + "left": "issue:5719", + "right": "issue:8898", + "accept": false, + "reason": "Both involve loading-path logic, but one is LoRA weight file precedence and the other is subfolder loading with model files; different conditions and fixes." + }, + { + "left": "issue:2459", + "right": "issue:4932", + "accept": false, + "reason": "xFormers enabling in LoRA training and missing pytorch_model.bin output from DreamBooth are separate training/export issues." + }, + { + "left": "issue:3565", + "right": "issue:7480", + "accept": false, + "reason": "Both are OOM reports, but they concern different examples/models and likely different memory causes." + }, + { + "left": "issue:3975", + "right": "issue:5128", + "accept": false, + "reason": "A typo in an error message and a failure loading a local safetensors file are not the same bug." + }, + { + "left": "issue:2089", + "right": "issue:3647", + "accept": false, + "reason": "Both are training OOMs, but they involve different models and workflows (AltDiffusion DreamBooth vs DeepFloyd DreamBooth)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 139, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10226", + "issue:10823", + "issue:11014", + "issue:11359", + "issue:11374", + "issue:11408", + "issue:1166", + "issue:1188", + "issue:1195", + "issue:12034", + "issue:12216", + "issue:12348", + "issue:1392", + "issue:1538", + "issue:1717", + "issue:2015", + "issue:2806", + "issue:2883", + "issue:3094", + "issue:3235", + "issue:3280", + "issue:3289", + "issue:3301", + "issue:3393", + "issue:3487", + "issue:3812", + "issue:3886", + "issue:4139", + "issue:4386", + "issue:4450", + "issue:4484", + "issue:4584", + "issue:4736", + "issue:4835", + "issue:5395", + "issue:5678", + "issue:6008", + "issue:6092", + "issue:6382", + "issue:6639", + "issue:6870", + "issue:6894", + "issue:6911", + "issue:7333", + "issue:8514", + "issue:9051", + "issue:9343", + "issue:9717" + ], + "soft_pairs": [ + "issue:10823|issue:8514", + "issue:2883|issue:7333", + "issue:3812|issue:6870", + "issue:11408|issue:9051", + "issue:11014|issue:3094", + "issue:2806|issue:6911", + "issue:4835|issue:5395", + "issue:12034|issue:12216", + "issue:1188|issue:2015", + "issue:1195|issue:1538", + "issue:3393|issue:4386", + "issue:6008|issue:6382", + "issue:9343|issue:9717", + "issue:1717|issue:5678", + "issue:10226|issue:12348", + "issue:1166|issue:3886", + "issue:4584|issue:4736", + "issue:11359|issue:11374", + "issue:1188|issue:1392", + "issue:3235|issue:4139", + "issue:3280|issue:3289", + "issue:4450|issue:6092", + "issue:3487|issue:6870", + "issue:3301|issue:6894", + "issue:4484|issue:6639" + ], + "prepared_review_unit_hash": "c0f6bc74fa1434573640566d89044aa4ade88c8b2748fb31198122481bdbfeaa", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34230, + "estimated_input_tokens": 8558, + "estimated_eval_tokens": 17372 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Loose keyword cluster around diffusion/LoRA/Flux/image-video problems, but the proposed pairings are mostly superficial (same terms like tensor mismatch, LoRA, image-to-video, or training) rather than the same underlying bug/change. I would not merge any of the soft edges.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 12348 is the clearest concrete bug report in the set (specific tensor mismatch, reproducible pipeline failure) and is more representative than the broader feature requests or unrelated training/runtime errors.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 12348 is the strongest single issue to anchor triage because it is specific, technically actionable, and not phrased as a broad request or ambiguous complaint.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10823", + "right": "issue:8514", + "accept": false, + "reason": "Both are training-related, but one is about missing inference artifacts after fine-tuning while the other is an EMAModel parallel fine-tuning error; different code paths and symptoms." + }, + { + "left": "issue:2883", + "right": "issue:7333", + "accept": false, + "reason": "One requests an image-to-video pipeline; the other is about a tuning-free noise rectification method for high-fidelity i2v generation. Same domain, different change." + }, + { + "left": "issue:3812", + "right": "issue:6870", + "accept": false, + "reason": "Both mention missing pipeline/config options, but they target different flags in different pipelines (`controlnet` config vs `enable_vae_tiling`)." + }, + { + "left": "issue:11408", + "right": "issue:9051", + "accept": false, + "reason": "`set_adapters` on compiled models and Flux batch inference support are unrelated feature/compatibility issues." + }, + { + "left": "issue:11014", + "right": "issue:3094", + "accept": false, + "reason": "Both are OOM reports, but they occur in different training scripts and environments; not enough evidence of the same bug." + }, + { + "left": "issue:2806", + "right": "issue:6911", + "accept": false, + "reason": "Different problems in different components: printing pipeline customization vs class labels not used in a 3D model." + }, + { + "left": "issue:4835", + "right": "issue:5395", + "accept": false, + "reason": "MultiControlNet img2img bug and PEFT LoRA-combination bug are separate code paths and separate failure modes." + }, + { + "left": "issue:12034", + "right": "issue:12216", + "accept": false, + "reason": "Both compare output quality, but they concern different models and workflows (Wan i2v vs Qwen image editing)." + }, + { + "left": "issue:1188", + "right": "issue:2015", + "accept": false, + "reason": "A CLIPSeg+Stable Diffusion pipeline request is unrelated to a categorical diffusion pipeline request." + }, + { + "left": "issue:1195", + "right": "issue:1538", + "accept": false, + "reason": "Memory-efficient attention with fp16 weights is a distinct issue from SD2-base failing in StableDiffusionMegaPipeline." + }, + { + "left": "issue:3393", + "right": "issue:4386", + "accept": false, + "reason": "Both involve LoRA conversion, but one asks about A1111 safetensors-to-bin conversion generally and the other is specifically about SDXL; too broad to merge." + }, + { + "left": "issue:6008", + "right": "issue:6382", + "accept": false, + "reason": "Both involve IP-Adapter and LCM LoRA, but one is a runtime compatibility complaint and the other is a training-script key error after distillation." + }, + { + "left": "issue:9343", + "right": "issue:9717", + "accept": false, + "reason": "Both are FLUX-related, but one is a loading argument mismatch and the other is a docker image error; not the same bug." + }, + { + "left": "issue:1717", + "right": "issue:5678", + "accept": false, + "reason": "Offline cache loading and higher-order derivative failures are unrelated." + }, + { + "left": "issue:10226", + "right": "issue:12348", + "accept": false, + "reason": "Both mention tensor mismatches, but they are different models/pipelines (Dreambooth Flux LoRA vs Wan image-to-video) with no shared code path shown." + }, + { + "left": "issue:1166", + "right": "issue:3886", + "accept": false, + "reason": "TPU save hangs in textual inversion training is unrelated to distributed watchdog timeouts during DDPM training." + }, + { + "left": "issue:4584", + "right": "issue:4736", + "accept": false, + "reason": "Same script, but one is a checkpoint-resume error and the other is OOM/cast instability; different concrete defects." + }, + { + "left": "issue:11359", + "right": "issue:11374", + "accept": false, + "reason": "Two separate feature requests for different models: faster inference for LTX-Video vs adding SkyReels-V2 support." + }, + { + "left": "issue:1188", + "right": "issue:1392", + "accept": false, + "reason": "A pipeline request and a general Stable Diffusion 2 issue are not the same bug/change." + }, + { + "left": "issue:3235", + "right": "issue:4139", + "accept": false, + "reason": "Both are Custom Diffusion-related, but one is about loading biased weights and the other about precision casting in an attention processor." + }, + { + "left": "issue:3280", + "right": "issue:3289", + "accept": false, + "reason": "Non-default scheduler failure and DeepFloyd super-resolution artifacts are separate pipeline bugs." + }, + { + "left": "issue:4450", + "right": "issue:6092", + "accept": false, + "reason": "Multi-text-inversion not working and LoRA example not training are different training features and failure modes." + }, + { + "left": "issue:3487", + "right": "issue:6870", + "accept": false, + "reason": "Inpaint quality with EulerAncestralDiscreteScheduler and missing `enable_vae_tiling` support are unrelated inpaint issues." + }, + { + "left": "issue:3301", + "right": "issue:6894", + "accept": false, + "reason": "Both are LoRA training complaints, but they involve different scripts, model scales, and symptoms." + }, + { + "left": "issue:4484", + "right": "issue:6639", + "accept": false, + "reason": "Weird SDXL Kohya LoRA results and differing outputs from `load_lora_weights` vs manual loading are related to LoRA behavior, but they are not the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 140, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10925", + "issue:11045", + "issue:11436", + "issue:11581", + "issue:11627", + "issue:12459", + "issue:1475", + "issue:1543", + "issue:1695", + "issue:1931", + "issue:2459", + "issue:2606", + "issue:2645", + "issue:3086", + "issue:3674", + "issue:3765", + "issue:3826", + "issue:4304", + "issue:4419", + "issue:4562", + "issue:4790", + "issue:5002", + "issue:5254", + "issue:5465", + "issue:556", + "issue:5594", + "issue:5606", + "issue:6277", + "issue:6689", + "issue:6832", + "issue:6974", + "issue:7221", + "issue:7380", + "issue:8414", + "issue:8514", + "issue:8817", + "issue:8832", + "issue:8955", + "issue:9051", + "issue:9089", + "issue:9109", + "issue:9370", + "issue:9667", + "issue:9673", + "issue:9744", + "issue:9873", + "issue:9953" + ], + "soft_pairs": [ + "issue:8955|issue:9667", + "issue:6832|issue:8817", + "issue:3826|issue:8414", + "issue:9744|issue:9873", + "issue:11045|issue:8832", + "issue:9051|issue:9089", + "issue:11581|issue:11627", + "issue:5465|issue:9109", + "issue:11436|issue:1931", + "issue:2645|issue:5254", + "issue:4419|issue:7221", + "issue:5606|issue:9673", + "issue:7380|issue:8414", + "issue:10925|issue:12459", + "issue:1475|issue:3765", + "issue:3674|issue:6277", + "issue:8514|issue:9953", + "issue:2459|issue:6974", + "issue:1543|issue:4790", + "issue:1695|issue:2606", + "issue:3086|issue:556", + "issue:4304|issue:5594", + "issue:5002|issue:9370", + "issue:4562|issue:6689" + ], + "prepared_review_unit_hash": "c5ebf153b96cbee13b2bb1b36f559791d3f620412b93524b3bc244a5014782c2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33530, + "estimated_input_tokens": 8383, + "estimated_eval_tokens": 17022 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly a grab bag of unrelated issues and feature requests; none of the soft-similarity pairs look like true duplicates or mergeable PRs.", + "confidence": 0.88, + "canonical_issue_reason": "No single canonical issue stands out because the items span distinct bugs, models, and feature requests rather than one underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 8955 is the strongest standalone report: it has the most detailed reproduction and a concrete CLIPTextModelWithProjection state_dict loading failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8955", + "right": "issue:9667", + "accept": false, + "reason": "Both involve FLUX loading, but one is a CLIPTextModelWithProjection state_dict error and the other is an fp8/local transformer file loading problem; different failure paths." + }, + { + "left": "issue:6832", + "right": "issue:8817", + "accept": false, + "reason": "Both mention IP-Adapter FaceID, but the concrete bugs differ: tuple.shape AttributeError vs a generic \"can't use\" report." + }, + { + "left": "issue:3826", + "right": "issue:8414", + "accept": false, + "reason": "Both are new model/pipeline requests, but they target different audio-generation models and different implementations." + }, + { + "left": "issue:9744", + "right": "issue:9873", + "accept": false, + "reason": "Different feature requests: Mochi video model vs OmniGen image generation model." + }, + { + "left": "issue:11045", + "right": "issue:8832", + "accept": false, + "reason": "One is a FLUX Dreambooth crash, the other is a generic pipe(prompt).images[0] runtime crash; no shared code path." + }, + { + "left": "issue:9051", + "right": "issue:9089", + "accept": false, + "reason": "Batch inference support for FluxPipeline is unrelated to FluxPipeline import/not-found errors." + }, + { + "left": "issue:11581", + "right": "issue:11627", + "accept": false, + "reason": "Both involve adapters, but one is a dtype kwarg TypeError in CLIPVisionModelWithProjection and the other is an SD3 pipeline integration issue." + }, + { + "left": "issue:5465", + "right": "issue:9109", + "accept": false, + "reason": "DiffEdit inpaint strength behavior is unrelated to the missing torchvision::nms operator error." + }, + { + "left": "issue:11436", + "right": "issue:1931", + "accept": false, + "reason": "Both are dtype mismatch errors, but they hit different models and code paths (Hidream inference vs SD2 training)." + }, + { + "left": "issue:2645", + "right": "issue:5254", + "accept": false, + "reason": "A mac M1 CUDA assertion is a bug; multi-ControlNet support is a feature request." + }, + { + "left": "issue:4419", + "right": "issue:7221", + "accept": false, + "reason": "Offset-noise training/inference mismatch is not the same as a noisy output from a specific checkpoint." + }, + { + "left": "issue:5606", + "right": "issue:9673", + "accept": false, + "reason": "LoRA alpha handling and high CPU usage when loading multiple LoRAs are different bugs." + }, + { + "left": "issue:7380", + "right": "issue:8414", + "accept": false, + "reason": "Both are diffusion-distillation research requests, but they describe different models and are not the same change." + }, + { + "left": "issue:10925", + "right": "issue:12459", + "accept": false, + "reason": "Dreambooth finetuning for FLUX dev CLIPTextModel is distinct from batch inference throughput for FLUX.1-Kontext-dev." + }, + { + "left": "issue:1475", + "right": "issue:3765", + "accept": false, + "reason": "Training with two image classes is unrelated to ControlNet stacking effects during training." + }, + { + "left": "issue:3674", + "right": "issue:6277", + "accept": false, + "reason": "Metadata.jsonl parsing during fine-tuning is not the same as broken gradients in the example LoRA script." + }, + { + "left": "issue:8514", + "right": "issue:9953", + "accept": false, + "reason": "EMAModel parallel fine-tuning errors and moving a quantized component to CUDA are different issues." + }, + { + "left": "issue:2459", + "right": "issue:6974", + "accept": false, + "reason": "xformers attention not working in one training script is unrelated to unload_textual_inversion not clearing tokenizer_2/text_encoder_2." + }, + { + "left": "issue:1543", + "right": "issue:4790", + "accept": false, + "reason": "Cannot set revision=fp16 is a separate conversion/configuration bug from round-tripping converted weights between scripts." + }, + { + "left": "issue:1695", + "right": "issue:2606", + "accept": false, + "reason": "Both involve DeepSpeed training, but one is multi-GPU setup and the other is a save_state hang; not the same concrete bug." + }, + { + "left": "issue:3086", + "right": "issue:556", + "accept": false, + "reason": "Input-audio/output-image is unrelated to seamless texture tiling." + }, + { + "left": "issue:4304", + "right": "issue:5594", + "accept": false, + "reason": "Support for SDXL Text-to-Image LoRA is a feature request; saving memory during SDXL fine-tuning is an optimization issue." + }, + { + "left": "issue:5002", + "right": "issue:9370", + "accept": false, + "reason": "Missing use_karras_sigmas in schedulers is unrelated to the missing pooled_projection argument in a different embedding module." + }, + { + "left": "issue:4562", + "right": "issue:6689", + "accept": false, + "reason": "Both are conversion failures, but they involve different checkpoints and likely different conversion paths; not enough to treat as one bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 141, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10514", + "issue:10866", + "issue:11370", + "issue:12147", + "issue:1251", + "issue:1310", + "issue:1392", + "issue:2107", + "issue:222", + "issue:2392", + "issue:2674", + "issue:2728", + "issue:3296", + "issue:3420", + "issue:3432", + "issue:3461", + "issue:3803", + "issue:4236", + "issue:4686", + "issue:4892", + "issue:4932", + "issue:5509", + "issue:5516", + "issue:5607", + "issue:5857", + "issue:6201", + "issue:6382", + "issue:6435", + "issue:6686", + "issue:6726", + "issue:6839", + "issue:6967", + "issue:6974", + "issue:7347", + "issue:7960", + "issue:7982", + "issue:8472", + "issue:8560", + "issue:8788", + "issue:8950", + "issue:8955", + "issue:9208", + "issue:9527", + "issue:9876", + "issue:9914", + "issue:9933", + "issue:9979" + ], + "soft_pairs": [ + "issue:3461|issue:6967", + "issue:6382|issue:9876", + "issue:3420|issue:6974", + "issue:1251|issue:2107", + "issue:6435|issue:9933", + "issue:2392|issue:6201", + "issue:2674|issue:5857", + "issue:4236|issue:6686", + "issue:10866|issue:12147", + "issue:1310|issue:1392", + "issue:4892|issue:6726", + "issue:8560|issue:8950", + "issue:222|issue:3432", + "issue:10514|issue:8788", + "issue:11370|issue:7982", + "issue:2728|issue:4686", + "issue:7347|issue:7960", + "issue:5509|issue:5607", + "issue:3803|issue:8472", + "issue:9527|issue:9979", + "issue:3296|issue:5516", + "issue:4932|issue:6974", + "issue:8955|issue:9914", + "issue:6839|issue:9208" + ], + "prepared_review_unit_hash": "d5c531810b517e117b1a5993fdb6dea4f9b9699ed6edfa9a26f0ee6b728672e6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33665, + "estimated_input_tokens": 8417, + "estimated_eval_tokens": 17090 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly heterogeneous issues with no PRs; the only strong duplicate-like pair is the prompt_embeds/prompt handling bug in Stable Diffusion pipelines.", + "confidence": 0.78, + "canonical_issue_reason": "issue:6201 is the better canonical issue for the prompt_embeds bug: it describes the core control-flow problem more generally, while issue:2392 is a narrower img2img manifestation of the same failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6201 best represents the one real duplicate pair in this set because it captures the underlying pipeline logic error rather than a single downstream symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3461", + "right": "issue:6967", + "accept": false, + "reason": "Both are training problems, but they concern different scripts and failure modes (performance/vRAM vs prior-preservation shape mismatch)." + }, + { + "left": "issue:6382", + "right": "issue:9876", + "accept": false, + "reason": "One is an IP-adapter key error after LCM LoRA training; the other is a VRAM-release concern after LoRA training. Different bugs." + }, + { + "left": "issue:3420", + "right": "issue:6974", + "accept": false, + "reason": "ControlNet/diffusion pipeline integration vs unloading textual inversion components; unrelated code paths." + }, + { + "left": "issue:1251", + "right": "issue:2107", + "accept": false, + "reason": "Both mention black images, but the triggers are different: RePaint on MPS vs low-resolution image handling." + }, + { + "left": "issue:6435", + "right": "issue:9933", + "accept": false, + "reason": "Both mention width/height, but one is default initialization in ControlNet and the other is a missing API parameter in SD3 img2img." + }, + { + "left": "issue:2392", + "right": "issue:6201", + "accept": true, + "reason": "Same underlying prompt handling bug: the pipeline wrongly requires/uses prompt when prompt_embeds (and negative_prompt_embeds) are provided." + }, + { + "left": "issue:2674", + "right": "issue:5857", + "accept": false, + "reason": "LoRA/safetensors packaging issue vs LoRA behavior on AnimateDiff SDXL; different problems." + }, + { + "left": "issue:4236", + "right": "issue:6686", + "accept": false, + "reason": "SDXL refiner text-encoder registration vs Dynamo compilation save/load performance; not the same bug." + }, + { + "left": "issue:10866", + "right": "issue:12147", + "accept": false, + "reason": "Both are LoRA loading issues, but for different models and failure contexts; too broad to treat as one bug." + }, + { + "left": "issue:1310", + "right": "issue:1392", + "accept": false, + "reason": "A K-Diffusion pipeline question is not the same as a Stable Diffusion 2 feature request." + }, + { + "left": "issue:4892", + "right": "issue:6726", + "accept": false, + "reason": "Different pipelines and tasks: TokenFlow video editing vs content-consistent super-resolution." + }, + { + "left": "issue:8560", + "right": "issue:8950", + "accept": false, + "reason": "Both involve SD3 training, but one is OOM and the other is a LoRA training bug; not the same issue." + }, + { + "left": "issue:222", + "right": "issue:3432", + "accept": false, + "reason": "Generic training RuntimeError vs a specific CPU Half-precision convolution limitation." + }, + { + "left": "issue:10514", + "right": "issue:8788", + "accept": false, + "reason": "Different model/tooling constraints: Sana resolution binning sample_size limit vs SD3 VAE tiling on non-power-of-two images." + }, + { + "left": "issue:11370", + "right": "issue:7982", + "accept": false, + "reason": "A converter file-load error is unrelated to a from_pipe MultiControlNetModel issue." + }, + { + "left": "issue:2728", + "right": "issue:4686", + "accept": false, + "reason": "Safetensors loading for UNet vs loading a saved upscale pipeline; same family of API but different concrete bugs." + }, + { + "left": "issue:7347", + "right": "issue:7960", + "accept": false, + "reason": "Both are training failures, but one is a post-update LoRA regression and the other is SDXL multi-GPU training failure." + }, + { + "left": "issue:5509", + "right": "issue:5607", + "accept": false, + "reason": "Both are shape mismatches, but the actual mismatched tensors and inputs are different; not enough to merge as one bug." + }, + { + "left": "issue:3803", + "right": "issue:8472", + "accept": false, + "reason": "Multi-GPU LoRA training bug vs a general question about simultaneous modules during pre-training." + }, + { + "left": "issue:9527", + "right": "issue:9979", + "accept": false, + "reason": "A dtype mismatch with ControlNet fp32/bf16 is not the same as a Flux img2img ControlNet channel error." + }, + { + "left": "issue:3296", + "right": "issue:5516", + "accept": false, + "reason": "Dreambooth checkpoint save/load with text_encoder vs LoRA behavior with t2i-adapter pipelines; different subsystems." + }, + { + "left": "issue:4932", + "right": "issue:6974", + "accept": false, + "reason": "Dreambooth artifact export issue vs unloading textual inversion components; unrelated." + }, + { + "left": "issue:8955", + "right": "issue:9914", + "accept": false, + "reason": "Both are load-time errors, but they involve different models and failure points (CLIPTextModelWithProjection state_dict vs Flux LoRA loading)." + }, + { + "left": "issue:6839", + "right": "issue:9208", + "accept": false, + "reason": "Both mention from_single_file, but one is StableVideoDiffusionPipeline and the other is ControlNet model loading; too broad to combine." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The summary stays conservative, and the only accepted soft pair (issue:2392 / issue:6201) is well supported by the titles as the same prompt/prompt_embeds handling bug, with 6201 as the broader canonical issue. The rejected pairs are also appropriately cautious and do not overclaim duplication from the packet evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 142, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10107", + "issue:11136", + "issue:11208", + "issue:11489", + "issue:12612", + "issue:13258", + "issue:1350", + "issue:1367", + "issue:1574", + "issue:1985", + "issue:2392", + "issue:2465", + "issue:2783", + "issue:2784", + "issue:2904", + "issue:3264", + "issue:3487", + "issue:3537", + "issue:3619", + "issue:4061", + "issue:4303", + "issue:4307", + "issue:4354", + "issue:4386", + "issue:4453", + "issue:4580", + "issue:4936", + "issue:5594", + "issue:5647", + "issue:5980", + "issue:6131", + "issue:6176", + "issue:6392", + "issue:6615", + "issue:6734", + "issue:6809", + "issue:6911", + "issue:7197", + "issue:7348", + "issue:7357", + "issue:7420", + "issue:8450", + "issue:8469", + "issue:8543", + "issue:9045", + "issue:9210", + "issue:9393" + ], + "soft_pairs": [ + "issue:1985|issue:2783", + "issue:6131|issue:8469", + "issue:10107|issue:8543", + "issue:3619|issue:6911", + "issue:4307|issue:4580", + "issue:7348|issue:8450", + "issue:6809|issue:9045", + "issue:11136|issue:11208", + "issue:2392|issue:6734", + "issue:4061|issue:4303", + "issue:3487|issue:5647", + "issue:12612|issue:13258", + "issue:1574|issue:2465", + "issue:1350|issue:2784", + "issue:4354|issue:6615", + "issue:4386|issue:5594", + "issue:4936|issue:6392", + "issue:7197|issue:9210", + "issue:11489|issue:9393", + "issue:1367|issue:2904", + "issue:3264|issue:3537", + "issue:4453|issue:5980", + "issue:6176|issue:6392", + "issue:7357|issue:7420" + ], + "prepared_review_unit_hash": "7fd6fa7860a4bf93fdf002aecaf98e0c923945a2c7ccf934a9a13f9d874cd5d1", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33469, + "estimated_input_tokens": 8368, + "estimated_eval_tokens": 16992 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is heterogeneous and the soft pairs are only superficially similar; none look like true duplicates or mergeable equivalents.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 8450 is the most concrete and detailed bug report in the only somewhat related inpainting-strength area, but the overall cluster is too mixed to treat it as a true canonical.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1985", + "right": "issue:2783", + "accept": false, + "reason": "Different feature requests: embedding loading vs adding an img2img/inpainting controlnet pipeline." + }, + { + "left": "issue:6131", + "right": "issue:8469", + "accept": false, + "reason": "Both involve IP-Adapter, but they describe different models and failure modes, not the same bug." + }, + { + "left": "issue:10107", + "right": "issue:8543", + "accept": false, + "reason": "Both mention gradient checkpointing, but one is about validation behavior in training examples and the other is an SD3 dreambooth example failure." + }, + { + "left": "issue:3619", + "right": "issue:6911", + "accept": false, + "reason": "Video-related, but cross-frame attention failure and class-label handling in a 3D UNet are different code paths." + }, + { + "left": "issue:4307", + "right": "issue:4580", + "accept": false, + "reason": "Unrelated issues: local_files_only loading error vs UniPC Karras schedule exception." + }, + { + "left": "issue:7348", + "right": "issue:8450", + "accept": false, + "reason": "Both are inpainting-quality bugs, but one is scheduler-specific ControlNet inpaint output and the other is SDXL inpaint noise at strength=1.0; too different to merge confidently." + }, + { + "left": "issue:6809", + "right": "issue:9045", + "accept": false, + "reason": "Both mention PEFT/LoRA-style loading, but one is adapter loading and the other is CUDA conversion under accelerate+peft." + }, + { + "left": "issue:11136", + "right": "issue:11208", + "accept": false, + "reason": "Both are unsupported-feature errors, but for different pipelines and different unsupported conditions." + }, + { + "left": "issue:2392", + "right": "issue:6734", + "accept": false, + "reason": "Different null-handling failures in different pipeline entry points." + }, + { + "left": "issue:4061", + "right": "issue:4303", + "accept": false, + "reason": "SDXL LoRA loading vs Flax SDXL loading crash are not the same bug or code path." + }, + { + "left": "issue:3487", + "right": "issue:5647", + "accept": false, + "reason": "Both concern inpaint strength/quality, but one is tied to EulerAncestral scheduler behavior and the other to strength==1 in SDXL inpaint." + }, + { + "left": "issue:12612", + "right": "issue:13258", + "accept": false, + "reason": "Different Wan model problems: latent NaNs vs a missing image-to-video model." + }, + { + "left": "issue:1574", + "right": "issue:2465", + "accept": false, + "reason": "Dreambooth training failure vs a request for a reproducible recipe; not duplicates." + }, + { + "left": "issue:1350", + "right": "issue:2784", + "accept": false, + "reason": "Completely different features: object detection vs text-to-video dreambooth support." + }, + { + "left": "issue:4354", + "right": "issue:6615", + "accept": false, + "reason": "Both involve loading from single file/original ckpt, but the failures are different: pipeline type mismatch vs local YAML config handling." + }, + { + "left": "issue:4386", + "right": "issue:5594", + "accept": false, + "reason": "LoRA conversion question vs memory-saving fine-tuning question; no shared bug." + }, + { + "left": "issue:4936", + "right": "issue:6392", + "accept": false, + "reason": "Resume-training regression and custom LoRA loading failure are separate issues." + }, + { + "left": "issue:7197", + "right": "issue:9210", + "accept": false, + "reason": "Different training errors: shortest-edge resume failure vs invalid image_column value." + }, + { + "left": "issue:11489", + "right": "issue:9393", + "accept": false, + "reason": "Both are Flux LoRA training issues, but one is a training-time error and the other is a model-save problem under deepspeed." + }, + { + "left": "issue:1367", + "right": "issue:2904", + "accept": false, + "reason": "Scheduler config serialization and AudioLDM scheduler NoneType error are unrelated." + }, + { + "left": "issue:3264", + "right": "issue:3537", + "accept": false, + "reason": "Different conversion directions and different model types." + }, + { + "left": "issue:4453", + "right": "issue:5980", + "accept": false, + "reason": "Both mention SDXL/LoRA/controlnet artifacts and formats, but one asks about conversion to safetensors and the other about output file type for ControlNet models." + }, + { + "left": "issue:6176", + "right": "issue:6392", + "accept": false, + "reason": "Both are LoRA loading problems, but the reports point to different scenarios and no shared concrete failure path." + }, + { + "left": "issue:7357", + "right": "issue:7420", + "accept": false, + "reason": "Performance regression with torch.compile/fuse_loras is different from a basic inability to use torch.compile." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 143, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1005", + "issue:10147", + "issue:10223", + "issue:10782", + "issue:1082", + "issue:10954", + "issue:10969", + "issue:11006", + "issue:11830", + "issue:12052", + "issue:1249", + "issue:13053", + "issue:13067", + "issue:1802", + "issue:1931", + "issue:2549", + "issue:3390", + "issue:3454", + "issue:3975", + "issue:4073", + "issue:4236", + "issue:4354", + "issue:4402", + "issue:4724", + "issue:4885", + "issue:5023", + "issue:5633", + "issue:5719", + "issue:5875", + "issue:5897", + "issue:5960", + "issue:6252", + "issue:6278", + "issue:6382", + "issue:7420", + "issue:746", + "issue:761", + "issue:794", + "issue:7948", + "issue:7991", + "issue:8320", + "issue:8532", + "issue:8589", + "issue:8970", + "issue:9045", + "issue:9156", + "issue:9341", + "issue:9927" + ], + "soft_pairs": [ + "issue:13053|issue:13067", + "issue:4885|issue:8532", + "issue:1249|issue:4073", + "issue:3454|issue:6278", + "issue:10954|issue:9341", + "issue:4724|issue:6252", + "issue:5897|issue:9045", + "issue:7420|issue:8589", + "issue:5960|issue:9927", + "issue:10147|issue:7948", + "issue:11830|issue:8970", + "issue:3975|issue:794", + "issue:4354|issue:4402", + "issue:5023|issue:5875", + "issue:5633|issue:7991", + "issue:10223|issue:6382", + "issue:1005|issue:746", + "issue:1802|issue:1931", + "issue:11006|issue:12052", + "issue:10782|issue:8320", + "issue:1082|issue:761", + "issue:10969|issue:9156", + "issue:2549|issue:4236", + "issue:3390|issue:5719" + ], + "prepared_review_unit_hash": "af40b1de0d903f2b63587eaf952481307df0ac0e0c00d2ccbe63ba3a2b6cdef4", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34103, + "estimated_input_tokens": 8526, + "estimated_eval_tokens": 17308 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a grab-bag of unrelated issues; nearly all soft pairs are only superficially similar and should stay separate. The only likely duplicate pair is the SDXL inpaint `from_single_file()` reports.", + "confidence": 0.78, + "canonical_issue_reason": "issue:4354 is the most concrete, specific bug report in the only likely duplicate pair, and it names the exact failing code path (`StableDiffusionXLInpaintPipeline.from_single_file`).", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4354 is the best canonical issue candidate because it is specific, actionable, and closest to a clear duplicate target; the rest of the items are unrelated feature requests or different failure modes.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13053", + "right": "issue:13067", + "accept": false, + "reason": "Different feature requests for different models/pipelines: Unipic 3.0 vs Anima." + }, + { + "left": "issue:4885", + "right": "issue:8532", + "accept": false, + "reason": "Both mention LoRA loading, but one is SDXL ControlNet img2img and the other is progressive memory growth during LoRA loading; different bugs." + }, + { + "left": "issue:1249", + "right": "issue:4073", + "accept": false, + "reason": "Different code paths and failures: ONNX pipeline device attribute error vs DeepFloyd IF safety checker `astype` error." + }, + { + "left": "issue:3454", + "right": "issue:6278", + "accept": false, + "reason": "Training-script problems in different scenarios: latest-version breakage vs multi-GPU setup failure." + }, + { + "left": "issue:10954", + "right": "issue:9341", + "accept": false, + "reason": "Both are FLUX LoRA load complaints, but the reported symptoms are not specific enough to say they are the same underlying bug." + }, + { + "left": "issue:4724", + "right": "issue:6252", + "accept": false, + "reason": "Different issues around SDXL img2img behavior vs checkpoint LoRA loading." + }, + { + "left": "issue:5897", + "right": "issue:9045", + "accept": false, + "reason": "Both involve device mismatch themes, but one is LoRA training and the other is SD3 pipeline `.to(cuda)` with accelerate/peft; not the same bug." + }, + { + "left": "issue:7420", + "right": "issue:8589", + "accept": false, + "reason": "`torch.compile` failure and prior-preservation support in SD3 DreamBooth are unrelated." + }, + { + "left": "issue:5960", + "right": "issue:9927", + "accept": false, + "reason": "Different failure modes: invalid cast warning in image conversion vs `HeaderTooLarge` during SDv3 controlnet training." + }, + { + "left": "issue:10147", + "right": "issue:7948", + "accept": false, + "reason": "Separate feature requests for different adapters/models: MV-Adapter multi-view generation vs MuLan language adaptation." + }, + { + "left": "issue:11830", + "right": "issue:8970", + "accept": false, + "reason": "Both produce black output, but on different pipelines and hardware paths (Framepack float16 vs Kolors MPS)." + }, + { + "left": "issue:3975", + "right": "issue:794", + "accept": false, + "reason": "A typo in an error message is not the same as a general unsupported-classes error-message improvement." + }, + { + "left": "issue:4354", + "right": "issue:4402", + "accept": true, + "reason": "Both concern `StableDiffusionXLInpaintPipeline.from_single_file()` misbehaving on the same pipeline; the titles indicate the same underlying construction bug." + }, + { + "left": "issue:5023", + "right": "issue:5875", + "accept": false, + "reason": "Import error for WuerstchenCombinedPipeline vs ONNX conversion prompt error are different problems." + }, + { + "left": "issue:5633", + "right": "issue:7991", + "accept": false, + "reason": "Different areas: a bug in `transformer_temporal.py` versus a training tutorial bug." + }, + { + "left": "issue:10223", + "right": "issue:6382", + "accept": false, + "reason": "Unrelated topics: where to get a LoRA ID for inference vs an IP-adapter key error after SDXL LCM LoRA training." + }, + { + "left": "issue:1005", + "right": "issue:746", + "accept": false, + "reason": "Different scheduler/model loading problems: timestep count constraint vs ONNX revision loading." + }, + { + "left": "issue:1802", + "right": "issue:1931", + "accept": false, + "reason": "Related to UNet usage/dtype, but one is `encoder_hidden_states=None` support and the other is a float16/float mismatch; not the same bug." + }, + { + "left": "issue:11006", + "right": "issue:12052", + "accept": false, + "reason": "Both involve video/wan pipelines and quantized/offloading behavior, but the failures differ: bad video output vs cross-device tensor multiplication." + }, + { + "left": "issue:10782", + "right": "issue:8320", + "accept": false, + "reason": "Different issues in Lumina Image 2.0 vs `from_single_file` returning an `AutoencoderKL` object without `__name__`." + }, + { + "left": "issue:1082", + "right": "issue:761", + "accept": false, + "reason": "Both are DreamBooth training failures, but one is a CUDA CUBLAS execution failure and the other is an 8-bit Adam optimizer failure." + }, + { + "left": "issue:10969", + "right": "issue:9156", + "accept": false, + "reason": "Different FLUX training issues: zero3 controlnet `weight` shape error vs CUDA OOM during LoRA training." + }, + { + "left": "issue:2549", + "right": "issue:4236", + "accept": false, + "reason": "Different SD2/SDXL model bugs: UNet2DConditionModel not working vs SDXL refiner text encoder registration." + }, + { + "left": "issue:3390", + "right": "issue:5719", + "accept": false, + "reason": "One is about loading local safetensors files, the other about LoRA weight-file detection order; same general area but not the same bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The assessment is well grounded in the packet and stays conservative on the soft edges. The only accepted pair, issue:4354 vs issue:4402, is plausibly the same StableDiffusionXLInpaintPipeline.from_single_file bug, and the rest of the verdicts do not overclaim duplicate status from the available titles/bodies. The summary and canonical choice are supported by the evidence provided." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 144, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10492", + "issue:10715", + "issue:10743", + "issue:11033", + "issue:11209", + "issue:11224", + "issue:11436", + "issue:11489", + "issue:11791", + "issue:12673", + "issue:1434", + "issue:1695", + "issue:2036", + "issue:2156", + "issue:2507", + "issue:3296", + "issue:4398", + "issue:4680", + "issue:4808", + "issue:4912", + "issue:5274", + "issue:5529", + "issue:5575", + "issue:5630", + "issue:5821", + "issue:6107", + "issue:6666", + "issue:6906", + "issue:692", + "issue:7082", + "issue:7146", + "issue:7147", + "issue:7263", + "issue:7306", + "issue:752", + "issue:7619", + "issue:8543", + "issue:8557", + "issue:8708", + "issue:8955", + "issue:8970", + "issue:9142", + "issue:9156", + "issue:9202", + "issue:9241", + "issue:9297", + "issue:9602" + ], + "soft_pairs": [ + "issue:4398|issue:6666", + "issue:4680|issue:692", + "issue:4912|issue:9241", + "issue:5821|issue:7306", + "issue:692|issue:752", + "issue:7263|issue:9202", + "issue:8557|issue:8708", + "issue:5529|issue:7146", + "issue:11033|issue:3296", + "issue:4808|issue:5630", + "issue:10492|issue:11209", + "issue:10492|issue:11224", + "issue:1434|issue:2507", + "issue:11489|issue:5274", + "issue:10715|issue:8970", + "issue:10743|issue:9156", + "issue:1695|issue:2156", + "issue:11436|issue:8955", + "issue:5575|issue:9602", + "issue:11791|issue:12673", + "issue:6906|issue:8543", + "issue:7082|issue:7147", + "issue:9142|issue:9297", + "issue:3296|issue:7619", + "issue:2036|issue:6107" + ], + "prepared_review_unit_hash": "30bd03d2778baed1dd75d3515b8adc3598a59d40aea71efee56966cb6ba6a89e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33382, + "estimated_input_tokens": 8346, + "estimated_eval_tokens": 16948 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly thematically related but not a true duplicate set. The only strong duplicate-looking pair is the Kolors black-image reports across ROCm and MPS; most other pairs share a subsystem or symptom but appear to be different bugs/requests.", + "confidence": 0.63, + "canonical_issue_reason": "issue:11209 is the best canonical issue because it matches issue:11224 exactly by title ('Error in loading Lora') and is the earlier report with inbound reference activity.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11209 is the strongest representative of the cluster\u2019s clearest duplicate topic (LoRA loading errors); it is open, earlier, and already referenced by another issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4398", + "right": "issue:6666", + "accept": false, + "reason": "Both mention black images, but one is SDXL LoRA-dreambooth generation and the other is validation images; different code paths and likely different bugs." + }, + { + "left": "issue:4680", + "right": "issue:692", + "accept": false, + "reason": "One is an SDXL inpainting fine-tuning issue and the other is a general request for a fine-tuning example; not the same underlying problem." + }, + { + "left": "issue:4912", + "right": "issue:9241", + "accept": false, + "reason": "Both are model-conversion related, but they target different source formats (CivitAI checkpoints vs XLab LoRA) and different scripts." + }, + { + "left": "issue:5821", + "right": "issue:7306", + "accept": false, + "reason": "Both are training/runtime errors, but they describe different failures (CUDA backward vs float16 in a specific pipeline)." + }, + { + "left": "issue:692", + "right": "issue:752", + "accept": false, + "reason": "A generic fine-tuning example request is not the same as a feature request for multiple concepts/classes in one Dreambooth model." + }, + { + "left": "issue:7263", + "right": "issue:9202", + "accept": false, + "reason": "Both involve IP-Adapter interactions, but one is attention slicing and the other is group offloading with a specific pipeline; different bug conditions." + }, + { + "left": "issue:8557", + "right": "issue:8708", + "accept": false, + "reason": "Different SD3 training issues: token limit versus missing VAE shift_factor configuration." + }, + { + "left": "issue:5529", + "right": "issue:7146", + "accept": false, + "reason": "Same script family, but one is a generic bug report and the other is a CUDA OOM during fine-tuning; not the same concrete defect." + }, + { + "left": "issue:11033", + "right": "issue:3296", + "accept": false, + "reason": "Both concern loading/saving in Dreambooth-related code, but one is from_single_file UNet loading and the other is checkpoint save/load with text_encoder." + }, + { + "left": "issue:4808", + "right": "issue:5630", + "accept": false, + "reason": "Both are img2img pipeline problems, but the failures differ (integration test on SHAP-E vs leftover noise with LMS scheduler on SDXL)." + }, + { + "left": "issue:10492", + "right": "issue:11209", + "accept": false, + "reason": "LoRA is mentioned, but this is torchao quantization/fuse behavior versus a LoRA loading error; different issues." + }, + { + "left": "issue:10492", + "right": "issue:11224", + "accept": false, + "reason": "Same as above: quantize/fuse issue is not the same as LoRA loading failure." + }, + { + "left": "issue:1434", + "right": "issue:2507", + "accept": false, + "reason": "One is an OOM during upscaling, the other is slow safetensors loading on GPU; unrelated performance problems." + }, + { + "left": "issue:11489", + "right": "issue:5274", + "accept": false, + "reason": "Both are training errors with mixed precision/dtype themes, but they involve different models, scripts, and failure modes." + }, + { + "left": "issue:10715", + "right": "issue:8970", + "accept": true, + "reason": "Both report Kolors pipelines producing black images on alternate backends (ROCm/MPS); this looks like the same underlying black-image bug affecting non-CUDA devices." + }, + { + "left": "issue:10743", + "right": "issue:9156", + "accept": false, + "reason": "These are different Flux training concerns: adding ZeRO-3 support versus a CUDA OOM during LoRA training." + }, + { + "left": "issue:1695", + "right": "issue:2156", + "accept": false, + "reason": "Multi-GPU Deepspeed training and CPU-vs-GPU usage are distinct training setup problems." + }, + { + "left": "issue:11436", + "right": "issue:8955", + "accept": false, + "reason": "A dtype mismatch in Hidream inference is not the same as a CLIPTextModel state_dict loading error." + }, + { + "left": "issue:5575", + "right": "issue:9602", + "accept": false, + "reason": "Hidden-size configuration in LoRA training is unrelated to adapter-name lookup failures." + }, + { + "left": "issue:11791", + "right": "issue:12673", + "accept": false, + "reason": "Both involve group offloading, but one is LoRA weights failing to load and the other is a pipeline compatibility issue with record_stream=True; not clearly the same bug." + }, + { + "left": "issue:6906", + "right": "issue:8543", + "accept": false, + "reason": "Different Dreambooth example failures: a broken LoRA example versus SD3 gradient checkpointing failure." + }, + { + "left": "issue:7082", + "right": "issue:7147", + "accept": false, + "reason": "These are unrelated import errors for different symbols from different modules." + }, + { + "left": "issue:9142", + "right": "issue:9297", + "accept": false, + "reason": "One is a controlnet loading AttributeError and the other is linear LoRA load-time slowdown; different issues." + }, + { + "left": "issue:3296", + "right": "issue:7619", + "accept": false, + "reason": "Both are Dreambooth.py issues, but one concerns checkpoint save/load with text_encoder and the other a channels mismatch at load time." + }, + { + "left": "issue:2036", + "right": "issue:6107", + "accept": false, + "reason": "An OSError on a model load is not the same as an IP-Adapter deployment/runtime issue on AWS EC2." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The canonical issue choice for 11209 vs 11224 is supported by the identical titles, earlier creation time, and inbound reference on 11209. The only accepted soft edge (10715/8970) is reasonably justified by the shared black-image symptom on non-CUDA backends. I don\u2019t see unsupported overstatement in the rejected pairs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 145, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10054", + "issue:10210", + "issue:10227", + "issue:10925", + "issue:10928", + "issue:11023", + "issue:11127", + "issue:11258", + "issue:11422", + "issue:11907", + "issue:1214", + "issue:1363", + "issue:1672", + "issue:1917", + "issue:2107", + "issue:2577", + "issue:2674", + "issue:3284", + "issue:3442", + "issue:3487", + "issue:3871", + "issue:3901", + "issue:4185", + "issue:4255", + "issue:4307", + "issue:4704", + "issue:4728", + "issue:4756", + "issue:4840", + "issue:5014", + "issue:5254", + "issue:5605", + "issue:6054", + "issue:6323", + "issue:6749", + "issue:6911", + "issue:7180", + "issue:7243", + "issue:7616", + "issue:771", + "issue:7948", + "issue:8309", + "issue:8414", + "issue:8738", + "issue:9202", + "issue:9514", + "issue:9841", + "issue:9856" + ], + "soft_pairs": [ + "issue:1672|issue:4185", + "issue:10210|issue:10227", + "issue:1214|issue:4840", + "issue:1363|issue:4307", + "issue:10925|issue:9514", + "issue:3284|issue:4704", + "issue:11907|issue:9202", + "issue:1917|issue:5605", + "issue:2674|issue:771", + "issue:7243|issue:7948", + "issue:2577|issue:3442", + "issue:3871|issue:4728", + "issue:11023|issue:7616", + "issue:11127|issue:11258", + "issue:6323|issue:9856", + "issue:10928|issue:3901", + "issue:6911|issue:8738", + "issue:8309|issue:8414", + "issue:10054|issue:6054", + "issue:4756|issue:5014", + "issue:5254|issue:7180", + "issue:11422|issue:9841", + "issue:2107|issue:4255", + "issue:3487|issue:6749" + ], + "prepared_review_unit_hash": "f09e600c607603d88edb39d7bd66b2811b04a9f779706dbf3f0c16081ae25a71", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34580, + "estimated_input_tokens": 8645, + "estimated_eval_tokens": 17546 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose similarity cluster, not a duplicate set: the items span unrelated scheduler bugs, training failures, model-loading issues, and feature requests. I reject every soft edge; none looks like the same underlying bug/change.", + "confidence": 0.94, + "canonical_issue_reason": "No strong canonical issue: the items are heterogeneous and do not share one underlying failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6054 is the cleanest concrete bug report in the set, but only as a loose anchor; it does not represent the rest of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1672", + "right": "issue:4185", + "accept": false, + "reason": "Different problems: ckpt-to-diffusers image-size regression vs SDXL ControlNet black images." + }, + { + "left": "issue:10210", + "right": "issue:10227", + "accept": false, + "reason": "Flux img2img near-identity output is unrelated to the LoRA feature limitation error." + }, + { + "left": "issue:1214", + "right": "issue:4840", + "accept": false, + "reason": "Half-precision CLIPTextModel failure is unrelated to FlaxAutoencoderKL checkpoint loading." + }, + { + "left": "issue:1363", + "right": "issue:4307", + "accept": false, + "reason": "ONNX scheduler error for long prompts is unrelated to local_files_only from_pretrained failure." + }, + { + "left": "issue:10925", + "right": "issue:9514", + "accept": false, + "reason": "Both mention FLUX, but one is DreamBooth/CLIPTextModel training and the other is fp8 LoRA loading after quantization." + }, + { + "left": "issue:3284", + "right": "issue:4704", + "accept": false, + "reason": "DreamBooth LoRA accelerate failure and custom diffusion training failure are different training scripts with different failure modes." + }, + { + "left": "issue:11907", + "right": "issue:9202", + "accept": false, + "reason": "Scheduler incompatibility in Flux is unrelated to the SDXL PAG/IP-Adapter FaceID issue." + }, + { + "left": "issue:1917", + "right": "issue:5605", + "accept": false, + "reason": "These are different feature requests: integration testing vs per-image captions." + }, + { + "left": "issue:2674", + "right": "issue:771", + "accept": false, + "reason": "LoRA safetensors packaging issue is unrelated to BF16 DreamBooth behavior." + }, + { + "left": "issue:7243", + "right": "issue:7948", + "accept": false, + "reason": "Different model proposals: ResAdapter vs MuLan language adapter." + }, + { + "left": "issue:2577", + "right": "issue:3442", + "accept": false, + "reason": "DDIM inverse scheduler problem is unrelated to textual inversion embedding dimension mismatch." + }, + { + "left": "issue:3871", + "right": "issue:4728", + "accept": false, + "reason": "Both mention resume training, but one is an error on resume and the other is resume OOM; not the same bug." + }, + { + "left": "issue:11023", + "right": "issue:7616", + "accept": false, + "reason": "Both are image-quality complaints, but one is blur in a custom pipeline and the other is low-denoise artifacts." + }, + { + "left": "issue:11127", + "right": "issue:11258", + "accept": false, + "reason": "Different single-file model support problems for different model families (Flux vs LTX)." + }, + { + "left": "issue:6323", + "right": "issue:9856", + "accept": false, + "reason": "TensorRT guidance_scale attribute error is unrelated to distributed port-collision startup failure." + }, + { + "left": "issue:10928", + "right": "issue:3901", + "accept": false, + "reason": "Missing model weight file and missing requirements.txt are unrelated filesystem/package issues." + }, + { + "left": "issue:6911", + "right": "issue:8738", + "accept": false, + "reason": "Different model-class bugs: class labels in UNetConditional3DModel vs SD3Transformer2DModel configuration." + }, + { + "left": "issue:8309", + "right": "issue:8414", + "accept": false, + "reason": "Both are distillation-related ideas, but they are distinct model/change requests." + }, + { + "left": "issue:10054", + "right": "issue:6054", + "accept": false, + "reason": "Both involve EulerDiscreteScheduler, but one is torch.linspace argument handling in tests and the other is torch.from_numpy misuse in set_timesteps; different bugs." + }, + { + "left": "issue:4756", + "right": "issue:5014", + "accept": false, + "reason": "Training-style support request is unrelated to poor DreamBooth LoRA results." + }, + { + "left": "issue:5254", + "right": "issue:7180", + "accept": false, + "reason": "Multi-ControlNet support request is unrelated to a CPU-offload failure in SDXL img2img ControlNet." + }, + { + "left": "issue:11422", + "right": "issue:9841", + "accept": false, + "reason": "Different DreamBooth/LoRA training bugs with different triggers and error types." + }, + { + "left": "issue:2107", + "right": "issue:4255", + "accept": false, + "reason": "Low-resolution glitches are not the same as generally worse results than sd-webui." + }, + { + "left": "issue:3487", + "right": "issue:6749", + "accept": false, + "reason": "Inpaint quality with EulerAncestralDiscreteScheduler is unrelated to InstantID community-example replication." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 146, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10055", + "issue:10066", + "issue:10178", + "issue:10282", + "issue:10392", + "issue:10722", + "issue:12157", + "issue:12458", + "issue:13079", + "issue:13140", + "issue:1617", + "issue:1883", + "issue:2392", + "issue:2695", + "issue:3101", + "issue:3175", + "issue:3218", + "issue:3379", + "issue:3405", + "issue:4206", + "issue:4349", + "issue:4726", + "issue:4885", + "issue:540", + "issue:5433", + "issue:5515", + "issue:5617", + "issue:5846", + "issue:5925", + "issue:5972", + "issue:6354", + "issue:6368", + "issue:6996", + "issue:7133", + "issue:7146", + "issue:7197", + "issue:7593", + "issue:7690", + "issue:7793", + "issue:8589", + "issue:8879", + "issue:9091", + "issue:9178", + "issue:9556", + "issue:9558", + "issue:9630", + "issue:9667", + "issue:9717" + ], + "soft_pairs": [ + "issue:4349|issue:5515", + "issue:3218|issue:5617", + "issue:8879|issue:9091", + "issue:10066|issue:6368", + "issue:10282|issue:12157", + "issue:1883|issue:540", + "issue:3379|issue:7690", + "issue:5433|issue:9556", + "issue:9667|issue:9717", + "issue:1617|issue:3175", + "issue:4726|issue:7593", + "issue:4885|issue:7793", + "issue:7197|issue:9558", + "issue:10178|issue:10722", + "issue:5972|issue:9630", + "issue:10055|issue:8589", + "issue:12458|issue:13140", + "issue:5925|issue:6354", + "issue:3379|issue:7133", + "issue:10392|issue:9178", + "issue:1617|issue:5846", + "issue:2695|issue:3101", + "issue:3405|issue:4206", + "issue:13079|issue:7146", + "issue:2392|issue:6996" + ], + "prepared_review_unit_hash": "f264fbeefe737a7e8efd905783f2366a91c8ba6fe6445a930a86956e3f3a142c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34663, + "estimated_input_tokens": 8666, + "estimated_eval_tokens": 17588 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the soft-similarity pairs mostly share broad themes like training, ControlNet, LoRA, or torch.compile, but they describe different bugs, models, and code paths. I would not merge any of the proposed pairs as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:4206 is the closest representative of the shared theme because it is a concrete ControlNet training regression with a clear model/context, unlike the more generic or unrelated errors elsewhere in the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4349", + "right": "issue:5515", + "accept": false, + "reason": "Different failures: one needs a text_encoder despite prompt_embeds, the other is an LCM init/scheduler loading issue." + }, + { + "left": "issue:3218", + "right": "issue:5617", + "accept": false, + "reason": "Both involve torch.compile, but one is graph-break behavior and the other is save/load after compilation; different code paths." + }, + { + "left": "issue:8879", + "right": "issue:9091", + "accept": false, + "reason": "Different problems: cudagraph/torch.compile failure vs VAE decoder compile support." + }, + { + "left": "issue:10066", + "right": "issue:6368", + "accept": false, + "reason": "Unpack error in self.attn is unrelated to the checkpoint/LoRA alpha unsupported error." + }, + { + "left": "issue:10282", + "right": "issue:12157", + "accept": false, + "reason": "Shape mismatch loading x_embedder weights is unrelated to the torch.device iteration error in Qwen training." + }, + { + "left": "issue:1883", + "right": "issue:540", + "accept": false, + "reason": "One is VAE gradient checkpointing support; the other is memory placement for UNet." + }, + { + "left": "issue:3379", + "right": "issue:7690", + "accept": false, + "reason": "Checkpoint loading failure and tokenizer loading failure are distinct issues." + }, + { + "left": "issue:5433", + "right": "issue:9556", + "accept": false, + "reason": "Scheduler artifacts in SDXL are unrelated to saved LoRA model issues with rslora." + }, + { + "left": "issue:9667", + "right": "issue:9717", + "accept": false, + "reason": "Flux fp8 model loading and Flux docker image errors are not the same bug." + }, + { + "left": "issue:1617", + "right": "issue:3175", + "accept": false, + "reason": "Missing CLI arguments in dreambooth.py is unrelated to a multi-face training complaint." + }, + { + "left": "issue:4726", + "right": "issue:7593", + "accept": false, + "reason": "Callback typing is a different API issue from missing IP-adapter image projection layers." + }, + { + "left": "issue:4885", + "right": "issue:7793", + "accept": false, + "reason": "LoRA loading in SDXL ControlNet img2img is different from unloading LoRA weights in DeepFloyd." + }, + { + "left": "issue:7197", + "right": "issue:9558", + "accept": false, + "reason": "Resume-from-checkpoint KeyError is unrelated to black validation images in TensorBoard." + }, + { + "left": "issue:10178", + "right": "issue:10722", + "accept": false, + "reason": "A UNet 3D block size mismatch is different from a DreamBooth prior-preservation tensor mismatch." + }, + { + "left": "issue:5972", + "right": "issue:9630", + "accept": false, + "reason": "Documentation for min-SNR weighting is unrelated to the flux-controlnet num_single_layers runtime error." + }, + { + "left": "issue:10055", + "right": "issue:8589", + "accept": false, + "reason": "Both mention SD3 training, but one is ControlNet training script failure and the other is DreamBooth prior-preservation behavior." + }, + { + "left": "issue:12458", + "right": "issue:13140", + "accept": false, + "reason": "Batch inference feature request is unrelated to the NoneType iteration failure in Qwen Edit pipelines." + }, + { + "left": "issue:5925", + "right": "issue:6354", + "accept": false, + "reason": "ZipLora multi-LoRA inference and ControlNet LoRA support are different feature requests." + }, + { + "left": "issue:3379", + "right": "issue:7133", + "accept": false, + "reason": "Checkpoint loading in DreamBooth and SDXL text-encoder size mismatch on resume are different resume/load bugs." + }, + { + "left": "issue:10392", + "right": "issue:9178", + "accept": false, + "reason": "Both concern FLUX LoRA training, but one is an fp8/Hyper-FLUX setup and the other is a peft target-module error." + }, + { + "left": "issue:1617", + "right": "issue:5846", + "accept": false, + "reason": "Missing required args in train_dreambooth.py is unrelated to the resume_from_checkpoint error in an SDXL LoRA script." + }, + { + "left": "issue:2695", + "right": "issue:3101", + "accept": false, + "reason": "TensorBoard validation failure and a generic ControlNet training example error do not appear to be the same bug." + }, + { + "left": "issue:3405", + "right": "issue:4206", + "accept": false, + "reason": "Both are ControlNet training issues, but one is generic while the other is specifically SDXL/main-branch breakage; not enough evidence of the same root cause." + }, + { + "left": "issue:13079", + "right": "issue:7146", + "accept": false, + "reason": "FLUX.2-klein looped-image CUDA OOM is not the same as SDXL fine-tuning OOM." + }, + { + "left": "issue:2392", + "right": "issue:6996", + "accept": false, + "reason": "Both involve NoneType-related errors, but they occur in different pipeline inputs/paths and are not the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 147, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10049", + "issue:10227", + "issue:10540", + "issue:10653", + "issue:10937", + "issue:11241", + "issue:11435", + "issue:11749", + "issue:1185", + "issue:11864", + "issue:13104", + "issue:1375", + "issue:1541", + "issue:175", + "issue:2668", + "issue:3296", + "issue:3379", + "issue:3468", + "issue:3765", + "issue:4000", + "issue:4107", + "issue:4241", + "issue:4324", + "issue:4530", + "issue:455", + "issue:4754", + "issue:5022", + "issue:5441", + "issue:5558", + "issue:5661", + "issue:5811", + "issue:6139", + "issue:6218", + "issue:6574", + "issue:6661", + "issue:7082", + "issue:7136", + "issue:7357", + "issue:752", + "issue:760", + "issue:7785", + "issue:8729", + "issue:8915", + "issue:9142", + "issue:9200", + "issue:954", + "issue:9822", + "issue:988" + ], + "soft_pairs": [ + "issue:4241|issue:752", + "issue:455|issue:760", + "issue:5661|issue:9200", + "issue:4530|issue:6574", + "issue:4107|issue:8729", + "issue:10227|issue:10653", + "issue:3379|issue:954", + "issue:5811|issue:7136", + "issue:3468|issue:4754", + "issue:5558|issue:9142", + "issue:1185|issue:1541", + "issue:4000|issue:7785", + "issue:11864|issue:13104", + "issue:3296|issue:4324", + "issue:5441|issue:6139", + "issue:11241|issue:11749", + "issue:2668|issue:6218", + "issue:3765|issue:6661", + "issue:10049|issue:8915", + "issue:10937|issue:7357", + "issue:1375|issue:5022", + "issue:175|issue:988", + "issue:11435|issue:7082", + "issue:10540|issue:9822" + ], + "prepared_review_unit_hash": "015f757e996be4cfc443d86222ca88ddf7c946b318b04fb22b14f5a3d90dec9a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34180, + "estimated_input_tokens": 8545, + "estimated_eval_tokens": 17346 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No strong duplicate cluster here: the soft pairs mostly share only a broad subsystem (imports, training, ControlNet, compile, MPS) but describe different concrete failures. I would not merge any of the proposed soft edges.", + "confidence": 0.93, + "canonical_issue_reason": "No single issue is a safe canonical representative; the items span unrelated bugs and feature requests rather than one underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a clearly better global representative because the cluster is heterogeneous and the soft-paired items do not share the same failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4241", + "right": "issue:752", + "accept": false, + "reason": "Different requests: LoRA support for unidiffusers vs multi-concept Dreambooth training." + }, + { + "left": "issue:455", + "right": "issue:760", + "accept": false, + "reason": "Different MPS problems: one is a missing op implementation, the other is incorrect StableDiffusion output." + }, + { + "left": "issue:5661", + "right": "issue:9200", + "accept": false, + "reason": "Import-time NameError for PreTrainedTokenizer vs SDXL ControlNet dtype mismatch." + }, + { + "left": "issue:4530", + "right": "issue:6574", + "accept": false, + "reason": "SDXL refiner image artifacts vs SVD pipeline warning about expected input range; different symptoms and code paths." + }, + { + "left": "issue:4107", + "right": "issue:8729", + "accept": false, + "reason": "Both involve training, but one is SDXL LoRA quality degradation while the other is a validation-image issue in ControlNet training." + }, + { + "left": "issue:10227", + "right": "issue:10653", + "accept": false, + "reason": "Unrelated bugs: LoRA feature-dimension support vs bitsandbytes 8-bit `.to()` to CPU unsupported." + }, + { + "left": "issue:3379", + "right": "issue:954", + "accept": false, + "reason": "Dreambooth checkpoint loading vs login/loading failure; no shared underlying bug." + }, + { + "left": "issue:5811", + "right": "issue:7136", + "accept": false, + "reason": "General callback API expansion vs a single pipeline asking for callback support; not the same change." + }, + { + "left": "issue:3468", + "right": "issue:4754", + "accept": false, + "reason": "Both are save/load file issues, but one is Dreambooth export metadata and the other is textual inversion model loading." + }, + { + "left": "issue:5558", + "right": "issue:9142", + "accept": false, + "reason": "Different multiple-ControlNet failures on different models and different errors (artifacts vs attribute/type error)." + }, + { + "left": "issue:1185", + "right": "issue:1541", + "accept": false, + "reason": "Different Dreambooth concerns: reproducibility vs model-type detection/loading checkpoints." + }, + { + "left": "issue:4000", + "right": "issue:7785", + "accept": false, + "reason": "Converting ckpt format is unrelated to first-inference slowness." + }, + { + "left": "issue:11864", + "right": "issue:13104", + "accept": false, + "reason": "Separate compile-time NameErrors in different modules: missing `torch` vs missing `logger`." + }, + { + "left": "issue:3296", + "right": "issue:4324", + "accept": false, + "reason": "Checkpoint save/load bug in Dreambooth vs a `NoneType` error in UNet forward." + }, + { + "left": "issue:5441", + "right": "issue:6139", + "accept": false, + "reason": "Related vocabulary (`attn_processors`), but one is a missing attribute in a Wuerstchen prior training script and the other is a model lacking the attribute entirely." + }, + { + "left": "issue:11241", + "right": "issue:11749", + "accept": false, + "reason": "Both mention floating point problems, but one is a training crash and the other is scheduler timestep precision." + }, + { + "left": "issue:2668", + "right": "issue:6218", + "accept": false, + "reason": "ControlNet image routing bug vs batched seed handling bug; different pipeline behavior." + }, + { + "left": "issue:3765", + "right": "issue:6661", + "accept": false, + "reason": "Stacking artifacts during ControlNet training vs MPS OOM; different failure modes." + }, + { + "left": "issue:10049", + "right": "issue:8915", + "accept": false, + "reason": "Both are import errors, but for different pipeline symbols and unrelated code paths." + }, + { + "left": "issue:10937", + "right": "issue:7357", + "accept": false, + "reason": "One is a `torch.compile` runtime error on `vae.encode`, the other is a performance complaint despite compile/fuse_loras." + }, + { + "left": "issue:1375", + "right": "issue:5022", + "accept": false, + "reason": "Different import breakages with different root causes and symbols." + }, + { + "left": "issue:175", + "right": "issue:988", + "accept": false, + "reason": "Manual seed correctness in example code vs FSDP training support; unrelated topics." + }, + { + "left": "issue:11435", + "right": "issue:7082", + "accept": false, + "reason": "Different import failures from different packages: `HiDreamImagePipeline` vs `CpuOffload`." + }, + { + "left": "issue:10540", + "right": "issue:9822", + "accept": false, + "reason": "Flux from_single_file loading regression vs loading SDXL LoRAs into Flux; not the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 148, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10049", + "issue:10382", + "issue:11321", + "issue:11574", + "issue:11989", + "issue:1774", + "issue:1802", + "issue:1925", + "issue:1976", + "issue:2163", + "issue:2203", + "issue:2214", + "issue:2480", + "issue:2581", + "issue:2774", + "issue:2844", + "issue:3346", + "issue:3454", + "issue:3461", + "issue:3674", + "issue:3717", + "issue:3803", + "issue:4030", + "issue:4031", + "issue:4125", + "issue:455", + "issue:4615", + "issue:4835", + "issue:4935", + "issue:5540", + "issue:5647", + "issue:5875", + "issue:5956", + "issue:6250", + "issue:6362", + "issue:6435", + "issue:6713", + "issue:6899", + "issue:7082", + "issue:7431", + "issue:7483", + "issue:776", + "issue:7785", + "issue:9063", + "issue:9178", + "issue:9558", + "issue:993" + ], + "soft_pairs": [ + "issue:2214|issue:776", + "issue:3346|issue:3454", + "issue:11989|issue:9558", + "issue:11321|issue:9178", + "issue:2844|issue:4030", + "issue:2203|issue:2480", + "issue:455|issue:993", + "issue:1976|issue:6435", + "issue:4615|issue:7082", + "issue:2581|issue:2774", + "issue:5956|issue:6362", + "issue:1774|issue:5540", + "issue:10382|issue:11574", + "issue:2163|issue:4935", + "issue:5875|issue:7483", + "issue:6713|issue:6899", + "issue:3674|issue:4125", + "issue:3717|issue:3803", + "issue:4031|issue:6250", + "issue:7785|issue:9063", + "issue:10049|issue:7082", + "issue:1802|issue:1925", + "issue:3461|issue:7431", + "issue:4835|issue:5647" + ], + "prepared_review_unit_hash": "d8b3b5849fe1f7218e5735e2a6de18eff4142cca326d69316ff094b563e540f6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33526, + "estimated_input_tokens": 8382, + "estimated_eval_tokens": 17020 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly an overmatch of unrelated issues sharing generic ML/diffusers vocabulary. A few pairs are thematically adjacent, but none are clearly the same underlying bug or change.", + "confidence": 0.88, + "canonical_issue_reason": "No single issue cleanly represents a true duplicate set here; the only somewhat related subset is around the Stable Diffusion conversion script, so #6899 is the broadest placeholder if one must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "#6899 is the most general, code-path-focused report in the closest subtheme (conversion script), but the cluster overall is too heterogeneous to treat it as a strong canonical issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2214", + "right": "issue:776", + "accept": false, + "reason": "Both are SD-related runtime problems, but one is an NSFW filter regression and the other is an image-to-image batch-size bug; different code paths." + }, + { + "left": "issue:3346", + "right": "issue:3454", + "accept": false, + "reason": "Both mention dreambooth LoRA training, but one is about resume_from_checkpoint support and the other is a generic latest-version failure." + }, + { + "left": "issue:11989", + "right": "issue:9558", + "accept": false, + "reason": "Both mention black images, but one is black inference output and the other is black validation images in TensorBoard; likely different causes." + }, + { + "left": "issue:11321", + "right": "issue:9178", + "accept": false, + "reason": "One is a README bug and the other is a Flux LoRA target-modules training error; not the same underlying issue." + }, + { + "left": "issue:2844", + "right": "issue:4030", + "accept": false, + "reason": "Both are embedding-related runtime errors, but they affect different pipelines and different failure modes." + }, + { + "left": "issue:2203", + "right": "issue:2480", + "accept": false, + "reason": "Both involve dreambooth training directories, but one is model download location and the other is checkpoint folder creation." + }, + { + "left": "issue:455", + "right": "issue:993", + "accept": false, + "reason": "Both are dtype/device stability issues, but one is an MPS index op failure and the other is fp16 type mismatch." + }, + { + "left": "issue:1976", + "right": "issue:6435", + "accept": false, + "reason": "These concern unrelated bugs: a wrong tokenizer file path versus ControlNet default width/height initialization." + }, + { + "left": "issue:4615", + "right": "issue:7082", + "accept": false, + "reason": "Both are import errors, but for unrelated modules and different dependency breakages." + }, + { + "left": "issue:2581", + "right": "issue:2774", + "accept": false, + "reason": "Both are about ControlNet conversion, but one is a feature request and the other is a specific safetensors conversion error." + }, + { + "left": "issue:5956", + "right": "issue:6362", + "accept": false, + "reason": "Both are SDXL fine-tuning complaints, but one is performance degradation and the other is dataset-mapping OOM." + }, + { + "left": "issue:1774", + "right": "issue:5540", + "accept": false, + "reason": "Both are OOM reports, but they are generic memory exhaustion symptoms without evidence of the same root cause." + }, + { + "left": "issue:10382", + "right": "issue:11574", + "accept": false, + "reason": "These are unrelated VAE/weight-loading bugs: a return-type mismatch versus a missing weight key." + }, + { + "left": "issue:2163", + "right": "issue:4935", + "accept": false, + "reason": "Both mention training scripts, but one is dtype mismatch in text-to-image training and the other is a LoRA layer-API question." + }, + { + "left": "issue:5875", + "right": "issue:7483", + "accept": false, + "reason": "These are different conversion problems: ONNX prompt issues versus LoRA-to-Diffusers conversion on SDXL base." + }, + { + "left": "issue:6713", + "right": "issue:6899", + "accept": false, + "reason": "Same utility file, but the titles point to different reports; insufficient evidence they are the same concrete bug." + }, + { + "left": "issue:3674", + "right": "issue:4125", + "accept": false, + "reason": "Data/metadata parsing in one issue and loss-curve expectations in the other are not the same bug." + }, + { + "left": "issue:3717", + "right": "issue:3803", + "accept": false, + "reason": "Both involve multi-GPU training, but one is OOM and the other is a model-sharing/distribution problem." + }, + { + "left": "issue:4031", + "right": "issue:6250", + "accept": false, + "reason": "Both are loading-related, but one is safety-checker local loading and the other is SDXL VAE single-file loading." + }, + { + "left": "issue:7785", + "right": "issue:9063", + "accept": false, + "reason": "Both mention startup/inference friction, but one is slow first inference and the other is tokenizer instantiation failure." + }, + { + "left": "issue:10049", + "right": "issue:7082", + "accept": false, + "reason": "Both are import failures, but they involve different packages and unrelated missing symbols." + }, + { + "left": "issue:1802", + "right": "issue:1925", + "accept": false, + "reason": "One is a UNet encoder_hidden_states handling bug and the other is an fp16 inpainting issue; not the same failure." + }, + { + "left": "issue:3461", + "right": "issue:7431", + "accept": false, + "reason": "Both are training-performance concerns, but one is DreamBooth efficiency and the other is DoRA text-encoder training issues." + }, + { + "left": "issue:4835", + "right": "issue:5647", + "accept": false, + "reason": "Both are pipeline bugs, but one is MultiControlNet handling and the other is SDXL inpaint strength==1 behavior." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 149, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10475", + "issue:10863", + "issue:11363", + "issue:11383", + "issue:11396", + "issue:12175", + "issue:12489", + "issue:1290", + "issue:1388", + "issue:1524", + "issue:1565", + "issue:1798", + "issue:1915", + "issue:1985", + "issue:2312", + "issue:2780", + "issue:2784", + "issue:2844", + "issue:3087", + "issue:3175", + "issue:3193", + "issue:3539", + "issue:381", + "issue:3871", + "issue:4241", + "issue:4304", + "issue:434", + "issue:4386", + "issue:4680", + "issue:5016", + "issue:5197", + "issue:5504", + "issue:5577", + "issue:5619", + "issue:5857", + "issue:5966", + "issue:6203", + "issue:6278", + "issue:6333", + "issue:7049", + "issue:7114", + "issue:7238", + "issue:7311", + "issue:7871", + "issue:7960", + "issue:8390", + "issue:9154", + "issue:9856" + ], + "soft_pairs": [ + "issue:10863|issue:9154", + "issue:11363|issue:12489", + "issue:11383|issue:11396", + "issue:1290|issue:1915", + "issue:1388|issue:1798", + "issue:1524|issue:2780", + "issue:1985|issue:3193", + "issue:2784|issue:3087", + "issue:3539|issue:6203", + "issue:381|issue:434", + "issue:3871|issue:5197", + "issue:4241|issue:4680", + "issue:4304|issue:4386", + "issue:5504|issue:6333", + "issue:5966|issue:8390", + "issue:6278|issue:7960", + "issue:7114|issue:7311", + "issue:3175|issue:5016", + "issue:7049|issue:7238", + "issue:5619|issue:5857", + "issue:5577|issue:7871", + "issue:10475|issue:12175", + "issue:2312|issue:2844", + "issue:1565|issue:9856" + ], + "prepared_review_unit_hash": "b8a28d90d77b985d7261ba7fd1a535a073674012e65788d6a2a4ab444270811b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34074, + "estimated_input_tokens": 8519, + "estimated_eval_tokens": 17294 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a grab bag of unrelated Diffusers issues: model-loading, ControlNet, LoRA/LoCon, SDXL training, IP-Adapter, and a few usage questions. None of the soft pairs look like the same underlying bug or change closely enough to merge.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a clear canonical for this set; the items are too heterogeneous, though training/inference bugs dominate overall.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10863", + "right": "issue:9154", + "accept": false, + "reason": "Both are loading-related, but one is a local safetensors load error and the other is a broken PT/VAE loading path; different artifact types and failure modes." + }, + { + "left": "issue:11363", + "right": "issue:12489", + "accept": false, + "reason": "CUDA OOM in a ControlNet example is unrelated to an sm_120 kernel compatibility error; different root causes and fixes." + }, + { + "left": "issue:11383", + "right": "issue:11396", + "accept": false, + "reason": "One asks for HiDream LoRA support; the other asks how to convert an existing HiDream LoRA for ComfyUI. Feature request vs. conversion question." + }, + { + "left": "issue:1290", + "right": "issue:1915", + "accept": false, + "reason": "DreamArtist community proposal is unrelated to unifying/uploading UnCLIP weights." + }, + { + "left": "issue:1388", + "right": "issue:1798", + "accept": false, + "reason": "Both mention SD2.0 support, but one is about a conversion script update and the other is about v_prediction support in a training script; not the same bug." + }, + { + "left": "issue:1524", + "right": "issue:2780", + "accept": false, + "reason": "Poor SD2.0 training quality and FP16 unscale errors are different issues with different symptoms and likely fixes." + }, + { + "left": "issue:1985", + "right": "issue:3193", + "accept": false, + "reason": "Both concern embeddings, but one is loading new embeddings in a pipeline and the other is batch-loading textual inversion embeddings; enhancement scope differs." + }, + { + "left": "issue:2784", + "right": "issue:3087", + "accept": false, + "reason": "Dreambooth for text2video and LyCORIS support are separate feature requests in the same broad training area." + }, + { + "left": "issue:3539", + "right": "issue:6203", + "accept": false, + "reason": "main_process_first is a distributed-training question; SDXL ControlNet image-size conditioning is a model-input behavior question." + }, + { + "left": "issue:381", + "right": "issue:434", + "accept": false, + "reason": "Both reference notebook/training breakage, but they target different docs and versions; not clearly the same bug." + }, + { + "left": "issue:3871", + "right": "issue:5197", + "accept": false, + "reason": "Checkpoint resume errors and multi-GPU slowdown are distinct training problems." + }, + { + "left": "issue:4241", + "right": "issue:4680", + "accept": false, + "reason": "LoRA support for unidiffusers and SDXL inpainting fine-tuning are different feature requests." + }, + { + "left": "issue:4304", + "right": "issue:4386", + "accept": false, + "reason": "SDXL LoRA training support and converting an SDXL LoRA safetensor are related, but not the same concrete bug/change." + }, + { + "left": "issue:5504", + "right": "issue:6333", + "accept": false, + "reason": "PEFT LoRA+xformers incompatibility is a runtime issue; finetuning a pretrained LoRA by PEFT is a different capability request." + }, + { + "left": "issue:5966", + "right": "issue:8390", + "accept": false, + "reason": "Stable Diffusion 2.1 ControlNet and Stable Cascade ControlNet training are different models and workflows." + }, + { + "left": "issue:6278", + "right": "issue:7960", + "accept": false, + "reason": "Both are multi-GPU training failures, but they are for different scripts/models and do not appear to share one code-path bug." + }, + { + "left": "issue:7114", + "right": "issue:7311", + "accept": false, + "reason": "CUDA OOM during SDXL training and failure to save checkpoints are separate failure points." + }, + { + "left": "issue:3175", + "right": "issue:5016", + "accept": false, + "reason": "Multi-face DreamBooth training and SDXL text-encoder training support are different training requests." + }, + { + "left": "issue:7049", + "right": "issue:7238", + "accept": false, + "reason": "IP-Adapter unpacking error and IP-Adapter masking are distinct issues; same subsystem but not same bug." + }, + { + "left": "issue:5619", + "right": "issue:5857", + "accept": false, + "reason": "Autoencoder image-encode failure in LCM img2img and weak LCM LoRA performance in AnimateDiff are unrelated." + }, + { + "left": "issue:5577", + "right": "issue:7871", + "accept": false, + "reason": "ControlNet single-file loading KeyError and blockwise-scale LoRA error are different code paths." + }, + { + "left": "issue:10475", + "right": "issue:12175", + "accept": false, + "reason": "Both complain about image quality, but they involve different models (SD3 vs Qwen Image) and likely different causes." + }, + { + "left": "issue:2312", + "right": "issue:2844", + "accept": false, + "reason": "EMA save-hook failure in text-to-image finetuning and StableUnCLIP image-embedding runtime error are unrelated." + }, + { + "left": "issue:1565", + "right": "issue:9856", + "accept": false, + "reason": "GPU selection/OOM and distributed port-in-use errors are different distributed/runtime configuration problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 150, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1005", + "issue:10337", + "issue:10972", + "issue:1181", + "issue:11833", + "issue:12012", + "issue:1261", + "issue:12689", + "issue:1407", + "issue:1524", + "issue:1865", + "issue:1969", + "issue:2250", + "issue:2439", + "issue:2444", + "issue:2507", + "issue:2775", + "issue:2786", + "issue:2819", + "issue:2893", + "issue:3061", + "issue:3198", + "issue:3468", + "issue:4186", + "issue:4366", + "issue:4782", + "issue:5023", + "issue:5224", + "issue:5226", + "issue:5327", + "issue:5737", + "issue:5892", + "issue:6239", + "issue:6383", + "issue:6435", + "issue:6713", + "issue:6815", + "issue:7357", + "issue:7978", + "issue:7985", + "issue:8450", + "issue:8832", + "issue:8854", + "issue:8955", + "issue:9497", + "issue:9810", + "issue:9876", + "issue:9913" + ], + "soft_pairs": [ + "issue:12012|issue:1407", + "issue:1524|issue:2819", + "issue:2250|issue:3468", + "issue:10337|issue:10972", + "issue:2507|issue:3198", + "issue:1181|issue:4366", + "issue:5226|issue:7978", + "issue:7357|issue:7985", + "issue:2439|issue:3061", + "issue:2893|issue:6435", + "issue:4782|issue:8450", + "issue:5023|issue:5892", + "issue:5737|issue:6713", + "issue:9876|issue:9913", + "issue:1261|issue:1969", + "issue:1865|issue:2444", + "issue:5327|issue:8854", + "issue:6815|issue:8832", + "issue:4186|issue:5224", + "issue:8955|issue:9497", + "issue:2775|issue:6239", + "issue:1005|issue:2786", + "issue:11833|issue:12689", + "issue:6383|issue:9810" + ], + "prepared_review_unit_hash": "0456e0f0b0178d5d46c4498100ffb317b7f3bb3be780b2b8eae57d60927cd222", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34001, + "estimated_input_tokens": 8501, + "estimated_eval_tokens": 17258 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a set of weakly related, non-duplicate issues. Several pairs share broad themes like LoRA or loading errors, but they describe different code paths, symptoms, or pipelines, so none should be merged as duplicates.", + "confidence": 0.92, + "canonical_issue_reason": "issue:12689 is the clearest concrete bug report in the set: LoRA adapter weights not loading from checkpoints. It is specific, reproducible, and broadly representative of the LoRA-loading failures here without being overly tied to one niche symptom.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:12689 is the best single issue to anchor this cluster because it names a direct loading failure and is easier to match against follow-up reports than the more vague or symptom-only issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12012", + "right": "issue:1407", + "accept": false, + "reason": "Both mention runtime errors in pipelines, but one is a Wan 2.2 context-state failure and the other is an ONNX SD 2.0 feature_extractor error; different subsystems and root causes." + }, + { + "left": "issue:1524", + "right": "issue:2819", + "accept": false, + "reason": "Textual inversion training quality vs a download/network failure for a community script are unrelated bugs." + }, + { + "left": "issue:2250", + "right": "issue:3468", + "accept": false, + "reason": "Dreambooth inference failure and missing model_index.json are different failure modes and likely different fixes." + }, + { + "left": "issue:10337", + "right": "issue:10972", + "accept": false, + "reason": "FLUX IPAdapter quantization failure and OneTrainer Flux LoRA loading failure are both FLUX-related but involve different components and symptoms." + }, + { + "left": "issue:2507", + "right": "issue:3198", + "accept": false, + "reason": "GPU safetensors loading speed vs textual inversion loading failure are not the same bug." + }, + { + "left": "issue:1181", + "right": "issue:4366", + "accept": false, + "reason": "Training from scratch vs an SDXL full U-Net fine-tuning script request are different topics, not duplicates." + }, + { + "left": "issue:5226", + "right": "issue:7978", + "accept": false, + "reason": "SDXL adapter LoRA loading incompatibility is unrelated to a missing torchvision nms operator." + }, + { + "left": "issue:7357", + "right": "issue:7985", + "accept": false, + "reason": "Slow inference with compile/fused LoRAs is a performance concern; IPAdapter not compatible with torch.compile is a different code-path incompatibility." + }, + { + "left": "issue:2439", + "right": "issue:3061", + "accept": false, + "reason": "Inpainting pipeline loading/running issue vs general reproducibility issue do not describe the same bug." + }, + { + "left": "issue:2893", + "right": "issue:6435", + "accept": false, + "reason": "ControlNet width/height mix-up is a functional bug; default width/height initialization is a separate setup issue." + }, + { + "left": "issue:4782", + "right": "issue:8450", + "accept": false, + "reason": "Mask handling in inpainting and SDXL inpaint strength=1.0 noise are different inpainting bugs." + }, + { + "left": "issue:5023", + "right": "issue:5892", + "accept": false, + "reason": "WuerstchenCombinedPipeline import regression and a generic import error are too vague to be the same concrete issue." + }, + { + "left": "issue:5737", + "right": "issue:6713", + "accept": false, + "reason": "SDXL conversion failure and HeaderTooLarge during conversion are both conversion-related but not the same underlying error." + }, + { + "left": "issue:9876", + "right": "issue:9913", + "accept": false, + "reason": "VRAM not released after LoRA training and set_lora_device not working are distinct LoRA lifecycle problems." + }, + { + "left": "issue:1261", + "right": "issue:1969", + "accept": false, + "reason": "Xformers backward assertion and tensor view/stride incompatibility are different low-level tensor errors." + }, + { + "left": "issue:1865", + "right": "issue:2444", + "accept": false, + "reason": "Zero stage 3 fine-tune dimension error and dreambooth inpaint script exit-status failure are not the same bug." + }, + { + "left": "issue:5327", + "right": "issue:8854", + "accept": false, + "reason": "SDXL textual inversion loading error and SDXL textual embedding loading failure are related only broadly; they likely involve different loaders and error causes." + }, + { + "left": "issue:6815", + "right": "issue:8832", + "accept": false, + "reason": "Black image / invalid cast during output conversion is not the same as a crash when calling the pipeline." + }, + { + "left": "issue:4186", + "right": "issue:5224", + "accept": false, + "reason": "SDXL inpaint with refiner and a Flax Stable Diffusion 2.1 error are unrelated models and code paths." + }, + { + "left": "issue:8955", + "right": "issue:9497", + "accept": false, + "reason": "CLIPTextModelWithProjection state-dict loading failure and Dreambooth Flux mat2 shape error are different model-loading/training issues." + }, + { + "left": "issue:2775", + "right": "issue:6239", + "accept": false, + "reason": "torch._dynamo.optimize misuse and FX support for VAE sampling are different torch-integration problems." + }, + { + "left": "issue:1005", + "right": "issue:2786", + "accept": false, + "reason": "Scheduler timestep limit question and numpy-input incompatibility in Depth2Img are unrelated." + }, + { + "left": "issue:11833", + "right": "issue:12689", + "accept": false, + "reason": "Both are LoRA-related, but one is a KeyError in set_lora_device and the other is missing adapter weights from checkpoints; different failure points." + }, + { + "left": "issue:6383", + "right": "issue:9810", + "accept": false, + "reason": "Multi-stage learning rates in get_scheduler and LoRA module_to_save support are unrelated feature requests/issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 151, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10066", + "issue:1100", + "issue:11448", + "issue:11839", + "issue:12009", + "issue:12195", + "issue:12574", + "issue:13386", + "issue:1407", + "issue:1572", + "issue:1976", + "issue:2321", + "issue:2554", + "issue:2645", + "issue:3182", + "issue:3296", + "issue:3346", + "issue:3803", + "issue:3808", + "issue:3978", + "issue:4073", + "issue:4217", + "issue:4389", + "issue:455", + "issue:4911", + "issue:4985", + "issue:5023", + "issue:5412", + "issue:5455", + "issue:5799", + "issue:6176", + "issue:6238", + "issue:6287", + "issue:6351", + "issue:6598", + "issue:6602", + "issue:6639", + "issue:7180", + "issue:7421", + "issue:7466", + "issue:7864", + "issue:8274", + "issue:8494", + "issue:8805", + "issue:8898", + "issue:8939", + "issue:9091", + "issue:9767" + ], + "soft_pairs": [ + "issue:10066|issue:1407", + "issue:3808|issue:5455", + "issue:2645|issue:7180", + "issue:5799|issue:8494", + "issue:11448|issue:12195", + "issue:2321|issue:3803", + "issue:4073|issue:455", + "issue:3296|issue:4911", + "issue:6598|issue:6639", + "issue:7864|issue:8274", + "issue:3978|issue:7421", + "issue:11839|issue:12009", + "issue:3346|issue:5412", + "issue:4985|issue:6238", + "issue:4389|issue:6602", + "issue:7466|issue:9091", + "issue:1572|issue:3182", + "issue:1976|issue:8939", + "issue:2554|issue:5023", + "issue:6176|issue:9767", + "issue:6287|issue:8805", + "issue:1100|issue:6351", + "issue:12574|issue:13386", + "issue:4217|issue:8898" + ], + "prepared_review_unit_hash": "5b82ffdcc4929ae8d5a082057f2fedbea2e8feebad69206aa9dabdb5a25f479a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34763, + "estimated_input_tokens": 8691, + "estimated_eval_tokens": 17638 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated bug reports across different pipelines, training scripts, and runtime environments. The soft pairs share only generic error wording or the same broad area, not the same underlying defect, so none should be merged as duplicates.", + "confidence": 0.87, + "canonical_issue_reason": "issue:10066 is the most concrete, narrowly scoped bug report in the set, centered on a specific self.attn unpacking failure; it is easier to represent than broader or more environment-specific reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10066 is a strong issue candidate because it names a precise runtime error and code path, making it highly actionable and representative of a real defect report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10066", + "right": "issue:1407", + "accept": false, + "reason": "Both mention ValueError-like failures, but one is a self.attn unpacking bug and the other is an ONNX pipeline feature_extractor regression; different code paths and causes." + }, + { + "left": "issue:3808", + "right": "issue:5455", + "accept": false, + "reason": "Both are Dreambooth/IF training failures, but one is about DeepFloyd IF stage 2 training support and the other is a subprocess error in a notebook; not the same bug." + }, + { + "left": "issue:2645", + "right": "issue:7180", + "accept": false, + "reason": "Both involve platform/offload issues, but one is a Mac M1 CUDA assertion and the other is an SDXL ControlNet cpu offload failure; different environments and failure modes." + }, + { + "left": "issue:5799", + "right": "issue:8494", + "accept": false, + "reason": "The titles both mention argument handling, but one is an inpainting strength keyword issue and the other is an SD3 force_hooks argument error; unrelated APIs." + }, + { + "left": "issue:11448", + "right": "issue:12195", + "accept": false, + "reason": "One is a missing fusing/stable-unclip-2-1-h-img2img model path, the other is a PyTorch 2.3.1 custom_op compatibility break; completely different problems." + }, + { + "left": "issue:2321", + "right": "issue:3803", + "accept": false, + "reason": "One concerns inference model copying and ema_weights, while the other is multi-GPU LoRA training model sharing; same training area but not the same defect." + }, + { + "left": "issue:4073", + "right": "issue:455", + "accept": false, + "reason": "Both are tensor/runtime errors on non-CUDA backends, but one is a DeepFloyd safety checker astype issue and the other is an MPS aten::index.Tensor NotImplementedError; distinct code paths." + }, + { + "left": "issue:3296", + "right": "issue:4911", + "accept": false, + "reason": "Both are loading/saving or custom encoder related, but one is dreambooth checkpoint saving with text_encoder and the other is custom text_encoder loading in from_single_file; not clearly the same bug." + }, + { + "left": "issue:6598", + "right": "issue:6639", + "accept": false, + "reason": "Both discuss output differences, but one compares img2img results versus A1111 and the other compares two LoRA loading methods in SDXL; different reproduction paths." + }, + { + "left": "issue:7864", + "right": "issue:8274", + "accept": false, + "reason": "Both are video-generation research items, but they describe different methods and no shared bug or implementation issue." + }, + { + "left": "issue:3978", + "right": "issue:7421", + "accept": false, + "reason": "Both are determinism/platform-adjacent pipeline issues, but one is MPS img2img failure and the other is seed handling in KDiffusion; not the same bug." + }, + { + "left": "issue:11839", + "right": "issue:12009", + "accept": false, + "reason": "Both involve loading newer model formats, but one is FluxKontextPipeline GGUF invalid input size and the other is Wan2.2 GGUF meta tensor copying failure; different loaders and errors." + }, + { + "left": "issue:3346", + "right": "issue:5412", + "accept": false, + "reason": "Both are checkpoint/resume training issues in dreambooth_lora scripts, but one requests resume_from_checkpoint support and the other reports a specific SDXL resume failure; not enough to treat as one defect." + }, + { + "left": "issue:4985", + "right": "issue:6238", + "accept": false, + "reason": "Both are SDXL training/runtime issues, but one is an InstructPix2Pix OOM and the other is an SDXL VAE problem; unrelated failure modes." + }, + { + "left": "issue:4389", + "right": "issue:6602", + "accept": false, + "reason": "Both involve model loading errors, but one is ControlNetModel loading and the other is loading after fuse_lora/save_pretrained; separate code paths." + }, + { + "left": "issue:7466", + "right": "issue:9091", + "accept": false, + "reason": "Both mention torch.compile, but one is multi-ControlNet SDXL inpaint compilation failure and the other is VAE decoder compile support; not the same concrete bug." + }, + { + "left": "issue:1572", + "right": "issue:3182", + "accept": false, + "reason": "Both are pipeline feature issues, but one is stable_diffusion_mega img2img/inpainting behavior and the other is CPU offload with a custom LPW pipeline; different causes." + }, + { + "left": "issue:1976", + "right": "issue:8939", + "accept": false, + "reason": "Both are tokenizer/file path loading errors, but one is a missing tokenizer/config.json path for stable-diffusion-v1-5 and the other is SD3 tokenizer loading for a different model family." + }, + { + "left": "issue:2554", + "right": "issue:5023", + "accept": false, + "reason": "One is accelerate version incompatibility for StableDiffusionPipeline; the other is a WuerstchenCombinedPipeline import error in a later version. Same broad release breakage theme, but not the same issue." + }, + { + "left": "issue:6176", + "right": "issue:9767", + "accept": false, + "reason": "Both concern LoRA-related training/loading, but one is loading LoRA weights after training dreambooth_lora and the other is a FluxImg2ImgPipeLine cuDNN frontend error; different subsystems." + }, + { + "left": "issue:6287", + "right": "issue:8805", + "accept": false, + "reason": "One is about unused setup.py dependencies, the other about a converter script no longer working; no shared bug." + }, + { + "left": "issue:1100", + "right": "issue:6351", + "accept": false, + "reason": "Both mention attribute errors, but one is a CLIPTextTransformer.device exception and the other is DDP missing disable_adapters during LCM distillation; unrelated." + }, + { + "left": "issue:12574", + "right": "issue:13386", + "accept": false, + "reason": "Both are video/training bugs, but one is WanVideoToVideoPipeline control-argument handling and the other is Dreambooth LoRA Qwen image concatenation; different code paths." + }, + { + "left": "issue:4217", + "right": "issue:8898", + "accept": false, + "reason": "Both are loading/performance issues, but one is slow safetensors loading and the other is from_pretrained subfolder path validation; not the same defect." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 152, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10705", + "issue:10782", + "issue:11436", + "issue:11651", + "issue:11922", + "issue:1543", + "issue:1925", + "issue:2554", + "issue:2947", + "issue:3811", + "issue:3910", + "issue:4257", + "issue:4450", + "issue:4488", + "issue:4675", + "issue:4840", + "issue:5502", + "issue:5504", + "issue:5529", + "issue:5886", + "issue:5956", + "issue:6008", + "issue:6333", + "issue:6336", + "issue:6382", + "issue:6407", + "issue:6793", + "issue:6911", + "issue:7146", + "issue:7168", + "issue:7311", + "issue:7441", + "issue:7483", + "issue:7848", + "issue:785", + "issue:8414", + "issue:8504", + "issue:8511", + "issue:8555", + "issue:8874", + "issue:9136", + "issue:9171", + "issue:9187", + "issue:9488", + "issue:9500", + "issue:9726", + "issue:9914" + ], + "soft_pairs": [ + "issue:6407|issue:8414", + "issue:6793|issue:7483", + "issue:10705|issue:11436", + "issue:5886|issue:7168", + "issue:6333|issue:9914", + "issue:11651|issue:11922", + "issue:9500|issue:9726", + "issue:6911|issue:7441", + "issue:8504|issue:8555", + "issue:5529|issue:7311", + "issue:5956|issue:7146", + "issue:6336|issue:6382", + "issue:3910|issue:4488", + "issue:8874|issue:9171", + "issue:1925|issue:5504", + "issue:4840|issue:7848", + "issue:2554|issue:785", + "issue:10782|issue:9488", + "issue:2947|issue:3811", + "issue:4257|issue:4675", + "issue:4450|issue:8511", + "issue:6008|issue:9187", + "issue:6382|issue:9136", + "issue:1543|issue:5502" + ], + "prepared_review_unit_hash": "ac9dec861ad640322f267a9c54497146d86136081806a9bb4347eb9ec3205c12", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33400, + "estimated_input_tokens": 8350, + "estimated_eval_tokens": 16956 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated diffusion/modeling issues; the soft pairs share only broad subsystem similarity (IP-Adapter, SDXL, LoRA, FLUX, training scripts) without the same concrete bug or fix path. No duplicate-worthy issue/PR cluster emerges.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:9500 is the strongest standalone report here: detailed, open, and tied to a concrete Flux checkpoint-saving failure. It is not, however, clearly a duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6407", + "right": "issue:8414", + "accept": false, + "reason": "Different products and problem domains: AnyText text generation/editing vs a new text-to-audio consistency model. Only superficial AI-model similarity." + }, + { + "left": "issue:6793", + "right": "issue:7483", + "accept": false, + "reason": "One is Hydra serialization/configuration, the other is LoRA conversion for SDXL base models. No shared bug path." + }, + { + "left": "issue:10705", + "right": "issue:11436", + "accept": false, + "reason": "Pipeline reuse slowdown vs a dtype mismatch in Hidream inference. Different failure modes and code paths." + }, + { + "left": "issue:5886", + "right": "issue:7168", + "accept": false, + "reason": "Both mention IP-Adapter, but one requests face-model support and the other asks about tensor shape/embeddings. Not the same bug." + }, + { + "left": "issue:6333", + "right": "issue:9914", + "accept": false, + "reason": "Different LoRA training/loading contexts: PEFT pre-trained LoRA finetuning vs Xlabs LoRA load error. Similar theme only." + }, + { + "left": "issue:11651", + "right": "issue:11922", + "accept": false, + "reason": "Flux Control LoRA loading problem vs a FLUX.1-Fill-dev runtime issue. Related model family, but not the same concrete defect." + }, + { + "left": "issue:9500", + "right": "issue:9726", + "accept": false, + "reason": "Both are Flux training issues, but one is checkpoint saving and the other is multigpu Dreambooth training. Not one underlying bug." + }, + { + "left": "issue:6911", + "right": "issue:7441", + "accept": false, + "reason": "Class-label handling in a 3D UNet vs pytest failing on Python 3.12. Completely unrelated." + }, + { + "left": "issue:8504", + "right": "issue:8555", + "accept": false, + "reason": "Feature request to choose SD3 text encoders vs FP8/text-encoder training availability. Overlapping area, but not the same issue." + }, + { + "left": "issue:5529", + "right": "issue:7311", + "accept": false, + "reason": "Bug in SDXL text-to-image training vs checkpoint save failure. Same script family, different concrete problems." + }, + { + "left": "issue:5956", + "right": "issue:7146", + "accept": false, + "reason": "SDXL fine-tuning performance degradation vs CUDA OOM during SDXL fine-tuning. One is quality/perf, the other is memory exhaustion." + }, + { + "left": "issue:6336", + "right": "issue:6382", + "accept": false, + "reason": "TensorRT VAE compilation with AutoencoderTiny vs an IP-Adapter key error during SDXL LCM-LoRA training/inference. Not the same code path." + }, + { + "left": "issue:3910", + "right": "issue:4488", + "accept": false, + "reason": "General LoRA loading failure vs SDXL checkpoint loading requiring pipeline_class. Different loading mechanisms and symptoms." + }, + { + "left": "issue:8874", + "right": "issue:9171", + "accept": false, + "reason": "General CKPT from_single_file unpickling failure vs a regression for SD 1.5 v_prediction handling. Different root causes." + }, + { + "left": "issue:1925", + "right": "issue:5504", + "accept": false, + "reason": "Half-precision inpainting issue vs PEFT LoRA with xformers. Same precision/LoRA ecosystem, but not the same bug." + }, + { + "left": "issue:4840", + "right": "issue:7848", + "accept": false, + "reason": "Flax autoencoder checkpoint conversion bug vs community SDXL decoder issue. Both involve decoding/loading, but distinct artifacts and failures." + }, + { + "left": "issue:2554", + "right": "issue:785", + "accept": false, + "reason": "Package/version requirement problem vs scheduler performance cost. No duplicate relationship." + }, + { + "left": "issue:10782", + "right": "issue:9488", + "accept": false, + "reason": "Different models and symptoms: Lumina Image 2.0 minor issue vs Lumina pipeline failing to generate any image. Too vague to merge." + }, + { + "left": "issue:2947", + "right": "issue:3811", + "accept": false, + "reason": "Missing UNet load_attn_procs attribute vs LoRA loaded but having no effect. Related to attention/LoRA, but not the same defect." + }, + { + "left": "issue:4257", + "right": "issue:4675", + "accept": false, + "reason": "Reference-only/outpainting support request for SDXL vs ControlNet reference-only for SD XL. Similar feature area, but not the same specific request." + }, + { + "left": "issue:4450", + "right": "issue:8511", + "accept": false, + "reason": "Multi text inversion example not working vs SD3 scripts not unloading text encoders. Different scripts and bugs." + }, + { + "left": "issue:6008", + "right": "issue:9187", + "accept": false, + "reason": "IP-Adapter plus face failing with LCM LoRA vs IP Adapter broken when image_encoder_folder=None after Kolors update. Both IP-Adapter-related, but distinct failure conditions." + }, + { + "left": "issue:6382", + "right": "issue:9136", + "accept": false, + "reason": "IP-Adapter key error in SDXL training/inference vs quality issues at some resolutions. Different symptom class and likely different root cause." + }, + { + "left": "issue:1543", + "right": "issue:5502", + "accept": false, + "reason": "textual_inversion fp16 revision issue vs custom diffusion training not supporting fp16. Same precision keyword, but different scripts and fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 153, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:1013", + "issue:10334", + "issue:10392", + "issue:10520", + "issue:10671", + "issue:1076", + "issue:10839", + "issue:10954", + "issue:1188", + "issue:1214", + "issue:12157", + "issue:12547", + "issue:1310", + "issue:1500", + "issue:1540", + "issue:1755", + "issue:1791", + "issue:1919", + "issue:2645", + "issue:2695", + "issue:3382", + "issue:3574", + "issue:3967", + "issue:4067", + "issue:4305", + "issue:4480", + "issue:4788", + "issue:4837", + "issue:5502", + "issue:5617", + "issue:5963", + "issue:6444", + "issue:6524", + "issue:6899", + "issue:6956", + "issue:7000", + "issue:7057", + "issue:7146", + "issue:7168", + "issue:7421", + "issue:7485", + "issue:7950", + "issue:8555", + "issue:8613", + "issue:8668", + "issue:8834", + "issue:9450" + ], + "soft_pairs": [ + "issue:2645|issue:5617", + "issue:10671|issue:8613", + "issue:6524|issue:7485", + "issue:7168|issue:9450", + "issue:1500|issue:4305", + "issue:1791|issue:4480", + "issue:10026|issue:3382", + "issue:12547|issue:7146", + "issue:2695|issue:6444", + "issue:4788|issue:6956", + "issue:10334|issue:10520", + "issue:1013|issue:3574", + "issue:1076|issue:1310", + "issue:3967|issue:7421", + "issue:4067|issue:7057", + "issue:4837|issue:6899", + "issue:5963|issue:8668", + "issue:8555|issue:8834", + "issue:10839|issue:12157", + "issue:1214|issue:5502", + "issue:7000|issue:7950", + "issue:1540|issue:1919", + "issue:10392|issue:10954", + "issue:1188|issue:1755" + ], + "prepared_review_unit_hash": "dbf72a4f58b11ad67fa4f42deb9b8de1ea2b8b3424b7c393e40080a31d3659f7", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34190, + "estimated_input_tokens": 8548, + "estimated_eval_tokens": 17352 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most items are distinct model-/pipeline-specific bugs, setup issues, or feature requests. None of the soft pairs look like the same underlying defect, so I rejected them all.", + "confidence": 0.78, + "canonical_issue_reason": "issue:10671 is the best umbrella issue here because it is the most general runtime device/dtype bug and is closest in theme to several other device-mismatch reports, even though the cluster is not actually duplicate-heavy.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10671 is the strongest representative for the cluster\u2019s loose device/dtype/runtime-error theme; it is still not a true duplicate target for the rest, but it is the clearest canonical issue candidate among this set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2645", + "right": "issue:5617", + "accept": false, + "reason": "Both involve ControlNet, but one is a Mac/M1 CUDA-support failure and the other is a torch.compile/save_pretrained loading issue; different code paths and failure modes." + }, + { + "left": "issue:10671", + "right": "issue:8613", + "accept": false, + "reason": "One is a deterministic dtype/device access bug in DiffusionPipeline; the other is an OOM during caption_column training. Different bug class." + }, + { + "left": "issue:6524", + "right": "issue:7485", + "accept": false, + "reason": "Loading LoRA a second time is unrelated to setting safety_checker to None and hitting a callable error. Not the same defect." + }, + { + "left": "issue:7168", + "right": "issue:9450", + "accept": false, + "reason": "IP-Adapter embed shape handling and Flux multi-GPU transformer device placement are different underlying problems." + }, + { + "left": "issue:1500", + "right": "issue:4305", + "accept": false, + "reason": "A community pipeline being incompatible with a newer diffusers version is not the same as a torch.compile graph break." + }, + { + "left": "issue:1791", + "right": "issue:4480", + "accept": false, + "reason": "Both are Mac/MPS-related, but one is unsupported pipeline/backend behavior and the other is a float16 kernel crash in StableDiffusionPipeline." + }, + { + "left": "issue:10026", + "right": "issue:3382", + "accept": false, + "reason": "Both mention multi-GPU memory, but the first is Flux ControlNet + DeepSpeed Stage-3 not reducing memory and the second is a text-to-image training OOM; not the same bug." + }, + { + "left": "issue:12547", + "right": "issue:7146", + "accept": false, + "reason": "Missing --instance_prompt in Dreambooth Flux Kontext I2I is an argument/CLI issue, while the SDXL item is an OOM during training." + }, + { + "left": "issue:2695", + "right": "issue:6444", + "accept": false, + "reason": "TensorBoard validation/report_to failure and missing peft_config during Dreambooth LoRA checkpointing are unrelated training bugs." + }, + { + "left": "issue:4788", + "right": "issue:6956", + "accept": false, + "reason": "Both are OOMs, but one is SDXL inpaint on low VRAM and the other is Stable Video Diffusion on SageMaker; different pipelines and constraints." + }, + { + "left": "issue:10334", + "right": "issue:10520", + "accept": false, + "reason": "Both are Sana-related, but gray/NaN outputs on MacOS MPS/CPU are not the same as an explicit cuda:0 vs cpu device mismatch." + }, + { + "left": "issue:1013", + "right": "issue:3574", + "accept": false, + "reason": "StableDiffusionMegaPipeline on Diffusers 6.0 and from_ckpt on Windows are unrelated compatibility issues." + }, + { + "left": "issue:1076", + "right": "issue:1310", + "accept": false, + "reason": "A generic Stable Diffusion question is not the same as a K-Diffusion + Stable Diffusion pipeline request." + }, + { + "left": "issue:3967", + "right": "issue:7421", + "accept": false, + "reason": "A deleted SD-XL branch issue and a seed reproducibility bug in StableDiffusionXLKDiffusionPipeline are different problems." + }, + { + "left": "issue:4067", + "right": "issue:7057", + "accept": false, + "reason": "LoRA scale inconsistency with prompt embeddings is unrelated to a checkpoint/model loading failure." + }, + { + "left": "issue:4837", + "right": "issue:6899", + "accept": false, + "reason": "Both mention the SD conversion script, but the titles do not point to the same concrete conversion failure; too little overlap to merge." + }, + { + "left": "issue:5963", + "right": "issue:8668", + "accept": false, + "reason": "Kandinsky 3.0 embed handling and SD3 img2img input-size alignment are different models and different validation paths." + }, + { + "left": "issue:8555", + "right": "issue:8834", + "accept": false, + "reason": "These are separate feature requests: FP8/text-encoder training vs SD3 ControlNet training code release." + }, + { + "left": "issue:10839", + "right": "issue:12157", + "accept": false, + "reason": "Dreambooth Flux last-step error and Qwen-Image quantization-config device iteration error are unrelated model/training bugs." + }, + { + "left": "issue:1214", + "right": "issue:5502", + "accept": false, + "reason": "Half-precision behavior in CLIPTextModel and lack of fp16 support in train_custom_diffusion.py are not the same bug." + }, + { + "left": "issue:7000", + "right": "issue:7950", + "accept": false, + "reason": "A PyTorch deprecation warning is not the same as a reported incompatibility with PyTorch 2.3.0." + }, + { + "left": "issue:1540", + "right": "issue:1919", + "accept": false, + "reason": "Missing MODEL_TYPE in Dreambooth/transformers and missing datasets in requirements.txt are unrelated setup problems." + }, + { + "left": "issue:10392", + "right": "issue:10954", + "accept": false, + "reason": "Both concern Flux LoRAs, but one is an fp8 training issue with Hyper-FLUX and the other is a ComfyUI-compatible LoRA load failure; not the same defect." + }, + { + "left": "issue:1188", + "right": "issue:1755", + "accept": false, + "reason": "A CLIPSeg + Stable Diffusion pipeline request is unrelated to a generic general-links issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 154, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1062", + "issue:13300", + "issue:1961", + "issue:2221", + "issue:2420", + "issue:2991", + "issue:3595", + "issue:3748", + "issue:3811", + "issue:3958", + "issue:4030", + "issue:4080", + "issue:4086", + "issue:4186", + "issue:4236", + "issue:4584", + "issue:4911", + "issue:5014", + "issue:556", + "issue:5593", + "issue:5606", + "issue:5678", + "issue:5873", + "issue:5890", + "issue:5963", + "issue:5972", + "issue:6351", + "issue:6382", + "issue:6790", + "issue:6794", + "issue:6832", + "issue:6865", + "issue:6928", + "issue:7082", + "issue:7286", + "issue:7322", + "issue:7492", + "issue:7801", + "issue:7962", + "issue:8604", + "issue:8626", + "issue:8880", + "issue:8975", + "issue:9178", + "issue:9393", + "issue:9420", + "issue:9515" + ], + "soft_pairs": [ + "issue:1961|issue:2221", + "issue:2991|issue:6790", + "issue:3811|issue:5606", + "issue:4080|issue:556", + "issue:7801|issue:9393", + "issue:5593|issue:5963", + "issue:4030|issue:4186", + "issue:6928|issue:7962", + "issue:9178|issue:9420", + "issue:4584|issue:7322", + "issue:6351|issue:6382", + "issue:3748|issue:3958", + "issue:7082|issue:8975", + "issue:2420|issue:5678", + "issue:5873|issue:9515", + "issue:6832|issue:8626", + "issue:3595|issue:4086", + "issue:5890|issue:8880", + "issue:4086|issue:5972", + "issue:1062|issue:5014", + "issue:4911|issue:7286", + "issue:4236|issue:8604", + "issue:7492|issue:7801", + "issue:6794|issue:6865", + "issue:13300|issue:5963" + ], + "prepared_review_unit_hash": "db6a810b63f062b632cc73d55b64bbac131ee1b76fb83c646b714b5195907a58", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 34274, + "estimated_input_tokens": 8569, + "estimated_eval_tokens": 17394 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are unrelated example/pipeline bugs, and the proposed soft links are only surface-level similarity. None of the candidate pairs look like the same concrete bug/change, so I would not merge them. If a representative issue is needed, the broadest user-facing one is the LoRA loading bug.", + "confidence": 0.62, + "canonical_issue_reason": "Issue 3811 is the broadest, most central bug report here: it describes LoRA loading having no effect, which is the closest recurring theme across the LoRA-related tickets in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 3811 is the best representative because it is concrete, user-facing, and general enough to cover the nearby LoRA/loading regressions better than the very specific or unrelated example issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1961", + "right": "issue:2221", + "accept": false, + "reason": "Same vague title, but no evidence they describe the same underlying colossalai failure." + }, + { + "left": "issue:2991", + "right": "issue:6790", + "accept": false, + "reason": "Different problems: mixed-precision ControlNet example vs LCM checkpoint key mismatch." + }, + { + "left": "issue:3811", + "right": "issue:5606", + "accept": false, + "reason": "Both involve LoRA loading, but one is no-effect generation and the other is missing alpha handling; likely distinct bugs." + }, + { + "left": "issue:4080", + "right": "issue:556", + "accept": false, + "reason": "Different features entirely: SDXL inpainting vs seamless texture tiling." + }, + { + "left": "issue:7801", + "right": "issue:9393", + "accept": false, + "reason": "Generic FP16 mismatch complaint vs a FLUX Dreambooth LoRA save-model issue; not the same bug." + }, + { + "left": "issue:5593", + "right": "issue:5963", + "accept": false, + "reason": "Both pipeline-related, but one is AutoPipeline conversion and the other is passing embeds to Kandinsky 3.0." + }, + { + "left": "issue:4030", + "right": "issue:4186", + "accept": false, + "reason": "Different SD/v2.1 embedding runtime error vs SDXL inpaint/refiner behavior." + }, + { + "left": "issue:6928", + "right": "issue:7962", + "accept": false, + "reason": "One is a documentation/option request for Marigold, the other is an API enhancement for instruct-pix2pix." + }, + { + "left": "issue:9178", + "right": "issue:9420", + "accept": false, + "reason": "FLUX LoRA PEFT target-module training error is unrelated to a saved transformer failing to load." + }, + { + "left": "issue:4584", + "right": "issue:7322", + "accept": false, + "reason": "Both are training-script errors, but one is resume-from-checkpoint and the other is SIGKILL during SDXL DreamBooth LoRA training." + }, + { + "left": "issue:6351", + "right": "issue:6382", + "accept": false, + "reason": "DDP missing disable_adapters and IP-adapter key errors are different failure modes in different code paths." + }, + { + "left": "issue:3748", + "right": "issue:3958", + "accept": false, + "reason": "Float16/float32 mismatch on LoRA weights is not the same as the enable_model_cpu_offload interaction bug." + }, + { + "left": "issue:7082", + "right": "issue:8975", + "accept": false, + "reason": "Separate import failures for different symbols from different modules." + }, + { + "left": "issue:2420", + "right": "issue:5678", + "accept": false, + "reason": "LoRA xFormers attention failure is unrelated to higher-order-derivative attention/autograd behavior." + }, + { + "left": "issue:5873", + "right": "issue:9515", + "accept": false, + "reason": "ONNX export support problem is not the same as missing RMSNorm in JointAttnProcessor2_0." + }, + { + "left": "issue:6832", + "right": "issue:8626", + "accept": false, + "reason": "A tuple-shape AttributeError in IP-Adapter Face ID is unrelated to guidance documentation for multiple IP adapter images/masks." + }, + { + "left": "issue:3595", + "right": "issue:4086", + "accept": false, + "reason": "from_ckpt loading failure and SDXL ControlNet training failure are distinct problems." + }, + { + "left": "issue:5890", + "right": "issue:8880", + "accept": false, + "reason": "Token length warning during training inference is unrelated to a broader inference speed regression." + }, + { + "left": "issue:4086", + "right": "issue:5972", + "accept": false, + "reason": "Training failure with SDXL ControlNet is not the same as a documentation update for Min-SNR weighting." + }, + { + "left": "issue:1062", + "right": "issue:5014", + "accept": false, + "reason": "Dreambooth paper-result reproducibility and LoRA training results are related only by topic, not by the same bug." + }, + { + "left": "issue:4911", + "right": "issue:7286", + "accept": false, + "reason": "Custom text-encoder loading from_single_file is unrelated to cross-attention weight loading in Custom Diffusion." + }, + { + "left": "issue:4236", + "right": "issue:8604", + "accept": false, + "reason": "SDXL refiner text-encoder registration and T5 float16 corruption are different code paths and symptoms." + }, + { + "left": "issue:7492", + "right": "issue:7801", + "accept": false, + "reason": "A missing repository URL/model lookup error is unrelated to mixed_precision not producing fp16 weights." + }, + { + "left": "issue:6794", + "right": "issue:6865", + "accept": false, + "reason": "Import error for cast_training_params and missing Image-instance check are separate utility issues." + }, + { + "left": "issue:13300", + "right": "issue:5963", + "accept": false, + "reason": "CPU generation failures in SD3.5/FLUX are unrelated to Kandinsky 3.0 embed-input handling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 155, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10025", + "issue:10066", + "issue:10311", + "issue:1197", + "issue:12609", + "issue:13258", + "issue:1985", + "issue:2324", + "issue:2599", + "issue:378", + "issue:4037", + "issue:4086", + "issue:4236", + "issue:4337", + "issue:435", + "issue:4405", + "issue:4933", + "issue:4964", + "issue:5227", + "issue:5433", + "issue:5594", + "issue:5737", + "issue:5932", + "issue:5953", + "issue:5966", + "issue:5967", + "issue:6093", + "issue:6258", + "issue:6382", + "issue:6388", + "issue:6441", + "issue:6790", + "issue:6794", + "issue:7057", + "issue:7420", + "issue:7466", + "issue:7483", + "issue:7882", + "issue:7985", + "issue:8547", + "issue:8555", + "issue:8645", + "issue:8668", + "issue:8673", + "issue:8879", + "issue:8887", + "issue:8906", + "issue:9101" + ], + "soft_pairs": [ + "issue:378|issue:435", + "issue:5953|issue:8668", + "issue:4337|issue:7985", + "issue:12609|issue:13258", + "issue:5433|issue:5737", + "issue:4964|issue:7483", + "issue:6382|issue:6790", + "issue:10066|issue:5932", + "issue:10025|issue:8673", + "issue:10311|issue:6794", + "issue:1197|issue:1985", + "issue:2324|issue:2599", + "issue:4037|issue:4086", + "issue:4405|issue:8887", + "issue:5227|issue:7466", + "issue:5594|issue:6388", + "issue:5966|issue:7882", + "issue:5967|issue:6258", + "issue:6093|issue:8547", + "issue:6441|issue:7057", + "issue:7420|issue:8879", + "issue:8555|issue:9101", + "issue:8645|issue:8906", + "issue:4236|issue:4933" + ], + "prepared_review_unit_hash": "5b8877ed2fd74c3fb58a1a3fcd54a50a4f90457f1742040f6d407a74d85bf89f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33807, + "estimated_input_tokens": 8452, + "estimated_eval_tokens": 17160 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft pair looks like the same underlying bug/change; the set is mostly unrelated training, conversion, and compile issues. No pull requests are present. The most representative issue is the SD3 attention-mask bug because it is concrete and actionable.", + "confidence": 0.78, + "canonical_issue_reason": "issue:8673 is the clearest concrete bug report here: missing attention masks in SD3 for padding tokens, with a specific model/component and a plausible targeted fix.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8673 is the strongest standalone issue for triage: specific, reproducible, and directly tied to a code-path bug rather than a broad feature request or vague failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:378", + "right": "issue:435", + "accept": false, + "reason": "Both are textual inversion/VAE-related, but they describe different failures and different symptoms; not the same underlying bug." + }, + { + "left": "issue:5953", + "right": "issue:8668", + "accept": false, + "reason": "Both involve image-size handling, but one is ControlNet img2img consistency and the other is SD3 alignment to 16; different code paths and fixes." + }, + { + "left": "issue:4337", + "right": "issue:7985", + "accept": false, + "reason": "Both mention torch.compile, but one is Community ControlNetReference and the other IPAdapter; too different to merge as one fix." + }, + { + "left": "issue:12609", + "right": "issue:13258", + "accept": false, + "reason": "Wan 2.2 latent NaNs versus a missing image-to-video model; unrelated problems." + }, + { + "left": "issue:5433", + "right": "issue:5737", + "accept": false, + "reason": "SDXL artifacting with schedulers versus SDXL conversion script failure; not the same bug." + }, + { + "left": "issue:4964", + "right": "issue:7483", + "accept": false, + "reason": "Both are conversion-related, but one is general SDXL conversion and the other is LoRA-to-Diffusers for SDXL base models; different workflows." + }, + { + "left": "issue:6382", + "right": "issue:6790", + "accept": false, + "reason": "Both concern LCM/SDXL training, but the errors are different state-dict/key problems in different scripts." + }, + { + "left": "issue:10066", + "right": "issue:5932", + "accept": false, + "reason": "Different runtime failures: unpacking in self.attn versus channel-mismatch in a convolution input." + }, + { + "left": "issue:10025", + "right": "issue:8673", + "accept": false, + "reason": "Both mention attention masks, but Flux transformer versus SD3 padding-token masking are different model-specific bugs." + }, + { + "left": "issue:10311", + "right": "issue:6794", + "accept": false, + "reason": "Both are import errors, but they point to different symbols and scripts; not a single underlying defect." + }, + { + "left": "issue:1197", + "right": "issue:1985", + "accept": false, + "reason": "A community pipeline request versus a request for easier embedding loading; unrelated feature work." + }, + { + "left": "issue:2324", + "right": "issue:2599", + "accept": false, + "reason": "Both are Dreambooth training issues, but one is instance data loading and the other is multi-subject quality degradation." + }, + { + "left": "issue:4037", + "right": "issue:4086", + "accept": false, + "reason": "Both are ControlNet training problems, but one is FSDP validation errors and the other is a general SDXL base-0.9 training failure." + }, + { + "left": "issue:4405", + "right": "issue:8887", + "accept": false, + "reason": "Different features and models: SDXL textual inversion script versus SD3 VAE encode bug." + }, + { + "left": "issue:5227", + "right": "issue:7466", + "accept": false, + "reason": "Both involve torch.compile, but one is LoRA loading and the other is multi-ControlNet inpainting; different code paths." + }, + { + "left": "issue:5594", + "right": "issue:6388", + "accept": false, + "reason": "Related SDXL-LoRA optimization topics, but one is memory usage generally and the other is gradient checkpointing being ignored." + }, + { + "left": "issue:5966", + "right": "issue:7882", + "accept": false, + "reason": "ControlNet for SD2.1 versus T2I-Adapter missing from_single_file; different features and implementations." + }, + { + "left": "issue:5967", + "right": "issue:6258", + "accept": false, + "reason": "An announcement versus a pipeline quality bug; no shared underlying defect." + }, + { + "left": "issue:6093", + "right": "issue:8547", + "accept": false, + "reason": "SDXL LoRA push_to_hub behavior versus an LCM scaling bug; unrelated." + }, + { + "left": "issue:6441", + "right": "issue:7057", + "accept": false, + "reason": "Both are model-loading problems, but one is from_pretrained in Colab and the other is checkpoint loading; different causes." + }, + { + "left": "issue:7420", + "right": "issue:8879", + "accept": false, + "reason": "Both mention torch.compile, but the failure modes are different and not clearly one mergeable fix." + }, + { + "left": "issue:8555", + "right": "issue:9101", + "accept": false, + "reason": "SD3 FP8/text encoder training request versus an InstructPix2Pix training script request; unrelated feature requests." + }, + { + "left": "issue:8645", + "right": "issue:8906", + "accept": false, + "reason": "Missing helper function in single_file_utils versus missing qk_norm in SD3Transformer2DModel; unrelated implementation gaps." + }, + { + "left": "issue:4236", + "right": "issue:4933", + "accept": false, + "reason": "SDXL refiner text-encoder registration versus Dreambooth output-file saving; different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 156, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10017", + "issue:10180", + "issue:10381", + "issue:10839", + "issue:1896", + "issue:2218", + "issue:2272", + "issue:2312", + "issue:2891", + "issue:3235", + "issue:3348", + "issue:3420", + "issue:3432", + "issue:3481", + "issue:3494", + "issue:3651", + "issue:4306", + "issue:4446", + "issue:4619", + "issue:4631", + "issue:4634", + "issue:490", + "issue:5399", + "issue:5433", + "issue:5510", + "issue:5546", + "issue:5846", + "issue:5956", + "issue:6164", + "issue:6224", + "issue:6536", + "issue:6540", + "issue:6996", + "issue:720", + "issue:7538", + "issue:7629", + "issue:7960", + "issue:7972", + "issue:8514", + "issue:8578", + "issue:8644", + "issue:8874", + "issue:8915", + "issue:9096", + "issue:9108", + "issue:9309", + "issue:9410", + "issue:9732" + ], + "soft_pairs": [ + "issue:5956|issue:6536", + "issue:7629|issue:9309", + "issue:10017|issue:7538", + "issue:2272|issue:3420", + "issue:10839|issue:9732", + "issue:3481|issue:3494", + "issue:2218|issue:3348", + "issue:8915|issue:9410", + "issue:2891|issue:490", + "issue:2312|issue:4306", + "issue:8578|issue:8644", + "issue:10180|issue:9108", + "issue:4631|issue:720", + "issue:4619|issue:7972", + "issue:10381|issue:9096", + "issue:3235|issue:4634", + "issue:6164|issue:6996", + "issue:6224|issue:8514", + "issue:5433|issue:5546", + "issue:5399|issue:6540", + "issue:1896|issue:5510", + "issue:4446|issue:8874", + "issue:3432|issue:5846", + "issue:3651|issue:7960" + ], + "prepared_review_unit_hash": "f719105a645b609284ca9fe9b132f784c8c39e79d8605eb5c9a52a06c0323a66", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34335, + "estimated_input_tokens": 8584, + "estimated_eval_tokens": 17424 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most items are distinct bug reports that only share broad terms like Flux, LoRA, training, or scheduler. I found no pair that looks like the same underlying issue/change, so all soft edges should be rejected.", + "confidence": 0.91, + "canonical_issue_reason": "No single canonical issue is justified here because the items span separate failures (dataset loading, LoRA/Flux inference, training errors, scheduler bugs, import errors, and memory/offload issues) rather than one duplicate report.", + "canonical_pr_reason": null, + "best_issue_reason": "If a representative issue is needed, issue 5433 is the strongest candidate: it has concrete symptoms, high discussion/inbound activity, and is a well-formed user-facing bug report with clear repro context.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5956", + "right": "issue:6536", + "accept": false, + "reason": "Different problems: SD-XL fine-tuning quality regression vs UNet3DConditionModel OOM." + }, + { + "left": "issue:7629", + "right": "issue:9309", + "accept": false, + "reason": "Unrelated bugs: PyTorch 1.13.1 OOM in SVD vs FluxControlNetPipeline on Flux schnell." + }, + { + "left": "issue:10017", + "right": "issue:7538", + "accept": false, + "reason": "Both mention dataset/training, but one is a load_dataset() issue and the other is a custom Fill50k training issue." + }, + { + "left": "issue:2272", + "right": "issue:3420", + "accept": false, + "reason": "Different code paths: ONNX conversion script failure vs ControlNet inpaint pipeline incompatibility." + }, + { + "left": "issue:10839", + "right": "issue:9732", + "accept": false, + "reason": "Both are Flux training reports, but one is a last-step error in Dreambooth LoRA training and the other is CUDA OOM in the README workflow." + }, + { + "left": "issue:3481", + "right": "issue:3494", + "accept": false, + "reason": "Related to generation/schedulers, but the symptoms and pipelines differ: nondeterminism across steps vs corrupted multidiffusion outputs." + }, + { + "left": "issue:2218", + "right": "issue:3348", + "accept": false, + "reason": "Overtrained-model behavior is not the same as xformers/Lora scale-parameter support." + }, + { + "left": "issue:8915", + "right": "issue:9410", + "accept": false, + "reason": "Both are import failures, but for different classes and symbols in different pipelines." + }, + { + "left": "issue:2891", + "right": "issue:490", + "accept": false, + "reason": "Docs/support mismatch for StableDiffusionUpscalePipeline embeds is unrelated to the memory issue in older Stable Diffusion." + }, + { + "left": "issue:2312", + "right": "issue:4306", + "accept": false, + "reason": "Different failures: EMA save-hook error vs StableDiffusionReferencePipeline list-index error." + }, + { + "left": "issue:8578", + "right": "issue:8644", + "accept": false, + "reason": "Both involve meta tensors/offload, but they are different setups and code paths (sd_turbo 1.5 vs SD3 T5 encoder offload)." + }, + { + "left": "issue:10180", + "right": "issue:9108", + "accept": false, + "reason": "Different Flux/LoRA training-vs-resume problems; one is loading multiple LoRAs, the other resuming with a quantized base model." + }, + { + "left": "issue:4631", + "right": "issue:720", + "accept": false, + "reason": "Similar scalar/Half runtime flavor, but they are different scheduler behaviors and different repros." + }, + { + "left": "issue:4619", + "right": "issue:7972", + "accept": false, + "reason": "Both are type-mismatch training issues, but for different scripts and failure modes." + }, + { + "left": "issue:10381", + "right": "issue:9096", + "accept": false, + "reason": "Both involve Flux inference instability, but one is quantization/LoRA/Quanto-related and the other is torch.half NaNs." + }, + { + "left": "issue:3235", + "right": "issue:4634", + "accept": false, + "reason": "Custom Diffusion loading vs safe-serialization support are different serialization issues." + }, + { + "left": "issue:6164", + "right": "issue:6996", + "accept": false, + "reason": "Different pipeline bugs: latent output handling vs NoneType added_cond_kwargs membership check." + }, + { + "left": "issue:6224", + "right": "issue:8514", + "accept": false, + "reason": "TorchDynamo fine-tuning issue and EMAModel/InstructP2P parallel fine-tuning error are separate training failures." + }, + { + "left": "issue:5433", + "right": "issue:5546", + "accept": false, + "reason": "Both mention SDXL schedulers, but one is visual artifacts without refiner and the other is wrong step counts with Karras sigmas." + }, + { + "left": "issue:5399", + "right": "issue:6540", + "accept": false, + "reason": "Different dtype/matrix errors in different finetuning/inference paths; not the same bug." + }, + { + "left": "issue:1896", + "right": "issue:5510", + "accept": false, + "reason": "Both are runtime dtype-ish errors, but in unrelated pipelines and with different root causes." + }, + { + "left": "issue:4446", + "right": "issue:8874", + "accept": false, + "reason": "Both concern from_single_file loading, but one is custom text encoder injection and the other is CKPT unpickling failure." + }, + { + "left": "issue:3432", + "right": "issue:5846", + "accept": false, + "reason": "Different problems: CPU half-precision conv failure vs resume_from_checkpoint error." + }, + { + "left": "issue:3651", + "right": "issue:7960", + "accept": false, + "reason": "Both are multi-GPU training failures, but they are distinct models and failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 157, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10030", + "issue:10795", + "issue:10937", + "issue:11232", + "issue:11285", + "issue:11370", + "issue:1166", + "issue:12547", + "issue:1377", + "issue:1704", + "issue:2225", + "issue:2414", + "issue:2420", + "issue:3064", + "issue:3826", + "issue:4448", + "issue:4460", + "issue:4480", + "issue:455", + "issue:4688", + "issue:4954", + "issue:4993", + "issue:5012", + "issue:5128", + "issue:5152", + "issue:528", + "issue:5337", + "issue:5442", + "issue:5617", + "issue:5821", + "issue:5845", + "issue:6363", + "issue:6726", + "issue:6894", + "issue:6996", + "issue:7168", + "issue:7420", + "issue:7480", + "issue:7722", + "issue:7848", + "issue:8624", + "issue:867", + "issue:8738", + "issue:9187", + "issue:9516", + "issue:9660", + "issue:993" + ], + "soft_pairs": [ + "issue:10030|issue:6996", + "issue:1166|issue:4954", + "issue:2420|issue:6363", + "issue:3064|issue:5128", + "issue:5845|issue:9516", + "issue:6894|issue:7848", + "issue:7168|issue:7480", + "issue:1704|issue:528", + "issue:4993|issue:5152", + "issue:5128|issue:7722", + "issue:11370|issue:9187", + "issue:4480|issue:455", + "issue:3826|issue:5442", + "issue:10795|issue:10937", + "issue:11232|issue:12547", + "issue:4688|issue:6726", + "issue:1377|issue:2414", + "issue:2225|issue:5337", + "issue:11285|issue:8624", + "issue:5617|issue:7420", + "issue:5012|issue:5821", + "issue:867|issue:993", + "issue:8738|issue:9660", + "issue:4448|issue:4460" + ], + "prepared_review_unit_hash": "77d4610275e0df1153393365cdec90e3dface97f457c37c3939858236fabc9eb", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33565, + "estimated_input_tokens": 8392, + "estimated_eval_tokens": 17040 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly false-positive topical similarity. The only plausible duplicate pair is the ControlNet/added_cond_kwargs None-handling regression; the rest appear to be different bugs, features, or unrelated pipeline/model issues.", + "confidence": 0.72, + "canonical_issue_reason": "issue:10030 is the clearest, most concrete report of the added_cond_kwargs null regression in ControlNet img2img, and it likely covers the symptom described in 6996.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10030 is the best representative issue: specific pipeline, clear failure mode, and easiest to map to a fix.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10030", + "right": "issue:6996", + "accept": true, + "reason": "Both describe the same underlying bug: code assumes added_cond_kwargs is a dict, but pipeline ControlNet/img2img can pass None, causing a null-membership/type error." + }, + { + "left": "issue:1166", + "right": "issue:4954", + "accept": false, + "reason": "Both are textual inversion issues, but one is TPU save hang and the other is an UnpicklingError; different failure modes." + }, + { + "left": "issue:2420", + "right": "issue:6363", + "accept": false, + "reason": "Different training problems: xformers attention bug vs FP16 gradient unscale error." + }, + { + "left": "issue:3064", + "right": "issue:5128", + "accept": false, + "reason": "Both involve .safetensors loading, but one is LoRA loading generally and the other is XLPipeline.from_single_file; not the same bug." + }, + { + "left": "issue:5845", + "right": "issue:9516", + "accept": false, + "reason": "Unrelated model families and failures: PixArt DreamBooth training vs FLUX joint_attention_kwargs propagation." + }, + { + "left": "issue:6894", + "right": "issue:7848", + "accept": false, + "reason": "Different outputs and code paths: LoRA export compatibility vs lpw SDXL decoding issue." + }, + { + "left": "issue:7168", + "right": "issue:7480", + "accept": false, + "reason": "IP-Adapter tensor-shape issue vs DeepFloyd IF OOM; no shared underlying bug." + }, + { + "left": "issue:1704", + "right": "issue:528", + "accept": false, + "reason": "Both are general questions, not the same topic or bug." + }, + { + "left": "issue:4993", + "right": "issue:5152", + "accept": false, + "reason": "Import-structure bug for CLIPImageProjection vs Wuerstchen alpha bug; unrelated." + }, + { + "left": "issue:5128", + "right": "issue:7722", + "accept": false, + "reason": "Local safetensors loading error vs Animatediff ControlNet/IPAdapter usage problem; different code paths." + }, + { + "left": "issue:11370", + "right": "issue:9187", + "accept": false, + "reason": "Different failures: invalid pickle/load-key in conversion script vs SDXL IP-Adapter image_encoder_folder=None regression." + }, + { + "left": "issue:4480", + "right": "issue:455", + "accept": false, + "reason": "Both are MPS-related, but one is a kernel crash on fp16 and the other is an aten::index.Tensor NotImplementedError." + }, + { + "left": "issue:3826", + "right": "issue:5442", + "accept": false, + "reason": "Both are new pipeline proposals, but they are entirely different pipelines and changes." + }, + { + "left": "issue:10795", + "right": "issue:10937", + "accept": false, + "reason": "Both mention torch.compile, but one is GGUF loading via from_single_file and the other is vae.encode; different concrete problems." + }, + { + "left": "issue:11232", + "right": "issue:12547", + "accept": false, + "reason": "Noise residuals during fine-tuning vs missing required --instance_prompt; unrelated training issues." + }, + { + "left": "issue:4688", + "right": "issue:6726", + "accept": false, + "reason": "Different research/pipeline topics with no shared bug or code path." + }, + { + "left": "issue:1377", + "right": "issue:2414", + "accept": false, + "reason": "One is a missing transformers symbol import, the other is a NameError from not importing transformers; different root causes." + }, + { + "left": "issue:2225", + "right": "issue:5337", + "accept": false, + "reason": "A usage question about safety checker vs a callback explanation question; not duplicates." + }, + { + "left": "issue:11285", + "right": "issue:8624", + "accept": false, + "reason": "Conversion/value errors versus a runtime shared-library symbol error; not the same issue." + }, + { + "left": "issue:5617", + "right": "issue:7420", + "accept": false, + "reason": "Loading a compiled ControlNet pipeline after save_pretrained vs general inability to use torch.compile; different specifics." + }, + { + "left": "issue:5012", + "right": "issue:5821", + "accept": false, + "reason": "LoRA text-encoder training behavior vs a generic CUDA backward failure; unrelated." + }, + { + "left": "issue:867", + "right": "issue:993", + "accept": false, + "reason": "GPU memory leak in FastAPI vs fp16 data type mismatch; different bugs." + }, + { + "left": "issue:8738", + "right": "issue:9660", + "accept": false, + "reason": "Model configuration failure in SD3Transformer2DModel vs from_pretrained not setting eval mode; different behavior." + }, + { + "left": "issue:4448", + "right": "issue:4460", + "accept": false, + "reason": "Both discuss output differences, but one is ControlNet outpainting quality and the other is a broad Stable Diffusion vs diffusers comparison; too vague to merge." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The accepted 10030\u21946996 pair is well supported by the titles, which both point to the same added_cond_kwargs None-handling regression in ControlNet/img2img. The remaining pairs are rejected cautiously without overclaiming duplicate status, and the overall summary appropriately characterizes the rest as likely false-positive topical similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 158, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10194", + "issue:10972", + "issue:11062", + "issue:11145", + "issue:12299", + "issue:12348", + "issue:1398", + "issue:2156", + "issue:2264", + "issue:2331", + "issue:2371", + "issue:2465", + "issue:2480", + "issue:3101", + "issue:3174", + "issue:3826", + "issue:3975", + "issue:4206", + "issue:4306", + "issue:4353", + "issue:4375", + "issue:4389", + "issue:4566", + "issue:4731", + "issue:4852", + "issue:5028", + "issue:5576", + "issue:5799", + "issue:6092", + "issue:6093", + "issue:6258", + "issue:6372", + "issue:7296", + "issue:7311", + "issue:7350", + "issue:7629", + "issue:7716", + "issue:7871", + "issue:7936", + "issue:8493", + "issue:867", + "issue:8900", + "issue:9044", + "issue:9204", + "issue:9501", + "issue:9613", + "issue:9902" + ], + "soft_pairs": [ + "issue:4389|issue:8493", + "issue:2156|issue:2480", + "issue:3826|issue:5576", + "issue:7311|issue:9501", + "issue:7871|issue:8900", + "issue:1398|issue:867", + "issue:3975|issue:7871", + "issue:7296|issue:9902", + "issue:3101|issue:4206", + "issue:2371|issue:2465", + "issue:4566|issue:7350", + "issue:12299|issue:12348", + "issue:5028|issue:5799", + "issue:10194|issue:11145", + "issue:2264|issue:4306", + "issue:10972|issue:11062", + "issue:2331|issue:6372", + "issue:6093|issue:9204", + "issue:4353|issue:6258", + "issue:6092|issue:9044", + "issue:3174|issue:4852", + "issue:7716|issue:7936", + "issue:7629|issue:9613", + "issue:4375|issue:4731" + ], + "prepared_review_unit_hash": "3a0919571159a037211db661eca6a7ceff4ee57570174e4921318d7b479af300", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33133, + "estimated_input_tokens": 8284, + "estimated_eval_tokens": 16824 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is heterogeneous: most items are separate bugs, feature requests, or training/loading problems that only share broad ML/Flux/SDXL vocabulary. I would not collapse any of the soft pairs as duplicates.", + "confidence": 0.88, + "canonical_issue_reason": "issue:10972 is the strongest concrete bug report in the broad Flux/LoRA area, with a specific failing code path (loading OneTrainer Flux LoRAs). It is still not a true cluster representative for the rest, which span unrelated pipelines and failure modes.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10972 has the clearest actionable reproduction and a narrowly defined failure mode, making it the best issue anchor among this mixed set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4389", + "right": "issue:8493", + "accept": false, + "reason": "Different loading failures: ControlNetModel init vs SD3Transformer2DModel state_dict mismatch. Not the same code path." + }, + { + "left": "issue:2156", + "right": "issue:2480", + "accept": false, + "reason": "One is GPU-vs-CPU training execution, the other is checkpoint directory layout/output saving. Separate bugs." + }, + { + "left": "issue:3826", + "right": "issue:5576", + "accept": false, + "reason": "A new pipeline request is not the same as a new scheduler request; different artifact and scope." + }, + { + "left": "issue:7311", + "right": "issue:9501", + "accept": false, + "reason": "Both involve training/save behavior, but one is a checkpoint save failure and the other is delayed model saving in Flux. Not one bug." + }, + { + "left": "issue:7871", + "right": "issue:8900", + "accept": false, + "reason": "Blockwise scale handling in a LoRA file is unrelated to loading an SDXL refiner from_single_file." + }, + { + "left": "issue:1398", + "right": "issue:867", + "accept": false, + "reason": "MPS/Apple Silicon compatibility and FastAPI GPU memory retention are unrelated runtime issues." + }, + { + "left": "issue:3975", + "right": "issue:7871", + "accept": false, + "reason": "A typo in an error message is unrelated to a LoRA/blockwise-scale loading failure." + }, + { + "left": "issue:7296", + "right": "issue:9902", + "accept": false, + "reason": "PEFT dependency pinning and a stuck conda version are distinct packaging/versioning problems." + }, + { + "left": "issue:3101", + "right": "issue:4206", + "accept": false, + "reason": "Both mention ControlNet training, but one is a training example error and the other is a broken SDXL main-branch behavior." + }, + { + "left": "issue:2371", + "right": "issue:2465", + "accept": false, + "reason": "Dreambooth output artifacts missing vs asking for a reproducible recipe are not duplicate bug reports." + }, + { + "left": "issue:4566", + "right": "issue:7350", + "accept": false, + "reason": "Checkpoint resume in SDXL LoRA training is unrelated to initializing ControlNet from a UNet with LoRA." + }, + { + "left": "issue:12299", + "right": "issue:12348", + "accept": false, + "reason": "Both are tensor mismatch reports, but they affect different pipelines and trigger conditions." + }, + { + "left": "issue:5028", + "right": "issue:5799", + "accept": false, + "reason": "Different API/runtime errors in different pipelines: hacked DownBlock2D scale kwarg vs inpainting strength kwarg." + }, + { + "left": "issue:10194", + "right": "issue:11145", + "accept": false, + "reason": "One is a Flux attention-masking feature request; the other is Flux Fill LoRA/DreamBooth support. Related theme, not same change." + }, + { + "left": "issue:2264", + "right": "issue:4306", + "accept": false, + "reason": "InstructPix2Pix seed handling and ReferencePipeline list-index failure are different pipeline bugs." + }, + { + "left": "issue:10972", + "right": "issue:11062", + "accept": false, + "reason": "Both are LoRA loading failures, but for different model families and error modes; not clearly the same defect." + }, + { + "left": "issue:2331", + "right": "issue:6372", + "accept": false, + "reason": "Adding a ControlNet pipeline and SVD+ControlNet support are different feature requests." + }, + { + "left": "issue:6093", + "right": "issue:9204", + "accept": false, + "reason": "Push-to-hub repo creation in one SDXL training script is unrelated to the FLUX.1 DreamBooth training script issue." + }, + { + "left": "issue:4353", + "right": "issue:6258", + "accept": false, + "reason": "Both involve image quality, but one is SDXL generation quality and the other is forward-chunking degradation in SVD." + }, + { + "left": "issue:6092", + "right": "issue:9044", + "accept": false, + "reason": "A LoRA example not training and a missing diffusion_pytorch_model.bin file are unrelated load/train problems." + }, + { + "left": "issue:3174", + "right": "issue:4852", + "accept": false, + "reason": "Both mention LoRA loading, but one is a broader regression from new LoRA features while the other is a standalone load error. Not enough to merge." + }, + { + "left": "issue:7716", + "right": "issue:7936", + "accept": false, + "reason": "Missing config.json in a directory and invalid JSON when loading a safetensors file are different file-format/path errors." + }, + { + "left": "issue:7629", + "right": "issue:9613", + "accept": false, + "reason": "OOM on PyTorch 1.13.1 and a missing scale attribute on Timesteps are unrelated compatibility bugs." + }, + { + "left": "issue:4375", + "right": "issue:4731", + "accept": false, + "reason": "Both involve torch.compile, but one is a general inference crash and the other is multi-ControlNet SDXL compile failure. Different code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 159, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10016", + "issue:10428", + "issue:10470", + "issue:10591", + "issue:10797", + "issue:10928", + "issue:1141", + "issue:1486", + "issue:1552", + "issue:1720", + "issue:1865", + "issue:1922", + "issue:1925", + "issue:1946", + "issue:2210", + "issue:2442", + "issue:3224", + "issue:3296", + "issue:3452", + "issue:409", + "issue:4125", + "issue:4376", + "issue:4427", + "issue:4468", + "issue:4773", + "issue:4975", + "issue:5125", + "issue:5447", + "issue:5606", + "issue:5661", + "issue:6176", + "issue:6183", + "issue:6280", + "issue:6362", + "issue:6382", + "issue:6385", + "issue:6527", + "issue:6602", + "issue:6632", + "issue:6722", + "issue:699", + "issue:7146", + "issue:7267", + "issue:7480", + "issue:7784", + "issue:785", + "issue:9803" + ], + "soft_pairs": [ + "issue:1922|issue:5661", + "issue:2210|issue:6722", + "issue:10016|issue:6183", + "issue:1552|issue:5447", + "issue:4125|issue:6362", + "issue:3296|issue:5125", + "issue:10428|issue:10470", + "issue:4376|issue:6527", + "issue:1486|issue:4468", + "issue:4773|issue:4975", + "issue:10470|issue:10797", + "issue:6385|issue:7146", + "issue:1925|issue:4427", + "issue:3224|issue:3452", + "issue:409|issue:785", + "issue:6176|issue:6602", + "issue:6280|issue:7784", + "issue:6382|issue:7480", + "issue:5606|issue:9803", + "issue:1720|issue:2442", + "issue:1865|issue:1946", + "issue:10591|issue:6632", + "issue:10928|issue:7267", + "issue:1141|issue:699" + ], + "prepared_review_unit_hash": "294acd5f937eb8e73fefe529554ec801070048f94330098cb8d056e2ae54679c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33332, + "estimated_input_tokens": 8333, + "estimated_eval_tokens": 16922 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous. Most pairings are clearly different bugs or feature requests from the titles alone. The only near-match is the two `from_single_file` shape-mismatch reports, but with title-only evidence and different models/pipelines, it should stay uncertain rather than accepted.", + "confidence": 0.81, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:6183 is the clearest representative of the `from_single_file` shape-mismatch theme, but the packet does not justify treating it as a confirmed duplicate canonical.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1922", + "right": "issue:5661", + "accept": false, + "reason": "Both involve missing names/imports, but they are different symbols in different contexts: notebook helper vs tokenizer typing/import issue." + }, + { + "left": "issue:2210", + "right": "issue:6722", + "accept": false, + "reason": "Both mention text encoders, but one is about training LoRA on the text encoder and the other is about using SigLIP as the text encoder." + }, + { + "left": "issue:10016", + "right": "issue:6183", + "accept": false, + "reason": "Close on `from_single_file` and shape mismatch, but the packet only gives titles and the pipelines/models differ (SD3.5M vs SDXL), so this is too uncertain to merge as a duplicate." + }, + { + "left": "issue:1552", + "right": "issue:5447", + "accept": false, + "reason": "Different failure modes in different pipeline/scheduler paths: attention slicing vs `denoising_start` with 2nd-order schedulers." + }, + { + "left": "issue:4125", + "right": "issue:6362", + "accept": false, + "reason": "One is about whether a loss curve looks expected; the other is an OOM during dataset mapping. Different problems." + }, + { + "left": "issue:3296", + "right": "issue:5125", + "accept": false, + "reason": "Checkpoint save/load with text encoder is unrelated to an XFormers CUDA invalid-configuration error." + }, + { + "left": "issue:10428", + "right": "issue:10470", + "accept": false, + "reason": "Both are Flux-related, but Ascend NPU inference and torchao inference are different environments and code paths." + }, + { + "left": "issue:4376", + "right": "issue:6527", + "accept": false, + "reason": "StableDiffusionXL textual inversion crash and EulerDiscreteScheduler `add_noise` bug are different subsystems." + }, + { + "left": "issue:1486", + "right": "issue:4468", + "accept": false, + "reason": "DDIMScheduler prediction type behavior is unrelated to multi-GPU vs single-GPU sample-code performance." + }, + { + "left": "issue:4773", + "right": "issue:4975", + "accept": false, + "reason": "Both are about LoRA loading, but one is a CivitAI loading issue and the other is inefficient datatype querying." + }, + { + "left": "issue:10470", + "right": "issue:10797", + "accept": false, + "reason": "Flux torchao inference and HunyuanVideo group offloading affect different pipelines and failure modes." + }, + { + "left": "issue:6385", + "right": "issue:7146", + "accept": false, + "reason": "Both are memory complaints, but StableVideoDiffusionPipeline VRAM usage and SDXL fine-tuning OOM are separate issues." + }, + { + "left": "issue:1925", + "right": "issue:4427", + "accept": false, + "reason": "Inpainting half-precision failure and T2I Multiadapter example breakage are different features and bugs." + }, + { + "left": "issue:3224", + "right": "issue:3452", + "accept": false, + "reason": "Learning-rate scaling duplication is unrelated to low-VRAM Dreambooth training." + }, + { + "left": "issue:409", + "right": "issue:785", + "accept": false, + "reason": "Missing `enable_attention_slicing` vs PNDMScheduler performance cost are not the same defect." + }, + { + "left": "issue:6176", + "right": "issue:6602", + "accept": false, + "reason": "Both involve LoRA/save-load workflows, but they hit different code paths: loading after Dreambooth-LoRA training vs loading after `fuse_lora()`/`save_pretrained()`." + }, + { + "left": "issue:6280", + "right": "issue:7784", + "accept": false, + "reason": "Missing `from_single_file` on one pipeline and missing `upcast_vae` on another are unrelated API issues." + }, + { + "left": "issue:6382", + "right": "issue:7480", + "accept": false, + "reason": "IP-adapter key error after LCM LoRA training and DeepFloyd IF OOM are unrelated." + }, + { + "left": "issue:5606", + "right": "issue:9803", + "accept": false, + "reason": "Both are LoRA loading failures, but one is about missing alpha handling and the other is a Flux-specific loading failure." + }, + { + "left": "issue:1720", + "right": "issue:2442", + "accept": false, + "reason": "M1 finetuning errors and an expected-loss-curve question for inpainting are different topics." + }, + { + "left": "issue:1865", + "right": "issue:1946", + "accept": false, + "reason": "Zero-stage-3 finetuning dimension errors and XFormers CUDA invalid-argument errors come from different subsystems." + }, + { + "left": "issue:10591", + "right": "issue:6632", + "accept": false, + "reason": "One reports a problem in an SD3 LoRA training script; the other asks whether `save_state` is necessary. Not a duplicate bug." + }, + { + "left": "issue:10928", + "right": "issue:7267", + "accept": false, + "reason": "Missing a specific UNet weight file and broken local dataset loading in `train_controlnet.py` are different file/loading failures." + }, + { + "left": "issue:1141", + "right": "issue:699", + "accept": false, + "reason": "Local file support in custom pipelines and negative prompts are unrelated feature requests." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Most of the negative pairwise judgments look conservative and are supported by the titles. However, the one accepted duplicate pair is a bit too strong for the evidence in the packet: issue:10016 and issue:6183 both involve `from_single_file` and shape mismatches, but the packet only gives titles, and the models/pipelines differ (SD3.5M vs SDXL) with no body-level confirmation that this is the same bug. I\u2019d keep that pair soft/uncertain rather than accepted." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 160, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10016", + "issue:10315", + "issue:10526", + "issue:10866", + "issue:11178", + "issue:11702", + "issue:12047", + "issue:13107", + "issue:1434", + "issue:1823", + "issue:2163", + "issue:228", + "issue:2696", + "issue:3223", + "issue:3379", + "issue:3452", + "issue:3658", + "issue:4080", + "issue:4316", + "issue:434", + "issue:435", + "issue:4446", + "issue:4724", + "issue:5044", + "issue:547", + "issue:5515", + "issue:6092", + "issue:6443", + "issue:6812", + "issue:6821", + "issue:7079", + "issue:7133", + "issue:7228", + "issue:7354", + "issue:7480", + "issue:7695", + "issue:8464", + "issue:8477", + "issue:8482", + "issue:8589", + "issue:8874", + "issue:9164", + "issue:9358", + "issue:945", + "issue:9514", + "issue:9767", + "issue:9914" + ], + "soft_pairs": [ + "issue:1823|issue:547", + "issue:228|issue:547", + "issue:3379|issue:945", + "issue:3452|issue:3658", + "issue:4080|issue:8482", + "issue:434|issue:435", + "issue:4724|issue:7695", + "issue:5515|issue:8464", + "issue:6812|issue:6821", + "issue:7133|issue:9358", + "issue:6092|issue:9164", + "issue:5044|issue:8874", + "issue:1434|issue:7480", + "issue:2163|issue:4316", + "issue:11702|issue:13107", + "issue:2696|issue:3223", + "issue:7079|issue:7228", + "issue:4446|issue:8589", + "issue:6443|issue:7354", + "issue:10526|issue:9514", + "issue:10016|issue:11178", + "issue:9767|issue:9914", + "issue:10315|issue:8477", + "issue:10866|issue:12047" + ], + "prepared_review_unit_hash": "1d405cb13478283e2f32598d4bd739ab6a373c78e1c8b8adff640d75d35e3ce6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33509, + "estimated_input_tokens": 8378, + "estimated_eval_tokens": 17012 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft pair looks like the same underlying bug/change. The set is a grab bag of unrelated loading, training, OOM, scheduler, and LoRA issues that only share broad diffusers/modeling vocabulary.", + "confidence": 0.97, + "canonical_issue_reason": "issue:10016 is the best representative only in the sense that it is a precise, concrete bug report with a clear failure mode and reproducible load path; the cluster as a whole is not a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10016 has the clearest scoped symptom and strongest bug-report quality among the issues, making it the least-bad representative if one must be chosen.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1823", + "right": "issue:547", + "accept": false, + "reason": "Swapped comments in depth2img is unrelated to documentation about loading stable diffusion locally." + }, + { + "left": "issue:228", + "right": "issue:547", + "accept": false, + "reason": "Broken model link is a release artifact problem, not a documentation issue." + }, + { + "left": "issue:3379", + "right": "issue:945", + "accept": false, + "reason": "Dreambooth checkpoint loading and ONNX custom checkpoint conversion are different code paths and failures." + }, + { + "left": "issue:3452", + "right": "issue:3658", + "accept": false, + "reason": "Low-VRAM Dreambooth training request is not the same as the additional-noise-latents bug in upsampler scripts." + }, + { + "left": "issue:4080", + "right": "issue:8482", + "accept": false, + "reason": "Adding SDXL inpainting support is a feature request, not the same as SD3.5 support." + }, + { + "left": "issue:434", + "right": "issue:435", + "accept": false, + "reason": "Training notebook breakage and AutoencoderKLOutput.sample error are related only by notebook context, not the same bug." + }, + { + "left": "issue:4724", + "right": "issue:7695", + "accept": false, + "reason": "SDXL img2img behavior and ip_adapter_masks warnings in attention are different problems." + }, + { + "left": "issue:5515", + "right": "issue:8464", + "accept": false, + "reason": "LCM scheduler initialization failure and SEGA seed/offload failure are different pipeline bugs." + }, + { + "left": "issue:6812", + "right": "issue:6821", + "accept": false, + "reason": "LoRA Q/K/V dtype mismatch and caption_column validation error are unrelated training-script issues." + }, + { + "left": "issue:7133", + "right": "issue:9358", + "accept": false, + "reason": "Checkpoint resume size mismatch in SDXL DreamBooth LoRA is not the same as redundant text-encoder reinit in Flux LoRA training." + }, + { + "left": "issue:6092", + "right": "issue:9164", + "accept": false, + "reason": "LoRA example not training and dog example not converging are both training complaints but not the same concrete defect." + }, + { + "left": "issue:5044", + "right": "issue:8874", + "accept": false, + "reason": "Kandinsky 2.2 load failure and CKPT unpickling failure are different load mechanisms and errors." + }, + { + "left": "issue:1434", + "right": "issue:7480", + "accept": false, + "reason": "Upscaling OOM in StableDiffusionUpscalePipeline is unrelated to DeepFloyd IF memory usage." + }, + { + "left": "issue:2163", + "right": "issue:4316", + "accept": false, + "reason": "UNet dtype mismatch in text-to-image training is not the same as validation-option failure in DreamBooth." + }, + { + "left": "issue:11702", + "right": "issue:13107", + "accept": false, + "reason": "AccVid LoRA key errors and Flux Klein LoRA missing keys are different model/LoRA compatibility issues." + }, + { + "left": "issue:2696", + "right": "issue:3223", + "accept": false, + "reason": "Torch CUDA not compiled and RAM not cleared after moving pipeline to cuda are unrelated environment/runtime issues." + }, + { + "left": "issue:7079", + "right": "issue:7228", + "accept": false, + "reason": "Transformer2DModel patch/norm error and traced-model missing dtype are distinct model API bugs." + }, + { + "left": "issue:4446", + "right": "issue:8589", + "accept": false, + "reason": "Custom text_encoder loading in from_single_file is not the same as SD3 DreamBooth prior-preservation training." + }, + { + "left": "issue:6443", + "right": "issue:7354", + "accept": false, + "reason": "SAGPipeline scheduler compatibility and PNDMScheduler incompatibility with DDPMPipeline are different scheduler/pipeline mismatches." + }, + { + "left": "issue:10526", + "right": "issue:9514", + "accept": false, + "reason": "optimum.quanto FP8 tensor construction error and LoRA loading after fp8 quantisation are related only broadly to quantization." + }, + { + "left": "issue:10016", + "right": "issue:11178", + "accept": false, + "reason": "Both are from_single_file loading errors, but one is SD3.5M pos_embed shape mismatch and the other is MistoLine ControlNet loading; different models and failure modes." + }, + { + "left": "issue:9767", + "right": "issue:9914", + "accept": false, + "reason": "FluxImg2Img cuDNN frontend error and Xlabs-trained LoRA loading fix proposal are not the same underlying bug." + }, + { + "left": "issue:10315", + "right": "issue:8477", + "accept": false, + "reason": "CogVideo training error and distributed consistency-script error are unrelated training failures." + }, + { + "left": "issue:10866", + "right": "issue:12047", + "accept": false, + "reason": "Lumina Image 2.0 LoRA incompatibility and Wan2.2 GGUF LoRA fusion failure are different model-family issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 161, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10231", + "issue:1037", + "issue:10639", + "issue:11136", + "issue:1141", + "issue:12047", + "issue:12329", + "issue:13079", + "issue:1407", + "issue:1537", + "issue:1541", + "issue:1974", + "issue:2107", + "issue:2156", + "issue:2444", + "issue:2465", + "issue:2781", + "issue:3182", + "issue:3420", + "issue:3984", + "issue:4134", + "issue:4236", + "issue:4398", + "issue:4488", + "issue:4639", + "issue:5447", + "issue:5520", + "issue:5557", + "issue:5619", + "issue:5919", + "issue:6230", + "issue:6492", + "issue:6791", + "issue:6815", + "issue:6832", + "issue:6870", + "issue:7418", + "issue:7492", + "issue:7591", + "issue:771", + "issue:7931", + "issue:8874", + "issue:8885", + "issue:8939", + "issue:9052", + "issue:9558", + "issue:9729", + "issue:9873" + ], + "soft_pairs": [ + "issue:4236|issue:771", + "issue:1407|issue:3984", + "issue:4398|issue:6492", + "issue:2107|issue:6815", + "issue:2444|issue:4639", + "issue:1037|issue:1141", + "issue:10639|issue:9873", + "issue:2156|issue:2465", + "issue:2781|issue:5919", + "issue:5557|issue:9729", + "issue:7418|issue:8939", + "issue:6832|issue:7931", + "issue:12047|issue:12329", + "issue:5520|issue:7591", + "issue:5619|issue:9558", + "issue:8874|issue:9052", + "issue:6870|issue:7492", + "issue:3182|issue:3420", + "issue:6230|issue:6791", + "issue:1537|issue:4134", + "issue:1541|issue:4488", + "issue:11136|issue:13079", + "issue:1974|issue:5447", + "issue:10231|issue:8885" + ], + "prepared_review_unit_hash": "6ebdcacae71824e7bb291e9e25f334f14a002b52a213f814f692497948484cc9", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34596, + "estimated_input_tokens": 8649, + "estimated_eval_tokens": 17554 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This looks like a false-positive cluster: the items span unrelated issues across tutorials, training scripts, pipeline support, model loading, and device/runtime errors. None of the soft pairs appear to be the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits this set; the items are thematically broad but not duplicate-convergent, so the cluster should be split rather than anchored to one issue.", + "canonical_pr_reason": null, + "best_issue_reason": "None of the issues is a strong global canonical for the whole set because they address different code paths and product areas; the safest triage is no global issue canonical.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4236", + "right": "issue:771", + "accept": false, + "reason": "Both mention SD/model incompatibility, but one is an SDXL refiner registration issue and the other is a BF16+Dreambooth problem; different bugs." + }, + { + "left": "issue:1407", + "right": "issue:3984", + "accept": false, + "reason": "One is an ONNX pipeline missing feature_extractor, the other is a class_labels validation error; unrelated failure modes." + }, + { + "left": "issue:4398", + "right": "issue:6492", + "accept": false, + "reason": "Both describe black images, but one is an SDXL LoRA-dreambooth pipeline and the other is generic image generation; too broad to treat as the same bug." + }, + { + "left": "issue:2107", + "right": "issue:6815", + "accept": false, + "reason": "Both involve bad low-resolution/black outputs, but one is about sub-512 image resolution glitches and the other is a tensorboard casting warning; not the same code path." + }, + { + "left": "issue:2444", + "right": "issue:4639", + "accept": false, + "reason": "Both are training-script errors, but one is dreambooth inpaint subprocess failure and the other is SDXL instruct-pix2pix dataloader multiprocessing; different causes." + }, + { + "left": "issue:1037", + "right": "issue:1141", + "accept": false, + "reason": "A community pipeline request and local-file support for custom pipelines are different feature requests." + }, + { + "left": "issue:10639", + "right": "issue:9873", + "accept": false, + "reason": "Both are model/pipeline additions, but Lumina-Image 2.0 and OmniGen are distinct model integrations." + }, + { + "left": "issue:2156", + "right": "issue:2465", + "accept": false, + "reason": "One reports CPU-only Dreambooth training; the other asks for a reproducible recipe. Same topic, not the same bug." + }, + { + "left": "issue:2781", + "right": "issue:5919", + "accept": false, + "reason": "Distributed training _cuda_setDevice error and LoRA layer attribute error are unrelated runtime issues." + }, + { + "left": "issue:5557", + "right": "issue:9729", + "accept": false, + "reason": "Both mention unexpected arguments, but one is use_reentrant and the other is save_model on T5EncoderModel; different API mismatches." + }, + { + "left": "issue:7418", + "right": "issue:8939", + "accept": false, + "reason": "The first is a bug in a community upscale script; the second is a tokenizer loading failure for SD3. Different subsystems." + }, + { + "left": "issue:6832", + "right": "issue:7931", + "accept": false, + "reason": "Both involve IP-Adapter-like image handling, but one is a tuple/shape crash and the other is ControlNet inpaint with ip_adapter_image_embeds; not the same defect." + }, + { + "left": "issue:12047", + "right": "issue:12329", + "accept": false, + "reason": "Both concern Wan model variants, but one is LoRA fusion failing on Wan2.2 GGUF and the other is incorrect transformer version selection; separate bugs." + }, + { + "left": "issue:5520", + "right": "issue:7591", + "accept": false, + "reason": "Both mention HeaderTooLarge, but one is generic deserialization and the other is from_single_file with local_files_only; not enough to merge." + }, + { + "left": "issue:5619", + "right": "issue:9558", + "accept": false, + "reason": "Both end up with black images, but one is an AutoencodeTiny LCM img2img encode bug and the other is tensorboard validation display; different failure points." + }, + { + "left": "issue:8874", + "right": "issue:9052", + "accept": false, + "reason": "Both are from_single_file checkpoint-loading problems, but one is a PyTorch unpickling/weights-only failure and the other is civit checkpoint compatibility; likely different root causes." + }, + { + "left": "issue:6870", + "right": "issue:7492", + "accept": false, + "reason": "One is missing enable_vae_tiling on img2img/inpaint, the other is a repository lookup failure for a T2I adapter URL; unrelated." + }, + { + "left": "issue:3182", + "right": "issue:3420", + "accept": false, + "reason": "Both are pipeline incompatibilities, but one is CPU offload with a custom LPW pipeline and the other is a ControlNet inpaint pipeline vs diffusion pipeline mismatch." + }, + { + "left": "issue:6230", + "right": "issue:6791", + "accept": false, + "reason": "Both are SDXL training script problems, but one is CUDA OOM and the other is a script that cannot run at all; not the same bug." + }, + { + "left": "issue:1537", + "right": "issue:4134", + "accept": false, + "reason": "Apple MPS scheduler behavior and ControlNet tensor-size mismatch are different device/model issues." + }, + { + "left": "issue:1541", + "right": "issue:4488", + "accept": false, + "reason": "Dreambooth CKPT model-type detection and SDXL checkpoint loading without pipeline_class are separate loading problems." + }, + { + "left": "issue:11136", + "right": "issue:13079", + "accept": false, + "reason": "A SanaSprint timestep limitation and a FLUX.2-klein CUDA OOM loop issue are unrelated." + }, + { + "left": "issue:1974", + "right": "issue:5447", + "accept": false, + "reason": "Euler generator-list reproducibility and SDXL img2img denoising_start with 2nd-order schedulers do not describe the same bug." + }, + { + "left": "issue:10231", + "right": "issue:8885", + "accept": false, + "reason": "Ubuntu tutorial ImportError and custom diffusion dependency/token loading issues are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 162, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10281", + "issue:1049", + "issue:11088", + "issue:11135", + "issue:1141", + "issue:11978", + "issue:12047", + "issue:12149", + "issue:12547", + "issue:12574", + "issue:12722", + "issue:13079", + "issue:13232", + "issue:1614", + "issue:1730", + "issue:2747", + "issue:2785", + "issue:3045", + "issue:3740", + "issue:4002", + "issue:4213", + "issue:4321", + "issue:457", + "issue:4892", + "issue:4974", + "issue:5004", + "issue:5475", + "issue:5605", + "issue:5628", + "issue:5873", + "issue:6093", + "issue:6232", + "issue:660", + "issue:6928", + "issue:7053", + "issue:7080", + "issue:7142", + "issue:7168", + "issue:7752", + "issue:7882", + "issue:7978", + "issue:9161", + "issue:9186", + "issue:9227", + "issue:9448", + "issue:9576", + "issue:9673", + "issue:9942" + ], + "soft_pairs": [ + "issue:3740|issue:4213", + "issue:11088|issue:12047", + "issue:1049|issue:2747", + "issue:1141|issue:5605", + "issue:1614|issue:4321", + "issue:4002|issue:7882", + "issue:5873|issue:7053", + "issue:12547|issue:12574", + "issue:7168|issue:9448", + "issue:4892|issue:5475", + "issue:9161|issue:9227", + "issue:1730|issue:7752", + "issue:10281|issue:11135", + "issue:12722|issue:13079", + "issue:3045|issue:4974", + "issue:7882|issue:9186", + "issue:457|issue:660", + "issue:11978|issue:13232", + "issue:12149|issue:9942", + "issue:7080|issue:7142", + "issue:9576|issue:9673", + "issue:5628|issue:6232", + "issue:2785|issue:6928", + "issue:5004|issue:6093", + "issue:4213|issue:7978" + ], + "prepared_review_unit_hash": "41e3de3109a8faa1f4108017684a9728dc826fe6acc47a344858c8ea39e7cddc", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34418, + "estimated_input_tokens": 8605, + "estimated_eval_tokens": 17466 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose grab-bag of unrelated issues and feature requests; the soft edges are driven by generic title overlap, not shared bugs or mergeable changes. No pair looks like a true duplicate.", + "confidence": 0.99, + "canonical_issue_reason": "No single canonical issue: the items span unrelated schedulers, pipeline requests, training-script bugs, conversion problems, and backend/device-specific failures.", + "canonical_pr_reason": "No PRs are present in this cluster.", + "best_issue_reason": "No strong global best issue; if forced, issue:12574 is the most concrete bug report, but it does not represent the rest of the cluster.", + "best_pr_reason": "No PRs are present in this cluster.", + "soft_edge_verdicts": [ + { + "left": "issue:3740", + "right": "issue:4213", + "accept": false, + "reason": "Different problems: legacy attention conversion with device_map=sequential vs SDXL from_single_file loading without accelerate." + }, + { + "left": "issue:11088", + "right": "issue:12047", + "accept": false, + "reason": "Both mention Wan/GGUF, but one is a load error and the other is LoRA fusion failure on a different model/version and code path." + }, + { + "left": "issue:1049", + "right": "issue:2747", + "accept": false, + "reason": "Both are docs-related, but one is specifically Euler scheduler docs and the other is a generic docs update request." + }, + { + "left": "issue:1141", + "right": "issue:5605", + "accept": false, + "reason": "Unrelated feature requests: local files in custom pipelines vs per-image caption support for SDXL dreambooth." + }, + { + "left": "issue:1614", + "right": "issue:4321", + "accept": false, + "reason": "Both show black images, but on different backends and causes: autocast on SD 2.1 vs ONNX FP16 DirectML." + }, + { + "left": "issue:4002", + "right": "issue:7882", + "accept": false, + "reason": "Both mention from_single_file, but they target different pipeline classes and likely different registration gaps." + }, + { + "left": "issue:5873", + "right": "issue:7053", + "accept": false, + "reason": "Completely different bugs: ONNX export of scaled_dot_product_attention vs UNet3DConditionModel class conditioning." + }, + { + "left": "issue:12547", + "right": "issue:12574", + "accept": false, + "reason": "Different training scripts, models, and missing arguments; not the same underlying failure mode." + }, + { + "left": "issue:7168", + "right": "issue:9448", + "accept": false, + "reason": "Both involve tuple/shape errors around IP-Adapter-style inputs, but they are different pipelines and different code paths." + }, + { + "left": "issue:4892", + "right": "issue:5475", + "accept": false, + "reason": "Different pipeline support requests: TokenFlow video editing vs LaVie video generation." + }, + { + "left": "issue:9161", + "right": "issue:9227", + "accept": false, + "reason": "Same script name, but one reports distributed training bugs while the other is a model-specific run issue; not a single concrete bug." + }, + { + "left": "issue:1730", + "right": "issue:7752", + "accept": false, + "reason": "Unrelated topics: model conversion to webui vs a training hyperparameter question." + }, + { + "left": "issue:10281", + "right": "issue:11135", + "accept": false, + "reason": "Different requests entirely: a new diffusion scheduler vs a new pipeline proposal." + }, + { + "left": "issue:12722", + "right": "issue:13079", + "accept": false, + "reason": "Unrelated: Kaggle usage question for qwen-image vs CUDA OOM in FLUX.2-klein." + }, + { + "left": "issue:3045", + "right": "issue:4974", + "accept": false, + "reason": "Both concern conversion to diffusers, but one is LoRA conversion and the other is general model conversion." + }, + { + "left": "issue:7882", + "right": "issue:9186", + "accept": false, + "reason": "Different feature areas: missing from_single_file for t2i adapter vs differential diffusion with ControlNet." + }, + { + "left": "issue:457", + "right": "issue:660", + "accept": false, + "reason": "Both are MPS-related, but one reports blue artifacts and the other is about running fp16 on M1; not the same bug." + }, + { + "left": "issue:11978", + "right": "issue:13232", + "accept": false, + "reason": "Different model-support requests: Bria 3.2 vs LTX-2.3." + }, + { + "left": "issue:12149", + "right": "issue:9942", + "accept": false, + "reason": "Both are packaging/version issues, but one is source-code mismatch after install and the other is install failure." + }, + { + "left": "issue:7080", + "right": "issue:7142", + "accept": false, + "reason": "Different model/pipeline contexts: SDXL controlnet training inference mismatch vs SDXL-lightning inference mismatch." + }, + { + "left": "issue:9576", + "right": "issue:9673", + "accept": false, + "reason": "Different loading bugs: safetensors propagation to submodules vs high CPU usage when loading multiple LoRAs." + }, + { + "left": "issue:5628", + "right": "issue:6232", + "accept": false, + "reason": "Scheduler bugs, but in different schedulers and different behaviors; not one shared defect." + }, + { + "left": "issue:2785", + "right": "issue:6928", + "accept": false, + "reason": "Unrelated feature requests: M1/M2 support for text-to-video vs disabling depthmap colorization in Marigold." + }, + { + "left": "issue:5004", + "right": "issue:6093", + "accept": false, + "reason": "Different SDXL training-script issues: checkpoint resume at fp16 vs push_to_hub repo creation behavior." + }, + { + "left": "issue:4213", + "right": "issue:7978", + "accept": false, + "reason": "Completely unrelated issues: from_single_file with SDXL vs missing torchvision::nms." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 163, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1037", + "issue:10447", + "issue:10550", + "issue:10594", + "issue:10650", + "issue:1125", + "issue:12396", + "issue:12459", + "issue:12533", + "issue:12536", + "issue:1434", + "issue:1811", + "issue:1931", + "issue:2107", + "issue:2673", + "issue:2838", + "issue:3182", + "issue:4104", + "issue:4185", + "issue:4368", + "issue:4402", + "issue:4476", + "issue:4605", + "issue:4726", + "issue:4806", + "issue:5215", + "issue:5399", + "issue:5509", + "issue:5630", + "issue:5719", + "issue:5897", + "issue:6164", + "issue:6362", + "issue:7003", + "issue:7267", + "issue:7354", + "issue:7660", + "issue:8464", + "issue:8729", + "issue:8792", + "issue:8826", + "issue:8839", + "issue:8900", + "issue:9523", + "issue:9637", + "issue:9717", + "issue:9767", + "issue:9856" + ], + "soft_pairs": [ + "issue:5630|issue:8792", + "issue:2107|issue:4104", + "issue:4402|issue:8900", + "issue:12459|issue:9717", + "issue:4726|issue:4806", + "issue:5897|issue:8826", + "issue:10650|issue:7003", + "issue:8464|issue:9637", + "issue:10447|issue:7267", + "issue:6362|issue:9523", + "issue:2673|issue:3182", + "issue:4368|issue:8729", + "issue:12396|issue:9856", + "issue:5399|issue:5509", + "issue:1037|issue:2838", + "issue:1125|issue:4605", + "issue:8839|issue:9767", + "issue:1811|issue:5719", + "issue:4185|issue:4476", + "issue:1931|issue:6164", + "issue:7354|issue:7660", + "issue:12533|issue:12536", + "issue:10550|issue:10594", + "issue:1434|issue:5215" + ], + "prepared_review_unit_hash": "397b5c2b4b5662c734651142296b856d881e407c875590851799c93f9607e703", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34685, + "estimated_input_tokens": 8672, + "estimated_eval_tokens": 17600 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Only one soft pair is a clear duplicate: the two issues about preferring .safetensors/.bin weights when both exist. The rest are topic-adjacent at best but describe different pipelines, symptoms, or code paths.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 1811 is the cleaner canonical because it states the exact file-selection bug in `from_pretrained` and is phrased as the underlying behavior to fix; 5719 is the same bug from the LoRA-loading path and is narrower.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1811 is the best representative overall: specific, actionable, and clearly about a single loader bug rather than a broad symptom report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5630", + "right": "issue:8792", + "accept": false, + "reason": "Both involve SDXL image quality, but one is about LMS scheduler leftovers in img2img and the other is a general migration-quality complaint; not the same concrete bug." + }, + { + "left": "issue:2107", + "right": "issue:4104", + "accept": false, + "reason": "Both mention black/glitchy images, but 2107 is resolution-specific while 4104 is tied to an invalid-cast warning; likely different failure modes." + }, + { + "left": "issue:4402", + "right": "issue:8900", + "accept": false, + "reason": "Both use `from_single_file`, but they concern different SDXL models and different loading problems; not one shared bug." + }, + { + "left": "issue:12459", + "right": "issue:9717", + "accept": false, + "reason": "Throughput in batch inference and a docker image launch error are unrelated." + }, + { + "left": "issue:4726", + "right": "issue:4806", + "accept": false, + "reason": "Both are type-related API issues, but they affect different symbols and likely different fixes." + }, + { + "left": "issue:5897", + "right": "issue:8826", + "accept": false, + "reason": "LoRA training device-mismatch and an empty-list `torch.cat` error are distinct runtime problems." + }, + { + "left": "issue:10650", + "right": "issue:7003", + "accept": false, + "reason": "Both are shape mismatch errors, but they occur on different models/pipelines and point to different expected tensor layouts." + }, + { + "left": "issue:8464", + "right": "issue:9637", + "accept": false, + "reason": "CPU offload plus seed handling is unrelated to xFormers attention-mask/bfloat shape constraints." + }, + { + "left": "issue:10447", + "right": "issue:7267", + "accept": false, + "reason": "A conversion script splitting UNet weights is unrelated to broken local data loading in a training script." + }, + { + "left": "issue:6362", + "right": "issue:9523", + "accept": false, + "reason": "Both mention dataset mapping, but one is an OOM during fine-tuning and the other is an error in specific controlnet training scripts; too broad to be the same bug." + }, + { + "left": "issue:2673", + "right": "issue:3182", + "accept": false, + "reason": "Upscale pipeline config/image input errors are unrelated to custom pipeline CPU offload incompatibility." + }, + { + "left": "issue:4368", + "right": "issue:8729", + "accept": false, + "reason": "LoRA training OOM and an unexpected validation image in a controlnet training script are different issues." + }, + { + "left": "issue:12396", + "right": "issue:9856", + "accept": false, + "reason": "Adapter deletion/leaf-level errors are unrelated to distributed port conflicts." + }, + { + "left": "issue:5399", + "right": "issue:5509", + "accept": false, + "reason": "A dtype mismatch in custom inpainting is not the same as image batch-size validation for numpy/cv2 inputs." + }, + { + "left": "issue:1037", + "right": "issue:2838", + "accept": false, + "reason": "A request to add a pipeline and a request to support a controlnet pipeline without control are not duplicates." + }, + { + "left": "issue:1125", + "right": "issue:4605", + "accept": false, + "reason": "Both mention inpaint-related errors, but the pipelines and reported failures differ; not enough evidence of one shared bug." + }, + { + "left": "issue:8839", + "right": "issue:9767", + "accept": false, + "reason": "LoRA weight errors on SD3 and a cuDNN frontend error on Flux Img2Img are unrelated." + }, + { + "left": "issue:1811", + "right": "issue:5719", + "accept": true, + "reason": "Both report the same loader-selection bug: when both `.bin` and `.safetensors` are present, the code should prefer one instead of treating them as multiple files." + }, + { + "left": "issue:4185", + "right": "issue:4476", + "accept": false, + "reason": "Both touch SDXL ControlNet, but one is about black images at inference and the other about unexpected results on a training dataset; different bugs." + }, + { + "left": "issue:1931", + "right": "issue:6164", + "accept": false, + "reason": "Float16 dtype handling and latent output-type handling are separate pipeline behaviors." + }, + { + "left": "issue:7354", + "right": "issue:7660", + "accept": false, + "reason": "Scheduler compatibility with DDPMPipeline is unrelated to a trace example failure." + }, + { + "left": "issue:12533", + "right": "issue:12536", + "accept": false, + "reason": "Both involve context parallelism, but one is a hook conflict with CPU offload and the other is a mesh-size divisibility assertion; different code paths." + }, + { + "left": "issue:10550", + "right": "issue:10594", + "accept": false, + "reason": "Loading LoRA into a quantized base model and using multiple controls with LoRA on FLUX are different problems." + }, + { + "left": "issue:1434", + "right": "issue:5215", + "accept": false, + "reason": "Upscale OOM at larger resolutions and brightness shifts for dark images are unrelated symptoms." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and the soft-edge calls are conservative. The only accepted pair, 1811/5719, is reasonably supported as the same loader-selection bug, while the rest are plausibly different bugs/code paths based on the titles and metadata provided." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 164, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10287", + "issue:10804", + "issue:10952", + "issue:11024", + "issue:11651", + "issue:11687", + "issue:11700", + "issue:12082", + "issue:12299", + "issue:12529", + "issue:13162", + "issue:1544", + "issue:1925", + "issue:202", + "issue:2036", + "issue:2068", + "issue:2234", + "issue:2462", + "issue:3595", + "issue:3886", + "issue:4001", + "issue:4080", + "issue:4186", + "issue:4265", + "issue:4453", + "issue:4488", + "issue:4801", + "issue:4840", + "issue:5344", + "issue:5545", + "issue:5577", + "issue:5783", + "issue:6234", + "issue:6823", + "issue:6894", + "issue:7057", + "issue:7210", + "issue:7441", + "issue:7690", + "issue:7985", + "issue:861", + "issue:8873", + "issue:9089", + "issue:9164", + "issue:9576", + "issue:9804" + ], + "soft_pairs": [ + "issue:1925|issue:2036", + "issue:202|issue:4001", + "issue:2068|issue:4488", + "issue:3595|issue:4186", + "issue:4265|issue:6823", + "issue:4453|issue:861", + "issue:5783|issue:6234", + "issue:7057|issue:7441", + "issue:1544|issue:2234", + "issue:10026|issue:12082", + "issue:11651|issue:13162", + "issue:12299|issue:12529", + "issue:5545|issue:5577", + "issue:4840|issue:7985", + "issue:3886|issue:4801", + "issue:10804|issue:10952", + "issue:5545|issue:7210", + "issue:11687|issue:11700", + "issue:2462|issue:4080", + "issue:5344|issue:9089", + "issue:6894|issue:9164", + "issue:10287|issue:11024", + "issue:7690|issue:9804", + "issue:8873|issue:9576" + ], + "prepared_review_unit_hash": "ce107ad420b6e6857d0967ab40dd6ae86d63c3ca535c01bff77bbdf66914afe2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33274, + "estimated_input_tokens": 8319, + "estimated_eval_tokens": 16894 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs, training, loading, ControlNet, LoRA, scheduler, and compile/memory issues across different model families. None of the soft pairs look like the same underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": "No strong canonical issue exists because the cluster is not cohesive. If forced to pick one, #10287 is the broadest, but it is too vague to represent the others as a true duplicate anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "#10287 is the most generic/broad item in the set, but it is still a weak representative because the rest of the cluster spans many unrelated failures.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1925", + "right": "issue:2036", + "accept": false, + "reason": "Both mention model loading problems, but one is SD2 inpainting half-precision behavior and the other is a stable-diffusion-v1-4 OSError; different failures and code paths." + }, + { + "left": "issue:202", + "right": "issue:4001", + "accept": false, + "reason": "A test request for unet_blocks is unrelated to a README/demo documentation update." + }, + { + "left": "issue:2068", + "right": "issue:4488", + "accept": false, + "reason": "Different bugs: dtype conversion in from_pretrained vs SDXL checkpoint loading without pipeline_class." + }, + { + "left": "issue:3595", + "right": "issue:4186", + "accept": false, + "reason": "convert-from-ckpt failure is not the same as SDXL inpaint/refiner behavior." + }, + { + "left": "issue:4265", + "right": "issue:6823", + "accept": false, + "reason": "One is a specific import error for StableDiffusionXLControlNetPipeline; the other is a general 'cannot use controlnet' report." + }, + { + "left": "issue:4453", + "right": "issue:861", + "accept": false, + "reason": "LoRA conversion for SDXL/A1111 is different from generic ckpt-to-diffusers conversion." + }, + { + "left": "issue:5783", + "right": "issue:6234", + "accept": false, + "reason": "EMA training behavior and a generic example issue are not the same bug." + }, + { + "left": "issue:7057", + "right": "issue:7441", + "accept": false, + "reason": "Model checkpoint loading failure vs pytest failing on Python 3.12 are unrelated." + }, + { + "left": "issue:1544", + "right": "issue:2234", + "accept": false, + "reason": "Missing get_velocity on DDPMScheduler is a scheduler API issue; xformers finetuning is a separate performance/training problem." + }, + { + "left": "issue:10026", + "right": "issue:12082", + "accept": false, + "reason": "Both are GPU-memory-related, but one is Flux ControlNet + DeepSpeed training and the other is Wan VAE memory after compile; different stacks." + }, + { + "left": "issue:11651", + "right": "issue:13162", + "accept": false, + "reason": "Both involve Flux LoRA support, but one is loading flux-fill-lora with FluxControl while the other is a request for Flux.2-klein-4b support; not the same concrete bug." + }, + { + "left": "issue:12299", + "right": "issue:12529", + "accept": false, + "reason": "Different models and errors: SD3 tensor mismatch with negative prompt embeddings vs Wan2.2 tiled VAE size mismatch." + }, + { + "left": "issue:5545", + "right": "issue:5577", + "accept": false, + "reason": "StableDiffusionInpaintPipeline issue and ControlNetModel.from_single_file KeyError are distinct loading paths." + }, + { + "left": "issue:4840", + "right": "issue:7985", + "accept": false, + "reason": "FlaxAutoencoderKL checkpoint loading and IPAdapter torch.compile incompatibility are unrelated." + }, + { + "left": "issue:3886", + "right": "issue:4801", + "accept": false, + "reason": "A distributed watchdog timeout is not the same as a mixed-precision example bug." + }, + { + "left": "issue:10804", + "right": "issue:10952", + "accept": false, + "reason": "Both are LoRA-loading related, but they fail in different pipelines with different adapter-name behavior; not clearly the same bug." + }, + { + "left": "issue:5545", + "right": "issue:7210", + "accept": false, + "reason": "Inpaint pipeline trouble and missing named_parameters on StableDiffusionXLPipeline are separate issues." + }, + { + "left": "issue:11687", + "right": "issue:11700", + "accept": false, + "reason": "One is a docs request for adding Pruna; the other is implementation work for loading through from_pretrained/pipeline." + }, + { + "left": "issue:2462", + "right": "issue:4080", + "accept": false, + "reason": "IterableDataset training support and adding an SDXL inpainting pipeline are unrelated changes." + }, + { + "left": "issue:5344", + "right": "issue:9089", + "accept": false, + "reason": "Missing omegaconf dependency and missing FluxPipeline symbol are different problems." + }, + { + "left": "issue:6894", + "right": "issue:9164", + "accept": false, + "reason": "Both are training-example LoRA complaints, but they target different models and symptoms, so they are not the same underlying bug." + }, + { + "left": "issue:10287", + "right": "issue:11024", + "accept": false, + "reason": "A generic docs issue is unrelated to the SiglipImageProcessor .to() method bug." + }, + { + "left": "issue:7690", + "right": "issue:9804", + "accept": false, + "reason": "Tokenizer loading failure for SDXL base and local SDXL ControlNet loading failure are different code paths." + }, + { + "left": "issue:8873", + "right": "issue:9576", + "accept": false, + "reason": "xformers runtime slowdown and use_safetensors not propagating to submodules are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 165, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1061", + "issue:10722", + "issue:10863", + "issue:11046", + "issue:1475", + "issue:1534", + "issue:1867", + "issue:1949", + "issue:1988", + "issue:2052", + "issue:2161", + "issue:2163", + "issue:228", + "issue:2577", + "issue:2674", + "issue:2774", + "issue:2880", + "issue:3064", + "issue:381", + "issue:3871", + "issue:3948", + "issue:4086", + "issue:4398", + "issue:4566", + "issue:5025", + "issue:5043", + "issue:5465", + "issue:5575", + "issue:5606", + "issue:5607", + "issue:5670", + "issue:579", + "issue:5925", + "issue:6130", + "issue:6196", + "issue:7592", + "issue:7782", + "issue:7793", + "issue:8469", + "issue:8493", + "issue:8667", + "issue:8828", + "issue:8874", + "issue:8950", + "issue:9410", + "issue:9611", + "issue:9767", + "issue:9895" + ], + "soft_pairs": [ + "issue:9611|issue:9895", + "issue:2161|issue:5575", + "issue:8828|issue:8950", + "issue:10722|issue:5607", + "issue:1534|issue:381", + "issue:2774|issue:579", + "issue:5043|issue:5925", + "issue:5670|issue:8667", + "issue:11046|issue:8469", + "issue:3948|issue:4398", + "issue:2577|issue:2674", + "issue:6130|issue:6196", + "issue:1988|issue:3871", + "issue:5025|issue:9410", + "issue:2052|issue:4566", + "issue:4086|issue:7592", + "issue:7782|issue:8874", + "issue:7793|issue:9767", + "issue:10863|issue:8493", + "issue:1061|issue:1475", + "issue:1867|issue:5465", + "issue:1949|issue:2163", + "issue:228|issue:2880", + "issue:3064|issue:5606" + ], + "prepared_review_unit_hash": "06af1b7014c88312b5554e7984df0172577f1794b48808e9362914996898dd3c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33945, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated diffusers bugs and feature requests across training, loading, inference, and notebooks. None of the soft pairs look like true duplicates; they mostly share only a subsystem name or a generic error pattern.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue stands out because the items are not one underlying bug; they span distinct models, pipelines, and failure modes.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10722 is the most concrete standalone bug report in the set: it has a specific runtime error, clear training context, and enough detail to be actionable.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9611", + "right": "issue:9895", + "accept": false, + "reason": "Both are Flux-related, but one is an image batching error and the other is a device-placement regression after a commit; different code paths." + }, + { + "left": "issue:2161", + "right": "issue:5575", + "accept": false, + "reason": "Both mention LoRA training, but one is about inpainting-model support and the other is about hidden-size configuration for a layer." + }, + { + "left": "issue:8828", + "right": "issue:8950", + "accept": false, + "reason": "Both involve SD3 training, but one is a generic 16GB training error and the other is a LoRA training bug; not the same failure." + }, + { + "left": "issue:10722", + "right": "issue:5607", + "accept": false, + "reason": "Both are tensor-size mismatch errors, but the shapes and training scenarios differ, so they do not clearly point to one underlying bug." + }, + { + "left": "issue:1534", + "right": "issue:381", + "accept": false, + "reason": "One is a DreamBooth notebook error and the other is a custom inference pipeline notebook issue; unrelated notebook paths." + }, + { + "left": "issue:2774", + "right": "issue:579", + "accept": false, + "reason": "ControlNet safetensors conversion and missing config.json in textual inversion are different loading problems." + }, + { + "left": "issue:5043", + "right": "issue:5925", + "accept": false, + "reason": "These are separate feature requests: adding LoRA finetuning for Wuerstchen vs adding ZipLora support." + }, + { + "left": "issue:5670", + "right": "issue:8667", + "accept": false, + "reason": "Both mention non-square image handling, but they target different pipelines/models and likely different shape assumptions." + }, + { + "left": "issue:11046", + "right": "issue:8469", + "accept": false, + "reason": "One is a complex Flux pipeline composition request/error, the other is corrupted faces from a specific IP-Adapter model; not the same bug." + }, + { + "left": "issue:3948", + "right": "issue:4398", + "accept": false, + "reason": "Both concern bad generated images, but pale images and all-black images are distinct symptoms with different likely causes." + }, + { + "left": "issue:2577", + "right": "issue:2674", + "accept": false, + "reason": "DDIMInverseScheduler usage and a LoRA dependency/requirements issue are unrelated." + }, + { + "left": "issue:6130", + "right": "issue:6196", + "accept": false, + "reason": "Both are loading failures, but one is generic LoRA loading and the other is ControlNetXS Depth SDXL loading; different artifacts and code paths." + }, + { + "left": "issue:1988", + "right": "issue:3871", + "accept": false, + "reason": "One is an fp16/bf16 dtype mismatch during image generation in a trainer example; the other is checkpoint-resume trouble in text-to-image training." + }, + { + "left": "issue:5025", + "right": "issue:9410", + "accept": false, + "reason": "An import error for randn_tensor and an inability to import a specific SD3 ControlNet inpainting pipeline are different import failures." + }, + { + "left": "issue:2052", + "right": "issue:4566", + "accept": false, + "reason": "Both mention resuming from checkpoint, but they affect different training scripts and model families." + }, + { + "left": "issue:4086", + "right": "issue:7592", + "accept": false, + "reason": "ControlNet training on SDXL and DoRA state_dict loading are separate problems involving different models and failure points." + }, + { + "left": "issue:7782", + "right": "issue:8874", + "accept": false, + "reason": "Both involve loading from files/checkpoints, but one is a bad safetensors conversion and the other is a CKPT unpickling weights-only failure." + }, + { + "left": "issue:7793", + "right": "issue:9767", + "accept": false, + "reason": "One is unloading LoRA weights in DeepFloyd, the other is a cuDNN frontend error during Flux LoRA inference; not the same bug." + }, + { + "left": "issue:10863", + "right": "issue:8493", + "accept": false, + "reason": "Both are state_dict loading errors, but one is a local safetensors file load failure and the other is SD3Transformer2DModel key mismatch." + }, + { + "left": "issue:1061", + "right": "issue:1475", + "accept": false, + "reason": "These are different training/tutorial issues: a docs update for textual inversion vs training with two class types in DreamBooth/textual inversion." + }, + { + "left": "issue:1867", + "right": "issue:5465", + "accept": false, + "reason": "Both involve image-to-image strength behavior, but img2img scheduler behavior and DiffEdit inpaint_strength support are distinct issues." + }, + { + "left": "issue:1949", + "right": "issue:2163", + "accept": false, + "reason": "Both are training example errors, but one is a custom-model tutorial problem and the other is a UNet dtype mismatch in train_text_to_image." + }, + { + "left": "issue:228", + "right": "issue:2880", + "accept": false, + "reason": "A broken release link and a Colab notebook that does not run are unrelated." + }, + { + "left": "issue:3064", + "right": "issue:5606", + "accept": false, + "reason": "Both mention LoRA and safetensors, but one is loading a .safetensors LoRA and the other is handling missing alpha metadata." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 166, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10520", + "issue:11046", + "issue:11185", + "issue:11824", + "issue:12348", + "issue:12536", + "issue:1988", + "issue:2596", + "issue:2603", + "issue:2669", + "issue:2673", + "issue:2991", + "issue:3412", + "issue:3910", + "issue:4031", + "issue:4037", + "issue:4134", + "issue:4697", + "issue:5025", + "issue:5227", + "issue:5458", + "issue:5595", + "issue:5596", + "issue:5862", + "issue:5905", + "issue:6162", + "issue:6435", + "issue:6791", + "issue:7114", + "issue:7185", + "issue:7267", + "issue:7420", + "issue:7556", + "issue:7636", + "issue:7799", + "issue:8237", + "issue:8450", + "issue:8482", + "issue:9045", + "issue:9115", + "issue:9227", + "issue:9309", + "issue:9341", + "issue:9378", + "issue:9402", + "issue:9515", + "issue:9670" + ], + "soft_pairs": [ + "issue:3412|issue:7556", + "issue:5905|issue:8482", + "issue:9378|issue:9402", + "issue:7799|issue:9115", + "issue:4031|issue:7267", + "issue:3910|issue:4697", + "issue:5595|issue:6435", + "issue:9227|issue:9309", + "issue:2669|issue:5596", + "issue:2991|issue:4037", + "issue:9341|issue:9670", + "issue:4134|issue:7636", + "issue:10520|issue:9045", + "issue:1988|issue:6162", + "issue:11185|issue:11824", + "issue:2596|issue:5025", + "issue:3910|issue:8237", + "issue:5458|issue:7185", + "issue:5862|issue:9515", + "issue:6791|issue:7114", + "issue:12348|issue:12536", + "issue:11046|issue:8450", + "issue:5227|issue:7420", + "issue:2603|issue:2673" + ], + "prepared_review_unit_hash": "9c074aa2cdb8b52b973fb9cfcf8443d375d7f64e3af92d1901a76e481311fb1a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33465, + "estimated_input_tokens": 8367, + "estimated_eval_tokens": 16990 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly one-off runtime, loading, or documentation issues across different pipelines and models. None of the soft pairs look like true duplicates of the same concrete bug/change; they mostly share a subsystem or symptom type rather than the same code-path failure.", + "confidence": 0.38, + "canonical_issue_reason": "No clear duplicate cluster emerges, so the best representative issue is issue 12348: it\u2019s a concrete, reproducible inference tensor-mismatch report for WanImageToVideoPipeline and is one of the most specific runtime-bug entries in the set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 12348 is the strongest standalone issue in this set because it describes an actionable pipeline failure with a clear symptom and scope. It is more representative than the vague feature requests or generic error reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3412", + "right": "issue:7556", + "accept": false, + "reason": "Both are docs-related, but one is a stable-diffusion docs update and the other is Animatediff scheduler docs; different scope and no shared bug/change." + }, + { + "left": "issue:5905", + "right": "issue:8482", + "accept": false, + "reason": "Different model requests: diffusion-GAN versus Stable Diffusion 3 support. Same broad repo area, not the same underlying change." + }, + { + "left": "issue:9378", + "right": "issue:9402", + "accept": false, + "reason": "Both involve Flux ControlNet, but one requests Xlabs ControlNet support and the other adds img2img/inpaint pipelines; separate features." + }, + { + "left": "issue:7799", + "right": "issue:9115", + "accept": false, + "reason": "Different pipelines and failure modes; both are image-generation errors but not the same concrete bug." + }, + { + "left": "issue:4031", + "right": "issue:7267", + "accept": false, + "reason": "Unrelated loading problems: safety checker local-file loading versus train_controlnet local data loading." + }, + { + "left": "issue:3910", + "right": "issue:4697", + "accept": false, + "reason": "Both mention LoRA loading, but the reports are too generic and likely different failure causes/regressions; not enough to treat as the same bug." + }, + { + "left": "issue:5595", + "right": "issue:6435", + "accept": false, + "reason": "One is a validation error for non-divisible dimensions; the other is a ControlNet default width/height initialization problem." + }, + { + "left": "issue:9227", + "right": "issue:9309", + "accept": false, + "reason": "Both are Flux-related, but one is a DreamBooth LoRA training issue and the other is a FluxControlNetPipeline runtime bug." + }, + { + "left": "issue:2669", + "right": "issue:5596", + "accept": false, + "reason": "Different attribute errors in different preprocessing/build paths; no clear shared code-path." + }, + { + "left": "issue:2991", + "right": "issue:4037", + "accept": false, + "reason": "Both are ControlNet training issues, but mixed-precision example failure and FSDP validation errors are distinct problems." + }, + { + "left": "issue:9341", + "right": "issue:9670", + "accept": false, + "reason": "Both concern LoRA loading, but one is about Flux LoRA key handling and the other is a missing PEFT backend requirement." + }, + { + "left": "issue:4134", + "right": "issue:7636", + "accept": false, + "reason": "A device-specific tensor size mismatch on MPS versus a how-to/support request for the SDXL tile ControlNet model." + }, + { + "left": "issue:10520", + "right": "issue:9045", + "accept": false, + "reason": "Both are device-placement issues, but they hit different models/pipelines and different code paths." + }, + { + "left": "issue:1988", + "right": "issue:6162", + "accept": false, + "reason": "Both are generation-time shape/type mismatches, but one is fp16/bf16 input-bias type mismatch and the other is an IP-Adapter shape mismatch." + }, + { + "left": "issue:11185", + "right": "issue:11824", + "accept": false, + "reason": "Different models and different error contexts; both are generic generation failures only at the title level." + }, + { + "left": "issue:2596", + "right": "issue:5025", + "accept": false, + "reason": "Both are diffusers.utils import/documentation problems, but they refer to different symbols and different underlying breakages." + }, + { + "left": "issue:3910", + "right": "issue:8237", + "accept": false, + "reason": "Both are LoRA-loading complaints, but one is a generic load failure and the other is a missing SDXL text-projection key." + }, + { + "left": "issue:5458", + "right": "issue:7185", + "accept": false, + "reason": "Local-files-only behavior in LoRA loading is unrelated to a broader PyTorch version support question." + }, + { + "left": "issue:5862", + "right": "issue:9515", + "accept": false, + "reason": "Different Transformer internals: gradient checkpointing support versus missing RMSNorm in a specific attention processor." + }, + { + "left": "issue:6791", + "right": "issue:7114", + "accept": false, + "reason": "Same training script, but one report says it can't run while the other is specifically CUDA OOM; not the same bug." + }, + { + "left": "issue:12348", + "right": "issue:12536", + "accept": false, + "reason": "Both are WanImageToVideoPipeline tensor-size complaints, but one is a resolution-dependent mismatch and the other is a context-parallelism mesh divisibility assertion; related theme, not clearly the same bug." + }, + { + "left": "issue:11046", + "right": "issue:8450", + "accept": false, + "reason": "Both involve Flux/SD-style inference, but one is a multi-feature Flux pipeline request and the other is an SDXL inpainting noise bug." + }, + { + "left": "issue:5227", + "right": "issue:7420", + "accept": false, + "reason": "Both mention torch.compile, but one is specifically load_lora_weight incompatibility and the other is a general inability to use torch.compile." + }, + { + "left": "issue:2603", + "right": "issue:2673", + "accept": false, + "reason": "Different pipeline issues: non-square image support versus an upscale pipeline config/image input error." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 167, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10575", + "issue:10972", + "issue:11103", + "issue:11209", + "issue:11224", + "issue:12319", + "issue:12329", + "issue:12565", + "issue:12673", + "issue:1304", + "issue:1647", + "issue:2185", + "issue:2210", + "issue:2645", + "issue:2689", + "issue:2841", + "issue:3111", + "issue:3347", + "issue:3689", + "issue:3724", + "issue:4080", + "issue:4354", + "issue:4647", + "issue:4667", + "issue:4701", + "issue:4837", + "issue:5170", + "issue:5226", + "issue:5337", + "issue:5387", + "issue:5593", + "issue:5647", + "issue:6415", + "issue:6610", + "issue:6713", + "issue:6731", + "issue:6758", + "issue:7214", + "issue:7229", + "issue:7441", + "issue:7592", + "issue:7974", + "issue:8541", + "issue:8615", + "issue:9237", + "issue:9454", + "issue:9729" + ], + "soft_pairs": [ + "issue:7229|issue:9237", + "issue:10575|issue:7214", + "issue:10972|issue:11209", + "issue:10972|issue:11224", + "issue:2185|issue:3347", + "issue:2645|issue:4701", + "issue:2841|issue:3724", + "issue:12319|issue:12673", + "issue:4837|issue:6713", + "issue:6415|issue:7974", + "issue:12329|issue:12565", + "issue:5226|issue:7592", + "issue:5647|issue:6610", + "issue:11103|issue:1647", + "issue:1304|issue:5337", + "issue:2210|issue:3111", + "issue:2645|issue:5593", + "issue:2689|issue:5387", + "issue:3689|issue:5170", + "issue:4080|issue:8615", + "issue:4354|issue:4667", + "issue:4647|issue:7441", + "issue:5226|issue:6758", + "issue:6731|issue:9729", + "issue:8541|issue:9454" + ], + "prepared_review_unit_hash": "5c075ccbe10b6b4c3a5a3c0523b8f20a2423b25379637f1192185327d5f75c6a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33452, + "estimated_input_tokens": 8363, + "estimated_eval_tokens": 16982 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly heterogeneous issues with only a couple of likely duplicate subclusters. The clearest matches are the LoRA-loading reports and the LoRA-removal reports; most other soft edges are only broad subsystem similarity and should stay separate.", + "confidence": 0.79, + "canonical_issue_reason": "issue:10972 is the strongest issue canonical in this set because it is the most specific LoRA-loading failure and has two near-identical follow-up reports (11209, 11224) pointing to the same problem class.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10972 is the best representative issue overall: concrete bug title, clear failure mode, and it anchors the only strong duplicate-like pairings in the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7229", + "right": "issue:9237", + "accept": false, + "reason": "Both mention training issues, but they concern different workflows and failure modes; not the same underlying bug." + }, + { + "left": "issue:10575", + "right": "issue:7214", + "accept": false, + "reason": "Same scheduler family, but one is an IndexError and the other is a karras_sigmas regression; too different to merge." + }, + { + "left": "issue:10972", + "right": "issue:11209", + "accept": true, + "reason": "Both are LoRA loading failures; 11209 looks like a generic report of the same underlying loading bug seen in 10972." + }, + { + "left": "issue:10972", + "right": "issue:11224", + "accept": true, + "reason": "Same LoRA-loading failure class as 10972, with 11224 appearing to be another report of the same bug." + }, + { + "left": "issue:2185", + "right": "issue:3347", + "accept": false, + "reason": "Training config files missing vs NaN image output are unrelated issues." + }, + { + "left": "issue:2645", + "right": "issue:4701", + "accept": false, + "reason": "Both are CUDA-related errors, but they occur in different code paths and are not the same bug." + }, + { + "left": "issue:2841", + "right": "issue:3724", + "accept": false, + "reason": "One is a scheduler feature request and the other is a watermarking feature idea; unrelated." + }, + { + "left": "issue:12319", + "right": "issue:12673", + "accept": false, + "reason": "Both involve group offloading, but the reported problems are different enough that they do not look like the same bug." + }, + { + "left": "issue:4837", + "right": "issue:6713", + "accept": false, + "reason": "Both mention the same conversion script, but the failures differ (SDXL conversion vs HeaderTooLarge)." + }, + { + "left": "issue:6415", + "right": "issue:7974", + "accept": false, + "reason": "Different IP-Adapter requests/bugs; not the same concrete issue." + }, + { + "left": "issue:12329", + "right": "issue:12565", + "accept": false, + "reason": "Same Wan image-to-video area, but one is a model-loading/version mismatch and the other is a running issue; not clearly duplicate." + }, + { + "left": "issue:5226", + "right": "issue:7592", + "accept": false, + "reason": "Both are LoRA/adapter loading problems, but they concern different mechanisms (load_lora_weights vs DoRA state_dict keys)." + }, + { + "left": "issue:5647", + "right": "issue:6610", + "accept": false, + "reason": "Both hit SDXL inpaint pipeline paths, but the symptoms and triggers differ." + }, + { + "left": "issue:11103", + "right": "issue:1647", + "accept": false, + "reason": "A repo-selection question and a color-correction question are unrelated." + }, + { + "left": "issue:1304", + "right": "issue:5337", + "accept": false, + "reason": "Training guidance vs callback semantics; no shared bug/change." + }, + { + "left": "issue:2210", + "right": "issue:3111", + "accept": false, + "reason": "Both are about conditioning/encoder ideas, but they are different feature requests." + }, + { + "left": "issue:2645", + "right": "issue:5593", + "accept": false, + "reason": "ControlNet CUDA failure vs AutoPipeline ControlNet wrapping failure; different issues." + }, + { + "left": "issue:2689", + "right": "issue:5387", + "accept": false, + "reason": "Both are broken-link reports, but they point to different links/doc pages." + }, + { + "left": "issue:3689", + "right": "issue:5170", + "accept": true, + "reason": "These describe the same underlying need to remove/unload LoRA weights after loading them." + }, + { + "left": "issue:4080", + "right": "issue:8615", + "accept": false, + "reason": "Different feature requests in different model families; not duplicates." + }, + { + "left": "issue:4354", + "right": "issue:4667", + "accept": false, + "reason": "Both involve from_single_file, but the pipeline class and failure mode differ." + }, + { + "left": "issue:4647", + "right": "issue:7441", + "accept": false, + "reason": "Model architecture bug vs test/runtime compatibility issue; unrelated." + }, + { + "left": "issue:5226", + "right": "issue:6758", + "accept": false, + "reason": "One is a LoRA loading bug and the other is a missing parameter support issue; not the same bug." + }, + { + "left": "issue:6731", + "right": "issue:9729", + "accept": false, + "reason": "Unhandled pipeline class vs unexpected saved model type in Flux training; unrelated." + }, + { + "left": "issue:8541", + "right": "issue:9454", + "accept": false, + "reason": "Both are model-not-found errors, but they refer to different models and repos." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted soft edges are defensible from the titles alone, especially the identical LoRA-loading reports and the unload/remove LoRA pair. The rejected edges are generally conservative and do not overclaim duplicates. No mergeability issues apply here because this is issue clustering rather than a PR pair." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 168, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10198", + "issue:10472", + "issue:10540", + "issue:10795", + "issue:11436", + "issue:11584", + "issue:11824", + "issue:11864", + "issue:11978", + "issue:12104", + "issue:13079", + "issue:13107", + "issue:13300", + "issue:1692", + "issue:1717", + "issue:1976", + "issue:2163", + "issue:2234", + "issue:2244", + "issue:3250", + "issue:3382", + "issue:3468", + "issue:4185", + "issue:4340", + "issue:4398", + "issue:4639", + "issue:4702", + "issue:5442", + "issue:5919", + "issue:6258", + "issue:6335", + "issue:6407", + "issue:6914", + "issue:6949", + "issue:7330", + "issue:7355", + "issue:7793", + "issue:7985", + "issue:8532", + "issue:8626", + "issue:8887", + "issue:8922", + "issue:9069", + "issue:9112", + "issue:9523", + "issue:954", + "issue:9744", + "issue:9857" + ], + "soft_pairs": [ + "issue:3382|issue:4639", + "issue:4185|issue:6949", + "issue:11584|issue:7985", + "issue:1692|issue:5919", + "issue:5442|issue:6407", + "issue:6914|issue:8626", + "issue:10795|issue:11864", + "issue:13300|issue:7355", + "issue:12104|issue:9112", + "issue:13079|issue:13107", + "issue:10198|issue:10540", + "issue:7793|issue:8532", + "issue:1976|issue:2234", + "issue:2163|issue:7330", + "issue:2244|issue:3468", + "issue:11978|issue:9744", + "issue:4702|issue:8887", + "issue:9523|issue:9857", + "issue:10472|issue:9069", + "issue:4340|issue:4398", + "issue:1717|issue:954", + "issue:11436|issue:11824", + "issue:6258|issue:8922", + "issue:3250|issue:6335" + ], + "prepared_review_unit_hash": "53b004aa764076ea5fb0e5a6f3149fed800b5356128772bef7f0a9b890f3fe10", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34451, + "estimated_input_tokens": 8613, + "estimated_eval_tokens": 17482 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated issues across training, inference, LoRA, torch.compile, model-loading, and feature requests. None of the soft edges look like true duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good canonical representative because the items do not share one underlying bug or change request. They span many distinct models, code paths, and symptoms.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3382", + "right": "issue:4639", + "accept": false, + "reason": "Different failures: multi-GPU text-to-image OOM vs multiprocessing data-loader error in a different script." + }, + { + "left": "issue:4185", + "right": "issue:6949", + "accept": false, + "reason": "Both are image-quality problems, but one is SDXL ControlNet black images and the other is DPMSolver even-step blank outputs; different code paths and symptoms." + }, + { + "left": "issue:11584", + "right": "issue:7985", + "accept": false, + "reason": "Different torch.compile incompatibilities in different pipelines/models; not the same underlying bug." + }, + { + "left": "issue:1692", + "right": "issue:5919", + "accept": false, + "reason": "Pickling a lambda and missing set_lora_layer on Linear are unrelated errors." + }, + { + "left": "issue:5442", + "right": "issue:6407", + "accept": false, + "reason": "One is a pipeline implementation request, the other is a request for a different model; not duplicates." + }, + { + "left": "issue:6914", + "right": "issue:8626", + "accept": false, + "reason": "One is an offload/runtime AttributeError in IP-Adapter-FaceID, the other is documentation/guidance for multiple IP-Adapter inputs." + }, + { + "left": "issue:10795", + "right": "issue:11864", + "accept": false, + "reason": "GGUF from_single_file torch.compile failure vs AutoencoderDC fullgraph name resolution error; unrelated code paths." + }, + { + "left": "issue:13300", + "right": "issue:7355", + "accept": false, + "reason": "Different models and failures: CPU generation errors on SD3.5/FLUX vs Stable Cascade bfloat16 arbitrary-resolution issue." + }, + { + "left": "issue:12104", + "right": "issue:9112", + "accept": false, + "reason": "Index out of bounds and tensor-size mismatch are different runtime bugs." + }, + { + "left": "issue:13079", + "right": "issue:13107", + "accept": false, + "reason": "CUDA OOM in a loop and LoRA missing-key loading errors are unrelated." + }, + { + "left": "issue:10198", + "right": "issue:10540", + "accept": false, + "reason": "Flux weight mismatch and Flux Dev from_single_file loading failure are different symptoms and likely different root causes." + }, + { + "left": "issue:7793", + "right": "issue:8532", + "accept": false, + "reason": "Both involve LoRA weights, but unloading errors and progressive memory growth are different problems; too broad to merge." + }, + { + "left": "issue:1976", + "right": "issue:2234", + "accept": false, + "reason": "Incorrect tokenizer file lookup and xformers finetuning failure are unrelated." + }, + { + "left": "issue:2163", + "right": "issue:7330", + "accept": false, + "reason": "Training dtype mismatch and FP16 unscale error are different failures in different parts of the training loop." + }, + { + "left": "issue:2244", + "right": "issue:3468", + "accept": false, + "reason": "Dreambooth CPU half-precision runtime error vs missing model_index.json are unrelated." + }, + { + "left": "issue:11978", + "right": "issue:9744", + "accept": false, + "reason": "Both are feature requests, but for different models and different additions." + }, + { + "left": "issue:4702", + "right": "issue:8887", + "accept": false, + "reason": "VQ-VAE example request and SD3 VAE encode bug are not the same change." + }, + { + "left": "issue:9523", + "right": "issue:9857", + "accept": false, + "reason": "Broad controlnet training error-mapping issue vs a specific FLUX controlnet embedding-size mismatch; not a duplicate." + }, + { + "left": "issue:10472", + "right": "issue:9069", + "accept": false, + "reason": "Unexpected lora_bias keyword and np.ndarray type mismatch are unrelated errors." + }, + { + "left": "issue:4340", + "right": "issue:4398", + "accept": false, + "reason": "SDXL LoRA training error with text encoder and black-image dreambooth inference are different bugs." + }, + { + "left": "issue:1717", + "right": "issue:954", + "accept": false, + "reason": "Offline cached-model loading and post-login loading failure are different authentication/cache problems." + }, + { + "left": "issue:11436", + "right": "issue:11824", + "accept": false, + "reason": "Hidream dtype mismatch vs FluxKontext multi-image input error are unrelated model-specific bugs." + }, + { + "left": "issue:6258", + "right": "issue:8922", + "accept": false, + "reason": "Forward-chunking quality degradation and quantized-T5 zero-output are different issues." + }, + { + "left": "issue:3250", + "right": "issue:6335", + "accept": false, + "reason": "StableDiffusionKDiffusionPipeline runtime failure and ControlNet output-size control bug are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 169, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10804", + "issue:1092", + "issue:10954", + "issue:11464", + "issue:11819", + "issue:12120", + "issue:12195", + "issue:1388", + "issue:1717", + "issue:222", + "issue:2324", + "issue:2822", + "issue:2947", + "issue:3112", + "issue:3468", + "issue:3623", + "issue:3824", + "issue:3937", + "issue:4117", + "issue:4183", + "issue:4194", + "issue:4332", + "issue:4340", + "issue:4366", + "issue:4520", + "issue:4727", + "issue:5226", + "issue:5474", + "issue:5488", + "issue:5545", + "issue:5647", + "issue:5862", + "issue:5897", + "issue:6093", + "issue:6203", + "issue:6839", + "issue:7041", + "issue:7377", + "issue:7644", + "issue:7681", + "issue:817", + "issue:8567", + "issue:8668", + "issue:8708", + "issue:8900", + "issue:9142", + "issue:9270" + ], + "soft_pairs": [ + "issue:1388|issue:4366", + "issue:2822|issue:6093", + "issue:5488|issue:8668", + "issue:5226|issue:5862", + "issue:10804|issue:9270", + "issue:1092|issue:4520", + "issue:3112|issue:5474", + "issue:3824|issue:6203", + "issue:3937|issue:4194", + "issue:4332|issue:7644", + "issue:4727|issue:5545", + "issue:5647|issue:7041", + "issue:4340|issue:5897", + "issue:6839|issue:8900", + "issue:1717|issue:3623", + "issue:3468|issue:4117", + "issue:2324|issue:817", + "issue:12195|issue:7681", + "issue:2947|issue:9142", + "issue:8567|issue:8708", + "issue:10954|issue:11464", + "issue:4183|issue:7377", + "issue:11819|issue:12120", + "issue:222|issue:4117" + ], + "prepared_review_unit_hash": "16f6dba662713b03d61034b9a7b8f9ed2c9871c8af168628a8e9677425a92a78", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33472, + "estimated_input_tokens": 8368, + "estimated_eval_tokens": 16992 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No PRs are present. Most soft pairs are clearly unrelated by title; the only mild theme overlap is among FLUX/LoRA reports, but they describe different pipelines, model variants, or training vs loading problems, so they should stay separate.", + "confidence": 0.76, + "canonical_issue_reason": "issue:10954 is the most general and current report of the FLUX1.dev LoRA loading failure, so it is the best umbrella duplicate target among the FLUX-related issues.", + "canonical_pr_reason": null, + "best_issue_reason": "Open, generic, and directly about the core symptom (FLUX1.dev LoRA fails to load), which makes it the best issue to anchor related follow-ups.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1388", + "right": "issue:4366", + "accept": false, + "reason": "Different topics: SD 2.0 conversion script vs SD-XL full U-Net fine-tuning." + }, + { + "left": "issue:2822", + "right": "issue:6093", + "accept": false, + "reason": "Training convergence complaint vs push_to_hub behavior; not the same bug." + }, + { + "left": "issue:5488", + "right": "issue:8668", + "accept": false, + "reason": "Demo image-generation failure is unrelated to SD3 img2img input-size alignment." + }, + { + "left": "issue:5226", + "right": "issue:5862", + "accept": false, + "reason": "LoRA loading in SDXL adapter pipeline vs gradient checkpointing support in Transformer2DModel." + }, + { + "left": "issue:10804", + "right": "issue:9270", + "accept": false, + "reason": "Both mention FLUX LoRA, but one is a load failure and the other is incomplete training-support for SimpleTuner." + }, + { + "left": "issue:1092", + "right": "issue:4520", + "accept": false, + "reason": "Euler/attention slicing issue is not the same as LoRA scale with attention slicing." + }, + { + "left": "issue:3112", + "right": "issue:5474", + "accept": false, + "reason": "Multi-ControlNet inpainting vs a new reference-only pipeline proposal; different changes." + }, + { + "left": "issue:3824", + "right": "issue:6203", + "accept": false, + "reason": "Legacy-vs-new inpaint pipeline question is unrelated to SDXL ControlNet image-size conditioning." + }, + { + "left": "issue:3937", + "right": "issue:4194", + "accept": false, + "reason": "ZeroscopeXL pipeline behavior vs from_single_file VAE loading; different failures." + }, + { + "left": "issue:4332", + "right": "issue:7644", + "accept": false, + "reason": "xformers on Kandinsky2.2 vs Stable Cascade non-power-of-two image handling." + }, + { + "left": "issue:4727", + "right": "issue:5545", + "accept": false, + "reason": "local_files_only auto-pipeline breakage is unrelated to an inpaint pipeline problem." + }, + { + "left": "issue:5647", + "right": "issue:7041", + "accept": false, + "reason": "Different failure modes and conditions in different pipelines; not the same underlying bug." + }, + { + "left": "issue:4340", + "right": "issue:5897", + "accept": false, + "reason": "Both are LoRA training errors, but they involve different scripts and different concrete failures." + }, + { + "left": "issue:6839", + "right": "issue:8900", + "accept": false, + "reason": "Both use from_single_file, but one is Stable Video Diffusion and the other is SDXL refiner; distinct loading paths." + }, + { + "left": "issue:1717", + "right": "issue:3623", + "accept": false, + "reason": "Offline cached-model loading is unrelated to generic large-model loading issues." + }, + { + "left": "issue:3468", + "right": "issue:4117", + "accept": false, + "reason": "Missing model_index.json is different from SDXL Dreambooth not working." + }, + { + "left": "issue:2324", + "right": "issue:817", + "accept": false, + "reason": "Instance-data loading bug vs mixed-precision issue in Dreambooth; unrelated." + }, + { + "left": "issue:12195", + "right": "issue:7681", + "accept": false, + "reason": "PyTorch custom_op incompatibility is unrelated to a lib/X509 import error." + }, + { + "left": "issue:2947", + "right": "issue:9142", + "accept": false, + "reason": "Missing load_attn_procs on UNet and multi-ControlNet __module__ error are different code paths." + }, + { + "left": "issue:8567", + "right": "issue:8708", + "accept": false, + "reason": "Training objective question for SD3 is not the same as a missing shift_factor in examples." + }, + { + "left": "issue:10954", + "right": "issue:11464", + "accept": false, + "reason": "Both are FLUX LoRA loading reports, but they target different model variants/pipelines and don't clearly share one concrete defect." + }, + { + "left": "issue:4183", + "right": "issue:7377", + "accept": false, + "reason": "ControlNet guidance-scale bug is unrelated to Stable Cascade num_images_per_prompt handling." + }, + { + "left": "issue:11819", + "right": "issue:12120", + "accept": false, + "reason": "Different FLUX LoRA-training questions: image-editing Kontext vs distilled flux-schnell." + }, + { + "left": "issue:222", + "right": "issue:4117", + "accept": false, + "reason": "Generic training runtime error is not the same as SDXL Dreambooth failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 170, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10210", + "issue:10518", + "issue:10722", + "issue:10967", + "issue:11182", + "issue:11256", + "issue:11489", + "issue:12022", + "issue:12535", + "issue:1807", + "issue:1985", + "issue:2244", + "issue:2296", + "issue:2392", + "issue:2462", + "issue:2806", + "issue:2844", + "issue:2904", + "issue:3040", + "issue:3175", + "issue:3774", + "issue:3802", + "issue:3954", + "issue:4306", + "issue:4480", + "issue:4566", + "issue:5226", + "issue:5577", + "issue:5862", + "issue:6107", + "issue:6110", + "issue:6324", + "issue:6382", + "issue:6518", + "issue:6548", + "issue:7223", + "issue:730", + "issue:7589", + "issue:760", + "issue:7644", + "issue:8228", + "issue:8305", + "issue:8309", + "issue:8588", + "issue:8950", + "issue:9142", + "issue:9156", + "issue:9673" + ], + "soft_pairs": [ + "issue:7223|issue:8309", + "issue:5577|issue:8228", + "issue:5226|issue:6110", + "issue:10210|issue:6548", + "issue:3175|issue:3774", + "issue:10722|issue:7644", + "issue:4566|issue:8950", + "issue:6107|issue:6382", + "issue:10518|issue:8588", + "issue:11489|issue:9156", + "issue:2244|issue:2844", + "issue:4306|issue:8305", + "issue:5862|issue:7589", + "issue:3802|issue:3954", + "issue:11182|issue:9673", + "issue:10967|issue:11256", + "issue:12022|issue:12535", + "issue:2392|issue:6518", + "issue:4480|issue:760", + "issue:1807|issue:2296", + "issue:1985|issue:2462", + "issue:2806|issue:730", + "issue:2904|issue:9142", + "issue:3040|issue:6324" + ], + "prepared_review_unit_hash": "9623ca72643682c7c86154ce99cf572f0bb769191964d1c7a06ec6da4047aff3", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34515, + "estimated_input_tokens": 8629, + "estimated_eval_tokens": 17514 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is highly heterogeneous: it mixes unrelated issues about img2img behavior, DreamBooth/LoRA training, model loading, scheduler math, MPS/flash-attn compatibility, and feature requests. None of the soft edges looks like the same underlying bug or change, so all are false positives.", + "confidence": 0.95, + "canonical_issue_reason": "No canonical issue: the items span many unrelated bugs/features and do not form a single duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good global representative for deduping because there is no coherent shared problem across the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7223", + "right": "issue:8309", + "accept": false, + "reason": "One is a video-generation model integration request; the other is a diffusion distillation feature request." + }, + { + "left": "issue:5577", + "right": "issue:8228", + "accept": false, + "reason": "Both involve `from_single_file`, but one is a ControlNet key error and the other is a CLIPTextModel loading failure." + }, + { + "left": "issue:5226", + "right": "issue:6110", + "accept": false, + "reason": "Both mention `load_lora_weights`, but one is an SDXL adapter compatibility bug and the other is an offline-loading problem." + }, + { + "left": "issue:10210", + "right": "issue:6548", + "accept": false, + "reason": "Both are img2img-related, but one reports near-identity output in Flux and the other is an RGBA latent-handling bug in SD/SDXL." + }, + { + "left": "issue:3175", + "right": "issue:3774", + "accept": false, + "reason": "Different training problems: multi-face DreamBooth behavior vs. Zero3 memory savings not appearing." + }, + { + "left": "issue:10722", + "right": "issue:7644", + "accept": false, + "reason": "A tensor-shape mismatch during DreamBooth training is not the same as Stable Cascade failing on non-power-of-two images." + }, + { + "left": "issue:4566", + "right": "issue:8950", + "accept": false, + "reason": "Both are LoRA training issues, but one is checkpoint resume and the other is an SD3 training bug." + }, + { + "left": "issue:6107", + "right": "issue:6382", + "accept": false, + "reason": "Both mention IP-Adapter, but one is an environment-specific app load failure and the other is a training-to-inference key mismatch." + }, + { + "left": "issue:10518", + "right": "issue:8588", + "accept": false, + "reason": "One reports a problem in a training script; the other says a conversion script is missing entirely." + }, + { + "left": "issue:11489", + "right": "issue:9156", + "accept": false, + "reason": "A generic Flux LoRA training error is not the same as a Flux LoRA CUDA OOM." + }, + { + "left": "issue:2244", + "right": "issue:2844", + "accept": false, + "reason": "Different pipelines and different failures: DreamBooth Half-precision CPU norm vs. explicit image-embedding handling in UnCLIP." + }, + { + "left": "issue:4306", + "right": "issue:8305", + "accept": false, + "reason": "A StableDiffusionReferencePipeline index error is unrelated to a scheduler `eta` range issue." + }, + { + "left": "issue:5862", + "right": "issue:7589", + "accept": false, + "reason": "Gradient-checkpointing support in Transformer2DModel is not the same as GLIGEN not working." + }, + { + "left": "issue:3802", + "right": "issue:3954", + "accept": false, + "reason": "One is about checkpoint limiting; the other is about incorrect LR scheduler step counts in multi-GPU examples." + }, + { + "left": "issue:11182", + "right": "issue:9673", + "accept": false, + "reason": "An unnecessary BF16 download is unrelated to high CPU usage when loading multiple LoRAs." + }, + { + "left": "issue:10967", + "right": "issue:11256", + "accept": false, + "reason": "These are separate feature requests for different models/methods, not the same integration." + }, + { + "left": "issue:12022", + "right": "issue:12535", + "accept": false, + "reason": "Flash-attn interface incompatibility is unrelated to Wan2.2 Lightning LoRA loading." + }, + { + "left": "issue:2392", + "right": "issue:6518", + "accept": false, + "reason": "An img2img prompt_embeds validation error is unrelated to a `NoneType // int` training crash." + }, + { + "left": "issue:4480", + "right": "issue:760", + "accept": false, + "reason": "Both are MPS-related, but one is a kernel crash with float16 and the other is incorrect output in an older version." + }, + { + "left": "issue:1807", + "right": "issue:2296", + "accept": false, + "reason": "Inference slowdown over time is not the same issue as xformers attention failing on T4." + }, + { + "left": "issue:1985", + "right": "issue:2462", + "accept": false, + "reason": "Loading embeddings and IterableDataset finetuning are different feature requests." + }, + { + "left": "issue:2806", + "right": "issue:730", + "accept": false, + "reason": "A pipeline-printing customization bug is unrelated to DreamBooth learning differently locally vs Colab." + }, + { + "left": "issue:2904", + "right": "issue:9142", + "accept": false, + "reason": "AudioLDM scheduler initialization failure and Hunyuan Dit multi-ControlNet loading are different code paths." + }, + { + "left": "issue:3040", + "right": "issue:6324", + "accept": false, + "reason": "Both mention `__module__`, but the failures come from different bad input types and different contexts." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 171, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10040", + "issue:10180", + "issue:1061", + "issue:10772", + "issue:11178", + "issue:11547", + "issue:12012", + "issue:12029", + "issue:12078", + "issue:12499", + "issue:1264", + "issue:2326", + "issue:2364", + "issue:2548", + "issue:2786", + "issue:3645", + "issue:3660", + "issue:3799", + "issue:4206", + "issue:4229", + "issue:4488", + "issue:4639", + "issue:4667", + "issue:4724", + "issue:4726", + "issue:4868", + "issue:4892", + "issue:4935", + "issue:499", + "issue:5150", + "issue:5295", + "issue:5840", + "issue:5862", + "issue:593", + "issue:6110", + "issue:6196", + "issue:6602", + "issue:6801", + "issue:6894", + "issue:7053", + "issue:7185", + "issue:7303", + "issue:7392", + "issue:7466", + "issue:7785", + "issue:9187", + "issue:9267", + "issue:9421" + ], + "soft_pairs": [ + "issue:3799|issue:4726", + "issue:4488|issue:4667", + "issue:5840|issue:6894", + "issue:7303|issue:7392", + "issue:2364|issue:499", + "issue:4868|issue:9267", + "issue:6801|issue:7785", + "issue:6110|issue:9187", + "issue:2326|issue:2548", + "issue:4229|issue:4639", + "issue:10180|issue:12078", + "issue:4206|issue:7466", + "issue:7053|issue:7185", + "issue:12012|issue:12499", + "issue:10040|issue:9421", + "issue:2786|issue:4935", + "issue:11178|issue:12029", + "issue:4892|issue:5295", + "issue:1061|issue:593", + "issue:10772|issue:11547", + "issue:1264|issue:3645", + "issue:3660|issue:4724", + "issue:5150|issue:5862", + "issue:6196|issue:6602" + ], + "prepared_review_unit_hash": "6f85aad33b48bd49fae1750d45fb681163addeb8237a485a32de106162fed483", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33994, + "estimated_input_tokens": 8499, + "estimated_eval_tokens": 17254 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a weakly related grab-bag of mostly unrelated diffusers issues across loading, training, scheduler, ControlNet, LoRA, and pipeline support. None of the proposed soft pairs look like the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue cleanly represents the cluster: the items cover many different code paths and features, so any one issue would be a misleading canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "There is no strong global best issue here; the cluster is not cohesive enough for one issue to serve as a reliable representative.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3799", + "right": "issue:4726", + "accept": false, + "reason": "Different problems: scheduler type inference vs a callback typing issue in a pipeline call signature." + }, + { + "left": "issue:4488", + "right": "issue:4667", + "accept": false, + "reason": "Both mention from_single_file, but one is SDXL checkpoint loading via ckpt conversion and the other is a generic v0.20.0 failure for StableDiffusionPipeline; not the same bug." + }, + { + "left": "issue:5840", + "right": "issue:6894", + "accept": false, + "reason": "Both concern SDXL LoRA training, but one is resume/freezing behavior and the other is A1111 compatibility after export; different failure modes." + }, + { + "left": "issue:7303", + "right": "issue:7392", + "accept": false, + "reason": "Scheduler zero-SNR support is unrelated to SDXL multi-aspect-ratio training." + }, + { + "left": "issue:2364", + "right": "issue:499", + "accept": false, + "reason": "Generic training-script failure versus unconditional training script not working; too broad and not clearly the same concrete bug." + }, + { + "left": "issue:4868", + "right": "issue:9267", + "accept": false, + "reason": "Different attribute errors in different components: a super().__getattr__ issue versus missing rope in FLUX transformer code." + }, + { + "left": "issue:6801", + "right": "issue:7785", + "accept": false, + "reason": "Scheduler loading error versus slow first inference; different symptoms and code paths." + }, + { + "left": "issue:6110", + "right": "issue:9187", + "accept": false, + "reason": "Offline LoRA loading and SDXL IP-Adapter image_encoder_folder behavior are unrelated." + }, + { + "left": "issue:2326", + "right": "issue:2548", + "accept": false, + "reason": "Both involve LoRA, but one is converting LoRA weights to safetensors/ckpt and the other is saving finetuning LoRA; not the same bug." + }, + { + "left": "issue:4229", + "right": "issue:4639", + "accept": false, + "reason": "Both are multi-GPU training errors, but they affect different scripts and failure modes." + }, + { + "left": "issue:10180", + "right": "issue:12078", + "accept": false, + "reason": "Different feature areas: loading multiple LoRAs with Flux Control LoRA versus a Flux Control finetuning validation-input example problem." + }, + { + "left": "issue:4206", + "right": "issue:7466", + "accept": false, + "reason": "ControlNet training with SDXL versus torch.compile failing for multi-ControlNet in SDXL inpaint; related subsystem, but not the same bug." + }, + { + "left": "issue:7053", + "right": "issue:7185", + "accept": false, + "reason": "Unet3D class conditioning bug is unrelated to PyTorch 2.2 support concerns." + }, + { + "left": "issue:12012", + "right": "issue:12499", + "accept": false, + "reason": "apply_first_block_cache context-state error is unrelated to Wan 2.2 I2V condition shape mismatch." + }, + { + "left": "issue:10040", + "right": "issue:9421", + "accept": false, + "reason": "Grey images in SD3 and inability to access an inpainting model are different issues." + }, + { + "left": "issue:2786", + "right": "issue:4935", + "accept": false, + "reason": "A numpy/transformers pipeline compatibility bug is unrelated to a question about AttnProcsLayers usage in an SDXL LoRA training script." + }, + { + "left": "issue:11178", + "right": "issue:12029", + "accept": false, + "reason": "MistoLine ControlNet loading and X-Omni-En model loading are different model-specific loading issues." + }, + { + "left": "issue:4892", + "right": "issue:5295", + "accept": false, + "reason": "TokenFlow video editing pipeline and LAVIE video generation are different pipeline proposals, not duplicates." + }, + { + "left": "issue:1061", + "right": "issue:593", + "accept": false, + "reason": "Textual inversion doc update versus repeated images in high-res stable diffusion; unrelated." + }, + { + "left": "issue:10772", + "right": "issue:11547", + "accept": false, + "reason": "Both request ControlNet support, but for different model families (Sana vs Flux fill dev); not the same concrete change." + }, + { + "left": "issue:1264", + "right": "issue:3645", + "accept": false, + "reason": "Hard-coded scheduler issue in dreambooth training is unrelated to a ControlNet inpaint training/help question." + }, + { + "left": "issue:3660", + "right": "issue:4724", + "accept": false, + "reason": "torch.compile with LoRA/xformers is unrelated to SDXL img2img behavior with a base checkpoint." + }, + { + "left": "issue:5150", + "right": "issue:5862", + "accept": false, + "reason": "from_single_file missing for UpscalePipeline is unrelated to Transformer2DModel gradient checkpointing support." + }, + { + "left": "issue:6196", + "right": "issue:6602", + "accept": false, + "reason": "Two different loading/saving failures: ControlNetXS Depth SDXL loading versus loading after fuse_lora/save_pretrained." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 172, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1100", + "issue:11700", + "issue:1197", + "issue:12195", + "issue:12535", + "issue:12575", + "issue:12689", + "issue:1271", + "issue:13142", + "issue:1350", + "issue:1544", + "issue:1616", + "issue:234", + "issue:2465", + "issue:2549", + "issue:2599", + "issue:2618", + "issue:2707", + "issue:3124", + "issue:3221", + "issue:3325", + "issue:3391", + "issue:3420", + "issue:347", + "issue:3660", + "issue:4080", + "issue:4185", + "issue:4325", + "issue:4664", + "issue:5103", + "issue:5379", + "issue:5442", + "issue:5544", + "issue:5905", + "issue:6492", + "issue:6722", + "issue:6911", + "issue:8532", + "issue:8543", + "issue:8555", + "issue:8726", + "issue:905", + "issue:9134", + "issue:9279", + "issue:9914", + "issue:9953", + "issue:9997" + ], + "soft_pairs": [ + "issue:8543|issue:8726", + "issue:4664|issue:5544", + "issue:1271|issue:347", + "issue:2707|issue:3325", + "issue:4185|issue:6492", + "issue:3221|issue:3660", + "issue:3420|issue:6911", + "issue:8532|issue:9914", + "issue:12535|issue:12689", + "issue:1197|issue:905", + "issue:234|issue:2549", + "issue:4325|issue:5103", + "issue:6722|issue:8555", + "issue:9134|issue:9997", + "issue:1100|issue:1544", + "issue:12195|issue:13142", + "issue:2465|issue:2599", + "issue:12575|issue:9953", + "issue:11700|issue:9279", + "issue:1350|issue:5442", + "issue:1616|issue:5379", + "issue:2618|issue:3391", + "issue:3124|issue:6911", + "issue:4080|issue:5905" + ], + "prepared_review_unit_hash": "139feee643d0f44638d1e32d1348622dc3311a7c6fe5e52c6768cacf7d6599ff", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33237, + "estimated_input_tokens": 8310, + "estimated_eval_tokens": 16876 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose topical bundle of diffusion, LoRA, pipeline, and training issues rather than true duplicates. None of the soft pairs look like the same underlying bug/change, so all are rejected.", + "confidence": 0.78, + "canonical_issue_reason": "issue:12689 is the most concrete, bug-focused report in the set: a clear loading failure for LoRA adapter weights from checkpoints with an actionable symptom statement.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:12689 is the best representative issue overall because it is specific, reproducible, and about a core failure mode (LoRA checkpoint loading) rather than a broad feature request or vague symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8543", + "right": "issue:8726", + "accept": false, + "reason": "Both involve SD3 text-encoder/training problems, but one is a gradient-checkpointing failure and the other is text encoder errors in training scripts; different concrete bugs." + }, + { + "left": "issue:4664", + "right": "issue:5544", + "accept": false, + "reason": "Both mention SDXL/LoRA-related pipeline behavior, but the symptoms differ: one is a LoRA loading inconsistency, the other is a FreeU disable error." + }, + { + "left": "issue:1271", + "right": "issue:347", + "accept": false, + "reason": "Both are dependency/install issues, but they concern different environments and failure modes; not the same underlying packaging bug." + }, + { + "left": "issue:2707", + "right": "issue:3325", + "accept": false, + "reason": "A syntax error in a conversion script and a CUDA runtime error during DreamBooth training are unrelated code paths." + }, + { + "left": "issue:4185", + "right": "issue:6492", + "accept": false, + "reason": "Both mention black images, but one is specifically SDXL ControlNet inference while the other is a generic image-generation symptom with unknown cause." + }, + { + "left": "issue:3221", + "right": "issue:3660", + "accept": false, + "reason": "These are different problems: attn processor loading in one version versus torch.compile compatibility with LoRA/xformers." + }, + { + "left": "issue:3420", + "right": "issue:6911", + "accept": false, + "reason": "Different models and failures: a ControlNet inpaint pipeline issue versus class-label handling in a 3D UNet." + }, + { + "left": "issue:8532", + "right": "issue:9914", + "accept": false, + "reason": "Both involve LoRA loading, but one is memory growth during loading and the other is a specific Flux/Xlabs loading error; not the same bug." + }, + { + "left": "issue:12535", + "right": "issue:12689", + "accept": false, + "reason": "Both are LoRA-loading complaints, but they target different model families and different failure modes, so they should not be merged." + }, + { + "left": "issue:1197", + "right": "issue:905", + "accept": false, + "reason": "These are separate community pipeline requests with different use cases; no shared underlying bug or implementation change." + }, + { + "left": "issue:234", + "right": "issue:2549", + "accept": false, + "reason": "Half-precision instability and an SD 2.0 UNet failure are not the same issue, even if both relate to Stable Diffusion models." + }, + { + "left": "issue:4325", + "right": "issue:5103", + "accept": false, + "reason": "Both produce black images, but the reported triggers differ (SDXL warning/cast path vs zero-SNR DDIM), so the underlying cause is not clearly the same." + }, + { + "left": "issue:6722", + "right": "issue:8555", + "accept": false, + "reason": "A feature request to use SigLIP as text encoder is not the same as SD3 text-encoder training availability/fp8 support." + }, + { + "left": "issue:9134", + "right": "issue:9997", + "accept": false, + "reason": "Both mention Flux LoRAs, but one is a specific incompatibility with SimpleTuner-trained LoRAs and the other is a broader support request for new LoRAs." + }, + { + "left": "issue:1100", + "right": "issue:1544", + "accept": false, + "reason": "Different attribute errors in different components: CLIPTextTransformer device access versus DDPMScheduler get_velocity." + }, + { + "left": "issue:12195", + "right": "issue:13142", + "accept": false, + "reason": "Both are version-compatibility regressions, but they involve different packages and distinct failing APIs." + }, + { + "left": "issue:2465", + "right": "issue:2599", + "accept": false, + "reason": "A request for a reproducible training recipe is not the same as a report that multi-subject DreamBooth training yields worse quality." + }, + { + "left": "issue:12575", + "right": "issue:9953", + "accept": false, + "reason": "These concern different failures: attention-backend checks in Chroma versus moving a quantized component to CUDA." + }, + { + "left": "issue:11700", + "right": "issue:9279", + "accept": false, + "reason": "One is a Pruna integration request for from_pretrained/pipeline loading, the other is dynamic LoRA loading with torch.compile; separate feature changes." + }, + { + "left": "issue:1350", + "right": "issue:5442", + "accept": false, + "reason": "Completely different domains: object detection (DiffusionDet) versus higher-resolution image generation (ScaleCrafter)." + }, + { + "left": "issue:1616", + "right": "issue:5379", + "accept": false, + "reason": "Different community pipelines with no shared bug or implementation target." + }, + { + "left": "issue:2618", + "right": "issue:3391", + "accept": false, + "reason": "A ControlNet image-input type issue and a DeepFloyd overexposure bug are unrelated pipeline problems." + }, + { + "left": "issue:3124", + "right": "issue:6911", + "accept": false, + "reason": "Stable Diffusion model conversion failure and a 3D UNet class-label bug are different code paths entirely." + }, + { + "left": "issue:4080", + "right": "issue:5905", + "accept": false, + "reason": "An SDXL inpainting pipeline request and a diffusion-GAN model request are separate feature additions, not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 173, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10928", + "issue:11247", + "issue:11351", + "issue:11584", + "issue:13258", + "issue:1363", + "issue:1616", + "issue:2163", + "issue:2192", + "issue:2250", + "issue:2707", + "issue:2838", + "issue:3177", + "issue:3648", + "issue:3881", + "issue:4520", + "issue:4835", + "issue:5233", + "issue:5509", + "issue:5578", + "issue:5606", + "issue:5753", + "issue:5783", + "issue:5862", + "issue:5966", + "issue:6323", + "issue:6351", + "issue:6427", + "issue:6549", + "issue:6561", + "issue:6791", + "issue:6934", + "issue:6974", + "issue:7549", + "issue:7592", + "issue:7619", + "issue:8233", + "issue:8493", + "issue:8518", + "issue:8732", + "issue:8832", + "issue:8909", + "issue:8958", + "issue:9084", + "issue:9422", + "issue:9534", + "issue:9927", + "issue:9979" + ], + "soft_pairs": [ + "issue:7592|issue:8493", + "issue:11584|issue:8518", + "issue:6549|issue:6561", + "issue:8732|issue:8909", + "issue:4835|issue:5606", + "issue:6351|issue:8832", + "issue:8958|issue:9534", + "issue:10928|issue:8233", + "issue:2163|issue:2707", + "issue:6427|issue:6934", + "issue:11247|issue:9979", + "issue:3648|issue:3881", + "issue:5753|issue:6323", + "issue:2250|issue:5509", + "issue:1363|issue:1616", + "issue:9422|issue:9927", + "issue:11351|issue:13258", + "issue:2838|issue:5966", + "issue:4520|issue:6974", + "issue:5233|issue:9084", + "issue:5578|issue:5862", + "issue:5783|issue:7549", + "issue:2192|issue:6791", + "issue:3177|issue:7619" + ], + "prepared_review_unit_hash": "ab8a689ba648803f9f46af465d02c934ebbd0f4633a8776e6a4802bbf9bb1b01", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34202, + "estimated_input_tokens": 8551, + "estimated_eval_tokens": 17358 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the issues span many unrelated bugs/features across model loading, training scripts, ControlNet/Flux/Wan/SD3, compilation, and file/path errors. The soft-link suggestions are mostly driven by superficial title similarity, not the same underlying problem.", + "confidence": 0.94, + "canonical_issue_reason": "No single issue is a safe canonical target; the items describe different code paths and failure modes, so collapsing them would be incorrect.", + "canonical_pr_reason": null, + "best_issue_reason": "None is a strong global representative because the set is heterogeneous rather than a near-duplicate cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7592", + "right": "issue:8493", + "accept": false, + "reason": "Both involve state_dict loading, but they target different models and failure modes (DoRA key coverage vs SD3Transformer2DModel mismatch)." + }, + { + "left": "issue:11584", + "right": "issue:8518", + "accept": false, + "reason": "Both mention torch.compile, but one is a group offloading incompatibility and the other is an SD3 dynamo unsupported-call error." + }, + { + "left": "issue:6549", + "right": "issue:6561", + "accept": false, + "reason": "Different LoRA/IP-Adapter loading problems: one is a lora_weight logging issue, the other is a missing config.json file path error." + }, + { + "left": "issue:8732", + "right": "issue:8909", + "accept": false, + "reason": "Both are training/save related, but one is checkpoint saving in SD1.5 advanced training and the other is text_encoder_ti saving in a Dreambooth LoRA script." + }, + { + "left": "issue:4835", + "right": "issue:5606", + "accept": false, + "reason": "ControlNet img2img with MultiControlNetModel is unrelated to LoRA loading with no alpha value." + }, + { + "left": "issue:6351", + "right": "issue:8832", + "accept": false, + "reason": "One is a DDP adapter attribute error in a training script; the other is a generic runtime crash on pipeline image generation." + }, + { + "left": "issue:8958", + "right": "issue:9534", + "accept": false, + "reason": "These are different PyTorch/Flux problems: a torch.library API mismatch versus FluxMultiControlNetModel missing config." + }, + { + "left": "issue:10928", + "right": "issue:8233", + "accept": false, + "reason": "Both mention fp16/model handling, but one is a missing model file and the other is a depth pipeline fp16 error." + }, + { + "left": "issue:2163", + "right": "issue:2707", + "accept": false, + "reason": "Different training/conversion failures: UNet dtype mismatch during text-to-image training vs a syntax error in a conversion script." + }, + { + "left": "issue:6427", + "right": "issue:6934", + "accept": false, + "reason": "Both are conversion bugs, but they refer to different converters and outputs (.ckpt vs original SDXL format) without evidence of the same defect." + }, + { + "left": "issue:11247", + "right": "issue:9979", + "accept": false, + "reason": "Both involve Flux controlnet/multi-GPU-ish topics, but one is transformer distribution across GPUs and the other is an img2img controlnet channels error." + }, + { + "left": "issue:3648", + "right": "issue:3881", + "accept": false, + "reason": "These are generic how-to questions about training size/epochs, not the same concrete bug." + }, + { + "left": "issue:5753", + "right": "issue:6323", + "accept": false, + "reason": "Different ControlNet/Lora integration issues: one is load_lora_weights failure, the other is a TensorRT img2img guidance_scale attribute problem." + }, + { + "left": "issue:2250", + "right": "issue:5509", + "accept": false, + "reason": "Dreambooth prompt-generation failure is unrelated to image batch-size validation for np.ndarray/cv2 inputs." + }, + { + "left": "issue:1363", + "right": "issue:1616", + "accept": false, + "reason": "Both concern LPW/community pipelines, but one is an ONNX scheduler error and the other is a general Diffusers version/support question." + }, + { + "left": "issue:9422", + "right": "issue:9927", + "accept": false, + "reason": "Different training instability bugs: one is NaNs during custom ControlNet SDXL backprop, the other is HeaderTooLarge when training controlnet with SD3." + }, + { + "left": "issue:11351", + "right": "issue:13258", + "accept": false, + "reason": "Both mention Wan video models, but one is a datatype question for a processor and the other is a missing Wan2.2 TI2V model/package issue." + }, + { + "left": "issue:2838", + "right": "issue:5966", + "accept": false, + "reason": "One asks for ControlNet pipeline behavior; the other is a request for Stable Diffusion 2.1 ControlNet support." + }, + { + "left": "issue:4520", + "right": "issue:6974", + "accept": false, + "reason": "Both are adapter-related, but one is a LoRA scale + attention slicing bug and the other is textual inversion unload missing tokenizer/text encoder cleanup." + }, + { + "left": "issue:5233", + "right": "issue:9084", + "accept": false, + "reason": "Different scheduler/callback semantics: wrong callback step value vs SD3 inpaint EulerA scale_model_input ordering." + }, + { + "left": "issue:5578", + "right": "issue:5862", + "accept": false, + "reason": "A missing model_index.json load error is unrelated to Transformer2DModel gradient checkpointing support." + }, + { + "left": "issue:5783", + "right": "issue:7549", + "accept": false, + "reason": "Both are SDXL training issues, but one is EMA not working in the training script and the other is a suspected UNet EMA bug; not clearly the same defect." + }, + { + "left": "issue:2192", + "right": "issue:6791", + "accept": false, + "reason": "Different training-script issues: missing CLI arguments in one script vs an SDXL training script that cannot run." + }, + { + "left": "issue:3177", + "right": "issue:7619", + "accept": false, + "reason": "DeepSpeed stage-3 offloading in Dreambooth is unrelated to a channel-number mismatch in advanced Dreambooth training." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 174, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10550", + "issue:11133", + "issue:11447", + "issue:11702", + "issue:1205", + "issue:12359", + "issue:1249", + "issue:1552", + "issue:1974", + "issue:2091", + "issue:2462", + "issue:255", + "issue:2581", + "issue:2991", + "issue:3019", + "issue:3103", + "issue:3191", + "issue:3445", + "issue:3540", + "issue:4000", + "issue:4892", + "issue:4974", + "issue:4993", + "issue:5023", + "issue:5152", + "issue:5432", + "issue:5687", + "issue:5889", + "issue:5966", + "issue:6157", + "issue:6382", + "issue:6407", + "issue:6811", + "issue:6949", + "issue:7033", + "issue:7079", + "issue:7255", + "issue:730", + "issue:761", + "issue:7616", + "issue:7948", + "issue:8732", + "issue:8738", + "issue:8873", + "issue:9454", + "issue:9546", + "issue:9763" + ], + "soft_pairs": [ + "issue:5687|issue:7616", + "issue:2091|issue:8732", + "issue:1205|issue:4974", + "issue:2462|issue:5966", + "issue:6811|issue:9454", + "issue:7079|issue:8873", + "issue:1249|issue:1974", + "issue:11447|issue:8873", + "issue:255|issue:3103", + "issue:3019|issue:6407", + "issue:5432|issue:6949", + "issue:7255|issue:7948", + "issue:2991|issue:761", + "issue:9546|issue:9763", + "issue:5023|issue:5152", + "issue:10550|issue:11702", + "issue:3540|issue:6157", + "issue:1552|issue:730", + "issue:6382|issue:7033", + "issue:2581|issue:4000", + "issue:4892|issue:5889", + "issue:3191|issue:3445", + "issue:4993|issue:5023", + "issue:10026|issue:8738", + "issue:11133|issue:12359" + ], + "prepared_review_unit_hash": "d3f74e51f642801bf80d5f4d59b5e49d9571652274ea83024ba44c94eeb9244b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34226, + "estimated_input_tokens": 8557, + "estimated_eval_tokens": 17370 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is largely heterogeneous: it mixes unrelated model-loading, scheduler, documentation, and training bugs. The only loose thematic concentration is around Flux/ControlNet training, but the specific failures differ enough that none of the soft pairs should be merged as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "issue:9546 is the best umbrella issue for the Flux ControlNet train-example memory-related reports in this set; it is broader and more representative than the more specific token-truncation or validation/OOM variants.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9546 has the clearest cluster-central topic and the most discussion among the Flux ControlNet training issues, so it is the best candidate to anchor that subtopic even though the set is not a true duplicate cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5687", + "right": "issue:7616", + "accept": false, + "reason": "Both mention bad img2img behavior, but one is a scheduler duplicate-timestep bug and the other is low-denoise artifacts in a specific model; not the same concrete failure." + }, + { + "left": "issue:2091", + "right": "issue:8732", + "accept": false, + "reason": "Checkpoint-resume failures and checkpoint-saving issues are related only broadly; they describe different training paths and different breakages." + }, + { + "left": "issue:1205", + "right": "issue:4974", + "accept": false, + "reason": "These are opposite conversion directions (SD to ONNX vs safetensors to diffusers) and are not the same change or bug." + }, + { + "left": "issue:2462", + "right": "issue:5966", + "accept": false, + "reason": "IterableDataset training support and ControlNet SD2.1 support are unrelated feature/usage topics." + }, + { + "left": "issue:6811", + "right": "issue:9454", + "accept": false, + "reason": "Both are 'model not found' reports, but for different models and likely different repository/config issues." + }, + { + "left": "issue:7079", + "right": "issue:8873", + "accept": false, + "reason": "Transformer patch/norm errors and xformers performance are distinct problems with no shared code-path evidence." + }, + { + "left": "issue:1249", + "right": "issue:1974", + "accept": false, + "reason": "ONNX pipeline ndarray device errors and Euler scheduler generator-list handling are unrelated." + }, + { + "left": "issue:11447", + "right": "issue:8873", + "accept": false, + "reason": "Offline checkpoint shard loading and xformers speed regressions are unrelated subsystems and failures." + }, + { + "left": "issue:255", + "right": "issue:3103", + "accept": false, + "reason": "Both concern image dimensions, but one is a Stable Diffusion unet_blocks error and the other is an inpainting height/width topic; not enough to be the same bug." + }, + { + "left": "issue:3019", + "right": "issue:6407", + "accept": false, + "reason": "These are different research-feature requests for text/image editing, not the same underlying issue." + }, + { + "left": "issue:5432", + "right": "issue:6949", + "accept": false, + "reason": "Both mention blank outputs, but one is blank validation images in a LoRA training script and the other is a scheduler-specific blank-image bug." + }, + { + "left": "issue:7255", + "right": "issue:7948", + "accept": false, + "reason": "Completely different model/proposal topics; no shared bug or change." + }, + { + "left": "issue:2991", + "right": "issue:761", + "accept": false, + "reason": "Mixed-precision ControlNet example failure and 8-bit Adam DreamBooth failure are different training configuration bugs." + }, + { + "left": "issue:9546", + "right": "issue:9763", + "accept": false, + "reason": "Both are Flux ControlNet training issues, but one is validation-step OOM while the other is token truncation beyond 77 tokens; different concrete bugs." + }, + { + "left": "issue:5023", + "right": "issue:5152", + "accept": false, + "reason": "Both are W\u00fcrstchen-related, but one is an import error in 0.21.0 and the other is a different bug in 0.21.3 alpha; not clearly the same issue." + }, + { + "left": "issue:10550", + "right": "issue:11702", + "accept": false, + "reason": "Both involve LoRA loading, but one is quantized-base-model loading and the other is an AccVid-specific key error; different code paths." + }, + { + "left": "issue:3540", + "right": "issue:6157", + "accept": false, + "reason": "TextToVideoZero prompt-embedding support and ControlNetImg2Img IP-Adapter behavior are unrelated pipeline bugs." + }, + { + "left": "issue:1552", + "right": "issue:730", + "accept": false, + "reason": "Both are DreamBooth/training-related, but the reported failures and conditions differ substantially." + }, + { + "left": "issue:6382", + "right": "issue:7033", + "accept": false, + "reason": "A training/loading key error with IP-Adapter and IP-Adapter attention masking warnings are not the same bug." + }, + { + "left": "issue:2581", + "right": "issue:4000", + "accept": false, + "reason": "ControlNet conversion support and SD2.0 ckpt-to-diffusers conversion are adjacent but not the same conversion task or failure." + }, + { + "left": "issue:4892", + "right": "issue:5889", + "accept": false, + "reason": "TokenFlow and Stable Video Diffusion are distinct video-generation features with no concrete duplicate evidence." + }, + { + "left": "issue:3191", + "right": "issue:3445", + "accept": false, + "reason": "DreamBooth+LoRA broken and text-encoder monkey-patching not working are related only at a very high level; different failure modes." + }, + { + "left": "issue:4993", + "right": "issue:5023", + "accept": false, + "reason": "CLIPImageProjection import-structure exposure and W\u00fcrstchenCombinedPipeline import error are separate import issues affecting different symbols." + }, + { + "left": "issue:10026", + "right": "issue:8738", + "accept": false, + "reason": "Flux ControlNet memory behavior and SD3Transformer2DModel configuration are different training/model-setup issues." + }, + { + "left": "issue:11133", + "right": "issue:12359", + "accept": false, + "reason": "CogVideoX image-to-video pipeline bug and Chroma guidance_scale documentation bug are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 175, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10040", + "issue:10428", + "issue:1100", + "issue:11006", + "issue:11104", + "issue:11136", + "issue:11359", + "issue:11410", + "issue:11830", + "issue:12120", + "issue:1536", + "issue:1695", + "issue:1802", + "issue:2089", + "issue:2107", + "issue:2179", + "issue:2495", + "issue:2695", + "issue:3213", + "issue:4030", + "issue:4186", + "issue:4321", + "issue:4752", + "issue:4892", + "issue:4936", + "issue:5227", + "issue:5274", + "issue:6090", + "issue:6258", + "issue:6385", + "issue:6540", + "issue:6549", + "issue:6594", + "issue:6858", + "issue:7002", + "issue:7731", + "issue:7950", + "issue:7974", + "issue:8182", + "issue:8472", + "issue:8518", + "issue:8551", + "issue:858", + "issue:8879", + "issue:8958", + "issue:9630", + "issue:9889", + "issue:9979" + ], + "soft_pairs": [ + "issue:6540|issue:6858", + "issue:11359|issue:12120", + "issue:11830|issue:4321", + "issue:1802|issue:4186", + "issue:2107|issue:2179", + "issue:6385|issue:8551", + "issue:11136|issue:9889", + "issue:10040|issue:6258", + "issue:2695|issue:9630", + "issue:2495|issue:4321", + "issue:11410|issue:7950", + "issue:6549|issue:7974", + "issue:4936|issue:8472", + "issue:5227|issue:8879", + "issue:11006|issue:11104", + "issue:1536|issue:3213", + "issue:4752|issue:5274", + "issue:8182|issue:8958", + "issue:8518|issue:8879", + "issue:1100|issue:858", + "issue:4892|issue:7731", + "issue:10428|issue:9979", + "issue:1695|issue:2089", + "issue:4030|issue:6594", + "issue:6090|issue:7002" + ], + "prepared_review_unit_hash": "45ad2ebb03a8073b9415e5615cc95178bc39ae6ba7620445d8e7f4e343e4092a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34795, + "estimated_input_tokens": 8699, + "estimated_eval_tokens": 17654 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of unrelated issues that only share broad symptom patterns (black/grey images, pipeline errors, training/runtime exceptions). None of the soft pairs look like safe duplicates under a strict same-code-path standard.", + "confidence": 0.37, + "canonical_issue_reason": "issue:10040 is the most general and representative visible-output corruption report in the set, with decent discussion and references, so it is the best canonical stand-in for the image-artifact subset.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10040 is concrete, recent, and broadly framed enough to anchor triage for the image-generation artifact reports; it is more representative than the highly specific one-off errors.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6540", + "right": "issue:6858", + "accept": false, + "reason": "Different failures: FSDP matrix-shape error in ControlNet fine-tuning vs a string/convert AttributeError during controlnet training." + }, + { + "left": "issue:11359", + "right": "issue:12120", + "accept": false, + "reason": "Feature request about LTX-Video speed vs a how-to question about training LoRA on a distilled Flux model; not the same bug." + }, + { + "left": "issue:11830", + "right": "issue:4321", + "accept": false, + "reason": "Both mention black output, but the pipelines and backends differ (FramePack float16 vs ONNX StableDiffusion DirectML), so this is too broad to merge as one bug." + }, + { + "left": "issue:1802", + "right": "issue:4186", + "accept": false, + "reason": "Different code paths and symptoms: encoder_hidden_states=None on unet_2d_condition vs SDXL inpaint/refiner behavior." + }, + { + "left": "issue:2107", + "right": "issue:2179", + "accept": false, + "reason": "Low-resolution glitch/black images vs FlaxDDPMScheduler black images; same broad symptom, but not clearly the same underlying defect." + }, + { + "left": "issue:6385", + "right": "issue:8551", + "accept": false, + "reason": "VRAM usage issue vs device placement bug for image latents; unrelated underlying problems." + }, + { + "left": "issue:11136", + "right": "issue:9889", + "accept": false, + "reason": "SanaSprint timestep support limitation vs SDXL-turbo segfault; different models and different failure modes." + }, + { + "left": "issue:10040", + "right": "issue:6258", + "accept": false, + "reason": "Grey image generation in SD3 vs quality degradation from forward chunking in Stable Video Diffusion; not the same bug." + }, + { + "left": "issue:2695", + "right": "issue:9630", + "accept": false, + "reason": "Tensorboard validation failure during training vs flux-controlnet num_single_layers handling on single GPU; unrelated." + }, + { + "left": "issue:2495", + "right": "issue:4321", + "accept": false, + "reason": "Both produce black images, but they are distinct pipelines/backends (MPS upscale vs ONNX DirectML), so they should not be merged." + }, + { + "left": "issue:11410", + "right": "issue:7950", + "accept": false, + "reason": "Attention unpacking ValueError vs PyTorch 2.3 compatibility issue; not the same concrete code-path." + }, + { + "left": "issue:6549", + "right": "issue:7974", + "accept": false, + "reason": "Warning-log typo/reporting issue vs IP-Adapter inpaint bug; different symptoms and fixes." + }, + { + "left": "issue:4936", + "right": "issue:8472", + "accept": false, + "reason": "Dreambooth resume regression vs a question about training LoRA with other modules; not duplicate bug reports." + }, + { + "left": "issue:5227", + "right": "issue:8879", + "accept": false, + "reason": "LoRA loading with torch.compile vs torch.compile+cudagraph failure on SDXL; related theme, but not the same underlying defect." + }, + { + "left": "issue:11006", + "right": "issue:11104", + "accept": false, + "reason": "Wan 2.1 I2V quantized-transformer video corruption vs LTX-Video offload methods broken; different pipelines and failure mechanisms." + }, + { + "left": "issue:1536", + "right": "issue:3213", + "accept": false, + "reason": "Both concern lpw_stable_diffusion, but one is a compatibility report and the other is a later breakage report; insufficient evidence they are the same bug." + }, + { + "left": "issue:4752", + "right": "issue:5274", + "accept": false, + "reason": "SDXL bias/input dtype mismatch in inference vs mixed-precision training dtype error; similar dtype flavor, but different code paths and likely different fixes." + }, + { + "left": "issue:8182", + "right": "issue:8958", + "accept": false, + "reason": "Logger API typo vs torch.library.register_fake missing in PyTorch 2.4; completely unrelated." + }, + { + "left": "issue:8518", + "right": "issue:8879", + "accept": false, + "reason": "SD3 torch.compile unsupported UserDefinedClassVariable vs SDXL torch.compile+cudagraph failure; different models and compiler failure sites." + }, + { + "left": "issue:1100", + "right": "issue:858", + "accept": false, + "reason": "CLIPTextTransformer device attribute issue vs Parameter.named_children AttributeError; different API breakages." + }, + { + "left": "issue:4892", + "right": "issue:7731", + "accept": false, + "reason": "Distinct research/pipeline features (TokenFlow vs StreamingT2V), not the same bug." + }, + { + "left": "issue:10428", + "right": "issue:9979", + "accept": false, + "reason": "Both mention Flux, but one is an Ascend NPU inference error and the other is a controlnet channels mismatch; different problems." + }, + { + "left": "issue:1695", + "right": "issue:2089", + "accept": false, + "reason": "Dreambooth multi-GPU/deepspeed training request vs a CUDA OOM in AltDiffusion DreamBooth; unrelated." + }, + { + "left": "issue:4030", + "right": "issue:6594", + "accept": false, + "reason": "Stable Diffusion embedding runtime error vs network failure fetching a remote config file; not duplicate issues." + }, + { + "left": "issue:6090", + "right": "issue:7002", + "accept": false, + "reason": "Channel-count mismatch in a model layer vs an invalid buffer size memory error; different root causes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 176, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1013", + "issue:10165", + "issue:10223", + "issue:1049", + "issue:10492", + "issue:10512", + "issue:1061", + "issue:1070", + "issue:10804", + "issue:10866", + "issue:10940", + "issue:11014", + "issue:11062", + "issue:11215", + "issue:11464", + "issue:11648", + "issue:11853", + "issue:2234", + "issue:2265", + "issue:259", + "issue:2880", + "issue:3149", + "issue:3382", + "issue:3412", + "issue:3566", + "issue:3595", + "issue:3886", + "issue:4398", + "issue:5064", + "issue:5075", + "issue:5128", + "issue:5215", + "issue:5258", + "issue:5433", + "issue:5630", + "issue:6154", + "issue:6297", + "issue:6741", + "issue:7168", + "issue:7516", + "issue:7982", + "issue:8832", + "issue:9136", + "issue:9208", + "issue:9525", + "issue:9558", + "issue:987" + ], + "soft_pairs": [ + "issue:7168|issue:8832", + "issue:2265|issue:3886", + "issue:10492|issue:11062", + "issue:11014|issue:3382", + "issue:10223|issue:4398", + "issue:5128|issue:9208", + "issue:5215|issue:5630", + "issue:3566|issue:987", + "issue:5630|issue:9558", + "issue:11853|issue:9136", + "issue:6154|issue:6741", + "issue:3595|issue:7982", + "issue:2234|issue:6297", + "issue:5258|issue:5433", + "issue:10165|issue:10866", + "issue:5064|issue:5075", + "issue:1013|issue:3149", + "issue:1049|issue:3412", + "issue:10512|issue:10804", + "issue:1061|issue:2880", + "issue:1070|issue:259", + "issue:10940|issue:7516", + "issue:11215|issue:9525", + "issue:11464|issue:11648" + ], + "prepared_review_unit_hash": "cfceafada8fdff137a5554f74a1b7eff8f5057a5e8bdacf9ffb372ba3c516474", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33213, + "estimated_input_tokens": 8304, + "estimated_eval_tokens": 16864 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: LoRA/Flux loading and scaling issues appear alongside unrelated training, scheduler, docs, and pipeline bugs. None of the soft pairs look like true duplicates; they\u2019re only loosely similar by subsystem.", + "confidence": 0.96, + "canonical_issue_reason": "Issue 9525 is the clearest concrete bug report in the densest recurring theme (Flux/LoRA behavior), but the overall cluster is too mixed to treat it as a real duplicate anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9525 is the most actionable standalone report: specific title, clear symptom, and a focused code-path (Flux LoRA scale handling).", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7168", + "right": "issue:8832", + "accept": false, + "reason": "Different problems: IP-Adapter image embedding tensor shape vs a generic pipeline crash on `pipe(prompt)`." + }, + { + "left": "issue:2265", + "right": "issue:3886", + "accept": false, + "reason": "Different failures: ONNXRuntime incompatibility in a training script vs distributed collective timeout during ddpm training." + }, + { + "left": "issue:10492", + "right": "issue:11062", + "accept": false, + "reason": "Related to quantization/LoRA, but one is a torchao quantize fuse error and the other is a CivitAI LoRA load error in a different setup." + }, + { + "left": "issue:11014", + "right": "issue:3382", + "accept": false, + "reason": "Both mention OOM, but the contexts differ: inference on Colab A100 vs multi-GPU training memory usage." + }, + { + "left": "issue:10223", + "right": "issue:4398", + "accept": false, + "reason": "Completely different topics: a question about a LoRA ID for inference vs all-black images in an SDXL LoRA dreambooth pipeline." + }, + { + "left": "issue:5128", + "right": "issue:9208", + "accept": false, + "reason": "Both involve loading files/models, but one is a local `.safetensors` load and the other is ControlNet loading from single-file checkpoints." + }, + { + "left": "issue:5215", + "right": "issue:5630", + "accept": false, + "reason": "Both are image-quality regressions, but one is brightness shift in upscale and the other is leftover noise in SDXL img2img with LMS." + }, + { + "left": "issue:3566", + "right": "issue:987", + "accept": false, + "reason": "Both report unexpected output colors/artifacts, but the pipelines and symptoms are different (ControlNet reference vs speech-to-image green output)." + }, + { + "left": "issue:5630", + "right": "issue:9558", + "accept": false, + "reason": "Different manifestations: noise leftover in SDXL img2img vs black validation images in tensorboard." + }, + { + "left": "issue:11853", + "right": "issue:9136", + "accept": false, + "reason": "Not the same bug: one is a controlnet+IP-Adapter dimension conflict, the other is a general quality issue at some resolutions." + }, + { + "left": "issue:6154", + "right": "issue:6741", + "accept": false, + "reason": "Different training problems: SDXL LoRA training on M2 Mac vs a missing `no_flip` argument in a Diffusion DPO script." + }, + { + "left": "issue:3595", + "right": "issue:7982", + "accept": false, + "reason": "Both are `from_*`/loading-related, but one is `from_ckpt` failure and the other is `from_pipe` with MultiControlNetModel." + }, + { + "left": "issue:2234", + "right": "issue:6297", + "accept": false, + "reason": "Both concern training, but one is xformers memory-efficient attention and the other is SDXL LoRA training on photos." + }, + { + "left": "issue:5258", + "right": "issue:5433", + "accept": false, + "reason": "Different SDXL/refiner issues: controlnet inpaint with refiner vs DPM++ artifacts without refiner." + }, + { + "left": "issue:10165", + "right": "issue:10866", + "accept": false, + "reason": "Both mention LoRA, but one is seed preservation when changing scale and the other is a Lumina/Civitai compatibility problem." + }, + { + "left": "issue:5064", + "right": "issue:5075", + "accept": false, + "reason": "Both involve SDXL refiner usage, but one is out-of-space inference and the other is a prompt-embedding shape mismatch." + }, + { + "left": "issue:1013", + "right": "issue:3149", + "accept": false, + "reason": "Different core bugs: a specific StableDiffusionMegaPipeline compatibility issue vs nondeterministic image generation." + }, + { + "left": "issue:1049", + "right": "issue:3412", + "accept": false, + "reason": "Both are docs-related, but they request different documentation updates (Euler schedulers vs stable diffusion next steps)." + }, + { + "left": "issue:10512", + "right": "issue:10804", + "accept": false, + "reason": "Related to FLUX LoRA loading, but one is Quanto Flux LoRA load failure and the other is a ComfyUI-compatible FLUX1.dev LoRA load failure." + }, + { + "left": "issue:1061", + "right": "issue:2880", + "accept": false, + "reason": "Different content entirely: a text inversion docs update vs a JAX Colab runtime failure." + }, + { + "left": "issue:1070", + "right": "issue:259", + "accept": false, + "reason": "Different pipelines and symptoms: inpainting bug in new SD vs progress bar behavior in Stable Diffusion." + }, + { + "left": "issue:10940", + "right": "issue:7516", + "accept": false, + "reason": "Unrelated import/runtime issues: missing `triton.ops` vs missing `DepthNormalEstimationPipeline` symbol." + }, + { + "left": "issue:11215", + "right": "issue:9525", + "accept": false, + "reason": "Both touch LoRA and quantization/Flux, but one is about slow quantization and ineffective LoRA support, not the same as `lora_scale` having no effect." + }, + { + "left": "issue:11464", + "right": "issue:11648", + "accept": false, + "reason": "Both are Flux/LoRA-related, but one is loading a specific Canny-dev LoRA in FluxControlPipeline and the other is fp8 transformer LoRA weight handling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 177, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1185", + "issue:11964", + "issue:1304", + "issue:1305", + "issue:1500", + "issue:1552", + "issue:1572", + "issue:1807", + "issue:2053", + "issue:2093", + "issue:2218", + "issue:2361", + "issue:2437", + "issue:2462", + "issue:2599", + "issue:2645", + "issue:2689", + "issue:2776", + "issue:2783", + "issue:2785", + "issue:2844", + "issue:2991", + "issue:3175", + "issue:3583", + "issue:3803", + "issue:3811", + "issue:383", + "issue:4064", + "issue:4080", + "issue:4117", + "issue:4376", + "issue:4868", + "issue:4933", + "issue:5014", + "issue:5153", + "issue:528", + "issue:540", + "issue:5595", + "issue:6158", + "issue:660", + "issue:668", + "issue:684", + "issue:7428", + "issue:776", + "issue:905", + "issue:9178", + "issue:9637" + ], + "soft_pairs": [ + "issue:1185|issue:2599", + "issue:11964|issue:9178", + "issue:1304|issue:2093", + "issue:1305|issue:5153", + "issue:1500|issue:2053", + "issue:1552|issue:1807", + "issue:1572|issue:4933", + "issue:2218|issue:5014", + "issue:2361|issue:4376", + "issue:2437|issue:4064", + "issue:2462|issue:540", + "issue:2645|issue:660", + "issue:2689|issue:776", + "issue:2776|issue:5595", + "issue:2783|issue:905", + "issue:2785|issue:4080", + "issue:2844|issue:668", + "issue:2991|issue:9637", + "issue:3175|issue:3803", + "issue:3583|issue:7428", + "issue:3811|issue:4868", + "issue:383|issue:528", + "issue:4080|issue:6158", + "issue:4117|issue:684" + ], + "prepared_review_unit_hash": "d09b840b81656e8e5645999b5000d223f5c3a87dea35c61653624fd27316622b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32722, + "estimated_input_tokens": 8181, + "estimated_eval_tokens": 16618 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a broad false positive: it mixes Dreambooth/LoRA training bugs, pipeline feature requests, platform-specific errors, and unrelated questions. None of the soft-edge pairs look like the same underlying issue or change.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 4117 is the most substantial bug report in the training-related subset and the closest thing to a cluster anchor, but it is still distinct from the other Dreambooth/LoRA reports.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 4117 is the strongest representative artifact: detailed repro, clear failure mode, and the broadest relevance among the training-bug issues in this mixed set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1185", + "right": "issue:2599", + "accept": false, + "reason": "Both concern Dreambooth results, but one is about reproducibility and the other about multi-subject quality degradation; different bugs." + }, + { + "left": "issue:11964", + "right": "issue:9178", + "accept": false, + "reason": "Both mention Flux LoRA, but one is a load-time missing-weights KeyError and the other is a training target-module error." + }, + { + "left": "issue:1304", + "right": "issue:2093", + "accept": false, + "reason": "These are both image-to-image related, but one is a how-to/training question and the other is a generic model-topic issue." + }, + { + "left": "issue:1305", + "right": "issue:5153", + "accept": false, + "reason": "Direct Inversion and FreeU are unrelated research/features; no shared bug or change path." + }, + { + "left": "issue:1500", + "right": "issue:2053", + "accept": false, + "reason": "One is a community pipeline incompatibility, the other is a VRAM regression; different symptoms and code paths." + }, + { + "left": "issue:1552", + "right": "issue:1807", + "accept": false, + "reason": "Attention slicing breakage is not the same as gradually slowing inference; different performance issues." + }, + { + "left": "issue:1572", + "right": "issue:4933", + "accept": false, + "reason": "Img2img/inpainting pipeline failure is unrelated to Dreambooth checkpoint export problems." + }, + { + "left": "issue:2218", + "right": "issue:5014", + "accept": false, + "reason": "Both describe poor training outcomes, but one is generic overtraining and the other is specifically LoRA results not showing up." + }, + { + "left": "issue:2361", + "right": "issue:4376", + "accept": false, + "reason": "Checkpoint-save crashing and textual inversion crashing are different failures in different scripts/pipelines." + }, + { + "left": "issue:2437", + "right": "issue:4064", + "accept": false, + "reason": "Both are pipeline feature requests, but for different pipeline types and different functionality." + }, + { + "left": "issue:2462", + "right": "issue:540", + "accept": false, + "reason": "IterableDataset support and GPU memory usage are unrelated training concerns." + }, + { + "left": "issue:2645", + "right": "issue:660", + "accept": false, + "reason": "Both involve Apple hardware, but one is a ControlNet CUDA assertion and the other is fp16/MPS execution." + }, + { + "left": "issue:2689", + "right": "issue:776", + "accept": false, + "reason": "A broken image link is not the same as an img2img batch-size bug." + }, + { + "left": "issue:2776", + "right": "issue:5595", + "accept": false, + "reason": "DeepSpeed/ZeRO runtime failure is unrelated to invalid image size dimensions." + }, + { + "left": "issue:2783", + "right": "issue:905", + "accept": false, + "reason": "Both mention img2img/inpainting, but one is a ControlNet pipeline request and the other is a broader community pipeline question." + }, + { + "left": "issue:2785", + "right": "issue:4080", + "accept": false, + "reason": "M1/M2 text-to-video support and SDXL inpainting are different feature requests for different pipelines." + }, + { + "left": "issue:2844", + "right": "issue:668", + "accept": false, + "reason": "StableUnCLIP image-embedding passing and Dreambooth on Colab are unrelated runtime errors." + }, + { + "left": "issue:2991", + "right": "issue:9637", + "accept": false, + "reason": "Both touch attention/mixed precision, but they affect different mechanisms: ControlNet mixed-precision vs XFormers attention-mask/bfloat16 edge case." + }, + { + "left": "issue:3175", + "right": "issue:3803", + "accept": false, + "reason": "Multi-face Dreambooth training and multi-GPU LoRA synchronization are different training problems." + }, + { + "left": "issue:3583", + "right": "issue:7428", + "accept": false, + "reason": "These are separate feature requests for different diffusion variants; no same implementation target." + }, + { + "left": "issue:3811", + "right": "issue:4868", + "accept": false, + "reason": "LoRA load having no effect on generation is not the same as a generic AttributeError in __getattr__." + }, + { + "left": "issue:383", + "right": "issue:528", + "accept": false, + "reason": "Both are conceptual questions about model classes, but they are about different frameworks and different topics." + }, + { + "left": "issue:4080", + "right": "issue:6158", + "accept": false, + "reason": "SDXL inpainting and animatediff SDXL support are separate pipeline features." + }, + { + "left": "issue:4117", + "right": "issue:684", + "accept": false, + "reason": "Dreambooth failing on SDXL is unrelated to AMD training memory optimization issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 178, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:12369", + "issue:12457", + "issue:12480", + "issue:12481", + "issue:12482", + "issue:12483", + "issue:2453", + "issue:4152", + "issue:4241", + "issue:4274", + "issue:4316", + "issue:4340", + "issue:4405", + "issue:4460", + "issue:4727", + "issue:4773", + "issue:4806", + "issue:5004", + "issue:5150", + "issue:5330", + "issue:5371", + "issue:5392", + "issue:5405", + "issue:5432", + "issue:5577", + "issue:5932", + "issue:6218", + "issue:6297", + "issue:6388", + "issue:6423", + "issue:6439", + "issue:6791", + "issue:6865", + "issue:7002", + "issue:7009", + "issue:7142", + "issue:7146", + "issue:7592", + "issue:7877", + "issue:8472", + "issue:8532", + "issue:9393", + "issue:9568", + "issue:9611", + "issue:9645", + "issue:9767", + "issue:986" + ], + "soft_pairs": [ + "issue:4152|issue:7877", + "issue:4274|issue:5371", + "issue:4316|issue:5432", + "issue:4340|issue:8472", + "issue:4405|issue:6388", + "issue:4460|issue:7142", + "issue:4727|issue:5577", + "issue:4773|issue:8532", + "issue:4806|issue:5330", + "issue:5004|issue:6439", + "issue:5150|issue:5405", + "issue:5392|issue:986", + "issue:5932|issue:7002", + "issue:6218|issue:7592", + "issue:6297|issue:9393", + "issue:6791|issue:7146", + "issue:6865|issue:9645", + "issue:7009|issue:9568", + "issue:9611|issue:9767", + "issue:12481|issue:12483", + "issue:12482|issue:12483", + "issue:12480|issue:12483", + "issue:2453|issue:6423", + "issue:2453|issue:4241", + "issue:12369|issue:12457" + ], + "prepared_review_unit_hash": "4e391ebdd7279f255cdf9d73a62083af3a4483abf58579417381147f78b3473e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33140, + "estimated_input_tokens": 8285, + "estimated_eval_tokens": 16826 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly unrelated issue reports, except for a tight duplicate set about `offload_state_dict` / `CLIPTextModel` / the diffusers-transformers dependency conflict. I would use issue 12483 as the canonical issue for that duplicate group.", + "confidence": 0.95, + "canonical_issue_reason": "12483 is the most complete and highest-level report in the duplicate set: it frames the problem as a diffusers 0.35.1 vs transformers 4.57.0 dependency conflict causing the `offload_state_dict` TypeError, while 12480-12482 are narrower near-duplicate titles of the same failure.", + "canonical_pr_reason": null, + "best_issue_reason": "12483 is the best triage anchor because it captures the underlying compatibility problem, not just the surface CLIPTextModel constructor error.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4152", + "right": "issue:7877", + "accept": false, + "reason": "Both concern attention APIs, but one asks for mixed attention kwargs per processor and the other for an `out_dim` option on Attention; different feature requests." + }, + { + "left": "issue:4274", + "right": "issue:5371", + "accept": false, + "reason": "Both mention BLIP, but one requests BLIP Diffusion support and the other BLIP inpainting pipeline support; separate additions." + }, + { + "left": "issue:4316", + "right": "issue:5432", + "accept": false, + "reason": "Both are training-script validation issues, but they affect different scripts and symptoms, so they are not the same bug." + }, + { + "left": "issue:4340", + "right": "issue:8472", + "accept": false, + "reason": "One is an SDXL LoRA training error and the other is a general question about training LoRA with other modules; not the same concrete problem." + }, + { + "left": "issue:4405", + "right": "issue:6388", + "accept": false, + "reason": "Textual inversion script request and gradient-checkpointing ignored in SDXL LoRA training are unrelated changes." + }, + { + "left": "issue:4460", + "right": "issue:7142", + "accept": false, + "reason": "Both compare diffusers output against another implementation, but they involve different models and failure modes." + }, + { + "left": "issue:4727", + "right": "issue:5577", + "accept": false, + "reason": "One is about `local_files_only` auto pipelines; the other is a `from_single_file` ControlNet key error. Different code paths." + }, + { + "left": "issue:4773", + "right": "issue:8532", + "accept": false, + "reason": "Both involve LoRA loading, but one is about loading a specific CivitAI LoRA and the other about progressive memory growth; different bugs." + }, + { + "left": "issue:4806", + "right": "issue:5330", + "accept": false, + "reason": "Both touch UNet2DModel, but one is a parameter type issue and the other is `class_labels` being ignored in forward; not the same defect." + }, + { + "left": "issue:5004", + "right": "issue:6439", + "accept": false, + "reason": "Checkpoint-resume training failure and a broken docs link are unrelated." + }, + { + "left": "issue:5150", + "right": "issue:5405", + "accept": false, + "reason": "Both mention `from_single_file`, but one is a missing method on StableDiffusionUpscalePipeline and the other is weights not being overwritten in SDXL loading." + }, + { + "left": "issue:5392", + "right": "issue:986", + "accept": false, + "reason": "A training question for latent diffusion and a butterfly demo question are unrelated." + }, + { + "left": "issue:5932", + "right": "issue:7002", + "accept": false, + "reason": "Both are runtime errors in pipelines, but they describe different symptoms and likely different root causes." + }, + { + "left": "issue:6218", + "right": "issue:7592", + "accept": false, + "reason": "Batch seed handling in pipelines vs incomplete DoRA state dict loading are unrelated." + }, + { + "left": "issue:6297", + "right": "issue:9393", + "accept": false, + "reason": "Both are LoRA training problems, but they involve different models, tooling, and failure points." + }, + { + "left": "issue:6791", + "right": "issue:7146", + "accept": false, + "reason": "Same training script, but one report says it cannot run while the other is specifically CUDA OOM; not the same issue." + }, + { + "left": "issue:6865", + "right": "issue:9645", + "accept": false, + "reason": "A load_image type-check bug and an import error for a loader mixin are unrelated." + }, + { + "left": "issue:7009", + "right": "issue:9568", + "accept": false, + "reason": "Different model families and different reconstruction/generation failures." + }, + { + "left": "issue:9611", + "right": "issue:9767", + "accept": false, + "reason": "Both involve FluxImg2ImgPipeline, but one is an image batching error and the other is a cuDNN frontend error during LoRA use; too different to merge." + }, + { + "left": "issue:12481", + "right": "issue:12483", + "accept": true, + "reason": "Same `offload_state_dict` TypeError family; 12481 is just another phrasing of the same diffusers/transformers incompatibility." + }, + { + "left": "issue:12482", + "right": "issue:12483", + "accept": true, + "reason": "Same underlying bug: `CLIPTextModel` receives unexpected `offload_state_dict`, matching the same compatibility problem." + }, + { + "left": "issue:12480", + "right": "issue:12483", + "accept": true, + "reason": "Same root cause and error path; 12480 is a duplicate report of the `offload_state_dict` constructor failure." + }, + { + "left": "issue:2453", + "right": "issue:6423", + "accept": false, + "reason": "General contribution question and a specific reproducibility issue are unrelated." + }, + { + "left": "issue:2453", + "right": "issue:4241", + "accept": false, + "reason": "A contribution question and a LoRA fine-tuning feature request are not duplicates." + }, + { + "left": "issue:12369", + "right": "issue:12457", + "accept": false, + "reason": "Both are CineScale-related, but the titles are too broad to confirm the same bug or change; insufficient evidence for a duplicate merge." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: 12369/12457 is a separate CineScale pair, and 12480/12481/12482/12483 are a clear duplicate cluster around the same `offload_state_dict` / `CLIPTextModel` constructor failure. The soft-edge rejections are conservative and the accepted duplicates are well supported by the near-identical titles and high similarity scores." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 179, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11519", + "issue:11560", + "issue:12369", + "issue:1755", + "issue:1962", + "issue:200", + "issue:202", + "issue:2195", + "issue:2311", + "issue:2453", + "issue:2462", + "issue:2838", + "issue:3385", + "issue:3427", + "issue:3576", + "issue:3593", + "issue:3698", + "issue:378", + "issue:384", + "issue:3902", + "issue:4241", + "issue:4445", + "issue:4609", + "issue:4716", + "issue:4747", + "issue:4750", + "issue:4876", + "issue:4913", + "issue:4922", + "issue:4978", + "issue:5274", + "issue:5425", + "issue:5447", + "issue:5628", + "issue:5630", + "issue:5932", + "issue:5947", + "issue:6423", + "issue:6572", + "issue:692", + "issue:7619", + "issue:9031", + "issue:9032", + "issue:9038", + "issue:9039", + "issue:9047", + "issue:9224", + "issue:9410" + ], + "soft_pairs": [ + "issue:3698|issue:4445", + "issue:4241|issue:4876", + "issue:4241|issue:6423", + "issue:4445|issue:4876", + "issue:4445|issue:692", + "issue:3576|issue:3593", + "issue:4913|issue:6572", + "issue:1755|issue:4609", + "issue:1755|issue:2311", + "issue:4922|issue:692", + "issue:4750|issue:4978", + "issue:200|issue:202", + "issue:11560|issue:9410", + "issue:9031|issue:9032", + "issue:4716|issue:4747", + "issue:5628|issue:5630", + "issue:378|issue:384", + "issue:1962|issue:3902", + "issue:9038|issue:9039", + "issue:2462|issue:5947", + "issue:11519|issue:12369", + "issue:5447|issue:5628", + "issue:9047|issue:9224", + "issue:2453|issue:2838", + "issue:5932|issue:7619", + "issue:5274|issue:5425", + "issue:1962|issue:2195", + "issue:3385|issue:3427" + ], + "prepared_review_unit_hash": "a23ea9e51211db08377febf02ae14ebd6430d169449931f451d33af42a611df0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 34301, + "estimated_input_tokens": 8576, + "estimated_eval_tokens": 17408 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous cluster of unrelated feature requests and bug reports. The only strong duplicate family I see is the DreamBooth channel-mismatch error pair (5932/7619), with a second textual-inversion/DiagonalGaussianDistribution pair (378/384) also looking like the same underlying bug. The other soft pairs are broader similarity matches but not true duplicates.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 5932 is the clearest concrete bug report in the set and has the strongest duplicate match (7619) with a reproducible channel-mismatch traceback.", + "canonical_pr_reason": null, + "best_issue_reason": "5932 is the best issue to keep as the canonical representative of its bug family; most other items are feature/support requests rather than the same defect.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3698", + "right": "issue:4445", + "accept": false, + "reason": "Different asks: converting an inpainting model to diffusers vs guidance on LoRA finetuning." + }, + { + "left": "issue:4241", + "right": "issue:4876", + "accept": false, + "reason": "Both mention LoRA, but one is about fine-tuning support and the other is about exposing a layer-std argument; not the same change." + }, + { + "left": "issue:4241", + "right": "issue:6423", + "accept": false, + "reason": "Unrelated feature request vs reproduction issue for fill50k results." + }, + { + "left": "issue:4445", + "right": "issue:4876", + "accept": false, + "reason": "Finetuning how-to vs LoRA initialization parameter request; different topics." + }, + { + "left": "issue:4445", + "right": "issue:692", + "accept": false, + "reason": "General fine-tuning example request is not the same as a LoRA finetuning question." + }, + { + "left": "issue:3576", + "right": "issue:3593", + "accept": false, + "reason": "Both involve xformers, but one is Textual Inversion and the other is LoRA/PyTorch 2.0 second-call failure; different code paths." + }, + { + "left": "issue:4913", + "right": "issue:6572", + "accept": false, + "reason": "Different pipeline/model requests: ControlNet openpose for SDXL vs ControlNetXS inpaint pipeline." + }, + { + "left": "issue:1755", + "right": "issue:4609", + "accept": false, + "reason": "No meaningful overlap beyond generic support-request noise; unrelated subjects." + }, + { + "left": "issue:1755", + "right": "issue:2311", + "accept": false, + "reason": "Generic unrelated issue vs UniPC scheduler request." + }, + { + "left": "issue:4922", + "right": "issue:692", + "accept": false, + "reason": "Android demo request is unrelated to diffusion fine-tuning examples." + }, + { + "left": "issue:4750", + "right": "issue:4978", + "accept": false, + "reason": "Both touch inpainting preprocessors, but they ask for different specific preprocessors and behaviors; not clearly the same change." + }, + { + "left": "issue:200", + "right": "issue:202", + "accept": false, + "reason": "Different test additions for different files/classes." + }, + { + "left": "issue:11560", + "right": "issue:9410", + "accept": false, + "reason": "Generic import error is too vague to treat as the same as the SD3 ControlNet inpainting import failure." + }, + { + "left": "issue:9031", + "right": "issue:9032", + "accept": false, + "reason": "Different FlaxStableDiffusionPipeline questions: safety checker vs TPU colab setup." + }, + { + "left": "issue:4716", + "right": "issue:4747", + "accept": false, + "reason": "SDXL long-prompt handling vs missing model_index.json after loading a fine-tuned LoRA; unrelated issues." + }, + { + "left": "issue:378", + "right": "issue:384", + "accept": true, + "reason": "Same Textual Inversion / DiagonalGaussianDistribution bug family; 384 reads like the concrete error manifestation of 378." + }, + { + "left": "issue:1962", + "right": "issue:3902", + "accept": false, + "reason": "Both are video-related, but one is a stable-diffusion video-generation request and the other is a Control-A-Video request." + }, + { + "left": "issue:9038", + "right": "issue:9039", + "accept": false, + "reason": "Prompt weighting and LoRA loading are separate API requests, not one duplicate bug." + }, + { + "left": "issue:2462", + "right": "issue:5947", + "accept": false, + "reason": "IterableDataset finetuning support and Diffusion-DPO policy discussion are unrelated." + }, + { + "left": "issue:11519", + "right": "issue:12369", + "accept": false, + "reason": "Different model support requests: MAGI-1 vs Wan 2.1 CineScale up-to-4K." + }, + { + "left": "issue:5447", + "right": "issue:5628", + "accept": false, + "reason": "Both concern denoising start/end behavior, but they target different scheduler configurations and symptoms; too broad to merge as one bug." + }, + { + "left": "issue:9047", + "right": "issue:9224", + "accept": false, + "reason": "MPS failure vs a specific float64-on-Metal limitation; related theme, but not clearly the same underlying defect from the available evidence." + }, + { + "left": "issue:2453", + "right": "issue:2838", + "accept": false, + "reason": "Contribution questions and ControlNet pipeline behavior are unrelated." + }, + { + "left": "issue:5932", + "right": "issue:7619", + "accept": true, + "reason": "Same concrete DreamBooth channel-count mismatch; 7619 is essentially the same error class as 5932." + }, + { + "left": "issue:5274", + "right": "issue:5425", + "accept": false, + "reason": "Different training/checkpointing problems; no shared concrete bug." + }, + { + "left": "issue:1962", + "right": "issue:2195", + "accept": false, + "reason": "Both mention video, but they are different feature requests with different targets." + }, + { + "left": "issue:3385", + "right": "issue:3427", + "accept": false, + "reason": "Distinct model-support requests for different JAX integrations." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and the soft-edge verdicts are generally conservative. The main duplicate family callout (5932/7619) is well supported, and the other pairs are mostly reasonable false negatives. Minor caution: avoid wording that implies extra evidence not shown in the packet (for example, a specific 'traceback') unless it appears in the issue text." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 180, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10049", + "issue:11432", + "issue:11555", + "issue:11560", + "issue:1158", + "issue:1160", + "issue:11630", + "issue:11878", + "issue:12097", + "issue:12499", + "issue:12567", + "issue:12732", + "issue:12738", + "issue:12896", + "issue:13105", + "issue:13232", + "issue:13351", + "issue:1755", + "issue:201", + "issue:2195", + "issue:2221", + "issue:2268", + "issue:2830", + "issue:2838", + "issue:2984", + "issue:338", + "issue:339", + "issue:3698", + "issue:3699", + "issue:4141", + "issue:4609", + "issue:4835", + "issue:4913", + "issue:5112", + "issue:5162", + "issue:5480", + "issue:5926", + "issue:5947", + "issue:6042", + "issue:6572", + "issue:671", + "issue:743", + "issue:7972", + "issue:8241", + "issue:9042", + "issue:9264", + "issue:9825" + ], + "soft_pairs": [ + "issue:7972|issue:8241", + "issue:671|issue:743", + "issue:3698|issue:5162", + "issue:2838|issue:5947", + "issue:1755|issue:5480", + "issue:9042|issue:9264", + "issue:11630|issue:11878", + "issue:201|issue:338", + "issue:201|issue:339", + "issue:2221|issue:2268", + "issue:10049|issue:11560", + "issue:4913|issue:5926", + "issue:13232|issue:13351", + "issue:1158|issue:1160", + "issue:9042|issue:9825", + "issue:4835|issue:5112", + "issue:6042|issue:6572", + "issue:2830|issue:4609", + "issue:3699|issue:4141", + "issue:12097|issue:12896", + "issue:12732|issue:12738", + "issue:12567|issue:13105", + "issue:11432|issue:11555", + "issue:2195|issue:2984", + "issue:12097|issue:12499" + ], + "prepared_review_unit_hash": "2b7e286bfe17b38d35cb4426e584e10db1be9f6a8c5a1e29b7e4d046c3546054", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 32732, + "estimated_input_tokens": 8183, + "estimated_eval_tokens": 16622 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated Diffusers issues and feature requests; none of the soft pairs look like true duplicates. The most representative issue is the broad FLUX pipeline request.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 9042 is the broadest, most central request in the FLUX subgroup, and it has nearby follow-up interest (e.g. FLUX inpainting/IPAdapter requests), so it is the best anchor for this cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9042 is the best overall representative because it is a clear, high-signal feature request with broader relevance than the more niche or noisy reports in the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7972", + "right": "issue:8241", + "accept": false, + "reason": "Same model family (LEDITS++) but different bugs: type mismatch vs abnormal model size." + }, + { + "left": "issue:671", + "right": "issue:743", + "accept": false, + "reason": "Both concern conversion, but one is about non-identical outputs while the other is a conversion failure; different concrete problems." + }, + { + "left": "issue:3698", + "right": "issue:5162", + "accept": false, + "reason": "Different requests: converting an inpainting model to Diffusers vs converting a folder checkpoint to safetensors." + }, + { + "left": "issue:2838", + "right": "issue:5947", + "accept": false, + "reason": "ControlNet pipeline behavior request and Diffusion-DPO policy discussion are unrelated." + }, + { + "left": "issue:1755", + "right": "issue:5480", + "accept": false, + "reason": "Generic title vs EDM implementation request; no sign of the same underlying issue." + }, + { + "left": "issue:9042", + "right": "issue:9264", + "accept": false, + "reason": "Both are FLUX-related, but one asks for the base pipeline and the other for an inpainting model; separate feature work." + }, + { + "left": "issue:11630", + "right": "issue:11878", + "accept": false, + "reason": "Support for from_single_file and a GGUF 1.3B runtime failure are different problems." + }, + { + "left": "issue:201", + "right": "issue:338", + "accept": false, + "reason": "Different tests for different components; not the same failure or change." + }, + { + "left": "issue:201", + "right": "issue:339", + "accept": false, + "reason": "Different test coverage targets (attention class vs Karras VE)." + }, + { + "left": "issue:2221", + "right": "issue:2268", + "accept": false, + "reason": "Both mention ColossalAI errors, but the titles are too generic to establish the same root cause." + }, + { + "left": "issue:10049", + "right": "issue:11560", + "accept": false, + "reason": "MochiPipeline import failure and a generic import error are not enough to identify the same bug." + }, + { + "left": "issue:4913", + "right": "issue:5926", + "accept": false, + "reason": "Different model/package requests entirely: controlnet-openpose-sdxl-1.0 vs Kandinsky 3.0." + }, + { + "left": "issue:13232", + "right": "issue:13351", + "accept": false, + "reason": "LTX-2.3 support and Wan/Flux controlnet support are distinct feature requests." + }, + { + "left": "issue:1158", + "right": "issue:1160", + "accept": false, + "reason": "Test interference in config tests and a hanging custom-pipeline test are different issues." + }, + { + "left": "issue:9042", + "right": "issue:9825", + "accept": false, + "reason": "FLUX base pipeline support vs IPAdapter support for FLUX pipelines; related area but not the same change." + }, + { + "left": "issue:4835", + "right": "issue:5112", + "accept": false, + "reason": "Same pipeline class, but one bug is MultiControlNet behavior and the other is a torch.Size float-shape error." + }, + { + "left": "issue:6042", + "right": "issue:6572", + "accept": false, + "reason": "SDXL-turbo distillation example request and ControlNetXS inpaint pipeline request are different features." + }, + { + "left": "issue:2830", + "right": "issue:4609", + "accept": false, + "reason": "A video-generation question and a model availability request are unrelated." + }, + { + "left": "issue:3699", + "right": "issue:4141", + "accept": false, + "reason": "Different bugs: CUDA-on-M2 assertion vs ControlNet inpaint not respecting the second controlnet." + }, + { + "left": "issue:12097", + "right": "issue:12896", + "accept": false, + "reason": "WAN 2.2 VRAM OOM at the end and an MPS-specific failure are different platform/runtime problems." + }, + { + "left": "issue:12732", + "right": "issue:12738", + "accept": false, + "reason": "Both are Z-Image pipeline errors, but one is generic inference failure and the other is a batch-size assertion; not clearly the same bug." + }, + { + "left": "issue:12567", + "right": "issue:13105", + "accept": false, + "reason": "Output noise under Ulysses Attention vs WAN 2.1 T2V noise are different model/code-path reports." + }, + { + "left": "issue:11432", + "right": "issue:11555", + "accept": false, + "reason": "torch_dtype='auto' not working and device_map='auto' support are different API concerns." + }, + { + "left": "issue:2195", + "right": "issue:2984", + "accept": false, + "reason": "Tune A Video and latent video diffusion long-video generation are different feature requests." + }, + { + "left": "issue:12097", + "right": "issue:12499", + "accept": false, + "reason": "WAN 2.2 VRAM OOM and WAN 2.2 I2V shape mismatch are different failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 181, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10120", + "issue:1076", + "issue:10760", + "issue:11315", + "issue:11436", + "issue:11653", + "issue:11821", + "issue:11878", + "issue:12606", + "issue:1755", + "issue:1962", + "issue:2015", + "issue:2195", + "issue:2453", + "issue:2635", + "issue:321", + "issue:322", + "issue:3728", + "issue:3788", + "issue:3838", + "issue:4042", + "issue:4225", + "issue:4232", + "issue:4269", + "issue:4406", + "issue:4615", + "issue:4913", + "issue:5017", + "issue:5284", + "issue:6008", + "issue:6042", + "issue:6287", + "issue:6393", + "issue:6423", + "issue:6444", + "issue:6482", + "issue:6572", + "issue:7303", + "issue:7408", + "issue:8477", + "issue:8493", + "issue:8560", + "issue:8738", + "issue:8803", + "issue:8894", + "issue:9218", + "issue:9220" + ], + "soft_pairs": [ + "issue:8493|issue:8738", + "issue:11315|issue:11436", + "issue:4269|issue:4615", + "issue:6287|issue:8803", + "issue:11436|issue:11653", + "issue:2453|issue:5017", + "issue:10760|issue:12606", + "issue:11821|issue:11878", + "issue:10120|issue:8560", + "issue:1076|issue:321", + "issue:3728|issue:7408", + "issue:4225|issue:4406", + "issue:2015|issue:6572", + "issue:321|issue:322", + "issue:3788|issue:7408", + "issue:6444|issue:6482", + "issue:4232|issue:5284", + "issue:6008|issue:6393", + "issue:9218|issue:9220", + "issue:1755|issue:2635", + "issue:8477|issue:8894", + "issue:1962|issue:3838", + "issue:4913|issue:6042", + "issue:1755|issue:4042", + "issue:5017|issue:6423", + "issue:5017|issue:7303", + "issue:2195|issue:3838" + ], + "prepared_review_unit_hash": "03903b07fdff4db092d807e9722b29e500d3d55f5665fe844bc48547628aecda", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 33657, + "estimated_input_tokens": 8415, + "estimated_eval_tokens": 17086 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set does not form a single duplicate cluster; it contains many unrelated issue reports. All soft-pair candidates are false positives and should be kept separate.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8493", + "right": "issue:8738", + "accept": false, + "reason": "Both concern SD3Transformer2DModel, but one is a state_dict loading error and the other is a configuration issue; similar subsystem, not the same bug." + }, + { + "left": "issue:11315", + "right": "issue:11436", + "accept": false, + "reason": "Both are HiDream dtype-related, but one is a float64 RoPE/MPS failure and the other is a Half vs BFloat16 matmul mismatch; different code paths and symptoms." + }, + { + "left": "issue:4269", + "right": "issue:4615", + "accept": false, + "reason": "Different problems: a TypeError from default args in check_inputs versus an import-name failure for ControlNetPipeline." + }, + { + "left": "issue:6287", + "right": "issue:8803", + "accept": false, + "reason": "One is dependency cleanup in setup.py, the other is a duplicate assignment in attention_processor.py; no duplicate bug/change." + }, + { + "left": "issue:11436", + "right": "issue:11653", + "accept": false, + "reason": "One reports a runtime dtype mismatch in HiDream inference; the other requests support for .alpha LoRA keys. Different issues." + }, + { + "left": "issue:2453", + "right": "issue:5017", + "accept": false, + "reason": "General contribution question versus a feature request for control mode in ControlNet; unrelated." + }, + { + "left": "issue:10760", + "right": "issue:12606", + "accept": false, + "reason": "Both mention imports, but they refer to different scripts and different import failures." + }, + { + "left": "issue:11821", + "right": "issue:11878", + "accept": false, + "reason": "WanVACE prompt-list support and WanVACE GGUF 1.3B loading are distinct issues." + }, + { + "left": "issue:10120", + "right": "issue:8560", + "accept": false, + "reason": "Both mention SD3 training memory/OOM, but one is DreamBooth+SD3 memory consumption and the other is a 4090 OOM during SD3 training; not the same concrete bug." + }, + { + "left": "issue:1076", + "right": "issue:321", + "accept": false, + "reason": "Completely different topics: stable-diffusion-v1-5 discussion versus blended diffusion." + }, + { + "left": "issue:3728", + "right": "issue:7408", + "accept": false, + "reason": "Different proposed pipelines/models: Shifted Diffusion versus FRESCO." + }, + { + "left": "issue:4225", + "right": "issue:4406", + "accept": false, + "reason": "ControlNet state_dict loading errors are not the same as a safetensor header deserialization failure." + }, + { + "left": "issue:2015", + "right": "issue:6572", + "accept": false, + "reason": "Categorical diffusion pipeline and ControlNetXS-SDXL-Inpaint-Pipeline are unrelated feature requests." + }, + { + "left": "issue:321", + "right": "issue:322", + "accept": false, + "reason": "Different diffusion methods/models: Blended Diffusion versus V-Diffusion." + }, + { + "left": "issue:3788", + "right": "issue:7408", + "accept": false, + "reason": "UniControl and FRESCO are separate pipeline requests with different functionality." + }, + { + "left": "issue:6444", + "right": "issue:6482", + "accept": false, + "reason": "Both are training-script failures, but they are different errors triggered by different conditions." + }, + { + "left": "issue:4232", + "right": "issue:5284", + "accept": false, + "reason": "Instruct-pix2pix via ControlNet condition is not the same change as ControlNet-LLLite support." + }, + { + "left": "issue:6008", + "right": "issue:6393", + "accept": false, + "reason": "IP-Adapter/LCM LoRA failure and private HF download failure through from_single_file are unrelated." + }, + { + "left": "issue:9218", + "right": "issue:9220", + "accept": false, + "reason": "PAG guess-mode support request versus an undefined-variable bug in PAG pipeline code; not the same bug." + }, + { + "left": "issue:1755", + "right": "issue:2635", + "accept": false, + "reason": "General discussion thread versus a scheduler naming question; no duplicate relationship." + }, + { + "left": "issue:8477", + "right": "issue:8894", + "accept": false, + "reason": "Both involve training scripts, but they reference different scripts and likely different distributed-training failures." + }, + { + "left": "issue:1962", + "right": "issue:3838", + "accept": false, + "reason": "Video generation with stable diffusion and Video Composer are distinct topics." + }, + { + "left": "issue:4913", + "right": "issue:6042", + "accept": false, + "reason": "ControlNet OpenPose SDXL model discussion is unrelated to an SDXL-turbo distillation example request." + }, + { + "left": "issue:1755", + "right": "issue:4042", + "accept": false, + "reason": "General thread versus a question about combining reference-only, inpainting, and depth control; not duplicates." + }, + { + "left": "issue:5017", + "right": "issue:6423", + "accept": false, + "reason": "Control mode request for ControlNet and inability to reproduce fill50k results are unrelated." + }, + { + "left": "issue:5017", + "right": "issue:7303", + "accept": false, + "reason": "Control mode for ControlNet and enabling zero_snr via from_pretrained are unrelated." + }, + { + "left": "issue:2195", + "right": "issue:3838", + "accept": false, + "reason": "Tune A Video and Video Composer are different video-generation projects/requests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 182, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10049", + "issue:10186", + "issue:10204", + "issue:10460", + "issue:10658", + "issue:10737", + "issue:10893", + "issue:11560", + "issue:1160", + "issue:12369", + "issue:12606", + "issue:12624", + "issue:13065", + "issue:13232", + "issue:1976", + "issue:3113", + "issue:3698", + "issue:3861", + "issue:3902", + "issue:4042", + "issue:4241", + "issue:4250", + "issue:4524", + "issue:4789", + "issue:5017", + "issue:5023", + "issue:5118", + "issue:5227", + "issue:5318", + "issue:5341", + "issue:5770", + "issue:6095", + "issue:6096", + "issue:6161", + "issue:6288", + "issue:6393", + "issue:8488", + "issue:8503", + "issue:8511", + "issue:8518", + "issue:8565", + "issue:8975", + "issue:9001", + "issue:9026", + "issue:9112", + "issue:9141", + "issue:9147" + ], + "soft_pairs": [ + "issue:6095|issue:6096", + "issue:5770|issue:6288", + "issue:3698|issue:6161", + "issue:11560|issue:8975", + "issue:12369|issue:13232", + "issue:10893|issue:13065", + "issue:10460|issue:9026", + "issue:5023|issue:9147", + "issue:9001|issue:9112", + "issue:4241|issue:5017", + "issue:5227|issue:8518", + "issue:10186|issue:10204", + "issue:8488|issue:8565", + "issue:8503|issue:8511", + "issue:3861|issue:4250", + "issue:5318|issue:5341", + "issue:1160|issue:1976", + "issue:10658|issue:12624", + "issue:4524|issue:5118", + "issue:4789|issue:6393", + "issue:3902|issue:4042", + "issue:10737|issue:9141", + "issue:10049|issue:12606", + "issue:3113|issue:3902" + ], + "prepared_review_unit_hash": "6f9542f464186bdc421a43910a3a8f58f746caa86b79dce5b1859ba48dfe395e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32878, + "estimated_input_tokens": 8220, + "estimated_eval_tokens": 16696 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the soft links mostly connect superficially similar diffusers issues (imports, Flux/SD3, controlnet, model support) that actually describe different bugs, features, or code paths.", + "confidence": 0.88, + "canonical_issue_reason": "No single canonical issue stands out because the set is heterogeneous and none of the soft pairs share the same concrete failure mode.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue for deduplication; if forced, issue:9141 is the cleanest self-contained bug report, but it is not representative of the rest of the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6095", + "right": "issue:6096", + "accept": false, + "reason": "Both involve SVD, but one is an IndexError in scheduling_euler_discrete and the other is a torch.compile DynamicOutputShapeException; different code paths." + }, + { + "left": "issue:5770", + "right": "issue:6288", + "accept": false, + "reason": "One is about dataset access preventing LCM distillation; the other is about the default learning rate being too small for LCM LoRA training." + }, + { + "left": "issue:3698", + "right": "issue:6161", + "accept": false, + "reason": "Inpainting model conversion to diffusers is unrelated to the sigma-scheduling bug about discarding the next-to-last sigma." + }, + { + "left": "issue:11560", + "right": "issue:8975", + "accept": false, + "reason": "Both mention import errors, but one is generic and the other is specifically about LoraLoaderMixin import removal; not enough to treat as the same bug." + }, + { + "left": "issue:12369", + "right": "issue:13232", + "accept": false, + "reason": "Different model requests entirely: Wan 2.1 CineScale up to 4K vs LTX-2.3 support." + }, + { + "left": "issue:10893", + "right": "issue:13065", + "accept": false, + "reason": "Both are control-related feature requests, but they target different model families and different controlnet capabilities." + }, + { + "left": "issue:10460", + "right": "issue:9026", + "accept": false, + "reason": "Missing documented UNet2DModel argument is unrelated to the load_checkpoint_and_dispatch unexpected force_hook argument error." + }, + { + "left": "issue:5023", + "right": "issue:9147", + "accept": false, + "reason": "WuerstchenCombinedPipeline import failure and Flux autopipeline error are distinct pipeline/import issues." + }, + { + "left": "issue:9001", + "right": "issue:9112", + "accept": false, + "reason": "Aspect ratio support request is not the same as a tensor-shape mismatch during Auraflow inference." + }, + { + "left": "issue:4241", + "right": "issue:5017", + "accept": false, + "reason": "LoRA fine-tuning for unidiffusers and control mode for controlnet are different feature requests." + }, + { + "left": "issue:5227", + "right": "issue:8518", + "accept": false, + "reason": "Both touch torch.compile, but one is about load_lora_weight failing under compile and the other is an unsupported SD3 compile trace; different failure modes." + }, + { + "left": "issue:10186", + "right": "issue:10204", + "accept": false, + "reason": "FluxFillPipeline import error and FluxFillPipeline inpainting quality degradation are unrelated issues." + }, + { + "left": "issue:8488", + "right": "issue:8565", + "accept": false, + "reason": "StableDiffusion3Pipeline from_pretrained exception and load_lora_weights IndexError affect different APIs and likely different fixes." + }, + { + "left": "issue:8503", + "right": "issue:8511", + "accept": false, + "reason": "Gradient checkpointing crash in SD3 is unrelated to the script not unloading text encoders." + }, + { + "left": "issue:3861", + "right": "issue:4250", + "accept": false, + "reason": "Both mention from_pretrained, but one is about the API form/naming and the other about incorrect model loading behavior; not the same bug." + }, + { + "left": "issue:5318", + "right": "issue:5341", + "accept": false, + "reason": "These are separate PEFT documentation/example updates, not the same underlying issue." + }, + { + "left": "issue:1160", + "right": "issue:1976", + "accept": false, + "reason": "A hanging pytest test is not the same as an incorrect tokenizer/config.json file lookup." + }, + { + "left": "issue:10658", + "right": "issue:12624", + "accept": false, + "reason": "onnxruntime provider_options and WanImageToVideoPipeline strength are unrelated API requests." + }, + { + "left": "issue:4524", + "right": "issue:5118", + "accept": false, + "reason": "AnimateDiff integration and ControlNet reference_only usage are different topics." + }, + { + "left": "issue:4789", + "right": "issue:6393", + "accept": false, + "reason": "Both involve downloads, but one is DeepFloyd IF download failure and the other is private HF link handling via from_single_file; different causes." + }, + { + "left": "issue:3902", + "right": "issue:4042", + "accept": false, + "reason": "control-a-video and combining reference-only with inpainting/depth control are distinct feature requests." + }, + { + "left": "issue:10737", + "right": "issue:9141", + "accept": false, + "reason": "Both concern FluxPipeline, but one lacks image_encoder in encode_image and the other lacks from_single_file; different APIs." + }, + { + "left": "issue:10049", + "right": "issue:12606", + "accept": false, + "reason": "Cannot import MochiPipeline is unrelated to wrong import behavior in scripts/convert_sd3_to_diffusers.py." + }, + { + "left": "issue:3113", + "right": "issue:3902", + "accept": false, + "reason": "Training-free layout control with cross-attention guidance is a different control idea from control-a-video." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 183, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10421", + "issue:10460", + "issue:10467", + "issue:10526", + "issue:10817", + "issue:10950", + "issue:11560", + "issue:11749", + "issue:12060", + "issue:12599", + "issue:12754", + "issue:13101", + "issue:1594", + "issue:1925", + "issue:2195", + "issue:2440", + "issue:272", + "issue:2737", + "issue:2822", + "issue:2830", + "issue:3140", + "issue:338", + "issue:3467", + "issue:3728", + "issue:3838", + "issue:3902", + "issue:4305", + "issue:4733", + "issue:4750", + "issue:4913", + "issue:4972", + "issue:5009", + "issue:5034", + "issue:5168", + "issue:5314", + "issue:5436", + "issue:5794", + "issue:6572", + "issue:6866", + "issue:6901", + "issue:8503", + "issue:8507", + "issue:9062", + "issue:9148", + "issue:9264", + "issue:9331", + "issue:9680", + "issue:9825" + ], + "soft_pairs": [ + "issue:2195|issue:2830", + "issue:2822|issue:3140", + "issue:12060|issue:13101", + "issue:1925|issue:2440", + "issue:1594|issue:3728", + "issue:8503|issue:8507", + "issue:10421|issue:12754", + "issue:10950|issue:9680", + "issue:10421|issue:12599", + "issue:4305|issue:5034", + "issue:9148|issue:9825", + "issue:3467|issue:4733", + "issue:5314|issue:9264", + "issue:2830|issue:3838", + "issue:11749|issue:9331", + "issue:4972|issue:5009", + "issue:11560|issue:9062", + "issue:272|issue:338", + "issue:10460|issue:6901", + "issue:1594|issue:3902", + "issue:2737|issue:6866", + "issue:4913|issue:5168", + "issue:10467|issue:10526", + "issue:4750|issue:6572", + "issue:10817|issue:8503", + "issue:5436|issue:5794" + ], + "prepared_review_unit_hash": "1757c7506c0119bbed7a2d973e045a082e689e5364e86f624cc7f32b24ee284f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 33651, + "estimated_input_tokens": 8413, + "estimated_eval_tokens": 17082 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated Diffusers issues; only a few soft pairs look like true duplicates. No PRs are present, and there is no single umbrella issue for the whole set.", + "confidence": 0.78, + "canonical_issue_reason": "issue:2440 is the clearest duplicate target in the only strong issue family (float16 inpainting load errors), but the overall set is too heterogeneous to have one true canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:2440 is the best standalone anchor for triage: specific, reproducible, and directly matches the strongest near-duplicate pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2195", + "right": "issue:2830", + "accept": false, + "reason": "Different tasks: video generation request vs a ModelScope video-use question." + }, + { + "left": "issue:2822", + "right": "issue:3140", + "accept": false, + "reason": "Training slowdown/error in ControlNet is unrelated to an import failure." + }, + { + "left": "issue:12060", + "right": "issue:13101", + "accept": false, + "reason": "A question about DiT block availability is unrelated to a malformed huggingface issue." + }, + { + "left": "issue:1925", + "right": "issue:2440", + "accept": true, + "reason": "Both report inpainting models failing in float16/half precision; same underlying loading/runtime bug." + }, + { + "left": "issue:1594", + "right": "issue:3728", + "accept": false, + "reason": "Img2Img video-generation discussion and a shifted-diffusion text-to-image pipeline request are different changes." + }, + { + "left": "issue:8503", + "right": "issue:8507", + "accept": false, + "reason": "Same SD3 area, but gradient-checkpointing crashes and train_text_encoder failures are different bugs." + }, + { + "left": "issue:10421", + "right": "issue:12754", + "accept": false, + "reason": "Multi-GPU memory leak and from_pipe dtype conversion are different problems and code paths." + }, + { + "left": "issue:10950", + "right": "issue:9680", + "accept": false, + "reason": "Broken docs for Accelerate inference and broken SDXL unconditional generation are unrelated." + }, + { + "left": "issue:10421", + "right": "issue:12599", + "accept": false, + "reason": "CUDA move slowness is not the same bug as a memory leak when moving pipelines across GPUs." + }, + { + "left": "issue:4305", + "right": "issue:5034", + "accept": false, + "reason": "torch.compile graph breaks and checkpoint-loading progress-bar glitches are unrelated." + }, + { + "left": "issue:9148", + "right": "issue:9825", + "accept": false, + "reason": "Flux inpainting pipeline support and FLUX IP-Adapter support are separate feature requests." + }, + { + "left": "issue:3467", + "right": "issue:4733", + "accept": false, + "reason": "Resume-step bug in ControlNet training is different from the DreamBooth LoRA optimizer-state error." + }, + { + "left": "issue:5314", + "right": "issue:9264", + "accept": false, + "reason": "ControlNet reference inpainting and Flux inpainting are distinct pipeline requests." + }, + { + "left": "issue:2830", + "right": "issue:3838", + "accept": false, + "reason": "A generic video-generation question and a Video Composer request are not the same concrete issue." + }, + { + "left": "issue:11749", + "right": "issue:9331", + "accept": true, + "reason": "Both describe the same FlowMatchEulerDiscreteScheduler timestep/indexing precision bug." + }, + { + "left": "issue:4972", + "right": "issue:5009", + "accept": true, + "reason": "Both are wuerstchen runtime/example failures and appear to be the same underlying issue." + }, + { + "left": "issue:11560", + "right": "issue:9062", + "accept": false, + "reason": "A vague import error is too broad to treat as the same issue as StableAudioPipeline import failure." + }, + { + "left": "issue:272", + "right": "issue:338", + "accept": false, + "reason": "K-LMS inpainting support and LMS-discrete test coverage are different tasks." + }, + { + "left": "issue:10460", + "right": "issue:6901", + "accept": false, + "reason": "Missing mid_block_type on UNet2DModel is unrelated to missing clip_sample on FlaxDDIMScheduler." + }, + { + "left": "issue:1594", + "right": "issue:3902", + "accept": false, + "reason": "Img2Img for video generation and control-a-video are different feature requests." + }, + { + "left": "issue:2737", + "right": "issue:6866", + "accept": false, + "reason": "CLIP guidance support and IP-Adapter support are separate feature changes." + }, + { + "left": "issue:4913", + "right": "issue:5168", + "accept": false, + "reason": "ControlNet-openpose SDXL and ControlNet-XS support are related only broadly, not the same concrete change." + }, + { + "left": "issue:10467", + "right": "issue:10526", + "accept": false, + "reason": "An FP8 example-code artifact issue and a WeightQBytesTensor constructor TypeError are different bugs." + }, + { + "left": "issue:4750", + "right": "issue:6572", + "accept": false, + "reason": "Inpaint + LaMa preprocessor and ControlNetXS-SDXL-Inpaint-Pipeline are different feature requests." + }, + { + "left": "issue:10817", + "right": "issue:8503", + "accept": false, + "reason": "SD3 ControlNet support and SD3 gradient-checkpointing crashes are unrelated." + }, + { + "left": "issue:5436", + "right": "issue:5794", + "accept": false, + "reason": "ControlNet XL training error and missing tokenizer/config.json are different problems." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: it correctly characterizes the set as mostly heterogeneous, notes there are no PRs, and only elevates a small number of duplicate-looking issue pairs. The accepted soft-edge verdicts are conservative and align with the issue titles, especially 1925\u21942440, 11749\u21949331, and 4972\u21945009. No clear evidence of overstatement beyond modest inference from titles." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 184, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10460", + "issue:10472", + "issue:10817", + "issue:10950", + "issue:11315", + "issue:11653", + "issue:12221", + "issue:12239", + "issue:13351", + "issue:2163", + "issue:2213", + "issue:2466", + "issue:2618", + "issue:2822", + "issue:2985", + "issue:321", + "issue:3431", + "issue:3451", + "issue:3566", + "issue:3958", + "issue:4063", + "issue:4376", + "issue:4611", + "issue:4779", + "issue:4885", + "issue:4888", + "issue:5054", + "issue:5075", + "issue:5109", + "issue:5205", + "issue:5258", + "issue:5318", + "issue:5320", + "issue:5339", + "issue:5395", + "issue:5824", + "issue:5926", + "issue:6681", + "issue:6688", + "issue:7178", + "issue:775", + "issue:8507", + "issue:9045", + "issue:9303", + "issue:938", + "issue:9825", + "issue:993" + ], + "soft_pairs": [ + "issue:3431|issue:3566", + "issue:5318|issue:5320", + "issue:4611|issue:4779", + "issue:6681|issue:7178", + "issue:12221|issue:13351", + "issue:2163|issue:2466", + "issue:6688|issue:9303", + "issue:5075|issue:5258", + "issue:2213|issue:3451", + "issue:10817|issue:8507", + "issue:2985|issue:5926", + "issue:10950|issue:9045", + "issue:5054|issue:5824", + "issue:4885|issue:5109", + "issue:4063|issue:5205", + "issue:4888|issue:5339", + "issue:775|issue:993", + "issue:321|issue:938", + "issue:11315|issue:11653", + "issue:12221|issue:9825", + "issue:10460|issue:10472", + "issue:12239|issue:9825", + "issue:4376|issue:5824", + "issue:2618|issue:2822", + "issue:3958|issue:5395" + ], + "prepared_review_unit_hash": "8a47365430f4e77200019c22e313de53f69138a2b123deb6dff7d8c6f8653ba4", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33214, + "estimated_input_tokens": 8304, + "estimated_eval_tokens": 16864 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly unrelated issues grouped by loose thematic similarity. I only consider two soft pairs as true duplicates: the fp16 dtype mismatch reports and the StableDiffusionXLControlNetImg2Img LoRA-loading reports.", + "confidence": 0.84, + "canonical_issue_reason": "No single issue cleanly represents the whole heterogeneous set; if one canonical issue must be chosen, #4885 is the clearest concrete bug report and anchors the strongest duplicate pair (#5109) around the same pipeline/LoRA-loading failure.", + "canonical_pr_reason": null, + "best_issue_reason": "#4885 is the best issue candidate because it is specific, reproducible, and narrowly scoped to a concrete pipeline bug, making it a strong triage anchor for the matching report in #5109.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3431", + "right": "issue:3566", + "accept": false, + "reason": "Both are ControlNet-related, but they describe different problems and different pipelines; too broad to be the same bug." + }, + { + "left": "issue:5318", + "right": "issue:5320", + "accept": false, + "reason": "One is docs/tutorials, the other is a PEFT API change; related topic, but not the same change request." + }, + { + "left": "issue:4611", + "right": "issue:4779", + "accept": false, + "reason": "Both mention ModelScope/video models, but they target different features and are not clearly the same underlying request." + }, + { + "left": "issue:6681", + "right": "issue:7178", + "accept": false, + "reason": "Same function name, but the reports are too generic to confirm the same concrete export bug." + }, + { + "left": "issue:12221", + "right": "issue:13351", + "accept": false, + "reason": "Different model families and different ControlNet support requests; not the same bug or change." + }, + { + "left": "issue:2163", + "right": "issue:2466", + "accept": false, + "reason": "Training dtype mismatch vs checkpoint retention behavior; unrelated issues." + }, + { + "left": "issue:6688", + "right": "issue:9303", + "accept": false, + "reason": "Different feature additions in the video space; no shared code-path bug is evident." + }, + { + "left": "issue:5075", + "right": "issue:5258", + "accept": false, + "reason": "Both involve SDXL/refiner, but one is a prompt-embedding shape error and the other is an inpaint/refiner pipeline issue." + }, + { + "left": "issue:2213", + "right": "issue:3451", + "accept": false, + "reason": "TPU distributed setup hang vs fp16 memory regression; unrelated failures." + }, + { + "left": "issue:10817", + "right": "issue:8507", + "accept": false, + "reason": "Both are SD3-related, but one is missing auto pipeline support and the other is a DreamBooth training failure." + }, + { + "left": "issue:2985", + "right": "issue:5926", + "accept": false, + "reason": "Different Kandinsky versions and different feature scopes; not duplicates." + }, + { + "left": "issue:10950", + "right": "issue:9045", + "accept": false, + "reason": "Documentation breakage vs runtime device-conversion failure; not the same underlying issue." + }, + { + "left": "issue:5054", + "right": "issue:5824", + "accept": false, + "reason": "Both are SDXL, but they concern different bugs in different methods." + }, + { + "left": "issue:4885", + "right": "issue:5109", + "accept": true, + "reason": "Same pipeline and same LoRA-loading path; one report says weights do not load, the other says the load call raises, which plausibly describes the same bug." + }, + { + "left": "issue:4063", + "right": "issue:5205", + "accept": false, + "reason": "Textual inversion unload vs inpainting/controlnet usage; different behaviors and code paths." + }, + { + "left": "issue:4888", + "right": "issue:5339", + "accept": false, + "reason": "Both concern local/offline safetensor loading, but the failure modes are too broad and not clearly the same bug." + }, + { + "left": "issue:775", + "right": "issue:993", + "accept": true, + "reason": "Both describe the same fp16/dtype mismatch class of Stable Diffusion failure, with nearly identical symptom framing." + }, + { + "left": "issue:321", + "right": "issue:938", + "accept": false, + "reason": "Completely different model domains: diffusion images vs text-to-motion." + }, + { + "left": "issue:11315", + "right": "issue:11653", + "accept": false, + "reason": "HiDream RoPE/MPS numeric issue vs LoRA key compatibility; different underlying problems." + }, + { + "left": "issue:12221", + "right": "issue:9825", + "accept": false, + "reason": "Both are feature requests around adapters/controlnets, but they target different model families and capabilities." + }, + { + "left": "issue:10460", + "right": "issue:10472", + "accept": false, + "reason": "A missing model init argument vs an unexpected LoRA keyword argument; unrelated bugs." + }, + { + "left": "issue:12239", + "right": "issue:9825", + "accept": false, + "reason": "InfiniteTalk support vs FLUX IPAdapter support; different features." + }, + { + "left": "issue:4376", + "right": "issue:5824", + "accept": false, + "reason": "Both are SDXL issues, but one is textual inversion loading and the other is a default-dimension bug." + }, + { + "left": "issue:2618", + "right": "issue:2822", + "accept": false, + "reason": "Input type handling for ControlNet vs training convergence performance; not the same issue." + }, + { + "left": "issue:3958", + "right": "issue:5395", + "accept": false, + "reason": "Both mention LoRA/Civitai/PEFT, but one is cpu-offload interaction and the other is combining LoRAs; different problems." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative: it avoids collapsing the heterogeneous set into one theme, and the two accepted soft pairs are reasonably supported by closely matching titles/symptoms. No soft-edge verdict looks overstated relative to the available evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 185, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10658", + "issue:1083", + "issue:10873", + "issue:1117", + "issue:11561", + "issue:12338", + "issue:1255", + "issue:1261", + "issue:13477", + "issue:1363", + "issue:1375", + "issue:1407", + "issue:1463", + "issue:1594", + "issue:1755", + "issue:1919", + "issue:1961", + "issue:2377", + "issue:2444", + "issue:2775", + "issue:2904", + "issue:3565", + "issue:3788", + "issue:460", + "issue:4789", + "issue:4912", + "issue:5008", + "issue:5317", + "issue:5318", + "issue:5480", + "issue:579", + "issue:5835", + "issue:6161", + "issue:6234", + "issue:6243", + "issue:7012", + "issue:7191", + "issue:7243", + "issue:7380", + "issue:7390", + "issue:7490", + "issue:7793", + "issue:8359", + "issue:8803", + "issue:9450", + "issue:9610", + "issue:9825", + "issue:989" + ], + "soft_pairs": [ + "issue:1261|issue:2444", + "issue:5317|issue:5318", + "issue:1755|issue:5008", + "issue:12338|issue:13477", + "issue:7390|issue:8359", + "issue:11561|issue:9825", + "issue:1594|issue:3788", + "issue:1255|issue:1375", + "issue:1363|issue:1407", + "issue:4789|issue:7793", + "issue:2904|issue:5835", + "issue:1463|issue:1919", + "issue:10658|issue:9610", + "issue:1083|issue:460", + "issue:10873|issue:9450", + "issue:1117|issue:2377", + "issue:5008|issue:5480", + "issue:5480|issue:7380", + "issue:6234|issue:8803", + "issue:6243|issue:7243", + "issue:7191|issue:7490", + "issue:3565|issue:7012", + "issue:1961|issue:579", + "issue:2775|issue:989", + "issue:4912|issue:6161" + ], + "prepared_review_unit_hash": "71afc8ffa10a11591e795aaa98ab6d70558a2dbb0dc22e60cf9e046313620518", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33821, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most candidate pairs are only superficially similar and do not describe the same bug or change. The only likely duplicate is the missing-`datasets` dependency report pair (1463/1919).", + "confidence": 0.72, + "canonical_issue_reason": "Issue 1919 is the better canonical for the `datasets` dependency problem because it identifies the root cause in `requirements.txt`, while 1463 mainly reports the resulting import error.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1919 is the clearest actionable issue in this set: it states the missing dependency/fix path directly and serves well as the duplicate target for the related import-error report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1261", + "right": "issue:2444", + "accept": false, + "reason": "Different failures: xformers backward kernel assert vs a DreamBooth script exit-status error." + }, + { + "left": "issue:5317", + "right": "issue:5318", + "accept": false, + "reason": "Same broad PEFT area, but one is API docs and the other is tutorials; not the same bug/change." + }, + { + "left": "issue:1755", + "right": "issue:5008", + "accept": false, + "reason": "Unrelated topics; no shared underlying bug or implementation." + }, + { + "left": "issue:12338", + "right": "issue:13477", + "accept": false, + "reason": "VAE slicing/decoding bug vs a callback `locals()` scope bug; different code paths." + }, + { + "left": "issue:7390", + "right": "issue:8359", + "accept": false, + "reason": "Both mention `from_single_file`, but they target different pipeline/custom-checkpoint scenarios and failures." + }, + { + "left": "issue:11561", + "right": "issue:9825", + "accept": false, + "reason": "Both involve IP-Adapter/FLUX support, but one is FluxFill-specific and the other is a broader FLUX-pipeline feature request." + }, + { + "left": "issue:1594", + "right": "issue:3788", + "accept": false, + "reason": "One asks for img2img video generation; the other requests a new pipeline implementation." + }, + { + "left": "issue:1255", + "right": "issue:1375", + "accept": false, + "reason": "Both are import-related, but one is a missing scheduler export and the other is a broader import regression from a PR." + }, + { + "left": "issue:1363", + "right": "issue:1407", + "accept": false, + "reason": "Different ONNX problems: long-prompt/LMS scheduler error vs missing `feature_extractor` in an SD2.0 ONNX pipeline." + }, + { + "left": "issue:4789", + "right": "issue:7793", + "accept": false, + "reason": "Download failure for DeepFloyd IF vs LoRA unload error; unrelated underlying bugs." + }, + { + "left": "issue:2904", + "right": "issue:5835", + "accept": false, + "reason": "AudioLDM scheduler attribute error vs general MusicLDM pipeline usage; not the same concrete issue." + }, + { + "left": "issue:1463", + "right": "issue:1919", + "accept": true, + "reason": "Both point to the same missing `datasets` dependency; 1919 names the root cause in `requirements.txt` and 1463 reports the import symptom." + }, + { + "left": "issue:10658", + "right": "issue:9610", + "accept": false, + "reason": "ONNXRuntime provider options vs adding LoRA loading to a Stable Diffusion 3 img2img pipeline; different changes." + }, + { + "left": "issue:1083", + "right": "issue:460", + "accept": false, + "reason": "Both are ONNX-related, but one is FP16 output correctness and the other is runtime/provider installation requirements." + }, + { + "left": "issue:10873", + "right": "issue:9450", + "accept": false, + "reason": "Different pipeline bugs: SanaPipeline issue vs FluxPipeline multi-GPU device placement error." + }, + { + "left": "issue:1117", + "right": "issue:2377", + "accept": false, + "reason": "Test-request for multiple negative prompts vs tests for EMAModel; unrelated scope." + }, + { + "left": "issue:5008", + "right": "issue:5480", + "accept": false, + "reason": "One is a specific TryOnDiffusion implementation issue; the other is a request for EDM support." + }, + { + "left": "issue:5480", + "right": "issue:7380", + "accept": false, + "reason": "Both are model/pipeline feature requests, but they are for different diffusion methods and not the same change." + }, + { + "left": "issue:6234", + "right": "issue:8803", + "accept": false, + "reason": "Different problems: SDXL text-to-image example issue vs a duplicate assignment in attention processor code." + }, + { + "left": "issue:6243", + "right": "issue:7243", + "accept": false, + "reason": "IP-Adapter Face ID support request vs ResAdapter implementation request; different features." + }, + { + "left": "issue:7191", + "right": "issue:7490", + "accept": false, + "reason": "Generic import problem vs a backward-pass problem; too little overlap to be the same bug." + }, + { + "left": "issue:3565", + "right": "issue:7012", + "accept": false, + "reason": "CUDA OOM during text-to-image training vs weird images with `v_prediction`; different failure modes." + }, + { + "left": "issue:1961", + "right": "issue:579", + "accept": false, + "reason": "Different missing-package/import issues and different affected workflows; not the same report." + }, + { + "left": "issue:2775", + "right": "issue:989", + "accept": false, + "reason": "`torch._dynamo.optimize` non-function error is unrelated to the `weight` must be 2-D runtime error." + }, + { + "left": "issue:4912", + "right": "issue:6161", + "accept": false, + "reason": "Custom SDXL checkpoint conversion request vs sigma-discard behavior; unrelated changes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only positive duplicate call, 1463/1919, is well supported by the titles and issue descriptions (missing `datasets` package vs missing dependency in `requirements.txt`). The other pair verdicts are appropriately cautious given the limited evidence in the packet. " + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 186, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10883", + "issue:10891", + "issue:10897", + "issue:11117", + "issue:11519", + "issue:1156", + "issue:12239", + "issue:12412", + "issue:1594", + "issue:2195", + "issue:222", + "issue:2673", + "issue:2676", + "issue:2722", + "issue:2799", + "issue:3178", + "issue:3221", + "issue:3537", + "issue:3623", + "issue:3728", + "issue:3817", + "issue:4232", + "issue:4250", + "issue:4524", + "issue:4702", + "issue:4837", + "issue:4969", + "issue:5034", + "issue:5168", + "issue:541", + "issue:5757", + "issue:5998", + "issue:6016", + "issue:6136", + "issue:7146", + "issue:7322", + "issue:7350", + "issue:7417", + "issue:8488", + "issue:8604", + "issue:8837", + "issue:8994", + "issue:9101", + "issue:9248", + "issue:9680", + "issue:9732", + "issue:9825" + ], + "soft_pairs": [ + "issue:2722|issue:3623", + "issue:12412|issue:9825", + "issue:10897|issue:9825", + "issue:4837|issue:6016", + "issue:10883|issue:11519", + "issue:1594|issue:2195", + "issue:7146|issue:9732", + "issue:2676|issue:4524", + "issue:8994|issue:9680", + "issue:10891|issue:11117", + "issue:4702|issue:9101", + "issue:6136|issue:7417", + "issue:5757|issue:7322", + "issue:11117|issue:8488", + "issue:8604|issue:8837", + "issue:1156|issue:222", + "issue:3537|issue:541", + "issue:3221|issue:7350", + "issue:3178|issue:5998", + "issue:4250|issue:5034", + "issue:3728|issue:3817", + "issue:4969|issue:9248", + "issue:10100|issue:12239", + "issue:2673|issue:2799", + "issue:4232|issue:5168" + ], + "prepared_review_unit_hash": "2d489d75bf7a2c841a4cb79fb786fce21dfb645a6eef81ee9bab2511cb2e6557", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33610, + "estimated_input_tokens": 8403, + "estimated_eval_tokens": 17062 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a collection of unrelated support requests and bug reports. The only close pair is the bf16/half type-mismatch reports, but the available evidence is still only soft similarity rather than a confirmed duplicate/root-cause match.", + "confidence": 0.64, + "canonical_issue_reason": "Issue 10891 is the broadest bfloat16-conversion bug report in the set, so it is the best tentative canonical issue, but the packet does not prove that 11117 is the same root cause.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10891 is the strongest issue candidate for grouping the bf16/half mismatch symptom, though only as a tentative umbrella for a related report rather than a confirmed duplicate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2722", + "right": "issue:3623", + "accept": false, + "reason": "Both are model-loading/conversion complaints, but they describe different failures and no shared concrete bug is evident." + }, + { + "left": "issue:12412", + "right": "issue:9825", + "accept": false, + "reason": "Separate feature requests for different model/pipeline support; same broad area, not the same change." + }, + { + "left": "issue:10897", + "right": "issue:9825", + "accept": false, + "reason": "Different requested model support items with no shared code-path or bug." + }, + { + "left": "issue:4837", + "right": "issue:6016", + "accept": false, + "reason": "Same conversion script, but the reported failures differ and do not clearly point to one underlying defect." + }, + { + "left": "issue:10883", + "right": "issue:11519", + "accept": false, + "reason": "Different model support/update requests; no common bug or merged change." + }, + { + "left": "issue:1594", + "right": "issue:2195", + "accept": false, + "reason": "Both concern video generation, but one is img2img video support and the other is Tune-A-Video; not the same artifact." + }, + { + "left": "issue:7146", + "right": "issue:9732", + "accept": false, + "reason": "Both mention OOM during training, but they are different training scripts/models and likely different memory issues." + }, + { + "left": "issue:2676", + "right": "issue:4524", + "accept": false, + "reason": "A scheduler compatibility question vs. an AnimateDiff integration request; unrelated changes." + }, + { + "left": "issue:8994", + "right": "issue:9680", + "accept": false, + "reason": "Both describe generation problems, but the symptoms and likely code paths are too different to merge as one bug." + }, + { + "left": "issue:10891", + "right": "issue:11117", + "accept": false, + "reason": "The titles suggest a related bf16/half mismatch, but the packet does not provide body evidence tying them to one confirmed duplicate or shared root cause; treat this as soft similarity only." + }, + { + "left": "issue:4702", + "right": "issue:9101", + "accept": false, + "reason": "Both ask for training scripts, but they target different models and training setups." + }, + { + "left": "issue:6136", + "right": "issue:7417", + "accept": false, + "reason": "Both involve NaN-related behavior, but one is Mac CPU inference and the other is a LoRA fuse test; not the same bug." + }, + { + "left": "issue:5757", + "right": "issue:7322", + "accept": false, + "reason": "Both are training errors, but they concern different scripts, versions, and failure modes." + }, + { + "left": "issue:11117", + "right": "issue:8488", + "accept": false, + "reason": "These are different StableDiffusion3Pipeline failures; one is a bf16 mismatch, the other is a from_pretrained exception." + }, + { + "left": "issue:8604", + "right": "issue:8837", + "accept": false, + "reason": "Different bugs in different components: corrupted output from T5 fp16 conversion vs. a sliced attention processor issue." + }, + { + "left": "issue:1156", + "right": "issue:222", + "accept": false, + "reason": "Generic training failures with insufficient evidence of a shared underlying defect." + }, + { + "left": "issue:3537", + "right": "issue:541", + "accept": false, + "reason": "A ControlNet conversion script request and a SageMaker support issue are unrelated." + }, + { + "left": "issue:3221", + "right": "issue:7350", + "accept": false, + "reason": "Different ControlNet/LoRA loading problems with distinct failure contexts." + }, + { + "left": "issue:3178", + "right": "issue:5998", + "accept": false, + "reason": "Same pipeline name, but one is a generic bug report and the other is a specific TyperError; not enough to treat as one change." + }, + { + "left": "issue:4250", + "right": "issue:5034", + "accept": false, + "reason": "Different symptoms: incorrect from_pretrained loading vs. a progress-bar display issue." + }, + { + "left": "issue:3728", + "right": "issue:3817", + "accept": false, + "reason": "Both are new pipeline requests, but they concern different projects and capabilities." + }, + { + "left": "issue:4969", + "right": "issue:9248", + "accept": false, + "reason": "An import error report and a pip install command request are not the same bug." + }, + { + "left": "issue:10100", + "right": "issue:12239", + "accept": false, + "reason": "Separate support requests for different models; only superficial similarity." + }, + { + "left": "issue:2673", + "right": "issue:2799", + "accept": false, + "reason": "One is an upscale pipeline usage error; the other is a request to raise a clearer mismatch error, not the same underlying bug." + }, + { + "left": "issue:4232", + "right": "issue:5168", + "accept": false, + "reason": "Both involve ControlNet-related work, but one is instruct-pix2pix training/inference and the other is ControlNet-XS support." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The writeup is mostly conservative on the unrelated pairs, but it overreaches on issue:10891 vs issue:11117. The packet does not provide body evidence linking them, and the soft-pair metadata marks that pair as deterministic_accept=false. Based on the available information, this should be treated as at most a plausible soft similarity, not a confirmed duplicate/root-cause match." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 187, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10150", + "issue:10737", + "issue:10795", + "issue:11374", + "issue:11642", + "issue:1188", + "issue:12599", + "issue:1263", + "issue:12695", + "issue:12728", + "issue:12754", + "issue:1305", + "issue:13279", + "issue:1375", + "issue:1633", + "issue:2737", + "issue:3043", + "issue:3171", + "issue:3325", + "issue:3725", + "issue:4063", + "issue:4213", + "issue:4232", + "issue:4302", + "issue:4390", + "issue:4450", + "issue:4480", + "issue:4496", + "issue:457", + "issue:4598", + "issue:4849", + "issue:4912", + "issue:4978", + "issue:541", + "issue:6008", + "issue:7569", + "issue:760", + "issue:761", + "issue:7646", + "issue:8435", + "issue:9063", + "issue:911", + "issue:9161", + "issue:9223", + "issue:9450", + "issue:9861", + "issue:9936" + ], + "soft_pairs": [ + "issue:6008|issue:9223", + "issue:1633|issue:2737", + "issue:761|issue:911", + "issue:11374|issue:12695", + "issue:3171|issue:4496", + "issue:4450|issue:4849", + "issue:4232|issue:4978", + "issue:12599|issue:13279", + "issue:4912|issue:541", + "issue:10795|issue:12728", + "issue:10737|issue:9063", + "issue:10150|issue:12754", + "issue:1305|issue:4063", + "issue:1263|issue:1375", + "issue:3725|issue:4302", + "issue:1188|issue:3043", + "issue:4480|issue:7569", + "issue:4213|issue:8435", + "issue:3325|issue:7646", + "issue:457|issue:760", + "issue:11642|issue:9861", + "issue:9161|issue:9936", + "issue:10737|issue:9450", + "issue:4390|issue:4598" + ], + "prepared_review_unit_hash": "31b130d70af874938b81ba1a3deedb1e5309014c667679f62aa568353af0ec65", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33086, + "estimated_input_tokens": 8272, + "estimated_eval_tokens": 16800 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is only loosely related by diffusion pipelines and transformer/loader issues, but the paired items are not the same underlying bug or change. Most are different pipelines, different hardware backends, or distinct feature requests, so none of the soft edges should be merged.", + "confidence": 0.31, + "canonical_issue_reason": "Issue #12754 is the clearest self-contained API bug in the set and is broadly representative of the pipeline-loading/casting theme, but the cluster overall is heterogeneous rather than a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #12754 is the strongest representative issue because it describes a specific, actionable regression in core pipeline behavior with clear expected-vs-actual semantics.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6008", + "right": "issue:9223", + "accept": false, + "reason": "Different problems: IP-Adapter/LCM face integration failure vs dependency/version conflict around pyarrow; unrelated symptoms and code paths." + }, + { + "left": "issue:1633", + "right": "issue:2737", + "accept": false, + "reason": "Both mention scheduler/features, but one is karras sigmas in a wrapper pipeline and the other is CLIP guidance support; different changes and APIs." + }, + { + "left": "issue:761", + "right": "issue:911", + "accept": false, + "reason": "Both are DreamBooth-related, but one is an 8-bit Adam training failure and the other is a bad config regression; not the same bug." + }, + { + "left": "issue:11374", + "right": "issue:12695", + "accept": false, + "reason": "SkyReels-V2 support and Hunyuan v1.5 support are separate model-integration feature requests, not the same change." + }, + { + "left": "issue:3171", + "right": "issue:4496", + "accept": false, + "reason": "LPW prompt support for ControlNet and prompt fusion are different prompt-handling features with no shared concrete bug." + }, + { + "left": "issue:4450", + "right": "issue:4849", + "accept": false, + "reason": "Multi textual inversion example failure and multi-GPU textual inversion training are unrelated issues despite both involving textual inversion." + }, + { + "left": "issue:4232", + "right": "issue:4978", + "accept": false, + "reason": "Instruct-pix2pix with ControlNet vs inpaint_global_harmonious preprocessor usage are different pipeline/preprocessor topics." + }, + { + "left": "issue:12599", + "right": "issue:13279", + "accept": false, + "reason": "Extremely slow CUDA move after loading vs AutoencoderRAE loading error with older transformers are different loading/runtime problems." + }, + { + "left": "issue:4912", + "right": "issue:541", + "accept": false, + "reason": "Custom SDXL checkpoint conversion and SageMaker support are separate workflow/environment requests, not the same bug or change." + }, + { + "left": "issue:10795", + "right": "issue:12728", + "accept": false, + "reason": "torch.compile on GGUF-loaded transformer models and Wan 2.2 compile/channels_last failures are distinct compiler/model-path issues." + }, + { + "left": "issue:10737", + "right": "issue:9063", + "accept": false, + "reason": "Missing image_encoder in FluxPipeline encode_image and tokenizer instantiation failure in FluxPipeline are different failures in different components." + }, + { + "left": "issue:10150", + "right": "issue:12754", + "accept": false, + "reason": "SDXL slow loading/inference is a performance complaint, while from_pipe float32 conversion is a dtype regression; not the same bug." + }, + { + "left": "issue:1305", + "right": "issue:4063", + "accept": false, + "reason": "Direct Inversion and unloading a textual inversion checkpoint are different features in the inversion/textual-inversion area, not duplicates." + }, + { + "left": "issue:1263", + "right": "issue:1375", + "accept": false, + "reason": "Importing StableDiffusionSafetyChecker and a PR breaking diffusers import are different import failures with different root causes." + }, + { + "left": "issue:3725", + "right": "issue:4302", + "accept": false, + "reason": "A1111 LoRA support failure and SDXL LoRA loading compatibility are both LoRA-related, but they are distinct compatibility problems." + }, + { + "left": "issue:1188", + "right": "issue:3043", + "accept": false, + "reason": "CLIPSeg x Stable Diffusion pipeline request and ODISE pipeline request are separate model integrations, not one change." + }, + { + "left": "issue:4480", + "right": "issue:7569", + "accept": false, + "reason": "Both are Mac/MPS-related, but one is a kernel crash during generation and the other is a failing test for FreeU; different concrete bugs." + }, + { + "left": "issue:4213", + "right": "issue:8435", + "accept": false, + "reason": "Both concern from_single_file, but one is specifically SD-XL without accelerate and the other is a generic runtime error; too broad to treat as the same bug." + }, + { + "left": "issue:3325", + "right": "issue:7646", + "accept": false, + "reason": "CUDA invalid-argument during DreamBooth training and dataloader_num_workers questions in a LoRA script are not the same underlying issue." + }, + { + "left": "issue:457", + "right": "issue:760", + "accept": false, + "reason": "Both are MPS-related, but one reports blue artifacts and the other unexpected StableDiffusionPipeline output; different symptoms and likely causes." + }, + { + "left": "issue:11642", + "right": "issue:9861", + "accept": false, + "reason": "prompt_embeds_scale having no effect in FluxPriorReduxPipeline and Flux training not updating the transformer are unrelated Flux issues." + }, + { + "left": "issue:9161", + "right": "issue:9936", + "accept": false, + "reason": "Distributed bugs in train_dreambooth_lora_flux.py and NCCL timeout in train_controlnet_flux.py are different multi-GPU training failures." + }, + { + "left": "issue:10737", + "right": "issue:9450", + "accept": false, + "reason": "Both are FluxPipeline issues, but one is a missing image encoder and the other is a multi-GPU device placement error; different code paths." + }, + { + "left": "issue:4390", + "right": "issue:4598", + "accept": false, + "reason": "Missing requirements and uninformative missing-dependency errors are both dependency-related, but one is a packaging issue and the other is an error-message improvement." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 188, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1076", + "issue:1082", + "issue:10952", + "issue:11050", + "issue:11133", + "issue:11182", + "issue:12052", + "issue:13034", + "issue:13488", + "issue:1367", + "issue:1608", + "issue:200", + "issue:2195", + "issue:2439", + "issue:2579", + "issue:2722", + "issue:2843", + "issue:3094", + "issue:3103", + "issue:3171", + "issue:322", + "issue:3348", + "issue:3406", + "issue:3630", + "issue:4186", + "issue:4232", + "issue:4445", + "issue:4747", + "issue:4949", + "issue:5002", + "issue:5118", + "issue:5258", + "issue:6141", + "issue:6901", + "issue:743", + "issue:7646", + "issue:8507", + "issue:8511", + "issue:8780", + "issue:9025", + "issue:9039", + "issue:9142", + "issue:938", + "issue:9539", + "issue:9576", + "issue:9660", + "issue:9895" + ], + "soft_pairs": [ + "issue:9539|issue:9895", + "issue:3630|issue:4747", + "issue:10952|issue:9142", + "issue:1082|issue:1608", + "issue:2722|issue:743", + "issue:4186|issue:5258", + "issue:4232|issue:5118", + "issue:2195|issue:4949", + "issue:2439|issue:3103", + "issue:11050|issue:11182", + "issue:9025|issue:9039", + "issue:2579|issue:3348", + "issue:8507|issue:8511", + "issue:9576|issue:9660", + "issue:3171|issue:6141", + "issue:4445|issue:8780", + "issue:11133|issue:13034", + "issue:1076|issue:322", + "issue:12052|issue:13488", + "issue:3094|issue:7646", + "issue:5002|issue:6901", + "issue:1367|issue:2843", + "issue:200|issue:3406", + "issue:322|issue:938" + ], + "prepared_review_unit_hash": "438da3e1339759bac4cb6cbdb67149cdb094ba0ae85cb06d299780f120d4d038", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33164, + "estimated_input_tokens": 8291, + "estimated_eval_tokens": 16838 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes many unrelated issues across different models, pipelines, training scripts, and years. None of the soft pairs look like the same concrete bug/change, so all should stay separate.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue: the items span unrelated topics (scheduler APIs, LoRA loading, offloading, ControlNet, video pipelines, training bugs, and model conversion) with no shared underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good cluster representative because the cluster is heterogeneous and does not center on one actionable problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9539", + "right": "issue:9895", + "accept": false, + "reason": "Both involve device/type failures in video pipelines, but one is Float8 index_select on CogVideoX and the other is a Flux offloading device-mismatch regression; different code paths and fixes." + }, + { + "left": "issue:3630", + "right": "issue:4747", + "accept": false, + "reason": "Both are runtime/load errors, but one is an unknown UNet2DModel issue and the other is missing model_index.json after loading an SDXL LoRA; not the same bug." + }, + { + "left": "issue:10952", + "right": "issue:9142", + "accept": false, + "reason": "Both mention adapter/model loading failures, but one is SDXL LoRA adapter-name validation and the other is Hunyuan DiT multiple-ControlNet loading; different mechanisms." + }, + { + "left": "issue:1082", + "right": "issue:1608", + "accept": false, + "reason": "Both are training-related failures, but one is a Dreambooth CUDA matmul crash and the other is training stopping at 500 steps; no shared root cause." + }, + { + "left": "issue:2722", + "right": "issue:743", + "accept": false, + "reason": "Both concern model conversion/loading, but one is ControlNet UNet mismatch after conversion and the other is generic ckpt-to-diffusers conversion failure; too broad to merge." + }, + { + "left": "issue:4186", + "right": "issue:5258", + "accept": false, + "reason": "Both mention SDXL inpaint/refiner, but one is inpainting with refiner and the other is ControlNet inpaint with refiner; different pipelines and failure modes." + }, + { + "left": "issue:4232", + "right": "issue:5118", + "accept": false, + "reason": "Both relate to controlnet/instruct-pix2pix style usage, but one is training/inference code and the other is a reference_only usage question; not the same bug." + }, + { + "left": "issue:2195", + "right": "issue:4949", + "accept": false, + "reason": "These are unrelated video topics: Tune-A-Video versus ProPainter in/outpainting; no common defect." + }, + { + "left": "issue:2439", + "right": "issue:3103", + "accept": false, + "reason": "Both involve inpainting, but one is loading/running the inpainting pipeline and the other is the height/width parameter behavior; different issues." + }, + { + "left": "issue:11050", + "right": "issue:11182", + "accept": false, + "reason": "One is a missing dtype cast in an SD3 training script; the other is an unnecessary BF16 variant download. Different code paths and fixes." + }, + { + "left": "issue:9025", + "right": "issue:9039", + "accept": false, + "reason": "Both are Flax pipeline questions, but one is from_single_file on TPU and the other is load_lora_weights in FlaxStableDiffusionPipeline; separate features." + }, + { + "left": "issue:2579", + "right": "issue:3348", + "accept": false, + "reason": "Both mention xFormers and LoRA, but one is a possible xFormers bug/misuse and the other is missing scale parameter support for LoRA with xFormers; not the same defect." + }, + { + "left": "issue:8507", + "right": "issue:8511", + "accept": false, + "reason": "Closely related SD3 training topics, but one is train_dreambooth_sd3.py failing with --train_text_encoder and the other is text encoders not being unloaded; distinct fixes." + }, + { + "left": "issue:9576", + "right": "issue:9660", + "accept": false, + "reason": "Both are pipeline-loading regressions, but one is use_safetensors not propagated to submodules and the other is modules not switched to eval mode; different behaviors." + }, + { + "left": "issue:3171", + "right": "issue:6141", + "accept": false, + "reason": "Both concern ControlNet parameters, but one is LPW prompt handling and the other is control_mode/resize_mode support; unrelated feature gaps." + }, + { + "left": "issue:4445", + "right": "issue:8780", + "accept": false, + "reason": "Both are vague model/data-type usage problems, but one asks how to finetune LoRA and the other reports model/input dtype mismatch; not the same bug." + }, + { + "left": "issue:11133", + "right": "issue:13034", + "accept": false, + "reason": "Both are Wan-related, but one is a cogvideox image-to-video bug and the other is WanPipeline height/width handling; different pipelines and symptoms." + }, + { + "left": "issue:1076", + "right": "issue:322", + "accept": false, + "reason": "Completely unrelated topics: stable-diffusion-v1-5 discussion versus V-Diffusion." + }, + { + "left": "issue:12052", + "right": "issue:13488", + "accept": false, + "reason": "Both mention offloading, but one is a concrete tensor-device mismatch in Wan 2.2 LightX2V offloading and the other is a broad diffusers 0.38.0.dev0 offloading regression; too broad to treat as one bug." + }, + { + "left": "issue:3094", + "right": "issue:7646", + "accept": false, + "reason": "Both are training script questions, but one is CUDA OOM on a finetuning example and the other is dataloader_num_workers behavior; different issues." + }, + { + "left": "issue:5002", + "right": "issue:6901", + "accept": false, + "reason": "Both are scheduler API gaps, but one is missing use_karras_sigmas and the other is missing clip_sample in FlaxDDIMScheduler; separate implementations." + }, + { + "left": "issue:1367", + "right": "issue:2843", + "accept": false, + "reason": "Both involve saving configuration/checkpoints, but one is scheduler config with custom trained_betas and the other is multi-GPU save checkpoint; not the same bug." + }, + { + "left": "issue:200", + "right": "issue:3406", + "accept": false, + "reason": "Not the same problem: one is a test request for resnet class, the other is Attention class behavior with only_cross_attention=True." + }, + { + "left": "issue:322", + "right": "issue:938", + "accept": false, + "reason": "Completely unrelated projects/models: Katherine's V-Diffusion and Human Motion Diffusion Model." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 189, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10144", + "issue:10336", + "issue:10460", + "issue:10608", + "issue:10653", + "issue:10658", + "issue:10710", + "issue:1082", + "issue:10848", + "issue:10873", + "issue:10883", + "issue:1099", + "issue:11117", + "issue:11396", + "issue:11700", + "issue:12009", + "issue:12037", + "issue:12666", + "issue:1405", + "issue:1478", + "issue:1552", + "issue:2392", + "issue:2830", + "issue:3451", + "issue:3624", + "issue:3728", + "issue:3944", + "issue:4269", + "issue:4686", + "issue:5128", + "issue:5205", + "issue:5605", + "issue:6357", + "issue:6503", + "issue:660", + "issue:6891", + "issue:713", + "issue:7243", + "issue:7333", + "issue:7646", + "issue:775", + "issue:8500", + "issue:8565", + "issue:8588", + "issue:8759", + "issue:8939", + "issue:9063", + "issue:9166" + ], + "soft_pairs": [ + "issue:8939|issue:9063", + "issue:3451|issue:7646", + "issue:3624|issue:4269", + "issue:3944|issue:5605", + "issue:10883|issue:7243", + "issue:10658|issue:11700", + "issue:10144|issue:12666", + "issue:8500|issue:8588", + "issue:1082|issue:713", + "issue:1478|issue:4686", + "issue:10653|issue:10710", + "issue:3728|issue:7333", + "issue:10460|issue:10848", + "issue:1405|issue:1552", + "issue:660|issue:775", + "issue:1099|issue:2392", + "issue:8759|issue:9166", + "issue:10608|issue:11396", + "issue:6503|issue:6891", + "issue:5128|issue:6357", + "issue:12009|issue:12037", + "issue:2830|issue:5205", + "issue:11117|issue:8565", + "issue:10336|issue:10873" + ], + "prepared_review_unit_hash": "335d6c27591f1b6471e461ab0ee7a519143b4e4fe66bb6d6b2aebb090ac94109", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34427, + "estimated_input_tokens": 8607, + "estimated_eval_tokens": 17470 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated Diffusers issues spanning different models, pipelines, training scripts, and feature requests. Only the attention-slicing reports look likely to be the same underlying bug; the rest are too model- or feature-specific to merge as duplicates.", + "confidence": 0.9, + "canonical_issue_reason": "Issue #1405 is the broadest and most representative report here: it describes the core attention-slicing breakage on main, which also subsumes the more specific SD2 report.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #1405 is the best single issue to keep as the cluster representative because it is the most general statement of the duplicate-worthy bug and points at the shared code path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8939", + "right": "issue:9063", + "accept": false, + "reason": "Both are tokenizer-loading failures, but for different pipelines/models and different error modes; not the same concrete bug." + }, + { + "left": "issue:3451", + "right": "issue:7646", + "accept": false, + "reason": "One is fp16 memory usage in training, the other is a dataloader worker question; unrelated." + }, + { + "left": "issue:3624", + "right": "issue:4269", + "accept": false, + "reason": "Both mention TypeError, but they hit different validation paths and different APIs." + }, + { + "left": "issue:3944", + "right": "issue:5605", + "accept": false, + "reason": "Both are Dreambooth-related feature requests, but one is LoRA rank customization and the other is per-image captions; distinct changes." + }, + { + "left": "issue:10883", + "right": "issue:7243", + "accept": false, + "reason": "Different pipelines/models and different problems; no shared underlying bug." + }, + { + "left": "issue:10658", + "right": "issue:11700", + "accept": false, + "reason": "One asks for provider_options in ONNX Runtime session creation; the other requests Pruna integration for from_pretrained/pipeline." + }, + { + "left": "issue:10144", + "right": "issue:12666", + "accept": false, + "reason": "Both are video-model issues, but they concern different model families and different problems." + }, + { + "left": "issue:8500", + "right": "issue:8588", + "accept": false, + "reason": "Both are SD3-related, but one is about token limits and the other about a missing conversion script." + }, + { + "left": "issue:1082", + "right": "issue:713", + "accept": false, + "reason": "Too generic to treat as the same bug; one is a specific CUDA matmul failure in Dreambooth and the other is a broad training error." + }, + { + "left": "issue:1478", + "right": "issue:4686", + "accept": false, + "reason": "Different StableDiffusionUpscalePipeline failures: output count vs loading a saved pipeline." + }, + { + "left": "issue:10653", + "right": "issue:10710", + "accept": false, + "reason": "Completely different topics: 8-bit module device transfer vs DDUF format support." + }, + { + "left": "issue:3728", + "right": "issue:7333", + "accept": false, + "reason": "Both are video-generation related, but they are different pipelines/feature requests rather than the same bug." + }, + { + "left": "issue:10460", + "right": "issue:10848", + "accept": false, + "reason": "Missing UNet argument and an unexpected keyword in IP-Adapter loading are separate API issues." + }, + { + "left": "issue:1405", + "right": "issue:1552", + "accept": true, + "reason": "These describe the same attention-slicing breakage; #1552 is the SD2-specific manifestation of the broader failure reported in #1405." + }, + { + "left": "issue:660", + "right": "issue:775", + "accept": false, + "reason": "Both involve fp16, but one is MPS-specific revision/dtype behavior and the other is a broader autocast-related problem." + }, + { + "left": "issue:1099", + "right": "issue:2392", + "accept": false, + "reason": "Unrelated Dreambooth class-name issue vs img2img prompt_embeds validation error." + }, + { + "left": "issue:8759", + "right": "issue:9166", + "accept": false, + "reason": "Both touch image_processor, but they concern different methods and symptoms." + }, + { + "left": "issue:10608", + "right": "issue:11396", + "accept": false, + "reason": "Different domains entirely: multi-GPU pipeline execution vs LoRA format conversion." + }, + { + "left": "issue:6503", + "right": "issue:6891", + "accept": false, + "reason": "Torch.compile training compatibility and model-card saving are unrelated training-script topics." + }, + { + "left": "issue:5128", + "right": "issue:6357", + "accept": false, + "reason": "Loading a local safetensors file is unrelated to API-vs-local result differences." + }, + { + "left": "issue:12009", + "right": "issue:12037", + "accept": false, + "reason": "Both are Wan 2.2 issues, but one is single-file GGUF loading and the other is LoRA compatibility; different code paths." + }, + { + "left": "issue:2830", + "right": "issue:5205", + "accept": false, + "reason": "No shared underlying bug; one asks about video generation, the other is a StableDiffusion repaint/ControlNet feature request." + }, + { + "left": "issue:11117", + "right": "issue:8565", + "accept": false, + "reason": "Same model family, but different failures: dtype mismatch at runtime vs an IndexError in LoRA loading." + }, + { + "left": "issue:10336", + "right": "issue:10873", + "accept": false, + "reason": "Both are Sana-related, but activation explosion/clamping and a specific SanaPipeline issue are not the same defect." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The representative choice and cluster summary are grounded in the packet, and the soft-edge verdicts stay conservative overall. The only accepted duplicate pair (#1405/#1552) is well supported by the shared attention-slicing failure, while the rest are reasonably rejected as model-, pipeline-, or feature-specific rather than clearly the same bug." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 190, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10045", + "issue:10111", + "issue:10382", + "issue:10392", + "issue:10512", + "issue:10772", + "issue:10950", + "issue:10952", + "issue:11229", + "issue:11675", + "issue:11790", + "issue:1195", + "issue:13065", + "issue:13137", + "issue:13221", + "issue:1478", + "issue:1538", + "issue:1594", + "issue:2689", + "issue:3162", + "issue:3838", + "issue:3884", + "issue:4000", + "issue:4210", + "issue:4349", + "issue:4356", + "issue:462", + "issue:4692", + "issue:4724", + "issue:4839", + "issue:5573", + "issue:5997", + "issue:6439", + "issue:6790", + "issue:7000", + "issue:7210", + "issue:7681", + "issue:7782", + "issue:8723", + "issue:8826", + "issue:8850", + "issue:8876", + "issue:9096", + "issue:9343", + "issue:9415", + "issue:9461", + "issue:9648", + "issue:993" + ], + "soft_pairs": [ + "issue:5573|issue:5997", + "issue:3884|issue:4839", + "issue:10111|issue:8876", + "issue:1195|issue:993", + "issue:4356|issue:7000", + "issue:6790|issue:7782", + "issue:4210|issue:4692", + "issue:10952|issue:8826", + "issue:11675|issue:11790", + "issue:10772|issue:13065", + "issue:2689|issue:6439", + "issue:9343|issue:9415", + "issue:10512|issue:9461", + "issue:462|issue:4724", + "issue:1594|issue:3838", + "issue:10382|issue:9648", + "issue:10950|issue:7210", + "issue:1538|issue:3162", + "issue:13137|issue:13221", + "issue:1478|issue:4349", + "issue:4000|issue:8723", + "issue:10045|issue:7681", + "issue:10950|issue:8850", + "issue:10392|issue:9461", + "issue:11229|issue:9096" + ], + "prepared_review_unit_hash": "1e8291432fdbd5c4db84ecf1419a27a89d2669f7b0c7f55651c3000140005fe6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34326, + "estimated_input_tokens": 8582, + "estimated_eval_tokens": 17420 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous; the soft pairs mostly share only broad subsystem keywords and do not look like the same underlying bug or change.", + "confidence": 0.87, + "canonical_issue_reason": "issue:10512 is the strongest representative issue in the cluster: it is a clear, highly active LoRA-loading failure report with concrete Flux/Quanto context and more discussion than the other similar-looking LoRA issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10512 is the best issue to keep as the representative bug report because it is specific, well-scoped, and has the most user activity among the candidate issue reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5573", + "right": "issue:5997", + "accept": false, + "reason": "Both are ControlNet/SDXL-adjacent, but one is a performance/init-time issue and the other is a model-type mismatch in a tutorial; different bugs." + }, + { + "left": "issue:3884", + "right": "issue:4839", + "accept": false, + "reason": "One is a training script failure and the other is a type error around lora_scale; no evidence they are the same concrete defect." + }, + { + "left": "issue:10111", + "right": "issue:8876", + "accept": false, + "reason": "Different pipelines and different symptoms; this is only shared FLUX/Pipeline terminology, not the same bug." + }, + { + "left": "issue:1195", + "right": "issue:993", + "accept": false, + "reason": "Both mention fp16/dtype problems, but one is about memory-efficient attention and the other is a general stable diffusion dtype mismatch." + }, + { + "left": "issue:4356", + "right": "issue:7000", + "accept": false, + "reason": "A module import breakage vs a PyTorch deprecation warning; unrelated issues." + }, + { + "left": "issue:6790", + "right": "issue:7782", + "accept": false, + "reason": "Unexpected state_dict keys during training is unrelated to a safetensors/UnpicklingError during conversion." + }, + { + "left": "issue:4210", + "right": "issue:4692", + "accept": false, + "reason": "One asks how to use attention_mask; the other is a memory issue. No shared bug." + }, + { + "left": "issue:10952", + "right": "issue:8826", + "accept": false, + "reason": "LoRA adapter-name loading error vs torch.cat empty-list runtime error; different code paths and failures." + }, + { + "left": "issue:11675", + "right": "issue:11790", + "accept": false, + "reason": "Both are LoRA-loading complaints, but the titles point to different contexts and no concrete evidence shows the same root cause." + }, + { + "left": "issue:10772", + "right": "issue:13065", + "accept": false, + "reason": "A ControlNet support request for Sana vs a Qwen-Image controlnet request; these are different feature requests." + }, + { + "left": "issue:2689", + "right": "issue:6439", + "accept": false, + "reason": "Both are broken links, but they are different docs issues with no indication of the same link or page." + }, + { + "left": "issue:9343", + "right": "issue:9415", + "accept": false, + "reason": "FLUX loading with mismatched sizes is unrelated to adding a new parameter module in CogVideoXTransformer3DModel." + }, + { + "left": "issue:10512", + "right": "issue:9461", + "accept": false, + "reason": "One is a specific LoRA load failure; the other is an \u0905\u0938\u094d\u092a\u0937\u094d\u091f/unfinished FLUX inference issue title with no matching bug detail." + }, + { + "left": "issue:462", + "right": "issue:4724", + "accept": false, + "reason": "Both involve img2img pipelines, but one is an Apple MPS regression and the other is an SDXL checkpoint behavior mismatch." + }, + { + "left": "issue:1594", + "right": "issue:3838", + "accept": false, + "reason": "A feature request for img2img video generation is not the same as a Video Composer request." + }, + { + "left": "issue:10382", + "right": "issue:9648", + "accept": false, + "reason": "A VAE forward return-type AttributeError is different from a VAE tiling memory overflow." + }, + { + "left": "issue:10950", + "right": "issue:7210", + "accept": false, + "reason": "A broken documentation section and a runtime accelerate dispatch error are unrelated." + }, + { + "left": "issue:1538", + "right": "issue:3162", + "accept": false, + "reason": "StableDiffusionMegaPipeline checkpoint compatibility and a classifier-free-sampling embedding error are different problems." + }, + { + "left": "issue:13137", + "right": "issue:13221", + "accept": false, + "reason": "Both are LoRA support/load issues, but they concern different model families and different error descriptions; not enough to treat as the same bug." + }, + { + "left": "issue:1478", + "right": "issue:4349", + "accept": false, + "reason": "Both are upscale-pipeline issues, but one is multi-image output handling and the other is text_encoder/prompt_embeds handling." + }, + { + "left": "issue:4000", + "right": "issue:8723", + "accept": false, + "reason": "Checkpoint-to-diffusers conversion is not the same as SD3ControlNetModel initialization failure." + }, + { + "left": "issue:10045", + "right": "issue:7681", + "accept": false, + "reason": "Both are import-related, but one is a generic pipeline import problem and the other is a specific OpenSSL/lib symbol failure." + }, + { + "left": "issue:10950", + "right": "issue:8850", + "accept": false, + "reason": "Broken docs section vs broken community pipelines mirror; different artifacts and failures." + }, + { + "left": "issue:10392", + "right": "issue:9461", + "accept": false, + "reason": "Both mention FLUX, but one is a specific fp8/lora setup and the other is an unclear FLUX inference issue; not the same bug." + }, + { + "left": "issue:11229", + "right": "issue:9096", + "accept": false, + "reason": "Both are NaN reports, but they affect different models, devices, and mechanisms (attention slicing on MPS vs half-precision Flux inference)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 191, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10025", + "issue:10382", + "issue:10514", + "issue:10710", + "issue:1083", + "issue:10920", + "issue:11023", + "issue:11275", + "issue:11283", + "issue:11709", + "issue:12126", + "issue:12529", + "issue:1475", + "issue:1534", + "issue:1565", + "issue:1608", + "issue:2618", + "issue:2941", + "issue:3140", + "issue:3250", + "issue:3405", + "issue:3728", + "issue:3891", + "issue:4159", + "issue:4314", + "issue:4390", + "issue:4542", + "issue:5429", + "issue:5462", + "issue:5510", + "issue:5523", + "issue:5606", + "issue:5769", + "issue:5772", + "issue:5877", + "issue:6368", + "issue:668", + "issue:6869", + "issue:7178", + "issue:7243", + "issue:7377", + "issue:7405", + "issue:7408", + "issue:7968", + "issue:7972", + "issue:8790", + "issue:9023", + "issue:9496" + ], + "soft_pairs": [ + "issue:10382|issue:9496", + "issue:4159|issue:4390", + "issue:12126|issue:12529", + "issue:11023|issue:11275", + "issue:10025|issue:7968", + "issue:1565|issue:1608", + "issue:5510|issue:7972", + "issue:10920|issue:9023", + "issue:11283|issue:11709", + "issue:5606|issue:6368", + "issue:1083|issue:2941", + "issue:5429|issue:5462", + "issue:10514|issue:10710", + "issue:1534|issue:668", + "issue:5769|issue:5772", + "issue:4542|issue:7377", + "issue:1475|issue:5877", + "issue:6869|issue:7178", + "issue:3891|issue:7408", + "issue:7243|issue:8790", + "issue:3728|issue:7405", + "issue:3405|issue:4314", + "issue:2618|issue:3140", + "issue:3250|issue:5523" + ], + "prepared_review_unit_hash": "676958d4fa95aeec2134876c4af9a28e8c5d649a6aa20780b61314bbb55f68df", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33664, + "estimated_input_tokens": 8416, + "estimated_eval_tokens": 17088 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly false-positive similarities: the items span unrelated bugs, feature requests, and model-specific failures, so there is no real duplicate cluster to collapse.", + "confidence": 0.96, + "canonical_issue_reason": "No clear canonical issue exists here because the candidates are not the same underlying bug or change; they cover different subsystems and failure modes.", + "canonical_pr_reason": "No pull requests are present in this set.", + "best_issue_reason": "No single issue is a good global representative for the cluster; if forced to pick a standalone diagnostic report, #10382 is the most detailed, but it does not match the rest.", + "best_pr_reason": "No pull requests are present in this set.", + "soft_edge_verdicts": [ + { + "left": "issue:10382", + "right": "issue:9496", + "accept": false, + "reason": "Both mention forward/type errors in VAE-related code, but they are different models and different failure paths." + }, + { + "left": "issue:4159", + "right": "issue:4390", + "accept": false, + "reason": "VRAM blow-up during DiffEdit is unrelated to missing requirements." + }, + { + "left": "issue:12126", + "right": "issue:12529", + "accept": false, + "reason": "Both are Wan2.2 issues, but one is a cache_context NoneType error and the other is a tiled VAE tensor mismatch." + }, + { + "left": "issue:11023", + "right": "issue:11275", + "accept": false, + "reason": "Both are image-quality/runtime problems, but they involve different pipelines and distinct root causes." + }, + { + "left": "issue:10025", + "right": "issue:7968", + "accept": false, + "reason": "They only share the word attention; one is about Flux transformer attention masks, the other about an UNet motion-module attention component." + }, + { + "left": "issue:1565", + "right": "issue:1608", + "accept": false, + "reason": "Both concern training on GPUs, but one is GPU selection/OOM and the other is a training stop at 500 steps." + }, + { + "left": "issue:5510", + "right": "issue:7972", + "accept": false, + "reason": "Both are type-mismatch reports, but they affect different pipelines and different tensor/bias handling." + }, + { + "left": "issue:10920", + "right": "issue:9023", + "accept": false, + "reason": "Both are generic runtime failures, but the scheduler inf bug is unrelated to sample code execution failures." + }, + { + "left": "issue:11283", + "right": "issue:11709", + "accept": false, + "reason": "A header deserialization error and a LoRA key error are different loading problems with no shared code path shown." + }, + { + "left": "issue:5606", + "right": "issue:6368", + "accept": false, + "reason": "Both involve LoRA alpha handling, but one is missing alpha value and the other is unsupported checkpoint layer naming." + }, + { + "left": "issue:1083", + "right": "issue:2941", + "accept": false, + "reason": "FP16 ONNX output correctness is unrelated to EulerAncestralDiscreteScheduler behavior in Stable Diffusion." + }, + { + "left": "issue:5429", + "right": "issue:5462", + "accept": false, + "reason": "One is about supporting models without VAE in examples, the other is about adding img2img for LCMs." + }, + { + "left": "issue:10514", + "right": "issue:10710", + "accept": false, + "reason": "Sana resolution binning support is unrelated to DDUF format support." + }, + { + "left": "issue:1534", + "right": "issue:668", + "accept": false, + "reason": "Both are DreamBooth notebook/training issues, but they do not describe the same failure mode." + }, + { + "left": "issue:5769", + "right": "issue:5772", + "accept": false, + "reason": "Mixed-precision training correctness and dataset performance are unrelated." + }, + { + "left": "issue:4542", + "right": "issue:7377", + "accept": false, + "reason": "Both are multi-image generation problems, but they affect different pipelines and different internal components." + }, + { + "left": "issue:1475", + "right": "issue:5877", + "accept": false, + "reason": "DreamBooth/textual inversion class-image handling is unrelated to subprocess errors in those training scripts." + }, + { + "left": "issue:6869", + "right": "issue:7178", + "accept": false, + "reason": "Both mention export_to_video, but one is a concrete unpacking ValueError and the other is a generic malfunction report; not enough to treat as the same bug." + }, + { + "left": "issue:3891", + "right": "issue:7408", + "accept": false, + "reason": "These are separate pipeline feature requests for different models (Tortoise TTS vs FRESCO)." + }, + { + "left": "issue:7243", + "right": "issue:8790", + "accept": false, + "reason": "ResAdapter and SDXL TensorRT/IP-Adapter are different model-integration topics." + }, + { + "left": "issue:3728", + "right": "issue:7405", + "accept": false, + "reason": "They are different new-model requests: Shifted Diffusion vs DynamiCrafter." + }, + { + "left": "issue:3405", + "right": "issue:4314", + "accept": false, + "reason": "ControlNet training errors and interrupted-download write protection are unrelated." + }, + { + "left": "issue:2618", + "right": "issue:3140", + "accept": false, + "reason": "Passing torch.FloatTensor image inputs to ControlNet is unrelated to the import error for StableDiffusionControlNetPipeline." + }, + { + "left": "issue:3250", + "right": "issue:5523", + "accept": false, + "reason": "A pipeline execution failure and inconsistent results are too generic and do not indicate the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 192, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10050", + "issue:10054", + "issue:10500", + "issue:10573", + "issue:10639", + "issue:10653", + "issue:10937", + "issue:11023", + "issue:11223", + "issue:11229", + "issue:11283", + "issue:11342", + "issue:11489", + "issue:11749", + "issue:12113", + "issue:12348", + "issue:12369", + "issue:12752", + "issue:3159", + "issue:3880", + "issue:4368", + "issue:4616", + "issue:5335", + "issue:5710", + "issue:6095", + "issue:6363", + "issue:6439", + "issue:6463", + "issue:6695", + "issue:7180", + "issue:7303", + "issue:7418", + "issue:7456", + "issue:7562", + "issue:7629", + "issue:7636", + "issue:7681", + "issue:7848", + "issue:7931", + "issue:8560", + "issue:8850", + "issue:9025", + "issue:9031", + "issue:9034", + "issue:9397", + "issue:9680", + "issue:9729", + "issue:9933" + ], + "soft_pairs": [ + "issue:12113|issue:12348", + "issue:7180|issue:7562", + "issue:10050|issue:12752", + "issue:10054|issue:9933", + "issue:3880|issue:5710", + "issue:4368|issue:8560", + "issue:6363|issue:6695", + "issue:6463|issue:7456", + "issue:7303|issue:9397", + "issue:7681|issue:9034", + "issue:6095|issue:7629", + "issue:10573|issue:7931", + "issue:10500|issue:10653", + "issue:8850|issue:9680", + "issue:11023|issue:7848", + "issue:5335|issue:7418", + "issue:9025|issue:9031", + "issue:4616|issue:7636", + "issue:11229|issue:11342", + "issue:10639|issue:12369", + "issue:11489|issue:9729", + "issue:3159|issue:6439", + "issue:10937|issue:11749", + "issue:11223|issue:11283" + ], + "prepared_review_unit_hash": "df8e04adb95524e6a787484ec613da6c4312ec78fb5f68566ec88c7e93b00f13", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34264, + "estimated_input_tokens": 8566, + "estimated_eval_tokens": 17388 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly heterogeneous issue set, with a handful of likely duplicates around WanImageToVideoPipeline tensor mismatches, SDXL ControlNet CPU offload, lpw_stable_diffusion_xl output corruption, Flux LoRA training, and HunyuanVideo bitsandbytes device handling.", + "confidence": 0.64, + "canonical_issue_reason": "Issue 12113 is the clearest concrete bug report in the set, and 12348 reads like a broader follow-up of the same WanImageToVideoPipeline resolution-specific tensor mismatch.", + "canonical_pr_reason": null, + "best_issue_reason": "12113 has the most specific repro/error message and is the best representative among the true duplicate-like pairs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12113", + "right": "issue:12348", + "accept": true, + "reason": "Same WanImageToVideoPipeline tensor/channel mismatch at certain resolutions; 12348 is the broader symptom report." + }, + { + "left": "issue:7180", + "right": "issue:7562", + "accept": true, + "reason": "Both describe SDXL ControlNet Img2Img breaking under CPU offload; 7562 is the multicontrolnet-specific form of the same bug." + }, + { + "left": "issue:10050", + "right": "issue:12752", + "accept": false, + "reason": "Different requests for different pipelines/models: img2img KDiffusion support vs Z-Image img2img/inpainting support." + }, + { + "left": "issue:10054", + "right": "issue:9933", + "accept": false, + "reason": "Scheduler torch.linspace parameter bug vs missing width/height parameters in SD3 img2img are unrelated." + }, + { + "left": "issue:3880", + "right": "issue:5710", + "accept": false, + "reason": "Inpainting quality degradation and missing sigmas initialization are different code paths." + }, + { + "left": "issue:4368", + "right": "issue:8560", + "accept": false, + "reason": "Both are OOM reports, but for different models and training setups (SDXL LoRA vs SD3 training)." + }, + { + "left": "issue:6363", + "right": "issue:6695", + "accept": false, + "reason": "train_text_to_image_lora AMP unscale error is unrelated to the subprocess CalledProcessError report." + }, + { + "left": "issue:6463", + "right": "issue:7456", + "accept": false, + "reason": "Missing audio_diffusion pipeline and a missing raise statement are not the same bug." + }, + { + "left": "issue:7303", + "right": "issue:9397", + "accept": false, + "reason": "Both mention schedulers, but one is about zero_snr creation and the other about Hyper LoRA support; not the same change." + }, + { + "left": "issue:7681", + "right": "issue:9034", + "accept": false, + "reason": "OpenSSL import failure and an unexpected sentencepiece dependency are unrelated issues." + }, + { + "left": "issue:6095", + "right": "issue:7629", + "accept": false, + "reason": "Euler discrete scheduling IndexError is distinct from PyTorch-version-specific OOM in SVD." + }, + { + "left": "issue:10573", + "right": "issue:7931", + "accept": false, + "reason": "Both are ControlNet inpaint issues, but one is padding_mask_crop with multiple controlnets and the other is IP-Adapter embeds." + }, + { + "left": "issue:10500", + "right": "issue:10653", + "accept": true, + "reason": "Same HunyuanVideo bitsandbytes/device mismatch class; both point to 8bit model CPU/GPU placement problems." + }, + { + "left": "issue:8850", + "right": "issue:9680", + "accept": false, + "reason": "Community pipelines mirror breakage is unrelated to SDXL unconditional generation." + }, + { + "left": "issue:11023", + "right": "issue:7848", + "accept": true, + "reason": "Both point to lpw_stable_diffusion_xl producing degraded/blurred output due to the same custom-pipeline decode issue." + }, + { + "left": "issue:5335", + "right": "issue:7418", + "accept": false, + "reason": "Local deployment/network access question is unrelated to a bug in a community upscale pipeline." + }, + { + "left": "issue:9025", + "right": "issue:9031", + "accept": false, + "reason": "One is about from_single_file on TPU VM; the other is about disabling the safety checker." + }, + { + "left": "issue:4616", + "right": "issue:7636", + "accept": false, + "reason": "SDXL LoRA conversion script failure is not the same as how to use the SDXL tile ControlNet model." + }, + { + "left": "issue:11229", + "right": "issue:11342", + "accept": false, + "reason": "Both involve MPS, but they are different pipelines with different failure modes." + }, + { + "left": "issue:10639", + "right": "issue:12369", + "accept": false, + "reason": "Different feature requests for different models: Lumina-Image 2.0 support vs Wan 2.1 4K scaling." + }, + { + "left": "issue:11489", + "right": "issue:9729", + "accept": true, + "reason": "Both are the same Flux LoRA advanced training failure; 9729 states the concrete T5EncoderModel save-model error." + }, + { + "left": "issue:3159", + "right": "issue:6439", + "accept": false, + "reason": "Docs typo in a code example and a 404 link in another guide are unrelated documentation issues." + }, + { + "left": "issue:10937", + "right": "issue:11749", + "accept": false, + "reason": "torch.compile on vae.encode and FlowMatchEulerDiscreteScheduler floating-point timestep errors are different bugs." + }, + { + "left": "issue:11223", + "right": "issue:11283", + "accept": false, + "reason": "WAN layer-wise upcasting OOM with LoRAs and WAN header deserialization are separate failures." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the provided packet and stays appropriately conservative. The accepted soft-edge pairs are all plausibly same-bug clusters based on the titles, and the rejected pairs are sensible distinctions rather than overclaims. No mergeability issue since these are issue pairs, not PR pairs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 193, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10057", + "issue:10282", + "issue:10321", + "issue:1037", + "issue:10398", + "issue:10757", + "issue:11023", + "issue:11063", + "issue:11519", + "issue:1207", + "issue:1290", + "issue:1433", + "issue:1445", + "issue:1941", + "issue:2044", + "issue:2311", + "issue:2635", + "issue:2728", + "issue:3019", + "issue:3043", + "issue:3085", + "issue:3623", + "issue:3901", + "issue:4159", + "issue:4446", + "issue:4609", + "issue:4615", + "issue:4648", + "issue:539", + "issue:6142", + "issue:6162", + "issue:6287", + "issue:6368", + "issue:6506", + "issue:6725", + "issue:6839", + "issue:7177", + "issue:7219", + "issue:7243", + "issue:7441", + "issue:8373", + "issue:8410", + "issue:8613", + "issue:8863", + "issue:8885", + "issue:8975", + "issue:9551", + "issue:9731" + ], + "soft_pairs": [ + "issue:6287|issue:7441", + "issue:6725|issue:9731", + "issue:3085|issue:539", + "issue:1941|issue:2728", + "issue:3623|issue:4446", + "issue:7243|issue:9551", + "issue:6162|issue:8863", + "issue:4648|issue:7177", + "issue:6506|issue:6839", + "issue:11519|issue:8410", + "issue:1433|issue:1445", + "issue:3019|issue:4609", + "issue:11023|issue:8613", + "issue:3901|issue:4159", + "issue:10057|issue:8885", + "issue:10321|issue:10398", + "issue:2311|issue:2635", + "issue:1037|issue:1290", + "issue:3043|issue:7219", + "issue:1207|issue:2044", + "issue:11063|issue:8373", + "issue:4615|issue:6142", + "issue:10757|issue:8975", + "issue:10282|issue:6368" + ], + "prepared_review_unit_hash": "e869ef94fe3c996985d3c4df189116059808e808228051c7fc310bf625040f38", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33984, + "estimated_input_tokens": 8496, + "estimated_eval_tokens": 17248 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a false-positive mix of unrelated Diffusers bugs, feature requests, and environment/loading errors. None of the soft pairs look like the same underlying bug or change, so I reject all of them.", + "confidence": 0.92, + "canonical_issue_reason": "No true duplicate set emerges here; if a representative issue must be chosen, issue:7219 is the most substantive and active standalone request in the group, but it is not a duplicate of the others.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue:7219 has the highest discussion activity and is a broad, concrete feature request, so it is the best single anchor for this mixed cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6287", + "right": "issue:7441", + "accept": false, + "reason": "Dependency cleanup/setup.py and Python 3.12 pytest breakage are unrelated issues." + }, + { + "left": "issue:6725", + "right": "issue:9731", + "accept": false, + "reason": "Different requests: LDMPipeline image generation vs Playground2.5 LoRA training guidance." + }, + { + "left": "issue:3085", + "right": "issue:539", + "accept": false, + "reason": "Both are API/version friction, but they concern different functions and different failure modes." + }, + { + "left": "issue:1941", + "right": "issue:2728", + "accept": false, + "reason": "Custom pipeline loading for UnCLIP vs safetensors loading for UNet2DConditionModel are distinct code paths." + }, + { + "left": "issue:3623", + "right": "issue:4446", + "accept": false, + "reason": "General large-model loading problem vs custom text_encoder support in from_single_file; not the same bug." + }, + { + "left": "issue:7243", + "right": "issue:9551", + "accept": false, + "reason": "Separate model-support requests: ResAdapter vs Flux ControlNet usage." + }, + { + "left": "issue:6162", + "right": "issue:8863", + "accept": false, + "reason": "IP-Adapter SDXL shape mismatch and tuple.shape AttributeError are different runtime failures." + }, + { + "left": "issue:4648", + "right": "issue:7177", + "accept": false, + "reason": "Missing config.json in a directory is not the same as StableCascadeDecoderPipeline.from_pretrained failing." + }, + { + "left": "issue:6506", + "right": "issue:6839", + "accept": false, + "reason": "One is an output_type bug, the other is from_single_file loading support for Stable Video Diffusion." + }, + { + "left": "issue:11519", + "right": "issue:8410", + "accept": false, + "reason": "Both are feature requests for new models, but they target different pipelines/models entirely." + }, + { + "left": "issue:1433", + "right": "issue:1445", + "accept": false, + "reason": "Different ONNX-related AttributeErrors in different pipeline classes." + }, + { + "left": "issue:3019", + "right": "issue:4609", + "accept": false, + "reason": "Unrelated community model requests: text-guided editing with SAM vs t2iadapter_style." + }, + { + "left": "issue:11023", + "right": "issue:8613", + "accept": false, + "reason": "Playground v2.5 blur/custom pipeline issue and a caption_column OOM are unrelated." + }, + { + "left": "issue:3901", + "right": "issue:4159", + "accept": false, + "reason": "Missing requirements.txt and DiffEdit VRAM usage are completely different problems." + }, + { + "left": "issue:10057", + "right": "issue:8885", + "accept": false, + "reason": "FluxTransformer2DModel.from_single_file token loading and Custom Diffusion modifier-token installation are different issues." + }, + { + "left": "issue:10321", + "right": "issue:10398", + "accept": false, + "reason": "Missing model files in LTX-Video vs missing target modules in a base model are separate loading/training errors." + }, + { + "left": "issue:2311", + "right": "issue:2635", + "accept": false, + "reason": "UniPC scheduler request and scheduler-name mapping question are related only broadly, not the same issue." + }, + { + "left": "issue:1037", + "right": "issue:1290", + "accept": false, + "reason": "Different community pipeline proposals; no shared bug or concrete implementation target." + }, + { + "left": "issue:3043", + "right": "issue:7219", + "accept": false, + "reason": "Different feature requests for different models/pipelines (ODISE vs SUPIR)." + }, + { + "left": "issue:1207", + "right": "issue:2044", + "accept": false, + "reason": "CUDA runtime symbol failure and DistributedDataParallel dtype attribute error are unrelated runtime/environment issues." + }, + { + "left": "issue:11063", + "right": "issue:8373", + "accept": false, + "reason": "Incorrect padding in prepare_attention_mask and a generic incorrect error message are not the same bug." + }, + { + "left": "issue:4615", + "right": "issue:6142", + "accept": false, + "reason": "Import error for ControlNet pipeline and undefined start_training are unrelated." + }, + { + "left": "issue:10757", + "right": "issue:8975", + "accept": false, + "reason": "A variable-name fix in a converter script is unrelated to the LoraLoaderMixin import error." + }, + { + "left": "issue:10282", + "right": "issue:6368", + "accept": false, + "reason": "Shape mismatch while loading x_embedder weights and unsupported lora alpha checkpoint layers are different checkpoint-loading problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 194, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10111", + "issue:10226", + "issue:10675", + "issue:10796", + "issue:10994", + "issue:11023", + "issue:11396", + "issue:1160", + "issue:1166", + "issue:1207", + "issue:1290", + "issue:12995", + "issue:1539", + "issue:1544", + "issue:1617", + "issue:1677", + "issue:197", + "issue:200", + "issue:2221", + "issue:2321", + "issue:2560", + "issue:3043", + "issue:3706", + "issue:3891", + "issue:4436", + "issue:4609", + "issue:462", + "issue:4648", + "issue:5300", + "issue:541", + "issue:5881", + "issue:6141", + "issue:6234", + "issue:6278", + "issue:6374", + "issue:6518", + "issue:6688", + "issue:7142", + "issue:785", + "issue:8588", + "issue:8880", + "issue:8918", + "issue:9258", + "issue:9450", + "issue:9895", + "issue:9904", + "issue:9958" + ], + "soft_pairs": [ + "issue:9450|issue:9904", + "issue:1207|issue:1544", + "issue:1539|issue:2321", + "issue:1617|issue:2221", + "issue:10010|issue:4648", + "issue:1290|issue:541", + "issue:11023|issue:12995", + "issue:10111|issue:10675", + "issue:10796|issue:11396", + "issue:197|issue:200", + "issue:2560|issue:462", + "issue:3043|issue:3891", + "issue:3043|issue:6688", + "issue:6234|issue:8588", + "issue:6374|issue:9258", + "issue:8880|issue:9895", + "issue:1160|issue:1166", + "issue:3706|issue:4609", + "issue:6141|issue:9958", + "issue:10226|issue:10994", + "issue:1677|issue:785", + "issue:5881|issue:7142", + "issue:6278|issue:6518", + "issue:10010|issue:10675", + "issue:4436|issue:5300", + "issue:6234|issue:8918" + ], + "prepared_review_unit_hash": "697d230343d49a19bd74747e0d7691fe2fd04a48c6e665e7630b448c5e754244", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 34382, + "estimated_input_tokens": 8596, + "estimated_eval_tokens": 17448 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a noisy mix of unrelated issues, with a few Flux-related reports that are close in theme but not the same underlying bug. I would not merge any of the soft-edge pairs as duplicates.", + "confidence": 0.35, + "canonical_issue_reason": "issue:10111 is the broadest, most umbrella Flux bug report in the set, so it is the best single issue anchor for this noisy cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10111 best represents the dominant Flux-related portion of the cluster without overfitting to one narrow symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9450", + "right": "issue:9904", + "accept": false, + "reason": "Both are FluxPipeline bugs, but one is a multi-GPU device placement error and the other is silent shape rounding; different code paths and fixes." + }, + { + "left": "issue:1207", + "right": "issue:1544", + "accept": false, + "reason": "One is a CUDA runtime symbol loading failure, the other is a missing scheduler method; unrelated failures." + }, + { + "left": "issue:1539", + "right": "issue:2321", + "accept": false, + "reason": "Conversion script type mismatch vs inference EMA/model copying semantics; not the same bug." + }, + { + "left": "issue:1617", + "right": "issue:2221", + "accept": false, + "reason": "Argument parsing error and a ColossalAI runtime error are distinct problems." + }, + { + "left": "issue:10010", + "right": "issue:4648", + "accept": false, + "reason": "Both mention missing files, but they refer to different assets and setup failures; not one underlying defect." + }, + { + "left": "issue:1290", + "right": "issue:541", + "accept": false, + "reason": "Community project request versus SageMaker support question; no shared bug/change." + }, + { + "left": "issue:11023", + "right": "issue:12995", + "accept": false, + "reason": "Different pipelines and different questions: blurry output in a custom pipeline vs video generation/reconstruction behavior." + }, + { + "left": "issue:10111", + "right": "issue:10675", + "accept": false, + "reason": "Related to Flux, but one is a general bug report and the other is a scheduler max_shift configuration discrepancy." + }, + { + "left": "issue:10796", + "right": "issue:11396", + "accept": false, + "reason": "Different model families and different documentation/format questions." + }, + { + "left": "issue:197", + "right": "issue:200", + "accept": false, + "reason": "Dead code cleanup request versus a test task; not duplicates." + }, + { + "left": "issue:2560", + "right": "issue:462", + "accept": false, + "reason": "Different pipelines and symptoms: generator-list nondeterminism vs MPS release breakage." + }, + { + "left": "issue:3043", + "right": "issue:3891", + "accept": false, + "reason": "Separate feature requests for entirely different pipelines/models." + }, + { + "left": "issue:3043", + "right": "issue:6688", + "accept": false, + "reason": "Both are feature additions, but for unrelated models and code paths." + }, + { + "left": "issue:6234", + "right": "issue:8588", + "accept": false, + "reason": "Different training/conversion script problems in different model families." + }, + { + "left": "issue:6374", + "right": "issue:9258", + "accept": false, + "reason": "Chinese-language LoRA weight-loading question vs training speed investigation; unrelated." + }, + { + "left": "issue:8880", + "right": "issue:9895", + "accept": false, + "reason": "Performance regression vs device mismatch error; different symptoms and fixes." + }, + { + "left": "issue:1160", + "right": "issue:1166", + "accept": false, + "reason": "Both hang, but in different tests and contexts, so not the same underlying bug." + }, + { + "left": "issue:3706", + "right": "issue:4609", + "accept": false, + "reason": "Different model/pipeline addition requests." + }, + { + "left": "issue:6141", + "right": "issue:9958", + "accept": false, + "reason": "ControlNet parameter support and VAE temporal decoder tiling are unrelated feature requests." + }, + { + "left": "issue:10226", + "right": "issue:10994", + "accept": false, + "reason": "Both are Flux LoRA training issues, but one is a tensor-size mismatch and the other is a batch-size crash; not clearly the same bug." + }, + { + "left": "issue:1677", + "right": "issue:785", + "accept": false, + "reason": "Wrong data type invocation versus scheduler performance cost; unrelated." + }, + { + "left": "issue:5881", + "right": "issue:7142", + "accept": false, + "reason": "Both compare output quality, but they concern different pipelines/models and different mismatch causes." + }, + { + "left": "issue:6278", + "right": "issue:6518", + "accept": false, + "reason": "Same script family, but one is a multi-GPU runtime issue and the other is a NoneType arithmetic crash; different root causes." + }, + { + "left": "issue:10010", + "right": "issue:10675", + "accept": false, + "reason": "A missing remote model file error is not the same as a Flux scheduler configuration mismatch." + }, + { + "left": "issue:4436", + "right": "issue:5300", + "accept": false, + "reason": "Conversion workflow versus internet-connection behavior; unrelated topics." + }, + { + "left": "issue:6234", + "right": "issue:8918", + "accept": false, + "reason": "Different training scripts and different failure reports; not one duplicate bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 195, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10054", + "issue:10231", + "issue:10855", + "issue:10920", + "issue:11014", + "issue:11178", + "issue:11540", + "issue:12436", + "issue:12483", + "issue:12761", + "issue:1350", + "issue:1539", + "issue:1617", + "issue:1974", + "issue:1976", + "issue:2083", + "issue:2161", + "issue:2676", + "issue:2830", + "issue:2984", + "issue:3073", + "issue:3974", + "issue:3984", + "issue:4178", + "issue:4194", + "issue:4213", + "issue:4605", + "issue:4611", + "issue:4647", + "issue:5108", + "issue:5557", + "issue:5731", + "issue:6572", + "issue:6958", + "issue:7183", + "issue:7456", + "issue:7513", + "issue:7686", + "issue:8465", + "issue:8485", + "issue:8624", + "issue:8790", + "issue:8834", + "issue:9096", + "issue:9350", + "issue:9799", + "issue:9895" + ], + "soft_pairs": [ + "issue:8485|issue:8834", + "issue:1974|issue:2083", + "issue:10231|issue:8624", + "issue:11014|issue:7513", + "issue:9350|issue:9799", + "issue:4213|issue:5108", + "issue:1350|issue:3073", + "issue:12436|issue:12483", + "issue:2830|issue:4611", + "issue:5731|issue:7456", + "issue:6572|issue:8790", + "issue:2676|issue:3974", + "issue:4647|issue:6958", + "issue:1539|issue:1976", + "issue:2161|issue:4605", + "issue:10920|issue:9096", + "issue:10054|issue:7183", + "issue:10855|issue:12761", + "issue:2984|issue:3073", + "issue:11540|issue:9895", + "issue:1617|issue:5557", + "issue:11178|issue:8465", + "issue:3984|issue:7686", + "issue:4178|issue:4194" + ], + "prepared_review_unit_hash": "d124da7a5250eb9215e89584d0b73d1ef09b434b645c5fcebd3d34ba20ec2dae", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33772, + "estimated_input_tokens": 8443, + "estimated_eval_tokens": 17142 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is a broad grab-bag of unrelated Diffusers issues. None of the soft pairs is close enough to treat as the same underlying bug or change; most are only superficially similar by subsystem or symptom.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 12483 is the strongest standalone representative: it is a concrete, well-scoped runtime compatibility bug with a clear version interaction and actionable reproduction context.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 12483 is the best global issue candidate because it is specific, technically grounded, and more diagnostic than the more symptom-level reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8485", + "right": "issue:8834", + "accept": false, + "reason": "SD3 LoRA support request vs SD3 ControlNet training-code request; related domain but different feature asks." + }, + { + "left": "issue:1974", + "right": "issue:2083", + "accept": false, + "reason": "Both involve schedulers and generators, but one is about Euler schedulers with generator lists and the other is DDIM eta handling; distinct bugs." + }, + { + "left": "issue:10231", + "right": "issue:8624", + "accept": false, + "reason": "Ubuntu tutorial ImportError vs SD3 fused layer norm undefined-symbol error; different failure modes and code paths." + }, + { + "left": "issue:11014", + "right": "issue:7513", + "accept": false, + "reason": "Generic CUDA OOM complaint vs specific SDXL LoRA DreamBooth on 3060 memory issue; not the same defect." + }, + { + "left": "issue:9350", + "right": "issue:9799", + "accept": false, + "reason": "txt_ids deprecation warning vs tensor shape mismatch runtime error; unrelated symptoms and likely unrelated fixes." + }, + { + "left": "issue:4213", + "right": "issue:5108", + "accept": false, + "reason": "from_single_file loading issue without accelerate vs T2I Multiadapter XL exceptions; different pipelines and error classes." + }, + { + "left": "issue:1350", + "right": "issue:3073", + "accept": false, + "reason": "DiffusionDet project discussion vs consistency-models feature request; not a duplicate bug/change." + }, + { + "left": "issue:12436", + "right": "issue:12483", + "accept": false, + "reason": "CLIPTextModel unexpected kwarg is a downstream symptom, while 12483 is a broader transformers/diffusers version conflict; similar area but not same issue." + }, + { + "left": "issue:2830", + "right": "issue:4611", + "accept": false, + "reason": "Modelscope video usage question vs VideoComposer merge topic; different content and no shared bug." + }, + { + "left": "issue:5731", + "right": "issue:7456", + "accept": false, + "reason": "AutoencoderKL missing logvar scalar vs missing raise statement; unrelated implementation issues." + }, + { + "left": "issue:6572", + "right": "issue:8790", + "accept": false, + "reason": "ControlNetXS-SDXL-Inpaint pipeline request vs SDXL TensorRT/IP-Adapter request; different features, not one bug." + }, + { + "left": "issue:2676", + "right": "issue:3974", + "accept": false, + "reason": "Scheduler-name question vs safetensors support bug in a specific pipeline; no same underlying defect." + }, + { + "left": "issue:4647", + "right": "issue:6958", + "accept": false, + "reason": "UNet2DModel mid_block num_layers plumbing bug vs unet_motion_model use_linear_projection omission; different model classes and code paths." + }, + { + "left": "issue:1539", + "right": "issue:1976", + "accept": false, + "reason": "Conversion script type bug vs tokenizer config file path error; unrelated." + }, + { + "left": "issue:2161", + "right": "issue:4605", + "accept": false, + "reason": "LoRA training on inpainting model request vs long-prompt-weighting pipeline error; different pipelines and failure conditions." + }, + { + "left": "issue:10920", + "right": "issue:9096", + "accept": false, + "reason": "DDIMInverseScheduler inf values from alpha/sample setting vs Flux half-precision NaNs; both numeric instability but different mechanisms." + }, + { + "left": "issue:10054", + "right": "issue:7183", + "accept": false, + "reason": "EulerDiscreteScheduler linspace parameter bug vs scheduler class-name display bug; different scheduler internals." + }, + { + "left": "issue:10855", + "right": "issue:12761", + "accept": false, + "reason": "Gradient accumulation step-frequency question vs templated attention backward gradient-count mismatch; unrelated training vs autograd bug." + }, + { + "left": "issue:2984", + "right": "issue:3073", + "accept": false, + "reason": "Latent video diffusion model discussion vs consistency models request; both research-related but not the same change." + }, + { + "left": "issue:11540", + "right": "issue:9895", + "accept": false, + "reason": "VAE slicing causing noise / unsupported in Flux img2img vs Flux device-mismatch regression; both Flux but different defects." + }, + { + "left": "issue:1617", + "right": "issue:5557", + "accept": false, + "reason": "Missing required args in training script vs unexpected use_reentrant keyword; different CLI/API errors." + }, + { + "left": "issue:11178", + "right": "issue:8465", + "accept": false, + "reason": "from_single_file loading MistoLine ControlNet issue vs StableCascadeUNet.from_single_file config.json error; similar loader surface, different models and missing files." + }, + { + "left": "issue:3984", + "right": "issue:7686", + "accept": false, + "reason": "class_labels requirement in one config path vs cross_attention_dim requirement in UNet2DModel; both validation errors but not the same bug." + }, + { + "left": "issue:4178", + "right": "issue:4194", + "accept": false, + "reason": "LoRA unload with XFormers vs from_single_file VAE loading problem; different pipeline components and failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 196, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10016", + "issue:10210", + "issue:10511", + "issue:1060", + "issue:10883", + "issue:10886", + "issue:10887", + "issue:10987", + "issue:11410", + "issue:11448", + "issue:11700", + "issue:12367", + "issue:1524", + "issue:2052", + "issue:2437", + "issue:3043", + "issue:3112", + "issue:3140", + "issue:3467", + "issue:3537", + "issue:3699", + "issue:3963", + "issue:4407", + "issue:4418", + "issue:4648", + "issue:4665", + "issue:4702", + "issue:4830", + "issue:5022", + "issue:5118", + "issue:5320", + "issue:5341", + "issue:5877", + "issue:5968", + "issue:6560", + "issue:7003", + "issue:7191", + "issue:7418", + "issue:743", + "issue:8234", + "issue:8637", + "issue:8809", + "issue:8837", + "issue:9101", + "issue:9147", + "issue:9171", + "issue:9933", + "issue:9958" + ], + "soft_pairs": [ + "issue:3963|issue:4665", + "issue:5320|issue:5341", + "issue:11448|issue:8637", + "issue:1524|issue:5877", + "issue:3699|issue:4418", + "issue:3140|issue:7191", + "issue:3112|issue:5118", + "issue:10016|issue:7003", + "issue:10210|issue:8809", + "issue:4830|issue:8234", + "issue:5968|issue:9101", + "issue:7418|issue:8837", + "issue:6560|issue:9171", + "issue:1060|issue:743", + "issue:5022|issue:9147", + "issue:10511|issue:10883", + "issue:10886|issue:10987", + "issue:10887|issue:11700", + "issue:11410|issue:9933", + "issue:12367|issue:9958", + "issue:2052|issue:3467", + "issue:2437|issue:3043", + "issue:3537|issue:4702", + "issue:4407|issue:4648" + ], + "prepared_review_unit_hash": "7f18cfb823bbe1d0af7bd2c86b4a5f71b46756509c370aa84501121c7485c64f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34048, + "estimated_input_tokens": 8512, + "estimated_eval_tokens": 17280 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is not a true duplicate cluster: the soft pairs span unrelated bugs, features, and docs/examples across different pipelines and components. All soft edges should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": "issue:10016 is the clearest self-contained bug report, but it is only a representative anchor; the rest of the items are heterogeneous and not duplicates of one another.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10016 is the strongest standalone issue to surface as a representative of the set because it is concrete and actionable, even though it is not a duplicate hub.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3963", + "right": "issue:4665", + "accept": false, + "reason": "Different failure modes: auto-download on Windows vs safetensors/alternate bin-file handling." + }, + { + "left": "issue:5320", + "right": "issue:5341", + "accept": false, + "reason": "One is a PEFT API enhancement request; the other is updating example scripts, not the same bug." + }, + { + "left": "issue:11448", + "right": "issue:8637", + "accept": false, + "reason": "Completely different issues: missing model path/name vs import error for a pipeline class." + }, + { + "left": "issue:1524", + "right": "issue:5877", + "accept": false, + "reason": "Training quality problem vs subprocess/checkpoint-resume failure; different root causes and scripts." + }, + { + "left": "issue:3699", + "right": "issue:4418", + "accept": false, + "reason": "OSX/CUDA platform error vs SDXL inpaint denoising_start behavior; unrelated code paths." + }, + { + "left": "issue:3140", + "right": "issue:7191", + "accept": false, + "reason": "Importing a specific pipeline class is not the same as an unspecified import problem." + }, + { + "left": "issue:3112", + "right": "issue:5118", + "accept": false, + "reason": "Multi-ControlNet inpainting and reference_only ControlNet usage are different features." + }, + { + "left": "issue:10016", + "right": "issue:7003", + "accept": false, + "reason": "Both are shape mismatches, but in different pipelines and tensors (SD3.5M pos_embed vs inpaint conv weights)." + }, + { + "left": "issue:10210", + "right": "issue:8809", + "accept": false, + "reason": "Img2img output similarity is unrelated to feedforward output inconsistency with sequence length." + }, + { + "left": "issue:4830", + "right": "issue:8234", + "accept": false, + "reason": "Different tensor errors: MultiControlNet size mismatch vs ip_adapter_scale dict multiplication." + }, + { + "left": "issue:5968", + "right": "issue:9101", + "accept": false, + "reason": "ONNX conversion request for video pipelines vs a training script request for SD3; not the same bug/change." + }, + { + "left": "issue:7418", + "right": "issue:8837", + "accept": false, + "reason": "Community upscale script bug and SlicedAttnProcessor bug are separate code paths." + }, + { + "left": "issue:6560", + "right": "issue:9171", + "accept": false, + "reason": "LoRA shape-size regression and SD1.5 v_prediction loading regression are different loader issues." + }, + { + "left": "issue:1060", + "right": "issue:743", + "accept": false, + "reason": "Safety-checker loading failure in one converter vs ckpt-to-diffusers conversion failure in another." + }, + { + "left": "issue:5022", + "right": "issue:9147", + "accept": false, + "reason": "Import-time tokenizer error vs Flux autopipeline error; unrelated subsystems." + }, + { + "left": "issue:10511", + "right": "issue:10883", + "accept": false, + "reason": "Scheduler redesign for SD3/Flux is unrelated to a Marigold feature update." + }, + { + "left": "issue:10886", + "right": "issue:10987", + "accept": false, + "reason": "Both are diffusion-model feature ideas, but they target different domains and are not the same change." + }, + { + "left": "issue:10887", + "right": "issue:11700", + "accept": false, + "reason": "Missing StableDiffusionXLTrainer import and Pruna integration are unrelated." + }, + { + "left": "issue:11410", + "right": "issue:9933", + "accept": false, + "reason": "Attention processor unpack error vs missing width/height parameters in SD3 img2img; different bugs." + }, + { + "left": "issue:12367", + "right": "issue:9958", + "accept": false, + "reason": "Qwen-Image Lora support request and AutoencoderKLTemporalDecoder tiling support are unrelated features." + }, + { + "left": "issue:2052", + "right": "issue:3467", + "accept": false, + "reason": "Both involve resuming training, but one is textual inversion checkpoint resume and the other is ControlNet step accounting." + }, + { + "left": "issue:2437", + "right": "issue:3043", + "accept": false, + "reason": "Different pipeline additions: T2I-Adapter vs ODISE." + }, + { + "left": "issue:3537", + "right": "issue:4702", + "accept": false, + "reason": "ControlNet conversion script and VQ-VAE training example are unrelated requests." + }, + { + "left": "issue:4407", + "right": "issue:4648", + "accept": false, + "reason": "Local hub_download storage question vs missing config.json in a ControlNet directory; different file-loading problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 197, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10003", + "issue:10382", + "issue:1082", + "issue:10893", + "issue:11221", + "issue:11468", + "issue:11574", + "issue:12113", + "issue:1290", + "issue:1530", + "issue:1919", + "issue:2091", + "issue:2185", + "issue:2552", + "issue:2661", + "issue:2662", + "issue:3162", + "issue:3647", + "issue:3788", + "issue:4321", + "issue:4830", + "issue:4935", + "issue:4975", + "issue:5037", + "issue:5233", + "issue:5476", + "issue:556", + "issue:5737", + "issue:5791", + "issue:5897", + "issue:5989", + "issue:6016", + "issue:6218", + "issue:6234", + "issue:6287", + "issue:6666", + "issue:7043", + "issue:7350", + "issue:7405", + "issue:7460", + "issue:8647", + "issue:8831", + "issue:8970", + "issue:9171", + "issue:9181", + "issue:9370", + "issue:9479", + "issue:9894" + ], + "soft_pairs": [ + "issue:4935|issue:7350", + "issue:5476|issue:7043", + "issue:6234|issue:6287", + "issue:1082|issue:3647", + "issue:10003|issue:9479", + "issue:2185|issue:2552", + "issue:1530|issue:3162", + "issue:6218|issue:9181", + "issue:4830|issue:5233", + "issue:11574|issue:12113", + "issue:5737|issue:6016", + "issue:3788|issue:7405", + "issue:10382|issue:9370", + "issue:11221|issue:11468", + "issue:4975|issue:8831", + "issue:10893|issue:9894", + "issue:5989|issue:7460", + "issue:1919|issue:3647", + "issue:1290|issue:556", + "issue:4321|issue:8970", + "issue:8647|issue:9171", + "issue:5037|issue:6666", + "issue:2091|issue:2661", + "issue:2091|issue:2662", + "issue:5791|issue:5897" + ], + "prepared_review_unit_hash": "83a26a3b0ee1ca2fa0fc5d0628cf99eebe887d7ba2bd70e9e16ec4c1ce7c5bf5", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34417, + "estimated_input_tokens": 8605, + "estimated_eval_tokens": 17466 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated bug reports or feature requests. Only two soft pairs look like true duplicates: the learned_classifier_free_sampling_embeddings errors, and the checkpoint write-error reports.", + "confidence": 0.88, + "canonical_issue_reason": "Issue 3162 is the clearest canonical target for the learned_classifier_free_sampling_embeddings bug because it states the exact runtime error directly and matches the paired report closely.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9171 is the strongest standalone report overall: it is a regression, names the exact entrypoint and model condition, and is more actionable than the vaguer reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4935", + "right": "issue:7350", + "accept": false, + "reason": "Both are about LoRA/ControlNet, but one is a usage question and the other is a different initialization failure." + }, + { + "left": "issue:5476", + "right": "issue:7043", + "accept": false, + "reason": "Different topics: xformers loading compatibility vs DreamBooth custom captions." + }, + { + "left": "issue:6234", + "right": "issue:6287", + "accept": false, + "reason": "One is a training-script issue; the other is about packaging/dependencies. Not the same bug." + }, + { + "left": "issue:1082", + "right": "issue:3647", + "accept": false, + "reason": "Both mention DreamBooth, but the CUDA matmul crash and DeepFloyd training are different problems." + }, + { + "left": "issue:10003", + "right": "issue:9479", + "accept": false, + "reason": "Different pipelines and symptoms: scheduler bug at 1000 steps vs inpaint mask leakage." + }, + { + "left": "issue:2185", + "right": "issue:2552", + "accept": false, + "reason": "Unrelated training-config omission vs an inpaint tutorial question." + }, + { + "left": "issue:1530", + "right": "issue:3162", + "accept": true, + "reason": "Same underlying bug: missing/expected learned_classifier_free_sampling_embeddings causing the same runtime error." + }, + { + "left": "issue:6218", + "right": "issue:9181", + "accept": false, + "reason": "Both are ControlNet-related, but batched seeds and strange noise artifacts are different issues." + }, + { + "left": "issue:4830", + "right": "issue:5233", + "accept": false, + "reason": "Tensor-size mismatch in MultiControlNet is unrelated to callback step numbering." + }, + { + "left": "issue:11574", + "right": "issue:12113", + "accept": false, + "reason": "Different failures: a VAE state/key error vs a 3D convolution channel mismatch in WanImageToVideoPipeline." + }, + { + "left": "issue:5737", + "right": "issue:6016", + "accept": false, + "reason": "Both involve the converter script, but the SDXL conversion failure and the torch.Size() issue are distinct symptoms." + }, + { + "left": "issue:3788", + "right": "issue:7405", + "accept": false, + "reason": "Two separate model/pipeline requests: UniControl vs DynamiCrafter." + }, + { + "left": "issue:10382", + "right": "issue:9370", + "accept": false, + "reason": "Different code paths: VAE encoder return-type mismatch vs missing pooled_projection argument." + }, + { + "left": "issue:11221", + "right": "issue:11468", + "accept": false, + "reason": "Non-convergence in autoencoder training is unrelated to float8_e4m3_tensor quantization discussion." + }, + { + "left": "issue:4975", + "right": "issue:8831", + "accept": false, + "reason": "Both mention LoRA, but inefficiency in loading and unsupported-layer continuation are not the same bug." + }, + { + "left": "issue:10893", + "right": "issue:9894", + "accept": false, + "reason": "Separate feature requests for different components: CogVideoX ControlNet vs DC-AE integration." + }, + { + "left": "issue:5989", + "right": "issue:7460", + "accept": false, + "reason": "A safetensors bug is unrelated to an AdaLayerNorm bug." + }, + { + "left": "issue:1919", + "right": "issue:3647", + "accept": false, + "reason": "Dependency requirements for a DreamBooth example are unrelated to DeepFloyd DreamBooth training." + }, + { + "left": "issue:1290", + "right": "issue:556", + "accept": false, + "reason": "DreamArtist community project vs seamless texture tiling are different topics." + }, + { + "left": "issue:4321", + "right": "issue:8970", + "accept": false, + "reason": "Both are black-image reports, but they affect different backends/models and do not look like one concrete bug." + }, + { + "left": "issue:8647", + "right": "issue:9171", + "accept": false, + "reason": "Different problems: a NoneType iteration error in ControlNet Img2Img vs an SD 1.5 v_prediction regression." + }, + { + "left": "issue:5037", + "right": "issue:6666", + "accept": false, + "reason": "Validation-image argument handling is not the same as black validation images." + }, + { + "left": "issue:2091", + "right": "issue:2661", + "accept": false, + "reason": "Resuming multi-GPU training from checkpoint is a different issue from checkpoint save/write failures." + }, + { + "left": "issue:2091", + "right": "issue:2662", + "accept": false, + "reason": "Same as above: checkpoint resuming and checkpoint write errors are distinct." + }, + { + "left": "issue:5791", + "right": "issue:5897", + "accept": false, + "reason": "DreamBooth prior preservation/SNR gamma discussion is unrelated to the LoRA device-mismatch error." + }, + { + "left": "issue:2661", + "right": "issue:2662", + "accept": true, + "reason": "Same underlying checkpoint write-error bug; one is just a more specific report of the same failure." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded in the packet overall, and the soft-edge calls stay conservative. The two accepted pairs are plausible duplicates and look mergeable as single bugfixes; the rejected pairs are mostly supported by distinct titles/symptoms rather than overclaimed matches." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 198, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10772", + "issue:11010", + "issue:11093", + "issue:11182", + "issue:11298", + "issue:1140", + "issue:11555", + "issue:11561", + "issue:11653", + "issue:11749", + "issue:12359", + "issue:12695", + "issue:12754", + "issue:1290", + "issue:13279", + "issue:13350", + "issue:13416", + "issue:1434", + "issue:1962", + "issue:2891", + "issue:3085", + "issue:3431", + "issue:3455", + "issue:3891", + "issue:4149", + "issue:4363", + "issue:4436", + "issue:4529", + "issue:4782", + "issue:4949", + "issue:5485", + "issue:5607", + "issue:5889", + "issue:6132", + "issue:6688", + "issue:7080", + "issue:7219", + "issue:7390", + "issue:7670", + "issue:8850", + "issue:8900", + "issue:9090", + "issue:9164", + "issue:9227", + "issue:9464", + "issue:9744" + ], + "soft_pairs": [ + "issue:3085|issue:4529", + "issue:11555|issue:12754", + "issue:9227|issue:9464", + "issue:11093|issue:8850", + "issue:7390|issue:8900", + "issue:4363|issue:4436", + "issue:5889|issue:9744", + "issue:3891|issue:7219", + "issue:1434|issue:2891", + "issue:11298|issue:11653", + "issue:1140|issue:1290", + "issue:1962|issue:4949", + "issue:11010|issue:12359", + "issue:11749|issue:9090", + "issue:9164|issue:9464", + "issue:5485|issue:7080", + "issue:10100|issue:11561", + "issue:10772|issue:12695", + "issue:6132|issue:6688", + "issue:13350|issue:13416", + "issue:3455|issue:7670", + "issue:4149|issue:5607", + "issue:11182|issue:13279", + "issue:3891|issue:6688", + "issue:3431|issue:4782" + ], + "prepared_review_unit_hash": "d53ab0228cfe0533529eafe2c0103c5af9760dd78a95dc9a5e8dececc0621451", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33078, + "estimated_input_tokens": 8270, + "estimated_eval_tokens": 16796 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a false-positive similarity cluster: the items share common Diffusers/model-support vocabulary, but they describe different models, pipelines, schedulers, or bug classes rather than the same underlying issue. No pull requests are present.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 11010 is the most substantive and broadly scoped item in the set, with the richest discussion and a concrete feature-support framing, so it is the best anchor if one issue must represent this heterogeneous cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "11010 is the clearest representative of the cluster\u2019s general \u201cadd/support a new model/pipeline\u201d theme, while most others are narrower one-off bugs or unrelated support requests.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3085", + "right": "issue:4529", + "accept": false, + "reason": "Both involve loading/API changes, but one is about a removed cache-folder argument and the other is a LoRA `network_alpha` keyword error; different code paths." + }, + { + "left": "issue:11555", + "right": "issue:12754", + "accept": false, + "reason": "Both touch pipeline loading behavior, but `device_map=\"auto\"` support and `from_pipe` dtype conversion are separate problems." + }, + { + "left": "issue:9227", + "right": "issue:9464", + "accept": false, + "reason": "Both mention FLUX DreamBooth LoRA training, but one is a convergence/example issue and the other is an `IndexError` with Prodigy/text-encoder training." + }, + { + "left": "issue:11093", + "right": "issue:8850", + "accept": false, + "reason": "Both say something is broken, but they refer to different artifacts: TorchAO example vs community pipelines mirror." + }, + { + "left": "issue:7390", + "right": "issue:8900", + "accept": false, + "reason": "Both are `from_single_file` loading errors, but they target different SDXL pipelines/models and likely different failure modes." + }, + { + "left": "issue:4363", + "right": "issue:4436", + "accept": false, + "reason": "Both are about SDXL safetensors loading/conversion, but one is a usage question and the other is a conversion workflow issue." + }, + { + "left": "issue:5889", + "right": "issue:9744", + "accept": false, + "reason": "Both are video-model related, but Stable Video Diffusion support and Mochi support are distinct model additions." + }, + { + "left": "issue:3891", + "right": "issue:7219", + "accept": false, + "reason": "Both are requests to add a pipeline, but Tortoise TTS and SUPIR are unrelated models and code paths." + }, + { + "left": "issue:1434", + "right": "issue:2891", + "accept": false, + "reason": "Same pipeline family, but one is an OOM problem and the other is a docs/API mismatch about embeds." + }, + { + "left": "issue:11298", + "right": "issue:11653", + "accept": false, + "reason": "Both involve LoRA metadata/key handling, but `peft` key errors during hotswapping and `.alpha` keys in HiDream LoRAs are different bugs." + }, + { + "left": "issue:1140", + "right": "issue:1290", + "accept": false, + "reason": "DreamArtist may be related to hypernetworks, but these are not clearly the same request or bug; one is generic hypernetwork support and the other is a specific community model proposal." + }, + { + "left": "issue:1962", + "right": "issue:4949", + "accept": false, + "reason": "Both concern video generation, but stable-diffusion video generation and ProPainter in/outpainting are separate features." + }, + { + "left": "issue:11010", + "right": "issue:12359", + "accept": false, + "reason": "Both mention Chroma, but one is a feature/support request and the other is a documentation bug about `guidance_scale`." + }, + { + "left": "issue:11749", + "right": "issue:9090", + "accept": false, + "reason": "Different schedulers and different timestep bugs; these are not the same underlying defect." + }, + { + "left": "issue:9164", + "right": "issue:9464", + "accept": false, + "reason": "Both are FLUX training issues, but one is non-convergence in an example and the other is a runtime `IndexError`." + }, + { + "left": "issue:5485", + "right": "issue:7080", + "accept": false, + "reason": "Both mention ControlNet/batch workflows, but one is slow batch inference and the other is a wandb output divergence issue." + }, + { + "left": "issue:10100", + "right": "issue:11561", + "accept": false, + "reason": "Different feature requests for different pipelines: ConsisID support vs loading IP-Adapter in FluxFill." + }, + { + "left": "issue:10772", + "right": "issue:12695", + "accept": false, + "reason": "Different model support requests: Sana ControlNet vs Hunyuan v1.5." + }, + { + "left": "issue:6132", + "right": "issue:6688", + "accept": false, + "reason": "One is about PyTorch loadable classes, the other is a request for MotionCntrl support; no shared underlying bug/change." + }, + { + "left": "issue:13350", + "right": "issue:13416", + "accept": false, + "reason": "Both are usage questions for new models, but they concern different models and different prompt/batch behavior." + }, + { + "left": "issue:3455", + "right": "issue:7670", + "accept": false, + "reason": "Unrelated topics: instruct-pix2pix training vs how to use IDDPM." + }, + { + "left": "issue:4149", + "right": "issue:5607", + "accept": false, + "reason": "Both are tensor mismatch errors, but the affected pipelines and failure contexts are different." + }, + { + "left": "issue:11182", + "right": "issue:13279", + "accept": false, + "reason": "Different loading issues: unnecessary BF16 download vs an `AutoencoderRAE`/older-transformers compatibility error." + }, + { + "left": "issue:3891", + "right": "issue:6688", + "accept": false, + "reason": "Different community feature requests for unrelated models: Tortoise TTS vs MotionCntrl." + }, + { + "left": "issue:3431", + "right": "issue:4782", + "accept": false, + "reason": "Both are inpainting-related, but poor ControlNet results and mask-overwrite behavior are distinct bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 199, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10514", + "issue:10639", + "issue:10817", + "issue:1156", + "issue:11973", + "issue:12358", + "issue:12574", + "issue:12666", + "issue:12752", + "issue:13062", + "issue:13429", + "issue:1790", + "issue:1934", + "issue:2299", + "issue:2324", + "issue:2361", + "issue:2737", + "issue:2755", + "issue:2891", + "issue:3570", + "issue:4159", + "issue:4236", + "issue:4876", + "issue:4913", + "issue:5034", + "issue:5118", + "issue:5168", + "issue:5197", + "issue:5258", + "issue:5463", + "issue:5576", + "issue:5656", + "issue:5981", + "issue:6117", + "issue:6132", + "issue:6357", + "issue:6493", + "issue:713", + "issue:7168", + "issue:7312", + "issue:8511", + "issue:8526", + "issue:8593", + "issue:8790", + "issue:9012", + "issue:9161", + "issue:9460" + ], + "soft_pairs": [ + "issue:5576|issue:6493", + "issue:1934|issue:4159", + "issue:10639|issue:12752", + "issue:2737|issue:7312", + "issue:2755|issue:3570", + "issue:4876|issue:5656", + "issue:5118|issue:6132", + "issue:10817|issue:8511", + "issue:10514|issue:12574", + "issue:2891|issue:7168", + "issue:11973|issue:9161", + "issue:2299|issue:5034", + "issue:2361|issue:5197", + "issue:4913|issue:8790", + "issue:5981|issue:6117", + "issue:1156|issue:713", + "issue:5463|issue:6357", + "issue:5118|issue:5168", + "issue:4236|issue:5258", + "issue:12358|issue:13429", + "issue:1790|issue:2324", + "issue:8526|issue:8593", + "issue:9012|issue:9460", + "issue:12666|issue:13062" + ], + "prepared_review_unit_hash": "13093cc1bbefd21464212334889e8ca69b93df4918557b9afc124edc8ebaf740", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32744, + "estimated_input_tokens": 8186, + "estimated_eval_tokens": 16628 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft pair looks like the same concrete bug/change; they mostly share broad keywords or subsystem names but target different pipelines, schedulers, training scripts, or model integrations. I\u2019d use issue:12574 as the best canonical issue because it is a specific, well-scoped bug report with a clear failure mode.", + "confidence": 0.84, + "canonical_issue_reason": "issue:12574 is the strongest canonical issue candidate: it describes a concrete runtime failure, names the affected pipeline and underlying VACE control-argument handling, and is more actionable than the broader feature/request items.", + "canonical_pr_reason": "No PRs are present in the set.", + "best_issue_reason": "issue:12574 is the best overall issue representative because it is specific, technically grounded, and easy to map to a concrete fix path.", + "best_pr_reason": "No PRs are present, so there is no best PR candidate.", + "soft_edge_verdicts": [ + { + "left": "issue:5576", + "right": "issue:6493", + "accept": false, + "reason": "Both mention FreeNoise, but one is a scheduler request and the other is a pipeline entry; they are related features, not the same underlying change." + }, + { + "left": "issue:1934", + "right": "issue:4159", + "accept": false, + "reason": "Both are VRAM-related complaints, but they concern different pipelines and memory footprints, so they are not the same bug." + }, + { + "left": "issue:10639", + "right": "issue:12752", + "accept": false, + "reason": "These are separate model-support requests for different pipelines; same broad theme, not one concrete issue." + }, + { + "left": "issue:2737", + "right": "issue:7312", + "accept": false, + "reason": "Different functionality: scheduler CLIP guidance support versus AlignProp/direct reward finetuning." + }, + { + "left": "issue:2755", + "right": "issue:3570", + "accept": false, + "reason": "Both involve checkpoint loading, but one is float16 support and the other is disabling downloads; different fixes." + }, + { + "left": "issue:4876", + "right": "issue:5656", + "accept": false, + "reason": "LoRA layer std as an argument is unrelated to adding a LoRA loader." + }, + { + "left": "issue:5118", + "right": "issue:6132", + "accept": false, + "reason": "ControlNet reference_only usage and loadable classes are unrelated code paths." + }, + { + "left": "issue:10817", + "right": "issue:8511", + "accept": false, + "reason": "Both are SD3-related, but one is auto_pipeline missing control nets and the other is Dreambooth/LoRA encoder unloading." + }, + { + "left": "issue:10514", + "right": "issue:12574", + "accept": false, + "reason": "Different models and failure modes: Sana resolution binning/sample size versus WanVideoToVideo control-argument handling." + }, + { + "left": "issue:2891", + "right": "issue:7168", + "accept": false, + "reason": "One is about pipeline embed support; the other is about IP-Adapter image embeddings tensor shape." + }, + { + "left": "issue:11973", + "right": "issue:9161", + "accept": false, + "reason": "Separate training-script bugs in different Flux scripts; not the same failure." + }, + { + "left": "issue:2299", + "right": "issue:5034", + "accept": false, + "reason": "Fast-stable-diffusion checkpoint support and progress-bar display are unrelated." + }, + { + "left": "issue:2361", + "right": "issue:5197", + "accept": false, + "reason": "Checkpoint-saving crashes and multi-GPU slowdown are different training issues." + }, + { + "left": "issue:4913", + "right": "issue:8790", + "accept": false, + "reason": "OpenPose ControlNet support and SDXL TensorRT/IP-Adapter integration are different features." + }, + { + "left": "issue:5981", + "right": "issue:6117", + "accept": false, + "reason": "Two distinct scheduler bugs affecting different algorithms and methods." + }, + { + "left": "issue:1156", + "right": "issue:713", + "accept": false, + "reason": "Both are vague training problems, but there is no evidence they describe the same root cause." + }, + { + "left": "issue:5463", + "right": "issue:6357", + "accept": false, + "reason": "An UnboundLocalError in one path is not the same as mismatched API vs local results." + }, + { + "left": "issue:5118", + "right": "issue:5168", + "accept": false, + "reason": "Reference_only usage and ControlNet-XS support are separate ControlNet topics." + }, + { + "left": "issue:4236", + "right": "issue:5258", + "accept": false, + "reason": "SDXL refiner text-encoder registration and ControlNet inpaint with refiner are different pipeline integration problems." + }, + { + "left": "issue:12358", + "right": "issue:13429", + "accept": false, + "reason": "Both request new pipeline support, but for different models and different changes." + }, + { + "left": "issue:1790", + "right": "issue:2324", + "accept": false, + "reason": "A general training question is not the same as Dreambooth not reading instance data." + }, + { + "left": "issue:8526", + "right": "issue:8593", + "accept": false, + "reason": "Different SD3 issues on different platforms: MPS scale handling versus CPU bfloat16 noise." + }, + { + "left": "issue:9012", + "right": "issue:9460", + "accept": false, + "reason": "Loading SD3ControlNet weights and assigning multiple controlnets to the pipeline are separate problems." + }, + { + "left": "issue:12666", + "right": "issue:13062", + "accept": false, + "reason": "Kandinsky video pipeline availability is unrelated to onnxruntime-windowsml support policy." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 200, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10147", + "issue:10231", + "issue:10848", + "issue:10952", + "issue:1099", + "issue:11223", + "issue:11435", + "issue:11721", + "issue:12078", + "issue:12221", + "issue:12367", + "issue:12458", + "issue:12738", + "issue:12995", + "issue:1305", + "issue:13079", + "issue:2213", + "issue:2294", + "issue:2310", + "issue:2319", + "issue:2646", + "issue:3094", + "issue:3113", + "issue:319", + "issue:321", + "issue:3350", + "issue:4042", + "issue:4232", + "issue:4524", + "issue:4619", + "issue:4815", + "issue:4846", + "issue:4907", + "issue:5545", + "issue:590", + "issue:6095", + "issue:6688", + "issue:6866", + "issue:7138", + "issue:8410", + "issue:8604", + "issue:8832", + "issue:8939", + "issue:8989", + "issue:9511", + "issue:9850", + "issue:9926" + ], + "soft_pairs": [ + "issue:4524|issue:8410", + "issue:2319|issue:590", + "issue:2646|issue:6095", + "issue:1305|issue:3113", + "issue:6866|issue:9511", + "issue:4619|issue:7138", + "issue:10147|issue:12221", + "issue:11223|issue:13079", + "issue:3350|issue:5545", + "issue:8604|issue:8989", + "issue:10100|issue:6688", + "issue:10231|issue:11435", + "issue:10848|issue:10952", + "issue:4232|issue:4846", + "issue:8939|issue:9850", + "issue:11721|issue:8832", + "issue:1099|issue:2294", + "issue:4815|issue:4907", + "issue:12078|issue:12995", + "issue:12367|issue:9926", + "issue:12458|issue:12738", + "issue:2213|issue:3094", + "issue:2310|issue:4042", + "issue:319|issue:321" + ], + "prepared_review_unit_hash": "bb4200927a2fcf77743490fa6c99da774e1f9235caa3746edf739d6598b7e969", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33793, + "estimated_input_tokens": 8449, + "estimated_eval_tokens": 17154 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a loose collection of unrelated feature requests and bug reports; the soft pairs are only superficially similar by title, not true duplicates. I would not merge any of the candidate pairs.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue #8410 is the broadest umbrella request (native Diffusers pipeline support), but the items are too heterogeneous for any issue to serve as a true canonical duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4524", + "right": "issue:8410", + "accept": false, + "reason": "Both are feature requests about adding pipeline support, but for different pipelines (AnimateDiff vs TextDiffuser); not the same underlying change." + }, + { + "left": "issue:2319", + "right": "issue:590", + "accept": false, + "reason": "Both mention typos, but they are in different functions and different code paths, so they are not duplicates." + }, + { + "left": "issue:2646", + "right": "issue:6095", + "accept": false, + "reason": "Both are IndexError reports, but one is tiled VAE and the other is SVD scheduling; distinct bugs." + }, + { + "left": "issue:1305", + "right": "issue:3113", + "accept": false, + "reason": "Both are image-editing/control-related papers, but they describe different methods and desired features." + }, + { + "left": "issue:6866", + "right": "issue:9511", + "accept": false, + "reason": "Same pipeline family, but one requests IP-Adapter support and the other batching for multi-controlnet; different problems." + }, + { + "left": "issue:4619", + "right": "issue:7138", + "accept": false, + "reason": "SDXL-related, but one is a training dtype mismatch and the other is a model structure mismatch; not the same bug." + }, + { + "left": "issue:10147", + "right": "issue:12221", + "accept": false, + "reason": "Different requested integrations/features: MV-Adapter multi-view generation vs DiffSynth ControlNet support." + }, + { + "left": "issue:11223", + "right": "issue:13079", + "accept": false, + "reason": "Both are OOM reports, but on different models and scenarios; no shared concrete code-path is evident." + }, + { + "left": "issue:3350", + "right": "issue:5545", + "accept": false, + "reason": "Different pipelines/models and different failure modes; too broad to be duplicates." + }, + { + "left": "issue:8604", + "right": "issue:8989", + "accept": false, + "reason": "Different bugs in different systems: T5 fp16 corruption vs Stable Audio offloading behavior." + }, + { + "left": "issue:10100", + "right": "issue:6688", + "accept": false, + "reason": "Both request support for different model families (ConsisID vs MotionCntrl); not the same change." + }, + { + "left": "issue:10231", + "right": "issue:11435", + "accept": false, + "reason": "Both are import-related, but they concern different missing symbols and likely different dependency issues." + }, + { + "left": "issue:10848", + "right": "issue:10952", + "accept": false, + "reason": "Both involve adapters, but one is a bad function signature and the other is adapter-name handling during LoRA load." + }, + { + "left": "issue:4232", + "right": "issue:4846", + "accept": false, + "reason": "One is instruct-pix2pix/controlnet training/inference; the other is a DreamBooth training result issue." + }, + { + "left": "issue:8939", + "right": "issue:9850", + "accept": false, + "reason": "Completely different topics: tokenizer loading for SD3 vs gradient checkpointing support." + }, + { + "left": "issue:11721", + "right": "issue:8832", + "accept": false, + "reason": "Both are crashes, but on different pipelines and likely different root causes." + }, + { + "left": "issue:1099", + "right": "issue:2294", + "accept": false, + "reason": "DreamBooth class-name issue vs missing StableDiffusionVideoInpaintPipeline import; unrelated." + }, + { + "left": "issue:4815", + "right": "issue:4907", + "accept": false, + "reason": "Latents-only input support vs an SDXL inpainting warning; not the same underlying bug or feature." + }, + { + "left": "issue:12078", + "right": "issue:12995", + "accept": false, + "reason": "Different products and scenarios: Flux finetuning validation input vs Cosmos-Predict2.5 video behavior." + }, + { + "left": "issue:12367", + "right": "issue:9926", + "accept": false, + "reason": "One asks for a specific LoRA feature, the other for gguf quantization/speed support; unrelated." + }, + { + "left": "issue:12458", + "right": "issue:12738", + "accept": false, + "reason": "Both mention batching, but they are different pipelines/models and different failure contexts." + }, + { + "left": "issue:2213", + "right": "issue:3094", + "accept": false, + "reason": "Same training script name, but one is TPU distributed hang and the other is CUDA OOM; distinct issues." + }, + { + "left": "issue:2310", + "right": "issue:4042", + "accept": false, + "reason": "Different compositional inpainting requests; no shared concrete bug or code path." + }, + { + "left": "issue:319", + "right": "issue:321", + "accept": false, + "reason": "Separate research-model support requests (VQ-Diffusion vs Blended Diffusion), not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 201, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10191", + "issue:10550", + "issue:10817", + "issue:10928", + "issue:11275", + "issue:11279", + "issue:11412", + "issue:11448", + "issue:11555", + "issue:11709", + "issue:12567", + "issue:13043", + "issue:13176", + "issue:1358", + "issue:1633", + "issue:1896", + "issue:1934", + "issue:2310", + "issue:2437", + "issue:2462", + "issue:2471", + "issue:2560", + "issue:3140", + "issue:3209", + "issue:3289", + "issue:3467", + "issue:4061", + "issue:4073", + "issue:4241", + "issue:4269", + "issue:4286", + "issue:4580", + "issue:4619", + "issue:4750", + "issue:4839", + "issue:5002", + "issue:5314", + "issue:5573", + "issue:5594", + "issue:5656", + "issue:5927", + "issue:6688", + "issue:7789", + "issue:7842", + "issue:8649", + "issue:9026", + "issue:9138", + "issue:9637" + ], + "soft_pairs": [ + "issue:4241|issue:5656", + "issue:11412|issue:11448", + "issue:11275|issue:9026", + "issue:3467|issue:5573", + "issue:4580|issue:5002", + "issue:13043|issue:13176", + "issue:7789|issue:7842", + "issue:10928|issue:13176", + "issue:11709|issue:12567", + "issue:1633|issue:5594", + "issue:2437|issue:6688", + "issue:10550|issue:11279", + "issue:2560|issue:3209", + "issue:1358|issue:2462", + "issue:4619|issue:4839", + "issue:3289|issue:4073", + "issue:4061|issue:4286", + "issue:3140|issue:4269", + "issue:10817|issue:8649", + "issue:10191|issue:9637", + "issue:2310|issue:5314", + "issue:4750|issue:5927", + "issue:11555|issue:9138", + "issue:1934|issue:3209", + "issue:1896|issue:2471" + ], + "prepared_review_unit_hash": "955edf8cf0c54dc2a0310e293e081a07b9d95d253039ad0cd6c5a401938bd314", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34020, + "estimated_input_tokens": 8505, + "estimated_eval_tokens": 17266 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issues across different features/subsystems. The only plausible duplicate pair is the SDXL LoRA-loading reports (#4061 and #4286); the rest are too different in concrete error, API, or code path to merge as duplicates.", + "confidence": 0.76, + "canonical_issue_reason": "#4286 is the clearest concrete report for the SDXL LoRA-loading failure (`load_lora_weights` not working in SDXL), while #4061 is a shorter, more generic symptom report.", + "canonical_pr_reason": null, + "best_issue_reason": "#4286 is the best representative issue because it names the specific failing API and model family, making it the most actionable canonical duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4241", + "right": "issue:5656", + "accept": false, + "reason": "Both mention LoRA, but one is about fine-tuning unidiffusers and the other is about adding a LoRA loader to community pipelines; different change requests." + }, + { + "left": "issue:11412", + "right": "issue:11448", + "accept": false, + "reason": "Both are model-not-found/404 reports, but for different repositories/models and different loader paths." + }, + { + "left": "issue:11275", + "right": "issue:9026", + "accept": false, + "reason": "A VAE addition bug and an unexpected keyword argument in checkpoint dispatch are unrelated code paths." + }, + { + "left": "issue:3467", + "right": "issue:5573", + "accept": false, + "reason": "Both are ControlNet training issues, but one is resume-step accounting and the other is initialization timing; not the same bug." + }, + { + "left": "issue:4580", + "right": "issue:5002", + "accept": false, + "reason": "Both involve Karras sigmas, but one is a UniPC schedule exception and the other is missing scheduler arguments in different classes." + }, + { + "left": "issue:13043", + "right": "issue:13176", + "accept": false, + "reason": "Different attention failures with different backends and failure modes; not the same underlying bug." + }, + { + "left": "issue:7789", + "right": "issue:7842", + "accept": false, + "reason": "One is an MPS float32 requirement, the other is an image-size tensor mismatch in DeepFloyd stage 2." + }, + { + "left": "issue:10928", + "right": "issue:13176", + "accept": false, + "reason": "A missing checkpoint file error and an attention runtime error are unrelated." + }, + { + "left": "issue:11709", + "right": "issue:12567", + "accept": false, + "reason": "LoRA key-loading failure versus Ulysses attention noise generation; different components and symptoms." + }, + { + "left": "issue:1633", + "right": "issue:5594", + "accept": false, + "reason": "A scheduler option request and SDXL LoRA memory-saving request are different features, not one bug." + }, + { + "left": "issue:2437", + "right": "issue:6688", + "accept": false, + "reason": "Two different pipeline feature requests with no shared concrete implementation bug." + }, + { + "left": "issue:10550", + "right": "issue:11279", + "accept": false, + "reason": "Quantized LoRA loading versus initializing new layers in partially pretrained models are different loading/training scenarios." + }, + { + "left": "issue:2560", + "right": "issue:3209", + "accept": false, + "reason": "Different failures: generator-list nondeterminism versus AutoencoderKL NaNs on large images." + }, + { + "left": "issue:1358", + "right": "issue:2462", + "accept": false, + "reason": "Safetensors support and IterableDataset finetuning support are unrelated enhancement requests." + }, + { + "left": "issue:4619", + "right": "issue:4839", + "accept": false, + "reason": "Training dtype mismatch and lora_scale type errors are distinct bugs in different parts of the LoRA path." + }, + { + "left": "issue:3289", + "right": "issue:4073", + "accept": false, + "reason": "DeepFloyd super-resolution resizing bug versus a safety-checker attribute error; different code paths." + }, + { + "left": "issue:4061", + "right": "issue:4286", + "accept": true, + "reason": "Both report SDXL LoRA loading failing, centered on the same underlying `load_lora_weights`/LoRA-in-SDXL path." + }, + { + "left": "issue:10817", + "right": "issue:8649", + "accept": false, + "reason": "Missing SD3 control nets and `num_images_per_prompt` behavior are separate SD3 pipeline issues." + }, + { + "left": "issue:10191", + "right": "issue:9637", + "accept": false, + "reason": "GPU device selection in training is unrelated to the xFormers attention-mask bfloat16 kernel bug." + }, + { + "left": "issue:2310", + "right": "issue:5314", + "accept": false, + "reason": "Both mention inpainting, but one is LDM/LAMA cleaner and the other is a ControlNet reference inpainting pipeline request." + }, + { + "left": "issue:4750", + "right": "issue:5927", + "accept": false, + "reason": "Both reference inpaint/lama concepts, but one is a preprocessor request and the other is a broader combined-feature request." + }, + { + "left": "issue:11555", + "right": "issue:9138", + "accept": false, + "reason": "`device_map=\"auto\"` support and an int-vs-str TypeError are not the same issue." + }, + { + "left": "issue:1934", + "right": "issue:3209", + "accept": false, + "reason": "CPU offload memory usage and AutoencoderKL NaNs are unrelated runtime problems." + }, + { + "left": "issue:1896", + "right": "issue:2471", + "accept": false, + "reason": "Different pipeline size-handling failures with different repros and code paths." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The only duplicate-like pair supported by the packet is indeed #4061/#4286 (both SDXL LoRA loading failures), and the rest are reasonably rejected as different bugs, features, or code paths. No evidence of overclaiming." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 202, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10111", + "issue:10636", + "issue:1106", + "issue:11222", + "issue:11283", + "issue:11644", + "issue:12195", + "issue:1543", + "issue:2185", + "issue:2310", + "issue:2437", + "issue:3073", + "issue:3221", + "issue:3576", + "issue:3643", + "issue:3975", + "issue:4210", + "issue:4445", + "issue:460", + "issue:4646", + "issue:4750", + "issue:5225", + "issue:5395", + "issue:5480", + "issue:5656", + "issue:5777", + "issue:5989", + "issue:6110", + "issue:6243", + "issue:6359", + "issue:7049", + "issue:746", + "issue:7695", + "issue:8320", + "issue:8432", + "issue:8493", + "issue:852", + "issue:8604", + "issue:8742", + "issue:8752", + "issue:8837", + "issue:911", + "issue:9297", + "issue:9358", + "issue:940", + "issue:9464", + "issue:9822", + "issue:9930" + ], + "soft_pairs": [ + "issue:11222|issue:11283", + "issue:852|issue:940", + "issue:1106|issue:3221", + "issue:2437|issue:6243", + "issue:460|issue:746", + "issue:5395|issue:9297", + "issue:3975|issue:5989", + "issue:4210|issue:5225", + "issue:8432|issue:8742", + "issue:7049|issue:7695", + "issue:460|issue:4646", + "issue:9358|issue:9464", + "issue:11644|issue:8493", + "issue:10111|issue:9822", + "issue:12195|issue:8320", + "issue:2185|issue:911", + "issue:3576|issue:3643", + "issue:10636|issue:8837", + "issue:8604|issue:9930", + "issue:6110|issue:8752", + "issue:2310|issue:4750", + "issue:3073|issue:5480", + "issue:4445|issue:5656", + "issue:5777|issue:6359", + "issue:1543|issue:746" + ], + "prepared_review_unit_hash": "5f83ef21c52ce1f20e695337ac935a7e7609e8a267ba6ec91b9ac529510f54a6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33790, + "estimated_input_tokens": 8448, + "estimated_eval_tokens": 17152 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated one-off bug reports or feature requests. The only plausible duplicate-like pair is the LaMa/inpaint preprocessing issues; everything else is too semantically different to merge.", + "confidence": 0.78, + "canonical_issue_reason": "issue:4750 is the clearest representative of the LaMa/inpaint preprocessing topic; issue:2310 appears to be the older, vaguer counterpart.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4750 is the strongest issue to anchor this cluster because it is the most specific, concrete report among the few near-matches.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11222", + "right": "issue:11283", + "accept": false, + "reason": "Feature request for more WAN LoRAs vs a deserialization error in WAN loading; different problems." + }, + { + "left": "issue:852", + "right": "issue:940", + "accept": false, + "reason": "Both mention MPS crashes, but the scheduler and failure mode are different enough to treat as separate bugs." + }, + { + "left": "issue:1106", + "right": "issue:3221", + "accept": false, + "reason": "PNDM repeatedly invoking UNet is unrelated to load_attn_procs failing in diffusers 0.16.0." + }, + { + "left": "issue:2437", + "right": "issue:6243", + "accept": false, + "reason": "T2I-Adapter pipeline request vs IP-Adapter Face ID request; different features." + }, + { + "left": "issue:460", + "right": "issue:746", + "accept": false, + "reason": "Both are ONNX-related, but one is about dependency selection and the other is an offline revision loading failure." + }, + { + "left": "issue:5395", + "right": "issue:9297", + "accept": false, + "reason": "Both involve LoRAs, but one is about broken PEFT merging and the other about linear load-time slowdown." + }, + { + "left": "issue:3975", + "right": "issue:5989", + "accept": false, + "reason": "Typo in an error message is unrelated to a safetensors.torch bug." + }, + { + "left": "issue:4210", + "right": "issue:5225", + "accept": false, + "reason": "Attention_mask usage in UNet is unrelated to SDXL VRAM exhaustion during forward pass." + }, + { + "left": "issue:8432", + "right": "issue:8742", + "accept": false, + "reason": "ImportError in FromOriginalVAEMixin and removal of LoRAAttnProcessor2_0 are different compatibility issues." + }, + { + "left": "issue:7049", + "right": "issue:7695", + "accept": false, + "reason": "Both touch IP-Adapter, but one is an unpacking error in the basic example and the other is ignored cross_attention_kwargs." + }, + { + "left": "issue:460", + "right": "issue:4646", + "accept": false, + "reason": "ONNX runtime dependency issue vs ControlNet-to-ONNX conversion failure; not the same bug." + }, + { + "left": "issue:9358", + "right": "issue:9464", + "accept": false, + "reason": "Both are FLUX training script issues, but one is redundant reinitialization and the other is a list index error." + }, + { + "left": "issue:11644", + "right": "issue:8493", + "accept": false, + "reason": "LTXConditionPipeline init failure and SD3Transformer2DModel state_dict loading are separate model-loading bugs." + }, + { + "left": "issue:10111", + "right": "issue:9822", + "accept": false, + "reason": "Broad FLUX pipeline bug report vs specifically loading SDXL LoRAs into Flux; not the same concrete change." + }, + { + "left": "issue:12195", + "right": "issue:8320", + "accept": false, + "reason": "PyTorch 2.3.1 custom_op failure in AutoencoderKL is unrelated to the from_single_file __name__ attribute error." + }, + { + "left": "issue:2185", + "right": "issue:911", + "accept": false, + "reason": "Missing training config files and Dreambooth returning bad config are different configuration issues." + }, + { + "left": "issue:3576", + "right": "issue:3643", + "accept": false, + "reason": "XFormers/Textual Inversion TypeError is unrelated to latent upsample INT_MAX decoding errors." + }, + { + "left": "issue:10636", + "right": "issue:8837", + "accept": false, + "reason": "value_guided_sampling.py bug and SlicedAttnProcessor bug are different code paths." + }, + { + "left": "issue:8604", + "right": "issue:9930", + "accept": false, + "reason": "T5 float16 corruption and PAG adaptive scale bug are unrelated." + }, + { + "left": "issue:6110", + "right": "issue:8752", + "accept": false, + "reason": "Both are offline-related, but one concerns loading LoRA weights and the other from_single_file SDXL loading." + }, + { + "left": "issue:2310", + "right": "issue:4750", + "accept": true, + "reason": "Both point to the same LaMa/inpaint preprocessing area; 4750 is the clearer concrete report and 2310 appears to describe the same functionality more vaguely." + }, + { + "left": "issue:3073", + "right": "issue:5480", + "accept": false, + "reason": "Consistency models request and EDM request are separate model-feature requests." + }, + { + "left": "issue:4445", + "right": "issue:5656", + "accept": false, + "reason": "General LoRA finetuning question vs adding a LoRA loader to community pipelines; different scope." + }, + { + "left": "issue:5777", + "right": "issue:6359", + "accept": false, + "reason": "PixArt dreambooth training error and EulerDiscreteScheduler velocity/get_noise issue are unrelated." + }, + { + "left": "issue:1543", + "right": "issue:746", + "accept": false, + "reason": "textual_inversion revision handling and ONNX img2img revision loading are different loader bugs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative: the only accepted pair is a plausible duplicate-like LaMa/inpaint preprocessing match, while the rest are appropriately rejected as different bugs or feature requests. The canonical issue choice is also defensible based on the clearer title of issue:4750." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 203, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10003", + "issue:11225", + "issue:11437", + "issue:12574", + "issue:1274", + "issue:13477", + "issue:1642", + "issue:1785", + "issue:2306", + "issue:2772", + "issue:3325", + "issue:3453", + "issue:3585", + "issue:3630", + "issue:4314", + "issue:4480", + "issue:4697", + "issue:4789", + "issue:4801", + "issue:4839", + "issue:4912", + "issue:5160", + "issue:5162", + "issue:5285", + "issue:5300", + "issue:5594", + "issue:5633", + "issue:5925", + "issue:6109", + "issue:6196", + "issue:6357", + "issue:6418", + "issue:6749", + "issue:6823", + "issue:7041", + "issue:709", + "issue:730", + "issue:7306", + "issue:7417", + "issue:817", + "issue:8509", + "issue:852", + "issue:8604", + "issue:8837", + "issue:898", + "issue:911", + "issue:9134", + "issue:9178" + ], + "soft_pairs": [ + "issue:4801|issue:817", + "issue:12574|issue:13477", + "issue:3453|issue:898", + "issue:9134|issue:9178", + "issue:6109|issue:6196", + "issue:3585|issue:852", + "issue:4480|issue:7417", + "issue:5633|issue:8837", + "issue:6823|issue:8509", + "issue:1274|issue:1785", + "issue:11225|issue:11437", + "issue:3325|issue:4839", + "issue:10003|issue:7041", + "issue:1642|issue:709", + "issue:2306|issue:3630", + "issue:4697|issue:6749", + "issue:4912|issue:5162", + "issue:5160|issue:5594", + "issue:5285|issue:5925", + "issue:5300|issue:6418", + "issue:730|issue:911", + "issue:7306|issue:8604", + "issue:2772|issue:6357", + "issue:4314|issue:4789" + ], + "prepared_review_unit_hash": "37b8d342bf856b99cc7f56f174a706c9abf6199a043e0514dbf6765934d1b7a5", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33346, + "estimated_input_tokens": 8337, + "estimated_eval_tokens": 16930 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of mostly unrelated Diffusers bug reports and feature requests around training, inference, LoRA/ControlNet, and platform-specific failures. None of the soft pairs look like strong same-bug duplicates, so I would keep the issues separated.", + "confidence": 0.91, + "canonical_issue_reason": "issue:10003 is the clearest, concrete runtime bug in a core Stable Diffusion scheduler/inference path, and it is more specific than the broader training/example reports in the set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10003 is the best representative issue because it describes a reproducible scheduler failure with a precise trigger, making it the most actionable anchor for the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4801", + "right": "issue:817", + "accept": false, + "reason": "Both mention mixed precision in training examples, but they are different scripts and likely different failure modes; too broad to treat as the same bug." + }, + { + "left": "issue:12574", + "right": "issue:13477", + "accept": false, + "reason": "Unrelated components and symptoms: WanVideo/VACE control-arg handling versus an ErnieImage callback scope issue." + }, + { + "left": "issue:3453", + "right": "issue:898", + "accept": false, + "reason": "Different backend/hardware failures: CUDA BF16 triu/tril support versus TPU BF16 scalar-type support." + }, + { + "left": "issue:9134", + "right": "issue:9178", + "accept": false, + "reason": "Both are Flux LoRA-related, but one concerns SimpleTuner-trained LoRAs and the other PEFT target-module errors; not the same concrete bug." + }, + { + "left": "issue:6109", + "right": "issue:6196", + "accept": false, + "reason": "Related topic area, but one is a ControlNet-XS depth feature/usage issue and the other is a loading error for a specific SDXL model." + }, + { + "left": "issue:3585", + "right": "issue:852", + "accept": false, + "reason": "Different problems: LoRA synthesis into SD 1.5 versus an MPS device_map keyword crash." + }, + { + "left": "issue:4480", + "right": "issue:7417", + "accept": false, + "reason": "Both are Mac-related, but one is an MPS kernel crash during generation and the other is a test failure around LoRA fuse NaNs." + }, + { + "left": "issue:5633", + "right": "issue:8837", + "accept": false, + "reason": "Different code paths and symptoms in temporal transformer versus sliced attention processor logic." + }, + { + "left": "issue:6823", + "right": "issue:8509", + "accept": false, + "reason": "One is a generic inability to use ControlNet; the other is a ControlNet++ feature request." + }, + { + "left": "issue:1274", + "right": "issue:1785", + "accept": false, + "reason": "Both mention num_images_per_prompt, but the pipelines and failure contexts differ enough that this is not clearly one bug." + }, + { + "left": "issue:11225", + "right": "issue:11437", + "accept": false, + "reason": "Validation OOM versus missing outputs/checkpoints-images during training are distinct issues, even if both occur in training flows." + }, + { + "left": "issue:3325", + "right": "issue:4839", + "accept": false, + "reason": "Different failures in different features: DreamBooth CUDA invalid argument versus lora_scale type error." + }, + { + "left": "issue:10003", + "right": "issue:7041", + "accept": false, + "reason": "Both involve inference-step crashes, but the described triggers and symptoms are different enough that they should not be merged without stronger evidence." + }, + { + "left": "issue:1642", + "right": "issue:709", + "accept": false, + "reason": "Scheduler implementation bug versus documentation typo; clearly unrelated." + }, + { + "left": "issue:2306", + "right": "issue:3630", + "accept": false, + "reason": "Generic runtime error versus unknown UNet2DModel error; no evidence of the same underlying defect." + }, + { + "left": "issue:4697", + "right": "issue:6749", + "accept": false, + "reason": "A specific LoRA load regression and an InstantID reproduction issue are not the same bug." + }, + { + "left": "issue:4912", + "right": "issue:5162", + "accept": false, + "reason": "Both are conversion-related, but one asks for SDXL checkpoint conversion and the other for folder-to-safetensors conversion; different requests." + }, + { + "left": "issue:5160", + "right": "issue:5594", + "accept": false, + "reason": "DreamBooth/LoRA masking feature request versus SDXL LoRA memory-saving request; not the same change." + }, + { + "left": "issue:5285", + "right": "issue:5925", + "accept": false, + "reason": "Docs translation work and a ZipLora feature request are unrelated." + }, + { + "left": "issue:5300", + "right": "issue:6418", + "accept": false, + "reason": "Internet-connection complaint versus stable video diffusion output collapse; no shared bug surface." + }, + { + "left": "issue:730", + "right": "issue:911", + "accept": false, + "reason": "Both concern DreamBooth, but one is a training-behavior report and the other is a bad config issue; too different to merge." + }, + { + "left": "issue:7306", + "right": "issue:8604", + "accept": false, + "reason": "Both mention float16, but they affect different pipelines/components and have different symptoms." + }, + { + "left": "issue:2772", + "right": "issue:6357", + "accept": false, + "reason": "Local dataset validation in train_controlnet.py and API-vs-local result differences are separate problems." + }, + { + "left": "issue:4314", + "right": "issue:4789", + "accept": false, + "reason": "Interrupted-download write-protection and inability to download DeepFloyd IF are different download issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 204, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10076", + "issue:10147", + "issue:10472", + "issue:10569", + "issue:10848", + "issue:10873", + "issue:10896", + "issue:10932", + "issue:11000", + "issue:1161", + "issue:11721", + "issue:12108", + "issue:1255", + "issue:13043", + "issue:13488", + "issue:13506", + "issue:1377", + "issue:2149", + "issue:2150", + "issue:2311", + "issue:2444", + "issue:2775", + "issue:305", + "issue:3073", + "issue:3097", + "issue:3769", + "issue:4000", + "issue:4731", + "issue:4808", + "issue:4912", + "issue:4964", + "issue:5091", + "issue:5576", + "issue:6008", + "issue:6818", + "issue:7766", + "issue:8518", + "issue:8588", + "issue:8759", + "issue:9026", + "issue:9166", + "issue:9241", + "issue:9258", + "issue:9479", + "issue:9602", + "issue:9867", + "issue:987", + "issue:993" + ], + "soft_pairs": [ + "issue:9479|issue:9867", + "issue:6818|issue:7766", + "issue:10873|issue:11721", + "issue:3073|issue:5576", + "issue:10147|issue:10896", + "issue:10569|issue:10932", + "issue:10472|issue:8759", + "issue:2311|issue:5576", + "issue:4731|issue:8518", + "issue:9026|issue:9166", + "issue:12108|issue:13488", + "issue:10848|issue:9602", + "issue:2444|issue:2775", + "issue:5091|issue:9258", + "issue:1255|issue:305", + "issue:4808|issue:6008", + "issue:2149|issue:2150", + "issue:4000|issue:4964", + "issue:10076|issue:9241", + "issue:11000|issue:8588", + "issue:1161|issue:4912", + "issue:1377|issue:305", + "issue:13043|issue:13506", + "issue:3097|issue:993", + "issue:3769|issue:987" + ], + "prepared_review_unit_hash": "6382d9f62cf7d186a0d18036a312e6707871b4fbbd802cee250ca9a2cc1c131c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34322, + "estimated_input_tokens": 8581, + "estimated_eval_tokens": 17418 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated feature requests, conversion scripts, pipeline-specific bugs, and dependency/runtime errors across many different models and code paths. No single issue or PR meaningfully represents the whole set, and none of the soft-edge pairs are close enough to be true duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9479", + "right": "issue:9867", + "accept": false, + "reason": "Both are inpaint/mask-related symptoms, but they affect different pipelines (StableDiffusion vs Flux) and do not point to the same concrete bug." + }, + { + "left": "issue:6818", + "right": "issue:7766", + "accept": false, + "reason": "One is an unload/runtime error with IP-Adapter; the other is a FaceID usage question. Different problem and code path." + }, + { + "left": "issue:10873", + "right": "issue:11721", + "accept": false, + "reason": "These Sana issues are not the same: one is model/pipeline-specific breakage, the other is a generic load crash." + }, + { + "left": "issue:3073", + "right": "issue:5576", + "accept": false, + "reason": "Both are requests, but one asks for consistency models and the other for a FreeNoise scheduler; separate features." + }, + { + "left": "issue:10147", + "right": "issue:10896", + "accept": false, + "reason": "Different support requests for different models/tasks: multi-view generation vs expressive portrait animation." + }, + { + "left": "issue:10569", + "right": "issue:10932", + "accept": false, + "reason": "A CPU memory issue for HunyuanVideo is unrelated to the torch.compile recompile_limit problem." + }, + { + "left": "issue:10472", + "right": "issue:8759", + "accept": false, + "reason": "Different failures: LoRA `lora_bias` init error versus an SD3 image processor warning." + }, + { + "left": "issue:2311", + "right": "issue:5576", + "accept": false, + "reason": "Both mention schedulers, but they request different ones (UniPC vs FreeNoise) and are not the same change." + }, + { + "left": "issue:4731", + "right": "issue:8518", + "accept": false, + "reason": "Both involve torch.compile, but they are distinct failures in different pipelines and code paths." + }, + { + "left": "issue:9026", + "right": "issue:9166", + "accept": false, + "reason": "One is a changed offloading API keyword error; the other is unexpected overlay behavior. Not the same bug." + }, + { + "left": "issue:12108", + "right": "issue:13488", + "accept": false, + "reason": "Both touch flow/offloading behavior, but one is a scheduler-parameter pipeline break and the other is a dev-release offloading regression." + }, + { + "left": "issue:10848", + "right": "issue:9602", + "accept": false, + "reason": "Both are adapter/load-related, but one is a `load_model_dict_into_meta(device)` keyword error and the other is an adapter-name state mismatch." + }, + { + "left": "issue:2444", + "right": "issue:2775", + "accept": false, + "reason": "Different runtime failures in different contexts: a training script exit vs a torch._dynamo optimize error." + }, + { + "left": "issue:5091", + "right": "issue:9258", + "accept": false, + "reason": "Multi-GPU/offload usage and Windows-vs-Linux performance are unrelated issues." + }, + { + "left": "issue:1255", + "right": "issue:305", + "accept": false, + "reason": "Both are import errors, but for different symbols/modules and different dependency problems." + }, + { + "left": "issue:4808", + "right": "issue:6008", + "accept": false, + "reason": "A SHAP-E integration test failure is unrelated to the IP-Adapter plus face/LCM LoRA regression." + }, + { + "left": "issue:2149", + "right": "issue:2150", + "accept": false, + "reason": "They are related in theme, but one is a workflow request and the other is img2img texture projection support; not the same concrete change." + }, + { + "left": "issue:4000", + "right": "issue:4964", + "accept": false, + "reason": "Both are conversion requests, but they target different model families and formats (SD2 ckpt vs SDXL)." + }, + { + "left": "issue:10076", + "right": "issue:9241", + "accept": false, + "reason": "Both mention conversion scripts, but they convert different things: HF-to-original format vs XLab LoRA-to-diffusers LoRA." + }, + { + "left": "issue:11000", + "right": "issue:8588", + "accept": false, + "reason": "A missing `shift` argument in WanImageToVideoPipeline is unrelated to a missing SD3 conversion script." + }, + { + "left": "issue:1161", + "right": "issue:4912", + "accept": false, + "reason": "Flax-to-PyTorch conversion and SDXL checkpoint-to-diffusers conversion are different tasks." + }, + { + "left": "issue:1377", + "right": "issue:305", + "accept": false, + "reason": "Different import errors from different packages/symbols; no shared underlying bug is evident." + }, + { + "left": "issue:13043", + "right": "issue:13506", + "accept": false, + "reason": "These are distinct attention-backend issues: Blackwell kernel support vs ring-attention incompatibility." + }, + { + "left": "issue:3097", + "right": "issue:993", + "accept": false, + "reason": "Textual inversion doing nothing and fp16 dtype mismatch are separate bugs with different causes." + }, + { + "left": "issue:3769", + "right": "issue:987", + "accept": false, + "reason": "Different pipeline defects: a quartered panorama output versus a green speech-to-image output." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 205, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:1013", + "issue:10321", + "issue:10514", + "issue:10520", + "issue:1062", + "issue:10674", + "issue:11033", + "issue:11041", + "issue:11555", + "issue:1185", + "issue:12995", + "issue:1572", + "issue:1617", + "issue:1708", + "issue:1866", + "issue:199", + "issue:201", + "issue:2202", + "issue:3073", + "issue:3116", + "issue:3162", + "issue:3301", + "issue:3891", + "issue:3974", + "issue:4349", + "issue:4676", + "issue:5008", + "issue:5344", + "issue:5565", + "issue:5840", + "issue:6028", + "issue:6427", + "issue:6688", + "issue:713", + "issue:7223", + "issue:7380", + "issue:7460", + "issue:7789", + "issue:7793", + "issue:7836", + "issue:8393", + "issue:917", + "issue:9170", + "issue:9306", + "issue:9778", + "issue:9895" + ], + "soft_pairs": [ + "issue:5344|issue:6427", + "issue:1572|issue:917", + "issue:5565|issue:9778", + "issue:1866|issue:3116", + "issue:5008|issue:7380", + "issue:7460|issue:7836", + "issue:7789|issue:7793", + "issue:10100|issue:7223", + "issue:1013|issue:3162", + "issue:1062|issue:1708", + "issue:3073|issue:3891", + "issue:3073|issue:6688", + "issue:10674|issue:9895", + "issue:2202|issue:3974", + "issue:6028|issue:9170", + "issue:10514|issue:11555", + "issue:10321|issue:10520", + "issue:1185|issue:3301", + "issue:11033|issue:12995", + "issue:11041|issue:9306", + "issue:1617|issue:713", + "issue:199|issue:201", + "issue:4349|issue:4676", + "issue:5840|issue:8393" + ], + "prepared_review_unit_hash": "b653d926b6f1c3cc482122ad4dcd6cd7b4bb406d2ea9653a60241d49f2541942", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32845, + "estimated_input_tokens": 8212, + "estimated_eval_tokens": 16680 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated Diffusers issues that only share broad keywords like pipelines, training, or model support. The proposed soft edges are all superficial similarities rather than the same concrete bug/change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5344", + "right": "issue:6427", + "accept": false, + "reason": "One is a missing dependency error; the other is a model conversion bug. Different failure modes and code paths." + }, + { + "left": "issue:1572", + "right": "issue:917", + "accept": false, + "reason": "Both mention inpainting, but one is a pipeline support gap and the other is an ONNX runtime error. Not the same bug." + }, + { + "left": "issue:5565", + "right": "issue:9778", + "accept": false, + "reason": "Both are run/tutorial related, but they concern different scripts and likely different setup problems." + }, + { + "left": "issue:1866", + "right": "issue:3116", + "accept": false, + "reason": "Both are scheduler-related img2img bugs, but they involve different schedulers and distinct incorrect behaviors." + }, + { + "left": "issue:5008", + "right": "issue:7380", + "accept": false, + "reason": "One is a specific architecture implementation request, the other is a separate distillation method request." + }, + { + "left": "issue:7460", + "right": "issue:7836", + "accept": false, + "reason": "AdaLayerNorm bug and CLIP overfitting are unrelated issues in different parts of training/inference." + }, + { + "left": "issue:7789", + "right": "issue:7793", + "accept": false, + "reason": "Different DeepFloyd problems: float32/MPS behavior versus LoRA unload errors." + }, + { + "left": "issue:10100", + "right": "issue:7223", + "accept": false, + "reason": "Both are feature requests for adding models, but they are different models and separate implementations." + }, + { + "left": "issue:1013", + "right": "issue:3162", + "accept": false, + "reason": "Both are compatibility-style errors, but one is a pipeline breakage and the other is a missing embedding argument." + }, + { + "left": "issue:1062", + "right": "issue:1708", + "accept": false, + "reason": "Dreambooth reproducibility/training quality is unrelated to config-file passing for inference." + }, + { + "left": "issue:3073", + "right": "issue:3891", + "accept": false, + "reason": "Both are add-a-pipeline requests, but for completely different model families and use cases." + }, + { + "left": "issue:3073", + "right": "issue:6688", + "accept": false, + "reason": "Same broad feature-request category, but not the same model or underlying change." + }, + { + "left": "issue:10674", + "right": "issue:9895", + "accept": false, + "reason": "Both involve FluxPipeline, but one is GGUF support and the other is a device placement regression." + }, + { + "left": "issue:2202", + "right": "issue:3974", + "accept": false, + "reason": "OpenVINO support request versus a safetensors loading bug in a different pipeline." + }, + { + "left": "issue:6028", + "right": "issue:9170", + "accept": false, + "reason": "Both mention high memory use, but one is an OOM report and the other asks about GPU requirements for SDXL+ControlNet." + }, + { + "left": "issue:10514", + "right": "issue:11555", + "accept": false, + "reason": "Sample-size/binning limitation and device_map auto support are unrelated pipeline features." + }, + { + "left": "issue:10321", + "right": "issue:10520", + "accept": false, + "reason": "One is a missing model file error; the other is a cross-device tensor placement error." + }, + { + "left": "issue:1185", + "right": "issue:3301", + "accept": false, + "reason": "Both are Dreambooth-related, but one is reproducibility and the other is a training script failure." + }, + { + "left": "issue:11033", + "right": "issue:12995", + "accept": false, + "reason": "Single-file SD1.5 loading and Cosmos-Predict2.5 video behavior are unrelated issues." + }, + { + "left": "issue:11041", + "right": "issue:9306", + "accept": false, + "reason": "Both mention device/offloading behavior, but they are different pipelines and different specific faults." + }, + { + "left": "issue:1617", + "right": "issue:713", + "accept": false, + "reason": "A missing-arguments CLI error is not the same as a generic training error." + }, + { + "left": "issue:199", + "right": "issue:201", + "accept": false, + "reason": "One is a dead-code cleanup request; the other is a test request." + }, + { + "left": "issue:4349", + "right": "issue:4676", + "accept": false, + "reason": "Text-encoder requirement bug and dark TAESD latents are different pipeline issues." + }, + { + "left": "issue:5840", + "right": "issue:8393", + "accept": false, + "reason": "Checkpoint resume/freezing bug is unrelated to slow shard loading." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 206, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10025", + "issue:10940", + "issue:11010", + "issue:11178", + "issue:11387", + "issue:11864", + "issue:11901", + "issue:12108", + "issue:12147", + "issue:1231", + "issue:12423", + "issue:1274", + "issue:1335", + "issue:1363", + "issue:1530", + "issue:2185", + "issue:2412", + "issue:2712", + "issue:3015", + "issue:3382", + "issue:3647", + "issue:3682", + "issue:4186", + "issue:4390", + "issue:4630", + "issue:4747", + "issue:4808", + "issue:490", + "issue:5108", + "issue:5197", + "issue:5224", + "issue:5595", + "issue:6095", + "issue:6101", + "issue:6370", + "issue:6374", + "issue:6393", + "issue:6695", + "issue:7348", + "issue:801", + "issue:852", + "issue:8918", + "issue:9001", + "issue:9034", + "issue:9424", + "issue:945", + "issue:962", + "issue:9822" + ], + "soft_pairs": [ + "issue:8918|issue:9424", + "issue:6101|issue:7348", + "issue:4186|issue:5108", + "issue:2185|issue:2712", + "issue:11387|issue:11864", + "issue:6095|issue:9001", + "issue:490|issue:801", + "issue:3682|issue:4630", + "issue:1335|issue:852", + "issue:2412|issue:3647", + "issue:4808|issue:6393", + "issue:10940|issue:9034", + "issue:1231|issue:5224", + "issue:3382|issue:5197", + "issue:10025|issue:9822", + "issue:1530|issue:4390", + "issue:3015|issue:6370", + "issue:4747|issue:6374", + "issue:5595|issue:6695", + "issue:11178|issue:12423", + "issue:11010|issue:12108", + "issue:1363|issue:945", + "issue:1274|issue:962", + "issue:11901|issue:12147" + ], + "prepared_review_unit_hash": "782fa1d436508aa7521e3956fe728f04ba44564b32cad4145d2acc8fb37285d2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33756, + "estimated_input_tokens": 8439, + "estimated_eval_tokens": 17134 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose topical cluster of Diffusers issues around model loading, Flux/Wan/Qwen support, schedulers, and training/runtime bugs. The soft pairs are mostly superficial similarity matches rather than true duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8918", + "right": "issue:9424", + "accept": false, + "reason": "Both mention broken training scripts, but one is SD3 Dreambooth and the other is a specific notebook regression; different code paths and fixes." + }, + { + "left": "issue:6101", + "right": "issue:7348", + "accept": false, + "reason": "Both involve ControlNet/inpaint behavior, but one is a performance complaint and the other is a scheduler output bug; not the same defect." + }, + { + "left": "issue:4186", + "right": "issue:5108", + "accept": false, + "reason": "Both are XL pipeline problems, but one is SDXL inpaint+refiner and the other is T2I multi-adapter exceptions; different features and failure modes." + }, + { + "left": "issue:2185", + "right": "issue:2712", + "accept": false, + "reason": "Both are config-related, but one is missing training config files and the other is missing a specific normalization config key; unrelated issues." + }, + { + "left": "issue:11387", + "right": "issue:11864", + "accept": false, + "reason": "One is an AutoencoderDC normalization implementation issue; the other is a torch.compile scoping error in encode. Different underlying bugs." + }, + { + "left": "issue:6095", + "right": "issue:9001", + "accept": false, + "reason": "Both are scheduler-related, but one is an IndexError in Euler discrete scheduling for SVD and the other is aspect-ratio handling in Auraflow; not the same change." + }, + { + "left": "issue:490", + "right": "issue:801", + "accept": false, + "reason": "Both discuss memory/attention behavior in image pipelines, but one is generic memory usage and the other is attention slicing in inpaint/img2img; too broad to merge." + }, + { + "left": "issue:3682", + "right": "issue:4630", + "accept": false, + "reason": "Both concern LoRAs, but one is loading weights and the other is changing LoRA weight after load; distinct functionality." + }, + { + "left": "issue:1335", + "right": "issue:852", + "accept": false, + "reason": "Both are constructor/runtime compatibility issues, but one is ONNX pipeline init and the other is MPS device_map handling; unrelated code paths." + }, + { + "left": "issue:2412", + "right": "issue:3647", + "accept": false, + "reason": "Both are Dreambooth/training-related, but one is a generic Dreambooth error and the other is specific to DeepFloyd support; not duplicates." + }, + { + "left": "issue:4808", + "right": "issue:6393", + "accept": false, + "reason": "Both involve loading-related failures, but one is an integration test for SHAP-E img2img and the other is private HF link download via from_single_file; different problems." + }, + { + "left": "issue:10940", + "right": "issue:9034", + "accept": false, + "reason": "Both mention import/requirements issues, but one is missing triton.ops and the other is an unnecessary sentencepiece requirement; not the same bug." + }, + { + "left": "issue:1231", + "right": "issue:5224", + "accept": false, + "reason": "Both mention schedulers/Flax, but one is a K-LMS scheduler feature request and the other is a Stable Diffusion 2.1 error; different contexts." + }, + { + "left": "issue:3382", + "right": "issue:5197", + "accept": false, + "reason": "Both are multi-GPU training performance issues, but one is CUDA OOM in a script and the other is slower-than-single-GPU throughput; not a duplicate fix." + }, + { + "left": "issue:10025", + "right": "issue:9822", + "accept": false, + "reason": "Both mention Flux, but one is about attention masks and the other is LoRA loading into Flux; separate functionality." + }, + { + "left": "issue:1530", + "right": "issue:4390", + "accept": false, + "reason": "Both are missing-configuration style reports, but one is a specific VQDiffusion pipeline field and the other is missing requirements in general; not the same issue." + }, + { + "left": "issue:3015", + "right": "issue:6370", + "accept": false, + "reason": "Both mention converting/loading LoRA or ControlNet assets, but one is ControlNet format conversion and the other is using diffusers LoRA in A1111; different tasks." + }, + { + "left": "issue:4747", + "right": "issue:6374", + "accept": false, + "reason": "Both concern loading fine-tuned/LoRA assets, but one is a missing model_index.json after load and the other is a Qwen LoRA weight question; unrelated." + }, + { + "left": "issue:5595", + "right": "issue:6695", + "accept": false, + "reason": "Both are runtime errors, but one is a shape divisibility ValueError and the other is a subprocess failure while starting training; no shared root cause." + }, + { + "left": "issue:11178", + "right": "issue:12423", + "accept": false, + "reason": "Both are single-file/model loading issues, but one is MistoLine ControlNet and the other is QwenImageTransformer2DModel.from_single_file; different models and loaders." + }, + { + "left": "issue:11010", + "right": "issue:12108", + "accept": false, + "reason": "Both relate to Flux/Chroma and flow-style models, but one is support for Chroma architecture changes and the other is a scheduler parameterization break; not mergeable as one bug." + }, + { + "left": "issue:1363", + "right": "issue:945", + "accept": false, + "reason": "One is an ONNX scheduler interaction error; the other is adding a custom checkpoint model for ONNX conversion. Different issue classes." + }, + { + "left": "issue:1274", + "right": "issue:962", + "accept": false, + "reason": "Both are SD pipeline shape/batch handling issues, but one is num_images_per_prompt regression and the other is inpaint size support; not the same defect." + }, + { + "left": "issue:11901", + "right": "issue:12147", + "accept": false, + "reason": "Both are Wan LoRA loading reports, but they target different model generations and likely different loaders/failure points; too weak for duplicate triage." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 207, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10191", + "issue:10227", + "issue:10302", + "issue:10520", + "issue:10590", + "issue:10891", + "issue:11220", + "issue:11436", + "issue:11569", + "issue:11644", + "issue:11653", + "issue:11710", + "issue:1889", + "issue:2053", + "issue:2635", + "issue:2676", + "issue:2905", + "issue:3199", + "issue:3300", + "issue:3975", + "issue:4159", + "issue:4265", + "issue:4303", + "issue:4923", + "issue:5317", + "issue:5320", + "issue:5327", + "issue:5509", + "issue:5558", + "issue:6101", + "issue:6423", + "issue:6452", + "issue:702", + "issue:7223", + "issue:7293", + "issue:7751", + "issue:7936", + "issue:7939", + "issue:8435", + "issue:8494", + "issue:8593", + "issue:8780", + "issue:9225", + "issue:9303", + "issue:9306", + "issue:9667", + "issue:9756", + "issue:9876" + ], + "soft_pairs": [ + "issue:11220|issue:9756", + "issue:5317|issue:5320", + "issue:10891|issue:11436", + "issue:8494|issue:8593", + "issue:5509|issue:9225", + "issue:10191|issue:9876", + "issue:1889|issue:2635", + "issue:7223|issue:9303", + "issue:10590|issue:10891", + "issue:2905|issue:6452", + "issue:4923|issue:6423", + "issue:10520|issue:9306", + "issue:10227|issue:11653", + "issue:11569|issue:11710", + "issue:11644|issue:8435", + "issue:2676|issue:702", + "issue:3300|issue:4265", + "issue:4303|issue:7751", + "issue:7936|issue:7939", + "issue:3975|issue:5327", + "issue:3199|issue:3300", + "issue:2053|issue:4159", + "issue:7293|issue:8780", + "issue:10302|issue:9667", + "issue:5558|issue:6101" + ], + "prepared_review_unit_hash": "be3273e6bf170122352579858d72eb7dcc497874dfc619c5fc7f82c53b65c72d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34180, + "estimated_input_tokens": 8545, + "estimated_eval_tokens": 17346 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated device-routing bugs, dtype/bfloat16 mismatches, feature requests, scheduler questions, docs issues, and model-loading failures. None of the soft edges look like true duplicates or the same fixable code path.", + "confidence": 0.96, + "canonical_issue_reason": "issue:8780 is the broadest generic symptom in the set ('Model and input data type is not same'), so it is the closest weak umbrella for the dtype-related subset, but the overall cluster is not cohesive enough to treat as one duplicate thread.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8780 is the most representative fallback because it is generic and open-ended, but it still only covers a subset of the issues and is not a strong global canonical for the whole cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11220", + "right": "issue:9756", + "accept": false, + "reason": "Docs page not working and missing notebook are unrelated documentation/navigation problems, not the same underlying bug." + }, + { + "left": "issue:5317", + "right": "issue:5320", + "accept": false, + "reason": "Both are PEFT-related, but one is documentation and the other is an API enhancement request; different changes." + }, + { + "left": "issue:10891", + "right": "issue:11436", + "accept": false, + "reason": "Both involve bfloat16, but they hit different components and failure modes: DCAE internal conversions vs Hidream matmul dtype mismatch." + }, + { + "left": "issue:8494", + "right": "issue:8593", + "accept": false, + "reason": "Unexpected force_hooks kwarg and CPU bfloat16 noise are different SD3 issues with no shared code path." + }, + { + "left": "issue:5509", + "right": "issue:9225", + "accept": false, + "reason": "Both are image-shape related, but one is prompt/image batch sizing and the other is RGBA latent sizing; different bugs." + }, + { + "left": "issue:10191", + "right": "issue:9876", + "accept": false, + "reason": "GPU device selection during training and VRAM not being released after LoRA training are different resource-management problems." + }, + { + "left": "issue:1889", + "right": "issue:2635", + "accept": false, + "reason": "A scheduler variant request and a scheduler-name mapping question are related only broadly, not the same issue or fix." + }, + { + "left": "issue:7223", + "right": "issue:9303", + "accept": false, + "reason": "Separate feature requests for different video models; not the same underlying change." + }, + { + "left": "issue:10590", + "right": "issue:10891", + "accept": false, + "reason": "Both mention bfloat16, but one is Sana tiled VAE decode and the other is DCAE internal type conversion; different code paths." + }, + { + "left": "issue:2905", + "right": "issue:6452", + "accept": false, + "reason": "Adding Karras sigmas to schedulers and adding masked_image_latents are unrelated feature requests." + }, + { + "left": "issue:4923", + "right": "issue:6423", + "accept": false, + "reason": "AudioLDM2 transcript support and fill50k reproduction are unrelated issues." + }, + { + "left": "issue:10520", + "right": "issue:9306", + "accept": false, + "reason": "Both are device-placement errors, but they concern different pipelines and likely different bugs." + }, + { + "left": "issue:10227", + "right": "issue:11653", + "accept": false, + "reason": "Both concern LoRA support, but one is a feature limitation on feature sizes and the other is .alpha-key compatibility; not the same fix." + }, + { + "left": "issue:11569", + "right": "issue:11710", + "accept": false, + "reason": "Different feature requests for different models/projects; no shared code defect." + }, + { + "left": "issue:11644", + "right": "issue:8435", + "accept": false, + "reason": "LTXConditionPipeline init failure and from_single_file runtime error are different loading paths and symptoms." + }, + { + "left": "issue:2676", + "right": "issue:702", + "accept": false, + "reason": "Scheduler naming question and reverse sampling proposal are about the same area but not the same underlying change." + }, + { + "left": "issue:3300", + "right": "issue:4265", + "accept": false, + "reason": "A general 'Diffusers on HuggingFace are broken' report and an import error for a specific pipeline are not the same bug." + }, + { + "left": "issue:4303", + "right": "issue:7751", + "accept": false, + "reason": "Flax SDXL loading crash and animatediff conversion errors are unrelated." + }, + { + "left": "issue:7936", + "right": "issue:7939", + "accept": false, + "reason": "Both are loading failures, but one is invalid JSON in a safetensors file and the other is loading from the same folder; different causes." + }, + { + "left": "issue:3975", + "right": "issue:5327", + "accept": false, + "reason": "A typo in an error message and SDXL textual inversion loading failure are unrelated." + }, + { + "left": "issue:3199", + "right": "issue:3300", + "accept": false, + "reason": "Custom pipelines broken and HF-hosted diffusers broken are too broad and refer to different problems." + }, + { + "left": "issue:2053", + "right": "issue:4159", + "accept": false, + "reason": "Both mention VRAM, but one is a version-wide regression and the other is DiffEdit's high memory use; not the same issue." + }, + { + "left": "issue:7293", + "right": "issue:8780", + "accept": false, + "reason": "AudioLDM2 model integration request and a generic dtype mismatch error are unrelated." + }, + { + "left": "issue:10302", + "right": "issue:9667", + "accept": false, + "reason": "Both involve FP8/transformer-flux context, but one reports inference noise without CPU offloading and the other a local file loading error; different fixes." + }, + { + "left": "issue:5558", + "right": "issue:6101", + "accept": false, + "reason": "Same pipeline family, but one is output artefacts with multiple ControlNets and the other is performance regression; not the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 208, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10591", + "issue:10772", + "issue:10855", + "issue:12147", + "issue:13279", + "issue:1343", + "issue:1961", + "issue:2150", + "issue:2220", + "issue:2312", + "issue:2331", + "issue:2390", + "issue:2424", + "issue:2453", + "issue:2644", + "issue:2783", + "issue:2843", + "issue:3480", + "issue:3689", + "issue:3694", + "issue:3740", + "issue:4453", + "issue:4724", + "issue:4782", + "issue:4874", + "issue:4907", + "issue:4923", + "issue:5197", + "issue:5829", + "issue:6055", + "issue:6069", + "issue:6090", + "issue:6288", + "issue:6357", + "issue:6399", + "issue:6570", + "issue:7210", + "issue:7895", + "issue:8323", + "issue:8477", + "issue:8514", + "issue:8543", + "issue:8994", + "issue:9069", + "issue:9225", + "issue:9293", + "issue:9511", + "issue:9979" + ], + "soft_pairs": [ + "issue:2150|issue:2783", + "issue:6069|issue:8323", + "issue:6570|issue:9225", + "issue:10772|issue:9511", + "issue:10855|issue:8543", + "issue:3694|issue:5197", + "issue:2331|issue:2390", + "issue:2453|issue:4923", + "issue:4724|issue:8994", + "issue:3740|issue:7210", + "issue:4782|issue:4907", + "issue:4874|issue:6357", + "issue:12147|issue:13279", + "issue:8477|issue:8514", + "issue:6090|issue:9979", + "issue:2220|issue:2644", + "issue:2312|issue:2843", + "issue:3480|issue:3689", + "issue:4453|issue:6399", + "issue:9069|issue:9293", + "issue:1961|issue:6055", + "issue:10591|issue:7895", + "issue:1343|issue:2424", + "issue:5829|issue:6288" + ], + "prepared_review_unit_hash": "69294bcaec44989ad1b27bbceaba42f1a8c202fed85d6977b4c741cfaaea570a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33639, + "estimated_input_tokens": 8410, + "estimated_eval_tokens": 17076 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a noisy, heterogeneous cluster of unrelated issue reports across training scripts, pipeline features, scheduler/runtime errors, and dependency/loading problems. None of the soft edges look like the same underlying bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2150", + "right": "issue:2783", + "accept": false, + "reason": "Both are ControlNet-adjacent, but one is about texture projection with img2img and the other is a request for an img2img/inpainting ControlNet pipeline; different feature work." + }, + { + "left": "issue:6069", + "right": "issue:8323", + "accept": false, + "reason": "Different failure modes: a scheduler math bug at t=0 versus a Chinese-language scheduler file recognition issue." + }, + { + "left": "issue:6570", + "right": "issue:9225", + "accept": false, + "reason": "Both involve SDXL InstructPix2Pix, but one is a tensor-size mismatch in the pipeline and the other is a latent-size issue specifically with RGBA images; related but not the same concrete bug." + }, + { + "left": "issue:10772", + "right": "issue:9511", + "accept": false, + "reason": "Sana ControlNet support request and multi-ControlNet batching for SDXL inpaint are separate feature requests for different pipelines." + }, + { + "left": "issue:10855", + "right": "issue:8543", + "accept": false, + "reason": "Gradient accumulation stepping every iteration is an optimizer loop bug, while the SD3 dreambooth issue is about gradient checkpointing failure; unrelated training problems." + }, + { + "left": "issue:3694", + "right": "issue:5197", + "accept": false, + "reason": "One is a missing model/config path during example fine-tuning; the other is multi-GPU training being slower than single GPU. Different symptoms and root causes." + }, + { + "left": "issue:2331", + "right": "issue:2390", + "accept": false, + "reason": "Both are pipeline feature requests, but ControlNet and T2I-Adapter are distinct integrations with different code paths." + }, + { + "left": "issue:2453", + "right": "issue:4923", + "accept": false, + "reason": "A general contribution question is unrelated to an AudioLDM 2 feature request." + }, + { + "left": "issue:4724", + "right": "issue:8994", + "accept": false, + "reason": "Both mention generation behavior, but one is SDXL img2img not working as expected and the other is steps stuck at 0; not the same bug." + }, + { + "left": "issue:3740", + "right": "issue:7210", + "accept": false, + "reason": "Different accelerate/device_map issues: legacy attention conversion with sequential dispatch versus missing named_parameters on a pipeline object." + }, + { + "left": "issue:4782", + "right": "issue:4907", + "accept": false, + "reason": "Inpainting mask leakage versus a warning about a new inpainting model under SDXL; different problems." + }, + { + "left": "issue:4874", + "right": "issue:6357", + "accept": false, + "reason": "Upscale pipeline local-loading bug and API-vs-local output differences are not the same issue." + }, + { + "left": "issue:12147", + "right": "issue:13279", + "accept": false, + "reason": "Different model-loading failures: Wan2.2 Lightning LoRA loading versus AutoencoderRAE loading with older transformers." + }, + { + "left": "issue:8477", + "right": "issue:8514", + "accept": false, + "reason": "Both are distributed training errors, but one is a DDP error in a consistency script and the other is an EMAModel/InstructP2P parallel fine-tuning failure; not the same code path." + }, + { + "left": "issue:6090", + "right": "issue:9979", + "accept": false, + "reason": "Both are tensor/channel mismatch errors, but they occur in different pipelines and likely different model-input contracts." + }, + { + "left": "issue:2220", + "right": "issue:2644", + "accept": false, + "reason": "Both ask about multi-GPU DreamBooth training, but one is general problems and the other is whether training on two 12GB GPUs is supported; not the same concrete bug." + }, + { + "left": "issue:2312", + "right": "issue:2843", + "accept": false, + "reason": "EMA hook save error versus multi-GPU checkpoint saving; both are saving-related but distinct failure points and contexts." + }, + { + "left": "issue:3480", + "right": "issue:3689", + "accept": false, + "reason": "A request for a text-encoder LoRA scale parameter is unrelated to unloading LoRA weights after loading." + }, + { + "left": "issue:4453", + "right": "issue:6399", + "accept": false, + "reason": "Both concern LoRA compatibility/visibility, but one is SDXL LoRA conversion for A1111 while the other is pytorch_lora_weights visibility in UIs; different interoperability problems." + }, + { + "left": "issue:9069", + "right": "issue:9293", + "accept": false, + "reason": "A NumPy type mismatch versus a missing Float8 storage backend are unrelated dependency/type errors." + }, + { + "left": "issue:1961", + "right": "issue:6055", + "accept": false, + "reason": "A ColossalAI-specific error and a generic accelerate error are too underspecified and appear unrelated." + }, + { + "left": "issue:10591", + "right": "issue:7895", + "accept": false, + "reason": "SD3 LoRA training-script problem and Flax memory-efficient attention bug are different frameworks and code paths." + }, + { + "left": "issue:1343", + "right": "issue:2424", + "accept": false, + "reason": "Enabling xformers by default is a feature/configuration change; flake8-comprehension checks are linting, not the same change." + }, + { + "left": "issue:5829", + "right": "issue:6288", + "accept": false, + "reason": "LCM training script missing an argument versus default LR being too small; both are training-script issues but not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 209, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10066", + "issue:10100", + "issue:1064", + "issue:11907", + "issue:12037", + "issue:1546", + "issue:1604", + "issue:1831", + "issue:1941", + "issue:2044", + "issue:2177", + "issue:2340", + "issue:2614", + "issue:2646", + "issue:2761", + "issue:3191", + "issue:3933", + "issue:3978", + "issue:4014", + "issue:4210", + "issue:4376", + "issue:4447", + "issue:4593", + "issue:4638", + "issue:5981", + "issue:5989", + "issue:6113", + "issue:6238", + "issue:6283", + "issue:6295", + "issue:6632", + "issue:6689", + "issue:7000", + "issue:7002", + "issue:709", + "issue:7219", + "issue:8465", + "issue:851", + "issue:8588", + "issue:9063", + "issue:9141", + "issue:9366", + "issue:9511", + "issue:9619", + "issue:968", + "issue:9895", + "issue:9904", + "issue:9972" + ], + "soft_pairs": [ + "issue:4447|issue:8465", + "issue:10066|issue:12037", + "issue:5981|issue:5989", + "issue:4014|issue:6295", + "issue:1064|issue:1546", + "issue:2044|issue:851", + "issue:2614|issue:968", + "issue:9511|issue:9619", + "issue:1831|issue:2340", + "issue:7002|issue:9972", + "issue:4210|issue:7000", + "issue:6113|issue:6689", + "issue:4376|issue:4638", + "issue:10100|issue:7219", + "issue:2646|issue:6238", + "issue:3933|issue:4593", + "issue:11907|issue:9366", + "issue:9141|issue:9904", + "issue:1604|issue:3191", + "issue:9063|issue:9895", + "issue:1941|issue:2177", + "issue:2761|issue:709", + "issue:3978|issue:6283", + "issue:6632|issue:8588" + ], + "prepared_review_unit_hash": "7118101572ceeff53b2af18f3bd4c1cd9824633c08a299ad15880063128fd060", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33854, + "estimated_input_tokens": 8464, + "estimated_eval_tokens": 17184 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly unrelated issues spanning different pipelines, schedulers, and feature requests. The only clear duplicate pair is the SDXL textual inversion / load_textual_inversion failure, where both reports describe the same underlying bug.", + "confidence": 0.89, + "canonical_issue_reason": "issue:4638 is the most general and direct statement of the bug (\u201cload_textual_inversion doesn't work on SDXL\u201d), making it a good canonical representative for the duplicate SDXL textual inversion reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4638 best represents the shared underlying bug because it is concise, specific to SDXL, and broad enough to cover the crash report in issue:4376.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4447", + "right": "issue:8465", + "accept": false, + "reason": "Different problems: one is a restricted-model access/auth issue, the other is a from_single_file config loading error." + }, + { + "left": "issue:10066", + "right": "issue:12037", + "accept": false, + "reason": "Both mention model incompatibility, but they are different model architectures and failure modes, not the same bug." + }, + { + "left": "issue:5981", + "right": "issue:5989", + "accept": false, + "reason": "Unrelated: scheduler bug vs safetensors.torch bug." + }, + { + "left": "issue:4014", + "right": "issue:6295", + "accept": false, + "reason": "Both involve image artifacts, but the titles point to different schedulers and likely different root causes." + }, + { + "left": "issue:1064", + "right": "issue:1546", + "accept": false, + "reason": "Different code paths: LMSDiscreteScheduler.add_noise() error vs batched-input truncation warning handling." + }, + { + "left": "issue:2044", + "right": "issue:851", + "accept": false, + "reason": "Both are dtype-related errors, but they are different runtime contexts and not clearly the same defect." + }, + { + "left": "issue:2614", + "right": "issue:968", + "accept": false, + "reason": "Download-directory configuration and model-load slowness are distinct issues." + }, + { + "left": "issue:9511", + "right": "issue:9619", + "accept": false, + "reason": "Different ControlNet requests: batching for inpaint vs interrupt support." + }, + { + "left": "issue:1831", + "right": "issue:2340", + "accept": false, + "reason": "Different pipelines and failures: Dreambooth class-image precision mismatch vs safety-checker error in tiny SD testing." + }, + { + "left": "issue:7002", + "right": "issue:9972", + "accept": false, + "reason": "Both are huge-buffer symptoms, but the affected pipelines and reported behavior differ too much to treat as one bug." + }, + { + "left": "issue:4210", + "right": "issue:7000", + "accept": false, + "reason": "A forward-signature/attention_mask question is unrelated to a PyTorch deprecation warning." + }, + { + "left": "issue:6113", + "right": "issue:6689", + "accept": false, + "reason": "Different tasks: loading control models from a repo vs converting a specific checkpoint to Diffusers." + }, + { + "left": "issue:4376", + "right": "issue:4638", + "accept": true, + "reason": "Same underlying bug: SDXL load_textual_inversion fails/crashes in both reports." + }, + { + "left": "issue:10100", + "right": "issue:7219", + "accept": false, + "reason": "Different feature requests for different models/upscalers." + }, + { + "left": "issue:2646", + "right": "issue:6238", + "accept": false, + "reason": "Both mention VAE, but one is tiled-VAE IndexError and the other is an SDXL VAE issue with different symptoms." + }, + { + "left": "issue:3933", + "right": "issue:4593", + "accept": false, + "reason": "Graph compile logging issue and LoRA rank mismatch are unrelated." + }, + { + "left": "issue:11907", + "right": "issue:9366", + "accept": false, + "reason": "Both involve schedulers, but one is a Flux pipeline incompatibility and the other is an img2img strength/step-count edge case." + }, + { + "left": "issue:9141", + "right": "issue:9904", + "accept": false, + "reason": "Both are FluxPipeline issues, but one is missing from_single_file() and the other is image-shape rounding." + }, + { + "left": "issue:1604", + "right": "issue:3191", + "accept": false, + "reason": "Different failures in different features: lpw_stable_diffusion breakage vs Dreambooth+LoRA breakage." + }, + { + "left": "issue:9063", + "right": "issue:9895", + "accept": false, + "reason": "Tokenizer instantiation failure and device-mismatch inference error are distinct bugs." + }, + { + "left": "issue:1941", + "right": "issue:2177", + "accept": false, + "reason": "Custom pipeline loading for UnCLIP is unrelated to AutoencoderKL loading failure." + }, + { + "left": "issue:2761", + "right": "issue:709", + "accept": false, + "reason": "Both are documentation-related, but they target different APIs and documentation gaps." + }, + { + "left": "issue:3978", + "right": "issue:6283", + "accept": false, + "reason": "Different img2img failures on different pipelines and platforms." + }, + { + "left": "issue:6632", + "right": "issue:8588", + "accept": false, + "reason": "A training-script question and a missing conversion script are not the same issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The output is grounded in the packet and stays conservative. The only accepted duplicate pair, issue:4376 vs issue:4638, is well supported by the titles, and the other pair rejections are phrased cautiously without overstating equivalence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 210, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10016", + "issue:10336", + "issue:10671", + "issue:1083", + "issue:11396", + "issue:11432", + "issue:1161", + "issue:12052", + "issue:12728", + "issue:13446", + "issue:1434", + "issue:1704", + "issue:2331", + "issue:2402", + "issue:2495", + "issue:3249", + "issue:3406", + "issue:4232", + "issue:4410", + "issue:4450", + "issue:4674", + "issue:4724", + "issue:4880", + "issue:5368", + "issue:5647", + "issue:5799", + "issue:5897", + "issue:6132", + "issue:6162", + "issue:6164", + "issue:6176", + "issue:6460", + "issue:6548", + "issue:6666", + "issue:7294", + "issue:7415", + "issue:7574", + "issue:7575", + "issue:759", + "issue:8469", + "issue:8729", + "issue:8809", + "issue:9148", + "issue:9303", + "issue:9476", + "issue:9637", + "issue:9680", + "issue:9792" + ], + "soft_pairs": [ + "issue:1083|issue:2495", + "issue:7415|issue:9476", + "issue:11432|issue:13446", + "issue:1434|issue:4674", + "issue:8809|issue:9637", + "issue:10016|issue:4410", + "issue:6176|issue:7574", + "issue:6666|issue:8729", + "issue:5368|issue:5897", + "issue:4450|issue:7294", + "issue:6132|issue:9303", + "issue:12052|issue:12728", + "issue:5647|issue:5799", + "issue:1161|issue:759", + "issue:4724|issue:6548", + "issue:8469|issue:9680", + "issue:3249|issue:6164", + "issue:10336|issue:10671", + "issue:11396|issue:9792", + "issue:1704|issue:3406", + "issue:6460|issue:9148", + "issue:2402|issue:4880", + "issue:2331|issue:4232", + "issue:6162|issue:7575" + ], + "prepared_review_unit_hash": "54f24c825176cd54f4100c5f3923694e30b493a11d77ed2b678047835c1f2e45", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34278, + "estimated_input_tokens": 8570, + "estimated_eval_tokens": 17396 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No duplicate cluster emerges here: the items span many unrelated bugs/features across loading, dtype, device placement, memory, training, and conversion. The soft pairs are mostly keyword-similar but describe different code paths or different model/pipeline families, so none should be merged.", + "confidence": 0.97, + "canonical_issue_reason": "No canonical issue: the set is heterogeneous and the paired items do not share the same underlying bug or repro path.", + "canonical_pr_reason": null, + "best_issue_reason": "No single best issue for deduplication: the collection is not a coherent duplicate group, so any one issue would be an arbitrary representative.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1083", + "right": "issue:2495", + "accept": false, + "reason": "Both are inference/output correctness issues, but one is FP16 ONNX and the other is an MPS upscale black-image bug; different backends and failure modes." + }, + { + "left": "issue:7415", + "right": "issue:9476", + "accept": false, + "reason": "Both involve mixed precision/bfloat16, but one is a TPU AutoencoderKL issue and the other is Flux LoRA validation dtype mismatch; different components and triggers." + }, + { + "left": "issue:11432", + "right": "issue:13446", + "accept": false, + "reason": "One is about torch_dtype='auto' on from_pretrained, the other about trust checks for custom_pipeline; unrelated API behaviors." + }, + { + "left": "issue:1434", + "right": "issue:4674", + "accept": false, + "reason": "OOM during upscaling is not the same as a width/height memory leak across calls; different symptom and likely different fix." + }, + { + "left": "issue:8809", + "right": "issue:9637", + "accept": false, + "reason": "Both mention tensor/shape-related behavior, but one is feedforward output variance with sequence length and the other is XFormers attention-mask/bfloat16 failure." + }, + { + "left": "issue:10016", + "right": "issue:4410", + "accept": false, + "reason": "Both are load-time shape mismatches, but for different models and parameters: SD3.5M pos_embed vs UNet conv_in.weight." + }, + { + "left": "issue:6176", + "right": "issue:7574", + "accept": false, + "reason": "Both are missing/mismatched weights, but one is LoRA loading after DreamBooth and the other is a specific checkpoint key error; not the same bug." + }, + { + "left": "issue:6666", + "right": "issue:8729", + "accept": false, + "reason": "Validation images being black is not the same as an unexpected validation image in a webdataset controlnet script; different pipeline stage and outcome." + }, + { + "left": "issue:5368", + "right": "issue:5897", + "accept": false, + "reason": "Both are dtype/device mismatches, but one is FP16 attention QKV dtype and the other is LoRA training across CPU/CUDA; different code paths." + }, + { + "left": "issue:4450", + "right": "issue:7294", + "accept": false, + "reason": "One is multi-text inversion not working, the other is a wrong text-encoder grad setting in Custom Diffusion; related area but distinct issues." + }, + { + "left": "issue:6132", + "right": "issue:9303", + "accept": false, + "reason": "Adding LOADABLE_CLASSES and adding a VEnhancer model are unrelated feature requests; no shared underlying bug/change." + }, + { + "left": "issue:12052", + "right": "issue:12728", + "accept": false, + "reason": "Both touch Wan 2.2, but one is offloading device-mismatch during multiplication and the other is torch.compile/channels_last with Dynamo unsupported behavior; distinct failures." + }, + { + "left": "issue:5647", + "right": "issue:5799", + "accept": false, + "reason": "Strength handling in SDXL inpainting is different from a general keyword-argument question about inpainting; not the same concrete bug." + }, + { + "left": "issue:1161", + "right": "issue:759", + "accept": false, + "reason": "Flax-to-PyTorch conversion and optional full-model saving for textual inversion are unrelated model/export topics." + }, + { + "left": "issue:4724", + "right": "issue:6548", + "accept": false, + "reason": "Both are img2img-related, but one concerns SDXL checkpoint behavior and the other misclassifies RGBA input as latents in prepare_latents; different defects." + }, + { + "left": "issue:8469", + "right": "issue:9680", + "accept": false, + "reason": "Corrupt IP-Adapter faces and broken SDXL unconditional generation are different generation bugs with no clear shared code-path." + }, + { + "left": "issue:3249", + "right": "issue:6164", + "accept": false, + "reason": "AttnUpBlock2D upsample_size argument support is unrelated to StableDiffusionPipeline latent output handling." + }, + { + "left": "issue:10336", + "right": "issue:10671", + "accept": false, + "reason": "Sana activation clamping/explosion and DiffusionPipeline dtype/device determinism are different runtime problems." + }, + { + "left": "issue:11396", + "right": "issue:9792", + "accept": false, + "reason": "A LoRA conversion/how-to question is unrelated to dataset cache mapping failure." + }, + { + "left": "issue:1704", + "right": "issue:3406", + "accept": false, + "reason": "A generic attention-block question is not the same as a bug where Attention fails for only_cross_attention=True." + }, + { + "left": "issue:6460", + "right": "issue:9148", + "accept": false, + "reason": "Endless loading for a specific inpainting model is unrelated to a request for adding a Flux inpainting pipeline." + }, + { + "left": "issue:2402", + "right": "issue:4880", + "accept": false, + "reason": "CUDA-not-enabled environment error and ControlNet image type validation are different classes of problems." + }, + { + "left": "issue:2331", + "right": "issue:4232", + "accept": false, + "reason": "Adding a ControlNet pipeline and training/inference code for instruct-pix2pix with ControlNet are different feature requests, not duplicate bugs." + }, + { + "left": "issue:6162", + "right": "issue:7575", + "accept": false, + "reason": "IP-Adapter SDXL shape mismatch and Transformer2DModel adanorm shape error are both shape-related but involve different modules and repros." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 211, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10382", + "issue:1062", + "issue:10672", + "issue:10695", + "issue:11223", + "issue:11569", + "issue:12110", + "issue:12633", + "issue:12783", + "issue:13105", + "issue:1463", + "issue:1566", + "issue:1672", + "issue:1931", + "issue:2266", + "issue:2440", + "issue:2549", + "issue:2695", + "issue:2984", + "issue:3045", + "issue:3087", + "issue:3139", + "issue:3467", + "issue:3647", + "issue:3689", + "issue:384", + "issue:3915", + "issue:3944", + "issue:4331", + "issue:460", + "issue:4692", + "issue:4912", + "issue:5000", + "issue:5402", + "issue:5518", + "issue:5574", + "issue:5575", + "issue:660", + "issue:6790", + "issue:7054", + "issue:7454", + "issue:7629", + "issue:8560", + "issue:861", + "issue:8613", + "issue:9216", + "issue:9264" + ], + "soft_pairs": [ + "issue:12783|issue:13105", + "issue:3915|issue:660", + "issue:2695|issue:3467", + "issue:3689|issue:7454", + "issue:5574|issue:5575", + "issue:1931|issue:2440", + "issue:2266|issue:460", + "issue:3045|issue:3139", + "issue:6790|issue:7054", + "issue:1062|issue:4331", + "issue:10382|issue:12110", + "issue:10695|issue:12633", + "issue:1566|issue:3647", + "issue:2984|issue:3087", + "issue:3944|issue:4912", + "issue:9216|issue:9264", + "issue:2549|issue:4692", + "issue:1672|issue:861", + "issue:11223|issue:7629", + "issue:1463|issue:5402", + "issue:8560|issue:8613", + "issue:10672|issue:11569", + "issue:1463|issue:384", + "issue:5000|issue:5518" + ], + "prepared_review_unit_hash": "dcc6f1f8f5d4dd7ca2f204e6f07be128d40fea0c53d536ed09c54333fa655885", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33220, + "estimated_input_tokens": 8305, + "estimated_eval_tokens": 16866 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are mostly weak title-level similarities across different models, schedulers, or training features. I don\u2019t see any pair that is safely the same underlying bug/change, so all soft edges should be rejected.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:12633 is the clearest standalone bug report here: it names a specific scheduler path, a precise mismatch condition (`timestep_spacing='linspace'`), and should be easy to triage.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12783", + "right": "issue:13105", + "accept": false, + "reason": "Both mention noisy/confetti output, but one is about Sage Attention sm90 and the other about a Wan model outputting noise; different failure contexts and likely different fixes." + }, + { + "left": "issue:3915", + "right": "issue:660", + "accept": false, + "reason": "`torch.compile` on text-to-video pipelines is unrelated to MPS fp16 image generation; shared wording is only about PyTorch/runtime settings." + }, + { + "left": "issue:2695", + "right": "issue:3467", + "accept": false, + "reason": "Both concern training workflows and validation/resume behavior, but one is Tensorboard validation failure in ControlNet training and the other is wrong resume step with gradient accumulation." + }, + { + "left": "issue:3689", + "right": "issue:7454", + "accept": false, + "reason": "Both are LoRA-related, but one asks how to unload weights after loading and the other is a generic feature request for LoRA support; not the same bug/change." + }, + { + "left": "issue:5574", + "right": "issue:5575", + "accept": false, + "reason": "Both are LoRA training questions, but one is about training part of UNet attention parameters and the other about hidden size for `transformer_in`; distinct configuration issues." + }, + { + "left": "issue:1931", + "right": "issue:2440", + "accept": false, + "reason": "Both involve float16-related failures, but one is a scalar type mismatch in Stable Diffusion 2 and the other is an inpainting model load error in float16; different code paths." + }, + { + "left": "issue:2266", + "right": "issue:460", + "accept": false, + "reason": "Both relate to performance/runtime on specific hardware setups, but one is sequential CPU offload slowdown and the other is ONNX runtime backend selection; not the same issue." + }, + { + "left": "issue:3045", + "right": "issue:3139", + "accept": false, + "reason": "One is about converting LoRA safetensors to diffusers, the other about using a safetensors file for generation; adjacent topics but different user needs." + }, + { + "left": "issue:6790", + "right": "issue:7054", + "accept": false, + "reason": "Both mention LoRA/LCM training state handling, but one is a missing key in a training state_dict and the other is input state_dict being emptied by `load_lora_weights`; different bugs." + }, + { + "left": "issue:1062", + "right": "issue:4331", + "accept": false, + "reason": "Both reference DreamBooth results, but one is a general inability to reproduce paper results and the other is a mismatch between two training scripts; not the same defect." + }, + { + "left": "issue:10382", + "right": "issue:12110", + "accept": false, + "reason": "One reports a VAE encoder return-type mismatch causing AttributeError; the other is a UNet1DModel GaussianFourier initialization bug. Different modules and failures." + }, + { + "left": "issue:10695", + "right": "issue:12633", + "accept": false, + "reason": "Both concern DDIM scheduler timestep handling, but one is DDIMInverseScheduler previous timestep calculation and the other is DDIMScheduler spacing mismatch; related area, not the same bug." + }, + { + "left": "issue:1566", + "right": "issue:3647", + "accept": false, + "reason": "Both touch DreamBooth training, but one crashes after saving with fp16 output and the other is training with DeepFloyd; different scenarios and failure modes." + }, + { + "left": "issue:2984", + "right": "issue:3087", + "accept": false, + "reason": "Latent Video Diffusion Models and LyCORIS support are separate feature requests with no shared concrete bug or code path." + }, + { + "left": "issue:3944", + "right": "issue:4912", + "accept": false, + "reason": "Both ask for more LoRA/customization support, but one is network rank customization in Dreambooth and the other is converting custom SDXL checkpoints; different requests." + }, + { + "left": "issue:9216", + "right": "issue:9264", + "accept": false, + "reason": "A general multi-model pipeline idea and a Flux inpainting feature request are unrelated feature requests." + }, + { + "left": "issue:2549", + "right": "issue:4692", + "accept": false, + "reason": "Both mention UNet2DConditionModel, but one says the model doesn't work for SD 2.0 and the other asks about memory usage; not the same bug." + }, + { + "left": "issue:1672", + "right": "issue:861", + "accept": false, + "reason": "Both are conversion-related, but one reports wrong image size after ckpt-to-diffusers conversion and the other cannot convert a ckpt model at all; different failure points." + }, + { + "left": "issue:11223", + "right": "issue:7629", + "accept": false, + "reason": "Both are OOM reports, but one is layer-wise upcasting with WAN and LoRAs, while the other is SVD on PyTorch 1.13.1 versus 2.x; different root causes." + }, + { + "left": "issue:1463", + "right": "issue:5402", + "accept": false, + "reason": "One is a missing `datasets` dependency, the other is an import error for `compute_snr`; both are import issues but unrelated packages/functions." + }, + { + "left": "issue:8560", + "right": "issue:8613", + "accept": false, + "reason": "Both are OOM reports, but one is SD3 training on a 4090 and the other is caption-column OOM during dataset processing; different stages and triggers." + }, + { + "left": "issue:10672", + "right": "issue:11569", + "accept": false, + "reason": "A pipeline callback feature request and a repo request for a specific model are unrelated enhancement requests." + }, + { + "left": "issue:1463", + "right": "issue:384", + "accept": false, + "reason": "Both are missing-module style errors, but one is absent `datasets` and the other is a `DiagonalGaussianDistribution` attribute error; different problems." + }, + { + "left": "issue:5000", + "right": "issue:5518", + "accept": false, + "reason": "Both involve LoRA-related regressions, but one is unfusing after fuse/unfuse and the other is a `ModuleList` missing `time`; not the same code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 212, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10147", + "issue:10887", + "issue:10893", + "issue:10987", + "issue:11010", + "issue:11374", + "issue:12149", + "issue:12575", + "issue:12666", + "issue:13067", + "issue:1909", + "issue:2806", + "issue:319", + "issue:322", + "issue:337", + "issue:4390", + "issue:4609", + "issue:5166", + "issue:5371", + "issue:5593", + "issue:5605", + "issue:5931", + "issue:5966", + "issue:6073", + "issue:6287", + "issue:6412", + "issue:6545", + "issue:6586", + "issue:6685", + "issue:7223", + "issue:7243", + "issue:7312", + "issue:7417", + "issue:7428", + "issue:8482", + "issue:8543", + "issue:8565", + "issue:8880", + "issue:8918", + "issue:8994", + "issue:917", + "issue:9216", + "issue:9258", + "issue:9362", + "issue:938", + "issue:9393", + "issue:9665", + "issue:9732" + ], + "soft_pairs": [ + "issue:8565|issue:9362", + "issue:7243|issue:9216", + "issue:337|issue:917", + "issue:5593|issue:7417", + "issue:11010|issue:12575", + "issue:1909|issue:2806", + "issue:10147|issue:7223", + "issue:10887|issue:11374", + "issue:10987|issue:8482", + "issue:4390|issue:5166", + "issue:5371|issue:7428", + "issue:6073|issue:9732", + "issue:6287|issue:8918", + "issue:6685|issue:8543", + "issue:8880|issue:9258", + "issue:6545|issue:6586", + "issue:319|issue:322", + "issue:5605|issue:7312", + "issue:9393|issue:9665", + "issue:5931|issue:8994", + "issue:10893|issue:13067", + "issue:12149|issue:12666", + "issue:4609|issue:938", + "issue:5966|issue:6412" + ], + "prepared_review_unit_hash": "0390b8ccd35da16fb178492bab8f56a29b5aa3b628966d6a6cfed5629a1b870f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33304, + "estimated_input_tokens": 8326, + "estimated_eval_tokens": 16908 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very heterogeneous: it mixes unrelated feature requests, model-support asks, training bugs, and API issues. None of the soft pairs look like the same underlying bug/change, so all should be rejected. The closest representative issue is the broad multi-view generation feature request, but even that is only a loose thematic fit, not a true duplicate anchor.", + "confidence": 0.31, + "canonical_issue_reason": "Issue 10147 is the broadest open feature request in the set and best matches the general 'add model support' theme, though the cluster is not a true duplicate group.", + "canonical_pr_reason": null, + "best_issue_reason": "10147 is the most generic, representative open issue among mostly model-support requests; it can serve as a loose anchor even though the set is not a coherent duplicate cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8565", + "right": "issue:9362", + "accept": false, + "reason": "Both mention IndexError, but they refer to different pipelines and failure modes; not the same bug." + }, + { + "left": "issue:7243", + "right": "issue:9216", + "accept": false, + "reason": "Both are about new pipeline/model composition, but they are different feature requests and code changes." + }, + { + "left": "issue:337", + "right": "issue:917", + "accept": false, + "reason": "Both concern inpainting, but one is an img2img/text-guided bug and the other is an ONNX inpainting error; not the same issue." + }, + { + "left": "issue:5593", + "right": "issue:7417", + "accept": false, + "reason": "ControlNet AutoPipeline failure and a Mac LoRA-fuse test failure are unrelated code paths." + }, + { + "left": "issue:11010", + "right": "issue:12575", + "accept": false, + "reason": "A Flux/Chroma model-support request is not the same as an attention-backend check error." + }, + { + "left": "issue:1909", + "right": "issue:2806", + "accept": false, + "reason": "Generator handling in schedulers and pipeline printing/customization are different API problems." + }, + { + "left": "issue:10147", + "right": "issue:7223", + "accept": false, + "reason": "Both are feature requests for new generation capabilities, but MV-Adapter multi-view support and Latte video support are distinct model integrations." + }, + { + "left": "issue:10887", + "right": "issue:11374", + "accept": false, + "reason": "Importing StableDiffusionXLTrainer and CogVideoX ControlNet integration are unrelated." + }, + { + "left": "issue:10987", + "right": "issue:8482", + "accept": false, + "reason": "A general spatio-temporal model request and SD3 model support are different features." + }, + { + "left": "issue:4390", + "right": "issue:5166", + "accept": false, + "reason": "Missing requirements and a missing VQVAE scaling factor are different bugs." + }, + { + "left": "issue:5371", + "right": "issue:7428", + "accept": false, + "reason": "BLIP inpainting support and layer diffusion are different feature requests." + }, + { + "left": "issue:6073", + "right": "issue:9732", + "accept": false, + "reason": "Both involve DreamBooth guides, but one is a beginner setup error and the other is CUDA OOM during Flux training." + }, + { + "left": "issue:6287", + "right": "issue:8918", + "accept": false, + "reason": "Setup dependency cleanup and a broken SD3 dreambooth script are unrelated." + }, + { + "left": "issue:6685", + "right": "issue:8543", + "accept": false, + "reason": "Absolute output_dir path handling and SD3 gradient checkpointing are different bugs." + }, + { + "left": "issue:8880", + "right": "issue:9258", + "accept": false, + "reason": "Both are performance complaints, but they concern different workloads and environments." + }, + { + "left": "issue:6545", + "right": "issue:6586", + "accept": false, + "reason": "A training/inference modularization tracker and SDXL micro-conditioning tracker are different changes." + }, + { + "left": "issue:319", + "right": "issue:322", + "accept": false, + "reason": "Both are early diffusion model requests, but VQ-Diffusion and V-Diffusion are distinct model families." + }, + { + "left": "issue:5605", + "right": "issue:7312", + "accept": false, + "reason": "Per-image caption support for DreamBooth and AlignProp support are different training features." + }, + { + "left": "issue:9393", + "right": "issue:9665", + "accept": false, + "reason": "Both involve DeepSpeed, but one is a save-model failure and the other is accelerator.prepare() behavior." + }, + { + "left": "issue:5931", + "right": "issue:8994", + "accept": false, + "reason": "Nondeterministic generation and generation steps stuck at 0 are different bugs." + }, + { + "left": "issue:10893", + "right": "issue:13067", + "accept": false, + "reason": "CogVideoX ControlNet integration and Anima support are separate model feature requests." + }, + { + "left": "issue:12149", + "right": "issue:12666", + "accept": false, + "reason": "Both mention release mismatch or missing pipelines, but they are different versioning/support problems." + }, + { + "left": "issue:4609", + "right": "issue:938", + "accept": false, + "reason": "A specific T2I-Adapter model request and a human motion diffusion model request are unrelated." + }, + { + "left": "issue:5966", + "right": "issue:6412", + "accept": false, + "reason": "ControlNet for SD2.1 and ControlNet annotators to ONNX are different parts of the stack and different changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 213, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10003", + "issue:10147", + "issue:10542", + "issue:10639", + "issue:1083", + "issue:11093", + "issue:11109", + "issue:11223", + "issue:11229", + "issue:11315", + "issue:11592", + "issue:12009", + "issue:12396", + "issue:12567", + "issue:12887", + "issue:13286", + "issue:1539", + "issue:2177", + "issue:2830", + "issue:3097", + "issue:3171", + "issue:3379", + "issue:3537", + "issue:3799", + "issue:4376", + "issue:4949", + "issue:4969", + "issue:5455", + "issue:5510", + "issue:5576", + "issue:5689", + "issue:6186", + "issue:656", + "issue:7368", + "issue:7535", + "issue:7606", + "issue:8482", + "issue:8504", + "issue:8641", + "issue:8667", + "issue:8790", + "issue:9069", + "issue:9166", + "issue:9225", + "issue:9298", + "issue:9362", + "issue:9370", + "issue:9496" + ], + "soft_pairs": [ + "issue:12009|issue:12887", + "issue:11592|issue:12396", + "issue:11109|issue:8667", + "issue:11592|issue:13286", + "issue:3097|issue:4376", + "issue:3537|issue:7368", + "issue:3799|issue:7535", + "issue:5576|issue:6186", + "issue:11223|issue:12567", + "issue:10003|issue:9362", + "issue:2830|issue:4949", + "issue:3171|issue:3537", + "issue:8482|issue:8504", + "issue:2177|issue:3379", + "issue:9370|issue:9496", + "issue:11229|issue:11315", + "issue:1083|issue:1539", + "issue:4969|issue:656", + "issue:5455|issue:7606", + "issue:9069|issue:9298", + "issue:5689|issue:9166", + "issue:10542|issue:11093", + "issue:5510|issue:9225", + "issue:10147|issue:8790", + "issue:10639|issue:8641" + ], + "prepared_review_unit_hash": "815f57ac2a638473d81dbe5c930554ca5953b3704818ed92ed88c9189d5a1fbe", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33615, + "estimated_input_tokens": 8404, + "estimated_eval_tokens": 17064 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All proposed soft edges are superficial title-level similarities rather than the same underlying bug or feature change. The set mixes unrelated models, pipelines, export/type issues, and feature requests, so every candidate pair is rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No true canonical issue stands out; the cluster is heterogeneous and the soft matches only share broad wording.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11229 is the best standalone representative because it is a concrete, reproducible bug report with clear SDXL/MPS context and actionable symptoms.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12009", + "right": "issue:12887", + "accept": false, + "reason": "Both involve from_single_file loading, but they target different models and failure modes (Wan2.2 GGUF meta-tensor copy error vs z-image-turbo load errors)." + }, + { + "left": "issue:11592", + "right": "issue:12396", + "accept": false, + "reason": "Both mention delete_adapters, but 11592 is a FluxLoraLoaderMixin regression while 12396 is a separate leaf_level + delete_adapters error path." + }, + { + "left": "issue:11109", + "right": "issue:8667", + "accept": false, + "reason": "Different bugs: depth LoRA transformer-layer mismatch vs SD3ControlNet image-size mismatch." + }, + { + "left": "issue:11592", + "right": "issue:13286", + "accept": false, + "reason": "Unrelated problems: adapter deletion regression vs unsupported torchao quantization version." + }, + { + "left": "issue:3097", + "right": "issue:4376", + "accept": false, + "reason": "Both touch textual inversion, but one is a no-op report and the other is a pipeline crash in SDXL; different concrete failures." + }, + { + "left": "issue:3537", + "right": "issue:7368", + "accept": false, + "reason": "A controlnet conversion script request and GLIGEN training support are different changes with no shared bug." + }, + { + "left": "issue:3799", + "right": "issue:7535", + "accept": false, + "reason": "Both are typing-related, but one is scheduler type inference and the other is a FloatTensor type-hint incompatibility." + }, + { + "left": "issue:5576", + "right": "issue:6186", + "accept": false, + "reason": "A new scheduler request and i2vgen-xl support are unrelated feature requests." + }, + { + "left": "issue:11223", + "right": "issue:12567", + "accept": false, + "reason": "Different symptoms and code paths: WAN LoRA OOM with layer-wise upcasting vs noise from Ulysses Attention." + }, + { + "left": "issue:10003", + "right": "issue:9362", + "accept": false, + "reason": "Scheduler bugs at 1000 steps and an index-out-of-bounds error are not the same underlying issue." + }, + { + "left": "issue:2830", + "right": "issue:4949", + "accept": false, + "reason": "Both are video-related questions, but one asks about image/text-to-video with ModelScope and the other is about ProPainter in/outpainting." + }, + { + "left": "issue:3171", + "right": "issue:3537", + "accept": false, + "reason": "LPW prompt support for ControlNet and a ControlNet conversion script are different requests." + }, + { + "left": "issue:8482", + "right": "issue:8504", + "accept": false, + "reason": "SD3-medium core support and choosing text encoders for SD3-medium are related only broadly; they are not the same concrete change." + }, + { + "left": "issue:2177", + "right": "issue:3379", + "accept": false, + "reason": "Different loading failures: pretrained AutoencoderKL vs Dreambooth intermediate checkpoint loading." + }, + { + "left": "issue:9370", + "right": "issue:9496", + "accept": false, + "reason": "Both are SD3 forward-path errors, but one is a missing pooled_projection argument and the other is a different ControlNet forward failure." + }, + { + "left": "issue:11229", + "right": "issue:11315", + "accept": false, + "reason": "Both mention MPS, but NaNs from attention slicing and float64 rope breaking on MPS have different causes and fixes." + }, + { + "left": "issue:1083", + "right": "issue:1539", + "accept": false, + "reason": "An FP16 ONNX correctness issue and a conversion-script type-int bug are unrelated." + }, + { + "left": "issue:4969", + "right": "issue:656", + "accept": false, + "reason": "CrossAttention import failure and a Dreambooth training error on 16GB GPUs are different issues." + }, + { + "left": "issue:5455", + "right": "issue:7606", + "accept": false, + "reason": "A subprocess failure in a notebook and a deprecated LoRA processor warning are not the same bug." + }, + { + "left": "issue:9069", + "right": "issue:9298", + "accept": false, + "reason": "These are different type/storage problems: ndarray type mismatch vs missing Float8 storage object." + }, + { + "left": "issue:5689", + "right": "issue:9166", + "accept": false, + "reason": "Residual noise in DPM++ output and unexpected apply_overlay behavior are separate issues in different subsystems." + }, + { + "left": "issue:10542", + "right": "issue:11093", + "accept": false, + "reason": "Hunyuan Video batch-size breakage and a broken TorchAO example are unrelated." + }, + { + "left": "issue:5510", + "right": "issue:9225", + "accept": false, + "reason": "Half/float bias mismatch in SDXL InstructPix2Pix and RGBA latent size mismatch are different pipeline bugs." + }, + { + "left": "issue:10147", + "right": "issue:8790", + "accept": false, + "reason": "A multi-view generation feature request and SDXL TensorRT + IP Adapter support are unrelated requests." + }, + { + "left": "issue:10639", + "right": "issue:8641", + "accept": false, + "reason": "Support for Lumina-Image 2.0 and integrating Lumina-T2X are different model additions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 214, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10165", + "issue:10202", + "issue:10428", + "issue:1101", + "issue:11088", + "issue:11109", + "issue:11283", + "issue:11412", + "issue:11536", + "issue:11540", + "issue:11709", + "issue:11779", + "issue:1185", + "issue:12071", + "issue:12110", + "issue:12653", + "issue:13477", + "issue:1672", + "issue:1677", + "issue:1934", + "issue:2210", + "issue:2392", + "issue:3073", + "issue:3441", + "issue:3944", + "issue:4005", + "issue:4210", + "issue:4331", + "issue:4349", + "issue:4631", + "issue:4871", + "issue:5002", + "issue:5171", + "issue:5661", + "issue:6242", + "issue:6423", + "issue:6446", + "issue:7219", + "issue:725", + "issue:8320", + "issue:8885", + "issue:9026", + "issue:9293", + "issue:9424", + "issue:9516", + "issue:9902", + "issue:9930" + ], + "soft_pairs": [ + "issue:11536|issue:11540", + "issue:4210|issue:5661", + "issue:6446|issue:8885", + "issue:1101|issue:725", + "issue:10202|issue:11109", + "issue:10010|issue:11412", + "issue:6242|issue:8320", + "issue:9424|issue:9902", + "issue:11088|issue:11709", + "issue:12110|issue:13477", + "issue:12071|issue:9516", + "issue:2210|issue:3944", + "issue:1672|issue:1677", + "issue:1934|issue:3441", + "issue:2392|issue:4349", + "issue:10165|issue:9930", + "issue:3073|issue:7219", + "issue:4005|issue:6423", + "issue:11283|issue:11779", + "issue:10428|issue:12653", + "issue:4871|issue:5171", + "issue:9026|issue:9293", + "issue:4631|issue:5002", + "issue:1185|issue:4331" + ], + "prepared_review_unit_hash": "a010149ec0c445d12d31f1ffff3d6ea2b3deb49df48f777c6c1a016347182efa", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34022, + "estimated_input_tokens": 8506, + "estimated_eval_tokens": 17268 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is heterogeneous: Flux, Wan, LoRA, scheduler, loading, and training issues are mixed together. None of the soft pairs look like true duplicates of the same underlying bug/change, so all should stay separate.", + "confidence": 0.22, + "canonical_issue_reason": "issue:11536 is the clearest broad, user-facing bug report in the dominant Flux theme, with concrete behavior and discussion; it is more representative than the narrower Flux/loading variants.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11536 is the best single issue to anchor this cluster because it is specific enough to be actionable yet broad enough to represent the main Flux-related regression pattern seen here.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11536", + "right": "issue:11540", + "accept": false, + "reason": "Both are Flux-related, but one is prompt adherence regression and the other is VAE slicing/noise behavior; different code paths and symptoms." + }, + { + "left": "issue:4210", + "right": "issue:5661", + "accept": false, + "reason": "UNet attention_mask usage question is unrelated to the missing PreTrainedTokenizer import error." + }, + { + "left": "issue:6446", + "right": "issue:8885", + "accept": false, + "reason": "Stable Diffusion Colab loading problem is unrelated to clip-retrieval install/modifier-token issues." + }, + { + "left": "issue:1101", + "right": "issue:725", + "accept": false, + "reason": "Scheduler addition request and memory-footprint/loading infrastructure request are different features." + }, + { + "left": "issue:10202", + "right": "issue:11109", + "accept": false, + "reason": "Flux Control LoRA unloading bug is not the same as depth LoRA transformer layer mismatch." + }, + { + "left": "issue:10010", + "right": "issue:11412", + "accept": false, + "reason": "Both mention 404s, but they point to different model repos/examples and likely different missing artifacts." + }, + { + "left": "issue:6242", + "right": "issue:8320", + "accept": false, + "reason": "Dataset log request is unrelated to the from_single_file AutoencoderKL attribute error." + }, + { + "left": "issue:9424", + "right": "issue:9902", + "accept": false, + "reason": "Notebook breakage after updates is not the same as Conda version pinning/stuck-version behavior." + }, + { + "left": "issue:11088", + "right": "issue:11709", + "accept": false, + "reason": "Wan2.1 GGUF deserialization/load error is different from CausVid LoRA key loading failure." + }, + { + "left": "issue:12110", + "right": "issue:13477", + "accept": false, + "reason": "UNet1DModel initialization bug and ErnieImage callback scope bug are unrelated code paths." + }, + { + "left": "issue:12071", + "right": "issue:9516", + "accept": false, + "reason": "Both are FLUX/transformer-adjacent, but one is changed forward behavior and the other is missing joint_attention_kwargs propagation." + }, + { + "left": "issue:2210", + "right": "issue:3944", + "accept": false, + "reason": "Training text encoder with LoRA is a feature request, not the same as choosing network rank for Dreambooth LoRA training." + }, + { + "left": "issue:1672", + "right": "issue:1677", + "accept": false, + "reason": "Wrong output size after ckpt conversion is unrelated to pipeline call data type validation." + }, + { + "left": "issue:1934", + "right": "issue:3441", + "accept": false, + "reason": "Both involve memory usage, but one is cpu_offload VRAM blowup and the other is PyTorch2/xformers VRAM regression." + }, + { + "left": "issue:2392", + "right": "issue:4349", + "accept": false, + "reason": "Both involve prompt_embeds-related validation, but they affect different pipelines and failure modes." + }, + { + "left": "issue:10165", + "right": "issue:9930", + "accept": false, + "reason": "LoRA scale breaking seed preservation is not the same as PAG adaptive-scale behavior." + }, + { + "left": "issue:3073", + "right": "issue:7219", + "accept": false, + "reason": "Requesting Consistency Models and adding a SUPIR upscaler are different feature requests." + }, + { + "left": "issue:4005", + "right": "issue:6423", + "accept": false, + "reason": "Inpaint aspect-ratio bug is unrelated to inability to reproduce fill50k results." + }, + { + "left": "issue:11283", + "right": "issue:11779", + "accept": false, + "reason": "Wan header deserialization failure is a loading/format issue, not the VACE mask-padding bug." + }, + { + "left": "issue:10428", + "right": "issue:12653", + "accept": false, + "reason": "Ascend NPU Flux inference error and PRX rope float64/MPS-NPU issue are different underlying device/dtype problems." + }, + { + "left": "issue:4871", + "right": "issue:5171", + "accept": false, + "reason": "How to run from_single_file is unrelated to the unexpected safety_checker keyword error." + }, + { + "left": "issue:9026", + "right": "issue:9293", + "accept": false, + "reason": "Unexpected force_hook argument and missing Float8 storage object are separate loading/runtime failures." + }, + { + "left": "issue:4631", + "right": "issue:5002", + "accept": false, + "reason": "Scheduler scalar-conversion crash with Karras sigmas is distinct from missing use_karras_sigmas support." + }, + { + "left": "issue:1185", + "right": "issue:4331", + "accept": false, + "reason": "Dreambooth reproducibility complaints are related at a high level, but these reference different training scripts and causes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 215, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10057", + "issue:10302", + "issue:10542", + "issue:10590", + "issue:10672", + "issue:10873", + "issue:11137", + "issue:11275", + "issue:1158", + "issue:11584", + "issue:12239", + "issue:12458", + "issue:1263", + "issue:12695", + "issue:13176", + "issue:13191", + "issue:13277", + "issue:13506", + "issue:1524", + "issue:305", + "issue:3537", + "issue:3884", + "issue:4002", + "issue:4005", + "issue:4061", + "issue:4354", + "issue:5328", + "issue:5371", + "issue:5390", + "issue:5593", + "issue:6055", + "issue:6079", + "issue:6132", + "issue:6141", + "issue:6331", + "issue:7210", + "issue:7219", + "issue:7389", + "issue:7569", + "issue:7751", + "issue:8323", + "issue:8594", + "issue:8850", + "issue:8939", + "issue:8989", + "issue:9111", + "issue:9731", + "issue:9768" + ], + "soft_pairs": [ + "issue:5593|issue:7569", + "issue:9731|issue:9768", + "issue:6132|issue:7219", + "issue:10057|issue:10873", + "issue:11275|issue:11584", + "issue:1158|issue:1524", + "issue:10302|issue:8989", + "issue:6079|issue:7389", + "issue:7751|issue:8939", + "issue:10590|issue:11137", + "issue:12239|issue:13191", + "issue:1263|issue:305", + "issue:3537|issue:6141", + "issue:5371|issue:6331", + "issue:8323|issue:9111", + "issue:10542|issue:8850", + "issue:12458|issue:13277", + "issue:6055|issue:7210", + "issue:4061|issue:5328", + "issue:13176|issue:13506", + "issue:3884|issue:8594", + "issue:4002|issue:4005", + "issue:4354|issue:5390", + "issue:10672|issue:12695" + ], + "prepared_review_unit_hash": "a3c70d113fe58593ad76fdfb5978f786ebe5cbf4e3415a6822b3cfc91b6f94e2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33656, + "estimated_input_tokens": 8414, + "estimated_eval_tokens": 17084 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicate cluster here: these are mostly unrelated issues across different pipelines, features, and failure modes. All soft edges look like false positives.", + "confidence": 0.99, + "canonical_issue_reason": "No issue is a good canonical representative: the items span unrelated bugs/features across different pipelines, scripts, and backends.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue stands out as a global best match; the cluster is heterogeneous and the soft pairs are all false positives.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5593", + "right": "issue:7569", + "accept": false, + "reason": "Different problems: AutoPipeline ControlNet handling vs a Mac FreeU test failure." + }, + { + "left": "issue:9731", + "right": "issue:9768", + "accept": false, + "reason": "Unrelated: Playground2.5 LoRA training guidance vs a CuDNN SDPA NaN backend bug." + }, + { + "left": "issue:6132", + "right": "issue:7219", + "accept": false, + "reason": "Different scope: serialization LOADABLE_CLASSES vs adding a new upscaler model." + }, + { + "left": "issue:10057", + "right": "issue:10873", + "accept": false, + "reason": "Different pipelines and symptoms: Flux from_single_file token issue vs Sana pipeline issue." + }, + { + "left": "issue:11275", + "right": "issue:11584", + "accept": false, + "reason": "VAE NoneType addition bug vs torch.compile/group offloading incompatibility." + }, + { + "left": "issue:1158", + "right": "issue:1524", + "accept": false, + "reason": "Test interference in config vs poor textual inversion training quality." + }, + { + "left": "issue:10302", + "right": "issue:8989", + "accept": false, + "reason": "FP8 inference noise vs sequential offloading bug on Stable Audio." + }, + { + "left": "issue:6079", + "right": "issue:7389", + "accept": false, + "reason": "Generic dtype mismatch vs a from_single_file loading error; not the same bug." + }, + { + "left": "issue:7751", + "right": "issue:8939", + "accept": false, + "reason": "Animatediff conversion errors vs SD3 tokenizer loading failure." + }, + { + "left": "issue:10590", + "right": "issue:11137", + "accept": false, + "reason": "Different pipeline failures: Sana BF16 tiled VAE decode vs Latte dtype mismatch." + }, + { + "left": "issue:12239", + "right": "issue:13191", + "accept": false, + "reason": "Support requests for different models; not the same concrete change." + }, + { + "left": "issue:1263", + "right": "issue:305", + "accept": false, + "reason": "Both are import errors, but different modules and root causes." + }, + { + "left": "issue:3537", + "right": "issue:6141", + "accept": false, + "reason": "ControlNet conversion script vs adding ControlNet parameters; unrelated changes." + }, + { + "left": "issue:5371", + "right": "issue:6331", + "accept": false, + "reason": "Different inpainting feature requests: BLIP support vs SAG for inpainting." + }, + { + "left": "issue:8323", + "right": "issue:9111", + "accept": false, + "reason": "Scheduler file recognition vs missing scheduler attribute; different failures." + }, + { + "left": "issue:10542", + "right": "issue:8850", + "accept": false, + "reason": "Hunyuan batch-size regression vs community pipeline mirror breakage." + }, + { + "left": "issue:12458", + "right": "issue:13277", + "accept": false, + "reason": "Both mention batch inputs, but one is a feature/request and the other a QwenImagePipeline failure." + }, + { + "left": "issue:6055", + "right": "issue:7210", + "accept": false, + "reason": "Too broad: generic accelerate error vs a specific named_parameters dispatch issue." + }, + { + "left": "issue:4061", + "right": "issue:5328", + "accept": false, + "reason": "Different LoRA complaints on different models; no clear shared code-path." + }, + { + "left": "issue:13176", + "right": "issue:13506", + "accept": false, + "reason": "Different attention backends and different error contexts; not one bug." + }, + { + "left": "issue:3884", + "right": "issue:8594", + "accept": false, + "reason": "Different training scripts and failure sites; not mergeable as one fix." + }, + { + "left": "issue:4002", + "right": "issue:4005", + "accept": false, + "reason": "One is missing from_single_file support, the other is aspect-ratio behavior." + }, + { + "left": "issue:4354", + "right": "issue:5390", + "accept": false, + "reason": "Different SDXL issues: wrong pipeline type vs unwanted text-encoder download." + }, + { + "left": "issue:10672", + "right": "issue:12695", + "accept": false, + "reason": "Callback API support request vs Hunyuan v1.5 model support; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 216, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10111", + "issue:10302", + "issue:10315", + "issue:10591", + "issue:10674", + "issue:11033", + "issue:12022", + "issue:12085", + "issue:12359", + "issue:12575", + "issue:12605", + "issue:13104", + "issue:2364", + "issue:2392", + "issue:2737", + "issue:2769", + "issue:3064", + "issue:3580", + "issue:3585", + "issue:4014", + "issue:4043", + "issue:4122", + "issue:4453", + "issue:4496", + "issue:4646", + "issue:5028", + "issue:5079", + "issue:5233", + "issue:5433", + "issue:5465", + "issue:5540", + "issue:5951", + "issue:6486", + "issue:6497", + "issue:7056", + "issue:7408", + "issue:843", + "issue:8535", + "issue:8593", + "issue:8604", + "issue:8708", + "issue:9001", + "issue:9063", + "issue:9163", + "issue:9170", + "issue:9514", + "issue:9904", + "issue:9936" + ], + "soft_pairs": [ + "issue:12359|issue:12575", + "issue:2737|issue:5079", + "issue:3064|issue:3585", + "issue:4496|issue:843", + "issue:4646|issue:8604", + "issue:5465|issue:9001", + "issue:12605|issue:9904", + "issue:10111|issue:10591", + "issue:10302|issue:9514", + "issue:5540|issue:9170", + "issue:6486|issue:8708", + "issue:8535|issue:8593", + "issue:10674|issue:12085", + "issue:4122|issue:5233", + "issue:9063|issue:9904", + "issue:2364|issue:5951", + "issue:9163|issue:9936", + "issue:11033|issue:13104", + "issue:10315|issue:7056", + "issue:12022|issue:12605", + "issue:4014|issue:5433", + "issue:2769|issue:4453", + "issue:3580|issue:5028", + "issue:2392|issue:4043", + "issue:6497|issue:7408" + ], + "prepared_review_unit_hash": "8451a17120db01b1a01760c9a461f275faff27aa1c85b6e8480accceac0a67a2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33922, + "estimated_input_tokens": 8481, + "estimated_eval_tokens": 17218 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are mostly broad thematic matches at best, but not true duplicates: they span different models, pipelines, training scripts, and failure modes. I would reject all soft edges and treat this as a heterogeneous set rather than one dedupe cluster.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:10674 is the closest representative only because it is a concrete FluxPipeline runtime/loading bug, but the overall set is too heterogeneous to have a real canonical issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12359", + "right": "issue:12575", + "accept": false, + "reason": "One is a Chroma docs bug about guidance_scale; the other is a Chroma attention-backend runtime error. Different problem and fix." + }, + { + "left": "issue:2737", + "right": "issue:5079", + "accept": false, + "reason": "Scheduler CLIP guidance support and LoRA loading support are unrelated features/issues." + }, + { + "left": "issue:3064", + "right": "issue:3585", + "accept": false, + "reason": "Both mention LoRA, but one is about loading .safetensors LoRAs and the other about a synthesized LoRA model failing. Different bug paths." + }, + { + "left": "issue:4496", + "right": "issue:843", + "accept": false, + "reason": "Prompt fusion feature request is unrelated to prompt syntax documentation." + }, + { + "left": "issue:4646", + "right": "issue:8604", + "accept": false, + "reason": "ControlNet ONNX conversion failure vs T5 fp16 corruption are different code paths and failure modes." + }, + { + "left": "issue:5465", + "right": "issue:9001", + "accept": false, + "reason": "DiffEdit inpaint_strength behavior and Auraflow aspect ratio handling are unrelated." + }, + { + "left": "issue:12605", + "right": "issue:9904", + "accept": false, + "reason": "Attention-mask backend bug and Flux image-shape rounding are distinct issues." + }, + { + "left": "issue:10111", + "right": "issue:10591", + "accept": false, + "reason": "Both are general bug reports, but FLUX pipeline bugs and SD3 LoRA training-script bugs are not the same underlying issue." + }, + { + "left": "issue:10302", + "right": "issue:9514", + "accept": false, + "reason": "FP8 inference noise without CPU offloading is not the same as LoRA loading after FP8 quantization." + }, + { + "left": "issue:5540", + "right": "issue:9170", + "accept": false, + "reason": "Both are memory-related, but generic OOM and a specific SDXL+ControlNet memory requirement report are not duplicates." + }, + { + "left": "issue:6486", + "right": "issue:8708", + "accept": false, + "reason": "Dreambooth embedding retraction bug and missing SD3 shift_factor in examples are unrelated training issues." + }, + { + "left": "issue:8535", + "right": "issue:8593", + "accept": false, + "reason": "xFormers attention enablement and bfloat16 CPU noise are different SD3 problems." + }, + { + "left": "issue:10674", + "right": "issue:12085", + "accept": false, + "reason": "Both involve GGUF loading, but they affect different model classes and likely different loader paths; too broad to merge as one bug." + }, + { + "left": "issue:4122", + "right": "issue:5233", + "accept": false, + "reason": "Timesteps mismatch in SDXL pipelines and callback step-value bug are related only at a very high level." + }, + { + "left": "issue:9063", + "right": "issue:9904", + "accept": false, + "reason": "Tokenizer instantiation failure and image-shape rounding are unrelated FluxPipeline issues." + }, + { + "left": "issue:2364", + "right": "issue:5951", + "accept": false, + "reason": "Unable to run training script is far too vague and does not match the PixArt training issue specifically." + }, + { + "left": "issue:9163", + "right": "issue:9936", + "accept": false, + "reason": "Both are Flux training issues, but tensor bugs and NCCL timeout in multigpu training are not the same concrete failure." + }, + { + "left": "issue:11033", + "right": "issue:13104", + "accept": false, + "reason": "SD1.5 from_single_file loading failure and a logger-not-defined error are unrelated." + }, + { + "left": "issue:10315", + "right": "issue:7056", + "accept": false, + "reason": "CogVideo training error and SDXL LoRA fine-tuning/train_text_encoder error are different models and code paths." + }, + { + "left": "issue:12022", + "right": "issue:12605", + "accept": false, + "reason": "Latest flash-attention interface incompatibility and attn_mask being ignored are separate attention backend bugs." + }, + { + "left": "issue:4014", + "right": "issue:5433", + "accept": false, + "reason": "Both mention image artifacts, but watermark artifacts and DPM++/no-refiner artifacts are different causes." + }, + { + "left": "issue:2769", + "right": "issue:4453", + "accept": false, + "reason": "Round-tripping ckpt/safetensors conversion and converting SDXL LoRA to safetensors are different workflows." + }, + { + "left": "issue:3580", + "right": "issue:5028", + "accept": false, + "reason": "XFormers TypeError and hacked DownBlock scale-arg TypeError are different runtime errors." + }, + { + "left": "issue:2392", + "right": "issue:4043", + "accept": false, + "reason": "Img2Img prompt_embeds validation and SDXL pooled_prompt_embeds requirement are related API areas but not the same bug." + }, + { + "left": "issue:6497", + "right": "issue:7408", + "accept": false, + "reason": "DragNUWA and FRESCO are unrelated pipeline/model issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 217, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10147", + "issue:1037", + "issue:10512", + "issue:11005", + "issue:1101", + "issue:11581", + "issue:12221", + "issue:12239", + "issue:13191", + "issue:1358", + "issue:1962", + "issue:2202", + "issue:2440", + "issue:2905", + "issue:3040", + "issue:3817", + "issue:4061", + "issue:4471", + "issue:4736", + "issue:4752", + "issue:5002", + "issue:5043", + "issue:5162", + "issue:5168", + "issue:5256", + "issue:5285", + "issue:5325", + "issue:539", + "issue:5515", + "issue:6094", + "issue:6486", + "issue:660", + "issue:6749", + "issue:6774", + "issue:6906", + "issue:7300", + "issue:7311", + "issue:7600", + "issue:7611", + "issue:7970", + "issue:8323", + "issue:8410", + "issue:8588", + "issue:8894", + "issue:9210", + "issue:9293", + "issue:9420" + ], + "soft_pairs": [ + "issue:1962|issue:3817", + "issue:2202|issue:5168", + "issue:6486|issue:7600", + "issue:7300|issue:7970", + "issue:4736|issue:9210", + "issue:1101|issue:5256", + "issue:4061|issue:4471", + "issue:5002|issue:539", + "issue:5515|issue:8323", + "issue:2440|issue:660", + "issue:3040|issue:6094", + "issue:4752|issue:5325", + "issue:11581|issue:9293", + "issue:7311|issue:8894", + "issue:6774|issue:7611", + "issue:10100|issue:8410", + "issue:10147|issue:12239", + "issue:1037|issue:2905", + "issue:10512|issue:9420", + "issue:11005|issue:8588", + "issue:12221|issue:13191", + "issue:1358|issue:5162", + "issue:5043|issue:5285", + "issue:6749|issue:6906" + ], + "prepared_review_unit_hash": "ce327b187f9c4d94f93e81a7f855bacf08eb825f789c9fcf21fa48edbfc9eeff", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33300, + "estimated_input_tokens": 8325, + "estimated_eval_tokens": 16906 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated feature requests, training/runtime bugs, and documentation/support items across many different pipelines and model families. All soft edges are false positives.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits the cluster because the items are semantically diverse and do not describe the same underlying bug or feature request.", + "canonical_pr_reason": null, + "best_issue_reason": "If one issue must stand in as a loose representative, issue 10147 is the broadest current open feature request, but it still does not cover the rest of the cluster well.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1962", + "right": "issue:3817", + "accept": false, + "reason": "Generic video generation with Stable Diffusion vs editable dance generation from music; different use cases and code paths." + }, + { + "left": "issue:2202", + "right": "issue:5168", + "accept": false, + "reason": "OpenVINO pipeline support and ControlNet-XS support are separate subsystem/model feature requests." + }, + { + "left": "issue:6486", + "right": "issue:7600", + "accept": false, + "reason": "Double embedding retraction in DreamBooth advanced vs not applying accelerator.accumulate in text_encoder; distinct training bugs." + }, + { + "left": "issue:7300", + "right": "issue:7970", + "accept": false, + "reason": "Both mention memory, but one is about set_default_attn_processor() behavior and the other is CPU offload RAM growth; different bugs." + }, + { + "left": "issue:4736", + "right": "issue:9210", + "accept": false, + "reason": "Train LoRA OOM/invalid-cast failure vs dataset image_column validation error; unrelated failure modes." + }, + { + "left": "issue:1101", + "right": "issue:5256", + "accept": false, + "reason": "Scheduler feature addition vs Insta-Flow training support; separate changes." + }, + { + "left": "issue:4061", + "right": "issue:4471", + "accept": false, + "reason": "Both are SDXL LoRA load problems, but they target different pipeline APIs and specific load failures." + }, + { + "left": "issue:5002", + "right": "issue:539", + "accept": false, + "reason": "Missing scheduler arguments vs snapshot_download allow_patterns keyword error; unrelated API compatibility issues." + }, + { + "left": "issue:5515", + "right": "issue:8323", + "accept": false, + "reason": "LCM pipeline initialization failure without scheduler vs scheduler file recognition error; different underlying problems." + }, + { + "left": "issue:2440", + "right": "issue:660", + "accept": false, + "reason": "Inpainting float16 load error vs MPS fp16 revision behavior; different environments and failure causes." + }, + { + "left": "issue:3040", + "right": "issue:6094", + "accept": false, + "reason": "Doc copy-paste AttributeError vs TextToVideoZero regression after a version update; unrelated issues." + }, + { + "left": "issue:4752", + "right": "issue:5325", + "accept": false, + "reason": "SDXL dtype mismatch in pipeline code vs packaging type information distribution; no shared bug." + }, + { + "left": "issue:11581", + "right": "issue:9293", + "accept": false, + "reason": "FLUX IP-Adapter dtype init error vs missing Float8 storage object; both dtype-related but different components and code paths." + }, + { + "left": "issue:7311", + "right": "issue:8894", + "accept": false, + "reason": "Checkpoint save failure in train_text_to_image_sdxl.py vs a separate training script error after a new commit." + }, + { + "left": "issue:6774", + "right": "issue:7611", + "accept": false, + "reason": "AttentionProcessor mixed-precision mismatch vs UNet tensor size mismatch; distinct runtime errors." + }, + { + "left": "issue:10100", + "right": "issue:8410", + "accept": false, + "reason": "ConsisID support and native TextDiffuser support are different model integrations." + }, + { + "left": "issue:10147", + "right": "issue:12239", + "accept": false, + "reason": "MV-Adapter multi-view generation vs InfiniteTalk support; separate feature requests." + }, + { + "left": "issue:1037", + "right": "issue:2905", + "accept": false, + "reason": "Community pipeline for Stable Diffusion Telephone vs Karras sigmas scheduler support; unrelated additions." + }, + { + "left": "issue:10512", + "right": "issue:9420", + "accept": false, + "reason": "Quanto Flux LoRA load failure vs transformer load failure after FLUX+Hyper-SD save; related to FLUX, but different concrete bugs." + }, + { + "left": "issue:11005", + "right": "issue:8588", + "accept": false, + "reason": "Docstring/default mismatch in pipeline_wan_i2v.py vs missing SD3 conversion script; unrelated." + }, + { + "left": "issue:12221", + "right": "issue:13191", + "accept": false, + "reason": "DiffSynth ControlNet support vs Elastic-DiT support; different model support requests." + }, + { + "left": "issue:1358", + "right": "issue:5162", + "accept": false, + "reason": "Safetensors support in the library vs a conversion script to safetensors; related topic, but not the same issue." + }, + { + "left": "issue:5043", + "right": "issue:5285", + "accept": false, + "reason": "W\u00fcrstchen LoRA finetuning vs Japanese docs translation; entirely different scopes." + }, + { + "left": "issue:6749", + "right": "issue:6906", + "accept": false, + "reason": "InstantID community example reproducibility vs broken LoRA DreamBooth example; different examples and failures." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 218, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10412", + "issue:10636", + "issue:10672", + "issue:11225", + "issue:11555", + "issue:11839", + "issue:12107", + "issue:1214", + "issue:12470", + "issue:12505", + "issue:13067", + "issue:1405", + "issue:1655", + "issue:1791", + "issue:1801", + "issue:1896", + "issue:1949", + "issue:2340", + "issue:2392", + "issue:2466", + "issue:2728", + "issue:2947", + "issue:3698", + "issue:3767", + "issue:3799", + "issue:4492", + "issue:4534", + "issue:457", + "issue:5368", + "issue:5395", + "issue:541", + "issue:5502", + "issue:5595", + "issue:6094", + "issue:6333", + "issue:6363", + "issue:6452", + "issue:646", + "issue:7606", + "issue:8373", + "issue:8482", + "issue:851", + "issue:8552", + "issue:8604", + "issue:8744", + "issue:8780", + "issue:8939", + "issue:9799" + ], + "soft_pairs": [ + "issue:5395|issue:6333", + "issue:1405|issue:2340", + "issue:8744|issue:8939", + "issue:10412|issue:11225", + "issue:1791|issue:457", + "issue:1896|issue:2392", + "issue:3799|issue:8373", + "issue:5595|issue:6363", + "issue:6094|issue:8552", + "issue:1655|issue:3767", + "issue:4492|issue:4534", + "issue:11839|issue:12107", + "issue:1949|issue:2466", + "issue:10672|issue:8482", + "issue:12470|issue:13067", + "issue:1801|issue:646", + "issue:3698|issue:541", + "issue:5368|issue:5502", + "issue:6452|issue:8780", + "issue:10636|issue:8604", + "issue:1214|issue:851", + "issue:2728|issue:2947", + "issue:7606|issue:9799", + "issue:11555|issue:12505" + ], + "prepared_review_unit_hash": "8a4af21d276fa053c960aa9898d16d18803a9c841864d78335b9d3240e6e33a8", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34022, + "estimated_input_tokens": 8506, + "estimated_eval_tokens": 17268 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly a heterogeneous set of unrelated bugs/features across different pipelines and training paths. I don\u2019t see any soft pair that clearly describes the same underlying issue, so all soft edges should be rejected.", + "confidence": 0.79, + "canonical_issue_reason": "There is no strong duplicate cluster here; if one issue must be used as an anchor, issue 10412 is the most detailed and self-contained report, but it does not subsume the others.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10412 is the strongest standalone issue in the set because it has the clearest problem statement and most context; the rest are mostly unrelated or too broad to serve as a better canonical anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5395", + "right": "issue:6333", + "accept": false, + "reason": "Both involve LoRA/PEFT, but one is about combining civitai LoRAs and the other about finetuning a pretrained LoRA; different problems and code paths." + }, + { + "left": "issue:1405", + "right": "issue:2340", + "accept": false, + "reason": "Attention slicing regression and a safety-checker failure are unrelated bugs; same project area, not the same issue." + }, + { + "left": "issue:8744", + "right": "issue:8939", + "accept": false, + "reason": "Loading from single-file training support and a tokenizer-loading error for SD3 are different failures." + }, + { + "left": "issue:10412", + "right": "issue:11225", + "accept": false, + "reason": "Both are OOM reports, but they concern different training examples, models, and triggering steps." + }, + { + "left": "issue:1791", + "right": "issue:457", + "accept": false, + "reason": "Both mention MPS, but one is UnCLIP unsupported on Mac/MPS and the other is image artifact corruption; not the same bug." + }, + { + "left": "issue:1896", + "right": "issue:2392", + "accept": false, + "reason": "Different runtime errors in different pipelines: one is a size/type error in CLIP-guided SD, the other is prompt_embeds validation in img2img." + }, + { + "left": "issue:3799", + "right": "issue:8373", + "accept": false, + "reason": "Type inference for `from_pretrained` and an incorrect error message are unrelated." + }, + { + "left": "issue:5595", + "right": "issue:6363", + "accept": false, + "reason": "One is an image-size divisibility validation error; the other is an fp16 gradient unscale failure during LoRA training." + }, + { + "left": "issue:6094", + "right": "issue:8552", + "accept": false, + "reason": "Both involve TextToVideoZeroPipeline, but one is a version regression and the other is a segmentation fault during inference; too different to merge." + }, + { + "left": "issue:1655", + "right": "issue:3767", + "accept": false, + "reason": "Pipeline initialization failure and FrozenDict mutability semantics are unrelated." + }, + { + "left": "issue:4492", + "right": "issue:4534", + "accept": false, + "reason": "A JSON/boolean NameError and an SDXL DreamBooth MPS validation crash are different bugs." + }, + { + "left": "issue:11839", + "right": "issue:12107", + "accept": false, + "reason": "GGUF loading for FluxKontextPipeline and `accelerator.init_trackers` with custom objects are unrelated." + }, + { + "left": "issue:1949", + "right": "issue:2466", + "accept": false, + "reason": "DreamBooth tutorial training errors and checkpoint limit enforcement are different example-script issues." + }, + { + "left": "issue:10672", + "right": "issue:8482", + "accept": false, + "reason": "Callback support for pipelines and SD3-medium model support are not the same change or bug." + }, + { + "left": "issue:12470", + "right": "issue:13067", + "accept": false, + "reason": "Quantizer API additions and adding Anima model support are unrelated feature requests." + }, + { + "left": "issue:1801", + "right": "issue:646", + "accept": false, + "reason": "`getattr` misuse and a missing `set_format` method on PNDMScheduler are different attribute errors." + }, + { + "left": "issue:3698", + "right": "issue:541", + "accept": false, + "reason": "Converting an inpainting model and running Diffusers on SageMaker are unrelated requests." + }, + { + "left": "issue:5368", + "right": "issue:5502", + "accept": false, + "reason": "Both mention fp16, but one is a dtype mismatch in attention and the other is lack of fp16 support in a training script." + }, + { + "left": "issue:6452", + "right": "issue:8780", + "accept": false, + "reason": "Adding masked image latents to ControlNet inpaint and a generic model/input dtype mismatch are not the same issue." + }, + { + "left": "issue:10636", + "right": "issue:8604", + "accept": false, + "reason": "A bug in `value_guided_sampling.py` and corrupted output from converting the T5 encoder to float16 are unrelated." + }, + { + "left": "issue:1214", + "right": "issue:851", + "accept": false, + "reason": "Both are half-precision/dtype complaints, but the reports are too vague and likely stem from different code paths to treat as duplicates." + }, + { + "left": "issue:2728", + "right": "issue:2947", + "accept": false, + "reason": "Cannot load safetensors in `from_pretrained` and missing `load_attn_procs` are distinct model-loading/API issues." + }, + { + "left": "issue:7606", + "right": "issue:9799", + "accept": false, + "reason": "A deprecated LoRA processor in Kandinsky 2.2 and a tensor-size mismatch runtime error are unrelated." + }, + { + "left": "issue:11555", + "right": "issue:12505", + "accept": false, + "reason": "`device_map=\"auto\"` support and a tgate error in img2img pipelines are different topics." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 219, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10178", + "issue:10350", + "issue:10382", + "issue:10512", + "issue:10887", + "issue:10896", + "issue:10954", + "issue:10986", + "issue:11569", + "issue:11687", + "issue:11779", + "issue:11922", + "issue:12108", + "issue:12575", + "issue:12624", + "issue:12738", + "issue:13277", + "issue:1358", + "issue:2195", + "issue:2266", + "issue:2838", + "issue:2883", + "issue:3960", + "issue:4014", + "issue:4269", + "issue:4406", + "issue:4790", + "issue:5538", + "issue:5628", + "issue:5731", + "issue:5824", + "issue:5849", + "issue:6016", + "issue:6374", + "issue:6513", + "issue:6570", + "issue:6866", + "issue:7267", + "issue:7640", + "issue:7936", + "issue:8577", + "issue:8624", + "issue:8939", + "issue:8968", + "issue:9101", + "issue:9680", + "issue:9690" + ], + "soft_pairs": [ + "issue:12108|issue:12575", + "issue:10350|issue:10382", + "issue:10178|issue:6570", + "issue:10887|issue:11569", + "issue:11687|issue:12624", + "issue:5849|issue:6374", + "issue:1358|issue:2838", + "issue:7267|issue:9680", + "issue:4790|issue:6016", + "issue:10986|issue:11779", + "issue:7936|issue:8968", + "issue:2195|issue:2883", + "issue:2266|issue:4269", + "issue:5538|issue:5731", + "issue:8624|issue:8939", + "issue:12738|issue:13277", + "issue:4014|issue:5824", + "issue:5628|issue:7640", + "issue:10100|issue:10896", + "issue:10954|issue:11922", + "issue:6513|issue:6866", + "issue:8577|issue:9101", + "issue:3960|issue:4406", + "issue:10512|issue:9690" + ], + "prepared_review_unit_hash": "e3a9ff55bc6ae657bc6cf82d94f64eede679eda4ff85755b879a8425caefa803", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34160, + "estimated_input_tokens": 8540, + "estimated_eval_tokens": 17336 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: feature requests, unrelated pipeline bugs, and a few load/shape/runtime errors. There is no cluster-wide duplicate set; the only clear duplicate-looking pair is the safetensors HeaderTooLarge LoRA reports.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue cleanly represents the whole cluster because the items span many unrelated bugs/features. If one issue must be chosen, issue:4406 is the best canonical for the only true duplicate pair because it states the exact safetensors HeaderTooLarge failure most explicitly.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4406 is the strongest issue candidate: it is concrete, specific, and matches another near-identical report (issue:3960) about the same LoRA safetensors HeaderTooLarge error.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12108", + "right": "issue:12575", + "accept": false, + "reason": "Different failures in Chroma/Qwen image pipelines: flow-matching scheduler behavior vs attention-backend checks." + }, + { + "left": "issue:10350", + "right": "issue:10382", + "accept": false, + "reason": "Both are VAE-related, but one is decoder memory use and the other is an encoder return-type AttributeError." + }, + { + "left": "issue:10178", + "right": "issue:6570", + "accept": false, + "reason": "Different tensor-shape bugs in different modules/models; only superficial size-mismatch similarity." + }, + { + "left": "issue:10887", + "right": "issue:11569", + "accept": false, + "reason": "Import error for StableDiffusionXLTrainer vs a feature request for a model repo; unrelated." + }, + { + "left": "issue:11687", + "right": "issue:12624", + "accept": false, + "reason": "Docs/framework addition request vs a new pipeline parameter request; not the same bug or change." + }, + { + "left": "issue:5849", + "right": "issue:6374", + "accept": false, + "reason": "Training-script argparse TypeError vs LoRA weight loading question; different code paths and symptoms." + }, + { + "left": "issue:1358", + "right": "issue:2838", + "accept": false, + "reason": "Generic safetensors support request vs a ControlNet pipeline feature request; unrelated." + }, + { + "left": "issue:7267", + "right": "issue:9680", + "accept": false, + "reason": "Broken local data loading in train_controlnet.py vs SDXL unconditional generation bug; different problems." + }, + { + "left": "issue:4790", + "right": "issue:6016", + "accept": false, + "reason": "Both involve conversion utilities, but one is round-trip file compatibility and the other is a torch.Size formatting/error case." + }, + { + "left": "issue:10986", + "right": "issue:11779", + "accept": false, + "reason": "MPS complex-dtype conversion error in transformer pipeline vs incorrect mask padding in Wan VACE; different pipelines and bugs." + }, + { + "left": "issue:7936", + "right": "issue:8968", + "accept": false, + "reason": "Invalid JSON/header parsing for a local safetensors file vs general pipeline safetensors loading support." + }, + { + "left": "issue:2195", + "right": "issue:2883", + "accept": false, + "reason": "Both are video-related feature requests, but they target different pipeline additions and are not the same change." + }, + { + "left": "issue:2266", + "right": "issue:4269", + "accept": false, + "reason": "Sequential CPU offload slowdown vs ControlNet input-validation TypeError; different concrete issues." + }, + { + "left": "issue:5538", + "right": "issue:5731", + "accept": false, + "reason": "Upscale pipeline latent-usage question vs missing trainable logvar in AutoencoderKL decoder; unrelated code paths." + }, + { + "left": "issue:8624", + "right": "issue:8939", + "accept": false, + "reason": "Undefined CUDA symbol in Stable Diffusion 3 setup vs tokenizer-loading OSError; different runtime failures." + }, + { + "left": "issue:12738", + "right": "issue:13277", + "accept": false, + "reason": "Both mention batching, but one is ZImage batch-size assertion and the other is QwenImage Ulysses SP batch handling." + }, + { + "left": "issue:4014", + "right": "issue:5824", + "accept": false, + "reason": "SDXL watermark artifacts vs incorrect text_encoder_projection_dim default; unrelated bugs." + }, + { + "left": "issue:5628", + "right": "issue:7640", + "accept": false, + "reason": "Scheduler denoising range offset vs from_single_file dimension runtime error; different root causes." + }, + { + "left": "issue:10100", + "right": "issue:10896", + "accept": false, + "reason": "Two different model-support requests (ConsisID vs SkyReels-A1); not the same feature." + }, + { + "left": "issue:10954", + "right": "issue:11922", + "accept": false, + "reason": "FLUX LoRA loading failure vs FLUX.1-Fill-dev runtime problem; same family name but different concrete bugs." + }, + { + "left": "issue:6513", + "right": "issue:6866", + "accept": false, + "reason": "Different adapter integrations: T2IAdapter model support vs IP-Adapter support for SDXL ControlNet inpaint." + }, + { + "left": "issue:8577", + "right": "issue:9101", + "accept": false, + "reason": "Differential diffusion feature for SD3 vs an InstructPix2Pix training script for SD3; different work items." + }, + { + "left": "issue:3960", + "right": "issue:4406", + "accept": true, + "reason": "Both report the same safetensors HeaderTooLarge problem while loading a LoRA; 4406 is the clearer description." + }, + { + "left": "issue:10512", + "right": "issue:9690", + "accept": false, + "reason": "LoRA loading failure in Flux/Quanto vs FP8 example-code cleanup; unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative. The one accepted soft-edge pair, issue:3960 and issue:4406, is supported by the matching safetensors HeaderTooLarge LoRA reports. The remaining pair verdicts are appropriately non-committal/negative given the evidence. No overstatement detected." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 220, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10336", + "issue:10542", + "issue:10639", + "issue:11519", + "issue:1181", + "issue:1197", + "issue:1205", + "issue:12060", + "issue:12239", + "issue:13430", + "issue:1350", + "issue:1440", + "issue:1594", + "issue:193", + "issue:1962", + "issue:2210", + "issue:2332", + "issue:3087", + "issue:318", + "issue:3325", + "issue:3444", + "issue:3698", + "issue:3891", + "issue:3975", + "issue:4014", + "issue:4232", + "issue:459", + "issue:4674", + "issue:4788", + "issue:4949", + "issue:4964", + "issue:5160", + "issue:5687", + "issue:5710", + "issue:5886", + "issue:5960", + "issue:6136", + "issue:6935", + "issue:7569", + "issue:7661", + "issue:7782", + "issue:7882", + "issue:8504", + "issue:8723", + "issue:9101", + "issue:9450", + "issue:9599", + "issue:9680" + ], + "soft_pairs": [ + "issue:1197|issue:459", + "issue:1205|issue:1440", + "issue:12060|issue:9599", + "issue:193|issue:3975", + "issue:4674|issue:4788", + "issue:8504|issue:9101", + "issue:10639|issue:12239", + "issue:10542|issue:9680", + "issue:11519|issue:13430", + "issue:1181|issue:3698", + "issue:2210|issue:5160", + "issue:2332|issue:4232", + "issue:4964|issue:7782", + "issue:1350|issue:3087", + "issue:1594|issue:4949", + "issue:1962|issue:3444", + "issue:318|issue:3891", + "issue:5687|issue:5710", + "issue:6935|issue:7882", + "issue:3325|issue:5960", + "issue:4014|issue:8723", + "issue:6136|issue:7569", + "issue:5886|issue:7661", + "issue:10336|issue:9450" + ], + "prepared_review_unit_hash": "cb1ccf974fcd160bf941fb2d7d55ac5fcdf55ccb9febcffd67704bf1b56c5532", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33161, + "estimated_input_tokens": 8291, + "estimated_eval_tokens": 16838 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated model-support requests, training/inference feature asks, and distinct bug reports across many different pipelines. The soft pairs are mostly only superficially similar by wording and should all be rejected as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1197", + "right": "issue:459", + "accept": false, + "reason": "Different requests: OpenAI Diffusion Pipeline vs Incremental Diffusion. Similar theme, not the same feature or bug." + }, + { + "left": "issue:1205", + "right": "issue:1440", + "accept": false, + "reason": "ONNX conversion request and deleting cached models are unrelated." + }, + { + "left": "issue:12060", + "right": "issue:9599", + "accept": false, + "reason": "A question about DiT blocks is not the same as asking for a LoRA-only FLUX finetune example." + }, + { + "left": "issue:193", + "right": "issue:3975", + "accept": false, + "reason": "Generic pipeline error report vs a typo in an error message; different problems." + }, + { + "left": "issue:4674", + "right": "issue:4788", + "accept": false, + "reason": "Memory leak when changing width/height is distinct from SDXL inpaint OOM on 6 GB VRAM." + }, + { + "left": "issue:8504", + "right": "issue:9101", + "accept": false, + "reason": "Selecting text encoders for SD3 training and an InstructPix2Pix SD3 training script are separate feature requests." + }, + { + "left": "issue:10639", + "right": "issue:12239", + "accept": false, + "reason": "Support for Lumina-Image 2.0 and support for InfiniteTalk are different model integrations." + }, + { + "left": "issue:10542", + "right": "issue:9680", + "accept": false, + "reason": "Hunyuan Video batch-size regression is unrelated to SDXL unconditional generation being broken." + }, + { + "left": "issue:11519", + "right": "issue:13430", + "accept": false, + "reason": "Support requests for MAGI-1 and JoyAI-Image-Edit target different models." + }, + { + "left": "issue:1181", + "right": "issue:3698", + "accept": false, + "reason": "Training from scratch is not the same as converting an inpainting model to diffusers." + }, + { + "left": "issue:2210", + "right": "issue:5160", + "accept": false, + "reason": "LoRA text-encoder training and Dreambooth/LoRA masking are different training features." + }, + { + "left": "issue:2332", + "right": "issue:4232", + "accept": false, + "reason": "Dreambooth inference error vs instruct-pix2pix ControlNet training/inference; not the same bug." + }, + { + "left": "issue:4964", + "right": "issue:7782", + "accept": false, + "reason": "SDXL model conversion and safetensors conversion failure are unrelated." + }, + { + "left": "issue:1350", + "right": "issue:3087", + "accept": false, + "reason": "DiffusionDet object detection support and LyCORIS/LoCon/LoHA support are different domains." + }, + { + "left": "issue:1594", + "right": "issue:4949", + "accept": false, + "reason": "Video img2img generation and ProPainter in/outpainting are not the same change." + }, + { + "left": "issue:1962", + "right": "issue:3444", + "accept": false, + "reason": "Video generation with stable diffusion is unrelated to audio-to-image integration with ImageBind." + }, + { + "left": "issue:318", + "right": "issue:3891", + "accept": false, + "reason": "Grad-TTS and Tortoise TTS are separate TTS pipeline requests." + }, + { + "left": "issue:5687", + "right": "issue:5710", + "accept": false, + "reason": "One is a low-strength img2img timestep bug; the other is a scheduler initialization bug. Different code paths." + }, + { + "left": "issue:6935", + "right": "issue:7882", + "accept": false, + "reason": "IP-Adapter face ID support and missing from_single_file on T2I-Adapter are unrelated." + }, + { + "left": "issue:3325", + "right": "issue:5960", + "accept": false, + "reason": "Dreambooth CUDA invalid argument and SDXL ControlNet NaN/uint8 cast warning are different failures." + }, + { + "left": "issue:4014", + "right": "issue:8723", + "accept": false, + "reason": "Watermark pixel artifacts in SDXL are unrelated to SD3ControlNetModel initialization errors." + }, + { + "left": "issue:6136", + "right": "issue:7569", + "accept": false, + "reason": "SD2.1 NaNs on Mac and a FreeU test failure on Mac are not the same issue." + }, + { + "left": "issue:5886", + "right": "issue:7661", + "accept": false, + "reason": "IP-Adapter face model support and enabling animate diff with IP-Adapter are distinct feature requests." + }, + { + "left": "issue:10336", + "right": "issue:9450", + "accept": false, + "reason": "Sana activation clamping and Flux multi-GPU device mismatch are different bugs in different models." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 221, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10054", + "issue:10210", + "issue:10513", + "issue:10650", + "issue:10800", + "issue:10932", + "issue:10962", + "issue:11006", + "issue:11352", + "issue:12396", + "issue:12504", + "issue:12606", + "issue:2203", + "issue:2228", + "issue:2364", + "issue:2947", + "issue:3032", + "issue:3085", + "issue:319", + "issue:3623", + "issue:3861", + "issue:3973", + "issue:4674", + "issue:4692", + "issue:4704", + "issue:4835", + "issue:490", + "issue:4936", + "issue:5044", + "issue:5372", + "issue:539", + "issue:5743", + "issue:5824", + "issue:5889", + "issue:5897", + "issue:6028", + "issue:671", + "issue:7232", + "issue:730", + "issue:7611", + "issue:7799", + "issue:7991", + "issue:8930", + "issue:938", + "issue:9392", + "issue:9485", + "issue:987" + ], + "soft_pairs": [ + "issue:3623|issue:671", + "issue:10054|issue:12606", + "issue:11352|issue:12504", + "issue:319|issue:938", + "issue:5743|issue:5897", + "issue:5372|issue:5824", + "issue:2364|issue:4704", + "issue:4674|issue:490", + "issue:2228|issue:3861", + "issue:4835|issue:8930", + "issue:5889|issue:7232", + "issue:539|issue:987", + "issue:7611|issue:7799", + "issue:3085|issue:3973", + "issue:2947|issue:4692", + "issue:10650|issue:10962", + "issue:10800|issue:10932", + "issue:10210|issue:11006", + "issue:5044|issue:6028", + "issue:2203|issue:730", + "issue:10513|issue:12396", + "issue:3032|issue:3861", + "issue:4936|issue:7991", + "issue:9392|issue:9485" + ], + "prepared_review_unit_hash": "a7b7b4a63d68d8dbe4096c6cf04d11861be708f7cb3f2b028c0ee62024f21410", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33386, + "estimated_input_tokens": 8347, + "estimated_eval_tokens": 16950 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft links are mostly false positives: these issues span unrelated pipelines, scheduler behavior, quantization/offload, shape mismatches, and docs/training edge cases. No pair looks like the same underlying bug or change.", + "confidence": 0.94, + "canonical_issue_reason": "issue:10210 is the most concrete and widely discussed user-facing bug report here, so it is the least arbitrary representative of this heterogeneous set.", + "canonical_pr_reason": null, + "best_issue_reason": "As a cluster representative, issue:10210 is the strongest fit by specificity and discussion activity, but it should not be treated as a true duplicate anchor for the whole group.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3623", + "right": "issue:671", + "accept": false, + "reason": "Both are model loading/behavior complaints, but one is about large-model loading failures and the other about converted-model fidelity; different underlying problems." + }, + { + "left": "issue:10054", + "right": "issue:12606", + "accept": false, + "reason": "Scheduler test parameter bug vs wrong script import; unrelated code paths." + }, + { + "left": "issue:11352", + "right": "issue:12504", + "accept": false, + "reason": "CPU move warning for 8bit quantized pipelines vs freeU ComplexHalf warning; different functionality and failure modes." + }, + { + "left": "issue:319", + "right": "issue:938", + "accept": false, + "reason": "Completely different domains: VQ-Diffusion model request vs human motion diffusion model request." + }, + { + "left": "issue:5743", + "right": "issue:5897", + "accept": false, + "reason": "Training stuck at 0% versus device-mismatch in LoRA training; not the same bug." + }, + { + "left": "issue:5372", + "right": "issue:5824", + "accept": false, + "reason": "Both mention SDXL pipeline internals, but one asks how to use safety_checker and the other reports a bad default in get_add_time_ids()." + }, + { + "left": "issue:2364", + "right": "issue:4704", + "accept": false, + "reason": "Generic training-script failure vs Custom Diffusion training failure; too broad and no shared concrete failure mode." + }, + { + "left": "issue:4674", + "right": "issue:490", + "accept": false, + "reason": "Memory leak when changing width/height across calls is not the same as general low-memory complaints." + }, + { + "left": "issue:2228", + "right": "issue:3861", + "accept": false, + "reason": "Loading-to-GPU issue vs API naming question about from_pretrained; different underlying issue." + }, + { + "left": "issue:4835", + "right": "issue:8930", + "accept": false, + "reason": "ControlNet img2img bug with MultiControlNetModel vs pose-control adherence issues; both about control images but not the same concrete defect." + }, + { + "left": "issue:5889", + "right": "issue:7232", + "accept": false, + "reason": "Stable video diffusion feature request/announcement vs support for multimodal diffusion transformers; not duplicates." + }, + { + "left": "issue:539", + "right": "issue:987", + "accept": false, + "reason": "Different unrelated TypeError reports from different APIs and pipelines." + }, + { + "left": "issue:7611", + "right": "issue:7799", + "accept": false, + "reason": "Shape mismatch in unet2dconditional vs matmul mismatch with IP-Adapter; both are tensor-shape errors but distinct code paths." + }, + { + "left": "issue:3085", + "right": "issue:3973", + "accept": false, + "reason": "Removed return_cache_folder argument vs unexpected vae_path keyword; both are parameter/API mismatches but not the same one." + }, + { + "left": "issue:2947", + "right": "issue:4692", + "accept": false, + "reason": "Missing load_attn_procs attribute vs generic UNet memory concerns; unrelated defects." + }, + { + "left": "issue:10650", + "right": "issue:10962", + "accept": false, + "reason": "Lumina reshape runtime error vs Cogview4 prompt-embed shape issues; similar symptom class only, not the same bug." + }, + { + "left": "issue:10800", + "right": "issue:10932", + "accept": false, + "reason": "Quantized model plus sequential CPU offload crash vs torch.compile recompile limit; different mechanisms and failures." + }, + { + "left": "issue:10210", + "right": "issue:11006", + "accept": false, + "reason": "FluxImg2Img output too similar to input is not the same as broken Wan 2.1 I2V output with a quantized transformer." + }, + { + "left": "issue:5044", + "right": "issue:6028", + "accept": false, + "reason": "Kandinsky 2.2 load failure vs Kandinsky 3.0 CUDA OOM; different models and different failure modes." + }, + { + "left": "issue:2203", + "right": "issue:730", + "accept": false, + "reason": "DreamBooth download location question vs DreamBooth training quality issue; unrelated." + }, + { + "left": "issue:10513", + "right": "issue:12396", + "accept": false, + "reason": "Colab FLUX errors vs leaf_level/delete_adapters errors; different subsystems and symptoms." + }, + { + "left": "issue:3032", + "right": "issue:3861", + "accept": false, + "reason": "ControlNet class lacking from_pretrained vs generic Pipeline.from_pretrained question; not the same defect." + }, + { + "left": "issue:4936", + "right": "issue:7991", + "accept": false, + "reason": "Dreambooth LoRA resume regression vs basic training tutorial bug; different workflows and root causes." + }, + { + "left": "issue:9392", + "right": "issue:9485", + "accept": false, + "reason": "Scheduler SNR shift design question vs request to move scheduler computation to GPU; related area but not the same underlying bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 222, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1013", + "issue:10202", + "issue:10334", + "issue:10336", + "issue:10411", + "issue:10490", + "issue:10514", + "issue:10529", + "issue:10596", + "issue:10671", + "issue:10817", + "issue:1083", + "issue:10873", + "issue:11046", + "issue:11148", + "issue:11886", + "issue:12052", + "issue:12321", + "issue:12459", + "issue:13506", + "issue:1791", + "issue:2089", + "issue:4210", + "issue:4332", + "issue:4534", + "issue:4616", + "issue:4631", + "issue:5026", + "issue:5545", + "issue:5595", + "issue:5646", + "issue:5689", + "issue:5951", + "issue:6091", + "issue:6222", + "issue:6518", + "issue:6783", + "issue:6794", + "issue:6928", + "issue:760", + "issue:7793", + "issue:7842", + "issue:834", + "issue:8589", + "issue:8612", + "issue:8647", + "issue:9953", + "issue:9972" + ], + "soft_pairs": [ + "issue:1791|issue:760", + "issue:10336|issue:10514", + "issue:11148|issue:9953", + "issue:1083|issue:834", + "issue:4616|issue:8647", + "issue:11886|issue:12459", + "issue:10411|issue:6518", + "issue:10817|issue:8589", + "issue:6091|issue:6222", + "issue:12052|issue:13506", + "issue:7793|issue:7842", + "issue:6783|issue:8612", + "issue:5545|issue:5595", + "issue:5646|issue:5689", + "issue:4210|issue:4332", + "issue:4534|issue:4631", + "issue:2089|issue:5951", + "issue:1013|issue:5026", + "issue:10202|issue:11046", + "issue:10334|issue:9972", + "issue:10490|issue:12321", + "issue:10529|issue:6928", + "issue:10596|issue:6794", + "issue:10671|issue:10873" + ], + "prepared_review_unit_hash": "e7fe80716a3256933b8fd6dee5645cd0d5dc9aa76344ba43b25233f710506ea9", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33866, + "estimated_input_tokens": 8467, + "estimated_eval_tokens": 17190 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is not a duplicate cluster: the items span many unrelated bugs, feature requests, and model-specific issues. None of the soft-similarity pairs are the same underlying problem closely enough to merge.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue stands out because the issues are heterogeneous and do not describe one shared bug or feature request.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a good global canonical for the set; the topics are too diverse (Sana, Flux, PEFT, schedulers, training, MPS, etc.).", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1791", + "right": "issue:760", + "accept": false, + "reason": "Both mention MPS, but one is UnCLIPPipeline support and the other is unexpected StableDiffusionPipeline output in an older release; different code paths and symptoms." + }, + { + "left": "issue:10336", + "right": "issue:10514", + "accept": false, + "reason": "Both are Sana-related, but one reports exploding activations/clamping while the other is a 4k resolution-binning sample_size limitation; not the same bug." + }, + { + "left": "issue:11148", + "right": "issue:9953", + "accept": false, + "reason": "Both involve loading/unloading adapters, but one is a PEFT state flag bug and the other is an error moving a pipeline with a quantized component to CUDA." + }, + { + "left": "issue:1083", + "right": "issue:834", + "accept": false, + "reason": "FP16 ONNX incorrect output is not the same as CPU underutilization; different functionality and failure mode." + }, + { + "left": "issue:4616", + "right": "issue:8647", + "accept": false, + "reason": "One is an SDXL LoRA conversion script issue, the other is a ControlNet img2img runtime NoneType error; unrelated." + }, + { + "left": "issue:11886", + "right": "issue:12459", + "accept": false, + "reason": "Both are FLUX/FluxKontext-related, but one is a pipeline issue and the other is a batch throughput concern; not the same concrete bug." + }, + { + "left": "issue:10411", + "right": "issue:6518", + "accept": false, + "reason": "Different training scripts and different errors: one asks how to call a consistency-stiffness LoRA training script, the other is a NoneType floor-division failure in a different distillation script." + }, + { + "left": "issue:10817", + "right": "issue:8589", + "accept": false, + "reason": "SD3 ControlNet missing in auto_pipeline is a support gap; SD3 DreamBooth prior preservation is a training limitation. Different features, different fixes." + }, + { + "left": "issue:6091", + "right": "issue:6222", + "accept": false, + "reason": "Both are scheduler/memory-related, but one concerns a consistency decoder size argument and the other DDPM batch-size dependency; not the same bug." + }, + { + "left": "issue:12052", + "right": "issue:13506", + "accept": false, + "reason": "Both involve attention/offloading on newer models, but one is a device-mismatch multiplication failure in Wan 2.2 and the other is Sage attention failing with ring-attention; distinct code paths." + }, + { + "left": "issue:7793", + "right": "issue:7842", + "accept": false, + "reason": "Both mention LoRA/deepfloyd, but one is unloading LoRA weights causing an error and the other is a tensor-size mismatch on non-8-divisible input dimensions; not the same issue." + }, + { + "left": "issue:6783", + "right": "issue:8612", + "accept": false, + "reason": "Both are PixArt/Pixart-sigma generation quality or precision issues, but one is messed-up generations and the other is mixed-precision incompatibility; different problems." + }, + { + "left": "issue:5545", + "right": "issue:5595", + "accept": false, + "reason": "Both are Stable Diffusion pipeline errors, but one is inpainting pipeline behavior and the other is a height/width divisibility validation error." + }, + { + "left": "issue:5646", + "right": "issue:5689", + "accept": false, + "reason": "Both concern diffusion scheduler behavior, but one is broad stability issues on a specific model and the other is residual noise in SDXL from DPM++; not the same concrete defect." + }, + { + "left": "issue:4210", + "right": "issue:4332", + "accept": false, + "reason": "Attention_mask usage in UNet2DConditionModel is unrelated to xformers failing in kandinsky2.2 pipeline." + }, + { + "left": "issue:4534", + "right": "issue:4631", + "accept": false, + "reason": "Both are runtime errors in generation/training, but one is SDXL dreambooth on MPS with a dimension mismatch and the other is a DPMSolverSinglestepScheduler scalar conversion error." + }, + { + "left": "issue:2089", + "right": "issue:5951", + "accept": false, + "reason": "DreamBooth OOM with AltDiffusion is a memory problem; Pixart training is a broad training issue with no matching failure mode." + }, + { + "left": "issue:1013", + "right": "issue:5026", + "accept": false, + "reason": "StableDiffusionMegaPipeline not working in Diffusers 6.0 and KeyError:'text_encoder' in 0.21.0 are version-specific but describe different failures in different code paths." + }, + { + "left": "issue:10202", + "right": "issue:11046", + "accept": false, + "reason": "A Flux Control LoRA unload bug is unrelated to FLUX pipeline inference with ControlNet/inpainting/IP-Adapter." + }, + { + "left": "issue:10334", + "right": "issue:9972", + "accept": false, + "reason": "Sana MacOS grey/NaN outputs are unrelated to CogX requesting an absurd 10TB buffer; different pipeline and failure mechanism." + }, + { + "left": "issue:10490", + "right": "issue:12321", + "accept": false, + "reason": "'support Infinity' and HunyuanImage-2.1 support are both feature requests, but for different models/features and not the same request." + }, + { + "left": "issue:10529", + "right": "issue:6928", + "accept": false, + "reason": "Missing linspace timestep spacing in DDIMInverseScheduler is unrelated to disabling depthmap colorization in Marigold." + }, + { + "left": "issue:10596", + "right": "issue:6794", + "accept": false, + "reason": "Invalid shape in FluxControlPipeline is a runtime shape bug; cast_training_params import error is a separate training-utils API issue." + }, + { + "left": "issue:10671", + "right": "issue:10873", + "accept": false, + "reason": "Deterministic dtype/device access in DiffusionPipeline and SanaPipeline issues are both pipeline bugs, but they are unrelated subsystems with different symptoms." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 223, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10760", + "issue:10767", + "issue:1101", + "issue:1117", + "issue:11351", + "issue:11352", + "issue:11519", + "issue:12108", + "issue:12338", + "issue:12358", + "issue:12599", + "issue:12645", + "issue:12666", + "issue:12754", + "issue:13034", + "issue:1305", + "issue:13191", + "issue:13377", + "issue:1398", + "issue:1530", + "issue:1541", + "issue:1919", + "issue:202", + "issue:2250", + "issue:2377", + "issue:2538", + "issue:2550", + "issue:2603", + "issue:2674", + "issue:2761", + "issue:2866", + "issue:2883", + "issue:2972", + "issue:3010", + "issue:3073", + "issue:3159", + "issue:3838", + "issue:3843", + "issue:4386", + "issue:4390", + "issue:5285", + "issue:5877", + "issue:6928", + "issue:7365", + "issue:8410", + "issue:940", + "issue:9471", + "issue:9744" + ], + "soft_pairs": [ + "issue:10760|issue:13034", + "issue:10767|issue:8410", + "issue:1101|issue:5285", + "issue:1117|issue:3010", + "issue:11351|issue:12666", + "issue:11352|issue:12599", + "issue:11519|issue:13191", + "issue:12108|issue:12645", + "issue:12338|issue:9471", + "issue:12358|issue:9744", + "issue:12754|issue:13377", + "issue:1305|issue:3073", + "issue:1398|issue:940", + "issue:1530|issue:1919", + "issue:1541|issue:5877", + "issue:202|issue:2377", + "issue:2250|issue:2603", + "issue:2538|issue:6928", + "issue:2550|issue:3843", + "issue:2674|issue:4390", + "issue:2761|issue:3159", + "issue:2866|issue:4386", + "issue:2883|issue:3838", + "issue:2972|issue:7365" + ], + "prepared_review_unit_hash": "2ba797a0bc4b539409cf083cb29a971b8bdeb64ba5acd865a2f115cbf2df2450", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33004, + "estimated_input_tokens": 8251, + "estimated_eval_tokens": 16758 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All proposed soft-edge pairs are weakly similar at best and mostly unrelated; they span different bugs, features, docs, models, and pipelines. I would not merge any pair as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue stands out: this set is heterogeneous, with unrelated model-support requests, bugs, docs issues, and pipeline-specific regressions.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue candidate: none of these issues clearly subsumes the others or represents the same underlying problem across the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10760", + "right": "issue:13034", + "accept": false, + "reason": "Different failures in different pipelines: import-time missing attribute vs WanPipeline height/width handling." + }, + { + "left": "issue:10767", + "right": "issue:8410", + "accept": false, + "reason": "FlashVideo integration request is unrelated to native TextDiffuser pipeline support." + }, + { + "left": "issue:1101", + "right": "issue:5285", + "accept": false, + "reason": "Scheduler feature request and Japanese docs translation are unrelated." + }, + { + "left": "issue:1117", + "right": "issue:3010", + "accept": false, + "reason": "Negative-prompt tests and distributed inference docs address different areas." + }, + { + "left": "issue:11351", + "right": "issue:12666", + "accept": false, + "reason": "Wan video processor dtype question is unrelated to Kandinsky 5.0 video pipeline availability." + }, + { + "left": "issue:11352", + "right": "issue:12599", + "accept": false, + "reason": "One is missing a warning on CPU move for 8-bit models; the other is slow CUDA transfer after loading." + }, + { + "left": "issue:11519", + "right": "issue:13191", + "accept": false, + "reason": "Two separate model support requests (MAGI-1 vs Elastic-DiT) with no shared bug/change." + }, + { + "left": "issue:12108", + "right": "issue:12645", + "accept": false, + "reason": "Flow-matching scheduler breakage is a functional bug; the other is a speed comparison complaint." + }, + { + "left": "issue:12338", + "right": "issue:9471", + "accept": false, + "reason": "AutoencoderDC VAE slicing bug is unrelated to a file-extension/resource-type parsing error." + }, + { + "left": "issue:12358", + "right": "issue:9744", + "accept": false, + "reason": "Different model support requests: Lumina-DiMOO vs Mochi video model." + }, + { + "left": "issue:12754", + "right": "issue:13377", + "accept": false, + "reason": "Both mention pipeline behavior, but one is float32 casting in from_pipe and the other is CFG suppression with negative_prompt_embeds." + }, + { + "left": "issue:1305", + "right": "issue:3073", + "accept": false, + "reason": "Direct Inversion method request is unrelated to adding consistency models." + }, + { + "left": "issue:1398", + "right": "issue:940", + "accept": false, + "reason": "Both touch MPS, but one is Stable Diffusion 2 support and the other is an LMSDiscreteScheduler crash." + }, + { + "left": "issue:1530", + "right": "issue:1919", + "accept": false, + "reason": "Missing VQDiffusion pipeline embedding field and missing DreamBooth requirements are unrelated." + }, + { + "left": "issue:1541", + "right": "issue:5877", + "accept": false, + "reason": "Both involve DreamBooth, but one is CKPT/model-type detection and the other is subprocess failures in training scripts." + }, + { + "left": "issue:202", + "right": "issue:2377", + "accept": false, + "reason": "Unet blocks tests and EMAModel tests are unrelated test additions." + }, + { + "left": "issue:2250", + "right": "issue:2603", + "accept": false, + "reason": "Different DreamBooth/Attend-And-Excite inference problems with different symptoms and code paths." + }, + { + "left": "issue:2538", + "right": "issue:6928", + "accept": false, + "reason": "Train-text-encoder option and Marigold depthmap colorization are separate feature requests." + }, + { + "left": "issue:2550", + "right": "issue:3843", + "accept": false, + "reason": "ControlNet model conversion and missing global_step key are different code-path issues." + }, + { + "left": "issue:2674", + "right": "issue:4390", + "accept": false, + "reason": "A missing safetensors dependency in LoRA workflows is not the same as a generic missing requirements report." + }, + { + "left": "issue:2761", + "right": "issue:3159", + "accept": false, + "reason": "Documentation request for UNet1DModel and a typo fix in docs are different doc issues." + }, + { + "left": "issue:2866", + "right": "issue:4386", + "accept": false, + "reason": "Both involve LoRA conversion, but one is bin-to-safetensors conversion and the other is an SDXL safetensor LoRA question." + }, + { + "left": "issue:2883", + "right": "issue:3838", + "accept": false, + "reason": "Image-to-video pipeline and Video Composer are broad feature requests, not the same concrete change." + }, + { + "left": "issue:2972", + "right": "issue:7365", + "accept": false, + "reason": "The 0.01 adjustment-loop bug is unrelated to pooled_prompt_embeds being overwritten." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 224, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:2984", + "issue:3093", + "issue:3111", + "issue:3113", + "issue:3171", + "issue:3221", + "issue:3338", + "issue:3406", + "issue:3444", + "issue:3728", + "issue:3817", + "issue:4052", + "issue:4264", + "issue:4311", + "issue:4312", + "issue:4386", + "issue:4524", + "issue:459", + "issue:4630", + "issue:4868", + "issue:4877", + "issue:4912", + "issue:5062", + "issue:5070", + "issue:5130", + "issue:5254", + "issue:5284", + "issue:5371", + "issue:5442", + "issue:5643", + "issue:5931", + "issue:6093", + "issue:6113", + "issue:6295", + "issue:6616", + "issue:6732", + "issue:692", + "issue:7147", + "issue:7264", + "issue:7377", + "issue:7440", + "issue:7456", + "issue:8410", + "issue:8451", + "issue:8495", + "issue:8792", + "issue:9001", + "issue:9793" + ], + "soft_pairs": [ + "issue:2984|issue:3817", + "issue:3093|issue:4630", + "issue:3111|issue:4912", + "issue:3111|issue:7264", + "issue:3113|issue:5442", + "issue:3171|issue:5070", + "issue:3221|issue:3338", + "issue:3406|issue:4311", + "issue:3406|issue:5062", + "issue:3444|issue:3728", + "issue:4052|issue:7440", + "issue:4264|issue:7456", + "issue:4312|issue:5371", + "issue:4386|issue:4877", + "issue:4524|issue:6616", + "issue:459|issue:692", + "issue:4868|issue:8495", + "issue:5130|issue:6732", + "issue:5254|issue:9001", + "issue:5284|issue:8410", + "issue:5643|issue:6113", + "issue:5931|issue:9793", + "issue:6093|issue:8451", + "issue:6295|issue:8792", + "issue:7147|issue:7377" + ], + "prepared_review_unit_hash": "3e021c1dca14cd8227729ef04a265755017fba3d368aa51187a5a1eb77645736", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33185, + "estimated_input_tokens": 8297, + "estimated_eval_tokens": 16850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated feature requests/questions across different Diffusers pipelines and utilities; the soft-similarity links are not strong enough to treat any pair as duplicates.", + "confidence": 0.34, + "canonical_issue_reason": "issue:4524 is the strongest representative issue here: it has the highest discussion activity and inbound references, and it is a concrete feature request rather than a narrow one-off question.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4524 is the best single issue to anchor the cluster because it is prominent, substantive, and clearly scoped to a real product enhancement.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2984", + "right": "issue:3817", + "accept": false, + "reason": "Unrelated topics: long-video diffusion model research vs editable dance generation from music." + }, + { + "left": "issue:3093", + "right": "issue:4630", + "accept": false, + "reason": "Noise manipulation during generation is different from changing LoRA weights after loading." + }, + { + "left": "issue:3111", + "right": "issue:4912", + "accept": false, + "reason": "Custom conditioning encoder support is unrelated to converting SDXL checkpoints." + }, + { + "left": "issue:3111", + "right": "issue:7264", + "accept": false, + "reason": "Custom conditioning encoder support is a different problem from serializing/loading custom Transformers models." + }, + { + "left": "issue:3113", + "right": "issue:5442", + "accept": false, + "reason": "Cross-attention layout control and ScaleCrafter are different feature requests for different code paths." + }, + { + "left": "issue:3171", + "right": "issue:5070", + "accept": false, + "reason": "Both mention ControlNet-style prompting, but one is about LPW prompts and the other about specific control flags in an example; not the same bug/change." + }, + { + "left": "issue:3221", + "right": "issue:3338", + "accept": false, + "reason": "A version-specific attention processor loading bug is unrelated to slow package import time." + }, + { + "left": "issue:3406", + "right": "issue:4311", + "accept": false, + "reason": "only_cross_attention behavior is unrelated to the local_files_only pipeline_class bug." + }, + { + "left": "issue:3406", + "right": "issue:5062", + "accept": false, + "reason": "Attention behavior for only_cross_attention is different from a typo in attention processor naming." + }, + { + "left": "issue:3444", + "right": "issue:3728", + "accept": false, + "reason": "Audio-to-image integration is unrelated to shifted diffusion text-to-image support." + }, + { + "left": "issue:4052", + "right": "issue:7440", + "accept": false, + "reason": "Missing from_single_file import and a missing dataset are unrelated issues." + }, + { + "left": "issue:4264", + "right": "issue:7456", + "accept": false, + "reason": "Missing example images is not the same as a missing raise statement." + }, + { + "left": "issue:4312", + "right": "issue:5371", + "accept": false, + "reason": "SAG for inpainting/img2img and BLIP inpainting support are different pipeline feature requests." + }, + { + "left": "issue:4386", + "right": "issue:4877", + "accept": false, + "reason": "LoRA conversion for SDXL is unrelated to exposing noise_init_strength as an input argument." + }, + { + "left": "issue:4524", + "right": "issue:6616", + "accept": false, + "reason": "AnimateDiff pipeline support and classifier guidance are different concrete features affecting different generation paths." + }, + { + "left": "issue:459", + "right": "issue:692", + "accept": false, + "reason": "Incremental diffusion and an example for fine-tuning are not the same underlying issue." + }, + { + "left": "issue:4868", + "right": "issue:8495", + "accept": false, + "reason": "A Python AttributeError in one path is unrelated to an SD3 tokenizer enum deserialization error." + }, + { + "left": "issue:5130", + "right": "issue:6732", + "accept": false, + "reason": "A scheduler usage question is not the same as undocumented trained_betas documentation." + }, + { + "left": "issue:5254", + "right": "issue:9001", + "accept": false, + "reason": "Multi-ControlNet inpaint support and Auraflow aspect ratios are separate pipeline enhancements." + }, + { + "left": "issue:5284", + "right": "issue:8410", + "accept": false, + "reason": "ControlNet-LLLite support and native TextDiffuser support are different model/pipeline additions." + }, + { + "left": "issue:5643", + "right": "issue:6113", + "accept": false, + "reason": "How to use IP-Adapter ControlNet is unrelated to loading models from the sd_control_collection repo." + }, + { + "left": "issue:5931", + "right": "issue:9793", + "accept": false, + "reason": "Generation nondeterminism and a progress bar update issue are different bugs." + }, + { + "left": "issue:6093", + "right": "issue:8451", + "accept": false, + "reason": "push_to_hub behavior in one training script is unrelated to missing training details for frozen UNet parameters." + }, + { + "left": "issue:6295", + "right": "issue:8792", + "accept": false, + "reason": "Artifacts with a specific sampler setting are not the same as general quality degradation migrating from A1111 to Diffusers." + }, + { + "left": "issue:7147", + "right": "issue:7377", + "accept": false, + "reason": "Import failure for StableCascadeDecoderPipeline is different from a runtime error when num_images_per_prompt > 1." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 225, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10204", + "issue:10382", + "issue:10447", + "issue:10584", + "issue:10737", + "issue:10833", + "issue:10928", + "issue:11010", + "issue:11109", + "issue:11223", + "issue:11258", + "issue:11329", + "issue:11363", + "issue:11570", + "issue:12338", + "issue:12656", + "issue:12981", + "issue:1815", + "issue:3177", + "issue:3748", + "issue:3765", + "issue:3880", + "issue:4229", + "issue:4520", + "issue:4631", + "issue:5313", + "issue:5710", + "issue:7237", + "issue:7322", + "issue:782", + "issue:7897", + "issue:8593", + "issue:8831", + "issue:9026", + "issue:9051", + "issue:9170", + "issue:9178", + "issue:9298", + "issue:9331", + "issue:9417", + "issue:9454", + "issue:9460", + "issue:9479", + "issue:9496", + "issue:954", + "issue:9637", + "issue:9902" + ], + "soft_pairs": [ + "issue:7237|issue:9902", + "issue:8593|issue:9454", + "issue:9026|issue:9298", + "issue:9331|issue:9417", + "issue:12338|issue:9637", + "issue:5313|issue:8831", + "issue:1815|issue:3177", + "issue:11329|issue:11570", + "issue:4229|issue:7322", + "issue:10584|issue:9331", + "issue:10447|issue:10928", + "issue:10204|issue:10833", + "issue:11109|issue:9178", + "issue:10737|issue:9051", + "issue:7897|issue:9170", + "issue:3765|issue:3880", + "issue:9460|issue:9496", + "issue:4631|issue:5710", + "issue:782|issue:954", + "issue:3748|issue:4520", + "issue:11223|issue:11363", + "issue:11010|issue:9479", + "issue:11258|issue:12981", + "issue:10382|issue:12656" + ], + "prepared_review_unit_hash": "c2203eb4bf353f7c8ed3d9fdf922062bd14ae063a68d9a285902cb69d4061c65", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33359, + "estimated_input_tokens": 8340, + "estimated_eval_tokens": 16936 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy and spans many unrelated bug reports and feature requests. The inpainting trio is the strongest thematic group, with #3880 as the best umbrella issue. For soft edges, only the DPMSolverSinglestepScheduler pair looks like a real duplicate-style match; the FlowMatch `index_for_timestep` pair should be rejected because one is a bug report and the other is a performance suggestion.", + "confidence": 0.79, + "canonical_issue_reason": "#3880 is the clearest broad inpainting-defect report and works best as the umbrella for related masked-area degradation complaints, including #9479 and the later FluxFill inpainting quality issue.", + "canonical_pr_reason": null, + "best_issue_reason": "#3880 is the most representative standalone bug in the set: it describes a concrete inpainting degradation problem with broad applicability.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7237", + "right": "issue:9902", + "accept": false, + "reason": "Unrelated: missing scheduler class vs a conda version pin/stuck version issue." + }, + { + "left": "issue:8593", + "right": "issue:9454", + "accept": false, + "reason": "Different bugs: SD3 CPU bfloat16 noise vs CogVideoX image2video model loading." + }, + { + "left": "issue:9026", + "right": "issue:9298", + "accept": false, + "reason": "Different failure modes: unexpected `force_hook` arg vs missing `Float8_e4m3fnStorage`." + }, + { + "left": "issue:9331", + "right": "issue:9417", + "accept": false, + "reason": "Same function name, but one is a correctness bug report and the other is a performance suggestion; not the same underlying defect." + }, + { + "left": "issue:12338", + "right": "issue:9637", + "accept": false, + "reason": "Different runtime issues: AutoencoderDC slicing decode bug vs xFormers attention-mask/bfloat16 failure." + }, + { + "left": "issue:5313", + "right": "issue:8831", + "accept": false, + "reason": "Both are LoRA-related, but they describe different problems: non-revertible fuse vs unsupported layers." + }, + { + "left": "issue:1815", + "right": "issue:3177", + "accept": false, + "reason": "Both involve DeepSpeed, but the reported training/offloading failures are different." + }, + { + "left": "issue:11329", + "right": "issue:11570", + "accept": false, + "reason": "Shared quantization context, but memory leak and `.to()` cast breakage are distinct bugs." + }, + { + "left": "issue:4229", + "right": "issue:7322", + "accept": false, + "reason": "Both are training failures, but the contexts and failure causes differ." + }, + { + "left": "issue:10584", + "right": "issue:9331", + "accept": false, + "reason": "Related scheduler area, but the specific reports do not clearly describe the same concrete bug." + }, + { + "left": "issue:10447", + "right": "issue:10928", + "accept": false, + "reason": "Model file handling is only a broad theme; conversion splitting and missing fp16 weights are different issues." + }, + { + "left": "issue:10204", + "right": "issue:10833", + "accept": false, + "reason": "Same pipeline family, but iterative inpainting degradation and quantization shape mismatch are different bugs." + }, + { + "left": "issue:11109", + "right": "issue:9178", + "accept": false, + "reason": "Different LoRA problems: transformer layer mismatch vs PEFT target-module errors." + }, + { + "left": "issue:10737", + "right": "issue:9051", + "accept": false, + "reason": "Both are FluxPipeline issues, but missing image encoder and lack of batch inference are different problems." + }, + { + "left": "issue:7897", + "right": "issue:9170", + "accept": false, + "reason": "Both mention resource pressure, but they are not the same underlying bug." + }, + { + "left": "issue:3765", + "right": "issue:3880", + "accept": false, + "reason": "Different tasks and symptoms: ControlNet training stacking effects vs inpainting mask degradation." + }, + { + "left": "issue:9460", + "right": "issue:9496", + "accept": false, + "reason": "Feature request to assign multiple ControlNet models vs a forward-function error." + }, + { + "left": "issue:4631", + "right": "issue:5710", + "accept": true, + "reason": "Same scheduler class and same sigma initialization/code-path problem; these look like the same underlying bug." + }, + { + "left": "issue:782", + "right": "issue:954", + "accept": false, + "reason": "Both are auth/download-related, but they are not the same failure." + }, + { + "left": "issue:3748", + "right": "issue:4520", + "accept": false, + "reason": "Both involve LoRA, but dtype mismatch on load and attention-slicing scale behavior are different bugs." + }, + { + "left": "issue:11223", + "right": "issue:11363", + "accept": false, + "reason": "OOM reports in different contexts without evidence of the same root cause." + }, + { + "left": "issue:11010", + "right": "issue:9479", + "accept": false, + "reason": "Architecture-support request vs a stable-diffusion inpainting bug; not the same issue." + }, + { + "left": "issue:11258", + "right": "issue:12981", + "accept": false, + "reason": "Same LTX family, but different versions/formats and different loading problems." + }, + { + "left": "issue:10382", + "right": "issue:12656", + "accept": false, + "reason": "Completely unrelated: VAE return-type mismatch vs AuraFlow positional embedding out-of-bounds." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Rejecting because one soft-edge verdict overstates the evidence: issue #9331 and #9417 share a function name, but the packet shows a bug report versus a performance suggestion, not the same underlying defect. The rest of the pairing judgments are mostly conservative, and the inpainting umbrella choice is plausible, but this accept is too strong for the available evidence." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 226, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10111", + "issue:10231", + "issue:10897", + "issue:11592", + "issue:1185", + "issue:12108", + "issue:12605", + "issue:12653", + "issue:13377", + "issue:1417", + "issue:1708", + "issue:1857", + "issue:1934", + "issue:2507", + "issue:2532", + "issue:2784", + "issue:2906", + "issue:2984", + "issue:337", + "issue:3444", + "issue:3728", + "issue:3769", + "issue:4390", + "issue:4480", + "issue:6112", + "issue:6256", + "issue:6385", + "issue:6506", + "issue:6533", + "issue:6901", + "issue:7082", + "issue:7380", + "issue:7492", + "issue:775", + "issue:834", + "issue:8410", + "issue:8593", + "issue:8604", + "issue:8832", + "issue:9023", + "issue:9171", + "issue:9209", + "issue:9320", + "issue:9460", + "issue:9516", + "issue:9610", + "issue:9719", + "issue:984" + ], + "soft_pairs": [ + "issue:12605|issue:13377", + "issue:6385|issue:6506", + "issue:2906|issue:7082", + "issue:6256|issue:9610", + "issue:2507|issue:834", + "issue:11592|issue:8832", + "issue:3728|issue:6533", + "issue:1934|issue:984", + "issue:7492|issue:9209", + "issue:12108|issue:9460", + "issue:12108|issue:12653", + "issue:10111|issue:9516", + "issue:2784|issue:7380", + "issue:4480|issue:8604", + "issue:1185|issue:1708", + "issue:337|issue:775", + "issue:10231|issue:9023", + "issue:6112|issue:8593", + "issue:9320|issue:9719", + "issue:10897|issue:8410", + "issue:1417|issue:1857", + "issue:2984|issue:3444", + "issue:9171|issue:9719", + "issue:4390|issue:6901", + "issue:2532|issue:3769" + ], + "prepared_review_unit_hash": "c1d21b56e926fe5d4f8834a7b6db046c18097fa27bab2dbb1a57cd85373e9f60", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34126, + "estimated_input_tokens": 8532, + "estimated_eval_tokens": 17320 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: the soft matches are broad topical similarities, not true duplicates. The only clear duplicate-like pair is the StableDiffusionPanoramaPipeline get_views / quartered-image bug.", + "confidence": 0.79, + "canonical_issue_reason": "Issue 2532 is the best root-cause anchor: it names the faulty `get_views` method, while 3769 is the downstream symptom report for the same panorama bug.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 2532 is the most actionable representative in the set because it points to the underlying code path rather than only the visual failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12605", + "right": "issue:13377", + "accept": false, + "reason": "Both are attention/mask-related FLUX bugs, but they affect different code paths: backend attention dispatch vs CFG being skipped when `mask` is `None`." + }, + { + "left": "issue:6385", + "right": "issue:6506", + "accept": false, + "reason": "Same broad StableVideoDiffusion area, but one is VRAM usage and the other is output type mismatch; not the same bug." + }, + { + "left": "issue:2906", + "right": "issue:7082", + "accept": false, + "reason": "Both are import/type-related failures, but they concern unrelated symbols (`PreTrainedTokenizer` vs `CpuOffload`) and different subsystems." + }, + { + "left": "issue:6256", + "right": "issue:9610", + "accept": false, + "reason": "Both are pipeline API feature requests, but they ask for different callbacks/methods and do not describe the same change." + }, + { + "left": "issue:2507", + "right": "issue:834", + "accept": false, + "reason": "Both mention performance, but one is safetensors load speed and the other is CPU underutilization during runtime; different problems." + }, + { + "left": "issue:11592", + "right": "issue:8832", + "accept": false, + "reason": "One is a FLUX LoRA adapter regression; the other is a generic crash when calling `pipe(prompt)`. No concrete shared code path." + }, + { + "left": "issue:3728", + "right": "issue:6533", + "accept": false, + "reason": "Both are requests for new pipelines, but they target different models/tasks and are not the same feature." + }, + { + "left": "issue:1934", + "right": "issue:984", + "accept": false, + "reason": "Unrelated performance/memory complaints in different code paths; no evidence of a shared underlying defect." + }, + { + "left": "issue:7492", + "right": "issue:9209", + "accept": false, + "reason": "Both involve loading models, but one is a missing repo error and the other is a network/mirror connectivity issue." + }, + { + "left": "issue:12108", + "right": "issue:9460", + "accept": false, + "reason": "Different pipelines and different failures: flow-matching scheduler breakage vs assigning multiple ControlNet models." + }, + { + "left": "issue:12108", + "right": "issue:12653", + "accept": false, + "reason": "Completely different problems: scheduler compatibility vs RoPE dtype causing MPS/NPU issues." + }, + { + "left": "issue:10111", + "right": "issue:9516", + "accept": false, + "reason": "Both mention FLUX, but one is a vague multi-bug report and the other is a specific missing `joint_attention_kwargs` pass-through; not the same bug." + }, + { + "left": "issue:2784", + "right": "issue:7380", + "accept": false, + "reason": "Both are feature requests, but they propose different diffusion capabilities and are not duplicates." + }, + { + "left": "issue:4480", + "right": "issue:8604", + "accept": false, + "reason": "Both are fp16/MPS-related image corruption reports, but they involve different models/components and different failure modes." + }, + { + "left": "issue:1185", + "right": "issue:1708", + "accept": false, + "reason": "Dreambooth is the only shared topic; one is reproducibility and the other is config-file usage. Not the same issue." + }, + { + "left": "issue:337", + "right": "issue:775", + "accept": false, + "reason": "Both are generation bugs, but one concerns img2img/inpainting behavior and the other fp16 without autocast." + }, + { + "left": "issue:10231", + "right": "issue:9023", + "accept": false, + "reason": "Both are generic run failures, but they are too unspecific and do not point to the same underlying problem." + }, + { + "left": "issue:6112", + "right": "issue:8593", + "accept": false, + "reason": "Both produce noise, but on different models and precisions; the bug reports do not indicate the same root cause." + }, + { + "left": "issue:9320", + "right": "issue:9719", + "accept": false, + "reason": "Saved/loaded finetuned weights in custom diffusion is unrelated to a progress-bar suppression regression." + }, + { + "left": "issue:10897", + "right": "issue:8410", + "accept": false, + "reason": "Both are feature requests for new model support, but they are about different pipelines/tasks." + }, + { + "left": "issue:1417", + "right": "issue:1857", + "accept": false, + "reason": "CPU offloading implementation concerns are unrelated to UnCLIP test coverage." + }, + { + "left": "issue:2984", + "right": "issue:3444", + "accept": false, + "reason": "Both request new pipeline/model support, but they are for different modalities and concrete changes." + }, + { + "left": "issue:9171", + "right": "issue:9719", + "accept": false, + "reason": "A single-file loading regression for SD 1.5 `v_prediction` is unrelated to `disable_progress_bar` being ignored." + }, + { + "left": "issue:4390", + "right": "issue:6901", + "accept": false, + "reason": "Missing requirements vs missing `clip_sample` implementation: both are 'missing' in name only, not the same defect." + }, + { + "left": "issue:2532", + "right": "issue:3769", + "accept": true, + "reason": "These describe the same StableDiffusionPanoramaPipeline bug: a `get_views` defect causing the quartered/incorrect panorama output." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The evaluation is grounded in the packet and stays conservative: the rejected soft pairs are only matched on broad topical similarity, while issue:2532 vs issue:3769 is a plausible same-bug pair based on the panorama/get_views titles. No overstatement that would require rejection." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 227, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10111", + "issue:10518", + "issue:10760", + "issue:10833", + "issue:10883", + "issue:10893", + "issue:11144", + "issue:11283", + "issue:11352", + "issue:11547", + "issue:11555", + "issue:11674", + "issue:12110", + "issue:12157", + "issue:12221", + "issue:1231", + "issue:12547", + "issue:13221", + "issue:13261", + "issue:2016", + "issue:2319", + "issue:2532", + "issue:3576", + "issue:4080", + "issue:4141", + "issue:4492", + "issue:4639", + "issue:5006", + "issue:5028", + "issue:5118", + "issue:5128", + "issue:5402", + "issue:5441", + "issue:5463", + "issue:5749", + "issue:5866", + "issue:5947", + "issue:6095", + "issue:6281", + "issue:7629", + "issue:8551", + "issue:9031", + "issue:9038", + "issue:9551", + "issue:9576", + "issue:9619" + ], + "soft_pairs": [ + "issue:4492|issue:5402", + "issue:11555|issue:8551", + "issue:1231|issue:2016", + "issue:10883|issue:9551", + "issue:5118|issue:5866", + "issue:5749|issue:6095", + "issue:10100|issue:12221", + "issue:10893|issue:9619", + "issue:13221|issue:13261", + "issue:4141|issue:6281", + "issue:11144|issue:9576", + "issue:11283|issue:11674", + "issue:5128|issue:5463", + "issue:12157|issue:12547", + "issue:10111|issue:10518", + "issue:10760|issue:12110", + "issue:4080|issue:5947", + "issue:9031|issue:9038", + "issue:5006|issue:7629", + "issue:10893|issue:11547", + "issue:4639|issue:5441", + "issue:2319|issue:2532", + "issue:3576|issue:5028", + "issue:10833|issue:11352" + ], + "prepared_review_unit_hash": "8ad9cc2dee261c6ba8d915d03a182bfc4aea49119f86cbb1e0b62b93c6ac57b8", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33512, + "estimated_input_tokens": 8378, + "estimated_eval_tokens": 17012 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: it mixes unrelated feature requests, import errors, training-script bugs, and pipeline-specific regressions. None of the soft pairs look like the same underlying issue or change, so all soft edges are rejected.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4492", + "right": "issue:5402", + "accept": false, + "reason": "Different problems: a JSON/boolean typo (`false` vs `False`) versus a missing `compute_snr` import. No shared code path." + }, + { + "left": "issue:11555", + "right": "issue:8551", + "accept": false, + "reason": "Both mention `device_map`, but one is a broad feature question and the other is a specific VAE device-placement bug in an inpaint pipeline." + }, + { + "left": "issue:1231", + "right": "issue:2016", + "accept": false, + "reason": "Both are Flax-related, but one is a scheduler request and the other is a sampling regression. Not the same bug/change." + }, + { + "left": "issue:10883", + "right": "issue:9551", + "accept": false, + "reason": "Different feature requests in different model families; no shared concrete implementation target." + }, + { + "left": "issue:5118", + "right": "issue:5866", + "accept": false, + "reason": "Both touch ControlNet usage, but one is about reference-only support and the other is AnimateDiff+ControlNet integration. Different code paths." + }, + { + "left": "issue:5749", + "right": "issue:6095", + "accept": false, + "reason": "Both are IndexErrors, but they affect different pipelines and likely different scheduler logic. Too broad to merge." + }, + { + "left": "issue:10100", + "right": "issue:12221", + "accept": false, + "reason": "Both are feature requests, but for unrelated models/integrations. No duplicate underlying change." + }, + { + "left": "issue:10893", + "right": "issue:9619", + "accept": false, + "reason": "One requests CogVideoX ControlNet support; the other requests interrupt support for StableDiffusionControlNetPipeline. Unrelated changes." + }, + { + "left": "issue:13221", + "right": "issue:13261", + "accept": false, + "reason": "Both mention LoRA, but one is a general support issue and the other is a specific FLUX.2-klein-9B LoRA/Lokr compatibility problem." + }, + { + "left": "issue:4141", + "right": "issue:6281", + "accept": false, + "reason": "One is a MultiControlNet behavior bug; the other is an import error for `MultiControlNetModel`. Related area, but not the same defect." + }, + { + "left": "issue:11144", + "right": "issue:9576", + "accept": false, + "reason": "Different concerns: dtype initialization in Flax UNet versus `use_safetensors` propagation when loading pipelines." + }, + { + "left": "issue:11283", + "right": "issue:11674", + "accept": false, + "reason": "Same Wan ecosystem, but one is a deserialization failure and the other is a feature request for VACE inputs. Not the same issue." + }, + { + "left": "issue:5128", + "right": "issue:5463", + "accept": false, + "reason": "Unrelated errors: loading a local safetensors file versus an `image_latents` uninitialized variable." + }, + { + "left": "issue:12157", + "right": "issue:12547", + "accept": false, + "reason": "Both are training-time errors, but caused by different scripts and different root causes (`torch.device` iterable bug vs missing `--instance_prompt`)." + }, + { + "left": "issue:10111", + "right": "issue:10518", + "accept": false, + "reason": "Different pipelines and different bug reports; no evidence they share the same root cause." + }, + { + "left": "issue:10760", + "right": "issue:12110", + "accept": false, + "reason": "Different code paths and error types: import failure for an ONNX pipeline class versus UNet1D initialization bug." + }, + { + "left": "issue:4080", + "right": "issue:5947", + "accept": false, + "reason": "A new SDXL inpainting pipeline request is unrelated to a Diffusion-DPO policy objective request." + }, + { + "left": "issue:9031", + "right": "issue:9038", + "accept": false, + "reason": "Both are Flax feature questions, but one is about disabling the safety checker and the other about prompt weighting. Separate features." + }, + { + "left": "issue:5006", + "right": "issue:7629", + "accept": false, + "reason": "Both involve OOM, but in different models, versions, and circumstances. Not a single concrete bug." + }, + { + "left": "issue:10893", + "right": "issue:11547", + "accept": false, + "reason": "Both reference ControlNet, but they target different model families and different feature requests." + }, + { + "left": "issue:4639", + "right": "issue:5441", + "accept": false, + "reason": "Different training scripts and failures: dataloader multiprocessing versus missing `attn_processors` on WuerstchenPrior." + }, + { + "left": "issue:2319", + "right": "issue:2532", + "accept": false, + "reason": "A typo fix in one method is unrelated to a panorama pipeline `get_views` bug." + }, + { + "left": "issue:3576", + "right": "issue:5028", + "accept": false, + "reason": "Both are TypeErrors, but caused by different mechanisms and pipeline implementations. Not mergeable as one fix." + }, + { + "left": "issue:10833", + "right": "issue:11352", + "accept": false, + "reason": "Both involve bitsandbytes quantization, but one is a matrix-shape runtime failure and the other is a missing CPU-move warning. Different outcomes and fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 228, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10302", + "issue:10467", + "issue:10757", + "issue:11045", + "issue:11721", + "issue:12308", + "issue:1304", + "issue:1335", + "issue:1434", + "issue:1519", + "issue:1672", + "issue:1963", + "issue:3301", + "issue:3624", + "issue:4331", + "issue:4349", + "issue:4436", + "issue:463", + "issue:4880", + "issue:5112", + "issue:5392", + "issue:5465", + "issue:5474", + "issue:5873", + "issue:6054", + "issue:6087", + "issue:6095", + "issue:6162", + "issue:6283", + "issue:6323", + "issue:7049", + "issue:7138", + "issue:7611", + "issue:7682", + "issue:785", + "issue:8523", + "issue:8565", + "issue:8708", + "issue:8744", + "issue:8939", + "issue:8997", + "issue:9147", + "issue:9297", + "issue:9397", + "issue:9505", + "issue:9713", + "issue:9902", + "issue:9941" + ], + "soft_pairs": [ + "issue:11045|issue:11721", + "issue:1963|issue:5392", + "issue:1519|issue:785", + "issue:8523|issue:8997", + "issue:3301|issue:4331", + "issue:10757|issue:8939", + "issue:1672|issue:6087", + "issue:10302|issue:10467", + "issue:1335|issue:3624", + "issue:4880|issue:6054", + "issue:5474|issue:9713", + "issue:5873|issue:9902", + "issue:5112|issue:6283", + "issue:6095|issue:8565", + "issue:6162|issue:7611", + "issue:1434|issue:4349", + "issue:6323|issue:9505", + "issue:12308|issue:9397", + "issue:1304|issue:463", + "issue:4436|issue:8708", + "issue:5465|issue:7049", + "issue:7138|issue:8744", + "issue:9147|issue:9941", + "issue:7682|issue:9297" + ], + "prepared_review_unit_hash": "248dc76b6833d784100e959c26b24b49eb928e510e30b122b36e0c0ff77f927c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34590, + "estimated_input_tokens": 8648, + "estimated_eval_tokens": 17552 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated issues across different pipelines, training scripts, schedulers, conversion utilities, and model families. I do not see a true duplicate set here, so all soft pairs are rejected.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical issue stands out because the items are not the same underlying bug; they cover distinct failures and feature requests across different code paths.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11045 is the most concrete, self-contained bug report in the set: it names a specific crash, a specific model, and a specific training script, making it a good representative issue if one must be chosen.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11045", + "right": "issue:11721", + "accept": false, + "reason": "Both are load-time crashes, but they involve different pipelines/models (Flux Schnell vs SanaPipeline) and different failure contexts." + }, + { + "left": "issue:1963", + "right": "issue:5392", + "accept": false, + "reason": "Same broad topic of unconditional latent diffusion, but one is a generation issue and the other is a how-to/training question; not the same bug." + }, + { + "left": "issue:1519", + "right": "issue:785", + "accept": false, + "reason": "No overlap in underlying problem: noisy batch outputs on Apple M2 vs scheduler performance cost." + }, + { + "left": "issue:8523", + "right": "issue:8997", + "accept": false, + "reason": "Different failures in different models/code paths; only a superficial \u201cCUDA error / scheduler error\u201d similarity." + }, + { + "left": "issue:3301", + "right": "issue:4331", + "accept": false, + "reason": "Both involve DreamBooth/LoRA training, but one is a generic failure report and the other is a behavioral discrepancy between two scripts." + }, + { + "left": "issue:10757", + "right": "issue:8939", + "accept": false, + "reason": "Unrelated: variable-name fix in a conversion script vs tokenizer loading failure for an SD3 model." + }, + { + "left": "issue:1672", + "right": "issue:6087", + "accept": false, + "reason": "Different subsystems and symptoms: checkpoint conversion image size mismatch vs LoRA config persistence causing wrong lora_alpha." + }, + { + "left": "issue:10302", + "right": "issue:10467", + "accept": false, + "reason": "FP8 inference noise is a runtime quality bug; the other appears to be about example-code/file-name output and is not the same issue." + }, + { + "left": "issue:1335", + "right": "issue:3624", + "accept": false, + "reason": "Different TypeErrors in different code paths; no evidence they share the same underlying defect." + }, + { + "left": "issue:4880", + "right": "issue:6054", + "accept": false, + "reason": "Different pipeline bugs: ControlNet image validation vs Euler scheduler timestep handling." + }, + { + "left": "issue:5474", + "right": "issue:9713", + "accept": false, + "reason": "A proposed new pipeline vs a controlnet-training implementation request; not duplicate bugs." + }, + { + "left": "issue:5873", + "right": "issue:9902", + "accept": false, + "reason": "ONNX export failure is unrelated to the conda version/package-management issue." + }, + { + "left": "issue:5112", + "right": "issue:6283", + "accept": false, + "reason": "Both are SDXL ControlNet img2img issues, but one is a float/shape TypeError and the other is a strength=0 edge-case; not clearly the same bug." + }, + { + "left": "issue:6095", + "right": "issue:8565", + "accept": false, + "reason": "Both mention IndexError, but they affect different pipelines and code paths (SVD scheduling vs SD3 LoRA loading)." + }, + { + "left": "issue:6162", + "right": "issue:7611", + "accept": false, + "reason": "Both are tensor/shape mismatch reports, but they concern different models and different components; too broad to merge." + }, + { + "left": "issue:1434", + "right": "issue:4349", + "accept": false, + "reason": "Same upscale pipeline family, but one is an OOM scaling issue and the other is a missing text_encoder compatibility bug." + }, + { + "left": "issue:6323", + "right": "issue:9505", + "accept": false, + "reason": "Completely different problems: a TensorRT attribute-setting issue vs a checkpoint key mismatch when upscaling." + }, + { + "left": "issue:12308", + "right": "issue:9397", + "accept": false, + "reason": "Different feature requests in unrelated areas: flash-attn support vs scheduler support for Hyper LoRAs." + }, + { + "left": "issue:1304", + "right": "issue:463", + "accept": false, + "reason": "A training-how-to question and a super-resolution model request are not duplicates." + }, + { + "left": "issue:4436", + "right": "issue:8708", + "accept": false, + "reason": "Both involve VAE/diffusers context, but one is conversion from civitai safetensors and the other is a missing SD3 training config field." + }, + { + "left": "issue:5465", + "right": "issue:7049", + "accept": false, + "reason": "Different pipeline bugs with different failure modes; no shared concrete code-path problem." + }, + { + "left": "issue:7138", + "right": "issue:8744", + "accept": false, + "reason": "SDXL UNet structure mismatch vs single-file loading for training are unrelated." + }, + { + "left": "issue:9147", + "right": "issue:9941", + "accept": false, + "reason": "Flux autopipeline error and a generic Colab stable-diffusion error do not describe the same defect." + }, + { + "left": "issue:7682", + "right": "issue:9297", + "accept": false, + "reason": "One is a multi-GPU runtime issue; the other is a LoRA loading performance complaint." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 229, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10025", + "issue:10282", + "issue:10334", + "issue:10336", + "issue:10712", + "issue:10896", + "issue:10952", + "issue:10954", + "issue:1100", + "issue:11050", + "issue:11137", + "issue:11169", + "issue:11222", + "issue:11422", + "issue:11519", + "issue:11721", + "issue:11724", + "issue:11779", + "issue:11907", + "issue:12195", + "issue:12771", + "issue:13300", + "issue:2294", + "issue:2444", + "issue:2606", + "issue:3338", + "issue:3651", + "issue:4043", + "issue:4116", + "issue:4349", + "issue:4399", + "issue:5034", + "issue:5523", + "issue:5897", + "issue:5934", + "issue:5947", + "issue:6351", + "issue:6866", + "issue:7185", + "issue:8482", + "issue:8885", + "issue:9248", + "issue:9374", + "issue:9393", + "issue:9471", + "issue:9516", + "issue:9958" + ], + "soft_pairs": [ + "issue:11422|issue:9393", + "issue:3651|issue:5523", + "issue:1100|issue:2444", + "issue:10952|issue:9471", + "issue:10282|issue:11169", + "issue:10896|issue:11519", + "issue:2294|issue:3338", + "issue:4116|issue:7185", + "issue:5947|issue:8482", + "issue:9374|issue:9958", + "issue:4043|issue:4349", + "issue:11721|issue:8885", + "issue:10334|issue:10336", + "issue:11907|issue:12771", + "issue:2606|issue:5897", + "issue:5034|issue:6351", + "issue:4399|issue:5934", + "issue:10025|issue:9516", + "issue:10712|issue:11050", + "issue:10954|issue:11724", + "issue:11137|issue:13300", + "issue:11222|issue:11779", + "issue:12195|issue:9248", + "issue:6866|issue:9958" + ], + "prepared_review_unit_hash": "c86a7d74b193c3fe1e73c30cef2ed60c371e92a32cedf320ed033856c0be3390", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33945, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly soft-similarity false positives: the linked issues cover unrelated bugs, feature requests, environment/import problems, and different model pipelines. None of the candidate pairs look like the same underlying defect/change, so no duplicates should be merged.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11422", + "right": "issue:9393", + "accept": false, + "reason": "Both mention LoRA/training, but one is a validation-prompt crash in HiDream LoRA training while the other is a Flux DreamBooth LoRA save failure; different pipelines and failure modes." + }, + { + "left": "issue:3651", + "right": "issue:5523", + "accept": false, + "reason": "Multi-GPU train_dreambooth failure is unrelated to run-to-run inconsistency; one is a launch/runtime bug, the other is nondeterminism." + }, + { + "left": "issue:1100", + "right": "issue:2444", + "accept": false, + "reason": "CLIPTextTransformer device attribute error is a model-object bug, while the other is a generic subprocess exit in inpaint training; no same code path." + }, + { + "left": "issue:10952", + "right": "issue:9471", + "accept": false, + "reason": "LoRA adapter-name lookup failure is unrelated to file-extension inference on a model path." + }, + { + "left": "issue:10282", + "right": "issue:11169", + "accept": false, + "reason": "Checkpoint shape mismatch for x_embedder.weight is unrelated to the NumPy binary-incompatibility import error." + }, + { + "left": "issue:10896", + "right": "issue:11519", + "accept": false, + "reason": "These are separate model-support requests for different models, not the same change." + }, + { + "left": "issue:2294", + "right": "issue:3338", + "accept": false, + "reason": "Missing pipeline import and slow package import are different issues with different symptoms and fixes." + }, + { + "left": "issue:4116", + "right": "issue:7185", + "accept": false, + "reason": "SDXL embedding-passing bug is unrelated to PyTorch version support question." + }, + { + "left": "issue:5947", + "right": "issue:8482", + "accept": false, + "reason": "Policy-objective discussion is unrelated to adding support for a specific model." + }, + { + "left": "issue:9374", + "right": "issue:9958", + "accept": false, + "reason": "AutoPipeline FLUX support and Autoencoder tiling support touch different features and code paths." + }, + { + "left": "issue:4043", + "right": "issue:4349", + "accept": false, + "reason": "Both involve embeddings in SDXL-family pipelines, but one is about requiring pooled embeddings and the other about omitting text_encoder; not the same concrete bug." + }, + { + "left": "issue:11721", + "right": "issue:8885", + "accept": false, + "reason": "SanaPipeline load crash is unrelated to clip-retrieval installation/modifier-token loading." + }, + { + "left": "issue:10334", + "right": "issue:10336", + "accept": false, + "reason": "MacOS/MPS gray images and CPU NaNs versus activation explosion/clamping are related at best, but not clearly the same defect." + }, + { + "left": "issue:11907", + "right": "issue:12771", + "accept": false, + "reason": "Scheduler incompatibility in FLUX with two schedulers is broader and different from the specific DPM solver out-of-bounds bug with Karras sigmas." + }, + { + "left": "issue:2606", + "right": "issue:5897", + "accept": false, + "reason": "DeepSpeed save_state hang and tensor device mismatch are distinct training failures." + }, + { + "left": "issue:5034", + "right": "issue:6351", + "accept": false, + "reason": "Checkpoint progress-bar behavior is unrelated to DDP missing disable_adapters." + }, + { + "left": "issue:4399", + "right": "issue:5934", + "accept": false, + "reason": "Both mention torch.compile, but one is specifically compile plus CPU offload, while the other is a general compile error." + }, + { + "left": "issue:10025", + "right": "issue:9516", + "accept": false, + "reason": "Both are FLUX-transformer keyword/attention topics, but one is about attention-mask support and the other about joint_attention_kwargs not being forwarded; not the same bug." + }, + { + "left": "issue:10712", + "right": "issue:11050", + "accept": false, + "reason": "RuntimeError using prompt_embeds in SD3 inference is different from bf16/fp16 dtype casting in a training script." + }, + { + "left": "issue:10954", + "right": "issue:11724", + "accept": false, + "reason": "A FLUX LoRA loading failure is unrelated to Chroma attention-mask behavior in the transformer." + }, + { + "left": "issue:11137", + "right": "issue:13300", + "accept": false, + "reason": "A Latte dtype mismatch and CPU failures in SD3.5/FLUX are different models and likely different root causes." + }, + { + "left": "issue:11222", + "right": "issue:11779", + "accept": false, + "reason": "WAN LoRA support request and Wan VACE mask-padding bug are different scope: feature request vs concrete pipeline bug." + }, + { + "left": "issue:12195", + "right": "issue:9248", + "accept": false, + "reason": "PyTorch custom_op compatibility failure is unrelated to a pip install/upgrade instruction issue." + }, + { + "left": "issue:6866", + "right": "issue:9958", + "accept": false, + "reason": "IP-Adapter support for an SDXL ControlNet inpaint pipeline is unrelated to AutoencoderKLTemporalDecoder tiling support." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 230, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10057", + "issue:10227", + "issue:10492", + "issue:10514", + "issue:10705", + "issue:10952", + "issue:11570", + "issue:13191", + "issue:13232", + "issue:1536", + "issue:172", + "issue:177", + "issue:181", + "issue:2712", + "issue:2838", + "issue:2844", + "issue:3284", + "issue:3679", + "issue:3799", + "issue:4005", + "issue:4542", + "issue:4672", + "issue:4779", + "issue:5009", + "issue:5168", + "issue:5284", + "issue:5325", + "issue:5399", + "issue:5645", + "issue:5857", + "issue:5881", + "issue:5963", + "issue:6096", + "issue:6141", + "issue:6444", + "issue:6593", + "issue:6809", + "issue:6870", + "issue:7629", + "issue:7877", + "issue:7982", + "issue:8410", + "issue:8429", + "issue:8801", + "issue:8834", + "issue:9648", + "issue:9904", + "issue:9972" + ], + "soft_pairs": [ + "issue:2838|issue:4005", + "issue:2844|issue:5009", + "issue:3679|issue:4779", + "issue:10057|issue:7982", + "issue:4542|issue:6593", + "issue:4672|issue:8834", + "issue:6870|issue:9648", + "issue:5399|issue:8429", + "issue:10492|issue:6809", + "issue:5168|issue:8410", + "issue:5284|issue:8801", + "issue:6141|issue:7877", + "issue:5963|issue:9972", + "issue:10952|issue:11570", + "issue:172|issue:177", + "issue:13191|issue:13232", + "issue:3799|issue:5325", + "issue:5857|issue:5881", + "issue:6096|issue:7629", + "issue:10227|issue:10514", + "issue:181|issue:3284", + "issue:2712|issue:6444", + "issue:10705|issue:9904", + "issue:1536|issue:5645" + ], + "prepared_review_unit_hash": "edccaa218385e273f50043cfb56194ffdec3ad597eb6e07443c8d4c78c49f831", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33822, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items do not form a duplicate cluster: they span many unrelated bugs, feature requests, and model-support issues across different pipelines and code paths. All soft edges should be rejected.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2838", + "right": "issue:4005", + "accept": false, + "reason": "Different requests: ControlNet pipeline behavior vs inpaint aspect-ratio preservation." + }, + { + "left": "issue:2844", + "right": "issue:5009", + "accept": false, + "reason": "Different runtime failures in different models/pipelines; no shared code-path bug." + }, + { + "left": "issue:3679", + "right": "issue:4779", + "accept": false, + "reason": "A model announcement/request vs an image-to-video pipeline feature request." + }, + { + "left": "issue:10057", + "right": "issue:7982", + "accept": false, + "reason": "Different APIs and failures: from_single_file token issue vs MultiControlNet from_pipe bug." + }, + { + "left": "issue:4542", + "right": "issue:6593", + "accept": false, + "reason": "One is about generating multiple images; the other is LoRA loading support for SDXL img2img." + }, + { + "left": "issue:4672", + "right": "issue:8834", + "accept": false, + "reason": "Training VRAM reduction vs a request to release SD3 ControlNet training code." + }, + { + "left": "issue:6870", + "right": "issue:9648", + "accept": false, + "reason": "Missing enable_vae_tiling support vs memory overflow when using it; related topic but not the same bug." + }, + { + "left": "issue:5399", + "right": "issue:8429", + "accept": false, + "reason": "Different pipelines and symptoms: dtype mismatch in inpainting vs I2VGen-XL script issues." + }, + { + "left": "issue:10492", + "right": "issue:6809", + "accept": false, + "reason": "Different LoRA/quantization problems with different failure modes and code paths." + }, + { + "left": "issue:5168", + "right": "issue:8410", + "accept": false, + "reason": "Support requests for different model families; not the same underlying change." + }, + { + "left": "issue:5284", + "right": "issue:8801", + "accept": false, + "reason": "ControlNet-LLLite support and Kolors support are separate model integrations." + }, + { + "left": "issue:6141", + "right": "issue:7877", + "accept": false, + "reason": "Unrelated features: ControlNet parameters vs Attention out_dim argument." + }, + { + "left": "issue:5963", + "right": "issue:9972", + "accept": false, + "reason": "Kandinsky embeds-vs-prompts runtime error is unrelated to CogX memory allocation on macOS." + }, + { + "left": "issue:10952", + "right": "issue:11570", + "accept": false, + "reason": "Both involve LoRA loading, but they are different bugs in different paths and not one mergeable fix." + }, + { + "left": "issue:172", + "right": "issue:177", + "accept": false, + "reason": "Progress bar options vs documentation for a model-loading argument." + }, + { + "left": "issue:13191", + "right": "issue:13232", + "accept": false, + "reason": "Support requests for different models; no shared concrete bug/change." + }, + { + "left": "issue:3799", + "right": "issue:5325", + "accept": false, + "reason": "Type inference bug vs packaging/type-distribution issue; unrelated." + }, + { + "left": "issue:5857", + "right": "issue:5881", + "accept": false, + "reason": "AnimateDiff/LCM-specific incompatibility vs a broad quality complaint about AnimateDiff." + }, + { + "left": "issue:6096", + "right": "issue:7629", + "accept": false, + "reason": "Both mention SVD, but one is a torch.compile shape exception and the other is an OOM issue on an older PyTorch version." + }, + { + "left": "issue:10227", + "right": "issue:10514", + "accept": false, + "reason": "Different model-specific errors: LoRA feature-size limitation vs Sana resolution-binning/sample_size support." + }, + { + "left": "issue:181", + "right": "issue:3284", + "accept": false, + "reason": "Different training scripts and failures; no common underlying bug." + }, + { + "left": "issue:2712", + "right": "issue:6444", + "accept": false, + "reason": "Missing config field warning vs CLIPTextModel peft_config AttributeError are separate problems." + }, + { + "left": "issue:10705", + "right": "issue:9904", + "accept": false, + "reason": "FluxPipeline performance degradation when reusing the pipeline vs silent image-shape rounding." + }, + { + "left": "issue:1536", + "right": "issue:5645", + "accept": false, + "reason": "Different custom-pipeline incompatibilities and different feature areas." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 231, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10139", + "issue:10231", + "issue:10382", + "issue:10722", + "issue:10952", + "issue:11644", + "issue:1261", + "issue:12610", + "issue:1314", + "issue:1657", + "issue:170", + "issue:1742", + "issue:1909", + "issue:1925", + "issue:201", + "issue:2518", + "issue:3643", + "issue:3915", + "issue:3975", + "issue:4002", + "issue:4042", + "issue:4405", + "issue:444", + "issue:4578", + "issue:4913", + "issue:4954", + "issue:5485", + "issue:5598", + "issue:5607", + "issue:5630", + "issue:5678", + "issue:5846", + "issue:5866", + "issue:5989", + "issue:6068", + "issue:6460", + "issue:6548", + "issue:6657", + "issue:6809", + "issue:7012", + "issue:7591", + "issue:7606", + "issue:8612", + "issue:8708", + "issue:8874", + "issue:8975", + "issue:929", + "issue:9477" + ], + "soft_pairs": [ + "issue:6548|issue:6657", + "issue:1261|issue:1314", + "issue:10231|issue:8975", + "issue:10382|issue:12610", + "issue:5630|issue:6068", + "issue:1314|issue:5678", + "issue:170|issue:444", + "issue:201|issue:929", + "issue:2518|issue:4002", + "issue:4913|issue:5866", + "issue:5598|issue:5989", + "issue:5846|issue:7012", + "issue:1742|issue:1909", + "issue:4042|issue:6460", + "issue:10952|issue:6809", + "issue:5485|issue:8612", + "issue:10722|issue:7606", + "issue:1925|issue:3915", + "issue:3975|issue:7591", + "issue:3643|issue:5607", + "issue:4954|issue:8874", + "issue:1657|issue:4405", + "issue:4578|issue:8708", + "issue:5989|issue:9477", + "issue:10139|issue:11644" + ], + "prepared_review_unit_hash": "d5e12cd7b7fd1967f64a3d4fdfe94a9aeae1dcf7a6da5c92a78cbd8eb7ecf744", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34609, + "estimated_input_tokens": 8653, + "estimated_eval_tokens": 17562 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft pairs are false positives: they share broad subsystem terms at most, but they describe different concrete bugs, different code paths, or different failure modes. No duplicate-worthy canonical issue or PR stands out from this set.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good canonical target here; the issues span unrelated failures (loading, attention/backward, LoRA, schedulers, docs, and training/runtime errors) rather than one underlying bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6548", + "right": "issue:6657", + "accept": false, + "reason": "Different bug classes: img2img RGBA latent handling vs UNet3D frame/channel layout." + }, + { + "left": "issue:1261", + "right": "issue:1314", + "accept": false, + "reason": "Both involve xformers/backward, but one is CUDA stride assert and the other is a Windows loss-backward error; not the same failure." + }, + { + "left": "issue:10231", + "right": "issue:8975", + "accept": false, + "reason": "Both are import-related, but one is a tutorial ImportError on Ubuntu and the other is a missing LoraLoaderMixin import." + }, + { + "left": "issue:10382", + "right": "issue:12610", + "accept": false, + "reason": "Unrelated code paths: VAE forward return-type mismatch vs Quanto/group offload device mismatch." + }, + { + "left": "issue:5630", + "right": "issue:6068", + "accept": false, + "reason": "Different scheduler issues with different symptoms and algorithms (LMS noise leftover vs DDIM incorrect samples)." + }, + { + "left": "issue:1314", + "right": "issue:5678", + "accept": false, + "reason": "Both touch attention/backward, but one is xformers on Windows and the other is higher-order-derivative failure in scaled_dot_product_attention." + }, + { + "left": "issue:170", + "right": "issue:444", + "accept": false, + "reason": "Unrelated errors: unexpected eta keyword vs index error at certain inference-step counts." + }, + { + "left": "issue:201", + "right": "issue:929", + "accept": false, + "reason": "One is a test issue, the other is an attention scaling implementation question; not the same bug." + }, + { + "left": "issue:2518", + "right": "issue:4002", + "accept": false, + "reason": "Different missing features: xformers support in an example script vs missing from_single_file on a pipeline." + }, + { + "left": "issue:4913", + "right": "issue:5866", + "accept": false, + "reason": "Both mention ControlNet, but one is a specific SDXL openpose model issue and the other is AnimateDiff + ControlNet integration." + }, + { + "left": "issue:5598", + "right": "issue:5989", + "accept": false, + "reason": "Different components: a text encoder modification bug vs a safetensors.torch bug." + }, + { + "left": "issue:5846", + "right": "issue:7012", + "accept": false, + "reason": "Resume-from-checkpoint training error is unrelated to v_prediction image quality issues." + }, + { + "left": "issue:1742", + "right": "issue:1909", + "accept": false, + "reason": "Performance timing for repeated prompts is unrelated to generator-argument support in schedulers." + }, + { + "left": "issue:4042", + "right": "issue:6460", + "accept": false, + "reason": "Different problems: combining pipelines/features vs endless loading while loading an inpainting model." + }, + { + "left": "issue:10952", + "right": "issue:6809", + "accept": false, + "reason": "Both are LoRA-related, but one is adapter-name resolution in a pipeline and the other is PEFT loading regressions." + }, + { + "left": "issue:5485", + "right": "issue:8612", + "accept": false, + "reason": "Batch ControlNet inference slowdown is unrelated to PixArt-Sigma mixed-precision incompatibility." + }, + { + "left": "issue:10722", + "right": "issue:7606", + "accept": false, + "reason": "Different training/runtime failures: DreamBooth tensor-size mismatch vs deprecated Kandinsky LoRA processor API." + }, + { + "left": "issue:1925", + "right": "issue:3915", + "accept": false, + "reason": "Unrelated contexts: SD2 inpainting half precision vs torch.compile on text-to-video pipelines." + }, + { + "left": "issue:3975", + "right": "issue:7591", + "accept": false, + "reason": "A typo in an error message is not the same as a HeaderTooLarge failure in from_single_file local loading." + }, + { + "left": "issue:3643", + "right": "issue:5607", + "accept": false, + "reason": "Different tensor-shape failures in different pipelines; no shared underlying bug." + }, + { + "left": "issue:4954", + "right": "issue:8874", + "accept": false, + "reason": "Both involve unpickling, but one is textual inversion loading and the other is ckpt from_single_file loading with weights-only failure." + }, + { + "left": "issue:1657", + "right": "issue:4405", + "accept": false, + "reason": "A question about a Kaggle notebook is unrelated to creating an SDXL textual inversion training script." + }, + { + "left": "issue:4578", + "right": "issue:8708", + "accept": false, + "reason": "Different documentation omissions: VAE not used in docs vs missing vae.config.shift_factor in DreamBooth examples." + }, + { + "left": "issue:5989", + "right": "issue:9477", + "accept": false, + "reason": "Different failures entirely: safetensors.torch bug vs GatheredParameters callable error." + }, + { + "left": "issue:10139", + "right": "issue:11644", + "accept": false, + "reason": "Both concern pipeline init, but they affect different pipelines with different initialization failures and no shared concrete code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 232, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10239", + "issue:10314", + "issue:10446", + "issue:10450", + "issue:10489", + "issue:1100", + "issue:11536", + "issue:11721", + "issue:11973", + "issue:12110", + "issue:1524", + "issue:1922", + "issue:2310", + "issue:2332", + "issue:2466", + "issue:2712", + "issue:2783", + "issue:2830", + "issue:3029", + "issue:3109", + "issue:3591", + "issue:3624", + "issue:3661", + "issue:4125", + "issue:490", + "issue:5109", + "issue:5171", + "issue:5344", + "issue:5670", + "issue:579", + "issue:5927", + "issue:6565", + "issue:7276", + "issue:7303", + "issue:7466", + "issue:7538", + "issue:7735", + "issue:8482", + "issue:8577", + "issue:8874", + "issue:9112", + "issue:9163", + "issue:9516", + "issue:953", + "issue:9637", + "issue:9778", + "issue:9876", + "issue:9895" + ], + "soft_pairs": [ + "issue:11536|issue:9895", + "issue:10489|issue:12110", + "issue:3624|issue:5171", + "issue:10239|issue:10314", + "issue:11721|issue:8874", + "issue:5670|issue:9112", + "issue:2310|issue:5927", + "issue:2332|issue:2830", + "issue:3029|issue:3591", + "issue:3109|issue:953", + "issue:3661|issue:5344", + "issue:1100|issue:5109", + "issue:1524|issue:2466", + "issue:2712|issue:579", + "issue:4125|issue:6565", + "issue:7538|issue:7735", + "issue:10446|issue:7466", + "issue:10450|issue:9516", + "issue:11973|issue:9163", + "issue:1922|issue:490", + "issue:2783|issue:7303", + "issue:7276|issue:9778", + "issue:8482|issue:8577", + "issue:9637|issue:9876" + ], + "prepared_review_unit_hash": "9bacd8f8eef7352c3c9c367aa8bbc71c9ebff4b1f1b0aae1e5461725fb49d5ba", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34056, + "estimated_input_tokens": 8514, + "estimated_eval_tokens": 17284 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicate cluster emerges here: the proposed pairs mostly share broad subsystems or similar error shapes, but they describe different bugs, features, or code paths. I would reject all soft-edge matches.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:10446 is the most concrete standalone bug report with a clear subsystem target, but it is not a duplicate of the other items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11536", + "right": "issue:9895", + "accept": false, + "reason": "Both involve FluxPipeline, but one is prompt adherence and the other is a device-mismatch regression; different failures." + }, + { + "left": "issue:10489", + "right": "issue:12110", + "accept": false, + "reason": "SanaPipeline example/load issues are unrelated to a UNet1DModel GaussianFourier initialization bug." + }, + { + "left": "issue:3624", + "right": "issue:5171", + "accept": false, + "reason": "Generic getattr TypeError is not the same as SDXL rejecting the safety_checker argument." + }, + { + "left": "issue:10239", + "right": "issue:10314", + "accept": false, + "reason": "HunyuanVideo offload-support limitation and NaN generation are different pipeline problems." + }, + { + "left": "issue:11721", + "right": "issue:8874", + "accept": false, + "reason": "SanaPipeline crash-on-load and single-file CKPT unpickling failure are distinct load paths." + }, + { + "left": "issue:5670", + "right": "issue:9112", + "accept": false, + "reason": "Non-square image support in PixArtAlpha is unrelated to Auraflow tensor-shape mismatch." + }, + { + "left": "issue:2310", + "right": "issue:5927", + "accept": false, + "reason": "LDM inpainting/LaMA cleaner and IP-Adapter + reference inpaint are different workflows." + }, + { + "left": "issue:2332", + "right": "issue:2830", + "accept": false, + "reason": "Dreambooth inference error and a video-model usage question are not the same bug." + }, + { + "left": "issue:3029", + "right": "issue:3591", + "accept": false, + "reason": "Custom ControlNet resolution support is not the same as an unclear loss curve during training." + }, + { + "left": "issue:3109", + "right": "issue:953", + "accept": false, + "reason": "Both ask about smaller images, but one is a concrete generation request and the other is general guidance." + }, + { + "left": "issue:3661", + "right": "issue:5344", + "accept": false, + "reason": "from_ckpt SD2 weight conversion failure is unrelated to a missing omegaconf dependency." + }, + { + "left": "issue:1100", + "right": "issue:5109", + "accept": false, + "reason": "CLIPTextTransformer.device and load_lora_weights exception in SDXL ControlNet img2img are different code paths." + }, + { + "left": "issue:1524", + "right": "issue:2466", + "accept": false, + "reason": "Poor textual inversion training quality is unrelated to checkpoint retention behavior." + }, + { + "left": "issue:2712", + "right": "issue:579", + "accept": false, + "reason": "Missing norm_num_groups in config and missing config.json are different configuration errors." + }, + { + "left": "issue:4125", + "right": "issue:6565", + "accept": false, + "reason": "LoRA loss-curve expectations and DPO diffusion instability are not the same underlying bug." + }, + { + "left": "issue:7538", + "right": "issue:7735", + "accept": false, + "reason": "Custom Fill50k training issue and missing caption_column are different dataset/training failures." + }, + { + "left": "issue:10446", + "right": "issue:7466", + "accept": false, + "reason": "FluxControlNetPipeline model-sharding incompatibility is distinct from multi-ControlNet torch.compile failure." + }, + { + "left": "issue:10450", + "right": "issue:9516", + "accept": false, + "reason": "Transformer block implementation differences and missing joint_attention_kwargs forwarding are separate issues." + }, + { + "left": "issue:11973", + "right": "issue:9163", + "accept": false, + "reason": "Missing required instance_prompt in the script is not the same as tensor bugs in train_dreambooth_lora_flux.py." + }, + { + "left": "issue:1922", + "right": "issue:490", + "accept": false, + "reason": "Undefined init_git_repo in a notebook is unrelated to memory usage differences between frameworks." + }, + { + "left": "issue:2783", + "right": "issue:7303", + "accept": false, + "reason": "Request to create an img2img ControlNet pipeline and scheduler zero_snr support are different topics." + }, + { + "left": "issue:7276", + "right": "issue:9778", + "accept": false, + "reason": "Network errors while using a downloaded pretrained model are not the same as a tutorial bug." + }, + { + "left": "issue:8482", + "right": "issue:8577", + "accept": false, + "reason": "General SD3 support and Differential Diffusion with SD3 are distinct feature requests." + }, + { + "left": "issue:9637", + "right": "issue:9876", + "accept": false, + "reason": "xFormers attention-mask bfloat bug and LoRA VRAM release concern are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 233, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10896", + "issue:11033", + "issue:11824", + "issue:12501", + "issue:1290", + "issue:13001", + "issue:13298", + "issue:1363", + "issue:1546", + "issue:1604", + "issue:1857", + "issue:259", + "issue:3086", + "issue:3089", + "issue:3279", + "issue:3427", + "issue:3551", + "issue:3612", + "issue:4804", + "issue:4830", + "issue:5402", + "issue:541", + "issue:5570", + "issue:5605", + "issue:5712", + "issue:6055", + "issue:6073", + "issue:6141", + "issue:6363", + "issue:6446", + "issue:6520", + "issue:6530", + "issue:6812", + "issue:7441", + "issue:746", + "issue:7460", + "issue:7675", + "issue:7751", + "issue:8551", + "issue:8939", + "issue:9227", + "issue:9309", + "issue:9320", + "issue:9358", + "issue:9397", + "issue:9630", + "issue:9825" + ], + "soft_pairs": [ + "issue:9227|issue:9358", + "issue:6363|issue:6812", + "issue:1363|issue:746", + "issue:1546|issue:6055", + "issue:5605|issue:7675", + "issue:11033|issue:13001", + "issue:6520|issue:9630", + "issue:12501|issue:13298", + "issue:4804|issue:4830", + "issue:3279|issue:7751", + "issue:5402|issue:6073", + "issue:10896|issue:9825", + "issue:11824|issue:9309", + "issue:1290|issue:3086", + "issue:3089|issue:7460", + "issue:3427|issue:6530", + "issue:3612|issue:541", + "issue:5570|issue:6520", + "issue:5712|issue:7441", + "issue:6141|issue:9397", + "issue:1604|issue:3551", + "issue:6446|issue:9320", + "issue:8551|issue:8939", + "issue:1857|issue:259" + ], + "prepared_review_unit_hash": "2bd8658dc2620f321f9e9e58c702f68210a15f9034cc41c52d14a5a90bfe1d6b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33495, + "estimated_input_tokens": 8374, + "estimated_eval_tokens": 17004 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated diffusers issues, not a true duplicate set. The soft-edge candidates are false positives driven by generic wording like \u201cerror\u201d, \u201cbug\u201d, and \u201csupport\u201d rather than the same underlying code-path bug or feature request.", + "confidence": 0.96, + "canonical_issue_reason": "No issue is a clean canonical duplicate target because the cluster is heterogeneous. If one placeholder must be chosen, issue 13298 is the broadest open runtime-bug report with some inbound references, while most others are narrower feature requests or unrelated failures.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 13298 is the best single representative only as a placeholder: it is open, broadly framed, and actionable as a runtime-generation bug. That said, the set overall does not share one underlying defect.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9227", + "right": "issue:9358", + "accept": false, + "reason": "Both mention train_dreambooth_lora_flux, but one is a model/run failure and the other is redundant text-encoder reinitialization; different bugs, not one duplicate." + }, + { + "left": "issue:6363", + "right": "issue:6812", + "accept": false, + "reason": "Both are LoRA training errors, but one is FP16 unscale failure and the other is Q/K/V dtype mismatch; distinct failure modes." + }, + { + "left": "issue:1363", + "right": "issue:746", + "accept": false, + "reason": "Different backends and symptoms: ONNX long-prompt weight error with LMS scheduler vs img2img ONNX revision loading problem." + }, + { + "left": "issue:1546", + "right": "issue:6055", + "accept": false, + "reason": "Truncation warning on batched inputs is unrelated to a generic accelerate error." + }, + { + "left": "issue:5605", + "right": "issue:7675", + "accept": false, + "reason": "One is SDXL DreamBooth caption support, the other is MLX quantized linear support; both are feature requests but for different components." + }, + { + "left": "issue:11033", + "right": "issue:13001", + "accept": false, + "reason": "Both are model-loading issues, but SD1.5 from_single_file loading and Flux2.Klein GGUF shape mismatch are different model families and code paths." + }, + { + "left": "issue:6520", + "right": "issue:9630", + "accept": false, + "reason": "Different scripts and symptoms: blank validation output in adapter training vs num_single_layers handling in flux-controlnet on single GPU." + }, + { + "left": "issue:12501", + "right": "issue:13298", + "accept": false, + "reason": "Both mention runtime/device behavior, but graph breaks in VAE decoder and CUDA generator device breakage are not the same bug." + }, + { + "left": "issue:4804", + "right": "issue:4830", + "accept": false, + "reason": "LoraLoaderMixin dict-iteration error and MultiControlNet tensor-size error affect different subsystems." + }, + { + "left": "issue:3279", + "right": "issue:7751", + "accept": false, + "reason": "Checkpoint resume errors in DreamBooth LoRA are unrelated to animatediff motion-module conversion failures." + }, + { + "left": "issue:5402", + "right": "issue:6073", + "accept": false, + "reason": "Missing compute_snr import is a package/API issue; the SDXL DreamBooth README error is a separate user-facing training problem." + }, + { + "left": "issue:10896", + "right": "issue:9825", + "accept": false, + "reason": "Both are support requests, but SkyReels-A1 portrait animation and FLUX IPAdapter support are different feature asks." + }, + { + "left": "issue:11824", + "right": "issue:9309", + "accept": false, + "reason": "Different FLUX pipelines and distinct failures: multi-image Kontext input vs FluxControlNet schnell bug." + }, + { + "left": "issue:1290", + "right": "issue:3086", + "accept": false, + "reason": "DreamArtist community pipeline and audio-to-image are unrelated topics." + }, + { + "left": "issue:3089", + "right": "issue:7460", + "accept": false, + "reason": "DDPM scheduler bug and AdaLayerNorm bug are different internal code-path issues." + }, + { + "left": "issue:3427", + "right": "issue:6530", + "accept": false, + "reason": "Different projects/features: makeavid-sd-jax vs SDXL K-diffusion ControlNet." + }, + { + "left": "issue:3612", + "right": "issue:541", + "accept": false, + "reason": "ControlNet preprocessor configuration and SageMaker usage are unrelated." + }, + { + "left": "issue:5570", + "right": "issue:6520", + "accept": false, + "reason": "Missing compute_snr packaging issue is not the same as blank log_validation output during training." + }, + { + "left": "issue:5712", + "right": "issue:7441", + "accept": false, + "reason": "Both involve Python 3.12, but one is pip install on Windows and the other is pytest execution; different problems." + }, + { + "left": "issue:6141", + "right": "issue:9397", + "accept": false, + "reason": "Control mode/resize mode support for ControlNet and SGMUniform scheduler support for Hyper LoRAs are separate feature requests." + }, + { + "left": "issue:1604", + "right": "issue:3551", + "accept": false, + "reason": "Broken lpw stable diffusion on diffusers 0.9.0 is unrelated to LoRA disabling xFormers." + }, + { + "left": "issue:6446", + "right": "issue:9320", + "accept": false, + "reason": "Minimal-loading Colab model load failure and custom-diffusion save/load problems are different training/loading paths." + }, + { + "left": "issue:8551", + "right": "issue:8939", + "accept": false, + "reason": "VAE device_map balanced latent placement bug and tokenizer loading failure for SD3 are unrelated components." + }, + { + "left": "issue:1857", + "right": "issue:259", + "accept": false, + "reason": "Poor UnCLIP testing and progress bar fix are not the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 234, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10045", + "issue:10231", + "issue:10569", + "issue:10800", + "issue:10952", + "issue:1099", + "issue:11199", + "issue:11486", + "issue:11700", + "issue:1207", + "issue:1260", + "issue:12624", + "issue:2132", + "issue:2646", + "issue:3010", + "issue:3401", + "issue:3412", + "issue:4229", + "issue:4376", + "issue:4438", + "issue:444", + "issue:4534", + "issue:4605", + "issue:4907", + "issue:4933", + "issue:5118", + "issue:5371", + "issue:5743", + "issue:5799", + "issue:6093", + "issue:6442", + "issue:6445", + "issue:6530", + "issue:7389", + "issue:761", + "issue:7682", + "issue:7722", + "issue:782", + "issue:7856", + "issue:8234", + "issue:8801", + "issue:8828", + "issue:9027", + "issue:9403", + "issue:9495", + "issue:9602", + "issue:9660", + "issue:9825" + ], + "soft_pairs": [ + "issue:1099|issue:1260", + "issue:11486|issue:8828", + "issue:10569|issue:10800", + "issue:2646|issue:444", + "issue:4376|issue:4438", + "issue:3401|issue:5743", + "issue:9602|issue:9660", + "issue:10231|issue:7682", + "issue:10045|issue:11199", + "issue:1207|issue:2132", + "issue:3010|issue:3412", + "issue:5118|issue:6530", + "issue:5371|issue:9403", + "issue:6093|issue:9027", + "issue:7722|issue:8234", + "issue:8801|issue:9825", + "issue:8234|issue:9495", + "issue:4907|issue:5799", + "issue:4229|issue:4534", + "issue:2646|issue:761", + "issue:11700|issue:12624", + "issue:4933|issue:782", + "issue:6442|issue:6445", + "issue:10952|issue:7856", + "issue:4605|issue:7389" + ], + "prepared_review_unit_hash": "b9ff7671ca36617e6b649aa9c2930a84c78fb35d914a3ca5ba6812c099ab566d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34031, + "estimated_input_tokens": 8508, + "estimated_eval_tokens": 17272 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a noisy cluster of mostly unrelated diffusers issues linked only by broad title similarity; I don\u2019t see any true duplicate pairs or shared fixes among the soft edges.", + "confidence": 0.18, + "canonical_issue_reason": "issue:10045 is the broadest, most umbrella-style problem report in the set, so it works best as a loose cluster anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10045 is the best representative issue because its generic import/pipeline wording is the closest match to the cluster\u2019s superficial theme.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1099", + "right": "issue:1260", + "accept": false, + "reason": "Both are import/runtime failures, but they are different APIs and different failure modes (DreamBooth class-name issue vs scheduler import error)." + }, + { + "left": "issue:11486", + "right": "issue:8828", + "accept": false, + "reason": "Different request and bug: xformers training-script support vs SD3 training memory error." + }, + { + "left": "issue:10569", + "right": "issue:10800", + "accept": false, + "reason": "Both involve HunyuanVideo CPU usage, but one is high memory consumption and the other is a BnB int4 + offload quantization type error." + }, + { + "left": "issue:2646", + "right": "issue:444", + "accept": false, + "reason": "Unrelated failures: tiled VAE IndexError vs inference-step indexing bug." + }, + { + "left": "issue:4376", + "right": "issue:4438", + "accept": false, + "reason": "Same pipeline family, but different crash contexts and no shared concrete code path identified." + }, + { + "left": "issue:3401", + "right": "issue:5743", + "accept": false, + "reason": "Class-label validation error is unrelated to LCD training stuck at 0%." + }, + { + "left": "issue:9602", + "right": "issue:9660", + "accept": false, + "reason": "Adapter-name lookup failure is different from from_pretrained leaving modules in train mode." + }, + { + "left": "issue:10231", + "right": "issue:7682", + "accept": false, + "reason": "ImportError in tutorial on Ubuntu is unrelated to multi-GPU memory/throughput behavior." + }, + { + "left": "issue:10045", + "right": "issue:11199", + "accept": false, + "reason": "Both mention imports, but one is generic pipeline importing and the other is a specific onnx_utils import failure." + }, + { + "left": "issue:1207", + "right": "issue:2132", + "accept": false, + "reason": "CUDA runtime symbol error and Accelerator trackers attribute error are different environment/library issues." + }, + { + "left": "issue:3010", + "right": "issue:3412", + "accept": false, + "reason": "Both are docs-related, but they are not the same document or fix." + }, + { + "left": "issue:5118", + "right": "issue:6530", + "accept": false, + "reason": "ControlNet reference_only usage question is not the same as SDXL K-diffusion + ControlNet error." + }, + { + "left": "issue:5371", + "right": "issue:9403", + "accept": false, + "reason": "Different feature requests: BLIP inpainting support vs Xlabs IPAdapter support for Flux." + }, + { + "left": "issue:6093", + "right": "issue:9027", + "accept": false, + "reason": "Push_to_hub repo-creation bug is unrelated to argument-name consistency discussion." + }, + { + "left": "issue:7722", + "right": "issue:8234", + "accept": false, + "reason": "Both touch IPAdapter, but one is Animatediff/ControlNet integration and the other is ip_adapter_scale type handling." + }, + { + "left": "issue:8801", + "right": "issue:9825", + "accept": false, + "reason": "Different model families and feature requests: Kolors support vs Flux IPAdapter support." + }, + { + "left": "issue:8234", + "right": "issue:9495", + "accept": false, + "reason": "Both involve IPAdapter, but one is a dict scaling TypeError and the other is SDXL PAG compatibility." + }, + { + "left": "issue:4907", + "right": "issue:5799", + "accept": false, + "reason": "Inpainting warning with SDXL is not the same as adding a strength argument to inpainting." + }, + { + "left": "issue:4229", + "right": "issue:4534", + "accept": false, + "reason": "Different training contexts and errors: multi-GPU Accelerate issue vs MPS validation crash." + }, + { + "left": "issue:2646", + "right": "issue:761", + "accept": false, + "reason": "Tiled VAE IndexError is unrelated to DreamBooth + 8-bit Adam failure." + }, + { + "left": "issue:11700", + "right": "issue:12624", + "accept": false, + "reason": "Pruna integration request and WaniImageToVideoPipeline strength request are unrelated feature asks." + }, + { + "left": "issue:4933", + "right": "issue:782", + "accept": false, + "reason": "DreamBooth output artifact problem is unrelated to model download auth-token removal." + }, + { + "left": "issue:6442", + "right": "issue:6445", + "accept": false, + "reason": "Both are training-time ValueErrors, but from different validation paths and different user inputs." + }, + { + "left": "issue:10952", + "right": "issue:7856", + "accept": false, + "reason": "Adapter-name mismatch during LoRA loading is unrelated to tokenizer multi-vector token collision." + }, + { + "left": "issue:4605", + "right": "issue:7389", + "accept": false, + "reason": "Different pipeline errors: long-prompt-weighting inpaint issue vs from_single_file loading error." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 235, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10016", + "issue:10336", + "issue:10518", + "issue:10952", + "issue:11127", + "issue:11286", + "issue:11342", + "issue:11437", + "issue:11584", + "issue:1161", + "issue:11790", + "issue:1181", + "issue:12096", + "issue:13104", + "issue:13249", + "issue:1367", + "issue:1388", + "issue:1934", + "issue:1964", + "issue:2294", + "issue:2312", + "issue:2507", + "issue:2774", + "issue:2995", + "issue:3209", + "issue:3338", + "issue:3391", + "issue:4071", + "issue:4445", + "issue:4615", + "issue:4675", + "issue:4789", + "issue:4912", + "issue:4964", + "issue:5325", + "issue:557", + "issue:6705", + "issue:6823", + "issue:7450", + "issue:7572", + "issue:7706", + "issue:7770", + "issue:7799", + "issue:8534", + "issue:9003", + "issue:9393", + "issue:9546", + "issue:993" + ], + "soft_pairs": [ + "issue:1161|issue:4445", + "issue:1181|issue:4445", + "issue:3338|issue:4615", + "issue:10016|issue:7770", + "issue:2507|issue:557", + "issue:2995|issue:7572", + "issue:1388|issue:4675", + "issue:4912|issue:9003", + "issue:11584|issue:12096", + "issue:1934|issue:557", + "issue:11286|issue:11790", + "issue:11127|issue:13249", + "issue:1964|issue:4071", + "issue:3391|issue:4789", + "issue:6705|issue:9393", + "issue:7706|issue:7799", + "issue:2774|issue:4964", + "issue:5325|issue:993", + "issue:1367|issue:2312", + "issue:2294|issue:6823", + "issue:3209|issue:7450", + "issue:11437|issue:9546", + "issue:10336|issue:11342", + "issue:10518|issue:8534", + "issue:10952|issue:13104" + ], + "prepared_review_unit_hash": "b7729bb3d1e76de661d7a9e2a74a9d6ee7a20e645a3f0d63749409fd5f0d6572", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33811, + "estimated_input_tokens": 8453, + "estimated_eval_tokens": 17162 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is noisy and spans several unrelated Diffusers bug families: LoRA/model loading, tensor shape mismatches, training/checkpointing, offloading/compile issues, and assorted pipeline-specific regressions. The soft pairs are mostly shared-subsystem similarity, not the same underlying bug, so they should stay separate.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 11286 is the broadest, most generic LoRA-loading report in the set, so it is the best umbrella issue if one must anchor this cluster; most other items are not true duplicates of it.", + "canonical_pr_reason": null, + "best_issue_reason": "11286 is the most representative user-facing report among the LoRA-loading subset, but it is still only a loose umbrella and not a precise duplicate match for the rest of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1161", + "right": "issue:4445", + "accept": false, + "reason": "Flax-to-PyTorch conversion is unrelated to a LoRA finetuning question." + }, + { + "left": "issue:1181", + "right": "issue:4445", + "accept": false, + "reason": "Training from scratch vs. how to finetune LoRA are different requests." + }, + { + "left": "issue:3338", + "right": "issue:4615", + "accept": false, + "reason": "Import performance regression is not the same as a missing pipeline import error." + }, + { + "left": "issue:10016", + "right": "issue:7770", + "accept": false, + "reason": "Both mention shape issues, but SD3.5 single-file pos_embed mismatch and model+LoRA tensor mismatch are different code paths." + }, + { + "left": "issue:2507", + "right": "issue:557", + "accept": false, + "reason": "Safetensors load speed and DDIM quality are unrelated problems." + }, + { + "left": "issue:2995", + "right": "issue:7572", + "accept": false, + "reason": "Matrix shape mismatch and dtype mismatch are different failure modes." + }, + { + "left": "issue:1388", + "right": "issue:4675", + "accept": false, + "reason": "SD 2.0 conversion scripting and SDXL ControlNet reference-only support are different changes." + }, + { + "left": "issue:4912", + "right": "issue:9003", + "accept": false, + "reason": "Checkpoint conversion to diffusers is unrelated to safety_checker strength control." + }, + { + "left": "issue:11584", + "right": "issue:12096", + "accept": false, + "reason": "Both involve offloading/compile behavior, but they target different models, pipelines, and failure modes." + }, + { + "left": "issue:1934", + "right": "issue:557", + "accept": false, + "reason": "CPU offload memory usage is unrelated to DDIM output quality." + }, + { + "left": "issue:11286", + "right": "issue:11790", + "accept": false, + "reason": "Both are LoRA loading complaints, but they are different scripts/models and report different concrete errors." + }, + { + "left": "issue:11127", + "right": "issue:13249", + "accept": false, + "reason": "CivitAI Flux single-file loading and Z-Image LoRA breakage are different model families and bugs." + }, + { + "left": "issue:1964", + "right": "issue:4071", + "accept": false, + "reason": "PyTorch support planning is unrelated to inpaint-only+lama feature support." + }, + { + "left": "issue:3391", + "right": "issue:4789", + "accept": false, + "reason": "Overexposed super-resolution output and download failure for DeepFloyd IF are different issues." + }, + { + "left": "issue:6705", + "right": "issue:9393", + "accept": false, + "reason": "Both involve saving during training, but accelerate+FSDP ckpt saving and deepspeed Flux LoRA model saving are not the same bug." + }, + { + "left": "issue:7706", + "right": "issue:7799", + "accept": false, + "reason": "Custom pipeline OOM and IP-Adapter matmul shape mismatch are unrelated." + }, + { + "left": "issue:2774", + "right": "issue:4964", + "accept": false, + "reason": "ControlNet safetensors conversion and SDXL model conversion are different conversion paths and errors." + }, + { + "left": "issue:5325", + "right": "issue:993", + "accept": false, + "reason": "Packaging type metadata distribution and fp16 dtype mismatch are unrelated." + }, + { + "left": "issue:1367", + "right": "issue:2312", + "accept": false, + "reason": "Scheduler config saving with trained_betas and EMA hook save failure are different code paths." + }, + { + "left": "issue:2294", + "right": "issue:6823", + "accept": false, + "reason": "Missing video inpaint import and inability to use ControlNet are unrelated." + }, + { + "left": "issue:3209", + "right": "issue:7450", + "accept": false, + "reason": "Large-image NaNs in AutoencoderKL and multithreading issues are different bugs." + }, + { + "left": "issue:11437", + "right": "issue:9546", + "accept": false, + "reason": "Missing training outputs and validation-step OOM are different symptoms and fixes." + }, + { + "left": "issue:10336", + "right": "issue:11342", + "accept": false, + "reason": "Both mention Sana, but activation/clamping explosions on one side and MPS failure on the other are different problems." + }, + { + "left": "issue:10518", + "right": "issue:8534", + "accept": false, + "reason": "A wrong example script and an SD3 logit-normal weighting bug are not the same underlying issue." + }, + { + "left": "issue:10952", + "right": "issue:13104", + "accept": false, + "reason": "LoRA adapter-name mismatch and an undefined logger error are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 236, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10045", + "issue:1037", + "issue:10514", + "issue:10675", + "issue:10833", + "issue:10994", + "issue:1101", + "issue:11342", + "issue:13034", + "issue:1522", + "issue:1817", + "issue:1961", + "issue:2098", + "issue:2268", + "issue:2299", + "issue:2645", + "issue:2747", + "issue:3085", + "issue:3249", + "issue:3391", + "issue:3487", + "issue:3631", + "issue:3901", + "issue:4134", + "issue:4305", + "issue:4390", + "issue:4445", + "issue:4554", + "issue:4665", + "issue:5335", + "issue:5429", + "issue:5592", + "issue:5687", + "issue:5689", + "issue:5703", + "issue:6141", + "issue:6517", + "issue:6818", + "issue:725", + "issue:7312", + "issue:7375", + "issue:7836", + "issue:785", + "issue:8523", + "issue:8898", + "issue:9397", + "issue:9799", + "issue:9822" + ], + "soft_pairs": [ + "issue:1522|issue:3249", + "issue:1817|issue:3085", + "issue:1961|issue:2268", + "issue:2299|issue:5592", + "issue:2645|issue:4134", + "issue:2747|issue:7375", + "issue:4445|issue:725", + "issue:5429|issue:9397", + "issue:6141|issue:7312", + "issue:10514|issue:11342", + "issue:3901|issue:4390", + "issue:6818|issue:8523", + "issue:2098|issue:3631", + "issue:3391|issue:3487", + "issue:10045|issue:6517", + "issue:10994|issue:9799", + "issue:1037|issue:1101", + "issue:10675|issue:9822", + "issue:4554|issue:7836", + "issue:4665|issue:8898", + "issue:5335|issue:785", + "issue:5689|issue:5703", + "issue:10833|issue:13034", + "issue:4305|issue:5687" + ], + "prepared_review_unit_hash": "57a3ba0cb7f56e91dac2d4306132fed908e1d41328bea999e1ecf171e25ddd9f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33796, + "estimated_input_tokens": 8449, + "estimated_eval_tokens": 17154 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No strong duplicate cluster here: the items span unrelated bug reports, feature requests, docs, and scheduler/model-specific issues. All soft-edge pairs should be rejected because they do not describe the same underlying defect or change.", + "confidence": 0.89, + "canonical_issue_reason": "issue:10045 is the broadest umbrella title in the set ('Problem with importing pipelines'), but this is only a weak anchor; the overall set is too heterogeneous to justify a true canonical duplicate issue.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10045 is the most generic and potentially reusable target, but it is still not a good global representative because the rest of the items are largely unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1522", + "right": "issue:3249", + "accept": false, + "reason": "Both are TypeError-style API mismatch reports, but for different arguments and different blocks; not the same bug." + }, + { + "left": "issue:1817", + "right": "issue:3085", + "accept": false, + "reason": "Both mention removed/unsupported scheduler-related args, but one is Dreambooth num_cycles and the other is cache folder; unrelated failures." + }, + { + "left": "issue:1961", + "right": "issue:2268", + "accept": false, + "reason": "Both are vague ColossalAI-related errors, but the reports do not indicate the same code path or fix." + }, + { + "left": "issue:2299", + "right": "issue:5592", + "accept": false, + "reason": "Different model-loading contexts and failure modes; one is fine-tuned checkpoints in Diffusers, the other is UniDiffuser checkpoint loading." + }, + { + "left": "issue:2645", + "right": "issue:4134", + "accept": false, + "reason": "Both are MPS/CUDA device issues in ControlNet, but one is a CUDA assertion and the other is a tensor shape mismatch; not the same bug." + }, + { + "left": "issue:2747", + "right": "issue:7375", + "accept": false, + "reason": "One is a generic docs update request; the other is a specific docs note about FreeInit outputs." + }, + { + "left": "issue:4445", + "right": "issue:725", + "accept": false, + "reason": "A how-to question about LoRA finetuning is not the same as a memory-footprint feature request for device map/accelerate." + }, + { + "left": "issue:5429", + "right": "issue:9397", + "accept": false, + "reason": "Both involve model/scheduler support additions, but they target different model families and different missing capabilities." + }, + { + "left": "issue:6141", + "right": "issue:7312", + "accept": false, + "reason": "One requests ControlNet control/resize modes; the other is about AlignProp direct reward finetuning. Different features." + }, + { + "left": "issue:10514", + "right": "issue:11342", + "accept": false, + "reason": "Both are Sana pipeline issues, but one is a resolution-binning/sample_size constraint and the other is an MPS failure; distinct bugs." + }, + { + "left": "issue:3901", + "right": "issue:4390", + "accept": false, + "reason": "Both are missing-file/environment setup errors, but one is missing requirements.txt and the other is generic missing requirements; not the same incident." + }, + { + "left": "issue:6818", + "right": "issue:8523", + "accept": false, + "reason": "Both mention runtime/back-end errors, but one is IP-Adapter unload construction and the other is a playground CUDA error; unrelated." + }, + { + "left": "issue:2098", + "right": "issue:3631", + "accept": false, + "reason": "A torch.randn argument mismatch is unrelated to the inpainting latents exception." + }, + { + "left": "issue:3391", + "right": "issue:3487", + "accept": false, + "reason": "Both are image-quality complaints, but one is overexposure in super-resolution and the other is poor results with a specific scheduler/strength combo." + }, + { + "left": "issue:10045", + "right": "issue:6517", + "accept": false, + "reason": "A generic pipeline import problem is not the same as a specific ControlNet SDXL import traceback." + }, + { + "left": "issue:10994", + "right": "issue:9799", + "accept": false, + "reason": "Different tensor-shape crashes in different training/inference paths; not the same underlying bug." + }, + { + "left": "issue:1037", + "right": "issue:1101", + "accept": false, + "reason": "Both are feature requests for community schedulers/pipelines, but they are different additions." + }, + { + "left": "issue:10675", + "right": "issue:9822", + "accept": false, + "reason": "Flux scheduler config max_shift and loading SDXL LoRAs into Flux are unrelated issues." + }, + { + "left": "issue:4554", + "right": "issue:7836", + "accept": false, + "reason": "One is an OOM on a specific example; the other is CLIP training overfitting. No shared bug." + }, + { + "left": "issue:4665", + "right": "issue:8898", + "accept": false, + "reason": "Both concern loading/safetensors behavior, but one is Windows alt-bin selection and the other is subfolder path validation; different defects." + }, + { + "left": "issue:5335", + "right": "issue:785", + "accept": false, + "reason": "A deployment/repo-access question is unrelated to scheduler performance." + }, + { + "left": "issue:5689", + "right": "issue:5703", + "accept": false, + "reason": "Residual noise with DPM++ SDXL is different from ddim_inverse noise reconstruction; distinct scheduler behavior." + }, + { + "left": "issue:10833", + "right": "issue:13034", + "accept": false, + "reason": "Different model/pipeline-specific shape or quantization issues; no evidence they share the same failure path." + }, + { + "left": "issue:4305", + "right": "issue:5687", + "accept": false, + "reason": "Both involve timesteps/schedulers, but one is a torch.compile graph break and the other is duplicate timesteps with Karras sigmas; different problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 237, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10287", + "issue:10511", + "issue:10772", + "issue:10958", + "issue:1188", + "issue:12239", + "issue:12501", + "issue:1566", + "issue:1785", + "issue:1790", + "issue:2026", + "issue:2098", + "issue:2390", + "issue:2462", + "issue:2465", + "issue:3176", + "issue:3181", + "issue:322", + "issue:3221", + "issue:3366", + "issue:4071", + "issue:4148", + "issue:4306", + "issue:4675", + "issue:4724", + "issue:5070", + "issue:5118", + "issue:5124", + "issue:5168", + "issue:5595", + "issue:5605", + "issue:5607", + "issue:5670", + "issue:6243", + "issue:6331", + "issue:6661", + "issue:7480", + "issue:8801", + "issue:8834", + "issue:8885", + "issue:8989", + "issue:9112", + "issue:9216", + "issue:9237", + "issue:9763", + "issue:9856", + "issue:9930" + ], + "soft_pairs": [ + "issue:10772|issue:8834", + "issue:2390|issue:6243", + "issue:4148|issue:6331", + "issue:5070|issue:5605", + "issue:5168|issue:8801", + "issue:1566|issue:5124", + "issue:10958|issue:9112", + "issue:1790|issue:2465", + "issue:5595|issue:5670", + "issue:8885|issue:9763", + "issue:2026|issue:4724", + "issue:4071|issue:4675", + "issue:5118|issue:9216", + "issue:6661|issue:7480", + "issue:12501|issue:9856", + "issue:3221|issue:3366", + "issue:8989|issue:9930", + "issue:10511|issue:12239", + "issue:1188|issue:322", + "issue:2462|issue:4071", + "issue:3176|issue:3181", + "issue:10287|issue:9237", + "issue:1785|issue:5607", + "issue:2098|issue:4306" + ], + "prepared_review_unit_hash": "21ea6d7c0eb555f692c1c387a546e5521d6c80c037b9ce2289c1cf3f061355ab", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33068, + "estimated_input_tokens": 8267, + "estimated_eval_tokens": 16790 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose grab-bag of mostly unrelated issues. The soft pairs share only broad subsystem words (ControlNet, Dreambooth, TensorRT, OOM, inpainting, etc.) but not the same concrete bug or change, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "If forced to pick one anchor, issue 10511 is the broadest open product request, but the set does not have a real common underlying problem, so no strong canonical issue exists.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10511 is the most general and forward-looking open request in the set, making it the least arbitrary issue to anchor the cluster, though it is not actually a duplicate hub.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10772", + "right": "issue:8834", + "accept": false, + "reason": "Both mention ControlNet/SD3, but one is asking for Sana ControlNet support and the other for SD3 controlnet training code release; different artifacts and goals." + }, + { + "left": "issue:2390", + "right": "issue:6243", + "accept": false, + "reason": "T2I-Adapter support and IP-Adapter Face ID are different adapter types and different feature requests." + }, + { + "left": "issue:4148", + "right": "issue:6331", + "accept": false, + "reason": "One is an SDXL inpainting config error, the other is a request for SAG inpainting support; not the same bug." + }, + { + "left": "issue:5070", + "right": "issue:5605", + "accept": false, + "reason": "ControlNet example parameter changes and SDXL DreamBooth per-image caption support are unrelated training/example features." + }, + { + "left": "issue:5168", + "right": "issue:8801", + "accept": false, + "reason": "ControlNet-XS support and Kolors support target different model families and changes." + }, + { + "left": "issue:1566", + "right": "issue:5124", + "accept": false, + "reason": "DreamBooth fp16 checkpoint crash and a general fine-tuning question about safetensors are not the same issue." + }, + { + "left": "issue:10958", + "right": "issue:9112", + "accept": false, + "reason": "Both are inference errors, but they affect different pipelines and failure modes (meta tensor copy vs tensor size mismatch)." + }, + { + "left": "issue:1790", + "right": "issue:2465", + "accept": false, + "reason": "These are both training questions, but one is a generic 'how training works' request and the other asks for a reproducible DreamBooth recipe." + }, + { + "left": "issue:5595", + "right": "issue:5670", + "accept": false, + "reason": "Both concern size/shape problems, but one is about image dimensions needing to be divisible by 8 and the other is a PixArtAlpha non-square image runtime error." + }, + { + "left": "issue:8885", + "right": "issue:9763", + "accept": false, + "reason": "Custom Diffusion dependency/install/token-loading problems are not the same as Flux ControlNet token truncation during training." + }, + { + "left": "issue:2026", + "right": "issue:4724", + "accept": false, + "reason": "PaintByExample cpu_offload and SDXL Img2Img unexpected behavior are separate pipeline issues." + }, + { + "left": "issue:4071", + "right": "issue:4675", + "accept": false, + "reason": "Inpaint-only+lama support and SDXL ControlNet reference-only are different feature requests." + }, + { + "left": "issue:5118", + "right": "issue:9216", + "accept": false, + "reason": "A question about ControlNet reference_only usage is unrelated to a custom pipeline that combines any number of models." + }, + { + "left": "issue:6661", + "right": "issue:7480", + "accept": false, + "reason": "Both mention OOM, but one is MPS training on specific image sizes and the other is DeepFloyd IF inference memory usage." + }, + { + "left": "issue:12501", + "right": "issue:9856", + "accept": false, + "reason": "A VAE graph-break bug and a distributed port conflict are entirely different failures." + }, + { + "left": "issue:3221", + "right": "issue:3366", + "accept": false, + "reason": "load_attn_procs failing in diffusers 0.16.0 and LoRALinearLayer breaking autocast are different code paths and symptoms." + }, + { + "left": "issue:8989", + "right": "issue:9930", + "accept": false, + "reason": "Sequential offloading in Stable Audio and PAG adaptive scale are unrelated issues." + }, + { + "left": "issue:10511", + "right": "issue:12239", + "accept": false, + "reason": "FlowMatchEuler scheduler redesign for SD3/Flux and InfiniteTalk support are different model-specific requests." + }, + { + "left": "issue:1188", + "right": "issue:322", + "accept": false, + "reason": "CLIPSeg x Stable Diffusion pipeline support and V-Diffusion are unrelated pipeline requests." + }, + { + "left": "issue:2462", + "right": "issue:4071", + "accept": false, + "reason": "IterableDataset training support and inpaint-only+lama support are different areas and not the same underlying change." + }, + { + "left": "issue:3176", + "right": "issue:3181", + "accept": false, + "reason": "Both are TensorRT-related, but one is a demo error and the other is a different example TypeError; not enough evidence of the same fix." + }, + { + "left": "issue:10287", + "right": "issue:9237", + "accept": false, + "reason": "Documentation example issues and Flux DreamBooth LoRA training problems are unrelated." + }, + { + "left": "issue:1785", + "right": "issue:5607", + "accept": false, + "reason": "UnCLIPPipeline num_images_per_prompt bug and a tensor size mismatch in another path are different failures." + }, + { + "left": "issue:2098", + "right": "issue:4306", + "accept": false, + "reason": "Torch.randn argument errors and StableDiffusionReferencePipeline list index errors are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 238, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10428", + "issue:10448", + "issue:11050", + "issue:1141", + "issue:11648", + "issue:11938", + "issue:12575", + "issue:1343", + "issue:13506", + "issue:1477", + "issue:1519", + "issue:1537", + "issue:201", + "issue:2161", + "issue:2203", + "issue:2377", + "issue:2440", + "issue:2518", + "issue:2552", + "issue:2737", + "issue:2772", + "issue:2784", + "issue:3043", + "issue:3073", + "issue:3694", + "issue:3986", + "issue:4213", + "issue:4488", + "issue:4593", + "issue:4846", + "issue:4912", + "issue:5515", + "issue:5553", + "issue:5731", + "issue:6321", + "issue:6463", + "issue:6561", + "issue:6742", + "issue:7012", + "issue:7519", + "issue:7661", + "issue:7841", + "issue:817", + "issue:8759", + "issue:9047", + "issue:9241", + "issue:9299", + "issue:9899" + ], + "soft_pairs": [ + "issue:2203|issue:817", + "issue:3694|issue:7012", + "issue:10428|issue:9299", + "issue:10448|issue:11648", + "issue:1141|issue:4912", + "issue:1343|issue:2518", + "issue:201|issue:2377", + "issue:2161|issue:2552", + "issue:5731|issue:6463", + "issue:6561|issue:7661", + "issue:7519|issue:7841", + "issue:9241|issue:9899", + "issue:5515|issue:5553", + "issue:12575|issue:13506", + "issue:1519|issue:1537", + "issue:1477|issue:2440", + "issue:11938|issue:9047", + "issue:11050|issue:8759", + "issue:2784|issue:4846", + "issue:4488|issue:6742", + "issue:2737|issue:3986", + "issue:2772|issue:4593", + "issue:3043|issue:3073", + "issue:4213|issue:6321" + ], + "prepared_review_unit_hash": "74e5c77475202ba1ac523cf314847a70f112ceb45ca1ccdbcd677c5cd0ea4e74", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33587, + "estimated_input_tokens": 8397, + "estimated_eval_tokens": 17050 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated issues with superficial lexical overlap only. None of the soft pairs look like the same underlying bug/change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical duplicate target stands out here; the issues span unrelated subsystems, hardware backends, training scripts, and feature requests.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:12575 is the most concrete self-contained bug report in the set, with a specific subsystem ([Chroma]), a clear failure mode, and enough detail to serve as the best representative issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2203", + "right": "issue:817", + "accept": false, + "reason": "Different problems: model download location vs mixed-precision training failure." + }, + { + "left": "issue:3694", + "right": "issue:7012", + "accept": false, + "reason": "One is about missing model/config files during finetuning; the other is about bad outputs with v-prediction." + }, + { + "left": "issue:10428", + "right": "issue:9299", + "accept": false, + "reason": "Both mention Flux, but one is an Ascend NPU inference error and the other is about CUDAGraphs for position embeddings." + }, + { + "left": "issue:10448", + "right": "issue:11648", + "accept": false, + "reason": "DDUF mmap loading is unrelated to loading LoRA weights into an FP8 transformer model." + }, + { + "left": "issue:1141", + "right": "issue:4912", + "accept": false, + "reason": "Local-file support in custom pipelines is a different concern from converting CivitAI SDXL checkpoints." + }, + { + "left": "issue:1343", + "right": "issue:2518", + "accept": false, + "reason": "General xformers default enablement is not the same as an example script missing xformers support." + }, + { + "left": "issue:201", + "right": "issue:2377", + "accept": false, + "reason": "These are different test requests for different components: attention class vs EMAModel." + }, + { + "left": "issue:2161", + "right": "issue:2552", + "accept": false, + "reason": "Training LoRA on inpainting models is distinct from asking for tutorial/docs about StableDiffusionInpaintPipeline." + }, + { + "left": "issue:5731", + "right": "issue:6463", + "accept": false, + "reason": "AutoencoderKL logvar behavior is unrelated to the missing audio_diffusion pipeline." + }, + { + "left": "issue:6561", + "right": "issue:7661", + "accept": false, + "reason": "Missing IP-Adapter config.json is a loading error; the other is about enabling animate diff with an IP-adapter-loaded UNet." + }, + { + "left": "issue:7519", + "right": "issue:7841", + "accept": false, + "reason": "An ignored xformers keyword is not the same bug as SDXL images showing red dots when enlarged." + }, + { + "left": "issue:9241", + "right": "issue:9899", + "accept": false, + "reason": "Converting XLab LoRA to diffusers LoRA is unrelated to converting a transformer to a checkpoint." + }, + { + "left": "issue:5515", + "right": "issue:5553", + "accept": false, + "reason": "LCM scheduler initialization failure and custom model loading failure are different code-paths." + }, + { + "left": "issue:12575", + "right": "issue:13506", + "accept": false, + "reason": "Both concern attention backends, but one is a Chroma backend-check error and the other is Sage attention failing with ring-attention; not the same bug." + }, + { + "left": "issue:1519", + "right": "issue:1537", + "accept": false, + "reason": "Apple M2 batch-image noise and Apple MPS scheduler issues are different symptoms and likely different root causes." + }, + { + "left": "issue:1477", + "right": "issue:2440", + "accept": false, + "reason": "A CUDA cublas batched GEMM error is not the same as failing to load an inpainting model in float16." + }, + { + "left": "issue:11938", + "right": "issue:9047", + "accept": false, + "reason": "RMSNorm registration bug and Flux not working on MPS are unrelated." + }, + { + "left": "issue:11050", + "right": "issue:8759", + "accept": false, + "reason": "Mixed-precision casting in a training script is different from a runtime warning in SD3 image postprocessing." + }, + { + "left": "issue:2784", + "right": "issue:4846", + "accept": false, + "reason": "A feature request for text-to-video dreambooth support is not the same as a report of strange dreambooth training results." + }, + { + "left": "issue:4488", + "right": "issue:6742", + "accept": false, + "reason": "SDXL checkpoint-loading behavior and DownBlockMotion checkpointing differences are different issues." + }, + { + "left": "issue:2737", + "right": "issue:3986", + "accept": false, + "reason": "Scheduler support for CLIP guidance is unrelated to adding mesh output in ShapEPipeline." + }, + { + "left": "issue:2772", + "right": "issue:4593", + "accept": false, + "reason": "Local dataset handling in train_controlnet.py and LoRA rank mismatch loading are different bugs." + }, + { + "left": "issue:3043", + "right": "issue:3073", + "accept": false, + "reason": "Add ODISE and add consistency models are separate feature requests." + }, + { + "left": "issue:4213", + "right": "issue:6321", + "accept": false, + "reason": "SDXL from_single_file without accelerate and missing openai/clip-vit-large-patch14 are unrelated loading issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 239, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10757", + "issue:11275", + "issue:1188", + "issue:11922", + "issue:12653", + "issue:13015", + "issue:1895", + "issue:1969", + "issue:1992", + "issue:201", + "issue:2326", + "issue:2571", + "issue:321", + "issue:3222", + "issue:3377", + "issue:3422", + "issue:3487", + "issue:3582", + "issue:3829", + "issue:3974", + "issue:4171", + "issue:4185", + "issue:4194", + "issue:4304", + "issue:4436", + "issue:4764", + "issue:4913", + "issue:5150", + "issue:5168", + "issue:5605", + "issue:5849", + "issue:5968", + "issue:6141", + "issue:6383", + "issue:6412", + "issue:6530", + "issue:6928", + "issue:7061", + "issue:7421", + "issue:7568", + "issue:7877", + "issue:7985", + "issue:8234", + "issue:8463", + "issue:9062", + "issue:9511", + "issue:9611", + "issue:9889" + ], + "soft_pairs": [ + "issue:11275|issue:8463", + "issue:2326|issue:4764", + "issue:3377|issue:4304", + "issue:3422|issue:6928", + "issue:3829|issue:7421", + "issue:4171|issue:7061", + "issue:5150|issue:6383", + "issue:5968|issue:6412", + "issue:7568|issue:9889", + "issue:7985|issue:8234", + "issue:2571|issue:5849", + "issue:12653|issue:13015", + "issue:1188|issue:321", + "issue:1992|issue:201", + "issue:3487|issue:3582", + "issue:3974|issue:5168", + "issue:6141|issue:9511", + "issue:11922|issue:9611", + "issue:5605|issue:7877", + "issue:10757|issue:9062", + "issue:4194|issue:4436", + "issue:4913|issue:6530", + "issue:1895|issue:1969", + "issue:3222|issue:4185" + ], + "prepared_review_unit_hash": "75054b666d0b614e537ad451613fc6a10e3749663cebd12f9993504e18dfb37c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33836, + "estimated_input_tokens": 8459, + "estimated_eval_tokens": 17174 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The list is a grab bag of unrelated bug reports and feature requests. The soft-similarity pairs are mostly superficial keyword matches; none look like the same underlying issue or change.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11275", + "right": "issue:8463", + "accept": false, + "reason": "Both are runtime errors, but they affect different code paths: VAE ResnetUpsampleBlock2D addition vs InstructP2P-SDXL tokenizer handling." + }, + { + "left": "issue:2326", + "right": "issue:4764", + "accept": false, + "reason": "Both mention LoRA conversion/loading, but one is about conversion to safetensors/ckpt and the other about loading a specific OFT-generated LoRA file." + }, + { + "left": "issue:3377", + "right": "issue:4304", + "accept": false, + "reason": "Training-related topics, but one asks about SNR weighting optimization and the other is a request for SDXL LoRA support." + }, + { + "left": "issue:3422", + "right": "issue:6928", + "accept": false, + "reason": "Both are about pipeline options, but disabling the safety checker is unrelated to adding a depthmap colorization toggle." + }, + { + "left": "issue:3829", + "right": "issue:7421", + "accept": false, + "reason": "Both involve determinism/seed behavior, but one is ONNX EulerAncestral output variance and the other is KDiffusion pipeline seed handling." + }, + { + "left": "issue:4171", + "right": "issue:7061", + "accept": false, + "reason": "These are documentation issues, but they concern different pages and different defects: homepage reference rendering vs an incorrect community pipeline docstring." + }, + { + "left": "issue:5150", + "right": "issue:6383", + "accept": false, + "reason": "One is missing a pipeline method, the other is about scheduler learning-rate support; not the same bug." + }, + { + "left": "issue:5968", + "right": "issue:6412", + "accept": false, + "reason": "Both are conversion/export requests, but text/video ONNX conversion and ControlNet annotator ONNX support are distinct features." + }, + { + "left": "issue:7568", + "right": "issue:9889", + "accept": false, + "reason": "Both involve inference behavior, but interpolation token handling is unrelated to an sdxl-turbo segmentation fault under a specific torch version." + }, + { + "left": "issue:7985", + "right": "issue:8234", + "accept": false, + "reason": "IPAdapter-related issues, but one is a torch.compile incompatibility and the other is a scale-argument type error with IPAdapter Plus." + }, + { + "left": "issue:2571", + "right": "issue:5849", + "accept": false, + "reason": "Both are import/CLI failures, but they stem from different causes: accelerate ProjectConfiguration import breakage vs a missing args parameter in an SDXL LoRA training entrypoint." + }, + { + "left": "issue:12653", + "right": "issue:13015", + "accept": false, + "reason": "Both are device/dtype portability problems, but one is a float64 rope reference breaking MPS and the other is an NPU tensor device mismatch in QwenImageEditPlus." + }, + { + "left": "issue:1188", + "right": "issue:321", + "accept": false, + "reason": "Different feature requests: adding a CLIPSeg+Stable Diffusion pipeline versus a community blended diffusion request." + }, + { + "left": "issue:1992", + "right": "issue:201", + "accept": false, + "reason": "One is a JAX attention refactor question, the other is a tests request for an attention class; not the same underlying change." + }, + { + "left": "issue:3487", + "right": "issue:3582", + "accept": false, + "reason": "Both involve inpaint strength behavior, but one reports poor results with a scheduler and the other says strength=1 does not work in ControlNet inpaint." + }, + { + "left": "issue:3974", + "right": "issue:5168", + "accept": false, + "reason": "Both mention ControlNet, but one is safetensors loading support and the other is ControlNet-XS support; different features." + }, + { + "left": "issue:6141", + "right": "issue:9511", + "accept": false, + "reason": "Both are about ControlNet options, but control_mode/resize_mode support is unrelated to multi-controlnet batching." + }, + { + "left": "issue:11922", + "right": "issue:9611", + "accept": false, + "reason": "Both are Flux pipeline problems, but one is a generic FLUX.1-Fill-dev run issue and the other is a specific batching error in img2img/inpaint pipelines." + }, + { + "left": "issue:5605", + "right": "issue:7877", + "accept": false, + "reason": "Both request pipeline/block extensions, but per-image caption support for SDXL DreamBooth is unrelated to adding an out_dim argument to Attention." + }, + { + "left": "issue:10757", + "right": "issue:9062", + "accept": false, + "reason": "Both are about names/imports, but one is a variable rename in a conversion script and the other is an import failure for StableAudioPipeline." + }, + { + "left": "issue:4194", + "right": "issue:4436", + "accept": false, + "reason": "Both are conversion/loading issues, but from_single_file VAE behavior and converting CIVITAI safetensors with VAE are different problems." + }, + { + "left": "issue:4913", + "right": "issue:6530", + "accept": false, + "reason": "Both mention ControlNet/SDXL, but one is a model/pipeline request and the other is about K-diffusion support with ControlNet." + }, + { + "left": "issue:1895", + "right": "issue:1969", + "accept": false, + "reason": "Both are runtime/training errors, but EMA multi-GPU update logic and tensor view/stride compatibility are unrelated." + }, + { + "left": "issue:3222", + "right": "issue:4185", + "accept": false, + "reason": "Both involve inpainting/control-related behavior, but missing custom diffusion output files and SDXL ControlNet black-image inference are not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 240, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10194", + "issue:10311", + "issue:1062", + "issue:10653", + "issue:10705", + "issue:10891", + "issue:10893", + "issue:11218", + "issue:11989", + "issue:12141", + "issue:12412", + "issue:13053", + "issue:13191", + "issue:1594", + "issue:1643", + "issue:1730", + "issue:1831", + "issue:1976", + "issue:2311", + "issue:2645", + "issue:2841", + "issue:2875", + "issue:2883", + "issue:3336", + "issue:3591", + "issue:3771", + "issue:3826", + "issue:3975", + "issue:4217", + "issue:4376", + "issue:4551", + "issue:460", + "issue:4773", + "issue:4782", + "issue:4815", + "issue:4826", + "issue:5002", + "issue:5166", + "issue:5605", + "issue:5719", + "issue:696", + "issue:7080", + "issue:7408", + "issue:834", + "issue:8752", + "issue:9063", + "issue:911", + "issue:9188" + ], + "soft_pairs": [ + "issue:3771|issue:7080", + "issue:10653|issue:11989", + "issue:2645|issue:4826", + "issue:4782|issue:4815", + "issue:10194|issue:9188", + "issue:10311|issue:11218", + "issue:10893|issue:13053", + "issue:1594|issue:2883", + "issue:1976|issue:911", + "issue:2311|issue:3336", + "issue:3826|issue:7408", + "issue:1062|issue:1730", + "issue:10705|issue:9063", + "issue:12412|issue:13191", + "issue:2875|issue:3591", + "issue:4376|issue:8752", + "issue:4217|issue:4773", + "issue:10891|issue:12141", + "issue:2841|issue:5605", + "issue:3975|issue:5719", + "issue:1643|issue:696", + "issue:460|issue:834", + "issue:5002|issue:5166", + "issue:1831|issue:4551" + ], + "prepared_review_unit_hash": "7cf54367f6b8de04ea1cb2e501b0db05c9dd61343c6f903acdb353aa5e9b3698", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33643, + "estimated_input_tokens": 8411, + "estimated_eval_tokens": 17078 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set; it mixes many unrelated feature requests and bugs across Dreambooth, ControlNet, schedulers, Flux, video pipelines, and loading/import errors. Most soft pairs share only very broad subsystem words and should not be merged.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 10194 is the broadest, still-open request in the set and best serves as a loose representative for the Flux-related subset, though the cluster overall is heterogeneous.", + "canonical_pr_reason": null, + "best_issue_reason": "10194 is the most general and current issue among the items; it is a reasonable anchor only because no tighter canonical duplicate exists.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3771", + "right": "issue:7080", + "accept": false, + "reason": "Both mention inference/output problems, but they concern different pipelines and failure modes; not the same bug." + }, + { + "left": "issue:10653", + "right": "issue:11989", + "accept": false, + "reason": "An 8-bit .to(cpu) limitation and a black-image rendering issue are unrelated symptoms in different code paths." + }, + { + "left": "issue:2645", + "right": "issue:4826", + "accept": false, + "reason": "Both are ControlNet-related, but one is a Mac/CUDA assertion and the other is a MultiControlNet/reference error; different bugs." + }, + { + "left": "issue:4782", + "right": "issue:4815", + "accept": false, + "reason": "Inpainting mask corruption and latents-only input support are different change requests, not the same underlying issue." + }, + { + "left": "issue:10194", + "right": "issue:9188", + "accept": false, + "reason": "Both are Flux-related feature requests, but attention masking and outpainting are distinct capabilities." + }, + { + "left": "issue:10311", + "right": "issue:11218", + "accept": false, + "reason": "These are separate import errors involving different modules and missing symbols." + }, + { + "left": "issue:10893", + "right": "issue:13053", + "accept": false, + "reason": "Unrelated model-support feature requests for different pipelines." + }, + { + "left": "issue:1594", + "right": "issue:2883", + "accept": false, + "reason": "Both concern video generation, but one asks for img2img support and the other for an image-to-video pipeline; not the same request." + }, + { + "left": "issue:1976", + "right": "issue:911", + "accept": false, + "reason": "Both are config-related, but one is a tokenizer file lookup error and the other is a Dreambooth bad-config report." + }, + { + "left": "issue:2311", + "right": "issue:3336", + "accept": false, + "reason": "Different scheduler feature requests; they are not the same algorithm or code-path change." + }, + { + "left": "issue:3826", + "right": "issue:7408", + "accept": false, + "reason": "Both are new pipeline requests, but for entirely different pipelines and use cases." + }, + { + "left": "issue:1062", + "right": "issue:1730", + "accept": false, + "reason": "Dreambooth-related, but one is about reproducing paper results and the other about CKPT conversion/webui inference." + }, + { + "left": "issue:10705", + "right": "issue:9063", + "accept": false, + "reason": "Pipeline reuse slowdown and Flux tokenizer instantiation errors are unrelated." + }, + { + "left": "issue:12412", + "right": "issue:13191", + "accept": false, + "reason": "Separate support requests for different model families; no shared bug or change." + }, + { + "left": "issue:2875", + "right": "issue:3591", + "accept": false, + "reason": "Training loss instability and ControlNet loss-curve confusion are different problems." + }, + { + "left": "issue:4376", + "right": "issue:8752", + "accept": false, + "reason": "Both are SDXL pipeline issues, but one is textual inversion crashing and the other is offline from_single_file loading." + }, + { + "left": "issue:4217", + "right": "issue:4773", + "accept": false, + "reason": "Both involve loading, but one is safetensors load time and the other is CivitAI LoRA loading; different issues." + }, + { + "left": "issue:10891", + "right": "issue:12141", + "accept": false, + "reason": "Both mention bfloat16, but they affect different models/pipelines and likely different dtype handling paths." + }, + { + "left": "issue:2841", + "right": "issue:5605", + "accept": false, + "reason": "Unrelated feature requests: scheduler sampling temperature vs SDXL Dreambooth captions." + }, + { + "left": "issue:3975", + "right": "issue:5719", + "accept": false, + "reason": "A typo in an error message and LoRA file-selection logic are unrelated." + }, + { + "left": "issue:1643", + "right": "issue:696", + "accept": false, + "reason": "Both are Dreambooth reports, but one is an xFormers API mismatch and the other is CUDA OOM." + }, + { + "left": "issue:460", + "right": "issue:834", + "accept": false, + "reason": "One is an ONNX runtime dependency problem; the other is a generic CPU utilization complaint." + }, + { + "left": "issue:5002", + "right": "issue:5166", + "accept": false, + "reason": "Different scheduler/VQ-VAE issues with no shared concrete bug." + }, + { + "left": "issue:1831", + "right": "issue:4551", + "accept": false, + "reason": "Both involve Dreambooth, but one is class-image precision mismatch and the other is an SDXL Dreambooth crash." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 241, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10227", + "issue:10710", + "issue:10928", + "issue:1106", + "issue:1156", + "issue:1478", + "issue:1657", + "issue:1805", + "issue:1909", + "issue:2083", + "issue:2214", + "issue:2277", + "issue:2447", + "issue:2480", + "issue:2769", + "issue:3113", + "issue:322", + "issue:3564", + "issue:3963", + "issue:4042", + "issue:4349", + "issue:4638", + "issue:4704", + "issue:4727", + "issue:5079", + "issue:5412", + "issue:5476", + "issue:5616", + "issue:5953", + "issue:5984", + "issue:6093", + "issue:6201", + "issue:6372", + "issue:6427", + "issue:6513", + "issue:684", + "issue:713", + "issue:7191", + "issue:7276", + "issue:7390", + "issue:7454", + "issue:8228", + "issue:8495", + "issue:8588", + "issue:8671", + "issue:9078", + "issue:9181", + "issue:9241" + ], + "soft_pairs": [ + "issue:7191|issue:9181", + "issue:5412|issue:5984", + "issue:7390|issue:8228", + "issue:1156|issue:684", + "issue:10227|issue:10710", + "issue:5616|issue:6427", + "issue:10928|issue:8495", + "issue:1106|issue:2480", + "issue:1657|issue:5476", + "issue:1805|issue:322", + "issue:1909|issue:2083", + "issue:2214|issue:2769", + "issue:2277|issue:6093", + "issue:2447|issue:713", + "issue:3113|issue:4042", + "issue:3564|issue:6372", + "issue:4349|issue:6201", + "issue:4638|issue:9078", + "issue:5079|issue:6513", + "issue:7454|issue:9241", + "issue:8588|issue:8671", + "issue:1478|issue:5953", + "issue:4704|issue:7276", + "issue:3963|issue:4727" + ], + "prepared_review_unit_hash": "0bd595aef088b0df683159f90193218d8302542a550451a593d6cc804a983d2a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33741, + "estimated_input_tokens": 8436, + "estimated_eval_tokens": 17128 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are only superficially similar; they concern different models, pipelines, or failure modes. The only clear duplicate is the multiple-generator/DDIMScheduler issue pair 1909\u21942083.", + "confidence": 0.92, + "canonical_issue_reason": "issue:1909 is the best canonical issue: it states the broad root problem (pipelines accept a list of generators, but schedulers do not), while issue:2083 is a narrower DDIMScheduler/eta>0 manifestation.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:1909 is the strongest global representative because it is the most general and actionable description of the shared bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7191", + "right": "issue:9181", + "accept": false, + "reason": "Different problems: an import failure versus a ControlNet pipeline artifact/noise issue." + }, + { + "left": "issue:5412", + "right": "issue:5984", + "accept": false, + "reason": "Both are DreamBooth/SDXL-related, but one is resume_from_checkpoint failure and the other is fp16 RAM usage." + }, + { + "left": "issue:7390", + "right": "issue:8228", + "accept": false, + "reason": "Both mention from_single_file, but they describe different load failures on different model components and paths." + }, + { + "left": "issue:1156", + "right": "issue:684", + "accept": false, + "reason": "Generic training failure vs AMD memory-optimization problems; no shared concrete bug." + }, + { + "left": "issue:10227", + "right": "issue:10710", + "accept": false, + "reason": "LoRA feature-support error and DDUF format support are unrelated." + }, + { + "left": "issue:5616", + "right": "issue:6427", + "accept": false, + "reason": "Both are model conversion/loading complaints, but they target different workflows and failure points." + }, + { + "left": "issue:10928", + "right": "issue:8495", + "accept": false, + "reason": "Missing fp16 weight file versus SD3 tokenizer enum parsing error; unrelated." + }, + { + "left": "issue:1106", + "right": "issue:2480", + "accept": false, + "reason": "Repeated UNet calls in PNDM is unrelated to a DreamBooth script checkpoint folder issue." + }, + { + "left": "issue:1657", + "right": "issue:5476", + "accept": false, + "reason": "Kaggle examples request versus xformers portability question; not the same bug or change." + }, + { + "left": "issue:1805", + "right": "issue:322", + "accept": false, + "reason": "Different feature requests for different diffusion approaches." + }, + { + "left": "issue:1909", + "right": "issue:2083", + "accept": true, + "reason": "Same underlying bug: multiple generators are passed through pipelines, but DDIMScheduler/eta>0 handling only supports a single generator." + }, + { + "left": "issue:2214", + "right": "issue:2769", + "accept": false, + "reason": "NSFW filter regression is unrelated to conversion round-tripping breaking a model." + }, + { + "left": "issue:2277", + "right": "issue:6093", + "accept": false, + "reason": "UNet config length validation issue versus push_to_hub behavior in a training script." + }, + { + "left": "issue:2447", + "right": "issue:713", + "accept": false, + "reason": "LoRA-specific error versus generic training error; too little concrete overlap." + }, + { + "left": "issue:3113", + "right": "issue:4042", + "accept": false, + "reason": "Both involve conditioning/control features, but they are distinct feature requests with different goals." + }, + { + "left": "issue:3564", + "right": "issue:6372", + "accept": false, + "reason": "Uni-ControlNet support and SVD+ControlNet are different pipeline features." + }, + { + "left": "issue:4349", + "right": "issue:6201", + "accept": false, + "reason": "Both touch prompt_embeds handling, but they affect different pipelines and different logic paths." + }, + { + "left": "issue:4638", + "right": "issue:9078", + "accept": false, + "reason": "Textual inversion on SDXL and Flux LoRA loading are unrelated." + }, + { + "left": "issue:5079", + "right": "issue:6513", + "accept": false, + "reason": "LoRA loading support versus T2IAdapter model support are separate feature gaps." + }, + { + "left": "issue:7454", + "right": "issue:9241", + "accept": false, + "reason": "General LoRA support request versus a LoRA conversion script; not the same concrete change." + }, + { + "left": "issue:8588", + "right": "issue:8671", + "accept": false, + "reason": "Missing SD3 conversion script and redundant DreamBooth code are unrelated." + }, + { + "left": "issue:1478", + "right": "issue:5953", + "accept": false, + "reason": "Upscale pipeline output-count bug versus ControlNet image-size mismatch crash." + }, + { + "left": "issue:4704", + "right": "issue:7276", + "accept": false, + "reason": "Custom diffusion training failure versus DreamBooth example network errors; different scripts and failure modes." + }, + { + "left": "issue:3963", + "right": "issue:4727", + "accept": false, + "reason": "Auto-download on Windows and local_files_only auto-pipeline handling are different download-related issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The conclusions are well grounded in the packet and stay conservative. The 1909\u21942083 pair is clearly the strongest duplicate-like match from the titles, and the other rejected pairs mostly do appear to involve different bugs, workflows, or components. I don\u2019t see any overstatement that would require rejection." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 242, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1033", + "issue:10448", + "issue:10520", + "issue:10798", + "issue:10954", + "issue:11033", + "issue:11307", + "issue:12037", + "issue:12052", + "issue:12423", + "issue:1246", + "issue:13034", + "issue:1315", + "issue:13162", + "issue:1375", + "issue:1572", + "issue:1934", + "issue:4141", + "issue:4375", + "issue:4399", + "issue:4453", + "issue:4480", + "issue:4506", + "issue:4726", + "issue:4806", + "issue:5016", + "issue:5574", + "issue:5685", + "issue:6201", + "issue:6398", + "issue:668", + "issue:7322", + "issue:8264", + "issue:8373", + "issue:8407", + "issue:8495", + "issue:8500", + "issue:8535", + "issue:8593", + "issue:8708", + "issue:8723", + "issue:8880", + "issue:9460", + "issue:964", + "issue:9686", + "issue:9902", + "issue:9942" + ], + "soft_pairs": [ + "issue:1033|issue:1572", + "issue:4806|issue:8373", + "issue:10520|issue:8264", + "issue:1572|issue:964", + "issue:10798|issue:13034", + "issue:12037|issue:13162", + "issue:1315|issue:1375", + "issue:11033|issue:12423", + "issue:4453|issue:5574", + "issue:9902|issue:9942", + "issue:5685|issue:8407", + "issue:4375|issue:8880", + "issue:4480|issue:4726", + "issue:5016|issue:7322", + "issue:4141|issue:4399", + "issue:1246|issue:668", + "issue:4506|issue:6201", + "issue:8495|issue:8593", + "issue:11307|issue:12052", + "issue:1934|issue:6398", + "issue:8500|issue:8535", + "issue:8708|issue:9686", + "issue:8723|issue:9460", + "issue:10448|issue:10954" + ], + "prepared_review_unit_hash": "98a1cf8a7810117c20f10f474855b72d7a240e4d6c7909d70e6294384b2a98a5", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33045, + "estimated_input_tokens": 8262, + "estimated_eval_tokens": 16780 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a grab-bag of unrelated issues that only share superficial terms like offloading, loading, device mismatch, or model names. I do not see any true duplicate pairs; all soft links look too weak and cross different code paths or different models.", + "confidence": 0.88, + "canonical_issue_reason": "Issue #12052 is the strongest anchor if one must be chosen: it is concrete, well-scoped, and describes a specific device-mismatch failure in a clearly identified Wan 2.2 + LightX2V offloading path.", + "canonical_pr_reason": null, + "best_issue_reason": "#12052 is the most actionable and diagnostic issue in the set, with a clear repro context and failure mode; it is a better representative than the more generic or title-only reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1033", + "right": "issue:1572", + "accept": false, + "reason": "Same broad Stable Diffusion Mega area, but different failure reports: an IndexError versus img2img/inpainting not working. Too weak to be the same bug." + }, + { + "left": "issue:4806", + "right": "issue:8373", + "accept": false, + "reason": "One is a type annotation/API issue on UNet2DModel; the other is an incorrect error message. Different bug classes and code paths." + }, + { + "left": "issue:10520", + "right": "issue:8264", + "accept": false, + "reason": "Runtime device mismatch in Sana 4K is unrelated to a training job stopping at 0%. No clear shared root cause." + }, + { + "left": "issue:1572", + "right": "issue:964", + "accept": false, + "reason": "Both mention inpainting, but one is a StableDiffusion Mega img2img/inpainting failure and the other is a generic inpainting example issue. Not enough evidence for duplication." + }, + { + "left": "issue:10798", + "right": "issue:13034", + "accept": false, + "reason": "Completely different topics: Flux quantization/device combinations versus WanPipeline height/width handling." + }, + { + "left": "issue:12037", + "right": "issue:13162", + "accept": false, + "reason": "Both involve LoRA support, but for different models and different compatibility problems. Not the same concrete bug." + }, + { + "left": "issue:1315", + "right": "issue:1375", + "accept": false, + "reason": "Import error from a missing symbol is not the same as a PR allegedly breaking imports; different symptom and likely different root cause." + }, + { + "left": "issue:11033", + "right": "issue:12423", + "accept": false, + "reason": "Both are from_single_file loading issues, but for different models (SD1.5 UNet vs QwenImageTransformer2DModel). Different code paths." + }, + { + "left": "issue:4453", + "right": "issue:5574", + "accept": false, + "reason": "LoRA conversion for SDXL safetensors is unrelated to training a subset of UNet attention parameters with LoRA." + }, + { + "left": "issue:9902", + "right": "issue:9942", + "accept": false, + "reason": "A conda version/install problem is not the same as a pip install failure for a diffusers dev version." + }, + { + "left": "issue:5685", + "right": "issue:8407", + "accept": false, + "reason": "Generic scheduler concerns versus a specific UniPC set_timesteps kwargs compatibility bug. Too broad to merge." + }, + { + "left": "issue:4375", + "right": "issue:8880", + "accept": false, + "reason": "Torch 2 compile crash during inference and inference slowdown are different symptoms; no concrete same bug path." + }, + { + "left": "issue:4480", + "right": "issue:4726", + "accept": false, + "reason": "MPS kernel crash on Mac M1 with fp16 is unrelated to a callback type issue." + }, + { + "left": "issue:5016", + "right": "issue:7322", + "accept": false, + "reason": "Training text encoders in SDXL LoRA is a functionality bug; SIGKILL during training is a resource/crash symptom. Not the same issue." + }, + { + "left": "issue:4141", + "right": "issue:4399", + "accept": false, + "reason": "Second controlnet handling and torch compile with CPU offload are separate problems in the broader pipeline ecosystem." + }, + { + "left": "issue:1246", + "right": "issue:668", + "accept": false, + "reason": "Both are dreambooth runtime problems, but one is an FP16 gradient/unscale issue and the other is a Colab Tesla T4 runtime error. Not the same bug." + }, + { + "left": "issue:4506", + "right": "issue:6201", + "accept": false, + "reason": "Both involve prompt embeddings, but one says a variable is overwritten while the other is about not re-encoding when prompt is absent. Different concrete behaviors." + }, + { + "left": "issue:8495", + "right": "issue:8593", + "accept": false, + "reason": "SD3 tokenizer parsing error and SD3 bfloat16 CPU noise are unrelated issues." + }, + { + "left": "issue:11307", + "right": "issue:12052", + "accept": false, + "reason": "Both mention offloading and device mismatches, but they affect different models and different tensor operations. Not mergeable as one PR-level fix." + }, + { + "left": "issue:1934", + "right": "issue:6398", + "accept": false, + "reason": "Memory usage with cpu_offload and a vague memory leak report are too generic and not clearly the same bug." + }, + { + "left": "issue:8500", + "right": "issue:8535", + "accept": false, + "reason": "Token limit questions and enabling xformers memory-efficient attention are unrelated." + }, + { + "left": "issue:8708", + "right": "issue:9686", + "accept": false, + "reason": "Missing SD3 dreambooth config field and a ControlNet pooled_projections bug are different features and code paths." + }, + { + "left": "issue:8723", + "right": "issue:9460", + "accept": false, + "reason": "SD3ControlNetModel initialization failure versus supporting a list of controlnets in the pipeline are different change requests." + }, + { + "left": "issue:10448", + "right": "issue:10954", + "accept": false, + "reason": "DDUF mmap loading and FLUX LoRA loading are unrelated model/loading concerns." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 243, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10227", + "issue:10470", + "issue:10596", + "issue:10950", + "issue:11093", + "issue:11127", + "issue:11221", + "issue:11314", + "issue:11555", + "issue:11648", + "issue:12738", + "issue:12905", + "issue:13073", + "issue:13104", + "issue:13286", + "issue:1398", + "issue:1439", + "issue:1519", + "issue:1805", + "issue:1918", + "issue:2439", + "issue:2500", + "issue:2581", + "issue:2646", + "issue:2663", + "issue:2712", + "issue:272", + "issue:3120", + "issue:321", + "issue:3694", + "issue:4031", + "issue:4086", + "issue:4282", + "issue:4312", + "issue:525", + "issue:6146", + "issue:6357", + "issue:6372", + "issue:7054", + "issue:7079", + "issue:7960", + "issue:8429", + "issue:8493", + "issue:8894", + "issue:9387", + "issue:969", + "issue:9690" + ], + "soft_pairs": [ + "issue:11314|issue:12905", + "issue:10470|issue:11093", + "issue:11555|issue:13286", + "issue:12738|issue:13073", + "issue:11221|issue:7960", + "issue:2646|issue:2663", + "issue:6146|issue:8894", + "issue:1918|issue:2439", + "issue:2581|issue:4086", + "issue:2712|issue:3694", + "issue:272|issue:4312", + "issue:6372|issue:9387", + "issue:10227|issue:11555", + "issue:10950|issue:8429", + "issue:1398|issue:1519", + "issue:4031|issue:6357", + "issue:11127|issue:13104", + "issue:3120|issue:4282", + "issue:2500|issue:969", + "issue:10596|issue:7079", + "issue:1439|issue:525", + "issue:1805|issue:321", + "issue:7054|issue:8493", + "issue:11648|issue:9690" + ], + "prepared_review_unit_hash": "8e485a89b2dbb69cb53e763e7dde7b517ae2478940327a2aac0735639afd6132", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32993, + "estimated_input_tokens": 8249, + "estimated_eval_tokens": 16754 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a loose thematic grouping around Flux/torchao, LoRA, and ZImage issues, but the soft pairs do not look like true duplicates; they describe different failure modes, pipelines, or docs/example problems rather than one shared bug.", + "confidence": 0.74, + "canonical_issue_reason": "issue:10470 is the broadest concrete bug report in the set, centered on Flux/torchao inference failure; the other issues are either narrower variants (LoRA, fp8, guidance, batch size) or clearly different subsystems.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10470 is the best representative issue because it captures the widest underlying Flux/torchao inference problem without being tied to a very specific symptom or example cleanup task.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11314", + "right": "issue:12905", + "accept": false, + "reason": "Both are Z-Image/Flux-style issues, but one is a controlnet training-script mistake and the other is a guidance-scale implementation bug; different code paths and fixes." + }, + { + "left": "issue:10470", + "right": "issue:11093", + "accept": false, + "reason": "Both mention torchao/Flux examples, but one is inference not working and the other is a broken example; too vague to treat as the same concrete bug." + }, + { + "left": "issue:11555", + "right": "issue:13286", + "accept": false, + "reason": "device_map='auto' support is an API/feature request, while torchao>=0.16.0 quantization not supported is a version-compatibility bug; not the same issue." + }, + { + "left": "issue:12738", + "right": "issue:13073", + "accept": false, + "reason": "Batch-size assertion failure and guidance_scale>1 support are different runtime conditions and different failures." + }, + { + "left": "issue:11221", + "right": "issue:7960", + "accept": false, + "reason": "Non-convergence in autoencoderkl training and SDXL multi-GPU training failure are unrelated training problems." + }, + { + "left": "issue:2646", + "right": "issue:2663", + "accept": false, + "reason": "Tiled VAE IndexError and safety-checker loading from safetensors are separate components and separate bugs." + }, + { + "left": "issue:6146", + "right": "issue:8894", + "accept": false, + "reason": "Both are training-script complaints, but they concern different scripts and failure modes, not one shared defect." + }, + { + "left": "issue:1918", + "right": "issue:2439", + "accept": false, + "reason": "DDIM scheduler behavior for a ddpm example is unrelated to inpainting pipeline loading/running issues." + }, + { + "left": "issue:2581", + "right": "issue:4086", + "accept": false, + "reason": "ControlNet checkpoint conversion and SDXL ControlNet training are different problems with different workflows." + }, + { + "left": "issue:2712", + "right": "issue:3694", + "accept": false, + "reason": "Missing config keys during model init and missing model_index/config files during finetuning are different loading/configuration errors." + }, + { + "left": "issue:272", + "right": "issue:4312", + "accept": false, + "reason": "K-LMS support for inpainting and a multi-ControlNet pipeline request are distinct feature requests." + }, + { + "left": "issue:6372", + "right": "issue:9387", + "accept": false, + "reason": "SVD+ControlNet and MimicMotion are unrelated pipeline feature requests." + }, + { + "left": "issue:10227", + "right": "issue:11555", + "accept": false, + "reason": "LoRA loading into fp8/quantized modules is a concrete model-compatibility error, while device_map='auto' is a pipeline placement feature request." + }, + { + "left": "issue:10950", + "right": "issue:8429", + "accept": false, + "reason": "Broken Accelerate docs and a malfunctioning I2VGen-XL inference script are unrelated." + }, + { + "left": "issue:1398", + "right": "issue:1519", + "accept": false, + "reason": "Both involve Apple Silicon/MPS, but one is SD2 support and the other is noisy batch outputs on M2; different bugs." + }, + { + "left": "issue:4031", + "right": "issue:6357", + "accept": false, + "reason": "Local safety-checker loading failure is not the same as API vs local result differences." + }, + { + "left": "issue:11127", + "right": "issue:13104", + "accept": false, + "reason": "A CivitAI Flux single-file loading problem and an undefined logger error are unrelated issues." + }, + { + "left": "issue:3120", + "right": "issue:4282", + "accept": false, + "reason": "Always-same-result ControlNet training and adding a Multi-ControlNet pipeline are different topics." + }, + { + "left": "issue:2500", + "right": "issue:969", + "accept": false, + "reason": "Meta tensor copy errors and multi-GPU accelerator test failures are not the same underlying bug." + }, + { + "left": "issue:10596", + "right": "issue:7079", + "accept": false, + "reason": "FluxControlPipeline shape errors and Transformer2DModel patch/norm_type errors are different model-path failures." + }, + { + "left": "issue:1439", + "right": "issue:525", + "accept": false, + "reason": "FeatureExtractionMixin import error and local-disk image loading in a notebook are unrelated." + }, + { + "left": "issue:1805", + "right": "issue:321", + "accept": false, + "reason": "Scalable Diffusion pipeline proposal and Blended Diffusion are different feature ideas." + }, + { + "left": "issue:7054", + "right": "issue:8493", + "accept": false, + "reason": "State-dict mutation in LoraLoaderMixin and SD3Transformer2DModel load errors are different loading bugs." + }, + { + "left": "issue:11648", + "right": "issue:9690", + "accept": false, + "reason": "LoRA weight loading with fp8 transformer models and Flux FP8 example cleanup are not the same bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 244, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10085", + "issue:10150", + "issue:10475", + "issue:10672", + "issue:1109", + "issue:12359", + "issue:12367", + "issue:13477", + "issue:1643", + "issue:1976", + "issue:2177", + "issue:2375", + "issue:2420", + "issue:2507", + "issue:2783", + "issue:2961", + "issue:3116", + "issue:3455", + "issue:3672", + "issue:3799", + "issue:3970", + "issue:4061", + "issue:4148", + "issue:4312", + "issue:5283", + "issue:5334", + "issue:5462", + "issue:5821", + "issue:5894", + "issue:6418", + "issue:6700", + "issue:7061", + "issue:7450", + "issue:7490", + "issue:7640", + "issue:785", + "issue:8803", + "issue:8805", + "issue:8876", + "issue:8970", + "issue:8997", + "issue:941", + "issue:954", + "issue:9844", + "issue:9926", + "issue:9958", + "issue:9972" + ], + "soft_pairs": [ + "issue:6700|issue:7640", + "issue:5894|issue:9972", + "issue:2375|issue:785", + "issue:2177|issue:954", + "issue:2783|issue:5462", + "issue:2961|issue:3455", + "issue:3116|issue:941", + "issue:4148|issue:4312", + "issue:5283|issue:6418", + "issue:5334|issue:8803", + "issue:5821|issue:7490", + "issue:8803|issue:8805", + "issue:1109|issue:2507", + "issue:10085|issue:10475", + "issue:3672|issue:7450", + "issue:3970|issue:4061", + "issue:8970|issue:8997", + "issue:12359|issue:13477", + "issue:10672|issue:12367", + "issue:9926|issue:9958", + "issue:1643|issue:2420", + "issue:10150|issue:9844", + "issue:1976|issue:3799", + "issue:7061|issue:8876" + ], + "prepared_review_unit_hash": "58841e656c59ab82b0ff87f6657cd9572473638ea440050b46b95581d6d6e098", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33099, + "estimated_input_tokens": 8275, + "estimated_eval_tokens": 16806 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a mixed bag of unrelated issues; the soft pairs mostly share only a broad subsystem or symptom word, not the same underlying bug. I would not merge any of the candidate pairs as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "issue:3672 is the strongest representative issue in this set: it has the highest discussion/inbound-reference activity and describes a concrete, well-scoped StableDiffusionPipeline concurrency failure.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:3672 is the most substantive and high-signal issue here, so it is the best overall issue representative for the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6700", + "right": "issue:7640", + "accept": false, + "reason": "Different failures: attention-processor temb TypeErrors vs SDXL from_single_file shape/dimension runtime error. Same general area, not the same bug." + }, + { + "left": "issue:5894", + "right": "issue:9972", + "accept": false, + "reason": "Both mention buffer sizes, but 5894 is a generic invalid-buffer error while 9972 is a MacOS-specific huge allocation/request issue. Not the same code-path problem." + }, + { + "left": "issue:2375", + "right": "issue:785", + "accept": false, + "reason": "safe_open NameError and PNDM scheduler performance are unrelated." + }, + { + "left": "issue:2177", + "right": "issue:954", + "accept": false, + "reason": "AutoencoderKL loading failure vs login/load issue; no shared bug mechanism." + }, + { + "left": "issue:2783", + "right": "issue:5462", + "accept": false, + "reason": "Both are img2img-related feature requests, but they target different pipeline additions and are not the same change." + }, + { + "left": "issue:2961", + "right": "issue:3455", + "accept": false, + "reason": "Training-example request for StableUnCLIP vs a separate instruct-pix2pix training request; different features." + }, + { + "left": "issue:3116", + "right": "issue:941", + "accept": false, + "reason": "Both report noise in img2img, but one is scheduler/inference-step specific and the other is MPS/num_images_per_prompt specific. Distinct bugs." + }, + { + "left": "issue:4148", + "right": "issue:4312", + "accept": false, + "reason": "SDXL inpainting config error vs SAG support request for inpainting/img2img. Not the same underlying issue." + }, + { + "left": "issue:5283", + "right": "issue:6418", + "accept": false, + "reason": "Completely different topics: media track merging vs stable video diffusion frame collapse." + }, + { + "left": "issue:5334", + "right": "issue:8803", + "accept": false, + "reason": "Accelerate multi-model training API request vs attention_processor hidden_states duplicate assignment bug. Unrelated." + }, + { + "left": "issue:5821", + "right": "issue:7490", + "accept": false, + "reason": "Both mention backward, but one is a CUDA training failure and the other is a generic backward problem. Not enough evidence of the same bug." + }, + { + "left": "issue:8803", + "right": "issue:8805", + "accept": false, + "reason": "Attention processor hidden_states bug vs SDXL conversion script failure; different code paths." + }, + { + "left": "issue:1109", + "right": "issue:2507", + "accept": false, + "reason": "CPU core utilization issue vs safetensors GPU load slowness; unrelated." + }, + { + "left": "issue:10085", + "right": "issue:10475", + "accept": false, + "reason": "LyCORIS batch-inference/skip_guidance_layers runtime error vs SD3 inference quality comparison during finetuning. Different problem classes." + }, + { + "left": "issue:3672", + "right": "issue:7450", + "accept": false, + "reason": "Both are threading-related, but one is StableDiffusionPipeline concurrency and the other is AutoencoderKL multithreading. Not the same concrete bug." + }, + { + "left": "issue:3970", + "right": "issue:4061", + "accept": false, + "reason": "SDXL prompt embeddings bug vs SDXL LoRA loading error; same model family, different failure modes." + }, + { + "left": "issue:8970", + "right": "issue:8997", + "accept": false, + "reason": "Both involve Kolors, but one is black images on MPS and the other is a scheduler usage error. Distinct issues." + }, + { + "left": "issue:12359", + "right": "issue:13477", + "accept": false, + "reason": "Documentation bug for Chroma guidance_scale vs ErnieImage callback locals() scope bug; unrelated." + }, + { + "left": "issue:10672", + "right": "issue:12367", + "accept": false, + "reason": "Callback-on-step-end support request vs Qwen-Image Eligen LoRA support request; different feature work." + }, + { + "left": "issue:9926", + "right": "issue:9958", + "accept": false, + "reason": "gguf quantize/speedup support and AutoencoderKLTemporalDecoder tiling are unrelated feature requests." + }, + { + "left": "issue:1643", + "right": "issue:2420", + "accept": false, + "reason": "Both mention xformers, but one is a DreamBooth method-signature error and the other is a LoRA training example issue. Not the same bug." + }, + { + "left": "issue:10150", + "right": "issue:9844", + "accept": false, + "reason": "Slow loading/inference for SDXL pipelines vs NaNs from SDXL VAE encoder; different symptoms and likely different causes." + }, + { + "left": "issue:1976", + "right": "issue:3799", + "accept": false, + "reason": "Incorrect tokenizer file lookup vs incorrect inferred type for scheduler.from_pretrained; unrelated." + }, + { + "left": "issue:7061", + "right": "issue:8876", + "accept": false, + "reason": "Both are pipeline/doc/config style issues, but they concern different pipelines and different concrete fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 245, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10128", + "issue:10144", + "issue:10194", + "issue:10591", + "issue:10653", + "issue:1082", + "issue:11000", + "issue:11486", + "issue:11833", + "issue:11973", + "issue:12367", + "issue:1249", + "issue:12540", + "issue:12620", + "issue:12905", + "issue:13073", + "issue:1363", + "issue:1536", + "issue:1672", + "issue:1941", + "issue:2560", + "issue:2988", + "issue:3391", + "issue:3452", + "issue:3908", + "issue:4159", + "issue:4520", + "issue:4557", + "issue:4839", + "issue:5028", + "issue:5465", + "issue:590", + "issue:6091", + "issue:6906", + "issue:7341", + "issue:7428", + "issue:7722", + "issue:7773", + "issue:8237", + "issue:8317", + "issue:8589", + "issue:8708", + "issue:9477", + "issue:9495", + "issue:954", + "issue:967", + "issue:9902", + "issue:9936" + ], + "soft_pairs": [ + "issue:7341|issue:9902", + "issue:1363|issue:1536", + "issue:1941|issue:954", + "issue:10194|issue:12367", + "issue:1672|issue:6091", + "issue:3452|issue:6906", + "issue:8317|issue:8708", + "issue:4159|issue:5465", + "issue:8237|issue:9477", + "issue:4520|issue:5028", + "issue:7722|issue:9495", + "issue:12540|issue:13073", + "issue:11973|issue:9936", + "issue:2988|issue:590", + "issue:3391|issue:4557", + "issue:2560|issue:3908", + "issue:10128|issue:7428", + "issue:10144|issue:7773", + "issue:10591|issue:12905", + "issue:10653|issue:11833", + "issue:1082|issue:4839", + "issue:11000|issue:12620", + "issue:11486|issue:8589", + "issue:1249|issue:967" + ], + "prepared_review_unit_hash": "ca75944e1c39f868de19073887a41b7c3b78c6858ba411d041630c1d76655bbe", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33870, + "estimated_input_tokens": 8468, + "estimated_eval_tokens": 17192 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a set of largely unrelated issue reports; none of the soft pairs look like true duplicates or the same fix/change. Most are only loosely related by broad themes like training scripts, schedulers, or guidance-scale support.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7341", + "right": "issue:9902", + "accept": false, + "reason": "Different problems: a TorchVersion/Version comparison TypeError vs a Conda version pinning/stuck-version report." + }, + { + "left": "issue:1363", + "right": "issue:1536", + "accept": false, + "reason": "Both involve custom/compatibility issues, but one is ONNX long-prompt weight + LMS scheduler, the other is lpw_stable_diffusion compatibility with newer diffusers." + }, + { + "left": "issue:1941", + "right": "issue:954", + "accept": false, + "reason": "Unrelated failures: custom pipeline loading for UnCLIP vs a login/load-after-login issue." + }, + { + "left": "issue:10194", + "right": "issue:12367", + "accept": false, + "reason": "Different feature requests for different models/pipelines: Flux attention masking vs Qwen-Image Eligen LoRA support." + }, + { + "left": "issue:1672", + "right": "issue:6091", + "accept": false, + "reason": "Both mention size, but one is image size after conversion and the other is a consistency decoder memory issue with a size argument." + }, + { + "left": "issue:3452", + "right": "issue:6906", + "accept": false, + "reason": "DreamBooth/training-related, but one is a low-VRAM question and the other is a broken LoRA DreamBooth example." + }, + { + "left": "issue:8317", + "right": "issue:8708", + "accept": false, + "reason": "Both concern SD3/VAE details, but one is a missing generator arg in AsymmetricAutoencoderKL decode and the other is missing shift_factor in training examples." + }, + { + "left": "issue:4159", + "right": "issue:5465", + "accept": false, + "reason": "Different DiffEdit bugs: excessive VRAM usage vs inpaint_strength only working at 0.8." + }, + { + "left": "issue:8237", + "right": "issue:9477", + "accept": false, + "reason": "Different issues: SDXL LoRA key loading vs a GatheredParameters callable error." + }, + { + "left": "issue:4520", + "right": "issue:5028", + "accept": false, + "reason": "Both mention scale/LoRA, but the failing code paths are different: attention slicing with lora_scale vs SDXL reference pipeline hacked block forward args." + }, + { + "left": "issue:7722", + "right": "issue:9495", + "accept": false, + "reason": "Both involve IPAdapter, but they are different pipelines and likely different bugs." + }, + { + "left": "issue:12540", + "right": "issue:13073", + "accept": false, + "reason": "Both are guidance-related, but one is SanaPipeline embeddings support and the other is ZImageControlNetPipeline guidance_scale behavior." + }, + { + "left": "issue:11973", + "right": "issue:9936", + "accept": false, + "reason": "Different training issues: missing instance_prompt arg vs NCCL timeout in multi-GPU controlnet training." + }, + { + "left": "issue:2988", + "right": "issue:590", + "accept": false, + "reason": "Both scheduler-related, but one is a crash at equal train/inference steps and the other is a Karras noise formula typo." + }, + { + "left": "issue:3391", + "right": "issue:4557", + "accept": false, + "reason": "Completely unrelated: overexposure in DeepFloyd super-resolution vs a __getattr__ AttributeError." + }, + { + "left": "issue:2560", + "right": "issue:3908", + "accept": false, + "reason": "Different domains: deterministic outputs with generator lists vs Streamlit model caching." + }, + { + "left": "issue:10128", + "right": "issue:7428", + "accept": false, + "reason": "Both are feature requests, but for unrelated features: fastercache vs layer diffusion." + }, + { + "left": "issue:10144", + "right": "issue:7773", + "accept": false, + "reason": "Different video-generation quality issues in different models/pipelines." + }, + { + "left": "issue:10591", + "right": "issue:12905", + "accept": false, + "reason": "Both are 'wrong' behavior reports, but they concern different components: SD3 LoRA training script vs Z_Image guidance scale implementation." + }, + { + "left": "issue:10653", + "right": "issue:11833", + "accept": false, + "reason": "Different device-management bugs: bitsandbytes 8bit .to(cpu) limitation vs set_lora_device KeyError." + }, + { + "left": "issue:1082", + "right": "issue:4839", + "accept": false, + "reason": "Unrelated errors: DreamBooth CUDA matmul failure vs a lora_scale type error." + }, + { + "left": "issue:11000", + "right": "issue:12620", + "accept": false, + "reason": "No overlap in bug/change: missing shift argument vs missing release assets." + }, + { + "left": "issue:11486", + "right": "issue:8589", + "accept": false, + "reason": "Different training concerns: enabling xformers memory-efficient attention vs SD3 prior preservation support." + }, + { + "left": "issue:1249", + "right": "issue:967", + "accept": false, + "reason": "Both are ONNX/scheduler related, but the reported failures and likely fixes are different." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 246, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10231", + "issue:10287", + "issue:1185", + "issue:12633", + "issue:1274", + "issue:13243", + "issue:1704", + "issue:1730", + "issue:1896", + "issue:1992", + "issue:2306", + "issue:2775", + "issue:3176", + "issue:3363", + "issue:3420", + "issue:3584", + "issue:3765", + "issue:3769", + "issue:3887", + "issue:3915", + "issue:4480", + "issue:4542", + "issue:4680", + "issue:4702", + "issue:4747", + "issue:4841", + "issue:4964", + "issue:5226", + "issue:5390", + "issue:5546", + "issue:5570", + "issue:5628", + "issue:5689", + "issue:5989", + "issue:6068", + "issue:6234", + "issue:6359", + "issue:6529", + "issue:6548", + "issue:6593", + "issue:6598", + "issue:7368", + "issue:7490", + "issue:7665", + "issue:7836", + "issue:8637", + "issue:8994", + "issue:9933" + ], + "soft_pairs": [ + "issue:1704|issue:1992", + "issue:3176|issue:6359", + "issue:3887|issue:8637", + "issue:4542|issue:6548", + "issue:4680|issue:6529", + "issue:4702|issue:7368", + "issue:4841|issue:6234", + "issue:4964|issue:5390", + "issue:5226|issue:6593", + "issue:5689|issue:6068", + "issue:5989|issue:7836", + "issue:6598|issue:9933", + "issue:3363|issue:3584", + "issue:2775|issue:5570", + "issue:5546|issue:5628", + "issue:1274|issue:1896", + "issue:10231|issue:10287", + "issue:2306|issue:4747", + "issue:1185|issue:1730", + "issue:3765|issue:3769", + "issue:3420|issue:3915", + "issue:7490|issue:7665", + "issue:12633|issue:13243", + "issue:4480|issue:8994" + ], + "prepared_review_unit_hash": "be6a7961ce506ec8bbe7b6ee40e2b4236b37bbd0cebd534752b6d5abd86a65ff", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33721, + "estimated_input_tokens": 8431, + "estimated_eval_tokens": 17118 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are mostly broad thematic matches rather than true duplicates; I would reject every edge. The strongest standalone issue in this set is the SDXL/DDIM sampling bug, which is the most concrete and reproducible report.", + "confidence": 0.86, + "canonical_issue_reason": "issue:6068 is the most technically precise and self-contained bug report: it names the exact scheduler, the SDXL code path, and the incorrect-sampling symptom with substantial discussion.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6068 is the best representative issue because it is specific, reproducible, and directly tied to a single code path rather than a vague feature/question.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1704", + "right": "issue:1992", + "accept": false, + "reason": "Both mention attention, but one is a general question and the other is a JAX refactor topic; not the same bug/change." + }, + { + "left": "issue:3176", + "right": "issue:6359", + "accept": false, + "reason": "TensorRT demo failure and a missing scheduler method are unrelated code paths." + }, + { + "left": "issue:3887", + "right": "issue:8637", + "accept": false, + "reason": "A quality/effect complaint about SD pipelines is unrelated to an import error for PixArtSigmaPipeline." + }, + { + "left": "issue:4542", + "right": "issue:6548", + "accept": false, + "reason": "One is about SDXL img2img generating only one image, the other is an RGBA latent-handling bug; different failure modes." + }, + { + "left": "issue:4680", + "right": "issue:6529", + "accept": false, + "reason": "General SDXL inpainting fine-tuning and a Fooocus inpaint request are not the same concrete issue." + }, + { + "left": "issue:4702", + "right": "issue:7368", + "accept": false, + "reason": "Training a VQ-VAE example and asking about GLIGEN training are different model/workflow requests." + }, + { + "left": "issue:4841", + "right": "issue:6234", + "accept": false, + "reason": "ControlNet image-size trouble and a text-to-image SDXL training script issue do not share the same bug." + }, + { + "left": "issue:4964", + "right": "issue:5390", + "accept": false, + "reason": "Model conversion to Diffusers and unwanted text-encoder downloads with local_files_only are unrelated." + }, + { + "left": "issue:5226", + "right": "issue:6593", + "accept": false, + "reason": "Both touch LoRA support, but they target different pipeline classes and different implementation paths." + }, + { + "left": "issue:5689", + "right": "issue:6068", + "accept": false, + "reason": "Both are scheduler-quality complaints, but DPM++ residual noise and DDIM incorrect samples are distinct scheduler bugs." + }, + { + "left": "issue:5989", + "right": "issue:7836", + "accept": false, + "reason": "A safetensors.torch bug and CLIP example overfitting are unrelated." + }, + { + "left": "issue:6598", + "right": "issue:9933", + "accept": false, + "reason": "Img2img parity differences and missing width/height parameters are different pipeline issues." + }, + { + "left": "issue:3363", + "right": "issue:3584", + "accept": false, + "reason": "Both involve DreamBooth, but one is a LoRA training failure and the other is a tutorial failure; not the same bug." + }, + { + "left": "issue:2775", + "right": "issue:5570", + "accept": false, + "reason": "A torch._dynamo optimization error and a missing compute_snr utility are unrelated." + }, + { + "left": "issue:5546", + "right": "issue:5628", + "accept": false, + "reason": "Both involve Karras sigmas, but one is wrong step counts in base+refiner while the other is denoising range offset in LMS; different code paths." + }, + { + "left": "issue:1274", + "right": "issue:1896", + "accept": false, + "reason": "A num_images_per_prompt failure and a CLIP-guided Stable Diffusion dict-size error are unrelated." + }, + { + "left": "issue:10231", + "right": "issue:10287", + "accept": false, + "reason": "Both are documentation/example complaints, but the first is a specific Ubuntu import error and the second is a generic docs issue." + }, + { + "left": "issue:2306", + "right": "issue:4747", + "accept": false, + "reason": "An INT_MAX runtime error and a missing model_index.json load error are different problems." + }, + { + "left": "issue:1185", + "right": "issue:1730", + "accept": false, + "reason": "DreamBooth reproducibility and CKPT conversion/webui incompatibility are different topics." + }, + { + "left": "issue:3765", + "right": "issue:3769", + "accept": false, + "reason": "Both mention image artifacts during training/inference, but controlnet stacking and panorama quartering are distinct issues." + }, + { + "left": "issue:3420", + "right": "issue:3915", + "accept": false, + "reason": "ControlNet inpaint pipeline incompatibility and torch.compile text-to-video failure are unrelated." + }, + { + "left": "issue:7490", + "right": "issue:7665", + "accept": false, + "reason": "A generic backward problem and an instruct-pix2pix issue are too vague and distinct to treat as the same bug." + }, + { + "left": "issue:12633", + "right": "issue:13243", + "accept": false, + "reason": "Both are scheduler timestep bugs, but one is DDIM linspace stepping and the other is FlowMatchEuler sigma shifting; different implementations." + }, + { + "left": "issue:4480", + "right": "issue:8994", + "accept": false, + "reason": "An MPS float16 kernel crash and generation steps stuck at 0 are not the same failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 247, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10311", + "issue:1082", + "issue:11581", + "issue:1161", + "issue:11659", + "issue:11700", + "issue:13311", + "issue:13547", + "issue:1574", + "issue:2068", + "issue:2213", + "issue:2772", + "issue:2784", + "issue:2891", + "issue:2985", + "issue:3040", + "issue:3381", + "issue:3584", + "issue:3591", + "issue:3954", + "issue:4353", + "issue:4354", + "issue:4471", + "issue:4753", + "issue:4880", + "issue:4913", + "issue:5128", + "issue:5412", + "issue:5480", + "issue:5567", + "issue:5619", + "issue:5714", + "issue:5905", + "issue:5925", + "issue:5931", + "issue:6201", + "issue:692", + "issue:7661", + "issue:7922", + "issue:7946", + "issue:7974", + "issue:8495", + "issue:8537", + "issue:8742", + "issue:898", + "issue:9224", + "issue:9331", + "issue:9610" + ], + "soft_pairs": [ + "issue:1574|issue:5412", + "issue:2068|issue:3591", + "issue:2784|issue:5480", + "issue:11581|issue:13547", + "issue:2985|issue:4913", + "issue:4471|issue:5714", + "issue:5905|issue:5925", + "issue:3040|issue:4880", + "issue:11700|issue:9610", + "issue:5619|issue:6201", + "issue:5567|issue:9331", + "issue:10311|issue:8742", + "issue:7661|issue:7974", + "issue:8537|issue:9224", + "issue:2891|issue:3381", + "issue:2772|issue:5128", + "issue:3954|issue:4753", + "issue:4353|issue:5931", + "issue:11659|issue:13311", + "issue:2213|issue:898", + "issue:4354|issue:7946", + "issue:7922|issue:8495", + "issue:1082|issue:3584", + "issue:1161|issue:692" + ], + "prepared_review_unit_hash": "51c84a21d239bc3b2f9ac0797910c78b2714d9e3fa6a77dc3ff6aff6ede848d1", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34041, + "estimated_input_tokens": 8511, + "estimated_eval_tokens": 17278 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is highly heterogeneous: it mixes DreamBooth training failures, LoRA/model-loading bugs, pipeline feature requests, and unrelated scheduler/dtype/docs issues. There is no true duplicate cluster here, so I would not merge any of the soft-edge pairs.", + "confidence": 0.86, + "canonical_issue_reason": "No single issue cleanly represents the whole set; if forced to pick an umbrella, issue 11659 is the broadest still-open LoRA-loading report, but it does not subsume the many unrelated training, TPU, scheduler, and docs bugs.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11659 is the closest global representative because it is generic, open, and centered on LoRA loading, which is the only theme that recurs in a few nearby reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1574", + "right": "issue:5412", + "accept": false, + "reason": "Both are DreamBooth-related, but they describe different failures in different scripts/situations (8GB GPU training instability vs resume_from_checkpoint in SDXL LoRA)." + }, + { + "left": "issue:2068", + "right": "issue:3591", + "accept": false, + "reason": "One is a dtype-conversion bug in from_pretrained; the other is a ControlNet loss-curve question. Different problems entirely." + }, + { + "left": "issue:2784", + "right": "issue:5480", + "accept": false, + "reason": "These are separate feature requests: DreamBooth text2video support vs implementing EDM." + }, + { + "left": "issue:11581", + "right": "issue:13547", + "accept": false, + "reason": "Both involve FLUX/LoRA loading, but one is a CLIPVisionModel dtype constructor error and the other is a Flux2 adapter key mismatch. Not the same code-path bug." + }, + { + "left": "issue:2985", + "right": "issue:4913", + "accept": false, + "reason": "Kandinsky 2.1 release discussion vs a specific ControlNet OpenPose SDXL model request; unrelated." + }, + { + "left": "issue:4471", + "right": "issue:5714", + "accept": false, + "reason": "SDXL img2img LoRA weight loading and maybe_raise_or_warn(PeftModel) are different defects in different parts of the stack." + }, + { + "left": "issue:5905", + "right": "issue:5925", + "accept": false, + "reason": "Both are model/inference feature requests, but for different features (diffusion-GAN vs ZipLoRA)." + }, + { + "left": "issue:3040", + "right": "issue:4880", + "accept": false, + "reason": "One is a docs/example AttributeError about a list object; the other is ControlNet single-image input validation. Different issues." + }, + { + "left": "issue:11700", + "right": "issue:9610", + "accept": false, + "reason": "Pruna integration for loading models and adding load_lora_weights to SD3 img2img are separate enhancements." + }, + { + "left": "issue:5619", + "right": "issue:6201", + "accept": false, + "reason": "AutoencodeTiny encoding failure and prompt_embeds handling in StableDiffusion are different pipeline bugs." + }, + { + "left": "issue:5567", + "right": "issue:9331", + "accept": false, + "reason": "accelerator.prepare overwriting unet.config is unrelated to FlowMatchEulerDiscreteScheduler index handling." + }, + { + "left": "issue:10311", + "right": "issue:8742", + "accept": false, + "reason": "Import error in train_dreambooth.py and removal of LoRAAttnProcessor2_0 are not the same bug or change." + }, + { + "left": "issue:7661", + "right": "issue:7974", + "accept": false, + "reason": "Both mention IP-Adapter, but one is about creating an AnimateDiff pipeline with IP-Adapter-loaded UNet and the other is an inpaint bug. Too different to merge." + }, + { + "left": "issue:8537", + "right": "issue:9224", + "accept": false, + "reason": "Both are dtype/platform compatibility issues, but one is missing float8 on torch and the other is float64 unsupported on MPS. Distinct failures." + }, + { + "left": "issue:2891", + "right": "issue:3381", + "accept": false, + "reason": "StableDiffusionUpscalePipeline embeds support/docs mismatch and inpaint discoloration are unrelated." + }, + { + "left": "issue:2772", + "right": "issue:5128", + "accept": false, + "reason": "ControlNet local dataset validation and XLPipeline.from_single_file safetensors loading are different pipeline issues." + }, + { + "left": "issue:3954", + "right": "issue:4753", + "accept": false, + "reason": "Learning-rate scheduler step counting and train_dreambooth_lora.py failures are separate training bugs." + }, + { + "left": "issue:4353", + "right": "issue:5931", + "accept": false, + "reason": "Poor SDXL image quality and non-deterministic StableDiffusionPipeline outputs are different symptom classes." + }, + { + "left": "issue:11659", + "right": "issue:13311", + "accept": false, + "reason": "These are related by LoRA loading, but one is a generic load issue and the other is a specific Flux2 Transformer key-prefix mismatch; not the same concrete bug." + }, + { + "left": "issue:2213", + "right": "issue:898", + "accept": false, + "reason": "TPU hang in a training script vs unsupported BFloat16 on TPU are distinct TPU problems." + }, + { + "left": "issue:4354", + "right": "issue:7946", + "accept": false, + "reason": "Wrong pipeline type from_single_file and custom_pipeline caching are different loading/caching bugs." + }, + { + "left": "issue:7922", + "right": "issue:8495", + "accept": false, + "reason": "Missing ORTStableDiffusionXLPipeline attribute and SD3 tokenizer enum parse failure are unrelated." + }, + { + "left": "issue:1082", + "right": "issue:3584", + "accept": false, + "reason": "DreamBooth CUDA execution failure and a failed DreamBooth tutorial are not the same underlying bug." + }, + { + "left": "issue:1161", + "right": "issue:692", + "accept": false, + "reason": "Converting Flax to PyTorch and asking for a fine-tuning example are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 248, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10204", + "issue:10800", + "issue:11489", + "issue:11569", + "issue:11651", + "issue:11885", + "issue:1207", + "issue:12175", + "issue:12501", + "issue:13067", + "issue:1367", + "issue:2614", + "issue:2669", + "issue:2799", + "issue:2905", + "issue:3019", + "issue:3045", + "issue:3113", + "issue:3174", + "issue:3233", + "issue:3699", + "issue:3728", + "issue:3960", + "issue:4314", + "issue:4578", + "issue:5009", + "issue:5026", + "issue:5205", + "issue:5334", + "issue:5578", + "issue:5710", + "issue:5763", + "issue:590", + "issue:6112", + "issue:6232", + "issue:6616", + "issue:7185", + "issue:725", + "issue:7370", + "issue:7389", + "issue:7600", + "issue:7841", + "issue:8759", + "issue:917", + "issue:938", + "issue:9858", + "issue:9953" + ], + "soft_pairs": [ + "issue:2614|issue:4314", + "issue:3019|issue:5205", + "issue:3113|issue:6616", + "issue:3699|issue:5763", + "issue:3728|issue:7370", + "issue:4578|issue:6112", + "issue:5205|issue:938", + "issue:5334|issue:7600", + "issue:2799|issue:917", + "issue:11651|issue:11885", + "issue:3174|issue:3960", + "issue:5578|issue:7841", + "issue:5710|issue:6232", + "issue:11569|issue:13067", + "issue:1367|issue:590", + "issue:5009|issue:7389", + "issue:5026|issue:7185", + "issue:1207|issue:2669", + "issue:10204|issue:12175", + "issue:11489|issue:9858", + "issue:2905|issue:725", + "issue:10800|issue:8759", + "issue:3045|issue:3233", + "issue:12501|issue:9953" + ], + "prepared_review_unit_hash": "e34fccbbd156bbd56ada0e37d26a3a51fefaf80a28fb288edcd080b0efce31d8", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33046, + "estimated_input_tokens": 8262, + "estimated_eval_tokens": 16780 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is largely heterogeneous; most pairs are only loosely related by subsystem or model family, not the same underlying bug/change. No soft edge is strong enough to merge on the evidence shown.", + "confidence": 0.68, + "canonical_issue_reason": "issue:11885 is the best representative because it is the most concrete, API-specific bug report (`load_lora_weights` on FLUX.1) and the closest to any other overlapping report in the set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11885 has the clearest actionable failure description and the most likely duplicate-target value among the listed issues; the rest are broader feature requests or unrelated bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2614", + "right": "issue:4314", + "accept": false, + "reason": "Both involve downloads, but one is about changing the download directory and the other about interrupted-download write protection; different bugs." + }, + { + "left": "issue:3019", + "right": "issue:5205", + "accept": false, + "reason": "Both are image-editing-related feature requests, but they target different capabilities and code paths." + }, + { + "left": "issue:3113", + "right": "issue:6616", + "accept": false, + "reason": "Both mention guidance, but one is layout control and the other classifier guidance implementation; not the same change." + }, + { + "left": "issue:3699", + "right": "issue:5763", + "accept": false, + "reason": "Completely different problems: CUDA on macOS versus safetensors serialization." + }, + { + "left": "issue:3728", + "right": "issue:7370", + "accept": false, + "reason": "Both are new-model/pipeline support requests, but for different architectures and tasks." + }, + { + "left": "issue:4578", + "right": "issue:6112", + "accept": false, + "reason": "Doc wording about VAE usage is unrelated to an SDXL-turbo image quality complaint." + }, + { + "left": "issue:5205", + "right": "issue:938", + "accept": false, + "reason": "Different application areas: Stable Diffusion repaint/control versus text-to-motion generation." + }, + { + "left": "issue:5334", + "right": "issue:7600", + "accept": false, + "reason": "Related to training utilities, but one is an Accelerate API enhancement and the other a DreamBooth script misuse bug." + }, + { + "left": "issue:2799", + "right": "issue:917", + "accept": false, + "reason": "Mismatch-checkpoint validation and ONNX inpainting are unrelated." + }, + { + "left": "issue:11651", + "right": "issue:11885", + "accept": false, + "reason": "Both concern FLUX LoRA loading, but one is FluxControl-specific and the other is a broader `load_lora_weights` bug; not enough evidence they are the same concrete failure." + }, + { + "left": "issue:3174", + "right": "issue:3960", + "accept": false, + "reason": "One is a LoRA/DreamBooth loading regression and the other is a header-size error; different symptoms and likely different causes." + }, + { + "left": "issue:5578", + "right": "issue:7841", + "accept": false, + "reason": "A download interruption write-protection issue is unrelated to the SDXL viewing artifact." + }, + { + "left": "issue:5710", + "right": "issue:6232", + "accept": false, + "reason": "Same scheduler class, but one is missing `sigmas` at initialization and the other is an `add_noise` malfunction under some settings; too weak to treat as the same bug." + }, + { + "left": "issue:11569", + "right": "issue:13067", + "accept": false, + "reason": "A feature request for a specific repo/model is not the same as a general request to add support for Anima." + }, + { + "left": "issue:1367", + "right": "issue:590", + "accept": false, + "reason": "Scheduler config saving with custom betas and a Karras formula typo are distinct issues." + }, + { + "left": "issue:5009", + "right": "issue:7389", + "accept": false, + "reason": "Wuerstchen runtime error and SDXL single-file loading error are separate model/pipeline failures." + }, + { + "left": "issue:5026", + "right": "issue:7185", + "accept": false, + "reason": "A KeyError on `text_encoder` is unrelated to a question about PyTorch 2.2 support." + }, + { + "left": "issue:1207", + "right": "issue:2669", + "accept": false, + "reason": "CUDA runtime symbol resolution and a `NoneType.replace` AttributeError are unrelated failures." + }, + { + "left": "issue:10204", + "right": "issue:12175", + "accept": false, + "reason": "Both mention poor output quality, but for different models/pipelines and without evidence of the same root cause." + }, + { + "left": "issue:11489", + "right": "issue:9858", + "accept": false, + "reason": "Both are training-related, but one is a Flux LoRA training error and the other is a generic `KeyError: 'train'`; not the same bug." + }, + { + "left": "issue:2905", + "right": "issue:725", + "accept": false, + "reason": "Scheduler Karras-sigma support and DiffusionPipeline device-map/accelerate memory reduction are unrelated enhancements." + }, + { + "left": "issue:10800", + "right": "issue:8759", + "accept": false, + "reason": "Quantization/offload dtype mismatch and SD3 image-processing cast warnings are different code paths." + }, + { + "left": "issue:3045", + "right": "issue:3233", + "accept": false, + "reason": "These are inverse conversion requests (LoRA to diffusers vs LoRA to safetensors), not the same task or bug." + }, + { + "left": "issue:12501", + "right": "issue:9953", + "accept": false, + "reason": "A VAE graph-break issue and a quantized-component CUDA move error are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 249, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1033", + "issue:10869", + "issue:11307", + "issue:11436", + "issue:12334", + "issue:12752", + "issue:1284", + "issue:13430", + "issue:1802", + "issue:1925", + "issue:2392", + "issue:2495", + "issue:2763", + "issue:2908", + "issue:3218", + "issue:3451", + "issue:3461", + "issue:3583", + "issue:3974", + "issue:4069", + "issue:4086", + "issue:4213", + "issue:4263", + "issue:4264", + "issue:4305", + "issue:4349", + "issue:487", + "issue:5108", + "issue:5371", + "issue:6242", + "issue:6392", + "issue:6393", + "issue:6790", + "issue:7199", + "issue:7333", + "issue:7509", + "issue:801", + "issue:8274", + "issue:8410", + "issue:8485", + "issue:8535", + "issue:8539", + "issue:8708", + "issue:9101", + "issue:967", + "issue:9792", + "issue:9895", + "issue:9946" + ], + "soft_pairs": [ + "issue:12334|issue:9946", + "issue:2763|issue:3218", + "issue:487|issue:801", + "issue:11307|issue:11436", + "issue:2495|issue:4349", + "issue:3583|issue:5371", + "issue:4069|issue:4263", + "issue:4213|issue:6393", + "issue:4264|issue:8708", + "issue:6242|issue:9792", + "issue:6392|issue:6790", + "issue:7333|issue:7509", + "issue:8485|issue:9101", + "issue:3451|issue:3461", + "issue:1802|issue:967", + "issue:1925|issue:2908", + "issue:1033|issue:2392", + "issue:7199|issue:8274", + "issue:1284|issue:4305", + "issue:10869|issue:9895", + "issue:8535|issue:8539", + "issue:12752|issue:13430", + "issue:3974|issue:8410", + "issue:4086|issue:5108" + ], + "prepared_review_unit_hash": "e093339d571fded8bf65ae361087f11dc971f2f28d72a44fc2ba82e5c301c02f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33946, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issue set; the only plausible duplicate is the xformers memory-efficient-attention report pair. All other soft edges look like different bugs or feature requests in nearby subsystems.", + "confidence": 0.91, + "canonical_issue_reason": "issue:9946 is the cleanest, most direct title for the xformers attention problem and reads like the broader canonical report versus the more symptom-focused issue:12334.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9946 is the best representative issue because it names the exact API/problem area and is the broader, more canonical phrasing for the only near-duplicate pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12334", + "right": "issue:9946", + "accept": true, + "reason": "Same xformers attention-enable API and same general bug area; these look like duplicate reports of the same problem." + }, + { + "left": "issue:2763", + "right": "issue:3218", + "accept": false, + "reason": "Both touch ControlNet, but one is VRAM growth with cudnn.benchmark and the other is torch.compile graph breaks; different failure modes." + }, + { + "left": "issue:487", + "right": "issue:801", + "accept": false, + "reason": "Both involve image pipelines, but M1 CPU failure and attention slicing not being honored are different bugs." + }, + { + "left": "issue:11307", + "right": "issue:11436", + "accept": false, + "reason": "HiDream group offloading at block level is not the same as the Half/BFloat16 dtype mismatch during inference." + }, + { + "left": "issue:2495", + "right": "issue:4349", + "accept": false, + "reason": "Both are upscale-pipeline issues, but black output on MPS and missing text_encoder despite prompt_embeds are different code-path problems." + }, + { + "left": "issue:3583", + "right": "issue:5371", + "accept": false, + "reason": "Unrelated feature requests: mixture-of-diffusers support vs BLIP inpainting pipe support." + }, + { + "left": "issue:4069", + "right": "issue:4263", + "accept": false, + "reason": "Both are unloading-related, but textual inversion unloading and LoRA unloading are different mechanisms and bugs." + }, + { + "left": "issue:4213", + "right": "issue:6393", + "accept": false, + "reason": "Both mention from_single_file, but one is missing accelerate and the other is private HF link download failure; different root causes." + }, + { + "left": "issue:4264", + "right": "issue:8708", + "accept": false, + "reason": "Missing example images and missing vae.config.shift_factor are unrelated documentation/training issues." + }, + { + "left": "issue:6242", + "right": "issue:9792", + "accept": false, + "reason": "Dataset log request vs cache-loading failure when mapping a dataset are not the same issue." + }, + { + "left": "issue:6392", + "right": "issue:6790", + "accept": false, + "reason": "Custom LoRA loading failure and unexpected state_dict key in LCM training are different problems." + }, + { + "left": "issue:7333", + "right": "issue:7509", + "accept": false, + "reason": "Both are video/inversion papers, but they target different algorithms and pipelines." + }, + { + "left": "issue:8485", + "right": "issue:9101", + "accept": false, + "reason": "Support for SD3 LoRA and an InstructPix2Pix training script for SD3 are separate feature requests." + }, + { + "left": "issue:3451", + "right": "issue:3461", + "accept": false, + "reason": "Both concern memory/performance, but text-to-image fp16 memory usage and Dreambooth throughput/VRAM are different reports." + }, + { + "left": "issue:1802", + "right": "issue:967", + "accept": false, + "reason": "UNet encoder_hidden_states handling and ONNX scheduler incompatibility are unrelated issues." + }, + { + "left": "issue:1925", + "right": "issue:2908", + "accept": false, + "reason": "Stable-diffusion inpainting fp16 problems and ControlNet training shape/mixed-precision errors are different bugs." + }, + { + "left": "issue:1033", + "right": "issue:2392", + "accept": false, + "reason": "IndexError in stable_diffusion_mega and img2img prompt_embeds validation failure are unrelated." + }, + { + "left": "issue:7199", + "right": "issue:8274", + "accept": false, + "reason": "Different research pipelines: ViewFusion multiview consistency vs FIFO-Diffusion infinite video generation." + }, + { + "left": "issue:1284", + "right": "issue:4305", + "accept": false, + "reason": "Batch-size slowdown and torch.compile graph breaks are unrelated performance vs compiler issues." + }, + { + "left": "issue:10869", + "right": "issue:9895", + "accept": false, + "reason": "Meta-tensor .item() failure and Flux device-placement mismatch are different runtime errors." + }, + { + "left": "issue:8535", + "right": "issue:8539", + "accept": false, + "reason": "Both are SD3-related, but xformers attention enablement and cross-attention configuration are not the same bug." + }, + { + "left": "issue:12752", + "right": "issue:13430", + "accept": false, + "reason": "Z-Image img2img/inpainting pipeline request and JoyAI-Image-Edit support are different model support requests." + }, + { + "left": "issue:3974", + "right": "issue:8410", + "accept": false, + "reason": "ControlNet inpaint .safetensors support and TextDiffuser native pipeline support are unrelated feature requests." + }, + { + "left": "issue:4086", + "right": "issue:5108", + "accept": false, + "reason": "SDXL ControlNet training failure and T2I Multiadapter XL exceptions are different pipeline/training problems." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative: the xformers pair (12334/9946) is the only clearly plausible duplicate, and the other pairs are reasonably treated as different bugs or feature requests. No overstatement detected." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 250, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10658", + "issue:10767", + "issue:10967", + "issue:11307", + "issue:11437", + "issue:11687", + "issue:12501", + "issue:12610", + "issue:1407", + "issue:1704", + "issue:2370", + "issue:2616", + "issue:2644", + "issue:2646", + "issue:2661", + "issue:2662", + "issue:2780", + "issue:2783", + "issue:2843", + "issue:3279", + "issue:3379", + "issue:3406", + "issue:3418", + "issue:3454", + "issue:3698", + "issue:4480", + "issue:460", + "issue:5043", + "issue:5062", + "issue:5368", + "issue:5395", + "issue:5557", + "issue:5997", + "issue:6250", + "issue:6321", + "issue:6858", + "issue:6879", + "issue:6958", + "issue:7197", + "issue:7438", + "issue:7574", + "issue:8744", + "issue:8837", + "issue:905", + "issue:9138", + "issue:9227", + "issue:9558" + ], + "soft_pairs": [ + "issue:7197|issue:7574", + "issue:4480|issue:7438", + "issue:5368|issue:5557", + "issue:5997|issue:8744", + "issue:12501|issue:12610", + "issue:2370|issue:2780", + "issue:2783|issue:3698", + "issue:2646|issue:6250", + "issue:10767|issue:10967", + "issue:2616|issue:2843", + "issue:3379|issue:6321", + "issue:5368|issue:6879", + "issue:11307|issue:9227", + "issue:2644|issue:3454", + "issue:5395|issue:8837", + "issue:10658|issue:11687", + "issue:11437|issue:9558", + "issue:1407|issue:460", + "issue:1704|issue:5062", + "issue:2661|issue:3279", + "issue:2662|issue:3279", + "issue:3406|issue:6958", + "issue:3418|issue:5043", + "issue:3698|issue:905", + "issue:6858|issue:9138" + ], + "prepared_review_unit_hash": "5bc90661fdb1ad888b9839d3cd22dd37a7c03bce93242d3a0ff81538cce3b391", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33344, + "estimated_input_tokens": 8336, + "estimated_eval_tokens": 16928 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the items span many unrelated bugs and feature requests across training, checkpointing, attention/dtype errors, ONNX, and docs. The soft pairs are mostly superficial keyword matches rather than the same underlying issue.", + "confidence": 0.97, + "canonical_issue_reason": "No canonical issue fits the cluster well; the issues are too heterogeneous, so picking one representative would be misleading.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good global representative because the cluster lacks a common underlying bug or feature change.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7197", + "right": "issue:7574", + "accept": false, + "reason": "Both mention checkpoint-like errors, but the failing keys and scenarios differ; not the same bug." + }, + { + "left": "issue:4480", + "right": "issue:7438", + "accept": false, + "reason": "MPS kernel crash on Mac and XPU generation stall are different platform/runtime problems." + }, + { + "left": "issue:5368", + "right": "issue:5557", + "accept": false, + "reason": "One is a mixed-precision dtype mismatch; the other is an unexpected keyword argument error." + }, + { + "left": "issue:5997", + "right": "issue:8744", + "accept": false, + "reason": "ONNX tutorial model-type mismatch and single-file training support are unrelated." + }, + { + "left": "issue:12501", + "right": "issue:12610", + "accept": false, + "reason": "Graph break in VAE decoder vs device mismatch from quanto/group offload are different code paths." + }, + { + "left": "issue:2370", + "right": "issue:2780", + "accept": false, + "reason": "A GPU memory/capacity question and an FP16 unscale error are not the same issue." + }, + { + "left": "issue:2783", + "right": "issue:3698", + "accept": false, + "reason": "Both touch inpainting, but one requests a pipeline feature and the other model conversion; not one bug/change." + }, + { + "left": "issue:2646", + "right": "issue:6250", + "accept": false, + "reason": "Tiled VAE indexing failure and SDXL VAE single-file KeyError are distinct loading/runtime bugs." + }, + { + "left": "issue:10767", + "right": "issue:10967", + "accept": false, + "reason": "Both are model-integration requests, but for different models and different tasks." + }, + { + "left": "issue:2616", + "right": "issue:2843", + "accept": false, + "reason": "NoneType in save_attn_procs and multi-GPU checkpoint saving errors are different failure modes." + }, + { + "left": "issue:3379", + "right": "issue:6321", + "accept": false, + "reason": "Checkpoint loading failure and missing CLIP detection are unrelated." + }, + { + "left": "issue:5368", + "right": "issue:6879", + "accept": false, + "reason": "Both are dtype-related, but they fail in different subsystems with different triggers." + }, + { + "left": "issue:11307", + "right": "issue:9227", + "accept": false, + "reason": "Group offloading device mismatch and FLUX dreambooth training issues are different problems." + }, + { + "left": "issue:2644", + "right": "issue:3454", + "accept": false, + "reason": "A general training capability question is not the same as a specific script regression." + }, + { + "left": "issue:5395", + "right": "issue:8837", + "accept": false, + "reason": "LoRA combination bug and sliced attention processor bug do not share the same code path." + }, + { + "left": "issue:10658", + "right": "issue:11687", + "accept": false, + "reason": "ONNX provider options and a docs request for pruna are unrelated." + }, + { + "left": "issue:11437", + "right": "issue:9558", + "accept": false, + "reason": "Missing validation outputs and black tensorboard validation images are different output/visualization issues." + }, + { + "left": "issue:1407", + "right": "issue:460", + "accept": false, + "reason": "Feature_extractor regression in an ONNX pipeline and CUDA-vs-CPU onnxruntime selection are different issues." + }, + { + "left": "issue:1704", + "right": "issue:5062", + "accept": false, + "reason": "A general question about attention blocks and a processor typo fix are not duplicates." + }, + { + "left": "issue:2661", + "right": "issue:3279", + "accept": false, + "reason": "Checkpoint write errors and resume-from-checkpoint errors are related only loosely; not the same bug." + }, + { + "left": "issue:2662", + "right": "issue:3279", + "accept": false, + "reason": "Same as above: saving write error versus resume failure are distinct." + }, + { + "left": "issue:3406", + "right": "issue:6958", + "accept": false, + "reason": "only_cross_attention behavior and use_linear_projection in a motion model are separate attention issues." + }, + { + "left": "issue:3418", + "right": "issue:5043", + "accept": false, + "reason": "Adding text-encoder LoRA support and adding W\u00fcrstchen LoRA fine-tuning are different feature requests." + }, + { + "left": "issue:3698", + "right": "issue:905", + "accept": false, + "reason": "Inpainting model conversion and a community image2image/inpainting pipeline request are related only broadly, not the same concrete change." + }, + { + "left": "issue:6858", + "right": "issue:9138", + "accept": false, + "reason": "A .convert attribute error in controlnet training and an int-vs-str comparison type error are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 251, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10003", + "issue:10194", + "issue:10470", + "issue:10520", + "issue:1064", + "issue:10869", + "issue:11104", + "issue:11144", + "issue:12157", + "issue:12645", + "issue:12656", + "issue:1543", + "issue:2052", + "issue:2068", + "issue:2310", + "issue:2661", + "issue:2662", + "issue:2772", + "issue:2786", + "issue:3112", + "issue:3348", + "issue:3350", + "issue:3951", + "issue:3984", + "issue:4229", + "issue:4403", + "issue:4727", + "issue:4841", + "issue:4975", + "issue:5000", + "issue:5062", + "issue:5395", + "issue:5884", + "issue:7136", + "issue:756", + "issue:7695", + "issue:775", + "issue:7773", + "issue:8469", + "issue:8594", + "issue:8644", + "issue:8850", + "issue:8880", + "issue:9109", + "issue:929", + "issue:9291", + "issue:9361", + "issue:9599" + ], + "soft_pairs": [ + "issue:12157|issue:12645", + "issue:9291|issue:9361", + "issue:10470|issue:8880", + "issue:4229|issue:8594", + "issue:11144|issue:9109", + "issue:3350|issue:4727", + "issue:10520|issue:10869", + "issue:2310|issue:3112", + "issue:8469|issue:8850", + "issue:1543|issue:775", + "issue:1064|issue:3951", + "issue:2786|issue:3348", + "issue:2068|issue:4975", + "issue:2052|issue:2661", + "issue:2052|issue:2662", + "issue:10194|issue:9599", + "issue:3984|issue:756", + "issue:4403|issue:7695", + "issue:4841|issue:7773", + "issue:5062|issue:929", + "issue:5884|issue:7136", + "issue:11104|issue:8644", + "issue:2772|issue:2786", + "issue:10003|issue:12656", + "issue:5000|issue:5395" + ], + "prepared_review_unit_hash": "066217b66d8afeb9b42573212255184c88d10b6538bf126e3cf7fd5c90d0d23b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33956, + "estimated_input_tokens": 8489, + "estimated_eval_tokens": 17234 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are not duplicates: they cover many unrelated models, pipelines, and failure modes. None of the soft pairs look safe to merge into one bug/change, so all are rejected.", + "confidence": 0.95, + "canonical_issue_reason": "issue:10470 is the strongest representative if one must be chosen: it is a concrete, self-contained bug report with clear repro context and follow-up discussion; the rest are too heterogeneous to serve as a shared canonical duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10470 is the best standalone issue in this set because it is specific, reproducible, and well-scoped, unlike the other reports which mostly describe different subsystems or feature requests.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12157", + "right": "issue:12645", + "accept": false, + "reason": "Quantization-config training error vs inference-speed complaint; different failure modes and code paths." + }, + { + "left": "issue:9291", + "right": "issue:9361", + "accept": false, + "reason": "Both mention FLUX LoRA, but one is key compatibility and the other is missing LoRA effect at inference; not the same bug." + }, + { + "left": "issue:10470", + "right": "issue:8880", + "accept": false, + "reason": "torchao inference breakage vs a later-version speed regression; correctness bug and performance regression are distinct." + }, + { + "left": "issue:4229", + "right": "issue:8594", + "accept": false, + "reason": "Different scripts and errors: multi-GPU accelerate failure vs UnboundLocalError in train_dreambooth_lora_sd3.py." + }, + { + "left": "issue:11144", + "right": "issue:9109", + "accept": false, + "reason": "Flax UNet dtype initialization issue is unrelated to missing torchvision::nms operator." + }, + { + "left": "issue:3350", + "right": "issue:4727", + "accept": false, + "reason": "Animov model errors vs local_files_only auto-pipeline loading; no shared underlying defect." + }, + { + "left": "issue:10520", + "right": "issue:10869", + "accept": false, + "reason": "Both are device-related, but one is cuda/cpu mismatch in Sana and the other is meta-tensor .item() failure." + }, + { + "left": "issue:2310", + "right": "issue:3112", + "accept": false, + "reason": "Inpainting cleaner issue vs multi-ControlNet inpainting feature request; not the same problem." + }, + { + "left": "issue:8469", + "right": "issue:8850", + "accept": false, + "reason": "Broken faces in IP-Adapter output vs community pipelines mirror broken are unrelated issues." + }, + { + "left": "issue:1543", + "right": "issue:775", + "accept": false, + "reason": "fp16/revision loading problem vs general fp16 without autocast issue; overlapping topic, but different concrete bugs." + }, + { + "left": "issue:1064", + "right": "issue:3951", + "accept": false, + "reason": "Scheduler add_noise error vs NaNs in demo code; different symptoms and mechanisms." + }, + { + "left": "issue:2786", + "right": "issue:3348", + "accept": false, + "reason": "numpy/transformers incompatibility in Depth2Img vs LoRA scale support with xformers; unrelated." + }, + { + "left": "issue:2068", + "right": "issue:4975", + "accept": false, + "reason": "One is incorrect dtype conversion behavior, the other is repeated datatype-query inefficiency; related area, different bug." + }, + { + "left": "issue:2052", + "right": "issue:2661", + "accept": false, + "reason": "Checkpoint resume failure vs checkpoint write error; distinct checkpointing problems." + }, + { + "left": "issue:2052", + "right": "issue:2662", + "accept": false, + "reason": "Same as above: resume-from-checkpoint and save/write failure are different issues." + }, + { + "left": "issue:10194", + "right": "issue:9599", + "accept": false, + "reason": "Flux attention masking support vs LoRA-only finetune example request; different feature requests." + }, + { + "left": "issue:3984", + "right": "issue:756", + "accept": false, + "reason": "class_labels requirement vs deprecation-tuple error; unrelated API failures." + }, + { + "left": "issue:4403", + "right": "issue:7695", + "accept": false, + "reason": "Half-precision LayerNorm kernel missing vs ignored ip_adapter_masks; different code paths." + }, + { + "left": "issue:4841", + "right": "issue:7773", + "accept": false, + "reason": "ControlNet image-size bug vs Stable Video Diffusion batch-size generation bug; not the same underlying defect." + }, + { + "left": "issue:5062", + "right": "issue:929", + "accept": false, + "reason": "Typo in attention processor vs redundant sqrt calculation; both in attention code, but not the same bug." + }, + { + "left": "issue:5884", + "right": "issue:7136", + "accept": false, + "reason": "IP-Adapter support expansion vs callback_on_step_end for LDM3D; separate feature requests." + }, + { + "left": "issue:11104", + "right": "issue:8644", + "accept": false, + "reason": "Both involve offload/meta-tensor problems, but they affect different models and loading flows." + }, + { + "left": "issue:2772", + "right": "issue:2786", + "accept": false, + "reason": "Local dataset validation in ControlNet training vs numpy/transformers incompatibility in Depth2Img; unrelated." + }, + { + "left": "issue:10003", + "right": "issue:12656", + "accept": false, + "reason": "DDIMScheduler 1000-step bug vs AuraFlow positional embedding OOB; different models and failures." + }, + { + "left": "issue:5000", + "right": "issue:5395", + "accept": false, + "reason": "Broken unfuse-after-fuse LoRA regression vs combining Civitai LoRAs with PEFT; same broad subsystem, but different concrete bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 252, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10025", + "issue:10412", + "issue:10675", + "issue:1076", + "issue:11135", + "issue:11148", + "issue:1398", + "issue:1524", + "issue:1642", + "issue:1734", + "issue:1866", + "issue:2392", + "issue:2984", + "issue:3089", + "issue:319", + "issue:3284", + "issue:337", + "issue:3418", + "issue:3748", + "issue:3788", + "issue:4086", + "issue:4213", + "issue:4337", + "issue:4438", + "issue:444", + "issue:4557", + "issue:4665", + "issue:4753", + "issue:5054", + "issue:5325", + "issue:5510", + "issue:5798", + "issue:5824", + "issue:6079", + "issue:6452", + "issue:6486", + "issue:660", + "issue:6783", + "issue:6967", + "issue:7370", + "issue:760", + "issue:7735", + "issue:8414", + "issue:8534", + "issue:9106", + "issue:9166", + "issue:9460" + ], + "soft_pairs": [ + "issue:5824|issue:7735", + "issue:2392|issue:5325", + "issue:11135|issue:8414", + "issue:1642|issue:3089", + "issue:1866|issue:337", + "issue:4086|issue:444", + "issue:4213|issue:6783", + "issue:1398|issue:760", + "issue:1524|issue:4753", + "issue:4337|issue:4665", + "issue:4557|issue:9106", + "issue:5510|issue:6079", + "issue:6486|issue:8534", + "issue:10412|issue:6967", + "issue:5798|issue:9166", + "issue:1734|issue:3284", + "issue:10025|issue:10675", + "issue:1076|issue:319", + "issue:11148|issue:9460", + "issue:2984|issue:7370", + "issue:3418|issue:6452", + "issue:3748|issue:660", + "issue:3788|issue:7370", + "issue:4438|issue:5054" + ], + "prepared_review_unit_hash": "b6d37ff1854f1de11ce7b2f5dd6934a5089c0f7c19647c8b7a864735eb4b9417", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33538, + "estimated_input_tokens": 8385, + "estimated_eval_tokens": 17026 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose mix of unrelated Diffusers issues that only share broad model/pipeline vocabulary; the soft pairs are superficial and should not be merged as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "issue:5824 is the best representative because it is a concrete, well-scoped SDXL pipeline bug with a clear code-path target; the rest of the cluster is too heterogeneous to serve as a better canonical issue.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:5824 is the strongest issue to anchor triage: specific, technically detailed, and centered on a reproducible pipeline behavior problem rather than a vague or generic report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5824", + "right": "issue:7735", + "accept": false, + "reason": "SDXL `get_add_time_ids()` default-value bug vs. a custom dataset caption-column error; different failure modes and code paths." + }, + { + "left": "issue:2392", + "right": "issue:5325", + "accept": false, + "reason": "`prompt_embeds` validation bug in img2img vs. packaging type distribution; unrelated symptoms and fixes." + }, + { + "left": "issue:11135", + "right": "issue:8414", + "accept": false, + "reason": "Two separate proposal-style items for different acceleration/model concepts, not the same bug or change." + }, + { + "left": "issue:1642", + "right": "issue:3089", + "accept": false, + "reason": "Different scheduler implementations with distinct missing/incorrect behavior; not one shared defect." + }, + { + "left": "issue:1866", + "right": "issue:337", + "accept": false, + "reason": "img2img strength handling in DPMSolverSinglestepScheduler vs. inpainting guidance behavior; only loosely related by pipeline type." + }, + { + "left": "issue:4086", + "right": "issue:444", + "accept": false, + "reason": "ControlNet training issue vs. timestep IndexError; different parts of the stack and different bugs." + }, + { + "left": "issue:4213", + "right": "issue:6783", + "accept": false, + "reason": "from_single_file() / accelerate dependency issue vs. PixArt generation corruption; not the same underlying problem." + }, + { + "left": "issue:1398", + "right": "issue:760", + "accept": false, + "reason": "Both mention MPS, but one is SD2 support on Apple Silicon and the other is unexpected SD1.5 output on an older diffusers version." + }, + { + "left": "issue:1524", + "right": "issue:4753", + "accept": false, + "reason": "Different DreamBooth/textual inversion training problems with different scripts and failure modes." + }, + { + "left": "issue:4337", + "right": "issue:4665", + "accept": false, + "reason": "torch.compile failure with a specific community ControlNet vs. Windows safetensors file selection; unrelated issues." + }, + { + "left": "issue:4557", + "right": "issue:9106", + "accept": false, + "reason": "Different AttributeError cases in different pipelines/models; no shared concrete code path." + }, + { + "left": "issue:5510", + "right": "issue:6079", + "accept": false, + "reason": "Both are dtype mismatch errors, but they occur in different pipelines and likely different layers, so they are not the same bug." + }, + { + "left": "issue:6486", + "right": "issue:8534", + "accept": false, + "reason": "Dreambooth embedding retraction bug vs. SD3 logit-normal weighting; unrelated training logic." + }, + { + "left": "issue:10412", + "right": "issue:6967", + "accept": false, + "reason": "Training memory blowup in SD3.5 DreamBooth vs. SDXL prior-preservation shape mismatch; different failure classes." + }, + { + "left": "issue:5798", + "right": "issue:9166", + "accept": false, + "reason": "ControlNet callback argument compatibility vs. image_processor overlay behavior; different APIs and defects." + }, + { + "left": "issue:1734", + "right": "issue:3284", + "accept": false, + "reason": "Both concern DreamBooth/accelerate performance or execution, but one is multi-GPU slowness and the other is a script failure; not the same bug." + }, + { + "left": "issue:10025", + "right": "issue:10675", + "accept": false, + "reason": "Flux attention-mask request vs. Flux scheduler config max_shift discrepancy; different components and changes." + }, + { + "left": "issue:1076", + "right": "issue:319", + "accept": false, + "reason": "Stable Diffusion v1.5 discussion vs. VQ-Diffusion model request; no duplicate relationship." + }, + { + "left": "issue:11148", + "right": "issue:9460", + "accept": false, + "reason": "PEFT adapter state bug vs. ControlNet list support; unrelated features in different subsystems." + }, + { + "left": "issue:2984", + "right": "issue:7370", + "accept": false, + "reason": "A finished paper/model entry vs. a request to support another model architecture; not the same change." + }, + { + "left": "issue:3418", + "right": "issue:6452", + "accept": false, + "reason": "Training script enhancement request vs. adding an inpaint pipeline input; different feature requests." + }, + { + "left": "issue:3748", + "right": "issue:660", + "accept": false, + "reason": "LoRA safetensor dtype mismatch vs. MPS fp16 revision usage; related only by dtype/platform, not the same bug." + }, + { + "left": "issue:3788", + "right": "issue:7370", + "accept": false, + "reason": "Two separate new-model/pipeline support requests for different architectures." + }, + { + "left": "issue:4438", + "right": "issue:5054", + "accept": false, + "reason": "Both are SDXL errors, but one is a generic pipeline runtime error and the other is a missing `final_offload_hook` attribute; different code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 253, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10054", + "issue:10303", + "issue:10659", + "issue:10674", + "issue:1106", + "issue:11398", + "issue:11642", + "issue:12299", + "issue:1305", + "issue:1310", + "issue:1405", + "issue:1805", + "issue:1990", + "issue:2068", + "issue:2331", + "issue:2459", + "issue:2532", + "issue:2579", + "issue:3043", + "issue:3113", + "issue:3391", + "issue:3861", + "issue:389", + "issue:4219", + "issue:4542", + "issue:4551", + "issue:462", + "issue:4868", + "issue:4933", + "issue:5046", + "issue:5570", + "issue:5743", + "issue:5897", + "issue:6008", + "issue:6584", + "issue:6809", + "issue:7276", + "issue:7460", + "issue:7492", + "issue:7497", + "issue:801", + "issue:9164", + "issue:9358", + "issue:9443", + "issue:954", + "issue:9548", + "issue:9974" + ], + "soft_pairs": [ + "issue:10303|issue:7460", + "issue:1305|issue:5046", + "issue:1805|issue:3043", + "issue:3391|issue:4868", + "issue:5570|issue:6008", + "issue:10054|issue:9974", + "issue:1310|issue:1990", + "issue:2331|issue:3043", + "issue:3113|issue:7497", + "issue:12299|issue:9443", + "issue:4542|issue:6809", + "issue:7276|issue:9548", + "issue:4933|issue:954", + "issue:9164|issue:9358", + "issue:1106|issue:2459", + "issue:2532|issue:2579", + "issue:389|issue:462", + "issue:4551|issue:5743", + "issue:10659|issue:10674", + "issue:11398|issue:11642", + "issue:2068|issue:3861", + "issue:1405|issue:801", + "issue:5897|issue:7492", + "issue:4219|issue:6584" + ], + "prepared_review_unit_hash": "eaeb18a6c5fc137cfc218785e7aa0f2a4ed0d04b88b2981122327984653ee218", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33221, + "estimated_input_tokens": 8306, + "estimated_eval_tokens": 16868 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly a grab bag of unrelated bugs/features. I only see one plausible duplicate pair around `enable_attention_slicing()` regressions; the rest target different models, pipelines, or failure modes.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 1405 is the broadest and earliest report of the attention-slicing regression, so it is the best umbrella representative for the one plausible duplicate pair.", + "canonical_pr_reason": null, + "best_issue_reason": "1405 best captures the shared code-path problem (`enable_attention_slicing()` not being honored) and is the most general reproduction among the issues here.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10303", + "right": "issue:7460", + "accept": false, + "reason": "Different components and failures: attention tensor permutation in `head_to_batch_dim` vs AdaLayerNorm behavior." + }, + { + "left": "issue:1305", + "right": "issue:5046", + "accept": false, + "reason": "Both are image-editing related, but they describe different methods/proposals (`Direct Inversion` vs `DragonDiffusion`) rather than the same bug or change." + }, + { + "left": "issue:1805", + "right": "issue:3043", + "accept": false, + "reason": "Unrelated requests: a scalable diffusion pipeline vs adding ODISE." + }, + { + "left": "issue:3391", + "right": "issue:4868", + "accept": false, + "reason": "Different failure modes and areas: overexposure in DeepFloyd SR vs an `__getattr__` AttributeError." + }, + { + "left": "issue:5570", + "right": "issue:6008", + "accept": false, + "reason": "Unrelated package/API issues: missing `compute_snr` vs IP-Adapter/LCM-LoRA inference failure." + }, + { + "left": "issue:10054", + "right": "issue:9974", + "accept": false, + "reason": "Different scheduler/config problems: wrong `torch.linspace` parameters in EulerDiscreteScheduler vs SD3 controlnet expected-parameter handling." + }, + { + "left": "issue:1310", + "right": "issue:1990", + "accept": false, + "reason": "Both are pipeline ideas, but they are distinct proposals for different workflows and do not describe the same change." + }, + { + "left": "issue:2331", + "right": "issue:3043", + "accept": false, + "reason": "Both mention pipelines, but one is ControlNet support and the other is ODISE; these are separate feature requests." + }, + { + "left": "issue:3113", + "right": "issue:7497", + "accept": false, + "reason": "Different research/feature requests with different goals and code paths." + }, + { + "left": "issue:12299", + "right": "issue:9443", + "accept": false, + "reason": "Distinct SD3 issues: tensor mismatch with negative prompt embeddings vs tokenizer slow/fast instantiation." + }, + { + "left": "issue:4542", + "right": "issue:6809", + "accept": false, + "reason": "Different bugs: SDXL img2img multi-image generation vs LoRA adapter loading under PEFT." + }, + { + "left": "issue:7276", + "right": "issue:9548", + "accept": false, + "reason": "Different Dreambooth/Flux problems with different symptoms; not the same underlying bug." + }, + { + "left": "issue:4933", + "right": "issue:954", + "accept": false, + "reason": "Completely different areas: missing model artifact after training vs login/load failure." + }, + { + "left": "issue:9164", + "right": "issue:9358", + "accept": false, + "reason": "Both involve Flux Dreambooth LoRA, but one is convergence on a sample prompt and the other is redundant text-encoder reinitialization; not clearly the same bug." + }, + { + "left": "issue:1106", + "right": "issue:2459", + "accept": false, + "reason": "Different code paths: PNDM repeatedly calling UNet vs xFormers attention in a training script." + }, + { + "left": "issue:2532", + "right": "issue:2579", + "accept": false, + "reason": "Panorama `get_views` bug and a possible xFormers issue are unrelated." + }, + { + "left": "issue:389", + "right": "issue:462", + "accept": false, + "reason": "README example breakage is not the same as the Apple MPS img2img regression." + }, + { + "left": "issue:4551", + "right": "issue:5743", + "accept": false, + "reason": "Different training failures: DreamBooth crash on SDXL base vs latent consistency distillation stuck at 0%." + }, + { + "left": "issue:10659", + "right": "issue:10674", + "accept": false, + "reason": "Different models and symptoms: poor ConsisID output vs FluxPipeline/GGUF incompatibility." + }, + { + "left": "issue:11398", + "right": "issue:11642", + "accept": false, + "reason": "Both are Flux-related, but one is missing config/set_default_attn_processor support and the other is `prompt_embeds_scale` having no effect; separate issues." + }, + { + "left": "issue:2068", + "right": "issue:3861", + "accept": false, + "reason": "`from_pretrained` dtype conversion behavior is unrelated to the `Pipeline.from_pretrained` naming/question." + }, + { + "left": "issue:1405", + "right": "issue:801", + "accept": true, + "reason": "Both report the same underlying regression: `enable_attention_slicing()` is not being respected in pipeline inference, with one broader report and one that calls out inpainting/img2img specifically." + }, + { + "left": "issue:5897", + "right": "issue:7492", + "accept": false, + "reason": "LoRA device-mismatch training error and a Hugging Face repository-not-found error are unrelated." + }, + { + "left": "issue:4219", + "right": "issue:6584", + "accept": false, + "reason": "Both concern conversion tooling, but they are different checkpoint/model cases and do not clearly indicate the same bug or change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative on the soft-edge pairs. The only accepted duplicate, issue:1405 vs issue:801, is well supported by the titles and the summary does not overclaim beyond that. The other pairwise rejections are plausible and appropriately cautious." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 254, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10112", + "issue:10382", + "issue:10752", + "issue:10855", + "issue:10897", + "issue:1106", + "issue:11220", + "issue:11627", + "issue:1188", + "issue:12071", + "issue:12149", + "issue:12895", + "issue:1298", + "issue:13034", + "issue:1675", + "issue:176", + "issue:1949", + "issue:2153", + "issue:2163", + "issue:2377", + "issue:2695", + "issue:2712", + "issue:2800", + "issue:2856", + "issue:3112", + "issue:338", + "issue:339", + "issue:3565", + "issue:3958", + "issue:4042", + "issue:4052", + "issue:4069", + "issue:4303", + "issue:4542", + "issue:477", + "issue:5374", + "issue:5405", + "issue:5607", + "issue:6113", + "issue:6793", + "issue:7428", + "issue:7562", + "issue:7686", + "issue:8236", + "issue:8801", + "issue:9822" + ], + "soft_pairs": [ + "issue:11627|issue:12895", + "issue:5607|issue:7686", + "issue:10382|issue:6793", + "issue:2695|issue:2712", + "issue:3958|issue:7562", + "issue:10010|issue:9822", + "issue:10112|issue:7428", + "issue:10752|issue:13034", + "issue:10855|issue:12071", + "issue:10897|issue:8801", + "issue:1106|issue:2163", + "issue:11220|issue:12149", + "issue:1188|issue:2800", + "issue:1298|issue:4542", + "issue:1675|issue:5374", + "issue:176|issue:4069", + "issue:1949|issue:4303", + "issue:2153|issue:477", + "issue:2377|issue:338", + "issue:2377|issue:339", + "issue:2856|issue:6113", + "issue:3112|issue:4042", + "issue:3565|issue:8236", + "issue:4052|issue:5405" + ], + "prepared_review_unit_hash": "7b4f5ef4b06c7779bbb31c1cd229db1e4a9c6f9e93e1c8f0116504caf3c3ecd0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33314, + "estimated_input_tokens": 8329, + "estimated_eval_tokens": 16914 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of unrelated diffusers issues spanning model support requests, training bugs, pipeline regressions, docs, and scheduler/config problems. The proposed soft pairs are mostly superficial title similarity and do not describe the same underlying defect or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11627", + "right": "issue:12895", + "accept": false, + "reason": "Both mention StableDiffusion3Pipeline, but one is a specific InstantX IP-Adapter support request and the other is a num_images_per_prompt batched-prompt crash; different bugs/changes." + }, + { + "left": "issue:5607", + "right": "issue:7686", + "accept": false, + "reason": "Both are runtime errors, but the first is a tensor shape mismatch in a pipeline and the second is a UNet config/cross-attention initialization error; unrelated code paths." + }, + { + "left": "issue:10382", + "right": "issue:6793", + "accept": false, + "reason": "One reports VAE forward return-type AttributeError, the other is a Hydra serialization problem; not the same issue." + }, + { + "left": "issue:2695", + "right": "issue:2712", + "accept": false, + "reason": "ControlNet validation/tensorboard behavior and missing norm_num_groups config are different training/configuration problems." + }, + { + "left": "issue:3958", + "right": "issue:7562", + "accept": false, + "reason": "Both involve offload and adapters, but one is LoRA with enable_model_cpu_offload and the other is multicontrolnet offload failure; not the same concrete bug." + }, + { + "left": "issue:10010", + "right": "issue:9822", + "accept": false, + "reason": "FLUX Fill model_index.json 404 vs loading SDXL LoRAs into Flux are unrelated compatibility/model-loading issues." + }, + { + "left": "issue:10112", + "right": "issue:7428", + "accept": false, + "reason": "The first is an unclear diffusers discussion issue, while the second is a request to support layer diffusion; no shared defect." + }, + { + "left": "issue:10752", + "right": "issue:13034", + "accept": false, + "reason": "FP16 gradient unscale failure and WanPipeline height/width handling are different bugs." + }, + { + "left": "issue:10855", + "right": "issue:12071", + "accept": false, + "reason": "Gradient accumulation causing optimizer.step every step is unrelated to Flux Kontext transformer forward behavior changes." + }, + { + "left": "issue:10897", + "right": "issue:8801", + "accept": false, + "reason": "Both are feature requests for different models (SkyReels-A1 vs Kolors); not duplicates." + }, + { + "left": "issue:1106", + "right": "issue:2163", + "accept": false, + "reason": "PNDM repeatedly calling UNet and a dtype mismatch during text-to-image training are separate problems." + }, + { + "left": "issue:11220", + "right": "issue:12149", + "accept": false, + "reason": "Broken docs page and pip/source version mismatch are distinct release/documentation issues." + }, + { + "left": "issue:1188", + "right": "issue:2800", + "accept": false, + "reason": "CLIPSeg x Stable Diffusion pipeline request and DiffEdit pipeline request are different additions." + }, + { + "left": "issue:1298", + "right": "issue:4542", + "accept": false, + "reason": "Both concern img2img pipelines, but one is a regression in Stable Diffusion img2img and the other is SDXL img2img output count limitation; different failures." + }, + { + "left": "issue:1675", + "right": "issue:5374", + "accept": false, + "reason": "Model conversion state-dict error and unexpected checkpoint_path argument are unrelated conversion/API issues." + }, + { + "left": "issue:176", + "right": "issue:4069", + "accept": false, + "reason": "Glide pipeline missing and unloading a textual inversion checkpoint are not the same bug." + }, + { + "left": "issue:1949", + "right": "issue:4303", + "accept": false, + "reason": "DreamBooth tutorial training errors and FlaxStableDiffusionPipeline loading SDXL are different subsystems and failures." + }, + { + "left": "issue:2153", + "right": "issue:477", + "accept": false, + "reason": "GPU float16 compatibility and a PNDM off-by-one scheduler error are unrelated." + }, + { + "left": "issue:2377", + "right": "issue:338", + "accept": false, + "reason": "Both ask for tests, but for different components (EMAModel vs LMS Discrete); not duplicates." + }, + { + "left": "issue:2377", + "right": "issue:339", + "accept": false, + "reason": "Same as above: different test targets, no shared underlying bug/change." + }, + { + "left": "issue:2856", + "right": "issue:6113", + "accept": false, + "reason": "Model conversion failure and how to use a control-model repo in diffusers are distinct user problems." + }, + { + "left": "issue:3112", + "right": "issue:4042", + "accept": false, + "reason": "Both touch ControlNet composition, but one requests multi-ControlNet inpainting and the other asks how to combine reference-only with inpainting and depth; related area, not the same change." + }, + { + "left": "issue:3565", + "right": "issue:8236", + "accept": false, + "reason": "OOM in text-to-image training and wrong LR scheduler step count on multi-GPU are different training issues." + }, + { + "left": "issue:4052", + "right": "issue:5405", + "accept": false, + "reason": "Missing from_single_file API and sdxl.from_single_file not overwriting weights are different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 255, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10767", + "issue:1109", + "issue:11148", + "issue:11410", + "issue:11945", + "issue:1195", + "issue:1323", + "issue:13279", + "issue:1519", + "issue:1541", + "issue:1611", + "issue:1674", + "issue:1811", + "issue:2661", + "issue:2662", + "issue:2769", + "issue:3064", + "issue:3116", + "issue:3296", + "issue:3428", + "issue:351", + "issue:3566", + "issue:3767", + "issue:4125", + "issue:4134", + "issue:4376", + "issue:4414", + "issue:460", + "issue:4880", + "issue:5372", + "issue:5761", + "issue:5798", + "issue:6068", + "issue:6090", + "issue:6091", + "issue:6157", + "issue:6288", + "issue:6463", + "issue:7428", + "issue:7619", + "issue:7802", + "issue:7872", + "issue:8589", + "issue:8708", + "issue:8837", + "issue:9092", + "issue:9551", + "issue:9895" + ], + "soft_pairs": [ + "issue:4125|issue:6288", + "issue:6157|issue:9092", + "issue:6463|issue:8708", + "issue:7428|issue:7802", + "issue:8589|issue:8837", + "issue:2661|issue:3296", + "issue:2662|issue:3296", + "issue:3428|issue:3566", + "issue:4880|issue:5798", + "issue:11945|issue:9895", + "issue:1109|issue:460", + "issue:4414|issue:5761", + "issue:11410|issue:7872", + "issue:1541|issue:3767", + "issue:11148|issue:13279", + "issue:3116|issue:6068", + "issue:6090|issue:7619", + "issue:1195|issue:1611", + "issue:1519|issue:4134", + "issue:1323|issue:1674", + "issue:4376|issue:5372", + "issue:10767|issue:9551", + "issue:1811|issue:3064", + "issue:2769|issue:351", + "issue:4414|issue:6091" + ], + "prepared_review_unit_hash": "988ab67530a5032173ad2bac14be7735e4c21350b3f59907b597d3bf5257357f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34295, + "estimated_input_tokens": 8574, + "estimated_eval_tokens": 17404 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous Diffusers issues with only a couple of plausible duplicate pairs. The strongest duplicate is the checkpoint write-error pair (2661/2662); a second likely duplicate is the DreamBooth channel-mismatch pair (6090/7619).", + "confidence": 0.91, + "canonical_issue_reason": "No single cluster-wide canonical issue fits well because the items span many unrelated bugs and feature requests; among the clear duplicates, issue 2662 is the broader representative for the checkpoint write-error bug.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 2662 is the best issue representative because it uses the broadest wording for the checkpoint save/write-error problem and is the cleanest anchor for deduping its pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4125", + "right": "issue:6288", + "accept": false, + "reason": "Both are LoRA-training questions, but one is about loss curves and the other about default learning rate; not the same bug/change." + }, + { + "left": "issue:6157", + "right": "issue:9092", + "accept": false, + "reason": "ControlNet/IP-Adapter behavior vs a missing model_index.json error are different problems." + }, + { + "left": "issue:6463", + "right": "issue:8708", + "accept": false, + "reason": "Unrelated topics: missing audio_diffusion pipeline vs SD3 dreambooth VAE shift_factor." + }, + { + "left": "issue:7428", + "right": "issue:7802", + "accept": false, + "reason": "Feature request for layer diffusion support is unrelated to an OLSS scheduler request." + }, + { + "left": "issue:8589", + "right": "issue:8837", + "accept": false, + "reason": "DreamBooth prior-preservation bug and SlicedAttnProcessor bug are different code paths." + }, + { + "left": "issue:2661", + "right": "issue:3296", + "accept": false, + "reason": "Both mention checkpointing, but 3296 is a DreamBooth text_encoder save/load issue, not the same write-error bug as 2661." + }, + { + "left": "issue:2662", + "right": "issue:3296", + "accept": false, + "reason": "Checkpoint write error vs DreamBooth text_encoder checkpoint save/load failure are distinct failures." + }, + { + "left": "issue:3428", + "right": "issue:3566", + "accept": false, + "reason": "Different pipelines and symptoms; this is not the same underlying bug." + }, + { + "left": "issue:4880", + "right": "issue:5798", + "accept": false, + "reason": "One is a ControlNet image-argument TypeError, the other is a callback keyword-argument API mismatch." + }, + { + "left": "issue:11945", + "right": "issue:9895", + "accept": false, + "reason": "Nightly PyTorch/CUDA floating-point exception vs FluxPipeline device mismatch are different runtime errors." + }, + { + "left": "issue:1109", + "right": "issue:460", + "accept": false, + "reason": "CPU core utilization and ONNX CUDA/CPU runtime selection are unrelated." + }, + { + "left": "issue:4414", + "right": "issue:5761", + "accept": false, + "reason": "Consistency-model sample request and consistency-decoder cost discussion are not the same issue." + }, + { + "left": "issue:11410", + "right": "issue:7872", + "accept": false, + "reason": "A specific attention __call__ unpacking ValueError is unrelated to a version-tag issue." + }, + { + "left": "issue:1541", + "right": "issue:3767", + "accept": false, + "reason": "DreamBooth model-type detection failure is unrelated to FrozenDict mutability." + }, + { + "left": "issue:11148", + "right": "issue:13279", + "accept": false, + "reason": "PEFT adapter state handling and AutoencoderRAE loading with older transformers are different bugs." + }, + { + "left": "issue:3116", + "right": "issue:6068", + "accept": false, + "reason": "KDPM2DiscreteScheduler image2image noise issue is not the same as SDXL DDIM sample correctness." + }, + { + "left": "issue:6090", + "right": "issue:7619", + "accept": true, + "reason": "Both report the same underlying channel-mismatch problem (8 channels expected, 4 provided) in DreamBooth/related diffusion code." + }, + { + "left": "issue:1195", + "right": "issue:1611", + "accept": false, + "reason": "fp16 memory-efficient attention and StableDiffusionMegaPipeline regressions are not the same failure." + }, + { + "left": "issue:1519", + "right": "issue:4134", + "accept": false, + "reason": "Apple MPS issues in both, but noisy batch outputs and ControlNet tensor-size mismatch are different bugs." + }, + { + "left": "issue:1323", + "right": "issue:1674", + "accept": false, + "reason": "Loading Flax models without dtype and DreamBooth scheduling/config loading are unrelated." + }, + { + "left": "issue:4376", + "right": "issue:5372", + "accept": false, + "reason": "Textual inversion crash in SDXL and safety_checker usage questions are different issues." + }, + { + "left": "issue:10767", + "right": "issue:9551", + "accept": false, + "reason": "Flux-related community pipeline topics, but not the same concrete bug or change." + }, + { + "left": "issue:1811", + "right": "issue:3064", + "accept": false, + "reason": "Weight-selection behavior in from_pretrained vs loading a safetensors LoRA are distinct problems." + }, + { + "left": "issue:2769", + "right": "issue:351", + "accept": false, + "reason": "Checkpoint conversion corruption and CUDA loading behavior are unrelated." + }, + { + "left": "issue:4414", + "right": "issue:6091", + "accept": false, + "reason": "Consistency-model sampling and consistency-decoder memory-size handling are different concerns." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: the cluster is heterogeneous, and 2662 is a reasonable broader representative for the checkpoint write-error duplicate pair. The soft-edge verdicts are mostly conservative, and the accepted 6090/7619 pair is defensible from the titles as a DreamBooth channel-mismatch duplicate." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 256, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1037", + "issue:10518", + "issue:1064", + "issue:10640", + "issue:11208", + "issue:11839", + "issue:12149", + "issue:12195", + "issue:12995", + "issue:1344", + "issue:1742", + "issue:1755", + "issue:1974", + "issue:1990", + "issue:2361", + "issue:2578", + "issue:3222", + "issue:3537", + "issue:4014", + "issue:4297", + "issue:4478", + "issue:4534", + "issue:4593", + "issue:4932", + "issue:5008", + "issue:5043", + "issue:5079", + "issue:5414", + "issue:562", + "issue:5857", + "issue:5968", + "issue:6518", + "issue:6548", + "issue:6616", + "issue:6749", + "issue:7210", + "issue:7589", + "issue:7666", + "issue:7757", + "issue:7895", + "issue:7972", + "issue:8463", + "issue:8615", + "issue:8671", + "issue:8786", + "issue:8894", + "issue:984" + ], + "soft_pairs": [ + "issue:5414|issue:7895", + "issue:5857|issue:6548", + "issue:4478|issue:7972", + "issue:4593|issue:6749", + "issue:1037|issue:5043", + "issue:10518|issue:8671", + "issue:10640|issue:7210", + "issue:12149|issue:12195", + "issue:1344|issue:1755", + "issue:3537|issue:5968", + "issue:5008|issue:6616", + "issue:7666|issue:8786", + "issue:4014|issue:4297", + "issue:11208|issue:7589", + "issue:11839|issue:12995", + "issue:1755|issue:1990", + "issue:2361|issue:4534", + "issue:5079|issue:8615", + "issue:6518|issue:8463", + "issue:1742|issue:1974", + "issue:3222|issue:4932", + "issue:7757|issue:8894", + "issue:1064|issue:562", + "issue:2578|issue:984" + ], + "prepared_review_unit_hash": "96705845007fa804b8dbda05f86e590393b292c0ca264025814e6ccc07a552c7", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33587, + "estimated_input_tokens": 8397, + "estimated_eval_tokens": 17050 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are not duplicates of one another; the cluster is a loose lexical-similarity bucket containing unrelated issues across different pipelines, schedulers, training scripts, and environment bugs. All soft edges should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:12195 is the strongest standalone report: it is concrete, version-specific, and clearly describes a reproducible runtime regression in AutoencoderKL with PyTorch 2.3.1.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5414", + "right": "issue:7895", + "accept": false, + "reason": "PEFT adapter weighting vs Flax attention are unrelated subsystems and failure modes." + }, + { + "left": "issue:5857", + "right": "issue:6548", + "accept": false, + "reason": "An LCM LoRA quality issue is unrelated to an img2img latent-vs-image RGBA bug." + }, + { + "left": "issue:4478", + "right": "issue:7972", + "accept": false, + "reason": "SDXL initialization dtype mismatch is not the same as LEDITS++ type mismatch." + }, + { + "left": "issue:4593", + "right": "issue:6749", + "accept": false, + "reason": "LoRA rank-loading error is unrelated to an InstantID reproduction issue." + }, + { + "left": "issue:1037", + "right": "issue:5043", + "accept": false, + "reason": "A community pipeline request is unrelated to a Wurstchen LoRA feature request." + }, + { + "left": "issue:10518", + "right": "issue:8671", + "accept": false, + "reason": "A broken SD3 Colab script is not the same as redundant code in train_dreambooth_lora_sd3.py." + }, + { + "left": "issue:10640", + "right": "issue:7210", + "accept": false, + "reason": "DataParallel on FluxInpaintPipeline is unrelated to accelerate.load_checkpoint_and_dispatch on SDXL." + }, + { + "left": "issue:12149", + "right": "issue:12195", + "accept": false, + "reason": "Source/package version mismatch is a different problem from the torch.library.custom_op regression." + }, + { + "left": "issue:1344", + "right": "issue:1755", + "accept": false, + "reason": "Multi-GPU attention support and a non-English generic issue are unrelated." + }, + { + "left": "issue:3537", + "right": "issue:5968", + "accept": false, + "reason": "A ControlNet conversion script request is unrelated to an ONNX video conversion script request." + }, + { + "left": "issue:5008", + "right": "issue:6616", + "accept": false, + "reason": "TryOnDiffusion implementation discussion is unrelated to adding classifier guidance." + }, + { + "left": "issue:7666", + "right": "issue:8786", + "accept": false, + "reason": "from_single_file support for a custom pipeline is unrelated to SD3 token-length behavior." + }, + { + "left": "issue:4014", + "right": "issue:4297", + "accept": false, + "reason": "General SDXL watermark artifacts and batched img2img fishnet artifacts are different concrete bugs." + }, + { + "left": "issue:11208", + "right": "issue:7589", + "accept": false, + "reason": "MultiControlNet support in SD3 inpainting is unrelated to GLIGEN layout-to-image failure." + }, + { + "left": "issue:11839", + "right": "issue:12995", + "accept": false, + "reason": "GGUF input-size loading failure is unrelated to a Cosmos-Predict2.5 pipeline usage question." + }, + { + "left": "issue:1755", + "right": "issue:1990", + "accept": false, + "reason": "A generic issue title is unrelated to a depth-to-image variation pipeline proposal." + }, + { + "left": "issue:2361", + "right": "issue:4534", + "accept": false, + "reason": "Checkpoint saving crash in a finetuning script is unrelated to SDXL validation on mps." + }, + { + "left": "issue:5079", + "right": "issue:8615", + "accept": false, + "reason": "LyCORIS loading support and SD3 fuse_lora support are different adapter/model features." + }, + { + "left": "issue:6518", + "right": "issue:8463", + "accept": false, + "reason": "LCM distillation NoneType error is unrelated to a tokenizer NoneType error in InstructP2P-SDXL." + }, + { + "left": "issue:1742", + "right": "issue:1974", + "accept": false, + "reason": "Inference speed variability for repeated prompts is unrelated to Euler scheduler generator-list handling." + }, + { + "left": "issue:3222", + "right": "issue:4932", + "accept": false, + "reason": "Missing custom diffusion weights after training is unrelated to Dreambooth not returning pytorch_model.bin." + }, + { + "left": "issue:7757", + "right": "issue:8894", + "accept": false, + "reason": "ControlNet SDXL multi-GPU OOM is unrelated to a train_cm_ct_unconditional script error." + }, + { + "left": "issue:1064", + "right": "issue:562", + "accept": false, + "reason": "LMSDiscreteScheduler.add_noise error and StableDiffusionInpaintPipeline/LMS incompatibility are related in theme but not clearly the same concrete bug." + }, + { + "left": "issue:2578", + "right": "issue:984", + "accept": false, + "reason": "Editable install failure is unrelated to an F.interpolate large-batch interpolation bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 257, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10290", + "issue:10737", + "issue:10953", + "issue:11497", + "issue:1161", + "issue:1214", + "issue:12547", + "issue:1314", + "issue:1875", + "issue:1922", + "issue:2192", + "issue:2331", + "issue:3279", + "issue:3383", + "issue:3403", + "issue:3689", + "issue:3717", + "issue:3767", + "issue:381", + "issue:3876", + "issue:4210", + "issue:435", + "issue:4665", + "issue:4701", + "issue:4876", + "issue:4957", + "issue:4969", + "issue:4985", + "issue:5064", + "issue:5124", + "issue:514", + "issue:5162", + "issue:5284", + "issue:5465", + "issue:5558", + "issue:5646", + "issue:6112", + "issue:6149", + "issue:6164", + "issue:6277", + "issue:6598", + "issue:7177", + "issue:776", + "issue:7939", + "issue:8228", + "issue:9089", + "issue:9611" + ], + "soft_pairs": [ + "issue:381|issue:435", + "issue:4957|issue:5064", + "issue:3279|issue:5124", + "issue:5465|issue:9611", + "issue:10737|issue:9089", + "issue:1161|issue:5162", + "issue:1314|issue:4701", + "issue:2331|issue:5284", + "issue:3689|issue:4876", + "issue:3876|issue:5558", + "issue:4969|issue:514", + "issue:4969|issue:7177", + "issue:11497|issue:12547", + "issue:5646|issue:6598", + "issue:1214|issue:3767", + "issue:1922|issue:4210", + "issue:2192|issue:6277", + "issue:3403|issue:776", + "issue:6112|issue:6149", + "issue:7939|issue:8228", + "issue:1875|issue:6164", + "issue:3717|issue:4985", + "issue:3383|issue:4665", + "issue:10290|issue:10953" + ], + "prepared_review_unit_hash": "4082e174959371c7b94cbb5ff009b06c23791ece841ca2d014b229f86af8f44a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33278, + "estimated_input_tokens": 8320, + "estimated_eval_tokens": 16896 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set; it\u2019s a bag of unrelated issues that only share superficial vocabulary around pipelines, training, Flux, ControlNet, and loading. I would reject all soft-edge merges and avoid selecting a canonical issue/PR for the cluster.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:381", + "right": "issue:435", + "accept": false, + "reason": "Different bugs: a notebook section error vs a textual inversion AutoencoderKLOutput attribute error." + }, + { + "left": "issue:4957", + "right": "issue:5064", + "accept": false, + "reason": "Both mention space issues, but one is disk space during ControlNet+inpainting-SDXL and the other is out-of-space using a refiner for inference; different workflows and failure modes." + }, + { + "left": "issue:3279", + "right": "issue:5124", + "accept": false, + "reason": "Checkpoint resume failure in Dreambooth LoRA is unrelated to a question about finetuning a .safetensor checkpoint." + }, + { + "left": "issue:5465", + "right": "issue:9611", + "accept": false, + "reason": "Different pipelines and symptoms: DiffEdit inpaint_strength limitation vs Flux image batching error." + }, + { + "left": "issue:10737", + "right": "issue:9089", + "accept": false, + "reason": "One is missing image_encoder in FluxPipeline.encode_image; the other is FluxPipeline not found at import time." + }, + { + "left": "issue:1161", + "right": "issue:5162", + "accept": false, + "reason": "Model conversion from Flax to PyTorch is unrelated to a script for converting a folder checkpoint to safetensors." + }, + { + "left": "issue:1314", + "right": "issue:4701", + "accept": false, + "reason": "Both are CUDA/precision-related, but one is a Windows xformers backward error and the other is AutoencoderKL decoder differentiation at float32." + }, + { + "left": "issue:2331", + "right": "issue:5284", + "accept": false, + "reason": "A general ControlNet pipeline request vs ControlNet-LLLite support; same broad area, but not the same change or bug." + }, + { + "left": "issue:3689", + "right": "issue:4876", + "accept": false, + "reason": "LoRA unload behavior is unrelated to making LoRA std configurable." + }, + { + "left": "issue:3876", + "right": "issue:5558", + "accept": false, + "reason": "Both involve ControlNet inpainting, but one is a demo error and the other is artifacts with multiple ControlNets; not the same concrete defect." + }, + { + "left": "issue:4969", + "right": "issue:514", + "accept": false, + "reason": "Different import/error contexts: missing CrossAttention vs enable_attention_slicing() throwing an error." + }, + { + "left": "issue:4969", + "right": "issue:7177", + "accept": false, + "reason": "Missing CrossAttention import is unrelated to StableCascadeDecoderPipeline.from_pretrained failure." + }, + { + "left": "issue:11497", + "right": "issue:12547", + "accept": false, + "reason": "Unrelated training-script issues: cache depending on irrelevant args vs missing required --instance_prompt argument." + }, + { + "left": "issue:5646", + "right": "issue:6598", + "accept": false, + "reason": "Scheduler stability on a specific model is unrelated to img2img result differences vs A1111." + }, + { + "left": "issue:1214", + "right": "issue:3767", + "accept": false, + "reason": "Half-precision CLIPTextModel failure is unrelated to FrozenDict mutability semantics." + }, + { + "left": "issue:1922", + "right": "issue:4210", + "accept": false, + "reason": "Notebook undefined function bug is unrelated to a question about attention_mask in UNet2DConditionModel.forward." + }, + { + "left": "issue:2192", + "right": "issue:6277", + "accept": false, + "reason": "Missing dataset_name/caption_column arguments in one script is unrelated to broken gradients in a LoRA training example." + }, + { + "left": "issue:3403", + "right": "issue:776", + "accept": false, + "reason": "TorchDynamo pipeline loading issues are not the same as batch_size>1 img2img breakage." + }, + { + "left": "issue:6112", + "right": "issue:6149", + "accept": false, + "reason": "Noisy SDXL-turbo output is unrelated to an ip-adapter-full-face_sd15 load error." + }, + { + "left": "issue:7939", + "right": "issue:8228", + "accept": false, + "reason": "Both are loading-related, but one is loading a model from the same folder and the other is from_single_file failing to load CLIPTextModel; different code paths." + }, + { + "left": "issue:1875", + "right": "issue:6164", + "accept": false, + "reason": "fp16 training type mismatch in Imagic is unrelated to latent output handling in StableDiffusionPipeline." + }, + { + "left": "issue:3717", + "right": "issue:4985", + "accept": false, + "reason": "GPU OOM on multi-GPU hardware is unrelated to SDXL InstructPix2Pix training OOM; different scenario and likely different scaling cause." + }, + { + "left": "issue:3383", + "right": "issue:4665", + "accept": false, + "reason": "Partial pipeline download handling is unrelated to Windows use_safetensors behavior with alternate .bin files." + }, + { + "left": "issue:10290", + "right": "issue:10953", + "accept": false, + "reason": "Torch.compile incompatibility for Flux+ControlNet is unrelated to a code-style/performance suggestion to use permute instead of transpose." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 258, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10057", + "issue:10165", + "issue:10239", + "issue:10767", + "issue:10860", + "issue:1111", + "issue:11247", + "issue:11592", + "issue:11864", + "issue:12108", + "issue:12157", + "issue:12221", + "issue:12338", + "issue:1574", + "issue:1790", + "issue:2264", + "issue:2402", + "issue:3481", + "issue:3651", + "issue:3655", + "issue:3683", + "issue:4080", + "issue:4366", + "issue:4506", + "issue:5225", + "issue:5284", + "issue:5395", + "issue:5463", + "issue:5489", + "issue:6091", + "issue:6335", + "issue:6398", + "issue:6452", + "issue:6530", + "issue:660", + "issue:7009", + "issue:7191", + "issue:7228", + "issue:7640", + "issue:7781", + "issue:7881", + "issue:8464", + "issue:8879", + "issue:8900", + "issue:9370", + "issue:953", + "issue:9810" + ], + "soft_pairs": [ + "issue:10767|issue:12221", + "issue:10860|issue:9810", + "issue:2402|issue:660", + "issue:4080|issue:6452", + "issue:7640|issue:8879", + "issue:11592|issue:8900", + "issue:5395|issue:5489", + "issue:12108|issue:12157", + "issue:1574|issue:1790", + "issue:4506|issue:5463", + "issue:6091|issue:6398", + "issue:6335|issue:7881", + "issue:10239|issue:7009", + "issue:11864|issue:12338", + "issue:3651|issue:3683", + "issue:7228|issue:7781", + "issue:10057|issue:11247", + "issue:5225|issue:9370", + "issue:1111|issue:953", + "issue:5284|issue:6530", + "issue:2264|issue:3481", + "issue:3655|issue:7191", + "issue:4080|issue:4366", + "issue:10165|issue:8464" + ], + "prepared_review_unit_hash": "4853dfb235d9bf16ee3e99343ac6149984f6e2fa23a6153ddc95050ce37e2fb8", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33322, + "estimated_input_tokens": 8331, + "estimated_eval_tokens": 16918 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy: the items share broad Diffusers/Flux/ControlNet/LoRA themes, but they describe different code paths, platforms, and failure modes rather than the same underlying bug or change.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 10057 is the most concrete, narrowly scoped bug report in the set (a specific FluxTransformer2DModel.from_single_file token problem) and is a reasonable cluster head, though the cluster itself is not truly coherent.", + "canonical_pr_reason": null, + "best_issue_reason": "10057 is the clearest actionable issue: it names a specific model class and method, implies a reproducible loading path, and is more bug-like than the feature requests or broad compatibility reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10767", + "right": "issue:12221", + "accept": false, + "reason": "Both mention diffusers integrations, but one is a FlashVideo/diffusers topic and the other is a request to support DiffSynth Controlnet; different concrete work." + }, + { + "left": "issue:10860", + "right": "issue:9810", + "accept": false, + "reason": "One is a ComfyUI compatibility failure for trained LoRA weights, the other is a module_to_save() support request; related area, not the same bug." + }, + { + "left": "issue:2402", + "right": "issue:660", + "accept": false, + "reason": "These are different backend/platform problems: CUDA-not-enabled vs MPS fp16 behavior on M1." + }, + { + "left": "issue:4080", + "right": "issue:6452", + "accept": false, + "reason": "Both concern inpainting, but one asks for a new SDXL inpainting pipeline while the other asks for a masked_image_latents input on ControlNet inpaint; not the same change." + }, + { + "left": "issue:7640", + "right": "issue:8879", + "accept": false, + "reason": "Both involve StableDiffusionXL, but the failures are different: from_single_file dimension handling vs torch.compile/cudagraph issues." + }, + { + "left": "issue:11592", + "right": "issue:8900", + "accept": false, + "reason": "Unrelated topics: a FluxLoraLoaderMixin regression versus a question about loading an SDXL refiner checkpoint." + }, + { + "left": "issue:5395", + "right": "issue:5489", + "accept": false, + "reason": "One is a specific bug combining Civitai LoRAs with PEFT; the other is a broad PEFT x diffusers integration umbrella issue." + }, + { + "left": "issue:12108", + "right": "issue:12157", + "accept": false, + "reason": "Different problem types: inference breakage with flow-matching schedulers vs a training-time quantization-config error." + }, + { + "left": "issue:1574", + "right": "issue:1790", + "accept": false, + "reason": "A Dreambooth GPU failure is not the same as a general question about how training works." + }, + { + "left": "issue:4506", + "right": "issue:5463", + "accept": false, + "reason": "Both are variable-scope bugs, but they affect different variables, code paths, and symptoms; not the same defect." + }, + { + "left": "issue:6091", + "right": "issue:6398", + "accept": false, + "reason": "One is a specific Consistency Decoder memory issue tied to size arguments; the other is an unspecific memory-leak report." + }, + { + "left": "issue:6335", + "right": "issue:7881", + "accept": false, + "reason": "Different pipeline problems: output-size handling in ControlNet Img2Img vs from_single_file loading for lpw_stable_diffusion." + }, + { + "left": "issue:10239", + "right": "issue:7009", + "accept": false, + "reason": "Both are video-related, but one is about sequential CPU offload support and the other about poor results from a specific TextToVideoSD model load." + }, + { + "left": "issue:11864", + "right": "issue:12338", + "accept": false, + "reason": "Same AutoencoderDC area, but different bugs: torch.compile fullgraph name resolution vs VAE slicing when decoding multiple images." + }, + { + "left": "issue:3651", + "right": "issue:3683", + "accept": false, + "reason": "Multi-GPU Dreambooth training and Multi-ControlNet save_pretrained() are unrelated changes." + }, + { + "left": "issue:7228", + "right": "issue:7781", + "accept": false, + "reason": "A traced-model dtype attribute error and a PIL serialization error are distinct failure modes." + }, + { + "left": "issue:10057", + "right": "issue:11247", + "accept": false, + "reason": "Both mention FluxTransformer2DModel, but one is a from_single_file token issue and the other is multi-GPU/controlnet distribution; different code paths." + }, + { + "left": "issue:5225", + "right": "issue:9370", + "accept": false, + "reason": "A VRAM/memory limitation is not the same as a missing pooled_projection argument in a forward method." + }, + { + "left": "issue:1111", + "right": "issue:953", + "accept": false, + "reason": "These are unrelated: a checkpoint dataset constraint versus guidance on generating smaller images." + }, + { + "left": "issue:5284", + "right": "issue:6530", + "accept": false, + "reason": "ControlNet-LLLite support is a feature request; SDXL K-diffusion with ControlNet is a separate pipeline compatibility issue." + }, + { + "left": "issue:2264", + "right": "issue:3481", + "accept": false, + "reason": "Both involve seeding/determinism, but one is an exception when setting the seed and the other is expected image variation across inference steps." + }, + { + "left": "issue:3655", + "right": "issue:7191", + "accept": false, + "reason": "Both are import-related, but the titles do not indicate the same module, backend, or failure mechanism." + }, + { + "left": "issue:4080", + "right": "issue:4366", + "accept": false, + "reason": "An SDXL inpainting pipeline request is not the same as fine-tuning the SDXL text-to-image script for a full U-Net." + }, + { + "left": "issue:10165", + "right": "issue:8464", + "accept": false, + "reason": "Both mention LoRA/seed behavior, but one is about preserving seeds when changing LoRA scale and the other is a SEGA CPU-offload seed failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 259, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10226", + "issue:10381", + "issue:10447", + "issue:10453", + "issue:10743", + "issue:10925", + "issue:1101", + "issue:12321", + "issue:12437", + "issue:12540", + "issue:13015", + "issue:13298", + "issue:1647", + "issue:2203", + "issue:2437", + "issue:2669", + "issue:3455", + "issue:4139", + "issue:4149", + "issue:4353", + "issue:4406", + "issue:4410", + "issue:4530", + "issue:4991", + "issue:5004", + "issue:5070", + "issue:5271", + "issue:5405", + "issue:5509", + "issue:5685", + "issue:5783", + "issue:6016", + "issue:6037", + "issue:6295", + "issue:6496", + "issue:6518", + "issue:6713", + "issue:7138", + "issue:7147", + "issue:7450", + "issue:7600", + "issue:761", + "issue:7798", + "issue:7856", + "issue:8612", + "issue:8918", + "issue:8939", + "issue:9450" + ], + "soft_pairs": [ + "issue:10226|issue:7138", + "issue:4530|issue:6295", + "issue:10453|issue:12540", + "issue:13015|issue:13298", + "issue:12321|issue:12437", + "issue:1647|issue:3455", + "issue:2203|issue:761", + "issue:2437|issue:6037", + "issue:4139|issue:8612", + "issue:4149|issue:5509", + "issue:4991|issue:5405", + "issue:5004|issue:7600", + "issue:5271|issue:6496", + "issue:5783|issue:8918", + "issue:7147|issue:7856", + "issue:10381|issue:9450", + "issue:4410|issue:6016", + "issue:10447|issue:8939", + "issue:4353|issue:7798", + "issue:4406|issue:6713", + "issue:5685|issue:7450", + "issue:10743|issue:10925", + "issue:2669|issue:6518", + "issue:1101|issue:5070" + ], + "prepared_review_unit_hash": "c8817a99115052b8d555e037601514fe083dae8d44827a311e03bb8954e6cc69", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34196, + "estimated_input_tokens": 8549, + "estimated_eval_tokens": 17354 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous, with a small Flux/device-mismatch subgroup but many unrelated training, conversion, scheduler, and feature-request issues. No soft pairs look like safe duplicates.", + "confidence": 0.36, + "canonical_issue_reason": "issue:9450 is the clearest and broadest Flux device-mismatch report in the set, so it is the best anchor for the small overlapping subgroup.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9450 is the strongest triage representative: explicit FluxPipeline device mismatch, concrete error text, and broader applicability than the narrower Flux/LoRA variants.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10226", + "right": "issue:7138", + "accept": false, + "reason": "Both mention model-shape mismatch, but one is Dreambooth Flux+LoRA and the other is SDXL UNet structure mismatch; different models and likely different bugs." + }, + { + "left": "issue:4530", + "right": "issue:6295", + "accept": false, + "reason": "Both are artifact-quality complaints, but the symptoms and likely causes differ: SDXL refiner color artifacts vs DPM++ 2M SDE Karras artifacts." + }, + { + "left": "issue:10453", + "right": "issue:12540", + "accept": false, + "reason": "Different pipelines and failures: Hunyuan Video batch-size limitation vs SanaCombinedTimestepGuidanceEmbeddings incompatibility with SanaPipeline." + }, + { + "left": "issue:13015", + "right": "issue:13298", + "accept": false, + "reason": "Both are device-mismatch errors, but one is QwenImageEditPlus on NPU and the other is Flux image generation with a CUDA generator; not the same concrete code path." + }, + { + "left": "issue:12321", + "right": "issue:12437", + "accept": false, + "reason": "Separate feature requests for different models: HunyuanImage-2.1 support vs OmniSVG support." + }, + { + "left": "issue:1647", + "right": "issue:3455", + "accept": false, + "reason": "A usage question about color-correction .pt files is unrelated to training InstructPix2Pix with a pretrained model." + }, + { + "left": "issue:2203", + "right": "issue:761", + "accept": false, + "reason": "Different problems: download location/path handling vs DreamBooth failure with 8-bit Adam." + }, + { + "left": "issue:2437", + "right": "issue:6037", + "accept": false, + "reason": "Different pipeline additions: T2I-Adapter vs InstaFlow." + }, + { + "left": "issue:4139", + "right": "issue:8612", + "accept": false, + "reason": "Both involve precision, but they affect different models and code paths: custom diffusion attention processor casting vs PixArt-sigma mixed-precision inference." + }, + { + "left": "issue:4149", + "right": "issue:5509", + "accept": false, + "reason": "Both are batch-dimension mismatches, but they occur in different pipelines and inputs; not enough evidence of one underlying bug." + }, + { + "left": "issue:4991", + "right": "issue:5405", + "accept": false, + "reason": "Training setup bug vs single-file SDXL weight-overwrite issue; unrelated failures." + }, + { + "left": "issue:5004", + "right": "issue:7600", + "accept": false, + "reason": "Checkpoint-resume fp16 bug and missing accelerate.accumulate on text_encoder are distinct training issues." + }, + { + "left": "issue:5271", + "right": "issue:6496", + "accept": false, + "reason": "Different DreamBooth requests: separate learning rates vs multiple people/faces example support." + }, + { + "left": "issue:5783", + "right": "issue:8918", + "accept": false, + "reason": "SDXL EMA handling and a broken SD3 DreamBooth script are unrelated training bugs." + }, + { + "left": "issue:7147", + "right": "issue:7856", + "accept": false, + "reason": "Import error for StableCascadeDecoderPipeline vs tokenizer multi-vector token collision; different subsystems." + }, + { + "left": "issue:10381", + "right": "issue:9450", + "accept": false, + "reason": "Both are Flux/device-related, but 10381 centers on quantization/LoRA/Quanto and zero-GPU spaces, while 9450 is a multi-GPU transformer device-placement issue." + }, + { + "left": "issue:4410", + "right": "issue:6016", + "accept": false, + "reason": "Both involve conversion/loading shape problems, but the specific failures differ: meta-device conv_in.weight mismatch vs a torch.Size() issue in the converter." + }, + { + "left": "issue:10447", + "right": "issue:8939", + "accept": false, + "reason": "Different issues in different tools: UNet split-file conversion breakage vs SD3 tokenizer load failure." + }, + { + "left": "issue:4353", + "right": "issue:7798", + "accept": false, + "reason": "Unrelated quality bug reports for different pipelines: SDXL refiner artifacts vs PixArtAlphaPipeline bug." + }, + { + "left": "issue:4406", + "right": "issue:6713", + "accept": false, + "reason": "Same error class appears in the title, but one is a generic safetensor deserialization issue and the other is specific to the conversion script; not enough evidence they are the same underlying bug." + }, + { + "left": "issue:5685", + "right": "issue:7450", + "accept": false, + "reason": "Both mention scheduler or threading concerns, but they are different issues: scheduler problems vs AutoencoderKL multi-threading." + }, + { + "left": "issue:10743", + "right": "issue:10925", + "accept": false, + "reason": "Flux zero-3 training support and Dreambooth finetuning FLUX CLIPTextModel are related to Flux training but are not the same change or bug." + }, + { + "left": "issue:2669", + "right": "issue:6518", + "accept": false, + "reason": "Different NoneType errors in different examples/scripts; similar exception shape but not the same root cause." + }, + { + "left": "issue:1101", + "right": "issue:5070", + "accept": false, + "reason": "DPM-Solver scheduler addition and a ControlNet example tweak are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 260, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1033", + "issue:11000", + "issue:11432", + "issue:11642", + "issue:1284", + "issue:1992", + "issue:2202", + "issue:2276", + "issue:2784", + "issue:3116", + "issue:3381", + "issue:3658", + "issue:3951", + "issue:4171", + "issue:444", + "issue:4453", + "issue:4664", + "issue:4885", + "issue:4975", + "issue:5062", + "issue:5079", + "issue:5485", + "issue:5798", + "issue:5969", + "issue:6176", + "issue:6454", + "issue:6790", + "issue:696", + "issue:7454", + "issue:7600", + "issue:7690", + "issue:7798", + "issue:7801", + "issue:8535", + "issue:8557", + "issue:9027", + "issue:9044", + "issue:9063", + "issue:9084", + "issue:9090", + "issue:9181", + "issue:9204", + "issue:9332", + "issue:9350", + "issue:9362", + "issue:9484", + "issue:9505", + "issue:9665" + ], + "soft_pairs": [ + "issue:3658|issue:7600", + "issue:4975|issue:9090", + "issue:5079|issue:7454", + "issue:6176|issue:9505", + "issue:6454|issue:7801", + "issue:6790|issue:9204", + "issue:7798|issue:9084", + "issue:11642|issue:9044", + "issue:9484|issue:9665", + "issue:9350|issue:9362", + "issue:1033|issue:444", + "issue:11000|issue:9027", + "issue:1992|issue:5062", + "issue:2202|issue:2784", + "issue:8535|issue:8557", + "issue:4664|issue:4885", + "issue:1284|issue:5485", + "issue:11432|issue:9332", + "issue:7690|issue:9063", + "issue:3116|issue:3381", + "issue:2276|issue:696", + "issue:5798|issue:9181", + "issue:3951|issue:5969", + "issue:4171|issue:4453" + ], + "prepared_review_unit_hash": "e8cd540d4d8989f0a00452487d50bc525660a2e1de90339e8475d414de1565de", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33814, + "estimated_input_tokens": 8454, + "estimated_eval_tokens": 17164 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are isolated bugs/feature requests across different pipelines, schedulers, and training scripts. I don\u2019t see any soft pair that is clearly the same underlying bug or change, so no merges are warranted.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "If one issue must anchor the cluster, #9332 is the most substantial and detailed, but it is still not a true duplicate hub for the rest of the items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3658", + "right": "issue:7600", + "accept": false, + "reason": "Both are DreamBooth-related, but one is about extra noise latents in upsampler scripts and the other about missing accelerator.accumulate on text_encoder; different bugs." + }, + { + "left": "issue:4975", + "right": "issue:9090", + "accept": false, + "reason": "LoRA loading inefficiency and repeated PNDM timesteps are unrelated subsystems and failure modes." + }, + { + "left": "issue:5079", + "right": "issue:7454", + "accept": false, + "reason": "One requests load_lora_weights support for LyCORIS, the other is a generic LORA support request; similar area, not the same change." + }, + { + "left": "issue:6176", + "right": "issue:9505", + "accept": false, + "reason": "DreamBooth LoRA load failure and an upscaler KeyError involve different pipelines and error paths." + }, + { + "left": "issue:6454", + "right": "issue:7801", + "accept": false, + "reason": "Both mention fp16, but one is an AMP gradient unscale error and the other is about mixed_precision not producing fp16 weights." + }, + { + "left": "issue:6790", + "right": "issue:9204", + "accept": false, + "reason": "State-dict key mismatch in LCM training is not the same as the FLUX DreamBooth training-script issue." + }, + { + "left": "issue:7798", + "right": "issue:9084", + "accept": false, + "reason": "PixArtAlphaPipeline bug and SD3 inpaint EulerA scheduler behavior are different code paths." + }, + { + "left": "issue:11642", + "right": "issue:9044", + "accept": false, + "reason": "Prompt embedding scaling not taking effect and SDXL model-file load failure are unrelated problems." + }, + { + "left": "issue:9484", + "right": "issue:9665", + "accept": false, + "reason": "Both involve training/DeepSpeed, but one is FLUX DreamBooth multigpu usage and the other is accelerator.prepare interaction; not the same concrete bug." + }, + { + "left": "issue:9350", + "right": "issue:9362", + "accept": false, + "reason": "A deprecation warning about txt_ids shape is not the same as an index-out-of-bounds runtime error." + }, + { + "left": "issue:1033", + "right": "issue:444", + "accept": false, + "reason": "Both are IndexErrors in diffusion contexts, but the described conditions and pipelines differ too much to treat as one bug." + }, + { + "left": "issue:11000", + "right": "issue:9027", + "accept": false, + "reason": "Missing shift argument in WanImageToVideoPipeline is unrelated to argument-name consistency complaints." + }, + { + "left": "issue:1992", + "right": "issue:5062", + "accept": false, + "reason": "JAX attention refactoring discussion and a processor typo are not duplicate bugs." + }, + { + "left": "issue:2202", + "right": "issue:2784", + "accept": false, + "reason": "OpenVINO support progress and DreamBooth support for text2video are different feature requests." + }, + { + "left": "issue:8535", + "right": "issue:8557", + "accept": false, + "reason": "xformers attention support in SD3 and SD3 training token-limit issues are separate concerns." + }, + { + "left": "issue:4664", + "right": "issue:4885", + "accept": false, + "reason": "Both are LoRA-loading reports, but they target different pipelines and symptoms; not clearly one underlying defect." + }, + { + "left": "issue:1284", + "right": "issue:5485", + "accept": false, + "reason": "General batch-size slowdown is not the same as ControlNet batch inference taking forever." + }, + { + "left": "issue:11432", + "right": "issue:9332", + "accept": false, + "reason": "Both concern from_pretrained behavior, but one is torch_dtype=\"auto\" handling and the other is loading the wrong/identical weights; different bugs." + }, + { + "left": "issue:7690", + "right": "issue:9063", + "accept": false, + "reason": "Tokenizer-loading failures in different models/pipelines are not the same concrete issue." + }, + { + "left": "issue:3116", + "right": "issue:3381", + "accept": false, + "reason": "A scheduler latent-noise bug in image2image is not the same as discolored inpaint batch inference output." + }, + { + "left": "issue:2276", + "right": "issue:696", + "accept": false, + "reason": "xformers installation problems and DreamBooth CUDA OOM are unrelated." + }, + { + "left": "issue:5798", + "right": "issue:9181", + "accept": false, + "reason": "callback_on_step_end API error and strange noise dots in ControlNet do not share a single code-path fix." + }, + { + "left": "issue:3951", + "right": "issue:5969", + "accept": false, + "reason": "Both mention NaNs, but they arise from different models/settings and are not clearly the same defect." + }, + { + "left": "issue:4171", + "right": "issue:4453", + "accept": false, + "reason": "Documentation rendering and LoRA conversion are completely unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 261, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10016", + "issue:10287", + "issue:10518", + "issue:11041", + "issue:11093", + "issue:11387", + "issue:11497", + "issue:11907", + "issue:12096", + "issue:12338", + "issue:12423", + "issue:12895", + "issue:12905", + "issue:1566", + "issue:170", + "issue:1842", + "issue:2083", + "issue:2091", + "issue:2495", + "issue:3116", + "issue:3420", + "issue:4497", + "issue:4676", + "issue:4788", + "issue:4880", + "issue:5226", + "issue:5379", + "issue:5399", + "issue:5480", + "issue:5515", + "issue:6054", + "issue:6167", + "issue:6548", + "issue:6616", + "issue:6700", + "issue:7220", + "issue:759", + "issue:7770", + "issue:7836", + "issue:8543", + "issue:9170", + "issue:9279", + "issue:9291", + "issue:9448", + "issue:9548", + "issue:9680", + "issue:9731", + "issue:9895" + ], + "soft_pairs": [ + "issue:4788|issue:9170", + "issue:5480|issue:6616", + "issue:7836|issue:8543", + "issue:5399|issue:7770", + "issue:10287|issue:9548", + "issue:12895|issue:9448", + "issue:1566|issue:2091", + "issue:11093|issue:9680", + "issue:170|issue:2083", + "issue:4497|issue:759", + "issue:9279|issue:9291", + "issue:11497|issue:9731", + "issue:5226|issue:6167", + "issue:11387|issue:12338", + "issue:2495|issue:3116", + "issue:5379|issue:5515", + "issue:3420|issue:7220", + "issue:1842|issue:6054", + "issue:4880|issue:6700", + "issue:11907|issue:12096", + "issue:11041|issue:9895", + "issue:10016|issue:12423", + "issue:4676|issue:6548", + "issue:10518|issue:12905" + ], + "prepared_review_unit_hash": "1afda699b9ed473b16009c6f9775dd78dfee54259b3406076f5aad34f63cc846", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34265, + "estimated_input_tokens": 8567, + "estimated_eval_tokens": 17390 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are unrelated pipeline/example/model bugs, so I do not see any true duplicate pairs to merge. The only reasonable representative issue is the specific SD3.5M from_single_file shape-mismatch crash, but the rest of the set spans distinct code paths and features.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 10016 is the most concrete and actionable report: it names a specific model, a specific API (`from_single_file`), and a precise failure (`mismatched shape for pos_embed.pos_embed`).", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10016 is the best single representative because it is narrowly scoped, reproducible, and clearly describes the affected loading path and error.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4788", + "right": "issue:9170", + "accept": false, + "reason": "Both are memory-related, but one is SDXL inpaint OOM and the other is a ControlNet+SDXL VRAM requirement question; different symptom and likely different root cause." + }, + { + "left": "issue:5480", + "right": "issue:6616", + "accept": false, + "reason": "These are separate feature requests: EDM implementation vs classifier-guidance implementation." + }, + { + "left": "issue:7836", + "right": "issue:8543", + "accept": false, + "reason": "Different example bugs in different training scripts; overfitting in CLIP training is not the same as gradient checkpointing failure in SD3 dreambooth." + }, + { + "left": "issue:5399", + "right": "issue:7770", + "accept": false, + "reason": "Both mention loading/inference errors, but one is a dtype mismatch in custom inpainting inference and the other is a model+LoRA shape mismatch; not the same bug." + }, + { + "left": "issue:10287", + "right": "issue:9548", + "accept": false, + "reason": "A docs-example issue and a Flux dreambooth LoRA training regression are unrelated." + }, + { + "left": "issue:12895", + "right": "issue:9448", + "accept": false, + "reason": "Different pipelines and failures: batched prompts with `num_images_per_prompt` versus IP-Adapter tuple/shape handling in ControlNet inpaint." + }, + { + "left": "issue:1566", + "right": "issue:2091", + "accept": false, + "reason": "One is fp16 checkpoint saving in Dreambooth; the other is resuming multi-GPU training from checkpoint. Same general area, but not the same concrete bug." + }, + { + "left": "issue:11093", + "right": "issue:9680", + "accept": false, + "reason": "Different broken examples/pipelines with no shared code-path failure." + }, + { + "left": "issue:170", + "right": "issue:2083", + "accept": false, + "reason": "Both involve eta, but one is an unexpected kwarg crash and the other is a multiple-generator/DDIMScheduler limitation; not the same underlying bug." + }, + { + "left": "issue:4497", + "right": "issue:759", + "accept": false, + "reason": "Dynamic thresholding and optional full-model saving for textual inversion are unrelated." + }, + { + "left": "issue:9279", + "right": "issue:9291", + "accept": false, + "reason": "Dynamic LoRA loading with `torch.compile` and FLUX LoRA key compatibility are different LoRA-related issues with different code paths." + }, + { + "left": "issue:11497", + "right": "issue:9731", + "accept": false, + "reason": "Dataset preprocessing cache invalidation and a Playground2.5 LoRA training question are unrelated." + }, + { + "left": "issue:5226", + "right": "issue:6167", + "accept": false, + "reason": "`load_lora_weights` support and `load_ip_adapter` addition are different adapter-loading features." + }, + { + "left": "issue:11387", + "right": "issue:12338", + "accept": false, + "reason": "Both touch AutoencoderDC, but one is a normalization-layer bug and the other is a slicing/multi-image decoding bug." + }, + { + "left": "issue:2495", + "right": "issue:3116", + "accept": false, + "reason": "MPS black-image output and KDPM2DiscreteScheduler latent noise are distinct failures." + }, + { + "left": "issue:5379", + "right": "issue:5515", + "accept": false, + "reason": "A community pipeline request and an LCM pipeline scheduler initialization failure are unrelated." + }, + { + "left": "issue:3420", + "right": "issue:7220", + "accept": false, + "reason": "Both mention ControlNet/IP-Adapter plumbing, but one is a diffusion-pipeline compatibility issue and the other is an `.unload_ip_adapter()` error; different bugs." + }, + { + "left": "issue:1842", + "right": "issue:6054", + "accept": false, + "reason": "Both mention scheduler/eta behavior, but one is a generic unexpected-eta kwarg crash and the other is a `torch.from_numpy` misuse in EulerDiscreteScheduler." + }, + { + "left": "issue:4880", + "right": "issue:6700", + "accept": false, + "reason": "Single-ControlNet image input validation and `temb` forwarding to attention processors are unrelated TypeError sources." + }, + { + "left": "issue:11907", + "right": "issue:12096", + "accept": false, + "reason": "Flux scheduler incompatibility and WanVACEPipeline group-offloading failure are different pipelines and different code paths." + }, + { + "left": "issue:11041", + "right": "issue:9895", + "accept": false, + "reason": "Wan2.1 group-offloading errors and Flux device-placement errors are unrelated." + }, + { + "left": "issue:10016", + "right": "issue:12423", + "accept": false, + "reason": "Both are `from_single_file` loading reports, but they affect different model families and different tensor shapes; not the same bug." + }, + { + "left": "issue:4676", + "right": "issue:6548", + "accept": false, + "reason": "TAESD latents being too dark and RGBA images being misread as latents are different preprocessing/encoding problems." + }, + { + "left": "issue:10518", + "right": "issue:12905", + "accept": false, + "reason": "A LoRA training example script issue and a guidance-scale implementation bug are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 262, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10223", + "issue:10281", + "issue:10917", + "issue:11127", + "issue:11214", + "issue:11387", + "issue:11859", + "issue:1444", + "issue:1543", + "issue:1695", + "issue:1979", + "issue:2083", + "issue:2265", + "issue:3177", + "issue:3249", + "issue:3583", + "issue:3730", + "issue:4000", + "issue:4122", + "issue:4148", + "issue:4303", + "issue:4331", + "issue:4554", + "issue:4648", + "issue:5014", + "issue:5026", + "issue:5028", + "issue:5084", + "issue:5595", + "issue:5603", + "issue:5687", + "issue:579", + "issue:5875", + "issue:6142", + "issue:6535", + "issue:6706", + "issue:6727", + "issue:6731", + "issue:6906", + "issue:6921", + "issue:7456", + "issue:8887", + "issue:9027", + "issue:9134", + "issue:9248", + "issue:9485", + "issue:9677", + "issue:9996" + ], + "soft_pairs": [ + "issue:10917|issue:8887", + "issue:11214|issue:9027", + "issue:1444|issue:5084", + "issue:1979|issue:4000", + "issue:2083|issue:579", + "issue:3583|issue:6706", + "issue:3730|issue:4148", + "issue:4303|issue:6921", + "issue:4554|issue:6906", + "issue:4648|issue:7456", + "issue:5595|issue:6731", + "issue:5603|issue:6142", + "issue:6535|issue:9485", + "issue:6727|issue:9248", + "issue:9134|issue:9677", + "issue:3249|issue:5028", + "issue:10223|issue:11387", + "issue:1543|issue:2265", + "issue:4122|issue:5687", + "issue:11127|issue:9996", + "issue:4331|issue:5014", + "issue:10281|issue:11859", + "issue:1695|issue:3177", + "issue:5026|issue:5875" + ], + "prepared_review_unit_hash": "e75f45c1481c4fca8fe2f0cb46b5eda2a818a749c0f6317fdd0f8550506c6762", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33531, + "estimated_input_tokens": 8383, + "estimated_eval_tokens": 17022 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes scheduler feature requests, training-script bugs, import/load errors, and model-specific runtime issues. No soft pair looks like the same underlying defect, so all soft edges should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9485 is the broadest scheduler-related anchor in the set, but it is still only a loose representative; the cluster is too mixed for a true duplicate canonical.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10917", + "right": "issue:8887", + "accept": false, + "reason": "Different models and code paths: lumina-2.0 script correctness vs SD3 training VAE encode." + }, + { + "left": "issue:11214", + "right": "issue:9027", + "accept": false, + "reason": "A specific UVit2DModel TypeError is not the same as a general API naming-consistency discussion." + }, + { + "left": "issue:1444", + "right": "issue:5084", + "accept": false, + "reason": "Both are import errors, but for different symbols and pipeline families; not the same bug." + }, + { + "left": "issue:1979", + "right": "issue:4000", + "accept": false, + "reason": "Both concern conversion, but one is generic SD conversion and the other is specifically SD2 ckpt conversion." + }, + { + "left": "issue:2083", + "right": "issue:579", + "accept": false, + "reason": "DDIM/generator behavior is unrelated to a missing config.json in textual inversion." + }, + { + "left": "issue:3583", + "right": "issue:6706", + "accept": false, + "reason": "Both are feature requests, but Mixture of Diffusers and cads support are unrelated changes." + }, + { + "left": "issue:3730", + "right": "issue:4148", + "accept": false, + "reason": "Inpainting blending timestep bug is different from SDXL inpainting config validation." + }, + { + "left": "issue:4303", + "right": "issue:6921", + "accept": false, + "reason": "Loader crash for SDXL in Flax is not the same as SDXL weight-shipping deviation." + }, + { + "left": "issue:4554", + "right": "issue:6906", + "accept": false, + "reason": "OOM in a Kandinsky example and a broken DreamBooth LoRA example are separate issues." + }, + { + "left": "issue:4648", + "right": "issue:7456", + "accept": false, + "reason": "Missing config.json and a missing raise statement are unrelated defects." + }, + { + "left": "issue:5595", + "right": "issue:6731", + "accept": false, + "reason": "Image-size validation and unhandled pipeline-class errors are different failures." + }, + { + "left": "issue:5603", + "right": "issue:6142", + "accept": false, + "reason": "An import error and a NameError in training setup are not the same underlying bug." + }, + { + "left": "issue:6535", + "right": "issue:9485", + "accept": false, + "reason": "Both mention schedulers, but one asks to change the scheduler and the other asks for device placement in scheduler code." + }, + { + "left": "issue:6727", + "right": "issue:9248", + "accept": false, + "reason": "A dtype mismatch in attention is unrelated to pip install / upgrade guidance." + }, + { + "left": "issue:9134", + "right": "issue:9677", + "accept": false, + "reason": "Both mention Flux, but one is a LoRA training issue and the other is a text-encoder precision policy change." + }, + { + "left": "issue:3249", + "right": "issue:5028", + "accept": false, + "reason": "Block-argument mismatch bugs, but on different blocks and different keywords; not one concrete code path." + }, + { + "left": "issue:10223", + "right": "issue:11387", + "accept": false, + "reason": "A LoRA inference-ID question and an AutoencoderDC normalization fix are unrelated." + }, + { + "left": "issue:1543", + "right": "issue:2265", + "accept": false, + "reason": "Textual inversion fp16 revision and ONNXRuntime incompatibility are separate problems." + }, + { + "left": "issue:4122", + "right": "issue:5687", + "accept": false, + "reason": "Both are scheduler timestep bugs, but they affect different schedulers and different failure modes." + }, + { + "left": "issue:11127", + "right": "issue:9996", + "accept": false, + "reason": "Flux single-file loading failure and NF4 transformer loading failure are different model-loading issues." + }, + { + "left": "issue:4331", + "right": "issue:5014", + "accept": false, + "reason": "Different DreamBooth/LoRA training symptoms; not the same bug." + }, + { + "left": "issue:10281", + "right": "issue:11859", + "accept": false, + "reason": "Both are scheduler feature requests, but for different scheduler algorithms." + }, + { + "left": "issue:1695", + "right": "issue:3177", + "accept": false, + "reason": "Both mention DreamBooth with DeepSpeed, but one is multi-GPU training and the other is stage-3 offload breakage." + }, + { + "left": "issue:5026", + "right": "issue:5875", + "accept": false, + "reason": "A text_encoder KeyError and an ONNX conversion prompt error are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 263, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10097", + "issue:10302", + "issue:10412", + "issue:10526", + "issue:11050", + "issue:1205", + "issue:12052", + "issue:1231", + "issue:12412", + "issue:13301", + "issue:1672", + "issue:1896", + "issue:1918", + "issue:2351", + "issue:2444", + "issue:2722", + "issue:2900", + "issue:3009", + "issue:3029", + "issue:3140", + "issue:3221", + "issue:3585", + "issue:3960", + "issue:4107", + "issue:4366", + "issue:4708", + "issue:4839", + "issue:5016", + "issue:5160", + "issue:5166", + "issue:542", + "issue:5515", + "issue:5605", + "issue:5760", + "issue:5968", + "issue:6142", + "issue:6372", + "issue:6439", + "issue:6497", + "issue:660", + "issue:6713", + "issue:7177", + "issue:7548", + "issue:7864", + "issue:835", + "issue:8556", + "issue:8732", + "issue:9527" + ], + "soft_pairs": [ + "issue:1672|issue:2722", + "issue:10302|issue:12052", + "issue:10097|issue:6439", + "issue:1205|issue:542", + "issue:2351|issue:5166", + "issue:3009|issue:3029", + "issue:3221|issue:7548", + "issue:5160|issue:5605", + "issue:10526|issue:8556", + "issue:3960|issue:6713", + "issue:4107|issue:5016", + "issue:4708|issue:7177", + "issue:6372|issue:6497", + "issue:10412|issue:8732", + "issue:1918|issue:5515", + "issue:1896|issue:660", + "issue:2900|issue:3585", + "issue:5760|issue:7864", + "issue:1231|issue:835", + "issue:12412|issue:13301", + "issue:3140|issue:6142", + "issue:4366|issue:5968", + "issue:11050|issue:9527", + "issue:2444|issue:4839" + ], + "prepared_review_unit_hash": "4652251a26a2018d18fe593e067cc0c241c8927cc17b538ed300a94690f04a29", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33483, + "estimated_input_tokens": 8371, + "estimated_eval_tokens": 16998 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set; the items span unrelated docs, model-support, conversion, scheduler, training, and dtype issues. All soft pairs are too broad or only superficially similar, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1672", + "right": "issue:2722", + "accept": false, + "reason": "Both involve conversion, but one is an output-size bug in ckpt-to-diffusers while the other is a ControlNet UNet mismatch after conversion; different failures and fixes." + }, + { + "left": "issue:10302", + "right": "issue:12052", + "accept": false, + "reason": "Both mention FP8/offloading, but one is inference noise and the other is a tensor device-mismatch crash; not the same code-path problem." + }, + { + "left": "issue:10097", + "right": "issue:6439", + "accept": false, + "reason": "Both are docs-related, but one is a missing optimization section and the other is a broken fp16 link; unrelated documentation issues." + }, + { + "left": "issue:1205", + "right": "issue:542", + "accept": false, + "reason": "Both are ONNX export topics, but one is converting full Stable Diffusion models and the other is a UNet export issue; too different to be one bug." + }, + { + "left": "issue:2351", + "right": "issue:5166", + "accept": false, + "reason": "A missing scheduler method versus a missing VQ-VAE scaling factor in a pipeline; different components and fixes." + }, + { + "left": "issue:3009", + "right": "issue:3029", + "accept": false, + "reason": "General ControlNet training advice is not the same as a custom-resolution question." + }, + { + "left": "issue:3221", + "right": "issue:7548", + "accept": false, + "reason": "One is an attention-procs loading API issue; the other is AnimateDiff flicker output quality. Same subsystem, different problems." + }, + { + "left": "issue:5160", + "right": "issue:5605", + "accept": false, + "reason": "Masking support and per-image caption support are distinct training features." + }, + { + "left": "issue:10526", + "right": "issue:8556", + "accept": false, + "reason": "Different models and error modes: optimum.quanto FP8 TypeError versus meta-tensor load failure in SD3 from_single_file." + }, + { + "left": "issue:3960", + "right": "issue:6713", + "accept": false, + "reason": "Both mention HeaderTooLarge, but one is about LoRA and the other about the conversion script; insufficient evidence of the same underlying bug." + }, + { + "left": "issue:4107", + "right": "issue:5016", + "accept": false, + "reason": "LoRA noise/reduced quality is unrelated to SDXL text encoder training not being enabled." + }, + { + "left": "issue:4708", + "right": "issue:7177", + "accept": false, + "reason": "Generic import error versus StableCascade decoder from_pretrained failure; not the same issue." + }, + { + "left": "issue:6372", + "right": "issue:6497", + "accept": false, + "reason": "SVD+ControlNet support and DragNUWA support are separate model/pipeline feature requests." + }, + { + "left": "issue:10412", + "right": "issue:8732", + "accept": false, + "reason": "VRAM overuse in SD3.5 DreamBooth training is unrelated to checkpoint-saving issues in SD1.5 advanced training." + }, + { + "left": "issue:1918", + "right": "issue:5515", + "accept": false, + "reason": "Both scheduler-related, but one is a DDIM example bug and the other is LCM pipeline initialization without a scheduler; different failure points." + }, + { + "left": "issue:1896", + "right": "issue:660", + "accept": false, + "reason": "A CLIP-guided Stable Diffusion size/type error is not the same as an MPS fp16 runtime issue." + }, + { + "left": "issue:2900", + "right": "issue:3585", + "accept": false, + "reason": "Both mention LoRA merging, but one asks how to merge/save LoRA weights and the other reports a failure using a synthesized-LoRA model; not clearly the same bug." + }, + { + "left": "issue:5760", + "right": "issue:7864", + "accept": false, + "reason": "Different research/demo projects in video generation; no shared concrete bug or change." + }, + { + "left": "issue:1231", + "right": "issue:835", + "accept": false, + "reason": "A Flax scheduler issue and a Flax numpy_to_pil failure are unrelated." + }, + { + "left": "issue:12412", + "right": "issue:13301", + "accept": false, + "reason": "Support requests for different model families and pipelines; not duplicates." + }, + { + "left": "issue:3140", + "right": "issue:6142", + "accept": false, + "reason": "StableDiffusionControlNetPipeline import failure versus missing start_training symbol; unrelated errors." + }, + { + "left": "issue:4366", + "right": "issue:5968", + "accept": false, + "reason": "Full U-Net SDXL fine-tuning and video-to-ONNX conversion are unrelated tasks." + }, + { + "left": "issue:11050", + "right": "issue:9527", + "accept": false, + "reason": "Both are dtype-related, but one is a missing cast in a specific training script and the other is a controlnet/mainpipe precision mismatch; not the same concrete fix." + }, + { + "left": "issue:2444", + "right": "issue:4839", + "accept": false, + "reason": "DreamBooth inpaint script failure and lora_scale type error are different problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 264, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10147", + "issue:10446", + "issue:10462", + "issue:1082", + "issue:11561", + "issue:12113", + "issue:12124", + "issue:1251", + "issue:13062", + "issue:1335", + "issue:1949", + "issue:2538", + "issue:2844", + "issue:2947", + "issue:3418", + "issue:3769", + "issue:4297", + "issue:4390", + "issue:4456", + "issue:4584", + "issue:4619", + "issue:4736", + "issue:4767", + "issue:4932", + "issue:4969", + "issue:4977", + "issue:5254", + "issue:5300", + "issue:5334", + "issue:5538", + "issue:5592", + "issue:5628", + "issue:5846", + "issue:5849", + "issue:6159", + "issue:6295", + "issue:6794", + "issue:7079", + "issue:7220", + "issue:8410", + "issue:8803", + "issue:9648", + "issue:9857", + "issue:9858", + "issue:988", + "issue:989", + "issue:9911" + ], + "soft_pairs": [ + "issue:9857|issue:9858", + "issue:4736|issue:5849", + "issue:10147|issue:10462", + "issue:10446|issue:9911", + "issue:12124|issue:13062", + "issue:2947|issue:4456", + "issue:4969|issue:5300", + "issue:5538|issue:8803", + "issue:4932|issue:7220", + "issue:4767|issue:5254", + "issue:4977|issue:5592", + "issue:5628|issue:6159", + "issue:4297|issue:6295", + "issue:11561|issue:8410", + "issue:1251|issue:3769", + "issue:1335|issue:4390", + "issue:2538|issue:3418", + "issue:5334|issue:988", + "issue:12113|issue:9648", + "issue:1949|issue:2844", + "issue:1082|issue:989", + "issue:5846|issue:5849", + "issue:6794|issue:7079", + "issue:4584|issue:4619" + ], + "prepared_review_unit_hash": "7e3763374682d3b746f41d6117337b990a92604555e7a6bdbe8c476b7f992982", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33746, + "estimated_input_tokens": 8437, + "estimated_eval_tokens": 17130 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated feature requests, training-example issues, pipeline/runtime bugs, and hardware/backend compatibility problems across many different code paths.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a safe canonical representative; the items are thematically broad but technically unrelated.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue is suitable because the cluster does not describe one underlying bug or change.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9857", + "right": "issue:9858", + "accept": false, + "reason": "One is a FLUX ControlNet embedding size mismatch, the other is a KeyError: 'train'; different failures." + }, + { + "left": "issue:4736", + "right": "issue:5849", + "accept": false, + "reason": "OOM/cast errors in SDXL LoRA training vs accelerate launch missing args; not the same bug." + }, + { + "left": "issue:10147", + "right": "issue:10462", + "accept": false, + "reason": "Multi-view MV-Adapter support request vs CogVideoX multi-ID customization; different features." + }, + { + "left": "issue:10446", + "right": "issue:9911", + "accept": false, + "reason": "FluxControlNetPipeline model-sharding incompatibility vs multi-ControlNet layer-length mismatch; different code paths." + }, + { + "left": "issue:12124", + "right": "issue:13062", + "accept": false, + "reason": "Dataloader shuffle question for qwen-image training vs unsupported onnxruntime-windowsml backend." + }, + { + "left": "issue:2947", + "right": "issue:4456", + "accept": false, + "reason": "UNet missing load_attn_procs attribute vs ModuleList indexing error in LoRA loading; unrelated loader issues." + }, + { + "left": "issue:4969", + "right": "issue:5300", + "accept": false, + "reason": "CrossAttention import error vs package connecting to internet; different problems." + }, + { + "left": "issue:5538", + "right": "issue:8803", + "accept": false, + "reason": "Upscale pipeline latent usage question vs duplicate assignment in attention processor; unrelated." + }, + { + "left": "issue:4932", + "right": "issue:7220", + "accept": false, + "reason": "Dreambooth output file issue vs unload_ip_adapter error in a ControlNet inpaint pipeline; not the same bug." + }, + { + "left": "issue:4767", + "right": "issue:5254", + "accept": false, + "reason": "Scheduler prediction_type support vs multi-ControlNet support in inpaint pipeline; different changes." + }, + { + "left": "issue:4977", + "right": "issue:5592", + "accept": false, + "reason": "Checkpoint removal in an example script vs UniDiffuser checkpoint loading failure; different artifacts." + }, + { + "left": "issue:5628", + "right": "issue:6159", + "accept": false, + "reason": "LMS scheduler denoising offset vs missing denoising_end support in ControlNetXSPipeline; different scheduler/pipeline issues." + }, + { + "left": "issue:4297", + "right": "issue:6295", + "accept": false, + "reason": "SDXL batched img2img artifacts vs DPM++ 2M SDE Karras artifacts; similar symptom but not the same concrete bug." + }, + { + "left": "issue:11561", + "right": "issue:8410", + "accept": false, + "reason": "FluxFill IP-Adapter support vs native TextDiffuser support; different feature requests." + }, + { + "left": "issue:1251", + "right": "issue:3769", + "accept": false, + "reason": "MPS black image in RePaintPipeline vs panorama quartered output; distinct pipeline bugs." + }, + { + "left": "issue:1335", + "right": "issue:4390", + "accept": false, + "reason": "Missing vae_encoder argument vs missing requirements; unrelated import/setup issues." + }, + { + "left": "issue:2538", + "right": "issue:3418", + "accept": false, + "reason": "Enabling train_text_encoder in one script vs adding text-encoder part in another training script; related theme but not the same change." + }, + { + "left": "issue:5334", + "right": "issue:988", + "accept": false, + "reason": "Accelerate multi-model handling request vs FSDP training examples; broad training infrastructure, but not the same issue." + }, + { + "left": "issue:12113", + "right": "issue:9648", + "accept": false, + "reason": "WanImageToVideo input channel mismatch vs VAE tiling memory overflow; different root causes." + }, + { + "left": "issue:1949", + "right": "issue:2844", + "accept": false, + "reason": "Dreambooth tutorial training errors vs explicit image embeddings in StableUnCLIPImg2ImgPipeline; unrelated." + }, + { + "left": "issue:1082", + "right": "issue:989", + "accept": false, + "reason": "Dreambooth cublasLtMatmul failure vs generic weight dimensionality error; not the same bug." + }, + { + "left": "issue:5846", + "right": "issue:5849", + "accept": false, + "reason": "resume_from_checkpoint error vs accelerate launch missing args; both training-script issues but different failures." + }, + { + "left": "issue:6794", + "right": "issue:7079", + "accept": false, + "reason": "Import error for cast_training_params vs Transformer2DModel patch/norm_type error; unrelated." + }, + { + "left": "issue:4584", + "right": "issue:4619", + "accept": false, + "reason": "resume_from_checkpoint error in SDXL LoRA training vs half/float bias mismatch; different training bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 265, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10933", + "issue:11117", + "issue:11396", + "issue:11661", + "issue:11779", + "issue:11901", + "issue:12358", + "issue:12567", + "issue:12783", + "issue:12895", + "issue:1305", + "issue:13137", + "issue:1546", + "issue:2202", + "issue:2471", + "issue:2695", + "issue:3087", + "issue:3689", + "issue:4030", + "issue:4107", + "issue:4269", + "issue:4332", + "issue:4368", + "issue:4478", + "issue:4534", + "issue:4615", + "issue:5079", + "issue:5245", + "issue:5327", + "issue:5573", + "issue:5607", + "issue:6109", + "issue:6439", + "issue:7138", + "issue:7191", + "issue:7420", + "issue:7432", + "issue:7572", + "issue:8241", + "issue:8827", + "issue:8837", + "issue:8930", + "issue:8989", + "issue:9227", + "issue:9479", + "issue:9531", + "issue:9813" + ], + "soft_pairs": [ + "issue:2695|issue:4534", + "issue:4478|issue:8241", + "issue:1546|issue:4269", + "issue:7432|issue:7572", + "issue:11661|issue:9227", + "issue:8837|issue:8989", + "issue:10933|issue:11396", + "issue:11117|issue:12895", + "issue:11779|issue:9479", + "issue:11901|issue:13137", + "issue:12567|issue:12783", + "issue:1305|issue:3087", + "issue:3689|issue:5079", + "issue:4332|issue:7420", + "issue:4615|issue:5245", + "issue:5245|issue:7191", + "issue:6439|issue:9531", + "issue:8827|issue:9813", + "issue:5573|issue:7138", + "issue:6109|issue:8930", + "issue:2471|issue:5607", + "issue:4030|issue:5327", + "issue:4107|issue:4368", + "issue:10100|issue:12358", + "issue:2202|issue:3087" + ], + "prepared_review_unit_hash": "8494d4b350226c1c3a9a992e67b248c0c9c4c5e40da3fcfdf075a1d52798904d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34075, + "estimated_input_tokens": 8519, + "estimated_eval_tokens": 17294 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is broadly heterogeneous; the soft pairs are mostly superficial topic matches rather than true duplicates. No PRs are present.", + "confidence": 0.22, + "canonical_issue_reason": "If a single issue must be chosen, issue 12783 is the most concrete, reproducible open bug report in the noisy-output/attention area, but the overall set is not a single duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 12783 is the strongest standalone issue: it has a clear symptom, current impact, and enough detail to be actionable.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2695", + "right": "issue:4534", + "accept": false, + "reason": "Both are training/validation crashes, but they affect different pipelines and error paths." + }, + { + "left": "issue:4478", + "right": "issue:8241", + "accept": false, + "reason": "Different models and different failures: dtype mismatch versus abnormal model size." + }, + { + "left": "issue:1546", + "right": "issue:4269", + "accept": false, + "reason": "Both involve input validation, but one is truncation/batched input behavior and the other is a TypeError in check_inputs." + }, + { + "left": "issue:7432", + "right": "issue:7572", + "accept": false, + "reason": "One is about image quality differences; the other is a concrete dtype mismatch error." + }, + { + "left": "issue:11661", + "right": "issue:9227", + "accept": false, + "reason": "Different training scripts and different bugs; only a broad LoRA/training theme overlaps." + }, + { + "left": "issue:8837", + "right": "issue:8989", + "accept": false, + "reason": "Both concern performance/attention-related behavior, but they are distinct code paths and symptoms." + }, + { + "left": "issue:10933", + "right": "issue:11396", + "accept": false, + "reason": "IP-Adapter loading error versus a LoRA conversion question; not the same underlying bug." + }, + { + "left": "issue:11117", + "right": "issue:12895", + "accept": false, + "reason": "Different Stable Diffusion 3 failures: dtype mismatch versus NoneType batching crash." + }, + { + "left": "issue:11779", + "right": "issue:9479", + "accept": false, + "reason": "Both involve masking, but one is Wan VACE mask padding and the other is inpaint overwriting unmasked regions." + }, + { + "left": "issue:11901", + "right": "issue:13137", + "accept": false, + "reason": "Model-specific LoRA loading failures with different models and error modes." + }, + { + "left": "issue:12567", + "right": "issue:12783", + "accept": false, + "reason": "Both mention noisy output from attention optimizations, but the implementations and failure details differ." + }, + { + "left": "issue:1305", + "right": "issue:3087", + "accept": false, + "reason": "Direct inversion support request and LyCORIS support are unrelated feature requests." + }, + { + "left": "issue:3689", + "right": "issue:5079", + "accept": false, + "reason": "Related to LoRA APIs, but one is unloading weights and the other is loading LyCORIS." + }, + { + "left": "issue:4332", + "right": "issue:7420", + "accept": false, + "reason": "xformers usage in a specific pipeline versus torch.compile enablement are unrelated." + }, + { + "left": "issue:4615", + "right": "issue:5245", + "accept": false, + "reason": "ImportError in diffusers versus a pipenv import-stage freeze; not the same issue." + }, + { + "left": "issue:5245", + "right": "issue:7191", + "accept": false, + "reason": "Generic import problems, but no shared code path or concrete failure mode." + }, + { + "left": "issue:6439", + "right": "issue:9531", + "accept": false, + "reason": "Docs link 404 and SDXL sigma tuning are unrelated." + }, + { + "left": "issue:8827", + "right": "issue:9813", + "accept": false, + "reason": "Both are IP-Adapter-related, but one is a FaceID loading error and the other is batch feature mixing." + }, + { + "left": "issue:5573", + "right": "issue:7138", + "accept": false, + "reason": "ControlNet initialization slowness versus SDXL UNet structure mismatch are different problems." + }, + { + "left": "issue:6109", + "right": "issue:8930", + "accept": false, + "reason": "Both involve ControlNet-like conditioning, but they target different pipelines and bugs." + }, + { + "left": "issue:2471", + "right": "issue:5607", + "accept": false, + "reason": "Both mention shape-related failures, but the underlying pipelines and tensor mismatch causes are different." + }, + { + "left": "issue:4030", + "right": "issue:5327", + "accept": false, + "reason": "Embedding/textual-inversion related, but one is a runtime embedding error and the other is an SDXL import/load issue." + }, + { + "left": "issue:4107", + "right": "issue:4368", + "accept": false, + "reason": "Noise/quality degradation and OOM during LoRA training are not the same bug." + }, + { + "left": "issue:10100", + "right": "issue:12358", + "accept": false, + "reason": "Both are feature requests for new pipelines, but they are for different models and changes." + }, + { + "left": "issue:2202", + "right": "issue:3087", + "accept": false, + "reason": "OpenVINO pipeline support and LyCORIS support are unrelated requests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 266, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10029", + "issue:10398", + "issue:10640", + "issue:10782", + "issue:11148", + "issue:11223", + "issue:11581", + "issue:11592", + "issue:11674", + "issue:11885", + "issue:12149", + "issue:12501", + "issue:1957", + "issue:2266", + "issue:2507", + "issue:2775", + "issue:2984", + "issue:3026", + "issue:3585", + "issue:3937", + "issue:4478", + "issue:4551", + "issue:4605", + "issue:4618", + "issue:5046", + "issue:5110", + "issue:5274", + "issue:5594", + "issue:5678", + "issue:5714", + "issue:5829", + "issue:6335", + "issue:6520", + "issue:6529", + "issue:6567", + "issue:6598", + "issue:6794", + "issue:7168", + "issue:7185", + "issue:7575", + "issue:7640", + "issue:7836", + "issue:9032", + "issue:9171", + "issue:9858", + "issue:9927" + ], + "soft_pairs": [ + "issue:2984|issue:5046", + "issue:9858|issue:9927", + "issue:10029|issue:10398", + "issue:2266|issue:2507", + "issue:4618|issue:5110", + "issue:11223|issue:11674", + "issue:10010|issue:7836", + "issue:1957|issue:5274", + "issue:5594|issue:6529", + "issue:5714|issue:7185", + "issue:3026|issue:3585", + "issue:7168|issue:7640", + "issue:10640|issue:6794", + "issue:3937|issue:6567", + "issue:4478|issue:4605", + "issue:2775|issue:6520", + "issue:4551|issue:5829", + "issue:11148|issue:12501", + "issue:6335|issue:6598", + "issue:5678|issue:7575", + "issue:10010|issue:9032", + "issue:10782|issue:12149", + "issue:11581|issue:11885", + "issue:11592|issue:9171" + ], + "prepared_review_unit_hash": "1b35112be0d07dfa1a98171a776bdca0a35c0e5fbb44dd84777b95834a99c4e3", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33862, + "estimated_input_tokens": 8466, + "estimated_eval_tokens": 17188 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All proposed soft links are false positives; the items are a heterogeneous mix of unrelated bugs, features, and regressions rather than one duplicate cluster.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical duplicate: the issues span unrelated code paths and product areas.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11581 is the most concrete standalone bug report in the set, with a specific runtime error and likely actionable fix.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2984", + "right": "issue:5046", + "accept": false, + "reason": "Different topics: a video diffusion paper/proposal vs a drag-style manipulation feature request." + }, + { + "left": "issue:9858", + "right": "issue:9927", + "accept": false, + "reason": "Both are training-related errors, but the failure modes differ completely (`KeyError: 'train'` vs `HeaderTooLarge`)." + }, + { + "left": "issue:10029", + "right": "issue:10398", + "accept": false, + "reason": "Both involve PEFT/adapter loading, but one is a missing `base_layer` attribute and the other is missing target modules in the base model." + }, + { + "left": "issue:2266", + "right": "issue:2507", + "accept": false, + "reason": "Both mention performance, but one is sequential CPU offload slowdown and the other is safetensors GPU loading being slower than CPU." + }, + { + "left": "issue:4618", + "right": "issue:5110", + "accept": false, + "reason": "Completely unrelated: SDXL ControlNet usage vs Hugging Face repo-id validation failure." + }, + { + "left": "issue:11223", + "right": "issue:11674", + "accept": false, + "reason": "Different requests: an OOM bug with LoRAs under layer-wise upcasting vs a feature request for Wan VACE inputs." + }, + { + "left": "issue:10010", + "right": "issue:7836", + "accept": false, + "reason": "Unrelated bugs: a FLUX Fill model download failure vs CLIP training overfitting." + }, + { + "left": "issue:1957", + "right": "issue:5274", + "accept": false, + "reason": "Different training problems: LR scheduler initialization vs mixed-precision dtype mismatch." + }, + { + "left": "issue:5594", + "right": "issue:6529", + "accept": false, + "reason": "Different concerns: SDXL LoRA memory-saving discussion vs Fooocus inpaint support." + }, + { + "left": "issue:5714", + "right": "issue:7185", + "accept": false, + "reason": "Unrelated issues: PeftModel handling in `maybe_raise_or_warn` vs PyTorch 2.2 compatibility." + }, + { + "left": "issue:3026", + "right": "issue:3585", + "accept": false, + "reason": "Different code paths and symptoms: instruct-pix2pix launch failure vs LoRA synthesis into SD1.5 failing." + }, + { + "left": "issue:7168", + "right": "issue:7640", + "accept": false, + "reason": "Both are image-generation issues, but IP-Adapter embedding shape handling is not the same as `from_single_file` dimension errors." + }, + { + "left": "issue:10640", + "right": "issue:6794", + "accept": false, + "reason": "Different subsystems: FluxInpaint/DataParallel GPU error vs import failure for `cast_training_params`." + }, + { + "left": "issue:3937", + "right": "issue:6567", + "accept": false, + "reason": "Unrelated: ZeroscopeXL model loading behavior vs an SD 1.5 import warning about `text_config_dict`." + }, + { + "left": "issue:4478", + "right": "issue:4605", + "accept": false, + "reason": "Both are SD model errors, but one is a dtype/bias mismatch in SDXL loading and the other is an inpaint pipeline error." + }, + { + "left": "issue:2775", + "right": "issue:6520", + "accept": false, + "reason": "Different failures: a Torch Dynamo optimization error vs blank validation outputs during adapter training." + }, + { + "left": "issue:4551", + "right": "issue:5829", + "accept": false, + "reason": "Different training crashes: SDXL DreamBooth failure vs missing `unet_time_cond_proj_dim` in LCM scripts." + }, + { + "left": "issue:11148", + "right": "issue:12501", + "accept": false, + "reason": "Both are bugs, but one is PEFT adapter state tracking and the other is a VAE decoder graph-break issue." + }, + { + "left": "issue:6335", + "right": "issue:6598", + "accept": false, + "reason": "Different img2img problems: output-size control vs differing results for identical parameters." + }, + { + "left": "issue:5678", + "right": "issue:7575", + "accept": false, + "reason": "Distinct model/math issues: higher-order derivative behavior in attention vs a shape error in Transformer2DModel with adanorm." + }, + { + "left": "issue:10010", + "right": "issue:9032", + "accept": false, + "reason": "Completely unrelated: FLUX Fill example download failure vs a Flax/TPU setup question." + }, + { + "left": "issue:10782", + "right": "issue:12149", + "accept": false, + "reason": "Different scopes: a Lumina Image 2.0 minor issue vs package/source version mismatch." + }, + { + "left": "issue:11581", + "right": "issue:11885", + "accept": false, + "reason": "Both concern FLUX adapters, but one is an IP-Adapter constructor dtype error and the other is a LoRA loading regression." + }, + { + "left": "issue:11592", + "right": "issue:9171", + "accept": false, + "reason": "Unrelated regressions: FluxLoraLoaderMixin adapter deletion vs SD 1.5 `v_prediction` handling in `from_single_file`." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 267, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10940", + "issue:11127", + "issue:1214", + "issue:1530", + "issue:2532", + "issue:2696", + "issue:3089", + "issue:3171", + "issue:4000", + "issue:4080", + "issue:4139", + "issue:4403", + "issue:4414", + "issue:447", + "issue:4542", + "issue:460", + "issue:4605", + "issue:4675", + "issue:4753", + "issue:5002", + "issue:5054", + "issue:5152", + "issue:5372", + "issue:5824", + "issue:5981", + "issue:5989", + "issue:6159", + "issue:6183", + "issue:6258", + "issue:6790", + "issue:7009", + "issue:7041", + "issue:7210", + "issue:7355", + "issue:7592", + "issue:8236", + "issue:8373", + "issue:8644", + "issue:8792", + "issue:8900", + "issue:9023", + "issue:9209", + "issue:9279", + "issue:9292", + "issue:962", + "issue:9856", + "issue:9953" + ], + "soft_pairs": [ + "issue:2532|issue:5824", + "issue:2696|issue:460", + "issue:3089|issue:5989", + "issue:3171|issue:447", + "issue:4000|issue:8373", + "issue:4080|issue:4675", + "issue:5054|issue:5372", + "issue:6790|issue:7592", + "issue:7041|issue:8792", + "issue:11127|issue:9856", + "issue:8644|issue:9953", + "issue:4403|issue:6159", + "issue:6183|issue:8373", + "issue:10940|issue:7210", + "issue:1214|issue:4139", + "issue:4542|issue:7355", + "issue:4753|issue:8236", + "issue:9279|issue:9292", + "issue:5152|issue:5981", + "issue:6258|issue:7009", + "issue:8900|issue:9209", + "issue:1530|issue:5002", + "issue:4414|issue:9023", + "issue:4605|issue:962" + ], + "prepared_review_unit_hash": "610c32b0b96ca39c913a7142487a73d6d99df01272d740ef5a84f1c7454fe4ed", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33464, + "estimated_input_tokens": 8366, + "estimated_eval_tokens": 16988 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the soft pairs mostly share only broad diffusion-model vocabulary or the same subsystem, but they describe different bugs, features, or backend/platform failures. I would not merge any of the proposed pairs.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2532", + "right": "issue:5824", + "accept": false, + "reason": "Both involve Stable Diffusion pipeline internals, but one is a panorama get_views bug and the other is an SDXL add_time_ids default issue; different code paths and symptoms." + }, + { + "left": "issue:2696", + "right": "issue:460", + "accept": false, + "reason": "CUDA installation/runtime error vs ONNX Runtime backend dependency selection; these are separate environment/backend problems." + }, + { + "left": "issue:3089", + "right": "issue:5989", + "accept": false, + "reason": "DDPM scheduler logic bug and safetensors.torch bug are unrelated components with different failure modes." + }, + { + "left": "issue:3171", + "right": "issue:447", + "accept": false, + "reason": "LPW prompt support for ControlNet vs prompt truncation detection/return values; different feature requests." + }, + { + "left": "issue:4000", + "right": "issue:8373", + "accept": false, + "reason": "Model ckpt conversion problem vs an incorrect error message; not the same underlying bug." + }, + { + "left": "issue:4080", + "right": "issue:4675", + "accept": false, + "reason": "SDXL inpainting pipeline request vs ControlNet Reference Only support; separate features." + }, + { + "left": "issue:5054", + "right": "issue:5372", + "accept": false, + "reason": "Missing final_offload_hook attribute and safety_checker usage are different StableDiffusionXL API issues." + }, + { + "left": "issue:6790", + "right": "issue:7592", + "accept": false, + "reason": "Both are state_dict loading problems, but one is an LCM example key mismatch and the other is DoRA key loading; not the same concrete bug." + }, + { + "left": "issue:7041", + "right": "issue:8792", + "accept": false, + "reason": "Segfault at num_inference_steps > 1 is a runtime crash; image-quality degradation during migration is a separate quality regression." + }, + { + "left": "issue:11127", + "right": "issue:9856", + "accept": false, + "reason": "Single-file FLUX model loading failure vs distributed port collision; unrelated errors." + }, + { + "left": "issue:8644", + "right": "issue:9953", + "accept": false, + "reason": "Both mention device/offload behavior, but one is T5 encoder sequential offload meta-tensor failure and the other is moving a quantized component to CUDA; different code paths." + }, + { + "left": "issue:4403", + "right": "issue:6159", + "accept": false, + "reason": "Half-precision LayerNorm failure vs missing denoising_end argument support; unrelated." + }, + { + "left": "issue:6183", + "right": "issue:8373", + "accept": false, + "reason": "Incorrect tensor shape from from_single_file is not the same as an incorrect error message." + }, + { + "left": "issue:10940", + "right": "issue:7210", + "accept": false, + "reason": "Missing triton.ops import and missing named_parameters under accelerate are different integration issues." + }, + { + "left": "issue:1214", + "right": "issue:4139", + "accept": false, + "reason": "Both concern precision, but CLIPTextModel half-precision support and custom attention processor casting are not the same bug." + }, + { + "left": "issue:4542", + "right": "issue:7355", + "accept": false, + "reason": "Cannot generate more than one image in SDXL img2img vs Stable Cascade arbitrary-resolution bfloat16 errors; unrelated." + }, + { + "left": "issue:4753", + "right": "issue:8236", + "accept": false, + "reason": "train_dreambooth_lora.py problem vs wrong multi-GPU LR scheduler step count; different training bugs." + }, + { + "left": "issue:9279", + "right": "issue:9292", + "accept": false, + "reason": "Dynamic LoRA loading with torch.compile and FLUX LoRA key incompatibility are separate issues." + }, + { + "left": "issue:5152", + "right": "issue:5981", + "accept": false, + "reason": "W\u00fcrstchen alpha bug vs DPMSolver scheduler bug; no concrete overlap." + }, + { + "left": "issue:6258", + "right": "issue:7009", + "accept": false, + "reason": "Forward chunking degrading quality in Stable Video Diffusion is not the same as poor results from a specific text-to-video checkpoint." + }, + { + "left": "issue:8900", + "right": "issue:9209", + "accept": false, + "reason": "Loading SDXL refiner from_single_file vs failing to connect to hf-mirror.com are unrelated." + }, + { + "left": "issue:1530", + "right": "issue:5002", + "accept": false, + "reason": "Missing learned_classifier_free_sampling_embeddings and missing use_karras_sigmas are different API gaps in different schedulers/pipelines." + }, + { + "left": "issue:4414", + "right": "issue:9023", + "accept": false, + "reason": "Consistency Model sample request vs generic sample code failure; too broad to be the same bug." + }, + { + "left": "issue:4605", + "right": "issue:962", + "accept": false, + "reason": "Long prompt weighting with an inpaint model error vs inpaint pipelines only supporting 512x512 are different inpaint limitations." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 268, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10066", + "issue:10198", + "issue:10650", + "issue:11169", + "issue:11489", + "issue:11661", + "issue:12126", + "issue:12565", + "issue:12645", + "issue:12995", + "issue:1444", + "issue:1709", + "issue:1909", + "issue:1919", + "issue:1922", + "issue:2098", + "issue:2737", + "issue:2838", + "issue:305", + "issue:3057", + "issue:3624", + "issue:4094", + "issue:4186", + "issue:4222", + "issue:4249", + "issue:459", + "issue:4768", + "issue:5110", + "issue:5120", + "issue:5485", + "issue:5605", + "issue:5689", + "issue:5821", + "issue:5989", + "issue:6042", + "issue:6055", + "issue:6313", + "issue:6513", + "issue:6616", + "issue:6686", + "issue:7000", + "issue:7490", + "issue:7798", + "issue:8552", + "issue:8790", + "issue:9331", + "issue:9516", + "issue:9974" + ], + "soft_pairs": [ + "issue:5821|issue:7000", + "issue:2737|issue:6513", + "issue:4222|issue:5605", + "issue:5989|issue:7798", + "issue:6042|issue:8790", + "issue:11489|issue:11661", + "issue:10198|issue:9516", + "issue:1909|issue:2098", + "issue:4186|issue:6686", + "issue:6313|issue:6616", + "issue:1709|issue:6055", + "issue:3624|issue:5110", + "issue:1919|issue:5120", + "issue:10650|issue:12995", + "issue:11169|issue:12645", + "issue:12126|issue:12565", + "issue:1444|issue:305", + "issue:1922|issue:5689", + "issue:2838|issue:459", + "issue:3057|issue:4249", + "issue:4094|issue:4768", + "issue:5485|issue:8552", + "issue:7490|issue:9331", + "issue:10066|issue:9974" + ], + "prepared_review_unit_hash": "8d6779f32c05601def4a185e5d303425b8903681d6a5000663b3d6c0dc473079", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33910, + "estimated_input_tokens": 8478, + "estimated_eval_tokens": 17212 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is largely heterogeneous: it mixes unrelated bugs, feature requests, and support questions across different pipelines and versions. None of the soft-edge pairs look like true duplicates of the same underlying bug/change.", + "confidence": 0.96, + "canonical_issue_reason": "No issue cleanly represents the whole set because the cluster is not cohesive. If one must be chosen, issue 11489 is the most concrete, actionable bug report (a specific Flux LoRA training failure) rather than a broad support question or a generic error report.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11489 is the best single triage candidate: it describes a specific, reproducible training error with a clear context. The rest are too mixed in scope to serve as a strong canonical representative.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5821", + "right": "issue:7000", + "accept": false, + "reason": "Different problems: CUDA/backward training failure vs a PyTorch deprecation warning." + }, + { + "left": "issue:2737", + "right": "issue:6513", + "accept": false, + "reason": "Both are feature requests, but for different capabilities: CLIP guidance vs T2IAdapter model support." + }, + { + "left": "issue:4222", + "right": "issue:5605", + "accept": false, + "reason": "Low-resolution dataset training question is unrelated to per-image caption support." + }, + { + "left": "issue:5989", + "right": "issue:7798", + "accept": false, + "reason": "Unrelated errors in different components: safetensors serialization vs PixArtAlphaPipeline bug." + }, + { + "left": "issue:6042", + "right": "issue:8790", + "accept": false, + "reason": "Different topics: SDXL-turbo distillation example vs SDXL TensorRT/IP-Adapter integration." + }, + { + "left": "issue:11489", + "right": "issue:11661", + "accept": false, + "reason": "Both mention Flux training, but the concrete failures differ and are not the same code-path bug." + }, + { + "left": "issue:10198", + "right": "issue:9516", + "accept": false, + "reason": "Weight mismatch is not the same as missing joint_attention_kwargs propagation." + }, + { + "left": "issue:1909", + "right": "issue:2098", + "accept": false, + "reason": "Generator API mismatch is unrelated to the randn argument error." + }, + { + "left": "issue:4186", + "right": "issue:6686", + "accept": false, + "reason": "Inpaint/refiner malfunction is unrelated to Dynamo save/load performance regression." + }, + { + "left": "issue:6313", + "right": "issue:6616", + "accept": false, + "reason": "Different requested features: null-text inversion pipeline vs classifier-guidance implementation." + }, + { + "left": "issue:1709", + "right": "issue:6055", + "accept": false, + "reason": "ONNX Tensor.astype error is unrelated to an accelerate issue." + }, + { + "left": "issue:3624", + "right": "issue:5110", + "accept": false, + "reason": "AttributeError from getattr is different from HF repo-id validation failure." + }, + { + "left": "issue:1919", + "right": "issue:5120", + "accept": false, + "reason": "Missing dependency in dreambooth docs is not the same as a prior-preservation training bug." + }, + { + "left": "issue:10650", + "right": "issue:12995", + "accept": false, + "reason": "Lumina shape invalid error is unrelated to Cosmos-Predict2.5 pipeline behavior." + }, + { + "left": "issue:11169", + "right": "issue:12645", + "accept": false, + "reason": "Binary incompatibility import failure is unrelated to Qwen-Image speed concerns." + }, + { + "left": "issue:12126", + "right": "issue:12565", + "accept": false, + "reason": "Both are Wan issues, but different versions and different failures; not the same concrete bug." + }, + { + "left": "issue:1444", + "right": "issue:305", + "accept": false, + "reason": "Missing StableDiffusionSafetyChecker import is not the same as the generic diffusers import error." + }, + { + "left": "issue:1922", + "right": "issue:5689", + "accept": false, + "reason": "Undefined notebook helper is unrelated to DPM++ residual noise in SDXL." + }, + { + "left": "issue:2838", + "right": "issue:459", + "accept": false, + "reason": "ControlNet pipeline behavior request is unrelated to incremental diffusion." + }, + { + "left": "issue:3057", + "right": "issue:4249", + "accept": false, + "reason": "Conversion-script import failure is unrelated to check_inputs prompt_embeds validation." + }, + { + "left": "issue:4094", + "right": "issue:4768", + "accept": false, + "reason": "Adding a T2I adapter training script is not the same as saving multiple adapters in one checkpoint." + }, + { + "left": "issue:5485", + "right": "issue:8552", + "accept": false, + "reason": "Batch inference slowness is unrelated to a TextToVideoZeroPipeline segfault." + }, + { + "left": "issue:7490", + "right": "issue:9331", + "accept": false, + "reason": "Generic backward problem is unrelated to FlowMatchEulerDiscreteScheduler index_for_timestep." + }, + { + "left": "issue:10066", + "right": "issue:9974", + "accept": false, + "reason": "Different bugs: attention unpacking error vs missing expected parameters in controlnet_sd3." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 269, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10111", + "issue:10147", + "issue:10382", + "issue:1055", + "issue:1083", + "issue:10952", + "issue:11160", + "issue:1158", + "issue:11659", + "issue:11732", + "issue:1205", + "issue:12097", + "issue:13162", + "issue:1445", + "issue:1572", + "issue:1574", + "issue:2288", + "issue:2480", + "issue:2799", + "issue:3045", + "issue:3065", + "issue:3391", + "issue:3717", + "issue:4528", + "issue:4557", + "issue:4585", + "issue:4972", + "issue:4977", + "issue:5034", + "issue:5037", + "issue:5046", + "issue:5395", + "issue:5432", + "issue:5442", + "issue:5763", + "issue:5875", + "issue:6548", + "issue:6839", + "issue:7133", + "issue:7147", + "issue:720", + "issue:7243", + "issue:835", + "issue:8801", + "issue:9090", + "issue:9387", + "issue:9686", + "issue:9846" + ], + "soft_pairs": [ + "issue:4585|issue:7147", + "issue:4977|issue:5034", + "issue:1055|issue:4557", + "issue:4528|issue:5763", + "issue:11659|issue:13162", + "issue:5046|issue:7243", + "issue:5395|issue:9090", + "issue:1574|issue:3717", + "issue:11160|issue:12097", + "issue:10111|issue:9686", + "issue:10382|issue:7133", + "issue:1083|issue:835", + "issue:10952|issue:11732", + "issue:1158|issue:2480", + "issue:1205|issue:3045", + "issue:1445|issue:5875", + "issue:2288|issue:3065", + "issue:3391|issue:6548", + "issue:5442|issue:9387", + "issue:6839|issue:9846", + "issue:4972|issue:720", + "issue:1572|issue:2799", + "issue:5037|issue:5432", + "issue:10147|issue:8801" + ], + "prepared_review_unit_hash": "3e1e7b4b33cd46eb79c4d6c27810457b17a03f3fd8329c7675f5359d40e0c3d8", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33871, + "estimated_input_tokens": 8468, + "estimated_eval_tokens": 17192 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are unrelated feature requests or distinct bug reports, and none of the soft pairs look like true duplicates of the same concrete issue. The only loose theme is LoRA/loading-related reports, but even those point to different failures and code paths.", + "confidence": 0.92, + "canonical_issue_reason": "issue:11659 is the broadest active issue in the set, but the cluster is not a genuine duplicate group; it only loosely matches the recurring LoRA-loading theme.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11659 is the most generic and umbrella-like issue title in the set, so it is the least bad representative if one must be chosen.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4585", + "right": "issue:7147", + "accept": false, + "reason": "Different problems: SDXL MultiControlNet documentation/usage vs missing import for StableCascadeDecoderPipeline." + }, + { + "left": "issue:4977", + "right": "issue:5034", + "accept": false, + "reason": "Both mention checkpoints, but one is about distributed checkpoint directories and the other about progress bar display during loading." + }, + { + "left": "issue:1055", + "right": "issue:4557", + "accept": false, + "reason": "Different failures: missing LatentDiffusionPipeline vs AttributeError on `super().__getattr__`." + }, + { + "left": "issue:4528", + "right": "issue:5763", + "accept": false, + "reason": "LoRA quality/output issue vs `save_pretrained` not writing safetensors; different code paths." + }, + { + "left": "issue:11659", + "right": "issue:13162", + "accept": false, + "reason": "Both are LoRA-related, but one is a generic load failure and the other is a model-specific support request for Flux.2-klein-4b." + }, + { + "left": "issue:5046", + "right": "issue:7243", + "accept": false, + "reason": "Different pipeline proposals: DragonDiffusion vs ResAdapter; not the same bug or change." + }, + { + "left": "issue:5395", + "right": "issue:9090", + "accept": false, + "reason": "LoRA combination/PEFT bug vs scheduler timestep repetition bug; unrelated failures." + }, + { + "left": "issue:1574", + "right": "issue:3717", + "accept": false, + "reason": "Different resource issues in different contexts: bare-metal 8GB Dreambooth training vs multi-GPU OOM on 3090Ti." + }, + { + "left": "issue:11160", + "right": "issue:12097", + "accept": false, + "reason": "Different Wan versions and different symptoms: inference inconsistency vs VRAM OOM." + }, + { + "left": "issue:10111", + "right": "issue:9686", + "accept": false, + "reason": "FLUX pipeline bugs vs SD3 ControlNet pooled projection bug; different subsystems and symptoms." + }, + { + "left": "issue:10382", + "right": "issue:7133", + "accept": false, + "reason": "VAE encoder return-type AttributeError vs SDXL Dreambooth LoRA text-encoder size mismatch; not the same bug." + }, + { + "left": "issue:1083", + "right": "issue:835", + "accept": false, + "reason": "FP16 ONNX output correctness vs flax pipeline `numpy_to_pil` failure; unrelated." + }, + { + "left": "issue:10952", + "right": "issue:11732", + "accept": false, + "reason": "Both are LoRA load errors, but the adapter-name mismatch and tuple-unpack error come from different failure modes." + }, + { + "left": "issue:1158", + "right": "issue:2480", + "accept": false, + "reason": "Test interference in `test_config.py` vs Dreambooth checkpoint folder layout; unrelated." + }, + { + "left": "issue:1205", + "right": "issue:3045", + "accept": false, + "reason": "Different conversion tasks: SD model to ONNX vs LoRA safetensors to diffusers." + }, + { + "left": "issue:1445", + "right": "issue:5875", + "accept": false, + "reason": "Both are ONNX-related, but they describe different pipeline errors and different conversion scenarios." + }, + { + "left": "issue:2288", + "right": "issue:3065", + "accept": false, + "reason": "Training-script implementation request vs adding a LoRA tuning option; separate feature requests." + }, + { + "left": "issue:3391", + "right": "issue:6548", + "accept": false, + "reason": "Different image-processing bugs: overexposure in DeepFloyd SR vs RGBA being mistaken for latents in img2img." + }, + { + "left": "issue:5442", + "right": "issue:9387", + "accept": false, + "reason": "Distinct pipeline features: ScaleCrafter vs MimicMotion." + }, + { + "left": "issue:6839", + "right": "issue:9846", + "accept": false, + "reason": "Different missing feature requests: from_single_file for StableVideoDiffusion vs FluxControlNetModel." + }, + { + "left": "issue:4972", + "right": "issue:720", + "accept": false, + "reason": "Example-code failure in Wuerstchen vs traced UNet dtype mismatch; different issues." + }, + { + "left": "issue:1572", + "right": "issue:2799", + "accept": false, + "reason": "One is a pipeline capability bug for img2img/inpainting, the other is a checkpoint-vs-pipeline mismatch guard." + }, + { + "left": "issue:5037", + "right": "issue:5432", + "accept": false, + "reason": "Second validation image argument error vs blank validation images; related area but different concrete bugs." + }, + { + "left": "issue:10147", + "right": "issue:8801", + "accept": false, + "reason": "Different model support requests: MV-Adapter multi-view generation vs Kolors support." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 270, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10107", + "issue:10749", + "issue:10950", + "issue:10962", + "issue:11010", + "issue:11642", + "issue:12574", + "issue:1742", + "issue:1823", + "issue:2221", + "issue:2228", + "issue:228", + "issue:2392", + "issue:2646", + "issue:3284", + "issue:3622", + "issue:3624", + "issue:3698", + "issue:3765", + "issue:3805", + "issue:4125", + "issue:4213", + "issue:4436", + "issue:4680", + "issue:4752", + "issue:5044", + "issue:5112", + "issue:5274", + "issue:5510", + "issue:579", + "issue:5989", + "issue:6087", + "issue:6094", + "issue:6142", + "issue:6164", + "issue:6176", + "issue:6517", + "issue:7535", + "issue:785", + "issue:8469", + "issue:8557", + "issue:8752", + "issue:8930", + "issue:8939", + "issue:9034", + "issue:9309", + "issue:9979" + ], + "soft_pairs": [ + "issue:6094|issue:6176", + "issue:6517|issue:9034", + "issue:5112|issue:8930", + "issue:3765|issue:6087", + "issue:10749|issue:11010", + "issue:1823|issue:3622", + "issue:2228|issue:785", + "issue:228|issue:3622", + "issue:3624|issue:6142", + "issue:3698|issue:4680", + "issue:4436|issue:5989", + "issue:4752|issue:7535", + "issue:10962|issue:12574", + "issue:11642|issue:8939", + "issue:3284|issue:3805", + "issue:5510|issue:6164", + "issue:1742|issue:2392", + "issue:2221|issue:579", + "issue:2646|issue:5044", + "issue:4125|issue:5274", + "issue:4213|issue:8752", + "issue:10107|issue:8557", + "issue:10950|issue:8469", + "issue:9309|issue:9979" + ], + "prepared_review_unit_hash": "b387869dd2c746764aa5aa6ccbf505e7290bc0b6bb94a35f1a5ded313dcd6446", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33676, + "estimated_input_tokens": 8419, + "estimated_eval_tokens": 17094 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issue set spanning docs, training, dependency, and many distinct pipeline bugs. The only plausible duplicate pair is the two Flux ControlNet reports; the rest should stay separate.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 9979 is the most concrete and actionable bug report in the only plausible duplicate subgroup, with an explicit channels error and clearer repro value.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9979 is the best anchor for deduping within this cluster because it names the concrete Flux ControlNet failure mode, while the broader set has no single shared root cause.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6094", + "right": "issue:6176", + "accept": false, + "reason": "Different pipelines and failures: TextToVideoZero regression after a version update vs LoRA weight loading after DreamBooth training." + }, + { + "left": "issue:6517", + "right": "issue:9034", + "accept": false, + "reason": "One is an import traceback for an SDXL ControlNet module; the other is an optional dependency requirement for AutoPipelineForText2Image." + }, + { + "left": "issue:5112", + "right": "issue:8930", + "accept": false, + "reason": "Different bug shapes: a runtime tensor shape/type error vs a control-image adherence quality problem." + }, + { + "left": "issue:3765", + "right": "issue:6087", + "accept": false, + "reason": "Training image stacking effects in ControlNet is unrelated to saving/loading LoRA config and wrong lora_alpha." + }, + { + "left": "issue:10749", + "right": "issue:11010", + "accept": false, + "reason": "Separate feature requests for different models and capabilities: GGUF support in Lumina2 vs architecture support for Chroma/Flux." + }, + { + "left": "issue:1823", + "right": "issue:3622", + "accept": false, + "reason": "A comments-swapped docs/content issue is unrelated to a tutorial import error." + }, + { + "left": "issue:2228", + "right": "issue:785", + "accept": false, + "reason": "Loading the pipeline onto GPU and PNDMScheduler runtime cost are different problems." + }, + { + "left": "issue:228", + "right": "issue:3622", + "accept": false, + "reason": "Broken release link and tutorial import failure are unrelated." + }, + { + "left": "issue:3624", + "right": "issue:6142", + "accept": false, + "reason": "Different exceptions in different code paths: getattr TypeError vs missing start_training NameError." + }, + { + "left": "issue:3698", + "right": "issue:4680", + "accept": false, + "reason": "Model conversion request vs fine-tuning SDXL inpainting discussion; not the same bug or change." + }, + { + "left": "issue:4436", + "right": "issue:5989", + "accept": false, + "reason": "Converting CIVITAI weights to diffusers is unrelated to a safetensors.torch bug report." + }, + { + "left": "issue:4752", + "right": "issue:7535", + "accept": false, + "reason": "Runtime half/float mismatch in SDXL is not the same as a type-hint incompatibility issue." + }, + { + "left": "issue:10962", + "right": "issue:12574", + "accept": false, + "reason": "Different model pipelines and failure modes: prompt_embed shape handling in CogView4 vs missing control-arg handling in WanVideoToVideo." + }, + { + "left": "issue:11642", + "right": "issue:8939", + "accept": false, + "reason": "Prompt scale not taking effect in FluxPriorRedux is unrelated to tokenizer loading/dependency resolution." + }, + { + "left": "issue:3284", + "right": "issue:3805", + "accept": false, + "reason": "An accelerate training-script failure is not the same as a Colab CalledProcessError." + }, + { + "left": "issue:5510", + "right": "issue:6164", + "accept": false, + "reason": "These are different pipeline regressions: SDXL instruct-pix2pix dtype mismatch vs latent output handling in StableDiffusionPipeline." + }, + { + "left": "issue:1742", + "right": "issue:2392", + "accept": false, + "reason": "Inference speed variation from prompt list content is unrelated to prompt_embeds validation raising NoneType errors." + }, + { + "left": "issue:2221", + "right": "issue:579", + "accept": false, + "reason": "ColossalAI integration errors and missing config.json in textual inversion are distinct issues." + }, + { + "left": "issue:2646", + "right": "issue:5044", + "accept": false, + "reason": "Tiled VAE IndexError is unrelated to Kandinsky 2.2 loading failure." + }, + { + "left": "issue:4125", + "right": "issue:5274", + "accept": false, + "reason": "A question about LoRA loss curves is not the same as a dtype mismatch training error." + }, + { + "left": "issue:4213", + "right": "issue:8752", + "accept": false, + "reason": "Both mention from_single_file, but one is about missing accelerate while the other is about offline server loading; different root causes." + }, + { + "left": "issue:10107", + "right": "issue:8557", + "accept": false, + "reason": "Training example validation checkpointing is unrelated to SD3 token-limit constraints." + }, + { + "left": "issue:10950", + "right": "issue:8469", + "accept": false, + "reason": "Documentation brokenness and broken face outputs are not the same underlying bug." + }, + { + "left": "issue:9309", + "right": "issue:9979", + "accept": true, + "reason": "Both point to Flux ControlNet pipeline failures on Flux 1.x, and the explicit channels error in 9979 plausibly describes the same underlying code-path bug reported more broadly in 9309." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: almost all pairs are clearly different by title/content, and the only plausible duplicate subgroup is the Flux ControlNet pair (9309/9979). The soft verdicts stay conservative overall, and the acceptance of that one pair is appropriately qualified as plausible rather than certain." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 271, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10672", + "issue:10869", + "issue:10883", + "issue:10967", + "issue:11022", + "issue:11035", + "issue:11222", + "issue:1161", + "issue:11630", + "issue:12147", + "issue:122", + "issue:13203", + "issue:1451", + "issue:1798", + "issue:1896", + "issue:2107", + "issue:2548", + "issue:2676", + "issue:2780", + "issue:2856", + "issue:3087", + "issue:3296", + "issue:3431", + "issue:3974", + "issue:4059", + "issue:5168", + "issue:5368", + "issue:5576", + "issue:5757", + "issue:5984", + "issue:599", + "issue:6054", + "issue:6399", + "issue:6958", + "issue:7238", + "issue:7370", + "issue:756", + "issue:757", + "issue:7661", + "issue:7877", + "issue:8309", + "issue:852", + "issue:8826", + "issue:9216", + "issue:941", + "issue:9551", + "issue:9637", + "issue:9958" + ], + "soft_pairs": [ + "issue:11022|issue:13203", + "issue:10869|issue:8826", + "issue:11035|issue:9958", + "issue:11222|issue:11630", + "issue:2780|issue:756", + "issue:6399|issue:6958", + "issue:852|issue:941", + "issue:2107|issue:3431", + "issue:5757|issue:5984", + "issue:5576|issue:8309", + "issue:7238|issue:7661", + "issue:10672|issue:7877", + "issue:122|issue:2856", + "issue:2548|issue:3296", + "issue:599|issue:757", + "issue:6054|issue:8826", + "issue:10883|issue:9216", + "issue:10967|issue:9551", + "issue:2676|issue:4059", + "issue:3087|issue:3974", + "issue:5368|issue:9637", + "issue:1161|issue:1798", + "issue:5168|issue:7370", + "issue:1451|issue:1896", + "issue:12147|issue:13203" + ], + "prepared_review_unit_hash": "bbe9fcf2405c2b90bec4cdb0f38a94c9e1952b7fe18f6ab81678a9113f1035e0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34224, + "estimated_input_tokens": 8556, + "estimated_eval_tokens": 17368 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are a mixed bag of unrelated bugs and feature requests; the soft pairs mostly share surface keywords or the same subsystem, but not the same underlying change. I would not merge any of the candidate pairs as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": "No single issue is a true canonical duplicate for this set because the cluster is heterogeneous. If one must be chosen, issue #11630 is the most actionable standalone request thanks to its explicit target API and concrete feature ask.", + "canonical_pr_reason": null, + "best_issue_reason": "#11630 is the strongest issue to keep as the representative item: it is specific, implementation-oriented, and names the exact missing support request for WanVACETransformer3DModel.from_single_file.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11022", + "right": "issue:13203", + "accept": false, + "reason": "Different models and different problems: WAN Quanto conversion support vs Zimage LoRA support." + }, + { + "left": "issue:10869", + "right": "issue:8826", + "accept": false, + "reason": "Meta tensor .item() failure is unrelated to torch.cat on an empty list." + }, + { + "left": "issue:11035", + "right": "issue:9958", + "accept": false, + "reason": "from_single_file download symlink handling is unrelated to AutoencoderKLTemporalDecoder tiling." + }, + { + "left": "issue:11222", + "right": "issue:11630", + "accept": false, + "reason": "Both are WAN-related, but one is LoRA coverage and the other is from_single_file model loading; different feature requests." + }, + { + "left": "issue:2780", + "right": "issue:756", + "accept": false, + "reason": "FP16 gradient unscale error and deprecation tuple misuse are unrelated training/runtime issues." + }, + { + "left": "issue:6399", + "right": "issue:6958", + "accept": false, + "reason": "LoRA visibility in UI is unrelated to a missing use_linear_projection setting in unet_motion_model." + }, + { + "left": "issue:852", + "right": "issue:941", + "accept": false, + "reason": "Both mention MPS, but they are distinct bugs in different code paths and symptoms." + }, + { + "left": "issue:2107", + "right": "issue:3431", + "accept": false, + "reason": "Low-resolution black images is a different problem from poor inpainting+ControlNet results." + }, + { + "left": "issue:5757", + "right": "issue:5984", + "accept": false, + "reason": "Checkpoint resume errors and fp16 RAM usage are separate training issues." + }, + { + "left": "issue:5576", + "right": "issue:8309", + "accept": false, + "reason": "Different feature requests: a FreeNoise scheduler versus conditional diffusion distillation." + }, + { + "left": "issue:7238", + "right": "issue:7661", + "accept": false, + "reason": "IP-Adapter masking and enabling AnimateDiff with IP-Adapter are related only by subsystem, not the same bug." + }, + { + "left": "issue:10672", + "right": "issue:7877", + "accept": false, + "reason": "callback_on_step_end support and Attention out_dim support are unrelated API requests." + }, + { + "left": "issue:122", + "right": "issue:2856", + "accept": false, + "reason": "Import error during training is unrelated to converting an SD model to diffusers." + }, + { + "left": "issue:2548", + "right": "issue:3296", + "accept": false, + "reason": "Both touch saving, but they concern different scripts and failure modes." + }, + { + "left": "issue:599", + "right": "issue:757", + "accept": false, + "reason": "Deepspeed integration and attention splicing for textual inversion are unrelated changes." + }, + { + "left": "issue:6054", + "right": "issue:8826", + "accept": false, + "reason": "Scheduler from_numpy misuse is unrelated to torch.cat on an empty list." + }, + { + "left": "issue:10883", + "right": "issue:9216", + "accept": false, + "reason": "A Marigold update request is unrelated to a custom multi-model pipeline concept." + }, + { + "left": "issue:10967", + "right": "issue:9551", + "accept": false, + "reason": "Integrating LaVie/Cinemo is unrelated to using x-labs Flux ControlNet models." + }, + { + "left": "issue:2676", + "right": "issue:4059", + "accept": false, + "reason": "Scheduler naming question is unrelated to the ControlNet ReferencePipeline bug/question." + }, + { + "left": "issue:3087", + "right": "issue:3974", + "accept": false, + "reason": "LyCORIS support and ControlNet inpaint safetensors handling are different requests." + }, + { + "left": "issue:5368", + "right": "issue:9637", + "accept": false, + "reason": "Both involve attention dtypes, but one is fp16 QKV dtype mismatch and the other is xFormers mask/bfloat/sequence-length behavior." + }, + { + "left": "issue:1161", + "right": "issue:1798", + "accept": false, + "reason": "General Flax-to-PyTorch conversion is different from adding v_prediction support to a Flax training script." + }, + { + "left": "issue:5168", + "right": "issue:7370", + "accept": false, + "reason": "ControlNet-XS support and OpenSora STDiT support are unrelated model integrations." + }, + { + "left": "issue:1451", + "right": "issue:1896", + "accept": false, + "reason": "Flax pipeline example failure is unrelated to the CLIP-guided size/type error." + }, + { + "left": "issue:12147", + "right": "issue:13203", + "accept": false, + "reason": "Wan2.2 Lightning LoRA loading failure is unrelated to Zimage LoRA support." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 272, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10057", + "issue:10120", + "issue:1083", + "issue:1106", + "issue:1109", + "issue:11859", + "issue:11978", + "issue:1205", + "issue:12412", + "issue:1556", + "issue:1811", + "issue:2514", + "issue:2822", + "issue:3405", + "issue:3683", + "issue:3891", + "issue:4149", + "issue:4186", + "issue:4217", + "issue:4269", + "issue:4530", + "issue:4665", + "issue:4773", + "issue:5128", + "issue:5258", + "issue:542", + "issue:5567", + "issue:5595", + "issue:5628", + "issue:5646", + "issue:5824", + "issue:5989", + "issue:6347", + "issue:6368", + "issue:6536", + "issue:6899", + "issue:7686", + "issue:7931", + "issue:8526", + "issue:8667", + "issue:8873", + "issue:8970", + "issue:9645", + "issue:9664", + "issue:9763", + "issue:9793", + "issue:9942", + "issue:9957" + ], + "soft_pairs": [ + "issue:5567|issue:5824", + "issue:1083|issue:1205", + "issue:1106|issue:542", + "issue:11859|issue:9957", + "issue:1811|issue:4665", + "issue:3405|issue:3683", + "issue:5989|issue:6899", + "issue:9645|issue:9942", + "issue:5258|issue:7931", + "issue:10120|issue:6536", + "issue:1109|issue:1556", + "issue:11978|issue:12412", + "issue:8526|issue:8970", + "issue:4186|issue:4530", + "issue:2514|issue:4217", + "issue:3891|issue:6347", + "issue:4149|issue:8667", + "issue:8873|issue:9664", + "issue:5595|issue:7686", + "issue:10057|issue:9763", + "issue:4773|issue:5128", + "issue:6368|issue:9793", + "issue:2822|issue:4269", + "issue:5628|issue:5646" + ], + "prepared_review_unit_hash": "812310177adcdb845fbdc99efa7a6e99fc8e6961ec0f4d305921ab1c387940d7", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33684, + "estimated_input_tokens": 8421, + "estimated_eval_tokens": 17098 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All candidates appear to be distinct issues rather than true duplicates. The cluster is noisy: many items mention similar subsystems (Flux/SDXL/MPS/schedulers/loading), but the concrete failure modes and targets differ.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5567", + "right": "issue:5824", + "accept": false, + "reason": "Different bugs: one is accelerator.prepare overwriting unet.config, the other is an SDXL time-ids default value problem." + }, + { + "left": "issue:1083", + "right": "issue:1205", + "accept": false, + "reason": "Both involve ONNX, but one reports incorrect FP16 ONNX output while the other is about converting Stable Diffusion to ONNX." + }, + { + "left": "issue:1106", + "right": "issue:542", + "accept": false, + "reason": "Unrelated code paths: repeated UNet calls in PNDM vs a UNet ONNX export issue." + }, + { + "left": "issue:11859", + "right": "issue:9957", + "accept": false, + "reason": "Both are scheduler ideas, but they are different scheduler implementations and not the same change." + }, + { + "left": "issue:1811", + "right": "issue:4665", + "accept": false, + "reason": "Related to weight-file selection, but one is about downloading both bin and safetensors, the other about use_safetensors failing on Windows with alternate bin files." + }, + { + "left": "issue:3405", + "right": "issue:3683", + "accept": false, + "reason": "Training bug vs serialization feature request; not the same underlying issue." + }, + { + "left": "issue:5989", + "right": "issue:6899", + "accept": false, + "reason": "One is a safetensors.torch bug, the other is a converter-script bug; different failure points." + }, + { + "left": "issue:9645", + "right": "issue:9942", + "accept": false, + "reason": "Import error for a loader mixin vs pip installation/versioning failure; distinct problems." + }, + { + "left": "issue:5258", + "right": "issue:7931", + "accept": false, + "reason": "Different pipeline features: SDXL ControlNet inpaint with refiner vs IP-Adapter image_embeds support." + }, + { + "left": "issue:10120", + "right": "issue:6536", + "accept": false, + "reason": "Both are memory-related, but one is DreamBooth/SD3 memory consumption and the other is UNet3DConditionModel OOM during fine-tuning." + }, + { + "left": "issue:1109", + "right": "issue:1556", + "accept": false, + "reason": "CPU utilization complaint vs black images with NSFW warning; unrelated symptoms and code paths." + }, + { + "left": "issue:11978", + "right": "issue:12412", + "accept": false, + "reason": "Different model-support requests: Bria 3.2 pipeline vs HunyuanImage-3.0 support." + }, + { + "left": "issue:8526", + "right": "issue:8970", + "accept": false, + "reason": "Both mention MPS, but one is ineffective joint_attention_kwargs scale on SD3 and the other is black images in Kolors." + }, + { + "left": "issue:4186", + "right": "issue:4530", + "accept": false, + "reason": "Both involve SDXL refiner, but one is inpaint-with-refiner failure and the other is artifacting in generated images." + }, + { + "left": "issue:2514", + "right": "issue:4217", + "accept": false, + "reason": "Both concern loading speed, but one is cached pipeline load time and the other is safetensors load time." + }, + { + "left": "issue:3891", + "right": "issue:6347", + "accept": false, + "reason": "Different model additions: Tortoise TTS pipeline vs GlueGen." + }, + { + "left": "issue:4149", + "right": "issue:8667", + "accept": false, + "reason": "Different tensor-mismatch triggers and models: SemanticStableDiffusion batch handling vs SD3ControlNet non-power-of-two image sizes." + }, + { + "left": "issue:8873", + "right": "issue:9664", + "accept": false, + "reason": "Performance complaint with xformers on A100 vs a training subprocess failure; no shared bug." + }, + { + "left": "issue:5595", + "right": "issue:7686", + "accept": false, + "reason": "Different validation errors: image size divisibility vs missing cross_attention_dim for a UNet block." + }, + { + "left": "issue:10057", + "right": "issue:9763", + "accept": false, + "reason": "Both mention Flux and token limits, but one is a from_single_file loading/token issue while the other is training truncation beyond 77 tokens." + }, + { + "left": "issue:4773", + "right": "issue:5128", + "accept": false, + "reason": "Different loading scenarios: CivitAI LoRA loading vs local .safetensors loading via XLPipeline.from_single_file." + }, + { + "left": "issue:6368", + "right": "issue:9793", + "accept": false, + "reason": "Checkpoint/lora alpha support error vs progress bar update problem; unrelated." + }, + { + "left": "issue:2822", + "right": "issue:4269", + "accept": false, + "reason": "Training convergence issue vs a TypeError in check_inputs; not the same bug." + }, + { + "left": "issue:5628", + "right": "issue:5646", + "accept": false, + "reason": "Both scheduler-related, but they describe different schedulers and different failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 273, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10066", + "issue:10315", + "issue:1055", + "issue:10653", + "issue:10659", + "issue:10674", + "issue:10749", + "issue:10757", + "issue:10782", + "issue:10962", + "issue:11005", + "issue:1158", + "issue:11779", + "issue:12299", + "issue:1895", + "issue:2026", + "issue:2891", + "issue:3761", + "issue:3805", + "issue:3816", + "issue:3886", + "issue:4314", + "issue:4492", + "issue:4840", + "issue:5002", + "issue:5016", + "issue:5628", + "issue:5630", + "issue:5737", + "issue:5907", + "issue:5984", + "issue:6427", + "issue:7147", + "issue:7621", + "issue:801", + "issue:8514", + "issue:8792", + "issue:9023", + "issue:9052", + "issue:9142", + "issue:9181", + "issue:9422", + "issue:9471", + "issue:962", + "issue:9704", + "issue:9850", + "issue:9867" + ], + "soft_pairs": [ + "issue:1158|issue:3886", + "issue:10066|issue:9471", + "issue:10653|issue:10659", + "issue:5002|issue:5907", + "issue:1895|issue:4492", + "issue:9142|issue:9704", + "issue:4840|issue:9052", + "issue:5737|issue:9850", + "issue:2026|issue:2891", + "issue:3761|issue:4314", + "issue:3805|issue:5984", + "issue:5016|issue:5984", + "issue:10962|issue:12299", + "issue:5630|issue:9181", + "issue:5628|issue:8792", + "issue:801|issue:962", + "issue:9023|issue:9422", + "issue:10315|issue:8514", + "issue:11779|issue:9867", + "issue:6427|issue:7621", + "issue:1055|issue:3816", + "issue:10674|issue:10749", + "issue:10757|issue:7147", + "issue:10782|issue:11005" + ], + "prepared_review_unit_hash": "a947f8f55feea4732520c119f92b283071c86f474d4c85b0cfe0dcf92ee21aea", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33564, + "estimated_input_tokens": 8391, + "estimated_eval_tokens": 17038 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated bug reports and feature requests from different pipelines, schedulers, and tooling paths. None of the soft pairs look like the same underlying issue, so all candidate duplicate links should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": "issue:10066 is the most concrete, actionable bug report in the set (specific traceback, clear failure mode), but there is no strong evidence that any other item is a true duplicate of it.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "issue:10066 is the best representative issue only because it is specific and well-scoped; however, the cluster overall does not form a coherent duplicate set.", + "best_pr_reason": "No pull requests are present, so there is no suitable canonical PR.", + "soft_edge_verdicts": [ + { + "left": "issue:1158", + "right": "issue:3886", + "accept": false, + "reason": "Different failures: test interference vs distributed training collective timeout; not the same bug." + }, + { + "left": "issue:10066", + "right": "issue:9471", + "accept": false, + "reason": "Unpack error in attention code vs file-extension inference error; unrelated code paths." + }, + { + "left": "issue:10653", + "right": "issue:10659", + "accept": false, + "reason": "bitsandbytes 8-bit device-move limitation vs poor output quality in ConsisID; different problems." + }, + { + "left": "issue:5002", + "right": "issue:5907", + "accept": false, + "reason": "Both mention scheduler arguments/docs, but one is missing an argument while the other disagrees with docs; not the same concrete bug." + }, + { + "left": "issue:1895", + "right": "issue:4492", + "accept": false, + "reason": "Multi-GPU EMA update issue vs JSON-style lowercase boolean NameError; unrelated." + }, + { + "left": "issue:9142", + "right": "issue:9704", + "accept": false, + "reason": "Multiple controlnets attribute error vs cuDNN execution-plan error; separate runtime failures." + }, + { + "left": "issue:4840", + "right": "issue:9052", + "accept": false, + "reason": "Checkpoint loading mismatch in FlaxAutoencoderKL vs single-file conversion failing on certain checkpoints; different loaders and symptoms." + }, + { + "left": "issue:5737", + "right": "issue:9850", + "accept": false, + "reason": "SDXL conversion failure vs gradient checkpointing with frozen model; no shared bug." + }, + { + "left": "issue:2026", + "right": "issue:2891", + "accept": false, + "reason": "cpu_offload issue vs embeds support/documentation mismatch; different pipeline behaviors." + }, + { + "left": "issue:3761", + "right": "issue:4314", + "accept": false, + "reason": "Community pipeline loading conflict vs interrupted-download write protection; unrelated filesystem/runtime issues." + }, + { + "left": "issue:3805", + "right": "issue:5984", + "accept": false, + "reason": "Colab subprocess failure vs fp16 RAM usage problem; not the same failure mode." + }, + { + "left": "issue:5016", + "right": "issue:5984", + "accept": false, + "reason": "Text encoder training not working vs RAM blow-up from fp16 variant; different training issues." + }, + { + "left": "issue:10962", + "right": "issue:12299", + "accept": false, + "reason": "Prompt-embedding shape issue in Cogview4 vs negative prompt embedding mismatch in SD3; similar theme but different pipelines and bugs." + }, + { + "left": "issue:5630", + "right": "issue:9181", + "accept": false, + "reason": "LMS/Karras denoising offset vs ControlNet noise dot artifact; different scheduler/pipeline symptoms." + }, + { + "left": "issue:5628", + "right": "issue:8792", + "accept": false, + "reason": "LMS denoising-end/start offset with Karras sigmas vs image quality degradation migrating from A1111; not the same concrete defect." + }, + { + "left": "issue:801", + "right": "issue:962", + "accept": false, + "reason": "Attention slicing not respected in inpaint/img2img vs inpaint pipelines not supporting non-512 sizes; different functional gaps." + }, + { + "left": "issue:9023", + "right": "issue:9422", + "accept": false, + "reason": "Sample code failure vs NaNs during ControlNet-SDXL training; too broad and not the same bug." + }, + { + "left": "issue:10315", + "right": "issue:8514", + "accept": false, + "reason": "CogVideo training error vs EMAModel parallel fine-tuning error; different training stacks and failure causes." + }, + { + "left": "issue:11779", + "right": "issue:9867", + "accept": false, + "reason": "Mask padding bug in Wan VACE vs FluxInpaint overwriting pixels outside mask; both are inpainting-related but distinct defects." + }, + { + "left": "issue:6427", + "right": "issue:7621", + "accept": false, + "reason": "diffusers-to-.ckpt conversion problem vs InstructPix2Pix with cosxl_edit compatibility issue; unrelated conversion vs inference bugs." + }, + { + "left": "issue:1055", + "right": "issue:3816", + "accept": false, + "reason": "Missing LatentDiffusionPipeline import availability vs missing CLIPImageProcessor attribute in transformers; different dependency/API issues." + }, + { + "left": "issue:10674", + "right": "issue:10749", + "accept": false, + "reason": "Flux GGUF support failure vs request to add GGUF support in Lumina2; one is a bug report, the other a feature request for a different pipeline." + }, + { + "left": "issue:10757", + "right": "issue:7147", + "accept": false, + "reason": "Variable naming cleanup in a converter script vs missing StableCascadeDecoderPipeline import; unrelated." + }, + { + "left": "issue:10782", + "right": "issue:11005", + "accept": false, + "reason": "Lumina Image minor issue vs docstring/default discrepancy in wan i2v args; different scopes and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 274, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10887", + "issue:11022", + "issue:1117", + "issue:11214", + "issue:11321", + "issue:11833", + "issue:11867", + "issue:11901", + "issue:12071", + "issue:1379", + "issue:1440", + "issue:1611", + "issue:1730", + "issue:1805", + "issue:1896", + "issue:1985", + "issue:1990", + "issue:2015", + "issue:2312", + "issue:2351", + "issue:2412", + "issue:2613", + "issue:2614", + "issue:2689", + "issue:2838", + "issue:3017", + "issue:3062", + "issue:3065", + "issue:3176", + "issue:3291", + "issue:337", + "issue:3412", + "issue:3444", + "issue:3788", + "issue:3951", + "issue:4057", + "issue:4405", + "issue:4580", + "issue:5026", + "issue:6548", + "issue:7119", + "issue:713", + "issue:7392", + "issue:7798", + "issue:8887", + "issue:9241", + "issue:9320", + "issue:938" + ], + "soft_pairs": [ + "issue:10887|issue:9241", + "issue:11022|issue:11901", + "issue:1117|issue:2613", + "issue:11214|issue:12071", + "issue:11321|issue:7798", + "issue:11833|issue:9320", + "issue:11867|issue:8887", + "issue:1379|issue:3062", + "issue:1440|issue:2614", + "issue:1611|issue:5026", + "issue:1730|issue:4057", + "issue:1805|issue:938", + "issue:1896|issue:337", + "issue:1985|issue:2838", + "issue:1990|issue:2015", + "issue:2312|issue:4580", + "issue:2351|issue:6548", + "issue:2412|issue:713", + "issue:2689|issue:7119", + "issue:3017|issue:3412", + "issue:3065|issue:4405", + "issue:3176|issue:3951", + "issue:3291|issue:7392", + "issue:3444|issue:3788" + ], + "prepared_review_unit_hash": "63f69cc433ab9aec4c9b3b337513b4a16b8b78f1713b81a9de23a0be84f89d9a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33484, + "estimated_input_tokens": 8371, + "estimated_eval_tokens": 16998 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: the suggested pairs span unrelated bugs, feature requests, docs, and model/pipeline-specific issues. None of the soft pairs look like the same underlying change or defect, so all should be rejected.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 11214 is the most concrete, self-contained bug report in the set (specific runtime TypeError with a clear code path). It is the best representative issue only in the weak sense that the cluster does not form a real duplicate group.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11214 is the most suitable single issue to anchor triage because it is open, specific, and actionable; the other items are mostly unrelated or more like feature requests/docs than duplicate bug reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10887", + "right": "issue:9241", + "accept": false, + "reason": "Different topics: SDXLTrainer import error vs XLab-to-diffusers LoRA conversion script." + }, + { + "left": "issue:11022", + "right": "issue:11901", + "accept": false, + "reason": "Both mention WAN 2.1, but one is Quanto transformation support and the other is LoRA loading on MPS; different failure modes and code paths." + }, + { + "left": "issue:1117", + "right": "issue:2613", + "accept": false, + "reason": "Negative prompts tests vs adding multiple LoRA layers; unrelated scope." + }, + { + "left": "issue:11214", + "right": "issue:12071", + "accept": false, + "reason": "Different runtime bugs in different models/components: UVit2DModel.forward keyword handling vs Flux Kontext single-block behavior." + }, + { + "left": "issue:11321", + "right": "issue:7798", + "accept": false, + "reason": "Flux ControlNet README bug is documentation, while PixArtAlphaPipeline bug is model/runtime behavior." + }, + { + "left": "issue:11833", + "right": "issue:9320", + "accept": false, + "reason": "LoRA device placement KeyError is a runtime state bug; train_custom_diffusion save/load is a separate persistence issue." + }, + { + "left": "issue:11867", + "right": "issue:8887", + "accept": false, + "reason": "Question about SD3 control vs SD3 VAE encode bug; not the same defect." + }, + { + "left": "issue:1379", + "right": "issue:3062", + "accept": false, + "reason": "Repeated operation in Transformer2DModel vs a docstring question about encoder_hidden_states; one is code behavior, the other docs." + }, + { + "left": "issue:1440", + "right": "issue:2614", + "accept": false, + "reason": "Deleting cached model vs changing pretrained-model download directory; unrelated utilities." + }, + { + "left": "issue:1611", + "right": "issue:5026", + "accept": false, + "reason": "StableDiffusionMegaPipeline regression vs KeyError 'text_encoder' in 0.21.0; different pipeline breakages." + }, + { + "left": "issue:1730", + "right": "issue:4057", + "accept": false, + "reason": "CKPT conversion/webui compatibility vs blurred finetuned inpainting output; different problems." + }, + { + "left": "issue:1805", + "right": "issue:938", + "accept": false, + "reason": "New scalable diffusion pipeline proposal vs human motion diffusion model proposal; distinct features." + }, + { + "left": "issue:1896", + "right": "issue:337", + "accept": false, + "reason": "CLIP-guided Stable Diffusion size error vs img2img/inpainting issue; different code paths." + }, + { + "left": "issue:1985", + "right": "issue:2838", + "accept": false, + "reason": "Embedding-loading feature request vs ControlNet pipeline behavior question; not the same change." + }, + { + "left": "issue:1990", + "right": "issue:2015", + "accept": false, + "reason": "Two separate pipeline proposals (Depth2ImgVariation vs categorical diffusion); unrelated." + }, + { + "left": "issue:2312", + "right": "issue:4580", + "accept": false, + "reason": "EMA save hook error in training vs UniPC schedule exception with karras sigmas; unrelated scheduler/training bugs." + }, + { + "left": "issue:2351", + "right": "issue:6548", + "accept": false, + "reason": "Missing get_velocity in DEISMultistepScheduler vs RGBA image misclassified as latents; different defects." + }, + { + "left": "issue:2412", + "right": "issue:713", + "accept": false, + "reason": "Dreambooth error vs generic training error; too broad and not the same underlying issue." + }, + { + "left": "issue:2689", + "right": "issue:7119", + "accept": false, + "reason": "Broken image link vs invalid docs link; both are link issues but not the same artifact or fix." + }, + { + "left": "issue:3017", + "right": "issue:3412", + "accept": false, + "reason": "Distributed inference docs request vs stable diffusion next-steps docs update; separate documentation tasks." + }, + { + "left": "issue:3065", + "right": "issue:4405", + "accept": false, + "reason": "DreamBooth LoRA text-encoder tuning vs SDXL textual inversion training script; distinct example scripts and goals." + }, + { + "left": "issue:3176", + "right": "issue:3951", + "accept": false, + "reason": "TensorRT demo error vs UniDiffuser NaNs; different subsystems and failure symptoms." + }, + { + "left": "issue:3291", + "right": "issue:7392", + "accept": false, + "reason": "Deepfloyd aspect ratios vs SDXL multiple aspect ratio training; similar theme but different models and requests." + }, + { + "left": "issue:3444", + "right": "issue:3788", + "accept": false, + "reason": "Audio-to-image integration request vs UniControl new pipeline proposal; unrelated feature work." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 275, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:3487", + "issue:3730", + "issue:3765", + "issue:3805", + "issue:3826", + "issue:3891", + "issue:4075", + "issue:4186", + "issue:4213", + "issue:4524", + "issue:4648", + "issue:4668", + "issue:4672", + "issue:4680", + "issue:4803", + "issue:4872", + "issue:4991", + "issue:5258", + "issue:5285", + "issue:5598", + "issue:5783", + "issue:5866", + "issue:5956", + "issue:5980", + "issue:5981", + "issue:6055", + "issue:6132", + "issue:6141", + "issue:6242", + "issue:6331", + "issue:6354", + "issue:6359", + "issue:6439", + "issue:6524", + "issue:6928", + "issue:6949", + "issue:6958", + "issue:7341", + "issue:7405", + "issue:7879", + "issue:8241", + "issue:8390", + "issue:8540", + "issue:8588", + "issue:8906", + "issue:9396", + "issue:962", + "issue:9958" + ], + "soft_pairs": [ + "issue:3487|issue:6949", + "issue:3730|issue:6331", + "issue:3765|issue:5956", + "issue:3805|issue:4872", + "issue:3826|issue:7405", + "issue:3891|issue:6132", + "issue:4075|issue:4803", + "issue:4186|issue:962", + "issue:4213|issue:6055", + "issue:4524|issue:5866", + "issue:4648|issue:8906", + "issue:4668|issue:6524", + "issue:4672|issue:8540", + "issue:4680|issue:5980", + "issue:4991|issue:6958", + "issue:5258|issue:8241", + "issue:5285|issue:6928", + "issue:5598|issue:5981", + "issue:5783|issue:8588", + "issue:6141|issue:8390", + "issue:6242|issue:7879", + "issue:6354|issue:9958", + "issue:6359|issue:7341", + "issue:6439|issue:9396" + ], + "prepared_review_unit_hash": "7efa34e0864eda513d2472fe3917d227767703a800986a7b661e9305764e2a3f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33302, + "estimated_input_tokens": 8326, + "estimated_eval_tokens": 16908 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All candidate soft edges are false positives: the cluster is a mix of unrelated issues spanning inpainting, schedulers, training, docs, datasets, and new pipeline/model requests.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3487", + "right": "issue:6949", + "accept": false, + "reason": "Both concern generation quality, but one is about inpaint strength with EulerAncestral and the other is blank outputs from DPMSolverSinglestep at even steps; different scheduler bugs." + }, + { + "left": "issue:3730", + "right": "issue:6331", + "accept": false, + "reason": "Inpainting time-step blending bug vs a SAG feature/request for inpainting; not the same underlying problem." + }, + { + "left": "issue:3765", + "right": "issue:5956", + "accept": false, + "reason": "Both mention training quality, but one is ControlNet stacking artifacts during training and the other is degraded SDXL U-Net fine-tuning performance; different issues." + }, + { + "left": "issue:3805", + "right": "issue:4872", + "accept": false, + "reason": "Both involve Colab failures, but one is a generic CalledProcessError and the other is CUDA OOM in SDXL finetuning; different failure modes." + }, + { + "left": "issue:3826", + "right": "issue:7405", + "accept": false, + "reason": "Both are new model/pipeline proposals, but they target completely different modalities: audio diffusion vs video diffusion." + }, + { + "left": "issue:3891", + "right": "issue:6132", + "accept": false, + "reason": "Tortoise TTS pipeline request vs adding LOADABLE_CLASSES support in PyTorch serialization; unrelated changes." + }, + { + "left": "issue:4075", + "right": "issue:4803", + "accept": false, + "reason": "Minor ShapE docstring typo vs a missing requirements file error; unrelated." + }, + { + "left": "issue:4186", + "right": "issue:962", + "accept": false, + "reason": "Both are inpainting-related, but one is SDXL inpaint with refiner and the other is a 512x512 size limitation in older inpaint pipelines; different bugs." + }, + { + "left": "issue:4213", + "right": "issue:6055", + "accept": false, + "reason": "SD-XL from_single_file() needing accelerate vs a generic accelerate error; too vague and not the same bug." + }, + { + "left": "issue:4524", + "right": "issue:5866", + "accept": false, + "reason": "AnimateDiff library request vs AnimateDiff + ControlNet feature request; related area but different change requests." + }, + { + "left": "issue:4648", + "right": "issue:8906", + "accept": false, + "reason": "Missing config.json during loading vs absence of qk_norm in SD3Transformer2DModel; unrelated." + }, + { + "left": "issue:4668", + "right": "issue:6524", + "accept": false, + "reason": "Weird images from Kohya-style LoRA vs LoRA breaking on second load; different LoRA problems." + }, + { + "left": "issue:4672", + "right": "issue:8540", + "accept": false, + "reason": "SDXL LoRA VRAM reduction request vs SD3 dreambooth embedding/VAE caching; different training optimizations." + }, + { + "left": "issue:4680", + "right": "issue:5980", + "accept": false, + "reason": "SDXL inpainting fine-tuning vs ControlNet SDXL model export format question; unrelated." + }, + { + "left": "issue:4991", + "right": "issue:6958", + "accept": false, + "reason": "Missing train_dataloader in accelerate.prepare() vs use_linear_projection absent in mid_block; different code paths." + }, + { + "left": "issue:5258", + "right": "issue:8241", + "accept": false, + "reason": "SDXL ControlNet inpaint with refiner vs abnormal LEDITS++ model size; unrelated model issues." + }, + { + "left": "issue:5285", + "right": "issue:6928", + "accept": false, + "reason": "Docs translation request vs Marigold depthmap colorization option; unrelated." + }, + { + "left": "issue:5598", + "right": "issue:5981", + "accept": false, + "reason": "_modify_text_encoder bug vs DPMSolverMultistepScheduler bug; different components." + }, + { + "left": "issue:5783", + "right": "issue:8588", + "accept": false, + "reason": "EMA not working in SDXL training script vs missing SD3 conversion script; unrelated." + }, + { + "left": "issue:6141", + "right": "issue:8390", + "accept": false, + "reason": "ControlNet parameter support request vs Stable Cascade ControlNet training issue; different change scopes." + }, + { + "left": "issue:6242", + "right": "issue:7879", + "accept": false, + "reason": "Dataset log request vs a specific captions dataset request; not the same issue." + }, + { + "left": "issue:6354", + "right": "issue:9958", + "accept": false, + "reason": "LoRA support for ControlNet vs AutoencoderKLTemporalDecoder tiling support; unrelated." + }, + { + "left": "issue:6359", + "right": "issue:7341", + "accept": false, + "reason": "EulerDiscreteScheduler velocity/add_noise bug vs a TorchVersion/Version comparison TypeError; different bugs." + }, + { + "left": "issue:6439", + "right": "issue:9396", + "accept": false, + "reason": "Broken fp16 doc link vs Korean docs translation; unrelated documentation tasks." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 276, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10191", + "issue:12012", + "issue:13073", + "issue:1323", + "issue:13300", + "issue:1495", + "issue:1608", + "issue:1672", + "issue:1946", + "issue:2578", + "issue:3116", + "issue:3222", + "issue:3338", + "issue:3908", + "issue:4043", + "issue:4356", + "issue:4615", + "issue:4646", + "issue:5281", + "issue:5619", + "issue:5770", + "issue:6351", + "issue:6481", + "issue:6725", + "issue:6742", + "issue:6949", + "issue:7054", + "issue:746", + "issue:7592", + "issue:7636", + "issue:8320", + "issue:8405", + "issue:8429", + "issue:8523", + "issue:8960", + "issue:9007", + "issue:9112", + "issue:9178", + "issue:9350", + "issue:9361", + "issue:9471", + "issue:9488", + "issue:9495", + "issue:9602", + "issue:9846", + "issue:9850", + "issue:9902", + "issue:9962" + ], + "soft_pairs": [ + "issue:6481|issue:9350", + "issue:6725|issue:9488", + "issue:6949|issue:9007", + "issue:7636|issue:8405", + "issue:8320|issue:9846", + "issue:8960|issue:9850", + "issue:9178|issue:9361", + "issue:9902|issue:9962", + "issue:7636|issue:9495", + "issue:12012|issue:9471", + "issue:8429|issue:9112", + "issue:3116|issue:3908", + "issue:7054|issue:7592", + "issue:10191|issue:9602", + "issue:1608|issue:1946", + "issue:4043|issue:5619", + "issue:13073|issue:13300", + "issue:6742|issue:9850", + "issue:1495|issue:2578", + "issue:1672|issue:3222", + "issue:5281|issue:8523", + "issue:5770|issue:6351", + "issue:1323|issue:4615", + "issue:3338|issue:4356", + "issue:4646|issue:746" + ], + "prepared_review_unit_hash": "eab199b693b10864f8f15951ab77c8ccf5a9f117c66c91f08176f7ab1b4688d2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34509, + "estimated_input_tokens": 8628, + "estimated_eval_tokens": 17512 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft-edge pair looks like the same underlying bug/change; the set is mostly unrelated issue reports. There are no PRs in the batch. If a representative issue must be chosen, issue 13073 is the clearest narrowly-scoped bug report.", + "confidence": 0.39, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 13073 is the most concrete and narrowly scoped bug report here, with a specific pipeline and behavior, so it is the best anchor if one issue must represent the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6481", + "right": "issue:9350", + "accept": false, + "reason": "Different problems: batch-size-related training failure vs a deprecation warning about 3D txt_ids tensors." + }, + { + "left": "issue:6725", + "right": "issue:9488", + "accept": false, + "reason": "Both are generation issues, but they involve different pipelines and unrelated failure modes." + }, + { + "left": "issue:6949", + "right": "issue:9007", + "accept": false, + "reason": "Same scheduler family, but different concrete symptoms and conditions: blank images at even steps vs black output with order 3." + }, + { + "left": "issue:7636", + "right": "issue:8405", + "accept": false, + "reason": "Both mention ControlNet, but one is about using an SDXL tile model and the other about multi-ControlNet tensor shape errors." + }, + { + "left": "issue:8320", + "right": "issue:9846", + "accept": false, + "reason": "Both concern from_single_file, but they are different models and different missing/invalid behavior." + }, + { + "left": "issue:8960", + "right": "issue:9850", + "accept": false, + "reason": "Training-example gradient accumulation bug is unrelated to frozen-model gradient checkpointing support." + }, + { + "left": "issue:9178", + "right": "issue:9361", + "accept": false, + "reason": "Training-time PEFT target-module errors and inference-time missing LoRA effect are different bugs." + }, + { + "left": "issue:9902", + "right": "issue:9962", + "accept": false, + "reason": "Completely unrelated: conda version pinning vs an unexpected keyword argument error." + }, + { + "left": "issue:7636", + "right": "issue:9495", + "accept": false, + "reason": "Different SDXL features and failures: ControlNet tile usage vs PAG with IPAdapter." + }, + { + "left": "issue:12012", + "right": "issue:9471", + "accept": false, + "reason": "Unrelated: Wan 2.2 cache-context error vs a CoreML file-extension inference error." + }, + { + "left": "issue:8429", + "right": "issue:9112", + "accept": false, + "reason": "Different models and failures: I2VGen-XL script malfunction vs Auraflow tensor-size mismatch." + }, + { + "left": "issue:3116", + "right": "issue:3908", + "accept": false, + "reason": "Scheduler latent-noise bug and Streamlit caching problem are unrelated." + }, + { + "left": "issue:7054", + "right": "issue:7592", + "accept": false, + "reason": "Both are adapter-loading issues, but one mutates an input state_dict and the other misses DoRA keys; not the same bug." + }, + { + "left": "issue:10191", + "right": "issue:9602", + "accept": false, + "reason": "GPU device selection during training is unrelated to an adapter-name validation error." + }, + { + "left": "issue:1608", + "right": "issue:1946", + "accept": false, + "reason": "Training halting at 500 steps is unrelated to xformers CUDA invalid-argument failures." + }, + { + "left": "issue:4043", + "right": "issue:5619", + "accept": false, + "reason": "Different pipelines and different missing-input problems; no same code path." + }, + { + "left": "issue:13073", + "right": "issue:13300", + "accept": false, + "reason": "Both are generation bugs, but they affect different pipelines/models and different failure conditions." + }, + { + "left": "issue:6742", + "right": "issue:9850", + "accept": false, + "reason": "Checkpointing-related, but one is a motion block behavior difference and the other is enabling checkpointing with a frozen model." + }, + { + "left": "issue:1495", + "right": "issue:2578", + "accept": false, + "reason": "Import failure for init_git_repo is unrelated to editable-install behavior." + }, + { + "left": "issue:1672", + "right": "issue:3222", + "accept": false, + "reason": "Wrong image size after conversion is unrelated to missing custom diffusion weight files after training." + }, + { + "left": "issue:5281", + "right": "issue:8523", + "accept": false, + "reason": "Both are CUDA/device errors, but they involve different workflows and no shared concrete bug." + }, + { + "left": "issue:5770", + "right": "issue:6351", + "accept": false, + "reason": "Dataset-access problem and DDP missing-disable_adapters error are unrelated." + }, + { + "left": "issue:1323", + "right": "issue:4615", + "accept": false, + "reason": "Flax dtype-loading failure is unrelated to a missing StableDiffusionControlnetPipeline import." + }, + { + "left": "issue:3338", + "right": "issue:4356", + "accept": false, + "reason": "Import-performance complaint and missing dummy module error are different issues." + }, + { + "left": "issue:4646", + "right": "issue:746", + "accept": false, + "reason": "Both touch ONNX, but one is a conversion failure and the other is an ONNX-revision loading problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 277, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10281", + "issue:10282", + "issue:10596", + "issue:10653", + "issue:1076", + "issue:1083", + "issue:10869", + "issue:11208", + "issue:11241", + "issue:11519", + "issue:11574", + "issue:12025", + "issue:12299", + "issue:12319", + "issue:12489", + "issue:12505", + "issue:12535", + "issue:12754", + "issue:13279", + "issue:13488", + "issue:1440", + "issue:1534", + "issue:1633", + "issue:193", + "issue:1934", + "issue:2153", + "issue:2437", + "issue:2763", + "issue:2858", + "issue:3401", + "issue:4059", + "issue:4104", + "issue:4297", + "issue:4551", + "issue:4631", + "issue:4646", + "issue:5433", + "issue:5953", + "issue:6533", + "issue:7223", + "issue:7370", + "issue:7629", + "issue:7660", + "issue:7895", + "issue:8552", + "issue:9007", + "issue:9516", + "issue:968" + ], + "soft_pairs": [ + "issue:11208|issue:7660", + "issue:4297|issue:5433", + "issue:1534|issue:193", + "issue:1083|issue:4646", + "issue:12319|issue:13488", + "issue:2437|issue:6533", + "issue:4631|issue:9007", + "issue:5953|issue:7629", + "issue:10869|issue:12299", + "issue:11241|issue:8552", + "issue:2153|issue:4104", + "issue:1934|issue:2763", + "issue:3401|issue:4551", + "issue:7223|issue:7370", + "issue:10653|issue:12489", + "issue:10281|issue:11519", + "issue:1440|issue:968", + "issue:1633|issue:2858", + "issue:7895|issue:9516", + "issue:10282|issue:11574", + "issue:12505|issue:12754", + "issue:12535|issue:13279", + "issue:10596|issue:12025", + "issue:1076|issue:4059" + ], + "prepared_review_unit_hash": "95603e57c7c592b4ccb56835e4bf76e64a9b012a3581c86a05d74a0bc9e3a423", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33844, + "estimated_input_tokens": 8461, + "estimated_eval_tokens": 17178 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated bug reports, feature requests, and model/pipeline support questions. None of the soft-similarity pairs look like true duplicates of the same underlying issue/change, so I would not merge any of them.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical duplicate stands out because the items span unrelated problems across different pipelines, schedulers, loading paths, and feature requests. If a representative issue must be chosen, #11208 is the strongest standalone bug report with broad impact and the most discussion.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "#11208 is the best standalone issue to anchor on: it is a concrete, user-facing bug with clear scope and higher engagement than most others here.", + "best_pr_reason": "No pull requests are present in this cluster.", + "soft_edge_verdicts": [ + { + "left": "issue:11208", + "right": "issue:7660", + "accept": false, + "reason": "Different problems: unsupported MultiControlNetModel in one specific pipeline vs a trace example failing to work." + }, + { + "left": "issue:4297", + "right": "issue:5433", + "accept": false, + "reason": "Both mention visual artifacts, but the failure modes and triggering conditions differ (batched img2img watermarks vs DPM++/no-refiner artifacts)." + }, + { + "left": "issue:1534", + "right": "issue:193", + "accept": false, + "reason": "Dreambooth notebook training error is unrelated to a generic pipeline error report." + }, + { + "left": "issue:1083", + "right": "issue:4646", + "accept": false, + "reason": "Incorrect FP16 ONNX output is a different bug from ControlNet ONNX conversion failure." + }, + { + "left": "issue:12319", + "right": "issue:13488", + "accept": false, + "reason": "Both concern offloading, but one is a block_level regression and the other is a version-wide breakage; not the same concrete code-path." + }, + { + "left": "issue:2437", + "right": "issue:6533", + "accept": false, + "reason": "A T2I-Adapter pipeline request is unrelated to the Marigold pipeline request." + }, + { + "left": "issue:4631", + "right": "issue:9007", + "accept": false, + "reason": "Same scheduler family, but one is a Tensor-to-scalar runtime error and the other is black output; too different to treat as one bug." + }, + { + "left": "issue:5953", + "right": "issue:7629", + "accept": false, + "reason": "ControlNet image/control size crash and SVD OOM on PyTorch 1.13 are unrelated." + }, + { + "left": "issue:10869", + "right": "issue:12299", + "accept": false, + "reason": "Meta-tensor item() failure and SD3 tensor mismatch with negative prompts are different shape/embedding bugs." + }, + { + "left": "issue:11241", + "right": "issue:8552", + "accept": false, + "reason": "Flux ControlNet training floating-point exception is unrelated to TextToVideoZeroPipeline segmentation fault." + }, + { + "left": "issue:2153", + "right": "issue:4104", + "accept": false, + "reason": "GTX 1660 float16 runtime issues are not the same as black-image invalid cast output." + }, + { + "left": "issue:1934", + "right": "issue:2763", + "accept": false, + "reason": "Both mention VRAM, but cpu_offload memory bloat and cudnn.benchmark ControlNet VRAM growth are distinct issues." + }, + { + "left": "issue:3401", + "right": "issue:4551", + "accept": false, + "reason": "class_labels / num_class_embeds validation is unrelated to a SDXL Dreambooth crash." + }, + { + "left": "issue:7223", + "right": "issue:7370", + "accept": false, + "reason": "These are different feature requests for different video models (Latte vs OpenSora STDiT)." + }, + { + "left": "issue:10653", + "right": "issue:12489", + "accept": false, + "reason": "Bitsandbytes .to()/device movement problem is unrelated to CUDA sm_120 kernel compatibility." + }, + { + "left": "issue:10281", + "right": "issue:11519", + "accept": false, + "reason": "A new scheduler request and a MAGI-1 support request are different feature asks." + }, + { + "left": "issue:1440", + "right": "issue:968", + "accept": false, + "reason": "Deleting cached models is unrelated to model-loading slowness." + }, + { + "left": "issue:1633", + "right": "issue:2858", + "accept": false, + "reason": "Karras sigmas support and seeding image support are separate pipeline enhancements." + }, + { + "left": "issue:7895", + "right": "issue:9516", + "accept": false, + "reason": "Flax memory-efficient attention bug and FLUX joint_attention_kwargs wiring are different code paths." + }, + { + "left": "issue:10282", + "right": "issue:11574", + "accept": false, + "reason": "Shape mismatch on x_embedder.weight and missing VAE key are different loading errors." + }, + { + "left": "issue:12505", + "right": "issue:12754", + "accept": false, + "reason": "tgate img2img error and from_pipe dtype conversion are unrelated issues." + }, + { + "left": "issue:12535", + "right": "issue:13279", + "accept": false, + "reason": "Wan2.2 LoRA loading failure and AutoencoderRAE compatibility with older transformers are distinct model-loading problems." + }, + { + "left": "issue:10596", + "right": "issue:12025", + "accept": false, + "reason": "FluxControlPipeline shape error and Cosmos VAE Encoder API error are unrelated." + }, + { + "left": "issue:1076", + "right": "issue:4059", + "accept": false, + "reason": "A general stable-diffusion-v1-5 question is unrelated to a ControlNet reference pipeline bug/question." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 278, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10282", + "issue:10411", + "issue:10450", + "issue:10514", + "issue:10591", + "issue:10659", + "issue:1070", + "issue:10920", + "issue:10940", + "issue:10950", + "issue:1195", + "issue:12022", + "issue:12060", + "issue:12221", + "issue:12319", + "issue:12373", + "issue:13430", + "issue:1811", + "issue:2097", + "issue:2392", + "issue:2437", + "issue:2967", + "issue:3347", + "issue:3622", + "issue:3698", + "issue:3876", + "issue:4720", + "issue:5546", + "issue:5607", + "issue:5777", + "issue:5927", + "issue:599", + "issue:6073", + "issue:6964", + "issue:7056", + "issue:725", + "issue:7264", + "issue:7538", + "issue:7732", + "issue:7935", + "issue:7950", + "issue:8277", + "issue:8824", + "issue:8832", + "issue:9109", + "issue:9161", + "issue:9171", + "issue:9464" + ], + "soft_pairs": [ + "issue:10940|issue:7935", + "issue:12060|issue:12373", + "issue:12221|issue:13430", + "issue:5546|issue:7732", + "issue:599|issue:725", + "issue:3876|issue:6964", + "issue:1070|issue:4720", + "issue:7538|issue:8824", + "issue:2437|issue:5927", + "issue:7950|issue:8832", + "issue:10514|issue:10659", + "issue:10920|issue:9171", + "issue:12022|issue:9109", + "issue:9161|issue:9464", + "issue:2392|issue:3347", + "issue:10282|issue:5607", + "issue:10450|issue:10591", + "issue:10950|issue:12319", + "issue:1195|issue:1811", + "issue:3622|issue:5777", + "issue:3698|issue:7264", + "issue:2097|issue:2967", + "issue:6073|issue:7056", + "issue:10411|issue:8277" + ], + "prepared_review_unit_hash": "3403a71205823bee90be2bfeba5605c50ba8bb0029884278968e1f99b8531b20", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34152, + "estimated_input_tokens": 8538, + "estimated_eval_tokens": 17332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set does not contain any clear duplicate issue groups; the suggested soft pairs are mostly superficial title similarity across different bugs, features, and docs/training topics. I\u2019d anchor the cluster on a concrete, well-scoped bug report rather than try to merge unrelated items.", + "confidence": 0.89, + "canonical_issue_reason": "Issue #12319 is a concrete, clearly scoped bug report with strong discussion/inbound references, making it the best anchor among mostly unrelated items.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #12319 is the most suitable representative: specific failure mode, actionable title, and higher engagement than the other issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10940", + "right": "issue:7935", + "accept": false, + "reason": "Both mention loading/import problems, but one is a Triton import error and the other is a missing config.json model-repo issue; different root causes and fixes." + }, + { + "left": "issue:12060", + "right": "issue:12373", + "accept": false, + "reason": "A general question about DiT blocks versus an API proposal for unlearn; unrelated intent and code change surface." + }, + { + "left": "issue:12221", + "right": "issue:13430", + "accept": false, + "reason": "Different feature requests for different model support; same broad theme of contribution requests, but not the same underlying bug/change." + }, + { + "left": "issue:5546", + "right": "issue:7732", + "accept": false, + "reason": "Both involve schedulers/SDXL-ish behavior, but one is a step-count bug in base+refiner ensembles and the other is a ControlNet/refiner runtime error; not the same issue." + }, + { + "left": "issue:599", + "right": "issue:725", + "accept": false, + "reason": "Both are memory/deployment related, but one requests DeepSpeed support and the other device-map/accelerate loading; separate changes." + }, + { + "left": "issue:3876", + "right": "issue:6964", + "accept": false, + "reason": "Both are inpainting-related, but they describe different pipeline failures and model-channel compatibility problems; not the same code-path bug." + }, + { + "left": "issue:1070", + "right": "issue:4720", + "accept": false, + "reason": "Older inpainting bug versus AutoencoderTiny incompatibility; same general area but different component and failure mechanism." + }, + { + "left": "issue:7538", + "right": "issue:8824", + "accept": false, + "reason": "Different training/data-prep questions for different pipelines; no shared concrete bug or patch target." + }, + { + "left": "issue:2437", + "right": "issue:5927", + "accept": false, + "reason": "A feature request for a new pipeline versus a usage/problem report combining existing components; not duplicates." + }, + { + "left": "issue:7950", + "right": "issue:8832", + "accept": false, + "reason": "Framework/library incompatibility versus a generic runtime crash when calling a pipeline; too vague and different root causes." + }, + { + "left": "issue:10514", + "right": "issue:10659", + "accept": false, + "reason": "Sana resolution-binning limitation versus ConsisID output quality complaint; different models and problems." + }, + { + "left": "issue:10920", + "right": "issue:9171", + "accept": false, + "reason": "Both are scheduler/pipeline regression reports, but one is DDIMInverseScheduler infs and the other is from_single_file prediction_type handling; different code paths." + }, + { + "left": "issue:12022", + "right": "issue:9109", + "accept": false, + "reason": "Flash-attention interface mismatch is unrelated to a torchvision nms operator missing error; different dependencies and fixes." + }, + { + "left": "issue:9161", + "right": "issue:9464", + "accept": false, + "reason": "Both affect Flux DreamBooth LoRA training, but one reports distributed bugs and the other a prodigy/train_text_encoder IndexError; not the same concrete bug." + }, + { + "left": "issue:2392", + "right": "issue:3347", + "accept": false, + "reason": "Prompt_embeds validation error in img2img versus NaN image output from a diffusion pipeline; different symptoms and likely different causes." + }, + { + "left": "issue:10282", + "right": "issue:5607", + "accept": false, + "reason": "Both are tensor shape/runtime failures, but the affected shapes and code paths differ; insufficient evidence of the same underlying bug." + }, + { + "left": "issue:10450", + "right": "issue:10591", + "accept": false, + "reason": "SingleStreamBlock implementation mismatch versus sd3 LoRA training script issues; same model family but not the same change or bug." + }, + { + "left": "issue:10950", + "right": "issue:12319", + "accept": false, + "reason": "Docs broken for accelerate inference versus a group offloading runtime bug; different artifact types and fixes." + }, + { + "left": "issue:1195", + "right": "issue:1811", + "accept": false, + "reason": "Memory-efficient attention fp16 support versus from_pretrained choosing between bin and safetensors; unrelated loading/attention problems." + }, + { + "left": "issue:3622", + "right": "issue:5777", + "accept": false, + "reason": "Tutorial import error versus PixArt DreamBooth training failure; different surfaces and root causes." + }, + { + "left": "issue:3698", + "right": "issue:7264", + "accept": false, + "reason": "Inpainting model conversion request versus custom transformers model serialization/loading problem; not the same issue." + }, + { + "left": "issue:2097", + "right": "issue:2967", + "accept": false, + "reason": "Both are UNet1D-related, but one is a general problem using UNet1D and the other cannot get a simple UNet1D to run; not enough evidence they are the same concrete bug." + }, + { + "left": "issue:6073", + "right": "issue:7056", + "accept": false, + "reason": "Beginner DreamBooth SDXL setup error versus a LoRA fine-tuning error with train_text_encoder; different scripts and failure modes." + }, + { + "left": "issue:10411", + "right": "issue:8277", + "accept": false, + "reason": "How-to training question versus a convergence problem in consistency training; different nature and no shared underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 279, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1037", + "issue:10653", + "issue:11448", + "issue:11872", + "issue:12216", + "issue:12359", + "issue:13446", + "issue:1556", + "issue:1931", + "issue:2068", + "issue:2444", + "issue:2534", + "issue:255", + "issue:2761", + "issue:280", + "issue:3238", + "issue:3251", + "issue:3574", + "issue:3799", + "issue:3861", + "issue:3948", + "issue:4264", + "issue:4391", + "issue:4605", + "issue:4665", + "issue:4724", + "issue:4767", + "issue:487", + "issue:5216", + "issue:5334", + "issue:5553", + "issue:5578", + "issue:5907", + "issue:6176", + "issue:6457", + "issue:6811", + "issue:6880", + "issue:6958", + "issue:7421", + "issue:7519", + "issue:7589", + "issue:8393", + "issue:8588", + "issue:8943", + "issue:9188", + "issue:9558", + "issue:9599", + "issue:984" + ], + "soft_pairs": [ + "issue:4767|issue:5907", + "issue:12216|issue:9558", + "issue:3251|issue:6958", + "issue:3574|issue:4665", + "issue:4724|issue:8393", + "issue:1556|issue:3948", + "issue:2444|issue:6880", + "issue:1931|issue:984", + "issue:5216|issue:7589", + "issue:5553|issue:6176", + "issue:12359|issue:13446", + "issue:1037|issue:280", + "issue:11448|issue:8943", + "issue:2068|issue:3238", + "issue:2534|issue:4605", + "issue:2761|issue:4391", + "issue:4264|issue:8588", + "issue:5334|issue:6457", + "issue:5578|issue:6811", + "issue:7421|issue:7519", + "issue:9188|issue:9599", + "issue:10653|issue:11872", + "issue:255|issue:487", + "issue:3799|issue:3861" + ], + "prepared_review_unit_hash": "e175dfc83d20daa516310a9746639bc3ad9784a09678529d67708cd8028fd34d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34053, + "estimated_input_tokens": 8514, + "estimated_eval_tokens": 17284 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly unrelated issue topics; the soft-similarity links are too broad to treat as duplicates, so all candidate merges should be rejected.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4767", + "right": "issue:5907", + "accept": false, + "reason": "Both mention prediction_type, but one is about EulerAncestralDiscreteScheduler support and the other is a DDPMScheduler docs/argument mismatch; different bugs and different code paths." + }, + { + "left": "issue:12216", + "right": "issue:9558", + "accept": false, + "reason": "Qwen-Image-Edit quality regression vs black validation images in TensorBoard; one is model output quality, the other is logging/display." + }, + { + "left": "issue:3251", + "right": "issue:6958", + "accept": false, + "reason": "Different models and failures: ControlNet Shuffle CFG handling vs a unet_motion_model mid_block projection mismatch." + }, + { + "left": "issue:3574", + "right": "issue:4665", + "accept": false, + "reason": "Both involve Windows loading problems, but one is from_ckpt behavior and the other is use_safetensors resolution when alternate bin files exist; not the same underlying bug." + }, + { + "left": "issue:4724", + "right": "issue:8393", + "accept": false, + "reason": "Img2Img output mismatch with an SDXL checkpoint vs slow checkpoint shard loading; unrelated symptoms and fixes." + }, + { + "left": "issue:1556", + "right": "issue:3948", + "accept": false, + "reason": "Black images with an NSFW warning is a safety-checker/output issue, while pale images is a color/quality complaint; not the same defect." + }, + { + "left": "issue:2444", + "right": "issue:6880", + "accept": false, + "reason": "Different scripts and errors: dreambooth_inpaint training failure vs custom diffusion retrieve.py behavior." + }, + { + "left": "issue:1931", + "right": "issue:984", + "accept": false, + "reason": "Float16 datatype mismatch during Stable Diffusion 2 loading is unrelated to F.interpolate breaking for large batch sizes." + }, + { + "left": "issue:5216", + "right": "issue:7589", + "accept": false, + "reason": "Both are GLIGEN-related, but one is a concrete in-painting 0-d tensor TypeError with a specific scheduler and the other is a general layout-to-image failure." + }, + { + "left": "issue:5553", + "right": "issue:6176", + "accept": false, + "reason": "Custom model loading failure and LoRA weight loading after DreamBooth training are distinct pipeline stages and failure modes." + }, + { + "left": "issue:12359", + "right": "issue:13446", + "accept": false, + "reason": "A documentation bug for guidance_scale is not the same as a security/trust-check issue for custom_pipeline." + }, + { + "left": "issue:1037", + "right": "issue:280", + "accept": false, + "reason": "Different feature requests: adding a pipeline vs adding a scheduler." + }, + { + "left": "issue:11448", + "right": "issue:8943", + "accept": false, + "reason": "Missing stable-unclip model file and missing libtorch CUDA shared library are unrelated loading/environment problems." + }, + { + "left": "issue:2068", + "right": "issue:3238", + "accept": false, + "reason": "Datatype conversion in from_pretrained is unrelated to converting LoRA safetensors to diffusers format." + }, + { + "left": "issue:2534", + "right": "issue:4605", + "accept": false, + "reason": "Scheduler switch errors and long-prompt-weighting inpaint errors are different pipeline failures." + }, + { + "left": "issue:2761", + "right": "issue:4391", + "accept": false, + "reason": "Both touch documentation/defaults, but for different models and parameters; not the same issue." + }, + { + "left": "issue:4264", + "right": "issue:8588", + "accept": false, + "reason": "Missing example images and a missing conversion script are both absent artifacts, but they are different files and requests." + }, + { + "left": "issue:5334", + "right": "issue:6457", + "accept": false, + "reason": "Accelerate API handling of multiple models is not the same as multi-GPU gradient accumulation behavior for SDXL-LoRA." + }, + { + "left": "issue:5578", + "right": "issue:6811", + "accept": false, + "reason": "Both mention model_index.json, but one is a local/random-module loading error and the other is a specific model repo URL missing the file." + }, + { + "left": "issue:7421", + "right": "issue:7519", + "accept": false, + "reason": "Seed reproducibility in StableDiffusionXLKDiffusionPipeline is unrelated to xFormers ignoring cross_attention_kwargs['scale']." + }, + { + "left": "issue:9188", + "right": "issue:9599", + "accept": false, + "reason": "Both are FLUX-related, but outpainting and LoRA-only finetuning are different feature requests." + }, + { + "left": "issue:10653", + "right": "issue:11872", + "accept": false, + "reason": "One is an unsupported .to() move for an 8-bit bitsandbytes module; the other is inefficient repeated CPU/GPU offloading loops." + }, + { + "left": "issue:255", + "right": "issue:487", + "accept": false, + "reason": "An image size divisibility bug in StableDiffusionPipeline is not the same as M1 CPU incompatibility." + }, + { + "left": "issue:3799", + "right": "issue:3861", + "accept": false, + "reason": "Inferred type issues for DDPMScheduler.from_pretrained and a Pipeline.from_pretrained naming question are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 280, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10085", + "issue:10111", + "issue:10204", + "issue:10210", + "issue:10475", + "issue:10518", + "issue:1127", + "issue:11432", + "issue:1207", + "issue:12781", + "issue:13249", + "issue:1750", + "issue:2201", + "issue:2402", + "issue:2880", + "issue:2891", + "issue:3081", + "issue:3366", + "issue:3450", + "issue:3481", + "issue:3954", + "issue:3978", + "issue:4198", + "issue:4217", + "issue:4480", + "issue:4668", + "issue:5215", + "issue:5392", + "issue:590", + "issue:5989", + "issue:6256", + "issue:6354", + "issue:6452", + "issue:6574", + "issue:7054", + "issue:7056", + "issue:7142", + "issue:7368", + "issue:7575", + "issue:7889", + "issue:8477", + "issue:8514", + "issue:8700", + "issue:8785", + "issue:8863", + "issue:9415", + "issue:9610", + "issue:989" + ], + "soft_pairs": [ + "issue:7056|issue:8477", + "issue:2201|issue:6574", + "issue:3954|issue:590", + "issue:2891|issue:5215", + "issue:4198|issue:989", + "issue:2880|issue:3081", + "issue:5392|issue:7368", + "issue:6354|issue:9610", + "issue:8514|issue:9415", + "issue:10210|issue:7054", + "issue:7575|issue:8863", + "issue:1127|issue:2402", + "issue:1207|issue:1750", + "issue:10111|issue:7889", + "issue:10518|issue:12781", + "issue:4217|issue:5989", + "issue:6256|issue:6452", + "issue:8700|issue:8785", + "issue:11432|issue:13249", + "issue:10204|issue:10475", + "issue:3978|issue:4480", + "issue:10085|issue:7142", + "issue:3366|issue:3450", + "issue:3481|issue:4668" + ], + "prepared_review_unit_hash": "e0578cf35171d0a437b239d3fd68177d64a634ae8d76951ce4afd5339ab7a117", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33905, + "estimated_input_tokens": 8477, + "estimated_eval_tokens": 17210 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the issues span unrelated pipelines, model families, training scripts, dtype/runtime errors, and feature requests. The soft-similarity pairs are not duplicate candidates for the same underlying bug/change.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 10210 is the most concrete, narrowly scoped bug report in the set (specific pipeline, clear symptom, reproducible behavior), so it is the best single issue representative if a canonical issue must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10210 is the strongest representative overall because it describes a precise, user-facing failure mode rather than a broad or speculative report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7056", + "right": "issue:8477", + "accept": false, + "reason": "Different failures in different training paths: SDXL LoRA fine-tuning error vs a Consistency Training script DDP error." + }, + { + "left": "issue:2201", + "right": "issue:6574", + "accept": false, + "reason": "Tokenizer padding behavior is unrelated to an SVD pipeline warning about expected input range." + }, + { + "left": "issue:3954", + "right": "issue:590", + "accept": false, + "reason": "Multi-GPU training-step counting is unrelated to a typo in a Karras scheduler noise formula." + }, + { + "left": "issue:2891", + "right": "issue:5215", + "accept": false, + "reason": "Docs/API mismatch for upscale embeds is not the same as a brightness regression on dark source images." + }, + { + "left": "issue:4198", + "right": "issue:989", + "accept": false, + "reason": "A dtype mismatch in a notebook is unrelated to a linear-layer weight shape error." + }, + { + "left": "issue:2880", + "right": "issue:3081", + "accept": false, + "reason": "Different Colab breakages in different projects; no shared code path or symptom." + }, + { + "left": "issue:5392", + "right": "issue:7368", + "accept": false, + "reason": "Training an unconditional latent diffusion model and training GLIGEN are different requests, not the same bug." + }, + { + "left": "issue:6354", + "right": "issue:9610", + "accept": false, + "reason": "Adding LoRA support for ControlNet is a different change from adding load_lora_weights to SD3 img2img." + }, + { + "left": "issue:8514", + "right": "issue:9415", + "accept": false, + "reason": "EMAModel parallel fine-tuning failure and CogVideoX parameter-module errors affect different subsystems and code paths." + }, + { + "left": "issue:10210", + "right": "issue:7054", + "accept": false, + "reason": "FluxImg2Img output collapse is unrelated to load_lora_weights mutating a passed state_dict." + }, + { + "left": "issue:7575", + "right": "issue:8863", + "accept": false, + "reason": "Transformer2DModel adanorm shape mismatch is not the same as a generic tuple-has-no-shape AttributeError." + }, + { + "left": "issue:1127", + "right": "issue:2402", + "accept": false, + "reason": "ONNX/DML import traceback is unrelated to Torch being built without CUDA." + }, + { + "left": "issue:1207", + "right": "issue:1750", + "accept": false, + "reason": "Both are symbol/linker errors, but they refer to different missing CUDA-related symbols on different environments." + }, + { + "left": "issue:10111", + "right": "issue:7889", + "accept": false, + "reason": "A vague FLUX pipeline bug summary is not the same underlying issue as single_file_utils checkpoint-loading bugs." + }, + { + "left": "issue:10518", + "right": "issue:12781", + "accept": false, + "reason": "An SD3 example script problem is unrelated to a Discord login/access issue." + }, + { + "left": "issue:4217", + "right": "issue:5989", + "accept": false, + "reason": "Slow safetensors loading is not the same as a safetensors.torch bug." + }, + { + "left": "issue:6256", + "right": "issue:6452", + "accept": false, + "reason": "These are separate feature requests for different pipeline callbacks versus masked-image-latents support." + }, + { + "left": "issue:8700", + "right": "issue:8785", + "accept": false, + "reason": "Both mention PAG, but they target different model families and different pipeline additions." + }, + { + "left": "issue:11432", + "right": "issue:13249", + "accept": false, + "reason": "torch_dtype='auto' loading behavior is unrelated to a LoRA regression on Z-Image." + }, + { + "left": "issue:10204", + "right": "issue:10475", + "accept": false, + "reason": "FluxFill inpainting drift over iterations is different from SD3 fine-tuning inference quality mismatch." + }, + { + "left": "issue:3978", + "right": "issue:4480", + "accept": false, + "reason": "Both are MPS-related, but one is img2img not working and the other is a float16 kernel crash; not the same bug." + }, + { + "left": "issue:10085", + "right": "issue:7142", + "accept": false, + "reason": "LyCORIS batch-inference/skip_guidance_layers runtime error is unrelated to SDXL-lightning inference differing from webui." + }, + { + "left": "issue:3366", + "right": "issue:3450", + "accept": false, + "reason": "LoRALinearLayer autocast behavior is unrelated to from_ckpt loading failures." + }, + { + "left": "issue:3481", + "right": "issue:4668", + "accept": false, + "reason": "Different problems: DDIM step-count sensitivity versus weird images from a Kohya-style LoRA." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 281, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10798", + "issue:10917", + "issue:11437", + "issue:12078", + "issue:12097", + "issue:12110", + "issue:1273", + "issue:13105", + "issue:1353", + "issue:1377", + "issue:1633", + "issue:1704", + "issue:2359", + "issue:2578", + "issue:2780", + "issue:3026", + "issue:3287", + "issue:3481", + "issue:4598", + "issue:5006", + "issue:5043", + "issue:5112", + "issue:5433", + "issue:5485", + "issue:5693", + "issue:5769", + "issue:6250", + "issue:6506", + "issue:6535", + "issue:6790", + "issue:6812", + "issue:6821", + "issue:7210", + "issue:725", + "issue:7365", + "issue:7421", + "issue:7535", + "issue:8233", + "issue:8534", + "issue:8612", + "issue:8671", + "issue:8673", + "issue:8752", + "issue:8792", + "issue:9166", + "issue:9306", + "issue:9539" + ], + "soft_pairs": [ + "issue:5693|issue:6250", + "issue:10917|issue:7365", + "issue:1273|issue:1704", + "issue:2359|issue:725", + "issue:2578|issue:4598", + "issue:5769|issue:8671", + "issue:7535|issue:8612", + "issue:5433|issue:8792", + "issue:6506|issue:6535", + "issue:6790|issue:9166", + "issue:11437|issue:12078", + "issue:2780|issue:6812", + "issue:12097|issue:13105", + "issue:5112|issue:5485", + "issue:5043|issue:725", + "issue:8534|issue:8673", + "issue:5006|issue:8233", + "issue:7210|issue:8752", + "issue:3481|issue:7421", + "issue:9306|issue:9539", + "issue:10798|issue:12110", + "issue:3026|issue:6821", + "issue:1353|issue:1377", + "issue:1633|issue:3287" + ], + "prepared_review_unit_hash": "463174fec3258b59d97d408e29ebe7896641ac4bfd5c3ba6963e76d90ebc7f2b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33480, + "estimated_input_tokens": 8370, + "estimated_eval_tokens": 16996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose mix of unrelated diffusers issues spanning training, inference, loading, and model-specific bugs. None of the soft pairs look like true duplicates or the same underlying change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:12110 is the most concrete, clearly scoped bug report in the set, but it should be treated only as an anchor; the surrounding items are not duplicates of it.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5693", + "right": "issue:6250", + "accept": false, + "reason": "Both concern VAE loading, but one is a generic load error and the other is an SDXL single-file KeyError; different failure modes and code paths." + }, + { + "left": "issue:10917", + "right": "issue:7365", + "accept": false, + "reason": "A script-correctness question about Lumina-2.0 is unrelated to a specific pooled_prompt_embeds overwrite bug." + }, + { + "left": "issue:1273", + "right": "issue:1704", + "accept": false, + "reason": "These are separate conceptual questions with no shared concrete bug or fix." + }, + { + "left": "issue:2359", + "right": "issue:725", + "accept": false, + "reason": "One is about latent output_type behavior in img2img; the other is about device maps and loading memory footprint." + }, + { + "left": "issue:2578", + "right": "issue:4598", + "accept": false, + "reason": "Editable install failure and missing-dependency error messaging are different problems." + }, + { + "left": "issue:5769", + "right": "issue:8671", + "accept": false, + "reason": "Mixed-precision training correctness and redundant code cleanup are not the same bug/change." + }, + { + "left": "issue:7535", + "right": "issue:8612", + "accept": false, + "reason": "Type-hint incompatibility is unrelated to PixArt-Sigma mixed-precision inference issues." + }, + { + "left": "issue:5433", + "right": "issue:8792", + "accept": false, + "reason": "Both are image-quality complaints, but they describe different pipelines, settings, and likely different root causes." + }, + { + "left": "issue:6506", + "right": "issue:6535", + "accept": false, + "reason": "A return-type bug in StableVideoDiffusionPipeline is unrelated to scheduler customization support." + }, + { + "left": "issue:6790", + "right": "issue:9166", + "accept": false, + "reason": "Unexpected state_dict keys during LCM training has nothing to do with image_processor.apply_overlay behavior." + }, + { + "left": "issue:11437", + "right": "issue:12078", + "accept": false, + "reason": "Both mention validation/output during training, but one is missing artifacts and the other is a problematic example validation input." + }, + { + "left": "issue:2780", + "right": "issue:6812", + "accept": false, + "reason": "Both are LoRA training errors, but the FP16 unscale error and Q/K/V dtype mismatch are distinct failures." + }, + { + "left": "issue:12097", + "right": "issue:13105", + "accept": false, + "reason": "Different Wan models with different symptoms: end-of-run OOM versus output noise." + }, + { + "left": "issue:5112", + "right": "issue:5485", + "accept": false, + "reason": "A shape/type error in SDXL ControlNet Img2Img is not the same as ControlNet batch inference slowness." + }, + { + "left": "issue:5043", + "right": "issue:725", + "accept": false, + "reason": "A feature request for W\u00fcrstchen LoRA finetuning is unrelated to device-map memory loading support." + }, + { + "left": "issue:8534", + "right": "issue:8673", + "accept": false, + "reason": "Both are SD3-related, but wrong logit-normal weighting and missing attention masks are separate training issues." + }, + { + "left": "issue:5006", + "right": "issue:8233", + "accept": false, + "reason": "These are different precision-related problems in different pipelines: BF16 end-of-training OOM versus an fp16 depth pipeline error." + }, + { + "left": "issue:7210", + "right": "issue:8752", + "accept": false, + "reason": "Missing named_parameters with accelerate dispatch and offline from_single_file loading are different pipeline-loading issues." + }, + { + "left": "issue:3481", + "right": "issue:7421", + "accept": false, + "reason": "Different determinism complaints in different pipelines; not the same reproducible bug." + }, + { + "left": "issue:9306", + "right": "issue:9539", + "accept": false, + "reason": "Both are CogVideoX issues, but wrong device selection and float8 index_select errors are distinct." + }, + { + "left": "issue:10798", + "right": "issue:12110", + "accept": false, + "reason": "Flux quantization/device-combination issues are unrelated to UNet1DModel GaussianFourier initialization." + }, + { + "left": "issue:3026", + "right": "issue:6821", + "accept": false, + "reason": "An accelerate-launch error in instruct_pix2pix is unrelated to caption_column validation." + }, + { + "left": "issue:1353", + "right": "issue:1377", + "accept": false, + "reason": "Combining a training script with LDMPipline and a missing transformers import are unrelated." + }, + { + "left": "issue:1633", + "right": "issue:3287", + "accept": false, + "reason": "Both are scheduler-related feature requests, but they target different algorithms and changes; not the same concrete PR-sized fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 282, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10729", + "issue:10932", + "issue:10950", + "issue:11092", + "issue:1125", + "issue:11885", + "issue:12338", + "issue:13311", + "issue:1677", + "issue:1889", + "issue:2098", + "issue:2500", + "issue:2501", + "issue:2838", + "issue:2858", + "issue:3052", + "issue:3065", + "issue:3149", + "issue:3336", + "issue:3481", + "issue:3638", + "issue:3811", + "issue:3951", + "issue:4213", + "issue:4584", + "issue:4638", + "issue:4923", + "issue:5225", + "issue:5313", + "issue:5371", + "issue:5594", + "issue:5605", + "issue:5656", + "issue:579", + "issue:5890", + "issue:5947", + "issue:6136", + "issue:6722", + "issue:7053", + "issue:7730", + "issue:8463", + "issue:8577", + "issue:911", + "issue:9397", + "issue:9496", + "issue:9610", + "issue:962" + ], + "soft_pairs": [ + "issue:2500|issue:2501", + "issue:2838|issue:4923", + "issue:2858|issue:5605", + "issue:3052|issue:3638", + "issue:3065|issue:5371", + "issue:3811|issue:5313", + "issue:5225|issue:9496", + "issue:5594|issue:6722", + "issue:5605|issue:9397", + "issue:5947|issue:8577", + "issue:10932|issue:12338", + "issue:11885|issue:13311", + "issue:1677|issue:2098", + "issue:4584|issue:5890", + "issue:579|issue:911", + "issue:10729|issue:8463", + "issue:4213|issue:4638", + "issue:1125|issue:962", + "issue:1889|issue:3336", + "issue:5656|issue:9610", + "issue:7053|issue:7730", + "issue:10950|issue:11092", + "issue:3149|issue:3481", + "issue:3951|issue:6136" + ], + "prepared_review_unit_hash": "cb565fa66f429ce16021b0d1ca3a56751bff93a491aeb4e9be51f8a528604f4d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33263, + "estimated_input_tokens": 8316, + "estimated_eval_tokens": 16888 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is heterogeneous: most pairings are broad keyword/theme matches, not true duplicates. The soft edges span unrelated bugs, feature requests, and docs changes; none look like the same underlying fixable change. No PRs are present.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:3811 is the strongest standalone bug report here: it describes a concrete, user-visible LoRA loading failure with a clear symptom and likely actionable code path, even though it is not a duplicate of the other items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2500", + "right": "issue:2501", + "accept": false, + "reason": "Specific meta-tensor copy failure vs broad multi-test failures; different root causes and scope." + }, + { + "left": "issue:2838", + "right": "issue:4923", + "accept": false, + "reason": "ControlNet pipeline behavior request is unrelated to AudioLDM2 transcript/text-to-speech support." + }, + { + "left": "issue:2858", + "right": "issue:5605", + "accept": false, + "reason": "Seed image support for unCLIP img2img and per-image captions for DreamBooth are different features." + }, + { + "left": "issue:3052", + "right": "issue:3638", + "accept": false, + "reason": "Versatile Diffusion control request and a new prompt-free diffusion pipeline are not the same change." + }, + { + "left": "issue:3065", + "right": "issue:5371", + "accept": false, + "reason": "DreamBooth LoRA text-encoder tuning and BLIP inpainting support are unrelated feature requests." + }, + { + "left": "issue:3811", + "right": "issue:5313", + "accept": false, + "reason": "Both mention LoRA, but one is loading having no effect and the other is about non-revertible fuse; different code paths." + }, + { + "left": "issue:5225", + "right": "issue:9496", + "accept": false, + "reason": "SDXL memory exhaustion and SD3ControlNet forward errors are distinct model-specific bugs." + }, + { + "left": "issue:5594", + "right": "issue:6722", + "accept": false, + "reason": "SDXL LoRA fine-tuning memory usage vs using SigLIP as a text encoder are unrelated topics." + }, + { + "left": "issue:5605", + "right": "issue:9397", + "accept": false, + "reason": "DreamBooth per-image captions and SGMUniform scheduler support for Hyper LoRAs are different requests." + }, + { + "left": "issue:5947", + "right": "issue:8577", + "accept": false, + "reason": "Diffusion-DPO policy objective and Differential Diffusion in SD3 do not share the same underlying bug/change." + }, + { + "left": "issue:10932", + "right": "issue:12338", + "accept": false, + "reason": "torch._dynamo recompile_limit on HunyuanVideo compile is unrelated to AutoencoderDC slicing multiple-image decoding." + }, + { + "left": "issue:11885", + "right": "issue:13311", + "accept": false, + "reason": "Both are Flux LoRA-loading issues, but they target different model generations and symptoms; not the same concrete bug." + }, + { + "left": "issue:1677", + "right": "issue:2098", + "accept": false, + "reason": "Wrong input type to StableDiffusionPipeline and invalid torch.randn arguments are different runtime errors." + }, + { + "left": "issue:4584", + "right": "issue:5890", + "accept": false, + "reason": "resume_from_checkpoint in an SDXL LoRA training script is unrelated to a token-length warning in a different training script." + }, + { + "left": "issue:579", + "right": "issue:911", + "accept": false, + "reason": "Missing textual inversion config.json and DreamBooth bad config are different configuration failures." + }, + { + "left": "issue:10729", + "right": "issue:8463", + "accept": false, + "reason": "_hf_hook deletion AttributeError and NoneType.tokenize in InstructP2P-SDXL are unrelated failures." + }, + { + "left": "issue:4213", + "right": "issue:4638", + "accept": false, + "reason": "from_single_file() without accelerate and load_textual_inversion on SDXL are separate issues." + }, + { + "left": "issue:1125", + "right": "issue:962", + "accept": false, + "reason": "Inpaint pipeline runtime error and size support limitation are related to inpainting but not the same bug." + }, + { + "left": "issue:1889", + "right": "issue:3336", + "accept": false, + "reason": "Different scheduler variants in the same family, but not the same concrete change or bug." + }, + { + "left": "issue:5656", + "right": "issue:9610", + "accept": false, + "reason": "Adding a LoRA loader to community pipelines vs adding load_lora_weights to a specific SD3 img2img pipeline are different feature requests." + }, + { + "left": "issue:7053", + "right": "issue:7730", + "accept": false, + "reason": "UNet3D class conditioning and request timeout handling are unrelated issues." + }, + { + "left": "issue:10950", + "right": "issue:11092", + "accept": false, + "reason": "Broken docs and empty timesteps in SDXL-Turbo img2img are unrelated." + }, + { + "left": "issue:3149", + "right": "issue:3481", + "accept": false, + "reason": "Both discuss determinism, but one is a version-wide nondeterminism regression while the other is expected step-dependent image differences." + }, + { + "left": "issue:3951", + "right": "issue:6136", + "accept": false, + "reason": "NaNs in UniDiffuser demo and NaNs in StableDiffusionPipeline on Mac CPU are different models/code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 283, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10191", + "issue:10334", + "issue:11047", + "issue:11148", + "issue:11275", + "issue:11486", + "issue:12104", + "issue:1314", + "issue:13286", + "issue:1709", + "issue:1919", + "issue:1923", + "issue:1931", + "issue:2020", + "issue:2044", + "issue:2068", + "issue:2370", + "issue:2844", + "issue:3381", + "issue:3481", + "issue:4149", + "issue:4353", + "issue:4361", + "issue:4530", + "issue:4631", + "issue:4668", + "issue:4913", + "issue:5374", + "issue:5578", + "issue:5903", + "issue:5969", + "issue:6092", + "issue:6359", + "issue:7021", + "issue:7636", + "issue:7974", + "issue:8441", + "issue:8493", + "issue:8509", + "issue:8514", + "issue:8609", + "issue:8894", + "issue:8960", + "issue:9154", + "issue:9218", + "issue:9343", + "issue:9768", + "issue:9841" + ], + "soft_pairs": [ + "issue:4353|issue:5903", + "issue:4530|issue:5969", + "issue:11047|issue:9218", + "issue:11486|issue:9841", + "issue:1919|issue:1923", + "issue:10191|issue:11275", + "issue:1314|issue:2844", + "issue:9154|issue:9343", + "issue:2044|issue:2068", + "issue:3481|issue:4149", + "issue:7636|issue:8609", + "issue:8493|issue:8960", + "issue:1709|issue:3381", + "issue:1931|issue:4631", + "issue:2020|issue:5374", + "issue:4668|issue:7974", + "issue:4913|issue:8509", + "issue:5578|issue:6359", + "issue:11148|issue:8514", + "issue:12104|issue:13286", + "issue:2370|issue:6092", + "issue:7021|issue:8441", + "issue:4361|issue:8894", + "issue:10334|issue:9768" + ], + "prepared_review_unit_hash": "a43f9acef4788b8ed0fe44fb59e665d97f29e961bc8cfa144006a5d3a2d0a772", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34129, + "estimated_input_tokens": 8533, + "estimated_eval_tokens": 17322 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a mixed bag of unrelated diffusers bugs/features across training, loading, scheduler, and environment issues. None of the proposed soft pairs looks like the same underlying bug or change, so no duplicate merges are justified.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue cleanly subsumes the cluster; the items span distinct problems rather than one shared bug.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11148 is one of the clearest, most concrete bug reports in the set, but it still is not representative enough to serve as a true umbrella issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4353", + "right": "issue:5903", + "accept": false, + "reason": "Both concern image quality, but 4353 is a broad SDXL quality complaint while 5903 is specifically about edge artifacts from non-divisible resolutions." + }, + { + "left": "issue:4530", + "right": "issue:5969", + "accept": false, + "reason": "One reports SDXL refiner color artifacts; the other is a scheduler NaN regression with ZSNR patch. Different failure modes." + }, + { + "left": "issue:11047", + "right": "issue:9218", + "accept": false, + "reason": "Both mention PAG, but 11047 is about a missing controlnet inpaint model in the pipeline, while 9218 is about PAG with guess mode. Not the same bug." + }, + { + "left": "issue:11486", + "right": "issue:9841", + "accept": false, + "reason": "One asks for xformers support in training; the other is a DreamBooth SDXL LoRA TypeError around use_dora=False. Different code paths." + }, + { + "left": "issue:1919", + "right": "issue:1923", + "accept": false, + "reason": "Unrelated setup problems: missing Python dependency for a DreamBooth example versus missing CUDA libraries in a Dockerfile." + }, + { + "left": "issue:10191", + "right": "issue:11275", + "accept": false, + "reason": "GPU device selection in training is unrelated to the VAE NoneType addition error in ResnetUpsampleBlock2D." + }, + { + "left": "issue:1314", + "right": "issue:2844", + "accept": false, + "reason": "xformers backward error on Windows is unrelated to explicitly passing image embeddings to StableUnCLIPImg2ImgPipeline." + }, + { + "left": "issue:9154", + "right": "issue:9343", + "accept": false, + "reason": "These are different loading bugs: PT VAE loading versus FLUX loading with low_cpu_mem_usage=False and ignore_mismatched_sizes=True." + }, + { + "left": "issue:2044", + "right": "issue:2068", + "accept": false, + "reason": "Both involve dtype wording, but one is a DDP wrapper missing dtype and the other is from_pretrained not converting model dtype. Not the same issue." + }, + { + "left": "issue:3481", + "right": "issue:4149", + "accept": false, + "reason": "Determinism across inference steps is a different problem from batch prompt/negative-prompt dimension mismatch." + }, + { + "left": "issue:7636", + "right": "issue:8609", + "accept": false, + "reason": "ControlNet SDXL tile usage and custom_pipeline HD painter with IP-Adapter are separate pipeline integration questions." + }, + { + "left": "issue:8493", + "right": "issue:8960", + "accept": false, + "reason": "State-dict loading for SD3Transformer2DModel is unrelated to gradient accumulation being nonfunctional in the basic training example." + }, + { + "left": "issue:1709", + "right": "issue:3381", + "accept": false, + "reason": "ONNX tensor astype error and inpaint batch discoloration are different bugs with different symptoms and code paths." + }, + { + "left": "issue:1931", + "right": "issue:4631", + "accept": false, + "reason": "Float16 mismatch in Stable Diffusion 2 is unrelated to DPMSolverSinglestepScheduler scalar conversion with Karras sigmas." + }, + { + "left": "issue:2020", + "right": "issue:5374", + "accept": false, + "reason": "Conversion-script failure and unexpected checkpoint_path argument are separate API/script issues." + }, + { + "left": "issue:4668", + "right": "issue:7974", + "accept": false, + "reason": "Weird LoRA images and inpaint with IP-Adapter are different generation scenarios and likely different root causes." + }, + { + "left": "issue:4913", + "right": "issue:8509", + "accept": false, + "reason": "OpenPose SDXL model topic versus ControlNet++ are not the same concrete bug or fix." + }, + { + "left": "issue:5578", + "right": "issue:6359", + "accept": false, + "reason": "Missing model_index.json and missing get_velocity on EulerDiscreteScheduler are unrelated loader versus scheduler API problems." + }, + { + "left": "issue:11148", + "right": "issue:8514", + "accept": false, + "reason": "PEFT adapter state handling and EMAModel parallel fine-tuning error are distinct training failures." + }, + { + "left": "issue:12104", + "right": "issue:13286", + "accept": false, + "reason": "Out-of-bounds indexing and torchao quantization support are unrelated errors." + }, + { + "left": "issue:2370", + "right": "issue:6092", + "accept": false, + "reason": "A question about whether LoRA training fits on 24 GB GPU is not the same as a LoRA example that does not train anything." + }, + { + "left": "issue:7021", + "right": "issue:8441", + "accept": false, + "reason": "Both involve HF_HOME, but one is an import error and the other is infinite recursion from relative paths. Different bugs." + }, + { + "left": "issue:4361", + "right": "issue:8894", + "accept": false, + "reason": "Shared-tensors safetensors save issue and train_cm_ct_unconditional failure are unrelated." + }, + { + "left": "issue:10334", + "right": "issue:9768", + "accept": false, + "reason": "MacOS/MPS greys and CPU NaNs in Sana are not the same as a torch 2.5 CuDNN SDPA NaN backend issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 284, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10790", + "issue:11000", + "issue:11136", + "issue:11162", + "issue:11163", + "issue:11661", + "issue:12012", + "issue:12047", + "issue:124", + "issue:12533", + "issue:12540", + "issue:13506", + "issue:1611", + "issue:1791", + "issue:2107", + "issue:2262", + "issue:2674", + "issue:2844", + "issue:2961", + "issue:3171", + "issue:3494", + "issue:3883", + "issue:3948", + "issue:4080", + "issue:4134", + "issue:4159", + "issue:4324", + "issue:4403", + "issue:455", + "issue:4589", + "issue:4720", + "issue:5225", + "issue:5593", + "issue:6157", + "issue:6320", + "issue:6807", + "issue:7142", + "issue:7456", + "issue:7535", + "issue:7842", + "issue:8234", + "issue:843", + "issue:852", + "issue:8577", + "issue:8950", + "issue:9298", + "issue:9511" + ], + "soft_pairs": [ + "issue:1611|issue:852", + "issue:3494|issue:7142", + "issue:6320|issue:9298", + "issue:1791|issue:455", + "issue:2674|issue:4159", + "issue:4324|issue:8234", + "issue:4403|issue:6157", + "issue:4720|issue:7535", + "issue:5593|issue:6807", + "issue:11661|issue:8950", + "issue:2844|issue:5225", + "issue:4134|issue:7842", + "issue:12047|issue:13506", + "issue:10790|issue:9511", + "issue:11000|issue:7456", + "issue:11136|issue:12540", + "issue:11162|issue:11163", + "issue:12012|issue:12533", + "issue:124|issue:2961", + "issue:2107|issue:3948", + "issue:2262|issue:3883", + "issue:3171|issue:843", + "issue:4080|issue:4589", + "issue:4080|issue:8577" + ], + "prepared_review_unit_hash": "6708d58f6623f008b9179aa31c4591cc03d45ef10a61989056ec32584fd7a94b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32953, + "estimated_input_tokens": 8239, + "estimated_eval_tokens": 16734 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly broad topical neighbors, not true duplicates. The soft pairs share themes like pipelines, offloading, MPS, VRAM, or Wan/Sana, but each points to a different bug, compatibility issue, or feature request.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue: the candidate pairs are heterogeneous and only loosely related by subsystem/theme, not by the same underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11162 is the most concrete and representative item here, centered on a specific WanImageToVideoPipeline offloading problem; it is actionable, but not the same bug as the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1611", + "right": "issue:852", + "accept": false, + "reason": "Both mention older model/runtime compatibility on MPS, but one is a pipeline/model regression and the other is a device_map keyword error; different failure modes." + }, + { + "left": "issue:3494", + "right": "issue:7142", + "accept": false, + "reason": "Both are image-quality complaints, but one is Multidiffusion corruption across schedulers and the other is SDXL-lightning output mismatch vs webui." + }, + { + "left": "issue:6320", + "right": "issue:9298", + "accept": false, + "reason": "Remote model lookup failure and Float8 storage lookup error are unrelated bugs." + }, + { + "left": "issue:1791", + "right": "issue:455", + "accept": false, + "reason": "Both involve MPS, but one is a pipeline not supported on Mac and the other is an aten::index Tensor backend error; not the same bug." + }, + { + "left": "issue:2674", + "right": "issue:4159", + "accept": false, + "reason": "One is a LoRA packaging/requirements issue; the other is DiffEdit extreme VRAM usage. Same broad area, different problem." + }, + { + "left": "issue:4324", + "right": "issue:8234", + "accept": false, + "reason": "A NoneType iteration crash in UNet forward is unrelated to ip_adapter_scale dict/Tensor multiplication." + }, + { + "left": "issue:4403", + "right": "issue:6157", + "accept": false, + "reason": "One is a torch half-precision LayerNorm backend limitation; the other is ip_adapter not working in a controlnet img2img pipeline." + }, + { + "left": "issue:4720", + "right": "issue:7535", + "accept": false, + "reason": "AutoencoderTiny pipeline incompatibility is not the same as a FloatTensor type-hint mismatch." + }, + { + "left": "issue:5593", + "right": "issue:6807", + "accept": false, + "reason": "AutoPipeline.from_pipe failing on ControlNet pipelines is unrelated to MPS test collection failures." + }, + { + "left": "issue:11661", + "right": "issue:8950", + "accept": false, + "reason": "Training-step None handling in flux-control examples is unrelated to SD3 LoRA training bugs." + }, + { + "left": "issue:2844", + "right": "issue:5225", + "accept": false, + "reason": "StableUnCLIP image embedding runtime error and SDXL forward-pass VRAM exhaustion are different issues." + }, + { + "left": "issue:4134", + "right": "issue:7842", + "accept": false, + "reason": "MPS ControlNet tensor-size mismatch and DeepFloyd size-mismatch on non-multiple-of-8 images are separate shape bugs." + }, + { + "left": "issue:12047", + "right": "issue:13506", + "accept": false, + "reason": "LoRA fusion on Wan2.2 GGUF and Sage attention/ring-attention failures are unrelated." + }, + { + "left": "issue:10790", + "right": "issue:9511", + "accept": false, + "reason": "Both concern ControlNet, but one is CPU offload VRAM usage and the other is multi-ControlNet batching support for a specific pipeline." + }, + { + "left": "issue:11000", + "right": "issue:7456", + "accept": false, + "reason": "Missing shift argument in WanImageToVideoPipeline and a generic missing raise statement are not the same bug." + }, + { + "left": "issue:11136", + "right": "issue:12540", + "accept": false, + "reason": "SanaSprint intermediate timesteps support and SanaCombinedTimestepGuidanceEmbeddings incompatibility are different Sana pipeline issues." + }, + { + "left": "issue:11162", + "right": "issue:11163", + "accept": false, + "reason": "Both are WanImageToVideoPipeline bugs, but one is about offloading support and the other is broken latent-preparation math; different code paths." + }, + { + "left": "issue:12012", + "right": "issue:12533", + "accept": false, + "reason": "First is a Wan 2.2 first-block-cache context error; second is a hooks conflict between context parallelism and CPU offload." + }, + { + "left": "issue:124", + "right": "issue:2961", + "accept": false, + "reason": "A dataset creation documentation request and a StableUnCLIP training example request are related to docs, not the same ask." + }, + { + "left": "issue:2107", + "right": "issue:3948", + "accept": false, + "reason": "Low-resolution glitchy/black images and pale generated images are different rendering/color issues." + }, + { + "left": "issue:2262", + "right": "issue:3883", + "accept": false, + "reason": "A safetensors LoRA usage question is unrelated to a realistic-vision dreambooth compatibility issue." + }, + { + "left": "issue:3171", + "right": "issue:843", + "accept": false, + "reason": "LPW prompt support for controlnet and prompt syntax documentation are different documentation/features." + }, + { + "left": "issue:4080", + "right": "issue:4589", + "accept": false, + "reason": "One requests SDXL inpainting support; the other concerns XL ControlNet img2img/inpaint pipeline variants. Related area, but not the same change." + }, + { + "left": "issue:4080", + "right": "issue:8577", + "accept": false, + "reason": "Inpainting pipeline request and Differential Diffusion for SD3 are distinct features." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 285, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10191", + "issue:1061", + "issue:11962", + "issue:175", + "issue:1896", + "issue:2024", + "issue:2392", + "issue:2439", + "issue:2772", + "issue:3194", + "issue:351", + "issue:4122", + "issue:4194", + "issue:4213", + "issue:4529", + "issue:4557", + "issue:4593", + "issue:4648", + "issue:4885", + "issue:5112", + "issue:5371", + "issue:5463", + "issue:5703", + "issue:5972", + "issue:5989", + "issue:6001", + "issue:6320", + "issue:6793", + "issue:7041", + "issue:7054", + "issue:7191", + "issue:7616", + "issue:7681", + "issue:775", + "issue:7757", + "issue:7802", + "issue:8523", + "issue:8546", + "issue:8551", + "issue:8556", + "issue:8943", + "issue:9134", + "issue:9181", + "issue:9248", + "issue:9331", + "issue:9613", + "issue:9731", + "issue:9941" + ], + "soft_pairs": [ + "issue:4194|issue:6001", + "issue:4648|issue:8943", + "issue:5371|issue:7802", + "issue:5989|issue:9248", + "issue:6320|issue:8551", + "issue:3194|issue:4557", + "issue:9613|issue:9731", + "issue:4885|issue:5112", + "issue:5703|issue:9181", + "issue:7191|issue:9331", + "issue:7681|issue:9941", + "issue:2772|issue:5463", + "issue:4122|issue:7041", + "issue:7616|issue:8523", + "issue:10191|issue:7757", + "issue:11962|issue:9134", + "issue:2392|issue:6793", + "issue:2024|issue:351", + "issue:4213|issue:775", + "issue:4529|issue:7054", + "issue:4593|issue:5972", + "issue:8546|issue:8556", + "issue:1896|issue:2439", + "issue:1061|issue:175" + ], + "prepared_review_unit_hash": "4afa09ffa1cb0d6bdf218283e3a611f41c2aeef4bfaa615d5dfb18a848f33ce7", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34193, + "estimated_input_tokens": 8549, + "estimated_eval_tokens": 17354 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a false positive: the items span many unrelated bugs, feature requests, and docs/training questions across different pipelines and subsystems. No pull requests are present. A single SD3 from_single_file pair is superficially close, but still looks like a feature-gap report versus a concrete loading error, not the same underlying defect.", + "confidence": 0.91, + "canonical_issue_reason": "issue:8551 is the most concrete, actionable bug report in the set: it names a specific pipeline, a specific device_map setting, and a clear incorrect device-transfer behavior for VAE image latents.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8551 is the strongest standalone issue for triage because it is narrowly scoped, reproducible, and tied to a specific code path rather than a broad support request or vague runtime error.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4194", + "right": "issue:6001", + "accept": false, + "reason": "Both mention from_single_file, but one is a VAE-loading failure and the other is a refiner returning the wrong pipeline class; different symptoms and likely different fixes." + }, + { + "left": "issue:4648", + "right": "issue:8943", + "accept": false, + "reason": "Missing config.json in a model directory is unrelated to a torch shared-object import/linker failure during installation." + }, + { + "left": "issue:5371", + "right": "issue:7802", + "accept": false, + "reason": "One is a request for BLIP inpainting support, the other is a scheduler support request; different features and code paths." + }, + { + "left": "issue:5989", + "right": "issue:9248", + "accept": false, + "reason": "A safetensors.torch bug and a pip install upgrade request are not the same underlying issue." + }, + { + "left": "issue:6320", + "right": "issue:8551", + "accept": false, + "reason": "Remote-model lookup failure for Emu2 is unrelated to SDXL inpaint image-latent device placement with device_map='balanced'." + }, + { + "left": "issue:3194", + "right": "issue:4557", + "accept": false, + "reason": "bitsandbytes package metadata missing is unrelated to a super().__getattr__ AttributeError." + }, + { + "left": "issue:9613", + "right": "issue:9731", + "accept": false, + "reason": "A Timesteps.scale attribute error in 0.30.x is unrelated to a question about training LoRA on a custom dataset." + }, + { + "left": "issue:4885", + "right": "issue:5112", + "accept": false, + "reason": "LoRA weights not loading in one pipeline and a torch.Size float-type error in another are different bugs." + }, + { + "left": "issue:5703", + "right": "issue:9181", + "accept": false, + "reason": "DDIM inverse noise reconstruction and a ControlNet noise artifact are different problems and code paths." + }, + { + "left": "issue:7191", + "right": "issue:9331", + "accept": false, + "reason": "A strange import problem is not the same as a scheduler index_for_timestep bug." + }, + { + "left": "issue:7681", + "right": "issue:9941", + "accept": false, + "reason": "An OpenSSL/lib attribute import failure is unrelated to a generic stable diffusion colab runtime error." + }, + { + "left": "issue:2772", + "right": "issue:5463", + "accept": false, + "reason": "Local dataset validation in train_controlnet.py is unrelated to an UnboundLocalError on image_latents." + }, + { + "left": "issue:4122", + "right": "issue:7041", + "accept": false, + "reason": "SDXL timestep-count mismatch in DPMSolverMultistepScheduler is not the same as a segmentation fault when num_inference_steps > 1." + }, + { + "left": "issue:7616", + "right": "issue:8523", + "accept": false, + "reason": "Low-denoise img2img artifacts and a CUDA error in playground are too broad and do not indicate the same defect." + }, + { + "left": "issue:10191", + "right": "issue:7757", + "accept": false, + "reason": "Forcing the wrong GPU device and multi-GPU OOM are distinct training issues with different failure modes." + }, + { + "left": "issue:11962", + "right": "issue:9134", + "accept": false, + "reason": "GGUF quantized model support for FLUX.1-Kontext-dev is unrelated to issues with FLUX LoRAs trained by SimpleTuner." + }, + { + "left": "issue:2392", + "right": "issue:6793", + "accept": false, + "reason": "prompt_embeds handling in img2img is unrelated to Hydra serialization failures." + }, + { + "left": "issue:2024", + "right": "issue:351", + "accept": false, + "reason": "A corrupted notebook file and a CUDA-loading concern are unrelated." + }, + { + "left": "issue:4213", + "right": "issue:775", + "accept": false, + "reason": "from_single_file without accelerate and fp16/autocast behavior are different issues." + }, + { + "left": "issue:4529", + "right": "issue:7054", + "accept": false, + "reason": "An unexpected network_alpha keyword error and state_dict mutation in load_lora_weights are separate LoRA-loading bugs." + }, + { + "left": "issue:4593", + "right": "issue:5972", + "accept": false, + "reason": "Automatic1111 LoRA rank limits and Min-SNR training documentation are unrelated." + }, + { + "left": "issue:8546", + "right": "issue:8556", + "accept": false, + "reason": "Both involve SD3 from_single_file, but one is a missing-support feature request while the other is a specific meta-tensor copy error; not enough evidence they are the same bug." + }, + { + "left": "issue:1896", + "right": "issue:2439", + "accept": false, + "reason": "A CLIP-guided Stable Diffusion size/type error is unrelated to an inpainting pipeline loading issue." + }, + { + "left": "issue:1061", + "right": "issue:175", + "accept": false, + "reason": "A Textual Inversion example update request is unrelated to whether torch.manual_seed is used correctly in training examples." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 286, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10198", + "issue:1033", + "issue:10887", + "issue:11418", + "issue:1304", + "issue:1323", + "issue:1439", + "issue:1566", + "issue:175", + "issue:1801", + "issue:2024", + "issue:2068", + "issue:2202", + "issue:2775", + "issue:2883", + "issue:3184", + "issue:3207", + "issue:3377", + "issue:3700", + "issue:3730", + "issue:4210", + "issue:4312", + "issue:4708", + "issue:4888", + "issue:5037", + "issue:525", + "issue:5256", + "issue:5284", + "issue:5432", + "issue:5515", + "issue:5685", + "issue:5829", + "issue:5905", + "issue:5968", + "issue:6359", + "issue:7054", + "issue:7264", + "issue:7267", + "issue:7408", + "issue:7548", + "issue:8578", + "issue:8645", + "issue:8970", + "issue:8975", + "issue:9096", + "issue:9147", + "issue:9224", + "issue:9241" + ], + "soft_pairs": [ + "issue:10887|issue:7264", + "issue:1439|issue:1801", + "issue:2202|issue:5284", + "issue:2883|issue:7408", + "issue:4888|issue:5432", + "issue:5968|issue:9241", + "issue:6359|issue:8645", + "issue:7054|issue:8975", + "issue:1323|issue:2068", + "issue:1566|issue:5829", + "issue:8970|issue:9224", + "issue:3700|issue:5037", + "issue:7548|issue:9096", + "issue:5515|issue:5685", + "issue:1304|issue:3377", + "issue:175|issue:3207", + "issue:2024|issue:525", + "issue:2775|issue:4210", + "issue:3730|issue:4312", + "issue:4708|issue:9147", + "issue:5256|issue:5905", + "issue:1033|issue:3184", + "issue:7267|issue:8578", + "issue:10198|issue:11418" + ], + "prepared_review_unit_hash": "26827a5cc0618eb64a53f670ba8dcde3253688219b261b67d1766d3d250ccb3d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33515, + "estimated_input_tokens": 8379, + "estimated_eval_tokens": 17014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: mostly unrelated bug reports, feature requests, and API questions across different models, pipelines, and training scripts. I do not see a credible duplicate core to unify them.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10887", + "right": "issue:7264", + "accept": false, + "reason": "Different bugs: one is an import error for `StableDiffusionXLTrainer`, the other is pipeline serialization/loading of custom transformers models." + }, + { + "left": "issue:1439", + "right": "issue:1801", + "accept": false, + "reason": "Unrelated errors: missing `FeatureExtractionMixin` in transformers vs `getattr()` receiving a non-string attribute name." + }, + { + "left": "issue:2202", + "right": "issue:5284", + "accept": false, + "reason": "Different feature requests for different subsystems: OpenVINO support vs ControlNet-LLLite support." + }, + { + "left": "issue:2883", + "right": "issue:7408", + "accept": false, + "reason": "Both are pipeline requests, but for different models/use cases; not the same underlying change." + }, + { + "left": "issue:4888", + "right": "issue:5432", + "accept": false, + "reason": "Different issues: offline safetensors loading vs blank validation images in the SDXL LoRA training script." + }, + { + "left": "issue:5968", + "right": "issue:9241", + "accept": false, + "reason": "Both involve conversion tooling, but they target different formats and model workflows; not the same code-path problem." + }, + { + "left": "issue:6359", + "right": "issue:8645", + "accept": false, + "reason": "Separate bugs: scheduler `get_velocity`/`add_noise` behavior vs a missing utility function in `single_file_utils.py`." + }, + { + "left": "issue:7054", + "right": "issue:8975", + "accept": false, + "reason": "Different failures: mutating the passed LoRA state dict vs an import error for `LoraLoaderMixin`." + }, + { + "left": "issue:1323", + "right": "issue:2068", + "accept": false, + "reason": "Related dtype theme, but different root causes: missing dtype on Flax model load vs `from_pretrained` not converting dtype." + }, + { + "left": "issue:1566", + "right": "issue:5829", + "accept": false, + "reason": "Different training crashes: fp16 checkpoint save failure in Dreambooth vs missing `unet_time_cond_proj_dim` for LCM scripts." + }, + { + "left": "issue:8970", + "right": "issue:9224", + "accept": false, + "reason": "Both mention MPS, but one is a black-image output bug and the other is an unsupported hardcoded float64 dtype." + }, + { + "left": "issue:3700", + "right": "issue:5037", + "accept": false, + "reason": "Distinct script issues: `min_snr` argument handling vs a validation-image path bug." + }, + { + "left": "issue:7548", + "right": "issue:9096", + "accept": false, + "reason": "Different inference problems: AnimateDiff flicker vs Flux half-precision NaNs." + }, + { + "left": "issue:5515", + "right": "issue:5685", + "accept": false, + "reason": "Scheduler-related, but not the same bug; one is missing scheduler initialization, the other is a broad scheduler-issues report." + }, + { + "left": "issue:1304", + "right": "issue:3377", + "accept": false, + "reason": "A training how-to question vs an SNR-weighting optimization question; no shared bug or change." + }, + { + "left": "issue:175", + "right": "issue:3207", + "accept": false, + "reason": "Unrelated topics: seed usage in example training code vs attention implementation compatibility with torch < 2.0." + }, + { + "left": "issue:2024", + "right": "issue:525", + "accept": false, + "reason": "Different notebook problems: corrupted notebook file vs loading images from local disk." + }, + { + "left": "issue:2775", + "right": "issue:4210", + "accept": false, + "reason": "Separate concerns: `torch._dynamo.optimize` misuse vs `attention_mask` usage in `UNet2DConditionModel`." + }, + { + "left": "issue:3730", + "right": "issue:4312", + "accept": false, + "reason": "Both touch inpainting, but one is a timestep blending bug and the other is a feature request for SAG in inpainting/img2img." + }, + { + "left": "issue:4708", + "right": "issue:9147", + "accept": false, + "reason": "Import error in `diffusers.utils` vs Flux autopipeline error; different code paths and symptoms." + }, + { + "left": "issue:5256", + "right": "issue:5905", + "accept": false, + "reason": "Both are feature requests, but for different models/training support; not the same underlying change." + }, + { + "left": "issue:1033", + "right": "issue:3184", + "accept": false, + "reason": "Different failures in different pipelines: Stable Diffusion Mega index error vs `lpw_stable_diffusion` generator-list handling." + }, + { + "left": "issue:7267", + "right": "issue:8578", + "accept": false, + "reason": "Unrelated bugs: local data loading in `train_controlnet.py` vs meta-tensor copy failure with `sd_turbo 1.5`." + }, + { + "left": "issue:10198", + "right": "issue:11418", + "accept": false, + "reason": "Flux weight mismatch vs how to add a Flux fp8 safetensors file; one is a bug, the other is a usage question." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 287, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10052", + "issue:10448", + "issue:10715", + "issue:11023", + "issue:11036", + "issue:11351", + "issue:11436", + "issue:11580", + "issue:1160", + "issue:11648", + "issue:12613", + "issue:12632", + "issue:12752", + "issue:13256", + "issue:13350", + "issue:13429", + "issue:1677", + "issue:1717", + "issue:2228", + "issue:2518", + "issue:2737", + "issue:3363", + "issue:3612", + "issue:3767", + "issue:4305", + "issue:4419", + "issue:4559", + "issue:4605", + "issue:4826", + "issue:4921", + "issue:5022", + "issue:5070", + "issue:5143", + "issue:5245", + "issue:5433", + "issue:5574", + "issue:5897", + "issue:6079", + "issue:6183", + "issue:6359", + "issue:6661", + "issue:7053", + "issue:7591", + "issue:7921", + "issue:8233", + "issue:8997", + "issue:9731" + ], + "soft_pairs": [ + "issue:11648|issue:8997", + "issue:10448|issue:11580", + "issue:11036|issue:11351", + "issue:1160|issue:5245", + "issue:2737|issue:4559", + "issue:3612|issue:7921", + "issue:4921|issue:5433", + "issue:3363|issue:5897", + "issue:4826|issue:8233", + "issue:1677|issue:2228", + "issue:13256|issue:5022", + "issue:10052|issue:6359", + "issue:10715|issue:11023", + "issue:12752|issue:13429", + "issue:1717|issue:3767", + "issue:2518|issue:5070", + "issue:4305|issue:5143", + "issue:4419|issue:7053", + "issue:4605|issue:6079", + "issue:5143|issue:5574", + "issue:12613|issue:13350", + "issue:6183|issue:7591", + "issue:6661|issue:9731", + "issue:11436|issue:12632" + ], + "prepared_review_unit_hash": "ed3eb69a81a13dde289569229efa7d499b55e06b20f700c7a80bd4f7130d52c0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33314, + "estimated_input_tokens": 8329, + "estimated_eval_tokens": 16914 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are not a true duplicate cluster; most soft pairs only overlap on generic terms like dtype, pipeline, or training, but describe different bugs or feature requests.", + "confidence": 0.95, + "canonical_issue_reason": "No clear canonical issue: the set is heterogeneous (loaders, schedulers, training, image artifacts, dtype mismatches, feature requests) and does not center on one underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a strong representative because there is no shared concrete failure mode across the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11648", + "right": "issue:8997", + "accept": false, + "reason": "Different problems: LoRA loading for an fp8 transformer model vs a scheduler error in Kolors." + }, + { + "left": "issue:10448", + "right": "issue:11580", + "accept": false, + "reason": "Unrelated topics: DDUF mmap loading vs FLUX fp8 support." + }, + { + "left": "issue:11036", + "right": "issue:11351", + "accept": false, + "reason": "Different code paths: latent-condition operations vs Wan i2v processor dtype behavior." + }, + { + "left": "issue:1160", + "right": "issue:5245", + "accept": false, + "reason": "Both mention freezes, but one is a pytest hang and the other is an import-stage pipenv freeze; not the same bug." + }, + { + "left": "issue:2737", + "right": "issue:4559", + "accept": false, + "reason": "Different feature requests: CLIP guidance support vs long weighted prompts." + }, + { + "left": "issue:3612", + "right": "issue:7921", + "accept": false, + "reason": "ControlNet creation/preprocessor request is not the same as IP-Adapter control_guidance_start/end usage." + }, + { + "left": "issue:4921", + "right": "issue:5433", + "accept": false, + "reason": "Both involve SDXL/refiner quality, but one is detail erasure and the other is scheduler artifacts without the refiner." + }, + { + "left": "issue:3363", + "right": "issue:5897", + "accept": false, + "reason": "Both are training failures, but they are different scripts and different root causes." + }, + { + "left": "issue:4826", + "right": "issue:8233", + "accept": false, + "reason": "Different pipelines and errors: MultiControlNet reference issues vs fp16 depth pipeline failure." + }, + { + "left": "issue:1677", + "right": "issue:2228", + "accept": false, + "reason": "Wrong data type on pipeline call is not the same as failing to load the pipe onto GPU." + }, + { + "left": "issue:13256", + "right": "issue:5022", + "accept": false, + "reason": "One is dependency/version compatibility, the other is an import NameError in UNet2DModel." + }, + { + "left": "issue:10052", + "right": "issue:6359", + "accept": false, + "reason": "Syntax error on ctx is unrelated to scheduler velocity/add_noise behavior." + }, + { + "left": "issue:10715", + "right": "issue:11023", + "accept": false, + "reason": "Different image-quality failures on different models and hardware; not one underlying bug." + }, + { + "left": "issue:12752", + "right": "issue:13429", + "accept": false, + "reason": "Feature request for Z-Image img2img/inpainting is not the same as adding JoyAI-Image-Edit." + }, + { + "left": "issue:1717", + "right": "issue:3767", + "accept": false, + "reason": "Offline cache loading and FrozenDict immutability are unrelated." + }, + { + "left": "issue:2518", + "right": "issue:5070", + "accept": false, + "reason": "Different example features: xformers support vs ControlNet flags." + }, + { + "left": "issue:4305", + "right": "issue:5143", + "accept": false, + "reason": "torch.compile graph breaks are a different issue from tiny numerical output differences." + }, + { + "left": "issue:4419", + "right": "issue:7053", + "accept": false, + "reason": "SDXL offset-noise inference behavior is unrelated to UNet3DConditionModel class conditioning." + }, + { + "left": "issue:4605", + "right": "issue:6079", + "accept": false, + "reason": "Inpaint long-prompt error is not the same as a generic Half-vs-Float dtype mismatch." + }, + { + "left": "issue:5143", + "right": "issue:5574", + "accept": false, + "reason": "Numerical output variance is unrelated to LoRA training of a subset of UNet attention parameters." + }, + { + "left": "issue:12613", + "right": "issue:13350", + "accept": false, + "reason": "Wan i2v NaNs and FLUX.2 batch inference are different model/runtime issues." + }, + { + "left": "issue:6183", + "right": "issue:7591", + "accept": false, + "reason": "Both use from_single_file, but one is a shape bug and the other is a local_files_only HeaderTooLarge error." + }, + { + "left": "issue:6661", + "right": "issue:9731", + "accept": false, + "reason": "OOM on MPS training at certain image sizes is different from how to train a Playground2.5 LoRA." + }, + { + "left": "issue:11436", + "right": "issue:12632", + "accept": false, + "reason": "Both are dtype mismatches, but they occur in different models and code paths (Hidream inference vs SDXL img2img)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 288, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10514", + "issue:10636", + "issue:1064", + "issue:10768", + "issue:10772", + "issue:1083", + "issue:11145", + "issue:11148", + "issue:1160", + "issue:12108", + "issue:12308", + "issue:12540", + "issue:1544", + "issue:1961", + "issue:1974", + "issue:2221", + "issue:2412", + "issue:2439", + "issue:2549", + "issue:2728", + "issue:2799", + "issue:2941", + "issue:3101", + "issue:3116", + "issue:3184", + "issue:3799", + "issue:4305", + "issue:4332", + "issue:4480", + "issue:4767", + "issue:487", + "issue:5046", + "issue:5274", + "issue:5436", + "issue:5570", + "issue:5578", + "issue:5595", + "issue:5630", + "issue:5760", + "issue:5951", + "issue:6016", + "issue:6560", + "issue:660", + "issue:7842", + "issue:8450", + "issue:8989", + "issue:984" + ], + "soft_pairs": [ + "issue:4480|issue:7842", + "issue:5630|issue:8450", + "issue:6016|issue:6560", + "issue:1974|issue:3184", + "issue:1064|issue:1961", + "issue:2549|issue:2728", + "issue:2941|issue:4767", + "issue:10636|issue:8989", + "issue:487|issue:660", + "issue:11148|issue:12108", + "issue:4332|issue:984", + "issue:5046|issue:5760", + "issue:5274|issue:5436", + "issue:1160|issue:5595", + "issue:1083|issue:3116", + "issue:4305|issue:5570", + "issue:2221|issue:2412", + "issue:1544|issue:3799", + "issue:3101|issue:5951", + "issue:2439|issue:2799", + "issue:10010|issue:5578", + "issue:10514|issue:12540", + "issue:10768|issue:11145", + "issue:10772|issue:12308" + ], + "prepared_review_unit_hash": "72774378900d22730f36a1956abdd3f7b80d9eb20d91e9e8624e03200965e06a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34059, + "estimated_input_tokens": 8515, + "estimated_eval_tokens": 17286 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a heterogeneous mix of Diffusers bugs/features across different models, schedulers, and platforms. I don\u2019t see any true duplicate pairs among the soft candidates; they are mostly thematically similar at a very high level but not the same underlying bug/change.", + "confidence": 0.78, + "canonical_issue_reason": "issue:2799 is the broadest, most generally phrased bug report in the set and is the best anchor if one issue must represent the cluster; the others are mostly narrower model-, pipeline-, or platform-specific failures.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:2799 is the most reusable representative of this mixed set because it targets a general checkpoint/pipeline mismatch problem rather than a narrow one-off failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4480", + "right": "issue:7842", + "accept": false, + "reason": "Both are runtime failures, but one is an M1 fp16/MPS crash and the other is a DeepFloyd image-size mismatch; different code paths and root causes." + }, + { + "left": "issue:5630", + "right": "issue:8450", + "accept": false, + "reason": "Both mention bad image output, but one is scheduler behavior in SDXL img2img and the other is inpainting strength=1.0 noise; not the same defect." + }, + { + "left": "issue:6016", + "right": "issue:6560", + "accept": false, + "reason": "Both involve size/shape errors, but one is conversion code and the other is LoRA loading/regression; different failing operations." + }, + { + "left": "issue:1974", + "right": "issue:3184", + "accept": false, + "reason": "Both mention generator lists, but one is Euler scheduler reproducibility and the other is a community pipeline input issue; not the same underlying bug." + }, + { + "left": "issue:1064", + "right": "issue:1961", + "accept": false, + "reason": "These are unrelated: an LMSDiscreteScheduler add_noise error versus a ColossalAI integration error." + }, + { + "left": "issue:2549", + "right": "issue:2728", + "accept": false, + "reason": "One is an SD2-based UNet incompatibility, the other is safetensors loading in from_pretrained; same model family, different failures." + }, + { + "left": "issue:2941", + "right": "issue:4767", + "accept": false, + "reason": "Both touch EulerAncestralDiscreteScheduler, but one reports poor image results in SD and the other asks for prediction_type=sample support; not a single concrete bug." + }, + { + "left": "issue:10636", + "right": "issue:8989", + "accept": false, + "reason": "Different components entirely: value_guided_sampling.py versus Stable Audio sequential offloading." + }, + { + "left": "issue:487", + "right": "issue:660", + "accept": false, + "reason": "Both are Apple/M1 float16-related, but one is CPU support and the other is MPS fp16 behavior; too different to treat as duplicates." + }, + { + "left": "issue:11148", + "right": "issue:12108", + "accept": false, + "reason": "PEFT adapter state tracking versus flow-matching scheduler breakage in Qwen/Chroma pipelines; unrelated bugs." + }, + { + "left": "issue:4332", + "right": "issue:984", + "accept": false, + "reason": "xformers support in Kandinsky2.2 is unrelated to F.interpolate breaking for large batch sizes." + }, + { + "left": "issue:5046", + "right": "issue:5760", + "accept": false, + "reason": "A no-code feature proposal for DragonDiffusion versus a video-to-audio synthesis paper entry; not duplicates." + }, + { + "left": "issue:5274", + "right": "issue:5436", + "accept": false, + "reason": "fp16 mixed-precision training error versus a ControlNet XL training-example failure; different scenarios and error sources." + }, + { + "left": "issue:1160", + "right": "issue:5595", + "accept": false, + "reason": "A hanging pytest versus a height/width divisibility validation error; unrelated." + }, + { + "left": "issue:1083", + "right": "issue:3116", + "accept": false, + "reason": "FP16 ONNX output correctness and KDPM2DiscreteScheduler img2img noise are different bugs." + }, + { + "left": "issue:4305", + "right": "issue:5570", + "accept": false, + "reason": "torch.compile graph breaks versus a missing compute_snr utility in packaging; unrelated code paths." + }, + { + "left": "issue:2221", + "right": "issue:2412", + "accept": false, + "reason": "Both are generic 'error' titles, but one is ColossalAI-related and the other is DreamBooth-related; no shared concrete bug." + }, + { + "left": "issue:1544", + "right": "issue:3799", + "accept": false, + "reason": "A missing get_velocity method at runtime versus an incorrect type inference issue; one is API availability, the other is typing." + }, + { + "left": "issue:3101", + "right": "issue:5951", + "accept": false, + "reason": "ControlNet example training error versus PixArt training discussion; too broad to be the same issue." + }, + { + "left": "issue:2439", + "right": "issue:2799", + "accept": false, + "reason": "Inpainting pipeline loading/running problems are not the same as explicitly raising on checkpoint/pipeline mismatch, even if they can overlap conceptually." + }, + { + "left": "issue:10010", + "right": "issue:5578", + "accept": false, + "reason": "Both mention model_index.json, but one is a missing remote HF file and the other is a local-folder OSError; different root causes." + }, + { + "left": "issue:10514", + "right": "issue:12540", + "accept": false, + "reason": "One is Sana resolution binning/sample_size support, the other is SanaCombinedTimestepGuidanceEmbeddings compatibility with SanaPipeline; distinct features/bugs." + }, + { + "left": "issue:10768", + "right": "issue:11145", + "accept": false, + "reason": "Hunyuan negative-prompt support and Flux Fill LoRA/dreambooth finetuning are unrelated model capabilities." + }, + { + "left": "issue:10772", + "right": "issue:12308", + "accept": false, + "reason": "Sana ControlNet support and flash-attn kernel support on non-Hopper GPUs are separate feature requests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 289, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11047", + "issue:11561", + "issue:1195", + "issue:12037", + "issue:1205", + "issue:12053", + "issue:1284", + "issue:1323", + "issue:13430", + "issue:1486", + "issue:1535", + "issue:1909", + "issue:1931", + "issue:201", + "issue:2414", + "issue:2844", + "issue:2857", + "issue:3162", + "issue:3198", + "issue:3441", + "issue:3660", + "issue:4075", + "issue:4217", + "issue:4308", + "issue:4416", + "issue:4436", + "issue:4483", + "issue:4789", + "issue:4911", + "issue:5034", + "issue:5515", + "issue:5598", + "issue:5633", + "issue:5645", + "issue:6028", + "issue:660", + "issue:6801", + "issue:6914", + "issue:7428", + "issue:7670", + "issue:7785", + "issue:8393", + "issue:8485", + "issue:8555", + "issue:9220", + "issue:9297", + "issue:9479", + "issue:9813" + ], + "soft_pairs": [ + "issue:11047|issue:9220", + "issue:11561|issue:13430", + "issue:1195|issue:3162", + "issue:12037|issue:12053", + "issue:1205|issue:4436", + "issue:1284|issue:1486", + "issue:1535|issue:2844", + "issue:1909|issue:4483", + "issue:201|issue:4416", + "issue:2857|issue:7428", + "issue:3198|issue:4911", + "issue:4075|issue:4308", + "issue:4217|issue:7785", + "issue:5598|issue:5633", + "issue:7428|issue:7670", + "issue:8393|issue:9297", + "issue:8485|issue:8555", + "issue:6914|issue:9813", + "issue:6028|issue:9479", + "issue:5515|issue:5645", + "issue:1931|issue:660", + "issue:1284|issue:5034", + "issue:1323|issue:4789", + "issue:2414|issue:6801", + "issue:3441|issue:3660" + ], + "prepared_review_unit_hash": "56c82977db5134f76302a91f0e29cd5de1b32b0dc1752c57a3e860b109aa007a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33665, + "estimated_input_tokens": 8417, + "estimated_eval_tokens": 17090 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated feature requests, docs/test issues, load-time regressions, and model-specific bugs across many subsystems. None of the soft pairs look like the same underlying defect or change, so there are no duplicate merges here.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue fits this cluster well; the items are too diverse and do not share one concrete bug or feature request.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9813 is the most detailed and actionable standalone bug report in the set, but it is still not a good duplicate anchor for the cluster as a whole.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11047", + "right": "issue:9220", + "accept": false, + "reason": "Both mention PAG, but one is about missing ControlNet PAG inpaint support and the other is an undefined variable in a PAG pipeline; different problems and likely different code paths." + }, + { + "left": "issue:11561", + "right": "issue:13430", + "accept": false, + "reason": "Both are feature requests around image-edit pipelines, but they target different models and APIs (FluxFill IP-Adapter loading vs JoyAI-Image-Edit support)." + }, + { + "left": "issue:1195", + "right": "issue:3162", + "accept": false, + "reason": "fp16 memory-efficient attention failures are unrelated to missing classifier-free-sampling embeddings; different bug classes." + }, + { + "left": "issue:12037", + "right": "issue:12053", + "accept": false, + "reason": "Different model families and failures: WanTransformer3D/Lightx2v LoRAs versus Flux1.Dev Kohya LoRA text-encoder layer support." + }, + { + "left": "issue:1205", + "right": "issue:4436", + "accept": false, + "reason": "ONNX conversion for SD models is a different workflow from converting CIVITAI safetensors with a VAE into diffusers." + }, + { + "left": "issue:1284", + "right": "issue:1486", + "accept": false, + "reason": "Longer sample time with bigger batches is a performance complaint, not the DDIMScheduler prediction_type='sample' correctness bug." + }, + { + "left": "issue:1535", + "right": "issue:2844", + "accept": false, + "reason": "Image Variation Pipeline support and explicit image-embedding errors in StableUnCLIPImg2Img are separate pipeline issues." + }, + { + "left": "issue:1909", + "right": "issue:4483", + "accept": false, + "reason": "Both involve generators, but one is about list-of-generators API support while the other is about SDXL nondeterminism; not the same bug." + }, + { + "left": "issue:201", + "right": "issue:4416", + "accept": false, + "reason": "These are unrelated test tasks for different components, not the same underlying change." + }, + { + "left": "issue:2857", + "right": "issue:7428", + "accept": false, + "reason": "Generic UniDiffuser support and layer diffusion support are different feature requests with different model targets." + }, + { + "left": "issue:3198", + "right": "issue:4911", + "accept": false, + "reason": "Safetensors textual inversion loading and custom text_encoder loading from_single_file are distinct loading paths." + }, + { + "left": "issue:4075", + "right": "issue:4308", + "accept": false, + "reason": "Both are docstring issues, but they concern different pipelines and different parameters." + }, + { + "left": "issue:4217", + "right": "issue:7785", + "accept": false, + "reason": "Slow safetensors loading and very slow first inference are performance problems in different stages of the workflow." + }, + { + "left": "issue:5598", + "right": "issue:5633", + "accept": false, + "reason": "Different internal modules and different bugs; no clear shared code-path defect." + }, + { + "left": "issue:7428", + "right": "issue:7670", + "accept": false, + "reason": "Layer diffusion support and IDDPM support are unrelated feature requests." + }, + { + "left": "issue:8393", + "right": "issue:9297", + "accept": false, + "reason": "Both mention load time, but one is about checkpoint shards and the other about loading multiple LoRAs; different objects and causes." + }, + { + "left": "issue:8485", + "right": "issue:8555", + "accept": false, + "reason": "Both are about SD3, but one asks for LoRA support while the other asks for FP8/text-encoder training; distinct capabilities." + }, + { + "left": "issue:6914", + "right": "issue:9813", + "accept": false, + "reason": "Both touch IP-Adapter, but one is an offload-related shape AttributeError and the other is batch feature mixing in prepare_ip_adapter_image_embeds; not the same concrete bug." + }, + { + "left": "issue:6028", + "right": "issue:9479", + "accept": false, + "reason": "Kandinsky 3.0 OOM and StableDiffusionInpaintPipeline masking errors are unrelated issues." + }, + { + "left": "issue:5515", + "right": "issue:5645", + "accept": false, + "reason": "Both involve LCM, but one is missing scheduler initialization and the other is a broader core support breakage; not a single mergeable fix." + }, + { + "left": "issue:1931", + "right": "issue:660", + "accept": false, + "reason": "Both are dtype-related, but one is SD2 float16 mismatch and the other is an MPS fp16 revision issue; different environments and root causes." + }, + { + "left": "issue:1284", + "right": "issue:5034", + "accept": false, + "reason": "Batch-size slowdown and progress-bar display bugs are unrelated." + }, + { + "left": "issue:1323", + "right": "issue:4789", + "accept": false, + "reason": "Flax dtype-loading and DeepFloyd IF download failures are separate problems." + }, + { + "left": "issue:2414", + "right": "issue:6801", + "accept": false, + "reason": "A missing transformers import and scheduler loading failure are unrelated." + }, + { + "left": "issue:3441", + "right": "issue:3660", + "accept": false, + "reason": "PyTorch 2 VRAM regressions without xformers and torch.compile incompatibilities with LoRA/xformers are related only broadly, not the same code-path bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 290, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10658", + "issue:10743", + "issue:1083", + "issue:12149", + "issue:12499", + "issue:12565", + "issue:1608", + "issue:1655", + "issue:1755", + "issue:1790", + "issue:2331", + "issue:2412", + "issue:2480", + "issue:2783", + "issue:2991", + "issue:3336", + "issue:3452", + "issue:3494", + "issue:3593", + "issue:3698", + "issue:3805", + "issue:3944", + "issue:3980", + "issue:4241", + "issue:4701", + "issue:4972", + "issue:499", + "issue:5050", + "issue:5120", + "issue:5166", + "issue:542", + "issue:543", + "issue:5687", + "issue:5761", + "issue:5772", + "issue:5905", + "issue:6141", + "issue:6688", + "issue:6928", + "issue:7053", + "issue:7240", + "issue:730", + "issue:9090", + "issue:9227", + "issue:9241", + "issue:988", + "issue:9902" + ], + "soft_pairs": [ + "issue:3494|issue:3980", + "issue:1083|issue:542", + "issue:2331|issue:6688", + "issue:2412|issue:5120", + "issue:5166|issue:7240", + "issue:1655|issue:499", + "issue:1655|issue:543", + "issue:10658|issue:6928", + "issue:10743|issue:9227", + "issue:3805|issue:730", + "issue:4701|issue:5761", + "issue:5687|issue:9090", + "issue:5772|issue:7053", + "issue:1608|issue:2991", + "issue:3593|issue:4972", + "issue:1790|issue:2480", + "issue:12149|issue:9902", + "issue:1755|issue:3336", + "issue:2783|issue:6141", + "issue:3452|issue:988", + "issue:3698|issue:5905", + "issue:3944|issue:4241", + "issue:5050|issue:9241", + "issue:12499|issue:12565" + ], + "prepared_review_unit_hash": "948532922f418b44cab882ce6974ff22b4963a0278aa289950c5eabfa8296079", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32619, + "estimated_input_tokens": 8155, + "estimated_eval_tokens": 16566 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a noisy mix of unrelated issue reports and feature requests; none of the soft-linked pairs look like the same underlying bug or change, so there is no clear canonical duplicate target.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3494", + "right": "issue:3980", + "accept": false, + "reason": "Both concern generation quality, but one is about Multidiffusion/scheduler corruption and the other about long-prompt behavior in A1111; different bugs and code paths." + }, + { + "left": "issue:1083", + "right": "issue:542", + "accept": false, + "reason": "Both mention ONNX, but one reports incorrect FP16 output while the other is an export issue for UNet; not the same failure mode." + }, + { + "left": "issue:2331", + "right": "issue:6688", + "accept": false, + "reason": "Different feature requests: adding a ControlNet pipeline vs adding MotionCtrl support." + }, + { + "left": "issue:2412", + "right": "issue:5120", + "accept": false, + "reason": "Both are DreamBooth-related, but one is a generic error and the other is a prior-preservation regression; not clearly the same bug." + }, + { + "left": "issue:5166", + "right": "issue:7240", + "accept": false, + "reason": "Different pipelines/models and different symptoms: missing VQ-VAE scaling in unconditional LDM vs low-quality output when converting ImageNet conditional LDM." + }, + { + "left": "issue:1655", + "right": "issue:499", + "accept": false, + "reason": "Pipeline initialization failure vs example training script failure; too different to be the same issue." + }, + { + "left": "issue:1655", + "right": "issue:543", + "accept": false, + "reason": "Pipeline init problem and a DDP stride warning are unrelated." + }, + { + "left": "issue:10658", + "right": "issue:6928", + "accept": false, + "reason": "Completely different topics: ONNX Runtime provider options vs Marigold depthmap colorization setting." + }, + { + "left": "issue:10743", + "right": "issue:9227", + "accept": false, + "reason": "Both are FLUX training-related, but one asks for ZeRO-3 support and the other reports a runtime problem with a specific FLUX model; not the same concrete bug." + }, + { + "left": "issue:3805", + "right": "issue:730", + "accept": false, + "reason": "Colab execution error vs DreamBooth learning behavior differs; these are not the same problem." + }, + { + "left": "issue:4701", + "right": "issue:5761", + "accept": false, + "reason": "AutoencoderKL CUDA/autograd error vs consistency decoder cost question; unrelated." + }, + { + "left": "issue:5687", + "right": "issue:9090", + "accept": false, + "reason": "Both mention repeated timesteps, but one is Karras-sigma/img2img low-strength behavior and the other is a PNDM scheduler timestep bug; too different to merge." + }, + { + "left": "issue:5772", + "right": "issue:7053", + "accept": false, + "reason": "Webdataset performance question vs UNet3D class-conditioning bug; unrelated." + }, + { + "left": "issue:1608", + "right": "issue:2991", + "accept": false, + "reason": "Training stopping at 500 steps on T4 and a mixed-precision ControlNet example failure are different issues." + }, + { + "left": "issue:3593", + "right": "issue:4972", + "accept": false, + "reason": "LoRA+xformers second-call failure vs wuerstchen example failing to run; different code paths and symptoms." + }, + { + "left": "issue:1790", + "right": "issue:2480", + "accept": false, + "reason": "A general 'how does training work' question is not the same as a specific checkpoint-folder bug in DreamBooth." + }, + { + "left": "issue:12149", + "right": "issue:9902", + "accept": false, + "reason": "Both are version-related complaints, but one is diffusers package/source mismatch and the other is a conda version pinning issue; unrelated." + }, + { + "left": "issue:1755", + "right": "issue:3336", + "accept": false, + "reason": "No meaningful overlap; one is a generic/non-English issue title and the other is a scheduler feature request." + }, + { + "left": "issue:2783", + "right": "issue:6141", + "accept": false, + "reason": "Both are ControlNet-related feature requests, but one asks for an img2img/inpainting pipeline and the other for control_mode/resize_mode parameters; different changes." + }, + { + "left": "issue:3452", + "right": "issue:988", + "accept": false, + "reason": "DreamBooth low-VRAM training vs FSDP support for examples are separate enablement topics." + }, + { + "left": "issue:3698", + "right": "issue:5905", + "accept": false, + "reason": "Different model support requests: inpainting model conversion vs adding diffusion-GAN." + }, + { + "left": "issue:3944", + "right": "issue:4241", + "accept": false, + "reason": "Both mention LoRA, but one is about network rank in DreamBooth training and the other about LoRA fine-tuning for unidiffusers; not the same change." + }, + { + "left": "issue:5050", + "right": "issue:9241", + "accept": false, + "reason": "LoRA config change request vs a conversion script from XLab LoRA to Diffusers LoRA; different artifacts and goals." + }, + { + "left": "issue:12499", + "right": "issue:12565", + "accept": false, + "reason": "Both concern Wan i2v, but they target different versions and different symptoms (shape mismatch vs general running issue)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 291, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10178", + "issue:10518", + "issue:10803", + "issue:10914", + "issue:11314", + "issue:11315", + "issue:11342", + "issue:11412", + "issue:11700", + "issue:1399", + "issue:1931", + "issue:2340", + "issue:2606", + "issue:2841", + "issue:2843", + "issue:2905", + "issue:3177", + "issue:3468", + "issue:3724", + "issue:4386", + "issue:4551", + "issue:459", + "issue:5577", + "issue:5907", + "issue:5917", + "issue:6113", + "issue:6686", + "issue:6709", + "issue:672", + "issue:6793", + "issue:7002", + "issue:7380", + "issue:7636", + "issue:7716", + "issue:7773", + "issue:8241", + "issue:8532", + "issue:8588", + "issue:8613", + "issue:8671", + "issue:8780", + "issue:8803", + "issue:8805", + "issue:8943", + "issue:9471", + "issue:9670", + "issue:9719" + ], + "soft_pairs": [ + "issue:4551|issue:6709", + "issue:10914|issue:8613", + "issue:5577|issue:6793", + "issue:1399|issue:5907", + "issue:7636|issue:8241", + "issue:10178|issue:7002", + "issue:10010|issue:8943", + "issue:10803|issue:7380", + "issue:11412|issue:7773", + "issue:2843|issue:6686", + "issue:3724|issue:459", + "issue:4386|issue:672", + "issue:5917|issue:6113", + "issue:8588|issue:8805", + "issue:8671|issue:8803", + "issue:2606|issue:3177", + "issue:11315|issue:11342", + "issue:1931|issue:2340", + "issue:11700|issue:8780", + "issue:2841|issue:2905", + "issue:8532|issue:9670", + "issue:9471|issue:9719", + "issue:3468|issue:7716", + "issue:10518|issue:11314" + ], + "prepared_review_unit_hash": "73629a617fbad027a44c3ee53317db4ed7b37014395bff7257d1d48adfa7b5a0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34060, + "estimated_input_tokens": 8515, + "estimated_eval_tokens": 17286 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are unrelated model/script/runtime bugs that only share shallow text similarity. None of the soft pairs look like the same underlying issue or a mergeable duplicate PR.", + "confidence": 0.72, + "canonical_issue_reason": "issue:10914 is the broadest, most representative runtime symptom in the set (unexpected CPU offloading), so it is the least specific anchor among otherwise unrelated reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10914 is the best single representative issue because it describes a general cross-model runtime behavior rather than a narrow model- or script-specific failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4551", + "right": "issue:6709", + "accept": false, + "reason": "Different failures and code paths: SDXL DreamBooth crash vs. custom diffusion concept-list training bug." + }, + { + "left": "issue:10914", + "right": "issue:8613", + "accept": false, + "reason": "Both involve resource symptoms, but one is unintended CPU offloading and the other is an OOM caused by caption-column training input handling." + }, + { + "left": "issue:5577", + "right": "issue:6793", + "accept": false, + "reason": "ControlNet checkpoint loading KeyError vs. Hydra serialization failure; unrelated subsystems." + }, + { + "left": "issue:1399", + "right": "issue:5907", + "accept": false, + "reason": "Unexpected UNet constructor arg in VersatileDiffusion vs. scheduler docs/API mismatch; not the same bug." + }, + { + "left": "issue:7636", + "right": "issue:8241", + "accept": false, + "reason": "A usage question about SDXL tile ControlNet vs. an abnormal model size report; different problems." + }, + { + "left": "issue:10178", + "right": "issue:7002", + "accept": false, + "reason": "3D block dimension mismatch vs. img2img invalid buffer size; different error contexts." + }, + { + "left": "issue:10010", + "right": "issue:8943", + "accept": false, + "reason": "Missing FLUX model_index.json from a repo URL vs. installation-time CUDA library error; unrelated." + }, + { + "left": "issue:10803", + "right": "issue:7380", + "accept": false, + "reason": "A proposal for SANARubber/SANA flexibility vs. a one-step diffusion implementation request; not the same change." + }, + { + "left": "issue:11412", + "right": "issue:7773", + "accept": false, + "reason": "404 for a specific Marigold model repo vs. a Stable Video Diffusion batch-size bug; different artifacts and code paths." + }, + { + "left": "issue:2843", + "right": "issue:6686", + "accept": false, + "reason": "Multi-GPU checkpoint saving vs. Dynamo compilation performance after save/load; different behaviors." + }, + { + "left": "issue:3724", + "right": "issue:459", + "accept": false, + "reason": "Watermarking module idea vs. incremental diffusion concept; both feature ideas, but not the same request." + }, + { + "left": "issue:4386", + "right": "issue:672", + "accept": false, + "reason": "SDXL LoRA conversion question vs. converting diffusers weights to original CompVis ckpt format; related theme, but not the same concrete task." + }, + { + "left": "issue:5917", + "right": "issue:6113", + "accept": false, + "reason": "Checkpoint conversion regression after upgrade vs. how to use a control repo in diffusers; not the same underlying bug." + }, + { + "left": "issue:8588", + "right": "issue:8805", + "accept": false, + "reason": "Missing SD3 conversion script vs. an SDXL conversion script that no longer works; both conversion-related, but different scripts and failures." + }, + { + "left": "issue:8671", + "right": "issue:8803", + "accept": false, + "reason": "Redundant code in a SD3 training script vs. duplicate assignment in attention_processor.py; unrelated code locations." + }, + { + "left": "issue:2606", + "right": "issue:3177", + "accept": false, + "reason": "DeepSpeed save_state hang vs. DreamBooth ZeRO-3 offloading being broken; both training-related, but different failure modes." + }, + { + "left": "issue:11315", + "right": "issue:11342", + "accept": false, + "reason": "Both are MPS issues, but one is a float64 RoPE implementation bug in HiDream and the other is a Sana-sprint pipeline failure." + }, + { + "left": "issue:1931", + "right": "issue:2340", + "accept": false, + "reason": "Float16 dtype mismatch in SD2 vs. tiny pipeline safety-checker error; unrelated code paths." + }, + { + "left": "issue:11700", + "right": "issue:8780", + "accept": false, + "reason": "Pruna integration request vs. a model/input dtype mismatch bug; different topics." + }, + { + "left": "issue:2841", + "right": "issue:2905", + "accept": false, + "reason": "Both touch schedulers, but one requests sampling temperature and the other requests Karras sigmas; separate feature changes." + }, + { + "left": "issue:8532", + "right": "issue:9670", + "accept": false, + "reason": "LoRA memory growth during loading vs. PEFT backend requirement failure; different loading problems." + }, + { + "left": "issue:9471", + "right": "issue:9719", + "accept": false, + "reason": "Multiple file extensions/config inference error vs. progress bar being ignored during shard loading; unrelated loader behaviors." + }, + { + "left": "issue:3468", + "right": "issue:7716", + "accept": false, + "reason": "Missing model_index.json after DreamBooth export vs. missing config.json in a different directory; similar symptom, but different artifacts and flows." + }, + { + "left": "issue:10518", + "right": "issue:11314", + "accept": false, + "reason": "Wrong SD3 training script vs. Flux ControlNet training script bug; different scripts and model families." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 292, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10281", + "issue:10640", + "issue:10986", + "issue:11088", + "issue:1109", + "issue:11448", + "issue:11989", + "issue:12108", + "issue:1344", + "issue:13446", + "issue:1375", + "issue:1495", + "issue:1867", + "issue:2163", + "issue:2213", + "issue:2659", + "issue:2851", + "issue:337", + "issue:3566", + "issue:3585", + "issue:3717", + "issue:4021", + "issue:4042", + "issue:4080", + "issue:4171", + "issue:4484", + "issue:4580", + "issue:4708", + "issue:4752", + "issue:5110", + "issue:5246", + "issue:5390", + "issue:5433", + "issue:5606", + "issue:5670", + "issue:5689", + "issue:6079", + "issue:6725", + "issue:7483", + "issue:7644", + "issue:7735", + "issue:7962", + "issue:8464", + "issue:8488", + "issue:9167", + "issue:9856", + "issue:9861", + "issue:9957" + ], + "soft_pairs": [ + "issue:10640|issue:7483", + "issue:10986|issue:11088", + "issue:11448|issue:9167", + "issue:1375|issue:1495", + "issue:11989|issue:9861", + "issue:5670|issue:7644", + "issue:2659|issue:6725", + "issue:4580|issue:8488", + "issue:1867|issue:337", + "issue:2163|issue:2213", + "issue:4708|issue:6079", + "issue:12108|issue:13446", + "issue:8464|issue:9856", + "issue:10281|issue:9957", + "issue:2851|issue:3717", + "issue:3585|issue:5110", + "issue:1109|issue:4752", + "issue:3566|issue:5689", + "issue:1344|issue:4042", + "issue:4021|issue:5390", + "issue:4080|issue:7962", + "issue:4171|issue:5246", + "issue:4484|issue:5433", + "issue:5606|issue:7735" + ], + "prepared_review_unit_hash": "8ca0713ac1b6fe26f93b95c4c1d53f13c0aca63effe7286b759b9ed754c6c1e6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34120, + "estimated_input_tokens": 8530, + "estimated_eval_tokens": 17316 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the suggested pairs are only loosely text-similar and do not share the same concrete bug or change. I reject all soft edges.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits this cluster; it mixes unrelated bugs, feature requests, import errors, scheduler issues, and training/runtime problems.", + "canonical_pr_reason": "No PRs are present in this cluster.", + "best_issue_reason": "No issue is a good global representative because the items do not converge on one underlying defect or feature.", + "best_pr_reason": "No PR is present to evaluate as a representative change.", + "soft_edge_verdicts": [ + { + "left": "issue:10640", + "right": "issue:7483", + "accept": false, + "reason": "Different problems: FluxInpaintPipeline DataParallel multi-GPU error vs SDXL LoRA conversion failure." + }, + { + "left": "issue:10986", + "right": "issue:11088", + "accept": false, + "reason": "MPS ComplexDouble conversion bug and wan2.1 GGUF load error are unrelated." + }, + { + "left": "issue:11448", + "right": "issue:9167", + "accept": false, + "reason": "Missing pipeline path vs missing StableDiffusion3Pipeline import; different issues." + }, + { + "left": "issue:1375", + "right": "issue:1495", + "accept": false, + "reason": "Both are import-related, but they fail on different symbols and root causes." + }, + { + "left": "issue:11989", + "right": "issue:9861", + "accept": false, + "reason": "Black output images and Flux training not updating the transformer are different bugs." + }, + { + "left": "issue:5670", + "right": "issue:7644", + "accept": false, + "reason": "Non-square PixArtAlpha failure is not the same as Stable Cascade non-power-of-two image failure." + }, + { + "left": "issue:2659", + "right": "issue:6725", + "accept": false, + "reason": "Batch-size variation in generated images is unrelated to a request to generate images with LDMPipeline." + }, + { + "left": "issue:4580", + "right": "issue:8488", + "accept": false, + "reason": "UniPC use_karras_sigmas exception and StableDiffusion3 from_pretrained exception are different code paths." + }, + { + "left": "issue:1867", + "right": "issue:337", + "accept": false, + "reason": "Both touch img2img, but one is a low-strength scheduler bug and the other is a text-guided inpainting problem." + }, + { + "left": "issue:2163", + "right": "issue:2213", + "accept": false, + "reason": "Training dtype mismatch in one script vs TPU distributed hang in another are unrelated." + }, + { + "left": "issue:4708", + "right": "issue:6079", + "accept": false, + "reason": "Import error for diffusers.utils is not the same as a Half-vs-Float tensor type mismatch." + }, + { + "left": "issue:12108", + "right": "issue:13446", + "accept": false, + "reason": "Flow-matching scheduler breakage and custom_pipeline trust checks are unrelated." + }, + { + "left": "issue:8464", + "right": "issue:9856", + "accept": false, + "reason": "Seed handling with CPU offload is unrelated to a distributed port-in-use error." + }, + { + "left": "issue:10281", + "right": "issue:9957", + "accept": false, + "reason": "Different scheduler feature requests: FreeScale vs BDIA-DDIM." + }, + { + "left": "issue:2851", + "right": "issue:3717", + "accept": false, + "reason": "TPU push_to_hub limitation and multi-GPU OOM are different runtime issues." + }, + { + "left": "issue:3585", + "right": "issue:5110", + "accept": false, + "reason": "LoRA synthesis failure and HFValidationError repo-id formatting are unrelated." + }, + { + "left": "issue:1109", + "right": "issue:4752", + "accept": false, + "reason": "CPU core utilization complaint and SDXL dtype mismatch do not match." + }, + { + "left": "issue:3566", + "right": "issue:5689", + "accept": false, + "reason": "ControlNet reference artifacts and residual noise in SDXL DPM++ are different symptoms and likely different causes." + }, + { + "left": "issue:1344", + "right": "issue:4042", + "accept": false, + "reason": "Single-GPU attention.py limitation is unrelated to combining reference-only with inpainting/depth control." + }, + { + "left": "issue:4021", + "right": "issue:5390", + "accept": false, + "reason": "Both involve downloads, but one is unnecessary CLIP download from_single_file and the other is local_files_only text-encoder fetch behavior." + }, + { + "left": "issue:4080", + "right": "issue:7962", + "accept": false, + "reason": "SDXL inpainting pipeline request and cross_attention_kwargs for instruct-pix2pix are separate feature asks." + }, + { + "left": "issue:4171", + "right": "issue:5246", + "accept": false, + "reason": "Reference page rendering bug and missing CLIPImageProjection attribute are unrelated." + }, + { + "left": "issue:4484", + "right": "issue:5433", + "accept": false, + "reason": "SDXL Kohya LoRA quality issues and DPM++ residual noise without refiner are not the same bug." + }, + { + "left": "issue:5606", + "right": "issue:7735", + "accept": false, + "reason": "LoRA alpha handling and caption_column dataset validation are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 293, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10239", + "issue:1061", + "issue:12022", + "issue:1477", + "issue:1633", + "issue:1831", + "issue:2266", + "issue:2276", + "issue:2437", + "issue:2577", + "issue:2782", + "issue:2843", + "issue:3580", + "issue:3612", + "issue:4349", + "issue:4414", + "issue:4468", + "issue:4677", + "issue:4767", + "issue:5465", + "issue:5628", + "issue:5630", + "issue:5760", + "issue:5881", + "issue:5897", + "issue:6087", + "issue:6132", + "issue:6162", + "issue:6452", + "issue:6709", + "issue:7009", + "issue:7333", + "issue:7357", + "issue:7415", + "issue:7728", + "issue:776", + "issue:7962", + "issue:8482", + "issue:8708", + "issue:8813", + "issue:8885", + "issue:8939", + "issue:9209", + "issue:9398", + "issue:9495", + "issue:9523", + "issue:9719" + ], + "soft_pairs": [ + "issue:6162|issue:9495", + "issue:6452|issue:7962", + "issue:8482|issue:8813", + "issue:8885|issue:9209", + "issue:6709|issue:8939", + "issue:2276|issue:3580", + "issue:4414|issue:4468", + "issue:5897|issue:7415", + "issue:7728|issue:9523", + "issue:1477|issue:1831", + "issue:2577|issue:5628", + "issue:5760|issue:7333", + "issue:8939|issue:9719", + "issue:2266|issue:2782", + "issue:2437|issue:6132", + "issue:4677|issue:7009", + "issue:4767|issue:5465", + "issue:5630|issue:5881", + "issue:8708|issue:9398", + "issue:10239|issue:12022", + "issue:2843|issue:6087", + "issue:4349|issue:7357", + "issue:1061|issue:776", + "issue:1633|issue:3612" + ], + "prepared_review_unit_hash": "cb746f5b39cacd37974468ec0c9fe099bd77d08fe7eab44e6e147721ae650b09", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33864, + "estimated_input_tokens": 8466, + "estimated_eval_tokens": 17188 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a real duplicate set: the issues span unrelated bugs and feature requests across pipelines, schedulers, training scripts, model loading, and docs. All soft pairs are superficial matches only, so every edge should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": "No strong canonical issue exists because the cluster is heterogeneous; if forced to anchor one, issue 10239 is the most concrete and recent offload-related report, but it does not match the rest well.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10239 is the most actionable and specific report in the set, but only as a loose triage anchor rather than a true duplicate representative.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6162", + "right": "issue:9495", + "accept": false, + "reason": "Both mention SDXL/IP-Adapter, but one is a shape-mismatch bug and the other is a PAG incompatibility; different failure modes and likely different fixes." + }, + { + "left": "issue:6452", + "right": "issue:7962", + "accept": false, + "reason": "Different pipelines and requests: ControlNet inpaint masked_image_latents versus adding cross_attention_kwargs to Instruct-Pix2Pix." + }, + { + "left": "issue:8482", + "right": "issue:8813", + "accept": false, + "reason": "One is general SD3-medium support, the other is a Differential Diffusion/Kolors issue; unrelated model/pipeline problems." + }, + { + "left": "issue:8885", + "right": "issue:9209", + "accept": false, + "reason": "Custom Diffusion/clip-retrieval setup issues are unrelated to an hf-mirror connectivity failure." + }, + { + "left": "issue:6709", + "right": "issue:8939", + "accept": false, + "reason": "Custom Diffusion training with concept lists is a different bug from SD3 tokenizer loading." + }, + { + "left": "issue:2276", + "right": "issue:3580", + "accept": false, + "reason": "One is an install problem for xformers in Colab, the other is a runtime TypeError using XFormers; not the same bug." + }, + { + "left": "issue:4414", + "right": "issue:4468", + "accept": false, + "reason": "A consistency-model sample request and a multi-GPU performance complaint are unrelated." + }, + { + "left": "issue:5897", + "right": "issue:7415", + "accept": false, + "reason": "LoRA device-mismatch during training and a bfloat16 TPU VAE issue are different code paths and symptoms." + }, + { + "left": "issue:7728", + "right": "issue:9523", + "accept": false, + "reason": "Both are training-script errors, but they concern different scripts and distinct mapping/Accelerate failures." + }, + { + "left": "issue:1477", + "right": "issue:1831", + "accept": false, + "reason": "CUDA cublas invalid-value and DreamBooth class-image precision mismatch are unrelated bugs." + }, + { + "left": "issue:2577", + "right": "issue:5628", + "accept": false, + "reason": "Both involve schedulers, but DDIMInverseScheduler failure and LMS/Karras denoising offset are different concrete issues." + }, + { + "left": "issue:5760", + "right": "issue:7333", + "accept": false, + "reason": "Video-to-audio synthesis and image-to-video generation are different projects and tasks." + }, + { + "left": "issue:8939", + "right": "issue:9719", + "accept": false, + "reason": "Tokenizer loading for SD3-medium versus progress-bar behavior while loading shards are unrelated." + }, + { + "left": "issue:2266", + "right": "issue:2782", + "accept": false, + "reason": "Both mention sequential offload, but one is slowdown and the other is a generic offloading error; not the same bug." + }, + { + "left": "issue:2437", + "right": "issue:6132", + "accept": false, + "reason": "A new T2I-Adapter pipeline request is unrelated to adding LOADABLE_CLASSES." + }, + { + "left": "issue:4677", + "right": "issue:7009", + "accept": false, + "reason": "A documentation script error and poor TextToVideoSDPipeline results are different issues." + }, + { + "left": "issue:4767", + "right": "issue:5465", + "accept": false, + "reason": "EulerAncestralDiscreteScheduler sample prediction support and DiffEdit inpaint_strength limits are separate feature/bug reports." + }, + { + "left": "issue:5630", + "right": "issue:5881", + "accept": false, + "reason": "Both are quality complaints, but they involve different pipelines/models and different underlying causes." + }, + { + "left": "issue:8708", + "right": "issue:9398", + "accept": false, + "reason": "Missing SD3 dreambooth config and a missing pooled_projection argument in a different embedding module are not the same defect." + }, + { + "left": "issue:10239", + "right": "issue:12022", + "accept": false, + "reason": "HunyuanVideoPipeline sequential CPU offload support and flash_attention_3 interface compatibility are unrelated problems." + }, + { + "left": "issue:2843", + "right": "issue:6087", + "accept": false, + "reason": "Multi-GPU checkpoint saving and LoRAConfig save/load producing wrong lora_alpha are distinct concerns." + }, + { + "left": "issue:4349", + "right": "issue:7357", + "accept": false, + "reason": "Missing text_encoder handling and slow inference with torch.compile/fuse_loras are different bugs." + }, + { + "left": "issue:1061", + "right": "issue:776", + "accept": false, + "reason": "A textual inversion example instruction update is unrelated to an image-to-image batch-size bug." + }, + { + "left": "issue:1633", + "right": "issue:3612", + "accept": false, + "reason": "Karras sigmas in the k_diffusion wrapper and ControlNet/reference_adain+attn guidance are unrelated feature requests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 294, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10050", + "issue:10566", + "issue:10673", + "issue:11010", + "issue:11178", + "issue:1141", + "issue:11627", + "issue:11779", + "issue:12012", + "issue:12025", + "issue:12071", + "issue:12536", + "issue:12574", + "issue:1273", + "issue:1304", + "issue:13107", + "issue:205", + "issue:2179", + "issue:2437", + "issue:2480", + "issue:2737", + "issue:2786", + "issue:3029", + "issue:307", + "issue:3455", + "issue:3481", + "issue:3958", + "issue:4111", + "issue:4116", + "issue:4264", + "issue:459", + "issue:5313", + "issue:5523", + "issue:5731", + "issue:5997", + "issue:6313", + "issue:6320", + "issue:6548", + "issue:6616", + "issue:6639", + "issue:7380", + "issue:7389", + "issue:7432", + "issue:7640", + "issue:8539", + "issue:8614", + "issue:8956", + "issue:9216" + ], + "soft_pairs": [ + "issue:2179|issue:3481", + "issue:3029|issue:5997", + "issue:307|issue:459", + "issue:4111|issue:4116", + "issue:4264|issue:5731", + "issue:6616|issue:7380", + "issue:6639|issue:7432", + "issue:11779|issue:12574", + "issue:2480|issue:5523", + "issue:10566|issue:12071", + "issue:10673|issue:12025", + "issue:7389|issue:7640", + "issue:11178|issue:8614", + "issue:2786|issue:6548", + "issue:11627|issue:8539", + "issue:3958|issue:5313", + "issue:12012|issue:12536", + "issue:6320|issue:8956", + "issue:10050|issue:9216", + "issue:11010|issue:13107", + "issue:1141|issue:2737", + "issue:1273|issue:3455", + "issue:1304|issue:205", + "issue:2437|issue:6313" + ], + "prepared_review_unit_hash": "026aa9258adac1c8e8440ca03c720a904f6dbd6117045feafe392211603a125a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34050, + "estimated_input_tokens": 8513, + "estimated_eval_tokens": 17282 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated feature requests or bug reports across different pipelines. The only near-duplicate pair is the ControlNet single-file loading/config.json reports.", + "confidence": 0.79, + "canonical_issue_reason": "issue:8614 is the broadest and most general statement of the ControlNet single-file loading problem, while issue:11178 is a specific symptom for one model.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8614 best represents the shared underlying bug because it describes the missing config.json / single-file ControlNet loading problem at the right abstraction level.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2179", + "right": "issue:3481", + "accept": false, + "reason": "Different problems: FlaxDDPMScheduler black images vs DDIM step-count sensitivity and inconsistent outputs." + }, + { + "left": "issue:3029", + "right": "issue:5997", + "accept": false, + "reason": "Both involve ControlNet tutorials, but one is custom resolution and the other is a model-type mismatch; not the same bug." + }, + { + "left": "issue:307", + "right": "issue:459", + "accept": false, + "reason": "One is about merging Stable Diffusion pipelines; the other is an incremental diffusion feature request." + }, + { + "left": "issue:4111", + "right": "issue:4116", + "accept": false, + "reason": "Same pipeline family, but different issues: VRAM/memory loading vs embeddings not working." + }, + { + "left": "issue:4264", + "right": "issue:5731", + "accept": false, + "reason": "Missing example images is unrelated to the AutoencoderKL trainable logvar implementation issue." + }, + { + "left": "issue:6616", + "right": "issue:7380", + "accept": false, + "reason": "Classifier guidance support and one-step diffusion distillation are different changes/problems." + }, + { + "left": "issue:6639", + "right": "issue:7432", + "accept": false, + "reason": "Both concern different outputs, but one is LoRA loading order in SDXL and the other is general inference differences vs WebUI." + }, + { + "left": "issue:11779", + "right": "issue:12574", + "accept": false, + "reason": "Both are Wan/VACE-related, but one is mask padding and the other is missing control-argument handling; distinct failure modes." + }, + { + "left": "issue:2480", + "right": "issue:5523", + "accept": false, + "reason": "Dreambooth checkpoint directory bug is unrelated to generic run-to-run inconsistency." + }, + { + "left": "issue:10566", + "right": "issue:12071", + "accept": false, + "reason": "Different models and code paths: CogVideoX forward work vs flux kontext single-block forward behavior." + }, + { + "left": "issue:10673", + "right": "issue:12025", + "accept": false, + "reason": "AuraFlow shape mismatch and Cosmos VAE encoder API error are unrelated." + }, + { + "left": "issue:7389", + "right": "issue:7640", + "accept": false, + "reason": "Both mention SDXL from_single_file, but one is a generic load error and the other is a dimensions-specific runtime error." + }, + { + "left": "issue:11178", + "right": "issue:8614", + "accept": true, + "reason": "Both describe the same ControlNet single-file loading failure centered on missing config.json / from_single_file support; 8614 is the broader root report." + }, + { + "left": "issue:2786", + "right": "issue:6548", + "accept": false, + "reason": "NumPy input incompatibility with transformers-4.27 is different from RGBA image handling in prepare_latents." + }, + { + "left": "issue:11627", + "right": "issue:8539", + "accept": false, + "reason": "StableDiffusion3 with IP-Adapter and SD3 cross-attention issues are separate model-integration problems." + }, + { + "left": "issue:3958", + "right": "issue:5313", + "accept": false, + "reason": "Different LoRA-related problems: cpu offload interaction vs non-revertible fuse behavior." + }, + { + "left": "issue:12012", + "right": "issue:12536", + "accept": false, + "reason": "Both Wan-related, but one is context/caching state management and the other is context-parallel sharding size divisibility." + }, + { + "left": "issue:6320", + "right": "issue:8956", + "accept": false, + "reason": "Emu2 remote model lookup and OVStableDiffusionXL model loading are unrelated." + }, + { + "left": "issue:10050", + "right": "issue:9216", + "accept": false, + "reason": "Img2img KDiffusion equivalent question is unrelated to a custom multi-model pipeline proposal." + }, + { + "left": "issue:11010", + "right": "issue:13107", + "accept": false, + "reason": "Flux Chroma architecture support and Flux Klein LoRA missing-key errors are different issues." + }, + { + "left": "issue:1141", + "right": "issue:2737", + "accept": false, + "reason": "Local files in custom pipelines is unrelated to CLIP guidance support for a scheduler." + }, + { + "left": "issue:1273", + "right": "issue:3455", + "accept": false, + "reason": "Dreambooth mean-loss question and using pretrained instruct-pix2pix in its own training are not duplicates." + }, + { + "left": "issue:1304", + "right": "issue:205", + "accept": false, + "reason": "Training img2img stable diffusion and adding interpolation example scripts are unrelated." + }, + { + "left": "issue:2437", + "right": "issue:6313", + "accept": false, + "reason": "T2I-Adapter pipeline request is unrelated to Null-Text Inversion pipeline implementation." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The canonical issue choice is supported by the packet: issue:8614 is the broader ControlNet single-file/config.json report, and issue:11178 is a narrower symptom of the same loading problem. The other soft-edge verdicts are phrased conservatively as non-duplicates and do not overclaim beyond the titles/body summaries shown." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 295, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10191", + "issue:10311", + "issue:10446", + "issue:10447", + "issue:10823", + "issue:10963", + "issue:2763", + "issue:2961", + "issue:3043", + "issue:3065", + "issue:3191", + "issue:3223", + "issue:3325", + "issue:3377", + "issue:3390", + "issue:3604", + "issue:3960", + "issue:4282", + "issue:4665", + "issue:4756", + "issue:4801", + "issue:4969", + "issue:5025", + "issue:5061", + "issue:5079", + "issue:5112", + "issue:514", + "issue:5399", + "issue:5432", + "issue:5654", + "issue:5873", + "issue:6132", + "issue:6157", + "issue:6354", + "issue:6359", + "issue:656", + "issue:6996", + "issue:7043", + "issue:7079", + "issue:770", + "issue:8241", + "issue:8824", + "issue:8939", + "issue:9047", + "issue:9178", + "issue:9397", + "issue:9424", + "issue:9926" + ], + "soft_pairs": [ + "issue:2763|issue:4282", + "issue:3043|issue:6132", + "issue:3065|issue:7043", + "issue:4756|issue:5432", + "issue:6354|issue:9926", + "issue:6359|issue:7079", + "issue:9047|issue:9178", + "issue:10191|issue:10446", + "issue:5399|issue:8241", + "issue:10447|issue:10823", + "issue:10311|issue:10963", + "issue:2961|issue:3604", + "issue:3191|issue:3960", + "issue:3223|issue:6157", + "issue:3325|issue:4801", + "issue:3377|issue:5654", + "issue:3390|issue:4665", + "issue:4969|issue:5061", + "issue:5025|issue:770", + "issue:5079|issue:9397", + "issue:5112|issue:6996", + "issue:514|issue:656", + "issue:5873|issue:9424", + "issue:8824|issue:8939" + ], + "prepared_review_unit_hash": "857a934ef5d37ecbb58948ec81ffed5f70b50ced45f8d3a4b23d01a82e1c7007", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34099, + "estimated_input_tokens": 8525, + "estimated_eval_tokens": 17306 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No duplicate cluster here: the items span unrelated training, import, pipeline, conversion, and scheduler issues. The soft pairs share only broad subsystem similarity, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": "If one issue had to anchor the set, #10191 is the clearest self-contained bug report with a specific expected-vs-actual regression (GPU device selection during training), but it does not represent the rest of the heterogeneous cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "#10191 is the best single issue to keep as a representative because it is concrete, well-scoped, and describes a clear reproducible training-device bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2763", + "right": "issue:4282", + "accept": false, + "reason": "ControlNet VRAM regression vs a Multi-ControlNet feature request; different problem types." + }, + { + "left": "issue:3043", + "right": "issue:6132", + "accept": false, + "reason": "ODISE addition vs LOADABLE_CLASSES support; unrelated features." + }, + { + "left": "issue:3065", + "right": "issue:7043", + "accept": false, + "reason": "DreamBooth LoRA option request vs custom captions support; related area, but different change requests." + }, + { + "left": "issue:4756", + "right": "issue:5432", + "accept": false, + "reason": "Both are DreamBooth/SDXL training-related, but one is about adding a training example and the other is a validation-image bug." + }, + { + "left": "issue:6354", + "right": "issue:9926", + "accept": false, + "reason": "ControlNet LoRA support vs GGUF quantize/speed-up support; unrelated enhancements." + }, + { + "left": "issue:6359", + "right": "issue:7079", + "accept": false, + "reason": "Scheduler velocity/add_noise bug vs Transformer2DModel patch/norm_type error; different code paths." + }, + { + "left": "issue:9047", + "right": "issue:9178", + "accept": false, + "reason": "MPS support for Flux vs FLUX LoRA PEFT target-module errors; distinct issues." + }, + { + "left": "issue:10191", + "right": "issue:10446", + "accept": false, + "reason": "GPU device selection during training vs FluxControlNetPipeline multi-GPU/model-sharding incompatibility; not the same bug." + }, + { + "left": "issue:5399", + "right": "issue:8241", + "accept": false, + "reason": "Custom inpainting dtype mismatch vs LEDITS++ model size question; unrelated." + }, + { + "left": "issue:10447", + "right": "issue:10823", + "accept": false, + "reason": "Converter splitting UNet into two files vs finetuning script output files for inference; similar output packaging theme, but different scripts and failures." + }, + { + "left": "issue:10311", + "right": "issue:10963", + "accept": false, + "reason": "Dreambooth import error vs AutoencoderKLWan import error; both are import issues, but for different symbols and modules." + }, + { + "left": "issue:2961", + "right": "issue:3604", + "accept": false, + "reason": "Request for a StableUnCLIP training example vs training loss not decreasing; one is documentation/example, the other a training-quality bug." + }, + { + "left": "issue:3191", + "right": "issue:3960", + "accept": false, + "reason": "DreamBooth+LoRA broken vs HeaderTooLarge for LoRA; different failure modes." + }, + { + "left": "issue:3223", + "right": "issue:6157", + "accept": false, + "reason": "Memory not cleared after moving pipeline to CUDA vs IP-Adapter not working in ControlNetImg2ImgPipeline; unrelated pipeline bugs." + }, + { + "left": "issue:3325", + "right": "issue:4801", + "accept": false, + "reason": "CUDA invalid-argument in train_dreambooth.py vs mixed-precision bug in train_unconditional.py; different scripts and errors." + }, + { + "left": "issue:3377", + "right": "issue:5654", + "accept": false, + "reason": "General question about SNR weighting vs a request to add Min-SNR weight for v-prediction; related topic, but not the same underlying issue." + }, + { + "left": "issue:3390", + "right": "issue:4665", + "accept": false, + "reason": "Local safetensors loading question vs Windows use_safetensors fallback bug; different scenarios." + }, + { + "left": "issue:4969", + "right": "issue:5061", + "accept": false, + "reason": "CrossAttention import error vs config_files argument error; separate API breakages." + }, + { + "left": "issue:5025", + "right": "issue:770", + "accept": false, + "reason": "randn_tensor import error vs Colab training startup failure; unrelated." + }, + { + "left": "issue:5079", + "right": "issue:9397", + "accept": false, + "reason": "LoRA loading support request vs SGMUniform scheduler support for Hyper LoRAs; both mention LoRA, but not the same change." + }, + { + "left": "issue:5112", + "right": "issue:6996", + "accept": false, + "reason": "Size/type error in SDXL ControlNet img2img vs NoneType added_cond_kwargs error; different runtime paths." + }, + { + "left": "issue:514", + "right": "issue:656", + "accept": false, + "reason": "Attention slicing error vs 16GB DreamBooth training error; too broad and not the same bug." + }, + { + "left": "issue:5873", + "right": "issue:9424", + "accept": false, + "reason": "ONNX export failure vs broken DreamBooth notebook; unrelated." + }, + { + "left": "issue:8824", + "right": "issue:8939", + "accept": false, + "reason": "Custom dataset guidance for instruct-pix2pix vs tokenizer load failure for SD3; different topics and failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 296, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10178", + "issue:1070", + "issue:11000", + "issue:11006", + "issue:11470", + "issue:12895", + "issue:13124", + "issue:1746", + "issue:1817", + "issue:1821", + "issue:1944", + "issue:2392", + "issue:2444", + "issue:2579", + "issue:2737", + "issue:3032", + "issue:3103", + "issue:3287", + "issue:3551", + "issue:3767", + "issue:3880", + "issue:4418", + "issue:4446", + "issue:4589", + "issue:4648", + "issue:4871", + "issue:4912", + "issue:4964", + "issue:5336", + "issue:5646", + "issue:5685", + "issue:5777", + "issue:5890", + "issue:617", + "issue:6481", + "issue:671", + "issue:7228", + "issue:7264", + "issue:7904", + "issue:8405", + "issue:8535", + "issue:8556", + "issue:8809", + "issue:917", + "issue:9674", + "issue:987", + "issue:9946", + "issue:9962" + ], + "soft_pairs": [ + "issue:10178|issue:8405", + "issue:12895|issue:2392", + "issue:3103|issue:917", + "issue:4964|issue:671", + "issue:11000|issue:9962", + "issue:11470|issue:13124", + "issue:2737|issue:3287", + "issue:4589|issue:5336", + "issue:4648|issue:5777", + "issue:7904|issue:8535", + "issue:4418|issue:5890", + "issue:4446|issue:4871", + "issue:2579|issue:3551", + "issue:6481|issue:9674", + "issue:7228|issue:8556", + "issue:1070|issue:3880", + "issue:11006|issue:8809", + "issue:1817|issue:1944", + "issue:3767|issue:617", + "issue:4912|issue:7264", + "issue:8535|issue:9946", + "issue:1746|issue:2444", + "issue:1821|issue:3032", + "issue:1821|issue:987", + "issue:5646|issue:5685" + ], + "prepared_review_unit_hash": "13487e560d80b1a96e0631a63fc2e0b266ebcf0878e61e78e21778d3ee452e73", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34266, + "estimated_input_tokens": 8567, + "estimated_eval_tokens": 17390 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous and does not look like a set of true duplicates. The only strong duplicate-like pair is the two xFormers memory-efficient-attention issues; the rest are related only at a broad subsystem level or are clearly different bugs/features.", + "confidence": 0.83, + "canonical_issue_reason": "issue:8535 is the best canonical issue because it is the clearest, most specific description of the only near-duplicate bug/change in the cluster: xFormers memory-efficient attention enablement for SD3.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:8535 is the strongest representative issue in this set: concrete, code-path-specific, and paired with a very likely duplicate (issue:9946).", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10178", + "right": "issue:8405", + "accept": false, + "reason": "Both mention size mismatches, but one is a 3D UNet block issue and the other is a multiple-ControlNet tensor sizing problem; different code paths." + }, + { + "left": "issue:12895", + "right": "issue:2392", + "accept": false, + "reason": "Both involve None/prompt handling, but they occur in different pipelines and under different APIs; not the same bug." + }, + { + "left": "issue:3103", + "right": "issue:917", + "accept": false, + "reason": "Both are inpainting-related, but one is about height/width handling and the other is an ONNX inpainting error; too different." + }, + { + "left": "issue:4964", + "right": "issue:671", + "accept": false, + "reason": "Both are conversion-related, but they target different model families and conversion flows; not a single duplicate bug." + }, + { + "left": "issue:11000", + "right": "issue:9962", + "accept": false, + "reason": "Missing shift argument vs unexpected use_cuda_graph keyword are unrelated API mismatches." + }, + { + "left": "issue:11470", + "right": "issue:13124", + "accept": false, + "reason": "These are separate training-script problems with different symptoms and triggers." + }, + { + "left": "issue:2737", + "right": "issue:3287", + "accept": false, + "reason": "A scheduler support request and a DDPM support request are different feature asks, not the same change." + }, + { + "left": "issue:4589", + "right": "issue:5336", + "accept": false, + "reason": "Both concern ControlNet img2img/inpaint, but one is about missing XL pipeline variants and the other about adding a strength parameter; different fixes." + }, + { + "left": "issue:4648", + "right": "issue:5777", + "accept": false, + "reason": "Missing config.json and a DreamBooth PixArt training error are unrelated." + }, + { + "left": "issue:7904", + "right": "issue:8535", + "accept": false, + "reason": "Scheduler sigmas mutating after forward is unrelated to enabling xFormers memory-efficient attention." + }, + { + "left": "issue:4418", + "right": "issue:5890", + "accept": false, + "reason": "Inpaint denoising_start behavior and a token-length warning during text-to-image training are not the same bug." + }, + { + "left": "issue:4446", + "right": "issue:4871", + "accept": false, + "reason": "One is about custom text_encoder support in from_single_file; the other is a general how-to question for from_single_file." + }, + { + "left": "issue:2579", + "right": "issue:3551", + "accept": false, + "reason": "Generic xFormers misuse/bug speculation is not the same as LoRA disabling xFormers." + }, + { + "left": "issue:6481", + "right": "issue:9674", + "accept": false, + "reason": "Batch-size training failure and SD3 ControlNet quality issues are different problems." + }, + { + "left": "issue:7228", + "right": "issue:8556", + "accept": false, + "reason": "Both are loading-time errors, but one is a missing dtype attribute on traced models and the other is a meta-tensor copy failure in SD3 from_single_file." + }, + { + "left": "issue:1070", + "right": "issue:3880", + "accept": false, + "reason": "Both are inpainting complaints, but the titles do not point to the same concrete bug; one is generic and the other is specifically about unmasked-area degradation." + }, + { + "left": "issue:11006", + "right": "issue:8809", + "accept": false, + "reason": "A Wan I2V video-output bug with quantized transformer is unrelated to a feedforward sequence-length inconsistency." + }, + { + "left": "issue:1817", + "right": "issue:1944", + "accept": false, + "reason": "DreamBooth scheduler argument handling and random DreamBooth outputs are different issues." + }, + { + "left": "issue:3767", + "right": "issue:617", + "accept": false, + "reason": "FrozenDict mutability and NNPACK hardware support are unrelated." + }, + { + "left": "issue:4912", + "right": "issue:7264", + "accept": false, + "reason": "Model conversion to diffusers and serialization/loading of custom transformers models are different code paths." + }, + { + "left": "issue:8535", + "right": "issue:9946", + "accept": true, + "reason": "These are essentially the same xFormers memory-efficient-attention issue; the titles describe the same concrete path and could be merged." + }, + { + "left": "issue:1746", + "right": "issue:2444", + "accept": false, + "reason": "A Flax pipeline error and a DreamBooth inpaint script failure are unrelated." + }, + { + "left": "issue:1821", + "right": "issue:3032", + "accept": false, + "reason": "Negative_prompt ineffectiveness in inpainting and a missing from_pretrained attribute are unrelated." + }, + { + "left": "issue:1821", + "right": "issue:987", + "accept": false, + "reason": "The inpainting negative_prompt complaint and a speech-to-image green-image problem are different bugs." + }, + { + "left": "issue:5646", + "right": "issue:5685", + "accept": false, + "reason": "Both discuss schedulers generally, but they are not the same concrete scheduler bug or fix." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only clearly supported duplicate-like pair is issue:8535 / issue:9946, and the other soft-edge rejections are appropriately cautious given the titles and issue contexts." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 297, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10154", + "issue:10302", + "issue:11574", + "issue:11833", + "issue:1249", + "issue:1434", + "issue:2091", + "issue:2722", + "issue:2851", + "issue:3097", + "issue:3405", + "issue:3468", + "issue:4232", + "issue:4376", + "issue:4480", + "issue:4530", + "issue:4646", + "issue:4649", + "issue:4826", + "issue:5124", + "issue:5152", + "issue:5509", + "issue:5515", + "issue:5607", + "issue:5753", + "issue:5866", + "issue:5917", + "issue:6090", + "issue:6109", + "issue:6280", + "issue:6321", + "issue:6639", + "issue:6686", + "issue:6807", + "issue:7240", + "issue:7415", + "issue:7562", + "issue:7589", + "issue:7891", + "issue:8526", + "issue:898", + "issue:9106", + "issue:9332", + "issue:9539", + "issue:9690", + "issue:9846", + "issue:9911" + ], + "soft_pairs": [ + "issue:3468|issue:6321", + "issue:11574|issue:11833", + "issue:1434|issue:3097", + "issue:2722|issue:5917", + "issue:4232|issue:5866", + "issue:4376|issue:5753", + "issue:5515|issue:9332", + "issue:4530|issue:6686", + "issue:4826|issue:7562", + "issue:10154|issue:6639", + "issue:4646|issue:7240", + "issue:5152|issue:9106", + "issue:4649|issue:7589", + "issue:6280|issue:9846", + "issue:4480|issue:8526", + "issue:2091|issue:5124", + "issue:9539|issue:9911", + "issue:7415|issue:7891", + "issue:4480|issue:6807", + "issue:5607|issue:6090", + "issue:10302|issue:9690", + "issue:3405|issue:6109", + "issue:1249|issue:5509", + "issue:2851|issue:898" + ], + "prepared_review_unit_hash": "ee7d17e6ec0c2dc038e633e2557e5f12cb6b9cfebea683175cf1313480582345", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33727, + "estimated_input_tokens": 8432, + "estimated_eval_tokens": 17120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous set of unrelated issues rather than a true duplicate cluster. All proposed soft pairs are different concrete bugs or feature requests and should be kept separate.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue is justified: the artifacts span unrelated topics (LoRA loading, ControlNet, TPU/MPS/FP8, conversion, training, and pipeline API gaps).", + "canonical_pr_reason": null, + "best_issue_reason": "No best representative issue stands out because the cluster is not a single underlying bug family.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3468", + "right": "issue:6321", + "accept": false, + "reason": "Different problems: missing `model_index.json` for DreamBooth vs CLIP model not being detected." + }, + { + "left": "issue:11574", + "right": "issue:11833", + "accept": false, + "reason": "Both mention KeyError, but one is a VAE weight lookup failure and the other is a LoRA device-setting error." + }, + { + "left": "issue:1434", + "right": "issue:3097", + "accept": false, + "reason": "OOM during upscaling is unrelated to `load_textual_inversion` doing nothing." + }, + { + "left": "issue:2722", + "right": "issue:5917", + "accept": false, + "reason": "Both involve conversion code, but one is ControlNet UNet mismatch after conversion and the other is `convert_ldm_unet_checkpoint` breaking after an upgrade." + }, + { + "left": "issue:4232", + "right": "issue:5866", + "accept": false, + "reason": "Different feature requests/bugs: instruct-pix2pix with ControlNet vs AnimateDiff + ControlNet." + }, + { + "left": "issue:4376", + "right": "issue:5753", + "accept": false, + "reason": "Both are SDXL/adapter-loading related, but one is textual inversion crashing and the other is LoRA loading failing in ControlNet." + }, + { + "left": "issue:5515", + "right": "issue:9332", + "accept": false, + "reason": "LCM scheduler initialization failure is not the same as `from_pretrained()` loading identical weights." + }, + { + "left": "issue:4530", + "right": "issue:6686", + "accept": false, + "reason": "Image artifacts in SDXL refiner are unrelated to Dynamo compile performance after save/load." + }, + { + "left": "issue:4826", + "right": "issue:7562", + "accept": false, + "reason": "Both mention MultiControlNet, but the failures are different: reference ControlNet errors vs CPU offload failure in an SDXL img2img pipeline." + }, + { + "left": "issue:10154", + "right": "issue:6639", + "accept": false, + "reason": "One is an attribute error from `load_lora_weights`; the other is output differences between loading LoRA in one step vs two steps." + }, + { + "left": "issue:4646", + "right": "issue:7240", + "accept": false, + "reason": "ControlNet ONNX conversion failure is unrelated to low-quality output from conditional LDM conversion." + }, + { + "left": "issue:5152", + "right": "issue:9106", + "accept": false, + "reason": "W\u00fcrstchen alpha bug and PixArt-alpha attribute error are different models and different failures." + }, + { + "left": "issue:4649", + "right": "issue:7589", + "accept": false, + "reason": "`NoneType` membership error in `added_cond_kwargs` is not the same as GLIGEN not working." + }, + { + "left": "issue:6280", + "right": "issue:9846", + "accept": false, + "reason": "Both ask for `from_single_file`, but they target different classes/pipelines and are separate API gaps." + }, + { + "left": "issue:4480", + "right": "issue:8526", + "accept": false, + "reason": "MPS kernel crash with fp16 is different from `joint_attention_kwargs['scale']` being ineffective without PEFT." + }, + { + "left": "issue:2091", + "right": "issue:5124", + "accept": false, + "reason": "Checkpoint resume with multi-GPU training is unrelated to how to fine-tune a `.safetensor` checkpoint." + }, + { + "left": "issue:9539", + "right": "issue:9911", + "accept": false, + "reason": "Float8 CUDA operator support in CogVideoX is unrelated to Flux MultiControlNet layer-length mismatch." + }, + { + "left": "issue:7415", + "right": "issue:7891", + "accept": false, + "reason": "TPU bfloat16 AutoencoderKL issues are not the same as mixed CPU/CUDA device placement errors." + }, + { + "left": "issue:4480", + "right": "issue:6807", + "accept": false, + "reason": "A runtime crash on MPS is not the same as test collection failures on MPS." + }, + { + "left": "issue:5607", + "right": "issue:6090", + "accept": false, + "reason": "Both are shape/channel mismatches, but they describe different tensor dimensions and likely different code paths." + }, + { + "left": "issue:10302", + "right": "issue:9690", + "accept": false, + "reason": "FP8 inference noise is not the same as FP8 example-code cleanup." + }, + { + "left": "issue:3405", + "right": "issue:6109", + "accept": false, + "reason": "Generic ControlNet training bug vs a specific ControlNet-xs depth-control issue." + }, + { + "left": "issue:1249", + "right": "issue:5509", + "accept": false, + "reason": "OnnxPipeline ndarray `.device` attribute error is different from image/prompt batch-size mismatch." + }, + { + "left": "issue:2851", + "right": "issue:898", + "accept": false, + "reason": "TPU push_to_hub training limitation is unrelated to unsupported BFloat16 scalar type on TPU." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 298, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10421", + "issue:10855", + "issue:10913", + "issue:11000", + "issue:1101", + "issue:11127", + "issue:1117", + "issue:11214", + "issue:11470", + "issue:11675", + "issue:1181", + "issue:1197", + "issue:11989", + "issue:12437", + "issue:12505", + "issue:12633", + "issue:1407", + "issue:1478", + "issue:177", + "issue:1821", + "issue:1857", + "issue:202", + "issue:2392", + "issue:244", + "issue:2462", + "issue:2737", + "issue:2782", + "issue:2871", + "issue:3162", + "issue:3280", + "issue:3391", + "issue:3418", + "issue:3428", + "issue:3480", + "issue:3769", + "issue:5170", + "issue:5336", + "issue:6958", + "issue:725", + "issue:7690", + "issue:7939", + "issue:8659", + "issue:9052", + "issue:9331", + "issue:939", + "issue:968", + "issue:9772" + ], + "soft_pairs": [ + "issue:7690|issue:7939", + "issue:10421|issue:12505", + "issue:10855|issue:12633", + "issue:10913|issue:11675", + "issue:11000|issue:11214", + "issue:1101|issue:3418", + "issue:11127|issue:9052", + "issue:1117|issue:202", + "issue:11470|issue:8659", + "issue:1181|issue:725", + "issue:1197|issue:939", + "issue:11989|issue:9331", + "issue:12437|issue:9772", + "issue:1407|issue:1821", + "issue:1478|issue:3769", + "issue:177|issue:244", + "issue:1857|issue:2871", + "issue:2392|issue:3428", + "issue:2462|issue:2737", + "issue:2782|issue:968", + "issue:3162|issue:6958", + "issue:3280|issue:3391", + "issue:3480|issue:5170", + "issue:3480|issue:5336" + ], + "prepared_review_unit_hash": "8cd42ff87cd6c4c43d42c0a3709726f0f1b90616309fbc3b566ce9524a628bcd", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33213, + "estimated_input_tokens": 8304, + "estimated_eval_tokens": 16864 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous set of unrelated issues; the only strong duplicate-like pair is the CivitAI/from_single_file loading failure.", + "confidence": 0.71, + "canonical_issue_reason": "issue:9052 is the clearest, most general statement of the shared bug class (from_single_file failing on certain CivitAI checkpoints), so it is the best duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9052 is specific enough to be actionable while still broad enough to subsume issue:11127's more specific Flux/CivitAI example.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7690", + "right": "issue:7939", + "accept": false, + "reason": "Both are loading-related, but one is a tokenizer load failure and the other is a model-from-folder load failure; too different to treat as the same bug." + }, + { + "left": "issue:10421", + "right": "issue:12505", + "accept": false, + "reason": "Different problems in different code paths: GPU memory leak during pipeline placement vs tgate errors in img2img pipelines." + }, + { + "left": "issue:10855", + "right": "issue:12633", + "accept": false, + "reason": "Gradient accumulation optimizer stepping is unrelated to DDIM timestep list mismatches." + }, + { + "left": "issue:10913", + "right": "issue:11675", + "accept": false, + "reason": "Dataset-from-disk loading and pretrained LoRA weight loading are unrelated failures." + }, + { + "left": "issue:11000", + "right": "issue:11214", + "accept": false, + "reason": "Both mention missing arguments, but they affect different models and signatures; not the same underlying bug." + }, + { + "left": "issue:1101", + "right": "issue:3418", + "accept": false, + "reason": "Scheduler feature request vs LoRA training-script enhancement; different changes." + }, + { + "left": "issue:11127", + "right": "issue:9052", + "accept": true, + "reason": "Both report from_single_file failing on certain CivitAI checkpoints; 11127 is a specific example of the broader loading bug described by 9052." + }, + { + "left": "issue:1117", + "right": "issue:202", + "accept": false, + "reason": "Different test additions for different areas, not the same defect or change." + }, + { + "left": "issue:11470", + "right": "issue:8659", + "accept": false, + "reason": "Training-script logging/images and checkpoint-resume/inference loading are separate issues." + }, + { + "left": "issue:1181", + "right": "issue:725", + "accept": false, + "reason": "Model-from-scratch training support and memory-footprint loading improvements are unrelated." + }, + { + "left": "issue:1197", + "right": "issue:939", + "accept": false, + "reason": "General community pipeline request vs using multiple community pipes in a list; not the same issue." + }, + { + "left": "issue:11989", + "right": "issue:9331", + "accept": false, + "reason": "Black inference images and a scheduler index_for_timestep bug do not describe the same failure mode." + }, + { + "left": "issue:12437", + "right": "issue:9772", + "accept": false, + "reason": "Different feature-support requests for different models/components." + }, + { + "left": "issue:1407", + "right": "issue:1821", + "accept": false, + "reason": "Onnx feature_extractor error and inpaint negative_prompt behavior are unrelated." + }, + { + "left": "issue:1478", + "right": "issue:3769", + "accept": false, + "reason": "Both are output-shape/image-generation problems, but they affect different pipelines and concrete bugs." + }, + { + "left": "issue:177", + "right": "issue:244", + "accept": false, + "reason": "Documentation tasks for different topics; not duplicates." + }, + { + "left": "issue:1857", + "right": "issue:2871", + "accept": false, + "reason": "Poor test coverage for UnCLIP and latent output handling in inpainting are separate concerns." + }, + { + "left": "issue:2392", + "right": "issue:3428", + "accept": false, + "reason": "Img2Img prompt_embeds validation is a concrete API bug; the other is a broader unexpected-result report with no clear same code-path." + }, + { + "left": "issue:2462", + "right": "issue:2737", + "accept": false, + "reason": "Different feature requests in different subsystems." + }, + { + "left": "issue:2782", + "right": "issue:968", + "accept": false, + "reason": "Sequential offloading error is not the same as slow model loading, even though both relate to loading/memory." + }, + { + "left": "issue:3162", + "right": "issue:6958", + "accept": false, + "reason": "Different model/configuration bugs with no shared code-path evidence." + }, + { + "left": "issue:3280", + "right": "issue:3391", + "accept": false, + "reason": "Both involve DeepFloyd IF, but one is scheduler compatibility and the other is overexposure in the super-resolution pipeline." + }, + { + "left": "issue:3480", + "right": "issue:5170", + "accept": false, + "reason": "LoRA scaling and LoRA removal after loading are different features." + }, + { + "left": "issue:3480", + "right": "issue:5336", + "accept": false, + "reason": "Both add pipeline parameters, but they target different parameters and pipelines, so they are not the same change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:11127 vs issue:9052, is reasonably supported by the titles as the same from_single_file / CivitAI checkpoint loading bug class. The rejected pairs are described at a high level without overclaiming evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 299, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11973", + "issue:1540", + "issue:1709", + "issue:1941", + "issue:2091", + "issue:2213", + "issue:2265", + "issue:2412", + "issue:2799", + "issue:2966", + "issue:3379", + "issue:3591", + "issue:3902", + "issue:4109", + "issue:4446", + "issue:4453", + "issue:4483", + "issue:4674", + "issue:4985", + "issue:5112", + "issue:5465", + "issue:5963", + "issue:5977", + "issue:6001", + "issue:6028", + "issue:6162", + "issue:6518", + "issue:6565", + "issue:6783", + "issue:6832", + "issue:6880", + "issue:7102", + "issue:7168", + "issue:7178", + "issue:7450", + "issue:7485", + "issue:7497", + "issue:8380", + "issue:8909", + "issue:9138", + "issue:9210", + "issue:9227", + "issue:9320", + "issue:9350", + "issue:9637", + "issue:9756", + "issue:988" + ], + "soft_pairs": [ + "issue:3902|issue:7497", + "issue:4109|issue:6001", + "issue:4446|issue:6880", + "issue:4453|issue:988", + "issue:5112|issue:7485", + "issue:5963|issue:7168", + "issue:5977|issue:9756", + "issue:6162|issue:6832", + "issue:6518|issue:9138", + "issue:7102|issue:7450", + "issue:8380|issue:9210", + "issue:2091|issue:2213", + "issue:2799|issue:5465", + "issue:1540|issue:2412", + "issue:9350|issue:9637", + "issue:6565|issue:8909", + "issue:3591|issue:6565", + "issue:4674|issue:6028", + "issue:1709|issue:2265", + "issue:4483|issue:6783", + "issue:2966|issue:4985", + "issue:1941|issue:3379", + "issue:11973|issue:9227", + "issue:7178|issue:9320" + ], + "prepared_review_unit_hash": "975c67863f16f5fd8b18a6f5f5c489c6be2058449427b39b8559ac482bffb533", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33575, + "estimated_input_tokens": 8394, + "estimated_eval_tokens": 17044 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are standalone bugs across different pipelines, scripts, and error modes. None of the soft pairs look like true duplicates or mergeable same-fix cases, so I would keep them separate.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9227 is the most concrete and actionable recent report in the set; it has a specific FLUX/Dreambooth failure, clear reproduction context, and an active discussion.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3902", + "right": "issue:7497", + "accept": false, + "reason": "Both mention control-related concepts, but one is about control-a-video and the other is a subject-attribute control request; different bugs and workflows." + }, + { + "left": "issue:4109", + "right": "issue:6001", + "accept": false, + "reason": "Both involve StableDiffusionXLInpaintPipeline, but one is a general failure and the other is a from_single_file refiner type-return bug; not the same code path." + }, + { + "left": "issue:4446", + "right": "issue:6880", + "accept": false, + "reason": "Custom text_encoder loading in from_single_file and Custom Diffusion retrieve.py are unrelated failures in different features." + }, + { + "left": "issue:4453", + "right": "issue:988", + "accept": false, + "reason": "LoRA export/compatibility with A1111 and FSDP training are different problems; same broad ecosystem only." + }, + { + "left": "issue:5112", + "right": "issue:7485", + "accept": false, + "reason": "One is an SDXL ControlNet shape conversion error, the other is a safety_checker callable error; different failure locations." + }, + { + "left": "issue:5963", + "right": "issue:7168", + "accept": false, + "reason": "Both mention embeds, but Kandinsky prompt/embeds handling and IP-Adapter image embeds are separate APIs and bugs." + }, + { + "left": "issue:5977", + "right": "issue:9756", + "accept": false, + "reason": "Saving LoRA keys and a missing notebook file are unrelated." + }, + { + "left": "issue:6162", + "right": "issue:6832", + "accept": false, + "reason": "Both are IP-Adapter related, but one is an SDXL generation shape mismatch and the other is a Face ID community pipeline tuple-shape error; not the same defect." + }, + { + "left": "issue:6518", + "right": "issue:9138", + "accept": false, + "reason": "Different scripts and different type errors; no shared underlying bug." + }, + { + "left": "issue:7102", + "right": "issue:7450", + "accept": false, + "reason": "ControlNet multi-prompt behavior and AutoencoderKL threading/memory issues are unrelated." + }, + { + "left": "issue:8380", + "right": "issue:9210", + "accept": false, + "reason": "ControlNet forward default-arg cleanup and a dataset column validation error are not the same issue." + }, + { + "left": "issue:2091", + "right": "issue:2213", + "accept": false, + "reason": "Both touch training/distributed execution, but one is checkpoint resume on multi-GPU and the other is TPU hang/no distributed environment; different root causes." + }, + { + "left": "issue:2799", + "right": "issue:5465", + "accept": false, + "reason": "Checkpoint/pipeline mismatch validation and DiffEdit inpaint_strength handling are unrelated." + }, + { + "left": "issue:1540", + "right": "issue:2412", + "accept": false, + "reason": "Both are Dreambooth-related, but the missing MODEL_TYPE startup failure and a generic Dreambooth error are too vague to treat as the same bug." + }, + { + "left": "issue:9350", + "right": "issue:9637", + "accept": false, + "reason": "A deprecation warning for txt_ids shape and an xFormers attention-mask/bfloat16 failure are different issues." + }, + { + "left": "issue:6565", + "right": "issue:8909", + "accept": false, + "reason": "DPO training instability and text_encoder_ti saving problems are unrelated training bugs." + }, + { + "left": "issue:3591", + "right": "issue:6565", + "accept": false, + "reason": "ControlNet loss-curve confusion is not the same as DPO diffusion loss spikes/noised samples." + }, + { + "left": "issue:4674", + "right": "issue:6028", + "accept": false, + "reason": "A DiffusionPipeline width/height memory leak and Kandinsky OOM are different memory issues in different code paths." + }, + { + "left": "issue:1709", + "right": "issue:2265", + "accept": false, + "reason": "Both reference ONNXRuntime, but the img2img tensor.astype error and train_unconditional incompatibility are distinct problems." + }, + { + "left": "issue:4483", + "right": "issue:6783", + "accept": false, + "reason": "Generator nondeterminism in SDXL and PixArt generations being messed up are not the same concrete bug." + }, + { + "left": "issue:2966", + "right": "issue:4985", + "accept": false, + "reason": "Multi-GPU train_instruct_pix2pix behavior and SDXL InstructPix2Pix OOM are different failure modes." + }, + { + "left": "issue:1941", + "right": "issue:3379", + "accept": false, + "reason": "UnCLIP custom pipeline loading and Dreambooth intermediate checkpoint loading are unrelated." + }, + { + "left": "issue:11973", + "right": "issue:9227", + "accept": false, + "reason": "Both are Flux Dreambooth training reports, but one is a missing required CLI argument and the other is a model-specific training failure; not the same underlying bug." + }, + { + "left": "issue:7178", + "right": "issue:9320", + "accept": false, + "reason": "Video export behavior and custom diffusion model save/load issues are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 300, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10302", + "issue:10467", + "issue:10803", + "issue:10804", + "issue:11169", + "issue:13035", + "issue:13256", + "issue:1407", + "issue:1784", + "issue:2185", + "issue:2213", + "issue:2250", + "issue:2410", + "issue:2737", + "issue:3381", + "issue:3788", + "issue:4037", + "issue:4139", + "issue:4219", + "issue:4456", + "issue:4488", + "issue:4720", + "issue:4868", + "issue:5017", + "issue:5178", + "issue:5628", + "issue:5794", + "issue:5875", + "issue:5932", + "issue:5957", + "issue:6176", + "issue:6363", + "issue:6368", + "issue:6445", + "issue:6517", + "issue:6533", + "issue:7012", + "issue:7183", + "issue:7243", + "issue:7330", + "issue:7569", + "issue:7708", + "issue:7751", + "issue:817", + "issue:8612", + "issue:9186", + "issue:9461", + "issue:9619" + ], + "soft_pairs": [ + "issue:13256|issue:7708", + "issue:4219|issue:4488", + "issue:4456|issue:4868", + "issue:2250|issue:3381", + "issue:2185|issue:5794", + "issue:1407|issue:5875", + "issue:5628|issue:7183", + "issue:4037|issue:7751", + "issue:6363|issue:6445", + "issue:2213|issue:2410", + "issue:10467|issue:10804", + "issue:7012|issue:7330", + "issue:10803|issue:7243", + "issue:11169|issue:13035", + "issue:1784|issue:2737", + "issue:3788|issue:6533", + "issue:4139|issue:817", + "issue:5017|issue:9186", + "issue:5957|issue:9619", + "issue:5932|issue:6176", + "issue:10302|issue:9461", + "issue:4720|issue:8612", + "issue:5178|issue:7569", + "issue:6368|issue:6517" + ], + "prepared_review_unit_hash": "b23dfef83c5acc298d2f2b6768bd4276443e09217a2b4f39028918dd29c1ed98", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34359, + "estimated_input_tokens": 8590, + "estimated_eval_tokens": 17436 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set: it groups many unrelated issues across inference, training, schedulers, LoRA, ControlNet, and dependency/import problems. The soft edges are mostly superficial text similarity and should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13256", + "right": "issue:7708", + "accept": false, + "reason": "Both mention tokenizer/package compatibility, but one is a vague version-compatibility note and the other is a missing tokenizer/config file in SDXL; different root causes." + }, + { + "left": "issue:4219", + "right": "issue:4488", + "accept": false, + "reason": "ControlNet checkpoint conversion support vs SDXL original-checkpoint loading behavior; not the same bug or change." + }, + { + "left": "issue:4456", + "right": "issue:4868", + "accept": false, + "reason": "Different loader/runtime failures: one is a LoRA ModuleList indexing error, the other is a generic __getattr__ AttributeError." + }, + { + "left": "issue:2250", + "right": "issue:3381", + "accept": false, + "reason": "Dreambooth inference not generating vs discolored inpaint batch output; both are inference issues but affect different pipelines and symptoms." + }, + { + "left": "issue:2185", + "right": "issue:5794", + "accept": false, + "reason": "Missing training config files vs missing tokenizer/config.json for SDXL; unrelated file-availability problems." + }, + { + "left": "issue:1407", + "right": "issue:5875", + "accept": false, + "reason": "ONNX pipeline feature_extractor error from a prior PR vs SDXL-to-ONNX prompt error; different pipeline and failure mode." + }, + { + "left": "issue:5628", + "right": "issue:7183", + "accept": false, + "reason": "LMS/Karras denoising-end offset bug vs scheduler class-name metadata mismatch; not the same underlying scheduler defect." + }, + { + "left": "issue:4037", + "right": "issue:7751", + "accept": false, + "reason": "ControlNet FSDP validation errors vs AnimateDiff conversion errors; unrelated training vs conversion paths." + }, + { + "left": "issue:6363", + "right": "issue:6445", + "accept": false, + "reason": "FP16 gradient unscale error during LoRA training vs dataset caption_column validation error; completely different causes." + }, + { + "left": "issue:2213", + "right": "issue:2410", + "accept": false, + "reason": "TPU distributed-environment hang in training script vs missing output artifacts in flax training; different training stages and symptoms." + }, + { + "left": "issue:10467", + "right": "issue:10804", + "accept": false, + "reason": "Different FLUX-related problems: one mentions example-code/compiler artifact text, the other a LoRA load failure; no shared concrete bug." + }, + { + "left": "issue:7012", + "right": "issue:7330", + "accept": false, + "reason": "Weird images with v_prediction vs FP16 gradient unscale error; one is model quality, the other is training precision handling." + }, + { + "left": "issue:10803", + "right": "issue:7243", + "accept": false, + "reason": "A new SANA/ResAdapter pipeline request vs a domain-consistent resolution adapter proposal; different feature requests." + }, + { + "left": "issue:11169", + "right": "issue:13035", + "accept": false, + "reason": "Numpy binary incompatibility when importing schedulers vs missing MT5Tokenizer from transformers; different dependency/import failures." + }, + { + "left": "issue:1784", + "right": "issue:2737", + "accept": false, + "reason": "Scheduler support for multiple timesteps vs CLIP guidance support for one scheduler; related only at a high level, not the same change." + }, + { + "left": "issue:3788", + "right": "issue:6533", + "accept": false, + "reason": "New UniControl pipeline request vs Marigold depth-estimation pipeline request; separate feature additions." + }, + { + "left": "issue:4139", + "right": "issue:817", + "accept": false, + "reason": "Custom diffusion attention precision casting vs dreambooth mixed-precision not working; similar theme but different code paths and targets." + }, + { + "left": "issue:5017", + "right": "issue:9186", + "accept": false, + "reason": "Request for ControlNet control mode vs differential diffusion with ControlNet pipeline; both are ControlNet-related but not the same feature." + }, + { + "left": "issue:5957", + "right": "issue:9619", + "accept": false, + "reason": "LoRACompatibleConv padding_mode support vs interrupt support in ControlNet pipeline; unrelated functionality." + }, + { + "left": "issue:5932", + "right": "issue:6176", + "accept": false, + "reason": "Channel-mismatch during SDXL inference vs LoRA weight loading after DreamBooth LoRA training; different failure points." + }, + { + "left": "issue:10302", + "right": "issue:9461", + "accept": false, + "reason": "FP8 inference noise bug vs vague Flux inference report; too little evidence and no matching concrete symptom." + }, + { + "left": "issue:4720", + "right": "issue:8612", + "accept": false, + "reason": "AutoencoderTiny incompatibility with img2img/inpainting vs PixArt-sigma mixed-precision inference issue; different models and root causes." + }, + { + "left": "issue:5178", + "right": "issue:7569", + "accept": false, + "reason": "DPM SDE resolution-change bug vs Mac-specific FreeU test failure; unrelated subsystems and environments." + }, + { + "left": "issue:6368", + "right": "issue:6517", + "accept": false, + "reason": "Unsupported LoRA checkpoint layer alpha vs controlnet SDXL img2img import error; different loaders and error types." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 301, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10035", + "issue:10638", + "issue:10963", + "issue:1099", + "issue:11047", + "issue:11885", + "issue:1255", + "issue:12606", + "issue:13162", + "issue:1519", + "issue:2097", + "issue:2674", + "issue:3901", + "issue:4266", + "issue:4578", + "issue:4631", + "issue:4701", + "issue:5485", + "issue:5553", + "issue:5598", + "issue:5605", + "issue:5743", + "issue:5917", + "issue:5919", + "issue:6093", + "issue:6162", + "issue:6399", + "issue:6529", + "issue:6574", + "issue:6801", + "issue:6949", + "issue:7061", + "issue:7221", + "issue:7239", + "issue:7418", + "issue:7575", + "issue:7695", + "issue:7735", + "issue:7950", + "issue:8451", + "issue:8502", + "issue:8551", + "issue:8613", + "issue:9336", + "issue:9477", + "issue:9568", + "issue:9889" + ], + "soft_pairs": [ + "issue:10638|issue:6801", + "issue:10963|issue:12606", + "issue:1099|issue:1255", + "issue:2674|issue:3901", + "issue:4578|issue:7221", + "issue:5919|issue:9477", + "issue:6093|issue:6399", + "issue:7061|issue:7418", + "issue:8451|issue:9336", + "issue:11885|issue:13162", + "issue:4631|issue:6949", + "issue:2097|issue:6162", + "issue:7735|issue:8613", + "issue:5553|issue:5917", + "issue:11047|issue:8502", + "issue:1519|issue:5485", + "issue:4701|issue:7575", + "issue:5598|issue:9477", + "issue:5605|issue:6529", + "issue:5743|issue:8551", + "issue:10035|issue:9568", + "issue:7950|issue:9889", + "issue:4266|issue:7239", + "issue:6574|issue:7695" + ], + "prepared_review_unit_hash": "6231154b344d7df6b6b3601b402f288a972efbed14d5d02611b365d0c4c6a7f2", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33314, + "estimated_input_tokens": 8329, + "estimated_eval_tokens": 16914 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The items are mostly unrelated bug reports spanning different models, pipelines, schedulers, and training scripts. The soft-edge pairs look like title-level similarity only, not the same underlying defect, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue is a clear canonical representative for this cluster because the set is heterogeneous and does not describe one underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11885 is the strongest standalone issue candidate because it has the most detailed report and a concrete, reproducible LoRA-loading failure in FLUX.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10638", + "right": "issue:6801", + "accept": false, + "reason": "Both mention schedulers, but one is about running a specific SDXL script and the other about loading a scheduler; different failure modes and code paths." + }, + { + "left": "issue:10963", + "right": "issue:12606", + "accept": false, + "reason": "Import failure for AutoencoderKLWan is unrelated to a wrong-import issue in a conversion script." + }, + { + "left": "issue:1099", + "right": "issue:1255", + "accept": false, + "reason": "Dreambooth class-name issue and missing DPMSolverMultistepScheduler import are different components and bugs." + }, + { + "left": "issue:2674", + "right": "issue:3901", + "accept": false, + "reason": "LoRA/safetensors requirement mismatch is unrelated to a missing requirements.txt file." + }, + { + "left": "issue:4578", + "right": "issue:7221", + "accept": false, + "reason": "Documentation complaint about VAE usage does not match a noisy-image inference problem." + }, + { + "left": "issue:5919", + "right": "issue:9477", + "accept": false, + "reason": "Missing set_lora_layer on Linear and a GatheredParameters callability error are different runtime failures." + }, + { + "left": "issue:6093", + "right": "issue:6399", + "accept": false, + "reason": "push_to_hub behavior in a training script is unrelated to LoRA visibility in UI tools." + }, + { + "left": "issue:7061", + "right": "issue:7418", + "accept": false, + "reason": "Incorrect docstring and a bug in a community pipeline file are not the same defect." + }, + { + "left": "issue:8451", + "right": "issue:9336", + "accept": false, + "reason": "Unet-freezing/training-script complaint is unrelated to IPAdapter gradients being left on." + }, + { + "left": "issue:11885", + "right": "issue:13162", + "accept": false, + "reason": "Both involve Flux and LoRA, but one is a loading bug and the other is support for a different model variant; not the same concrete change." + }, + { + "left": "issue:4631", + "right": "issue:6949", + "accept": false, + "reason": "Both involve DPMSolverSinglestepScheduler, but one is a scalar-conversion runtime error and the other produces blank images at even step counts." + }, + { + "left": "issue:2097", + "right": "issue:6162", + "accept": false, + "reason": "UNet1D usage and IP-Adapter SDXL shape mismatch are unrelated model-specific issues." + }, + { + "left": "issue:7735", + "right": "issue:8613", + "accept": false, + "reason": "Missing caption column and OOM while using caption_column are different dataset/training problems." + }, + { + "left": "issue:5553", + "right": "issue:5917", + "accept": false, + "reason": "Custom model loading failure and convert_ldm_unet_checkpoint regression are distinct conversion/loading issues." + }, + { + "left": "issue:11047", + "right": "issue:8502", + "accept": false, + "reason": "ControlNet PAG inpaint support and SD3 512x512 generation are different feature/issues." + }, + { + "left": "issue:1519", + "right": "issue:5485", + "accept": false, + "reason": "Apple M2 noisy outputs and slow ControlNet batch inference are different performance/correctness bugs." + }, + { + "left": "issue:4701", + "right": "issue:7575", + "accept": false, + "reason": "AutoencoderKL CUDA differentiation error and Transformer2DModel adanorm shape error are separate code paths." + }, + { + "left": "issue:5598", + "right": "issue:9477", + "accept": false, + "reason": "Text-encoder modification bug and GatheredParameters callability error are unrelated." + }, + { + "left": "issue:5605", + "right": "issue:6529", + "accept": false, + "reason": "SDXL dreambooth caption support and Fooocus inpaint are not the same bug or change." + }, + { + "left": "issue:5743", + "right": "issue:8551", + "accept": false, + "reason": "Training stuck at 0% in LCD and VAE device placement in inpaint are different problems." + }, + { + "left": "issue:10035", + "right": "issue:9568", + "accept": false, + "reason": "Both are CogVideoX-related, but one is a memory regression and the other is a reconstruction failure; different symptoms and likely different fixes." + }, + { + "left": "issue:7950", + "right": "issue:9889", + "accept": false, + "reason": "General PyTorch/diffusers incompatibility and a specific SDXL-turbo segmentation fault are not the same issue." + }, + { + "left": "issue:4266", + "right": "issue:7239", + "accept": false, + "reason": "Both mention SDXL training errors, but one is a broadcast-compatibility problem on Mac and the other is a matrix-shape mismatch in Dreambooth LoRA." + }, + { + "left": "issue:6574", + "right": "issue:7695", + "accept": false, + "reason": "A warning about SVD pipeline tensor range and ignored ip_adapter_masks are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 302, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10303", + "issue:10596", + "issue:11436", + "issue:11497", + "issue:11839", + "issue:12610", + "issue:1746", + "issue:1964", + "issue:2044", + "issue:2392", + "issue:2786", + "issue:2966", + "issue:3182", + "issue:3287", + "issue:3401", + "issue:3958", + "issue:3970", + "issue:3975", + "issue:4206", + "issue:4241", + "issue:4255", + "issue:4269", + "issue:4353", + "issue:4383", + "issue:4438", + "issue:4993", + "issue:5079", + "issue:5171", + "issue:5335", + "issue:5368", + "issue:5372", + "issue:5429", + "issue:5570", + "issue:579", + "issue:5980", + "issue:6161", + "issue:6486", + "issue:7914", + "issue:7968", + "issue:8393", + "issue:8482", + "issue:8507", + "issue:8543", + "issue:861", + "issue:8615", + "issue:8850", + "issue:9241", + "issue:9926" + ], + "soft_pairs": [ + "issue:4206|issue:8850", + "issue:10596|issue:11839", + "issue:1964|issue:5429", + "issue:2044|issue:2966", + "issue:2786|issue:861", + "issue:3287|issue:4241", + "issue:3287|issue:5079", + "issue:3970|issue:4353", + "issue:3975|issue:7914", + "issue:4241|issue:8615", + "issue:4269|issue:5171", + "issue:4383|issue:5980", + "issue:4438|issue:5372", + "issue:4993|issue:5570", + "issue:6161|issue:9241", + "issue:6486|issue:8543", + "issue:8482|issue:9926", + "issue:1746|issue:579", + "issue:3401|issue:5368", + "issue:3182|issue:3958", + "issue:11436|issue:12610", + "issue:2392|issue:5335", + "issue:4255|issue:8393", + "issue:10303|issue:7968", + "issue:11497|issue:8507" + ], + "prepared_review_unit_hash": "7218ee6030fe919c6541a128a10c546ad9d6a2f1fec9b1225eeccdf96d21d628", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34053, + "estimated_input_tokens": 8514, + "estimated_eval_tokens": 17284 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is a loose collection of mostly unrelated issues; there is no strong duplicate cluster or single canonical bug/change to consolidate around.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4206", + "right": "issue:8850", + "accept": false, + "reason": "Different problems: SDXL controlnet training breakage vs community-pipelines mirror availability." + }, + { + "left": "issue:10596", + "right": "issue:11839", + "accept": false, + "reason": "Both are Flux pipeline shape/input errors, but they target different pipelines and failure modes." + }, + { + "left": "issue:1964", + "right": "issue:5429", + "accept": false, + "reason": "PyTorch 2.0 support request vs support for models without VAE; unrelated feature areas." + }, + { + "left": "issue:2044", + "right": "issue:2966", + "accept": false, + "reason": "DDP dtype attribute error vs multi-GPU example training issue; not the same bug." + }, + { + "left": "issue:2786", + "right": "issue:861", + "accept": false, + "reason": "Depth2Img numpy/transformers compatibility issue vs checkpoint conversion failure." + }, + { + "left": "issue:3287", + "right": "issue:4241", + "accept": false, + "reason": "General DDPM model support vs LoRA fine-tuning support; different changes." + }, + { + "left": "issue:3287", + "right": "issue:5079", + "accept": false, + "reason": "OpenAI improved DDPM support vs loading a specific Lyrocis LoRA format." + }, + { + "left": "issue:3970", + "right": "issue:4353", + "accept": false, + "reason": "Prompt embeddings not working vs poor image quality; same model family, different issue." + }, + { + "left": "issue:3975", + "right": "issue:7914", + "accept": false, + "reason": "Error-message typo vs textual inversion failure under device_map='balanced'." + }, + { + "left": "issue:4241", + "right": "issue:8615", + "accept": false, + "reason": "LoRA fine-tuning support request vs fuse_lora support on SD3; related theme but not the same concrete change." + }, + { + "left": "issue:4269", + "right": "issue:5171", + "accept": false, + "reason": "ControlNet check_inputs TypeError vs SDXL init rejecting safety_checker kwarg." + }, + { + "left": "issue:4383", + "right": "issue:5980", + "accept": false, + "reason": "How to use a trained ControlNet in A1111 vs file format question for SDXL ControlNet models." + }, + { + "left": "issue:4438", + "right": "issue:5372", + "accept": false, + "reason": "Generic SDXL pipeline error vs question about using safety_checker; not the same defect." + }, + { + "left": "issue:4993", + "right": "issue:5570", + "accept": false, + "reason": "Missing CLIPImageProjection export vs missing compute_snr helper in training_utils." + }, + { + "left": "issue:6161", + "right": "issue:9241", + "accept": false, + "reason": "Sigma-selection algorithm issue vs LoRA conversion script request; unrelated." + }, + { + "left": "issue:6486", + "right": "issue:8543", + "accept": false, + "reason": "Dreambooth embedding retraction duplicated vs SD3 dreambooth gradient checkpointing failure." + }, + { + "left": "issue:8482", + "right": "issue:9926", + "accept": false, + "reason": "Support for SD3-medium model vs GGUF quantize/speed support; different feature requests." + }, + { + "left": "issue:1746", + "right": "issue:579", + "accept": false, + "reason": "Flax pipeline error vs missing config.json in textual inversion; unrelated." + }, + { + "left": "issue:3401", + "right": "issue:5368", + "accept": false, + "reason": "num_class_embeds/class_labels validation vs fp16 attention dtype mismatch." + }, + { + "left": "issue:3182", + "right": "issue:3958", + "accept": false, + "reason": "lpw_stable_diffusion CPU offload bug vs LoRA offload issue; both offload-related but distinct code paths." + }, + { + "left": "issue:11436", + "right": "issue:12610", + "accept": false, + "reason": "Half/BFloat16 matmul dtype mismatch vs CPU/GPU device mismatch under group offload; different underlying causes." + }, + { + "left": "issue:2392", + "right": "issue:5335", + "accept": false, + "reason": "prompt_embeds handling bug vs local deployment/network access question." + }, + { + "left": "issue:4255", + "right": "issue:8393", + "accept": false, + "reason": "Image quality comparison vs slow checkpoint shard loading; unrelated." + }, + { + "left": "issue:10303", + "right": "issue:7968", + "accept": false, + "reason": "Attention tensor permutation bug vs attention implementation in the motion module; not enough evidence of the same defect." + }, + { + "left": "issue:11497", + "right": "issue:8507", + "accept": false, + "reason": "SD3 ControlNet dataset cache-key bug vs SD3 DreamBooth text-encoder training failure; same project area, different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 303, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11329", + "issue:11437", + "issue:11874", + "issue:13221", + "issue:1405", + "issue:1486", + "issue:1990", + "issue:2331", + "issue:2332", + "issue:2437", + "issue:3233", + "issue:3643", + "issue:3699", + "issue:3808", + "issue:3854", + "issue:3975", + "issue:4116", + "issue:4117", + "issue:4325", + "issue:4414", + "issue:4846", + "issue:497", + "issue:5016", + "issue:5017", + "issue:5233", + "issue:5284", + "issue:5485", + "issue:5866", + "issue:6093", + "issue:6222", + "issue:6242", + "issue:6250", + "issue:6288", + "issue:6412", + "issue:6533", + "issue:6570", + "issue:6958", + "issue:7142", + "issue:7197", + "issue:7440", + "issue:7640", + "issue:9350", + "issue:9476", + "issue:9530", + "issue:970", + "issue:9704", + "issue:9889", + "issue:9901" + ], + "soft_pairs": [ + "issue:1405|issue:3854", + "issue:1486|issue:4414", + "issue:1990|issue:2437", + "issue:4116|issue:6958", + "issue:5016|issue:6288", + "issue:5017|issue:6412", + "issue:6242|issue:7440", + "issue:9704|issue:9901", + "issue:11437|issue:9476", + "issue:6570|issue:9350", + "issue:11329|issue:9530", + "issue:3643|issue:7640", + "issue:2331|issue:6533", + "issue:4325|issue:5233", + "issue:7142|issue:9889", + "issue:3699|issue:970", + "issue:11874|issue:13221", + "issue:2332|issue:4846", + "issue:3233|issue:6093", + "issue:3808|issue:4117", + "issue:5284|issue:5866", + "issue:3975|issue:497", + "issue:5485|issue:6222", + "issue:6250|issue:7197" + ], + "prepared_review_unit_hash": "52f786f536c724305ac1875e45025a5444e25454277012fcbbfa0ad610fcb7c3", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33666, + "estimated_input_tokens": 8417, + "estimated_eval_tokens": 17090 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a loose topical collection of unrelated diffusion/training issues, not a true duplicate cluster. No soft pair looks like the same underlying bug or change.", + "confidence": 0.95, + "canonical_issue_reason": "No strong duplicate anchor emerges; issue:9476 is the most detailed, specific, and current open bug report among the set, so it is the best fallback representative.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue:9476 has the clearest concrete failure mode, traceback, and active status, making it the most actionable issue to keep as the cluster representative.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1405", + "right": "issue:3854", + "accept": false, + "reason": "Different failures: attention slicing regression vs a ckpt-loading hang/lock; only broad pipeline/runtime overlap." + }, + { + "left": "issue:1486", + "right": "issue:4414", + "accept": false, + "reason": "One is a scheduler bug with prediction_type='sample'; the other is a new pipeline proposal/sample request. Not the same bug." + }, + { + "left": "issue:1990", + "right": "issue:2437", + "accept": false, + "reason": "Depth2ImgVariation proposal and ControlNet pipeline request are separate feature ideas." + }, + { + "left": "issue:4116", + "right": "issue:6958", + "accept": false, + "reason": "Img2Img embedding passing failure vs a motion model configuration mismatch; different code paths and symptoms." + }, + { + "left": "issue:5016", + "right": "issue:6288", + "accept": false, + "reason": "Text-encoder training failure in SDXL LoRA vs a learning-rate tuning complaint for LCM LoRA; not the same defect." + }, + { + "left": "issue:5017", + "right": "issue:6412", + "accept": false, + "reason": "Control mode request vs ONNX annotators request; unrelated feature asks." + }, + { + "left": "issue:6242", + "right": "issue:7440", + "accept": false, + "reason": "A request for logs vs dataset availability issue; no shared bug/change." + }, + { + "left": "issue:9704", + "right": "issue:9901", + "accept": false, + "reason": "Different runtime errors in different components: cuDNN execution plan failure vs tuple/shape error with AttnProcessor." + }, + { + "left": "issue:11437", + "right": "issue:9476", + "accept": false, + "reason": "Training outputs not being created vs a flux validation dtype mismatch; separate training problems." + }, + { + "left": "issue:6570", + "right": "issue:9350", + "accept": false, + "reason": "Tensor size mismatch in SDXL InstructPix2Pix vs a deprecation warning about txt_ids shape; not the same issue." + }, + { + "left": "issue:11329", + "right": "issue:9530", + "accept": false, + "reason": "Both mention memory, but one is a bitsandbytes quantization leak and the other is GPU memory retention after torch.compile; different causes." + }, + { + "left": "issue:3643", + "right": "issue:7640", + "accept": false, + "reason": "Both touch latent decoding/dimensions, but the errors are distinct: INT_MAX overflow vs from_single_file dimension/runtime failure." + }, + { + "left": "issue:2331", + "right": "issue:6533", + "accept": false, + "reason": "ControlNet pipeline addition vs Marigold pipeline proposal; separate pipeline requests." + }, + { + "left": "issue:4325", + "right": "issue:5233", + "accept": false, + "reason": "Black-image/NaN casting issue vs incorrect step value passed to callback; unrelated bugs." + }, + { + "left": "issue:7142", + "right": "issue:9889", + "accept": false, + "reason": "Both are SDXL inference-related, but one is output divergence and the other is a segfault; not the same code-path bug." + }, + { + "left": "issue:3699", + "right": "issue:970", + "accept": false, + "reason": "A platform/CUDA availability error vs failing integration tests; no shared underlying defect." + }, + { + "left": "issue:11874", + "right": "issue:13221", + "accept": false, + "reason": "Different LoRA-related requests/bugs: exclude_modules matching vs Zimage LoRA support." + }, + { + "left": "issue:2332", + "right": "issue:4846", + "accept": false, + "reason": "Inference-after-training error vs strange training result; same general workflow but not the same concrete failure." + }, + { + "left": "issue:3233", + "right": "issue:6093", + "accept": false, + "reason": "Exporting trained LoRA to safetensors vs push_to_hub repo creation behavior; unrelated functionality." + }, + { + "left": "issue:3808", + "right": "issue:4117", + "accept": false, + "reason": "Different Dreambooth problems on different model families and stages; not one bug." + }, + { + "left": "issue:5284", + "right": "issue:5866", + "accept": false, + "reason": "ControlNet-LLLite support request vs AnimateDiff+ControlNet support request; separate feature asks." + }, + { + "left": "issue:3975", + "right": "issue:497", + "accept": false, + "reason": "Both are typo fixes, but they concern different error messages and methods." + }, + { + "left": "issue:5485", + "right": "issue:6222", + "accept": false, + "reason": "Batch inference performance issue vs scheduler batch-size dependency; different failure modes." + }, + { + "left": "issue:6250", + "right": "issue:7197", + "accept": false, + "reason": "Different KeyError contexts: SDXL VAE single-file loading vs training resume with shortest-edge preprocessing." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 304, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10227", + "issue:12009", + "issue:1537", + "issue:1672", + "issue:1708", + "issue:2185", + "issue:2659", + "issue:3112", + "issue:3355", + "issue:338", + "issue:3380", + "issue:339", + "issue:3683", + "issue:3730", + "issue:3937", + "issue:3948", + "issue:3975", + "issue:4279", + "issue:4321", + "issue:4403", + "issue:4416", + "issue:4554", + "issue:457", + "issue:5108", + "issue:5135", + "issue:5153", + "issue:5256", + "issue:5284", + "issue:5575", + "issue:5578", + "issue:5607", + "issue:5925", + "issue:5977", + "issue:6252", + "issue:6363", + "issue:6399", + "issue:6454", + "issue:6731", + "issue:6901", + "issue:7333", + "issue:7450", + "issue:8264", + "issue:9032", + "issue:9278", + "issue:9320", + "issue:9370", + "issue:9471" + ], + "soft_pairs": [ + "issue:3683|issue:7450", + "issue:10227|issue:12009", + "issue:3112|issue:5284", + "issue:338|issue:4416", + "issue:3380|issue:3730", + "issue:339|issue:4416", + "issue:3937|issue:4321", + "issue:4279|issue:5108", + "issue:4403|issue:5578", + "issue:4554|issue:9032", + "issue:5135|issue:5575", + "issue:5153|issue:7333", + "issue:5256|issue:5925", + "issue:5977|issue:6399", + "issue:6454|issue:6731", + "issue:6901|issue:9370", + "issue:8264|issue:9278", + "issue:6363|issue:9471", + "issue:1537|issue:457", + "issue:1708|issue:2185", + "issue:6252|issue:9320", + "issue:3355|issue:3975", + "issue:1672|issue:3948", + "issue:2659|issue:5607" + ], + "prepared_review_unit_hash": "505ea36e93a34aabb0b0c3209a3761c36ec4694d21e831764492a31a0819a59d", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33038, + "estimated_input_tokens": 8260, + "estimated_eval_tokens": 16776 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a set of soft-similarity false positives: the issues span many unrelated topics (LoRA loading/training, schedulers, MPS/ONNX, config files, image-size bugs, etc.), so none of the proposed pairs look like the same underlying bug.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a safe canonical representative; the items are thematically scattered and the soft links do not indicate true duplication.", + "best_pr_reason": "No PRs are present in this cluster.", + "soft_edge_verdicts": [ + { + "left": "issue:3683", + "right": "issue:7450", + "accept": false, + "reason": "Different problems: saving Multi-ControlNet vs AutoencoderKL multithreading issues." + }, + { + "left": "issue:10227", + "right": "issue:12009", + "accept": false, + "reason": "Unrelated loading failures: LoRA feature-size limitation vs Wan2.2 GGUF meta-tensor copy error." + }, + { + "left": "issue:3112", + "right": "issue:5284", + "accept": false, + "reason": "Different features: multi-ControlNet inpainting vs ControlNet-LLLite support." + }, + { + "left": "issue:338", + "right": "issue:4416", + "accept": false, + "reason": "One is LMS scheduler test coverage; the other is T2I Adapter tests, not the same bug/change." + }, + { + "left": "issue:3380", + "right": "issue:3730", + "accept": false, + "reason": "Embedding-size/config mismatch is not the same as incorrect inpainting timestep blending." + }, + { + "left": "issue:339", + "right": "issue:4416", + "accept": false, + "reason": "Different test targets: Karras VE tests vs T2I Adapter tests." + }, + { + "left": "issue:3937", + "right": "issue:4321", + "accept": false, + "reason": "Different runtime failures: ZeroscopeXL model loading vs ONNX DirectML black images on FP16." + }, + { + "left": "issue:4279", + "right": "issue:5108", + "accept": false, + "reason": "Both involve adapters/LoRAs, but the concrete issues differ: SDXL LoRA loading vs T2I Multiadapter exceptions." + }, + { + "left": "issue:4403", + "right": "issue:5578", + "accept": false, + "reason": "Different error classes and code paths: PyTorch Half LayerNorm support vs missing model_index.json." + }, + { + "left": "issue:4554", + "right": "issue:9032", + "accept": false, + "reason": "Kandinsky example OOM is unrelated to a TPU minimum working example request." + }, + { + "left": "issue:5135", + "right": "issue:5575", + "accept": false, + "reason": "LoRA quality/xformers discussion is unrelated to setting transformer_in hidden size during LoRA training." + }, + { + "left": "issue:5153", + "right": "issue:7333", + "accept": false, + "reason": "FreeU feature request is unrelated to image-to-video noise rectification." + }, + { + "left": "issue:5256", + "right": "issue:5925", + "accept": false, + "reason": "Different changes: Insta-Flow training support vs adding ZipLoRA for multi-LoRA inference." + }, + { + "left": "issue:5977", + "right": "issue:6399", + "accept": false, + "reason": "Different save/load problems: training-time key naming vs LoRA visibility in external UIs." + }, + { + "left": "issue:6454", + "right": "issue:6731", + "accept": false, + "reason": "Both are ValueErrors, but they stem from different bugs: FP16 gradient scaling vs unhandled pipeline class." + }, + { + "left": "issue:6901", + "right": "issue:9370", + "accept": false, + "reason": "Missing clip_sample in FlaxDDIMScheduler is unrelated to a missing pooled_projection argument." + }, + { + "left": "issue:8264", + "right": "issue:9278", + "accept": false, + "reason": "Train stopped at 0% is not the same as FLUX multigpu Dreambooth training." + }, + { + "left": "issue:6363", + "right": "issue:9471", + "accept": false, + "reason": "Different failure modes: FP16 gradient unscale during LoRA training vs CoreML resource-type inference error." + }, + { + "left": "issue:1537", + "right": "issue:457", + "accept": false, + "reason": "Different MPS issues: scheduler-specific behavior vs blue artifacts on MPS." + }, + { + "left": "issue:1708", + "right": "issue:2185", + "accept": false, + "reason": "One asks how to pass a YAML config for inference; the other reports missing training config files." + }, + { + "left": "issue:6252", + "right": "issue:9320", + "accept": false, + "reason": "Both mention finetuned weights, but the concrete problems differ: checkpoint LoRA loading vs saving/loading custom diffusion finetunes." + }, + { + "left": "issue:3355", + "right": "issue:3975", + "accept": false, + "reason": "Safetensors loading inconsistency is unrelated to a typo in an error message." + }, + { + "left": "issue:1672", + "right": "issue:3948", + "accept": false, + "reason": "Wrong output size after ckpt-to-diffusers conversion is unrelated to pale generated images." + }, + { + "left": "issue:2659", + "right": "issue:5607", + "accept": false, + "reason": "Batch-size-dependent image variation is a different bug than tensor-size mismatch during concatenation." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 305, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10239", + "issue:10653", + "issue:10865", + "issue:10954", + "issue:11343", + "issue:12147", + "issue:13261", + "issue:1363", + "issue:1633", + "issue:1730", + "issue:2599", + "issue:2988", + "issue:3008", + "issue:3280", + "issue:3303", + "issue:3441", + "issue:3880", + "issue:4111", + "issue:4720", + "issue:5079", + "issue:514", + "issue:5429", + "issue:5607", + "issue:5678", + "issue:5685", + "issue:5989", + "issue:6222", + "issue:6385", + "issue:6431", + "issue:6506", + "issue:6513", + "issue:6572", + "issue:6639", + "issue:6956", + "issue:7142", + "issue:7183", + "issue:7185", + "issue:7214", + "issue:7217", + "issue:7355", + "issue:7549", + "issue:8970", + "issue:9167", + "issue:9361", + "issue:9387", + "issue:9476", + "issue:9488" + ], + "soft_pairs": [ + "issue:3280|issue:7214", + "issue:10239|issue:10653", + "issue:2988|issue:5685", + "issue:7183|issue:7217", + "issue:3880|issue:4720", + "issue:10865|issue:7185", + "issue:10954|issue:9361", + "issue:12147|issue:13261", + "issue:1633|issue:5079", + "issue:3303|issue:514", + "issue:5429|issue:6513", + "issue:6572|issue:9387", + "issue:6639|issue:7142", + "issue:6506|issue:6956", + "issue:1730|issue:2599", + "issue:3441|issue:5678", + "issue:1363|issue:3008", + "issue:6385|issue:6956", + "issue:7355|issue:9476", + "issue:5607|issue:6222", + "issue:11343|issue:9167", + "issue:4111|issue:6431", + "issue:5989|issue:7549", + "issue:8970|issue:9488" + ], + "prepared_review_unit_hash": "f107cbdf30a90a935eef0105bde75e8e735200f2fb26abee5ca3f3c8f185052b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33337, + "estimated_input_tokens": 8335, + "estimated_eval_tokens": 16926 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous and does not look like one duplicate set. The only strong duplicate-like pair is the StableVideoDiffusionPipeline memory/regression reports; the rest are distinct issues across different models, schedulers, and pipelines.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 6385 is the best representative of the only clear duplicate-like theme here: a broad StableVideoDiffusionPipeline VRAM regression, with 6956 reading like an environment-specific OOM manifestation of the same problem.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6385 is the strongest global representative because it is the broadest, clearest statement of the shared failure mode in this group and can serve as the umbrella for the matching memory/OOM report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3280", + "right": "issue:7214", + "accept": false, + "reason": "Both mention schedulers, but one is a DeepFloyd IF non-default-scheduler problem and the other is a DPMSolverMultistepScheduler karras_sigmas regression; different bugs." + }, + { + "left": "issue:10239", + "right": "issue:10653", + "accept": false, + "reason": "Both are about HunyuanVideo offload/move-to-CPU behavior, but the concrete failures differ: unsupported sequential CPU offload vs bitsandbytes 8-bit `.to()` behavior." + }, + { + "left": "issue:2988", + "right": "issue:5685", + "accept": false, + "reason": "Both are scheduler-related, but one is a hard crash at equal train/inference steps and the other is a vague scheduler-issues report with no clear same code path." + }, + { + "left": "issue:7183", + "right": "issue:7217", + "accept": false, + "reason": "One is a scheduler config display bug; the other is DDIM config incompatibility. Not the same underlying defect." + }, + { + "left": "issue:3880", + "right": "issue:4720", + "accept": false, + "reason": "Both involve inpainting, but one is degraded unmasked regions and the other is AutoencoderTiny incompatibility with img2img/inpainting pipelines." + }, + { + "left": "issue:10865", + "right": "issue:7185", + "accept": false, + "reason": "SDXL model choice for inpainting is unrelated to pytorch-version support." + }, + { + "left": "issue:10954", + "right": "issue:9361", + "accept": false, + "reason": "Both involve FLUX LoRA loading, but the reports point to different symptoms and contexts; not enough to treat as the same concrete bug." + }, + { + "left": "issue:12147", + "right": "issue:13261", + "accept": false, + "reason": "Different models and different LoRA-loading failures; only the broad LoRA theme matches." + }, + { + "left": "issue:1633", + "right": "issue:5079", + "accept": false, + "reason": "karras sigmas support in a wrapper pipeline is unrelated to Lyrocis LoRA loading support." + }, + { + "left": "issue:3303", + "right": "issue:514", + "accept": false, + "reason": "Missing triton in Modelscope and attention slicing errors are unrelated." + }, + { + "left": "issue:5429", + "right": "issue:6513", + "accept": false, + "reason": "One is about models without a VAE in examples; the other is T2IAdapter support for a specific model. Different feature requests." + }, + { + "left": "issue:6572", + "right": "issue:9387", + "accept": false, + "reason": "Different pipelines entirely: ControlNetXS-SDXL-Inpaint vs MimicMotion." + }, + { + "left": "issue:6639", + "right": "issue:7142", + "accept": false, + "reason": "Both are SDXL output mismatch reports, but one is about load_lora_weights behavior and the other compares lightning inference against webui; not the same bug." + }, + { + "left": "issue:6506", + "right": "issue:6956", + "accept": false, + "reason": "One is an output-type API bug and the other is a CUDA OOM/memory issue; unrelated." + }, + { + "left": "issue:1730", + "right": "issue:2599", + "accept": false, + "reason": "Dreambooth-to-CKPT conversion and multi-subject training quality are different problems." + }, + { + "left": "issue:3441", + "right": "issue:5678", + "accept": false, + "reason": "Higher VRAM usage in PyTorch 2 and higher-order-derivative attention errors are unrelated." + }, + { + "left": "issue:1363", + "right": "issue:3008", + "accept": false, + "reason": "ONNX prompt-weight scheduler error and invalid torch.randn arguments in lpw_stable_diffusion are not the same bug." + }, + { + "left": "issue:6385", + "right": "issue:6956", + "accept": true, + "reason": "Both describe the same StableVideoDiffusionPipeline memory regression: unusually high VRAM use / CUDA OOM on the same pipeline." + }, + { + "left": "issue:7355", + "right": "issue:9476", + "accept": false, + "reason": "Stable Cascade bfloat16/resolution failure and Flux LoRA validation dtype mismatch are different code paths and symptoms." + }, + { + "left": "issue:5607", + "right": "issue:6222", + "accept": false, + "reason": "Tensor size mismatch in a model path is not the same as DDPM scheduler batch-size dependence." + }, + { + "left": "issue:11343", + "right": "issue:9167", + "accept": false, + "reason": "Import breakage after a transformers update is unrelated to a missing StableDiffusion3Pipeline import name." + }, + { + "left": "issue:4111", + "right": "issue:6431", + "accept": false, + "reason": "One is an SDXL Img2Img VRAM issue; the other is a broad pipeline-loading regression." + }, + { + "left": "issue:5989", + "right": "issue:7549", + "accept": false, + "reason": "A safetensors bug and a UNet EMA bug in training are unrelated." + }, + { + "left": "issue:8970", + "right": "issue:9488", + "accept": false, + "reason": "Different pipelines and different failure modes: Kolors MPS black output vs Lumina producing no image." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The summary correctly treats the cluster as mostly heterogeneous and only elevates the StableVideoDiffusionPipeline memory/OOM pair as a plausible duplicate-like match. The soft-edge verdicts are cautious and the rationales do not overclaim identity where only broad thematic overlap exists." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 306, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10039", + "issue:10227", + "issue:11569", + "issue:11687", + "issue:13067", + "issue:1335", + "issue:1405", + "issue:1445", + "issue:1572", + "issue:1831", + "issue:2089", + "issue:228", + "issue:2324", + "issue:2961", + "issue:3010", + "issue:3301", + "issue:3452", + "issue:3597", + "issue:3631", + "issue:409", + "issue:4269", + "issue:444", + "issue:4446", + "issue:4456", + "issue:4557", + "issue:4677", + "issue:4702", + "issue:4720", + "issue:5022", + "issue:5138", + "issue:5931", + "issue:6234", + "issue:6335", + "issue:6552", + "issue:6584", + "issue:6891", + "issue:6934", + "issue:7438", + "issue:7441", + "issue:7456", + "issue:7556", + "issue:8317", + "issue:8435", + "issue:8809", + "issue:8887", + "issue:9034", + "issue:9106", + "issue:9371" + ], + "soft_pairs": [ + "issue:1405|issue:409", + "issue:4446|issue:8435", + "issue:3597|issue:6335", + "issue:5931|issue:7438", + "issue:1572|issue:4677", + "issue:3631|issue:4720", + "issue:5022|issue:9034", + "issue:6584|issue:6934", + "issue:1831|issue:2089", + "issue:1445|issue:444", + "issue:10227|issue:8809", + "issue:10039|issue:6234", + "issue:11569|issue:9371", + "issue:11687|issue:13067", + "issue:1335|issue:4269", + "issue:228|issue:3452", + "issue:2324|issue:3301", + "issue:2961|issue:4702", + "issue:3010|issue:7556", + "issue:4456|issue:4557", + "issue:5138|issue:8887", + "issue:6552|issue:6891", + "issue:7441|issue:9106", + "issue:7456|issue:8317" + ], + "prepared_review_unit_hash": "90b466088543c365593a7b45a9bd99e09f2d3add44045fd6a370d928625b1764", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33880, + "estimated_input_tokens": 8470, + "estimated_eval_tokens": 17196 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No duplicate cluster emerges here: the soft pairs are mostly broad topical similarities, but they describe different bugs, features, or docs changes. All soft edges should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the items are heterogeneous and do not converge on one underlying bug or change.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good global representative; the set is too diverse and there is no true duplicate center.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1405", + "right": "issue:409", + "accept": false, + "reason": "Both mention attention slicing, but one is a missing/absent API symptom and the other is a broken call on main; not the same concrete bug." + }, + { + "left": "issue:4446", + "right": "issue:8435", + "accept": false, + "reason": "Both involve from_single_file runtime problems, but they report different failure modes and likely different code paths." + }, + { + "left": "issue:3597", + "right": "issue:6335", + "accept": false, + "reason": "One is an ONNX pipeline not working, the other is output-size control in img2img; too different to be duplicates." + }, + { + "left": "issue:5931", + "right": "issue:7438", + "accept": false, + "reason": "Determinism vs XPU loading/generation stalling are unrelated issues." + }, + { + "left": "issue:1572", + "right": "issue:4677", + "accept": false, + "reason": "Both touch inpainting/docs, but one is a mega pipeline functionality bug and the other is a doc script error." + }, + { + "left": "issue:3631", + "right": "issue:4720", + "accept": false, + "reason": "Both are inpainting-related, but one is latent handling and the other is AutoencoderTiny compatibility; different bugs." + }, + { + "left": "issue:5022", + "right": "issue:9034", + "accept": false, + "reason": "Import error for UNet2DModel and an optional dependency requirement problem are different issues." + }, + { + "left": "issue:6584", + "right": "issue:6934", + "accept": false, + "reason": "Both are conversion-related, but they target different model formats and directions." + }, + { + "left": "issue:1831", + "right": "issue:2089", + "accept": false, + "reason": "DreamBooth precision mismatch for class images and AltDiffusion OOM are unrelated." + }, + { + "left": "issue:1445", + "right": "issue:444", + "accept": false, + "reason": "Legacy ONNX inpaint attribute error and a num_inference_steps index error are different failures." + }, + { + "left": "issue:10227", + "right": "issue:8809", + "accept": false, + "reason": "LoRA feature-size support and feedforward output inconsistency are unrelated code-path problems." + }, + { + "left": "issue:10039", + "right": "issue:6234", + "accept": false, + "reason": "A missing output sample size argument in UNet blocks is unrelated to an example script issue." + }, + { + "left": "issue:11569", + "right": "issue:9371", + "accept": false, + "reason": "A feature request for a model repo and a Flax pipeline API refactor are not duplicates." + }, + { + "left": "issue:11687", + "right": "issue:13067", + "accept": false, + "reason": "Docs request for pruna and feature support for Anima are separate requests." + }, + { + "left": "issue:1335", + "right": "issue:4269", + "accept": false, + "reason": "Both are TypeErrors, but they affect different pipelines and different parameters." + }, + { + "left": "issue:228", + "right": "issue:3452", + "accept": false, + "reason": "Broken release link and DreamBooth low-VRAM training are unrelated." + }, + { + "left": "issue:2324", + "right": "issue:3301", + "accept": false, + "reason": "Both are DreamBooth training complaints, but the reported failure modes are different and not clearly the same bug." + }, + { + "left": "issue:2961", + "right": "issue:4702", + "accept": false, + "reason": "StableUnCLIP training example and VQ-VAE training example are different feature requests." + }, + { + "left": "issue:3010", + "right": "issue:7556", + "accept": false, + "reason": "Both are docs issues, but they concern different topics and changes." + }, + { + "left": "issue:4456", + "right": "issue:4557", + "accept": false, + "reason": "Different AttributeError symptoms in different loader/model code paths; not the same fix." + }, + { + "left": "issue:5138", + "right": "issue:8887", + "accept": false, + "reason": "Training utility refactor and wrong VAE encode in training script are unrelated." + }, + { + "left": "issue:6552", + "right": "issue:6891", + "accept": false, + "reason": "A training-resume FP16 tracker and a model-card utility tracker are distinct tasks." + }, + { + "left": "issue:7441", + "right": "issue:9106", + "accept": false, + "reason": "pytest on Python 3.12 and a PixArt-alpha time_step attribute error are unrelated." + }, + { + "left": "issue:7456", + "right": "issue:8317", + "accept": false, + "reason": "A missing raise statement and a missing generator argument in AsymmetricAutoencoderKL are different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 307, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10591", + "issue:10659", + "issue:10866", + "issue:10967", + "issue:11050", + "issue:12606", + "issue:13104", + "issue:1478", + "issue:1785", + "issue:1895", + "issue:222", + "issue:2359", + "issue:2375", + "issue:3366", + "issue:3540", + "issue:3748", + "issue:3750", + "issue:3933", + "issue:4213", + "issue:4304", + "issue:4400", + "issue:4419", + "issue:4450", + "issue:4605", + "issue:5254", + "issue:5314", + "issue:5433", + "issue:5594", + "issue:6068", + "issue:6087", + "issue:6094", + "issue:6157", + "issue:6549", + "issue:6686", + "issue:7061", + "issue:7228", + "issue:7420", + "issue:8410", + "issue:8493", + "issue:892", + "issue:8935", + "issue:9051", + "issue:9387", + "issue:9530", + "issue:9793", + "issue:984", + "issue:9944" + ], + "soft_pairs": [ + "issue:1478|issue:1785", + "issue:5433|issue:6068", + "issue:892|issue:984", + "issue:3366|issue:4450", + "issue:3750|issue:6549", + "issue:10659|issue:10866", + "issue:6157|issue:9793", + "issue:4605|issue:5254", + "issue:6087|issue:6686", + "issue:8493|issue:9944", + "issue:8935|issue:9051", + "issue:10591|issue:12606", + "issue:10967|issue:8410", + "issue:6686|issue:9530", + "issue:3540|issue:6094", + "issue:3933|issue:7420", + "issue:11050|issue:13104", + "issue:1895|issue:222", + "issue:2359|issue:5594", + "issue:2375|issue:4213", + "issue:3748|issue:7228", + "issue:4304|issue:4400", + "issue:4419|issue:7061", + "issue:5314|issue:9387" + ], + "prepared_review_unit_hash": "943175f7544cdf23ca323c1a9fd0bcb3d59a5292f7727346505fbd6d6f3b27c3", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33134, + "estimated_input_tokens": 8284, + "estimated_eval_tokens": 16824 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the items span unrelated pipelines, training scripts, import/logging problems, and feature requests. The soft edges are only title-level similarity, not the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11050 is the most concrete and actionable report in the set (specific file, exact dtype-cast bug, clear reproduction context), but it still does not represent the rest of the cluster because the items are about different subsystems and symptoms.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1478", + "right": "issue:1785", + "accept": false, + "reason": "Both are multi-image/output-count complaints, but they target different pipelines and failure modes: StableDiffusionUpscalePipeline output handling vs UnCLIP num_images_per_prompt behavior." + }, + { + "left": "issue:5433", + "right": "issue:6068", + "accept": false, + "reason": "SDXL scheduler artifacts are a sampling-quality issue; DDIM incorrect samples is a different algorithmic bug in a separate code path." + }, + { + "left": "issue:892", + "right": "issue:984", + "accept": false, + "reason": "These are unrelated runtime regressions: Docker/container compatibility vs F.interpolate breaking for large batch sizes." + }, + { + "left": "issue:3366", + "right": "issue:4450", + "accept": false, + "reason": "Autocast being blocked by LoRALinearLayer is a precision/context-manager bug, while multi-text inversion not working is a different training feature issue." + }, + { + "left": "issue:3750", + "right": "issue:6549", + "accept": false, + "reason": "One is lack of LoRA loading support in a pipeline; the other is a warning/logging mismatch when combining IP-Adapter and LoRA weights." + }, + { + "left": "issue:10659", + "right": "issue:10866", + "accept": false, + "reason": "Both mention poor LoRA results, but one is about ConsisID generation quality and the other about a Lumina Image 2.0 LoRA compatibility/loading problem." + }, + { + "left": "issue:6157", + "right": "issue:9793", + "accept": false, + "reason": "IP-Adapter not working in a specific pipeline is unrelated to a progress bar update bug." + }, + { + "left": "issue:4605", + "right": "issue:5254", + "accept": false, + "reason": "Long-prompt inpaint model errors and multi-ControlNet support in controlnet inpaint are distinct pipeline capabilities, not the same bug." + }, + { + "left": "issue:6087", + "right": "issue:6686", + "accept": false, + "reason": "Saving/loading LoraConfig metadata is a LoRA serialization bug; Dynamo save/load performance loss affects compiled SDXL pipelines and is a different issue." + }, + { + "left": "issue:8493", + "right": "issue:9944", + "accept": false, + "reason": "Both concern SD3 model loading, but one is a state_dict load error in the transformer and the other is SD3.5 Medium size-mismatch loading; different model versions and failure causes." + }, + { + "left": "issue:8935", + "right": "issue:9051", + "accept": false, + "reason": "These are separate batching/device issues in different pipelines: AuraFlow long-prompt tensor placement vs Flux batch inference support." + }, + { + "left": "issue:10591", + "right": "issue:12606", + "accept": false, + "reason": "SD3 LoRA training-script concerns are unrelated to a wrong import in a conversion script." + }, + { + "left": "issue:10967", + "right": "issue:8410", + "accept": false, + "reason": "A request to integrate LaVie/Cinemo into diffusers is a feature proposal, while TextDiffuser native support is a separate pipeline integration request." + }, + { + "left": "issue:6686", + "right": "issue:9530", + "accept": false, + "reason": "SDXL compile save/load performance regression is different from freeing GPU memory after torch.compile on the UNet." + }, + { + "left": "issue:3540", + "right": "issue:6094", + "accept": false, + "reason": "Prompt embeds not working in TextToVideoZeroPipeline is an input-handling bug, whereas the 0.24 update issue is a broader pipeline breakage after a version change." + }, + { + "left": "issue:3933", + "right": "issue:7420", + "accept": false, + "reason": "Graph compile breaking on logging.info is a specific logging/compile interaction; unable to use torch.compile is a broader, unspecific compile problem." + }, + { + "left": "issue:11050", + "right": "issue:13104", + "accept": false, + "reason": "The first is a bf16/fp16 dtype casting bug in a training example; the second is a missing logger definition error in a different code path." + }, + { + "left": "issue:1895", + "right": "issue:222", + "accept": false, + "reason": "EMA not updating in multi-GPU training and a generic training RuntimeError are different bugs with no shared code-path evidence." + }, + { + "left": "issue:2359", + "right": "issue:5594", + "accept": false, + "reason": "Latent output_type support in img2img is a pipeline behavior request; SDXL LoRA memory-saving is a training optimization topic." + }, + { + "left": "issue:2375", + "right": "issue:4213", + "accept": false, + "reason": "safe_open NameError and from_single_file() requiring accelerate are unrelated loading/environment issues." + }, + { + "left": "issue:3748", + "right": "issue:7228", + "accept": false, + "reason": "LoRA safetensor float16/float32 mismatch is a LoRA load issue; missing dtype on traced StableDiffusionPipeline is a different traced-model attribute bug." + }, + { + "left": "issue:4304", + "right": "issue:4400", + "accept": false, + "reason": "SDXL text-to-image LoRA support request and text_encoder support for inpaint LoRA are different feature scopes." + }, + { + "left": "issue:4419", + "right": "issue:7061", + "accept": false, + "reason": "Offset noise being omitted in inference is a model-usage bug; incorrect docstring text for a community pipeline is documentation-only." + }, + { + "left": "issue:5314", + "right": "issue:9387", + "accept": false, + "reason": "ControlNet reference inpainting and MimicMotion are unrelated pipeline feature requests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 308, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10659", + "issue:1072", + "issue:10797", + "issue:1082", + "issue:11005", + "issue:11133", + "issue:11229", + "issue:11410", + "issue:11486", + "issue:1158", + "issue:11827", + "issue:12329", + "issue:12653", + "issue:13043", + "issue:13073", + "issue:13386", + "issue:1500", + "issue:1574", + "issue:1734", + "issue:1961", + "issue:1990", + "issue:2444", + "issue:2578", + "issue:281", + "issue:2883", + "issue:3774", + "issue:4210", + "issue:4250", + "issue:4353", + "issue:4354", + "issue:4630", + "issue:4708", + "issue:4907", + "issue:5110", + "issue:5604", + "issue:5835", + "issue:6354", + "issue:6598", + "issue:6958", + "issue:7418", + "issue:7762", + "issue:7784", + "issue:7889", + "issue:9106", + "issue:9186", + "issue:9471", + "issue:9756", + "issue:9856" + ], + "soft_pairs": [ + "issue:6958|issue:9756", + "issue:11133|issue:11827", + "issue:11410|issue:9471", + "issue:6598|issue:7762", + "issue:1072|issue:4250", + "issue:281|issue:4630", + "issue:6354|issue:9186", + "issue:2578|issue:5604", + "issue:12329|issue:13073", + "issue:11229|issue:12653", + "issue:11005|issue:7418", + "issue:1158|issue:1500", + "issue:1574|issue:3774", + "issue:1990|issue:2883", + "issue:4354|issue:5835", + "issue:4708|issue:5110", + "issue:13043|issue:13386", + "issue:1082|issue:1961", + "issue:10659|issue:10797", + "issue:4210|issue:7889", + "issue:4353|issue:4907", + "issue:11486|issue:9856", + "issue:1734|issue:2444", + "issue:7784|issue:9106" + ], + "prepared_review_unit_hash": "9dce3c08e67957a8604acbfeb2bce208757cd10687d0f73ca16508fc8624e379", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34401, + "estimated_input_tokens": 8601, + "estimated_eval_tokens": 17458 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose, heterogeneous set of issues spanning unrelated bugs, feature requests, and docs/test complaints. There is no strong duplicate cluster here; issue:281 is only a weak canonical because it is the broadest and highest-signal item.", + "confidence": 0.22, + "canonical_issue_reason": "issue:281 is the most central/high-signal issue in the set: broad model-loading optimization request, highest discussion/inbound refs, and widest potential relevance. It is still only a weak fit because the cluster is not truly cohesive.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:281 is the best representative in this mixed set for the same reasons: broad scope, strong engagement, and cross-cutting model-loading context. It is the least narrow among the issues, though not a true duplicate anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6958", + "right": "issue:9756", + "accept": false, + "reason": "Different subjects: an unet_motion_model config bug vs a missing notebook/file-path issue." + }, + { + "left": "issue:11133", + "right": "issue:11827", + "accept": false, + "reason": "Different pipelines and failure modes: CogVideoX image-to-video bug vs safetensors header corruption." + }, + { + "left": "issue:11410", + "right": "issue:9471", + "accept": false, + "reason": "Attention processor unpacking error and multiple-extension resource detection are unrelated code paths." + }, + { + "left": "issue:6598", + "right": "issue:7762", + "accept": false, + "reason": "Both mention poor image results, but one is parity with A1111 while the other is a general quality complaint; not the same concrete bug." + }, + { + "left": "issue:1072", + "right": "issue:4250", + "accept": false, + "reason": "General advice on loading multiple fine-tuned models is not the same as an incorrect from_pretrained loading bug." + }, + { + "left": "issue:281", + "right": "issue:4630", + "accept": false, + "reason": "Model-loading optimization request and post-load LoRA weight editing are different features." + }, + { + "left": "issue:6354", + "right": "issue:9186", + "accept": false, + "reason": "ControlNet LoRA support request vs differential diffusion with ControlNet are distinct feature ideas." + }, + { + "left": "issue:2578", + "right": "issue:5604", + "accept": false, + "reason": "Editable-install failure and unfusing multiple LoRA are unrelated problems." + }, + { + "left": "issue:12329", + "right": "issue:13073", + "accept": false, + "reason": "Wan single-file transformer version selection and ZImageControlNet guidance_scale behavior are separate issues." + }, + { + "left": "issue:11229", + "right": "issue:12653", + "accept": false, + "reason": "Both involve MPS, but one is attention slicing NaNs and the other is a rope float64 reference break; different code paths." + }, + { + "left": "issue:11005", + "right": "issue:7418", + "accept": false, + "reason": "Docstring/default mismatch is not the same as a bug in a community upscale pipeline file." + }, + { + "left": "issue:1158", + "right": "issue:1500", + "accept": false, + "reason": "Test interference and community-pipeline compatibility with a new diffusers release are unrelated." + }, + { + "left": "issue:1574", + "right": "issue:3774", + "accept": false, + "reason": "Both are training/memory-related, but they describe different scripts and failure mechanisms." + }, + { + "left": "issue:1990", + "right": "issue:2883", + "accept": false, + "reason": "A pipeline proposal and an image-to-video pipeline request are different feature discussions." + }, + { + "left": "issue:4354", + "right": "issue:5835", + "accept": false, + "reason": "from_single_file returning the wrong pipeline type is unrelated to using a pretrained MusicLDM pipeline." + }, + { + "left": "issue:4708", + "right": "issue:5110", + "accept": false, + "reason": "Import failure in diffusers.utils and HF repo-id validation are different errors." + }, + { + "left": "issue:13043", + "right": "issue:13386", + "accept": false, + "reason": "RTX 5090 kernel-image support and qwen prior-preservation concat crashes are unrelated." + }, + { + "left": "issue:1082", + "right": "issue:1961", + "accept": false, + "reason": "Dreambooth cublas matmul failure and a generic colossalai error are not the same bug." + }, + { + "left": "issue:10659", + "right": "issue:10797", + "accept": false, + "reason": "ConsisID output-quality complaint and HunyuanVideo group_offloading failure are different pipeline issues." + }, + { + "left": "issue:4210", + "right": "issue:7889", + "accept": false, + "reason": "A UNet attention_mask API question and single_file_utils bugs are not the same underlying problem." + }, + { + "left": "issue:4353", + "right": "issue:4907", + "accept": false, + "reason": "General SDXL quality complaint and an inpainting-model warning are distinct." + }, + { + "left": "issue:11486", + "right": "issue:9856", + "accept": false, + "reason": "Training-script xformers feature request and distributed port collision are unrelated." + }, + { + "left": "issue:1734", + "right": "issue:2444", + "accept": false, + "reason": "Multi-GPU slowdown in Dreambooth and a train_dreambooth_inpaint subprocess failure are different issues." + }, + { + "left": "issue:7784", + "right": "issue:9106", + "accept": false, + "reason": "Missing upcast_vae on one pipeline and a PixArt-alpha attribute error on another are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 309, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10039", + "issue:10147", + "issue:10697", + "issue:10722", + "issue:10861", + "issue:11005", + "issue:11232", + "issue:1572", + "issue:1614", + "issue:1969", + "issue:1988", + "issue:2363", + "issue:2471", + "issue:2603", + "issue:2618", + "issue:3645", + "issue:3986", + "issue:4141", + "issue:4269", + "issue:4310", + "issue:434", + "issue:4445", + "issue:4648", + "issue:4675", + "issue:4782", + "issue:4876", + "issue:5050", + "issue:5168", + "issue:5300", + "issue:5538", + "issue:5763", + "issue:5972", + "issue:6113", + "issue:6363", + "issue:6383", + "issue:6407", + "issue:6439", + "issue:6530", + "issue:6866", + "issue:7264", + "issue:7686", + "issue:7773", + "issue:7900", + "issue:835", + "issue:8873", + "issue:9062", + "issue:9674", + "issue:989" + ], + "soft_pairs": [ + "issue:7900|issue:8873", + "issue:6363|issue:7686", + "issue:1614|issue:835", + "issue:5972|issue:9674", + "issue:1572|issue:4782", + "issue:10039|issue:7773", + "issue:10147|issue:6407", + "issue:3645|issue:4675", + "issue:4648|issue:5538", + "issue:5300|issue:6383", + "issue:10722|issue:11232", + "issue:4141|issue:5763", + "issue:1969|issue:989", + "issue:2471|issue:434", + "issue:2618|issue:4269", + "issue:1988|issue:2603", + "issue:10697|issue:11005", + "issue:10861|issue:9062", + "issue:2363|issue:6113", + "issue:3986|issue:6866", + "issue:4310|issue:6439", + "issue:4445|issue:7264", + "issue:4876|issue:5050", + "issue:5168|issue:6530" + ], + "prepared_review_unit_hash": "f53dfcb77fc3eee3c35884d6e472e6b744b451bf81dc0dd50a4ad11b4eb61299", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33936, + "estimated_input_tokens": 8484, + "estimated_eval_tokens": 17224 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous and the soft links are only topical. None of the proposed pairs look like the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": "issue:10039 is the most concrete, code-localized bug report in the set (specific function/file and missing argument), so it is the best representative issue for the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10039 is the strongest issue candidate because it is narrowly scoped, actionable, and clearly tied to a specific code path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7900", + "right": "issue:8873", + "accept": false, + "reason": "One is an __init__.py import/export problem; the other is an xformers performance complaint. Different symptoms and code paths." + }, + { + "left": "issue:6363", + "right": "issue:7686", + "accept": false, + "reason": "FP16 gradient scaling in training is unrelated to a UNet cross_attention_dim configuration error." + }, + { + "left": "issue:1614", + "right": "issue:835", + "accept": false, + "reason": "Black-image output with autocast is a different failure mode from flax numpy_to_pil conversion issues." + }, + { + "left": "issue:5972", + "right": "issue:9674", + "accept": false, + "reason": "Documentation around min-SNR weighting is not the same as a bad SD3 ControlNet training result." + }, + { + "left": "issue:1572", + "right": "issue:4782", + "accept": false, + "reason": "The first is a pipeline support bug for img2img/inpainting; the second is a masking correctness bug in inpainting." + }, + { + "left": "issue:10039", + "right": "issue:7773", + "accept": false, + "reason": "Missing sample size in AttnUpBlock2D is unrelated to a Stable Video Diffusion batch-size generation bug." + }, + { + "left": "issue:10147", + "right": "issue:6407", + "accept": false, + "reason": "Multi-view generation adapter support and AnyText multilingual text generation are different features." + }, + { + "left": "issue:3645", + "right": "issue:4675", + "accept": false, + "reason": "Training/finetuning guidance for an inpaint pipeline is not the same as SDXL reference-only ControlNet support." + }, + { + "left": "issue:4648", + "right": "issue:5538", + "accept": false, + "reason": "Missing config.json is a loading/setup error; encoder-decoder latent usage is a model-pipeline implementation issue." + }, + { + "left": "issue:5300", + "right": "issue:6383", + "accept": false, + "reason": "Network connectivity behavior and multi-stage learning rate scheduling are unrelated." + }, + { + "left": "issue:10722", + "right": "issue:11232", + "accept": false, + "reason": "Tensor-size mismatch in DreamBooth prior preservation is not the same as noise residual behavior during T2I fine-tuning." + }, + { + "left": "issue:4141", + "right": "issue:5763", + "accept": false, + "reason": "Second-ControlNet handling and safetensors serialization are separate issues in different subsystems." + }, + { + "left": "issue:1969", + "right": "issue:989", + "accept": false, + "reason": "A tensor view/stride runtime error is a different problem from a 2-D weight shape error." + }, + { + "left": "issue:2471", + "right": "issue:434", + "accept": false, + "reason": "Attend-and-excite non-default size support is unrelated to a training notebook compatibility regression." + }, + { + "left": "issue:2618", + "right": "issue:4269", + "accept": false, + "reason": "ControlNet image input type handling and check_inputs default-argument TypeError are not the same bug." + }, + { + "left": "issue:1988", + "right": "issue:2603", + "accept": false, + "reason": "fp16/bf16 input-bias dtype mismatch is unrelated to non-square image generation support." + }, + { + "left": "issue:10697", + "right": "issue:11005", + "accept": false, + "reason": "Random transform mismatch in instruct-pix2pix is a data-preprocessing bug, not a docstring/default discrepancy." + }, + { + "left": "issue:10861", + "right": "issue:9062", + "accept": false, + "reason": "Latent scaling in StableAudioPipeline and an import error for StableAudioPipeline are entirely different issues." + }, + { + "left": "issue:2363", + "right": "issue:6113", + "accept": false, + "reason": "LoRA conversion tooling and using sd_control_collection models in diffusers do not describe the same change." + }, + { + "left": "issue:3986", + "right": "issue:6866", + "accept": false, + "reason": "ShapE mesh output support and IP-Adapter support for SDXL ControlNet inpaint are unrelated feature requests." + }, + { + "left": "issue:4310", + "right": "issue:6439", + "accept": false, + "reason": "SDXL staircase artifacts and a broken docs link are not the same underlying problem." + }, + { + "left": "issue:4445", + "right": "issue:7264", + "accept": false, + "reason": "LoRA finetuning help and custom transformers model serialization/loading are unrelated." + }, + { + "left": "issue:4876", + "right": "issue:5050", + "accept": false, + "reason": "LoRA layer std as a parameter and LoRA config changes are different feature requests with different scopes." + }, + { + "left": "issue:5168", + "right": "issue:6530", + "accept": false, + "reason": "ControlNet-XS support and SDXL K-diffusion with ControlNet are distinct feature efforts." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 310, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10191", + "issue:10412", + "issue:10658", + "issue:10673", + "issue:11221", + "issue:12574", + "issue:13077", + "issue:13300", + "issue:1677", + "issue:1785", + "issue:1925", + "issue:2554", + "issue:2763", + "issue:347", + "issue:3954", + "issue:4030", + "issue:4266", + "issue:4403", + "issue:4534", + "issue:4830", + "issue:5334", + "issue:5546", + "issue:5617", + "issue:5835", + "issue:5924", + "issue:5989", + "issue:6141", + "issue:6689", + "issue:6725", + "issue:7009", + "issue:7431", + "issue:7440", + "issue:7519", + "issue:7695", + "issue:7841", + "issue:7897", + "issue:7962", + "issue:8488", + "issue:8502", + "issue:851", + "issue:8543", + "issue:8826", + "issue:8873", + "issue:8876", + "issue:9488", + "issue:9516", + "issue:9780", + "issue:9926" + ], + "soft_pairs": [ + "issue:6141|issue:9926", + "issue:7009|issue:7440", + "issue:8502|issue:8543", + "issue:1925|issue:851", + "issue:4830|issue:5546", + "issue:10191|issue:7897", + "issue:2554|issue:347", + "issue:4403|issue:5617", + "issue:5334|issue:9516", + "issue:5835|issue:9488", + "issue:5989|issue:6689", + "issue:10658|issue:7962", + "issue:10673|issue:8876", + "issue:13300|issue:5924", + "issue:6725|issue:7009", + "issue:7519|issue:8873", + "issue:11221|issue:8826", + "issue:1785|issue:4030", + "issue:10412|issue:7431", + "issue:7695|issue:7841", + "issue:12574|issue:13077", + "issue:1677|issue:3954", + "issue:2763|issue:5617", + "issue:4266|issue:4534", + "issue:8488|issue:9780" + ], + "prepared_review_unit_hash": "80ae461c3a1e5c79f6cfe7f4d84ff79d57a8e97b951949b1522ff150d773b95a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34401, + "estimated_input_tokens": 8601, + "estimated_eval_tokens": 17458 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a title-similarity bucket spanning many unrelated bugs (training, pipelines, loading, memory, CPU/MPS, xformers, etc.). I found no soft pair that clearly describes the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.66, + "canonical_issue_reason": "issue:7897 is the broadest representative of the cluster\u2019s few multi-GPU/training-related reports, and it is an actual open bug report rather than a narrow symptom question.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:7897 is a reasonable anchor because it is open, concrete, and broad enough to cover nearby device-placement/multi-GPU training failures better than the more specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6141", + "right": "issue:9926", + "accept": false, + "reason": "One is ControlNet API support, the other is GGUF quantization/speed support; different features and code paths." + }, + { + "left": "issue:7009", + "right": "issue:7440", + "accept": false, + "reason": "Text-to-video quality/runtime issue vs dataset availability problem; not the same bug." + }, + { + "left": "issue:8502", + "right": "issue:8543", + "accept": false, + "reason": "Both involve SD3, but one is image generation behavior and the other is dreambooth gradient checkpointing; different failures." + }, + { + "left": "issue:1925", + "right": "issue:851", + "accept": false, + "reason": "Both mention dtype/half precision, but one is a specific inpainting model issue and the other is a generic dtype mismatch report." + }, + { + "left": "issue:4830", + "right": "issue:5546", + "accept": false, + "reason": "Wrong tensor size in MultiControlNet vs scheduler step-count bug in base+refiner ensemble; unrelated problems." + }, + { + "left": "issue:10191", + "right": "issue:7897", + "accept": false, + "reason": "Both touch multi-GPU training, but one is GPU device selection being ignored while the other is a generic multi-GPU failure; not the same concrete code-path." + }, + { + "left": "issue:2554", + "right": "issue:347", + "accept": false, + "reason": "Different dependency/install issues: accelerate version requirement vs huggingface-hub installation failure." + }, + { + "left": "issue:4403", + "right": "issue:5617", + "accept": false, + "reason": "Half-precision LayerNorm runtime error vs torch.compile save/load problem; unrelated." + }, + { + "left": "issue:5334", + "right": "issue:9516", + "accept": false, + "reason": "Training/accelerate multi-model handling vs FLUX joint_attention_kwargs forwarding; different fixes." + }, + { + "left": "issue:5835", + "right": "issue:9488", + "accept": false, + "reason": "MusicLDM usage question vs Lumina image-generation failure; unrelated pipelines." + }, + { + "left": "issue:5989", + "right": "issue:6689", + "accept": false, + "reason": "Generic safetensors bug vs model conversion request; too different to be duplicates." + }, + { + "left": "issue:10658", + "right": "issue:7962", + "accept": false, + "reason": "onnxruntime session provider options vs Instruct-pix2pix cross_attention_kwargs support; unrelated APIs." + }, + { + "left": "issue:10673", + "right": "issue:8876", + "accept": false, + "reason": "AuraFlow shape reshape failure vs PixArt Sigma resolution microconditioning omission; different model-specific bugs." + }, + { + "left": "issue:13300", + "right": "issue:5924", + "accept": false, + "reason": "CPU inference failure for SD3.5/FLUX vs VAE memory spike; different symptoms and code paths." + }, + { + "left": "issue:6725", + "right": "issue:7009", + "accept": false, + "reason": "Generic LDM image generation request vs text-to-video result quality issue; not the same defect." + }, + { + "left": "issue:7519", + "right": "issue:8873", + "accept": false, + "reason": "Ignored cross_attention_kwargs in xFormers attention vs xFormers performance regression; distinct issues." + }, + { + "left": "issue:11221", + "right": "issue:8826", + "accept": false, + "reason": "AutoencoderKL non-convergence vs empty tensor list cat runtime error; unrelated." + }, + { + "left": "issue:1785", + "right": "issue:4030", + "accept": false, + "reason": "UnCLIP num_images_per_prompt bug vs Stable Diffusion v2.1 embedding runtime error; different code paths." + }, + { + "left": "issue:10412", + "right": "issue:7431", + "accept": false, + "reason": "DreamBooth VRAM usage vs DoRA/text-encoder training issues; both training-related but not the same bug." + }, + { + "left": "issue:7695", + "right": "issue:7841", + "accept": false, + "reason": "Ignored ip_adapter_masks kwargs vs red-dot rendering artifact; unrelated." + }, + { + "left": "issue:12574", + "right": "issue:13077", + "accept": false, + "reason": "WanVideoToVideo control-argument handling vs shallow-copy transformer weights in ZImageControlNet; different bugs." + }, + { + "left": "issue:1677", + "right": "issue:3954", + "accept": false, + "reason": "Wrong input data type for pipeline call vs multi-GPU scheduler step count bug; unrelated." + }, + { + "left": "issue:2763", + "right": "issue:5617", + "accept": false, + "reason": "ControlNet VRAM regression with cudnn.benchmark vs torch.compile save/load failure; not the same issue." + }, + { + "left": "issue:4266", + "right": "issue:4534", + "accept": false, + "reason": "Both are SDXL on Apple/MPS, but one is a broadcast type mismatch and the other crashes during validation; too different." + }, + { + "left": "issue:8488", + "right": "issue:9780", + "accept": false, + "reason": "General StableDiffusion3Pipeline from_pretrained exception vs missing norm_added_k attribute when loading SD3.5; likely related surface area, but not clearly the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 311, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10003", + "issue:10337", + "issue:1056", + "issue:1064", + "issue:10839", + "issue:1101", + "issue:11047", + "issue:11127", + "issue:11422", + "issue:11489", + "issue:11519", + "issue:12022", + "issue:1251", + "issue:1802", + "issue:2262", + "issue:228", + "issue:272", + "issue:2737", + "issue:2780", + "issue:2941", + "issue:3301", + "issue:3537", + "issue:3658", + "issue:3672", + "issue:3689", + "issue:3728", + "issue:389", + "issue:4086", + "issue:4324", + "issue:487", + "issue:4964", + "issue:5043", + "issue:5075", + "issue:5143", + "issue:5281", + "issue:557", + "issue:6161", + "issue:6288", + "issue:6357", + "issue:6486", + "issue:7003", + "issue:7440", + "issue:7889", + "issue:7948", + "issue:7955", + "issue:9108", + "issue:9171" + ], + "soft_pairs": [ + "issue:2941|issue:557", + "issue:4964|issue:6357", + "issue:10337|issue:9108", + "issue:1056|issue:1802", + "issue:5075|issue:5281", + "issue:12022|issue:9171", + "issue:4086|issue:7003", + "issue:10003|issue:7889", + "issue:1064|issue:1251", + "issue:10839|issue:11127", + "issue:1101|issue:5043", + "issue:11047|issue:7440", + "issue:11422|issue:11489", + "issue:11519|issue:7948", + "issue:2262|issue:3689", + "issue:228|issue:389", + "issue:272|issue:2737", + "issue:2780|issue:6288", + "issue:3301|issue:4324", + "issue:3537|issue:6161", + "issue:3658|issue:5143", + "issue:3658|issue:6486", + "issue:3672|issue:487", + "issue:3728|issue:7955" + ], + "prepared_review_unit_hash": "dfd08b62d7708cce06bb13d2d4386bc9d37dd542bd29a3c591a00ebf8828c28e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33375, + "estimated_input_tokens": 8344, + "estimated_eval_tokens": 16944 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft edges look like true duplicates; the pairs mostly share a subsystem or buzzword (Flux, LoRA, quantization, schedulers) but describe different bugs or feature requests. No PRs are present, and there isn\u2019t a strong duplicate cluster to collapse into a single canonical item.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 9108 is the most substantive open bug in the set: it is clearly framed, actively discussed, and describes a concrete training failure with quantized base models.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9108 is the best representative issue here because it is open, well-scoped, and has the strongest discussion signal among the listed issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2941", + "right": "issue:557", + "accept": false, + "reason": "Both concern scheduler quality, but they are different schedulers with different symptoms and are not the same bug." + }, + { + "left": "issue:4964", + "right": "issue:6357", + "accept": false, + "reason": "Model conversion to Diffusers is unrelated to API-vs-local result differences." + }, + { + "left": "issue:10337", + "right": "issue:9108", + "accept": false, + "reason": "Both involve quantization, but one is FLUX IPAdapter inference failure and the other is training resume with a quantized base model." + }, + { + "left": "issue:1056", + "right": "issue:1802", + "accept": false, + "reason": "Same file, but distinct problems: MPS support versus encoder_hidden_states=None handling." + }, + { + "left": "issue:5075", + "right": "issue:5281", + "accept": false, + "reason": "Different failures: SDXL refiner prompt-embedding shape mismatch versus multi-pipeline CPU offload device mismatch." + }, + { + "left": "issue:12022", + "right": "issue:9171", + "accept": false, + "reason": "One is a flash-attn interface compatibility issue; the other is a from_single_file v_prediction regression." + }, + { + "left": "issue:4086", + "right": "issue:7003", + "accept": false, + "reason": "Training ControlNet on SDXL base-0.9 is not the same underlying bug as an inpaint pipeline weight-shape mismatch." + }, + { + "left": "issue:10003", + "right": "issue:7889", + "accept": false, + "reason": "DDIM/DDPM scheduler bug with 1000 steps is unrelated to single-file checkpoint loading bugs." + }, + { + "left": "issue:1064", + "right": "issue:1251", + "accept": false, + "reason": "LMSDiscreteScheduler.add_noise() error is distinct from RePaintPipeline producing black images on MPS." + }, + { + "left": "issue:10839", + "right": "issue:11127", + "accept": false, + "reason": "Both mention Flux, but one is a Dreambooth LoRA training last-step error and the other is a single-file model loading problem." + }, + { + "left": "issue:1101", + "right": "issue:5043", + "accept": false, + "reason": "These are separate feature requests for different capabilities and do not share the same underlying change." + }, + { + "left": "issue:11047", + "right": "issue:7440", + "accept": false, + "reason": "ControlNet inpaint model support is unrelated to dataset availability for text-to-image." + }, + { + "left": "issue:11422", + "right": "issue:11489", + "accept": false, + "reason": "Both are training crashes, but they target different models and different failure modes." + }, + { + "left": "issue:11519", + "right": "issue:7948", + "accept": false, + "reason": "Different support requests for different models/adapters." + }, + { + "left": "issue:2262", + "right": "issue:3689", + "accept": false, + "reason": "How to use safetensors LoRA is not the same as unloading LoRA weights after loading." + }, + { + "left": "issue:228", + "right": "issue:389", + "accept": false, + "reason": "A broken release link and broken README code are separate documentation issues." + }, + { + "left": "issue:272", + "right": "issue:2737", + "accept": false, + "reason": "Both are scheduler-related feature requests, but they ask for different scheduler capabilities in different contexts." + }, + { + "left": "issue:2780", + "right": "issue:6288", + "accept": false, + "reason": "FP16 unscale runtime error is unrelated to the LCM LoRA default learning-rate discussion." + }, + { + "left": "issue:3301", + "right": "issue:4324", + "accept": false, + "reason": "The generic training failure could hide many causes, but this is not enough to equate it with the specific NoneType error." + }, + { + "left": "issue:3537", + "right": "issue:6161", + "accept": false, + "reason": "A conversion script request is unrelated to always discarding the next-to-last sigma." + }, + { + "left": "issue:3658", + "right": "issue:5143", + "accept": false, + "reason": "Extra noise latents in training scripts is a different bug from tiny numerical differences producing output drift." + }, + { + "left": "issue:3658", + "right": "issue:6486", + "accept": false, + "reason": "These are both Dreambooth-related, but one is an input-noise bug and the other is double embedding retraction." + }, + { + "left": "issue:3672", + "right": "issue:487", + "accept": false, + "reason": "Concurrent inference failures are not the same as M1 CPU incompatibility." + }, + { + "left": "issue:3728", + "right": "issue:7955", + "accept": false, + "reason": "A new pipeline request and a portrait-generation model request are separate feature ideas." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 312, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11286", + "issue:1185", + "issue:11865", + "issue:1249", + "issue:1251", + "issue:13107", + "issue:1463", + "issue:1519", + "issue:1556", + "issue:1655", + "issue:1851", + "issue:3040", + "issue:3494", + "issue:3954", + "issue:3970", + "issue:4043", + "issue:4080", + "issue:4325", + "issue:4460", + "issue:4733", + "issue:490", + "issue:4912", + "issue:4921", + "issue:5014", + "issue:5064", + "issue:5245", + "issue:5538", + "issue:5546", + "issue:5573", + "issue:5685", + "issue:5960", + "issue:6418", + "issue:6529", + "issue:6783", + "issue:6901", + "issue:6955", + "issue:760", + "issue:7682", + "issue:7922", + "issue:8560", + "issue:8813", + "issue:8915", + "issue:9032", + "issue:9038", + "issue:9359", + "issue:9844", + "issue:9846", + "issue:9933" + ], + "soft_pairs": [ + "issue:3954|issue:5685", + "issue:3970|issue:4325", + "issue:4043|issue:6783", + "issue:4080|issue:8813", + "issue:4912|issue:6529", + "issue:4921|issue:5546", + "issue:5014|issue:8560", + "issue:5245|issue:8915", + "issue:5538|issue:6418", + "issue:6901|issue:9933", + "issue:6955|issue:7922", + "issue:9032|issue:9038", + "issue:9359|issue:9844", + "issue:5573|issue:5960", + "issue:1249|issue:3040", + "issue:1519|issue:760", + "issue:1463|issue:4733", + "issue:1655|issue:490", + "issue:5064|issue:7682", + "issue:1185|issue:1851", + "issue:11865|issue:9846", + "issue:1251|issue:1556", + "issue:3494|issue:4460", + "issue:11286|issue:13107" + ], + "prepared_review_unit_hash": "24f63d133002afb54e4dc91460aaa053e639da159dca6ed4721226635f1e592b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34070, + "estimated_input_tokens": 8518, + "estimated_eval_tokens": 17292 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a noisy mix of unrelated diffusers issues. None of the proposed soft pairs look like the same concrete bug/change, so I would not merge any of them as duplicates.", + "confidence": 0.9, + "canonical_issue_reason": "If a representative issue must be chosen, issue:5685 is the broadest and most central because it is a generic scheduler-bug report; however, the cluster is too heterogeneous to treat it as a true canonical duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:5685 is the best broad representative of the limited scheduler-related subtheme in this set, but it does not actually unify the cluster into one underlying defect.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3954", + "right": "issue:5685", + "accept": false, + "reason": "Both mention schedulers, but one is a multi-gpu LR step-count bug and the other is a vague scheduler-internals report; not the same defect." + }, + { + "left": "issue:3970", + "right": "issue:4325", + "accept": false, + "reason": "Prompt-embeddings failure and black-image/runtime-warning behavior are different code paths and symptoms." + }, + { + "left": "issue:4043", + "right": "issue:6783", + "accept": false, + "reason": "SDXL embedding-argument requirements are unrelated to PixArt generations being corrupted." + }, + { + "left": "issue:4080", + "right": "issue:8813", + "accept": false, + "reason": "A new SDXL inpainting pipeline request is not the same as a Differential Diffusion issue for Kolors." + }, + { + "left": "issue:4912", + "right": "issue:6529", + "accept": false, + "reason": "Checkpoint conversion support and SDXL Fooocus inpaint are different feature requests with no shared bug." + }, + { + "left": "issue:4921", + "right": "issue:5546", + "accept": false, + "reason": "Refiner detail loss and wrong step counts with base+refiner ensembles are both refiner-related, but not the same underlying bug." + }, + { + "left": "issue:5014", + "right": "issue:8560", + "accept": false, + "reason": "Dreambooth LoRA quality complaints and SD3 training OOM are separate problems." + }, + { + "left": "issue:5245", + "right": "issue:8915", + "accept": false, + "reason": "Import freeze in pipenv and a missing import symbol are different failure modes." + }, + { + "left": "issue:5538", + "right": "issue:6418", + "accept": false, + "reason": "Upscale latent-usage behavior and stable-video finetuning output collapse are unrelated pipeline bugs." + }, + { + "left": "issue:6901", + "right": "issue:9933", + "accept": false, + "reason": "A missing Flax scheduler method and missing img2img width/height params are different API gaps." + }, + { + "left": "issue:6955", + "right": "issue:7922", + "accept": false, + "reason": "A Transformers AddedToken attribute error and a missing ORT pipeline attribute are distinct import/runtime issues." + }, + { + "left": "issue:9032", + "right": "issue:9038", + "accept": false, + "reason": "Both are FlaxStableDiffusionPipeline questions, but one is a TPU setup request and the other is about prompt weighting." + }, + { + "left": "issue:9359", + "right": "issue:9844", + "accept": false, + "reason": "A scheduler step ValueError and SDXL VAE NaNs are different stages of the pipeline." + }, + { + "left": "issue:5573", + "right": "issue:5960", + "accept": false, + "reason": "Slow ControlNet weight init and invalid-value casting in SDXL ControlNet training are not the same bug." + }, + { + "left": "issue:1249", + "right": "issue:3040", + "accept": false, + "reason": "OnnxPipeline ndarray/device handling and a docs copy-paste '__module__' error are unrelated." + }, + { + "left": "issue:1519", + "right": "issue:760", + "accept": false, + "reason": "Both involve MPS, but one is batch-input noise on M2 and the other is unexpected output on older diffusers; too different to merge." + }, + { + "left": "issue:1463", + "right": "issue:4733", + "accept": false, + "reason": "Missing the datasets package and a dreambooth checkpoint-resume optimizer assertion are unrelated." + }, + { + "left": "issue:1655", + "right": "issue:490", + "accept": false, + "reason": "Pipeline initialization failure and general memory issues are not the same defect." + }, + { + "left": "issue:5064", + "right": "issue:7682", + "accept": false, + "reason": "Refiner out-of-space and multi-GPU scaling issues are separate resource/performance problems." + }, + { + "left": "issue:1185", + "right": "issue:1851", + "accept": false, + "reason": "Dreambooth reproducibility and multi-GPU slowdown are different training concerns." + }, + { + "left": "issue:11865", + "right": "issue:9846", + "accept": false, + "reason": "Missing from_single_file on two different model classes is a similar API gap, but not one concrete bug." + }, + { + "left": "issue:1251", + "right": "issue:1556", + "accept": false, + "reason": "Both mention black images, but they affect different pipelines/hardware paths and do not look like one bug." + }, + { + "left": "issue:3494", + "right": "issue:4460", + "accept": false, + "reason": "Corrupted Multidiffusion outputs with schedulers and generic stable-diffusion/diffusers result differences are too broad to be duplicates." + }, + { + "left": "issue:11286", + "right": "issue:13107", + "accept": false, + "reason": "Both are LoRA-loading errors, but one is generic and the other is a specific Flux Klein missing-key report; not clearly the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 313, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1013", + "issue:10198", + "issue:10591", + "issue:10675", + "issue:1083", + "issue:10962", + "issue:11644", + "issue:12126", + "issue:125", + "issue:12896", + "issue:13105", + "issue:13232", + "issue:13301", + "issue:13477", + "issue:1462", + "issue:1574", + "issue:1708", + "issue:2015", + "issue:2035", + "issue:2439", + "issue:2548", + "issue:3026", + "issue:3418", + "issue:3803", + "issue:3886", + "issue:3887", + "issue:3910", + "issue:4014", + "issue:4337", + "issue:4366", + "issue:4504", + "issue:5124", + "issue:5458", + "issue:562", + "issue:6238", + "issue:6533", + "issue:6727", + "issue:7936", + "issue:8503", + "issue:8594", + "issue:8723", + "issue:8832", + "issue:8960", + "issue:9171", + "issue:945", + "issue:9670", + "issue:9686", + "issue:9768" + ], + "soft_pairs": [ + "issue:1462|issue:562", + "issue:9171|issue:9768", + "issue:3886|issue:8594", + "issue:125|issue:2548", + "issue:12896|issue:13105", + "issue:3910|issue:5458", + "issue:2439|issue:6238", + "issue:4504|issue:7936", + "issue:8723|issue:8960", + "issue:1013|issue:3887", + "issue:10198|issue:10675", + "issue:10591|issue:9686", + "issue:10962|issue:8832", + "issue:13232|issue:13301", + "issue:2015|issue:6533", + "issue:3026|issue:3803", + "issue:3418|issue:4366", + "issue:4337|issue:6727", + "issue:5458|issue:9670", + "issue:1083|issue:945", + "issue:11644|issue:8503", + "issue:2035|issue:4014", + "issue:12126|issue:13477", + "issue:5124|issue:945", + "issue:1574|issue:1708" + ], + "prepared_review_unit_hash": "71cda6875e5b5dd20147f731e4e62f70eb5313cb9a1411b335a6d6f7de4c450e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34007, + "estimated_input_tokens": 8502, + "estimated_eval_tokens": 17260 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: the paired issues share only broad topical similarity (mostly pipeline/training/runtime errors) and do not appear to describe the same underlying bug or change. No canonical issue/PR stands out.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1462", + "right": "issue:562", + "accept": false, + "reason": "Both are inpainting-related, but one is a multi-image/multi-mask unassigned-variable bug and the other is an LMSDiscreteScheduler compatibility issue; different failure modes and code paths." + }, + { + "left": "issue:9171", + "right": "issue:9768", + "accept": false, + "reason": "Both concern image quality/regression in generation, but one is an SDXL from_single_file v_prediction loading issue and the other is a torch 2.5 CuDNN SDPA NaN problem; not the same bug." + }, + { + "left": "issue:3886", + "right": "issue:8594", + "accept": false, + "reason": "Both are training crashes, but one is a distributed collective timeout and the other is an UnboundLocalError in an SD3 dreambooth script; unrelated root causes." + }, + { + "left": "issue:125", + "right": "issue:2548", + "accept": false, + "reason": "Both involve training/accelerate usage, but one is an accelerate launch argument error and the other is a LoRA saving failure; different problems." + }, + { + "left": "issue:12896", + "right": "issue:13105", + "accept": false, + "reason": "Both are Wan model inference issues, but one is an MPS-specific transformer switch/cache_context error and the other is output noise in a different model/version; not the same code-path bug." + }, + { + "left": "issue:3910", + "right": "issue:5458", + "accept": false, + "reason": "Both mention LoRA loading, but one says certain LoRAs will not load while the other is about local_files_only not being respected; different defects." + }, + { + "left": "issue:2439", + "right": "issue:6238", + "accept": false, + "reason": "Both are pipeline-related, but one is inpainting load/run failure and the other is an SDXL VAE issue; too broad to be the same bug." + }, + { + "left": "issue:4504", + "right": "issue:7936", + "accept": false, + "reason": "Both concern safetensors loading, but one is a local variable assignment error in LoRA loading and the other is invalid JSON parsing for a local safetensors file; different failure points." + }, + { + "left": "issue:8723", + "right": "issue:8960", + "accept": false, + "reason": "Both involve SD3/training behavior, but one is SD3 ControlNet initialization and the other is gradient accumulation in a basic training example; not mergeable as one fix." + }, + { + "left": "issue:1013", + "right": "issue:3887", + "accept": false, + "reason": "Both are about pipeline output differences, but one is StableDiffusionMegaPipeline on Diffusers 6.0 and the other is a generic SD pipeline effect mismatch; insufficiently specific and likely unrelated." + }, + { + "left": "issue:10198", + "right": "issue:10675", + "accept": false, + "reason": "Both are Flux-related config issues, but one is a weight mismatch and the other is a scheduler max_shift difference; distinct configuration problems." + }, + { + "left": "issue:10591", + "right": "issue:9686", + "accept": false, + "reason": "Both touch model training/inference bugs, but one is an SD3 LoRA training-script issue and the other is an SD3 ControlNet pooled projections bug; different components." + }, + { + "left": "issue:10962", + "right": "issue:8832", + "accept": false, + "reason": "Both mention shape/runtime failures, but one is Cogview4 prompt-embed shape handling and the other is a generic pipe(prompt).images[0] crash; not clearly the same bug." + }, + { + "left": "issue:13232", + "right": "issue:13301", + "accept": false, + "reason": "One is a support request for LTX-2.3, the other is a feature request for PixArtAlpha modular pipeline support; not the same underlying issue." + }, + { + "left": "issue:2015", + "right": "issue:6533", + "accept": false, + "reason": "These are unrelated pipeline feature requests: categorical diffusion vs. Marigold monocular depth estimation." + }, + { + "left": "issue:3026", + "right": "issue:3803", + "accept": false, + "reason": "Both are training-script related, but one is an instruct_pix2pix accelerate-launch error and the other is multi-GPU LoRA sharing; different scripts and failures." + }, + { + "left": "issue:3418", + "right": "issue:4366", + "accept": false, + "reason": "Both are training script enhancements, but one asks for text-encoder support in a LoRA script and the other is full U-Net fine-tuning for SDXL; not the same change." + }, + { + "left": "issue:4337", + "right": "issue:6727", + "accept": false, + "reason": "Both are runtime errors, but one is torch.compile with ControlNetReference and the other is a QKV dtype mismatch in attention; unrelated code paths." + }, + { + "left": "issue:5458", + "right": "issue:9670", + "accept": false, + "reason": "Both involve LoRA loading/backends, but one is about local_files_only handling and the other is a PEFT backend requirement; different problems." + }, + { + "left": "issue:1083", + "right": "issue:945", + "accept": false, + "reason": "Both concern ONNX conversion/output correctness, but one is FP16 ONNX incorrect output and the other is adding a custom checkpoint to ONNX conversion; distinct issues." + }, + { + "left": "issue:11644", + "right": "issue:8503", + "accept": false, + "reason": "Both are initialization/training failures, but one is LTXConditionPipeline init-from-pretrained and the other is SD3 gradient checkpointing crash; not the same underlying bug." + }, + { + "left": "issue:2035", + "right": "issue:4014", + "accept": false, + "reason": "Both mention poor output quality, but one is tiled upscaling artifacts and the other is SDXL watermark pixel artifacts; different features and fixes." + }, + { + "left": "issue:12126", + "right": "issue:13477", + "accept": false, + "reason": "Both are pipeline bugs with object-scope style failures, but one is wan2.2 transformer/cache_context switching and the other is ErnieImage callback locals() scope; unrelated components." + }, + { + "left": "issue:5124", + "right": "issue:945", + "accept": false, + "reason": "Both reference checkpoints/safetensors/ONNX workflows, but one is how to fine tune a checkpoint .safetensor and the other is converting a custom checkpoint to ONNX; not the same bug." + }, + { + "left": "issue:1574", + "right": "issue:1708", + "accept": false, + "reason": "One is Dreambooth training failing on 8GB bare metal Linux, the other is a question about passing a YAML config for inference; different topics and not a duplicate." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 314, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10039", + "issue:10145", + "issue:10573", + "issue:10743", + "issue:10994", + "issue:11396", + "issue:11464", + "issue:11661", + "issue:11821", + "issue:12096", + "issue:12221", + "issue:13301", + "issue:1466", + "issue:2402", + "issue:2552", + "issue:2844", + "issue:3124", + "issue:3238", + "issue:3300", + "issue:3803", + "issue:4107", + "issue:4206", + "issue:4453", + "issue:4615", + "issue:4677", + "issue:5246", + "issue:5327", + "issue:5544", + "issue:5934", + "issue:6087", + "issue:6323", + "issue:6366", + "issue:6481", + "issue:7009", + "issue:7368", + "issue:7418", + "issue:7460", + "issue:7798", + "issue:7968", + "issue:8509", + "issue:8588", + "issue:8609", + "issue:9147", + "issue:9306", + "issue:9476", + "issue:9488", + "issue:9551" + ], + "soft_pairs": [ + "issue:7009|issue:9488", + "issue:10743|issue:11661", + "issue:5246|issue:7968", + "issue:6087|issue:6366", + "issue:7460|issue:7798", + "issue:3300|issue:4615", + "issue:5544|issue:9147", + "issue:5327|issue:5544", + "issue:5934|issue:9306", + "issue:10039|issue:8588", + "issue:10145|issue:11396", + "issue:12221|issue:13301", + "issue:1466|issue:2402", + "issue:2844|issue:4677", + "issue:4107|issue:6481", + "issue:4453|issue:7368", + "issue:6323|issue:8609", + "issue:8509|issue:9551", + "issue:2552|issue:3803", + "issue:3124|issue:3238", + "issue:4206|issue:7418", + "issue:10994|issue:9476", + "issue:11821|issue:12096", + "issue:10573|issue:11464" + ], + "prepared_review_unit_hash": "25305ff556ffbbf23fb24b6b88c4c692b6e179af8d0249f0c8b9d41a09951c14", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33247, + "estimated_input_tokens": 8312, + "estimated_eval_tokens": 16880 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated bug reports and feature requests that only share broad subsystem names (e.g. FLUX, LoRA, pipelines). No soft pair looks like the same concrete underlying issue/change, so all soft-edge candidates should be rejected.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 10994 is a concrete, reproducible bug report in a specific training script (train_dreambooth_lora_flux) with a clear failure mode and active discussion, making it a strong representative issue.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10994 is the clearest standalone bug report here: specific scope, concrete crash, and enough detail to anchor triage.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7009", + "right": "issue:9488", + "accept": false, + "reason": "Different pipelines and symptoms: one reports poor generations in TextToVideoSDPipeline, the other a Lumina pipeline failing to generate any image." + }, + { + "left": "issue:10743", + "right": "issue:11661", + "accept": false, + "reason": "One is a training feature request for ZeRO-3 in FLUX training; the other is a script bug about args.max_train_steps when None in flux-control." + }, + { + "left": "issue:5246", + "right": "issue:7968", + "accept": false, + "reason": "Unrelated code paths: missing CLIPImageProjection attribute vs attention logic in UNetMotionModel's motion module." + }, + { + "left": "issue:6087", + "right": "issue:6366", + "accept": false, + "reason": "Both involve LoRA training/loading, but they are different concrete bugs in different scripts and failure points, so they should not be merged." + }, + { + "left": "issue:7460", + "right": "issue:7798", + "accept": false, + "reason": "Different model/components entirely: AdaLayerNorm bug versus a PixArtAlphaPipeline bug." + }, + { + "left": "issue:3300", + "right": "issue:4615", + "accept": false, + "reason": "Generic brokenness complaint versus a specific ImportError for a controlnet pipeline; not the same issue." + }, + { + "left": "issue:5544", + "right": "issue:9147", + "accept": false, + "reason": "Different features and failures: FreeU disable error in SDXL versus an autopipeline error for FLUX." + }, + { + "left": "issue:5327", + "right": "issue:5544", + "accept": false, + "reason": "Different concerns: textual inversion loading in SDXL versus FreeU toggling behavior." + }, + { + "left": "issue:5934", + "right": "issue:9306", + "accept": false, + "reason": "Different bug classes and code paths: torch.compile error versus wrong device selection in CogVideoXPipeline.from_pretrained." + }, + { + "left": "issue:10039", + "right": "issue:8588", + "accept": false, + "reason": "Completely different areas: UNet block argument wiring versus a missing conversion script for SD3." + }, + { + "left": "issue:10145", + "right": "issue:11396", + "accept": false, + "reason": "OOM during FLUX Dreambooth LoRA training is unrelated to converting a HiDream LoRA for ComfyUI." + }, + { + "left": "issue:12221", + "right": "issue:13301", + "accept": false, + "reason": "Both are feature requests, but they request different model support and different pipeline integrations." + }, + { + "left": "issue:1466", + "right": "issue:2402", + "accept": false, + "reason": "Different problems: OS compatibility for the Dreambooth example versus CUDA-not-compiled runtime environment failure." + }, + { + "left": "issue:2844", + "right": "issue:4677", + "accept": false, + "reason": "Different pipelines and error conditions: explicit image embeddings in StableUnCLIPImg2ImgPipeline versus an inpainting doc script error." + }, + { + "left": "issue:4107", + "right": "issue:6481", + "accept": false, + "reason": "Both are training-related, but one is SDXL LoRA quality/noise regression and the other is batch-size>1 training failure." + }, + { + "left": "issue:4453", + "right": "issue:7368", + "accept": false, + "reason": "LoRA conversion for AUTO1111 is unrelated to a GLIGEN training question." + }, + { + "left": "issue:6323", + "right": "issue:8609", + "accept": false, + "reason": "Different pipeline features: TensorRT Img2Img guidance_scale attribute bug versus custom pipeline/IP-Adapter compatibility." + }, + { + "left": "issue:8509", + "right": "issue:9551", + "accept": false, + "reason": "Both mention controlnets, but one is a ControlNet++ request and the other asks about x-labs FLUX controlnet models; not the same change." + }, + { + "left": "issue:2552", + "right": "issue:3803", + "accept": false, + "reason": "Tutorial question for StableDiffusionInpaintPipeline versus multi-GPU LoRA training sharing bug; unrelated." + }, + { + "left": "issue:3124", + "right": "issue:3238", + "accept": false, + "reason": "Both are conversion-related, but one is general Stable Diffusion model conversion and the other is converting LoRA safetensors to diffusers format." + }, + { + "left": "issue:4206", + "right": "issue:7418", + "accept": false, + "reason": "Different bug reports in different files/features: SDXL controlnet training breakage versus a community upscale pipeline bug." + }, + { + "left": "issue:10994", + "right": "issue:9476", + "accept": false, + "reason": "Same training script family, but different failures: batch-size crash versus validation dtype mismatch." + }, + { + "left": "issue:11821", + "right": "issue:12096", + "accept": false, + "reason": "Same pipeline name, but different unsupported feature/error: list-of-prompts handling versus apply_group_offloading incompatibility." + }, + { + "left": "issue:10573", + "right": "issue:11464", + "accept": false, + "reason": "Different pipelines and subsystems: multi-controlnet padding_mask_crop bug versus loading a FLUX LoRA into a FluxControlPipeline." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 315, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10191", + "issue:10427", + "issue:10472", + "issue:11343", + "issue:11374", + "issue:11547", + "issue:1156", + "issue:11938", + "issue:12459", + "issue:13277", + "issue:1375", + "issue:1572", + "issue:1655", + "issue:1672", + "issue:1730", + "issue:2100", + "issue:2414", + "issue:2663", + "issue:3140", + "issue:3149", + "issue:3162", + "issue:3441", + "issue:4213", + "issue:4303", + "issue:4701", + "issue:5380", + "issue:5616", + "issue:5969", + "issue:6160", + "issue:6323", + "issue:6363", + "issue:6481", + "issue:6552", + "issue:6610", + "issue:6821", + "issue:684", + "issue:7300", + "issue:7341", + "issue:7390", + "issue:7736", + "issue:785", + "issue:8472", + "issue:9186", + "issue:9619", + "issue:9645", + "issue:9660", + "issue:9729", + "issue:9971" + ], + "soft_pairs": [ + "issue:11343|issue:9645", + "issue:11374|issue:7736", + "issue:3162|issue:7341", + "issue:6481|issue:6821", + "issue:8472|issue:9660", + "issue:2663|issue:4303", + "issue:12459|issue:13277", + "issue:3149|issue:785", + "issue:1156|issue:1730", + "issue:6610|issue:7390", + "issue:10472|issue:9729", + "issue:3441|issue:7300", + "issue:6363|issue:6552", + "issue:4213|issue:5616", + "issue:4701|issue:6160", + "issue:10191|issue:6323", + "issue:10427|issue:9619", + "issue:11547|issue:9186", + "issue:11938|issue:9971", + "issue:1375|issue:2100", + "issue:1572|issue:5380", + "issue:1655|issue:684", + "issue:1672|issue:5969", + "issue:2414|issue:3140" + ], + "prepared_review_unit_hash": "436ddae95cd122b7aff94c56c50ebcfa410981f555c36c26b5f5849816af8791", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34067, + "estimated_input_tokens": 8517, + "estimated_eval_tokens": 17290 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No convincing duplicate cluster here: the items span many unrelated bugs, feature requests, and training/import/runtime issues. All soft-edge pairs look like false positives and should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11343", + "right": "issue:9645", + "accept": false, + "reason": "Both are import failures, but one is a broad transformers-version compatibility problem while the other is a specific missing `StableDiffusionLoraLoaderMixin` symbol; different underlying causes." + }, + { + "left": "issue:11374", + "right": "issue:7736", + "accept": false, + "reason": "Feature requests in different areas: SkyReels-V2 support vs adding callbacks support. Not the same change." + }, + { + "left": "issue:3162", + "right": "issue:7341", + "accept": false, + "reason": "Different errors and code paths: missing `learned_classifier_free_sampling_embeddings` vs `TorchVersion`/`Version` comparison failure." + }, + { + "left": "issue:6481", + "right": "issue:6821", + "accept": false, + "reason": "Unrelated training problems: batch size > 1 Dreambooth failure vs invalid `caption_column` value validation error." + }, + { + "left": "issue:8472", + "right": "issue:9660", + "accept": false, + "reason": "One asks about pre-training LoRA with other modules, the other reports `from_pretrained` not setting modules to eval mode; not the same bug." + }, + { + "left": "issue:2663", + "right": "issue:4303", + "accept": false, + "reason": "Both involve loading models, but one is a safetensors/safety-checker loading failure and the other is SDXL crashing in Flax; different subsystems." + }, + { + "left": "issue:12459", + "right": "issue:13277", + "accept": false, + "reason": "Both mention performance/batch behavior, but one is FLUX.1-Kontext-dev throughput and the other is QwenImagePipeline with Ulysses SP batch inputs." + }, + { + "left": "issue:3149", + "right": "issue:785", + "accept": false, + "reason": "Non-determinism in outputs vs scheduler performance cost; clearly different issues." + }, + { + "left": "issue:1156", + "right": "issue:1730", + "accept": false, + "reason": "Generic training failure vs converted Dreambooth CKPT compatibility with webui; not the same underlying problem." + }, + { + "left": "issue:6610", + "right": "issue:7390", + "accept": false, + "reason": "Both touch `from_single_file` in inpaint/controlnet contexts, but the symptoms differ: `num_in_channels` ignored vs explicit controlnet inpaint error." + }, + { + "left": "issue:10472", + "right": "issue:9729", + "accept": false, + "reason": "Different LoRA/Flux problems: unexpected `lora_bias` argument vs unexpected save-model handling for T5EncoderModel." + }, + { + "left": "issue:3441", + "right": "issue:7300", + "accept": false, + "reason": "Both are memory-related, but one is PyTorch 2 without xformers under certain situations and the other is `set_default_attn_processor()` increasing UNet memory; different triggers." + }, + { + "left": "issue:6363", + "right": "issue:6552", + "accept": false, + "reason": "Training FP16 issues, but one is `Attempting to unscale FP16 gradients` in a LoRA script and the other is a resuming problem tracker for FP16 examples." + }, + { + "left": "issue:4213", + "right": "issue:5616", + "accept": false, + "reason": "Different loading failures: `from_single_file()` needing accelerate vs downloaded SDXL safetensors error." + }, + { + "left": "issue:4701", + "right": "issue:6160", + "accept": false, + "reason": "AutoencoderKL CUDA float32 gradient error vs textual inversion Colab trainer error; unrelated code paths." + }, + { + "left": "issue:10191", + "right": "issue:6323", + "accept": false, + "reason": "GPU device selection bug vs TensorRT pipeline attribute-setting bug; no shared concrete defect." + }, + { + "left": "issue:10427", + "right": "issue:9619", + "accept": false, + "reason": "GGUF loading for AuraFlow vs adding interrupt support to ControlNet pipeline; unrelated changes." + }, + { + "left": "issue:11547", + "right": "issue:9186", + "accept": false, + "reason": "Both involve ControlNet/Flux themes, but one is a feature request for Flux fill dev support and the other is differential diffusion with ControlNet pipeline behavior." + }, + { + "left": "issue:11938", + "right": "issue:9971", + "accept": false, + "reason": "RMSNorm parameter registration bug vs DEIS scheduler not working on FLUX; different functionality." + }, + { + "left": "issue:1375", + "right": "issue:2100", + "accept": false, + "reason": "Both are import errors, but one is a general breakage from PR #1365 and the other is a missing `UnCLIPImageVariationPipeline` import." + }, + { + "left": "issue:1572", + "right": "issue:5380", + "accept": false, + "reason": "Stable Diffusion Mega img2img/inpainting failure vs community pipeline mixture tiling; unrelated pipeline features." + }, + { + "left": "issue:1655", + "right": "issue:684", + "accept": false, + "reason": "Initialization failure for `StableDiffusionImageVariationPipeline` vs AMD training memory optimization issue; different bugs." + }, + { + "left": "issue:1672", + "right": "issue:5969", + "accept": false, + "reason": "Wrong image size after conversion vs DDIM NaNs with ZSNR patch; unrelated symptoms and fixes." + }, + { + "left": "issue:2414", + "right": "issue:3140", + "accept": false, + "reason": "`transformers` NameError while loading a model vs inability to import ControlNet pipeline; different import problems with different causes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 316, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10120", + "issue:10752", + "issue:10798", + "issue:11302", + "issue:11315", + "issue:12011", + "issue:12034", + "issue:12110", + "issue:1251", + "issue:12529", + "issue:13107", + "issue:1407", + "issue:1546", + "issue:1556", + "issue:1792", + "issue:2299", + "issue:2780", + "issue:2907", + "issue:3086", + "issue:3338", + "issue:3428", + "issue:3674", + "issue:3908", + "issue:4841", + "issue:4972", + "issue:5365", + "issue:5963", + "issue:6028", + "issue:6132", + "issue:6362", + "issue:6452", + "issue:6593", + "issue:6964", + "issue:7180", + "issue:7220", + "issue:7440", + "issue:7485", + "issue:8182", + "issue:8495", + "issue:8673", + "issue:8752", + "issue:9007", + "issue:9239", + "issue:9448", + "issue:9479", + "issue:9637", + "issue:9732" + ], + "soft_pairs": [ + "issue:2780|issue:3674", + "issue:3086|issue:6452", + "issue:4841|issue:9239", + "issue:5963|issue:9007", + "issue:7440|issue:8673", + "issue:9479|issue:9637", + "issue:4972|issue:7180", + "issue:11302|issue:11315", + "issue:6964|issue:7220", + "issue:6593|issue:8752", + "issue:12011|issue:9732", + "issue:5365|issue:6028", + "issue:8182|issue:8495", + "issue:10120|issue:6362", + "issue:1407|issue:1792", + "issue:2299|issue:3338", + "issue:12034|issue:12529", + "issue:1251|issue:2907", + "issue:10798|issue:13107", + "issue:7485|issue:9448", + "issue:10752|issue:12110", + "issue:3428|issue:3908", + "issue:10100|issue:6132", + "issue:1546|issue:1556" + ], + "prepared_review_unit_hash": "8a9606e26e037c2021264d8636c61b620adcf3b981c699ff6444bf7562599c30", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33867, + "estimated_input_tokens": 8467, + "estimated_eval_tokens": 17190 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous; most soft edges are only broad subsystem similarity, not duplicates. The only plausible duplicate is the feature_extractor/safety_checker pipeline-init failure pair (1407 \u2194 1792).", + "confidence": 0.52, + "canonical_issue_reason": "issue:1792 is the clearest and broadest report of the shared initialization failure around missing feature_extractor/safety_checker; it generalizes the same code-path problem more than the PR-specific regression report in 1407.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:1792 has a concrete repro and captures the underlying pipeline initialization bug most generally.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2780", + "right": "issue:3674", + "accept": false, + "reason": "Different failures: FP16 gradient unscale vs dataset metadata parsing." + }, + { + "left": "issue:3086", + "right": "issue:6452", + "accept": false, + "reason": "Unrelated pipeline behaviors: audio-to-image vs adding masked_image_latents input." + }, + { + "left": "issue:4841", + "right": "issue:9239", + "accept": false, + "reason": "Both involve image generation issues, but one is image-size handling and the other is black output with ControlNet-Canny." + }, + { + "left": "issue:5963", + "right": "issue:9007", + "accept": false, + "reason": "Different bugs: Kandinsky embeds-vs-prompts handling vs DPMSolver black output at order 3." + }, + { + "left": "issue:7440", + "right": "issue:8673", + "accept": false, + "reason": "Dataset availability is unrelated to SD3 attention-mask handling." + }, + { + "left": "issue:9479", + "right": "issue:9637", + "accept": false, + "reason": "Inpaint mask preservation and XFormers attention-mask/bfloat16 failure are distinct code paths." + }, + { + "left": "issue:4972", + "right": "issue:7180", + "accept": false, + "reason": "Example runtime failure vs SDXL ControlNet CPU offload failure; not the same bug." + }, + { + "left": "issue:11302", + "right": "issue:11315", + "accept": false, + "reason": "Quantized model save issue and MPS float64 rope init bug are unrelated." + }, + { + "left": "issue:6964", + "right": "issue:7220", + "accept": false, + "reason": "Same pipeline family, but one is 4-channel UNet compatibility and the other is unload_ip_adapter cleanup." + }, + { + "left": "issue:6593", + "right": "issue:8752", + "accept": false, + "reason": "LoRA loading request for SDXL Img2Img is unrelated to offline from_single_file loading." + }, + { + "left": "issue:12011", + "right": "issue:9732", + "accept": false, + "reason": "Both mention OOM, but on different models/tasks and likely different causes." + }, + { + "left": "issue:5365", + "right": "issue:6028", + "accept": false, + "reason": "num_inference_steps validation in txt2video is unrelated to Kandinsky OOM." + }, + { + "left": "issue:8182", + "right": "issue:8495", + "accept": false, + "reason": "Logger typo and SD3 tokenizer enum parsing failure are unrelated." + }, + { + "left": "issue:10120", + "right": "issue:6362", + "accept": false, + "reason": "Both are memory complaints, but one is DreamBooth+SD3 and the other is dataset mapping during SDXL fine-tuning." + }, + { + "left": "issue:1407", + "right": "issue:1792", + "accept": true, + "reason": "Both describe the same underlying initialization bug: pipelines error when feature_extractor/safety_checker are missing from SD/SD2-style variants." + }, + { + "left": "issue:2299", + "right": "issue:3338", + "accept": false, + "reason": "Fast-stable-diffusion checkpoint support request vs slow import performance; different issues." + }, + { + "left": "issue:12034", + "right": "issue:12529", + "accept": false, + "reason": "Both are Wan 2.2 i2v reports, but one is poor quality and the other is a tiled VAE tensor mismatch." + }, + { + "left": "issue:1251", + "right": "issue:2907", + "accept": false, + "reason": "Black image on MPS in RePaintPipeline vs calling enable_model_cpu_offload twice; not the same bug." + }, + { + "left": "issue:10798", + "right": "issue:13107", + "accept": false, + "reason": "Flux quantization device-combination bug and Flux Klein LoRA missing-key errors are distinct." + }, + { + "left": "issue:7485", + "right": "issue:9448", + "accept": false, + "reason": "Safety checker NoneType callable error vs IP-Adapter tuple shape AttributeError in ControlNet inpaint." + }, + { + "left": "issue:10752", + "right": "issue:12110", + "accept": false, + "reason": "FP16 gradient unscale bug and UNet1DModel GaussianFourier init bug are unrelated." + }, + { + "left": "issue:3428", + "right": "issue:3908", + "accept": false, + "reason": "Different concerns: unexpected img2img behavior vs caching a model in Streamlit." + }, + { + "left": "issue:10100", + "right": "issue:6132", + "accept": false, + "reason": "ConsisID support request is unrelated to adding LOADABLE_CLASSES." + }, + { + "left": "issue:1546", + "right": "issue:1556", + "accept": false, + "reason": "Truncation warning causing batched-input errors is not the same as black images with NSFW warning." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The output is conservative and grounded in the packet. The only accepted duplicate pair, issue:1407 \u2194 issue:1792, is well supported by the titles describing the same missing feature_extractor/safety_checker initialization failure. The other soft-edge rejections are appropriately cautious and do not overclaim duplication." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 317, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10526", + "issue:1068", + "issue:1082", + "issue:1866", + "issue:1925", + "issue:2035", + "issue:235", + "issue:2410", + "issue:2577", + "issue:3103", + "issue:3646", + "issue:3853", + "issue:3954", + "issue:4069", + "issue:4530", + "issue:4630", + "issue:4638", + "issue:4686", + "issue:4692", + "issue:4747", + "issue:4885", + "issue:490", + "issue:4907", + "issue:4911", + "issue:5226", + "issue:525", + "issue:5335", + "issue:5520", + "issue:5572", + "issue:5963", + "issue:6087", + "issue:6232", + "issue:6234", + "issue:6536", + "issue:6657", + "issue:6807", + "issue:7080", + "issue:725", + "issue:7506", + "issue:7694", + "issue:7708", + "issue:8708", + "issue:8831", + "issue:8930", + "issue:9063", + "issue:9147", + "issue:9293", + "issue:967" + ], + "soft_pairs": [ + "issue:235|issue:525", + "issue:3853|issue:5520", + "issue:4630|issue:725", + "issue:4911|issue:9147", + "issue:7080|issue:8930", + "issue:3954|issue:6087", + "issue:4686|issue:5335", + "issue:4885|issue:8831", + "issue:10526|issue:9293", + "issue:1925|issue:2577", + "issue:7708|issue:9063", + "issue:6232|issue:7506", + "issue:1068|issue:1082", + "issue:3646|issue:4747", + "issue:2035|issue:4530", + "issue:2410|issue:5572", + "issue:3103|issue:4907", + "issue:4069|issue:7694", + "issue:5963|issue:6807", + "issue:6234|issue:8708", + "issue:6536|issue:6657", + "issue:1866|issue:967", + "issue:4692|issue:490", + "issue:4638|issue:5226" + ], + "prepared_review_unit_hash": "61a9c4dd3ce67e744b40c02a232608f29625f8251521fca549c808d59a34a01c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34093, + "estimated_input_tokens": 8524, + "estimated_eval_tokens": 17304 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The listed issues are broadly heterogeneous; the proposed soft pairs mostly share generic ML/diffusers wording but not the same underlying bug or fix path. No single issue stands out as a safe canonical duplicate target for the whole set.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue: the collection mixes unrelated problems (CUDA/runtime failures, scheduler math, model loading/serialization, LoRA/Textual Inversion, ControlNet, Flux, SDXL, and architecture bugs), so collapsing them into one duplicate target would be unsafe.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue: none of the candidates cleanly represents the rest of the set, and the closest-looking items still point to distinct code paths and failure modes.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:235", + "right": "issue:525", + "accept": false, + "reason": "Both mention training/CUDA-adjacent problems, but one is a standalone CUDA porting issue and the other is a local image loading notebook problem; different bug and fix path." + }, + { + "left": "issue:3853", + "right": "issue:5520", + "accept": false, + "reason": "Both are about weight/header deserialization, but the concrete failures differ: LoRA weight loading vs generic HeaderTooLarge serialization error." + }, + { + "left": "issue:4630", + "right": "issue:725", + "accept": false, + "reason": "One asks to modify LoRA weights after loading; the other is about pipeline device mapping / accelerate memory reduction. Related area, but not the same bug." + }, + { + "left": "issue:4911", + "right": "issue:9147", + "accept": false, + "reason": "Both involve Flux/autopipeline loading errors, but one is a custom text_encoder loading failure and the other is a general Flux autopipeline error; not clearly the same code-path issue." + }, + { + "left": "issue:7080", + "right": "issue:8930", + "accept": false, + "reason": "Both concern SDXL ControlNet behavior, but one is a training/inference output mismatch while the other is pose-image conditioning not being respected; different symptoms and likely different causes." + }, + { + "left": "issue:3954", + "right": "issue:6087", + "accept": false, + "reason": "One is a wrong step-count bug in multi-GPU scheduler training, the other is incorrect LoRA alpha persistence on save/load; same broad training area, not same defect." + }, + { + "left": "issue:4686", + "right": "issue:5335", + "accept": false, + "reason": "One is about loading a locally saved pipeline, the other about deploying behind a network block; these are unrelated operational/loading issues." + }, + { + "left": "issue:4885", + "right": "issue:8831", + "accept": false, + "reason": "Both touch LoRA support, but one is an SDXL ControlNet pipeline load issue and the other is unsupported LoRA layers continuation; not the same underlying change." + }, + { + "left": "issue:10526", + "right": "issue:9293", + "accept": false, + "reason": "Both are low-level tensor/storage errors, but one is an optimum.quanto WeightQBytesTensor constructor failure and the other is a missing Float8 storage type; different stack and bug." + }, + { + "left": "issue:1925", + "right": "issue:2577", + "accept": false, + "reason": "One is half-precision inpainting failure, the other is DDIMInverseScheduler behavior; both are diffusion issues but not the same bug." + }, + { + "left": "issue:7708", + "right": "issue:9063", + "accept": false, + "reason": "Both involve Flux/SDXL-style model loading, but one is a missing tokenizer/config file and the other is a tokenizer instantiation compatibility error; distinct failures." + }, + { + "left": "issue:6232", + "right": "issue:7506", + "accept": false, + "reason": "Both mention scheduler/model loading problems, but one is a DPMSolver add_noise edge case and the other is a meta-tensor copy/model init mismatch; unrelated code paths." + }, + { + "left": "issue:1068", + "right": "issue:1082", + "accept": false, + "reason": "Both are runtime failures, but one is a distributed launch/elastic error and the other is a CUDA cublasLtMatmul failure in Dreambooth; not the same defect." + }, + { + "left": "issue:3646", + "right": "issue:4747", + "accept": false, + "reason": "Both are model/API loading issues, but one is ControlNetModel.from_unet missing and the other is missing model_index.json after loading a fine-tuned SDXL LoRA; different targets." + }, + { + "left": "issue:2035", + "right": "issue:4530", + "accept": false, + "reason": "Both mention visual artifacts, but one is tiled upscaling launch/tiling behavior and the other is SDXL refiner color artifacts; too different to merge." + }, + { + "left": "issue:2410", + "right": "issue:5572", + "accept": false, + "reason": "One is a Flax training script output-file issue, the other is a FrozenDict missing scaling_factor attribute in a pipeline; unrelated." + }, + { + "left": "issue:3103", + "right": "issue:4907", + "accept": false, + "reason": "Both relate to inpainting, but one is height/width handling and the other is an SDXL inpainting model warning; not the same concrete bug." + }, + { + "left": "issue:4069", + "right": "issue:7694", + "accept": false, + "reason": "One is about unloading textual inversion checkpoints from a pipeline; the other is checkpoint key updates for ControlNetXS. Same ecosystem, different changes." + }, + { + "left": "issue:5963", + "right": "issue:6807", + "accept": false, + "reason": "One is Kandinsky embeddings vs prompts behavior, the other is MPS test collection failure; no shared bug or fix." + }, + { + "left": "issue:6234", + "right": "issue:8708", + "accept": false, + "reason": "One is an example script issue for SDXL training, the other is a missing VAE shift_factor in SD3 dreambooth examples; different model family and issue." + }, + { + "left": "issue:6536", + "right": "issue:6657", + "accept": false, + "reason": "One is UNet3DConditionModel fine-tuning OOM, the other is frames/channels being swapped in UNet3DConditionModel; memory vs tensor-shape bug." + }, + { + "left": "issue:1866", + "right": "issue:967", + "accept": false, + "reason": "Both are scheduler incompatibilities, but one is img2img strength handling in DPMSolverSinglestepScheduler and the other is OnnxStableDiffusionPipeline unsupported operand errors; not the same fix." + }, + { + "left": "issue:4692", + "right": "issue:490", + "accept": false, + "reason": "Both concern memory, but one is a UNet2DConditionModel memory issue and the other is a broad pipeline memory comparison; too generic to treat as duplicate." + }, + { + "left": "issue:4638", + "right": "issue:5226", + "accept": false, + "reason": "Both are about loading LoRA/textual inversion on SDXL pipelines, but one is textual inversion loading and the other is adapter pipeline LoRA support; different pipeline paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 318, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10003", + "issue:10010", + "issue:10194", + "issue:10392", + "issue:10450", + "issue:1062", + "issue:10782", + "issue:11117", + "issue:11570", + "issue:11830", + "issue:12082", + "issue:12783", + "issue:1298", + "issue:13137", + "issue:1478", + "issue:1790", + "issue:2132", + "issue:2276", + "issue:2332", + "issue:2410", + "issue:2728", + "issue:2783", + "issue:2803", + "issue:3086", + "issue:3149", + "issue:3171", + "issue:3451", + "issue:4709", + "issue:4733", + "issue:5315", + "issue:5605", + "issue:5931", + "issue:6232", + "issue:6727", + "issue:6790", + "issue:6906", + "issue:7415", + "issue:761", + "issue:7895", + "issue:8440", + "issue:8578", + "issue:9393", + "issue:9670", + "issue:9756", + "issue:9763", + "issue:9858", + "issue:9926", + "issue:9953" + ], + "soft_pairs": [ + "issue:1062|issue:2332", + "issue:2132|issue:4733", + "issue:5315|issue:6232", + "issue:6790|issue:6906", + "issue:11117|issue:7415", + "issue:2410|issue:3451", + "issue:11570|issue:9953", + "issue:2728|issue:2803", + "issue:10003|issue:8578", + "issue:10010|issue:9756", + "issue:10450|issue:7895", + "issue:10782|issue:13137", + "issue:1298|issue:1478", + "issue:1790|issue:761", + "issue:2783|issue:3171", + "issue:3086|issue:5605", + "issue:3149|issue:5931", + "issue:6727|issue:8440", + "issue:12082|issue:9763", + "issue:2276|issue:4709", + "issue:11830|issue:12783", + "issue:9393|issue:9858", + "issue:10194|issue:9926", + "issue:10392|issue:9670" + ], + "prepared_review_unit_hash": "913085fb97cc5e8bb47aeb7c2eb95d8fbda379ceafff007b044ceb2152db1c86", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33677, + "estimated_input_tokens": 8420, + "estimated_eval_tokens": 17096 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft pair looks like a safe duplicate merge candidate; the set is a heterogeneous mix of unrelated scheduler, Flux, DreamBooth, quantization, docs, and feature-request issues. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": "No single issue is a safe canonical here: the items span unrelated bugs and feature requests across different pipelines, model families, and tooling.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a strong global representative for duplication triage because the cluster does not center on one underlying defect or change.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1062", + "right": "issue:2332", + "accept": false, + "reason": "Both mention DreamBooth, but one is about training repeatability and the other about inference after training; different failure modes and code paths." + }, + { + "left": "issue:2132", + "right": "issue:4733", + "accept": false, + "reason": "Both are DreamBooth training issues, but 'Accelerator.trackers' missing and checkpoint resume optimizer checks are unrelated bugs." + }, + { + "left": "issue:5315", + "right": "issue:6232", + "accept": false, + "reason": "Both are scheduler bugs, but they target different schedulers and different methods (DDIMInverseScheduler.step vs DPMSolverSinglestepScheduler.add_noise)." + }, + { + "left": "issue:6790", + "right": "issue:6906", + "accept": false, + "reason": "LCM state_dict key mismatch and a broken LoRA DreamBooth example are different training problems, not one underlying defect." + }, + { + "left": "issue:11117", + "right": "issue:7415", + "accept": false, + "reason": "Both are dtype-related, but they affect different pipelines/devices and have different mismatch scenarios (SD3 Half/BFloat16 vs AutoencoderKL bfloat16 on TPU)." + }, + { + "left": "issue:2410", + "right": "issue:3451", + "accept": false, + "reason": "Missing model output files in a Flax training script is unrelated to fp16 memory usage in another training script." + }, + { + "left": "issue:11570", + "right": "issue:9953", + "accept": false, + "reason": "Both involve quantized components, but one breaks LoRA weight loading on bitsandbytes text encoders while the other breaks moving a quantized pipeline to CUDA." + }, + { + "left": "issue:2728", + "right": "issue:2803", + "accept": false, + "reason": "A safetensors loading issue and a local docs build failure are clearly unrelated." + }, + { + "left": "issue:10003", + "right": "issue:8578", + "accept": false, + "reason": "A scheduler bug at 1000 inference steps and a meta-tensor copy error during loading are different bugs with different triggers." + }, + { + "left": "issue:10010", + "right": "issue:9756", + "accept": false, + "reason": "A missing model_index.json for FLUX Fill and a missing notebook file are unrelated path/resource issues." + }, + { + "left": "issue:10450", + "right": "issue:7895", + "accept": false, + "reason": "Different Flux/Flax blocks and a broken memory-efficient attention path are not the same concrete problem." + }, + { + "left": "issue:10782", + "right": "issue:13137", + "accept": false, + "reason": "These concern different models and different adapter/LoRA behaviors, so they are not duplicates." + }, + { + "left": "issue:1298", + "right": "issue:1478", + "accept": false, + "reason": "One is a regression in Stable Diffusion img2img; the other is an output-count limitation in the upscale pipeline." + }, + { + "left": "issue:1790", + "right": "issue:761", + "accept": false, + "reason": "A generic training question is not the same as a DreamBooth 8-bit optimizer failure." + }, + { + "left": "issue:2783", + "right": "issue:3171", + "accept": false, + "reason": "Both are ControlNet feature requests, but they ask for different features and do not describe the same change." + }, + { + "left": "issue:3086", + "right": "issue:5605", + "accept": false, + "reason": "Input-audio/output-image and per-image caption support are unrelated feature requests." + }, + { + "left": "issue:3149", + "right": "issue:5931", + "accept": false, + "reason": "Both mention nondeterminism, but they are broad symptom reports in different contexts and versions; not clearly the same underlying bug." + }, + { + "left": "issue:6727", + "right": "issue:8440", + "accept": false, + "reason": "These are different ValueError contexts in different APIs; the titles do not indicate one shared bug." + }, + { + "left": "issue:12082", + "right": "issue:9763", + "accept": false, + "reason": "GPU memory growth after compile and token truncation in Flux ControlNet training are unrelated issues." + }, + { + "left": "issue:2276", + "right": "issue:4709", + "accept": false, + "reason": "A xformers installation problem in a Colab notebook is unrelated to ControlNet guess mode support." + }, + { + "left": "issue:11830", + "right": "issue:12783", + "accept": false, + "reason": "Both are visual-output problems, but black output in framepack fp16 and confetti/noisy output on sm90 are different symptoms with different likely causes." + }, + { + "left": "issue:9393", + "right": "issue:9858", + "accept": false, + "reason": "Saving a Flux DreamBooth LoRA with DeepSpeed and a KeyError: 'train' are not the same bug." + }, + { + "left": "issue:10194", + "right": "issue:9926", + "accept": false, + "reason": "Flux attention masking support and gguf quantize/speed-up support are different feature requests." + }, + { + "left": "issue:10392", + "right": "issue:9670", + "accept": false, + "reason": "A Flux fp8/LoRA compatibility issue and a PEFT-backend requirement for loading LoRA are distinct adapter-loading problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 319, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10277", + "issue:10673", + "issue:11093", + "issue:11581", + "issue:11874", + "issue:12034", + "issue:12108", + "issue:12175", + "issue:13203", + "issue:2371", + "issue:2614", + "issue:2772", + "issue:2776", + "issue:3019", + "issue:3073", + "issue:3087", + "issue:3570", + "issue:3631", + "issue:4057", + "issue:4337", + "issue:4398", + "issue:4554", + "issue:4649", + "issue:4668", + "issue:4782", + "issue:4874", + "issue:5178", + "issue:5593", + "issue:6125", + "issue:6313", + "issue:6667", + "issue:6700", + "issue:7168", + "issue:7348", + "issue:7378", + "issue:7417", + "issue:7661", + "issue:8469", + "issue:8831", + "issue:9223", + "issue:9298", + "issue:9534", + "issue:9660", + "issue:9665", + "issue:9674", + "issue:9704", + "issue:9933" + ], + "soft_pairs": [ + "issue:11093|issue:8469", + "issue:12034|issue:12175", + "issue:2614|issue:3570", + "issue:2776|issue:9223", + "issue:3073|issue:3087", + "issue:3631|issue:4782", + "issue:3631|issue:7378", + "issue:4337|issue:4554", + "issue:4398|issue:8831", + "issue:7168|issue:7661", + "issue:9534|issue:9665", + "issue:3019|issue:6313", + "issue:10673|issue:9704", + "issue:6667|issue:9674", + "issue:2371|issue:4057", + "issue:5593|issue:9660", + "issue:6125|issue:9933", + "issue:10277|issue:12108", + "issue:11874|issue:13203", + "issue:5178|issue:7417", + "issue:2772|issue:4874", + "issue:4668|issue:7348", + "issue:4649|issue:6700", + "issue:11581|issue:9298" + ], + "prepared_review_unit_hash": "aec09ecac663944e4ba13573a8937db2b49beee94d76805425b6be451115debe", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33366, + "estimated_input_tokens": 8342, + "estimated_eval_tokens": 16940 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This looks like a false-positive semantic cluster: the items span unrelated model, pipeline, training, and dependency issues. None of the soft pairs describe the same concrete bug or change, so I would not merge any of them as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue stands out; the set is heterogeneous and the soft-paired items are not the same underlying bug. If forced, 12108 is only the closest thematic match because it is one of the few scheduler/flow-matching reports, but it still does not represent the whole set.", + "canonical_pr_reason": null, + "best_issue_reason": "There is no good global best issue for dedupe here because the cluster is not a true duplicate group. The issues are too diverse in model, subsystem, and failure mode to use one as the representative canonical.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11093", + "right": "issue:8469", + "accept": false, + "reason": "Different problems: TorchAO example breakage vs IP-Adapter face corruption." + }, + { + "left": "issue:12034", + "right": "issue:12175", + "accept": false, + "reason": "Both mention quality, but they are different models and likely different root causes." + }, + { + "left": "issue:2614", + "right": "issue:3570", + "accept": false, + "reason": "Download-directory configuration vs disabling downloads when loading ckpt are separate behaviors." + }, + { + "left": "issue:2776", + "right": "issue:9223", + "accept": false, + "reason": "DeepSpeed/runtime setup issue vs Python dependency conflict; unrelated." + }, + { + "left": "issue:3073", + "right": "issue:3087", + "accept": false, + "reason": "Feature request for consistency models vs LoRA/LyCORIS support request." + }, + { + "left": "issue:3631", + "right": "issue:4782", + "accept": false, + "reason": "Inpainting latents exception is not the same as mask leakage during inpainting." + }, + { + "left": "issue:3631", + "right": "issue:7378", + "accept": false, + "reason": "Inpainting exception vs AnimateDiff video-to-video memory usage; different code paths." + }, + { + "left": "issue:4337", + "right": "issue:4554", + "accept": false, + "reason": "torch.compile ControlNetReference failure is distinct from Kandinsky OOM." + }, + { + "left": "issue:4398", + "right": "issue:8831", + "accept": false, + "reason": "SDXL LORA dreambooth black images and unsupported-layers continuation are not the same bug." + }, + { + "left": "issue:7168", + "right": "issue:7661", + "accept": false, + "reason": "IP-Adapter embedding shape issue vs request to support IP-Adapter in AnimateDiff." + }, + { + "left": "issue:9534", + "right": "issue:9665", + "accept": false, + "reason": "Missing config attribute in FluxMultiControlNetModel is unrelated to accelerator.prepare being affected by DeepSpeed config." + }, + { + "left": "issue:3019", + "right": "issue:6313", + "accept": false, + "reason": "A model request and a pipeline implementation request are different changes." + }, + { + "left": "issue:10673", + "right": "issue:9704", + "accept": false, + "reason": "Shape mismatch in AuraFlow pipeline vs cuDNN frontend execution-plan error; different failure modes." + }, + { + "left": "issue:6667", + "right": "issue:9674", + "accept": false, + "reason": "controlnet-xs-inpaint request is not the same as poor SD3 controlnet training results." + }, + { + "left": "issue:2371", + "right": "issue:4057", + "accept": false, + "reason": "Dreambooth output omission vs blurred finetuned inpainting output are distinct issues." + }, + { + "left": "issue:5593", + "right": "issue:9660", + "accept": false, + "reason": "AutoPipeline.from_pipe with ControlNet vs from_pretrained not setting eval mode are separate bugs." + }, + { + "left": "issue:6125", + "right": "issue:9933", + "accept": false, + "reason": "UnetMotionModel parameter handling is unrelated to missing width/height in SD3 img2img." + }, + { + "left": "issue:10277", + "right": "issue:12108", + "accept": false, + "reason": "Both mention schedulers/flow matching, but they target different pipelines and concrete failure contexts." + }, + { + "left": "issue:11874", + "right": "issue:13203", + "accept": false, + "reason": "LoRA exclude_modules behavior is a different support issue than Zimage LoRA support." + }, + { + "left": "issue:5178", + "right": "issue:7417", + "accept": false, + "reason": "DPM SDE resolution-change bug is unrelated to the Mac lora_fuse_nan test failure." + }, + { + "left": "issue:2772", + "right": "issue:4874", + "accept": false, + "reason": "Local dataset validation in train_controlnet.py is not the same as low_res_scheduler behavior in upscale pipeline." + }, + { + "left": "issue:4668", + "right": "issue:7348", + "accept": false, + "reason": "Weird images from Kohya-style LoRA vs weird results with EulerAncestralControlNetInpaint are not the same concrete bug." + }, + { + "left": "issue:4649", + "right": "issue:6700", + "accept": false, + "reason": "Both are TypeErrors in model internals, but they arise from different parameters and code paths." + }, + { + "left": "issue:11581", + "right": "issue:9298", + "accept": false, + "reason": "IP-Adapter dtype kwarg error is unrelated to missing Float8 storage type." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 320, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10550", + "issue:10640", + "issue:10658", + "issue:11321", + "issue:11408", + "issue:12470", + "issue:1255", + "issue:1290", + "issue:1685", + "issue:181", + "issue:3062", + "issue:307", + "issue:3085", + "issue:3251", + "issue:3382", + "issue:3643", + "issue:3645", + "issue:3674", + "issue:3824", + "issue:3910", + "issue:4075", + "issue:4213", + "issue:4222", + "issue:4453", + "issue:4483", + "issue:4578", + "issue:4585", + "issue:4676", + "issue:4705", + "issue:4756", + "issue:5109", + "issue:5209", + "issue:5328", + "issue:5405", + "issue:5578", + "issue:5645", + "issue:5769", + "issue:5786", + "issue:6164", + "issue:6274", + "issue:6476", + "issue:6742", + "issue:7021", + "issue:7178", + "issue:7813", + "issue:8393", + "issue:8930", + "issue:9046" + ], + "soft_pairs": [ + "issue:6476|issue:7178", + "issue:4213|issue:6742", + "issue:5109|issue:8930", + "issue:181|issue:3382", + "issue:3643|issue:6164", + "issue:10550|issue:11408", + "issue:10640|issue:7021", + "issue:10658|issue:12470", + "issue:11321|issue:9046", + "issue:1255|issue:4585", + "issue:1290|issue:307", + "issue:1685|issue:3251", + "issue:3062|issue:4075", + "issue:3085|issue:4483", + "issue:3645|issue:4705", + "issue:3674|issue:4756", + "issue:3824|issue:7813", + "issue:3910|issue:5328", + "issue:4222|issue:5786", + "issue:4453|issue:5209", + "issue:4578|issue:5405", + "issue:4676|issue:8393", + "issue:5578|issue:5769", + "issue:5645|issue:6274" + ], + "prepared_review_unit_hash": "1a2aa09a712c355ef9219e55d96caf26a4ed2836f8c5f4cf4b01ea5490c7f0db", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33764, + "estimated_input_tokens": 8441, + "estimated_eval_tokens": 17138 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is a loose collection of unrelated issues; the soft-matched pairs only share vague topic similarity and do not describe the same underlying bug or change. All soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6476", + "right": "issue:7178", + "accept": false, + "reason": "Both mention export_to_video, but one is an OpenCV-version crash and the other is a general malfunction report; no evidence they are the same failure mode." + }, + { + "left": "issue:4213", + "right": "issue:6742", + "accept": false, + "reason": "SDXL single-file loading without accelerate is unrelated to DownBlockMotion checkpointing behavior; different code paths and symptoms." + }, + { + "left": "issue:5109", + "right": "issue:8930", + "accept": false, + "reason": "One is a LoRA-loading exception in a specific ControlNet img2img pipeline, the other is poor pose adherence in another pipeline; not the same bug." + }, + { + "left": "issue:181", + "right": "issue:3382", + "accept": false, + "reason": "Both concern training, but one is image writing failure and the other is multi-GPU OOM; different problems and fixes." + }, + { + "left": "issue:3643", + "right": "issue:6164", + "accept": false, + "reason": "INT_MAX overflow during latent decoding is a low-level runtime error, while latent output handling is a pipeline API/behavior issue; unrelated." + }, + { + "left": "issue:10550", + "right": "issue:11408", + "accept": false, + "reason": "LoRA into a quantized base model and set_adapters on compiled models are separate adapter/quantization concerns with different failure points." + }, + { + "left": "issue:10640", + "right": "issue:7021", + "accept": false, + "reason": "DataParallel error in FluxInpaintPipeline is unrelated to HF_HOME import errors." + }, + { + "left": "issue:10658", + "right": "issue:12470", + "accept": false, + "reason": "ONNX Runtime provider_options request and DiffusersQuantizer API additions are different feature requests with no shared code-path bug." + }, + { + "left": "issue:11321", + "right": "issue:9046", + "accept": false, + "reason": "A README bug for Flux controlnet training is documentation-only; Flux VAE latent_channels is a model-architecture question." + }, + { + "left": "issue:1255", + "right": "issue:4585", + "accept": false, + "reason": "Importing DPMSolverMultistepScheduler and SDXL MultiControlNet documentation are unrelated." + }, + { + "left": "issue:1290", + "right": "issue:307", + "accept": false, + "reason": "Community DreamArtist project post is not the same as a request to merge Stable Diffusion pipelines into one." + }, + { + "left": "issue:1685", + "right": "issue:3251", + "accept": false, + "reason": "Telemetry opt-out and ControlNet Shuffle guidance are unrelated issues." + }, + { + "left": "issue:3062", + "right": "issue:4075", + "accept": false, + "reason": "Different docstring questions for different classes; no same underlying bug or change." + }, + { + "left": "issue:3085", + "right": "issue:4483", + "accept": false, + "reason": "Removed cache-folder argument and SDXL generator nondeterminism are separate API/behavior issues." + }, + { + "left": "issue:3645", + "right": "issue:4705", + "accept": false, + "reason": "Training/finetuning questions for inpainting pipelines are not the same problem as modifying train_controlnet.py for inpainting." + }, + { + "left": "issue:3674", + "right": "issue:4756", + "accept": false, + "reason": "metadata.jsonl parsing in LoRA training and a request for style-training support are different training-data concerns." + }, + { + "left": "issue:3824", + "right": "issue:7813", + "accept": false, + "reason": "Question about inpaint pipeline variants and a TODO about passing timesteps are unrelated." + }, + { + "left": "issue:3910", + "right": "issue:5328", + "accept": false, + "reason": "Certain LoRA weights failing to load and Civital LoRAs all failing likely refer to different loader/asset issues without concrete overlap." + }, + { + "left": "issue:4222", + "right": "issue:5786", + "accept": false, + "reason": "Low-resolution dataset training guidance and loading a precomputed cache on another machine are different usage questions." + }, + { + "left": "issue:4453", + "right": "issue:5209", + "accept": false, + "reason": "Converting SDXL LoRA to safetensors for A1111 and black validation images in a script are unrelated." + }, + { + "left": "issue:4578", + "right": "issue:5405", + "accept": false, + "reason": "Documentation claim about VAE usage and from_single_file not overwriting weights are distinct problems." + }, + { + "left": "issue:4676", + "right": "issue:8393", + "accept": false, + "reason": "TAESD latent brightness and checkpoint-shard loading time are unrelated performance/quality topics." + }, + { + "left": "issue:5578", + "right": "issue:5769", + "accept": false, + "reason": "Missing model_index.json and mixed-precision training correctness are unrelated." + }, + { + "left": "issue:5645", + "right": "issue:6274", + "accept": false, + "reason": "LCM support broken and multi-adapter T2I-Adapter behavior are different features with different failure surfaces." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 321, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10798", + "issue:1614", + "issue:1672", + "issue:2201", + "issue:255", + "issue:2704", + "issue:2904", + "issue:3194", + "issue:3249", + "issue:3651", + "issue:3683", + "issue:409", + "issue:4286", + "issue:4450", + "issue:4868", + "issue:5109", + "issue:5178", + "issue:5523", + "issue:5678", + "issue:5791", + "issue:5845", + "issue:5849", + "issue:5969", + "issue:6110", + "issue:6681", + "issue:6869", + "issue:6906", + "issue:6967", + "issue:7000", + "issue:7102", + "issue:7209", + "issue:7333", + "issue:7354", + "issue:7516", + "issue:7548", + "issue:7955", + "issue:7972", + "issue:8477", + "issue:8671", + "issue:8918", + "issue:8943", + "issue:9055", + "issue:9161", + "issue:9366", + "issue:9415", + "issue:9476", + "issue:9477", + "issue:9856" + ], + "soft_pairs": [ + "issue:5845|issue:6906", + "issue:6681|issue:6869", + "issue:7333|issue:7955", + "issue:8671|issue:8918", + "issue:8943|issue:9477", + "issue:6967|issue:7972", + "issue:1672|issue:2201", + "issue:3194|issue:4868", + "issue:3249|issue:7354", + "issue:8477|issue:9415", + "issue:255|issue:409", + "issue:3683|issue:4450", + "issue:7516|issue:9055", + "issue:2904|issue:5523", + "issue:7000|issue:7548", + "issue:3651|issue:5178", + "issue:1614|issue:5969", + "issue:5678|issue:5791", + "issue:5849|issue:9856", + "issue:6110|issue:7209", + "issue:2704|issue:7102", + "issue:4286|issue:5109", + "issue:9161|issue:9476", + "issue:10798|issue:9366" + ], + "prepared_review_unit_hash": "048347806f56ba71778fbc64410a3f5b1b1aea2f649a4caca55ea3549cd56ba3", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34415, + "estimated_input_tokens": 8604, + "estimated_eval_tokens": 17464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated issues that only share broad diffusers/Stable Diffusion themes. I only see one plausible duplicate edge around export_to_video; the rest are different bugs, features, or scripts and should stay separate.", + "confidence": 0.58, + "canonical_issue_reason": "issue:6869 is the most concrete report in the only plausible duplicate pair: it names the exact export_to_video failure and error, while issue:6681 is just a generic version of the same problem.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6869 is the best representative issue for this cluster because it is specific, reproducible, and the clearest candidate for deduplication; most other items are too heterogeneous to serve as a canonical bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5845", + "right": "issue:6906", + "accept": false, + "reason": "Both concern DreamBooth/LoRA training, but they target different setups (PixArt Alpha vs the generic LoRA example) and do not look like the same underlying bug." + }, + { + "left": "issue:6681", + "right": "issue:6869", + "accept": true, + "reason": "Both are about export_to_video failing; 6869 provides the concrete ValueError while 6681 is the generic report, so they likely describe the same bug." + }, + { + "left": "issue:7333", + "right": "issue:7955", + "accept": false, + "reason": "Completely different research/model topics with no shared failure mode or code path." + }, + { + "left": "issue:8671", + "right": "issue:8918", + "accept": false, + "reason": "One is about redundant code in a script, the other about a broken training script; same family of files but not the same bug." + }, + { + "left": "issue:8943", + "right": "issue:9477", + "accept": false, + "reason": "A missing shared library at install time is unrelated to the GatheredParameters callable error." + }, + { + "left": "issue:6967", + "right": "issue:7972", + "accept": false, + "reason": "Different training/inference features and different failure types; no evidence of a shared concrete defect." + }, + { + "left": "issue:1672", + "right": "issue:2201", + "accept": false, + "reason": "Wrong output image size after conversion is unrelated to tokenizer padding differences." + }, + { + "left": "issue:3194", + "right": "issue:4868", + "accept": false, + "reason": "Bitsandbytes package metadata missing is unrelated to the super().__getattr__ AttributeError." + }, + { + "left": "issue:3249", + "right": "issue:7354", + "accept": false, + "reason": "These are distinct scheduler/block API issues with no shared code-path bug." + }, + { + "left": "issue:8477", + "right": "issue:9415", + "accept": false, + "reason": "A distributed training error and a model-parameter API error are not the same underlying issue." + }, + { + "left": "issue:255", + "right": "issue:409", + "accept": false, + "reason": "Shape-validation error in StableDiffusionPipeline is unrelated to the missing enable_attention_slicing attribute." + }, + { + "left": "issue:3683", + "right": "issue:4450", + "accept": false, + "reason": "Implementing save_pretrained for Multi-ControlNet is a different change from multi text inversion not working." + }, + { + "left": "issue:7516", + "right": "issue:9055", + "accept": false, + "reason": "Missing DepthNormalEstimationPipeline export is unrelated to missing StableDiffusionLoraLoaderMixin import." + }, + { + "left": "issue:2904", + "right": "issue:5523", + "accept": false, + "reason": "AudioLDM scheduler initialization failure is unrelated to nondeterministic results across runs." + }, + { + "left": "issue:7000", + "right": "issue:7548", + "accept": false, + "reason": "A PyTorch deprecation warning is not the same as flickering AnimateDiff outputs." + }, + { + "left": "issue:3651", + "right": "issue:5178", + "accept": false, + "reason": "Multi-GPU training failure and DPM-SDE resolution-change failure are different bugs." + }, + { + "left": "issue:1614", + "right": "issue:5969", + "accept": false, + "reason": "Blank black images under autocast and NaNs with ZSNR patch are different numerical issues." + }, + { + "left": "issue:5678", + "right": "issue:5791", + "accept": false, + "reason": "Higher-order derivative failure in attention is unrelated to DreamBooth prior preservation/SNR gamma training behavior." + }, + { + "left": "issue:5849", + "right": "issue:9856", + "accept": false, + "reason": "A missing main() args error is unrelated to distributed port contention." + }, + { + "left": "issue:6110", + "right": "issue:7209", + "accept": false, + "reason": "Offline LoRA loading and offline single-file loading are related thematically but not the same concrete failure." + }, + { + "left": "issue:2704", + "right": "issue:7102", + "accept": false, + "reason": "Multi-ControlNet batching and multi-prompt controlnet bugs are different issues even though they touch the same subsystem." + }, + { + "left": "issue:4286", + "right": "issue:5109", + "accept": false, + "reason": "Both mention load_lora_weights, but they are different pipelines and the reported failures are not clearly the same bug." + }, + { + "left": "issue:9161", + "right": "issue:9476", + "accept": false, + "reason": "Distributed bugs in Flux LoRA training and a validation dtype mismatch are distinct problems." + }, + { + "left": "issue:10798", + "right": "issue:9366", + "accept": false, + "reason": "Flux quantization/device-combination issues and DPMSolver step/strength failures are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The only accepted soft edge, issue:6681 \u2194 issue:6869, is supported by the packet: both are about export_to_video and 6869 provides the concrete ValueError while 6681 is generic. The remaining rejections do not overstate evidence, and the summary appropriately treats the rest of the cluster as heterogeneous." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 322, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10209", + "issue:10337", + "issue:10636", + "issue:10760", + "issue:10891", + "issue:10967", + "issue:11137", + "issue:11962", + "issue:12613", + "issue:12653", + "issue:12666", + "issue:13311", + "issue:1802", + "issue:215", + "issue:2581", + "issue:2645", + "issue:2676", + "issue:280", + "issue:2905", + "issue:2973", + "issue:3336", + "issue:3570", + "issue:4116", + "issue:4186", + "issue:4263", + "issue:4789", + "issue:487", + "issue:5006", + "issue:5064", + "issue:5523", + "issue:5714", + "issue:6383", + "issue:656", + "issue:6632", + "issue:6956", + "issue:7441", + "issue:8320", + "issue:8534", + "issue:8535", + "issue:8589", + "issue:8641", + "issue:8958", + "issue:9031", + "issue:9039", + "issue:9178", + "issue:9467", + "issue:9515" + ], + "soft_pairs": [ + "issue:12613|issue:13311", + "issue:10209|issue:9467", + "issue:2676|issue:3336", + "issue:3570|issue:4789", + "issue:10636|issue:8589", + "issue:11962|issue:9178", + "issue:1802|issue:487", + "issue:10209|issue:12666", + "issue:10337|issue:11137", + "issue:215|issue:656", + "issue:280|issue:2905", + "issue:4116|issue:8320", + "issue:4263|issue:6632", + "issue:5714|issue:7441", + "issue:6383|issue:9515", + "issue:8534|issue:8535", + "issue:9031|issue:9039", + "issue:2973|issue:5523", + "issue:10760|issue:8958", + "issue:5006|issue:6956", + "issue:10891|issue:12653", + "issue:4186|issue:5064", + "issue:10967|issue:8641", + "issue:2581|issue:2645" + ], + "prepared_review_unit_hash": "59fe3be7c957a5b966121e72e4eea7f3ffd11f153b2d4ab8bfa59dc24e11274f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32960, + "estimated_input_tokens": 8240, + "estimated_eval_tokens": 16736 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set; it mixes unrelated issues across different models, pipelines, schedulers, and backend/device bugs. All soft-edge pairs should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a valid canonical duplicate target because the items describe distinct problems or feature requests with different code paths.", + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12613", + "right": "issue:13311", + "accept": false, + "reason": "Different problems: WAN 2.2 i2v latent NaNs vs FLUX2 LoRA key loading prefix mismatch." + }, + { + "left": "issue:10209", + "right": "issue:9467", + "accept": false, + "reason": "Unrelated model/pipeline bugs: CogVideoX fine-tuning error vs CogVideoX I2V num_frames guard-rail." + }, + { + "left": "issue:2676", + "right": "issue:3336", + "accept": false, + "reason": "Scheduler-name question vs DPM-Solver inverse scheduler feature request; not the same issue." + }, + { + "left": "issue:3570", + "right": "issue:4789", + "accept": false, + "reason": "Loading control/download behavior vs DeepFloyd IF download failure; different code paths." + }, + { + "left": "issue:10636", + "right": "issue:8589", + "accept": false, + "reason": "General bug in value_guided_sampling.py vs SD3 DreamBooth prior-preservation training issue." + }, + { + "left": "issue:11962", + "right": "issue:9178", + "accept": false, + "reason": "GGUF quantized FLUX.1-Kontext support request vs Flux LoRA PEFT target-module error." + }, + { + "left": "issue:1802", + "right": "issue:487", + "accept": false, + "reason": "Different generations and symptoms: unet_2d_condition encoder_hidden_states handling vs M1 CPU failure." + }, + { + "left": "issue:10209", + "right": "issue:12666", + "accept": false, + "reason": "Different pipelines: CogVideoX fine-tuning error vs Kandinsky 5.0 video pipeline availability in a release." + }, + { + "left": "issue:10337", + "right": "issue:11137", + "accept": false, + "reason": "FLUX IPAdapter quantization failure vs LattePipeline dtype mismatch; separate models/code paths." + }, + { + "left": "issue:215", + "right": "issue:656", + "accept": false, + "reason": "Generic from_pretrained kill vs DreamBooth 16GB GPU error; too broad and not the same bug." + }, + { + "left": "issue:280", + "right": "issue:2905", + "accept": false, + "reason": "Add DEIS scheduler vs add Karras sigmas to schedulers; related area but distinct changes." + }, + { + "left": "issue:4116", + "right": "issue:8320", + "accept": false, + "reason": "SDXL img2img embeddings bug vs from_single_file AutoencoderKL __name__ attribute regression." + }, + { + "left": "issue:4263", + "right": "issue:6632", + "accept": false, + "reason": "LoRA unloading bug vs question about save_state in LoRA training scripts; different concerns." + }, + { + "left": "issue:5714", + "right": "issue:7441", + "accept": false, + "reason": "PeftModel support in maybe_raise_or_warn vs pytest failing on Python 3.12; unrelated." + }, + { + "left": "issue:6383", + "right": "issue:9515", + "accept": false, + "reason": "Multi-stage learning-rate scheduler support vs SD3 JointAttnProcessor RMSNorm omission." + }, + { + "left": "issue:8534", + "right": "issue:8535", + "accept": false, + "reason": "SD3 dreambooth logit-normal weighting bug vs SD3 xformers attention support; separate issues." + }, + { + "left": "issue:9031", + "right": "issue:9039", + "accept": false, + "reason": "Flax pipeline safety checker question vs Flax LoRA loading question; not the same underlying bug." + }, + { + "left": "issue:2973", + "right": "issue:5523", + "accept": false, + "reason": "enable_model_cpu_offload changes results vs inconsistent runs; too generic and not clearly the same defect." + }, + { + "left": "issue:10760", + "right": "issue:8958", + "accept": false, + "reason": "ONNX import attribute bug vs PyTorch 2.4 torch.library.register_fake incompatibility; unrelated." + }, + { + "left": "issue:5006", + "right": "issue:6956", + "accept": false, + "reason": "SDXL BF16 training OOM vs StableVideoDiffusion SageMaker OOM; same symptom class, different workloads/code paths." + }, + { + "left": "issue:10891", + "right": "issue:12653", + "accept": false, + "reason": "DCAE bfloat16 type conversion bug vs PRX rope float64 reference on MPS/npu; different models and failures." + }, + { + "left": "issue:4186", + "right": "issue:5064", + "accept": false, + "reason": "SDXL inpaint/refiner interaction vs refiner inference out-of-space; different symptoms and fixes." + }, + { + "left": "issue:10967", + "right": "issue:8641", + "accept": false, + "reason": "Integrating LaVie/Cinemo vs integrating Lumina-T2X; separate feature requests." + }, + { + "left": "issue:2581", + "right": "issue:2645", + "accept": false, + "reason": "ControlNet checkpoint conversion feature vs Mac M1 CUDA assertion error; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 323, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10030", + "issue:10121", + "issue:10584", + "issue:11225", + "issue:11574", + "issue:11601", + "issue:12052", + "issue:13281", + "issue:1746", + "issue:1933", + "issue:2668", + "issue:3116", + "issue:3209", + "issue:3406", + "issue:3441", + "issue:3960", + "issue:4075", + "issue:4104", + "issue:4249", + "issue:4483", + "issue:4701", + "issue:4801", + "issue:4911", + "issue:5081", + "issue:5224", + "issue:5313", + "issue:5365", + "issue:5703", + "issue:5981", + "issue:6091", + "issue:6131", + "issue:7365", + "issue:7418", + "issue:7460", + "issue:7636", + "issue:8309", + "issue:8649", + "issue:8839", + "issue:8898", + "issue:9052", + "issue:9092", + "issue:929", + "issue:9476", + "issue:9516", + "issue:9822", + "issue:984", + "issue:9873" + ], + "soft_pairs": [ + "issue:3116|issue:3441", + "issue:3209|issue:984", + "issue:6131|issue:9092", + "issue:11574|issue:8839", + "issue:11225|issue:9476", + "issue:4483|issue:5081", + "issue:8309|issue:9873", + "issue:2668|issue:4104", + "issue:3116|issue:5703", + "issue:4701|issue:6091", + "issue:3960|issue:5313", + "issue:12052|issue:13281", + "issue:11601|issue:8898", + "issue:1746|issue:5224", + "issue:1933|issue:4075", + "issue:3406|issue:929", + "issue:4249|issue:7365", + "issue:4801|issue:5981", + "issue:4911|issue:9052", + "issue:5365|issue:8649", + "issue:7418|issue:7460", + "issue:9516|issue:9822", + "issue:10030|issue:7636", + "issue:10121|issue:10584" + ], + "prepared_review_unit_hash": "7a032c4fce90cd33daf1666e87e83a69dae106d9aedcc2f222ba1eb1fec5b657", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33265, + "estimated_input_tokens": 8317, + "estimated_eval_tokens": 16890 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous cluster of mostly unrelated diffusers issues across ControlNet, Flux, schedulers, loaders, training examples, and documentation. None of the proposed soft pairs looks like the same underlying bug/change.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #10030 is the most concrete, code-path-specific bug report in the set: it names the failing pipeline and the missing ControlNet `added_cond_kwargs` contract clearly.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #10030 is the best representative issue because it is narrowly scoped, reproducible, and tied to a specific pipeline mismatch rather than a broad symptom or feature request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3116", + "right": "issue:3441", + "accept": false, + "reason": "Both involve diffusion internals, but one is a scheduler/noise bug and the other is a VRAM regression; different symptoms and code paths." + }, + { + "left": "issue:3209", + "right": "issue:984", + "accept": false, + "reason": "Both mention large inputs/tensor shape issues, but one is an AutoencoderKL NaN problem and the other is an interpolation/batch-size failure; not the same bug." + }, + { + "left": "issue:6131", + "right": "issue:9092", + "accept": false, + "reason": "IP-Adapter Plus not working and Flux model_index.json missing are separate loading/configuration issues." + }, + { + "left": "issue:11574", + "right": "issue:8839", + "accept": false, + "reason": "Both are weight-loading errors, but they affect different models and failure modes (missing VAE key vs SD3 LoRA weight mismatch)." + }, + { + "left": "issue:11225", + "right": "issue:9476", + "accept": false, + "reason": "Training OOM on validation and Flux validation dtype mismatch are unrelated training failures." + }, + { + "left": "issue:4483", + "right": "issue:5081", + "accept": false, + "reason": "Determinism of the generator and a matmul shape error are different problems in different parts of the pipeline." + }, + { + "left": "issue:8309", + "right": "issue:9873", + "accept": false, + "reason": "A feature request for conditional diffusion distillation is not the same change as adding an OmniGen pipeline." + }, + { + "left": "issue:2668", + "right": "issue:4104", + "accept": false, + "reason": "ControlNet image misrouted as Img2Img input and black-image casting warnings are different bugs with different mechanisms." + }, + { + "left": "issue:3116", + "right": "issue:5703", + "accept": false, + "reason": "Both are scheduler-related, but one is latent noise in img2img and the other is ddim_inverse reconstruction failure; different code paths." + }, + { + "left": "issue:4701", + "right": "issue:6091", + "accept": false, + "reason": "AutoencoderKL differentiation CUDA error and consistency-decoder memory usage are unrelated." + }, + { + "left": "issue:3960", + "right": "issue:5313", + "accept": false, + "reason": "A LoRA header-too-large loading error and non-revertible LoRA fuse are different LoRA issues." + }, + { + "left": "issue:12052", + "right": "issue:13281", + "accept": false, + "reason": "Both describe offloading/device mismatch, but they involve different models and different offloading implementations; too broad to treat as the same bug." + }, + { + "left": "issue:11601", + "right": "issue:8898", + "accept": false, + "reason": "Missing `use_safetensors` propagation and subfolder loading ValueError are separate loader bugs." + }, + { + "left": "issue:1746", + "right": "issue:5224", + "accept": false, + "reason": "Two Flax Stable Diffusion errors, but no shared failing behavior or code path is evident from the titles." + }, + { + "left": "issue:1933", + "right": "issue:4075", + "accept": false, + "reason": "Both are docstring-only issues, but they touch different classes and unrelated text changes." + }, + { + "left": "issue:3406", + "right": "issue:929", + "accept": false, + "reason": "Both mention attention, but one is a functional `only_cross_attention` bug and the other is a duplicate square-root calculation; not the same issue." + }, + { + "left": "issue:4249", + "right": "issue:7365", + "accept": false, + "reason": "Both concern pipeline input handling, but one is `prompt_embeds` validation in upscale and the other is pooled embeddings being overwritten; not close enough." + }, + { + "left": "issue:4801", + "right": "issue:5981", + "accept": false, + "reason": "Mixed-precision training example bug and DPMSolverMultistepScheduler bug are unrelated." + }, + { + "left": "issue:4911", + "right": "issue:9052", + "accept": false, + "reason": "Both involve `from_single_file`, but one is custom text encoder loading and the other is checkpoint compatibility with certain civit files; different underlying causes." + }, + { + "left": "issue:5365", + "right": "issue:8649", + "accept": false, + "reason": "Txt2video inference-step handling and SD3 `num_images_per_prompt` handling are unrelated pipeline parameter bugs." + }, + { + "left": "issue:7418", + "right": "issue:7460", + "accept": false, + "reason": "An example script bug and an AdaLayerNorm bug are clearly different." + }, + { + "left": "issue:9516", + "right": "issue:9822", + "accept": false, + "reason": "Missing `joint_attention_kwargs` forwarding into FLUX and loading SDXL LoRAs into Flux are different FLUX-related issues." + }, + { + "left": "issue:10030", + "right": "issue:7636", + "accept": false, + "reason": "Both mention ControlNet, but one is a concrete `added_cond_kwargs` pipeline bug and the other is a usage question about the SDXL tile model." + }, + { + "left": "issue:10121", + "right": "issue:10584", + "accept": false, + "reason": "High-resolution Flux generation failure and missing timestep indices in a scheduler helper are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 324, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1009", + "issue:10194", + "issue:11083", + "issue:11092", + "issue:11136", + "issue:11221", + "issue:11321", + "issue:11569", + "issue:11978", + "issue:12120", + "issue:12369", + "issue:12624", + "issue:1486", + "issue:2579", + "issue:2758", + "issue:2875", + "issue:3209", + "issue:3348", + "issue:3354", + "issue:3441", + "issue:3960", + "issue:4159", + "issue:4424", + "issue:4616", + "issue:4767", + "issue:4875", + "issue:4933", + "issue:4935", + "issue:5103", + "issue:5544", + "issue:5546", + "issue:5621", + "issue:5997", + "issue:6101", + "issue:6552", + "issue:6586", + "issue:6686", + "issue:6964", + "issue:7220", + "issue:7300", + "issue:8451", + "issue:8873", + "issue:9051", + "issue:9297", + "issue:9670", + "issue:9799", + "issue:9907" + ], + "soft_pairs": [ + "issue:5997|issue:9907", + "issue:2579|issue:3441", + "issue:3441|issue:4159", + "issue:6101|issue:6964", + "issue:10194|issue:12120", + "issue:11221|issue:9799", + "issue:4875|issue:9051", + "issue:6552|issue:6586", + "issue:11321|issue:8451", + "issue:11569|issue:12624", + "issue:2875|issue:3209", + "issue:4767|issue:5621", + "issue:5546|issue:6686", + "issue:9297|issue:9670", + "issue:11092|issue:11136", + "issue:1009|issue:1486", + "issue:3348|issue:3960", + "issue:4933|issue:7220", + "issue:11083|issue:8873", + "issue:11978|issue:12369", + "issue:2758|issue:5103", + "issue:3354|issue:4424", + "issue:4616|issue:5544", + "issue:4935|issue:7300" + ], + "prepared_review_unit_hash": "1cf5569ba6cc86016275ab76992dfbe706b164d8cfc26d426f62136daab57c40", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33417, + "estimated_input_tokens": 8355, + "estimated_eval_tokens": 16966 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose similarity bucket, not a true duplicate set: the items span unrelated training bugs, scheduler/timestep issues, memory/performance regressions, and feature requests across different pipelines and models. None of the soft pairs look like the same underlying bug/change.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue fits the cluster; the only commonality is superficial topic overlap in titles, not a shared code-path or bug.", + "canonical_pr_reason": null, + "best_issue_reason": "No globally best issue for deduplication. The closest tracker-style issues (for example 6586) are still too narrow and do not represent the rest of the heterogeneous set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5997", + "right": "issue:9907", + "accept": false, + "reason": "Different problems: a ControlNet SDXL tutorial type-mismatch versus a Flux controlnet sequence-length warning." + }, + { + "left": "issue:2579", + "right": "issue:3441", + "accept": false, + "reason": "Both mention xFormers/PyTorch performance, but one is a misuse/bug report and the other is a VRAM regression under PyTorch 2." + }, + { + "left": "issue:3441", + "right": "issue:4159", + "accept": false, + "reason": "Not the same bug: higher VRAM with PyTorch 2 is distinct from DiffEdit\u2019s extreme VRAM requirement." + }, + { + "left": "issue:6101", + "right": "issue:6964", + "accept": false, + "reason": "Different code paths: ControlNet inpaint performance versus incompatibility with 4-channel UNets." + }, + { + "left": "issue:10194", + "right": "issue:12120", + "accept": false, + "reason": "A feature request for Flux attention masking is unrelated to a question about LoRA training on distilled Flux models." + }, + { + "left": "issue:11221", + "right": "issue:9799", + "accept": false, + "reason": "Training non-convergence for AutoencoderKL is unrelated to a tensor shape mismatch runtime error." + }, + { + "left": "issue:4875", + "right": "issue:9051", + "accept": false, + "reason": "Scheduler crash when batch size changes is not the same as lack of batch inference support in FluxPipeline." + }, + { + "left": "issue:6552", + "right": "issue:6586", + "accept": false, + "reason": "Both are trackers, but they cover different training fixes: FP16 resume issues vs SDXL micro-conditioning." + }, + { + "left": "issue:11321", + "right": "issue:8451", + "accept": false, + "reason": "A README bug for Flux ControlNet training is unrelated to documentation about frozen UNet parameters / missing SD1.x training script." + }, + { + "left": "issue:11569", + "right": "issue:12624", + "accept": false, + "reason": "A repo-request feature asks for a model repository; the other requests adding a strength parameter to a video pipeline." + }, + { + "left": "issue:2875", + "right": "issue:3209", + "accept": false, + "reason": "NaN training loss and AutoencoderKL encoder NaNs are different failures with different triggers." + }, + { + "left": "issue:4767", + "right": "issue:5621", + "accept": false, + "reason": "Scheduler prediction_type support is unrelated to SDXL ControlNet encoder_hid_proj support." + }, + { + "left": "issue:5546", + "right": "issue:6686", + "accept": false, + "reason": "Wrong step counts in schedulers is a different problem from Dynamo compilation performance not persisting after save/load." + }, + { + "left": "issue:9297", + "right": "issue:9670", + "accept": false, + "reason": "Linear LoRA load-time slowdown is unrelated to the PEFT-backend-required load failure." + }, + { + "left": "issue:11092", + "right": "issue:11136", + "accept": false, + "reason": "Both involve timesteps, but they affect different pipelines/models and different failure modes." + }, + { + "left": "issue:1009", + "right": "issue:1486", + "accept": false, + "reason": "Textual inversion weight-decay bug is unrelated to a DDIMScheduler prediction_type='sample' bug." + }, + { + "left": "issue:3348", + "right": "issue:3960", + "accept": false, + "reason": "Support for xFormers scale in LoRA models is not the same as a HeaderTooLarge error." + }, + { + "left": "issue:4933", + "right": "issue:7220", + "accept": false, + "reason": "Dreambooth missing pytorch_model.bin after training is unrelated to unload_ip_adapter() failing in ControlNet inpaint." + }, + { + "left": "issue:11083", + "right": "issue:8873", + "accept": false, + "reason": "DDP stride-mismatch warning and xFormers being slower on A100 are different performance/debug issues." + }, + { + "left": "issue:11978", + "right": "issue:12369", + "accept": false, + "reason": "These are separate feature requests for different video/image models and pipelines." + }, + { + "left": "issue:2758", + "right": "issue:5103", + "accept": false, + "reason": "Both mention black images, but one is ROCm torch.compile warning behavior and the other is zero-SNR DDIM sampling." + }, + { + "left": "issue:3354", + "right": "issue:4424", + "accept": false, + "reason": "Unet being cast to fp16 in a LoRA training script is distinct from unconditional VAE upcasting in SDXL Instruct Pix2Pix." + }, + { + "left": "issue:4616", + "right": "issue:5544", + "accept": false, + "reason": "SDXL LoRA conversion failing is unrelated to disabling FreeU when it is not enabled." + }, + { + "left": "issue:4935", + "right": "issue:7300", + "accept": false, + "reason": "A question about AttnProcsLayers in an SDXL LoRA script is unrelated to high GPU memory after set_default_attn_processor()." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 325, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10039", + "issue:1009", + "issue:10112", + "issue:10472", + "issue:11169", + "issue:11497", + "issue:12358", + "issue:1262", + "issue:1314", + "issue:13176", + "issue:13301", + "issue:1537", + "issue:1540", + "issue:1896", + "issue:1918", + "issue:2306", + "issue:2579", + "issue:3282", + "issue:3405", + "issue:3643", + "issue:3660", + "issue:4593", + "issue:4916", + "issue:4923", + "issue:5016", + "issue:5017", + "issue:5271", + "issue:5474", + "issue:5515", + "issue:5593", + "issue:5678", + "issue:5710", + "issue:579", + "issue:5897", + "issue:6366", + "issue:6412", + "issue:6700", + "issue:6742", + "issue:7859", + "issue:8390", + "issue:8488", + "issue:8644", + "issue:8803", + "issue:9007", + "issue:9186", + "issue:9216", + "issue:941", + "issue:964" + ], + "soft_pairs": [ + "issue:5017|issue:8390", + "issue:5474|issue:9216", + "issue:5593|issue:6700", + "issue:5710|issue:9007", + "issue:2306|issue:3643", + "issue:10472|issue:11497", + "issue:11169|issue:13176", + "issue:12358|issue:13301", + "issue:8488|issue:8644", + "issue:5515|issue:6742", + "issue:1314|issue:4916", + "issue:1537|issue:941", + "issue:1540|issue:579", + "issue:1918|issue:964", + "issue:6412|issue:9186", + "issue:5016|issue:6366", + "issue:4923|issue:5271", + "issue:1262|issue:1896", + "issue:2579|issue:3660", + "issue:4593|issue:5678", + "issue:3282|issue:5897", + "issue:10039|issue:8803", + "issue:1009|issue:3405", + "issue:10112|issue:7859" + ], + "prepared_review_unit_hash": "3d108e0afe08035696f4506efef07f0b7412a82c9ca3a191113b2a0206b7eb08", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34356, + "estimated_input_tokens": 8589, + "estimated_eval_tokens": 17434 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft links mostly connect unrelated diffusers issues across different subsystems (pipelines, schedulers, training scripts, environment/import errors, and feature requests). I do not see any pair that clearly describes the same underlying bug/change, so all soft edges should be rejected.", + "confidence": 0.91, + "canonical_issue_reason": "Issue:10039 is the sharpest concrete code bug in the set (a specific UNet attention call in unet_2d_blocks.py), so it is the best available anchor, though the items are not actually one duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue:10039 is the most concrete and code-path-specific item here; the others are mostly broader feature requests or unrelated runtime/environment failures.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5017", + "right": "issue:8390", + "accept": false, + "reason": "ControlNet control mode request vs Stable Cascade ControlNet training bug; different problem and scope." + }, + { + "left": "issue:5474", + "right": "issue:9216", + "accept": false, + "reason": "Two different pipeline feature requests for different models; not the same change." + }, + { + "left": "issue:5593", + "right": "issue:6700", + "accept": false, + "reason": "AutoPipeline conversion failure and attention temb TypeError are different code paths." + }, + { + "left": "issue:5710", + "right": "issue:9007", + "accept": false, + "reason": "Scheduler initialization bug vs black output with order=3; distinct scheduler defects." + }, + { + "left": "issue:2306", + "right": "issue:3643", + "accept": false, + "reason": "Both mention INT_MAX, but one is a generic size error and the other is a specific latent decoding limit; not clearly the same bug." + }, + { + "left": "issue:10472", + "right": "issue:11497", + "accept": false, + "reason": "LoRA bias constructor error vs preprocessing cache dependence on unrelated args; unrelated bugs." + }, + { + "left": "issue:11169", + "right": "issue:13176", + "accept": false, + "reason": "Binary incompatibility on import vs UNet attention error; different failure modes." + }, + { + "left": "issue:12358", + "right": "issue:13301", + "accept": false, + "reason": "Separate pipeline support requests for different models; not duplicates." + }, + { + "left": "issue:8488", + "right": "issue:8644", + "accept": false, + "reason": "Both involve SD3 loading, but one is a generic from_pretrained exception and the other is a specific offload/meta-tensor bug." + }, + { + "left": "issue:5515", + "right": "issue:6742", + "accept": false, + "reason": "LCM pipeline without scheduler and DownBlockMotion checkpointing behavior are unrelated." + }, + { + "left": "issue:1314", + "right": "issue:4916", + "accept": false, + "reason": "Windows xformers backward failure vs checkpointing/ZeRO3 training failure; different interactions." + }, + { + "left": "issue:1537", + "right": "issue:941", + "accept": false, + "reason": "MPS scheduler issue vs img2img noise with num_images_per_prompt; separate bugs." + }, + { + "left": "issue:1540", + "right": "issue:579", + "accept": false, + "reason": "Missing MODEL_TYPE in Dreambooth/transformers vs missing config.json in textual inversion; different config-loading issues." + }, + { + "left": "issue:1918", + "right": "issue:964", + "accept": false, + "reason": "Different example failures (DDIM scheduler vs inpainting example); not the same underlying defect." + }, + { + "left": "issue:6412", + "right": "issue:9186", + "accept": false, + "reason": "ControlNet annotator ONNX request vs differential diffusion with ControlNet pipeline; different feature areas." + }, + { + "left": "issue:5016", + "right": "issue:6366", + "accept": false, + "reason": "SDXL LoRA text encoder training bug vs checkpoint restore bug; same script family but different concrete failures." + }, + { + "left": "issue:4923", + "right": "issue:5271", + "accept": false, + "reason": "AudioLDM2 transcript feature request vs dreambooth learning-rate feature request; unrelated." + }, + { + "left": "issue:1262", + "right": "issue:1896", + "accept": false, + "reason": "Rectangle-image training question vs CLIP-guided Stable Diffusion size/type error; unrelated." + }, + { + "left": "issue:2579", + "right": "issue:3660", + "accept": false, + "reason": "Ambiguous xFormers misuse/bug report vs torch.compile incompatibility with LoRA/xformers; not the same bug." + }, + { + "left": "issue:4593", + "right": "issue:5678", + "accept": false, + "reason": "LoRA rank validation error vs higher-order-derivative attention bug; different code paths." + }, + { + "left": "issue:3282", + "right": "issue:5897", + "accept": false, + "reason": "xFormers operator dispatch failure vs device mismatch in LoRA training; unrelated runtime errors." + }, + { + "left": "issue:10039", + "right": "issue:8803", + "accept": false, + "reason": "Both touch attention internals, but one is a missing output sample size in AttnUpBlock2D and the other is a duplicate hidden_states assignment; different bugs." + }, + { + "left": "issue:1009", + "right": "issue:3405", + "accept": false, + "reason": "Textual inversion weight decay bug vs ControlNet training bug; unrelated." + }, + { + "left": "issue:10112", + "right": "issue:7859", + "accept": false, + "reason": "Diffuse 'Detail-Daemon' issue vs FlashFace implementation request; different topics." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 326, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10266", + "issue:10337", + "issue:10381", + "issue:10591", + "issue:10782", + "issue:10901", + "issue:1101", + "issue:1117", + "issue:11653", + "issue:11790", + "issue:12012", + "issue:12083", + "issue:12610", + "issue:1746", + "issue:1994", + "issue:201", + "issue:2192", + "issue:2412", + "issue:2785", + "issue:2907", + "issue:3045", + "issue:3706", + "issue:3761", + "issue:4194", + "issue:4468", + "issue:4616", + "issue:5079", + "issue:5135", + "issue:5441", + "issue:5489", + "issue:5538", + "issue:5777", + "issue:5989", + "issue:6176", + "issue:6230", + "issue:6250", + "issue:6441", + "issue:6686", + "issue:7157", + "issue:7525", + "issue:7574", + "issue:8567", + "issue:8579", + "issue:8588", + "issue:8604", + "issue:9178", + "issue:9810" + ], + "soft_pairs": [ + "issue:10266|issue:10337", + "issue:10381|issue:8604", + "issue:10591|issue:11790", + "issue:10591|issue:9810", + "issue:10782|issue:12083", + "issue:10901|issue:9178", + "issue:1101|issue:2785", + "issue:1117|issue:201", + "issue:1994|issue:5079", + "issue:2192|issue:6230", + "issue:2907|issue:3761", + "issue:3045|issue:5989", + "issue:3706|issue:7525", + "issue:4194|issue:6441", + "issue:4468|issue:6686", + "issue:5135|issue:6176", + "issue:5489|issue:7157", + "issue:5538|issue:8588", + "issue:5777|issue:8567", + "issue:11653|issue:8579", + "issue:12012|issue:12610", + "issue:1746|issue:2412", + "issue:4616|issue:5441", + "issue:6250|issue:7574" + ], + "prepared_review_unit_hash": "7fad6dd7b23982ca8f04ee448c5b955a00db496e3ae7c36cda04533472a66136", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32858, + "estimated_input_tokens": 8215, + "estimated_eval_tokens": 16686 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All candidate pairs look like distinct issues or feature requests across different models, scripts, and failure modes. The shared keywords are too broad to treat any pair as the same underlying bug/change, so every soft edge is rejected.", + "confidence": 0.96, + "canonical_issue_reason": "No clear canonical issue: the items span unrelated topics and none of the pairings convincingly collapse to one bug or change.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a safe duplicate hub for this set; picking one would overfit to superficial keyword overlap.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10266", + "right": "issue:10337", + "accept": false, + "reason": "Different pipelines and failures: UniPC/FlowMatch index error vs FLUX IPAdapter quantization failure." + }, + { + "left": "issue:10381", + "right": "issue:8604", + "accept": false, + "reason": "Both mention Flux and quantization, but one is a Flux LoRA/Quanto inference issue and the other is a T5 float16 corruption bug." + }, + { + "left": "issue:10591", + "right": "issue:11790", + "accept": false, + "reason": "Both are LoRA training-script complaints, but they target different scripts and different failure points." + }, + { + "left": "issue:10591", + "right": "issue:9810", + "accept": false, + "reason": "SD3 training-script bug vs a general LoRA module_to_save limitation; not the same issue." + }, + { + "left": "issue:10782", + "right": "issue:12083", + "accept": false, + "reason": "Different models and symptoms: Lumina Image 2.0 minor issue vs Qwen-Image long-prompt error." + }, + { + "left": "issue:10901", + "right": "issue:9178", + "accept": false, + "reason": "HunyuanVideo negative_prompt behavior and FLUX LoRA PEFT target-module errors are unrelated code paths." + }, + { + "left": "issue:1101", + "right": "issue:2785", + "accept": false, + "reason": "Add a scheduler vs add M1/M2 support; separate feature requests." + }, + { + "left": "issue:1117", + "right": "issue:201", + "accept": false, + "reason": "Both are test-related, but they cover different behaviors and do not describe the same test failure/change." + }, + { + "left": "issue:1994", + "right": "issue:5079", + "accept": false, + "reason": "AWS/from_pretrained storage support is unrelated to loading LyCORIS weights." + }, + { + "left": "issue:2192", + "right": "issue:6230", + "accept": false, + "reason": "Missing CLI arguments in one training script vs CUDA OOM in another; distinct problems." + }, + { + "left": "issue:2907", + "right": "issue:3761", + "accept": false, + "reason": "Repeated CPU offload and loading multiple community pipelines are different lifecycle bugs." + }, + { + "left": "issue:3045", + "right": "issue:5989", + "accept": false, + "reason": "LoRA conversion utility issue vs a safetensors.torch bug; no concrete overlap." + }, + { + "left": "issue:3706", + "right": "issue:7525", + "accept": false, + "reason": "A new model announcement and a discrete scheduler issue are not the same change." + }, + { + "left": "issue:4194", + "right": "issue:6441", + "accept": false, + "reason": "Single-file VAE loading failure vs Google Colab from_pretrained failure; different contexts and symptoms." + }, + { + "left": "issue:4468", + "right": "issue:6686", + "accept": false, + "reason": "Multi-GPU performance discrepancy and Dynamo save/load performance regression are different issues." + }, + { + "left": "issue:5135", + "right": "issue:6176", + "accept": false, + "reason": "LoRA quality degradation with xformers is unrelated to loading LoRA weights after training." + }, + { + "left": "issue:5489", + "right": "issue:7157", + "accept": false, + "reason": "Broad PEFT integration work vs unexpected keys when loading LCM LoRA; not the same concrete bug." + }, + { + "left": "issue:5538", + "right": "issue:8588", + "accept": false, + "reason": "Up-scaler latent usage bug vs missing SD3 conversion script; different files and failures." + }, + { + "left": "issue:5777", + "right": "issue:8567", + "accept": false, + "reason": "PixArt dreambooth training error vs SD3 training-objective question; not the same underlying problem." + }, + { + "left": "issue:11653", + "right": "issue:8579", + "accept": false, + "reason": "Both involve LoRA support, but one is about .alpha keys in HiDream LoRAs and the other is SD3 LoRA functionality." + }, + { + "left": "issue:12012", + "right": "issue:12610", + "accept": false, + "reason": "Context-state error with Wan 2.2 caching vs Quanto/group-offload device mismatch; different mechanisms." + }, + { + "left": "issue:1746", + "right": "issue:2412", + "accept": false, + "reason": "Flax pipeline error and a generic dreambooth error are too broad to be the same bug." + }, + { + "left": "issue:4616", + "right": "issue:5441", + "accept": false, + "reason": "SDXL LoRA conversion failure vs WuerstchenPrior missing attn_processors; different models and scripts." + }, + { + "left": "issue:6250", + "right": "issue:7574", + "accept": false, + "reason": "Different KeyErrors in different loading paths; no evidence they share the same root cause." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 327, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10428", + "issue:10460", + "issue:11732", + "issue:11872", + "issue:12338", + "issue:13137", + "issue:13386", + "issue:1339", + "issue:13488", + "issue:2163", + "issue:2370", + "issue:2392", + "issue:2410", + "issue:2581", + "issue:2610", + "issue:2988", + "issue:3040", + "issue:3149", + "issue:3194", + "issue:3250", + "issue:3450", + "issue:347", + "issue:3958", + "issue:4101", + "issue:4232", + "issue:4279", + "issue:4376", + "issue:444", + "issue:4480", + "issue:4488", + "issue:4664", + "issue:4753", + "issue:4756", + "issue:4837", + "issue:4904", + "issue:5178", + "issue:5339", + "issue:5365", + "issue:5957", + "issue:6457", + "issue:6870", + "issue:8296", + "issue:8429", + "issue:8494", + "issue:8509", + "issue:8752", + "issue:9163", + "issue:9613" + ], + "soft_pairs": [ + "issue:2392|issue:3040", + "issue:11872|issue:13488", + "issue:2610|issue:4480", + "issue:2988|issue:444", + "issue:10460|issue:8494", + "issue:3250|issue:5178", + "issue:3450|issue:347", + "issue:5339|issue:8752", + "issue:6457|issue:8296", + "issue:10428|issue:8429", + "issue:3958|issue:4279", + "issue:3194|issue:6870", + "issue:4753|issue:9163", + "issue:2581|issue:4488", + "issue:4101|issue:4837", + "issue:12338|issue:13386", + "issue:2163|issue:2410", + "issue:4376|issue:4664", + "issue:5365|issue:9613", + "issue:11732|issue:13137", + "issue:1339|issue:3149", + "issue:2370|issue:4756", + "issue:4232|issue:8509", + "issue:4904|issue:5957" + ], + "prepared_review_unit_hash": "b73dab6ce2f0d2000536b83d611b2944cc00c108ddd680f9984afb8eb0984f05", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33972, + "estimated_input_tokens": 8493, + "estimated_eval_tokens": 17242 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly heterogeneous and the soft edges are largely false positives from superficial title similarity. The only likely true duplicate pair is the seed-determinism reports (#1339 and #3149).", + "confidence": 0.89, + "canonical_issue_reason": "No single canonical issue fits the full cluster: it mixes unrelated bugs across flux, LoRA loading, offloading, training scripts, conversion tools, and scheduler behavior. If a subcluster representative were needed, #1339 is the broadest determinism report.", + "canonical_pr_reason": null, + "best_issue_reason": "No globally best issue for the whole cluster; the items are too unrelated for one issue to represent them well. #1339 is only the best representative for the determinism subcluster, not the cluster overall.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2392", + "right": "issue:3040", + "accept": false, + "reason": "Both involve prompt-related errors, but one is an img2img prompt_embeds ValueError and the other is a docs-copy/paste AttributeError; different bugs." + }, + { + "left": "issue:11872", + "right": "issue:13488", + "accept": false, + "reason": "Both mention offloading, but one is a loop inefficiency from repeated CPU/GPU transfers and the other is a regression breaking offloading in a specific dev version." + }, + { + "left": "issue:2610", + "right": "issue:4480", + "accept": false, + "reason": "Mac-related failures, but one is a local setup/clang problem and the other is an MPS float16 kernel crash during generation." + }, + { + "left": "issue:2988", + "right": "issue:444", + "accept": false, + "reason": "Both are scheduler/step-count problems, but one crashes when training and inference steps are equal while the other is an index error for specific step values." + }, + { + "left": "issue:10460", + "right": "issue:8494", + "accept": false, + "reason": "Different models and failures: a missing documented UNet argument versus an unexpected keyword argument in SD3 offloading/hooks." + }, + { + "left": "issue:3250", + "right": "issue:5178", + "accept": false, + "reason": "Both are pipeline/runtime issues, but one is about StableDiffusionKDiffusionPipeline not running and the other is a DPM-SDE resolution-change failure." + }, + { + "left": "issue:3450", + "right": "issue:347", + "accept": false, + "reason": "One is a from_ckpt runtime regression; the other is an installation dependency resolution failure. Not the same bug." + }, + { + "left": "issue:5339", + "right": "issue:8752", + "accept": false, + "reason": "Both are loading problems, but one is a local safetensor loading complaint and the other is an offline from_single_file/SDXL resolution issue." + }, + { + "left": "issue:6457", + "right": "issue:8296", + "accept": false, + "reason": "Both touch training bookkeeping, but one is gradient accumulation on multi-GPU SDXL-LoRA and the other is checkpoint global-step confusion across scripts." + }, + { + "left": "issue:10428", + "right": "issue:8429", + "accept": false, + "reason": "Different domains and code paths: Ascend NPU Flux inference versus an I2VGen-XL inference-script malfunction." + }, + { + "left": "issue:3958", + "right": "issue:4279", + "accept": false, + "reason": "Both mention CivitAI LoRAs, but one is tied to enable_model_cpu_offload and the other to SDXL 0.9 loading failure; not clearly the same concrete bug." + }, + { + "left": "issue:3194", + "right": "issue:6870", + "accept": false, + "reason": "Completely different failures: bitsandbytes package metadata missing versus enable_vae_tiling missing in specific pipelines." + }, + { + "left": "issue:4753", + "right": "issue:9163", + "accept": false, + "reason": "Both are training-script bugs, but they affect different scripts and different tensor issues; not the same underlying defect." + }, + { + "left": "issue:2581", + "right": "issue:4488", + "accept": false, + "reason": "One is about converting ControlNet checkpoints to diffusers; the other is SDXL checkpoint loading from single file in offline mode." + }, + { + "left": "issue:4101", + "right": "issue:4837", + "accept": false, + "reason": "Both involve conversion to diffusers, but one is a custom ControlNet conversion failure and the other is an SDXL conversion failure; too different to merge." + }, + { + "left": "issue:12338", + "right": "issue:13386", + "accept": false, + "reason": "Different subsystems and symptoms: VAE slicing/AutoencoderDC decoding versus a Qwen image Dreambooth prior-preservation concatenation crash." + }, + { + "left": "issue:2163", + "right": "issue:2410", + "accept": false, + "reason": "One is a dtype mismatch in a training example; the other is missing output artifacts in a Flax training script. Not the same issue." + }, + { + "left": "issue:4376", + "right": "issue:4664", + "accept": false, + "reason": "Both concern SDXL loading, but one is textual inversion crashing and the other is LoRA loading on a differently named pipeline class." + }, + { + "left": "issue:5365", + "right": "issue:9613", + "accept": false, + "reason": "Both are inference-time errors, but one is a num_inference_steps issue in txt2video and the other is a missing scale attribute on a custom Timesteps object." + }, + { + "left": "issue:11732", + "right": "issue:13137", + "accept": false, + "reason": "Both mention load_lora_weights, but one is a generic unpacking ValueError and the other is a Flux 2 Klein LoKr-specific loading error." + }, + { + "left": "issue:1339", + "right": "issue:3149", + "accept": true, + "reason": "Same underlying symptom: nondeterministic outputs despite the same seed. #3149 is a version-specific restatement of #1339's determinism problem." + }, + { + "left": "issue:2370", + "right": "issue:4756", + "accept": false, + "reason": "Different topics: hardware feasibility for LoRA training versus a request for an SDXL dreambooth-lora example script." + }, + { + "left": "issue:4232", + "right": "issue:8509", + "accept": false, + "reason": "Both are feature requests around control conditions, but they concern different models and capabilities, not the same change." + }, + { + "left": "issue:4904", + "right": "issue:5957", + "accept": false, + "reason": "Unrelated changes: modularizing GLIGEN versus fixing LoRACompatibleConv padding_mode behavior." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: the cluster is clearly heterogeneous, and the only soft edge with direct support from the titles is #1339/#3149. The other pair verdicts stay conservative and avoid overclaiming duplicate status from superficial title similarity. No mergeability issues since there are no PR pairs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 328, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10121", + "issue:10518", + "issue:10526", + "issue:10594", + "issue:10958", + "issue:11083", + "issue:11104", + "issue:11160", + "issue:11222", + "issue:11886", + "issue:13286", + "issue:1434", + "issue:1534", + "issue:1556", + "issue:1643", + "issue:1811", + "issue:2035", + "issue:2459", + "issue:2495", + "issue:2534", + "issue:2800", + "issue:4043", + "issue:4149", + "issue:4307", + "issue:460", + "issue:4668", + "issue:4692", + "issue:490", + "issue:4911", + "issue:5009", + "issue:5628", + "issue:5984", + "issue:6154", + "issue:6297", + "issue:6570", + "issue:6732", + "issue:6880", + "issue:696", + "issue:7066", + "issue:7240", + "issue:7333", + "issue:7616", + "issue:8264", + "issue:8609", + "issue:8880", + "issue:9611", + "issue:9756", + "issue:9858" + ], + "soft_pairs": [ + "issue:5628|issue:7240", + "issue:2035|issue:490", + "issue:2534|issue:5009", + "issue:4043|issue:460", + "issue:4668|issue:7616", + "issue:10526|issue:10958", + "issue:10518|issue:11083", + "issue:1434|issue:4692", + "issue:1534|issue:696", + "issue:1643|issue:2459", + "issue:6732|issue:9756", + "issue:11160|issue:11222", + "issue:6154|issue:9858", + "issue:10594|issue:11886", + "issue:1556|issue:2495", + "issue:4911|issue:6880", + "issue:7066|issue:8609", + "issue:10121|issue:9611", + "issue:2800|issue:7333", + "issue:4149|issue:6570", + "issue:5984|issue:6297", + "issue:11104|issue:13286", + "issue:8264|issue:8880", + "issue:1811|issue:4307" + ], + "prepared_review_unit_hash": "912c17f86674bc2552a5bfa5450ed278fd7d061316dacd9379815c372c1ed6e6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34079, + "estimated_input_tokens": 8520, + "estimated_eval_tokens": 17296 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose grab-bag of unrelated Diffusers bugs/features (Flux, LoRA training, schedulers, offload, quantization, pipeline shape mismatches). None of the soft candidate pairs look like the same underlying bug or change, so all should be kept separate.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 9611 is the most concrete, self-contained regression report in the set: it names the exact Flux pipelines, the batching failure, and a version. The rest of the items are too disparate to serve as a better cluster anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9611 is the strongest standalone issue to retain for follow-up because it is specific, reproducible, and actionable, with a clear pipeline scope and failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5628", + "right": "issue:7240", + "accept": false, + "reason": "Different problems: scheduler denoising offset vs low-quality ImageNet LDM conversion output." + }, + { + "left": "issue:2035", + "right": "issue:490", + "accept": false, + "reason": "Both mention upscaling, but one is artifact quality and the other is an OOM/memory blowout; not the same bug." + }, + { + "left": "issue:2534", + "right": "issue:5009", + "accept": false, + "reason": "Scheduler-switching error and Wuerstchen error are unrelated failures in different code paths." + }, + { + "left": "issue:4043", + "right": "issue:460", + "accept": false, + "reason": "SDXL prompt-embed tensor requirements and ONNX Runtime dependency selection are unrelated issues." + }, + { + "left": "issue:4668", + "right": "issue:7616", + "accept": false, + "reason": "Both are image-artifact reports, but they involve different models, pipelines, and trigger conditions." + }, + { + "left": "issue:10526", + "right": "issue:10958", + "accept": false, + "reason": "Flux FP8 quantization TypeError and CogView4 meta-tensor copy failure are distinct bugs in different subsystems." + }, + { + "left": "issue:10518", + "right": "issue:11083", + "accept": false, + "reason": "Training-notebook bug vs DDP stride-mismatch warning; no shared concrete code path." + }, + { + "left": "issue:1434", + "right": "issue:4692", + "accept": false, + "reason": "Both are memory-related, but one is StableDiffusionUpscalePipeline OOM and the other is a general UNet memory concern." + }, + { + "left": "issue:1534", + "right": "issue:696", + "accept": false, + "reason": "Dreambooth notebook error and training OOM are different failure modes, not the same underlying bug." + }, + { + "left": "issue:1643", + "right": "issue:2459", + "accept": false, + "reason": "Both reference xformers attention, but the reports are for different scripts and are not clearly the same concrete defect." + }, + { + "left": "issue:6732", + "right": "issue:9756", + "accept": false, + "reason": "A scheduler documentation issue and a missing notebook file are unrelated." + }, + { + "left": "issue:11160", + "right": "issue:11222", + "accept": false, + "reason": "WAN inference inconsistency and WAN LoRA support are different issues." + }, + { + "left": "issue:6154", + "right": "issue:9858", + "accept": false, + "reason": "SDXL LoRA training on M2 Mac and a KeyError:'train' report do not describe the same bug." + }, + { + "left": "issue:10594", + "right": "issue:11886", + "accept": false, + "reason": "Both are Flux-related, but one is multi-control + LoRA behavior and the other is a FluxKontextPipeline issue." + }, + { + "left": "issue:1556", + "right": "issue:2495", + "accept": false, + "reason": "Black-image reports on CUDA vs MPS and different pipelines are not the same underlying failure." + }, + { + "left": "issue:4911", + "right": "issue:6880", + "accept": false, + "reason": "Custom text-encoder loading and Custom Diffusion retrieve.py are separate features/issues." + }, + { + "left": "issue:7066", + "right": "issue:8609", + "accept": false, + "reason": "Multi-image img2img support and a custom pipeline/IP-Adapter integration issue are unrelated." + }, + { + "left": "issue:10121", + "right": "issue:9611", + "accept": false, + "reason": "Both are Flux issues, but one is high-resolution generation failure and the other is image batching in img2img/inpaint; different code paths." + }, + { + "left": "issue:2800", + "right": "issue:7333", + "accept": false, + "reason": "DiffEdit semantic editing and a tuning-free noise-rectification I2V topic are not the same bug/change." + }, + { + "left": "issue:4149", + "right": "issue:6570", + "accept": false, + "reason": "Both mention tensor shape mismatches, but they affect different pipelines and conditioning flows." + }, + { + "left": "issue:5984", + "right": "issue:6297", + "accept": false, + "reason": "Both touch train_dreambooth_lora_sdxl, but one is a RAM issue and the other is a training-quality/problem report." + }, + { + "left": "issue:11104", + "right": "issue:13286", + "accept": false, + "reason": "LTX-Video offload breakage and torchao quantization support are unrelated." + }, + { + "left": "issue:8264", + "right": "issue:8880", + "accept": false, + "reason": "Training stuck at 0% and inference speed regression are different symptoms in different workflows." + }, + { + "left": "issue:1811", + "right": "issue:4307", + "accept": false, + "reason": "Weight-download behavior and local_files_only error are separate from_pretrained issues with different causes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 329, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10121", + "issue:1037", + "issue:10540", + "issue:10695", + "issue:10697", + "issue:11041", + "issue:1140", + "issue:1231", + "issue:12609", + "issue:13311", + "issue:1367", + "issue:1403", + "issue:1486", + "issue:3089", + "issue:3431", + "issue:3573", + "issue:3593", + "issue:3730", + "issue:4104", + "issue:4484", + "issue:4668", + "issue:4975", + "issue:5125", + "issue:5126", + "issue:5135", + "issue:5153", + "issue:5225", + "issue:5412", + "issue:5537", + "issue:5703", + "issue:6366", + "issue:7061", + "issue:7165", + "issue:7440", + "issue:7575", + "issue:8323", + "issue:8450", + "issue:8503", + "issue:8898", + "issue:8960", + "issue:9101", + "issue:9178", + "issue:9180", + "issue:9225", + "issue:9241", + "issue:938", + "issue:9539", + "issue:9677" + ], + "soft_pairs": [ + "issue:3730|issue:7165", + "issue:5153|issue:938", + "issue:5412|issue:6366", + "issue:5537|issue:9180", + "issue:5703|issue:8323", + "issue:9101|issue:9241", + "issue:5125|issue:7575", + "issue:3431|issue:4484", + "issue:4104|issue:8450", + "issue:3573|issue:3593", + "issue:4975|issue:5135", + "issue:8503|issue:8960", + "issue:1231|issue:1367", + "issue:12609|issue:13311", + "issue:5126|issue:5225", + "issue:10540|issue:8898", + "issue:11041|issue:9539", + "issue:9178|issue:9677", + "issue:10121|issue:9225", + "issue:1037|issue:1140", + "issue:10695|issue:7061", + "issue:10697|issue:7440", + "issue:1403|issue:4668", + "issue:1486|issue:3089" + ], + "prepared_review_unit_hash": "6551cf81b36b2e3c2db7e73d8da0241b171015c1c8212577d31f7c63356de059", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33824, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Only one soft edge is a clear same-bug match: the SDXL DreamBooth LoRA checkpoint-resume/loading reports (5412 \u2194 6366). The rest are better treated as distinct issues; several are in the same general area (schedulers, LoRA/training, Flux/WAN) but do not look like the same underlying bug or change.", + "confidence": 0.84, + "canonical_issue_reason": "issue:6366 is the best canonical issue because it describes the broader checkpoint-loading/resume failure across both SDXL DreamBooth LoRA training scripts, making it the most representative root report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6366 is the strongest global issue candidate since it is broader and more actionable than the narrower resume-from-checkpoint report while covering the shared failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3730", + "right": "issue:7165", + "accept": false, + "reason": "Both are scheduler-related, but they describe different bugs in different implementations and do not look like the same fix." + }, + { + "left": "issue:5153", + "right": "issue:938", + "accept": false, + "reason": "Different topics entirely: a U-Net optimization idea versus a community text-to-motion pipeline request." + }, + { + "left": "issue:5412", + "right": "issue:6366", + "accept": true, + "reason": "These appear to report the same checkpoint-resume/loading problem in SDXL DreamBooth LoRA training, with 6366 being the broader description." + }, + { + "left": "issue:5537", + "right": "issue:9180", + "accept": false, + "reason": "Different code paths and symptoms: a notebook not working versus a pipeline attribute error." + }, + { + "left": "issue:5703", + "right": "issue:8323", + "accept": false, + "reason": "Both are scheduler-adjacent, but the reported problems are distinct and not the same bug." + }, + { + "left": "issue:9101", + "right": "issue:9241", + "accept": false, + "reason": "Different functionality: an InstructPix2Pix training script versus a LoRA conversion script." + }, + { + "left": "issue:5125", + "right": "issue:7575", + "accept": false, + "reason": "Distinct failures in different model/components: xFormers CUDA error versus a Transformer2DModel shape error." + }, + { + "left": "issue:3431", + "right": "issue:4484", + "accept": false, + "reason": "Both concern poor image quality, but they involve different pipelines/models and different failure modes." + }, + { + "left": "issue:4104", + "right": "issue:8450", + "accept": false, + "reason": "Both mention bad image output, but the titles point to different concrete problems and likely different fixes." + }, + { + "left": "issue:3573", + "right": "issue:3593", + "accept": false, + "reason": "Local upscaler class-label handling is unrelated to a LoRA xformers second-call failure." + }, + { + "left": "issue:4975", + "right": "issue:5135", + "accept": false, + "reason": "Performance regression in LoRA loading versus LoRA quality degradation are related only loosely, not the same bug." + }, + { + "left": "issue:8503", + "right": "issue:8960", + "accept": false, + "reason": "Different training-time issues: gradient checkpointing crash versus gradient accumulation not working." + }, + { + "left": "issue:1231", + "right": "issue:1367", + "accept": false, + "reason": "Scheduler-related, but the titles indicate different bugs and different affected behavior." + }, + { + "left": "issue:12609", + "right": "issue:13311", + "accept": false, + "reason": "Different model families and different failures: NaNs in latents versus LoRA key/prefix mismatch." + }, + { + "left": "issue:5126", + "right": "issue:5225", + "accept": false, + "reason": "A Conv2d forward signature error is not the same as an SDXL VRAM exhaustion issue." + }, + { + "left": "issue:10540", + "right": "issue:8898", + "accept": false, + "reason": "Both are loading issues, but one is a from_single_file Flux regression and the other is from_pretrained path validation." + }, + { + "left": "issue:11041", + "right": "issue:9539", + "accept": false, + "reason": "Different models and code paths: WAN group offloading versus CogVideoX float8 indexing." + }, + { + "left": "issue:9178", + "right": "issue:9677", + "accept": false, + "reason": "Both are Flux/LoRA-adjacent, but one is a PEFT target-module error and the other is about text encoder precision." + }, + { + "left": "issue:10121", + "right": "issue:9225", + "accept": false, + "reason": "Different pipelines and different symptoms: Flux high-resolution generation versus SDXL InstructPix2Pix latent-size mismatch." + }, + { + "left": "issue:1037", + "right": "issue:1140", + "accept": false, + "reason": "A community pipeline request and a community hypernetwork issue are not the same bug or change." + }, + { + "left": "issue:10695", + "right": "issue:7061", + "accept": false, + "reason": "A scheduler timestep calculation bug and a docstring correction are unrelated." + }, + { + "left": "issue:10697", + "right": "issue:7440", + "accept": false, + "reason": "Training-transform inconsistency and dataset availability are different kinds of problems." + }, + { + "left": "issue:1403", + "right": "issue:4668", + "accept": false, + "reason": "Safety checker output differences are not the same as LoRA-generated weird images." + }, + { + "left": "issue:1486", + "right": "issue:3089", + "accept": false, + "reason": "Both are scheduler bugs, but they concern different schedulers and different calculations." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Mostly conservative, but the writeup overstates the evidence in a few places. The summary says all non-matching pairs are \"cross-topic\" false positives, yet several are actually same-domain but different bugs (e.g., scheduler-related or LoRA/training-related), so that wording is too strong. A few pair reasons also rely on specifics not visible in the packet, so the grounding is not fully verifiable here. The 5412\u21946366 edge is the only clearly supported same-bug match; the rest should be described more narrowly as distinct issues rather than cross-topic false positives." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 330, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:11144", + "issue:1188", + "issue:2276", + "issue:2319", + "issue:2560", + "issue:2663", + "issue:2716", + "issue:2776", + "issue:2803", + "issue:3073", + "issue:319", + "issue:3347", + "issue:3585", + "issue:3593", + "issue:3672", + "issue:3761", + "issue:3933", + "issue:4310", + "issue:4332", + "issue:4453", + "issue:4593", + "issue:5046", + "issue:5205", + "issue:5226", + "issue:5516", + "issue:579", + "issue:5925", + "issue:5967", + "issue:6112", + "issue:6363", + "issue:6640", + "issue:709", + "issue:7168", + "issue:761", + "issue:7708", + "issue:7766", + "issue:785", + "issue:7985", + "issue:835", + "issue:8744", + "issue:8885", + "issue:9171", + "issue:9178", + "issue:9439", + "issue:9539", + "issue:9610", + "issue:9717", + "issue:9813" + ], + "soft_pairs": [ + "issue:2319|issue:709", + "issue:2663|issue:3761", + "issue:2716|issue:3585", + "issue:2776|issue:6363", + "issue:3933|issue:7985", + "issue:4310|issue:4453", + "issue:5205|issue:6640", + "issue:5925|issue:9610", + "issue:761|issue:835", + "issue:4593|issue:8885", + "issue:9178|issue:9439", + "issue:9539|issue:9717", + "issue:1188|issue:319", + "issue:2276|issue:3593", + "issue:2560|issue:3347", + "issue:3073|issue:5046", + "issue:3672|issue:785", + "issue:5226|issue:5516", + "issue:7766|issue:9813", + "issue:11144|issue:9171", + "issue:2803|issue:579", + "issue:4332|issue:7168", + "issue:5967|issue:6112", + "issue:7708|issue:8744" + ], + "prepared_review_unit_hash": "b96c26914beb6997c5471ffc2fd6d5616c88c5d60abffab0e8e1b56c58b8b78c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33712, + "estimated_input_tokens": 8428, + "estimated_eval_tokens": 17112 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicate group emerges here; the soft pairs mostly share broad diffusion/LoRA themes but describe different failure modes, code paths, or feature requests. I\u2019d treat this set as mostly unrelated issues with one representative bug report chosen for triage.", + "confidence": 0.29, + "canonical_issue_reason": "issue:9171 is the most concrete, actionable bug report in the set: it names a specific regression, exact API (`StableDiffusionPipeline.from_single_file()`), model family, and failure condition (`v_prediction`).", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9171 is the best overall issue candidate because it is narrowly scoped, reproducible, and tied to a clear code path; it\u2019s a stronger triage anchor than the more generic questions/feature requests.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2319", + "right": "issue:709", + "accept": false, + "reason": "Different problems: a typo in a method implementation vs a documentation/docstring issue." + }, + { + "left": "issue:2663", + "right": "issue:3761", + "accept": false, + "reason": "One is a safety-checker safetensors loading failure; the other is loading two community pipelines in the same process." + }, + { + "left": "issue:2716", + "right": "issue:3585", + "accept": false, + "reason": "Checkpoint-loading help request vs LoRA synthesis failure; no shared bug mechanism." + }, + { + "left": "issue:2776", + "right": "issue:6363", + "accept": false, + "reason": "DreamBooth/DeepSpeed version compatibility on 8GB hardware vs FP16 gradient unscale error in LoRA training." + }, + { + "left": "issue:3933", + "right": "issue:7985", + "accept": false, + "reason": "Both mention compile-related issues, but one is logging breaking graph compile and the other is IP-Adapter incompatibility with `torch.compile`." + }, + { + "left": "issue:4310", + "right": "issue:4453", + "accept": false, + "reason": "SDXL scanline artifacts are an image-quality bug; the other is a conversion/how-to question about LoRA safetensors." + }, + { + "left": "issue:5205", + "right": "issue:6640", + "accept": false, + "reason": "ControlNet/Repaint with LoRA-textual inversion is a pipeline-composition request; Unet3DConditional LoRA is a different model integration task." + }, + { + "left": "issue:5925", + "right": "issue:9610", + "accept": false, + "reason": "ZipLora multi-LoRA enhancement request vs adding `load_lora_weights` to an SD3 img2img pipeline; related area, different change." + }, + { + "left": "issue:761", + "right": "issue:835", + "accept": false, + "reason": "Different code paths: 8-bit Adam training failure vs flax pipeline image conversion failure." + }, + { + "left": "issue:4593", + "right": "issue:8885", + "accept": false, + "reason": "A1111 LoRA rank constraint error is unrelated to clip-retrieval installation and modifier-token loading in the custom diffusion example." + }, + { + "left": "issue:9178", + "right": "issue:9439", + "accept": false, + "reason": "Both are Flux-related, but one is a LoRA PEFT target-module training error and the other is inconsistent outputs when loading the model differently." + }, + { + "left": "issue:9539", + "right": "issue:9717", + "accept": false, + "reason": "Float8 CUDA op limitation in CogVideoX is unrelated to a Flux-dev docker image error." + }, + { + "left": "issue:1188", + "right": "issue:319", + "accept": false, + "reason": "Two separate feature requests for different model/pipeline additions, not the same change." + }, + { + "left": "issue:2276", + "right": "issue:3593", + "accept": false, + "reason": "xformers installation on a notebook vs second-call failure when enabling xformers with LoRA; different failure surface." + }, + { + "left": "issue:2560", + "right": "issue:3347", + "accept": false, + "reason": "List-of-generators nondeterminism in img2img is not the same as NaN image output from text prompts." + }, + { + "left": "issue:3073", + "right": "issue:5046", + "accept": false, + "reason": "Consistency models request and DragonDiffusion request are distinct model-feature proposals." + }, + { + "left": "issue:3672", + "right": "issue:785", + "accept": false, + "reason": "Concurrent-thread failure vs scheduler performance cost; not the same bug." + }, + { + "left": "issue:5226", + "right": "issue:5516", + "accept": false, + "reason": "Both involve LoRA in adapter-like pipelines, but the concrete incompatibilities and affected pipelines differ." + }, + { + "left": "issue:7766", + "right": "issue:9813", + "accept": false, + "reason": "Usage questions about IP-Adapter FaceID Plus vs a specific `prepare_ip_adapter_image_embeds` batch-mixing bug." + }, + { + "left": "issue:11144", + "right": "issue:9171", + "accept": false, + "reason": "Different models and failure modes: Flax UNet dtype initialization vs single-file loading regression for SD 1.5 v-prediction." + }, + { + "left": "issue:2803", + "right": "issue:579", + "accept": false, + "reason": "Local docs build failure on WSL2 is unrelated to a missing `config.json` for textual inversion." + }, + { + "left": "issue:4332", + "right": "issue:7168", + "accept": false, + "reason": "xformers support for Kandinsky 2.2 vs incorrect IP-Adapter image-embed tensor shape; different bugs." + }, + { + "left": "issue:5967", + "right": "issue:6112", + "accept": false, + "reason": "An announcement and a user report of noisy output are not the same underlying issue." + }, + { + "left": "issue:7708", + "right": "issue:8744", + "accept": false, + "reason": "Missing tokenizer config for an SDXL base model vs a request to use single-file loading for advanced model training." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 331, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10209", + "issue:1062", + "issue:10795", + "issue:11083", + "issue:11389", + "issue:11569", + "issue:12047", + "issue:12085", + "issue:12096", + "issue:12120", + "issue:12437", + "issue:13274", + "issue:1614", + "issue:1823", + "issue:1857", + "issue:2179", + "issue:2416", + "issue:2596", + "issue:2761", + "issue:2803", + "issue:2871", + "issue:3097", + "issue:3455", + "issue:3540", + "issue:3565", + "issue:3967", + "issue:4014", + "issue:4171", + "issue:447", + "issue:4496", + "issue:457", + "issue:4578", + "issue:4708", + "issue:4841", + "issue:5138", + "issue:5300", + "issue:6535", + "issue:6574", + "issue:7009", + "issue:7147", + "issue:761", + "issue:776", + "issue:7889", + "issue:801", + "issue:8588", + "issue:9044", + "issue:964" + ], + "soft_pairs": [ + "issue:801|issue:964", + "issue:1062|issue:761", + "issue:6535|issue:6574", + "issue:4014|issue:457", + "issue:12047|issue:12085", + "issue:13274|issue:9044", + "issue:10795|issue:12096", + "issue:10209|issue:7009", + "issue:11083|issue:8588", + "issue:11389|issue:12437", + "issue:11569|issue:12120", + "issue:1614|issue:2179", + "issue:1823|issue:2871", + "issue:1857|issue:776", + "issue:2416|issue:4841", + "issue:2596|issue:2761", + "issue:2803|issue:7147", + "issue:3097|issue:3540", + "issue:3455|issue:5138", + "issue:3565|issue:7889", + "issue:3967|issue:4578", + "issue:4171|issue:5300", + "issue:447|issue:4496", + "issue:4708|issue:5300" + ], + "prepared_review_unit_hash": "961bd2a7d1a27d42463530b8f2f3b39bbe507b89c9f650d031e936ddb37fa390", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33012, + "estimated_input_tokens": 8253, + "estimated_eval_tokens": 16762 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is not a true duplicate set; it contains many unrelated issue reports spanning different models, features, and bugs, with only superficial title similarity from the automatic soft matcher. No soft pair should be merged.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue fits this cluster: the items describe distinct problems across unrelated pipelines/models (GGUF loading, scheduler changes, watermark artifacts, docs, training, etc.).", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "issue:10795 is the strongest standalone issue to represent a cluster member because it is detailed, reproducible, and points to a specific code path (`from_single_file` + GGUF + `torch.compile`), but it is not a true canonical for the whole set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:801", + "right": "issue:964", + "accept": false, + "reason": "Both are image-generation bugs, but one is about attention slicing and the other is about inpainting/image-to-image batch handling; different failure modes and likely different fixes." + }, + { + "left": "issue:1062", + "right": "issue:761", + "accept": false, + "reason": "Both are DreamBooth/training related, but one is about reproducing paper results and the other is an 8-bit Adam training failure; not the same bug." + }, + { + "left": "issue:6535", + "right": "issue:6574", + "accept": false, + "reason": "Both concern Stable Video Diffusion scheduling/behavior, but one asks about swapping the scheduler and the other reports a warning about input range; distinct issues." + }, + { + "left": "issue:4014", + "right": "issue:457", + "accept": false, + "reason": "Both mention visual artifacts, but one is SDXL watermark pixel artifacts and the other is MPS blue artifacts; different hardware/path and symptom." + }, + { + "left": "issue:12047", + "right": "issue:12085", + "accept": false, + "reason": "Both are GGUF/lightx2v related, but one is LoRA fusion on Wan2.2 GGUF and the other is model loading for WanVACETransformer3DModel; different operations and code paths." + }, + { + "left": "issue:13274", + "right": "issue:9044", + "accept": false, + "reason": "Both are model-loading/runtime errors, but one is an AMD recursion error on stable-audio-open and the other is an OSError about missing SDXL weight files; unrelated root causes." + }, + { + "left": "issue:10795", + "right": "issue:12096", + "accept": false, + "reason": "Both involve new model/pipeline support, but one is GGUF + `torch.compile` for transformer models and the other is group offloading in WanVACEPipeline; not the same defect." + }, + { + "left": "issue:10209", + "right": "issue:7009", + "accept": false, + "reason": "The titles are different model-specific problems (`cogvideox-1.5-5b` and TextToVideoSDPipeline with ali-vilab model); no evidence they are the same bug." + }, + { + "left": "issue:11083", + "right": "issue:8588", + "accept": false, + "reason": "One is a DDP stride-mismatch warning during runtime; the other is a missing docs conversion script. Different subsystem and issue type." + }, + { + "left": "issue:11389", + "right": "issue:12437", + "accept": false, + "reason": "Scheduler API refactor request and OmniSVG support request are unrelated feature requests." + }, + { + "left": "issue:11569", + "right": "issue:12120", + "accept": false, + "reason": "Both are about model/distilled-flux/ltxv usage, but one requests a repo for a specific distilled model and the other asks how to train a LoRA for distilled Flux; different asks." + }, + { + "left": "issue:1614", + "right": "issue:2179", + "accept": false, + "reason": "Both report black outputs, but one is Stable Diffusion 2.1 with autocast and the other is FlaxDDPMScheduler; different implementations and likely different bugs." + }, + { + "left": "issue:1823", + "right": "issue:2871", + "accept": false, + "reason": "One is swapped comments in depth2img docs/code, the other is incorrect handling of `output_type='latent'` in inpainting; unrelated." + }, + { + "left": "issue:1857", + "right": "issue:776", + "accept": false, + "reason": "UnCLIP pipeline testing quality and batch_size>1 image-to-image breakage are unrelated issues." + }, + { + "left": "issue:2416", + "right": "issue:4841", + "accept": false, + "reason": "Training slowdown with xformers and a ControlNet image-size issue are different performance/shape problems." + }, + { + "left": "issue:2596", + "right": "issue:2761", + "accept": false, + "reason": "Both are documentation-related, but one is a missing `load_image` reference and the other is documentation for `UNet1DModel.extra_in_channels`; different docs gaps." + }, + { + "left": "issue:2803", + "right": "issue:7147", + "accept": false, + "reason": "WSL2 docs build failure and missing `StableCascadeDecoderPipeline` import are unrelated environment/import problems." + }, + { + "left": "issue:3097", + "right": "issue:3540", + "accept": false, + "reason": "Both concern prompt embeddings/conditioning, but one says `load_textual_inversion` does nothing while the other is `prompt_embeds`/`negative_prompt_embeds` not working in TextToVideoZeroPipeline; different pipeline behavior." + }, + { + "left": "issue:3455", + "right": "issue:5138", + "accept": false, + "reason": "One is a training recipe request for instruct-pix2pix; the other is a training utility refactor request. Not the same bug/change." + }, + { + "left": "issue:3565", + "right": "issue:7889", + "accept": false, + "reason": "Text-to-image training OOM and `load_single_file_model_checkpoint` bugs are not the same failure mode or code path." + }, + { + "left": "issue:3967", + "right": "issue:4578", + "accept": false, + "reason": "One is an SDXL branch/dependency issue; the other is a docs problem about VAE usage. No shared underlying bug." + }, + { + "left": "issue:4171", + "right": "issue:5300", + "accept": false, + "reason": "Reference page rendering and unexpected internet connectivity are unrelated documentation/UI vs networking/configuration issues." + }, + { + "left": "issue:447", + "right": "issue:4496", + "accept": false, + "reason": "Prompt truncation detection and prompt fusion are both prompt-related, but they solve different problems and would not be merged into one fix." + }, + { + "left": "issue:4708", + "right": "issue:5300", + "accept": false, + "reason": "Import error in `diffusers.utils` and unexpected internet connections are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 332, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1037", + "issue:11422", + "issue:1750", + "issue:2272", + "issue:2578", + "issue:2774", + "issue:4176", + "issue:4297", + "issue:4375", + "issue:4888", + "issue:4912", + "issue:5233", + "issue:5447", + "issue:5449", + "issue:5510", + "issue:5619", + "issue:5645", + "issue:5661", + "issue:5824", + "issue:5924", + "issue:5977", + "issue:6054", + "issue:6196", + "issue:6218", + "issue:6275", + "issue:6323", + "issue:6336", + "issue:6791", + "issue:6958", + "issue:699", + "issue:7168", + "issue:7378", + "issue:7421", + "issue:7629", + "issue:7848", + "issue:7946", + "issue:7985", + "issue:8518", + "issue:8589", + "issue:8609", + "issue:8879", + "issue:9034", + "issue:9309", + "issue:9336", + "issue:9350", + "issue:9398", + "issue:9495", + "issue:9813" + ], + "soft_pairs": [ + "issue:5977|issue:6958", + "issue:6196|issue:9495", + "issue:7985|issue:9336", + "issue:6336|issue:7378", + "issue:5619|issue:9350", + "issue:1750|issue:2578", + "issue:5645|issue:6275", + "issue:2272|issue:6323", + "issue:5661|issue:9034", + "issue:9309|issue:9398", + "issue:7168|issue:9813", + "issue:4375|issue:8518", + "issue:2774|issue:4888", + "issue:4297|issue:6218", + "issue:5233|issue:5824", + "issue:5447|issue:7848", + "issue:7629|issue:8609", + "issue:6054|issue:8879", + "issue:1037|issue:699", + "issue:4176|issue:4912", + "issue:5510|issue:5924", + "issue:7421|issue:7946", + "issue:11422|issue:8589", + "issue:5449|issue:6791" + ], + "prepared_review_unit_hash": "f159f6c7f636f6a212a247969ed0e432ac9b96f261290920c72f3c1b4fab555c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34050, + "estimated_input_tokens": 8513, + "estimated_eval_tokens": 17282 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated Diffusers issues spanning training, pipelines, compilation, schedulers, and utilities. None of the soft pairs look like the same underlying bug or change, so all soft edges are rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5977", + "right": "issue:6958", + "accept": false, + "reason": "Different bugs: LoRA checkpoint key naming vs a missing `use_linear_projection` setting in `unet_motion_model`." + }, + { + "left": "issue:6196", + "right": "issue:9495", + "accept": false, + "reason": "Different pipelines and failures: ControlNetXS depth loading vs SDXL PAG with IP-Adapter." + }, + { + "left": "issue:7985", + "right": "issue:9336", + "accept": false, + "reason": "Both involve IP-Adapter/compile, but one is a torch.compile incompatibility and the other is training parameters remaining `requires_grad=True`." + }, + { + "left": "issue:6336", + "right": "issue:7378", + "accept": false, + "reason": "Unrelated: TensorRT VAE compilation with AutoencoderTiny vs AnimateDiff video-to-video latent decoding memory usage." + }, + { + "left": "issue:5619", + "right": "issue:9350", + "accept": false, + "reason": "Different issues: AutoencodeTiny img2img encode behavior vs a deprecation warning about 3D `txt_ids` tensors." + }, + { + "left": "issue:1750", + "right": "issue:2578", + "accept": false, + "reason": "Different failure modes: Debian VM symbol lookup error vs editable-install packaging problem." + }, + { + "left": "issue:5645", + "right": "issue:6275", + "accept": false, + "reason": "LCM core behavior being broken is not the same as T2I-Adapters not supporting all official models." + }, + { + "left": "issue:2272", + "right": "issue:6323", + "accept": false, + "reason": "Different code paths: checkpoint-to-ONNX conversion vs TensorRT img2img `guidance_scale` attribute handling." + }, + { + "left": "issue:5661", + "right": "issue:9034", + "accept": false, + "reason": "Missing `PreTrainedTokenizer` import is unrelated to an AutoPipeline pulling in `sentencepiece`." + }, + { + "left": "issue:9309", + "right": "issue:9398", + "accept": false, + "reason": "FluxControlNet on `schnell` and a missing `pooled_projection` argument are separate model/runtime issues." + }, + { + "left": "issue:7168", + "right": "issue:9813", + "accept": false, + "reason": "Both are IP-Adapter related, but one is about incorrect image embeds and the other about batch feature mixing in `prepare_ip_adapter_image_embeds`." + }, + { + "left": "issue:4375", + "right": "issue:8518", + "accept": false, + "reason": "Both mention `torch.compile`, but they describe different crashes in different models/code paths." + }, + { + "left": "issue:2774", + "right": "issue:4888", + "accept": false, + "reason": "ControlNet safetensors conversion failure is unrelated to fully offline safetensors model loading." + }, + { + "left": "issue:4297", + "right": "issue:6218", + "accept": false, + "reason": "Different batching bugs: SDXL img2img artifacts vs batched seeds handling in ControlNet/img2img pipelines." + }, + { + "left": "issue:5233", + "right": "issue:5824", + "accept": false, + "reason": "Callback step value mismatch is distinct from the SDXL `text_encoder_projection_dim` default issue." + }, + { + "left": "issue:5447", + "right": "issue:7848", + "accept": false, + "reason": "Different pipelines and symptoms: SDXL img2img scheduler/denoising-start bug vs a community LPW decoding issue." + }, + { + "left": "issue:7629", + "right": "issue:8609", + "accept": false, + "reason": "Unrelated: SVD OOM on PyTorch 1.13.1 vs custom pipeline/IP-Adapter incompatibility." + }, + { + "left": "issue:6054", + "right": "issue:8879", + "accept": false, + "reason": "Euler scheduler NumPy conversion misuse is not the same as a torch.compile + cudagraph failure." + }, + { + "left": "issue:1037", + "right": "issue:699", + "accept": false, + "reason": "Community pipeline addition and negative prompt support are different feature requests." + }, + { + "left": "issue:4176", + "right": "issue:4912", + "accept": false, + "reason": "Mandatory authentication is unrelated to converting custom CivitAI SDXL checkpoints to diffusers." + }, + { + "left": "issue:5510", + "right": "issue:5924", + "accept": false, + "reason": "Different runtime problems: dtype mismatch in SDXL InstructPix2Pix vs VAE memory spike." + }, + { + "left": "issue:7421", + "right": "issue:7946", + "accept": false, + "reason": "Seed handling in K-Diffusion is unrelated to custom pipeline caching." + }, + { + "left": "issue:11422", + "right": "issue:8589", + "accept": false, + "reason": "Different training bugs: HiDream LoRA validation prompt crash vs SD3 DreamBooth prior preservation support." + }, + { + "left": "issue:5449", + "right": "issue:6791", + "accept": false, + "reason": "Pickling `preprocess_train` in a training script is not the same as `train_text_to_image_sdxl.py` failing to run." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 333, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10512", + "issue:10569", + "issue:10901", + "issue:11109", + "issue:11144", + "issue:1814", + "issue:2228", + "issue:2459", + "issue:2722", + "issue:2782", + "issue:2783", + "issue:2822", + "issue:2858", + "issue:3418", + "issue:3808", + "issue:4194", + "issue:4222", + "issue:4314", + "issue:4578", + "issue:4675", + "issue:4756", + "issue:4922", + "issue:4969", + "issue:5110", + "issue:5339", + "issue:5628", + "issue:5963", + "issue:6176", + "issue:6354", + "issue:6393", + "issue:6593", + "issue:7053", + "issue:7312", + "issue:7519", + "issue:7770", + "issue:7982", + "issue:8451", + "issue:8834", + "issue:8871", + "issue:9038", + "issue:9052", + "issue:9332", + "issue:9511", + "issue:9648", + "issue:988", + "issue:9930", + "issue:9941", + "issue:9958" + ], + "soft_pairs": [ + "issue:10569|issue:10901", + "issue:1814|issue:2782", + "issue:2822|issue:988", + "issue:2858|issue:4222", + "issue:3418|issue:4756", + "issue:3808|issue:7770", + "issue:4194|issue:5339", + "issue:4578|issue:7519", + "issue:6354|issue:9511", + "issue:9052|issue:9332", + "issue:8871|issue:9930", + "issue:5110|issue:6593", + "issue:6176|issue:9941", + "issue:10512|issue:11109", + "issue:2783|issue:4675", + "issue:4922|issue:8834", + "issue:4969|issue:8451", + "issue:7053|issue:9038", + "issue:7312|issue:9958", + "issue:2228|issue:5628", + "issue:5963|issue:9648", + "issue:4314|issue:6393", + "issue:11144|issue:7982", + "issue:2459|issue:2722" + ], + "prepared_review_unit_hash": "12179fb52e47941d74b87dadc66c911e4c2605465a3e4058a36ac68e95973a3a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33650, + "estimated_input_tokens": 8413, + "estimated_eval_tokens": 17082 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a real duplicate set: it mixes unrelated bugs and feature requests across LoRA loading, HunyuanVideo, ControlNet, schedulers, Flax/JAX, and training scripts. No soft pair looks like the same underlying issue closely enough to merge.", + "confidence": 0.97, + "canonical_issue_reason": "No true duplicate canonical stands out here; if a single representative is needed, issue 9332 is the clearest standalone bug report because it is open, detailed, and reproducible.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 9332 is the strongest self-contained report in the set: it describes a concrete loading failure with enough detail to be actionable.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10569", + "right": "issue:10901", + "accept": false, + "reason": "Different HunyuanVideo problems: CPU memory use vs wrong output when using negative_prompt." + }, + { + "left": "issue:1814", + "right": "issue:2782", + "accept": false, + "reason": "Unrelated requests: CLIP model support vs a sequential offloading error." + }, + { + "left": "issue:2822", + "right": "issue:988", + "accept": false, + "reason": "Different training concerns: convergence quality vs FSDP training setup." + }, + { + "left": "issue:2858", + "right": "issue:4222", + "accept": false, + "reason": "Different img2img/training contexts: seeding image support vs low-resolution dataset training." + }, + { + "left": "issue:3418", + "right": "issue:4756", + "accept": false, + "reason": "Both involve LoRA training, but one asks for text-encoder training code and the other for SDXL style training support; not the same bug or change." + }, + { + "left": "issue:3808", + "right": "issue:7770", + "accept": false, + "reason": "Different failure modes: DeepFloyd IF stage 2 training vs tensor shape mismatch when loading model and LoRA." + }, + { + "left": "issue:4194", + "right": "issue:5339", + "accept": false, + "reason": "Both are loading-related, but one is VAE handling in from_single_file and the other is local safetensor loading failure; different code paths." + }, + { + "left": "issue:4578", + "right": "issue:7519", + "accept": false, + "reason": "Documentation issue vs xformers attention-processor behavior; not the same bug." + }, + { + "left": "issue:6354", + "right": "issue:9511", + "accept": false, + "reason": "Feature request for ControlNet LoRA support vs batching support for SDXL ControlNet inpaint; different scope and problem." + }, + { + "left": "issue:9052", + "right": "issue:9332", + "accept": false, + "reason": "Both are model-loading bugs, but one is from_single_file failing on some checkpoints and the other is from_pretrained loading the wrong weights; different API paths and symptoms." + }, + { + "left": "issue:8871", + "right": "issue:9930", + "accept": false, + "reason": "Different training bugs: FP16 unscale error during LoRA training vs PAG adaptive scale behavior." + }, + { + "left": "issue:5110", + "right": "issue:6593", + "accept": false, + "reason": "Repo-id validation error is unrelated to adding load_lora_weights support in SDXL img2img." + }, + { + "left": "issue:6176", + "right": "issue:9941", + "accept": false, + "reason": "LoRA weight loading after DreamBooth training vs a general Colab runtime error; not the same underlying bug." + }, + { + "left": "issue:10512", + "right": "issue:11109", + "accept": false, + "reason": "Different LoRA issues: Quanto Flux LoRA loading vs depth LoRA transformer-layer mismatch." + }, + { + "left": "issue:2783", + "right": "issue:4675", + "accept": false, + "reason": "Different ControlNet feature requests: a new img2img/inpainting pipeline vs SDXL reference-only support." + }, + { + "left": "issue:4922", + "right": "issue:8834", + "accept": false, + "reason": "Unrelated requests: Android demo vs SD3 ControlNet training code release." + }, + { + "left": "issue:4969", + "right": "issue:8451", + "accept": false, + "reason": "ImportError for CrossAttention is unrelated to training-script documentation about frozen UNet parameters." + }, + { + "left": "issue:7053", + "right": "issue:9038", + "accept": false, + "reason": "Different model families and behaviors: UNet3D class conditioning vs prompt weighting in FlaxStableDiffusionPipeline." + }, + { + "left": "issue:7312", + "right": "issue:9958", + "accept": false, + "reason": "AlignProp direct reward finetuning is unrelated to AutoencoderKLTemporalDecoder tiling support." + }, + { + "left": "issue:2228", + "right": "issue:5628", + "accept": false, + "reason": "Pipeline GPU placement issue vs scheduler denoising offset bug; different code paths." + }, + { + "left": "issue:5963", + "right": "issue:9648", + "accept": false, + "reason": "Kandinsky embeds/prompt handling is unrelated to VAE tiling memory overflow." + }, + { + "left": "issue:4314", + "right": "issue:6393", + "accept": false, + "reason": "Interrupted-download write protection vs private HF link failure in from_single_file." + }, + { + "left": "issue:11144", + "right": "issue:7982", + "accept": false, + "reason": "Flax dtype initialization bug vs from_pipe with MultiControlNetModel; different subsystems and symptoms." + }, + { + "left": "issue:2459", + "right": "issue:2722", + "accept": false, + "reason": "LoRA xformers training issue vs ControlNet conversion mismatch; not the same underlying problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 334, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10100", + "issue:10194", + "issue:10427", + "issue:1084", + "issue:10954", + "issue:11644", + "issue:11830", + "issue:12308", + "issue:12505", + "issue:12695", + "issue:13105", + "issue:13203", + "issue:1407", + "issue:1546", + "issue:1566", + "issue:1677", + "issue:2560", + "issue:2716", + "issue:2772", + "issue:2785", + "issue:3296", + "issue:3381", + "issue:3551", + "issue:3570", + "issue:3740", + "issue:4073", + "issue:4236", + "issue:4349", + "issue:4488", + "issue:4615", + "issue:4977", + "issue:5325", + "issue:5399", + "issue:5628", + "issue:5693", + "issue:5710", + "issue:6154", + "issue:6256", + "issue:6560", + "issue:7498", + "issue:7548", + "issue:8405", + "issue:843", + "issue:8514", + "issue:8801", + "issue:8832", + "issue:9628", + "issue:9804" + ], + "soft_pairs": [ + "issue:6560|issue:9628", + "issue:10100|issue:8801", + "issue:10194|issue:12695", + "issue:10427|issue:12308", + "issue:1084|issue:843", + "issue:12505|issue:9804", + "issue:2785|issue:6256", + "issue:3570|issue:4615", + "issue:3740|issue:4236", + "issue:1566|issue:3296", + "issue:2772|issue:4349", + "issue:3381|issue:5399", + "issue:10954|issue:13203", + "issue:11644|issue:8514", + "issue:11830|issue:13105", + "issue:1407|issue:1546", + "issue:2716|issue:5693", + "issue:4073|issue:8405", + "issue:6154|issue:7498", + "issue:1677|issue:5325", + "issue:2560|issue:3551", + "issue:4488|issue:4977", + "issue:5628|issue:5710", + "issue:7548|issue:8832" + ], + "prepared_review_unit_hash": "87c38422b7ae7943f37372f5751ec540802d94cd9e8ae897c9321a82bea14cb0", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33712, + "estimated_input_tokens": 8428, + "estimated_eval_tokens": 17112 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated feature requests and bug reports around different diffusion models, pipelines, training scripts, and loading paths. The soft pairs share only broad keywords like \"support\", \"load\", or \"error\", not the same concrete bug/change, so none should be merged as duplicates.", + "confidence": 0.9, + "canonical_issue_reason": "issue:9804 is the clearest concrete, self-contained bug report in the set: it names a specific API (`ControlNetModel.from_pretrained`), a specific failure mode, and a specific model class (SDXL ControlNet local loading).", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9804 is the best representative issue because it is specific, actionable, and tied to a precise code path; several others are either broader feature requests or more ambiguous reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6560", + "right": "issue:9628", + "accept": false, + "reason": "Both involve LoRA/training stability, but they describe different scripts, symptoms, and failure modes; not the same underlying bug." + }, + { + "left": "issue:10100", + "right": "issue:8801", + "accept": false, + "reason": "Both are model-support requests, but for different architectures and different work items." + }, + { + "left": "issue:10194", + "right": "issue:12695", + "accept": false, + "reason": "Both are \"support\" requests, but they target unrelated features/models and do not point to one shared change." + }, + { + "left": "issue:10427", + "right": "issue:12308", + "accept": false, + "reason": "One is GGUF loading for AuraFlow; the other is flash-attn kernel support on non-Hopper GPUs. Different problems." + }, + { + "left": "issue:1084", + "right": "issue:843", + "accept": false, + "reason": "Both are documentation-related, but they cover different topics and would not be fixed by one PR." + }, + { + "left": "issue:12505", + "right": "issue:9804", + "accept": false, + "reason": "Both mention pipeline/loading errors, but one is a tgate/img2img bug and the other is ControlNet local loading; different code paths." + }, + { + "left": "issue:2785", + "right": "issue:6256", + "accept": false, + "reason": "M1/M2 support and callback API support are unrelated feature requests." + }, + { + "left": "issue:3570", + "right": "issue:4615", + "accept": false, + "reason": "One is about disabling downloads when loading checkpoints; the other is an import/export symbol issue. Different bugs." + }, + { + "left": "issue:3740", + "right": "issue:4236", + "accept": false, + "reason": "Different failures: device_map conversion with accelerate vs SDXL refiner text-encoder registration." + }, + { + "left": "issue:1566", + "right": "issue:3296", + "accept": false, + "reason": "Both involve checkpoint save/load in Dreambooth, but the specific failure conditions differ enough that they are not the same bug." + }, + { + "left": "issue:2772", + "right": "issue:4349", + "accept": false, + "reason": "A local dataset validation issue in a training script is not the same as missing text_encoder handling in an upscale pipeline." + }, + { + "left": "issue:3381", + "right": "issue:5399", + "accept": false, + "reason": "Both touch inpainting/dtype behavior, but one is batch discoloration and a docs concern, while the other is a custom-model matmul dtype mismatch." + }, + { + "left": "issue:10954", + "right": "issue:13203", + "accept": false, + "reason": "Both are LoRA-related, but they concern different models and different load failures; too broad to merge." + }, + { + "left": "issue:11644", + "right": "issue:8514", + "accept": false, + "reason": "Different pipeline/training errors with different triggers and components; no shared fix is evident." + }, + { + "left": "issue:11830", + "right": "issue:13105", + "accept": false, + "reason": "Both are output-quality problems, but one is black output under float16 and the other is noise in a different model; not the same bug." + }, + { + "left": "issue:1407", + "right": "issue:1546", + "accept": false, + "reason": "An ONNX feature_extractor regression and a truncation warning batch-input error are distinct issues." + }, + { + "left": "issue:2716", + "right": "issue:5693", + "accept": false, + "reason": "General checkpoint-loading help and a VAE loading error are not the same underlying issue." + }, + { + "left": "issue:4073", + "right": "issue:8405", + "accept": false, + "reason": "Safety-checker dtype error vs multi-ControlNet tensor-size mismatch: different subsystems and failure modes." + }, + { + "left": "issue:6154", + "right": "issue:7498", + "accept": false, + "reason": "Both are Apple-silicon training issues, but they affect different models/scripts and do not share one concrete bug." + }, + { + "left": "issue:1677", + "right": "issue:5325", + "accept": false, + "reason": "Wrong data type when calling a pipeline is unrelated to packaging type information distribution." + }, + { + "left": "issue:2560", + "right": "issue:3551", + "accept": false, + "reason": "Different regressions: generator-list determinism vs LoRA disabling xFormers." + }, + { + "left": "issue:4488", + "right": "issue:4977", + "accept": false, + "reason": "Checkpoint conversion/loading behavior and checkpoint cleanup in a training example are unrelated." + }, + { + "left": "issue:5628", + "right": "issue:5710", + "accept": false, + "reason": "The former is a denoising offset bug; the latter is missing sigma initialization. Related area, but not the same code-path problem." + }, + { + "left": "issue:7548", + "right": "issue:8832", + "accept": false, + "reason": "AnimateDiff flicker and a pipeline crash on image access are different issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 335, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10381", + "issue:10446", + "issue:10893", + "issue:11032", + "issue:11700", + "issue:11828", + "issue:12437", + "issue:1655", + "issue:2410", + "issue:2444", + "issue:2659", + "issue:2784", + "issue:3428", + "issue:3455", + "issue:3679", + "issue:3944", + "issue:3974", + "issue:4232", + "issue:4405", + "issue:4446", + "issue:4484", + "issue:4649", + "issue:4672", + "issue:4803", + "issue:5392", + "issue:5485", + "issue:5510", + "issue:5518", + "issue:5926", + "issue:6050", + "issue:617", + "issue:6335", + "issue:6441", + "issue:6685", + "issue:7177", + "issue:7456", + "issue:746", + "issue:7658", + "issue:8228", + "issue:8434", + "issue:8450", + "issue:8708", + "issue:9067", + "issue:9223", + "issue:9398", + "issue:9476", + "issue:9933" + ], + "soft_pairs": [ + "issue:3428|issue:746", + "issue:4446|issue:8228", + "issue:11828|issue:9223", + "issue:2659|issue:5485", + "issue:4484|issue:8450", + "issue:10381|issue:10446", + "issue:10893|issue:11700", + "issue:11032|issue:12437", + "issue:11828|issue:9476", + "issue:1655|issue:617", + "issue:2410|issue:2444", + "issue:2784|issue:3974", + "issue:3455|issue:4405", + "issue:3679|issue:5926", + "issue:3944|issue:4672", + "issue:4232|issue:5392", + "issue:4649|issue:5510", + "issue:4803|issue:6050", + "issue:5518|issue:9067", + "issue:6335|issue:9933", + "issue:6441|issue:7177", + "issue:6685|issue:8708", + "issue:7456|issue:9398", + "issue:7658|issue:8434" + ], + "prepared_review_unit_hash": "a2c11e95ea470399ef8111decd0b76798838384ee4cdb2520cf89fc565e5e74e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33668, + "estimated_input_tokens": 8417, + "estimated_eval_tokens": 17090 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated bug reports, feature requests, training issues, install problems, and pipeline-specific regressions. The soft pairs mostly share only broad Diffusers vocabulary, not the same underlying bug or change, so none should be merged.", + "confidence": 0.94, + "canonical_issue_reason": "No single canonical issue stands out; the items cover many unrelated pipelines, models, and request types, so this cluster should not be collapsed to one issue.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10381 is the broadest concrete bug report in the set (Flux quantization/LoRA inference), but it is still not a true umbrella for the rest of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3428", + "right": "issue:746", + "accept": false, + "reason": "Different problems: one is an unexpected img2img pipeline result, the other is an ONNX revision loading issue." + }, + { + "left": "issue:4446", + "right": "issue:8228", + "accept": false, + "reason": "Both mention from_single_file/text encoder loading, but one is about custom text_encoder override support while the other is a CLIPTextModel load failure." + }, + { + "left": "issue:11828", + "right": "issue:9223", + "accept": false, + "reason": "Completely unrelated: a type error in ControlNet conditioning arguments versus a dependency/version conflict on pyarrow." + }, + { + "left": "issue:2659", + "right": "issue:5485", + "accept": false, + "reason": "Different symptoms and scope: batch-size-related output variation versus ControlNet batch inference slowness." + }, + { + "left": "issue:4484", + "right": "issue:8450", + "accept": false, + "reason": "Both concern image quality, but one is about SDXL Kohya LoRA training artifacts and the other is an inpainting noise regression at strength=1.0." + }, + { + "left": "issue:10381", + "right": "issue:10446", + "accept": false, + "reason": "Different code paths: Flux quantization/LoRA inference error versus multi-GPU/model-sharding incompatibility with FluxControlNetPipeline." + }, + { + "left": "issue:10893", + "right": "issue:11700", + "accept": false, + "reason": "Both are feature requests, but they target different integrations: CogvideoX ControlNet support versus Pruna loading integration." + }, + { + "left": "issue:11032", + "right": "issue:12437", + "accept": false, + "reason": "Unrelated requests: TensorRT backend support versus OmniSVG model support." + }, + { + "left": "issue:11828", + "right": "issue:9476", + "accept": false, + "reason": "Different failures: a ControlNet argument type bug versus a Flux LoRA validation dtype mismatch." + }, + { + "left": "issue:1655", + "right": "issue:617", + "accept": false, + "reason": "Different issues entirely: pipeline initialization failure versus an NNPACK hardware warning from AutoencoderKL." + }, + { + "left": "issue:2410", + "right": "issue:2444", + "accept": false, + "reason": "Different training script failures: missing saved model artifacts versus a subprocess exit error in dreambooth inpaint." + }, + { + "left": "issue:2784", + "right": "issue:3974", + "accept": false, + "reason": "Different requests: Dreambooth support for text2video models versus safetensors support for ControlNet inpaint." + }, + { + "left": "issue:3455", + "right": "issue:4405", + "accept": false, + "reason": "Different training features: pretrained instruct-pix2pix usage versus a textual inversion training script for SDXL." + }, + { + "left": "issue:3679", + "right": "issue:5926", + "accept": false, + "reason": "Different model requests: specific text2video model additions versus Kandinsky 3.0 support." + }, + { + "left": "issue:3944", + "right": "issue:4672", + "accept": false, + "reason": "Not the same change: network-rank configurability for LoRA training versus reducing VRAM usage." + }, + { + "left": "issue:4232", + "right": "issue:5392", + "accept": false, + "reason": "Different tasks: instruct-pix2pix with ControlNet conditioning versus training an unconditional latent diffusion model." + }, + { + "left": "issue:4649", + "right": "issue:5510", + "accept": false, + "reason": "Different runtime errors in different pipelines: NoneType added_cond_kwargs versus dtype mismatch in SDXL InstructPix2Pix." + }, + { + "left": "issue:4803", + "right": "issue:6050", + "accept": false, + "reason": "Different installation failures: missing requirements file versus pip install subprocess-exited-with-error." + }, + { + "left": "issue:5518", + "right": "issue:9067", + "accept": false, + "reason": "Different import/runtime problems: ModuleList missing time attribute versus missing diffusers.hub_utils module." + }, + { + "left": "issue:6335", + "right": "issue:9933", + "accept": false, + "reason": "Related only at a high level; these are different pipeline API issues on different models and parameters, not the same bug." + }, + { + "left": "issue:6441", + "right": "issue:7177", + "accept": false, + "reason": "Both involve from_pretrained, but one is a Google Colab issue and the other is a StableCascadeDecoderPipeline-specific load error." + }, + { + "left": "issue:6685", + "right": "issue:8708", + "accept": false, + "reason": "Different Dreambooth topics: absolute output_dir path handling versus a missing SD3 VAE shift_factor in examples." + }, + { + "left": "issue:7456", + "right": "issue:9398", + "accept": false, + "reason": "Different bugs: a missing raise statement versus a missing pooled_projection argument in a projection embedding module." + }, + { + "left": "issue:7658", + "right": "issue:8434", + "accept": false, + "reason": "Both mention multiple ControlNets, but one is a question about whether multiple conditioning is possible and the other is a feature request for specific XSPipelines; not the same concrete bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 336, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10186", + "issue:10428", + "issue:11092", + "issue:1158", + "issue:11581", + "issue:11872", + "issue:12574", + "issue:12610", + "issue:13298", + "issue:1695", + "issue:1807", + "issue:1842", + "issue:1851", + "issue:1896", + "issue:2444", + "issue:2507", + "issue:2554", + "issue:2799", + "issue:3175", + "issue:3431", + "issue:3740", + "issue:3887", + "issue:3967", + "issue:4598", + "issue:4789", + "issue:488", + "issue:4972", + "issue:5009", + "issue:5152", + "issue:5372", + "issue:5523", + "issue:557", + "issue:5570", + "issue:5951", + "issue:5977", + "issue:6068", + "issue:6158", + "issue:6196", + "issue:6392", + "issue:6869", + "issue:713", + "issue:7210", + "issue:7312", + "issue:782", + "issue:8507", + "issue:8708", + "issue:9052", + "issue:9972" + ], + "soft_pairs": [ + "issue:8507|issue:9972", + "issue:1896|issue:4972", + "issue:2444|issue:713", + "issue:3431|issue:557", + "issue:1842|issue:5009", + "issue:11872|issue:12574", + "issue:1851|issue:3175", + "issue:5372|issue:7210", + "issue:10428|issue:11092", + "issue:2799|issue:488", + "issue:1158|issue:5523", + "issue:4789|issue:782", + "issue:12610|issue:13298", + "issue:6068|issue:6196", + "issue:2554|issue:3740", + "issue:5152|issue:6869", + "issue:10186|issue:11581", + "issue:1807|issue:2507", + "issue:4598|issue:5570", + "issue:1695|issue:5951", + "issue:3887|issue:3967", + "issue:5977|issue:8708", + "issue:6158|issue:7312", + "issue:6392|issue:9052" + ], + "prepared_review_unit_hash": "c01acc4ef5bf30985055068e89781a52231d6be9f2d5e66f8f181c7109bed718", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33769, + "estimated_input_tokens": 8443, + "estimated_eval_tokens": 17142 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is a grab bag of unrelated diffusers issues; none of the soft-similarity pairs look like the same underlying bug or change. All soft edges should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 6068 is the strongest representative bug report: concrete, reproducible, and high-signal (SDXL/DDIM incorrect sampling) with substantial discussion.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6068 is the best overall issue candidate because it describes a specific code-path defect with clear scope and impact, making it a strong canonical reference.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8507", + "right": "issue:9972", + "accept": false, + "reason": "Different failures: SD3 DreamBooth training vs CogX/macOS buffer allocation; no shared bug path." + }, + { + "left": "issue:1896", + "right": "issue:4972", + "accept": false, + "reason": "Different pipelines and errors; CLIP-guided SD size/type issue is unrelated to wuerstchen example startup failure." + }, + { + "left": "issue:2444", + "right": "issue:713", + "accept": false, + "reason": "Generic training error is too broad to equate with a specific DreamBooth inpaint subprocess failure." + }, + { + "left": "issue:3431", + "right": "issue:557", + "accept": false, + "reason": "Different symptoms and contexts: ControlNet inpainting quality vs DDIM quality comparison." + }, + { + "left": "issue:1842", + "right": "issue:5009", + "accept": false, + "reason": "Unexpected `eta` kwarg is a scheduler API mismatch, not the same as a generic wuerstchen error." + }, + { + "left": "issue:11872", + "right": "issue:12574", + "accept": false, + "reason": "Performance/offloading inefficiency is not the same as WanVideo control-argument handling failure." + }, + { + "left": "issue:1851", + "right": "issue:3175", + "accept": false, + "reason": "Both are training-related, but they describe different multi-GPU cases and different scripts." + }, + { + "left": "issue:5372", + "right": "issue:7210", + "accept": false, + "reason": "Safety checker usage question is unrelated to missing `named_parameters` under accelerate dispatch." + }, + { + "left": "issue:10428", + "right": "issue:11092", + "accept": false, + "reason": "Flux inference on Ascend NPU is a hardware/runtime error; empty timesteps in SDXL-Turbo is a scheduler edge case." + }, + { + "left": "issue:2799", + "right": "issue:488", + "accept": false, + "reason": "Checkpoint/pipeline mismatch error is unrelated to grayscale `numpy_to_pil` handling." + }, + { + "left": "issue:1158", + "right": "issue:5523", + "accept": false, + "reason": "Test interference and inconsistent results are both broad, but not the same concrete bug." + }, + { + "left": "issue:4789", + "right": "issue:782", + "accept": false, + "reason": "DeepFloyd download failure is unrelated to the auth-token API removal issue." + }, + { + "left": "issue:12610", + "right": "issue:13298", + "accept": false, + "reason": "Both mention device mismatch, but one is quant/offload weight placement and the other is generator device handling; not the same bug." + }, + { + "left": "issue:6068", + "right": "issue:6196", + "accept": false, + "reason": "Incorrect DDIM sampling in SDXL is a different failure from ControlNetXS Depth SDXL loading." + }, + { + "left": "issue:2554", + "right": "issue:3740", + "accept": false, + "reason": "Unreleased accelerate dependency is unrelated to legacy attention block conversion with sequential device_map." + }, + { + "left": "issue:5152", + "right": "issue:6869", + "accept": false, + "reason": "W\u00fcrstchen alpha bug and `export_to_video` unpacking error are different code paths." + }, + { + "left": "issue:10186", + "right": "issue:11581", + "accept": false, + "reason": "ImportError for `FluxFillPipeline` is not the same as an IP Adapter dtype-constructor error in FLUX." + }, + { + "left": "issue:1807", + "right": "issue:2507", + "accept": false, + "reason": "Both are performance complaints, but one is generation slowdown and the other is safetensors load speed." + }, + { + "left": "issue:4598", + "right": "issue:5570", + "accept": false, + "reason": "Missing dependency messaging is a UX/error-reporting issue, not the same as missing `compute_snr` in training utils." + }, + { + "left": "issue:1695", + "right": "issue:5951", + "accept": false, + "reason": "DreamBooth multi-GPU training and Pixart training are different workloads with different failure modes." + }, + { + "left": "issue:3887", + "right": "issue:3967", + "accept": false, + "reason": "Quality/regression complaint is unrelated to a branch/deleted-code runtime failure." + }, + { + "left": "issue:5977", + "right": "issue:8708", + "accept": false, + "reason": "LoRA checkpoint key-saving and SD3 DreamBooth `shift_factor` missing are different training issues." + }, + { + "left": "issue:6158", + "right": "issue:7312", + "accept": false, + "reason": "Animatediff SDXL support request is unrelated to AlignProp reward finetuning support." + }, + { + "left": "issue:6392", + "right": "issue:9052", + "accept": false, + "reason": "Custom LoRA loading and single-file civit checkpoint conversion are both loading flows, but not the same failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 337, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10039", + "issue:10350", + "issue:1101", + "issue:11329", + "issue:1140", + "issue:11424", + "issue:11489", + "issue:11833", + "issue:1197", + "issue:12436", + "issue:13140", + "issue:2044", + "issue:2596", + "issue:3300", + "issue:341", + "issue:3880", + "issue:3944", + "issue:4183", + "issue:4356", + "issue:4419", + "issue:4835", + "issue:4876", + "issue:5079", + "issue:5109", + "issue:5224", + "issue:5335", + "issue:5897", + "issue:5954", + "issue:6087", + "issue:6709", + "issue:6725", + "issue:6732", + "issue:6812", + "issue:7142", + "issue:7264", + "issue:7312", + "issue:8477", + "issue:8532", + "issue:8551", + "issue:8707", + "issue:8780", + "issue:8805", + "issue:8826", + "issue:8939", + "issue:8958", + "issue:8994", + "issue:9602", + "issue:9856" + ], + "soft_pairs": [ + "issue:10350|issue:8551", + "issue:4356|issue:8994", + "issue:12436|issue:13140", + "issue:8477|issue:8707", + "issue:9602|issue:9856", + "issue:8826|issue:8958", + "issue:11424|issue:11833", + "issue:10039|issue:8805", + "issue:1101|issue:341", + "issue:1140|issue:1197", + "issue:3880|issue:6087", + "issue:3944|issue:4876", + "issue:4183|issue:6725", + "issue:4419|issue:6732", + "issue:5079|issue:7312", + "issue:5897|issue:6812", + "issue:11489|issue:8939", + "issue:11329|issue:8532", + "issue:2044|issue:5335", + "issue:2596|issue:3300", + "issue:7264|issue:8780", + "issue:5224|issue:6709", + "issue:4835|issue:5109", + "issue:5954|issue:7142" + ], + "prepared_review_unit_hash": "48f7874eea34b261181d1563aea2a75eb9beabfd75d003e983fd593703d7757c", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34445, + "estimated_input_tokens": 8612, + "estimated_eval_tokens": 17480 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: it groups many unrelated issues across schedulers, LoRA training, VAE/device-map bugs, pipeline failures, and docs. All soft pairs are only superficially similar and should stay separate.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue fits this cluster; the items are heterogeneous and do not describe the same underlying bug or change.", + "canonical_pr_reason": null, + "best_issue_reason": "No strong best issue candidate either; the cluster is too mixed for one issue to safely represent the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10350", + "right": "issue:8551", + "accept": false, + "reason": "Both involve VAE decoding, but one is high memory use in SD2.1 and the other is a device-map latent transfer bug in SDXL inpaint; different failure modes." + }, + { + "left": "issue:4356", + "right": "issue:8994", + "accept": false, + "reason": "0.19.2 missing dummy module import is unrelated to generation steps stuck at 0." + }, + { + "left": "issue:12436", + "right": "issue:13140", + "accept": false, + "reason": "Different errors and code paths: CLIPTextModel init keyword mismatch vs Qwen pipeline NoneType iteration." + }, + { + "left": "issue:8477", + "right": "issue:8707", + "accept": false, + "reason": "Both training-related, but one is a DDP error in a research script and the other is an SDXL text-to-image training failure; not the same bug." + }, + { + "left": "issue:9602", + "right": "issue:9856", + "accept": false, + "reason": "Adapter registration error and distributed port conflict are unrelated." + }, + { + "left": "issue:8826", + "right": "issue:8958", + "accept": false, + "reason": "torch.cat on an empty list is a different runtime issue from torch.library.register_fake missing." + }, + { + "left": "issue:11424", + "right": "issue:11833", + "accept": false, + "reason": "VAE group_norm decode failure is unrelated to a LoRA device-setting KeyError." + }, + { + "left": "issue:10039", + "right": "issue:8805", + "accept": false, + "reason": "Forgotten output sample size in a UNet block is not the same as the SDXL conversion script regression." + }, + { + "left": "issue:1101", + "right": "issue:341", + "accept": false, + "reason": "Both are scheduler requests, but DPM-Solver and K-LMS in-paint are distinct features, not one bug/change." + }, + { + "left": "issue:1140", + "right": "issue:1197", + "accept": false, + "reason": "Separate community feature requests with no shared code path." + }, + { + "left": "issue:3880", + "right": "issue:6087", + "accept": false, + "reason": "Inpainting degradation and LoRAConfig save/load metadata issues are unrelated." + }, + { + "left": "issue:3944", + "right": "issue:4876", + "accept": false, + "reason": "Both mention LoRA, but one requests network-rank customization for training and the other asks for a std argument for LoRA layers; different changes." + }, + { + "left": "issue:4183", + "right": "issue:6725", + "accept": false, + "reason": "ControlNet guidance_scale bug and LDMPipeline generation request are unrelated." + }, + { + "left": "issue:4419", + "right": "issue:6732", + "accept": false, + "reason": "Offset-noise inference mismatch is unrelated to documentation for trained_betas." + }, + { + "left": "issue:5079", + "right": "issue:7312", + "accept": false, + "reason": "Loading a specific LoRA type and AlignProp support are different features." + }, + { + "left": "issue:5897", + "right": "issue:6812", + "accept": false, + "reason": "Both are LoRA training errors, but device mismatch and dtype mismatch are different underlying bugs." + }, + { + "left": "issue:11489", + "right": "issue:8939", + "accept": false, + "reason": "Flux LoRA training failure and tokenizer load failure for SD3 are unrelated." + }, + { + "left": "issue:11329", + "right": "issue:8532", + "accept": false, + "reason": "Memory leak in bitsandbytes quantization and progressive memory increase when loading LoRA weights are different code paths." + }, + { + "left": "issue:2044", + "right": "issue:5335", + "accept": false, + "reason": "DDP dtype attribute error has nothing to do with local deployment behind blocked Hugging Face access." + }, + { + "left": "issue:2596", + "right": "issue:3300", + "accept": false, + "reason": "Documentation typo about load_image is unrelated to the broader 'Diffusers on HuggingFace are broken' report." + }, + { + "left": "issue:7264", + "right": "issue:8780", + "accept": false, + "reason": "Serialization/loading of custom transformers models is unrelated to a model/input dtype mismatch." + }, + { + "left": "issue:5224", + "right": "issue:6709", + "accept": false, + "reason": "Flax SD2.1 error and custom diffusion concept-list training bug are different issues." + }, + { + "left": "issue:4835", + "right": "issue:5109", + "accept": false, + "reason": "Same pipeline family, but MultiControlNetModel img2img failure and load_lora_weights exception are distinct concrete problems." + }, + { + "left": "issue:5954", + "right": "issue:7142", + "accept": false, + "reason": "Kandinsky22 inference error and SDXL-lightning output mismatch are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 338, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10121", + "issue:10282", + "issue:10591", + "issue:10895", + "issue:11302", + "issue:11436", + "issue:11938", + "issue:175", + "issue:181", + "issue:2052", + "issue:2276", + "issue:2332", + "issue:2676", + "issue:3159", + "issue:3380", + "issue:3453", + "issue:3593", + "issue:4292", + "issue:4403", + "issue:4585", + "issue:4615", + "issue:4709", + "issue:4753", + "issue:4872", + "issue:5124", + "issue:5315", + "issue:5799", + "issue:6162", + "issue:6517", + "issue:6634", + "issue:6732", + "issue:7009", + "issue:7165", + "issue:7341", + "issue:7365", + "issue:7562", + "issue:7572", + "issue:7651", + "issue:7931", + "issue:8451", + "issue:8534", + "issue:8551", + "issue:8975", + "issue:9051", + "issue:9090", + "issue:9181", + "issue:9822", + "issue:9844" + ], + "soft_pairs": [ + "issue:7651|issue:9090", + "issue:4403|issue:7572", + "issue:10282|issue:6517", + "issue:11302|issue:11436", + "issue:2052|issue:5124", + "issue:8551|issue:9844", + "issue:3593|issue:7562", + "issue:10591|issue:8451", + "issue:10895|issue:7365", + "issue:11938|issue:9051", + "issue:175|issue:3159", + "issue:181|issue:3453", + "issue:2332|issue:2676", + "issue:3380|issue:7165", + "issue:4292|issue:5799", + "issue:4615|issue:9181", + "issue:4709|issue:7931", + "issue:4753|issue:8534", + "issue:5315|issue:6634", + "issue:6732|issue:7341", + "issue:7009|issue:9822", + "issue:10121|issue:6162", + "issue:4585|issue:8975", + "issue:2276|issue:4872" + ], + "prepared_review_unit_hash": "7cbb23eec8e072cb22f86750396be1e8d45bac3f5c8d5ce87335124b3ab30e83", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 34199, + "estimated_input_tokens": 8550, + "estimated_eval_tokens": 17356 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated issues across inference, training scripts, schedulers, docs, import errors, and model-specific bugs. No soft-edge pair looks like the same underlying defect/change.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 7651 is the most central and well-developed bug report here: it targets a concrete pipeline timing defect, has the highest discussion activity, and is more representative than the narrower one-off reports.", + "canonical_pr_reason": null, + "best_issue_reason": "7651 is the best standalone issue to keep as the cluster anchor because it describes a clear, reproducible core pipeline bug rather than a peripheral docs/import/training variant.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7651", + "right": "issue:9090", + "accept": false, + "reason": "Both mention timesteps, but 7651 is about img2img timestep calculation while 9090 is a PNDMScheduler repeated-timestep bug; different code paths and symptoms." + }, + { + "left": "issue:4403", + "right": "issue:7572", + "accept": false, + "reason": "Both are dtype-related, but one is LayerNorm half-kernel support and the other is input/bias dtype mismatch; not the same bug." + }, + { + "left": "issue:10282", + "right": "issue:6517", + "accept": false, + "reason": "A model weight shape mismatch during loading is unrelated to a pipeline import failure in SDXL ControlNet img2img." + }, + { + "left": "issue:11302", + "right": "issue:11436", + "accept": false, + "reason": "One is about saving a quantized HiDream model; the other is an inference dtype mismatch in HiDream. Same model family, different failure modes." + }, + { + "left": "issue:2052", + "right": "issue:5124", + "accept": false, + "reason": "Checkpoint resume behavior in textual inversion is not the same as a question about fine-tuning from a .safetensors checkpoint." + }, + { + "left": "issue:8551", + "right": "issue:9844", + "accept": false, + "reason": "Device placement for VAE latents with balanced device_map is unrelated to SDXL VAE NaN outputs." + }, + { + "left": "issue:3593", + "right": "issue:7562", + "accept": false, + "reason": "LoRA+xformers second-call failure is a different problem from multi-ControlNet CPU offload failure in img2img." + }, + { + "left": "issue:10591", + "right": "issue:8451", + "accept": false, + "reason": "These concern different training-script issues: SD3 LoRA script correctness versus sd1.x training details and frozen parameters." + }, + { + "left": "issue:10895", + "right": "issue:7365", + "accept": false, + "reason": "Deprecation handling is unrelated to pooled_prompt_embeds being overwritten by prompt_embeds[0]." + }, + { + "left": "issue:11938", + "right": "issue:9051", + "accept": false, + "reason": "RMSNorm submodule registration is a model-initialization bug; Flux batch inference support is a separate feature/behavior issue." + }, + { + "left": "issue:175", + "right": "issue:3159", + "accept": false, + "reason": "A training-code seeding question is not the same as a documentation typo." + }, + { + "left": "issue:181", + "right": "issue:3453", + "accept": false, + "reason": "Image writing failure in unconditional training is distinct from a bfloat16 CUDA kernel limitation in text-to-image training." + }, + { + "left": "issue:2332", + "right": "issue:2676", + "accept": false, + "reason": "Dreambooth inference errors and scheduler-name lookup questions are unrelated." + }, + { + "left": "issue:3380", + "right": "issue:7165", + "accept": false, + "reason": "Hard-coded embedding size and JAX dynamic-thresholding scale are different components with different failure modes." + }, + { + "left": "issue:4292", + "right": "issue:5799", + "accept": false, + "reason": "M1 inpainting crashes are not the same as a strength keyword-argument API issue." + }, + { + "left": "issue:4615", + "right": "issue:9181", + "accept": false, + "reason": "Importing a missing pipeline class is unrelated to a ControlNet noise-artifact bug." + }, + { + "left": "issue:4709", + "right": "issue:7931", + "accept": false, + "reason": "Guess mode support in SDXL ControlNet is different from IP-Adapter embed handling in the inpaint pipeline." + }, + { + "left": "issue:4753", + "right": "issue:8534", + "accept": false, + "reason": "Both touch training scripts, but they target different scripts and different bugs." + }, + { + "left": "issue:5315", + "right": "issue:6634", + "accept": false, + "reason": "Both involve DDIMInverseScheduler, but one is a step-function bug and the other is about use_clipped_model_output; not clearly the same defect." + }, + { + "left": "issue:6732", + "right": "issue:7341", + "accept": false, + "reason": "Missing documentation for trained_betas is unrelated to a TorchVersion/Version comparison TypeError." + }, + { + "left": "issue:7009", + "right": "issue:9822", + "accept": false, + "reason": "Poor results from a text-to-video pipeline are not the same as loading SDXL LoRAs into Flux." + }, + { + "left": "issue:10121", + "right": "issue:6162", + "accept": false, + "reason": "Flux high-resolution generation limitations and SDXL IP-Adapter shape mismatch are different model/pipeline bugs." + }, + { + "left": "issue:4585", + "right": "issue:8975", + "accept": false, + "reason": "A documentation inconsistency about MultiControlNet is unrelated to an import error for LoraLoaderMixin." + }, + { + "left": "issue:2276", + "right": "issue:4872", + "accept": false, + "reason": "xformers installation failure and CUDA OOM during finetuning are separate problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 339, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10054", + "issue:10085", + "issue:10150", + "issue:10722", + "issue:1080", + "issue:11497", + "issue:11749", + "issue:1323", + "issue:13386", + "issue:1734", + "issue:1988", + "issue:2053", + "issue:2294", + "issue:2618", + "issue:3029", + "issue:3103", + "issue:3441", + "issue:3717", + "issue:4139", + "issue:4815", + "issue:4841", + "issue:4849", + "issue:5334", + "issue:5395", + "issue:5436", + "issue:5557", + "issue:5605", + "issue:5737", + "issue:5860", + "issue:6109", + "issue:617", + "issue:6335", + "issue:6392", + "issue:6399", + "issue:6482", + "issue:6527", + "issue:6732", + "issue:6921", + "issue:8503", + "issue:8556", + "issue:8579", + "issue:8615", + "issue:8708", + "issue:8850", + "issue:8989", + "issue:9467", + "issue:9664", + "issue:9680" + ], + "soft_pairs": [ + "issue:11749|issue:13386", + "issue:1323|issue:2294", + "issue:2053|issue:3441", + "issue:5860|issue:8850", + "issue:11497|issue:8503", + "issue:3029|issue:5436", + "issue:3717|issue:4849", + "issue:6921|issue:9680", + "issue:8556|issue:8579", + "issue:3103|issue:4815", + "issue:4139|issue:6392", + "issue:1988|issue:617", + "issue:6109|issue:6335", + "issue:1080|issue:5334", + "issue:2618|issue:4841", + "issue:5557|issue:6482", + "issue:5605|issue:8615", + "issue:5737|issue:9664", + "issue:6399|issue:6732", + "issue:8708|issue:9467", + "issue:10722|issue:1734", + "issue:5395|issue:8989", + "issue:10054|issue:6527", + "issue:10085|issue:10150" + ], + "prepared_review_unit_hash": "4088aa53ebc6ffa72f0787d4e512a2435c167f83bbb0eca5435dd7bc2b23614e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33852, + "estimated_input_tokens": 8463, + "estimated_eval_tokens": 17182 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All candidate pairs are weak title-level similarities rather than true duplicates. The set spans many unrelated bugs across schedulers, SD3/SDXL training, ControlNet, LoRA, and pipeline loading; none of the soft pairs appear to describe the same concrete code path.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 11497 is the most broadly phrased, concrete, and actionable bug report in the set: it identifies a specific SD3 ControlNet preprocessing/cache dependency problem that is clearly reproducible and could anchor related training-script issues.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11497 is the best standalone issue to keep because it is specific enough to triage, has a clear failure mechanism, and is broader than the more narrowly scoped or vaguer reports around it.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11749", + "right": "issue:13386", + "accept": false, + "reason": "Different domains and failures: scheduler floating-point timesteps vs DreamBooth prior-preservation tensor concatenation." + }, + { + "left": "issue:1323", + "right": "issue:2294", + "accept": false, + "reason": "Unrelated problems: Flax dtype loading vs missing export/import for a video inpaint pipeline." + }, + { + "left": "issue:2053", + "right": "issue:3441", + "accept": false, + "reason": "Both mention VRAM, but one is a version regression and the other is a PyTorch 2/xformers-specific memory issue." + }, + { + "left": "issue:5860", + "right": "issue:8850", + "accept": false, + "reason": "Different features: Tiny Autoencoder+ControlNet failure vs broken community pipeline mirrors." + }, + { + "left": "issue:11497", + "right": "issue:8503", + "accept": false, + "reason": "Unrelated SD3 training-script cache bug vs gradient-checkpointing crash." + }, + { + "left": "issue:3029", + "right": "issue:5436", + "accept": false, + "reason": "Both touch ControlNet, but one is custom resolution support and the other is an SDXL training example error." + }, + { + "left": "issue:3717", + "right": "issue:4849", + "accept": false, + "reason": "OOM during multi-GPU training is not the same as a general question about using multiple GPUs for textual inversion." + }, + { + "left": "issue:6921", + "right": "issue:9680", + "accept": false, + "reason": "Different issues: shipped SDXL weights diverging from official weights vs broken unconditional generation." + }, + { + "left": "issue:8556", + "right": "issue:8579", + "accept": false, + "reason": "Different SD3 failures: meta-tensor loading error vs nonfunctional LoRA support." + }, + { + "left": "issue:3103", + "right": "issue:4815", + "accept": false, + "reason": "Both involve inpainting, but one is output height/width handling and the other is latent-only input support." + }, + { + "left": "issue:4139", + "right": "issue:6392", + "accept": false, + "reason": "Different code paths: custom diffusion attention precision casting vs custom LoRA loading failure." + }, + { + "left": "issue:1988", + "right": "issue:617", + "accept": false, + "reason": "Unrelated failures: fp16/bf16 generation dtype mismatch vs NNPACK hardware support warning." + }, + { + "left": "issue:6109", + "right": "issue:6335", + "accept": false, + "reason": "Different ControlNet topics: depth-based ControlNet-xs behavior vs img2img output size parameter handling." + }, + { + "left": "issue:1080", + "right": "issue:5334", + "accept": false, + "reason": "A usage question about launching DreamBooth from Python is not the same as Accelerate needing multiple models passed to prepare/accumulate." + }, + { + "left": "issue:2618", + "right": "issue:4841", + "accept": false, + "reason": "Different problems: FloatTensor image input acceptance vs ControlNet image-size handling." + }, + { + "left": "issue:5557", + "right": "issue:6482", + "accept": false, + "reason": "Unrelated errors: unexpected use_reentrant kwarg vs resume-from-checkpoint FP16 unscale failure." + }, + { + "left": "issue:5605", + "right": "issue:8615", + "accept": false, + "reason": "Different feature requests: per-image captions for SDXL DreamBooth vs fuse_lora support for SD3." + }, + { + "left": "issue:5737", + "right": "issue:9664", + "accept": false, + "reason": "Different code paths: conversion script failure for SDXL vs DreamBooth LoRA training subprocess failure." + }, + { + "left": "issue:6399", + "right": "issue:6732", + "accept": false, + "reason": "Unrelated: LoRA visibility in UIs vs scheduler documentation for trained_betas." + }, + { + "left": "issue:8708", + "right": "issue:9467", + "accept": false, + "reason": "Different guardrails: missing SD3 VAE shift_factor in training examples vs num_frames validation for CogVideoX I2V." + }, + { + "left": "issue:10722", + "right": "issue:1734", + "accept": false, + "reason": "Different symptoms and likely causes: prior-preservation tensor mismatch vs slow multi-GPU DreamBooth training." + }, + { + "left": "issue:5395", + "right": "issue:8989", + "accept": false, + "reason": "Unrelated: PEFT LoRA-merging bug vs Stable Audio sequential offloading bug." + }, + { + "left": "issue:10054", + "right": "issue:6527", + "accept": false, + "reason": "Same scheduler family but different concrete bugs: wrong torch.linspace parameters in tests vs add_noise behavior." + }, + { + "left": "issue:10085", + "right": "issue:10150", + "accept": false, + "reason": "Different problems: LyCORIS/batch inference/skip_guidance_layers runtime error vs slow SDXL pipeline loading/inference." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 340, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10120", + "issue:10121", + "issue:10302", + "issue:10315", + "issue:11163", + "issue:11540", + "issue:11581", + "issue:11657", + "issue:11659", + "issue:2659", + "issue:2838", + "issue:2851", + "issue:3026", + "issue:3097", + "issue:3194", + "issue:3576", + "issue:3631", + "issue:3951", + "issue:3975", + "issue:4484", + "issue:4618", + "issue:5270", + "issue:5592", + "issue:5884", + "issue:5984", + "issue:6091", + "issue:6230", + "issue:6232", + "issue:6598", + "issue:6610", + "issue:7210", + "issue:7322", + "issue:7355", + "issue:8323", + "issue:8450", + "issue:8707", + "issue:8710", + "issue:8759", + "issue:898", + "issue:9106", + "issue:9163", + "issue:9292", + "issue:9361", + "issue:939", + "issue:9464", + "issue:9664", + "issue:9944" + ], + "soft_pairs": [ + "issue:2659|issue:6091", + "issue:6232|issue:8323", + "issue:11581|issue:11659", + "issue:10120|issue:6230", + "issue:11163|issue:9106", + "issue:11657|issue:11659", + "issue:5270|issue:6091", + "issue:10315|issue:8707", + "issue:2851|issue:3026", + "issue:4484|issue:6598", + "issue:10302|issue:11540", + "issue:3194|issue:7210", + "issue:8759|issue:9944", + "issue:10121|issue:7355", + "issue:2838|issue:939", + "issue:3097|issue:3576", + "issue:3631|issue:898", + "issue:3951|issue:5592", + "issue:3975|issue:4618", + "issue:5884|issue:8710", + "issue:5984|issue:9664", + "issue:6610|issue:8450", + "issue:9163|issue:9464", + "issue:9292|issue:9361", + "issue:5984|issue:7322" + ], + "prepared_review_unit_hash": "dfa7bcde0d1cfb1e382b4550d220e8657496e10b9658d7d2c579e46c3f03c763", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33915, + "estimated_input_tokens": 8479, + "estimated_eval_tokens": 17214 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose thematic cluster of diffusion/model bugs, not a true duplicate set. The soft pairs mostly share only broad subsystem similarity, so none should be merged as duplicates.", + "confidence": 0.61, + "canonical_issue_reason": "issue:11540 is the clearest, most concrete bug report in the set: it names an exact pipeline regression and a second closely related pipeline limitation, making it the best representative issue even though the cluster itself is heterogeneous.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11540 is the strongest standalone issue for triage because it is specific, reproducible, and actionable, with an explicit symptom and affected APIs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2659", + "right": "issue:6091", + "accept": false, + "reason": "Different problems: batch-size nondeterminism in SD vs consistency decoder memory usage with a size argument." + }, + { + "left": "issue:6232", + "right": "issue:8323", + "accept": false, + "reason": "Both are scheduler-related only in the broadest sense; one is a DPMSolver add_noise bug, the other is an unreadable scheduler filename issue." + }, + { + "left": "issue:11581", + "right": "issue:11659", + "accept": false, + "reason": "Both involve LoRA/IP-adapter style loading on FLUX/Wan, but the concrete failures and code paths differ." + }, + { + "left": "issue:10120", + "right": "issue:6230", + "accept": false, + "reason": "Dreambooth SD3 memory consumption is not the same as SDXL training out-of-memory." + }, + { + "left": "issue:11163", + "right": "issue:9106", + "accept": false, + "reason": "Wan image-to-video latent math bug is unrelated to PixArt time_step attribute errors." + }, + { + "left": "issue:11657", + "right": "issue:11659", + "accept": false, + "reason": "Both mention LoRA loading, but one is a silent Wan LoRA key mismatch and the other is a generic load issue with different context." + }, + { + "left": "issue:5270", + "right": "issue:6091", + "accept": false, + "reason": "Notebook memory leak and consistency decoder size-related memory issue are different symptoms and code paths." + }, + { + "left": "issue:10315", + "right": "issue:8707", + "accept": false, + "reason": "CogVideo training error and SDXL text-to-image training error are unrelated training failures." + }, + { + "left": "issue:2851", + "right": "issue:3026", + "accept": false, + "reason": "TPU push_to_hub failure is unrelated to instruct-pix2pix single-GPU launch errors." + }, + { + "left": "issue:4484", + "right": "issue:6598", + "accept": false, + "reason": "Weird SDXL Kohya LoRA results and img2img differences vs A1111 are different model/workflow behaviors." + }, + { + "left": "issue:10302", + "right": "issue:11540", + "accept": false, + "reason": "FP8 noise without CPU offloading is a precision/offloading issue, not the same as VAE slicing noise or missing slicing support." + }, + { + "left": "issue:3194", + "right": "issue:7210", + "accept": false, + "reason": "bitsandbytes packaging error is unrelated to accelerate dispatch on StableDiffusionXLPipeline." + }, + { + "left": "issue:8759", + "right": "issue:9944", + "accept": false, + "reason": "SD3 image_processor NaN warning and SD3.5 size mismatch are different failure modes." + }, + { + "left": "issue:10121", + "right": "issue:7355", + "accept": false, + "reason": "Flux high-resolution generation failure does not match Stable Cascade arbitrary-resolution+bfloat16 errors." + }, + { + "left": "issue:2838", + "right": "issue:939", + "accept": false, + "reason": "ControlNet pipeline support question is unrelated to using multiple community pipes in a list." + }, + { + "left": "issue:3097", + "right": "issue:3576", + "accept": false, + "reason": "Textual inversion doing nothing is not the same as the xformers TypeError during textual inversion." + }, + { + "left": "issue:3631", + "right": "issue:898", + "accept": false, + "reason": "Inpainting latents exception and TPU bfloat16 scalar type exception are different pipelines and errors." + }, + { + "left": "issue:3951", + "right": "issue:5592", + "accept": false, + "reason": "UniDiffuser NaNs in demo code and load-from-checkpoints failure are distinct issues." + }, + { + "left": "issue:3975", + "right": "issue:4618", + "accept": false, + "reason": "Safetensors error-message typo is unrelated to ControlNet usage guidance." + }, + { + "left": "issue:5884", + "right": "issue:8710", + "accept": false, + "reason": "IP-Adapter support expansion and PAG support request are different feature requests, not the same change." + }, + { + "left": "issue:5984", + "right": "issue:9664", + "accept": false, + "reason": "Colab system RAM issue in SDXL LoRA training and a subprocess error in a different training script are not the same bug." + }, + { + "left": "issue:6610", + "right": "issue:8450", + "accept": false, + "reason": "from_single_file num_in_channels ignored and SDXL inpainting strength=1.0 noise are different inpainting bugs." + }, + { + "left": "issue:9163", + "right": "issue:9464", + "accept": false, + "reason": "Tensor bugs in FLUX dreambooth LoRA training and a Prodigy/train_text_encoder index error are separate training failures." + }, + { + "left": "issue:9292", + "right": "issue:9361", + "accept": false, + "reason": "X-Labs FLUX LoRA key compatibility and FLUXPipeline LoRA effect missing are related only at a high level; the concrete incompatibilities differ." + }, + { + "left": "issue:5984", + "right": "issue:7322", + "accept": false, + "reason": "SDXL LoRA Colab RAM issue and dreambooth_lora_sdxl SIGKILL training error are both resource-related but not the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 341, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10710", + "issue:10798", + "issue:1084", + "issue:1106", + "issue:11092", + "issue:11374", + "issue:11447", + "issue:11448", + "issue:11790", + "issue:12373", + "issue:12606", + "issue:1730", + "issue:2299", + "issue:2414", + "issue:2442", + "issue:251", + "issue:2552", + "issue:2716", + "issue:307", + "issue:4031", + "issue:4043", + "issue:4186", + "issue:4399", + "issue:4476", + "issue:4641", + "issue:4697", + "issue:4874", + "issue:4977", + "issue:5081", + "issue:5135", + "issue:5412", + "issue:5553", + "issue:5763", + "issue:6008", + "issue:6176", + "issue:6636", + "issue:6686", + "issue:6783", + "issue:6921", + "issue:776", + "issue:7939", + "issue:8373", + "issue:8956", + "issue:9112", + "issue:9200", + "issue:9224", + "issue:9844" + ], + "soft_pairs": [ + "issue:6921|issue:9200", + "issue:10798|issue:9224", + "issue:2299|issue:4977", + "issue:1106|issue:2552", + "issue:11790|issue:12606", + "issue:251|issue:776", + "issue:4186|issue:6783", + "issue:4399|issue:5763", + "issue:10710|issue:8373", + "issue:6686|issue:9844", + "issue:1084|issue:307", + "issue:11374|issue:12373", + "issue:11447|issue:11448", + "issue:1730|issue:2442", + "issue:2414|issue:6636", + "issue:4031|issue:4874", + "issue:4476|issue:5412", + "issue:4641|issue:4697", + "issue:5135|issue:6008", + "issue:7939|issue:8956", + "issue:11092|issue:9112", + "issue:2716|issue:6176", + "issue:4031|issue:5553", + "issue:4043|issue:5081" + ], + "prepared_review_unit_hash": "7acc606cf2659159f5658406d6d68c8fac9b689fb10bead4be9d1049bdbb0667", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33186, + "estimated_input_tokens": 8297, + "estimated_eval_tokens": 16850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous and mostly false-positive grouping: it mixes unrelated loading, training, inference, docs, and feature-request issues. If a representative issue is needed, the broadest loader-related report is the best fit, but there is no strong true duplicate center.", + "confidence": 0.93, + "canonical_issue_reason": "issue:5553 is the broadest and most generic loader-failure report in the set, so it is the closest weak representative. That said, the cluster is not a cohesive duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:5553 best matches the widest subset of titles about model/pipeline loading failures, making it the most plausible cluster anchor despite low semantic cohesion.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6921", + "right": "issue:9200", + "accept": false, + "reason": "Official SDXL weight mismatch vs ControlNet dtype error; different bug classes and code paths." + }, + { + "left": "issue:10798", + "right": "issue:9224", + "accept": false, + "reason": "Bitsandbytes flux quantization/device-combo bug vs hardcoded float64 on MPS; both precision-related, but not the same issue." + }, + { + "left": "issue:2299", + "right": "issue:4977", + "accept": false, + "reason": "Checkpoint conversion/compatibility problem vs example cleanup about checkpoint removal; unrelated changes." + }, + { + "left": "issue:1106", + "right": "issue:2552", + "accept": false, + "reason": "PNDM repeatedly invoking UNet vs inpainting model tutorial question; different subjects entirely." + }, + { + "left": "issue:11790", + "right": "issue:12606", + "accept": false, + "reason": "LoRA weight loading failure vs wrong import in an SD3 conversion script; distinct failures." + }, + { + "left": "issue:251", + "right": "issue:776", + "accept": false, + "reason": "Latent distribution fragility vs batch_size>1 img2img breakage; both image generation issues, but not the same bug." + }, + { + "left": "issue:4186", + "right": "issue:6783", + "accept": false, + "reason": "SDXL inpaint/refiner failure vs PixArt generations corrupted; different models and failure modes." + }, + { + "left": "issue:4399", + "right": "issue:5763", + "accept": false, + "reason": "torch.compile plus CPU offload incompatibility vs save_pretrained not writing safetensors; unrelated subsystems." + }, + { + "left": "issue:10710", + "right": "issue:8373", + "accept": false, + "reason": "Feature request for DDUF support vs incorrect error message; not the same underlying change." + }, + { + "left": "issue:6686", + "right": "issue:9844", + "accept": false, + "reason": "Dynamo compile perf regression after save/load vs VAE encoder NaNs; different modules and symptoms." + }, + { + "left": "issue:1084", + "right": "issue:307", + "accept": false, + "reason": "Documentation around weight reuse vs merging pipelines into one; related theme, but not one concrete bug/change." + }, + { + "left": "issue:11374", + "right": "issue:12373", + "accept": false, + "reason": "SkyReels-V2 integration request vs Unlearn API request; separate feature proposals." + }, + { + "left": "issue:11447", + "right": "issue:11448", + "accept": false, + "reason": "Offline checkpoint shard lookup bug vs missing fusing/stable-unclip model path; both loading-related, but different concrete problems." + }, + { + "left": "issue:1730", + "right": "issue:2442", + "accept": false, + "reason": "Model conversion to CKPT/webui compatibility vs training loss-graph question; different workflows and issues." + }, + { + "left": "issue:2414", + "right": "issue:6636", + "accept": false, + "reason": "Missing transformers import during hub load vs missing DIFFUSERS_SLOW_IMPORT symbol; distinct import failures." + }, + { + "left": "issue:4031", + "right": "issue:4874", + "accept": false, + "reason": "Local safety-checker load failure vs low_res_scheduler bug when loading from local; different components." + }, + { + "left": "issue:4476", + "right": "issue:5412", + "accept": false, + "reason": "Inference quality issue with SDXL+ControlNet vs training resume_from_checkpoint failure; unrelated code paths." + }, + { + "left": "issue:4641", + "right": "issue:4697", + "accept": false, + "reason": "Local-file LoRA loading regression vs specific LoRA failing after an update; same broad area, but too unspecific to treat as the same bug." + }, + { + "left": "issue:5135", + "right": "issue:6008", + "accept": false, + "reason": "LoRA quality degradation possibly from xformers vs IP-Adapter Plus Face failing with LCM LoRA; different features and symptoms." + }, + { + "left": "issue:7939", + "right": "issue:8956", + "accept": false, + "reason": "Same-folder model load failure vs OVStableDiffusionXLPipeline model load failure; both loading issues, but not the same path or error." + }, + { + "left": "issue:11092", + "right": "issue:9112", + "accept": false, + "reason": "SDXL Img2Img empty timesteps bug vs Auraflow tensor-size mismatch; unrelated pipelines and errors." + }, + { + "left": "issue:2716", + "right": "issue:6176", + "accept": false, + "reason": "General question about loading checkpoints vs error loading LoRA after DreamBooth training; not a duplicate." + }, + { + "left": "issue:4031", + "right": "issue:5553", + "accept": false, + "reason": "Safety checker local-file load failure vs custom model load failure; both are loader bugs, but different components and failure modes." + }, + { + "left": "issue:4043", + "right": "issue:5081", + "accept": false, + "reason": "Missing pooled prompt embeddings for SDXL vs LPW XL matrix-shape mismatch; both shape-related, but different model/input requirements." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 342, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10695", + "issue:10768", + "issue:10772", + "issue:10893", + "issue:12761", + "issue:2201", + "issue:2392", + "issue:2550", + "issue:272", + "issue:2991", + "issue:3181", + "issue:3403", + "issue:3467", + "issue:3578", + "issue:3624", + "issue:3643", + "issue:3682", + "issue:3876", + "issue:3935", + "issue:3951", + "issue:4001", + "issue:4118", + "issue:4304", + "issue:4306", + "issue:4675", + "issue:4804", + "issue:5043", + "issue:5365", + "issue:5794", + "issue:5875", + "issue:6335", + "issue:6598", + "issue:7237", + "issue:725", + "issue:7286", + "issue:7736", + "issue:7813", + "issue:8390", + "issue:8503", + "issue:8667", + "issue:8855", + "issue:8898", + "issue:8930", + "issue:9225", + "issue:9460", + "issue:9539", + "issue:9756" + ], + "soft_pairs": [ + "issue:8667|issue:9460", + "issue:10695|issue:12761", + "issue:10768|issue:10772", + "issue:10893|issue:8390", + "issue:2201|issue:6598", + "issue:2392|issue:3951", + "issue:2550|issue:5875", + "issue:272|issue:4304", + "issue:3181|issue:3624", + "issue:3578|issue:4118", + "issue:3682|issue:725", + "issue:3935|issue:4001", + "issue:4675|issue:8390", + "issue:4804|issue:9225", + "issue:5043|issue:7736", + "issue:5794|issue:8898", + "issue:7237|issue:9756", + "issue:7813|issue:8855", + "issue:8503|issue:9539", + "issue:3876|issue:4306", + "issue:6335|issue:8930", + "issue:3643|issue:5365", + "issue:3403|issue:7286", + "issue:2991|issue:3467" + ], + "prepared_review_unit_hash": "89b9f6f46c6c1d3cb06f448d113ff949f78f4cd9062df905a25a02529c244dd6", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33323, + "estimated_input_tokens": 8331, + "estimated_eval_tokens": 16918 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the soft pairs mostly connect unrelated bugs/features across different pipelines, schedulers, loaders, and docs. None look like true duplicates of the same underlying change or defect.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue is a good canonical here; the items span unrelated topics, so duplicate triage should stay separate.", + "canonical_pr_reason": "No pull requests are present in this cluster.", + "best_issue_reason": "No issue stands out as a global best representative because the cluster does not share one concrete bug or feature area.", + "best_pr_reason": "No pull requests are present in this cluster.", + "soft_edge_verdicts": [ + { + "left": "issue:8667", + "right": "issue:9460", + "accept": false, + "reason": "Different problems: SD3ControlNet tensor-size mismatch vs adding list support for multiple ControlNet models." + }, + { + "left": "issue:10695", + "right": "issue:12761", + "accept": false, + "reason": "Scheduler timestep calculation bug vs custom attention backward gradient-count error; unrelated code paths." + }, + { + "left": "issue:10768", + "right": "issue:10772", + "accept": false, + "reason": "Negative-prompt support question for Hunyuan video vs Sana ControlNet support request." + }, + { + "left": "issue:10893", + "right": "issue:8390", + "accept": false, + "reason": "CogVideoX ControlNet integration request vs Stable Cascade ControlNet training request; different features/models." + }, + { + "left": "issue:2201", + "right": "issue:6598", + "accept": false, + "reason": "Tokenizer padding behavior mismatch vs img2img prompt_embeds reproducibility; not the same defect." + }, + { + "left": "issue:2392", + "right": "issue:3951", + "accept": false, + "reason": "Img2img prompt_embeds validation error vs UniDiffuser NaNs; different pipelines and symptoms." + }, + { + "left": "issue:2550", + "right": "issue:5875", + "accept": false, + "reason": "ControlNet-to-diffusers conversion issue vs ONNX conversion prompt error; different conversion paths." + }, + { + "left": "issue:272", + "right": "issue:4304", + "accept": false, + "reason": "K-LMS inpainting support vs SDXL LoRA training support; unrelated feature requests." + }, + { + "left": "issue:3181", + "right": "issue:3624", + "accept": false, + "reason": "TensorRT example TypeError vs generic getattr attribute-name TypeError; different failures." + }, + { + "left": "issue:3578", + "right": "issue:4118", + "accept": false, + "reason": "AutoencoderKL checkpoint loading failure vs missing from_single_file attribute; related component, different bug." + }, + { + "left": "issue:3682", + "right": "issue:725", + "accept": false, + "reason": "LoRA loading weights vs device_map/accelerate loading memory footprint; separate loading concerns." + }, + { + "left": "issue:3935", + "right": "issue:4001", + "accept": false, + "reason": "Docstrings for feed-forward chunking vs README demo-space docs; documentation tasks are unrelated." + }, + { + "left": "issue:4675", + "right": "issue:8390", + "accept": false, + "reason": "SDXL reference-only ControlNet feature vs Stable Cascade ControlNet training; different model families." + }, + { + "left": "issue:4804", + "right": "issue:9225", + "accept": false, + "reason": "LoRA loader iteration bug vs rgba latent size mismatch; different runtime issues." + }, + { + "left": "issue:5043", + "right": "issue:7736", + "accept": false, + "reason": "W\u00fcrstchen LoRA finetuning request vs supported callbacks API; not the same change." + }, + { + "left": "issue:5794", + "right": "issue:8898", + "accept": false, + "reason": "Missing tokenizer/config.json in SDXL base vs from_pretrained subfolder path resolution error; different loading bugs." + }, + { + "left": "issue:7237", + "right": "issue:9756", + "accept": false, + "reason": "Missing EDMEulerScheduler class vs missing notebook file; unrelated." + }, + { + "left": "issue:7813", + "right": "issue:8855", + "accept": false, + "reason": "Question about passing timesteps as tensors vs joint_attention_kwargs not reaching the AttentionProcessor." + }, + { + "left": "issue:8503", + "right": "issue:9539", + "accept": false, + "reason": "SD3 gradient checkpointing crash vs CogVideoX Float8 index_select error; different model/runtime issues." + }, + { + "left": "issue:3876", + "right": "issue:4306", + "accept": false, + "reason": "ControlNet inpaint demo error vs StableDiffusionReferencePipeline list-index error; different pipelines." + }, + { + "left": "issue:6335", + "right": "issue:8930", + "accept": false, + "reason": "Output-size control in ControlNet img2img vs pose-conditioning not being respected; both ControlNet but not the same bug." + }, + { + "left": "issue:3643", + "right": "issue:5365", + "accept": false, + "reason": "Latent decode INT_MAX overflow vs txt2video num_inference_steps error; unrelated symptoms." + }, + { + "left": "issue:3403", + "right": "issue:7286", + "accept": false, + "reason": "TorchDynamo pipeline-loading issue vs custom diffusion cross-attention weight loading; different components." + }, + { + "left": "issue:2991", + "right": "issue:3467", + "accept": false, + "reason": "Mixed-precision ControlNet example failure vs wrong-step resume with gradient accumulation; distinct training bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 343, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1205", + "issue:1214", + "issue:1271", + "issue:1617", + "issue:2326", + "issue:234", + "issue:235", + "issue:2364", + "issue:2596", + "issue:2606", + "issue:2763", + "issue:2782", + "issue:2880", + "issue:2905", + "issue:2908", + "issue:3081", + "issue:3222", + "issue:3224", + "issue:3453", + "issue:3573", + "issue:3591", + "issue:3740", + "issue:3884", + "issue:4185", + "issue:4616", + "issue:4709", + "issue:4826", + "issue:4936", + "issue:5538", + "issue:5605", + "issue:5630", + "issue:6162", + "issue:6164", + "issue:6287", + "issue:6336", + "issue:6354", + "issue:6520", + "issue:7168", + "issue:7300", + "issue:7415", + "issue:7483", + "issue:7600", + "issue:8226", + "issue:8551", + "issue:8659", + "issue:9001", + "issue:9610", + "issue:970" + ], + "soft_pairs": [ + "issue:1214|issue:2908", + "issue:4185|issue:8659", + "issue:2763|issue:4826", + "issue:1205|issue:2782", + "issue:2905|issue:6354", + "issue:4616|issue:970", + "issue:5630|issue:9001", + "issue:6162|issue:6336", + "issue:2606|issue:3884", + "issue:1617|issue:2326", + "issue:4936|issue:6520", + "issue:7168|issue:8551", + "issue:4709|issue:6164", + "issue:235|issue:2880", + "issue:2596|issue:3081", + "issue:3222|issue:7300", + "issue:3224|issue:7600", + "issue:3453|issue:7415", + "issue:5538|issue:6287", + "issue:5605|issue:9610", + "issue:1271|issue:2364", + "issue:3740|issue:7483", + "issue:234|issue:3573", + "issue:3591|issue:8226" + ], + "prepared_review_unit_hash": "719d2a95506d5dc90be907cf843f56898f638baf0b3af74c3735b15c78ddb77f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33801, + "estimated_input_tokens": 8451, + "estimated_eval_tokens": 17158 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most items are distinct bugs or feature requests spanning conversion, training scripts, ControlNet, schedulers, and docs. There is only loose thematic overlap around precision/memory/runtime issues, but no true duplicate pair stands out.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 234 is the broadest representative of the half-precision/runtime bug theme and is the closest fit as a canonical issue, though the overall set is not a tight duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 234 is the most general user-facing failure report here and overlaps thematically with several precision-related complaints better than the narrower, script-specific issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1214", + "right": "issue:2908", + "accept": false, + "reason": "Both mention mixed/half precision, but one is CLIPTextModel inference and the other is a ControlNet training-script shape issue; different code paths." + }, + { + "left": "issue:4185", + "right": "issue:8659", + "accept": false, + "reason": "SDXL/ControlNet context overlaps, but one is black-image inference and the other is checkpoint resume/load behavior in a training script." + }, + { + "left": "issue:2763", + "right": "issue:4826", + "accept": false, + "reason": "Both are ControlNet-related, but one is VRAM usage with cudnn benchmarking and the other is MultiControlNet reference handling errors; not the same bug." + }, + { + "left": "issue:1205", + "right": "issue:2782", + "accept": false, + "reason": "One is SD-to-ONNX conversion and the other is a sequential offloading error; different features and failure modes." + }, + { + "left": "issue:2905", + "right": "issue:6354", + "accept": false, + "reason": "A scheduler feature request and a ControlNet LoRA support request are unrelated change requests." + }, + { + "left": "issue:4616", + "right": "issue:970", + "accept": false, + "reason": "LoRA conversion for SDXL and failed integration tests are unrelated; no shared concrete code-path bug." + }, + { + "left": "issue:5630", + "right": "issue:9001", + "accept": false, + "reason": "Both concern pipeline behavior, but one is LMS noise in SDXL Img2Img and the other is Auraflow aspect ratios; different models and bugs." + }, + { + "left": "issue:6162", + "right": "issue:6336", + "accept": false, + "reason": "IP-Adapter shape mismatch in SDXL is unrelated to TensorRT VAE compilation with AutoencoderTiny." + }, + { + "left": "issue:2606", + "right": "issue:3884", + "accept": false, + "reason": "Both involve training scripts failing, but one is DeepSpeed save_state hang and the other is a different LoRA training-script error." + }, + { + "left": "issue:1617", + "right": "issue:2326", + "accept": false, + "reason": "Both are conversion/training utility issues, but the first is missing required CLI args and the second is LoRA safetensors/ckpt conversion output errors." + }, + { + "left": "issue:4936", + "right": "issue:6520", + "accept": false, + "reason": "Both are DreamBooth/adapter training reports, but one is resume broken in Dreambooth LoRA and the other is blank validation output in T2I adapter training." + }, + { + "left": "issue:7168", + "right": "issue:8551", + "accept": false, + "reason": "Both mention image embeddings/latents, but one is about IP-Adapter embedding tensor shape and the other is VAE device placement in SDXL inpaint." + }, + { + "left": "issue:4709", + "right": "issue:6164", + "accept": false, + "reason": "One is SDXL ControlNet guess-mode support and the other is latent output handling in StableDiffusionPipeline; feature gap vs bug in different paths." + }, + { + "left": "issue:235", + "right": "issue:2880", + "accept": false, + "reason": "Both mention runtime portability/cuda-like environment issues, but one is a standalone Python CUDA problem and the other is a JAX colab execution failure." + }, + { + "left": "issue:2596", + "right": "issue:3081", + "accept": false, + "reason": "Docs referencing a missing utility function and a broken Textual Inversion Colab are separate issues." + }, + { + "left": "issue:3222", + "right": "issue:7300", + "accept": false, + "reason": "Missing custom diffusion weights after training and high GPU memory after changing attention processor are different bugs." + }, + { + "left": "issue:3224", + "right": "issue:7600", + "accept": false, + "reason": "Learning-rate scaling duplication in DreamBooth LoRA and missing accelerator.accumulate on text_encoder are unrelated training-script defects." + }, + { + "left": "issue:3453", + "right": "issue:7415", + "accept": false, + "reason": "Both are precision-related, but one is a CUDA bfloat16 matmul/triu issue in text-to-image training and the other is an AutoencoderKL TPU bfloat16 bug." + }, + { + "left": "issue:5538", + "right": "issue:6287", + "accept": false, + "reason": "A pipeline implementation question and setup.py dependency cleanup are unrelated." + }, + { + "left": "issue:5605", + "right": "issue:9610", + "accept": false, + "reason": "Per-image caption support for SDXL DreamBooth and adding load_lora_weights to SD3 img2img are distinct feature requests." + }, + { + "left": "issue:1271", + "right": "issue:2364", + "accept": false, + "reason": "Both are install/run problems, but one is dependency installation on GCP notebooks and the other is a generic training-script launch failure." + }, + { + "left": "issue:3740", + "right": "issue:7483", + "accept": false, + "reason": "Both involve conversion scripts, but one is legacy attention block conversion with accelerate device_map and the other is LoRA-to-Diffusers for SDXL 1.0 base models." + }, + { + "left": "issue:234", + "right": "issue:3573", + "accept": false, + "reason": "Half-precision pipeline failure versus a local upscaler class_labels requirement; different pipelines and error causes." + }, + { + "left": "issue:3591", + "right": "issue:8226", + "accept": false, + "reason": "ControlNet loss-curve confusion is a training/UX question, while VectorQuantizer Loss is a separate model-loss issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 344, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10016", + "issue:1055", + "issue:1099", + "issue:1111", + "issue:11144", + "issue:11749", + "issue:1273", + "issue:1821", + "issue:2068", + "issue:2097", + "issue:2228", + "issue:2250", + "issue:2578", + "issue:3222", + "issue:3539", + "issue:3970", + "issue:4618", + "issue:4676", + "issue:4726", + "issue:4972", + "issue:5075", + "issue:5150", + "issue:5162", + "issue:540", + "issue:5405", + "issue:5630", + "issue:5919", + "issue:5931", + "issue:5968", + "issue:5998", + "issue:6321", + "issue:6439", + "issue:6446", + "issue:6561", + "issue:6685", + "issue:6811", + "issue:6966", + "issue:7049", + "issue:7535", + "issue:7616", + "issue:7762", + "issue:7939", + "issue:7945", + "issue:851", + "issue:9069", + "issue:9187", + "issue:9613" + ], + "soft_pairs": [ + "issue:1821|issue:3222", + "issue:2097|issue:5075", + "issue:4726|issue:7535", + "issue:11144|issue:9187", + "issue:4676|issue:7049", + "issue:5931|issue:7945", + "issue:1111|issue:540", + "issue:2228|issue:6446", + "issue:5162|issue:5968", + "issue:5630|issue:7762", + "issue:5998|issue:9069", + "issue:5919|issue:6966", + "issue:6561|issue:6811", + "issue:10016|issue:7939", + "issue:2068|issue:851", + "issue:2578|issue:6321", + "issue:2250|issue:4972", + "issue:1055|issue:1099", + "issue:11749|issue:9613", + "issue:1273|issue:3539", + "issue:3970|issue:4618", + "issue:5150|issue:6685", + "issue:5405|issue:6439", + "issue:5630|issue:7616" + ], + "prepared_review_unit_hash": "0f24b2018100e0bacf406d1f8c28ff20ac695639c1be2a064658e7ce2d8dc53b", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33194, + "estimated_input_tokens": 8299, + "estimated_eval_tokens": 16854 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated issue reports that only share superficial diffusers/model-loading vocabulary. I do not see any true duplicate pairs among the soft edges.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 10016 is the most concrete, self-contained bug report in the set and the best representative of the cluster\u2019s strongest theme (model loading / single-file shape mismatch).", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10016 is the best standalone issue to keep because it describes a specific reproducible failure and is more actionable than the many questions, feature requests, or broad symptom reports here.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1821", + "right": "issue:3222", + "accept": false, + "reason": "Negative prompt behavior in inpaint is unrelated to missing custom diffusion output files." + }, + { + "left": "issue:2097", + "right": "issue:5075", + "accept": false, + "reason": "UNet1D usage problems and SDXL refiner prompt-embedding shape mismatch are different code paths and failures." + }, + { + "left": "issue:4726", + "right": "issue:7535", + "accept": false, + "reason": "A callback type-hint issue is unrelated to a tensor type-hint incompatibility." + }, + { + "left": "issue:11144", + "right": "issue:9187", + "accept": false, + "reason": "Flax UNet dtype initialization and SDXL IP-Adapter image encoder handling are distinct bugs." + }, + { + "left": "issue:4676", + "right": "issue:7049", + "accept": false, + "reason": "TAESD latent brightness and IP-Adapter unpacking errors are unrelated runtime problems." + }, + { + "left": "issue:5931", + "right": "issue:7945", + "accept": false, + "reason": "Generation nondeterminism is not the same as a community-pipeline composition request." + }, + { + "left": "issue:1111", + "right": "issue:540", + "accept": false, + "reason": "A checkpoint licensing/data question is unrelated to memory-placement behavior for Stable Diffusion." + }, + { + "left": "issue:2228", + "right": "issue:6446", + "accept": false, + "reason": "Pipeline GPU placement/loading issues are similar at a broad level but not the same concrete failure." + }, + { + "left": "issue:5162", + "right": "issue:5968", + "accept": false, + "reason": "A checkpoint-to-safetensors converter request is unrelated to video-to-ONNX export." + }, + { + "left": "issue:5630", + "right": "issue:7762", + "accept": false, + "reason": "SDXL img2img noise leftover and generic bad results are too broad to be the same bug." + }, + { + "left": "issue:5998", + "right": "issue:9069", + "accept": false, + "reason": "VersatileDiffusion argument mismatch and NumPy type error are different incompatibilities." + }, + { + "left": "issue:5919", + "right": "issue:6966", + "accept": false, + "reason": "Missing LoRA layer support on Linear and missing attn_processors after forward pass are different model-internal issues." + }, + { + "left": "issue:6561", + "right": "issue:6811", + "accept": false, + "reason": "IP-Adapter local config-file lookup and i2vgen-xl model_index.json not found are separate missing-file/load-path failures." + }, + { + "left": "issue:10016", + "right": "issue:7939", + "accept": false, + "reason": "Single-file SD3.5M shape mismatch is not the same as failing to load a model from the same folder." + }, + { + "left": "issue:2068", + "right": "issue:851", + "accept": false, + "reason": "from_pretrained dtype conversion complaint and a Half-vs-Float mismatch are related only loosely, not the same bug." + }, + { + "left": "issue:2578", + "right": "issue:6321", + "accept": false, + "reason": "Editable install failure and CLIP model detection failure are unrelated environment/setup problems." + }, + { + "left": "issue:2250", + "right": "issue:4972", + "accept": false, + "reason": "Dreambooth prompt generation failure is not the same as a wuerstchen example runtime failure." + }, + { + "left": "issue:1055", + "right": "issue:1099", + "accept": false, + "reason": "Missing LatentDiffusionPipeline and missing dreambooth class name are different missing-API issues." + }, + { + "left": "issue:11749", + "right": "issue:9613", + "accept": false, + "reason": "Floating-point timestep errors and a missing scale attribute on Timesteps are different scheduler bugs." + }, + { + "left": "issue:1273", + "right": "issue:3539", + "accept": false, + "reason": "A mean-loss calculation question is not the same as a question about main_process_first semantics." + }, + { + "left": "issue:3970", + "right": "issue:4618", + "accept": false, + "reason": "SDXL prompt embeddings not working and using a specific checkpoint with ControlNet are not the same underlying problem." + }, + { + "left": "issue:5150", + "right": "issue:6685", + "accept": false, + "reason": "Missing from_single_file support and unsupported absolute output paths in Advanced Dreambooth are unrelated." + }, + { + "left": "issue:5405", + "right": "issue:6439", + "accept": false, + "reason": "SDXL from_single_file overwriting weights and a broken docs link are not the same issue." + }, + { + "left": "issue:5630", + "right": "issue:7616", + "accept": false, + "reason": "Both concern SDXL img2img quality, but one is scheduler-specific noise and the other is low-denoise artifacts; not identical enough." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 345, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1070", + "issue:11088", + "issue:11363", + "issue:11547", + "issue:11878", + "issue:12060", + "issue:12082", + "issue:12113", + "issue:12900", + "issue:13065", + "issue:1363", + "issue:1403", + "issue:1541", + "issue:1730", + "issue:1883", + "issue:1918", + "issue:1969", + "issue:2026", + "issue:2761", + "issue:3171", + "issue:347", + "issue:4701", + "issue:4773", + "issue:4919", + "issue:5425", + "issue:547", + "issue:5574", + "issue:5578", + "issue:6242", + "issue:6297", + "issue:6412", + "issue:6441", + "issue:6506", + "issue:7054", + "issue:7147", + "issue:7183", + "issue:7322", + "issue:7450", + "issue:759", + "issue:7782", + "issue:7946", + "issue:8906", + "issue:9515", + "issue:9516", + "issue:9648", + "issue:9660", + "issue:9670", + "issue:9836" + ], + "soft_pairs": [ + "issue:6242|issue:6441", + "issue:7054|issue:9660", + "issue:7782|issue:9670", + "issue:8906|issue:9515", + "issue:11088|issue:12082", + "issue:1363|issue:1918", + "issue:4701|issue:7450", + "issue:6297|issue:7322", + "issue:1883|issue:759", + "issue:4773|issue:4919", + "issue:5574|issue:9516", + "issue:5578|issue:7946", + "issue:11363|issue:12900", + "issue:6506|issue:9648", + "issue:1403|issue:2026", + "issue:1541|issue:1730", + "issue:1969|issue:5425", + "issue:7147|issue:7183", + "issue:11878|issue:12113", + "issue:1070|issue:347", + "issue:11547|issue:13065", + "issue:12060|issue:9836", + "issue:2761|issue:547", + "issue:3171|issue:6412" + ], + "prepared_review_unit_hash": "c42fc934543e499c8cbf2f27516992defc4010e728707c25499ffbb8a5287854", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33821, + "estimated_input_tokens": 8456, + "estimated_eval_tokens": 17168 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The items are a heterogeneous set of unrelated bug reports and feature/support questions connected only by shallow title similarity. None of the soft-paired artifacts look like the same underlying issue, so all soft edges should be rejected and there is no good global canonical issue/PR for this cluster.", + "confidence": 0.98, + "canonical_issue_reason": "No single issue is a valid canonical representative: the cluster mixes unrelated topics (loading errors, OOMs, controlnet, LoRA, schedulers, docs, features) with no evidence of one shared bug.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is globally suitable as the cluster anchor because the soft links are lexical-only and do not indicate one underlying defect shared by the group.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6242", + "right": "issue:6441", + "accept": false, + "reason": "Unrelated topics: a dataset log request versus a from_pretrained/Colab loading failure." + }, + { + "left": "issue:7054", + "right": "issue:9660", + "accept": false, + "reason": "Different bugs: LoRA state_dict mutation versus modules not being set to eval mode." + }, + { + "left": "issue:7782", + "right": "issue:9670", + "accept": false, + "reason": "Unrelated loading/conversion error versus LoRA backend requirement error." + }, + { + "left": "issue:8906", + "right": "issue:9515", + "accept": false, + "reason": "Different SD3 transformer concerns: absence of qk_norm versus RMSNorm in JointAttnProcessor2_0." + }, + { + "left": "issue:11088", + "right": "issue:12082", + "accept": false, + "reason": "Different model/load issues: GGUF load error versus VAE memory increase after compile." + }, + { + "left": "issue:1363", + "right": "issue:1918", + "accept": false, + "reason": "Both mention schedulers, but one is an ONNX long-prompt error and the other is a DDIM example issue; not the same bug." + }, + { + "left": "issue:4701", + "right": "issue:7450", + "accept": false, + "reason": "Both are AutoencoderKL-related, but one is a CUDA differentiation error and the other is a multithreading issue." + }, + { + "left": "issue:6297", + "right": "issue:7322", + "accept": false, + "reason": "Different training failures: SDXL Dreambooth training behavior versus SIGKILL during a specific script/version combo." + }, + { + "left": "issue:1883", + "right": "issue:759", + "accept": false, + "reason": "Gradient checkpointing for VAE is unrelated to optional full-model saving for textual inversion." + }, + { + "left": "issue:4773", + "right": "issue:4919", + "accept": false, + "reason": "Loading a CivitAI LoRA is not the same as adding multi-LoRA fuse support." + }, + { + "left": "issue:5574", + "right": "issue:9516", + "accept": false, + "reason": "Different LoRA/attention topics: training partial UNet attention parameters versus FLUX joint_attention_kwargs propagation." + }, + { + "left": "issue:5578", + "right": "issue:7946", + "accept": false, + "reason": "Model_index.json loading failure is unrelated to custom_pipeline caching." + }, + { + "left": "issue:11363", + "right": "issue:12900", + "accept": false, + "reason": "Both are OOM reports, but for different models and scenarios: ControlNet inference versus Qwen-Image at 2048px." + }, + { + "left": "issue:6506", + "right": "issue:9648", + "accept": false, + "reason": "Different memory/output issues: StableVideoDiffusion output type bug versus VAE tiling causing overflow." + }, + { + "left": "issue:1403", + "right": "issue:2026", + "accept": false, + "reason": "Safety checker output mismatch is unrelated to cpu_offload not working in PaintByExample." + }, + { + "left": "issue:1541", + "right": "issue:1730", + "accept": false, + "reason": "Related to Dreambooth/CKPT loading, but one is MODEL_TYPE detection and the other is a converted-model compatibility question." + }, + { + "left": "issue:1969", + "right": "issue:5425", + "accept": false, + "reason": "Tensor view-stride runtime error is unrelated to checkpoint saving compatibility with accelerate/PT 2.1." + }, + { + "left": "issue:7147", + "right": "issue:7183", + "accept": false, + "reason": "Importing StableCascadeDecoderPipeline is unrelated to scheduler.config reporting the wrong class name." + }, + { + "left": "issue:11878", + "right": "issue:12113", + "accept": false, + "reason": "Both involve Wan models, but one is a GGUF 1.3B load problem and the other is a channel-mismatch runtime error in WanImageToVideoPipeline." + }, + { + "left": "issue:1070", + "right": "issue:347", + "accept": false, + "reason": "Stable diffusion inpainting bug is unrelated to installing diffusers/huggingface-hub dependency failures." + }, + { + "left": "issue:11547", + "right": "issue:13065", + "accept": false, + "reason": "Different ControlNet/Qwen-Image feature requests; one is Flux fill dev support, the other is Qwen-Image inpaint+canny support." + }, + { + "left": "issue:12060", + "right": "issue:9836", + "accept": false, + "reason": "A question about DiT block definitions is unrelated to recording layer_id for DiT models." + }, + { + "left": "issue:2761", + "right": "issue:547", + "accept": false, + "reason": "Documentation for UNet1DModel.extra_in_channels is not the same as general stable diffusion local loading docs." + }, + { + "left": "issue:3171", + "right": "issue:6412", + "accept": false, + "reason": "LPW prompt support for ControlNet is unrelated to converting ControlNet annotators to ONNX." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 346, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10381", + "issue:10526", + "issue:10782", + "issue:10920", + "issue:10928", + "issue:11055", + "issue:11083", + "issue:11447", + "issue:11872", + "issue:1197", + "issue:12433", + "issue:1284", + "issue:13256", + "issue:1865", + "issue:1922", + "issue:2234", + "issue:2439", + "issue:3287", + "issue:3574", + "issue:3643", + "issue:3728", + "issue:3853", + "issue:3908", + "issue:4236", + "issue:4250", + "issue:4263", + "issue:4363", + "issue:4468", + "issue:5064", + "issue:5395", + "issue:5989", + "issue:6297", + "issue:6366", + "issue:660", + "issue:7492", + "issue:7591", + "issue:7773", + "issue:7785", + "issue:7864", + "issue:7891", + "issue:7945", + "issue:7955", + "issue:8541", + "issue:8637", + "issue:8918", + "issue:9450", + "issue:9792" + ], + "soft_pairs": [ + "issue:3728|issue:7864", + "issue:4263|issue:8918", + "issue:4363|issue:5989", + "issue:5395|issue:7945", + "issue:7773|issue:9792", + "issue:3643|issue:7891", + "issue:6297|issue:6366", + "issue:1865|issue:2234", + "issue:2439|issue:660", + "issue:10920|issue:7785", + "issue:11872|issue:9450", + "issue:4236|issue:5064", + "issue:7864|issue:7955", + "issue:1922|issue:3574", + "issue:3853|issue:7591", + "issue:3908|issue:4250", + "issue:10381|issue:10526", + "issue:12433|issue:13256", + "issue:1284|issue:4468", + "issue:10928|issue:7492", + "issue:10782|issue:11055", + "issue:11083|issue:8637", + "issue:11447|issue:8541", + "issue:1197|issue:3287" + ], + "prepared_review_unit_hash": "00e3ba985f5ffa6e960c6ecf232c975893f98bb5f55315623eb1435c75940e11", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33538, + "estimated_input_tokens": 8385, + "estimated_eval_tokens": 17026 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a coherent duplicate cluster: the items span many unrelated bugs, docs, and feature requests. All soft edges look like false positives rather than the same underlying issue.", + "confidence": 0.94, + "canonical_issue_reason": "No canonical issue: the list spans unrelated bugs, docs, training, loading, scheduler, and performance reports, so there isn't one issue that safely represents a duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue: none of the items is a strong representative for the full set because the topics diverge too much.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3728", + "right": "issue:7864", + "accept": false, + "reason": "Different requests: a new text-to-image pipeline versus MotionMaster camera-motion transfer." + }, + { + "left": "issue:4263", + "right": "issue:8918", + "accept": false, + "reason": "Different bugs: LoRA unloading versus SD3 training script breakage." + }, + { + "left": "issue:4363", + "right": "issue:5989", + "accept": false, + "reason": "Both involve loading failures, but one is a model-loading question and the other is a safetensors library bug; not the same code-path." + }, + { + "left": "issue:5395", + "right": "issue:7945", + "accept": false, + "reason": "PEFT/LoRA combination bug versus a community-pipeline combination request." + }, + { + "left": "issue:7773", + "right": "issue:9792", + "accept": false, + "reason": "Different failure modes: Stable Video Diffusion batch-size issue versus dataset cache loading." + }, + { + "left": "issue:3643", + "right": "issue:7891", + "accept": false, + "reason": "Different runtime problems: latent upsampling tensor-size limit versus cross-device tensor placement." + }, + { + "left": "issue:6297", + "right": "issue:6366", + "accept": false, + "reason": "Different training issues: can't train on photos versus checkpoint restore/load logic." + }, + { + "left": "issue:1865", + "right": "issue:2234", + "accept": false, + "reason": "Different paths: ZeRO stage-3 finetuning dimension error versus xformers attention finetuning failure." + }, + { + "left": "issue:2439", + "right": "issue:660", + "accept": false, + "reason": "Inpainting pipeline loading versus MPS fp16/revision handling; not the same bug." + }, + { + "left": "issue:10920", + "right": "issue:7785", + "accept": false, + "reason": "Scheduler inf-values bug versus slow first-inference performance regression." + }, + { + "left": "issue:11872", + "right": "issue:9450", + "accept": false, + "reason": "Both mention GPU movement, but one is offloading inefficiency and the other is a device mismatch when passing a transformer." + }, + { + "left": "issue:4236", + "right": "issue:5064", + "accept": false, + "reason": "Refiner text-encoder registration issue versus out-of-space during inference." + }, + { + "left": "issue:7864", + "right": "issue:7955", + "accept": false, + "reason": "Different community models: camera motion transfer versus identity-preserving portrait generation." + }, + { + "left": "issue:1922", + "right": "issue:3574", + "accept": false, + "reason": "Notebook undefined-function bug versus Windows checkpoint conversion/loading issue." + }, + { + "left": "issue:3853", + "right": "issue:7591", + "accept": false, + "reason": "Similar error text, but the loading paths differ enough that this is not clearly the same underlying bug." + }, + { + "left": "issue:3908", + "right": "issue:4250", + "accept": false, + "reason": "Streamlit caching behavior versus incorrect from_pretrained model loading." + }, + { + "left": "issue:10381", + "right": "issue:10526", + "accept": false, + "reason": "Related Flux/Quanto area, but one is a broad inference report with LoRA/Zero GPU spaces while the other is a specific WeightQBytesTensor TypeError; not clearly one bug." + }, + { + "left": "issue:12433", + "right": "issue:13256", + "accept": false, + "reason": "Docs/version-compatibility complaint versus a generic package compatibility claim; not the same issue." + }, + { + "left": "issue:1284", + "right": "issue:4468", + "accept": false, + "reason": "Batch-size slowdown versus multi-GPU performance difference." + }, + { + "left": "issue:10928", + "right": "issue:7492", + "accept": false, + "reason": "Missing local shard file versus repository-not-found on the hub." + }, + { + "left": "issue:10782", + "right": "issue:11055", + "accept": false, + "reason": "Lumina minor issue versus unconditional training producing colorized images." + }, + { + "left": "issue:11083", + "right": "issue:8637", + "accept": false, + "reason": "DDP stride mismatch warning versus missing PixArtSigmaPipeline import." + }, + { + "left": "issue:11447", + "right": "issue:8541", + "accept": false, + "reason": "Offline shard-file lookup failure versus model repository not found." + }, + { + "left": "issue:1197", + "right": "issue:3287", + "accept": false, + "reason": "OpenAI Diffusion Pipeline proposal versus OpenAI improved DDPM support request." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 347, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10085", + "issue:10227", + "issue:1066", + "issue:10710", + "issue:10860", + "issue:10963", + "issue:11062", + "issue:12610", + "issue:13001", + "issue:244", + "issue:2459", + "issue:280", + "issue:2841", + "issue:2966", + "issue:3781", + "issue:3885", + "issue:4057", + "issue:4418", + "issue:4453", + "issue:459", + "issue:4680", + "issue:4788", + "issue:5150", + "issue:5160", + "issue:5225", + "issue:5442", + "issue:5567", + "issue:5710", + "issue:5763", + "issue:5846", + "issue:6313", + "issue:6444", + "issue:6533", + "issue:6949", + "issue:7593", + "issue:7752", + "issue:8594", + "issue:8759", + "issue:8786", + "issue:9025", + "issue:9051", + "issue:9138", + "issue:9241", + "issue:9267", + "issue:9387", + "issue:9523", + "issue:9953" + ], + "soft_pairs": [ + "issue:244|issue:280", + "issue:2459|issue:5567", + "issue:3885|issue:8594", + "issue:4057|issue:7752", + "issue:4453|issue:4680", + "issue:4788|issue:5225", + "issue:5150|issue:9025", + "issue:5160|issue:9241", + "issue:5442|issue:6313", + "issue:6533|issue:9387", + "issue:7593|issue:9267", + "issue:8759|issue:8786", + "issue:4418|issue:5763", + "issue:5846|issue:9523", + "issue:10085|issue:9051", + "issue:11062|issue:13001", + "issue:5710|issue:6949", + "issue:10227|issue:9138", + "issue:12610|issue:9953", + "issue:2841|issue:459", + "issue:2966|issue:6444", + "issue:1066|issue:3781", + "issue:10710|issue:9138", + "issue:10860|issue:10963" + ], + "prepared_review_unit_hash": "cfaee5345e68204eaa6e92b65d46a4433a018aa16c32314342c145e785b63db4", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33523, + "estimated_input_tokens": 8381, + "estimated_eval_tokens": 17018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: many items are in adjacent Diffusers areas (LoRA, schedulers, pipelines, quantization), but the paired issues do not cleanly describe the same underlying bug or change. I would not merge any of the suggested soft pairs.", + "confidence": 0.82, + "canonical_issue_reason": "Issue 10227 is the clearest self-contained bug report in the LoRA subgroup, with a concrete exception and an actionable technical description. That said, the overall cluster is too mixed for a true duplicate canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10227 is the strongest single representative issue because it is precise, technical, and easy to reproduce/search against, but it does not subsume the other cluster members.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:244", + "right": "issue:280", + "accept": false, + "reason": "Formatter-docs guidance and a DEIS scheduler feature request are unrelated." + }, + { + "left": "issue:2459", + "right": "issue:5567", + "accept": false, + "reason": "xformers attention in a LoRA training script and an accelerator/config overwrite bug are different code-paths." + }, + { + "left": "issue:3885", + "right": "issue:8594", + "accept": false, + "reason": "VRAM reduction advice for training is not the same bug as an UnboundLocalError in an SD3 training script." + }, + { + "left": "issue:4057", + "right": "issue:7752", + "accept": false, + "reason": "Finetuned inpainting blur and a training hyperparameter question are unrelated." + }, + { + "left": "issue:4453", + "right": "issue:4680", + "accept": false, + "reason": "LoRA conversion to safetensors and SDXL inpainting fine-tuning are different problems." + }, + { + "left": "issue:4788", + "right": "issue:5225", + "accept": false, + "reason": "Both involve memory limits, but they are separate OOM reports in different contexts and not the same bug." + }, + { + "left": "issue:5150", + "right": "issue:9025", + "accept": false, + "reason": "Both mention from_single_file, but they target different pipeline classes and do not show the same missing implementation bug." + }, + { + "left": "issue:5160", + "right": "issue:9241", + "accept": false, + "reason": "Dreambooth/LoRA masking and an XLab-to-Diffusers conversion script are unrelated." + }, + { + "left": "issue:5442", + "right": "issue:6313", + "accept": false, + "reason": "ScaleCrafter and Null-Text Inversion are distinct pipeline features with no shared failure mode." + }, + { + "left": "issue:6533", + "right": "issue:9387", + "accept": false, + "reason": "Marigold depth estimation and MimicMotion are separate pipelines." + }, + { + "left": "issue:7593", + "right": "issue:9267", + "accept": false, + "reason": "IP-Adapter image embed preparation error and missing rope attribute in Flux are different failures." + }, + { + "left": "issue:8759", + "right": "issue:8786", + "accept": false, + "reason": "A runtime warning in SD3 image postprocessing is not the same as max_sequence_length having no effect on token count." + }, + { + "left": "issue:4418", + "right": "issue:5763", + "accept": false, + "reason": "SDXL inpaint denoising_start behavior and safetensors serialization are unrelated." + }, + { + "left": "issue:5846", + "right": "issue:9523", + "accept": false, + "reason": "Checkpoint resume error in one training script and mapping errors across several training scripts are not the same concrete bug." + }, + { + "left": "issue:10085", + "right": "issue:9051", + "accept": false, + "reason": "Batch inference/skip_guidance_layers with LyCORIS is more specific than a general FluxPipeline batch-inference limitation; not the same underlying issue." + }, + { + "left": "issue:11062", + "right": "issue:13001", + "accept": false, + "reason": "A CivitAI LoRA loading error and a Flux2.Klein GGUF shape mismatch are distinct model-loading failures." + }, + { + "left": "issue:5710", + "right": "issue:6949", + "accept": false, + "reason": "Both touch DPMSolverSinglestepScheduler, but missing sigmas initialization and even-step blank images are different symptoms without a clear same-root-cause match." + }, + { + "left": "issue:10227", + "right": "issue:9138", + "accept": false, + "reason": "A LoRA feature-size NotImplementedError and an int/str TypeError are unrelated." + }, + { + "left": "issue:12610", + "right": "issue:9953", + "accept": false, + "reason": "Both involve quantized pipelines/device movement, but one is a Quanto+group-offload matmul device mismatch and the other is a broader cuda-move error; too imprecise to merge as the same fix." + }, + { + "left": "issue:2841", + "right": "issue:459", + "accept": false, + "reason": "Sampling temperature support and incremental diffusion are different feature requests." + }, + { + "left": "issue:2966", + "right": "issue:6444", + "accept": false, + "reason": "Multi-GPU training issues in different scripts with different stack traces are not the same bug." + }, + { + "left": "issue:1066", + "right": "issue:3781", + "accept": false, + "reason": "Docs generation in forks and loading multiple LoRAs are unrelated." + }, + { + "left": "issue:10710", + "right": "issue:9138", + "accept": false, + "reason": "A question about DDUF support and a TypeError report are unrelated." + }, + { + "left": "issue:10860", + "right": "issue:10963", + "accept": false, + "reason": "A LoRA compatibility/loading problem and a missing AutoencoderKLWan import are different issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 348, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10638", + "issue:11321", + "issue:11343", + "issue:1156", + "issue:12666", + "issue:12881", + "issue:1323", + "issue:1572", + "issue:1798", + "issue:2233", + "issue:2645", + "issue:2941", + "issue:2961", + "issue:3103", + "issue:3249", + "issue:3296", + "issue:3431", + "issue:3467", + "issue:3871", + "issue:4340", + "issue:4598", + "issue:4615", + "issue:487", + "issue:5002", + "issue:5166", + "issue:5365", + "issue:5429", + "issue:5475", + "issue:5693", + "issue:579", + "issue:6481", + "issue:6632", + "issue:6726", + "issue:7057", + "issue:7400", + "issue:7438", + "issue:7836", + "issue:7950", + "issue:7991", + "issue:8507", + "issue:8637", + "issue:8708", + "issue:8788", + "issue:9090", + "issue:9370", + "issue:9793", + "issue:9844", + "issue:9889" + ], + "soft_pairs": [ + "issue:11321|issue:7836", + "issue:11343|issue:8637", + "issue:2941|issue:3431", + "issue:4598|issue:5002", + "issue:4615|issue:7950", + "issue:5166|issue:9370", + "issue:6481|issue:7057", + "issue:6632|issue:8708", + "issue:7400|issue:9889", + "issue:7438|issue:9793", + "issue:5475|issue:6726", + "issue:3467|issue:3871", + "issue:7991|issue:9090", + "issue:1323|issue:579", + "issue:1572|issue:3103", + "issue:4340|issue:8507", + "issue:5693|issue:8788", + "issue:1156|issue:3296", + "issue:3249|issue:487", + "issue:2645|issue:5365", + "issue:10638|issue:9844", + "issue:12666|issue:12881", + "issue:1798|issue:5429", + "issue:2233|issue:2961" + ], + "prepared_review_unit_hash": "42a9660ae60509780225b7aeaeb4e4783b50608f9a5a512b6ad75c031a7a7e1e", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33720, + "estimated_input_tokens": 8430, + "estimated_eval_tokens": 17116 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose grouping of unrelated issue reports that only share broad diffusers/training/inference vocabulary. None of the soft pairs look like the same underlying bug.", + "confidence": 0.39, + "canonical_issue_reason": "issue:7991 is the best representative of the cluster\u2019s broad training/tutorial-bug theme and is specific enough to serve as a canonical anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:7991 is the most generally applicable and actionable issue in the set; the others are either narrower, more version-specific, or about different pipelines altogether.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11321", + "right": "issue:7836", + "accept": false, + "reason": "One is a FLUX ControlNet README problem; the other is a CLIP training overfitting example bug. Different topic and failure mode." + }, + { + "left": "issue:11343", + "right": "issue:8637", + "accept": false, + "reason": "Both are import errors, but they involve different packages/pipelines and likely different dependency breakages." + }, + { + "left": "issue:2941", + "right": "issue:3431", + "accept": false, + "reason": "EulerAncestral scheduler output quality is unrelated to Inpainting+ControlNet poor results; different code paths." + }, + { + "left": "issue:4598", + "right": "issue:5002", + "accept": false, + "reason": "Missing dependency error messaging is not the same as schedulers missing a Karras argument." + }, + { + "left": "issue:4615", + "right": "issue:7950", + "accept": false, + "reason": "ControlNet import error versus PyTorch 2.3 incompatibility; one is a symbol/export issue, the other a version compatibility issue." + }, + { + "left": "issue:5166", + "right": "issue:9370", + "accept": false, + "reason": "VQ-VAE scaling factor in unconditional LDM is unrelated to missing pooled_projection in timestep guidance embeddings." + }, + { + "left": "issue:6481", + "right": "issue:7057", + "accept": false, + "reason": "Dreambooth batch-size bug and a failing checkpoint load are different training failures." + }, + { + "left": "issue:6632", + "right": "issue:8708", + "accept": false, + "reason": "save_state question for LoRA scripts does not match missing vae.config.shift_factor in SD3 dreambooth examples." + }, + { + "left": "issue:7400", + "right": "issue:9889", + "accept": false, + "reason": "Black images in stable video diffusion and an SDXL-turbo segmentation fault are distinct inference/runtime problems." + }, + { + "left": "issue:7438", + "right": "issue:9793", + "accept": false, + "reason": "XPU generation stall is not the same as a progress bar refresh bug." + }, + { + "left": "issue:5475", + "right": "issue:6726", + "accept": false, + "reason": "These are different model papers/features: LaVie video generation versus CCSR super-resolution." + }, + { + "left": "issue:3467", + "right": "issue:3871", + "accept": false, + "reason": "ControlNet resume-step accounting with gradient accumulation is different from a generic checkpoint resume error in text-to-image training." + }, + { + "left": "issue:7991", + "right": "issue:9090", + "accept": false, + "reason": "Training tutorial bug and PNDMScheduler repeated timesteps are unrelated subsystems." + }, + { + "left": "issue:1323", + "right": "issue:579", + "accept": false, + "reason": "Flax model dtype loading and missing config.json during textual inversion are different loader failures." + }, + { + "left": "issue:1572", + "right": "issue:3103", + "accept": false, + "reason": "stable_diffusion_mega img2img/inpainting support and inpainting height/width handling are not the same bug." + }, + { + "left": "issue:4340", + "right": "issue:8507", + "accept": false, + "reason": "SDXL LoRA train_text_encoder error and SD3 dreambooth train_text_encoder failure are different model-specific training issues." + }, + { + "left": "issue:5693", + "right": "issue:8788", + "accept": false, + "reason": "VAE load error versus SD3 VAE tiling support are separate VAE problems." + }, + { + "left": "issue:1156", + "right": "issue:3296", + "accept": false, + "reason": "A generic 'training doesn't work' report and checkpoint save/load failure with text_encoder are not clearly the same defect." + }, + { + "left": "issue:3249", + "right": "issue:487", + "accept": false, + "reason": "AttnUpBlock2D upsample_size API mismatch is unrelated to StableDiffusionPipeline not working on M1 CPU." + }, + { + "left": "issue:2645", + "right": "issue:5365", + "accept": false, + "reason": "ControlNet on M1 with CUDA assertion and txt2video num_inference_steps error are different platform and scheduler issues." + }, + { + "left": "issue:10638", + "right": "issue:9844", + "accept": false, + "reason": "SDXL schedule script problem is not the same as SDXL VAE encoder producing NaNs." + }, + { + "left": "issue:12666", + "right": "issue:12881", + "accept": false, + "reason": "Kandinsky 5.0 Video Pipeline missing from a release is unrelated to a prompt2prompt word-replacement bug." + }, + { + "left": "issue:1798", + "right": "issue:5429", + "accept": false, + "reason": "Flax dreambooth v_prediction support and a generic request to support models without VAE are different feature gaps." + }, + { + "left": "issue:2233", + "right": "issue:2961", + "accept": false, + "reason": "Textual inversion validation/gradient accumulation bugs do not match the StableUnCLIP training example request." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 349, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10045", + "issue:10829", + "issue:11045", + "issue:11232", + "issue:11486", + "issue:1572", + "issue:181", + "issue:2440", + "issue:2444", + "issue:2673", + "issue:2737", + "issue:3698", + "issue:3967", + "issue:4264", + "issue:4374", + "issue:4400", + "issue:4453", + "issue:4580", + "issue:4686", + "issue:4806", + "issue:4830", + "issue:4841", + "issue:5004", + "issue:5079", + "issue:5109", + "issue:5567", + "issue:5647", + "issue:6109", + "issue:6801", + "issue:6958", + "issue:7043", + "issue:7080", + "issue:7185", + "issue:725", + "issue:7300", + "issue:7350", + "issue:7456", + "issue:759", + "issue:7686", + "issue:8511", + "issue:8543", + "issue:9161", + "issue:9225", + "issue:9237", + "issue:9366", + "issue:9467", + "issue:9534", + "issue:9568" + ], + "soft_pairs": [ + "issue:3967|issue:6958", + "issue:7300|issue:7350", + "issue:7456|issue:9467", + "issue:11232|issue:8511", + "issue:4580|issue:5109", + "issue:2673|issue:4686", + "issue:5567|issue:9534", + "issue:11045|issue:9161", + "issue:4806|issue:7686", + "issue:11486|issue:9237", + "issue:5647|issue:9366", + "issue:10045|issue:6801", + "issue:10829|issue:7043", + "issue:1572|issue:2440", + "issue:181|issue:2444", + "issue:2737|issue:4400", + "issue:3698|issue:759", + "issue:4264|issue:4453", + "issue:4374|issue:4841", + "issue:4830|issue:9225", + "issue:5004|issue:8543", + "issue:5079|issue:725", + "issue:6109|issue:7080", + "issue:7185|issue:9568" + ], + "prepared_review_unit_hash": "7b345a93006340096cf504667af476def1b89cf4a1dc37d0dcd1f065dcee9f80", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33553, + "estimated_input_tokens": 8389, + "estimated_eval_tokens": 17034 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are a loose set of unrelated Diffusers bugs/feature requests. The soft pairs mostly share broad subsystem words (pipeline, ControlNet, DreamBooth, SDXL) but not the same concrete failure or fix path, so I would not merge any of them as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "No clear canonical issue: the collection is heterogeneous and the proposed soft pairs do not describe the same underlying bug.", + "canonical_pr_reason": "No pull requests are present in the cluster.", + "best_issue_reason": "No single issue stands out as the best representative because the items are not a duplicate set and cover distinct problems.", + "best_pr_reason": "No pull requests are present in the cluster.", + "soft_edge_verdicts": [ + { + "left": "issue:3967", + "right": "issue:6958", + "accept": false, + "reason": "Both involve SDXL/unet-related failures, but one is a branch/deployment complaint and the other is a specific mid_block config mismatch in unet_motion_model. Different root causes." + }, + { + "left": "issue:7300", + "right": "issue:7350", + "accept": false, + "reason": "One reports high GPU memory after set_default_attn_processor(); the other is about initializing ControlNet from a LoRA-tuned UNet. Same area, different bugs." + }, + { + "left": "issue:7456", + "right": "issue:9467", + "accept": false, + "reason": "Both mention missing guard-rail/raise-style errors, but they are different products and code paths: a generic missing raise statement vs CogVideoX num_frames validation." + }, + { + "left": "issue:11232", + "right": "issue:8511", + "accept": false, + "reason": "Training-related only at a high level; one is about noise residuals during text-to-image fine-tuning, the other about SD3 scripts not unloading text encoders. Not the same defect." + }, + { + "left": "issue:4580", + "right": "issue:5109", + "accept": false, + "reason": "unipc/Karras scheduler exception and SDXL ControlNet img2img load_lora_weights exception are unrelated failures." + }, + { + "left": "issue:2673", + "right": "issue:4686", + "accept": false, + "reason": "Both touch StableDiffusionUpscalePipeline, but one is a runtime config/image verification failure and the other is loading a saved pipeline with from_pretrained. Different issue shape." + }, + { + "left": "issue:5567", + "right": "issue:9534", + "accept": false, + "reason": "Both mention config attributes, but one is accelerator.prepare overwriting unet.config and the other is FluxMultiControlNetModel lacking config. Not the same code-path bug." + }, + { + "left": "issue:11045", + "right": "issue:9161", + "accept": false, + "reason": "Both are Flux/DreamBooth LoRA training reports, but one is a model load crash and the other is distributed-training bugs. Not one concrete bug." + }, + { + "left": "issue:4806", + "right": "issue:7686", + "accept": false, + "reason": "Both concern UNet2DModel initialization, but one is an incorrect parameter type and the other is a missing cross_attention_dim requirement. Different validation errors." + }, + { + "left": "issue:11486", + "right": "issue:9237", + "accept": false, + "reason": "One requests enabling xformers in the training script; the other is a broad Flux DreamBooth LoRA training issue. No shared concrete failure." + }, + { + "left": "issue:5647", + "right": "issue:9366", + "accept": false, + "reason": "Both involve strength-related img2img behavior, but 5647 is a specific inpaint strength==1 bug while 9366 is a scheduler/step-count edge case. Not mergeable as one fix." + }, + { + "left": "issue:10045", + "right": "issue:6801", + "accept": false, + "reason": "Importing pipelines and loading schedulers are different initialization problems, even though both happen during model loading." + }, + { + "left": "issue:10829", + "right": "issue:7043", + "accept": false, + "reason": "OOM during final inference in SDXL DreamBooth LoRA is unrelated to custom captions support in Dreambooth." + }, + { + "left": "issue:1572", + "right": "issue:2440", + "accept": false, + "reason": "Both mention inpainting, but one is a mega pipeline compatibility issue and the other is float16 model loading. Not the same bug." + }, + { + "left": "issue:181", + "right": "issue:2444", + "accept": false, + "reason": "Train_unconditional image writing failure and dreambooth_inpaint subprocess exit error are unrelated training-script issues." + }, + { + "left": "issue:2737", + "right": "issue:4400", + "accept": false, + "reason": "CLIP guidance scheduler support and text_encoder support in an inpaint LoRA script are different feature requests." + }, + { + "left": "issue:3698", + "right": "issue:759", + "accept": false, + "reason": "Converting an inpainting model to diffusers and making full-model saving optional for textual inversion are unrelated tasks." + }, + { + "left": "issue:4264", + "right": "issue:4453", + "accept": false, + "reason": "Missing example images for SDXL ensemble of experts and SDXL LoRA conversion to safetensors are unrelated documentation/tooling issues." + }, + { + "left": "issue:4374", + "right": "issue:4841", + "accept": false, + "reason": "Both involve ControlNet and image size, but one is an inpaint pipeline issue and the other is a generic ControlNet image-size problem. Too different to treat as duplicates." + }, + { + "left": "issue:4830", + "right": "issue:9225", + "accept": false, + "reason": "Wrong tensor size in MultiControlNet and latent-size mismatch with rgba images are both shape issues, but they affect different pipelines and failure mechanisms." + }, + { + "left": "issue:5004", + "right": "issue:8543", + "accept": false, + "reason": "Checkpoint resume at fp16 and gradient checkpointing failure are both training reliability issues, but they are separate bugs." + }, + { + "left": "issue:5079", + "right": "issue:725", + "accept": false, + "reason": "LoRA loading support and device-map/accelerate memory reduction are different model-loading enhancements." + }, + { + "left": "issue:6109", + "right": "issue:7080", + "accept": false, + "reason": "ControlNet-xs depth support and SDXL ControlNet training inference output drift from wandb are unrelated ControlNet problems." + }, + { + "left": "issue:7185", + "right": "issue:9568", + "accept": false, + "reason": "PyTorch 2.2 compatibility question and CogVideoX autoencoder reconstruction failure have no meaningful overlap beyond being user reports." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 350, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10165", + "issue:10287", + "issue:10950", + "issue:11321", + "issue:11547", + "issue:12308", + "issue:1246", + "issue:1335", + "issue:1358", + "issue:1530", + "issue:1717", + "issue:2024", + "issue:2163", + "issue:2177", + "issue:2265", + "issue:2359", + "issue:2803", + "issue:3403", + "issue:3612", + "issue:3627", + "issue:3661", + "issue:3933", + "issue:4236", + "issue:4256", + "issue:4615", + "issue:4677", + "issue:4975", + "issue:5009", + "issue:5245", + "issue:5327", + "issue:5466", + "issue:6238", + "issue:6321", + "issue:646", + "issue:6921", + "issue:699", + "issue:7061", + "issue:7147", + "issue:7267", + "issue:7968", + "issue:8593", + "issue:8673", + "issue:8786", + "issue:9023", + "issue:9239", + "issue:9599", + "issue:9613", + "issue:9619" + ], + "soft_pairs": [ + "issue:8593|issue:8786", + "issue:10287|issue:9023", + "issue:1717|issue:2177", + "issue:10950|issue:7267", + "issue:2803|issue:4615", + "issue:1246|issue:3627", + "issue:1335|issue:1530", + "issue:5327|issue:5466", + "issue:6238|issue:6921", + "issue:11321|issue:9239", + "issue:11547|issue:9599", + "issue:12308|issue:9619", + "issue:1358|issue:699", + "issue:2024|issue:646", + "issue:2359|issue:3612", + "issue:3403|issue:7061", + "issue:3933|issue:4975", + "issue:7968|issue:8673", + "issue:3661|issue:6321", + "issue:4236|issue:4677", + "issue:5245|issue:7147", + "issue:2163|issue:2265", + "issue:4256|issue:5009", + "issue:10165|issue:9613" + ], + "prepared_review_unit_hash": "840801755755459cce1d76b2084bdca31a95f2b4b09af0feb15ec958b1950286", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33521, + "estimated_input_tokens": 8381, + "estimated_eval_tokens": 17018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated docs, training, loading, and runtime issues across different pipelines/models. There is no convincing duplicate group; if one issue must be picked as the broadest representative, #10287 is the most generic docs/example report.", + "confidence": 0.17, + "canonical_issue_reason": "#10287 is the broadest and most generic issue here, framed around an example-code/documentation problem rather than a narrow model- or pipeline-specific bug.", + "canonical_pr_reason": null, + "best_issue_reason": "#10287 is the best overall issue representative because its scope is broad and documentation-focused, but it still should not be treated as a duplicate anchor for the rest of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8593", + "right": "issue:8786", + "accept": false, + "reason": "Different SD3 bugs: one is bfloat16 noise on CPU, the other is a max_sequence_length token-count issue." + }, + { + "left": "issue:10287", + "right": "issue:9023", + "accept": false, + "reason": "Both mention example/sample code, but they are distinct documentation/runtime failures with no shared code path shown." + }, + { + "left": "issue:1717", + "right": "issue:2177", + "accept": false, + "reason": "Offline cache loading vs pretrained AutoencoderKL loading are different model-loading bugs." + }, + { + "left": "issue:10950", + "right": "issue:7267", + "accept": false, + "reason": "Doc section brokenness and local training-data loading failure are unrelated issues." + }, + { + "left": "issue:2803", + "right": "issue:4615", + "accept": false, + "reason": "Docs build on WSL2 and an import error for a ControlNet pipeline are separate problems." + }, + { + "left": "issue:1246", + "right": "issue:3627", + "accept": false, + "reason": "Different example failures: FP16 gradient unscale in Dreambooth vs retrieve.py/custom diffusion error." + }, + { + "left": "issue:1335", + "right": "issue:1530", + "accept": false, + "reason": "These are different missing-argument/missing-field bugs in different pipelines, not the same underlying defect." + }, + { + "left": "issue:5327", + "right": "issue:5466", + "accept": false, + "reason": "SDXL textual inversion loading and LoRA weight-shape mismatch are distinct loading paths." + }, + { + "left": "issue:6238", + "right": "issue:6921", + "accept": false, + "reason": "An SDXL VAE issue is not the same as shipped weights deviating from official weights." + }, + { + "left": "issue:11321", + "right": "issue:9239", + "accept": false, + "reason": "README/documentation bug vs black-image runtime output from ControlNet-Canny are different." + }, + { + "left": "issue:11547", + "right": "issue:9599", + "accept": false, + "reason": "Feature request for Flux controlnet support and request for a LoRA-only example are not the same change." + }, + { + "left": "issue:12308", + "right": "issue:9619", + "accept": false, + "reason": "Flash-attn kernel support for non-Hopper GPUs is unrelated to adding interrupt support in a ControlNet pipeline." + }, + { + "left": "issue:1358", + "right": "issue:699", + "accept": false, + "reason": "Safetensors support and negative-prompt support are unrelated feature requests." + }, + { + "left": "issue:2024", + "right": "issue:646", + "accept": false, + "reason": "Corrupted notebook file and PNDMScheduler API missing attribute are different classes of bugs." + }, + { + "left": "issue:2359", + "right": "issue:3612", + "accept": false, + "reason": "Img2img latent output type and ControlNet creation/preprocessor questions do not share the same code path." + }, + { + "left": "issue:3403", + "right": "issue:7061", + "accept": false, + "reason": "Pipeline loading with TorchDynamo and an incorrect docstring are unrelated." + }, + { + "left": "issue:3933", + "right": "issue:4975", + "accept": false, + "reason": "Graph compile/logging failure and LoRA loading inefficiency are separate concerns." + }, + { + "left": "issue:7968", + "right": "issue:8673", + "accept": false, + "reason": "Both are attention-related, but they affect different models and different mechanisms (motion module vs SD3 text padding masks)." + }, + { + "left": "issue:3661", + "right": "issue:6321", + "accept": false, + "reason": "Checkpoint conversion for SD2 weights and missing CLIP detection are unrelated loading issues." + }, + { + "left": "issue:4236", + "right": "issue:4677", + "accept": false, + "reason": "SDXL refiner text-encoder registration and an inpainting doc script error are not the same bug." + }, + { + "left": "issue:5245", + "right": "issue:7147", + "accept": false, + "reason": "A pipenv import freeze and a missing pipeline import are different failures." + }, + { + "left": "issue:2163", + "right": "issue:2265", + "accept": false, + "reason": "UNet dtype mismatch during text-to-image training and ONNXRuntime incompatibility are separate example/runtime issues." + }, + { + "left": "issue:4256", + "right": "issue:5009", + "accept": false, + "reason": "Different model-specific training/runtime errors: SDXL Dreambooth vs Wuerstchen." + }, + { + "left": "issue:10165", + "right": "issue:9613", + "accept": false, + "reason": "LoRA seed preservation on scale change and a Timesteps.scale attribute error on a custom diffusion model are unrelated bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 351, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10017", + "issue:1101", + "issue:11035", + "issue:11136", + "issue:1141", + "issue:11447", + "issue:12060", + "issue:12195", + "issue:12695", + "issue:13225", + "issue:181", + "issue:2616", + "issue:2856", + "issue:3026", + "issue:3174", + "issue:3209", + "issue:3432", + "issue:3808", + "issue:4061", + "issue:4134", + "issue:4263", + "issue:4562", + "issue:4764", + "issue:5070", + "issue:5079", + "issue:525", + "issue:5313", + "issue:5368", + "issue:5466", + "issue:5577", + "issue:5689", + "issue:5923", + "issue:6520", + "issue:6921", + "issue:6928", + "issue:6949", + "issue:7000", + "issue:7012", + "issue:7049", + "issue:7197", + "issue:7619", + "issue:7851", + "issue:7877", + "issue:7991", + "issue:8604", + "issue:8732", + "issue:9613" + ], + "soft_pairs": [ + "issue:1101|issue:1141", + "issue:2616|issue:525", + "issue:2856|issue:4562", + "issue:3174|issue:5313", + "issue:5070|issue:6928", + "issue:5079|issue:7877", + "issue:5689|issue:6921", + "issue:6520|issue:6949", + "issue:3432|issue:7012", + "issue:10017|issue:6520", + "issue:11136|issue:9613", + "issue:3808|issue:7619", + "issue:4134|issue:7851", + "issue:3209|issue:7000", + "issue:4764|issue:5466", + "issue:5923|issue:8732", + "issue:7991|issue:8604", + "issue:5368|issue:7049", + "issue:5577|issue:7197", + "issue:11035|issue:12695", + "issue:11447|issue:12195", + "issue:12060|issue:13225", + "issue:181|issue:3026", + "issue:4061|issue:4263" + ], + "prepared_review_unit_hash": "d59060458951c43e9b477976cff9a3990d0d48325648a8793e363d55580992bf", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33437, + "estimated_input_tokens": 8360, + "estimated_eval_tokens": 16976 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of unrelated diffusers issues; none of the soft-edge pairs look like true duplicates or the same fix. The shared wording is usually only at the subsystem level (LoRA, SDXL, training, timesteps), which is too broad to merge.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue stands out: the items span many unrelated bugs/features, so this is not a coherent duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a strong global representative because the cluster is heterogeneous; at most they share the broader diffusers domain, not one underlying problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1101", + "right": "issue:1141", + "accept": false, + "reason": "Different requests: adding a scheduler vs supporting local files in custom pipelines." + }, + { + "left": "issue:2616", + "right": "issue:525", + "accept": false, + "reason": "Both mention loading/saving workflows, but one is a LoRA save error and the other is local-disk image loading in notebooks." + }, + { + "left": "issue:2856", + "right": "issue:4562", + "accept": false, + "reason": "Both are model-conversion failures, but they concern different conversion paths, formats, and error modes." + }, + { + "left": "issue:3174", + "right": "issue:5313", + "accept": false, + "reason": "Both involve LoRA, but one is loading broken by new features and the other is non-revertible fuse behavior." + }, + { + "left": "issue:5070", + "right": "issue:6928", + "accept": false, + "reason": "Unrelated: ControlNet training flags vs a Marigold depthmap colorization option." + }, + { + "left": "issue:5079", + "right": "issue:7877", + "accept": false, + "reason": "Loading a specific LoRA style is unrelated to adding an attention out_dim argument." + }, + { + "left": "issue:5689", + "right": "issue:6921", + "accept": false, + "reason": "Both touch SDXL quality, but one is a sampler noise issue and the other is a shipped-weights mismatch." + }, + { + "left": "issue:6520", + "right": "issue:6949", + "accept": false, + "reason": "Blank validation output is not the same bug as even-step blank images from a specific scheduler." + }, + { + "left": "issue:3432", + "right": "issue:7012", + "accept": false, + "reason": "CPU half-precision convolution failure is unrelated to v_prediction image artifacts." + }, + { + "left": "issue:10017", + "right": "issue:6520", + "accept": false, + "reason": "Dataset loading with load_dataset() is unrelated to blank validation results during adapter training." + }, + { + "left": "issue:11136", + "right": "issue:9613", + "accept": false, + "reason": "Different timesteps-related failures with different models and errors; not the same code path." + }, + { + "left": "issue:3808", + "right": "issue:7619", + "accept": false, + "reason": "Both are training issues, but they affect different pipelines and different underlying failures." + }, + { + "left": "issue:4134", + "right": "issue:7851", + "accept": false, + "reason": "MPS tensor-shape mismatch in ControlNet is unrelated to a concurrent-pipeline index error." + }, + { + "left": "issue:3209", + "right": "issue:7000", + "accept": false, + "reason": "NaN encoder outputs and a PyTorch deprecation warning are not the same bug." + }, + { + "left": "issue:4764", + "right": "issue:5466", + "accept": false, + "reason": "Both are LoRA load errors, but they involve different file types/models and different incompatibility causes." + }, + { + "left": "issue:5923", + "right": "issue:8732", + "accept": false, + "reason": "Multi-GPU training hang is unrelated to checkpoint-saving trouble." + }, + { + "left": "issue:7991", + "right": "issue:8604", + "accept": false, + "reason": "A tutorial bug is unrelated to corrupt images from float16 T5 conversion." + }, + { + "left": "issue:5368", + "right": "issue:7049", + "accept": false, + "reason": "Mixed-precision dtype mismatch is unrelated to an IP-Adapter unpacking error." + }, + { + "left": "issue:5577", + "right": "issue:7197", + "accept": false, + "reason": "Both are KeyErrors, but they are different missing keys in different loading/resume scenarios." + }, + { + "left": "issue:11035", + "right": "issue:12695", + "accept": false, + "reason": "A symlink/download option request is unrelated to Hunyuan v1.5 support." + }, + { + "left": "issue:11447", + "right": "issue:12195", + "accept": false, + "reason": "Offline shard-file handling and a Torch custom_op compatibility failure are unrelated." + }, + { + "left": "issue:12060", + "right": "issue:13225", + "accept": false, + "reason": "A question about whether DiT blocks exist is not the same as support for published RAE+DiT checkpoints." + }, + { + "left": "issue:181", + "right": "issue:3026", + "accept": false, + "reason": "Image-writing failure in a training script is unrelated to a one-GPU accelerate launch error." + }, + { + "left": "issue:4061", + "right": "issue:4263", + "accept": false, + "reason": "Both involve LoRA, but one is loading LoRA in SDXL and the other is unloading/fusing behavior." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 352, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1009", + "issue:10198", + "issue:10772", + "issue:10897", + "issue:11220", + "issue:11343", + "issue:12173", + "issue:12369", + "issue:1604", + "issue:1923", + "issue:1957", + "issue:2265", + "issue:2412", + "issue:2459", + "issue:3120", + "issue:3452", + "issue:3570", + "issue:3748", + "issue:3954", + "issue:4178", + "issue:4279", + "issue:4354", + "issue:4438", + "issue:4615", + "issue:4771", + "issue:4804", + "issue:4912", + "issue:5084", + "issue:5518", + "issue:5803", + "issue:5824", + "issue:6218", + "issue:6703", + "issue:7102", + "issue:7370", + "issue:7606", + "issue:7736", + "issue:782", + "issue:7962", + "issue:8464", + "issue:8493", + "issue:8502", + "issue:861", + "issue:8915", + "issue:9186", + "issue:9332", + "issue:962", + "issue:9822" + ], + "soft_pairs": [ + "issue:4178|issue:4354", + "issue:4912|issue:7962", + "issue:782|issue:861", + "issue:1923|issue:2412", + "issue:3120|issue:5803", + "issue:4279|issue:5518", + "issue:8464|issue:9332", + "issue:10198|issue:9822", + "issue:10897|issue:7370", + "issue:11220|issue:8502", + "issue:12173|issue:12369", + "issue:3452|issue:6703", + "issue:3570|issue:861", + "issue:3748|issue:7606", + "issue:4438|issue:5824", + "issue:4615|issue:5084", + "issue:4912|issue:7736", + "issue:6218|issue:7102", + "issue:1604|issue:2265", + "issue:1957|issue:3954", + "issue:4771|issue:8493", + "issue:4804|issue:962", + "issue:1009|issue:2459", + "issue:10772|issue:9186", + "issue:11343|issue:8915" + ], + "prepared_review_unit_hash": "eb0a81000588768e2afab9e550c7635bb720195dc6d7dda60d7b044549596cbe", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33986, + "estimated_input_tokens": 8497, + "estimated_eval_tokens": 17250 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The items are mostly unrelated issues; the soft-similarity links look like broad topical overlap rather than true duplicates. I would not merge any of the proposed soft edges.", + "confidence": 0.86, + "canonical_issue_reason": "issue:9332 is the strongest standalone issue to anchor the set: it has the most detailed reproduction, a concrete loading symptom, and a broad but specific code path around model loading.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9332 is the best representative issue overall because it is detailed, actionable, and centered on a concrete bug rather than a vague feature request or environment-specific complaint.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4178", + "right": "issue:4354", + "accept": false, + "reason": "Both mention LoRA/pipeline behavior, but one is about unloading weights with XFormers and the other is about a wrong pipeline type from `from_single_file`; different bugs and code paths." + }, + { + "left": "issue:4912", + "right": "issue:7962", + "accept": false, + "reason": "One is about converting SDXL CivitAI checkpoints to diffusers, the other about adding `cross_attention_kwargs` to Instruct-Pix2Pix; this is a feature request vs a conversion problem." + }, + { + "left": "issue:782", + "right": "issue:861", + "accept": false, + "reason": "These are separate loading/conversion failures: model download auth handling vs checkpoint-to-diffusers conversion." + }, + { + "left": "issue:1923", + "right": "issue:2412", + "accept": false, + "reason": "Missing CUDA libraries in a Docker/bitsandbytes setup is unrelated to a DreamBooth training error." + }, + { + "left": "issue:3120", + "right": "issue:5803", + "accept": false, + "reason": "ControlNet training producing unchanged outputs is a distinct bug from reproducing black images in SD generation." + }, + { + "left": "issue:4279", + "right": "issue:5518", + "accept": false, + "reason": "Both involve LoRA loading, but one is SDXL CivitAI loading failure and the other is an `AttributeError` involving `ModuleList`; not the same defect." + }, + { + "left": "issue:8464", + "right": "issue:9332", + "accept": false, + "reason": "SEGA cpu-offload seed failure is a different pipeline/runtime bug from `from_pretrained()` loading the wrong weights." + }, + { + "left": "issue:10198", + "right": "issue:9822", + "accept": false, + "reason": "Both mention Flux/SDXL weights, but one is a Flux.1-dev weight mismatch and the other is loading SDXL LoRAs into Flux; related area, different concrete problem." + }, + { + "left": "issue:10897", + "right": "issue:7370", + "accept": false, + "reason": "These are separate support requests for different video/image models, not the same underlying change." + }, + { + "left": "issue:11220", + "right": "issue:8502", + "accept": false, + "reason": "Documentation page malfunction is not the same as an SD3 medium 512x512 generation issue." + }, + { + "left": "issue:12173", + "right": "issue:12369", + "accept": false, + "reason": "Both are Wan-related, but one is latent upscale i2i behavior and the other is a 4K CineScale support request; different capabilities and likely different fixes." + }, + { + "left": "issue:3452", + "right": "issue:6703", + "accept": false, + "reason": "DreamBooth 4GB VRAM training concerns are unrelated to a textual inversion token being weakened by negative prompts." + }, + { + "left": "issue:3570", + "right": "issue:861", + "accept": false, + "reason": "Disabling downloads when loading checkpoints is a different loading-path issue from checkpoint conversion to diffusers." + }, + { + "left": "issue:3748", + "right": "issue:7606", + "accept": false, + "reason": "A float16/float32 mismatch while loading LoRA safetensors is unrelated to a deprecated Kandinsky LoRA processor class." + }, + { + "left": "issue:4438", + "right": "issue:5824", + "accept": false, + "reason": "A generic SDXL pipeline error and a wrong `text_encoder_projection_dim` default are not clearly the same bug." + }, + { + "left": "issue:4615", + "right": "issue:5084", + "accept": false, + "reason": "These are import errors for different pipeline symbols and different backends (ControlNet vs Flax SDXL)." + }, + { + "left": "issue:4912", + "right": "issue:7736", + "accept": false, + "reason": "Custom checkpoint conversion and adding callbacks are unrelated changes." + }, + { + "left": "issue:6218", + "right": "issue:7102", + "accept": false, + "reason": "Both involve ControlNet, but one is batched seeds in ControlNet/Img2Img and the other is multi-ControlNet with multi-prompt behavior; not the same concrete bug." + }, + { + "left": "issue:1604", + "right": "issue:2265", + "accept": false, + "reason": "A broken LPW script on diffusers 0.9.0 is unrelated to ONNXRuntime incompatibility in the unconditional training script." + }, + { + "left": "issue:1957", + "right": "issue:3954", + "accept": false, + "reason": "Both touch LR scheduling, but one is scheduler initialization in unconditional training and the other is step counting under multi-GPU; different failure modes." + }, + { + "left": "issue:4771", + "right": "issue:8493", + "accept": false, + "reason": "Loading a fine-tuned diffusion model is too broad to match the specific SD3Transformer2DModel state_dict error." + }, + { + "left": "issue:4804", + "right": "issue:962", + "accept": false, + "reason": "A LoRA loader dictionary mutation error is unrelated to inpaint pipeline size limitations." + }, + { + "left": "issue:1009", + "right": "issue:2459", + "accept": false, + "reason": "Textual inversion weight decay and LoRA xformers support in a different training script are separate training issues." + }, + { + "left": "issue:10772", + "right": "issue:9186", + "accept": false, + "reason": "A feature request for Sana ControlNet support is not the same as a Differential Diffusion ControlNet pipeline issue." + }, + { + "left": "issue:11343", + "right": "issue:8915", + "accept": false, + "reason": "Both are import failures, but the affected modules and root causes differ: a transformers update breaking diffusers imports versus a missing `AutoPipelineForText2Image` import." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 353, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10382", + "issue:10511", + "issue:10573", + "issue:11022", + "issue:11135", + "issue:11885", + "issue:11962", + "issue:11989", + "issue:1284", + "issue:2822", + "issue:2830", + "issue:2905", + "issue:3113", + "issue:3347", + "issue:3366", + "issue:3570", + "issue:3674", + "issue:3679", + "issue:3787", + "issue:4158", + "issue:4217", + "issue:4398", + "issue:4414", + "issue:4419", + "issue:4724", + "issue:4977", + "issue:5153", + "issue:5256", + "issue:5410", + "issue:5630", + "issue:5953", + "issue:5977", + "issue:6001", + "issue:6348", + "issue:6363", + "issue:7057", + "issue:7168", + "issue:7519", + "issue:7945", + "issue:8440", + "issue:8534", + "issue:8922", + "issue:9042", + "issue:9546", + "issue:9686", + "issue:9930", + "issue:9958" + ], + "soft_pairs": [ + "issue:1284|issue:4414", + "issue:2830|issue:3679", + "issue:4398|issue:4977", + "issue:5953|issue:9546", + "issue:6348|issue:9958", + "issue:7168|issue:7945", + "issue:4158|issue:7168", + "issue:2905|issue:5256", + "issue:3570|issue:7057", + "issue:4419|issue:7519", + "issue:10382|issue:9930", + "issue:3347|issue:3787", + "issue:3674|issue:6363", + "issue:5410|issue:5630", + "issue:10573|issue:11989", + "issue:11962|issue:9042", + "issue:3113|issue:5153", + "issue:3366|issue:4724", + "issue:5977|issue:8440", + "issue:6001|issue:8922", + "issue:8534|issue:9686", + "issue:10511|issue:11135", + "issue:11022|issue:11885", + "issue:2822|issue:4217" + ], + "prepared_review_unit_hash": "7c7ed2ff413f174889504efdfdc92fd8f3dc7edaac42389a500a7e9352faa32f", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33479, + "estimated_input_tokens": 8370, + "estimated_eval_tokens": 16996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of unrelated Diffusers issues and feature requests across different pipelines, schedulers, model families, and utility bugs. None of the soft pairs look like the same underlying defect or change, so they should stay separate.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:10382 is the clearest standalone bug report here: it names a specific module, a concrete return-type mismatch, and an AttributeError symptom, making it the strongest triage anchor among otherwise unrelated items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1284", + "right": "issue:4414", + "accept": false, + "reason": "Batch-size slowdown and Consistency Model sampling are different performance/problem domains with no shared bug path." + }, + { + "left": "issue:2830", + "right": "issue:3679", + "accept": false, + "reason": "A question about image/text-to-video use is unrelated to announcing new text-to-video models." + }, + { + "left": "issue:4398", + "right": "issue:4977", + "accept": false, + "reason": "Black SDXL LoRA outputs and removing checkpoints in a training example are unrelated issues." + }, + { + "left": "issue:5953", + "right": "issue:9546", + "accept": false, + "reason": "Control image size mismatch crash and Flux validation OOM are different controlnet/training problems." + }, + { + "left": "issue:6348", + "right": "issue:9958", + "accept": false, + "reason": "3D medical/video AutoencoderKL support and temporal-decoder tiling are different feature requests for different classes." + }, + { + "left": "issue:7168", + "right": "issue:7945", + "accept": false, + "reason": "IP-Adapter image-embed tensor-shape troubleshooting is unrelated to a new community pipeline proposal." + }, + { + "left": "issue:4158", + "right": "issue:7168", + "accept": false, + "reason": "from_single_file stalling and IP-Adapter embedding shape issues are separate code paths and symptoms." + }, + { + "left": "issue:2905", + "right": "issue:5256", + "accept": false, + "reason": "Scheduler-wide Karras sigmas support and Insta-Flow training support are unrelated feature requests." + }, + { + "left": "issue:3570", + "right": "issue:7057", + "accept": false, + "reason": "Model download disabling and a checkpoint load failure are both loading-related but not the same concrete bug." + }, + { + "left": "issue:4419", + "right": "issue:7519", + "accept": false, + "reason": "Offset-noise inference behavior and XFormers scale-kwarg handling are different attention/inference concerns." + }, + { + "left": "issue:10382", + "right": "issue:9930", + "accept": false, + "reason": "VAE encoder return-type mismatch with AttributeError is unrelated to PAG adaptive scale behavior." + }, + { + "left": "issue:3347", + "right": "issue:3787", + "accept": false, + "reason": "NaN image generation and a circular import ImportError are distinct failure modes." + }, + { + "left": "issue:3674", + "right": "issue:6363", + "accept": false, + "reason": "metadata.jsonl parsing and FP16 gradient unscale errors occur in different training stages and code paths." + }, + { + "left": "issue:5410", + "right": "issue:5630", + "accept": false, + "reason": "Blurry latent upscale and leftover noise in SDXL Img2Img are both quality complaints, but they point to different pipeline/scheduler bugs." + }, + { + "left": "issue:10573", + "right": "issue:11989", + "accept": false, + "reason": "Multiple-controlnet padding_mask_crop failure and a black image/weight suspicion are unrelated symptoms." + }, + { + "left": "issue:11962", + "right": "issue:9042", + "accept": false, + "reason": "GGUF support for FLUX.1-Kontext-dev and adding the Flux pipeline are different model-support requests." + }, + { + "left": "issue:3113", + "right": "issue:5153", + "accept": false, + "reason": "Cross-attention layout control and FreeU are unrelated diffusion technique feature requests." + }, + { + "left": "issue:3366", + "right": "issue:4724", + "accept": false, + "reason": "Autocast breakage in LoRALinearLayer and SDXL Img2Img misbehavior with a checkpoint are different bugs." + }, + { + "left": "issue:5977", + "right": "issue:8440", + "accept": false, + "reason": "Saving unwanted LoRA keys and FromOriginalModelMixin subclass ValueError are separate implementation issues." + }, + { + "left": "issue:6001", + "right": "issue:8922", + "accept": false, + "reason": "Incorrect pipeline class returned from from_single_file and black output from quantized T5 are different problems." + }, + { + "left": "issue:8534", + "right": "issue:9686", + "accept": false, + "reason": "Wrong SD3 dreambooth weighting and SD3 ControlNet pooled projection bugs are separate SD3 issues in different parts of the stack." + }, + { + "left": "issue:10511", + "right": "issue:11135", + "accept": false, + "reason": "Scheduler redesign for SD3/Flux and a new SmoothCache pipeline are unrelated proposals." + }, + { + "left": "issue:11022", + "right": "issue:11885", + "accept": false, + "reason": "Quanto support for WAN 2.1 and a FLUX.1 LoRA loading bug are unrelated model-specific topics." + }, + { + "left": "issue:2822", + "right": "issue:4217", + "accept": false, + "reason": "Training convergence slowness and safetensors loading slowness are different performance issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 354, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10085", + "issue:10712", + "issue:10937", + "issue:11077", + "issue:11363", + "issue:1161", + "issue:1608", + "issue:1851", + "issue:1883", + "issue:2276", + "issue:2462", + "issue:2552", + "issue:2838", + "issue:2966", + "issue:3073", + "issue:3811", + "issue:4057", + "issue:4061", + "issue:4064", + "issue:4109", + "issue:4321", + "issue:4491", + "issue:4872", + "issue:5108", + "issue:5135", + "issue:5567", + "issue:5574", + "issue:5604", + "issue:5835", + "issue:6008", + "issue:6154", + "issue:6274", + "issue:6320", + "issue:6435", + "issue:6774", + "issue:7180", + "issue:730", + "issue:752", + "issue:7706", + "issue:7735", + "issue:7946", + "issue:7982", + "issue:8429", + "issue:8472", + "issue:8759", + "issue:9667", + "issue:9686" + ], + "soft_pairs": [ + "issue:2838|issue:752", + "issue:7735|issue:8759", + "issue:2966|issue:6154", + "issue:6320|issue:9667", + "issue:10937|issue:11077", + "issue:3811|issue:5604", + "issue:1161|issue:1883", + "issue:2462|issue:752", + "issue:3073|issue:4064", + "issue:4321|issue:5835", + "issue:11363|issue:8429", + "issue:5135|issue:8472", + "issue:5567|issue:7982", + "issue:7180|issue:7706", + "issue:1851|issue:730", + "issue:4109|issue:7946", + "issue:5574|issue:6008", + "issue:1608|issue:4872", + "issue:2552|issue:4057", + "issue:5108|issue:6274", + "issue:6435|issue:9686", + "issue:10085|issue:10712", + "issue:2276|issue:6774", + "issue:4061|issue:4491" + ], + "prepared_review_unit_hash": "fdac0d2b1b31a8f526a5e1421f455f178ccd15d288d34557170575b69e7064a7", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32973, + "estimated_input_tokens": 8244, + "estimated_eval_tokens": 16744 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set; it mixes unrelated issues across training, LoRA, ControlNet, SDXL, VAE, and pipeline utilities. All soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 10712 is the cleanest concrete bug report in the set, but it is not a canonical duplicate target because the rest of the cluster is unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2838", + "right": "issue:752", + "accept": false, + "reason": "Different requests: ControlNet pipeline behavior vs DreamBooth multi-concept support." + }, + { + "left": "issue:7735", + "right": "issue:8759", + "accept": false, + "reason": "Training dataset caption-column error vs SD3 image postprocessing warning; unrelated code paths." + }, + { + "left": "issue:2966", + "right": "issue:6154", + "accept": false, + "reason": "Multi-GPU instruct-pix2pix training issue vs SDXL LoRA training on M2 Mac; different failures." + }, + { + "left": "issue:6320", + "right": "issue:9667", + "accept": false, + "reason": "Remote model loading failure vs local flux fp8 transformer file loading; different artifacts and symptoms." + }, + { + "left": "issue:10937", + "right": "issue:11077", + "accept": false, + "reason": "torch.compile VAE encode error vs VAE conversion-script question; not the same bug." + }, + { + "left": "issue:3811", + "right": "issue:5604", + "accept": false, + "reason": "Both LoRA-related, but one is ineffective LoRA loading and the other is unfusing multiple LoRA; different problems." + }, + { + "left": "issue:1161", + "right": "issue:1883", + "accept": false, + "reason": "Model conversion workflow vs VAE gradient checkpointing; unrelated issues." + }, + { + "left": "issue:2462", + "right": "issue:752", + "accept": false, + "reason": "IterableDataset training support request vs DreamBooth multi-concept support; different feature asks." + }, + { + "left": "issue:3073", + "right": "issue:4064", + "accept": false, + "reason": "Consistency-model request vs PSLD pipeline request; different features." + }, + { + "left": "issue:4321", + "right": "issue:5835", + "accept": false, + "reason": "Black images in ONNX DirectML vs pretrained MusicLDM usage; unrelated." + }, + { + "left": "issue:11363", + "right": "issue:8429", + "accept": false, + "reason": "ControlNet example OOM vs I2VGen-XL script malfunction; different pipelines and failures." + }, + { + "left": "issue:5135", + "right": "issue:8472", + "accept": false, + "reason": "Both LoRA-adjacent, but one is quality degradation and the other is about simultaneous pretraining with other modules." + }, + { + "left": "issue:5567", + "right": "issue:7982", + "accept": false, + "reason": "Accelerate overwriting unet.config vs MultiControlNet from_pipe failure; unrelated code paths." + }, + { + "left": "issue:7180", + "right": "issue:7706", + "accept": false, + "reason": "ControlNet cpu-offload failure vs custom pipeline OOM; both memory-related but not the same bug." + }, + { + "left": "issue:1851", + "right": "issue:730", + "accept": false, + "reason": "Multi-GPU slowdown vs local DreamBooth not learning; different symptoms and causes." + }, + { + "left": "issue:4109", + "right": "issue:7946", + "accept": false, + "reason": "SDXL inpaint pipeline failure vs custom_pipeline caching; unrelated." + }, + { + "left": "issue:5574", + "right": "issue:6008", + "accept": false, + "reason": "LoRA parameter-training question vs IP-Adapter/LCM LoRA failure; different concrete issues." + }, + { + "left": "issue:1608", + "right": "issue:4872", + "accept": false, + "reason": "Training halts at 500 steps vs CUDA OOM during finetuning; both on T4, but distinct failures." + }, + { + "left": "issue:2552", + "right": "issue:4057", + "accept": false, + "reason": "Tutorial question vs blurred output in finetuned inpainting model; not a duplicate." + }, + { + "left": "issue:5108", + "right": "issue:6274", + "accept": false, + "reason": "Both involve multiple adapters, but one is T2I Multiadapter XL exceptions and the other is generic multiple-adapter breakage; not enough to be the same bug." + }, + { + "left": "issue:6435", + "right": "issue:9686", + "accept": false, + "reason": "ControlNet default dimensions init bug vs SD3 controlnet pooled projections bug; different failures." + }, + { + "left": "issue:10085", + "right": "issue:10712", + "accept": false, + "reason": "Runtime errors, but one is LyCORIS batch inference with skip_guidance_layers and the other is SD3 prompt_embeds; different paths." + }, + { + "left": "issue:2276", + "right": "issue:6774", + "accept": false, + "reason": "xformers installation issue vs xformers mixed-precision dtype mismatch; unrelated." + }, + { + "left": "issue:4061", + "right": "issue:4491", + "accept": false, + "reason": "SDXL LoRA loading error vs SDXL refiner broken in latest diffusers; same area, different bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 355, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10026", + "issue:10277", + "issue:1070", + "issue:11127", + "issue:12009", + "issue:1214", + "issue:12689", + "issue:13001", + "issue:13077", + "issue:1546", + "issue:1616", + "issue:1791", + "issue:2244", + "issue:2312", + "issue:2447", + "issue:2578", + "issue:2858", + "issue:2907", + "issue:3199", + "issue:3213", + "issue:3631", + "issue:3884", + "issue:4282", + "issue:4665", + "issue:4686", + "issue:4854", + "issue:4923", + "issue:5619", + "issue:5794", + "issue:5829", + "issue:5967", + "issue:6240", + "issue:6793", + "issue:7102", + "issue:7801", + "issue:785", + "issue:7851", + "issue:7945", + "issue:7946", + "issue:8407", + "issue:8898", + "issue:9134", + "issue:9291", + "issue:9479", + "issue:9674", + "issue:9680", + "issue:9704", + "issue:9889" + ], + "soft_pairs": [ + "issue:4665|issue:5794", + "issue:9134|issue:9291", + "issue:12009|issue:13001", + "issue:1214|issue:2244", + "issue:7801|issue:9674", + "issue:7945|issue:9680", + "issue:7946|issue:8898", + "issue:1791|issue:2578", + "issue:3199|issue:3213", + "issue:4282|issue:7102", + "issue:5619|issue:5829", + "issue:7851|issue:9479", + "issue:12689|issue:13077", + "issue:1070|issue:4854", + "issue:6240|issue:9704", + "issue:4686|issue:6793", + "issue:1616|issue:2907", + "issue:2447|issue:3884", + "issue:5967|issue:9889", + "issue:1546|issue:785", + "issue:2858|issue:4923", + "issue:2312|issue:3631", + "issue:10026|issue:11127", + "issue:10277|issue:8407" + ], + "prepared_review_unit_hash": "90d631ecee4595e31983f28669bc9853b93eb6c9e8735b1b8933a10e70e86388", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 33786, + "estimated_input_tokens": 8447, + "estimated_eval_tokens": 17150 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: the paired issues share only broad keywords or subsystems, not the same concrete bug. All soft edges should be rejected.", + "confidence": 0.91, + "canonical_issue_reason": "No single issue is a good canonical for the whole set; the items span unrelated failures in loading, training, inference, schedulers, and pipeline utilities.", + "canonical_pr_reason": "No PRs are present in this cluster.", + "best_issue_reason": "If forced to pick one representative, issue 13077 is the most concrete and well-engaged bug report, but it still does not represent the rest of the cluster well.", + "best_pr_reason": "No PRs are present, so there is no best PR candidate.", + "soft_edge_verdicts": [ + { + "left": "issue:4665", + "right": "issue:5794", + "accept": false, + "reason": "Both concern loading assets, but one is a Windows safetensors/bin selection issue and the other is a missing tokenizer config in SDXL; different bugs." + }, + { + "left": "issue:9134", + "right": "issue:9291", + "accept": false, + "reason": "Both mention FLUX LoRAs, but one is a training issue with SimpleTuner and the other is key-format incompatibility from kohya-ss; not the same failure." + }, + { + "left": "issue:12009", + "right": "issue:13001", + "accept": false, + "reason": "Both are GGUF loading problems, but they involve different models and different errors: meta-tensor copy failure vs shape mismatch." + }, + { + "left": "issue:1214", + "right": "issue:2244", + "accept": false, + "reason": "Both involve half-precision, but one is CLIPTextModel precision behavior and the other is a Dreambooth CPU norm implementation error; not the same code path." + }, + { + "left": "issue:7801", + "right": "issue:9674", + "accept": false, + "reason": "One says fp16 mixed precision does not actually make the model fp16; the other is poor SD3 ControlNet training results. Different issues." + }, + { + "left": "issue:7945", + "right": "issue:9680", + "accept": false, + "reason": "One is a feature request about combining community pipelines; the other is SDXL unconditional generation being broken." + }, + { + "left": "issue:7946", + "right": "issue:8898", + "accept": false, + "reason": "Custom pipeline caching and subfolder from_pretrained loading are unrelated bugs." + }, + { + "left": "issue:1791", + "right": "issue:2578", + "accept": false, + "reason": "Mac/MPS support for UnCLIP and editable-install packaging failure are unrelated." + }, + { + "left": "issue:3199", + "right": "issue:3213", + "accept": false, + "reason": "Both mention broken pipelines, but one is a general custom-pipeline breakage and the other is a specific LPW pipeline regression." + }, + { + "left": "issue:4282", + "right": "issue:7102", + "accept": false, + "reason": "Adding Multi-ControlNet is a feature request; the other is a bug with multicontrolnet and multi-prompt handling." + }, + { + "left": "issue:5619", + "right": "issue:5829", + "accept": false, + "reason": "Both are LCM-related, but one is an AutoencodeTiny img2img encode bug and the other is a missing train-script config argument." + }, + { + "left": "issue:7851", + "right": "issue:9479", + "accept": false, + "reason": "Concurrent pipeline-call indexing errors and inpainting mask leakage are different bugs." + }, + { + "left": "issue:12689", + "right": "issue:13077", + "accept": false, + "reason": "One is checkpoint LoRA adapter weights not loading; the other is transformer weights being shallow-copied in ZImageControlNet. Different underlying problems." + }, + { + "left": "issue:1070", + "right": "issue:4854", + "accept": false, + "reason": "Both are inpainting-related, but one is a new-version bug and the other only affects the first image in a batch; not the same symptom or fix." + }, + { + "left": "issue:6240", + "right": "issue:9704", + "accept": false, + "reason": "device_map auto loading errors and a cuDNN frontend graph execution failure are unrelated." + }, + { + "left": "issue:4686", + "right": "issue:6793", + "accept": false, + "reason": "Local pipeline reload failure and Hydra serialization failure are different concerns." + }, + { + "left": "issue:1616", + "right": "issue:2907", + "accept": false, + "reason": "LPW pipeline compatibility and calling enable_model_cpu_offload twice are unrelated." + }, + { + "left": "issue:2447", + "right": "issue:3884", + "accept": false, + "reason": "Both are LoRA-related, but one is a generic LoRA error and the other is a specific training-script failure." + }, + { + "left": "issue:5967", + "right": "issue:9889", + "accept": false, + "reason": "An announcement about Stable Diffusion Turbo is not the same as a segfault in sdxl-turbo inference." + }, + { + "left": "issue:1546", + "right": "issue:785", + "accept": false, + "reason": "Truncation-warning batching error and PNDMScheduler performance cost are unrelated." + }, + { + "left": "issue:2858", + "right": "issue:4923", + "accept": false, + "reason": "UnCLIP seeding-image feature request and AudioLDM 2 transcript/text-to-speech feature request are different models and changes." + }, + { + "left": "issue:2312", + "right": "issue:3631", + "accept": false, + "reason": "EMA save-hook error and inpainting-latents exception are different training/inference failures." + }, + { + "left": "issue:10026", + "right": "issue:11127", + "accept": false, + "reason": "Multi-GPU DeepSpeed memory behavior and a single-file Flux model load failure are unrelated." + }, + { + "left": "issue:10277", + "right": "issue:8407", + "accept": false, + "reason": "Using Euler in fluxfill and UniPC scheduler kwargs support are separate scheduler issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 356, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10337", + "issue:10500", + "issue:10526", + "issue:10590", + "issue:1127", + "issue:1214", + "issue:1271", + "issue:13015", + "issue:1335", + "issue:13386", + "issue:1785", + "issue:1895", + "issue:2495", + "issue:2644", + "issue:3094", + "issue:3181", + "issue:3209", + "issue:3280", + "issue:3624", + "issue:3683", + "issue:3752", + "issue:3787", + "issue:4014", + "issue:4122", + "issue:4134", + "issue:4438", + "issue:4598", + "issue:4697", + "issue:4874", + "issue:499", + "issue:5000", + "issue:5353", + "issue:5463", + "issue:5509", + "issue:5546", + "issue:5596", + "issue:5646", + "issue:6278", + "issue:6639", + "issue:6661", + "issue:6807", + "issue:6869", + "issue:7450", + "issue:7548", + "issue:8609", + "issue:9461", + "issue:9530" + ], + "soft_pairs": [ + "issue:3624|issue:5509", + "issue:10500|issue:6869", + "issue:1214|issue:2644", + "issue:3094|issue:5596", + "issue:1271|issue:3181", + "issue:2495|issue:6661", + "issue:4122|issue:5546", + "issue:4874|issue:5463", + "issue:10526|issue:9461", + "issue:1895|issue:499", + "issue:3683|issue:6278", + "issue:3787|issue:4598", + "issue:4697|issue:6661", + "issue:13015|issue:13386", + "issue:1127|issue:1335", + "issue:3209|issue:7548", + "issue:3280|issue:3752", + "issue:4014|issue:4438", + "issue:4134|issue:6807", + "issue:6661|issue:6807", + "issue:10337|issue:10590", + "issue:5353|issue:8609", + "issue:5646|issue:7450", + "issue:1785|issue:5000", + "issue:6639|issue:9530" + ], + "prepared_review_unit_hash": "7aa38c32a591aba342a00f970d874c3ed720b5396aa393f79954f1eb02bdf338", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33945, + "estimated_input_tokens": 8487, + "estimated_eval_tokens": 17230 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the items cover unrelated model families, devices, and failure modes. None of the soft pairs looks like the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a valid duplicate canonical here; the titles span distinct bugs across different pipelines and error types. If forced to pick a representative, issue:10500 is the clearest runtime failure report, but it does not subsume the rest.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10500 is the best standalone issue to keep as a representative because it is concrete and well-scoped, but it is not a good merge target for the other issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3624", + "right": "issue:5509", + "accept": false, + "reason": "Different failures: generic getattr TypeError vs image/prompt batch-size mismatch." + }, + { + "left": "issue:10500", + "right": "issue:6869", + "accept": false, + "reason": "Both are runtime errors, but one is a device-mismatch crash and the other is an export_to_video unpacking bug." + }, + { + "left": "issue:1214", + "right": "issue:2644", + "accept": false, + "reason": "Half-precision CLIPTextModel bug vs a multi-GPU training question; not the same underlying problem." + }, + { + "left": "issue:3094", + "right": "issue:5596", + "accept": false, + "reason": "OOM during finetuning vs dict.convert error while building captions; unrelated code paths." + }, + { + "left": "issue:1271", + "right": "issue:3181", + "accept": false, + "reason": "Dependency installation problem vs TensorRT example TypeError; different subsystems." + }, + { + "left": "issue:2495", + "right": "issue:6661", + "accept": false, + "reason": "MPS black-image bug vs MPS OOM during training; same platform, different issues." + }, + { + "left": "issue:4122", + "right": "issue:5546", + "accept": false, + "reason": "Both mention scheduler step counts, but they target different pipeline setups and likely different bugs." + }, + { + "left": "issue:4874", + "right": "issue:5463", + "accept": false, + "reason": "Local low_res_scheduler bug vs unbound image_latents variable; not the same defect." + }, + { + "left": "issue:10526", + "right": "issue:9461", + "accept": false, + "reason": "Flux FP8 quantization TypeError vs a generic Flux inference issue; no shared concrete failure." + }, + { + "left": "issue:1895", + "right": "issue:499", + "accept": false, + "reason": "EMA multi-GPU update bug vs example training script failure; distinct behavior and scope." + }, + { + "left": "issue:3683", + "right": "issue:6278", + "accept": false, + "reason": "save_pretrained support request vs multi-GPU training run failure; unrelated changes." + }, + { + "left": "issue:3787", + "right": "issue:4598", + "accept": false, + "reason": "Partially initialized import traceback vs missing-dependency error messaging; different problem classes." + }, + { + "left": "issue:4697", + "right": "issue:6661", + "accept": false, + "reason": "LoRA loading regression vs MPS training OOM; not the same underlying bug." + }, + { + "left": "issue:13015", + "right": "issue:13386", + "accept": false, + "reason": "NPU same-device tensor error vs prior-preservation concatenation crash; different code paths." + }, + { + "left": "issue:1127", + "right": "issue:1335", + "accept": false, + "reason": "ONNX import traceback vs constructor missing vae_encoder; both ONNX-related, but not the same bug." + }, + { + "left": "issue:3209", + "right": "issue:7548", + "accept": false, + "reason": "AutoencoderKL NaNs vs flicker in AnimateDiff outputs; different symptoms and mechanisms." + }, + { + "left": "issue:3280", + "right": "issue:3752", + "accept": false, + "reason": "Scheduler incompatibility in DeepFloyd IF vs finetuned DeepFloyd IF inference issue; too broad to merge." + }, + { + "left": "issue:4014", + "right": "issue:4438", + "accept": false, + "reason": "Watermark artifacts vs generic SDXL error; no concrete shared failure." + }, + { + "left": "issue:4134", + "right": "issue:6807", + "accept": false, + "reason": "ControlNet tensor-shape mismatch on MPS vs MPS test collection failure; different layers of failure." + }, + { + "left": "issue:6661", + "right": "issue:6807", + "accept": false, + "reason": "Training OOM on MPS vs test collection failure on MPS; platform overlap only." + }, + { + "left": "issue:10337", + "right": "issue:10590", + "accept": false, + "reason": "Quantized Flux IPAdapter failure vs Sana BF16 tiled VAE decode issue; different models and failure modes." + }, + { + "left": "issue:5353", + "right": "issue:8609", + "accept": false, + "reason": "Usage question about FreeU vs custom_pipeline/IP_adapter incompatibility; not duplicate bugs." + }, + { + "left": "issue:5646", + "right": "issue:7450", + "accept": false, + "reason": "Scheduler stability concerns vs AutoencoderKL multithreading issues; unrelated subsystems." + }, + { + "left": "issue:1785", + "right": "issue:5000", + "accept": false, + "reason": "UnCLIP num_images_per_prompt bug vs LoRA fuse/unfuse regression; different features." + }, + { + "left": "issue:6639", + "right": "issue:9530", + "accept": false, + "reason": "Different SDXL LoRA-loading outputs vs GPU memory freeing after torch.compile; unrelated concerns." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 357, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10337", + "issue:10526", + "issue:10533", + "issue:10550", + "issue:10848", + "issue:11570", + "issue:1249", + "issue:1552", + "issue:1918", + "issue:1944", + "issue:1957", + "issue:2041", + "issue:2161", + "issue:2201", + "issue:2244", + "issue:2326", + "issue:2402", + "issue:2507", + "issue:2560", + "issue:2722", + "issue:3454", + "issue:3908", + "issue:3958", + "issue:4427", + "issue:4692", + "issue:4888", + "issue:5109", + "issue:5112", + "issue:5328", + "issue:579", + "issue:5805", + "issue:6028", + "issue:6110", + "issue:6218", + "issue:6222", + "issue:6297", + "issue:6335", + "issue:6517", + "issue:7354", + "issue:746", + "issue:7485", + "issue:7851", + "issue:8565", + "issue:9170", + "issue:9523", + "issue:954", + "issue:9927", + "issue:9941" + ], + "soft_pairs": [ + "issue:3958|issue:5328", + "issue:10337|issue:11570", + "issue:2161|issue:2326", + "issue:4692|issue:9170", + "issue:6218|issue:7485", + "issue:2402|issue:6028", + "issue:4888|issue:6110", + "issue:1552|issue:2201", + "issue:2041|issue:3454", + "issue:6517|issue:9941", + "issue:1957|issue:6297", + "issue:5109|issue:8565", + "issue:1944|issue:579", + "issue:746|issue:954", + "issue:9523|issue:9927", + "issue:2507|issue:3908", + "issue:2560|issue:2722", + "issue:1249|issue:1918", + "issue:6222|issue:6335", + "issue:10550|issue:10848", + "issue:2722|issue:5805", + "issue:2244|issue:5805", + "issue:4427|issue:7354", + "issue:10526|issue:10533", + "issue:5112|issue:7851" + ], + "prepared_review_unit_hash": "cb6d2980c060a9270c51cb3184981042a526baee3842bcc7720d70cf21ffcd1a", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 34220, + "estimated_input_tokens": 8555, + "estimated_eval_tokens": 17366 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set; it mixes many distinct bugs around quantization, LoRA/IP-Adapter loading, schedulers, offline loading, ControlNet, and generic runtime errors. I would not merge any soft pairs.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue: the items cover several unrelated failure modes, so any representative would be misleading as a duplicate anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10337 is the clearest representative of the quantized-adapter loading theme, but only as a loose topical anchor\u2014not a duplicate target for the whole cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3958", + "right": "issue:5328", + "accept": false, + "reason": "Both mention LoRA/Civitai and offload-related failures, but one is about model CPU offload and the other is a broader LoRA regression; not the same concrete bug." + }, + { + "left": "issue:10337", + "right": "issue:11570", + "accept": false, + "reason": "Different code paths and symptoms: IPAdapter + quantized transformers vs load_lora_weights with bitsandbytes quantized text_encoder." + }, + { + "left": "issue:2161", + "right": "issue:2326", + "accept": false, + "reason": "Training LoRA on an inpainting model is unrelated to converting a LoRA to safetensors/ckpt." + }, + { + "left": "issue:4692", + "right": "issue:9170", + "accept": false, + "reason": "Both are memory-related, but one is a UNet memory question and the other is SDXL+ControlNet VRAM requirements; not the same bug." + }, + { + "left": "issue:6218", + "right": "issue:7485", + "accept": false, + "reason": "Batched seeds in ControlNet pipelines and a NoneType safety_checker error are separate pipeline issues." + }, + { + "left": "issue:2402", + "right": "issue:6028", + "accept": false, + "reason": "CUDA not enabled is an environment/setup error, while Kandinsky 3.0 OOM is a runtime memory exhaustion issue." + }, + { + "left": "issue:4888", + "right": "issue:6110", + "accept": false, + "reason": "Both involve offline loading, but one is generic safetensors model loading and the other is LoRA weights specifically; too broad to treat as one bug." + }, + { + "left": "issue:1552", + "right": "issue:2201", + "accept": false, + "reason": "Attention slicing failure and tokenizer padding differences are different functional areas." + }, + { + "left": "issue:2041", + "right": "issue:3454", + "accept": false, + "reason": "Windows compatibility for Dreambooth is not the same as train_dreambooth_lora.py failing on latest versions." + }, + { + "left": "issue:6517", + "right": "issue:9941", + "accept": false, + "reason": "Import failure for a specific ControlNet SDXL module is unrelated to a general Colab runtime failure." + }, + { + "left": "issue:1957", + "right": "issue:6297", + "accept": false, + "reason": "LR scheduler init in unconditional training is unrelated to Dreambooth SDXL training not working on photos." + }, + { + "left": "issue:5109", + "right": "issue:8565", + "accept": false, + "reason": "Both are LoRA-loading errors, but they hit different pipelines and different failure modes (exception vs IndexError)." + }, + { + "left": "issue:1944", + "right": "issue:579", + "accept": false, + "reason": "Dreambooth output quality and missing config.json in textual inversion are unrelated." + }, + { + "left": "issue:746", + "right": "issue:954", + "accept": false, + "reason": "Onnx revision loading and post-login loading failures are different problems." + }, + { + "left": "issue:9523", + "right": "issue:9927", + "accept": false, + "reason": "ControlNet training mapping errors and HeaderTooLarge during SD3 ControlNet training are distinct errors." + }, + { + "left": "issue:2507", + "right": "issue:3908", + "accept": false, + "reason": "Safetensors GPU load performance and Streamlit model caching are unrelated." + }, + { + "left": "issue:2560", + "right": "issue:2722", + "accept": false, + "reason": "List-of-generators output differences and ControlNet UNet conversion mismatch are different bugs." + }, + { + "left": "issue:1249", + "right": "issue:1918", + "accept": false, + "reason": "An OnnxPipeline numpy/device AttributeError is unrelated to a DDIM scheduler example issue." + }, + { + "left": "issue:6222", + "right": "issue:6335", + "accept": false, + "reason": "Batch-size dependence in DDPM scheduler and ControlNet img2img output sizing are separate pipeline behaviors." + }, + { + "left": "issue:10550", + "right": "issue:10848", + "accept": false, + "reason": "Quantized-base-model LoRA loading and load_model_dict_into_meta device keyword handling are different loading paths." + }, + { + "left": "issue:2722", + "right": "issue:5805", + "accept": false, + "reason": "ControlNet conversion mismatch and an XLA float/half type error are not the same failure." + }, + { + "left": "issue:2244", + "right": "issue:5805", + "accept": false, + "reason": "CPU half-precision normalization failure and XLA input/bias type mismatch are different runtime errors." + }, + { + "left": "issue:4427", + "right": "issue:7354", + "accept": false, + "reason": "A broken MultiAdapter example and PNDM/DDPM compatibility are unrelated." + }, + { + "left": "issue:10526", + "right": "issue:10533", + "accept": false, + "reason": "Different Flux problems: one is an optimum.quanto FP8 tensor construction TypeError, the other is black-image generation in fp16." + }, + { + "left": "issue:5112", + "right": "issue:7851", + "accept": false, + "reason": "A torch.Size float-item error in ControlNet img2img and an out-of-bounds index under concurrent calls are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 358, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10010", + "issue:10052", + "issue:10120", + "issue:1014", + "issue:10202", + "issue:10204", + "issue:10490", + "issue:10596", + "issue:1061", + "issue:10638", + "issue:10672", + "issue:10673", + "issue:1076", + "issue:10796", + "issue:1088", + "issue:10895", + "issue:11383", + "issue:11939", + "issue:1261", + "issue:12712", + "issue:1805", + "issue:1831", + "issue:2091", + "issue:2444", + "issue:3622", + "issue:3694", + "issue:4361", + "issue:4534", + "issue:4566", + "issue:4728", + "issue:5064", + "issue:5254", + "issue:5951", + "issue:6258", + "issue:6336", + "issue:6602", + "issue:7312", + "issue:7802", + "issue:8469", + "issue:8950", + "issue:9115", + "issue:9309", + "issue:9366", + "issue:9374", + "issue:9420", + "issue:9731", + "issue:9813" + ], + "soft_pairs": [ + "issue:9366|issue:9731", + "issue:8469|issue:9813", + "issue:1014|issue:1261", + "issue:2091|issue:5951", + "issue:4566|issue:4728", + "issue:5254|issue:6336", + "issue:3694|issue:5064", + "issue:4361|issue:6602", + "issue:1831|issue:4534", + "issue:10010|issue:9420", + "issue:10052|issue:10596", + "issue:10120|issue:8950", + "issue:10202|issue:9309", + "issue:10204|issue:6258", + "issue:10490|issue:7802", + "issue:1061|issue:3622", + "issue:10638|issue:9115", + "issue:10672|issue:7312", + "issue:10672|issue:9374", + "issue:10673|issue:12712", + "issue:1076|issue:1805", + "issue:10796|issue:11383", + "issue:1088|issue:2444", + "issue:10895|issue:11939" + ], + "prepared_review_unit_hash": "652ccde4e8922bb0dae2eed2e1debbb516eded724fb6b2b6cd67abef2d6e15f1", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 33220, + "estimated_input_tokens": 8305, + "estimated_eval_tokens": 16866 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose mix of unrelated Diffusers issues and feature requests, with several Flux-related items but no clear duplicate group. The soft pairs are mostly broad topical similarities rather than the same concrete bug or change, so none should be merged.", + "confidence": 0.91, + "canonical_issue_reason": "No single issue is a reliable canonical duplicate target here: the items span different pipelines, models, and failure modes. If forced to pick a representative issue, issue:9374 is the broadest recent Flux API-support request, but it is not a true duplicate anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9374 is the best representative of the recent Flux subgroup because it covers multiple Flux pipelines in one API-support request, but the cluster overall is too heterogeneous for a strong canonical choice.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9366", + "right": "issue:9731", + "accept": false, + "reason": "Different problem types: scheduler failure in SDXL image-to-image vs a training/how-to style request for Playground2.5 Lora use." + }, + { + "left": "issue:8469", + "right": "issue:9813", + "accept": false, + "reason": "Both touch IP-Adapter, but one reports corrupt face outputs and the other a batch-embedding mixing bug in a helper function; not the same code-path bug." + }, + { + "left": "issue:1014", + "right": "issue:1261", + "accept": false, + "reason": "Upgrade-related regression report vs xformers CUDA backward assertion; unrelated symptoms and subsystems." + }, + { + "left": "issue:2091", + "right": "issue:5951", + "accept": false, + "reason": "Checkpoint resume in multi-GPU text-to-image training vs PixArt training question; different scripts and issues." + }, + { + "left": "issue:4566", + "right": "issue:4728", + "accept": false, + "reason": "Both involve resuming training, but one is SDXL LoRA checkpoint resume and the other is GPU OOM on resume in text_to_image; not the same bug." + }, + { + "left": "issue:5254", + "right": "issue:6336", + "accept": false, + "reason": "Multi-ControlNet support request for an inpaint pipeline vs TensorRT VAE compilation error with AutoencoderTiny; unrelated changes." + }, + { + "left": "issue:3694", + "right": "issue:5064", + "accept": false, + "reason": "Missing model/config files during finetuning example setup vs out-of-space during refiner inference; different failures." + }, + { + "left": "issue:4361", + "right": "issue:6602", + "accept": false, + "reason": "Both are save/load related, but shared-tensor safetensors serialization and fuse_lora/save_pretrained loading are distinct mechanisms and bugs." + }, + { + "left": "issue:1831", + "right": "issue:4534", + "accept": false, + "reason": "Dreambooth class-image precision mismatch vs SDXL Dreambooth validation crash on MPS; different pipelines and failure modes." + }, + { + "left": "issue:10010", + "right": "issue:9420", + "accept": false, + "reason": "Both mention FLUX model loading, but one is a missing example URL/model_index.json and the other is a saved model that cannot be reloaded after FLUX+Hyper-SD LoRA; not the same fix." + }, + { + "left": "issue:10052", + "right": "issue:10596", + "accept": false, + "reason": "Syntax error on ctx vs invalid shape in FluxControlPipeline; completely different bugs." + }, + { + "left": "issue:10120", + "right": "issue:8950", + "accept": false, + "reason": "Memory consumption in Dreambooth+SD3 vs SD3 LoRA training bug; not the same underlying issue." + }, + { + "left": "issue:10202", + "right": "issue:9309", + "accept": false, + "reason": "Flux Control LoRA unloading state bug vs FluxControlNetPipeline failure on Flux schnell; related area but different concrete problems." + }, + { + "left": "issue:10204", + "right": "issue:6258", + "accept": false, + "reason": "Both describe quality degradation, but one is FluxFill inpainting over iterations and the other is Stable Video Diffusion forward chunking; different pipelines and causes." + }, + { + "left": "issue:10490", + "right": "issue:7802", + "accept": false, + "reason": "Infinity support request vs OLSS scheduler support request; separate feature requests." + }, + { + "left": "issue:1061", + "right": "issue:3622", + "accept": false, + "reason": "Docs/tutorial update request vs an import error in a tutorial; one is editorial, the other is a runtime issue." + }, + { + "left": "issue:10638", + "right": "issue:9115", + "accept": false, + "reason": "DPMSolverMultistepScheduler example failure vs AnimateDiff-Lightning/SDXL motion adapter error; different pipelines and bugs." + }, + { + "left": "issue:10672", + "right": "issue:7312", + "accept": false, + "reason": "callback_on_step_end support across pipelines vs AlignProp direct reward finetuning support; both feature requests, but not the same change." + }, + { + "left": "issue:10672", + "right": "issue:9374", + "accept": false, + "reason": "Both are API-support requests, but one is a callback hook and the other is AutoPipeline support for Flux pipelines; no shared concrete implementation target." + }, + { + "left": "issue:10673", + "right": "issue:12712", + "accept": false, + "reason": "AuraFlow tensor reshape error vs missing FLUX.2-dev-bnb-4bit model; unrelated issues." + }, + { + "left": "issue:1076", + "right": "issue:1805", + "accept": false, + "reason": "Stable-diffusion-v1-5 question vs adding a scalable diffusion pipeline; not the same underlying request." + }, + { + "left": "issue:10796", + "right": "issue:11383", + "accept": false, + "reason": "Both ask about LoRA docs/support, but for different models (HunyuanVideo vs HiDream); not duplicates." + }, + { + "left": "issue:1088", + "right": "issue:2444", + "accept": false, + "reason": "Both are Dreambooth training problems, but different scripts and errors (flax vs inpaint subprocess failure)." + }, + { + "left": "issue:10895", + "right": "issue:11939", + "accept": false, + "reason": "Deprecation cleanup request vs Flux ControlNet training code update; different scopes and fix paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 359, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:10932", + "issue:10963", + "issue:11047", + "issue:11055", + "issue:1117", + "issue:11220", + "issue:11343", + "issue:11362", + "issue:11418", + "issue:11547", + "issue:11830", + "issue:11839", + "issue:1197", + "issue:1205", + "issue:1214", + "issue:12195", + "issue:12308", + "issue:1249", + "issue:1255", + "issue:1290", + "issue:13001", + "issue:1304", + "issue:1310", + "issue:13484", + "issue:1358", + "issue:1363", + "issue:1375", + "issue:1810", + "issue:1869", + "issue:305", + "issue:3213", + "issue:3300", + "issue:338", + "issue:339", + "issue:3724", + "issue:4071", + "issue:4646", + "issue:4702", + "issue:8317", + "issue:8557", + "issue:9188", + "issue:9223", + "issue:9224", + "issue:9239", + "issue:9438", + "issue:9637", + "issue:967", + "issue:9958" + ], + "soft_pairs": [ + "issue:10932|issue:9223", + "issue:10932|issue:9637", + "issue:10963|issue:11343", + "issue:11047|issue:12195", + "issue:11047|issue:9239", + "issue:11055|issue:8557", + "issue:1117|issue:338", + "issue:1117|issue:339", + "issue:11220|issue:9438", + "issue:11362|issue:8317", + "issue:11418|issue:13484", + "issue:11547|issue:9188", + "issue:11830|issue:9224", + "issue:11839|issue:13001", + "issue:1197|issue:1290", + "issue:1197|issue:3724", + "issue:1205|issue:4646", + "issue:1214|issue:967", + "issue:12308|issue:9958", + "issue:1249|issue:1810", + "issue:1255|issue:3300", + "issue:1304|issue:4702", + "issue:1310|issue:1869", + "issue:1358|issue:4071", + "issue:1363|issue:3213", + "issue:1375|issue:305" + ], + "prepared_review_unit_hash": "956b1c830d4bb02cbba2e5ee88622d46a2528b9845a8f50175fc0b6b2de4e265", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 33590, + "estimated_input_tokens": 8398, + "estimated_eval_tokens": 17052 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a broad false positive: the items span unrelated bugs, docs, model-loading, and pipeline issues across different code paths. There is no single underlying defect or change to treat as a canonical duplicate.", + "confidence": 0.99, + "canonical_issue_reason": "No clear canonical issue: the cluster mixes unrelated topics (compilation limits, import errors, controlnet/inpaint, training artifacts, docs, GGUF shape mismatches, etc.) without a shared code path.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue in this set is a good global representative, because any choice would misrepresent the rest of the cluster as the same bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10932", + "right": "issue:9223", + "accept": false, + "reason": "Different problems: torch._dynamo recompilation in HunyuanVideo vs dependency version conflicts around pyarrow/datasets." + }, + { + "left": "issue:10932", + "right": "issue:9637", + "accept": false, + "reason": "Unrelated code paths: HunyuanVideo compile/recompile_limit vs XFormers attention-mask bfloat16 edge case." + }, + { + "left": "issue:10963", + "right": "issue:11343", + "accept": false, + "reason": "Both are import failures, but one is a missing symbol and the other is a transformers-version compatibility break; not the same bug." + }, + { + "left": "issue:11047", + "right": "issue:12195", + "accept": false, + "reason": "Different model/component and failure: missing ControlNet pag inpaint model vs AutoencoderKL custom_op on older PyTorch." + }, + { + "left": "issue:11047", + "right": "issue:9239", + "accept": false, + "reason": "ControlNet model availability issue is unrelated to black-image output from ControlNet-Canny." + }, + { + "left": "issue:11055", + "right": "issue:8557", + "accept": false, + "reason": "Training colorization artifact for unconditional image generation is not the same as SD3 token-limit training behavior." + }, + { + "left": "issue:1117", + "right": "issue:338", + "accept": false, + "reason": "Both mention negatives/schedulers in tests, but one is multiple negative prompts and the other is LMS Discrete tests; different scopes." + }, + { + "left": "issue:1117", + "right": "issue:339", + "accept": false, + "reason": "Different scheduler test targets: negative prompt behavior vs Karras VE tests." + }, + { + "left": "issue:11220", + "right": "issue:9438", + "accept": false, + "reason": "Docs page not working is not the same as a quicktour page 404 caused by model_name in the URL." + }, + { + "left": "issue:11362", + "right": "issue:8317", + "accept": false, + "reason": "Both involve Autoencoder-style models, but one is an unexpected keyword in HiDreamImageTransformer2DModel and the other is a missing generator arg in AsymmetricAutoencoderKL." + }, + { + "left": "issue:11418", + "right": "issue:13484", + "accept": false, + "reason": "Adding a Flux fill model file is not the same as LoRA loading failures for Flux 2 dev FAL LoRAs." + }, + { + "left": "issue:11547", + "right": "issue:9188", + "accept": false, + "reason": "ControlNet support request for Flux fill dev is different from Flux outpainting feature request." + }, + { + "left": "issue:11830", + "right": "issue:9224", + "accept": false, + "reason": "Black output in framepack with float16 is unrelated to hardcoded float64 on MPS." + }, + { + "left": "issue:11839", + "right": "issue:13001", + "accept": false, + "reason": "Both are GGUF loading issues, but they describe different pipelines/models and different errors (invalid input size vs shape mismatch)." + }, + { + "left": "issue:1197", + "right": "issue:1290", + "accept": false, + "reason": "Two community pipeline proposals with different concepts: OpenAI Diffusion vs DreamArtist." + }, + { + "left": "issue:1197", + "right": "issue:3724", + "accept": false, + "reason": "OpenAI Diffusion pipeline proposal is unrelated to a watermarking module discussion." + }, + { + "left": "issue:1205", + "right": "issue:4646", + "accept": false, + "reason": "ONNX conversion of SD models is unrelated to controlnet-to-onnx conversion failure." + }, + { + "left": "issue:1214", + "right": "issue:967", + "accept": false, + "reason": "Half-precision CLIPTextModel issue is not the same as scheduler incompatibility in OnnxStableDiffusionPipeline." + }, + { + "left": "issue:12308", + "right": "issue:9958", + "accept": false, + "reason": "flash-attn support for non-Hopper GPUs is unrelated to AutoencoderKLTemporalDecoder tiling support." + }, + { + "left": "issue:1249", + "right": "issue:1810", + "accept": false, + "reason": "OnnxPipeline numpy/device error is not the same as deprecated numpy type usage." + }, + { + "left": "issue:1255", + "right": "issue:3300", + "accept": false, + "reason": "Import error for one scheduler symbol is unrelated to a broad report that HuggingFace-hosted diffusers are broken." + }, + { + "left": "issue:1304", + "right": "issue:4702", + "accept": false, + "reason": "Training text-guided image-to-image diffusion is different from an example script request for training a VQ-VAE." + }, + { + "left": "issue:1310", + "right": "issue:1869", + "accept": false, + "reason": "K-Diffusion + Stable Diffusion pipeline is not the same underlying change as UnCLIP image/text interpolations." + }, + { + "left": "issue:1358", + "right": "issue:4071", + "accept": false, + "reason": "Safetensors support is unrelated to inpaint-only + LaMa support." + }, + { + "left": "issue:1363", + "right": "issue:3213", + "accept": false, + "reason": "ONNX long-prompt/LMS discrete scheduler error is unrelated to lpw_stable_diffusion being broken." + }, + { + "left": "issue:1375", + "right": "issue:305", + "accept": false, + "reason": "A PR breaking imports may resemble import issues generally, but these refer to different failures and likely different fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 360, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:1482", + "issue:1486", + "issue:1535", + "issue:1536", + "issue:1546", + "issue:1566", + "issue:1594", + "issue:1604", + "issue:1647", + "issue:1692", + "issue:1704", + "issue:1851", + "issue:1876", + "issue:1909", + "issue:1962", + "issue:1969", + "issue:199", + "issue:2068", + "issue:2163", + "issue:2192", + "issue:2225", + "issue:2319", + "issue:2359", + "issue:2371", + "issue:2375", + "issue:2390", + "issue:2416", + "issue:2532", + "issue:2616", + "issue:2618", + "issue:2775", + "issue:2800", + "issue:2908", + "issue:3087", + "issue:3162", + "issue:3279", + "issue:3444", + "issue:3674", + "issue:3694", + "issue:3817", + "issue:463", + "issue:4969", + "issue:543", + "issue:5462", + "issue:5824", + "issue:782", + "issue:834", + "issue:861" + ], + "soft_pairs": [ + "issue:1482|issue:4969", + "issue:1486|issue:2532", + "issue:1535|issue:1536", + "issue:1546|issue:543", + "issue:1566|issue:3279", + "issue:1594|issue:3444", + "issue:1594|issue:3817", + "issue:1604|issue:782", + "issue:1647|issue:1704", + "issue:1692|issue:2616", + "issue:1704|issue:199", + "issue:1851|issue:2371", + "issue:1876|issue:861", + "issue:1909|issue:3162", + "issue:1962|issue:2800", + "issue:1969|issue:2775", + "issue:2068|issue:834", + "issue:2163|issue:2908", + "issue:2192|issue:3674", + "issue:2225|issue:463", + "issue:2319|issue:5824", + "issue:2359|issue:5462", + "issue:2375|issue:3694", + "issue:2390|issue:3087", + "issue:2416|issue:2618" + ], + "prepared_review_unit_hash": "d6521d9c1c077224b5caa8a6f6d634d498d10b0578ba5331ab174554837dd125", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33722, + "estimated_input_tokens": 8431, + "estimated_eval_tokens": 17118 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are a loose collection of unrelated Diffusers issues spanning different pipelines, bugs, and feature requests. The soft pairs mostly share only superficial subsystem overlap, not the same concrete problem, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 3087 is the broadest, highest-activity request in the set and is the best representative if one issue must be chosen, but the cluster itself does not show a true duplicate theme.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 3087 has the strongest discussion signal and a broad community-facing feature scope compared with the other mostly narrow bug reports and questions.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1482", + "right": "issue:4969", + "accept": false, + "reason": "Different topics: a Stable Diffusion upscale pipeline request vs an import error for CrossAttention." + }, + { + "left": "issue:1486", + "right": "issue:2532", + "accept": false, + "reason": "DDIMScheduler prediction_type bug and panorama get_views bug are unrelated code paths." + }, + { + "left": "issue:1535", + "right": "issue:1536", + "accept": false, + "reason": "General image variation pipeline request vs lpw_stable_diffusion compatibility break." + }, + { + "left": "issue:1546", + "right": "issue:543", + "accept": false, + "reason": "Batched truncation warning error is not the same as DDP grad-stride warning." + }, + { + "left": "issue:1566", + "right": "issue:3279", + "accept": false, + "reason": "Both involve Dreambooth checkpoints, but one crashes after saving fp16 output and the other fails on resume from checkpoint." + }, + { + "left": "issue:1594", + "right": "issue:3444", + "accept": false, + "reason": "Video img2img request and audio-to-image integration are different generative tasks." + }, + { + "left": "issue:1594", + "right": "issue:3817", + "accept": false, + "reason": "Video generation from img2img and music-to-dance generation are distinct features." + }, + { + "left": "issue:1604", + "right": "issue:782", + "accept": false, + "reason": "LPW pipeline break in diffusers 0.9.0 is unrelated to the model-download auth_token regression." + }, + { + "left": "issue:1647", + "right": "issue:1704", + "accept": false, + "reason": "A color-correction usage question does not match an attention block question." + }, + { + "left": "issue:1692", + "right": "issue:2616", + "accept": false, + "reason": "Pickling a lambda and NoneType.replace during save_attn_procs are different failures." + }, + { + "left": "issue:1704", + "right": "issue:199", + "accept": false, + "reason": "A question about attention blocks is not a dead-code cleanup issue." + }, + { + "left": "issue:1851", + "right": "issue:2371", + "accept": false, + "reason": "Multi-GPU slowdown in Dreambooth training is unrelated to missing objects in Dreambooth output." + }, + { + "left": "issue:1876", + "right": "issue:861", + "accept": false, + "reason": "Both mention conversion, but one is about missing state_dict in checkpoints and the other about a ckpt-to-diffusers conversion failure." + }, + { + "left": "issue:1909", + "right": "issue:3162", + "accept": false, + "reason": "Generator list handling in pipelines and missing learned_classifier_free_sampling_embeddings are unrelated API issues." + }, + { + "left": "issue:1962", + "right": "issue:2800", + "accept": false, + "reason": "General video generation request and DiffEdit semantic editing are different use cases." + }, + { + "left": "issue:1969", + "right": "issue:2775", + "accept": false, + "reason": "Tensor view stride error and torch._dynamo.optimize misuse are separate runtime errors." + }, + { + "left": "issue:2068", + "right": "issue:834", + "accept": false, + "reason": "Model dtype conversion bug is not the same as CPU under-utilization." + }, + { + "left": "issue:2163", + "right": "issue:2908", + "accept": false, + "reason": "Both are training-script errors, but they affect different scripts and different mixed-precision/shaping failures." + }, + { + "left": "issue:2192", + "right": "issue:3674", + "accept": false, + "reason": "Missing CLI args in dreambooth_lora and metadata.jsonl parsing in text_to_image_lora are separate issues." + }, + { + "left": "issue:2225", + "right": "issue:463", + "accept": false, + "reason": "Safety checker usage outside a pipeline is not the same as a super-resolution model request." + }, + { + "left": "issue:2319", + "right": "issue:5824", + "accept": false, + "reason": "A typo in an original-stable-diffusion loader method is unrelated to SDXL time-id default behavior." + }, + { + "left": "issue:2359", + "right": "issue:5462", + "accept": false, + "reason": "Allowing latent output_type in img2img is different from adding img2img support for LCMs." + }, + { + "left": "issue:2375", + "right": "issue:3694", + "accept": false, + "reason": "safe_open NameError and missing model_index/config.json are unrelated file-loading failures." + }, + { + "left": "issue:2390", + "right": "issue:3087", + "accept": false, + "reason": "T2I-Adapter support and LyCORIS support are different model-family feature requests." + }, + { + "left": "issue:2416", + "right": "issue:2618", + "accept": false, + "reason": "xformers training slowdown and ControlNet FloatTensor image input are unrelated problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 361, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:2462", + "issue:2507", + "issue:2663", + "issue:272", + "issue:2755", + "issue:2763", + "issue:2785", + "issue:2800", + "issue:2871", + "issue:2883", + "issue:2907", + "issue:3029", + "issue:3085", + "issue:3116", + "issue:3139", + "issue:3142", + "issue:3182", + "issue:3223", + "issue:3251", + "issue:3283", + "issue:3366", + "issue:3403", + "issue:3406", + "issue:3411", + "issue:3444", + "issue:3461", + "issue:3481", + "issue:3551", + "issue:3574", + "issue:3774", + "issue:3817", + "issue:4071", + "issue:4103", + "issue:4400", + "issue:4450", + "issue:4476", + "issue:459", + "issue:4949", + "issue:5002", + "issue:5112", + "issue:5128", + "issue:5592", + "issue:5989", + "issue:6218", + "issue:6560", + "issue:6732", + "issue:938" + ], + "soft_pairs": [ + "issue:2462|issue:459", + "issue:2507|issue:2663", + "issue:272|issue:4400", + "issue:2755|issue:5989", + "issue:2763|issue:5112", + "issue:2785|issue:4071", + "issue:2800|issue:938", + "issue:2871|issue:6732", + "issue:2883|issue:4949", + "issue:2907|issue:5592", + "issue:3029|issue:4476", + "issue:3085|issue:5002", + "issue:3116|issue:3481", + "issue:3139|issue:3283", + "issue:3142|issue:3223", + "issue:3182|issue:6218", + "issue:3251|issue:6732", + "issue:3366|issue:3574", + "issue:3403|issue:5128", + "issue:3406|issue:4103", + "issue:3411|issue:4450", + "issue:3444|issue:3817", + "issue:3461|issue:6560", + "issue:3551|issue:3774" + ], + "prepared_review_unit_hash": "3fe7ab1870604fdb75593c2450b9f1a175a8777dcd43013222b87dbc46ae5398", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32811, + "estimated_input_tokens": 8203, + "estimated_eval_tokens": 16662 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set: it mixes many unrelated Stable Diffusion/diffusers bugs, feature requests, and API questions across loading, schedulers, ControlNet, LoRA, pipelines, and docs. None of the soft pairs look like the same underlying change or defect.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2462", + "right": "issue:459", + "accept": false, + "reason": "Different topics: iterable dataset fine-tuning support vs a general diffusion model concept request." + }, + { + "left": "issue:2507", + "right": "issue:2663", + "accept": false, + "reason": "Both mention safetensors, but one is a performance issue on GPU loading and the other is a model-loading failure for a specific safety checker file." + }, + { + "left": "issue:272", + "right": "issue:4400", + "accept": false, + "reason": "Inpainting K-LMS scheduler support is unrelated to DreamBooth inpaint LoRA text-encoder support." + }, + { + "left": "issue:2755", + "right": "issue:5989", + "accept": false, + "reason": "One is a Stable Diffusion float16 loading bug; the other is a generic safetensors.torch bug with no clear same code path." + }, + { + "left": "issue:2763", + "right": "issue:5112", + "accept": false, + "reason": "ControlNet VRAM/cudnn behavior is a different problem from an SDXL ControlNet dtype/shape error." + }, + { + "left": "issue:2785", + "right": "issue:4071", + "accept": false, + "reason": "Text-to-video M1/M2 support and inpaint-only lama support are separate feature requests." + }, + { + "left": "issue:2800", + "right": "issue:938", + "accept": false, + "reason": "DiffEdit semantic image editing and human motion diffusion are unrelated features." + }, + { + "left": "issue:2871", + "right": "issue:6732", + "accept": false, + "reason": "Stable Diffusion inpainting latent output handling is unrelated to scheduler documentation." + }, + { + "left": "issue:2883", + "right": "issue:4949", + "accept": false, + "reason": "Image-to-video and ProPainter video in/outpainting are different video-generation problems." + }, + { + "left": "issue:2907", + "right": "issue:5592", + "accept": false, + "reason": "Model CPU offload being called twice is unrelated to checkpoint loading for UniDiffuser." + }, + { + "left": "issue:3029", + "right": "issue:4476", + "accept": false, + "reason": "ControlNet custom resolution question is not the same as an SDXL ControlNet fill50k result mismatch." + }, + { + "left": "issue:3085", + "right": "issue:5002", + "accept": false, + "reason": "Cache-folder argument removal and missing scheduler argument are unrelated API breakages." + }, + { + "left": "issue:3116", + "right": "issue:3481", + "accept": false, + "reason": "KDPM2DiscreteScheduler image2image latent noise bug is not the same as DDIM step-dependent image dissimilarity." + }, + { + "left": "issue:3139", + "right": "issue:3283", + "accept": false, + "reason": "How to use safetensors for generation vs how to convert LoRA safetensors to .bin are different usage questions." + }, + { + "left": "issue:3142", + "right": "issue:3223", + "accept": false, + "reason": "Safety checker placement and RAM not clearing after moving pipeline to CUDA are separate memory/inference issues." + }, + { + "left": "issue:3182", + "right": "issue:6218", + "accept": false, + "reason": "LPW Stable Diffusion CPU offload incompatibility is not the same as batched seeds failing in ControlNet pipelines." + }, + { + "left": "issue:3251", + "right": "issue:6732", + "accept": false, + "reason": "ControlNet Shuffle behavior is unrelated to scheduler documentation." + }, + { + "left": "issue:3366", + "right": "issue:3574", + "accept": false, + "reason": "LoRALinearLayer autocast regression and Windows from_ckpt loading issues are different bugs." + }, + { + "left": "issue:3403", + "right": "issue:5128", + "accept": false, + "reason": "TorchDynamo pipeline loading and local safetensors loading failure are unrelated loading paths." + }, + { + "left": "issue:3406", + "right": "issue:4103", + "accept": false, + "reason": "Attention class only_cross_attention bug and AutoencoderKL scaling_factor handling are separate model components." + }, + { + "left": "issue:3411", + "right": "issue:4450", + "accept": false, + "reason": "DreamBooth LoRA training test coverage and multi-text inversion failure are different training/example issues." + }, + { + "left": "issue:3444", + "right": "issue:3817", + "accept": false, + "reason": "ImageBind/StableUnCLIP audio-to-image integration and editable dance generation from music are distinct projects." + }, + { + "left": "issue:3461", + "right": "issue:6560", + "accept": false, + "reason": "DreamBooth script performance/memory comparison is unrelated to LoRA shape-size regression." + }, + { + "left": "issue:3551", + "right": "issue:3774", + "accept": false, + "reason": "LoRA disabling xFormers and zero3 not saving GPU memory are different optimization/runtime problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 362, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:3566", + "issue:3580", + "issue:3624", + "issue:3658", + "issue:3677", + "issue:3688", + "issue:3693", + "issue:3750", + "issue:3788", + "issue:3809", + "issue:3816", + "issue:3946", + "issue:3967", + "issue:3986", + "issue:4001", + "issue:4043", + "issue:4069", + "issue:4071", + "issue:4171", + "issue:4194", + "issue:4222", + "issue:4241", + "issue:4349", + "issue:4353", + "issue:4354", + "issue:4436", + "issue:4456", + "issue:5075", + "issue:5079", + "issue:5215", + "issue:5281", + "issue:5320", + "issue:5403", + "issue:5408", + "issue:5410", + "issue:5476", + "issue:557", + "issue:6529", + "issue:6685", + "issue:7061", + "issue:7183", + "issue:7392", + "issue:7418", + "issue:7509", + "issue:8320", + "issue:8440", + "issue:8637" + ], + "soft_pairs": [ + "issue:3566|issue:557", + "issue:3580|issue:5075", + "issue:3624|issue:7183", + "issue:3658|issue:5403", + "issue:3677|issue:3688", + "issue:3677|issue:5476", + "issue:3693|issue:7061", + "issue:3750|issue:5408", + "issue:3788|issue:7509", + "issue:3809|issue:7418", + "issue:3816|issue:4456", + "issue:3946|issue:8637", + "issue:3967|issue:8320", + "issue:3986|issue:5079", + "issue:4001|issue:5320", + "issue:4043|issue:5281", + "issue:4069|issue:4354", + "issue:4071|issue:6529", + "issue:4171|issue:4436", + "issue:4194|issue:8440", + "issue:4222|issue:7392", + "issue:4241|issue:7392", + "issue:4349|issue:5215", + "issue:4353|issue:5410", + "issue:4354|issue:6685" + ], + "prepared_review_unit_hash": "99e2951d7779c2e22f044bca4f66adcf6ff86d43094a8304e3ca28c085471ba8", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 33304, + "estimated_input_tokens": 8326, + "estimated_eval_tokens": 16908 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicate pairs stand out here: the items span unrelated bugs, feature requests, and docs issues across different pipelines and code paths. All soft edges look like superficial title similarity only.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3566", + "right": "issue:557", + "accept": false, + "reason": "Different topics: a community controlnet_reference pipeline issue vs a DDIM quality comparison; no shared bug path." + }, + { + "left": "issue:3580", + "right": "issue:5075", + "accept": false, + "reason": "Both mention inference/runtime errors, but one is an xFormers TypeError and the other is an SDXL refiner shape mismatch; distinct failures." + }, + { + "left": "issue:3624", + "right": "issue:7183", + "accept": false, + "reason": "Unrelated errors: getattr TypeError vs scheduler class-name metadata reporting." + }, + { + "left": "issue:3658", + "right": "issue:5403", + "accept": false, + "reason": "Dreambooth upsampler training bug vs T2I adapter training script question; different scripts and symptoms." + }, + { + "left": "issue:3677", + "right": "issue:3688", + "accept": false, + "reason": "General Kandinsky feature request vs a specific LoRA/PT 2.0 optimization issue; not the same change." + }, + { + "left": "issue:3677", + "right": "issue:5476", + "accept": false, + "reason": "Kandinsky support request vs a DreamBooth/xformers compatibility question; unrelated scope." + }, + { + "left": "issue:3693", + "right": "issue:7061", + "accept": false, + "reason": "Both are doc-related, but one is Kandinsky docstrings and the other is imagic pipeline inference docs; different files and problems." + }, + { + "left": "issue:3750", + "right": "issue:5408", + "accept": false, + "reason": "KDiffusion LoRA-loading bug vs bad inference with kohya-style LoRA; separate issues in different paths." + }, + { + "left": "issue:3788", + "right": "issue:7509", + "accept": false, + "reason": "New pipeline request (UniControl) vs ReNoise pipeline proposal; clearly different additions." + }, + { + "left": "issue:3809", + "right": "issue:7418", + "accept": false, + "reason": "DDP stride-mismatch warning vs a specific bug in an ldm3d community script; not the same issue." + }, + { + "left": "issue:3816", + "right": "issue:4456", + "accept": false, + "reason": "Different import/load failures: missing CLIPImageProcessor vs ModuleList attribute error in LoRA loading." + }, + { + "left": "issue:3946", + "right": "issue:8637", + "accept": false, + "reason": "Both involve naming/import mechanics, but one is about from_ckpt naming and the other an import failure for PixArtSigmaPipeline; unrelated." + }, + { + "left": "issue:3967", + "right": "issue:8320", + "accept": false, + "reason": "SDXL branch/deletion regression vs from_single_file __name__ error; distinct regressions." + }, + { + "left": "issue:3986", + "right": "issue:5079", + "accept": false, + "reason": "Support for mesh output in ShapEPipeline vs loading LyCORIS weights; different features." + }, + { + "left": "issue:4001", + "right": "issue:5320", + "accept": false, + "reason": "README/demo docs request vs PEFT API change for fuse_lora; not the same artifact or bug." + }, + { + "left": "issue:4043", + "right": "issue:5281", + "accept": false, + "reason": "SDXL prompt/pool embedding requirement vs multi-pipeline CPU offload device mismatch; separate SDXL bugs." + }, + { + "left": "issue:4069", + "right": "issue:4354", + "accept": false, + "reason": "Textual inversion unload behavior vs from_single_file returning wrong pipeline type; different code paths." + }, + { + "left": "issue:4071", + "right": "issue:6529", + "accept": false, + "reason": "Inpaint-only+lama support request vs SDXL Fooocus Inpaint discussion; one is a feature request, the other a pipeline-specific topic." + }, + { + "left": "issue:4171", + "right": "issue:4436", + "accept": false, + "reason": "Docs rendering issue on the reference page vs CIVITAI safetensors conversion; unrelated." + }, + { + "left": "issue:4194", + "right": "issue:8440", + "accept": false, + "reason": "from_single_file VAE loading issue vs FromOriginalModelMixin subclass ValueError; different mechanics." + }, + { + "left": "issue:4222", + "right": "issue:7392", + "accept": false, + "reason": "General LDM training on low-resolution data vs SDXL multi-aspect-ratio training question; not a duplicate bug report." + }, + { + "left": "issue:4241", + "right": "issue:7392", + "accept": false, + "reason": "LoRA fine-tuning support for unidiffusers vs SDXL aspect-ratio training; different requests." + }, + { + "left": "issue:4349", + "right": "issue:5215", + "accept": false, + "reason": "Upscale pipeline missing text_encoder despite prompt_embeds vs brightness shift on dark images; separate failures." + }, + { + "left": "issue:4353", + "right": "issue:5410", + "accept": false, + "reason": "General SDXL poor-quality report vs latent upscale blur; too broad and not the same concrete bug." + }, + { + "left": "issue:4354", + "right": "issue:6685", + "accept": false, + "reason": "Wrong pipeline type from_single_file vs DreamBooth absolute output_dir paths; unrelated bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 363, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:4363", + "issue:4445", + "issue:4453", + "issue:4468", + "issue:4504", + "issue:4593", + "issue:4634", + "issue:4638", + "issue:4641", + "issue:4667", + "issue:4672", + "issue:4675", + "issue:4681", + "issue:4708", + "issue:4724", + "issue:4764", + "issue:4877", + "issue:5022", + "issue:5061", + "issue:5081", + "issue:5084", + "issue:5128", + "issue:5138", + "issue:525", + "issue:5328", + "issue:5374", + "issue:5429", + "issue:5760", + "issue:5777", + "issue:5783", + "issue:5791", + "issue:5956", + "issue:5972", + "issue:6388", + "issue:6412", + "issue:6496", + "issue:6791", + "issue:7177", + "issue:7405", + "issue:7454", + "issue:770", + "issue:8537", + "issue:8742", + "issue:8771", + "issue:8803", + "issue:8831", + "issue:9101" + ], + "soft_pairs": [ + "issue:4363|issue:4453", + "issue:4445|issue:7454", + "issue:4468|issue:5956", + "issue:4504|issue:5328", + "issue:4593|issue:8831", + "issue:4634|issue:4638", + "issue:4641|issue:5328", + "issue:4667|issue:8537", + "issue:4672|issue:9101", + "issue:4675|issue:6412", + "issue:4681|issue:4877", + "issue:4708|issue:5061", + "issue:4724|issue:5081", + "issue:4764|issue:6791", + "issue:5022|issue:8771", + "issue:5084|issue:8742", + "issue:5128|issue:5777", + "issue:5138|issue:6388", + "issue:525|issue:770", + "issue:5374|issue:7177", + "issue:5429|issue:6496", + "issue:5760|issue:7405", + "issue:5783|issue:8803", + "issue:5791|issue:5972" + ], + "prepared_review_unit_hash": "d844052030b6421b647b23f64565c2fb872ce65e2fbd33a067a374a0d92dbda7", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 32787, + "estimated_input_tokens": 8197, + "estimated_eval_tokens": 16650 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of SDXL/LoRA/training/loading issues, not true duplicates. The strongest single issue is the concrete LoRA safetensor loading bug, but the rest are different bugs or feature requests in nearby areas.", + "confidence": 0.86, + "canonical_issue_reason": "issue:4504 is the most concrete, reproducible bug report in the set: a specific LoRA safetensor load path failing with a Python exception and enough detail to diagnose.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:4504 is the best representative issue overall because it is a precise code-path failure rather than a broad question or feature request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4363", + "right": "issue:4453", + "accept": false, + "reason": "Both mention SDXL and safetensors, but one is about loading a checkpoint and the other about converting LoRA weights for A1111; different tasks and failure modes." + }, + { + "left": "issue:4445", + "right": "issue:7454", + "accept": false, + "reason": "A how-to question about finetuning LoRA is not the same as a broad LoRA support feature request." + }, + { + "left": "issue:4468", + "right": "issue:5956", + "accept": false, + "reason": "Both involve performance, but one is multi-GPU sample-code performance and the other is degraded SD-XL U-Net fine-tuning quality/performance; not the same bug." + }, + { + "left": "issue:4504", + "right": "issue:5328", + "accept": false, + "reason": "Both are LoRA loading complaints, but 4504 is a specific state_dict exception while 5328 is a broad regression report about CivitAI LoRAs failing; too vague to call the same underlying bug." + }, + { + "left": "issue:4593", + "right": "issue:8831", + "accept": false, + "reason": "Both concern LoRA layer compatibility, but they describe different errors and different code paths." + }, + { + "left": "issue:4634", + "right": "issue:4638", + "accept": false, + "reason": "Custom diffusion safe serialization and SDXL textual inversion loading are unrelated loader issues." + }, + { + "left": "issue:4641", + "right": "issue:5328", + "accept": false, + "reason": "Both are LoRA loading problems, but one is a local-file load_lora_weights issue and the other is a broad \u201call failing\u201d report; insufficient evidence they are the same bug." + }, + { + "left": "issue:4667", + "right": "issue:8537", + "accept": false, + "reason": "Both are from from_single_file, but they target different pipelines and different root causes." + }, + { + "left": "issue:4672", + "right": "issue:9101", + "accept": false, + "reason": "Training-memory reduction for SDXL LoRA is unrelated to adding an InstructPix2Pix training script for SD3." + }, + { + "left": "issue:4675", + "right": "issue:6412", + "accept": false, + "reason": "ControlNet Reference Only for SD XL and ControlNet annotators to ONNX are distinct features." + }, + { + "left": "issue:4681", + "right": "issue:4877", + "accept": false, + "reason": "Transformer2DModel positional embeddings and making noise_init_strength configurable are unrelated changes." + }, + { + "left": "issue:4708", + "right": "issue:5061", + "accept": false, + "reason": "An import error in diffusers.utils is not the same as an argument error in a conversion script." + }, + { + "left": "issue:4724", + "right": "issue:5081", + "accept": false, + "reason": "Both involve SDXL inference, but one is an img2img checkpoint behavior issue and the other is a tensor-shape mismatch in LPW XL." + }, + { + "left": "issue:4764", + "right": "issue:6791", + "accept": false, + "reason": "A LoRA weight compatibility problem is unrelated to a training script failing to run." + }, + { + "left": "issue:5022", + "right": "issue:8771", + "accept": false, + "reason": "One is a missing-name import error; the other is about removing LoRAAttnProcessor and downstream dependency breakage." + }, + { + "left": "issue:5084", + "right": "issue:8742", + "accept": false, + "reason": "FlaxStableDiffusionXLPipeline import failure and LoRAAttnProcessor2_0 removal are different compatibility problems." + }, + { + "left": "issue:5128", + "right": "issue:5777", + "accept": false, + "reason": "Loading a local safetensors file is unrelated to a PixArt DreamBooth training error." + }, + { + "left": "issue:5138", + "right": "issue:6388", + "accept": false, + "reason": "Moving compute_snr to training_utils is a refactor request, not the same as gradient checkpointing being ignored." + }, + { + "left": "issue:525", + "right": "issue:770", + "accept": false, + "reason": "Local-disk image loading in a notebook and Colab startup failures are different training-environment issues." + }, + { + "left": "issue:5374", + "right": "issue:7177", + "accept": false, + "reason": "An unexpected keyword argument and a StableCascade from_pretrained error are different failure modes." + }, + { + "left": "issue:5429", + "right": "issue:6496", + "accept": false, + "reason": "Support for non-DreamBooth examples without VAE is not the same as a multiple-person DreamBooth example request." + }, + { + "left": "issue:5760", + "right": "issue:7405", + "accept": false, + "reason": "These are entirely different research-model issues with no shared bug context." + }, + { + "left": "issue:5783", + "right": "issue:8803", + "accept": false, + "reason": "EMA not working in an SDXL training script is unrelated to a duplicate assignment in attention_processor.py." + }, + { + "left": "issue:5791", + "right": "issue:5972", + "accept": false, + "reason": "A training-parameter interaction report is not the same as a documentation note about Min-SNR weighting." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 102, + "review_unit_index": 364, + "review_unit_count": 364, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-50-3014", + "nodes": [ + "issue:5805", + "issue:5905", + "issue:5925", + "issue:5947", + "issue:5977", + "issue:5980", + "issue:6079", + "issue:6091", + "issue:6092", + "issue:6093", + "issue:6203", + "issue:6240", + "issue:6256", + "issue:6313", + "issue:6388", + "issue:6457", + "issue:6753", + "issue:7112", + "issue:7220", + "issue:7438", + "issue:7440", + "issue:7469", + "issue:7798", + "issue:7836", + "issue:7841", + "issue:7871", + "issue:7962", + "issue:8320", + "issue:8485", + "issue:8500", + "issue:8504", + "issue:8534", + "issue:8540", + "issue:8543", + "issue:8546", + "issue:8645", + "issue:8649", + "issue:8731", + "issue:9032", + "issue:9039", + "issue:9069", + "issue:9220", + "issue:9448", + "issue:9477", + "issue:9511", + "issue:9674", + "issue:9873", + "issue:9974" + ], + "soft_pairs": [ + "issue:5805|issue:6079", + "issue:5905|issue:5947", + "issue:5925|issue:9511", + "issue:5977|issue:6093", + "issue:5980|issue:6093", + "issue:6091|issue:6753", + "issue:6092|issue:9674", + "issue:6203|issue:6388", + "issue:6240|issue:7871", + "issue:6256|issue:7962", + "issue:6313|issue:9873", + "issue:6457|issue:8543", + "issue:7112|issue:7440", + "issue:7220|issue:9448", + "issue:7438|issue:8645", + "issue:7469|issue:9220", + "issue:7798|issue:7836", + "issue:7841|issue:8731", + "issue:8320|issue:9477", + "issue:8485|issue:8540", + "issue:8500|issue:8534", + "issue:8504|issue:8540", + "issue:8546|issue:8649", + "issue:9032|issue:9039", + "issue:9069|issue:9974" + ], + "prepared_review_unit_hash": "301c6d5f3c950188f57f2ae8bef5cee252ee2a960a218d888190524d243aa413", + "component_budget": { + "node_count": 3014, + "item_count": 3014, + "soft_pair_count": 8931, + "serialized_chars": 4765605, + "estimated_input_tokens": 1191402, + "estimated_eval_tokens": 2383060 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 33362, + "estimated_input_tokens": 8341, + "estimated_eval_tokens": 16938 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the issues span many unrelated bugs and feature requests across SDXL, SD3, LoRA, ControlNet, IP-Adapter, Flax, and training scripts. None of the soft pairs looks like the same underlying change or defect.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue fits the cluster because the items are heterogeneous and do not share one root cause or code path.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue is appropriate here; any single issue would misrepresent the mixed set of unrelated reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5805", + "right": "issue:6079", + "accept": false, + "reason": "Both are dtype-mismatch runtime errors, but they occur in different pipelines/contexts and do not clearly point to the same bug." + }, + { + "left": "issue:5905", + "right": "issue:5947", + "accept": false, + "reason": "One is a request to add a diffusion-GAN model; the other is a policy/objective request for Diffusion-DPO. Different changes." + }, + { + "left": "issue:5925", + "right": "issue:9511", + "accept": false, + "reason": "ZipLoRA multi-LoRA inference and multi-ControlNet batching are unrelated features in different subsystems." + }, + { + "left": "issue:5977", + "right": "issue:6093", + "accept": false, + "reason": "Saving extra `unet.unet` keys and push_to_hub repo creation are different LoRA training issues with different fixes." + }, + { + "left": "issue:5980", + "right": "issue:6093", + "accept": false, + "reason": "ControlNet SDXL output file format question is unrelated to the consistency-model LoRA push_to_hub bug." + }, + { + "left": "issue:6091", + "right": "issue:6753", + "accept": false, + "reason": "Consistency decoder memory with size argument and VAE configuration are different problems." + }, + { + "left": "issue:6092", + "right": "issue:9674", + "accept": false, + "reason": "LoRA example not training and SD3 ControlNet producing bad results are separate training-path issues." + }, + { + "left": "issue:6203", + "right": "issue:6388", + "accept": false, + "reason": "SDXL ControlNet image-size conditioning and ignored gradient checkpointing in SDXL+LoRA are unrelated." + }, + { + "left": "issue:6240", + "right": "issue:7871", + "accept": false, + "reason": "AutoPipeline `device_map=\"auto\"` error and blockwise-scale LoRA error affect different code paths." + }, + { + "left": "issue:6256", + "right": "issue:7962", + "accept": false, + "reason": "Both are pipeline API feature requests, but for different callbacks/kwargs and not the same change." + }, + { + "left": "issue:6313", + "right": "issue:9873", + "accept": false, + "reason": "Null-text inversion pipeline and OmniGen pipeline are distinct model/pipeline requests." + }, + { + "left": "issue:6457", + "right": "issue:8543", + "accept": false, + "reason": "Multi-GPU gradient accumulation for SDXL-LoRA and SD3 dreambooth gradient-checkpointing are separate training bugs." + }, + { + "left": "issue:7112", + "right": "issue:7440", + "accept": false, + "reason": "IP-Adapter image embedding dimensions and dataset availability are unrelated." + }, + { + "left": "issue:7220", + "right": "issue:9448", + "accept": false, + "reason": "Both involve IP-Adapter in a ControlNet inpaint pipeline, but one is unload-time cleanup and the other is a shape error during use; different code paths." + }, + { + "left": "issue:7438", + "right": "issue:8645", + "accept": false, + "reason": "XPU generation stall and missing `create_diffusers_controlnet_model_from_ldm` are unrelated." + }, + { + "left": "issue:7469", + "right": "issue:9220", + "accept": false, + "reason": "AutoPipeline variable naming and PAG pipeline undefined variable are different bugs in different code." + }, + { + "left": "issue:7798", + "right": "issue:7836", + "accept": false, + "reason": "PixArtAlphaPipeline bug and CLIP training overfitting are unrelated." + }, + { + "left": "issue:7841", + "right": "issue:8731", + "accept": false, + "reason": "Visual artifact in SDXL output and batched text-to-image slowness are different symptoms and fixes." + }, + { + "left": "issue:8320", + "right": "issue:9477", + "accept": false, + "reason": "Single-file AutoencoderKL `__name__` error and `GatheredParameters` callable error are unrelated." + }, + { + "left": "issue:8485", + "right": "issue:8540", + "accept": false, + "reason": "SD3 LoRA support and SD3 dreambooth embedding/VAE caching are separate feature requests." + }, + { + "left": "issue:8500", + "right": "issue:8534", + "accept": false, + "reason": "SD3 token-limit question and wrong logit-normal weighting are different SD3 training concerns." + }, + { + "left": "issue:8504", + "right": "issue:8540", + "accept": false, + "reason": "Choosing text encoders for SD3-medium and caching embeddings/VAE representations are distinct enhancements." + }, + { + "left": "issue:8546", + "right": "issue:8649", + "accept": false, + "reason": "Missing SD3 from-single-file support and `num_images_per_prompt` regression are unrelated issues." + }, + { + "left": "issue:9032", + "right": "issue:9039", + "accept": false, + "reason": "Both are Flax pipeline questions, but TPU minimal example and LoRA loading are not the same underlying bug/change." + }, + { + "left": "issue:9069", + "right": "issue:9974", + "accept": false, + "reason": "A NumPy type error and adding expected parameters to `controlnet_sd3` are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 2, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10327-2", + "nodes": [ + "issue:10327", + "issue:10328" + ], + "soft_pairs": [ + "issue:10327|issue:10328" + ], + "prepared_review_unit_hash": "6f7f962f8abfd105de32b66fa98c7266889f3dbbf8c24e06621cc1e75d90faaf", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1577, + "estimated_input_tokens": 395, + "estimated_eval_tokens": 1046 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1577, + "estimated_input_tokens": 395, + "estimated_eval_tokens": 1046 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are closely related but not duplicates: one is a broad request to propagate applicable quantization settings during model loading, while the other is a specific request to support `optimum.quanto` as a load-time backend.", + "confidence": 0.83, + "canonical_issue_reason": "Issue 10327 is the broader umbrella: it describes the general load-time `quantization_config` application problem and can subsume backend-specific support requests like `optimum.quanto`.", + "canonical_pr_reason": null, + "best_issue_reason": "10327 is the better representative issue because it captures the underlying mechanism rather than one particular quantization backend.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10327", + "right": "issue:10328", + "accept": false, + "reason": "Related scope, but not the same underlying request: 10327 is a general model-loading quantization propagation issue, while 10328 is specifically about adding `optimum.quanto` support." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 3, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10375-2", + "nodes": [ + "issue:10375", + "issue:10433" + ], + "soft_pairs": [ + "issue:10375|issue:10433" + ], + "prepared_review_unit_hash": "95e8965e25668b8e7a1aa854f1354d44fd4981e7567122569c0056c436d41b2b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1528, + "estimated_input_tokens": 382, + "estimated_eval_tokens": 1020 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1528, + "estimated_input_tokens": 382, + "estimated_eval_tokens": 1020 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are closed documentation-link issues, but the evidence only shows broad topical overlap, not the same specific broken links or doc section. Treat them as related but not duplicates.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 10433 is the more specific report (\u201cBroken Links in a Section of Documentation\u201d) and likely represents the better canonical discussion target for this doc-links cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "10433 has the clearer scope and more detailed body, so it is the best issue to anchor the cluster if these are grouped together.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10375", + "right": "issue:10433", + "accept": false, + "reason": "Both are about broken documentation links, but neither identifies the same exact pages or failure mode; the overlap is too broad to treat as the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 4, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6309-3", + "nodes": [ + "issue:10414", + "issue:6309", + "issue:9120" + ], + "soft_pairs": [ + "issue:6309|issue:9120", + "issue:10414|issue:9120" + ], + "prepared_review_unit_hash": "488a2127cc3d37ea80d1ac3273f6be5fe048180d20e385e839bc71614bbf3b0e", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2422, + "estimated_input_tokens": 606, + "estimated_eval_tokens": 1468 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2422, + "estimated_input_tokens": 606, + "estimated_eval_tokens": 1468 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are separate translation requests for different target languages, not duplicates of the same issue. They share a template and topic, but each one is a distinct language-specific task.", + "confidence": 0.97, + "canonical_issue_reason": "No single canonical issue: each ticket targets a different language (Chinese, Japanese, Arabic), so they represent separate work items rather than the same bug/change.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 6309 is the best representative only in the sense that it is the most active/open translation request, but it is not a valid canonical for the others because the language target differs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6309", + "right": "issue:9120", + "accept": false, + "reason": "Different target languages (Japanese vs Arabic); these are separate translation requests, not the same underlying issue." + }, + { + "left": "issue:10414", + "right": "issue:9120", + "accept": false, + "reason": "Different target languages (Chinese vs Arabic); same template, but distinct work items." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 5, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9900-2", + "nodes": [ + "issue:10457", + "issue:9900" + ], + "soft_pairs": [ + "issue:10457|issue:9900" + ], + "prepared_review_unit_hash": "b1b9cb4ed84b8af2dfa62f237ed31a8be24817d8d43131a1acec8e21ac73095c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1471, + "estimated_input_tokens": 368, + "estimated_eval_tokens": 992 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1471, + "estimated_input_tokens": 368, + "estimated_eval_tokens": 992 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are both vague bug reports, but they appear to concern different areas: one is specifically about FluxPriorReduxPipeline comments, while the other is a generic repaint bug. I would not cluster them as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": "Issue #10457 is the more specific and informative report, with a concrete subsystem reference in the title.", + "canonical_pr_reason": null, + "best_issue_reason": "#10457 is the best representative issue because it is specific enough to anchor the cluster, whereas #9900 is too generic to serve as a canonical example.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10457", + "right": "issue:9900", + "accept": false, + "reason": "The titles point to different underlying problems and different code areas; the overlap is only generic bug language, not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 6, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1046-2", + "nodes": [ + "issue:1046", + "issue:3148" + ], + "soft_pairs": [ + "issue:1046|issue:3148" + ], + "prepared_review_unit_hash": "54e780f9838862f0e09d3f365125be59c4fd476d71448a874b722ded53814b31", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1497, + "estimated_input_tokens": 375, + "estimated_eval_tokens": 1006 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1497, + "estimated_input_tokens": 375, + "estimated_eval_tokens": 1006 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one is about loading custom pipelines via from_pretrained, the other about promoting community ControlNet pipelines. They share only a very broad pipeline/theme overlap.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1046 is the broader, more foundational topic; 3148 is a narrower community-ControlNet follow-up and not the same underlying problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1046", + "right": "issue:3148", + "accept": false, + "reason": "Different requests and code paths: custom pipeline loading/from_pretrained versus surfacing community ControlNet pipelines. The similarity is only at the high-level pipeline concept." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 7, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10562-2", + "nodes": [ + "issue:10562", + "issue:10599" + ], + "soft_pairs": [ + "issue:10562|issue:10599" + ], + "prepared_review_unit_hash": "6b78a850502c3f79064d013f8bfe43b3dd551b9065df416cc201a0fda17e74f6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1577, + "estimated_input_tokens": 395, + "estimated_eval_tokens": 1046 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1577, + "estimated_input_tokens": 395, + "estimated_eval_tokens": 1046 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues look closely related at the title level, but the packet only provides soft similarity and no body-level evidence, so they should not be treated as definite duplicates. Issue 10599 is still the better canonical representative because it is more specific.", + "confidence": 0.68, + "canonical_issue_reason": "issue:10599 is the clearer, more specific report: it names the exact error text and the Civitai/popular LoRA context.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10599 is the best representative issue because it is narrower and more actionable, but the available evidence is not strong enough to prove the two issues are the same bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10562", + "right": "issue:10599", + "accept": false, + "reason": "Only title-level similarity and a moderate soft-similarity score are available, with no body evidence, explicit target, or shared file/target evidence. This is enough to say the reports may be related, but not enough to accept them as the same underlying bug." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The representative choice is not sufficiently grounded. The packet only provides title-level similarity and a soft-similarity score (jaccard ~0.45) with no body evidence, so treating these as definite duplicates or the same underlying bug is too strong. The canonical-issue preference for 10599 is fine, but the soft-edge verdict should be conservative (e.g. \"possibly related\" rather than accepted as same bug)." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 8, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10621-2", + "nodes": [ + "issue:10621", + "issue:11752" + ], + "soft_pairs": [ + "issue:10621|issue:11752" + ], + "prepared_review_unit_hash": "6c7ea69710dcf176964aa7bb1d416dd2dcef51a4f1be4954e17e65ec59c8e34c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1594, + "estimated_input_tokens": 399, + "estimated_eval_tokens": 1054 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1594, + "estimated_input_tokens": 399, + "estimated_eval_tokens": 1054 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "#10621 remains the best canonical issue because it captures the exact TorchaoLoraLinear constructor failure, while #11752 is a broader int8 LoRA-loading report with only title-level overlap.", + "confidence": 0.79, + "canonical_issue_reason": "#10621 is the stronger canonical issue because it names the precise error path (`TorchaoLoraLinear.__init__()` missing `get_apply_tensor_subclass`) and ties it to quantized-model LoRA loading.", + "canonical_pr_reason": null, + "best_issue_reason": "#10621 is the best representative for the cluster since it gives the concrete failure signature; #11752 is too general to serve as the main issue without clearer body-level matching.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10621", + "right": "issue:11752", + "accept": false, + "reason": "Conservative reject: the packet only shows soft similarity and no shared targets/files or body evidence proving the same failure path. The titles suggest related LoRA/int8 loading problems, but that is not enough to treat them as definitively the same bug." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The canonical issue choice is grounded in the packet, but the soft-edge verdict is too strong. The packet only shows title-level similarity and a `soft_similarity` edge with `deterministic_accept: false`; it does not establish that #10621 and #11752 are definitively the same bug. Please keep that pair conservative unless the body text clearly matches the same failure path." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 9, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2413-8", + "nodes": [ + "issue:10680", + "issue:11898", + "issue:11899", + "issue:11900", + "issue:2413", + "issue:2441", + "issue:5929", + "issue:6652" + ], + "soft_pairs": [ + "issue:2413|issue:2441", + "issue:11898|issue:2413", + "issue:11899|issue:2413", + "issue:11900|issue:2413", + "issue:2413|issue:6652", + "issue:2441|issue:6652", + "issue:10680|issue:2441", + "issue:11898|issue:2441", + "issue:11899|issue:2441", + "issue:2413|issue:5929", + "issue:2441|issue:5929", + "issue:11898|issue:5929", + "issue:11899|issue:5929", + "issue:11900|issue:5929", + "issue:10680|issue:6652", + "issue:11898|issue:6652", + "issue:11899|issue:6652", + "issue:10680|issue:2413", + "issue:10680|issue:11898", + "issue:10680|issue:11899", + "issue:11898|issue:11899", + "issue:11898|issue:11900", + "issue:11899|issue:11900", + "issue:11900|issue:2441", + "issue:11900|issue:6652" + ], + "prepared_review_unit_hash": "fe8c5a5cd038a6b26709c80d026634bae527464a6fb9ad27fc6460fca641b80d", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 25, + "serialized_chars": 12960, + "estimated_input_tokens": 3240, + "estimated_eval_tokens": 6736 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 25, + "serialized_chars": 12960, + "estimated_input_tokens": 3240, + "estimated_eval_tokens": 6736 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is dominated by duplicate reports about missing fp16 artifacts in Stable Diffusion model repos. Exact duplicates for stabilityai/stable-diffusion-2 should merge; cross-model/version reports (2 vs 2-1 vs v1-5, or base variants) are related but not the same issue.", + "confidence": 0.87, + "canonical_issue_reason": "Issue 2413 is the oldest exact-match report for the stable-diffusion-2 fp16-file absence and has multiple direct duplicates pointing to it.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 2413 is the best representative of the cluster\u2019s clearest duplicate set: repeated stable-diffusion-2 missing-fp16 reports with the broadest direct duplication signal.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2413", + "right": "issue:2441", + "accept": false, + "reason": "Different model targets: stable-diffusion-2 vs stable-diffusion-2-1. Same general symptom, but not the same underlying issue." + }, + { + "left": "issue:11898", + "right": "issue:2413", + "accept": true, + "reason": "Same model and same missing-fp16 problem; this is a direct duplicate of the stable-diffusion-2 report." + }, + { + "left": "issue:11899", + "right": "issue:2413", + "accept": true, + "reason": "Same title and same stable-diffusion-2 missing-fp16 complaint; duplicate." + }, + { + "left": "issue:11900", + "right": "issue:2413", + "accept": true, + "reason": "Same stable-diffusion-2 missing-fp16 issue; duplicate of the older report." + }, + { + "left": "issue:2413", + "right": "issue:6652", + "accept": false, + "reason": "Different model/version and different artifact wording; related category, but not the same bug report." + }, + { + "left": "issue:2441", + "right": "issue:6652", + "accept": true, + "reason": "Both are about stable-diffusion-2-1 missing fp16 artifacts, with the latter naming a specific fp16 safetensors file; same underlying issue." + }, + { + "left": "issue:10680", + "right": "issue:2441", + "accept": false, + "reason": "Base variant vs the general 2-1 model report; too specific/different to treat as the same duplicate." + }, + { + "left": "issue:11898", + "right": "issue:2441", + "accept": false, + "reason": "stable-diffusion-2 vs stable-diffusion-2-1 are different model targets." + }, + { + "left": "issue:11899", + "right": "issue:2441", + "accept": false, + "reason": "stable-diffusion-2 vs stable-diffusion-2-1 are different model targets." + }, + { + "left": "issue:2413", + "right": "issue:5929", + "accept": false, + "reason": "Different model families (stable-diffusion-2 vs runwayml/stable-diffusion-v1-5); same symptom only at a broad category level." + }, + { + "left": "issue:2441", + "right": "issue:5929", + "accept": false, + "reason": "Different model families and versions; not the same underlying issue." + }, + { + "left": "issue:11898", + "right": "issue:5929", + "accept": false, + "reason": "Different model family/version; only the symptom overlaps." + }, + { + "left": "issue:11899", + "right": "issue:5929", + "accept": false, + "reason": "Different model family/version; only the symptom overlaps." + }, + { + "left": "issue:11900", + "right": "issue:5929", + "accept": false, + "reason": "Different model family/version; only the symptom overlaps." + }, + { + "left": "issue:10680", + "right": "issue:6652", + "accept": false, + "reason": "Base-variant 2-1-base report does not cleanly match the general 2-1 safetensors report." + }, + { + "left": "issue:11898", + "right": "issue:6652", + "accept": false, + "reason": "Different model target (stable-diffusion-2 vs stable-diffusion-2-1) and different file-specific complaint." + }, + { + "left": "issue:11899", + "right": "issue:6652", + "accept": false, + "reason": "Different model target (stable-diffusion-2 vs stable-diffusion-2-1) and different file-specific complaint." + }, + { + "left": "issue:10680", + "right": "issue:2413", + "accept": false, + "reason": "stable-diffusion-2-1-base is a different artifact from stable-diffusion-2; related but not a duplicate." + }, + { + "left": "issue:10680", + "right": "issue:11898", + "accept": false, + "reason": "Different model targets: 2-1-base vs 2." + }, + { + "left": "issue:10680", + "right": "issue:11899", + "accept": false, + "reason": "Different model targets: 2-1-base vs 2." + }, + { + "left": "issue:11898", + "right": "issue:11899", + "accept": true, + "reason": "Exact same stable-diffusion-2 missing-fp16 report; clear duplicate." + }, + { + "left": "issue:11898", + "right": "issue:11900", + "accept": true, + "reason": "Exact same stable-diffusion-2 missing-fp16 report; clear duplicate." + }, + { + "left": "issue:11899", + "right": "issue:11900", + "accept": true, + "reason": "Exact same stable-diffusion-2 missing-fp16 report; clear duplicate." + }, + { + "left": "issue:11900", + "right": "issue:2441", + "accept": false, + "reason": "Different model versions: stable-diffusion-2 vs stable-diffusion-2-1." + }, + { + "left": "issue:11900", + "right": "issue:6652", + "accept": false, + "reason": "Different model target and file-specific complaint; not the same issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: 2413 is the oldest stable-diffusion-2 report, and 11898/11899/11900 are exact-title duplicates of that same model-specific complaint. The cross-model pairings are handled conservatively as related but not identical, and the accepted soft-edge matches are limited to clearly same-model/same-symptom cases." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 10, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1074-5", + "nodes": [ + "issue:1074", + "issue:1075", + "issue:3329", + "issue:4643", + "issue:5551" + ], + "soft_pairs": [ + "issue:1074|issue:1075", + "issue:1074|issue:3329", + "issue:3329|issue:4643", + "issue:3329|issue:5551" + ], + "prepared_review_unit_hash": "dba09d4f9224dab04dbe499252c8601bab37222045edf5a66f811d8f62bac271", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 4137, + "estimated_input_tokens": 1035, + "estimated_eval_tokens": 2326 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 4137, + "estimated_input_tokens": 1035, + "estimated_eval_tokens": 2326 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are all loosely related support questions about JAX/model usage, but they do not describe the same underlying bug or change. No duplicate cluster emerges here.", + "confidence": 0.91, + "canonical_issue_reason": "No clear canonical issue: the five issues ask different how-to questions (saving models, inference, inpaint model compatibility, ControlNet inference code, and model inversion).", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1075 is the broadest representative of the cluster\u2019s general JAX-inference theme, but it is not a true duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1074", + "right": "issue:1075", + "accept": false, + "reason": "Saving JAX models and using JAX inference are different tasks, not the same defect or request." + }, + { + "left": "issue:1074", + "right": "issue:3329", + "accept": false, + "reason": "Model saving and inpaint-model compatibility are unrelated support questions." + }, + { + "left": "issue:3329", + "right": "issue:4643", + "accept": false, + "reason": "Inpaint model loading vs. ControlNet inference implementation are different code-path questions." + }, + { + "left": "issue:3329", + "right": "issue:5551", + "accept": false, + "reason": "Loading a specific inpaint model is not the same as asking whether any pretrained model can be inverted." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 11, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-807-4", + "nodes": [ + "issue:1093", + "issue:2084", + "issue:807", + "issue:918" + ], + "soft_pairs": [ + "issue:807|issue:918", + "issue:2084|issue:807", + "issue:1093|issue:918" + ], + "prepared_review_unit_hash": "d9f5541b9c3fdad945c5bf06e0db7bf026f1baad254dda90b0f2a68dc70a17a5", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 3321, + "estimated_input_tokens": 831, + "estimated_eval_tokens": 1918 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 3321, + "estimated_input_tokens": 831, + "estimated_eval_tokens": 1918 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are thematically related to Dreambooth/textual inversion training failures, but they describe different underlying problems: low-VRAM limits, train-text-encoder type errors, and a float16/float32 dtype mismatch. I would not merge any of the soft edges.", + "confidence": 0.88, + "canonical_issue_reason": "Issue 807 is the broadest and most central Dreambooth training-failure report in the set, so it is the best umbrella representative even though the others are not true duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 807 best captures the cluster\u2019s main theme of Dreambooth training not working, but its symptom is still distinct from the other concrete error reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:807", + "right": "issue:918", + "accept": false, + "reason": "Low-VRAM training failure vs. a specific --train-text-encoder type error in train_dreambooth.py; different bug modes." + }, + { + "left": "issue:2084", + "right": "issue:807", + "accept": false, + "reason": "Float16/float32 UNet dtype mismatch is a different concrete failure from an 8GB training/memory issue." + }, + { + "left": "issue:1093", + "right": "issue:918", + "accept": false, + "reason": "Textual inversion with DeepSpeed fallback is a different code path and symptom than the Dreambooth train-text-encoder type error." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 12, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1094-3", + "nodes": [ + "issue:1094", + "issue:1288", + "issue:3227" + ], + "soft_pairs": [ + "issue:1094|issue:1288", + "issue:1288|issue:3227" + ], + "prepared_review_unit_hash": "5f510d1cdc875abf8a27cabaaffa83dfe646862d9a155e61e17a93222c2c058f", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2451, + "estimated_input_tokens": 613, + "estimated_eval_tokens": 1482 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2451, + "estimated_input_tokens": 613, + "estimated_eval_tokens": 1482 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are all about model/inference speedups, but they describe different optimization ideas rather than the same concrete change. Issue 3227 is the broadest umbrella and best representative of the cluster.", + "confidence": 0.73, + "canonical_issue_reason": "Issue 3227 is the most general speed-optimization umbrella (\u201cWinograd Convolution & other Speed Is All You Need optimizations\u201d), while 1094 and 1288 are narrower exploratory ideas around specific third-party acceleration paths.", + "canonical_pr_reason": null, + "best_issue_reason": "3227 best represents the shared theme of performance work because it can encompass multiple speedup approaches without being tied to a single vendor/tool experiment.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1094", + "right": "issue:1288", + "accept": false, + "reason": "Both are exploratory speed-improvement ideas, but one targets kernl and the other NVIDIA TransformerEngine; they are different proposed solutions, not the same underlying change." + }, + { + "left": "issue:1288", + "right": "issue:3227", + "accept": false, + "reason": "They share a performance theme, but 1288 is specifically about TransformerEngine while 3227 is a broader set of optimization ideas, so they are not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 13, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11008-5", + "nodes": [ + "issue:11008", + "issue:11304", + "issue:11326", + "issue:12138", + "issue:12465" + ], + "soft_pairs": [ + "issue:11008|issue:11326", + "issue:11008|issue:12138", + "issue:11304|issue:12465", + "issue:11304|issue:11326", + "issue:11326|issue:12465" + ], + "prepared_review_unit_hash": "29d206ab40aec8161fe5e1f0d11d7d67fe6e411f0fcc60b2ceb4ed55f74fabbd", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 5, + "serialized_chars": 4442, + "estimated_input_tokens": 1111, + "estimated_eval_tokens": 2478 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 5, + "serialized_chars": 4442, + "estimated_input_tokens": 1111, + "estimated_eval_tokens": 2478 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Loose topical cluster around Wan video and Remote VAE, but the evidence only supports relatedness, not duplicate/mergeable identity. None of the issue pairs have enough concrete overlap to accept as the same underlying bug/change.", + "confidence": 0.69, + "canonical_issue_reason": "issue:11326 is the most central and specific issue title in the cluster because it directly names both Wan Video and RemoteVAE, matching the common theme more closely than the broader support/feedback/requirements reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:11326 is still the best single issue to represent the cluster, but only as the focal topic; the surrounding issues are related context, not confirmed duplicates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11008", + "right": "issue:11326", + "accept": false, + "reason": "Related theme only: both mention Wan video / RemoteVAE, but the titles alone do not show the same concrete bug or change. Without body or diff evidence, this is too weak to accept as a duplicate." + }, + { + "left": "issue:11008", + "right": "issue:12138", + "accept": false, + "reason": "The titles point to different concerns: model support versus a requirements/usage question. Related topic, not the same underlying issue." + }, + { + "left": "issue:11304", + "right": "issue:12465", + "accept": false, + "reason": "Both are about Remote VAE, but one is a specific unpacked-latents request and the other is generic feedback. Not enough evidence that they are the same actionable problem." + }, + { + "left": "issue:11304", + "right": "issue:11326", + "accept": false, + "reason": "Same subsystem, but one is a narrow unpacked-latents request and the other is a Wan Video RemoteVAE topic. Too different to treat as duplicates." + }, + { + "left": "issue:11326", + "right": "issue:12465", + "accept": false, + "reason": "Generic Remote VAE feedback is related to Wan Video RemoteVAE, but the available evidence does not show the same concrete issue or a mergeable PR target." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Most of the summary is grounded, but one soft-edge verdict is too strong for the available evidence: issue:11008 vs issue:11326 is only clearly a related Wan video / RemoteVAE theme from the titles, not a conservative duplicate/mergeable match. With no body or diff evidence, that pair should stay unresolved or be marked related, not accepted. The other pairwise rejections are appropriately conservative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 14, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9085-5", + "nodes": [ + "issue:11019", + "issue:12899", + "issue:9085", + "issue:9151", + "issue:9910" + ], + "soft_pairs": [ + "issue:11019|issue:9085", + "issue:12899|issue:9085", + "issue:9085|issue:9151", + "issue:9151|issue:9910", + "issue:11019|issue:12899" + ], + "prepared_review_unit_hash": "fe69388c0faedb3ef620ec6cf380b662e1f14264998b449a4fa298f84f344ad9", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 5, + "serialized_chars": 4517, + "estimated_input_tokens": 1130, + "estimated_eval_tokens": 2516 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 5, + "serialized_chars": 4517, + "estimated_input_tokens": 1130, + "estimated_eval_tokens": 2516 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are thematically related ControlNet feature requests, but they target different pipelines/models (Flux, Sana, Z-Image, StableDiffusion latent upscale, img2img). They are not the same underlying bug/change, so they should not be deduplicated.", + "confidence": 0.81, + "canonical_issue_reason": "issue:9085 is the best representative because it is the clearest, most central ControlNet request in this cluster and has the most discussion, but it is still model-specific (Flux) rather than a true umbrella duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9085 is the strongest cluster anchor: it explicitly asks for a ControlNet training script for Flux, which is the most concrete and broadly aligned request among the issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11019", + "right": "issue:9085", + "accept": false, + "reason": "Both mention ControlNet training, but one is for Sana and the other for Flux; these are different model-specific requests, not the same change." + }, + { + "left": "issue:12899", + "right": "issue:9085", + "accept": false, + "reason": "Z-Image ControlNet training and Flux ControlNet training are distinct model targets and would require different implementation work." + }, + { + "left": "issue:9085", + "right": "issue:9151", + "accept": false, + "reason": "One requests a training script, the other requests an img2img pipeline; related subsystem, but not the same underlying feature." + }, + { + "left": "issue:9151", + "right": "issue:9910", + "accept": false, + "reason": "ControlNet img2img support and prompt_embeds support in a latent upscale pipeline are unrelated feature requests." + }, + { + "left": "issue:11019", + "right": "issue:12899", + "accept": false, + "reason": "Both are ControlNet-related, but they target different models (Sana vs Z-Image) and different asks, so they are not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 15, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11121-2", + "nodes": [ + "issue:11121", + "issue:11168" + ], + "soft_pairs": [], + "prepared_review_unit_hash": "d58c53e5539858791e9950572647387145ab9e55f79437f26f106245c467a960", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 0, + "serialized_chars": 1203, + "estimated_input_tokens": 301, + "estimated_eval_tokens": 858 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 0, + "serialized_chars": 1203, + "estimated_input_tokens": 301, + "estimated_eval_tokens": 858 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues are about adding SageAttention support for the diffuser library, so they are the same feature request. The newer, more specific open issue is the better canonical record.", + "confidence": 0.93, + "canonical_issue_reason": "Issue #11168 is more specific (\u201cSage Attention for diffuser library\u201d), has the richer discussion, and is still open; #11121 is a shorter generic duplicate request.", + "canonical_pr_reason": null, + "best_issue_reason": "#11168 is the best issue to keep as the cluster representative because it is clearer, more detailed, and better reflects the exact requested change.", + "best_pr_reason": null, + "soft_edge_verdicts": [] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 16, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7925-2", + "nodes": [ + "issue:11181", + "issue:7925" + ], + "soft_pairs": [ + "issue:11181|issue:7925" + ], + "prepared_review_unit_hash": "9bb8ab8ec7ab374baef63f9154bd10335b00ce5abd50832577bedff610d1e54d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1585, + "estimated_input_tokens": 397, + "estimated_eval_tokens": 1050 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1585, + "estimated_input_tokens": 397, + "estimated_eval_tokens": 1050 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are both controlnet-training issue reports, but they describe different problems: one is Flux ControlNet training producing bad results on a toy dataset, while the other concerns ControlNet training with LoRA. They overlap in subsystem, not in a clearly shared bug.", + "confidence": 0.83, + "canonical_issue_reason": "Issue 11181 is the more specific and concrete report, naming the exact script (`train_controlnet_flux.py`) and symptom (bad results on fill50k), so it is the better representative issue.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 11181 is the best issue in this cluster because it is the most actionable and narrowly scoped description of the training failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11181", + "right": "issue:7925", + "accept": false, + "reason": "Both are about ControlNet training problems, but one is Flux-specific bad output and the other is a LoRA-related training issue; they do not look like the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 17, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11267-2", + "nodes": [ + "issue:11267", + "issue:12542" + ], + "soft_pairs": [ + "issue:11267|issue:12542" + ], + "prepared_review_unit_hash": "150b62bfadf0e417bfe897b619de1ec94b4b30d847ebf4dee351a384434b6494", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1504, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1504, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are unrelated: one requests adding `onnxruntime-qnn` to a list, the other requests custom `output_params` for `export_to_video`. They do not describe the same bug or change.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11267", + "right": "issue:12542", + "accept": false, + "reason": "Different feature requests in different code areas; no shared underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 18, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2419-12", + "nodes": [ + "issue:11295", + "issue:11460", + "issue:12125", + "issue:12253", + "issue:13061", + "issue:2419", + "issue:2601", + "issue:2611", + "issue:5492", + "issue:5863", + "issue:5918", + "issue:7198" + ], + "soft_pairs": [ + "issue:11295|issue:2419", + "issue:11460|issue:2419", + "issue:12125|issue:2419", + "issue:2419|issue:2601", + "issue:2419|issue:5492", + "issue:11295|issue:2601", + "issue:11460|issue:2601", + "issue:12125|issue:2601", + "issue:2601|issue:5492", + "issue:2611|issue:5492", + "issue:2611|issue:5918", + "issue:12253|issue:2611", + "issue:2419|issue:2611", + "issue:2601|issue:2611", + "issue:12253|issue:5492", + "issue:5492|issue:5918", + "issue:2419|issue:5863", + "issue:11295|issue:5863", + "issue:11460|issue:5863", + "issue:12125|issue:5863", + "issue:12253|issue:5863", + "issue:12253|issue:5918", + "issue:2419|issue:5918", + "issue:2601|issue:5918", + "issue:2419|issue:7198", + "issue:11295|issue:7198", + "issue:11460|issue:7198", + "issue:12125|issue:7198", + "issue:2601|issue:7198", + "issue:11295|issue:11460", + "issue:11295|issue:12125", + "issue:11295|issue:5492", + "issue:11460|issue:12125", + "issue:11460|issue:12253", + "issue:12125|issue:5492", + "issue:12253|issue:13061", + "issue:12253|issue:2419", + "issue:13061|issue:5863", + "issue:11295|issue:13061", + "issue:11460|issue:13061", + "issue:12125|issue:13061" + ], + "prepared_review_unit_hash": "0bcf1952944095f96f2565d4e142eaa92cb106c18efa193c21fec18fd4dd8464", + "component_budget": { + "node_count": 12, + "item_count": 12, + "soft_pair_count": 41, + "serialized_chars": 20972, + "estimated_input_tokens": 5243, + "estimated_eval_tokens": 10742 + }, + "budget": { + "node_count": 12, + "item_count": 12, + "soft_pair_count": 41, + "serialized_chars": 20972, + "estimated_input_tokens": 5243, + "estimated_eval_tokens": 10742 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly several duplicate issue families about missing fp16 model files. The largest and clearest family is CompVis/stable-diffusion-v1-4; a second family is stabilityai/stable-diffusion-2-base; the remaining issues are different repos or different missing-file problems.", + "confidence": 0.88, + "canonical_issue_reason": "issue:2419 is the earliest, cleanest, exact-title report for the main stable-diffusion-v1-4 fp16 missing-files bug, and it anchors the largest duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:2419 is the best global representative because it is the canonical early duplicate for the dominant issue family and has the most straightforward title.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11295", + "right": "issue:2419", + "accept": true, + "reason": "Same repo/model family and same missing-fp16-files problem." + }, + { + "left": "issue:11460", + "right": "issue:2419", + "accept": true, + "reason": "Same stable-diffusion-v1-4 fp16 missing-files issue." + }, + { + "left": "issue:12125", + "right": "issue:2419", + "accept": true, + "reason": "Same underlying stable-diffusion-v1-4 fp16 missing-files bug." + }, + { + "left": "issue:2419", + "right": "issue:2601", + "accept": false, + "reason": "Different model repo; same symptom but not the same concrete bug." + }, + { + "left": "issue:2419", + "right": "issue:5492", + "accept": false, + "reason": "Different Stable Diffusion checkpoint/repo, so this is a separate missing-files report." + }, + { + "left": "issue:11295", + "right": "issue:2601", + "accept": false, + "reason": "Different model repo and checkpoint." + }, + { + "left": "issue:11460", + "right": "issue:2601", + "accept": false, + "reason": "Different repo/model; only the fp16 symptom overlaps." + }, + { + "left": "issue:12125", + "right": "issue:2601", + "accept": false, + "reason": "Different repo/model, not one mergeable bug." + }, + { + "left": "issue:2601", + "right": "issue:5492", + "accept": false, + "reason": "Different checkpoints; not the same artifact missing file set." + }, + { + "left": "issue:2611", + "right": "issue:5492", + "accept": false, + "reason": "One is a missing torch_model.bin in depth model; the other is fp16 files in base model." + }, + { + "left": "issue:2611", + "right": "issue:5918", + "accept": false, + "reason": "Different repo and different missing filename/type." + }, + { + "left": "issue:12253", + "right": "issue:2611", + "accept": false, + "reason": "Different model family and different missing file." + }, + { + "left": "issue:2419", + "right": "issue:2611", + "accept": false, + "reason": "Different concrete missing-file problem." + }, + { + "left": "issue:2601", + "right": "issue:2611", + "accept": false, + "reason": "Different repository and different missing artifact." + }, + { + "left": "issue:12253", + "right": "issue:5492", + "accept": true, + "reason": "Both are the same stabilityai/stable-diffusion-2-base fp16 missing-files issue." + }, + { + "left": "issue:5492", + "right": "issue:5918", + "accept": false, + "reason": "Different model versions and different missing artifact specifics." + }, + { + "left": "issue:2419", + "right": "issue:5863", + "accept": true, + "reason": "Same stable-diffusion-v1-4 fp16 filenames problem, just phrased more specifically." + }, + { + "left": "issue:11295", + "right": "issue:5863", + "accept": true, + "reason": "Same stable-diffusion-v1-4 missing-fp16-files bug." + }, + { + "left": "issue:11460", + "right": "issue:5863", + "accept": true, + "reason": "Same underlying v1-4 fp16 filename issue." + }, + { + "left": "issue:12125", + "right": "issue:5863", + "accept": true, + "reason": "Same repo/model and same missing fp16 files." + }, + { + "left": "issue:12253", + "right": "issue:5863", + "accept": false, + "reason": "Different checkpoint family; not the same bug." + }, + { + "left": "issue:12253", + "right": "issue:5918", + "accept": false, + "reason": "Different model family and different specific missing file." + }, + { + "left": "issue:2419", + "right": "issue:5918", + "accept": false, + "reason": "Different checkpoint/version and a different filename-specific issue." + }, + { + "left": "issue:2601", + "right": "issue:5918", + "accept": false, + "reason": "Different repo/model family." + }, + { + "left": "issue:2419", + "right": "issue:7198", + "accept": false, + "reason": "Different repo/model; only the fp16 symptom is shared." + }, + { + "left": "issue:11295", + "right": "issue:7198", + "accept": false, + "reason": "Different checkpoint family." + }, + { + "left": "issue:11460", + "right": "issue:7198", + "accept": false, + "reason": "Different repo/model, not one concrete bug." + }, + { + "left": "issue:12125", + "right": "issue:7198", + "accept": false, + "reason": "Different checkpoint/repo." + }, + { + "left": "issue:2601", + "right": "issue:7198", + "accept": false, + "reason": "Different model repository and artifact set." + }, + { + "left": "issue:11295", + "right": "issue:11460", + "accept": true, + "reason": "Exact same stable-diffusion-v1-4 missing-fp16-files report." + }, + { + "left": "issue:11295", + "right": "issue:12125", + "accept": true, + "reason": "Exact same underlying v1-4 fp16 missing-files issue." + }, + { + "left": "issue:11295", + "right": "issue:5492", + "accept": false, + "reason": "Different model checkpoint family." + }, + { + "left": "issue:11460", + "right": "issue:12125", + "accept": true, + "reason": "Same repo/model and same fp16 missing-files bug." + }, + { + "left": "issue:11460", + "right": "issue:12253", + "accept": false, + "reason": "Different checkpoint family." + }, + { + "left": "issue:12125", + "right": "issue:5492", + "accept": false, + "reason": "Different model repo/version." + }, + { + "left": "issue:12253", + "right": "issue:13061", + "accept": false, + "reason": "Different repos and different failure mode; not the same concrete issue." + }, + { + "left": "issue:12253", + "right": "issue:2419", + "accept": false, + "reason": "Different checkpoint family; same broad symptom only." + }, + { + "left": "issue:13061", + "right": "issue:5863", + "accept": false, + "reason": "Different model and a different fp16-related error path." + }, + { + "left": "issue:11295", + "right": "issue:13061", + "accept": false, + "reason": "Different repository/model." + }, + { + "left": "issue:11460", + "right": "issue:13061", + "accept": false, + "reason": "Different repository/model and issue type." + }, + { + "left": "issue:12125", + "right": "issue:13061", + "accept": false, + "reason": "Different model repo and unrelated filename availability problem." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The grouping is grounded in the packet: 2419/11295/11460/12125/5863 are clearly the same CompVis/stable-diffusion-v1-4 fp16-files issue, while 5492/12253 is a separate stable-diffusion-2-base family, and the remaining issues are different repos or different missing-artifact bugs. The rejected cross-family pairs are conservative, not overstated." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 19, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11390-2", + "nodes": [ + "issue:11390", + "issue:11730" + ], + "soft_pairs": [ + "issue:11390|issue:11730" + ], + "prepared_review_unit_hash": "459921b05124092936a290a909b7a091f795d82c1b6b2d3e730863960249444c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1560, + "estimated_input_tokens": 390, + "estimated_eval_tokens": 1036 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1560, + "estimated_input_tokens": 390, + "estimated_eval_tokens": 1036 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two closed issues about training-script follow-ups, but they target different features: image interpolation vs LoRA alpha/metadata handling. They are related in theme, not duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "issue #11390 is the better canonical issue because it has substantially more discussion and inbound references, suggesting broader community attention.", + "canonical_pr_reason": null, + "best_issue_reason": "#11390 is the strongest representative issue for the cluster since it appears more established and widely referenced than #11730.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11390", + "right": "issue:11730", + "accept": false, + "reason": "Both are follow-up issues in training scripts, but they concern different concrete changes. One is about image interpolation; the other is about `--lora_alpha` and metadata handling. Same subsystem, not the same bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 20, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-478-5", + "nodes": [ + "issue:1145", + "issue:1238", + "issue:478", + "issue:522", + "issue:523" + ], + "soft_pairs": [ + "issue:478|issue:523", + "issue:522|issue:523", + "issue:1145|issue:522", + "issue:1145|issue:1238" + ], + "prepared_review_unit_hash": "d44b284630057cac98aa37c67947996b56c8a5719d2733f1cd20d8186ffcbd37", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 4098, + "estimated_input_tokens": 1025, + "estimated_eval_tokens": 2306 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 4098, + "estimated_input_tokens": 1025, + "estimated_eval_tokens": 2306 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The set is mostly unrelated Flax issues. The only clear duplicate pair is the PT<->Flax conversion-script request: issues 478 and 523. The other issues concern version breakage, loss instability, and dropout in UNet, which are separate topics.", + "confidence": 0.88, + "canonical_issue_reason": "Issue 478 is the better anchor for the duplicate conversion-script request because it states the feature more completely ('automatic conversion PT <=> Flax script'); issue 523 is essentially the same request phrased more briefly.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 478 is the strongest representative of the only true duplicate subgroup in this set. The remaining issues do not match closely enough to serve as duplicates of it or of each other.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:478", + "right": "issue:523", + "accept": true, + "reason": "Same underlying request: a PT<->Flax conversion script. The titles are near-paraphrases and describe the same feature change." + }, + { + "left": "issue:522", + "right": "issue:523", + "accept": false, + "reason": "Different topics: dropout for Flax UNet versus a PT/Flax conversion script. They do not describe the same bug or change." + }, + { + "left": "issue:1145", + "right": "issue:522", + "accept": false, + "reason": "One is about Flax 0.7.0 not working; the other is a dropout feature request. No shared concrete problem." + }, + { + "left": "issue:1145", + "right": "issue:1238", + "accept": false, + "reason": "Both mention Flax, but one reports version breakage and the other reports crazy loss; these are different failure modes and not the same issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is well-grounded in the packet and stays conservative: it accepts only the clearly matching PT<->Flax conversion-script pair (478/523) and rejects the other soft similarities as different topics. The summary does not overstate the evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 21, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-960-5", + "nodes": [ + "issue:11542", + "issue:12185", + "issue:2720", + "issue:4146", + "issue:960" + ], + "soft_pairs": [ + "issue:2720|issue:960", + "issue:11542|issue:2720", + "issue:2720|issue:4146", + "issue:12185|issue:2720" + ], + "prepared_review_unit_hash": "ab9f0b00b514768a147478efe7fcdc84990377dcc97ab18cb127a7d11af84f8b", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 4340, + "estimated_input_tokens": 1085, + "estimated_eval_tokens": 2426 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 4340, + "estimated_input_tokens": 1085, + "estimated_eval_tokens": 2426 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All items are loosely related \u201cwhat\u2019s the difference between X and Y?\u201d questions, but they do not describe the same underlying bug, change, or code-path issue. No soft duplicates should be merged here.", + "confidence": 0.91, + "canonical_issue_reason": "issue:2720 is the earliest and broadest of the LoRA/DreamBooth comparison questions, so if one issue had to stand in for this topic cluster it would be the best anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:2720 is the best representative issue for the cluster because it covers the core LoRA vs DreamBooth script comparison and predates the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2720", + "right": "issue:960", + "accept": false, + "reason": "Different topics entirely: LoRA training-script comparison vs scheduler comparison (PNDM vs PLMS)." + }, + { + "left": "issue:11542", + "right": "issue:2720", + "accept": false, + "reason": "Both are LoRA-related comparisons, but they ask about different scripts and a different distinction; not the same underlying issue." + }, + { + "left": "issue:2720", + "right": "issue:4146", + "accept": false, + "reason": "One compares training scripts; the other compares DreamBooth LoRA vs DreamBooth. Related conceptually, but not the same concrete question." + }, + { + "left": "issue:12185", + "right": "issue:2720", + "accept": false, + "reason": "Different comparison target: DreamBooth LoRA vs traditional LoRA, versus text-to-image LoRA vs DreamBooth LoRA." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 22, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11691-2", + "nodes": [ + "issue:11691", + "issue:11692" + ], + "soft_pairs": [ + "issue:11691|issue:11692" + ], + "prepared_review_unit_hash": "a992d89f67d0a1290f61f98a81d88b06316f57565b67e6890cf24359e1faa6c5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1426, + "estimated_input_tokens": 357, + "estimated_eval_tokens": 970 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1426, + "estimated_input_tokens": 357, + "estimated_eval_tokens": 970 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are effectively identical closed spam issues, so they cluster as the same issue class; issue #11691 is the best canonical representative by virtue of being older.", + "confidence": 0.98, + "canonical_issue_reason": "#11691 and #11692 have the same title, same body length, no code references, and no distinguishing metadata; pick the older one as canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "#11691 is as representative as #11692, with identical spam characteristics and earlier creation time, making it the cleanest anchor for the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11691", + "right": "issue:11692", + "accept": true, + "reason": "Identical spam-like issues with no substantive differences; they appear to be the same underlying artifact." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary and canonical choice are grounded in the packet: the two issues have identical titles, equal body length, no refs/filenames, and only differ by timestamp, so using the older one as the canonical representative is conservative. The soft-edge accept is also reasonable given the extremely strong similarity evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 23, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4047-2", + "nodes": [ + "issue:11766", + "issue:4047" + ], + "soft_pairs": [ + "issue:11766|issue:4047" + ], + "prepared_review_unit_hash": "e6d28abdcfd9bbec32380e37efcf75ae489dfd6e6b24282937b8bb470d228d91", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1497, + "estimated_input_tokens": 375, + "estimated_eval_tokens": 1006 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1497, + "estimated_input_tokens": 375, + "estimated_eval_tokens": 1006 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are about related LoRA/adapters functionality, but they do not describe the same underlying bug: one is a concrete bug with adapters under compiled models, the other is a usage question about LoRA scale during loading.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 11766 is the better canonical issue because it is a specific, actionable bug report with a clear failure mode; issue 4047 is more of a general question about LoRA scaling rather than the same defect.", + "canonical_pr_reason": null, + "best_issue_reason": "11766 is the strongest representative for the cluster because it is the more concrete and diagnostic report; 4047 is broader and not clearly the same problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11766", + "right": "issue:4047", + "accept": false, + "reason": "They touch the same LoRA/adapters area, but 11766 concerns setting adapters on a compiled model, while 4047 asks how to set LoRA scale when loading a model. Different behavior, different failure mode, not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 24, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11798-2", + "nodes": [ + "issue:11798", + "issue:12748" + ], + "soft_pairs": [ + "issue:11798|issue:12748" + ], + "prepared_review_unit_hash": "90b138774427459f7432dba78a630bd128d0bd6d52885a9893f7d0e52df28bc2", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1529, + "estimated_input_tokens": 383, + "estimated_eval_tokens": 1022 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1529, + "estimated_input_tokens": 383, + "estimated_eval_tokens": 1022 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are both feature requests for GGUF support, but they target different model families (Cosmos-Predict2 vs Z-Image-Turbo) and do not look like the same underlying bug or change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 12748 is the cleaner standalone request, but it is still model-specific and not a safe canonical for the pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11798", + "right": "issue:12748", + "accept": false, + "reason": "Both ask for GGUF/model-format support, but for different models and likely separate integration work; this is a shared feature theme, not the same duplicate issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 25, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9383-2", + "nodes": [ + "issue:11817", + "issue:9383" + ], + "soft_pairs": [ + "issue:11817|issue:9383" + ], + "prepared_review_unit_hash": "e90c22d0903e138ecc2fccbd8c919014b84570bf10f46c493f2596a2983bded6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1488, + "estimated_input_tokens": 372, + "estimated_eval_tokens": 1000 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1488, + "estimated_input_tokens": 372, + "estimated_eval_tokens": 1000 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two closed issues are both Flux-related feature requests, but they ask for different capabilities: one for flux_control support on Flux Kontext Dev, and one for bitsandbytes/LoRA model support. They do not appear to be the same underlying bug or change.", + "confidence": 0.89, + "canonical_issue_reason": "Neither issue is a clear duplicate of the other. If one must be chosen, #11817 is the more concrete and narrowly scoped request.", + "canonical_pr_reason": null, + "best_issue_reason": "#11817 is the better representative because it states a specific Flux/Control-related request, while #9383 is a separate LoRA/bitsandbytes support request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11817", + "right": "issue:9383", + "accept": false, + "reason": "Both mention Flux, but they target different features and code paths: flux_control support for Kontext Dev versus bitsandbytes support for LoRA models. This is not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 26, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9585-2", + "nodes": [ + "issue:11956", + "issue:9585" + ], + "soft_pairs": [ + "issue:11956|issue:9585" + ], + "prepared_review_unit_hash": "420f7c7a4a3b754eb0f84d91ee3298f1c3259374d4c74e750e2ad122a067b0a1", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1567, + "estimated_input_tokens": 392, + "estimated_eval_tokens": 1040 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1567, + "estimated_input_tokens": 392, + "estimated_eval_tokens": 1040 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are related in theme (high-guidance artifacts in diffusion models) but do not look like the same underlying request or bug: one is a specific FDG proposal, the other is a broader APG issue about oversaturation/artifacts. I would not merge them as duplicates.", + "confidence": 0.74, + "canonical_issue_reason": "issue:9585 is the broader, more established problem statement about high guidance-scale artifacts; issue:11956 appears to be a narrower proposal (FDG) that could belong under it but is not the same issue.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:9585 is the best single representative because its title and scope cover the general defect class, whereas issue:11956 is a specific technique proposal rather than the same open problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:11956", + "right": "issue:9585", + "accept": false, + "reason": "They share the same broad diffusion-guidance topic, but FDG vs APG indicates different methods/proposals and not a clearly identical underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 27, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12295-3", + "nodes": [ + "issue:12295", + "issue:12344", + "issue:12893" + ], + "soft_pairs": [ + "issue:12295|issue:12344", + "issue:12344|issue:12893" + ], + "prepared_review_unit_hash": "3506a32d05227a5283a7be2f7c5618bfc8732a0d3a55b055fb1ee80f0c033e03", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2389, + "estimated_input_tokens": 598, + "estimated_eval_tokens": 1452 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2389, + "estimated_input_tokens": 598, + "estimated_eval_tokens": 1452 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are not duplicates as a cluster. #12344 is the clearest, most specific Qwen Image bug report; #12295 is a broad umbrella issue; #12893 appears to be a related-but-different Z-Image text-length problem.", + "confidence": 0.84, + "canonical_issue_reason": "#12344 is the most concrete and actionable report: it names a specific Qwen Image field (`txt_seq_lens`) and the alleged problem. #12295 is too vague/broad, and #12893 is about Z-Image rather than Qwen Image.", + "canonical_pr_reason": null, + "best_issue_reason": "#12344 is the best representative issue because it targets a specific, likely root-cause bug in Qwen Image, whereas the others are either too generic or for a different model.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12295", + "right": "issue:12344", + "accept": false, + "reason": "The first issue is an umbrella 'Various' report, while the second is a specific `txt_seq_lens` complaint. They do not clearly describe the same underlying bug." + }, + { + "left": "issue:12344", + "right": "issue:12893", + "accept": false, + "reason": "Both mention text sequence length, but one is for Qwen Image and the other for Z-Image; that is too weak to treat as the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 28, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-282-2", + "nodes": [ + "issue:1232", + "issue:282" + ], + "soft_pairs": [ + "issue:1232|issue:282" + ], + "prepared_review_unit_hash": "89a32dbaa5c6153939c3271711126822226c4505385951690cd170a86e8a083c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1517, + "estimated_input_tokens": 380, + "estimated_eval_tokens": 1016 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1517, + "estimated_input_tokens": 380, + "estimated_eval_tokens": 1016 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are about different bugs in different parts of the product: one is Dreambooth learning-rate scheduling, the other is inpainting on CPU. They should not be clustered as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": "Issue #1232 is the more self-contained and specific bug report, so it is the better representative if a single issue must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #1232 is the clearest standalone bug description; issue #282 is unrelated and not suitable as the cluster representative.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1232", + "right": "issue:282", + "accept": false, + "reason": "Different underlying problems and code paths: Dreambooth cosine_with_restarts behavior versus CPU support for inpainting. They do not describe the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 29, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12375-2", + "nodes": [ + "issue:12375", + "issue:12990" + ], + "soft_pairs": [ + "issue:12375|issue:12990" + ], + "prepared_review_unit_hash": "76570bb067f21af3fe76e08b1f06cd8c310dabae3cec82bffc15d75b04de4f20", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1550, + "estimated_input_tokens": 388, + "estimated_eval_tokens": 1032 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1550, + "estimated_input_tokens": 388, + "estimated_eval_tokens": 1032 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues appear related and topic-overlapping, but the packet only shows title-level/soft similarity, so duplication is not established with confidence.", + "confidence": 0.48, + "canonical_issue_reason": "#12375 is the earlier, broader thread and is the safer canonical representative for this topic, but the available evidence is not strong enough to treat the pair as confirmed duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "#12990 is the more explicit and actionable phrasing of the topic, so it is a slightly better representative of the request, though still not enough to prove it is the same underlying issue as #12375.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12375", + "right": "issue:12990", + "accept": false, + "reason": "The evidence only supports soft topical similarity: both mention kernels in Diffusers, but there is no explicit linkage, shared targets, or enough detail to conclude they ask the same underlying request." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The packet only provides title-level similarity plus a soft-similarity signal; it does not support the stronger claim that both issues ask the same underlying request or are near-duplicates with confidence. A more conservative framing would be that they appear related and topic-overlapping, but duplication is not established from the available evidence." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 30, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12410-2", + "nodes": [ + "issue:12410", + "issue:13112" + ], + "soft_pairs": [ + "issue:12410|issue:13112" + ], + "prepared_review_unit_hash": "70c706a919c332b8dfb74480e3edc83bff72ba177758a7692181b12d8e2e53ad", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1558, + "estimated_input_tokens": 390, + "estimated_eval_tokens": 1036 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1558, + "estimated_input_tokens": 390, + "estimated_eval_tokens": 1036 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are unrelated: one is a WanVACEPipeline runtime limitation around low-noise-only operation, the other is a SkyReelsV2DiffusionForcingPipeline package-dependency check problem. The soft duplicate link should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": "issue #12410 is the better canonical representative only in the sense that it is a concrete functional bug report; #13112 is a separate dependency-validation issue in a different pipeline.", + "canonical_pr_reason": null, + "best_issue_reason": "#12410 is the strongest issue to anchor the cluster because it describes a specific pipeline failure mode, while #13112 concerns missing package checks and does not share the same underlying bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12410", + "right": "issue:13112", + "accept": false, + "reason": "Different pipelines and different failure classes: low-noise-stage behavior vs missing available-package checks. They do not describe the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 31, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12569-3", + "nodes": [ + "issue:12569", + "issue:12570", + "issue:12589" + ], + "soft_pairs": [ + "issue:12569|issue:12570", + "issue:12569|issue:12589", + "issue:12570|issue:12589" + ], + "prepared_review_unit_hash": "570a8e9ad155cb10531e945b201e2823a614338630439c5aff6e4cabe070911e", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2729, + "estimated_input_tokens": 683, + "estimated_eval_tokens": 1622 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2729, + "estimated_input_tokens": 683, + "estimated_eval_tokens": 1622 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three different feature requests\u2014TaylorSeer, unified attention, and TeaCache\u2014not duplicates of the same underlying change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "No true canonical issue here; if one must be chosen, 12589 is the only open item, but it is a distinct TeaCache request rather than a duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12569", + "right": "issue:12570", + "accept": false, + "reason": "Different requested features with different mechanisms; only a generic similarity in being feature asks." + }, + { + "left": "issue:12569", + "right": "issue:12589", + "accept": false, + "reason": "TaylorSeer and TeaCache are separate feature ideas, not the same bug or code change." + }, + { + "left": "issue:12570", + "right": "issue:12589", + "accept": false, + "reason": "Unified attention and TeaCache target different model components and do not look mergeable as one change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 32, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1257-3", + "nodes": [ + "issue:1257", + "issue:2950", + "issue:3147" + ], + "soft_pairs": [ + "issue:1257|issue:2950", + "issue:2950|issue:3147" + ], + "prepared_review_unit_hash": "12ef6f353bdc17f8d3edd126539c235396a863f5887dd7da7a5f390717aa2fe9", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2396, + "estimated_input_tokens": 599, + "estimated_eval_tokens": 1454 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2396, + "estimated_input_tokens": 599, + "estimated_eval_tokens": 1454 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Issues 1257 and 2950 both center on img2img/inpaint pipeline API unification and whether img2img should be a separate pipeline; 3147 is a separate performance concern about reducing model loads during pipeline switches.", + "confidence": 0.84, + "canonical_issue_reason": "#1257 is the most concrete and broadly representative of the shared img2img pipeline/API unification theme, with a clearer actionable scope than the more open-ended architectural question in #2950.", + "canonical_pr_reason": null, + "best_issue_reason": "#1257 is the best standalone representative because it names the specific API convention (`init_image`) and spans the same underlying pipeline-structure discussion seen in #2950.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1257", + "right": "issue:2950", + "accept": true, + "reason": "Both issues discuss the same underlying img2img/inpaint pipeline design and image-input API unification; they are close enough to be treated as the same change/theme." + }, + { + "left": "issue:2950", + "right": "issue:3147", + "accept": false, + "reason": "They only overlap at a broad pipeline level: #2950 is about img2img pipeline structure, while #3147 is specifically about reducing model loads on pipeline switches. Different concrete problems and fixes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and the soft-edge verdicts stay conservative. #1257 and #2950 are reasonably tied by the shared img2img/inpaint pipeline structure/API theme, and #3147 is correctly treated as a separate pipeline-switch performance issue rather than the same bugfix." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 33, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1266-2", + "nodes": [ + "issue:1266", + "issue:1295" + ], + "soft_pairs": [ + "issue:1266|issue:1295" + ], + "prepared_review_unit_hash": "c007648f93773a66d713ebd40a392339eec9f2a816ba0ccd0d93ce364ac352a4", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1507, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1507, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both closed issues concern prompt-running failures in the Stable Diffusion/JAX notebook space, but one is generic and the other is a narrower multi-prompt notebook report. They look related, yet the available evidence is too thin to treat them as the same exact bug with high confidence.", + "confidence": 0.55, + "canonical_issue_reason": "Issue 1295 is the more specific and actionable report, with the clearer reproduction context ('multiple prompts' in a named notebook).", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1295 is the best representative because its title better identifies the failing workflow and scope of the problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1266", + "right": "issue:1295", + "accept": false, + "reason": "The titles overlap around prompt execution, but 1266 is generic while 1295 is specifically about multiple prompts in one notebook; without shared targets, filenames, or body details, this is not enough to \u092a\u0941\u0937\u094d\u091f\u093f the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 34, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12864-2", + "nodes": [ + "issue:12864", + "issue:12935" + ], + "soft_pairs": [ + "issue:12864|issue:12935" + ], + "prepared_review_unit_hash": "a4406f42548b6b892d669c0a35d8e69b3cf75eca59572798234d46c489120918", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1590, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1590, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues request the same feature: a Diffusers skill for Claude and a Claude Code plugin. They differ only in wording and specificity, so they should be treated as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": "Issue #12935 is the clearest wording of the feature request and explicitly names Hugging Face Diffusers, Claude, and Claude Code.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #12935 is the best representative because it is more explicit while describing the same request as the other issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12864", + "right": "issue:12935", + "accept": true, + "reason": "Same underlying feature request with only minor wording differences; both ask for a Diffusers skill for Claude and a Claude Code plugin." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both issues ask for a Diffusers/Hugging Face Diffusers skill for Claude plus a Claude Code plugin. The soft-edge accept is conservative because the titles are near-paraphrases with only minor wording differences, and the canonical issue choice is supported by the clearer title." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 35, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12925-2", + "nodes": [ + "issue:12925", + "issue:12926" + ], + "soft_pairs": [ + "issue:12925|issue:12926" + ], + "prepared_review_unit_hash": "914390db2e9649f274d90bd5a854a054d643f94b5e2db279e406601e9e50e0a0", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1490, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1490, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two closed issues are about different LTX-2 topics: one requests distilled checkpoint support, the other concerns the condition pipeline. They share only broad project context, not the same underlying problem.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue 12925 is the more concrete, self-contained feature request, but it is not a duplicate of 12926.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:12925", + "right": "issue:12926", + "accept": false, + "reason": "Different scopes: distilled checkpoint support versus condition pipeline. No shared target, filename, or concrete bug/change indicating the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 36, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1296-2", + "nodes": [ + "issue:1296", + "issue:3134" + ], + "soft_pairs": [ + "issue:1296|issue:3134" + ], + "prepared_review_unit_hash": "2be9816aba3c0610185f7d3c77b830dd868b266fa81e3f6c3c2db1307a05f34b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1435, + "estimated_input_tokens": 359, + "estimated_eval_tokens": 974 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1435, + "estimated_input_tokens": 359, + "estimated_eval_tokens": 974 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are related by topic/name, but they do not look like the same underlying bug or request. \"Paella v3\" reads like a distinct follow-up or revision rather than a duplicate of the original \"Paella\" issue.", + "confidence": 0.78, + "canonical_issue_reason": "issue #3134 is the better representative: its title is more specific, it is newer, and it has inbound references, suggesting it is the more central tracking item.", + "canonical_pr_reason": null, + "best_issue_reason": "#3134 is the strongest issue to keep as the cluster anchor because it is the more explicit and better-connected Paella topic.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1296", + "right": "issue:3134", + "accept": false, + "reason": "Shared topic only; the titles suggest a related follow-up/versioning relationship, not the same concrete issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 37, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-871-9", + "nodes": [ + "issue:1333", + "issue:871", + "issue:872", + "issue:873", + "issue:875", + "issue:876", + "issue:877", + "issue:878", + "issue:895" + ], + "soft_pairs": [ + "issue:871|issue:895", + "issue:871|issue:872", + "issue:871|issue:875", + "issue:871|issue:876", + "issue:871|issue:877", + "issue:872|issue:875", + "issue:872|issue:876", + "issue:872|issue:877", + "issue:872|issue:878", + "issue:871|issue:873", + "issue:872|issue:873", + "issue:873|issue:875", + "issue:873|issue:876", + "issue:873|issue:877", + "issue:875|issue:876", + "issue:875|issue:877", + "issue:875|issue:878", + "issue:876|issue:878", + "issue:876|issue:895", + "issue:1333|issue:876", + "issue:876|issue:877", + "issue:877|issue:878", + "issue:872|issue:895", + "issue:873|issue:895", + "issue:875|issue:895", + "issue:1333|issue:878", + "issue:1333|issue:895", + "issue:1333|issue:871", + "issue:1333|issue:872" + ], + "prepared_review_unit_hash": "e6db404f70a6bdddac415d720a6a1dced063bb2104252d82764fe207ec67160b", + "component_budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 29, + "serialized_chars": 14818, + "estimated_input_tokens": 3705, + "estimated_eval_tokens": 7666 + }, + "budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 29, + "serialized_chars": 14818, + "estimated_input_tokens": 3705, + "estimated_eval_tokens": 7666 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are distinct Community Pipeline feature requests, not duplicates of the same bug or change. The overlap is mostly the shared project/theme wording, so all soft edges should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": "No single issue is a true canonical duplicate target here; the cluster is heterogeneous and each item describes a different pipeline or feature request.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 871 is the best representative if one must be chosen: it has the strongest external reference activity and is a broad, high-visibility Community Pipeline request, though it is not a duplicate canonical.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:871", + "right": "issue:895", + "accept": false, + "reason": "Speech-to-image and Imagic are different tasks: audio-conditioned generation vs text-based real-image editing." + }, + { + "left": "issue:871", + "right": "issue:872", + "accept": false, + "reason": "Speech-to-image and Wildcards are unrelated pipeline ideas with different functionality and code paths." + }, + { + "left": "issue:871", + "right": "issue:875", + "accept": false, + "reason": "Speech-to-image and Seed resize address different behaviors; one is modality conversion, the other seed handling." + }, + { + "left": "issue:871", + "right": "issue:876", + "accept": false, + "reason": "Speech-to-image and Composable Diffusion are separate community pipeline concepts, not the same underlying change." + }, + { + "left": "issue:871", + "right": "issue:877", + "accept": false, + "reason": "Speech-to-image and Checkpoint merging solve different problems and would not be merged into one PR." + }, + { + "left": "issue:872", + "right": "issue:875", + "accept": false, + "reason": "Wildcards and Seed resize are unrelated features; they do not describe the same bug or implementation." + }, + { + "left": "issue:872", + "right": "issue:876", + "accept": false, + "reason": "Wildcards and Composable Diffusion are different pipeline features with different intended code changes." + }, + { + "left": "issue:872", + "right": "issue:877", + "accept": false, + "reason": "Wildcards and Checkpoint merging are distinct changes; shared community-pipeline context is too broad." + }, + { + "left": "issue:872", + "right": "issue:878", + "accept": false, + "reason": "Wildcards and Modified Cross-Attention are different features and not the same concrete fix/change." + }, + { + "left": "issue:871", + "right": "issue:873", + "accept": false, + "reason": "Speech-to-image and SD v1-v4 compare are separate requests with different goals and implementation scope." + }, + { + "left": "issue:872", + "right": "issue:873", + "accept": false, + "reason": "Wildcards and SD v1-v4 compare are unrelated features; no shared bug or code path." + }, + { + "left": "issue:873", + "right": "issue:875", + "accept": false, + "reason": "SD v1-v4 compare and Seed resize are distinct pipeline features, not a single change." + }, + { + "left": "issue:873", + "right": "issue:876", + "accept": false, + "reason": "SD v1-v4 compare and Composable Diffusion are different requests with different behavior and code paths." + }, + { + "left": "issue:873", + "right": "issue:877", + "accept": false, + "reason": "SD v1-v4 compare and Checkpoint merging are unrelated changes beyond the broad Stable Diffusion context." + }, + { + "left": "issue:875", + "right": "issue:876", + "accept": false, + "reason": "Seed resize and Composable Diffusion address different functionality and should not be merged as duplicates." + }, + { + "left": "issue:875", + "right": "issue:877", + "accept": false, + "reason": "Seed resize and Checkpoint merging are separate pipeline concerns with no shared concrete bug." + }, + { + "left": "issue:875", + "right": "issue:878", + "accept": false, + "reason": "Seed resize and Modified Cross-Attention are different features and code paths." + }, + { + "left": "issue:876", + "right": "issue:878", + "accept": false, + "reason": "Composable Diffusion and Modified Cross-Attention are both compositional diffusion ideas, but they target different mechanisms and are not the same change." + }, + { + "left": "issue:876", + "right": "issue:895", + "accept": false, + "reason": "Composable Diffusion and Imagic are different community pipeline proposals with different goals." + }, + { + "left": "issue:1333", + "right": "issue:876", + "accept": false, + "reason": "RePaint + Stable Diffusion and Composable Diffusion are distinct pipelines; the overlap is only the shared Stable Diffusion umbrella." + }, + { + "left": "issue:876", + "right": "issue:877", + "accept": false, + "reason": "Composable Diffusion and Checkpoint merging are different changes, not one concrete code-path fix." + }, + { + "left": "issue:877", + "right": "issue:878", + "accept": false, + "reason": "Checkpoint merging and Modified Cross-Attention are unrelated features and should not be deduplicated." + }, + { + "left": "issue:872", + "right": "issue:895", + "accept": false, + "reason": "Wildcards and Imagic are unrelated community pipeline ideas." + }, + { + "left": "issue:873", + "right": "issue:895", + "accept": false, + "reason": "SD v1-v4 compare and Imagic solve different problems and use different implementations." + }, + { + "left": "issue:875", + "right": "issue:895", + "accept": false, + "reason": "Seed resize and Imagic are separate features with no shared underlying bug." + }, + { + "left": "issue:1333", + "right": "issue:878", + "accept": false, + "reason": "RePaint + Stable Diffusion and Modified Cross-Attention are different pipeline concepts and code paths." + }, + { + "left": "issue:1333", + "right": "issue:895", + "accept": false, + "reason": "RePaint + Stable Diffusion and Imagic are different tasks: inpainting/editing pipeline vs text-based real-image editing." + }, + { + "left": "issue:1333", + "right": "issue:871", + "accept": false, + "reason": "RePaint + Stable Diffusion and Speech-to-image are unrelated Community Pipeline requests." + }, + { + "left": "issue:1333", + "right": "issue:872", + "accept": false, + "reason": "RePaint + Stable Diffusion and Wildcards are different feature requests, not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 38, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13362-2", + "nodes": [ + "issue:13362", + "issue:13394" + ], + "soft_pairs": [ + "issue:13362|issue:13394" + ], + "prepared_review_unit_hash": "eb84a297700bf5a481e33c4db7293bb788de9efa71ea5557e6a0be56ab89829f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1601, + "estimated_input_tokens": 401, + "estimated_eval_tokens": 1058 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1601, + "estimated_input_tokens": 401, + "estimated_eval_tokens": 1058 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are related at a high level (validation gaps in diffusion components) but they describe different bugs in different code paths: one is DDIMPipeline eta-range validation, the other is DDPMScheduler allowing num_inference_steps=0.", + "confidence": 0.93, + "canonical_issue_reason": "issue 13394 is the better representative for this cluster because it is open, concrete, and about a direct scheduler validation inconsistency; issue 13362 is a separate, already-closed DDIMPipeline constraint problem.", + "canonical_pr_reason": null, + "best_issue_reason": "issue 13394 is the strongest standalone issue to anchor on: it states a specific missing validation, names the affected scheduler, and describes the inconsistency clearly.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13362", + "right": "issue:13394", + "accept": false, + "reason": "Not the same underlying bug: 13362 is about validating eta in DDIMPipeline, while 13394 is about num_inference_steps=0 in DDPMScheduler. Similar theme, different parameter and code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 39, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1364-4", + "nodes": [ + "issue:1364", + "issue:2280", + "issue:6853", + "issue:6854" + ], + "soft_pairs": [ + "issue:1364|issue:6853", + "issue:1364|issue:6854", + "issue:2280|issue:6853", + "issue:2280|issue:6854", + "issue:6853|issue:6854" + ], + "prepared_review_unit_hash": "2868716321c2538e17faa7393ff3adb96aea627c244f7ebaa651a0a4c32f9dca", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 5, + "serialized_chars": 4018, + "estimated_input_tokens": 1005, + "estimated_eval_tokens": 2266 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 5, + "serialized_chars": 4018, + "estimated_input_tokens": 1005, + "estimated_eval_tokens": 2266 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Most items are unrelated community questions. The only clear duplicate pair is 6853 and 6854, which share the same inpainting-training request, with 6854 \u044f\u0432\u043d\u043e pointing back to #6853.", + "confidence": 0.96, + "canonical_issue_reason": "Issue 6853 is the substantive report/request; 6854 appears to be a duplicate or follow-up that literally references #6853 in the title and has no body.", + "canonical_pr_reason": null, + "best_issue_reason": "6853 is the best representative of the only true duplicate cluster because it is the original, non-empty issue describing the inpainting training request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1364", + "right": "issue:6853", + "accept": false, + "reason": "A generic question about pre-training Stable Diffusion is not the same underlying request as replacing training for the inpainting model." + }, + { + "left": "issue:1364", + "right": "issue:6854", + "accept": false, + "reason": "Different topics: general pre-training code question vs a specific inpainting training replacement request." + }, + { + "left": "issue:2280", + "right": "issue:6853", + "accept": false, + "reason": "Converting Stable Diffusion to inpainting is a different task from replacing training for the inpainting model." + }, + { + "left": "issue:2280", + "right": "issue:6854", + "accept": false, + "reason": "Related to inpainting, but a conversion script request is not the same underlying change as training replacement." + }, + { + "left": "issue:6853", + "right": "issue:6854", + "accept": true, + "reason": "Same concrete request: both ask to replace training for the Stable Diffusion inpainting model, and 6854 explicitly references 6853 as the target." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The verdict is grounded in the packet: 6853 and 6854 have highly overlapping titles, and 6854 explicitly references #6853 while having no body. The other pairs are conservatively rejected as only soft-similar, with no evidence of the same underlying request." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 40, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-138-2", + "nodes": [ + "issue:138", + "issue:3349" + ], + "soft_pairs": [ + "issue:138|issue:3349" + ], + "prepared_review_unit_hash": "3be9f2b2192a1234d386785f2959f35cc74f87ec45d84fc1aa2166c2a9aee621", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1568, + "estimated_input_tokens": 392, + "estimated_eval_tokens": 1040 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1568, + "estimated_input_tokens": 392, + "estimated_eval_tokens": 1040 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are related to missing transformers, but they describe different problem classes: a concrete runtime ImportError versus missing installation documentation. I would not treat them as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": "Issue #138 is the more concrete, bug-focused report: it describes the actual failure path and error message when the dependency is absent.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #138 is the best representative of the cluster because it captures the actionable runtime problem, whereas #3349 is primarily a documentation/install-instructions complaint.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:138", + "right": "issue:3349", + "accept": false, + "reason": "Reject: #138 is a runtime ImportError in a pipeline code path; #3349 is about missing dependency mention in install docs. Related, but not the same underlying bug/change and not something that would be merged as one fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 41, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1422-2", + "nodes": [ + "issue:1422", + "issue:5193" + ], + "soft_pairs": [ + "issue:1422|issue:5193" + ], + "prepared_review_unit_hash": "f08eff611b2683a6587594f1700103f28a9b8f94c54004ec91b69b6ecbc0b6a3", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1606, + "estimated_input_tokens": 402, + "estimated_eval_tokens": 1060 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1606, + "estimated_input_tokens": 402, + "estimated_eval_tokens": 1060 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: #1422 reports a broken/mosaic output when using CLIPGuidedStableDiffusion with stable-diffusion-2, while #5193 asks how to use a Chinese CLIP model. Different problem statements and likely different resolutions.", + "confidence": 0.96, + "canonical_issue_reason": "If a representative issue is needed, #1422 is the more concrete bug report; however it is not the same underlying problem as #5193.", + "canonical_pr_reason": null, + "best_issue_reason": "#1422 is the clearest standalone issue in this pair because it describes an explicit failure mode, but it should not be merged with #5193.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1422", + "right": "issue:5193", + "accept": false, + "reason": "Different topics: one is a bug about mosaic results on SD2, the other is a usage question about a Chinese CLIP model; not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 42, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-410-2", + "nodes": [ + "issue:1443", + "issue:410" + ], + "soft_pairs": [ + "issue:1443|issue:410" + ], + "prepared_review_unit_hash": "26225fcab8149f8fb9f8459dc4378f265ecdec56745aabaf6a446e664710c916", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1551, + "estimated_input_tokens": 388, + "estimated_eval_tokens": 1032 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1551, + "estimated_input_tokens": 388, + "estimated_eval_tokens": 1032 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one is a design request about moving prediction type onto the UNet, while the other reports a hard limitation on UNet internal block channel sizes. They touch different problems and would not be merged as the same underlying issue.", + "confidence": 0.95, + "canonical_issue_reason": "Issue #1443 is the clearer, more actionable representative of the cluster because it states a specific architectural change request and is more self-contained than #410.", + "canonical_pr_reason": null, + "best_issue_reason": "#1443 is the better canonical issue candidate if one must be chosen, since it has a narrower, more explicit scope and reads like a concrete product/design task.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1443", + "right": "issue:410", + "accept": false, + "reason": "The titles and implied problems are unrelated: prediction-type placement on UNet versus channel-size constraints in UNet blocks. They do not describe the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 43, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-146-2", + "nodes": [ + "issue:146", + "issue:2626" + ], + "soft_pairs": [ + "issue:146|issue:2626" + ], + "prepared_review_unit_hash": "f540ecc88f38052a27a13f9c4cd2d3fa2eddd2e927dce2f654bdee909ee65e9a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1507, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1507, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both artifacts are feature requests for training examples, but they target different pipelines: a superresolution training example versus a ControlNet training example. They are related in theme, not the same underlying request.", + "confidence": 0.71, + "canonical_issue_reason": "Issue 2626 is the more specific and clearly scoped request, making it a better canonical representative for the training-example feature request cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 2626 is the best single issue because it states the concrete desired example directly and is easier to map to a future implementation.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:146", + "right": "issue:2626", + "accept": false, + "reason": "Different target workflows and model types; they are both example requests, but not the same underlying feature/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 44, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1473-3", + "nodes": [ + "issue:1473", + "issue:4594", + "issue:5990" + ], + "soft_pairs": [ + "issue:1473|issue:4594", + "issue:4594|issue:5990" + ], + "prepared_review_unit_hash": "059e02839c5470a0e0f07d931b15386d47ddb2984165a22ecd55b4845dd78f4b", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2440, + "estimated_input_tokens": 610, + "estimated_eval_tokens": 1476 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2440, + "estimated_input_tokens": 610, + "estimated_eval_tokens": 1476 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three unrelated issue reports: one generic help request, one custom-pipeline gradient question, and one Stable Video Diffusion inpainting implementation problem. None appear to be duplicates of the same underlying bug.", + "confidence": 0.93, + "canonical_issue_reason": "Issue #5990 is the most concrete and actionable report, with a specific pipeline and implementation problem rather than a broad help request.", + "canonical_pr_reason": null, + "best_issue_reason": "#5990 is the best representative issue because it describes a specific technical problem in a named pipeline and is more likely to anchor related discussion than the generic or vague reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1473", + "right": "issue:4594", + "accept": false, + "reason": "#1473 is a generic request for code review/help, while #4594 is a specific custom pipeline gradient issue; they do not describe the same bug." + }, + { + "left": "issue:4594", + "right": "issue:5990", + "accept": false, + "reason": "Both involve pipeline-related troubleshooting, but one is about latents.requires_grad in a custom pipeline and the other about Stable Video Diffusion inpainting implementation issues; these are different problem reports." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 45, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1629-3", + "nodes": [ + "issue:1629", + "issue:1630", + "issue:1631" + ], + "soft_pairs": [ + "issue:1629|issue:1630", + "issue:1629|issue:1631", + "issue:1630|issue:1631" + ], + "prepared_review_unit_hash": "7791ecd730bbd336331f8404d7389134044169f0fe9c239c8d105b0b0ddeb902", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2811, + "estimated_input_tokens": 703, + "estimated_eval_tokens": 1662 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2811, + "estimated_input_tokens": 703, + "estimated_eval_tokens": 1662 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three issues concern flaky/non-deterministic test outputs, but they target different test suites and scenarios (AltDiffusion, common fast tests on MPS, and depth estimator in StableDiffusiondepth2img). They are related in theme, not the same underlying bug.", + "confidence": 0.87, + "canonical_issue_reason": "Issue #1630 is the broadest and most representative description of the cluster: it captures inconsistent outputs in common fast tests on MPS, which fits the shared reproducibility theme better than the two more specialized reports.", + "canonical_pr_reason": null, + "best_issue_reason": "#1630 is the best cluster representative because it is general, clearly states the symptom, and is not tied to one narrowly scoped pipeline test.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1629", + "right": "issue:1630", + "accept": false, + "reason": "Both are flaky-test reports, but one is AltDiffusion-specific and the other is a broader MPS inconsistency report; they do not describe the same concrete failure." + }, + { + "left": "issue:1629", + "right": "issue:1631", + "accept": false, + "reason": "These involve different pipelines and different test components, so they are not the same underlying bug." + }, + { + "left": "issue:1630", + "right": "issue:1631", + "accept": false, + "reason": "Shared nondeterminism theme only; the affected test paths and likely causes differ, so they should not be merged as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 46, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1684-2", + "nodes": [ + "issue:1684", + "issue:1696" + ], + "soft_pairs": [ + "issue:1684|issue:1696" + ], + "prepared_review_unit_hash": "6c4a6d6313db6a2d91539adb45b582f6f2654c8578ac26bdb4d8048f14e46cfd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1578, + "estimated_input_tokens": 395, + "estimated_eval_tokens": 1046 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1578, + "estimated_input_tokens": 395, + "estimated_eval_tokens": 1046 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items report floating point exception crashes, but they describe different triggers: generic inference vs. changing the scheduler. The overlap is only in the symptom, not clearly the same underlying bug.", + "confidence": 0.84, + "canonical_issue_reason": "Issue #1684 is the broader report of the crash during inference, so it is the better umbrella issue if these need to be tracked together.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #1684 is the best representative because its title describes the more general inference-path failure, making it the most reusable anchor for related reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1684", + "right": "issue:1696", + "accept": false, + "reason": "Reject: both mention a floating point exception, but one is tied to inference generally and the other specifically to changing the scheduler. That suggests different code paths and possibly different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 47, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1712-2", + "nodes": [ + "issue:1712", + "issue:5861" + ], + "soft_pairs": [ + "issue:1712|issue:5861" + ], + "prepared_review_unit_hash": "35a4cbed0489fc2822c4d4f9563805356c4599b4b12aa23fe75223cce31ade7c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1490, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1490, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are not duplicates: one is about GPU/CPU setup problems in certain modules, while the other is specifically about using ControlNet only with `UNet2DModel`.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue cleanly represents the other: #5861 is more specific and technically distinct, while #1712 is about a different class of setup/runtime problem. No single canonical issue for this cluster is appropriate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1712", + "right": "issue:5861", + "accept": false, + "reason": "Different underlying problems and code paths: GPU/CPU setup issues vs. ControlNet compatibility with `UNet2DModel`. Similarity is only superficial." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 48, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-329-3", + "nodes": [ + "issue:1748", + "issue:329", + "issue:4463" + ], + "soft_pairs": [ + "issue:1748|issue:329", + "issue:329|issue:4463" + ], + "prepared_review_unit_hash": "6825a888a1475d4309e5373fb73e11fd9478fc362adf17469c3e7c55083c6243", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2424, + "estimated_input_tokens": 606, + "estimated_eval_tokens": 1468 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2424, + "estimated_input_tokens": 606, + "estimated_eval_tokens": 1468 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These three issues are related only by broad Stable Diffusion support questions, but they ask for different capabilities: SD2 support, K-LMS inpainting/img2img support, and guide diffusion/unet_1d_condition support. They do not look like duplicates of the same underlying bug or feature request.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 329 is the most concrete and implementation-specific request, while the others are broader or different support questions; it is the best representative if one issue must anchor the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 329 is the strongest standalone issue because it names a specific missing capability and pipeline context, making it the clearest actionable request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1748", + "right": "issue:329", + "accept": false, + "reason": "Both mention Stable Diffusion support, but one is a general SD v2 compatibility question and the other is a specific K-LMS inpainting/img2img feature request; different underlying changes." + }, + { + "left": "issue:329", + "right": "issue:4463", + "accept": false, + "reason": "These request different support areas and code paths: K-LMS for inpainting/img2img versus guide diffusion and unet_1d_condition support." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 49, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1760-2", + "nodes": [ + "issue:1760", + "issue:4981" + ], + "soft_pairs": [ + "issue:1760|issue:4981" + ], + "prepared_review_unit_hash": "30c52bef685c5c85f279d2a91f306c2594715191d9a92e79eb4d145e88853c77", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1674, + "estimated_input_tokens": 419, + "estimated_eval_tokens": 1094 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1674, + "estimated_input_tokens": 419, + "estimated_eval_tokens": 1094 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues concern ONNX Runtime running with the OpenVINO execution provider, but they describe different levels of specificity: one is a generic OpenVINO-EP subgraph runtime exception, while the other is a StableDiffusionPipeline failure using OpenVINO EP. I don\u2019t see enough evidence that they are the same concrete bug.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 1760 is the more direct bug report: it names the concrete ONNXRuntime/OpenVINO-EP runtime exception and is less application-specific.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1760 is the best cluster representative because its title captures the underlying runtime failure more explicitly and broadly than the StableDiffusionPipeline-specific report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1760", + "right": "issue:4981", + "accept": false, + "reason": "Both mention OpenVINO EP with ONNX Runtime, but the evidence only shows a broad subsystem overlap. 4981 is tied to StableDiffusionPipeline and 1760 to a specific subgraph runtime exception, so I can\u2019t confirm the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 50, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1779-2", + "nodes": [ + "issue:1779", + "issue:2582" + ], + "soft_pairs": [ + "issue:1779|issue:2582" + ], + "prepared_review_unit_hash": "832746e07ef5b6b729236fccbdd0f4e18a6b5c96ba86f86383fa57f813839782", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1501, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1501, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two closed issues have the same title and no distinguishing scope signals, so they are best treated as duplicates; #1779 is the natural canonical issue because it was created first.", + "confidence": 0.95, + "canonical_issue_reason": "#1779 is the earlier-created issue, making it the best canonical representative for this duplicate report.", + "canonical_pr_reason": null, + "best_issue_reason": "#1779 is the strongest issue anchor because it has the same title as #2582 and was opened earlier.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1779", + "right": "issue:2582", + "accept": true, + "reason": "Identical titles and no evidence of distinct scope or targets; they appear to be duplicate reports of the same request." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster decision is mostly grounded, but the reasoning overstates the evidence: both issues have the same discussion_activity, so saying #1779 has 'more discussion context' is not supported by the packet. If you keep the cluster, justify the canonical pick only by earlier creation time (or leave the canonical rationale neutral)." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 51, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1797-2", + "nodes": [ + "issue:1797", + "issue:6182" + ], + "soft_pairs": [ + "issue:1797|issue:6182" + ], + "prepared_review_unit_hash": "f0f2e0b92142ba1e2ca305245e8e9804b73569a54dca82ea9b07cc0df13217fb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1484, + "estimated_input_tokens": 371, + "estimated_eval_tokens": 998 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1484, + "estimated_input_tokens": 371, + "estimated_eval_tokens": 998 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The pair is only weakly similar on metadata, with no body-content or path-level evidence to support a firm duplicate conclusion. If a canonical issue must be chosen, #6182 is the better metadata-based candidate, but the duplicate judgment should stay tentative.", + "confidence": 0.48, + "canonical_issue_reason": "#6182 is the stronger canonical candidate on metadata alone: it is newer, has more inbound references, and its title is broader/more descriptive.", + "canonical_pr_reason": null, + "best_issue_reason": "#6182 is the best issue to anchor future triage, but only tentatively, because the available evidence does not confirm the two reports describe the same bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1797", + "right": "issue:6182", + "accept": false, + "reason": "The only linkage is a low soft-similarity score (Jaccard 0.23) plus similar titles; there is no body-content evidence showing the same bug or code path, so this is too weak to accept as a duplicate." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The conclusion is too strong for the packet. We only have titles, dates, activity counts, and a soft-similarity edge (jaccard 0.23); there is no body-content evidence that both issues describe the same bug or the same code path. Saying they are 'likely duplicates of the same underlying problem' overstates the evidence. #6182 can be a better canonical candidate on metadata grounds, but the duplicate/merge judgment should remain tentative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 52, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1837-4", + "nodes": [ + "issue:1837", + "issue:2628", + "issue:3599", + "issue:4575" + ], + "soft_pairs": [ + "issue:1837|issue:2628", + "issue:2628|issue:3599", + "issue:2628|issue:4575" + ], + "prepared_review_unit_hash": "03cc288fecf9b78cc47a59355d366aa7f9a510430ff90fba5eb39b425ea35237", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 3242, + "estimated_input_tokens": 811, + "estimated_eval_tokens": 1878 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 3242, + "estimated_input_tokens": 811, + "estimated_eval_tokens": 1878 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are four unrelated feature/enhancement issues that only loosely share a training/scheduler theme. They do not look like duplicates or the same underlying change.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue: each ticket asks for a different feature in a different area (textual inversion logging, LR decay scheduling, DAAM, and scheduler self-recurrence).", + "canonical_pr_reason": null, + "best_issue_reason": "If one must be chosen, issue 2628 is the broadest/general enhancement request, but it is still not representative of the others as a duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1837", + "right": "issue:2628", + "accept": false, + "reason": "Different requested changes and code paths: image logging for textual inversion vs cosine-decay scheduling." + }, + { + "left": "issue:2628", + "right": "issue:3599", + "accept": false, + "reason": "Both are enhancements, but they target different functionality: scheduler decay vs DAAM attention visualization." + }, + { + "left": "issue:2628", + "right": "issue:4575", + "accept": false, + "reason": "Not the same change: cosine decay scheduling vs adding a self-recurrence trick to schedulers; too distinct to merge as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 53, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-853-2", + "nodes": [ + "issue:1877", + "issue:853" + ], + "soft_pairs": [ + "issue:1877|issue:853" + ], + "prepared_review_unit_hash": "de3a59027a9493f62b81810110bb22d636d5264689d4bb5086a14f610a5bb77a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1668, + "estimated_input_tokens": 417, + "estimated_eval_tokens": 1090 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1668, + "estimated_input_tokens": 417, + "estimated_eval_tokens": 1090 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are both about checkpoint/config conversion around LatentDiffusion, but they describe different problems: one is a model state_dict size-mismatch when loading a generated ckpt/yaml, the other asks whether ckpt conversion should change when training with 8GB VRAM settings. They are related topic-wise but not the same bug report.", + "confidence": 0.86, + "canonical_issue_reason": "Issue #1877 is the more concrete failure report: it includes the exact loading error and asks how to generate ckpt/yaml from a diffuser model. That makes it the better representative of the cluster\u2019s core problem surface.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #1877 is the best standalone issue because it has the clearest reproducible symptom and the most actionable description.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1877", + "right": "issue:853", + "accept": false, + "reason": "Both discuss ckpt conversion, but #1877 is a specific state_dict loading mismatch while #853 is a broader question about whether conversion should change for 8GB VRAM training settings. They do not look like the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 54, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-398-2", + "nodes": [ + "issue:1887", + "issue:398" + ], + "soft_pairs": [ + "issue:1887|issue:398" + ], + "prepared_review_unit_hash": "8fd643d9eb3a88bd57c22762f4532dbb5f7dbb56b66ea356ab72e05c39fa16ac", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1481, + "estimated_input_tokens": 371, + "estimated_eval_tokens": 998 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1481, + "estimated_input_tokens": 371, + "estimated_eval_tokens": 998 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items discuss sampler-related functionality, but they are not clearly the same underlying issue: one is a specific request about using DPM++ 2M Karras in Diffusers, while the other is a broad \"Sampler change\" ticket. I would not merge them as duplicates on the available evidence.", + "confidence": 0.62, + "canonical_issue_reason": "Issue 1887 is the more specific and descriptive sampler request, making it the better canonical issue for this cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 1887 is the strongest representative because it states the concrete sampler variant and intended library, whereas issue 398 is too generic to anchor the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1887", + "right": "issue:398", + "accept": false, + "reason": "They are related to samplers, but the evidence does not show the same concrete bug or change request; 1887 is a specific Diffusers question, while 398 is a broad sampler-change issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 55, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1975-2", + "nodes": [ + "issue:1975", + "issue:4522" + ], + "soft_pairs": [ + "issue:1975|issue:4522" + ], + "prepared_review_unit_hash": "362a578ceceb158065e95ba3e711f60c2634d4ea9850ea9892c72d627384ab9d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1503, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1503, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related to inpainting, but they describe different problems: textual inversion support vs. using custom inpainting models in the ControlNet inpaint pipeline. Not a duplicate pair.", + "confidence": 0.88, + "canonical_issue_reason": "Issue 4522 is the more concrete and specific report, with a clearer target pipeline and model-loading problem.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 4522 is the best representative because it names the exact pipeline and the practical failure mode more precisely than the broader feature request in 1975.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:1975", + "right": "issue:4522", + "accept": false, + "reason": "They overlap on inpainting, but one is about textual inversion support and the other about custom inpainting models in StableDiffusionControlNetInpaintPipeline; different underlying changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 56, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2135-2", + "nodes": [ + "issue:2135", + "issue:2136" + ], + "soft_pairs": [ + "issue:2135|issue:2136" + ], + "prepared_review_unit_hash": "0c6c4b5cacb38033f7b2f34904ae01141a08a019f83bc9c8c5f99b03af7f46dd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1554, + "estimated_input_tokens": 389, + "estimated_eval_tokens": 1034 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1554, + "estimated_input_tokens": 389, + "estimated_eval_tokens": 1034 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are related to Diffusers model support, but they describe different underlying requests: multilingual/AltDiffusion support vs. lifting the 77-token limit. They should not be treated as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical duplicate; the issues target different problems and should remain separate.", + "canonical_pr_reason": null, + "best_issue_reason": "issue #2136 is the more concrete technical limitation and is better suited as a tracking issue for implementation work.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2135", + "right": "issue:2136", + "accept": false, + "reason": "Different underlying requests: one is out-of-the-box support for multilingual models, the other is removing the 77-token prompt length limit. They may intersect in downstream usage, but they are not the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 57, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2160-3", + "nodes": [ + "issue:2160", + "issue:3185", + "issue:3482" + ], + "soft_pairs": [ + "issue:2160|issue:3185", + "issue:2160|issue:3482", + "issue:3185|issue:3482" + ], + "prepared_review_unit_hash": "5e5f8beb263f98bdc30fa900383feabdf18b8b2fb4ca3bfa5b72fd608e0ede74", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2782, + "estimated_input_tokens": 696, + "estimated_eval_tokens": 1648 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2782, + "estimated_input_tokens": 696, + "estimated_eval_tokens": 1648 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are thematically related Stable Diffusion training/fine-tuning questions, but they do not describe the same underlying bug or change, so no duplicates are accepted.", + "confidence": 0.89, + "canonical_issue_reason": "Issue 3482 is the broadest and most representative of the shared super-resolution/fine-tuning theme, but the three issues are still distinct enough that there is no true canonical duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 3482 is the best single representative for the cluster because it is the most general Stable Diffusion super-resolution fine-tuning question and closest to the overall topic.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2160", + "right": "issue:3185", + "accept": false, + "reason": "2160 is about a LoRA fine-tuning error tied to vae.config.scaling_factor; 3185 is about training a Stable Diffusion upscale-4x model. Same broad domain, but not the same concrete bug or code path." + }, + { + "left": "issue:2160", + "right": "issue:3482", + "accept": false, + "reason": "2160 reports a specific LoRA/VAE scaling-factor error, while 3482 asks about fine-tuning a Stable Diffusion super-resolution model. Different problem statements and no evidence of the same underlying issue." + }, + { + "left": "issue:3185", + "right": "issue:3482", + "accept": false, + "reason": "Both mention super-resolution/upscaling, but 3185 is about training a particular upscale-4x model and 3482 is a general fine-tuning question. Related topic, not a duplicate bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 58, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2170-3", + "nodes": [ + "issue:2170", + "issue:2283", + "issue:4691" + ], + "soft_pairs": [ + "issue:2170|issue:4691", + "issue:2170|issue:2283", + "issue:2283|issue:4691" + ], + "prepared_review_unit_hash": "ea8e63e13247872bfd4862b3e02e6a703620748e7fcd212884c8b1498f700ef5", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2996, + "estimated_input_tokens": 749, + "estimated_eval_tokens": 1754 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2996, + "estimated_input_tokens": 749, + "estimated_eval_tokens": 1754 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three items are ONNX-conversion-related, but they describe different failures: opset-version support for stable-diffusion-2-1, a CLIPVisionConfig attribute error for stable-diffusion-v1-5, and unsupported aten::scaled_dot_product_attention export at opset 14. They look like related symptom reports, not duplicates.", + "confidence": 0.89, + "canonical_issue_reason": "Issue 4691 is the clearest concrete export failure and the most actionable representative of the cluster; it has a specific unsupported operator and explicit opset context.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 4691 is the best global issue candidate because it states a precise ONNX export limitation with a concrete operator/opset mismatch, unlike the other two which are model-specific setup/config problems or broader opset questions.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2170", + "right": "issue:4691", + "accept": false, + "reason": "Different underlying problems: 2170 asks whether stable-diffusion-2-1 can export with opset 13, while 4691 reports aten::scaled_dot_product_attention failing at opset 14. Same area, not the same bug." + }, + { + "left": "issue:2170", + "right": "issue:2283", + "accept": false, + "reason": "2170 is about opset support during export; 2283 is a CLIPVisionConfig missing num_channels attribute. These are distinct failure modes." + }, + { + "left": "issue:2283", + "right": "issue:4691", + "accept": false, + "reason": "2283 is a configuration/attribute error in model conversion, while 4691 is an unsupported operator export error. They do not look like one underlying code-path issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 59, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2189-4", + "nodes": [ + "issue:2189", + "issue:2739", + "issue:3732", + "issue:4902" + ], + "soft_pairs": [ + "issue:2189|issue:4902", + "issue:2189|issue:2739", + "issue:2739|issue:3732" + ], + "prepared_review_unit_hash": "8960eaf198ba81582decc71895469b053d9649a46a516f452bb4e68a99907720", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 3258, + "estimated_input_tokens": 815, + "estimated_eval_tokens": 1886 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 3258, + "estimated_input_tokens": 815, + "estimated_eval_tokens": 1886 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Only one pair looks like the same underlying issue: both 2189 and 2739 ask whether multiple LoRA adapters/models can be loaded for inference. The other pairs are too broad or refer to safetensors loading in general, not the same LoRA-specific problem.", + "confidence": 0.88, + "canonical_issue_reason": "Issue 2189 is the clearest, most concrete statement of the LoRA multi-load inference problem; 2739 is essentially the same question in less specific wording.", + "canonical_pr_reason": null, + "best_issue_reason": "2189 is the best cluster representative because it names the exact behavior needed and is more actionable than the other issue titles.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2189", + "right": "issue:4902", + "accept": false, + "reason": "Both mention loading, but 4902 is about safetensors loading in general, not specifically loading multiple LoRA weights during inference." + }, + { + "left": "issue:2189", + "right": "issue:2739", + "accept": true, + "reason": "These are the same underlying question: loading multiple LoRA models/weights during inference." + }, + { + "left": "issue:2739", + "right": "issue:3732", + "accept": false, + "reason": "2739 is about multiple LoRA adapters, while 3732 is about using a local safetensors model for inference; different problem." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the only accepted soft-edge pair (2189 \u2194 2739) is conservative given the near-identical LoRA-loading wording in the titles. The rejected pairs are also reasonably cautious: 4902 is about safetensors loading generally, and 3732 is about local safetensors inference, so neither is clearly the same LoRA-multi-load issue. No overstatement is apparent." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 60, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2219-2", + "nodes": [ + "issue:2219", + "issue:5774" + ], + "soft_pairs": [ + "issue:2219|issue:5774" + ], + "prepared_review_unit_hash": "0f787ec6904183186a54ce3e341d95ab69c06db6f6b7f68de4d850fd5cc06a9f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1536, + "estimated_input_tokens": 384, + "estimated_eval_tokens": 1024 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1536, + "estimated_input_tokens": 384, + "estimated_eval_tokens": 1024 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two related but not duplicate how-to questions about Stable Diffusion fine-tuning; they ask about different training scenarios (super-resolution vs custom captioned image dataset).", + "confidence": 0.83, + "canonical_issue_reason": "issue:5774 is the broader and more general fine-tuning question, so it is the better representative of the cluster if one issue must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:5774 is the strongest single issue to keep because it is more general and better captures the common fine-tuning topic without being tied to the narrower super-resolution variant.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2219", + "right": "issue:5774", + "accept": false, + "reason": "Both are about fine-tuning Stable Diffusion, but they target different use cases and do not describe the same underlying bug or request; this is topical similarity, not duplication." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 61, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2255-2", + "nodes": [ + "issue:2255", + "issue:2256" + ], + "soft_pairs": [ + "issue:2255|issue:2256" + ], + "prepared_review_unit_hash": "528445be577b59a54c3ba2c776247f5a0d58ed579bb0f339d2ef63ed3702aabf", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1593, + "estimated_input_tokens": 399, + "estimated_eval_tokens": 1054 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1593, + "estimated_input_tokens": 399, + "estimated_eval_tokens": 1054 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Issues #2255 and #2256 are the same question with identical titles and very similar bodies, so they should be clustered as duplicates. #2255 is the better canonical issue because it has slightly more discussion activity.", + "confidence": 0.99, + "canonical_issue_reason": "#2255 and #2256 ask the exact same question; #2255 has more discussion activity, so it is the best canonical representative.", + "canonical_pr_reason": null, + "best_issue_reason": "#2255 is the stronger representative because it is the same issue as #2256 and has marginally more engagement.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2255", + "right": "issue:2256", + "accept": true, + "reason": "Identical title and highly similar content indicate the same underlying issue/question." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded in the packet: the titles are identical and the pair has strong soft similarity (jaccard 0.87). The canonical choice is also conservative, since #2255 has more discussion activity than #2256." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 62, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2270-2", + "nodes": [ + "issue:2270", + "issue:2353" + ], + "soft_pairs": [ + "issue:2270|issue:2353" + ], + "prepared_review_unit_hash": "095632f8a662b9f9223a5b0f5429020fb9132cfc2a2bb6f4b009846440c3b569", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1492, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1492, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: one asks about negative prompt support, while the other reports missing denoise support in a specific ONNX inpaint pipeline. They concern different features/code paths.", + "confidence": 0.91, + "canonical_issue_reason": "Issue #2353 is the more concrete, actionable bug report with a specific pipeline and missing capability.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #2353 best represents the cluster because it identifies a concrete implementation gap rather than a general usage question.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2270", + "right": "issue:2353", + "accept": false, + "reason": "Different functionality: negative prompt support is a separate feature request from denoise support in OnnxStableDiffusionInpaintPipeline. No clear same-bug/same-change evidence." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 63, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2411-4", + "nodes": [ + "issue:2411", + "issue:2777", + "issue:4253", + "issue:6178" + ], + "soft_pairs": [ + "issue:2411|issue:4253", + "issue:2777|issue:4253", + "issue:2777|issue:6178" + ], + "prepared_review_unit_hash": "26050084cb04c3eab3d0a732acb3bc0a012e4e1aa0de4c8cc23e3a8869d400ce", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 3344, + "estimated_input_tokens": 836, + "estimated_eval_tokens": 1928 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 3344, + "estimated_input_tokens": 836, + "estimated_eval_tokens": 1928 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "No duplicate cluster here: the issues only share superficial Stable Diffusion/Dreambooth wording, but each asks about a different problem or feature.", + "confidence": 0.96, + "canonical_issue_reason": "No canonical issue: the items are not the same underlying bug/request, so none should represent a duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "No single best representative for deduplication; the topics split across performance, Dreambooth support, image-guided generation, and DDPM training.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2411", + "right": "issue:4253", + "accept": false, + "reason": "2411 is a performance complaint about slow image generation; 4253 asks about using an input image with a Dreambooth-fine-tuned model. Different underlying issues." + }, + { + "left": "issue:2777", + "right": "issue:4253", + "accept": false, + "reason": "2777 requests Dreambooth support for Alt Diffusion; 4253 asks about image-guided generation after Dreambooth fine-tuning. Related vocabulary, but not the same change or bug." + }, + { + "left": "issue:2777", + "right": "issue:6178", + "accept": false, + "reason": "2777 is a Dreambooth/Alt Diffusion feature request, while 6178 asks how to train Stable Diffusion with DDPM. Different training methods and goals." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 64, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2709-2", + "nodes": [ + "issue:2709", + "issue:3731" + ], + "soft_pairs": [ + "issue:2709|issue:3731" + ], + "prepared_review_unit_hash": "a9eb727f96862f5a6777c242f464c98ce4bf4b21bbed12590e6cabefe7430849", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1523, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1523, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one is about saving PyTorch 2.0-compiled models, while the other is a missing-module import error for `models.unet_3d_condition`. They affect different code paths and have different failure modes.", + "confidence": 0.93, + "canonical_issue_reason": "Issue #2709 is the better canonical issue because it describes a broader, user-facing serialization failure that is more likely to represent the main cluster theme.", + "canonical_pr_reason": null, + "best_issue_reason": "#2709 is the strongest representative for this cluster since it is the more general bug report and not a narrow import-path symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2709", + "right": "issue:3731", + "accept": false, + "reason": "Different underlying problems: model saving/serialization for torch.compile in #2709 versus a missing module import in #3731. The titles and likely code paths do not indicate the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 65, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-271-2", + "nodes": [ + "issue:271", + "issue:277" + ], + "soft_pairs": [ + "issue:271|issue:277" + ], + "prepared_review_unit_hash": "8d5ece0a95f4e072595060956e56ec4d8a966c6ac10a8e773dcda47d2854f2e9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1471, + "estimated_input_tokens": 368, + "estimated_eval_tokens": 992 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1471, + "estimated_input_tokens": 368, + "estimated_eval_tokens": 992 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues are related thematically (samplers/schedulers), but they do not look like the same underlying request. #277 is a concrete feature request for additional samplers/schedulers, while #271 is a broader samplers/documentation topic.", + "confidence": 0.84, + "canonical_issue_reason": "#277 is the more specific and actionable issue: it asks for additional samplers/schedulers (e.g. Euler). #271 is broader and appears centered on documentation around samplers rather than the same concrete change.", + "canonical_pr_reason": null, + "best_issue_reason": "#277 is the best representative because it describes a clear, implementable feature request; #271 is too generic/broad to serve as the canonical issue for this cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:271", + "right": "issue:277", + "accept": false, + "reason": "They share the same general sampler theme, but one is documentation-focused and the other is a specific request for more samplers/schedulers. That is not enough to treat them as the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 66, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2750-3", + "nodes": [ + "issue:2750", + "issue:4679", + "issue:6429" + ], + "soft_pairs": [ + "issue:2750|issue:4679", + "issue:4679|issue:6429" + ], + "prepared_review_unit_hash": "4c6786db9cfee47668168d819217b3f05ddfcd391b0e904d91451013c91bf984", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2376, + "estimated_input_tokens": 594, + "estimated_eval_tokens": 1444 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2376, + "estimated_input_tokens": 594, + "estimated_eval_tokens": 1444 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly about ControlNet-related feature requests, but it splits into two topics: a specific face-landmarks ControlNet enhancement, and a LoRA-with-ControlNet request. The LoRA issues look like duplicates of each other; the face-landmarks request is separate.", + "confidence": 0.82, + "canonical_issue_reason": "Issue 4679 is the best canonical issue because it is the most general and still-open request for ControlNet LoRA support, making it the most useful representative for the duplicated LoRA thread.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 4679 best represents the cluster\u2019s duplicate pair because its title is broad, directly matches the LoRA-with-ControlNet theme, and it is still open.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2750", + "right": "issue:4679", + "accept": false, + "reason": "Both mention ControlNet, but 2750 is specifically about adding face-landmarks support, while 4679 is about ControlNet LoRA support. They are different requested features, not the same underlying change." + }, + { + "left": "issue:4679", + "right": "issue:6429", + "accept": true, + "reason": "These two issues describe the same underlying request: using or supporting LoRA with the ControlNet pipeline. The wording differs, but the feature target is the same." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The issue selection is grounded in the packet: 4679 is the broad, still-open ControlNet LoRA request, and 6429 is plausibly the same request phrased differently. The rejection of 2750 vs 4679 is appropriately conservative because it is a different ControlNet feature (face-landmarks). The summary stays within the evidence, noting only likely duplication for the LoRA issues and separating the face-landmarks request." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 67, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2820-2", + "nodes": [ + "issue:2820", + "issue:5140" + ], + "soft_pairs": [ + "issue:2820|issue:5140" + ], + "prepared_review_unit_hash": "7ccc1c9ef6f7b090dc3b1e8214aa006c7d3b2967a542129aa15aec117f072da0", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1511, + "estimated_input_tokens": 378, + "estimated_eval_tokens": 1012 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1511, + "estimated_input_tokens": 378, + "estimated_eval_tokens": 1012 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items discuss multi-ControlNet support, but they are not the same request: #2820 is a general img2img feature inquiry, while #5140 is specifically about BLIP-Diffusion lacking multi-ControlNet support.", + "confidence": 0.86, + "canonical_issue_reason": "#5140 is the more concrete and specific issue, making it the better representative of the topic.", + "canonical_pr_reason": null, + "best_issue_reason": "#5140 has a clearer problem statement and is more actionable than the generic question in #2820.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:2820", + "right": "issue:5140", + "accept": false, + "reason": "Related theme, but not the same underlying bug/change: one is a broad img2img feature request, the other is a BLIP-Diffusion-specific limitation." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 68, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3307-2", + "nodes": [ + "issue:3307", + "issue:3356" + ], + "soft_pairs": [ + "issue:3307|issue:3356" + ], + "prepared_review_unit_hash": "8181f82388232db21d43e045a6b608bdc97144c7564b22cf1a29fab291209263", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1522, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1522, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are related DeepFloyd IF fine-tuning questions, but they ask about different things: one is a general fine-tuning question, the other is specifically about `predicted_variance` for textual inversion tuning. They are not duplicates.", + "confidence": 0.91, + "canonical_issue_reason": "Issue 3356 is the broader, more representative question about fine-tuning DeepFloyd IF; issue 3307 is a narrower subtopic rather than the same issue.", + "canonical_pr_reason": null, + "best_issue_reason": "3356 is the best single issue to represent this cluster because it captures the general fine-tuning problem space more cleanly than the very specific `predicted_variance` question.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3307", + "right": "issue:3356", + "accept": false, + "reason": "Related topic, but not the same underlying problem: 3307 asks about `predicted_variance` in textual inversion tuning, while 3356 asks generally how to fine-tune DeepFloyd IF." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 69, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3317-3", + "nodes": [ + "issue:3317", + "issue:3502", + "issue:4823" + ], + "soft_pairs": [ + "issue:3317|issue:3502", + "issue:3502|issue:4823" + ], + "prepared_review_unit_hash": "df83233b3d8c79a867846f64979d4ae7803d9a4f061d3fcd4a2327cc1d8c2578", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2451, + "estimated_input_tokens": 613, + "estimated_eval_tokens": 1482 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 2451, + "estimated_input_tokens": 613, + "estimated_eval_tokens": 1482 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is only loosely related by vague pipeline/diffusion wording. The first two issues may be adjacent, but the available evidence does not show the same concrete bug; the third looks like a separate feature idea.", + "confidence": 0.61, + "canonical_issue_reason": "issue #3317 is the most concrete, bug-shaped report (scheduler not applying a denoising step). The other issues are broader/vaguer and do not clearly collapse into the same defect.", + "canonical_pr_reason": null, + "best_issue_reason": "#3317 is the best representative issue because it names a specific failing code path and is the most actionable of the three.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3317", + "right": "issue:3502", + "accept": false, + "reason": "Both mention pipelines, but the titles do not identify the same failure mode; #3317 is about a scheduler skipping denoising, while #3502 is a vague rules/pipeline complaint. Not enough evidence they are duplicates." + }, + { + "left": "issue:3502", + "right": "issue:4823", + "accept": false, + "reason": "#3502 sounds like a broken-pipeline complaint, while #4823 is framed as a feature/concept proposal for more flexible diffusers. These do not look like the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 70, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-369-2", + "nodes": [ + "issue:3352", + "issue:369" + ], + "soft_pairs": [ + "issue:3352|issue:369" + ], + "prepared_review_unit_hash": "326d9f3f71407986d830412b40885066204c0e6a84eeed2f49c2006c2e183c05", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1523, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1523, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues touch multi-token / multi-vector textual inversion, but the evidence is too thin to treat them as the same request with confidence. #369 is the clearer feature request; #3352 reads more like a usage/help question.", + "confidence": 0.64, + "canonical_issue_reason": "#369 is the broader, more direct feature request about representing a learned concept with more than one token.", + "canonical_pr_reason": null, + "best_issue_reason": "#369 is the best representative because it states the underlying enhancement clearly, whereas #3352 is framed as how to apply the technique.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3352", + "right": "issue:369", + "accept": false, + "reason": "The topics are related, but the packet does not firmly show the same underlying issue: #3352 is a how-to/support question and #369 is a feature request. I would treat this as a plausible relation, not a confirmed duplicate." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The relationship is plausible from the titles, but the soft-edge verdict is a bit too strong given the limited evidence in the packet. #3352 reads as a how-to/support question, while #369 is a feature request; treating them as the same underlying request is not fully grounded and should be stated more cautiously." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 71, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3423-2", + "nodes": [ + "issue:3423", + "issue:4490" + ], + "soft_pairs": [ + "issue:3423|issue:4490" + ], + "prepared_review_unit_hash": "c3b8b8c08517efb1880d593081bb730b7c444510f4ec579d0d4483d6017bfffa", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1523, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1523, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two closed issues do not look like duplicates: one describes a launch hang when creating a model, while the other is a checkpoint-to-\u0926iffusers conversion error involving text_config_dict. They share only a broad model-related theme.", + "confidence": 0.93, + "canonical_issue_reason": "No canonical issue: the reports describe different failure modes and code paths, so neither clearly represents the other.", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a good global representative for the cluster because they are not the same underlying bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3423", + "right": "issue:4490", + "accept": false, + "reason": "Different symptoms and likely different subsystems: a startup/launch stall versus a conversion-time config error. The overlap is only broad model tooling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 72, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3538-2", + "nodes": [ + "issue:3538", + "issue:5756" + ], + "soft_pairs": [ + "issue:3538|issue:5756" + ], + "prepared_review_unit_hash": "757fc8c96c8f684fac6dfe4030071c2f88993f96da6c8ae9f63c4a718bea34eb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1539, + "estimated_input_tokens": 385, + "estimated_eval_tokens": 1026 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1539, + "estimated_input_tokens": 385, + "estimated_eval_tokens": 1026 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are both about LoRA-related workflows, but they ask different questions: one about finetuning ControlNet when merging SD models with different LoRAs, and the other about generating an LCM LoRA from an existing model. They do not look like duplicates of the same underlying bug or request.", + "confidence": 0.9, + "canonical_issue_reason": "Issue #5756 is the more substantive and discussion-heavy support question, but the cluster is not truly duplicate-worthy because the topics differ.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #5756 is the best representative of the cluster only by activity and specificity; it is the clearest standalone issue here.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3538", + "right": "issue:5756", + "accept": false, + "reason": "Different requests and different workflows: merging SD/ControlNet with LoRAs vs creating an LCM LoRA. They share the LoRA keyword, but not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 73, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3544-2", + "nodes": [ + "issue:3544", + "issue:4785" + ], + "soft_pairs": [ + "issue:3544|issue:4785" + ], + "prepared_review_unit_hash": "f6523f1b9f63f061cf8d7866b03727978c5a8f9303a1310f27c8477f16711ba6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1504, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1504, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: #3544 asks a conceptual question about whether latents should be scaled by the scheduler, while #4785 is a separate typo report in the Euler discrete scheduler. They mention the same general area, but not the same underlying bug or change.", + "confidence": 0.93, + "canonical_issue_reason": "#3544 is the more substantive and general scheduler-related issue; #4785 is a narrow typo report and does not appear to share the same problem statement.", + "canonical_pr_reason": null, + "best_issue_reason": "#3544 is the best representative of the cluster because it covers the broader technical question, whereas #4785 is unrelated beyond the scheduler context.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3544", + "right": "issue:4785", + "accept": false, + "reason": "Different topics: one is about latent scaling semantics, the other is a typo in a specific scheduler implementation. Shared subsystem alone is not enough to mark them duplicate." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 74, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-358-2", + "nodes": [ + "issue:358", + "issue:363" + ], + "soft_pairs": [ + "issue:358|issue:363" + ], + "prepared_review_unit_hash": "f8aad0a933d1eedc88261e1fcb6ea662af3ef0ae1a44ba65ce78ce6ce0d34047", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1520, + "estimated_input_tokens": 380, + "estimated_eval_tokens": 1016 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1520, + "estimated_input_tokens": 380, + "estimated_eval_tokens": 1016 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues both mention Apple MPS, but they describe different failure modes: one is a concrete error in `unet_2d_condition.py`, while the other says prompt arrays fail to generate images at all. I do not think they are the same underlying bug.", + "confidence": 0.82, + "canonical_issue_reason": "Issue #358 is the more concrete report: it names the specific failing file (`unet_2d_condition.py`) and describes an explicit MPS error, making it the better anchor for this cluster if a single issue must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #358 is the best issue candidate because it is more specific and actionable than #363, which is broader and symptom-level.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:358", + "right": "issue:363", + "accept": false, + "reason": "Reject: both mention Apple MPS, but one is a file-specific runtime error and the other is a prompt-array generation failure; the symptoms and likely code paths are not clearly the same." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 75, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3684-2", + "nodes": [ + "issue:3684", + "issue:4150" + ], + "soft_pairs": [ + "issue:3684|issue:4150" + ], + "prepared_review_unit_hash": "8a05264a4aaaefe33883aca23ffdaa951103378a4bcc98762e59e2a4f82f27e5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related only at a very high level (both ask about model training), but they describe different tasks: img2img fine-tuning vs. training a text-to-image model based on SDXL. They do not look like duplicates of the same underlying issue.", + "confidence": 0.93, + "canonical_issue_reason": "Issue #4150 is the more specific and actionable training question (SDXL text-to-image), so it is the better representative issue if one must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "#4150 is the best issue candidate because it names the concrete model family and training setup, making it easier to anchor discussion than the broader img2img fine-tuning question.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3684", + "right": "issue:4150", + "accept": false, + "reason": "Different training goals and model types: one asks about img2img fine-tuning, the other about training a text-to-image SDXL model. Shared theme is too broad to treat as the same issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 76, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-380-2", + "nodes": [ + "issue:380", + "issue:465" + ], + "soft_pairs": [ + "issue:380|issue:465" + ], + "prepared_review_unit_hash": "a9c935f3a0984d6fa2e20a8e83adf1ff0e0268dc47e175e6f9993b7603102886", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1490, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1490, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are related to the scheduler, but they point at different problems: one is about unused `init` parameters, the other about inconsistent `offset` handling. They do not look like the same underlying bug.", + "confidence": 0.89, + "canonical_issue_reason": "Issue #465 is the more concrete bug report: it names a specific scheduler behavior discrepancy (`offset` usage) rather than a generic cleanup concern.", + "canonical_pr_reason": null, + "best_issue_reason": "#465 is the best anchor for the cluster because it describes the actionable functional issue and has more inbound references.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:380", + "right": "issue:465", + "accept": false, + "reason": "Different problem statements: unused `init` parameters vs `offset` usage discrepancy. No shared filenames/targets, and the similarity is too weak to treat as the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 77, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-405-2", + "nodes": [ + "issue:3827", + "issue:405" + ], + "soft_pairs": [ + "issue:3827|issue:405" + ], + "prepared_review_unit_hash": "cf4e86ec096a68dd4a232a1f0f44c5e91e02e2e35dc97959ca9ced1fe790501f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1532, + "estimated_input_tokens": 383, + "estimated_eval_tokens": 1022 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1532, + "estimated_input_tokens": 383, + "estimated_eval_tokens": 1022 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues look possibly related at the title level, but the packet does not provide enough evidence to treat them as the same underlying bug. The match is weak: no shared targets, filenames, or body-level details tie them to one concrete root cause.", + "confidence": 0.61, + "canonical_issue_reason": "Issue 405 is still the better canonical issue because it is the broader, earlier report and uses more generic wording that can absorb related tokenizer-error reports, but it should be treated cautiously rather than as a confirmed duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 405 is the best issue representative for clustering because its wording is more general and likely to match related reports, while 3827 appears more specific to one usage scenario.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3827", + "right": "issue:405", + "accept": false, + "reason": "Reject as a duplicate edge: the evidence is limited to modest title similarity and a soft similarity signal. There are no shared targets, filenames, or body details showing the same root cause, so they should only be considered possibly related." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The merge/duplication claim is too strong for the packet evidence. We only have title-level similarity and a modest Jaccard score, with no shared targets, filenames, or body evidence showing the same root cause. The summary and soft-edge verdict should stay more conservative (e.g., 'possibly related' rather than 'same underlying bug')." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 78, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3988-2", + "nodes": [ + "issue:3988", + "issue:5503" + ], + "soft_pairs": [ + "issue:3988|issue:5503" + ], + "prepared_review_unit_hash": "18656b2a62f44c66b725b072fb27c39dc634ca29b278e84b0a745c760a076fc2", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1549, + "estimated_input_tokens": 388, + "estimated_eval_tokens": 1032 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1549, + "estimated_input_tokens": 388, + "estimated_eval_tokens": 1032 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are both about ControlNet, but they describe different user problems: one asks how to use part of a ControlNet model with an inpaint pipeline, while the other is specifically about ControlNet Scribble with SDXL. They do not look like the same underlying bug or change.", + "confidence": 0.87, + "canonical_issue_reason": "No clear duplicate cluster among the issues; if one must be treated as the representative issue, #5503 is the more concrete, product-specific report.", + "canonical_pr_reason": null, + "best_issue_reason": "#5503 is the better representative issue because it is a more specific, actionable problem statement tied to a particular ControlNet/SDXL configuration.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:3988", + "right": "issue:5503", + "accept": false, + "reason": "Different use cases and failure modes: one is a general question about partial ControlNet usage with an inpaint pipeline, the other is a specific SDXL Scribble ControlNet issue. Similar subsystem, but not the same underlying problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 79, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4017-2", + "nodes": [ + "issue:4017", + "issue:7231" + ], + "soft_pairs": [ + "issue:4017|issue:7231" + ], + "prepared_review_unit_hash": "ca7d70cfe321a136b1b279ffb9e6675618590c8516662446b039a8a0d4d8b4f4", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1490, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1490, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues look related around blockwise/block-based LoRA weighting, but the evidence is limited to title similarity. There is not enough body-level detail to confirm they are duplicates, so the soft link should be treated as a likely relation rather than a merge-worthy match.", + "confidence": 0.64, + "canonical_issue_reason": "Issue 7231 is the clearest and most specific wording of the apparent request, so it is the better representative if these are grouped together.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 7231 is the best issue representative because it is more explicit and readable, but the similarity alone is not enough to call it a confirmed duplicate of 4017.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4017", + "right": "issue:7231", + "accept": false, + "reason": "The titles are broadly similar, but there is no supporting body evidence, explicit linkage, or shared target/file evidence to confirm the same underlying request. They appear related, not proven duplicates." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The linkage is plausible from the titles, but the packet only provides soft similarity and no body-level evidence. The analyst states the issues are the same request too definitively; a more conservative verdict would be that they appear related, not confirmed duplicates." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 80, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4161-2", + "nodes": [ + "issue:4161", + "issue:4162" + ], + "soft_pairs": [ + "issue:4161|issue:4162" + ], + "prepared_review_unit_hash": "d32dca72efad0e732d0b7e9522b388dbdc6b0adc08c167684c8f407bef1b7fa3", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1476, + "estimated_input_tokens": 369, + "estimated_eval_tokens": 994 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1476, + "estimated_input_tokens": 369, + "estimated_eval_tokens": 994 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two different issue reports in SD.Next, not duplicates: one about Kandinsky Prior and one about Img2Img.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue: the reports describe different problems and different code paths.", + "canonical_pr_reason": null, + "best_issue_reason": "4161 is the more specific report, but it is not a duplicate target for 4162.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4161", + "right": "issue:4162", + "accept": false, + "reason": "The titles point to unrelated failures in different features (Kandinsky Prior vs Img2Img), so they do not appear to be the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 81, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4367-2", + "nodes": [ + "issue:4367", + "issue:6863" + ], + "soft_pairs": [ + "issue:4367|issue:6863" + ], + "prepared_review_unit_hash": "b8f883838666efc1bd24ea428e65c8f1a206dd118f30ba5bbc220dbb7fd074e0", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1512, + "estimated_input_tokens": 378, + "estimated_eval_tokens": 1012 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1512, + "estimated_input_tokens": 378, + "estimated_eval_tokens": 1012 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues request the same missing AutoPipeline `from_single_file()` support, phrased as a feature gap in the same API surface. They are likely duplicates, with 6863 being the clearer, later report.", + "confidence": 0.91, + "canonical_issue_reason": "issue:6863 is the more direct and specific report (\u201cfrom_single_file missing in AutoPipeline API\u201d), making it the best representative of the underlying request.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:6863 best captures the concrete API gap in current wording and is the clearest canonical issue for deduping.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4367", + "right": "issue:6863", + "accept": true, + "reason": "Same underlying request: AutoPipeline should support `from_single_file()`. The titles differ only in wording, not in substance." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both issues describe the same AutoPipeline `from_single_file()` gap, and the duplicate/representative call is conservative. No evidence is overstated." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 82, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4601-2", + "nodes": [ + "issue:4601", + "issue:4602" + ], + "soft_pairs": [ + "issue:4601|issue:4602" + ], + "prepared_review_unit_hash": "16761e15c47f064def76d90ab3fe932d9c1ae4f16164af494c6f0d02658fcf83", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1488, + "estimated_input_tokens": 372, + "estimated_eval_tokens": 1000 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1488, + "estimated_input_tokens": 372, + "estimated_eval_tokens": 1000 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related ControlNet SDXL checkpoint requests, but they ask for different model variants and do not appear to be duplicates.", + "confidence": 0.95, + "canonical_issue_reason": "Issue 4602 is the more general request about SDXL refiner+base support, so it is the better representative if one must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a strong duplicate target; 4602 is slightly broader and more central, while 4601 is a specific NormalBae/Tile/MLSD checkpoint request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4601", + "right": "issue:4602", + "accept": false, + "reason": "They request different ControlNet SDXL checkpoint variants (NormalBae/Tile/MLSD vs refiner+base), so they are not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 83, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4658-2", + "nodes": [ + "issue:4658", + "issue:6763" + ], + "soft_pairs": [ + "issue:4658|issue:6763" + ], + "prepared_review_unit_hash": "f9538dde15a1875ec808c6c08a20042e26de496f5deec9b1f73335b4be8a34b9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1508, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1508, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one asks whether AudioLDM should use a separate UNet class, while the other reports that `timestep` should be forced to a tensor in UNet. They concern different code paths and different kinds of changes.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the issues are about unrelated problems (pipeline/class design vs UNet input typing).", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a good cluster anchor because they do not describe the same underlying bug or change.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4658", + "right": "issue:6763", + "accept": false, + "reason": "Different subjects and remedies: AudioLDM pipeline class separation vs forcing `timestep` to tensor in UNet; no shared concrete bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 84, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4732-2", + "nodes": [ + "issue:4732", + "issue:4970" + ], + "soft_pairs": [ + "issue:4732|issue:4970" + ], + "prepared_review_unit_hash": "c1e58fa4e1b872dbd75e8a172ddba4dee7619a36c2b11354f4dbc8862a2b4cd9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1494, + "estimated_input_tokens": 374, + "estimated_eval_tokens": 1004 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1494, + "estimated_input_tokens": 374, + "estimated_eval_tokens": 1004 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are about different topics and do not look like duplicates: one is about prompt precedence in sd-webui-controlnet/diffusers, the other is about tile_colorfix in diffusers.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 4732 is the more general and clearly stated user-facing problem, but it is not a canonical duplicate of 4970 because the underlying bugs differ.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4732", + "right": "issue:4970", + "accept": false, + "reason": "The titles describe unrelated behaviors: prompt importance/controlnet precedence versus tile_colorfix. No shared code path or concrete duplicate signal is present." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 85, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4900-2", + "nodes": [ + "issue:4900", + "issue:5691" + ], + "soft_pairs": [ + "issue:4900|issue:5691" + ], + "prepared_review_unit_hash": "4040c8309fa719d6c02482785a17707bc6964802fdf75521f605aaaa82814cdb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1506, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1506, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are closed issues about inpaint behavior/settings, but they appear to cover different user questions: one about choosing a starting point, the other about explaining inpaint modes (fill/original/latent options). They are related, but not clearly duplicates of the same underlying bug or request.", + "confidence": 0.76, + "canonical_issue_reason": "Issue #5691 is the broader, more descriptive inpaint-mode issue and is the better canonical reference for this cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "#5691 is the best representative because it explicitly covers the inpaint mode options and is broader than the narrower starting-point request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:4900", + "right": "issue:5691", + "accept": false, + "reason": "Related inpaint topics, but the titles suggest different requests: selecting a starting point vs understanding/changing inpaint mode variants. Not enough evidence they are the same bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 86, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5141-2", + "nodes": [ + "issue:5141", + "issue:5566" + ], + "soft_pairs": [ + "issue:5141|issue:5566" + ], + "prepared_review_unit_hash": "d51ff64c2fb533033f3406e564bf0830fb089f5de265d9fca372b3b945bfd471", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1495, + "estimated_input_tokens": 374, + "estimated_eval_tokens": 1004 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1495, + "estimated_input_tokens": 374, + "estimated_eval_tokens": 1004 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: one asks for configuring how many checkpoints to save, while the other asks to save multiple models into a single file. They touch related storage behavior but different user goals and likely different implementations.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 5566 is the more concrete representative of the storage/serialization theme, but it should not be treated as a duplicate of 5141.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5141", + "right": "issue:5566", + "accept": false, + "reason": "Different requests: checkpoint retention count vs packing multiple models into one file. Related area, but not the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 87, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5228-13", + "nodes": [ + "issue:5228", + "issue:5229", + "issue:5230", + "issue:5231", + "issue:5241", + "issue:5252", + "issue:5257", + "issue:5260", + "issue:5262", + "issue:5263", + "issue:5264", + "issue:5302", + "issue:5548" + ], + "soft_pairs": [ + "issue:5228|issue:5229", + "issue:5228|issue:5230", + "issue:5228|issue:5231", + "issue:5228|issue:5241", + "issue:5228|issue:5302", + "issue:5229|issue:5230", + "issue:5229|issue:5231", + "issue:5229|issue:5241", + "issue:5229|issue:5302", + "issue:5230|issue:5231", + "issue:5230|issue:5241", + "issue:5230|issue:5302", + "issue:5231|issue:5241", + "issue:5231|issue:5302", + "issue:5241|issue:5302", + "issue:5252|issue:5548", + "issue:5228|issue:5257", + "issue:5229|issue:5257", + "issue:5230|issue:5257", + "issue:5231|issue:5257", + "issue:5241|issue:5257", + "issue:5260|issue:5262", + "issue:5260|issue:5263", + "issue:5260|issue:5264", + "issue:5228|issue:5260", + "issue:5229|issue:5260", + "issue:5262|issue:5263", + "issue:5262|issue:5264", + "issue:5228|issue:5262", + "issue:5229|issue:5262", + "issue:5263|issue:5264", + "issue:5228|issue:5263", + "issue:5229|issue:5263", + "issue:5228|issue:5264", + "issue:5229|issue:5264", + "issue:5228|issue:5548", + "issue:5229|issue:5548", + "issue:5230|issue:5548", + "issue:5231|issue:5548" + ], + "prepared_review_unit_hash": "c26f49091c2e99c7959774545c7ed1b1ab5f7e0eedf4818975255bb3cea1bba2", + "component_budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 39, + "serialized_chars": 20752, + "estimated_input_tokens": 5188, + "estimated_eval_tokens": 10632 + }, + "budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 39, + "serialized_chars": 20752, + "estimated_input_tokens": 5188, + "estimated_eval_tokens": 10632 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are mostly separate HacktoberFest issues with similar wording but different concrete targets (different models, functions, pipelines, or language translations). I would not collapse any of the soft pairs as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 5228 is the broadest representative of the 'Missing Docstrings' task set, but it is still a distinct issue rather than a duplicate of the others.", + "canonical_pr_reason": null, + "best_issue_reason": "5228 is the best single issue to anchor the cluster because it is a clear, central docstring task and the most general among the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5228", + "right": "issue:5229", + "accept": false, + "reason": "Different targets: UNet2DConditionModel vs Autoencoder encode." + }, + { + "left": "issue:5228", + "right": "issue:5230", + "accept": false, + "reason": "Different targets: UNet2DConditionModel vs Autoencoder decode." + }, + { + "left": "issue:5228", + "right": "issue:5231", + "accept": false, + "reason": "Different targets: UNet2DConditionModel vs Adapters." + }, + { + "left": "issue:5228", + "right": "issue:5241", + "accept": false, + "reason": "Different targets: UNet2DConditionModel vs W\u00fcrstchen." + }, + { + "left": "issue:5228", + "right": "issue:5302", + "accept": false, + "reason": "Different tasks: missing docstrings for a model vs DeepFloyd IF docstrings." + }, + { + "left": "issue:5229", + "right": "issue:5230", + "accept": false, + "reason": "Related autoencoder methods, but encode and decode are distinct functions and not the same underlying task." + }, + { + "left": "issue:5229", + "right": "issue:5231", + "accept": false, + "reason": "Different targets: Autoencoder encode vs Adapters." + }, + { + "left": "issue:5229", + "right": "issue:5241", + "accept": false, + "reason": "Different targets: Autoencoder encode vs W\u00fcrstchen." + }, + { + "left": "issue:5229", + "right": "issue:5302", + "accept": false, + "reason": "Different tasks: Autoencoder encode docstrings vs DeepFloyd IF docstrings." + }, + { + "left": "issue:5230", + "right": "issue:5231", + "accept": false, + "reason": "Different targets: Autoencoder decode vs Adapters." + }, + { + "left": "issue:5230", + "right": "issue:5241", + "accept": false, + "reason": "Different targets: Autoencoder decode vs W\u00fcrstchen." + }, + { + "left": "issue:5230", + "right": "issue:5302", + "accept": false, + "reason": "Different tasks: Autoencoder decode docstrings vs DeepFloyd IF docstrings." + }, + { + "left": "issue:5231", + "right": "issue:5241", + "accept": false, + "reason": "Different targets: Adapters vs W\u00fcrstchen." + }, + { + "left": "issue:5231", + "right": "issue:5302", + "accept": false, + "reason": "Different tasks: Adapters docstrings vs DeepFloyd IF docstrings." + }, + { + "left": "issue:5241", + "right": "issue:5302", + "accept": false, + "reason": "Different targets: W\u00fcrstchen vs DeepFloyd IF docstrings." + }, + { + "left": "issue:5252", + "right": "issue:5548", + "accept": false, + "reason": "Same broad documentation area, but different languages (Thai vs Portuguese) and separate translation work." + }, + { + "left": "issue:5228", + "right": "issue:5257", + "accept": false, + "reason": "One is docstrings for a model; the other is removal of outdated schedulers." + }, + { + "left": "issue:5229", + "right": "issue:5257", + "accept": false, + "reason": "One is docstrings for an autoencoder method; the other is removal of outdated schedulers." + }, + { + "left": "issue:5230", + "right": "issue:5257", + "accept": false, + "reason": "One is docstrings for an autoencoder method; the other is removal of outdated schedulers." + }, + { + "left": "issue:5231", + "right": "issue:5257", + "accept": false, + "reason": "One is docstrings for adapters; the other is removal of outdated schedulers." + }, + { + "left": "issue:5241", + "right": "issue:5257", + "accept": false, + "reason": "One is docstrings for W\u00fcrstchen; the other is removal of outdated schedulers." + }, + { + "left": "issue:5260", + "right": "issue:5262", + "accept": false, + "reason": "Same test-sizing theme, but different pipelines: StableDiffusionXLPipeline vs StableDiffusionPipeline." + }, + { + "left": "issue:5260", + "right": "issue:5263", + "accept": false, + "reason": "Same test-sizing theme, but different pipelines: StableDiffusionXLPipeline vs StableDiffusionControlNetPipeline." + }, + { + "left": "issue:5260", + "right": "issue:5264", + "accept": false, + "reason": "Same test-sizing theme, but different pipelines: StableDiffusionXLPipeline vs StableDiffusionControlNetImg2ImgPipeline." + }, + { + "left": "issue:5228", + "right": "issue:5260", + "accept": false, + "reason": "Different tasks: docstrings for UNet2DConditionModel vs shrinking a fast test for StableDiffusionXLPipeline." + }, + { + "left": "issue:5229", + "right": "issue:5260", + "accept": false, + "reason": "Different tasks: autoencoder encode docstrings vs shrinking a fast test for StableDiffusionXLPipeline." + }, + { + "left": "issue:5262", + "right": "issue:5263", + "accept": false, + "reason": "Same test-sizing theme, but different pipelines: StableDiffusionPipeline vs StableDiffusionControlNetPipeline." + }, + { + "left": "issue:5262", + "right": "issue:5264", + "accept": false, + "reason": "Same test-sizing theme, but different pipelines: StableDiffusionPipeline vs StableDiffusionControlNetImg2ImgPipeline." + }, + { + "left": "issue:5228", + "right": "issue:5262", + "accept": false, + "reason": "Different tasks: docstrings for UNet2DConditionModel vs shrinking a fast test for StableDiffusionPipeline." + }, + { + "left": "issue:5229", + "right": "issue:5262", + "accept": false, + "reason": "Different tasks: autoencoder encode docstrings vs shrinking a fast test for StableDiffusionPipeline." + }, + { + "left": "issue:5263", + "right": "issue:5264", + "accept": false, + "reason": "Same test-sizing theme, but different pipelines: StableDiffusionControlNetPipeline vs StableDiffusionControlNetImg2ImgPipeline." + }, + { + "left": "issue:5228", + "right": "issue:5263", + "accept": false, + "reason": "Different tasks: docstrings for UNet2DConditionModel vs shrinking a fast test for StableDiffusionControlNetPipeline." + }, + { + "left": "issue:5229", + "right": "issue:5263", + "accept": false, + "reason": "Different tasks: autoencoder encode docstrings vs shrinking a fast test for StableDiffusionControlNetPipeline." + }, + { + "left": "issue:5228", + "right": "issue:5264", + "accept": false, + "reason": "Different tasks: docstrings for UNet2DConditionModel vs shrinking a fast test for StableDiffusionControlNetImg2ImgPipeline." + }, + { + "left": "issue:5229", + "right": "issue:5264", + "accept": false, + "reason": "Different tasks: autoencoder encode docstrings vs shrinking a fast test for StableDiffusionControlNetImg2ImgPipeline." + }, + { + "left": "issue:5228", + "right": "issue:5548", + "accept": false, + "reason": "Different documentation tasks: missing docstrings vs Portuguese translation work." + }, + { + "left": "issue:5229", + "right": "issue:5548", + "accept": false, + "reason": "Different documentation tasks: autoencoder encode docstrings vs Portuguese translation work." + }, + { + "left": "issue:5230", + "right": "issue:5548", + "accept": false, + "reason": "Different documentation tasks: autoencoder decode docstrings vs Portuguese translation work." + }, + { + "left": "issue:5231", + "right": "issue:5548", + "accept": false, + "reason": "Different documentation tasks: adapters docstrings vs Portuguese translation work." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 88, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5311-2", + "nodes": [ + "issue:5311", + "issue:7366" + ], + "soft_pairs": [ + "issue:5311|issue:7366" + ], + "prepared_review_unit_hash": "9baa15ce715d4bc4c18dd5b8396d7a1c035ada17a0d724c727c80cd770458816", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1546, + "estimated_input_tokens": 387, + "estimated_eval_tokens": 1030 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1546, + "estimated_input_tokens": 387, + "estimated_eval_tokens": 1030 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two closed issues are related only at a broad API level (`load_lora_weights`), but they describe different failure modes: one is a missing attribute on `StableDiffusionXLPipeline`, the other is a dora-specific loading error. I would not treat them as duplicates.", + "confidence": 0.87, + "canonical_issue_reason": "Issue #5311 is the cleaner canonical issue because it names a concrete, specific exception (`StableDiffusionXLPipeline` missing `load_lora_weights`) and is the more directly actionable bug report.", + "canonical_pr_reason": null, + "best_issue_reason": "#5311 is the best representative of the cluster since it captures the core API/attribute problem most precisely; #7366 is narrower and appears to involve a different error path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5311", + "right": "issue:7366", + "accept": false, + "reason": "Both mention `load_lora_weights`, but the reported problems are not the same underlying bug: one is an AttributeError on SDXL pipeline, the other is a dora-related failure. Too broad to merge as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 89, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5333-2", + "nodes": [ + "issue:5333", + "issue:5375" + ], + "soft_pairs": [ + "issue:5333|issue:5375" + ], + "prepared_review_unit_hash": "0deae05513793c00509fb1782ebd4d0e53e0a7ff0b44f709d25f2d27484bc701", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1637, + "estimated_input_tokens": 410, + "estimated_eval_tokens": 1076 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1637, + "estimated_input_tokens": 410, + "estimated_eval_tokens": 1076 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These issues report a similar missing `from_single_file` attribute, but they affect different pipeline classes and there\u2019s not enough evidence they\u2019re the same concrete bug instance.", + "confidence": 0.58, + "canonical_issue_reason": "#5333 is the better canonical issue because it was opened earlier and already describes the missing `from_single_file` problem across multiple ControlNet inpaint pipeline variants.", + "canonical_pr_reason": null, + "best_issue_reason": "#5333 is the broadest and most representative report of the defect, so it is the best issue to anchor triage.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5333", + "right": "issue:5375", + "accept": false, + "reason": "Both mention `from_single_file`, but they concern different pipeline classes (`ControlNetInpaint*` vs `StableDiffusionXLControlNetPipeline`). That is too broad to treat as the same underlying bug without a shared code-path or fix evidence." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 90, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5762-2", + "nodes": [ + "issue:5762", + "issue:9733" + ], + "soft_pairs": [ + "issue:5762|issue:9733" + ], + "prepared_review_unit_hash": "c56561610ac3c46e0a376adeb069497581a101378c3f64ab7e958a8d359731f7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1557, + "estimated_input_tokens": 390, + "estimated_eval_tokens": 1036 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1557, + "estimated_input_tokens": 390, + "estimated_eval_tokens": 1036 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues describe unrelated problems: one is a TemporalConvLayer dimension-mismatch error, the other is a FluxControlNetInpaintPipeline compatibility problem with XLabs controlnets. They are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #9733 is the broader, more specific user-facing bug report with clearer context and newer discussion activity.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #9733 is the better canonical issue because it identifies a concrete pipeline incompatibility and appears to capture the current actionable problem more clearly than the older, narrower TemporalConvLayer error report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:5762", + "right": "issue:9733", + "accept": false, + "reason": "These reports concern different code paths and symptoms: a TemporalConvLayer dimension mismatch versus a FluxControlNetInpaintPipeline/XLabs controlnet incompatibility. They do not look like the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 91, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6033-2", + "nodes": [ + "issue:6033", + "issue:6034" + ], + "soft_pairs": [ + "issue:6033|issue:6034" + ], + "prepared_review_unit_hash": "88cc9339d70cac902f4469d8a81ccd71373766a651b5e4c1b42ae1c4ddc61e08", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1669, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1669, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "6034 is the better-documented issue to keep as canonical, but the match with 6033 is only a soft one based on title similarity and the shared error signature.", + "confidence": 0.74, + "canonical_issue_reason": "6034 is the stronger report because it includes the triggering context (converting a VAE safetensor to diffusers) in addition to the same UnicodeDecodeError title seen in 6033.", + "canonical_pr_reason": null, + "best_issue_reason": "6034 is the best issue to keep as the representative report since it has more actionable context and the same visible error signature as 6033.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6033", + "right": "issue:6034", + "accept": false, + "reason": "The packet only shows title-level overlap and a soft_similarity signal; there is not enough body evidence to confirm they are the same underlying bug." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The canonical choice is plausible, but the writeup is a bit too confident for the evidence in the packet. We only have title-level overlap plus a soft_similarity signal; there is no body evidence tying both reports to the same root cause. Saying they are the same underlying failure / clearly the same conversion failure overstates what is directly supported. A more conservative phrasing would treat 6034 as the better-documented candidate while noting the match is only soft." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 92, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-615-2", + "nodes": [ + "issue:615", + "issue:765" + ], + "soft_pairs": [ + "issue:615|issue:765" + ], + "prepared_review_unit_hash": "1b5cb7d587897acf0456ca31a307fcffe02130f90f0e96a01a97cb0f0402ecab", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1494, + "estimated_input_tokens": 374, + "estimated_eval_tokens": 1004 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1494, + "estimated_input_tokens": 374, + "estimated_eval_tokens": 1004 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two unrelated closed feature/help questions: one asks about an outpainting demo, the other about adding undesired content to the txt2img pipe. They do not describe the same underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:615", + "right": "issue:765", + "accept": false, + "reason": "Different topics and user intents; no shared concrete problem, code path, or change request beyond both being pipeline-related questions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 93, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6406-2", + "nodes": [ + "issue:6406", + "issue:6430" + ], + "soft_pairs": [ + "issue:6406|issue:6430" + ], + "prepared_review_unit_hash": "55ed4235ef36a12592b8e5d7078a141fb02ccac33a4b5126ba8331d75b639fe2", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1543, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1543, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are related to custom pipeline support, but they describe different problems: one is a concrete feature/bug in `lpw_stable_diffusion_xl`, the other is documentation inaccuracy for custom pipelines. They should not be clustered as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": "#6406 is the more concrete, actionable bug report: it identifies a specific pipeline and the missing support for new run parameters.", + "canonical_pr_reason": null, + "best_issue_reason": "#6406 is the best representative issue because it describes a specific runtime behavior problem rather than a documentation complaint.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6406", + "right": "issue:6430", + "accept": false, + "reason": "Related topic, but not the same underlying issue: one is a pipeline feature gap, the other says the custom pipeline docs are inaccurate." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 94, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6882-2", + "nodes": [ + "issue:6882", + "issue:7765" + ], + "soft_pairs": [ + "issue:6882|issue:7765" + ], + "prepared_review_unit_hash": "ac689aedd7b6dcf220b5e1f8efb8cc6996f6afb7a1562ca71240f145b31f22ca", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1558, + "estimated_input_tokens": 390, + "estimated_eval_tokens": 1036 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1558, + "estimated_input_tokens": 390, + "estimated_eval_tokens": 1036 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: #6882 reports slow SDXL inference on Cloud TPU v5e for non-1024 image sizes, while #7765 reports fp16 NaN outputs in FlaxStableDiffusionXLPipeline. They describe different failure modes, triggers, and likely code paths.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue: the reports describe distinct bugs rather than one underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "#7765 is the more concrete, reproducible bug report, but it is still separate from #6882.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:6882", + "right": "issue:7765", + "accept": false, + "reason": "Different symptoms and contexts: TPU performance regression at non-1024 sizes versus fp16 NaN generation. Not the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 95, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7383-2", + "nodes": [ + "issue:7383", + "issue:8596" + ], + "soft_pairs": [ + "issue:7383|issue:8596" + ], + "prepared_review_unit_hash": "e3233197e266ab28fa7afe4c560df93d8d2e7a485322dca2498fe1cefc5b7a26", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1513, + "estimated_input_tokens": 379, + "estimated_eval_tokens": 1014 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1513, + "estimated_input_tokens": 379, + "estimated_eval_tokens": 1014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are closely related ControlNet/Playground v2.5 training requests, but the evidence is only title-level soft similarity and does not support a firm duplicate conclusion. No PRs are present.", + "confidence": 0.61, + "canonical_issue_reason": "Issue 8596 is the clearer, more specific phrasing of the shared request, so it is the better issue to use as the representative when grouping these two.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 8596 is the most readable and specific title for the shared topic, but the match is still only a soft relatedness signal rather than a confirmed duplicate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7383", + "right": "issue:8596", + "accept": false, + "reason": "The titles are related, but the packet only shows soft similarity and no stronger evidence that they are the same underlying feature request. This is not enough for a duplicate-style accept." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary is a bit too definitive for the evidence available. The packet only shows title-level soft similarity between two issues, and the pair is explicitly not a deterministic accept. It is reasonable to say they are related, but not safe to conclude they are the same feature request with high confidence. The \"no PRs are present\" part is grounded, but the soft-edge verdict is not conservative enough." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 96, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7608-2", + "nodes": [ + "issue:7608", + "issue:9661" + ], + "soft_pairs": [ + "issue:7608|issue:9661" + ], + "prepared_review_unit_hash": "d76aff7a3939a41f943459e44bdf56fc6ebd5a05fbda8771bbcd9ca72b84841b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1508, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1508, + "estimated_input_tokens": 377, + "estimated_eval_tokens": 1010 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues both concern `from_pretrained`, but they describe different API concerns: one asks for `strict=False`, the other asks whether the `filename` argument was removed. They are related, but not the same underlying bug/request.", + "confidence": 0.43, + "canonical_issue_reason": "Issue #7608 is the broader, more general `from_pretrained` behavior request and is the better representative of this small cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "#7608 is the clearest cluster representative because it is a direct feature request around loading behavior, while #9661 is a narrower parameter-regression question.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7608", + "right": "issue:9661", + "accept": false, + "reason": "Same method, but different concerns: `strict=False` support vs. a missing/removed `filename` argument. They are not clearly the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 97, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7677-2", + "nodes": [ + "issue:7677", + "issue:8907" + ], + "soft_pairs": [ + "issue:7677|issue:8907" + ], + "prepared_review_unit_hash": "41c991290503fa7dcdb365b635c4a5fac854dca303c854f668ab689bae6c06f9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1525, + "estimated_input_tokens": 382, + "estimated_eval_tokens": 1020 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1525, + "estimated_input_tokens": 382, + "estimated_eval_tokens": 1020 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are both in the tests area, but they address different goals: speeding up fast pipeline tests vs improving transformer model test coverage. They do not describe the same underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #7677 is the more central tests-performance thread, with higher discussion and inbound references, so it is the better representative issue if one must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #7677 is the best overall issue to anchor the cluster because it is broader and more actively referenced, while #8907 is a separate coverage-improvement request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7677", + "right": "issue:8907", + "accept": false, + "reason": "Reject: the titles and bodies point to different test workstreams\u2014pipeline speedups versus model test coverage\u2014so they are not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 98, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7928-2", + "nodes": [ + "issue:7928", + "issue:9081" + ], + "soft_pairs": [ + "issue:7928|issue:9081" + ], + "prepared_review_unit_hash": "1ab4e12028da8a76b1ad49aec2cfa2e97ab61dfe1b8c88ab8c85202dcfa4cf95", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1458, + "estimated_input_tokens": 365, + "estimated_eval_tokens": 986 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1458, + "estimated_input_tokens": 365, + "estimated_eval_tokens": 986 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two unrelated feature requests: one for Kandinsky 3.1 support and one for ControlNet support for Flux.1. They should not be clustered as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the two issues request support for different model/features and do not describe the same underlying problem.", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a good canonical representative because they are distinct requests; if forced, 9081 is slightly more concrete, but they should remain separate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:7928", + "right": "issue:9081", + "accept": false, + "reason": "Different feature requests for different models (Kandinsky 3.1 vs Flux.1 ControlNet); not the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 99, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8619-2", + "nodes": [ + "issue:8619", + "issue:8748" + ], + "soft_pairs": [ + "issue:8619|issue:8748" + ], + "prepared_review_unit_hash": "201c0dc46fc5718fd7570d4f1b5812470125a6058f4bd05211a43436a14ae0b7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1543, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1543, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues concern SD3 finetuning/training quality, but they describe different failure modes: one is specifically batch-size-induced destabilization during Dreambooth, the other is a general inability to finetune to better facial/hand quality. They are related, but not clearly the same underlying bug.", + "confidence": 0.68, + "canonical_issue_reason": "Issue 8619 is more specific and actionable, with a concrete trigger (high batch size) and a defined training instability symptom.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 8619 is the better canonical issue because it isolates a reproducible configuration problem rather than a broad quality complaint.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8619", + "right": "issue:8748", + "accept": false, + "reason": "Reject: both are about SD3 finetuning quality, but one targets batch-size destabilization and the other describes general poor finetuning outcomes (face/hand deformation). The underlying bug is not sufficiently specific or shared to treat them as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 100, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8933-2", + "nodes": [ + "issue:8933", + "issue:9315" + ], + "soft_pairs": [ + "issue:8933|issue:9315" + ], + "prepared_review_unit_hash": "a7ef4de9919b5a74aa9ba2428d88151140b996af64ba8d2a907454d551607cca", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1489, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1489, + "estimated_input_tokens": 373, + "estimated_eval_tokens": 1002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are both about Kolors, but they request different capabilities: LoRA training versus ControlNet/IP-Adapter FaceID support. They are not duplicates.", + "confidence": 0.89, + "canonical_issue_reason": "issue #9315 is the better representative because it is the more concrete, actionable feature request; #8933 is a much vaguer training ask.", + "canonical_pr_reason": null, + "best_issue_reason": "#9315 is the clearest issue in the set, with a specific target and broader implementation scope than the terse #8933 request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:8933", + "right": "issue:9315", + "accept": false, + "reason": "Reject: these are different Kolors feature requests. LoRA training and ControlNet/IP-Adapter FaceID support are separate capabilities and do not describe the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 101, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9076-2", + "nodes": [ + "issue:9076", + "issue:9124" + ], + "soft_pairs": [ + "issue:9076|issue:9124" + ], + "prepared_review_unit_hash": "8c1f550baf3a9d3957f6bc33a028beaae0bbc6af89530c522fbd08b799fd5c18", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1539, + "estimated_input_tokens": 385, + "estimated_eval_tokens": 1026 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1539, + "estimated_input_tokens": 385, + "estimated_eval_tokens": 1026 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: one is a feature request for improving `callback_on_step_end` in `FluxPipeline`, the other is a bug report that `negative_prompt` cannot be used in `FluxPipeline`. They concern different API areas and different underlying changes.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 9124 is the more concrete bug report with a clear failure mode (`negative_prompt` not working in `FluxPipeline`); issue 9076 is a separate enhancement request.", + "canonical_pr_reason": null, + "best_issue_reason": "9124 is the best representative issue because it describes a specific, user-facing bug in the pipeline rather than a general API improvement request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:9076", + "right": "issue:9124", + "accept": false, + "reason": "Same subsystem, but different problems: `callback_on_step_end` customization vs `negative_prompt` support. They do not look like the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 102, + "component_count": 102, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9326-2", + "nodes": [ + "issue:9326", + "issue:9329" + ], + "soft_pairs": [ + "issue:9326|issue:9329" + ], + "prepared_review_unit_hash": "57d7d08e80dedc571190863c92186c24ce3f3d1ff88bb33dc8a261701b424325", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1522, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1522, + "estimated_input_tokens": 381, + "estimated_eval_tokens": 1018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues concern AnimateDiff + ControlNet pipeline behavior, but they target different adjacent workflows: VideoToVideo in #9326 and SparseControl in #9329. They look related at a high level, but not like the same underlying bug.", + "confidence": 0.89, + "canonical_issue_reason": "#9329 is the better representative issue because it is more active and its title is slightly more general within the shared AnimateDiff/ControlNet pipeline area.", + "canonical_pr_reason": null, + "best_issue_reason": "#9329 is the best single issue to surface for the cluster since it has substantially more discussion and is still open, making it the more current tracking thread.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue #9326 [Pipeline] AnimateDiff + VideoToVideo + ControlNet", + "right": "issue #9329 [Pipeline] AnimateDiff + SparseControl + ControlNet", + "accept": false, + "reason": "Reject: both mention AnimateDiff and ControlNet, but the middle component differs and likely indicates different pipeline paths (VideoToVideo vs SparseControl). No evidence they describe the same concrete bug or fixable change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10261", + "pull_request:10364", + "pull_request:10365", + "pull_request:10498", + "pull_request:10568", + "pull_request:10570", + "pull_request:11251", + "pull_request:11395", + "pull_request:11426", + "pull_request:11455", + "pull_request:11463", + "pull_request:11465", + "pull_request:11471", + "pull_request:11492", + "pull_request:12400", + "pull_request:12467", + "pull_request:12534", + "pull_request:12698", + "pull_request:12897", + "pull_request:4855", + "pull_request:4927", + "pull_request:5280", + "pull_request:5286", + "pull_request:5478", + "pull_request:5535", + "pull_request:5744", + "pull_request:5748", + "pull_request:6525", + "pull_request:6556", + "pull_request:6941", + "pull_request:7030", + "pull_request:7031", + "pull_request:7442", + "pull_request:7550", + "pull_request:8725", + "pull_request:8916", + "pull_request:8919", + "pull_request:8987", + "pull_request:9386", + "pull_request:9582", + "pull_request:9583", + "pull_request:9606", + "pull_request:9608", + "pull_request:9659", + "pull_request:9724", + "pull_request:9787", + "pull_request:9815", + "pull_request:9843" + ], + "soft_pairs": [ + "pull_request:10261|pull_request:9787", + "pull_request:8725|pull_request:9386", + "pull_request:6525|pull_request:6556", + "pull_request:11395|pull_request:11463", + "pull_request:9606|pull_request:9724", + "pull_request:11426|pull_request:11492", + "pull_request:5286|pull_request:5478", + "pull_request:10365|pull_request:10498", + "pull_request:9583|pull_request:9608", + "pull_request:5280|pull_request:5535", + "pull_request:10364|pull_request:10498", + "pull_request:7442|pull_request:7550", + "pull_request:6941|pull_request:7030", + "pull_request:6941|pull_request:7031", + "pull_request:9606|pull_request:9815", + "pull_request:11251|pull_request:12400", + "pull_request:9659|pull_request:9843", + "pull_request:12467|pull_request:12698", + "pull_request:9583|pull_request:9815", + "pull_request:11463|pull_request:11492", + "pull_request:9582|pull_request:9724", + "pull_request:4855|pull_request:4927", + "pull_request:10568|pull_request:10570", + "pull_request:11395|pull_request:11471", + "pull_request:12534|pull_request:12897", + "pull_request:11395|pull_request:11455", + "pull_request:11395|pull_request:11465", + "pull_request:5744|pull_request:5748", + "pull_request:11455|pull_request:11471", + "pull_request:11465|pull_request:11471", + "pull_request:8919|pull_request:8987", + "pull_request:8916|pull_request:8987" + ], + "prepared_review_unit_hash": "ee55a4ae50b4eb9b81be18c2ea0156c4c9d48797a30d2c9093270b241a2a62b9", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 109783, + "estimated_input_tokens": 27446, + "estimated_eval_tokens": 55148 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous cluster of unrelated or only loosely related PRs. The only clear duplicate-like overlap is the Flux IP-Adapter pair; the rest are same-issue follow-ups across different files/pipelines or broader refactors, so I rejected them.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "10261 is the cleanest merged representative of the Flux IP-Adapter change; 9787 appears to be the overlapping draft/partial version of the same feature work.", + "best_issue_reason": null, + "best_pr_reason": "10498 is the broadest umbrella item in the set (PyTorch/XLA support across many pipelines), so it is the best single representative overall even though the cluster is not a single duplicate family.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10261", + "right": "pull_request:9787", + "accept": true, + "reason": "Same Flux IP-Adapter feature, same file set, same issue target; this looks like the same underlying change, with 9787 being the overlapping draft/partial version." + }, + { + "left": "pull_request:8725", + "right": "pull_request:9386", + "accept": false, + "reason": "Same issue target, but these add PAG support for different pipeline variants; related feature work, not the same concrete code path." + }, + { + "left": "pull_request:6525", + "right": "pull_request:6556", + "accept": false, + "reason": "Both are torch.compile-related training-script tweaks, but they touch different scripts and fixes, so they are not the same change." + }, + { + "left": "pull_request:11395", + "right": "pull_request:11463", + "accept": false, + "reason": "Both switch interpolation defaults, but they are separate training scripts and not the same concrete code-path fix." + }, + { + "left": "pull_request:9606", + "right": "pull_request:9724", + "accept": false, + "reason": "Both are docstring/refactor work in training utilities, but they are not the same change or bug fix." + }, + { + "left": "pull_request:11426", + "right": "pull_request:11492", + "accept": false, + "reason": "Same general interpolation theme, but different scripts and implementations; this is broader follow-up work, not a duplicate." + }, + { + "left": "pull_request:5286", + "right": "pull_request:5478", + "accept": false, + "reason": "The first is an initial Japanese docs page, the second is the fuller docs enablement; related but not the same underlying change." + }, + { + "left": "pull_request:10365", + "right": "pull_request:10498", + "accept": false, + "reason": "10498 is a broader XLA sweep across many pipelines, while 10365 is a single-pipeline addition; related but not duplicate enough." + }, + { + "left": "pull_request:9583", + "right": "pull_request:9608", + "accept": false, + "reason": "Different files and cleanup areas; both are documentation/validation refactors, not the same bug or change." + }, + { + "left": "pull_request:5280", + "right": "pull_request:5535", + "accept": false, + "reason": "Both optimize tests, but they are separate test-configuration edits rather than the same underlying fix." + }, + { + "left": "pull_request:10364", + "right": "pull_request:10498", + "accept": false, + "reason": "10498 subsumes XLA support across multiple pipelines, but 10364 is a distinct single-pipeline patch, so they are not the same PR-level change." + }, + { + "left": "pull_request:7442", + "right": "pull_request:7550", + "accept": false, + "reason": "One is a research-project README/implementation, the other is a community pipeline; same theme, different artifacts and code paths." + }, + { + "left": "pull_request:6941", + "right": "pull_request:7030", + "accept": false, + "reason": "These cover different parts of IP-Adapter support for the same pipeline; one adds support, the other weight-loading integration, so they are complementary not duplicates." + }, + { + "left": "pull_request:6941", + "right": "pull_request:7031", + "accept": false, + "reason": "Same as above: related IP-Adapter work on the same pipeline, but not the same concrete code change." + }, + { + "left": "pull_request:9606", + "right": "pull_request:9815", + "accept": false, + "reason": "Both touch training-utils docstrings, but 9815 is a separate refactor pass with different edits." + }, + { + "left": "pull_request:11251", + "right": "pull_request:12400", + "accept": false, + "reason": "Both concern SD3 multi-controlnet support, but one is constructor/setup work and the other is runtime handling; related but not identical." + }, + { + "left": "pull_request:9659", + "right": "pull_request:9843", + "accept": false, + "reason": "One adds SD3 Img2Img LoRA support, the other is a test fix; not the same underlying change." + }, + { + "left": "pull_request:12467", + "right": "pull_request:12698", + "accept": false, + "reason": "12698 is a broad batch-normalization sweep across several pipelines, not the same concrete Qwen-Image-Edit fix as 12467." + }, + { + "left": "pull_request:9583", + "right": "pull_request:9815", + "accept": false, + "reason": "Both are docstring cleanups in utility code, but they are separate refactors with different scope." + }, + { + "left": "pull_request:4855", + "right": "pull_request:4927", + "accept": false, + "reason": "4927 is a temporary revert of 4855, not a duplicate implementation of the same fix." + }, + { + "left": "pull_request:10568", + "right": "pull_request:10570", + "accept": false, + "reason": "Both change GPU CI tokens in the same workflow, but they target different jobs and are not the same code-path fix." + }, + { + "left": "pull_request:11395", + "right": "pull_request:11471", + "accept": false, + "reason": "Same interpolation theme, but applied to different training scripts; too broad to treat as a duplicate." + }, + { + "left": "pull_request:12534", + "right": "pull_request:12897", + "accept": false, + "reason": "12897 is a broader NPU pos-embed rewrite across multiple transformer models, so it is not the same single-file change as 12534." + }, + { + "left": "pull_request:11395", + "right": "pull_request:11455", + "accept": false, + "reason": "Both add lanczos defaults, but they are separate script updates rather than the same underlying PR." + }, + { + "left": "pull_request:11395", + "right": "pull_request:11465", + "accept": false, + "reason": "Same as above: same theme, different training script and patch." + }, + { + "left": "pull_request:11463", + "right": "pull_request:11492", + "accept": false, + "reason": "Same default-interpolation theme, but different scripts and different patch details." + }, + { + "left": "pull_request:9582", + "right": "pull_request:9608", + "accept": false, + "reason": "Different utility/documentation refactors; not the same bug or change." + }, + { + "left": "pull_request:11455", + "right": "pull_request:11471", + "accept": false, + "reason": "Same interpolation sweep, but separate scripts and distinct edits." + }, + { + "left": "pull_request:11465", + "right": "pull_request:11471", + "accept": false, + "reason": "Both are script-specific interpolation defaults, not the same concrete patch." + }, + { + "left": "pull_request:5744", + "right": "pull_request:5748", + "accept": false, + "reason": "Both are docs typo sweeps, but they cover different page sets and are not the same change." + }, + { + "left": "pull_request:9582", + "right": "pull_request:9724", + "accept": false, + "reason": "Different utility/docstring edits in different files; not the same underlying issue." + }, + { + "left": "pull_request:8919", + "right": "pull_request:8987", + "accept": false, + "reason": "These are test coverage additions for different model families, so they are unrelated beyond being test PRs." + }, + { + "left": "pull_request:8916", + "right": "pull_request:8987", + "accept": false, + "reason": "Different model test additions; same broad area, but not the same change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary stays grounded in the packet and avoids overstating duplicates. The single accepted Flux IP-Adapter pair is well supported by the overlapping file set and shared feature target, while the other pairs are conservatively rejected as same-theme but different concrete changes or broader follow-ups. The representative PR choices are also reasonable for this heterogeneous cluster." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 2, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10259", + "pull_request:10437", + "pull_request:10508", + "pull_request:10578", + "pull_request:11271", + "pull_request:11395", + "pull_request:11421", + "pull_request:11455", + "pull_request:11465", + "pull_request:11472", + "pull_request:11491", + "pull_request:11492", + "pull_request:11578", + "pull_request:11728", + "pull_request:12451", + "pull_request:12856", + "pull_request:12857", + "pull_request:2602", + "pull_request:3633", + "pull_request:4536", + "pull_request:456", + "pull_request:4910", + "pull_request:533", + "pull_request:6332", + "pull_request:6525", + "pull_request:6558", + "pull_request:6576", + "pull_request:6577", + "pull_request:7113", + "pull_request:7117", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7833", + "pull_request:7846", + "pull_request:7898", + "pull_request:8239", + "pull_request:8630", + "pull_request:8755", + "pull_request:9300", + "pull_request:9346", + "pull_request:9579", + "pull_request:9591", + "pull_request:9682", + "pull_request:9706", + "pull_request:9817", + "pull_request:9896" + ], + "soft_pairs": [ + "pull_request:11395|pull_request:11472", + "pull_request:11455|pull_request:11491", + "pull_request:11465|pull_request:11491", + "pull_request:11455|pull_request:11472", + "pull_request:11465|pull_request:11472", + "pull_request:9579|pull_request:9591", + "pull_request:6525|pull_request:6558", + "pull_request:6525|pull_request:6576", + "pull_request:6332|pull_request:8239", + "pull_request:8630|pull_request:8755", + "pull_request:11271|pull_request:11578", + "pull_request:9817|pull_request:9896", + "pull_request:7113|pull_request:7117", + "pull_request:4536|pull_request:4910", + "pull_request:12856|pull_request:12857", + "pull_request:9579|pull_request:9706", + "pull_request:10259|pull_request:10578", + "pull_request:456|pull_request:533", + "pull_request:8755|pull_request:9682", + "pull_request:11728|pull_request:12451", + "pull_request:6525|pull_request:6577", + "pull_request:7797|pull_request:7898", + "pull_request:7803|pull_request:7898", + "pull_request:7804|pull_request:7898", + "pull_request:7833|pull_request:7898", + "pull_request:7846|pull_request:7898", + "pull_request:10437|pull_request:10508", + "pull_request:11421|pull_request:11492", + "pull_request:2602|pull_request:3633", + "pull_request:9300|pull_request:9346" + ], + "prepared_review_unit_hash": "ba872b79af7bdd971fb6ac024626d3109b14d46d28da9bb147bdd384988daff0", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 100186, + "estimated_input_tokens": 25047, + "estimated_eval_tokens": 50350 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous and appears to contain many superficial similarity matches rather than true duplicates. The soft edges mostly connect PRs that share a filename or broad theme but implement different fixes, scripts, or even reversions. I would not consolidate them into a single canonical change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out: the cluster spans unrelated areas (Flux LoRA, test infra, DreamBooth interpolation, LTX single-file support, docs, Docker, and revert PRs), so it is not a true duplicate group.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR for duplicate triage; the items do not describe one underlying bug or change, so selecting a representative PR would be misleading.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11395", + "right": "pull_request:11472", + "accept": false, + "reason": "Same narrow theme (image interpolation defaults) but different training scripts/files and separate changes, so not the same PR-worthy fix." + }, + { + "left": "pull_request:11455", + "right": "pull_request:11491", + "accept": false, + "reason": "Both add LANCZOS interpolation options, but they target different scripts and code paths; not mergeable as one duplicate fix." + }, + { + "left": "pull_request:11465", + "right": "pull_request:11491", + "accept": false, + "reason": "Different SDXL LoRA script vs Lumina2 DreamBooth LoRA script; similar pattern but distinct changes, not one underlying bug." + }, + { + "left": "pull_request:11455", + "right": "pull_request:11472", + "accept": false, + "reason": "Both are interpolation-mode additions, but for different pipelines/scripts and separate implementations." + }, + { + "left": "pull_request:11465", + "right": "pull_request:11472", + "accept": false, + "reason": "Same general resizing-option theme, but the scripts and affected training flows are different." + }, + { + "left": "pull_request:9579", + "right": "pull_request:9591", + "accept": false, + "reason": "Both are documentation PRs, but they update different docs and cover unrelated model docs content." + }, + { + "left": "pull_request:6525", + "right": "pull_request:6558", + "accept": false, + "reason": "Both add torch.compile compatibility, but for different training scripts and distinct code paths." + }, + { + "left": "pull_request:6525", + "right": "pull_request:6576", + "accept": false, + "reason": "Same broad compile-compat theme, but this is a different InstructPix2Pix SDXL script and not the same fix." + }, + { + "left": "pull_request:6332", + "right": "pull_request:8239", + "accept": false, + "reason": "Both touch examples/community README, but they add different community pipelines with unrelated implementations." + }, + { + "left": "pull_request:8630", + "right": "pull_request:8755", + "accept": false, + "reason": "Both concern SD3 DreamBooth LoRA, but one adds text-encoder training and the other fixes that feature; not the same change." + }, + { + "left": "pull_request:11271", + "right": "pull_request:11578", + "accept": false, + "reason": "Both edit single_file_utils for LTX models, but they target different model versions and different loading failures." + }, + { + "left": "pull_request:9817", + "right": "pull_request:9896", + "accept": false, + "reason": "This is a change versus its explicit revert; they are opposites, not duplicates." + }, + { + "left": "pull_request:7113", + "right": "pull_request:7117", + "accept": false, + "reason": "Both add Slack notifications, but one is for Docker CI and the other for nightly tests; different workflows and purposes." + }, + { + "left": "pull_request:4536", + "right": "pull_request:4910", + "accept": false, + "reason": "Both are inpainting docs, but one adds a full example and the other fixes a typo in that example; not the same underlying change." + }, + { + "left": "pull_request:12856", + "right": "pull_request:12857", + "accept": false, + "reason": "These are related Z-Image implementation PRs, but one adds docs and the other adds the actual pipeline/model code; not one duplicate fix." + }, + { + "left": "pull_request:9579", + "right": "pull_request:9706", + "accept": false, + "reason": "Both are documentation-only PRs, but they cover different modules and unrelated content edits." + }, + { + "left": "pull_request:10259", + "right": "pull_request:10578", + "accept": false, + "reason": "Both touch LoRA loading, but one is for Flux regular LoRAs and the other for 4-bit quantized Flux models; different concrete bugs." + }, + { + "left": "pull_request:456", + "right": "pull_request:533", + "accept": false, + "reason": "This is a type-hint addition and its later revert; opposite changes, not duplicates." + }, + { + "left": "pull_request:8755", + "right": "pull_request:9682", + "accept": false, + "reason": "Both are SD3 DreamBooth LoRA changes, but one is a bug-fix follow-up around text encoder training and the other adds latent caching/small updates." + }, + { + "left": "pull_request:11728", + "right": "pull_request:12451", + "accept": false, + "reason": "Both modify the doc-builder Dockerfile, but they add different packages for different needs and are not the same change." + }, + { + "left": "pull_request:6525", + "right": "pull_request:6577", + "accept": false, + "reason": "Both are torch.compile compatibility fixes, but they apply to different training scripts and code paths." + }, + { + "left": "pull_request:7797", + "right": "pull_request:7898", + "accept": false, + "reason": "Both reduce test model size, but they target different tests and different model architectures." + }, + { + "left": "pull_request:7803", + "right": "pull_request:7898", + "accept": false, + "reason": "Different tests and different fast-test reductions; only broad theme matches." + }, + { + "left": "pull_request:7804", + "right": "pull_request:7898", + "accept": false, + "reason": "These are separate test-size reductions for different pipelines, not one underlying bug." + }, + { + "left": "pull_request:7833", + "right": "pull_request:7898", + "accept": false, + "reason": "Different pipeline test adjustments; the similarity is only that they shrink fast tests." + }, + { + "left": "pull_request:7846", + "right": "pull_request:7898", + "accept": false, + "reason": "Same broad test-optimization theme, but different model/test code and no shared concrete fix." + }, + { + "left": "pull_request:10437", + "right": "pull_request:10508", + "accept": false, + "reason": "Both make tests more device-agnostic, but they are separate test-maintenance PRs with different file sets and changes." + }, + { + "left": "pull_request:11421", + "right": "pull_request:11492", + "accept": false, + "reason": "Both set LANCZOS defaults, but they target different DreamBooth training scripts and are distinct changes." + }, + { + "left": "pull_request:2602", + "right": "pull_request:3633", + "accept": false, + "reason": "Both pin huggingface-hub in setup.py, but they are different version tests with no shared underlying bug." + }, + { + "left": "pull_request:9300", + "right": "pull_request:9346", + "accept": false, + "reason": "Both are CI/test fixes, but they affect different model/test classes and solve different failures." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 3, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10318", + "pull_request:10378", + "pull_request:11395", + "pull_request:11421", + "pull_request:11449", + "pull_request:11471", + "pull_request:11472", + "pull_request:11490", + "pull_request:11491", + "pull_request:11492", + "pull_request:11737", + "pull_request:1488", + "pull_request:3642", + "pull_request:4155", + "pull_request:4188", + "pull_request:4216", + "pull_request:5027", + "pull_request:5407", + "pull_request:5674", + "pull_request:5720", + "pull_request:6035", + "pull_request:6171", + "pull_request:6380", + "pull_request:643", + "pull_request:644", + "pull_request:645", + "pull_request:7233", + "pull_request:7679", + "pull_request:7697", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7833", + "pull_request:7846", + "pull_request:7865", + "pull_request:8675", + "pull_request:9257", + "pull_request:9300", + "pull_request:9353", + "pull_request:9355", + "pull_request:942", + "pull_request:9592", + "pull_request:9598", + "pull_request:9646", + "pull_request:9682", + "pull_request:9714", + "pull_request:9715" + ], + "soft_pairs": [ + "pull_request:9353|pull_request:9598", + "pull_request:5674|pull_request:5720", + "pull_request:7797|pull_request:7865", + "pull_request:7803|pull_request:7865", + "pull_request:7804|pull_request:7865", + "pull_request:7833|pull_request:7865", + "pull_request:7846|pull_request:7865", + "pull_request:11421|pull_request:11449", + "pull_request:11421|pull_request:11491", + "pull_request:9592|pull_request:9714", + "pull_request:3642|pull_request:4216", + "pull_request:9257|pull_request:9646", + "pull_request:643|pull_request:644", + "pull_request:9592|pull_request:9715", + "pull_request:643|pull_request:645", + "pull_request:11472|pull_request:11492", + "pull_request:11395|pull_request:11449", + "pull_request:9646|pull_request:9682", + "pull_request:10318|pull_request:9598", + "pull_request:5027|pull_request:5407", + "pull_request:11471|pull_request:11490", + "pull_request:7679|pull_request:7697", + "pull_request:6035|pull_request:6171", + "pull_request:9300|pull_request:9355", + "pull_request:10378|pull_request:11737", + "pull_request:1488|pull_request:942", + "pull_request:6380|pull_request:8675", + "pull_request:4155|pull_request:4188", + "pull_request:6035|pull_request:7233" + ], + "prepared_review_unit_hash": "00ed665516f2d55fc5d317ed0c22ba7c812c9cd0e4fd2a4b82d7d8b9edcd293a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 98237, + "estimated_input_tokens": 24560, + "estimated_eval_tokens": 49376 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: several unrelated docs, test-size reduction, and training-script interpolation tweaks are mixed together. The only clear duplicate-like subgroup is the textual inversion checkpoint-saving PRs (643/644/645), which are iterations of the same feature change.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 645 is the cleanest representative: it is the finalized merged version of the textual-inversion checkpoint-saving feature, whereas 643/644 are earlier WIP variants.", + "best_issue_reason": null, + "best_pr_reason": "PR 645 is the best overall anchor because it is the completed, merged implementation of the only true duplicate cluster in this set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9353", + "right": "pull_request:9598", + "accept": false, + "reason": "Both touch Flux LoRA tests, but one adds integration coverage and the other only adds a version note; different changes, not the same bugfix." + }, + { + "left": "pull_request:5674", + "right": "pull_request:5720", + "accept": false, + "reason": "Both are docs-only, but they edit different documentation scopes and are not the same underlying change." + }, + { + "left": "pull_request:7797", + "right": "pull_request:7865", + "accept": false, + "reason": "Both reduce fast-test model sizes, but they apply to different pipelines and independent test fixtures." + }, + { + "left": "pull_request:7803", + "right": "pull_request:7865", + "accept": false, + "reason": "Different pipelines and different test updates; same general pattern, but not the same change." + }, + { + "left": "pull_request:7804", + "right": "pull_request:7865", + "accept": false, + "reason": "These are separate fast-test reductions for different Amused/Dance codepaths, not one mergeable fix." + }, + { + "left": "pull_request:7833", + "right": "pull_request:7865", + "accept": false, + "reason": "AudioLDM test shrink vs Dance Diffusion test shrink are distinct changes in different files." + }, + { + "left": "pull_request:7846", + "right": "pull_request:7865", + "accept": false, + "reason": "Different test targets and model configs; not the same concrete fix." + }, + { + "left": "pull_request:11421", + "right": "pull_request:11449", + "accept": false, + "reason": "Both add lanczos interpolation options, but they change different training scripts and are separate feature ports." + }, + { + "left": "pull_request:11421", + "right": "pull_request:11491", + "accept": false, + "reason": "Same broad interpolation theme, but different scripts/codepaths; not one PR-worthy duplicate." + }, + { + "left": "pull_request:9592", + "right": "pull_request:9714", + "accept": false, + "reason": "Both refactor docs/docstrings, but they are for different model files and unrelated edits." + }, + { + "left": "pull_request:3642", + "right": "pull_request:4216", + "accept": false, + "reason": "Same inpaint pipeline area, but one adds the pipeline and the other changes timestep handling; different changes." + }, + { + "left": "pull_request:9257", + "right": "pull_request:9646", + "accept": false, + "reason": "Both involve Dreambooth Flux, but they fix different features: prompt/tokenization alignment vs configurable target modules." + }, + { + "left": "pull_request:643", + "right": "pull_request:644", + "accept": true, + "reason": "Same textual inversion checkpoint-saving feature: both add periodic learned_embeds.bin saving in the same training script, with 644 appearing to refine 643." + }, + { + "left": "pull_request:9592", + "right": "pull_request:9715", + "accept": false, + "reason": "Different docstring refactors in different source files; same style, not same change." + }, + { + "left": "pull_request:643", + "right": "pull_request:645", + "accept": true, + "reason": "Same underlying textual inversion save-checkpoints change; 645 is the finalized version of 643's WIP feature." + }, + { + "left": "pull_request:11472", + "right": "pull_request:11492", + "accept": false, + "reason": "Both add lanczos defaults, but they target different advanced training scripts and are separate ports." + }, + { + "left": "pull_request:11395", + "right": "pull_request:11449", + "accept": false, + "reason": "Same interpolation-default theme, but different training scripts and codepaths." + }, + { + "left": "pull_request:9646", + "right": "pull_request:9682", + "accept": false, + "reason": "Both are Dreambooth/SD3 follow-up changes, but they address different model features and bugfixes." + }, + { + "left": "pull_request:10318", + "right": "pull_request:9598", + "accept": false, + "reason": "One adds Flux LoRA integration tests while the other only annotates test-version requirements; not the same fix." + }, + { + "left": "pull_request:5027", + "right": "pull_request:5407", + "accept": false, + "reason": "Both touch unconditional training, but one fixes mixed precision behavior and the other changes dtype handling for a related example update." + }, + { + "left": "pull_request:11471", + "right": "pull_request:11490", + "accept": false, + "reason": "Same lanczos interpolation theme, but different SDXL/Dreambooth script changes and separate implementations." + }, + { + "left": "pull_request:7679", + "right": "pull_request:7697", + "accept": false, + "reason": "Both adjust PEFT installation in CI, but they modify different workflows and are not the same concrete change." + }, + { + "left": "pull_request:6035", + "right": "pull_request:6171", + "accept": false, + "reason": "Same advanced SDXL script, but one improves help text and the other changes save-model-card behavior; not duplicates." + }, + { + "left": "pull_request:9300", + "right": "pull_request:9355", + "accept": false, + "reason": "One changes fast GPU test behavior and the other adds workflow_dispatch; unrelated CI edits." + }, + { + "left": "pull_request:10378", + "right": "pull_request:11737", + "accept": false, + "reason": "Both are training-script DS-support changes, but for different models and different save hooks." + }, + { + "left": "pull_request:1488", + "right": "pull_request:942", + "accept": false, + "reason": "Both address MPS dtype issues, but they patch different schedulers/models and are separate codepaths." + }, + { + "left": "pull_request:6380", + "right": "pull_request:8675", + "accept": false, + "reason": "Both edit the fast-diffusion doc, but they add different notes and are not the same change." + }, + { + "left": "pull_request:4155", + "right": "pull_request:4188", + "accept": false, + "reason": "Same SDXL ControlNet area, but one is a pipeline follow-up and the other is a training-cache fingerprint fix." + }, + { + "left": "pull_request:6035", + "right": "pull_request:7233", + "accept": false, + "reason": "Different changes to the same advanced SDXL script: help text vs adding DoRA support." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative on the soft-edge pairs. The only clearly duplicate-like subgroup is the textual inversion checkpoint-saving trio, and the accepted pairs there are defensible as mergeable variants of the same feature/fix. The other verdicts appropriately treat similar-but-distinct docs, test reductions, interpolation ports, and training-script changes as non-duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 4, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10288", + "pull_request:10318", + "pull_request:10921", + "pull_request:10922", + "pull_request:10991", + "pull_request:11025", + "pull_request:11449", + "pull_request:11455", + "pull_request:11463", + "pull_request:11465", + "pull_request:11471", + "pull_request:11588", + "pull_request:11670", + "pull_request:11776", + "pull_request:12086", + "pull_request:12269", + "pull_request:2677", + "pull_request:2734", + "pull_request:3169", + "pull_request:3419", + "pull_request:3472", + "pull_request:3642", + "pull_request:377", + "pull_request:388", + "pull_request:5720", + "pull_request:5744", + "pull_request:6021", + "pull_request:6035", + "pull_request:6464", + "pull_request:6550", + "pull_request:6555", + "pull_request:6599", + "pull_request:676", + "pull_request:677", + "pull_request:7182", + "pull_request:7235", + "pull_request:8230", + "pull_request:8773", + "pull_request:8774", + "pull_request:8775", + "pull_request:8976", + "pull_request:8986", + "pull_request:9005", + "pull_request:9207", + "pull_request:9353", + "pull_request:9593", + "pull_request:9707", + "pull_request:9775" + ], + "soft_pairs": [ + "pull_request:8773|pull_request:8774", + "pull_request:10318|pull_request:9353", + "pull_request:6035|pull_request:7235", + "pull_request:11670|pull_request:11776", + "pull_request:8774|pull_request:8976", + "pull_request:6464|pull_request:6599", + "pull_request:11449|pull_request:11455", + "pull_request:11449|pull_request:11465", + "pull_request:3419|pull_request:3642", + "pull_request:5720|pull_request:5744", + "pull_request:11025|pull_request:11588", + "pull_request:12086|pull_request:12269", + "pull_request:10288|pull_request:10991", + "pull_request:11463|pull_request:11471", + "pull_request:2677|pull_request:3472", + "pull_request:2734|pull_request:3169", + "pull_request:377|pull_request:388", + "pull_request:6021|pull_request:7182", + "pull_request:8230|pull_request:9207", + "pull_request:8775|pull_request:9775", + "pull_request:8986|pull_request:9005", + "pull_request:9593|pull_request:9707", + "pull_request:10921|pull_request:10922", + "pull_request:676|pull_request:677", + "pull_request:6550|pull_request:6555" + ], + "prepared_review_unit_hash": "2672dc6f0a41dca0e995494859a1c64e75991e9e1aef54bf9f48820bc3cd41dd", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 110163, + "estimated_input_tokens": 27541, + "estimated_eval_tokens": 55338 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mix of unrelated PRs that only share broad themes or filenames. The only clear duplicate-style pair is the Wan 2.1 feature PR and its follow-up continuation (#10921/#10922).", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "#10922 is the merged, completed follow-up to the WIP Wan 2.1 support work in #10921, so it is the best canonical PR for the only true duplicate-style pair here.", + "best_issue_reason": null, + "best_pr_reason": "#10922 is the strongest representative PR overall: merged, complete, and the natural continuation of the same Wan 2.1 support change started in #10921.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8773", + "right": "pull_request:8774", + "accept": false, + "reason": "Same feature area, but one is a revert and the other introduces the mixin; they are opposite changes, not duplicates." + }, + { + "left": "pull_request:10318", + "right": "pull_request:9353", + "accept": false, + "reason": "Both add Flux LoRA tests, but they cover different test scenarios and are not the same underlying change." + }, + { + "left": "pull_request:6035", + "right": "pull_request:7235", + "accept": false, + "reason": "Help-text cleanup versus adding DoRA training support; different changes in the same script." + }, + { + "left": "pull_request:11670", + "right": "pull_request:11776", + "accept": false, + "reason": "Both are torch.compile-related tests, but they target different failure modes and are not the same code-path fix." + }, + { + "left": "pull_request:8774", + "right": "pull_request:8976", + "accept": false, + "reason": "Same mixin story, but one is a revert of the prior change and the other is the reintroduction/rework; not a duplicate change." + }, + { + "left": "pull_request:6464", + "right": "pull_request:6599", + "accept": false, + "reason": "Both touch the same advanced DreamBooth LoRA script, but they fix different bugs." + }, + { + "left": "pull_request:11449", + "right": "pull_request:11455", + "accept": false, + "reason": "Same interpolation-mode idea, but applied to different training scripts; not the same concrete PR." + }, + { + "left": "pull_request:11449", + "right": "pull_request:11465", + "accept": false, + "reason": "Same as above: same pattern, different script and code path." + }, + { + "left": "pull_request:3419", + "right": "pull_request:3642", + "accept": false, + "reason": "Different community pipelines (img2img vs inpainting), so not the same underlying change." + }, + { + "left": "pull_request:5720", + "right": "pull_request:5744", + "accept": false, + "reason": "Both are docs cleanup PRs, but they update different pages and are not the same change." + }, + { + "left": "pull_request:11025", + "right": "pull_request:11588", + "accept": false, + "reason": "Both concern Wan LoRA support, but they implement different mechanisms in different parts of the LoRA pipeline." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12269", + "accept": false, + "reason": "Both are modular Flux work, but one adds I2I support and the other adds Flux Kontext support; not mergeable as one change." + }, + { + "left": "pull_request:10288", + "right": "pull_request:10991", + "accept": false, + "reason": "Single-file support for different models (Flux vs Wan); same area, different feature." + }, + { + "left": "pull_request:11463", + "right": "pull_request:11471", + "accept": false, + "reason": "Both set LANCZOS defaults, but in different training scripts and with different surrounding fixes." + }, + { + "left": "pull_request:2677", + "right": "pull_request:3472", + "accept": false, + "reason": "Both add docs entries, but for unrelated training topics and pages." + }, + { + "left": "pull_request:2734", + "right": "pull_request:3169", + "accept": false, + "reason": "Both edit Habana docs, but they update different version/benchmark details rather than the same change." + }, + { + "left": "pull_request:377", + "right": "pull_request:388", + "accept": false, + "reason": "Related latent diffusion docs, but one is conditional and the other unconditional; different pages and artifacts." + }, + { + "left": "pull_request:6021", + "right": "pull_request:7182", + "accept": false, + "reason": "Different enhancements to the same script: one adds a variant arg, the other adds EDM-style training." + }, + { + "left": "pull_request:8230", + "right": "pull_request:9207", + "accept": false, + "reason": "Both discuss CPU offloading usage, but they update different docs/examples and are not the same bug fix." + }, + { + "left": "pull_request:8775", + "right": "pull_request:9775", + "accept": false, + "reason": "Unrelated README/doc updates; no shared underlying change." + }, + { + "left": "pull_request:8986", + "right": "pull_request:9005", + "accept": false, + "reason": "Both are CI runner/workflow updates, but they change different workflow configurations." + }, + { + "left": "pull_request:9593", + "right": "pull_request:9707", + "accept": false, + "reason": "Docstring refactors in different files; not the same underlying change." + }, + { + "left": "pull_request:10921", + "right": "pull_request:10922", + "accept": true, + "reason": "Direct continuation of the same Wan 2.1 support work; #10922 completes the WIP implementation from #10921 and could plausibly be one PR." + }, + { + "left": "pull_request:676", + "right": "pull_request:677", + "accept": false, + "reason": "Both are resnet.py cleanups, but they are separate refactors rather than the same concrete bug or feature change." + }, + { + "left": "pull_request:6550", + "right": "pull_request:6555", + "accept": false, + "reason": "Both address torch.compile compatibility, but for different training scripts and different issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the soft-edge verdicts stay conservative overall. The only accepted pair, #10921/#10922, does look like a split implementation of the same Wan 2.1 support work that could plausibly be merged into one PR. No soft-edge reason overstates duplicate evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 5, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10288", + "pull_request:10298", + "pull_request:10437", + "pull_request:10508", + "pull_request:10781", + "pull_request:11058", + "pull_request:11516", + "pull_request:12095", + "pull_request:12170", + "pull_request:12236", + "pull_request:12272", + "pull_request:12439", + "pull_request:12566", + "pull_request:12790", + "pull_request:1846", + "pull_request:1927", + "pull_request:2479", + "pull_request:2487", + "pull_request:2588", + "pull_request:2677", + "pull_request:3297", + "pull_request:3707", + "pull_request:3711", + "pull_request:3979", + "pull_request:4188", + "pull_request:4223", + "pull_request:4237", + "pull_request:4238", + "pull_request:424", + "pull_request:4481", + "pull_request:4946", + "pull_request:509", + "pull_request:5097", + "pull_request:5172", + "pull_request:5914", + "pull_request:5979", + "pull_request:6021", + "pull_request:6035", + "pull_request:6528", + "pull_request:6566", + "pull_request:6884", + "pull_request:7166", + "pull_request:7182", + "pull_request:7189", + "pull_request:7413", + "pull_request:8926", + "pull_request:9083", + "pull_request:9400" + ], + "soft_pairs": [ + "pull_request:10508|pull_request:11058", + "pull_request:2479|pull_request:2588", + "pull_request:424|pull_request:509", + "pull_request:6884|pull_request:7189", + "pull_request:11516|pull_request:12095", + "pull_request:2677|pull_request:4946", + "pull_request:3979|pull_request:5914", + "pull_request:4481|pull_request:5979", + "pull_request:5097|pull_request:5172", + "pull_request:10437|pull_request:9400", + "pull_request:4188|pull_request:4223", + "pull_request:6035|pull_request:7182", + "pull_request:12170|pull_request:12790", + "pull_request:3707|pull_request:3711", + "pull_request:4237|pull_request:4238", + "pull_request:6021|pull_request:6035", + "pull_request:12236|pull_request:12439", + "pull_request:1846|pull_request:3297", + "pull_request:10288|pull_request:10781", + "pull_request:1927|pull_request:2487", + "pull_request:6035|pull_request:6566", + "pull_request:6528|pull_request:7413", + "pull_request:7166|pull_request:8926", + "pull_request:10298|pull_request:9083", + "pull_request:12272|pull_request:12566", + "pull_request:5914|pull_request:6035" + ], + "prepared_review_unit_hash": "8a3be2f0b8e8f3ce28e97bde7978487d8b746f2f97bddf0fafc8c5ccee72e730", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 110535, + "estimated_input_tokens": 27634, + "estimated_eval_tokens": 55524 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft-edge candidates are mostly unrelated incremental PRs within the same repo areas, not duplicate fixes. I would reject all pairings; several are even opposite changes or different model/version releases rather than the same underlying bug/change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10508", + "right": "pull_request:11058", + "accept": false, + "reason": "Both are test/device-agnosticity work, but one targets XPU test support while the other changes textual inversion offload device handling; different bugs and code paths." + }, + { + "left": "pull_request:2479", + "right": "pull_request:2588", + "accept": false, + "reason": "One adds LoRA support for the CLIP text encoder in a training script; the other adds a research-project LoRA example and README. Not the same change." + }, + { + "left": "pull_request:424", + "right": "pull_request:509", + "accept": false, + "reason": "One tweaks a reproducibility assertion and dataset loading in tests; the other adds a shared image-loading helper. Related area, but not the same bug." + }, + { + "left": "pull_request:6884", + "right": "pull_request:7189", + "accept": false, + "reason": "Both touch IP-Adapter behavior, but one validates scale length while the other fixes embed repetition for plus adapters. Separate issues." + }, + { + "left": "pull_request:11516", + "right": "pull_request:12095", + "accept": false, + "reason": "These are different LTX model releases/feature sets (0.9.7 vs 0.9.8), not duplicate fixes." + }, + { + "left": "pull_request:2677", + "right": "pull_request:4946", + "accept": false, + "reason": "Both are docs updates, but one adds ControlNet training docs and the other adds T2I-Adapter overview entries. Different documentation topics." + }, + { + "left": "pull_request:3979", + "right": "pull_request:5914", + "accept": false, + "reason": "One is an SDXL inference bug fix; the other is a DreamBooth LoRA README template cleanup. Not the same underlying problem." + }, + { + "left": "pull_request:4481", + "right": "pull_request:5979", + "accept": false, + "reason": "Different model families entirely: EDM pipeline/scheduler vs DDGAN pipeline/scheduler. Do not merge as duplicates." + }, + { + "left": "pull_request:5097", + "right": "pull_request:5172", + "accept": false, + "reason": "Both adjust tests, but one is float16 fast-test tolerances and the other is LoRA/CUDA test fixes. No shared concrete bug." + }, + { + "left": "pull_request:10437", + "right": "pull_request:9400", + "accept": false, + "reason": "One makes tests XPU-aware; the other updates single-file pipeline tests to pass an explicit device. Different test infrastructure changes." + }, + { + "left": "pull_request:4188", + "right": "pull_request:4223", + "accept": false, + "reason": "Same training script, but one fixes dataset fingerprinting during embedding caching and the other adjusts prompt/auth handling. Distinct fixes." + }, + { + "left": "pull_request:6035", + "right": "pull_request:7182", + "accept": false, + "reason": "One improves help text for custom captions; the other adds EDM-style training support. Different scope and intent." + }, + { + "left": "pull_request:12170", + "right": "pull_request:12790", + "accept": false, + "reason": "Both are LoRA conversion changes, but for different model ecosystems and key-normalization problems (Qwen vs ZImage). Not the same code-path issue." + }, + { + "left": "pull_request:3707", + "right": "pull_request:3711", + "accept": false, + "reason": "The second PR reverts the first; they are opposite changes, not duplicates." + }, + { + "left": "pull_request:4237", + "right": "pull_request:4238", + "accept": false, + "reason": "Both touch ControlNet SDXL, but one fixes prompt embedding dtype and the other fixes CPU offload sequencing. Separate bugs." + }, + { + "left": "pull_request:6021", + "right": "pull_request:6035", + "accept": false, + "reason": "Same file, but one adds a variant argument while the other improves help text for custom captions. Not the same change." + }, + { + "left": "pull_request:12236", + "right": "pull_request:12439", + "accept": false, + "reason": "One adds FlashAttention-3 hub-kernel support; the other adds sage attention plus FA2 support through kernels. Different backend integrations." + }, + { + "left": "pull_request:1846", + "right": "pull_request:3297", + "accept": false, + "reason": "The first is a WIP retriever/RDM architecture PR; the second is a concrete RDM example pipeline. Different maturity and purpose." + }, + { + "left": "pull_request:10288", + "right": "pull_request:10781", + "accept": false, + "reason": "Different model support: Flux single-file support vs Lumina single-file/GGUF support. Not the same underlying feature." + }, + { + "left": "pull_request:1927", + "right": "pull_request:2487", + "accept": false, + "reason": "Both are docs PRs, but one imports a Stable Diffusion guide and the other adds a JAX/Flax guide. Unrelated content." + }, + { + "left": "pull_request:6035", + "right": "pull_request:6566", + "accept": false, + "reason": "Same advanced DreamBooth file, but one is help-tag/documentation cleanup and the other fixes training resume state handling. Different bugs." + }, + { + "left": "pull_request:6528", + "right": "pull_request:7413", + "accept": false, + "reason": "Both mention IP-Adapter, but one fixes ControlNet img2img missing code and the other updates an AnimateDiff community pipeline. Different code paths." + }, + { + "left": "pull_request:7166", + "right": "pull_request:8926", + "accept": false, + "reason": "Both are DreamBooth LoRA SDXL fixes, but one corrects loading state and the other fixes checkpoint saving. Not the same bug." + }, + { + "left": "pull_request:10298", + "right": "pull_request:9083", + "accept": false, + "reason": "One adds LTX GGUF/single-file support; the other adds Flux transformer single-file support. Different models and loaders." + }, + { + "left": "pull_request:12272", + "right": "pull_request:12566", + "accept": false, + "reason": "One fixes Flux modular pipeline height/width propagation; the other adds tests for Flux modular pipelines. Test coverage is not the same fix." + }, + { + "left": "pull_request:5914", + "right": "pull_request:6035", + "accept": false, + "reason": "One fixes a DreamBooth SDXL README template; the other improves advanced DreamBooth help text. Different documentation tweaks, not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 6, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10179", + "pull_request:10199", + "pull_request:10298", + "pull_request:10437", + "pull_request:10508", + "pull_request:10570", + "pull_request:10985", + "pull_request:11292", + "pull_request:1194", + "pull_request:12737", + "pull_request:12909", + "pull_request:1307", + "pull_request:3633", + "pull_request:3795", + "pull_request:3979", + "pull_request:4607", + "pull_request:5052", + "pull_request:5175", + "pull_request:5506", + "pull_request:5508", + "pull_request:5608", + "pull_request:5906", + "pull_request:6021", + "pull_request:6119", + "pull_request:6279", + "pull_request:6402", + "pull_request:6511", + "pull_request:6523", + "pull_request:6526", + "pull_request:6556", + "pull_request:6588", + "pull_request:6589", + "pull_request:6757", + "pull_request:6795", + "pull_request:6816", + "pull_request:6848", + "pull_request:7018", + "pull_request:7124", + "pull_request:7235", + "pull_request:7258", + "pull_request:8819", + "pull_request:9255", + "pull_request:9355", + "pull_request:9400", + "pull_request:9401", + "pull_request:9593", + "pull_request:9714", + "pull_request:9715" + ], + "soft_pairs": [ + "pull_request:6119|pull_request:7018", + "pull_request:10985|pull_request:11292", + "pull_request:10179|pull_request:10199", + "pull_request:10570|pull_request:9255", + "pull_request:1194|pull_request:1307", + "pull_request:3633|pull_request:5175", + "pull_request:3979|pull_request:5906", + "pull_request:5052|pull_request:6556", + "pull_request:6523|pull_request:6848", + "pull_request:6526|pull_request:6589", + "pull_request:9593|pull_request:9714", + "pull_request:6119|pull_request:6402", + "pull_request:6511|pull_request:6526", + "pull_request:12737|pull_request:12909", + "pull_request:9593|pull_request:9715", + "pull_request:10437|pull_request:9401", + "pull_request:10508|pull_request:9400", + "pull_request:10570|pull_request:9355", + "pull_request:3795|pull_request:4607", + "pull_request:6021|pull_request:6795", + "pull_request:6279|pull_request:6523", + "pull_request:6588|pull_request:6757", + "pull_request:7124|pull_request:7258", + "pull_request:9400|pull_request:9401", + "pull_request:5506|pull_request:5608", + "pull_request:6816|pull_request:7235", + "pull_request:5508|pull_request:6021", + "pull_request:10298|pull_request:8819" + ], + "prepared_review_unit_hash": "598ad5a08d02bef328e9b8ededf141a3de6573489ee4a12c49f97b6a317de080", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 100842, + "estimated_input_tokens": 25211, + "estimated_eval_tokens": 50678 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a grab bag of unrelated docs, CI, test, loader, and training changes. The only close duplicate-style pattern in the packet is the AuraFlow docstring refactor pair 9714/9715, but that exact pair is not one of the provided soft-edge candidates; the 9593 links are a different hd_painter docstring cleanup, so all provided soft edges should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR; the cluster is heterogeneous, and the likely duplicate-style AuraFlow pair (9714/9715) is not among the supplied soft-edge candidates.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR. The packet\u2019s closest duplicate-style match is AuraFlow 9714/9715, but the candidate edges only involve 9593, which is a separate hd_painter docstring refactor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6119", + "right": "pull_request:7018", + "accept": false, + "reason": "Both are LoRA/training fixes, but they address different bugs in different scripts and are not the same change." + }, + { + "left": "pull_request:10985", + "right": "pull_request:11292", + "accept": false, + "reason": "Same loader utility, but one adds Flux/ComfyUI LoRA support and the other fixes SDXL SGM-vs-UNet key handling; different code-path problems." + }, + { + "left": "pull_request:10179", + "right": "pull_request:10199", + "accept": false, + "reason": "One is doc/example cleanup, the other adds eta decay behavior in the RF inversion pipeline; not the same fix." + }, + { + "left": "pull_request:10570", + "right": "pull_request:9255", + "accept": false, + "reason": "Both touch the workflow file, but one changes a secret and the other changes branch triggers; unrelated CI edits." + }, + { + "left": "pull_request:1194", + "right": "pull_request:1307", + "accept": false, + "reason": "This is a docs-workflow update versus its revert, not the same underlying change." + }, + { + "left": "pull_request:3633", + "right": "pull_request:5175", + "accept": false, + "reason": "Different concerns: dependency pinning in setup.py versus setting HF_HOME in CI." + }, + { + "left": "pull_request:3979", + "right": "pull_request:5906", + "accept": false, + "reason": "Different files and bugs: SDXL attention logic versus a DreamBooth README/template doc fix." + }, + { + "left": "pull_request:5052", + "right": "pull_request:6556", + "accept": false, + "reason": "Both are training scripts, but for different model families and different feature changes; not mergeable as one PR." + }, + { + "left": "pull_request:6523", + "right": "pull_request:6848", + "accept": false, + "reason": "Same LCM distillation area, but one fixes guidance embeddings and the other fixes target U-Net initialization." + }, + { + "left": "pull_request:6526", + "right": "pull_request:6589", + "accept": false, + "reason": "Both are torch.compile-related, but they fix different scripts and different failure points." + }, + { + "left": "pull_request:9593", + "right": "pull_request:9714", + "accept": false, + "reason": "Different docstring refactors in different files; 9593 is hd_painter, while 9714 is AuraFlow." + }, + { + "left": "pull_request:6119", + "right": "pull_request:6402", + "accept": false, + "reason": "Both are LoRA training script changes, but they fix different bugs: fp16 unscale versus ignored gradient checkpointing." + }, + { + "left": "pull_request:6511", + "right": "pull_request:6526", + "accept": false, + "reason": "Same broad torch.compile theme, but different training scripts and different bugs." + }, + { + "left": "pull_request:12737", + "right": "pull_request:12909", + "accept": false, + "reason": "Both mention parallelism, but they fix different model code paths in different transformers." + }, + { + "left": "pull_request:9593", + "right": "pull_request:9715", + "accept": false, + "reason": "Different docstring cleanup PRs in different files; 9593 is unrelated to the AuraFlow transformer refactor." + }, + { + "left": "pull_request:10437", + "right": "pull_request:9401", + "accept": false, + "reason": "Both are device-agnostic test work, but they target different test suites and code paths." + }, + { + "left": "pull_request:10508", + "right": "pull_request:9400", + "accept": false, + "reason": "Related testing cleanup, but one is accelerator-aware model tests and the other is single-file pipeline tests." + }, + { + "left": "pull_request:10570", + "right": "pull_request:9355", + "accept": false, + "reason": "Same workflow file, but one updates a token secret and the other adds workflow_dispatch; unrelated changes." + }, + { + "left": "pull_request:3795", + "right": "pull_request:4607", + "accept": false, + "reason": "Both are docs PRs, but they add different overview entries and touch different features." + }, + { + "left": "pull_request:6021", + "right": "pull_request:6795", + "accept": false, + "reason": "Different DreamBooth/SDXL changes: variant support versus micro-conditioning." + }, + { + "left": "pull_request:6279", + "right": "pull_request:6523", + "accept": false, + "reason": "Same LCM distillation scripts, but they fix different bugs in different parts of the training flow." + }, + { + "left": "pull_request:6588", + "right": "pull_request:6757", + "accept": false, + "reason": "One updates a slow-test expectation; the other changes the pipeline implementation to move sigmas to device." + }, + { + "left": "pull_request:7124", + "right": "pull_request:7258", + "accept": false, + "reason": "Different fixes in the same script: a validation kwarg versus model-card/save-card text changes." + }, + { + "left": "pull_request:9400", + "right": "pull_request:9401", + "accept": false, + "reason": "Both are device-agnostic test efforts, but they cover different test scopes and files." + }, + { + "left": "pull_request:5506", + "right": "pull_request:5608", + "accept": false, + "reason": "Different LoRA/PEFT bugs: xformers attention handling versus empty network-alpha dict handling." + }, + { + "left": "pull_request:6816", + "right": "pull_request:7235", + "accept": false, + "reason": "Different SDXL DreamBooth changes: DDP+PEFT compatibility versus DoRA training support." + }, + { + "left": "pull_request:5508", + "right": "pull_request:6021", + "accept": false, + "reason": "Different SDXL training-script additions: new training features versus variant support." + }, + { + "left": "pull_request:10298", + "right": "pull_request:8819", + "accept": false, + "reason": "Both add from_single_file support, but for different model families and loaders, so they are not the same change." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary is not fully grounded in the packet: it misidentifies the only duplicate-style pair as AuraFlow docstring refactors, but the packet shows the AuraFlow pair is 9714/9715, while 9593 is a different hd_painter docstring PR. The soft-edge verdict `9593` vs `9715` is therefore overstated and not conservative. Because of that incorrect accept, I can\u2019t accept the evaluation." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 7, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10183", + "pull_request:10268", + "pull_request:10298", + "pull_request:11358", + "pull_request:11426", + "pull_request:11427", + "pull_request:11463", + "pull_request:11472", + "pull_request:11747", + "pull_request:12407", + "pull_request:1418", + "pull_request:1682", + "pull_request:1898", + "pull_request:2154", + "pull_request:2336", + "pull_request:288", + "pull_request:359", + "pull_request:3642", + "pull_request:3945", + "pull_request:4596", + "pull_request:4946", + "pull_request:4947", + "pull_request:5052", + "pull_request:6035", + "pull_request:6511", + "pull_request:6525", + "pull_request:6532", + "pull_request:6555", + "pull_request:6589", + "pull_request:6795", + "pull_request:6851", + "pull_request:6875", + "pull_request:6960", + "pull_request:7233", + "pull_request:7235", + "pull_request:7260", + "pull_request:7467", + "pull_request:7984", + "pull_request:7988", + "pull_request:8725", + "pull_request:8810", + "pull_request:8864", + "pull_request:8926", + "pull_request:912", + "pull_request:9463", + "pull_request:9917", + "pull_request:9993" + ], + "soft_pairs": [ + "pull_request:6525|pull_request:6589", + "pull_request:10183|pull_request:10298", + "pull_request:11426|pull_request:12407", + "pull_request:11427|pull_request:12407", + "pull_request:11463|pull_request:11472", + "pull_request:1418|pull_request:2336", + "pull_request:1898|pull_request:2154", + "pull_request:3945|pull_request:6035", + "pull_request:4946|pull_request:4947", + "pull_request:6035|pull_request:7467", + "pull_request:6875|pull_request:6960", + "pull_request:8810|pull_request:9917", + "pull_request:8864|pull_request:9917", + "pull_request:8725|pull_request:9463", + "pull_request:6795|pull_request:7233", + "pull_request:7235|pull_request:8926", + "pull_request:8926|pull_request:9993", + "pull_request:6851|pull_request:7260", + "pull_request:6511|pull_request:6532", + "pull_request:6795|pull_request:7235", + "pull_request:10268|pull_request:10298", + "pull_request:11358|pull_request:11747", + "pull_request:1682|pull_request:912", + "pull_request:288|pull_request:359", + "pull_request:3642|pull_request:4596", + "pull_request:5052|pull_request:6555", + "pull_request:6525|pull_request:7984", + "pull_request:6525|pull_request:7988" + ], + "prepared_review_unit_hash": "3437586443e45bc8e0feebe029e6a07ae902f0da10522a2ad6c1f18405788be9", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 101135, + "estimated_input_tokens": 25284, + "estimated_eval_tokens": 50824 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a noisy thematic cluster with several unrelated PRs. I only see two true duplicate-style pairs: the HiDream license header fix and the PAG ControlNet Img2Img pipeline addition. Everything else is the same broad area but a different script, model, or follow-up fix, so I rejected them.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9917 is the clearest representative of the one concrete duplicate-like feature subgroup (PAG ControlNet Img2Img); the rest of the set is too heterogeneous for a single cluster-wide canonical.", + "best_issue_reason": null, + "best_pr_reason": "PR 9917 is the best single PR representative among the near-duplicate candidates because it names the exact pipeline and is the most directly aligned with a concrete code-path change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6525", + "right": "pull_request:6589", + "accept": false, + "reason": "Same general torch.compile theme, but different training scripts and save hooks; not the same concrete fix." + }, + { + "left": "pull_request:10183", + "right": "pull_request:10298", + "accept": false, + "reason": "Both are single-file support PRs, but for different models (AutoencoderDC vs Mochi Transformer)." + }, + { + "left": "pull_request:11426", + "right": "pull_request:12407", + "accept": false, + "reason": "Different scripts and goals: interpolation-mode follow-up vs UV compatibility docs/packaging update." + }, + { + "left": "pull_request:11427", + "right": "pull_request:12407", + "accept": false, + "reason": "Different scripts and goals: interpolation-mode follow-up vs UV compatibility docs/packaging update." + }, + { + "left": "pull_request:11463", + "right": "pull_request:11472", + "accept": false, + "reason": "Same interpolation idea, but different training scripts and model families; not one duplicate bug/change." + }, + { + "left": "pull_request:1418", + "right": "pull_request:2336", + "accept": false, + "reason": "Both are Versatile Diffusion test fixes, but they touch different test data/content and are not the same change." + }, + { + "left": "pull_request:1898", + "right": "pull_request:2154", + "accept": false, + "reason": "2154 is a typo fix within the env-version feature added by 1898, not the same underlying PR/change." + }, + { + "left": "pull_request:3945", + "right": "pull_request:6035", + "accept": false, + "reason": "Different Dreambooth LoRA changes in the same script area, but separate features/help text updates." + }, + { + "left": "pull_request:4946", + "right": "pull_request:4947", + "accept": false, + "reason": "One updates docs and the other adds tests; not the same concrete bug or code-path change." + }, + { + "left": "pull_request:6035", + "right": "pull_request:7467", + "accept": false, + "reason": "Different training-script enhancements (help text vs tag shuffling/drop) in the same subsystem." + }, + { + "left": "pull_request:6875", + "right": "pull_request:6960", + "accept": false, + "reason": "Both are IP-Adapter test edits, but one adjusts expected values and the other marks flakiness; not duplicates." + }, + { + "left": "pull_request:8810", + "right": "pull_request:9917", + "accept": true, + "reason": "Same concrete feature: adding StableDiffusionControlNetPAGImg2ImgPipeline support; 9917 is essentially the same pipeline addition in a newer/broader PR." + }, + { + "left": "pull_request:8864", + "right": "pull_request:9917", + "accept": false, + "reason": "Related PAG ControlNet Img2Img work, but 8864 is a broader/open variant and not the same exact PR change as 9917." + }, + { + "left": "pull_request:8725", + "right": "pull_request:9463", + "accept": false, + "reason": "Both add PAG support, but for different pipeline variants (text-to-image vs Img2Img)." + }, + { + "left": "pull_request:6795", + "right": "pull_request:7233", + "accept": false, + "reason": "Same Dreambooth LoRA SDXL script, but one adds micro-conditioning and the other adds DoRA; different features." + }, + { + "left": "pull_request:7235", + "right": "pull_request:8926", + "accept": false, + "reason": "Different Dreambooth LoRA scripts and different fixes: DoRA feature vs checkpoint-save bug." + }, + { + "left": "pull_request:8926", + "right": "pull_request:9993", + "accept": false, + "reason": "Different models/scripts (SDXL advanced vs SD3) and different checkpoint-save logic." + }, + { + "left": "pull_request:6851", + "right": "pull_request:7260", + "accept": false, + "reason": "Different pipelines (PIA vs SAG) and different adapter integration work." + }, + { + "left": "pull_request:6511", + "right": "pull_request:6532", + "accept": false, + "reason": "Both address torch.compile compatibility, but in different training scripts and code paths." + }, + { + "left": "pull_request:6795", + "right": "pull_request:7235", + "accept": false, + "reason": "Same Dreambooth LoRA SDXL script, but different features: micro-conditioning vs DoRA." + }, + { + "left": "pull_request:10268", + "right": "pull_request:10298", + "accept": false, + "reason": "Both extend single-file loading, but for different model families and conversion paths." + }, + { + "left": "pull_request:11358", + "right": "pull_request:11747", + "accept": true, + "reason": "Same HiDream license-header fix in the same files; 11358 is the broader ancestor, 11747 is the focused duplicate-style follow-up." + }, + { + "left": "pull_request:1682", + "right": "pull_request:912", + "accept": false, + "reason": "Both are version bumps, but for different release versions and different eras." + }, + { + "left": "pull_request:288", + "right": "pull_request:359", + "accept": false, + "reason": "Same Karras VE area, but one is pipeline typing and the other is scheduler typing; not the same change." + }, + { + "left": "pull_request:3642", + "right": "pull_request:4596", + "accept": false, + "reason": "Completely different functionality: TensorRT inpainting pipeline vs SDXL conversion script." + }, + { + "left": "pull_request:5052", + "right": "pull_request:6555", + "accept": false, + "reason": "Different training families: Wuerstchen prior/text-to-image vs SD text-to-image LoRA compile compatibility." + }, + { + "left": "pull_request:6525", + "right": "pull_request:7984", + "accept": false, + "reason": "ControlNet compile fix vs VAE compile fix; related theme but different code paths." + }, + { + "left": "pull_request:6525", + "right": "pull_request:7988", + "accept": false, + "reason": "ControlNet compile fix vs IP-Adapter compile fix; different model components and code paths." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: the accepted pairs both share the same concrete pipeline/license fix, while the rejected pairs are described as related but distinct changes. I don\u2019t see an overstatement of evidence or an obviously non-mergeable accepted pair." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 8, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10077", + "pull_request:10298", + "pull_request:10633", + "pull_request:10991", + "pull_request:11271", + "pull_request:11550", + "pull_request:12215", + "pull_request:12325", + "pull_request:288", + "pull_request:325", + "pull_request:3642", + "pull_request:3952", + "pull_request:4595", + "pull_request:5057", + "pull_request:5175", + "pull_request:5517", + "pull_request:5914", + "pull_request:5938", + "pull_request:6035", + "pull_request:6303", + "pull_request:644", + "pull_request:645", + "pull_request:6511", + "pull_request:6534", + "pull_request:6556", + "pull_request:6566", + "pull_request:6691", + "pull_request:6816", + "pull_request:6968", + "pull_request:7233", + "pull_request:7258", + "pull_request:754", + "pull_request:7580", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7833", + "pull_request:7846", + "pull_request:869", + "pull_request:8755", + "pull_request:8770", + "pull_request:8775", + "pull_request:8926", + "pull_request:8973", + "pull_request:8983", + "pull_request:8987", + "pull_request:9483", + "pull_request:9787" + ], + "soft_pairs": [ + "pull_request:754|pull_request:869", + "pull_request:5517|pull_request:6303", + "pull_request:8973|pull_request:8987", + "pull_request:6691|pull_request:6968", + "pull_request:6816|pull_request:7233", + "pull_request:10633|pull_request:9787", + "pull_request:10298|pull_request:11550", + "pull_request:6511|pull_request:6534", + "pull_request:644|pull_request:645", + "pull_request:5057|pull_request:5938", + "pull_request:12215|pull_request:12325", + "pull_request:10077|pull_request:11271", + "pull_request:10298|pull_request:10991", + "pull_request:288|pull_request:325", + "pull_request:3642|pull_request:4595", + "pull_request:3952|pull_request:5175", + "pull_request:5914|pull_request:6556", + "pull_request:6035|pull_request:7258", + "pull_request:6566|pull_request:8755", + "pull_request:7580|pull_request:7797", + "pull_request:7580|pull_request:7803", + "pull_request:7580|pull_request:7804", + "pull_request:7580|pull_request:7833", + "pull_request:7580|pull_request:7846", + "pull_request:8770|pull_request:8775", + "pull_request:8983|pull_request:9483", + "pull_request:7233|pull_request:8926" + ], + "prepared_review_unit_hash": "9cf7625171687ad55045c95bc909e02ebfc7e29a7dd4e2af269ef44fb825a356", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 97579, + "estimated_input_tokens": 24395, + "estimated_eval_tokens": 49046 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Only one soft pair looks like the same underlying change: the textual inversion progress-saving WIP and its finalized merged follow-up. The rest are different model/features, different pipelines, or unrelated docs/test/CI edits.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 645 is the finalized merged version of the textual inversion save-progress feature; PR 644 is a WIP precursor with the same functionality.", + "best_issue_reason": null, + "best_pr_reason": "PR 645 is the cleanest canonical choice: merged, narrower, and it subsumes the WIP in PR 644 without extra unrelated churn.", + "soft_edge_verdicts": [ + { + "left": "pull_request:754", + "right": "pull_request:869", + "accept": false, + "reason": "Both touch community-pipeline import handling, but the diffs/titles suggest separate cleanup fixes plus README changes, not one concrete duplicate bug." + }, + { + "left": "pull_request:5517", + "right": "pull_request:6303", + "accept": false, + "reason": "Different projects/features: LCM training script vs ADD training scripts. No shared underlying change." + }, + { + "left": "pull_request:8973", + "right": "pull_request:8987", + "accept": false, + "reason": "Different test areas and goals: Lumina pipeline size reduction vs a new transformer model test suite." + }, + { + "left": "pull_request:6691", + "right": "pull_request:6968", + "accept": false, + "reason": "Same file, but different issues: general advanced SDXL LoRA feature/bug work vs a specific prior-preservation fix." + }, + { + "left": "pull_request:6816", + "right": "pull_request:7233", + "accept": false, + "reason": "One fixes DDP+PEFT behavior; the other adds DoRA support. Related file, different change." + }, + { + "left": "pull_request:10633", + "right": "pull_request:9787", + "accept": false, + "reason": "Docs example for Flux IP-Adapter vs implementation of Flux IP-Adapter. Not the same patch." + }, + { + "left": "pull_request:10298", + "right": "pull_request:11550", + "accept": false, + "reason": "Both add single-file/GGUF support, but for different model families (LTX vs HiDream)." + }, + { + "left": "pull_request:6511", + "right": "pull_request:6534", + "accept": false, + "reason": "Both are torch.compile compatibility fixes, but for different DreamBooth scripts and code paths." + }, + { + "left": "pull_request:644", + "right": "pull_request:645", + "accept": true, + "reason": "Same textual-inversion save-progress feature; 645 is the refined merged follow-up to the WIP in 644." + }, + { + "left": "pull_request:5057", + "right": "pull_request:5938", + "accept": false, + "reason": "SDXL clip-skip support and ControlNet LoRA are unrelated features." + }, + { + "left": "pull_request:12215", + "right": "pull_request:12325", + "accept": false, + "reason": "Both are Qwen-Image ControlNet work, but for different pipeline generations and separate implementation changes." + }, + { + "left": "pull_request:10077", + "right": "pull_request:11271", + "accept": false, + "reason": "Different single-file fixes for different models (SD3.5 vs LTX 0.9.5)." + }, + { + "left": "pull_request:10298", + "right": "pull_request:10991", + "accept": false, + "reason": "Both single-file support changes, but for different target models and code paths." + }, + { + "left": "pull_request:288", + "right": "pull_request:325", + "accept": false, + "reason": "Both are type-hint cleanups, but for different pipelines and unrelated code paths." + }, + { + "left": "pull_request:3642", + "right": "pull_request:4595", + "accept": false, + "reason": "Unrelated community inpainting pipeline vs an unfinished XL conversion work-in-progress." + }, + { + "left": "pull_request:3952", + "right": "pull_request:5175", + "accept": false, + "reason": "Test dependency pinning and HF_HOME permission testing are unrelated." + }, + { + "left": "pull_request:5914", + "right": "pull_request:6556", + "accept": false, + "reason": "Textual inversion saving vs SDXL LoRA compile compatibility are different training-script fixes." + }, + { + "left": "pull_request:6035", + "right": "pull_request:7258", + "accept": false, + "reason": "One changes help text/docs; the other fixes model-card generation logic. Not the same bug." + }, + { + "left": "pull_request:6566", + "right": "pull_request:8755", + "accept": false, + "reason": "Different DreamBooth scripts and different issues: resume handling vs SD3 text-encoder training." + }, + { + "left": "pull_request:7580", + "right": "pull_request:7797", + "accept": false, + "reason": "Different pipelines and test cases; only both reduce fast-test model sizes." + }, + { + "left": "pull_request:7580", + "right": "pull_request:7803", + "accept": false, + "reason": "Different pipelines and test cases; only both reduce fast-test model sizes." + }, + { + "left": "pull_request:7580", + "right": "pull_request:7804", + "accept": false, + "reason": "Different pipelines and test cases; only both reduce fast-test model sizes." + }, + { + "left": "pull_request:7580", + "right": "pull_request:7833", + "accept": false, + "reason": "Different pipelines and test cases; only both reduce fast-test model sizes." + }, + { + "left": "pull_request:7580", + "right": "pull_request:7846", + "accept": false, + "reason": "Different pipelines and test cases; only both reduce fast-test model sizes." + }, + { + "left": "pull_request:8770", + "right": "pull_request:8775", + "accept": false, + "reason": "Both edit README files, but they add different links/content and do not fix the same issue." + }, + { + "left": "pull_request:8983", + "right": "pull_request:9483", + "accept": false, + "reason": "CI parallelism fix vs nightly model-test updates are unrelated." + }, + { + "left": "pull_request:7233", + "right": "pull_request:8926", + "accept": false, + "reason": "DoRA support and checkpoint-save bug fix are separate changes in the same script." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The result is grounded in the packet. The only accepted pair, PR 644/645, is a conservative same-feature match: both modify the textual inversion training script to save learned embeddings during training, with 645 reading as the refined merged follow-up to the WIP in 644. The rejected soft pairs are also cautiously framed and do not overstate equivalence from the available diffs/titles." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 9, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1012", + "pull_request:10991", + "pull_request:11059", + "pull_request:11175", + "pull_request:12170", + "pull_request:12301", + "pull_request:12325", + "pull_request:12496", + "pull_request:12909", + "pull_request:2237", + "pull_request:3127", + "pull_request:3201", + "pull_request:3265", + "pull_request:344", + "pull_request:346", + "pull_request:3686", + "pull_request:4495", + "pull_request:4498", + "pull_request:4621", + "pull_request:4862", + "pull_request:4946", + "pull_request:5090", + "pull_request:5448", + "pull_request:5559", + "pull_request:5993", + "pull_request:6004", + "pull_request:6021", + "pull_request:6035", + "pull_request:6099", + "pull_request:6330", + "pull_request:6511", + "pull_request:6525", + "pull_request:6556", + "pull_request:6566", + "pull_request:6691", + "pull_request:6764", + "pull_request:6795", + "pull_request:6864", + "pull_request:7072", + "pull_request:7124", + "pull_request:7166", + "pull_request:7182", + "pull_request:7196", + "pull_request:7233", + "pull_request:7984", + "pull_request:7988", + "pull_request:9754", + "pull_request:9806" + ], + "soft_pairs": [ + "pull_request:5448|pull_request:5559", + "pull_request:4946|pull_request:7196", + "pull_request:6511|pull_request:6556", + "pull_request:6004|pull_request:6330", + "pull_request:6035|pull_request:6691", + "pull_request:5993|pull_request:6099", + "pull_request:6795|pull_request:7072", + "pull_request:11059|pull_request:12170", + "pull_request:12301|pull_request:12325", + "pull_request:1012|pull_request:2237", + "pull_request:10991|pull_request:11175", + "pull_request:3201|pull_request:3686", + "pull_request:3265|pull_request:3686", + "pull_request:344|pull_request:346", + "pull_request:4495|pull_request:4498", + "pull_request:6021|pull_request:7233", + "pull_request:6511|pull_request:6525", + "pull_request:6511|pull_request:7984", + "pull_request:6511|pull_request:7988", + "pull_request:6566|pull_request:7182", + "pull_request:7124|pull_request:7166", + "pull_request:9754|pull_request:9806", + "pull_request:3127|pull_request:4862", + "pull_request:4621|pull_request:5090", + "pull_request:6764|pull_request:6864", + "pull_request:12496|pull_request:12909" + ], + "prepared_review_unit_hash": "6d8045325c1a308f27d411280f66dacea776283a55d3ea34d1b3de75f85384cc", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 98705, + "estimated_input_tokens": 24677, + "estimated_eval_tokens": 49610 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the PRs span unrelated areas (DreamBooth, Wan, Qwen, SVD docs, adapters, LoRA, compile compatibility). The soft pairs are mostly superficial similarities from shared filenames or broad subsystem overlap, not the same concrete bug or change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR is suitable here; the items are heterogeneous and do not represent one underlying bug/change, so picking any one would be arbitrary.", + "best_issue_reason": null, + "best_pr_reason": "No PR in this set is a clear global representative duplicate target. The closest matches are only loosely related by file or subsystem, not by the same fix or feature.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5448", + "right": "pull_request:5559", + "accept": false, + "reason": "Same LCM family, but one adds the base pipeline and the other adds an img2img variant; different concrete changes." + }, + { + "left": "pull_request:4946", + "right": "pull_request:7196", + "accept": false, + "reason": "Training overview docs vs IP-Adapter training scripts/docs; unrelated features." + }, + { + "left": "pull_request:6511", + "right": "pull_request:6556", + "accept": false, + "reason": "Both are torch.compile-related, but they fix different training scripts and different code paths; not mergeable into one PR." + }, + { + "left": "pull_request:6004", + "right": "pull_request:6330", + "accept": false, + "reason": "Both touch SVD docs, but they fix different documentation issues and different content." + }, + { + "left": "pull_request:6035", + "right": "pull_request:6691", + "accept": false, + "reason": "Same advanced training README, but one is help-text cleanup while the other adds substantial new features and fixes." + }, + { + "left": "pull_request:5993", + "right": "pull_request:6099", + "accept": false, + "reason": "Similar superclass-args bug pattern, but in different pipelines and different implementations." + }, + { + "left": "pull_request:6795", + "right": "pull_request:7072", + "accept": false, + "reason": "Same training script family, but micro-conditioning and DoRA are different features, not the same change." + }, + { + "left": "pull_request:11059", + "right": "pull_request:12170", + "accept": false, + "reason": "Both are LoRA conversion utilities, but they target different model families and different conversion problems." + }, + { + "left": "pull_request:12301", + "right": "pull_request:12325", + "accept": false, + "reason": "Both are Qwen-Image ControlNet variants, but one is inpaint and the other is edit; different pipeline classes and code paths." + }, + { + "left": "pull_request:1012", + "right": "pull_request:2237", + "accept": false, + "reason": "Same DreamBooth script, but one sets text encoder train mode and the other adds learning-rate args; different fixes." + }, + { + "left": "pull_request:10991", + "right": "pull_request:11175", + "accept": false, + "reason": "Wan-related, but single-file loading support and image-embeds support are unrelated changes." + }, + { + "left": "pull_request:3201", + "right": "pull_request:3686", + "accept": false, + "reason": "A revert of one community pipeline update vs an IPEX pipeline update; not the same underlying bug or change." + }, + { + "left": "pull_request:3265", + "right": "pull_request:3686", + "accept": false, + "reason": "Again, these are unrelated edits to the same example file but they revert/add different content for different reasons." + }, + { + "left": "pull_request:344", + "right": "pull_request:346", + "accept": false, + "reason": "Both touch VAE type hints, but one is broad type-hint cleanup and the other is a narrow encode signature annotation." + }, + { + "left": "pull_request:4495", + "right": "pull_request:4498", + "accept": false, + "reason": "One adds the example pipeline code; the other adds docs for it. Related, but not the same change." + }, + { + "left": "pull_request:6021", + "right": "pull_request:7233", + "accept": false, + "reason": "Both affect SDXL DreamBooth LoRA, but variant support and DoRA support are different features." + }, + { + "left": "pull_request:6511", + "right": "pull_request:6525", + "accept": false, + "reason": "Both address checkpointing/compile compatibility, but in different training scripts and different model families." + }, + { + "left": "pull_request:6511", + "right": "pull_request:7984", + "accept": false, + "reason": "Different code paths: DreamBooth checkpointing vs VAE decode compile compatibility." + }, + { + "left": "pull_request:6511", + "right": "pull_request:7988", + "accept": false, + "reason": "Both mention torch.compile, but IPAdapter compile fixes are a separate model/code path from SDXL DreamBooth." + }, + { + "left": "pull_request:6566", + "right": "pull_request:7182", + "accept": false, + "reason": "Same advanced DreamBooth script, but one fixes resume/loading and the other adds EDM-style training." + }, + { + "left": "pull_request:7124", + "right": "pull_request:7166", + "accept": false, + "reason": "Both are SDXL DreamBooth LoRA bug fixes, but they address different failures in different parts of the script." + }, + { + "left": "pull_request:9754", + "right": "pull_request:9806", + "accept": false, + "reason": "Same Flux quantization script, but one introduces the script and the other tweaks optimizer support." + }, + { + "left": "pull_request:3127", + "right": "pull_request:4862", + "accept": false, + "reason": "Stable Diffusion safety-checker img2img bug vs SDXL ControlNet MultiControlNet shape handling; different bugs." + }, + { + "left": "pull_request:4621", + "right": "pull_request:5090", + "accept": false, + "reason": "Both hit adapter code, but one validates downscale factors while the other changes weight accumulation behavior." + }, + { + "left": "pull_request:6764", + "right": "pull_request:6864", + "accept": false, + "reason": "Both are PEFT inference docs, but one covers serialization after fusing adapters and the other covers torch.compile usage." + }, + { + "left": "pull_request:12496", + "right": "pull_request:12909", + "accept": false, + "reason": "Both are WAN I2V-related, but shape mismatch and context-parallel fixes are separate issues in different code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 10, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10222", + "pull_request:10275", + "pull_request:10359", + "pull_request:10365", + "pull_request:10419", + "pull_request:10464", + "pull_request:10585", + "pull_request:1103", + "pull_request:11325", + "pull_request:11348", + "pull_request:1162", + "pull_request:11731", + "pull_request:11740", + "pull_request:11851", + "pull_request:1230", + "pull_request:12757", + "pull_request:13082", + "pull_request:13254", + "pull_request:13283", + "pull_request:13315", + "pull_request:13335", + "pull_request:13344", + "pull_request:13371", + "pull_request:136", + "pull_request:1394", + "pull_request:1423", + "pull_request:168", + "pull_request:1763", + "pull_request:178", + "pull_request:1795", + "pull_request:2487", + "pull_request:325", + "pull_request:326", + "pull_request:3541", + "pull_request:386", + "pull_request:4632", + "pull_request:4749", + "pull_request:4774", + "pull_request:5057", + "pull_request:715", + "pull_request:7166", + "pull_request:869", + "pull_request:8691", + "pull_request:8797", + "pull_request:912", + "pull_request:934", + "pull_request:936", + "pull_request:9751" + ], + "soft_pairs": [ + "pull_request:7166|pull_request:8691", + "pull_request:325|pull_request:326", + "pull_request:4632|pull_request:4749", + "pull_request:4774|pull_request:5057", + "pull_request:10222|pull_request:10365", + "pull_request:10275|pull_request:10585", + "pull_request:10359|pull_request:8797", + "pull_request:10419|pull_request:11325", + "pull_request:10464|pull_request:9751", + "pull_request:1103|pull_request:136", + "pull_request:11348|pull_request:11851", + "pull_request:1162|pull_request:869", + "pull_request:11731|pull_request:11740", + "pull_request:1230|pull_request:934", + "pull_request:12757|pull_request:13283", + "pull_request:13082|pull_request:13371", + "pull_request:13254|pull_request:13344", + "pull_request:13315|pull_request:13344", + "pull_request:13335|pull_request:13344", + "pull_request:1394|pull_request:912", + "pull_request:1423|pull_request:2487", + "pull_request:168|pull_request:386", + "pull_request:1763|pull_request:3541", + "pull_request:178|pull_request:936", + "pull_request:1795|pull_request:715" + ], + "prepared_review_unit_hash": "0972ebf9f81c6298a3e128bdc9adff0159c6c4a692feee203c5917456a2ac55a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 98174, + "estimated_input_tokens": 24544, + "estimated_eval_tokens": 49344 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a noisy mix of unrelated PRs across docs, tests, and small fixes; I don\u2019t see a single underlying duplicate bug/change tying them together.", + "confidence": 0.2, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue artifact exists here, and there is no single duplicated change across the PRs. If a representative must be chosen, PR 10365 is the cleanest focused code fix with an explicit issue target.", + "best_issue_reason": null, + "best_pr_reason": "PR 10365 is a small, well-scoped merged code change with a direct issue target, making it the best representative among otherwise mostly unrelated docs/tests/chore PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7166", + "right": "pull_request:8691", + "accept": false, + "reason": "Both touch LoRA loading, but they fix different model/script paths and different loading behavior; too broad to be the same duplicate change." + }, + { + "left": "pull_request:325", + "right": "pull_request:326", + "accept": false, + "reason": "Different pipelines and different type-hint cleanup; same style of edit, not the same code-path problem." + }, + { + "left": "pull_request:4632", + "right": "pull_request:4749", + "accept": false, + "reason": "Both concern SDXL LoRA training/checkpointing, but one is text-to-image and the other dreambooth; not the same concrete bug." + }, + { + "left": "pull_request:4774", + "right": "pull_request:5057", + "accept": false, + "reason": "Both are SDXL feature work, but one adds negative size/position conditioning and the other adds clip skip; distinct changes." + }, + { + "left": "pull_request:10222", + "right": "pull_request:10365", + "accept": false, + "reason": "Both add torch_xla handling, but for different pipelines and different call sites; not the same underlying fix." + }, + { + "left": "pull_request:10275", + "right": "pull_request:10585", + "accept": false, + "reason": "One is copyright-header churn and the other is a VAE annotation fix; unrelated despite shared files." + }, + { + "left": "pull_request:10359", + "right": "pull_request:8797", + "accept": false, + "reason": "Both are test fixes, but for different test suites and failure modes; not mergeable as one duplicate PR." + }, + { + "left": "pull_request:10419", + "right": "pull_request:11325", + "accept": false, + "reason": "Both are doc updates, but they address different models and different docstring gaps." + }, + { + "left": "pull_request:10464", + "right": "pull_request:9751", + "accept": false, + "reason": "Similar NPU-adaptation titles, but they touch different training scripts and model paths; not the same concrete change." + }, + { + "left": "pull_request:1103", + "right": "pull_request:136", + "accept": false, + "reason": "Model-card URL fix vs manifest packaging fix; unrelated." + }, + { + "left": "pull_request:11348", + "right": "pull_request:11851", + "accept": false, + "reason": "Both are docs-only, but about unrelated topics in different docs pages." + }, + { + "left": "pull_request:1162", + "right": "pull_request:869", + "accept": false, + "reason": "Community-pipeline docs/link cleanup vs an import bug in dynamic modules; different underlying issues." + }, + { + "left": "pull_request:11731", + "right": "pull_request:11740", + "accept": false, + "reason": "Docs tip vs compile/offload tests; different artifact types and different changes." + }, + { + "left": "pull_request:1230", + "right": "pull_request:934", + "accept": false, + "reason": "Both touch MPS testing, but one is a generator/device-specific fast-test fix and the other is broader MPS failing tests." + }, + { + "left": "pull_request:12757", + "right": "pull_request:13283", + "accept": false, + "reason": "Same docs file, but one is a formatting/link tweak and the other changes the FlashAttention-3 support note." + }, + { + "left": "pull_request:13082", + "right": "pull_request:13371", + "accept": false, + "reason": "Both are test refactors for Wan-related models, but they cover different model classes and test scopes." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13344", + "accept": false, + "reason": "Different model test refactors; no shared concrete bug or change." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13344", + "accept": false, + "reason": "Different model test refactors; same pattern, different target models." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13344", + "accept": false, + "reason": "Cosmos vs GLM image tests are unrelated refactors, not one duplicate fix." + }, + { + "left": "pull_request:1394", + "right": "pull_request:912", + "accept": false, + "reason": "Both are version bumps, but to different versions and in different release moments." + }, + { + "left": "pull_request:1423", + "right": "pull_request:2487", + "accept": false, + "reason": "One adds Flax code support and the other adds a how-to guide; code change vs documentation." + }, + { + "left": "pull_request:168", + "right": "pull_request:386", + "accept": false, + "reason": "Stable Diffusion implementation PR vs Stable Diffusion docs PR; not the same artifact change." + }, + { + "left": "pull_request:1763", + "right": "pull_request:3541", + "accept": false, + "reason": "Attention-mask fixes in different code paths: a model attention module vs dreambooth training data collation." + }, + { + "left": "pull_request:178", + "right": "pull_request:936", + "accept": false, + "reason": "Adding Stable Diffusion tests is not the same as splitting tests into separate files; distinct test-maintenance changes." + }, + { + "left": "pull_request:1795", + "right": "pull_request:715", + "accept": false, + "reason": "Both address environment/import behavior, but one is a JAX-installed OOM workaround and the other is a PyTorch import guard; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 11, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1960", + "pull_request:2487", + "pull_request:2539", + "pull_request:2559", + "pull_request:2604", + "pull_request:2757", + "pull_request:2839", + "pull_request:2887", + "pull_request:3166", + "pull_request:3168", + "pull_request:3170", + "pull_request:3525", + "pull_request:3633", + "pull_request:3680", + "pull_request:3687", + "pull_request:3819", + "pull_request:4113", + "pull_request:4145", + "pull_request:4258", + "pull_request:4317", + "pull_request:4330", + "pull_request:4339", + "pull_request:4472", + "pull_request:4574", + "pull_request:4626", + "pull_request:4717", + "pull_request:5411", + "pull_request:5720", + "pull_request:5721", + "pull_request:5838", + "pull_request:6106", + "pull_request:6122", + "pull_request:6282", + "pull_request:6837", + "pull_request:7360", + "pull_request:7425", + "pull_request:7903", + "pull_request:7910", + "pull_request:8064", + "pull_request:8483", + "pull_request:874", + "pull_request:8775", + "pull_request:889", + "pull_request:8923", + "pull_request:897", + "pull_request:9386", + "pull_request:9553" + ], + "soft_pairs": [ + "pull_request:1960|pull_request:3525", + "pull_request:2487|pull_request:889", + "pull_request:2539|pull_request:4339", + "pull_request:2559|pull_request:4717", + "pull_request:2604|pull_request:3680", + "pull_request:2757|pull_request:4113", + "pull_request:2839|pull_request:3819", + "pull_request:2887|pull_request:4317", + "pull_request:3166|pull_request:5411", + "pull_request:3168|pull_request:3687", + "pull_request:3170|pull_request:3633", + "pull_request:3170|pull_request:4145", + "pull_request:4258|pull_request:4472", + "pull_request:4317|pull_request:6106", + "pull_request:4317|pull_request:6282", + "pull_request:4330|pull_request:4574", + "pull_request:4574|pull_request:4626", + "pull_request:5720|pull_request:6122", + "pull_request:5721|pull_request:5838", + "pull_request:6106|pull_request:6282", + "pull_request:6837|pull_request:7360", + "pull_request:6837|pull_request:7910", + "pull_request:7360|pull_request:7425", + "pull_request:7903|pull_request:8923", + "pull_request:8064|pull_request:8923", + "pull_request:8483|pull_request:9386", + "pull_request:874|pull_request:897", + "pull_request:8775|pull_request:9553" + ], + "prepared_review_unit_hash": "434d9e2e59d019e646b92e65a569046ff4449d6f17b71d2fc3f10cb8c4a83d66", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 94782, + "estimated_input_tokens": 23696, + "estimated_eval_tokens": 47648 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is a grab bag of mostly unrelated docs/example cleanup, CI dependency tweaks, and a few unrelated code fixes. The soft-similarity signals are mostly superficial (generic README/docs edits or same broad subsystem), so none of the candidate pairs look like true duplicates of the same concrete change.", + "confidence": 0.18, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong canonical representative: the items span unrelated changes across docs, examples, CI, and core code, so there is no single underlying fix/change to anchor the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1960", + "right": "pull_request:3525", + "accept": false, + "reason": "Both are README/docs updates, but for different projects and different content; not the same change." + }, + { + "left": "pull_request:2487", + "right": "pull_request:889", + "accept": false, + "reason": "One adds a JAX/Flax guide, the other adds the DiffusionPipeline components method; unrelated docs/features." + }, + { + "left": "pull_request:2539", + "right": "pull_request:4339", + "accept": false, + "reason": "Both touch guidance_scale docs, but in different files for different pipelines and with different values/context." + }, + { + "left": "pull_request:2559", + "right": "pull_request:4717", + "accept": false, + "reason": "ControlNet docs formatting vs a missing tag fix; different documentation issues." + }, + { + "left": "pull_request:2604", + "right": "pull_request:3680", + "accept": false, + "reason": "Both are link fixes, but in different docs sections and for different targets." + }, + { + "left": "pull_request:2757", + "right": "pull_request:4113", + "accept": false, + "reason": "Different bug fixes in different pipelines: depth batching vs dtype handling." + }, + { + "left": "pull_request:2839", + "right": "pull_request:3819", + "accept": false, + "reason": "Same README file, but one adds example entries while the other adds an FAQ section; not one merged change." + }, + { + "left": "pull_request:2887", + "right": "pull_request:4317", + "accept": false, + "reason": "ControlNet validation batching and DreamBooth validation config are unrelated training-example fixes." + }, + { + "left": "pull_request:3166", + "right": "pull_request:5411", + "accept": false, + "reason": "Textual inversion token/doc example fix vs broken training-doc link; different underlying issues." + }, + { + "left": "pull_request:3168", + "right": "pull_request:3687", + "accept": false, + "reason": "Both are doc-builder test edits, but they change different page titles/files and are not the same patch." + }, + { + "left": "pull_request:3170", + "right": "pull_request:3633", + "accept": false, + "reason": "Both tweak setup.py pins, but they target different dependency versions and were separate test edits." + }, + { + "left": "pull_request:3170", + "right": "pull_request:4145", + "accept": false, + "reason": "setup.py dependency pinning is unrelated to the UnCLIP constructor comment changes." + }, + { + "left": "pull_request:4258", + "right": "pull_request:4472", + "accept": false, + "reason": "Both edit README_sdxl.md, but one rewrites the notes section while the other removes login instructions." + }, + { + "left": "pull_request:4317", + "right": "pull_request:6106", + "accept": false, + "reason": "DreamBooth validation config vs an advanced training script example fix; different code paths." + }, + { + "left": "pull_request:4317", + "right": "pull_request:6282", + "accept": false, + "reason": "Both are DreamBooth script changes, but one fixes tracker config copying while the other removes a VAE helper; different bugs." + }, + { + "left": "pull_request:4330", + "right": "pull_request:4574", + "accept": false, + "reason": "ControlNet README header correction vs InstructPix2Pix README mention updates; unrelated docs changes." + }, + { + "left": "pull_request:4574", + "right": "pull_request:4626", + "accept": false, + "reason": "InstructPix2Pix README edits vs text2image link fixes; different docs and content." + }, + { + "left": "pull_request:5720", + "right": "pull_request:6122", + "accept": false, + "reason": "Both are broad docs typo sweeps, but they are separate content sets and not the same underlying fix." + }, + { + "left": "pull_request:5721", + "right": "pull_request:5838", + "accept": false, + "reason": "Both adjust CI installs, but in different workflows and with different dependency goals." + }, + { + "left": "pull_request:6106", + "right": "pull_request:6282", + "accept": false, + "reason": "Advanced training example correction vs DreamBooth runtime error fix; different scripts and issues." + }, + { + "left": "pull_request:6837", + "right": "pull_request:7360", + "accept": false, + "reason": "i2vgen docs dependency import order vs TCD scheduler docs/type cleanup; unrelated." + }, + { + "left": "pull_request:6837", + "right": "pull_request:7910", + "accept": false, + "reason": "Different files and domains: i2vgen docs vs image_processor docs additions." + }, + { + "left": "pull_request:7360", + "right": "pull_request:7425", + "accept": false, + "reason": "Scheduler doc tweaks vs PEFT inference doc formatting; not the same change." + }, + { + "left": "pull_request:7903", + "right": "pull_request:8923", + "accept": false, + "reason": "VideoProcessor test scaffolding vs CI Hub debug changes; unrelated." + }, + { + "left": "pull_request:8064", + "right": "pull_request:8923", + "accept": false, + "reason": "Marigold test PR vs CI Hub debug PR; different areas and purposes." + }, + { + "left": "pull_request:8483", + "right": "pull_request:9386", + "accept": false, + "reason": "Different feature additions: Stable Diffusion 3 vs PAG support for SD inpaint." + }, + { + "left": "pull_request:874", + "right": "pull_request:897", + "accept": false, + "reason": "Same community README, but different added example entries and different goals." + }, + { + "left": "pull_request:8775", + "right": "pull_request:9553", + "accept": false, + "reason": "Colab link fix for an SD3 README vs a distributed inference doc update; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 12, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10063", + "pull_request:10323", + "pull_request:10346", + "pull_request:10402", + "pull_request:10968", + "pull_request:11243", + "pull_request:11292", + "pull_request:11431", + "pull_request:11451", + "pull_request:11516", + "pull_request:11571", + "pull_request:12095", + "pull_request:12225", + "pull_request:12249", + "pull_request:1508", + "pull_request:1513", + "pull_request:2766", + "pull_request:3170", + "pull_request:4944", + "pull_request:5175", + "pull_request:5448", + "pull_request:5506", + "pull_request:5517", + "pull_request:5807", + "pull_request:5915", + "pull_request:6225", + "pull_request:6267", + "pull_request:6714", + "pull_request:6757", + "pull_request:6760", + "pull_request:6787", + "pull_request:6962", + "pull_request:7062", + "pull_request:7166", + "pull_request:7206", + "pull_request:7372", + "pull_request:7966", + "pull_request:8572", + "pull_request:9005", + "pull_request:9008", + "pull_request:9070", + "pull_request:9257", + "pull_request:934", + "pull_request:9408", + "pull_request:9483", + "pull_request:9881", + "pull_request:9940", + "pull_request:9993" + ], + "soft_pairs": [ + "pull_request:9070|pull_request:9408", + "pull_request:5448|pull_request:5517", + "pull_request:11571|pull_request:12095", + "pull_request:4944|pull_request:5915", + "pull_request:5506|pull_request:6225", + "pull_request:9257|pull_request:9993", + "pull_request:6267|pull_request:6757", + "pull_request:6787|pull_request:7062", + "pull_request:10968|pull_request:11516", + "pull_request:5807|pull_request:7062", + "pull_request:10346|pull_request:9881", + "pull_request:10063|pull_request:9940", + "pull_request:11243|pull_request:11292", + "pull_request:11431|pull_request:11451", + "pull_request:8572|pull_request:9008", + "pull_request:12225|pull_request:12249", + "pull_request:10323|pull_request:10402", + "pull_request:2766|pull_request:934", + "pull_request:3170|pull_request:5175", + "pull_request:6714|pull_request:6760", + "pull_request:6962|pull_request:7966", + "pull_request:9005|pull_request:9483", + "pull_request:7206|pull_request:7372", + "pull_request:1508|pull_request:1513", + "pull_request:6225|pull_request:7166" + ], + "prepared_review_unit_hash": "7298db4569d9fba545cf6cbd6ab1d3bee8d98f60022cfeb0f5f63ac73b2af15f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 103705, + "estimated_input_tokens": 25927, + "estimated_eval_tokens": 52110 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is heterogeneous; the soft pairs are mostly same subsystem or adjacent follow-up work, but not the same concrete bug/change. I would not collapse any pair as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical representative: these items span unrelated docs, model features, tests, and bug fixes rather than one underlying change.", + "best_issue_reason": null, + "best_pr_reason": "No PR stands out as globally canonical for the cluster because there is no single shared defect or feature across the items.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9070", + "right": "pull_request:9408", + "accept": false, + "reason": "Both are FLUX pipeline additions, but one adds base Img2Img while the other adds ControlNet img2img/inpaint; different code paths and feature scope." + }, + { + "left": "pull_request:5448", + "right": "pull_request:5517", + "accept": false, + "reason": "A released pipeline and a draft distillation script are not the same change; one adds inference support, the other is training code." + }, + { + "left": "pull_request:11571", + "right": "pull_request:12095", + "accept": false, + "reason": "Different LTX releases and different functionality; 0.9.7-distilled docs/improvements are not the same change as 0.9.8 model support." + }, + { + "left": "pull_request:4944", + "right": "pull_request:5915", + "accept": false, + "reason": "IP-Adapter and IP-Adapter Plus are related, but they are distinct implementations with different loading/image-projection logic, not one fix." + }, + { + "left": "pull_request:5506", + "right": "pull_request:6225", + "accept": false, + "reason": "Both involve LoRA, but one fixes PEFT+xformers attention behavior while the other fixes alpha parsing/loading; separate bugs." + }, + { + "left": "pull_request:9257", + "right": "pull_request:9993", + "accept": false, + "reason": "Different DreamBooth scripts and different bugs: Flux prompt/tokenization handling versus SD3 checkpoint saving logic." + }, + { + "left": "pull_request:6267", + "right": "pull_request:6757", + "accept": false, + "reason": "A refactor moving k-diffusion code is not the same as a device-placement fix for sigma tensors." + }, + { + "left": "pull_request:6787", + "right": "pull_request:7062", + "accept": false, + "reason": "Completely unrelated: Marigold depth input normalization versus an Imagic docstring cleanup." + }, + { + "left": "pull_request:10968", + "right": "pull_request:11516", + "accept": false, + "reason": "Both are LTX releases, but they introduce different model versions and pipeline additions; not one mergeable fix." + }, + { + "left": "pull_request:5807", + "right": "pull_request:7062", + "accept": false, + "reason": "A community pipeline feature fix is unrelated to an Imagic docstring edit." + }, + { + "left": "pull_request:10346", + "right": "pull_request:9881", + "accept": false, + "reason": "Same file, but different IP-Adapter concerns: optional masks versus xformers attention processor support." + }, + { + "left": "pull_request:10063", + "right": "pull_request:9940", + "accept": false, + "reason": "Both are docs nits, but they fix different pages and different wording issues." + }, + { + "left": "pull_request:11243", + "right": "pull_request:11292", + "accept": false, + "reason": "Both touch LoRA conversion utilities, but they target different adapter formats/model families and different conversion paths." + }, + { + "left": "pull_request:11431", + "right": "pull_request:11451", + "accept": false, + "reason": "Same general test pattern, but different models and different test targets; not the same concrete bug." + }, + { + "left": "pull_request:8572", + "right": "pull_request:9008", + "accept": false, + "reason": "Both are broad typo/errata sweeps, not the same underlying documentation change." + }, + { + "left": "pull_request:12225", + "right": "pull_request:12249", + "accept": false, + "reason": "Two different new pipelines for different models; they do not fix the same code-path problem." + }, + { + "left": "pull_request:10323", + "right": "pull_request:10402", + "accept": false, + "reason": "Both are documentation fixes, but for different sections and different broken links." + }, + { + "left": "pull_request:2766", + "right": "pull_request:934", + "accept": false, + "reason": "Both address MPS-related test issues, but the affected tests and expected outputs differ; not the same bug." + }, + { + "left": "pull_request:3170", + "right": "pull_request:5175", + "accept": false, + "reason": "Unrelated setup.py pin versus CI HF_HOME permission path change." + }, + { + "left": "pull_request:6714", + "right": "pull_request:6760", + "accept": false, + "reason": "One adds IP-Adapter tests to the slow matrix; the other updates slow-test expectations. Different changes, not a single fix." + }, + { + "left": "pull_request:6962", + "right": "pull_request:7966", + "accept": false, + "reason": "Different PixArt model test suites and different expected outputs; not one shared bug." + }, + { + "left": "pull_request:9005", + "right": "pull_request:9483", + "accept": false, + "reason": "Both are CI/test maintenance, but the workflows and model expectations are unrelated." + }, + { + "left": "pull_request:7206", + "right": "pull_request:7372", + "accept": false, + "reason": "Model-card tagging cleanup versus adding EDM/scheduler metadata; related area, but not the same underlying change." + }, + { + "left": "pull_request:1508", + "right": "pull_request:1513", + "accept": false, + "reason": "Same README, but different docs fixes: prior-preservation clarification versus accelerate config bootstrapping." + }, + { + "left": "pull_request:6225", + "right": "pull_request:7166", + "accept": false, + "reason": "Both are DreamBooth LoRA-related, but one fixes alpha parsing while the other fixes SDXL checkpoint loading; separate bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 13, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10206", + "pull_request:10222", + "pull_request:10246", + "pull_request:10253", + "pull_request:10364", + "pull_request:1073", + "pull_request:10819", + "pull_request:11059", + "pull_request:1130", + "pull_request:11537", + "pull_request:1167", + "pull_request:1169", + "pull_request:11825", + "pull_request:11920", + "pull_request:11955", + "pull_request:121", + "pull_request:1230", + "pull_request:13071", + "pull_request:13340", + "pull_request:1418", + "pull_request:1420", + "pull_request:1423", + "pull_request:1467", + "pull_request:16", + "pull_request:171", + "pull_request:1727", + "pull_request:1729", + "pull_request:1769", + "pull_request:1849", + "pull_request:1960", + "pull_request:1998", + "pull_request:2487", + "pull_request:252", + "pull_request:2526", + "pull_request:260", + "pull_request:2604", + "pull_request:3819", + "pull_request:386", + "pull_request:3989", + "pull_request:531", + "pull_request:555", + "pull_request:62", + "pull_request:642", + "pull_request:874", + "pull_request:879", + "pull_request:889", + "pull_request:942", + "pull_request:9598" + ], + "soft_pairs": [ + "pull_request:10206|pull_request:11059", + "pull_request:10222|pull_request:10364", + "pull_request:10246|pull_request:10819", + "pull_request:10253|pull_request:9598", + "pull_request:1073|pull_request:121", + "pull_request:1073|pull_request:252", + "pull_request:1130|pull_request:874", + "pull_request:11537|pull_request:11920", + "pull_request:11537|pull_request:11955", + "pull_request:1167|pull_request:121", + "pull_request:1167|pull_request:252", + "pull_request:1169|pull_request:942", + "pull_request:11825|pull_request:11920", + "pull_request:11825|pull_request:11955", + "pull_request:121|pull_request:16", + "pull_request:121|pull_request:1960", + "pull_request:1230|pull_request:1727", + "pull_request:13071|pull_request:13340", + "pull_request:1418|pull_request:1849", + "pull_request:1420|pull_request:1423", + "pull_request:1420|pull_request:555", + "pull_request:1420|pull_request:889", + "pull_request:1467|pull_request:2526", + "pull_request:16|pull_request:252", + "pull_request:171|pull_request:62", + "pull_request:1729|pull_request:1769", + "pull_request:1729|pull_request:531", + "pull_request:1960|pull_request:252", + "pull_request:1998|pull_request:3819", + "pull_request:2487|pull_request:386", + "pull_request:2526|pull_request:3989", + "pull_request:260|pull_request:879", + "pull_request:2604|pull_request:642" + ], + "prepared_review_unit_hash": "98de8fe2f9523f507a9a988ecd968b6d5292c1fce618c97b19533dc387a729a9", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 97580, + "estimated_input_tokens": 24395, + "estimated_eval_tokens": 49046 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly false-positive similarity: the PRs span unrelated docs edits, test updates, and distinct pipeline/runtime fixes. I would reject all soft edges; none look like the same underlying change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the set is heterogeneous and no PR subsumes the others or fixes the same concrete bug/change.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:1169 is the strongest standalone fix in the set because it makes a concrete runtime correction across scheduler/pipeline code and includes aligned tests, but it still does not match the other items closely enough to serve as a duplicate canonical.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10206", + "right": "pull_request:11059", + "accept": false, + "reason": "Different LoRA features for different model families (Flux Control vs WAN); not the same bug or patch." + }, + { + "left": "pull_request:10222", + "right": "pull_request:10364", + "accept": false, + "reason": "Both add torch_xla support, but to different pipelines and code paths; separate changes." + }, + { + "left": "pull_request:10246", + "right": "pull_request:10819", + "accept": false, + "reason": "Unrelated docs updates in different API sections; no shared bug/change." + }, + { + "left": "pull_request:10253", + "right": "pull_request:9598", + "accept": false, + "reason": "Different documentation note additions in different files; not duplicates." + }, + { + "left": "pull_request:1073", + "right": "pull_request:121", + "accept": false, + "reason": "Both touch README, but one is a typo/command-line fix and the other adds example code from a different era." + }, + { + "left": "pull_request:1073", + "right": "pull_request:252", + "accept": false, + "reason": "README edits only share the file name; the content and intent are unrelated." + }, + { + "left": "pull_request:1130", + "right": "pull_request:874", + "accept": false, + "reason": "Different community README changes: one adds LoRA/offload support, the other adds a generic inference example." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11920", + "accept": false, + "reason": "Different test suites for different models (UNet2DConditionModel vs Wan pipelines)." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11955", + "accept": false, + "reason": "Same broad theme of tests, but for different models and assertions; not the same fix." + }, + { + "left": "pull_request:1167", + "right": "pull_request:121", + "accept": false, + "reason": "Both are README edits, but they address unrelated content in different sections and contexts." + }, + { + "left": "pull_request:1167", + "right": "pull_request:252", + "accept": false, + "reason": "Shared README file only; one is an Intel XPU note, the other is example expansion." + }, + { + "left": "pull_request:1169", + "right": "pull_request:942", + "accept": false, + "reason": "Related MPS dtype issues, but they fix different scheduler/pipeline paths and are not one mergeable PR." + }, + { + "left": "pull_request:11825", + "right": "pull_request:11920", + "accept": false, + "reason": "Compilation/resolution-change tests vs Wan test slices; different problems." + }, + { + "left": "pull_request:11825", + "right": "pull_request:11955", + "accept": false, + "reason": "Different test concerns and different pipelines; not the same underlying issue." + }, + { + "left": "pull_request:121", + "right": "pull_request:16", + "accept": false, + "reason": "Both README updates, but they are from different parts of the project and unrelated content." + }, + { + "left": "pull_request:121", + "right": "pull_request:1960", + "accept": false, + "reason": "One adds usage examples, the other updates a research-project README; no shared code/change." + }, + { + "left": "pull_request:1230", + "right": "pull_request:1727", + "accept": false, + "reason": "MPS fast-test fixes vs ONNX img2img preprocessing/coverage; unrelated changes." + }, + { + "left": "pull_request:13071", + "right": "pull_request:13340", + "accept": false, + "reason": "Different model test refactors for Flux2 vs SD3; not the same concrete fix." + }, + { + "left": "pull_request:1418", + "right": "pull_request:1849", + "accept": false, + "reason": "Versatile Diffusion test/cleanup work vs a cross_attention_kwargs plumbing fix; different problems." + }, + { + "left": "pull_request:1420", + "right": "pull_request:1423", + "accept": false, + "reason": "Stable Diffusion 2 test additions vs Flax support implementation; unrelated." + }, + { + "left": "pull_request:1420", + "right": "pull_request:555", + "accept": false, + "reason": "Test additions vs Flax VAE implementation; not the same bug/change." + }, + { + "left": "pull_request:1420", + "right": "pull_request:889", + "accept": false, + "reason": "Test file vs pipeline utility feature; different scope and purpose." + }, + { + "left": "pull_request:1467", + "right": "pull_request:2526", + "accept": false, + "reason": "Both adjust slow tests, but they target different pipelines and assertions." + }, + { + "left": "pull_request:16", + "right": "pull_request:252", + "accept": false, + "reason": "Both are README expansions, but they are separate documentation changes with no common underlying bug." + }, + { + "left": "pull_request:171", + "right": "pull_request:62", + "accept": false, + "reason": "Both touch latent diffusion, but one fixes eta handling and the other fixes unconditional decoding; distinct bugs." + }, + { + "left": "pull_request:1729", + "right": "pull_request:1769", + "accept": false, + "reason": "Different test fixes for different pipelines (latent diffusion vs UnCLIP)." + }, + { + "left": "pull_request:1729", + "right": "pull_request:531", + "accept": false, + "reason": "Latent diffusion test/device handling vs spatial transformer GPU test fixes; unrelated." + }, + { + "left": "pull_request:1960", + "right": "pull_request:252", + "accept": false, + "reason": "Different docs areas and different kinds of README updates; not duplicates." + }, + { + "left": "pull_request:1998", + "right": "pull_request:3819", + "accept": false, + "reason": "Both DreamBooth README edits, but one adds a blog post and the other adds a FAQ section; separate content." + }, + { + "left": "pull_request:2487", + "right": "pull_request:386", + "accept": false, + "reason": "JAX/Flax guide addition vs Stable Diffusion pipeline docs; different documentation work." + }, + { + "left": "pull_request:2526", + "right": "pull_request:3989", + "accept": false, + "reason": "Slow-test threshold tweaks vs SDXL offload support; completely different changes." + }, + { + "left": "pull_request:260", + "right": "pull_request:879", + "accept": false, + "reason": "Both community docs, but one fixes a typo and the other changes a table entry; not the same change." + }, + { + "left": "pull_request:2604", + "right": "pull_request:642", + "accept": false, + "reason": "Both fix links in docs, but they point to different targets and are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 14, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2677", + "pull_request:2788", + "pull_request:2839", + "pull_request:2897", + "pull_request:3030", + "pull_request:3128", + "pull_request:3168", + "pull_request:3170", + "pull_request:3625", + "pull_request:3819", + "pull_request:4175", + "pull_request:4330", + "pull_request:4351", + "pull_request:4540", + "pull_request:4574", + "pull_request:4612", + "pull_request:4655", + "pull_request:4696", + "pull_request:4862", + "pull_request:4946", + "pull_request:5332", + "pull_request:5626", + "pull_request:5747", + "pull_request:5844", + "pull_request:5852", + "pull_request:5982", + "pull_request:6106", + "pull_request:6389", + "pull_request:6403", + "pull_request:6440", + "pull_request:6475", + "pull_request:6592", + "pull_request:6775", + "pull_request:7124", + "pull_request:7360", + "pull_request:7495", + "pull_request:7542", + "pull_request:7544", + "pull_request:7910", + "pull_request:832", + "pull_request:874", + "pull_request:8775", + "pull_request:8923", + "pull_request:9070", + "pull_request:9131", + "pull_request:9135", + "pull_request:9760", + "pull_request:9775" + ], + "soft_pairs": [ + "pull_request:2677|pull_request:2788", + "pull_request:2839|pull_request:4351", + "pull_request:2839|pull_request:4574", + "pull_request:2897|pull_request:874", + "pull_request:3030|pull_request:832", + "pull_request:3128|pull_request:4862", + "pull_request:3168|pull_request:3625", + "pull_request:3170|pull_request:4175", + "pull_request:3819|pull_request:4351", + "pull_request:4330|pull_request:4540", + "pull_request:4612|pull_request:6440", + "pull_request:4655|pull_request:5747", + "pull_request:4696|pull_request:4946", + "pull_request:5332|pull_request:7124", + "pull_request:5626|pull_request:5852", + "pull_request:5844|pull_request:5982", + "pull_request:6106|pull_request:6403", + "pull_request:6389|pull_request:6475", + "pull_request:6403|pull_request:6592", + "pull_request:6775|pull_request:7360", + "pull_request:6775|pull_request:7910", + "pull_request:7495|pull_request:8775", + "pull_request:7542|pull_request:7544", + "pull_request:8923|pull_request:9131", + "pull_request:9070|pull_request:9135", + "pull_request:9760|pull_request:9775" + ], + "prepared_review_unit_hash": "5a2d1d2bbc7fb94dd2f905dc8863ef2c0d7b923e9f1d68d1d58d35d8e66b0de6", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 90921, + "estimated_input_tokens": 22731, + "estimated_eval_tokens": 45718 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly unrelated docs, example, and test PRs. The only clear duplicate-like pair is the repeated `data_dir` dataset-loading fix in 4655 and 5747; the rest should stay separate.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 5747 is the best canonical PR candidate because it is the merged, final version of the same `data_dir` fix that 4655 also applies in the same two training scripts.", + "best_issue_reason": null, + "best_pr_reason": "5747 is the strongest representative fix in the set: it is a concrete, merged code-path change rather than docs or test churn, and it cleanly captures the shared `data_dir` bug.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2677", + "right": "pull_request:2788", + "accept": false, + "reason": "Both touch docs navigation, but one adds a ControlNet docs entry while the other deduplicates the training section; different changes." + }, + { + "left": "pull_request:2839", + "right": "pull_request:4351", + "accept": false, + "reason": "Both edit README.md, but one adds example rows and the other changes an image path; unrelated changes." + }, + { + "left": "pull_request:2839", + "right": "pull_request:4574", + "accept": false, + "reason": "One updates the examples README table, the other updates InstructPix2Pix README text; not the same change." + }, + { + "left": "pull_request:2897", + "right": "pull_request:874", + "accept": false, + "reason": "Different docs goals: StableUnCLIP usage example vs generic community pipeline example." + }, + { + "left": "pull_request:3030", + "right": "pull_request:832", + "accept": false, + "reason": "Different bugs in different pipelines; panorama CFG handling is not the Flax NSFW image-blocking fix." + }, + { + "left": "pull_request:3128", + "right": "pull_request:4862", + "accept": false, + "reason": "Both are timestep/shape-related, but they fix different pipelines and different failure modes." + }, + { + "left": "pull_request:3168", + "right": "pull_request:3625", + "accept": false, + "reason": "Both are doc-builder test PRs, but they touch different workflow files and unrelated test scaffolding." + }, + { + "left": "pull_request:3170", + "right": "pull_request:4175", + "accept": false, + "reason": "Setup dependency pinning and a ControlNet webdataset example are unrelated changes." + }, + { + "left": "pull_request:3819", + "right": "pull_request:4351", + "accept": false, + "reason": "Both modify README.md, but one adds FAQ content while the other changes a PyPI-friendly image URL." + }, + { + "left": "pull_request:4330", + "right": "pull_request:4540", + "accept": false, + "reason": "Both are README tweaks, but one corrects the SDXL ControlNet header and the other adds a Colab notebook link." + }, + { + "left": "pull_request:4612", + "right": "pull_request:6440", + "accept": false, + "reason": "Both fix docs links, but they target different pages and different broken references." + }, + { + "left": "pull_request:4655", + "right": "pull_request:5747", + "accept": true, + "reason": "Same underlying bug: both add `data_dir` to the same dataset-loading calls in the same two training scripts." + }, + { + "left": "pull_request:4696", + "right": "pull_request:4946", + "accept": false, + "reason": "One adds T2I-Adapter XL models; the other updates training overview docs to list T2I adapters. Different scope." + }, + { + "left": "pull_request:5332", + "right": "pull_request:7124", + "accept": false, + "reason": "Different DreamBooth LoRA edits: a comment typo vs a kwarg rename in validation flow." + }, + { + "left": "pull_request:5626", + "right": "pull_request:5852", + "accept": false, + "reason": "Both involve tests, but they address different test files and different failures." + }, + { + "left": "pull_request:5844", + "right": "pull_request:5982", + "accept": false, + "reason": "Different documentation additions for different pipelines; no shared underlying bug or feature." + }, + { + "left": "pull_request:6106", + "right": "pull_request:6403", + "accept": false, + "reason": "Both touch the advanced training script, but one fixes a saved example and the other adds WebUI format support; not the same change." + }, + { + "left": "pull_request:6389", + "right": "pull_request:6475", + "accept": false, + "reason": "Different Stable Video Diffusion doc/code fixes: parameter type vs VAE type annotation." + }, + { + "left": "pull_request:6403", + "right": "pull_request:6592", + "accept": false, + "reason": "WebUI format support in an existing script is unrelated to adding a new inpainting training script." + }, + { + "left": "pull_request:6775", + "right": "pull_request:7360", + "accept": false, + "reason": "Both are scheduler docs updates, but they address different schedulers and different missing docs details." + }, + { + "left": "pull_request:6775", + "right": "pull_request:7910", + "accept": false, + "reason": "Unrelated docs additions: DDPM scheduler docs vs image processor docs." + }, + { + "left": "pull_request:7495", + "right": "pull_request:8775", + "accept": false, + "reason": "Both add Colab links, but for different READMEs and different training projects." + }, + { + "left": "pull_request:7542", + "right": "pull_request:7544", + "accept": false, + "reason": "Both are CPU offload tests, but one fixes a typo and the other changes assertion logic; not the same underlying fix." + }, + { + "left": "pull_request:8923", + "right": "pull_request:9131", + "accept": false, + "reason": "CI/debug workflow edits are unrelated to a tokenizers dependency pin." + }, + { + "left": "pull_request:9070", + "right": "pull_request:9135", + "accept": false, + "reason": "9135 adds Flux img2img plus inpaint, while 9070 is only a draft img2img pipeline; not the same concrete change." + }, + { + "left": "pull_request:9760", + "right": "pull_request:9775", + "accept": false, + "reason": "Both edit READMEs, but one is a general wording cleanup and the other adds a Flux quantization note." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, 4655/5747, is a plausible duplicate-like bugfix for the same `data_dir` dataset-loading calls in the same two scripts. The other pairwise rejections are appropriately cautious and do not overstate sameness." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 15, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10192", + "pull_request:10268", + "pull_request:10330", + "pull_request:10351", + "pull_request:10578", + "pull_request:1063", + "pull_request:10718", + "pull_request:10728", + "pull_request:10867", + "pull_request:10968", + "pull_request:11039", + "pull_request:11451", + "pull_request:11512", + "pull_request:11550", + "pull_request:11588", + "pull_request:11595", + "pull_request:11655", + "pull_request:11892", + "pull_request:12074", + "pull_request:12225", + "pull_request:12301", + "pull_request:12909", + "pull_request:12970", + "pull_request:3952", + "pull_request:4145", + "pull_request:4451", + "pull_request:4526", + "pull_request:4714", + "pull_request:4783", + "pull_request:5051", + "pull_request:5506", + "pull_request:5697", + "pull_request:5915", + "pull_request:6308", + "pull_request:6526", + "pull_request:6550", + "pull_request:6795", + "pull_request:6864", + "pull_request:7131", + "pull_request:7189", + "pull_request:8290", + "pull_request:8627", + "pull_request:8715", + "pull_request:8725", + "pull_request:8875", + "pull_request:948", + "pull_request:9985", + "pull_request:9987" + ], + "soft_pairs": [ + "pull_request:4783|pull_request:6864", + "pull_request:10192|pull_request:9985", + "pull_request:11451|pull_request:11512", + "pull_request:12225|pull_request:12301", + "pull_request:6526|pull_request:6550", + "pull_request:6795|pull_request:8715", + "pull_request:8290|pull_request:8627", + "pull_request:11588|pull_request:12074", + "pull_request:12909|pull_request:12970", + "pull_request:11039|pull_request:11892", + "pull_request:10330|pull_request:10968", + "pull_request:10578|pull_request:11655", + "pull_request:10728|pull_request:10867", + "pull_request:4451|pull_request:4526", + "pull_request:8725|pull_request:8875", + "pull_request:10351|pull_request:11595", + "pull_request:10718|pull_request:9987", + "pull_request:4714|pull_request:5506", + "pull_request:5915|pull_request:7189", + "pull_request:10268|pull_request:11550", + "pull_request:1063|pull_request:948", + "pull_request:3952|pull_request:4145", + "pull_request:5051|pull_request:7131", + "pull_request:5697|pull_request:6308" + ], + "prepared_review_unit_hash": "2924a07c54e8a2af202f5946cfa7ce80762405fcaaafe06ea4cc82802c5988a1", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 108461, + "estimated_input_tokens": 27116, + "estimated_eval_tokens": 54488 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of mostly unrelated PRs across Flux, LTX, Qwen-Image, Wan, SD3, LoRA, docs, and tests. The soft pairs share broad subsystems or filenames in a few cases, but they do not look like the same underlying change or bug, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:4783", + "right": "pull_request:6864", + "accept": false, + "reason": "Both touch SDXL/ControlNet docs, but one adds embedding-forcing docs and the other adds a torch.compile note for PEFT/LoRA; different user problems and code paths." + }, + { + "left": "pull_request:10192", + "right": "pull_request:9985", + "accept": false, + "reason": "Both are Flux-related, but 10192 adds Flux Control Inpaint docs/pipeline while 9985 adds Flux Fill/Canny/Depth/Redux; separate model families and PR scopes." + }, + { + "left": "pull_request:11451", + "right": "pull_request:11512", + "accept": false, + "reason": "Both are torch.compile test work, but they target different transformer models and different test implementations; not the same bug fix." + }, + { + "left": "pull_request:12225", + "right": "pull_request:12301", + "accept": false, + "reason": "Both are Qwen-Image inpaint variants, but one is edit-inpaint and the other is ControlNet-inpaint; distinct pipelines and behavior." + }, + { + "left": "pull_request:11039", + "right": "pull_request:11892", + "accept": false, + "reason": "One changes single-file loading/quantization handling for Kolors, the other removes k-diffusion from core dependencies; unrelated." + }, + { + "left": "pull_request:6795", + "right": "pull_request:8715", + "accept": false, + "reason": "Both touch SDXL DreamBooth training, but one adds micro-conditioning/crop args and the other adds clip_skip support; different training features." + }, + { + "left": "pull_request:8290", + "right": "pull_request:8627", + "accept": false, + "reason": "Both are broad Hunyuan/SD3-related refactors or docs/test sizing, but they address different model integration work and are not the same change." + }, + { + "left": "pull_request:11588", + "right": "pull_request:12074", + "accept": false, + "reason": "Both involve LoRA loading, but 11588 is Wan I2V LoRA support while 12074 adds Wan 2.2 second-transformer LoRA loading; different model behavior." + }, + { + "left": "pull_request:12909", + "right": "pull_request:12970", + "accept": false, + "reason": "Both mention context parallel, but one fixes Wan attention parallelism and the other fixes Qwen-Image series context parallel; different code paths and models." + }, + { + "left": "pull_request:10330", + "right": "pull_request:10968", + "accept": false, + "reason": "Both are LTX Video updates, but they are separate version bumps/releases with different converter/model changes; not the same underlying bug or feature." + }, + { + "left": "pull_request:10578", + "right": "pull_request:11655", + "accept": false, + "reason": "Both modify LoRA loading for quantized Flux, but one adds 4-bit support and the other adds 8-bit support; separate quantization cases." + }, + { + "left": "pull_request:10728", + "right": "pull_request:10867", + "accept": false, + "reason": "One removes a dtype reset in Flux IP-Adapter loading, the other adds multi-IP-Adapter support; different loader behaviors." + }, + { + "left": "pull_request:4451", + "right": "pull_request:4526", + "accept": false, + "reason": "Both rename slow tests to nightly, but they are different test batches and housekeeping changes, not one duplicate patch." + }, + { + "left": "pull_request:8725", + "right": "pull_request:8875", + "accept": false, + "reason": "Both are PAG-related, but one adds PAG for SD architecture and the other adds ControlNet PAG inpaint support; different pipelines and capabilities." + }, + { + "left": "pull_request:10351", + "right": "pull_request:11595", + "accept": false, + "reason": "One is a test fix in LoRA utilities, the other changes load_lora_weights behavior for Flux and updates a test; related area but different fixes." + }, + { + "left": "pull_request:10718", + "right": "pull_request:9987", + "accept": false, + "reason": "Both involve SD3/IP-Adapter, but 10718 is runtime checkpoint conversion for SD3 IP-Adapter while 9987 integrates SD3.5 IP-Adapter pipelines; not the same change." + }, + { + "left": "pull_request:4714", + "right": "pull_request:5506", + "accept": false, + "reason": "Both are small attention/utility fixes, but one changes dummy-object backend handling and the other fixes PEFT+xformers LoRA argument passing; unrelated bugs." + }, + { + "left": "pull_request:5915", + "right": "pull_request:7189", + "accept": false, + "reason": "Both concern IP-Adapter Plus, but 5915 adds the feature broadly while 7189 fixes repeated-embed handling across pipelines; not mergeable into one concrete fix." + }, + { + "left": "pull_request:10268", + "right": "pull_request:11550", + "accept": false, + "reason": "Both are single-file loader support, but one adds Mochi transformer support and the other adds GGUF/HiDream support; different models and loaders." + }, + { + "left": "pull_request:1063", + "right": "pull_request:948", + "accept": false, + "reason": "A variable-name typo fix in attention code is unrelated to a comment typo in DDPM; no shared bug." + }, + { + "left": "pull_request:3952", + "right": "pull_request:4145", + "accept": false, + "reason": "One is a temporary dependency pin for huggingface-hub, the other is a draft UnCLIP pipeline signature experiment; completely different and not duplicate-worthy." + }, + { + "left": "pull_request:5051", + "right": "pull_request:7131", + "accept": false, + "reason": "One fixes Wuerstchen docs typos, the other fixes broader docs typos and consistency decoder docs; same general docs area but not the same underlying change." + }, + { + "left": "pull_request:5697", + "right": "pull_request:6308", + "accept": false, + "reason": "Both are LoRA/PEFT related, but 5697 fixes xformers+PEFT attention scaling while 6308 fixes dummy components in LoRA tests; distinct issues and scopes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 16, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10005", + "pull_request:10188", + "pull_request:10246", + "pull_request:10255", + "pull_request:10279", + "pull_request:10316", + "pull_request:10323", + "pull_request:10351", + "pull_request:10465", + "pull_request:10524", + "pull_request:10854", + "pull_request:10858", + "pull_request:10911", + "pull_request:10995", + "pull_request:11505", + "pull_request:11655", + "pull_request:11786", + "pull_request:11823", + "pull_request:12041", + "pull_request:12562", + "pull_request:12660", + "pull_request:12737", + "pull_request:12739", + "pull_request:13212", + "pull_request:3847", + "pull_request:3921", + "pull_request:4276", + "pull_request:4632", + "pull_request:4847", + "pull_request:4944", + "pull_request:5887", + "pull_request:6215", + "pull_request:6550", + "pull_request:6576", + "pull_request:6729", + "pull_request:6759", + "pull_request:6864", + "pull_request:6939", + "pull_request:7246", + "pull_request:8630", + "pull_request:8685", + "pull_request:8691", + "pull_request:8875", + "pull_request:9086", + "pull_request:9290", + "pull_request:9463", + "pull_request:9751", + "pull_request:9827" + ], + "soft_pairs": [ + "pull_request:8630|pull_request:9086", + "pull_request:9086|pull_request:9290", + "pull_request:11655|pull_request:11823", + "pull_request:3921|pull_request:4847", + "pull_request:4944|pull_request:5887", + "pull_request:10351|pull_request:10911", + "pull_request:4632|pull_request:6759", + "pull_request:8691|pull_request:9827", + "pull_request:6729|pull_request:6939", + "pull_request:10465|pull_request:9751", + "pull_request:10854|pull_request:11786", + "pull_request:12739|pull_request:13212", + "pull_request:6550|pull_request:6576", + "pull_request:6864|pull_request:8685", + "pull_request:3847|pull_request:4276", + "pull_request:6215|pull_request:7246", + "pull_request:8875|pull_request:9463", + "pull_request:12041|pull_request:12562", + "pull_request:12660|pull_request:12737", + "pull_request:10005|pull_request:10246", + "pull_request:10188|pull_request:9827", + "pull_request:10255|pull_request:10995", + "pull_request:10279|pull_request:10858", + "pull_request:10316|pull_request:10524", + "pull_request:10323|pull_request:11505" + ], + "prepared_review_unit_hash": "52401c52269a91f0489cb0309dd82bd59d61f6351728f2a5b2fda870f7177236", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 96678, + "estimated_input_tokens": 24170, + "estimated_eval_tokens": 48596 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab-bag of superficially similar docs, tests, LoRA, and pipeline PRs, but the soft edges do not indicate true duplicates. The paired items usually touch different models, different docs pages, or different bug fixes/features within the same broad subsystem.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:8630", + "right": "pull_request:9086", + "accept": false, + "reason": "Both mention FLUX/DreamBooth LoRA, but one adds SD3 text-encoder training while the other adds Flux DreamBooth/LoRA scripts and README; different models and changes." + }, + { + "left": "pull_request:9086", + "right": "pull_request:9290", + "accept": false, + "reason": "The shared README_flux.md is misleadingly similar, but 9086 introduces Flux training scripts while 9290 only improves documentation links/guidance." + }, + { + "left": "pull_request:11655", + "right": "pull_request:11823", + "accept": false, + "reason": "Both modify lora_pipeline.py, but 11655 adds bitsandbytes 8-bit support and 11823 adds Fal Kontext LoRA conversion; different code paths and features." + }, + { + "left": "pull_request:3921", + "right": "pull_request:4847", + "accept": false, + "reason": "Both are community examples, but they implement unrelated pipelines/features: blended text inpainting vs prompt-free diffusion." + }, + { + "left": "pull_request:4944", + "right": "pull_request:5887", + "accept": false, + "reason": "Both are about IP-Adapter, but 4944 is a broad support implementation in loaders/model code while 5887 is a specific ControlNet inpaint pipeline integration." + }, + { + "left": "pull_request:10351", + "right": "pull_request:10911", + "accept": false, + "reason": "Both are LoRA test fixes, but they target different models and failures (generic test utility vs Lumina2-specific fuse_nan behavior)." + }, + { + "left": "pull_request:4632", + "right": "pull_request:6759", + "accept": false, + "reason": "One fixes SDXL LoRA training checkpointing/casting, the other fixes InstantID prompt image embedding shape; unrelated bugs." + }, + { + "left": "pull_request:8691", + "right": "pull_request:9827", + "accept": false, + "reason": "Both concern LoRA loading, but 8691 changes vanilla LoRA loading semantics while 9827 fixes device_map/offloading behavior during distributed inference." + }, + { + "left": "pull_request:6729", + "right": "pull_request:6939", + "accept": false, + "reason": "Both standardize model cards, but they apply to different training scripts and are separate documentation refactors." + }, + { + "left": "pull_request:10465", + "right": "pull_request:9751", + "accept": false, + "reason": "Both are NPU/FLUX related, but 10465 removes/adjusts training-script NPU flags while 9751 adds NPU adaptations for FLUX training runtime; not the same concrete fix." + }, + { + "left": "pull_request:10854", + "right": "pull_request:11786", + "accept": false, + "reason": "Different problem classes: Hunyuan DiT test module rename versus a broad CI marker overhaul across many accelerator tests." + }, + { + "left": "pull_request:12739", + "right": "pull_request:13212", + "accept": false, + "reason": "Both are Z-Image support fixes, but 12739 is a type-check/export shim and 13212 changes classifier-free guidance thresholds across pipelines; different defects." + }, + { + "left": "pull_request:6550", + "right": "pull_request:6576", + "accept": false, + "reason": "Both are compile-compatibility training-script edits, but they target different scripts and distinct model workflows (text-to-image vs InstructPix2Pix SDXL)." + }, + { + "left": "pull_request:6864", + "right": "pull_request:8685", + "accept": false, + "reason": "Both are docs updates, but one adds a PEFT/torch.compile note and the other adds an SD3 float8 checkpoint note; no shared underlying change." + }, + { + "left": "pull_request:3847", + "right": "pull_request:4276", + "accept": false, + "reason": "The first adds a dependency-check workflow and soft-dependency test, while the second adds a pipeline-import test; related area, but not the same bug/change." + }, + { + "left": "pull_request:6215", + "right": "pull_request:7246", + "accept": false, + "reason": "Both touch tests, but one fixes PixArt test determinism and the other fixes VAE tiling device selection; unrelated failures." + }, + { + "left": "pull_request:8875", + "right": "pull_request:9463", + "accept": false, + "reason": "Both add PAG support, but for different pipeline variants: ControlNet inpaint vs SD img2img. Not the same concrete feature." + }, + { + "left": "pull_request:12041", + "right": "pull_request:12562", + "accept": false, + "reason": "WAN VAE patchify logic and Wan context-parallel crash handling are different code paths and problems." + }, + { + "left": "pull_request:12660", + "right": "pull_request:12737", + "accept": false, + "reason": "Both are context-parallel fixes, but for different models and bugs: ChronoEdit attention behavior vs FLUX.2 input sharding metadata." + }, + { + "left": "pull_request:10005", + "right": "pull_request:10246", + "accept": false, + "reason": "Both are docs-only, but one adds missing Flux pipelines and the other adds missing attention processors; unrelated content." + }, + { + "left": "pull_request:10188", + "right": "pull_request:9827", + "accept": false, + "reason": "One fixes LoRA expansion error messaging and quantized-weight handling, the other fixes offloading/device_map loading; different failures." + }, + { + "left": "pull_request:10255", + "right": "pull_request:10995", + "accept": false, + "reason": "Both tweak LoRA loader comments, but they are distinct comment corrections in different contexts and not the same underlying code change." + }, + { + "left": "pull_request:10279", + "right": "pull_request:10858", + "accept": false, + "reason": "Quantization docs link fixes versus a pipelines overview classification fix; unrelated docs updates." + }, + { + "left": "pull_request:10316", + "right": "pull_request:10524", + "accept": false, + "reason": "Both update docs, but for different pipelines and different issues: removing a bad generator arg from LTX and correcting Hunyuan video checkpoint info." + }, + { + "left": "pull_request:10323", + "right": "pull_request:11505", + "accept": false, + "reason": "Both are link-fix docs PRs, but they modify different pages and different link targets; not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 17, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10394", + "pull_request:10418", + "pull_request:10434", + "pull_request:10632", + "pull_request:10845", + "pull_request:10978", + "pull_request:11537", + "pull_request:1178", + "pull_request:11825", + "pull_request:11954", + "pull_request:12827", + "pull_request:12971", + "pull_request:13071", + "pull_request:13082", + "pull_request:13253", + "pull_request:13346", + "pull_request:1437", + "pull_request:1967", + "pull_request:2381", + "pull_request:2391", + "pull_request:2541", + "pull_request:2566", + "pull_request:2677", + "pull_request:2678", + "pull_request:2764", + "pull_request:2825", + "pull_request:2839", + "pull_request:296", + "pull_request:3132", + "pull_request:3170", + "pull_request:3226", + "pull_request:3370", + "pull_request:3389", + "pull_request:3440", + "pull_request:3577", + "pull_request:3633", + "pull_request:3839", + "pull_request:4145", + "pull_request:4258", + "pull_request:4317", + "pull_request:4330", + "pull_request:4350", + "pull_request:4540", + "pull_request:4946", + "pull_request:5101", + "pull_request:948", + "pull_request:9655" + ], + "soft_pairs": [ + "pull_request:10394|pull_request:9655", + "pull_request:10418|pull_request:10434", + "pull_request:10632|pull_request:9655", + "pull_request:10845|pull_request:10978", + "pull_request:11537|pull_request:11954", + "pull_request:1178|pull_request:948", + "pull_request:11825|pull_request:11954", + "pull_request:12827|pull_request:12971", + "pull_request:13071|pull_request:13346", + "pull_request:13082|pull_request:13253", + "pull_request:1437|pull_request:948", + "pull_request:1967|pull_request:3170", + "pull_request:2381|pull_request:296", + "pull_request:2391|pull_request:2678", + "pull_request:2541|pull_request:2566", + "pull_request:2677|pull_request:3226", + "pull_request:2764|pull_request:3370", + "pull_request:2825|pull_request:4317", + "pull_request:2839|pull_request:4540", + "pull_request:3132|pull_request:3839", + "pull_request:3389|pull_request:3440", + "pull_request:3577|pull_request:5101", + "pull_request:3633|pull_request:4145", + "pull_request:4258|pull_request:4330", + "pull_request:4350|pull_request:4946" + ], + "prepared_review_unit_hash": "bcfc328e3aca427bf02b465f0af5c4c018df09f62a2024003543fae038d7f506", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 90256, + "estimated_input_tokens": 22564, + "estimated_eval_tokens": 45384 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a collection of unrelated docs/test/maintenance PRs. The only clear duplicate-like pair is the tensorboard tracker fix in `examples/unconditional_image_generation/train_unconditional.py` (2541 vs 2566), which is the same underlying code-path change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10434 is the strongest representative overall: it is merged, has an explicit issue target, and makes a broad, coherent docs link fix across many files.", + "best_issue_reason": null, + "best_pr_reason": "PR 10434 is the most representative merged PR in the set, with explicit issue linkage and the clearest substantive scope.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10394", + "right": "pull_request:9655", + "accept": false, + "reason": "Both are docs link/image fixes, but they target different pages and different broken references; not the same underlying change." + }, + { + "left": "pull_request:10418", + "right": "pull_request:10434", + "accept": false, + "reason": "Both touch the same tutorial/docs area, but one is a single internal link fix while the other is a broader broken-links sweep across many docs files." + }, + { + "left": "pull_request:10632", + "right": "pull_request:9655", + "accept": false, + "reason": "Different docs pages and different asset paths; no shared bug beyond generic link cleanup." + }, + { + "left": "pull_request:10845", + "right": "pull_request:10978", + "accept": false, + "reason": "Both are Flux LoRA-related, but one fixes training embedding handling and the other fixes LoRA conversion/loading for OneTrainer; different code paths." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11954", + "accept": false, + "reason": "Unrelated test additions for different models and failure modes; not mergeable as one PR." + }, + { + "left": "pull_request:1178", + "right": "pull_request:948", + "accept": false, + "reason": "These are independent typo fixes in different pipeline files." + }, + { + "left": "pull_request:11825", + "right": "pull_request:11954", + "accept": false, + "reason": "Compilation/hotswapping docs+tests versus Hunyuan video slice tests; different concerns and code paths." + }, + { + "left": "pull_request:12827", + "right": "pull_request:12971", + "accept": false, + "reason": "Same doc file, but one fixes code-block syntax while the other removes/repositions a section; not the same underlying bug/change." + }, + { + "left": "pull_request:13071", + "right": "pull_request:13346", + "accept": false, + "reason": "Both are test refactors, but for different model families and different test scopes." + }, + { + "left": "pull_request:13082", + "right": "pull_request:13253", + "accept": false, + "reason": "Different transformer test refactors for different models; no common concrete bug." + }, + { + "left": "pull_request:1437", + "right": "pull_request:948", + "accept": false, + "reason": "Both are typo fixes, but in different pipelines and unrelated text." + }, + { + "left": "pull_request:1967", + "right": "pull_request:3170", + "accept": false, + "reason": "One is a runtime code change, the other is a dependency/version pin; completely different issues." + }, + { + "left": "pull_request:2381", + "right": "pull_request:296", + "accept": false, + "reason": "Dummy-import packaging change versus auth-token test fixes; unrelated." + }, + { + "left": "pull_request:2391", + "right": "pull_request:2678", + "accept": false, + "reason": "Different test fixes in different areas; one is slow-test data, the other is EMA serialization behavior." + }, + { + "left": "pull_request:2541", + "right": "pull_request:2566", + "accept": true, + "reason": "Same example file and same exact fix: use `accelerator.get_tracker(\"tensorboard\")` in the non-new-API branch. This is the same underlying bug/change." + }, + { + "left": "pull_request:2677", + "right": "pull_request:3226", + "accept": false, + "reason": "Both involve ControlNet docs, but one adds training docs entries while the other documents ControlNet v1.1; different deliverables." + }, + { + "left": "pull_request:2764", + "right": "pull_request:3370", + "accept": false, + "reason": "Both edit `torch2.0.mdx`, but they update different claims and sections; not the same change." + }, + { + "left": "pull_request:2825", + "right": "pull_request:4317", + "accept": false, + "reason": "Different DreamBooth example fixes with different behavior changes and files; not duplicates." + }, + { + "left": "pull_request:2839", + "right": "pull_request:4540", + "accept": false, + "reason": "Both update examples docs, but one is README table expansion and the other is an SDXL Colab notebook link addition." + }, + { + "left": "pull_request:3132", + "right": "pull_request:3839", + "accept": false, + "reason": "One corrects a constructor type annotation, the other changes SDE noise handling logic; unrelated." + }, + { + "left": "pull_request:3389", + "right": "pull_request:3440", + "accept": false, + "reason": "Both touch CI/workflows, but they are different test-intent changes and not the same fix." + }, + { + "left": "pull_request:3577", + "right": "pull_request:5101", + "accept": false, + "reason": "Both are documentation link updates, but for different docs pages and different broken targets." + }, + { + "left": "pull_request:3633", + "right": "pull_request:4145", + "accept": false, + "reason": "Dependency pin test versus an UnCLIP pipeline constructor/comment change; unrelated." + }, + { + "left": "pull_request:4258", + "right": "pull_request:4330", + "accept": false, + "reason": "Different README_sdxl doc edits with different content changes, not the same underlying issue." + }, + { + "left": "pull_request:4350", + "right": "pull_request:4946", + "accept": false, + "reason": "Adapter dtype fix versus adding a training overview entry; unrelated docs changes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The one accepted duplicate-like pair (2541 vs 2566) is supported by the identical tracker call fix in the same file, and the rejected pairs are plausibly separate bugs/edits. No overstatement stands out." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 18, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11532", + "pull_request:11823", + "pull_request:12170", + "pull_request:12269", + "pull_request:2602", + "pull_request:3127", + "pull_request:3128", + "pull_request:4513", + "pull_request:4574", + "pull_request:4596", + "pull_request:4607", + "pull_request:4707", + "pull_request:4717", + "pull_request:4795", + "pull_request:4946", + "pull_request:5298", + "pull_request:5626", + "pull_request:5697", + "pull_request:5750", + "pull_request:5871", + "pull_request:6017", + "pull_request:6018", + "pull_request:6100", + "pull_request:6106", + "pull_request:6122", + "pull_request:6209", + "pull_request:6225", + "pull_request:6282", + "pull_request:6403", + "pull_request:6405", + "pull_request:6604", + "pull_request:6729", + "pull_request:6941", + "pull_request:7118", + "pull_request:7150", + "pull_request:7260", + "pull_request:7360", + "pull_request:7451", + "pull_request:7495", + "pull_request:7542", + "pull_request:7613", + "pull_request:7827", + "pull_request:7910", + "pull_request:8775", + "pull_request:9005", + "pull_request:9152", + "pull_request:9255", + "pull_request:9274" + ], + "soft_pairs": [ + "pull_request:4513|pull_request:5871", + "pull_request:4574|pull_request:6209", + "pull_request:4596|pull_request:5750", + "pull_request:4607|pull_request:4946", + "pull_request:4707|pull_request:6106", + "pull_request:4707|pull_request:6282", + "pull_request:4717|pull_request:6604", + "pull_request:5626|pull_request:7827", + "pull_request:6017|pull_request:6405", + "pull_request:6018|pull_request:6405", + "pull_request:6100|pull_request:6403", + "pull_request:6122|pull_request:7451", + "pull_request:6209|pull_request:7495", + "pull_request:6941|pull_request:7260", + "pull_request:7150|pull_request:9152", + "pull_request:7360|pull_request:7910", + "pull_request:7542|pull_request:7827", + "pull_request:8775|pull_request:9152", + "pull_request:9005|pull_request:9274", + "pull_request:9255|pull_request:9274", + "pull_request:11532|pull_request:12170", + "pull_request:6729|pull_request:7613", + "pull_request:2602|pull_request:4513", + "pull_request:3127|pull_request:3128", + "pull_request:4795|pull_request:5298", + "pull_request:6122|pull_request:7118", + "pull_request:5697|pull_request:6225", + "pull_request:11823|pull_request:12269" + ], + "prepared_review_unit_hash": "f3d9b78b01899e6c3ebba1d5d1e723d021be0ddb65f164997131d126d9b7dea3", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 96264, + "estimated_input_tokens": 24066, + "estimated_eval_tokens": 48388 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of unrelated PRs that only look similar at a high level (LoRA/docs/CI/offload/ControlNet themes). None of the soft edges appear to be the same underlying bug or change, so all should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out: the items span unrelated bug fixes, docs updates, feature work, and CI changes. Similar filenames or subsystem names are too broad to treat as duplicates.", + "best_issue_reason": null, + "best_pr_reason": "No globally best PR for deduping this cluster; the overlap is superficial and not centered on one concrete code-path change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4513", + "right": "pull_request:5871", + "accept": false, + "reason": "Both touch packaging/dependency surfaces, but one is a safetensors pre-release pin and the other is a deprecated scheduler rebase; unrelated changes." + }, + { + "left": "pull_request:4574", + "right": "pull_request:6209", + "accept": false, + "reason": "Both are example README edits, but they address different examples and different fixes: push_to_hub mentions vs a text-to-image test script/import cleanup." + }, + { + "left": "pull_request:4596", + "right": "pull_request:5750", + "accept": false, + "reason": "One adds an SDXL conversion script; the other is a doc-intro placeholder with no shared bug or code path." + }, + { + "left": "pull_request:4607", + "right": "pull_request:4946", + "accept": false, + "reason": "Both are docs overviews, but one adds PushToHubMixin API docs and the other updates the training overview list; not the same change." + }, + { + "left": "pull_request:4707", + "right": "pull_request:6106", + "accept": false, + "reason": "Same repository area only at a broad README level; one fixes custom diffusion training instructions, the other fixes an advanced dreambooth example path/reference." + }, + { + "left": "pull_request:4707", + "right": "pull_request:6282", + "accept": false, + "reason": "Custom diffusion README edits are unrelated to the dreambooth runtime-error fix; different scripts and problems." + }, + { + "left": "pull_request:4717", + "right": "pull_request:6604", + "accept": false, + "reason": "One is a ControlNet docs typo fix, the other exposes AutoencoderKL API entries; no shared bug or change." + }, + { + "left": "pull_request:5626", + "right": "pull_request:7827", + "accept": false, + "reason": "Both mention CPU offload, but one fixes a test assertion and the other updates a docstring example; not the same code-path problem." + }, + { + "left": "pull_request:6017", + "right": "pull_request:6405", + "accept": false, + "reason": "Both are doc-builder tests, but these are different WIP workflow/doc-builder experiments with different files and intent." + }, + { + "left": "pull_request:6018", + "right": "pull_request:6405", + "accept": false, + "reason": "Same as above: unrelated doc-builder test runs, not one concrete change." + }, + { + "left": "pull_request:6100", + "right": "pull_request:6403", + "accept": false, + "reason": "Both edit the same advanced training script, but one adds widget text while the other adds WebUI format support; distinct features, not duplicates." + }, + { + "left": "pull_request:6122", + "right": "pull_request:7451", + "accept": false, + "reason": "Both are large docs typo sweeps, but they target different document sets and corrections; same class of cleanup, not the same change." + }, + { + "left": "pull_request:6209", + "right": "pull_request:7495", + "accept": false, + "reason": "Different README fixes for different examples: one repairs a test script in text-to-image docs, the other fixes Colab links in advanced training docs." + }, + { + "left": "pull_request:6941", + "right": "pull_request:7260", + "accept": false, + "reason": "Both add IP-Adapter support, but to different pipelines and code paths (SDXL ControlNet inpaint vs SAG pipeline), so they are not mergeable as one PR." + }, + { + "left": "pull_request:7150", + "right": "pull_request:9152", + "accept": false, + "reason": "One updates a callback code example; the other expands distributed inference docs. They are unrelated documentation changes." + }, + { + "left": "pull_request:7360", + "right": "pull_request:7910", + "accept": false, + "reason": "TCD scheduler docstring/type cleanup and image-processor API docs are separate features with no shared underlying bug." + }, + { + "left": "pull_request:7542", + "right": "pull_request:7827", + "accept": false, + "reason": "A CPU offload test typo fix and a docstring CPU-offload example fix are not the same concrete issue." + }, + { + "left": "pull_request:8775", + "right": "pull_request:9152", + "accept": false, + "reason": "Both are docs updates, but one is a Colab link fix for SD3 training and the other is distributed inference guidance; unrelated." + }, + { + "left": "pull_request:9005", + "right": "pull_request:9274", + "accept": false, + "reason": "Both touch CI workflows, but one changes runner configuration across many workflows while the other adds a release-test workflow; different operational changes." + }, + { + "left": "pull_request:9255", + "right": "pull_request:9274", + "accept": false, + "reason": "Both involve push/release test workflows, but one modifies branch triggers for push tests and the other creates a separate release workflow; not one shared fix." + }, + { + "left": "pull_request:11532", + "right": "pull_request:12170", + "accept": false, + "reason": "Both are LoRA conversion helpers, but they target different model families and conversion logic (HiDream vs Qwen)." + }, + { + "left": "pull_request:6729", + "right": "pull_request:7613", + "accept": false, + "reason": "Both standardize model cards, but for different training scripts and different card templates; too broad to be duplicates." + }, + { + "left": "pull_request:2602", + "right": "pull_request:4513", + "accept": false, + "reason": "Different dependency pins for different packages and purposes; no shared underlying bug or change." + }, + { + "left": "pull_request:3127", + "right": "pull_request:3128", + "accept": false, + "reason": "Both touch img2img-related code/tests, but one fixes safety-checker output handling and the other fixes timestep selection; distinct bugs." + }, + { + "left": "pull_request:4795", + "right": "pull_request:5298", + "accept": false, + "reason": "Both address torch.compile compatibility, but in different modules and mechanisms (ControlNet forward logic vs LoRA conv forward implementation)." + }, + { + "left": "pull_request:6122", + "right": "pull_request:7118", + "accept": false, + "reason": "A broad typo sweep and a smaller docs typo fix are the same class of maintenance, but not the same concrete change." + }, + { + "left": "pull_request:5697", + "right": "pull_request:6225", + "accept": false, + "reason": "Both concern LoRA/PEFT, but one fixes xformers interaction and the other fixes alpha parsing; separate bugs." + }, + { + "left": "pull_request:11823", + "right": "pull_request:12269", + "accept": false, + "reason": "Both mention Flux Kontext, but one adds LoRA conversion support and the other adds modular pipeline support; different features and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 19, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10494", + "pull_request:10759", + "pull_request:10943", + "pull_request:10951", + "pull_request:11025", + "pull_request:11059", + "pull_request:11522", + "pull_request:11835", + "pull_request:11994", + "pull_request:12074", + "pull_request:12395", + "pull_request:12397", + "pull_request:1937", + "pull_request:3080", + "pull_request:3587", + "pull_request:399", + "pull_request:4278", + "pull_request:4595", + "pull_request:4596", + "pull_request:5033", + "pull_request:5298", + "pull_request:5750", + "pull_request:5914", + "pull_request:6140", + "pull_request:6217", + "pull_request:6550", + "pull_request:6573", + "pull_request:6577", + "pull_request:6729", + "pull_request:6908", + "pull_request:6909", + "pull_request:6910", + "pull_request:6946", + "pull_request:7189", + "pull_request:7226", + "pull_request:7467", + "pull_request:7919", + "pull_request:7934", + "pull_request:821", + "pull_request:8422", + "pull_request:8932", + "pull_request:9086", + "pull_request:9160", + "pull_request:9601", + "pull_request:9682", + "pull_request:9827", + "pull_request:9993" + ], + "soft_pairs": [ + "pull_request:6946|pull_request:7189", + "pull_request:12395|pull_request:12397", + "pull_request:5914|pull_request:7467", + "pull_request:6140|pull_request:7226", + "pull_request:6550|pull_request:6577", + "pull_request:6729|pull_request:6910", + "pull_request:9086|pull_request:9160", + "pull_request:11835|pull_request:11994", + "pull_request:4595|pull_request:5750", + "pull_request:4278|pull_request:5033", + "pull_request:6573|pull_request:7189", + "pull_request:9682|pull_request:9993", + "pull_request:6729|pull_request:6908", + "pull_request:6729|pull_request:6909", + "pull_request:6573|pull_request:6946", + "pull_request:11025|pull_request:12074", + "pull_request:1937|pull_request:3080", + "pull_request:7919|pull_request:9827", + "pull_request:7934|pull_request:9601", + "pull_request:8422|pull_request:8932", + "pull_request:10494|pull_request:11522", + "pull_request:3587|pull_request:4596", + "pull_request:5298|pull_request:6217", + "pull_request:10943|pull_request:11059", + "pull_request:10759|pull_request:10951", + "pull_request:399|pull_request:821" + ], + "prepared_review_unit_hash": "00dcca80d4e4eea5709bfcc211554673a815c8ab16011015857fe1b49b157559", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 102412, + "estimated_input_tokens": 25603, + "estimated_eval_tokens": 51462 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated PRs across different features, fixes, tests, and docs. The soft-similarity pairs share broad subsystems or filenames, but they do not describe the same concrete bug/change, so none should be merged as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:6946", + "right": "pull_request:7189", + "accept": false, + "reason": "Both touch IP-Adapter loading, but one adds low_cpu_mem_usage support and the other fixes plus/embeds repetition logic across several pipelines; different bugs." + }, + { + "left": "pull_request:12395", + "right": "pull_request:12397", + "accept": false, + "reason": "Both edit CI workflows, but one installs prerelease huggingface_hub for transformers-from-main and the other disables those installs entirely; opposite maintenance actions." + }, + { + "left": "pull_request:5914", + "right": "pull_request:7467", + "accept": false, + "reason": "Both affect SDXL LoRA training docs/scripts, but one fixes a model card template and the other changes tag shuffling/drop behavior; unrelated changes." + }, + { + "left": "pull_request:6140", + "right": "pull_request:7226", + "accept": false, + "reason": "Both touch IP-Adapter docs, but one adds API docs and the other adds image embedding/loader behavior; not the same underlying issue." + }, + { + "left": "pull_request:6550", + "right": "pull_request:6577", + "accept": false, + "reason": "Both add torch.compile compatibility, but for different training scripts and model types (SDXL text-to-image vs T2I Adapter SDXL)." + }, + { + "left": "pull_request:6729", + "right": "pull_request:6910", + "accept": false, + "reason": "Both standardize model cards, but on different training scripts and components (dreambooth/custom diffusion vs ControlNet); different change sets." + }, + { + "left": "pull_request:9086", + "right": "pull_request:9160", + "accept": false, + "reason": "Both are FLUX DreamBooth-related, but one introduces the training scripts and the other adds latent caching/precision behavior; separate features." + }, + { + "left": "pull_request:11835", + "right": "pull_request:11994", + "accept": false, + "reason": "Both address test failures, but one fixes float16 CUDA test assertions and the other updates Hub auth/install docs and scripts; unrelated." + }, + { + "left": "pull_request:4595", + "right": "pull_request:5750", + "accept": false, + "reason": "The first is an abandoned WIP converter idea and the second is unrelated reference-only documentation; no shared concrete bug or change." + }, + { + "left": "pull_request:4278", + "right": "pull_request:5033", + "accept": false, + "reason": "Both fix prompt/image batch handling in inpaint/upscale pipelines, but they target different pipelines and failure modes; not the same bug." + }, + { + "left": "pull_request:6573", + "right": "pull_request:7189", + "accept": false, + "reason": "Both involve IP-Adapter embedding handling, but 6573 adds multi-IP-Adapter support broadly while 7189 fixes embed repetition shape issues for plus adapters; distinct changes." + }, + { + "left": "pull_request:9682", + "right": "pull_request:9993", + "accept": false, + "reason": "Both are SD3 DreamBooth-LoRA updates, but one adds latent caching and training script changes while the other fixes checkpoint saving for text encoders; different bugs." + }, + { + "left": "pull_request:6729", + "right": "pull_request:6908", + "accept": false, + "reason": "Shared model-card standardization theme, but one is for DreamBooth/custom diffusion and the other for ControlNet SDXL; not a duplicate." + }, + { + "left": "pull_request:6729", + "right": "pull_request:6909", + "accept": false, + "reason": "Same as above: different training script and framework (Flax ControlNet), so not the same change." + }, + { + "left": "pull_request:6573", + "right": "pull_request:6946", + "accept": false, + "reason": "Both touch IP-Adapter internals, but one is multi-adapter support across many pipelines and the other is low_cpu_mem_usage loading optimization; separate fixes." + }, + { + "left": "pull_request:11025", + "right": "pull_request:12074", + "accept": false, + "reason": "Both are Wan LoRA work, but 11025 adds i2v LoRA conversion/support and 12074 adds second-transformer loading plus LightX2V support; related area, different functionality." + }, + { + "left": "pull_request:1937", + "right": "pull_request:3080", + "accept": false, + "reason": "Both are conversion/save logic fixes, but for different code paths and eras: EMA extraction in SD conversion vs save_pretrained handling of None values." + }, + { + "left": "pull_request:7919", + "right": "pull_request:9827", + "accept": false, + "reason": "Both mention offload/device-map behavior, but 7919 fixes offload handling when device_map is enabled while 9827 is a broader open LoRA device_map loading fix; not the same patch." + }, + { + "left": "pull_request:7934", + "right": "pull_request:9601", + "accept": false, + "reason": "Both concern loader/config handling, but 7934 fixes single-file dict config loading and 9601 fixes VAE dtype during SD3 training with mixed precision; unrelated." + }, + { + "left": "pull_request:8422", + "right": "pull_request:8932", + "accept": false, + "reason": "Both add test coverage for subcomponents, but they cover different model families and test goals; not duplicates." + }, + { + "left": "pull_request:10494", + "right": "pull_request:11522", + "accept": false, + "reason": "Both are CI/test compatibility fixes, but one fixes tokenizers installation in LoRA tests and the other adapts AudioLDM2 to newer transformers API; different issues." + }, + { + "left": "pull_request:3587", + "right": "pull_request:4596", + "accept": false, + "reason": "Both are community/example additions, but they implement different pipelines and examples; no shared underlying bug or change." + }, + { + "left": "pull_request:5298", + "right": "pull_request:6217", + "accept": false, + "reason": "Both relate to torch.compile compatibility, but one fixes LoRA conv forward behavior and the other fixes SVD VAE introspection under compile; different code paths." + }, + { + "left": "pull_request:10943", + "right": "pull_request:11059", + "accept": false, + "reason": "Both are Wan LoRA support work, but 10943 adds Wan loader/model support and 11059 adjusts non-diffusers Wan T2V LoRA conversion logic; separate changes." + }, + { + "left": "pull_request:10759", + "right": "pull_request:10951", + "accept": false, + "reason": "Both update community docs/readme around pipelines, but they introduce different community pipelines and example code." + }, + { + "left": "pull_request:399", + "right": "pull_request:821", + "accept": false, + "reason": "Both are early pipeline additions, but one is the ONNX exporter/pipeline and the other is an all-in-one Stable Diffusion community pipeline; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 20, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10005", + "pull_request:1004", + "pull_request:10232", + "pull_request:1026", + "pull_request:10279", + "pull_request:10288", + "pull_request:10299", + "pull_request:10402", + "pull_request:1041", + "pull_request:10418", + "pull_request:10434", + "pull_request:10524", + "pull_request:1073", + "pull_request:10750", + "pull_request:10845", + "pull_request:1087", + "pull_request:10947", + "pull_request:11099", + "pull_request:11102", + "pull_request:11236", + "pull_request:11327", + "pull_request:1136", + "pull_request:11402", + "pull_request:11439", + "pull_request:11505", + "pull_request:11537", + "pull_request:11546", + "pull_request:1162", + "pull_request:1167", + "pull_request:11693", + "pull_request:11740", + "pull_request:11825", + "pull_request:11953", + "pull_request:12074", + "pull_request:12086", + "pull_request:121", + "pull_request:12337", + "pull_request:12454", + "pull_request:1346", + "pull_request:252", + "pull_request:2678", + "pull_request:2839", + "pull_request:3157", + "pull_request:3299", + "pull_request:72", + "pull_request:9175", + "pull_request:9282", + "pull_request:934" + ], + "soft_pairs": [ + "pull_request:10750|pull_request:9175", + "pull_request:11546|pull_request:12074", + "pull_request:10005|pull_request:9282", + "pull_request:1004|pull_request:72", + "pull_request:10232|pull_request:11236", + "pull_request:1026|pull_request:121", + "pull_request:1026|pull_request:252", + "pull_request:10279|pull_request:10434", + "pull_request:10288|pull_request:10947", + "pull_request:10299|pull_request:10524", + "pull_request:10402|pull_request:11505", + "pull_request:1041|pull_request:934", + "pull_request:10418|pull_request:11505", + "pull_request:1073|pull_request:2839", + "pull_request:10845|pull_request:11439", + "pull_request:1087|pull_request:2678", + "pull_request:11099|pull_request:11102", + "pull_request:11327|pull_request:11693", + "pull_request:1136|pull_request:3299", + "pull_request:11402|pull_request:12337", + "pull_request:11537|pull_request:11740", + "pull_request:11537|pull_request:11953", + "pull_request:1162|pull_request:3157", + "pull_request:1167|pull_request:2839", + "pull_request:11825|pull_request:11953", + "pull_request:12086|pull_request:12454", + "pull_request:121|pull_request:1346" + ], + "prepared_review_unit_hash": "e5bf91e81a3faad76bbac22e02ac893f54090fb72a2e743278ca5c48c657428b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 93580, + "estimated_input_tokens": 23395, + "estimated_eval_tokens": 47046 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs fixes, tests, CI workflow changes, and feature work across different models and subsystems. I do not see any true duplicate pairs among the soft edges.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are not one underlying change, but a collection of unrelated docs, test, CI, and feature PRs.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good cluster representative; PR 12086 is the largest code change, but it is not a duplicate of the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10750", + "right": "pull_request:9175", + "accept": false, + "reason": "Both are Flux ControlNet work, but one adds img2img support and the other adds Union ControlNet; different code paths and features." + }, + { + "left": "pull_request:11546", + "right": "pull_request:12074", + "accept": false, + "reason": "Different LoRA targets and model families: HunyuanVideo FramePack vs Wan 2.2; not the same change." + }, + { + "left": "pull_request:10005", + "right": "pull_request:9282", + "accept": false, + "reason": "Both are docs, but one adds missing flux pipelines and the other updates the pipeline table; different edits." + }, + { + "left": "pull_request:1004", + "right": "pull_request:72", + "accept": false, + "reason": "One refactors 1D blocks, the other adds tests for 1D up/downsample blocks; not the same underlying fix." + }, + { + "left": "pull_request:10232", + "right": "pull_request:11236", + "accept": false, + "reason": "Different Sana docs pages and different edits; no shared bug or change." + }, + { + "left": "pull_request:1026", + "right": "pull_request:121", + "accept": false, + "reason": "Different docs/example updates from different eras; not a duplicate change." + }, + { + "left": "pull_request:1026", + "right": "pull_request:252", + "accept": false, + "reason": "Both touch READMEs, but the content and target changes are unrelated." + }, + { + "left": "pull_request:10279", + "right": "pull_request:10434", + "accept": false, + "reason": "Both fix docs links, but on different pages with different broken targets and scope." + }, + { + "left": "pull_request:10288", + "right": "pull_request:10947", + "accept": false, + "reason": "Same utility file, but one adds Flux single-file support and the other adds SANA Transformer loading; different model support." + }, + { + "left": "pull_request:10299", + "right": "pull_request:10524", + "accept": false, + "reason": "Different docs corrections: GGUF generator wording vs HunyuanVideo checkpoint-id text." + }, + { + "left": "pull_request:10402", + "right": "pull_request:11505", + "accept": false, + "reason": "Both are link-fix docs PRs, but they target different docs sets and different broken links." + }, + { + "left": "pull_request:1041", + "right": "pull_request:934", + "accept": false, + "reason": "Different test-failure fixes in different parts of the repo and different root causes." + }, + { + "left": "pull_request:10418", + "right": "pull_request:11505", + "accept": false, + "reason": "Internal PEFT docs link fix vs broad invalid-links cleanup; unrelated changes." + }, + { + "left": "pull_request:1073", + "right": "pull_request:2839", + "accept": false, + "reason": "Dreambooth README typo cleanup vs examples README expansion; not the same issue." + }, + { + "left": "pull_request:10845", + "right": "pull_request:11439", + "accept": false, + "reason": "Different LoRA/validation bugs in different training scripts and model families." + }, + { + "left": "pull_request:1087", + "right": "pull_request:2678", + "accept": false, + "reason": "Both touch tests/config, but one fixes slow tests while the other changes EMA serialization behavior; unrelated." + }, + { + "left": "pull_request:11099", + "right": "pull_request:11102", + "accept": false, + "reason": "Same workflow file, but one gates jobs on quality and the other fixes dependency uninstalling; different CI changes." + }, + { + "left": "pull_request:11327", + "right": "pull_request:11693", + "accept": false, + "reason": "Both mention compilation in docs, but they cover different quantization/inference topics." + }, + { + "left": "pull_request:1136", + "right": "pull_request:3299", + "accept": false, + "reason": "README example-link cleanup vs LoRA training docs/example sync; different documentation problems." + }, + { + "left": "pull_request:11402", + "right": "pull_request:12337", + "accept": false, + "reason": "Flux docs example argument fix vs xformers attention enablement bug; different layers of the stack." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11740", + "accept": false, + "reason": "Both are tests about compilation, but one is UNet2DConditionModel and the other is GGUF compile/offload coverage." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11953", + "accept": false, + "reason": "Different test suites and different goals: torch.compile coverage vs fast slice tests for HiDream." + }, + { + "left": "pull_request:1162", + "right": "pull_request:3157", + "accept": false, + "reason": "Both are README/docs style changes, but one fixes community links and the other fixes a TensorRT device helper call." + }, + { + "left": "pull_request:1167", + "right": "pull_request:2839", + "accept": false, + "reason": "Intel XPU README note vs examples README table update; unrelated documentation changes." + }, + { + "left": "pull_request:11825", + "right": "pull_request:11953", + "accept": false, + "reason": "Compilation/hotswapping test work vs HiDream fast slice test; different features and code paths." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12454", + "accept": false, + "reason": "Both are Flux modular pipeline work, but one adds I2I support and the other adds Kontext modular I2I/T2I support; not the same PR." + }, + { + "left": "pull_request:121", + "right": "pull_request:1346", + "accept": false, + "reason": "Early README example snippets vs community example typo fix; unrelated changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 21, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1210", + "pull_request:13069", + "pull_request:13340", + "pull_request:1346", + "pull_request:1360", + "pull_request:1420", + "pull_request:1625", + "pull_request:1667", + "pull_request:1960", + "pull_request:2257", + "pull_request:2271", + "pull_request:2333", + "pull_request:2396", + "pull_request:2405", + "pull_request:252", + "pull_request:2543", + "pull_request:2595", + "pull_request:2602", + "pull_request:2678", + "pull_request:2699", + "pull_request:2839", + "pull_request:2897", + "pull_request:2974", + "pull_request:3068", + "pull_request:3080", + "pull_request:3170", + "pull_request:3187", + "pull_request:3587", + "pull_request:3633", + "pull_request:3642", + "pull_request:3680", + "pull_request:377", + "pull_request:3819", + "pull_request:386", + "pull_request:3952", + "pull_request:3989", + "pull_request:4113", + "pull_request:4175", + "pull_request:4465", + "pull_request:4595", + "pull_request:5039", + "pull_request:5675", + "pull_request:5986", + "pull_request:882", + "pull_request:897", + "pull_request:908", + "pull_request:930", + "pull_request:936" + ], + "soft_pairs": [ + "pull_request:1210|pull_request:2678", + "pull_request:13069|pull_request:13340", + "pull_request:1346|pull_request:252", + "pull_request:1360|pull_request:1625", + "pull_request:1420|pull_request:936", + "pull_request:1667|pull_request:882", + "pull_request:1667|pull_request:908", + "pull_request:1960|pull_request:2839", + "pull_request:1960|pull_request:3819", + "pull_request:2257|pull_request:930", + "pull_request:2271|pull_request:2678", + "pull_request:2333|pull_request:2699", + "pull_request:2396|pull_request:3989", + "pull_request:2405|pull_request:2602", + "pull_request:2543|pull_request:3187", + "pull_request:2595|pull_request:3080", + "pull_request:2897|pull_request:897", + "pull_request:2974|pull_request:3642", + "pull_request:3068|pull_request:5039", + "pull_request:3170|pull_request:3952", + "pull_request:3587|pull_request:4595", + "pull_request:3633|pull_request:4175", + "pull_request:3642|pull_request:4465", + "pull_request:3680|pull_request:5986", + "pull_request:377|pull_request:386", + "pull_request:4113|pull_request:5675" + ], + "prepared_review_unit_hash": "807d9dac6304aab199c6f06ae5cebf5b8e250416245e126ba03afbde4b5c663f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 94126, + "estimated_input_tokens": 23532, + "estimated_eval_tokens": 47320 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of unrelated docs, test, scheduler, and pipeline PRs. The soft-similarity matches are superficial (same area or similar wording) and do not indicate the same underlying bug/change, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear canonical PR: the items span many unrelated changes, so there is no single artifact that represents the cluster.", + "best_issue_reason": null, + "best_pr_reason": "No best PR overall; the set is heterogeneous and lacks one change that meaningfully subsumes the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1210", + "right": "pull_request:2678", + "accept": false, + "reason": "Both mention test fixes, but one changes Cycle Diffusion/schedulers and the other changes EMA serialization tests; unrelated concrete problems." + }, + { + "left": "pull_request:13069", + "right": "pull_request:13340", + "accept": false, + "reason": "Both are model test refactors, but for different models and test suites (QwenImage vs SD3); not the same change." + }, + { + "left": "pull_request:1346", + "right": "pull_request:252", + "accept": false, + "reason": "Both touch README/example text, but they fix different documentation content and examples; not a duplicate edit." + }, + { + "left": "pull_request:1360", + "right": "pull_request:1625", + "accept": false, + "reason": "Both are about k-diffusion, but one adds a community pipeline and the other changes default dimensions in the stable diffusion k-diffusion pipeline; different issues." + }, + { + "left": "pull_request:1420", + "right": "pull_request:936", + "accept": false, + "reason": "One adds SD2 v-prediction tests, the other reorganizes stable diffusion tests; same test area, not the same concrete work." + }, + { + "left": "pull_request:1667", + "right": "pull_request:882", + "accept": false, + "reason": "Scheduler-related, but one fixes conversion config for DDIM while the other prevents PNDM list growth; different bugs." + }, + { + "left": "pull_request:1667", + "right": "pull_request:908", + "accept": false, + "reason": "These target different problems: scheduler conversion config vs using the correct inpaint pipeline in the community mega example." + }, + { + "left": "pull_request:1960", + "right": "pull_request:2839", + "accept": false, + "reason": "Both update docs, but on different README files and unrelated content." + }, + { + "left": "pull_request:1960", + "right": "pull_request:3819", + "accept": false, + "reason": "Both are README/docs edits, but they address different documentation sections with no shared underlying fix." + }, + { + "left": "pull_request:2257", + "right": "pull_request:930", + "accept": false, + "reason": "One adds an UnCLIP text interpolation pipeline; the other updates docs/toctree structure. Not the same change." + }, + { + "left": "pull_request:2271", + "right": "pull_request:2678", + "accept": false, + "reason": "Both mention slow tests, but they affect different test files and different pipeline/model behavior." + }, + { + "left": "pull_request:2333", + "right": "pull_request:2699", + "accept": false, + "reason": "Same docs file, but the edits are different: formatting/preamble cleanup vs adding a safety checker entry." + }, + { + "left": "pull_request:2396", + "right": "pull_request:3989", + "accept": false, + "reason": "Both are test-related, but one fixes scheduler warnings/UniPC tests and the other fixes SDXL slow tests; unrelated." + }, + { + "left": "pull_request:2405", + "right": "pull_request:2602", + "accept": false, + "reason": "Both change version pins, but they target different release/testing versions and different intents; not the same patch." + }, + { + "left": "pull_request:2543", + "right": "pull_request:3187", + "accept": false, + "reason": "Both concern offload behavior, but 2543 is a broad warning/offload cleanup and 3187 is a specific x4 upscaler offload addition." + }, + { + "left": "pull_request:2595", + "right": "pull_request:3080", + "accept": false, + "reason": "Scheduler switching/config changes versus pipeline save behavior; completely different code paths." + }, + { + "left": "pull_request:2897", + "right": "pull_request:897", + "accept": false, + "reason": "One adds StableUnCLIP docs/examples, the other adds a speech-to-image community pipeline; only broad topical overlap." + }, + { + "left": "pull_request:2974", + "right": "pull_request:3642", + "accept": false, + "reason": "Both add TensorRT-related community pipelines, but they are different pipelines and implementations (txt2img/utility vs inpainting)." + }, + { + "left": "pull_request:3068", + "right": "pull_request:5039", + "accept": false, + "reason": "Both involve textual inversion docs, but one is for general training docs and the other is specifically SDXL textual inversion guidance." + }, + { + "left": "pull_request:3170", + "right": "pull_request:3952", + "accept": false, + "reason": "Both are dependency-pin test PRs, but for different hub release candidates and different versions; not duplicate." + }, + { + "left": "pull_request:3587", + "right": "pull_request:4595", + "accept": false, + "reason": "A community pipeline PR versus an unrelated WIP conversion PR; no shared concrete bug or feature." + }, + { + "left": "pull_request:3633", + "right": "pull_request:4175", + "accept": false, + "reason": "One is a test dependency pin, the other is a ControlNet webdataset training example; unrelated." + }, + { + "left": "pull_request:3642", + "right": "pull_request:4465", + "accept": false, + "reason": "Both mention TensorRT/controlnet/community examples, but they implement different pipelines and conversion tooling." + }, + { + "left": "pull_request:3680", + "right": "pull_request:5986", + "accept": false, + "reason": "Different documentation fixes in different files: loader-method link versus video link." + }, + { + "left": "pull_request:377", + "right": "pull_request:386", + "accept": false, + "reason": "Both are pipeline docs, but for different models (Latent Diffusion vs Stable Diffusion) and different content." + }, + { + "left": "pull_request:4113", + "right": "pull_request:5675", + "accept": false, + "reason": "Different bug fixes in different pipelines: SDXL inpaint dtype handling versus Flax NSFW-filter exception handling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 22, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4145", + "pull_request:4211", + "pull_request:4278", + "pull_request:4540", + "pull_request:4564", + "pull_request:4574", + "pull_request:4582", + "pull_request:4612", + "pull_request:4644", + "pull_request:4717", + "pull_request:4745", + "pull_request:4797", + "pull_request:4946", + "pull_request:5035", + "pull_request:5175", + "pull_request:5207", + "pull_request:5411", + "pull_request:5499", + "pull_request:5626", + "pull_request:5699", + "pull_request:5754", + "pull_request:5986", + "pull_request:6045", + "pull_request:6225", + "pull_request:6308", + "pull_request:6340", + "pull_request:6402", + "pull_request:6534", + "pull_request:6680", + "pull_request:6871", + "pull_request:7182", + "pull_request:7335", + "pull_request:7542", + "pull_request:7764", + "pull_request:7857", + "pull_request:7868", + "pull_request:8470", + "pull_request:8630", + "pull_request:869", + "pull_request:8715", + "pull_request:8720", + "pull_request:874", + "pull_request:8791", + "pull_request:903", + "pull_request:904", + "pull_request:9066", + "pull_request:9775" + ], + "soft_pairs": [ + "pull_request:4145|pull_request:5175", + "pull_request:4211|pull_request:4582", + "pull_request:4540|pull_request:4574", + "pull_request:4564|pull_request:4644", + "pull_request:4582|pull_request:4717", + "pull_request:4612|pull_request:5499", + "pull_request:4797|pull_request:6402", + "pull_request:4946|pull_request:5207", + "pull_request:5035|pull_request:6045", + "pull_request:5411|pull_request:5986", + "pull_request:5626|pull_request:7335", + "pull_request:5699|pull_request:5754", + "pull_request:6225|pull_request:6308", + "pull_request:6340|pull_request:7764", + "pull_request:6534|pull_request:8630", + "pull_request:6680|pull_request:8720", + "pull_request:6871|pull_request:8791", + "pull_request:7182|pull_request:8715", + "pull_request:7542|pull_request:7868", + "pull_request:7857|pull_request:9066", + "pull_request:8470|pull_request:9775", + "pull_request:869|pull_request:874", + "pull_request:903|pull_request:904", + "pull_request:4278|pull_request:4745" + ], + "prepared_review_unit_hash": "48a559d5e483985ed154c8b48baf404a74987e32ef3108cf5f051b3310999f1c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 95791, + "estimated_input_tokens": 23948, + "estimated_eval_tokens": 48152 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the soft-edge pairs are mostly unrelated docs, test tweaks, and separate bug fixes. I would reject all soft edges; none look like the same underlying change or bug.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out: the items span unrelated documentation edits, test adjustments, and distinct code fixes, so none safely represents the whole set.", + "best_issue_reason": null, + "best_pr_reason": "No clear global best PR: the candidates are too diverse to treat any one as the representative duplicate target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4145", + "right": "pull_request:5175", + "accept": false, + "reason": "UnCLIP constructor/comment cleanup vs CI HF_HOME env change; unrelated files and no shared bug." + }, + { + "left": "pull_request:4211", + "right": "pull_request:4582", + "accept": false, + "reason": "SDXL img2img/inpaint docstring wording vs ControlNet SDXL docstring example fix; different pipelines and changes." + }, + { + "left": "pull_request:4540", + "right": "pull_request:4574", + "accept": false, + "reason": "DreamBooth README Colab link vs InstructPix2Pix README mentions; different docs and purpose." + }, + { + "left": "pull_request:4564", + "right": "pull_request:4644", + "accept": false, + "reason": "Both touch the same SDXL ControlNet doc, but one fixes the model card link and the other expands checkpoint guidance; not the same change." + }, + { + "left": "pull_request:4582", + "right": "pull_request:4717", + "accept": false, + "reason": "ControlNet SDXL docstring example fix vs a training doc Tip tag fix; unrelated bug surfaces." + }, + { + "left": "pull_request:4612", + "right": "pull_request:5499", + "accept": false, + "reason": "Instruct Pix2Pix link correction vs broad doc link rewrites across many pages; same theme, different edits." + }, + { + "left": "pull_request:4797", + "right": "pull_request:6402", + "accept": false, + "reason": "LoRA SGM block mapping in loaders vs gradient-checkpointing handling in SDXL LoRA training; different code paths and bugs." + }, + { + "left": "pull_request:4946", + "right": "pull_request:5207", + "accept": false, + "reason": "Training overview entry addition vs a t2i adapter README filename fix; unrelated documentation changes." + }, + { + "left": "pull_request:5035", + "right": "pull_request:6045", + "accept": false, + "reason": "Optional transformers import cleanup vs PEFT linear layer behavior in LoRA training; different underlying issues." + }, + { + "left": "pull_request:5411", + "right": "pull_request:5986", + "accept": false, + "reason": "Broken textual inversion training link vs a video URL update in SVD docs; unrelated doc fixes." + }, + { + "left": "pull_request:5626", + "right": "pull_request:7335", + "accept": false, + "reason": "CPU offload test assertion cleanup vs ControlNet SDXL config/docs test changes; not the same failure." + }, + { + "left": "pull_request:5699", + "right": "pull_request:5754", + "accept": false, + "reason": "Generic style cleanup in pipeline code vs formatting cleanup in training scripts; not the same change." + }, + { + "left": "pull_request:6225", + "right": "pull_request:6308", + "accept": false, + "reason": "LoRA alpha parsing bug fix vs dummy test component setup; test support is not the same underlying fix." + }, + { + "left": "pull_request:6340", + "right": "pull_request:7764", + "accept": false, + "reason": "AnimateDiff conversion script addition vs later safetensor loading support in those scripts; related area but distinct changes." + }, + { + "left": "pull_request:6534", + "right": "pull_request:8630", + "accept": false, + "reason": "Torch.compile compatibility for DreamBooth LoRA vs SD3 CLIP text encoder training; different features and code paths." + }, + { + "left": "pull_request:6680", + "right": "pull_request:8720", + "accept": false, + "reason": "DPM slow-test stabilization vs precision-threshold updates in various slow tests; different test issues." + }, + { + "left": "pull_request:6871", + "right": "pull_request:8791", + "accept": false, + "reason": "VAE tiling/slicing for img2img and inpaint vs VAE tiling option for SD3; same broad feature class, but different pipeline targets." + }, + { + "left": "pull_request:7182", + "right": "pull_request:8715", + "accept": false, + "reason": "EDM-style DreamBooth training support vs adding clip_skip; unrelated advanced training enhancements." + }, + { + "left": "pull_request:7542", + "right": "pull_request:7868", + "accept": false, + "reason": "Two offload test fixes, but one is a typo correction and the other changes device assertions; not the same bug." + }, + { + "left": "pull_request:7857", + "right": "pull_request:9066", + "accept": false, + "reason": "PixArtSigma docs addition vs StableCascadeUNet docs addition; same toctree file but different docs pages." + }, + { + "left": "pull_request:8470", + "right": "pull_request:9775", + "accept": false, + "reason": "MaPO project README rewrite vs Flux LoRA quantization README note; unrelated repositories/examples." + }, + { + "left": "pull_request:869", + "right": "pull_request:874", + "accept": false, + "reason": "Community README/import bug fix vs a generic inference example addition; different changes with no shared bug." + }, + { + "left": "pull_request:903", + "right": "pull_request:904", + "accept": false, + "reason": "Inpainting pipeline deprecation vs the inpainting implementation itself; one removes/redirects, the other defines behavior." + }, + { + "left": "pull_request:4278", + "right": "pull_request:4745", + "accept": false, + "reason": "StableDiffusionUpscale prompt=None bug fix vs from_pretrained optional-components handling; different failure modes despite overlap in tests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 23, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10261", + "pull_request:10351", + "pull_request:10352", + "pull_request:10728", + "pull_request:11025", + "pull_request:11243", + "pull_request:11726", + "pull_request:12086", + "pull_request:12170", + "pull_request:12416", + "pull_request:12714", + "pull_request:13100", + "pull_request:13119", + "pull_request:13148", + "pull_request:3742", + "pull_request:3877", + "pull_request:4062", + "pull_request:4597", + "pull_request:4749", + "pull_request:4847", + "pull_request:4899", + "pull_request:5608", + "pull_request:5914", + "pull_request:5938", + "pull_request:6303", + "pull_request:6402", + "pull_request:6564", + "pull_request:6691", + "pull_request:6816", + "pull_request:7018", + "pull_request:7058", + "pull_request:7086", + "pull_request:7166", + "pull_request:7182", + "pull_request:7235", + "pull_request:7277", + "pull_request:7326", + "pull_request:7372", + "pull_request:8630", + "pull_request:8755", + "pull_request:9086", + "pull_request:9257", + "pull_request:9346", + "pull_request:9355", + "pull_request:9646", + "pull_request:9838", + "pull_request:9993" + ], + "soft_pairs": [ + "pull_request:4847|pull_request:6303", + "pull_request:9646|pull_request:9838", + "pull_request:12086|pull_request:12416", + "pull_request:6564|pull_request:7086", + "pull_request:3877|pull_request:4749", + "pull_request:10351|pull_request:10352", + "pull_request:12714|pull_request:13119", + "pull_request:6691|pull_request:7182", + "pull_request:6816|pull_request:8755", + "pull_request:7235|pull_request:9257", + "pull_request:10261|pull_request:10728", + "pull_request:8630|pull_request:9993", + "pull_request:3742|pull_request:4062", + "pull_request:11025|pull_request:11726", + "pull_request:4597|pull_request:5938", + "pull_request:11243|pull_request:12170", + "pull_request:13100|pull_request:13148", + "pull_request:5914|pull_request:6402", + "pull_request:7372|pull_request:9086", + "pull_request:9346|pull_request:9355", + "pull_request:7018|pull_request:7166", + "pull_request:7277|pull_request:7326", + "pull_request:4597|pull_request:4899", + "pull_request:5608|pull_request:7058" + ], + "prepared_review_unit_hash": "836dd27e31c33d959ad21715cc56d221e1112afbe12f10c8350a22f267525360", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 100507, + "estimated_input_tokens": 25127, + "estimated_eval_tokens": 50510 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a mixed bag of unrelated PRs that only overlap superficially by subsystem/name (LoRA, Flux, Shap-E, modular docs, tests). None of the soft edges look like true duplicates; most are separate bugs, follow-up enhancements, or test/doc-only changes.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR is a good duplicate representative here: the set is heterogeneous, and the apparent overlaps are mostly superficial or follow-up changes rather than the same underlying change.", + "best_issue_reason": null, + "best_pr_reason": "If one PR must serve as a representative anchor, #10261 is the strongest candidate because it is the broadest, most substantive feature PR in the group, has explicit issue links, and touches the main Flux IP-Adapter code path across multiple files. That said, it is still not a duplicate match for the rest of the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4847", + "right": "pull_request:6303", + "accept": false, + "reason": "Different projects and goals: Prompt Free Diffusion community example vs ADD training scripts." + }, + { + "left": "pull_request:9646", + "right": "pull_request:9838", + "accept": false, + "reason": "Both are Flux LoRA-related, but they fix different things: configurable target modules vs an advanced training memory/bug fix." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12416", + "accept": false, + "reason": "Unrelated modular-pipeline features for different model families (Flux I2I vs QwenImage Edit Plus)." + }, + { + "left": "pull_request:6564", + "right": "pull_request:7086", + "accept": false, + "reason": "Both mention IP-Adapter, but one fixes path handling while the other updates AnimateDiff img2vid/controlnet examples." + }, + { + "left": "pull_request:3877", + "right": "pull_request:4749", + "accept": false, + "reason": "Different DreamBooth LoRA changes: SDXL script addition vs checkpoint-loading fix." + }, + { + "left": "pull_request:10351", + "right": "pull_request:10352", + "accept": false, + "reason": "Same test-fix theme, but for different pipelines and different underlying issues." + }, + { + "left": "pull_request:12714", + "right": "pull_request:13119", + "accept": false, + "reason": "Flux2 test NaN handling vs Flux2 non-diffusers key conversion; related area, but not the same bug." + }, + { + "left": "pull_request:6691", + "right": "pull_request:7182", + "accept": false, + "reason": "Same advanced DreamBooth area, but one is a broad feature/bugfix bundle and the other is specifically EDM-style training support." + }, + { + "left": "pull_request:6816", + "right": "pull_request:8755", + "accept": false, + "reason": "Both involve SDXL/SD3 LoRA training, but they address different model families and different training bugs." + }, + { + "left": "pull_request:7235", + "right": "pull_request:9257", + "accept": false, + "reason": "DoRA training for SDXL DreamBooth vs a Flux DreamBooth script bug fix are unrelated changes." + }, + { + "left": "pull_request:10261", + "right": "pull_request:10728", + "accept": false, + "reason": "Same Flux/IP-Adapter neighborhood, but one adds support and the other removes a device/dtype call for quantized Flux." + }, + { + "left": "pull_request:8630", + "right": "pull_request:9993", + "accept": false, + "reason": "SD3 text-encoder training feature vs a later checkpoint-saving fix in the same script; distinct changes." + }, + { + "left": "pull_request:3742", + "right": "pull_request:4062", + "accept": false, + "reason": "4062 extends Shap-E with mesh output on top of 3742; related, but not the same underlying change." + }, + { + "left": "pull_request:11025", + "right": "pull_request:11726", + "accept": false, + "reason": "Both are Wan LoRA conversion work, but 11726 is a later extension for VACE variants, not a duplicate of the original support PR." + }, + { + "left": "pull_request:4597", + "right": "pull_request:5938", + "accept": false, + "reason": "MultiControlNet SDXL support and ControlNet LoRA support are different feature areas." + }, + { + "left": "pull_request:11243", + "right": "pull_request:12170", + "accept": false, + "reason": "Both add LoRA conversion support, but for different model families and different key-mapping logic (Wan vs Qwen)." + }, + { + "left": "pull_request:13100", + "right": "pull_request:13148", + "accept": false, + "reason": "Different modular-doc updates: PipelineState docs vs pipeline blocks docs." + }, + { + "left": "pull_request:5914", + "right": "pull_request:6402", + "accept": false, + "reason": "README template typo fix vs gradient-checkpointing being ignored in LoRA training are unrelated." + }, + { + "left": "pull_request:7372", + "right": "pull_request:9086", + "accept": false, + "reason": "Model-card metadata/scheduler-type work vs Flux DreamBooth LoRA scripts are different changes." + }, + { + "left": "pull_request:9346", + "right": "pull_request:9355", + "accept": false, + "reason": "Separate CI changes: GPU test fixes vs adding workflow dispatch to the push-tests workflow." + }, + { + "left": "pull_request:7018", + "right": "pull_request:7166", + "accept": false, + "reason": "Both are SDXL DreamBooth LoRA fixes, but they target different bugs: gradient unscaling vs loading the wrong text encoder." + }, + { + "left": "pull_request:7277", + "right": "pull_request:7326", + "accept": false, + "reason": "Deprecated-test cleanup and warning fixes are different maintenance changes, not the same bug." + }, + { + "left": "pull_request:5608", + "right": "pull_request:7058", + "accept": false, + "reason": "Both touch PEFT/LoRA loading, but one fixes empty network_alpha handling and the other prevents in-place mutation of passed state dicts." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 24, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10222", + "pull_request:10279", + "pull_request:10291", + "pull_request:10312", + "pull_request:10316", + "pull_request:10394", + "pull_request:10458", + "pull_request:10524", + "pull_request:10781", + "pull_request:10810", + "pull_request:1087", + "pull_request:10985", + "pull_request:11271", + "pull_request:11431", + "pull_request:11481", + "pull_request:11505", + "pull_request:11529", + "pull_request:11545", + "pull_request:11550", + "pull_request:11552", + "pull_request:11681", + "pull_request:12006", + "pull_request:12562", + "pull_request:1467", + "pull_request:2046", + "pull_request:2915", + "pull_request:3170", + "pull_request:4278", + "pull_request:4714", + "pull_request:4799", + "pull_request:5697", + "pull_request:5738", + "pull_request:5887", + "pull_request:6556", + "pull_request:6573", + "pull_request:689", + "pull_request:7072", + "pull_request:7898", + "pull_request:8406", + "pull_request:8630", + "pull_request:8816", + "pull_request:8820", + "pull_request:8916", + "pull_request:9408", + "pull_request:9463", + "pull_request:9517", + "pull_request:9598", + "pull_request:9705" + ], + "soft_pairs": [ + "pull_request:10312|pull_request:9517", + "pull_request:10781|pull_request:11550", + "pull_request:11431|pull_request:11552", + "pull_request:1467|pull_request:689", + "pull_request:2046|pull_request:3170", + "pull_request:4714|pull_request:5697", + "pull_request:6556|pull_request:8630", + "pull_request:7898|pull_request:8916", + "pull_request:8406|pull_request:9705", + "pull_request:4278|pull_request:4799", + "pull_request:11481|pull_request:11681", + "pull_request:5738|pull_request:7072", + "pull_request:8816|pull_request:9463", + "pull_request:8820|pull_request:9463", + "pull_request:12006|pull_request:12562", + "pull_request:5887|pull_request:6573", + "pull_request:10222|pull_request:11545", + "pull_request:10279|pull_request:10394", + "pull_request:10279|pull_request:11505", + "pull_request:10291|pull_request:9408", + "pull_request:10316|pull_request:11529", + "pull_request:10458|pull_request:9598", + "pull_request:10524|pull_request:11529", + "pull_request:10781|pull_request:11271", + "pull_request:10810|pull_request:10985", + "pull_request:1087|pull_request:2915", + "pull_request:1087|pull_request:689" + ], + "prepared_review_unit_hash": "1068e37fe745313bb10979987b0a4b7e2accd146287609287cb3981fc62cea2d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 97644, + "estimated_input_tokens": 24411, + "estimated_eval_tokens": 49078 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs edits, test updates, and feature/fix PRs across many different pipelines and models. The soft-similarity links are mostly superficial (same subsystem or same file), not the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out: there is no shared concrete code-path problem across the items, so picking one would be arbitrary.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR for duplicate triage: the set spans unrelated changes, so none is a strong representative canonical target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10312", + "right": "pull_request:9517", + "accept": false, + "reason": "Different bugs in different areas: Attention.head_to_batch_dim docs/behavior vs FLUX joint_attention_kwargs plumbing." + }, + { + "left": "pull_request:10781", + "right": "pull_request:11550", + "accept": false, + "reason": "Both add single-file support, but for different models/pipelines (Lumina2 vs HiDream) and separate integration paths." + }, + { + "left": "pull_request:11431", + "right": "pull_request:11552", + "accept": false, + "reason": "Different transformer test suites for different models; not the same bug or change." + }, + { + "left": "pull_request:1467", + "right": "pull_request:689", + "accept": false, + "reason": "Both mention slow tests, but one changes test tolerances while the other changes attention implementation performance." + }, + { + "left": "pull_request:2046", + "right": "pull_request:3170", + "accept": false, + "reason": "Both pin huggingface-hub RC versions, but they are separate dependency experiments with different versions and contexts." + }, + { + "left": "pull_request:4714", + "right": "pull_request:5697", + "accept": false, + "reason": "Distinct fixes in different code paths: import_utils error handling vs xformers/PEFT attention scaling." + }, + { + "left": "pull_request:6556", + "right": "pull_request:8630", + "accept": false, + "reason": "Both are LoRA training-related, but one is SDXL training script compile support and the other is SD3 text-encoder training." + }, + { + "left": "pull_request:7898", + "right": "pull_request:8916", + "accept": false, + "reason": "Different model tests and different scope; not the same test failure or model behavior." + }, + { + "left": "pull_request:8406", + "right": "pull_request:9705", + "accept": false, + "reason": "Both edit advanced-training READMEs, but they cover different training workflows and content." + }, + { + "left": "pull_request:4278", + "right": "pull_request:4799", + "accept": false, + "reason": "Both are pipeline bugfixes, but for different pipelines and different input-handling issues." + }, + { + "left": "pull_request:11481", + "right": "pull_request:11681", + "accept": false, + "reason": "Test renames/chore vs device_map documentation clarification; unrelated changes." + }, + { + "left": "pull_request:5738", + "right": "pull_request:7072", + "accept": false, + "reason": "LoRA adapter deletion feature vs DoRA training documentation/features; different functionality and code paths." + }, + { + "left": "pull_request:8816", + "right": "pull_request:9463", + "accept": false, + "reason": "Both add PAG support, but for different pipeline families (SD1.5 ControlNet vs SD img2img) and separate implementations." + }, + { + "left": "pull_request:8820", + "right": "pull_request:9463", + "accept": false, + "reason": "Same as above: ControlNet PAG support vs SD img2img PAG support are distinct features." + }, + { + "left": "pull_request:12006", + "right": "pull_request:12562", + "accept": false, + "reason": "Wan i2v feature addition vs context-parallel crash fix; not the same underlying issue." + }, + { + "left": "pull_request:5887", + "right": "pull_request:6573", + "accept": false, + "reason": "Both touch IP-Adapter/ControlNet areas, but one adds IP-Adapter to a specific inpaint pipeline while the other adds multi-IP-Adapter support across pipelines." + }, + { + "left": "pull_request:10222", + "right": "pull_request:11545", + "accept": false, + "reason": "Unrelated pipeline fixes: XLA support in ControlNet vs missing VAE slicing/tiling in FLUX img2img." + }, + { + "left": "pull_request:10279", + "right": "pull_request:10394", + "accept": false, + "reason": "Both are docs-only link fixes, but they target different docs and different broken links." + }, + { + "left": "pull_request:10279", + "right": "pull_request:11505", + "accept": false, + "reason": "Both are docs-only link fixes, but for different documentation sections and unrelated link targets." + }, + { + "left": "pull_request:10291", + "right": "pull_request:9408", + "accept": false, + "reason": "Both are FLUX ControlNet-related, but one fixes a config bug in inpainting while the other adds img2img/inpaint pipelines." + }, + { + "left": "pull_request:10316", + "right": "pull_request:11529", + "accept": false, + "reason": "One removes an outdated generator argument from a docs example; the other has no visible diff and no shared underlying change." + }, + { + "left": "pull_request:10458", + "right": "pull_request:9598", + "accept": false, + "reason": "Test markers for SD3 LoRA integration vs a note on Flux LoRA test environment versions; unrelated maintenance." + }, + { + "left": "pull_request:10524", + "right": "pull_request:11529", + "accept": false, + "reason": "Checkpoint-id docs correction vs an empty docs update; no shared bug or change." + }, + { + "left": "pull_request:10781", + "right": "pull_request:11271", + "accept": false, + "reason": "Both are single-file support work, but for different model families (Lumina2/HiDream vs LTX 0.9.5) and different loader logic." + }, + { + "left": "pull_request:10810", + "right": "pull_request:10985", + "accept": false, + "reason": "Both modify LoRA conversion utilities, but they address different conversion problems and broader Flux LoRA support is not the same concrete fix." + }, + { + "left": "pull_request:1087", + "right": "pull_request:2915", + "accept": false, + "reason": "Both tweak slow-test expectations, but they are separate test files and unrelated adjustments." + }, + { + "left": "pull_request:1087", + "right": "pull_request:689", + "accept": false, + "reason": "Test-threshold adjustment vs attention performance rewrite; different changes and different failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 25, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10991", + "pull_request:11107", + "pull_request:1130", + "pull_request:11426", + "pull_request:11427", + "pull_request:11449", + "pull_request:11451", + "pull_request:11511", + "pull_request:11537", + "pull_request:1167", + "pull_request:11825", + "pull_request:11920", + "pull_request:11955", + "pull_request:1210", + "pull_request:1234", + "pull_request:12464", + "pull_request:12739", + "pull_request:1275", + "pull_request:13069", + "pull_request:13076", + "pull_request:1319", + "pull_request:13253", + "pull_request:1386", + "pull_request:1420", + "pull_request:1960", + "pull_request:1967", + "pull_request:1998", + "pull_request:2271", + "pull_request:2384", + "pull_request:2734", + "pull_request:2836", + "pull_request:2846", + "pull_request:2887", + "pull_request:2897", + "pull_request:2915", + "pull_request:3201", + "pull_request:3217", + "pull_request:3265", + "pull_request:3633", + "pull_request:3863", + "pull_request:4227", + "pull_request:4749", + "pull_request:4893", + "pull_request:564", + "pull_request:689", + "pull_request:930", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:10991|pull_request:11107", + "pull_request:1130|pull_request:3201", + "pull_request:1130|pull_request:3265", + "pull_request:1130|pull_request:930", + "pull_request:11426|pull_request:11449", + "pull_request:11427|pull_request:11449", + "pull_request:11451|pull_request:11920", + "pull_request:11451|pull_request:11955", + "pull_request:11511|pull_request:11537", + "pull_request:11537|pull_request:11825", + "pull_request:1167|pull_request:1998", + "pull_request:1210|pull_request:2915", + "pull_request:1210|pull_request:689", + "pull_request:1234|pull_request:2887", + "pull_request:12464|pull_request:13076", + "pull_request:12739|pull_request:13253", + "pull_request:1275|pull_request:975", + "pull_request:13069|pull_request:13253", + "pull_request:1319|pull_request:564", + "pull_request:1386|pull_request:1420", + "pull_request:1960|pull_request:1998", + "pull_request:1967|pull_request:3633", + "pull_request:2271|pull_request:2915", + "pull_request:2271|pull_request:689", + "pull_request:2384|pull_request:3217", + "pull_request:2734|pull_request:3863", + "pull_request:2836|pull_request:4749", + "pull_request:2846|pull_request:2915", + "pull_request:2846|pull_request:689", + "pull_request:2897|pull_request:4227", + "pull_request:2897|pull_request:4893", + "pull_request:2897|pull_request:930" + ], + "prepared_review_unit_hash": "001ef87af3fad0372d8421550bba08cc186ebbbcff6b2a668582f272b652abce", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 94365, + "estimated_input_tokens": 23592, + "estimated_eval_tokens": 47440 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are soft-similarity false positives: the PRs mostly touch different models, examples, docs, or tests, and do not look like the same underlying change or bug.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "10991 is the strongest standalone PR here: a merged, substantial Wan single-file feature with code, docs, and tests, even though it is not a duplicate anchor for the rest.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10991", + "right": "pull_request:11107", + "accept": false, + "reason": "Wan single-file support vs Wan LoRA compatibility; different features and code paths." + }, + { + "left": "pull_request:1130", + "right": "pull_request:3201", + "accept": false, + "reason": "Same community pipeline file, but one adds offload/xformers support while the other is a revert/update of that pipeline." + }, + { + "left": "pull_request:1130", + "right": "pull_request:3265", + "accept": false, + "reason": "One is a pipeline enhancement, the other is a revert of a revert; not the same change." + }, + { + "left": "pull_request:1130", + "right": "pull_request:930", + "accept": false, + "reason": "Pipeline code change vs docs/toctree restructuring; unrelated." + }, + { + "left": "pull_request:11426", + "right": "pull_request:11449", + "accept": false, + "reason": "Both add interpolation options, but to different training scripts and code paths." + }, + { + "left": "pull_request:11427", + "right": "pull_request:11449", + "accept": false, + "reason": "Same broad idea, but different scripts and implementations; not one duplicate fix." + }, + { + "left": "pull_request:11451", + "right": "pull_request:11920", + "accept": false, + "reason": "Wan torch.compile test vs Wan test-slice update; different purposes." + }, + { + "left": "pull_request:11451", + "right": "pull_request:11955", + "accept": false, + "reason": "Different models and test goals: Wan compile test vs Cosmos slices." + }, + { + "left": "pull_request:11511", + "right": "pull_request:11537", + "accept": false, + "reason": "Compile tests for different model classes; not the same bug or change." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11825", + "accept": false, + "reason": "UNet compile test vs LoRA hotswapping/resolution-change compilation test; related theme, different issue." + }, + { + "left": "pull_request:1167", + "right": "pull_request:1998", + "accept": false, + "reason": "Unrelated README/docs edits for different topics." + }, + { + "left": "pull_request:1210", + "right": "pull_request:2915", + "accept": false, + "reason": "Both are slow-test fixes, but for different pipelines and assertions." + }, + { + "left": "pull_request:1210", + "right": "pull_request:689", + "accept": false, + "reason": "Scheduler/pipeline noise handling vs attention implementation optimization; unrelated." + }, + { + "left": "pull_request:1234", + "right": "pull_request:2887", + "accept": false, + "reason": "Different examples and behaviors: unconditional sampling script vs ControlNet validation batching." + }, + { + "left": "pull_request:12464", + "right": "pull_request:13076", + "accept": false, + "reason": "Separate docs syntax fixes in unrelated areas." + }, + { + "left": "pull_request:12739", + "right": "pull_request:13253", + "accept": false, + "reason": "Type-check export for ZImage vs ZImage test refactor; different maintenance tasks." + }, + { + "left": "pull_request:1275", + "right": "pull_request:975", + "accept": false, + "reason": "Unet doc typo fix vs OpenVINO title typo fix; unrelated." + }, + { + "left": "pull_request:13069", + "right": "pull_request:13253", + "accept": false, + "reason": "Test refactors for different models; not the same underlying change." + }, + { + "left": "pull_request:1319", + "right": "pull_request:564", + "accept": false, + "reason": "Both Flax scheduler-related, but one fixes subfolder loading and the other updates scheduler semantics/docs; not a duplicate." + }, + { + "left": "pull_request:1386", + "right": "pull_request:1420", + "accept": false, + "reason": "Scheduler v-prediction support vs follow-up tests for SD2; feature and tests are distinct changes." + }, + { + "left": "pull_request:1960", + "right": "pull_request:1998", + "accept": false, + "reason": "Different documentation updates in different example folders." + }, + { + "left": "pull_request:1967", + "right": "pull_request:3633", + "accept": false, + "reason": "CPU offload example tweak vs dependency pin test; unrelated." + }, + { + "left": "pull_request:2271", + "right": "pull_request:2915", + "accept": false, + "reason": "Slow-test fixes in different pipelines and with different expected outputs." + }, + { + "left": "pull_request:2271", + "right": "pull_request:689", + "accept": false, + "reason": "Test-fix PR vs core attention math change; unrelated." + }, + { + "left": "pull_request:2384", + "right": "pull_request:3217", + "accept": false, + "reason": "Pix2Pix-zero helper revert vs pt_to_pil utility addition; opposite/unrelated changes." + }, + { + "left": "pull_request:2734", + "right": "pull_request:3863", + "accept": false, + "reason": "Both touch Habana docs, but they update different versions/configuration details and are not the same change." + }, + { + "left": "pull_request:2836", + "right": "pull_request:4749", + "accept": false, + "reason": "DreamBooth LoRA guide link fix vs SDXL LoRA checkpointing fix; unrelated." + }, + { + "left": "pull_request:2846", + "right": "pull_request:2915", + "accept": false, + "reason": "Slow-test fixes for different pipelines; not the same bug." + }, + { + "left": "pull_request:2846", + "right": "pull_request:689", + "accept": false, + "reason": "Stable UnCLIP test adjustment vs attention performance change; unrelated." + }, + { + "left": "pull_request:2897", + "right": "pull_request:4227", + "accept": false, + "reason": "StableUnCLIP docs example vs textual inversion README import fix; unrelated." + }, + { + "left": "pull_request:2897", + "right": "pull_request:4893", + "accept": false, + "reason": "StableUnCLIP docs example vs LoRA fusion docs; different documentation topics." + }, + { + "left": "pull_request:2897", + "right": "pull_request:930", + "accept": false, + "reason": "StableUnCLIP docs example vs toctree/docs navigation update; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 26, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10349", + "pull_request:3163", + "pull_request:3165", + "pull_request:3680", + "pull_request:3941", + "pull_request:3962", + "pull_request:4175", + "pull_request:4258", + "pull_request:4267", + "pull_request:4365", + "pull_request:4574", + "pull_request:4582", + "pull_request:4632", + "pull_request:4644", + "pull_request:4707", + "pull_request:5144", + "pull_request:5175", + "pull_request:5329", + "pull_request:5411", + "pull_request:5499", + "pull_request:5550", + "pull_request:5738", + "pull_request:6035", + "pull_request:6106", + "pull_request:6282", + "pull_request:6361", + "pull_request:6440", + "pull_request:6526", + "pull_request:6837", + "pull_request:7233", + "pull_request:7301", + "pull_request:7335", + "pull_request:7916", + "pull_request:7956", + "pull_request:7963", + "pull_request:8765", + "pull_request:8810", + "pull_request:8864", + "pull_request:9139", + "pull_request:9386", + "pull_request:9408", + "pull_request:9473", + "pull_request:9553", + "pull_request:9627", + "pull_request:9705", + "pull_request:9775", + "pull_request:9827", + "pull_request:9843" + ], + "soft_pairs": [ + "pull_request:3163|pull_request:3165", + "pull_request:3680|pull_request:5411", + "pull_request:3941|pull_request:5411", + "pull_request:3962|pull_request:5550", + "pull_request:4175|pull_request:5175", + "pull_request:4258|pull_request:4574", + "pull_request:4267|pull_request:4582", + "pull_request:4365|pull_request:4632", + "pull_request:4644|pull_request:6526", + "pull_request:4707|pull_request:5329", + "pull_request:5144|pull_request:6837", + "pull_request:5499|pull_request:6440", + "pull_request:5738|pull_request:7233", + "pull_request:6035|pull_request:6106", + "pull_request:6035|pull_request:6282", + "pull_request:6106|pull_request:6361", + "pull_request:7301|pull_request:7335", + "pull_request:7916|pull_request:7956", + "pull_request:7963|pull_request:9408", + "pull_request:8810|pull_request:9386", + "pull_request:8864|pull_request:9386", + "pull_request:9139|pull_request:9705", + "pull_request:9473|pull_request:9827", + "pull_request:9553|pull_request:9775", + "pull_request:9627|pull_request:9843", + "pull_request:10349|pull_request:8765" + ], + "prepared_review_unit_hash": "8ec867efbaf684a546c8ba98204aeac994cf6d52a7646d4189818a0411763b9e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 98615, + "estimated_input_tokens": 24654, + "estimated_eval_tokens": 49564 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: mostly unrelated docs/example fixes, test tweaks, and feature PRs. None of the soft-edge pairs look like the same underlying change, so no duplicate merges are justified.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical representative because the cluster spans unrelated topics and code paths rather than one shared bug or feature.", + "best_issue_reason": null, + "best_pr_reason": "No PR stands out as a true cluster anchor; even the closest-looking pairs are only loosely related and should stay separate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3163", + "right": "pull_request:3165", + "accept": false, + "reason": "Same docs file, but one adds author credit while the other changes training instructions; different changes." + }, + { + "left": "pull_request:3680", + "right": "pull_request:5411", + "accept": false, + "reason": "Both are doc fixes, but they target different pages and different broken links." + }, + { + "left": "pull_request:3941", + "right": "pull_request:5411", + "accept": false, + "reason": "Different documentation pages and different link/code fixes; not the same underlying issue." + }, + { + "left": "pull_request:3962", + "right": "pull_request:5550", + "accept": false, + "reason": "Both fix checkpoint links in docs, but for different pipelines/pages and different URLs." + }, + { + "left": "pull_request:4175", + "right": "pull_request:5175", + "accept": false, + "reason": "One is a ControlNet WebDataset training feature; the other is a CI environment variable tweak." + }, + { + "left": "pull_request:4258", + "right": "pull_request:4574", + "accept": false, + "reason": "Both touch example READMEs, but they fix different example scripts and different content." + }, + { + "left": "pull_request:4267", + "right": "pull_request:4582", + "accept": false, + "reason": "Unrelated doc image correction versus ControlNet SDXL docstring fix." + }, + { + "left": "pull_request:4365", + "right": "pull_request:4632", + "accept": false, + "reason": "Both involve the SDXL LoRA example area, but one introduces the script and docs while the other fixes runtime bugs in that script." + }, + { + "left": "pull_request:4644", + "right": "pull_request:6526", + "accept": false, + "reason": "Different SDXL-related docs/training changes; not the same concrete code-path or issue." + }, + { + "left": "pull_request:4707", + "right": "pull_request:5329", + "accept": false, + "reason": "Custom Diffusion README changes are unrelated to the later inference bug fix in the training script." + }, + { + "left": "pull_request:5144", + "right": "pull_request:6837", + "accept": false, + "reason": "Different pipeline docs and unrelated fixes; only broad doc similarity." + }, + { + "left": "pull_request:5499", + "right": "pull_request:6440", + "accept": false, + "reason": "Both are link-fix PRs, but they affect different documentation sets and different broken links." + }, + { + "left": "pull_request:5738", + "right": "pull_request:7233", + "accept": false, + "reason": "LoRA adapter deletion and DoRA training are different features, not the same change." + }, + { + "left": "pull_request:6035", + "right": "pull_request:6106", + "accept": false, + "reason": "Same file, but one improves CLI help text and the other fixes an example snippet; separate issues." + }, + { + "left": "pull_request:6035", + "right": "pull_request:6282", + "accept": false, + "reason": "Help-text cleanup versus DreamBooth runtime fix are unrelated." + }, + { + "left": "pull_request:6106", + "right": "pull_request:6361", + "accept": false, + "reason": "Both touch the same advanced SDXL script, but one fixes an example while the other changes LoRA serialization keys." + }, + { + "left": "pull_request:7301", + "right": "pull_request:7335", + "accept": false, + "reason": "A VAE scaling constant test fix and a ControlNet SDXL config test are unrelated." + }, + { + "left": "pull_request:7916", + "right": "pull_request:7956", + "accept": false, + "reason": "Training-script NPU support and adding an NPU Dockerfile are related infrastructure work, but not the same concrete change." + }, + { + "left": "pull_request:7963", + "right": "pull_request:9408", + "accept": false, + "reason": "Both add pipeline variants with custom timestep handling, but they are for different families and separate code paths." + }, + { + "left": "pull_request:8810", + "right": "pull_request:9386", + "accept": false, + "reason": "Both add PAG ControlNet img2img-related docs/code, but img2img and inpaint are distinct pipelines and changes." + }, + { + "left": "pull_request:8864", + "right": "pull_request:9386", + "accept": false, + "reason": "Same PAG/ControlNet area, but still different pipeline variants and separate additions." + }, + { + "left": "pull_request:9139", + "right": "pull_request:9705", + "accept": false, + "reason": "Flux Dreambooth/LoRA README updates do not match the broader bug-fix and update set in the later PR." + }, + { + "left": "pull_request:9473", + "right": "pull_request:9827", + "accept": false, + "reason": "One fixes a Flux Prodigy learning-rate bug; the other fixes LoRA device-map/offload behavior." + }, + { + "left": "pull_request:9553", + "right": "pull_request:9775", + "accept": false, + "reason": "Distributed inference docs versus Flux LoRA quantization README are unrelated." + }, + { + "left": "pull_request:9627", + "right": "pull_request:9843", + "accept": false, + "reason": "Different test files and different LoRA scenarios; not the same bug." + }, + { + "left": "pull_request:10349", + "right": "pull_request:8765", + "accept": false, + "reason": "One fixes single-file model-type inference code; the other fixes documentation examples for single-file loading." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 27, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10261", + "pull_request:10867", + "pull_request:10985", + "pull_request:11059", + "pull_request:11067", + "pull_request:11243", + "pull_request:11292", + "pull_request:11512", + "pull_request:11552", + "pull_request:11670", + "pull_request:11704", + "pull_request:1229", + "pull_request:12330", + "pull_request:12353", + "pull_request:12541", + "pull_request:12660", + "pull_request:12970", + "pull_request:13099", + "pull_request:192", + "pull_request:2128", + "pull_request:2172", + "pull_request:2557", + "pull_request:2595", + "pull_request:4227", + "pull_request:4795", + "pull_request:5615", + "pull_request:5738", + "pull_request:5807", + "pull_request:6174", + "pull_request:6217", + "pull_request:6327", + "pull_request:6787", + "pull_request:6864", + "pull_request:6965", + "pull_request:6968", + "pull_request:7189", + "pull_request:7235", + "pull_request:7820", + "pull_request:8282", + "pull_request:8798", + "pull_request:8902", + "pull_request:8919", + "pull_request:8985", + "pull_request:903", + "pull_request:9061", + "pull_request:9257", + "pull_request:9682", + "pull_request:9709" + ], + "soft_pairs": [ + "pull_request:11704|pull_request:12541", + "pull_request:2128|pull_request:2172", + "pull_request:8919|pull_request:8985", + "pull_request:9257|pull_request:9682", + "pull_request:7189|pull_request:7820", + "pull_request:11670|pull_request:12330", + "pull_request:10985|pull_request:11243", + "pull_request:11059|pull_request:11292", + "pull_request:5615|pull_request:6965", + "pull_request:5738|pull_request:7235", + "pull_request:5807|pull_request:6787", + "pull_request:8798|pull_request:8902", + "pull_request:11067|pull_request:9709", + "pull_request:11512|pull_request:11552", + "pull_request:1229|pull_request:2557", + "pull_request:12353|pull_request:13099", + "pull_request:192|pull_request:903", + "pull_request:2128|pull_request:2595", + "pull_request:4227|pull_request:6174", + "pull_request:4795|pull_request:6217", + "pull_request:6864|pull_request:8282", + "pull_request:8902|pull_request:9061", + "pull_request:10261|pull_request:10867", + "pull_request:12660|pull_request:12970", + "pull_request:6327|pull_request:6968" + ], + "prepared_review_unit_hash": "ad8cf8a8476b593d5b88ab6d43d5925b6f3165fefa79c2eda6600a1042e109c8", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 96551, + "estimated_input_tokens": 24138, + "estimated_eval_tokens": 48532 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the items span many unrelated PRs (Flux/IP-Adapter, LoRA converters, torch.compile tests, docs, CI, scheduler/loading fixes). The soft edges mostly reflect broad topical similarity, not the same underlying change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a valid canonical for this set; the cluster mixes unrelated features, bug fixes, tests, and docs.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global representative for the whole set; only a few pairs are loosely adjacent, but the cluster is overall heterogeneous.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11704", + "right": "pull_request:12541", + "accept": false, + "reason": "Both touch Wan LoRA conversion, but they fix different formats and code paths: AccVideo low-magnitude diff weights vs latest few-step head.head weights." + }, + { + "left": "pull_request:2128", + "right": "pull_request:2172", + "accept": false, + "reason": "Different changes: disabling LoRA after use vs adding validation support in multi-GPU training scripts." + }, + { + "left": "pull_request:8919", + "right": "pull_request:8985", + "accept": false, + "reason": "Unrelated test changes for different models; one adds Latte transformer coverage, the other shrinks Lumina test size." + }, + { + "left": "pull_request:9257", + "right": "pull_request:9682", + "accept": false, + "reason": "Different DreamBooth/script fixes: Flux prompt tokenization alignment vs SD3 latent caching and training updates." + }, + { + "left": "pull_request:7189", + "right": "pull_request:7820", + "accept": false, + "reason": "Both involve IP-Adapter, but one fixes plus-embed repetition across many pipelines while the other adds InstructPix2Pix IP-Adapter support." + }, + { + "left": "pull_request:11670", + "right": "pull_request:12330", + "accept": false, + "reason": "Shared group-offloading theme, but one adds a torch.compile + group offloading test while the other cleans up pipeline-level group-offloading tests." + }, + { + "left": "pull_request:10985", + "right": "pull_request:11243", + "accept": false, + "reason": "Different LoRA conversion fixes for different ecosystems: ComfyUI Flux loras vs musubi Wan loras." + }, + { + "left": "pull_request:11059", + "right": "pull_request:11292", + "accept": false, + "reason": "Both edit LoRA conversion utils, but one targets non-diffusers Wan t2v loras and the other supports more SDXL loras." + }, + { + "left": "pull_request:5615", + "right": "pull_request:6965", + "accept": false, + "reason": "One reverts SDXL Adapter load_lora handling; the other moves an SDXL T2I Adapter LoRA test into the PEFT workflow. Not the same change." + }, + { + "left": "pull_request:5738", + "right": "pull_request:7235", + "accept": false, + "reason": "Different features: LoRA adapter deletion vs DoRA training support." + }, + { + "left": "pull_request:5807", + "right": "pull_request:6787", + "accept": false, + "reason": "Unrelated fixes in different areas: LPW SDXL prompt embedding duplication vs Marigold depth input range normalization." + }, + { + "left": "pull_request:8798", + "right": "pull_request:8902", + "accept": false, + "reason": "Both are sharded-checkpoint loading fixes, but they address different bugs: subfolder handling vs variant sibling selection." + }, + { + "left": "pull_request:11067", + "right": "pull_request:9709", + "accept": false, + "reason": "Unrelated CI changes: benchmarking transformer pin vs max torch version pin." + }, + { + "left": "pull_request:11512", + "right": "pull_request:11552", + "accept": false, + "reason": "Same test-coverage theme, but they target different models and are separate torch.compile test additions." + }, + { + "left": "pull_request:1229", + "right": "pull_request:2557", + "accept": false, + "reason": "Unrelated docs/example edits: bitsandbytes install wording vs DreamBooth intermediate logging." + }, + { + "left": "pull_request:12353", + "right": "pull_request:13099", + "accept": false, + "reason": "One implements AutoModel custom-code loading; the other adds docs. Related feature, but not the same concrete change." + }, + { + "left": "pull_request:192", + "right": "pull_request:903", + "accept": false, + "reason": "Completely different changes: README parameter tweak vs inpainting pipeline deprecation." + }, + { + "left": "pull_request:2128", + "right": "pull_request:2595", + "accept": false, + "reason": "Different code paths: LoRA disable/removal vs scheduler configuration switching." + }, + { + "left": "pull_request:4227", + "right": "pull_request:6174", + "accept": false, + "reason": "Both are docs/example fixes, but for unrelated examples and unrelated mistakes." + }, + { + "left": "pull_request:4795", + "right": "pull_request:6217", + "accept": false, + "reason": "Separate torch.compile bugs in different components: ControlNet vs Stable Video Diffusion VAE." + }, + { + "left": "pull_request:6864", + "right": "pull_request:8282", + "accept": false, + "reason": "One is a PEFT/torch.compile docs note, the other is I2VGenXL docs. Not the same change." + }, + { + "left": "pull_request:8902", + "right": "pull_request:9061", + "accept": false, + "reason": "Both concern sharded checkpoints, but the concrete fixes differ: part II variant handling vs loading from cached subfolders." + }, + { + "left": "pull_request:10261", + "right": "pull_request:10867", + "accept": false, + "reason": "Closely related Flux/IP-Adapter work, but 10261 adds Flux IP-Adapter support while 10867 adds multi-IP-Adapter support; these are separate changes." + }, + { + "left": "pull_request:12660", + "right": "pull_request:12970", + "accept": false, + "reason": "Different model-specific context-parallel bug fixes: ChronoEdit vs Qwen-Image." + }, + { + "left": "pull_request:6327", + "right": "pull_request:6968", + "accept": false, + "reason": "Different DreamBooth-LoRA fixes: advanced SDXL script bug fix vs prior-preservation-data bug fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 28, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10183", + "pull_request:1026", + "pull_request:10279", + "pull_request:10299", + "pull_request:10323", + "pull_request:10376", + "pull_request:10418", + "pull_request:10544", + "pull_request:10633", + "pull_request:10975", + "pull_request:11025", + "pull_request:1130", + "pull_request:11338", + "pull_request:11451", + "pull_request:11473", + "pull_request:11529", + "pull_request:11541", + "pull_request:11550", + "pull_request:11623", + "pull_request:11778", + "pull_request:11954", + "pull_request:12183", + "pull_request:12681", + "pull_request:12750", + "pull_request:13039", + "pull_request:13069", + "pull_request:13209", + "pull_request:13344", + "pull_request:1346", + "pull_request:2839", + "pull_request:2994", + "pull_request:3197", + "pull_request:3299", + "pull_request:4632", + "pull_request:4799", + "pull_request:4910", + "pull_request:5906", + "pull_request:5986", + "pull_request:6004", + "pull_request:6402", + "pull_request:6999", + "pull_request:7934", + "pull_request:8571", + "pull_request:8755", + "pull_request:9008", + "pull_request:9139", + "pull_request:9827" + ], + "soft_pairs": [ + "pull_request:8755|pull_request:9139", + "pull_request:2994|pull_request:4632", + "pull_request:3299|pull_request:4910", + "pull_request:5906|pull_request:6402", + "pull_request:5986|pull_request:6004", + "pull_request:8571|pull_request:9008", + "pull_request:10376|pull_request:11025", + "pull_request:7934|pull_request:9827", + "pull_request:4799|pull_request:6999", + "pull_request:12681|pull_request:13039", + "pull_request:10183|pull_request:11550", + "pull_request:1026|pull_request:2839", + "pull_request:10279|pull_request:11473", + "pull_request:10299|pull_request:11529", + "pull_request:10323|pull_request:10418", + "pull_request:10544|pull_request:10633", + "pull_request:10975|pull_request:11778", + "pull_request:1130|pull_request:3197", + "pull_request:11338|pull_request:12183", + "pull_request:11451|pull_request:11954", + "pull_request:11541|pull_request:11623", + "pull_request:12750|pull_request:13209", + "pull_request:13069|pull_request:13344", + "pull_request:1346|pull_request:2839" + ], + "prepared_review_unit_hash": "50b61f20847d2fa0893845fde603095fdf8596c48d5a2dc2917adb992f578d05", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 93724, + "estimated_input_tokens": 23431, + "estimated_eval_tokens": 47118 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated docs, tests, and feature PRs. None of the soft pairs look like the same concrete bug or change, so all soft edges should be rejected. If a representative PR is needed, 12750 is the strongest standalone example because it is the broadest merged feature change in the set.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "12750 is the best representative PR here: it is a substantial merged feature change with code, docs, and tests; the other items are smaller, unrelated fixes or docs edits.", + "best_issue_reason": null, + "best_pr_reason": "12750 is the most substantial and self-contained merged PR in the list, making it the best standalone representative even though the cluster is not a true duplicate set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8755", + "right": "pull_request:9139", + "accept": false, + "reason": "Both are Dreambooth-related, but they target different model families and different issues; not the same underlying change." + }, + { + "left": "pull_request:2994", + "right": "pull_request:4632", + "accept": false, + "reason": "Both touch training examples, but one fixes ControlNet type casting while the other fixes LoRA checkpointing/casting in SDXL training; different bugs." + }, + { + "left": "pull_request:3299", + "right": "pull_request:4910", + "accept": false, + "reason": "One is LoRA training/docs work and the other is an inpainting docs typo fix; unrelated changes." + }, + { + "left": "pull_request:5906", + "right": "pull_request:6402", + "accept": false, + "reason": "One fixes an SDXL LoRA README template issue, the other fixes a gradient-checkpointing bug in training; different code paths." + }, + { + "left": "pull_request:5986", + "right": "pull_request:6004", + "accept": false, + "reason": "Same SVD doc file, but the edits are different follow-up documentation changes rather than the same fix." + }, + { + "left": "pull_request:8571", + "right": "pull_request:9008", + "accept": false, + "reason": "Both are errata-style cleanups, but they change different files and address different typo/style issues." + }, + { + "left": "pull_request:10376", + "right": "pull_request:11025", + "accept": false, + "reason": "Both involve LoRA conversion code for Wan/HunyuanVideo, but they add support for different model-specific formats and code paths." + }, + { + "left": "pull_request:7934", + "right": "pull_request:9827", + "accept": false, + "reason": "Both are loader-related, but one fixes single-file config dict handling while the other fixes LoRA device-map/offload behavior; not the same bug." + }, + { + "left": "pull_request:4799", + "right": "pull_request:6999", + "accept": false, + "reason": "Different pipelines and different bugs: ControlNet SDXL guess_mode versus Stable Video Diffusion image-tensor handling." + }, + { + "left": "pull_request:12681", + "right": "pull_request:13039", + "accept": false, + "reason": "One fixes failing CPU pipeline tests across several models; the other changes Wan Animate upcasting logic in core model code." + }, + { + "left": "pull_request:10183", + "right": "pull_request:11550", + "accept": false, + "reason": "Both are single-file loading support, but for different models and formats (AutoencoderDC vs HiDream GGUF); not mergeable as one fix." + }, + { + "left": "pull_request:1026", + "right": "pull_request:2839", + "accept": false, + "reason": "Both are README/docs updates, but they cover different example sets and unrelated content changes." + }, + { + "left": "pull_request:10279", + "right": "pull_request:11473", + "accept": false, + "reason": "Both are quantization docs fixes, but they address different pages and different mistakes." + }, + { + "left": "pull_request:10299", + "right": "pull_request:11529", + "accept": false, + "reason": "The first is a specific GGUF docs example fix; the second has no diff preview and no evidence it addresses the same issue." + }, + { + "left": "pull_request:10323", + "right": "pull_request:10418", + "accept": false, + "reason": "Both are internal-link docs cleanups, but for different pages and different link targets." + }, + { + "left": "pull_request:10544", + "right": "pull_request:10633", + "accept": false, + "reason": "Both add docs for separate features, ParaAttention and IP-Adapter; they are not the same change." + }, + { + "left": "pull_request:10975", + "right": "pull_request:11778", + "accept": false, + "reason": "Both are Wan-related, but they fix different pipeline bugs in different components (i2v pipeline vs VACE mask padding)." + }, + { + "left": "pull_request:1130", + "right": "pull_request:3197", + "accept": false, + "reason": "Same community pipeline family, but one is an incremental feature fix while the other is a broader pipeline rewrite; not the same underlying change." + }, + { + "left": "pull_request:11338", + "right": "pull_request:12183", + "accept": false, + "reason": "One updates LoRA docs utilities, the other adds a LoRA badge to QwenImage docs; unrelated docs changes." + }, + { + "left": "pull_request:11451", + "right": "pull_request:11954", + "accept": false, + "reason": "Both are tests, but for different models and different behaviors; no shared bug or merged fix path." + }, + { + "left": "pull_request:11541", + "right": "pull_request:11623", + "accept": false, + "reason": "These are both tiny typo fixes, but in unrelated files and unrelated code/docs locations." + }, + { + "left": "pull_request:12750", + "right": "pull_request:13209", + "accept": false, + "reason": "Both are ZImage LoRA-related, but one adds the pipeline support and the other adjusts conversion normalization for broader LoRA compatibility; different concrete changes." + }, + { + "left": "pull_request:13069", + "right": "pull_request:13344", + "accept": false, + "reason": "Different model test suites with different goals; no same underlying test or model bug." + }, + { + "left": "pull_request:1346", + "right": "pull_request:2839", + "accept": false, + "reason": "Both touch README content, but they are unrelated example documentation edits from different areas." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 29, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:178", + "pull_request:1927", + "pull_request:2330", + "pull_request:2464", + "pull_request:2526", + "pull_request:2559", + "pull_request:2678", + "pull_request:2699", + "pull_request:2767", + "pull_request:3397", + "pull_request:3458", + "pull_request:3460", + "pull_request:3472", + "pull_request:3504", + "pull_request:3568", + "pull_request:3626", + "pull_request:3852", + "pull_request:386", + "pull_request:3945", + "pull_request:3971", + "pull_request:4237", + "pull_request:4238", + "pull_request:4258", + "pull_request:4540", + "pull_request:4556", + "pull_request:4582", + "pull_request:4612", + "pull_request:4652", + "pull_request:4893", + "pull_request:4947", + "pull_request:5051", + "pull_request:5078", + "pull_request:5090", + "pull_request:5101", + "pull_request:5144", + "pull_request:5180", + "pull_request:5244", + "pull_request:5568", + "pull_request:5674", + "pull_request:5914", + "pull_request:6106", + "pull_request:6282", + "pull_request:6402", + "pull_request:6556", + "pull_request:6775", + "pull_request:7086", + "pull_request:7196", + "pull_request:903" + ], + "soft_pairs": [ + "pull_request:178|pull_request:903", + "pull_request:1927|pull_request:386", + "pull_request:2330|pull_request:2699", + "pull_request:2464|pull_request:2767", + "pull_request:2526|pull_request:2678", + "pull_request:2559|pull_request:4582", + "pull_request:3397|pull_request:3852", + "pull_request:3458|pull_request:3626", + "pull_request:3460|pull_request:3626", + "pull_request:3472|pull_request:4893", + "pull_request:3504|pull_request:5568", + "pull_request:3568|pull_request:5101", + "pull_request:3945|pull_request:6282", + "pull_request:3971|pull_request:4237", + "pull_request:3971|pull_request:4238", + "pull_request:4258|pull_request:4540", + "pull_request:4556|pull_request:4582", + "pull_request:4612|pull_request:5180", + "pull_request:4652|pull_request:5914", + "pull_request:4652|pull_request:6402", + "pull_request:4947|pull_request:7196", + "pull_request:5051|pull_request:5674", + "pull_request:5078|pull_request:6556", + "pull_request:5090|pull_request:7086", + "pull_request:5144|pull_request:6775", + "pull_request:5244|pull_request:6106", + "pull_request:5244|pull_request:6282" + ], + "prepared_review_unit_hash": "c0201fa73044543320292160984ebd81fde0e54ba6542317599bfa55f5bf23b2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 97850, + "estimated_input_tokens": 24463, + "estimated_eval_tokens": 49182 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: it mixes unrelated docs, tests, and code fixes across many different areas. I would not merge any of the soft edges.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 3504 is the most central substantive code change in the set and includes tests, but the cluster is too heterogeneous to have a true duplicate canonical PR.", + "best_issue_reason": null, + "best_pr_reason": "PR 3504 is the best representative only in a loose sense because it is a real code-path change rather than a doc-only or tiny test tweak.", + "soft_edge_verdicts": [ + { + "left": "pull_request:178", + "right": "pull_request:903", + "accept": false, + "reason": "Stable Diffusion test update vs inpainting pipeline deprecation; different changes." + }, + { + "left": "pull_request:1927", + "right": "pull_request:386", + "accept": false, + "reason": "General Stable Diffusion guide vs API docs page; unrelated doc content." + }, + { + "left": "pull_request:2330", + "right": "pull_request:2699", + "accept": false, + "reason": "Foundation ethical-guidelines page vs later safety-checker subsection in the same file; follow-up, not duplicate." + }, + { + "left": "pull_request:2464", + "right": "pull_request:2767", + "accept": false, + "reason": "Different training scripts and fixes: xformers warning removal vs PEFT flag rename." + }, + { + "left": "pull_request:2526", + "right": "pull_request:2678", + "accept": false, + "reason": "Separate test suites and failures; one adjusts tolerances, the other changes serialization setup." + }, + { + "left": "pull_request:2559", + "right": "pull_request:4582", + "accept": false, + "reason": "Docs formatting fix vs ControlNet SDXL docstring example correction." + }, + { + "left": "pull_request:3397", + "right": "pull_request:3852", + "accept": false, + "reason": "Different docs additions on unrelated pages." + }, + { + "left": "pull_request:3458", + "right": "pull_request:3626", + "accept": false, + "reason": "Both mention LoRA scale, but one is inpaint support and the other is a broader draft for text-encoder scale across many files." + }, + { + "left": "pull_request:3460", + "right": "pull_request:3626", + "accept": false, + "reason": "Broad multi-file LoRA scale draft is not the same concrete change as inpaint lora-scale support." + }, + { + "left": "pull_request:3472", + "right": "pull_request:4893", + "accept": false, + "reason": "DreamBooth IF training docs vs LoRA-fusion scale guidance; different documentation topics." + }, + { + "left": "pull_request:3504", + "right": "pull_request:5568", + "accept": false, + "reason": "Same subsystem, but distinct custom-pipeline loading bugs in different code paths; not one mergeable change." + }, + { + "left": "pull_request:3568", + "right": "pull_request:5101", + "accept": false, + "reason": "Broken-link fix vs SDXL refine-doc link update." + }, + { + "left": "pull_request:3945", + "right": "pull_request:6282", + "accept": false, + "reason": "Different DreamBooth script issues: adding rank vs removing a runtime-error code path." + }, + { + "left": "pull_request:3971", + "right": "pull_request:4237", + "accept": false, + "reason": "Docs workflow/page update vs SDXL prompt-encoding bug fix." + }, + { + "left": "pull_request:3971", + "right": "pull_request:4238", + "accept": false, + "reason": "Docs workflow/page update vs SDXL ControlNet inference fix." + }, + { + "left": "pull_request:4258", + "right": "pull_request:4540", + "accept": false, + "reason": "Same README file, but different additions: hyperparameter note vs Colab notebook link." + }, + { + "left": "pull_request:4556", + "right": "pull_request:4582", + "accept": false, + "reason": "New ControlNet SDXL docs page vs a docstring example fix." + }, + { + "left": "pull_request:4612", + "right": "pull_request:5180", + "accept": false, + "reason": "Different docs pages and different formatting fixes." + }, + { + "left": "pull_request:4652", + "right": "pull_request:5914", + "accept": false, + "reason": "LoRA SDXL tests vs README template text." + }, + { + "left": "pull_request:4652", + "right": "pull_request:6402", + "accept": false, + "reason": "Tests vs a training-script gradient-checkpointing bug fix." + }, + { + "left": "pull_request:4947", + "right": "pull_request:7196", + "accept": false, + "reason": "T2I-Adapter training tests vs IP-Adapter training/docs addition." + }, + { + "left": "pull_request:5051", + "right": "pull_request:5674", + "accept": false, + "reason": "Wuerstchen typo fix vs broader docs cleanup across many files." + }, + { + "left": "pull_request:5078", + "right": "pull_request:6556", + "accept": false, + "reason": "Training bug fix vs torch.compile compatibility changes." + }, + { + "left": "pull_request:5090", + "right": "pull_request:7086", + "accept": false, + "reason": "Adapter bug fix vs AnimateDiff/IP-Adapter community update." + }, + { + "left": "pull_request:5144", + "right": "pull_request:6775", + "accept": false, + "reason": "BLIP Diffusion doc sample fix vs scheduler parameter docs." + }, + { + "left": "pull_request:5244", + "right": "pull_request:6106", + "accept": false, + "reason": "Kandinsky training bug fix vs SDXL advanced example code fix." + }, + { + "left": "pull_request:5244", + "right": "pull_request:6282", + "accept": false, + "reason": "Completely different scripts and issues; no shared concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 30, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11532", + "pull_request:11660", + "pull_request:11994", + "pull_request:12439", + "pull_request:12681", + "pull_request:12687", + "pull_request:12812", + "pull_request:12914", + "pull_request:1370", + "pull_request:2193", + "pull_request:3779", + "pull_request:3859", + "pull_request:4098", + "pull_request:4295", + "pull_request:4417", + "pull_request:5506", + "pull_request:5945", + "pull_request:6035", + "pull_request:6145", + "pull_request:6174", + "pull_request:6184", + "pull_request:6361", + "pull_request:6403", + "pull_request:6440", + "pull_request:6537", + "pull_request:6566", + "pull_request:6655", + "pull_request:6837", + "pull_request:6875", + "pull_request:6962", + "pull_request:7029", + "pull_request:7124", + "pull_request:7227", + "pull_request:7233", + "pull_request:7236", + "pull_request:7260", + "pull_request:7425", + "pull_request:830", + "pull_request:833", + "pull_request:8616", + "pull_request:869", + "pull_request:8715", + "pull_request:874", + "pull_request:879", + "pull_request:9017", + "pull_request:9257", + "pull_request:930" + ], + "soft_pairs": [ + "pull_request:5945|pull_request:6962", + "pull_request:6035|pull_request:6403", + "pull_request:6035|pull_request:7124", + "pull_request:6145|pull_request:6537", + "pull_request:6174|pull_request:6837", + "pull_request:6174|pull_request:7425", + "pull_request:6361|pull_request:6403", + "pull_request:6440|pull_request:7029", + "pull_request:6566|pull_request:8715", + "pull_request:6655|pull_request:7124", + "pull_request:6875|pull_request:7260", + "pull_request:7227|pull_request:7236", + "pull_request:830|pull_request:833", + "pull_request:869|pull_request:879", + "pull_request:874|pull_request:930", + "pull_request:4295|pull_request:4417", + "pull_request:2193|pull_request:3779", + "pull_request:5506|pull_request:6184", + "pull_request:12439|pull_request:12812", + "pull_request:11994|pull_request:12681", + "pull_request:3859|pull_request:4098", + "pull_request:7233|pull_request:9257", + "pull_request:8616|pull_request:9017", + "pull_request:11532|pull_request:11660", + "pull_request:12687|pull_request:12914", + "pull_request:1370|pull_request:874" + ], + "prepared_review_unit_hash": "b8b112a4d6dc8b324e697c7ed5d9d34fa2f70d915af1eb545dcaf647968ad460", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 98805, + "estimated_input_tokens": 24702, + "estimated_eval_tokens": 49660 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly weakly similar but not duplicate-prone: it mixes unrelated docs, test, release, and feature PRs. I found no soft edges strong enough to merge; the closest ones are same-file follow-ups, but they still change different code paths or different feature scopes.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11532 is the most self-contained concrete feature change here: a merged HiDream LoRA conversion/pipeline support PR with a single clear code path and no broader cross-cutting edits.", + "best_issue_reason": null, + "best_pr_reason": "PR 9017 is the broadest substantive PR in the set, but it is still a distinct AuraFlow feature rather than a duplicate of the others; if one artifact had to anchor the cluster, this is the strongest representative by scope.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5945", + "right": "pull_request:6962", + "accept": false, + "reason": "Both touch PixArt tests, but one removes slow-suite fast tests while the other changes a slow-test assertion; same area, different test failures and fixes." + }, + { + "left": "pull_request:6035", + "right": "pull_request:6403", + "accept": false, + "reason": "One is a help-text/docstring cleanup; the other adds WebUI format support and loader logic. Same file, different change." + }, + { + "left": "pull_request:6035", + "right": "pull_request:7124", + "accept": false, + "reason": "Help text improvements and a kwarg rename in training are unrelated fixes, not the same bug/change." + }, + { + "left": "pull_request:6145", + "right": "pull_request:6537", + "accept": false, + "reason": "Both are LCM distillation scripts, but one cleans comments/validation logging and the other adds a target-model option. Different functionality." + }, + { + "left": "pull_request:6174", + "right": "pull_request:6837", + "accept": false, + "reason": "A docs punctuation fix and a separate dependency-order tweak in a pipeline doc are not the same issue." + }, + { + "left": "pull_request:6174", + "right": "pull_request:7425", + "accept": false, + "reason": "Missing backslashes in LoRA docs and missing output image content are different documentation fixes." + }, + { + "left": "pull_request:6361", + "right": "pull_request:6403", + "accept": false, + "reason": "One fixes LoRA key conversion on save; the other adds WebUI-format import/export support. Related topic, different concrete change." + }, + { + "left": "pull_request:6440", + "right": "pull_request:7029", + "accept": false, + "reason": "These are separate docs fixes on different pages: local link cleanup vs alt-text/image URL cleanup." + }, + { + "left": "pull_request:6566", + "right": "pull_request:8715", + "accept": false, + "reason": "Both edit the advanced DreamBooth LoRA script, but one fixes resume behavior and the other adds clip_skip support. Not the same underlying bug." + }, + { + "left": "pull_request:6655", + "right": "pull_request:7124", + "accept": false, + "reason": "One modularizes training-time inference; the other renames a kwarg in a validation call. Different scope and code path." + }, + { + "left": "pull_request:6875", + "right": "pull_request:7260", + "accept": false, + "reason": "A test expectation update for IP Adapter is not the same as adding IP Adapter support to SAG pipeline code." + }, + { + "left": "pull_request:7227", + "right": "pull_request:7236", + "accept": false, + "reason": "One adds DoRA parser/LoRA config support; the other fixes embedding save paths. Same file, different features." + }, + { + "left": "pull_request:830", + "right": "pull_request:833", + "accept": false, + "reason": "These are sequential release version bumps (0.5.0 to 0.5.1), not duplicates of the same change." + }, + { + "left": "pull_request:869", + "right": "pull_request:879", + "accept": false, + "reason": "Both update the community README, but one fixes an import bug and expands README content while the other adjusts the examples table. Different edits." + }, + { + "left": "pull_request:874", + "right": "pull_request:930", + "accept": false, + "reason": "A generic community pipeline example and a broader custom-pipeline docs restructure are not the same change." + }, + { + "left": "pull_request:4295", + "right": "pull_request:4417", + "accept": false, + "reason": "Zero123 community pipeline implementation and broader Zero123 support/docs changes overlap by topic, but they are not the same concrete PR change." + }, + { + "left": "pull_request:2193", + "right": "pull_request:3779", + "accept": false, + "reason": "Adding SAG Stable Diffusion and adding segmentation output are different feature requests for different pipeline behavior." + }, + { + "left": "pull_request:5506", + "right": "pull_request:6184", + "accept": false, + "reason": "One fixes PEFT/xformers attention calls; the other fixes offline LoRA weight discovery. Different code paths and failures." + }, + { + "left": "pull_request:12439", + "right": "pull_request:12812", + "accept": false, + "reason": "Both touch kernels-based attention dispatch, but one adds backend support while the other enables checkpointing wrappers. Same subsystem, different concrete change." + }, + { + "left": "pull_request:11994", + "right": "pull_request:12681", + "accept": false, + "reason": "A huggingface-hub test/doc compatibility fix and a CPU pipeline test fix are unrelated CI issues." + }, + { + "left": "pull_request:3859", + "right": "pull_request:4098", + "accept": false, + "reason": "SDXL pipeline addition and SDXL inpainting addition are related releases, but not the same code-path change." + }, + { + "left": "pull_request:7233", + "right": "pull_request:9257", + "accept": false, + "reason": "DoRA training for SDXL LoRA and Flux dreambooth alignment are different scripts/models, not duplicates." + }, + { + "left": "pull_request:8616", + "right": "pull_request:9017", + "accept": false, + "reason": "SD3 LoRA fusion and AuraFlow LoRA support are distinct model-feature implementations." + }, + { + "left": "pull_request:11532", + "right": "pull_request:11660", + "accept": false, + "reason": "The second PR extends HiDream LoRA handling with a new format-detection path; it is a follow-up enhancement, not the same concrete change as the first." + }, + { + "left": "pull_request:12687", + "right": "pull_request:12914", + "accept": false, + "reason": "A typo fix in autoencoder_dc and a revision-string typo fix in Cosmos docs/code are unrelated." + }, + { + "left": "pull_request:1370", + "right": "pull_request:874", + "accept": false, + "reason": "Both touch community README content, but one updates an inpainting example and the other adds a generic pipeline example. Different examples and edits." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 31, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10259", + "pull_request:10376", + "pull_request:10378", + "pull_request:10481", + "pull_request:10517", + "pull_request:11546", + "pull_request:11572", + "pull_request:12188", + "pull_request:12190", + "pull_request:13056", + "pull_request:13255", + "pull_request:1758", + "pull_request:2254", + "pull_request:2345", + "pull_request:2994", + "pull_request:3135", + "pull_request:3248", + "pull_request:3795", + "pull_request:3945", + "pull_request:420", + "pull_request:4223", + "pull_request:4408", + "pull_request:4461", + "pull_request:4596", + "pull_request:4633", + "pull_request:4694", + "pull_request:4762", + "pull_request:5275", + "pull_request:5369", + "pull_request:5376", + "pull_request:5517", + "pull_request:5697", + "pull_request:5914", + "pull_request:6021", + "pull_request:6035", + "pull_request:6360", + "pull_request:6361", + "pull_request:6402", + "pull_request:6403", + "pull_request:6526", + "pull_request:6556", + "pull_request:6717", + "pull_request:6795", + "pull_request:6884", + "pull_request:6946", + "pull_request:7182", + "pull_request:7786", + "pull_request:9682" + ], + "soft_pairs": [ + "pull_request:1758|pull_request:420", + "pull_request:2254|pull_request:2345", + "pull_request:3795|pull_request:5369", + "pull_request:3945|pull_request:6021", + "pull_request:4461|pull_request:4596", + "pull_request:4633|pull_request:6526", + "pull_request:5376|pull_request:6360", + "pull_request:5697|pull_request:7786", + "pull_request:5914|pull_request:6361", + "pull_request:6035|pull_request:6402", + "pull_request:6035|pull_request:6795", + "pull_request:6402|pull_request:6556", + "pull_request:6403|pull_request:7182", + "pull_request:4408|pull_request:4694", + "pull_request:10259|pull_request:11546", + "pull_request:10376|pull_request:11572", + "pull_request:5517|pull_request:6717", + "pull_request:12188|pull_request:12190", + "pull_request:2994|pull_request:4223", + "pull_request:3135|pull_request:3248", + "pull_request:6884|pull_request:6946", + "pull_request:10378|pull_request:9682", + "pull_request:10481|pull_request:10517", + "pull_request:13056|pull_request:13255", + "pull_request:4762|pull_request:5275" + ], + "prepared_review_unit_hash": "3c2cc9903bc438b5d7a5d10c03b2781993c7f35c160cb829de00f95d358cfc55", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 98099, + "estimated_input_tokens": 24525, + "estimated_eval_tokens": 49306 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it contains many unrelated PRs spanning docs, training scripts, and model-specific loader fixes. The soft pairs mostly share a broad file or subsystem but address different code paths or models, so they should not be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out because the items are unrelated changes rather than duplicates of one underlying fix or feature.", + "best_issue_reason": null, + "best_pr_reason": "No best PR: the set is heterogeneous, so there is no single representative change that safely subsumes the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1758", + "right": "pull_request:420", + "accept": false, + "reason": "Both are docs-related, but one adds checkpoint inference guidance for DreamBooth while the other is a general optimization-section cleanup." + }, + { + "left": "pull_request:2254", + "right": "pull_request:2345", + "accept": false, + "reason": "One documents xFormers training caveats; the other adds a runtime warning in training scripts. Same topic, different change and scope." + }, + { + "left": "pull_request:3795", + "right": "pull_request:5369", + "accept": false, + "reason": "Both are minor docs/pipeline edits, but they target different examples and unrelated issues." + }, + { + "left": "pull_request:3945", + "right": "pull_request:6021", + "accept": false, + "reason": "One adds a rank argument to DreamBooth LoRA; the other adds a variant argument to an advanced SDXL script. Different functionality." + }, + { + "left": "pull_request:4461", + "right": "pull_request:4596", + "accept": false, + "reason": "Both touch conversion scripts, but one adds pipeline_class_name to the original SD conversion script while the other adds the reverse SDXL-to-original conversion script." + }, + { + "left": "pull_request:4633", + "right": "pull_request:6526", + "accept": false, + "reason": "One adds an SDXL ControlNet training example; the other makes the SDXL ControlNet training script torch.compile-compatible. Different artifacts and goals." + }, + { + "left": "pull_request:5376", + "right": "pull_request:6360", + "accept": false, + "reason": "Both involve inpaint/control pipelines, but one adds FreeU across many pipelines and the other adds padding_mask_crop handling for inpainting." + }, + { + "left": "pull_request:5697", + "right": "pull_request:7786", + "accept": false, + "reason": "Both are bug fixes in different pipelines; one fixes PEFT LoRA with xformers, the other fixes Stable Video Diffusion num_videos_per_prompt behavior." + }, + { + "left": "pull_request:5914", + "right": "pull_request:6361", + "accept": false, + "reason": "One fixes a README template in SDXL LoRA training; the other changes LoRA key conversion for advanced training output. Not the same bug." + }, + { + "left": "pull_request:6035", + "right": "pull_request:6402", + "accept": false, + "reason": "One improves help text; the other fixes gradient-checkpointing handling in LoRA training scripts. Different underlying change." + }, + { + "left": "pull_request:6035", + "right": "pull_request:6795", + "accept": false, + "reason": "One is help-text cleanup, while the other adds SDXL micro-conditioning support. No shared bug or fix." + }, + { + "left": "pull_request:6402", + "right": "pull_request:6556", + "accept": false, + "reason": "Both affect LoRA training scripts, but one restores gradient checkpointing while the other makes SDXL LoRA torch.compile compatible." + }, + { + "left": "pull_request:6403", + "right": "pull_request:7182", + "accept": false, + "reason": "Both touch the advanced SDXL DreamBooth script, but one adds WebUI-format support and the other adds EDM-style training. Different features." + }, + { + "left": "pull_request:4408", + "right": "pull_request:4694", + "accept": false, + "reason": "One adds SDXL single-file inpainting support; the other adds ControlNet SDXL inpainting support. Same broad area, different pipeline." + }, + { + "left": "pull_request:10259", + "right": "pull_request:11546", + "accept": false, + "reason": "One enables loading regular Flux LoRAs into Flux Control/Fill; the other supports HunyuanVideo FramePack LoRA loading. Different model families." + }, + { + "left": "pull_request:10376", + "right": "pull_request:11572", + "accept": false, + "reason": "Both add non-diffusers LoRA conversion support, but for different models: HunyuanVideo original-format LoRAs vs LTX-Video LoRAs." + }, + { + "left": "pull_request:5517", + "right": "pull_request:6717", + "accept": false, + "reason": "One is a WIP LCM distillation script, the other is a consistency training example. Different training methods and codepaths." + }, + { + "left": "pull_request:12188", + "right": "pull_request:12190", + "accept": false, + "reason": "Both edit Qwen-Image-Edit, but one fixes a width/height bug and the other removes auto-resize/preset-resolution logic for performance. Not the same fix." + }, + { + "left": "pull_request:2994", + "right": "pull_request:4223", + "accept": false, + "reason": "One fixes type casting in ControlNet training; the other fixes SDXL ControlNet training script argument/config issues. Different bugs." + }, + { + "left": "pull_request:3135", + "right": "pull_request:3248", + "accept": false, + "reason": "Both are ControlNet training script changes, but one enforces divisible-by-8 resizing while the other fixes logging image shape/channel handling." + }, + { + "left": "pull_request:6884", + "right": "pull_request:6946", + "accept": false, + "reason": "Both touch IP-Adapter scaling/loading, but one validates scale length and the other adds low_cpu_mem_usage support. Separate issues." + }, + { + "left": "pull_request:10378", + "right": "pull_request:9682", + "accept": false, + "reason": "Both involve DreamBooth training, but one is an accelerator/distributed-type save hook fix for SD3 and the other is a broader SD3 training updates/bug-fixes PR." + }, + { + "left": "pull_request:10481", + "right": "pull_request:10517", + "accept": false, + "reason": "One changes Sana's default clean_caption flag; the other is a separate Sana 4K pipeline bug fix with no shared patch." + }, + { + "left": "pull_request:13056", + "right": "pull_request:13255", + "accept": false, + "reason": "One introduces Z-Image DreamBooth training assets; the other fixes Z-Image non-diffusers LoRA loading. Related ecosystem, but different changes." + }, + { + "left": "pull_request:4762", + "right": "pull_request:5275", + "accept": false, + "reason": "One updates an SDXL text-to-image README with a VAE note; the other adds a JAX/TPU SDXL research example. Not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 32, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10259", + "pull_request:10318", + "pull_request:10376", + "pull_request:10576", + "pull_request:10854", + "pull_request:11508", + "pull_request:11602", + "pull_request:11670", + "pull_request:11754", + "pull_request:11767", + "pull_request:11796", + "pull_request:12164", + "pull_request:12188", + "pull_request:12190", + "pull_request:12225", + "pull_request:12839", + "pull_request:2037", + "pull_request:243", + "pull_request:4461", + "pull_request:4595", + "pull_request:4666", + "pull_request:5807", + "pull_request:5887", + "pull_request:5915", + "pull_request:6035", + "pull_request:6179", + "pull_request:6190", + "pull_request:6762", + "pull_request:6864", + "pull_request:6975", + "pull_request:7182", + "pull_request:7196", + "pull_request:7212", + "pull_request:7372", + "pull_request:7476", + "pull_request:7720", + "pull_request:7786", + "pull_request:7932", + "pull_request:8696", + "pull_request:8825", + "pull_request:8902", + "pull_request:9139", + "pull_request:9290", + "pull_request:9318", + "pull_request:9445", + "pull_request:9483", + "pull_request:9816" + ], + "soft_pairs": [ + "pull_request:5807|pull_request:7786", + "pull_request:6035|pull_request:7372", + "pull_request:6190|pull_request:6864", + "pull_request:11767|pull_request:11796", + "pull_request:11602|pull_request:12225", + "pull_request:12164|pull_request:12190", + "pull_request:12188|pull_request:12839", + "pull_request:6762|pull_request:8696", + "pull_request:7182|pull_request:7372", + "pull_request:7196|pull_request:8825", + "pull_request:9139|pull_request:9290", + "pull_request:9318|pull_request:9483", + "pull_request:5887|pull_request:5915", + "pull_request:9445|pull_request:9816", + "pull_request:11508|pull_request:11670", + "pull_request:10318|pull_request:10576", + "pull_request:4461|pull_request:4595", + "pull_request:4666|pull_request:6179", + "pull_request:7212|pull_request:7720", + "pull_request:6975|pull_request:7476", + "pull_request:10854|pull_request:11754", + "pull_request:2037|pull_request:243", + "pull_request:7932|pull_request:8902", + "pull_request:10259|pull_request:10376" + ], + "prepared_review_unit_hash": "b05d013161daa45ef6c769470e27bedf93fd94ecc051494bd84afae8f67c97e7", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 97279, + "estimated_input_tokens": 24320, + "estimated_eval_tokens": 48896 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a heterogeneous grab-bag of unrelated PRs (features, tests, docs, CI, and follow-up fixes). None of the soft pairs look like true duplicates of the same underlying change, so all are rejected.", + "confidence": 0.66, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue artifacts are present; among the PRs, #12164 is the most central standalone feature change and underpins later Qwen-Image Edit follow-ups.", + "best_issue_reason": null, + "best_pr_reason": "#12164 is the strongest representative PR by scope and impact, with a broad code change across model, pipeline, and exports rather than a narrow follow-up or docs/test tweak.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5807", + "right": "pull_request:7786", + "accept": false, + "reason": "Different problems: LPW SDXL embedding duplication vs Stable Video Diffusion latent duplication." + }, + { + "left": "pull_request:6035", + "right": "pull_request:7372", + "accept": false, + "reason": "Help-text cleanup in a training script vs adding EDM tags/model-card metadata; related file, different change." + }, + { + "left": "pull_request:6190", + "right": "pull_request:6864", + "accept": false, + "reason": "One is torch.compile test enablement; the other is docs guidance. Not the same bug/change." + }, + { + "left": "pull_request:11767", + "right": "pull_request:11796", + "accept": false, + "reason": "Different fixes: unique memory address for group offloading with disk vs group-offloading/layerwise casting test interaction." + }, + { + "left": "pull_request:11602", + "right": "pull_request:12225", + "accept": false, + "reason": "Separate features for different pipelines: Sana Sprint img2img vs Qwen Image Edit inpainting." + }, + { + "left": "pull_request:12164", + "right": "pull_request:12190", + "accept": false, + "reason": "Base Qwen-Image Edit support vs a later performance tweak in the same file; not the same concrete change." + }, + { + "left": "pull_request:12188", + "right": "pull_request:12839", + "accept": false, + "reason": "Emergency fix for Qwen-Image-Edit vs later 2511 model support; different code-path problems." + }, + { + "left": "pull_request:6762", + "right": "pull_request:8696", + "accept": false, + "reason": "Unrelated flaky-test annotations for different model test suites." + }, + { + "left": "pull_request:7182", + "right": "pull_request:7372", + "accept": false, + "reason": "Training-feature support vs docs/model-card tagging for that training flow; not one mergeable change." + }, + { + "left": "pull_request:7196", + "right": "pull_request:8825", + "accept": false, + "reason": "IP-Adapter training docs/scripts vs adding ControlNet/IP-Adapter support to a K-Diffusion pipeline." + }, + { + "left": "pull_request:9139", + "right": "pull_request:9290", + "accept": false, + "reason": "Flux Dreambooth LoRA code fixes vs README guidance updates; docs-only follow-up, not the same change." + }, + { + "left": "pull_request:9318", + "right": "pull_request:9483", + "accept": false, + "reason": "Nightly CI token change vs model-test expectation updates; unrelated maintenance." + }, + { + "left": "pull_request:5887", + "right": "pull_request:5915", + "accept": false, + "reason": "Specific inpaint-pipeline IP-Adapter support vs broad IP-Adapter Plus support across many pipelines." + }, + { + "left": "pull_request:9445", + "right": "pull_request:9816", + "accept": false, + "reason": "Different Flux community pipelines with different purposes and code paths." + }, + { + "left": "pull_request:11508", + "right": "pull_request:11670", + "accept": false, + "reason": "Torch.compile CI/test infrastructure changes vs a group-offloading test addition; different issues." + }, + { + "left": "pull_request:10318", + "right": "pull_request:10576", + "accept": false, + "reason": "Flux LoRA integration tests vs 8-bit quantized model LoRA-loading tests; same broad topic, different concrete behavior." + }, + { + "left": "pull_request:4461", + "right": "pull_request:4595", + "accept": false, + "reason": "Concrete pipeline_class_name implementation vs a WIP reverse-conversion PR with no diff; not duplicates." + }, + { + "left": "pull_request:4666", + "right": "pull_request:6179", + "accept": false, + "reason": "SDXL ControlNet LoRA support vs fused attention projection support for other pipelines; separate features." + }, + { + "left": "pull_request:7212", + "right": "pull_request:7720", + "accept": false, + "reason": "TemporalBasicTransformerBlock cleanup vs unrelated dead-code removal across scripts and utilities." + }, + { + "left": "pull_request:6975", + "right": "pull_request:7476", + "accept": false, + "reason": "Single-file ControlNet inpaint fix vs a multi-ControlNet torch.compile bug in SDXL inpaint." + }, + { + "left": "pull_request:10854", + "right": "pull_request:11754", + "accept": false, + "reason": "Test-module rename for Hunyuan DiT vs HiDream test-module rename; same pattern, different targets and changes." + }, + { + "left": "pull_request:2037", + "right": "pull_request:243", + "accept": false, + "reason": "Community Stable UnCLIP addition vs a WIP RePaint pipeline draft; not the same artifact." + }, + { + "left": "pull_request:7932", + "right": "pull_request:8902", + "accept": false, + "reason": "LoRA test expectation refresh vs sharded-checkpoint loading fixes; unrelated." + }, + { + "left": "pull_request:10259", + "right": "pull_request:10376", + "accept": false, + "reason": "Flux Control/Fill LoRA-loading support vs HunyuanVideo LoRA format conversion support; different models and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 33, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1012", + "pull_request:10279", + "pull_request:10349", + "pull_request:10402", + "pull_request:10418", + "pull_request:10578", + "pull_request:1063", + "pull_request:1073", + "pull_request:10845", + "pull_request:10922", + "pull_request:10943", + "pull_request:10947", + "pull_request:11130", + "pull_request:11243", + "pull_request:11327", + "pull_request:11348", + "pull_request:11373", + "pull_request:11451", + "pull_request:11552", + "pull_request:1162", + "pull_request:1167", + "pull_request:11740", + "pull_request:1178", + "pull_request:11876", + "pull_request:11953", + "pull_request:121", + "pull_request:12230", + "pull_request:12737", + "pull_request:1283", + "pull_request:12933", + "pull_request:12970", + "pull_request:12983", + "pull_request:1346", + "pull_request:16", + "pull_request:1605", + "pull_request:1960", + "pull_request:1998", + "pull_request:4944", + "pull_request:6573", + "pull_request:6729", + "pull_request:6907", + "pull_request:694", + "pull_request:754", + "pull_request:879", + "pull_request:883", + "pull_request:9542", + "pull_request:9655" + ], + "soft_pairs": [ + "pull_request:6729|pull_request:6907", + "pull_request:12737|pull_request:12970", + "pull_request:4944|pull_request:6573", + "pull_request:12933|pull_request:12983", + "pull_request:11130|pull_request:11876", + "pull_request:1012|pull_request:883", + "pull_request:10279|pull_request:9655", + "pull_request:10349|pull_request:10947", + "pull_request:10402|pull_request:10418", + "pull_request:10578|pull_request:9542", + "pull_request:1063|pull_request:1178", + "pull_request:1073|pull_request:879", + "pull_request:10845|pull_request:11373", + "pull_request:10922|pull_request:12230", + "pull_request:10943|pull_request:11243", + "pull_request:11327|pull_request:11348", + "pull_request:11451|pull_request:11740", + "pull_request:11451|pull_request:11953", + "pull_request:11552|pull_request:11740", + "pull_request:1162|pull_request:754", + "pull_request:1167|pull_request:16", + "pull_request:1167|pull_request:1960", + "pull_request:121|pull_request:694", + "pull_request:1283|pull_request:1605", + "pull_request:1346|pull_request:1998" + ], + "prepared_review_unit_hash": "e10721e034a437b9c90cbf61b86dcdeb26c6e7c0561402035c707377e2d65ab3", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 95879, + "estimated_input_tokens": 23970, + "estimated_eval_tokens": 48196 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes docs-only fixes, model support/features, tests, and unrelated bug fixes. None of the soft pairs look like true duplicates or the same concrete change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits: the items span unrelated topics (DreamBooth training, GGUF docs, Wan/LoRA support, quantization, xDiT docs, etc.) and do not share one underlying fix.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong cluster-wide representative because this is not a duplicate set; the closest pairs still address different code paths or different doc updates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6729", + "right": "pull_request:6907", + "accept": false, + "reason": "Both are model-card/docs maintenance, but they update different training examples and different code paths." + }, + { + "left": "pull_request:12737", + "right": "pull_request:12970", + "accept": false, + "reason": "Both touch parallel/context code, but they fix different transformer implementations and distinct bugs." + }, + { + "left": "pull_request:4944", + "right": "pull_request:6573", + "accept": false, + "reason": "Both involve adapter loading, but one adds IP-Adapter support while the other adds multi-IP-Adapter behavior across many pipelines." + }, + { + "left": "pull_request:12933", + "right": "pull_request:12983", + "accept": false, + "reason": "Both are LTX-2 related, but one adds LoRA support and the other adds single-file loading support." + }, + { + "left": "pull_request:11130", + "right": "pull_request:11876", + "accept": false, + "reason": "Both concern pipeline-level quantization, but one adds the feature and the other makes QoL/pipeline loading refinements." + }, + { + "left": "pull_request:1012", + "right": "pull_request:883", + "accept": false, + "reason": "Both mention DreamBooth text-encoder training, but one sets train mode and the other adds training support/docs; not the same fix." + }, + { + "left": "pull_request:10279", + "right": "pull_request:9655", + "accept": false, + "reason": "Both are docs fixes, but they cover unrelated documentation areas and different link/image corrections." + }, + { + "left": "pull_request:10349", + "right": "pull_request:10947", + "accept": false, + "reason": "Both touch single-file loading, but for different model families and different conversion logic." + }, + { + "left": "pull_request:10402", + "right": "pull_request:10418", + "accept": false, + "reason": "Both are internal-link docs fixes, but they target different docs pages and unrelated broken links." + }, + { + "left": "pull_request:10578", + "right": "pull_request:9542", + "accept": false, + "reason": "Both are LoRA-related, but one enables 4-bit Flux LoRA loading while the other handles Kohya Flux LoRAs with text encoders." + }, + { + "left": "pull_request:1063", + "right": "pull_request:1178", + "accept": false, + "reason": "Both are tiny correctness/typo fixes, but they change unrelated files and different code paths." + }, + { + "left": "pull_request:1073", + "right": "pull_request:879", + "accept": false, + "reason": "Both edit DreamBooth/community docs, but they fix different README issues and are not the same change." + }, + { + "left": "pull_request:10845", + "right": "pull_request:11373", + "accept": false, + "reason": "Both are Flux-related training docs/code, but one fixes T5 embedding handling and the other addresses ControlNet training stability/validation." + }, + { + "left": "pull_request:10922", + "right": "pull_request:12230", + "accept": false, + "reason": "Both are Wan docs, but one introduces the Wan implementation and the other only renames/adjusts docs wording." + }, + { + "left": "pull_request:10943", + "right": "pull_request:11243", + "accept": false, + "reason": "Both are Wan LoRA-related, but one adds Wan LoRA support and the other adds a specific musubi-Wan LoRA conversion path." + }, + { + "left": "pull_request:11327", + "right": "pull_request:11348", + "accept": false, + "reason": "Both update AuraFlow docs, but one adds a compile snippet and the other adds an explanatory note." + }, + { + "left": "pull_request:11451", + "right": "pull_request:11740", + "accept": false, + "reason": "One is a Wan torch.compile test and the other is GGUF compile/offload tests; different models and goals." + }, + { + "left": "pull_request:11451", + "right": "pull_request:11953", + "accept": false, + "reason": "One tests Wan torch.compile behavior, the other adds fast inference slices for HiDream; unrelated." + }, + { + "left": "pull_request:11552", + "right": "pull_request:11740", + "accept": false, + "reason": "Both are tests for different models; no shared bug or change path." + }, + { + "left": "pull_request:1162", + "right": "pull_request:754", + "accept": false, + "reason": "Both are docs/community oriented but fix unrelated content and different files." + }, + { + "left": "pull_request:1167", + "right": "pull_request:16", + "accept": false, + "reason": "Both touch README.md, but they are unrelated README edits years apart." + }, + { + "left": "pull_request:1167", + "right": "pull_request:1960", + "accept": false, + "reason": "Both are README updates, but they affect different examples and different issues." + }, + { + "left": "pull_request:121", + "right": "pull_request:694", + "accept": false, + "reason": "Both are early README/README-like docs edits, but they do not address the same change." + }, + { + "left": "pull_request:1283", + "right": "pull_request:1605", + "accept": false, + "reason": "Both are Versatile Diffusion changes, but one adds the model and the other only adds an upcast_attention parameter." + }, + { + "left": "pull_request:1346", + "right": "pull_request:1998", + "accept": false, + "reason": "Both are README updates, but they fix different example snippets and documentation content." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 34, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1360", + "pull_request:1394", + "pull_request:1415", + "pull_request:1496", + "pull_request:1622", + "pull_request:168", + "pull_request:1682", + "pull_request:1723", + "pull_request:1758", + "pull_request:1763", + "pull_request:1769", + "pull_request:178", + "pull_request:2009", + "pull_request:2062", + "pull_request:2406", + "pull_request:252", + "pull_request:2526", + "pull_request:2566", + "pull_request:2583", + "pull_request:2604", + "pull_request:2760", + "pull_request:2915", + "pull_request:3123", + "pull_request:3201", + "pull_request:325", + "pull_request:3265", + "pull_request:330", + "pull_request:3321", + "pull_request:333", + "pull_request:3334", + "pull_request:3458", + "pull_request:3460", + "pull_request:3472", + "pull_request:3531", + "pull_request:3542", + "pull_request:365", + "pull_request:3795", + "pull_request:3945", + "pull_request:400", + "pull_request:4952", + "pull_request:555", + "pull_request:645", + "pull_request:689", + "pull_request:694", + "pull_request:888", + "pull_request:889", + "pull_request:903" + ], + "soft_pairs": [ + "pull_request:1360|pull_request:3201", + "pull_request:1360|pull_request:3265", + "pull_request:1394|pull_request:1682", + "pull_request:1415|pull_request:1622", + "pull_request:1496|pull_request:178", + "pull_request:1496|pull_request:555", + "pull_request:1496|pull_request:889", + "pull_request:168|pull_request:903", + "pull_request:1723|pull_request:333", + "pull_request:1758|pull_request:3795", + "pull_request:1763|pull_request:3531", + "pull_request:1769|pull_request:2062", + "pull_request:2009|pull_request:888", + "pull_request:2406|pull_request:3321", + "pull_request:252|pull_request:694", + "pull_request:2526|pull_request:2915", + "pull_request:2526|pull_request:689", + "pull_request:2566|pull_request:3123", + "pull_request:2583|pull_request:2604", + "pull_request:2760|pull_request:645", + "pull_request:325|pull_request:400", + "pull_request:330|pull_request:365", + "pull_request:3334|pull_request:4952", + "pull_request:3458|pull_request:3542", + "pull_request:3460|pull_request:3542", + "pull_request:3472|pull_request:3945" + ], + "prepared_review_unit_hash": "33fb2f18989d3633d2f6458b513b0fcd6ee648fe21cab9b4e3a5665d7f510d13", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 98238, + "estimated_input_tokens": 24560, + "estimated_eval_tokens": 49376 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated Stable Diffusion/docs/version/test PRs; there is one likely duplicate pair around inpaint LoRA support, but the rest are not the same underlying change.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #2009 is the strongest anchor if a canonical PR is needed: it is the broadest substantive Stable Diffusion implementation change in the set, with real code impact and high discussion, whereas most other items are narrow docs/version/test edits.", + "best_issue_reason": null, + "best_pr_reason": "PR #2009 is the best overall representative of the cluster because it is a core code change in the Stable Diffusion area and has the most substance; the other PRs are mostly peripheral fixes or unrelated follow-ups.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1360", + "right": "pull_request:3201", + "accept": false, + "reason": "Different changes: #1360 adds a community K-Diffusion pipeline, while #3201 reverts and rewrites lpw_stable_diffusion imports/code. Same area, but not the same bug/change." + }, + { + "left": "pull_request:1360", + "right": "pull_request:3265", + "accept": false, + "reason": "#1360 is a new community pipeline addition; #3265 is a revert-of-revert for lpw_stable_diffusion. Not the same underlying change." + }, + { + "left": "pull_request:1394", + "right": "pull_request:1682", + "accept": false, + "reason": "Both bump versions, but to different releases and in different commits; this is not a duplicate bug/fix." + }, + { + "left": "pull_request:1415", + "right": "pull_request:1622", + "accept": false, + "reason": "#1415 is a broad stable-diffusion bug-fix sweep; #1622 fixes LPW pipeline compatibility with PIL imports/version handling. Different concrete changes." + }, + { + "left": "pull_request:1496", + "right": "pull_request:178", + "accept": false, + "reason": "Docs for Habana Gaudi vs adding Stable Diffusion tests; unrelated edits." + }, + { + "left": "pull_request:1496", + "right": "pull_request:555", + "accept": false, + "reason": "Habana docs and Flax VAE implementation are unrelated." + }, + { + "left": "pull_request:1496", + "right": "pull_request:889", + "accept": false, + "reason": "Habana docs vs adding pipeline components support; not the same change." + }, + { + "left": "pull_request:168", + "right": "pull_request:903", + "accept": false, + "reason": "#168 introduces Stable Diffusion pipeline support; #903 deprecates the old inpainting pipeline. Related subsystem, but different code-path changes." + }, + { + "left": "pull_request:1723", + "right": "pull_request:333", + "accept": false, + "reason": "Latent Diffusion latents input support vs Latent Diffusion uncond type hints. Same file family, different change." + }, + { + "left": "pull_request:1758", + "right": "pull_request:3795", + "accept": false, + "reason": "Both are docs, but one is Dreambooth-specific and the other is pipeline overview updates. Not the same underlying change." + }, + { + "left": "pull_request:1763", + "right": "pull_request:3531", + "accept": false, + "reason": "Both touch attention mask handling, but one removes an unused mask from versatile attention and the other changes padding validation in attention_processor. Different bug fixes." + }, + { + "left": "pull_request:1769", + "right": "pull_request:2062", + "accept": false, + "reason": "#1769 only adjusts UnCLIP tests/memory thresholds; #2062 fixes paint-by-example CPU offload behavior in code. Not duplicates." + }, + { + "left": "pull_request:2009", + "right": "pull_request:888", + "accept": false, + "reason": "#2009 adds textual inversion loading across stable diffusion pipelines; #888 adds CycleDiffusion. Same ecosystem, different feature work." + }, + { + "left": "pull_request:2406", + "right": "pull_request:3321", + "accept": false, + "reason": "Both edit the controlling-generation docs, but they add different content at different times; not the same change." + }, + { + "left": "pull_request:252", + "right": "pull_request:694", + "accept": false, + "reason": "README edits in different locations and for different content (general examples vs schedulers README). Not a duplicate." + }, + { + "left": "pull_request:2526", + "right": "pull_request:2915", + "accept": false, + "reason": "Both relax slow-test expectations, but they target different tests/pipelines and different expected outputs." + }, + { + "left": "pull_request:2526", + "right": "pull_request:689", + "accept": false, + "reason": "#2526 is test-threshold churn across several pipelines; #689 changes attention math implementation. Different scope and purpose." + }, + { + "left": "pull_request:2566", + "right": "pull_request:3123", + "accept": false, + "reason": "TensorBoard logger fix in training script vs a JAX textual inversion typo fix. Unrelated." + }, + { + "left": "pull_request:2583", + "right": "pull_request:2604", + "accept": false, + "reason": "Both fix docs links, but one is a dataset URL in text-to-image docs and the other is a Colab link in the index. Different edits." + }, + { + "left": "pull_request:2760", + "right": "pull_request:645", + "accept": false, + "reason": "Different textual inversion-related projects: one is a new quantization/distillation example, the other adds checkpoint saving during training." + }, + { + "left": "pull_request:325", + "right": "pull_request:400", + "accept": false, + "reason": "#325 is type hints for Score SDE VE; #400 is docs plus pipeline/training references. Not the same concrete change." + }, + { + "left": "pull_request:330", + "right": "pull_request:365", + "accept": false, + "reason": "Both are type-hint cleanups, but on different model classes (UNet vs VAE). Not duplicates." + }, + { + "left": "pull_request:3334", + "right": "pull_request:4952", + "accept": false, + "reason": "Both touch pipeline_utils docs, but they fix different docstrings/method descriptions. Not the same underlying change." + }, + { + "left": "pull_request:3458", + "right": "pull_request:3542", + "accept": false, + "reason": "Both mention LoRA support, but #3458 targets Stable Diffusion inpaint while #3542 targets ControlNet. Different pipelines and code paths." + }, + { + "left": "pull_request:3460", + "right": "pull_request:3542", + "accept": false, + "reason": "#3460 adds inpaint LoRA-scale support; #3542 adds ControlNet LoRA support. Same theme, different implementation targets." + }, + { + "left": "pull_request:3472", + "right": "pull_request:3945", + "accept": false, + "reason": "Dreambooth IF docs vs adding a rank parameter to LoRA training script. Different artifacts and purposes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 35, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3642", + "pull_request:3795", + "pull_request:386", + "pull_request:3945", + "pull_request:4044", + "pull_request:4113", + "pull_request:4211", + "pull_request:4227", + "pull_request:4503", + "pull_request:4595", + "pull_request:4652", + "pull_request:4696", + "pull_request:4717", + "pull_request:4745", + "pull_request:4749", + "pull_request:4760", + "pull_request:4847", + "pull_request:4893", + "pull_request:4946", + "pull_request:4947", + "pull_request:5033", + "pull_request:5051", + "pull_request:5144", + "pull_request:5361", + "pull_request:5411", + "pull_request:5511", + "pull_request:5671", + "pull_request:5750", + "pull_request:6140", + "pull_request:6145", + "pull_request:6174", + "pull_request:6202", + "pull_request:6216", + "pull_request:6245", + "pull_request:6361", + "pull_request:6376", + "pull_request:6401", + "pull_request:6403", + "pull_request:6499", + "pull_request:6523", + "pull_request:6528", + "pull_request:6566", + "pull_request:6724", + "pull_request:6775", + "pull_request:6999", + "pull_request:7118", + "pull_request:7360", + "pull_request:903" + ], + "soft_pairs": [ + "pull_request:3642|pull_request:5750", + "pull_request:3795|pull_request:4227", + "pull_request:386|pull_request:903", + "pull_request:3945|pull_request:6403", + "pull_request:4044|pull_request:4211", + "pull_request:4113|pull_request:6202", + "pull_request:4503|pull_request:6724", + "pull_request:4595|pull_request:4847", + "pull_request:4595|pull_request:6376", + "pull_request:4652|pull_request:4749", + "pull_request:4696|pull_request:4947", + "pull_request:4717|pull_request:6528", + "pull_request:4745|pull_request:5033", + "pull_request:4760|pull_request:5411", + "pull_request:4893|pull_request:4946", + "pull_request:5051|pull_request:7118", + "pull_request:5144|pull_request:7360", + "pull_request:5361|pull_request:5671", + "pull_request:5511|pull_request:6999", + "pull_request:6140|pull_request:6499", + "pull_request:6145|pull_request:6523", + "pull_request:6174|pull_request:6775", + "pull_request:6216|pull_request:6401", + "pull_request:6245|pull_request:6361", + "pull_request:6361|pull_request:6566" + ], + "prepared_review_unit_hash": "1e13107558ff550275e6091ede07969f39c5bcaa34eec9632ab71cf1df132a08", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 93993, + "estimated_input_tokens": 23499, + "estimated_eval_tokens": 47254 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are unrelated docs/example tweaks or distinct bug fixes across different pipelines and training scripts. The soft edges are driven by broad keyword overlap (Stable Diffusion, LoRA, docs, inpaint) rather than the same underlying change, so none should be merged as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents the whole cluster; the items span unrelated documentation edits and separate bug fixes, so there is no safe canonical PR for duplicate triage.", + "best_issue_reason": null, + "best_pr_reason": "PR 5033 is the strongest standalone bug fix in the set because it fixes a concrete inpaint code-path across multiple pipelines and tests, but it is still not a duplicate of the other PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3642", + "right": "pull_request:5750", + "accept": false, + "reason": "Different changes: one adds an SD TensorRT inpainting community pipeline, the other is a draft reference-only tech introduction with no diff preview." + }, + { + "left": "pull_request:3795", + "right": "pull_request:4227", + "accept": false, + "reason": "Both are docs-related, but one updates pipeline overview pages while the other adds a missing import in a textual inversion example." + }, + { + "left": "pull_request:386", + "right": "pull_request:903", + "accept": false, + "reason": "PR 386 adds Stable Diffusion pipeline docs and code, while PR 903 deprecates the old inpainting pipeline; related area but different change and intent." + }, + { + "left": "pull_request:3945", + "right": "pull_request:6403", + "accept": false, + "reason": "Both touch LoRA training scripts, but one adds a rank parameter and the other adds WebUI format support with different state-dict handling." + }, + { + "left": "pull_request:4044", + "right": "pull_request:4211", + "accept": false, + "reason": "A docstring typo fix in AudioLDM is unrelated to an SDXL inpainting docstring/example fix." + }, + { + "left": "pull_request:4113", + "right": "pull_request:6202", + "accept": false, + "reason": "Both involve Stable Diffusion XL inpaint-related code, but one fixes dtype handling while the other changes prompt embedding flow in Stable Diffusion/Alt Diffusion code." + }, + { + "left": "pull_request:4503", + "right": "pull_request:6724", + "accept": false, + "reason": "Both edit lora.md, but they address different documentation topics: SDXL support clarification vs rank description." + }, + { + "left": "pull_request:4595", + "right": "pull_request:4847", + "accept": false, + "reason": "Unrelated: one is a WIP pipeline conversion idea, the other adds a community Prompt Free Diffusion example." + }, + { + "left": "pull_request:4595", + "right": "pull_request:6376", + "accept": false, + "reason": "No shared bug/change: one is a WIP conversion PR, the other is a stable video diffusion + ControlNet community pipeline." + }, + { + "left": "pull_request:4652", + "right": "pull_request:4749", + "accept": false, + "reason": "Both are LoRA-related tests/examples, but one changes test serialization behavior and the other fixes SDXL LoRA checkpointing." + }, + { + "left": "pull_request:4696", + "right": "pull_request:4947", + "accept": false, + "reason": "PR 4696 adds T2I-Adapter-XL models/docs; PR 4947 adds training tests and an adapter model arg. Same subsystem, different change." + }, + { + "left": "pull_request:4717", + "right": "pull_request:6528", + "accept": false, + "reason": "A docs formatting fix for ControlNet is unrelated to an IP-Adapter code fix in ControlNet img2img." + }, + { + "left": "pull_request:4745", + "right": "pull_request:5033", + "accept": false, + "reason": "Different bug fixes: optional-component loading in from_pretrained vs inpaint masking/latents handling." + }, + { + "left": "pull_request:4760", + "right": "pull_request:5411", + "accept": false, + "reason": "Separate docs fixes: one corrects a broken CLIP doc link, the other fixes a textual inversion training-doc link." + }, + { + "left": "pull_request:4893", + "right": "pull_request:4946", + "accept": false, + "reason": "Both touch lora/training docs, but one explains LoRA fusion scale behavior and the other updates the training overview entries." + }, + { + "left": "pull_request:5051", + "right": "pull_request:7118", + "accept": false, + "reason": "Both are docs typos, but for different subjects: Wuerstchen docs vs IP-Adapter usage docs." + }, + { + "left": "pull_request:5144", + "right": "pull_request:7360", + "accept": false, + "reason": "BLIP Diffusion sample-code parenthesis fix is unrelated to TCD scheduler docs/type annotations." + }, + { + "left": "pull_request:5361", + "right": "pull_request:5671", + "accept": false, + "reason": "These are separate CI/dependency workflow changes with different purposes and file scope." + }, + { + "left": "pull_request:5511", + "right": "pull_request:6999", + "accept": false, + "reason": "Different Stable Video/SDXL scheduler and pipeline fixes; one targets ensemble-of-experts timesteps, the other image-tensor preprocessing." + }, + { + "left": "pull_request:6140", + "right": "pull_request:6499", + "accept": false, + "reason": "Both are loader docs, but one adds IP-Adapter API docs and the other updates PEFT adapter API wording." + }, + { + "left": "pull_request:6145", + "right": "pull_request:6523", + "accept": false, + "reason": "Both touch LCM distillation scripts, but one is a comments cleanup and the other fixes a time_cond_proj_dim bug." + }, + { + "left": "pull_request:6174", + "right": "pull_request:6775", + "accept": false, + "reason": "A markdown backslash fix in LoRA docs is unrelated to adding a missing DDPMScheduler parameter docstring." + }, + { + "left": "pull_request:6216", + "right": "pull_request:6401", + "accept": false, + "reason": "Both edit the PEFT LoRA test suite, but one fixes a version-specific test failure and the other removes unused test scaffolding." + }, + { + "left": "pull_request:6245", + "right": "pull_request:6361", + "accept": false, + "reason": "Wuerstchen prior fp16/LoRA argument correction is unrelated to advanced SDXL DreamBooth LoRA key-format conversion." + }, + { + "left": "pull_request:6361", + "right": "pull_request:6566", + "accept": false, + "reason": "Same file and broad area, but different fixes: one converts keys for LoRA format, the other fixes checkpoint resume behavior." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 36, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10206", + "pull_request:10271", + "pull_request:10351", + "pull_request:10810", + "pull_request:11595", + "pull_request:11755", + "pull_request:11994", + "pull_request:12069", + "pull_request:12269", + "pull_request:12416", + "pull_request:13042", + "pull_request:13119", + "pull_request:13209", + "pull_request:13313", + "pull_request:13469", + "pull_request:1788", + "pull_request:2415", + "pull_request:2795", + "pull_request:325", + "pull_request:360", + "pull_request:3641", + "pull_request:4188", + "pull_request:4262", + "pull_request:4272", + "pull_request:5164", + "pull_request:5506", + "pull_request:5792", + "pull_request:5943", + "pull_request:6426", + "pull_request:6448", + "pull_request:6528", + "pull_request:6555", + "pull_request:6564", + "pull_request:6816", + "pull_request:6875", + "pull_request:6941", + "pull_request:7086", + "pull_request:7634", + "pull_request:788", + "pull_request:8630", + "pull_request:8825", + "pull_request:8926", + "pull_request:9152", + "pull_request:9251", + "pull_request:9257", + "pull_request:9705", + "pull_request:9775" + ], + "soft_pairs": [ + "pull_request:6426|pull_request:6448", + "pull_request:6528|pull_request:6875", + "pull_request:6555|pull_request:8630", + "pull_request:6816|pull_request:8926", + "pull_request:6941|pull_request:8825", + "pull_request:7086|pull_request:8825", + "pull_request:9152|pull_request:9775", + "pull_request:9257|pull_request:9705", + "pull_request:12269|pull_request:12416", + "pull_request:10271|pull_request:9251", + "pull_request:13119|pull_request:13313", + "pull_request:13209|pull_request:13313", + "pull_request:1788|pull_request:788", + "pull_request:2415|pull_request:2795", + "pull_request:3641|pull_request:5506", + "pull_request:4188|pull_request:4262", + "pull_request:4272|pull_request:5792", + "pull_request:5164|pull_request:5943", + "pull_request:6564|pull_request:7634", + "pull_request:10206|pull_request:9257", + "pull_request:10810|pull_request:11595", + "pull_request:325|pull_request:360", + "pull_request:13042|pull_request:13469", + "pull_request:10351|pull_request:11755", + "pull_request:11994|pull_request:12069" + ], + "prepared_review_unit_hash": "fd5aa1f8b3cf5223074a32bdbeb67f7c51861fa6b9b577bfdcc4b06345aae886", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 99552, + "estimated_input_tokens": 24888, + "estimated_eval_tokens": 50032 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated small PRs across LoRA loading/conversion, training-script fixes, docs, tests, and model-specific support. The soft pairs share superficial themes or filenames in a few cases, but they do not look like the same concrete bug/change.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "If one PR has to anchor the cluster, 13313 is the best representative of the core LoRA theme: it makes a concrete loader/conversion fix in the main code path rather than just touching tests/docs. That said, the cluster is too heterogeneous to treat as a real duplicate set.", + "best_issue_reason": null, + "best_pr_reason": "13313 is the strongest standalone PR here because it addresses an actual production code path with a focused LoRA loading/conversion fix and is more central than the surrounding test/doc-only changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6426", + "right": "pull_request:6448", + "accept": false, + "reason": "Both are LoRA-related and share a test file, but 6426 is a deprecation/backend cleanup across models while 6448 is only test assertion maintenance. Not the same bug/change." + }, + { + "left": "pull_request:6528", + "right": "pull_request:6875", + "accept": false, + "reason": "6528 adds IP-Adapter code for ControlNet img2img; 6875 only updates regression expectations. Same area, different change." + }, + { + "left": "pull_request:6555", + "right": "pull_request:8630", + "accept": false, + "reason": "Different scripts, different model families, and different fixes: torch.compile compatibility vs SD3 CLIP text-encoder training support." + }, + { + "left": "pull_request:6816", + "right": "pull_request:8926", + "accept": false, + "reason": "Both are SDXL training-related, but one fixes DDP+PEFT behavior and the other fixes checkpoint saving in a separate advanced script. Not the same code-path problem." + }, + { + "left": "pull_request:6941", + "right": "pull_request:8825", + "accept": false, + "reason": "Both involve ControlNet/IP-Adapter, but they target different pipelines and different feature sets. Too broad to merge as duplicates." + }, + { + "left": "pull_request:7086", + "right": "pull_request:8825", + "accept": false, + "reason": "Community AnimateDiff img2vid/controlnet fix vs SDXL K-Diffusion ControlNet/IP-Adapter support are unrelated concrete changes." + }, + { + "left": "pull_request:9152", + "right": "pull_request:9775", + "accept": false, + "reason": "One is distributed inference docs, the other is a README note for flux quantization. Same general documentation bucket only." + }, + { + "left": "pull_request:9257", + "right": "pull_request:9705", + "accept": false, + "reason": "Different Flux training docs/code changes: one fixes the dreambooth script, the other updates a README. Not the same bug." + }, + { + "left": "pull_request:12269", + "right": "pull_request:12416", + "accept": false, + "reason": "Both touch modular pipelines, but one adds Flux Kontext support and the other adds QwenImage Edit Plus support. Different model integrations, not duplicates." + }, + { + "left": "pull_request:10271", + "right": "pull_request:9251", + "accept": false, + "reason": "A test refactor in TorchAO serialization is unrelated to the workflow change that triggers fast tests on patch branches." + }, + { + "left": "pull_request:13119", + "right": "pull_request:13313", + "accept": false, + "reason": "Both are LoRA/Flux conversion-related, but 13119 handles non-diffusers Flux2 key conversion while 13313 implements Klein/Kohya loading support. Related area, different concrete fixes." + }, + { + "left": "pull_request:13209", + "right": "pull_request:13313", + "accept": false, + "reason": "13209 normalizes ZImage LoRA key names; 13313 adds Klein LoRA loading. Same utility file, but not the same underlying bug." + }, + { + "left": "pull_request:1788", + "right": "pull_request:788", + "accept": false, + "reason": "DDIM scheduler behavior fix vs PIL sampler warning cleanup are unrelated." + }, + { + "left": "pull_request:2415", + "right": "pull_request:2795", + "accept": false, + "reason": "Pipeline module version-check cleanup and a LoRA training-script tweak are not the same change." + }, + { + "left": "pull_request:3641", + "right": "pull_request:5506", + "accept": false, + "reason": "Inpainting latent handling and PEFT/xFormers attention handling are different code paths and bugs." + }, + { + "left": "pull_request:4188", + "right": "pull_request:4262", + "accept": false, + "reason": "Both are training-script fixes, but one is dataset fingerprinting and the other is SDXL DreamBooth LoRA logic. Not a duplicate." + }, + { + "left": "pull_request:4272", + "right": "pull_request:5792", + "accept": false, + "reason": "Missing optional imports cleanup and import-path fixes are in the same broad area, but not the same bug or feature." + }, + { + "left": "pull_request:5164", + "right": "pull_request:5943", + "accept": false, + "reason": "FreeU mechanism addition and transformer2d gradient checkpointing support are distinct core features." + }, + { + "left": "pull_request:6564", + "right": "pull_request:7634", + "accept": false, + "reason": "IP Adapter path generation fix and high-res UNet fix are unrelated pipeline/model issues." + }, + { + "left": "pull_request:10206", + "right": "pull_request:9257", + "accept": false, + "reason": "LoRA unload support for Flux Control and a dreambooth script bug fix are different parts of the stack." + }, + { + "left": "pull_request:10810", + "right": "pull_request:11595", + "accept": false, + "reason": "Both involve Flux LoRA support, but 10810 is conversion-utility work and 11595 is a load_lora_weights base-layer selection fix. Different concrete problems." + }, + { + "left": "pull_request:325", + "right": "pull_request:360", + "accept": false, + "reason": "These are separate type-hint cleanups for different classes/files, not a duplicate bug or change." + }, + { + "left": "pull_request:13042", + "right": "pull_request:13469", + "accept": false, + "reason": "QwenImage prompt-isolation test fixes are unrelated to deprecated attention-processor test handling." + }, + { + "left": "pull_request:10351", + "right": "pull_request:11755", + "accept": false, + "reason": "One is a LoRA test helper change; the other is a CPU offload fix for an LTX pipeline. Not the same underlying issue." + }, + { + "left": "pull_request:11994", + "right": "pull_request:12069", + "accept": false, + "reason": "Hugging Face auth/docs and AudioLDM2 test fixes are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 37, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10127", + "pull_request:102", + "pull_request:1020", + "pull_request:10232", + "pull_request:10250", + "pull_request:10278", + "pull_request:10319", + "pull_request:1041", + "pull_request:10429", + "pull_request:10504", + "pull_request:10505", + "pull_request:10517", + "pull_request:10609", + "pull_request:10756", + "pull_request:10826", + "pull_request:10858", + "pull_request:1087", + "pull_request:10888", + "pull_request:10978", + "pull_request:10990", + "pull_request:1104", + "pull_request:11124", + "pull_request:11142", + "pull_request:113", + "pull_request:11325", + "pull_request:1136", + "pull_request:11469", + "pull_request:11531", + "pull_request:11634", + "pull_request:11755", + "pull_request:11802", + "pull_request:11847", + "pull_request:11879", + "pull_request:12030", + "pull_request:12035", + "pull_request:12360", + "pull_request:12714", + "pull_request:17", + "pull_request:2062", + "pull_request:22", + "pull_request:2208", + "pull_request:2369", + "pull_request:2476", + "pull_request:408", + "pull_request:774", + "pull_request:9266", + "pull_request:9832" + ], + "soft_pairs": [ + "pull_request:10127|pull_request:11469", + "pull_request:11755|pull_request:12714", + "pull_request:11802|pull_request:12714", + "pull_request:2369|pull_request:2476", + "pull_request:102|pull_request:1020", + "pull_request:1020|pull_request:2208", + "pull_request:10232|pull_request:9266", + "pull_request:10250|pull_request:10429", + "pull_request:10278|pull_request:10319", + "pull_request:10278|pull_request:11325", + "pull_request:10319|pull_request:11325", + "pull_request:1041|pull_request:17", + "pull_request:1041|pull_request:408", + "pull_request:10504|pull_request:10609", + "pull_request:10504|pull_request:11879", + "pull_request:10505|pull_request:10609", + "pull_request:10505|pull_request:11879", + "pull_request:10517|pull_request:11142", + "pull_request:10756|pull_request:9832", + "pull_request:10826|pull_request:12030", + "pull_request:10858|pull_request:11847", + "pull_request:1087|pull_request:1104", + "pull_request:10888|pull_request:12035", + "pull_request:10978|pull_request:10990", + "pull_request:1104|pull_request:22", + "pull_request:11124|pull_request:11755", + "pull_request:113|pull_request:774", + "pull_request:11325|pull_request:11847", + "pull_request:11325|pull_request:12360", + "pull_request:1136|pull_request:2062", + "pull_request:11531|pull_request:11634" + ], + "prepared_review_unit_hash": "940be00f67def3f16bb71dc3c1289c768327a10fd1df25d80bbf7063a2563d7e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 96123, + "estimated_input_tokens": 24031, + "estimated_eval_tokens": 48318 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most of the soft pairs are only superficially similar (same subsystem, same docs/tests area, or same buzzwords) but do not fix the same concrete bug/change. The only clear duplicate is the two identical PRs removing print statements from tests, where one is effectively a duplicate/branch variant of the other.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10505 is the best canonical PR because it is the merged, finalized version of the exact same test-print removal change shown in 10504.", + "best_issue_reason": null, + "best_pr_reason": "PR 10505 is the strongest cluster representative because it is the only exact duplicate-style match and has the complete merged implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10127", + "right": "pull_request:11469", + "accept": false, + "reason": "Both touch test xfails, but they are unrelated test failures for different pipelines and different conditions." + }, + { + "left": "pull_request:11755", + "right": "pull_request:12714", + "accept": false, + "reason": "Different fixes in different areas: LTX cpu offload vs Flux2 LoRA NaN test." + }, + { + "left": "pull_request:11802", + "right": "pull_request:12714", + "accept": false, + "reason": "Both involve LoRA-related testing, but one removes deprecated utilities and the other fixes Flux2 LoRA behavior; not the same bug." + }, + { + "left": "pull_request:2369", + "right": "pull_request:2476", + "accept": false, + "reason": "Same pipeline family, but one adds the pipeline and docs while the other changes image-size behavior; different changes." + }, + { + "left": "pull_request:102", + "right": "pull_request:1020", + "accept": false, + "reason": "Both update training examples/docs, but they modify different content and workflows; not the same change." + }, + { + "left": "pull_request:1020", + "right": "pull_request:2208", + "accept": false, + "reason": "Both are docs updates, but one is a broad training-doc refresh and the other is xFormers installation guidance; unrelated." + }, + { + "left": "pull_request:10232", + "right": "pull_request:9266", + "accept": false, + "reason": "Both are minor docs/community edits, but they change different pages and different content." + }, + { + "left": "pull_request:10250", + "right": "pull_request:10429", + "accept": false, + "reason": "Different code paths: a 1D UNet downsample bug versus an NPU rotary-embedding dtype fix." + }, + { + "left": "pull_request:10278", + "right": "pull_request:10319", + "accept": false, + "reason": "Different doc corrections in different files; no shared underlying bug or change." + }, + { + "left": "pull_request:10278", + "right": "pull_request:11325", + "accept": false, + "reason": "Both are doc fixes, but they address unrelated examples/docstring issues." + }, + { + "left": "pull_request:10319", + "right": "pull_request:11325", + "accept": false, + "reason": "Unrelated docstring corrections in different pipelines." + }, + { + "left": "pull_request:1041", + "right": "pull_request:17", + "accept": false, + "reason": "A test-fix PR and a typo/docs PR are not the same underlying change." + }, + { + "left": "pull_request:1041", + "right": "pull_request:408", + "accept": false, + "reason": "Different scopes: test/runtime fixes versus general docs typos." + }, + { + "left": "pull_request:10504", + "right": "pull_request:10609", + "accept": false, + "reason": "One removes print statements from tests; the other edits docs and unrelated code cleanup." + }, + { + "left": "pull_request:10504", + "right": "pull_request:11879", + "accept": false, + "reason": "Test print cleanup is unrelated to deleting deprecated pipeline tests." + }, + { + "left": "pull_request:10505", + "right": "pull_request:10609", + "accept": false, + "reason": "Different change types and files; only superficial similarity in being small chores." + }, + { + "left": "pull_request:10505", + "right": "pull_request:11879", + "accept": false, + "reason": "Different change types: test print cleanup versus removal of deprecated pipeline tests." + }, + { + "left": "pull_request:10517", + "right": "pull_request:11142", + "accept": false, + "reason": "Same broad Sana area, but one is a 4K pipeline bug and the other is a conversion-script fix for Sana-Sprint." + }, + { + "left": "pull_request:10756", + "right": "pull_request:9832", + "accept": false, + "reason": "Both relate to quantization, but one adds a backend while the other adds CI coverage; not the same fix." + }, + { + "left": "pull_request:10826", + "right": "pull_request:12030", + "accept": false, + "reason": "Both mention IP-Adapter, but they target different pipelines and different feature work." + }, + { + "left": "pull_request:10858", + "right": "pull_request:11847", + "accept": false, + "reason": "Both are docs, but one fixes the pipelines overview and the other fixes a single-file example." + }, + { + "left": "pull_request:1087", + "right": "pull_request:1104", + "accept": false, + "reason": "Different test/CI changes: slow-test fixes versus adding slow MPS test jobs." + }, + { + "left": "pull_request:10888", + "right": "pull_request:12035", + "accept": false, + "reason": "Both involve controlnet/autopipeline naming, but they add support for different pipelines and features." + }, + { + "left": "pull_request:10978", + "right": "pull_request:10990", + "accept": false, + "reason": "Both touch LoRA/quantization-related code, but they fix different concrete issues and code paths." + }, + { + "left": "pull_request:1104", + "right": "pull_request:22", + "accept": false, + "reason": "A CI test-job addition and a model-loading warning fix are unrelated." + }, + { + "left": "pull_request:11124", + "right": "pull_request:11755", + "accept": false, + "reason": "Different pipelines and problems: a CogView4 test update versus an LTX offload fix." + }, + { + "left": "pull_request:113", + "right": "pull_request:774", + "accept": false, + "reason": "Both are test fixes, but they address completely different behaviors and eras of the codebase." + }, + { + "left": "pull_request:11325", + "right": "pull_request:11847", + "accept": false, + "reason": "Different docs updates in different pipeline pages; not the same underlying issue." + }, + { + "left": "pull_request:11325", + "right": "pull_request:12360", + "accept": false, + "reason": "One is a docstring/doc example cleanup; the other is a Chroma guidance text fix." + }, + { + "left": "pull_request:1136", + "right": "pull_request:2062", + "accept": false, + "reason": "One fixes example README links; the other changes paint-by-example CPU-offload behavior." + }, + { + "left": "pull_request:11531", + "right": "pull_request:11634", + "accept": false, + "reason": "Both update torchao docs, but they fix different documentation details and are not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 38, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1162", + "pull_request:11847", + "pull_request:11983", + "pull_request:12069", + "pull_request:121", + "pull_request:1218", + "pull_request:12265", + "pull_request:1230", + "pull_request:12360", + "pull_request:12418", + "pull_request:12420", + "pull_request:12425", + "pull_request:12438", + "pull_request:12490", + "pull_request:12530", + "pull_request:12552", + "pull_request:12731", + "pull_request:12757", + "pull_request:12775", + "pull_request:12782", + "pull_request:1291", + "pull_request:12971", + "pull_request:12984", + "pull_request:13039", + "pull_request:13055", + "pull_request:13254", + "pull_request:13315", + "pull_request:13335", + "pull_request:13368", + "pull_request:13369", + "pull_request:13424", + "pull_request:13469", + "pull_request:1373", + "pull_request:1394", + "pull_request:1406", + "pull_request:1408", + "pull_request:1412", + "pull_request:1418", + "pull_request:1425", + "pull_request:1490", + "pull_request:17", + "pull_request:1861", + "pull_request:21", + "pull_request:22", + "pull_request:260", + "pull_request:286", + "pull_request:3241", + "pull_request:3345" + ], + "soft_pairs": [ + "pull_request:1162|pull_request:3345", + "pull_request:11847|pull_request:12360", + "pull_request:11983|pull_request:12731", + "pull_request:11983|pull_request:12757", + "pull_request:11983|pull_request:12971", + "pull_request:12069|pull_request:12418", + "pull_request:121|pull_request:1861", + "pull_request:121|pull_request:260", + "pull_request:1218|pull_request:1408", + "pull_request:12265|pull_request:12552", + "pull_request:1230|pull_request:3241", + "pull_request:12418|pull_request:12530", + "pull_request:12420|pull_request:12425", + "pull_request:12438|pull_request:13055", + "pull_request:12490|pull_request:13424", + "pull_request:12775|pull_request:12984", + "pull_request:12782|pull_request:13469", + "pull_request:1291|pull_request:286", + "pull_request:13039|pull_request:13469", + "pull_request:13254|pull_request:13368", + "pull_request:13254|pull_request:13369", + "pull_request:13315|pull_request:13368", + "pull_request:13315|pull_request:13369", + "pull_request:13335|pull_request:13368", + "pull_request:13335|pull_request:13369", + "pull_request:1373|pull_request:1418", + "pull_request:1394|pull_request:1490", + "pull_request:1406|pull_request:1408", + "pull_request:1412|pull_request:21", + "pull_request:1412|pull_request:22", + "pull_request:1425|pull_request:17" + ], + "prepared_review_unit_hash": "0925dc1bb41b88b5d534ae7d3e98ff57bce88929cb87f838de977b140d7f13ef", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 97824, + "estimated_input_tokens": 24456, + "estimated_eval_tokens": 49168 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Loose cluster of unrelated low-risk maintenance PRs: docs typos/link fixes, test/CI flakiness tweaks, and a few feature/docs additions. None of the soft pairs look like the same underlying change, so I rejected them all.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12265 is the broadest representative of the cluster's dominant minor-maintenance theme, spanning multiple typo/cleanup edits across files; there is no shared concrete bug or change tying the whole set together.", + "best_issue_reason": null, + "best_pr_reason": "PR 12265 is the best overall representative because it is a wide, low-risk cleanup PR and most closely matches the cluster's common pattern of small documentation/code polish rather than a specific feature or bug fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1162", + "right": "pull_request:3345", + "accept": false, + "reason": "Both are docs/link fixes, but they touch different pages and different targets; not the same underlying change." + }, + { + "left": "pull_request:11847", + "right": "pull_request:12360", + "accept": false, + "reason": "Unrelated docs edits: a single-file WAN example fix vs Chroma guidance wording changes." + }, + { + "left": "pull_request:11983", + "right": "pull_request:12731", + "accept": false, + "reason": "Both are documentation cleanups, but they modify different install/usage docs with different content changes." + }, + { + "left": "pull_request:11983", + "right": "pull_request:12757", + "accept": false, + "reason": "Different docs pages and edits; one is installation instructions, the other is attention-backend wording/formatting." + }, + { + "left": "pull_request:11983", + "right": "pull_request:12971", + "accept": false, + "reason": "Installation docs vs distributed-inference docs repositioning; not the same fix." + }, + { + "left": "pull_request:12069", + "right": "pull_request:12418", + "accept": false, + "reason": "AudioLDM2 code/test fix vs an unrelated CI xfail for a different test suite." + }, + { + "left": "pull_request:121", + "right": "pull_request:1861", + "accept": false, + "reason": "README examples update vs training-loss computation changes in example scripts; different changes and code paths." + }, + { + "left": "pull_request:121", + "right": "pull_request:260", + "accept": false, + "reason": "Both are README/example polish, but they affect different files and unrelated example sections." + }, + { + "left": "pull_request:1218", + "right": "pull_request:1408", + "accept": false, + "reason": "Same script file, but the changes are about different conversion behaviors and different bugs." + }, + { + "left": "pull_request:12265", + "right": "pull_request:12552", + "accept": false, + "reason": "Both are typo/cleanup PRs, but they are separate docs edits in different files with no shared concrete issue." + }, + { + "left": "pull_request:1230", + "right": "pull_request:3241", + "accept": false, + "reason": "Different test fixes for different pipelines and failure modes; not mergeable as one change." + }, + { + "left": "pull_request:12418", + "right": "pull_request:12530", + "accept": false, + "reason": "Both xfail tests, but they target different failing suites and different transformers-version issues." + }, + { + "left": "pull_request:12420", + "right": "pull_request:12425", + "accept": false, + "reason": "Related imports, but one is attention dispatch and the other is context parallel hooks; distinct code paths." + }, + { + "left": "pull_request:12438", + "right": "pull_request:13055", + "accept": false, + "reason": "Different loader vs Wan pipeline fixes; same broad area only, not the same bug." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13424", + "accept": false, + "reason": "Separate test-suite adjustments for different autoencoder/pipeline components." + }, + { + "left": "pull_request:12775", + "right": "pull_request:12984", + "accept": false, + "reason": "Two different feature/docs additions for different models; no shared underlying change." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13469", + "accept": false, + "reason": "Different test fixes in unrelated suites: Hunyuan offloading vs attention-processor deprecation handling." + }, + { + "left": "pull_request:1291", + "right": "pull_request:286", + "accept": false, + "reason": "Both touch links, but they fix different broken references in different README files." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13469", + "accept": false, + "reason": "Wan CPU upcasting fix vs attention-processor test deprecation; unrelated code paths." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13368", + "accept": false, + "reason": "Different transformer/autoencoder test refactors; not the same model or failure." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13369", + "accept": false, + "reason": "LTX transformer test refactor vs AutoencoderDC test refactor; unrelated changes." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13368", + "accept": false, + "reason": "Sana transformer tests vs AutoencoderKL tests; different components and fixes." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13369", + "accept": false, + "reason": "Sana transformer tests vs AutoencoderDC tests; not the same issue." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13368", + "accept": false, + "reason": "Cosmos transformer tests vs AutoencoderKL tests; different test targets and motivations." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13369", + "accept": false, + "reason": "Cosmos transformer tests vs AutoencoderDC tests; unrelated refactors." + }, + { + "left": "pull_request:1373", + "right": "pull_request:1418", + "accept": false, + "reason": "Versatile Diffusion extension work vs fixes for remaining tests; different scope and code paths." + }, + { + "left": "pull_request:1394", + "right": "pull_request:1490", + "accept": false, + "reason": "A version bump and a later deprecation sweep are repository-wide maintenance, but they are not the same change." + }, + { + "left": "pull_request:1406", + "right": "pull_request:1408", + "accept": false, + "reason": "Related Stable Diffusion conversion-era work, but one is deprecating sample size and the other fixes a conversion script bug." + }, + { + "left": "pull_request:1412", + "right": "pull_request:21", + "accept": false, + "reason": "Stable Diffusion 2 inpainting integration tests vs core model-test additions; unrelated artifacts." + }, + { + "left": "pull_request:1412", + "right": "pull_request:22", + "accept": false, + "reason": "SD2 inpainting tests vs a GLIDE UNet logging fix; different subsystems and purposes." + }, + { + "left": "pull_request:1425", + "right": "pull_request:17", + "accept": false, + "reason": "Both are typo cleanups, but they are from different docs eras/files and do not represent the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 39, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1555", + "pull_request:1558", + "pull_request:1610", + "pull_request:1690", + "pull_request:17", + "pull_request:178", + "pull_request:1861", + "pull_request:1937", + "pull_request:2006", + "pull_request:2010", + "pull_request:2077", + "pull_request:21", + "pull_request:2134", + "pull_request:2137", + "pull_request:2172", + "pull_request:2175", + "pull_request:22", + "pull_request:2290", + "pull_request:2320", + "pull_request:2333", + "pull_request:252", + "pull_request:2526", + "pull_request:2540", + "pull_request:2542", + "pull_request:2566", + "pull_request:2569", + "pull_request:260", + "pull_request:2873", + "pull_request:299", + "pull_request:3066", + "pull_request:3266", + "pull_request:3318", + "pull_request:353", + "pull_request:3549", + "pull_request:3710", + "pull_request:3848", + "pull_request:392", + "pull_request:407", + "pull_request:408", + "pull_request:421", + "pull_request:4320", + "pull_request:4822", + "pull_request:568", + "pull_request:710", + "pull_request:798", + "pull_request:840", + "pull_request:889", + "pull_request:904" + ], + "soft_pairs": [ + "pull_request:1555|pull_request:2006", + "pull_request:1558|pull_request:260", + "pull_request:1558|pull_request:353", + "pull_request:1558|pull_request:407", + "pull_request:1558|pull_request:798", + "pull_request:1610|pull_request:178", + "pull_request:1690|pull_request:840", + "pull_request:17|pull_request:568", + "pull_request:178|pull_request:21", + "pull_request:178|pull_request:22", + "pull_request:178|pull_request:299", + "pull_request:178|pull_request:904", + "pull_request:1861|pull_request:252", + "pull_request:1937|pull_request:2172", + "pull_request:2006|pull_request:3266", + "pull_request:2010|pull_request:421", + "pull_request:2077|pull_request:3318", + "pull_request:2134|pull_request:260", + "pull_request:2134|pull_request:353", + "pull_request:2134|pull_request:407", + "pull_request:2134|pull_request:798", + "pull_request:2137|pull_request:408", + "pull_request:2137|pull_request:568", + "pull_request:2175|pull_request:3549", + "pull_request:2175|pull_request:889", + "pull_request:2290|pull_request:710", + "pull_request:2320|pull_request:4320", + "pull_request:2333|pull_request:408", + "pull_request:252|pull_request:260", + "pull_request:2526|pull_request:3066", + "pull_request:2540|pull_request:2873", + "pull_request:2540|pull_request:392", + "pull_request:2542|pull_request:3848", + "pull_request:2566|pull_request:4320", + "pull_request:2566|pull_request:4822", + "pull_request:2569|pull_request:3710" + ], + "prepared_review_unit_hash": "c6db2294197a1a937018fcdec34f20d0237ca744cf2c6d9c14f356930e03b732", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 36, + "serialized_chars": 94484, + "estimated_input_tokens": 23621, + "estimated_eval_tokens": 47498 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose grab-bag of typo/doc/example/test/utility PRs. Despite some superficial title/file overlap, they address different files and different concrete changes, so I would not merge any pair as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a true canonical for the whole cluster because the items are heterogeneous. If forced to pick one representative, #568 is the broadest typo/docs cleanup and best matches the dominant theme, but it is not a duplicate umbrella for the rest.", + "best_issue_reason": null, + "best_pr_reason": "#568 is the closest thing to a representative PR: it is a broad typo cleanup across multiple docs/code files and best captures the recurring maintenance theme. Still, the cluster is not actually one underlying change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1555", + "right": "pull_request:2006", + "accept": false, + "reason": "Both touch community utilities, but one fixes custom-pipeline cache_dir handling and the other fixes checkpoint merger behavior; different code paths and bugs." + }, + { + "left": "pull_request:1558", + "right": "pull_request:260", + "accept": false, + "reason": "Both are typo fixes, but in different files and unrelated text; no shared underlying bug." + }, + { + "left": "pull_request:1558", + "right": "pull_request:353", + "accept": false, + "reason": "Different typo locations in different pipeline code; unrelated corrections." + }, + { + "left": "pull_request:1558", + "right": "pull_request:407", + "accept": false, + "reason": "Same broad stable-diffusion area, but one is docs typo cleanup and the other changes an API call for attention slicing; not the same fix." + }, + { + "left": "pull_request:1558", + "right": "pull_request:798", + "accept": false, + "reason": "Both are docstring/typo edits, but they fix separate unet docstrings in different modules." + }, + { + "left": "pull_request:1610", + "right": "pull_request:178", + "accept": false, + "reason": "Tests-related only at a high level; one adds pipeline attention logging behavior, the other updates stable diffusion test expectations." + }, + { + "left": "pull_request:1690", + "right": "pull_request:840", + "accept": false, + "reason": "Same example file family, but #840 introduces the one-step dummy pipeline while #1690 changes its return value for testing; not duplicate changes." + }, + { + "left": "pull_request:17", + "right": "pull_request:568", + "accept": false, + "reason": "Both are README/docs typos, but they target different text and different files; not the same underlying issue." + }, + { + "left": "pull_request:178", + "right": "pull_request:21", + "accept": false, + "reason": "Both modify tests, but they cover different model paths and different expected outputs." + }, + { + "left": "pull_request:178", + "right": "pull_request:22", + "accept": false, + "reason": "Same test file, but different model tests and unrelated assertions." + }, + { + "left": "pull_request:178", + "right": "pull_request:299", + "accept": false, + "reason": "One is a stable-diffusion pipeline test, the other is a scheduler implementation change; unrelated." + }, + { + "left": "pull_request:178", + "right": "pull_request:904", + "accept": false, + "reason": "Different layers entirely: pipeline test expectations versus inpainting implementation refactor." + }, + { + "left": "pull_request:1861", + "right": "pull_request:252", + "accept": false, + "reason": "Both touch examples/docs style, but one changes loss computation in training scripts while the other updates README examples; unrelated fixes." + }, + { + "left": "pull_request:1937", + "right": "pull_request:2172", + "accept": false, + "reason": "Both involve training/conversion utilities, but one fixes EMA extraction logic and the other fixes multi-GPU validation setup; different problems." + }, + { + "left": "pull_request:2006", + "right": "pull_request:3266", + "accept": false, + "reason": "Both are community-pipeline related, but one fixes checkpoint merging and the other fixes version loading for community pipelines; not the same bug." + }, + { + "left": "pull_request:2010", + "right": "pull_request:421", + "accept": false, + "reason": "Completely different subsystems: LR scheduler init versus scheduler docs typo." + }, + { + "left": "pull_request:2077", + "right": "pull_request:3318", + "accept": false, + "reason": "Both touch textual inversion, but #2077 adds validation images while #3318 is a WIP IF/textual-inversion feature branch; not the same concrete change." + }, + { + "left": "pull_request:2134", + "right": "pull_request:260", + "accept": false, + "reason": "Both are typo/doc fixes, but in different documents and unrelated wording." + }, + { + "left": "pull_request:2134", + "right": "pull_request:353", + "accept": false, + "reason": "Different typo in a README versus a model error message; unrelated." + }, + { + "left": "pull_request:2134", + "right": "pull_request:407", + "accept": false, + "reason": "One is a README typo, the other changes stable-diffusion attention-slicing API usage; not duplicate." + }, + { + "left": "pull_request:2134", + "right": "pull_request:798", + "accept": false, + "reason": "Both are wording fixes, but they patch different docstrings in different UNet variants." + }, + { + "left": "pull_request:2137", + "right": "pull_request:408", + "accept": false, + "reason": "Both are language fixes, but one is loaders.py copy and the other is broader docs phrasing; different changes." + }, + { + "left": "pull_request:2137", + "right": "pull_request:568", + "accept": false, + "reason": "Same general typo theme, but they edit different files and different statements; not the same bug." + }, + { + "left": "pull_request:2175", + "right": "pull_request:3549", + "accept": false, + "reason": "Both are docs updates, but one adds components to the pipeline API and the other revises evaluation content; unrelated." + }, + { + "left": "pull_request:2175", + "right": "pull_request:889", + "accept": false, + "reason": "One updates docs, the other adds a new API and tests for components; different scope and intent." + }, + { + "left": "pull_request:2290", + "right": "pull_request:710", + "accept": false, + "reason": "Both fix docstrings, but in different pipelines/models and with different text." + }, + { + "left": "pull_request:2320", + "right": "pull_request:4320", + "accept": false, + "reason": "Different typo fixes in different docs/API paths; no shared underlying bug." + }, + { + "left": "pull_request:2333", + "right": "pull_request:408", + "accept": false, + "reason": "Ethical-guidelines doc rewrite versus general docs typo cleanup; unrelated content." + }, + { + "left": "pull_request:252", + "right": "pull_request:260", + "accept": false, + "reason": "Both are README/example updates, but they address different wording and sections." + }, + { + "left": "pull_request:2526", + "right": "pull_request:3066", + "accept": false, + "reason": "Both are slow-test expectation adjustments, but they affect different test suites and models." + }, + { + "left": "pull_request:2540", + "right": "pull_request:2873", + "accept": false, + "reason": "Different docs sections and different edits; one is ControlNet docs, the other is CONTRIBUTING phrasing." + }, + { + "left": "pull_request:2540", + "right": "pull_request:392", + "accept": false, + "reason": "Both are docs, but one is a small ControlNet docs fix while the other is a large conceptual contribution rewrite." + }, + { + "left": "pull_request:2542", + "right": "pull_request:3848", + "accept": false, + "reason": "One is docs/index navigation and the other is VAE conversion code; unrelated." + }, + { + "left": "pull_request:2566", + "right": "pull_request:4320", + "accept": false, + "reason": "Both mention training/docs fixes, but one changes tensorboard tracker handling and the other is an ONNX docs typo; unrelated." + }, + { + "left": "pull_request:2566", + "right": "pull_request:4822", + "accept": false, + "reason": "Different code paths: tensorboard logging in training versus a LoRA loader docstring/warning typo." + }, + { + "left": "pull_request:2569", + "right": "pull_request:3710", + "accept": false, + "reason": "OpenVINO docs versus image-processor docs; both documentation, but not the same topic or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 40, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2587", + "pull_request:260", + "pull_request:2637", + "pull_request:2639", + "pull_request:2642", + "pull_request:2647", + "pull_request:2684", + "pull_request:2869", + "pull_request:2911", + "pull_request:2912", + "pull_request:2913", + "pull_request:2922", + "pull_request:2924", + "pull_request:2925", + "pull_request:2926", + "pull_request:2927", + "pull_request:2929", + "pull_request:2930", + "pull_request:2931", + "pull_request:2932", + "pull_request:2933", + "pull_request:2935", + "pull_request:2936", + "pull_request:2937", + "pull_request:2938", + "pull_request:2939", + "pull_request:3310", + "pull_request:353", + "pull_request:3549", + "pull_request:407", + "pull_request:798", + "pull_request:972", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:2587|pull_request:3549", + "pull_request:260|pull_request:353", + "pull_request:260|pull_request:407", + "pull_request:260|pull_request:798", + "pull_request:260|pull_request:972", + "pull_request:260|pull_request:975", + "pull_request:2637|pull_request:2869", + "pull_request:2637|pull_request:2911", + "pull_request:2637|pull_request:2912", + "pull_request:2637|pull_request:2913", + "pull_request:2637|pull_request:2922", + "pull_request:2637|pull_request:2924", + "pull_request:2637|pull_request:2925", + "pull_request:2637|pull_request:2926", + "pull_request:2637|pull_request:2927", + "pull_request:2637|pull_request:2929", + "pull_request:2637|pull_request:2930", + "pull_request:2637|pull_request:2931", + "pull_request:2637|pull_request:2932", + "pull_request:2637|pull_request:2933", + "pull_request:2637|pull_request:2935", + "pull_request:2637|pull_request:2936", + "pull_request:2637|pull_request:2937", + "pull_request:2637|pull_request:2938", + "pull_request:2637|pull_request:2939", + "pull_request:2639|pull_request:3310", + "pull_request:2642|pull_request:2869", + "pull_request:2642|pull_request:2911", + "pull_request:2642|pull_request:2912", + "pull_request:2642|pull_request:2913", + "pull_request:2642|pull_request:2924", + "pull_request:2642|pull_request:2925", + "pull_request:2642|pull_request:2926", + "pull_request:2642|pull_request:2927", + "pull_request:2642|pull_request:2929", + "pull_request:2642|pull_request:2930", + "pull_request:2642|pull_request:2931", + "pull_request:2642|pull_request:2932", + "pull_request:2642|pull_request:2933", + "pull_request:2642|pull_request:2935", + "pull_request:2642|pull_request:2936", + "pull_request:2642|pull_request:2937", + "pull_request:2642|pull_request:2938", + "pull_request:2642|pull_request:2939", + "pull_request:2647|pull_request:2869", + "pull_request:2647|pull_request:2911", + "pull_request:2647|pull_request:2912", + "pull_request:2647|pull_request:2913", + "pull_request:2647|pull_request:2924", + "pull_request:2647|pull_request:2925", + "pull_request:2647|pull_request:2926", + "pull_request:2647|pull_request:2927", + "pull_request:2647|pull_request:2929", + "pull_request:2647|pull_request:2930", + "pull_request:2647|pull_request:2931", + "pull_request:2647|pull_request:2932", + "pull_request:2647|pull_request:2933", + "pull_request:2647|pull_request:2935", + "pull_request:2647|pull_request:2936", + "pull_request:2647|pull_request:2937", + "pull_request:2647|pull_request:2938", + "pull_request:2647|pull_request:2939", + "pull_request:2684|pull_request:2869", + "pull_request:2684|pull_request:2911", + "pull_request:2684|pull_request:2912", + "pull_request:2684|pull_request:2913", + "pull_request:2684|pull_request:2926", + "pull_request:2684|pull_request:2927", + "pull_request:2684|pull_request:2929", + "pull_request:2684|pull_request:2930", + "pull_request:2684|pull_request:2931", + "pull_request:2684|pull_request:2932" + ], + "prepared_review_unit_hash": "b4ed489bcb952ac91076366f81142068254f39cdd0908fd6795c5aedd2808447", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 33, + "item_count": 33, + "soft_pair_count": 72, + "serialized_chars": 71847, + "estimated_input_tokens": 17962, + "estimated_eval_tokens": 36180 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All candidates are unrelated documentation/typo cleanup PRs in different files; none look like the same underlying change or a mergeable duplicate pair.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are isolated docs/typo edits across different files and topics, not one shared fix.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good global representative; the cluster is too heterogeneous to anchor on one PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2587", + "right": "pull_request:3549", + "accept": false, + "reason": "Translated docs addition vs evaluation-doc content update; different pages and different changes." + }, + { + "left": "pull_request:260", + "right": "pull_request:353", + "accept": false, + "reason": "README typo fix vs code typo fix in unet blocks; unrelated artifacts and code paths." + }, + { + "left": "pull_request:260", + "right": "pull_request:407", + "accept": false, + "reason": "README typo fix vs stable diffusion attention-slicing code fix; not the same change." + }, + { + "left": "pull_request:260", + "right": "pull_request:798", + "accept": false, + "reason": "README typo fix vs UNet docstring typo; unrelated files and scopes." + }, + { + "left": "pull_request:260", + "right": "pull_request:972", + "accept": false, + "reason": "README typo fix vs README parameter typo fix, but in a different file/context and not the same issue." + }, + { + "left": "pull_request:260", + "right": "pull_request:975", + "accept": false, + "reason": "README typo fix vs docs title capitalization fix; different targets and edits." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2869", + "accept": false, + "reason": "Quicktour wording tweak vs Paint by Example paper punctuation; different docs pages." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2911", + "accept": false, + "reason": "Quicktour wording tweak vs image_variation paper punctuation; unrelated documentation edits." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2912", + "accept": false, + "reason": "Quicktour wording tweak vs ControlNet note punctuation; different docs sections." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2913", + "accept": false, + "reason": "Quicktour wording tweak vs stable_unclip prose punctuation; unrelated doc cleanup." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2922", + "accept": false, + "reason": "Quicktour wording tweak vs unclip paper punctuation; different docs pages." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2924", + "accept": false, + "reason": "Quicktour wording tweak vs VQ diffusion paper punctuation; unrelated docs cleanup." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2925", + "accept": false, + "reason": "Quicktour wording tweak vs schedulers overview punctuation; different page and content." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2926", + "accept": false, + "reason": "Quicktour wording tweak vs DDIM heading capitalization; not the same issue." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2927", + "accept": false, + "reason": "Quicktour wording tweak vs DDIM inverse sentence punctuation; unrelated docs edit." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2929", + "accept": false, + "reason": "Quicktour wording tweak vs DDPM heading capitalization; different page and change." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2930", + "accept": false, + "reason": "Quicktour wording tweak vs DPM discrete punctuation; unrelated docs cleanup." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2931", + "accept": false, + "reason": "Quicktour wording tweak vs DPM ancestral punctuation; different scheduler doc page." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2932", + "accept": false, + "reason": "Quicktour wording tweak vs Euler ancestral link formatting; unrelated docs edit." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2933", + "accept": false, + "reason": "Quicktour wording tweak vs Heun punctuation; different docs page and wording." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2935", + "accept": false, + "reason": "Quicktour wording tweak vs RePaint punctuation; unrelated documentation changes." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2936", + "accept": false, + "reason": "Quicktour wording tweak vs UniPC link formatting; different docs page and fix." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2937", + "accept": false, + "reason": "Quicktour wording tweak vs ScoreSdeVe heading capitalization; unrelated docs edit." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2938", + "accept": false, + "reason": "Quicktour wording tweak vs ScoreSdeVp heading capitalization; different page and change." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2939", + "accept": false, + "reason": "Quicktour wording tweak vs VQ diffusion punctuation; unrelated docs cleanup." + }, + { + "left": "pull_request:2639", + "right": "pull_request:3310", + "accept": false, + "reason": "Basic training docs edit vs stable diffusion docs edit; different tutorials and fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2869", + "accept": false, + "reason": "Loading docs punctuation fix vs Paint by Example paper punctuation; unrelated pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2911", + "accept": false, + "reason": "Loading docs punctuation fix vs image_variation punctuation; different docs pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2912", + "accept": false, + "reason": "Loading docs punctuation fix vs ControlNet punctuation; unrelated documentation edits." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2913", + "accept": false, + "reason": "Loading docs punctuation fix vs stable_unclip punctuation; different docs sections." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2924", + "accept": false, + "reason": "Loading docs punctuation fix vs VQ diffusion punctuation; unrelated pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2925", + "accept": false, + "reason": "Loading docs punctuation fix vs schedulers overview punctuation; different content and scope." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2926", + "accept": false, + "reason": "Loading docs punctuation fix vs DDIM capitalization; not the same underlying fix." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2927", + "accept": false, + "reason": "Loading docs punctuation fix vs DDIM inverse punctuation; unrelated docs edit." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2929", + "accept": false, + "reason": "Loading docs punctuation fix vs DDPM capitalization; different page and edit." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2930", + "accept": false, + "reason": "Loading docs punctuation fix vs DPM discrete punctuation; unrelated docs cleanup." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2931", + "accept": false, + "reason": "Loading docs punctuation fix vs DPM ancestral punctuation; different docs page." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2932", + "accept": false, + "reason": "Loading docs punctuation fix vs Euler ancestral link formatting; unrelated content." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2933", + "accept": false, + "reason": "Loading docs punctuation fix vs Heun punctuation; different docs page and change." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2935", + "accept": false, + "reason": "Loading docs punctuation fix vs RePaint punctuation; unrelated docs edits." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2936", + "accept": false, + "reason": "Loading docs punctuation fix vs UniPC link formatting; different page and fix." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2937", + "accept": false, + "reason": "Loading docs punctuation fix vs ScoreSdeVe capitalization; unrelated docs change." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2938", + "accept": false, + "reason": "Loading docs punctuation fix vs ScoreSdeVp capitalization; different docs page." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2939", + "accept": false, + "reason": "Loading docs punctuation fix vs VQ diffusion punctuation; unrelated cleanup." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2869", + "accept": false, + "reason": "Schedulers docs wording tweak vs Paint by Example paper punctuation; unrelated pages." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2911", + "accept": false, + "reason": "Schedulers docs wording tweak vs image_variation punctuation; different docs pages." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2912", + "accept": false, + "reason": "Schedulers docs wording tweak vs ControlNet punctuation; unrelated docs cleanup." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2913", + "accept": false, + "reason": "Schedulers docs wording tweak vs stable_unclip punctuation; different content." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2924", + "accept": false, + "reason": "Schedulers docs wording tweak vs VQ diffusion punctuation; unrelated change." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2925", + "accept": false, + "reason": "Schedulers docs wording tweak vs schedulers overview punctuation; different page and wording." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2926", + "accept": false, + "reason": "Schedulers docs wording tweak vs DDIM capitalization; not the same issue." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2927", + "accept": false, + "reason": "Schedulers docs wording tweak vs DDIM inverse punctuation; unrelated docs fix." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2929", + "accept": false, + "reason": "Schedulers docs wording tweak vs DDPM capitalization; different page and change." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2930", + "accept": false, + "reason": "Schedulers docs wording tweak vs DPM discrete punctuation; unrelated docs edit." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2931", + "accept": false, + "reason": "Schedulers docs wording tweak vs DPM ancestral punctuation; different scheduler doc." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2932", + "accept": false, + "reason": "Schedulers docs wording tweak vs Euler ancestral link formatting; unrelated docs cleanup." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2933", + "accept": false, + "reason": "Schedulers docs wording tweak vs Heun punctuation; different page and change." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2935", + "accept": false, + "reason": "Schedulers docs wording tweak vs RePaint punctuation; unrelated documentation fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2936", + "accept": false, + "reason": "Schedulers docs wording tweak vs UniPC link formatting; different page and issue." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2937", + "accept": false, + "reason": "Schedulers docs wording tweak vs ScoreSdeVe capitalization; unrelated docs change." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2938", + "accept": false, + "reason": "Schedulers docs wording tweak vs ScoreSdeVp capitalization; different page." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2939", + "accept": false, + "reason": "Schedulers docs wording tweak vs VQ diffusion punctuation; unrelated cleanup." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2869", + "accept": false, + "reason": "Custom pipeline docs punctuation vs Paint by Example paper punctuation; unrelated pages." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2911", + "accept": false, + "reason": "Custom pipeline docs punctuation vs image_variation punctuation; different docs topics." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2912", + "accept": false, + "reason": "Custom pipeline docs punctuation vs ControlNet punctuation; unrelated docs edit." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2913", + "accept": false, + "reason": "Custom pipeline docs punctuation vs stable_unclip punctuation; different page and change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2926", + "accept": false, + "reason": "Custom pipeline docs punctuation vs DDIM capitalization; not the same fix." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2927", + "accept": false, + "reason": "Custom pipeline docs punctuation vs DDIM inverse punctuation; unrelated documentation edit." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2929", + "accept": false, + "reason": "Custom pipeline docs punctuation vs DDPM capitalization; different page." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2930", + "accept": false, + "reason": "Custom pipeline docs punctuation vs DPM discrete punctuation; unrelated docs cleanup." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2931", + "accept": false, + "reason": "Custom pipeline docs punctuation vs DPM ancestral punctuation; different scheduler doc." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2932", + "accept": false, + "reason": "Custom pipeline docs punctuation vs Euler ancestral link formatting; unrelated change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 41, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2684", + "pull_request:2757", + "pull_request:2762", + "pull_request:2808", + "pull_request:2812", + "pull_request:2813", + "pull_request:2823", + "pull_request:2833", + "pull_request:286", + "pull_request:2873", + "pull_request:2915", + "pull_request:2933", + "pull_request:2936", + "pull_request:2937", + "pull_request:2938", + "pull_request:2958", + "pull_request:3068", + "pull_request:309", + "pull_request:3132", + "pull_request:3166", + "pull_request:3226", + "pull_request:3241", + "pull_request:3331", + "pull_request:3397", + "pull_request:3446", + "pull_request:3458", + "pull_request:3460", + "pull_request:353", + "pull_request:3554", + "pull_request:3555", + "pull_request:3614", + "pull_request:3641", + "pull_request:37", + "pull_request:3806", + "pull_request:3842", + "pull_request:3850", + "pull_request:4058", + "pull_request:407", + "pull_request:4203", + "pull_request:4822", + "pull_request:4862", + "pull_request:5292", + "pull_request:55", + "pull_request:798", + "pull_request:904", + "pull_request:972", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:2684|pull_request:2933", + "pull_request:2684|pull_request:2936", + "pull_request:2684|pull_request:2937", + "pull_request:2684|pull_request:2938", + "pull_request:2757|pull_request:904", + "pull_request:2762|pull_request:2833", + "pull_request:2808|pull_request:3842", + "pull_request:2812|pull_request:2823", + "pull_request:2813|pull_request:2823", + "pull_request:286|pull_request:309", + "pull_request:2873|pull_request:4058", + "pull_request:2915|pull_request:3614", + "pull_request:2958|pull_request:3226", + "pull_request:2958|pull_request:3458", + "pull_request:2958|pull_request:3460", + "pull_request:3068|pull_request:3166", + "pull_request:3068|pull_request:3554", + "pull_request:3068|pull_request:3555", + "pull_request:3132|pull_request:4822", + "pull_request:3241|pull_request:5292", + "pull_request:3331|pull_request:4203", + "pull_request:3397|pull_request:3806", + "pull_request:3446|pull_request:3641", + "pull_request:353|pull_request:407", + "pull_request:353|pull_request:798", + "pull_request:353|pull_request:972", + "pull_request:353|pull_request:975", + "pull_request:3614|pull_request:3850", + "pull_request:3641|pull_request:4862", + "pull_request:37|pull_request:55" + ], + "prepared_review_unit_hash": "d7c714f6611152ccbeeadea44bbe782a6edc2be66d34848586b4fb4c3c0ff35a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 89192, + "estimated_input_tokens": 22298, + "estimated_eval_tokens": 44852 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly unrelated small docs/test/code PRs that share only broad topical similarity. I don\u2019t see any issue artifacts, and none of the soft-edge pairs look like the same concrete change; the only clear overlap is superficial doc similarity, not a mergeable duplicate.", + "confidence": 0.83, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue artifacts are present; among the PRs, #3460 is the strongest standalone bugfix because it fixes a concrete inpainting-pipeline behavior and includes tests, unlike the many tiny docs/typo edits.", + "best_issue_reason": null, + "best_pr_reason": "#3460 is the best single PR candidate by substance and specificity: it addresses an actual runtime bug in the inpainting pipeline and is well-scoped with test coverage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2684", + "right": "pull_request:2933", + "accept": false, + "reason": "Both are tiny docs edits, but they touch different pages and different wording; not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2936", + "accept": false, + "reason": "One updates the custom pipeline overview docs; the other changes UniPC scheduler docs. Same broad docs category, but different artifacts and changes." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2937", + "accept": false, + "reason": "Different docs pages and unrelated wording fixes; no shared code-path or same change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2938", + "accept": false, + "reason": "Same broad docs hygiene theme, but these are separate scheduler docs pages with distinct edits." + }, + { + "left": "pull_request:2757", + "right": "pull_request:904", + "accept": false, + "reason": "Both are test-related, but they fix different pipelines and different bugs: depth batching vs inpainting implementation." + }, + { + "left": "pull_request:2762", + "right": "pull_request:2833", + "accept": false, + "reason": "ControlNet docs vs a community example implementation; not the same concrete bug or change." + }, + { + "left": "pull_request:2808", + "right": "pull_request:3842", + "accept": false, + "reason": "Both relax tests, but for different models and different failures; not mergeable as one fix." + }, + { + "left": "pull_request:2812", + "right": "pull_request:2823", + "accept": false, + "reason": "Same stable-unCLIP docs file, but one adds background/paper/ckpt info while the other rewrites the image-variation example; different sections and purposes." + }, + { + "left": "pull_request:2813", + "right": "pull_request:2823", + "accept": false, + "reason": "Same as above: different stable-unCLIP doc sections, not the same underlying change." + }, + { + "left": "pull_request:286", + "right": "pull_request:309", + "accept": false, + "reason": "Both are README/docs link tweaks, but they fix different links in different places and are not the same change." + }, + { + "left": "pull_request:2873", + "right": "pull_request:4058", + "accept": false, + "reason": "One corrects contributor-guide wording; the other changes SDXL docs image layout. Different docs and different edits." + }, + { + "left": "pull_request:2915", + "right": "pull_request:3614", + "accept": false, + "reason": "Both touch stable-diffusion tests, but one loosens a textual inversion tolerance and the other changes a deterministic attn test setup; distinct bugs." + }, + { + "left": "pull_request:2958", + "right": "pull_request:3226", + "accept": false, + "reason": "ControlNet inpaint code/docs vs ControlNet v1.1 docs; unrelated changes." + }, + { + "left": "pull_request:2958", + "right": "pull_request:3458", + "accept": false, + "reason": "The inpaint lora-scale PR adds API support, while #2958 adds the inpaint pipeline itself; not the same fix." + }, + { + "left": "pull_request:2958", + "right": "pull_request:3460", + "accept": false, + "reason": "These are different inpainting-related changes: pipeline addition vs a later API/support patch. Not the same underlying bug." + }, + { + "left": "pull_request:3068", + "right": "pull_request:3166", + "accept": false, + "reason": "Both concern textual inversion docs/examples, but one is documentation expansion and the other changes the loader example token argument; different fixes." + }, + { + "left": "pull_request:3068", + "right": "pull_request:3554", + "accept": false, + "reason": "Textual inversion docs vs a broad Moreh training integration PR; not the same change." + }, + { + "left": "pull_request:3068", + "right": "pull_request:3555", + "accept": false, + "reason": "Same broad area as #3554, but still a different integration PR rather than the same textual inversion docs fix." + }, + { + "left": "pull_request:3132", + "right": "pull_request:4822", + "accept": false, + "reason": "One fixes a type annotation typo in a pipeline constructor; the other fixes a LoRA warning string. Different code paths and issues." + }, + { + "left": "pull_request:3241", + "right": "pull_request:5292", + "accept": false, + "reason": "Both adjust tests, but for different pipelines and performance/determinism reasons; not the same concrete bug." + }, + { + "left": "pull_request:3331", + "right": "pull_request:4203", + "accept": false, + "reason": "Both are docs typo fixes, but one is inpainting docs and the other is DreamBooth README wording; unrelated." + }, + { + "left": "pull_request:3397", + "right": "pull_request:3806", + "accept": false, + "reason": "Both are training-example docs, but one adds a local directory note for LoRA while the other explains loading from checkpoints for text-to-image training; distinct docs changes." + }, + { + "left": "pull_request:3446", + "right": "pull_request:3641", + "accept": false, + "reason": "Warning-message formatting vs initial-latents inpainting behavior; different code paths and no shared bug." + }, + { + "left": "pull_request:353", + "right": "pull_request:407", + "accept": false, + "reason": "A UNet block typo fix and an attention-slicing API typo fix are unrelated." + }, + { + "left": "pull_request:353", + "right": "pull_request:798", + "accept": false, + "reason": "Different files and issues: UNet block error message vs UNet2D docstring formatting." + }, + { + "left": "pull_request:353", + "right": "pull_request:972", + "accept": false, + "reason": "UNet block typo vs README parameter typo; same generic typo theme only." + }, + { + "left": "pull_request:353", + "right": "pull_request:975", + "accept": false, + "reason": "Different typo fixes in different docs/code locations; not the same underlying change." + }, + { + "left": "pull_request:3614", + "right": "pull_request:3850", + "accept": false, + "reason": "Both are test fixes, but one targets stable diffusion attention determinism and the other audio diffusion mel-resolution setup; unrelated." + }, + { + "left": "pull_request:3641", + "right": "pull_request:4862", + "accept": false, + "reason": "Both touch inpainting-related code, but one fixes initial latents in inpaint pipelines and the other fixes MultiControlNetModel image sizing in SDXL img2img; different bugs." + }, + { + "left": "pull_request:37", + "right": "pull_request:55", + "accept": false, + "reason": "Both are old model refactors, but they merge different architectures (attention vs resnet) and different codebases." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 42, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3744", + "pull_request:3745", + "pull_request:382", + "pull_request:3821", + "pull_request:3839", + "pull_request:386", + "pull_request:3912", + "pull_request:407", + "pull_request:408", + "pull_request:42", + "pull_request:421", + "pull_request:4320", + "pull_request:4330", + "pull_request:4352", + "pull_request:4380", + "pull_request:4396", + "pull_request:4570", + "pull_request:46", + "pull_request:4666", + "pull_request:4749", + "pull_request:4822", + "pull_request:4833", + "pull_request:4855", + "pull_request:4887", + "pull_request:4934", + "pull_request:5", + "pull_request:5066", + "pull_request:5267", + "pull_request:5286", + "pull_request:5454", + "pull_request:55", + "pull_request:5506", + "pull_request:5640", + "pull_request:568", + "pull_request:5682", + "pull_request:59", + "pull_request:5943", + "pull_request:6", + "pull_request:62", + "pull_request:6438", + "pull_request:6592", + "pull_request:6669", + "pull_request:710", + "pull_request:798", + "pull_request:971", + "pull_request:972", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:3744|pull_request:3745", + "pull_request:382|pull_request:386", + "pull_request:3821|pull_request:5286", + "pull_request:3839|pull_request:4822", + "pull_request:3912|pull_request:4396", + "pull_request:407|pull_request:798", + "pull_request:407|pull_request:972", + "pull_request:407|pull_request:975", + "pull_request:408|pull_request:568", + "pull_request:42|pull_request:6", + "pull_request:42|pull_request:62", + "pull_request:421|pull_request:710", + "pull_request:4320|pull_request:5066", + "pull_request:4330|pull_request:5454", + "pull_request:4330|pull_request:6438", + "pull_request:4352|pull_request:4887", + "pull_request:4380|pull_request:5640", + "pull_request:4570|pull_request:5682", + "pull_request:46|pull_request:55", + "pull_request:46|pull_request:59", + "pull_request:4666|pull_request:5943", + "pull_request:4749|pull_request:4833", + "pull_request:4855|pull_request:5506", + "pull_request:4934|pull_request:6592", + "pull_request:5|pull_request:6", + "pull_request:5|pull_request:971", + "pull_request:5267|pull_request:6669" + ], + "prepared_review_unit_hash": "700b0005683bfe735b661979c7254301ecd1a558aa89991dcb64d8a0b5819ac1", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 96258, + "estimated_input_tokens": 24065, + "estimated_eval_tokens": 48386 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated docs typos, workflow token changes, scheduler/model fixes, and larger feature PRs. None of the soft-edge pairs look like the same underlying bug or change, so there is no meaningful canonical duplicate to select.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:3744", + "right": "pull_request:3745", + "accept": false, + "reason": "Both change documentation-upload tokens, but for different workflows/files and different secrets; separate fixes." + }, + { + "left": "pull_request:382", + "right": "pull_request:386", + "accept": false, + "reason": "Both are pipeline docs PRs, but they target different pipelines and add different content." + }, + { + "left": "pull_request:3821", + "right": "pull_request:5286", + "accept": false, + "reason": "Both add translated docs, but for different languages and different pages; not the same change." + }, + { + "left": "pull_request:3839", + "right": "pull_request:4822", + "accept": false, + "reason": "One fixes SDE-VE noise handling in a scheduler, the other fixes a LoRA warning string; unrelated." + }, + { + "left": "pull_request:3912", + "right": "pull_request:4396", + "accept": false, + "reason": "One adds LoRA text-encoder finetuning, the other broadens prompt encoding to work without a text encoder instance; different code paths." + }, + { + "left": "pull_request:407", + "right": "pull_request:798", + "accept": false, + "reason": "Both are tiny typo fixes in Stable Diffusion-related code, but they touch different methods and bugs." + }, + { + "left": "pull_request:407", + "right": "pull_request:972", + "accept": false, + "reason": "A code typo fix versus a README API-parameter typo; unrelated artifacts." + }, + { + "left": "pull_request:407", + "right": "pull_request:975", + "accept": false, + "reason": "Different typo fixes in different files and subsystems; no shared bug." + }, + { + "left": "pull_request:408", + "right": "pull_request:568", + "accept": false, + "reason": "Both are docs typo cleanup, but one is documentation prose and the other spans code/config typo fixes; not one change." + }, + { + "left": "pull_request:42", + "right": "pull_request:6", + "accept": false, + "reason": "Both are early modeling changes, but one adds a test and the other registers UNetLDM in init; different purposes." + }, + { + "left": "pull_request:42", + "right": "pull_request:62", + "accept": false, + "reason": "Both touch test_modeling_utils, but one adds LDM test coverage while the other changes LDM uncond pipeline behavior." + }, + { + "left": "pull_request:421", + "right": "pull_request:710", + "accept": false, + "reason": "Both fix docstring/typo issues, but in different files and different classes." + }, + { + "left": "pull_request:4320", + "right": "pull_request:5066", + "accept": false, + "reason": "One fixes ONNX docs text; the other fixes a LoRA/attention_processor typo in code and related examples." + }, + { + "left": "pull_request:4330", + "right": "pull_request:5454", + "accept": false, + "reason": "Both are documentation edits, but for different pages and different wording issues." + }, + { + "left": "pull_request:4330", + "right": "pull_request:6438", + "accept": false, + "reason": "Both touch docs, but one corrects a README_sdxl heading and the other adds installation guidance; unrelated." + }, + { + "left": "pull_request:4352", + "right": "pull_request:4887", + "accept": false, + "reason": "Both are Kandinsky-related, but one adds combined-pipeline methods/tests while the other removes a latent input from a prior pipeline; different concrete changes." + }, + { + "left": "pull_request:4380", + "right": "pull_request:5640", + "accept": false, + "reason": "Both are docs updates, but they cover different training topics and different text." + }, + { + "left": "pull_request:4570", + "right": "pull_request:5682", + "accept": false, + "reason": "The second is a larger AnimateDiff implementation/checks PR; the first is an earlier draft pipeline skeleton. Not the same mergeable change." + }, + { + "left": "pull_request:46", + "right": "pull_request:55", + "accept": false, + "reason": "Both refactor ResNet-related code, but they merge different source families and include different structural changes." + }, + { + "left": "pull_request:46", + "right": "pull_request:59", + "accept": false, + "reason": "Shared ResNet file, but one merges LDM into ResNet and the other finalizes a different 2D ResNet variant; not the same bug/change." + }, + { + "left": "pull_request:4666", + "right": "pull_request:5943", + "accept": false, + "reason": "One adds LoRA support for SDXL ControlNet; the other adds gradient checkpointing to Transformer2DModel. Different functionality." + }, + { + "left": "pull_request:4749", + "right": "pull_request:4833", + "accept": false, + "reason": "One fixes SDXL DreamBooth LoRA checkpoint loading, the other fixes LoRA unfuse/save behavior; related domain but different concrete problems." + }, + { + "left": "pull_request:4855", + "right": "pull_request:5506", + "accept": false, + "reason": "Both involve offloading/PEFT-related behavior, but one is accelerate hook handling during side loading and the other is xformers/LoRA scaling in attention." + }, + { + "left": "pull_request:4934", + "right": "pull_request:6592", + "accept": false, + "reason": "Both are training-example PRs, but one adds T2I-Adapter training docs/scripts and the other adds an inpainting SDXL script; different features." + }, + { + "left": "pull_request:5", + "right": "pull_request:6", + "accept": false, + "reason": "Linked by the same UNetLDM lineage, but one adds the model file and the other only exports it in package init; not duplicate changes." + }, + { + "left": "pull_request:5", + "right": "pull_request:971", + "accept": false, + "reason": "A core model implementation PR versus a community example README/code PR; unrelated." + }, + { + "left": "pull_request:5267", + "right": "pull_request:6669", + "accept": false, + "reason": "One updates contributor README code fences, the other updates repo statistics in the main README; different documentation fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 43, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5499", + "pull_request:55", + "pull_request:5506", + "pull_request:555", + "pull_request:5698", + "pull_request:5707", + "pull_request:5727", + "pull_request:5745", + "pull_request:5818", + "pull_request:5823", + "pull_request:5855", + "pull_request:59", + "pull_request:5983", + "pull_request:6002", + "pull_request:6198", + "pull_request:6314", + "pull_request:6433", + "pull_request:6462", + "pull_request:6475", + "pull_request:6669", + "pull_request:6837", + "pull_request:6998", + "pull_request:7015", + "pull_request:7181", + "pull_request:7319", + "pull_request:7324", + "pull_request:7325", + "pull_request:74", + "pull_request:7425", + "pull_request:7472", + "pull_request:7709", + "pull_request:7797", + "pull_request:7867", + "pull_request:798", + "pull_request:824", + "pull_request:8403", + "pull_request:8426", + "pull_request:8470", + "pull_request:8600", + "pull_request:8851", + "pull_request:8932", + "pull_request:9", + "pull_request:9024", + "pull_request:9066", + "pull_request:9107", + "pull_request:9266", + "pull_request:972", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:5499|pull_request:6998", + "pull_request:55|pull_request:59", + "pull_request:5506|pull_request:5823", + "pull_request:555|pull_request:824", + "pull_request:5698|pull_request:6198", + "pull_request:5707|pull_request:5727", + "pull_request:5745|pull_request:6475", + "pull_request:5745|pull_request:7015", + "pull_request:5818|pull_request:7324", + "pull_request:5855|pull_request:6669", + "pull_request:5983|pull_request:6002", + "pull_request:6314|pull_request:7472", + "pull_request:6433|pull_request:8426", + "pull_request:6462|pull_request:7472", + "pull_request:6837|pull_request:8851", + "pull_request:7181|pull_request:7325", + "pull_request:7319|pull_request:9066", + "pull_request:74|pull_request:9", + "pull_request:7425|pull_request:8851", + "pull_request:7709|pull_request:8403", + "pull_request:7797|pull_request:8932", + "pull_request:7867|pull_request:9024", + "pull_request:798|pull_request:972", + "pull_request:798|pull_request:975", + "pull_request:8470|pull_request:9266", + "pull_request:8600|pull_request:9107" + ], + "prepared_review_unit_hash": "39755e78e47f6f890733196c0174fb9b8ea7b5354753c962832c7f613aa4a27d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 97307, + "estimated_input_tokens": 24327, + "estimated_eval_tokens": 48910 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous set of unrelated PRs spanning docs fixes, CI tweaks, model changes, and feature work. The soft pairs are mostly superficial title/file similarity rather than the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:5499", + "right": "pull_request:6998", + "accept": false, + "reason": "Both are docs link/callout edits, but they fix different pages and different link formatting issues." + }, + { + "left": "pull_request:55", + "right": "pull_request:59", + "accept": false, + "reason": "Both touch resnet code, but one merges glide resnet into the general resnet while the other finalizes a different resnet variant; not the same change." + }, + { + "left": "pull_request:5506", + "right": "pull_request:5823", + "accept": false, + "reason": "Both involve LoRA/PEFT attention code, but one fixes xformers scale handling and the other adds adapter_names to fuse_lora; different code paths." + }, + { + "left": "pull_request:555", + "right": "pull_request:824", + "accept": false, + "reason": "Both are Flax-related model utilities, but one adds Flax VAE support and the other adds a Flax test and utility cleanup; unrelated changes." + }, + { + "left": "pull_request:5698", + "right": "pull_request:6198", + "accept": false, + "reason": "Both mention LCM, but one fixes img2img behavior in the pipeline and the other adds benchmark reporting; different functionality." + }, + { + "left": "pull_request:5707", + "right": "pull_request:5727", + "accept": false, + "reason": "One is a fix for LCM loading across pipelines; the other adds LCM training scripts and README. Not the same bug/change." + }, + { + "left": "pull_request:5745", + "right": "pull_request:6475", + "accept": false, + "reason": "Both are ConsistencyDecoder/SVD docs-only tweaks, but they address different documentation details." + }, + { + "left": "pull_request:5745", + "right": "pull_request:7015", + "accept": false, + "reason": "A doc type/example fix versus a single-file loading doc/example fix; different artifacts and problems." + }, + { + "left": "pull_request:5818", + "right": "pull_request:7324", + "accept": false, + "reason": "Both are CI/test workflow related, but one updates PR test matrix setup while the other updates Stable Cascade tests; not the same issue." + }, + { + "left": "pull_request:5855", + "right": "pull_request:6669", + "accept": false, + "reason": "Both are README/doc errata, but they change different example/community docs and different mistakes." + }, + { + "left": "pull_request:5983", + "right": "pull_request:6002", + "accept": false, + "reason": "Both are SVD docs/pipeline fixes, but one edits documentation while the other changes guidance-scale logic in the pipeline." + }, + { + "left": "pull_request:6314", + "right": "pull_request:7472", + "accept": false, + "reason": "Both are typo/doc fixes, but they target different docs and unrelated wording/link issues." + }, + { + "left": "pull_request:6433", + "right": "pull_request:8426", + "accept": false, + "reason": "One adds a community pipeline entry; the other changes workflow dependency installation. No shared underlying bug." + }, + { + "left": "pull_request:6462", + "right": "pull_request:7472", + "accept": false, + "reason": "Different docs updates in different files; one fixes a paper link, the other a broken examples link." + }, + { + "left": "pull_request:6837", + "right": "pull_request:8851", + "accept": false, + "reason": "Both are docs additions, but one is an i2vgen XL docs dependency/import tweak and the other adds AuraFlow docs and code." + }, + { + "left": "pull_request:7181", + "right": "pull_request:7325", + "accept": false, + "reason": "Both are typo/doc cleanups, but they touch many different files and unrelated text fixes." + }, + { + "left": "pull_request:7319", + "right": "pull_request:9066", + "accept": false, + "reason": "Both add scheduler/model docs, but one introduces EDM scheduler pages and the other adds Stable Cascade UNet docs." + }, + { + "left": "pull_request:74", + "right": "pull_request:9", + "accept": false, + "reason": "Both concern Grad-TTS/resnet-era model work, but one adds a mid-block to Grad-TTS while the other introduces the Grad-TTS model/pipeline itself." + }, + { + "left": "pull_request:7425", + "right": "pull_request:8851", + "accept": false, + "reason": "Both are docs-related, but one fixes LoRA tutorial text while the other adds AuraFlow docs and implementation." + }, + { + "left": "pull_request:7709", + "right": "pull_request:8403", + "accept": false, + "reason": "Both touch CI runner/workflow infrastructure, but one adds tailscale to push tests and the other updates the tailscale action version in SSH runner." + }, + { + "left": "pull_request:7797", + "right": "pull_request:8932", + "accept": false, + "reason": "Both reduce test coverage/runtime by adjusting tiny test models, but they are for different test suites and different model classes." + }, + { + "left": "pull_request:7867", + "right": "pull_request:9024", + "accept": false, + "reason": "Both modify workflow cache settings, but they target different CI changes and 9024 spans many workflows plus dependency updates." + }, + { + "left": "pull_request:798", + "right": "pull_request:972", + "accept": false, + "reason": "Both are small docstring/readme fixes, but they address different typos in different files." + }, + { + "left": "pull_request:798", + "right": "pull_request:975", + "accept": false, + "reason": "Different documentation typo fixes in different files; no shared underlying change." + }, + { + "left": "pull_request:8470", + "right": "pull_request:9266", + "accept": false, + "reason": "One rewrites a research project README; the other adds an entry to community projects docs. Same broad area, but not the same change." + }, + { + "left": "pull_request:8600", + "right": "pull_request:9107", + "accept": false, + "reason": "Both are SD3 DreamBooth-related, but one fixes text-encoder cleanup in two scripts while the other fixes a loading hook state-dict filter; different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 44, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10205", + "pull_request:10810", + "pull_request:11067", + "pull_request:11894", + "pull_request:11994", + "pull_request:12086", + "pull_request:12445", + "pull_request:12530", + "pull_request:12782", + "pull_request:13039", + "pull_request:13164", + "pull_request:13323", + "pull_request:1355", + "pull_request:1564", + "pull_request:1582", + "pull_request:1583", + "pull_request:1966", + "pull_request:2339", + "pull_request:2464", + "pull_request:2470", + "pull_request:3037", + "pull_request:5312", + "pull_request:5759", + "pull_request:6217", + "pull_request:6646", + "pull_request:6898", + "pull_request:6908", + "pull_request:6999", + "pull_request:7632", + "pull_request:8627", + "pull_request:8660", + "pull_request:8684", + "pull_request:8797", + "pull_request:8883", + "pull_request:9021", + "pull_request:9024", + "pull_request:9048", + "pull_request:9060", + "pull_request:9175", + "pull_request:9257", + "pull_request:9400", + "pull_request:9506", + "pull_request:9507", + "pull_request:9696", + "pull_request:9705", + "pull_request:980", + "pull_request:981" + ], + "soft_pairs": [ + "pull_request:8627|pull_request:9400", + "pull_request:8660|pull_request:9060", + "pull_request:8684|pull_request:9506", + "pull_request:8797|pull_request:9696", + "pull_request:8883|pull_request:9024", + "pull_request:9021|pull_request:9257", + "pull_request:9048|pull_request:9705", + "pull_request:980|pull_request:981", + "pull_request:1355|pull_request:1966", + "pull_request:12086|pull_request:12445", + "pull_request:5312|pull_request:5759", + "pull_request:6898|pull_request:6908", + "pull_request:9175|pull_request:9507", + "pull_request:11994|pull_request:12782", + "pull_request:12782|pull_request:13323", + "pull_request:13039|pull_request:13323", + "pull_request:1564|pull_request:2339", + "pull_request:2470|pull_request:3037", + "pull_request:6217|pull_request:7632", + "pull_request:6646|pull_request:6999", + "pull_request:10205|pull_request:10810", + "pull_request:11067|pull_request:11894", + "pull_request:12530|pull_request:13164", + "pull_request:1582|pull_request:2464", + "pull_request:1583|pull_request:2464" + ], + "prepared_review_unit_hash": "ee041b9da1d5558ad73beb3db1491f4f59605061c466dd33deb8bd7613651f45", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 98371, + "estimated_input_tokens": 24593, + "estimated_eval_tokens": 49442 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a false positive mix: the candidate PRs span unrelated docs, CI, tests, and model/pipeline fixes. None of the soft edges look like the same underlying bug or change, so all should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a valid canonical representative here: the set is heterogeneous and the pairwise candidates do not converge on one underlying fix or change.", + "best_issue_reason": null, + "best_pr_reason": "No best PR stands out for duplicate triage; the items are not duplicates of one another, so selecting a canonical PR would be arbitrary.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8627", + "right": "pull_request:9400", + "accept": false, + "reason": "Different scopes: SD3 ControlNet fast-test/docs work vs single-file pipeline device-agnostic test cleanup." + }, + { + "left": "pull_request:8660", + "right": "pull_request:9060", + "accept": false, + "reason": "Unrelated changes: SD3 ControlNet input validation vs removing a debug print in Flux transformer code." + }, + { + "left": "pull_request:8684", + "right": "pull_request:9506", + "accept": false, + "reason": "Both touch docs, but they fix different documentation issues in different pipelines and examples." + }, + { + "left": "pull_request:8797", + "right": "pull_request:9696", + "accept": false, + "reason": "One fixes sharded checkpoint tests; the other updates a PR-testing workflow. Different problems entirely." + }, + { + "left": "pull_request:8883", + "right": "pull_request:9024", + "accept": false, + "reason": "Both are CI-related, but one changes nightly test execution and the other adjusts cache groups and dependency pins." + }, + { + "left": "pull_request:9021", + "right": "pull_request:9257", + "accept": false, + "reason": "Different code paths: SD3 ControlNet transformer loading vs Dreambooth Flux prompt/tokenization logic." + }, + { + "left": "pull_request:9048", + "right": "pull_request:9705", + "accept": false, + "reason": "These are separate Flux README/doc tweaks, not the same concrete bug or change." + }, + { + "left": "pull_request:980", + "right": "pull_request:981", + "accept": false, + "reason": "Same area, but one is an FP16 dtype fix and the other is a naming/API doc change for audio length." + }, + { + "left": "pull_request:1355", + "right": "pull_request:1966", + "accept": false, + "reason": "Both add Flax pipelines, but img2img and inpaint are distinct feature implementations." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12445", + "accept": false, + "reason": "Related Flux modular code, but one adds I2I support and the other realigns the modular structure with Qwen." + }, + { + "left": "pull_request:5312", + "right": "pull_request:5759", + "accept": false, + "reason": "One is targeted typehint/docs cleanup in models; the other is a broad docs/type-hints sweep across many modules." + }, + { + "left": "pull_request:6898", + "right": "pull_request:6908", + "accept": false, + "reason": "Both standardize model cards, but one is a broad multi-example refactor and the other is a specific ControlNet SDXL follow-up." + }, + { + "left": "pull_request:9175", + "right": "pull_request:9507", + "accept": false, + "reason": "One adds Flux ControlNet support; the other fixes a control-mode batching bug in existing Flux ControlNet code." + }, + { + "left": "pull_request:11994", + "right": "pull_request:12782", + "accept": false, + "reason": "Different failures: huggingface-hub test breakage vs HunyuanVideo offloading test adjustments." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13323", + "accept": false, + "reason": "Different test fixes in different components; no shared underlying bug." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13323", + "accept": false, + "reason": "WAN CPU upcasting fix vs device/dtype test cleanup are unrelated." + }, + { + "left": "pull_request:1564", + "right": "pull_request:2339", + "accept": false, + "reason": "Default accelerate config docs vs Dreambooth checkpointing docs/tests; not the same change." + }, + { + "left": "pull_request:2470", + "right": "pull_request:3037", + "accept": false, + "reason": "Both are MPS test-related, but they address different model/pipeline failures." + }, + { + "left": "pull_request:6217", + "right": "pull_request:7632", + "accept": false, + "reason": "Stable Video Diffusion torch-compile fix vs Dreambooth VAE-loading issue are different bugs." + }, + { + "left": "pull_request:6646", + "right": "pull_request:6999", + "accept": false, + "reason": "Both concern SVD, but one changes resolution handling and the other fixes tensor-image input behavior." + }, + { + "left": "pull_request:10205", + "right": "pull_request:10810", + "accept": false, + "reason": "Different LoRA-related work in different examples and converter code paths." + }, + { + "left": "pull_request:11067", + "right": "pull_request:11894", + "accept": false, + "reason": "Both pin dependencies for CI, but they pin different packages for different reasons." + }, + { + "left": "pull_request:12530", + "right": "pull_request:13164", + "accept": false, + "reason": "One xfails a flaky test; the other strips test deps from CI workflows. Different maintenance actions." + }, + { + "left": "pull_request:1582", + "right": "pull_request:2464", + "accept": false, + "reason": "RePaint scheduler bug fix vs removal of xformers handling in text-to-image LoRA training are unrelated." + }, + { + "left": "pull_request:1583", + "right": "pull_request:2464", + "accept": false, + "reason": "Legacy inpaint masking fix vs text-to-image LoRA training cleanup are not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 45, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11823", + "pull_request:11843", + "pull_request:12315", + "pull_request:12454", + "pull_request:1788", + "pull_request:2009", + "pull_request:2345", + "pull_request:2398", + "pull_request:2956", + "pull_request:3127", + "pull_request:3160", + "pull_request:3214", + "pull_request:3446", + "pull_request:4078", + "pull_request:4242", + "pull_request:4272", + "pull_request:4373", + "pull_request:4395", + "pull_request:4666", + "pull_request:4700", + "pull_request:4722", + "pull_request:4749", + "pull_request:4787", + "pull_request:4862", + "pull_request:5090", + "pull_request:5157", + "pull_request:5164", + "pull_request:5220", + "pull_request:5608", + "pull_request:5823", + "pull_request:6437", + "pull_request:6448", + "pull_request:6759", + "pull_request:6787", + "pull_request:7050", + "pull_request:7103", + "pull_request:7627", + "pull_request:7738", + "pull_request:8422", + "pull_request:8572", + "pull_request:8618", + "pull_request:8702", + "pull_request:8798", + "pull_request:8812", + "pull_request:8901", + "pull_request:8992", + "pull_request:9021", + "pull_request:9597" + ], + "soft_pairs": [ + "pull_request:4272|pull_request:4373", + "pull_request:6437|pull_request:7738", + "pull_request:7103|pull_request:9021", + "pull_request:8992|pull_request:9597", + "pull_request:8812|pull_request:8901", + "pull_request:11843|pull_request:12315", + "pull_request:1788|pull_request:3214", + "pull_request:2345|pull_request:3446", + "pull_request:4666|pull_request:5164", + "pull_request:4666|pull_request:5220", + "pull_request:4787|pull_request:6448", + "pull_request:4862|pull_request:5157", + "pull_request:6787|pull_request:7627", + "pull_request:8422|pull_request:8798", + "pull_request:5608|pull_request:5823", + "pull_request:1788|pull_request:3127", + "pull_request:2009|pull_request:2398", + "pull_request:4078|pull_request:4242", + "pull_request:4395|pull_request:4749", + "pull_request:5090|pull_request:6759", + "pull_request:8618|pull_request:8702", + "pull_request:2956|pull_request:3160", + "pull_request:11823|pull_request:12454", + "pull_request:4700|pull_request:4722", + "pull_request:7050|pull_request:8572" + ], + "prepared_review_unit_hash": "f68f7a9642584c58fa08069ec5f835e3a50513bae21c82e53cedde00cfe78620", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 103422, + "estimated_input_tokens": 25856, + "estimated_eval_tokens": 51968 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The candidate links are mostly false positives: they pair unrelated feature PRs, test-only PRs, or docs updates across different subsystems. I would not merge any soft edge as a duplicate.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8812 is the most central, substantial feature PR in the set: it adds the main Kolors implementation, has an explicit issue target, and anchors a later follow-up (#8901).", + "best_issue_reason": null, + "best_pr_reason": "PR #8812 is the best representative artifact if one must be chosen, because it is the broadest core code change rather than a narrow test/docs tweak.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4272", + "right": "pull_request:4373", + "accept": false, + "reason": "Optional import cleanup vs SDXL resize bug fix; different code paths and different problems." + }, + { + "left": "pull_request:6437", + "right": "pull_request:7738", + "accept": false, + "reason": "CI exclusion tweak vs LoRA device-placement bug; not the same underlying issue." + }, + { + "left": "pull_request:7103", + "right": "pull_request:9021", + "accept": false, + "reason": "ControlNet input validation fix vs SD3ControlNet num_layers loading fix; unrelated behavior." + }, + { + "left": "pull_request:8992", + "right": "pull_request:9597", + "accept": false, + "reason": "Different community pipelines and different differential-diffusion implementations; not a duplicate." + }, + { + "left": "pull_request:8812", + "right": "pull_request:8901", + "accept": false, + "reason": "Main Kolors integration vs adding IP Adapter support for Kolors; related ecosystem, but separate features." + }, + { + "left": "pull_request:11843", + "right": "pull_request:12315", + "accept": false, + "reason": "Wan VACE exclude_modules test coverage vs scheduler simplification in LoRA tests; unrelated." + }, + { + "left": "pull_request:1788", + "right": "pull_request:3214", + "accept": false, + "reason": "DDIM prediction_type bug vs attention reshape mismatch bug; different components and fixes." + }, + { + "left": "pull_request:2345", + "right": "pull_request:3446", + "accept": false, + "reason": "xFormers warning text vs pipeline loading warning wording; not the same bug or change." + }, + { + "left": "pull_request:4666", + "right": "pull_request:5164", + "accept": false, + "reason": "SDXL ControlNet LoRA support vs FreeU mechanism; separate features in different areas." + }, + { + "left": "pull_request:4666", + "right": "pull_request:5220", + "accept": false, + "reason": "ControlNet LoRA support vs moving tests to nightly; unrelated." + }, + { + "left": "pull_request:4787", + "right": "pull_request:6448", + "accept": false, + "reason": "LoRA checkpoint filtering logic vs LoRA test assertion updates; one is code behavior, the other is tests." + }, + { + "left": "pull_request:4862", + "right": "pull_request:5157", + "accept": false, + "reason": "MultiControlNet SDXL img2img height/width bug vs LoRA conversion mapping fix; different issues." + }, + { + "left": "pull_request:6787", + "right": "pull_request:7627", + "accept": false, + "reason": "Marigold image normalization bug vs Playground img2img latent stats handling; separate pipelines and bugs." + }, + { + "left": "pull_request:8422", + "right": "pull_request:8798", + "accept": false, + "reason": "A broad subfolder-loading test suite vs a specific sharded-checkpoint loading fix; not the same concrete change." + }, + { + "left": "pull_request:5608", + "right": "pull_request:5823", + "accept": false, + "reason": "Empty network_alpha dict handling vs adapter_names support in fuse_lora; different LoRA APIs." + }, + { + "left": "pull_request:1788", + "right": "pull_request:3127", + "accept": false, + "reason": "Scheduler step bug vs safety checker tensor output handling; unrelated code paths." + }, + { + "left": "pull_request:2009", + "right": "pull_request:2398", + "accept": false, + "reason": "Textual inversion loading feature vs Pix2Pix Zero docs update; not duplicates." + }, + { + "left": "pull_request:4078", + "right": "pull_request:4242", + "accept": false, + "reason": "Require accelerate for from_single_file vs allowing VAE loading; different loader behavior changes." + }, + { + "left": "pull_request:4395", + "right": "pull_request:4749", + "accept": false, + "reason": "DreamBooth class prompt preprocessing bug vs SDXL LoRA checkpointing fix; separate issues." + }, + { + "left": "pull_request:5090", + "right": "pull_request:6759", + "accept": false, + "reason": "T2I-Adapter weighting bug fix vs InstantID prompt image repetition fix; different pipelines and failures." + }, + { + "left": "pull_request:8618", + "right": "pull_request:8702", + "accept": false, + "reason": "SD3 tiny autoencoder shift_factor change vs SD3 conversion script; one is model API, the other is tooling." + }, + { + "left": "pull_request:2956", + "right": "pull_request:3160", + "accept": false, + "reason": "Karras sigmas added to different schedulers; same theme, but not the same fix or code path." + }, + { + "left": "pull_request:11823", + "right": "pull_request:12454", + "accept": false, + "reason": "Flux Kontext LoRA support vs modular i2i/t2i support for Kontext; related domain, but distinct implementations." + }, + { + "left": "pull_request:4700", + "right": "pull_request:4722", + "accept": false, + "reason": "Shap-E docs guide vs DiffEdit docs guide; different documentation topics." + }, + { + "left": "pull_request:7050", + "right": "pull_request:8572", + "accept": false, + "reason": "Text-to-image typo cleanup vs general contributing/docs errata; not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 46, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10105", + "pull_request:10117", + "pull_request:10205", + "pull_request:10318", + "pull_request:10494", + "pull_request:10497", + "pull_request:10589", + "pull_request:10717", + "pull_request:11243", + "pull_request:11353", + "pull_request:1147", + "pull_request:11997", + "pull_request:12541", + "pull_request:13209", + "pull_request:1527", + "pull_request:1945", + "pull_request:2009", + "pull_request:2062", + "pull_request:2172", + "pull_request:2386", + "pull_request:2588", + "pull_request:2595", + "pull_request:2892", + "pull_request:3001", + "pull_request:3160", + "pull_request:326", + "pull_request:3299", + "pull_request:343", + "pull_request:3673", + "pull_request:3847", + "pull_request:3850", + "pull_request:408", + "pull_request:4352", + "pull_request:4396", + "pull_request:4761", + "pull_request:4855", + "pull_request:5672", + "pull_request:5695", + "pull_request:5943", + "pull_request:6691", + "pull_request:6795", + "pull_request:7118", + "pull_request:7632", + "pull_request:8564", + "pull_request:8572", + "pull_request:8691", + "pull_request:8755", + "pull_request:9426" + ], + "soft_pairs": [ + "pull_request:7118|pull_request:8572", + "pull_request:3001|pull_request:3160", + "pull_request:10497|pull_request:9426", + "pull_request:5672|pull_request:5695", + "pull_request:11353|pull_request:11997", + "pull_request:8564|pull_request:8755", + "pull_request:10105|pull_request:10117", + "pull_request:10205|pull_request:11243", + "pull_request:12541|pull_request:13209", + "pull_request:1945|pull_request:3847", + "pull_request:2062|pull_request:3299", + "pull_request:2172|pull_request:3673", + "pull_request:326|pull_request:343", + "pull_request:3850|pull_request:4761", + "pull_request:4855|pull_request:5943", + "pull_request:6691|pull_request:6795", + "pull_request:7632|pull_request:8691", + "pull_request:10318|pull_request:10494", + "pull_request:10589|pull_request:10717", + "pull_request:1147|pull_request:2588", + "pull_request:1527|pull_request:2009", + "pull_request:2386|pull_request:408", + "pull_request:2595|pull_request:4396", + "pull_request:2892|pull_request:4352" + ], + "prepared_review_unit_hash": "0b049e26065ce45b0ea552b98f1ddc686d47356515ca385bc80f71128c7de95e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 103495, + "estimated_input_tokens": 25874, + "estimated_eval_tokens": 52004 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of superficially similar PRs (docs typos, CI tweaks, LoRA/pipeline example changes, scheduler refactors). None of the soft pairs are convincing duplicates of the same concrete change, so I would not merge any of them together.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits this cluster; the items span unrelated docs, tests, CI, LoRA conversion, and pipeline feature work rather than one shared bug/change.", + "best_issue_reason": null, + "best_pr_reason": "No strong global best PR: the cluster is heterogeneous, so any one PR would misrepresent the rest rather than serve as a true canonical example.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7118", + "right": "pull_request:8572", + "accept": false, + "reason": "Both touch docs, but one is IP-Adapter typo cleanup and the other is broad errata/contributing edits; not the same underlying change." + }, + { + "left": "pull_request:3001", + "right": "pull_request:3160", + "accept": false, + "reason": "Both relate to schedulers/Karras sigmas, but they modify different schedulers and different behavior; not one mergeable fix." + }, + { + "left": "pull_request:10497", + "right": "pull_request:9426", + "accept": false, + "reason": "Both edit CI workflow install steps, but one changes tokenizers install source and the other updates artifact uploader versions; unrelated fixes." + }, + { + "left": "pull_request:5672", + "right": "pull_request:5695", + "accept": false, + "reason": "Both are PixArt-related, but one fixes non-square image support and the other fixes masking logic in prompt embedding; different bugs." + }, + { + "left": "pull_request:11353", + "right": "pull_request:11997", + "accept": false, + "reason": "Both target Wan/FLF2V support, but one adds user-facing docs and conversion/pipeline support while the other rewires modular pipeline blocks; too broad and distinct." + }, + { + "left": "pull_request:8564", + "right": "pull_request:8755", + "accept": false, + "reason": "Both are SD3 training changes, but one adds T5 token-limit support and the other fixes text-encoder training logic; separate concerns." + }, + { + "left": "pull_request:10105", + "right": "pull_request:10117", + "accept": false, + "reason": "Both fix README links, but they are separate doc cleanup PRs with different link targets and scope; not the same concrete bug." + }, + { + "left": "pull_request:10205", + "right": "pull_request:11243", + "accept": false, + "reason": "These are different LoRA/example changes: PEFT refactor for instructpix2pix vs musubi Wan LoRA support; no shared code-path fix." + }, + { + "left": "pull_request:12541", + "right": "pull_request:13209", + "accept": false, + "reason": "Both touch lora_conversion_utils, but one fixes latest few-step Wan LoRA head mapping and the other normalizes ZImage LoRA key names; distinct converters." + }, + { + "left": "pull_request:1945", + "right": "pull_request:3847", + "accept": false, + "reason": "One adds UNet block tests, the other adds dependency-soft-check tests; same broad area of tests but not the same issue." + }, + { + "left": "pull_request:2062", + "right": "pull_request:3299", + "accept": false, + "reason": "Different fixes in different components: paint-by-example CPU offload vs LoRA/text-encoder docs and training example sync." + }, + { + "left": "pull_request:2172", + "right": "pull_request:3673", + "accept": false, + "reason": "Both mention multi-GPU DreamBooth, but they patch different parts of the training scripts and fix different bugs." + }, + { + "left": "pull_request:326", + "right": "pull_request:343", + "accept": false, + "reason": "Both are type-hint/scheduler-era changes, but one is a large PNDM-related pipeline/type-hint PR and the other is a narrow DDIM signature/type-hint update." + }, + { + "left": "pull_request:3850", + "right": "pull_request:4761", + "accept": false, + "reason": "Both are test/image-encoder related, but one fixes audio diffusion test setup and the other changes Paint-by-Example encoder initialization." + }, + { + "left": "pull_request:4855", + "right": "pull_request:5943", + "accept": false, + "reason": "Both mention offload/checkpointing-ish core changes, but they address different mechanisms and components; not the same fix." + }, + { + "left": "pull_request:6691", + "right": "pull_request:6795", + "accept": false, + "reason": "Both are advanced DreamBooth LoRA work, but one is a broad feature/bug-fix bundle and the other is specifically SDXL micro-conditioning." + }, + { + "left": "pull_request:7632", + "right": "pull_request:8691", + "accept": false, + "reason": "Both are DreamBooth/LoRA training fixes, but one addresses VAE loading and the other fixes vanilla LoRA loading behavior; separate bugs." + }, + { + "left": "pull_request:10318", + "right": "pull_request:10494", + "accept": false, + "reason": "One adds Flux LoRA integration tests; the other changes tokenizers installation in CI. Shared test file context is not enough to make them the same change." + }, + { + "left": "pull_request:10589", + "right": "pull_request:10717", + "accept": false, + "reason": "Both are IP-Adapter-related, but they target different pipelines (SD3 img2img vs Flux img2img) and distinct implementation paths." + }, + { + "left": "pull_request:1147", + "right": "pull_request:2588", + "accept": false, + "reason": "One loads the text encoder from a subfolder in Flax examples; the other adds LoRA support for the text encoder. Different features." + }, + { + "left": "pull_request:1527", + "right": "pull_request:2009", + "accept": false, + "reason": "The shared inpaint file is incidental: one allows tensor image/mask inputs, the other adds textual inversion loading/offload handling across pipelines." + }, + { + "left": "pull_request:2386", + "right": "pull_request:408", + "accept": false, + "reason": "Both are docs-only, but they are unrelated typo/documentation cleanup PRs with no common bug or change." + }, + { + "left": "pull_request:2595", + "right": "pull_request:4396", + "accept": false, + "reason": "One fixes scheduler config switching between DEIS/DPM/UniPC; the other makes encode_prompt work without a text encoder instance variable. Different code paths." + }, + { + "left": "pull_request:2892", + "right": "pull_request:4352", + "accept": false, + "reason": "One fixes upscaler input validation to allow embeds; the other adds Kandinsky combined pipeline utilities. No meaningful overlap in the underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 47, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11786", + "pull_request:11994", + "pull_request:12236", + "pull_request:12490", + "pull_request:12530", + "pull_request:12549", + "pull_request:13323", + "pull_request:2398", + "pull_request:2928", + "pull_request:2994", + "pull_request:3127", + "pull_request:3214", + "pull_request:3397", + "pull_request:4078", + "pull_request:4188", + "pull_request:4267", + "pull_request:4607", + "pull_request:4622", + "pull_request:4749", + "pull_request:4947", + "pull_request:5003", + "pull_request:5035", + "pull_request:5090", + "pull_request:5164", + "pull_request:5186", + "pull_request:5367", + "pull_request:5407", + "pull_request:5697", + "pull_request:5741", + "pull_request:6291", + "pull_request:6330", + "pull_request:6564", + "pull_request:6578", + "pull_request:6645", + "pull_request:6787", + "pull_request:6875", + "pull_request:6898", + "pull_request:6910", + "pull_request:7062", + "pull_request:7068", + "pull_request:7166", + "pull_request:7634", + "pull_request:8452", + "pull_request:8467", + "pull_request:8691", + "pull_request:8798", + "pull_request:9008" + ], + "soft_pairs": [ + "pull_request:2928|pull_request:4188", + "pull_request:3127|pull_request:3214", + "pull_request:4267|pull_request:5367", + "pull_request:4607|pull_request:4622", + "pull_request:5090|pull_request:6564", + "pull_request:5697|pull_request:5741", + "pull_request:6291|pull_request:6875", + "pull_request:7068|pull_request:9008", + "pull_request:12236|pull_request:12549", + "pull_request:11786|pull_request:12530", + "pull_request:11994|pull_request:12490", + "pull_request:12490|pull_request:13323", + "pull_request:2398|pull_request:3397", + "pull_request:4078|pull_request:5035", + "pull_request:4749|pull_request:5003", + "pull_request:4947|pull_request:5090", + "pull_request:5035|pull_request:5186", + "pull_request:5164|pull_request:5186", + "pull_request:5407|pull_request:7634", + "pull_request:6330|pull_request:6578", + "pull_request:6645|pull_request:7062", + "pull_request:6787|pull_request:8467", + "pull_request:6898|pull_request:6910", + "pull_request:7166|pull_request:8798", + "pull_request:8452|pull_request:8691", + "pull_request:2928|pull_request:2994" + ], + "prepared_review_unit_hash": "b83a5f30fa8dd06ec703e868f21cb69a84dab74c1e73d23acb5e207c1d15552f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 95601, + "estimated_input_tokens": 23901, + "estimated_eval_tokens": 48058 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is noisy: it contains many unrelated docs, CI, dependency, and feature PRs that only share broad subsystems. None of the soft-edge pairs look like the same underlying bug or change, so all should be rejected.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:2928", + "right": "pull_request:4188", + "accept": false, + "reason": "Both touch ControlNet training, but one changes local dataset loading and the other fixes SDXL embedding fingerprinting; different problems and code paths." + }, + { + "left": "pull_request:3127", + "right": "pull_request:3214", + "accept": false, + "reason": "Unrelated fixes: img2img safety-checker tensor handling vs an attention reshape shape mismatch." + }, + { + "left": "pull_request:4267", + "right": "pull_request:5367", + "accept": false, + "reason": "Both are docs edits, but one fixes an SDXL image reference and the other corrects tutorial pipeline instantiation; not the same change." + }, + { + "left": "pull_request:4607", + "right": "pull_request:4622", + "accept": false, + "reason": "Related to PushToHub docs, but one updates overview autodo c entries while the other adds a full guide; different scope and content." + }, + { + "left": "pull_request:5090", + "right": "pull_request:6564", + "accept": false, + "reason": "Different fixes in different adapter systems: T2I adapter community pipeline vs IP Adapter path construction." + }, + { + "left": "pull_request:5697", + "right": "pull_request:5741", + "accept": false, + "reason": "Both involve PEFT/LoRA, but one fixes xformers scaling behavior and the other speeds up unloading; not the same bug." + }, + { + "left": "pull_request:6291", + "right": "pull_request:6875", + "accept": false, + "reason": "Test-only changes for different components: LoRA unfusion assertion vs IP Adapter output expectations." + }, + { + "left": "pull_request:7068", + "right": "pull_request:9008", + "accept": false, + "reason": "Both are typo/errata cleanup, but they edit different files and unrelated text; not a duplicate change." + }, + { + "left": "pull_request:12236", + "right": "pull_request:12549", + "accept": false, + "reason": "Both add attention backends, but they implement different backends with different integrations and hardware targets." + }, + { + "left": "pull_request:11786", + "right": "pull_request:12530", + "accept": false, + "reason": "Separate CI test maintenance: big-accelerator marker cleanup vs xfail for one failing test." + }, + { + "left": "pull_request:11994", + "right": "pull_request:12490", + "accept": false, + "reason": "Docs/auth test-failure fix vs an Audioldm2 ClapConfig test update; unrelated changes." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13323", + "accept": false, + "reason": "Different test fixes in different areas: Audioldm2 config setup vs generic device/to-dtype test handling." + }, + { + "left": "pull_request:2398", + "right": "pull_request:3397", + "accept": false, + "reason": "Both are docs, but one adds Pix2Pix Zero content and the other updates LoRA local-path guidance." + }, + { + "left": "pull_request:4078", + "right": "pull_request:5035", + "accept": false, + "reason": "Both touch imports/dependencies, but one requires accelerate for single-file loading and the other stops forcing transformers; opposite and unrelated concerns." + }, + { + "left": "pull_request:4749", + "right": "pull_request:5003", + "accept": false, + "reason": "SDXL DreamBooth LoRA checkpoint loading vs LoRA fuse/unfuse mechanics; different code paths and fixes." + }, + { + "left": "pull_request:4947", + "right": "pull_request:5090", + "accept": false, + "reason": "T2I adapter training tests vs a T2I adapter runtime fix; one is test coverage, the other is a bug fix." + }, + { + "left": "pull_request:5035", + "right": "pull_request:5186", + "accept": false, + "reason": "Optional-import cleanup vs a draft FreeU proof-of-concept; unrelated changes." + }, + { + "left": "pull_request:5164", + "right": "pull_request:5186", + "accept": false, + "reason": "Same feature area, but 5186 is a draft POC while 5164 is the merged FreeU implementation; not a mergeable duplicate change." + }, + { + "left": "pull_request:5407", + "right": "pull_request:7634", + "accept": false, + "reason": "Different example/model changes: mixed-precision training fix vs a high-resolution generation feature." + }, + { + "left": "pull_request:6330", + "right": "pull_request:6578", + "accept": false, + "reason": "Docs-only fixes in different tutorials: SVD rendering/content vs ControlNet heading formatting." + }, + { + "left": "pull_request:6645", + "right": "pull_request:7062", + "accept": false, + "reason": "Unrelated documentation cleanups in different files and topics." + }, + { + "left": "pull_request:6787", + "right": "pull_request:8467", + "accept": false, + "reason": "Different runtime fixes in unrelated pipelines: Marigold input normalization vs SEGA device placement." + }, + { + "left": "pull_request:6898", + "right": "pull_request:6910", + "accept": false, + "reason": "Both standardize model cards, but one is a broad multi-example refactor and the other is a ControlNet-specific follow-up; not the same underlying change." + }, + { + "left": "pull_request:7166", + "right": "pull_request:8798", + "accept": false, + "reason": "Different loading bugs: SDXL LoRA DreamBooth checkpoint restoration vs sharded checkpoint subfolder resolution." + }, + { + "left": "pull_request:8452", + "right": "pull_request:8691", + "accept": false, + "reason": "Single-file loading error fix vs LoRA load selection logic; distinct issues." + }, + { + "left": "pull_request:2928", + "right": "pull_request:2994", + "accept": false, + "reason": "Both are ControlNet training scripts, but one changes dataset loading and the other fixes residual type casting; different bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 48, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10222", + "pull_request:10402", + "pull_request:10404", + "pull_request:10825", + "pull_request:10858", + "pull_request:11248", + "pull_request:11303", + "pull_request:11505", + "pull_request:11835", + "pull_request:12018", + "pull_request:12231", + "pull_request:12527", + "pull_request:12552", + "pull_request:12775", + "pull_request:12782", + "pull_request:12883", + "pull_request:13382", + "pull_request:1513", + "pull_request:16", + "pull_request:1977", + "pull_request:2172", + "pull_request:2396", + "pull_request:2415", + "pull_request:2490", + "pull_request:260", + "pull_request:2642", + "pull_request:2647", + "pull_request:2684", + "pull_request:2685", + "pull_request:2954", + "pull_request:3128", + "pull_request:327", + "pull_request:3471", + "pull_request:3499", + "pull_request:3546", + "pull_request:3568", + "pull_request:3577", + "pull_request:360", + "pull_request:3716", + "pull_request:3825", + "pull_request:3921", + "pull_request:3942", + "pull_request:4612", + "pull_request:5499", + "pull_request:5979", + "pull_request:9286", + "pull_request:9760" + ], + "soft_pairs": [ + "pull_request:3499|pull_request:3546", + "pull_request:3921|pull_request:5979", + "pull_request:10222|pull_request:9286", + "pull_request:10402|pull_request:10858", + "pull_request:10404|pull_request:11248", + "pull_request:10404|pull_request:11303", + "pull_request:10825|pull_request:9760", + "pull_request:11505|pull_request:12018", + "pull_request:11835|pull_request:12782", + "pull_request:12231|pull_request:12883", + "pull_request:12527|pull_request:12552", + "pull_request:12775|pull_request:13382", + "pull_request:1513|pull_request:3471", + "pull_request:16|pull_request:260", + "pull_request:1977|pull_request:2415", + "pull_request:2172|pull_request:3128", + "pull_request:2396|pull_request:3942", + "pull_request:2490|pull_request:4612", + "pull_request:2642|pull_request:3825", + "pull_request:2647|pull_request:3825", + "pull_request:2684|pull_request:3825", + "pull_request:2685|pull_request:3825", + "pull_request:2954|pull_request:3716", + "pull_request:327|pull_request:360", + "pull_request:3568|pull_request:4612", + "pull_request:3577|pull_request:5499" + ], + "prepared_review_unit_hash": "cafd536d04bad468c33ff9770c2b0c96665e69e7a42345f7ab5e9d76bbd58016", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 94252, + "estimated_input_tokens": 23563, + "estimated_eval_tokens": 47382 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The pairs are mostly superficial matches: same filenames, same docs-heavy maintenance patterns, or same broad subsystem, but they do not describe the same concrete bug/change. I found no convincing duplicate PR pair in this cluster.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:3499", + "right": "pull_request:3546", + "accept": false, + "reason": "Same panorama file, but 3499 fixes a dpmsolver-specific scheduler bug while 3546 broadens scheduler support; different underlying changes." + }, + { + "left": "pull_request:3921", + "right": "pull_request:5979", + "accept": false, + "reason": "Different features entirely: Blended Latent Diffusion community pipeline vs DDGAN pipeline/scheduler." + }, + { + "left": "pull_request:10222", + "right": "pull_request:9286", + "accept": false, + "reason": "ControlNet PyTorch XLA support vs exporting FluxMultiControlNetModel in __init__.py; unrelated changes." + }, + { + "left": "pull_request:10402", + "right": "pull_request:10858", + "accept": false, + "reason": "Both are docs PRs, but one fixes hunyuan broken links and the other updates the overview table entry for I2VGen-XL; not the same issue." + }, + { + "left": "pull_request:10404", + "right": "pull_request:11248", + "accept": false, + "reason": "Both are typo/docs cleanup, but they touch different pages and different content; not the same underlying fix." + }, + { + "left": "pull_request:10404", + "right": "pull_request:11303", + "accept": false, + "reason": "Different docs files and different typos; no shared concrete bug." + }, + { + "left": "pull_request:10825", + "right": "pull_request:9760", + "accept": false, + "reason": "One adds checkpoint inference guidance to training docs; the other edits README headings/wording. Not the same change." + }, + { + "left": "pull_request:11505", + "right": "pull_request:12018", + "accept": false, + "reason": "Both fix docs wording, but in different documents and on different topics; too broad to be duplicates." + }, + { + "left": "pull_request:11835", + "right": "pull_request:12782", + "accept": false, + "reason": "Different test suites for different models and failure modes; not the same bug." + }, + { + "left": "pull_request:12231", + "right": "pull_request:12883", + "accept": false, + "reason": "Both are typo fixes, but across unrelated example/docs files and different text." + }, + { + "left": "pull_request:12527", + "right": "pull_request:12552", + "accept": false, + "reason": "Code-path fix in Kandinsky5 CFG handling vs docs typo cleanup for Kandinsky5; not mergeable as one fix." + }, + { + "left": "pull_request:12775", + "right": "pull_request:13382", + "accept": false, + "reason": "Both add docs content, but one adds Z-Image pages and the other adds modular auto-docstring documentation; unrelated additions." + }, + { + "left": "pull_request:1513", + "right": "pull_request:3471", + "accept": false, + "reason": "Both touch dreambooth README, but one adds accelerate notebook setup instructions and the other adds a torch.compile note; different edits." + }, + { + "left": "pull_request:16", + "right": "pull_request:260", + "accept": false, + "reason": "General README update vs a small typo fix in examples/inference/readme.md; not the same underlying change." + }, + { + "left": "pull_request:1977", + "right": "pull_request:2415", + "accept": false, + "reason": "Completely different code areas: training/scheduler/model changes vs pipeline_utils import/version handling." + }, + { + "left": "pull_request:2172", + "right": "pull_request:3128", + "accept": false, + "reason": "LoRA validation multi-GPU fix vs img2img timestep selection bug; unrelated." + }, + { + "left": "pull_request:2396", + "right": "pull_request:3942", + "accept": false, + "reason": "Different test failures in different contexts; one is scheduler warning cleanup, the other np/xformers test handling." + }, + { + "left": "pull_request:2490", + "right": "pull_request:4612", + "accept": false, + "reason": "MultiDiffusion docs addition vs controlling_generation link fixes; same docs area but distinct content." + }, + { + "left": "pull_request:2642", + "right": "pull_request:3825", + "accept": false, + "reason": "Separate broken-link docs fixes in different pages; not the same specific link set." + }, + { + "left": "pull_request:2647", + "right": "pull_request:3825", + "accept": false, + "reason": "Scheduler docs wording fix vs control_brightness link correction; unrelated docs changes." + }, + { + "left": "pull_request:2684", + "right": "pull_request:3825", + "accept": false, + "reason": "Custom pipeline overview link fix vs control_brightness link fix; different files and targets." + }, + { + "left": "pull_request:2685", + "right": "pull_request:3825", + "accept": false, + "reason": "KerasCV docs punctuation/newline cleanup vs control_brightness link fix; unrelated." + }, + { + "left": "pull_request:2954", + "right": "pull_request:3716", + "accept": false, + "reason": "Both are pipeline additions, but TextToVideoZero and ParaDiGMS are different features with different codepaths." + }, + { + "left": "pull_request:327", + "right": "pull_request:360", + "accept": false, + "reason": "Both are type-hint cleanups, but for different schedulers/pipeline code and not the same concrete bug." + }, + { + "left": "pull_request:3568", + "right": "pull_request:4612", + "accept": false, + "reason": "Both fix broken links, but in different docs pages and different link targets." + }, + { + "left": "pull_request:3577", + "right": "pull_request:5499", + "accept": false, + "reason": "README link rewrite vs large docs link fix across many pages; similar theme but not the same PR change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 49, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11469", + "pull_request:1229", + "pull_request:12418", + "pull_request:13352", + "pull_request:13536", + "pull_request:1508", + "pull_request:2892", + "pull_request:3334", + "pull_request:382", + "pull_request:388", + "pull_request:3923", + "pull_request:3987", + "pull_request:4188", + "pull_request:4242", + "pull_request:4352", + "pull_request:4373", + "pull_request:4574", + "pull_request:4707", + "pull_request:4749", + "pull_request:4833", + "pull_request:4859", + "pull_request:5003", + "pull_request:5015", + "pull_request:5051", + "pull_request:5129", + "pull_request:5157", + "pull_request:5202", + "pull_request:5236", + "pull_request:5298", + "pull_request:5420", + "pull_request:5470", + "pull_request:5833", + "pull_request:5834", + "pull_request:6215", + "pull_request:6342", + "pull_request:7050", + "pull_request:7068", + "pull_request:7118", + "pull_request:7632", + "pull_request:8490", + "pull_request:8642", + "pull_request:8860", + "pull_request:9070", + "pull_request:9083", + "pull_request:9086", + "pull_request:9100", + "pull_request:9754", + "pull_request:9760" + ], + "soft_pairs": [ + "pull_request:382|pull_request:388", + "pull_request:3923|pull_request:5236", + "pull_request:3987|pull_request:4707", + "pull_request:4574|pull_request:6342", + "pull_request:5003|pull_request:5298", + "pull_request:5051|pull_request:7068", + "pull_request:5157|pull_request:5833", + "pull_request:7050|pull_request:7118", + "pull_request:8490|pull_request:9760", + "pull_request:8642|pull_request:9100", + "pull_request:8860|pull_request:9760", + "pull_request:9070|pull_request:9083", + "pull_request:9086|pull_request:9754", + "pull_request:4352|pull_request:5470", + "pull_request:5129|pull_request:5202", + "pull_request:1229|pull_request:1508", + "pull_request:13352|pull_request:13536", + "pull_request:2892|pull_request:4373", + "pull_request:3334|pull_request:5015", + "pull_request:4188|pull_request:5420", + "pull_request:4242|pull_request:5833", + "pull_request:4242|pull_request:5834", + "pull_request:4749|pull_request:4859", + "pull_request:4833|pull_request:5157", + "pull_request:6215|pull_request:7632", + "pull_request:11469|pull_request:12418" + ], + "prepared_review_unit_hash": "8947f5d71275297cbc8705bd4ca9b26ae1fc5959644c8f57f0364748918efc09", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 91378, + "estimated_input_tokens": 22845, + "estimated_eval_tokens": 45946 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of unrelated docs, test, CI, and a few LoRA/Flux/controlnet fixes. The soft pairs are mostly title/file-similarity false positives rather than the same underlying bug or change, so I would not merge any of them as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:382", + "right": "pull_request:388", + "accept": false, + "reason": "Both are docs-only, but they update different pipeline pages for different models; no shared bug or change." + }, + { + "left": "pull_request:3923", + "right": "pull_request:5236", + "accept": false, + "reason": "Both are documentation fixes, but for unrelated pages/languages and different content issues." + }, + { + "left": "pull_request:3987", + "right": "pull_request:4707", + "accept": false, + "reason": "Audio Diffusion snippet fix and Custom Diffusion README rewrite are unrelated examples/docs changes." + }, + { + "left": "pull_request:4574", + "right": "pull_request:6342", + "accept": false, + "reason": "One updates InstructPix2Pix README examples; the other adds an installation note. Different issues." + }, + { + "left": "pull_request:5003", + "right": "pull_request:5298", + "accept": false, + "reason": "Both touch LoRA code, but one fixes fuse/unfuse logic while the other fixes torch.compile conv behavior; different code-path bugs." + }, + { + "left": "pull_request:5051", + "right": "pull_request:7068", + "accept": false, + "reason": "Broken docs link vs typo/docstring cleanup; same broad docs area, but not the same change." + }, + { + "left": "pull_request:5157", + "right": "pull_request:5833", + "accept": false, + "reason": "ControlNet LoRA name conversion fix and single-file loading import fix are different loader paths and bugs." + }, + { + "left": "pull_request:7050", + "right": "pull_request:7118", + "accept": false, + "reason": "Both are typo/doc tweaks, but in different examples/docs sections with no shared underlying defect." + }, + { + "left": "pull_request:8490", + "right": "pull_request:9760", + "accept": false, + "reason": "DreamBooth SD3 README script-name correction and top-level README wording cleanup are unrelated docs edits." + }, + { + "left": "pull_request:8642", + "right": "pull_request:9100", + "accept": false, + "reason": "Stable Diffusion 3 return-type typo and text-to-image help-text typo are separate docs fixes." + }, + { + "left": "pull_request:8860", + "right": "pull_request:9760", + "accept": false, + "reason": "One is a README_sd3 update, the other is a broad README wording tweak; not the same change." + }, + { + "left": "pull_request:9070", + "right": "pull_request:9083", + "accept": false, + "reason": "Adding a Flux img2img pipeline and adding single-file support for FluxTransformer2DModel are distinct features." + }, + { + "left": "pull_request:9086", + "right": "pull_request:9754", + "accept": false, + "reason": "Both are Flux-related, but one adds DreamBooth LoRA training scripts while the other adds a quantized research training setup; different concrete work." + }, + { + "left": "pull_request:4352", + "right": "pull_request:5470", + "accept": false, + "reason": "Kandinsky CPU offload/xformers support and ControlNet inpaint pipeline/docs fixes are unrelated." + }, + { + "left": "pull_request:5129", + "right": "pull_request:5202", + "accept": false, + "reason": "One adds multi-LoRA fuse/unfuse support; the other adds Kohya checkpoint mapping entries. Different functionality." + }, + { + "left": "pull_request:1229", + "right": "pull_request:1508", + "accept": false, + "reason": "Both edit DreamBooth README text, but they address different guidance topics and not the same issue." + }, + { + "left": "pull_request:13352", + "right": "pull_request:13536", + "accept": false, + "reason": "One adds a checkout step; the other changes workflow permissions/instructions. Same file, different workflow behavior." + }, + { + "left": "pull_request:2892", + "right": "pull_request:4373", + "accept": false, + "reason": "Upcaler input validation for embeds and SDXL resize/default-size handling are separate pipeline bugs." + }, + { + "left": "pull_request:3334", + "right": "pull_request:5015", + "accept": false, + "reason": "Docstring formatting cleanup and broken-link fix are unrelated documentation fixes." + }, + { + "left": "pull_request:4188", + "right": "pull_request:5420", + "accept": false, + "reason": "ControlNet SDXL dataset fingerprinting and basic training doc example correction are different problems." + }, + { + "left": "pull_request:4242", + "right": "pull_request:5833", + "accept": false, + "reason": "Allowing VAE in from_single_file and fixing ControlNet import paths are different loader issues." + }, + { + "left": "pull_request:4242", + "right": "pull_request:5834", + "accept": false, + "reason": "VAE loading support versus single-file ControlNet import correction are different code paths." + }, + { + "left": "pull_request:4749", + "right": "pull_request:4859", + "accept": false, + "reason": "SDXL LoRA checkpointing and SDXL inpaint fast-test setup are unrelated changes." + }, + { + "left": "pull_request:4833", + "right": "pull_request:5157", + "accept": false, + "reason": "LoRA fuse/unfuse internals and ControlNet LoRA conversion mappings are different bugs in different areas." + }, + { + "left": "pull_request:6215", + "right": "pull_request:7632", + "accept": false, + "reason": "PixArt test initialization and DreamBooth VAE path handling are unrelated changes." + }, + { + "left": "pull_request:11469", + "right": "pull_request:12418", + "accept": false, + "reason": "Both are test xfails, but for different pipelines and different failure modes; not the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 50, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10132", + "pull_request:10359", + "pull_request:10497", + "pull_request:10717", + "pull_request:10750", + "pull_request:10867", + "pull_request:11025", + "pull_request:11353", + "pull_request:11716", + "pull_request:12055", + "pull_request:12164", + "pull_request:1229", + "pull_request:12526", + "pull_request:12551", + "pull_request:12647", + "pull_request:12714", + "pull_request:13058", + "pull_request:1513", + "pull_request:1920", + "pull_request:2398", + "pull_request:2588", + "pull_request:2595", + "pull_request:3128", + "pull_request:4272", + "pull_request:4276", + "pull_request:4396", + "pull_request:4621", + "pull_request:4847", + "pull_request:4862", + "pull_request:4886", + "pull_request:5470", + "pull_request:6419", + "pull_request:6448", + "pull_request:6564", + "pull_request:6898", + "pull_request:6907", + "pull_request:6939", + "pull_request:7118", + "pull_request:7131", + "pull_request:7638", + "pull_request:7717", + "pull_request:7738", + "pull_request:7932", + "pull_request:8825", + "pull_request:9207", + "pull_request:9638", + "pull_request:9696" + ], + "soft_pairs": [ + "pull_request:11716|pull_request:12714", + "pull_request:12551|pull_request:12647", + "pull_request:12526|pull_request:13058", + "pull_request:10359|pull_request:9696", + "pull_request:4862|pull_request:5470", + "pull_request:6898|pull_request:6939", + "pull_request:10750|pull_request:9638", + "pull_request:7638|pull_request:7717", + "pull_request:11025|pull_request:11353", + "pull_request:12055|pull_request:12164", + "pull_request:1920|pull_request:2398", + "pull_request:2588|pull_request:4396", + "pull_request:4272|pull_request:4276", + "pull_request:4621|pull_request:6564", + "pull_request:6564|pull_request:8825", + "pull_request:7118|pull_request:9207", + "pull_request:7131|pull_request:9207", + "pull_request:7932|pull_request:9696", + "pull_request:6898|pull_request:6907", + "pull_request:1229|pull_request:1513", + "pull_request:2595|pull_request:3128", + "pull_request:4862|pull_request:4886", + "pull_request:6448|pull_request:7738", + "pull_request:10717|pull_request:10867", + "pull_request:10132|pull_request:10497", + "pull_request:4847|pull_request:6419" + ], + "prepared_review_unit_hash": "c7bbbead87453643b2bd78a71e0cf20bc49f1e17d73fe6208718395c340ab203", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 103774, + "estimated_input_tokens": 25944, + "estimated_eval_tokens": 52144 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most of the cluster is unrelated feature/docs/test work. The only grounded duplicate-like pair is the tokenizers CI workflow follow-up; the Sana layerwise-casting skip pair is not the same fix and should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10132 is the base tokenizers workflow fix; PR 10497 is a narrow follow-up correcting that same install step.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong global representative for this heterogeneous cluster; 10132 is only the best local anchor because it owns the only grounded duplicate pair.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11716", + "right": "pull_request:12714", + "accept": false, + "reason": "Both are LoRA test/code changes, but one fixes Flux metadata handling and the other fixes a Flux2 NaN test; different bugs/code paths." + }, + { + "left": "pull_request:12551", + "right": "pull_request:12647", + "accept": false, + "reason": "Sana test skip vs autoencoder_dc layerwise-casting skip in a different test file; not the same bug or mergeable fix." + }, + { + "left": "pull_request:12526", + "right": "pull_request:13058", + "accept": false, + "reason": "Base Qwen-Image support vs Qwen Image Animate pipeline; separate feature additions." + }, + { + "left": "pull_request:10359", + "right": "pull_request:9696", + "accept": false, + "reason": "A Sana test cleanup and a PR workflow change; unrelated." + }, + { + "left": "pull_request:4862", + "right": "pull_request:5470", + "accept": false, + "reason": "SDXL MultiControlNet height/width handling vs inpaint pipeline docs/fixes; different issues." + }, + { + "left": "pull_request:6898", + "right": "pull_request:6939", + "accept": false, + "reason": "Both standardize model cards, but they target different example scripts and aren\u2019t one concrete change." + }, + { + "left": "pull_request:10750", + "right": "pull_request:9638", + "accept": false, + "reason": "Flux ControlNet img2img support vs Xlabs ControlNet support; separate implementations." + }, + { + "left": "pull_request:7638", + "right": "pull_request:7717", + "accept": false, + "reason": "Wrong-link fix vs docstring style cleanup; unrelated." + }, + { + "left": "pull_request:11025", + "right": "pull_request:11353", + "accept": false, + "reason": "Wan LoRA conversion vs FLF2V pipeline support; separate features." + }, + { + "left": "pull_request:12055", + "right": "pull_request:12164", + "accept": false, + "reason": "Qwen-Image base package support vs Qwen Image Edit support; different feature sets." + }, + { + "left": "pull_request:1920", + "right": "pull_request:2398", + "accept": false, + "reason": "Different docs edits in unrelated areas." + }, + { + "left": "pull_request:2588", + "right": "pull_request:4396", + "accept": false, + "reason": "Text-encoder LoRA support vs encode_prompt fallback behavior; different problems." + }, + { + "left": "pull_request:4272", + "right": "pull_request:4276", + "accept": false, + "reason": "Missing optional imports vs import test; related topic, not the same change." + }, + { + "left": "pull_request:4621", + "right": "pull_request:6564", + "accept": false, + "reason": "Multi-adapter validation vs IP-Adapter path handling; different subsystems." + }, + { + "left": "pull_request:6564", + "right": "pull_request:8825", + "accept": false, + "reason": "IP-Adapter path fix vs SDXL K-Diffusion ControlNet/IP-Adapter support; not the same bug." + }, + { + "left": "pull_request:7118", + "right": "pull_request:9207", + "accept": false, + "reason": "IP-Adapter docs typos vs CPU-offloading docs; unrelated." + }, + { + "left": "pull_request:7131", + "right": "pull_request:9207", + "accept": false, + "reason": "Separate documentation edits in different areas." + }, + { + "left": "pull_request:7932", + "right": "pull_request:9696", + "accept": false, + "reason": "LoRA test expectation tweaks vs PR workflow update; unrelated." + }, + { + "left": "pull_request:6898", + "right": "pull_request:6907", + "accept": false, + "reason": "Different model-card standardizations for different example families; too broad for one duplicate." + }, + { + "left": "pull_request:1229", + "right": "pull_request:1513", + "accept": false, + "reason": "Dreambooth README setup changes for different instructions." + }, + { + "left": "pull_request:2595", + "right": "pull_request:3128", + "accept": false, + "reason": "Scheduler config switching vs img2img timestep selection; separate bugs." + }, + { + "left": "pull_request:4862", + "right": "pull_request:4886", + "accept": false, + "reason": "MultiControlNet shape handling vs upscale safety-checker input processing; different bugs." + }, + { + "left": "pull_request:6448", + "right": "pull_request:7738", + "accept": false, + "reason": "Two LoRA test/code adjustments, but they fix different assertions and paths." + }, + { + "left": "pull_request:10717", + "right": "pull_request:10867", + "accept": false, + "reason": "Flux img2img IP-Adapter support vs multi-IP-Adapter support; separate changes." + }, + { + "left": "pull_request:10132", + "right": "pull_request:10497", + "accept": true, + "reason": "Same tokenizers install line in the same workflow; 10497 is a direct correction to 10132's install target." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Reject: one accepted soft edge is not grounded. PR 12551 is a Sana test skip, while PR 12647 is an autoencoder_dc layerwise-casting skip in a different test file; they do not look like a single mergeable bugfix. The summary also overstates that this is the same Sana CI flake family. The tokenizers workflow follow-up (10132/10497) is grounded, but the cluster-level claim that only two soft edges are real duplicates is too strong given the unsupported second acceptance." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 51, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10115", + "pull_request:10192", + "pull_request:1020", + "pull_request:10222", + "pull_request:10230", + "pull_request:10232", + "pull_request:10253", + "pull_request:10259", + "pull_request:10278", + "pull_request:10318", + "pull_request:10340", + "pull_request:10358", + "pull_request:10376", + "pull_request:10394", + "pull_request:1041", + "pull_request:10431", + "pull_request:10434", + "pull_request:10498", + "pull_request:10531", + "pull_request:10548", + "pull_request:1077", + "pull_request:10802", + "pull_request:10819", + "pull_request:10888", + "pull_request:11204", + "pull_request:11319", + "pull_request:11546", + "pull_request:11623", + "pull_request:11634", + "pull_request:11770", + "pull_request:11789", + "pull_request:11847", + "pull_request:11851", + "pull_request:1194", + "pull_request:12983", + "pull_request:13121", + "pull_request:1412", + "pull_request:2788", + "pull_request:531", + "pull_request:5979", + "pull_request:6303", + "pull_request:8481", + "pull_request:8600", + "pull_request:9116", + "pull_request:9196", + "pull_request:9734", + "pull_request:9860" + ], + "soft_pairs": [ + "pull_request:12983|pull_request:13121", + "pull_request:5979|pull_request:6303", + "pull_request:10192|pull_request:10259", + "pull_request:10340|pull_request:10431", + "pull_request:10115|pull_request:8481", + "pull_request:1020|pull_request:1194", + "pull_request:1020|pull_request:2788", + "pull_request:10222|pull_request:8600", + "pull_request:10230|pull_request:10318", + "pull_request:10232|pull_request:9734", + "pull_request:10253|pull_request:9196", + "pull_request:10278|pull_request:11634", + "pull_request:10358|pull_request:11319", + "pull_request:10376|pull_request:11546", + "pull_request:10394|pull_request:9116", + "pull_request:1041|pull_request:531", + "pull_request:10434|pull_request:11847", + "pull_request:10498|pull_request:11204", + "pull_request:10531|pull_request:11851", + "pull_request:10548|pull_request:11789", + "pull_request:1077|pull_request:1412", + "pull_request:10802|pull_request:11623", + "pull_request:10819|pull_request:11770", + "pull_request:10819|pull_request:11851", + "pull_request:10888|pull_request:9860" + ], + "prepared_review_unit_hash": "6ba40b6176c33a9d2dd02455129087c8caf5bc1cfb69a8a0f8a404a5dd3eefed", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 95854, + "estimated_input_tokens": 23964, + "estimated_eval_tokens": 48184 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly unrelated PRs that only look similar by title or broad subsystem. I found no true duplicate pairs; the soft edges are all false positives across docs, typos, feature work, and separate bug fixes.", + "confidence": 0.24, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10498 is the strongest representative PR here: it has a clear concrete runtime change, touches multiple code paths in a single feature area (PyTorch/XLA support), and is linked to an explicit issue target.", + "best_issue_reason": null, + "best_pr_reason": "PR 10498 is the best overall PR candidate because it is substantive, self-contained, and clearly scoped to one underlying change, unlike the many small docs/typo PRs in the set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12983", + "right": "pull_request:13121", + "accept": false, + "reason": "Different LTX-2 changes: one adds single-file support, the other fixes batching/CFG inference. Same area, not the same bug or change." + }, + { + "left": "pull_request:5979", + "right": "pull_request:6303", + "accept": false, + "reason": "Completely different work: DDGAN pipeline/scheduler vs ADD training scripts. No shared code-path problem." + }, + { + "left": "pull_request:10192", + "right": "pull_request:10259", + "accept": false, + "reason": "Both Flux-related, but one adds a Control/Inpaint pipeline while the other adds LoRA loading support for Flux Control/Fill. Different features." + }, + { + "left": "pull_request:10340", + "right": "pull_request:10431", + "accept": false, + "reason": "Both Sana fixes, but they address different configuration/positional-embedding issues and different model variants." + }, + { + "left": "pull_request:10115", + "right": "pull_request:8481", + "accept": false, + "reason": "Unrelated bugs in different subsystems: Flux rotary embeddings vs image processor input validation." + }, + { + "left": "pull_request:1020", + "right": "pull_request:1194", + "accept": false, + "reason": "Docs update vs GitHub Actions workflow update; not the same underlying change." + }, + { + "left": "pull_request:1020", + "right": "pull_request:2788", + "accept": false, + "reason": "Both touch docs structure, but one updates prose while the other deduplicates the training section tree. Distinct changes." + }, + { + "left": "pull_request:10222", + "right": "pull_request:8600", + "accept": false, + "reason": "PyTorch/XLA runtime support in ControlNet is unrelated to DreamBooth SD3 training cleanup." + }, + { + "left": "pull_request:10230", + "right": "pull_request:10318", + "accept": false, + "reason": "Docs list expansion vs Flux LoRA integration tests. Related topic, but not the same change or bug." + }, + { + "left": "pull_request:10232", + "right": "pull_request:9734", + "accept": false, + "reason": "Separate Sana documentation correction vs broader SD3 doc/test update." + }, + { + "left": "pull_request:10253", + "right": "pull_request:9196", + "accept": false, + "reason": "Contributor-note docs change vs PixArt attention processor feature; unrelated." + }, + { + "left": "pull_request:10278", + "right": "pull_request:11634", + "accept": false, + "reason": "Same doc file, but one fixes the example and the other adds a link reference. Not the same issue." + }, + { + "left": "pull_request:10358", + "right": "pull_request:11319", + "accept": false, + "reason": "Both add community projects, but they are different project entries with no shared bug/change." + }, + { + "left": "pull_request:10376", + "right": "pull_request:11546", + "accept": false, + "reason": "Both LoRA-related, but one converts original HunyuanVideo LoRAs and the other adds FramePack LoRA loading support. Different code paths." + }, + { + "left": "pull_request:10394", + "right": "pull_request:9116", + "accept": false, + "reason": "Both are link fixes, but they fix different docs and different links." + }, + { + "left": "pull_request:1041", + "right": "pull_request:531", + "accept": false, + "reason": "Different test fixes in different areas and eras; no shared defect." + }, + { + "left": "pull_request:10434", + "right": "pull_request:11847", + "accept": false, + "reason": "Docs link repair across many pages vs a single-file example correction. Not the same underlying change." + }, + { + "left": "pull_request:10498", + "right": "pull_request:11204", + "accept": false, + "reason": "Both involve PyTorch/XLA, but one adds inference/runtime support across pipelines while the other adds an SDXL training example. Different problems." + }, + { + "left": "pull_request:10531", + "right": "pull_request:11851", + "accept": false, + "reason": "Negative-prompt docs for Flux vs a note about `_keep_in_fp32_modules`; unrelated documentation changes." + }, + { + "left": "pull_request:10548", + "right": "pull_request:11789", + "accept": false, + "reason": "HunyuanVideo batch-size bug fix vs Framepack device-test adjustments. Different artifacts and different issues." + }, + { + "left": "pull_request:1077", + "right": "pull_request:1412", + "accept": false, + "reason": "Scheduler test refactor vs SD2 inpainting integration tests. Different changes." + }, + { + "left": "pull_request:10802", + "right": "pull_request:11623", + "accept": false, + "reason": "Both typo fixes, but in different files and with different wording; not a duplicate change." + }, + { + "left": "pull_request:10819", + "right": "pull_request:11770", + "accept": false, + "reason": "Both touch LoRA docs, but one adds missing entries while the other performs broader cleanup and code/doc refactors. Not the same fix." + }, + { + "left": "pull_request:10819", + "right": "pull_request:11851", + "accept": false, + "reason": "LoRA docs additions vs an fp32-module documentation note; unrelated." + }, + { + "left": "pull_request:10888", + "right": "pull_request:9860", + "accept": false, + "reason": "Both concern SD3 ControlNet, but one wires AutoPipeline support while the other updates the training example for SD3.5. Different scopes and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 52, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10938", + "pull_request:11096", + "pull_request:11236", + "pull_request:11237", + "pull_request:11248", + "pull_request:11249", + "pull_request:11251", + "pull_request:11303", + "pull_request:11407", + "pull_request:11416", + "pull_request:11476", + "pull_request:1162", + "pull_request:11746", + "pull_request:11838", + "pull_request:11847", + "pull_request:11857", + "pull_request:11875", + "pull_request:11879", + "pull_request:11958", + "pull_request:11983", + "pull_request:12068", + "pull_request:1218", + "pull_request:12183", + "pull_request:12193", + "pull_request:12247", + "pull_request:12265", + "pull_request:12307", + "pull_request:12337", + "pull_request:12360", + "pull_request:12487", + "pull_request:12560", + "pull_request:1269", + "pull_request:12716", + "pull_request:12731", + "pull_request:12797", + "pull_request:12890", + "pull_request:12902", + "pull_request:12984", + "pull_request:13115", + "pull_request:13135", + "pull_request:13148", + "pull_request:1527", + "pull_request:1555", + "pull_request:2257", + "pull_request:3063", + "pull_request:3266", + "pull_request:9760", + "pull_request:98" + ], + "soft_pairs": [ + "pull_request:10938|pull_request:11983", + "pull_request:10938|pull_request:9760", + "pull_request:11096|pull_request:12193", + "pull_request:11236|pull_request:11237", + "pull_request:11248|pull_request:11983", + "pull_request:11248|pull_request:9760", + "pull_request:11249|pull_request:11251", + "pull_request:11303|pull_request:11983", + "pull_request:11303|pull_request:9760", + "pull_request:11407|pull_request:12265", + "pull_request:11416|pull_request:12265", + "pull_request:11476|pull_request:12265", + "pull_request:1162|pull_request:1555", + "pull_request:1162|pull_request:2257", + "pull_request:1162|pull_request:3063", + "pull_request:1162|pull_request:3266", + "pull_request:11746|pull_request:12560", + "pull_request:11838|pull_request:11879", + "pull_request:11847|pull_request:12068", + "pull_request:11847|pull_request:12487", + "pull_request:11857|pull_request:12337", + "pull_request:11875|pull_request:11958", + "pull_request:11983|pull_request:12890", + "pull_request:1218|pull_request:98", + "pull_request:12183|pull_request:12984", + "pull_request:12247|pull_request:12307", + "pull_request:12247|pull_request:13115", + "pull_request:12360|pull_request:12902", + "pull_request:12360|pull_request:13135", + "pull_request:12487|pull_request:12716", + "pull_request:12560|pull_request:13148", + "pull_request:1269|pull_request:1527", + "pull_request:12731|pull_request:12797" + ], + "prepared_review_unit_hash": "7e036f93460154b7d76c8161824993a95847852faf3bdf7836c6ccee12939674", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 102154, + "estimated_input_tokens": 25539, + "estimated_eval_tokens": 51334 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a false-positive pileup: unrelated docs typos, feature additions, modular-pipeline refactors, and a few genuine code fixes. None of the suggested soft edges look like the same underlying change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10938", + "right": "pull_request:11983", + "accept": false, + "reason": "Both are docs-only, but one updates evaluation guidance and the other fixes installation instructions; unrelated changes." + }, + { + "left": "pull_request:10938", + "right": "pull_request:9760", + "accept": false, + "reason": "Evaluation-doc update vs README heading formatting; same file type, different content and intent." + }, + { + "left": "pull_request:11096", + "right": "pull_request:12193", + "accept": false, + "reason": "Flux TPU README performance wording vs Qwen DreamBooth script name correction; unrelated model/docs edits." + }, + { + "left": "pull_request:11236", + "right": "pull_request:11237", + "accept": false, + "reason": "One renames a Sana pipeline doc heading, the other fixes a dtype-map example; different docs and issues." + }, + { + "left": "pull_request:11248", + "right": "pull_request:11983", + "accept": false, + "reason": "Flux ControlNet README code fix vs installation guide update; no shared bug or code path." + }, + { + "left": "pull_request:11248", + "right": "pull_request:9760", + "accept": false, + "reason": "A README code formatting fix and a general README title tweak are not the same change." + }, + { + "left": "pull_request:11249", + "right": "pull_request:11251", + "accept": false, + "reason": "Both touch ControlNet, but one adds XLab controlnet support in Flux image-to-image while the other adds SD3 multi-controlnet inpainting support; distinct features and code paths." + }, + { + "left": "pull_request:11303", + "right": "pull_request:11983", + "accept": false, + "reason": "Autoencoder docs class-name fix vs installation docs; unrelated doc corrections." + }, + { + "left": "pull_request:11303", + "right": "pull_request:9760", + "accept": false, + "reason": "Autoencoder doc correction vs README formatting cleanup; different topics." + }, + { + "left": "pull_request:11407", + "right": "pull_request:12265", + "accept": false, + "reason": "Both are typo sweeps, but they target different file sets and unrelated edits; not the same underlying fix." + }, + { + "left": "pull_request:11416", + "right": "pull_request:12265", + "accept": false, + "reason": "Docs/comments typo cleanup vs broader typo cleanup across code and notebooks; too general to be the same change." + }, + { + "left": "pull_request:11476", + "right": "pull_request:12265", + "accept": false, + "reason": "Another typo-fix sweep, but the affected files and exact edits differ; not mergeable as one PR." + }, + { + "left": "pull_request:1162", + "right": "pull_request:1555", + "accept": false, + "reason": "Community links cleanup in docs vs a cache-dir bug in dynamic module loading; different problems." + }, + { + "left": "pull_request:1162", + "right": "pull_request:2257", + "accept": false, + "reason": "Docs link fix vs adding a new community pipeline feature; unrelated." + }, + { + "left": "pull_request:1162", + "right": "pull_request:3063", + "accept": false, + "reason": "Docs link fix vs pipeline __setattr__ bug fix; different code paths and intents." + }, + { + "left": "pull_request:1162", + "right": "pull_request:3266", + "accept": false, + "reason": "Docs link fix vs dynamic module version-loading bug; not the same issue." + }, + { + "left": "pull_request:11746", + "right": "pull_request:12560", + "accept": false, + "reason": "License-year refresh vs modular pipeline refactor; unrelated maintenance work." + }, + { + "left": "pull_request:11838", + "right": "pull_request:11879", + "accept": false, + "reason": "Deprecated-pipeline docs warnings vs removing tests for deprecated pipelines; related theme, but not the same change." + }, + { + "left": "pull_request:11847", + "right": "pull_request:12068", + "accept": false, + "reason": "Wan single_file example fix vs Qwen docs corrections; different models and docs." + }, + { + "left": "pull_request:11847", + "right": "pull_request:12487", + "accept": false, + "reason": "Wan example fix vs broad broken-link cleanup across many docs; separate doc tasks." + }, + { + "left": "pull_request:11857", + "right": "pull_request:12337", + "accept": false, + "reason": "Flux docs reference update vs a Flux xformers runtime fix; one is docs-only, the other changes code behavior." + }, + { + "left": "pull_request:11875", + "right": "pull_request:11958", + "accept": false, + "reason": "Kontext auto-pipeline addition vs F5 TTS pipeline addition; different features and codebases." + }, + { + "left": "pull_request:11983", + "right": "pull_request:12890", + "accept": false, + "reason": "Installation command correction vs removing hfoptions markup in wan docs; unrelated docs edits." + }, + { + "left": "pull_request:1218", + "right": "pull_request:98", + "accept": false, + "reason": "Different conversion scripts for different model families; not the same migration or bug fix." + }, + { + "left": "pull_request:12183", + "right": "pull_request:12984", + "accept": false, + "reason": "QwenImage LoRA-button docs update vs Flux2 Klein docs addition; unrelated documentation changes." + }, + { + "left": "pull_request:12247", + "right": "pull_request:12307", + "accept": false, + "reason": "AutoGuidance validation fix vs a broad modular cleanup PR; different scope and code." + }, + { + "left": "pull_request:12247", + "right": "pull_request:13115", + "accept": false, + "reason": "Specific AutoGuidance validation bug vs generic modular pipeline fixes; not the same underlying problem." + }, + { + "left": "pull_request:12360", + "right": "pull_request:12902", + "accept": false, + "reason": "Chroma docs wording update vs distributed inference device_map fix; unrelated." + }, + { + "left": "pull_request:12360", + "right": "pull_request:13135", + "accept": false, + "reason": "Chroma docs update vs LTX2 docstring fix; different pipelines and edits." + }, + { + "left": "pull_request:12487", + "right": "pull_request:12716", + "accept": false, + "reason": "General broken-link cleanup vs Flux2 link correction; both docs, but not the same concrete issue." + }, + { + "left": "pull_request:12560", + "right": "pull_request:13148", + "accept": false, + "reason": "Different modular-doc/refactor updates touching different structures; too broad to be duplicates." + }, + { + "left": "pull_request:1269", + "right": "pull_request:1527", + "accept": false, + "reason": "Stable Diffusion refactor vs inpaint tensor-input support; distinct pipeline behavior changes." + }, + { + "left": "pull_request:12731", + "right": "pull_request:12797", + "accept": false, + "reason": "Bria Fibo minor docs fixes vs Kandinsky docs fixes; unrelated model documentation." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 53, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:12883", + "pull_request:12984", + "pull_request:13135", + "pull_request:13158", + "pull_request:13253", + "pull_request:13254", + "pull_request:13315", + "pull_request:13335", + "pull_request:13368", + "pull_request:13369", + "pull_request:13371", + "pull_request:1341", + "pull_request:13567", + "pull_request:136", + "pull_request:1380", + "pull_request:1386", + "pull_request:1408", + "pull_request:1418", + "pull_request:1487", + "pull_request:1505", + "pull_request:1510", + "pull_request:1582", + "pull_request:1583", + "pull_request:1601", + "pull_request:1627", + "pull_request:168", + "pull_request:1701", + "pull_request:1761", + "pull_request:1763", + "pull_request:1769", + "pull_request:178", + "pull_request:190", + "pull_request:2415", + "pull_request:2795", + "pull_request:299", + "pull_request:3", + "pull_request:3074", + "pull_request:3079", + "pull_request:3183", + "pull_request:3549", + "pull_request:43", + "pull_request:5", + "pull_request:61", + "pull_request:63", + "pull_request:904", + "pull_request:971", + "pull_request:98" + ], + "soft_pairs": [ + "pull_request:12883|pull_request:13135", + "pull_request:12984|pull_request:13158", + "pull_request:12984|pull_request:13567", + "pull_request:13253|pull_request:13368", + "pull_request:13253|pull_request:13369", + "pull_request:13254|pull_request:13371", + "pull_request:13315|pull_request:13371", + "pull_request:13335|pull_request:13371", + "pull_request:1341|pull_request:178", + "pull_request:136|pull_request:43", + "pull_request:1380|pull_request:3074", + "pull_request:1386|pull_request:1505", + "pull_request:1408|pull_request:61", + "pull_request:1408|pull_request:63", + "pull_request:1418|pull_request:1763", + "pull_request:1487|pull_request:1627", + "pull_request:1487|pull_request:3549", + "pull_request:1510|pull_request:98", + "pull_request:1582|pull_request:2415", + "pull_request:1582|pull_request:2795", + "pull_request:1583|pull_request:2415", + "pull_request:1583|pull_request:2795", + "pull_request:1601|pull_request:3183", + "pull_request:168|pull_request:299", + "pull_request:168|pull_request:904", + "pull_request:1701|pull_request:3079", + "pull_request:1761|pull_request:1769", + "pull_request:178|pull_request:5", + "pull_request:178|pull_request:971", + "pull_request:190|pull_request:3" + ], + "prepared_review_unit_hash": "412236cf554ef57b7426b086ec56f07798e28c4c425c35e4e82e4262a51a10b5", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 96001, + "estimated_input_tokens": 24001, + "estimated_eval_tokens": 48258 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: it mixes unrelated docs fixes, test refactors, scheduler/code bugfixes, and conversion-script changes that only share broad maintenance language. No single PR clearly represents the same underlying change across the cluster.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:12883", + "right": "pull_request:13135", + "accept": false, + "reason": "Both are small docs edits, but they target different pipelines and different content; no shared bug or change." + }, + { + "left": "pull_request:12984", + "right": "pull_request:13158", + "accept": false, + "reason": "Unrelated documentation additions for different pipelines; same docs area is too broad to merge as duplicates." + }, + { + "left": "pull_request:12984", + "right": "pull_request:13567", + "accept": false, + "reason": "Klein docs vs repo agent-doc updates are unrelated changes." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13368", + "accept": false, + "reason": "Different test refactors for different model families; superficial CI-style similarity only." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13369", + "accept": false, + "reason": "Different model test suites and different refactor goals; not the same underlying bug/change." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13371", + "accept": false, + "reason": "LTX transformer tests and WAN autoencoder tests are unrelated refactors." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13371", + "accept": false, + "reason": "Different test modules for different architectures; no shared concrete fix." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13371", + "accept": false, + "reason": "Cosmos transformer test refactor vs WAN autoencoder test refactor are separate maintenance changes." + }, + { + "left": "pull_request:1341", + "right": "pull_request:178", + "accept": false, + "reason": "Stable Diffusion RePaint implementation vs stable-diffusion pipeline test update are unrelated." + }, + { + "left": "pull_request:136", + "right": "pull_request:43", + "accept": false, + "reason": "Manifest packaging fix and model-card-template addition are different changes." + }, + { + "left": "pull_request:1380", + "right": "pull_request:3074", + "accept": false, + "reason": "One fixes a Transformer2DModel code path, the other only corrects a docstring; not duplicate fixes." + }, + { + "left": "pull_request:1386", + "right": "pull_request:1505", + "accept": false, + "reason": "Both relate to v-prediction, but they touch different scheduler sets and appear as separate feature expansions, not one duplicate PR." + }, + { + "left": "pull_request:1408", + "right": "pull_request:61", + "accept": false, + "reason": "Different conversion scripts for different model families; no shared bug or change." + }, + { + "left": "pull_request:1408", + "right": "pull_request:63", + "accept": false, + "reason": "Sample-size conversion fix vs BDDM conversion script are unrelated converter changes." + }, + { + "left": "pull_request:1418", + "right": "pull_request:1763", + "accept": false, + "reason": "Versatile Diffusion tests/docs fix vs attention-mask behavior fix in attention.py are separate issues." + }, + { + "left": "pull_request:1487", + "right": "pull_request:1627", + "accept": false, + "reason": "Docs-only xformers updates versus actual model helper changes; same topic, different code-path and intent." + }, + { + "left": "pull_request:1487", + "right": "pull_request:3549", + "accept": false, + "reason": "Both are docs PRs, but for different subjects and pipelines; not the same change." + }, + { + "left": "pull_request:1510", + "right": "pull_request:98", + "accept": false, + "reason": "Different conversion-tooling PRs for different model types and scripts; not duplicates." + }, + { + "left": "pull_request:1582", + "right": "pull_request:2415", + "accept": false, + "reason": "RePaint scheduler fix and pipeline_utils import/version handling are unrelated." + }, + { + "left": "pull_request:1582", + "right": "pull_request:2795", + "accept": false, + "reason": "RePaint bugfix vs LoRA training-script tweak are unrelated changes." + }, + { + "left": "pull_request:1583", + "right": "pull_request:2415", + "accept": false, + "reason": "Stable Diffusion inpaint-legacy bugfix vs pipeline_utils version logic are unrelated." + }, + { + "left": "pull_request:1583", + "right": "pull_request:2795", + "accept": false, + "reason": "Different codebases and different problems; no duplicate overlap." + }, + { + "left": "pull_request:1601", + "right": "pull_request:3183", + "accept": false, + "reason": "Pipeline classifier-free-guidance bugfix vs DreamBooth LoRA checkpointing fix are unrelated." + }, + { + "left": "pull_request:168", + "right": "pull_request:299", + "accept": false, + "reason": "Stable Diffusion pipeline introduction and a later PNDM hotfix are not the same underlying change." + }, + { + "left": "pull_request:168", + "right": "pull_request:904", + "accept": false, + "reason": "Broad Stable Diffusion pipeline work vs inpainting implementation are separate features." + }, + { + "left": "pull_request:1701", + "right": "pull_request:3079", + "accept": false, + "reason": "RePaint API/test cleanup and attend-and-excite fast-test tuning are unrelated test/pipeline changes." + }, + { + "left": "pull_request:1761", + "right": "pull_request:1769", + "accept": false, + "reason": "Both concern UnCLIP, but one adds CPU offloading in code while the other only adjusts test settings and thresholds." + }, + { + "left": "pull_request:178", + "right": "pull_request:5", + "accept": false, + "reason": "Stable Diffusion pipeline test update vs latent-diffusion UNet implementation are unrelated." + }, + { + "left": "pull_request:178", + "right": "pull_request:971", + "accept": false, + "reason": "Pipeline test changes and community bit-diffusion example/docs are unrelated." + }, + { + "left": "pull_request:190", + "right": "pull_request:3", + "accept": false, + "reason": "SciPy-import guard in a pipeline is unrelated to the from-hub configuration/import infrastructure fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 54, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2077", + "pull_request:2086", + "pull_request:21", + "pull_request:2118", + "pull_request:2137", + "pull_request:2138", + "pull_request:2175", + "pull_request:2193", + "pull_request:22", + "pull_request:2208", + "pull_request:2278", + "pull_request:2290", + "pull_request:2320", + "pull_request:2329", + "pull_request:2433", + "pull_request:2446", + "pull_request:2539", + "pull_request:2540", + "pull_request:2566", + "pull_request:2587", + "pull_request:260", + "pull_request:2604", + "pull_request:2624", + "pull_request:2715", + "pull_request:2952", + "pull_request:299", + "pull_request:3063", + "pull_request:3068", + "pull_request:3118", + "pull_request:3389", + "pull_request:3474", + "pull_request:353", + "pull_request:3549", + "pull_request:3554", + "pull_request:3555", + "pull_request:3572", + "pull_request:3971", + "pull_request:407", + "pull_request:421", + "pull_request:4275", + "pull_request:428", + "pull_request:4856", + "pull_request:715", + "pull_request:72", + "pull_request:798", + "pull_request:910", + "pull_request:948" + ], + "soft_pairs": [ + "pull_request:2077|pull_request:3068", + "pull_request:2077|pull_request:3554", + "pull_request:2077|pull_request:3555", + "pull_request:2086|pull_request:428", + "pull_request:21|pull_request:72", + "pull_request:2118|pull_request:715", + "pull_request:2137|pull_request:2715", + "pull_request:2137|pull_request:3118", + "pull_request:2138|pull_request:948", + "pull_request:2175|pull_request:3474", + "pull_request:2193|pull_request:2952", + "pull_request:22|pull_request:72", + "pull_request:2208|pull_request:910", + "pull_request:2278|pull_request:299", + "pull_request:2290|pull_request:3971", + "pull_request:2320|pull_request:260", + "pull_request:2320|pull_request:353", + "pull_request:2320|pull_request:407", + "pull_request:2320|pull_request:798", + "pull_request:2329|pull_request:3389", + "pull_request:2433|pull_request:3549", + "pull_request:2446|pull_request:421", + "pull_request:2446|pull_request:4275", + "pull_request:2539|pull_request:3063", + "pull_request:2540|pull_request:3572", + "pull_request:2566|pull_request:407", + "pull_request:2566|pull_request:798", + "pull_request:2587|pull_request:3474", + "pull_request:2604|pull_request:4856", + "pull_request:2624|pull_request:3474" + ], + "prepared_review_unit_hash": "0eed1beea527077faaa73921d602cdd5959010eb0afa6602cdf888742a0d77c2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 90655, + "estimated_input_tokens": 22664, + "estimated_eval_tokens": 45584 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is mostly a grab-bag of unrelated docs, typo, and test PRs that only look superficially similar. None of the provided soft edges are true duplicate/change matches; the only obvious near-duplicate in the broader set is the pair of identical Moreh textual inversion PRs, but that pair is not one of the listed soft edges.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 3554 is the best representative PR here because it is the earlier copy of the only near-identical duplicate-style change in the set (3554/3555), and there are no issue artifacts to anchor on.", + "best_issue_reason": null, + "best_pr_reason": "PR 3554 is the strongest standalone PR candidate for the cluster: it is the earliest clear duplicate-like submission and is more representative than the many unrelated docs/typo/test PRs around it.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2077", + "right": "pull_request:3068", + "accept": false, + "reason": "Textual-inversion validation-image support vs docs finishing textual inversion; different change scope and code path." + }, + { + "left": "pull_request:2077", + "right": "pull_request:3554", + "accept": false, + "reason": "Validation-image feature work is unrelated to the Moreh textual inversion environment/script bundle." + }, + { + "left": "pull_request:2077", + "right": "pull_request:3555", + "accept": false, + "reason": "Validation-image feature work is unrelated to the Moreh textual inversion environment/script bundle." + }, + { + "left": "pull_request:2086", + "right": "pull_request:428", + "accept": false, + "reason": "LoRA docs addition is not the same as the broad docs restructuring in Using Diffusers." + }, + { + "left": "pull_request:21", + "right": "pull_request:72", + "accept": false, + "reason": "Different test additions for different model blocks; they do not fix the same underlying behavior." + }, + { + "left": "pull_request:2118", + "right": "pull_request:715", + "accept": false, + "reason": "One fixes an import-utils spelling entry; the other guards testing-utils imports when PyTorch is absent." + }, + { + "left": "pull_request:2137", + "right": "pull_request:2715", + "accept": false, + "reason": "Loaders typo/doc cleanup vs a broad typo sweep across many modules; not the same concrete change." + }, + { + "left": "pull_request:2137", + "right": "pull_request:3118", + "accept": false, + "reason": "Loaders typo/doc cleanup vs ControlNet default-argument fix; unrelated code paths." + }, + { + "left": "pull_request:2138", + "right": "pull_request:948", + "accept": false, + "reason": "Docs typo in loaders vs comment typo in DDPM pipeline; unrelated artifacts." + }, + { + "left": "pull_request:2175", + "right": "pull_request:3474", + "accept": false, + "reason": "Diffusion pipeline docs update vs attention-processor docs page; separate documentation additions." + }, + { + "left": "pull_request:2193", + "right": "pull_request:2952", + "accept": false, + "reason": "SAG pipeline introduction vs later SAG docs update; the latter is a follow-up, not the same PR-worthy change." + }, + { + "left": "pull_request:22", + "right": "pull_request:72", + "accept": false, + "reason": "Different test targets and different code areas; not the same bug or change." + }, + { + "left": "pull_request:2208", + "right": "pull_request:910", + "accept": false, + "reason": "xFormers installation docs vs inpainting pipeline launch docs; unrelated content." + }, + { + "left": "pull_request:2278", + "right": "pull_request:299", + "accept": false, + "reason": "Video Stable Diffusion project work vs a PNDM scheduler hotfix; completely different changes." + }, + { + "left": "pull_request:2290", + "right": "pull_request:3971", + "accept": false, + "reason": "Pix2Pix doc formatting fix vs SDXL docs/workflow changes; unrelated." + }, + { + "left": "pull_request:2320", + "right": "pull_request:260", + "accept": false, + "reason": "Stable Diffusion checkpoint-loading typo fix vs README typo fix; different files and behavior." + }, + { + "left": "pull_request:2320", + "right": "pull_request:353", + "accept": false, + "reason": "Checkpoint-loading typo fix vs UNet block error-message typo; unrelated." + }, + { + "left": "pull_request:2320", + "right": "pull_request:407", + "accept": false, + "reason": "Checkpoint-loading typo fix vs attention-slicing typo fix; different code paths." + }, + { + "left": "pull_request:2320", + "right": "pull_request:798", + "accept": false, + "reason": "Checkpoint-loading typo fix vs UNet2DCondition docstring typo; unrelated." + }, + { + "left": "pull_request:2329", + "right": "pull_request:3389", + "accept": false, + "reason": "Workflow refactor vs attention-refactor integration-test PR; not the same concrete change." + }, + { + "left": "pull_request:2433", + "right": "pull_request:3549", + "accept": false, + "reason": "SDEdit note in img2img docs vs qualitative evaluation docs; unrelated documentation changes." + }, + { + "left": "pull_request:2446", + "right": "pull_request:421", + "accept": false, + "reason": "Pix2Pix docs snippet correction vs scheduler docs typo fix; separate topics." + }, + { + "left": "pull_request:2446", + "right": "pull_request:4275", + "accept": false, + "reason": "Pix2Pix docs snippet correction vs AutoPipeline docs fix; unrelated docs pages." + }, + { + "left": "pull_request:2539", + "right": "pull_request:3063", + "accept": false, + "reason": "Pix2Pix Zero doc default-value correction vs pipeline __setattr__ behavior fix; not the same bug." + }, + { + "left": "pull_request:2540", + "right": "pull_request:3572", + "accept": false, + "reason": "ControlNet docs correction vs inpainting ControlNet example rewrite; different docs changes." + }, + { + "left": "pull_request:2566", + "right": "pull_request:407", + "accept": false, + "reason": "TensorBoard tracker typo fix vs attention slicing typo fix; unrelated behavior." + }, + { + "left": "pull_request:2566", + "right": "pull_request:798", + "accept": false, + "reason": "TensorBoard tracker typo fix vs a docstring typo in UNet2DCondition; unrelated." + }, + { + "left": "pull_request:2587", + "right": "pull_request:3474", + "accept": false, + "reason": "Chinese translation docs PR vs attention-processor docs page; different documentation sets." + }, + { + "left": "pull_request:2604", + "right": "pull_request:4856", + "accept": false, + "reason": "Colab link fix in index vs SDXL API-to-guide link fix; not the same change." + }, + { + "left": "pull_request:2624", + "right": "pull_request:3474", + "accept": false, + "reason": "Pipeline cache_dir doc addition vs attention-processor docs page; unrelated docs content." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 55, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2639", + "pull_request:2701", + "pull_request:2715", + "pull_request:2760", + "pull_request:2767", + "pull_request:2792", + "pull_request:2808", + "pull_request:2839", + "pull_request:2869", + "pull_request:2873", + "pull_request:2894", + "pull_request:2911", + "pull_request:2912", + "pull_request:2913", + "pull_request:2922", + "pull_request:2924", + "pull_request:2925", + "pull_request:2926", + "pull_request:2927", + "pull_request:2929", + "pull_request:2930", + "pull_request:2931", + "pull_request:2932", + "pull_request:2933", + "pull_request:2935", + "pull_request:2936", + "pull_request:2937", + "pull_request:2938", + "pull_request:2939", + "pull_request:299", + "pull_request:3012", + "pull_request:3063", + "pull_request:309", + "pull_request:3132", + "pull_request:3318", + "pull_request:3345", + "pull_request:3359", + "pull_request:3449", + "pull_request:3458", + "pull_request:3460", + "pull_request:3744", + "pull_request:3755", + "pull_request:3820", + "pull_request:4472", + "pull_request:568", + "pull_request:824", + "pull_request:972", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:2639|pull_request:2869", + "pull_request:2639|pull_request:2911", + "pull_request:2639|pull_request:2912", + "pull_request:2639|pull_request:2913", + "pull_request:2639|pull_request:2922", + "pull_request:2639|pull_request:2924", + "pull_request:2639|pull_request:2925", + "pull_request:2639|pull_request:2926", + "pull_request:2639|pull_request:2927", + "pull_request:2639|pull_request:2929", + "pull_request:2639|pull_request:2930", + "pull_request:2639|pull_request:2931", + "pull_request:2639|pull_request:2932", + "pull_request:2639|pull_request:2933", + "pull_request:2639|pull_request:2935", + "pull_request:2639|pull_request:2936", + "pull_request:2639|pull_request:2937", + "pull_request:2639|pull_request:2938", + "pull_request:2639|pull_request:2939", + "pull_request:2701|pull_request:3458", + "pull_request:2701|pull_request:3460", + "pull_request:2715|pull_request:568", + "pull_request:2760|pull_request:3318", + "pull_request:2767|pull_request:3359", + "pull_request:2792|pull_request:3449", + "pull_request:2808|pull_request:3755", + "pull_request:2839|pull_request:4472", + "pull_request:2873|pull_request:3744", + "pull_request:2894|pull_request:824", + "pull_request:299|pull_request:309", + "pull_request:3012|pull_request:3820", + "pull_request:3063|pull_request:3345", + "pull_request:3132|pull_request:3345", + "pull_request:3132|pull_request:972", + "pull_request:3132|pull_request:975" + ], + "prepared_review_unit_hash": "de142ece76f3d6df4a5cfd2836dbf18ceb79a38685b160a41a76970be13a2c96", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 35, + "serialized_chars": 86992, + "estimated_input_tokens": 21748, + "estimated_eval_tokens": 43752 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a pile of unrelated small documentation edits, test tweaks, and a few independent code fixes. The soft matches are largely title/topic similarity rather than the same underlying change, so they should stay separate.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out: the items span unrelated docs nits, test adjustments, and code fixes, with only superficial similarity in wording or subsystem.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global duplicate representative here; the strongest-looking pairs are still different files or different concrete behaviors, so there is no shared mergeable change to anchor the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2639", + "right": "pull_request:2869", + "accept": false, + "reason": "Both are tiny docs edits, but they update different pages with different wording fixes; same style, not the same change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2911", + "accept": false, + "reason": "Different documentation files and unrelated sentence punctuation fixes." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2912", + "accept": false, + "reason": "Different docs page; only the PR titles are similarly generic." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2913", + "accept": false, + "reason": "One edits basic_training.mdx, the other stable_unclip.mdx; unrelated documentation changes." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2922", + "accept": false, + "reason": "Different docs targets with no shared underlying bug or content." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2924", + "accept": false, + "reason": "Different docs pages; both are minor prose fixes only." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2925", + "accept": false, + "reason": "Same generic docs-update pattern, but separate files and separate edits." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2926", + "accept": false, + "reason": "Different scheduler docs page; not the same underlying change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2927", + "accept": false, + "reason": "Unrelated docs edits in different API pages." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2929", + "accept": false, + "reason": "Different documentation file and different typo cleanup." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2930", + "accept": false, + "reason": "Same superficial 'update docs' theme, but distinct file and wording." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2931", + "accept": false, + "reason": "Different scheduler docs page; no shared concrete fix." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2932", + "accept": false, + "reason": "Different docs page; both are punctuation/formatting nits only." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2933", + "accept": false, + "reason": "Separate docs pages with unrelated copy fixes." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2935", + "accept": false, + "reason": "Different docs file and different sentence-level correction." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2936", + "accept": false, + "reason": "Different docs page; no shared code path or content change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2937", + "accept": false, + "reason": "Different scheduler doc, only similarly small prose edits." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2938", + "accept": false, + "reason": "Different docs page with unrelated wording cleanup." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2939", + "accept": false, + "reason": "Different API docs page; not the same underlying issue." + }, + { + "left": "pull_request:2701", + "right": "pull_request:3458", + "accept": false, + "reason": "One adds ONNX Runtime ROCm support; the other adds inpaint LoRA/cross-attention support. Different features, different code paths." + }, + { + "left": "pull_request:2701", + "right": "pull_request:3460", + "accept": false, + "reason": "ONNXRuntime backend support is unrelated to inpaint LoRA scale support in Stable Diffusion." + }, + { + "left": "pull_request:2715", + "right": "pull_request:568", + "accept": false, + "reason": "Both are broad typo fixes, but they touch different files and different kinds of edits; not the same concrete change." + }, + { + "left": "pull_request:2760", + "right": "pull_request:3318", + "accept": false, + "reason": "Both mention textual inversion, but one is a quantization/distillation example and the other is a WIP IF textual inversion feature; different implementations and goals." + }, + { + "left": "pull_request:2767", + "right": "pull_request:3359", + "accept": false, + "reason": "One fixes `use_peft` logic in the LoRA trainer; the other disables safety checker in a different training script." + }, + { + "left": "pull_request:2792", + "right": "pull_request:3449", + "accept": false, + "reason": "Both involve tests/CI on MPS, but one skips specific tests while the other reworks broader MPS and ONNX test workflows; not one change." + }, + { + "left": "pull_request:2808", + "right": "pull_request:3755", + "accept": false, + "reason": "Both relax tests, but they cover different models and different failure modes." + }, + { + "left": "pull_request:2839", + "right": "pull_request:4472", + "accept": false, + "reason": "Different README updates with no shared bug or change set." + }, + { + "left": "pull_request:2873", + "right": "pull_request:3744", + "accept": false, + "reason": "Both are docs cleanups, but one changes phrasing in contributing docs and the other updates a workflow secret name." + }, + { + "left": "pull_request:2894", + "right": "pull_request:824", + "accept": false, + "reason": "Adding a Flax requirement is unrelated to adding a Flax test harness/change." + }, + { + "left": "pull_request:299", + "right": "pull_request:309", + "accept": false, + "reason": "The hotfix in PNDM scheduler code is unrelated to the README notebook link fix." + }, + { + "left": "pull_request:3012", + "right": "pull_request:3820", + "accept": false, + "reason": "A comment typo fix is not the same as the LDM3D pipeline/docs update." + }, + { + "left": "pull_request:3063", + "right": "pull_request:3345", + "accept": false, + "reason": "The pipeline `__setattr__` bugfix is unrelated to a README example class rename." + }, + { + "left": "pull_request:3132", + "right": "pull_request:3345", + "accept": false, + "reason": "A type annotation fix in the Score SDE pipeline does not match a README example change." + }, + { + "left": "pull_request:3132", + "right": "pull_request:972", + "accept": false, + "reason": "Unrelated: score_sde pipeline typing versus a Stable Diffusion README typo fix." + }, + { + "left": "pull_request:3132", + "right": "pull_request:975", + "accept": false, + "reason": "Different docs/config file and different kind of change; no duplicate basis." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 56, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3183", + "pull_request:3345", + "pull_request:3446", + "pull_request:3458", + "pull_request:3460", + "pull_request:3577", + "pull_request:3641", + "pull_request:3709", + "pull_request:3710", + "pull_request:3786", + "pull_request:3819", + "pull_request:3821", + "pull_request:3894", + "pull_request:3923", + "pull_request:3971", + "pull_request:3985", + "pull_request:4012", + "pull_request:408", + "pull_request:4172", + "pull_request:4275", + "pull_request:4320", + "pull_request:4350", + "pull_request:4352", + "pull_request:4472", + "pull_request:4596", + "pull_request:4626", + "pull_request:4721", + "pull_request:4746", + "pull_request:4822", + "pull_request:4853", + "pull_request:4869", + "pull_request:4887", + "pull_request:4965", + "pull_request:5250", + "pull_request:5286", + "pull_request:5454", + "pull_request:5505", + "pull_request:5568", + "pull_request:5640", + "pull_request:5707", + "pull_request:5983", + "pull_request:6311", + "pull_request:6339", + "pull_request:6342", + "pull_request:6592", + "pull_request:6787", + "pull_request:710", + "pull_request:957" + ], + "soft_pairs": [ + "pull_request:3183|pull_request:957", + "pull_request:3345|pull_request:5568", + "pull_request:3446|pull_request:4352", + "pull_request:3458|pull_request:5250", + "pull_request:3460|pull_request:5250", + "pull_request:3577|pull_request:4626", + "pull_request:3641|pull_request:4352", + "pull_request:3709|pull_request:4853", + "pull_request:3710|pull_request:3821", + "pull_request:3710|pull_request:5286", + "pull_request:3786|pull_request:3894", + "pull_request:3786|pull_request:4172", + "pull_request:3819|pull_request:5454", + "pull_request:3923|pull_request:5983", + "pull_request:3971|pull_request:4721", + "pull_request:3971|pull_request:6339", + "pull_request:3985|pull_request:5640", + "pull_request:4012|pull_request:5640", + "pull_request:408|pull_request:710", + "pull_request:4275|pull_request:4965", + "pull_request:4320|pull_request:5505", + "pull_request:4350|pull_request:6339", + "pull_request:4472|pull_request:6342", + "pull_request:4596|pull_request:6311", + "pull_request:4746|pull_request:6592", + "pull_request:4822|pull_request:5505", + "pull_request:4869|pull_request:5707", + "pull_request:4887|pull_request:6787" + ], + "prepared_review_unit_hash": "0935faf877a7a4c8df834da6ae3bd862faffb471dc22403f2bb07a88376bf487", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 94972, + "estimated_input_tokens": 23743, + "estimated_eval_tokens": 47742 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft links are mostly superficial overlaps (docs/typos/shared filenames) rather than the same underlying bug or change. I would reject all candidate duplicate pairs.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue artifacts are present. If a PR must be chosen as the anchor, pull_request:3460 is the strongest representative because it is a merged, self-contained behavioral fix with accompanying tests and a clear concrete code path.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:3460 is the best PR anchor among this set: it has a focused functional change, tests, and a clear problem statement, unlike the mostly docs-only or loosely related fixes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3183", + "right": "pull_request:957", + "accept": false, + "reason": "Both are bugfix PRs, but they fix different code paths: DreamBooth LoRA checkpointing vs DDPMScheduler args in text-to-image training." + }, + { + "left": "pull_request:3345", + "right": "pull_request:5568", + "accept": false, + "reason": "One is a README example class typo; the other fixes custom pipeline loading logic. Same broad area, not the same bug." + }, + { + "left": "pull_request:3446", + "right": "pull_request:4352", + "accept": false, + "reason": "A warning-message wording fix is unrelated to Kandinsky combined-pipeline functionality." + }, + { + "left": "pull_request:3458", + "right": "pull_request:5250", + "accept": false, + "reason": "Inpaint LoRA scale support and DDPO documentation are unrelated changes." + }, + { + "left": "pull_request:3460", + "right": "pull_request:5250", + "accept": false, + "reason": "Same as above: inpaint LoRA support vs DDPO docs, no shared underlying change." + }, + { + "left": "pull_request:3577", + "right": "pull_request:4626", + "accept": false, + "reason": "Both touch docs links, but they update different pages and different broken links." + }, + { + "left": "pull_request:3641", + "right": "pull_request:4352", + "accept": false, + "reason": "Initial-latents inpainting fix is unrelated to Kandinsky combined-pipeline method additions." + }, + { + "left": "pull_request:3709", + "right": "pull_request:4853", + "accept": false, + "reason": "Adding image-processor docs is unrelated to the image-processor width bug fix." + }, + { + "left": "pull_request:3710", + "right": "pull_request:3821", + "accept": false, + "reason": "Different documentation content and locale: EN image-processor docs vs Korean translation of another guide." + }, + { + "left": "pull_request:3710", + "right": "pull_request:5286", + "accept": false, + "reason": "These are different localized docs additions, not the same artifact or bug." + }, + { + "left": "pull_request:3786", + "right": "pull_request:3894", + "accept": false, + "reason": "PR template addition is unrelated to Flax UNet dropout behavior." + }, + { + "left": "pull_request:3786", + "right": "pull_request:4172", + "accept": false, + "reason": "PR template addition is unrelated to Flax type-casting changes." + }, + { + "left": "pull_request:3819", + "right": "pull_request:5454", + "accept": false, + "reason": "An FAQ README edit and a schedulers docs wording fix are unrelated." + }, + { + "left": "pull_request:3923", + "right": "pull_request:5983", + "accept": false, + "reason": "Torch import fix in Stable Diffusion docs is unrelated to SVD documentation formatting." + }, + { + "left": "pull_request:3971", + "right": "pull_request:4721", + "accept": false, + "reason": "SDXL docs/workflow fixes and Korean DreamBooth docs changes are unrelated." + }, + { + "left": "pull_request:3971", + "right": "pull_request:6339", + "accept": false, + "reason": "Docs build / SDXL doc fixes are unrelated to AnimateDiff docs wording updates." + }, + { + "left": "pull_request:3985", + "right": "pull_request:5640", + "accept": false, + "reason": "SDXL documentation tweaks and AnimateDiff documentation tweaks are different changes." + }, + { + "left": "pull_request:4012", + "right": "pull_request:5640", + "accept": false, + "reason": "Dance Diffusion doc wording changes are unrelated to AnimateDiff doc nits." + }, + { + "left": "pull_request:408", + "right": "pull_request:710", + "accept": false, + "reason": "General docs typo cleanup is unrelated to the specific docstring formatting fix." + }, + { + "left": "pull_request:4275", + "right": "pull_request:4965", + "accept": false, + "reason": "AutoPipeline docs corrections and W\u00fcrstchen docs formatting are different documentation changes." + }, + { + "left": "pull_request:4320", + "right": "pull_request:5505", + "accept": false, + "reason": "ONNX docs wording and ControlNet naming / loader changes are not the same fix." + }, + { + "left": "pull_request:4350", + "right": "pull_request:6339", + "accept": false, + "reason": "T2I adapter docs dtype fix and AnimateDiff docs updates are unrelated." + }, + { + "left": "pull_request:4472", + "right": "pull_request:6342", + "accept": false, + "reason": "SDXL README cleanup and installation.md note are different documentation updates." + }, + { + "left": "pull_request:4596", + "right": "pull_request:6311", + "accept": false, + "reason": "SDXL conversion script and diffusion-fast tutorial are completely different additions." + }, + { + "left": "pull_request:4746", + "right": "pull_request:6592", + "accept": false, + "reason": "Both mention inpainting, but one is a broad WIP SDXL inpainting training refactor and the other adds a separate inpainting training script; not the same concrete change." + }, + { + "left": "pull_request:4822", + "right": "pull_request:5505", + "accept": false, + "reason": "Both touch loaders/docs, but they address different typos and different functionality." + }, + { + "left": "pull_request:4869", + "right": "pull_request:5707", + "accept": false, + "reason": "Combined-pipeline tests and LCM tests are different test additions for different features." + }, + { + "left": "pull_request:4887", + "right": "pull_request:6787", + "accept": false, + "reason": "Kandinsky prior-emb2emb latent-input removal and Marigold input-range normalization are unrelated bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 57, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5", + "pull_request:5051", + "pull_request:5184", + "pull_request:5267", + "pull_request:5332", + "pull_request:5454", + "pull_request:5533", + "pull_request:555", + "pull_request:5640", + "pull_request:5686", + "pull_request:5707", + "pull_request:5723", + "pull_request:574", + "pull_request:5855", + "pull_request:5901", + "pull_request:5982", + "pull_request:6002", + "pull_request:6174", + "pull_request:6207", + "pull_request:6215", + "pull_request:6217", + "pull_request:6314", + "pull_request:6322", + "pull_request:6360", + "pull_request:6401", + "pull_request:6419", + "pull_request:6433", + "pull_request:6453", + "pull_request:6466", + "pull_request:6475", + "pull_request:6511", + "pull_request:6633", + "pull_request:6752", + "pull_request:6760", + "pull_request:6775", + "pull_request:6787", + "pull_request:6847", + "pull_request:6904", + "pull_request:6941", + "pull_request:7150", + "pull_request:7181", + "pull_request:7353", + "pull_request:7747", + "pull_request:7932", + "pull_request:7963", + "pull_request:8427", + "pull_request:8851", + "pull_request:889" + ], + "soft_pairs": [ + "pull_request:5|pull_request:555", + "pull_request:5|pull_request:889", + "pull_request:5051|pull_request:7181", + "pull_request:5184|pull_request:5707", + "pull_request:5267|pull_request:7150", + "pull_request:5332|pull_request:6322", + "pull_request:5454|pull_request:7150", + "pull_request:5533|pull_request:5686", + "pull_request:555|pull_request:574", + "pull_request:5640|pull_request:6466", + "pull_request:5707|pull_request:5723", + "pull_request:5855|pull_request:7150", + "pull_request:5901|pull_request:6847", + "pull_request:5901|pull_request:6941", + "pull_request:5982|pull_request:6174", + "pull_request:6002|pull_request:6475", + "pull_request:6207|pull_request:6401", + "pull_request:6215|pull_request:6217", + "pull_request:6314|pull_request:6453", + "pull_request:6360|pull_request:7963", + "pull_request:6419|pull_request:6752", + "pull_request:6433|pull_request:8427", + "pull_request:6475|pull_request:7747", + "pull_request:6511|pull_request:7932", + "pull_request:6633|pull_request:7353", + "pull_request:6760|pull_request:6847", + "pull_request:6775|pull_request:8851", + "pull_request:6787|pull_request:6904" + ], + "prepared_review_unit_hash": "ff5125ad2b0e80988598ae7005ea9fa05e134d7f5d33927aa9ed7de98a38eb19", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 96048, + "estimated_input_tokens": 24012, + "estimated_eval_tokens": 48280 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "A noisy cluster of mostly unrelated docs, tests, and feature PRs across different subsystems; the shared similarity is mostly generic wording, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 6847 is the strongest single representative here because it is the most substantial and heavily reviewed change, but it still addresses a distinct IP-Adapter masking feature rather than the same change as the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5", + "right": "pull_request:555", + "accept": false, + "reason": "Different model components and scopes: UNet for latent diffusion vs Flax VAE implementation/fix." + }, + { + "left": "pull_request:5", + "right": "pull_request:889", + "accept": false, + "reason": "UNet model addition vs diffusion pipeline API/docs components exposure; not the same code-path change." + }, + { + "left": "pull_request:5051", + "right": "pull_request:7181", + "accept": false, + "reason": "Both are typo/docs cleanup, but they fix different text across different files and are not one concrete duplicate change." + }, + { + "left": "pull_request:5184", + "right": "pull_request:5707", + "accept": false, + "reason": "Scheduler test augmentation vs LCM feature/test additions; different test targets and intent." + }, + { + "left": "pull_request:5267", + "right": "pull_request:7150", + "accept": false, + "reason": "Unrelated documentation edits in different pages and contexts." + }, + { + "left": "pull_request:5332", + "right": "pull_request:6322", + "accept": false, + "reason": "DreamBooth LoRA typo fix vs training-script attention/scheduler changes; different fixes." + }, + { + "left": "pull_request:5454", + "right": "pull_request:7150", + "accept": false, + "reason": "Broken scheduler docs link vs callback example formatting; unrelated documentation issues." + }, + { + "left": "pull_request:5533", + "right": "pull_request:5686", + "accept": false, + "reason": "Different fast-test updates for different pipelines; not the same bug or mergeable PR." + }, + { + "left": "pull_request:555", + "right": "pull_request:574", + "accept": false, + "reason": "Same general Flax VAE area, but 555 adds the model/subfolder plumbing while 574 tweaks decode behavior; not the same concrete fix." + }, + { + "left": "pull_request:5640", + "right": "pull_request:6466", + "accept": false, + "reason": "AnimateDiff doc nits vs SVD doc edits; unrelated docs." + }, + { + "left": "pull_request:5707", + "right": "pull_request:5723", + "accept": false, + "reason": "LCM test/loader changes vs LCM documentation; different artifacts and code paths." + }, + { + "left": "pull_request:5855", + "right": "pull_request:7150", + "accept": false, + "reason": "Community README typo fix vs callback example formatting; unrelated docs changes." + }, + { + "left": "pull_request:5901", + "right": "pull_request:6847", + "accept": false, + "reason": "Different IP-Adapter work: ControlNet img2img integration vs attention masking feature; not one concrete PR." + }, + { + "left": "pull_request:5901", + "right": "pull_request:6941", + "accept": false, + "reason": "Different pipelines and features: ControlNet img2img IP-Adapter support vs SDXL ControlNet inpaint support." + }, + { + "left": "pull_request:5982", + "right": "pull_request:6174", + "accept": false, + "reason": "SDXL Turbo docs vs LoRA docs typo; unrelated documentation fixes." + }, + { + "left": "pull_request:6002", + "right": "pull_request:6475", + "accept": false, + "reason": "Stable Video Diffusion guidance-scale code fix vs VAE type documentation; different layers entirely." + }, + { + "left": "pull_request:6207", + "right": "pull_request:6401", + "accept": false, + "reason": "Removing deprecated training-script code vs LoRA PEFT test cleanup; different targets." + }, + { + "left": "pull_request:6215", + "right": "pull_request:6217", + "accept": false, + "reason": "PixArt test init changes vs Stable Video Diffusion torch.compile fix; unrelated test/code paths." + }, + { + "left": "pull_request:6314", + "right": "pull_request:6453", + "accept": false, + "reason": "Two separate docs typo/link fixes for different papers and pages." + }, + { + "left": "pull_request:6360", + "right": "pull_request:7963", + "accept": false, + "reason": "Both touch inpaint pipelines, but one adds padding_mask_crop while the other adds custom timesteps/sigmas; different features." + }, + { + "left": "pull_request:6419", + "right": "pull_request:6752", + "accept": false, + "reason": "WIP prompt-diffusion community pipeline vs finalized PromptDiffusion community pipeline; not the same artifact." + }, + { + "left": "pull_request:6433", + "right": "pull_request:8427", + "accept": false, + "reason": "Community pipeline addition vs workflow debug logging; no shared bug/change." + }, + { + "left": "pull_request:6475", + "right": "pull_request:7747", + "accept": false, + "reason": "SVD VAE type doc fix vs generic VAE tiling test fix; unrelated." + }, + { + "left": "pull_request:6511", + "right": "pull_request:7932", + "accept": false, + "reason": "Torch.compile checkpointing support vs LoRA test expectation updates; different code paths." + }, + { + "left": "pull_request:6633", + "right": "pull_request:7353", + "accept": false, + "reason": "Checkpoint YAML loading fix vs ControlNet SDXL img2img loader mixin wiring; unrelated." + }, + { + "left": "pull_request:6760", + "right": "pull_request:6847", + "accept": false, + "reason": "Both involve IP-Adapter, but one adjusts test expectations while the other adds attention masking; not the same concrete change." + }, + { + "left": "pull_request:6775", + "right": "pull_request:8851", + "accept": false, + "reason": "Scheduler docs parameter addition vs AuraFlow docs/model/pipeline addition; unrelated." + }, + { + "left": "pull_request:6787", + "right": "pull_request:6904", + "accept": false, + "reason": "Marigold input-range bug fix vs load_image PIL support; different utility bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 58, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11081", + "pull_request:11682", + "pull_request:11835", + "pull_request:12004", + "pull_request:12069", + "pull_request:12551", + "pull_request:12562", + "pull_request:12970", + "pull_request:12988", + "pull_request:13039", + "pull_request:6868", + "pull_request:6946", + "pull_request:6961", + "pull_request:710", + "pull_request:7150", + "pull_request:737", + "pull_request:7410", + "pull_request:7550", + "pull_request:7638", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7833", + "pull_request:7919", + "pull_request:7977", + "pull_request:798", + "pull_request:806", + "pull_request:822", + "pull_request:8303", + "pull_request:839", + "pull_request:8478", + "pull_request:8498", + "pull_request:8591", + "pull_request:8631", + "pull_request:8642", + "pull_request:8755", + "pull_request:8775", + "pull_request:8992", + "pull_request:91", + "pull_request:9266", + "pull_request:9408", + "pull_request:9473", + "pull_request:9553", + "pull_request:9591", + "pull_request:9592", + "pull_request:9699", + "pull_request:971", + "pull_request:9832" + ], + "soft_pairs": [ + "pull_request:6961|pull_request:7797", + "pull_request:6961|pull_request:7803", + "pull_request:6961|pull_request:7804", + "pull_request:6961|pull_request:7833", + "pull_request:710|pull_request:798", + "pull_request:7150|pull_request:9266", + "pull_request:737|pull_request:839", + "pull_request:7410|pull_request:7919", + "pull_request:7638|pull_request:8303", + "pull_request:7977|pull_request:8642", + "pull_request:806|pull_request:971", + "pull_request:822|pull_request:91", + "pull_request:8478|pull_request:9408", + "pull_request:8498|pull_request:8642", + "pull_request:8591|pull_request:8755", + "pull_request:8631|pull_request:9473", + "pull_request:8775|pull_request:9266", + "pull_request:9266|pull_request:9553", + "pull_request:9591|pull_request:9592", + "pull_request:9699|pull_request:9832", + "pull_request:12004|pull_request:12562", + "pull_request:11081|pull_request:11682", + "pull_request:7550|pull_request:8992", + "pull_request:6868|pull_request:6946", + "pull_request:12551|pull_request:13039", + "pull_request:12970|pull_request:12988", + "pull_request:11835|pull_request:12069" + ], + "prepared_review_unit_hash": "6b56d030acf405ef7a5cafde3a168168612cd431c0b33167c9188e0ed493ec61", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 99148, + "estimated_input_tokens": 24787, + "estimated_eval_tokens": 49830 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated false-positive similarity pairs across different subsystems (memory offloading, WAN, Flux, SD3 training, docs/tests, CI, etc.). There is no single coherent duplicate target for the whole cluster; only one pair looks like a plausible same-bug docstring fix.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:6961", + "right": "pull_request:7797", + "accept": false, + "reason": "Different work: single-file loader test scaffolding vs DDPM fast-test model downsizing." + }, + { + "left": "pull_request:6961", + "right": "pull_request:7803", + "accept": false, + "reason": "Different work: single-file loader tests vs DDIM fast-test model downsizing." + }, + { + "left": "pull_request:6961", + "right": "pull_request:7804", + "accept": false, + "reason": "Different work: single-file loader tests vs Amused fast-test model downsizing." + }, + { + "left": "pull_request:6961", + "right": "pull_request:7833", + "accept": false, + "reason": "Different work: single-file loader tests vs AudioLDM fast-test model downsizing." + }, + { + "left": "pull_request:710", + "right": "pull_request:798", + "accept": true, + "reason": "Same underlying docstring typo fix (`r\"\"\"`) in closely related UNet model files; these could plausibly be combined." + }, + { + "left": "pull_request:7150", + "right": "pull_request:9266", + "accept": false, + "reason": "Unrelated docs edits: callback example cleanup vs community projects list update." + }, + { + "left": "pull_request:737", + "right": "pull_request:839", + "accept": false, + "reason": "Both remove `use_auth_token`, but one is broad repo cleanup and the other is a DreamBooth README-only tweak." + }, + { + "left": "pull_request:7410", + "right": "pull_request:7919", + "accept": false, + "reason": "Different bugs: FreeInit progress-bar/refactor changes vs offload behavior when `hf_device_map` is enabled." + }, + { + "left": "pull_request:7638", + "right": "pull_request:8303", + "accept": false, + "reason": "Both docs typos, but unrelated files and wording fixes with no shared concrete change." + }, + { + "left": "pull_request:7977", + "right": "pull_request:8642", + "accept": false, + "reason": "Different docs typos in different Stable Diffusion 3 pages; not the same change." + }, + { + "left": "pull_request:806", + "right": "pull_request:971", + "accept": false, + "reason": "DreamBooth EMA training change vs adding a community bit-diffusion example." + }, + { + "left": "pull_request:822", + "right": "pull_request:91", + "accept": false, + "reason": "Both are scheduler/API related, but they fix different generation stacks and different concrete code paths." + }, + { + "left": "pull_request:8478", + "right": "pull_request:9408", + "accept": false, + "reason": "Different feature additions: HunyuanDiT ControlNet vs Flux ControlNet img2img/inpaint." + }, + { + "left": "pull_request:8498", + "right": "pull_request:8642", + "accept": false, + "reason": "Different typo fixes in different Stable Diffusion 3 docs files; not the same artifact." + }, + { + "left": "pull_request:8591", + "right": "pull_request:8755", + "accept": false, + "reason": "Different SD3 training changes: utility refactor vs text-encoder training bug fix." + }, + { + "left": "pull_request:8631", + "right": "pull_request:9473", + "accept": false, + "reason": "Different fixes in different single-file / Flux training contexts; no shared concrete bug." + }, + { + "left": "pull_request:8775", + "right": "pull_request:9266", + "accept": false, + "reason": "README Colab link update vs community projects table update; unrelated docs changes." + }, + { + "left": "pull_request:9266", + "right": "pull_request:9553", + "accept": false, + "reason": "Different docs updates: community project listing vs distributed inference device-map tip." + }, + { + "left": "pull_request:9591", + "right": "pull_request:9592", + "accept": false, + "reason": "Unrelated documentation refactors in different modules." + }, + { + "left": "pull_request:9699", + "right": "pull_request:9832", + "accept": false, + "reason": "Different CI additions: runner availability vs quantization nightly test job." + }, + { + "left": "pull_request:12004", + "right": "pull_request:12562", + "accept": false, + "reason": "Same Wan family, but one adds Wan2.2 support while the other fixes a CP crash in a specific Wan model." + }, + { + "left": "pull_request:11081", + "right": "pull_request:11682", + "accept": false, + "reason": "Same group-offloading area, but one adds `record_stream` and the other adds disk/NVMe offload; distinct changes." + }, + { + "left": "pull_request:7550", + "right": "pull_request:8992", + "accept": false, + "reason": "Both are community diffusion pipelines, but they target different models and different features." + }, + { + "left": "pull_request:6868", + "right": "pull_request:6946", + "accept": false, + "reason": "Both touch IP-Adapter loading, but one adds image-embed plumbing and the other adds low-CPU-memory loading support." + }, + { + "left": "pull_request:12551", + "right": "pull_request:13039", + "accept": false, + "reason": "Different Wan-related fixes: CI skip for Sana layerwise casting vs CPU upcasting fix in Wan Animate." + }, + { + "left": "pull_request:12970", + "right": "pull_request:12988", + "accept": false, + "reason": "Both are context-parallel fixes, but one is Qwen-Image-specific and the other is a general seq-len divisibility crash fix." + }, + { + "left": "pull_request:11835", + "right": "pull_request:12069", + "accept": false, + "reason": "Different test fixes in different pipelines; one is float16 CUDA tests, the other is AudioLDM2 cache-position handling." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The false-edge verdicts are well supported by the cited diffs, and the lone accepted pair (710/798) is plausibly the same docstring typo fix across related UNet files. No pair is overstated as mergeable beyond what the diffs show." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 59, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10318", + "pull_request:10581", + "pull_request:10717", + "pull_request:10911", + "pull_request:11353", + "pull_request:11588", + "pull_request:11786", + "pull_request:1215", + "pull_request:12714", + "pull_request:13313", + "pull_request:13365", + "pull_request:13536", + "pull_request:2062", + "pull_request:2892", + "pull_request:2915", + "pull_request:3446", + "pull_request:3968", + "pull_request:4034", + "pull_request:4083", + "pull_request:4099", + "pull_request:4498", + "pull_request:4572", + "pull_request:4629", + "pull_request:4632", + "pull_request:4666", + "pull_request:4714", + "pull_request:4761", + "pull_request:4783", + "pull_request:4944", + "pull_request:5015", + "pull_request:5217", + "pull_request:5407", + "pull_request:5506", + "pull_request:5615", + "pull_request:5750", + "pull_request:5852", + "pull_request:6330", + "pull_request:6816", + "pull_request:6946", + "pull_request:7118", + "pull_request:7458", + "pull_request:7893", + "pull_request:8571", + "pull_request:9175", + "pull_request:9257", + "pull_request:9506", + "pull_request:9676", + "pull_request:9770" + ], + "soft_pairs": [ + "pull_request:10581|pull_request:10717", + "pull_request:3446|pull_request:4714", + "pull_request:4034|pull_request:4666", + "pull_request:4572|pull_request:4632", + "pull_request:4629|pull_request:4783", + "pull_request:4761|pull_request:5407", + "pull_request:5506|pull_request:6816", + "pull_request:4944|pull_request:6946", + "pull_request:10318|pull_request:9257", + "pull_request:10911|pull_request:11786", + "pull_request:1215|pull_request:2062", + "pull_request:12714|pull_request:13313", + "pull_request:13365|pull_request:13536", + "pull_request:2892|pull_request:4099", + "pull_request:2915|pull_request:4666", + "pull_request:3968|pull_request:4083", + "pull_request:4498|pull_request:5750", + "pull_request:4666|pull_request:5615", + "pull_request:5015|pull_request:6330", + "pull_request:5217|pull_request:5852", + "pull_request:7118|pull_request:8571", + "pull_request:7458|pull_request:7893", + "pull_request:9175|pull_request:9770", + "pull_request:9506|pull_request:9676", + "pull_request:11353|pull_request:11588" + ], + "prepared_review_unit_hash": "26307511ef122e25ce888dd36bbd6160a2628c9332afd01059ad3262f6fbe152", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 98746, + "estimated_input_tokens": 24687, + "estimated_eval_tokens": 49630 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No duplicate cluster emerges here: the items span unrelated fixes across pipelines, docs, CI, and LoRA utilities. Even the similar-looking pairs usually touch different concrete code paths or different products, so they should stay separate.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10581", + "right": "pull_request:10717", + "accept": false, + "reason": "Different pipelines and fixes: SD3 inpaint IP-Adapter support vs Flux img2img negative embed ordering. Same broad area, not the same bug/change." + }, + { + "left": "pull_request:3446", + "right": "pull_request:4714", + "accept": false, + "reason": "Both are small error-message/import adjustments, but they fix different issues in different code paths and are not mergeable as one change." + }, + { + "left": "pull_request:4034", + "right": "pull_request:4666", + "accept": false, + "reason": "LoRA unload utility vs SDXL ControlNet LoRA support are distinct features in different subsystems." + }, + { + "left": "pull_request:4572", + "right": "pull_request:4632", + "accept": false, + "reason": "Both touch an SDXL training script, but one fixes network_alpha naming while the other fixes checkpointing/casting behavior; not the same underlying bug." + }, + { + "left": "pull_request:4629", + "right": "pull_request:4783", + "accept": false, + "reason": "Docs/navigation cleanup vs SDXL prompt-embedding semantics. Overlap is only on SDXL docs, not the concrete change." + }, + { + "left": "pull_request:4761", + "right": "pull_request:5407", + "accept": false, + "reason": "Paint-by-example encoder sizing fix is unrelated to unconditional training mixed-precision handling." + }, + { + "left": "pull_request:5506", + "right": "pull_request:6816", + "accept": false, + "reason": "Both are PEFT/LoRA-related, but one fixes xformers interaction in attention layers and the other fixes an SDXL Dreambooth training script. Different bugs." + }, + { + "left": "pull_request:4944", + "right": "pull_request:6946", + "accept": false, + "reason": "General IP-Adapter support work vs a specific low_cpu_mem_usage loading enhancement. Same subsystem, different change." + }, + { + "left": "pull_request:10318", + "right": "pull_request:9257", + "accept": false, + "reason": "Flux LoRA integration tests vs Dreambooth Flux prompt/tokenization bug fix are unrelated." + }, + { + "left": "pull_request:10911", + "right": "pull_request:11786", + "accept": false, + "reason": "Lumina2 LoRA NaN test fix vs CI marker/workflow changes; no shared bug or code path." + }, + { + "left": "pull_request:1215", + "right": "pull_request:2062", + "accept": false, + "reason": "DDIM generator/eta bug fix is unrelated to paint-by-example CPU offload support." + }, + { + "left": "pull_request:12714", + "right": "pull_request:13313", + "accept": false, + "reason": "Flux2 LoRA NaN test fix vs Klein LoRA conversion/loading support are different problems." + }, + { + "left": "pull_request:13365", + "right": "pull_request:13536", + "accept": false, + "reason": "Both alter the Claude review workflow, but one enables fork reviews and the other allows opening PRs; separate behaviors and changes." + }, + { + "left": "pull_request:2892", + "right": "pull_request:4099", + "accept": false, + "reason": "Stable Diffusion upscaler input validation vs SDXL prompt embed handling are different pipeline bugs." + }, + { + "left": "pull_request:2915", + "right": "pull_request:4666", + "accept": false, + "reason": "A flaky test tolerance adjustment is unrelated to SDXL ControlNet LoRA support." + }, + { + "left": "pull_request:3968", + "right": "pull_request:4083", + "accept": false, + "reason": "Large SDXL improvement/docs sweep vs low-precision VAE support. They share a theme but not the same concrete change." + }, + { + "left": "pull_request:4498", + "right": "pull_request:5750", + "accept": false, + "reason": "One adds SDXL reference-image functionality, the other is only a technology intro doc. Not the same fix/change." + }, + { + "left": "pull_request:4666", + "right": "pull_request:5615", + "accept": false, + "reason": "SDXL ControlNet LoRA enablement vs a revert in the T2I adapter pipeline; different code paths and goals." + }, + { + "left": "pull_request:5015", + "right": "pull_request:6330", + "accept": false, + "reason": "Broken link fix in docs vs SVD video-rendering doc fix; both docs-only but unrelated." + }, + { + "left": "pull_request:5217", + "right": "pull_request:5852", + "accept": false, + "reason": "PEFT import typo fix vs CI silent-behavior fix in workflow config; different defects." + }, + { + "left": "pull_request:7118", + "right": "pull_request:8571", + "accept": false, + "reason": "Docs typos in IP-Adapter vs broad errata/style cleanup; not the same underlying issue." + }, + { + "left": "pull_request:7458", + "right": "pull_request:7893", + "accept": false, + "reason": "Python-3.12 test skipping vs Dockerfile Python 3.10 upgrade. Different maintenance changes." + }, + { + "left": "pull_request:9175", + "right": "pull_request:9770", + "accept": false, + "reason": "Flux ControlNet union support vs a specific Xlabs ControlNet inpainting adjustment. Same area, different capability." + }, + { + "left": "pull_request:9506", + "right": "pull_request:9676", + "accept": false, + "reason": "Docs import/path fix and benchmark import fix are unrelated." + }, + { + "left": "pull_request:11353", + "right": "pull_request:11588", + "accept": false, + "reason": "Wan-FLF2V feature support vs Wan LoRA I2V support. Same model family, different concrete features and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 60, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10206", + "pull_request:10359", + "pull_request:10365", + "pull_request:10648", + "pull_request:10885", + "pull_request:10911", + "pull_request:11124", + "pull_request:11582", + "pull_request:11595", + "pull_request:11726", + "pull_request:12584", + "pull_request:13029", + "pull_request:13120", + "pull_request:13229", + "pull_request:1496", + "pull_request:2001", + "pull_request:2223", + "pull_request:2734", + "pull_request:2959", + "pull_request:3446", + "pull_request:3641", + "pull_request:3675", + "pull_request:3716", + "pull_request:4370", + "pull_request:4373", + "pull_request:4695", + "pull_request:4714", + "pull_request:483", + "pull_request:4862", + "pull_request:5275", + "pull_request:530", + "pull_request:5506", + "pull_request:5788", + "pull_request:6265", + "pull_request:6573", + "pull_request:6757", + "pull_request:6795", + "pull_request:6864", + "pull_request:689", + "pull_request:7016", + "pull_request:7113", + "pull_request:7246", + "pull_request:8510", + "pull_request:888", + "pull_request:9206", + "pull_request:9386", + "pull_request:9473", + "pull_request:978" + ], + "soft_pairs": [ + "pull_request:6573|pull_request:7016", + "pull_request:10359|pull_request:9206", + "pull_request:11582|pull_request:11726", + "pull_request:1496|pull_request:2734", + "pull_request:2001|pull_request:483", + "pull_request:2001|pull_request:978", + "pull_request:3716|pull_request:4695", + "pull_request:4370|pull_request:4862", + "pull_request:5275|pull_request:6795", + "pull_request:13029|pull_request:13120", + "pull_request:2223|pull_request:888", + "pull_request:2959|pull_request:3675", + "pull_request:3446|pull_request:4373", + "pull_request:3641|pull_request:4373", + "pull_request:4714|pull_request:5788", + "pull_request:530|pull_request:689", + "pull_request:5506|pull_request:7246", + "pull_request:6265|pull_request:6757", + "pull_request:6864|pull_request:7113", + "pull_request:8510|pull_request:9473", + "pull_request:10911|pull_request:11124", + "pull_request:12584|pull_request:13229", + "pull_request:10206|pull_request:11595", + "pull_request:10365|pull_request:9386", + "pull_request:10648|pull_request:10885" + ], + "prepared_review_unit_hash": "db81b25be6fac494a96e3449bee908a066d788853fb7f5a2d31a376d9bbd7221", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 103196, + "estimated_input_tokens": 25799, + "estimated_eval_tokens": 51854 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous cluster with only superficial overlap: most soft edges share a broad subsystem, file path, or test/docs context, but they do not fix the same concrete bug or implement the same change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents this cluster; the items span unrelated docs, tests, pipeline additions, and LoRA fixes.", + "best_issue_reason": null, + "best_pr_reason": "No global PR is a good canonical representative because the soft pairs are mostly superficial overlaps, not the same underlying change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6573", + "right": "pull_request:7016", + "accept": false, + "reason": "Both touch IP-Adapter code, but one adds multi-adapter support while the other refactors embedding prep and image-encoder loading; different changes, not one bug." + }, + { + "left": "pull_request:10359", + "right": "pull_request:9206", + "accept": false, + "reason": "Both are test fixes, but they target different failures in different pipelines and xformers paths." + }, + { + "left": "pull_request:11582", + "right": "pull_request:11726", + "accept": false, + "reason": "Wan VACE pipeline work vs Wan LoRA conversion support are related only by project name, not the same code-path problem." + }, + { + "left": "pull_request:1496", + "right": "pull_request:2734", + "accept": false, + "reason": "Both are Habana docs, but one adds documentation and the other updates benchmark/version numbers; not duplicates." + }, + { + "left": "pull_request:2001", + "right": "pull_request:483", + "accept": false, + "reason": "A ColossalAI example typo fix is unrelated to the repo-wide typo cleanup/action PR." + }, + { + "left": "pull_request:2001", + "right": "pull_request:978", + "accept": false, + "reason": "Example typo fix vs general typo cleanup across docs/code; different scope and intent." + }, + { + "left": "pull_request:3716", + "right": "pull_request:4695", + "accept": false, + "reason": "Both add pipelines, but they implement different pipelines with different code paths and features." + }, + { + "left": "pull_request:4370", + "right": "pull_request:4862", + "accept": false, + "reason": "Dummy import naming cleanup is unrelated to the SDXL ControlNet MultiControlNetModel bug fix." + }, + { + "left": "pull_request:5275", + "right": "pull_request:6795", + "accept": false, + "reason": "SDXL JAX/TPU example and SDXL micro-conditioning training script are separate features, not one change." + }, + { + "left": "pull_request:13029", + "right": "pull_request:13120", + "accept": false, + "reason": "Modular docs quickstart and modular loading tests address different artifacts and problems." + }, + { + "left": "pull_request:2223", + "right": "pull_request:888", + "accept": false, + "reason": "Both add pipelines, but CycleDiffusion and semantic guidance are distinct pipeline implementations." + }, + { + "left": "pull_request:2959", + "right": "pull_request:3675", + "accept": false, + "reason": "These are unrelated test tweaks for different models and failure modes." + }, + { + "left": "pull_request:3446", + "right": "pull_request:4373", + "accept": false, + "reason": "Pipeline loading warning-message cleanup is unrelated to the SDXL resize bug." + }, + { + "left": "pull_request:3641", + "right": "pull_request:4373", + "accept": false, + "reason": "Inpainting latent handling and SDXL resize are different code paths and issues." + }, + { + "left": "pull_request:4714", + "right": "pull_request:5788", + "accept": false, + "reason": "Both improve error messages, but for different subsystems and failures." + }, + { + "left": "pull_request:530", + "right": "pull_request:689", + "accept": false, + "reason": "Both are test-related, but they fix different flaky/slow test issues." + }, + { + "left": "pull_request:5506", + "right": "pull_request:7246", + "accept": false, + "reason": "PEFT/xformers attention fix and VAE tiling device fix are unrelated." + }, + { + "left": "pull_request:6265", + "right": "pull_request:6757", + "accept": false, + "reason": "GLIGEN refactor and SDXL k-diffusion sigma handling are separate changes." + }, + { + "left": "pull_request:6864", + "right": "pull_request:7113", + "accept": false, + "reason": "A docs note about torch.compile and a Docker CI Slack notification are unrelated." + }, + { + "left": "pull_request:8510", + "right": "pull_request:9473", + "accept": false, + "reason": "Both involve DreamBooth scripts, but they fix different learning-rate behaviors in different Flux vs non-Flux scripts." + }, + { + "left": "pull_request:10911", + "right": "pull_request:11124", + "accept": false, + "reason": "Lumina2 LoRA fuse_nan test and CogView4 test update are unrelated test changes." + }, + { + "left": "pull_request:12584", + "right": "pull_request:13229", + "accept": false, + "reason": "Both concern Sana-Video, but one adds the pipeline while the other swaps in an LTX VAE; they are not the same change." + }, + { + "left": "pull_request:10206", + "right": "pull_request:11595", + "accept": false, + "reason": "Both touch LoRA loading/unloading in Flux, but they fix different functions and different behaviors in the load/unload paths." + }, + { + "left": "pull_request:10365", + "right": "pull_request:9386", + "accept": false, + "reason": "Torch/XLA support for AuraFlow and PAG support for SD inpaint are unrelated pipeline additions." + }, + { + "left": "pull_request:10648", + "right": "pull_request:10885", + "accept": false, + "reason": "Both are docs PRs, but they document unrelated scheduler features." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 61, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11248", + "pull_request:11303", + "pull_request:11473", + "pull_request:11755", + "pull_request:11835", + "pull_request:11891", + "pull_request:12018", + "pull_request:12193", + "pull_request:12360", + "pull_request:12464", + "pull_request:2004", + "pull_request:2029", + "pull_request:2339", + "pull_request:2536", + "pull_request:2559", + "pull_request:2976", + "pull_request:3128", + "pull_request:3214", + "pull_request:3226", + "pull_request:3471", + "pull_request:3506", + "pull_request:3642", + "pull_request:4211", + "pull_request:4248", + "pull_request:4457", + "pull_request:4595", + "pull_request:4707", + "pull_request:4797", + "pull_request:4934", + "pull_request:5015", + "pull_request:5041", + "pull_request:5707", + "pull_request:5807", + "pull_request:5844", + "pull_request:6002", + "pull_request:6216", + "pull_request:6279", + "pull_request:6311", + "pull_request:6389", + "pull_request:652", + "pull_request:6680", + "pull_request:6848", + "pull_request:6962", + "pull_request:7627", + "pull_request:7747", + "pull_request:7910", + "pull_request:8272", + "pull_request:9100" + ], + "soft_pairs": [ + "pull_request:11248|pull_request:11891", + "pull_request:11303|pull_request:11891", + "pull_request:11473|pull_request:12018", + "pull_request:11755|pull_request:11835", + "pull_request:11891|pull_request:12193", + "pull_request:12360|pull_request:12464", + "pull_request:2004|pull_request:2029", + "pull_request:2339|pull_request:3471", + "pull_request:2536|pull_request:652", + "pull_request:2559|pull_request:3226", + "pull_request:2976|pull_request:5015", + "pull_request:3128|pull_request:3214", + "pull_request:3506|pull_request:4248", + "pull_request:3642|pull_request:4457", + "pull_request:4211|pull_request:4797", + "pull_request:4595|pull_request:6311", + "pull_request:4707|pull_request:4934", + "pull_request:5041|pull_request:5707", + "pull_request:5807|pull_request:7627", + "pull_request:5844|pull_request:7910", + "pull_request:6002|pull_request:6389", + "pull_request:6216|pull_request:7747", + "pull_request:6279|pull_request:6848", + "pull_request:6680|pull_request:6962", + "pull_request:8272|pull_request:9100" + ], + "prepared_review_unit_hash": "29e8589d03941c1d746d316a08b557970999a6f909e1f558b441a042334b72c2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 93334, + "estimated_input_tokens": 23334, + "estimated_eval_tokens": 46924 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs typo fixes, test adjustments, and distinct bug fixes across many pipelines and examples.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the items are heterogeneous and the soft edges are only superficial title-level similarities.", + "best_issue_reason": null, + "best_pr_reason": "PR 6848 is the strongest standalone code fix with an explicit issue target and a clear, concrete change, but it still does not represent the rest of the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11248", + "right": "pull_request:11891", + "accept": false, + "reason": "Both are README/doc typo fixes, but they touch different examples and fix different text errors; not the same change." + }, + { + "left": "pull_request:11303", + "right": "pull_request:11891", + "accept": false, + "reason": "Different docs files and different corrections: one fixes an autoencoder class name, the other fixes a method call typo." + }, + { + "left": "pull_request:11473", + "right": "pull_request:12018", + "accept": false, + "reason": "Both are quantization docs edits, but the specific wording/label fixes are unrelated and do not share the same underlying bug." + }, + { + "left": "pull_request:11755", + "right": "pull_request:11835", + "accept": false, + "reason": "One fixes LTX cpu-offload support in a pipeline; the other changes float16 test behavior. Different code paths and purposes." + }, + { + "left": "pull_request:11891", + "right": "pull_request:12193", + "accept": false, + "reason": "A Chroma docs typo fix and a Qwen DreamBooth script name correction are unrelated." + }, + { + "left": "pull_request:12360", + "right": "pull_request:12464", + "accept": false, + "reason": "Both are docs-oriented, but they target different pipeline docs and different issues; one is Chroma docs, the other is Marigold docs/code syntax cleanup." + }, + { + "left": "pull_request:2004", + "right": "pull_request:2029", + "accept": false, + "reason": "DreamBooth README wording changes are unrelated to code-side validation-message typo fixes in pipelines and schedulers." + }, + { + "left": "pull_request:2339", + "right": "pull_request:3471", + "accept": false, + "reason": "Both touch DreamBooth docs/script help text, but they fix different issues: checkpointing behavior vs a torch.compile note." + }, + { + "left": "pull_request:2536", + "right": "pull_request:652", + "accept": false, + "reason": "Enabling VAE gradient checkpointing is a feature/code fix; lowering a test tolerance is a separate testing change." + }, + { + "left": "pull_request:2559", + "right": "pull_request:3226", + "accept": false, + "reason": "Both are docs formatting updates around ControlNet, but they are separate documentation fixes with different targets." + }, + { + "left": "pull_request:2976", + "right": "pull_request:5015", + "accept": false, + "reason": "Both are link-navigation docs edits, but they correct different pages and different broken links." + }, + { + "left": "pull_request:3128", + "right": "pull_request:3214", + "accept": false, + "reason": "Both are bug fixes, but one repairs timestep slicing across img2img pipelines while the other fixes attention reshape logic; not the same underlying bug." + }, + { + "left": "pull_request:3506", + "right": "pull_request:4248", + "accept": false, + "reason": "Checkpoint-resume step scaling and accelerator tracker config handling are distinct training-script bugs." + }, + { + "left": "pull_request:3642", + "right": "pull_request:4457", + "accept": false, + "reason": "One adds a TensorRT inpainting pipeline; the other makes minor fixes to the existing community inpaint example. Same area, different change." + }, + { + "left": "pull_request:4211", + "right": "pull_request:4797", + "accept": false, + "reason": "A docstring/example fix for SDXL inpainting is unrelated to the LoRA loader block-mapping change." + }, + { + "left": "pull_request:4595", + "right": "pull_request:6311", + "accept": false, + "reason": "A WIP conversion project and a new fast-diffusion tutorial are unrelated documentation/features." + }, + { + "left": "pull_request:4707", + "right": "pull_request:4934", + "accept": false, + "reason": "Custom diffusion README training instructions and T2I-Adapter training docs/script are different example features, not one bug." + }, + { + "left": "pull_request:5041", + "right": "pull_request:5707", + "accept": false, + "reason": "Both involve tests, but one adds SDXL refiner tests while the other adds LCM-related tests and loader fixes; different scopes and code paths." + }, + { + "left": "pull_request:5807", + "right": "pull_request:7627", + "accept": false, + "reason": "Community LPW prompt-duplication support and Playground img2img latent normalization are unrelated fixes." + }, + { + "left": "pull_request:5844", + "right": "pull_request:7910", + "accept": false, + "reason": "A custom-pipeline overview example and image-processor API docs are different documentation additions." + }, + { + "left": "pull_request:6002", + "right": "pull_request:6389", + "accept": false, + "reason": "The first changes Stable Video Diffusion guidance-scale behavior; the second only corrects a parameter type annotation/docs text." + }, + { + "left": "pull_request:6216", + "right": "pull_request:7747", + "accept": false, + "reason": "A PEFT-specific LoRA test compatibility fix and a VAE tiling test fix are unrelated." + }, + { + "left": "pull_request:6279", + "right": "pull_request:6848", + "accept": false, + "reason": "Both touch LCM distillation scripts, but they fix different bugs: guidance-scale embeddings vs target U-Net initialization." + }, + { + "left": "pull_request:6680", + "right": "pull_request:6962", + "accept": false, + "reason": "A DPM-related slow-test adjustment and PixArt slow-test fixes are separate test issues." + }, + { + "left": "pull_request:8272", + "right": "pull_request:9100", + "accept": false, + "reason": "These are unrelated typo/grammar fixes in different files and different user-facing messages." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 62, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10318", + "pull_request:10968", + "pull_request:11174", + "pull_request:1136", + "pull_request:11743", + "pull_request:11814", + "pull_request:11913", + "pull_request:12005", + "pull_request:1215", + "pull_request:12611", + "pull_request:12723", + "pull_request:13208", + "pull_request:1788", + "pull_request:2588", + "pull_request:2693", + "pull_request:2836", + "pull_request:3691", + "pull_request:3912", + "pull_request:4099", + "pull_request:4395", + "pull_request:4666", + "pull_request:4749", + "pull_request:4859", + "pull_request:5090", + "pull_request:5407", + "pull_request:5928", + "pull_request:6119", + "pull_request:6215", + "pull_request:6509", + "pull_request:6564", + "pull_request:6645", + "pull_request:6787", + "pull_request:6898", + "pull_request:6909", + "pull_request:7247", + "pull_request:7353", + "pull_request:7738", + "pull_request:7764", + "pull_request:7932", + "pull_request:8660", + "pull_request:8810", + "pull_request:8819", + "pull_request:8864", + "pull_request:9070", + "pull_request:9206", + "pull_request:9411", + "pull_request:9473", + "pull_request:9601" + ], + "soft_pairs": [ + "pull_request:8810|pull_request:9070", + "pull_request:8864|pull_request:9070", + "pull_request:9473|pull_request:9601", + "pull_request:6898|pull_request:6909", + "pull_request:12723|pull_request:13208", + "pull_request:10968|pull_request:11174", + "pull_request:11913|pull_request:12611", + "pull_request:5928|pull_request:6509", + "pull_request:7738|pull_request:9411", + "pull_request:10318|pull_request:11743", + "pull_request:1136|pull_request:2836", + "pull_request:11814|pull_request:12005", + "pull_request:1215|pull_request:1788", + "pull_request:2588|pull_request:3691", + "pull_request:2588|pull_request:3912", + "pull_request:2693|pull_request:4859", + "pull_request:4099|pull_request:4395", + "pull_request:4666|pull_request:4749", + "pull_request:5090|pull_request:6119", + "pull_request:5407|pull_request:6564", + "pull_request:6215|pull_request:7247", + "pull_request:6645|pull_request:6787", + "pull_request:7353|pull_request:8660", + "pull_request:7764|pull_request:8819", + "pull_request:7932|pull_request:9206" + ], + "prepared_review_unit_hash": "c9733e34c0ab650e17a6d59a774d732fc75b62ae22227ada8039f50e0c86e51f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 102117, + "estimated_input_tokens": 25530, + "estimated_eval_tokens": 51316 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the pairs mostly share broad subsystems or file paths, but not the same concrete bug/change. I found no safe duplicate-merge candidates among the soft edges.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:8810", + "right": "pull_request:9070", + "accept": false, + "reason": "Different features in different pipelines: SD ControlNet PAG img2img vs FLUX img2img. Same broad area only." + }, + { + "left": "pull_request:8864", + "right": "pull_request:9070", + "accept": false, + "reason": "Unrelated PRs; one adds SD ControlNet PAG img2img support, the other adds FLUX img2img pipeline code." + }, + { + "left": "pull_request:9473", + "right": "pull_request:9601", + "accept": false, + "reason": "Different fixes in different training scripts and codepaths: Flux DreamBooth LoRA LR handling vs SD3 ControlNet VAE dtype." + }, + { + "left": "pull_request:6898", + "right": "pull_request:6909", + "accept": false, + "reason": "Both standardize model cards, but they touch separate example families and are not the same concrete change." + }, + { + "left": "pull_request:12723", + "right": "pull_request:13208", + "accept": false, + "reason": "Distinct model launches: LongSANA video generation vs Helios video generation. No shared bug or change." + }, + { + "left": "pull_request:10968", + "right": "pull_request:11174", + "accept": false, + "reason": "Both LTX-related, but one is the 0.9.5 release integration and the other is a later pipeline refactor; not the same fix." + }, + { + "left": "pull_request:11913", + "right": "pull_request:12611", + "accept": false, + "reason": "Both concern Wan modular pipelines, but they appear to be separate generation/version efforts rather than one mergeable fix." + }, + { + "left": "pull_request:5928", + "right": "pull_request:6509", + "accept": false, + "reason": "Different community pipelines and README updates; related area, not the same underlying change." + }, + { + "left": "pull_request:7738", + "right": "pull_request:9411", + "accept": false, + "reason": "Both touch LoRA device movement, but one is a test/device guard fix and the other is a DoRA-specific adapter movement fix." + }, + { + "left": "pull_request:10318", + "right": "pull_request:11743", + "accept": false, + "reason": "Tests for Flux LoRA expansion vs a new shortened Flux training script; unrelated changes." + }, + { + "left": "pull_request:1136", + "right": "pull_request:2836", + "accept": false, + "reason": "Both are documentation link fixes, but they fix different pages and different broken links." + }, + { + "left": "pull_request:11814", + "right": "pull_request:12005", + "accept": false, + "reason": "Different core bugs: gradient-checkpointing kwargs plumbing vs sharded checkpoint local_files_only handling." + }, + { + "left": "pull_request:1215", + "right": "pull_request:1788", + "accept": false, + "reason": "Both modify DDIM, but they fix different scheduler bugs with different triggers and code paths." + }, + { + "left": "pull_request:2588", + "right": "pull_request:3691", + "accept": false, + "reason": "LoRA text-encoder support vs selecting the PT2 LoRA processor class; related subsystem, but not the same change." + }, + { + "left": "pull_request:2588", + "right": "pull_request:3912", + "accept": false, + "reason": "One adds text-encoder LoRA support, the other updates training docs/script to finetune the text encoder; not duplicates." + }, + { + "left": "pull_request:2693", + "right": "pull_request:4859", + "accept": false, + "reason": "A doc image-link fix and an SDXL inpaint test adjustment are unrelated." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4395", + "accept": false, + "reason": "Different prompt-embedding and DreamBooth class-prompt bugs in separate codepaths." + }, + { + "left": "pull_request:4666", + "right": "pull_request:4749", + "accept": false, + "reason": "Both SDXL/LoRA-related, but one enables LoRA for ControlNet and the other fixes DreamBooth LoRA checkpointing." + }, + { + "left": "pull_request:5090", + "right": "pull_request:6119", + "accept": false, + "reason": "A T2I-Adapter fix and a LoRA fp16 gradient fix are unrelated." + }, + { + "left": "pull_request:5407", + "right": "pull_request:6564", + "accept": false, + "reason": "Mixed-precision training example fix vs IP-Adapter path handling; no shared underlying bug." + }, + { + "left": "pull_request:6215", + "right": "pull_request:7247", + "accept": false, + "reason": "PixArt test initialization vs single-file config-checking tests; unrelated." + }, + { + "left": "pull_request:6645", + "right": "pull_request:6787", + "accept": false, + "reason": "Community README cleanup vs Marigold input-range normalization fix; unrelated." + }, + { + "left": "pull_request:7353", + "right": "pull_request:8660", + "accept": false, + "reason": "Separate pipeline updates for SDXL ControlNet img2img and SD3 ControlNet; same domain only." + }, + { + "left": "pull_request:7764", + "right": "pull_request:8819", + "accept": false, + "reason": "Both involve AnimateDiff tooling, but conversion-script safetensor loading and single-file loading support are different changes." + }, + { + "left": "pull_request:7932", + "right": "pull_request:9206", + "accept": false, + "reason": "LoRA test expectation updates vs xformers test fixes; no common bug or codepath." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 63, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11522", + "pull_request:12387", + "pull_request:12395", + "pull_request:13209", + "pull_request:13255", + "pull_request:13280", + "pull_request:2037", + "pull_request:2385", + "pull_request:2397", + "pull_request:3245", + "pull_request:325", + "pull_request:3321", + "pull_request:333", + "pull_request:335", + "pull_request:3471", + "pull_request:3589", + "pull_request:365", + "pull_request:3859", + "pull_request:3866", + "pull_request:3952", + "pull_request:3968", + "pull_request:4022", + "pull_request:4058", + "pull_request:4078", + "pull_request:4196", + "pull_request:4223", + "pull_request:4240", + "pull_request:4262", + "pull_request:4272", + "pull_request:4301", + "pull_request:4343", + "pull_request:4397", + "pull_request:4408", + "pull_request:4717", + "pull_request:5057", + "pull_request:5298", + "pull_request:5386", + "pull_request:5671", + "pull_request:5833", + "pull_request:5834", + "pull_request:6078", + "pull_request:6081", + "pull_request:6144", + "pull_request:6236", + "pull_request:6633", + "pull_request:7068", + "pull_request:7451" + ], + "soft_pairs": [ + "pull_request:2385|pull_request:2397", + "pull_request:11522|pull_request:12395", + "pull_request:4408|pull_request:5057", + "pull_request:12387|pull_request:13280", + "pull_request:3859|pull_request:3968", + "pull_request:13209|pull_request:13255", + "pull_request:2037|pull_request:3589", + "pull_request:3245|pull_request:4022", + "pull_request:325|pull_request:365", + "pull_request:3321|pull_request:4301", + "pull_request:333|pull_request:335", + "pull_request:3471|pull_request:5298", + "pull_request:3866|pull_request:4343", + "pull_request:3952|pull_request:5671", + "pull_request:4058|pull_request:4196", + "pull_request:4078|pull_request:5833", + "pull_request:4078|pull_request:5834", + "pull_request:4223|pull_request:4262", + "pull_request:4240|pull_request:4717", + "pull_request:4272|pull_request:5834", + "pull_request:4397|pull_request:6236", + "pull_request:5386|pull_request:7068", + "pull_request:5386|pull_request:7451", + "pull_request:6078|pull_request:6081", + "pull_request:6144|pull_request:6633" + ], + "prepared_review_unit_hash": "04e6f4fbdffae45ebb120621ebde3463324c0ffc12e28757a747675e60211e6f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 98879, + "estimated_input_tokens": 24720, + "estimated_eval_tokens": 49696 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the soft pairs mostly share a file, subsystem, or theme, but they do not describe the same concrete bug/change. I would not merge any of the soft edges as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a canonical duplicate target here; the items span unrelated docs, training scripts, loading fixes, and scheduler/model changes.", + "best_issue_reason": null, + "best_pr_reason": "No PR stands out as a global representative duplicate target, because the cluster does not center on one underlying fix or feature.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2385", + "right": "pull_request:2397", + "accept": false, + "reason": "Same pipeline file, but 2385 adds a utility helper while 2397 adds DDIM inversion support and docs; different changes." + }, + { + "left": "pull_request:11522", + "right": "pull_request:12395", + "accept": false, + "reason": "Completely unrelated PRs: AudioLDM2 transformers compatibility vs workflow changes for prerelease installs." + }, + { + "left": "pull_request:4408", + "right": "pull_request:5057", + "accept": false, + "reason": "Both touch SDXL-related code, but one fixes from_single_file loading and the other adds clip-skip support; different code paths." + }, + { + "left": "pull_request:12387", + "right": "pull_request:13280", + "accept": false, + "reason": "Both concern attention backends, but they add different kernels features (flash_hub vs flash_4_hub); not the same change." + }, + { + "left": "pull_request:3859", + "right": "pull_request:3968", + "accept": false, + "reason": "Large SDXL PRs, but one adds new pipelines and the other improves existing SDXL docs/loaders; not a single duplicate fix." + }, + { + "left": "pull_request:13209", + "right": "pull_request:13255", + "accept": false, + "reason": "Same helper file, but one normalizes ZImage key names and the other handles missing alpha tensors; distinct LoRA conversion bugs." + }, + { + "left": "pull_request:2037", + "right": "pull_request:3589", + "accept": false, + "reason": "Both are community README updates, but they add different pipelines/docs and different reference-trick notes." + }, + { + "left": "pull_request:3245", + "right": "pull_request:4022", + "accept": false, + "reason": "Separate translation updates for different languages, so they are not duplicates." + }, + { + "left": "pull_request:325", + "right": "pull_request:365", + "accept": false, + "reason": "Both are type-hint cleanups, but they affect different models/pipelines and are unrelated changes." + }, + { + "left": "pull_request:3321", + "right": "pull_request:4301", + "accept": false, + "reason": "Different documentation areas: controlling-generation overview vs Kandinsky docs." + }, + { + "left": "pull_request:333", + "right": "pull_request:335", + "accept": false, + "reason": "Both are type-hint PRs, but one targets a pipeline and the other a scheduler; separate changes." + }, + { + "left": "pull_request:3471", + "right": "pull_request:5298", + "accept": false, + "reason": "DreamBooth docs note vs LoRA convolution compile fix; unrelated changes." + }, + { + "left": "pull_request:3866", + "right": "pull_request:4343", + "accept": false, + "reason": "Both add a LoRA rank argument, but to different training scripts and flows; not the same concrete change." + }, + { + "left": "pull_request:3952", + "right": "pull_request:5671", + "accept": false, + "reason": "Unrelated test/workflow changes; no shared underlying bug or feature." + }, + { + "left": "pull_request:4058", + "right": "pull_request:4196", + "accept": false, + "reason": "Both are docs edits around SDXL, but they fix different documentation sections and examples." + }, + { + "left": "pull_request:4078", + "right": "pull_request:5833", + "accept": false, + "reason": "One enforces accelerate for from_single_file, the other fixes import paths for ControlNet single-file loading; different problems." + }, + { + "left": "pull_request:4078", + "right": "pull_request:5834", + "accept": false, + "reason": "Different single-file loading bug: dependency check versus import path correction." + }, + { + "left": "pull_request:4223", + "right": "pull_request:4262", + "accept": false, + "reason": "Both are training-script fixes, but for different scripts and issues; not mergeable as one PR." + }, + { + "left": "pull_request:4240", + "right": "pull_request:4717", + "accept": false, + "reason": "Both are docs fixes, but they touch different pages and different errors." + }, + { + "left": "pull_request:4272", + "right": "pull_request:5834", + "accept": false, + "reason": "Import cleanup in pipeline folders versus a single-file loading import bug; related theme, but different code paths." + }, + { + "left": "pull_request:4397", + "right": "pull_request:6236", + "accept": false, + "reason": "SDXL docstring tweak vs ControlNet inpainting docs example; unrelated documentation fixes." + }, + { + "left": "pull_request:5386", + "right": "pull_request:7068", + "accept": false, + "reason": "Both are typo-fix sweeps, but they are broad, separate PRs across different files and times." + }, + { + "left": "pull_request:5386", + "right": "pull_request:7451", + "accept": false, + "reason": "Another broad typo-fix sweep; similar theme, but not the same underlying change." + }, + { + "left": "pull_request:6078", + "right": "pull_request:6081", + "accept": false, + "reason": "Both mention sigmas, but one fixes Euler Discrete scheduler behavior and the other adjusts SDXL pipeline handling for torch.compile." + }, + { + "left": "pull_request:6144", + "right": "pull_request:6633", + "accept": false, + "reason": "Unrelated fixes: saving LoRA weights from an unwrapped model versus reading checkpoint YAML contents." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 64, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10378", + "pull_request:10510", + "pull_request:10532", + "pull_request:10595", + "pull_request:10717", + "pull_request:10728", + "pull_request:10808", + "pull_request:10951", + "pull_request:11188", + "pull_request:11704", + "pull_request:11896", + "pull_request:12074", + "pull_request:12117", + "pull_request:12170", + "pull_request:12249", + "pull_request:1226", + "pull_request:12416", + "pull_request:12439", + "pull_request:12562", + "pull_request:12634", + "pull_request:13120", + "pull_request:13134", + "pull_request:13229", + "pull_request:13280", + "pull_request:13321", + "pull_request:13410", + "pull_request:1570", + "pull_request:1622", + "pull_request:2009", + "pull_request:2464", + "pull_request:2489", + "pull_request:2921", + "pull_request:3506", + "pull_request:3806", + "pull_request:3852", + "pull_request:3921", + "pull_request:4034", + "pull_request:4099", + "pull_request:4370", + "pull_request:5127", + "pull_request:5823", + "pull_request:5938", + "pull_request:6303", + "pull_request:7227", + "pull_request:9168", + "pull_request:9985", + "pull_request:9988" + ], + "soft_pairs": [ + "pull_request:7227|pull_request:9168", + "pull_request:12439|pull_request:13280", + "pull_request:10717|pull_request:10728", + "pull_request:13120|pull_request:13410", + "pull_request:2464|pull_request:2921", + "pull_request:3506|pull_request:3806", + "pull_request:4034|pull_request:5823", + "pull_request:10951|pull_request:11188", + "pull_request:9985|pull_request:9988", + "pull_request:12074|pull_request:12562", + "pull_request:10510|pull_request:10595", + "pull_request:1226|pull_request:2009", + "pull_request:4099|pull_request:4370", + "pull_request:12117|pull_request:12249", + "pull_request:12416|pull_request:12439", + "pull_request:5127|pull_request:5938", + "pull_request:10532|pull_request:10808", + "pull_request:10378|pull_request:11704", + "pull_request:13134|pull_request:13321", + "pull_request:1570|pull_request:1622", + "pull_request:2489|pull_request:3852", + "pull_request:3921|pull_request:6303", + "pull_request:11896|pull_request:12170", + "pull_request:12634|pull_request:13229" + ], + "prepared_review_unit_hash": "97120932c1ea37dea9ef865ce08ac8309f3884e952b7dc682d232d394090d42e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 101872, + "estimated_input_tokens": 25468, + "estimated_eval_tokens": 51192 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set: it contains many unrelated PRs across training scripts, LoRA loading, Flux, QwenImage, Sana, docs, and attention backends. The soft pairs are mostly false positives from broad subsystem overlap or nearby filenames, not the same underlying bug/change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the cluster spans many unrelated feature additions and bug fixes, so none of the PRs cleanly represents the whole set as a duplicate target.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global representative because the items do not share one concrete code-path problem or one mergeable change set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7227", + "right": "pull_request:9168", + "accept": false, + "reason": "Both touch DreamBooth/LoRA script args, but they are different training scripts and different changes; not the same bug or patch." + }, + { + "left": "pull_request:12439", + "right": "pull_request:13280", + "accept": false, + "reason": "Both extend attention backends, but one adds kernels-based FA2/sage support and the other adds FA4 support; distinct backend changes." + }, + { + "left": "pull_request:10717", + "right": "pull_request:10728", + "accept": false, + "reason": "One adds IP-Adapter support to Flux Img2Img pipelines; the other adjusts quantized Flux IP-Adapter loading. Related area, different code paths and goals." + }, + { + "left": "pull_request:13120", + "right": "pull_request:13410", + "accept": false, + "reason": "One adds tests for robust model loading; the other is an agent/modular doc update. Not the same underlying change." + }, + { + "left": "pull_request:2464", + "right": "pull_request:2921", + "accept": false, + "reason": "Different fixes: training script xformers handling vs CPU offload indentation/behavior. Shared pipeline context only." + }, + { + "left": "pull_request:3506", + "right": "pull_request:3806", + "accept": false, + "reason": "One fixes checkpoint resume step scaling in code; the other adds README guidance for loading from checkpoints. Docs vs code, not duplicates." + }, + { + "left": "pull_request:4034", + "right": "pull_request:5823", + "accept": false, + "reason": "Both are LoRA utilities, but one adds unload_lora_weights and the other adds adapter_names to fuse_lora. Separate features." + }, + { + "left": "pull_request:10951", + "right": "pull_request:11188", + "accept": false, + "reason": "Both are community pipeline additions, but they implement different pipelines and touch the shared README only incidentally." + }, + { + "left": "pull_request:9985", + "right": "pull_request:9988", + "accept": false, + "reason": "Flux family additions overlap, but 9985 is a broader multi-variant release while 9988 is Redux-only; not one concrete change." + }, + { + "left": "pull_request:12074", + "right": "pull_request:12562", + "accept": false, + "reason": "Wan 2.2 LoRA loading support and a context-parallel crash fix are unrelated problems in different layers." + }, + { + "left": "pull_request:10510", + "right": "pull_request:10595", + "accept": false, + "reason": "DC-AE tiling support and bf16 attention dtype handling are separate fixes, even though both involve SANA/DC-AE." + }, + { + "left": "pull_request:1226", + "right": "pull_request:2009", + "accept": false, + "reason": "Padding/truncation fixes in stable diffusion are unrelated to adding textual inversion loader support." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4370", + "accept": false, + "reason": "One fixes prompt embedding handling in SDXL; the other corrects a dummy import path. No shared bug." + }, + { + "left": "pull_request:12117", + "right": "pull_request:12249", + "accept": false, + "reason": "QwenImage pipeline additions and Step1X-Edit pipeline creation are entirely different feature areas." + }, + { + "left": "pull_request:12416", + "right": "pull_request:12439", + "accept": false, + "reason": "Modular QwenImage Edit Plus support is unrelated to attention backend kernel support." + }, + { + "left": "pull_request:5127", + "right": "pull_request:5938", + "accept": false, + "reason": "Multi-adapter image preprocessing in SDXL AdapterPipeline and ControlNet LoRA support are different pipeline features." + }, + { + "left": "pull_request:10532", + "right": "pull_request:10808", + "accept": false, + "reason": "LoRA state-dict parsing and PEFT config-key handling are both loader-related, but they fix different parsing/configuration bugs." + }, + { + "left": "pull_request:10378", + "right": "pull_request:11704", + "accept": false, + "reason": "DreamBooth LoRA DS support and Wan AccVideo LoRA conversion are unrelated training/conversion changes." + }, + { + "left": "pull_request:13134", + "right": "pull_request:13321", + "accept": false, + "reason": "A typing annotation cleanup in Cosmos is unrelated to guarding a torchvision import in a Cosmos pipeline." + }, + { + "left": "pull_request:1570", + "right": "pull_request:1622", + "accept": false, + "reason": "Both touch lpw_stable_diffusion community code, but one is a refactor/compatibility cleanup and the other is a later compatibility fix; not the same concrete patch." + }, + { + "left": "pull_request:2489", + "right": "pull_request:3852", + "accept": false, + "reason": "Core ML docs addition and a contributor note in paradigms docs are unrelated documentation-only changes." + }, + { + "left": "pull_request:3921", + "right": "pull_request:6303", + "accept": false, + "reason": "A blended text inpainting community pipeline and an ADD distillation training script are different features with no shared bug." + }, + { + "left": "pull_request:11896", + "right": "pull_request:12170", + "accept": false, + "reason": "One fixes disabling hooks when loading LoRAs; the other broadens Qwen LoRA conversion support. Same subsystem, different code paths." + }, + { + "left": "pull_request:12634", + "right": "pull_request:13229", + "accept": false, + "reason": "SANA-Video I2V pipeline support and adding an LTX2 VAE are related to SANA-Video, but they address different enhancements and are not mergeable as one fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 65, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10136", + "pull_request:10494", + "pull_request:11243", + "pull_request:11384", + "pull_request:11534", + "pull_request:11588", + "pull_request:11684", + "pull_request:11802", + "pull_request:11843", + "pull_request:11869", + "pull_request:11997", + "pull_request:12425", + "pull_request:12714", + "pull_request:13209", + "pull_request:13255", + "pull_request:13275", + "pull_request:13330", + "pull_request:1623", + "pull_request:1937", + "pull_request:2092", + "pull_request:2994", + "pull_request:3128", + "pull_request:3921", + "pull_request:4481", + "pull_request:4572", + "pull_request:5015", + "pull_request:5090", + "pull_request:5186", + "pull_request:6144", + "pull_request:6215", + "pull_request:6229", + "pull_request:6578", + "pull_request:6961", + "pull_request:7062", + "pull_request:7113", + "pull_request:7116", + "pull_request:7353", + "pull_request:748", + "pull_request:7825", + "pull_request:7938", + "pull_request:826", + "pull_request:8684", + "pull_request:8825", + "pull_request:9257", + "pull_request:9400", + "pull_request:9676", + "pull_request:9687" + ], + "soft_pairs": [ + "pull_request:10494|pull_request:9257", + "pull_request:11243|pull_request:11843", + "pull_request:12425|pull_request:13209", + "pull_request:13275|pull_request:13330", + "pull_request:1623|pull_request:2092", + "pull_request:3921|pull_request:4481", + "pull_request:5015|pull_request:6578", + "pull_request:5090|pull_request:7062", + "pull_request:5186|pull_request:7113", + "pull_request:6144|pull_request:7353", + "pull_request:6215|pull_request:7825", + "pull_request:6229|pull_request:6961", + "pull_request:7116|pull_request:7938", + "pull_request:748|pull_request:826", + "pull_request:7825|pull_request:9400", + "pull_request:8684|pull_request:9676", + "pull_request:8825|pull_request:9687", + "pull_request:11588|pull_request:11997", + "pull_request:10136|pull_request:11534", + "pull_request:11684|pull_request:11869", + "pull_request:10494|pull_request:11843", + "pull_request:11384|pull_request:11802", + "pull_request:12714|pull_request:13255", + "pull_request:1937|pull_request:3128", + "pull_request:2994|pull_request:4572" + ], + "prepared_review_unit_hash": "988b89b1e7fd234098767d64d5d945745bdec764d1a37073423c147e8b41cd0c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 94001, + "estimated_input_tokens": 23501, + "estimated_eval_tokens": 47258 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The candidates are a heterogeneous set of unrelated PRs that only share broad topical similarity (LoRA, CI, docs, or model-specific changes). None of the soft pairs look like the same underlying bug or change, so no duplicates should be merged.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10494", + "right": "pull_request:9257", + "accept": false, + "reason": "Different fixes: tokenizers install in LoRA test workflow vs DreamBooth Flux script logic. No shared code path." + }, + { + "left": "pull_request:11243", + "right": "pull_request:11843", + "accept": false, + "reason": "Musubi Wan LoRA conversion support and Wan VACE exclude_modules tests are distinct changes; one is production conversion code, the other is a test file." + }, + { + "left": "pull_request:12425", + "right": "pull_request:13209", + "accept": false, + "reason": "Torch.distributed import guard is unrelated to ZImage LoRA conversion normalization." + }, + { + "left": "pull_request:13275", + "right": "pull_request:13330", + "accept": false, + "reason": "Hunyuan test kwargs cleanup vs torchao test/workflow fixes are unrelated test maintenance changes." + }, + { + "left": "pull_request:1623", + "right": "pull_request:2092", + "accept": false, + "reason": "Both touch requirements files, but one upgrades transformers and the other removes wandb in a different file and timeframe." + }, + { + "left": "pull_request:3921", + "right": "pull_request:4481", + "accept": false, + "reason": "Community inpainting pipeline work vs EDM pipeline/scheduler implementation are separate features." + }, + { + "left": "pull_request:5015", + "right": "pull_request:6578", + "accept": false, + "reason": "Broken docs link fix and ControlNet heading/indent fix are unrelated documentation edits." + }, + { + "left": "pull_request:5090", + "right": "pull_request:7062", + "accept": false, + "reason": "T2I Adapter accumulation bug and Imagic docstring cleanup do not affect the same code path." + }, + { + "left": "pull_request:5186", + "right": "pull_request:7113", + "accept": false, + "reason": "FreeU proof-of-concept and Docker Slack notification workflow changes are unrelated." + }, + { + "left": "pull_request:6144", + "right": "pull_request:7353", + "accept": false, + "reason": "DreamBooth LoRA checkpoint save fix vs adding FromSingleFileMixin to a ControlNet XL pipeline are unrelated." + }, + { + "left": "pull_request:6215", + "right": "pull_request:7825", + "accept": false, + "reason": "PixArt test seeding/expectation update and model device-map test guards are separate test fixes." + }, + { + "left": "pull_request:6229", + "right": "pull_request:6961", + "accept": false, + "reason": "Fast push test dependency addition vs single-file loader fast tests are different CI/test concerns." + }, + { + "left": "pull_request:7116", + "right": "pull_request:7938", + "accept": false, + "reason": "Switching workflow installs to uv and changing Docker status reporting are unrelated CI changes." + }, + { + "left": "pull_request:748", + "right": "pull_request:826", + "accept": false, + "reason": "Both are DreamBooth-related, but one fixes push_to_hub invocation while the other fixes fp16 loss casting; different bugs." + }, + { + "left": "pull_request:7825", + "right": "pull_request:9400", + "accept": false, + "reason": "Model offload/device-map test guards and single-file pipeline device-agnostic tests are not the same issue." + }, + { + "left": "pull_request:8684", + "right": "pull_request:9676", + "accept": false, + "reason": "ControlNet inpaint docs import fix and EntryNotFoundError import fix are unrelated." + }, + { + "left": "pull_request:8825", + "right": "pull_request:9687", + "accept": false, + "reason": "k-diffusion ControlNet/IP-Adapter support and Xlabs ControlNet support are different model integrations." + }, + { + "left": "pull_request:11588", + "right": "pull_request:11997", + "accept": false, + "reason": "Wan I2V LoRA support for one implementation and modular Wan I2V/FLF2V pipeline refactor are related by subsystem only, not the same concrete change." + }, + { + "left": "pull_request:10136", + "right": "pull_request:11534", + "accept": false, + "reason": "Hunyuan Video core pipeline support and Framepack F1 extension are different features, with different entry points and filenames." + }, + { + "left": "pull_request:11684", + "right": "pull_request:11869", + "accept": false, + "reason": "GGUF single-file loading support and GGUF CUDA kernel acceleration are separate features." + }, + { + "left": "pull_request:10494", + "right": "pull_request:11843", + "accept": false, + "reason": "Tokenizers workflow install fix and Wan VACE exclude_modules tests are unrelated." + }, + { + "left": "pull_request:11384", + "right": "pull_request:11802", + "accept": false, + "reason": "HiDream deprecation version bump and LoRA deprecation cleanup are different code paths and concerns." + }, + { + "left": "pull_request:12714", + "right": "pull_request:13255", + "accept": false, + "reason": "Flux2 LoRA NaN test fix and ZImage non-diffusers LoRA loading fix are unrelated model families and bugs." + }, + { + "left": "pull_request:1937", + "right": "pull_request:3128", + "accept": false, + "reason": "EMA extraction logic and img2img timestep slicing are distinct conversion/pipeline bugs." + }, + { + "left": "pull_request:2994", + "right": "pull_request:4572", + "accept": false, + "reason": "ControlNet training dtype casting and network_alpha->network_alphas rename are separate example-script fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 66, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10205", + "pull_request:10206", + "pull_request:10597", + "pull_request:10718", + "pull_request:10750", + "pull_request:10911", + "pull_request:11113", + "pull_request:11124", + "pull_request:11755", + "pull_request:11786", + "pull_request:11802", + "pull_request:11823", + "pull_request:11835", + "pull_request:11994", + "pull_request:12004", + "pull_request:12041", + "pull_request:12490", + "pull_request:3321", + "pull_request:3334", + "pull_request:3675", + "pull_request:3798", + "pull_request:3924", + "pull_request:4099", + "pull_request:4629", + "pull_request:4659", + "pull_request:4666", + "pull_request:4714", + "pull_request:4744", + "pull_request:4783", + "pull_request:4847", + "pull_request:4859", + "pull_request:5506", + "pull_request:5807", + "pull_request:5979", + "pull_request:6259", + "pull_request:6673", + "pull_request:6764", + "pull_request:6904", + "pull_request:6972", + "pull_request:7507", + "pull_request:7786", + "pull_request:8467", + "pull_request:8685", + "pull_request:8913", + "pull_request:9257", + "pull_request:9473", + "pull_request:9838", + "pull_request:9993" + ], + "soft_pairs": [ + "pull_request:3321|pull_request:3924", + "pull_request:3675|pull_request:4859", + "pull_request:3798|pull_request:4714", + "pull_request:4099|pull_request:4859", + "pull_request:4859|pull_request:6972", + "pull_request:6764|pull_request:8685", + "pull_request:7786|pull_request:8467", + "pull_request:8913|pull_request:9993", + "pull_request:6673|pull_request:7507", + "pull_request:10205|pull_request:10206", + "pull_request:10206|pull_request:9473", + "pull_request:3334|pull_request:4744", + "pull_request:4629|pull_request:4666", + "pull_request:4659|pull_request:4783", + "pull_request:5506|pull_request:5807", + "pull_request:6259|pull_request:6904", + "pull_request:9257|pull_request:9838", + "pull_request:12004|pull_request:12041", + "pull_request:10597|pull_request:10718", + "pull_request:11113|pull_request:9838", + "pull_request:11124|pull_request:11786", + "pull_request:11755|pull_request:11994", + "pull_request:11835|pull_request:12490", + "pull_request:10911|pull_request:11802", + "pull_request:4847|pull_request:5979", + "pull_request:10750|pull_request:11823" + ], + "prepared_review_unit_hash": "011d2910399597b3484c9e1e1276f21b7306a63b32643fbca16644425a9bdd08", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 97094, + "estimated_input_tokens": 24274, + "estimated_eval_tokens": 48804 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No real duplicate cluster here: the soft edges are mostly false positives across unrelated docs, tests, CI, and model-specific fixes.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No safe canonical PR: the items span unrelated subsystems and change types, so any single PR would be an arbitrary representative rather than a true duplicate center.", + "best_issue_reason": null, + "best_pr_reason": "No single PR cleanly represents the cluster; the set is heterogeneous and the apparent similarities are only surface-level.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3321", + "right": "pull_request:3924", + "accept": false, + "reason": "Different docs and different content: controlling-generation overview vs a basic-training code example update." + }, + { + "left": "pull_request:3675", + "right": "pull_request:4859", + "accept": false, + "reason": "Unrelated test cases for different pipelines; no shared bug or code path." + }, + { + "left": "pull_request:3798", + "right": "pull_request:4714", + "accept": false, + "reason": "A training-script comment cleanup is not the same as a missing-transformers import error fix." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4859", + "accept": false, + "reason": "SDXL prompt-embedding logic bug vs an inpaint fast-test fixture change; different underlying issues." + }, + { + "left": "pull_request:4859", + "right": "pull_request:6972", + "accept": false, + "reason": "Both are tests, but they cover different features: inpaint input setup vs IP-Adapter unload behavior." + }, + { + "left": "pull_request:6764", + "right": "pull_request:8685", + "accept": false, + "reason": "Both are docs, but one is about LoRA serialization and the other about float8 support notes." + }, + { + "left": "pull_request:7786", + "right": "pull_request:8467", + "accept": false, + "reason": "Different pipeline bugs: SVD prompt batching vs SEGA device selection." + }, + { + "left": "pull_request:8913", + "right": "pull_request:9993", + "accept": false, + "reason": "Sharded-checkpoint subfolder loading is unrelated to SD3 LoRA checkpoint-saving logic." + }, + { + "left": "pull_request:6673", + "right": "pull_request:7507", + "accept": false, + "reason": "Same family of InstantID pipelines, but these are separate additions (base vs img2img), not one concrete fix." + }, + { + "left": "pull_request:10205", + "right": "pull_request:10206", + "accept": false, + "reason": "Different changes: InstructPix2Pix LoRA support refactor vs Flux Control LoRA unload support." + }, + { + "left": "pull_request:10206", + "right": "pull_request:9473", + "accept": false, + "reason": "Flux LoRA unload support and Flux Dreambooth LR handling are distinct training/runtime fixes." + }, + { + "left": "pull_request:3334", + "right": "pull_request:4744", + "accept": false, + "reason": "Both touch docs/docstrings, but they address different modules and different wording changes." + }, + { + "left": "pull_request:4629", + "right": "pull_request:4666", + "accept": false, + "reason": "Same SDXL ControlNet area, but one adds long-prompt support and the other adds LoRA support; different code-path problems." + }, + { + "left": "pull_request:4659", + "right": "pull_request:4783", + "accept": false, + "reason": "A long-prompt pipeline addition is not the same as an SDXL docs/configuration note about zeroing embeddings." + }, + { + "left": "pull_request:5506", + "right": "pull_request:5807", + "accept": false, + "reason": "PEFT/xformers attention fix vs LPW SDXL prompt batching; different failure modes and fixes." + }, + { + "left": "pull_request:6259", + "right": "pull_request:6904", + "accept": false, + "reason": "A mixed-precision validation fix is unrelated to loading PIL images." + }, + { + "left": "pull_request:9257", + "right": "pull_request:9838", + "accept": false, + "reason": "Both are Flux Dreambooth-related, but they fix different training-script issues and can\u2019t be merged as one concrete fix." + }, + { + "left": "pull_request:12004", + "right": "pull_request:12041", + "accept": false, + "reason": "Wan2.2 model integration is much broader than the VAE patchify bug; not the same underlying change." + }, + { + "left": "pull_request:10597", + "right": "pull_request:10718", + "accept": false, + "reason": "The model-id docs update and the runtime IP-Adapter checkpoint conversion are separate changes." + }, + { + "left": "pull_request:11113", + "right": "pull_request:9838", + "accept": false, + "reason": "AutoencoderKL training/discriminator fix is unrelated to Flux Dreambooth prompt-memory handling." + }, + { + "left": "pull_request:11124", + "right": "pull_request:11786", + "accept": false, + "reason": "CogView4 test fixture update and big-GPU marker CI fix are different concerns." + }, + { + "left": "pull_request:11755", + "right": "pull_request:11994", + "accept": false, + "reason": "LTX cpu-offload test adjustment and huggingface-hub auth/docs fixes are unrelated." + }, + { + "left": "pull_request:11835", + "right": "pull_request:12490", + "accept": false, + "reason": "Two separate test fixes in different models; no shared concrete bug." + }, + { + "left": "pull_request:10911", + "right": "pull_request:11802", + "accept": false, + "reason": "Lumina2 LoRA fuse-nan test and LoRA deprecation cleanup are different issues." + }, + { + "left": "pull_request:4847", + "right": "pull_request:5979", + "accept": false, + "reason": "Both are WIP pipeline additions, but they implement different generative methods and code paths." + }, + { + "left": "pull_request:10750", + "right": "pull_request:11823", + "accept": false, + "reason": "Flux ControlNet img2img support and Flux Kontext LoRA conversion are distinct features, not the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 67, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10109", + "pull_request:10115", + "pull_request:10197", + "pull_request:1020", + "pull_request:10206", + "pull_request:10230", + "pull_request:10246", + "pull_request:10250", + "pull_request:1027", + "pull_request:10273", + "pull_request:10275", + "pull_request:10278", + "pull_request:1031", + "pull_request:10312", + "pull_request:10316", + "pull_request:10318", + "pull_request:10319", + "pull_request:10351", + "pull_request:10378", + "pull_request:10420", + "pull_request:10429", + "pull_request:10494", + "pull_request:10504", + "pull_request:10505", + "pull_request:10632", + "pull_request:10648", + "pull_request:10783", + "pull_request:10854", + "pull_request:10885", + "pull_request:11416", + "pull_request:11439", + "pull_request:1234", + "pull_request:12915", + "pull_request:12983", + "pull_request:2764", + "pull_request:5887", + "pull_request:6946", + "pull_request:7016", + "pull_request:7186", + "pull_request:806", + "pull_request:8583", + "pull_request:8812", + "pull_request:8836", + "pull_request:9005", + "pull_request:9266", + "pull_request:9274", + "pull_request:9423", + "pull_request:9760" + ], + "soft_pairs": [ + "pull_request:5887|pull_request:6946", + "pull_request:8812|pull_request:9423", + "pull_request:12915|pull_request:12983", + "pull_request:7016|pull_request:7186", + "pull_request:10109|pull_request:8583", + "pull_request:10115|pull_request:10250", + "pull_request:10115|pull_request:10429", + "pull_request:10197|pull_request:9005", + "pull_request:10197|pull_request:9274", + "pull_request:1020|pull_request:2764", + "pull_request:10206|pull_request:10318", + "pull_request:10230|pull_request:10378", + "pull_request:10246|pull_request:10648", + "pull_request:10246|pull_request:10885", + "pull_request:10250|pull_request:10783", + "pull_request:10250|pull_request:11439", + "pull_request:1027|pull_request:1031", + "pull_request:1027|pull_request:1234", + "pull_request:1027|pull_request:806", + "pull_request:10273|pull_request:10504", + "pull_request:10273|pull_request:10505", + "pull_request:10275|pull_request:10420", + "pull_request:10278|pull_request:10632", + "pull_request:10278|pull_request:11416", + "pull_request:10312|pull_request:8836", + "pull_request:10316|pull_request:9266", + "pull_request:10316|pull_request:9760", + "pull_request:10319|pull_request:10632", + "pull_request:10319|pull_request:11416", + "pull_request:10351|pull_request:10494", + "pull_request:10351|pull_request:10854" + ], + "prepared_review_unit_hash": "8c020dad013cefbbcaa952980d7af33cbbe0b6be7f61c500374de9d0dc1933a8", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 99863, + "estimated_input_tokens": 24966, + "estimated_eval_tokens": 50188 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicate cluster emerges here: the soft pairs are mostly same-subsystem but distinct docs/CI/features/bugfixes, so all should be rejected.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 7186 is the best representative merged PR here: it has an explicit issue link, touches the core IP-Adapter code path across loaders, embeddings, tests, and docs, and is a coherent feature change rather than a doc/CI cleanup.", + "best_issue_reason": null, + "best_pr_reason": "Same as above; among this mixed set, 7186 is the most substantive and self-contained change with the clearest downstream impact.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5887", + "right": "pull_request:6946", + "accept": false, + "reason": "Both are IP-Adapter related, but one adds ControlNetInpaint support and the other changes loading behavior; different code paths and not mergeable as one fix." + }, + { + "left": "pull_request:8812", + "right": "pull_request:9423", + "accept": false, + "reason": "Different features in the same product area: adding Kolors core vs adding Differential Diffusion to Kolors." + }, + { + "left": "pull_request:12915", + "right": "pull_request:12983", + "accept": false, + "reason": "Both are LTX 2 work, but one adds the pipelines while the other adds single-file loading support; distinct changes that could land separately." + }, + { + "left": "pull_request:7016", + "right": "pull_request:7186", + "accept": false, + "reason": "Related IP-Adapter work, but 7016 refactors embed preparation/loading while 7186 moves Face ID into core and adds model/test/docs changes; not the same concrete change." + }, + { + "left": "pull_request:10109", + "right": "pull_request:8583", + "accept": false, + "reason": "Different problems in different components: Stable Audio XLA support vs SD3 xFormers support." + }, + { + "left": "pull_request:10115", + "right": "pull_request:10250", + "accept": false, + "reason": "Different bugs in different modules: Flux positional embedding vs MidResTemporalBlock1D downsample." + }, + { + "left": "pull_request:10115", + "right": "pull_request:10429", + "accept": false, + "reason": "Same file, but the fixes are unrelated: FluxPosEmbed theta handling vs NPU float-dtype handling." + }, + { + "left": "pull_request:10197", + "right": "pull_request:9005", + "accept": false, + "reason": "Both touch CI runners, but they update different workflow infrastructure and scopes; not the same change." + }, + { + "left": "pull_request:10197", + "right": "pull_request:9274", + "accept": false, + "reason": "Different CI updates: TPU runner changes vs release-test workflow changes." + }, + { + "left": "pull_request:1020", + "right": "pull_request:2764", + "accept": false, + "reason": "Both are docs, but they cover unrelated topics and different documentation surfaces." + }, + { + "left": "pull_request:10206", + "right": "pull_request:10318", + "accept": false, + "reason": "One implements Flux LoRA unload support; the other adds Flux LoRA integration tests for expansion/turbo cases." + }, + { + "left": "pull_request:10230", + "right": "pull_request:10378", + "accept": false, + "reason": "Docs-only mixin list update vs DreamBooth LoRA training support; different artifacts and purposes." + }, + { + "left": "pull_request:10246", + "right": "pull_request:10648", + "accept": false, + "reason": "Both are docs updates, but one is AttnProcessor API coverage and the other is DDPM scheduler docs; unrelated." + }, + { + "left": "pull_request:10246", + "right": "pull_request:10885", + "accept": false, + "reason": "Different documentation additions for different APIs; no shared underlying bug or change." + }, + { + "left": "pull_request:10250", + "right": "pull_request:10783", + "accept": false, + "reason": "Unrelated bug fixes in different code paths: UNet 1D downsample vs fp16 unscaling in SDXL LoRA training." + }, + { + "left": "pull_request:10250", + "right": "pull_request:11439", + "accept": false, + "reason": "Different components and failure modes; only both are validation/runtime fixes." + }, + { + "left": "pull_request:1027", + "right": "pull_request:1031", + "accept": false, + "reason": "Same training example, but one adds a dataloader arg and the other changes the training target; separate changes." + }, + { + "left": "pull_request:1027", + "right": "pull_request:1234", + "accept": false, + "reason": "Both are about the unconditional example, but one is a training-arg tweak and the other adds a sampling script." + }, + { + "left": "pull_request:1027", + "right": "pull_request:806", + "accept": false, + "reason": "Different dreambooth/unconditional training changes with no single concrete bug or patch overlap." + }, + { + "left": "pull_request:10273", + "right": "pull_request:10504", + "accept": false, + "reason": "Both are test cleanup, but one removes null-op import checks while the other removes prints; different edits." + }, + { + "left": "pull_request:10273", + "right": "pull_request:10505", + "accept": false, + "reason": "Same as above: unrelated test hygiene changes, not the same underlying fix." + }, + { + "left": "pull_request:10275", + "right": "pull_request:10420", + "accept": false, + "reason": "Licensing-header cleanup vs LTX checkpoint-id correction; unrelated changes." + }, + { + "left": "pull_request:10278", + "right": "pull_request:10632", + "accept": false, + "reason": "Separate docs fixes for unrelated sections and examples." + }, + { + "left": "pull_request:10278", + "right": "pull_request:11416", + "accept": false, + "reason": "Docs example correction vs broad typo cleanup; different scope and intent." + }, + { + "left": "pull_request:10312", + "right": "pull_request:8836", + "accept": false, + "reason": "Different attention bugs in the same module, but distinct code paths and fixes." + }, + { + "left": "pull_request:10316", + "right": "pull_request:9266", + "accept": false, + "reason": "One changes an LTX video example, the other updates community projects docs; not the same change." + }, + { + "left": "pull_request:10316", + "right": "pull_request:9760", + "accept": false, + "reason": "Different README/docs maintenance with no shared underlying bug." + }, + { + "left": "pull_request:10319", + "right": "pull_request:10632", + "accept": false, + "reason": "Docstring typo fix vs image-path docs fix; unrelated." + }, + { + "left": "pull_request:10319", + "right": "pull_request:11416", + "accept": false, + "reason": "Docstring correction vs broad typo cleanup; not the same artifact." + }, + { + "left": "pull_request:10351", + "right": "pull_request:10494", + "accept": false, + "reason": "Test fixture change vs CI dependency install fix; different issues." + }, + { + "left": "pull_request:10351", + "right": "pull_request:10854", + "accept": false, + "reason": "Unrelated test adjustments in different areas." + }, + { + "left": "pull_request:10429", + "right": "pull_request:10115", + "accept": false, + "reason": "Same file, but distinct fixes for different device-specific behavior and embedding parameters." + }, + { + "left": "pull_request:10504", + "right": "pull_request:10505", + "accept": false, + "reason": "Not a soft-edge duplicate pair in the provided candidates, and in any case these are separate cleanup PRs rather than one concrete bug fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 68, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10364", + "pull_request:10378", + "pull_request:1038", + "pull_request:10386", + "pull_request:10404", + "pull_request:1041", + "pull_request:10429", + "pull_request:10454", + "pull_request:10517", + "pull_request:10524", + "pull_request:10531", + "pull_request:10538", + "pull_request:10539", + "pull_request:10585", + "pull_request:1063", + "pull_request:10632", + "pull_request:10651", + "pull_request:10678", + "pull_request:10756", + "pull_request:1077", + "pull_request:10783", + "pull_request:10784", + "pull_request:10802", + "pull_request:10819", + "pull_request:10849", + "pull_request:10852", + "pull_request:10977", + "pull_request:113", + "pull_request:1137", + "pull_request:11439", + "pull_request:11534", + "pull_request:11693", + "pull_request:11731", + "pull_request:11789", + "pull_request:11847", + "pull_request:11993", + "pull_request:12129", + "pull_request:169", + "pull_request:353", + "pull_request:407", + "pull_request:9107", + "pull_request:9266", + "pull_request:9386", + "pull_request:9513", + "pull_request:9687", + "pull_request:9754", + "pull_request:9760", + "pull_request:9765" + ], + "soft_pairs": [ + "pull_request:10364|pull_request:9386", + "pull_request:10378|pull_request:9687", + "pull_request:1038|pull_request:1137", + "pull_request:10386|pull_request:9107", + "pull_request:10404|pull_request:9266", + "pull_request:10404|pull_request:9760", + "pull_request:1041|pull_request:113", + "pull_request:10429|pull_request:10783", + "pull_request:10429|pull_request:11439", + "pull_request:10454|pull_request:11534", + "pull_request:10517|pull_request:10538", + "pull_request:10524|pull_request:9266", + "pull_request:10531|pull_request:10819", + "pull_request:10531|pull_request:11693", + "pull_request:10531|pull_request:11731", + "pull_request:10539|pull_request:9513", + "pull_request:10585|pull_request:10849", + "pull_request:10585|pull_request:10852", + "pull_request:1063|pull_request:353", + "pull_request:1063|pull_request:407", + "pull_request:10632|pull_request:11847", + "pull_request:10651|pull_request:10678", + "pull_request:10651|pull_request:10784", + "pull_request:10756|pull_request:9754", + "pull_request:1077|pull_request:169", + "pull_request:10802|pull_request:12129", + "pull_request:10802|pull_request:9765", + "pull_request:10819|pull_request:11693", + "pull_request:10819|pull_request:11993", + "pull_request:10977|pull_request:11789" + ], + "prepared_review_unit_hash": "369f1b0034045449a50777185a8f7c8816e4f57ed9ce273d5caf40787a50f794", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 92594, + "estimated_input_tokens": 23149, + "estimated_eval_tokens": 46554 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs updates, pipeline feature additions, training-script fixes, hardware-specific bugfixes, and test changes. There is no single duplicate theme tying these PRs together, so all soft edges should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits this cluster; the items cover unrelated codepaths and documentation areas rather than one underlying change.", + "best_issue_reason": null, + "best_pr_reason": "PR 10756 is the most substantial standalone change, but it is only a weak cluster anchor because the rest of the items are not duplicates of it.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10364", + "right": "pull_request:9386", + "accept": false, + "reason": "Different changes: torch_xla support in Sana vs adding PAG support to SD inpaint; unrelated pipelines and codepaths." + }, + { + "left": "pull_request:10378", + "right": "pull_request:9687", + "accept": false, + "reason": "Different features in different training/controlnet scripts; not the same bug or change." + }, + { + "left": "pull_request:1038", + "right": "pull_request:1137", + "accept": false, + "reason": "Unrelated fixes: Flax dtype handling vs safety-checker dtype in legacy inpaint." + }, + { + "left": "pull_request:10386", + "right": "pull_request:9107", + "accept": false, + "reason": "Different scripts and problems: multi-process data loading vs a transformer loading-hook fix." + }, + { + "left": "pull_request:10404", + "right": "pull_request:9266", + "accept": false, + "reason": "Both are docs-oriented, but one is typo cleanup across pipeline docs and the other is a community projects entry; not the same change." + }, + { + "left": "pull_request:10404", + "right": "pull_request:9760", + "accept": false, + "reason": "Unrelated docs cleanup versus README heading edits; no shared underlying issue." + }, + { + "left": "pull_request:1041", + "right": "pull_request:113", + "accept": false, + "reason": "Tests cleanup versus an old scheduler initialization change; different areas and intents." + }, + { + "left": "pull_request:10429", + "right": "pull_request:10783", + "accept": false, + "reason": "Both are platform-specific fixes, but they affect different subsystems and bug conditions." + }, + { + "left": "pull_request:10429", + "right": "pull_request:11439", + "accept": false, + "reason": "Ascend NPU rotary-embed dtype fix is unrelated to Hi-Dream validation cleanup." + }, + { + "left": "pull_request:10454", + "right": "pull_request:11534", + "accept": false, + "reason": "Attention mask dimension fix in Hunyuan Video is not the same as the Framepack feature/docs PR." + }, + { + "left": "pull_request:10517", + "right": "pull_request:10538", + "accept": false, + "reason": "A bugfix for Sana 4K vs adding Sana to auto pipeline registration; different change types." + }, + { + "left": "pull_request:10524", + "right": "pull_request:9266", + "accept": false, + "reason": "Different documentation targets: HunyuanVideo checkpoint text versus community projects listing." + }, + { + "left": "pull_request:10531", + "right": "pull_request:10819", + "accept": false, + "reason": "Unrelated docs additions: negative-prompt docs for Flux versus LoRA docs entries." + }, + { + "left": "pull_request:10531", + "right": "pull_request:11693", + "accept": false, + "reason": "Different docs topics and codepaths; Flux negative-prompt docs do not duplicate bitsandbytes compile docs." + }, + { + "left": "pull_request:10531", + "right": "pull_request:11731", + "accept": false, + "reason": "Both are docs, but one is Flux pipeline docs and the other is memory/offload guidance; not a duplicate." + }, + { + "left": "pull_request:10539", + "right": "pull_request:9513", + "accept": false, + "reason": "Shared Flux/CFG theme, but one changes core pipeline logic and the other is a community example; not mergeable as one concrete fix." + }, + { + "left": "pull_request:10585", + "right": "pull_request:10849", + "accept": false, + "reason": "Mochi VAE annotation cleanup is unrelated to creating a remote VAE issue template." + }, + { + "left": "pull_request:10585", + "right": "pull_request:10852", + "accept": false, + "reason": "Different files and purposes inside issue templates vs pipeline typing docs." + }, + { + "left": "pull_request:1063", + "right": "pull_request:353", + "accept": false, + "reason": "Both are typo fixes, but in unrelated modules with no shared underlying bug." + }, + { + "left": "pull_request:1063", + "right": "pull_request:407", + "accept": false, + "reason": "Different typo fixes in different attention/pipeline codepaths; not the same change." + }, + { + "left": "pull_request:10632", + "right": "pull_request:11847", + "accept": false, + "reason": "Different docs fixes: image path update in optimization docs vs Wan single-file example corrections." + }, + { + "left": "pull_request:10651", + "right": "pull_request:10678", + "accept": false, + "reason": "Very similar pattern, but they are separate pipeline-specific input-validation fixes; not one duplicate PR." + }, + { + "left": "pull_request:10651", + "right": "pull_request:10784", + "accept": false, + "reason": "Same general height/width validation pattern, but applied to different pipelines and codepaths." + }, + { + "left": "pull_request:10756", + "right": "pull_request:9754", + "accept": false, + "reason": "Quantization backend addition versus a Flux LoRA quantization research example; not the same underlying change." + }, + { + "left": "pull_request:1077", + "right": "pull_request:169", + "accept": false, + "reason": "Scheduler test refactor versus scheduler implementation work; unrelated changes." + }, + { + "left": "pull_request:10802", + "right": "pull_request:12129", + "accept": false, + "reason": "Both are small typo fixes, but in unrelated scripts and subsystems." + }, + { + "left": "pull_request:10802", + "right": "pull_request:9765", + "accept": false, + "reason": "ControlNet training comment typo and SD3 docstring typo are unrelated." + }, + { + "left": "pull_request:10819", + "right": "pull_request:11693", + "accept": false, + "reason": "Different docs areas: LoRA loader docs versus bitsandbytes compilation guidance." + }, + { + "left": "pull_request:10819", + "right": "pull_request:11993", + "accept": false, + "reason": "Both mention LoRA docs, but one adds loader entries and the other adds a hotswap tutorial section; not the same fix." + }, + { + "left": "pull_request:10977", + "right": "pull_request:11789", + "accept": false, + "reason": "Both touch tests, but they validate different behaviors and pipelines; not a duplicate bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 69, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10977", + "pull_request:10978", + "pull_request:11037", + "pull_request:1104", + "pull_request:11053", + "pull_request:11238", + "pull_request:11249", + "pull_request:11262", + "pull_request:11290", + "pull_request:1130", + "pull_request:11325", + "pull_request:11416", + "pull_request:11505", + "pull_request:1157", + "pull_request:11576", + "pull_request:11623", + "pull_request:11739", + "pull_request:1174", + "pull_request:11778", + "pull_request:11782", + "pull_request:11783", + "pull_request:11789", + "pull_request:11838", + "pull_request:11851", + "pull_request:11857", + "pull_request:1187", + "pull_request:11920", + "pull_request:12070", + "pull_request:12183", + "pull_request:1230", + "pull_request:12337", + "pull_request:12360", + "pull_request:12391", + "pull_request:12479", + "pull_request:1341", + "pull_request:2175", + "pull_request:2587", + "pull_request:2624", + "pull_request:2792", + "pull_request:2974", + "pull_request:429", + "pull_request:500", + "pull_request:897", + "pull_request:9721", + "pull_request:9765", + "pull_request:9860", + "pull_request:9993", + "pull_request:9994" + ], + "soft_pairs": [ + "pull_request:10977|pull_request:12070", + "pull_request:10978|pull_request:11857", + "pull_request:11037|pull_request:11920", + "pull_request:1104|pull_request:1230", + "pull_request:1104|pull_request:2792", + "pull_request:11053|pull_request:11920", + "pull_request:11238|pull_request:9993", + "pull_request:11249|pull_request:9860", + "pull_request:11249|pull_request:9994", + "pull_request:11262|pull_request:11778", + "pull_request:11290|pull_request:9721", + "pull_request:1130|pull_request:1341", + "pull_request:1130|pull_request:2974", + "pull_request:1130|pull_request:897", + "pull_request:11325|pull_request:9765", + "pull_request:11416|pull_request:12360", + "pull_request:11505|pull_request:12360", + "pull_request:1157|pull_request:1187", + "pull_request:11576|pull_request:12391", + "pull_request:11623|pull_request:12479", + "pull_request:11739|pull_request:11782", + "pull_request:11739|pull_request:11783", + "pull_request:1174|pull_request:2175", + "pull_request:1174|pull_request:2587", + "pull_request:1174|pull_request:2624", + "pull_request:1174|pull_request:429", + "pull_request:1174|pull_request:500", + "pull_request:11778|pull_request:12337", + "pull_request:11789|pull_request:12070", + "pull_request:11838|pull_request:11857", + "pull_request:11851|pull_request:12183" + ], + "prepared_review_unit_hash": "44ec32e41cd8df31d0eb7a34f320eefddce76dff3a75a46cf6de035bc202e371", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 99520, + "estimated_input_tokens": 24880, + "estimated_eval_tokens": 50016 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs updates, test adjustments, CI workflow tweaks, and separate model/pipeline feature fixes. The soft similarities are superficial (same broad repo areas like docs, tests, MPS, Wan, or LoRA) and do not indicate the same underlying bug or change.", + "confidence": 0.12, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out because the items are not duplicates of one another. If forced to anchor the cluster, PR 11053 is the most substantial merged feature PR, but it does not represent the rest of the items.", + "best_issue_reason": null, + "best_pr_reason": "PR 11053 is the broadest self-contained merged change here (new Wan Video2Video pipeline with docs/tests), so it is the best representative anchor, though not a duplicate target for the other PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10977", + "right": "pull_request:12070", + "accept": false, + "reason": "Different model test files and different fixes: save/load optional components vs Qwen slice updates." + }, + { + "left": "pull_request:10978", + "right": "pull_request:11857", + "accept": false, + "reason": "Flux LoRA loader bugfix vs Flux docs/reference wording; unrelated change types and code paths." + }, + { + "left": "pull_request:11037", + "right": "pull_request:11920", + "accept": false, + "reason": "Different test concerns in different pipelines; peft decorator/nightly config is unrelated to Wan test slices." + }, + { + "left": "pull_request:1104", + "right": "pull_request:1230", + "accept": false, + "reason": "Both touch MPS-related testing, but one adds slow MPS CI coverage while the other fixes generator/device behavior in DDIM/DDPM tests." + }, + { + "left": "pull_request:1104", + "right": "pull_request:2792", + "accept": false, + "reason": "Both are MPS test-related, but one is CI coverage and the other skips specific text-to-video tests; not the same bug." + }, + { + "left": "pull_request:11053", + "right": "pull_request:11920", + "accept": false, + "reason": "Wan Video2Video pipeline addition vs Wan test slice updates; tests only mirror a pipeline family, not the same concrete change." + }, + { + "left": "pull_request:11238", + "right": "pull_request:9993", + "accept": false, + "reason": "SD3 ControlNet validation/runtime fix is unrelated to SD3 DreamBooth LoRA checkpoint saving." + }, + { + "left": "pull_request:11249", + "right": "pull_request:9860", + "accept": false, + "reason": "Both are ControlNet-related, but one adds Xlab ControlNet support in Flux and the other updates SD3.5 ControlNet example support." + }, + { + "left": "pull_request:11249", + "right": "pull_request:9994", + "accept": false, + "reason": "Flux ControlNet support and Omnigen model/pipeline work are unrelated code changes." + }, + { + "left": "pull_request:11262", + "right": "pull_request:11778", + "accept": false, + "reason": "Wan ftfy import guard and Wan VACE mask padding fix are separate issues in different pipeline files." + }, + { + "left": "pull_request:11290", + "right": "pull_request:9721", + "accept": false, + "reason": "Different Dockerfile dependency policy changes in different contexts; one unpins onnx CPU torch deps, the other pins torch versions across several dockerfiles." + }, + { + "left": "pull_request:1130", + "right": "pull_request:1341", + "accept": false, + "reason": "Community pipeline docs/test tweaks vs introducing Stable Diffusion Repaint pipeline; unrelated." + }, + { + "left": "pull_request:1130", + "right": "pull_request:2974", + "accept": false, + "reason": "Different community/example pipeline additions with no shared concrete bug or code path." + }, + { + "left": "pull_request:1130", + "right": "pull_request:897", + "accept": false, + "reason": "Unrelated community example additions." + }, + { + "left": "pull_request:11325", + "right": "pull_request:9765", + "accept": false, + "reason": "Both are docstring fixes, but for different pipelines and different text; not the same underlying issue." + }, + { + "left": "pull_request:11416", + "right": "pull_request:12360", + "accept": false, + "reason": "General docs/typo cleanup across many files is unrelated to Chroma docs corrections." + }, + { + "left": "pull_request:11505", + "right": "pull_request:12360", + "accept": false, + "reason": "Invalid-link docs cleanup and Chroma docs rewrite are separate documentation changes." + }, + { + "left": "pull_request:1157", + "right": "pull_request:1187", + "accept": false, + "reason": "Both are Flax example script edits, but they fix different training scripts and different behavior." + }, + { + "left": "pull_request:11576", + "right": "pull_request:12391", + "accept": false, + "reason": "A docs tip about quantization/offloading is not the same as the quantization docs reorganization itself." + }, + { + "left": "pull_request:11623", + "right": "pull_request:12479", + "accept": false, + "reason": "A pipeline doc typo fix is unrelated to the constants environment-variable logic fix." + }, + { + "left": "pull_request:11739", + "right": "pull_request:11782", + "accept": false, + "reason": "SDXL DreamBooth LoRA alpha/metadata fix is a different script and issue from the SD3 DreamBooth LoRA change." + }, + { + "left": "pull_request:11739", + "right": "pull_request:11783", + "accept": false, + "reason": "Same as above: different training script, not the same bug or change." + }, + { + "left": "pull_request:1174", + "right": "pull_request:2175", + "accept": false, + "reason": "Loading docs rewrite and adding components to API docs are separate documentation tasks." + }, + { + "left": "pull_request:1174", + "right": "pull_request:2587", + "accept": false, + "reason": "Loading docs and translated docs are unrelated." + }, + { + "left": "pull_request:1174", + "right": "pull_request:2624", + "accept": false, + "reason": "Loading docs rewrite vs adding cache_dir docs; different doc fixes." + }, + { + "left": "pull_request:1174", + "right": "pull_request:429", + "accept": false, + "reason": "Different docs pages and scope: loading guide vs configuration API docs." + }, + { + "left": "pull_request:1174", + "right": "pull_request:500", + "accept": false, + "reason": "Loading guide vs subfolder docs in config utilities; not the same underlying change." + }, + { + "left": "pull_request:11778", + "right": "pull_request:12337", + "accept": false, + "reason": "Wan VACE mask padding bugfix and Flux xformers attention fix are unrelated pipeline internals." + }, + { + "left": "pull_request:11789", + "right": "pull_request:12070", + "accept": false, + "reason": "Different pipeline test suites and different test updates; no shared bug." + }, + { + "left": "pull_request:11838", + "right": "pull_request:11857", + "accept": false, + "reason": "Deprecated-pipelines docs and Flux references docs are unrelated documentation changes." + }, + { + "left": "pull_request:11851", + "right": "pull_request:12183", + "accept": false, + "reason": "Scheduler docs note and QwenImage docs badge are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 70, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11857", + "pull_request:1186", + "pull_request:1187", + "pull_request:119", + "pull_request:11920", + "pull_request:1194", + "pull_request:11955", + "pull_request:11994", + "pull_request:12070", + "pull_request:12129", + "pull_request:1221", + "pull_request:12228", + "pull_request:12231", + "pull_request:12272", + "pull_request:1230", + "pull_request:12307", + "pull_request:12337", + "pull_request:12360", + "pull_request:12391", + "pull_request:12418", + "pull_request:1243", + "pull_request:12464", + "pull_request:12487", + "pull_request:12552", + "pull_request:12675", + "pull_request:12731", + "pull_request:12757", + "pull_request:1283", + "pull_request:12883", + "pull_request:12902", + "pull_request:1291", + "pull_request:12971", + "pull_request:1299", + "pull_request:13040", + "pull_request:13076", + "pull_request:1311", + "pull_request:13115", + "pull_request:13135", + "pull_request:13181", + "pull_request:13400", + "pull_request:2009", + "pull_request:2208", + "pull_request:2539", + "pull_request:2604", + "pull_request:2792", + "pull_request:2910", + "pull_request:395", + "pull_request:910" + ], + "soft_pairs": [ + "pull_request:11857|pull_request:12902", + "pull_request:1186|pull_request:1187", + "pull_request:119|pull_request:395", + "pull_request:11920|pull_request:12070", + "pull_request:1194|pull_request:910", + "pull_request:11955|pull_request:12070", + "pull_request:11994|pull_request:12418", + "pull_request:12129|pull_request:12883", + "pull_request:1221|pull_request:2208", + "pull_request:12228|pull_request:12883", + "pull_request:12231|pull_request:12675", + "pull_request:12272|pull_request:12307", + "pull_request:12272|pull_request:12337", + "pull_request:12272|pull_request:13115", + "pull_request:1230|pull_request:2792", + "pull_request:12360|pull_request:13040", + "pull_request:12360|pull_request:13076", + "pull_request:12360|pull_request:13181", + "pull_request:12391|pull_request:13076", + "pull_request:1243|pull_request:2539", + "pull_request:12464|pull_request:12487", + "pull_request:12464|pull_request:12552", + "pull_request:12487|pull_request:12902", + "pull_request:12487|pull_request:13135", + "pull_request:12552|pull_request:12902", + "pull_request:12552|pull_request:13135", + "pull_request:12731|pull_request:12757", + "pull_request:12731|pull_request:12971", + "pull_request:12731|pull_request:13400", + "pull_request:12757|pull_request:12971", + "pull_request:12757|pull_request:13400", + "pull_request:1283|pull_request:2009", + "pull_request:12883|pull_request:13040", + "pull_request:12883|pull_request:13181", + "pull_request:1291|pull_request:2604", + "pull_request:1291|pull_request:2910", + "pull_request:12971|pull_request:13400", + "pull_request:1299|pull_request:1311" + ], + "prepared_review_unit_hash": "f16ea957575541d28f38998dfa1420c336725572fa909553d75491e082897c68", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 38, + "serialized_chars": 105627, + "estimated_input_tokens": 26407, + "estimated_eval_tokens": 53070 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a noisy mix of unrelated docs tweaks, test snapshot updates, and code refactors; none of the soft-edge pairs look like the same underlying change.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items span unrelated fixes across docs, tests, and different codepaths, so there is no duplicate anchor.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good global representative; even the broader changes are about different subsystems and would not merge as one PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11857", + "right": "pull_request:12902", + "accept": false, + "reason": "Flux docs reference cleanup and distributed inference device_map behavior are unrelated changes." + }, + { + "left": "pull_request:1186", + "right": "pull_request:1187", + "accept": false, + "reason": "Both touch the same file, but one adds reporting/push-to-hub plumbing while the other removes copy-pasted setup code." + }, + { + "left": "pull_request:119", + "right": "pull_request:395", + "accept": false, + "reason": "Different variable-name typo fixes in different modules; no shared bug or change." + }, + { + "left": "pull_request:11920", + "right": "pull_request:12070", + "accept": false, + "reason": "Different pipeline test-slice updates for different models (Wan vs QwenImage)." + }, + { + "left": "pull_request:1194", + "right": "pull_request:910", + "accept": false, + "reason": "Documentation workflow updates and inpainting docs are unrelated." + }, + { + "left": "pull_request:11955", + "right": "pull_request:12070", + "accept": false, + "reason": "Cosmos test slices and QwenImage test slices are separate test updates for different pipelines." + }, + { + "left": "pull_request:11994", + "right": "pull_request:12418", + "accept": false, + "reason": "huggingface-hub login docs and CI xfail handling are unrelated fixes." + }, + { + "left": "pull_request:12129", + "right": "pull_request:12883", + "accept": false, + "reason": "NPU attention typo fix and torchao docs typo fix are different issues." + }, + { + "left": "pull_request:1221", + "right": "pull_request:2208", + "accept": false, + "reason": "Weight reuse docs and xFormers installation docs do not address the same change." + }, + { + "left": "pull_request:12228", + "right": "pull_request:12883", + "accept": false, + "reason": "LoRA punctuation/typo fix and torchao typo fix are unrelated docs edits." + }, + { + "left": "pull_request:12231", + "right": "pull_request:12675", + "accept": false, + "reason": "General typo cleanup in community examples and Sana video doc rewording are distinct changes." + }, + { + "left": "pull_request:12272", + "right": "pull_request:12307", + "accept": false, + "reason": "Both are modular-pipeline related, but they fix different Flux pipeline problems and are not mergeable as one change." + }, + { + "left": "pull_request:12272", + "right": "pull_request:12337", + "accept": false, + "reason": "Flux modular pipeline defaults and xFormers attention enabling are different codepaths." + }, + { + "left": "pull_request:12272", + "right": "pull_request:13115", + "accept": false, + "reason": "Shared modular filename only; one changes Flux block defaults, the other exports modular types." + }, + { + "left": "pull_request:1230", + "right": "pull_request:2792", + "accept": false, + "reason": "MPS generator test handling and text-to-video test skipping are different CI/test fixes." + }, + { + "left": "pull_request:12360", + "right": "pull_request:13040", + "accept": false, + "reason": "Chroma docs guidance and fp16_safetensors CLI warning wording are unrelated." + }, + { + "left": "pull_request:12360", + "right": "pull_request:13076", + "accept": false, + "reason": "Chroma docs and torchao quantization syntax are different documentation fixes." + }, + { + "left": "pull_request:12360", + "right": "pull_request:13181", + "accept": false, + "reason": "Chroma docs and torchrun command ordering do not share the same bug." + }, + { + "left": "pull_request:12391", + "right": "pull_request:13076", + "accept": false, + "reason": "Quantization section reorganization and a torchao syntax correction are separate doc changes." + }, + { + "left": "pull_request:1243", + "right": "pull_request:2539", + "accept": false, + "reason": "UNet docstring shape correction and pix2pix-zero default guidance value are unrelated documentation fixes." + }, + { + "left": "pull_request:12464", + "right": "pull_request:12487", + "accept": false, + "reason": "Marigold syntax fixes and broken-link fixes are different doc issues." + }, + { + "left": "pull_request:12464", + "right": "pull_request:12552", + "accept": false, + "reason": "Marigold docs syntax and Kandinsky typo cleanup are unrelated." + }, + { + "left": "pull_request:12487", + "right": "pull_request:12902", + "accept": false, + "reason": "Broken-link cleanup and distributed inference command changes are unrelated docs work." + }, + { + "left": "pull_request:12487", + "right": "pull_request:13135", + "accept": false, + "reason": "Broken-link cleanup and LTX2 docstring fixes are different documentation changes." + }, + { + "left": "pull_request:12552", + "right": "pull_request:12902", + "accept": false, + "reason": "Kandinsky typo fixes and torchrun/device_map docs changes are unrelated." + }, + { + "left": "pull_request:12552", + "right": "pull_request:13135", + "accept": false, + "reason": "Kandinsky docs typos and LTX2 image-to-video docstring corrections are not the same change." + }, + { + "left": "pull_request:12731", + "right": "pull_request:12757", + "accept": false, + "reason": "Bria Fibo wording cleanup and attention backend formatting are separate docs updates." + }, + { + "left": "pull_request:12731", + "right": "pull_request:12971", + "accept": false, + "reason": "Bria Fibo note rewording and distributed inference section reordering are unrelated." + }, + { + "left": "pull_request:12731", + "right": "pull_request:13400", + "accept": false, + "reason": "Bria Fibo docs edits and profiling README acknowledgements are not the same underlying change." + }, + { + "left": "pull_request:12757", + "right": "pull_request:12971", + "accept": false, + "reason": "Attention backend doc formatting and distributed inference doc restructuring are unrelated." + }, + { + "left": "pull_request:12757", + "right": "pull_request:13400", + "accept": false, + "reason": "Attention-backends wording and profiling acknowledgements are separate docs changes." + }, + { + "left": "pull_request:1283", + "right": "pull_request:2009", + "accept": false, + "reason": "Adding Versatile Diffusion and adding textual inversion loading are different feature PRs." + }, + { + "left": "pull_request:12883", + "right": "pull_request:13040", + "accept": false, + "reason": "torchao docs typo fix and fp16_safetensors CLI warning docs are unrelated." + }, + { + "left": "pull_request:12883", + "right": "pull_request:13181", + "accept": false, + "reason": "torchao syntax/typo fix and torchrun argument ordering are different docs issues." + }, + { + "left": "pull_request:1291", + "right": "pull_request:2604", + "accept": false, + "reason": "Stable Diffusion notebook link fix and colab link fix are separate link corrections." + }, + { + "left": "pull_request:1291", + "right": "pull_request:2910", + "accept": false, + "reason": "Stable Diffusion notebook link and Semantic Stable Diffusion link are different targets." + }, + { + "left": "pull_request:12971", + "right": "pull_request:13400", + "accept": false, + "reason": "Distributed inference section repositioning and profiling README additions are unrelated." + }, + { + "left": "pull_request:1299", + "right": "pull_request:1311", + "accept": false, + "reason": "Adding AltDiffusion and adding AltDiffusion tests are related project work but not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 71, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1307", + "pull_request:1311", + "pull_request:13119", + "pull_request:1312", + "pull_request:13209", + "pull_request:13253", + "pull_request:1331", + "pull_request:13371", + "pull_request:1341", + "pull_request:136", + "pull_request:1401", + "pull_request:1412", + "pull_request:1415", + "pull_request:1418", + "pull_request:1423", + "pull_request:1487", + "pull_request:1555", + "pull_request:1561", + "pull_request:1576", + "pull_request:1587", + "pull_request:1588", + "pull_request:1589", + "pull_request:1627", + "pull_request:1640", + "pull_request:1668", + "pull_request:168", + "pull_request:169", + "pull_request:1713", + "pull_request:178", + "pull_request:1989", + "pull_request:2046", + "pull_request:21", + "pull_request:22", + "pull_request:2598", + "pull_request:2897", + "pull_request:2915", + "pull_request:3201", + "pull_request:3265", + "pull_request:3266", + "pull_request:3318", + "pull_request:3400", + "pull_request:3474", + "pull_request:473", + "pull_request:580", + "pull_request:60", + "pull_request:689", + "pull_request:806", + "pull_request:908" + ], + "soft_pairs": [ + "pull_request:1307|pull_request:3201", + "pull_request:1307|pull_request:3265", + "pull_request:1311|pull_request:21", + "pull_request:1311|pull_request:22", + "pull_request:13119|pull_request:13209", + "pull_request:1312|pull_request:2046", + "pull_request:13253|pull_request:13371", + "pull_request:1331|pull_request:473", + "pull_request:1341|pull_request:2897", + "pull_request:136|pull_request:60", + "pull_request:1401|pull_request:1589", + "pull_request:1412|pull_request:3400", + "pull_request:1415|pull_request:168", + "pull_request:1418|pull_request:178", + "pull_request:1423|pull_request:178", + "pull_request:1487|pull_request:3474", + "pull_request:1555|pull_request:3266", + "pull_request:1561|pull_request:908", + "pull_request:1576|pull_request:3318", + "pull_request:1587|pull_request:2598", + "pull_request:1588|pull_request:2915", + "pull_request:1588|pull_request:689", + "pull_request:1627|pull_request:1640", + "pull_request:1668|pull_request:806", + "pull_request:169|pull_request:580", + "pull_request:1713|pull_request:1989" + ], + "prepared_review_unit_hash": "4c303ed4d9ccdb070b65f3cd626f9d7755a4567b8d6b2451ba9e970ca90da864", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 97709, + "estimated_input_tokens": 24428, + "estimated_eval_tokens": 49112 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a duplicate cluster in the strict sense: the items span unrelated PRs across docs, tests, pipelines, dependency bumps, and small bug fixes. The soft links are mostly superficial same-file or same-subsystem overlaps, not the same underlying change.", + "confidence": 0.78, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR cleanly represents a duplicate set here; the cluster is heterogeneous. If one must be chosen, PR 1341 is the most substantive and discussion-heavy Stable Diffusion-related change, but it is not a true duplicate hub.", + "best_issue_reason": null, + "best_pr_reason": "PR 1341 is the closest thing to a representative PR because it is a large, substantive feature implementation with high review/discussion activity, unlike the many narrow test/docs/dependency PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1307", + "right": "pull_request:3201", + "accept": false, + "reason": "Revert of doc-workflow action update vs. community pipeline revert; unrelated changes and code paths." + }, + { + "left": "pull_request:1307", + "right": "pull_request:3265", + "accept": false, + "reason": "Doc workflow revert and lpw_stable_diffusion revert are different fixes in different areas." + }, + { + "left": "pull_request:1311", + "right": "pull_request:21", + "accept": false, + "reason": "AltDiffusion tests vs. generic model tests; same repo, not the same bug/change." + }, + { + "left": "pull_request:1311", + "right": "pull_request:22", + "accept": false, + "reason": "AltDiffusion test additions and GLIDE test additions are separate test work." + }, + { + "left": "pull_request:13119", + "right": "pull_request:13209", + "accept": false, + "reason": "Both touch lora_conversion_utils, but one fixes Flux2 non-diffusers keys and the other ZImage-specific normalization; different concrete problems." + }, + { + "left": "pull_request:1312", + "right": "pull_request:2046", + "accept": false, + "reason": "Both test hub prereleases, but for different versions and setups; not the same change." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13371", + "accept": false, + "reason": "Both are test refactors, but for different model families and unrelated test rewrites." + }, + { + "left": "pull_request:1331", + "right": "pull_request:473", + "accept": false, + "reason": "Both involve inpainting, but one fixes latent scaling/order in the inpaint pipeline and the other targets a different sample/pipeline-utils issue." + }, + { + "left": "pull_request:1341", + "right": "pull_request:2897", + "accept": false, + "reason": "Repaint pipeline implementation vs. Stable UnCLIP docs example; unrelated changes." + }, + { + "left": "pull_request:136", + "right": "pull_request:60", + "accept": false, + "reason": "Manifest packaging fix vs. UNet SDE model change; no shared underlying bug." + }, + { + "left": "pull_request:1401", + "right": "pull_request:1589", + "accept": false, + "reason": "Optional height/width across many pipelines vs. a Paint-by-Example width/height ordering fix; different bugs." + }, + { + "left": "pull_request:1412", + "right": "pull_request:3400", + "accept": false, + "reason": "SD2 inpainting integration tests vs. adding omegaconf test deps; unrelated." + }, + { + "left": "pull_request:1415", + "right": "pull_request:168", + "accept": false, + "reason": "Broad stable diffusion fixes vs. initial stable diffusion pipeline introduction; not the same change." + }, + { + "left": "pull_request:1418", + "right": "pull_request:178", + "accept": false, + "reason": "Versatile Diffusion test fixes vs. initial Stable Diffusion model tests; different subsystems and goals." + }, + { + "left": "pull_request:1423", + "right": "pull_request:178", + "accept": false, + "reason": "Flax SD2 support vs. Stable Diffusion model tests; not the same underlying bug/change." + }, + { + "left": "pull_request:1487", + "right": "pull_request:3474", + "accept": false, + "reason": "Both are docs, but one adds xformers docs for pipelines and the other documents attention processors; different content." + }, + { + "left": "pull_request:1555", + "right": "pull_request:3266", + "accept": false, + "reason": "Both touch dynamic module loading, but cache-dir behavior and version resolution are distinct bugs." + }, + { + "left": "pull_request:1561", + "right": "pull_request:908", + "accept": false, + "reason": "Mega community pipeline config change vs. switching its inpaint pipeline class; different fixes." + }, + { + "left": "pull_request:1576", + "right": "pull_request:3318", + "accept": false, + "reason": "ORT integration into training examples vs. IF textual inversion work; unrelated examples." + }, + { + "left": "pull_request:1587", + "right": "pull_request:2598", + "accept": false, + "reason": "Code fix for Paint-by-Example defaults vs. docs update for the same pipeline; not the same concrete change." + }, + { + "left": "pull_request:1588", + "right": "pull_request:2915", + "accept": false, + "reason": "FP16 common-test fix vs. relaxing a textual inversion slow-test tolerance; different test issues." + }, + { + "left": "pull_request:1588", + "right": "pull_request:689", + "accept": false, + "reason": "FP16 save/load test fix vs. attention performance tweak; unrelated." + }, + { + "left": "pull_request:1627", + "right": "pull_request:1640", + "accept": false, + "reason": "One re-adds xformers enablement APIs, the other stops auto-enabling xformers in scripts; opposite intents." + }, + { + "left": "pull_request:1668", + "right": "pull_request:806", + "accept": false, + "reason": "Dreambooth checkpoint/resume support vs. EMA training options; both training scripts but different features." + }, + { + "left": "pull_request:169", + "right": "pull_request:580", + "accept": false, + "reason": "JAX PNDM scheduler work vs. Flax scheduler tests; same broad area, different concrete changes." + }, + { + "left": "pull_request:1713", + "right": "pull_request:1989", + "accept": false, + "reason": "Conversion-script cleanup vs. safetensors loading support; distinct conversion improvements." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 72, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1727", + "pull_request:1729", + "pull_request:1861", + "pull_request:1920", + "pull_request:1932", + "pull_request:1989", + "pull_request:1996", + "pull_request:2002", + "pull_request:2010", + "pull_request:2077", + "pull_request:2086", + "pull_request:21", + "pull_request:2101", + "pull_request:2118", + "pull_request:2139", + "pull_request:2154", + "pull_request:2197", + "pull_request:22", + "pull_request:2209", + "pull_request:2290", + "pull_request:230", + "pull_request:2330", + "pull_request:2333", + "pull_request:2337", + "pull_request:2362", + "pull_request:2388", + "pull_request:2396", + "pull_request:2490", + "pull_request:2588", + "pull_request:2598", + "pull_request:2839", + "pull_request:296", + "pull_request:3397", + "pull_request:3449", + "pull_request:3458", + "pull_request:3460", + "pull_request:3629", + "pull_request:375", + "pull_request:3852", + "pull_request:3923", + "pull_request:3971", + "pull_request:3983", + "pull_request:406", + "pull_request:4320", + "pull_request:688", + "pull_request:774", + "pull_request:802", + "pull_request:948" + ], + "soft_pairs": [ + "pull_request:1727|pull_request:774", + "pull_request:1729|pull_request:774", + "pull_request:1861|pull_request:2839", + "pull_request:1920|pull_request:3397", + "pull_request:1920|pull_request:3852", + "pull_request:1932|pull_request:230", + "pull_request:1989|pull_request:1996", + "pull_request:2002|pull_request:2588", + "pull_request:2010|pull_request:3983", + "pull_request:2077|pull_request:375", + "pull_request:2086|pull_request:3629", + "pull_request:21|pull_request:296", + "pull_request:2101|pull_request:2598", + "pull_request:2118|pull_request:3923", + "pull_request:2118|pull_request:688", + "pull_request:2139|pull_request:3458", + "pull_request:2139|pull_request:3460", + "pull_request:2154|pull_request:4320", + "pull_request:2197|pull_request:948", + "pull_request:22|pull_request:296", + "pull_request:2209|pull_request:802", + "pull_request:2290|pull_request:406", + "pull_request:2330|pull_request:2333", + "pull_request:2333|pull_request:3971", + "pull_request:2337|pull_request:2396", + "pull_request:2362|pull_request:3449", + "pull_request:2388|pull_request:2490" + ], + "prepared_review_unit_hash": "5651d8d48d87501f38af68c1cfbd9b47a70ae1e884a946439422d590543ede72", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 94857, + "estimated_input_tokens": 23715, + "estimated_eval_tokens": 47686 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of unrelated docs, tests, and code fixes; none of the soft-similarity pairs are the same concrete change, so all soft edges should be rejected.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No true canonical duplicate stands out here; if forced to pick a representative, PR 3983 is the strongest because it fixes a concrete scheduler-step bug across many training scripts and has an explicit issue target.", + "best_issue_reason": null, + "best_pr_reason": "PR 3983 is the best overall representative by scope and specificity: it makes a concrete, cross-cutting training-scheduler fix with an explicit issue link and coordinated script updates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1727", + "right": "pull_request:774", + "accept": false, + "reason": "ONNX img2img preprocessing fix vs Stable Diffusion test tweaks; different code paths and different bug classes." + }, + { + "left": "pull_request:1729", + "right": "pull_request:774", + "accept": false, + "reason": "Latent diffusion test-device gating vs unrelated Stable Diffusion callback test adjustments; not the same change." + }, + { + "left": "pull_request:1861", + "right": "pull_request:2839", + "accept": false, + "reason": "Training loss computation change vs README/example list update; unrelated changes." + }, + { + "left": "pull_request:1920", + "right": "pull_request:3397", + "accept": false, + "reason": "New TRANSLATING docs vs a small note added to LoRA docs; same general area, but not the same underlying change." + }, + { + "left": "pull_request:1920", + "right": "pull_request:3852", + "accept": false, + "reason": "A new docs file vs a contributor note in a pipeline doc; different documentation changes." + }, + { + "left": "pull_request:1932", + "right": "pull_request:230", + "accept": false, + "reason": "DDIM scheduler conversion loading vs generic fp16 loading support; different loading behaviors and code paths." + }, + { + "left": "pull_request:1989", + "right": "pull_request:1996", + "accept": false, + "reason": "One adds safetensors loading, the other adds safetensors saving; related theme but opposite directions and different scripts." + }, + { + "left": "pull_request:2002", + "right": "pull_request:2588", + "accept": false, + "reason": "LoRA support docs/infra vs a research-project LoRA training script; same broad topic but not the same concrete change." + }, + { + "left": "pull_request:2010", + "right": "pull_request:3983", + "accept": false, + "reason": "A single LR scheduler init fix for one script vs a broad multi-script num_processes scheduler fix; not identical enough to dedupe." + }, + { + "left": "pull_request:2077", + "right": "pull_request:375", + "accept": false, + "reason": "Textual inversion validation-image support vs adding Colab links to the README; unrelated docs/example changes." + }, + { + "left": "pull_request:2086", + "right": "pull_request:3629", + "accept": false, + "reason": "LoRA training doc expansion vs A1111 LoRA loading docs; both LoRA-related, but different documentation additions." + }, + { + "left": "pull_request:21", + "right": "pull_request:296", + "accept": false, + "reason": "Early model tests vs auth-token fixes for SD tests; unrelated test/code changes." + }, + { + "left": "pull_request:2101", + "right": "pull_request:2598", + "accept": false, + "reason": "Pix2Pix example doc update vs Paint-by-Example doc fix; different pipelines and different docs." + }, + { + "left": "pull_request:2118", + "right": "pull_request:3923", + "accept": false, + "reason": "Import-utils typo fix vs Stable Diffusion docs import example; unrelated." + }, + { + "left": "pull_request:2118", + "right": "pull_request:688", + "accept": false, + "reason": "Import-utils typo fix vs Flax/PyTorch import handling; different issue and different subsystem behavior." + }, + { + "left": "pull_request:2139", + "right": "pull_request:3458", + "accept": false, + "reason": "LoRA scale docs vs inpaint LoRA scale code support; docs and implementation are not the same change." + }, + { + "left": "pull_request:2139", + "right": "pull_request:3460", + "accept": false, + "reason": "Docs-only LoRA scale note vs inpaint code/tests adding cross_attention_kwargs support; related but not duplicate." + }, + { + "left": "pull_request:2154", + "right": "pull_request:4320", + "accept": false, + "reason": "CLI env typo fix vs ONNX docs wording cleanup; unrelated text fixes in different files." + }, + { + "left": "pull_request:2197", + "right": "pull_request:948", + "accept": false, + "reason": "Inpaint pipeline arg/kwargs fix vs a DDPM comment typo; not the same bug or code path." + }, + { + "left": "pull_request:22", + "right": "pull_request:296", + "accept": false, + "reason": "Modeling-utils logging typo/tests vs auth-token workflow/test fix; unrelated." + }, + { + "left": "pull_request:2209", + "right": "pull_request:802", + "accept": false, + "reason": "Incorrect argument name in img2img docs vs indentation fix in a custom pipeline example; different issues." + }, + { + "left": "pull_request:2290", + "right": "pull_request:406", + "accept": false, + "reason": "Pix2Pix doc formatting fix vs toctree title typo; different documentation corrections." + }, + { + "left": "pull_request:2330", + "right": "pull_request:2333", + "accept": false, + "reason": "Both touch the ethical-guidelines page, but one adds the page and the other only tweaks the intro formatting; not the same underlying change." + }, + { + "left": "pull_request:2333", + "right": "pull_request:3971", + "accept": false, + "reason": "Minor ethical-guidelines text cleanup vs SD XL docs/workflow fix; unrelated." + }, + { + "left": "pull_request:2337", + "right": "pull_request:2396", + "accept": false, + "reason": "Removing unnecessary tests vs fixing UniPC tests/warnings; both test-related but clearly different changes." + }, + { + "left": "pull_request:2362", + "right": "pull_request:3449", + "accept": false, + "reason": "Adding an MPS skip decorator vs reworking MPS/ONNX test jobs; same testing area but not the same concrete fix." + }, + { + "left": "pull_request:2388", + "right": "pull_request:2490", + "accept": false, + "reason": "Initial controlling-generation docs vs adding a MultiDiffusion subsection to that page; same file, but not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 73, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2396", + "pull_request:2398", + "pull_request:2404", + "pull_request:2433", + "pull_request:2446", + "pull_request:2487", + "pull_request:2489", + "pull_request:2539", + "pull_request:2540", + "pull_request:2569", + "pull_request:2598", + "pull_request:2639", + "pull_request:2715", + "pull_request:2757", + "pull_request:2760", + "pull_request:2766", + "pull_request:2814", + "pull_request:2815", + "pull_request:2817", + "pull_request:2897", + "pull_request:2922", + "pull_request:2959", + "pull_request:299", + "pull_request:2994", + "pull_request:3022", + "pull_request:3068", + "pull_request:3165", + "pull_request:3241", + "pull_request:3449", + "pull_request:3474", + "pull_request:3554", + "pull_request:3555", + "pull_request:3800", + "pull_request:3923", + "pull_request:3987", + "pull_request:4058", + "pull_request:408", + "pull_request:420", + "pull_request:4479", + "pull_request:4607", + "pull_request:4859", + "pull_request:5188", + "pull_request:802", + "pull_request:824", + "pull_request:889", + "pull_request:910", + "pull_request:936", + "pull_request:979" + ], + "soft_pairs": [ + "pull_request:2396|pull_request:3241", + "pull_request:2398|pull_request:2489", + "pull_request:2404|pull_request:3800", + "pull_request:2433|pull_request:3474", + "pull_request:2433|pull_request:4607", + "pull_request:2446|pull_request:408", + "pull_request:2446|pull_request:802", + "pull_request:2487|pull_request:824", + "pull_request:2539|pull_request:3923", + "pull_request:2540|pull_request:4058", + "pull_request:2540|pull_request:420", + "pull_request:2569|pull_request:2817", + "pull_request:2598|pull_request:2897", + "pull_request:2598|pull_request:910", + "pull_request:2639|pull_request:3165", + "pull_request:2715|pull_request:4479", + "pull_request:2757|pull_request:3987", + "pull_request:2757|pull_request:889", + "pull_request:2760|pull_request:3068", + "pull_request:2760|pull_request:3554", + "pull_request:2760|pull_request:3555", + "pull_request:2766|pull_request:3449", + "pull_request:2814|pull_request:2815", + "pull_request:2815|pull_request:2922", + "pull_request:2959|pull_request:4859", + "pull_request:299|pull_request:936", + "pull_request:2994|pull_request:5188", + "pull_request:3022|pull_request:979" + ], + "prepared_review_unit_hash": "ad0576c6fcb0cde5887c99be816e231ee387f5549c0d80705ab0641579d7e04a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 94884, + "estimated_input_tokens": 23721, + "estimated_eval_tokens": 47698 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set; it mixes unrelated docs updates, test fixes, CI/workflow edits, and a few substantive code changes. No pair looks like the same concrete bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a credible canonical duplicate representative because the cluster spans unrelated docs, tests, CI, and implementation changes.", + "best_issue_reason": null, + "best_pr_reason": "PR 5188 is the most substantial code-focused change in the set, but it still does not represent the other items well enough to serve as a duplicate canonical.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2396", + "right": "pull_request:3241", + "accept": false, + "reason": "Different test areas and failures: UniPC test warnings vs UnCLIP attention-slicing tolerance." + }, + { + "left": "pull_request:2398", + "right": "pull_request:2489", + "accept": false, + "reason": "Both are docs, but for unrelated pipelines/topics: Pix2Pix Zero usage vs Core ML docs addition." + }, + { + "left": "pull_request:2404", + "right": "pull_request:3800", + "accept": false, + "reason": "Typo fix in AttnProcessor symbol import vs a panorama batching bug; unrelated changes." + }, + { + "left": "pull_request:2433", + "right": "pull_request:3474", + "accept": false, + "reason": "SDEdit note in img2img docs vs a new Attention Processor docs page; different doc topics." + }, + { + "left": "pull_request:2433", + "right": "pull_request:4607", + "accept": false, + "reason": "Img2img docs note vs API docs for PushToHubMixin; no shared underlying change." + }, + { + "left": "pull_request:2446", + "right": "pull_request:408", + "accept": false, + "reason": "Instruct Pix2Pix code snippet fix vs general typo cleanup across docs; not the same change." + }, + { + "left": "pull_request:2446", + "right": "pull_request:802", + "accept": false, + "reason": "Pix2Pix docs snippet correction vs indentation fix in a custom pipeline example; unrelated." + }, + { + "left": "pull_request:2487", + "right": "pull_request:824", + "accept": false, + "reason": "Stable Diffusion JAX/Flax guide vs Flax test/pipeline setup; different artifact types and goals." + }, + { + "left": "pull_request:2539", + "right": "pull_request:3923", + "accept": false, + "reason": "Doc default-value correction in Pix2Pix Zero vs adding a missing torch import in Stable Diffusion docs." + }, + { + "left": "pull_request:2540", + "right": "pull_request:4058", + "accept": false, + "reason": "ControlNet section-doc correction vs SDXL docs layout/content update; unrelated docs fixes." + }, + { + "left": "pull_request:2540", + "right": "pull_request:420", + "accept": false, + "reason": "ControlNet docs correction vs old optimization-section cleanup; no common bug/change." + }, + { + "left": "pull_request:2569", + "right": "pull_request:2817", + "accept": false, + "reason": "OpenVINO documentation expansion vs UNet1D extra_in_channels docs; unrelated features." + }, + { + "left": "pull_request:2598", + "right": "pull_request:2897", + "accept": false, + "reason": "Paint-by-example docs update vs Stable UnCLIP docs and example; different pipelines." + }, + { + "left": "pull_request:2598", + "right": "pull_request:910", + "accept": false, + "reason": "Paint-by-example docs rewrite vs initial in-painting docs update; separate pipelines and content." + }, + { + "left": "pull_request:2639", + "right": "pull_request:3165", + "accept": false, + "reason": "Basic training tutorial tweak vs custom diffusion docs formatting/update; not the same issue." + }, + { + "left": "pull_request:2715", + "right": "pull_request:4479", + "accept": false, + "reason": "Both touch typos, but across different files and concrete text changes; not one shared fix." + }, + { + "left": "pull_request:2757", + "right": "pull_request:3987", + "accept": false, + "reason": "Stable Diffusion depth batching fix vs audio diffusion snippet fix; unrelated code paths." + }, + { + "left": "pull_request:2757", + "right": "pull_request:889", + "accept": false, + "reason": "Depth-map batching bug vs adding DiffusionPipeline components API; unrelated implementation changes." + }, + { + "left": "pull_request:2760", + "right": "pull_request:3068", + "accept": false, + "reason": "Quantization/distillation example addition vs textual inversion docs finishing; different scopes." + }, + { + "left": "pull_request:2760", + "right": "pull_request:3554", + "accept": false, + "reason": "Textual inversion DFQ example vs a broader examples repo change; not the same concrete change." + }, + { + "left": "pull_request:2760", + "right": "pull_request:3555", + "accept": false, + "reason": "Textual inversion DFQ example vs another unrelated Moreh example bundle; no duplicate evidence." + }, + { + "left": "pull_request:2766", + "right": "pull_request:3449", + "accept": false, + "reason": "MPS torch-2.0 test expectation fix vs CI workflow rework for MPS/ONNX; different problems." + }, + { + "left": "pull_request:2814", + "right": "pull_request:2815", + "accept": false, + "reason": "Stable UnCLIP implementation completion vs docs update for Stable UnCLIP; code and docs are not the same fix." + }, + { + "left": "pull_request:2815", + "right": "pull_request:2922", + "accept": false, + "reason": "Stable UnCLIP docs update vs a punctuation tweak in unclip docs; separate edits." + }, + { + "left": "pull_request:2959", + "right": "pull_request:4859", + "accept": false, + "reason": "UNet3D normalization test change vs SDXL inpaint fast-test input tweak; unrelated models/tests." + }, + { + "left": "pull_request:299", + "right": "pull_request:936", + "accept": false, + "reason": "PNDM scheduler hotfix vs splitting Stable Diffusion tests into files; unrelated code vs test refactor." + }, + { + "left": "pull_request:2994", + "right": "pull_request:5188", + "accept": false, + "reason": "ControlNet training type-casting fix vs moving compute_snr into training utils; different training issues and code paths." + }, + { + "left": "pull_request:3022", + "right": "pull_request:979", + "accept": false, + "reason": "Missing AttnProcessor2_0 union entry vs missing README import; unrelated fixes in different layers." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 74, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3030", + "pull_request:3063", + "pull_request:3069", + "pull_request:3099", + "pull_request:3118", + "pull_request:3132", + "pull_request:316", + "pull_request:3166", + "pull_request:3183", + "pull_request:325", + "pull_request:3266", + "pull_request:3331", + "pull_request:335", + "pull_request:3474", + "pull_request:3547", + "pull_request:3554", + "pull_request:3555", + "pull_request:3572", + "pull_request:3577", + "pull_request:3581", + "pull_request:3608", + "pull_request:3710", + "pull_request:3744", + "pull_request:3786", + "pull_request:3820", + "pull_request:3894", + "pull_request:3921", + "pull_request:3923", + "pull_request:3952", + "pull_request:4172", + "pull_request:4612", + "pull_request:4741", + "pull_request:4853", + "pull_request:4965", + "pull_request:4984", + "pull_request:5039", + "pull_request:5065", + "pull_request:5066", + "pull_request:5303", + "pull_request:5448", + "pull_request:5549", + "pull_request:5723", + "pull_request:5745", + "pull_request:5780", + "pull_request:5871", + "pull_request:822", + "pull_request:971" + ], + "soft_pairs": [ + "pull_request:3030|pull_request:3183", + "pull_request:3063|pull_request:3099", + "pull_request:3069|pull_request:822", + "pull_request:3118|pull_request:5066", + "pull_request:3132|pull_request:5065", + "pull_request:3132|pull_request:5066", + "pull_request:316|pull_request:971", + "pull_request:3166|pull_request:4965", + "pull_request:325|pull_request:335", + "pull_request:3266|pull_request:5303", + "pull_request:3331|pull_request:3572", + "pull_request:3474|pull_request:5723", + "pull_request:3547|pull_request:5549", + "pull_request:3554|pull_request:5039", + "pull_request:3555|pull_request:5039", + "pull_request:3572|pull_request:3744", + "pull_request:3577|pull_request:4612", + "pull_request:3581|pull_request:5780", + "pull_request:3608|pull_request:3820", + "pull_request:3608|pull_request:5066", + "pull_request:3710|pull_request:4853", + "pull_request:3786|pull_request:4984", + "pull_request:3894|pull_request:4172", + "pull_request:3921|pull_request:5448", + "pull_request:3923|pull_request:4741", + "pull_request:3923|pull_request:5745", + "pull_request:3952|pull_request:5871" + ], + "prepared_review_unit_hash": "5805706261319717a3b01d69e2c9944e96de973d16d4bdf00875598f98a0118d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 89401, + "estimated_input_tokens": 22351, + "estimated_eval_tokens": 44958 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly composed of small, similarly worded but unrelated fixes (docs typos, type hints, test updates, and one-off code corrections) across different files and subsystems. The soft links are similarity-only and do not look like the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a true duplicate center here; the items span unrelated changes, so there is no good canonical PR to collapse onto.", + "best_issue_reason": null, + "best_pr_reason": "No best PR for deduping: the cluster is heterogeneous, and any single PR would be an arbitrary exemplar rather than the same fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3030", + "right": "pull_request:3183", + "accept": false, + "reason": "Different bugs in different areas: panorama CFG handling vs DreamBooth LoRA checkpoint registration." + }, + { + "left": "pull_request:3063", + "right": "pull_request:3099", + "accept": false, + "reason": "One fixes __setattr__ None handling; the other changes a default argument in Attend-and-Excite." + }, + { + "left": "pull_request:3069", + "right": "pull_request:822", + "accept": false, + "reason": "Flax TPU test expectation update is unrelated to the Flax scheduler API change." + }, + { + "left": "pull_request:3118", + "right": "pull_request:5066", + "accept": false, + "reason": "ControlNet default-argument fix and attention_processor typo fix are unrelated code paths." + }, + { + "left": "pull_request:3132", + "right": "pull_request:5065", + "accept": false, + "reason": "Score SDE type-hint cleanup is unrelated to the LoRA attention processor typo fix." + }, + { + "left": "pull_request:3132", + "right": "pull_request:5066", + "accept": false, + "reason": "Type hints in Score SDE do not match the attention_processor typo fix." + }, + { + "left": "pull_request:316", + "right": "pull_request:971", + "accept": false, + "reason": "README contribution text and community bit diffusion docs are different changes." + }, + { + "left": "pull_request:3166", + "right": "pull_request:4965", + "accept": false, + "reason": "Textual inversion docs update and Wuerstchen docs formatting are separate documentation edits." + }, + { + "left": "pull_request:325", + "right": "pull_request:335", + "accept": false, + "reason": "Both are type-hint PRs, but they target different scheduler implementations and are not the same change." + }, + { + "left": "pull_request:3266", + "right": "pull_request:5303", + "accept": false, + "reason": "Community-pipeline version loading bug and DeepFloyd IF docstring cleanup are unrelated." + }, + { + "left": "pull_request:3331", + "right": "pull_request:3572", + "accept": false, + "reason": "Both touch inpainting docs, but they fix different examples and are not the same underlying change." + }, + { + "left": "pull_request:3474", + "right": "pull_request:5723", + "accept": false, + "reason": "Attention Processor docs and LCM docs are different documentation additions." + }, + { + "left": "pull_request:3547", + "right": "pull_request:5549", + "accept": false, + "reason": "Bibtex reference file creation and Portuguese docs translation are unrelated docs infrastructure changes." + }, + { + "left": "pull_request:3554", + "right": "pull_request:5039", + "accept": false, + "reason": "The Moreh example bundle and SDXL textual inversion docs are unrelated." + }, + { + "left": "pull_request:3555", + "right": "pull_request:5039", + "accept": false, + "reason": "Same as above: unrelated example bundle vs SDXL textual inversion documentation." + }, + { + "left": "pull_request:3572", + "right": "pull_request:3744", + "accept": false, + "reason": "Inpainting docs typo fix and docs-upload token workflow fix are unrelated." + }, + { + "left": "pull_request:3577", + "right": "pull_request:4612", + "accept": false, + "reason": "Both update links, but they affect different docs pages and different link targets." + }, + { + "left": "pull_request:3581", + "right": "pull_request:5780", + "accept": false, + "reason": "bit_diffusion example cleanup and checkpoint_merger prose fix are unrelated community examples." + }, + { + "left": "pull_request:3608", + "right": "pull_request:3820", + "accept": false, + "reason": "Training script typo fix and LDM3D docs/code typo fix are different changes." + }, + { + "left": "pull_request:3608", + "right": "pull_request:5066", + "accept": false, + "reason": "Training script input_perturbation fix is unrelated to attention_processor type checks." + }, + { + "left": "pull_request:3710", + "right": "pull_request:4853", + "accept": false, + "reason": "Adding image_processor docs is not the same as fixing the image_processor width bug." + }, + { + "left": "pull_request:3786", + "right": "pull_request:4984", + "accept": false, + "reason": "Initial PR template creation and later template text correction are different PRs, not duplicates." + }, + { + "left": "pull_request:3894", + "right": "pull_request:4172", + "accept": false, + "reason": "Flax UNet dropout addition and FlaxModelMixin casting logic are separate Flax changes." + }, + { + "left": "pull_request:3921", + "right": "pull_request:5448", + "accept": false, + "reason": "Blended latent diffusion community pipeline work is unrelated to adding LCM." + }, + { + "left": "pull_request:3923", + "right": "pull_request:4741", + "accept": false, + "reason": "Docs import fix and AudioLDM2 dummy import fix address different problems." + }, + { + "left": "pull_request:3923", + "right": "pull_request:5745", + "accept": false, + "reason": "Docs import fix and ConsistencyDecoder doc type fix are unrelated." + }, + { + "left": "pull_request:3952", + "right": "pull_request:5871", + "accept": false, + "reason": "Dependency pin test PR and deprecated scheduler rebase are unrelated changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 75, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3983", + "pull_request:3985", + "pull_request:3989", + "pull_request:4058", + "pull_request:406", + "pull_request:4082", + "pull_request:421", + "pull_request:4211", + "pull_request:4267", + "pull_request:4275", + "pull_request:4320", + "pull_request:4330", + "pull_request:4350", + "pull_request:4351", + "pull_request:4397", + "pull_request:4479", + "pull_request:4495", + "pull_request:4546", + "pull_request:4570", + "pull_request:4607", + "pull_request:4626", + "pull_request:4632", + "pull_request:4644", + "pull_request:4707", + "pull_request:4746", + "pull_request:4783", + "pull_request:4794", + "pull_request:4797", + "pull_request:4952", + "pull_request:5051", + "pull_request:5063", + "pull_request:5407", + "pull_request:5454", + "pull_request:5745", + "pull_request:5818", + "pull_request:5852", + "pull_request:5961", + "pull_request:5982", + "pull_request:6143", + "pull_request:6209", + "pull_request:6343", + "pull_request:6344", + "pull_request:6438", + "pull_request:6440", + "pull_request:6475", + "pull_request:6596", + "pull_request:6620" + ], + "soft_pairs": [ + "pull_request:3983|pull_request:6143", + "pull_request:3985|pull_request:4644", + "pull_request:3989|pull_request:4546", + "pull_request:3989|pull_request:5852", + "pull_request:4058|pull_request:4267", + "pull_request:406|pull_request:421", + "pull_request:4082|pull_request:4211", + "pull_request:4082|pull_request:4797", + "pull_request:4211|pull_request:4267", + "pull_request:4211|pull_request:4746", + "pull_request:4211|pull_request:4952", + "pull_request:4211|pull_request:5063", + "pull_request:4275|pull_request:4794", + "pull_request:4275|pull_request:6440", + "pull_request:4275|pull_request:6620", + "pull_request:4320|pull_request:5961", + "pull_request:4330|pull_request:6596", + "pull_request:4350|pull_request:5745", + "pull_request:4350|pull_request:6475", + "pull_request:4351|pull_request:5454", + "pull_request:4351|pull_request:6438", + "pull_request:4397|pull_request:4783", + "pull_request:4479|pull_request:5051", + "pull_request:4495|pull_request:5982", + "pull_request:4570|pull_request:5818", + "pull_request:4607|pull_request:6343", + "pull_request:4626|pull_request:6344", + "pull_request:4632|pull_request:5407", + "pull_request:4707|pull_request:6209" + ], + "prepared_review_unit_hash": "de4f9d93338ed05355d73ff9e1dc1335b5dee27b310d61f97a6856f910397756", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 93865, + "estimated_input_tokens": 23467, + "estimated_eval_tokens": 47190 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a mixed cluster of small docs/typo/example PRs around Diffusers, with a few substantive training/pipeline fixes. The items are generally too broad or too different in code path to treat as duplicates; no soft-edge pair looks safely mergeable as one underlying change.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 6143 is the strongest representative of the cluster\u2019s substantive bug-fix theme: it addresses an actual training-step counting error in the text-to-image LoRA/SDXL scripts and has an explicit issue target, unlike the many docs/typo-only PRs.", + "best_issue_reason": null, + "best_pr_reason": "PR 6143 is the best overall representative because it fixes a concrete functional training bug rather than a documentation nit, and it is more central to the cluster than the scattered typo/doc updates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3983", + "right": "pull_request:6143", + "accept": false, + "reason": "Both concern training-step math, but they fix different formulas in different example scripts and are not the same concrete bug or code path." + }, + { + "left": "pull_request:3985", + "right": "pull_request:4644", + "accept": false, + "reason": "Both are SDXL docs edits, but they touch different pages and different content; this is not one underlying change." + }, + { + "left": "pull_request:3989", + "right": "pull_request:4546", + "accept": false, + "reason": "Unrelated changes: SDXL CPU offload logic vs a typo fix in a ControlNet conversion script." + }, + { + "left": "pull_request:3989", + "right": "pull_request:5852", + "accept": false, + "reason": "Different code paths and purposes: SDXL pipeline offload logic vs CI workflow variable typo." + }, + { + "left": "pull_request:4058", + "right": "pull_request:4267", + "accept": false, + "reason": "Same documentation file, but one updates prose/table formatting while the other fixes image links; separate edits." + }, + { + "left": "pull_request:406", + "right": "pull_request:421", + "accept": false, + "reason": "Both are docs fixes, but they affect unrelated files and unrelated typos." + }, + { + "left": "pull_request:4082", + "right": "pull_request:4211", + "accept": false, + "reason": "Requirements pin fix vs SDXL inpainting docstring fix; not the same bug or change." + }, + { + "left": "pull_request:4082", + "right": "pull_request:4797", + "accept": false, + "reason": "Different scopes entirely: dependency version correction vs LoRA block-mapping logic." + }, + { + "left": "pull_request:4211", + "right": "pull_request:4267", + "accept": false, + "reason": "Both touch SDXL docs, but one is a docstring/example issue and the other is an image-link fix." + }, + { + "left": "pull_request:4211", + "right": "pull_request:4746", + "accept": false, + "reason": "Docstring cleanup vs a large WIP SDXL training/example implementation; not the same underlying fix." + }, + { + "left": "pull_request:4211", + "right": "pull_request:4952", + "accept": false, + "reason": "Different targets: SDXL inpainting docstring vs generic pipeline offload docstring." + }, + { + "left": "pull_request:4211", + "right": "pull_request:5063", + "accept": false, + "reason": "Both are docstring/docs changes, but they concern different parameters and different pipeline families." + }, + { + "left": "pull_request:4275", + "right": "pull_request:4794", + "accept": false, + "reason": "AutoPipeline docs vs compel syntax docs; unrelated documentation fixes." + }, + { + "left": "pull_request:4275", + "right": "pull_request:6440", + "accept": false, + "reason": "Different docs areas and issues: AutoPipeline example vs local-link cleanup." + }, + { + "left": "pull_request:4275", + "right": "pull_request:6620", + "accept": false, + "reason": "AutoPipeline docs vs SDXL LoRA converter usage docs; not the same change." + }, + { + "left": "pull_request:4320", + "right": "pull_request:5961", + "accept": false, + "reason": "Different documentation pages and different typos; no shared code path." + }, + { + "left": "pull_request:4330", + "right": "pull_request:6596", + "accept": false, + "reason": "Both adjust headings in markdown, but they are unrelated files and unrelated content." + }, + { + "left": "pull_request:4350", + "right": "pull_request:5745", + "accept": false, + "reason": "Both are type/example doc fixes, but they concern different components and different errors." + }, + { + "left": "pull_request:4350", + "right": "pull_request:6475", + "accept": false, + "reason": "Different model/pipeline docs with different type annotations; not one bug." + }, + { + "left": "pull_request:4351", + "right": "pull_request:5454", + "accept": false, + "reason": "README image path fix vs scheduler docs link fix; unrelated." + }, + { + "left": "pull_request:4351", + "right": "pull_request:6438", + "accept": false, + "reason": "README asset-path tweak vs installation instructions update; different documentation tasks." + }, + { + "left": "pull_request:4397", + "right": "pull_request:4783", + "accept": false, + "reason": "Both mention SDXL defaults/docs, but they change different parameters in different docs and pipeline files." + }, + { + "left": "pull_request:4479", + "right": "pull_request:5051", + "accept": false, + "reason": "General typo cleanup across pipelines vs Wuerstchen doc typos; not the same concrete issue." + }, + { + "left": "pull_request:4495", + "right": "pull_request:5982", + "accept": false, + "reason": "Completely different scopes: a new SDXL reference example vs adding SDXL Turbo docs." + }, + { + "left": "pull_request:4570", + "right": "pull_request:5818", + "accept": false, + "reason": "AnimateDiff implementation work vs CI workflow restructuring; unrelated changes." + }, + { + "left": "pull_request:4607", + "right": "pull_request:6343", + "accept": false, + "reason": "Docs overview additions for PushToHubMixin vs Amused pipeline docs; different features and files." + }, + { + "left": "pull_request:4626", + "right": "pull_request:6344", + "accept": false, + "reason": "Different documentation sets for different models; same broad doc maintenance theme is not enough." + }, + { + "left": "pull_request:4632", + "right": "pull_request:5407", + "accept": false, + "reason": "Different training examples and different bugs: SDXL LoRA checkpoint/casting vs unconditional mixed-precision handling." + }, + { + "left": "pull_request:4707", + "right": "pull_request:6209", + "accept": false, + "reason": "Custom diffusion README training example changes vs text-to-image README test-script fix; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 76, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4717", + "pull_request:4721", + "pull_request:4722", + "pull_request:4766", + "pull_request:4794", + "pull_request:4822", + "pull_request:4853", + "pull_request:4856", + "pull_request:4999", + "pull_request:5041", + "pull_request:5063", + "pull_request:5065", + "pull_request:517", + "pull_request:5236", + "pull_request:5267", + "pull_request:5298", + "pull_request:5386", + "pull_request:5411", + "pull_request:5639", + "pull_request:568", + "pull_request:5727", + "pull_request:580", + "pull_request:5807", + "pull_request:5823", + "pull_request:5828", + "pull_request:5831", + "pull_request:5961", + "pull_request:5982", + "pull_request:5983", + "pull_request:5991", + "pull_request:6209", + "pull_request:6229", + "pull_request:6308", + "pull_request:6314", + "pull_request:6339", + "pull_request:6340", + "pull_request:6462", + "pull_request:6604", + "pull_request:6620", + "pull_request:6998", + "pull_request:7181", + "pull_request:7495", + "pull_request:7747", + "pull_request:7919", + "pull_request:822", + "pull_request:885", + "pull_request:978" + ], + "soft_pairs": [ + "pull_request:4717|pull_request:5831", + "pull_request:4721|pull_request:4794", + "pull_request:4721|pull_request:6620", + "pull_request:4721|pull_request:6998", + "pull_request:4722|pull_request:4766", + "pull_request:4794|pull_request:4999", + "pull_request:4822|pull_request:5961", + "pull_request:4853|pull_request:5065", + "pull_request:4856|pull_request:6314", + "pull_request:4856|pull_request:6462", + "pull_request:5041|pull_request:5982", + "pull_request:5063|pull_request:5961", + "pull_request:517|pull_request:885", + "pull_request:5236|pull_request:5983", + "pull_request:5267|pull_request:6209", + "pull_request:5267|pull_request:7495", + "pull_request:5298|pull_request:5991", + "pull_request:5386|pull_request:7181", + "pull_request:5411|pull_request:6998", + "pull_request:5639|pull_request:6339", + "pull_request:568|pull_request:978", + "pull_request:5727|pull_request:6340", + "pull_request:580|pull_request:822", + "pull_request:5807|pull_request:7919", + "pull_request:5823|pull_request:6229", + "pull_request:5823|pull_request:6308", + "pull_request:5828|pull_request:7747", + "pull_request:5831|pull_request:6604" + ], + "prepared_review_unit_hash": "a216075412015095839bacd8f18627ec97aa5d8966cde878b01da9f517f1d66c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 95930, + "estimated_input_tokens": 23983, + "estimated_eval_tokens": 48222 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly unrelated docs/typo PRs plus a few distinct code fixes; none of the soft pairs look like the same underlying bug/change, so I reject them all.", + "confidence": 0.31, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 5991 is the best representative because it fixes a concrete LoRA serialization/load bug in code, whereas most other items are minor docs or typo edits.", + "best_issue_reason": null, + "best_pr_reason": "PR 5991 is the strongest standalone PR in the set: merged, code-level, and clearly addresses a real behavior bug rather than a superficial doc change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4717", + "right": "pull_request:5831", + "accept": false, + "reason": "Both are doc TOC/callout edits, but they touch different documentation topics and fix different formatting issues." + }, + { + "left": "pull_request:4721", + "right": "pull_request:4794", + "accept": false, + "reason": "One is a Korean DreamBooth doc reflow; the other is a weighted-prompts syntax example. Same broad docs area, but not the same change." + }, + { + "left": "pull_request:4721", + "right": "pull_request:6620", + "accept": false, + "reason": "Different docs pages and different fixes; only superficial similarity as typo/doc cleanup." + }, + { + "left": "pull_request:4721", + "right": "pull_request:6998", + "accept": false, + "reason": "Unrelated documentation edits in different guides; no shared underlying bug or change." + }, + { + "left": "pull_request:4722", + "right": "pull_request:4766", + "accept": false, + "reason": "Both add pipeline guides to docs, but they are for different pipelines and different content." + }, + { + "left": "pull_request:4794", + "right": "pull_request:4999", + "accept": false, + "reason": "Both mention compel, but 4794 fixes a docs syntax example while 4999 fixes Wuerstchen compel usage in code/docs; not the same code path." + }, + { + "left": "pull_request:4822", + "right": "pull_request:5961", + "accept": false, + "reason": "Different typo fixes in unrelated strings/docstrings; no same bug or code path." + }, + { + "left": "pull_request:4853", + "right": "pull_request:5065", + "accept": false, + "reason": "Image processor width bug and guidance_rescale docstring default are unrelated." + }, + { + "left": "pull_request:4856", + "right": "pull_request:6314", + "accept": false, + "reason": "These are separate hyperlink corrections in different docs." + }, + { + "left": "pull_request:4856", + "right": "pull_request:6462", + "accept": false, + "reason": "Separate paper-link fixes in different pages; no shared underlying issue." + }, + { + "left": "pull_request:5041", + "right": "pull_request:5982", + "accept": false, + "reason": "SDXL refiner-only tests and SDXL Turbo docs are unrelated changes." + }, + { + "left": "pull_request:5063", + "right": "pull_request:5961", + "accept": false, + "reason": "Different docstring default-value corrections in different pipelines; not the same fix." + }, + { + "left": "pull_request:517", + "right": "pull_request:885", + "accept": false, + "reason": "Workflow test-report plumbing and a Flax scheduler implementation change are unrelated." + }, + { + "left": "pull_request:5236", + "right": "pull_request:5983", + "accept": false, + "reason": "Both are doc formatting fixes, but they affect different guides and different content." + }, + { + "left": "pull_request:5267", + "right": "pull_request:6209", + "accept": false, + "reason": "Both touch README docs, but one is markdown syntax guidance and the other is an example inference script fix." + }, + { + "left": "pull_request:5267", + "right": "pull_request:7495", + "accept": false, + "reason": "Unrelated README/link cleanup PRs; no shared change." + }, + { + "left": "pull_request:5298", + "right": "pull_request:5991", + "accept": false, + "reason": "Both LoRA-related, but 5298 fixes torch.compile conv forwarding while 5991 fixes LoRA prefix handling during serialization/loading." + }, + { + "left": "pull_request:5386", + "right": "pull_request:7181", + "accept": false, + "reason": "Both are typo cleanups, but they are broad, unrelated sweeps rather than the same underlying issue." + }, + { + "left": "pull_request:5411", + "right": "pull_request:6998", + "accept": false, + "reason": "Broken training-doc link and callout markup fix are unrelated documentation changes." + }, + { + "left": "pull_request:5639", + "right": "pull_request:6339", + "accept": false, + "reason": "Both are AnimateDiff docs, but one adds Motion LoRAs guidance and the other fixes example text/model references; not the same concrete change." + }, + { + "left": "pull_request:568", + "right": "pull_request:978", + "accept": false, + "reason": "Both are typo fixes, but they are separate broad cleanup PRs across different files and scopes." + }, + { + "left": "pull_request:5727", + "right": "pull_request:6340", + "accept": false, + "reason": "Both add conversion/example scripts, but for different models and different conversion tasks." + }, + { + "left": "pull_request:580", + "right": "pull_request:822", + "accept": false, + "reason": "Different Flax scheduler changes and different affected code paths; not mergeable as one fix." + }, + { + "left": "pull_request:5807", + "right": "pull_request:7919", + "accept": false, + "reason": "One adds num_images_per_prompt support to a community pipeline; the other fixes offload behavior with device maps. Different bugs." + }, + { + "left": "pull_request:5823", + "right": "pull_request:6229", + "accept": false, + "reason": "Feature work for fuse_lora(adapter_names) versus adding a test dependency; clearly unrelated." + }, + { + "left": "pull_request:5823", + "right": "pull_request:6308", + "accept": false, + "reason": "A LoRA API feature and LoRA test-fixture cleanup are not the same underlying change." + }, + { + "left": "pull_request:5828", + "right": "pull_request:7747", + "accept": false, + "reason": "These are separate test fixes for different VAE behaviors; no shared bug." + }, + { + "left": "pull_request:5831", + "right": "pull_request:6604", + "accept": false, + "reason": "ToC title cleanup and API autodoc exposure are unrelated documentation changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 77, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5838", + "pull_request:5855", + "pull_request:5901", + "pull_request:5911", + "pull_request:5961", + "pull_request:5982", + "pull_request:5986", + "pull_request:60", + "pull_request:6081", + "pull_request:61", + "pull_request:6106", + "pull_request:6122", + "pull_request:6144", + "pull_request:6147", + "pull_request:6163", + "pull_request:6209", + "pull_request:6221", + "pull_request:6229", + "pull_request:6276", + "pull_request:6291", + "pull_request:63", + "pull_request:6308", + "pull_request:6311", + "pull_request:6332", + "pull_request:6339", + "pull_request:6403", + "pull_request:6620", + "pull_request:6760", + "pull_request:6952", + "pull_request:6961", + "pull_request:6965", + "pull_request:7015", + "pull_request:7062", + "pull_request:7181", + "pull_request:7260", + "pull_request:7301", + "pull_request:7472", + "pull_request:7495", + "pull_request:7697", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7857", + "pull_request:7913", + "pull_request:7977", + "pull_request:8426", + "pull_request:89", + "pull_request:98" + ], + "soft_pairs": [ + "pull_request:5838|pull_request:7697", + "pull_request:5855|pull_request:6209", + "pull_request:5855|pull_request:7495", + "pull_request:5901|pull_request:6760", + "pull_request:5901|pull_request:7260", + "pull_request:5911|pull_request:6403", + "pull_request:5961|pull_request:7977", + "pull_request:5982|pull_request:7857", + "pull_request:5986|pull_request:7472", + "pull_request:60|pull_request:89", + "pull_request:6081|pull_request:7913", + "pull_request:61|pull_request:98", + "pull_request:6106|pull_request:7015", + "pull_request:6122|pull_request:7181", + "pull_request:6144|pull_request:6952", + "pull_request:6147|pull_request:6961", + "pull_request:6163|pull_request:6221", + "pull_request:6229|pull_request:6311", + "pull_request:6229|pull_request:7797", + "pull_request:6229|pull_request:7803", + "pull_request:6229|pull_request:7804", + "pull_request:6276|pull_request:8426", + "pull_request:6291|pull_request:7301", + "pull_request:63|pull_request:98", + "pull_request:6308|pull_request:6965", + "pull_request:6332|pull_request:7062", + "pull_request:6339|pull_request:6620" + ], + "prepared_review_unit_hash": "2600ff50487d3f6145c149ceda37d8c7a7fdaf9c61caee528321c95b8298a413", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 95272, + "estimated_input_tokens": 23818, + "estimated_eval_tokens": 47892 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly unrelated docs, tests, workflow, and feature PRs; none of the soft pairs look like the same underlying change closely enough to merge as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 5911 is the best representative PR here: it has an explicit issue target, substantive code/docs/test changes, and a clear feature scope (IPAdapterFull).", + "best_issue_reason": null, + "best_pr_reason": "PR 5911 is the strongest standalone PR in the set because it is feature-complete, well-scoped, and tied to an explicit issue, unlike the mostly typo/test/workflow-only PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5838", + "right": "pull_request:7697", + "accept": false, + "reason": "Both touch dependency setup, but they change different workflows and solve different CI environments; not the same bug/change." + }, + { + "left": "pull_request:5855", + "right": "pull_request:6209", + "accept": false, + "reason": "Different README files and different fixes; one updates community docs, the other fixes a text-to-image example script." + }, + { + "left": "pull_request:5855", + "right": "pull_request:7495", + "accept": false, + "reason": "Both are README edits, but they target different docs and different link/content corrections." + }, + { + "left": "pull_request:5901", + "right": "pull_request:6760", + "accept": false, + "reason": "Same broad IP-Adapter area, but one adds ControlNet Img2Img support and the other updates slow tests; not the same code-path fix." + }, + { + "left": "pull_request:5901", + "right": "pull_request:7260", + "accept": false, + "reason": "Both relate to IP Adapter support, but they modify different pipelines and different integration paths." + }, + { + "left": "pull_request:5911", + "right": "pull_request:6403", + "accept": false, + "reason": "Different features in different files: IPAdapterFull support versus WebUI-format training script support." + }, + { + "left": "pull_request:5961", + "right": "pull_request:7977", + "accept": false, + "reason": "Both are typo/docstring edits, but they fix different text in different files; not a duplicate change." + }, + { + "left": "pull_request:5982", + "right": "pull_request:7857", + "accept": false, + "reason": "Both add docs navigation entries, but they cover different model families and different documentation pages." + }, + { + "left": "pull_request:5986", + "right": "pull_request:7472", + "accept": false, + "reason": "Both are broken-link fixes, but they point to different pages and different link targets." + }, + { + "left": "pull_request:60", + "right": "pull_request:89", + "accept": false, + "reason": "PR 89 is a broader SDE-to-unconditional merge, while PR 60 is a narrower earlier fix; they overlap in files but are not the same concrete change." + }, + { + "left": "pull_request:6081", + "right": "pull_request:7913", + "accept": false, + "reason": "One edits SDXL pipeline sigma handling for torch.compile, the other adds custom sigmas/timesteps to ControlNet XL; different behavior and code paths." + }, + { + "left": "pull_request:61", + "right": "pull_request:98", + "accept": false, + "reason": "Both are conversion scripts, but they target different model families and conversion workflows." + }, + { + "left": "pull_request:6106", + "right": "pull_request:7015", + "accept": false, + "reason": "Both touch example docs/code around loading pipelines, but they fix different examples and different single-file loading issues." + }, + { + "left": "pull_request:6122", + "right": "pull_request:7181", + "accept": false, + "reason": "Both are broad typo/doc cleanups, but they span different files and unrelated edits." + }, + { + "left": "pull_request:6144", + "right": "pull_request:6952", + "accept": false, + "reason": "Different training examples with different bug fixes; not the same underlying change." + }, + { + "left": "pull_request:6147", + "right": "pull_request:6961", + "accept": false, + "reason": "One fixes single-file loading for SDXL inpainting/refiner, the other adds fast tests for single-file checkpoint loaders." + }, + { + "left": "pull_request:6163", + "right": "pull_request:6221", + "accept": false, + "reason": "Both are test fixes, but they cover different pipelines and different slow-test behaviors." + }, + { + "left": "pull_request:6229", + "right": "pull_request:6311", + "accept": false, + "reason": "Different CI/example concerns: one adds a PEFT dependency, the other adds Fast Diffusion docs." + }, + { + "left": "pull_request:6229", + "right": "pull_request:7797", + "accept": false, + "reason": "One is CI dependency setup, the other shrinks DDPM test models; unrelated changes." + }, + { + "left": "pull_request:6229", + "right": "pull_request:7803", + "accept": false, + "reason": "One is CI dependency setup, the other shrinks DDIM test models; unrelated changes." + }, + { + "left": "pull_request:6229", + "right": "pull_request:7804", + "accept": false, + "reason": "One is CI dependency setup, the other shrinks Amused test models; unrelated changes." + }, + { + "left": "pull_request:6276", + "right": "pull_request:8426", + "accept": false, + "reason": "A community pipeline feature PR versus a workflow cleanup PR; no shared bug/change." + }, + { + "left": "pull_request:6291", + "right": "pull_request:7301", + "accept": false, + "reason": "Different test suites and different assertions; both are test fixes but not the same issue." + }, + { + "left": "pull_request:63", + "right": "pull_request:98", + "accept": false, + "reason": "Different conversion utilities for different pipelines/models; not the same change." + }, + { + "left": "pull_request:6308", + "right": "pull_request:6965", + "accept": false, + "reason": "Both involve LoRA/PEFT tests, but they target different integration scenarios and different failures." + }, + { + "left": "pull_request:6332", + "right": "pull_request:7062", + "accept": false, + "reason": "Different community pipelines and different edits; not the same feature or bug." + }, + { + "left": "pull_request:6339", + "right": "pull_request:6620", + "accept": false, + "reason": "Both are docs-related, but one fixes AnimateDiff example docs and the other only adjusts script arguments/comments." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 78, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:6424", + "pull_request:6573", + "pull_request:6669", + "pull_request:6680", + "pull_request:6764", + "pull_request:6789", + "pull_request:6868", + "pull_request:7068", + "pull_request:7113", + "pull_request:7118", + "pull_request:7236", + "pull_request:7247", + "pull_request:7325", + "pull_request:7360", + "pull_request:7451", + "pull_request:7542", + "pull_request:7544", + "pull_request:7709", + "pull_request:7719", + "pull_request:7746", + "pull_request:7786", + "pull_request:7868", + "pull_request:7919", + "pull_request:7932", + "pull_request:7934", + "pull_request:7977", + "pull_request:8230", + "pull_request:8426", + "pull_request:8470", + "pull_request:8483", + "pull_request:8490", + "pull_request:8498", + "pull_request:8583", + "pull_request:8600", + "pull_request:8642", + "pull_request:8675", + "pull_request:8691", + "pull_request:8851", + "pull_request:8860", + "pull_request:8883", + "pull_request:8984", + "pull_request:9066", + "pull_request:9077", + "pull_request:9100", + "pull_request:9249", + "pull_request:9286", + "pull_request:9765", + "pull_request:9860" + ], + "soft_pairs": [ + "pull_request:6424|pull_request:6789", + "pull_request:6573|pull_request:6868", + "pull_request:6669|pull_request:8470", + "pull_request:6680|pull_request:7746", + "pull_request:6680|pull_request:8883", + "pull_request:6764|pull_request:7113", + "pull_request:7068|pull_request:7325", + "pull_request:7068|pull_request:9077", + "pull_request:7118|pull_request:7325", + "pull_request:7236|pull_request:9286", + "pull_request:7247|pull_request:7932", + "pull_request:7360|pull_request:8851", + "pull_request:7451|pull_request:9077", + "pull_request:7542|pull_request:8230", + "pull_request:7542|pull_request:8642", + "pull_request:7544|pull_request:7868", + "pull_request:7544|pull_request:8230", + "pull_request:7709|pull_request:7719", + "pull_request:7786|pull_request:7919", + "pull_request:7934|pull_request:8691", + "pull_request:7977|pull_request:9100", + "pull_request:8426|pull_request:9765", + "pull_request:8470|pull_request:8490", + "pull_request:8470|pull_request:8860", + "pull_request:8483|pull_request:8675", + "pull_request:8483|pull_request:9066", + "pull_request:8498|pull_request:9100", + "pull_request:8583|pull_request:8984", + "pull_request:8583|pull_request:9249", + "pull_request:8583|pull_request:9860", + "pull_request:8600|pull_request:9286" + ], + "prepared_review_unit_hash": "9698391277a72974b12514fbc6a8d9e8689f1cb1080be9bec5b75c78c76526b3", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 102104, + "estimated_input_tokens": 25526, + "estimated_eval_tokens": 51308 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly heterogeneous PRs spanning docs, CI, tests, and unrelated feature work. Only two soft edges look like the same underlying test/workflow change; the rest are different bugs or unrelated edits.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a true canonical duplicate target because the cluster is not one coherent change set. If a representative is needed, 8483 is the broadest and most central PR in the set.", + "best_issue_reason": null, + "best_pr_reason": "8483 is the largest and most consequential item here (SD3 core feature work with broad code coverage and high review activity), so it's the best representative PR, though not a duplicate anchor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6424", + "right": "pull_request:6789", + "accept": false, + "reason": "Related PixArt work, but one is a converter fix and the other adds 256px inference/pipeline changes; not the same concrete change." + }, + { + "left": "pull_request:6573", + "right": "pull_request:6868", + "accept": false, + "reason": "Both are IP-Adapter related, but they address different features: multi-adapter support vs passing image embeddings through pipelines." + }, + { + "left": "pull_request:6669", + "right": "pull_request:8470", + "accept": false, + "reason": "Both touch README content, but they are unrelated documentation updates for different projects/topics." + }, + { + "left": "pull_request:6680", + "right": "pull_request:7746", + "accept": false, + "reason": "Different test fixes for different components; not the same bug or code path." + }, + { + "left": "pull_request:6680", + "right": "pull_request:8883", + "accept": false, + "reason": "A scheduler test fix and a CI overhaul are unrelated." + }, + { + "left": "pull_request:6764", + "right": "pull_request:7113", + "accept": false, + "reason": "One is a docs note about LoRA serialization; the other adds Docker CI Slack notifications." + }, + { + "left": "pull_request:7068", + "right": "pull_request:7325", + "accept": false, + "reason": "Both are typo sweeps, but they touch different docs/code areas and fix different text." + }, + { + "left": "pull_request:7068", + "right": "pull_request:9077", + "accept": false, + "reason": "Same broad typo category, but not the same concrete correction set or target files." + }, + { + "left": "pull_request:7118", + "right": "pull_request:7325", + "accept": false, + "reason": "Docs typo cleanup vs a much broader multi-file maintenance PR; not the same underlying change." + }, + { + "left": "pull_request:7236", + "right": "pull_request:9286", + "accept": false, + "reason": "A training script update and an __init__ export addition are unrelated." + }, + { + "left": "pull_request:7247", + "right": "pull_request:7932", + "accept": false, + "reason": "Different test suites and different failure modes; not duplicate fixes." + }, + { + "left": "pull_request:7360", + "right": "pull_request:8851", + "accept": false, + "reason": "A scheduler docs tweak and AuraFlow docs addition are unrelated." + }, + { + "left": "pull_request:7451", + "right": "pull_request:9077", + "accept": false, + "reason": "Both are typo fixes, but they are separate sweeps across different files and content." + }, + { + "left": "pull_request:7542", + "right": "pull_request:8230", + "accept": false, + "reason": "A single CPU-offload test typo vs a broader offloading usage update; not the same concrete issue." + }, + { + "left": "pull_request:7542", + "right": "pull_request:8642", + "accept": false, + "reason": "Different typo fixes in different tests/docs; not the same change." + }, + { + "left": "pull_request:7544", + "right": "pull_request:7868", + "accept": true, + "reason": "Both fix the same CPU offload test expectations in the same test file and could plausibly be merged into one PR." + }, + { + "left": "pull_request:7544", + "right": "pull_request:8230", + "accept": false, + "reason": "One is a test assertion cleanup; the other is a broad CPU-offloading usage/docs PR." + }, + { + "left": "pull_request:7709", + "right": "pull_request:7719", + "accept": true, + "reason": "Both are sequential edits to the same push_tests workflow around Tailscale setup and failure handling; they fit one PR." + }, + { + "left": "pull_request:7786", + "right": "pull_request:7919", + "accept": false, + "reason": "Stable Video Diffusion prompt batching and device-map offload behavior are different bugs." + }, + { + "left": "pull_request:7934", + "right": "pull_request:8691", + "accept": false, + "reason": "Single-file config loading and LoRA loading behavior are unrelated." + }, + { + "left": "pull_request:7977", + "right": "pull_request:9100", + "accept": false, + "reason": "Both are typo fixes, but in unrelated docs/code files." + }, + { + "left": "pull_request:8426", + "right": "pull_request:9765", + "accept": false, + "reason": "Mirror community pipeline CI changes and SD3 docstring cleanup are unrelated." + }, + { + "left": "pull_request:8470", + "right": "pull_request:8490", + "accept": false, + "reason": "Both update README files, but for different projects and purposes." + }, + { + "left": "pull_request:8470", + "right": "pull_request:8860", + "accept": false, + "reason": "Different README updates; no shared underlying bug/change." + }, + { + "left": "pull_request:8483", + "right": "pull_request:8675", + "accept": false, + "reason": "Large SD3 feature work vs a fast-diffusion docs note; unrelated." + }, + { + "left": "pull_request:8483", + "right": "pull_request:9066", + "accept": false, + "reason": "Adding SD3 core support and adding Stable Cascade docs are separate changes." + }, + { + "left": "pull_request:8498", + "right": "pull_request:9100", + "accept": false, + "reason": "Both are small typo fixes, but they target different content and are not the same correction." + }, + { + "left": "pull_request:8583", + "right": "pull_request:8984", + "accept": false, + "reason": "XFormers support for SD3 and ONNX conversion/pipeline support are different code paths and changes." + }, + { + "left": "pull_request:8583", + "right": "pull_request:9249", + "accept": false, + "reason": "SD3 attention backend support and a ControlNet example are unrelated." + }, + { + "left": "pull_request:8583", + "right": "pull_request:9860", + "accept": false, + "reason": "Different SD3-related features in different subsystems; not mergeable as one PR." + }, + { + "left": "pull_request:8600", + "right": "pull_request:9286", + "accept": false, + "reason": "A training script fix and an exported class addition are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The two accepted soft edges both look plausibly mergeable into a single PR for the same area of fix (CPU offload tests and the push_tests Tailscale workflow), while the rejected pairs are appropriately treated as distinct changes rather than duplicates. The representative PR choice and the note that there is no canonical duplicate anchor are both supported by the evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 79, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10494", + "pull_request:11059", + "pull_request:11124", + "pull_request:11522", + "pull_request:11595", + "pull_request:11835", + "pull_request:11843", + "pull_request:12040", + "pull_request:12170", + "pull_request:1513", + "pull_request:1838", + "pull_request:2183", + "pull_request:2339", + "pull_request:3968", + "pull_request:4098", + "pull_request:4099", + "pull_request:4659", + "pull_request:4661", + "pull_request:5058", + "pull_request:5117", + "pull_request:5202", + "pull_request:5407", + "pull_request:5470", + "pull_request:5697", + "pull_request:5807", + "pull_request:6236", + "pull_request:6573", + "pull_request:7186", + "pull_request:8642", + "pull_request:8845", + "pull_request:8937", + "pull_request:8965", + "pull_request:8984", + "pull_request:9244", + "pull_request:9426", + "pull_request:9571", + "pull_request:9620", + "pull_request:9658", + "pull_request:9687", + "pull_request:9706", + "pull_request:9709", + "pull_request:9754", + "pull_request:9765", + "pull_request:9827", + "pull_request:9832", + "pull_request:9860", + "pull_request:9994" + ], + "soft_pairs": [ + "pull_request:8642|pull_request:8845", + "pull_request:8845|pull_request:9765", + "pull_request:8965|pull_request:9994", + "pull_request:8984|pull_request:9860", + "pull_request:8984|pull_request:9994", + "pull_request:9571|pull_request:9658", + "pull_request:9620|pull_request:9860", + "pull_request:9754|pull_request:9832", + "pull_request:12040|pull_request:12170", + "pull_request:5470|pull_request:6236", + "pull_request:11522|pull_request:11835", + "pull_request:1838|pull_request:2183", + "pull_request:3968|pull_request:4098", + "pull_request:9426|pull_request:9709", + "pull_request:5058|pull_request:5202", + "pull_request:1513|pull_request:2339", + "pull_request:4099|pull_request:4659", + "pull_request:4099|pull_request:4661", + "pull_request:5117|pull_request:5407", + "pull_request:5697|pull_request:5807", + "pull_request:8937|pull_request:9706", + "pull_request:9244|pull_request:9687", + "pull_request:6573|pull_request:7186", + "pull_request:10494|pull_request:9827", + "pull_request:11059|pull_request:11843", + "pull_request:11124|pull_request:11595" + ], + "prepared_review_unit_hash": "9bcebd7458211b00eca3c26aed800dccd5a251311e42bbd943edfa9a64aaac3e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 104331, + "estimated_input_tokens": 26083, + "estimated_eval_tokens": 52422 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All proposed soft edges look like false positives: they share broad themes (LoRA, ControlNet, SDXL, tests/docs, CI, typos) but not the same concrete bug or change path.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out; the candidates are heterogeneous and do not collapse into one underlying change.", + "best_issue_reason": null, + "best_pr_reason": "No clear global best PR for deduping in this set, since none of the PRs are duplicates of each other.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8642", + "right": "pull_request:8845", + "accept": false, + "reason": "Both are typo fixes, but one is SD3 docstring text and the other is PAG test assertions; unrelated changes." + }, + { + "left": "pull_request:8845", + "right": "pull_request:9765", + "accept": false, + "reason": "Same generic 'typo fix' theme, but they touch different features and code paths (PAG tests vs SD3 pipeline docs)." + }, + { + "left": "pull_request:8965", + "right": "pull_request:9994", + "accept": false, + "reason": "One is a small pipeline instantiation rename; the other adds OmniGen model/pipeline support. Not the same change." + }, + { + "left": "pull_request:8984", + "right": "pull_request:9860", + "accept": false, + "reason": "SD3 ONNX export support and SD3.5 ControlNet example updates are different features with no shared concrete bug." + }, + { + "left": "pull_request:8984", + "right": "pull_request:9994", + "accept": false, + "reason": "ONNX conversion for SD3 and Omnigen model support are unrelated additions." + }, + { + "left": "pull_request:9571", + "right": "pull_request:9658", + "accept": false, + "reason": "Flux ControlNet guidance scheduling and a community demo script are not the same code-path fix." + }, + { + "left": "pull_request:9620", + "right": "pull_request:9860", + "accept": false, + "reason": "ControlNet interrupt handling and SD3.5 ControlNet example support are different issues entirely." + }, + { + "left": "pull_request:9754", + "right": "pull_request:9832", + "accept": false, + "reason": "A Flux quantization training script and quantization CI/tests are related only broadly; not the same underlying change." + }, + { + "left": "pull_request:12040", + "right": "pull_request:12170", + "accept": false, + "reason": "Both touch LoRA conversion utilities, but they support different model families and different concrete conversion logic." + }, + { + "left": "pull_request:5470", + "right": "pull_request:6236", + "accept": false, + "reason": "One fixes ControlNet inpaint pipeline code; the other corrects a documentation example. Not duplicates." + }, + { + "left": "pull_request:11522", + "right": "pull_request:11835", + "accept": false, + "reason": "Audioldm2 transformers-main compatibility and float16 CUDA test cleanup are separate fixes." + }, + { + "left": "pull_request:1838", + "right": "pull_request:2183", + "accept": false, + "reason": "Both involve textual inversion, but one adds image logging and the other patches pipeline/tokenizer setup; different problems." + }, + { + "left": "pull_request:3968", + "right": "pull_request:4098", + "accept": false, + "reason": "SDXL broad improvements versus adding SDXL inpainting are distinct feature sets." + }, + { + "left": "pull_request:9426", + "right": "pull_request:9709", + "accept": false, + "reason": "Artifact uploader version bump and torch version pin are unrelated CI maintenance changes." + }, + { + "left": "pull_request:5058", + "right": "pull_request:5202", + "accept": false, + "reason": "PEFT text-encoder integration and Kohya checkpoint mapping are different LoRA-related changes, not one bug fix." + }, + { + "left": "pull_request:1513", + "right": "pull_request:2339", + "accept": false, + "reason": "DreamBooth README guidance and checkpointing docs/tests are separate documentation updates." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4659", + "accept": false, + "reason": "SDXL prompt-embed bugfix and long weighted prompt pipeline/docs overhaul are not the same underlying change." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4661", + "accept": false, + "reason": "Prompt-embed bugfix versus community long weighted prompt pipeline addition: related domain, different fix." + }, + { + "left": "pull_request:5117", + "right": "pull_request:5407", + "accept": false, + "reason": "Training resume log handling and mixed-precision training fixes are unrelated example-script changes." + }, + { + "left": "pull_request:5697", + "right": "pull_request:5807", + "accept": false, + "reason": "Peft/xformers attention fix and LPW SDXL num_images_per_prompt support are unrelated code paths." + }, + { + "left": "pull_request:8937", + "right": "pull_request:9706", + "accept": false, + "reason": "AuraFlow long-prompt handling and a docstring wording cleanup in another model are not duplicates." + }, + { + "left": "pull_request:9244", + "right": "pull_request:9687", + "accept": false, + "reason": "Flux single-file support and Xlabs ControlNet support are separate Flux features, not the same change." + }, + { + "left": "pull_request:6573", + "right": "pull_request:7186", + "accept": false, + "reason": "Both are IP-Adapter-related, but one adds multiple-adapter support and the other moves Face ID to core; different concrete changes." + }, + { + "left": "pull_request:10494", + "right": "pull_request:9827", + "accept": false, + "reason": "ControlNet interrupt support and LoRA device_map handling are unrelated." + }, + { + "left": "pull_request:11059", + "right": "pull_request:11843", + "accept": false, + "reason": "Wan t2v LoRA conversion support and Wan VACE exclude_modules tests are different model targets and different changes." + }, + { + "left": "pull_request:11124", + "right": "pull_request:11595", + "accept": false, + "reason": "CogView4 test fixture update and Flux LoRA load fix are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 80, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10096", + "pull_request:10544", + "pull_request:11039", + "pull_request:11067", + "pull_request:11238", + "pull_request:11739", + "pull_request:11743", + "pull_request:11765", + "pull_request:11814", + "pull_request:11985", + "pull_request:12269", + "pull_request:12445", + "pull_request:12714", + "pull_request:13126", + "pull_request:13169", + "pull_request:13209", + "pull_request:1788", + "pull_request:2216", + "pull_request:2223", + "pull_request:2928", + "pull_request:2994", + "pull_request:3641", + "pull_request:3691", + "pull_request:3798", + "pull_request:4034", + "pull_request:4223", + "pull_request:4272", + "pull_request:4352", + "pull_request:4373", + "pull_request:4714", + "pull_request:4833", + "pull_request:4886", + "pull_request:5188", + "pull_request:5201", + "pull_request:5298", + "pull_request:6564", + "pull_request:7018", + "pull_request:7096", + "pull_request:7103", + "pull_request:7153", + "pull_request:7161", + "pull_request:7476", + "pull_request:819", + "pull_request:9638", + "pull_request:9662", + "pull_request:9735", + "pull_request:9770" + ], + "soft_pairs": [ + "pull_request:11739|pull_request:11765", + "pull_request:11743|pull_request:11985", + "pull_request:12714|pull_request:13209", + "pull_request:2216|pull_request:2223", + "pull_request:2928|pull_request:4223", + "pull_request:2994|pull_request:3798", + "pull_request:3641|pull_request:4714", + "pull_request:3691|pull_request:4034", + "pull_request:4272|pull_request:4714", + "pull_request:4352|pull_request:4373", + "pull_request:4833|pull_request:5298", + "pull_request:4886|pull_request:7103", + "pull_request:5188|pull_request:7096", + "pull_request:5201|pull_request:7018", + "pull_request:6564|pull_request:7153", + "pull_request:9638|pull_request:9770", + "pull_request:7161|pull_request:7476", + "pull_request:12269|pull_request:12445", + "pull_request:13126|pull_request:13169", + "pull_request:10096|pull_request:11039", + "pull_request:10544|pull_request:11814", + "pull_request:11067|pull_request:9662", + "pull_request:11238|pull_request:9735", + "pull_request:1788|pull_request:819" + ], + "prepared_review_unit_hash": "03eb3144ae9afd67dac145f05b6b598a35c032b256987077efe36aebe07e4e19", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 96077, + "estimated_input_tokens": 24020, + "estimated_eval_tokens": 48296 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most candidate pairs only share broad subsystem or wording similarity, not the same concrete bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits: these PRs span unrelated docs, CI, LoRA, ControlNet, modular pipeline, and scheduler changes.", + "best_issue_reason": null, + "best_pr_reason": "No PR is globally suitable as a cluster canonical because there is no single underlying change represented across the set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11739", + "right": "pull_request:11765", + "accept": false, + "reason": "Both are LoRA metadata/alpha edits, but they target different training scripts and models (SDXL vs HiDream), so they are not the same change." + }, + { + "left": "pull_request:11743", + "right": "pull_request:11985", + "accept": false, + "reason": "Different features: a Flux LoRA quantization training script versus Flux Kontext modular pipeline support." + }, + { + "left": "pull_request:12714", + "right": "pull_request:13209", + "accept": false, + "reason": "Both involve LoRA-related testing/conversion, but they fix different code paths for different models/formats (Flux2 NaN test vs Z-Image conversion)." + }, + { + "left": "pull_request:2216", + "right": "pull_request:2223", + "accept": false, + "reason": "Completely different pipelines: DifFace versus semantic stable diffusion." + }, + { + "left": "pull_request:2928", + "right": "pull_request:4223", + "accept": false, + "reason": "Both touch ControlNet examples, but one adds local dataset loading and the other fixes SDXL script details; not the same bug." + }, + { + "left": "pull_request:2994", + "right": "pull_request:3798", + "accept": false, + "reason": "Type-casting in ControlNet training and comment cleanup are unrelated changes." + }, + { + "left": "pull_request:3641", + "right": "pull_request:4714", + "accept": false, + "reason": "Inpainting latent handling and missing-transformers error messaging are unrelated code paths." + }, + { + "left": "pull_request:3691", + "right": "pull_request:4034", + "accept": false, + "reason": "Both are LoRA-related, but one changes text-encoder attention processor selection while the other adds unload/fuse utility behavior." + }, + { + "left": "pull_request:4272", + "right": "pull_request:4714", + "accept": false, + "reason": "Optional import fixes and missing-transformers error handling are separate issues." + }, + { + "left": "pull_request:4352", + "right": "pull_request:4373", + "accept": false, + "reason": "Kandinsky combined pipeline offload/xformers tweaks and SDXL resize logic are unrelated." + }, + { + "left": "pull_request:4833", + "right": "pull_request:5298", + "accept": false, + "reason": "Both touch LoRA internals, but one fixes unload/fuse behavior and the other fixes torch.compile behavior for LoRA convs." + }, + { + "left": "pull_request:4886", + "right": "pull_request:7103", + "accept": false, + "reason": "StableDiffusionUpscale safety checking and ControlNet image validation are different bugs in different pipelines." + }, + { + "left": "pull_request:5188", + "right": "pull_request:7096", + "accept": false, + "reason": "Compute-SNR training utility work is unrelated to ControlNet example updates." + }, + { + "left": "pull_request:5201", + "right": "pull_request:7018", + "accept": false, + "reason": "Both are LoRA-related, but one fixes xformers LoRA handling and the other fixes gradient unscaling in an advanced training script." + }, + { + "left": "pull_request:6564", + "right": "pull_request:7153", + "accept": false, + "reason": "IP Adapter path handling and documentation workflow path filtering are unrelated CI/runtime changes." + }, + { + "left": "pull_request:9638", + "right": "pull_request:9770", + "accept": false, + "reason": "Both concern Xlabs controlnets, but they modify different pipelines and code paths; they are not mergeable as one concrete fix." + }, + { + "left": "pull_request:7161", + "right": "pull_request:7476", + "accept": false, + "reason": "A docs example for torch.compile and a multi-ControlNet inpaint bug fix are unrelated." + }, + { + "left": "pull_request:12269", + "right": "pull_request:12445", + "accept": false, + "reason": "Both are Flux modular changes, but they address different goals and code paths; not the same underlying change." + }, + { + "left": "pull_request:13126", + "right": "pull_request:13169", + "accept": false, + "reason": "Different LoRA formats and conversion logic: Z-Image distilled loras versus Flux Klein/PEFT lora format." + }, + { + "left": "pull_request:10096", + "right": "pull_request:11039", + "accept": false, + "reason": "Stale-label handling and torch_dtype/quantization loading behavior are unrelated." + }, + { + "left": "pull_request:10544", + "right": "pull_request:11814", + "accept": false, + "reason": "Docs for ParaAttention and a Flux gradient-checkpointing fix are unrelated changes." + }, + { + "left": "pull_request:11067", + "right": "pull_request:9662", + "accept": false, + "reason": "Benchmark dependency pinning and a LoRA test version bump are unrelated CI/test maintenance." + }, + { + "left": "pull_request:11238", + "right": "pull_request:9735", + "accept": false, + "reason": "One fixes SD3 ControlNet validation runtime; the other updates SD3 ControlNet example/docs." + }, + { + "left": "pull_request:1788", + "right": "pull_request:819", + "accept": false, + "reason": "Same file, but different DDIM bugs: prediction_type handling versus Windows int64 timestep dtype." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 81, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10058", + "pull_request:10318", + "pull_request:10750", + "pull_request:10911", + "pull_request:10985", + "pull_request:11179", + "pull_request:11532", + "pull_request:11715", + "pull_request:11786", + "pull_request:11896", + "pull_request:12086", + "pull_request:12277", + "pull_request:12288", + "pull_request:12439", + "pull_request:12714", + "pull_request:12812", + "pull_request:12919", + "pull_request:13055", + "pull_request:13058", + "pull_request:2393", + "pull_request:2721", + "pull_request:2836", + "pull_request:2919", + "pull_request:3859", + "pull_request:4083", + "pull_request:4276", + "pull_request:4498", + "pull_request:4744", + "pull_request:4783", + "pull_request:5015", + "pull_request:5127", + "pull_request:5186", + "pull_request:5396", + "pull_request:5713", + "pull_request:5744", + "pull_request:5887", + "pull_request:5915", + "pull_request:6355", + "pull_request:6573", + "pull_request:6759", + "pull_request:6828", + "pull_request:6946", + "pull_request:7615", + "pull_request:7825", + "pull_request:7932", + "pull_request:8480", + "pull_request:8566", + "pull_request:9770" + ], + "soft_pairs": [ + "pull_request:2836|pull_request:5015", + "pull_request:2919|pull_request:4276", + "pull_request:4498|pull_request:4783", + "pull_request:4498|pull_request:5186", + "pull_request:7825|pull_request:7932", + "pull_request:5713|pull_request:6573", + "pull_request:12086|pull_request:12439", + "pull_request:11532|pull_request:11896", + "pull_request:10750|pull_request:9770", + "pull_request:6355|pull_request:7615", + "pull_request:2393|pull_request:2721", + "pull_request:3859|pull_request:4083", + "pull_request:5127|pull_request:5915", + "pull_request:10058|pull_request:11179", + "pull_request:5396|pull_request:6759", + "pull_request:8480|pull_request:8566", + "pull_request:12812|pull_request:12919", + "pull_request:12277|pull_request:13055", + "pull_request:4744|pull_request:5744", + "pull_request:6828|pull_request:6946", + "pull_request:12288|pull_request:13058", + "pull_request:5887|pull_request:6828", + "pull_request:10318|pull_request:10985", + "pull_request:10911|pull_request:11715", + "pull_request:11786|pull_request:12714" + ], + "prepared_review_unit_hash": "f47173f3ea2aaf2fa25b9e15b475127880fa63161a2dffa1457f668d5b6a5f67", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 107929, + "estimated_input_tokens": 26983, + "estimated_eval_tokens": 54222 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs fixes, test-only changes, CI tweaks, and distinct feature work across many subsystems. The soft pairs are mostly superficial matches by topic or filename pattern, not the same underlying bug/change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out; the items span unrelated docs, tests, CI, and feature additions, so there isn\u2019t one common concrete change to anchor the cluster.", + "best_issue_reason": null, + "best_pr_reason": "No PR in this set is a good global representative for duplicate triage because the cluster is too heterogeneous and the soft matches do not share one fixable code path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2836", + "right": "pull_request:5015", + "accept": false, + "reason": "Both are docs link fixes, but they target different pages and different broken links; not the same change." + }, + { + "left": "pull_request:2919", + "right": "pull_request:4276", + "accept": false, + "reason": "One speeds up a unit test, the other adds import coverage; different test changes and no shared bug." + }, + { + "left": "pull_request:4498", + "right": "pull_request:4783", + "accept": false, + "reason": "Both touch docs around pipelines, but one adds a new SDXL reference example and the other adds SDXL prompt-embedding docs; unrelated changes." + }, + { + "left": "pull_request:4498", + "right": "pull_request:5186", + "accept": false, + "reason": "A community example addition versus a FreeU proof-of-concept in core models; clearly different features." + }, + { + "left": "pull_request:7825", + "right": "pull_request:7932", + "accept": false, + "reason": "Both adjust tests, but for different models and different failure modes; not the same underlying issue." + }, + { + "left": "pull_request:5713", + "right": "pull_request:6573", + "accept": false, + "reason": "Both are IP-Adapter related, but the latter adds multi-adapter support on top of the former's initial feature; not duplicate fixes." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12439", + "accept": false, + "reason": "Different feature areas: modular Flux I2I support versus attention backend kernel support." + }, + { + "left": "pull_request:11532", + "right": "pull_request:11896", + "accept": false, + "reason": "One adds HiDream LoRA conversion support, the other fixes hook disabling when loading LoRAs; different code paths." + }, + { + "left": "pull_request:10750", + "right": "pull_request:9770", + "accept": false, + "reason": "Both are Flux ControlNet work, but one adds img2img support and the other adds Xlabs inpainting support; separate features." + }, + { + "left": "pull_request:6355", + "right": "pull_request:7615", + "accept": false, + "reason": "Same example file family, but the changes address different SDXL training/model-card concerns and are not the same fix." + }, + { + "left": "pull_request:2393", + "right": "pull_request:2721", + "accept": false, + "reason": "Both add pipeline docs/examples, but they introduce different pipelines with distinct implementations." + }, + { + "left": "pull_request:3859", + "right": "pull_request:4083", + "accept": false, + "reason": "SDXL-related broad changes, but one adds new pipelines and the other changes VAE precision behavior; not one bug." + }, + { + "left": "pull_request:5127", + "right": "pull_request:5915", + "accept": false, + "reason": "Different adapter features: multi-adapter support versus IP-Adapter Plus loading." + }, + { + "left": "pull_request:10058", + "right": "pull_request:11179", + "accept": false, + "reason": "Both are docs-only fixes, but one fixes a training link and the other corrects an environment variable value; unrelated." + }, + { + "left": "pull_request:5396", + "right": "pull_request:6759", + "accept": false, + "reason": "One fixes stale.py timezone handling, the other fixes an example pipeline tensor repeat; no shared code path." + }, + { + "left": "pull_request:8480", + "right": "pull_request:8566", + "accept": false, + "reason": "HunyuanDiT ControlNet support versus SD3 ControlNet support; related domain but distinct models and implementations." + }, + { + "left": "pull_request:12812", + "right": "pull_request:12919", + "accept": false, + "reason": "Both touch attention_dispatch, but they enable different backend capabilities for different hardware paths." + }, + { + "left": "pull_request:12277", + "right": "pull_request:13055", + "accept": false, + "reason": "Both are test/pipeline tweaks, but they fix different pipelines with unrelated conditions." + }, + { + "left": "pull_request:4744", + "right": "pull_request:5744", + "accept": false, + "reason": "Both adjust docs text, but they are separate docstring/title cleanup PRs with no common bug." + }, + { + "left": "pull_request:6828", + "right": "pull_request:6946", + "accept": false, + "reason": "One updates UNetMotionModel IP-adapter logic, the other adds low_cpu_mem_usage for IP-Adapter loading; different problems." + }, + { + "left": "pull_request:12288", + "right": "pull_request:13058", + "accept": false, + "reason": "One adds InfiniteTalk pipeline code, the other adds LTX2 condition pipeline support; unrelated feature work." + }, + { + "left": "pull_request:5887", + "right": "pull_request:6828", + "accept": false, + "reason": "IP-Adapter in ControlNet inpaint versus UNetMotionModel IP-adapter handling; different integration points." + }, + { + "left": "pull_request:10318", + "right": "pull_request:10985", + "accept": false, + "reason": "Different LoRA test/logic areas: Flux LoRA integration tests versus ComfyUI LoRA conversion support." + }, + { + "left": "pull_request:10911", + "right": "pull_request:11715", + "accept": false, + "reason": "Both are version-gating test fixes, but they target different models and packages." + }, + { + "left": "pull_request:11786", + "right": "pull_request:12714", + "accept": false, + "reason": "A CI marker refactor versus a Flux2 LoRA NaN test fix; not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 82, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10318", + "pull_request:10837", + "pull_request:10983", + "pull_request:11797", + "pull_request:12207", + "pull_request:1250", + "pull_request:12681", + "pull_request:12839", + "pull_request:12921", + "pull_request:13296", + "pull_request:13323", + "pull_request:13364", + "pull_request:1527", + "pull_request:192", + "pull_request:2009", + "pull_request:2037", + "pull_request:2415", + "pull_request:2921", + "pull_request:2994", + "pull_request:3245", + "pull_request:3294", + "pull_request:3471", + "pull_request:3691", + "pull_request:4315", + "pull_request:4659", + "pull_request:4666", + "pull_request:4684", + "pull_request:4706", + "pull_request:473", + "pull_request:4947", + "pull_request:5506", + "pull_request:5823", + "pull_request:5852", + "pull_request:6105", + "pull_request:6133", + "pull_request:6225", + "pull_request:6303", + "pull_request:6752", + "pull_request:6762", + "pull_request:6893", + "pull_request:7044", + "pull_request:7115", + "pull_request:7155", + "pull_request:7200", + "pull_request:7542", + "pull_request:7786", + "pull_request:826", + "pull_request:9244" + ], + "soft_pairs": [ + "pull_request:12839|pull_request:12921", + "pull_request:192|pull_request:2009", + "pull_request:3294|pull_request:3691", + "pull_request:4947|pull_request:6762", + "pull_request:5852|pull_request:7542", + "pull_request:10837|pull_request:10983", + "pull_request:6105|pull_request:6752", + "pull_request:2994|pull_request:826", + "pull_request:4659|pull_request:4666", + "pull_request:4706|pull_request:5506", + "pull_request:5823|pull_request:6225", + "pull_request:6893|pull_request:7044", + "pull_request:7200|pull_request:7786", + "pull_request:6133|pull_request:6303", + "pull_request:7115|pull_request:7155", + "pull_request:12681|pull_request:13296", + "pull_request:10318|pull_request:9244", + "pull_request:11797|pull_request:12207", + "pull_request:1250|pull_request:2037", + "pull_request:13323|pull_request:13364", + "pull_request:1527|pull_request:473", + "pull_request:2415|pull_request:2921", + "pull_request:3245|pull_request:4684", + "pull_request:3471|pull_request:4315" + ], + "prepared_review_unit_hash": "661ff5b3ddce9d13bae2aadcb8642e7886009bc1385bf9cb25550f8fbd674507", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 103250, + "estimated_input_tokens": 25813, + "estimated_eval_tokens": 51882 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: the soft-similarity pairs mostly share only broad subsystems or filenames, but they do not describe the same underlying bug/change. I reject all soft edges and do not see a valid canonical duplicate PR in this set.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:12839", + "right": "pull_request:12921", + "accept": false, + "reason": "Different model launches/features (Qwen-image edit support vs GLM-Image support); not the same change." + }, + { + "left": "pull_request:192", + "right": "pull_request:2009", + "accept": false, + "reason": "A README parameter tweak is unrelated to adding textual inversion loading support." + }, + { + "left": "pull_request:3294", + "right": "pull_request:3691", + "accept": false, + "reason": "Both touch LoRA loading, but one adds prefix handling and the other changes processor selection; different code-path problems." + }, + { + "left": "pull_request:4947", + "right": "pull_request:6762", + "accept": false, + "reason": "Training-script adapter argument support and a flaky Kandinsky test fix are unrelated." + }, + { + "left": "pull_request:5852", + "right": "pull_request:7542", + "accept": false, + "reason": "CI workflow variable fix is unrelated to a CPU offload test typo." + }, + { + "left": "pull_request:10837", + "right": "pull_request:10983", + "accept": false, + "reason": "Both are HunyuanVideo-related, but one adds broader T2V/I2V support while the other adds official I2V model support; distinct features, not one bug/change." + }, + { + "left": "pull_request:6105", + "right": "pull_request:6752", + "accept": false, + "reason": "Different community pipelines with no shared concrete fix or feature." + }, + { + "left": "pull_request:2994", + "right": "pull_request:826", + "accept": false, + "reason": "Both involve numeric casting, but they fix different training scripts and different failure modes." + }, + { + "left": "pull_request:4659", + "right": "pull_request:4666", + "accept": false, + "reason": "They both touch SDXL/controlnet areas, but one is a broader pipeline/docs change and the other adds LoRA support for SDXL ControlNets; not the same PR-worthy change." + }, + { + "left": "pull_request:4706", + "right": "pull_request:5506", + "accept": false, + "reason": "LoRA alpha fallback and PEFT/xformers attention handling are different bugs." + }, + { + "left": "pull_request:5823", + "right": "pull_request:6225", + "accept": false, + "reason": "Both are LoRA/PEFT-related, but adapter fusion support and correct alpha parsing are separate changes." + }, + { + "left": "pull_request:6893", + "right": "pull_request:7044", + "accept": false, + "reason": "These are different UNet refactor efforts; not the same concrete change." + }, + { + "left": "pull_request:7200", + "right": "pull_request:7786", + "accept": false, + "reason": "Transformers2D norm-type validation and SVD video prompt handling are unrelated." + }, + { + "left": "pull_request:6133", + "right": "pull_request:6303", + "accept": false, + "reason": "UFOGen scheduler/pipeline support and ADD training scripts are unrelated features." + }, + { + "left": "pull_request:7115", + "right": "pull_request:7155", + "accept": false, + "reason": "Both touch Docker/CI, but one removes uv from a TPU Dockerfile and the other pins uv plus changes Slack notification logic; not the same change." + }, + { + "left": "pull_request:12681", + "right": "pull_request:13296", + "accept": false, + "reason": "CPU test stabilization and release-test transformer version updates are unrelated CI fixes." + }, + { + "left": "pull_request:10318", + "right": "pull_request:9244", + "accept": false, + "reason": "Flux LoRA integration tests and Flux single-file pipeline support are different changes." + }, + { + "left": "pull_request:11797", + "right": "pull_request:12207", + "accept": false, + "reason": "Higgs quantization and Nunchaku incorporation are different quantization backends and code paths." + }, + { + "left": "pull_request:1250", + "right": "pull_request:2037", + "accept": false, + "reason": "Different community pipelines; no shared underlying bug or feature." + }, + { + "left": "pull_request:13323", + "right": "pull_request:13364", + "accept": false, + "reason": "A test-device check fix and a MotionConv2d dtype cast fix are unrelated." + }, + { + "left": "pull_request:1527", + "right": "pull_request:473", + "accept": false, + "reason": "Both touch inpainting, but one broadens tensor inputs while the other fixes a sample/doc issue; not the same bug." + }, + { + "left": "pull_request:2415", + "right": "pull_request:2921", + "accept": false, + "reason": "Pipeline loader version parsing and stable diffusion pipeline offload changes are unrelated." + }, + { + "left": "pull_request:3245", + "right": "pull_request:4684", + "accept": false, + "reason": "These are translation/documentation updates in different locales, not duplicates." + }, + { + "left": "pull_request:3471", + "right": "pull_request:4315", + "accept": false, + "reason": "Dreambooth docs note and torch.compile graph-break fix are different artifacts and not the same code change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 83, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10378", + "pull_request:11743", + "pull_request:11995", + "pull_request:12445", + "pull_request:13042", + "pull_request:13330", + "pull_request:2001", + "pull_request:2693", + "pull_request:3299", + "pull_request:333", + "pull_request:349", + "pull_request:3921", + "pull_request:3924", + "pull_request:3995", + "pull_request:4132", + "pull_request:4237", + "pull_request:4242", + "pull_request:4498", + "pull_request:4499", + "pull_request:4783", + "pull_request:4859", + "pull_request:4862", + "pull_request:5015", + "pull_request:5056", + "pull_request:5102", + "pull_request:5151", + "pull_request:5186", + "pull_request:5202", + "pull_request:5386", + "pull_request:5506", + "pull_request:5807", + "pull_request:5833", + "pull_request:5842", + "pull_request:5898", + "pull_request:5943", + "pull_request:6009", + "pull_request:6035", + "pull_request:6105", + "pull_request:6293", + "pull_request:6311", + "pull_request:6645", + "pull_request:6787", + "pull_request:6960", + "pull_request:7206", + "pull_request:7246", + "pull_request:7411", + "pull_request:7632" + ], + "soft_pairs": [ + "pull_request:3995|pull_request:4237", + "pull_request:5015|pull_request:5898", + "pull_request:5386|pull_request:7411", + "pull_request:5506|pull_request:6009", + "pull_request:5807|pull_request:6645", + "pull_request:5833|pull_request:7632", + "pull_request:5842|pull_request:6787", + "pull_request:6035|pull_request:7206", + "pull_request:6293|pull_request:6960", + "pull_request:7246|pull_request:7632", + "pull_request:5151|pull_request:5202", + "pull_request:11995|pull_request:12445", + "pull_request:5102|pull_request:5202", + "pull_request:10378|pull_request:11743", + "pull_request:13042|pull_request:13330", + "pull_request:2001|pull_request:3299", + "pull_request:2693|pull_request:3924", + "pull_request:333|pull_request:349", + "pull_request:3921|pull_request:6105", + "pull_request:4132|pull_request:4242", + "pull_request:4498|pull_request:5943", + "pull_request:4499|pull_request:4783", + "pull_request:4499|pull_request:5186", + "pull_request:4859|pull_request:4862", + "pull_request:5056|pull_request:6311" + ], + "prepared_review_unit_hash": "f45aa27f79ccf83853b7d33b333a91a43eba0f09953c40b700f5d7989ec9f112", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 100594, + "estimated_input_tokens": 25149, + "estimated_eval_tokens": 50554 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicate cluster is present here: the soft pairs are mostly superficial similarities (shared files, same subsystem, or docs/tests), but they address different bugs or unrelated changes. I would reject every soft edge.", + "confidence": 0.98, + "canonical_issue_reason": "No issue artifacts are present in this cluster.", + "canonical_pr_reason": "No single PR is a valid canonical duplicate target; the PRs span unrelated fixes, features, docs, tests, and examples rather than one underlying change.", + "best_issue_reason": "No issue artifact is suitable as a global canonical for this cluster.", + "best_pr_reason": "No PR stands out as a canonical representative because the items are heterogeneous and do not share one concrete code-path problem or change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3995", + "right": "pull_request:4237", + "accept": false, + "reason": "Both touch SDXL prompt encoding, but one adds prompt embeddings while the other fixes a dtype bug; different changes and not mergeable as one PR." + }, + { + "left": "pull_request:5015", + "right": "pull_request:5898", + "accept": false, + "reason": "One fixes a broken docs link, the other rewrites unconditional-image docs; same area, but unrelated edits." + }, + { + "left": "pull_request:5386", + "right": "pull_request:7411", + "accept": false, + "reason": "Both are typo/doc cleanup PRs, but they touch different files and have no shared underlying bug/change." + }, + { + "left": "pull_request:5506", + "right": "pull_request:6009", + "accept": false, + "reason": "PEFT/xFormers attention scaling fix vs TensorRT constructor argument fix; unrelated code paths." + }, + { + "left": "pull_request:5807", + "right": "pull_request:6645", + "accept": false, + "reason": "Community pipeline code vs community README housekeeping; not the same change." + }, + { + "left": "pull_request:5833", + "right": "pull_request:7632", + "accept": false, + "reason": "PixArt docs about low-VRAM usage vs DreamBooth VAE loading bug; unrelated issues." + }, + { + "left": "pull_request:5842", + "right": "pull_request:6787", + "accept": false, + "reason": "Both are model-specific bug fixes, but for different pipelines and different failure modes." + }, + { + "left": "pull_request:6035", + "right": "pull_request:7206", + "accept": false, + "reason": "One updates help text in an SDXL advanced training script; the other adds training tags across many scripts. Same file overlap is incidental." + }, + { + "left": "pull_request:6293", + "right": "pull_request:6960", + "accept": false, + "reason": "IP-Adapter feature work for SDXL ControlNet vs a flaky IP-Adapter test fix; different scopes and code changes." + }, + { + "left": "pull_request:7246", + "right": "pull_request:7632", + "accept": false, + "reason": "VAE tiling test fix vs DreamBooth VAE loading bug; not the same defect." + }, + { + "left": "pull_request:5151", + "right": "pull_request:5202", + "accept": false, + "reason": "Both relate to LoRA/PEFT, but one integrates PEFT into UNet while the other adds Kohya checkpoint mapping support; different concrete changes." + }, + { + "left": "pull_request:11995", + "right": "pull_request:12445", + "accept": false, + "reason": "Both modify Flux modular pipeline files, but one introduces the pipeline and the other aligns it with Qwen modular; not a single bug fix." + }, + { + "left": "pull_request:5102", + "right": "pull_request:5202", + "accept": false, + "reason": "LyCORIS support and Kohya checkpoint support are distinct loader enhancements." + }, + { + "left": "pull_request:10378", + "right": "pull_request:11743", + "accept": false, + "reason": "DreamBooth LoRA DeepSpeed support vs a shortened Flux quantized training script; unrelated examples." + }, + { + "left": "pull_request:13042", + "right": "pull_request:13330", + "accept": false, + "reason": "QwenImage prompt isolation test fix vs torchao test/CI adjustment; different artifacts and behavior." + }, + { + "left": "pull_request:2001", + "right": "pull_request:3299", + "accept": false, + "reason": "ColossalAI typo fix vs LoRA docs/readme sync; unrelated changes." + }, + { + "left": "pull_request:2693", + "right": "pull_request:3924", + "accept": false, + "reason": "Inpaint doc image link fix vs training tutorial parameter update; no shared underlying issue." + }, + { + "left": "pull_request:333", + "right": "pull_request:349", + "accept": false, + "reason": "Two separate type-hinting PRs for different classes/modules; not the same change." + }, + { + "left": "pull_request:3921", + "right": "pull_request:6105", + "accept": false, + "reason": "A new community pipeline implementation vs a different community pipeline addition; same section, but distinct changes." + }, + { + "left": "pull_request:4132", + "right": "pull_request:4242", + "accept": false, + "reason": "Both touch single-file loading/conversion, but one makes accelerate optional while the other adds VAE loading; different fixes." + }, + { + "left": "pull_request:4498", + "right": "pull_request:5943", + "accept": false, + "reason": "SDXL reference pipeline support vs gradient checkpointing in Transformer2DModel; unrelated." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4783", + "accept": false, + "reason": "Reference-attention feature work vs docs for forcing embeddings to zero; different changes." + }, + { + "left": "pull_request:4499", + "right": "pull_request:5186", + "accept": false, + "reason": "Reference-attention feature work vs a FreeU proof-of-concept; unrelated." + }, + { + "left": "pull_request:4859", + "right": "pull_request:4862", + "accept": false, + "reason": "One fixes an SDXL inpaint test, the other fixes ControlNet img2img with MultiControlNetModel; different bugs." + }, + { + "left": "pull_request:5056", + "right": "pull_request:6311", + "accept": false, + "reason": "Multiple-LoRA docs vs a fast diffusion tutorial; both documentation, but not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 84, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10206", + "pull_request:10310", + "pull_request:10508", + "pull_request:10578", + "pull_request:10648", + "pull_request:10728", + "pull_request:10854", + "pull_request:10938", + "pull_request:11099", + "pull_request:11292", + "pull_request:11473", + "pull_request:11508", + "pull_request:11522", + "pull_request:11744", + "pull_request:11755", + "pull_request:11756", + "pull_request:11789", + "pull_request:12231", + "pull_request:12397", + "pull_request:12418", + "pull_request:12551", + "pull_request:1256", + "pull_request:12775", + "pull_request:12921", + "pull_request:13056", + "pull_request:13296", + "pull_request:13344", + "pull_request:1400", + "pull_request:1945", + "pull_request:2486", + "pull_request:4481", + "pull_request:4930", + "pull_request:5792", + "pull_request:5834", + "pull_request:5915", + "pull_request:6633", + "pull_request:6764", + "pull_request:6828", + "pull_request:6946", + "pull_request:7016", + "pull_request:7235", + "pull_request:8282", + "pull_request:8660", + "pull_request:8702", + "pull_request:9536", + "pull_request:9696", + "pull_request:9775" + ], + "soft_pairs": [ + "pull_request:5792|pull_request:5834", + "pull_request:6633|pull_request:8660", + "pull_request:6764|pull_request:8282", + "pull_request:7235|pull_request:8702", + "pull_request:9536|pull_request:9696", + "pull_request:5915|pull_request:6946", + "pull_request:1400|pull_request:1945", + "pull_request:4481|pull_request:4930", + "pull_request:12418|pull_request:12551", + "pull_request:5915|pull_request:7016", + "pull_request:6828|pull_request:7016", + "pull_request:10206|pull_request:11292", + "pull_request:10310|pull_request:11508", + "pull_request:10508|pull_request:11099", + "pull_request:10578|pull_request:10728", + "pull_request:10648|pull_request:11744", + "pull_request:10854|pull_request:11755", + "pull_request:10938|pull_request:9775", + "pull_request:11473|pull_request:12231", + "pull_request:11522|pull_request:12397", + "pull_request:11756|pull_request:11789", + "pull_request:12418|pull_request:13296", + "pull_request:1256|pull_request:2486", + "pull_request:12775|pull_request:13056", + "pull_request:12921|pull_request:13344" + ], + "prepared_review_unit_hash": "0260c7aecdab7058d1f5480763ae76958f0d473266aa31a21ec8d224bdd2de1b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 101723, + "estimated_input_tokens": 25431, + "estimated_eval_tokens": 51118 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are mostly superficial matches across unrelated PRs (docs, CI, tests, and separate feature work). None look like the same underlying change or bug fix, so I would not merge any of them as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out: the items span many unrelated changes, so there is no shared underlying change to centralize.", + "best_issue_reason": null, + "best_pr_reason": "No best PR for the cluster overall; the PRs are heterogeneous and do not form one duplicate set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5792", + "right": "pull_request:5834", + "accept": false, + "reason": "Both are import-related fixes, but they touch different loading paths and solve different problems." + }, + { + "left": "pull_request:6633", + "right": "pull_request:8660", + "accept": false, + "reason": "Unrelated changes: one fixes original checkpoint YAML reading, the other adds a max_sequence_length guard in SD3 controlnet input validation." + }, + { + "left": "pull_request:6764", + "right": "pull_request:8282", + "accept": false, + "reason": "Both are documentation edits, but they cover different topics and different pages." + }, + { + "left": "pull_request:7235", + "right": "pull_request:8702", + "accept": false, + "reason": "Different scopes entirely: DreamBooth DoRA docs/support versus an SD3 conversion script." + }, + { + "left": "pull_request:9536", + "right": "pull_request:9696", + "accept": false, + "reason": "Tokenizer padding API tweak vs CI workflow changes; not the same code-path or fix." + }, + { + "left": "pull_request:5915", + "right": "pull_request:6946", + "accept": false, + "reason": "Same broad IP-Adapter area, but one adds IP-Adapter Plus support and the other adds low_cpu_mem_usage loading; distinct changes." + }, + { + "left": "pull_request:1400", + "right": "pull_request:1945", + "accept": false, + "reason": "One updates Versatile Diffusion attention parameters; the other adds UNet block tests. Different artifacts and goals." + }, + { + "left": "pull_request:4481", + "right": "pull_request:4930", + "accept": false, + "reason": "EDM pipeline/scheduler implementation versus Heun scheduler improvements; not the same concrete change." + }, + { + "left": "pull_request:12418", + "right": "pull_request:12551", + "accept": false, + "reason": "Both are CI/test stability tweaks, but they target different failing tests and unrelated skips/xfails." + }, + { + "left": "pull_request:5915", + "right": "pull_request:7016", + "accept": false, + "reason": "IP-Adapter Plus support and IP-Adapter embedding refactor are related broadly, but they are separate feature changes." + }, + { + "left": "pull_request:6828", + "right": "pull_request:7016", + "accept": false, + "reason": "UNetMotionModel IP-Adapter code change versus a wider IP-Adapter embedding refactor; not the same bug or patch." + }, + { + "left": "pull_request:10206", + "right": "pull_request:11292", + "accept": false, + "reason": "Different LoRA changes: Flux unload behavior vs more SDXL LoRA conversion support." + }, + { + "left": "pull_request:10310", + "right": "pull_request:11508", + "accept": false, + "reason": "Both affect CI workflows, but one gates fast GPU tests on PRs while the other changes torch.compile CI/test setup." + }, + { + "left": "pull_request:10508", + "right": "pull_request:11099", + "accept": false, + "reason": "Device-agnostic tests and PR-GPU workflow conditioning are unrelated test infrastructure changes." + }, + { + "left": "pull_request:10578", + "right": "pull_request:10728", + "accept": false, + "reason": "Both mention quantized Flux/IP-Adapter, but they solve different concrete problems in different loaders." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11744", + "accept": false, + "reason": "Docs-only scheduler update versus DreamBooth LoRA metadata/alpha support; unrelated changes." + }, + { + "left": "pull_request:10854", + "right": "pull_request:11755", + "accept": false, + "reason": "One renames Hunyuan DiT test modules, the other fixes LTX CPU offload behavior; different code paths." + }, + { + "left": "pull_request:10938", + "right": "pull_request:9775", + "accept": false, + "reason": "Different documentation updates with no shared underlying bug or change." + }, + { + "left": "pull_request:11473", + "right": "pull_request:12231", + "accept": false, + "reason": "A specific torchao docs typo fix versus a broad typo cleanup across community examples." + }, + { + "left": "pull_request:11522", + "right": "pull_request:12397", + "accept": false, + "reason": "A pipeline compatibility fix for transformers main versus CI changes disabling transformers-from-main installs; related environment, but not the same change." + }, + { + "left": "pull_request:11756", + "right": "pull_request:11789", + "accept": false, + "reason": "Both are test fixes, but they target different pipelines and different numerical/device issues." + }, + { + "left": "pull_request:12418", + "right": "pull_request:13296", + "accept": false, + "reason": "One xfails a failing pipeline test; the other updates transformers version handling in release tests. Separate CI fixes." + }, + { + "left": "pull_request:1256", + "right": "pull_request:2486", + "accept": false, + "reason": "Negative prompt test adjustment and mixin skip relocation are unrelated test changes." + }, + { + "left": "pull_request:12775", + "right": "pull_request:13056", + "accept": false, + "reason": "Z-Image docs and Z-Image LoRA training are related by model family, but not the same underlying change." + }, + { + "left": "pull_request:12921", + "right": "pull_request:13344", + "accept": false, + "reason": "The latter adds GLM Image model tests and a small model metadata tweak, but it is not the same change as introducing GLM Image support itself." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 85, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1561", + "pull_request:1668", + "pull_request:1687", + "pull_request:2415", + "pull_request:2746", + "pull_request:2767", + "pull_request:288", + "pull_request:2974", + "pull_request:3197", + "pull_request:3266", + "pull_request:3474", + "pull_request:3568", + "pull_request:3839", + "pull_request:3979", + "pull_request:3987", + "pull_request:401", + "pull_request:4211", + "pull_request:4227", + "pull_request:4546", + "pull_request:4762", + "pull_request:4862", + "pull_request:4934", + "pull_request:5051", + "pull_request:5144", + "pull_request:5275", + "pull_request:5367", + "pull_request:5386", + "pull_request:5499", + "pull_request:5506", + "pull_request:5626", + "pull_request:5702", + "pull_request:5741", + "pull_request:5982", + "pull_request:6035", + "pull_request:6140", + "pull_request:6216", + "pull_request:6291", + "pull_request:6578", + "pull_request:6795", + "pull_request:6808", + "pull_request:6875", + "pull_request:7141", + "pull_request:7335", + "pull_request:7857", + "pull_request:7916", + "pull_request:806", + "pull_request:837", + "pull_request:874" + ], + "soft_pairs": [ + "pull_request:1561|pull_request:2974", + "pull_request:1668|pull_request:1687", + "pull_request:1687|pull_request:806", + "pull_request:2415|pull_request:2767", + "pull_request:2746|pull_request:837", + "pull_request:288|pull_request:401", + "pull_request:2974|pull_request:874", + "pull_request:3197|pull_request:3266", + "pull_request:3474|pull_request:4227", + "pull_request:3568|pull_request:5499", + "pull_request:3839|pull_request:4546", + "pull_request:3979|pull_request:5702", + "pull_request:3987|pull_request:5144", + "pull_request:4211|pull_request:5367", + "pull_request:4762|pull_request:6795", + "pull_request:4862|pull_request:6578", + "pull_request:4934|pull_request:6035", + "pull_request:5051|pull_request:5386", + "pull_request:5051|pull_request:7141", + "pull_request:5275|pull_request:5982", + "pull_request:5506|pull_request:5741", + "pull_request:5626|pull_request:6808", + "pull_request:5982|pull_request:7916", + "pull_request:6140|pull_request:7857", + "pull_request:6216|pull_request:6875", + "pull_request:6291|pull_request:7335" + ], + "prepared_review_unit_hash": "36a3c495659de281c23601aa30979d21bbc1d58b2c5601fa423c523f753293c2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 94227, + "estimated_input_tokens": 23557, + "estimated_eval_tokens": 47370 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly unrelated small docs, example, test, and bugfix PRs. No soft edge looks like the same underlying change, so I rejected them all. The most representative PR in the set is the broad typo/docs cleanup PR.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 5386 is the broadest repo-wide cleanup in the set, touching many docs/examples with the same minor-fix character as several candidates.", + "best_issue_reason": null, + "best_pr_reason": "PR 5386 is the best representative for this cluster because it is a wide, merged cleanup PR and most closely matches the general theme of small documentation fixes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1561", + "right": "pull_request:2974", + "accept": false, + "reason": "Different changes: Stable Diffusion Mega pipeline vs TensorRT community pipeline/docs. Same area (community pipelines) is too broad." + }, + { + "left": "pull_request:1668", + "right": "pull_request:1687", + "accept": false, + "reason": "Related checkpointing theme, but 1668 is Dreambooth save/restore state while 1687 extends checkpointing to other scripts; not the same PR-level change." + }, + { + "left": "pull_request:1687", + "right": "pull_request:806", + "accept": false, + "reason": "Both touch Dreambooth, but one is checkpointing rollout and the other is an older WIP EMA/reporting change. Different underlying changes." + }, + { + "left": "pull_request:2415", + "right": "pull_request:2767", + "accept": false, + "reason": "Unrelated code paths: pipeline loading/version handling vs LoRA training script PEFT flag usage." + }, + { + "left": "pull_request:2746", + "right": "pull_request:837", + "accept": false, + "reason": "Both are docs-only fixes, but one is fp16 prose and the other is a speedup-ratio correction. Not the same change." + }, + { + "left": "pull_request:288", + "right": "pull_request:401", + "accept": false, + "reason": "Type hints for a pipeline implementation vs docs for the same pipeline family. Different artifact types and scope." + }, + { + "left": "pull_request:2974", + "right": "pull_request:874", + "accept": false, + "reason": "Both affect community README content, but one adds a TensorRT pipeline and the other adds a generic inference example. Not duplicate." + }, + { + "left": "pull_request:3197", + "right": "pull_request:3266", + "accept": false, + "reason": "One is a large lpw_stable_diffusion pipeline update; the other fixes dynamic module version loading. No shared bug/change." + }, + { + "left": "pull_request:3474", + "right": "pull_request:4227", + "accept": false, + "reason": "Docs for attention processors vs a missing import in textual inversion README. Different topics and files." + }, + { + "left": "pull_request:3568", + "right": "pull_request:5499", + "accept": false, + "reason": "Both fix broken links, but in different documentation pages and with different link targets. Too generic to call duplicate." + }, + { + "left": "pull_request:3839", + "right": "pull_request:4546", + "accept": false, + "reason": "Scheduler add_noise bugfix vs a typo in a conversion script docstring. Completely different changes." + }, + { + "left": "pull_request:3979", + "right": "pull_request:5702", + "accept": false, + "reason": "Different bugs in different pipelines: SDXL attention/precision logic vs AnimateDiff prompt default handling." + }, + { + "left": "pull_request:3987", + "right": "pull_request:5144", + "accept": false, + "reason": "Separate documentation snippet fixes in unrelated pipeline pages." + }, + { + "left": "pull_request:4211", + "right": "pull_request:5367", + "accept": false, + "reason": "SDXL docstring cleanup in two pipeline files vs a tutorial code creation fix. Not the same underlying change." + }, + { + "left": "pull_request:4762", + "right": "pull_request:6795", + "accept": false, + "reason": "Different SDXL training docs/examples changes: VAE variable naming vs micro-conditioning/crop args." + }, + { + "left": "pull_request:4862", + "right": "pull_request:6578", + "accept": false, + "reason": "ControlNet SDXL MultiControlNet bugfix vs a docs heading indentation fix. No overlap in actual change." + }, + { + "left": "pull_request:4934", + "right": "pull_request:6035", + "accept": false, + "reason": "T2I-Adapter training script/docs vs advanced DreamBooth LoRA help text. Same training umbrella, different changes." + }, + { + "left": "pull_request:5051", + "right": "pull_request:5386", + "accept": false, + "reason": "Both are typo/docs cleanups, but 5051 is Wuerstchen-specific while 5386 is broad repo-wide cleanup. Not the same fix." + }, + { + "left": "pull_request:5051", + "right": "pull_request:7141", + "accept": false, + "reason": "Docs typo cleanup vs Stable Video Diffusion step numbering. Different pages and issues." + }, + { + "left": "pull_request:5275", + "right": "pull_request:5982", + "accept": false, + "reason": "Different features entirely: SDXL Flax example vs SDXL Turbo docs." + }, + { + "left": "pull_request:5506", + "right": "pull_request:5741", + "accept": false, + "reason": "Same PEFT subsystem, but one fixes xformers+LoRA scaling and the other speeds up LoRA unload. Different concrete problems." + }, + { + "left": "pull_request:5626", + "right": "pull_request:6808", + "accept": false, + "reason": "Different tests for different environments: CPU offload assertion vs MPS test collection fix." + }, + { + "left": "pull_request:5982", + "right": "pull_request:7916", + "accept": false, + "reason": "SDXL Turbo docs vs Ascend NPU support in the SDXL training script. No shared change." + }, + { + "left": "pull_request:6140", + "right": "pull_request:7857", + "accept": false, + "reason": "Both add docs pages, but for unrelated features (IP-Adapter vs PixArtSigma). Not duplicate." + }, + { + "left": "pull_request:6216", + "right": "pull_request:6875", + "accept": false, + "reason": "Both are test adjustments, but for different features and failure modes: PEFT version handling vs IP-Adapter output changes." + }, + { + "left": "pull_request:6291", + "right": "pull_request:7335", + "accept": false, + "reason": "LoRA unfusion test fix vs ControlNet SDXL config test fix. Different tests and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 86, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10589", + "pull_request:10630", + "pull_request:10728", + "pull_request:10970", + "pull_request:11273", + "pull_request:11572", + "pull_request:11715", + "pull_request:12021", + "pull_request:12142", + "pull_request:12143", + "pull_request:12687", + "pull_request:2919", + "pull_request:3214", + "pull_request:3586", + "pull_request:3847", + "pull_request:4099", + "pull_request:4242", + "pull_request:4417", + "pull_request:4629", + "pull_request:4886", + "pull_request:6147", + "pull_request:6342", + "pull_request:6573", + "pull_request:6724", + "pull_request:6808", + "pull_request:6828", + "pull_request:7050", + "pull_request:7124", + "pull_request:7291", + "pull_request:7411", + "pull_request:7487", + "pull_request:7542", + "pull_request:7747", + "pull_request:7827", + "pull_request:7868", + "pull_request:8272", + "pull_request:8490", + "pull_request:8571", + "pull_request:8583", + "pull_request:8630", + "pull_request:8715", + "pull_request:8775", + "pull_request:8860", + "pull_request:9008", + "pull_request:9100", + "pull_request:9257", + "pull_request:9807" + ], + "soft_pairs": [ + "pull_request:6342|pull_request:6724", + "pull_request:6808|pull_request:7542", + "pull_request:7050|pull_request:8571", + "pull_request:7124|pull_request:8715", + "pull_request:7291|pull_request:7411", + "pull_request:7487|pull_request:8715", + "pull_request:7542|pull_request:9100", + "pull_request:7747|pull_request:7868", + "pull_request:7827|pull_request:7868", + "pull_request:8272|pull_request:9008", + "pull_request:8490|pull_request:8775", + "pull_request:8583|pull_request:8630", + "pull_request:8775|pull_request:8860", + "pull_request:9008|pull_request:9100", + "pull_request:9257|pull_request:9807", + "pull_request:11572|pull_request:12021", + "pull_request:10589|pull_request:10728", + "pull_request:6573|pull_request:6828", + "pull_request:10630|pull_request:11715", + "pull_request:10970|pull_request:11273", + "pull_request:12142|pull_request:12687", + "pull_request:12143|pull_request:12687", + "pull_request:2919|pull_request:3847", + "pull_request:3214|pull_request:4886", + "pull_request:3586|pull_request:4417", + "pull_request:4099|pull_request:4629", + "pull_request:4242|pull_request:6147" + ], + "prepared_review_unit_hash": "9133728ef1099a1d9fbdd0ae299cc80573331d1d3627b893a85fa1406a93315f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 96457, + "estimated_input_tokens": 24115, + "estimated_eval_tokens": 48486 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous set of small, unrelated docs/tests/style or feature PRs. The only clear duplicate-like pair is the two Wan VACE bf16/fp16 mask-casting fixes.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12143 is the cleaner canonical for the Wan VACE dtype fix: it keeps the mask cast in the right place and adds regression coverage, while 12142 is the earlier near-duplicate variant.", + "best_issue_reason": null, + "best_pr_reason": "PR 12143 is the best representative of the only true duplicate pair in this set because it fixes the same code path more completely and includes tests.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6342", + "right": "pull_request:6724", + "accept": false, + "reason": "Unrelated docs updates: installation guidance vs LoRA rank wording." + }, + { + "left": "pull_request:6808", + "right": "pull_request:7542", + "accept": false, + "reason": "Both are test fixes, but they address different failures and code paths." + }, + { + "left": "pull_request:7050", + "right": "pull_request:8571", + "accept": false, + "reason": "Typos in example docs/tests versus a broad style/errata PR; not the same change." + }, + { + "left": "pull_request:7124", + "right": "pull_request:8715", + "accept": false, + "reason": "Different DreamBooth scripts and different fixes: a kwarg rename vs adding clip_skip support." + }, + { + "left": "pull_request:7291", + "right": "pull_request:7411", + "accept": false, + "reason": "Documentation typo fix in UNet2DConditionModel vs unrelated typo cleanup across many schedulers/docs." + }, + { + "left": "pull_request:7487", + "right": "pull_request:8715", + "accept": false, + "reason": "Different features: SDXL DreamBooth full-weight training vs clip_skip argument for advanced LoRA." + }, + { + "left": "pull_request:7542", + "right": "pull_request:9100", + "accept": false, + "reason": "Different test typo/usage fixes in different files; no shared bug." + }, + { + "left": "pull_request:7747", + "right": "pull_request:7868", + "accept": false, + "reason": "Same test file, but one fixes VAE tiling assertions and the other fixes offload-device expectations." + }, + { + "left": "pull_request:7827", + "right": "pull_request:7868", + "accept": false, + "reason": "Docstring CPU-offload example fix is unrelated to the offload test assertion change." + }, + { + "left": "pull_request:8272", + "right": "pull_request:9008", + "accept": false, + "reason": "Grammar cleanup in scheduler error messages versus broad typo/style cleanup across workflows/docs." + }, + { + "left": "pull_request:8490", + "right": "pull_request:8775", + "accept": false, + "reason": "Both touch README content, but one fixes SD3 DreamBooth docs and the other adds a Colab link; different changes." + }, + { + "left": "pull_request:8583", + "right": "pull_request:8630", + "accept": false, + "reason": "Different SD3 attention backend support vs SD3 DreamBooth LoRA text-encoder training." + }, + { + "left": "pull_request:8775", + "right": "pull_request:8860", + "accept": false, + "reason": "README link update vs a separate README edit with no evidence of the same underlying change." + }, + { + "left": "pull_request:9008", + "right": "pull_request:9100", + "accept": false, + "reason": "Broad typo/style cleanup across many files versus a single typo fix in one training script." + }, + { + "left": "pull_request:9257", + "right": "pull_request:9807", + "accept": false, + "reason": "Different DreamBooth/Flux adapter fixes in the same script family, but not the same bug or code path." + }, + { + "left": "pull_request:11572", + "right": "pull_request:12021", + "accept": false, + "reason": "Different LTX work: LoRA conversion support vs attention backend support." + }, + { + "left": "pull_request:10589", + "right": "pull_request:10728", + "accept": false, + "reason": "Both mention IP-Adapter, but they fix different pipelines and code paths (SD3 img2img vs Flux quantization loader)." + }, + { + "left": "pull_request:6573", + "right": "pull_request:6828", + "accept": false, + "reason": "IP-Adapter support in different components; one is multi-adapter plumbing, the other UNetMotionModel handling." + }, + { + "left": "pull_request:10630", + "right": "pull_request:11715", + "accept": false, + "reason": "Both adjust tests for bitsandbytes/compilation, but they target different test classes and versions." + }, + { + "left": "pull_request:10970", + "right": "pull_request:11273", + "accept": false, + "reason": "Different areas entirely: workflow automation vs Slack notification timeout handling." + }, + { + "left": "pull_request:12142", + "right": "pull_request:12687", + "accept": false, + "reason": "Different bugs and files: Wan VACE mask dtype issue vs a typo rename in autoencoder_dc." + }, + { + "left": "pull_request:12143", + "right": "pull_request:12687", + "accept": false, + "reason": "Same as above; unrelated Wan VACE fix versus autoencoder_dc typo correction." + }, + { + "left": "pull_request:2919", + "right": "pull_request:3847", + "accept": false, + "reason": "Test speedup in UNet3DConditionModel versus dependency-check workflow; unrelated." + }, + { + "left": "pull_request:3214", + "right": "pull_request:4886", + "accept": false, + "reason": "Attention reshape bug versus upscale pipeline safety-checker behavior; different code paths." + }, + { + "left": "pull_request:3586", + "right": "pull_request:4417", + "accept": false, + "reason": "Community pipeline addition versus Zero123 support/docs PR; not the same change." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4629", + "accept": false, + "reason": "SDXL prompt embedding bug fix versus unrelated documentation removal/addition work." + }, + { + "left": "pull_request:4242", + "right": "pull_request:6147", + "accept": false, + "reason": "Both touch single-file loading, but they fix different model-loading cases (VAE loading vs SDXL inpainting/refiner args)." + }, + { + "left": "pull_request:12142", + "right": "pull_request:12143", + "accept": true, + "reason": "Both patch the same Wan VACE bf16/fp16 mask-casting bug in `pipeline_wan_vace.py`; 12143 is the more complete version with the cast placed correctly and test coverage." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 87, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10318", + "pull_request:10431", + "pull_request:10517", + "pull_request:10547", + "pull_request:10810", + "pull_request:11524", + "pull_request:12420", + "pull_request:12782", + "pull_request:13039", + "pull_request:13330", + "pull_request:1977", + "pull_request:2464", + "pull_request:2655", + "pull_request:406", + "pull_request:453", + "pull_request:4666", + "pull_request:4787", + "pull_request:4847", + "pull_request:5015", + "pull_request:5499", + "pull_request:5559", + "pull_request:5698", + "pull_request:5738", + "pull_request:5887", + "pull_request:5928", + "pull_request:6208", + "pull_request:6217", + "pull_request:6489", + "pull_request:6523", + "pull_request:6673", + "pull_request:6721", + "pull_request:6752", + "pull_request:6757", + "pull_request:689", + "pull_request:6904", + "pull_request:7016", + "pull_request:7186", + "pull_request:7246", + "pull_request:7507", + "pull_request:7820", + "pull_request:823", + "pull_request:8564", + "pull_request:8696", + "pull_request:8749", + "pull_request:9021", + "pull_request:9244", + "pull_request:9682", + "pull_request:9696" + ], + "soft_pairs": [ + "pull_request:4666|pull_request:5738", + "pull_request:6208|pull_request:6904", + "pull_request:6217|pull_request:7246", + "pull_request:6721|pull_request:6757", + "pull_request:689|pull_request:823", + "pull_request:5887|pull_request:7820", + "pull_request:5887|pull_request:7186", + "pull_request:5559|pull_request:6673", + "pull_request:5887|pull_request:7016", + "pull_request:10431|pull_request:10517", + "pull_request:10547|pull_request:9696", + "pull_request:11524|pull_request:12420", + "pull_request:1977|pull_request:2464", + "pull_request:2655|pull_request:4787", + "pull_request:406|pull_request:453", + "pull_request:5015|pull_request:5499", + "pull_request:5559|pull_request:7507", + "pull_request:5698|pull_request:6523", + "pull_request:5928|pull_request:6489", + "pull_request:8749|pull_request:9021", + "pull_request:4847|pull_request:5559", + "pull_request:6489|pull_request:6752", + "pull_request:8564|pull_request:9682", + "pull_request:10318|pull_request:8696", + "pull_request:10810|pull_request:9244", + "pull_request:12782|pull_request:13330", + "pull_request:13039|pull_request:13330" + ], + "prepared_review_unit_hash": "63394693f3674c15c5ea2d1cc1940376f5a6cb682b16f8f1ef589fb439eacff9", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 96639, + "estimated_input_tokens": 24160, + "estimated_eval_tokens": 48576 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are unrelated PRs touching different subsystems, so there is no clear duplicate canonical. All soft edges are rejected under a strict same-bug/same-change standard.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:4666", + "right": "pull_request:5738", + "accept": false, + "reason": "Different LoRA work: SDXL ControlNet inpaint LoRA support vs LoRA adapter deletion. Same broad area, but not the same underlying change or bug." + }, + { + "left": "pull_request:6208", + "right": "pull_request:6904", + "accept": false, + "reason": "Unrelated fixes: a dtype mismatch in Stable Diffusion decoding vs allowing PIL.Image input in load_image." + }, + { + "left": "pull_request:6217", + "right": "pull_request:7246", + "accept": false, + "reason": "Different problems: torch compile/VAE signature handling vs VAE tiling test device selection." + }, + { + "left": "pull_request:6721", + "right": "pull_request:6757", + "accept": false, + "reason": "AnimateDiff SDXL support/docs versus moving sigmas to device in SDXL k-diffusion. Not the same code path or bug." + }, + { + "left": "pull_request:689", + "right": "pull_request:823", + "accept": false, + "reason": "Slow attention test/performance rewrite vs negative-prompt type-mismatch fixes. Completely different changes." + }, + { + "left": "pull_request:5887", + "right": "pull_request:7820", + "accept": false, + "reason": "Both involve IP-Adapter, but they fix different pipeline integrations (ControlNet inpaint vs InstructPix2Pix) and are not the same concrete bug." + }, + { + "left": "pull_request:5887", + "right": "pull_request:7186", + "accept": false, + "reason": "IP-Adapter support in ControlNet inpaint vs moving Face ID support to core. Different features and files." + }, + { + "left": "pull_request:5559", + "right": "pull_request:6673", + "accept": false, + "reason": "Latent Consistency Model Img2Img pipeline addition vs InstantID pipeline addition. Separate features." + }, + { + "left": "pull_request:5887", + "right": "pull_request:7016", + "accept": false, + "reason": "7016 is a broad refactor of IP-Adapter image embedding handling across many pipelines; 5887 is a specific ControlNet inpaint IP-Adapter integration. Not the same bug/change." + }, + { + "left": "pull_request:10431", + "right": "pull_request:10517", + "accept": false, + "reason": "Same Sana family, but one fixes 1k positional embeddings and the other a 4K pag pipeline bug. Distinct issues." + }, + { + "left": "pull_request:10547", + "right": "pull_request:9696", + "accept": false, + "reason": "Offload test adjustments for CogVideoX/CogView3 vs workflow changes for running tests from a PR. Unrelated." + }, + { + "left": "pull_request:11524", + "right": "pull_request:12420", + "accept": false, + "reason": "Conditional torchvision import in Cosmos transformer vs conditional torch.distributed import in attention dispatch. Different subsystems." + }, + { + "left": "pull_request:1977", + "right": "pull_request:2464", + "accept": false, + "reason": "Broad training/script updates vs a specific xformers-related LoRA training cleanup. Not the same change." + }, + { + "left": "pull_request:2655", + "right": "pull_request:4787", + "accept": false, + "reason": "Both touch LoRA loading, but 2655 fixes save/load behavior generally while 4787 narrows weight-name selection. Different defects." + }, + { + "left": "pull_request:406", + "right": "pull_request:453", + "accept": false, + "reason": "Two separate documentation typos in different files." + }, + { + "left": "pull_request:5015", + "right": "pull_request:5499", + "accept": false, + "reason": "Specific broken-link fix vs a wider docs link cleanup. Related area, but not the same change." + }, + { + "left": "pull_request:5559", + "right": "pull_request:7507", + "accept": false, + "reason": "Different community pipelines: Latent Consistency Img2Img vs InstantID SDXL Img2Img." + }, + { + "left": "pull_request:5698", + "right": "pull_request:6523", + "accept": false, + "reason": "LCM img2img pipeline fix vs LCM distillation script bug. Same project family, different code paths and bugs." + }, + { + "left": "pull_request:5928", + "right": "pull_request:6489", + "accept": false, + "reason": "Different community pipeline additions with separate functionality." + }, + { + "left": "pull_request:8749", + "right": "pull_request:9021", + "accept": false, + "reason": "9021 is a follow-up fix inside from_transformer, but 8749 is the broader initial SD3 ControlNet from_transformer support. Not the same exact bug/change." + }, + { + "left": "pull_request:4847", + "right": "pull_request:5559", + "accept": false, + "reason": "Prompt Free Diffusion community example vs LCM Img2Img pipeline. Unrelated features." + }, + { + "left": "pull_request:6489", + "right": "pull_request:6752", + "accept": false, + "reason": "StyleAligned pipeline vs PromptDiffusion pipeline. Different community additions." + }, + { + "left": "pull_request:8564", + "right": "pull_request:9682", + "accept": false, + "reason": "Both concern SD3 training, but 8564 is a T5 token-limit update while 9682 is a broader dreambooth-lora bug-fix patch." + }, + { + "left": "pull_request:10318", + "right": "pull_request:8696", + "accept": false, + "reason": "Flux LoRA integration tests vs Marigold flaky-test annotations. No overlap in underlying change." + }, + { + "left": "pull_request:10810", + "right": "pull_request:9244", + "accept": false, + "reason": "Flux LoRA conversion/support improvements vs adding single-file support for Flux pipelines. Different changes." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13330", + "accept": false, + "reason": "HunyuanVideo offloading test settings vs torchao test fixes. Different areas and failures." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13330", + "accept": false, + "reason": "WAN CPU upcasting fix vs torchao test fixes. Separate bugs and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 88, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10378", + "pull_request:10750", + "pull_request:10943", + "pull_request:11655", + "pull_request:12055", + "pull_request:12063", + "pull_request:12383", + "pull_request:13237", + "pull_request:1788", + "pull_request:2009", + "pull_request:2415", + "pull_request:2915", + "pull_request:3294", + "pull_request:3339", + "pull_request:343", + "pull_request:349", + "pull_request:4397", + "pull_request:4499", + "pull_request:4706", + "pull_request:473", + "pull_request:4749", + "pull_request:4761", + "pull_request:4855", + "pull_request:5506", + "pull_request:5668", + "pull_request:5697", + "pull_request:5823", + "pull_request:5880", + "pull_request:5915", + "pull_request:5943", + "pull_request:6231", + "pull_request:6752", + "pull_request:6757", + "pull_request:6952", + "pull_request:7153", + "pull_request:7156", + "pull_request:7627", + "pull_request:7692", + "pull_request:7790", + "pull_request:7820", + "pull_request:8452", + "pull_request:8902", + "pull_request:8913", + "pull_request:8932", + "pull_request:9060", + "pull_request:9244", + "pull_request:9257", + "pull_request:9662" + ], + "soft_pairs": [ + "pull_request:1788|pull_request:473", + "pull_request:2415|pull_request:3339", + "pull_request:2915|pull_request:4761", + "pull_request:343|pull_request:349", + "pull_request:4397|pull_request:4749", + "pull_request:4499|pull_request:5943", + "pull_request:4855|pull_request:5697", + "pull_request:5506|pull_request:6231", + "pull_request:5697|pull_request:5823", + "pull_request:6952|pull_request:9060", + "pull_request:7153|pull_request:8913", + "pull_request:7156|pull_request:9257", + "pull_request:8452|pull_request:9244", + "pull_request:8932|pull_request:9662", + "pull_request:12383|pull_request:13237", + "pull_request:5915|pull_request:7820", + "pull_request:10378|pull_request:10943", + "pull_request:2009|pull_request:3294", + "pull_request:4706|pull_request:5697", + "pull_request:6757|pull_request:7790", + "pull_request:7627|pull_request:8902", + "pull_request:12055|pull_request:12063", + "pull_request:5668|pull_request:5880", + "pull_request:6752|pull_request:7692", + "pull_request:10750|pull_request:11655" + ], + "prepared_review_unit_hash": "dbf7482a7b2665e97b728106c8955a5356c60974a6382919a1992caec9b2c2e5", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 100025, + "estimated_input_tokens": 25007, + "estimated_eval_tokens": 50270 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft edges are mostly superficial similarities: same subsystem, same file, or similar feature area, but not the same concrete bug/change. No pair looks safe to merge as a duplicate, so there is no clear canonical PR for this set.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1788", + "right": "pull_request:473", + "accept": false, + "reason": "DDIM scheduler sample-prediction bug fix vs Stable Diffusion inpainting sample/docs ONNX handling; unrelated code paths." + }, + { + "left": "pull_request:2415", + "right": "pull_request:3339", + "accept": false, + "reason": "Pipeline version check guard vs setup.py urllib3 pin; different maintenance concerns." + }, + { + "left": "pull_request:2915", + "right": "pull_request:4761", + "accept": false, + "reason": "Textual inversion test tolerance tweak vs Paint-by-Example encoder init fix; unrelated changes." + }, + { + "left": "pull_request:343", + "right": "pull_request:349", + "accept": false, + "reason": "Both are scheduler type-hint updates, but for different classes (DDIM vs DDPM) and not the same bug or change." + }, + { + "left": "pull_request:4397", + "right": "pull_request:4749", + "accept": false, + "reason": "SDXL docstring default update vs SDXL Dreambooth LoRA checkpointing fix; unrelated." + }, + { + "left": "pull_request:4499", + "right": "pull_request:5943", + "accept": false, + "reason": "Community SDXL reference-attn example vs Transformer2D gradient checkpointing support; different features." + }, + { + "left": "pull_request:4855", + "right": "pull_request:5697", + "accept": false, + "reason": "Textual inversion offloading hook handling vs PEFT/xformers attention scale bug; different failure modes." + }, + { + "left": "pull_request:5506", + "right": "pull_request:6231", + "accept": false, + "reason": "PEFT/xformers attention fix vs fp16 LoRA training upcast fix; not the same underlying issue." + }, + { + "left": "pull_request:5697", + "right": "pull_request:5823", + "accept": false, + "reason": "xformers PEFT LoRA bugfix vs fuse_lora adapter_names feature/documentation; different changes." + }, + { + "left": "pull_request:6952", + "right": "pull_request:9060", + "accept": false, + "reason": "Textual inversion model-card update vs removing a debug print in Flux transformer; unrelated." + }, + { + "left": "pull_request:7153", + "right": "pull_request:8913", + "accept": false, + "reason": "Docs workflow path filter fix vs sharded checkpoint subfolder loading bug; different areas." + }, + { + "left": "pull_request:7156", + "right": "pull_request:9257", + "accept": false, + "reason": "Stalebot label normalization vs Dreambooth Flux tokenization/encoding bug fix; unrelated." + }, + { + "left": "pull_request:8452", + "right": "pull_request:9244", + "accept": false, + "reason": "Single-file loading strictness fix vs Flux single-file pipeline support; different code paths and scope." + }, + { + "left": "pull_request:8932", + "right": "pull_request:9662", + "accept": false, + "reason": "Temporal transformer test coverage vs LoRA low_cpu_mem_usage version requirement bump; unrelated." + }, + { + "left": "pull_request:12383", + "right": "pull_request:13237", + "accept": false, + "reason": "Both touch single_file_utils, but one adds modular single-file support while the other only extends FLUX.2 Klein model mappings; not the same change." + }, + { + "left": "pull_request:5915", + "right": "pull_request:7820", + "accept": false, + "reason": "IP-Adapter Plus support vs instruct-pix2pix IP-adapter fix; related subsystem but different concrete problems." + }, + { + "left": "pull_request:10378", + "right": "pull_request:10943", + "accept": false, + "reason": "SD3 LoRA training/save-hook changes vs Wan LoRA support; distinct model families and features." + }, + { + "left": "pull_request:2009", + "right": "pull_request:3294", + "accept": false, + "reason": "Textual inversion loader integration vs LoRA prefix naming support; both loader-related but not the same bug or feature." + }, + { + "left": "pull_request:4706", + "right": "pull_request:5697", + "accept": false, + "reason": "Fallback alpha handling in LoRA MLP loading vs xformers PEFT attention bug; different fixes." + }, + { + "left": "pull_request:6757", + "right": "pull_request:7790", + "accept": false, + "reason": "SDXL k-diffusion sigma device placement vs masked img2img SDXL community pipeline adaptation; unrelated." + }, + { + "left": "pull_request:7627", + "right": "pull_request:8902", + "accept": false, + "reason": "Playground img2img latents mean/std handling vs sharded checkpoint loading path fix; unrelated." + }, + { + "left": "pull_request:12055", + "right": "pull_request:12063", + "accept": false, + "reason": "Qwen-Image model addition vs Qwen-Image VAE license header update; same project area but entirely different change." + }, + { + "left": "pull_request:5668", + "right": "pull_request:5880", + "accept": false, + "reason": "Transformers-optional compatibility work vs VAE config/latent-diffusion support; different underlying problems." + }, + { + "left": "pull_request:6752", + "right": "pull_request:7692", + "accept": false, + "reason": "PromptDiffusion community pipeline vs PAG community pipeline; both community examples, but different algorithms and codebases." + }, + { + "left": "pull_request:10750", + "right": "pull_request:11655", + "accept": false, + "reason": "Flux ControlNet img2img support vs bnb 8-bit Flux Control LoRA support; same model family, but not the same bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 89, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10005", + "pull_request:10099", + "pull_request:10109", + "pull_request:10183", + "pull_request:1020", + "pull_request:10232", + "pull_request:10246", + "pull_request:10250", + "pull_request:10253", + "pull_request:10255", + "pull_request:1027", + "pull_request:10273", + "pull_request:10275", + "pull_request:10276", + "pull_request:10299", + "pull_request:10312", + "pull_request:10316", + "pull_request:10319", + "pull_request:10322", + "pull_request:10332", + "pull_request:10365", + "pull_request:10394", + "pull_request:10429", + "pull_request:10531", + "pull_request:10538", + "pull_request:10585", + "pull_request:10819", + "pull_request:10858", + "pull_request:10938", + "pull_request:10978", + "pull_request:11105", + "pull_request:11248", + "pull_request:11303", + "pull_request:13187", + "pull_request:13217", + "pull_request:2887", + "pull_request:2897", + "pull_request:8675", + "pull_request:8775", + "pull_request:8845", + "pull_request:8851", + "pull_request:9015", + "pull_request:910", + "pull_request:9107", + "pull_request:9266", + "pull_request:9481", + "pull_request:9553", + "pull_request:9760" + ], + "soft_pairs": [ + "pull_request:13187|pull_request:13217", + "pull_request:10005|pull_request:10531", + "pull_request:10099|pull_request:8845", + "pull_request:10109|pull_request:11105", + "pull_request:10183|pull_request:10365", + "pull_request:10183|pull_request:11105", + "pull_request:1020|pull_request:910", + "pull_request:10232|pull_request:10938", + "pull_request:10232|pull_request:11248", + "pull_request:10232|pull_request:11303", + "pull_request:10232|pull_request:8775", + "pull_request:10232|pull_request:9553", + "pull_request:10246|pull_request:8851", + "pull_request:10250|pull_request:10312", + "pull_request:10253|pull_request:8675", + "pull_request:10255|pull_request:10978", + "pull_request:1027|pull_request:2887", + "pull_request:1027|pull_request:2897", + "pull_request:10273|pull_request:9481", + "pull_request:10275|pull_request:10276", + "pull_request:10276|pull_request:10585", + "pull_request:10299|pull_request:9266", + "pull_request:10299|pull_request:9760", + "pull_request:10312|pull_request:10429", + "pull_request:10316|pull_request:10858", + "pull_request:10319|pull_request:9015", + "pull_request:10322|pull_request:10538", + "pull_request:10332|pull_request:10819", + "pull_request:10394|pull_request:10978", + "pull_request:10394|pull_request:9107" + ], + "prepared_review_unit_hash": "625d24d9cf05dded7fdea12cf5f61687a0f13ea11f728d0b49af8e096aa554a1", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 93279, + "estimated_input_tokens": 23320, + "estimated_eval_tokens": 46896 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: mostly unrelated docs, test, and feature PRs with a few superficial filename overlaps. I would not merge any of the suggested soft edges as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out as a duplicate representative; the items span unrelated changes across docs, tests, loaders, and pipelines.", + "best_issue_reason": null, + "best_pr_reason": "PR 13217 is the strongest anchor if one must pick one artifact: it is the largest substantive change, adds LTX-2.3 support across many files, and has the broadest review/inbound activity.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13187", + "right": "pull_request:13217", + "accept": false, + "reason": "Both touch LTX2 image/video, but 13187 fixes two-stage latent masking while 13217 adds LTX-2.3 model support across many components." + }, + { + "left": "pull_request:10005", + "right": "pull_request:10531", + "accept": false, + "reason": "Both are docs PRs, but one adds missing flux pipelines and the other fixes negative-prompt docs for Flux/SD3; different changes." + }, + { + "left": "pull_request:10099", + "right": "pull_request:8845", + "accept": false, + "reason": "Both are typo fixes in tests, but they target different test files and different issues." + }, + { + "left": "pull_request:10109", + "right": "pull_request:11105", + "accept": false, + "reason": "Stable Audio XLA support and AutoencoderKLWan gradient checkpointing are unrelated code paths and changes." + }, + { + "left": "pull_request:10183", + "right": "pull_request:10365", + "accept": false, + "reason": "AutoencoderDC single-file support and AuraFlow XLA support are unrelated features in different modules." + }, + { + "left": "pull_request:10183", + "right": "pull_request:11105", + "accept": false, + "reason": "Single-file AutoencoderDC support is unrelated to gradient checkpointing in AutoencoderKLWan." + }, + { + "left": "pull_request:1020", + "right": "pull_request:910", + "accept": false, + "reason": "Both are documentation updates, but they cover different docs and different topics." + }, + { + "left": "pull_request:10232", + "right": "pull_request:10938", + "accept": false, + "reason": "Minor Sana docs correction vs evaluation conceptual docs update; no shared underlying bug or change." + }, + { + "left": "pull_request:10232", + "right": "pull_request:11248", + "accept": false, + "reason": "Sana docs correction and a Flux controlnet README fix are unrelated documentation changes." + }, + { + "left": "pull_request:10232", + "right": "pull_request:11303", + "accept": false, + "reason": "Sana docs correction and AutoencoderKLAllegro docs correction are unrelated." + }, + { + "left": "pull_request:10232", + "right": "pull_request:8775", + "accept": false, + "reason": "Sana docs correction and a Colab-link README update are different docs tasks." + }, + { + "left": "pull_request:10232", + "right": "pull_request:9553", + "accept": false, + "reason": "These update different docs pages for different purposes; not the same change." + }, + { + "left": "pull_request:10246", + "right": "pull_request:8851", + "accept": false, + "reason": "Attention processor docs additions and AuraFlow docs additions are separate documentation expansions." + }, + { + "left": "pull_request:10250", + "right": "pull_request:10312", + "accept": false, + "reason": "Different bugs in different model components: downsample assignment in UNet blocks vs attention reshape behavior." + }, + { + "left": "pull_request:10253", + "right": "pull_request:8675", + "accept": false, + "reason": "Contributor-note docs and fast-diffusion benchmarking docs are unrelated." + }, + { + "left": "pull_request:10255", + "right": "pull_request:10978", + "accept": false, + "reason": "A Mochi comment/docstring fix is unrelated to the OneTrainer Flux LoRA loading bug." + }, + { + "left": "pull_request:1027", + "right": "pull_request:2887", + "accept": false, + "reason": "Both affect training examples, but one adds a CLI argument and the other changes validation sampling behavior." + }, + { + "left": "pull_request:1027", + "right": "pull_request:2897", + "accept": false, + "reason": "DDPM example CLI changes and Stable UnCLIP docs additions are unrelated." + }, + { + "left": "pull_request:10273", + "right": "pull_request:9481", + "accept": false, + "reason": "Both refactor LoRA tests, but they target different test structure changes and are not the same fix." + }, + { + "left": "pull_request:10275", + "right": "pull_request:10276", + "accept": false, + "reason": "Licensing header fixes in pipelines and a README filename typo fix are unrelated." + }, + { + "left": "pull_request:10276", + "right": "pull_request:10585", + "accept": false, + "reason": "README rename typo fix and Mochi VAE annotation fix are different changes." + }, + { + "left": "pull_request:10299", + "right": "pull_request:9266", + "accept": false, + "reason": "GGUF docs generator handling and community projects listing are unrelated docs edits." + }, + { + "left": "pull_request:10299", + "right": "pull_request:9760", + "accept": false, + "reason": "GGUF docs generator change and README wording cleanup are not the same underlying change." + }, + { + "left": "pull_request:10312", + "right": "pull_request:10429", + "accept": false, + "reason": "One fixes attention reshape docs/behavior, the other adjusts rotary embedding dtype for NPU; different bugs." + }, + { + "left": "pull_request:10316", + "right": "pull_request:10858", + "accept": false, + "reason": "LTX video docs generator removal and pipelines overview classification are unrelated documentation updates." + }, + { + "left": "pull_request:10319", + "right": "pull_request:9015", + "accept": false, + "reason": "Docstring typo fix in HunyuanVideo and PIA example rewrite are unrelated." + }, + { + "left": "pull_request:10322", + "right": "pull_request:10538", + "accept": false, + "reason": "Sana converter/model support and auto-pipeline registration are different pieces of Sana support, not one duplicate change." + }, + { + "left": "pull_request:10332", + "right": "pull_request:10819", + "accept": false, + "reason": "Model-loading error-message fix and LoRA docs entries are unrelated." + }, + { + "left": "pull_request:10394", + "right": "pull_request:10978", + "accept": false, + "reason": "Peft docs link fix and Flux LoRA loading bug fix are unrelated." + }, + { + "left": "pull_request:10394", + "right": "pull_request:9107", + "accept": false, + "reason": "Docs link correction and SD3 DreamBooth LoRA loading-hook fix are different changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 90, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10420", + "pull_request:10494", + "pull_request:10524", + "pull_request:10531", + "pull_request:10544", + "pull_request:10585", + "pull_request:10632", + "pull_request:10648", + "pull_request:10678", + "pull_request:10730", + "pull_request:10802", + "pull_request:10819", + "pull_request:10858", + "pull_request:1087", + "pull_request:10885", + "pull_request:10941", + "pull_request:10978", + "pull_request:10990", + "pull_request:10998", + "pull_request:11027", + "pull_request:11029", + "pull_request:1103", + "pull_request:11037", + "pull_request:11257", + "pull_request:11271", + "pull_request:11473", + "pull_request:11558", + "pull_request:11595", + "pull_request:11600", + "pull_request:11701", + "pull_request:11747", + "pull_request:11778", + "pull_request:11802", + "pull_request:11851", + "pull_request:11857", + "pull_request:11954", + "pull_request:12228", + "pull_request:260", + "pull_request:268", + "pull_request:43", + "pull_request:8947", + "pull_request:8965", + "pull_request:9066", + "pull_request:9473", + "pull_request:9598", + "pull_request:9705", + "pull_request:9765" + ], + "soft_pairs": [ + "pull_request:10420|pull_request:10524", + "pull_request:10420|pull_request:11271", + "pull_request:10494|pull_request:11802", + "pull_request:10524|pull_request:10858", + "pull_request:10531|pull_request:8947", + "pull_request:10531|pull_request:9066", + "pull_request:10544|pull_request:9473", + "pull_request:10585|pull_request:10678", + "pull_request:10585|pull_request:10998", + "pull_request:10585|pull_request:8965", + "pull_request:10632|pull_request:11857", + "pull_request:10648|pull_request:11851", + "pull_request:10678|pull_request:11778", + "pull_request:10678|pull_request:9765", + "pull_request:10730|pull_request:9473", + "pull_request:10802|pull_request:11473", + "pull_request:10819|pull_request:11747", + "pull_request:1087|pull_request:268", + "pull_request:10885|pull_request:11851", + "pull_request:10941|pull_request:11595", + "pull_request:10978|pull_request:11600", + "pull_request:10978|pull_request:12228", + "pull_request:10990|pull_request:9598", + "pull_request:10990|pull_request:9705", + "pull_request:10998|pull_request:11778", + "pull_request:11027|pull_request:9765", + "pull_request:11029|pull_request:11257", + "pull_request:11029|pull_request:11701", + "pull_request:1103|pull_request:260", + "pull_request:1103|pull_request:43", + "pull_request:11037|pull_request:11558", + "pull_request:11037|pull_request:11954" + ], + "prepared_review_unit_hash": "724fc7c441a00b4ca95d247913900c5e7cc636fa1ef1524b6593ecd0e2199f60", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 90928, + "estimated_input_tokens": 22732, + "estimated_eval_tokens": 45720 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated doc typos, CI/workflow tweaks, and small pipeline/model fixes. The soft links are superficial (same docs area, same workflow file, same broad LoRA/Flux/Wan context) rather than the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 10990 is the strongest anchor candidate: it fixes a concrete Flux LoRA quantization path, spans code plus tests, and has explicit issue linkage. Still, it is not a duplicate representative for the whole mixed cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10420", + "right": "pull_request:10524", + "accept": false, + "reason": "Both are docs-only checkpoint-id edits, but they target different video pipelines (LTX vs Hunyuan) and different model IDs." + }, + { + "left": "pull_request:10420", + "right": "pull_request:11271", + "accept": false, + "reason": "LTX checkpoint-id docs fix is unrelated to the single-file LTX 0.9.5 mapping/code-path change." + }, + { + "left": "pull_request:10494", + "right": "pull_request:11802", + "accept": false, + "reason": "Different CI/doc concerns: tokenizers install in LoRA tests versus LoRA deprecation cleanup in lora_base." + }, + { + "left": "pull_request:10524", + "right": "pull_request:10858", + "accept": false, + "reason": "Both are docs updates, but one fixes Hunyuan checkpoint docs and the other corrects the I2VGen-XL pipeline category." + }, + { + "left": "pull_request:10531", + "right": "pull_request:8947", + "accept": false, + "reason": "Docstring negative-prompt cleanup for SD3/Flux is unrelated to latent-upscale prompt-embed support." + }, + { + "left": "pull_request:10531", + "right": "pull_request:9066", + "accept": false, + "reason": "Different documentation targets: negative-prompt docs versus adding StableCascadeUNet docs." + }, + { + "left": "pull_request:10544", + "right": "pull_request:9473", + "accept": false, + "reason": "ParaAttention documentation is unrelated to the Flux DreamBooth learning-rate bug fix." + }, + { + "left": "pull_request:10585", + "right": "pull_request:10678", + "accept": false, + "reason": "Mochi VAE annotation fix and Auraflow input validation fix address different pipelines and different issues." + }, + { + "left": "pull_request:10585", + "right": "pull_request:10998", + "accept": false, + "reason": "Mochi VAE typing/docs cleanup is unrelated to Wan VAE scaling relocation." + }, + { + "left": "pull_request:10585", + "right": "pull_request:8965", + "accept": false, + "reason": "Mochi VAE annotation fix and pipeline instantiation rename are unrelated changes." + }, + { + "left": "pull_request:10632", + "right": "pull_request:11857", + "accept": false, + "reason": "Both touch docs, but one fixes a ParaAttention image path and the other updates Flux pipeline references/descriptions." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11851", + "accept": false, + "reason": "DDPM scheduler docs for sigmoid schedule are unrelated to the fp32-module note in schedulers docs." + }, + { + "left": "pull_request:10678", + "right": "pull_request:11778", + "accept": false, + "reason": "Auraflow height/width validation is unrelated to Wan VACE mask padding." + }, + { + "left": "pull_request:10678", + "right": "pull_request:9765", + "accept": false, + "reason": "Auraflow input checks and SD3 docstring corrections are different pipelines and different bug classes." + }, + { + "left": "pull_request:10730", + "right": "pull_request:9473", + "accept": false, + "reason": "Hook-unwrapping fix in pipeline_utils is unrelated to the Flux DreamBooth LR change." + }, + { + "left": "pull_request:10802", + "right": "pull_request:11473", + "accept": false, + "reason": "A typo fix in a training example is unrelated to torchao fp8 documentation corrections." + }, + { + "left": "pull_request:10819", + "right": "pull_request:11747", + "accept": false, + "reason": "LoRA docs entries addition is unrelated to adding the HiDream license headers." + }, + { + "left": "pull_request:1087", + "right": "pull_request:268", + "accept": false, + "reason": "These are separate historical slow-test fixes touching different code paths; not the same regression." + }, + { + "left": "pull_request:10885", + "right": "pull_request:11851", + "accept": false, + "reason": "Adding CogVideoX scheduler docs is unrelated to the _keep_in_fp32_modules note in schedulers docs." + }, + { + "left": "pull_request:10941", + "right": "pull_request:11595", + "accept": false, + "reason": "Flux copy-fix in IP adapter/img2img/inpaint is unrelated to the LoRA load_state_dict logic and test update." + }, + { + "left": "pull_request:10978", + "right": "pull_request:11600", + "accept": false, + "reason": "One fixes OneTrainer Flux LoRA conversion; the other only renames CI failure-report filenames." + }, + { + "left": "pull_request:10978", + "right": "pull_request:12228", + "accept": false, + "reason": "Flux LoRA conversion bug and a LoRA tutorial typo fix are unrelated." + }, + { + "left": "pull_request:10990", + "right": "pull_request:9598", + "accept": false, + "reason": "Flux quantized LoRA support is unrelated to a Flux LoRA integration-test note about torch/CUDA versions." + }, + { + "left": "pull_request:10990", + "right": "pull_request:9705", + "accept": false, + "reason": "Flux quantized LoRA code is unrelated to README wording tweaks for an advanced Flux LoRA tutorial." + }, + { + "left": "pull_request:10998", + "right": "pull_request:11778", + "accept": false, + "reason": "Wan VAE scaling move and Wan VACE mask padding fix are different pipeline bugs." + }, + { + "left": "pull_request:11027", + "right": "pull_request:9765", + "accept": false, + "reason": "IP-Adapter feature-extractor placement fix is unrelated to SD3 docstring cleanup." + }, + { + "left": "pull_request:11029", + "right": "pull_request:11257", + "accept": false, + "reason": "Both touch the style-bot workflow, but they change different secrets and are not the same fix." + }, + { + "left": "pull_request:11029", + "right": "pull_request:11701", + "accept": false, + "reason": "Same workflow file, but one adds make fix-copies and the other swaps to a different bot token." + }, + { + "left": "pull_request:1103", + "right": "pull_request:260", + "accept": false, + "reason": "These are unrelated README typo fixes from different eras and different files." + }, + { + "left": "pull_request:1103", + "right": "pull_request:43", + "accept": false, + "reason": "README typo fix and model-card template introduction are not the same change." + }, + { + "left": "pull_request:11037", + "right": "pull_request:11558", + "accept": false, + "reason": "Nightly test dependency wiring is unrelated to the layerwise-upcasting/group-offloading test addition." + }, + { + "left": "pull_request:11037", + "right": "pull_request:11954", + "accept": false, + "reason": "Quantization CI dependency changes are unrelated to Hunyuan Video test slice updates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 91, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1104", + "pull_request:11053", + "pull_request:11099", + "pull_request:11187", + "pull_request:1123", + "pull_request:11259", + "pull_request:11262", + "pull_request:113", + "pull_request:1130", + "pull_request:1131", + "pull_request:11325", + "pull_request:11365", + "pull_request:11386", + "pull_request:11391", + "pull_request:11416", + "pull_request:11434", + "pull_request:11439", + "pull_request:11473", + "pull_request:11476", + "pull_request:11558", + "pull_request:1157", + "pull_request:11576", + "pull_request:1162", + "pull_request:11731", + "pull_request:11920", + "pull_request:11955", + "pull_request:11983", + "pull_request:12018", + "pull_request:12230", + "pull_request:12247", + "pull_request:12360", + "pull_request:12391", + "pull_request:12487", + "pull_request:12552", + "pull_request:1394", + "pull_request:1561", + "pull_request:1640", + "pull_request:1769", + "pull_request:2010", + "pull_request:2257", + "pull_request:2766", + "pull_request:531", + "pull_request:60", + "pull_request:688", + "pull_request:715", + "pull_request:874", + "pull_request:9760" + ], + "soft_pairs": [ + "pull_request:1104|pull_request:2766", + "pull_request:11053|pull_request:11187", + "pull_request:11099|pull_request:11259", + "pull_request:11187|pull_request:11920", + "pull_request:1123|pull_request:1640", + "pull_request:1123|pull_request:715", + "pull_request:11262|pull_request:11434", + "pull_request:113|pull_request:1769", + "pull_request:113|pull_request:2010", + "pull_request:113|pull_request:531", + "pull_request:113|pull_request:60", + "pull_request:1130|pull_request:2257", + "pull_request:1131|pull_request:1394", + "pull_request:11325|pull_request:12018", + "pull_request:11365|pull_request:12230", + "pull_request:11386|pull_request:11983", + "pull_request:11386|pull_request:9760", + "pull_request:11391|pull_request:11983", + "pull_request:11416|pull_request:12487", + "pull_request:11439|pull_request:12247", + "pull_request:11473|pull_request:12360", + "pull_request:11473|pull_request:12391", + "pull_request:11476|pull_request:12552", + "pull_request:11558|pull_request:11920", + "pull_request:11558|pull_request:11955", + "pull_request:1157|pull_request:688", + "pull_request:11576|pull_request:11731", + "pull_request:1162|pull_request:1561", + "pull_request:1162|pull_request:874" + ], + "prepared_review_unit_hash": "d35e7614e71e25fb6115f477ec1700e059ef22477cb2c227097f5384257e2f49", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 93239, + "estimated_input_tokens": 23310, + "estimated_eval_tokens": 46876 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab-bag of unrelated PRs across CI, docs, tests, model code, and release/version maintenance. The soft similarities are mostly lexical (e.g. \u201cfix\u201d, \u201cdocs\u201d, \u201ctests\u201d) and do not indicate the same underlying change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1104", + "right": "pull_request:2766", + "accept": false, + "reason": "Both touch tests and MPS-related behavior, but one adds nightly slow MPS CI and the other fixes expected outputs in unit tests; different changes." + }, + { + "left": "pull_request:11053", + "right": "pull_request:11187", + "accept": false, + "reason": "Wan Video2Video pipeline addition vs adding CacheMixin to Wan/LTX transformers; separate code paths and goals." + }, + { + "left": "pull_request:11099", + "right": "pull_request:11259", + "accept": false, + "reason": "CI quality gating change vs a dummy styling-trigger PR; not the same change." + }, + { + "left": "pull_request:11187", + "right": "pull_request:11920", + "accept": false, + "reason": "Transformer cache support vs Wan test slices; implementation feature versus test fixture update." + }, + { + "left": "pull_request:1123", + "right": "pull_request:1640", + "accept": false, + "reason": "MPS setup docs vs automatic xformers enablement in training scripts; unrelated subsystems." + }, + { + "left": "pull_request:1123", + "right": "pull_request:715", + "accept": false, + "reason": "MPS documentation update vs import guarding when PyTorch is absent; different problems." + }, + { + "left": "pull_request:11262", + "right": "pull_request:11434", + "accept": false, + "reason": "Wan ftfy import guard vs a test import cleanup; unrelated fixes." + }, + { + "left": "pull_request:113", + "right": "pull_request:1769", + "accept": false, + "reason": "SDE-VE scheduler initialization fix vs UnCLIP offload test threshold change; different code and behavior." + }, + { + "left": "pull_request:113", + "right": "pull_request:2010", + "accept": false, + "reason": "Scheduler initialization bug fix vs LR scheduler setup in an example training script; not the same issue." + }, + { + "left": "pull_request:113", + "right": "pull_request:531", + "accept": false, + "reason": "Scheduler init fix vs GPU attention test slice updates; unrelated." + }, + { + "left": "pull_request:113", + "right": "pull_request:60", + "accept": false, + "reason": "Both are model/scheduler-related, but they fix different code paths and bugs." + }, + { + "left": "pull_request:1130", + "right": "pull_request:2257", + "accept": false, + "reason": "LPW Stable Diffusion community pipeline tweaks vs UnCLIP text interpolation pipeline; different community features." + }, + { + "left": "pull_request:1131", + "right": "pull_request:1394", + "accept": false, + "reason": "Both are version bumps, but they are separate release bumps to different versions, not one merged change." + }, + { + "left": "pull_request:11325", + "right": "pull_request:12018", + "accept": false, + "reason": "HiDream docstring cleanup vs quantization docs link wording; unrelated doc updates." + }, + { + "left": "pull_request:11365", + "right": "pull_request:12230", + "accept": false, + "reason": "Wan conversion-script flow_shift logic vs Wan docs rename; different artifacts and intent." + }, + { + "left": "pull_request:11386", + "right": "pull_request:11983", + "accept": false, + "reason": "HiDream README requirements fix vs installation.md command correction; unrelated docs." + }, + { + "left": "pull_request:11386", + "right": "pull_request:9760", + "accept": false, + "reason": "HiDream README update vs general README title formatting; not the same content or bug." + }, + { + "left": "pull_request:11391", + "right": "pull_request:11983", + "accept": false, + "reason": "Dummy README_hidream PR vs installation docs update; no shared underlying change." + }, + { + "left": "pull_request:11416", + "right": "pull_request:12487", + "accept": false, + "reason": "Both are docs cleanup, but one is broad typo/comment fixes and the other is broken link fixes across different pages; not the same change." + }, + { + "left": "pull_request:11439", + "right": "pull_request:12247", + "accept": false, + "reason": "HiDream LoRA validation bug fix vs AutoGuidance validation logic fix; same word, different failure modes and code paths." + }, + { + "left": "pull_request:11473", + "right": "pull_request:12360", + "accept": false, + "reason": "TorchAO docs typo fix vs Chroma docs update; unrelated documentation changes." + }, + { + "left": "pull_request:11473", + "right": "pull_request:12391", + "accept": false, + "reason": "TorchAO typo correction vs quantization docs reorganization; same general area but different edits and scope." + }, + { + "left": "pull_request:11476", + "right": "pull_request:12552", + "accept": false, + "reason": "Generic typo cleanup across code/docs vs Kandinsky5 docs typo fixes; unrelated edit sets." + }, + { + "left": "pull_request:11558", + "right": "pull_request:11920", + "accept": false, + "reason": "Group offloading/layerwise casting tests vs Wan slice tests; different models and behaviors under test." + }, + { + "left": "pull_request:11558", + "right": "pull_request:11955", + "accept": false, + "reason": "Layerwise casting/group offloading tests vs Cosmos slice tests; unrelated test additions." + }, + { + "left": "pull_request:1157", + "right": "pull_request:688", + "accept": false, + "reason": "Flax training workflow fix vs Flax import handling without PyTorch; separate issues." + }, + { + "left": "pull_request:11576", + "right": "pull_request:11731", + "accept": false, + "reason": "Both add memory-optimization tips, but one is about quantization+offloading and the other about compile+offload; distinct guidance." + }, + { + "left": "pull_request:1162", + "right": "pull_request:1561", + "accept": false, + "reason": "Community pipeline link fixes vs adding the Stable Diffusion Mega community pipeline; unrelated changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 92, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11623", + "pull_request:11673", + "pull_request:11693", + "pull_request:11697", + "pull_request:1171", + "pull_request:11731", + "pull_request:11737", + "pull_request:1174", + "pull_request:11740", + "pull_request:1178", + "pull_request:1186", + "pull_request:1187", + "pull_request:11891", + "pull_request:1194", + "pull_request:11954", + "pull_request:11983", + "pull_request:11993", + "pull_request:12069", + "pull_request:12070", + "pull_request:1210", + "pull_request:12129", + "pull_request:1218", + "pull_request:12183", + "pull_request:12193", + "pull_request:1221", + "pull_request:12228", + "pull_request:12233", + "pull_request:12343", + "pull_request:12675", + "pull_request:12681", + "pull_request:12731", + "pull_request:12757", + "pull_request:12971", + "pull_request:1487", + "pull_request:1989", + "pull_request:2433", + "pull_request:2598", + "pull_request:260", + "pull_request:268", + "pull_request:3012", + "pull_request:353", + "pull_request:407", + "pull_request:61", + "pull_request:63", + "pull_request:688", + "pull_request:798", + "pull_request:972" + ], + "soft_pairs": [ + "pull_request:11623|pull_request:12129", + "pull_request:11623|pull_request:12228", + "pull_request:11623|pull_request:12343", + "pull_request:11623|pull_request:12675", + "pull_request:11673|pull_request:11737", + "pull_request:11693|pull_request:12183", + "pull_request:11697|pull_request:11740", + "pull_request:1171|pull_request:260", + "pull_request:1171|pull_request:353", + "pull_request:1171|pull_request:407", + "pull_request:1171|pull_request:798", + "pull_request:1171|pull_request:972", + "pull_request:11731|pull_request:12183", + "pull_request:11731|pull_request:12233", + "pull_request:1174|pull_request:1487", + "pull_request:1174|pull_request:2433", + "pull_request:1174|pull_request:61", + "pull_request:1174|pull_request:63", + "pull_request:1178|pull_request:260", + "pull_request:1178|pull_request:3012", + "pull_request:1178|pull_request:353", + "pull_request:1186|pull_request:688", + "pull_request:1187|pull_request:688", + "pull_request:11891|pull_request:11983", + "pull_request:1194|pull_request:1221", + "pull_request:11954|pull_request:12070", + "pull_request:11993|pull_request:12183", + "pull_request:12069|pull_request:12681", + "pull_request:1210|pull_request:268", + "pull_request:1218|pull_request:1989", + "pull_request:12193|pull_request:12731", + "pull_request:12193|pull_request:12757", + "pull_request:12193|pull_request:12971", + "pull_request:1221|pull_request:2598" + ], + "prepared_review_unit_hash": "5fc88e91429bd88685e89846fa013cfd43cd245c2e7e8ec9a18aeadc4e57a26c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 34, + "serialized_chars": 92373, + "estimated_input_tokens": 23094, + "estimated_eval_tokens": 46444 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mixed bag of unrelated tiny docs/typo updates plus a few distinct test or bugfix PRs; none of the soft pairs look like the same underlying change, so all are rejected. The best single representative PR is 12343 because it is the most concrete bugfix with an explicit issue target.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12343 is the strongest single representative: a concrete VAE slicing bugfix with an explicit issue target and a focused code-path change.", + "best_issue_reason": null, + "best_pr_reason": "PR 12343 is the most substantive and well-scoped PR in the set, so it is the best representative of the cluster even though the cluster is not truly duplicate-heavy.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11623", + "right": "pull_request:12129", + "accept": false, + "reason": "Both are small typo fixes, but they touch different code paths and unrelated text, so they are not the same change." + }, + { + "left": "pull_request:11623", + "right": "pull_request:12228", + "accept": false, + "reason": "One fixes a pipeline doc typo; the other fixes LoRA tutorial formatting and punctuation. Different edits, not duplicates." + }, + { + "left": "pull_request:11623", + "right": "pull_request:12343", + "accept": false, + "reason": "A typo fix in a pipeline doc is unrelated to the VAE slicing bugfix." + }, + { + "left": "pull_request:11623", + "right": "pull_request:12675", + "accept": false, + "reason": "These are unrelated doc typo/cleanup changes in different files and topics." + }, + { + "left": "pull_request:11673", + "right": "pull_request:11737", + "accept": false, + "reason": "HiDream expert-loss model changes and DeepSpeed support in the training script are different features." + }, + { + "left": "pull_request:11693", + "right": "pull_request:12183", + "accept": false, + "reason": "Both touch docs, but one is about bitsandbytes torch.compile guidance and the other adds a LoRA badge; unrelated changes." + }, + { + "left": "pull_request:11697", + "right": "pull_request:11740", + "accept": false, + "reason": "Both concern compile tests, but for different quantization backends and different test scopes." + }, + { + "left": "pull_request:1171", + "right": "pull_request:260", + "accept": false, + "reason": "These are typo fixes in different examples and different text; not the same underlying bug." + }, + { + "left": "pull_request:1171", + "right": "pull_request:353", + "accept": false, + "reason": "One is a broad typo sweep across many pipelines; the other fixes a specific error message string." + }, + { + "left": "pull_request:1171", + "right": "pull_request:407", + "accept": false, + "reason": "A typo sweep in docs/examples is not the same as changing the attention slicing API call in code." + }, + { + "left": "pull_request:1171", + "right": "pull_request:798", + "accept": false, + "reason": "These are separate docstring typo fixes in different components." + }, + { + "left": "pull_request:1171", + "right": "pull_request:972", + "accept": false, + "reason": "Different typo fixes in different files; no shared code-path bug." + }, + { + "left": "pull_request:11731", + "right": "pull_request:12183", + "accept": false, + "reason": "One adds a compile/offload docs tip, the other adds a LoRA badge to QwenImage docs; unrelated." + }, + { + "left": "pull_request:11731", + "right": "pull_request:12233", + "accept": false, + "reason": "Both are docs about compile/offload, but one adds a tip and the other renames a section title; not the same change." + }, + { + "left": "pull_request:1174", + "right": "pull_request:1487", + "accept": false, + "reason": "Both edit loading docs, but they are different documentation expansions with different content and intent." + }, + { + "left": "pull_request:1174", + "right": "pull_request:2433", + "accept": false, + "reason": "General loading docs expansion is unrelated to adding an SDEdit note for img2img." + }, + { + "left": "pull_request:1174", + "right": "pull_request:61", + "accept": false, + "reason": "A docs rewrite is unrelated to adding a conversion script for LatentDiffusionUncondPipeline." + }, + { + "left": "pull_request:1174", + "right": "pull_request:63", + "accept": false, + "reason": "A docs rewrite is unrelated to adding a BDDM conversion script." + }, + { + "left": "pull_request:1178", + "right": "pull_request:260", + "accept": false, + "reason": "These are separate typo fixes in different files and contexts." + }, + { + "left": "pull_request:1178", + "right": "pull_request:3012", + "accept": false, + "reason": "Different typo corrections in unrelated comments/docstrings." + }, + { + "left": "pull_request:1178", + "right": "pull_request:353", + "accept": false, + "reason": "A small config logger typo fix is unrelated to the UNet block error-message typo." + }, + { + "left": "pull_request:1186", + "right": "pull_request:688", + "accept": false, + "reason": "Flax Dreambooth training changes are unrelated to the Flax import-without-PyTorch fix." + }, + { + "left": "pull_request:1187", + "right": "pull_request:688", + "accept": false, + "reason": "These touch the same training script file family but fix different Flax push-to-hub behavior and import availability issues." + }, + { + "left": "pull_request:11891", + "right": "pull_request:11983", + "accept": false, + "reason": "One fixes a code example typo in Chroma docs; the other fixes installation instructions. Different docs, different issue." + }, + { + "left": "pull_request:1194", + "right": "pull_request:1221", + "accept": false, + "reason": "These are unrelated docs/workflow updates; one changes doc-builder actions, the other adds weight reuse docs." + }, + { + "left": "pull_request:11954", + "right": "pull_request:12070", + "accept": false, + "reason": "Different test fixtures and different pipelines; both are test-slice updates, not the same bug." + }, + { + "left": "pull_request:11993", + "right": "pull_request:12183", + "accept": false, + "reason": "A LoRA inference docs deep-dive is unrelated to adding a LoRA badge to QwenImage docs." + }, + { + "left": "pull_request:12069", + "right": "pull_request:12681", + "accept": false, + "reason": "One fixes AudioLDM2 cache-position handling; the other adjusts CPU test setups across several pipelines. Not the same change." + }, + { + "left": "pull_request:1210", + "right": "pull_request:268", + "accept": false, + "reason": "Both touch tests/pipeline behavior, but 1210 changes CycleDiffusion/schedulers for test speed while 268 updates expected outputs after VAE/pipeline behavior changes." + }, + { + "left": "pull_request:1218", + "right": "pull_request:1989", + "accept": false, + "reason": "Both modify the conversion script, but they add different features in different eras and are not the same bugfix." + }, + { + "left": "pull_request:12193", + "right": "pull_request:12731", + "accept": false, + "reason": "These are unrelated README/docs fixes for different models and content." + }, + { + "left": "pull_request:12193", + "right": "pull_request:12757", + "accept": false, + "reason": "Different docs updates: a README filename fix versus attention-backend documentation formatting." + }, + { + "left": "pull_request:12193", + "right": "pull_request:12971", + "accept": false, + "reason": "A README command rename is unrelated to reordering distributed inference documentation sections." + }, + { + "left": "pull_request:1221", + "right": "pull_request:2598", + "accept": false, + "reason": "Weight reuse docs and PaintByExample docs cleanup are unrelated documentation changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 93, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:12247", + "pull_request:1227", + "pull_request:12272", + "pull_request:1230", + "pull_request:12307", + "pull_request:12343", + "pull_request:12387", + "pull_request:12416", + "pull_request:12479", + "pull_request:12487", + "pull_request:12552", + "pull_request:12618", + "pull_request:12675", + "pull_request:12731", + "pull_request:12757", + "pull_request:12775", + "pull_request:1282", + "pull_request:12890", + "pull_request:12971", + "pull_request:12980", + "pull_request:12984", + "pull_request:13040", + "pull_request:13076", + "pull_request:13082", + "pull_request:13115", + "pull_request:13135", + "pull_request:13158", + "pull_request:13174", + "pull_request:13181", + "pull_request:13253", + "pull_request:13294", + "pull_request:13344", + "pull_request:13368", + "pull_request:13369", + "pull_request:13382", + "pull_request:13400", + "pull_request:13499", + "pull_request:13567", + "pull_request:1751", + "pull_request:2175", + "pull_request:2362", + "pull_request:2587", + "pull_request:2624", + "pull_request:2836", + "pull_request:3079", + "pull_request:429", + "pull_request:500" + ], + "soft_pairs": [ + "pull_request:12247|pull_request:12272", + "pull_request:1227|pull_request:2175", + "pull_request:1227|pull_request:2587", + "pull_request:1227|pull_request:2624", + "pull_request:1227|pull_request:429", + "pull_request:1227|pull_request:500", + "pull_request:1230|pull_request:1751", + "pull_request:1230|pull_request:2362", + "pull_request:1230|pull_request:3079", + "pull_request:12307|pull_request:12980", + "pull_request:12343|pull_request:12479", + "pull_request:12387|pull_request:12416", + "pull_request:12487|pull_request:13040", + "pull_request:12487|pull_request:13076", + "pull_request:12487|pull_request:13181", + "pull_request:12552|pull_request:13040", + "pull_request:12552|pull_request:13076", + "pull_request:12552|pull_request:13181", + "pull_request:12618|pull_request:13294", + "pull_request:12675|pull_request:13135", + "pull_request:12731|pull_request:12890", + "pull_request:12757|pull_request:12890", + "pull_request:12775|pull_request:13158", + "pull_request:12775|pull_request:13253", + "pull_request:12775|pull_request:13567", + "pull_request:1282|pull_request:2836", + "pull_request:12890|pull_request:12971", + "pull_request:12890|pull_request:13400", + "pull_request:12980|pull_request:13115", + "pull_request:12984|pull_request:13382", + "pull_request:12984|pull_request:13499", + "pull_request:13082|pull_request:13368", + "pull_request:13082|pull_request:13369", + "pull_request:13174|pull_request:13344" + ], + "prepared_review_unit_hash": "b6fd98bb938469b86869524514f4e9a2419a3d3a2d8fc26b7ef68d360bd91bc4", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 34, + "serialized_chars": 97670, + "estimated_input_tokens": 24418, + "estimated_eval_tokens": 49092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a mixed bag of unrelated docs, test, and code cleanup PRs. The soft links are all superficial topical overlap, not the same underlying bug or change.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "If a representative PR is needed, pull_request:12307 is the broadest substantive code change, but it still does not anchor the rest of the items as duplicates.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:12307 is the closest thing to a cluster anchor because it is a multi-file modular pipeline fix with real code impact; however, the cluster is too heterogeneous for a true duplicate canonical.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12247", + "right": "pull_request:12272", + "accept": false, + "reason": "Different fixes in different code paths: AutoGuidance validation vs FLUX modular pipeline defaults." + }, + { + "left": "pull_request:1227", + "right": "pull_request:2175", + "accept": false, + "reason": "Both are docs, but one adds modular docs and the other updates API docs; no shared change." + }, + { + "left": "pull_request:1227", + "right": "pull_request:2587", + "accept": false, + "reason": "Unrelated documentation updates for different sections and languages." + }, + { + "left": "pull_request:1227", + "right": "pull_request:2624", + "accept": false, + "reason": "No same bug or feature; docs for community examples vs cache_dir docs." + }, + { + "left": "pull_request:1227", + "right": "pull_request:429", + "accept": false, + "reason": "Different documentation targets and different content areas." + }, + { + "left": "pull_request:1227", + "right": "pull_request:500", + "accept": false, + "reason": "Both are docs changes, but about unrelated API/config topics." + }, + { + "left": "pull_request:1230", + "right": "pull_request:1751", + "accept": false, + "reason": "Different runtime issues: MPS test setup vs UnCLIP generator handling." + }, + { + "left": "pull_request:1230", + "right": "pull_request:2362", + "accept": false, + "reason": "Different test fixes for different pipelines and different failure modes." + }, + { + "left": "pull_request:1230", + "right": "pull_request:3079", + "accept": false, + "reason": "Both touch tests, but they target different pipelines and unrelated flakiness/performance issues." + }, + { + "left": "pull_request:12307", + "right": "pull_request:12980", + "accept": false, + "reason": "Shared modular-pipeline area, but one is broad pipeline wiring and the other is Mellon param/docstring cleanup." + }, + { + "left": "pull_request:12343", + "right": "pull_request:12479", + "accept": false, + "reason": "Different bugs in different files: VAE slicing logic vs an environment constant parsing fix." + }, + { + "left": "pull_request:12387", + "right": "pull_request:12416", + "accept": false, + "reason": "Different feature work: kernels flash-attention support vs QwenImage Edit Plus modular support." + }, + { + "left": "pull_request:12487", + "right": "pull_request:13040", + "accept": false, + "reason": "Both are documentation/grammar fixes, but they affect unrelated pages and issues." + }, + { + "left": "pull_request:12487", + "right": "pull_request:13076", + "accept": false, + "reason": "Unrelated docs fixes: broken links vs quantization syntax correction." + }, + { + "left": "pull_request:12487", + "right": "pull_request:13181", + "accept": false, + "reason": "Different docs topics and different corrections." + }, + { + "left": "pull_request:12552", + "right": "pull_request:13040", + "accept": false, + "reason": "Typos in different docs files; no single underlying problem." + }, + { + "left": "pull_request:12552", + "right": "pull_request:13076", + "accept": false, + "reason": "Separate doc typo cleanup from a quantization config syntax fix." + }, + { + "left": "pull_request:12552", + "right": "pull_request:13181", + "accept": false, + "reason": "Different docs pages and unrelated edits." + }, + { + "left": "pull_request:12618", + "right": "pull_request:13294", + "accept": false, + "reason": "CI workflow typo fix is unrelated to WanVACE flaky test adjustments." + }, + { + "left": "pull_request:12675", + "right": "pull_request:13135", + "accept": false, + "reason": "Different docstring/docs fixes for different pipelines." + }, + { + "left": "pull_request:12731", + "right": "pull_request:12890", + "accept": false, + "reason": "Both edit docs, but FIBO guidance changes are unrelated to Wan docs cleanup." + }, + { + "left": "pull_request:12757", + "right": "pull_request:12890", + "accept": false, + "reason": "Different docs topics: attention backend formatting vs Wan pipeline options cleanup." + }, + { + "left": "pull_request:12775", + "right": "pull_request:13158", + "accept": false, + "reason": "Z-Image docs addition is unrelated to QwenImageLayered docs." + }, + { + "left": "pull_request:12775", + "right": "pull_request:13253", + "accept": false, + "reason": "Docs addition vs test refactor; not the same change." + }, + { + "left": "pull_request:12775", + "right": "pull_request:13567", + "accept": false, + "reason": "Z-Image docs are unrelated to agent docs / pipeline guidance updates." + }, + { + "left": "pull_request:1282", + "right": "pull_request:2836", + "accept": false, + "reason": "Both are docs link fixes, but they concern different guides and different target links." + }, + { + "left": "pull_request:12890", + "right": "pull_request:12971", + "accept": false, + "reason": "Wan markdown cleanup vs distributed inference section reordering are unrelated docs changes." + }, + { + "left": "pull_request:12890", + "right": "pull_request:13400", + "accept": false, + "reason": "Different documentation pages and different content updates." + }, + { + "left": "pull_request:12980", + "right": "pull_request:13115", + "accept": false, + "reason": "Both are modular-pipeline related, but one is Mellon docstring cleanup and the other is Flux export/init changes." + }, + { + "left": "pull_request:12984", + "right": "pull_request:13382", + "accept": false, + "reason": "Flux2 Klein docs/code addition is unrelated to auto-docstring template documentation." + }, + { + "left": "pull_request:12984", + "right": "pull_request:13499", + "accept": false, + "reason": "Different docs/code areas: Flux2 Klein pipeline vs speed-memory optimization guidance." + }, + { + "left": "pull_request:13082", + "right": "pull_request:13368", + "accept": false, + "reason": "Wan model test refactor vs AutoencoderKL test refactor are different model families and test goals." + }, + { + "left": "pull_request:13082", + "right": "pull_request:13369", + "accept": false, + "reason": "Same general test-refactor theme, but different autoencoder targets and code paths." + }, + { + "left": "pull_request:13174", + "right": "pull_request:13344", + "accept": false, + "reason": "Attention-backend test additions are unrelated to GLM image transformer tests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 94, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1319", + "pull_request:1330", + "pull_request:1341", + "pull_request:1373", + "pull_request:1394", + "pull_request:1401", + "pull_request:1408", + "pull_request:1412", + "pull_request:1423", + "pull_request:1437", + "pull_request:1467", + "pull_request:1490", + "pull_request:1496", + "pull_request:15", + "pull_request:1510", + "pull_request:154", + "pull_request:1558", + "pull_request:1576", + "pull_request:1605", + "pull_request:1627", + "pull_request:1640", + "pull_request:168", + "pull_request:1727", + "pull_request:1849", + "pull_request:1989", + "pull_request:260", + "pull_request:3066", + "pull_request:3068", + "pull_request:316", + "pull_request:3166", + "pull_request:3241", + "pull_request:3389", + "pull_request:3417", + "pull_request:3554", + "pull_request:3555", + "pull_request:395", + "pull_request:407", + "pull_request:421", + "pull_request:641", + "pull_request:72", + "pull_request:798", + "pull_request:831", + "pull_request:848", + "pull_request:874", + "pull_request:908", + "pull_request:948", + "pull_request:98" + ], + "soft_pairs": [ + "pull_request:1319|pull_request:421", + "pull_request:1330|pull_request:908", + "pull_request:1341|pull_request:874", + "pull_request:1373|pull_request:1849", + "pull_request:1394|pull_request:831", + "pull_request:1401|pull_request:848", + "pull_request:1408|pull_request:98", + "pull_request:1412|pull_request:3389", + "pull_request:1412|pull_request:72", + "pull_request:1423|pull_request:168", + "pull_request:1437|pull_request:260", + "pull_request:1437|pull_request:407", + "pull_request:1437|pull_request:798", + "pull_request:1467|pull_request:3066", + "pull_request:1490|pull_request:641", + "pull_request:1496|pull_request:154", + "pull_request:1496|pull_request:1605", + "pull_request:15|pull_request:395", + "pull_request:1510|pull_request:1989", + "pull_request:1558|pull_request:948", + "pull_request:1576|pull_request:3068", + "pull_request:1576|pull_request:3166", + "pull_request:1576|pull_request:3554", + "pull_request:1576|pull_request:3555", + "pull_request:1627|pull_request:316", + "pull_request:1640|pull_request:3417", + "pull_request:1727|pull_request:3241" + ], + "prepared_review_unit_hash": "50ee5c8691d5cae0264c1a01865025939951ae9768e9f558ef894fc49b90171e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 97696, + "estimated_input_tokens": 24424, + "estimated_eval_tokens": 49104 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a false-positive collection: the PRs share broad themes like diffusion pipelines, schedulers, tests, or docs, but none of the soft-edge pairs fix the same concrete bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a valid canonical representative; the cluster spans unrelated feature, typo, doc, test, and release changes.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong global best choice because the set is heterogeneous; any pick would be arbitrary and misleading.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1319", + "right": "pull_request:421", + "accept": false, + "reason": "Scheduler-loading fix vs scheduler docs typo; unrelated change scopes." + }, + { + "left": "pull_request:1330", + "right": "pull_request:908", + "accept": false, + "reason": "Training/inpaint scaling fix vs community pipeline inpaint routing; different problems." + }, + { + "left": "pull_request:1341", + "right": "pull_request:874", + "accept": false, + "reason": "New repaint pipeline feature vs generic README example; not the same change." + }, + { + "left": "pull_request:1373", + "right": "pull_request:1849", + "accept": false, + "reason": "Versatile Diffusion feature expansion vs one attention-kwargs bugfix; different code paths." + }, + { + "left": "pull_request:1394", + "right": "pull_request:831", + "accept": false, + "reason": "Different release version bumps; not duplicates." + }, + { + "left": "pull_request:1401", + "right": "pull_request:848", + "accept": false, + "reason": "Broad optional height/width API change vs Flax shape bug in one pipeline." + }, + { + "left": "pull_request:1408", + "right": "pull_request:98", + "accept": false, + "reason": "Stable Diffusion converter fix vs NCSNPP converter script; unrelated scripts." + }, + { + "left": "pull_request:1412", + "right": "pull_request:3389", + "accept": false, + "reason": "Integration tests for SD2 inpainting vs attention-refactor integration tests; unrelated." + }, + { + "left": "pull_request:1412", + "right": "pull_request:72", + "accept": false, + "reason": "SD2 inpaint tests vs 1D up/downsample tests; different test areas." + }, + { + "left": "pull_request:1423", + "right": "pull_request:168", + "accept": false, + "reason": "Stable Diffusion 2 Flax support vs original Stable Diffusion pipeline introduction." + }, + { + "left": "pull_request:1437", + "right": "pull_request:260", + "accept": false, + "reason": "Pipeline typo fix vs README typo fix; unrelated files and bugs." + }, + { + "left": "pull_request:1437", + "right": "pull_request:407", + "accept": false, + "reason": "Different typos in the same file, but not the same concrete issue." + }, + { + "left": "pull_request:1437", + "right": "pull_request:798", + "accept": false, + "reason": "Pipeline typo vs UNet docstring typo; unrelated." + }, + { + "left": "pull_request:1467", + "right": "pull_request:3066", + "accept": false, + "reason": "Both relax slow tests, but they touch different assertions and test cases." + }, + { + "left": "pull_request:1490", + "right": "pull_request:641", + "accept": false, + "reason": "Version/deprecation bump vs temporary dependency pin; not the same fix." + }, + { + "left": "pull_request:1496", + "right": "pull_request:154", + "accept": false, + "reason": "Habana docs vs Stable Diffusion conversion script; unrelated." + }, + { + "left": "pull_request:1496", + "right": "pull_request:1605", + "accept": false, + "reason": "Habana docs vs Versatile Diffusion attention option; unrelated." + }, + { + "left": "pull_request:15", + "right": "pull_request:395", + "accept": false, + "reason": "README output-path tweak vs variable renaming in resnet; unrelated." + }, + { + "left": "pull_request:1510", + "right": "pull_request:1989", + "accept": false, + "reason": "Both touch the conversion script, but they address different enhancements." + }, + { + "left": "pull_request:1558", + "right": "pull_request:948", + "accept": false, + "reason": "Stable Diffusion typo cleanup vs DDPM comment typo; unrelated." + }, + { + "left": "pull_request:1576", + "right": "pull_request:3068", + "accept": false, + "reason": "ORT example integration vs textual inversion docs; different changes." + }, + { + "left": "pull_request:1576", + "right": "pull_request:3166", + "accept": false, + "reason": "ORT example integration vs textual inversion loader docs fix; unrelated." + }, + { + "left": "pull_request:1576", + "right": "pull_request:3554", + "accept": false, + "reason": "ORT example integration vs Moreh textual inversion adaptation; unrelated." + }, + { + "left": "pull_request:1576", + "right": "pull_request:3555", + "accept": false, + "reason": "ORT example integration vs Moreh textual inversion adaptation; unrelated." + }, + { + "left": "pull_request:1627", + "right": "pull_request:316", + "accept": false, + "reason": "xformers enable support vs README contribution text; unrelated." + }, + { + "left": "pull_request:1640", + "right": "pull_request:3417", + "accept": false, + "reason": "xformers auto-enable policy vs unCLIP scheduler warning note; different concerns." + }, + { + "left": "pull_request:1727", + "right": "pull_request:3241", + "accept": false, + "reason": "ONNX img2img preprocessing fix vs UnCLIP attention-slicing test tolerance; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 95, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:178", + "pull_request:1838", + "pull_request:1885", + "pull_request:190", + "pull_request:1932", + "pull_request:1965", + "pull_request:1995", + "pull_request:2006", + "pull_request:2009", + "pull_request:2077", + "pull_request:2078", + "pull_request:2085", + "pull_request:2086", + "pull_request:2103", + "pull_request:2134", + "pull_request:2138", + "pull_request:2139", + "pull_request:2144", + "pull_request:2151", + "pull_request:2154", + "pull_request:2175", + "pull_request:2181", + "pull_request:2278", + "pull_request:2539", + "pull_request:2540", + "pull_request:260", + "pull_request:2760", + "pull_request:3030", + "pull_request:309", + "pull_request:3187", + "pull_request:3285", + "pull_request:3388", + "pull_request:3458", + "pull_request:3460", + "pull_request:3470", + "pull_request:353", + "pull_request:3587", + "pull_request:3839", + "pull_request:407", + "pull_request:421", + "pull_request:4320", + "pull_request:688", + "pull_request:72", + "pull_request:798", + "pull_request:869", + "pull_request:925", + "pull_request:948", + "pull_request:972" + ], + "soft_pairs": [ + "pull_request:178|pull_request:2278", + "pull_request:178|pull_request:309", + "pull_request:178|pull_request:72", + "pull_request:1838|pull_request:2009", + "pull_request:1885|pull_request:260", + "pull_request:1885|pull_request:353", + "pull_request:1885|pull_request:407", + "pull_request:1885|pull_request:798", + "pull_request:1885|pull_request:972", + "pull_request:190|pull_request:688", + "pull_request:1932|pull_request:3285", + "pull_request:1965|pull_request:2006", + "pull_request:1995|pull_request:2539", + "pull_request:1995|pull_request:260", + "pull_request:1995|pull_request:353", + "pull_request:2006|pull_request:869", + "pull_request:2006|pull_request:925", + "pull_request:2009|pull_request:3587", + "pull_request:2077|pull_request:2760", + "pull_request:2078|pull_request:2139", + "pull_request:2085|pull_request:3470", + "pull_request:2086|pull_request:3458", + "pull_request:2086|pull_request:3460", + "pull_request:2103|pull_request:3187", + "pull_request:2103|pull_request:3458", + "pull_request:2103|pull_request:3460", + "pull_request:2134|pull_request:948", + "pull_request:2138|pull_request:3839", + "pull_request:2144|pull_request:3030", + "pull_request:2151|pull_request:4320", + "pull_request:2154|pull_request:260", + "pull_request:2154|pull_request:353", + "pull_request:2154|pull_request:407", + "pull_request:2154|pull_request:798", + "pull_request:2175|pull_request:3388", + "pull_request:2181|pull_request:2540", + "pull_request:2181|pull_request:421" + ], + "prepared_review_unit_hash": "0813ce3097555e177f71c464c4b2f65c5abefbfe56a27e98054ff17a4e1ea3a6", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 37, + "serialized_chars": 93354, + "estimated_input_tokens": 23339, + "estimated_eval_tokens": 46934 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a mix of unrelated documentation fixes, example-script tweaks, and a few real code changes across different subsystems. The soft-similarity links do not indicate duplicate bugs or mergeable PRs, so no pair should be collapsed.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:178", + "right": "pull_request:2278", + "accept": false, + "reason": "Stable Diffusion test update vs large video-stable-diffusion feature work; unrelated changes." + }, + { + "left": "pull_request:178", + "right": "pull_request:309", + "accept": false, + "reason": "Stable Diffusion test data update vs README link fix; different scope." + }, + { + "left": "pull_request:178", + "right": "pull_request:72", + "accept": false, + "reason": "Pipeline test update vs 1D up/downsample tests; different code paths." + }, + { + "left": "pull_request:1838", + "right": "pull_request:2009", + "accept": false, + "reason": "Textual inversion logging/docs work vs loader API support for embeddings; not the same bug/change." + }, + { + "left": "pull_request:1885", + "right": "pull_request:260", + "accept": false, + "reason": "DreamBooth loop variable fix vs README typo fix; unrelated." + }, + { + "left": "pull_request:1885", + "right": "pull_request:353", + "accept": false, + "reason": "Training script typo vs UNet block error-message typo; different files and issues." + }, + { + "left": "pull_request:1885", + "right": "pull_request:407", + "accept": false, + "reason": "DreamBooth training typo vs attention-slicing typo; unrelated subsystems." + }, + { + "left": "pull_request:1885", + "right": "pull_request:798", + "accept": false, + "reason": "DreamBooth training typo vs docstring typo; not duplicate work." + }, + { + "left": "pull_request:1885", + "right": "pull_request:972", + "accept": false, + "reason": "Training loop typo vs README torch_dtype typo; different artifacts." + }, + { + "left": "pull_request:190", + "right": "pull_request:688", + "accept": false, + "reason": "PyTorch import guard for Stable Diffusion vs Flax-without-PyTorch import handling; distinct compatibility fixes." + }, + { + "left": "pull_request:1932", + "right": "pull_request:3285", + "accept": false, + "reason": "DDIM scheduler conversion fix vs disk offload support; unrelated functionality." + }, + { + "left": "pull_request:1965", + "right": "pull_request:2006", + "accept": false, + "reason": "Both touch checkpoint merger, but one is import cleanup and the other is broader merger behavior fixes; not the same concrete change." + }, + { + "left": "pull_request:1995", + "right": "pull_request:2539", + "accept": false, + "reason": "Img2img strength range typo vs pix2pix-zero guidance default doc fix; unrelated." + }, + { + "left": "pull_request:1995", + "right": "pull_request:260", + "accept": false, + "reason": "Img2img parameter validation vs README typo fix; unrelated." + }, + { + "left": "pull_request:1995", + "right": "pull_request:353", + "accept": false, + "reason": "Img2img validation message vs UNet block typo; different code paths." + }, + { + "left": "pull_request:2006", + "right": "pull_request:869", + "accept": false, + "reason": "Checkpoint merger behavior fix vs community README/import bug cleanup; unrelated." + }, + { + "left": "pull_request:2006", + "right": "pull_request:925", + "accept": false, + "reason": "Checkpoint merger behavior fix vs prompt-padding bug in LPW community pipeline; different problem." + }, + { + "left": "pull_request:2009", + "right": "pull_request:3587", + "accept": false, + "reason": "Textual inversion loader support vs community pipeline README/example addition; unrelated." + }, + { + "left": "pull_request:2077", + "right": "pull_request:2760", + "accept": false, + "reason": "Validation-image support in textual inversion vs new DFQ example project; not duplicate." + }, + { + "left": "pull_request:2078", + "right": "pull_request:2139", + "accept": false, + "reason": "TensorBoard image logging in LoRA training vs LoRA alpha/scale docs section; different change type and target." + }, + { + "left": "pull_request:2085", + "right": "pull_request:3470", + "accept": false, + "reason": "DreamBooth crop option docs vs IF DreamBooth docs; unrelated documentation." + }, + { + "left": "pull_request:2086", + "right": "pull_request:3458", + "accept": false, + "reason": "LoRA docs addition vs inpaint LoRA-scale support implementation; different code and intent." + }, + { + "left": "pull_request:2086", + "right": "pull_request:3460", + "accept": false, + "reason": "LoRA docs addition vs inpaint LoRA-scale support implementation/tests; not the same change." + }, + { + "left": "pull_request:2103", + "right": "pull_request:3187", + "accept": false, + "reason": "Tagging model cards with lora vs x4 upscaler offload support; unrelated." + }, + { + "left": "pull_request:2103", + "right": "pull_request:3458", + "accept": false, + "reason": "LoRA tag metadata vs inpaint cross-attention support; different purpose." + }, + { + "left": "pull_request:2103", + "right": "pull_request:3460", + "accept": false, + "reason": "LoRA tag metadata vs inpaint support implementation/tests; unrelated." + }, + { + "left": "pull_request:2134", + "right": "pull_request:948", + "accept": false, + "reason": "README typo fix in DreamBooth vs pipeline_ddpm comment typo; unrelated docs-only fixes." + }, + { + "left": "pull_request:2138", + "right": "pull_request:3839", + "accept": false, + "reason": "Docs wording fix vs SDE VE add_noise behavior fix; different artifacts and code paths." + }, + { + "left": "pull_request:2144", + "right": "pull_request:3030", + "accept": false, + "reason": "Transformer 2D projection dimension bug vs panorama CFG bug; different models and failure modes." + }, + { + "left": "pull_request:2151", + "right": "pull_request:4320", + "accept": false, + "reason": "EMAModel state_dict typo fix vs ONNX docs typo; unrelated." + }, + { + "left": "pull_request:2154", + "right": "pull_request:260", + "accept": false, + "reason": "CLI environment version variable typo vs README typo; different files and issues." + }, + { + "left": "pull_request:2154", + "right": "pull_request:353", + "accept": false, + "reason": "CLI env version typo vs UNet block typo; unrelated." + }, + { + "left": "pull_request:2154", + "right": "pull_request:407", + "accept": false, + "reason": "CLI env version typo vs attention-slicing typo; unrelated." + }, + { + "left": "pull_request:2154", + "right": "pull_request:798", + "accept": false, + "reason": "CLI env version typo vs docstring typo; unrelated." + }, + { + "left": "pull_request:2175", + "right": "pull_request:3388", + "accept": false, + "reason": "DiffusionPipeline docs tweak vs installation docs adding transformers; unrelated docs." + }, + { + "left": "pull_request:2181", + "right": "pull_request:2540", + "accept": false, + "reason": "Flax scheduler docs section vs ControlNet docs correction; unrelated documentation." + }, + { + "left": "pull_request:2181", + "right": "pull_request:421", + "accept": false, + "reason": "Flax scheduler docs addition vs scheduler docs typo cleanup; different scope and content." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 96, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2208", + "pull_request:2216", + "pull_request:2271", + "pull_request:2290", + "pull_request:2333", + "pull_request:2362", + "pull_request:2381", + "pull_request:2396", + "pull_request:2405", + "pull_request:2488", + "pull_request:2498", + "pull_request:2503", + "pull_request:2530", + "pull_request:2533", + "pull_request:2557", + "pull_request:2563", + "pull_request:2587", + "pull_request:2590", + "pull_request:2604", + "pull_request:2624", + "pull_request:2629", + "pull_request:2631", + "pull_request:2637", + "pull_request:2639", + "pull_request:2642", + "pull_request:2647", + "pull_request:268", + "pull_request:2684", + "pull_request:2685", + "pull_request:2764", + "pull_request:3099", + "pull_request:3310", + "pull_request:3388", + "pull_request:3400", + "pull_request:3680", + "pull_request:3755", + "pull_request:4022", + "pull_request:406", + "pull_request:408", + "pull_request:4112", + "pull_request:4515", + "pull_request:4740", + "pull_request:4794", + "pull_request:483", + "pull_request:703", + "pull_request:830", + "pull_request:888", + "pull_request:979" + ], + "soft_pairs": [ + "pull_request:2208|pull_request:4022", + "pull_request:2216|pull_request:888", + "pull_request:2271|pull_request:268", + "pull_request:2290|pull_request:3680", + "pull_request:2333|pull_request:406", + "pull_request:2362|pull_request:3400", + "pull_request:2381|pull_request:979", + "pull_request:2396|pull_request:408", + "pull_request:2405|pull_request:830", + "pull_request:2488|pull_request:4515", + "pull_request:2498|pull_request:3099", + "pull_request:2503|pull_request:2764", + "pull_request:2530|pull_request:3755", + "pull_request:2533|pull_request:4740", + "pull_request:2557|pull_request:4112", + "pull_request:2563|pull_request:703", + "pull_request:2587|pull_request:3388", + "pull_request:2590|pull_request:483", + "pull_request:2604|pull_request:4794", + "pull_request:2624|pull_request:3388", + "pull_request:2629|pull_request:2631", + "pull_request:2637|pull_request:2639", + "pull_request:2637|pull_request:3310", + "pull_request:2639|pull_request:2642", + "pull_request:2639|pull_request:2647", + "pull_request:2639|pull_request:2684", + "pull_request:2639|pull_request:2685" + ], + "prepared_review_unit_hash": "e2cc74f6ac8f18b4340e0dc75593a0010b08113c0857720781174d425842418b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 96454, + "estimated_input_tokens": 24114, + "estimated_eval_tokens": 48484 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: unrelated docs fixes, test updates, dependency/CI changes, and a few feature PRs. None of the soft pairs look like the same concrete bug or change, so all should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a good canonical representative; the items span unrelated maintenance and feature work rather than one duplicate change.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong global duplicate target because there is no shared underlying bug or change across the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2208", + "right": "pull_request:4022", + "accept": false, + "reason": "xFormers docs update vs Korean translation overhaul; different documentation changes." + }, + { + "left": "pull_request:2216", + "right": "pull_request:888", + "accept": false, + "reason": "DifFace pipeline addition vs CycleDiffusion pipeline addition; different features and code paths." + }, + { + "left": "pull_request:2271", + "right": "pull_request:268", + "accept": false, + "reason": "Different test maintenance: one adjusts slow test expectations, the other updates model/pipeline expected outputs across several components." + }, + { + "left": "pull_request:2290", + "right": "pull_request:3680", + "accept": false, + "reason": "Pix2Pix doc formatting fix vs A1111 LoRA loader-link fix; unrelated docs changes." + }, + { + "left": "pull_request:2333", + "right": "pull_request:406", + "accept": false, + "reason": "Ethical guidelines wording change vs toctree typo fix; unrelated docs edits." + }, + { + "left": "pull_request:2362", + "right": "pull_request:3400", + "accept": false, + "reason": "MPS skip decorator/test changes vs dependency addition for tests; different code and intent." + }, + { + "left": "pull_request:2381", + "right": "pull_request:979", + "accept": false, + "reason": "Stable Diffusion dummy-import guard vs README missing import; unrelated fixes in different files." + }, + { + "left": "pull_request:2396", + "right": "pull_request:408", + "accept": false, + "reason": "Scheduler test warning cleanup vs documentation typo fixes; not the same change." + }, + { + "left": "pull_request:2405", + "right": "pull_request:830", + "accept": false, + "reason": "Revert of release bump vs release version bump; opposite versioning actions, not duplicates." + }, + { + "left": "pull_request:2488", + "right": "pull_request:4515", + "accept": false, + "reason": "Pipeline mixin refactor for num_images_per_prompt vs moving pipeline.to() for validation; different training/runtime behavior." + }, + { + "left": "pull_request:2498", + "right": "pull_request:3099", + "accept": false, + "reason": "Attend-and-excite test timeout adjustment vs changing a pipeline default parameter; unrelated test vs API change." + }, + { + "left": "pull_request:2503", + "right": "pull_request:2764", + "accept": false, + "reason": "Different torch2.0 docs edits: benchmark/title wording vs installation guidance; same file family but not the same change." + }, + { + "left": "pull_request:2530", + "right": "pull_request:3755", + "accept": false, + "reason": "New EMAModel test suite vs relaxing a flaky test tolerance; distinct test changes." + }, + { + "left": "pull_request:2533", + "right": "pull_request:4740", + "accept": false, + "reason": "Temporary CI/test setup for safetensors rc1 vs a later safetensors rc1 sanity-check WIP; different dependency-testing changes." + }, + { + "left": "pull_request:2557", + "right": "pull_request:4112", + "accept": false, + "reason": "DreamBooth intermediate logging vs adding rank support to LoRA training; separate training-script enhancements." + }, + { + "left": "pull_request:2563", + "right": "pull_request:703", + "accept": false, + "reason": "Removing cuda cache emptying in EMA step vs CI cache testing workflow; unrelated fixes." + }, + { + "left": "pull_request:2587", + "right": "pull_request:3388", + "accept": false, + "reason": "Korean docs translation update vs adding transformers to installation docs; different documentation work." + }, + { + "left": "pull_request:2590", + "right": "pull_request:483", + "accept": false, + "reason": "CI dependency-table check addition vs typo-fix workflow/docs changes; different maintenance areas." + }, + { + "left": "pull_request:2604", + "right": "pull_request:4794", + "accept": false, + "reason": "Colab link fix in docs vs compel syntax/example update; unrelated docs edits." + }, + { + "left": "pull_request:2624", + "right": "pull_request:3388", + "accept": false, + "reason": "cache_dir docs addition in pipeline utils vs installation docs adding transformers; different documentation targets." + }, + { + "left": "pull_request:2629", + "right": "pull_request:2631", + "accept": false, + "reason": "StableDiffusionSAG non-square image support vs DreamBooth main-process checks; unrelated code paths." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2639", + "accept": false, + "reason": "Quicktour wording cleanup vs basic training tutorial import addition; different docs pages and fixes." + }, + { + "left": "pull_request:2637", + "right": "pull_request:3310", + "accept": false, + "reason": "Quicktour wording cleanup vs stable_diffusion tutorial import fix; unrelated docs changes." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2642", + "accept": false, + "reason": "Basic training tutorial import fix vs loading docs wording change; separate docs edits." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2647", + "accept": false, + "reason": "Basic training tutorial import fix vs schedulers docs grammar correction; unrelated documentation changes." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2684", + "accept": false, + "reason": "Basic training tutorial import fix vs custom pipeline overview punctuation; different docs locations and fixes." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2685", + "accept": false, + "reason": "Basic training tutorial import fix vs KerasCV docs punctuation fix; unrelated docs changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 97, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2639", + "pull_request:2642", + "pull_request:2647", + "pull_request:2684", + "pull_request:2686", + "pull_request:2687", + "pull_request:2688", + "pull_request:2690", + "pull_request:2730", + "pull_request:2740", + "pull_request:2741", + "pull_request:2742", + "pull_request:2746", + "pull_request:2749", + "pull_request:2751", + "pull_request:2752", + "pull_request:2764", + "pull_request:2766", + "pull_request:2788", + "pull_request:2792", + "pull_request:2815", + "pull_request:2817", + "pull_request:2823", + "pull_request:286", + "pull_request:2862", + "pull_request:2863", + "pull_request:2864", + "pull_request:2865", + "pull_request:2869", + "pull_request:2870", + "pull_request:2873", + "pull_request:2894", + "pull_request:2911", + "pull_request:2912", + "pull_request:2913", + "pull_request:2919", + "pull_request:2923", + "pull_request:3119", + "pull_request:3310", + "pull_request:3472", + "pull_request:3572", + "pull_request:3821", + "pull_request:3894", + "pull_request:4715", + "pull_request:642", + "pull_request:650", + "pull_request:910", + "pull_request:983" + ], + "soft_pairs": [ + "pull_request:2639|pull_request:2686", + "pull_request:2639|pull_request:2687", + "pull_request:2639|pull_request:2688", + "pull_request:2639|pull_request:2690", + "pull_request:2639|pull_request:2730", + "pull_request:2639|pull_request:2740", + "pull_request:2639|pull_request:2741", + "pull_request:2639|pull_request:2742", + "pull_request:2639|pull_request:2746", + "pull_request:2639|pull_request:2749", + "pull_request:2639|pull_request:2751", + "pull_request:2639|pull_request:2752", + "pull_request:2639|pull_request:2862", + "pull_request:2639|pull_request:2863", + "pull_request:2639|pull_request:2864", + "pull_request:2639|pull_request:2865", + "pull_request:2639|pull_request:2870", + "pull_request:2639|pull_request:2923", + "pull_request:2642|pull_request:3310", + "pull_request:2647|pull_request:3310", + "pull_request:2684|pull_request:3310", + "pull_request:2742|pull_request:650", + "pull_request:2746|pull_request:983", + "pull_request:2764|pull_request:2766", + "pull_request:2764|pull_request:910", + "pull_request:2766|pull_request:2792", + "pull_request:2788|pull_request:3472", + "pull_request:2815|pull_request:2823", + "pull_request:2817|pull_request:3821", + "pull_request:286|pull_request:642", + "pull_request:2869|pull_request:3119", + "pull_request:2873|pull_request:3572", + "pull_request:2894|pull_request:3894", + "pull_request:2911|pull_request:3119", + "pull_request:2912|pull_request:3119", + "pull_request:2913|pull_request:3119", + "pull_request:2919|pull_request:4715" + ], + "prepared_review_unit_hash": "cb5a7e4c9d29fc455fce1a51963fb3f016eccec772d5381dbf94e73b489813f2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 37, + "serialized_chars": 92989, + "estimated_input_tokens": 23248, + "estimated_eval_tokens": 46752 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated documentation typo/phrasing PRs, plus a few separate test/code changes; there is no single underlying duplicate change to collapse.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 910 is the broadest, most substantive docs update in the set (multi-file inpainting guidance), so it is the best anchor if one must pick a representative PR, but the cluster is not a true duplicate group.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2639", + "right": "pull_request:2686", + "accept": false, + "reason": "Different docs pages and different tweaks: basic_training imports `os`, while unconditional_image_generation only adds a period." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2687", + "accept": false, + "reason": "Separate tutorial docs with unrelated wording fixes; not the same underlying change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2688", + "accept": false, + "reason": "basic_training import fix vs img2img section punctuation/formatting edits; different changes." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2690", + "accept": false, + "reason": "Unrelated docs pages and edits; one is a code snippet import, the other a sentence punctuation fix." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2730", + "accept": false, + "reason": "Different training docs and different edits; no shared concrete change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2740", + "accept": false, + "reason": "basic_training typo/import cleanup is unrelated to the dreambooth/text_inversion docs revisions in 2740." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2741", + "accept": false, + "reason": "Different files and different typo fixes; not one merged change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2742", + "accept": false, + "reason": "basic_training import edit vs dreambooth wording fix; separate doc updates." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2746", + "accept": false, + "reason": "Different docs and topics; one is tutorial code, the other fp16 prose cleanup." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2749", + "accept": false, + "reason": "Different docs pages; no evidence of a shared underlying bug/change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2751", + "accept": false, + "reason": "Separate training docs typo fixes; too broad and unrelated to be duplicates." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2752", + "accept": false, + "reason": "Different conceptual docs and different edits; no same change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2862", + "accept": false, + "reason": "basic_training import cleanup is unrelated to evaluation.mdx wording fixes." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2863", + "accept": false, + "reason": "Different docs pages; typo fixes only, not the same underlying change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2864", + "accept": false, + "reason": "Separate docs updates in different files; no duplicate bug/change." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2865", + "accept": false, + "reason": "Unrelated doc pages and edits; only broad docs similarity." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2870", + "accept": false, + "reason": "Different docs pages; one is a snippet/import edit, the other stable_diffusion_safe wording cleanup." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2923", + "accept": false, + "reason": "Different training tutorial files and separate wording fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:3310", + "accept": false, + "reason": "loading.mdx variant wording change vs stable_diffusion.mdx PIL import fix; unrelated docs updates." + }, + { + "left": "pull_request:2647", + "right": "pull_request:3310", + "accept": false, + "reason": "Schedulers docs punctuation vs stable_diffusion tutorial import fix; not one change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:3310", + "accept": false, + "reason": "custom_pipeline_overview punctuation fix is unrelated to stable_diffusion.md import correction." + }, + { + "left": "pull_request:2742", + "right": "pull_request:650", + "accept": false, + "reason": "Same general DreamBooth area, but different files and different changes; not a duplicate PR." + }, + { + "left": "pull_request:2746", + "right": "pull_request:983", + "accept": false, + "reason": "fp16 prose cleanup vs an empty debug PR; no shared concrete change." + }, + { + "left": "pull_request:2764", + "right": "pull_request:2766", + "accept": false, + "reason": "Docs update about torch 2.0 vs a test fix for MPS outputs; different artifact types and changes." + }, + { + "left": "pull_request:2764", + "right": "pull_request:910", + "accept": false, + "reason": "Torch 2.0 docs benchmark update and inpainting docs overhaul are unrelated changes." + }, + { + "left": "pull_request:2766", + "right": "pull_request:2792", + "accept": false, + "reason": "Both touch MPS-related tests, but they fix different test suites and different failures; not the same bug." + }, + { + "left": "pull_request:2788", + "right": "pull_request:3472", + "accept": false, + "reason": "Training toctree cleanup is unrelated to adding IF content in DreamBooth docs." + }, + { + "left": "pull_request:2815", + "right": "pull_request:2823", + "accept": false, + "reason": "Both edit stable_unclip.mdx, but 2815 adds background/checkpoints while 2823 rewrites the example code; separate doc updates." + }, + { + "left": "pull_request:2817", + "right": "pull_request:3821", + "accept": false, + "reason": "UNet1DModel docs addition vs a full Korean translation page; not the same change." + }, + { + "left": "pull_request:286", + "right": "pull_request:642", + "accept": false, + "reason": "Different README/link fixes in different files and contexts; not duplicates." + }, + { + "left": "pull_request:2869", + "right": "pull_request:3119", + "accept": false, + "reason": "Different pipeline docs and different wording/interface updates; not the same underlying change." + }, + { + "left": "pull_request:2873", + "right": "pull_request:3572", + "accept": false, + "reason": "General phrasing cleanup vs controlnet inpainting doc correction; unrelated edits." + }, + { + "left": "pull_request:2894", + "right": "pull_request:3894", + "accept": false, + "reason": "New flax requirements file vs Flax UNet dropout implementation; completely different changes." + }, + { + "left": "pull_request:2911", + "right": "pull_request:3119", + "accept": false, + "reason": "Stable-diffusion image_variation punctuation fix is unrelated to repaint interface docs." + }, + { + "left": "pull_request:2912", + "right": "pull_request:3119", + "accept": false, + "reason": "ControlNet docs punctuation fix vs repaint interface update; different changes." + }, + { + "left": "pull_request:2913", + "right": "pull_request:3119", + "accept": false, + "reason": "Stable-unCLIP docs punctuation/final newline fix is not the same as repaint docs." + }, + { + "left": "pull_request:2919", + "right": "pull_request:4715", + "accept": false, + "reason": "MPS test speedup and example script renames are different kinds of changes with no shared bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 98, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:296", + "pull_request:299", + "pull_request:2994", + "pull_request:3012", + "pull_request:3030", + "pull_request:3079", + "pull_request:3099", + "pull_request:3132", + "pull_request:3166", + "pull_request:3169", + "pull_request:3183", + "pull_request:3226", + "pull_request:3388", + "pull_request:3470", + "pull_request:3472", + "pull_request:3506", + "pull_request:3525", + "pull_request:356", + "pull_request:3572", + "pull_request:3577", + "pull_request:3641", + "pull_request:3663", + "pull_request:3680", + "pull_request:3744", + "pull_request:3755", + "pull_request:3795", + "pull_request:3798", + "pull_request:3821", + "pull_request:3842", + "pull_request:388", + "pull_request:3979", + "pull_request:4058", + "pull_request:4203", + "pull_request:4285", + "pull_request:4503", + "pull_request:4721", + "pull_request:4729", + "pull_request:4794", + "pull_request:5", + "pull_request:5117", + "pull_request:5157", + "pull_request:5250", + "pull_request:5332", + "pull_request:5639", + "pull_request:5723", + "pull_request:904", + "pull_request:948", + "pull_request:979" + ], + "soft_pairs": [ + "pull_request:296|pull_request:979", + "pull_request:299|pull_request:356", + "pull_request:2994|pull_request:5157", + "pull_request:3012|pull_request:948", + "pull_request:3030|pull_request:3979", + "pull_request:3079|pull_request:3099", + "pull_request:3132|pull_request:5332", + "pull_request:3166|pull_request:4794", + "pull_request:3169|pull_request:3663", + "pull_request:3183|pull_request:4729", + "pull_request:3226|pull_request:3572", + "pull_request:3388|pull_request:3795", + "pull_request:3470|pull_request:4203", + "pull_request:3472|pull_request:5250", + "pull_request:3472|pull_request:5723", + "pull_request:3506|pull_request:3641", + "pull_request:3525|pull_request:4203", + "pull_request:3525|pull_request:4503", + "pull_request:356|pull_request:904", + "pull_request:3577|pull_request:5639", + "pull_request:3680|pull_request:4721", + "pull_request:3744|pull_request:4058", + "pull_request:3755|pull_request:3842", + "pull_request:3798|pull_request:5117", + "pull_request:3821|pull_request:4285", + "pull_request:388|pull_request:5" + ], + "prepared_review_unit_hash": "12c0bfbdab66f2db02ead3db26b0cc3ec41d4dcdb3e9439bbaac48763d482749", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 92169, + "estimated_input_tokens": 23043, + "estimated_eval_tokens": 46342 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated documentation edits, example-script fixes, test adjustments, and core pipeline/scheduler bug fixes. There isn\u2019t a single underlying bug or change that unifies the items into one duplicate set.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR serves as a canonical representative because the cluster spans many unrelated changes across docs, tests, examples, and core code paths.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global canonical choice; the items do not share one concrete fix or feature, so any PR would misrepresent the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:296", + "right": "pull_request:979", + "accept": false, + "reason": "Different changes: auth-token test/workflow fix vs missing import + dtype README update." + }, + { + "left": "pull_request:299", + "right": "pull_request:356", + "accept": false, + "reason": "PNDM scheduler hotfix is unrelated to the stable-diffusion fine-tuning example PR." + }, + { + "left": "pull_request:2994", + "right": "pull_request:5157", + "accept": false, + "reason": "ControlNet training type-casting fix and LoRA name-conversion fix touch different code paths and bugs." + }, + { + "left": "pull_request:3012", + "right": "pull_request:948", + "accept": false, + "reason": "Both are typo/comment fixes, but in different pipelines and unrelated files; not the same change." + }, + { + "left": "pull_request:3030", + "right": "pull_request:3979", + "accept": false, + "reason": "Panorama CFG bug fix is unrelated to the SDXL attention processor/type-check bug." + }, + { + "left": "pull_request:3079", + "right": "pull_request:3099", + "accept": false, + "reason": "One speeds up tests; the other changes a pipeline default value. Different artifacts and purposes." + }, + { + "left": "pull_request:3132", + "right": "pull_request:5332", + "accept": false, + "reason": "A type-hint fix in score_sde_ve is unrelated to a DreamBooth comment typo." + }, + { + "left": "pull_request:3166", + "right": "pull_request:4794", + "accept": false, + "reason": "Textual inversion docs/example tweak vs compel syntax fix; both docs, but different issues." + }, + { + "left": "pull_request:3169", + "right": "pull_request:3663", + "accept": false, + "reason": "Habana docs update and Compel textual-inversion docs are separate documentation topics." + }, + { + "left": "pull_request:3183", + "right": "pull_request:4729", + "accept": false, + "reason": "DreamBooth LoRA checkpointing bug and a community script progress-bar bug are unrelated." + }, + { + "left": "pull_request:3226", + "right": "pull_request:3572", + "accept": false, + "reason": "ControlNet v1.1 docs addition is not the same as correcting the inpainting ControlNet docs example." + }, + { + "left": "pull_request:3388", + "right": "pull_request:3795", + "accept": false, + "reason": "Installation docs change vs overview-page pipeline additions; unrelated documentation scopes." + }, + { + "left": "pull_request:3470", + "right": "pull_request:4203", + "accept": false, + "reason": "Shared README file, but one adds IF training docs while the other only fixes a typo." + }, + { + "left": "pull_request:3472", + "right": "pull_request:5250", + "accept": false, + "reason": "DreamBooth IF docs and DDPO docs are separate new documentation sections for different workflows." + }, + { + "left": "pull_request:3472", + "right": "pull_request:5723", + "accept": false, + "reason": "DreamBooth IF docs and LCM docs describe different models and usage patterns." + }, + { + "left": "pull_request:3506", + "right": "pull_request:3641", + "accept": false, + "reason": "Checkpoint resume-step fix for ControlNet training is unrelated to inpainting initial-latents handling." + }, + { + "left": "pull_request:3525", + "right": "pull_request:4203", + "accept": false, + "reason": "README update is broad/tainted content, while the other is a typo fix; not the same underlying change." + }, + { + "left": "pull_request:3525", + "right": "pull_request:4503", + "accept": false, + "reason": "README update and LoRA docs clarification are different documentation changes." + }, + { + "left": "pull_request:356", + "right": "pull_request:904", + "accept": false, + "reason": "Stable diffusion fine-tuning repo/docs are unrelated to the inpainting pipeline implementation." + }, + { + "left": "pull_request:3577", + "right": "pull_request:5639", + "accept": false, + "reason": "Broken README links and AnimateDiff Motion LoRA docs are unrelated documentation updates." + }, + { + "left": "pull_request:3680", + "right": "pull_request:4721", + "accept": false, + "reason": "Loader-method link fix and Korean DreamBooth translation are different docs changes." + }, + { + "left": "pull_request:3744", + "right": "pull_request:4058", + "accept": false, + "reason": "Docs upload token fix is unrelated to SDXL documentation content changes." + }, + { + "left": "pull_request:3755", + "right": "pull_request:3842", + "accept": false, + "reason": "Both relax tests, but they cover different models/components and are not the same flaky test fix." + }, + { + "left": "pull_request:3798", + "right": "pull_request:5117", + "accept": false, + "reason": "Both touch training scripts, but one is comment cleanup and the other fixes resume-step logging/logic across many scripts." + }, + { + "left": "pull_request:3821", + "right": "pull_request:4285", + "accept": false, + "reason": "Translated documentation addition is unrelated to ONNX/OpenVINO/SDXL docs." + }, + { + "left": "pull_request:388", + "right": "pull_request:5", + "accept": false, + "reason": "A docs/pipeline page for latent diffusion is unrelated to adding the original UNet model implementation." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 99, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3979", + "pull_request:406", + "pull_request:407", + "pull_request:408", + "pull_request:4082", + "pull_request:4106", + "pull_request:4203", + "pull_request:421", + "pull_request:4262", + "pull_request:4267", + "pull_request:4275", + "pull_request:4285", + "pull_request:4330", + "pull_request:4350", + "pull_request:4380", + "pull_request:4437", + "pull_request:4457", + "pull_request:4461", + "pull_request:4472", + "pull_request:4479", + "pull_request:4503", + "pull_request:4563", + "pull_request:4582", + "pull_request:4595", + "pull_request:4612", + "pull_request:4626", + "pull_request:4696", + "pull_request:4721", + "pull_request:4763", + "pull_request:4796", + "pull_request:4797", + "pull_request:5013", + "pull_request:5063", + "pull_request:5078", + "pull_request:5267", + "pull_request:5286", + "pull_request:564", + "pull_request:5828", + "pull_request:5831", + "pull_request:5855", + "pull_request:6122", + "pull_request:6265", + "pull_request:6314", + "pull_request:6339", + "pull_request:6342", + "pull_request:6433", + "pull_request:948" + ], + "soft_pairs": [ + "pull_request:3979|pull_request:4796", + "pull_request:3979|pull_request:4797", + "pull_request:3979|pull_request:5078", + "pull_request:406|pull_request:408", + "pull_request:407|pull_request:948", + "pull_request:4082|pull_request:4267", + "pull_request:4082|pull_request:4582", + "pull_request:4082|pull_request:5013", + "pull_request:4106|pull_request:4595", + "pull_request:4203|pull_request:5267", + "pull_request:4203|pull_request:5855", + "pull_request:421|pull_request:564", + "pull_request:4262|pull_request:4763", + "pull_request:4267|pull_request:6339", + "pull_request:4275|pull_request:5831", + "pull_request:4285|pull_request:5286", + "pull_request:4330|pull_request:6342", + "pull_request:4350|pull_request:4696", + "pull_request:4350|pull_request:6314", + "pull_request:4380|pull_request:4457", + "pull_request:4437|pull_request:5828", + "pull_request:4461|pull_request:6265", + "pull_request:4472|pull_request:4503", + "pull_request:4479|pull_request:6122", + "pull_request:4503|pull_request:5267", + "pull_request:4563|pull_request:6433", + "pull_request:4582|pull_request:5063", + "pull_request:4612|pull_request:4626", + "pull_request:4721|pull_request:5831" + ], + "prepared_review_unit_hash": "0915bead843fb9a1779eb3d2415d2424a8e61afca3e99657e3d27db0c31d94d6", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 93360, + "estimated_input_tokens": 23340, + "estimated_eval_tokens": 46936 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs typo cleanups, SDXL runtime bugfixes, feature additions, and refactors. I do not see a single underlying issue/change spanning the items, so the soft matches should be rejected.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:3979", + "right": "pull_request:4796", + "accept": false, + "reason": "Both touch SDXL pipeline float/dtype code, but one is a small attention-processor check in upscale/XL pipelines while the other fixes broader SDXL latent upcasting behavior across multiple pipelines." + }, + { + "left": "pull_request:3979", + "right": "pull_request:4797", + "accept": false, + "reason": "Different bugs: attention-processor detection in a pipeline vs LoRA block remapping in loaders." + }, + { + "left": "pull_request:3979", + "right": "pull_request:5078", + "accept": false, + "reason": "Unrelated fixes: SDXL VAE/attention handling vs text-encoder training gradient-checkpointing setup." + }, + { + "left": "pull_request:406", + "right": "pull_request:408", + "accept": false, + "reason": "Both are docs typo cleanup, but they fix different files and different text; not the same change." + }, + { + "left": "pull_request:407", + "right": "pull_request:948", + "accept": false, + "reason": "One fixes an attention-slicing API typo; the other fixes a DDPM comment typo." + }, + { + "left": "pull_request:4082", + "right": "pull_request:4267", + "accept": false, + "reason": "Requirement pin change vs SDXL docs image-link correction; no shared bug or code path." + }, + { + "left": "pull_request:4082", + "right": "pull_request:4582", + "accept": false, + "reason": "Dependency/version fix vs ControlNet SDXL docstring correction; unrelated." + }, + { + "left": "pull_request:4082", + "right": "pull_request:5013", + "accept": false, + "reason": "Package requirement adjustment vs SDXL instruct-pix2pix dtype bug fix; different artifacts and problems." + }, + { + "left": "pull_request:4106", + "right": "pull_request:4595", + "accept": false, + "reason": "A large Tortoise TTS feature addition vs an unfinished converter script for SDXL; not the same change." + }, + { + "left": "pull_request:4203", + "right": "pull_request:5267", + "accept": false, + "reason": "DreamBooth README typo fix vs docs README formatting guidance; different docs and edits." + }, + { + "left": "pull_request:4203", + "right": "pull_request:5855", + "accept": false, + "reason": "DreamBooth README typo vs community README import typo; both docs-only but unrelated." + }, + { + "left": "pull_request:421", + "right": "pull_request:564", + "accept": false, + "reason": "Scheduler docs typo cleanup vs Flax scheduler API/documentation expansion; different scope and content." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4763", + "accept": false, + "reason": "SDXL DreamBooth LoRA training fixes vs a LoRA-to-safetensors conversion script; separate functionality." + }, + { + "left": "pull_request:4267", + "right": "pull_request:6339", + "accept": false, + "reason": "Different docs fixes for different features: SDXL image swap vs AnimateDiff docs." + }, + { + "left": "pull_request:4275", + "right": "pull_request:5831", + "accept": false, + "reason": "Autopipeline docs correction vs a docs tree title fix; unrelated doc edits." + }, + { + "left": "pull_request:4285", + "right": "pull_request:5286", + "accept": false, + "reason": "ONNX/OpenVINO documentation update vs adding Japanese docs; not the same underlying change." + }, + { + "left": "pull_request:4330", + "right": "pull_request:6342", + "accept": false, + "reason": "ControlNet README header correction vs installation docs note; unrelated documentation fixes." + }, + { + "left": "pull_request:4350", + "right": "pull_request:4696", + "accept": false, + "reason": "Base T2I-Adapter docs fp16 fix vs adding T2I-Adapter-XL support; one is a doc tweak, the other is a feature/code addition." + }, + { + "left": "pull_request:4350", + "right": "pull_request:6314", + "accept": false, + "reason": "Different docs issues in the same broad area, but one is dtype in adapter docs and the other is a paper-link typo." + }, + { + "left": "pull_request:4380", + "right": "pull_request:4457", + "accept": false, + "reason": "Minor docs fixes across training guides vs a TensorRT inpaint pipeline README/code fix; not the same change." + }, + { + "left": "pull_request:4437", + "right": "pull_request:5828", + "accept": false, + "reason": "Consistency-model sampling bugfix vs a VAE test memory annotation; different code paths." + }, + { + "left": "pull_request:4461", + "right": "pull_request:6265", + "accept": false, + "reason": "Converter script argument addition vs GLIGEN refactor; unrelated changes." + }, + { + "left": "pull_request:4472", + "right": "pull_request:4503", + "accept": false, + "reason": "DreamBooth README edit vs LoRA docs clarification; different documents and content." + }, + { + "left": "pull_request:4479", + "right": "pull_request:6122", + "accept": false, + "reason": "Pipeline typo fixes in code vs a broad docs typo sweep; same general theme but not the same bug/change." + }, + { + "left": "pull_request:4503", + "right": "pull_request:5267", + "accept": false, + "reason": "LoRA SDXL support clarification vs docs README code-block markup changes; unrelated." + }, + { + "left": "pull_request:4563", + "right": "pull_request:6433", + "accept": false, + "reason": "Prompt2Prompt community pipeline addition vs Gluegen community pipeline addition; separate features." + }, + { + "left": "pull_request:4582", + "right": "pull_request:5063", + "accept": false, + "reason": "ControlNet SDXL docstring fix vs guidance_rescale docstring updates across multiple pipelines; different doc issues." + }, + { + "left": "pull_request:4612", + "right": "pull_request:4626", + "accept": false, + "reason": "Link-path correction in controlling_generation vs link updates in text2image docs; related topic but different files and fixes." + }, + { + "left": "pull_request:4721", + "right": "pull_request:5831", + "accept": false, + "reason": "Korean DreamBooth docs cleanup vs English docs tree title fix; unrelated localization/docs maintenance." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 100, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4741", + "pull_request:4796", + "pull_request:4847", + "pull_request:4862", + "pull_request:4946", + "pull_request:498", + "pull_request:5041", + "pull_request:5065", + "pull_request:5066", + "pull_request:5184", + "pull_request:5207", + "pull_request:5236", + "pull_request:5286", + "pull_request:5407", + "pull_request:5478", + "pull_request:5505", + "pull_request:5652", + "pull_request:5695", + "pull_request:5697", + "pull_request:5698", + "pull_request:5727", + "pull_request:5744", + "pull_request:5745", + "pull_request:5828", + "pull_request:5831", + "pull_request:5844", + "pull_request:591", + "pull_request:6009", + "pull_request:6021", + "pull_request:6105", + "pull_request:6140", + "pull_request:6215", + "pull_request:6330", + "pull_request:6339", + "pull_request:6424", + "pull_request:6475", + "pull_request:6647", + "pull_request:6680", + "pull_request:6796", + "pull_request:6847", + "pull_request:6941", + "pull_request:7015", + "pull_request:7096", + "pull_request:7335", + "pull_request:7916", + "pull_request:827", + "pull_request:925" + ], + "soft_pairs": [ + "pull_request:4741|pull_request:6796", + "pull_request:4796|pull_request:5745", + "pull_request:4796|pull_request:6475", + "pull_request:4796|pull_request:6647", + "pull_request:4847|pull_request:6105", + "pull_request:4862|pull_request:5698", + "pull_request:4862|pull_request:6424", + "pull_request:4946|pull_request:5727", + "pull_request:498|pull_request:925", + "pull_request:5041|pull_request:5184", + "pull_request:5065|pull_request:5505", + "pull_request:5066|pull_request:5505", + "pull_request:5207|pull_request:5652", + "pull_request:5236|pull_request:5745", + "pull_request:5236|pull_request:7015", + "pull_request:5286|pull_request:5844", + "pull_request:5407|pull_request:7096", + "pull_request:5478|pull_request:5844", + "pull_request:5695|pull_request:6215", + "pull_request:5697|pull_request:6009", + "pull_request:5744|pull_request:6330", + "pull_request:5828|pull_request:6680", + "pull_request:5828|pull_request:7335", + "pull_request:5831|pull_request:6339", + "pull_request:591|pull_request:827", + "pull_request:6021|pull_request:7916", + "pull_request:6140|pull_request:6847", + "pull_request:6140|pull_request:6941" + ], + "prepared_review_unit_hash": "0ad0b1af00b18e72c099b748f038db5542ddaa97d237bc7b36bfb3549a73738d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 98434, + "estimated_input_tokens": 24609, + "estimated_eval_tokens": 49474 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of largely unrelated PRs across docs, tests, examples, and pipeline fixes. The soft links look like broad keyword similarity rather than the same underlying bug or change, so I rejected all candidate duplicate edges.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear canonical PR: the cluster spans unrelated fixes across docs, tests, training scripts, and pipelines.", + "best_issue_reason": null, + "best_pr_reason": "PR 6847 is the most substantial standalone change here, with broad code/test/docs impact and high discussion, but it still is not a safe duplicate anchor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4741", + "right": "pull_request:6796", + "accept": false, + "reason": "AudioLDM2 dummy import fix vs torchvision import gating; unrelated bugs." + }, + { + "left": "pull_request:4796", + "right": "pull_request:5745", + "accept": false, + "reason": "SDXL type-mismatch fix vs a docstring type correction; different artifacts and problems." + }, + { + "left": "pull_request:4796", + "right": "pull_request:6475", + "accept": false, + "reason": "Pipeline float16 upcasting issue vs VAE type doc update; not the same code-path bug." + }, + { + "left": "pull_request:4796", + "right": "pull_request:6647", + "accept": false, + "reason": "SDXL pipeline dtype issue vs a scheduler test expectation update; unrelated." + }, + { + "left": "pull_request:4847", + "right": "pull_request:6105", + "accept": false, + "reason": "Prompt Free Diffusion example vs SDE Drag pipeline addition; different community examples." + }, + { + "left": "pull_request:4862", + "right": "pull_request:5698", + "accept": false, + "reason": "MultiControlNet height/width bug vs LCM img2img fix; different pipelines and failures." + }, + { + "left": "pull_request:4862", + "right": "pull_request:6424", + "accept": false, + "reason": "ControlNet img2img fix vs PixArt weight conversion bug; unrelated changes." + }, + { + "left": "pull_request:4946", + "right": "pull_request:5727", + "accept": false, + "reason": "Training overview docs vs LCM script additions; not the same change." + }, + { + "left": "pull_request:498", + "right": "pull_request:925", + "accept": false, + "reason": "Attention-slicing disable fix vs lpw token padding logic; unrelated community/pipeline bugs." + }, + { + "left": "pull_request:5041", + "right": "pull_request:5184", + "accept": false, + "reason": "SDXL refiner-only tests vs scheduler consistency tests; different test targets." + }, + { + "left": "pull_request:5065", + "right": "pull_request:5505", + "accept": false, + "reason": "Attention-processor typo/docs cleanup vs ControlNet naming typo; same subsystem label but different fixes." + }, + { + "left": "pull_request:5066", + "right": "pull_request:5505", + "accept": false, + "reason": "Attention-processor typo cleanup vs ControlNet naming typo; unrelated concrete changes." + }, + { + "left": "pull_request:5207", + "right": "pull_request:5652", + "accept": false, + "reason": "T2I adapter README filename fix vs PixArt pipeline class-name fix; different files and issues." + }, + { + "left": "pull_request:5236", + "right": "pull_request:5745", + "accept": false, + "reason": "Korean docs formatting cleanup vs ConsistencyDecoder doc type fix; unrelated." + }, + { + "left": "pull_request:5236", + "right": "pull_request:7015", + "accept": false, + "reason": "Korean docs formatting cleanup vs single-file SDXL doc example fix; unrelated docs changes." + }, + { + "left": "pull_request:5286", + "right": "pull_request:5844", + "accept": false, + "reason": "Japanese docs addition vs an English docs reference example; not the same change." + }, + { + "left": "pull_request:5407", + "right": "pull_request:7096", + "accept": false, + "reason": "Mixed-precision training fix for unconditional generation vs ControlNet training enhancements; unrelated examples." + }, + { + "left": "pull_request:5478", + "right": "pull_request:5844", + "accept": false, + "reason": "Japanese docs workflow/content addition vs an English docs example addition; different documentation tasks." + }, + { + "left": "pull_request:5695", + "right": "pull_request:6215", + "accept": false, + "reason": "PixArt mask-feature condition bug vs PixArt test seeding/expected slice updates; not the same underlying issue." + }, + { + "left": "pull_request:5697", + "right": "pull_request:6009", + "accept": false, + "reason": "PEFT LoRA/xformers attention bug fix vs TensorRT super-args error; different code paths." + }, + { + "left": "pull_request:5744", + "right": "pull_request:6330", + "accept": false, + "reason": "API docs typo sweep vs SVD video-rendering doc fix; unrelated documentation edits." + }, + { + "left": "pull_request:5828", + "right": "pull_request:6680", + "accept": false, + "reason": "Consistency Decoder VAE test memory fix vs DPM scheduler slow-test fix; unrelated tests." + }, + { + "left": "pull_request:5828", + "right": "pull_request:7335", + "accept": false, + "reason": "VAE test memory fix vs ControlNet SDXL config test; different modules and failures." + }, + { + "left": "pull_request:5831", + "right": "pull_request:6339", + "accept": false, + "reason": "Generic docs title fix vs AnimateDiff docs fix; not the same artifact." + }, + { + "left": "pull_request:591", + "right": "pull_request:827", + "accept": false, + "reason": "Flax safety-checker save/load refactor vs PT/Flax API alignment; both Flax-related but different problems." + }, + { + "left": "pull_request:6021", + "right": "pull_request:7916", + "accept": false, + "reason": "Dreambooth LoRA variant argument vs Ascend NPU support for SDXL; unrelated feature work." + }, + { + "left": "pull_request:6140", + "right": "pull_request:6847", + "accept": false, + "reason": "IP-Adapter API docs page vs IP-Adapter attention masking feature; docs versus code change, not one bug." + }, + { + "left": "pull_request:6140", + "right": "pull_request:6941", + "accept": false, + "reason": "IP-Adapter docs page vs ControlNet Inpaint IP-Adapter support; same topic area but different changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 101, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:62", + "pull_request:6205", + "pull_request:6209", + "pull_request:6244", + "pull_request:626", + "pull_request:6276", + "pull_request:6293", + "pull_request:6314", + "pull_request:6339", + "pull_request:6669", + "pull_request:6754", + "pull_request:6762", + "pull_request:6765", + "pull_request:6766", + "pull_request:6767", + "pull_request:6768", + "pull_request:6796", + "pull_request:6808", + "pull_request:6837", + "pull_request:6847", + "pull_request:6871", + "pull_request:6941", + "pull_request:6962", + "pull_request:7015", + "pull_request:7120", + "pull_request:715", + "pull_request:7206", + "pull_request:7226", + "pull_request:7259", + "pull_request:7272", + "pull_request:7302", + "pull_request:7319", + "pull_request:7360", + "pull_request:7564", + "pull_request:7746", + "pull_request:7747", + "pull_request:7779", + "pull_request:7780", + "pull_request:8422", + "pull_request:8427", + "pull_request:848", + "pull_request:8491", + "pull_request:8630", + "pull_request:8675", + "pull_request:8775", + "pull_request:8882", + "pull_request:8896", + "pull_request:9015" + ], + "soft_pairs": [ + "pull_request:62|pull_request:848", + "pull_request:6205|pull_request:6314", + "pull_request:6209|pull_request:7746", + "pull_request:6244|pull_request:6847", + "pull_request:6244|pull_request:6941", + "pull_request:626|pull_request:715", + "pull_request:6276|pull_request:8427", + "pull_request:6293|pull_request:6847", + "pull_request:6293|pull_request:6941", + "pull_request:6339|pull_request:7779", + "pull_request:6669|pull_request:8775", + "pull_request:6754|pull_request:8882", + "pull_request:6754|pull_request:8896", + "pull_request:6762|pull_request:8422", + "pull_request:6765|pull_request:8882", + "pull_request:6765|pull_request:8896", + "pull_request:6766|pull_request:8882", + "pull_request:6766|pull_request:8896", + "pull_request:6767|pull_request:8882", + "pull_request:6767|pull_request:8896", + "pull_request:6768|pull_request:8882", + "pull_request:6768|pull_request:8896", + "pull_request:6796|pull_request:8491", + "pull_request:6808|pull_request:7746", + "pull_request:6837|pull_request:8675", + "pull_request:6871|pull_request:7747", + "pull_request:6962|pull_request:7564", + "pull_request:7015|pull_request:9015", + "pull_request:7120|pull_request:7319", + "pull_request:7206|pull_request:8630", + "pull_request:7226|pull_request:7780", + "pull_request:7259|pull_request:7360", + "pull_request:7272|pull_request:7302" + ], + "prepared_review_unit_hash": "6d7716a3231a17b45b75ad8e1fc18537b28eb598f5c6c80c00de69d969c96474", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 96659, + "estimated_input_tokens": 24165, + "estimated_eval_tokens": 48586 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is highly heterogeneous: it contains unrelated docs-only fixes, test tweaks, dependency/import cleanup, scheduler additions, and separate IP-Adapter/controlnet changes. There isn\u2019t a single underlying bug or change tying the cluster together.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits: the artifacts span unrelated documentation edits, tests, refactors, and feature work across different subsystems.", + "best_issue_reason": null, + "best_pr_reason": "No best PR for deduplication; the items do not appear to be duplicates or mergeable into one concrete fix/change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:62", + "right": "pull_request:848", + "accept": false, + "reason": "Both are old pipeline fixes, but they address different code paths: one changes latent diffusion decoding, the other fixes Flax latent shape calculation." + }, + { + "left": "pull_request:6205", + "right": "pull_request:6314", + "accept": false, + "reason": "Both edit the same docs file, but one is a blog URL correction and the other fixes a paper link typo; unrelated documentation changes." + }, + { + "left": "pull_request:6209", + "right": "pull_request:7746", + "accept": false, + "reason": "One fixes an example README script, the other fixes a VAE test; same project area but not the same underlying bug." + }, + { + "left": "pull_request:6244", + "right": "pull_request:6847", + "accept": false, + "reason": "Both touch IP-Adapter-related code, but 6244 is an embedding/refactor PR while 6847 adds attention masking support; different features." + }, + { + "left": "pull_request:6244", + "right": "pull_request:6941", + "accept": false, + "reason": "Both involve IP-Adapter support in ControlNet-ish pipelines, but they implement different pipelines/features and are not the same fix." + }, + { + "left": "pull_request:626", + "right": "pull_request:715", + "accept": false, + "reason": "One is an ONNX typing fix, the other is a PyTorch import guard; unrelated maintenance changes." + }, + { + "left": "pull_request:6276", + "right": "pull_request:8427", + "accept": false, + "reason": "One adds a community IP-Adapter FaceID pipeline, the other only prints env vars in a workflow; no shared bug/change." + }, + { + "left": "pull_request:6293", + "right": "pull_request:6847", + "accept": false, + "reason": "Both are IP-Adapter related, but one adds IP-Adapter to SDXL ControlNet img2img and the other adds attention masking; different concrete changes." + }, + { + "left": "pull_request:6293", + "right": "pull_request:6941", + "accept": false, + "reason": "These target different ControlNet pipelines and different integration points for IP-Adapter support; not the same code-path fix." + }, + { + "left": "pull_request:6339", + "right": "pull_request:7779", + "accept": false, + "reason": "One updates AnimateDiff docs/example code, the other fixes AutoPipeline docstrings; unrelated documentation work." + }, + { + "left": "pull_request:6669", + "right": "pull_request:8775", + "accept": false, + "reason": "Both are README updates, but one is repo stats wording and the other is a Colab link in a research README; unrelated." + }, + { + "left": "pull_request:6754", + "right": "pull_request:8882", + "accept": false, + "reason": "The DPO training loss fix is unrelated to an IP-Adapter loader docstring correction." + }, + { + "left": "pull_request:6754", + "right": "pull_request:8896", + "accept": false, + "reason": "The DPO training loss fix and the single-file controlnet utility call fix are different training/runtime issues." + }, + { + "left": "pull_request:6762", + "right": "pull_request:8422", + "accept": false, + "reason": "One relaxes/flakes a Kandinsky test, the other adds subfolder-loading test coverage; different test concerns." + }, + { + "left": "pull_request:6765", + "right": "pull_request:8882", + "accept": false, + "reason": "Path-handling cache update in hub_utils is unrelated to an IP-Adapter loader docstring fix." + }, + { + "left": "pull_request:6765", + "right": "pull_request:8896", + "accept": false, + "reason": "Cache-path handling and single-file controlnet config wiring address different runtime paths." + }, + { + "left": "pull_request:6766", + "right": "pull_request:8882", + "accept": false, + "reason": "Constants path normalization is unrelated to the IP-Adapter loader docstring change." + }, + { + "left": "pull_request:6766", + "right": "pull_request:8896", + "accept": false, + "reason": "A constants.py path update does not match the single-file controlnet utility fix." + }, + { + "left": "pull_request:6767", + "right": "pull_request:8882", + "accept": false, + "reason": "Flax model path handling and an IP-Adapter loader docstring correction are unrelated." + }, + { + "left": "pull_request:6767", + "right": "pull_request:8896", + "accept": false, + "reason": "Different modules and different failures: Flax loading vs controlnet single-file config conversion." + }, + { + "left": "pull_request:6768", + "right": "pull_request:8882", + "accept": false, + "reason": "Configuration utils path handling and IP-Adapter loader docs are unrelated changes." + }, + { + "left": "pull_request:6768", + "right": "pull_request:8896", + "accept": false, + "reason": "These touch different utilities and fix different issues." + }, + { + "left": "pull_request:6796", + "right": "pull_request:8491", + "accept": false, + "reason": "One adds a torchvision availability guard; the other fixes a model import path. Same broad import area, but not the same bug." + }, + { + "left": "pull_request:6808", + "right": "pull_request:7746", + "accept": false, + "reason": "Testing utility device normalization is unrelated to the VAE test consistency fix." + }, + { + "left": "pull_request:6837", + "right": "pull_request:8675", + "accept": false, + "reason": "One updates I2VGenXL docs dependencies; the other changes a fast-diffusion tutorial note. Both docs, but unrelated content." + }, + { + "left": "pull_request:6871", + "right": "pull_request:7747", + "accept": false, + "reason": "VAE slicing/tiling feature work is different from a test assertion fix for tiled VAE decoding." + }, + { + "left": "pull_request:6941", + "right": "pull_request:6762", + "accept": false, + "reason": "ControlNet SDXL IP-Adapter support and a Kandinsky test flake fix are unrelated." + }, + { + "left": "pull_request:6962", + "right": "pull_request:7564", + "accept": false, + "reason": "PixArt slow test fixes are unrelated to the broader Animatediff/test compatibility cleanup." + }, + { + "left": "pull_request:7015", + "right": "pull_request:9015", + "accept": false, + "reason": "Different doc examples for different pipelines; no shared underlying bug or change." + }, + { + "left": "pull_request:7120", + "right": "pull_request:7319", + "accept": false, + "reason": "EDM scheduler code addition versus docs-only scheduler pages are not the same change." + }, + { + "left": "pull_request:7206", + "right": "pull_request:8630", + "accept": false, + "reason": "Training-script tagging across many examples is unrelated to SD3 DreamBooth LoRA text-encoder training." + }, + { + "left": "pull_request:7226", + "right": "pull_request:7780", + "accept": false, + "reason": "Both touch IP-Adapter docs, but 7226 adds API/embedding docs while 7780 edits masking wording/example text; not the same fix." + }, + { + "left": "pull_request:7259", + "right": "pull_request:7360", + "accept": false, + "reason": "TCD intro docs and TCD scheduler docstring cleanup are different artifacts with different purposes." + }, + { + "left": "pull_request:7272", + "right": "pull_request:7302", + "accept": false, + "reason": "Custom Diffusion PEFT loader fix and the training-script gradient-index bug are different code paths and different issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 102, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:12006", + "pull_request:12086", + "pull_request:12387", + "pull_request:12909", + "pull_request:7384", + "pull_request:7425", + "pull_request:7464", + "pull_request:7530", + "pull_request:7627", + "pull_request:7719", + "pull_request:7746", + "pull_request:7827", + "pull_request:7918", + "pull_request:798", + "pull_request:8427", + "pull_request:8470", + "pull_request:8479", + "pull_request:8481", + "pull_request:8497", + "pull_request:8498", + "pull_request:8720", + "pull_request:8737", + "pull_request:8836", + "pull_request:8838", + "pull_request:8845", + "pull_request:8851", + "pull_request:8946", + "pull_request:8951", + "pull_request:8984", + "pull_request:9008", + "pull_request:9066", + "pull_request:9077", + "pull_request:9100", + "pull_request:9107", + "pull_request:9152", + "pull_request:9196", + "pull_request:9266", + "pull_request:9408", + "pull_request:948", + "pull_request:9553", + "pull_request:9571", + "pull_request:9627", + "pull_request:9658", + "pull_request:9683", + "pull_request:972", + "pull_request:9739", + "pull_request:9765" + ], + "soft_pairs": [ + "pull_request:7384|pull_request:8851", + "pull_request:7425|pull_request:8737", + "pull_request:7425|pull_request:9066", + "pull_request:7464|pull_request:8498", + "pull_request:7530|pull_request:8946", + "pull_request:7627|pull_request:8720", + "pull_request:7719|pull_request:7918", + "pull_request:7746|pull_request:9627", + "pull_request:7746|pull_request:9683", + "pull_request:7827|pull_request:9765", + "pull_request:798|pull_request:948", + "pull_request:8427|pull_request:9765", + "pull_request:8470|pull_request:9553", + "pull_request:8479|pull_request:9408", + "pull_request:8481|pull_request:8836", + "pull_request:8497|pull_request:9107", + "pull_request:8838|pull_request:9196", + "pull_request:8845|pull_request:9100", + "pull_request:8845|pull_request:9683", + "pull_request:8951|pull_request:9765", + "pull_request:8984|pull_request:9765", + "pull_request:9008|pull_request:9077", + "pull_request:9008|pull_request:9739", + "pull_request:9152|pull_request:9266", + "pull_request:9408|pull_request:9571", + "pull_request:9408|pull_request:9658", + "pull_request:948|pull_request:972", + "pull_request:12086|pull_request:12387", + "pull_request:12006|pull_request:12909" + ], + "prepared_review_unit_hash": "90756c137c34d066f0977bdaa51e97fdfc07a44657a626db56e8485d09bc1266", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 93898, + "estimated_input_tokens": 23475, + "estimated_eval_tokens": 47206 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Loose cluster of mostly unrelated small docs, tests, workflow, and feature PRs; no soft pair looks like a true duplicate of the same underlying change.", + "confidence": 0.16, + "canonical_issue_reason": null, + "canonical_pr_reason": "If forced to pick one PR as the closest umbrella representative, PR 9008 is the broadest errata/cleanup item; however, the cluster is too heterogeneous for a real canonical PR.", + "best_issue_reason": null, + "best_pr_reason": "PR 9008 is the best single representative only because it is the broadest generic cleanup/docs PR in the set, not because it matches the others as a duplicate fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7384", + "right": "pull_request:8851", + "accept": false, + "reason": "Both are docs-related, but one adds FreeInit outputs and the other adds AuraFlow docs; different features." + }, + { + "left": "pull_request:7425", + "right": "pull_request:8737", + "accept": false, + "reason": "Missing output image docs vs a large workflow/repo cleanup PR; not the same change." + }, + { + "left": "pull_request:7425", + "right": "pull_request:9066", + "accept": false, + "reason": "Both touch docs, but one is a tutorial image fix and the other adds Stable Cascade docs; unrelated." + }, + { + "left": "pull_request:7464", + "right": "pull_request:8498", + "accept": false, + "reason": "Small LDM3D import fix vs an SD3 docstring typo; different artifacts and code paths." + }, + { + "left": "pull_request:7530", + "right": "pull_request:8946", + "accept": false, + "reason": "Training autocast bugfix vs a test expectation change; unrelated." + }, + { + "left": "pull_request:7627", + "right": "pull_request:8720", + "accept": false, + "reason": "ControlNet img2img latent handling vs slow test precision updates; not the same bug." + }, + { + "left": "pull_request:7719", + "right": "pull_request:7918", + "accept": false, + "reason": "Workflow secret fix vs a Flax attention reshape bugfix; completely different changes." + }, + { + "left": "pull_request:7746", + "right": "pull_request:9627", + "accept": false, + "reason": "VAE test dtype adjustment vs LoRA warning-test reordering; unrelated test fixes." + }, + { + "left": "pull_request:7746", + "right": "pull_request:9683", + "accept": false, + "reason": "VAE test dtype fix vs CogVideoX test rename/unskip; different test issues." + }, + { + "left": "pull_request:7827", + "right": "pull_request:9765", + "accept": false, + "reason": "Both are doc changes, but one fixes CPU offload wording and the other fixes SD3 pipeline docstrings; not the same change." + }, + { + "left": "pull_request:798", + "right": "pull_request:948", + "accept": false, + "reason": "Unet docstring typo vs DDPM comment typo; unrelated." + }, + { + "left": "pull_request:8427", + "right": "pull_request:9765", + "accept": false, + "reason": "Workflow debug print vs SD3 docstring typo; no shared bug." + }, + { + "left": "pull_request:8470", + "right": "pull_request:9553", + "accept": false, + "reason": "Different documentation updates for different topics; not one underlying change." + }, + { + "left": "pull_request:8479", + "right": "pull_request:9408", + "accept": false, + "reason": "Both add controlnet-related features, but for different models and code paths; too broad to merge as one PR." + }, + { + "left": "pull_request:8481", + "right": "pull_request:8836", + "accept": false, + "reason": "Image processor error message vs sliced attention loop off-by-one fix; unrelated." + }, + { + "left": "pull_request:8497", + "right": "pull_request:9107", + "accept": false, + "reason": "Scheduler spelling fix vs SD3 loading-hook logic bug; different artifacts." + }, + { + "left": "pull_request:8838", + "right": "pull_request:9196", + "accept": false, + "reason": "SD3 forward chunking support vs PixArt attention-processor helper; different model features." + }, + { + "left": "pull_request:8845", + "right": "pull_request:9100", + "accept": false, + "reason": "PAG test typo vs a LoRA example typo; unrelated test/docs edits." + }, + { + "left": "pull_request:8845", + "right": "pull_request:9683", + "accept": false, + "reason": "PAG test typo vs CogVideoX test rename/unskip; not the same issue." + }, + { + "left": "pull_request:8951", + "right": "pull_request:9765", + "accept": false, + "reason": "SD3 training code deletion logic bug vs SD3 pipeline docstring typo; code vs docs." + }, + { + "left": "pull_request:8984", + "right": "pull_request:9765", + "accept": false, + "reason": "SD3 ONNX conversion support vs a docstring fix; different scope entirely." + }, + { + "left": "pull_request:9008", + "right": "pull_request:9077", + "accept": false, + "reason": "Broad errata/docs cleanup vs isolated typo fixes; not one duplicate change." + }, + { + "left": "pull_request:9008", + "right": "pull_request:9739", + "accept": false, + "reason": "General errata cleanup vs SD3 docs typo and single-file checkpoint docs; unrelated." + }, + { + "left": "pull_request:9152", + "right": "pull_request:9266", + "accept": false, + "reason": "Distributed inference tutorial expansion vs community projects list update; different docs additions." + }, + { + "left": "pull_request:9408", + "right": "pull_request:9571", + "accept": false, + "reason": "Flux ControlNet pipeline additions vs Flux ControlNet guidance-window parameters; same subsystem, but different concrete features." + }, + { + "left": "pull_request:9408", + "right": "pull_request:9658", + "accept": false, + "reason": "Flux ControlNet pipeline feature vs a community example script; not the same underlying change." + }, + { + "left": "pull_request:948", + "right": "pull_request:972", + "accept": false, + "reason": "DDPM comment typo vs README argument-name typo; unrelated." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12387", + "accept": false, + "reason": "Flux modular I2I support vs FlashAttention-kernels support; different features." + }, + { + "left": "pull_request:12006", + "right": "pull_request:12909", + "accept": false, + "reason": "Wan 5B I2V feature addition vs a WAN context-parallel fix; related area, but not the same bug or merged change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 103, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10230", + "pull_request:10318", + "pull_request:10544", + "pull_request:11479", + "pull_request:11595", + "pull_request:11843", + "pull_request:12074", + "pull_request:12324", + "pull_request:12416", + "pull_request:12451", + "pull_request:12541", + "pull_request:12681", + "pull_request:12782", + "pull_request:13042", + "pull_request:13211", + "pull_request:13280", + "pull_request:2172", + "pull_request:2367", + "pull_request:2464", + "pull_request:2467", + "pull_request:3080", + "pull_request:3299", + "pull_request:3339", + "pull_request:3652", + "pull_request:4315", + "pull_request:4508", + "pull_request:4761", + "pull_request:4795", + "pull_request:4862", + "pull_request:4899", + "pull_request:5127", + "pull_request:5298", + "pull_request:5738", + "pull_request:5948", + "pull_request:5958", + "pull_request:6217", + "pull_request:6868", + "pull_request:6946", + "pull_request:7189", + "pull_request:7632", + "pull_request:7786", + "pull_request:823", + "pull_request:826", + "pull_request:9061", + "pull_request:9073", + "pull_request:9213", + "pull_request:9964" + ], + "soft_pairs": [ + "pull_request:6946|pull_request:9073", + "pull_request:6868|pull_request:7189", + "pull_request:9213|pull_request:9964", + "pull_request:12416|pull_request:13280", + "pull_request:12074|pull_request:12324", + "pull_request:2367|pull_request:3652", + "pull_request:7632|pull_request:9061", + "pull_request:2467|pull_request:823", + "pull_request:3299|pull_request:4761", + "pull_request:4315|pull_request:5298", + "pull_request:4508|pull_request:5948", + "pull_request:4795|pull_request:4862", + "pull_request:5738|pull_request:5958", + "pull_request:6217|pull_request:7786", + "pull_request:823|pull_request:826", + "pull_request:4899|pull_request:5127", + "pull_request:10230|pull_request:10544", + "pull_request:10318|pull_request:11595", + "pull_request:11479|pull_request:12451", + "pull_request:11843|pull_request:12541", + "pull_request:12681|pull_request:13042", + "pull_request:12782|pull_request:13211", + "pull_request:2172|pull_request:3080", + "pull_request:2464|pull_request:3339" + ], + "prepared_review_unit_hash": "58cf577745bd1d362482e76ea0c3bbc7518db968da1ea13cdcd5efd42e311308", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 101725, + "estimated_input_tokens": 25432, + "estimated_eval_tokens": 51120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated PRs across docs, quantization, Wan/Qwen, checkpoints, tests, and hardware-specific fixes. The only plausible duplicate-like pair is the two LoRA torch.compile fixes, which address the same underlying compile-compatibility problem in the LoRA conv path.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a good canonical for the whole cluster; the items are too heterogeneous, with only a small LoRA compile-fix subcluster overlapping.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR for the cluster because there is no single underlying bug/change tying the items together.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6946", + "right": "pull_request:9073", + "accept": false, + "reason": "Different features in different areas: IP-Adapter low_cpu_mem_usage loading vs SparseControlNet single-file loading support." + }, + { + "left": "pull_request:6868", + "right": "pull_request:7189", + "accept": false, + "reason": "Both are IP-Adapter related, but one adds image_embeds input support and the other fixes plus-version embedding handling; not the same concrete change." + }, + { + "left": "pull_request:9213", + "right": "pull_request:9964", + "accept": false, + "reason": "Different quantization backends and changes: bitsandbytes quantization support vs GGUF support." + }, + { + "left": "pull_request:12416", + "right": "pull_request:13280", + "accept": false, + "reason": "Unrelated subsystems: QwenImage Edit Plus modular pipeline support vs FlashAttention-4 support." + }, + { + "left": "pull_request:12074", + "right": "pull_request:12324", + "accept": false, + "reason": "Both touch Wan, but one adds second-transformer LoRA loading and the other adds Wan2.2 VACE-Fun conversion/model support; different changes." + }, + { + "left": "pull_request:2367", + "right": "pull_request:3652", + "accept": false, + "reason": "Both mention checkpoints, but one adds a checkpoint limit argument and the other refactors checkpoint naming; not the same bug or change." + }, + { + "left": "pull_request:7632", + "right": "pull_request:9061", + "accept": false, + "reason": "Different problems: DreamBooth VAE loading path issue vs sharded-checkpoint loading with variants." + }, + { + "left": "pull_request:2467", + "right": "pull_request:823", + "accept": false, + "reason": "Different pipeline bugs: latent output type addition vs negative-prompt type mismatch handling." + }, + { + "left": "pull_request:3299", + "right": "pull_request:4761", + "accept": false, + "reason": "Docs/example sync for LoRA training vs Paint-by-Example image encoder fix; unrelated." + }, + { + "left": "pull_request:4315", + "right": "pull_request:5298", + "accept": true, + "reason": "Same underlying LoRA torch.compile compatibility problem in the conv path; one fixes the no-LoRA branch and the other fixes the LoRA-enabled branch, so they plausibly belong in one PR." + }, + { + "left": "pull_request:4508", + "right": "pull_request:5948", + "accept": false, + "reason": "Different pipeline_utils issues: use_onnx propagation vs variant warning logic." + }, + { + "left": "pull_request:4795", + "right": "pull_request:4862", + "accept": false, + "reason": "Different ControlNet bugs: torch.compile condition handling vs SDXL ControlNet img2img shape handling for MultiControlNetModel." + }, + { + "left": "pull_request:5738", + "right": "pull_request:5958", + "accept": false, + "reason": "Different LoRA features: delete adapters vs load_lora support refactor." + }, + { + "left": "pull_request:6217", + "right": "pull_request:7786", + "accept": false, + "reason": "Same Stable Video Diffusion area but different issues: VAE compile compatibility vs num_videos_per_prompt behavior." + }, + { + "left": "pull_request:823", + "right": "pull_request:826", + "accept": false, + "reason": "Different bugs in different code paths: negative-prompt type checking vs DreamBooth fp16 loss casting." + }, + { + "left": "pull_request:4899", + "right": "pull_request:5127", + "accept": false, + "reason": "Related adapter/controlnet ecosystem, but one is ControlNet LoRA support and the other is Multi-Adapter support; not the same change." + }, + { + "left": "pull_request:10230", + "right": "pull_request:10544", + "accept": false, + "reason": "Both docs PRs, but they cover unrelated topics: LoRA loader docs vs ParaAttention optimization docs." + }, + { + "left": "pull_request:10318", + "right": "pull_request:11595", + "accept": false, + "reason": "Flux LoRA integration tests vs a LoRA loading fix; same broad domain but different concrete changes." + }, + { + "left": "pull_request:11479", + "right": "pull_request:12451", + "accept": false, + "reason": "Unrelated docs/config changes: Intel Gaudi wording update vs doc-builder Dockerfile packages." + }, + { + "left": "pull_request:11843", + "right": "pull_request:12541", + "accept": false, + "reason": "Both Wan LoRA related, but one is tests for exclude_modules and the other is support for latest few-step Wan LoRA; different changes." + }, + { + "left": "pull_request:12681", + "right": "pull_request:13042", + "accept": false, + "reason": "Different CI/test fixes in unrelated areas: CPU pipeline tests vs QwenImage prompt isolation." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13211", + "accept": false, + "reason": "Both involve offloading tests/fixes, but for different models and different bugs; not the same code-path problem." + }, + { + "left": "pull_request:2172", + "right": "pull_request:3080", + "accept": false, + "reason": "Different training/pipeline concerns: multi-GPU validation behavior vs save_pretrained handling of None entries." + }, + { + "left": "pull_request:2464", + "right": "pull_request:3339", + "accept": false, + "reason": "Unrelated changes: xformers handling in a LoRA training script vs urllib3 dependency pinning." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: it correctly characterizes the cluster as mostly heterogeneous and only treats the LoRA torch.compile pair as plausibly mergeable for the same bugfix. The accepted soft-edge verdict on 4315/5298 is supported by the shared file and matching compile-fix theme, and the false verdicts avoid overclaiming duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 104, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10261", + "pull_request:10366", + "pull_request:10717", + "pull_request:10854", + "pull_request:11124", + "pull_request:11756", + "pull_request:12024", + "pull_request:12041", + "pull_request:1269", + "pull_request:1527", + "pull_request:192", + "pull_request:2915", + "pull_request:2983", + "pull_request:3127", + "pull_request:3316", + "pull_request:3707", + "pull_request:3847", + "pull_request:3850", + "pull_request:4083", + "pull_request:4098", + "pull_request:4845", + "pull_request:4886", + "pull_request:4989", + "pull_request:4999", + "pull_request:5010", + "pull_request:5093", + "pull_request:5477", + "pull_request:5559", + "pull_request:6346", + "pull_request:6355", + "pull_request:6787", + "pull_request:6828", + "pull_request:6864", + "pull_request:7016", + "pull_request:7096", + "pull_request:7103", + "pull_request:7156", + "pull_request:7246", + "pull_request:7328", + "pull_request:7458", + "pull_request:7613", + "pull_request:7820", + "pull_request:7932", + "pull_request:8797", + "pull_request:8901", + "pull_request:9175", + "pull_request:9542" + ], + "soft_pairs": [ + "pull_request:2983|pull_request:3847", + "pull_request:3127|pull_request:4886", + "pull_request:6346|pull_request:7096", + "pull_request:6864|pull_request:7328", + "pull_request:7103|pull_request:7156", + "pull_request:7246|pull_request:7458", + "pull_request:7932|pull_request:8797", + "pull_request:5477|pull_request:5559", + "pull_request:5010|pull_request:6787", + "pull_request:6355|pull_request:7613", + "pull_request:6828|pull_request:7820", + "pull_request:4083|pull_request:4098", + "pull_request:9175|pull_request:9542", + "pull_request:12024|pull_request:12041", + "pull_request:7016|pull_request:8901", + "pull_request:10261|pull_request:10717", + "pull_request:10366|pull_request:11756", + "pull_request:10854|pull_request:11124", + "pull_request:1269|pull_request:192", + "pull_request:1527|pull_request:192", + "pull_request:2915|pull_request:5093", + "pull_request:3316|pull_request:3707", + "pull_request:3850|pull_request:4845", + "pull_request:4989|pull_request:4999" + ], + "prepared_review_unit_hash": "dc544295298a335045f60b76e828c05ea2af90425b409bc0a7e154a79e46001f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 97728, + "estimated_input_tokens": 24432, + "estimated_eval_tokens": 49120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated PRs that only weakly resemble one another by broad keywords like tests, docs, IP-Adapter, or model-family names. I found no true duplicate pair under a strict same-bug/same-change standard.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 7016 is the strongest representative of the only recurring technical theme here (IP-Adapter plumbing): it is broad, well-reviewed, and its changes are reused by later, narrower IP-Adapter follow-ups.", + "best_issue_reason": null, + "best_pr_reason": "PR 7016 is the best global representative because it has the widest functional surface area and the clearest downstream relevance among the items; the rest are mostly isolated fixes, docs, or tests.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2983", + "right": "pull_request:3847", + "accept": false, + "reason": "Docs for installing test deps vs adding a dependency-check test workflow; different changes and different failure modes." + }, + { + "left": "pull_request:3127", + "right": "pull_request:4886", + "accept": false, + "reason": "Both touch safety checker behavior, but one fixes img2img processing and the other fixes upscale safety-checker input handling; not the same code path." + }, + { + "left": "pull_request:6346", + "right": "pull_request:7096", + "accept": false, + "reason": "Training-example follow-up vs ControlNet training enhancements; unrelated scripts and functionality." + }, + { + "left": "pull_request:6864", + "right": "pull_request:7328", + "accept": false, + "reason": "One is a docs note about torch.compile with LoRA fusion; the other is a packaging workflow tweak adding torch to a publish step." + }, + { + "left": "pull_request:7103", + "right": "pull_request:7156", + "accept": false, + "reason": "ControlNet input validation bug vs stale-bot label normalization; no shared underlying bug." + }, + { + "left": "pull_request:7246", + "right": "pull_request:7458", + "accept": false, + "reason": "VAE tiling test-device fix vs skipping dynamo tests on Python 3.12; unrelated test maintenance." + }, + { + "left": "pull_request:7932", + "right": "pull_request:8797", + "accept": false, + "reason": "LoRA test expectation update vs sharded checkpoint test fixes; different test areas and causes." + }, + { + "left": "pull_request:5477", + "right": "pull_request:5559", + "accept": false, + "reason": "Related family name, but one adds a community pipeline and the other adds an img2img pipeline; different features and code paths." + }, + { + "left": "pull_request:5010", + "right": "pull_request:6787", + "accept": false, + "reason": "GLIGEN import/export fix vs Marigold input-range correction; unrelated subsystems." + }, + { + "left": "pull_request:6355", + "right": "pull_request:7613", + "accept": false, + "reason": "Both are advanced-training docs/card tweaks, but they fix different scripts and different metadata problems." + }, + { + "left": "pull_request:6828", + "right": "pull_request:7820", + "accept": false, + "reason": "Both mention IP-Adapter, but one changes UNetMotionModel internals and the other wires IP-Adapter into InstructPix2Pix pipelines; not one concrete bug." + }, + { + "left": "pull_request:4083", + "right": "pull_request:4098", + "accept": false, + "reason": "SDXL VAE precision support vs adding SDXL inpainting; these are separate features and not mergeable as one fix." + }, + { + "left": "pull_request:9175", + "right": "pull_request:9542", + "accept": false, + "reason": "Union ControlNet support vs Kohya Flux LoRA text-encoder conversion; completely different model/tooling changes." + }, + { + "left": "pull_request:12024", + "right": "pull_request:12041", + "accept": false, + "reason": "Wan2.2 follow-up pipeline/model fix vs VAE patchify correction; different bugs in different components." + }, + { + "left": "pull_request:7016", + "right": "pull_request:8901", + "accept": false, + "reason": "Both are IP-Adapter-related, but 7016 is a broad loader/API refactor while 8901 adds Kolors support with its own image-encoder requirements; not the same change." + }, + { + "left": "pull_request:10261", + "right": "pull_request:10717", + "accept": false, + "reason": "Both concern Flux IP-Adapter, but one is the core support PR and the other adds img2img/control-pipeline wiring; separate code paths." + }, + { + "left": "pull_request:10366", + "right": "pull_request:11756", + "accept": false, + "reason": "Sana transformer/offload fix vs SANA test cleanup; unrelated." + }, + { + "left": "pull_request:10854", + "right": "pull_request:11124", + "accept": false, + "reason": "Hunyuan DiT test-module rename vs CogView4 test update; different test targets." + }, + { + "left": "pull_request:1269", + "right": "pull_request:192", + "accept": false, + "reason": "Stable Diffusion refactor cleanup vs README example parameter update; unrelated scope." + }, + { + "left": "pull_request:1527", + "right": "pull_request:192", + "accept": false, + "reason": "Inpaint tensor-input support vs README example parameter update; one is code behavior, the other is documentation." + }, + { + "left": "pull_request:2915", + "right": "pull_request:5093", + "accept": false, + "reason": "Text inversion tolerance test tweak vs moving many slow tests to nightly; separate test-policy changes." + }, + { + "left": "pull_request:3316", + "right": "pull_request:3707", + "accept": false, + "reason": "Both are documentation workflow changes, but one changes release branch matching and the other switches secrets/workflow triggering; distinct fixes." + }, + { + "left": "pull_request:3850", + "right": "pull_request:4845", + "accept": false, + "reason": "AudioDiffusion test adjustments vs Stable Diffusion inpaint input handling; unrelated pipelines." + }, + { + "left": "pull_request:4989", + "right": "pull_request:4999", + "accept": false, + "reason": "Two Wuerstchen PRs, but one fixes num_images_per_prompt and the other fixes compel usage; different bugs and merge intents." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 105, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10205", + "pull_request:10292", + "pull_request:10378", + "pull_request:10454", + "pull_request:10544", + "pull_request:11059", + "pull_request:11428", + "pull_request:11485", + "pull_request:11726", + "pull_request:12249", + "pull_request:12451", + "pull_request:12679", + "pull_request:13100", + "pull_request:2479", + "pull_request:4062", + "pull_request:4396", + "pull_request:4910", + "pull_request:4944", + "pull_request:5030", + "pull_request:5057", + "pull_request:5157", + "pull_request:5470", + "pull_request:5702", + "pull_request:6009", + "pull_request:6099", + "pull_request:6191", + "pull_request:6241", + "pull_request:6330", + "pull_request:6499", + "pull_request:6828", + "pull_request:6904", + "pull_request:6972", + "pull_request:7158", + "pull_request:7226", + "pull_request:7476", + "pull_request:7550", + "pull_request:8490", + "pull_request:8688", + "pull_request:8691", + "pull_request:8916", + "pull_request:8985", + "pull_request:9017", + "pull_request:9040", + "pull_request:9244", + "pull_request:9311", + "pull_request:9367", + "pull_request:9542", + "pull_request:9993" + ], + "soft_pairs": [ + "pull_request:5030|pull_request:5702", + "pull_request:6009|pull_request:6099", + "pull_request:6191|pull_request:8490", + "pull_request:6499|pull_request:7226", + "pull_request:8688|pull_request:8691", + "pull_request:8916|pull_request:8985", + "pull_request:11726|pull_request:12679", + "pull_request:4062|pull_request:5057", + "pull_request:4944|pull_request:6828", + "pull_request:10454|pull_request:11428", + "pull_request:11485|pull_request:12249", + "pull_request:10205|pull_request:10378", + "pull_request:10205|pull_request:11059", + "pull_request:10378|pull_request:9993", + "pull_request:10544|pull_request:9244", + "pull_request:12451|pull_request:13100", + "pull_request:2479|pull_request:4396", + "pull_request:4910|pull_request:6330", + "pull_request:5157|pull_request:5470", + "pull_request:6241|pull_request:6904", + "pull_request:6972|pull_request:7158", + "pull_request:7476|pull_request:9311", + "pull_request:9017|pull_request:9542", + "pull_request:7550|pull_request:9040", + "pull_request:10292|pull_request:9367" + ], + "prepared_review_unit_hash": "e185837165f20f52bae84828e86edaf1db0ef4e2fc4f3ce8844ba523cd96698e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 96552, + "estimated_input_tokens": 24138, + "estimated_eval_tokens": 48532 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs, loader fixes, pipeline features, and example scripts across different subsystems. The soft links are title/subsystem similarity only, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:5030", + "right": "pull_request:5702", + "accept": false, + "reason": "Different bugs: model offload key handling vs AnimateDiff prompt handling." + }, + { + "left": "pull_request:6009", + "right": "pull_request:6099", + "accept": false, + "reason": "Different TensorRT and TextToVideoZero constructor fixes; unrelated code paths." + }, + { + "left": "pull_request:6191", + "right": "pull_request:8490", + "accept": false, + "reason": "Both are README updates for different examples, not the same change." + }, + { + "left": "pull_request:6499", + "right": "pull_request:7226", + "accept": false, + "reason": "Different docs/API work: PEFT docs vs IP-Adapter embedding support." + }, + { + "left": "pull_request:8688", + "right": "pull_request:8691", + "accept": false, + "reason": "Both touch LoRA loading, but one fixes DoRA conversion and the other vanilla loading logic; not the same bug." + }, + { + "left": "pull_request:8916", + "right": "pull_request:8985", + "accept": false, + "reason": "Different targets: adding Hunyuan DiT tests vs resizing Lumina test models." + }, + { + "left": "pull_request:11726", + "right": "pull_request:12679", + "accept": false, + "reason": "Both are LoRA-related, but Wan LoRA conversion and ChronoEdit multi-LoRA support are separate features." + }, + { + "left": "pull_request:4062", + "right": "pull_request:5057", + "accept": false, + "reason": "Shap-E mesh output vs SDXL clip-skip support are unrelated pipeline features." + }, + { + "left": "pull_request:4944", + "right": "pull_request:6828", + "accept": false, + "reason": "IP-Adapter support and UNetMotionModel image-conditioning fixes are different implementations." + }, + { + "left": "pull_request:10454", + "right": "pull_request:11428", + "accept": false, + "reason": "Hunyuan Video attention-mask dim fix vs Framepack pipeline addition are not the same code-path problem." + }, + { + "left": "pull_request:11485", + "right": "pull_request:12249", + "accept": false, + "reason": "Both add pipelines, but Flex.2-preview and Step1X-Edit are distinct models and codebases." + }, + { + "left": "pull_request:10205", + "right": "pull_request:10378", + "accept": false, + "reason": "InstructPix2Pix LoRA refactor and SD3 LoRA training distributed support are unrelated training scripts." + }, + { + "left": "pull_request:10205", + "right": "pull_request:11059", + "accept": false, + "reason": "Different LoRA work: InstructPix2Pix training refactor vs Wan LoRA conversion support." + }, + { + "left": "pull_request:10378", + "right": "pull_request:9993", + "accept": false, + "reason": "Same file family, but different SD3 checkpoint-saving fixes and distinct save-path logic." + }, + { + "left": "pull_request:10544", + "right": "pull_request:9244", + "accept": false, + "reason": "ParaAttention docs vs Flux single-file pipeline support are unrelated changes." + }, + { + "left": "pull_request:12451", + "right": "pull_request:13100", + "accept": false, + "reason": "Dockerfile dependency update vs ModularPipeline docs/runtime refactor are unrelated." + }, + { + "left": "pull_request:2479", + "right": "pull_request:4396", + "accept": false, + "reason": "Text-encoder LoRA training support vs prompt-encoding fallback logic are not the same bug." + }, + { + "left": "pull_request:4910", + "right": "pull_request:6330", + "accept": false, + "reason": "Inpainting docs typo fix vs SVD docs rendering fix are separate documentation edits." + }, + { + "left": "pull_request:5157", + "right": "pull_request:5470", + "accept": false, + "reason": "Different ControlNet/LoRA conversion fixes; one is kohya conversion parsing, the other is SDXL inpaint docs/code cleanup." + }, + { + "left": "pull_request:6241", + "right": "pull_request:6904", + "accept": false, + "reason": "Runtime error report vs load_image PIL support fix; no shared underlying change." + }, + { + "left": "pull_request:6972", + "right": "pull_request:7158", + "accept": false, + "reason": "IP-Adapter unload test correction vs LCM benchmark fix are unrelated." + }, + { + "left": "pull_request:7476", + "right": "pull_request:9311", + "accept": false, + "reason": "Different SDXL ControlNet issues: multi-controlnet compile behavior vs CFG cutoff callbacks." + }, + { + "left": "pull_request:9017", + "right": "pull_request:9542", + "accept": false, + "reason": "AuraFlow LoRA support and Flux LoRA text-encoder handling are separate model integrations." + }, + { + "left": "pull_request:7550", + "right": "pull_request:9040", + "accept": false, + "reason": "Different community pipelines for different models (SDXL differential diffusion vs HunyuanDiT differential diffusion)." + }, + { + "left": "pull_request:10292", + "right": "pull_request:9367", + "accept": false, + "reason": "Both touch AutoPipeline and Flux, but one adds Flux Control and the other adds Flux inpaint/img2img/controlnet routing; not the same concrete fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 106, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10136", + "pull_request:10199", + "pull_request:10211", + "pull_request:10270", + "pull_request:10286", + "pull_request:10316", + "pull_request:10481", + "pull_request:10524", + "pull_request:10578", + "pull_request:10581", + "pull_request:10648", + "pull_request:10825", + "pull_request:1104", + "pull_request:11058", + "pull_request:11099", + "pull_request:1129", + "pull_request:11428", + "pull_request:11545", + "pull_request:11660", + "pull_request:11716", + "pull_request:11802", + "pull_request:11835", + "pull_request:12018", + "pull_request:12170", + "pull_request:12275", + "pull_request:1230", + "pull_request:12418", + "pull_request:12464", + "pull_request:1701", + "pull_request:1765", + "pull_request:2129", + "pull_request:2476", + "pull_request:2956", + "pull_request:3001", + "pull_request:3020", + "pull_request:4417", + "pull_request:5807", + "pull_request:5938", + "pull_request:6828", + "pull_request:7186", + "pull_request:7944", + "pull_request:796", + "pull_request:8797", + "pull_request:9015", + "pull_request:9070", + "pull_request:9379", + "pull_request:9881" + ], + "soft_pairs": [ + "pull_request:5938|pull_request:7944", + "pull_request:10581|pull_request:9881", + "pull_request:11716|pull_request:11802", + "pull_request:11835|pull_request:12418", + "pull_request:4417|pull_request:5807", + "pull_request:2129|pull_request:2476", + "pull_request:8797|pull_request:9379", + "pull_request:6828|pull_request:7186", + "pull_request:10136|pull_request:11428", + "pull_request:3001|pull_request:3020", + "pull_request:2956|pull_request:3001", + "pull_request:10199|pull_request:9070", + "pull_request:10211|pull_request:9015", + "pull_request:10270|pull_request:11716", + "pull_request:10286|pull_request:10481", + "pull_request:10316|pull_request:10825", + "pull_request:10524|pull_request:10825", + "pull_request:10578|pull_request:11660", + "pull_request:10648|pull_request:11545", + "pull_request:1104|pull_request:796", + "pull_request:11058|pull_request:11099", + "pull_request:1129|pull_request:1765", + "pull_request:12018|pull_request:12464", + "pull_request:12170|pull_request:12275", + "pull_request:1230|pull_request:1701" + ], + "prepared_review_unit_hash": "a6254a62523a283c00d47ba6fe26dcbc77d2e9b7a6bfb33c4db2f49198af2091", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 100602, + "estimated_input_tokens": 25151, + "estimated_eval_tokens": 50558 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose set of superficially similar but mostly unrelated PRs across docs, tests, schedulers, LoRA, pipelines, and CI. I would not treat any soft edge as a true duplicate/merge candidate; all should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the cluster spans unrelated features and fixes across different subsystems, so there is no representative PR that subsumes the rest.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global representative for this heterogeneous cluster; picking one would incorrectly imply the rest are duplicates of the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5938", + "right": "pull_request:7944", + "accept": false, + "reason": "Different changes: ControlNet LoRA support vs PAG support. They only share broad loader/pipeline plumbing, not the same bug or change." + }, + { + "left": "pull_request:10581", + "right": "pull_request:9881", + "accept": false, + "reason": "Different code paths: SD3 inpaint IP-Adapter support vs IP-Adapter XFormers attention processor support. Same broad area, but not one mergeable fix." + }, + { + "left": "pull_request:11716", + "right": "pull_request:11802", + "accept": false, + "reason": "Flux LoRA metadata/output handling vs lora deprecation cleanup. The second is API cleanup, not the same underlying bug." + }, + { + "left": "pull_request:11835", + "right": "pull_request:12418", + "accept": false, + "reason": "One fixes float16 CUDA test behavior; the other xfails a legacy-variant download test due to a transformers import issue. Unrelated test failures." + }, + { + "left": "pull_request:4417", + "right": "pull_request:5807", + "accept": false, + "reason": "Zero123 community pipeline support vs LPW SDXL prompt-embedding duplication. Different features and unrelated code paths." + }, + { + "left": "pull_request:2129", + "right": "pull_request:2476", + "accept": false, + "reason": "LoRA passthrough in pipelines vs Attend-and-Excite image-size support. Both touch pipelines, but they fix different problems." + }, + { + "left": "pull_request:8797", + "right": "pull_request:9379", + "accept": false, + "reason": "Sharded checkpoint test fixes vs Flux LoRA training/test GPU fixes. Shared theme is tests, but not the same defect." + }, + { + "left": "pull_request:6828", + "right": "pull_request:7186", + "accept": false, + "reason": "UNetMotionModel IP-Adapter handling update vs moving IP-Adapter Face ID into core. Related ecosystem, but different concrete functionality." + }, + { + "left": "pull_request:10136", + "right": "pull_request:11428", + "accept": false, + "reason": "Hunyuan Video base pipeline/model integration vs Hunyuan Video Framepack extension. Same family, different pipelines and changesets." + }, + { + "left": "pull_request:3001", + "right": "pull_request:3020", + "accept": false, + "reason": "Both are scheduler-related, but one adds Karras handling to DPMSolverMultistep while the other adds a stochastic DPM++ SDE sampler. Different algorithms." + }, + { + "left": "pull_request:2956", + "right": "pull_request:3001", + "accept": false, + "reason": "Both mention Karras sigmas, but they modify different schedulers and different implementations. Not the same concrete code-path fix." + }, + { + "left": "pull_request:10199", + "right": "pull_request:9070", + "accept": false, + "reason": "RF inversion eta decay tweak in a community pipeline vs introducing the Flux img2img pipeline. Completely different functionality." + }, + { + "left": "pull_request:10211", + "right": "pull_request:9015", + "accept": false, + "reason": "Checkpoint-id example fix for CogView3Plus vs PIA example/docs update. Both are example/doc corrections, not the same artifact or bug." + }, + { + "left": "pull_request:10270", + "right": "pull_request:11716", + "accept": false, + "reason": "CogVideoX embedding initialization fix vs Flux LoRA metadata return fix. No shared underlying issue." + }, + { + "left": "pull_request:10286", + "right": "pull_request:10481", + "accept": false, + "reason": "Sana README checkpoint path update vs Sana pipeline default clean_caption change. One is docs, the other is runtime behavior." + }, + { + "left": "pull_request:10316", + "right": "pull_request:10825", + "accept": false, + "reason": "LTX video docs tweak vs text2image training docs update. Both are documentation-only and unrelated topics." + }, + { + "left": "pull_request:10524", + "right": "pull_request:10825", + "accept": false, + "reason": "Hunyuan Video docs checkpoint correction vs text2image training docs addition. Different docs for different pipelines." + }, + { + "left": "pull_request:10578", + "right": "pull_request:11660", + "accept": false, + "reason": "Flux 4bit LoRA loading support vs HiDream LoRA conversion support. Same loader subsystem, but different model families and code paths." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11545", + "accept": false, + "reason": "DDPM docs update for sigmoid schedule vs Flux img2img VAE slicing/tiling methods. One is docs-only, the other is pipeline behavior." + }, + { + "left": "pull_request:1104", + "right": "pull_request:796", + "accept": false, + "reason": "Both touch CI/M1 testing, but one adds slow MPS tests while the other sets up Apple M1 tests. Separate infrastructure changes." + }, + { + "left": "pull_request:11058", + "right": "pull_request:11099", + "accept": false, + "reason": "Device-agnostic test cleanup vs conditioning PR GPU tests on styling/quality checks. Different test infrastructure fixes." + }, + { + "left": "pull_request:1129", + "right": "pull_request:1765", + "accept": false, + "reason": "Both involve DreamBooth Flax, but one fixes sample batch sizing while the other is a broader Flax fixes PR. Not the same bug." + }, + { + "left": "pull_request:12018", + "right": "pull_request:12464", + "accept": false, + "reason": "Quantization docs link rename vs Marigold docs syntax cleanup. Completely unrelated documentation changes." + }, + { + "left": "pull_request:12170", + "right": "pull_request:12275", + "accept": false, + "reason": "Qwen LoRA conversion support vs TorchAO config-class support. Different feature areas and unrelated changes." + }, + { + "left": "pull_request:1230", + "right": "pull_request:1701", + "accept": false, + "reason": "MPS generator fast-test fix vs RePaint fast tests/API conformance. Both are tests/pipeline cleanup, but not the same issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 107, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:12856", + "pull_request:13056", + "pull_request:1408", + "pull_request:1416", + "pull_request:1436", + "pull_request:1605", + "pull_request:168", + "pull_request:1849", + "pull_request:1879", + "pull_request:1927", + "pull_request:2103", + "pull_request:2400", + "pull_request:2446", + "pull_request:2693", + "pull_request:2974", + "pull_request:3389", + "pull_request:3398", + "pull_request:3680", + "pull_request:3755", + "pull_request:392", + "pull_request:3945", + "pull_request:3971", + "pull_request:402", + "pull_request:4211", + "pull_request:4258", + "pull_request:4343", + "pull_request:4457", + "pull_request:4498", + "pull_request:4503", + "pull_request:4582", + "pull_request:4595", + "pull_request:4659", + "pull_request:4661", + "pull_request:4763", + "pull_request:4795", + "pull_request:4856", + "pull_request:4934", + "pull_request:5157", + "pull_request:5180", + "pull_request:5275", + "pull_request:5289", + "pull_request:5367", + "pull_request:5499", + "pull_request:6289", + "pull_request:6305", + "pull_request:6757", + "pull_request:688", + "pull_request:874" + ], + "soft_pairs": [ + "pull_request:12856|pull_request:13056", + "pull_request:1408|pull_request:1416", + "pull_request:1436|pull_request:688", + "pull_request:1605|pull_request:1849", + "pull_request:168|pull_request:1927", + "pull_request:1879|pull_request:3389", + "pull_request:2103|pull_request:3945", + "pull_request:2400|pull_request:874", + "pull_request:2446|pull_request:3398", + "pull_request:2693|pull_request:4856", + "pull_request:2974|pull_request:4457", + "pull_request:3680|pull_request:5499", + "pull_request:3755|pull_request:5289", + "pull_request:392|pull_request:402", + "pull_request:3945|pull_request:4934", + "pull_request:3971|pull_request:5180", + "pull_request:4211|pull_request:5275", + "pull_request:4258|pull_request:4503", + "pull_request:4343|pull_request:6305", + "pull_request:4498|pull_request:4763", + "pull_request:4582|pull_request:5367", + "pull_request:4595|pull_request:6289", + "pull_request:4659|pull_request:6757", + "pull_request:4661|pull_request:6757", + "pull_request:4795|pull_request:5157" + ], + "prepared_review_unit_hash": "351d4a73af617569e50964d71a0476e178aa3ad74133d9d879ca6612f39d6be7", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 96757, + "estimated_input_tokens": 24190, + "estimated_eval_tokens": 48636 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous set of unrelated diffusers PRs spanning docs, example scripts, pipeline additions, refactors, and scheduler/loaders fixes. I don\u2019t see a single underlying bug or change tying the items together, so none of the soft edges should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:12856", + "right": "pull_request:13056", + "accept": false, + "reason": "Same model family, but one adds Z-Image implementation docs and the other adds a Z-Image LoRA training example; different artifacts and changes." + }, + { + "left": "pull_request:1408", + "right": "pull_request:1416", + "accept": false, + "reason": "PR 1408 fixes sample-size conversion; PR 1416 fixes ONNX conversion/inference behavior and safety checker handling. Different code paths." + }, + { + "left": "pull_request:1436", + "right": "pull_request:688", + "accept": false, + "reason": "Both touch Flax loading, but 1436 is a small wording/import cleanup while 688 adds broader Flax-without-PyTorch support. Not the same concrete fix." + }, + { + "left": "pull_request:1605", + "right": "pull_request:1849", + "accept": false, + "reason": "One adds upcast_attention plumbing, the other passes cross_attention_kwargs through Versatile Diffusion attention. Different features in the same area." + }, + { + "left": "pull_request:168", + "right": "pull_request:1927", + "accept": false, + "reason": "PR 168 adds Stable Diffusion pipeline code; PR 1927 rewrites a Stable Diffusion docs guide. Not duplicate changes." + }, + { + "left": "pull_request:1879", + "right": "pull_request:3389", + "accept": false, + "reason": "1879 is the attention-file refactor; 3389 is a draft integration-test run for that refactor. Related topic, but not the same change." + }, + { + "left": "pull_request:2103", + "right": "pull_request:3945", + "accept": false, + "reason": "Both edit DreamBooth LoRA scripts, but one only adds a model tag while the other introduces a rank argument. Distinct changes." + }, + { + "left": "pull_request:2400", + "right": "pull_request:874", + "accept": false, + "reason": "Both affect community docs, but they document different community examples and different behaviors. Too broad to be the same issue." + }, + { + "left": "pull_request:2446", + "right": "pull_request:3398", + "accept": false, + "reason": "One fixes an instruct-pix2pix code snippet; the other corrects a typo in stable_diffusion.mdx. Different doc fixes." + }, + { + "left": "pull_request:2693", + "right": "pull_request:4856", + "accept": false, + "reason": "One fixes an inpaint doc image link; the other fixes an SDXL guide link. Unrelated documentation issues." + }, + { + "left": "pull_request:2974", + "right": "pull_request:4457", + "accept": false, + "reason": "Both involve TensorRT/community pipeline docs, but one adds an SD/txt2img pipeline while the other fixes an inpaint pipeline. Different pipelines." + }, + { + "left": "pull_request:3680", + "right": "pull_request:5499", + "accept": false, + "reason": "3680 is a specific loader-method link fix; 5499 is a broad links cleanup across many API docs. Not the same bug." + }, + { + "left": "pull_request:3755", + "right": "pull_request:5289", + "accept": false, + "reason": "Both relax test tolerances, but they are in different test modules for different pipelines. Separate flaky-test fixes." + }, + { + "left": "pull_request:392", + "right": "pull_request:402", + "accept": false, + "reason": "Both are docs overhauls, but one is a conceptual contribution page and the other is a toctree/intro restructure. Not the same underlying change." + }, + { + "left": "pull_request:3945", + "right": "pull_request:4934", + "accept": false, + "reason": "3945 adds a DreamBooth LoRA rank parameter; 4934 adds T2I-Adapter training docs/scripts. Different training workflows." + }, + { + "left": "pull_request:3971", + "right": "pull_request:5180", + "accept": false, + "reason": "3971 is an SDXL docs/build workflow fix; 5180 is a Wuerstchen citation/doc-builder fix. Different docs problems." + }, + { + "left": "pull_request:4211", + "right": "pull_request:5275", + "accept": false, + "reason": "4211 fixes SDXL inpainting docstrings; 5275 adds an SDXL research example. Different purpose and code." + }, + { + "left": "pull_request:4258", + "right": "pull_request:4503", + "accept": false, + "reason": "Both mention SDXL/LoRA docs, but one revises an SDXL README note and the other clarifies LoRA/Kohya support. Different documentation changes." + }, + { + "left": "pull_request:4343", + "right": "pull_request:6305", + "accept": false, + "reason": "4343 adds a DreamBooth LoRA rank argument; 6305 adds zero-terminal-SNR rescaling to DDPMScheduler. Different concrete code paths." + }, + { + "left": "pull_request:4498", + "right": "pull_request:4763", + "accept": false, + "reason": "4498 adds an SDXL reference pipeline; 4763 adds a LoRA-to-safetensors conversion script. Completely different functionality." + }, + { + "left": "pull_request:4582", + "right": "pull_request:5367", + "accept": false, + "reason": "4582 fixes a ControlNet SDXL docstring example; 5367 fixes SDXL tutorial pipeline creation. Different documentation targets." + }, + { + "left": "pull_request:4595", + "right": "pull_request:6289", + "accept": false, + "reason": "4595 is an SDXL conversion-pipeline WIP; 6289 adds AnimateDiff SparseCntrl. Different features and codebases." + }, + { + "left": "pull_request:4659", + "right": "pull_request:6757", + "accept": false, + "reason": "4659 adds an SDXL long-weighted-prompt pipeline; 6757 moves sigmas to device in k-diffusion. Unrelated changes." + }, + { + "left": "pull_request:4661", + "right": "pull_request:6757", + "accept": false, + "reason": "4661 also adds the long-weighted-prompt community pipeline, while 6757 is a scheduler/device fix. No duplicate bug or change." + }, + { + "left": "pull_request:4795", + "right": "pull_request:5157", + "accept": false, + "reason": "4795 fixes torch.compile behavior in ControlNet; 5157 fixes LoRA naming/conversion logic in loaders. Different code paths and symptoms." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 108, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10589", + "pull_request:10867", + "pull_request:10941", + "pull_request:11481", + "pull_request:11516", + "pull_request:11572", + "pull_request:11684", + "pull_request:11716", + "pull_request:11767", + "pull_request:12269", + "pull_request:12275", + "pull_request:12315", + "pull_request:12490", + "pull_request:12681", + "pull_request:12736", + "pull_request:12782", + "pull_request:12973", + "pull_request:130", + "pull_request:13169", + "pull_request:13323", + "pull_request:13330", + "pull_request:1579", + "pull_request:2009", + "pull_request:2325", + "pull_request:2994", + "pull_request:365", + "pull_request:4115", + "pull_request:473", + "pull_request:5039", + "pull_request:5117", + "pull_request:5511", + "pull_request:5698", + "pull_request:5982", + "pull_request:6163", + "pull_request:6291", + "pull_request:6311", + "pull_request:6528", + "pull_request:6875", + "pull_request:6999", + "pull_request:7029", + "pull_request:7050", + "pull_request:7181", + "pull_request:7301", + "pull_request:7319", + "pull_request:7550", + "pull_request:8856", + "pull_request:9015", + "pull_request:9423" + ], + "soft_pairs": [ + "pull_request:5039|pull_request:5982", + "pull_request:5698|pull_request:6528", + "pull_request:6163|pull_request:6291", + "pull_request:6311|pull_request:7319", + "pull_request:6875|pull_request:7301", + "pull_request:7029|pull_request:9015", + "pull_request:7050|pull_request:7181", + "pull_request:11516|pull_request:11572", + "pull_request:4115|pull_request:5511", + "pull_request:10589|pull_request:10867", + "pull_request:12269|pull_request:13169", + "pull_request:7550|pull_request:9423", + "pull_request:6999|pull_request:8856", + "pull_request:10941|pull_request:11716", + "pull_request:12736|pull_request:12973", + "pull_request:2994|pull_request:5117", + "pull_request:11684|pull_request:12275", + "pull_request:11481|pull_request:12315", + "pull_request:11767|pull_request:12782", + "pull_request:12490|pull_request:13330", + "pull_request:12681|pull_request:13323", + "pull_request:130|pull_request:1579", + "pull_request:2009|pull_request:473", + "pull_request:2325|pull_request:365" + ], + "prepared_review_unit_hash": "91be231d9894af617e30cb22cdbb0158d77402ff2eb14a8349a5c0c06447a4f4", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 102683, + "estimated_input_tokens": 25671, + "estimated_eval_tokens": 51598 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These PRs are mostly unrelated; the soft-similarity pairs share broad themes (docs, tests, IP-Adapter, LoRA, scheduling) but not the same concrete bug or change.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "No true duplicate cluster emerges here. If forced to pick one representative PR, 10867 is the strongest anchor because it has an explicit issue target and a substantial multi-file code change; the rest are distinct fixes, docs, or tests.", + "best_issue_reason": null, + "best_pr_reason": "10867 is the best single PR to represent the set only in the weak sense that it is the most substantive and clearly scoped change, but it is not a duplicate of the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5039", + "right": "pull_request:5982", + "accept": false, + "reason": "Both are docs, but one is SDXL textual inversion and the other is SDXL Turbo docs; different topics and deliverables." + }, + { + "left": "pull_request:5698", + "right": "pull_request:6528", + "accept": false, + "reason": "Different bugs in different pipelines: LCM img2img vs ControlNet img2img IP-Adapter handling." + }, + { + "left": "pull_request:6163", + "right": "pull_request:6291", + "accept": false, + "reason": "Both are LoRA tests, but one adds xformers-attention coverage while the other changes the unfusion tolerance check; not the same underlying fix." + }, + { + "left": "pull_request:6311", + "right": "pull_request:7319", + "accept": false, + "reason": "Both touch docs navigation, but one adds a fast diffusion tutorial and the other documents EDM schedulers." + }, + { + "left": "pull_request:6875", + "right": "pull_request:7301", + "accept": false, + "reason": "Different test fixes in different subsystems: IP-Adapter output slice vs VAE scaling constant." + }, + { + "left": "pull_request:7029", + "right": "pull_request:9015", + "accept": false, + "reason": "Both are documentation-only, but one fixes AnimateLCM doc links/alt text while the other updates a PIA example." + }, + { + "left": "pull_request:7050", + "right": "pull_request:7181", + "accept": false, + "reason": "Typos/examples cleanup versus training resume-log behavior; unrelated changes." + }, + { + "left": "pull_request:11516", + "right": "pull_request:11572", + "accept": false, + "reason": "Same model family, but one is an LTX Video release/docs update and the other is LoRA format support; different concrete changes." + }, + { + "left": "pull_request:4115", + "right": "pull_request:5511", + "accept": false, + "reason": "Related area, but not the same bug: one adds discrete timestep support for SDXL ensemble-of-experts, the other fixes a second-order scheduler edge case." + }, + { + "left": "pull_request:10589", + "right": "pull_request:10867", + "accept": false, + "reason": "Both involve IP-Adapter support, but they target different pipelines and capabilities (SD3 img2img vs Flux multi-adapter support)." + }, + { + "left": "pull_request:12269", + "right": "pull_request:13169", + "accept": false, + "reason": "Different Flux features: modular Kontext support versus PEFT LoRA format conversion." + }, + { + "left": "pull_request:7550", + "right": "pull_request:9423", + "accept": false, + "reason": "Both are differential diffusion community pipelines, but for different models/pipelines (SDXL vs Kolors)." + }, + { + "left": "pull_request:6999", + "right": "pull_request:8856", + "accept": false, + "reason": "Unrelated img2img/input handling bugs in different pipelines: Stable Video Diffusion tensor normalization vs ControlNet/SDXL image batch duplication." + }, + { + "left": "pull_request:10941", + "right": "pull_request:11716", + "accept": false, + "reason": "One is a copy/fix pass across Flux pipelines; the other is a LoRA loader metadata fix. Different code paths and outcomes." + }, + { + "left": "pull_request:12736", + "right": "pull_request:12973", + "accept": false, + "reason": "Different models and features: regional compilation for Z-Image versus adding GLM-Image model support." + }, + { + "left": "pull_request:2994", + "right": "pull_request:5117", + "accept": false, + "reason": "ControlNet training type-casting and training resume logging are separate issues." + }, + { + "left": "pull_request:11684", + "right": "pull_request:12275", + "accept": false, + "reason": "Different subsystems: GGUF single-file checkpoint loading vs TorchAO quantization config support." + }, + { + "left": "pull_request:11481", + "right": "pull_request:12315", + "accept": false, + "reason": "A test rename is not the same change as collapsing LoRA tests to a single scheduler." + }, + { + "left": "pull_request:11767", + "right": "pull_request:12782", + "accept": false, + "reason": "Implementation fix for group offloading with disk versus a test tweak for HunyuanVideo offloading." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13330", + "accept": false, + "reason": "Different test issues: CLAP config setup in AudioLDM2 vs TorchAO test adjustments." + }, + { + "left": "pull_request:12681", + "right": "pull_request:13323", + "accept": false, + "reason": "Both are CI/test fixes, but they address different failures and code paths." + }, + { + "left": "pull_request:130", + "right": "pull_request:1579", + "accept": false, + "reason": "Both touch examples docs, but one adds an accelerate setup step and the other adds Intel optimization examples." + }, + { + "left": "pull_request:2009", + "right": "pull_request:473", + "accept": false, + "reason": "Textual inversion loader support and a Stable Diffusion inpaint sample fix are unrelated." + }, + { + "left": "pull_request:2325", + "right": "pull_request:365", + "accept": false, + "reason": "ONNX VAE converter support and VAE type-hint cleanup are different changes with no shared bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 109, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10312", + "pull_request:10359", + "pull_request:10376", + "pull_request:10589", + "pull_request:10911", + "pull_request:11113", + "pull_request:11655", + "pull_request:11786", + "pull_request:11892", + "pull_request:12378", + "pull_request:12551", + "pull_request:12618", + "pull_request:13099", + "pull_request:13296", + "pull_request:1945", + "pull_request:2393", + "pull_request:3268", + "pull_request:3337", + "pull_request:3716", + "pull_request:3766", + "pull_request:4132", + "pull_request:4301", + "pull_request:4397", + "pull_request:4499", + "pull_request:4500", + "pull_request:4501", + "pull_request:4502", + "pull_request:4666", + "pull_request:4761", + "pull_request:4862", + "pull_request:5038", + "pull_request:5506", + "pull_request:5898", + "pull_request:6215", + "pull_request:6231", + "pull_request:6578", + "pull_request:7058", + "pull_request:7166", + "pull_request:7932", + "pull_request:823", + "pull_request:8679", + "pull_request:8992", + "pull_request:9257", + "pull_request:9268", + "pull_request:9423", + "pull_request:9473", + "pull_request:9536", + "pull_request:9881" + ], + "soft_pairs": [ + "pull_request:3766|pull_request:4301", + "pull_request:4397|pull_request:4862", + "pull_request:4499|pull_request:4666", + "pull_request:4500|pull_request:4666", + "pull_request:4501|pull_request:4666", + "pull_request:4502|pull_request:4666", + "pull_request:4761|pull_request:6215", + "pull_request:5898|pull_request:6578", + "pull_request:6231|pull_request:7166", + "pull_request:7932|pull_request:9536", + "pull_request:10376|pull_request:11655", + "pull_request:10359|pull_request:10911", + "pull_request:10589|pull_request:9881", + "pull_request:11786|pull_request:12618", + "pull_request:3268|pull_request:3337", + "pull_request:4132|pull_request:5038", + "pull_request:5506|pull_request:7058", + "pull_request:8679|pull_request:9423", + "pull_request:2393|pull_request:3716", + "pull_request:8992|pull_request:9268", + "pull_request:10312|pull_request:9257", + "pull_request:11113|pull_request:9473", + "pull_request:11892|pull_request:12551", + "pull_request:12378|pull_request:13099", + "pull_request:12551|pull_request:13296", + "pull_request:1945|pull_request:823" + ], + "prepared_review_unit_hash": "0db7304d9aaa52e51265af720978f585a6753e1e66b0fc40c2e0b4d77cfe03ed", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 101399, + "estimated_input_tokens": 25350, + "estimated_eval_tokens": 50956 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose similarity cluster, not a true duplicate set. The PRs span unrelated fixes across docs, CI, tests, pipelines, and model loaders; none of the soft edges look like the same concrete bug/change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out as a duplicate target. If one representative is needed, PR 4502 is the most complete and merged SDXL-related change in the set, but the cluster as a whole is not a true duplicate group.", + "best_issue_reason": null, + "best_pr_reason": "PR 4502 is the strongest standalone PR here: merged, substantial, and with inbound references. It is still not a duplicate anchor for the other PRs in this cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3766", + "right": "pull_request:4301", + "accept": false, + "reason": "Kandinsky UNet conversion script vs Kandinsky docs update; same broad area, but not the same bug or change." + }, + { + "left": "pull_request:4397", + "right": "pull_request:4862", + "accept": false, + "reason": "SDXL docstring wording vs a ControlNet img2img bug fix; unrelated code paths." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4666", + "accept": false, + "reason": "SDXL reference-attn community example/docs vs SDXL ControlNet LoRA support; different features and code paths." + }, + { + "left": "pull_request:4500", + "right": "pull_request:4666", + "accept": false, + "reason": "Same as above: SDXL reference example/docs vs ControlNet LoRA handling are not the same underlying change." + }, + { + "left": "pull_request:4501", + "right": "pull_request:4666", + "accept": false, + "reason": "Same as above: a reference-image SDXL example is unrelated to enabling LoRA in ControlNet." + }, + { + "left": "pull_request:4502", + "right": "pull_request:4666", + "accept": false, + "reason": "Same as above: different SDXL features, different files, different fix scopes." + }, + { + "left": "pull_request:4761", + "right": "pull_request:6215", + "accept": false, + "reason": "Paint-by-example image encoder fix vs PixArt test seeding/expected-values update; different bugs." + }, + { + "left": "pull_request:5898", + "right": "pull_request:6578", + "accept": false, + "reason": "Two separate documentation edits with different content and scope; not the same change." + }, + { + "left": "pull_request:6231", + "right": "pull_request:7166", + "accept": false, + "reason": "SDXL LoRA training fp16 gradient fix vs SDXL DreamBooth LoRA loading fix; distinct training/loading bugs." + }, + { + "left": "pull_request:7932", + "right": "pull_request:9536", + "accept": false, + "reason": "LoRA test expected-value update vs ChatGLM tokenizer padding-side API tweak; unrelated." + }, + { + "left": "pull_request:10376", + "right": "pull_request:11655", + "accept": false, + "reason": "Both touch LoRA loading, but one adds HunyuanVideo original-format support and the other adds bnb 8-bit handling for Flux Control LoRA; different concrete changes." + }, + { + "left": "pull_request:10359", + "right": "pull_request:10911", + "accept": false, + "reason": "Different test fixes in different model suites; no shared underlying bug." + }, + { + "left": "pull_request:10589", + "right": "pull_request:9881", + "accept": false, + "reason": "SD3 Img2Img IP-Adapter support vs IP-Adapter XFormers attention processor support; related subsystem, but not one fix." + }, + { + "left": "pull_request:11786", + "right": "pull_request:12618", + "accept": false, + "reason": "Big GPU marker CI changes vs uv install typo fix; completely separate CI issues." + }, + { + "left": "pull_request:3268", + "right": "pull_request:3337", + "accept": false, + "reason": "Img2Img postprocessing refactor vs broader postprocessing refactor for other pipelines; same theme, but not the same change." + }, + { + "left": "pull_request:4132", + "right": "pull_request:5038", + "accept": false, + "reason": "Optional accelerate handling in checkpoint conversion vs explicit LoRA scale argument passing; unrelated code paths." + }, + { + "left": "pull_request:5506", + "right": "pull_request:7058", + "accept": false, + "reason": "PEFT/xformers LoRA call-signature fix vs copying state dict before load_lora_weights; different defects." + }, + { + "left": "pull_request:8679", + "right": "pull_request:9423", + "accept": false, + "reason": "Two different differential-diffusion community pipelines for different models; not the same underlying bug or mergeable change." + }, + { + "left": "pull_request:2393", + "right": "pull_request:3716", + "accept": false, + "reason": "Both add pipelines, but panorama and ParaDiGMS are different features with different implementations." + }, + { + "left": "pull_request:8992", + "right": "pull_request:9268", + "accept": false, + "reason": "Different differential-diffusion community pipelines for HunyuanDiT vs Flux; not duplicates." + }, + { + "left": "pull_request:10312", + "right": "pull_request:9257", + "accept": false, + "reason": "Attention head-to-batch bug fix vs Flux DreamBooth prompt/tokenization change; unrelated." + }, + { + "left": "pull_request:11113", + "right": "pull_request:9473", + "accept": false, + "reason": "AutoencoderKL train-script fix vs Flux DreamBooth LoRA optimizer-lr fix; separate training-script bugs." + }, + { + "left": "pull_request:11892", + "right": "pull_request:12551", + "accept": false, + "reason": "Removing k-diffusion dependency vs skipping Sana CI tests; unrelated maintenance." + }, + { + "left": "pull_request:12378", + "right": "pull_request:13099", + "accept": false, + "reason": "Custom-code loading logic bug fix vs documentation for custom models; docs only, not the same code change." + }, + { + "left": "pull_request:12551", + "right": "pull_request:13296", + "accept": false, + "reason": "Skipping Sana CI tests vs release-test transformer version update; different CI maintenance tasks." + }, + { + "left": "pull_request:1945", + "right": "pull_request:823", + "accept": false, + "reason": "UNet block tests vs negative-prompt type mismatch fix; unrelated test coverage vs runtime bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 110, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10723", + "pull_request:3462", + "pull_request:3766", + "pull_request:4010", + "pull_request:4034", + "pull_request:4078", + "pull_request:4132", + "pull_request:4201", + "pull_request:4237", + "pull_request:4238", + "pull_request:4262", + "pull_request:4461", + "pull_request:4847", + "pull_request:4856", + "pull_request:4862", + "pull_request:4886", + "pull_request:5003", + "pull_request:5015", + "pull_request:5035", + "pull_request:5085", + "pull_request:5117", + "pull_request:5386", + "pull_request:5448", + "pull_request:5738", + "pull_request:5833", + "pull_request:5834", + "pull_request:5979", + "pull_request:6021", + "pull_request:6119", + "pull_request:6144", + "pull_request:6147", + "pull_request:6259", + "pull_request:6419", + "pull_request:6721", + "pull_request:6997", + "pull_request:7018", + "pull_request:7156", + "pull_request:7325", + "pull_request:7371", + "pull_request:7790", + "pull_request:8616", + "pull_request:8769", + "pull_request:8791", + "pull_request:8825", + "pull_request:9175", + "pull_request:9463", + "pull_request:9659", + "pull_request:9893" + ], + "soft_pairs": [ + "pull_request:3462|pull_request:4262", + "pull_request:3766|pull_request:4461", + "pull_request:4010|pull_request:4237", + "pull_request:4010|pull_request:4238", + "pull_request:4034|pull_request:5738", + "pull_request:4078|pull_request:6147", + "pull_request:4132|pull_request:5833", + "pull_request:4132|pull_request:5834", + "pull_request:4201|pull_request:4847", + "pull_request:4856|pull_request:5015", + "pull_request:4862|pull_request:6147", + "pull_request:4886|pull_request:7156", + "pull_request:5003|pull_request:7018", + "pull_request:5035|pull_request:5085", + "pull_request:5117|pull_request:6119", + "pull_request:5386|pull_request:7325", + "pull_request:5979|pull_request:6419", + "pull_request:6021|pull_request:6997", + "pull_request:6144|pull_request:6259", + "pull_request:6721|pull_request:7790", + "pull_request:8769|pull_request:8791", + "pull_request:8825|pull_request:9893", + "pull_request:7371|pull_request:8616", + "pull_request:10723|pull_request:9175", + "pull_request:5448|pull_request:5979", + "pull_request:9463|pull_request:9659" + ], + "prepared_review_unit_hash": "d1113691dfe998f50a20bfe013ee4780ec645599d64683ba656850b8dab312ff", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 103944, + "estimated_input_tokens": 25986, + "estimated_eval_tokens": 52228 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous cluster of unrelated pull requests across ControlNet/SDXL fixes, LoRA loader/training changes, conversion scripts, docs typos, and new pipeline features. No soft edge looks like the same underlying bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the cluster contains many unrelated changes (docs, loaders, training, conversion, and new pipelines) rather than one underlying fix.", + "best_issue_reason": null, + "best_pr_reason": "No clear best PR for deduping this cluster; the items are not variants of one change, so any choice would be arbitrary.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3462", + "right": "pull_request:4262", + "accept": false, + "reason": "Both are DreamBooth/LoRA training PRs, but one improves orig UNet robustness while the other fixes fp16 gradient/unscale behavior; different bugs." + }, + { + "left": "pull_request:3766", + "right": "pull_request:4461", + "accept": false, + "reason": "Same conversion-script area, but Kandinsky UNet config updates vs. adding a pipeline_class_name argument for SD conversion are unrelated changes." + }, + { + "left": "pull_request:4010", + "right": "pull_request:4237", + "accept": false, + "reason": "Both touch SDXL img2img/offload code, but 4010 fixes sequential CPU offload while 4237 fixes prompt embedding dtype in encode_prompt." + }, + { + "left": "pull_request:4010", + "right": "pull_request:4238", + "accept": false, + "reason": "Same broad SDXL area, but 4010 is about sequential offload in img2img and 4238 is about ControlNet inference/offload sequencing; not the same bug." + }, + { + "left": "pull_request:4034", + "right": "pull_request:5738", + "accept": false, + "reason": "Both modify loaders.py, but 4034 adds LoRA unloading utilities/docs and 5738 adds adapter deletion; distinct features." + }, + { + "left": "pull_request:4078", + "right": "pull_request:6147", + "accept": false, + "reason": "Both involve from_single_file/convert_from_ckpt, but 4078 forces accelerate while 6147 adds SDXL inpainting/refiner single-file support; different code-path problems." + }, + { + "left": "pull_request:4132", + "right": "pull_request:5833", + "accept": false, + "reason": "Both concern single-file loading, but 4132 makes accelerate optional and 5833 fixes ControlNet import paths; unrelated issues." + }, + { + "left": "pull_request:4132", + "right": "pull_request:5834", + "accept": false, + "reason": "Both concern single-file loading, but 4132 changes accelerate handling and 5834 fixes relative imports for ControlNet classes; not duplicates." + }, + { + "left": "pull_request:4201", + "right": "pull_request:4847", + "accept": false, + "reason": "FABRIC community example/docs vs. Prompt Free Diffusion community example; different pipelines and purposes." + }, + { + "left": "pull_request:4856", + "right": "pull_request:5015", + "accept": false, + "reason": "Both are docs-link fixes, but they correct different broken links on different pages." + }, + { + "left": "pull_request:4862", + "right": "pull_request:6147", + "accept": false, + "reason": "4862 fixes MultiControlNetModel img2img shape handling, while 6147 adds SDXL refiner support in single-file loading; different bugs." + }, + { + "left": "pull_request:4886", + "right": "pull_request:7156", + "accept": false, + "reason": "StableDiffusionUpscalePipeline safety-checker fix vs. stalebot label normalization; completely unrelated." + }, + { + "left": "pull_request:5003", + "right": "pull_request:7018", + "accept": false, + "reason": "Both are LoRA-related, but 5003 fixes fuse/unfuse internals while 7018 fixes gradient unscaling in advanced LoRA training; not the same change." + }, + { + "left": "pull_request:5035", + "right": "pull_request:5085", + "accept": false, + "reason": "Both touch loaders.py, but 5035 removes a hard transformers dependency and 5085 prevents offloading breakage for incompatible LoRA checkpoints; separate issues." + }, + { + "left": "pull_request:5117", + "right": "pull_request:6119", + "accept": false, + "reason": "Training-log resume fixes across scripts vs. fp16 gradient unscaling for LoRA training; different underlying problems." + }, + { + "left": "pull_request:5386", + "right": "pull_request:7325", + "accept": false, + "reason": "Both are typo/doc cleanup, but they affect different files and languages; not the same artifact or bug." + }, + { + "left": "pull_request:5979", + "right": "pull_request:6419", + "accept": false, + "reason": "DDGAN pipeline/scheduler addition vs. prompt-diffusion in-context learning example; unrelated feature work." + }, + { + "left": "pull_request:6021", + "right": "pull_request:6997", + "accept": false, + "reason": "DreamBooth LoRA advanced variant support vs. LoRA support for UNetSpatioTemporalConditionModel; different models and changes." + }, + { + "left": "pull_request:6144", + "right": "pull_request:6259", + "accept": false, + "reason": "Both in text-to-image LoRA training, but one fixes saving unwrapped LoRA weights and the other fixes autocast-related runtime errors during validation; distinct bugs." + }, + { + "left": "pull_request:6721", + "right": "pull_request:7790", + "accept": false, + "reason": "AnimateDiff SDXL pipeline vs. SDXL masked im2im community pipeline; different features, not the same underlying change." + }, + { + "left": "pull_request:8769", + "right": "pull_request:8791", + "accept": false, + "reason": "Both modify AutoencoderKL, but one adds attentionless VAE support and the other adds tiling support for SD3; separate changes." + }, + { + "left": "pull_request:8825", + "right": "pull_request:9893", + "accept": false, + "reason": "K-diffusion SDXL ControlNet/IP-Adapter support vs. a community reference pipeline for SDXL ControlNet; different pipeline code paths." + }, + { + "left": "pull_request:7371", + "right": "pull_request:8616", + "accept": false, + "reason": "Both are LoRA feature work, but 7371 adds DoRA support and 8616 adds LoRA fusion for SD3; not the same concrete fix." + }, + { + "left": "pull_request:10723", + "right": "pull_request:9175", + "accept": false, + "reason": "Both mention ControlNet Union, but 10723 adds per-control-type scaling while 9175 adds Flux ControlNet Union support; different model families and functionality." + }, + { + "left": "pull_request:5448", + "right": "pull_request:5979", + "accept": false, + "reason": "LCM pipeline/scheduler vs. DDGAN pipeline/scheduler; both are new pipelines but not the same underlying feature." + }, + { + "left": "pull_request:9463", + "right": "pull_request:9659", + "accept": false, + "reason": "PAG support for SD Img2Img vs. LoRA support for SD3 Img2Img; unrelated pipeline enhancements." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 111, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1012", + "pull_request:10359", + "pull_request:10481", + "pull_request:10547", + "pull_request:10576", + "pull_request:11142", + "pull_request:11159", + "pull_request:11238", + "pull_request:1147", + "pull_request:11532", + "pull_request:11726", + "pull_request:11843", + "pull_request:12040", + "pull_request:12074", + "pull_request:12215", + "pull_request:12225", + "pull_request:12490", + "pull_request:12681", + "pull_request:1282", + "pull_request:1283", + "pull_request:2002", + "pull_request:243", + "pull_request:2489", + "pull_request:3397", + "pull_request:3589", + "pull_request:4070", + "pull_request:4370", + "pull_request:4499", + "pull_request:4500", + "pull_request:4501", + "pull_request:4706", + "pull_request:4749", + "pull_request:4763", + "pull_request:4944", + "pull_request:5750", + "pull_request:5833", + "pull_request:5900", + "pull_request:6225", + "pull_request:7016", + "pull_request:7820", + "pull_request:7924", + "pull_request:8777", + "pull_request:8825", + "pull_request:888", + "pull_request:8981", + "pull_request:9206", + "pull_request:9968" + ], + "soft_pairs": [ + "pull_request:12215|pull_request:12225", + "pull_request:11532|pull_request:12040", + "pull_request:10576|pull_request:11843", + "pull_request:11238|pull_request:9968", + "pull_request:1282|pull_request:2002", + "pull_request:243|pull_request:888", + "pull_request:4370|pull_request:5833", + "pull_request:4499|pull_request:4763", + "pull_request:4500|pull_request:4763", + "pull_request:4501|pull_request:4763", + "pull_request:7924|pull_request:8825", + "pull_request:8777|pull_request:8981", + "pull_request:5900|pull_request:7820", + "pull_request:11726|pull_request:12074", + "pull_request:10359|pull_request:11159", + "pull_request:12490|pull_request:12681", + "pull_request:4944|pull_request:7016", + "pull_request:1012|pull_request:1147", + "pull_request:10481|pull_request:11142", + "pull_request:10547|pull_request:9206", + "pull_request:1283|pull_request:888", + "pull_request:2489|pull_request:3397", + "pull_request:3589|pull_request:5750", + "pull_request:4070|pull_request:4749", + "pull_request:4706|pull_request:6225" + ], + "prepared_review_unit_hash": "f4edf3d6808fb3b71421c596d4ad7a2e56db1775efaef82f0f74827ec0400739", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 102734, + "estimated_input_tokens": 25684, + "estimated_eval_tokens": 51624 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated PRs across model additions, docs, loaders, training examples, and test-only fixes. None of the soft pairs look like the same underlying bug or change, so I reject all soft edges.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:12215", + "right": "pull_request:12225", + "accept": false, + "reason": "Different Qwen-Image features: ControlNet support vs edit inpainting pipeline." + }, + { + "left": "pull_request:11532", + "right": "pull_request:12040", + "accept": false, + "reason": "Different model families and changes: HiDream LoRA support vs Wan lightx2v/2nd-denoiser LoRA support." + }, + { + "left": "pull_request:10576", + "right": "pull_request:11843", + "accept": false, + "reason": "Unrelated test coverage for different subsystems and models." + }, + { + "left": "pull_request:11238", + "right": "pull_request:9968", + "accept": false, + "reason": "One is a specific SD3 ControlNet validation fix; the other is a broad ControlNet refactor." + }, + { + "left": "pull_request:1282", + "right": "pull_request:2002", + "accept": false, + "reason": "Docs link correction versus a new LoRA training feature; not the same change." + }, + { + "left": "pull_request:243", + "right": "pull_request:888", + "accept": false, + "reason": "Different pipelines: RePaint inpainting vs CycleDiffusion." + }, + { + "left": "pull_request:4370", + "right": "pull_request:5833", + "accept": false, + "reason": "Unrelated import/name cleanup vs single-file loading import fix." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4763", + "accept": false, + "reason": "Different SDXL work: reference-attn example vs LoRA conversion utility script." + }, + { + "left": "pull_request:4500", + "right": "pull_request:4763", + "accept": false, + "reason": "Different SDXL work: reference-attn example vs LoRA conversion utility script." + }, + { + "left": "pull_request:4501", + "right": "pull_request:4763", + "accept": false, + "reason": "Different SDXL work: reference-attn example vs LoRA conversion utility script." + }, + { + "left": "pull_request:7924", + "right": "pull_request:8825", + "accept": false, + "reason": "Different code paths: ControlNet Inpaint SDXL IP-Adapter bug fix vs adding ControlNet/IP-Adapter support to SDXL K-Diffusion." + }, + { + "left": "pull_request:8777", + "right": "pull_request:8981", + "accept": false, + "reason": "Dummy attention processor classes vs adding LoraLoaderMixin to package inits." + }, + { + "left": "pull_request:5900", + "right": "pull_request:7820", + "accept": false, + "reason": "Broad IP-Adapter support across many pipelines vs a narrow InstructPix2Pix-specific fix." + }, + { + "left": "pull_request:11726", + "right": "pull_request:12074", + "accept": false, + "reason": "Both touch Wan LoRA conversion, but they solve different concrete problems and are not one mergeable change." + }, + { + "left": "pull_request:10359", + "right": "pull_request:11159", + "accept": false, + "reason": "Different test-fix areas: LoRA layer test cleanup vs quantization dtype/test updates." + }, + { + "left": "pull_request:12490", + "right": "pull_request:12681", + "accept": false, + "reason": "AudioLDM2 ClapConfig test tweak is unrelated to the broader CI CPU test fixes." + }, + { + "left": "pull_request:4944", + "right": "pull_request:7016", + "accept": false, + "reason": "IP-Adapter feature work vs later refactor/embedding API changes; not the same underlying change." + }, + { + "left": "pull_request:1012", + "right": "pull_request:1147", + "accept": false, + "reason": "Different frameworks and bugs: PyTorch DreamBooth text-encoder train mode vs Flax examples loading text encoder from subfolder." + }, + { + "left": "pull_request:10481", + "right": "pull_request:11142", + "accept": false, + "reason": "Different Sana fixes: clean_caption default change vs conversion-script checkpoint/scheduler updates." + }, + { + "left": "pull_request:10547", + "right": "pull_request:9206", + "accept": false, + "reason": "Different test failures and models; only both are test fixes." + }, + { + "left": "pull_request:1283", + "right": "pull_request:888", + "accept": false, + "reason": "Different pipeline additions: Versatile Diffusion vs CycleDiffusion." + }, + { + "left": "pull_request:2489", + "right": "pull_request:3397", + "accept": false, + "reason": "Unrelated docs additions; no shared bug or implementation change." + }, + { + "left": "pull_request:3589", + "right": "pull_request:5750", + "accept": false, + "reason": "Reference-tricks README notes are not the same as adding a reference-only SDXL introduction." + }, + { + "left": "pull_request:4070", + "right": "pull_request:4749", + "accept": false, + "reason": "SDXL docs cleanup and SDXL DreamBooth LoRA checkpointing are separate issues." + }, + { + "left": "pull_request:4706", + "right": "pull_request:6225", + "accept": false, + "reason": "Related LoRA alpha theme, but one is loader internals and the other is example config parsing; not the same concrete fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 112, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10494", + "pull_request:11081", + "pull_request:11159", + "pull_request:11605", + "pull_request:11715", + "pull_request:12021", + "pull_request:12091", + "pull_request:1215", + "pull_request:12277", + "pull_request:12288", + "pull_request:12416", + "pull_request:12551", + "pull_request:12882", + "pull_request:13056", + "pull_request:2976", + "pull_request:3135", + "pull_request:4215", + "pull_request:4223", + "pull_request:4481", + "pull_request:5539", + "pull_request:5608", + "pull_request:5943", + "pull_request:6213", + "pull_request:6217", + "pull_request:6303", + "pull_request:6330", + "pull_request:6448", + "pull_request:6845", + "pull_request:7103", + "pull_request:7113", + "pull_request:7277", + "pull_request:7353", + "pull_request:7411", + "pull_request:742", + "pull_request:7590", + "pull_request:7712", + "pull_request:7825", + "pull_request:7932", + "pull_request:8630", + "pull_request:9040", + "pull_request:9139", + "pull_request:9268", + "pull_request:9379", + "pull_request:9473", + "pull_request:9536", + "pull_request:9646", + "pull_request:9821", + "pull_request:9827" + ], + "soft_pairs": [ + "pull_request:5539|pull_request:7411", + "pull_request:5608|pull_request:7103", + "pull_request:9473|pull_request:9646", + "pull_request:12091|pull_request:12288", + "pull_request:9040|pull_request:9268", + "pull_request:3135|pull_request:4223", + "pull_request:12021|pull_request:12416", + "pull_request:11159|pull_request:11715", + "pull_request:1215|pull_request:742", + "pull_request:12277|pull_request:12551", + "pull_request:12882|pull_request:13056", + "pull_request:2976|pull_request:4215", + "pull_request:4481|pull_request:6303", + "pull_request:5943|pull_request:7113", + "pull_request:6213|pull_request:6845", + "pull_request:6217|pull_request:6330", + "pull_request:6448|pull_request:7932", + "pull_request:7277|pull_request:7712", + "pull_request:7353|pull_request:7590", + "pull_request:8630|pull_request:9139", + "pull_request:11081|pull_request:11605", + "pull_request:7825|pull_request:9379", + "pull_request:9821|pull_request:9827", + "pull_request:10494|pull_request:9536" + ], + "prepared_review_unit_hash": "2ac83cb7cc84e61b90adafb7e8547f547091ed07818f1713528b4c15a65ff663", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 104399, + "estimated_input_tokens": 26100, + "estimated_eval_tokens": 52456 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a noisy topical-similarity group, not a true duplicate set: it mixes unrelated docs cleanups, test adjustments, feature additions, and separate bug fixes across many subsystems. No single PR cleanly represents a shared underlying change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good global representative; the cluster is too heterogeneous for any one PR to serve as the canonical change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5539", + "right": "pull_request:7411", + "accept": false, + "reason": "Both are minor cleanup/docs-oriented PRs, but they touch different files and unrelated changes (DreamBooth typos vs broad typo/docs sweep)." + }, + { + "left": "pull_request:5608", + "right": "pull_request:7103", + "accept": false, + "reason": "Both fix bugs, but one is a PEFT/LoRA empty alpha bug while the other is a ControlNet image validation bug; different code paths and symptoms." + }, + { + "left": "pull_request:9473", + "right": "pull_request:9646", + "accept": false, + "reason": "Both affect Flux Dreambooth scripts, but one fixes optimizer LR handling for T5 and the other adds configurable LoRA target modules; not the same bug/change." + }, + { + "left": "pull_request:12091", + "right": "pull_request:12288", + "accept": false, + "reason": "Both add new model/pipeline exports, but they introduce different products (Dream vs InfiniteTalk) and unrelated implementations." + }, + { + "left": "pull_request:9040", + "right": "pull_request:9268", + "accept": false, + "reason": "Both are differential-diffusion community pipelines, but they target different models and separate implementations; not mergeable as one fix." + }, + { + "left": "pull_request:3135", + "right": "pull_request:4223", + "accept": false, + "reason": "Both touch ControlNet training, but one enforces resolution divisibility and the other fixes SDXL training-script call/import behavior; distinct issues." + }, + { + "left": "pull_request:12021", + "right": "pull_request:12416", + "accept": false, + "reason": "Both are model/pipeline integration PRs, but they add different features for different architectures (LTX attention backends vs QwenImage Edit Plus modular support)." + }, + { + "left": "pull_request:11159", + "right": "pull_request:11715", + "accept": false, + "reason": "Both are quantization test-related, but one fixes dtype handling in single-file loading and the other adds bitsandbytes version guards for compile tests." + }, + { + "left": "pull_request:1215", + "right": "pull_request:742", + "accept": false, + "reason": "Same scheduler file, but the fixes address different DDIM noise-generation bugs (CPU-generator path vs dtype correctness)." + }, + { + "left": "pull_request:12277", + "right": "pull_request:12551", + "accept": false, + "reason": "Both skip tests in CI, but they target different pipelines and different flaky behaviors; not the same underlying defect." + }, + { + "left": "pull_request:12882", + "right": "pull_request:13056", + "accept": false, + "reason": "Both are Z-Image related, but one adds a community differential img2img pipeline and the other adds DreamBooth training docs/script; separate deliverables." + }, + { + "left": "pull_request:2976", + "right": "pull_request:4215", + "accept": false, + "reason": "Both are docs link/path fixes, but they affect unrelated documentation pages and do not share the same bug." + }, + { + "left": "pull_request:4481", + "right": "pull_request:6303", + "accept": false, + "reason": "Both are large new pipeline/algorithm PRs, but EDM implementation and ADD training scripts are entirely different changes." + }, + { + "left": "pull_request:5943", + "right": "pull_request:7113", + "accept": false, + "reason": "Completely unrelated: transformer gradient checkpointing versus Docker Slack notifications." + }, + { + "left": "pull_request:6213", + "right": "pull_request:6845", + "accept": false, + "reason": "Both are refactors/cleanup in video-related code, but they concern different pipelines and different code paths." + }, + { + "left": "pull_request:6217", + "right": "pull_request:6330", + "accept": false, + "reason": "One fixes torch.compile behavior in Stable Video Diffusion, the other is a docs update for SVD rendering; not the same change." + }, + { + "left": "pull_request:6448", + "right": "pull_request:7932", + "accept": false, + "reason": "Both are LoRA test fixes, but they address different regressions in different tests and outputs." + }, + { + "left": "pull_request:7277", + "right": "pull_request:7712", + "accept": false, + "reason": "Both are general cleanup PRs, but they modify unrelated files and issues (typos/docs vs imports)." + }, + { + "left": "pull_request:7353", + "right": "pull_request:7590", + "accept": false, + "reason": "These are unrelated: adding FromSingleFileMixin support to a ControlNet pipeline versus expanding Python version support in setup.py." + }, + { + "left": "pull_request:8630", + "right": "pull_request:9139", + "accept": false, + "reason": "Both touch DreamBooth docs/training, but one adds SD3 text-encoder training and the other is a Flux DreamBooth bugfix/update set." + }, + { + "left": "pull_request:11081", + "right": "pull_request:11605", + "accept": false, + "reason": "Both involve group offloading, but one adds record_stream support while the other makes it compatible with torch.compile; different underlying fixes." + }, + { + "left": "pull_request:7825", + "right": "pull_request:9379", + "accept": false, + "reason": "Both are test adjustments, but one fixes device-map/offload tests and the other fixes Flux GPU test inputs/device handling." + }, + { + "left": "pull_request:9821", + "right": "pull_request:9827", + "accept": false, + "reason": "Related theme, but not duplicates: 9821 changes the general device_map detection utility, while 9827 uses that behavior to fix LoRA loading with device maps." + }, + { + "left": "pull_request:10494", + "right": "pull_request:9536", + "accept": false, + "reason": "One changes tokenizers installation in CI, the other fixes a tokenizer API in Kolors; unrelated code and symptom." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 113, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10730", + "pull_request:11896", + "pull_request:12142", + "pull_request:12143", + "pull_request:1215", + "pull_request:12790", + "pull_request:12914", + "pull_request:13100", + "pull_request:1327", + "pull_request:13410", + "pull_request:1588", + "pull_request:2002", + "pull_request:2479", + "pull_request:2557", + "pull_request:2584", + "pull_request:2861", + "pull_request:2919", + "pull_request:2928", + "pull_request:2994", + "pull_request:3016", + "pull_request:3128", + "pull_request:3446", + "pull_request:3614", + "pull_request:3691", + "pull_request:3912", + "pull_request:4070", + "pull_request:4215", + "pull_request:4428", + "pull_request:4749", + "pull_request:4862", + "pull_request:4886", + "pull_request:5157", + "pull_request:6085", + "pull_request:6147", + "pull_request:6578", + "pull_request:6946", + "pull_request:7018", + "pull_request:7166", + "pull_request:7186", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:816", + "pull_request:819", + "pull_request:8422", + "pull_request:9099", + "pull_request:9175", + "pull_request:9601" + ], + "soft_pairs": [ + "pull_request:12142|pull_request:12914", + "pull_request:12143|pull_request:12914", + "pull_request:1215|pull_request:819", + "pull_request:1588|pull_request:3614", + "pull_request:2584|pull_request:4862", + "pull_request:2861|pull_request:2919", + "pull_request:2994|pull_request:4749", + "pull_request:4215|pull_request:6578", + "pull_request:7797|pull_request:8422", + "pull_request:7803|pull_request:8422", + "pull_request:7804|pull_request:8422", + "pull_request:9099|pull_request:9175", + "pull_request:11896|pull_request:12790", + "pull_request:6946|pull_request:7186", + "pull_request:10730|pull_request:9601", + "pull_request:13100|pull_request:13410", + "pull_request:1327|pull_request:2928", + "pull_request:1327|pull_request:3446", + "pull_request:2002|pull_request:816", + "pull_request:2479|pull_request:3691", + "pull_request:2479|pull_request:3912", + "pull_request:2557|pull_request:3016", + "pull_request:3128|pull_request:4886", + "pull_request:4070|pull_request:4428", + "pull_request:4862|pull_request:6085", + "pull_request:5157|pull_request:7018", + "pull_request:6147|pull_request:7166" + ], + "prepared_review_unit_hash": "5dc76b1dbdb1248cb532d11aabc2541c437cc60fb1f35dfce9f89f1f46a115f7", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 97513, + "estimated_input_tokens": 24379, + "estimated_eval_tokens": 49014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a title-similarity false positive set: the PRs span unrelated fixes across LoRA offloading, WAN dtype handling, DDIM/other schedulers, docs-only updates, and various test tweaks rather than one shared bug/change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:12142", + "right": "pull_request:12914", + "accept": false, + "reason": "Different code paths: WAN VACE mask dtype fix vs. a Cosmos doc typo." + }, + { + "left": "pull_request:12143", + "right": "pull_request:12914", + "accept": false, + "reason": "Different code paths: WAN VACE mask dtype fix vs. a Cosmos doc typo." + }, + { + "left": "pull_request:1215", + "right": "pull_request:819", + "accept": false, + "reason": "Both touch DDIM, but one fixes eta/generator noise handling and the other fixes timestep dtype on Windows; separate bugs." + }, + { + "left": "pull_request:1588", + "right": "pull_request:3614", + "accept": false, + "reason": "One is a test compatibility tweak for FP16, the other is a documentation/test cleanup; not the same bug." + }, + { + "left": "pull_request:2584", + "right": "pull_request:4862", + "accept": false, + "reason": "Different features/bugs: a community ControlNet pipeline addition vs. SDXL MultiControlNet img2img sizing fix." + }, + { + "left": "pull_request:2861", + "right": "pull_request:2919", + "accept": false, + "reason": "Stable UnCLIP image_embeds handling vs. shrinking a 3D UNet test for speed; unrelated." + }, + { + "left": "pull_request:2994", + "right": "pull_request:4749", + "accept": false, + "reason": "ControlNet training type-casting fix vs. SDXL LoRA checkpoint loading fix; different training scripts and failure modes." + }, + { + "left": "pull_request:4215", + "right": "pull_request:6578", + "accept": false, + "reason": "Docs path update for ControlNet vs. heading/formatting fix in ControlNet docs; not the same change." + }, + { + "left": "pull_request:7797", + "right": "pull_request:8422", + "accept": false, + "reason": "A DDPM fast-test size reduction vs. a generic subfolder-loading test suite; unrelated test maintenance." + }, + { + "left": "pull_request:7803", + "right": "pull_request:8422", + "accept": false, + "reason": "A DDIM fast-test size reduction vs. a generic subfolder-loading test suite; unrelated test maintenance." + }, + { + "left": "pull_request:7804", + "right": "pull_request:8422", + "accept": false, + "reason": "Amused fast-test size reduction vs. a generic subfolder-loading test suite; unrelated test maintenance." + }, + { + "left": "pull_request:9099", + "right": "pull_request:9175", + "accept": false, + "reason": "Both concern ControlNet/Flux support, but one adds SD3 controlnet inpainting while the other adds Union ControlNet for Flux; distinct features." + }, + { + "left": "pull_request:11896", + "right": "pull_request:12790", + "accept": false, + "reason": "Different LoRA issues: hook removal/offloading when loading LoRAs vs. LoRA key normalization for ZImage models." + }, + { + "left": "pull_request:6946", + "right": "pull_request:7186", + "accept": false, + "reason": "Related IP-Adapter area, but one adds low_cpu_mem_usage loading and the other moves Face ID to core with model/embedding changes; not one concrete fix." + }, + { + "left": "pull_request:10730", + "right": "pull_request:9601", + "accept": false, + "reason": "Unwrap hooks removal in pipeline utils vs. VAE dtype control in SD3 training; unrelated problems." + }, + { + "left": "pull_request:13100", + "right": "pull_request:13410", + "accept": false, + "reason": "Modular pipeline docs update vs. agent docs/skill restructuring; different documentation targets and intent." + }, + { + "left": "pull_request:1327", + "right": "pull_request:2928", + "accept": false, + "reason": "Custom pipeline local loading convenience vs. ControlNet local dataset loading; different features." + }, + { + "left": "pull_request:1327", + "right": "pull_request:3446", + "accept": false, + "reason": "Custom pipeline local loading vs. a pipeline loading warning-message wording fix; not the same bug." + }, + { + "left": "pull_request:2002", + "right": "pull_request:816", + "accept": false, + "reason": "LoRA training example support vs. deepspeed FP16 loss casting fix; unrelated training concerns." + }, + { + "left": "pull_request:2479", + "right": "pull_request:3691", + "accept": false, + "reason": "LoRA text-encoder training support vs. choosing LoRAAttnProcessor2_0 for PT 2.0; distinct LoRA enhancements." + }, + { + "left": "pull_request:2479", + "right": "pull_request:3912", + "accept": false, + "reason": "Both touch the LoRA text-to-image script, but one adds CLI support while the other adds text-encoder finetuning/docs; could not be merged as one concrete bug fix." + }, + { + "left": "pull_request:2557", + "right": "pull_request:3016", + "accept": false, + "reason": "DreamBooth logging vs. ControlNet Min-SNR/gradient-accumulation docs and flax training updates; unrelated." + }, + { + "left": "pull_request:3128", + "right": "pull_request:4886", + "accept": false, + "reason": "Img2img timestep selection bug vs. upscale safety-checker input processing; different pipelines and defects." + }, + { + "left": "pull_request:4070", + "right": "pull_request:4428", + "accept": false, + "reason": "Minor SDXL doc link/wording fix vs. broad SDXL docs reorganization; not the same change." + }, + { + "left": "pull_request:4862", + "right": "pull_request:6085", + "accept": false, + "reason": "ControlNet img2img MultiControlNet shape handling vs. scheduler add_noise indexing bug; unrelated." + }, + { + "left": "pull_request:5157", + "right": "pull_request:7018", + "accept": false, + "reason": "LoRA conversion name-mapping fix for ControlNet/LyCORIS vs. advanced LoRA gradient unscaling fix; different layers of the stack." + }, + { + "left": "pull_request:6147", + "right": "pull_request:7166", + "accept": false, + "reason": "Single-file SDXL inpainting/refiner loading vs. SDXL LoRA checkpoint loading; related ecosystem, but different concrete fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 114, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10396", + "pull_request:10397", + "pull_request:10911", + "pull_request:11159", + "pull_request:11485", + "pull_request:11522", + "pull_request:11588", + "pull_request:11755", + "pull_request:12225", + "pull_request:12416", + "pull_request:12496", + "pull_request:1250", + "pull_request:12919", + "pull_request:12983", + "pull_request:12996", + "pull_request:13089", + "pull_request:13237", + "pull_request:2034", + "pull_request:243", + "pull_request:3850", + "pull_request:4408", + "pull_request:4481", + "pull_request:4498", + "pull_request:4774", + "pull_request:4795", + "pull_request:4847", + "pull_request:4862", + "pull_request:5915", + "pull_request:5922", + "pull_request:6197", + "pull_request:6864", + "pull_request:6884", + "pull_request:689", + "pull_request:7295", + "pull_request:7476", + "pull_request:8389", + "pull_request:8631", + "pull_request:8688", + "pull_request:8755", + "pull_request:8856", + "pull_request:8859", + "pull_request:8902", + "pull_request:9244", + "pull_request:9268", + "pull_request:9423", + "pull_request:9696", + "pull_request:9807" + ], + "soft_pairs": [ + "pull_request:7295|pull_request:9244", + "pull_request:8631|pull_request:8902", + "pull_request:8755|pull_request:9807", + "pull_request:10396|pull_request:10397", + "pull_request:1250|pull_request:243", + "pull_request:2034|pull_request:3850", + "pull_request:2034|pull_request:689", + "pull_request:4795|pull_request:6864", + "pull_request:8688|pull_request:8755", + "pull_request:12983|pull_request:13237", + "pull_request:7476|pull_request:8856", + "pull_request:7476|pull_request:8859", + "pull_request:11485|pull_request:12225", + "pull_request:10911|pull_request:11755", + "pull_request:12416|pull_request:13089", + "pull_request:4862|pull_request:5922", + "pull_request:9268|pull_request:9423", + "pull_request:4408|pull_request:4774", + "pull_request:8389|pull_request:9696", + "pull_request:4481|pull_request:4847", + "pull_request:4498|pull_request:6197", + "pull_request:5915|pull_request:6884", + "pull_request:11159|pull_request:11522", + "pull_request:11588|pull_request:12496", + "pull_request:12919|pull_request:12996" + ], + "prepared_review_unit_hash": "e353a4b09ca4cdb85e5466b6ac4d451f00173d4c8ae8fa331965ca1babf64df3", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 99090, + "estimated_input_tokens": 24773, + "estimated_eval_tokens": 49802 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is noisy and largely heterogeneous. The soft pairs mostly share broad subsystem terms (LoRA, SDXL, single-file, pipelines, tests) but do not look like the same underlying bug/change, so none should be merged as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear canonical PR: the items span unrelated fixes and features across different pipelines, loaders, tests, and docs.", + "best_issue_reason": null, + "best_pr_reason": "PR 10397 is the strongest standalone representative in the set because it is a merged, concrete Flux LoRA fix with user-facing impact, but it still is not a true duplicate anchor for the whole cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7295", + "right": "pull_request:9244", + "accept": false, + "reason": "Both touch single-file loading, but one is a UNet loader cleanup and the other adds Flux pipeline single-file support; different models and code paths." + }, + { + "left": "pull_request:8631", + "right": "pull_request:8902", + "accept": false, + "reason": "One fixes a return value for T5 checkpoint conversion; the other changes sharded checkpoint sibling detection. Same utility area, different bugs." + }, + { + "left": "pull_request:8755", + "right": "pull_request:9807", + "accept": false, + "reason": "SD3 Dreambooth LoRA text-encoder training and AutoencoderKL LoRA adapter support are unrelated changes in different subsystems." + }, + { + "left": "pull_request:10396", + "right": "pull_request:10397", + "accept": false, + "reason": "Both are Flux LoRA-related, but 10396 fixes state-dict expansion indexing while 10397 changes unload/reset behavior and docs; not the same concrete bug." + }, + { + "left": "pull_request:1250", + "right": "pull_request:243", + "accept": false, + "reason": "Community CLIPSeg+inpainting example and RePaint inpainting pipeline are different features with no shared code-path bug." + }, + { + "left": "pull_request:2034", + "right": "pull_request:3850", + "accept": false, + "reason": "These are separate test fixes for DiT and audio diffusion; same test-fix pattern but different components." + }, + { + "left": "pull_request:2034", + "right": "pull_request:689", + "accept": false, + "reason": "DiT test adjustments are unrelated to the attention performance refactor in the slow-tests PR." + }, + { + "left": "pull_request:4795", + "right": "pull_request:6864", + "accept": false, + "reason": "One fixes ControlNet torch-compile behavior; the other is documentation about torch.compile usage with PEFT." + }, + { + "left": "pull_request:8688", + "right": "pull_request:8755", + "accept": false, + "reason": "DoRA conversion logic and SD3 Dreambooth text-encoder training are different LoRA-related areas and different bugs." + }, + { + "left": "pull_request:12983", + "right": "pull_request:13237", + "accept": false, + "reason": "LTX2 single-file support and FLUX.2 Klein single-file mapping are different model families and separate additions." + }, + { + "left": "pull_request:7476", + "right": "pull_request:8856", + "accept": false, + "reason": "MultiControlNet SDXL img2img compile handling and generic img2img image-batch duplication are different bug fixes." + }, + { + "left": "pull_request:7476", + "right": "pull_request:8859", + "accept": false, + "reason": "These affect different pipelines and failure modes: ControlNet shape handling versus AuraFlow compile compatibility." + }, + { + "left": "pull_request:11485", + "right": "pull_request:12225", + "accept": false, + "reason": "Flex.2 preview pipeline addition and Qwen-Image edit inpainting pipeline are separate feature PRs for different models." + }, + { + "left": "pull_request:10911", + "right": "pull_request:11755", + "accept": false, + "reason": "Lumina2 LoRA fuse test coverage and LTX latent upsample CPU offload behavior are unrelated test changes." + }, + { + "left": "pull_request:12416", + "right": "pull_request:13089", + "accept": false, + "reason": "Both mention QwenImage modular code, but one adds Edit Plus support while the other adds a new guider and broader pipeline updates; not the same change." + }, + { + "left": "pull_request:4862", + "right": "pull_request:5922", + "accept": false, + "reason": "SDXL MultiControlNet img2img shape handling and MultiAdapter inpainting preprocessing are different adapter paths and different bugs." + }, + { + "left": "pull_request:9268", + "right": "pull_request:9423", + "accept": false, + "reason": "Both are differential-diffusion examples, but they target different pipelines (Flux vs Kolors) and are separate community implementations." + }, + { + "left": "pull_request:4408", + "right": "pull_request:4774", + "accept": false, + "reason": "SDXL from_single_file inpainting support and negative size/crop conditioning are different user-facing features." + }, + { + "left": "pull_request:8389", + "right": "pull_request:9696", + "accept": false, + "reason": "Benchmark workflow maintenance and PR-test workflow input validation are unrelated CI changes." + }, + { + "left": "pull_request:4481", + "right": "pull_request:4847", + "accept": false, + "reason": "EDM pipeline/scheduler implementation and Prompt Free Diffusion community example are distinct additions with no shared bug." + }, + { + "left": "pull_request:4498", + "right": "pull_request:6197", + "accept": false, + "reason": "An SDXL reference example and DeepCache support for Stable Diffusion are not the same underlying change." + }, + { + "left": "pull_request:5915", + "right": "pull_request:6884", + "accept": false, + "reason": "IP-Adapter Plus support and scale-length validation in IP-Adapter are related but separate issues, not one duplicate fix." + }, + { + "left": "pull_request:11159", + "right": "pull_request:11522", + "accept": false, + "reason": "Mixed-int8/quantization dtype fixes and AudioLDM2 transformers-main compatibility are unrelated changes." + }, + { + "left": "pull_request:11588", + "right": "pull_request:12496", + "accept": false, + "reason": "WAN I2V LoRA support and WAN I2V condition-shape mismatch are different code paths and different bugs." + }, + { + "left": "pull_request:12919", + "right": "pull_request:12996", + "accept": false, + "reason": "NPU attention enablement and Ulysses Anything Attention are both attention-related, but they solve different parallel/back-end problems and should not be merged as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 115, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10005", + "pull_request:10099", + "pull_request:10102", + "pull_request:10109", + "pull_request:10115", + "pull_request:10179", + "pull_request:10183", + "pull_request:10188", + "pull_request:1020", + "pull_request:10278", + "pull_request:10332", + "pull_request:10364", + "pull_request:10517", + "pull_request:10581", + "pull_request:10648", + "pull_request:10728", + "pull_request:10783", + "pull_request:10977", + "pull_request:11025", + "pull_request:11271", + "pull_request:11439", + "pull_request:11546", + "pull_request:11595", + "pull_request:11804", + "pull_request:11823", + "pull_request:12040", + "pull_request:12091", + "pull_request:1221", + "pull_request:12526", + "pull_request:12737", + "pull_request:12988", + "pull_request:4632", + "pull_request:5057", + "pull_request:5127", + "pull_request:5185", + "pull_request:5417", + "pull_request:6464", + "pull_request:7550", + "pull_request:8675", + "pull_request:8679", + "pull_request:9175", + "pull_request:9268", + "pull_request:9367", + "pull_request:9445", + "pull_request:9507", + "pull_request:9597", + "pull_request:9843" + ], + "soft_pairs": [ + "pull_request:10581|pull_request:10728", + "pull_request:12091|pull_request:12526", + "pull_request:4632|pull_request:6464", + "pull_request:5057|pull_request:5127", + "pull_request:11546|pull_request:11804", + "pull_request:5185|pull_request:5417", + "pull_request:7550|pull_request:9268", + "pull_request:8679|pull_request:9597", + "pull_request:11823|pull_request:12040", + "pull_request:9367|pull_request:9507", + "pull_request:9175|pull_request:9445", + "pull_request:11025|pull_request:11823", + "pull_request:12737|pull_request:12988", + "pull_request:10005|pull_request:10332", + "pull_request:10005|pull_request:8675", + "pull_request:10099|pull_request:10977", + "pull_request:10102|pull_request:9843", + "pull_request:10109|pull_request:10648", + "pull_request:10115|pull_request:10783", + "pull_request:10115|pull_request:11439", + "pull_request:10179|pull_request:10278", + "pull_request:10179|pull_request:10517", + "pull_request:10183|pull_request:10364", + "pull_request:10183|pull_request:11271", + "pull_request:10188|pull_request:11595", + "pull_request:1020|pull_request:1221" + ], + "prepared_review_unit_hash": "4236e6da58ffe7bd4f5f506c5eb1b77c85a96798df175f6c480c5bed5fad413f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 96508, + "estimated_input_tokens": 24127, + "estimated_eval_tokens": 48510 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab-bag of mostly unrelated PRs across docs, tests, LoRA loading, single-file support, and several different model families. The apparent similarities are usually just broad subsystem overlap, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10581", + "right": "pull_request:10728", + "accept": false, + "reason": "Both touch Flux-adjacent code, but one adds IP-Adapter support for SD3 inpaint and the other is a Flux quantization/IP-Adapter tweak; different pipelines and different problems." + }, + { + "left": "pull_request:12091", + "right": "pull_request:12526", + "accept": false, + "reason": "Both are large model-addition PRs, but Dream 7B and Wan2.2 Animate are different models/pipelines with unrelated code paths." + }, + { + "left": "pull_request:4632", + "right": "pull_request:6464", + "accept": false, + "reason": "Both are training-script bug fixes, but they target different Dreambooth LoRA scripts and different checkpoint/save issues." + }, + { + "left": "pull_request:5057", + "right": "pull_request:5127", + "accept": false, + "reason": "SDXL clip-skip support and MultiAdapter support are separate features touching some of the same pipeline files but fixing different behavior." + }, + { + "left": "pull_request:11546", + "right": "pull_request:11804", + "accept": false, + "reason": "Both involve LoRA loading infrastructure, but one is a Hunyuan/FramePack PEFT mapping addition and the other is group-offloading dynamic load/unload support." + }, + { + "left": "pull_request:5185", + "right": "pull_request:5417", + "accept": false, + "reason": "Both relate to LoRA scaling, but 5185 is a PEFT CI fix while 5417 changes scale/unscale behavior across many pipelines; not the same bug." + }, + { + "left": "pull_request:7550", + "right": "pull_request:9268", + "accept": false, + "reason": "Both are differential-diffusion community pipelines, but SDXL img2img and Flux soft inpainting are different implementations and change sets." + }, + { + "left": "pull_request:8679", + "right": "pull_request:9597", + "accept": false, + "reason": "Both are community diffusion examples, but one is SD3 differential img2img and the other is Auraflow differential diffusion; unrelated pipelines." + }, + { + "left": "pull_request:11823", + "right": "pull_request:12040", + "accept": false, + "reason": "Both add LoRA conversion support, but for different targets: Fal Kontext vs Wan/lightx2v. Similar subsystem, different conversion logic and model families." + }, + { + "left": "pull_request:9367", + "right": "pull_request:9507", + "accept": false, + "reason": "Both are Flux-related, but one extends auto-pipeline registrations and the other fixes Flux ControlNet batching/control modes; distinct code paths." + }, + { + "left": "pull_request:9175", + "right": "pull_request:9445", + "accept": false, + "reason": "Both are Flux features, but Union ControlNet support and Flux with CFG are separate pipeline capabilities, not one concrete fix." + }, + { + "left": "pull_request:11025", + "right": "pull_request:11823", + "accept": false, + "reason": "Both touch LoRA conversion utilities, but they support different external formats/models (Wan vs Fal Kontext) and do not fix the same conversion bug." + }, + { + "left": "pull_request:12737", + "right": "pull_request:12988", + "accept": false, + "reason": "Both are context-parallel related, but one fixes Flux.2 context-parallel settings and the other changes generic sharding assertions; not the same issue." + }, + { + "left": "pull_request:10005", + "right": "pull_request:10332", + "accept": false, + "reason": "One adds missing Flux pipeline docs; the other fixes a shape attribute in loading errors. Same repo area only by coincidence." + }, + { + "left": "pull_request:10005", + "right": "pull_request:8675", + "accept": false, + "reason": "Flux pipeline docs versus a general fast-diffusion docs tip update are unrelated documentation changes." + }, + { + "left": "pull_request:10099", + "right": "pull_request:10977", + "accept": false, + "reason": "One is a typo fix in xfail conditions; the other adds save/load optional-component tests across multiple pipelines." + }, + { + "left": "pull_request:10102", + "right": "pull_request:9843", + "accept": false, + "reason": "Both are test additions, but one covers SD3 skip_guidance_layers behavior and the other updates SD3 LoRA test data and expectations." + }, + { + "left": "pull_request:10109", + "right": "pull_request:10648", + "accept": false, + "reason": "Stable Audio torch_xla support and DDPM sigmoid-doc updates are different features with no shared bug." + }, + { + "left": "pull_request:10115", + "right": "pull_request:10783", + "accept": false, + "reason": "FluxPosEmbed bugfix versus Dreambooth fp16 unscaling bug are unrelated code paths." + }, + { + "left": "pull_request:10115", + "right": "pull_request:11439", + "accept": false, + "reason": "FluxPosEmbed fix and Hi-Dream validation cleanup address different problems in different model families." + }, + { + "left": "pull_request:10179", + "right": "pull_request:10278", + "accept": false, + "reason": "A community example doc fix for rf-inversion is unrelated to a TorchAo Flux quantization example fix." + }, + { + "left": "pull_request:10179", + "right": "pull_request:10517", + "accept": false, + "reason": "A doc/example correction and a Sana 4K pipeline bug report are not the same underlying issue." + }, + { + "left": "pull_request:10183", + "right": "pull_request:10364", + "accept": false, + "reason": "AutoencoderDC single-file support and Sana torch_xla support are unrelated features." + }, + { + "left": "pull_request:10183", + "right": "pull_request:11271", + "accept": false, + "reason": "Both touch single-file utilities, but one adds AutoencoderDC support and the other fixes LTX 0.9.5 model inference/config mapping." + }, + { + "left": "pull_request:10188", + "right": "pull_request:11595", + "accept": false, + "reason": "Both concern Flux LoRA loading, but 10188 handles mixed normal/expanding LoRAs while 11595 fixes load_lora_weights base-weight selection; different concrete bugs." + }, + { + "left": "pull_request:1020", + "right": "pull_request:1221", + "accept": false, + "reason": "Both are README/docs updates, but one updates training docs broadly and the other adds a specific weight-reuse section." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 116, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10205", + "pull_request:10220", + "pull_request:10222", + "pull_request:10232", + "pull_request:10246", + "pull_request:10250", + "pull_request:10253", + "pull_request:10255", + "pull_request:10268", + "pull_request:1027", + "pull_request:10271", + "pull_request:10275", + "pull_request:10278", + "pull_request:10299", + "pull_request:10319", + "pull_request:10323", + "pull_request:10332", + "pull_request:10359", + "pull_request:10365", + "pull_request:10394", + "pull_request:10404", + "pull_request:1041", + "pull_request:10419", + "pull_request:10429", + "pull_request:1044", + "pull_request:10494", + "pull_request:10585", + "pull_request:10632", + "pull_request:10825", + "pull_request:10858", + "pull_request:11105", + "pull_request:11292", + "pull_request:11325", + "pull_request:11439", + "pull_request:11545", + "pull_request:11600", + "pull_request:11747", + "pull_request:1727", + "pull_request:1729", + "pull_request:178", + "pull_request:874", + "pull_request:8836", + "pull_request:8838", + "pull_request:9066", + "pull_request:9072", + "pull_request:9655", + "pull_request:9683", + "pull_request:9808" + ], + "soft_pairs": [ + "pull_request:10205|pull_request:11292", + "pull_request:10220|pull_request:9683", + "pull_request:10222|pull_request:11105", + "pull_request:10232|pull_request:10404", + "pull_request:10246|pull_request:11545", + "pull_request:10250|pull_request:8836", + "pull_request:10253|pull_request:8838", + "pull_request:10255|pull_request:10275", + "pull_request:10255|pull_request:10585", + "pull_request:10255|pull_request:11439", + "pull_request:10255|pull_request:11600", + "pull_request:10268|pull_request:11105", + "pull_request:1027|pull_request:874", + "pull_request:10271|pull_request:9808", + "pull_request:10278|pull_request:10419", + "pull_request:10278|pull_request:9655", + "pull_request:10299|pull_request:10825", + "pull_request:10299|pull_request:10858", + "pull_request:10319|pull_request:10419", + "pull_request:10319|pull_request:9072", + "pull_request:10319|pull_request:9655", + "pull_request:10323|pull_request:11325", + "pull_request:10332|pull_request:11747", + "pull_request:10359|pull_request:10494", + "pull_request:10365|pull_request:11105", + "pull_request:10394|pull_request:10632", + "pull_request:10394|pull_request:9066", + "pull_request:1041|pull_request:1727", + "pull_request:1041|pull_request:1729", + "pull_request:10419|pull_request:10858", + "pull_request:10429|pull_request:8836", + "pull_request:1044|pull_request:178" + ], + "prepared_review_unit_hash": "34bb4374bb858fe7b6f59f1596b7862cd1979ec477a77d37d242660bcb43ba37", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 94868, + "estimated_input_tokens": 23717, + "estimated_eval_tokens": 47690 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is a loose collection of mostly unrelated docs/tests/chore PRs that only share broad terminology (e.g. Mochi, LoRA, docs, tests) or superficial filename overlap. I do not see one underlying bug/change tying them together, so the soft edges should all be rejected.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10205", + "right": "pull_request:11292", + "accept": false, + "reason": "Different changes: InstructPix2Pix LoRA PEFT refactor vs SDXL LoRA conversion utility; only broad LoRA theme overlaps." + }, + { + "left": "pull_request:10220", + "right": "pull_request:9683", + "accept": false, + "reason": "Unrelated test renames in different pipelines (Mochi vs CogVideoX image2video)." + }, + { + "left": "pull_request:10222", + "right": "pull_request:11105", + "accept": false, + "reason": "Different problems: XLA step marking in ControlNet pipeline vs gradient checkpointing support for AutoencoderKLWan." + }, + { + "left": "pull_request:10232", + "right": "pull_request:10404", + "accept": false, + "reason": "Both touch docs, but one is a Sana attribution/correction and the other is a broad typo pass across many pipeline docs." + }, + { + "left": "pull_request:10246", + "right": "pull_request:11545", + "accept": false, + "reason": "Different scope: adding missing attention processors docs/code vs adding VAE slicing/tiling methods to Flux img2img." + }, + { + "left": "pull_request:10250", + "right": "pull_request:8836", + "accept": false, + "reason": "Different code paths: MidResTemporalBlock1D downsample bug vs SlicedAttnProcessor loop fix." + }, + { + "left": "pull_request:10253", + "right": "pull_request:8838", + "accept": false, + "reason": "Unrelated chores: contribution note in docs vs adding disable_forward_chunking to SD3 transformer." + }, + { + "left": "pull_request:10255", + "right": "pull_request:10275", + "accept": false, + "reason": "Comment/docstring cleanup in Mochi LoRA loader is unrelated to licensing-header fixes across Mochi and LTX." + }, + { + "left": "pull_request:10255", + "right": "pull_request:10585", + "accept": false, + "reason": "Different Mochi follow-ups: LoRA loader comment fix vs VAE annotation correction." + }, + { + "left": "pull_request:10255", + "right": "pull_request:11439", + "accept": false, + "reason": "Different Mochi-related fixes: loader docstring cleanup vs Hi-Dream LoRA validation bug." + }, + { + "left": "pull_request:10255", + "right": "pull_request:11600", + "accept": false, + "reason": "Different area entirely: Mochi LoRA loader docs vs CI filename display fix." + }, + { + "left": "pull_request:10268", + "right": "pull_request:11105", + "accept": false, + "reason": "Single-file support for Mochi Transformer is unrelated to WAN autoencoder gradient checkpointing." + }, + { + "left": "pull_request:1027", + "right": "pull_request:874", + "accept": false, + "reason": "Both are example/community docs, but one adds a DDPM dataloader option and the other adds a generic community pipeline inference example." + }, + { + "left": "pull_request:10271", + "right": "pull_request:9808", + "accept": false, + "reason": "Both are tests, but one is TorchAO serialization refactoring and the other is a broad VAE test refactor across many models." + }, + { + "left": "pull_request:10278", + "right": "pull_request:10419", + "accept": false, + "reason": "Different docs fixes: TorchAO example correction vs scheduler docstrings/parameters update." + }, + { + "left": "pull_request:10278", + "right": "pull_request:9655", + "accept": false, + "reason": "Different documentation topics: TorchAO example vs xDiT image-path updates." + }, + { + "left": "pull_request:10299", + "right": "pull_request:10825", + "accept": false, + "reason": "Different docs topics: GGUF example cleanup vs training text2image checkpoint-loading guidance." + }, + { + "left": "pull_request:10299", + "right": "pull_request:10858", + "accept": false, + "reason": "GGUF example change is not the same issue as the pipelines overview category rename for I2VGen-XL." + }, + { + "left": "pull_request:10319", + "right": "pull_request:10419", + "accept": false, + "reason": "Docstring typo in HunyuanVideo pipeline is unrelated to scheduler parameter documentation." + }, + { + "left": "pull_request:10319", + "right": "pull_request:9072", + "accept": false, + "reason": "One is a HunyuanVideo docstring correction; the other is a grammar fix in ControlNet docs." + }, + { + "left": "pull_request:10319", + "right": "pull_request:9655", + "accept": false, + "reason": "HunyuanVideo docstring fix and xDiT doc image-path updates are unrelated." + }, + { + "left": "pull_request:10323", + "right": "pull_request:11325", + "accept": false, + "reason": "Quantization link cleanup is unrelated to HiDream docstring replacement/doc examples." + }, + { + "left": "pull_request:10332", + "right": "pull_request:11747", + "accept": false, + "reason": "Shape error message fix in model loading is unrelated to missing HiDream license headers." + }, + { + "left": "pull_request:10359", + "right": "pull_request:10494", + "accept": false, + "reason": "Broad test fixes in multiple files vs CI tokenizers install change; not the same bug." + }, + { + "left": "pull_request:10365", + "right": "pull_request:11105", + "accept": false, + "reason": "Both mention XLA/torch_xla in different pipeline code paths, but they fix different pipelines and behaviors." + }, + { + "left": "pull_request:10394", + "right": "pull_request:10632", + "accept": false, + "reason": "Docs link fix in PEFT inference tutorial vs image-path correction in para attention docs." + }, + { + "left": "pull_request:10394", + "right": "pull_request:9066", + "accept": false, + "reason": "PEFT doc link correction is unrelated to adding the StableCascadeUNet docs page." + }, + { + "left": "pull_request:1041", + "right": "pull_request:1727", + "accept": false, + "reason": "Very different issues: older Stable Diffusion test fixes vs ONNX img2img preprocessing bug." + }, + { + "left": "pull_request:1041", + "right": "pull_request:1729", + "accept": false, + "reason": "Different test areas and failure modes: stable diffusion test updates vs latent diffusion CPU/GPU test adjustments." + }, + { + "left": "pull_request:10419", + "right": "pull_request:10858", + "accept": false, + "reason": "Scheduler docstring parameter fixes are unrelated to the pipelines overview table category rename." + }, + { + "left": "pull_request:10429", + "right": "pull_request:8836", + "accept": false, + "reason": "Ascend NPU rotary embedding dtype fix is unrelated to sliced attention processor loop handling." + }, + { + "left": "pull_request:1044", + "right": "pull_request:178", + "accept": false, + "reason": "Music Spectrogram diffusion pipeline addition is unrelated to adding stable diffusion pipeline tests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 117, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10458", + "pull_request:10464", + "pull_request:10491", + "pull_request:10504", + "pull_request:10505", + "pull_request:10531", + "pull_request:10534", + "pull_request:10544", + "pull_request:10585", + "pull_request:1063", + "pull_request:10630", + "pull_request:10632", + "pull_request:10633", + "pull_request:10648", + "pull_request:10684", + "pull_request:10783", + "pull_request:10802", + "pull_request:10819", + "pull_request:10845", + "pull_request:10858", + "pull_request:10885", + "pull_request:10888", + "pull_request:10938", + "pull_request:10947", + "pull_request:10975", + "pull_request:10990", + "pull_request:10991", + "pull_request:11248", + "pull_request:11303", + "pull_request:11355", + "pull_request:11439", + "pull_request:11550", + "pull_request:11693", + "pull_request:11731", + "pull_request:11737", + "pull_request:11743", + "pull_request:11755", + "pull_request:11768", + "pull_request:11778", + "pull_request:12193", + "pull_request:1437", + "pull_request:9553", + "pull_request:9598", + "pull_request:9655", + "pull_request:9683", + "pull_request:9765", + "pull_request:9987" + ], + "soft_pairs": [ + "pull_request:10458|pull_request:10888", + "pull_request:10464|pull_request:10534", + "pull_request:10491|pull_request:9683", + "pull_request:10504|pull_request:11355", + "pull_request:10504|pull_request:9598", + "pull_request:10505|pull_request:11355", + "pull_request:10505|pull_request:9598", + "pull_request:10531|pull_request:10648", + "pull_request:10531|pull_request:10885", + "pull_request:10544|pull_request:11743", + "pull_request:10585|pull_request:11778", + "pull_request:1063|pull_request:1437", + "pull_request:10630|pull_request:11755", + "pull_request:10632|pull_request:9765", + "pull_request:10633|pull_request:9987", + "pull_request:10648|pull_request:11693", + "pull_request:10648|pull_request:11731", + "pull_request:10684|pull_request:10802", + "pull_request:10783|pull_request:11439", + "pull_request:10819|pull_request:10885", + "pull_request:10819|pull_request:11737", + "pull_request:10819|pull_request:11768", + "pull_request:10845|pull_request:10990", + "pull_request:10858|pull_request:9655", + "pull_request:10885|pull_request:11693", + "pull_request:10938|pull_request:11248", + "pull_request:10938|pull_request:11303", + "pull_request:10938|pull_request:12193", + "pull_request:10938|pull_request:9553", + "pull_request:10947|pull_request:11550", + "pull_request:10975|pull_request:10991" + ], + "prepared_review_unit_hash": "b24b07ba9d376e3dafb9dafd9e19d56c2709476bca2ad1366ae2d537acd0258c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 94104, + "estimated_input_tokens": 23526, + "estimated_eval_tokens": 47308 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: it mixes unrelated doc-only edits, test cleanups, pipeline bug fixes, and model/loading feature work. None of the soft pairs look like the same underlying bug or change, so there is no true duplicate set here.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10458", + "right": "pull_request:10888", + "accept": false, + "reason": "One only adds test markers for SD3 LoRA integration tests; the other adds SD3 ControlNet to AutoPipeline. Different code paths and purposes." + }, + { + "left": "pull_request:10464", + "right": "pull_request:10534", + "accept": false, + "reason": "Both mention NPU, but one is FLUX NPU adaptation across several files and the other is an RMSNorm-specific normalization change. Not the same bug/change." + }, + { + "left": "pull_request:10491", + "right": "pull_request:9683", + "accept": false, + "reason": "CI path fix for a misnamed test file vs. CogVideoX I2V test renaming/unskipping. Different targets and unrelated changes." + }, + { + "left": "pull_request:10504", + "right": "pull_request:11355", + "accept": false, + "reason": "Both touch tests, but one removes debug prints and the other adjusts bitsandbytes test behavior/naming. Not the same underlying issue." + }, + { + "left": "pull_request:10504", + "right": "pull_request:9598", + "accept": false, + "reason": "Test-print cleanup is unrelated to adding version notes in Flux LoRA integration tests." + }, + { + "left": "pull_request:10505", + "right": "pull_request:11355", + "accept": false, + "reason": "General test print removal vs. bitsandbytes test consistency updates; different files and fixes." + }, + { + "left": "pull_request:10505", + "right": "pull_request:9598", + "accept": false, + "reason": "Removing prints from tests is unrelated to documenting Flux LoRA test version requirements." + }, + { + "left": "pull_request:10531", + "right": "pull_request:10648", + "accept": false, + "reason": "Both are docs changes, but one documents negative prompts for Flux/SD3 pipelines while the other documents DDPM sigmoid scheduling. Different functionality." + }, + { + "left": "pull_request:10531", + "right": "pull_request:10885", + "accept": false, + "reason": "Flux negative-prompt docs and CogVideoX scheduler docs are separate documentation updates for different APIs." + }, + { + "left": "pull_request:10544", + "right": "pull_request:11743", + "accept": false, + "reason": "ParaAttention documentation is unrelated to adding a shortened Flux training script with quantization." + }, + { + "left": "pull_request:10585", + "right": "pull_request:11778", + "accept": false, + "reason": "Mochi VAE annotation fix and Wan VACE mask-padding fix affect different pipelines and different bugs." + }, + { + "left": "pull_request:1063", + "right": "pull_request:1437", + "accept": false, + "reason": "Both are typo/variable-name fixes, but in different attention/pipeline code paths and unrelated issues." + }, + { + "left": "pull_request:10630", + "right": "pull_request:11755", + "accept": false, + "reason": "A failing Flax scheduler slice update is unrelated to fixing CPU offload behavior for LTX latent upsample." + }, + { + "left": "pull_request:10632", + "right": "pull_request:9765", + "accept": false, + "reason": "Docs image-path correction for ParaAttention vs. SD3 pipeline docstring typo fix; unrelated documentation items." + }, + { + "left": "pull_request:10633", + "right": "pull_request:9987", + "accept": false, + "reason": "Flux IP-Adapter docs example and SD3.5 IP-Adapter pipeline integration are related only thematically, not the same concrete change." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11693", + "accept": false, + "reason": "Both are docs, but DDPM sigmoid scheduler docs and bitsandbytes compile docs cover different features and code paths." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11731", + "accept": false, + "reason": "Scheduler docs vs. memory/offload compile tip; same broad optimization/docs area, but not the same change." + }, + { + "left": "pull_request:10684", + "right": "pull_request:10802", + "accept": false, + "reason": "Hunyuan-Video input validation typo and a controlnet training comment fix are unrelated typo fixes in different files." + }, + { + "left": "pull_request:10783", + "right": "pull_request:11439", + "accept": false, + "reason": "Different training scripts and different bugs: fp16 unscaling fix vs. Hi-Dream validation cleanup." + }, + { + "left": "pull_request:10819", + "right": "pull_request:10885", + "accept": false, + "reason": "LoRA docs entries and CogVideoX scheduler docs are separate documentation additions." + }, + { + "left": "pull_request:10819", + "right": "pull_request:11737", + "accept": false, + "reason": "Docs-only LoRA additions vs. DeepSpeed support in Hi-Dream LoRA training are different changes." + }, + { + "left": "pull_request:10819", + "right": "pull_request:11768", + "accept": false, + "reason": "LoRA docs updates do not match the hook-removal logic fix in loaders." + }, + { + "left": "pull_request:10845", + "right": "pull_request:10990", + "accept": false, + "reason": "Flux LoRA training bugfix and Flux quantized LoRA support are different feature areas and code paths." + }, + { + "left": "pull_request:10858", + "right": "pull_request:9655", + "accept": false, + "reason": "API overview table wording fix is unrelated to xDiT image-path documentation repair." + }, + { + "left": "pull_request:10885", + "right": "pull_request:11693", + "accept": false, + "reason": "CogVideoX scheduler docs and bitsandbytes compile docs are unrelated docs changes." + }, + { + "left": "pull_request:10938", + "right": "pull_request:11248", + "accept": false, + "reason": "Evaluation docs update and a README typo fix are unrelated documentation edits." + }, + { + "left": "pull_request:10938", + "right": "pull_request:11303", + "accept": false, + "reason": "Evaluation guidance and Allegro autoencoder docs are separate documentation topics." + }, + { + "left": "pull_request:10938", + "right": "pull_request:12193", + "accept": false, + "reason": "Evaluation docs and Qwen README path fix do not target the same issue." + }, + { + "left": "pull_request:10938", + "right": "pull_request:9553", + "accept": false, + "reason": "Evaluation conceptual docs are unrelated to distributed inference device-map documentation." + }, + { + "left": "pull_request:10947", + "right": "pull_request:11550", + "accept": false, + "reason": "Both are single-file/loading related, but one adds Sana transformer support and the other adds HiDream GGUF/single-file support. Different models and implementations." + }, + { + "left": "pull_request:10975", + "right": "pull_request:10991", + "accept": false, + "reason": "Wan pipeline bug fixes and Wan single-file support are distinct changes with different goals and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 118, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10977", + "pull_request:10978", + "pull_request:10991", + "pull_request:1103", + "pull_request:11037", + "pull_request:1104", + "pull_request:11053", + "pull_request:11113", + "pull_request:11204", + "pull_request:11212", + "pull_request:11248", + "pull_request:113", + "pull_request:11303", + "pull_request:11325", + "pull_request:11434", + "pull_request:11439", + "pull_request:1147", + "pull_request:11473", + "pull_request:11478", + "pull_request:11485", + "pull_request:11529", + "pull_request:11537", + "pull_request:11558", + "pull_request:11600", + "pull_request:11602", + "pull_request:11623", + "pull_request:11634", + "pull_request:11697", + "pull_request:11740", + "pull_request:11789", + "pull_request:11953", + "pull_request:11954", + "pull_request:11983", + "pull_request:12070", + "pull_request:12193", + "pull_request:12259", + "pull_request:12343", + "pull_request:12487", + "pull_request:12552", + "pull_request:1311", + "pull_request:169", + "pull_request:2479", + "pull_request:400", + "pull_request:60", + "pull_request:9627", + "pull_request:9646", + "pull_request:9683", + "pull_request:9754" + ], + "soft_pairs": [ + "pull_request:10977|pull_request:11434", + "pull_request:10977|pull_request:9683", + "pull_request:10978|pull_request:9627", + "pull_request:10991|pull_request:11053", + "pull_request:1103|pull_request:60", + "pull_request:11037|pull_request:11740", + "pull_request:11037|pull_request:11953", + "pull_request:1104|pull_request:1311", + "pull_request:11113|pull_request:9646", + "pull_request:11204|pull_request:9754", + "pull_request:11212|pull_request:11634", + "pull_request:11248|pull_request:11303", + "pull_request:11248|pull_request:12193", + "pull_request:113|pull_request:169", + "pull_request:113|pull_request:400", + "pull_request:11303|pull_request:12193", + "pull_request:11325|pull_request:12259", + "pull_request:11434|pull_request:11789", + "pull_request:11434|pull_request:12070", + "pull_request:11439|pull_request:11600", + "pull_request:11439|pull_request:12343", + "pull_request:1147|pull_request:2479", + "pull_request:11473|pull_request:11623", + "pull_request:11473|pull_request:11634", + "pull_request:11473|pull_request:12487", + "pull_request:11473|pull_request:12552", + "pull_request:11478|pull_request:11953", + "pull_request:11485|pull_request:11602", + "pull_request:11529|pull_request:11983", + "pull_request:11537|pull_request:11697", + "pull_request:11558|pull_request:11954" + ], + "prepared_review_unit_hash": "8b3a847c847f1a783a833dd966763921298e82d1d680d061c300652db9bc236c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 95721, + "estimated_input_tokens": 23931, + "estimated_eval_tokens": 48118 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous set of mostly unrelated PRs. The soft links are superficial overlaps in docs, tests, or broad subsystem names, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "10991 is the strongest representative PR: it adds Wan single-file support across loaders, models, tests, and docs, giving it the clearest self-contained scope among the set.", + "best_issue_reason": null, + "best_pr_reason": "10991 is the best standalone representative if one PR must anchor the cluster, because it is feature-complete and internally coherent; the others are mostly tiny typo/test/CI/doc fixes or separate features.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10977", + "right": "pull_request:11434", + "accept": false, + "reason": "Different test fixes in unrelated files: save/load optional-components coverage vs an import adjustment in common test utilities." + }, + { + "left": "pull_request:10977", + "right": "pull_request:9683", + "accept": false, + "reason": "Both are test-related, but they target different pipelines and different behaviors; not the same bug or change." + }, + { + "left": "pull_request:10978", + "right": "pull_request:9627", + "accept": false, + "reason": "LoRA conversion logic for OneTrainer Flux is unrelated to an SDXL test timing/warning fix." + }, + { + "left": "pull_request:10991", + "right": "pull_request:11053", + "accept": false, + "reason": "Same Wan docs file, but one adds single-file loading while the other adds a video-to-video pipeline; distinct features and code paths." + }, + { + "left": "pull_request:1103", + "right": "pull_request:60", + "accept": false, + "reason": "A README URL/license fix is unrelated to a scheduler initialization rewrite." + }, + { + "left": "pull_request:11037", + "right": "pull_request:11740", + "accept": false, + "reason": "CI dependency plumbing for bnb tests is unrelated to adding GGUF compile/offload tests." + }, + { + "left": "pull_request:11037", + "right": "pull_request:11953", + "accept": false, + "reason": "Workflow dependency changes do not match a HiDream test-slice update." + }, + { + "left": "pull_request:1104", + "right": "pull_request:1311", + "accept": false, + "reason": "Slow MPS CI setup is unrelated to AltDiffusion test additions." + }, + { + "left": "pull_request:11113", + "right": "pull_request:9646", + "accept": false, + "reason": "AutoencoderKL training-script bug fix and Flux Dreambooth LoRA target-module configurability are separate changes." + }, + { + "left": "pull_request:11204", + "right": "pull_request:9754", + "accept": false, + "reason": "PyTorch/XLA SDXL training support is unrelated to Flux quantized training example work." + }, + { + "left": "pull_request:11212", + "right": "pull_request:11634", + "accept": false, + "reason": "Both are docs-only, but one is an MPS badge update and the other is a torchao link update; not the same issue." + }, + { + "left": "pull_request:11248", + "right": "pull_request:11303", + "accept": false, + "reason": "A Flux README typo fix and an autoencoder docs symbol correction are unrelated documentation edits." + }, + { + "left": "pull_request:11248", + "right": "pull_request:12193", + "accept": false, + "reason": "Different README corrections for different examples; no shared underlying bug." + }, + { + "left": "pull_request:113", + "right": "pull_request:169", + "accept": false, + "reason": "Score SDE VE scheduler initialization and PNDM JAXification are different scheduler changes." + }, + { + "left": "pull_request:113", + "right": "pull_request:400", + "accept": false, + "reason": "The doc PR describes Score SDE VE, but the code fix is a scheduler init change; not the same concrete change." + }, + { + "left": "pull_request:11303", + "right": "pull_request:12193", + "accept": false, + "reason": "Autoencoder docs class-name fix and Qwen README filename fix are unrelated docs updates." + }, + { + "left": "pull_request:11325", + "right": "pull_request:12259", + "accept": false, + "reason": "HiDream docstring cleanup and distributed-inference VAE scale-factor correction are separate docs fixes." + }, + { + "left": "pull_request:11434", + "right": "pull_request:11789", + "accept": false, + "reason": "Import cleanup in test utilities is unrelated to HunyuanVideo device-test behavior." + }, + { + "left": "pull_request:11434", + "right": "pull_request:12070", + "accept": false, + "reason": "Common test import fix does not match Qwen inference slice updates." + }, + { + "left": "pull_request:11439", + "right": "pull_request:11600", + "accept": false, + "reason": "HiDream LoRA validation logic and CI failure-file naming are unrelated." + }, + { + "left": "pull_request:11439", + "right": "pull_request:12343", + "accept": false, + "reason": "HiDream validation cleanup is unrelated to VAE slicing decode behavior." + }, + { + "left": "pull_request:1147", + "right": "pull_request:2479", + "accept": false, + "reason": "Flax text-encoder loading-from-subfolder and LoRA support for the clip text encoder are different training-script changes." + }, + { + "left": "pull_request:11473", + "right": "pull_request:11623", + "accept": false, + "reason": "torchao documentation typo fix and a Flux pipeline typo fix are separate doc/code edits." + }, + { + "left": "pull_request:11473", + "right": "pull_request:11634", + "accept": false, + "reason": "Same torchao doc file, but different content fixes: quantization shorthand typo vs external link update." + }, + { + "left": "pull_request:11473", + "right": "pull_request:12487", + "accept": false, + "reason": "torchao docs typo fix is not the same as a broad broken-links sweep." + }, + { + "left": "pull_request:11473", + "right": "pull_request:12552", + "accept": false, + "reason": "torchao quantization-doc wording and Kandinsky5 typo fixes are unrelated documentation changes." + }, + { + "left": "pull_request:11478", + "right": "pull_request:11953", + "accept": false, + "reason": "HiDream transformer tests and HiDream image fast-test slices are both for the model family, but they are different test additions, not one duplicate change." + }, + { + "left": "pull_request:11485", + "right": "pull_request:11602", + "accept": false, + "reason": "A new Flex.2 pipeline and a Sana Sprint img2img pipeline are entirely different features." + }, + { + "left": "pull_request:11529", + "right": "pull_request:11983", + "accept": false, + "reason": "Unspecified docs edit in inpaint.md is unrelated to an installation command fix." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11697", + "accept": false, + "reason": "UNet torch.compile test coverage is unrelated to TorchAO/bitsandbytes compile-offloading tests." + }, + { + "left": "pull_request:11558", + "right": "pull_request:11954", + "accept": false, + "reason": "Layerwise casting plus group offloading tests are unrelated to Hunyuan Video test-slice updates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 119, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1157", + "pull_request:11600", + "pull_request:11613", + "pull_request:1162", + "pull_request:11634", + "pull_request:1167", + "pull_request:1169", + "pull_request:1171", + "pull_request:11715", + "pull_request:1173", + "pull_request:11847", + "pull_request:1186", + "pull_request:1187", + "pull_request:11953", + "pull_request:11979", + "pull_request:11983", + "pull_request:120", + "pull_request:12018", + "pull_request:12070", + "pull_request:1211", + "pull_request:12129", + "pull_request:1218", + "pull_request:12193", + "pull_request:12197", + "pull_request:1221", + "pull_request:12228", + "pull_request:12247", + "pull_request:12259", + "pull_request:12360", + "pull_request:12675", + "pull_request:12775", + "pull_request:12827", + "pull_request:12883", + "pull_request:12890", + "pull_request:12902", + "pull_request:12980", + "pull_request:1360", + "pull_request:1416", + "pull_request:154", + "pull_request:16", + "pull_request:1601", + "pull_request:1960", + "pull_request:2764", + "pull_request:3132", + "pull_request:41", + "pull_request:848", + "pull_request:925" + ], + "soft_pairs": [ + "pull_request:1157|pull_request:848", + "pull_request:11600|pull_request:12228", + "pull_request:11613|pull_request:11979", + "pull_request:1162|pull_request:1360", + "pull_request:1162|pull_request:925", + "pull_request:11634|pull_request:12018", + "pull_request:1167|pull_request:120", + "pull_request:1169|pull_request:1601", + "pull_request:1171|pull_request:3132", + "pull_request:11715|pull_request:11979", + "pull_request:1173|pull_request:1416", + "pull_request:11847|pull_request:12259", + "pull_request:1186|pull_request:848", + "pull_request:1187|pull_request:848", + "pull_request:11953|pull_request:12070", + "pull_request:11953|pull_request:12775", + "pull_request:11983|pull_request:12827", + "pull_request:120|pull_request:16", + "pull_request:120|pull_request:1960", + "pull_request:12018|pull_request:12902", + "pull_request:1211|pull_request:41", + "pull_request:12129|pull_request:12228", + "pull_request:12129|pull_request:12675", + "pull_request:1218|pull_request:154", + "pull_request:12193|pull_request:12890", + "pull_request:12197|pull_request:12775", + "pull_request:1221|pull_request:2764", + "pull_request:12228|pull_request:12675", + "pull_request:12247|pull_request:12980", + "pull_request:12259|pull_request:12360", + "pull_request:12259|pull_request:12883" + ], + "prepared_review_unit_hash": "97eb0f5d3bfb2ad467e4de6575110c908c32441db15a8173a67717eed789452b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 93678, + "estimated_input_tokens": 23420, + "estimated_eval_tokens": 47096 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly superficial similarity rather than true duplication: the paired PRs span unrelated docs, test tweaks, and bugfixes across different code paths. I would reject all soft edges.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1157", + "right": "pull_request:848", + "accept": false, + "reason": "Both are Flax-related, but one fixes training arg handling/report_to/push_to_hub while the other fixes latent shape from width/height; different bugs." + }, + { + "left": "pull_request:11600", + "right": "pull_request:12228", + "accept": false, + "reason": "CI filename display fix vs LoRA docs typo; no shared code path or underlying bug." + }, + { + "left": "pull_request:11613", + "right": "pull_request:11979", + "accept": false, + "reason": "Both touch torch.compile tests, but 11613 reverts CI/test changes while 11979 only adds version guards; different fixes." + }, + { + "left": "pull_request:1162", + "right": "pull_request:1360", + "accept": false, + "reason": "README/community link cleanup vs adding a K-Diffusion community pipeline; unrelated changes." + }, + { + "left": "pull_request:1162", + "right": "pull_request:925", + "accept": false, + "reason": "Community links/docs cleanup and LPW token padding logic are unrelated." + }, + { + "left": "pull_request:11634", + "right": "pull_request:12018", + "accept": false, + "reason": "Both are quantization docs edits, but they fix different link/text wording issues in separate pages." + }, + { + "left": "pull_request:1167", + "right": "pull_request:120", + "accept": false, + "reason": "README note about Intel XPU vs main README examples update; not the same underlying change." + }, + { + "left": "pull_request:1169", + "right": "pull_request:1601", + "accept": false, + "reason": "Different pipeline bugs: MPS scheduler dtype handling vs classifier-free guidance branching." + }, + { + "left": "pull_request:1171", + "right": "pull_request:3132", + "accept": false, + "reason": "Typo cleanup in community pipeline docs vs a score_sde scheduler type annotation fix; unrelated." + }, + { + "left": "pull_request:11715", + "right": "pull_request:11979", + "accept": false, + "reason": "Bitsandbytes test version annotations are unrelated to torch.compile test version gating." + }, + { + "left": "pull_request:1173", + "right": "pull_request:1416", + "accept": false, + "reason": "Both are ONNX-related, but one targets scheduler compatibility/safety checker and the other fixes conversion/inference feature_extractor handling; distinct concrete bugs." + }, + { + "left": "pull_request:11847", + "right": "pull_request:12259", + "accept": false, + "reason": "Wan single-file docs example vs distributed inference VAE scale-factor docs; unrelated." + }, + { + "left": "pull_request:1186", + "right": "pull_request:848", + "accept": false, + "reason": "Dreambooth Flax report_to/push_to_hub setup vs Flax pipeline width/height bug; different code paths." + }, + { + "left": "pull_request:1187", + "right": "pull_request:848", + "accept": false, + "reason": "Dreambooth Flax cleanup vs latent shape handling in a different Flax pipeline; not the same bug." + }, + { + "left": "pull_request:11953", + "right": "pull_request:12070", + "accept": false, + "reason": "Both update test slices, but for different models (HiDream vs Qwen) and different expected outputs." + }, + { + "left": "pull_request:11953", + "right": "pull_request:12775", + "accept": false, + "reason": "HiDream test slices vs Z-Image docs addition; unrelated." + }, + { + "left": "pull_request:11983", + "right": "pull_request:12827", + "accept": false, + "reason": "Both touch docs syntax, but one is uv install wording and the other is shell block formatting; no shared bug." + }, + { + "left": "pull_request:120", + "right": "pull_request:16", + "accept": false, + "reason": "Both edit README.md, but one adds examples while the other fixes wording; not the same change." + }, + { + "left": "pull_request:120", + "right": "pull_request:1960", + "accept": false, + "reason": "Main README examples update vs ColossalAI README install instructions; unrelated." + }, + { + "left": "pull_request:12018", + "right": "pull_request:12902", + "accept": false, + "reason": "Quantization docs wording vs distributed inference device_map usage; different documentation fixes." + }, + { + "left": "pull_request:1211", + "right": "pull_request:41", + "accept": false, + "reason": "Flax usage comments vs resnet comment cleanup; unrelated." + }, + { + "left": "pull_request:12129", + "right": "pull_request:12228", + "accept": false, + "reason": "NPU attention typo fix vs LoRA docs typo; different subsystems and purposes." + }, + { + "left": "pull_request:12129", + "right": "pull_request:12675", + "accept": false, + "reason": "NPU attention kernel typo fix vs Sana video docs typo; unrelated." + }, + { + "left": "pull_request:1218", + "right": "pull_request:154", + "accept": false, + "reason": "Both concern conversion scripts, but 154 adds the original SD converter and 1218 extends it with new scheduler options; not the same bug." + }, + { + "left": "pull_request:12193", + "right": "pull_request:12890", + "accept": false, + "reason": "Qwen Dreambooth filename/docs tweak vs Wan docs cleanup; unrelated." + }, + { + "left": "pull_request:12197", + "right": "pull_request:12775", + "accept": false, + "reason": "Flux Kontext inpainting docs vs Z-Image docs; unrelated features." + }, + { + "left": "pull_request:1221", + "right": "pull_request:2764", + "accept": false, + "reason": "Weight reuse README docs vs torch 2.0 benchmark docs; no shared change." + }, + { + "left": "pull_request:12228", + "right": "pull_request:12675", + "accept": false, + "reason": "Two independent typo/doc fixes in different pages; not the same underlying issue." + }, + { + "left": "pull_request:12247", + "right": "pull_request:12980", + "accept": false, + "reason": "AutoGuidance validation logic bug vs Mellon param/docstring refactor; unrelated." + }, + { + "left": "pull_request:12259", + "right": "pull_request:12360", + "accept": false, + "reason": "Distributed inference VAE scale-factor docs vs Chroma guidance docs rewrite; different documentation topics." + }, + { + "left": "pull_request:12259", + "right": "pull_request:12883", + "accept": false, + "reason": "Distributed inference docs fix vs torchao quantization typo fix; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 120, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1227", + "pull_request:1229", + "pull_request:1256", + "pull_request:12675", + "pull_request:12681", + "pull_request:1269", + "pull_request:12731", + "pull_request:12775", + "pull_request:1283", + "pull_request:12856", + "pull_request:12883", + "pull_request:1291", + "pull_request:12914", + "pull_request:12971", + "pull_request:12984", + "pull_request:13029", + "pull_request:13040", + "pull_request:13099", + "pull_request:131", + "pull_request:1311", + "pull_request:13120", + "pull_request:13134", + "pull_request:13158", + "pull_request:13181", + "pull_request:1319", + "pull_request:13192", + "pull_request:13253", + "pull_request:13283", + "pull_request:13293", + "pull_request:13344", + "pull_request:13400", + "pull_request:13410", + "pull_request:13499", + "pull_request:13567", + "pull_request:1373", + "pull_request:1487", + "pull_request:1605", + "pull_request:178", + "pull_request:2002", + "pull_request:21", + "pull_request:2209", + "pull_request:2433", + "pull_request:316", + "pull_request:3400", + "pull_request:473", + "pull_request:580", + "pull_request:72", + "pull_request:824" + ], + "soft_pairs": [ + "pull_request:1227|pull_request:1487", + "pull_request:1227|pull_request:2433", + "pull_request:1229|pull_request:2002", + "pull_request:1256|pull_request:824", + "pull_request:12675|pull_request:13040", + "pull_request:12675|pull_request:13181", + "pull_request:12681|pull_request:13293", + "pull_request:1269|pull_request:473", + "pull_request:12731|pull_request:13283", + "pull_request:12775|pull_request:13344", + "pull_request:1283|pull_request:1373", + "pull_request:1283|pull_request:21", + "pull_request:12856|pull_request:13253", + "pull_request:12883|pull_request:13499", + "pull_request:1291|pull_request:2209", + "pull_request:12914|pull_request:13134", + "pull_request:12971|pull_request:13283", + "pull_request:12984|pull_request:13099", + "pull_request:13029|pull_request:13410", + "pull_request:131|pull_request:316", + "pull_request:131|pull_request:72", + "pull_request:1311|pull_request:178", + "pull_request:1311|pull_request:3400", + "pull_request:13120|pull_request:13192", + "pull_request:13158|pull_request:13567", + "pull_request:1319|pull_request:580", + "pull_request:13283|pull_request:13400", + "pull_request:1373|pull_request:1605" + ], + "prepared_review_unit_hash": "b2c8a8ea1a0279dfa2c9e7ea985d11bd63955bee9c5d237f50c0d9df883d0502", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 97877, + "estimated_input_tokens": 24470, + "estimated_eval_tokens": 49196 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly unrelated PRs spanning docs typos, test updates, and separate feature additions. There is no clear duplicate issue/PR at the cluster level; the only loosely related items are in the Versatile Diffusion / modular docs areas, but they still address different changes.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1227", + "right": "pull_request:1487", + "accept": false, + "reason": "Both are docs PRs, but one adds multi-modal examples while the other expands Stable Diffusion xformers docs; different scope and change intent." + }, + { + "left": "pull_request:1227", + "right": "pull_request:2433", + "accept": false, + "reason": "Different docs topics: multi-modal examples vs an SDEdit note for img2img." + }, + { + "left": "pull_request:1229", + "right": "pull_request:2002", + "accept": false, + "reason": "One is a small DreamBooth README install note, the other is a broad LoRA fine-tuning support PR touching many files." + }, + { + "left": "pull_request:1256", + "right": "pull_request:824", + "accept": false, + "reason": "One adds a negative_prompt test in PyTorch pipelines; the other is a Flax scheduler/pipeline test change. Not the same bug or change." + }, + { + "left": "pull_request:12675", + "right": "pull_request:13040", + "accept": false, + "reason": "Both are typo/wording fixes, but in unrelated docs/code paths: Sana video docs vs fp16_safetensors CLI warnings." + }, + { + "left": "pull_request:12675", + "right": "pull_request:13181", + "accept": false, + "reason": "Sana video docs typo versus torchrun argument order docs; unrelated edits." + }, + { + "left": "pull_request:12681", + "right": "pull_request:13293", + "accept": false, + "reason": "Both relate to tests/CI, but they fix different failing pipelines and different code paths." + }, + { + "left": "pull_request:1269", + "right": "pull_request:473", + "accept": false, + "reason": "Shared Stable Diffusion area, but 1269 is a refactor across multiple pipelines while 473 is a specific inpainting sample fix." + }, + { + "left": "pull_request:12731", + "right": "pull_request:13283", + "accept": false, + "reason": "Both are docs, but one is Bria Fibo guidance and the other updates FlashAttention-3 support wording; no shared underlying change." + }, + { + "left": "pull_request:12775", + "right": "pull_request:13344", + "accept": false, + "reason": "Z-Image docs versus GLM Image transformer tests; unrelated artifacts and purposes." + }, + { + "left": "pull_request:1283", + "right": "pull_request:1373", + "accept": false, + "reason": "Both are Versatile Diffusion-related, but 1283 adds the base model integration while 1373 extends it with additional pipelines; not the same concrete change." + }, + { + "left": "pull_request:1283", + "right": "pull_request:21", + "accept": false, + "reason": "Versatile Diffusion model integration versus an early Stable Diffusion test PR; unrelated." + }, + { + "left": "pull_request:12856", + "right": "pull_request:13253", + "accept": false, + "reason": "One adds the z-image-omni-base implementation, the other refactors Z Image transformer tests. Related subsystem, but different underlying change." + }, + { + "left": "pull_request:12883", + "right": "pull_request:13499", + "accept": false, + "reason": "Both mention quantization-related docs, but one fixes a torchao typo and the other adds general torchao mentions in speed/memory docs." + }, + { + "left": "pull_request:1291", + "right": "pull_request:2209", + "accept": false, + "reason": "Stable Diffusion notebook link fix versus an img2img argument-name fix; different examples and bugs." + }, + { + "left": "pull_request:12914", + "right": "pull_request:13134", + "accept": false, + "reason": "Cosmos pipeline doc typo versus a transformer typing annotation fix; unrelated changes." + }, + { + "left": "pull_request:12971", + "right": "pull_request:13283", + "accept": false, + "reason": "Distributed inference docs reordering versus attention backend support wording; different documentation issues." + }, + { + "left": "pull_request:12984", + "right": "pull_request:13099", + "accept": false, + "reason": "Flux2 Klein docs addition versus AutoModel custom-code guide; separate docs/features." + }, + { + "left": "pull_request:13029", + "right": "pull_request:13410", + "accept": false, + "reason": "The modular quickstart guide and the AI agent modular-doc restructuring are different documentation efforts, not the same change." + }, + { + "left": "pull_request:131", + "right": "pull_request:316", + "accept": false, + "reason": "Both touch README.md, but one adds syntax highlighting and the other adds installation/contribution content; not duplicates." + }, + { + "left": "pull_request:131", + "right": "pull_request:72", + "accept": false, + "reason": "README formatting change versus 1D up/downsample test coverage; unrelated." + }, + { + "left": "pull_request:1311", + "right": "pull_request:178", + "accept": false, + "reason": "AltDiffusion test/dependency work versus an early Stable Diffusion pipeline test; different models and goals." + }, + { + "left": "pull_request:1311", + "right": "pull_request:3400", + "accept": false, + "reason": "Shared dependency/setup files, but 1311 adds AltDiffusion tests while 3400 adds omegaconf for tests; not the same bug/change." + }, + { + "left": "pull_request:13120", + "right": "pull_request:13192", + "accept": false, + "reason": "Both add modular pipeline tests, but one checks load-id tagging and the other checks saved index consistency; different behaviors." + }, + { + "left": "pull_request:13158", + "right": "pull_request:13567", + "accept": false, + "reason": "Qwen-Image docs versus AI agent docs; no overlap in underlying change." + }, + { + "left": "pull_request:1319", + "right": "pull_request:580", + "accept": false, + "reason": "Both are Flax-related, but one fixes scheduler subfolder loading and the other is an initial Flax scheduler test/change set." + }, + { + "left": "pull_request:13283", + "right": "pull_request:13400", + "accept": false, + "reason": "Attention backend support note versus profiling guide acknowledgements; unrelated docs changes." + }, + { + "left": "pull_request:1373", + "right": "pull_request:1605", + "accept": false, + "reason": "Versatile Diffusion expansion versus an upcast_attention parameter addition in one helper file; different concrete changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 121, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1415", + "pull_request:1437", + "pull_request:1476", + "pull_request:1487", + "pull_request:1510", + "pull_request:154", + "pull_request:1561", + "pull_request:1564", + "pull_request:1576", + "pull_request:1586", + "pull_request:1587", + "pull_request:1591", + "pull_request:1601", + "pull_request:1625", + "pull_request:1640", + "pull_request:1668", + "pull_request:168", + "pull_request:1687", + "pull_request:1727", + "pull_request:1729", + "pull_request:1756", + "pull_request:1763", + "pull_request:1769", + "pull_request:1778", + "pull_request:178", + "pull_request:1849", + "pull_request:1967", + "pull_request:2077", + "pull_request:2209", + "pull_request:2257", + "pull_request:2278", + "pull_request:2428", + "pull_request:2557", + "pull_request:2629", + "pull_request:299", + "pull_request:309", + "pull_request:3118", + "pull_request:3388", + "pull_request:3449", + "pull_request:3472", + "pull_request:356", + "pull_request:375", + "pull_request:62", + "pull_request:75", + "pull_request:832", + "pull_request:904", + "pull_request:942", + "pull_request:957" + ], + "soft_pairs": [ + "pull_request:1415|pull_request:1849", + "pull_request:1415|pull_request:356", + "pull_request:1437|pull_request:3118", + "pull_request:1476|pull_request:3472", + "pull_request:1487|pull_request:3388", + "pull_request:1510|pull_request:154", + "pull_request:154|pull_request:299", + "pull_request:154|pull_request:904", + "pull_request:1561|pull_request:2257", + "pull_request:1564|pull_request:2557", + "pull_request:1576|pull_request:2077", + "pull_request:1576|pull_request:375", + "pull_request:1586|pull_request:2629", + "pull_request:1587|pull_request:2209", + "pull_request:1591|pull_request:75", + "pull_request:1601|pull_request:832", + "pull_request:1601|pull_request:957", + "pull_request:1625|pull_request:178", + "pull_request:1640|pull_request:942", + "pull_request:1668|pull_request:3472", + "pull_request:168|pull_request:2278", + "pull_request:168|pull_request:309", + "pull_request:1687|pull_request:3472", + "pull_request:1727|pull_request:3449", + "pull_request:1729|pull_request:62", + "pull_request:1756|pull_request:1763", + "pull_request:1763|pull_request:1849", + "pull_request:1769|pull_request:1967", + "pull_request:1778|pull_request:2428" + ], + "prepared_review_unit_hash": "a67ed98e280d4371e35305140b28aee17cf25d68edd56fc6953fc579e933fe6f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 97066, + "estimated_input_tokens": 24267, + "estimated_eval_tokens": 48790 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated diffusers PRs; the soft links are driven by broad naming/subsystem similarity rather than the same concrete fix. I would not merge any of the suggested soft edges.", + "confidence": 0.31, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue artifacts are present. If a PR must represent the cluster, #1415 is the broadest merged stable-diffusion code fix and the most central example among the candidates.", + "best_issue_reason": null, + "best_pr_reason": "#1415 is the best representative PR because it is a merged, multi-file stable-diffusion fix touching several pipelines/tests, whereas the others are narrower one-off docs/tests/script changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1415", + "right": "pull_request:1849", + "accept": false, + "reason": "Both touch attention-related code, but #1415 is a broad stable-diffusion cleanup and #1849 is a separate Versatile Diffusion cross_attention_kwargs fix. Different bug/change." + }, + { + "left": "pull_request:1415", + "right": "pull_request:356", + "accept": false, + "reason": "#1415 fixes stable-diffusion pipeline behavior; #356 adds training example docs/script. Not the same underlying change." + }, + { + "left": "pull_request:1437", + "right": "pull_request:3118", + "accept": false, + "reason": "A typo fix in one pipeline file is unrelated to a ControlNet API-default fix. Different scope and code path." + }, + { + "left": "pull_request:1476", + "right": "pull_request:3472", + "accept": false, + "reason": "Training dependency docs and DreamBooth IF docs are separate documentation updates, not the same change." + }, + { + "left": "pull_request:1487", + "right": "pull_request:3388", + "accept": false, + "reason": "xformers docs and install-docs changes are unrelated documentation fixes." + }, + { + "left": "pull_request:1510", + "right": "pull_request:154", + "accept": false, + "reason": "They both mention the conversion script, but #154 introduces it and #1510 later extends the whole stable-diffusion stack. Not the same concrete change." + }, + { + "left": "pull_request:154", + "right": "pull_request:299", + "accept": false, + "reason": "The original conversion script addition and a PNDM scheduler hotfix are unrelated." + }, + { + "left": "pull_request:154", + "right": "pull_request:904", + "accept": false, + "reason": "Conversion-script creation vs inpainting pipeline rewrite are different changes." + }, + { + "left": "pull_request:1561", + "right": "pull_request:2257", + "accept": false, + "reason": "Both are community pipeline PRs, but they implement different pipelines with different behavior." + }, + { + "left": "pull_request:1564", + "right": "pull_request:2557", + "accept": false, + "reason": "Docs-only default accelerate config vs DreamBooth logging/training-script changes are unrelated." + }, + { + "left": "pull_request:1576", + "right": "pull_request:2077", + "accept": false, + "reason": "ORT integration for training examples and textual-inversion validation support are separate feature changes." + }, + { + "left": "pull_request:1576", + "right": "pull_request:375", + "accept": false, + "reason": "ORT integration and Colab-link docs are not the same underlying change." + }, + { + "left": "pull_request:1586", + "right": "pull_request:2629", + "accept": false, + "reason": "Both address non-square images, but for different pipelines and different fixes. Too broad to be one PR." + }, + { + "left": "pull_request:1587", + "right": "pull_request:2209", + "accept": false, + "reason": "Default image width logic in Paint by Example is unrelated to an img2img docs argument-name correction." + }, + { + "left": "pull_request:1591", + "right": "pull_request:75", + "accept": false, + "reason": "Shared file does not make them duplicates: one fixes sliced-attention upcasting, the other changes Glide attention initialization logic." + }, + { + "left": "pull_request:1601", + "right": "pull_request:832", + "accept": false, + "reason": "Classifier-free-guidance bug in Stable/Alt Diffusion is unrelated to a Flax NSFW masking fix." + }, + { + "left": "pull_request:1601", + "right": "pull_request:957", + "accept": false, + "reason": "A diffusion-pipeline control-flow bug and a scheduler-argument cleanup in training code are unrelated." + }, + { + "left": "pull_request:1625", + "right": "pull_request:178", + "accept": false, + "reason": "k-diffusion height/width defaults and a stable-diffusion test update do not fix the same problem." + }, + { + "left": "pull_request:1640", + "right": "pull_request:942", + "accept": false, + "reason": "Auto-enabling xformers in training scripts is unrelated to the MPS float64 scheduler issue." + }, + { + "left": "pull_request:1668", + "right": "pull_request:3472", + "accept": false, + "reason": "DreamBooth checkpointing and DreamBooth IF docs are different artifacts and different changes." + }, + { + "left": "pull_request:168", + "right": "pull_request:2278", + "accept": false, + "reason": "The original Stable Diffusion pipeline intro and a later video-stable-diffusion project are not the same change." + }, + { + "left": "pull_request:168", + "right": "pull_request:309", + "accept": false, + "reason": "Base pipeline introduction vs a README link fix are unrelated." + }, + { + "left": "pull_request:1687", + "right": "pull_request:3472", + "accept": false, + "reason": "Checkpointing across training scripts and IF DreamBooth docs address different needs." + }, + { + "left": "pull_request:1727", + "right": "pull_request:3449", + "accept": false, + "reason": "ONNX img2img preprocessing fix is unrelated to a workflow/test rework PR." + }, + { + "left": "pull_request:1729", + "right": "pull_request:62", + "accept": false, + "reason": "A test-only LDM CPU/GPU adjustment and an LDM unconditional pipeline fix are not the same bug." + }, + { + "left": "pull_request:1756", + "right": "pull_request:1763", + "accept": false, + "reason": "Both mention attention masks, but one adds mask handling for UCLIP while the other removes an unused mask in Versatile Diffusion. Opposite and different code paths." + }, + { + "left": "pull_request:1763", + "right": "pull_request:1849", + "accept": false, + "reason": "Unused attention-mask cleanup and cross_attention_kwargs plumbing are distinct Versatile Diffusion fixes." + }, + { + "left": "pull_request:1769", + "right": "pull_request:1967", + "accept": false, + "reason": "UnCLIP CPU-offload test changes and a WIP CPU-offload example are unrelated." + }, + { + "left": "pull_request:1778", + "right": "pull_request:2428", + "accept": false, + "reason": "Removing hardcoded names in training scripts is unrelated to removing author names in docs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 122, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:178", + "pull_request:1788", + "pull_request:191", + "pull_request:192", + "pull_request:1920", + "pull_request:1932", + "pull_request:1951", + "pull_request:1960", + "pull_request:1995", + "pull_request:1996", + "pull_request:2006", + "pull_request:2009", + "pull_request:2022", + "pull_request:2040", + "pull_request:2085", + "pull_request:2101", + "pull_request:2118", + "pull_request:2138", + "pull_request:2151", + "pull_request:2181", + "pull_request:2393", + "pull_request:2404", + "pull_request:2433", + "pull_request:2489", + "pull_request:260", + "pull_request:2733", + "pull_request:2788", + "pull_request:2839", + "pull_request:296", + "pull_request:3063", + "pull_request:3099", + "pull_request:3215", + "pull_request:3439", + "pull_request:353", + "pull_request:3800", + "pull_request:3863", + "pull_request:388", + "pull_request:3941", + "pull_request:407", + "pull_request:421", + "pull_request:533", + "pull_request:580", + "pull_request:798", + "pull_request:869", + "pull_request:879", + "pull_request:888", + "pull_request:912", + "pull_request:972" + ], + "soft_pairs": [ + "pull_request:178|pull_request:296", + "pull_request:178|pull_request:388", + "pull_request:1788|pull_request:533", + "pull_request:191|pull_request:421", + "pull_request:192|pull_request:888", + "pull_request:1920|pull_request:2489", + "pull_request:1932|pull_request:3215", + "pull_request:1951|pull_request:2839", + "pull_request:1960|pull_request:879", + "pull_request:1995|pull_request:3063", + "pull_request:1995|pull_request:3099", + "pull_request:1996|pull_request:2733", + "pull_request:2006|pull_request:879", + "pull_request:2009|pull_request:3439", + "pull_request:2022|pull_request:912", + "pull_request:2040|pull_request:2393", + "pull_request:2085|pull_request:2433", + "pull_request:2101|pull_request:3863", + "pull_request:2118|pull_request:3941", + "pull_request:2118|pull_request:869", + "pull_request:2138|pull_request:2404", + "pull_request:2138|pull_request:3800", + "pull_request:2151|pull_request:260", + "pull_request:2151|pull_request:353", + "pull_request:2151|pull_request:407", + "pull_request:2151|pull_request:798", + "pull_request:2151|pull_request:972", + "pull_request:2181|pull_request:2788", + "pull_request:2181|pull_request:580" + ], + "prepared_review_unit_hash": "327b28f10d42a6d891b6d23688904e006e96952d4bdbc87376b75f9b0cc35cbf", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 93077, + "estimated_input_tokens": 23270, + "estimated_eval_tokens": 46796 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose set of unrelated docs, typo, example, and feature PRs; there are no true duplicates here. The most substantial code change is the textual inversion loader PR, but it is not a duplicate of the other items.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 2009 is the strongest representative because it is the largest cross-cutting code change in the set, with broad loader/pipeline impact and substantial review, unlike the many small docs/typo fixes.", + "best_issue_reason": null, + "best_pr_reason": "PR 2009 is the best overall representative of the cluster: it is the most substantive, code-centric change and the closest thing to a central artifact among mostly unrelated small PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:178", + "right": "pull_request:296", + "accept": false, + "reason": "Both mention Stable Diffusion tests, but one adds a test while the other fixes auth tokens in CI; different problems and code paths." + }, + { + "left": "pull_request:178", + "right": "pull_request:388", + "accept": false, + "reason": "Unrelated artifacts: a Stable Diffusion test update vs. an Unconditional Latent Diffusion docs/pipeline item." + }, + { + "left": "pull_request:1788", + "right": "pull_request:533", + "accept": false, + "reason": "Same file, but different changes: a real DDIM scheduler bug fix vs. a revert of type hints. Not the same underlying issue." + }, + { + "left": "pull_request:191", + "right": "pull_request:421", + "accept": false, + "reason": "Import fix for scipy/dummy objects vs. scheduler documentation cleanup; unrelated." + }, + { + "left": "pull_request:192", + "right": "pull_request:888", + "accept": false, + "reason": "Stable Diffusion README parameter tweak vs. adding CycleDiffusion pipeline; different scope and problem." + }, + { + "left": "pull_request:1920", + "right": "pull_request:2489", + "accept": false, + "reason": "Both are docs additions, but they cover different topics and files with no shared bug or change." + }, + { + "left": "pull_request:1932", + "right": "pull_request:3215", + "accept": false, + "reason": "DDIM loading behavior fix vs. 8-bit loading checks in pipeline_utils; unrelated code paths." + }, + { + "left": "pull_request:1951", + "right": "pull_request:2839", + "accept": false, + "reason": "ColossalAI example update vs. examples README update; related only at a high level, not the same change." + }, + { + "left": "pull_request:1960", + "right": "pull_request:879", + "accept": false, + "reason": "Different README/community docs edits with no shared bug or concrete fix." + }, + { + "left": "pull_request:1995", + "right": "pull_request:3063", + "accept": false, + "reason": "Different pipeline issues: strength range validation typo vs. __setattr__ None-handling bug." + }, + { + "left": "pull_request:1995", + "right": "pull_request:3099", + "accept": false, + "reason": "Separate stable diffusion pipeline changes; one is a validation-message fix, the other changes a default attention resolution." + }, + { + "left": "pull_request:1996", + "right": "pull_request:2733", + "accept": false, + "reason": "Both touch conversion scripts, but one adds safetensors output and the other adds fp16 support for a different conversion direction." + }, + { + "left": "pull_request:2006", + "right": "pull_request:879", + "accept": false, + "reason": "Community merger logic fix vs. README table formatting; unrelated." + }, + { + "left": "pull_request:2009", + "right": "pull_request:3439", + "accept": false, + "reason": "Both involve textual inversion, but one adds loading embeddings to pipelines and the other enables state-dict input in the loader; different features." + }, + { + "left": "pull_request:2022", + "right": "pull_request:912", + "accept": false, + "reason": "Dependency/version gating for k-diffusion vs. a project version bump; not the same issue." + }, + { + "left": "pull_request:2040", + "right": "pull_request:2393", + "accept": false, + "reason": "Both add major pipelines, but they are different pipelines with different implementations and docs." + }, + { + "left": "pull_request:2085", + "right": "pull_request:2433", + "accept": false, + "reason": "Separate docs tweaks for different training/pipeline topics." + }, + { + "left": "pull_request:2101", + "right": "pull_request:3863", + "accept": false, + "reason": "Pix2Pix example update vs. Habana optimization docs; unrelated." + }, + { + "left": "pull_request:2118", + "right": "pull_request:3941", + "accept": false, + "reason": "Import spelling fix vs. docs/code address fix; unrelated." + }, + { + "left": "pull_request:2118", + "right": "pull_request:869", + "accept": false, + "reason": "Both touch imports at a broad level, but they fix different modules and problems." + }, + { + "left": "pull_request:2138", + "right": "pull_request:2404", + "accept": false, + "reason": "Different typo fixes in unrelated docs/code paths." + }, + { + "left": "pull_request:2138", + "right": "pull_request:3800", + "accept": false, + "reason": "Docs typo vs. panorama code typo; not the same artifact." + }, + { + "left": "pull_request:2151", + "right": "pull_request:260", + "accept": false, + "reason": "EMAModel load_state_dict bug fix vs. README typo; unrelated." + }, + { + "left": "pull_request:2151", + "right": "pull_request:353", + "accept": false, + "reason": "Both are small fixes, but they affect different code paths and error conditions." + }, + { + "left": "pull_request:2151", + "right": "pull_request:407", + "accept": false, + "reason": "Different fixes in different modules: EMAModel state loading vs. attention slicing behavior." + }, + { + "left": "pull_request:2151", + "right": "pull_request:798", + "accept": false, + "reason": "Different typos in unrelated modules; no shared underlying bug." + }, + { + "left": "pull_request:2151", + "right": "pull_request:972", + "accept": false, + "reason": "EMAModel bug fix vs. README parameter typo; unrelated." + }, + { + "left": "pull_request:2181", + "right": "pull_request:2788", + "accept": false, + "reason": "Docs-only changes in different sections: Flax scheduler usage vs. training TOC deduplication." + }, + { + "left": "pull_request:2181", + "right": "pull_request:580", + "accept": false, + "reason": "Docs on Flax scheduler usage vs. an experimental Flax scheduler test PR; not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 123, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2197", + "pull_request:2209", + "pull_request:2257", + "pull_request:2278", + "pull_request:2287", + "pull_request:2290", + "pull_request:2320", + "pull_request:2333", + "pull_request:2362", + "pull_request:2367", + "pull_request:2383", + "pull_request:2395", + "pull_request:2396", + "pull_request:2398", + "pull_request:2400", + "pull_request:2433", + "pull_request:2446", + "pull_request:2487", + "pull_request:2498", + "pull_request:2503", + "pull_request:2530", + "pull_request:2533", + "pull_request:2557", + "pull_request:2629", + "pull_request:2762", + "pull_request:2766", + "pull_request:3079", + "pull_request:309", + "pull_request:3168", + "pull_request:3388", + "pull_request:3473", + "pull_request:3549", + "pull_request:3680", + "pull_request:3839", + "pull_request:3852", + "pull_request:386", + "pull_request:3894", + "pull_request:3971", + "pull_request:4317", + "pull_request:4546", + "pull_request:4766", + "pull_request:642", + "pull_request:796", + "pull_request:869", + "pull_request:874", + "pull_request:889", + "pull_request:934", + "pull_request:948" + ], + "soft_pairs": [ + "pull_request:2197|pull_request:3839", + "pull_request:2209|pull_request:4317", + "pull_request:2257|pull_request:874", + "pull_request:2278|pull_request:309", + "pull_request:2278|pull_request:386", + "pull_request:2287|pull_request:2629", + "pull_request:2290|pull_request:642", + "pull_request:2320|pull_request:4546", + "pull_request:2320|pull_request:948", + "pull_request:2333|pull_request:3680", + "pull_request:2362|pull_request:2766", + "pull_request:2362|pull_request:796", + "pull_request:2367|pull_request:2557", + "pull_request:2383|pull_request:889", + "pull_request:2395|pull_request:2762", + "pull_request:2396|pull_request:2530", + "pull_request:2396|pull_request:934", + "pull_request:2398|pull_request:3473", + "pull_request:2398|pull_request:3852", + "pull_request:2400|pull_request:869", + "pull_request:2433|pull_request:3388", + "pull_request:2446|pull_request:3971", + "pull_request:2487|pull_request:3549", + "pull_request:2487|pull_request:3894", + "pull_request:2487|pull_request:4766", + "pull_request:2498|pull_request:3079", + "pull_request:2503|pull_request:2766", + "pull_request:2533|pull_request:3168" + ], + "prepared_review_unit_hash": "ef59239664c2faacf500a4df566cbbbffe7088e4aedffdef578941467a9f218f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 94326, + "estimated_input_tokens": 23582, + "estimated_eval_tokens": 47420 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These PRs are only superficially similar (mostly typos, docs, or test tweaks) and do not form a true duplicate cluster around one underlying bug/change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the cluster mixes unrelated documentation, test, and code changes with no single underlying fix or feature.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a strong representative because the items are heterogeneous and not mergeable into one coherent change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2197", + "right": "pull_request:3839", + "accept": false, + "reason": "Different bugs in different areas: inpaint pipeline argument typo vs SDE VE noise handling." + }, + { + "left": "pull_request:2209", + "right": "pull_request:4317", + "accept": false, + "reason": "Docs argument-name fix in img2img example vs DreamBooth validation option fix; unrelated." + }, + { + "left": "pull_request:2257", + "right": "pull_request:874", + "accept": false, + "reason": "Both touch community docs, but one adds UnCLIP interpolation and the other adds a generic inference example." + }, + { + "left": "pull_request:2278", + "right": "pull_request:309", + "accept": false, + "reason": "Video Stable Diffusion pipeline addition vs README link fix; not the same change." + }, + { + "left": "pull_request:2278", + "right": "pull_request:386", + "accept": false, + "reason": "Video Stable Diffusion community pipeline vs stable diffusion API docs; unrelated." + }, + { + "left": "pull_request:2287", + "right": "pull_request:2629", + "accept": false, + "reason": "W&B logging for unconditional image generation vs SAG non-square support; different code paths and goals." + }, + { + "left": "pull_request:2290", + "right": "pull_request:642", + "accept": false, + "reason": "Pix2Pix docs indentation fix vs a docs link update for an unrelated training notebook." + }, + { + "left": "pull_request:2320", + "right": "pull_request:4546", + "accept": false, + "reason": "Different typo fixes in different conversion/script files; not the same underlying issue." + }, + { + "left": "pull_request:2320", + "right": "pull_request:948", + "accept": false, + "reason": "Checkpoint conversion filename typo vs a DDPM comment typo; unrelated." + }, + { + "left": "pull_request:2333", + "right": "pull_request:3680", + "accept": false, + "reason": "Ethical guidelines formatting cleanup vs loader-method link fix; unrelated docs changes." + }, + { + "left": "pull_request:2362", + "right": "pull_request:2766", + "accept": false, + "reason": "Both concern MPS/test behavior, but one adds a skip decorator while the other changes expected slices for torch 2.0." + }, + { + "left": "pull_request:2362", + "right": "pull_request:796", + "accept": false, + "reason": "MPS skip decorator vs Apple M1 test setup; different fixes." + }, + { + "left": "pull_request:2367", + "right": "pull_request:2557", + "accept": false, + "reason": "Both DreamBooth training scripts, but one adds checkpoint limits and the other adds intermediate logging." + }, + { + "left": "pull_request:2383", + "right": "pull_request:889", + "accept": false, + "reason": "Pix2Pix Zero edit-vector utility vs DiffusionPipeline components support; unrelated." + }, + { + "left": "pull_request:2395", + "right": "pull_request:2762", + "accept": false, + "reason": "Pix2Pix Zero docs on embeddings vs ControlNet multi-conditioning docs; unrelated features." + }, + { + "left": "pull_request:2396", + "right": "pull_request:2530", + "accept": false, + "reason": "Scheduler test warning cleanup vs EMAModel test suite; unrelated." + }, + { + "left": "pull_request:2396", + "right": "pull_request:934", + "accept": false, + "reason": "UniPC scheduler test fixes vs MPS failing tests; different scopes." + }, + { + "left": "pull_request:2398", + "right": "pull_request:3473", + "accept": false, + "reason": "Pix2Pix Zero inversion docs vs textual inversion inference docs; same broad topic, different pipelines and content." + }, + { + "left": "pull_request:2398", + "right": "pull_request:3852", + "accept": false, + "reason": "Pix2Pix Zero docs vs a contributor note in paradigms docs; unrelated." + }, + { + "left": "pull_request:2400", + "right": "pull_request:869", + "accept": false, + "reason": "Both touch community README, but one adds image interpolation and the other fixes README/import behavior." + }, + { + "left": "pull_request:2433", + "right": "pull_request:3388", + "accept": false, + "reason": "SDEdit note in img2img docs vs installation docs update; unrelated." + }, + { + "left": "pull_request:2446", + "right": "pull_request:3971", + "accept": false, + "reason": "Pix2Pix docs snippet fix vs SD XL docs/workflow fixes; unrelated." + }, + { + "left": "pull_request:2487", + "right": "pull_request:3549", + "accept": false, + "reason": "Stable Diffusion JAX/Flax guide vs evaluation docs update; different documentation topics." + }, + { + "left": "pull_request:2487", + "right": "pull_request:3894", + "accept": false, + "reason": "Docs guide vs Flax dropout implementation change; not the same underlying change." + }, + { + "left": "pull_request:2487", + "right": "pull_request:4766", + "accept": false, + "reason": "Both docs, but one is a JAX/Flax how-to and the other is a Pix2Pix Zero guide; unrelated." + }, + { + "left": "pull_request:2498", + "right": "pull_request:3079", + "accept": false, + "reason": "Both touch attend-and-excite tests, but one addresses timeout by shrinking batch coverage while the other speeds up fast tests by changing model/input settings." + }, + { + "left": "pull_request:2503", + "right": "pull_request:2766", + "accept": false, + "reason": "PyTorch 2.0 benchmark/docs change vs MPS test expectation fix; unrelated." + }, + { + "left": "pull_request:2533", + "right": "pull_request:3168", + "accept": false, + "reason": "RC1 testing workflow/package pinning vs doc-builder testing WIP; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 124, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2536", + "pull_request:2539", + "pull_request:2540", + "pull_request:2543", + "pull_request:2559", + "pull_request:2563", + "pull_request:2566", + "pull_request:2598", + "pull_request:2602", + "pull_request:2637", + "pull_request:2639", + "pull_request:2642", + "pull_request:2647", + "pull_request:2677", + "pull_request:2678", + "pull_request:2684", + "pull_request:2685", + "pull_request:2693", + "pull_request:2748", + "pull_request:2766", + "pull_request:2767", + "pull_request:2812", + "pull_request:2817", + "pull_request:2836", + "pull_request:2845", + "pull_request:2861", + "pull_request:2958", + "pull_request:3163", + "pull_request:3581", + "pull_request:3710", + "pull_request:3789", + "pull_request:3847", + "pull_request:3923", + "pull_request:3941", + "pull_request:3942", + "pull_request:3961", + "pull_request:4022", + "pull_request:4145", + "pull_request:4203", + "pull_request:421", + "pull_request:4275", + "pull_request:4546", + "pull_request:4721", + "pull_request:4853", + "pull_request:641", + "pull_request:651", + "pull_request:948" + ], + "soft_pairs": [ + "pull_request:2536|pull_request:3789", + "pull_request:2539|pull_request:3941", + "pull_request:2540|pull_request:2677", + "pull_request:2543|pull_request:651", + "pull_request:2559|pull_request:2677", + "pull_request:2559|pull_request:421", + "pull_request:2559|pull_request:4275", + "pull_request:2559|pull_request:4721", + "pull_request:2563|pull_request:4145", + "pull_request:2566|pull_request:4546", + "pull_request:2566|pull_request:948", + "pull_request:2598|pull_request:4022", + "pull_request:2598|pull_request:4203", + "pull_request:2602|pull_request:641", + "pull_request:2637|pull_request:3163", + "pull_request:2637|pull_request:3961", + "pull_request:2639|pull_request:2748", + "pull_request:2642|pull_request:3163", + "pull_request:2642|pull_request:3961", + "pull_request:2647|pull_request:3163", + "pull_request:2647|pull_request:3961", + "pull_request:2677|pull_request:2958", + "pull_request:2678|pull_request:3942", + "pull_request:2684|pull_request:3163", + "pull_request:2684|pull_request:3961", + "pull_request:2685|pull_request:3961", + "pull_request:2693|pull_request:2836", + "pull_request:2766|pull_request:3923", + "pull_request:2767|pull_request:3581", + "pull_request:2812|pull_request:4022", + "pull_request:2817|pull_request:3710", + "pull_request:2845|pull_request:4853", + "pull_request:2861|pull_request:3847" + ], + "prepared_review_unit_hash": "ef1da14cb2819bd5e2c2aa695bce9ebc20f6d0db89d190d7ff317713ab293329", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 33, + "serialized_chars": 93715, + "estimated_input_tokens": 23429, + "estimated_eval_tokens": 47114 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicate cluster emerges here: the soft pairs are mostly unrelated docs/typo/test/code changes across different files and bugs, so all candidate edges should be rejected.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a valid canonical duplicate target; the items are heterogeneous and do not resolve to one underlying change. If forced to pick a standalone representative, PR #2845 is the clearest concrete bug fix, but it is not a duplicate hub.", + "best_issue_reason": null, + "best_pr_reason": "PR #2845 is the strongest standalone PR in the set: it has an explicit issue target, a concrete code-path fix, and a focused behavioral change rather than a minor doc/test typo.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2536", + "right": "pull_request:3789", + "accept": false, + "reason": "Different features: VAE gradient checkpointing vs 3D U-Net gradient checkpointing." + }, + { + "left": "pull_request:2539", + "right": "pull_request:3941", + "accept": false, + "reason": "Unrelated doc tweaks in different pipelines; one changes a default doc value, the other adds import guidance." + }, + { + "left": "pull_request:2540", + "right": "pull_request:2677", + "accept": false, + "reason": "Different documentation sections: ControlNet API docs vs training docs entry." + }, + { + "left": "pull_request:2543", + "right": "pull_request:651", + "accept": false, + "reason": "Both mention warnings, but they target different subsystems and different deprecation issues." + }, + { + "left": "pull_request:2559", + "right": "pull_request:2677", + "accept": false, + "reason": "Different docs pages and different edits; no shared underlying bug/change." + }, + { + "left": "pull_request:2559", + "right": "pull_request:421", + "accept": false, + "reason": "Both are docs, but about different topics and files; not the same change." + }, + { + "left": "pull_request:2559", + "right": "pull_request:4275", + "accept": false, + "reason": "Different docs fixes; no concrete overlap beyond being documentation edits." + }, + { + "left": "pull_request:2559", + "right": "pull_request:4721", + "accept": false, + "reason": "Different language/docs updates with no shared bug or text change." + }, + { + "left": "pull_request:2563", + "right": "pull_request:4145", + "accept": false, + "reason": "EMA cache cleanup is unrelated to the UnCLIP pipeline placeholder change." + }, + { + "left": "pull_request:2566", + "right": "pull_request:4546", + "accept": false, + "reason": "Separate typo fixes in unrelated files and code paths." + }, + { + "left": "pull_request:2566", + "right": "pull_request:948", + "accept": false, + "reason": "Training tensorboard tracker typo vs a pipeline comment typo; not the same fix." + }, + { + "left": "pull_request:2598", + "right": "pull_request:4022", + "accept": false, + "reason": "Both are docs, but one is PaintByExample docs and the other is Korean translation work." + }, + { + "left": "pull_request:2598", + "right": "pull_request:4203", + "accept": false, + "reason": "Different documentation targets and content changes; not duplicates." + }, + { + "left": "pull_request:2602", + "right": "pull_request:641", + "accept": false, + "reason": "Both pin huggingface-hub in setup.py, but to different versions for different test releases; not the same change." + }, + { + "left": "pull_request:2637", + "right": "pull_request:3163", + "accept": false, + "reason": "Quicktour wording fix vs custom diffusion attribution/update; unrelated docs edits." + }, + { + "left": "pull_request:2637", + "right": "pull_request:3961", + "accept": false, + "reason": "Different doc pages and different text corrections." + }, + { + "left": "pull_request:2639", + "right": "pull_request:2748", + "accept": false, + "reason": "Basic training docs vs torch 2.0 optimization docs; no shared underlying issue." + }, + { + "left": "pull_request:2642", + "right": "pull_request:3163", + "accept": false, + "reason": "Loading docs wording fix vs custom diffusion attribution/code update; unrelated." + }, + { + "left": "pull_request:2642", + "right": "pull_request:3961", + "accept": false, + "reason": "Different docs topics and edits." + }, + { + "left": "pull_request:2647", + "right": "pull_request:3163", + "accept": false, + "reason": "Schedulers docs wording fix vs custom diffusion attribution/update; no duplicate basis." + }, + { + "left": "pull_request:2647", + "right": "pull_request:3961", + "accept": false, + "reason": "Different docs pages and textual changes only." + }, + { + "left": "pull_request:2677", + "right": "pull_request:2958", + "accept": false, + "reason": "Training docs expansion vs adding a controlnet inpaint pipeline; different artifacts and goals." + }, + { + "left": "pull_request:2678", + "right": "pull_request:3942", + "accept": false, + "reason": "Both touch tests, but one changes EMA serialization setup and the other fixes xformers NP test handling." + }, + { + "left": "pull_request:2684", + "right": "pull_request:3163", + "accept": false, + "reason": "Custom pipeline overview wording vs custom diffusion credit/update; unrelated." + }, + { + "left": "pull_request:2684", + "right": "pull_request:3961", + "accept": false, + "reason": "Different docs pages with no shared bug or edit." + }, + { + "left": "pull_request:2685", + "right": "pull_request:3961", + "accept": false, + "reason": "KerasCV docs punctuation fix vs contributor credit change; unrelated." + }, + { + "left": "pull_request:2693", + "right": "pull_request:2836", + "accept": false, + "reason": "Inpaint image-link fix vs DreamBooth LoRA link fix; same broad docs area but different issues." + }, + { + "left": "pull_request:2766", + "right": "pull_request:3923", + "accept": false, + "reason": "Both add/import torch in docs/tests contexts, but they fix different files and failures." + }, + { + "left": "pull_request:2767", + "right": "pull_request:3581", + "accept": false, + "reason": "LoRA training args refactor vs bit_diffusion import cleanup; no shared code-path problem." + }, + { + "left": "pull_request:2812", + "right": "pull_request:4022", + "accept": false, + "reason": "Stable unCLIP docs update vs Korean translation update; not the same change." + }, + { + "left": "pull_request:2817", + "right": "pull_request:3710", + "accept": false, + "reason": "UNet1DModel docs for extra_in_channels vs image processor docs entry; unrelated." + }, + { + "left": "pull_request:2845", + "right": "pull_request:4853", + "accept": false, + "reason": "StableUnCLIP image_embeds handling vs image_processor width bug; completely different code paths." + }, + { + "left": "pull_request:2861", + "right": "pull_request:3847", + "accept": false, + "reason": "StableUnCLIP test for image_embeds=None vs dependency-check workflow/tests; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 125, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2869", + "pull_request:2897", + "pull_request:2911", + "pull_request:2912", + "pull_request:2958", + "pull_request:2974", + "pull_request:3023", + "pull_request:3063", + "pull_request:3118", + "pull_request:3121", + "pull_request:3123", + "pull_request:3132", + "pull_request:3165", + "pull_request:3215", + "pull_request:3226", + "pull_request:3262", + "pull_request:3266", + "pull_request:3271", + "pull_request:3334", + "pull_request:3345", + "pull_request:3370", + "pull_request:3388", + "pull_request:3398", + "pull_request:3417", + "pull_request:3525", + "pull_request:3542", + "pull_request:3549", + "pull_request:3563", + "pull_request:3568", + "pull_request:3690", + "pull_request:3691", + "pull_request:4227", + "pull_request:4258", + "pull_request:4333", + "pull_request:4531", + "pull_request:4546", + "pull_request:4564", + "pull_request:4626", + "pull_request:4696", + "pull_request:4796", + "pull_request:4910", + "pull_request:4934", + "pull_request:5323", + "pull_request:5671", + "pull_request:5844", + "pull_request:848", + "pull_request:897", + "pull_request:948" + ], + "soft_pairs": [ + "pull_request:2869|pull_request:3165", + "pull_request:2897|pull_request:3690", + "pull_request:2911|pull_request:3165", + "pull_request:2912|pull_request:3165", + "pull_request:2912|pull_request:4564", + "pull_request:2958|pull_request:3121", + "pull_request:2974|pull_request:897", + "pull_request:3023|pull_request:3691", + "pull_request:3063|pull_request:848", + "pull_request:3118|pull_request:4333", + "pull_request:3118|pull_request:948", + "pull_request:3123|pull_request:3132", + "pull_request:3132|pull_request:4546", + "pull_request:3132|pull_request:948", + "pull_request:3215|pull_request:4696", + "pull_request:3226|pull_request:3542", + "pull_request:3262|pull_request:4564", + "pull_request:3266|pull_request:4796", + "pull_request:3271|pull_request:4934", + "pull_request:3334|pull_request:4910", + "pull_request:3345|pull_request:4531", + "pull_request:3370|pull_request:4564", + "pull_request:3388|pull_request:4227", + "pull_request:3398|pull_request:5323", + "pull_request:3417|pull_request:5671", + "pull_request:3525|pull_request:4258", + "pull_request:3549|pull_request:5844", + "pull_request:3563|pull_request:3690", + "pull_request:3568|pull_request:4626" + ], + "prepared_review_unit_hash": "dce12af40e1091fe675c0556c6d638534c53cd72bab6feb08d59a947881e683d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 92053, + "estimated_input_tokens": 23014, + "estimated_eval_tokens": 46284 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated docs typos, link fixes, and a few separate code changes across different subsystems. None of the soft pairs look like the same underlying bug or change, so all should be rejected as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 4796 is the most substantial and concrete fix in the set, but it is still not representative of the others; the cluster is not a true duplicate family.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2869", + "right": "pull_request:3165", + "accept": false, + "reason": "Both are tiny docs edits, but they touch different pages and unrelated wording; not the same change." + }, + { + "left": "pull_request:2897", + "right": "pull_request:3690", + "accept": false, + "reason": "Stable UnCLIP docs/example addition and a Kandinsky pipeline example fix are unrelated changes." + }, + { + "left": "pull_request:2911", + "right": "pull_request:3165", + "accept": false, + "reason": "Different documentation pages and different typo fixes; no shared underlying bug." + }, + { + "left": "pull_request:2912", + "right": "pull_request:3165", + "accept": false, + "reason": "ControlNet docs punctuation fix vs a custom diffusion docs edit; unrelated content." + }, + { + "left": "pull_request:2912", + "right": "pull_request:4564", + "accept": false, + "reason": "A docs punctuation fix is not the same as updating SDXL ControlNet model/repo names." + }, + { + "left": "pull_request:2958", + "right": "pull_request:3121", + "accept": false, + "reason": "ControlNet inpaint pipeline addition and global pooling in ControlNet model are separate code changes." + }, + { + "left": "pull_request:2974", + "right": "pull_request:897", + "accept": false, + "reason": "Community README/TensorRT pipeline work is not the same as adding a speech-to-image example." + }, + { + "left": "pull_request:3023", + "right": "pull_request:3691", + "accept": false, + "reason": "Both relate to attention/torch 2.0, but they fix different code paths and different bugs." + }, + { + "left": "pull_request:3063", + "right": "pull_request:848", + "accept": false, + "reason": "Pipeline __setattr__ None-handling and Flax width/height handling are unrelated bugs." + }, + { + "left": "pull_request:3118", + "right": "pull_request:4333", + "accept": false, + "reason": "ControlNet default-arg fix and DreamBooth validation-scheduler change are different features/issues." + }, + { + "left": "pull_request:3118", + "right": "pull_request:948", + "accept": false, + "reason": "A ControlNet API fix and a DDPM comment typo are clearly unrelated." + }, + { + "left": "pull_request:3123", + "right": "pull_request:3132", + "accept": false, + "reason": "Textual inversion doc typo and ScoreSdeVe type hint fix are unrelated." + }, + { + "left": "pull_request:3132", + "right": "pull_request:4546", + "accept": false, + "reason": "ScoreSdeVe type hints and a ControlNet converter docstring typo do not share a concrete bug." + }, + { + "left": "pull_request:3132", + "right": "pull_request:948", + "accept": false, + "reason": "Different files, different subsystems, and no common fix." + }, + { + "left": "pull_request:3215", + "right": "pull_request:4696", + "accept": false, + "reason": "8-bit loading checks in pipeline_utils and T2I-Adapter-XL model additions are unrelated." + }, + { + "left": "pull_request:3226", + "right": "pull_request:3542", + "accept": false, + "reason": "ControlNet v1.1 docs and ControlNet LoRA support are different changes, not one bug fix." + }, + { + "left": "pull_request:3262", + "right": "pull_request:4564", + "accept": false, + "reason": "Updating IF naming to XL and updating the SDXL ControlNet repo name are separate documentation/model-name changes." + }, + { + "left": "pull_request:3266", + "right": "pull_request:4796", + "accept": false, + "reason": "Community pipeline version loading and SDXL type-mismatch fixes are different bugs in different code paths." + }, + { + "left": "pull_request:3271", + "right": "pull_request:4934", + "accept": false, + "reason": "Linking training docs is not the same as adding the T2I-Adapter training script and docs." + }, + { + "left": "pull_request:3334", + "right": "pull_request:4910", + "accept": false, + "reason": "Docstring cleanup and an inpainting example variable fix are unrelated." + }, + { + "left": "pull_request:3345", + "right": "pull_request:4531", + "accept": false, + "reason": "README pipeline-class correction and reference-pipeline indexing fix are different issues." + }, + { + "left": "pull_request:3370", + "right": "pull_request:4564", + "accept": false, + "reason": "PyTorch 2.0 docs updates and SDXL ControlNet repo-name updates are unrelated documentation changes." + }, + { + "left": "pull_request:3388", + "right": "pull_request:4227", + "accept": false, + "reason": "Installation docs update and a missing import in textual inversion are different fixes." + }, + { + "left": "pull_request:3398", + "right": "pull_request:5323", + "accept": false, + "reason": "Stable Diffusion typo fix and adapter-doc link fix are unrelated docs edits." + }, + { + "left": "pull_request:3417", + "right": "pull_request:5671", + "accept": false, + "reason": "UnCLIP scheduler note and a dependency test workflow tweak are not the same change." + }, + { + "left": "pull_request:3525", + "right": "pull_request:4258", + "accept": false, + "reason": "README marketing/content changes and an SDXL hyperparameter note are unrelated." + }, + { + "left": "pull_request:3549", + "right": "pull_request:5844", + "accept": false, + "reason": "Evaluation-doc prose update and a Japanese SDXL reference addition are different docs changes." + }, + { + "left": "pull_request:3563", + "right": "pull_request:3690", + "accept": false, + "reason": "A VQ pipeline block fix and a Kandinsky example fix are unrelated code paths." + }, + { + "left": "pull_request:3568", + "right": "pull_request:4626", + "accept": false, + "reason": "Broken internal links in tutorials and text2image training-link fixes are not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 126, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3577", + "pull_request:3592", + "pull_request:3625", + "pull_request:3685", + "pull_request:3710", + "pull_request:3755", + "pull_request:3786", + "pull_request:382", + "pull_request:3839", + "pull_request:3852", + "pull_request:3903", + "pull_request:392", + "pull_request:3941", + "pull_request:3942", + "pull_request:3962", + "pull_request:3971", + "pull_request:3979", + "pull_request:3987", + "pull_request:3994", + "pull_request:4022", + "pull_request:406", + "pull_request:4070", + "pull_request:4082", + "pull_request:4099", + "pull_request:4113", + "pull_request:420", + "pull_request:4267", + "pull_request:4285", + "pull_request:4320", + "pull_request:4397", + "pull_request:4437", + "pull_request:4729", + "pull_request:4740", + "pull_request:4741", + "pull_request:4886", + "pull_request:4965", + "pull_request:5013", + "pull_request:5065", + "pull_request:5367", + "pull_request:5411", + "pull_request:5549", + "pull_request:5550", + "pull_request:5745", + "pull_request:5844", + "pull_request:5911", + "pull_request:710", + "pull_request:910" + ], + "soft_pairs": [ + "pull_request:3577|pull_request:5411", + "pull_request:3592|pull_request:5065", + "pull_request:3625|pull_request:4740", + "pull_request:3685|pull_request:5911", + "pull_request:3710|pull_request:4285", + "pull_request:3755|pull_request:3942", + "pull_request:3786|pull_request:3994", + "pull_request:382|pull_request:910", + "pull_request:3839|pull_request:4320", + "pull_request:3852|pull_request:5844", + "pull_request:3903|pull_request:5550", + "pull_request:392|pull_request:420", + "pull_request:3941|pull_request:4741", + "pull_request:3941|pull_request:5745", + "pull_request:3962|pull_request:4437", + "pull_request:3971|pull_request:4965", + "pull_request:3979|pull_request:4267", + "pull_request:3979|pull_request:4729", + "pull_request:3979|pull_request:5013", + "pull_request:3987|pull_request:4113", + "pull_request:4022|pull_request:5549", + "pull_request:406|pull_request:710", + "pull_request:4070|pull_request:5367", + "pull_request:4082|pull_request:5367", + "pull_request:4099|pull_request:4397", + "pull_request:4099|pull_request:4886" + ], + "prepared_review_unit_hash": "c9caf1d8ce1ab85ce78ae11b83b43254f7f65448832a75c2fbf375c19214e629", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 92491, + "estimated_input_tokens": 23123, + "estimated_eval_tokens": 46502 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a mixed bag of unrelated docs, workflow, test, and code-fix PRs. The soft pairs mostly match on broad keywords like \"docs\", \"fix\", or the same subsystem, but they do not describe the same underlying bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits: the set spans many independent fixes across different files and subsystems, so there is no shared underlying change to canonicalize.", + "best_issue_reason": null, + "best_pr_reason": "PR 4099 is the strongest standalone bug-fix PR in the set because it has an explicit issue target, a concrete code-path correction, and test updates; however, it is not a duplicate of the other items.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3577", + "right": "pull_request:5411", + "accept": false, + "reason": "Both touch docs, but 3577 fixes broken README links while 5411 fixes a broken textual inversion training link and related code/docs; different content and purpose." + }, + { + "left": "pull_request:3592", + "right": "pull_request:5065", + "accept": false, + "reason": "3592 removes debug prints in attention_processor.py; 5065 is a broad LoRA/attention-processor typo fix with many docs/tests/file changes. Same area, not the same bug." + }, + { + "left": "pull_request:3625", + "right": "pull_request:4740", + "accept": false, + "reason": "A doc-builder workflow test PR and a safetensors dependency pin are unrelated changes with no shared code path." + }, + { + "left": "pull_request:3685", + "right": "pull_request:5911", + "accept": false, + "reason": "3685 adds thread inference support to StableDiffusionPipeline; 5911 adds IP-Adapter Full support. Different features and code paths." + }, + { + "left": "pull_request:3710", + "right": "pull_request:4285", + "accept": false, + "reason": "Both are documentation PRs, but they add different pages for different features (image processor vs ONNX/OpenVINO docs)." + }, + { + "left": "pull_request:3755", + "right": "pull_request:3942", + "accept": false, + "reason": "Both adjust tests, but they target different test files and different failures; not the same failing test or bug." + }, + { + "left": "pull_request:3786", + "right": "pull_request:3994", + "accept": false, + "reason": "A PR template addition is unrelated to adding an act_fn parameter to OutValueFunctionBlock." + }, + { + "left": "pull_request:382", + "right": "pull_request:910", + "accept": false, + "reason": "These are separate documentation updates for different pipelines and eras; no shared underlying change." + }, + { + "left": "pull_request:3839", + "right": "pull_request:4320", + "accept": false, + "reason": "3839 fixes scheduler noise handling in code, while 4320 is a documentation typo fix. Different artifact types and problems." + }, + { + "left": "pull_request:3852", + "right": "pull_request:5844", + "accept": false, + "reason": "Both are docs, but one adds a contributor note to paradigms docs and the other adds a Japanese SDXL reference example. Unrelated." + }, + { + "left": "pull_request:3903", + "right": "pull_request:5550", + "accept": false, + "reason": "3903 is a large Kandinsky2.2 implementation PR; 5550 only corrects a checkpoint URL in the docs. Not the same change." + }, + { + "left": "pull_request:392", + "right": "pull_request:420", + "accept": false, + "reason": "General docs/contribution content versus optimization-section cleanup; different pages and objectives." + }, + { + "left": "pull_request:3941", + "right": "pull_request:4741", + "accept": false, + "reason": "3941 fixes a broken loading-doc code snippet; 4741 fixes dummy imports for AudioLDM2. Different bugs and files." + }, + { + "left": "pull_request:3941", + "right": "pull_request:5745", + "accept": false, + "reason": "A docs code-snippet/import fix is unrelated to a consistency-decoder docstring type correction." + }, + { + "left": "pull_request:3962", + "right": "pull_request:4437", + "accept": false, + "reason": "3962 changes a doc URL; 4437 changes model sampling behavior for consistency models. Not the same issue." + }, + { + "left": "pull_request:3971", + "right": "pull_request:4965", + "accept": false, + "reason": "Both are documentation-only, but they update different docs pages for different features; no duplicate change." + }, + { + "left": "pull_request:3979", + "right": "pull_request:4267", + "accept": false, + "reason": "3979 fixes SDXL pipeline code; 4267 only swaps images in SDXL docs. Code bug vs documentation asset update." + }, + { + "left": "pull_request:3979", + "right": "pull_request:4729", + "accept": false, + "reason": "Both are runtime code fixes, but they affect different pipelines and different symptoms; they are not mergeable as one fix." + }, + { + "left": "pull_request:3979", + "right": "pull_request:5013", + "accept": false, + "reason": "3979 targets SDXL upscale attention handling, while 5013 fixes SDXL InstructPix2Pix image latent dtype. Same family, different concrete bug." + }, + { + "left": "pull_request:3987", + "right": "pull_request:4113", + "accept": false, + "reason": "Audio Diffusion doc snippet fix versus SDXL inpaint dtype fix; unrelated." + }, + { + "left": "pull_request:4022", + "right": "pull_request:5549", + "accept": false, + "reason": "Different translation rollouts (Korean vs initial Portuguese) with separate doc tree/content changes." + }, + { + "left": "pull_request:406", + "right": "pull_request:710", + "accept": false, + "reason": "406 fixes a toctree typo; 710 fixes docstring formatting for issue #709. No shared bug." + }, + { + "left": "pull_request:4070", + "right": "pull_request:5367", + "accept": false, + "reason": "4070 makes minor SDXL docs wording/link changes; 5367 fixes SDXL tutorial pipeline creation code. Different artifact and scope." + }, + { + "left": "pull_request:4082", + "right": "pull_request:5367", + "accept": false, + "reason": "A dependency file tweak for SDXL training is unrelated to the tutorial pipeline creation fix." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4397", + "accept": false, + "reason": "4099 fixes prompt-embedding handling in SDXL pipelines; 4397 only updates an SDXL docstring default. Different code paths and severity." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4886", + "accept": false, + "reason": "Both are pipeline bug fixes, but one corrects prompt embedding duplication in SDXL while the other fixes safety-checker preprocessing in UpscalePipeline. Not the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 127, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4196", + "pull_request:421", + "pull_request:4216", + "pull_request:4238", + "pull_request:4258", + "pull_request:4267", + "pull_request:4276", + "pull_request:4285", + "pull_request:4315", + "pull_request:4350", + "pull_request:4351", + "pull_request:4380", + "pull_request:4457", + "pull_request:4499", + "pull_request:4500", + "pull_request:4501", + "pull_request:4502", + "pull_request:4582", + "pull_request:4612", + "pull_request:4626", + "pull_request:4629", + "pull_request:4652", + "pull_request:4659", + "pull_request:4661", + "pull_request:4666", + "pull_request:4696", + "pull_request:4712", + "pull_request:4717", + "pull_request:4965", + "pull_request:5013", + "pull_request:5104", + "pull_request:5236", + "pull_request:5309", + "pull_request:5369", + "pull_request:5454", + "pull_request:5982", + "pull_request:6197", + "pull_request:6205", + "pull_request:6217", + "pull_request:642", + "pull_request:6438", + "pull_request:6440", + "pull_request:6638", + "pull_request:6647", + "pull_request:6669", + "pull_request:6721", + "pull_request:6757", + "pull_request:6962" + ], + "soft_pairs": [ + "pull_request:4196|pull_request:5982", + "pull_request:421|pull_request:642", + "pull_request:4216|pull_request:4457", + "pull_request:4238|pull_request:4717", + "pull_request:4258|pull_request:5454", + "pull_request:4258|pull_request:6438", + "pull_request:4267|pull_request:6647", + "pull_request:4276|pull_request:6638", + "pull_request:4285|pull_request:4696", + "pull_request:4315|pull_request:6217", + "pull_request:4350|pull_request:4965", + "pull_request:4350|pull_request:6205", + "pull_request:4351|pull_request:6669", + "pull_request:4380|pull_request:5309", + "pull_request:4380|pull_request:5369", + "pull_request:4499|pull_request:6197", + "pull_request:4500|pull_request:6197", + "pull_request:4501|pull_request:6197", + "pull_request:4502|pull_request:6197", + "pull_request:4582|pull_request:6647", + "pull_request:4612|pull_request:4965", + "pull_request:4626|pull_request:5236", + "pull_request:4626|pull_request:6440", + "pull_request:4629|pull_request:6757", + "pull_request:4652|pull_request:5013", + "pull_request:4652|pull_request:5104", + "pull_request:4652|pull_request:6962", + "pull_request:4659|pull_request:6721", + "pull_request:4661|pull_request:6721", + "pull_request:4666|pull_request:4712" + ], + "prepared_review_unit_hash": "64bc4da16c90dc0dd0c11d9dfa78eda689d570e20e1c02594bda117843a79180", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 98260, + "estimated_input_tokens": 24565, + "estimated_eval_tokens": 49386 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of SDXL-related docs, tests, and feature PRs rather than true duplicates. I found no strong duplicate PR pairs to merge, and the best representative PR is the broad T2I-Adapter-XL feature addition.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4696 is the most central and substantive item: it adds T2I-Adapter-XL support across code, docs, and tests, with extensive review activity and broad impact.", + "best_issue_reason": null, + "best_pr_reason": "PR 4696 is the strongest canonical representative because it is the largest, most complete SDXL-related change in the set and covers a concrete feature rather than a small docs-only tweak.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4196", + "right": "pull_request:5982", + "accept": false, + "reason": "Both are docs PRs, but one improves SDXL pipeline docs and the other adds SDXL Turbo docs; different artifacts and different change scopes." + }, + { + "left": "pull_request:421", + "right": "pull_request:642", + "accept": false, + "reason": "Both touch documentation, but they fix unrelated docs pages in different areas of the project." + }, + { + "left": "pull_request:4216", + "right": "pull_request:4457", + "accept": false, + "reason": "They both involve the TensorRT inpaint example, but 4216 changes timestep handling/strength behavior while 4457 is a separate minor example/README cleanup." + }, + { + "left": "pull_request:4238", + "right": "pull_request:4717", + "accept": false, + "reason": "One fixes SDXL ControlNet inference code; the other is a docs formatting fix. Not the same underlying bug or change." + }, + { + "left": "pull_request:4258", + "right": "pull_request:5454", + "accept": false, + "reason": "Both are docs-only changes, but they concern unrelated documentation pages and link updates." + }, + { + "left": "pull_request:4258", + "right": "pull_request:6438", + "accept": false, + "reason": "Different docs targets: one rewrites SDXL DreamBooth notes, the other updates installation guidance." + }, + { + "left": "pull_request:4267", + "right": "pull_request:6647", + "accept": false, + "reason": "One fixes an SDXL docs image reference; the other updates an SDXL k-diffusion test baseline. Different kinds of changes." + }, + { + "left": "pull_request:4276", + "right": "pull_request:6638", + "accept": false, + "reason": "Both are tests, but they validate different behaviors in different code paths and are not the same fix." + }, + { + "left": "pull_request:4285", + "right": "pull_request:4696", + "accept": false, + "reason": "ONNX/OpenVINO documentation work is unrelated to T2I-Adapter-XL model and pipeline implementation." + }, + { + "left": "pull_request:4315", + "right": "pull_request:6217", + "accept": false, + "reason": "Both mention torch compile, but they fix different components and failure modes in different pipelines." + }, + { + "left": "pull_request:4350", + "right": "pull_request:4965", + "accept": false, + "reason": "Both are docs edits, but they target unrelated topics: T2I-Adapter fp16 usage vs Wuerstchen formatting." + }, + { + "left": "pull_request:4350", + "right": "pull_request:6205", + "accept": false, + "reason": "Same broad T2I-adapter area, but different docs and different issues: dtype usage vs blog link correction." + }, + { + "left": "pull_request:4351", + "right": "pull_request:6669", + "accept": false, + "reason": "Both touch README.md, but they change unrelated sections: image URL vs repository stats text." + }, + { + "left": "pull_request:4380", + "right": "pull_request:5309", + "accept": false, + "reason": "Both are minor docs fixes, but they address different documentation pages and content errors." + }, + { + "left": "pull_request:4380", + "right": "pull_request:5369", + "accept": false, + "reason": "These are separate docs cleanups in different guides, not the same underlying change." + }, + { + "left": "pull_request:4499", + "right": "pull_request:6197", + "accept": false, + "reason": "One adds SDXL reference-attn/adain support; the other adds DeepCache support. Different features and code paths." + }, + { + "left": "pull_request:4500", + "right": "pull_request:6197", + "accept": false, + "reason": "Same as above: SDXL reference pipeline work is unrelated to DeepCache support." + }, + { + "left": "pull_request:4501", + "right": "pull_request:6197", + "accept": false, + "reason": "SDXL reference-attn/adain support and DeepCache support are distinct feature additions." + }, + { + "left": "pull_request:4502", + "right": "pull_request:6197", + "accept": false, + "reason": "The merged SDXL reference PR is not the same change as DeepCache support." + }, + { + "left": "pull_request:4582", + "right": "pull_request:6647", + "accept": false, + "reason": "One corrects a ControlNet SDXL docstring; the other updates an SDXL test expectation. Not the same bug." + }, + { + "left": "pull_request:4612", + "right": "pull_request:4965", + "accept": false, + "reason": "They are unrelated docs updates for different pipeline families." + }, + { + "left": "pull_request:4626", + "right": "pull_request:5236", + "accept": false, + "reason": "Both are link/docs fixes, but in different languages and different pages with no shared concrete change." + }, + { + "left": "pull_request:4626", + "right": "pull_request:6440", + "accept": false, + "reason": "Both edit links, but they address different docs and different local-link problems." + }, + { + "left": "pull_request:4629", + "right": "pull_request:6757", + "accept": false, + "reason": "Long weighted prompt pipeline docs/code and SDXL k-diffusion sigma placement are unrelated." + }, + { + "left": "pull_request:4652", + "right": "pull_request:5013", + "accept": false, + "reason": "One fixes LoRA test serialization; the other fixes SDXL InstructPix2Pix latent dtype handling. Different issues." + }, + { + "left": "pull_request:4652", + "right": "pull_request:5104", + "accept": false, + "reason": "A LoRA test fix and a new timestep-bias training feature are not the same change." + }, + { + "left": "pull_request:4652", + "right": "pull_request:6962", + "accept": false, + "reason": "These are tests for different pipelines and failure modes, so they are not duplicates." + }, + { + "left": "pull_request:4659", + "right": "pull_request:6721", + "accept": false, + "reason": "A weighted-prompt pipeline/docs replacement is unrelated to AnimateDiff SDXL support." + }, + { + "left": "pull_request:4661", + "right": "pull_request:6721", + "accept": false, + "reason": "Same reasoning: weighted-prompt pipeline work is not the same as AnimateDiff SDXL." + }, + { + "left": "pull_request:4666", + "right": "pull_request:4712", + "accept": false, + "reason": "SDXL ControlNet LoRA support is a pipeline feature; tests_fetcher is unrelated infrastructure work." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 128, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4684", + "pull_request:4696", + "pull_request:4717", + "pull_request:4740", + "pull_request:4748", + "pull_request:4796", + "pull_request:4797", + "pull_request:4822", + "pull_request:4856", + "pull_request:4869", + "pull_request:4934", + "pull_request:4946", + "pull_request:4952", + "pull_request:5013", + "pull_request:5051", + "pull_request:5065", + "pull_request:5083", + "pull_request:5085", + "pull_request:509", + "pull_request:5097", + "pull_request:5144", + "pull_request:5180", + "pull_request:5186", + "pull_request:5201", + "pull_request:5207", + "pull_request:5217", + "pull_request:5250", + "pull_request:5298", + "pull_request:5303", + "pull_request:531", + "pull_request:5332", + "pull_request:5505", + "pull_request:5679", + "pull_request:5740", + "pull_request:5818", + "pull_request:5952", + "pull_request:5961", + "pull_request:5986", + "pull_request:5996", + "pull_request:6018", + "pull_request:6433", + "pull_request:6511", + "pull_request:6569", + "pull_request:6604", + "pull_request:6620", + "pull_request:6647", + "pull_request:6998", + "pull_request:7050" + ], + "soft_pairs": [ + "pull_request:4684|pull_request:5952", + "pull_request:4696|pull_request:5207", + "pull_request:4717|pull_request:5144", + "pull_request:4740|pull_request:6018", + "pull_request:4748|pull_request:5996", + "pull_request:4796|pull_request:5303", + "pull_request:4797|pull_request:5201", + "pull_request:4822|pull_request:5217", + "pull_request:4856|pull_request:5986", + "pull_request:4856|pull_request:6604", + "pull_request:4869|pull_request:5818", + "pull_request:4934|pull_request:6569", + "pull_request:4946|pull_request:5250", + "pull_request:4952|pull_request:5180", + "pull_request:5013|pull_request:6647", + "pull_request:5051|pull_request:6620", + "pull_request:5051|pull_request:6998", + "pull_request:5051|pull_request:7050", + "pull_request:5065|pull_request:5961", + "pull_request:5083|pull_request:6433", + "pull_request:5085|pull_request:5186", + "pull_request:509|pull_request:531", + "pull_request:5097|pull_request:5679", + "pull_request:5144|pull_request:6604", + "pull_request:5180|pull_request:6998", + "pull_request:5207|pull_request:5740", + "pull_request:5298|pull_request:6511", + "pull_request:5332|pull_request:5505" + ], + "prepared_review_unit_hash": "f107e3fed3fd79b7e365e3042fea51f5dc84819e8661a4ee9a5be44a5ea98c9e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 95137, + "estimated_input_tokens": 23785, + "estimated_eval_tokens": 47826 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set: it mixes unrelated docs tweaks, training-example updates, workflow/config changes, and several distinct code fixes. The only similarity is superficial (small typo/docstring-style edits or shared broad files like `setup.py` / `loaders.py`), not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No single PR cleanly represents the cluster, because the items do not describe one concrete bug or feature. The closest matches are still materially different in intent and code path, so there is no good canonical PR for duplicate triage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4684", + "right": "pull_request:5952", + "accept": false, + "reason": "Both are docs-related, but one is a Korean translation sitemap update while the other adds a pipeline list entry for Kandinsky 3; unrelated content and purpose." + }, + { + "left": "pull_request:4696", + "right": "pull_request:5207", + "accept": false, + "reason": "Different changes entirely: T2I-Adapter-XL model support vs correcting a filename in the T2I adapter README." + }, + { + "left": "pull_request:4717", + "right": "pull_request:5144", + "accept": false, + "reason": "One fixes a missing `` in ControlNet docs; the other corrects a sample code resize call in BLIP Diffusion docs. Different documents and issues." + }, + { + "left": "pull_request:4740", + "right": "pull_request:6018", + "accept": false, + "reason": "Both touch setup/docs around packaging, but one pins `safetensors` for a test sanity check while the other is a throwaway doc-builder test edit; not the same change." + }, + { + "left": "pull_request:4748", + "right": "pull_request:5996", + "accept": false, + "reason": "Both edit `setup.py`, but one changes dependency/version-checking behavior and the other edits package metadata/cleanup text. Same file, different intent." + }, + { + "left": "pull_request:4796", + "right": "pull_request:5303", + "accept": false, + "reason": "Different targets: SDXL pipeline dtype fixes versus DeepFloyd IF docstring cleanup. No shared bug or code path." + }, + { + "left": "pull_request:4797", + "right": "pull_request:5201", + "accept": false, + "reason": "Both are LoRA-related, but one fixes SDXL LoRA block mapping while the other changes xformers/LoRA processor removal behavior across multiple modules. Distinct fixes." + }, + { + "left": "pull_request:4822", + "right": "pull_request:5217", + "accept": false, + "reason": "Both touch `loaders.py`, but one fixes a user-facing LoRA warning string and the other fixes a PEFT import typo. Separate problems." + }, + { + "left": "pull_request:4856", + "right": "pull_request:5986", + "accept": false, + "reason": "One fixes an internal cross-doc link in SDXL API docs; the other changes a video URL in SVD docs. Same broad docs category only." + }, + { + "left": "pull_request:4856", + "right": "pull_request:6604", + "accept": false, + "reason": "One is a link correction in SDXL docs; the other adds missing autodoc members for AutoencoderKL. Different documentation tasks." + }, + { + "left": "pull_request:4869", + "right": "pull_request:5818", + "accept": false, + "reason": "Combined pipeline tests are unrelated to a workflow PR that restructures PR test matrix setup. Different subsystems and goals." + }, + { + "left": "pull_request:4934", + "right": "pull_request:6569", + "accept": false, + "reason": "T2I adapter training script/docs versus modularizing InstructPix2Pix SDXL validation during training; not the same feature or code path." + }, + { + "left": "pull_request:4946", + "right": "pull_request:5250", + "accept": false, + "reason": "Both update training docs indexes, but one adds T2I-Adapters and the other adds DDPO. Distinct training topics." + }, + { + "left": "pull_request:4952", + "right": "pull_request:5180", + "accept": false, + "reason": "One fixes a pipeline offload docstring; the other repairs Wuerstchen docs citation formatting. No shared underlying issue." + }, + { + "left": "pull_request:5013", + "right": "pull_request:6647", + "accept": false, + "reason": "Both relate to SDXL testing/dtype behavior, but one fixes image-latent dtype handling in InstructPix2Pix SDXL and the other only updates a test expectation for k-diffusion outputs. Different code paths." + }, + { + "left": "pull_request:5051", + "right": "pull_request:6620", + "accept": false, + "reason": "Wuerstchen docs typo fixes are unrelated to a script-argument example change in the SDXL LoRA conversion script." + }, + { + "left": "pull_request:5051", + "right": "pull_request:6998", + "accept": false, + "reason": "Wuerstchen docs typo cleanup versus a ControlNet/IP-Adapter callout formatting fix; different docs and content." + }, + { + "left": "pull_request:5051", + "right": "pull_request:7050", + "accept": false, + "reason": "Wuerstchen docs typos and text-to-image example typos are both doc fixes, but they are unrelated files and edits." + }, + { + "left": "pull_request:5065", + "right": "pull_request:5961", + "accept": false, + "reason": "The first is a large Wuerstchen-related code/docs/model PR; the second is a tiny PixArt docstring default-value correction. No overlap in the underlying change." + }, + { + "left": "pull_request:5083", + "right": "pull_request:6433", + "accept": false, + "reason": "Issue-template guidance for community pipelines is unrelated to adding the Gluegen community pipeline itself." + }, + { + "left": "pull_request:5085", + "right": "pull_request:5186", + "accept": false, + "reason": "LoRA offloading/import handling fixes are not the same as a FreeU proof-of-concept in UNet blocks; different mechanisms and scope." + }, + { + "left": "pull_request:509", + "right": "pull_request:531", + "accept": false, + "reason": "Both are test helpers, but one adds image-loading utilities for SD tests while the other fixes numeric expectations on GPU. Different test failures." + }, + { + "left": "pull_request:5097", + "right": "pull_request:5679", + "accept": false, + "reason": "Both are test tweaks for xformers/determinism, but one adds float16 inference coverage and the other revises attn-processor determinism assertions. Not the same bug." + }, + { + "left": "pull_request:5144", + "right": "pull_request:6604", + "accept": false, + "reason": "BLIP Diffusion sample-code parenthesis fix versus AutoencoderKL API docs additions. Different artifacts and issues." + }, + { + "left": "pull_request:5180", + "right": "pull_request:6998", + "accept": false, + "reason": "Wuerstchen citation/doc builder cleanup does not match an IP-Adapter callout formatting correction." + }, + { + "left": "pull_request:5207", + "right": "pull_request:5740", + "accept": false, + "reason": "A README filename correction for T2I adapters is unrelated to a RealFill distributed-training code fix." + }, + { + "left": "pull_request:5298", + "right": "pull_request:6511", + "accept": false, + "reason": "Both involve compile/LoRA-ish training contexts, but one fixes LoRA conv forward behavior and the other makes SDXL DreamBooth checkpointing compatible with `torch.compile`. Different components." + }, + { + "left": "pull_request:5332", + "right": "pull_request:5505", + "accept": false, + "reason": "DreamBooth LoRA comment typo versus ControlNet API naming fixes; no shared bug or mergeable code change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 129, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5385", + "pull_request:5386", + "pull_request:5411", + "pull_request:5506", + "pull_request:555", + "pull_request:5552", + "pull_request:5615", + "pull_request:5626", + "pull_request:564", + "pull_request:5672", + "pull_request:5723", + "pull_request:5741", + "pull_request:5744", + "pull_request:5788", + "pull_request:580", + "pull_request:5814", + "pull_request:5818", + "pull_request:5823", + "pull_request:5828", + "pull_request:5842", + "pull_request:5898", + "pull_request:5901", + "pull_request:5922", + "pull_request:5961", + "pull_request:5983", + "pull_request:5986", + "pull_request:6140", + "pull_request:6184", + "pull_request:6209", + "pull_request:6244", + "pull_request:6293", + "pull_request:6308", + "pull_request:6314", + "pull_request:6389", + "pull_request:6437", + "pull_request:6588", + "pull_request:6808", + "pull_request:6875", + "pull_request:6947", + "pull_request:7015", + "pull_request:7050", + "pull_request:7103", + "pull_request:7131", + "pull_request:7384", + "pull_request:7564", + "pull_request:7827", + "pull_request:7857", + "pull_request:936" + ], + "soft_pairs": [ + "pull_request:5385|pull_request:5552", + "pull_request:5386|pull_request:7050", + "pull_request:5386|pull_request:7131", + "pull_request:5411|pull_request:7015", + "pull_request:5506|pull_request:6437", + "pull_request:555|pull_request:936", + "pull_request:5615|pull_request:6947", + "pull_request:5626|pull_request:7564", + "pull_request:564|pull_request:580", + "pull_request:5672|pull_request:5842", + "pull_request:5723|pull_request:7384", + "pull_request:5741|pull_request:6308", + "pull_request:5744|pull_request:5898", + "pull_request:5788|pull_request:6184", + "pull_request:5814|pull_request:7857", + "pull_request:5818|pull_request:6588", + "pull_request:5823|pull_request:6437", + "pull_request:5828|pull_request:6209", + "pull_request:5828|pull_request:6808", + "pull_request:5901|pull_request:6140", + "pull_request:5901|pull_request:6244", + "pull_request:5901|pull_request:6293", + "pull_request:5901|pull_request:6875", + "pull_request:5922|pull_request:7103", + "pull_request:5961|pull_request:7827", + "pull_request:5983|pull_request:6389", + "pull_request:5986|pull_request:6314" + ], + "prepared_review_unit_hash": "987de50d6d73b2edc0c06fe1eed1b5bc9af676bfb62f308a4bef8bf3d1dfdb04", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 99887, + "estimated_input_tokens": 24972, + "estimated_eval_tokens": 50200 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs typo cleanups, scheduler/model fixes, and distinct feature work. None of the soft-paired PRs look like the same underlying bug/change, so all soft edges should be rejected.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a good canonical representative because the cluster spans multiple unrelated themes (docs typos, IP-Adapter, PixArt, LoRA, ControlNet, tests, and schedulers).", + "best_issue_reason": null, + "best_pr_reason": "No PR is globally suitable as a duplicate canonical: the items are too diverse and do not converge on one concrete code-path or change request.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5385", + "right": "pull_request:5552", + "accept": false, + "reason": "Different fixes in different areas: copy-signature CI utility vs FreeU disable logic. No shared underlying bug." + }, + { + "left": "pull_request:5386", + "right": "pull_request:7050", + "accept": false, + "reason": "Both are typo-related, but they fix different docs/examples in different files and do not address the same issue." + }, + { + "left": "pull_request:5386", + "right": "pull_request:7131", + "accept": false, + "reason": "Separate docs cleanup PRs with different file sets and different edits; not the same concrete change." + }, + { + "left": "pull_request:5411", + "right": "pull_request:7015", + "accept": false, + "reason": "Broken training doc link fix vs from_single_file example/doc update. Different problems." + }, + { + "left": "pull_request:5506", + "right": "pull_request:6437", + "accept": false, + "reason": "PEFT/xformers runtime fix vs CI job exclusion change. One is code behavior, the other is test workflow." + }, + { + "left": "pull_request:555", + "right": "pull_request:936", + "accept": false, + "reason": "Flax VAE implementation addition vs splitting Stable Diffusion tests. Entirely different work." + }, + { + "left": "pull_request:5615", + "right": "pull_request:6947", + "accept": false, + "reason": "SDXL adapter LoRA revert vs T2I-Adapter model card standardization. Unrelated." + }, + { + "left": "pull_request:5626", + "right": "pull_request:7564", + "accept": false, + "reason": "Both touch tests, but they fix different test failures in different subsystems and are not one underlying bug." + }, + { + "left": "pull_request:564", + "right": "pull_request:580", + "accept": false, + "reason": "Related Flax scheduler area, but one is a scheduler fix and the other is initial scheduler test scaffolding; not the same change." + }, + { + "left": "pull_request:5672", + "right": "pull_request:5842", + "accept": false, + "reason": "Both concern PixArt, but one adds non-square image support and the other fixes a 512px resolution issue. Different bugs." + }, + { + "left": "pull_request:5723", + "right": "pull_request:7384", + "accept": false, + "reason": "Unrelated docs additions: LCM docs vs AnimateDiff FreeInit docs." + }, + { + "left": "pull_request:5741", + "right": "pull_request:6308", + "accept": false, + "reason": "LoRA unload speedup vs dummy-component test fix. Different problems." + }, + { + "left": "pull_request:5744", + "right": "pull_request:5898", + "accept": false, + "reason": "General API docs typo cleanup vs a specific unconditional image generation space/link wording fix. Not the same issue." + }, + { + "left": "pull_request:5788", + "right": "pull_request:6184", + "accept": false, + "reason": "LCM scheduler error handling vs LoRA offline weight lookup. Different code paths and bugs." + }, + { + "left": "pull_request:5814", + "right": "pull_request:7857", + "accept": false, + "reason": "PixArt docs about 8-bit inference vs PixArt-Sigma docs/pipeline addition. Different documentation/features." + }, + { + "left": "pull_request:5818", + "right": "pull_request:6588", + "accept": false, + "reason": "Test workflow update vs SDXL k-diffusion slow-test expectation update. Not the same bug." + }, + { + "left": "pull_request:5823", + "right": "pull_request:6437", + "accept": false, + "reason": "Feature work adding adapter_names to fuse_lora vs CI exclusion for PEFT tests. Different goals." + }, + { + "left": "pull_request:5828", + "right": "pull_request:6209", + "accept": false, + "reason": "Consistency decoder memory test fix vs text-to-image README script fix. No shared underlying issue." + }, + { + "left": "pull_request:5828", + "right": "pull_request:6808", + "accept": false, + "reason": "Different test-related fixes: one is a VAE test, the other MPS collection handling." + }, + { + "left": "pull_request:5901", + "right": "pull_request:6140", + "accept": false, + "reason": "Code feature for ControlNet Img2Img IP-Adapter vs standalone IP-Adapter API docs. Docs and implementation are not duplicates." + }, + { + "left": "pull_request:5901", + "right": "pull_request:6244", + "accept": false, + "reason": "Both involve IP-Adapter, but one integrates it into ControlNet Img2Img while the other refactors embeddings across models. Not one concrete fix." + }, + { + "left": "pull_request:5901", + "right": "pull_request:6293", + "accept": false, + "reason": "Similar feature family, but SD v1 ControlNet Img2Img vs SDXL ControlNet Img2Img are distinct pipelines and changes." + }, + { + "left": "pull_request:5901", + "right": "pull_request:6875", + "accept": false, + "reason": "Implementation PR vs test expectation update for IP-Adapter. Different artifacts and not mergeable as one change." + }, + { + "left": "pull_request:5922", + "right": "pull_request:7103", + "accept": false, + "reason": "MultiAdapter inpainting fix vs ControlNet input validation fix. Different bug reports." + }, + { + "left": "pull_request:5961", + "right": "pull_request:7827", + "accept": false, + "reason": "PixArt docstring typo vs ControlNet CPU offload docstring fix. Different files and issues." + }, + { + "left": "pull_request:5983", + "right": "pull_request:6389", + "accept": false, + "reason": "SVD docs cleanup vs Stable Video Diffusion parameter type fix. Different change types and code paths." + }, + { + "left": "pull_request:5986", + "right": "pull_request:6314", + "accept": false, + "reason": "Broken video link fix vs T2I-Adapter paper link typo. Unrelated docs issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 130, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5986", + "pull_request:6021", + "pull_request:6085", + "pull_request:6106", + "pull_request:6122", + "pull_request:6140", + "pull_request:6184", + "pull_request:62", + "pull_request:6209", + "pull_request:6244", + "pull_request:6245", + "pull_request:6247", + "pull_request:6282", + "pull_request:6293", + "pull_request:6305", + "pull_request:6337", + "pull_request:6462", + "pull_request:6528", + "pull_request:6592", + "pull_request:6604", + "pull_request:6670", + "pull_request:6680", + "pull_request:6760", + "pull_request:6775", + "pull_request:6787", + "pull_request:6795", + "pull_request:6808", + "pull_request:6837", + "pull_request:6871", + "pull_request:6875", + "pull_request:6912", + "pull_request:6997", + "pull_request:7068", + "pull_request:7103", + "pull_request:7156", + "pull_request:7200", + "pull_request:7291", + "pull_request:7325", + "pull_request:7425", + "pull_request:7719", + "pull_request:7747", + "pull_request:7868", + "pull_request:7937", + "pull_request:8470", + "pull_request:8675", + "pull_request:869", + "pull_request:9070" + ], + "soft_pairs": [ + "pull_request:5986|pull_request:6462", + "pull_request:5986|pull_request:7937", + "pull_request:6021|pull_request:6305", + "pull_request:6021|pull_request:6670", + "pull_request:6085|pull_request:7103", + "pull_request:6106|pull_request:6592", + "pull_request:6122|pull_request:7325", + "pull_request:6140|pull_request:6760", + "pull_request:6184|pull_request:7200", + "pull_request:62|pull_request:869", + "pull_request:6209|pull_request:7747", + "pull_request:6209|pull_request:8470", + "pull_request:6244|pull_request:6760", + "pull_request:6245|pull_request:6247", + "pull_request:6282|pull_request:6592", + "pull_request:6293|pull_request:6528", + "pull_request:6293|pull_request:6760", + "pull_request:6337|pull_request:6912", + "pull_request:6604|pull_request:6837", + "pull_request:6604|pull_request:7425", + "pull_request:6680|pull_request:7719", + "pull_request:6760|pull_request:6875", + "pull_request:6775|pull_request:8675", + "pull_request:6787|pull_request:7156", + "pull_request:6795|pull_request:6997", + "pull_request:6808|pull_request:7747", + "pull_request:6871|pull_request:9070", + "pull_request:6875|pull_request:7868", + "pull_request:7068|pull_request:7291" + ], + "prepared_review_unit_hash": "257c05f0cecc83bcf09920754e4ac763e2a80b677688ecef233663f325132b86", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 95478, + "estimated_input_tokens": 23870, + "estimated_eval_tokens": 47996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated one-off docs, examples, tests, or small bug fixes. I don\u2019t see any true duplicate PRs among the soft pairs; several are in the same subsystem, but they address different concrete problems or files.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 6085 is a clear, substantive bug fix in core scheduler behavior with concrete code-path changes and broad impact; it is the strongest representative of the set.", + "best_issue_reason": null, + "best_pr_reason": "PR 6085 is the best standalone PR in this cluster because it fixes an actual runtime bug in `add_noise` across multiple scheduler implementations, rather than a docs/test-only or narrow example change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5986", + "right": "pull_request:6462", + "accept": false, + "reason": "Both are docs fixes, but one updates a video link and the other corrects a paper reference for aMUSEd; different content and no shared bug." + }, + { + "left": "pull_request:5986", + "right": "pull_request:7937", + "accept": false, + "reason": "Unrelated documentation changes: a broken SVD video link versus adding a VideoProcessor docs header." + }, + { + "left": "pull_request:6021", + "right": "pull_request:6305", + "accept": false, + "reason": "Both touch training/config args, but one adds `variant` to a DreamBooth LoRA script and the other adds zero-SNR rescaling to DDPMScheduler; different code paths." + }, + { + "left": "pull_request:6021", + "right": "pull_request:6670", + "accept": false, + "reason": "Both involve `variant`, but one is a training script arg and the other updates checkpoint_merger loading; not the same underlying change." + }, + { + "left": "pull_request:6085", + "right": "pull_request:7103", + "accept": false, + "reason": "Different bugs: scheduler `add_noise` step indexing versus ControlNet image validation logic." + }, + { + "left": "pull_request:6106", + "right": "pull_request:6592", + "accept": false, + "reason": "One fixes a generated example in a training script; the other is a new inpainting training script WIP. No duplicate bug/change." + }, + { + "left": "pull_request:6122", + "right": "pull_request:7325", + "accept": false, + "reason": "Both are typo/docs cleanup, but they affect different files and unrelated content areas." + }, + { + "left": "pull_request:6140", + "right": "pull_request:6760", + "accept": false, + "reason": "Adding IP-Adapter API docs is unrelated to updating slow-test expectations for IP-Adapter." + }, + { + "left": "pull_request:6184", + "right": "pull_request:7200", + "accept": false, + "reason": "LoRA offline weight-name handling is unrelated to Transformer2D layernorm/patch-input validation." + }, + { + "left": "pull_request:62", + "right": "pull_request:869", + "accept": false, + "reason": "Very different changes: latent diffusion pipeline behavior versus community README/import bug fixes." + }, + { + "left": "pull_request:6209", + "right": "pull_request:7747", + "accept": false, + "reason": "One fixes a README test script; the other adjusts a VAE tiling test assertion. Same docs/tests area, but not the same bug." + }, + { + "left": "pull_request:6209", + "right": "pull_request:8470", + "accept": false, + "reason": "Example README fix versus research-project README rewrite; unrelated documentation changes." + }, + { + "left": "pull_request:6244", + "right": "pull_request:6760", + "accept": false, + "reason": "IP-Adapter refactor in model/pipeline code versus slow-test expectation updates; the latter is test maintenance, not the same code-path bug." + }, + { + "left": "pull_request:6245", + "right": "pull_request:6247", + "accept": false, + "reason": "Same Wuerstchen area, but one fixes fp16 training/LoRA config and the other adds a dreambooth LoRA example; different changes." + }, + { + "left": "pull_request:6282", + "right": "pull_request:6592", + "accept": false, + "reason": "Dreambooth runtime-error cleanup is unrelated to adding a new inpainting training script." + }, + { + "left": "pull_request:6293", + "right": "pull_request:6528", + "accept": false, + "reason": "Both IP-Adapter-related, but one adds IP-Adapter support to SDXL ControlNet Img2Img while the other fixes missing code in ControlNet img2img. Different pipelines and bugs." + }, + { + "left": "pull_request:6293", + "right": "pull_request:6760", + "accept": false, + "reason": "Feature addition versus slow-test threshold adjustment; not the same underlying issue." + }, + { + "left": "pull_request:6337", + "right": "pull_request:6912", + "accept": false, + "reason": "One fixes `retrieve_latents` usage in a TensorRT img2img example; the other fixes truthiness checks for `denoising_start` across several pipelines." + }, + { + "left": "pull_request:6604", + "right": "pull_request:6837", + "accept": false, + "reason": "Missing API docs for AutoencoderKL versus docs-only dependency order tweaks in I2VGenXL; unrelated." + }, + { + "left": "pull_request:6604", + "right": "pull_request:7425", + "accept": false, + "reason": "AutoencoderKL API docs versus a PEFT inference tutorial fix; different documentation targets." + }, + { + "left": "pull_request:6680", + "right": "pull_request:7719", + "accept": false, + "reason": "Slow test failure fix for DPM scheduler versus adding a Tailscale auth key to workflow failure handling; unrelated." + }, + { + "left": "pull_request:6760", + "right": "pull_request:6875", + "accept": false, + "reason": "Both update IP-Adapter tests, but they cover different expected outputs and scenarios; not a single duplicate bug." + }, + { + "left": "pull_request:6775", + "right": "pull_request:8675", + "accept": false, + "reason": "A DDPMScheduler parameter docs change is unrelated to a fast-diffusion note/caching docs update." + }, + { + "left": "pull_request:6787", + "right": "pull_request:7156", + "accept": false, + "reason": "Marigold input normalization bug fix versus stalebot label handling; completely different code paths." + }, + { + "left": "pull_request:6795", + "right": "pull_request:6997", + "accept": false, + "reason": "DreamBooth SDXL micro-conditioning vs adding LoRA support to UNetSpatioTemporalConditionModel; different features." + }, + { + "left": "pull_request:6808", + "right": "pull_request:7747", + "accept": false, + "reason": "MPS test-collection fix in testing utilities versus a VAE tiling assertion fix; unrelated test issues." + }, + { + "left": "pull_request:6871", + "right": "pull_request:9070", + "accept": false, + "reason": "Adding VAE tiling/slicing to SD img2img/inpaint is unrelated to a new FLUX img2img pipeline." + }, + { + "left": "pull_request:6875", + "right": "pull_request:7868", + "accept": false, + "reason": "IP-Adapter output expectation updates versus offload test assertions; different test behaviors and failures." + }, + { + "left": "pull_request:6912", + "right": "pull_request:7068", + "accept": false, + "reason": "Different bug classes: denoising_start truthiness versus typo/doc cleanup across loaders and pipeline examples." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 131, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:7120", + "pull_request:7150", + "pull_request:7156", + "pull_request:7164", + "pull_request:7206", + "pull_request:7233", + "pull_request:7235", + "pull_request:7259", + "pull_request:7291", + "pull_request:7335", + "pull_request:7360", + "pull_request:7384", + "pull_request:7411", + "pull_request:7451", + "pull_request:7473", + "pull_request:7495", + "pull_request:7530", + "pull_request:7542", + "pull_request:7564", + "pull_request:7590", + "pull_request:7594", + "pull_request:7632", + "pull_request:7747", + "pull_request:7767", + "pull_request:7832", + "pull_request:7903", + "pull_request:7916", + "pull_request:7966", + "pull_request:7977", + "pull_request:8064", + "pull_request:810", + "pull_request:822", + "pull_request:8272", + "pull_request:8470", + "pull_request:8498", + "pull_request:8702", + "pull_request:8720", + "pull_request:8908", + "pull_request:8926", + "pull_request:9021", + "pull_request:9060", + "pull_request:9077", + "pull_request:910", + "pull_request:9242", + "pull_request:9318", + "pull_request:9627", + "pull_request:9683" + ], + "soft_pairs": [ + "pull_request:7120|pull_request:7360", + "pull_request:7150|pull_request:7594", + "pull_request:7156|pull_request:9021", + "pull_request:7164|pull_request:7564", + "pull_request:7206|pull_request:7235", + "pull_request:7233|pull_request:8702", + "pull_request:7259|pull_request:7384", + "pull_request:7291|pull_request:7451", + "pull_request:7335|pull_request:8908", + "pull_request:7411|pull_request:9077", + "pull_request:7473|pull_request:8272", + "pull_request:7495|pull_request:8470", + "pull_request:7530|pull_request:7903", + "pull_request:7530|pull_request:8064", + "pull_request:7542|pull_request:7977", + "pull_request:7542|pull_request:8498", + "pull_request:7564|pull_request:7966", + "pull_request:7564|pull_request:8720", + "pull_request:7590|pull_request:9060", + "pull_request:7632|pull_request:8926", + "pull_request:7747|pull_request:9627", + "pull_request:7747|pull_request:9683", + "pull_request:7767|pull_request:8498", + "pull_request:7832|pull_request:9318", + "pull_request:7916|pull_request:9242", + "pull_request:7966|pull_request:8720", + "pull_request:810|pull_request:822", + "pull_request:810|pull_request:910" + ], + "prepared_review_unit_hash": "fc67326e7adbed62c7f514f44e0320a730bc17cc4b879239693e9dbb0fb7e3a3", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 98637, + "estimated_input_tokens": 24660, + "estimated_eval_tokens": 49576 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "A heterogeneous set of unrelated PRs (docs typos, tests, CI, scheduler additions, training feature work, and bug fixes). The soft pairs mostly share superficial keywords or filenames but do not fix the same concrete problem, so none should be clustered as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits this cluster: the items are not one underlying bug/change but many independent PRs across docs, tests, CI, and features.", + "best_issue_reason": null, + "best_pr_reason": "PR 7530 is the strongest standalone representative because it fixes a concrete runtime bug, has an explicit issue target, and is substantially more substantive than the docs/test-only changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7120", + "right": "pull_request:7360", + "accept": false, + "reason": "Different scheduler work: one adds an EDM DPM solver scheduler, the other tweaks TCD scheduler docs/signatures. Same broad area, not the same change." + }, + { + "left": "pull_request:7150", + "right": "pull_request:7594", + "accept": false, + "reason": "Both touch callback docs, but one updates a code example and the other fixes a separate example bug. Superficial overlap only." + }, + { + "left": "pull_request:7156", + "right": "pull_request:9021", + "accept": false, + "reason": "Unrelated changes: stalebot label handling vs SD3 ControlNet layer-loading logic." + }, + { + "left": "pull_request:7164", + "right": "pull_request:7564", + "accept": false, + "reason": "CI Dockerfile pinning vs pipeline scheduler deprecation handling; no shared bug or code path." + }, + { + "left": "pull_request:7206", + "right": "pull_request:7235", + "accept": false, + "reason": "Both are training-script related, but one adds tags across many scripts while the other adds DoRA support to one script. Different goals and diffs." + }, + { + "left": "pull_request:7233", + "right": "pull_request:8702", + "accept": false, + "reason": "DoRA training option in DreamBooth LoRA vs SD3 checkpoint conversion script. Completely different code paths." + }, + { + "left": "pull_request:7259", + "right": "pull_request:7384", + "accept": false, + "reason": "Separate docs additions: TCD intro page vs FreeInit outputs in AnimateDiff docs." + }, + { + "left": "pull_request:7291", + "right": "pull_request:7451", + "accept": false, + "reason": "A documentation default-value typo fix is unrelated to broad docs typo cleanup across many files." + }, + { + "left": "pull_request:7335", + "right": "pull_request:8908", + "accept": false, + "reason": "ControlNet SDXL config test/docs ordering fix vs a request-caching skip annotation. Different test failures." + }, + { + "left": "pull_request:7411", + "right": "pull_request:9077", + "accept": false, + "reason": "One is a general typo cleanup across schedulers/docs, the other mixes unrelated typo fixes across several files; no single underlying bug." + }, + { + "left": "pull_request:7473", + "right": "pull_request:8272", + "accept": false, + "reason": "Both fix incorrect `raise` statements, but they target different files and different input-validation sites; not the same concrete bug." + }, + { + "left": "pull_request:7495", + "right": "pull_request:8470", + "accept": false, + "reason": "README link cleanup vs replacing a research-project README with a different project description. Not the same change." + }, + { + "left": "pull_request:7530", + "right": "pull_request:7903", + "accept": false, + "reason": "Autocast behavior in training scripts vs a video processor input-validation test harness. Different subsystems and bugs." + }, + { + "left": "pull_request:7530", + "right": "pull_request:8064", + "accept": false, + "reason": "Autocast bug fix vs Marigold pipeline feature/test addition. No shared code path." + }, + { + "left": "pull_request:7542", + "right": "pull_request:7977", + "accept": false, + "reason": "CPU offload test typo vs docs typo in an unrelated optimization guide." + }, + { + "left": "pull_request:7542", + "right": "pull_request:8498", + "accept": false, + "reason": "Test typo fix vs a docs string/prompt typo fix. No underlying duplicate." + }, + { + "left": "pull_request:7564", + "right": "pull_request:7966", + "accept": false, + "reason": "Scheduler deprecation/test updates vs PixArt Sigma test expectation updates. Both are test-related but different failures." + }, + { + "left": "pull_request:7564", + "right": "pull_request:8720", + "accept": false, + "reason": "Pipeline deprecation/test adjustments vs precision-threshold updates in slow pipeline tests. Not the same bug." + }, + { + "left": "pull_request:7590", + "right": "pull_request:9060", + "accept": false, + "reason": "Python packaging version-range change is unrelated to removing a debug print in a model forward pass." + }, + { + "left": "pull_request:7632", + "right": "pull_request:8926", + "accept": false, + "reason": "Both concern DreamBooth LoRA scripts, but one fixes VAE path loading and the other fixes checkpoint saving for advanced SDXL LoRA. Different issues." + }, + { + "left": "pull_request:7747", + "right": "pull_request:9627", + "accept": false, + "reason": "VAE tiling test fix vs DoRA LoRA warning test ordering. Different test cases and behaviors." + }, + { + "left": "pull_request:7747", + "right": "pull_request:9683", + "accept": false, + "reason": "VAE tiling test fix vs CogVideoX integration test rename/unskip. Not duplicates." + }, + { + "left": "pull_request:7767", + "right": "pull_request:8498", + "accept": false, + "reason": "PixArt pipeline refactor/removal vs a single docs typo fix. No shared bug/change." + }, + { + "left": "pull_request:7832", + "right": "pull_request:9318", + "accept": false, + "reason": "CI cache mount path update vs nightly hub token secret rename. Both CI-only, but different infra problems." + }, + { + "left": "pull_request:7916", + "right": "pull_request:9242", + "accept": false, + "reason": "Ascend NPU support in SDXL training is unrelated to xpu test backend support." + }, + { + "left": "pull_request:7966", + "right": "pull_request:8720", + "accept": false, + "reason": "PixArt Sigma test expectation changes vs broader precision-related slow test fixes. Similar test domain, but not the same concrete change." + }, + { + "left": "pull_request:810", + "right": "pull_request:822", + "accept": false, + "reason": "These are related to the same Flax scheduler API transition, but they address different stages of the update and different files; not the same duplicate change." + }, + { + "left": "pull_request:810", + "right": "pull_request:910", + "accept": false, + "reason": "Flax scheduler API adaptation vs initial inpainting docs update. Completely unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 132, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10122", + "pull_request:1527", + "pull_request:3298", + "pull_request:4847", + "pull_request:6752", + "pull_request:828", + "pull_request:8290", + "pull_request:832", + "pull_request:8422", + "pull_request:8496", + "pull_request:8583", + "pull_request:8600", + "pull_request:8627", + "pull_request:864", + "pull_request:8696", + "pull_request:8755", + "pull_request:8838", + "pull_request:8851", + "pull_request:889", + "pull_request:8913", + "pull_request:8926", + "pull_request:9005", + "pull_request:9021", + "pull_request:9060", + "pull_request:9070", + "pull_request:9196", + "pull_request:9249", + "pull_request:9251", + "pull_request:9257", + "pull_request:9282", + "pull_request:9286", + "pull_request:9307", + "pull_request:9355", + "pull_request:936", + "pull_request:9379", + "pull_request:9385", + "pull_request:9442", + "pull_request:957", + "pull_request:9571", + "pull_request:9598", + "pull_request:9620", + "pull_request:9658", + "pull_request:9699", + "pull_request:9754", + "pull_request:9832", + "pull_request:9838", + "pull_request:9860", + "pull_request:9980" + ], + "soft_pairs": [ + "pull_request:828|pull_request:864", + "pull_request:8290|pull_request:9307", + "pull_request:832|pull_request:957", + "pull_request:8422|pull_request:8696", + "pull_request:8496|pull_request:8755", + "pull_request:8583|pull_request:8838", + "pull_request:8600|pull_request:9980", + "pull_request:8627|pull_request:9251", + "pull_request:8838|pull_request:9249", + "pull_request:8851|pull_request:9282", + "pull_request:889|pull_request:936", + "pull_request:8913|pull_request:8926", + "pull_request:9005|pull_request:9699", + "pull_request:9021|pull_request:9838", + "pull_request:9060|pull_request:9286", + "pull_request:9070|pull_request:9754", + "pull_request:9196|pull_request:9598", + "pull_request:9355|pull_request:9832", + "pull_request:9379|pull_request:9442", + "pull_request:9385|pull_request:9571", + "pull_request:9385|pull_request:9620", + "pull_request:9385|pull_request:9658", + "pull_request:9385|pull_request:9860", + "pull_request:4847|pull_request:6752", + "pull_request:10122|pull_request:9257", + "pull_request:1527|pull_request:3298" + ], + "prepared_review_unit_hash": "67867e152aeaf8c6aac153e0f03e69360cb10c3d5653f5726370a53b85eec2ce", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 96820, + "estimated_input_tokens": 24205, + "estimated_eval_tokens": 48666 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a mixed set of unrelated PRs across Flux, SD3, ControlNet, CI, docs, and tests. The soft edges are only broad topical similarities or shared filenames, not the same concrete bug/change, so they should all be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents this cluster; the items are independent changes in different subsystems, so there is no meaningful canonical PR.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR stands out because the cluster is not one duplicate family; picking any single PR would misrepresent the rest.", + "soft_edge_verdicts": [ + { + "left": "pull_request:828", + "right": "pull_request:864", + "accept": false, + "reason": "Different Flax test/scheduler changes; not the same bug or fix." + }, + { + "left": "pull_request:8290", + "right": "pull_request:9307", + "accept": false, + "reason": "Shared embeddings file, but unrelated HunyuanDIT refactor vs rotary-embedding fix." + }, + { + "left": "pull_request:832", + "right": "pull_request:957", + "accept": false, + "reason": "Unrelated bugs in Flax nsfw handling vs text-to-image scheduler args." + }, + { + "left": "pull_request:8422", + "right": "pull_request:8696", + "accept": false, + "reason": "Subfolder-loading tests vs Marigold flaky-test annotations; different concerns." + }, + { + "left": "pull_request:8496", + "right": "pull_request:8755", + "accept": false, + "reason": "Transformer warning-log tweak vs SD3 dreambooth training bug; unrelated." + }, + { + "left": "pull_request:8583", + "right": "pull_request:8838", + "accept": false, + "reason": "Both touch SD3 transformer, but one adds XFormers support and the other forward chunking." + }, + { + "left": "pull_request:8600", + "right": "pull_request:9980", + "accept": false, + "reason": "Dreambooth script cleanup vs LoRA conversion mapping; different code paths." + }, + { + "left": "pull_request:8627", + "right": "pull_request:9251", + "accept": false, + "reason": "SD3 ControlNet docs/code vs CI trigger change; unrelated." + }, + { + "left": "pull_request:8838", + "right": "pull_request:9249", + "accept": false, + "reason": "SD3 transformer utility vs SD3 ControlNet example; not the same change." + }, + { + "left": "pull_request:8851", + "right": "pull_request:9282", + "accept": false, + "reason": "AuraFlow docs vs pipeline table update; different doc changes." + }, + { + "left": "pull_request:889", + "right": "pull_request:936", + "accept": false, + "reason": "Both touch pipeline utilities/tests, but address different features and bugs." + }, + { + "left": "pull_request:8913", + "right": "pull_request:8926", + "accept": false, + "reason": "Checkpoint loading fix vs checkpoint saving fix; different failures." + }, + { + "left": "pull_request:9005", + "right": "pull_request:9699", + "accept": false, + "reason": "Runner config refactor vs adding a new runner option; not the same issue." + }, + { + "left": "pull_request:9021", + "right": "pull_request:9838", + "accept": false, + "reason": "SD3ControlNet num_layers bug vs Flux training memory fix; unrelated." + }, + { + "left": "pull_request:9060", + "right": "pull_request:9286", + "accept": false, + "reason": "Removing a debug print vs exporting FluxMultiControlNetModel; unrelated." + }, + { + "left": "pull_request:9070", + "right": "pull_request:9754", + "accept": false, + "reason": "Flux img2img pipeline addition vs Flux quantized training example; different artifacts." + }, + { + "left": "pull_request:9196", + "right": "pull_request:9598", + "accept": false, + "reason": "PixArt attention API change vs Flux LoRA test notes; unrelated." + }, + { + "left": "pull_request:9355", + "right": "pull_request:9832", + "accept": false, + "reason": "Both are CI-related, but one adds manual dispatch and the other adds quantization tests." + }, + { + "left": "pull_request:9379", + "right": "pull_request:9442", + "accept": false, + "reason": "Fast GPU test fixes for different failures; not one concrete bug." + }, + { + "left": "pull_request:9385", + "right": "pull_request:9571", + "accept": false, + "reason": "Different controlnet implementations and different feature additions." + }, + { + "left": "pull_request:9385", + "right": "pull_request:9620", + "accept": false, + "reason": "Xlabs ControlNet support vs interrupt handling; separate problems." + }, + { + "left": "pull_request:9385", + "right": "pull_request:9658", + "accept": false, + "reason": "Xlabs ControlNet support vs Flux distilled ControlNet example; unrelated." + }, + { + "left": "pull_request:9385", + "right": "pull_request:9860", + "accept": false, + "reason": "Different model families: Xlabs Flux ControlNet vs SD3.5 ControlNet example." + }, + { + "left": "pull_request:4847", + "right": "pull_request:6752", + "accept": false, + "reason": "Prompt Free Diffusion community example vs PromptDiffusion research project; distinct projects." + }, + { + "left": "pull_request:10122", + "right": "pull_request:9257", + "accept": false, + "reason": "Flux attention-mask fix vs Flux Dreambooth training bug; different code paths." + }, + { + "left": "pull_request:1527", + "right": "pull_request:3298", + "accept": false, + "reason": "Stable Diffusion inpaint tensor input fix vs IF super-resolution aspect-ratio support; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 133, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10132", + "pull_request:10148", + "pull_request:10197", + "pull_request:10200", + "pull_request:10271", + "pull_request:10370", + "pull_request:10547", + "pull_request:10554", + "pull_request:10597", + "pull_request:10850", + "pull_request:11052", + "pull_request:1137", + "pull_request:11588", + "pull_request:11639", + "pull_request:11813", + "pull_request:11818", + "pull_request:12062", + "pull_request:1219", + "pull_request:12360", + "pull_request:12450", + "pull_request:12637", + "pull_request:12797", + "pull_request:13146", + "pull_request:1424", + "pull_request:1468", + "pull_request:1527", + "pull_request:1564", + "pull_request:159", + "pull_request:2655", + "pull_request:3042", + "pull_request:397", + "pull_request:415", + "pull_request:5188", + "pull_request:5393", + "pull_request:718", + "pull_request:7634", + "pull_request:7893", + "pull_request:8404", + "pull_request:8702", + "pull_request:8722", + "pull_request:8978", + "pull_request:9077", + "pull_request:9426", + "pull_request:9524", + "pull_request:9721", + "pull_request:9739", + "pull_request:9987" + ], + "soft_pairs": [ + "pull_request:1564|pull_request:3042", + "pull_request:5188|pull_request:5393", + "pull_request:7634|pull_request:8404", + "pull_request:7893|pull_request:9721", + "pull_request:11588|pull_request:11639", + "pull_request:10132|pull_request:9426", + "pull_request:10597|pull_request:9987", + "pull_request:10148|pull_request:8702", + "pull_request:10197|pull_request:8978", + "pull_request:10200|pull_request:8722", + "pull_request:10271|pull_request:9524", + "pull_request:10370|pull_request:10850", + "pull_request:10547|pull_request:11052", + "pull_request:10554|pull_request:9077", + "pull_request:10554|pull_request:9739", + "pull_request:1137|pull_request:1527", + "pull_request:11813|pull_request:11818", + "pull_request:12062|pull_request:12450", + "pull_request:1219|pull_request:397", + "pull_request:1219|pull_request:415", + "pull_request:12360|pull_request:12797", + "pull_request:12637|pull_request:13146", + "pull_request:1424|pull_request:397", + "pull_request:1424|pull_request:415", + "pull_request:1468|pull_request:2655", + "pull_request:159|pull_request:718" + ], + "prepared_review_unit_hash": "af236d6d74539f49c60b0d190140881db1e194641b06f01abee59bf1d88c42fe", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 103382, + "estimated_input_tokens": 25846, + "estimated_eval_tokens": 51948 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft edges look like false positives: they mostly share broad subsystems or filenames, but the underlying changes are different bugs, features, or docs/CI updates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the soft links span unrelated fixes across CI, docs, model additions, tests, and refactors, so none is a safe cluster representative.", + "best_issue_reason": null, + "best_pr_reason": "No PR in this set is globally representative of the whole cluster; the items are heterogeneous and should stay separate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1564", + "right": "pull_request:3042", + "accept": false, + "reason": "Both touch training docs, but one adds a default accelerate config note and the other raises accelerate minimum versions for example requirements; different changes." + }, + { + "left": "pull_request:5188", + "right": "pull_request:5393", + "accept": false, + "reason": "Both affect training examples, but one centralizes compute_snr and the other updates HFApi usage in tutorials/examples; not the same fix." + }, + { + "left": "pull_request:7634", + "right": "pull_request:8404", + "accept": false, + "reason": "High-res fix for SD vs adding Latte video model support; only overlap is generic model registration/docs." + }, + { + "left": "pull_request:7893", + "right": "pull_request:9721", + "accept": false, + "reason": "Both are Dockerfile environment changes, but one upgrades Python while the other pins torch versions; unrelated fixes." + }, + { + "left": "pull_request:11588", + "right": "pull_request:11639", + "accept": false, + "reason": "Same Wan area, but one expands LoRA weights for I2V and the other standardizes VAE sampling mode in video-to-video; different code paths." + }, + { + "left": "pull_request:10132", + "right": "pull_request:9426", + "accept": false, + "reason": "Both touch CI, but one installs tokenizers for a backend test job and the other updates artifact uploader versions; separate changes." + }, + { + "left": "pull_request:10597", + "right": "pull_request:9987", + "accept": false, + "reason": "Both mention SD3, but one updates a docs model ID while the other integrates SD3.5 IP-Adapter pipeline support; different functionality." + }, + { + "left": "pull_request:10148", + "right": "pull_request:8702", + "accept": false, + "reason": "OmniGen feature addition versus an SD3 conversion script; completely different artifacts and code paths." + }, + { + "left": "pull_request:10197", + "right": "pull_request:8978", + "accept": false, + "reason": "Both are broad workflow/example updates, but one changes TPU CI setup and the other updates DreamBooth/Imagic examples; unrelated." + }, + { + "left": "pull_request:10200", + "right": "pull_request:8722", + "accept": false, + "reason": "Different subsystems: ControlNet union refactor versus embeddings refactor; no shared concrete bug or change." + }, + { + "left": "pull_request:10271", + "right": "pull_request:9524", + "accept": false, + "reason": "One refactors TorchAO serialization tests; the other refactors CogVideoX VAE internals; unrelated test/model changes." + }, + { + "left": "pull_request:10370", + "right": "pull_request:10850", + "accept": false, + "reason": "TorchAO quantizer style cleanup versus hook helper utilities; different purposes and code paths." + }, + { + "left": "pull_request:10547", + "right": "pull_request:11052", + "accept": false, + "reason": "Both are test-related, but one fixes offload tests for specific models and the other adds compute-capability gating; distinct issues." + }, + { + "left": "pull_request:10554", + "right": "pull_request:9077", + "accept": false, + "reason": "SD3 LoRA example fix versus generic typo fixes; not the same underlying bug." + }, + { + "left": "pull_request:10554", + "right": "pull_request:9739", + "accept": false, + "reason": "Both touch SD3-related docs/examples, but one fixes a training script state-dict filter and the other adds single-file SD3.5 docs; different changes." + }, + { + "left": "pull_request:1137", + "right": "pull_request:1527", + "accept": false, + "reason": "Same inpaint family, but one fixes safety-checker dtype handling and the other allows tensor image/mask inputs; separate bugs." + }, + { + "left": "pull_request:11813", + "right": "pull_request:11818", + "accept": false, + "reason": "Kontext training changes versus a runtime sync optimization in Flux denoising; unrelated." + }, + { + "left": "pull_request:12062", + "right": "pull_request:12450", + "accept": false, + "reason": "Both CI-related, but one enables all GPUs in workflows and the other revisits installation steps; different operational fixes." + }, + { + "left": "pull_request:1219", + "right": "pull_request:397", + "accept": false, + "reason": "DDIM docs versus Quicktour docs; unrelated documentation." + }, + { + "left": "pull_request:1219", + "right": "pull_request:415", + "accept": false, + "reason": "DDIM docs versus training overview/docs; unrelated documentation updates." + }, + { + "left": "pull_request:12360", + "right": "pull_request:12797", + "accept": false, + "reason": "Different docs for different models: Chroma guidance wording versus Kandinsky minor fixes." + }, + { + "left": "pull_request:12637", + "right": "pull_request:13146", + "accept": false, + "reason": "Sana video transformer copy cleanup versus tokenizer type fix in PAG HunyuanDiT; unrelated code changes." + }, + { + "left": "pull_request:1424", + "right": "pull_request:397", + "accept": false, + "reason": "Stable Diffusion 2 docs versus Quicktour docs; different documentation additions." + }, + { + "left": "pull_request:1424", + "right": "pull_request:415", + "accept": false, + "reason": "Stable Diffusion 2 docs versus training overview/docs; unrelated documentation." + }, + { + "left": "pull_request:1468", + "right": "pull_request:2655", + "accept": false, + "reason": "Saving trained betas across schedulers versus LoRA save/load behavior; different config and loader problems." + }, + { + "left": "pull_request:159", + "right": "pull_request:718", + "accept": false, + "reason": "Both are typo/wording fixes, but they target different files and unrelated text changes; too broad to merge as one change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 134, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1945", + "pull_request:1967", + "pull_request:1982", + "pull_request:2138", + "pull_request:2271", + "pull_request:2322", + "pull_request:2327", + "pull_request:2345", + "pull_request:2398", + "pull_request:2553", + "pull_request:2759", + "pull_request:3003", + "pull_request:3006", + "pull_request:3168", + "pull_request:3262", + "pull_request:3292", + "pull_request:3400", + "pull_request:3536", + "pull_request:3607", + "pull_request:3637", + "pull_request:3642", + "pull_request:3653", + "pull_request:3766", + "pull_request:386", + "pull_request:388", + "pull_request:390", + "pull_request:3916", + "pull_request:3957", + "pull_request:397", + "pull_request:402", + "pull_request:404", + "pull_request:415", + "pull_request:417", + "pull_request:418", + "pull_request:4187", + "pull_request:4239", + "pull_request:453", + "pull_request:4887", + "pull_request:531", + "pull_request:555", + "pull_request:5622", + "pull_request:5754", + "pull_request:5895", + "pull_request:6405", + "pull_request:6410", + "pull_request:774", + "pull_request:909" + ], + "soft_pairs": [ + "pull_request:1945|pull_request:2327", + "pull_request:1967|pull_request:3168", + "pull_request:1982|pull_request:531", + "pull_request:2138|pull_request:4239", + "pull_request:2271|pull_request:3400", + "pull_request:2322|pull_request:3262", + "pull_request:2345|pull_request:4187", + "pull_request:2398|pull_request:4187", + "pull_request:2553|pull_request:555", + "pull_request:2759|pull_request:774", + "pull_request:2759|pull_request:909", + "pull_request:3003|pull_request:3006", + "pull_request:3292|pull_request:3536", + "pull_request:3607|pull_request:5754", + "pull_request:3637|pull_request:3916", + "pull_request:3642|pull_request:5895", + "pull_request:3653|pull_request:5622", + "pull_request:3766|pull_request:4887", + "pull_request:386|pull_request:390", + "pull_request:388|pull_request:404", + "pull_request:388|pull_request:417", + "pull_request:3957|pull_request:6405", + "pull_request:397|pull_request:415", + "pull_request:397|pull_request:418", + "pull_request:402|pull_request:453", + "pull_request:415|pull_request:418", + "pull_request:4239|pull_request:6410" + ], + "prepared_review_unit_hash": "f0dddf1d50c12fc8e0c6d8dc8105f0b1ac3db2bb829c646c468523339d41a132", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 98748, + "estimated_input_tokens": 24687, + "estimated_eval_tokens": 49630 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is highly heterogeneous: mostly unrelated docs, tests, and core code changes. I don\u2019t see any soft-edge pair that clearly reflects the same underlying bug or change, so I would reject all of them as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a clear canonical representative; the items span unrelated subsystems and change types.", + "best_issue_reason": null, + "best_pr_reason": "PR 2327 is the strongest standalone core bugfix-style change, but it is still not a duplicate representative for the rest of the set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1945", + "right": "pull_request:2327", + "accept": false, + "reason": "Tests for UNet blocks vs. a UNet input-validation fix; related area, but not the same concrete change." + }, + { + "left": "pull_request:1967", + "right": "pull_request:3168", + "accept": false, + "reason": "CPU offload example code vs. doc-builder workflow; unrelated." + }, + { + "left": "pull_request:1982", + "right": "pull_request:531", + "accept": false, + "reason": "Both are test tweaks, but they exercise different components and are not the same bug." + }, + { + "left": "pull_request:2138", + "right": "pull_request:4239", + "accept": false, + "reason": "A doc typo fix vs. a DreamBooth/text-encoder code change; unrelated." + }, + { + "left": "pull_request:2271", + "right": "pull_request:3400", + "accept": false, + "reason": "Slow-test numeric updates vs. adding a dependency; unrelated." + }, + { + "left": "pull_request:2322", + "right": "pull_request:3262", + "accept": false, + "reason": "Large example/inpainting update vs. IF model renaming; different features." + }, + { + "left": "pull_request:2345", + "right": "pull_request:4187", + "accept": false, + "reason": "xFormers warning in training scripts vs. scheduler timestep indexing; unrelated bugs." + }, + { + "left": "pull_request:2398", + "right": "pull_request:4187", + "accept": false, + "reason": "Pix2Pix Zero docs vs. DPMSolver scheduler internals; unrelated." + }, + { + "left": "pull_request:2553", + "right": "pull_request:555", + "accept": false, + "reason": "Flax CFG flag work vs. adding Flax VAE support; different changes." + }, + { + "left": "pull_request:2759", + "right": "pull_request:774", + "accept": false, + "reason": "Broad test-run/CI changes vs. a specific test fix; not the same change." + }, + { + "left": "pull_request:2759", + "right": "pull_request:909", + "accept": false, + "reason": "Broad test-run/CI changes vs. a specific inpaint test fix; not the same change." + }, + { + "left": "pull_request:3003", + "right": "pull_request:3006", + "accept": false, + "reason": "Both touch schedulers, but one is dynamic thresholding and the other is variance clamping; different bugs." + }, + { + "left": "pull_request:3292", + "right": "pull_request:3536", + "accept": false, + "reason": "Input perturbation training flag vs. IF super-resolution normalization; unrelated." + }, + { + "left": "pull_request:3607", + "right": "pull_request:5754", + "accept": false, + "reason": "Attention implementation fix vs. formatting-only cleanup; unrelated." + }, + { + "left": "pull_request:3637", + "right": "pull_request:3916", + "accept": false, + "reason": "Both are branch-sync/meta PRs with broad diffs, but not the same underlying bug or change." + }, + { + "left": "pull_request:3642", + "right": "pull_request:5895", + "accept": false, + "reason": "TensorRT inpainting pipeline vs. Stable Video Diffusion addition; different features." + }, + { + "left": "pull_request:3653", + "right": "pull_request:5622", + "accept": false, + "reason": "Lint cleanup vs. optimization/task-ops PR; unrelated." + }, + { + "left": "pull_request:3766", + "right": "pull_request:4887", + "accept": false, + "reason": "Kandinsky conversion-script changes vs. removing a prior_emb2emb latent input; related project, different code paths." + }, + { + "left": "pull_request:386", + "right": "pull_request:390", + "accept": false, + "reason": "Stable Diffusion docs vs. MPS optimization docs; different documentation topics." + }, + { + "left": "pull_request:388", + "right": "pull_request:404", + "accept": false, + "reason": "Different docs pages and scopes; not the same change." + }, + { + "left": "pull_request:388", + "right": "pull_request:417", + "accept": false, + "reason": "Both are inference docs, but they cover different pipeline guides and rewrites." + }, + { + "left": "pull_request:3957", + "right": "pull_request:6405", + "accept": false, + "reason": "Attn-name correction work vs. doc-builder workflow testing; unrelated." + }, + { + "left": "pull_request:397", + "right": "pull_request:415", + "accept": false, + "reason": "Quicktour docs vs. training docs; both docs, but not the same change." + }, + { + "left": "pull_request:397", + "right": "pull_request:418", + "accept": false, + "reason": "Quicktour docs vs. DiffusionPipeline API docs; different documentation artifacts." + }, + { + "left": "pull_request:402", + "right": "pull_request:453", + "accept": false, + "reason": "Intro-section reorg vs. a one-line installation typo fix; not the same underlying change." + }, + { + "left": "pull_request:415", + "right": "pull_request:418", + "accept": false, + "reason": "Training docs vs. DiffusionPipeline API docs; different scopes." + }, + { + "left": "pull_request:4239", + "right": "pull_request:6410", + "accept": false, + "reason": "DreamBooth text-encoder fixes vs. RL module registration; unrelated code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 135, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1038", + "pull_request:10454", + "pull_request:10482", + "pull_request:11273", + "pull_request:11433", + "pull_request:11469", + "pull_request:1147", + "pull_request:11516", + "pull_request:12021", + "pull_request:12530", + "pull_request:1527", + "pull_request:1913", + "pull_request:3160", + "pull_request:3476", + "pull_request:4294", + "pull_request:4347", + "pull_request:5552", + "pull_request:5707", + "pull_request:5916", + "pull_request:6007", + "pull_request:6140", + "pull_request:6150", + "pull_request:6215", + "pull_request:6229", + "pull_request:6484", + "pull_request:6643", + "pull_request:6754", + "pull_request:6765", + "pull_request:6766", + "pull_request:6767", + "pull_request:6768", + "pull_request:6800", + "pull_request:6864", + "pull_request:6999", + "pull_request:7247", + "pull_request:7627", + "pull_request:7632", + "pull_request:7941", + "pull_request:8322", + "pull_request:8386", + "pull_request:8478", + "pull_request:8572", + "pull_request:8696", + "pull_request:8978", + "pull_request:8990", + "pull_request:9506", + "pull_request:9601", + "pull_request:9620" + ], + "soft_pairs": [ + "pull_request:4294|pull_request:5916", + "pull_request:4347|pull_request:5707", + "pull_request:5552|pull_request:6150", + "pull_request:6007|pull_request:6864", + "pull_request:6140|pull_request:6643", + "pull_request:6215|pull_request:8386", + "pull_request:6754|pull_request:8978", + "pull_request:6765|pull_request:8978", + "pull_request:6766|pull_request:8978", + "pull_request:6767|pull_request:8978", + "pull_request:6768|pull_request:8978", + "pull_request:6800|pull_request:8990", + "pull_request:7247|pull_request:8696", + "pull_request:7627|pull_request:9506", + "pull_request:8322|pull_request:8572", + "pull_request:8478|pull_request:9620", + "pull_request:3160|pull_request:3476", + "pull_request:10454|pull_request:10482", + "pull_request:11516|pull_request:12021", + "pull_request:1038|pull_request:1147", + "pull_request:11469|pull_request:12530", + "pull_request:6484|pull_request:6999", + "pull_request:6229|pull_request:7941", + "pull_request:7632|pull_request:9601", + "pull_request:11273|pull_request:11433", + "pull_request:1527|pull_request:1913" + ], + "prepared_review_unit_hash": "e571268b4f9bcf87ad81a3d496322906a51db157d3235eb9779f55fc1e1f7591", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 101599, + "estimated_input_tokens": 25400, + "estimated_eval_tokens": 51056 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: it mixes unrelated docs, examples, tests, and core bugfix PRs. I only see one plausible near-duplicate pair in the HunyuanVideo attention-mask fixes; everything else should be kept separate.", + "confidence": 0.62, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10454 is the strongest single representative only because it has the clearest explicit bugfix/issue linkage in the one repeated code path (HunyuanVideo attention mask); the rest of the cluster does not form a coherent duplicate set.", + "best_issue_reason": null, + "best_pr_reason": "PR 10454 is the best standalone PR to anchor this cluster: it is a concrete fix with an explicit issue target and a narrowly scoped code-path change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4294", + "right": "pull_request:5916", + "accept": false, + "reason": "Both are rename/cleanup docs or args changes, but they touch different SDXL vs DreamBooth surfaces and solve unrelated problems." + }, + { + "left": "pull_request:4347", + "right": "pull_request:5707", + "accept": false, + "reason": "Both involve scheduler behavior, but one adds a step-index counter and the other adds LCM tests; not the same fix." + }, + { + "left": "pull_request:5552", + "right": "pull_request:6150", + "accept": false, + "reason": "Both are core loader/model changes, but one fixes FreeU disable logic while the other adds IP-Adapter conversion support." + }, + { + "left": "pull_request:6007", + "right": "pull_request:6864", + "accept": false, + "reason": "Both are docs-only, but they cover unrelated topics: contribution guidance vs PEFT/torch.compile notes." + }, + { + "left": "pull_request:6140", + "right": "pull_request:6643", + "accept": false, + "reason": "Both add docs entries, but IP-Adapter and UViT2D are different features with no shared bug/change." + }, + { + "left": "pull_request:6215", + "right": "pull_request:8386", + "accept": false, + "reason": "Both adjust test initialization, but they target different models and different test harness behavior." + }, + { + "left": "pull_request:6754", + "right": "pull_request:8978", + "accept": false, + "reason": "Different research/example updates: diffusion DPO loss math vs DreamBooth notebook/example overhaul." + }, + { + "left": "pull_request:6765", + "right": "pull_request:8978", + "accept": false, + "reason": "Cache-path utility cleanup and DreamBooth notebook/example rewrite are unrelated changes." + }, + { + "left": "pull_request:6766", + "right": "pull_request:8978", + "accept": false, + "reason": "Constants path handling and DreamBooth notebook/example rewrite are unrelated." + }, + { + "left": "pull_request:6767", + "right": "pull_request:8978", + "accept": false, + "reason": "Flax model path handling and DreamBooth notebook/example rewrite are unrelated fixes." + }, + { + "left": "pull_request:6768", + "right": "pull_request:8978", + "accept": false, + "reason": "Configuration path handling and DreamBooth notebook/example rewrite are unrelated." + }, + { + "left": "pull_request:6800", + "right": "pull_request:8990", + "accept": false, + "reason": "Import availability plumbing and adding a new PAG ControlNet pipeline are different changes." + }, + { + "left": "pull_request:7247", + "right": "pull_request:8696", + "accept": false, + "reason": "One fixes single-file config checking, the other marks Marigold tests flaky; different code paths and goals." + }, + { + "left": "pull_request:7627", + "right": "pull_request:9506", + "accept": false, + "reason": "Both touch img2img-ish docs/code, but one fixes latent scaling in ControlNet/SDXL while the other is a docs import/path update." + }, + { + "left": "pull_request:8322", + "right": "pull_request:8572", + "accept": false, + "reason": "Both are broad errata/docs sweep PRs, but they cover different files and different cleanup sets." + }, + { + "left": "pull_request:8478", + "right": "pull_request:9620", + "accept": false, + "reason": "HunyuanDiT ControlNet/pipeline addition is unrelated to ControlNet interrupt support." + }, + { + "left": "pull_request:3160", + "right": "pull_request:3476", + "accept": false, + "reason": "Both add Karras sigmas support, but they are separate schedulers and not the same concrete change." + }, + { + "left": "pull_request:10454", + "right": "pull_request:10482", + "accept": true, + "reason": "Both modify the same HunyuanVideo attention-mask code path to fix masking-related failures; they are close enough to be merged into one PR." + }, + { + "left": "pull_request:11516", + "right": "pull_request:12021", + "accept": false, + "reason": "LTX Video release/docs/model wiring is unrelated to LTX attention-backend support." + }, + { + "left": "pull_request:1038", + "right": "pull_request:1147", + "accept": false, + "reason": "Same Flax example area, but different bugs: dtype selection vs loading the text encoder from the correct subfolder." + }, + { + "left": "pull_request:11469", + "right": "pull_request:12530", + "accept": false, + "reason": "Both add xfails, but they target different failing tests and different upstream version constraints." + }, + { + "left": "pull_request:6484", + "right": "pull_request:6999", + "accept": false, + "reason": "Same SVD pipeline file, but different bugs: GPU-generated tensors vs image-as-tensor preprocessing." + }, + { + "left": "pull_request:6229", + "right": "pull_request:7941", + "accept": false, + "reason": "Adding PEFT to CI and marking a single test class slow are unrelated test/CI changes." + }, + { + "left": "pull_request:7632", + "right": "pull_request:9601", + "accept": false, + "reason": "Both involve DreamBooth/SD3 training, but one fixes VAE path handling and the other makes VAE dtype conditional on mixed precision." + }, + { + "left": "pull_request:11273", + "right": "pull_request:11433", + "accept": false, + "reason": "Notification bot timeout cleanup is unrelated to a test decorator rename/fix." + }, + { + "left": "pull_request:1527", + "right": "pull_request:1913", + "accept": false, + "reason": "Both are pipeline input-handling changes, but stable diffusion inpaint tensor support and img2img input validation are different bugs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, 10454/10482, is plausibly mergeable because both edit the same HunyuanVideo attention-mask code path; the rest of the pairwise rejections are consistent with distinct bugs/features rather than duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 136, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10086", + "pull_request:10966", + "pull_request:11089", + "pull_request:11243", + "pull_request:11366", + "pull_request:11614", + "pull_request:11841", + "pull_request:1229", + "pull_request:12315", + "pull_request:12450", + "pull_request:12493", + "pull_request:1468", + "pull_request:152", + "pull_request:1554", + "pull_request:1913", + "pull_request:1945", + "pull_request:1968", + "pull_request:2706", + "pull_request:2928", + "pull_request:2959", + "pull_request:2968", + "pull_request:3037", + "pull_request:3298", + "pull_request:3536", + "pull_request:3589", + "pull_request:3798", + "pull_request:4451", + "pull_request:4855", + "pull_request:5035", + "pull_request:5111", + "pull_request:5788", + "pull_request:5792", + "pull_request:5946", + "pull_request:5948", + "pull_request:6261", + "pull_request:6432", + "pull_request:6448", + "pull_request:6904", + "pull_request:7044", + "pull_request:7068", + "pull_request:7277", + "pull_request:7455", + "pull_request:788", + "pull_request:8467", + "pull_request:8510", + "pull_request:8512", + "pull_request:8761", + "pull_request:8937" + ], + "soft_pairs": [ + "pull_request:1554|pull_request:1968", + "pull_request:3589|pull_request:5111", + "pull_request:3798|pull_request:5948", + "pull_request:4451|pull_request:6261", + "pull_request:4855|pull_request:5788", + "pull_request:5946|pull_request:6448", + "pull_request:10086|pull_request:8467", + "pull_request:10966|pull_request:11366", + "pull_request:11089|pull_request:11841", + "pull_request:11614|pull_request:12450", + "pull_request:1229|pull_request:152", + "pull_request:12315|pull_request:12493", + "pull_request:1468|pull_request:3298", + "pull_request:1913|pull_request:3536", + "pull_request:1945|pull_request:2706", + "pull_request:2928|pull_request:3298", + "pull_request:2968|pull_request:788", + "pull_request:5035|pull_request:5792", + "pull_request:6432|pull_request:8510", + "pull_request:6904|pull_request:8937", + "pull_request:7044|pull_request:8761", + "pull_request:7455|pull_request:8512", + "pull_request:11243|pull_request:12315", + "pull_request:7068|pull_request:7277", + "pull_request:2959|pull_request:3037" + ], + "prepared_review_unit_hash": "423e493a6ac3b35396fa2cf807fadd4978243b78e2999f69dbdf132ad98787f5", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 102678, + "estimated_input_tokens": 25670, + "estimated_eval_tokens": 51596 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The candidate pairs are mostly superficial similarities across unrelated docs, tests, CI, and pipeline fixes. I don\u2019t see any pair that fixes the same concrete bug/change, so no duplicates are accepted and there is no meaningful canonical PR/issue for the cluster.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1554", + "right": "pull_request:1968", + "accept": false, + "reason": "Docs-only correction for UNet text shapes vs a CPU offload behavior fix; different files and problems." + }, + { + "left": "pull_request:3589", + "right": "pull_request:5111", + "accept": false, + "reason": "Reference-tricks README update vs adding karras sigmas to KDPM2 schedulers; unrelated changes." + }, + { + "left": "pull_request:3798", + "right": "pull_request:5948", + "accept": false, + "reason": "Comment/typo cleanup across training scripts vs a warning-condition fix in from_pretrained; not the same bug." + }, + { + "left": "pull_request:4451", + "right": "pull_request:6261", + "accept": false, + "reason": "Moving tests to nightly is a test-runner change; moving Attend-and-Excite out of stable_diffusion is a code refactor." + }, + { + "left": "pull_request:4855", + "right": "pull_request:5788", + "accept": false, + "reason": "Offload-hook handling with textual inversion side loading vs an LCM timesteps error check; unrelated code paths." + }, + { + "left": "pull_request:5946", + "right": "pull_request:6448", + "accept": false, + "reason": "Example-test refactor vs LoRA test assertion fixes; both test-related but different underlying issues." + }, + { + "left": "pull_request:10086", + "right": "pull_request:8467", + "accept": false, + "reason": "StableDiffusion3 skip-guidance batch-size bug vs SEGA device-placement fix; different pipelines and failures." + }, + { + "left": "pull_request:10966", + "right": "pull_request:11366", + "accept": false, + "reason": "CogView4 input validation vs Hidream transformer output-shape change; unrelated functionality." + }, + { + "left": "pull_request:11089", + "right": "pull_request:11841", + "accept": false, + "reason": "Wan docs memory-optimization content vs generic batch-generation docs; different documentation topics." + }, + { + "left": "pull_request:11614", + "right": "pull_request:12450", + "accept": false, + "reason": "Docker CI security fix vs CI installation cleanup; both CI but not the same issue." + }, + { + "left": "pull_request:1229", + "right": "pull_request:152", + "accept": false, + "reason": "DreamBooth README installation note vs local-folder training support; unrelated training features." + }, + { + "left": "pull_request:12315", + "right": "pull_request:12493", + "accept": false, + "reason": "LoRA test simplification vs LTX-Video 0.9.8 model/docs updates; no shared bug or change." + }, + { + "left": "pull_request:1468", + "right": "pull_request:3298", + "accept": false, + "reason": "Saving trained betas in schedulers vs arbitrary aspect-ratio support in IF super-resolution; different subsystems." + }, + { + "left": "pull_request:1913", + "right": "pull_request:3536", + "accept": false, + "reason": "Img2img input-type validation vs PIL normalization for IF super-resolution; separate bugs and pipelines." + }, + { + "left": "pull_request:1945", + "right": "pull_request:2706", + "accept": false, + "reason": "Adding UNet block tests vs preserving/restoring EMA weights in training; tests vs runtime training logic." + }, + { + "left": "pull_request:2928", + "right": "pull_request:3298", + "accept": false, + "reason": "ControlNet local-dataset loading vs IF super-resolution aspect-ratio support; unrelated pipeline features." + }, + { + "left": "pull_request:2968", + "right": "pull_request:788", + "accept": false, + "reason": "Post-processing indentation/control-flow fix vs PIL resampling deprecation warnings; different bug fixes." + }, + { + "left": "pull_request:5035", + "right": "pull_request:5792", + "accept": false, + "reason": "Dropping forced transformers dependency vs import-path fixes in examples; unrelated packaging/import issues." + }, + { + "left": "pull_request:6432", + "right": "pull_request:8510", + "accept": false, + "reason": "README learning-rate example update vs LR scheduler math fixes in advanced training scripts; not the same change." + }, + { + "left": "pull_request:6904", + "right": "pull_request:8937", + "accept": false, + "reason": "load_image PIL support vs long-prompt handling in AuraFlow; unrelated code paths." + }, + { + "left": "pull_request:7044", + "right": "pull_request:8761", + "accept": false, + "reason": "Re-adding UNet refactor plumbing vs Hunyuan-DiT embedding refactor; different model internals." + }, + { + "left": "pull_request:7455", + "right": "pull_request:8512", + "accept": false, + "reason": "Removing distutils dependencies vs removing deprecated prepare_mask_and_masked_image; unrelated cleanup changes." + }, + { + "left": "pull_request:11243", + "right": "pull_request:12315", + "accept": false, + "reason": "Musubi Wan LoRA conversion support vs LoRA test scheduler simplification; one is feature support, the other test maintenance." + }, + { + "left": "pull_request:7068", + "right": "pull_request:7277", + "accept": false, + "reason": "General typo fixes vs deprecated-parameter test updates plus typos; overlapping wording only, not the same underlying fix." + }, + { + "left": "pull_request:2959", + "right": "pull_request:3037", + "accept": false, + "reason": "UNet3D norm-group test adjustment vs skipping an unstable MPS test; separate test issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 137, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10021", + "pull_request:10140", + "pull_request:10148", + "pull_request:10326", + "pull_request:10430", + "pull_request:10454", + "pull_request:10478", + "pull_request:10537", + "pull_request:10579", + "pull_request:10837", + "pull_request:10943", + "pull_request:10968", + "pull_request:10981", + "pull_request:11052", + "pull_request:11089", + "pull_request:11091", + "pull_request:11142", + "pull_request:11252", + "pull_request:11434", + "pull_request:12086", + "pull_request:12490", + "pull_request:13022", + "pull_request:2595", + "pull_request:2928", + "pull_request:3499", + "pull_request:3632", + "pull_request:3708", + "pull_request:382", + "pull_request:3928", + "pull_request:401", + "pull_request:4112", + "pull_request:4408", + "pull_request:4819", + "pull_request:5393", + "pull_request:5407", + "pull_request:5747", + "pull_request:5952", + "pull_request:6065", + "pull_request:6308", + "pull_request:6311", + "pull_request:6405", + "pull_request:7247", + "pull_request:7461", + "pull_request:7771", + "pull_request:8618", + "pull_request:8652", + "pull_request:8722", + "pull_request:9918" + ], + "soft_pairs": [ + "pull_request:3499|pull_request:3632", + "pull_request:10021|pull_request:10968", + "pull_request:10537|pull_request:11142", + "pull_request:2595|pull_request:3928", + "pull_request:6311|pull_request:6405", + "pull_request:4408|pull_request:4819", + "pull_request:12490|pull_request:13022", + "pull_request:10943|pull_request:10981", + "pull_request:10326|pull_request:11252", + "pull_request:10478|pull_request:11434", + "pull_request:2928|pull_request:3708", + "pull_request:382|pull_request:401", + "pull_request:4112|pull_request:5747", + "pull_request:5393|pull_request:5407", + "pull_request:5952|pull_request:6065", + "pull_request:8618|pull_request:8722", + "pull_request:10140|pull_request:10430", + "pull_request:6308|pull_request:7461", + "pull_request:7247|pull_request:7771", + "pull_request:10148|pull_request:8652", + "pull_request:10454|pull_request:10837", + "pull_request:10579|pull_request:9918", + "pull_request:11052|pull_request:11089", + "pull_request:11091|pull_request:12086" + ], + "prepared_review_unit_hash": "bad33a2d9c2bb8cf9ba88b31fa4b4804d44c6a43497eef886d6046f2bda9a322", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 106130, + "estimated_input_tokens": 26533, + "estimated_eval_tokens": 53322 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These PRs are mostly unrelated feature additions, docs updates, small fixes, and test/util changes across different subsystems; none of the soft edges look like true duplicates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR for the cluster: the items span unrelated changes, so there isn't one duplicate source of truth.", + "best_issue_reason": null, + "best_pr_reason": "If one representative must be chosen, PR 8652 is the most substantial self-contained core feature addition with broad docs/tests and a clear scope, but it still does not duplicate the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3499", + "right": "pull_request:3632", + "accept": false, + "reason": "Same panorama file, but 3499 fixes a dpmsolver-related bug while 3632 adds view batching/docs; different changes." + }, + { + "left": "pull_request:10021", + "right": "pull_request:10968", + "accept": false, + "reason": "Both are LTX-related, but 10021 is the initial LTX Video integration and 10968 is the later LTX 0.9.5 update with different components and behavior." + }, + { + "left": "pull_request:10537", + "right": "pull_request:11142", + "accept": false, + "reason": "Both touch Sana assets, but one is a Sana-4K pipeline PR and the other is a Sana Sprint converter fix; not the same bug/change." + }, + { + "left": "pull_request:2595", + "right": "pull_request:3928", + "accept": false, + "reason": "Shared scheduler/controlnet files are broad overlap, but 2595 is about scheduler config switching and 3928 fixes controlnet guidance range handling; not duplicates." + }, + { + "left": "pull_request:6311", + "right": "pull_request:6405", + "accept": false, + "reason": "One is a docs tutorial addition and the other is a doc-builder workflow tweak; unrelated." + }, + { + "left": "pull_request:4408", + "right": "pull_request:4819", + "accept": false, + "reason": "Different code paths: SDXL from-single-file support versus inpainting masked latent handling." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13022", + "accept": false, + "reason": "A test fixture tweak versus Dockerfile/runtime environment fixes; unrelated." + }, + { + "left": "pull_request:10943", + "right": "pull_request:10981", + "accept": false, + "reason": "Both add LoRA support, but for different model families (Wan vs CogView4) and different code paths." + }, + { + "left": "pull_request:10326", + "right": "pull_request:11252", + "accept": false, + "reason": "Workflow quoting fix versus timeout constant change; unrelated." + }, + { + "left": "pull_request:10478", + "right": "pull_request:11434", + "accept": false, + "reason": "Style-only cleanup in pipelines versus a test import fix; not the same bug." + }, + { + "left": "pull_request:2928", + "right": "pull_request:3708", + "accept": false, + "reason": "ControlNet local dataset loading and custom releases version sorting are unrelated changes." + }, + { + "left": "pull_request:382", + "right": "pull_request:401", + "accept": false, + "reason": "Both are docs PRs, but for different pipelines and different content." + }, + { + "left": "pull_request:4112", + "right": "pull_request:5747", + "accept": false, + "reason": "Dreambooth LoRA rank support versus dataset loading parameter propagation; different features." + }, + { + "left": "pull_request:5393", + "right": "pull_request:5407", + "accept": false, + "reason": "Examples/docs HFApi update versus mixed-precision training fix in a different example; not duplicates." + }, + { + "left": "pull_request:5952", + "right": "pull_request:6065", + "accept": false, + "reason": "Pipeline overview doc update versus SDXL Turbo docs; related area but different content and scope." + }, + { + "left": "pull_request:8618", + "right": "pull_request:8722", + "accept": false, + "reason": "SD3 tiny autoencoder shift_factor addition versus a large embeddings refactor; not the same issue." + }, + { + "left": "pull_request:10140", + "right": "pull_request:10430", + "accept": false, + "reason": "ConsisID model/pipeline integration versus TeaCache hook infrastructure; unrelated." + }, + { + "left": "pull_request:6308", + "right": "pull_request:7461", + "accept": false, + "reason": "LoRA dummy component test changes versus a utility bool parser fix; unrelated." + }, + { + "left": "pull_request:7247", + "right": "pull_request:7771", + "accept": false, + "reason": "Config checking test expectations versus marking a different test flaky; not duplicates." + }, + { + "left": "pull_request:10454", + "right": "pull_request:10837", + "accept": false, + "reason": "Hunyuan attention mask dimension fix versus SkyReels Hunyuan model/pipeline addition; different change intent." + }, + { + "left": "pull_request:10579", + "right": "pull_request:9918", + "accept": false, + "reason": "A community example pipeline versus a server example; different features." + }, + { + "left": "pull_request:11052", + "right": "pull_request:11089", + "accept": false, + "reason": "Testing utility decorator addition versus Wan documentation updates; unrelated." + }, + { + "left": "pull_request:11091", + "right": "pull_request:12086", + "accept": false, + "reason": "Flux remote encode support and modular Flux I2I are both Flux-related but different implementations and problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 138, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11979", + "pull_request:12005", + "pull_request:12137", + "pull_request:1215", + "pull_request:12420", + "pull_request:12736", + "pull_request:13081", + "pull_request:13120", + "pull_request:13150", + "pull_request:13472", + "pull_request:1468", + "pull_request:152", + "pull_request:1525", + "pull_request:1554", + "pull_request:1920", + "pull_request:1937", + "pull_request:2655", + "pull_request:2693", + "pull_request:2892", + "pull_request:2928", + "pull_request:2970", + "pull_request:2983", + "pull_request:3001", + "pull_request:3054", + "pull_request:3292", + "pull_request:3321", + "pull_request:3397", + "pull_request:3476", + "pull_request:3536", + "pull_request:3561", + "pull_request:3657", + "pull_request:3708", + "pull_request:3746", + "pull_request:402", + "pull_request:4112", + "pull_request:4396", + "pull_request:4787", + "pull_request:4887", + "pull_request:4963", + "pull_request:51", + "pull_request:5298", + "pull_request:530", + "pull_request:5331", + "pull_request:5407", + "pull_request:5946", + "pull_request:788", + "pull_request:8563", + "pull_request:9721" + ], + "soft_pairs": [ + "pull_request:12005|pull_request:12420", + "pull_request:12137|pull_request:13081", + "pull_request:13120|pull_request:13150", + "pull_request:13120|pull_request:13472", + "pull_request:1468|pull_request:2892", + "pull_request:152|pull_request:788", + "pull_request:1525|pull_request:51", + "pull_request:1554|pull_request:2655", + "pull_request:1554|pull_request:402", + "pull_request:1920|pull_request:402", + "pull_request:1937|pull_request:3536", + "pull_request:2693|pull_request:3708", + "pull_request:2928|pull_request:3657", + "pull_request:2970|pull_request:4396", + "pull_request:2983|pull_request:4963", + "pull_request:3054|pull_request:3321", + "pull_request:3292|pull_request:4887", + "pull_request:3321|pull_request:5407", + "pull_request:3397|pull_request:3746", + "pull_request:3561|pull_request:4112", + "pull_request:4787|pull_request:5946", + "pull_request:5298|pull_request:5331", + "pull_request:8563|pull_request:9721", + "pull_request:1215|pull_request:530", + "pull_request:3001|pull_request:3476", + "pull_request:11979|pull_request:12736" + ], + "prepared_review_unit_hash": "10a77d2f02ff7799d7d5e53b0186f860d3ae9bcb1ab11cbe69e5fa3ea35b751c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 98430, + "estimated_input_tokens": 24608, + "estimated_eval_tokens": 49472 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly unrelated pull requests that happen to cluster around tests, docs, loaders, and scheduler fixes; there is no clear duplicate issue/PR group here. I would treat them as separate artifacts rather than one underlying change set.", + "confidence": 0.19, + "canonical_issue_reason": null, + "canonical_pr_reason": "If one PR must represent the cluster, PR #12005 is the cleanest standalone bugfix: it fixes sharded checkpoint loading to respect `local_files_only`, with a narrow and concrete code path.", + "best_issue_reason": null, + "best_pr_reason": "PR #12005 is the best single PR candidate because it addresses a specific user-facing bug, has a tight scope, and is clearly distinct from the mostly test/doc-only or unrelated maintenance changes in the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12005", + "right": "pull_request:12420", + "accept": false, + "reason": "Different bugs: sharded checkpoint handling vs conditional import of `torch.distributed`; same broad area only." + }, + { + "left": "pull_request:12137", + "right": "pull_request:13081", + "accept": false, + "reason": "Unrelated fixes: constants/parallel loading flag vs Qwen hidden-state contiguity for torchao." + }, + { + "left": "pull_request:13120", + "right": "pull_request:13150", + "accept": false, + "reason": "Both are test changes, but they target different behaviors and code paths." + }, + { + "left": "pull_request:13120", + "right": "pull_request:13472", + "accept": false, + "reason": "A test helper tweak and a docs-only gotcha addition are not the same underlying change." + }, + { + "left": "pull_request:1468", + "right": "pull_request:2892", + "accept": false, + "reason": "Separate functional areas: saving trained betas vs upscaler input validation/embeds." + }, + { + "left": "pull_request:152", + "right": "pull_request:788", + "accept": false, + "reason": "Local image-folder support is unrelated to PIL resampling warning fixes." + }, + { + "left": "pull_request:1525", + "right": "pull_request:51", + "accept": false, + "reason": "Different scope and era: upscaler batch-size bugfix vs packaging metadata typo." + }, + { + "left": "pull_request:1554", + "right": "pull_request:2655", + "accept": false, + "reason": "Docs correction vs LoRA loading/saving behavior are not the same change." + }, + { + "left": "pull_request:1554", + "right": "pull_request:402", + "accept": false, + "reason": "Both touch docs, but they address different documentation reorganizations." + }, + { + "left": "pull_request:1920", + "right": "pull_request:402", + "accept": false, + "reason": "Both are docs PRs, but one adds translating guidance and the other reorganizes the intro docs tree." + }, + { + "left": "pull_request:1937", + "right": "pull_request:3536", + "accept": false, + "reason": "EMA-weight extraction logic is unrelated to IF super-resolution image normalization." + }, + { + "left": "pull_request:2693", + "right": "pull_request:3708", + "accept": false, + "reason": "A doc image link fix is unrelated to version sorting for custom releases." + }, + { + "left": "pull_request:2928", + "right": "pull_request:3657", + "accept": false, + "reason": "Both mention ControlNet, but one adds local dataset loading and the other tweaks checkpoint parser arguments." + }, + { + "left": "pull_request:2970", + "right": "pull_request:4396", + "accept": false, + "reason": "Variable rename in a DreamBooth script vs prompt-encoding behavior across pipelines are different bugs." + }, + { + "left": "pull_request:2983", + "right": "pull_request:4963", + "accept": false, + "reason": "Installing test dependencies is separate from pinning accelerate/main in workflows." + }, + { + "left": "pull_request:3054", + "right": "pull_request:3321", + "accept": false, + "reason": "Different docs pages and different content updates; not a duplicate change." + }, + { + "left": "pull_request:3292", + "right": "pull_request:4887", + "accept": false, + "reason": "Input perturbation training option vs removing latent input from a different pipeline are unrelated." + }, + { + "left": "pull_request:3321", + "right": "pull_request:5407", + "accept": false, + "reason": "Docs update vs mixed-precision training fix; same repo area only by coincidence." + }, + { + "left": "pull_request:3397", + "right": "pull_request:3746", + "accept": false, + "reason": "Documentation note on local paths vs code fix for `from_ckpt` path joining are distinct." + }, + { + "left": "pull_request:3561", + "right": "pull_request:4112", + "accept": false, + "reason": "Both touch DreamBooth LoRA files, but one is IF stage-II upscaling and the other adds rank handling." + }, + { + "left": "pull_request:4787", + "right": "pull_request:5946", + "accept": false, + "reason": "LoRA checkpoint filename filtering is unrelated to refactoring example tests." + }, + { + "left": "pull_request:5298", + "right": "pull_request:5331", + "accept": false, + "reason": "Torch.compile fix for LoRA conv is different from switching DreamBooth LoRA training to LoRALinear." + }, + { + "left": "pull_request:8563", + "right": "pull_request:9721", + "accept": false, + "reason": "Version pins in Python deps are unrelated to Docker torch version pins." + }, + { + "left": "pull_request:1215", + "right": "pull_request:530", + "accept": false, + "reason": "Different pipelines and different failure modes: DDIM CPU generator handling vs img2img test flakiness." + }, + { + "left": "pull_request:3001", + "right": "pull_request:3476", + "accept": false, + "reason": "Related scheduler family, but not the same concrete change or bug." + }, + { + "left": "pull_request:11979", + "right": "pull_request:12736", + "accept": false, + "reason": "Torch version gating in tests is unrelated to enabling regional compilation on a model." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 139, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10200", + "pull_request:10354", + "pull_request:10704", + "pull_request:10850", + "pull_request:11254", + "pull_request:11455", + "pull_request:11465", + "pull_request:11916", + "pull_request:11995", + "pull_request:12004", + "pull_request:12074", + "pull_request:12092", + "pull_request:12236", + "pull_request:12419", + "pull_request:12812", + "pull_request:130", + "pull_request:13169", + "pull_request:13280", + "pull_request:1513", + "pull_request:1525", + "pull_request:2314", + "pull_request:2968", + "pull_request:3561", + "pull_request:388", + "pull_request:390", + "pull_request:4262", + "pull_request:4347", + "pull_request:5559", + "pull_request:6114", + "pull_request:6140", + "pull_request:6305", + "pull_request:6450", + "pull_request:6886", + "pull_request:689", + "pull_request:6994", + "pull_request:7007", + "pull_request:7115", + "pull_request:7319", + "pull_request:748", + "pull_request:7652", + "pull_request:8600", + "pull_request:8860", + "pull_request:9175", + "pull_request:9307", + "pull_request:9638", + "pull_request:9676", + "pull_request:9696" + ], + "soft_pairs": [ + "pull_request:5559|pull_request:6114", + "pull_request:12812|pull_request:13280", + "pull_request:12419|pull_request:13169", + "pull_request:9175|pull_request:9638", + "pull_request:10200|pull_request:9307", + "pull_request:10704|pull_request:11455", + "pull_request:10704|pull_request:11465", + "pull_request:10850|pull_request:12092", + "pull_request:11254|pull_request:9696", + "pull_request:11916|pull_request:12236", + "pull_request:130|pull_request:1513", + "pull_request:1525|pull_request:3561", + "pull_request:2314|pull_request:689", + "pull_request:2968|pull_request:748", + "pull_request:388|pull_request:390", + "pull_request:4262|pull_request:6450", + "pull_request:4347|pull_request:6305", + "pull_request:6140|pull_request:7319", + "pull_request:6450|pull_request:7007", + "pull_request:6886|pull_request:6994", + "pull_request:7115|pull_request:7652", + "pull_request:8600|pull_request:8860", + "pull_request:12004|pull_request:12074", + "pull_request:11995|pull_request:12419", + "pull_request:10354|pull_request:9676" + ], + "prepared_review_unit_hash": "07be13443aac773caf70db10cc9943fd60f44ea819dba8ca228de4f00f323e3b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 96569, + "estimated_input_tokens": 24143, + "estimated_eval_tokens": 48542 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose similarity bucket, not a true duplicate set. The PRs span unrelated docs, training scripts, loaders, attention backends, Flux/Wan work, and workflow chores; none of the soft edges look like the same underlying change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical choice because the grouped PRs do not share one concrete bugfix or feature; they only share broad subsystem keywords or filename overlap.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global representative for duplicate triage: the cluster is heterogeneous and the soft matches are mostly false positives.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5559", + "right": "pull_request:6114", + "accept": false, + "reason": "Both are pipeline-related, but one adds Latent Consistency img2img and the other adds lpw-xl img2img/inpaint support; different changes and code paths." + }, + { + "left": "pull_request:12812", + "right": "pull_request:13280", + "accept": false, + "reason": "They both touch attention dispatch, but one adds CP wrapping for kernels-based backends while the other adds FA4 support/docs; not the same fix." + }, + { + "left": "pull_request:12419", + "right": "pull_request:13169", + "accept": false, + "reason": "Flux modular-pipeline mellon prep and Flux Klein PEFT LoRA format support are separate concerns with different code paths." + }, + { + "left": "pull_request:9175", + "right": "pull_request:9638", + "accept": false, + "reason": "Both are Flux ControlNet related, but one introduces Flux Union ControlNet support and the other adds Xlabs ControlNets; not mergeable as one duplicate change." + }, + { + "left": "pull_request:10200", + "right": "pull_request:9307", + "accept": false, + "reason": "ControlNet Union refactor and rotary embedding CPU fix are unrelated changes in different parts of the library." + }, + { + "left": "pull_request:10704", + "right": "pull_request:11455", + "accept": false, + "reason": "Docs for img2img/upscaler usage versus a training-script interpolation default change; same theme, not the same underlying artifact." + }, + { + "left": "pull_request:10704", + "right": "pull_request:11465", + "accept": false, + "reason": "The docs PR and the LoRA training-script PR address different files and behaviors, so they are not duplicates." + }, + { + "left": "pull_request:10850", + "right": "pull_request:12092", + "accept": false, + "reason": "Hook utility methods and GGUF docs are unrelated; no shared bug or code path." + }, + { + "left": "pull_request:11254", + "right": "pull_request:9696", + "accept": false, + "reason": "ConsisID import fixes and the test-workflow input rename are unrelated maintenance changes." + }, + { + "left": "pull_request:11916", + "right": "pull_request:12236", + "accept": false, + "reason": "Shared attention_dispatch.py overlap, but one is a large refactor and the other adds kernels-backed FLASH_3 hub support; different concrete goals." + }, + { + "left": "pull_request:130", + "right": "pull_request:1513", + "accept": false, + "reason": "Both are documentation-only, but one adds accelerate config to examples README and the other adds notebook config guidance for dreambooth; not the same document change." + }, + { + "left": "pull_request:1525", + "right": "pull_request:3561", + "accept": false, + "reason": "Batch-size fix in upscaling versus Dreambooth IF stage-II upscaling training changes; different pipelines and bugs." + }, + { + "left": "pull_request:2314", + "right": "pull_request:689", + "accept": false, + "reason": "One adjusts fast-test expectations, the other changes attention math for performance; not the same issue." + }, + { + "left": "pull_request:2968", + "right": "pull_request:748", + "accept": false, + "reason": "Post-processing bugfixes and push_to_hub call fixes are unrelated." + }, + { + "left": "pull_request:388", + "right": "pull_request:390", + "accept": false, + "reason": "These are separate docs over different topics (latent diffusion vs MPS optimization), with no shared underlying change." + }, + { + "left": "pull_request:4262", + "right": "pull_request:6450", + "accept": false, + "reason": "SDXL DreamBooth LoRA fixes are unrelated to LoRA deprecation test cleanup." + }, + { + "left": "pull_request:4347", + "right": "pull_request:6305", + "accept": false, + "reason": "A scheduler step-index counter and zero-terminal-SNR beta rescaling are distinct scheduler features." + }, + { + "left": "pull_request:6140", + "right": "pull_request:7319", + "accept": false, + "reason": "Adding IP-Adapter docs and adding EDM scheduler docs are separate documentation additions." + }, + { + "left": "pull_request:6450", + "right": "pull_request:7007", + "accept": false, + "reason": "Test renaming/cleanup versus LoRA attention-processor deprecation are different changes." + }, + { + "left": "pull_request:6886", + "right": "pull_request:6994", + "accept": false, + "reason": "One removes torch_dtype from to(), the other passes torch_dtype into single-file loaders; related API area, but not the same fix." + }, + { + "left": "pull_request:7115", + "right": "pull_request:7652", + "accept": false, + "reason": "Dockerfile package-install changes and workflow buildx enforcement are unrelated infrastructure updates." + }, + { + "left": "pull_request:8600", + "right": "pull_request:8860", + "accept": false, + "reason": "The SD3 DreamBooth update and the README_sd3.md-only PR do not fix the same concrete bug or change." + }, + { + "left": "pull_request:12004", + "right": "pull_request:12074", + "accept": false, + "reason": "Wan 2.2 core model support and Wan 2.2 LoRA loading support are related project work, but they are separate features and not one duplicate PR." + }, + { + "left": "pull_request:11995", + "right": "pull_request:12419", + "accept": false, + "reason": "Both are Flux modular-pipeline work, but one adds modular Flux text-to-image and the other prepares Flux for mellon; different concrete objectives." + }, + { + "left": "pull_request:10354", + "right": "pull_request:9676", + "accept": false, + "reason": "Flux IP-adapter tester fix and a benchmark import cleanup are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 140, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10086", + "pull_request:10127", + "pull_request:10342", + "pull_request:10351", + "pull_request:11039", + "pull_request:11252", + "pull_request:1128", + "pull_request:11639", + "pull_request:11716", + "pull_request:11726", + "pull_request:12041", + "pull_request:12424", + "pull_request:12521", + "pull_request:12806", + "pull_request:12812", + "pull_request:1283", + "pull_request:1299", + "pull_request:13022", + "pull_request:1355", + "pull_request:1527", + "pull_request:1571", + "pull_request:1968", + "pull_request:2129", + "pull_request:2286", + "pull_request:2452", + "pull_request:3313", + "pull_request:3797", + "pull_request:4242", + "pull_request:5213", + "pull_request:5361", + "pull_request:5653", + "pull_request:5852", + "pull_request:5915", + "pull_request:6330", + "pull_request:7246", + "pull_request:750", + "pull_request:7632", + "pull_request:7714", + "pull_request:7771", + "pull_request:7825", + "pull_request:788", + "pull_request:8452", + "pull_request:8510", + "pull_request:8696", + "pull_request:8797", + "pull_request:9060", + "pull_request:9506", + "pull_request:9676" + ], + "soft_pairs": [ + "pull_request:7771|pull_request:7825", + "pull_request:1128|pull_request:3313", + "pull_request:2129|pull_request:2286", + "pull_request:12806|pull_request:12812", + "pull_request:11639|pull_request:11726", + "pull_request:10342|pull_request:9060", + "pull_request:10351|pull_request:11252", + "pull_request:11252|pull_request:9676", + "pull_request:1527|pull_request:788", + "pull_request:1968|pull_request:3797", + "pull_request:4242|pull_request:5213", + "pull_request:5361|pull_request:5852", + "pull_request:6330|pull_request:7714", + "pull_request:7632|pull_request:9506", + "pull_request:11039|pull_request:11716", + "pull_request:2452|pull_request:750", + "pull_request:12041|pull_request:12521", + "pull_request:10127|pull_request:8696", + "pull_request:10127|pull_request:8797", + "pull_request:12424|pull_request:13022", + "pull_request:1283|pull_request:1299", + "pull_request:5653|pull_request:5915", + "pull_request:7246|pull_request:8696", + "pull_request:1355|pull_request:1571", + "pull_request:10351|pull_request:11716", + "pull_request:10086|pull_request:8452", + "pull_request:10086|pull_request:8510" + ], + "prepared_review_unit_hash": "9ee042164eb2a121d77a596c836eaa614ecd2a020079119b5a7ca356a10f7e4c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 103654, + "estimated_input_tokens": 25914, + "estimated_eval_tokens": 52084 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly false positives: the cluster mixes unrelated bug fixes, feature additions, docs, and test-only tweaks that merely share broad subsystems or filenames. None of the soft pairs look like the same concrete change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR; the items span unrelated fixes across loaders, pipelines, tests, docs, and infra.", + "best_issue_reason": null, + "best_pr_reason": "PR 11039 is the strongest representative because it is a substantive core loader fix touching multiple code paths, unlike most others here which are test/doc/utility-only or clearly different bugs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7771", + "right": "pull_request:7825", + "accept": false, + "reason": "Different test issues: one marks a flaky UNetControlNetXS test, the other skips offload/device-map tests for models without split modules." + }, + { + "left": "pull_request:1128", + "right": "pull_request:3313", + "accept": false, + "reason": "One adds tensor input support for inpaint; the other fixes torch.compile breaks across ControlNet/UNet code paths." + }, + { + "left": "pull_request:2129", + "right": "pull_request:2286", + "accept": false, + "reason": "Related LoRA/xformers area, but different changes: LoRA pipeline support vs making LoRA work with xformers attention processors." + }, + { + "left": "pull_request:12806", + "right": "pull_request:12812", + "accept": false, + "reason": "Different problems: distributed progress/logging handling vs enabling CP for kernel-based attention backends." + }, + { + "left": "pull_request:11639", + "right": "pull_request:11726", + "accept": false, + "reason": "Different Wan fixes: VAE sampling-mode standardization vs LoRA conversion/loader support." + }, + { + "left": "pull_request:10342", + "right": "pull_request:9060", + "accept": false, + "reason": "Unrelated: dtype inference logic vs removing a stray debug print in transformer_flux." + }, + { + "left": "pull_request:10351", + "right": "pull_request:11252", + "accept": false, + "reason": "Different scope entirely: a LoRA test helper tweak vs fixing a timeout constant in a release-branch utility script." + }, + { + "left": "pull_request:11252", + "right": "pull_request:9676", + "accept": false, + "reason": "Both are small maintenance fixes, but they address unrelated code paths: request timeout vs benchmark import path." + }, + { + "left": "pull_request:1527", + "right": "pull_request:788", + "accept": false, + "reason": "Inpaint tensor-input support vs PIL sampler warning cleanup in img2img/textual-inversion; not the same bug." + }, + { + "left": "pull_request:1968", + "right": "pull_request:3797", + "accept": false, + "reason": "Docs/CPU-offload corrections versus a core attention naming refactor; different underlying issues." + }, + { + "left": "pull_request:4242", + "right": "pull_request:5213", + "accept": false, + "reason": "Different VAE-related bugs: single-file VAE loading support vs scaling-factor handling in latent diffusion." + }, + { + "left": "pull_request:5361", + "right": "pull_request:5852", + "accept": false, + "reason": "PR test workflow fix is unrelated to the PEFT integration test harness changes." + }, + { + "left": "pull_request:6330", + "right": "pull_request:7714", + "accept": false, + "reason": "One fixes SVD docs rendering; the other is an AutoPipeline documentation update." + }, + { + "left": "pull_request:7632", + "right": "pull_request:9506", + "accept": false, + "reason": "DreamBooth VAE loading path fix is unrelated to docs/link cleanup and imageio import changes." + }, + { + "left": "pull_request:11039", + "right": "pull_request:11716", + "accept": false, + "reason": "Both involve loaders, but one changes single-file torch_dtype/quantization behavior while the other fixes LoRA metadata preparation." + }, + { + "left": "pull_request:2452", + "right": "pull_request:750", + "accept": false, + "reason": "Draft docs page addition and LMS timestep back-compat are unrelated." + }, + { + "left": "pull_request:12041", + "right": "pull_request:12521", + "accept": false, + "reason": "Same model family, but different concrete Wan autoencoder bugs: patchify logic vs tiled decode crash handling." + }, + { + "left": "pull_request:10127", + "right": "pull_request:8696", + "accept": false, + "reason": "One xfails incompatible SD config tests; the other marks Marigold depth tests flaky. Different test failures." + }, + { + "left": "pull_request:10127", + "right": "pull_request:8797", + "accept": false, + "reason": "Unrelated test changes: incompatible SD config xfails vs sharding test fixes." + }, + { + "left": "pull_request:12424", + "right": "pull_request:13022", + "accept": false, + "reason": "Both touch Dockerfiles, but they are different infra cleanups with different images and fixes." + }, + { + "left": "pull_request:1283", + "right": "pull_request:1299", + "accept": false, + "reason": "Distinct features: adding Versatile Diffusion versus adding AltDiffusion." + }, + { + "left": "pull_request:5653", + "right": "pull_request:5915", + "accept": false, + "reason": "PEFT-aware maybe_raise_or_warn support is unrelated to IP-Adapter Plus model support." + }, + { + "left": "pull_request:7246", + "right": "pull_request:8696", + "accept": false, + "reason": "VAE tiling test device fix and Marigold flaky-test annotations are unrelated." + }, + { + "left": "pull_request:1355", + "right": "pull_request:1571", + "accept": false, + "reason": "Flax img2img pipeline addition versus SD Img2Img resize-to-multiple-of-8 bug fix are different changes." + }, + { + "left": "pull_request:10351", + "right": "pull_request:11716", + "accept": false, + "reason": "Test cleanup in LoRA utils is unrelated to LoRA loader metadata handling." + }, + { + "left": "pull_request:10086", + "right": "pull_request:8452", + "accept": false, + "reason": "StableDiffusion3 guidance-layer fix and single-file loading strictness are unrelated bugs." + }, + { + "left": "pull_request:10086", + "right": "pull_request:8510", + "accept": false, + "reason": "Different fixes: SD3 skip-guidance-layer handling versus DreamBooth LR scheduler logic." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 141, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10271", + "pull_request:10360", + "pull_request:10547", + "pull_request:11269", + "pull_request:11292", + "pull_request:11999", + "pull_request:12062", + "pull_request:12092", + "pull_request:1229", + "pull_request:12331", + "pull_request:12425", + "pull_request:12491", + "pull_request:130", + "pull_request:13211", + "pull_request:1508", + "pull_request:152", + "pull_request:1527", + "pull_request:2037", + "pull_request:2062", + "pull_request:2279", + "pull_request:2341", + "pull_request:2345", + "pull_request:2415", + "pull_request:2467", + "pull_request:249", + "pull_request:2595", + "pull_request:2655", + "pull_request:2968", + "pull_request:3042", + "pull_request:3536", + "pull_request:3586", + "pull_request:3657", + "pull_request:3708", + "pull_request:3806", + "pull_request:402", + "pull_request:4040", + "pull_request:4187", + "pull_request:4248", + "pull_request:5035", + "pull_request:5053", + "pull_request:5111", + "pull_request:5665", + "pull_request:6261", + "pull_request:676", + "pull_request:7044", + "pull_request:833", + "pull_request:9918" + ], + "soft_pairs": [ + "pull_request:10271|pull_request:10547", + "pull_request:10360|pull_request:10547", + "pull_request:11269|pull_request:9918", + "pull_request:11292|pull_request:11999", + "pull_request:12062|pull_request:12491", + "pull_request:12092|pull_request:12331", + "pull_request:1229|pull_request:130", + "pull_request:12425|pull_request:13211", + "pull_request:1508|pull_request:402", + "pull_request:152|pull_request:1527", + "pull_request:2037|pull_request:3586", + "pull_request:2062|pull_request:3708", + "pull_request:2279|pull_request:249", + "pull_request:2341|pull_request:2655", + "pull_request:2345|pull_request:833", + "pull_request:2415|pull_request:676", + "pull_request:2467|pull_request:4187", + "pull_request:2595|pull_request:3536", + "pull_request:2968|pull_request:5053", + "pull_request:3042|pull_request:5035", + "pull_request:3657|pull_request:5665", + "pull_request:3806|pull_request:5111", + "pull_request:4040|pull_request:4248", + "pull_request:4248|pull_request:5665", + "pull_request:6261|pull_request:7044" + ], + "prepared_review_unit_hash": "82926a2cf352975562b6a2383dc66e6f7a80edb53d7de14bdb0b3b67938f7ca9", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 96561, + "estimated_input_tokens": 24141, + "estimated_eval_tokens": 48538 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These PRs are mostly unrelated, with only superficial overlap from shared subsystems like docs, CI, LoRA, or schedulers. I don\u2019t see any pair that clearly fixes the same concrete bug/change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10271", + "right": "pull_request:10547", + "accept": false, + "reason": "Both touch tests, but they exercise different features: TorchAO serialization refactoring vs offload test parameters for CogVideoX/CogView3." + }, + { + "left": "pull_request:10360", + "right": "pull_request:10547", + "accept": false, + "reason": "Release workflow TestPyPI install fix is unrelated to model offload test tweaks." + }, + { + "left": "pull_request:11269", + "right": "pull_request:9918", + "accept": false, + "reason": "ONNX Runtime package detection changes and the server example are unrelated changes." + }, + { + "left": "pull_request:11292", + "right": "pull_request:11999", + "accept": false, + "reason": "Both are LoRA-related, but one expands SDXL LoRA mapping support while the other adapts to a new PEFT injection API; different code paths and fixes." + }, + { + "left": "pull_request:12062", + "right": "pull_request:12491", + "accept": false, + "reason": "CI GPU container settings and documentation link checking are separate workflow changes." + }, + { + "left": "pull_request:12092", + "right": "pull_request:12331", + "accept": false, + "reason": "GGUF documentation additions and parallel-inference docs updates are different documentation topics." + }, + { + "left": "pull_request:1229", + "right": "pull_request:130", + "accept": false, + "reason": "DreamBooth README wording and the accelerate config note are separate example-doc edits." + }, + { + "left": "pull_request:12425", + "right": "pull_request:13211", + "accept": false, + "reason": "torch.distributed import guarding is unrelated to the group-offloading runtime bug fix." + }, + { + "left": "pull_request:1508", + "right": "pull_request:402", + "accept": false, + "reason": "DreamBooth class-image clarification and the intro docs rewrite are unrelated documentation changes." + }, + { + "left": "pull_request:152", + "right": "pull_request:1527", + "accept": false, + "reason": "Local-folder training support and tensor image/mask input support solve different example/pipeline behaviors." + }, + { + "left": "pull_request:2037", + "right": "pull_request:3586", + "accept": false, + "reason": "Both add community examples, but they are different pipelines/features and not the same underlying change." + }, + { + "left": "pull_request:2062", + "right": "pull_request:3708", + "accept": false, + "reason": "Paint-by-example CPU offload and custom release sorting are unrelated fixes." + }, + { + "left": "pull_request:2279", + "right": "pull_request:249", + "accept": false, + "reason": "Both involve tooling, but one replaces flake8 with ruff while the other pins black; not the same concrete change." + }, + { + "left": "pull_request:2341", + "right": "pull_request:2655", + "accept": false, + "reason": "EMA checkpoint saving and LoRA load/save behavior are different training vs loader issues." + }, + { + "left": "pull_request:2345", + "right": "pull_request:833", + "accept": false, + "reason": "xFormers training warning and the 0.5.1 release bump are unrelated." + }, + { + "left": "pull_request:2415", + "right": "pull_request:676", + "accept": false, + "reason": "Pipeline version-check logic and a resnet variable rename are not the same bug/change." + }, + { + "left": "pull_request:2467", + "right": "pull_request:4187", + "accept": false, + "reason": "Latent output support and DPMSolver index handling are unrelated scheduler/pipeline changes." + }, + { + "left": "pull_request:2595", + "right": "pull_request:3536", + "accept": false, + "reason": "Scheduler config switching and IF super-resolution normalization are different code paths." + }, + { + "left": "pull_request:2968", + "right": "pull_request:5053", + "accept": false, + "reason": "Post-processing control flow and the validation_image CLI argument are unrelated." + }, + { + "left": "pull_request:3042", + "right": "pull_request:5035", + "accept": false, + "reason": "They both adjust dependency/import handling, but for different packages and problems; not the same underlying fix." + }, + { + "left": "pull_request:3657", + "right": "pull_request:5665", + "accept": false, + "reason": "ControlNet checkpoint parsing tweaks and image preprocessing size propagation are distinct issues." + }, + { + "left": "pull_request:3806", + "right": "pull_request:5111", + "accept": false, + "reason": "Checkpoint loading docs and Karras sigma scheduler support are unrelated." + }, + { + "left": "pull_request:4040", + "right": "pull_request:4248", + "accept": false, + "reason": "A config default-type fix and removing a list from tracker config are different configuration-related bugs." + }, + { + "left": "pull_request:4248", + "right": "pull_request:5665", + "accept": false, + "reason": "Accelerator tracker config cleanup and ControlNet image preprocessing are unrelated changes." + }, + { + "left": "pull_request:6261", + "right": "pull_request:7044", + "accept": false, + "reason": "Moving Attend-and-Excite out of stable_diffusion and re-adding a UNet refactor are different refactorings, not one merged change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 142, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10105", + "pull_request:10122", + "pull_request:10318", + "pull_request:10342", + "pull_request:10454", + "pull_request:10983", + "pull_request:11044", + "pull_request:11269", + "pull_request:11290", + "pull_request:113", + "pull_request:11366", + "pull_request:11535", + "pull_request:11571", + "pull_request:11994", + "pull_request:12005", + "pull_request:12021", + "pull_request:12092", + "pull_request:12360", + "pull_request:12424", + "pull_request:12450", + "pull_request:12451", + "pull_request:12467", + "pull_request:12487", + "pull_request:12552", + "pull_request:12586", + "pull_request:12797", + "pull_request:12960", + "pull_request:13120", + "pull_request:13313", + "pull_request:13365", + "pull_request:1982", + "pull_request:4276", + "pull_request:4987", + "pull_request:5010", + "pull_request:6241", + "pull_request:6426", + "pull_request:6886", + "pull_request:7116", + "pull_request:7216", + "pull_request:7455", + "pull_request:7679", + "pull_request:8522", + "pull_request:8563", + "pull_request:8631", + "pull_request:8719", + "pull_request:9473", + "pull_request:9601", + "pull_request:9940" + ], + "soft_pairs": [ + "pull_request:6426|pull_request:8719", + "pull_request:6886|pull_request:7455", + "pull_request:7116|pull_request:7679", + "pull_request:7216|pull_request:8631", + "pull_request:8522|pull_request:8563", + "pull_request:8563|pull_request:9601", + "pull_request:10122|pull_request:9473", + "pull_request:4276|pull_request:4987", + "pull_request:11290|pull_request:12424", + "pull_request:11994|pull_request:12960", + "pull_request:10105|pull_request:9940", + "pull_request:5010|pull_request:6241", + "pull_request:11571|pull_request:12021", + "pull_request:10454|pull_request:10983", + "pull_request:10318|pull_request:11269", + "pull_request:10342|pull_request:11366", + "pull_request:11044|pull_request:11535", + "pull_request:113|pull_request:1982", + "pull_request:12005|pull_request:12092", + "pull_request:12360|pull_request:12467", + "pull_request:12450|pull_request:13365", + "pull_request:12451|pull_request:12586", + "pull_request:12487|pull_request:12797", + "pull_request:12552|pull_request:12797", + "pull_request:13120|pull_request:13313" + ], + "prepared_review_unit_hash": "1c0a1a62951b34119f1ce68cd4fb864ca13989d968126f128f76a883bb4b3025", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 95462, + "estimated_input_tokens": 23866, + "estimated_eval_tokens": 47988 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the suggested soft matches are mostly superficial title or subsystem similarities, not the same underlying bug/change. Reject all soft edges.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents the cluster; the artifacts span unrelated docs, CI, loader, model, and example fixes.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global representative because this is not a true duplicate set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6426", + "right": "pull_request:8719", + "accept": false, + "reason": "Both mention deprecation cleanup, but one is broad LoRA/API deprecation handling and the other is a VQ model output doc/type cleanup; different changes." + }, + { + "left": "pull_request:6886", + "right": "pull_request:7455", + "accept": false, + "reason": "One removes deprecated `torch_dtype` usage in scripts; the other removes `distutils` usage. Same style of cleanup, different code paths." + }, + { + "left": "pull_request:7116", + "right": "pull_request:7679", + "accept": false, + "reason": "Both touch CI installs, but one is a workflow-wide uv migration and the other is a specific temporary pip fallback for peft; not the same fix." + }, + { + "left": "pull_request:7216", + "right": "pull_request:8631", + "accept": false, + "reason": "Both relate to single-file loading, but one updates the public signature while the other fixes a T5 checkpoint return bug; different underlying changes." + }, + { + "left": "pull_request:8522", + "right": "pull_request:8563", + "accept": false, + "reason": "Both update dependency pins, but they are separate version bumps with different scope and rationale." + }, + { + "left": "pull_request:8563", + "right": "pull_request:9601", + "accept": false, + "reason": "Dependency version pinning and SD3 VAE dtype handling are unrelated problems." + }, + { + "left": "pull_request:10122", + "right": "pull_request:9473", + "accept": false, + "reason": "Flux attention-mask wiring and Flux DreamBooth learning-rate logic are different code-path fixes." + }, + { + "left": "pull_request:4276", + "right": "pull_request:4987", + "accept": false, + "reason": "One adds import tests; the other removes a duplicate GLIGEN export. No shared bug." + }, + { + "left": "pull_request:11290", + "right": "pull_request:12424", + "accept": false, + "reason": "A Dockerfile torch-pin change and a broad Dockerfile definition cleanup are not the same change." + }, + { + "left": "pull_request:11994", + "right": "pull_request:12960", + "accept": false, + "reason": "Hub login/docs updates and `is_fsdp` detection in training scripts are unrelated." + }, + { + "left": "pull_request:10105", + "right": "pull_request:9940", + "accept": false, + "reason": "Both are docs-oriented, but one fixes a broken link and the other updates textual inversion docs/example wording; not the same issue." + }, + { + "left": "pull_request:5010", + "right": "pull_request:6241", + "accept": false, + "reason": "These are unrelated: one is a GLIGEN import/export fix, the other is a runtime dtype mismatch report with no matching patch shown." + }, + { + "left": "pull_request:11571", + "right": "pull_request:12021", + "accept": false, + "reason": "LTX docs improvements and LTX attention backend support are distinct changes." + }, + { + "left": "pull_request:10454", + "right": "pull_request:10983", + "accept": false, + "reason": "Both touch Hunyuan video, but 10454 fixes attention-mask dimensions while 10983 is a much larger feature/enablement PR; not mergeable as one duplicate fix." + }, + { + "left": "pull_request:10318", + "right": "pull_request:11269", + "accept": false, + "reason": "Flux LoRA integration tests and ONNX Runtime backend support are unrelated." + }, + { + "left": "pull_request:10342", + "right": "pull_request:11366", + "accept": false, + "reason": "`get_parameter_dtype` behavior and Hidream transformer outputs are unrelated code paths." + }, + { + "left": "pull_request:11044", + "right": "pull_request:11535", + "accept": false, + "reason": "LoRA no-op warning changes and Framepack transformer initialization order are different changes." + }, + { + "left": "pull_request:113", + "right": "pull_request:1982", + "accept": false, + "reason": "An SDE-VE scheduler init fix and an example test path update are unrelated." + }, + { + "left": "pull_request:12005", + "right": "pull_request:12092", + "accept": false, + "reason": "Sharded-checkpoint `local_files_only` behavior and GGUF documentation are unrelated." + }, + { + "left": "pull_request:12360", + "right": "pull_request:12467", + "accept": false, + "reason": "Chroma docs wording and QwenImage batch handling are unrelated." + }, + { + "left": "pull_request:12450", + "right": "pull_request:13365", + "accept": false, + "reason": "CI installation changes and Claude review fork support are different workflow problems." + }, + { + "left": "pull_request:12451", + "right": "pull_request:12586", + "accept": false, + "reason": "One adds doc-builder dependencies; the other just reorders a toctree entry." + }, + { + "left": "pull_request:12487", + "right": "pull_request:12797", + "accept": false, + "reason": "Both are doc fixes, but they target different models and different link/wording issues." + }, + { + "left": "pull_request:12552", + "right": "pull_request:12797", + "accept": false, + "reason": "Kandinsky5 typo fixes and broader Kandinsky doc tweaks are not the same underlying change." + }, + { + "left": "pull_request:13120", + "right": "pull_request:13313", + "accept": false, + "reason": "Modular pipeline tests and LoRA conversion utilities are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 143, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:12714", + "pull_request:13321", + "pull_request:1513", + "pull_request:1913", + "pull_request:1982", + "pull_request:2209", + "pull_request:2279", + "pull_request:230", + "pull_request:2327", + "pull_request:2365", + "pull_request:2415", + "pull_request:2470", + "pull_request:2487", + "pull_request:2553", + "pull_request:2894", + "pull_request:2958", + "pull_request:2970", + "pull_request:3054", + "pull_request:326", + "pull_request:3471", + "pull_request:3656", + "pull_request:3681", + "pull_request:3806", + "pull_request:3957", + "pull_request:402", + "pull_request:4078", + "pull_request:4262", + "pull_request:4669", + "pull_request:5057", + "pull_request:5220", + "pull_request:538", + "pull_request:5478", + "pull_request:5493", + "pull_request:5698", + "pull_request:5702", + "pull_request:5724", + "pull_request:5946", + "pull_request:6163", + "pull_request:6215", + "pull_request:6340", + "pull_request:6447", + "pull_request:6448", + "pull_request:6643", + "pull_request:7024", + "pull_request:8438", + "pull_request:8510", + "pull_request:8571", + "pull_request:980" + ], + "soft_pairs": [ + "pull_request:1513|pull_request:402", + "pull_request:1913|pull_request:326", + "pull_request:1982|pull_request:2209", + "pull_request:2279|pull_request:538", + "pull_request:230|pull_request:980", + "pull_request:2327|pull_request:3681", + "pull_request:2365|pull_request:3471", + "pull_request:2365|pull_request:3806", + "pull_request:2415|pull_request:3054", + "pull_request:2470|pull_request:4262", + "pull_request:2487|pull_request:2553", + "pull_request:2894|pull_request:2958", + "pull_request:2970|pull_request:3957", + "pull_request:3656|pull_request:5220", + "pull_request:3681|pull_request:4078", + "pull_request:4669|pull_request:6448", + "pull_request:5057|pull_request:6447", + "pull_request:5478|pull_request:6643", + "pull_request:5493|pull_request:5946", + "pull_request:5698|pull_request:5702", + "pull_request:5724|pull_request:6163", + "pull_request:6215|pull_request:8510", + "pull_request:6340|pull_request:7024", + "pull_request:8438|pull_request:8571", + "pull_request:12714|pull_request:13321" + ], + "prepared_review_unit_hash": "7a77e4ba17a0c90cba22e634e2754aacc1d740e4f5b3c347b83d87c90b3f5c0a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 102788, + "estimated_input_tokens": 25697, + "estimated_eval_tokens": 51650 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These PRs are only superficially similar; the soft edges mostly connect unrelated docs, tests, dependency, and pipeline changes, so none should be deduplicated.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical duplicate target; the candidate pairs do not share the same underlying bug or change, only broad keyword overlap.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong global best candidate because the set is heterogeneous and the pairings are not true duplicates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1513", + "right": "pull_request:402", + "accept": false, + "reason": "Both are docs-related, but one adds notebook setup guidance while the other restructures the intro docs tree; different content and purpose." + }, + { + "left": "pull_request:1913", + "right": "pull_request:326", + "accept": false, + "reason": "One validates img2img input types in specific pipelines; the other is a broad typing/UNet API PR across many files, not the same fix." + }, + { + "left": "pull_request:1982", + "right": "pull_request:2209", + "accept": false, + "reason": "Example test path correction vs. an image-to-image docs argument-name fix; unrelated changes." + }, + { + "left": "pull_request:2279", + "right": "pull_request:538", + "accept": false, + "reason": "CI/lint tool migration is unrelated to a package version bump." + }, + { + "left": "pull_request:230", + "right": "pull_request:980", + "accept": false, + "reason": "Both touch fp16/dtype themes, but one changes loading APIs and the other adjusts Dance Diffusion pipeline/runtime dtype handling." + }, + { + "left": "pull_request:2327", + "right": "pull_request:3681", + "accept": false, + "reason": "UNet constructor validation is unrelated to checkpoint limit handling in training scripts." + }, + { + "left": "pull_request:2365", + "right": "pull_request:3471", + "accept": false, + "reason": "A scheduler-type note in code is not the same as a DreamBooth docs note about torch.compile." + }, + { + "left": "pull_request:2365", + "right": "pull_request:3806", + "accept": false, + "reason": "Different docs topics: scheduler typing vs. checkpoint loading guidance." + }, + { + "left": "pull_request:2415", + "right": "pull_request:3054", + "accept": false, + "reason": "Pipeline loader compatibility logic and a docs wording cleanup are unrelated." + }, + { + "left": "pull_request:2470", + "right": "pull_request:4262", + "accept": false, + "reason": "Both are test-oriented, but one is an MPS skip fix and the other is SDXL DreamBooth LoRA training changes." + }, + { + "left": "pull_request:2487", + "right": "pull_request:2553", + "accept": false, + "reason": "A JAX/Flax docs addition is unrelated to adding a CFG flag in Flax pipelines." + }, + { + "left": "pull_request:2894", + "right": "pull_request:2958", + "accept": false, + "reason": "A requirements file addition and a new ControlNet Inpaint feature are different changes." + }, + { + "left": "pull_request:2970", + "right": "pull_request:3957", + "accept": false, + "reason": "Variable rename in one DreamBooth script does not match the attn-name correction WIP." + }, + { + "left": "pull_request:3656", + "right": "pull_request:5220", + "accept": false, + "reason": "Core activation refactor and moving tests to nightly are unrelated." + }, + { + "left": "pull_request:3681", + "right": "pull_request:4078", + "accept": false, + "reason": "Training checkpoint-limit logic and from_single_file accelerate gating are separate code paths." + }, + { + "left": "pull_request:4669", + "right": "pull_request:6448", + "accept": false, + "reason": "LoRA rank handling in loaders is not the same as LoRA test assertion cleanup." + }, + { + "left": "pull_request:5057", + "right": "pull_request:6447", + "accept": false, + "reason": "SDXL clip-skip support and a K-Diffusion pipeline addition are distinct features." + }, + { + "left": "pull_request:5478", + "right": "pull_request:6643", + "accept": false, + "reason": "Japanese docs rollout and UViT2D docs are separate documentation additions." + }, + { + "left": "pull_request:5493", + "right": "pull_request:5946", + "accept": false, + "reason": "A core activation/normalization refactor is not the same as refactoring example tests." + }, + { + "left": "pull_request:5698", + "right": "pull_request:5702", + "accept": false, + "reason": "LCM img2img bugfix and AnimateDiff prompt bugfix affect different pipelines and failure modes." + }, + { + "left": "pull_request:5724", + "right": "pull_request:6163", + "accept": false, + "reason": "Both are LoRA test-related, but one adds backend gating and the other tweaks test coverage/behavior; not the same underlying fix." + }, + { + "left": "pull_request:6215", + "right": "pull_request:8510", + "accept": false, + "reason": "PixArt test initialization fix and DreamBooth LoRA scheduler fixes are unrelated." + }, + { + "left": "pull_request:6340", + "right": "pull_request:7024", + "accept": false, + "reason": "AnimateDiff conversion scripts and AnimateLCM docs are different artifacts and changes." + }, + { + "left": "pull_request:8438", + "right": "pull_request:8571", + "accept": false, + "reason": "Both are broad errata/style PRs, but they touch different files and distinct content edits; not one underlying change." + }, + { + "left": "pull_request:12714", + "right": "pull_request:13321", + "accept": false, + "reason": "Flux2 LoRA NaN test fixes and an unguarded torchvision import fix in Cosmos are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 144, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10114", + "pull_request:10188", + "pull_request:10200", + "pull_request:10245", + "pull_request:10431", + "pull_request:10538", + "pull_request:10539", + "pull_request:11238", + "pull_request:1137", + "pull_request:11479", + "pull_request:11522", + "pull_request:11614", + "pull_request:11814", + "pull_request:12363", + "pull_request:12424", + "pull_request:12438", + "pull_request:12490", + "pull_request:12491", + "pull_request:12960", + "pull_request:13081", + "pull_request:13134", + "pull_request:13330", + "pull_request:1338", + "pull_request:13410", + "pull_request:13472", + "pull_request:1508", + "pull_request:152", + "pull_request:1525", + "pull_request:1527", + "pull_request:1706", + "pull_request:1788", + "pull_request:2415", + "pull_request:2902", + "pull_request:3292", + "pull_request:3473", + "pull_request:3536", + "pull_request:3708", + "pull_request:4245", + "pull_request:4278", + "pull_request:453", + "pull_request:51", + "pull_request:748", + "pull_request:8425", + "pull_request:8591", + "pull_request:8798", + "pull_request:9592", + "pull_request:9593", + "pull_request:9676" + ], + "soft_pairs": [ + "pull_request:2902|pull_request:4278", + "pull_request:11522|pull_request:12438", + "pull_request:12424|pull_request:13134", + "pull_request:12490|pull_request:12960", + "pull_request:10114|pull_request:8425", + "pull_request:13410|pull_request:13472", + "pull_request:3473|pull_request:4245", + "pull_request:10188|pull_request:8798", + "pull_request:10200|pull_request:8591", + "pull_request:10245|pull_request:9592", + "pull_request:10245|pull_request:9593", + "pull_request:10431|pull_request:10538", + "pull_request:10539|pull_request:11814", + "pull_request:11238|pull_request:9676", + "pull_request:1137|pull_request:51", + "pull_request:11479|pull_request:12363", + "pull_request:11614|pull_request:12491", + "pull_request:13081|pull_request:13330", + "pull_request:1338|pull_request:748", + "pull_request:1508|pull_request:152", + "pull_request:1525|pull_request:453", + "pull_request:1527|pull_request:3292", + "pull_request:1527|pull_request:3536", + "pull_request:1706|pull_request:2415", + "pull_request:1788|pull_request:3708" + ], + "prepared_review_unit_hash": "89443ecaf70c628ef89400b1ee95ee86f57ef633c2a1e463cda14eb545f8a9d4", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 91971, + "estimated_input_tokens": 22993, + "estimated_eval_tokens": 46242 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster does not contain a real duplicate set; the soft pairs are mostly unrelated formatting, docs, CI, or small fixes in different code paths. None of the candidate pairs looks mergeable as one PR or the same underlying bug/change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:2902", + "right": "pull_request:4278", + "accept": false, + "reason": "Both touch image pipelines, but they fix different bugs in different classes and inputs; not the same concrete code-path issue." + }, + { + "left": "pull_request:11522", + "right": "pull_request:12438", + "accept": false, + "reason": "Both are version/compatibility fixes around loading, but one is AudioLDM2 transformers handling and the other is generic offload_state_dict cleanup; different failure modes." + }, + { + "left": "pull_request:12424", + "right": "pull_request:13134", + "accept": false, + "reason": "Unrelated scopes: Dockerfile/workflow cleanup versus Cosmos transformer typing; no shared bug or change." + }, + { + "left": "pull_request:12490", + "right": "pull_request:12960", + "accept": false, + "reason": "Audioldm2 test fixture change versus Flux FSDP check; both are tests/examples but not the same issue." + }, + { + "left": "pull_request:10114", + "right": "pull_request:8425", + "accept": false, + "reason": "Stable Audio formatting fix and workflow name rename are unrelated cosmetic edits." + }, + { + "left": "pull_request:13410", + "right": "pull_request:13472", + "accept": false, + "reason": "Different documentation updates: modular-pipeline guidance versus float64 gotcha; not duplicate content." + }, + { + "left": "pull_request:3473", + "right": "pull_request:4245", + "accept": false, + "reason": "Both are docs, but one adds textual inversion inference and the other adds safetensors examples across many docs; different changes." + }, + { + "left": "pull_request:10188", + "right": "pull_request:8798", + "accept": false, + "reason": "Both concern loading behavior, but one is LoRA expansion error handling and the other is sharded checkpoint subfolder loading; distinct bugs." + }, + { + "left": "pull_request:10200", + "right": "pull_request:8591", + "accept": false, + "reason": "ControlNet Union refactor versus SD3 training density/weighting utilities; different subsystems and purposes." + }, + { + "left": "pull_request:10245", + "right": "pull_request:9592", + "accept": false, + "reason": "PEFT docs tweak versus Flax embedding docstring refactor; not the same underlying change." + }, + { + "left": "pull_request:10245", + "right": "pull_request:9593", + "accept": false, + "reason": "PEFT docs tweak versus community docstring cleanup; unrelated documentation work." + }, + { + "left": "pull_request:10431", + "right": "pull_request:10538", + "accept": false, + "reason": "Sana transformer bug fix versus auto-pipeline registration for Sana; one fixes model internals, the other pipeline discovery." + }, + { + "left": "pull_request:10539", + "right": "pull_request:11814", + "accept": false, + "reason": "Flux true-CFG condition fix versus gradient-checkpointing kwargs propagation; same model family but different code paths and bugs." + }, + { + "left": "pull_request:11238", + "right": "pull_request:9676", + "accept": false, + "reason": "SD3 ControlNet validation/runtime fix versus benchmark import cleanup; unrelated." + }, + { + "left": "pull_request:1137", + "right": "pull_request:51", + "accept": false, + "reason": "Different eras and bugs: dtype cast in safety checker versus install_requires typo." + }, + { + "left": "pull_request:11479", + "right": "pull_request:12363", + "accept": false, + "reason": "Intel Gaudi doc rename versus QwenImage Edit Plus docs/tests; no common issue." + }, + { + "left": "pull_request:11614", + "right": "pull_request:12491", + "accept": false, + "reason": "Docker build script security cleanup versus documentation link-check CI; both CI-related but not same change." + }, + { + "left": "pull_request:13081", + "right": "pull_request:13330", + "accept": false, + "reason": "Qwen hidden-state contiguity fix versus torchao test backend updates; one is a model bug, the other is test infrastructure." + }, + { + "left": "pull_request:1338", + "right": "pull_request:748", + "accept": false, + "reason": "ONNX inpainting cast order fix versus push_to_hub API cleanup; unrelated pipeline bug and training utility fix." + }, + { + "left": "pull_request:1508", + "right": "pull_request:152", + "accept": false, + "reason": "DreamBooth prior-preservation docs clarification versus local image-folder training support; docs vs feature change." + }, + { + "left": "pull_request:1525", + "right": "pull_request:453", + "accept": false, + "reason": "Upscaling batch-size fix versus installation typo; completely unrelated." + }, + { + "left": "pull_request:1527", + "right": "pull_request:3292", + "accept": false, + "reason": "Stable Diffusion inpaint tensor-input support versus input perturbation option in text-to-image training; different pipelines and behavior." + }, + { + "left": "pull_request:1527", + "right": "pull_request:3536", + "accept": false, + "reason": "Stable Diffusion inpaint tensor-input support versus DeepFloyd IF PIL normalization; different pipeline families and bugs." + }, + { + "left": "pull_request:1706", + "right": "pull_request:2415", + "accept": false, + "reason": "Main docs update versus pipeline_utils version handling; unrelated documentation and code maintenance." + }, + { + "left": "pull_request:1788", + "right": "pull_request:3708", + "accept": false, + "reason": "DDIM prediction_type logic bug versus custom release sorting/version parsing; different modules and failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 145, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10136", + "pull_request:10359", + "pull_request:10567", + "pull_request:10810", + "pull_request:10985", + "pull_request:11469", + "pull_request:11999", + "pull_request:12151", + "pull_request:2655", + "pull_request:3473", + "pull_request:3589", + "pull_request:3656", + "pull_request:3723", + "pull_request:3797", + "pull_request:3850", + "pull_request:3957", + "pull_request:4070", + "pull_request:4451", + "pull_request:449", + "pull_request:4749", + "pull_request:5506", + "pull_request:5747", + "pull_request:5948", + "pull_request:6150", + "pull_request:6195", + "pull_request:6215", + "pull_request:6373", + "pull_request:6426", + "pull_request:6564", + "pull_request:676", + "pull_request:6886", + "pull_request:6961", + "pull_request:6972", + "pull_request:6994", + "pull_request:7202", + "pull_request:7216", + "pull_request:7410", + "pull_request:7458", + "pull_request:7634", + "pull_request:7662", + "pull_request:8425", + "pull_request:8512", + "pull_request:8618", + "pull_request:8627", + "pull_request:8755", + "pull_request:8761", + "pull_request:9307" + ], + "soft_pairs": [ + "pull_request:2655|pull_request:3797", + "pull_request:3473|pull_request:4070", + "pull_request:3589|pull_request:3723", + "pull_request:3656|pull_request:4451", + "pull_request:3850|pull_request:5948", + "pull_request:3957|pull_request:6195", + "pull_request:4749|pull_request:6972", + "pull_request:5506|pull_request:7458", + "pull_request:5747|pull_request:7634", + "pull_request:6150|pull_request:6564", + "pull_request:6215|pull_request:8425", + "pull_request:6373|pull_request:7202", + "pull_request:6426|pull_request:6886", + "pull_request:6961|pull_request:8627", + "pull_request:6994|pull_request:7216", + "pull_request:7410|pull_request:8761", + "pull_request:7458|pull_request:9307", + "pull_request:7662|pull_request:8512", + "pull_request:8618|pull_request:8755", + "pull_request:10359|pull_request:11469", + "pull_request:449|pull_request:676", + "pull_request:10136|pull_request:10567", + "pull_request:10810|pull_request:12151", + "pull_request:10985|pull_request:11999" + ], + "prepared_review_unit_hash": "4f180f56ec9342484c11e6598904ad2c7f38f7363988362e9a84b26d9a1d3b99", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 100013, + "estimated_input_tokens": 25004, + "estimated_eval_tokens": 50264 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a broad false-positive cluster: the PRs span unrelated docs, tests, workflows, LoRA, Hunyuan Video, IP-Adapter, and single-file loader work. None of the soft edges looks like the same underlying change or bug fix.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: there isn\u2019t a single duplicate thread here, just many unrelated PRs that happen to share broad subsystem keywords.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global representative because the cluster is not coherent; PR 10136 is large, but it is still unrelated to most others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2655", + "right": "pull_request:3797", + "accept": false, + "reason": "LoRA save/load fix vs attention-head naming correction across model configs; different bugs and code paths." + }, + { + "left": "pull_request:3473", + "right": "pull_request:4070", + "accept": false, + "reason": "Textual inversion docs vs SDXL docs wording cleanup; both are documentation, but not the same change." + }, + { + "left": "pull_request:3589", + "right": "pull_request:3723", + "accept": false, + "reason": "Community README notes vs adding np.ndarray support to reference pipelines; same example area, different concrete changes." + }, + { + "left": "pull_request:3656", + "right": "pull_request:4451", + "accept": false, + "reason": "Activation helper refactor vs moving tests to nightly; unrelated refactor and test-scheduling change." + }, + { + "left": "pull_request:3850", + "right": "pull_request:5948", + "accept": false, + "reason": "AudioDiffusion test fixture fix vs variant-loading warning tweak; no shared bug or code path." + }, + { + "left": "pull_request:3957", + "right": "pull_request:6195", + "accept": false, + "reason": "WIP attention-head correction vs AnimateDiff SDXL feature work; shared file path is not enough to merge them." + }, + { + "left": "pull_request:4749", + "right": "pull_request:6972", + "accept": false, + "reason": "SDXL LoRA checkpointing vs IP-Adapter unload test; different components and failure modes." + }, + { + "left": "pull_request:5506", + "right": "pull_request:7458", + "accept": false, + "reason": "PEFT/xformers LoRA bugfix vs skipping dynamo tests on Python 3.12; unrelated." + }, + { + "left": "pull_request:5747", + "right": "pull_request:7634", + "accept": false, + "reason": "Dataset data_dir plumbing vs high-res generation model addition; different features." + }, + { + "left": "pull_request:6150", + "right": "pull_request:6564", + "accept": false, + "reason": "IP-Adapter converter method vs path handling in image-encoder loading; same loader family, but different concrete fixes." + }, + { + "left": "pull_request:6215", + "right": "pull_request:8425", + "accept": false, + "reason": "PixArt test determinism fix vs workflow name rename; unrelated." + }, + { + "left": "pull_request:6373", + "right": "pull_request:7202", + "accept": false, + "reason": "Deleting doc-comment workflows vs removing huggingface-cli from example requirements; no common bug." + }, + { + "left": "pull_request:6426", + "right": "pull_request:6886", + "accept": false, + "reason": "LoRA deprecation cleanup vs removing torch_dtype deprecation in to(); different deprecation paths." + }, + { + "left": "pull_request:6961", + "right": "pull_request:8627", + "accept": false, + "reason": "Single-file loader test additions vs SD3 ControlNet test-size reduction; both tests, but different targets and issues." + }, + { + "left": "pull_request:6994", + "right": "pull_request:7216", + "accept": false, + "reason": "Passing torch_dtype through single-file loaders vs updating from_single_file signature/docs; related area, but not the same concrete fix." + }, + { + "left": "pull_request:7410", + "right": "pull_request:8761", + "accept": false, + "reason": "FreeInit behavior fix vs Hunyuan embedding refactor; unrelated pipeline internals." + }, + { + "left": "pull_request:7458", + "right": "pull_request:9307", + "accept": false, + "reason": "Skipping dynamo tests on Python 3.12 vs rotary embedding CPU/device refactor; different problems." + }, + { + "left": "pull_request:7662", + "right": "pull_request:8512", + "accept": false, + "reason": "Flax CI cleanup vs removing deprecated prepare_mask_and_masked_image; unrelated workflows/code paths." + }, + { + "left": "pull_request:8618", + "right": "pull_request:8755", + "accept": false, + "reason": "SD3 tiny autoencoder shift_factor addition vs SD3 DreamBooth LoRA training fix; same product family, different components." + }, + { + "left": "pull_request:10359", + "right": "pull_request:11469", + "accept": false, + "reason": "General test fixes vs targeted xfails for specific pipeline methods; not the same test failure." + }, + { + "left": "pull_request:449", + "right": "pull_request:676", + "accept": false, + "reason": "Variable renaming in attention vs variable renaming in resnet; stylistically similar, but separate refactors." + }, + { + "left": "pull_request:10136", + "right": "pull_request:10567", + "accept": false, + "reason": "Hunyuan Video core model/pipeline addition vs adding true CFG support to that pipeline; same module, different feature." + }, + { + "left": "pull_request:10810", + "right": "pull_request:12151", + "accept": false, + "reason": "Flux LoRA conversion support vs deprecating Flax support; completely different subsystems." + }, + { + "left": "pull_request:10985", + "right": "pull_request:11999", + "accept": false, + "reason": "Flux ComfyUI LoRA conversion support vs new LoRA config injection method; related theme, but not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 146, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10119", + "pull_request:10246", + "pull_request:10418", + "pull_request:10808", + "pull_request:11089", + "pull_request:11250", + "pull_request:11366", + "pull_request:11421", + "pull_request:11535", + "pull_request:11896", + "pull_request:12142", + "pull_request:12143", + "pull_request:1513", + "pull_request:152", + "pull_request:1913", + "pull_request:2398", + "pull_request:2641", + "pull_request:3042", + "pull_request:327", + "pull_request:3410", + "pull_request:3681", + "pull_request:3707", + "pull_request:3846", + "pull_request:4622", + "pull_request:4635", + "pull_request:4695", + "pull_request:5053", + "pull_request:5057", + "pull_request:5438", + "pull_request:5439", + "pull_request:5477", + "pull_request:5713", + "pull_request:5823", + "pull_request:5928", + "pull_request:6225", + "pull_request:6289", + "pull_request:6328", + "pull_request:6698", + "pull_request:6743", + "pull_request:6828", + "pull_request:6918", + "pull_request:7058", + "pull_request:7246", + "pull_request:7771", + "pull_request:9212", + "pull_request:9234", + "pull_request:9909" + ], + "soft_pairs": [ + "pull_request:3410|pull_request:3846", + "pull_request:4622|pull_request:4635", + "pull_request:5053|pull_request:5439", + "pull_request:5823|pull_request:7058", + "pull_request:5928|pull_request:6328", + "pull_request:6289|pull_request:6698", + "pull_request:9212|pull_request:9234", + "pull_request:11421|pull_request:12142", + "pull_request:11421|pull_request:12143", + "pull_request:1513|pull_request:152", + "pull_request:1913|pull_request:327", + "pull_request:2398|pull_request:2641", + "pull_request:2641|pull_request:3707", + "pull_request:3042|pull_request:3681", + "pull_request:6225|pull_request:6743", + "pull_request:6743|pull_request:6918", + "pull_request:5438|pull_request:5477", + "pull_request:7246|pull_request:7771", + "pull_request:10808|pull_request:11896", + "pull_request:4695|pull_request:5057", + "pull_request:5713|pull_request:6828", + "pull_request:10119|pull_request:11089", + "pull_request:10246|pull_request:9909", + "pull_request:10418|pull_request:11250", + "pull_request:11366|pull_request:11535" + ], + "prepared_review_unit_hash": "ca41964ca9c5b43d080c7adb9d20aad55e7f21e3a713f34495a948748acbb584", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 99143, + "estimated_input_tokens": 24786, + "estimated_eval_tokens": 49828 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a false-positive similarity group: the PRs span unrelated docs, example scripts, workflows, and separate model/pipeline fixes. None of the suggested soft edges looks like the same underlying change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are too heterogeneous and do not converge on one concrete bug or feature change.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good global representative; the cluster is a mix of unrelated changes, so any one choice would be arbitrary.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3410", + "right": "pull_request:3846", + "accept": false, + "reason": "Both touch training utilities, but one fixes DeepSpeed plugin retrieval and the other changes random-module seeding; different bugs and code paths." + }, + { + "left": "pull_request:4622", + "right": "pull_request:4635", + "accept": false, + "reason": "Docs-only similarity is too broad; PushToHub docs and MultiControlNet docs/example are unrelated changes." + }, + { + "left": "pull_request:5053", + "right": "pull_request:5439", + "accept": false, + "reason": "One fixes an argument parsing interface in a training script, the other fixes tokenizer/config loading in from_single_file; unrelated." + }, + { + "left": "pull_request:5823", + "right": "pull_request:7058", + "accept": false, + "reason": "Both are LoRA-related, but they fix different mechanisms: adapter fusion by names vs copying an input state dict before loading." + }, + { + "left": "pull_request:5928", + "right": "pull_request:6328", + "accept": false, + "reason": "One adds a community README entry, the other adds a new AnimateDiff video-to-video pipeline; not the same concrete change." + }, + { + "left": "pull_request:6289", + "right": "pull_request:6698", + "accept": false, + "reason": "AnimateDiff SparseCntrl and PIA are distinct model/pipeline additions with different architectures and goals." + }, + { + "left": "pull_request:9212", + "right": "pull_request:9234", + "accept": false, + "reason": "Kolors img2img auto-pipeline mapping and safetensors compatibility regression are unrelated code paths." + }, + { + "left": "pull_request:11421", + "right": "pull_request:12142", + "accept": false, + "reason": "Dreambooth interpolation-mode handling and Wan VACE bf16/fp16 mask casting are unrelated fixes." + }, + { + "left": "pull_request:11421", + "right": "pull_request:12143", + "accept": false, + "reason": "These are separate concerns: an interpolation CLI option vs a Wan VACE dtype fix; not the same bug." + }, + { + "left": "pull_request:1513", + "right": "pull_request:152", + "accept": false, + "reason": "Dreambooth notebook setup docs and local-image-folder training support are different documentation/features." + }, + { + "left": "pull_request:1913", + "right": "pull_request:327", + "accept": false, + "reason": "Input-type validation across image pipelines is unrelated to type-hint cleanup for the PNDM pipeline." + }, + { + "left": "pull_request:2398", + "right": "pull_request:2641", + "accept": false, + "reason": "Pix2Pix Zero docs and documentation-build workflow changes are unrelated." + }, + { + "left": "pull_request:2641", + "right": "pull_request:3707", + "accept": false, + "reason": "Both affect docs workflows, but one adds notebook build plumbing and the other changes secret handling/deletion workflow; not the same change." + }, + { + "left": "pull_request:3042", + "right": "pull_request:3681", + "accept": false, + "reason": "Accelerate version pinning and manual checkpoint-limit handling solve different training-script issues." + }, + { + "left": "pull_request:6225", + "right": "pull_request:6743", + "accept": false, + "reason": "LoRA alpha parsing and hflip argument correction are separate training-script bugs." + }, + { + "left": "pull_request:6743", + "right": "pull_request:6918", + "accept": false, + "reason": "A flip-flag fix and hub-token exposure guard are unrelated." + }, + { + "left": "pull_request:5438", + "right": "pull_request:5477", + "accept": false, + "reason": "These are related only at the topic level; one adds a community txt2img entry, the other adds an img2img pipeline implementation." + }, + { + "left": "pull_request:7246", + "right": "pull_request:7771", + "accept": false, + "reason": "One fixes VAE tiling test device usage, the other marks a ControlNetXS test flaky; different test issues." + }, + { + "left": "pull_request:10808", + "right": "pull_request:11896", + "accept": false, + "reason": "Both concern LoRA/offloading, but one restricts PEFT config key updates while the other fixes hook removal during loading; distinct bugs." + }, + { + "left": "pull_request:4695", + "right": "pull_request:5057", + "accept": false, + "reason": "Text-to-Video Zero SDXL support and SDXL clip-skip support are separate features." + }, + { + "left": "pull_request:5713", + "right": "pull_request:6828", + "accept": false, + "reason": "IP-Adapter feature work and a UNetMotionModel IP-adapter code fix are not the same concrete change." + }, + { + "left": "pull_request:10119", + "right": "pull_request:11089", + "accept": false, + "reason": "Docs for load_lora_adapter and Wan memory-optimization docs are unrelated." + }, + { + "left": "pull_request:10246", + "right": "pull_request:9909", + "accept": false, + "reason": "One updates docs for attention processors, the other changes the type union in code; related topic, but not the same PR-level change." + }, + { + "left": "pull_request:10418", + "right": "pull_request:11250", + "accept": false, + "reason": "Internal-link docs cleanup and AutoModel docs addition are unrelated documentation changes." + }, + { + "left": "pull_request:11366", + "right": "pull_request:11535", + "accept": false, + "reason": "Hidream transformer output shape change and Framepack initialization-order change affect different models and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 147, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11588", + "pull_request:11913", + "pull_request:12425", + "pull_request:13022", + "pull_request:1494", + "pull_request:1564", + "pull_request:1788", + "pull_request:1850", + "pull_request:2430", + "pull_request:2464", + "pull_request:2470", + "pull_request:2557", + "pull_request:2593", + "pull_request:2655", + "pull_request:3054", + "pull_request:3241", + "pull_request:3561", + "pull_request:3852", + "pull_request:4215", + "pull_request:4498", + "pull_request:4787", + "pull_request:5220", + "pull_request:5289", + "pull_request:5639", + "pull_request:5946", + "pull_request:6065", + "pull_request:6114", + "pull_request:6147", + "pull_request:6195", + "pull_request:6261", + "pull_request:6340", + "pull_request:6405", + "pull_request:6643", + "pull_request:6762", + "pull_request:6775", + "pull_request:6961", + "pull_request:711", + "pull_request:715", + "pull_request:748", + "pull_request:7627", + "pull_request:7779", + "pull_request:833", + "pull_request:8386", + "pull_request:8445", + "pull_request:8452", + "pull_request:8591", + "pull_request:9447", + "pull_request:9735" + ], + "soft_pairs": [ + "pull_request:12425|pull_request:13022", + "pull_request:1494|pull_request:2655", + "pull_request:1564|pull_request:3561", + "pull_request:1564|pull_request:748", + "pull_request:1788|pull_request:711", + "pull_request:1850|pull_request:2470", + "pull_request:2430|pull_request:715", + "pull_request:2464|pull_request:3054", + "pull_request:2557|pull_request:748", + "pull_request:2593|pull_request:833", + "pull_request:3241|pull_request:5289", + "pull_request:3852|pull_request:4215", + "pull_request:4498|pull_request:6114", + "pull_request:4787|pull_request:6762", + "pull_request:5220|pull_request:6261", + "pull_request:5639|pull_request:6340", + "pull_request:5946|pull_request:6762", + "pull_request:6065|pull_request:6643", + "pull_request:6147|pull_request:8452", + "pull_request:6195|pull_request:6405", + "pull_request:6775|pull_request:7779", + "pull_request:6961|pull_request:8386", + "pull_request:7627|pull_request:9447", + "pull_request:8445|pull_request:9447", + "pull_request:8591|pull_request:9735", + "pull_request:11588|pull_request:11913" + ], + "prepared_review_unit_hash": "dc745efae11436d7b0047c986fd26e6ade5de343a2378559da107f250b07389b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 95419, + "estimated_input_tokens": 23855, + "estimated_eval_tokens": 47966 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated PRs across docs, tests, packaging, and a few feature/fix areas. The soft edges share broad keywords at best, but they do not line up on the same concrete bug or merged change path.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a good canonical duplicate target here; the items span many unrelated changes, so any one PR would be misleading as a representative.", + "best_issue_reason": null, + "best_pr_reason": "PR 11913 is the broadest, most self-contained feature PR in the set (Wan modular diffusers), so it is the least arbitrary single PR to use as a representative if one must be chosen.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12425", + "right": "pull_request:13022", + "accept": false, + "reason": "Both touch import/platform issues, but one is a torch.distributed guard and the other is Dockerfile dependency pinning; unrelated changes." + }, + { + "left": "pull_request:1494", + "right": "pull_request:2655", + "accept": false, + "reason": "Safe serialization support and LoRA loading fixes are different code paths and different problems." + }, + { + "left": "pull_request:1564", + "right": "pull_request:3561", + "accept": false, + "reason": "Docs for accelerate defaults versus Dreambooth IF upscaling logic; not the same bug or change." + }, + { + "left": "pull_request:1564", + "right": "pull_request:748", + "accept": false, + "reason": "One is a docs tweak, the other fixes push_to_hub behavior in training scripts; different scope." + }, + { + "left": "pull_request:1788", + "right": "pull_request:711", + "accept": false, + "reason": "A concrete DDIM sampling bug fix does not match a broad scheduler design proposal." + }, + { + "left": "pull_request:1850", + "right": "pull_request:2470", + "accept": false, + "reason": "Both are tests, but they cover different models and different MPS-related adjustments." + }, + { + "left": "pull_request:2430", + "right": "pull_request:715", + "accept": false, + "reason": "One changes transformers version labeling, the other guards PyTorch imports; unrelated." + }, + { + "left": "pull_request:2464", + "right": "pull_request:3054", + "accept": false, + "reason": "Training-script xformers handling is unrelated to a docs punctuation/contribution edit." + }, + { + "left": "pull_request:2557", + "right": "pull_request:748", + "accept": false, + "reason": "Both involve Dreambooth scripts, but one adds validation logging and the other fixes hub pushing; separate changes." + }, + { + "left": "pull_request:2593", + "right": "pull_request:833", + "accept": false, + "reason": "ControlNet conversion tooling and a version-bump release PR are not duplicates." + }, + { + "left": "pull_request:3241", + "right": "pull_request:5289", + "accept": false, + "reason": "Both are test adjustments, but for different pipelines and different failure modes." + }, + { + "left": "pull_request:3852", + "right": "pull_request:4215", + "accept": false, + "reason": "Different documentation edits in different files with no shared underlying issue." + }, + { + "left": "pull_request:4498", + "right": "pull_request:6114", + "accept": false, + "reason": "Both mention SDXL community examples, but one adds a reference demo while the other adds img2img/inpaint support; different features." + }, + { + "left": "pull_request:4787", + "right": "pull_request:6762", + "accept": false, + "reason": "LoRA loading filename filtering and a flaky Kandinsky test fix are unrelated." + }, + { + "left": "pull_request:5220", + "right": "pull_request:6261", + "accept": false, + "reason": "One retags slow tests, the other refactors pipeline exports; no shared concrete change." + }, + { + "left": "pull_request:5639", + "right": "pull_request:6340", + "accept": false, + "reason": "Motion LoRA docs and AnimateDiff conversion scripts are different artifacts and different tasks." + }, + { + "left": "pull_request:5946", + "right": "pull_request:6762", + "accept": false, + "reason": "A broad example-test refactor is unrelated to a single flaky Kandinsky test adjustment." + }, + { + "left": "pull_request:6065", + "right": "pull_request:6643", + "accept": false, + "reason": "SDXL Turbo docs and UViT2D docs are separate documentation additions." + }, + { + "left": "pull_request:6147", + "right": "pull_request:8452", + "accept": false, + "reason": "Both are about single-file loading, but they address different failures and different code paths; too broad to merge as duplicates." + }, + { + "left": "pull_request:6195", + "right": "pull_request:6405", + "accept": false, + "reason": "A WIP AnimateDiff SDXL feature PR and a doc-builder workflow test PR are unrelated." + }, + { + "left": "pull_request:6775", + "right": "pull_request:7779", + "accept": false, + "reason": "Both are docstring/docs fixes, but for different APIs and different issues." + }, + { + "left": "pull_request:6961", + "right": "pull_request:8386", + "accept": false, + "reason": "Single-file checkpoint loader tests and output_shape-based model test cleanup are unrelated test maintenance." + }, + { + "left": "pull_request:7627", + "right": "pull_request:9447", + "accept": false, + "reason": "An img2img pipeline fix for Playground and a save_pretrained shard-size default change do not share the same bug." + }, + { + "left": "pull_request:8445", + "right": "pull_request:9447", + "accept": false, + "reason": "Both alter max_shard_size defaults, but one changes model saving shards and the other changes pipeline save semantics; not the same concrete fix." + }, + { + "left": "pull_request:8591", + "right": "pull_request:9735", + "accept": false, + "reason": "Both are SD3 training-related, but one refactors weighting utilities while the other updates a ControlNet example/training script; different changes." + }, + { + "left": "pull_request:11588", + "right": "pull_request:11913", + "accept": false, + "reason": "Both are Wan-related, but one adds LoRA support for I2V while the other introduces the modular Wan pipeline; separate code paths and features." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 148, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10069", + "pull_request:10117", + "pull_request:10126", + "pull_request:10271", + "pull_request:10326", + "pull_request:10354", + "pull_request:10376", + "pull_request:10482", + "pull_request:10704", + "pull_request:1089", + "pull_request:10922", + "pull_request:11052", + "pull_request:11059", + "pull_request:11066", + "pull_request:11067", + "pull_request:11196", + "pull_request:11273", + "pull_request:11290", + "pull_request:1147", + "pull_request:11614", + "pull_request:11728", + "pull_request:11823", + "pull_request:12062", + "pull_request:1215", + "pull_request:12419", + "pull_request:12425", + "pull_request:13364", + "pull_request:1496", + "pull_request:2915", + "pull_request:2956", + "pull_request:2959", + "pull_request:3020", + "pull_request:3410", + "pull_request:4927", + "pull_request:4963", + "pull_request:6490", + "pull_request:6645", + "pull_request:6752", + "pull_request:7086", + "pull_request:7714", + "pull_request:7771", + "pull_request:7941", + "pull_request:833", + "pull_request:8412", + "pull_request:8709", + "pull_request:8755", + "pull_request:9099", + "pull_request:9721" + ], + "soft_pairs": [ + "pull_request:8709|pull_request:9099", + "pull_request:10376|pull_request:11066", + "pull_request:10117|pull_request:10704", + "pull_request:10326|pull_request:10354", + "pull_request:2959|pull_request:3410", + "pull_request:4927|pull_request:4963", + "pull_request:6645|pull_request:7086", + "pull_request:7714|pull_request:8412", + "pull_request:7771|pull_request:7941", + "pull_request:11823|pull_request:12419", + "pull_request:6490|pull_request:6752", + "pull_request:10376|pull_request:10482", + "pull_request:2956|pull_request:3020", + "pull_request:10069|pull_request:11273", + "pull_request:10126|pull_request:8755", + "pull_request:10271|pull_request:11052", + "pull_request:1089|pull_request:1496", + "pull_request:10922|pull_request:11059", + "pull_request:11067|pull_request:9721", + "pull_request:11196|pull_request:11273", + "pull_request:11290|pull_request:12425", + "pull_request:1147|pull_request:2915", + "pull_request:11614|pull_request:11728", + "pull_request:11614|pull_request:12062", + "pull_request:1215|pull_request:833", + "pull_request:12425|pull_request:13364" + ], + "prepared_review_unit_hash": "1883a9ff8f6b0880473f00a22e5dce8359289c72ac35067fc18cbe70338bee08", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 97146, + "estimated_input_tokens": 24287, + "estimated_eval_tokens": 48830 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set; it\u2019s a grab bag of unrelated PRs spanning docs, CI/workflows, model code, LoRA conversions, and tests. All soft edges look like superficial similarity only, not the same underlying bug/change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are heterogeneous and do not resolve to one underlying fix or feature.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good global representative for deduping this cluster; the PRs cover unrelated areas and changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8709", + "right": "pull_request:9099", + "accept": false, + "reason": "Both are SD3-related pipeline additions, but they target different features: SD3 inpaint vs SD3 ControlNet inpainting." + }, + { + "left": "pull_request:10376", + "right": "pull_request:11066", + "accept": false, + "reason": "Both touch HunyuanVideo, but one is LoRA format conversion support and the other is a new HunyuanVideo-I2V model/pipeline." + }, + { + "left": "pull_request:10117", + "right": "pull_request:10704", + "accept": false, + "reason": "Both are docs fixes, but one repairs broken README links while the other updates image-to-image usage examples." + }, + { + "left": "pull_request:10326", + "right": "pull_request:10354", + "accept": false, + "reason": "Unrelated CI/test changes: one fixes an MPS workflow install command, the other fixes a Flux IP-Adapter test helper." + }, + { + "left": "pull_request:2959", + "right": "pull_request:3410", + "accept": false, + "reason": "Different test fixes in different code paths: UNet3D norm-group test vs DeepSpeed plugin retrieval in training examples." + }, + { + "left": "pull_request:4927", + "right": "pull_request:4963", + "accept": false, + "reason": "Both involve offloading/loaders, but one reverts accelerate hook handling while the other reinstalls accelerate main in workflows; not the same change." + }, + { + "left": "pull_request:6645", + "right": "pull_request:7086", + "accept": false, + "reason": "Both update community docs, but they cover different pipelines and different functional content." + }, + { + "left": "pull_request:7714", + "right": "pull_request:8412", + "accept": false, + "reason": "Both are docs work, but AutoPipeline documentation and single-file documentation are distinct topics." + }, + { + "left": "pull_request:7771", + "right": "pull_request:7941", + "accept": false, + "reason": "Both are test annotations, but one marks a ControlNetXS test flaky while the other marks a StableDiffusion21 single-file class as slow." + }, + { + "left": "pull_request:11823", + "right": "pull_request:12419", + "accept": false, + "reason": "Both are LoRA-related, but they support different model families and conversion paths: Flux Kontext vs modular Flux readiness." + }, + { + "left": "pull_request:6490", + "right": "pull_request:6752", + "accept": false, + "reason": "Both add community/research project content, but they introduce different projects and codepaths." + }, + { + "left": "pull_request:10376", + "right": "pull_request:10482", + "accept": false, + "reason": "Both are HunyuanVideo changes, but one adds LoRA conversion support and the other fixes a NaN/attention mask bug." + }, + { + "left": "pull_request:2956", + "right": "pull_request:3020", + "accept": false, + "reason": "Both relate to schedulers, but they implement different scheduler features in different algorithms." + }, + { + "left": "pull_request:10069", + "right": "pull_request:11273", + "accept": false, + "reason": "Completely different concerns: BnB quantizer device handling vs notification bot timeout behavior." + }, + { + "left": "pull_request:10126", + "right": "pull_request:8755", + "accept": false, + "reason": "Different domains and fixes: LoRA API deprecation vs SD3 DreamBooth LoRA training logic." + }, + { + "left": "pull_request:10271", + "right": "pull_request:11052", + "accept": false, + "reason": "TorchAO test refactor vs a new compute-capability test decorator; related to tests but not the same bug/change." + }, + { + "left": "pull_request:1089", + "right": "pull_request:1496", + "accept": false, + "reason": "Both are docs, but one expands scheduler usage docs and the other adds Habana Gaudi documentation." + }, + { + "left": "pull_request:10922", + "right": "pull_request:11059", + "accept": false, + "reason": "Same broad Wan area, but one adds Wan model/pipeline support and the other converts non-diffusers Wan LoRAs; different changes." + }, + { + "left": "pull_request:11067", + "right": "pull_request:9721", + "accept": false, + "reason": "Both are Docker/benchmark config changes, but one pins transformers for benchmarking and the other pins torch versions across Dockerfiles." + }, + { + "left": "pull_request:11196", + "right": "pull_request:11273", + "accept": false, + "reason": "Model save_pretrained behavior vs Slack release-notification timeout; unrelated code paths." + }, + { + "left": "pull_request:11290", + "right": "pull_request:12425", + "accept": false, + "reason": "Dockerfile dependency pinning vs torch.distributed import guarding; no shared underlying bug." + }, + { + "left": "pull_request:1147", + "right": "pull_request:2915", + "accept": false, + "reason": "Flax example loading fix vs a tolerance adjustment in a slow test; unrelated." + }, + { + "left": "pull_request:11614", + "right": "pull_request:11728", + "accept": false, + "reason": "CI workflow hardening vs doc-builder Dockerfile adding quant libs; different files and goals." + }, + { + "left": "pull_request:11614", + "right": "pull_request:12062", + "accept": false, + "reason": "Both touch CI, but one is a workflow loop/quoting fix and the other enables all GPUs across multiple workflows; not the same issue." + }, + { + "left": "pull_request:1215", + "right": "pull_request:833", + "accept": false, + "reason": "DDIM generator/device bugfix vs a version bump release PR; unrelated." + }, + { + "left": "pull_request:12425", + "right": "pull_request:13364", + "accept": false, + "reason": "Torch distributed import guarding vs MotionConv2d dtype casting; completely different code paths and fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 149, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10132", + "pull_request:11044", + "pull_request:11767", + "pull_request:1331", + "pull_request:1706", + "pull_request:2464", + "pull_request:2892", + "pull_request:2968", + "pull_request:3337", + "pull_request:3397", + "pull_request:3589", + "pull_request:3590", + "pull_request:4272", + "pull_request:4669", + "pull_request:4783", + "pull_request:4787", + "pull_request:5053", + "pull_request:5090", + "pull_request:5111", + "pull_request:6007", + "pull_request:6081", + "pull_request:6119", + "pull_request:6262", + "pull_request:6263", + "pull_request:6426", + "pull_request:6448", + "pull_request:6743", + "pull_request:6752", + "pull_request:6787", + "pull_request:6961", + "pull_request:7295", + "pull_request:7410", + "pull_request:7442", + "pull_request:7650", + "pull_request:7932", + "pull_request:7941", + "pull_request:816", + "pull_request:8267", + "pull_request:8370", + "pull_request:8452", + "pull_request:8496", + "pull_request:8591", + "pull_request:8652", + "pull_request:8660", + "pull_request:8719", + "pull_request:8761", + "pull_request:8937", + "pull_request:8973" + ], + "soft_pairs": [ + "pull_request:1331|pull_request:3337", + "pull_request:1706|pull_request:2464", + "pull_request:2892|pull_request:5053", + "pull_request:2968|pull_request:816", + "pull_request:3337|pull_request:4272", + "pull_request:3397|pull_request:5111", + "pull_request:3589|pull_request:3590", + "pull_request:3590|pull_request:5090", + "pull_request:4272|pull_request:6081", + "pull_request:4669|pull_request:4787", + "pull_request:4783|pull_request:6007", + "pull_request:6119|pull_request:7932", + "pull_request:6262|pull_request:7410", + "pull_request:6263|pull_request:7410", + "pull_request:6426|pull_request:8370", + "pull_request:6448|pull_request:8496", + "pull_request:6743|pull_request:7650", + "pull_request:6787|pull_request:8937", + "pull_request:6961|pull_request:7941", + "pull_request:7295|pull_request:8452", + "pull_request:8267|pull_request:8719", + "pull_request:8591|pull_request:8761", + "pull_request:8652|pull_request:8973", + "pull_request:6752|pull_request:7442", + "pull_request:10132|pull_request:8660", + "pull_request:11044|pull_request:11767" + ], + "prepared_review_unit_hash": "13d06b5c8209c5ad521343c130fb53d8997226349f53ac9dcaccd93fbdb3151a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 100496, + "estimated_input_tokens": 25124, + "estimated_eval_tokens": 50504 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No duplicate cluster emerges here: the soft edges mostly connect unrelated PRs that happen to touch similar subsystems, docs, or tests. I would not merge any of these pairs as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1331", + "right": "pull_request:3337", + "accept": false, + "reason": "Both touch inpaint/postprocessing, but 1331 fixes ONNX inpainting scheduler input order while 3337 is a broader postprocessing refactor across multiple pipelines." + }, + { + "left": "pull_request:1706", + "right": "pull_request:2464", + "accept": false, + "reason": "One is main docs cleanup; the other removes xformers training code from a LoRA example. Same repo area, different changes." + }, + { + "left": "pull_request:2892", + "right": "pull_request:5053", + "accept": false, + "reason": "Upscaler input validation for embeds is unrelated to the t2i-adapter validation_image CLI argument change." + }, + { + "left": "pull_request:2968", + "right": "pull_request:816", + "accept": false, + "reason": "Both are pipeline/training fixes, but one is img2img post-processing and the other is deepspeed float casting in text-to-image training." + }, + { + "left": "pull_request:3337", + "right": "pull_request:4272", + "accept": false, + "reason": "Postprocessing refactor is unrelated to optional import cleanup; different goals and code paths." + }, + { + "left": "pull_request:3397", + "right": "pull_request:5111", + "accept": false, + "reason": "Docs note for local directory paths has nothing to do with K-Diffusion generator passthrough or scheduler changes." + }, + { + "left": "pull_request:3589", + "right": "pull_request:3590", + "accept": false, + "reason": "Both update community docs, but they add different community pipelines/features and are not the same change." + }, + { + "left": "pull_request:3590", + "right": "pull_request:5090", + "accept": false, + "reason": "Stable Diffusion Mixture Canvas docs/code are unrelated to T2I-Adapter bugfixes." + }, + { + "left": "pull_request:4272", + "right": "pull_request:6081", + "accept": false, + "reason": "Optional import statement fixes are unrelated to the SDXL sigma/torch.compile pipeline tweak." + }, + { + "left": "pull_request:4669", + "right": "pull_request:4787", + "accept": false, + "reason": "Both are LoRA-related loader changes, but one handles mixed LoRA ranks and the other tightens checkpoint file filtering." + }, + { + "left": "pull_request:4783", + "right": "pull_request:6007", + "accept": false, + "reason": "SDXL embedding docs and the copied-from docs refactor are different documentation changes with no shared bug or behavior." + }, + { + "left": "pull_request:6119", + "right": "pull_request:7932", + "accept": false, + "reason": "Both mention LoRA tests, but they address different issues: fp16 gradient scaling vs RGB ordering side effects." + }, + { + "left": "pull_request:6262", + "right": "pull_request:7410", + "accept": false, + "reason": "Moving panorama out of stable_diffusion is unrelated to fixing FreeInit behavior in animation/PIA pipelines." + }, + { + "left": "pull_request:6263", + "right": "pull_request:7410", + "accept": false, + "reason": "LDM3D refactor is a pipeline move; FreeInit fix is a separate runtime behavior change." + }, + { + "left": "pull_request:6426", + "right": "pull_request:8370", + "accept": false, + "reason": "LoRA deprecation cleanup is unrelated to safely handling norm_type remapping in model loading utilities." + }, + { + "left": "pull_request:6448", + "right": "pull_request:8496", + "accept": false, + "reason": "LoRA test assertion updates do not match the Transformer SD3 warning-log guard change." + }, + { + "left": "pull_request:6743", + "right": "pull_request:7650", + "accept": false, + "reason": "These are different training arguments in different scripts: hflip flag correction vs SDXL ip2p cosxl-edit support." + }, + { + "left": "pull_request:6787", + "right": "pull_request:8937", + "accept": false, + "reason": "Marigold depth input scaling fix is unrelated to AuraFlow long-prompt handling." + }, + { + "left": "pull_request:6961", + "right": "pull_request:7941", + "accept": false, + "reason": "Single-file loader fast-test additions are not the same as marking a separate single-file test class as slow." + }, + { + "left": "pull_request:7295", + "right": "pull_request:8452", + "accept": false, + "reason": "UNet single-file loader docs cleanup is unrelated to the single-file loading strictness fix." + }, + { + "left": "pull_request:8267", + "right": "pull_request:8719", + "accept": false, + "reason": "Both involve single-file loading, but one changes resume_download handling and the other updates VQModelOutput deprecation wording." + }, + { + "left": "pull_request:8591", + "right": "pull_request:8761", + "accept": false, + "reason": "SD3 training weighting refactor is unrelated to Hunyuan embedding refactoring." + }, + { + "left": "pull_request:8652", + "right": "pull_request:8973", + "accept": false, + "reason": "Lumina-T2X feature addition and the Lumina test-size reduction are not the same underlying change." + }, + { + "left": "pull_request:10132", + "right": "pull_request:8660", + "accept": false, + "reason": "Workflow tokenizers pinning is unrelated to SD3 ControlNet input validation." + }, + { + "left": "pull_request:11044", + "right": "pull_request:11767", + "accept": false, + "reason": "LoRA no-op logging change and group-offloading unique-memory-address fix address completely different code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 150, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10838", + "pull_request:10970", + "pull_request:11261", + "pull_request:1171", + "pull_request:11975", + "pull_request:1206", + "pull_request:1229", + "pull_request:12451", + "pull_request:12578", + "pull_request:12739", + "pull_request:12921", + "pull_request:13120", + "pull_request:13352", + "pull_request:1554", + "pull_request:1945", + "pull_request:2327", + "pull_request:2334", + "pull_request:2385", + "pull_request:2590", + "pull_request:3201", + "pull_request:3245", + "pull_request:3265", + "pull_request:3275", + "pull_request:3398", + "pull_request:340", + "pull_request:3614", + "pull_request:3711", + "pull_request:3825", + "pull_request:4098", + "pull_request:4187", + "pull_request:453", + "pull_request:4819", + "pull_request:5024", + "pull_request:51", + "pull_request:5111", + "pull_request:5396", + "pull_request:5456", + "pull_request:5837", + "pull_request:5858", + "pull_request:6217", + "pull_request:6915", + "pull_request:6972", + "pull_request:711", + "pull_request:8422", + "pull_request:8425", + "pull_request:886", + "pull_request:934" + ], + "soft_pairs": [ + "pull_request:1171|pull_request:1206", + "pull_request:1229|pull_request:2590", + "pull_request:1229|pull_request:3275", + "pull_request:12578|pull_request:13120", + "pull_request:12739|pull_request:12921", + "pull_request:1554|pull_request:453", + "pull_request:1945|pull_request:3275", + "pull_request:2327|pull_request:2590", + "pull_request:3201|pull_request:3711", + "pull_request:3245|pull_request:3825", + "pull_request:3265|pull_request:3711", + "pull_request:3398|pull_request:3825", + "pull_request:3614|pull_request:5396", + "pull_request:3711|pull_request:5858", + "pull_request:4187|pull_request:5024", + "pull_request:4187|pull_request:5456", + "pull_request:453|pull_request:51", + "pull_request:5837|pull_request:6915", + "pull_request:6217|pull_request:8425", + "pull_request:6972|pull_request:8422", + "pull_request:886|pull_request:934", + "pull_request:4098|pull_request:4819", + "pull_request:2334|pull_request:2385", + "pull_request:10970|pull_request:11975", + "pull_request:12451|pull_request:13352", + "pull_request:340|pull_request:711", + "pull_request:5024|pull_request:5111", + "pull_request:10838|pull_request:11261" + ], + "prepared_review_unit_hash": "1486f83954114840f14958292a59eb9a2ee167314bd8572f55cc2cf20fbe1c40", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 95845, + "estimated_input_tokens": 23962, + "estimated_eval_tokens": 48180 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a bag of unrelated PRs across docs, CI, tests, schedulers, and new model features. I don\u2019t see a single underlying bug/change tying them together, so none of the soft pairs should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1171", + "right": "pull_request:1206", + "accept": false, + "reason": "Docs typo fix vs LDM layer-name fix; different files and changes." + }, + { + "left": "pull_request:1229", + "right": "pull_request:2590", + "accept": false, + "reason": "Dreambooth README wording vs CI deps-table check; unrelated." + }, + { + "left": "pull_request:1229", + "right": "pull_request:3275", + "accept": false, + "reason": "README/install docs vs UNet upsample_size API change; not the same change." + }, + { + "left": "pull_request:12578", + "right": "pull_request:13120", + "accept": false, + "reason": "CPU offload error handling vs model-loading tests; different code paths." + }, + { + "left": "pull_request:12739", + "right": "pull_request:12921", + "accept": false, + "reason": "Dummy export for ZImage vs full GLM-Image model support; same area but different feature work." + }, + { + "left": "pull_request:1554", + "right": "pull_request:453", + "accept": false, + "reason": "Docs correction in pipeline docs vs installation typo fix; unrelated." + }, + { + "left": "pull_request:1945", + "right": "pull_request:3275", + "accept": false, + "reason": "Block tests vs adding upsample_size to blocks; related subsystem but not the same underlying change." + }, + { + "left": "pull_request:2327", + "right": "pull_request:2590", + "accept": false, + "reason": "UNet input validation vs CI workflow update; unrelated." + }, + { + "left": "pull_request:3201", + "right": "pull_request:3711", + "accept": false, + "reason": "Community pipeline revert vs doc-build workflow revert; no shared bug." + }, + { + "left": "pull_request:3245", + "right": "pull_request:3825", + "accept": false, + "reason": "Chinese docs translation update vs a single English docs typo fix; unrelated." + }, + { + "left": "pull_request:3265", + "right": "pull_request:3711", + "accept": false, + "reason": "lpw community pipeline revert vs doc-build workflow revert; unrelated." + }, + { + "left": "pull_request:3398", + "right": "pull_request:3825", + "accept": false, + "reason": "Stable diffusion docs typo vs control_brightness docs update; different doc edits." + }, + { + "left": "pull_request:3614", + "right": "pull_request:5396", + "accept": false, + "reason": "Test cleanup in stable diffusion vs stale-bot timezone fix; unrelated." + }, + { + "left": "pull_request:3711", + "right": "pull_request:5858", + "accept": false, + "reason": "Workflow revert vs training-doc timestamp dtype fix; unrelated." + }, + { + "left": "pull_request:4187", + "right": "pull_request:5024", + "accept": false, + "reason": "Different scheduler bugs in different scheduler classes; not mergeable as one PR." + }, + { + "left": "pull_request:4187", + "right": "pull_request:5456", + "accept": false, + "reason": "DPMSolverMultistep scheduler change vs RealFill feature addition; unrelated." + }, + { + "left": "pull_request:453", + "right": "pull_request:51", + "accept": false, + "reason": "Install docs typo vs setup.py dependency pin fix; similar theme but different artifacts and fixes." + }, + { + "left": "pull_request:5837", + "right": "pull_request:6915", + "accept": false, + "reason": "Textual inversion multi-token support vs attention processor signature handling; unrelated." + }, + { + "left": "pull_request:6217", + "right": "pull_request:8425", + "accept": false, + "reason": "Torch-compile SVD VAE fix vs workflow rename; unrelated." + }, + { + "left": "pull_request:6972", + "right": "pull_request:8422", + "accept": false, + "reason": "IPAdapter unload test expectation vs subfolder-loading test suite; different test goals." + }, + { + "left": "pull_request:886", + "right": "pull_request:934", + "accept": false, + "reason": "VAE test seeding fix vs broader MPS test adjustments; same family of tests but not the same bug/change." + }, + { + "left": "pull_request:4098", + "right": "pull_request:4819", + "accept": false, + "reason": "SDXL inpainting feature add vs masked-latent input handling across inpainting pipelines; overlapping area but distinct changes." + }, + { + "left": "pull_request:2334", + "right": "pull_request:2385", + "accept": false, + "reason": "Pix2Pix Zero pipeline introduction vs a utility helper added later in the same file; not the same underlying change." + }, + { + "left": "pull_request:10970", + "right": "pull_request:11975", + "accept": false, + "reason": "Workflow reuse change vs a style fix in a training script; unrelated." + }, + { + "left": "pull_request:12451", + "right": "pull_request:13352", + "accept": false, + "reason": "Doc-builder Dockerfile packages vs Claude review checkout step; unrelated CI infra changes." + }, + { + "left": "pull_request:340", + "right": "pull_request:711", + "accept": false, + "reason": "Img2Img scheduler support fix vs a broad scheduler design proposal; not the same code-path bug." + }, + { + "left": "pull_request:5024", + "right": "pull_request:5111", + "accept": false, + "reason": "DPM-Solver-SDE step index/noise sampler vs KDPM2 Karras-sigmas support; different scheduler fixes." + }, + { + "left": "pull_request:10838", + "right": "pull_request:11261", + "accept": false, + "reason": "FlashVideo text-to-video pipeline vs VidTok autoencoder support; both video-related but unrelated changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 151, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10326", + "pull_request:10354", + "pull_request:10810", + "pull_request:10941", + "pull_request:11231", + "pull_request:11243", + "pull_request:11254", + "pull_request:11532", + "pull_request:11814", + "pull_request:11892", + "pull_request:12099", + "pull_request:12117", + "pull_request:12151", + "pull_request:12301", + "pull_request:12496", + "pull_request:12530", + "pull_request:12611", + "pull_request:12618", + "pull_request:12747", + "pull_request:13081", + "pull_request:13211", + "pull_request:13296", + "pull_request:1494", + "pull_request:152", + "pull_request:2365", + "pull_request:2489", + "pull_request:2706", + "pull_request:2959", + "pull_request:3001", + "pull_request:3351", + "pull_request:3397", + "pull_request:3471", + "pull_request:3561", + "pull_request:3766", + "pull_request:3968", + "pull_request:3989", + "pull_request:402", + "pull_request:4040", + "pull_request:4115", + "pull_request:4259", + "pull_request:4412", + "pull_request:504", + "pull_request:7328", + "pull_request:8797", + "pull_request:8798", + "pull_request:8981", + "pull_request:9676" + ], + "soft_pairs": [ + "pull_request:12117|pull_request:12301", + "pull_request:12496|pull_request:12611", + "pull_request:11243|pull_request:12151", + "pull_request:12530|pull_request:12618", + "pull_request:12530|pull_request:13296", + "pull_request:2489|pull_request:402", + "pull_request:3001|pull_request:3351", + "pull_request:3989|pull_request:4259", + "pull_request:7328|pull_request:8981", + "pull_request:3968|pull_request:4115", + "pull_request:10354|pull_request:8797", + "pull_request:10941|pull_request:11892", + "pull_request:10326|pull_request:8798", + "pull_request:10810|pull_request:11814", + "pull_request:11231|pull_request:11532", + "pull_request:11254|pull_request:9676", + "pull_request:12099|pull_request:13081", + "pull_request:12747|pull_request:13211", + "pull_request:1494|pull_request:152", + "pull_request:2365|pull_request:3397", + "pull_request:2365|pull_request:4040", + "pull_request:2489|pull_request:504", + "pull_request:2706|pull_request:3766", + "pull_request:2959|pull_request:4412", + "pull_request:3471|pull_request:3561" + ], + "prepared_review_unit_hash": "2f2b90b10a089b94eb43e4611b29867b246bd9d2a439891156bcf57e114614d5", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 98224, + "estimated_input_tokens": 24556, + "estimated_eval_tokens": 49368 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a broad grab-bag of unrelated PRs across CI, docs, LoRA conversion, Qwen/Flux/Wan pipelines, Flax deprecation, and assorted test fixes. The soft edges look like topic overlap only, not the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are heterogeneous and cover distinct subsystems and fixes, so any single representative would be arbitrary.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a strong global duplicate representative here; the cluster is too heterogeneous for a meaningful best-PR choice.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12117", + "right": "pull_request:12301", + "accept": false, + "reason": "Both are Qwen-Image feature additions, but one adds img2img/inpaint pipelines while the other adds ControlNet inpainting; different changes." + }, + { + "left": "pull_request:12496", + "right": "pull_request:12611", + "accept": false, + "reason": "Wan i2v shape-mismatch fix vs. broad Wan modular refactor; unrelated concrete problems." + }, + { + "left": "pull_request:11243", + "right": "pull_request:12151", + "accept": false, + "reason": "LoRA support for Musubi Wan vs. Flax deprecation; no shared bug/change." + }, + { + "left": "pull_request:12530", + "right": "pull_request:12618", + "accept": false, + "reason": "Both touch CI, but one xfails a flaky test and the other fixes a uv install typo; not the same issue." + }, + { + "left": "pull_request:12530", + "right": "pull_request:13296", + "accept": false, + "reason": "CI-related only; xfail for a test vs. release-test dependency update are different fixes." + }, + { + "left": "pull_request:2489", + "right": "pull_request:402", + "accept": false, + "reason": "Docs coreml page addition vs. intro-section docs restructure; unrelated." + }, + { + "left": "pull_request:3001", + "right": "pull_request:3351", + "accept": false, + "reason": "Both involve Karras-related scheduler work, but they change different schedulers and different APIs." + }, + { + "left": "pull_request:3989", + "right": "pull_request:4259", + "accept": false, + "reason": "Slow-test fixes in SDXL vs. circular-import fix in loaders; unrelated." + }, + { + "left": "pull_request:7328", + "right": "pull_request:8981", + "accept": false, + "reason": "PyPI workflow dependency tweak vs. LoRA loader init/docs changes; no shared code-path bug." + }, + { + "left": "pull_request:3968", + "right": "pull_request:4115", + "accept": false, + "reason": "Both touch SDXL, but one is a workflow/docs/model cleanup and the other fixes denoising_start/end semantics; not the same change." + }, + { + "left": "pull_request:10354", + "right": "pull_request:8797", + "accept": false, + "reason": "Flux tester mixin fix vs. sharded checkpoint test fix; different test issues." + }, + { + "left": "pull_request:10941", + "right": "pull_request:11892", + "accept": false, + "reason": "Flux copy fixes vs. removing k-diffusion dependency; unrelated." + }, + { + "left": "pull_request:10326", + "right": "pull_request:8798", + "accept": false, + "reason": "MPS workflow package quoting fix vs. checkpoint subfolder loading fix; unrelated." + }, + { + "left": "pull_request:10810", + "right": "pull_request:11814", + "accept": false, + "reason": "Flux LoRA conversion support vs. gradient-checkpointing kwargs passthrough; different code paths." + }, + { + "left": "pull_request:11231", + "right": "pull_request:11532", + "accept": false, + "reason": "HiDream model/pipeline addition vs. HiDream LoRA conversion support; same ecosystem but different features." + }, + { + "left": "pull_request:11254", + "right": "pull_request:9676", + "accept": false, + "reason": "ConsisID import/dependency fix vs. EntryNotFoundError import fix; unrelated dependency cleanups." + }, + { + "left": "pull_request:12099", + "right": "pull_request:13081", + "accept": false, + "reason": "Qwen device placement typo vs. Qwen contiguous-tensor fix; separate bugs." + }, + { + "left": "pull_request:12747", + "right": "pull_request:13211", + "accept": false, + "reason": "Both concern group offloading, but one adds pinning features and the other patches a fallback bug; not one mergeable change." + }, + { + "left": "pull_request:1494", + "right": "pull_request:152", + "accept": false, + "reason": "Safetensors saving support vs. local image-folder training example; unrelated." + }, + { + "left": "pull_request:2365", + "right": "pull_request:3397", + "accept": false, + "reason": "Scheduler type-note cleanup vs. LoRA docs note; no shared bug/change." + }, + { + "left": "pull_request:2365", + "right": "pull_request:4040", + "accept": false, + "reason": "Scheduler typing note vs. config _use_default_values list fix; unrelated." + }, + { + "left": "pull_request:2489", + "right": "pull_request:504", + "accept": false, + "reason": "CoreML docs addition vs. docs link-path fix; unrelated." + }, + { + "left": "pull_request:2706", + "right": "pull_request:3766", + "accept": false, + "reason": "Dreambooth EMA save/restore fix vs. Kandinsky conversion script update; different code paths." + }, + { + "left": "pull_request:2959", + "right": "pull_request:4412", + "accept": false, + "reason": "UNet3D norm-group test adjustment vs. float16 inference test fix; different test failures." + }, + { + "left": "pull_request:3471", + "right": "pull_request:3561", + "accept": false, + "reason": "Dreambooth docs note vs. Dreambooth/IF stage-II upscaling implementation; not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 152, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10069", + "pull_request:10114", + "pull_request:10122", + "pull_request:10206", + "pull_request:10271", + "pull_request:10293", + "pull_request:1030", + "pull_request:10358", + "pull_request:11588", + "pull_request:11728", + "pull_request:11975", + "pull_request:12363", + "pull_request:12397", + "pull_request:12450", + "pull_request:12611", + "pull_request:12642", + "pull_request:12919", + "pull_request:12982", + "pull_request:12998", + "pull_request:13017", + "pull_request:13169", + "pull_request:2902", + "pull_request:3189", + "pull_request:3313", + "pull_request:3590", + "pull_request:4112", + "pull_request:4214", + "pull_request:4262", + "pull_request:4572", + "pull_request:5186", + "pull_request:5384", + "pull_request:5420", + "pull_request:5552", + "pull_request:5792", + "pull_request:6489", + "pull_request:6759", + "pull_request:7044", + "pull_request:7442", + "pull_request:7679", + "pull_request:8285", + "pull_request:8467", + "pull_request:8660", + "pull_request:9060", + "pull_request:9065", + "pull_request:9400", + "pull_request:9426", + "pull_request:9593" + ], + "soft_pairs": [ + "pull_request:3590|pull_request:5384", + "pull_request:4112|pull_request:4262", + "pull_request:4214|pull_request:5792", + "pull_request:4572|pull_request:5420", + "pull_request:5186|pull_request:5552", + "pull_request:6759|pull_request:8660", + "pull_request:7044|pull_request:9065", + "pull_request:7679|pull_request:8285", + "pull_request:9060|pull_request:9426", + "pull_request:2902|pull_request:3313", + "pull_request:12919|pull_request:13017", + "pull_request:6489|pull_request:7442", + "pull_request:12982|pull_request:13169", + "pull_request:11588|pull_request:12611", + "pull_request:10069|pull_request:9400", + "pull_request:10114|pull_request:8467", + "pull_request:10122|pull_request:10206", + "pull_request:10271|pull_request:10293", + "pull_request:1030|pull_request:3189", + "pull_request:10358|pull_request:9593", + "pull_request:11728|pull_request:12363", + "pull_request:11728|pull_request:12642", + "pull_request:11975|pull_request:12998", + "pull_request:12397|pull_request:12450" + ], + "prepared_review_unit_hash": "78266a825859431563a0846a8da7fe39e8408cf94ffd0db3aa8db2da8c1363fd", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 99026, + "estimated_input_tokens": 24757, + "estimated_eval_tokens": 49770 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set; it mixes many unrelated PRs across pipelines, docs, CI, LoRA, quantization, and NPU work. The soft-similarity links are mostly superficial (same broad area or same file) rather than the same concrete bug/change.", + "confidence": 0.76, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR really fits because the cluster is heterogeneous; if one representative is needed, PR 12611 is the broadest and most substantive merged change in the set.", + "best_issue_reason": null, + "best_pr_reason": "PR 12611 is the most substantial standalone PR here, with the widest scope and the most core code touched, so it is the best representative artifact despite the cluster not being a true duplicate group.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3590", + "right": "pull_request:5384", + "accept": false, + "reason": "Unrelated changes: Stable Diffusion Mixture Canvas/community example work vs hypernetwork loader support." + }, + { + "left": "pull_request:4112", + "right": "pull_request:4262", + "accept": false, + "reason": "Both are DreamBooth-related, but they fix different things: rank plumbing vs SDXL DreamBooth batch/prior-preservation fixes." + }, + { + "left": "pull_request:4214", + "right": "pull_request:5792", + "accept": false, + "reason": "Custom Diffusion bf16 error handling and import-path cleanup are unrelated changes." + }, + { + "left": "pull_request:4572", + "right": "pull_request:5420", + "accept": false, + "reason": "One fixes LoRA loading argument names; the other is a tutorial/doc snippet correction for create_repo return value." + }, + { + "left": "pull_request:5186", + "right": "pull_request:5552", + "accept": false, + "reason": "Related FreeU area, but not the same underlying fix: a FreeU PoC vs correcting disable_freeu behavior." + }, + { + "left": "pull_request:6759", + "right": "pull_request:8660", + "accept": false, + "reason": "InstantID image embedding repetition and SD3 ControlNet max_sequence_length validation are different code paths and bugs." + }, + { + "left": "pull_request:7044", + "right": "pull_request:9065", + "accept": false, + "reason": "UNet refactor restoration and sentencepiece soft-dependency support are unrelated." + }, + { + "left": "pull_request:7679", + "right": "pull_request:8285", + "accept": false, + "reason": "CI peft-install tweak and installing wget in the doc-builder Dockerfile are unrelated maintenance changes." + }, + { + "left": "pull_request:9060", + "right": "pull_request:9426", + "accept": false, + "reason": "Removing a debug print in transformer_flux is unrelated to updating artifact uploader versions in CI." + }, + { + "left": "pull_request:2902", + "right": "pull_request:3313", + "accept": false, + "reason": "Img2Img cross_attention_kwargs bugfix and broad torch.compile break fixes are different concrete problems." + }, + { + "left": "pull_request:12919", + "right": "pull_request:13017", + "accept": false, + "reason": "Both touch NPU attention, but one adds a backend op for Ulysses while the other fixes QwenImageEditPlus-specific NPU masking/shape behavior." + }, + { + "left": "pull_request:6489", + "right": "pull_request:7442", + "accept": false, + "reason": "StyleAligned pipeline and Differential Diffusion are separate community projects with different functionality." + }, + { + "left": "pull_request:12982", + "right": "pull_request:13169", + "accept": false, + "reason": "Adding Flux2 klein support and converting Flux Klein PEFT LoRA format are related by model name but are distinct features/fixes." + }, + { + "left": "pull_request:11588", + "right": "pull_request:12611", + "accept": false, + "reason": "LoRA unload support for Flux Control and the Wan modular pipeline are different features in different subsystems." + }, + { + "left": "pull_request:10069", + "right": "pull_request:9400", + "accept": false, + "reason": "BnB quantizer device-type validation and device-agnostic test updates are not the same bug/change." + }, + { + "left": "pull_request:10114", + "right": "pull_request:8467", + "accept": false, + "reason": "Formatting cleanup in Stable Audio is unrelated to the SEGA pipeline execution-device fix." + }, + { + "left": "pull_request:10122", + "right": "pull_request:10206", + "accept": false, + "reason": "Passing attn_mask into SDPA for Flux and unloading LoRA weights for Flux Control are different code-path fixes." + }, + { + "left": "pull_request:10271", + "right": "pull_request:10293", + "accept": false, + "reason": "TorchAO test refactoring and minimum-version bump are related to the same package but are different changes, not one concrete fix." + }, + { + "left": "pull_request:1030", + "right": "pull_request:3189", + "accept": false, + "reason": "DreamBooth padding behavior and AudioLDM waveform dtype are unrelated." + }, + { + "left": "pull_request:10358", + "right": "pull_request:9593", + "accept": false, + "reason": "Community projects table update and docstring refactoring are unrelated documentation changes." + }, + { + "left": "pull_request:11728", + "right": "pull_request:12363", + "accept": false, + "reason": "Doc-builder quant-lib installation and QwenImageEdit Plus docs/tests are unrelated." + }, + { + "left": "pull_request:11728", + "right": "pull_request:12642", + "accept": false, + "reason": "Doc-builder Dockerfile changes and doc TOC updates do not fix the same issue." + }, + { + "left": "pull_request:11975", + "right": "pull_request:12998", + "accept": false, + "reason": "These are separate style-only cleanups in different training/pipeline files." + }, + { + "left": "pull_request:12397", + "right": "pull_request:12450", + "accept": false, + "reason": "Temporarily disabling main-branch installs in specific CI jobs and broadly revisiting CI installation strategy are related at a high level but not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 153, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10387", + "pull_request:11534", + "pull_request:11797", + "pull_request:11950", + "pull_request:12591", + "pull_request:1468", + "pull_request:1527", + "pull_request:1564", + "pull_request:1937", + "pull_request:2655", + "pull_request:3673", + "pull_request:3675", + "pull_request:3691", + "pull_request:3797", + "pull_request:3798", + "pull_request:3806", + "pull_request:3850", + "pull_request:453", + "pull_request:4862", + "pull_request:4963", + "pull_request:5697", + "pull_request:5754", + "pull_request:6078", + "pull_request:6119", + "pull_request:6261", + "pull_request:6564", + "pull_request:6772", + "pull_request:6886", + "pull_request:689", + "pull_request:7116", + "pull_request:7295", + "pull_request:7410", + "pull_request:7662", + "pull_request:7679", + "pull_request:7778", + "pull_request:816", + "pull_request:8267", + "pull_request:8303", + "pull_request:8425", + "pull_request:8618", + "pull_request:8627", + "pull_request:8652", + "pull_request:8719", + "pull_request:9060", + "pull_request:9323", + "pull_request:9489", + "pull_request:9696", + "pull_request:9918" + ], + "soft_pairs": [ + "pull_request:1468|pull_request:1527", + "pull_request:1564|pull_request:453", + "pull_request:1937|pull_request:3675", + "pull_request:2655|pull_request:3673", + "pull_request:3691|pull_request:3806", + "pull_request:3797|pull_request:3798", + "pull_request:3850|pull_request:6078", + "pull_request:4862|pull_request:6772", + "pull_request:4963|pull_request:5754", + "pull_request:5697|pull_request:7679", + "pull_request:6119|pull_request:8303", + "pull_request:6261|pull_request:7410", + "pull_request:6564|pull_request:8425", + "pull_request:6886|pull_request:7662", + "pull_request:6886|pull_request:8267", + "pull_request:689|pull_request:816", + "pull_request:7116|pull_request:7662", + "pull_request:7295|pull_request:8719", + "pull_request:7778|pull_request:9696", + "pull_request:8618|pull_request:8627", + "pull_request:8652|pull_request:9918", + "pull_request:9060|pull_request:9323", + "pull_request:11797|pull_request:11950", + "pull_request:11534|pull_request:12591", + "pull_request:10387|pull_request:9489" + ], + "prepared_review_unit_hash": "4c150f5fd59423a7dbd10d27ffcc58ac06d3d8b9d9515b1a45c5f48c0e85dd53", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 97892, + "estimated_input_tokens": 24473, + "estimated_eval_tokens": 49202 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These PRs are not true duplicates; they only share broad topical similarity (docs, workflows, training scripts, quantization, pipelines) and do not target the same concrete bug or change. No single canonical PR or issue stands out for the cluster.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1468", + "right": "pull_request:1527", + "accept": false, + "reason": "Unrelated fixes: saving trained betas vs. inpaint tensor input handling." + }, + { + "left": "pull_request:1564", + "right": "pull_request:453", + "accept": false, + "reason": "Different docs typos/notes in separate example READMEs." + }, + { + "left": "pull_request:1937", + "right": "pull_request:3675", + "accept": false, + "reason": "Conversion EMA extraction logic vs. a UniDiffuser test fix; no shared code path." + }, + { + "left": "pull_request:2655", + "right": "pull_request:3673", + "accept": false, + "reason": "LoRA save/load logic vs. DreamBooth multi-GPU fix; different subsystems and bugs." + }, + { + "left": "pull_request:3691", + "right": "pull_request:3806", + "accept": false, + "reason": "Text-encoder LoRA processor selection vs. checkpoint loading note; unrelated changes." + }, + { + "left": "pull_request:3797", + "right": "pull_request:3798", + "accept": false, + "reason": "Different model naming/constructor fixes vs. comment cleanup; not the same bug." + }, + { + "left": "pull_request:3850", + "right": "pull_request:6078", + "accept": false, + "reason": "AudioDiffusion test setup fix vs. Euler sigma scheduler fix; different targets." + }, + { + "left": "pull_request:4862", + "right": "pull_request:6772", + "accept": false, + "reason": "SDXL MultiControlNet height/width bug vs. ControlNet-XS implementation; distinct code paths." + }, + { + "left": "pull_request:4963", + "right": "pull_request:5754", + "accept": false, + "reason": "Workflow/installing accelerate main vs. unrelated formatting cleanup in training scripts." + }, + { + "left": "pull_request:5697", + "right": "pull_request:7679", + "accept": false, + "reason": "PEFT/xformers attention scaling bug vs. workflow package-install change; not mergeable as one fix." + }, + { + "left": "pull_request:6119", + "right": "pull_request:8303", + "accept": false, + "reason": "LoRA fp16 gradient handling vs. a documentation wording change." + }, + { + "left": "pull_request:6261", + "right": "pull_request:7410", + "accept": false, + "reason": "Attend-and-Excite refactor vs. FreeInit behavior fix; different pipelines and problems." + }, + { + "left": "pull_request:6564", + "right": "pull_request:8425", + "accept": false, + "reason": "IP Adapter path bug fix vs. workflow name rename; unrelated." + }, + { + "left": "pull_request:6886", + "right": "pull_request:7662", + "accept": false, + "reason": "Deprecated `to(torch_dtype=...)` cleanup vs. removing redundant apt install in Flax tests." + }, + { + "left": "pull_request:6886", + "right": "pull_request:8267", + "accept": false, + "reason": "Deprecated `to(torch_dtype=...)` cleanup vs. `resume_download` deprecation handling." + }, + { + "left": "pull_request:689", + "right": "pull_request:816", + "accept": false, + "reason": "Attention implementation performance tweak vs. Deepspeed mixed-precision training fix." + }, + { + "left": "pull_request:7116", + "right": "pull_request:7662", + "accept": false, + "reason": "Workflow dependency installation overhaul vs. removing redundant package install in one job." + }, + { + "left": "pull_request:7295", + "right": "pull_request:8719", + "accept": false, + "reason": "Single-file UNet loader docs cleanup vs. VQModelOutput deprecation messaging." + }, + { + "left": "pull_request:7778", + "right": "pull_request:9696", + "accept": false, + "reason": "Debugging SSH workflow vs. PR test workflow input changes; different workflows and purposes." + }, + { + "left": "pull_request:8618", + "right": "pull_request:8627", + "accept": false, + "reason": "SD3 tiny autoencoder shift_factor change vs. SD3 ControlNet example/pipeline addition; different features." + }, + { + "left": "pull_request:8652", + "right": "pull_request:9918", + "accept": false, + "reason": "Lumina feature addition vs. server example docs/code; only broad release-related similarity." + }, + { + "left": "pull_request:9060", + "right": "pull_request:9323", + "accept": false, + "reason": "Remove debug print in Flux transformer vs. changing single-file default checkpoints; unrelated." + }, + { + "left": "pull_request:11797", + "right": "pull_request:11950", + "accept": false, + "reason": "Higgs quantization support vs. SVDQuant support; both quantization-related but distinct methods and implementations." + }, + { + "left": "pull_request:11534", + "right": "pull_request:12591", + "accept": false, + "reason": "Framepack docs/pipeline addition vs. attention dispatch backend support for Hunyuan Video; not the same change." + }, + { + "left": "pull_request:10387", + "right": "pull_request:9489", + "accept": false, + "reason": "Different SD3 controlnet training fixes: custom data loading vs. DDP/find_unused_parameters." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 154, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10576", + "pull_request:11039", + "pull_request:1128", + "pull_request:11290", + "pull_request:11583", + "pull_request:11595", + "pull_request:11614", + "pull_request:11618", + "pull_request:11655", + "pull_request:11699", + "pull_request:12277", + "pull_request:12331", + "pull_request:12397", + "pull_request:12419", + "pull_request:12451", + "pull_request:12806", + "pull_request:13164", + "pull_request:13165", + "pull_request:13280", + "pull_request:13454", + "pull_request:13562", + "pull_request:2040", + "pull_request:2795", + "pull_request:3852", + "pull_request:3912", + "pull_request:3928", + "pull_request:4248", + "pull_request:4301", + "pull_request:4352", + "pull_request:4744", + "pull_request:5157", + "pull_request:5299", + "pull_request:5771", + "pull_request:5943", + "pull_request:6003", + "pull_request:6014", + "pull_request:6241", + "pull_request:6484", + "pull_request:6787", + "pull_request:6961", + "pull_request:7134", + "pull_request:7158", + "pull_request:7638", + "pull_request:7650", + "pull_request:8290", + "pull_request:8684", + "pull_request:8685", + "pull_request:8761" + ], + "soft_pairs": [ + "pull_request:11614|pull_request:12397", + "pull_request:11655|pull_request:12419", + "pull_request:11699|pull_request:12331", + "pull_request:12331|pull_request:13165", + "pull_request:13164|pull_request:13454", + "pull_request:13164|pull_request:13562", + "pull_request:2795|pull_request:3912", + "pull_request:3928|pull_request:4248", + "pull_request:5299|pull_request:7134", + "pull_request:5771|pull_request:6961", + "pull_request:5943|pull_request:7650", + "pull_request:8290|pull_request:8761", + "pull_request:11290|pull_request:12451", + "pull_request:3852|pull_request:4744", + "pull_request:4301|pull_request:4352", + "pull_request:5157|pull_request:7158", + "pull_request:6014|pull_request:7638", + "pull_request:8684|pull_request:8685", + "pull_request:6241|pull_request:6787", + "pull_request:10576|pull_request:11595", + "pull_request:11039|pull_request:12277", + "pull_request:11583|pull_request:11618", + "pull_request:6003|pull_request:6484", + "pull_request:12806|pull_request:13280", + "pull_request:1128|pull_request:2040" + ], + "prepared_review_unit_hash": "cfb6711ab8555e1103a31881fcce4a1f02e68caa315a4a19500755523b9c8374", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 98200, + "estimated_input_tokens": 24550, + "estimated_eval_tokens": 49356 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a grab bag of unrelated PRs. I only found one true duplicate-like pair: two CONTRIBUTING.md link fixes for the supported Python versions reference. Everything else is either a different feature, different bug, or just the same broad subsystem without the same concrete change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 7638 is the cleanest canonical candidate: it directly fixes the broken supported-Python link in CONTRIBUTING.md with no extra unrelated changes, making it the most precise representative of that duplicate pair.", + "best_issue_reason": null, + "best_pr_reason": "PR 7638 is the best global PR pick because it is a focused, self-contained fix for the specific bad-link problem, whereas the nearby PR 6014 also bundles an unrelated Makefile formatting change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11614", + "right": "pull_request:12397", + "accept": false, + "reason": "Both touch CI workflows, but one is about build-docker security and the other disables installing main-branch transformers/accelerate. Different concrete fixes." + }, + { + "left": "pull_request:11655", + "right": "pull_request:12419", + "accept": false, + "reason": "One adds bnb 8-bit LoRA support and tests; the other is a Flux/Mellon modular-pipeline change. Same broad area, not the same bug or change." + }, + { + "left": "pull_request:11699", + "right": "pull_request:12331", + "accept": false, + "reason": "Docs-only PRs in different areas: fp8 hardware tips vs parallel inference docs restructuring. No shared concrete issue." + }, + { + "left": "pull_request:12331", + "right": "pull_request:13165", + "accept": false, + "reason": "Parallel inference docs vs FA3 hub wrapper implementation. Not the same artifact or fix." + }, + { + "left": "pull_request:13164", + "right": "pull_request:13454", + "accept": false, + "reason": "Different CI/test changes: dropping test deps vs xfail-ing CLIP-related tests. Not a duplicate." + }, + { + "left": "pull_request:13164", + "right": "pull_request:13562", + "accept": false, + "reason": "Both are CI workflow edits, but one removes test deps and the other pins tokenizers. Different underlying change." + }, + { + "left": "pull_request:2795", + "right": "pull_request:3912", + "accept": false, + "reason": "Both relate to LoRA training, but they target different scripts and different stages of the feature. Not the same concrete code-path fix." + }, + { + "left": "pull_request:3928", + "right": "pull_request:4248", + "accept": false, + "reason": "ControlNet timestep guidance bug fix vs accelerator tracker input cleanup. Unrelated issues." + }, + { + "left": "pull_request:5299", + "right": "pull_request:7134", + "accept": false, + "reason": "Both are type-hint/doc cleanup in model code, but they touch different model files and different changes. Too broad to treat as one fix." + }, + { + "left": "pull_request:5771", + "right": "pull_request:6961", + "accept": false, + "reason": "A broad loaders refactor is not the same as single-file loader tests and conversion-script tweaks. Different underlying work." + }, + { + "left": "pull_request:5943", + "right": "pull_request:7650", + "accept": false, + "reason": "Gradient-checkpointing support in Transformer2DModel and an SDXL InstructPix2Pix arg addition are unrelated." + }, + { + "left": "pull_request:8290", + "right": "pull_request:8761", + "accept": false, + "reason": "Both are Hunyuan/Hunyuan-DIT changes, but one is a large refactor and the other is a specific embedding refactor. Not the same bug or patch." + }, + { + "left": "pull_request:11290", + "right": "pull_request:12451", + "accept": false, + "reason": "Dockerfile changes for different images and different dependencies; no shared concrete fix." + }, + { + "left": "pull_request:3852", + "right": "pull_request:4744", + "accept": false, + "reason": "Contributor-note docs vs AudioLDM docstring correction. Different documentation targets and problems." + }, + { + "left": "pull_request:4301", + "right": "pull_request:4352", + "accept": false, + "reason": "Kandinsky docs update vs combined-pipeline feature fixes. Same domain, but not the same change." + }, + { + "left": "pull_request:5157", + "right": "pull_request:7158", + "accept": false, + "reason": "LoRA conversion bug fix vs benchmark cleanup. Completely different purpose." + }, + { + "left": "pull_request:6014", + "right": "pull_request:7638", + "accept": true, + "reason": "Both fix the same broken CONTRIBUTING.md link to supported Python versions. 6014 also tweaks Makefile formatting, but the core artifact is the same link correction." + }, + { + "left": "pull_request:8684", + "right": "pull_request:8685", + "accept": false, + "reason": "Both are docs-level patches around different topics (Pillow import example vs float8 note). No duplicate bug/change." + }, + { + "left": "pull_request:6241", + "right": "pull_request:6787", + "accept": false, + "reason": "One addresses a float/half runtime error, the other fixes Marigold input normalization. Different code paths." + }, + { + "left": "pull_request:10576", + "right": "pull_request:11595", + "accept": false, + "reason": "Both mention LoRA loading with quantization, but one is a test addition for 8-bit bnb and the other is a Flux-specific loader fix. Not the same concrete fix." + }, + { + "left": "pull_request:11039", + "right": "pull_request:12277", + "accept": false, + "reason": "Single-file torch_dtype handling vs disabling xformers tests for some pipelines. Different concerns entirely." + }, + { + "left": "pull_request:11583", + "right": "pull_request:11618", + "accept": false, + "reason": "Both are docs/navigation updates, but one removes a tutorial and the other rewires PyTorch 2.0 doc links. Not a duplicate." + }, + { + "left": "pull_request:6003", + "right": "pull_request:6484", + "accept": false, + "reason": "Same SVD pipeline file, but one adds SVD inpaint and the other fixes GPU generator handling. Different concrete changes." + }, + { + "left": "pull_request:12806", + "right": "pull_request:13280", + "accept": false, + "reason": "Distributed logging/progress handling vs FlashAttention-4 support. Different subsystems and fixes." + }, + { + "left": "pull_request:1128", + "right": "pull_request:2040", + "accept": false, + "reason": "Inpaint tensor-input support vs adding InstructPix2Pix pipeline. Unrelated features." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The only accepted soft-edge pair, 6014/7638, is plausibly mergeable as a single CONTRIBUTING.md link fix despite 6014 also including a minor Makefile whitespace cleanup. The rest of the pairwise rejections are appropriately cautious and supported by the listed diffs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 155, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10000", + "pull_request:10008", + "pull_request:10069", + "pull_request:10101", + "pull_request:10114", + "pull_request:10122", + "pull_request:10132", + "pull_request:10192", + "pull_request:102", + "pull_request:10218", + "pull_request:10554", + "pull_request:10728", + "pull_request:10808", + "pull_request:11081", + "pull_request:11479", + "pull_request:11639", + "pull_request:11904", + "pull_request:11913", + "pull_request:12028", + "pull_request:12249", + "pull_request:12416", + "pull_request:12523", + "pull_request:12909", + "pull_request:1971", + "pull_request:4062", + "pull_request:415", + "pull_request:433", + "pull_request:4408", + "pull_request:5448", + "pull_request:5653", + "pull_request:5887", + "pull_request:6484", + "pull_request:6646", + "pull_request:6752", + "pull_request:6868", + "pull_request:7476", + "pull_request:8631", + "pull_request:8941", + "pull_request:9077", + "pull_request:9175", + "pull_request:9234", + "pull_request:9542", + "pull_request:9721", + "pull_request:9739", + "pull_request:9840", + "pull_request:9843", + "pull_request:9896", + "pull_request:9934" + ], + "soft_pairs": [ + "pull_request:11904|pull_request:12028", + "pull_request:7476|pull_request:9234", + "pull_request:11639|pull_request:11913", + "pull_request:10192|pull_request:10728", + "pull_request:5448|pull_request:6752", + "pull_request:12249|pull_request:12416", + "pull_request:5887|pull_request:6868", + "pull_request:11081|pull_request:9840", + "pull_request:4062|pull_request:4408", + "pull_request:12523|pull_request:12909", + "pull_request:10808|pull_request:9542", + "pull_request:4408|pull_request:5653", + "pull_request:6484|pull_request:6646", + "pull_request:10000|pull_request:9934", + "pull_request:10008|pull_request:8941", + "pull_request:10069|pull_request:9721", + "pull_request:10101|pull_request:9843", + "pull_request:10114|pull_request:8631", + "pull_request:10122|pull_request:9896", + "pull_request:10132|pull_request:11479", + "pull_request:10192|pull_request:9175", + "pull_request:102|pull_request:1971", + "pull_request:102|pull_request:415", + "pull_request:102|pull_request:433", + "pull_request:10218|pull_request:10554", + "pull_request:10218|pull_request:9077", + "pull_request:10218|pull_request:9739" + ], + "prepared_review_unit_hash": "ed7f81e62309902f57d6b74f50524e97d2a35d0bb9db458b8b468c9cb5e746e0", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 104203, + "estimated_input_tokens": 26051, + "estimated_eval_tokens": 52358 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a noisy mix of unrelated PRs that only look similar at a broad subsystem level (model loading, Flux/Wan, ControlNet, docs, tests). None of the soft-edge pairs look like the same underlying change, so I would not merge any of them into a duplicate cluster.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:11904", + "right": "pull_request:12028", + "accept": false, + "reason": "Both touch loading utilities, but 11904 is a non-blocking single-file load speedup while 12028 is parallel shard loading; different mechanisms and bug targets." + }, + { + "left": "pull_request:7476", + "right": "pull_request:9234", + "accept": false, + "reason": "7476 fixes ControlNetUnion callback behavior in SDXL inpaint; 9234 fixes a safetensors compatibility regression. Same area, different problem." + }, + { + "left": "pull_request:11639", + "right": "pull_request:11913", + "accept": false, + "reason": "11639 changes Wan video-to-video latent sampling; 11913 adds the Wan modular pipeline. Not the same code-path or change." + }, + { + "left": "pull_request:10192", + "right": "pull_request:10728", + "accept": false, + "reason": "10192 adds Flux Control/Inpaint support; 10728 changes Flux IP-Adapter handling for quantized models. Different features and fixes." + }, + { + "left": "pull_request:5448", + "right": "pull_request:6752", + "accept": false, + "reason": "Both add new research pipelines, but they are entirely different models and implementations, not the same underlying change." + }, + { + "left": "pull_request:12249", + "right": "pull_request:12416", + "accept": false, + "reason": "12249 adds Step1X-Edit; 12416 adds QwenImage Edit Plus modular support. Distinct pipelines/models." + }, + { + "left": "pull_request:5887", + "right": "pull_request:6868", + "accept": false, + "reason": "5887 adds IP-Adapter to ControlNet inpaint, while 6868 propagates ip_adapter_image_embeds through SD pipelines generally. Related topic, but not the same fix." + }, + { + "left": "pull_request:11081", + "right": "pull_request:9840", + "accept": false, + "reason": "11081 implements record_stream for group offloading; 9840 relaxes CUDA placement restrictions for bnb-loaded pipelines. Different behavior and code paths." + }, + { + "left": "pull_request:4062", + "right": "pull_request:4408", + "accept": false, + "reason": "4062 adds Shap-E mesh output; 4408 supports SDXL inpainting from single-file checkpoints. Unrelated changes." + }, + { + "left": "pull_request:12523", + "right": "pull_request:12909", + "accept": false, + "reason": "12523 fixes AutoencoderKLWan skip keys; 12909 fixes Wan 2.1 context parallel attention. Separate bugs in different layers." + }, + { + "left": "pull_request:10808", + "right": "pull_request:9542", + "accept": false, + "reason": "10808 tightens PEFT config key handling; 9542 improves conversion of Kohya Flux LoRAs with text encoder keys. Not the same bug." + }, + { + "left": "pull_request:4408", + "right": "pull_request:5653", + "accept": false, + "reason": "4408 is SDXL single-file inpainting support; 5653 is a PEFT-related maybe_raise_or_warn enhancement. No shared underlying change." + }, + { + "left": "pull_request:6484", + "right": "pull_request:6646", + "accept": false, + "reason": "6484 fixes GPU-created generators in SVD; 6646 changes SVD resolution constraints. Same model family, different issues." + }, + { + "left": "pull_request:10000", + "right": "pull_request:9934", + "accept": false, + "reason": "10000 adds Mochi video-to-video support; 9934 updates Mochi docs. One is code, the other documentation." + }, + { + "left": "pull_request:10008", + "right": "pull_request:8941", + "accept": false, + "reason": "10008 fixes a CogVideoX docs table; 8941 fixes an SD3 guidance_scale docstring mismatch. Both docs-only and unrelated." + }, + { + "left": "pull_request:10069", + "right": "pull_request:9721", + "accept": false, + "reason": "10069 corrects a BnB device type check; 9721 pins torch versions in Dockerfiles. Different concerns." + }, + { + "left": "pull_request:10101", + "right": "pull_request:9843", + "accept": false, + "reason": "10101 adds a generic sigmas test; 9843 adjusts an SD3 LoRA test fixture. Testing-related, but not the same concrete issue." + }, + { + "left": "pull_request:10114", + "right": "pull_request:8631", + "accept": false, + "reason": "10114 is a formatting-only stable audio cleanup; 8631 fixes from_single_file for T5 checkpoints. Completely different." + }, + { + "left": "pull_request:10122", + "right": "pull_request:9896", + "accept": false, + "reason": "10122 passes attn_mask into Flux attention; 9896 reverts Flux device/typecasting changes. Different code paths and fixes." + }, + { + "left": "pull_request:10132", + "right": "pull_request:11479", + "accept": false, + "reason": "10132 updates tokenizers in CI; 11479 renames Habana Gaudi docs to Intel Gaudi. Unrelated." + }, + { + "left": "pull_request:10192", + "right": "pull_request:9175", + "accept": false, + "reason": "Both are Flux ControlNet-related, but 10192 adds Flux Control/Inpaint docs and pipeline support while 9175 adds Union ControlNet support. Different features, not duplicates." + }, + { + "left": "pull_request:102", + "right": "pull_request:1971", + "accept": false, + "reason": "102 updates training example commands; 1971 only edits a scheduler docstring. Not the same change." + }, + { + "left": "pull_request:102", + "right": "pull_request:415", + "accept": false, + "reason": "102 updates examples/README and training scripts, while 415 restructures training docs. Same broad docs area, but different artifacts and scope." + }, + { + "left": "pull_request:102", + "right": "pull_request:433", + "accept": false, + "reason": "102 is training example maintenance; 433 updates black and unrelated code/style files. Not the same underlying bug or change." + }, + { + "left": "pull_request:10218", + "right": "pull_request:10554", + "accept": false, + "reason": "10218 fixes ControlNetUnion callback tensor inputs; 10554 fixes an SD3 LoRA training script key filter. Different pipelines and bugs." + }, + { + "left": "pull_request:10218", + "right": "pull_request:9077", + "accept": false, + "reason": "10218 is a functional ControlNetUnion fix; 9077 is a typo-only cleanup. No overlap in the actual change." + }, + { + "left": "pull_request:10218", + "right": "pull_request:9739", + "accept": false, + "reason": "10218 changes ControlNetUnion callback inputs; 9739 adds SD3 single-file docs. Different purpose and files." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 156, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10230", + "pull_request:10316", + "pull_request:10322", + "pull_request:10376", + "pull_request:10378", + "pull_request:1043", + "pull_request:10430", + "pull_request:10431", + "pull_request:10434", + "pull_request:10473", + "pull_request:10491", + "pull_request:10524", + "pull_request:10554", + "pull_request:10567", + "pull_request:10579", + "pull_request:10810", + "pull_request:10834", + "pull_request:10850", + "pull_request:1087", + "pull_request:10894", + "pull_request:10922", + "pull_request:10941", + "pull_request:10990", + "pull_request:11087", + "pull_request:11113", + "pull_request:11194", + "pull_request:11212", + "pull_request:11238", + "pull_request:11451", + "pull_request:11556", + "pull_request:11812", + "pull_request:11813", + "pull_request:11838", + "pull_request:11975", + "pull_request:11999", + "pull_request:2759", + "pull_request:3108", + "pull_request:774", + "pull_request:8755", + "pull_request:8845", + "pull_request:8920", + "pull_request:9036", + "pull_request:9116", + "pull_request:9687", + "pull_request:9696", + "pull_request:9736", + "pull_request:9918", + "pull_request:9934" + ], + "soft_pairs": [ + "pull_request:10230|pull_request:11556", + "pull_request:10230|pull_request:8755", + "pull_request:10316|pull_request:9934", + "pull_request:10322|pull_request:10431", + "pull_request:10376|pull_request:10567", + "pull_request:10378|pull_request:11238", + "pull_request:1043|pull_request:2759", + "pull_request:1043|pull_request:774", + "pull_request:10430|pull_request:9736", + "pull_request:10434|pull_request:8920", + "pull_request:10473|pull_request:11813", + "pull_request:10491|pull_request:8845", + "pull_request:10524|pull_request:9934", + "pull_request:10554|pull_request:9036", + "pull_request:10554|pull_request:9116", + "pull_request:10579|pull_request:9687", + "pull_request:10810|pull_request:11999", + "pull_request:10834|pull_request:11451", + "pull_request:10850|pull_request:9687", + "pull_request:10850|pull_request:9918", + "pull_request:1087|pull_request:3108", + "pull_request:10894|pull_request:9696", + "pull_request:10922|pull_request:11087", + "pull_request:10941|pull_request:11975", + "pull_request:10990|pull_request:11812", + "pull_request:11113|pull_request:11975", + "pull_request:11194|pull_request:11838", + "pull_request:11212|pull_request:9934" + ], + "prepared_review_unit_hash": "79307d1870402690e1183df572775b9daca14c8fa811912d33cf9bb4ae2e6cca", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 101040, + "estimated_input_tokens": 25260, + "estimated_eval_tokens": 50776 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose similarity group, not a true duplicate set: it mixes unrelated docs updates, training-script fixes, model additions, and pipeline features. None of the soft-edge pairs look like the same underlying bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a good canonical for the cluster; the items span unrelated codepaths and content areas (docs, CI, training scripts, model/pipeline feature work).", + "best_issue_reason": null, + "best_pr_reason": "No PR stands out as a global representative duplicate target because the cluster is not centered on one concrete fix or feature.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10230", + "right": "pull_request:11556", + "accept": false, + "reason": "Both are docs-only, but one adds LoRA loader mixins and the other adds regional compilation docs; unrelated content and codepaths." + }, + { + "left": "pull_request:10230", + "right": "pull_request:8755", + "accept": false, + "reason": "LoRA docs update vs SD3 DreamBooth training logic fix; same broad area only by keyword, not the same change." + }, + { + "left": "pull_request:10316", + "right": "pull_request:9934", + "accept": false, + "reason": "Different documentation fixes for different pipelines; no shared bug or underlying change." + }, + { + "left": "pull_request:10322", + "right": "pull_request:10431", + "accept": false, + "reason": "Sana converter/model support vs Sana transformer PE fix; both Sana-related but distinct problems and files." + }, + { + "left": "pull_request:10376", + "right": "pull_request:10567", + "accept": false, + "reason": "HunyuanVideo LoRA conversion support vs HunyuanVideo true-CFG pipeline feature; unrelated code changes." + }, + { + "left": "pull_request:10378", + "right": "pull_request:11238", + "accept": false, + "reason": "Both touch training scripts, but one is SD3 LoRA saving and the other is SD3 ControlNet validation on A100; different fixes." + }, + { + "left": "pull_request:1043", + "right": "pull_request:2759", + "accept": false, + "reason": "The shared tests are superficial; one adds torch_all_close helper, the other is a broad attention/model refactor." + }, + { + "left": "pull_request:1043", + "right": "pull_request:774", + "accept": false, + "reason": "Test-helper/print improvements vs unrelated historical pipeline test tweaks; not the same defect." + }, + { + "left": "pull_request:10430", + "right": "pull_request:9736", + "accept": false, + "reason": "TeaCache core hooks work vs Allegro model/pipeline addition; completely different changes." + }, + { + "left": "pull_request:10434", + "right": "pull_request:8920", + "accept": false, + "reason": "Broken docs links across many pages vs a small PAG guide wording fix; same docs area only, not same issue." + }, + { + "left": "pull_request:10473", + "right": "pull_request:11813", + "accept": false, + "reason": "General advanced training README text vs Flux Kontext training code and docs; unrelated feature/documentation work." + }, + { + "left": "pull_request:10491", + "right": "pull_request:8845", + "accept": false, + "reason": "CI filename fix vs PAG test typo fix; both tests/CI, but unrelated concrete changes." + }, + { + "left": "pull_request:10524", + "right": "pull_request:9934", + "accept": false, + "reason": "HunyuanVideo checkpoint-id correction vs Mochi docs overhaul; different pipelines and no common bug." + }, + { + "left": "pull_request:10554", + "right": "pull_request:9036", + "accept": false, + "reason": "SD3 miniature training loader bugfix vs nightly dependency install fix; unrelated scopes." + }, + { + "left": "pull_request:10554", + "right": "pull_request:9116", + "accept": false, + "reason": "SD3 training-script fix vs a dead-link docs fix; no shared codepath." + }, + { + "left": "pull_request:10579", + "right": "pull_request:9687", + "accept": false, + "reason": "A new community pipeline README/code addition vs Flux ControlNet support; different features." + }, + { + "left": "pull_request:10810", + "right": "pull_request:11999", + "accept": false, + "reason": "Flux LoRA conversion logic vs PEFT version/config-injection adaptation; the latter is dependency compatibility, not the same bug." + }, + { + "left": "pull_request:10834", + "right": "pull_request:11451", + "accept": false, + "reason": "GGUF isinstance bugfix vs Wan transformer compile test; one fixes runtime logic, the other adds test coverage." + }, + { + "left": "pull_request:10850", + "right": "pull_request:9687", + "accept": false, + "reason": "Hooks utility refactor vs Xlabs ControlNet support; not the same concrete code-path problem." + }, + { + "left": "pull_request:10850", + "right": "pull_request:9918", + "accept": false, + "reason": "Hooks utility refactor vs server example/docs addition; unrelated." + }, + { + "left": "pull_request:1087", + "right": "pull_request:3108", + "accept": false, + "reason": "Slow-test adjustments vs an empty/ckpt-fix PR with no comparable change; not duplicates." + }, + { + "left": "pull_request:10894", + "right": "pull_request:9696", + "accept": false, + "reason": "Docs toctree sorting vs PR workflow test runner changes; unrelated documentation vs CI automation edits." + }, + { + "left": "pull_request:10922", + "right": "pull_request:11087", + "accept": false, + "reason": "Wan model/pipeline addition vs Wan I2V quality fix; same product line, but different code paths and merge scope." + }, + { + "left": "pull_request:10941", + "right": "pull_request:11975", + "accept": false, + "reason": "Flux pipeline copy-fixes vs a style fix in a Flux advanced training script; unrelated changes." + }, + { + "left": "pull_request:10990", + "right": "pull_request:11812", + "accept": false, + "reason": "Flux LoRA quantization support vs Flux Kontext pipeline feature; different underlying functionality." + }, + { + "left": "pull_request:11113", + "right": "pull_request:11975", + "accept": false, + "reason": "AutoencoderKL training bugfix vs Flux training script style fix; unrelated models and issues." + }, + { + "left": "pull_request:11194", + "right": "pull_request:11838", + "accept": false, + "reason": "Loading guide docs for dtype maps vs deprecated-pipelines warning docs; both docs-only but unrelated subjects." + }, + { + "left": "pull_request:11212", + "right": "pull_request:9934", + "accept": false, + "reason": "MPS badge/docs update across many pipeline pages vs Mochi docs; same docs theme only, not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 157, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11237", + "pull_request:11254", + "pull_request:1133", + "pull_request:11416", + "pull_request:11472", + "pull_request:115", + "pull_request:11556", + "pull_request:1162", + "pull_request:11621", + "pull_request:11635", + "pull_request:11684", + "pull_request:11767", + "pull_request:118", + "pull_request:11814", + "pull_request:11838", + "pull_request:1184", + "pull_request:11848", + "pull_request:11892", + "pull_request:11894", + "pull_request:11904", + "pull_request:11916", + "pull_request:1194", + "pull_request:11975", + "pull_request:120", + "pull_request:12115", + "pull_request:12231", + "pull_request:12277", + "pull_request:12331", + "pull_request:12340", + "pull_request:12398", + "pull_request:12419", + "pull_request:12519", + "pull_request:12644", + "pull_request:12747", + "pull_request:12960", + "pull_request:1765", + "pull_request:1971", + "pull_request:2101", + "pull_request:2322", + "pull_request:2691", + "pull_request:2808", + "pull_request:498", + "pull_request:705", + "pull_request:824", + "pull_request:840", + "pull_request:9688", + "pull_request:9827", + "pull_request:9878" + ], + "soft_pairs": [ + "pull_request:11237|pull_request:9688", + "pull_request:11254|pull_request:9827", + "pull_request:1133|pull_request:2808", + "pull_request:1133|pull_request:705", + "pull_request:1133|pull_request:824", + "pull_request:11416|pull_request:12231", + "pull_request:11472|pull_request:9878", + "pull_request:115|pull_request:498", + "pull_request:11556|pull_request:12331", + "pull_request:1162|pull_request:840", + "pull_request:11621|pull_request:12340", + "pull_request:11621|pull_request:12398", + "pull_request:11635|pull_request:12115", + "pull_request:11684|pull_request:11904", + "pull_request:11767|pull_request:11814", + "pull_request:11767|pull_request:12747", + "pull_request:118|pull_request:1765", + "pull_request:11838|pull_request:12644", + "pull_request:1184|pull_request:2691", + "pull_request:11848|pull_request:12644", + "pull_request:11892|pull_request:12277", + "pull_request:11894|pull_request:12519", + "pull_request:11916|pull_request:12419", + "pull_request:1194|pull_request:2101", + "pull_request:1194|pull_request:2322", + "pull_request:11975|pull_request:12960", + "pull_request:120|pull_request:1971" + ], + "prepared_review_unit_hash": "4b154bba76fc9fe27acf2a566862a7b7526f56d37fdeff2ef0a2ea6de10f00ed", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 106819, + "estimated_input_tokens": 26705, + "estimated_eval_tokens": 53666 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous grab-bag of unrelated PRs across docs, examples, dependency tweaks, pipeline additions, and separate bug fixes. None of the suggested soft edges look like the same underlying change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 11904 is the strongest standalone representative: it is a broad, self-contained code change that improves model loading across multiple loaders and utility paths, unlike the mostly narrow docs/example tweaks elsewhere in the set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11237", + "right": "pull_request:9688", + "accept": false, + "reason": "One is a tiny docs dtype-map example update; the other is a Flux ControlNet docs change. Different topics and no shared code path." + }, + { + "left": "pull_request:11254", + "right": "pull_request:9827", + "accept": false, + "reason": "ConsisID import fixes and LoRA device_map loading fixes are separate dependency/runtime issues in different areas." + }, + { + "left": "pull_request:1133", + "right": "pull_request:2808", + "accept": false, + "reason": "A Windows CI workflow and a test relaxation for DiT are unrelated changes." + }, + { + "left": "pull_request:1133", + "right": "pull_request:705", + "accept": false, + "reason": "Both are CI-related, but one adds a Windows workflow while the other changes push-test GPU setup; not the same bug or change." + }, + { + "left": "pull_request:1133", + "right": "pull_request:824", + "accept": false, + "reason": "CI workflow setup and a Flax test addition are unrelated." + }, + { + "left": "pull_request:11416", + "right": "pull_request:12231", + "accept": false, + "reason": "Both are docs typo cleanups, but they touch different files and different text; not one underlying bug." + }, + { + "left": "pull_request:11472", + "right": "pull_request:9878", + "accept": false, + "reason": "DreamBooth interpolation defaults and gradient-checkpointing behavior in model forwards are unrelated code paths." + }, + { + "left": "pull_request:115", + "right": "pull_request:498", + "accept": false, + "reason": "Pipeline renames and attention-slicing bug fixes are different changes." + }, + { + "left": "pull_request:11556", + "right": "pull_request:12331", + "accept": false, + "reason": "One adds regional compilation docs; the other reorganizes parallel inference docs. Same broad docs area, but not the same change." + }, + { + "left": "pull_request:1162", + "right": "pull_request:840", + "accept": false, + "reason": "A links/doc update and a new community example are unrelated." + }, + { + "left": "pull_request:11621", + "right": "pull_request:12340", + "accept": false, + "reason": "Two different pipeline additions for different models; not the same feature or code path." + }, + { + "left": "pull_request:11621", + "right": "pull_request:12398", + "accept": false, + "reason": "Two separate new pipelines for different architectures/models, not a duplicate change." + }, + { + "left": "pull_request:11635", + "right": "pull_request:12115", + "accept": false, + "reason": "Flux Kontext pipeline work and an AnimateDiff test are entirely different artifacts." + }, + { + "left": "pull_request:11684", + "right": "pull_request:11904", + "accept": false, + "reason": "Both touch single-file loading, but one adds GGUF checkpoint support while the other speeds up loading and sync behavior; different concrete behavior." + }, + { + "left": "pull_request:11767", + "right": "pull_request:11814", + "accept": false, + "reason": "Group-offloading memory-address fixes and Flux gradient-checkpointing kwargs plumbing are unrelated." + }, + { + "left": "pull_request:11767", + "right": "pull_request:12747", + "accept": false, + "reason": "Same subsystem, but one fixes disk-offload address collisions and the other adds pin-group feature semantics; not the same underlying bug." + }, + { + "left": "pull_request:118", + "right": "pull_request:1765", + "accept": false, + "reason": "A ScoreSdeVe pipeline rename and Flax dreambooth fixes are unrelated." + }, + { + "left": "pull_request:11838", + "right": "pull_request:12644", + "accept": false, + "reason": "Deprecated-pipelines docs and AutoModel docs are separate documentation changes." + }, + { + "left": "pull_request:1184", + "right": "pull_request:2691", + "accept": false, + "reason": "Adding flash attention and renaming attention APIs are different code changes, even though both touch attention-related files." + }, + { + "left": "pull_request:11848", + "right": "pull_request:12644", + "accept": false, + "reason": "LoRA metadata docs and AutoModel docs are unrelated docs updates." + }, + { + "left": "pull_request:11892", + "right": "pull_request:12277", + "accept": false, + "reason": "Removing k-diffusion from core dependencies and disabling xformers tests are unrelated maintenance changes." + }, + { + "left": "pull_request:11894", + "right": "pull_request:12519", + "accept": false, + "reason": "Pinning k-diffusion for CI and fixing toctree sorting are unrelated." + }, + { + "left": "pull_request:11916", + "right": "pull_request:12419", + "accept": false, + "reason": "A large Flux/Chroma single-file refactor and a modular-pipeline metadata tweak for Flux are not the same concrete change." + }, + { + "left": "pull_request:1194", + "right": "pull_request:2101", + "accept": false, + "reason": "Documentation workflow updates and a pix2pix example doc/code update are unrelated." + }, + { + "left": "pull_request:1194", + "right": "pull_request:2322", + "accept": false, + "reason": "PR docs action workflow changes and a broad example/inpainting update are unrelated." + }, + { + "left": "pull_request:11975", + "right": "pull_request:12960", + "accept": false, + "reason": "A style-only DreamBooth Flux change and an FSDP detection fix in Flux2 scripts are different issues." + }, + { + "left": "pull_request:120", + "right": "pull_request:1971", + "accept": false, + "reason": "A README expansion and a scheduler docstring cleanup are not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 158, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:120", + "pull_request:12099", + "pull_request:12109", + "pull_request:1219", + "pull_request:12244", + "pull_request:12265", + "pull_request:1229", + "pull_request:12340", + "pull_request:12387", + "pull_request:12398", + "pull_request:1240", + "pull_request:12420", + "pull_request:12451", + "pull_request:12467", + "pull_request:12487", + "pull_request:1250", + "pull_request:12530", + "pull_request:12541", + "pull_request:12552", + "pull_request:12604", + "pull_request:12618", + "pull_request:12637", + "pull_request:12686", + "pull_request:1269", + "pull_request:12775", + "pull_request:12782", + "pull_request:1283", + "pull_request:12914", + "pull_request:12961", + "pull_request:13029", + "pull_request:13039", + "pull_request:13081", + "pull_request:1311", + "pull_request:13146", + "pull_request:13164", + "pull_request:13294", + "pull_request:13321", + "pull_request:13338", + "pull_request:13472", + "pull_request:13562", + "pull_request:1386", + "pull_request:2892", + "pull_request:2926", + "pull_request:3337", + "pull_request:433", + "pull_request:483", + "pull_request:748", + "pull_request:79" + ], + "soft_pairs": [ + "pull_request:120|pull_request:433", + "pull_request:12099|pull_request:12914", + "pull_request:12109|pull_request:12387", + "pull_request:1219|pull_request:2926", + "pull_request:12244|pull_request:12604", + "pull_request:12265|pull_request:12637", + "pull_request:12265|pull_request:13146", + "pull_request:1229|pull_request:748", + "pull_request:12340|pull_request:12961", + "pull_request:12398|pull_request:12961", + "pull_request:1240|pull_request:79", + "pull_request:12420|pull_request:13081", + "pull_request:12451|pull_request:13338", + "pull_request:12467|pull_request:12487", + "pull_request:12467|pull_request:12552", + "pull_request:1250|pull_request:2892", + "pull_request:12530|pull_request:13294", + "pull_request:12541|pull_request:13039", + "pull_request:12618|pull_request:13562", + "pull_request:12686|pull_request:12775", + "pull_request:1269|pull_request:3337", + "pull_request:12782|pull_request:13321", + "pull_request:1283|pull_request:483", + "pull_request:13029|pull_request:13472", + "pull_request:1311|pull_request:1386", + "pull_request:13164|pull_request:13294" + ], + "prepared_review_unit_hash": "6995bec11d9ac229ca4bf5ef087ea5a70c1726adf2967415c988c10284741dc3", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 102119, + "estimated_input_tokens": 25530, + "estimated_eval_tokens": 51316 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the items span unrelated docs, CI, model, scheduler, and pipeline changes. The soft edges are mostly superficial similarity (same subsystem, shared filenames, or both being typo/docs PRs) rather than the same underlying bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are a heterogeneous mix of unrelated docs, CI, model, scheduler, and pipeline changes.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good global representative; none of the items consolidates the cluster into one concrete fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:120", + "right": "pull_request:433", + "accept": false, + "reason": "Both touch README/docs, but one is a README example update and the other is a broad black-version/tooling refresh; not the same change." + }, + { + "left": "pull_request:12099", + "right": "pull_request:12914", + "accept": false, + "reason": "Both are tiny typo/device-revision fixes, but they affect different pipelines and different code paths." + }, + { + "left": "pull_request:12109", + "right": "pull_request:12387", + "accept": false, + "reason": "Scheduler feature work vs attention-backend support; same broad area only, not the same bug or fix." + }, + { + "left": "pull_request:1219", + "right": "pull_request:2926", + "accept": false, + "reason": "Both concern DDIM docs, but one is pipeline docs and the other is scheduler docs with unrelated wording cleanup." + }, + { + "left": "pull_request:12244", + "right": "pull_request:12604", + "accept": false, + "reason": "Both are doc-only, but one removes Flax from installation/docs while the other cleans up modular docs content." + }, + { + "left": "pull_request:12265", + "right": "pull_request:12637", + "accept": false, + "reason": "Both are typo-style cleanups, but they touch different files and unrelated content." + }, + { + "left": "pull_request:12265", + "right": "pull_request:13146", + "accept": false, + "reason": "A generic typo sweep is not the same as fixing an MT5Tokenizer type annotation in PAG HunYuanDiT." + }, + { + "left": "pull_request:1229", + "right": "pull_request:748", + "accept": false, + "reason": "Dreambooth README install guidance and push_to_hub call fixes are unrelated changes." + }, + { + "left": "pull_request:12340", + "right": "pull_request:12961", + "accept": false, + "reason": "Adding LucyEditPipeline and adding PhotoDoodle are different pipeline additions with no shared code-path bug." + }, + { + "left": "pull_request:12398", + "right": "pull_request:12961", + "accept": false, + "reason": "Mirage pipeline implementation and PhotoDoodle community README content are unrelated." + }, + { + "left": "pull_request:1240", + "right": "pull_request:79", + "accept": false, + "reason": "Both are early top-level model/pipeline additions, but they add different functionality and different code paths." + }, + { + "left": "pull_request:12420", + "right": "pull_request:13081", + "accept": false, + "reason": "Conditional torch-distributed import and contiguous Qwen hidden states fix separate runtime issues." + }, + { + "left": "pull_request:12451", + "right": "pull_request:13338", + "accept": false, + "reason": "Doc-builder Dockerfile packages vs GitHub workflow permissions are unrelated CI maintenance changes." + }, + { + "left": "pull_request:12467", + "right": "pull_request:12487", + "accept": false, + "reason": "QwenImage batch handling bug fix vs docs broken-link cleanup; not the same artifact or bug." + }, + { + "left": "pull_request:12467", + "right": "pull_request:12552", + "accept": false, + "reason": "A QwenImage runtime fix and Kandinsky5 doc typo fixes do not share the same underlying problem." + }, + { + "left": "pull_request:1250", + "right": "pull_request:2892", + "accept": false, + "reason": "Community pipeline addition and upscaler input-validation fix are different changes." + }, + { + "left": "pull_request:12530", + "right": "pull_request:13294", + "accept": false, + "reason": "Both address test stability, but they target different tests and different failure modes." + }, + { + "left": "pull_request:12541", + "right": "pull_request:13039", + "accept": false, + "reason": "Wan LoRA conversion logic and WanVACE CPU upcasting are separate Wan-related fixes, not one code-path issue." + }, + { + "left": "pull_request:12618", + "right": "pull_request:13562", + "accept": false, + "reason": "Both adjust CI install commands, but they fix different dependency/version problems." + }, + { + "left": "pull_request:12686", + "right": "pull_request:12775", + "accept": false, + "reason": "Both are docs-related and share toctree changes, but one adds Kandinsky5 pipelines while the other adds Z-Image docs." + }, + { + "left": "pull_request:1269", + "right": "pull_request:3337", + "accept": false, + "reason": "Both involve stable-diffusion-era refactors, but they refactor different code paths and different pipeline sets." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13321", + "accept": false, + "reason": "Test offloading tweaks for HunyuanVideo and a guarded torchvision import for Cosmos are unrelated." + }, + { + "left": "pull_request:1283", + "right": "pull_request:483", + "accept": false, + "reason": "Versatile Diffusion model addition vs typo/typo-check docs work are completely different changes." + }, + { + "left": "pull_request:13029", + "right": "pull_request:13472", + "accept": false, + "reason": "A modular quickstart rewrite and an agent-docs float64 gotcha note are both docs, but not the same documentation fix." + }, + { + "left": "pull_request:1311", + "right": "pull_request:1386", + "accept": false, + "reason": "AltDiffusion tests and scheduler v-prediction implementation are unrelated." + }, + { + "left": "pull_request:13164", + "right": "pull_request:13294", + "accept": false, + "reason": "CI dependency cleanup and a flaky WanVACE test fix are both test-related, but they address different failures and are not mergeable as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 159, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1325", + "pull_request:13296", + "pull_request:1341", + "pull_request:13536", + "pull_request:13562", + "pull_request:1393", + "pull_request:1402", + "pull_request:1406", + "pull_request:1415", + "pull_request:1458", + "pull_request:1468", + "pull_request:1487", + "pull_request:1548", + "pull_request:1588", + "pull_request:1667", + "pull_request:1706", + "pull_request:1713", + "pull_request:1724", + "pull_request:1751", + "pull_request:1757", + "pull_request:1761", + "pull_request:1765", + "pull_request:1766", + "pull_request:1850", + "pull_request:1868", + "pull_request:1939", + "pull_request:1965", + "pull_request:2009", + "pull_request:2314", + "pull_request:2322", + "pull_request:2341", + "pull_request:286", + "pull_request:2919", + "pull_request:3230", + "pull_request:3321", + "pull_request:3343", + "pull_request:34", + "pull_request:3402", + "pull_request:3433", + "pull_request:3653", + "pull_request:3712", + "pull_request:397", + "pull_request:41", + "pull_request:415", + "pull_request:418", + "pull_request:423", + "pull_request:538", + "pull_request:768" + ], + "soft_pairs": [ + "pull_request:1325|pull_request:3343", + "pull_request:1325|pull_request:3402", + "pull_request:1325|pull_request:3433", + "pull_request:1325|pull_request:41", + "pull_request:13296|pull_request:13562", + "pull_request:1341|pull_request:3230", + "pull_request:13536|pull_request:13562", + "pull_request:1393|pull_request:1406", + "pull_request:1402|pull_request:34", + "pull_request:1415|pull_request:1751", + "pull_request:1458|pull_request:2009", + "pull_request:1468|pull_request:2341", + "pull_request:1487|pull_request:3230", + "pull_request:1548|pull_request:2322", + "pull_request:1588|pull_request:2314", + "pull_request:1667|pull_request:1965", + "pull_request:1706|pull_request:3321", + "pull_request:1713|pull_request:423", + "pull_request:1724|pull_request:397", + "pull_request:1724|pull_request:415", + "pull_request:1724|pull_request:418", + "pull_request:1757|pull_request:3230", + "pull_request:1761|pull_request:3230", + "pull_request:1765|pull_request:768", + "pull_request:1766|pull_request:538", + "pull_request:1850|pull_request:2919", + "pull_request:1868|pull_request:286", + "pull_request:1868|pull_request:3653", + "pull_request:1868|pull_request:3712", + "pull_request:1939|pull_request:286", + "pull_request:1939|pull_request:3653", + "pull_request:1939|pull_request:3712" + ], + "prepared_review_unit_hash": "8fb2b8940c1026a8f9ad939f0810f80b434238eb61775aacddde7658fe643b32", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 106188, + "estimated_input_tokens": 26547, + "estimated_eval_tokens": 53350 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated feature additions, CI/workflow tweaks, docs updates, and test fixes. None of the soft edges should be accepted.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits: the artifacts span unrelated changes across pipelines, schedulers, CI, docs, and tests.", + "best_issue_reason": null, + "best_pr_reason": "PR 3230 is the broadest and most substantial artifact here, so it is the closest thing to a representative PR, but the cluster is too heterogeneous to treat as one change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1325", + "right": "pull_request:3343", + "accept": false, + "reason": "Different changes in different areas: OneFlow image2image support vs a DPM-Solver config fix." + }, + { + "left": "pull_request:1325", + "right": "pull_request:3402", + "accept": false, + "reason": "Unrelated codepaths: OneFlow pipeline work vs a Dockerfile cleanup." + }, + { + "left": "pull_request:1325", + "right": "pull_request:3433", + "accept": false, + "reason": "Different concerns entirely: OneFlow image2image support vs documentation formatting/style fixes." + }, + { + "left": "pull_request:1325", + "right": "pull_request:41", + "accept": false, + "reason": "Completely unrelated: OneFlow image2image work vs comment-only resnet cleanup." + }, + { + "left": "pull_request:13296", + "right": "pull_request:13562", + "accept": false, + "reason": "Both touch CI, but they fix different package/version issues in the same workflow, not the same bug or change." + }, + { + "left": "pull_request:1341", + "right": "pull_request:3230", + "accept": false, + "reason": "Both add pipeline exports, but they implement different features: Stable Diffusion repaint vs IF support." + }, + { + "left": "pull_request:13536", + "right": "pull_request:13562", + "accept": false, + "reason": "Workflow permission/behavior changes are unrelated to the tokenizers/transformers CI install update." + }, + { + "left": "pull_request:1393", + "right": "pull_request:1406", + "accept": false, + "reason": "Both are deprecation-related, but they target different config behaviors and different codepaths." + }, + { + "left": "pull_request:1402", + "right": "pull_request:34", + "accept": false, + "reason": "A pipeline-loading bug fix is unrelated to a typo/comment fix in a scheduler file." + }, + { + "left": "pull_request:1415", + "right": "pull_request:1751", + "accept": false, + "reason": "Both affect Stable Diffusion-family pipelines, but one fixes all Stable Diffusion code paths while the other fixes UnCLIP generator handling." + }, + { + "left": "pull_request:1458", + "right": "pull_request:2009", + "accept": false, + "reason": "ONNX runtime candidate detection is unrelated to textual inversion loader support." + }, + { + "left": "pull_request:1468", + "right": "pull_request:2341", + "accept": false, + "reason": "Saving trained betas in schedulers is unrelated to EMA saving in a training example." + }, + { + "left": "pull_request:1487", + "right": "pull_request:3230", + "accept": false, + "reason": "Docs for xFormers are unrelated to adding the IF pipeline." + }, + { + "left": "pull_request:1548", + "right": "pull_request:2322", + "accept": false, + "reason": "Bug-report template wording is unrelated to the example/pipeline code added in the other PR." + }, + { + "left": "pull_request:1588", + "right": "pull_request:2314", + "accept": false, + "reason": "Fast-test FP16 fixes are unrelated to image-slice tolerance updates." + }, + { + "left": "pull_request:1667", + "right": "pull_request:1965", + "accept": false, + "reason": "Scheduler config handling for SD conversion is unrelated to the checkpoint merger import fix." + }, + { + "left": "pull_request:1706", + "right": "pull_request:3321", + "accept": false, + "reason": "General docs refresh is unrelated to the controlling-generation documentation update." + }, + { + "left": "pull_request:1713", + "right": "pull_request:423", + "accept": false, + "reason": "Different changes: Stable Diffusion conversion cleanup vs output/documentation syntax improvements." + }, + { + "left": "pull_request:1724", + "right": "pull_request:397", + "accept": false, + "reason": "Both are docs, but one is xFormers guidance and the other is the quicktour rewrite; not the same change." + }, + { + "left": "pull_request:1724", + "right": "pull_request:415", + "accept": false, + "reason": "xFormers recommendations are unrelated to training overview docs refresh." + }, + { + "left": "pull_request:1724", + "right": "pull_request:418", + "accept": false, + "reason": "xFormers docs and DiffusionPipeline docs are separate documentation efforts." + }, + { + "left": "pull_request:1757", + "right": "pull_request:3230", + "accept": false, + "reason": "Adding resnet_time_scale_shift in versatile diffusion is unrelated to the IF pipeline feature." + }, + { + "left": "pull_request:1761", + "right": "pull_request:3230", + "accept": false, + "reason": "UnCLIP CPU offloading and IF pipeline addition are different features and codepaths." + }, + { + "left": "pull_request:1765", + "right": "pull_request:768", + "accept": false, + "reason": "Dreambooth Flax fixes are unrelated to the Dreambooth enhancement example." + }, + { + "left": "pull_request:1766", + "right": "pull_request:538", + "accept": false, + "reason": "UnCLIP version gating is unrelated to upgrading the huggingface-hub dependency." + }, + { + "left": "pull_request:1850", + "right": "pull_request:2919", + "accept": false, + "reason": "ResnetBlock2D tests and a 3D UNet test speedup are different test changes." + }, + { + "left": "pull_request:1868", + "right": "pull_request:286", + "accept": false, + "reason": "EMA decay logic in a training script is unrelated to a README link fix." + }, + { + "left": "pull_request:1868", + "right": "pull_request:3653", + "accept": false, + "reason": "EMA decay changes are unrelated to a linting cleanup in a community example." + }, + { + "left": "pull_request:1868", + "right": "pull_request:3712", + "accept": false, + "reason": "Training-script EMA logic is unrelated to documentation workflow secret renames." + }, + { + "left": "pull_request:1939", + "right": "pull_request:286", + "accept": false, + "reason": "README image path correction is unrelated to the stable-diffusion README link fix." + }, + { + "left": "pull_request:1939", + "right": "pull_request:3653", + "accept": false, + "reason": "A README asset-path fix is unrelated to a community-example lint fix." + }, + { + "left": "pull_request:1939", + "right": "pull_request:3712", + "accept": false, + "reason": "A README path correction is unrelated to documentation workflow secret changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 160, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1971", + "pull_request:2001", + "pull_request:2006", + "pull_request:2007", + "pull_request:21", + "pull_request:2138", + "pull_request:217", + "pull_request:2190", + "pull_request:2197", + "pull_request:22", + "pull_request:2208", + "pull_request:2227", + "pull_request:2295", + "pull_request:2314", + "pull_request:2322", + "pull_request:2329", + "pull_request:2362", + "pull_request:2388", + "pull_request:2401", + "pull_request:2436", + "pull_request:2475", + "pull_request:252", + "pull_request:2526", + "pull_request:2542", + "pull_request:2655", + "pull_request:2706", + "pull_request:286", + "pull_request:2912", + "pull_request:2958", + "pull_request:3069", + "pull_request:3172", + "pull_request:3262", + "pull_request:3400", + "pull_request:3402", + "pull_request:3581", + "pull_request:3708", + "pull_request:3712", + "pull_request:3757", + "pull_request:3786", + "pull_request:3916", + "pull_request:397", + "pull_request:4081", + "pull_request:415", + "pull_request:4239", + "pull_request:7", + "pull_request:774", + "pull_request:909", + "pull_request:979" + ], + "soft_pairs": [ + "pull_request:1971|pull_request:2007", + "pull_request:1971|pull_request:3262", + "pull_request:2001|pull_request:3708", + "pull_request:2006|pull_request:3712", + "pull_request:2007|pull_request:3581", + "pull_request:21|pull_request:7", + "pull_request:21|pull_request:909", + "pull_request:2138|pull_request:2227", + "pull_request:2138|pull_request:3402", + "pull_request:217|pull_request:286", + "pull_request:2190|pull_request:3916", + "pull_request:2197|pull_request:4239", + "pull_request:22|pull_request:7", + "pull_request:22|pull_request:909", + "pull_request:2208|pull_request:3172", + "pull_request:2295|pull_request:2706", + "pull_request:2314|pull_request:2655", + "pull_request:2322|pull_request:252", + "pull_request:2322|pull_request:3069", + "pull_request:2329|pull_request:774", + "pull_request:2362|pull_request:3786", + "pull_request:2388|pull_request:397", + "pull_request:2388|pull_request:415", + "pull_request:2401|pull_request:4081", + "pull_request:2436|pull_request:979", + "pull_request:2475|pull_request:2542", + "pull_request:2475|pull_request:2912", + "pull_request:2475|pull_request:2958", + "pull_request:2475|pull_request:3757", + "pull_request:2526|pull_request:3400" + ], + "prepared_review_unit_hash": "3f7317e577ce08128800498ea668fc7b8baaff6f0db81d7390b94ade6be28cdf", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 96641, + "estimated_input_tokens": 24161, + "estimated_eval_tokens": 48578 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: mostly unrelated docs/typo/test/example maintenance PRs, plus a few ControlNet-related feature PRs and other isolated changes. I don\u2019t see a single underlying bug or change that ties them together, so no soft edges are accepted.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR fits the whole cluster; the items span unrelated docs, tests, examples, workflows, and feature work.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong global representative. PR 2475 is the largest substantive item, but it only matches a small subset of the cluster and not the rest.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1971", + "right": "pull_request:2007", + "accept": false, + "reason": "Docstring cleanup vs broad black-formatting sweep; different changes." + }, + { + "left": "pull_request:1971", + "right": "pull_request:3262", + "accept": false, + "reason": "Docstring wording fix vs IF model-name update; unrelated." + }, + { + "left": "pull_request:2001", + "right": "pull_request:3708", + "accept": false, + "reason": "ColossalAI seed launch fix vs version-sorting utility change; different code paths." + }, + { + "left": "pull_request:2006", + "right": "pull_request:3712", + "accept": false, + "reason": "Checkpoint merger fix vs workflow secret rename; unrelated." + }, + { + "left": "pull_request:2007", + "right": "pull_request:3581", + "accept": false, + "reason": "Black formatting PR vs bit_diffusion import cleanup; not the same change." + }, + { + "left": "pull_request:21", + "right": "pull_request:7", + "accept": false, + "reason": "Model test additions vs initial VQ-VAE implementation; different artifacts." + }, + { + "left": "pull_request:21", + "right": "pull_request:909", + "accept": false, + "reason": "Different test updates on different model paths; not a duplicate change." + }, + { + "left": "pull_request:2138", + "right": "pull_request:2227", + "accept": false, + "reason": "Docs typo fix vs pipeline API typos across many files; different scope and content." + }, + { + "left": "pull_request:2138", + "right": "pull_request:3402", + "accept": false, + "reason": "Docs typo fix vs Dockerfile fix; unrelated." + }, + { + "left": "pull_request:217", + "right": "pull_request:286", + "accept": false, + "reason": "Safety-checker logic bug vs README link correction; different concerns." + }, + { + "left": "pull_request:2190", + "right": "pull_request:3916", + "accept": false, + "reason": "Unrelated branch syncs touching different example sets." + }, + { + "left": "pull_request:2197", + "right": "pull_request:4239", + "accept": false, + "reason": "Inpaint pipeline typo/input fix vs LoRA text-encoder support fix; different code paths." + }, + { + "left": "pull_request:22", + "right": "pull_request:7", + "accept": false, + "reason": "Modeling-utils warning typo vs VQ-VAE addition; unrelated." + }, + { + "left": "pull_request:22", + "right": "pull_request:909", + "accept": false, + "reason": "Warning-logging typo fix vs inpaint test/model-id update; different changes." + }, + { + "left": "pull_request:2208", + "right": "pull_request:3172", + "accept": false, + "reason": "Separate documentation updates on different topics." + }, + { + "left": "pull_request:2295", + "right": "pull_request:2706", + "accept": false, + "reason": "Configurable transformer-depth feature vs EMA save/restore training fix; not the same bug." + }, + { + "left": "pull_request:2314", + "right": "pull_request:2655", + "accept": false, + "reason": "Slow-test threshold tweaks vs LoRA loading/saving logic; unrelated." + }, + { + "left": "pull_request:2322", + "right": "pull_request:252", + "accept": false, + "reason": "New inpaint example vs README examples update; different deliverables." + }, + { + "left": "pull_request:2322", + "right": "pull_request:3069", + "accept": false, + "reason": "Example script addition vs Flax TPU test updates; unrelated." + }, + { + "left": "pull_request:2329", + "right": "pull_request:774", + "accept": false, + "reason": "Workflow refactor vs callback test adjustments; different code paths." + }, + { + "left": "pull_request:2362", + "right": "pull_request:3786", + "accept": false, + "reason": "MPS skip decorator vs PR template addition; unrelated." + }, + { + "left": "pull_request:2388", + "right": "pull_request:397", + "accept": false, + "reason": "Different documentation pages and scopes, not the same change." + }, + { + "left": "pull_request:2388", + "right": "pull_request:415", + "accept": false, + "reason": "Control docs vs training docs; unrelated documentation work." + }, + { + "left": "pull_request:2401", + "right": "pull_request:4081", + "accept": false, + "reason": "Docs index overhaul vs README table entry; not a duplicate." + }, + { + "left": "pull_request:2436", + "right": "pull_request:979", + "accept": false, + "reason": "Demo-link docs change vs missing import in README example; different artifacts." + }, + { + "left": "pull_request:2475", + "right": "pull_request:2542", + "accept": false, + "reason": "ControlNet feature work vs small docs fixes; unrelated." + }, + { + "left": "pull_request:2475", + "right": "pull_request:2912", + "accept": false, + "reason": "Major ControlNet addition vs punctuation/doc tweak; not the same change." + }, + { + "left": "pull_request:2475", + "right": "pull_request:2958", + "accept": false, + "reason": "Base ControlNet feature vs ControlNet inpaint variant; related subsystem but different concrete PRs." + }, + { + "left": "pull_request:2475", + "right": "pull_request:3757", + "accept": false, + "reason": "Base ControlNet feature vs conditioning-channels config change; different functionality." + }, + { + "left": "pull_request:2526", + "right": "pull_request:3400", + "accept": false, + "reason": "Test-threshold adjustments vs dependency/Dockerfile changes; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 161, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2533", + "pull_request:2590", + "pull_request:2693", + "pull_request:2701", + "pull_request:2759", + "pull_request:2762", + "pull_request:2811", + "pull_request:2814", + "pull_request:286", + "pull_request:2970", + "pull_request:3042", + "pull_request:3063", + "pull_request:3108", + "pull_request:3169", + "pull_request:3217", + "pull_request:3230", + "pull_request:3295", + "pull_request:3318", + "pull_request:3343", + "pull_request:34", + "pull_request:3402", + "pull_request:3433", + "pull_request:3474", + "pull_request:3524", + "pull_request:3525", + "pull_request:3608", + "pull_request:3707", + "pull_request:3709", + "pull_request:3931", + "pull_request:3955", + "pull_request:4081", + "pull_request:41", + "pull_request:4495", + "pull_request:4547", + "pull_request:4748", + "pull_request:4963", + "pull_request:4973", + "pull_request:5267", + "pull_request:5286", + "pull_request:5478", + "pull_request:5622", + "pull_request:5855", + "pull_request:693", + "pull_request:705", + "pull_request:828", + "pull_request:909", + "pull_request:948", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:2533|pull_request:4547", + "pull_request:2590|pull_request:2970", + "pull_request:2693|pull_request:3707", + "pull_request:2701|pull_request:3230", + "pull_request:2759|pull_request:828", + "pull_request:2762|pull_request:3230", + "pull_request:2811|pull_request:705", + "pull_request:2814|pull_request:909", + "pull_request:286|pull_request:41", + "pull_request:286|pull_request:693", + "pull_request:3042|pull_request:4963", + "pull_request:3063|pull_request:3955", + "pull_request:3108|pull_request:3931", + "pull_request:3169|pull_request:975", + "pull_request:3217|pull_request:3230", + "pull_request:3230|pull_request:3709", + "pull_request:3230|pull_request:4081", + "pull_request:3230|pull_request:4495", + "pull_request:3230|pull_request:4973", + "pull_request:3230|pull_request:5286", + "pull_request:3295|pull_request:3608", + "pull_request:3318|pull_request:4547", + "pull_request:3343|pull_request:5622", + "pull_request:34|pull_request:948", + "pull_request:3402|pull_request:5622", + "pull_request:3433|pull_request:5622", + "pull_request:3474|pull_request:5478", + "pull_request:3524|pull_request:3525", + "pull_request:3524|pull_request:3931", + "pull_request:3524|pull_request:4748", + "pull_request:3524|pull_request:5267", + "pull_request:3524|pull_request:5855" + ], + "prepared_review_unit_hash": "bfb48f8c2be7228a509717dad28da84e4ad5a30eb0c68b2c13c55fe37e205b34", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 93442, + "estimated_input_tokens": 23361, + "estimated_eval_tokens": 46978 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No duplicate cluster here: the PRs span unrelated docs, CI, example scripts, scheduler fixes, and feature work. The soft links are mostly superficial similarity on filenames or titles, not the same concrete change.", + "confidence": 0.97, + "canonical_issue_reason": "No issue artifacts are present in this cluster.", + "canonical_pr_reason": "No single PR is a credible canonical duplicate target; the items are heterogeneous and do not share one underlying bug or change.", + "best_issue_reason": "No issue artifacts are present in this cluster.", + "best_pr_reason": "No PR stands out as a global representative duplicate; the cluster is too mixed across unrelated changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2533", + "right": "pull_request:4547", + "accept": false, + "reason": "WIP testing rc1 vs Flax scheduler implementation; unrelated changes and code paths." + }, + { + "left": "pull_request:2590", + "right": "pull_request:2970", + "accept": false, + "reason": "CI dependency-table check vs a variable rename in a dreambooth example; not the same change." + }, + { + "left": "pull_request:2693", + "right": "pull_request:3707", + "accept": false, + "reason": "Doc image-link fix vs doc-build secrets/workflow changes; different files and problems." + }, + { + "left": "pull_request:2701", + "right": "pull_request:3230", + "accept": false, + "reason": "ONNX Runtime support list update vs the IF pipeline landing; shared utility area but unrelated functionality." + }, + { + "left": "pull_request:2759", + "right": "pull_request:828", + "accept": false, + "reason": "Large test-run/workflow debug PR vs Flax test completion; different scopes and code paths." + }, + { + "left": "pull_request:2762", + "right": "pull_request:3230", + "accept": false, + "reason": "ControlNet docs for multi-conditioning vs IF feature implementation; both docs/feature work but not the same change." + }, + { + "left": "pull_request:2811", + "right": "pull_request:705", + "accept": false, + "reason": "onnx PNDM test slice update vs CUDA CI workflow/dependency changes; unrelated." + }, + { + "left": "pull_request:2814", + "right": "pull_request:909", + "accept": false, + "reason": "Stable UnCLIP completion vs Stable Diffusion inpaint test/model-path updates; different pipeline work." + }, + { + "left": "pull_request:286", + "right": "pull_request:41", + "accept": false, + "reason": "Stable Diffusion README link fix vs ancient resnet comment cleanup; no shared bug." + }, + { + "left": "pull_request:286", + "right": "pull_request:693", + "accept": false, + "reason": "README link fix vs citation-format typo; only superficial README overlap." + }, + { + "left": "pull_request:3042", + "right": "pull_request:4963", + "accept": false, + "reason": "Accelerate minimum-version updates in example requirements vs installing accelerate main in CI; related dependency area but different concrete fixes." + }, + { + "left": "pull_request:3063", + "right": "pull_request:3955", + "accept": false, + "reason": "Pipeline __setattr__ None handling vs Kandinsky parameter docs; code fix vs documentation." + }, + { + "left": "pull_request:3108", + "right": "pull_request:3931", + "accept": false, + "reason": "Empty/unknown checkpoint fix vs ckpt loading overhaul; no evidence of same underlying bug." + }, + { + "left": "pull_request:3169", + "right": "pull_request:975", + "accept": false, + "reason": "Habana docs update vs OpenVINO title typo; different docs and unrelated content." + }, + { + "left": "pull_request:3217", + "right": "pull_request:3230", + "accept": false, + "reason": "Add pt_to_pil utility refactor vs IF feature addition; same module area but distinct changes." + }, + { + "left": "pull_request:3230", + "right": "pull_request:3709", + "accept": false, + "reason": "IF pipeline addition vs adding image-processor docs; unrelated docs/feature work." + }, + { + "left": "pull_request:3230", + "right": "pull_request:4081", + "accept": false, + "reason": "IF pipeline landing vs Kandinsky README table row; different features and files." + }, + { + "left": "pull_request:3230", + "right": "pull_request:4495", + "accept": false, + "reason": "IF pipeline addition vs SDXL reference community example; unrelated pipeline work." + }, + { + "left": "pull_request:3230", + "right": "pull_request:4973", + "accept": false, + "reason": "IF pipeline feature vs monthly install badge in README; no shared underlying change." + }, + { + "left": "pull_request:3230", + "right": "pull_request:5286", + "accept": false, + "reason": "IF pipeline addition vs Japanese docs skeleton; unrelated." + }, + { + "left": "pull_request:3295", + "right": "pull_request:3608", + "accept": false, + "reason": "Tutorial typo in PIL conversion vs train_text_to_image arg typo; both typo fixes but different code locations and behavior." + }, + { + "left": "pull_request:3318", + "right": "pull_request:4547", + "accept": false, + "reason": "WIP IF textual inversion vs Flax scheduler implementation; different model families and changes." + }, + { + "left": "pull_request:3343", + "right": "pull_request:5622", + "accept": false, + "reason": "DPMSolver config bugfix vs example-training optimization cleanup; unrelated." + }, + { + "left": "pull_request:34", + "right": "pull_request:948", + "accept": false, + "reason": "Comment typo cleanup in DDPM vs another DDPM comment typo fix years later; not the same concrete patch." + }, + { + "left": "pull_request:3402", + "right": "pull_request:5622", + "accept": false, + "reason": "Dockerfile dependency install fix vs large examples/optimization refactor; unrelated." + }, + { + "left": "pull_request:3433", + "right": "pull_request:5622", + "accept": false, + "reason": "Docs style rendering cleanup vs examples/optimization changes; not the same issue." + }, + { + "left": "pull_request:3474", + "right": "pull_request:5478", + "accept": false, + "reason": "Attention processor docs vs Japanese docs rollout; both documentation but completely different artifacts." + }, + { + "left": "pull_request:3524", + "right": "pull_request:3525", + "accept": false, + "reason": "Two README edits, but one rewrites documentation sections and the other tweaks a usage snippet and badges; not one merged change." + }, + { + "left": "pull_request:3524", + "right": "pull_request:3931", + "accept": false, + "reason": "README rewrite vs ckpt loading logic; unrelated." + }, + { + "left": "pull_request:3524", + "right": "pull_request:4748", + "accept": false, + "reason": "README rewrite vs setup.py release-publishing instructions; different documentation targets and intent." + }, + { + "left": "pull_request:3524", + "right": "pull_request:5267", + "accept": false, + "reason": "README rewrite vs docs/README guidance updates; both docs but not the same concrete change." + }, + { + "left": "pull_request:3524", + "right": "pull_request:5855", + "accept": false, + "reason": "README rewrite vs community README typo fix; same file type but distinct edits and scopes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 162, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3525", + "pull_request:3563", + "pull_request:3641", + "pull_request:3687", + "pull_request:3723", + "pull_request:3798", + "pull_request:3806", + "pull_request:3820", + "pull_request:3835", + "pull_request:386", + "pull_request:390", + "pull_request:391", + "pull_request:392", + "pull_request:3955", + "pull_request:396", + "pull_request:397", + "pull_request:3989", + "pull_request:404", + "pull_request:4082", + "pull_request:41", + "pull_request:4113", + "pull_request:415", + "pull_request:416", + "pull_request:417", + "pull_request:418", + "pull_request:4203", + "pull_request:422", + "pull_request:4239", + "pull_request:4273", + "pull_request:4275", + "pull_request:4276", + "pull_request:428", + "pull_request:4284", + "pull_request:4498", + "pull_request:4807", + "pull_request:4822", + "pull_request:5729", + "pull_request:5792", + "pull_request:5948", + "pull_request:6078", + "pull_request:6150", + "pull_request:6337", + "pull_request:6410", + "pull_request:6491", + "pull_request:6499", + "pull_request:6591", + "pull_request:662" + ], + "soft_pairs": [ + "pull_request:3525|pull_request:5729", + "pull_request:3563|pull_request:3989", + "pull_request:3641|pull_request:5948", + "pull_request:3687|pull_request:4807", + "pull_request:3723|pull_request:4498", + "pull_request:3798|pull_request:6078", + "pull_request:3806|pull_request:6150", + "pull_request:3820|pull_request:3955", + "pull_request:3820|pull_request:4284", + "pull_request:3835|pull_request:4273", + "pull_request:386|pull_request:396", + "pull_request:386|pull_request:416", + "pull_request:390|pull_request:416", + "pull_request:391|pull_request:392", + "pull_request:391|pull_request:404", + "pull_request:391|pull_request:417", + "pull_request:397|pull_request:422", + "pull_request:397|pull_request:428", + "pull_request:4082|pull_request:6410", + "pull_request:41|pull_request:662", + "pull_request:4113|pull_request:4822", + "pull_request:415|pull_request:422", + "pull_request:415|pull_request:428", + "pull_request:418|pull_request:422", + "pull_request:418|pull_request:428", + "pull_request:4203|pull_request:6491", + "pull_request:4239|pull_request:6337", + "pull_request:4239|pull_request:6591", + "pull_request:4273|pull_request:6499", + "pull_request:4275|pull_request:6410", + "pull_request:4276|pull_request:5792" + ], + "prepared_review_unit_hash": "f923be4b9582f58e86d3d7a682c1f596ab4291910c51dbb12ec811919f9b2bcb", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 97774, + "estimated_input_tokens": 24444, + "estimated_eval_tokens": 49144 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated docs, typo, example, test, and small bugfix PRs. The soft edges are driven by superficial similarity (same repo area, docs, or tiny edits) rather than the same underlying change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 3835 is the broadest and most central documentation/API sweep in the set, so it is the closest thing to a representative anchor, though the cluster is too heterogeneous to treat as a true duplicate group.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3525", + "right": "pull_request:5729", + "accept": false, + "reason": "README wording/usage examples vs a large training-script version check update; different change intents and code paths." + }, + { + "left": "pull_request:3563", + "right": "pull_request:3989", + "accept": false, + "reason": "A UNet block bugfix for vq pipeline vs SDXL test/offload changes; not the same underlying bug." + }, + { + "left": "pull_request:3641", + "right": "pull_request:5948", + "accept": false, + "reason": "Inpainting latent handling fix vs a from_pretrained deprecation warning tweak; unrelated functionality." + }, + { + "left": "pull_request:3687", + "right": "pull_request:4807", + "accept": false, + "reason": "Doc-builder test edit vs Chinese documentation additions; unrelated artifacts." + }, + { + "left": "pull_request:3723", + "right": "pull_request:4498", + "accept": false, + "reason": "Reference-image example updates vs a new SDXL reference pipeline feature; not the same change." + }, + { + "left": "pull_request:3798", + "right": "pull_request:6078", + "accept": false, + "reason": "Comment cleanup in training scripts vs Euler scheduler sigma logic; different issues." + }, + { + "left": "pull_request:3806", + "right": "pull_request:6150", + "accept": false, + "reason": "Checkpoint-loading README note vs IP-Adapter conversion code; unrelated." + }, + { + "left": "pull_request:3820", + "right": "pull_request:3955", + "accept": false, + "reason": "LDM3D typo/docs cleanup vs Kandinsky prior parameter/default changes; different pipelines and bugs." + }, + { + "left": "pull_request:3820", + "right": "pull_request:4284", + "accept": false, + "reason": "Both touch LDM3D, but one is a small typo/example cleanup and the other renames the output class/docs; not one concrete fix." + }, + { + "left": "pull_request:3835", + "right": "pull_request:4273", + "accept": false, + "reason": "Broad API docs refactor vs AutoPipeline tutorial/docs; same docs area but not the same change." + }, + { + "left": "pull_request:386", + "right": "pull_request:396", + "accept": false, + "reason": "Stable Diffusion pipeline docs vs DDIM docs; different models and pages." + }, + { + "left": "pull_request:386", + "right": "pull_request:416", + "accept": false, + "reason": "Stable Diffusion pipeline docs vs models API docs; unrelated documentation targets." + }, + { + "left": "pull_request:390", + "right": "pull_request:416", + "accept": false, + "reason": "MPS optimization docs vs models API docs; unrelated topics." + }, + { + "left": "pull_request:391", + "right": "pull_request:392", + "accept": false, + "reason": "PNDM docs vs contribution docs; no shared underlying bug or feature." + }, + { + "left": "pull_request:391", + "right": "pull_request:404", + "accept": false, + "reason": "PNDM docs vs fp16 optimization docs; different documentation subjects." + }, + { + "left": "pull_request:391", + "right": "pull_request:417", + "accept": false, + "reason": "PNDM docs vs inference pipeline tutorials; unrelated docs changes." + }, + { + "left": "pull_request:397", + "right": "pull_request:422", + "accept": false, + "reason": "Quicktour rewrite vs outputs API docs; different documentation areas." + }, + { + "left": "pull_request:397", + "right": "pull_request:428", + "accept": false, + "reason": "Quicktour overhaul vs Using Diffusers docs consolidation; both docs, but not the same concrete change." + }, + { + "left": "pull_request:4082", + "right": "pull_request:6410", + "accept": false, + "reason": "SDXL dependency pin vs RL module registration fix; unrelated code paths." + }, + { + "left": "pull_request:41", + "right": "pull_request:662", + "accept": false, + "reason": "Resnet comment cleanup vs textual inversion example/assets update; clearly different." + }, + { + "left": "pull_request:4113", + "right": "pull_request:4822", + "accept": false, + "reason": "SDXL inpaint dtype fix vs LoRA warning typo; unrelated bug fixes." + }, + { + "left": "pull_request:415", + "right": "pull_request:422", + "accept": false, + "reason": "Training docs overhaul vs outputs API docs; different subjects." + }, + { + "left": "pull_request:415", + "right": "pull_request:428", + "accept": false, + "reason": "Training docs vs Using Diffusers guides; both docs, but not the same change or page." + }, + { + "left": "pull_request:418", + "right": "pull_request:422", + "accept": false, + "reason": "DiffusionPipeline docs vs outputs API docs; separate foundational docs pages." + }, + { + "left": "pull_request:418", + "right": "pull_request:428", + "accept": false, + "reason": "DiffusionPipeline docs vs Using Diffusers docs; related only at a high level, not the same change." + }, + { + "left": "pull_request:4203", + "right": "pull_request:6491", + "accept": false, + "reason": "Dreambooth README typo fix vs a zero-diff typo PR; not a duplicate fix." + }, + { + "left": "pull_request:4239", + "right": "pull_request:6337", + "accept": false, + "reason": "Text-encoder loading fix vs TensorRT latent retrieval change; different code paths and problems." + }, + { + "left": "pull_request:4239", + "right": "pull_request:6591", + "accept": false, + "reason": "SDXL LoRA/text_encoder fix vs broad typo cleanup across examples and pipelines; unrelated." + }, + { + "left": "pull_request:4273", + "right": "pull_request:6499", + "accept": false, + "reason": "AutoPipeline tutorial/docs vs PEFT adapter API docs; separate documentation features." + }, + { + "left": "pull_request:4275", + "right": "pull_request:6410", + "accept": false, + "reason": "AutoPipeline docs correction vs RL register_modules refactor; unrelated." + }, + { + "left": "pull_request:4276", + "right": "pull_request:5792", + "accept": false, + "reason": "Pipeline import test vs import-path fixes in examples; test coverage change is not the same as the example import cleanup." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 163, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4301", + "pull_request:4320", + "pull_request:433", + "pull_request:4396", + "pull_request:4472", + "pull_request:4547", + "pull_request:4746", + "pull_request:4748", + "pull_request:4783", + "pull_request:4787", + "pull_request:4887", + "pull_request:4937", + "pull_request:51", + "pull_request:5122", + "pull_request:5164", + "pull_request:5267", + "pull_request:5312", + "pull_request:5369", + "pull_request:5393", + "pull_request:5478", + "pull_request:5552", + "pull_request:5602", + "pull_request:5622", + "pull_request:5624", + "pull_request:5682", + "pull_request:5697", + "pull_request:5698", + "pull_request:5699", + "pull_request:5707", + "pull_request:5729", + "pull_request:5916", + "pull_request:5982", + "pull_request:6053", + "pull_request:6065", + "pull_request:6237", + "pull_request:6262", + "pull_request:6305", + "pull_request:6410", + "pull_request:6432", + "pull_request:6437", + "pull_request:6775", + "pull_request:6886", + "pull_request:70", + "pull_request:7134", + "pull_request:7200", + "pull_request:7510", + "pull_request:7650", + "pull_request:788" + ], + "soft_pairs": [ + "pull_request:4301|pull_request:4887", + "pull_request:4320|pull_request:6410", + "pull_request:433|pull_request:70", + "pull_request:4396|pull_request:6262", + "pull_request:4472|pull_request:5729", + "pull_request:4547|pull_request:4746", + "pull_request:4547|pull_request:5682", + "pull_request:4748|pull_request:5729", + "pull_request:4783|pull_request:6237", + "pull_request:4787|pull_request:6437", + "pull_request:4887|pull_request:6886", + "pull_request:4937|pull_request:6305", + "pull_request:51|pull_request:788", + "pull_request:5122|pull_request:5164", + "pull_request:5267|pull_request:5729", + "pull_request:5312|pull_request:7134", + "pull_request:5369|pull_request:6775", + "pull_request:5393|pull_request:6432", + "pull_request:5478|pull_request:5982", + "pull_request:5552|pull_request:7650", + "pull_request:5602|pull_request:6065", + "pull_request:5622|pull_request:7510", + "pull_request:5624|pull_request:5916", + "pull_request:5697|pull_request:7200", + "pull_request:5698|pull_request:5699", + "pull_request:5707|pull_request:6053" + ], + "prepared_review_unit_hash": "dd73365133cd72f28b61b9038aabf5f1d224e37637aa6ecc3de39a02bde7c78c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 104051, + "estimated_input_tokens": 26013, + "estimated_eval_tokens": 52282 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of mostly unrelated PRs that were linked only by weak textual similarity. The soft edges mostly connect docs, training-script, scheduler, and model changes that do not fix the same underlying bug or implement the same change, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:4301", + "right": "pull_request:4887", + "accept": false, + "reason": "Kandinsky docs cleanup vs Kandinsky prior_emb2emb API removal; one is documentation, the other changes pipeline behavior." + }, + { + "left": "pull_request:4320", + "right": "pull_request:6410", + "accept": false, + "reason": "Typo/doc wording fix vs RL value-guided sampling refactor; unrelated code paths and goals." + }, + { + "left": "pull_request:433", + "right": "pull_request:70", + "accept": false, + "reason": "Black version bump and formatting cleanup vs an early attention/unet refactor; no shared bug or feature." + }, + { + "left": "pull_request:4396", + "right": "pull_request:6262", + "accept": false, + "reason": "Prompt-encoding fallback fix vs moving panorama out of stable_diffusion; unrelated pipeline changes." + }, + { + "left": "pull_request:4472", + "right": "pull_request:5729", + "accept": false, + "reason": "DreamBooth README edit vs broad setup/version updates in many scripts; not the same concrete change." + }, + { + "left": "pull_request:4547", + "right": "pull_request:4746", + "accept": false, + "reason": "Flax scheduler WIP vs SDXL inpainting work; different features and code paths." + }, + { + "left": "pull_request:4547", + "right": "pull_request:5682", + "accept": false, + "reason": "Flax scheduler WIP vs AnimateDiff checks; unrelated implementations." + }, + { + "left": "pull_request:4748", + "right": "pull_request:5729", + "accept": false, + "reason": "Packaging/release instruction cleanup vs example script minimum-version updates; only loosely related to setup/versioning." + }, + { + "left": "pull_request:4783", + "right": "pull_request:6237", + "accept": false, + "reason": "SDXL embedding-force docs vs batched seed documentation; different user-facing topics." + }, + { + "left": "pull_request:4787", + "right": "pull_request:6437", + "accept": false, + "reason": "LoRA weight-file selection logic vs CI test exclusion for PEFT env; same broad area, but not the same bug or mergeable fix." + }, + { + "left": "pull_request:4887", + "right": "pull_request:6886", + "accept": false, + "reason": "Kandinsky pipeline API change vs removing torch_dtype deprecation usage; unrelated changes." + }, + { + "left": "pull_request:4937", + "right": "pull_request:6305", + "accept": false, + "reason": "Adapter conditioning-factor addition vs DDPM beta rescaling; different scheduler/model functionality." + }, + { + "left": "pull_request:51", + "right": "pull_request:788", + "accept": false, + "reason": "install_requires pin fix vs PIL sampler warning fix; no common underlying issue." + }, + { + "left": "pull_request:5122", + "right": "pull_request:5164", + "accept": false, + "reason": "CI Python version bump vs adding FreeU mechanism; unrelated maintenance vs new model feature." + }, + { + "left": "pull_request:5267", + "right": "pull_request:5729", + "accept": false, + "reason": "Markdown code-fence cleanup in docs vs broad example/setup version changes; not the same artifact or bug." + }, + { + "left": "pull_request:5312", + "right": "pull_request:7134", + "accept": false, + "reason": "Broad model docs/type-hint cleanup vs specific UNet type-hint adjustments; same general theme, but not the same concrete change." + }, + { + "left": "pull_request:5369", + "right": "pull_request:6775", + "accept": false, + "reason": "Minor docs fixes in img2img/inpaint vs adding a missing DDPM parameter doc; different documentation edits." + }, + { + "left": "pull_request:5393", + "right": "pull_request:6432", + "accept": false, + "reason": "HFApi tutorial migration vs DreamBooth learning-rate tweak; unrelated example updates." + }, + { + "left": "pull_request:5478", + "right": "pull_request:5982", + "accept": false, + "reason": "Japanese docs rollout vs SDXL Turbo docs; different doc scopes and features." + }, + { + "left": "pull_request:5552", + "right": "pull_request:7650", + "accept": false, + "reason": "FreeU disable-method bugfix vs SDXL InstructPix2Pix arg addition; unrelated pipeline/model changes." + }, + { + "left": "pull_request:5602", + "right": "pull_request:6065", + "accept": false, + "reason": "Whitespace cleanup in docs vs SDXL Turbo docs; not the same change." + }, + { + "left": "pull_request:5622", + "right": "pull_request:7510", + "accept": false, + "reason": "Example training script/README and ignore-file cleanup vs VAE tiling OOM test fix; unrelated." + }, + { + "left": "pull_request:5624", + "right": "pull_request:5916", + "accept": false, + "reason": "Both touch DreamBooth, but one is a broad argument/README update and the other only renames the output_dir default; not the same underlying fix." + }, + { + "left": "pull_request:5697", + "right": "pull_request:7200", + "accept": false, + "reason": "PEFT LoRA+xformers bugfix vs Transformer2D validation/type behavior; different code paths." + }, + { + "left": "pull_request:5698", + "right": "pull_request:5699", + "accept": false, + "reason": "LCM img2img functional fix vs stable unCLIP styling cleanup; unrelated despite adjacency." + }, + { + "left": "pull_request:5707", + "right": "pull_request:6053", + "accept": false, + "reason": "LCM tests vs logging step_loss in text-to-image training; different purposes and files." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 164, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5727", + "pull_request:5729", + "pull_request:5833", + "pull_request:5834", + "pull_request:5899", + "pull_request:5937", + "pull_request:594", + "pull_request:6051", + "pull_request:6114", + "pull_request:6192", + "pull_request:6205", + "pull_request:6262", + "pull_request:6263", + "pull_request:6296", + "pull_request:6299", + "pull_request:6300", + "pull_request:6301", + "pull_request:6308", + "pull_request:6311", + "pull_request:6322", + "pull_request:6330", + "pull_request:6343", + "pull_request:6453", + "pull_request:6656", + "pull_request:6759", + "pull_request:6800", + "pull_request:6861", + "pull_request:7050", + "pull_request:7077", + "pull_request:7115", + "pull_request:7118", + "pull_request:7153", + "pull_request:7181", + "pull_request:7277", + "pull_request:7339", + "pull_request:7455", + "pull_request:7461", + "pull_request:748", + "pull_request:7627", + "pull_request:7650", + "pull_request:816", + "pull_request:825", + "pull_request:8290", + "pull_request:8378", + "pull_request:8412", + "pull_request:8652", + "pull_request:8722", + "pull_request:9247" + ], + "soft_pairs": [ + "pull_request:5727|pull_request:6051", + "pull_request:5729|pull_request:6861", + "pull_request:5833|pull_request:7461", + "pull_request:5834|pull_request:7461", + "pull_request:5899|pull_request:5937", + "pull_request:594|pull_request:825", + "pull_request:6114|pull_request:7627", + "pull_request:6192|pull_request:6800", + "pull_request:6205|pull_request:7077", + "pull_request:6205|pull_request:7181", + "pull_request:6262|pull_request:8290", + "pull_request:6263|pull_request:8290", + "pull_request:6296|pull_request:6343", + "pull_request:6296|pull_request:6453", + "pull_request:6299|pull_request:6322", + "pull_request:6300|pull_request:6322", + "pull_request:6301|pull_request:6322", + "pull_request:6308|pull_request:7339", + "pull_request:6311|pull_request:8652", + "pull_request:6330|pull_request:8412", + "pull_request:6656|pull_request:6800", + "pull_request:6759|pull_request:8378", + "pull_request:7050|pull_request:7277", + "pull_request:7115|pull_request:7455", + "pull_request:7118|pull_request:7277", + "pull_request:7153|pull_request:9247", + "pull_request:748|pull_request:816", + "pull_request:7627|pull_request:7650", + "pull_request:7627|pull_request:8722" + ], + "prepared_review_unit_hash": "5de4d9545da3b0072a5c79ea547f3cad42e6f7b127d0a0da4b7225cd41d85d01", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 104789, + "estimated_input_tokens": 26198, + "estimated_eval_tokens": 52652 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous mix of unrelated PRs (docs, CI, examples, and distinct model/pipeline fixes). The only true duplicate-looking subcluster is the repeated text-to-image training-script scheduler tweak around 6299/6300/6301/6322; everything else should stay separate.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR for the whole cluster; the only genuine duplicate subcluster is the repeated text-to-image training fix, where PR 6322 is the clearest representative.", + "best_issue_reason": null, + "best_pr_reason": "PR 6322 is the best representative of the only real duplicate subcluster because it matches the same training-script code path and appears to be the most consolidated version of that change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5727", + "right": "pull_request:6051", + "accept": false, + "reason": "Both are LCM-related, but one adds example scripts while the other adds Flax scheduler support; different changes and code paths." + }, + { + "left": "pull_request:5729", + "right": "pull_request:6861", + "accept": false, + "reason": "Both touch training/pipeline setup, but one is a version-check update across examples and the other adds cfg_end support in SDXL pipelines." + }, + { + "left": "pull_request:5833", + "right": "pull_request:7461", + "accept": false, + "reason": "Docs for PixArt low-VRAM usage vs a testing-utils helper fix; unrelated artifacts." + }, + { + "left": "pull_request:5834", + "right": "pull_request:7461", + "accept": false, + "reason": "Single-file loading import path fix vs textual inversion boolean helper; not the same bug or change." + }, + { + "left": "pull_request:5899", + "right": "pull_request:5937", + "accept": false, + "reason": "One adds Kandinsky3 code, the other only documents it; docs follow-up, not a duplicate code change." + }, + { + "left": "pull_request:594", + "right": "pull_request:825", + "accept": false, + "reason": "Both are Flax maintenance, but one fixes UNet/DDIM embedding math and the other removes Flax safety-checker dummy handling." + }, + { + "left": "pull_request:6114", + "right": "pull_request:7627", + "accept": false, + "reason": "Both involve img2img/inpaint, but they target different pipelines and distinct bugs." + }, + { + "left": "pull_request:6192", + "right": "pull_request:6800", + "accept": false, + "reason": "IP-Adapter unload support vs a torchvision availability helper; unrelated changes." + }, + { + "left": "pull_request:6205", + "right": "pull_request:7077", + "accept": false, + "reason": "Docs URL typo vs attention tensor reshaping bug; unrelated." + }, + { + "left": "pull_request:6205", + "right": "pull_request:7181", + "accept": false, + "reason": "Both are small fixes, but one is a blog-link correction and the other is a broad typo sweep across multiple files." + }, + { + "left": "pull_request:6262", + "right": "pull_request:8290", + "accept": false, + "reason": "Panorama refactor vs HunyuanDIT refactor; different modules and changes." + }, + { + "left": "pull_request:6263", + "right": "pull_request:8290", + "accept": false, + "reason": "LDM3D refactor vs HunyuanDIT refactor; different code paths and goals." + }, + { + "left": "pull_request:6296", + "right": "pull_request:6343", + "accept": false, + "reason": "Initial aMUSEd docs/implementation PR vs later docs expansion for extra pipelines; follow-up documentation, not duplicate change." + }, + { + "left": "pull_request:6296", + "right": "pull_request:6453", + "accept": false, + "reason": "Initial aMUSEd docs/implementation PR vs paper-link doc tweak; not the same underlying change." + }, + { + "left": "pull_request:6299", + "right": "pull_request:6322", + "accept": true, + "reason": "Same text-to-image training-script scheduler adjustment in the same files; plausibly one merged PR." + }, + { + "left": "pull_request:6300", + "right": "pull_request:6322", + "accept": true, + "reason": "Same training-script change and file set as the other text-to-image validation scheduler fix." + }, + { + "left": "pull_request:6301", + "right": "pull_request:6322", + "accept": true, + "reason": "Same code-path and diff theme as the other text-to-image training updates; looks duplicate." + }, + { + "left": "pull_request:6308", + "right": "pull_request:7339", + "accept": false, + "reason": "LoRA test fixture fix vs release-workflow dependency changes; unrelated." + }, + { + "left": "pull_request:6311", + "right": "pull_request:8652", + "accept": false, + "reason": "A docs/tutorial addition vs a major model/pipeline addition; only share generic documentation tree edits." + }, + { + "left": "pull_request:6330", + "right": "pull_request:8412", + "accept": false, + "reason": "SVD docs rendering fix vs single-file usage docs rewrite; different documentation topics." + }, + { + "left": "pull_request:6656", + "right": "pull_request:6800", + "accept": false, + "reason": "Textual inversion unload API vs torchvision availability helper; unrelated." + }, + { + "left": "pull_request:6759", + "right": "pull_request:8378", + "accept": false, + "reason": "InstantID example fix vs asymmetric autoencoder forward fix; different areas and bugs." + }, + { + "left": "pull_request:7050", + "right": "pull_request:7277", + "accept": false, + "reason": "Both mention typos, but one is text_to_image examples and the other is a broader tests/docs cleanup with a deprecated-parameter update." + }, + { + "left": "pull_request:7115", + "right": "pull_request:7455", + "accept": false, + "reason": "Dockerfile packaging change vs distutils removal across setup/testing utilities; unrelated." + }, + { + "left": "pull_request:7118", + "right": "pull_request:7277", + "accept": false, + "reason": "Docs typo cleanup vs tests/deprecated-parameter update; not the same change." + }, + { + "left": "pull_request:7153", + "right": "pull_request:9247", + "accept": false, + "reason": "Documentation workflow path filters vs pipeline-loading variant detection fix; unrelated." + }, + { + "left": "pull_request:748", + "right": "pull_request:816", + "accept": false, + "reason": "Push-to-hub API cleanup vs mixed-precision/deepspeed training loss fix; different problems." + }, + { + "left": "pull_request:7627", + "right": "pull_request:7650", + "accept": false, + "reason": "Both are SDXL-related, but one fixes img2img latents handling and the other adds an is_cosxl_edit argument for instruct-pix2pix." + }, + { + "left": "pull_request:825", + "right": "pull_request:8722", + "accept": false, + "reason": "Flax safety-checker cleanup vs a large embeddings refactor; different code paths." + }, + { + "left": "pull_request:8290", + "right": "pull_request:8722", + "accept": false, + "reason": "HunyuanDIT refactor vs embeddings refactor; both broad refactors but not the same underlying change." + }, + { + "left": "pull_request:8378", + "right": "pull_request:8722", + "accept": false, + "reason": "AsymmetricAutoencoderKL forward bugfix vs embeddings refactor; unrelated." + }, + { + "left": "pull_request:8412", + "right": "pull_request:8652", + "accept": false, + "reason": "Single-file docs vs Lumina model/pipeline addition; different artifacts and scope." + }, + { + "left": "pull_request:8652", + "right": "pull_request:8722", + "accept": false, + "reason": "Lumina-T2X feature addition vs embeddings refactor; not the same code-path problem." + }, + { + "left": "pull_request:8722", + "right": "pull_request:9247", + "accept": false, + "reason": "Large embeddings refactor vs a small model-variant discovery bugfix; unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: most pairs are clearly unrelated, and the only plausible duplicate subcluster is the repeated text-to-image training-script scheduler fix around 6299/6300/6301/6322. The conservative rejections look appropriate, and the single accepted subcluster is supported by the identical file set and diff theme." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 165, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10396", + "pull_request:10578", + "pull_request:12086", + "pull_request:12419", + "pull_request:12439", + "pull_request:12549", + "pull_request:2421", + "pull_request:3034", + "pull_request:4597", + "pull_request:4799", + "pull_request:5127", + "pull_request:5653", + "pull_request:5887", + "pull_request:6003", + "pull_request:6975", + "pull_request:6999", + "pull_request:7637", + "pull_request:7717", + "pull_request:7893", + "pull_request:7977", + "pull_request:8308", + "pull_request:8310", + "pull_request:840", + "pull_request:8403", + "pull_request:8479", + "pull_request:8481", + "pull_request:8660", + "pull_request:8719", + "pull_request:8761", + "pull_request:8978", + "pull_request:8984", + "pull_request:9", + "pull_request:9060", + "pull_request:909", + "pull_request:91", + "pull_request:9269", + "pull_request:9289", + "pull_request:9307", + "pull_request:9426", + "pull_request:9483", + "pull_request:9620", + "pull_request:9676", + "pull_request:9699", + "pull_request:9734", + "pull_request:9739", + "pull_request:9938", + "pull_request:9994" + ], + "soft_pairs": [ + "pull_request:7637|pull_request:7717", + "pull_request:7893|pull_request:9307", + "pull_request:7977|pull_request:9739", + "pull_request:8308|pull_request:9289", + "pull_request:8310|pull_request:9289", + "pull_request:840|pull_request:9", + "pull_request:8403|pull_request:8978", + "pull_request:8479|pull_request:9620", + "pull_request:8481|pull_request:9739", + "pull_request:8660|pull_request:9426", + "pull_request:8719|pull_request:9676", + "pull_request:8761|pull_request:9269", + "pull_request:8978|pull_request:9734", + "pull_request:8984|pull_request:9620", + "pull_request:9060|pull_request:9938", + "pull_request:909|pull_request:91", + "pull_request:9483|pull_request:9699", + "pull_request:9620|pull_request:9994", + "pull_request:4597|pull_request:5887", + "pull_request:10396|pull_request:10578", + "pull_request:12439|pull_request:12549", + "pull_request:12086|pull_request:12419", + "pull_request:6003|pull_request:6999", + "pull_request:4799|pull_request:6975", + "pull_request:5127|pull_request:5653", + "pull_request:2421|pull_request:3034" + ], + "prepared_review_unit_hash": "91ce99af5b432fff383e7ff40e70e7417aba370a68bc4ec3c77a1814b1f8858f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 101541, + "estimated_input_tokens": 25386, + "estimated_eval_tokens": 51028 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: LoRA, ControlNet, Flux, Hunyuan, SD3, CI, docs, and unrelated maintenance changes. The soft edges are mostly superficial overlaps in filenames or broad subsystem names, not true duplicate bugs or mergeable duplicate PRs.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out; the items are unrelated changes spanning many subsystems, so there is no meaningful duplicate anchor.", + "best_issue_reason": null, + "best_pr_reason": "No globally best PR for deduplication: the set is too mixed, so any single PR would be arbitrary rather than representative of the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7637", + "right": "pull_request:7717", + "accept": false, + "reason": "Different maintenance areas: Ruff config deprecation cleanup vs docstring style fixes." + }, + { + "left": "pull_request:7893", + "right": "pull_request:9307", + "accept": false, + "reason": "Unrelated changes: Python 3.10 Dockerfile upgrade vs rotary embedding CPU/device refactor." + }, + { + "left": "pull_request:7977", + "right": "pull_request:9739", + "accept": false, + "reason": "One fixes a typo in docs; the other adds an SD3 single-file loading example. Not the same change." + }, + { + "left": "pull_request:8308", + "right": "pull_request:9289", + "accept": false, + "reason": "Different models/features: Hunyuan refactor vs AnimateDiff model work." + }, + { + "left": "pull_request:8310", + "right": "pull_request:9289", + "accept": false, + "reason": "HunyuanDiT refactor and AnimateDiff model work are unrelated code paths." + }, + { + "left": "pull_request:840", + "right": "pull_request:9", + "accept": false, + "reason": "One-step U-Net community example vs Grad TTS model/pipeline addition are unrelated." + }, + { + "left": "pull_request:8403", + "right": "pull_request:8978", + "accept": false, + "reason": "Updating a Tailscale action is unrelated to DreamBooth/imagic content changes." + }, + { + "left": "pull_request:8479", + "right": "pull_request:9620", + "accept": false, + "reason": "Both involve ControlNet, but one adds Hunyuan ControlNet support and the other adds interrupt handling across ControlNet pipelines." + }, + { + "left": "pull_request:8481", + "right": "pull_request:9739", + "accept": false, + "reason": "Image input error-message tweak vs SD3 docs/typo fixes are unrelated." + }, + { + "left": "pull_request:8660", + "right": "pull_request:9426", + "accept": false, + "reason": "SD3 ControlNet validation change vs CI artifact uploader upgrade are different concerns." + }, + { + "left": "pull_request:8719", + "right": "pull_request:9676", + "accept": false, + "reason": "VQModelOutput deprecation/docs cleanup vs benchmark import fix; no shared bug or change." + }, + { + "left": "pull_request:8761", + "right": "pull_request:9269", + "accept": false, + "reason": "Shared embeddings.py file, but these patch different functions for different models (Hunyuan embedding refactor vs CogVideoX 3D rope refactor)." + }, + { + "left": "pull_request:8978", + "right": "pull_request:9734", + "accept": false, + "reason": "DreamBooth/imagic update and SD3 doc/test update are unrelated documentation/example changes." + }, + { + "left": "pull_request:8984", + "right": "pull_request:9620", + "accept": false, + "reason": "ONNX conversion/pipeline support for SD3 is unrelated to ControlNet interrupt handling." + }, + { + "left": "pull_request:9060", + "right": "pull_request:9938", + "accept": false, + "reason": "Removing a debug print in Flux is unrelated to SDXL reference pipeline updates." + }, + { + "left": "pull_request:909", + "right": "pull_request:91", + "accept": false, + "reason": "Different scopes and eras: test model ID cleanup vs a broad scheduler API finish." + }, + { + "left": "pull_request:9483", + "right": "pull_request:9699", + "accept": false, + "reason": "Nightly test numeric expectation updates vs adding a new SSH runner are unrelated CI changes." + }, + { + "left": "pull_request:9620", + "right": "pull_request:9994", + "accept": false, + "reason": "ControlNet interrupt support and Omnigen model/pipeline support are unrelated features." + }, + { + "left": "pull_request:4597", + "right": "pull_request:5887", + "accept": false, + "reason": "Both touch ControlNet, but one adds MultiControlNet SDXL support and the other adds IP-Adapter support for inpaint; distinct PRs." + }, + { + "left": "pull_request:10396", + "right": "pull_request:10578", + "accept": false, + "reason": "Both are LoRA-related, but one handles missing base layers in LoRA expansion while the other supports 4-bit quantized Flux LoRA loading." + }, + { + "left": "pull_request:12439", + "right": "pull_request:12549", + "accept": false, + "reason": "Both concern attention backends, but one adds kernels-based Sage/FA2 support while the other adds AITER backend support." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12419", + "accept": false, + "reason": "Both touch Flux modular pipelines, but one adds Flux I2I support and the other adds Mellon-specific kwargs typing." + }, + { + "left": "pull_request:6003", + "right": "pull_request:6999", + "accept": false, + "reason": "Both are Stable Video Diffusion, but one adds inpaint support and the other fixes tensor image input handling." + }, + { + "left": "pull_request:4799", + "right": "pull_request:6975", + "accept": false, + "reason": "SDXL ControlNet guess_mode bugfix vs ControlNet inpaint single-file conversion fix are separate issues." + }, + { + "left": "pull_request:5127", + "right": "pull_request:5653", + "accept": false, + "reason": "Multi-adapter SDXL support and PEFT model unwrapping are unrelated changes." + }, + { + "left": "pull_request:2421", + "right": "pull_request:3034", + "accept": false, + "reason": "Deprecation call cleanup in cross_attention is unrelated to fixing an accelerate parameter name." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 166, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10008", + "pull_request:10136", + "pull_request:10435", + "pull_request:10482", + "pull_request:10567", + "pull_request:10578", + "pull_request:10719", + "pull_request:10837", + "pull_request:11130", + "pull_request:11814", + "pull_request:11985", + "pull_request:12074", + "pull_request:12416", + "pull_request:12445", + "pull_request:12519", + "pull_request:12530", + "pull_request:12679", + "pull_request:13029", + "pull_request:13100", + "pull_request:1513", + "pull_request:1554", + "pull_request:1579", + "pull_request:1942", + "pull_request:2002", + "pull_request:2129", + "pull_request:2200", + "pull_request:2393", + "pull_request:2584", + "pull_request:2721", + "pull_request:2928", + "pull_request:3042", + "pull_request:3128", + "pull_request:3590", + "pull_request:3746", + "pull_request:4062", + "pull_request:4214", + "pull_request:4659", + "pull_request:4661", + "pull_request:4715", + "pull_request:5117", + "pull_request:5833", + "pull_request:5834", + "pull_request:6009", + "pull_request:6133", + "pull_request:6447", + "pull_request:6650", + "pull_request:689", + "pull_request:9918" + ], + "soft_pairs": [ + "pull_request:12074|pull_request:12679", + "pull_request:4062|pull_request:6447", + "pull_request:10578|pull_request:11130", + "pull_request:4715|pull_request:5833", + "pull_request:4715|pull_request:5834", + "pull_request:10482|pull_request:10567", + "pull_request:2129|pull_request:2721", + "pull_request:2200|pull_request:2393", + "pull_request:12519|pull_request:12530", + "pull_request:1579|pull_request:2002", + "pull_request:10136|pull_request:10837", + "pull_request:1942|pull_request:689", + "pull_request:2928|pull_request:5117", + "pull_request:12416|pull_request:12445", + "pull_request:6133|pull_request:6650", + "pull_request:10008|pull_request:10435", + "pull_request:10719|pull_request:9918", + "pull_request:11814|pull_request:11985", + "pull_request:13029|pull_request:13100", + "pull_request:1513|pull_request:3042", + "pull_request:1554|pull_request:3128", + "pull_request:2584|pull_request:3590", + "pull_request:3746|pull_request:4659", + "pull_request:3746|pull_request:4661", + "pull_request:4214|pull_request:6009" + ], + "prepared_review_unit_hash": "a5abc4e5a115cb3bc398c447a78bbf91538a9f4499995bbf4f7205f60f31aa50", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 106313, + "estimated_input_tokens": 26579, + "estimated_eval_tokens": 53414 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: docs tweaks, example additions, model/pipeline features, and unrelated bug fixes. None of the suggested soft pairs look like the same underlying change or bug.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR fits well; the cluster spans unrelated features, docs, and fixes rather than one duplicated change.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a strong representative for duplicate triage because there is no shared concrete code-path or bug across the items.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12074", + "right": "pull_request:12679", + "accept": false, + "reason": "Both involve LoRA support, but one is Wan 2.2 second-transformer loading and the other is ChronoEdit multi-LoRA docs/loaders; different models and changes." + }, + { + "left": "pull_request:4062", + "right": "pull_request:6447", + "accept": false, + "reason": "Shap-E mesh output support is unrelated to adding K-Diffusion pipelines." + }, + { + "left": "pull_request:10578", + "right": "pull_request:11130", + "accept": false, + "reason": "4bit Flux LoRA loading and pipeline-level quantization config are different feature sets, not the same fix." + }, + { + "left": "pull_request:4715", + "right": "pull_request:5833", + "accept": false, + "reason": "File renames for community tests vs fixing imports in single-file loading are unrelated changes." + }, + { + "left": "pull_request:4715", + "right": "pull_request:5834", + "accept": false, + "reason": "File renames for community tests vs fixing imports in single-file loading are unrelated changes." + }, + { + "left": "pull_request:10482", + "right": "pull_request:10567", + "accept": false, + "reason": "NaN fix in HunyuanVideo attention masking is unrelated to the true-cfg feature work." + }, + { + "left": "pull_request:2129", + "right": "pull_request:2721", + "accept": false, + "reason": "Allowing LoRA from pipeline and adding ModelEditing pipeline are different capabilities." + }, + { + "left": "pull_request:2200", + "right": "pull_request:2393", + "accept": false, + "reason": "Both are panorama-related, but one is a community multi-diffusion pipeline and the other is an official stable-diffusion panorama pipeline; not the same concrete change." + }, + { + "left": "pull_request:12519", + "right": "pull_request:12530", + "accept": false, + "reason": "Toctree sorting/quality fix and xfail-ing a flaky test solve different CI issues." + }, + { + "left": "pull_request:1579", + "right": "pull_request:2002", + "accept": false, + "reason": "Intel-optimized examples are unrelated to LoRA fine-tuning support." + }, + { + "left": "pull_request:10136", + "right": "pull_request:10837", + "accept": false, + "reason": "Base Hunyuan Video support and SkyReels Hunyuan T2V/I2V are distinct model integrations." + }, + { + "left": "pull_request:1942", + "right": "pull_request:689", + "accept": false, + "reason": "Fixing an upcast_attention flag in a conversion script is unrelated to the attention performance change." + }, + { + "left": "pull_request:2928", + "right": "pull_request:5117", + "accept": false, + "reason": "Custom local dataset loading for ControlNet training and training resume log handling are unrelated." + }, + { + "left": "pull_request:12416", + "right": "pull_request:12445", + "accept": false, + "reason": "QwenImage Edit Plus modular support and Flux modular alignment are different model families and codepaths." + }, + { + "left": "pull_request:6133", + "right": "pull_request:6650", + "accept": false, + "reason": "UFOGen pipeline/scheduler implementation and a community example for the scheduler are not the same change." + }, + { + "left": "pull_request:10008", + "right": "pull_request:10435", + "accept": false, + "reason": "Table formatting docs for CogVideoX and doc variable-name fixes are unrelated documentation edits." + }, + { + "left": "pull_request:10719", + "right": "pull_request:9918", + "accept": false, + "reason": "Adding provider_options to pipeline loading and adding a server example are unrelated." + }, + { + "left": "pull_request:11814", + "right": "pull_request:11985", + "accept": false, + "reason": "Passing joint_attention_kwargs in gradient checkpointing and adding Flux Kontext are different features." + }, + { + "left": "pull_request:13029", + "right": "pull_request:13100", + "accept": false, + "reason": "Both are modular docs, but one is a quickstart guide and the other is a broader ModularPipeline doc update; not the same change." + }, + { + "left": "pull_request:1513", + "right": "pull_request:3042", + "accept": false, + "reason": "Dreambooth notebook config guidance and accelerate minimum-version updates are different docs/dependency fixes." + }, + { + "left": "pull_request:1554", + "right": "pull_request:3128", + "accept": false, + "reason": "Docs corrections for UNet text shapes are unrelated to the img2img timestep bug fix." + }, + { + "left": "pull_request:2584", + "right": "pull_request:3590", + "accept": false, + "reason": "Community ControlNet img2img and Stable Diffusion Mixture Canvas are separate community pipelines." + }, + { + "left": "pull_request:3746", + "right": "pull_request:4659", + "accept": false, + "reason": "Path-join fix in from_ckpt is unrelated to the SDXL long weighted prompt pipeline/docs changes." + }, + { + "left": "pull_request:3746", + "right": "pull_request:4661", + "accept": false, + "reason": "Path-join fix in from_ckpt is unrelated to the SDXL long weighted prompt pipeline/docs changes." + }, + { + "left": "pull_request:4214", + "right": "pull_request:6009", + "accept": false, + "reason": "bf16 error fix in Custom Diffusion is unrelated to the TensorRT super-args error fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 167, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10058", + "pull_request:10114", + "pull_request:10122", + "pull_request:10126", + "pull_request:10544", + "pull_request:10941", + "pull_request:11583", + "pull_request:11950", + "pull_request:12207", + "pull_request:12541", + "pull_request:13100", + "pull_request:13120", + "pull_request:1331", + "pull_request:13352", + "pull_request:2111", + "pull_request:2345", + "pull_request:2467", + "pull_request:2693", + "pull_request:3278", + "pull_request:3313", + "pull_request:3607", + "pull_request:3691", + "pull_request:4272", + "pull_request:4301", + "pull_request:4412", + "pull_request:4695", + "pull_request:4715", + "pull_request:4887", + "pull_request:4987", + "pull_request:5033", + "pull_request:5299", + "pull_request:5979", + "pull_request:6085", + "pull_request:6197", + "pull_request:6261", + "pull_request:6447", + "pull_request:6650", + "pull_request:6658", + "pull_request:6975", + "pull_request:7371", + "pull_request:8290", + "pull_request:8452", + "pull_request:8510", + "pull_request:8691", + "pull_request:8816", + "pull_request:8820", + "pull_request:9601" + ], + "soft_pairs": [ + "pull_request:4412|pull_request:6085", + "pull_request:4887|pull_request:4987", + "pull_request:5299|pull_request:6658", + "pull_request:6261|pull_request:8290", + "pull_request:6975|pull_request:8816", + "pull_request:6975|pull_request:8820", + "pull_request:3313|pull_request:5033", + "pull_request:11950|pull_request:12207", + "pull_request:4695|pull_request:6447", + "pull_request:5979|pull_request:6650", + "pull_request:6197|pull_request:7371", + "pull_request:10058|pull_request:11583", + "pull_request:10114|pull_request:8452", + "pull_request:10114|pull_request:8510", + "pull_request:10122|pull_request:10941", + "pull_request:10126|pull_request:8691", + "pull_request:10544|pull_request:9601", + "pull_request:12541|pull_request:13352", + "pull_request:13100|pull_request:13120", + "pull_request:1331|pull_request:2467", + "pull_request:2111|pull_request:3607", + "pull_request:2345|pull_request:3691", + "pull_request:2467|pull_request:4272", + "pull_request:2693|pull_request:4301", + "pull_request:3278|pull_request:4715" + ], + "prepared_review_unit_hash": "629bb44bdb346522d2f2fa78877f410d82168d28f20620217fab3172dbdfbdf9", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 101102, + "estimated_input_tokens": 25276, + "estimated_eval_tokens": 50808 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is not a duplicate cluster; it\u2019s a grab bag of mostly unrelated Diffusers PRs spanning docs, small bug fixes, loader/quantization work, and pipeline refactors. The soft pairs are mostly same-subsystem but different concrete changes, so they should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:4412", + "right": "pull_request:6085", + "accept": false, + "reason": "Both touch test/common utility code, but one is a float16 inference test fix and the other is a scheduler add_noise bug fix. Different code paths and different failures." + }, + { + "left": "pull_request:4887", + "right": "pull_request:4987", + "accept": false, + "reason": "These are unrelated model API cleanup changes: one removes a latent input from Kandinsky prior_emb2emb, the other removes a duplicate GLIGEN export." + }, + { + "left": "pull_request:5299", + "right": "pull_request:6658", + "accept": false, + "reason": "Both add utility methods, but one is about typehints/docs in models and the other introduces a memory-footprint helper for models/pipelines. Not the same underlying change." + }, + { + "left": "pull_request:6261", + "right": "pull_request:8290", + "accept": false, + "reason": "Both affect pipeline import/docs organization, but one refactors Attend-and-Excite out of stable_diffusion while the other is a broad HunyuanDiT/docs workflow refactor. No shared bug or change." + }, + { + "left": "pull_request:6975", + "right": "pull_request:8816", + "accept": false, + "reason": "One fixes single-file controlnet inpaint loading; the other adds PAG support for SD1.5 ControlNet. Different features and code paths." + }, + { + "left": "pull_request:6975", + "right": "pull_request:8820", + "accept": false, + "reason": "Single-file controlnet inpaint loading fix versus PAG support docs/code for SD1.5 ControlNet. Same broad area, different concrete work." + }, + { + "left": "pull_request:3313", + "right": "pull_request:5033", + "accept": false, + "reason": "Both involve inpainting, but one is a torch.compile compatibility sweep and the other fixes regular text2image UNet handling in inpaint. Not duplicative." + }, + { + "left": "pull_request:11950", + "right": "pull_request:12207", + "accept": false, + "reason": "Both are quantization-related and touch loader/import plumbing, but one introduces SVDQuant support and the other incorporates Nunchaku. Different backends and functionality." + }, + { + "left": "pull_request:4695", + "right": "pull_request:6447", + "accept": false, + "reason": "Both add pipeline documentation and exports, but they introduce entirely different pipelines: TextToVideoZero SDXL vs StableDiffusionXL K-Diffusion." + }, + { + "left": "pull_request:5979", + "right": "pull_request:6650", + "accept": false, + "reason": "Both add new pipeline/scheduler code, but DDGAN and UFOGen are distinct models and implementations. Not the same bug or change." + }, + { + "left": "pull_request:6197", + "right": "pull_request:7371", + "accept": false, + "reason": "DeepCache support for Stable Diffusion versus DoRA LoRA support. Different features, different subsystems." + }, + { + "left": "pull_request:10058", + "right": "pull_request:11583", + "accept": false, + "reason": "Both are docs-only edits, but one fixes a training dataset link while the other removes a fast diffusion tutorial. No shared underlying issue." + }, + { + "left": "pull_request:10114", + "right": "pull_request:8452", + "accept": false, + "reason": "A formatting cleanup in stable audio versus a single-file loading error fix. Completely different problem statements." + }, + { + "left": "pull_request:10114", + "right": "pull_request:8510", + "accept": false, + "reason": "Formatting cleanup in a pipeline versus learning-rate scheduler math fixes in dreambooth LoRA training. Not the same code path." + }, + { + "left": "pull_request:10122", + "right": "pull_request:10941", + "accept": false, + "reason": "Flux attention-mask plumbing in the attention processor is unrelated to flux pipeline copy/refactor cleanup." + }, + { + "left": "pull_request:10126", + "right": "pull_request:8691", + "accept": false, + "reason": "One deprecates save_attn_procs and updates tests; the other fixes vanilla fine-tuned LoRA loading behavior. Different loader paths and goals." + }, + { + "left": "pull_request:10544", + "right": "pull_request:9601", + "accept": false, + "reason": "Docs for ParaAttention optimization versus an SD3 training flag for VAE dtype. No duplicate bug or change." + }, + { + "left": "pull_request:12541", + "right": "pull_request:13352", + "accept": false, + "reason": "Wan LoRA conversion support and a CI workflow checkout step are unrelated changes." + }, + { + "left": "pull_request:13100", + "right": "pull_request:13120", + "accept": false, + "reason": "ModularPipeline docs update versus a test for AutoModel load-id tagging. Related project area, but not the same issue or change." + }, + { + "left": "pull_request:1331", + "right": "pull_request:2467", + "accept": false, + "reason": "Both mention output-type/inpainting-related code, but one is an inpainting scheduler/path fix and the other adds latent output type across many pipelines. Different feature scope." + }, + { + "left": "pull_request:2111", + "right": "pull_request:3607", + "accept": false, + "reason": "Both affect attention-related code, but one fuses attention masking in cross_attention while the other changes temb attention and adds docker/testing updates. Not the same concrete fix." + }, + { + "left": "pull_request:2345", + "right": "pull_request:3691", + "accept": false, + "reason": "Both are xFormers/PT2-era loader changes, but one is a warning about xformers 0.0.16 training issues and the other chooses LoRAAttnProcessor2_0 for text encoder LoRA. Different behavior." + }, + { + "left": "pull_request:2467", + "right": "pull_request:4272", + "accept": false, + "reason": "Latent output-type plumbing across pipelines versus missing optional imports in pipeline folders. Same repository area, not the same bug or change." + }, + { + "left": "pull_request:2693", + "right": "pull_request:4301", + "accept": false, + "reason": "Both are docs fixes, but one is an image link fix in inpaint docs and the other updates Kandinsky docs and pipeline links. No shared underlying defect." + }, + { + "left": "pull_request:3278", + "right": "pull_request:4715", + "accept": false, + "reason": "One temporarily disables spectrogram diffusion tests and dependency entries; the other renames community example test files to run scripts. Different maintenance tasks." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 168, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10148", + "pull_request:10230", + "pull_request:10326", + "pull_request:10539", + "pull_request:10547", + "pull_request:10850", + "pull_request:10966", + "pull_request:11269", + "pull_request:11479", + "pull_request:11602", + "pull_request:11916", + "pull_request:12301", + "pull_request:12439", + "pull_request:12534", + "pull_request:12919", + "pull_request:2334", + "pull_request:2397", + "pull_request:3673", + "pull_request:3850", + "pull_request:4099", + "pull_request:4621", + "pull_request:4947", + "pull_request:5186", + "pull_request:5526", + "pull_request:5853", + "pull_request:6007", + "pull_request:6195", + "pull_request:6237", + "pull_request:6864", + "pull_request:6961", + "pull_request:7044", + "pull_request:7134", + "pull_request:7202", + "pull_request:7206", + "pull_request:7295", + "pull_request:7328", + "pull_request:7410", + "pull_request:8404", + "pull_request:8522", + "pull_request:8650", + "pull_request:8684", + "pull_request:8698", + "pull_request:8755", + "pull_request:8932", + "pull_request:9247", + "pull_request:9257", + "pull_request:9524", + "pull_request:9896" + ], + "soft_pairs": [ + "pull_request:3673|pull_request:4621", + "pull_request:3850|pull_request:5526", + "pull_request:4099|pull_request:6195", + "pull_request:4947|pull_request:5853", + "pull_request:5186|pull_request:7328", + "pull_request:6007|pull_request:6237", + "pull_request:6007|pull_request:7134", + "pull_request:6864|pull_request:8684", + "pull_request:6961|pull_request:7295", + "pull_request:7044|pull_request:7134", + "pull_request:7202|pull_request:8698", + "pull_request:7206|pull_request:8650", + "pull_request:7410|pull_request:9247", + "pull_request:8404|pull_request:8932", + "pull_request:8522|pull_request:8932", + "pull_request:9257|pull_request:9896", + "pull_request:11602|pull_request:12301", + "pull_request:2334|pull_request:2397", + "pull_request:11916|pull_request:12439", + "pull_request:12534|pull_request:12919", + "pull_request:10148|pull_request:10850", + "pull_request:10230|pull_request:11269", + "pull_request:10326|pull_request:8755", + "pull_request:10539|pull_request:9257", + "pull_request:10547|pull_request:9524", + "pull_request:10966|pull_request:11479" + ], + "prepared_review_unit_hash": "594bc55285b00f03361b8c5563511461a4e33aff3dbdbddb793fc3f148056a64", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 102578, + "estimated_input_tokens": 25645, + "estimated_eval_tokens": 51546 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: the soft links mostly connect unrelated PRs that happen to share a subsystem, doc area, or test target. None look like true duplicates or a single mergeable code change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:3673", + "right": "pull_request:4621", + "accept": false, + "reason": "Different fixes: DreamBooth multi-GPU logic vs T2I-Adapter downscale-factor validation." + }, + { + "left": "pull_request:3850", + "right": "pull_request:5526", + "accept": false, + "reason": "Unrelated test cleanup vs scheduler/inpaint timestep behavior fix." + }, + { + "left": "pull_request:4099", + "right": "pull_request:6195", + "accept": false, + "reason": "Both touch UNet/SDXL-related code, but one fixes prompt-embed handling and the other adds AnimateDiff SDXL support." + }, + { + "left": "pull_request:4947", + "right": "pull_request:5853", + "accept": false, + "reason": "Training-script tests vs doctest workflow infrastructure; no shared bug/change." + }, + { + "left": "pull_request:5186", + "right": "pull_request:7328", + "accept": false, + "reason": "FreeU proof-of-concept vs PyPI workflow dependency fix are unrelated." + }, + { + "left": "pull_request:6007", + "right": "pull_request:6237", + "accept": false, + "reason": "Docs wording change vs batched seed guidance; different documentation updates." + }, + { + "left": "pull_request:6007", + "right": "pull_request:7134", + "accept": false, + "reason": "Contribution docs update vs UNet type-hint cleanup; unrelated." + }, + { + "left": "pull_request:6864", + "right": "pull_request:8684", + "accept": false, + "reason": "PEFT inference docs vs Pillow import fix in a controlnet example; unrelated." + }, + { + "left": "pull_request:6961", + "right": "pull_request:7295", + "accept": false, + "reason": "Single-file loader test utilities vs UNet single-file loader copy-paste cleanup; different targets." + }, + { + "left": "pull_request:7044", + "right": "pull_request:7134", + "accept": false, + "reason": "Same file family, but one re-adds UNet refactor code and the other is type-hint cleanup." + }, + { + "left": "pull_request:7202", + "right": "pull_request:8698", + "accept": false, + "reason": "Requirements-file cleanup vs transformer output-class deprecation removal; unrelated." + }, + { + "left": "pull_request:7206", + "right": "pull_request:8650", + "accept": false, + "reason": "Training-script tag metadata vs SD3 DreamBooth test suite; different concerns." + }, + { + "left": "pull_request:7410", + "right": "pull_request:9247", + "accept": false, + "reason": "FreeInit behavior fix vs pipeline variant-identification bug; unrelated code paths." + }, + { + "left": "pull_request:8404", + "right": "pull_request:8932", + "accept": false, + "reason": "Latte model/pipeline addition vs temporal transformer test coverage; not the same change." + }, + { + "left": "pull_request:8522", + "right": "pull_request:8932", + "accept": false, + "reason": "Transformers version pinning vs temporal transformer tests; unrelated." + }, + { + "left": "pull_request:9257", + "right": "pull_request:9896", + "accept": false, + "reason": "Flux DreamBooth training-script bug fix vs reverting explicit device transfers in Flux pipelines; different issues." + }, + { + "left": "pull_request:11602", + "right": "pull_request:12301", + "accept": false, + "reason": "Sana Sprint img2img pipeline addition vs Qwen-Image ControlNet inpaint pipeline addition; different features." + }, + { + "left": "pull_request:2334", + "right": "pull_request:2397", + "accept": false, + "reason": "Both are Pix2Pix Zero-related, but one adds the pipeline and the other adds DDIM inversion support; not the same concrete change." + }, + { + "left": "pull_request:11916", + "right": "pull_request:12439", + "accept": false, + "reason": "Attention dispatcher refactor vs adding SageAttention/FA2 kernels support; different backend changes." + }, + { + "left": "pull_request:12534", + "right": "pull_request:12919", + "accept": false, + "reason": "Ascend NPU positional-embedding handling vs NPU attention backend implementation; different problems." + }, + { + "left": "pull_request:10148", + "right": "pull_request:10850", + "accept": false, + "reason": "OmniGen feature addition vs hook utility-method refactor; unrelated." + }, + { + "left": "pull_request:10230", + "right": "pull_request:11269", + "accept": false, + "reason": "Docs for LoRA loader mixins vs ONNX Runtime package detection updates; unrelated." + }, + { + "left": "pull_request:10326", + "right": "pull_request:8755", + "accept": false, + "reason": "CI workflow quoting fix vs SD3 DreamBooth LoRA training bug fix; unrelated." + }, + { + "left": "pull_request:10539", + "right": "pull_request:9257", + "accept": false, + "reason": "Flux true-CFG condition fix vs Flux DreamBooth script bug fix; both Flux but different code paths and failures." + }, + { + "left": "pull_request:10547", + "right": "pull_request:9524", + "accept": false, + "reason": "Offload-test parameterization vs CogVideoX VAE conv_cache refactor; different areas." + }, + { + "left": "pull_request:10966", + "right": "pull_request:11479", + "accept": false, + "reason": "CogView4 input validation fix vs Intel Gaudi docs rename; unrelated." + }, + { + "left": "pull_request:110?", + "right": "pull_request:110?", + "accept": false, + "reason": "No matching pair provided." + }, + { + "left": "pull_request:7206", + "right": "pull_request:7206", + "accept": false, + "reason": "No duplicate edge here; listed for completeness in the candidate set." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 169, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11269", + "pull_request:11449", + "pull_request:11495", + "pull_request:12092", + "pull_request:12586", + "pull_request:12911", + "pull_request:13022", + "pull_request:13211", + "pull_request:1324", + "pull_request:1579", + "pull_request:1623", + "pull_request:1850", + "pull_request:2119", + "pull_request:2470", + "pull_request:2983", + "pull_request:3077", + "pull_request:3471", + "pull_request:3675", + "pull_request:3723", + "pull_request:3957", + "pull_request:4112", + "pull_request:4248", + "pull_request:4396", + "pull_request:4595", + "pull_request:51", + "pull_request:5164", + "pull_request:538", + "pull_request:6007", + "pull_request:6081", + "pull_request:6261", + "pull_request:6265", + "pull_request:6311", + "pull_request:6864", + "pull_request:689", + "pull_request:7050", + "pull_request:7652", + "pull_request:7712", + "pull_request:7778", + "pull_request:8404", + "pull_request:9065", + "pull_request:9074", + "pull_request:9207", + "pull_request:9400", + "pull_request:9489", + "pull_request:9579", + "pull_request:9687", + "pull_request:9988" + ], + "soft_pairs": [ + "pull_request:11269|pull_request:9687", + "pull_request:11449|pull_request:11495", + "pull_request:12092|pull_request:12586", + "pull_request:12092|pull_request:12911", + "pull_request:13022|pull_request:13211", + "pull_request:1623|pull_request:538", + "pull_request:1850|pull_request:2983", + "pull_request:2119|pull_request:538", + "pull_request:2470|pull_request:2983", + "pull_request:3077|pull_request:3675", + "pull_request:3471|pull_request:4112", + "pull_request:3723|pull_request:4112", + "pull_request:3957|pull_request:4595", + "pull_request:4248|pull_request:6081", + "pull_request:4396|pull_request:6261", + "pull_request:51|pull_request:689", + "pull_request:5164|pull_request:6007", + "pull_request:6265|pull_request:8404", + "pull_request:6311|pull_request:7778", + "pull_request:6864|pull_request:9065", + "pull_request:7050|pull_request:7712", + "pull_request:7050|pull_request:9207", + "pull_request:7652|pull_request:9400", + "pull_request:9489|pull_request:9579", + "pull_request:9074|pull_request:9988", + "pull_request:1324|pull_request:1579" + ], + "prepared_review_unit_hash": "dbcd6e8cd8f9f14d22ad7756dc0d038aa61b4be44b573e836c3aff04e7409fd6", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 96688, + "estimated_input_tokens": 24172, + "estimated_eval_tokens": 48600 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set: it mixes unrelated docs, example-script edits, build/test fixes, and Flux/ControlNet/FreeU feature work. All soft edges are rejected because they do not match the same concrete bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:11269", + "right": "pull_request:9687", + "accept": false, + "reason": "Different scopes: ONNXRuntime package detection vs Flux Redux model/pipeline conversion changes." + }, + { + "left": "pull_request:11449", + "right": "pull_request:11495", + "accept": false, + "reason": "Both touch image interpolation defaults, but for different training scripts and separate code paths." + }, + { + "left": "pull_request:12092", + "right": "pull_request:12586", + "accept": false, + "reason": "GGUF docs content vs a docs tree sorting/move change; not the same change." + }, + { + "left": "pull_request:12092", + "right": "pull_request:12911", + "accept": false, + "reason": "GGUF conversion docs are unrelated to discrete diffusion docs/examples." + }, + { + "left": "pull_request:13022", + "right": "pull_request:13211", + "accept": false, + "reason": "Dockerfile dependency/runtime fixes vs group-offloading runtime bug fix are unrelated." + }, + { + "left": "pull_request:1623", + "right": "pull_request:538", + "accept": false, + "reason": "Requirements bumps for example scripts vs core hub dependency version update; different artifacts and purpose." + }, + { + "left": "pull_request:1850", + "right": "pull_request:2983", + "accept": false, + "reason": "A model-layer test addition is unrelated to documentation about installing test dependencies." + }, + { + "left": "pull_request:2119", + "right": "pull_request:538", + "accept": false, + "reason": "Local-files-only Hub access guard vs dependency version bump; different problems." + }, + { + "left": "pull_request:2470", + "right": "pull_request:2983", + "accept": false, + "reason": "MPS-specific test skips vs contributing-doc test setup guidance are unrelated." + }, + { + "left": "pull_request:3077", + "right": "pull_request:3675", + "accept": false, + "reason": "Flax dataset preprocessing change vs a V100 test fix; no shared underlying bug." + }, + { + "left": "pull_request:3471", + "right": "pull_request:4112", + "accept": false, + "reason": "DreamBooth README note vs LoRA rank support in training code are distinct changes." + }, + { + "left": "pull_request:3723", + "right": "pull_request:4112", + "accept": false, + "reason": "Reference image input enhancement vs DreamBooth LoRA rank support; different features." + }, + { + "left": "pull_request:3957", + "right": "pull_request:4595", + "accept": false, + "reason": "WIP attention-name correction vs XL-to-SD conversion work are unrelated." + }, + { + "left": "pull_request:4248", + "right": "pull_request:6081", + "accept": false, + "reason": "Trackers config fix in DreamBooth vs SDXL scheduler-sigmas handling; different code paths." + }, + { + "left": "pull_request:4396", + "right": "pull_request:6261", + "accept": false, + "reason": "Prompt dtype fallback fix vs pipeline package refactor; not the same bug." + }, + { + "left": "pull_request:51", + "right": "pull_request:689", + "accept": false, + "reason": "Install requirements fix vs attention implementation performance tweak are unrelated." + }, + { + "left": "pull_request:5164", + "right": "pull_request:6007", + "accept": false, + "reason": "FreeU core mechanism vs docs for the copied-from convention; completely different changes." + }, + { + "left": "pull_request:6265", + "right": "pull_request:8404", + "accept": false, + "reason": "Both touch import exposure, but one splits out GLIGEN while the other adds Latte; different features." + }, + { + "left": "pull_request:6311", + "right": "pull_request:7778", + "accept": false, + "reason": "Fast diffusion tutorial docs vs debugging workflow YAML are unrelated." + }, + { + "left": "pull_request:6864", + "right": "pull_request:9065", + "accept": false, + "reason": "PEFT torch.compile docs vs sentencepiece soft-dependency support; not the same issue." + }, + { + "left": "pull_request:7050", + "right": "pull_request:7712", + "accept": false, + "reason": "Typo cleanup in examples vs import-path fixes; unrelated." + }, + { + "left": "pull_request:7050", + "right": "pull_request:9207", + "accept": false, + "reason": "Text-to-image typo fixes vs CPU offloading docs fix; different changes." + }, + { + "left": "pull_request:7652", + "right": "pull_request:9400", + "accept": false, + "reason": "Docker buildx workflow setup vs device-agnostic test adjustments; unrelated." + }, + { + "left": "pull_request:9489", + "right": "pull_request:9579", + "accept": false, + "reason": "SD3 ControlNet training bug fix vs embeddings doc expansion; not the same underlying change." + }, + { + "left": "pull_request:9074", + "right": "pull_request:9988", + "accept": false, + "reason": "Both are Flux-related, but one is a core follow-up bug/processor fix while the other is a broader Redux/conversion feature PR; too distinct to merge as one fix." + }, + { + "left": "pull_request:1324", + "right": "pull_request:1579", + "accept": false, + "reason": "INT8 Stable Diffusion quantization docs/feature vs Intel-optimized example set; related theme but not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 170, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10259", + "pull_request:10326", + "pull_request:10354", + "pull_request:10557", + "pull_request:10808", + "pull_request:10894", + "pull_request:1129", + "pull_request:11655", + "pull_request:11663", + "pull_request:11726", + "pull_request:11738", + "pull_request:11869", + "pull_request:11913", + "pull_request:1206", + "pull_request:12119", + "pull_request:12324", + "pull_request:12387", + "pull_request:12418", + "pull_request:12534", + "pull_request:13134", + "pull_request:13208", + "pull_request:13294", + "pull_request:13296", + "pull_request:13365", + "pull_request:13562", + "pull_request:2279", + "pull_request:3746", + "pull_request:3968", + "pull_request:4346", + "pull_request:4944", + "pull_request:5298", + "pull_request:5439", + "pull_request:5713", + "pull_request:5887", + "pull_request:6757", + "pull_request:6868", + "pull_request:6929", + "pull_request:7062", + "pull_request:7068", + "pull_request:7712", + "pull_request:8419", + "pull_request:8452", + "pull_request:8564", + "pull_request:8652", + "pull_request:8684", + "pull_request:8836", + "pull_request:9206", + "pull_request:9473" + ], + "soft_pairs": [ + "pull_request:11663|pull_request:11738", + "pull_request:1129|pull_request:1206", + "pull_request:13134|pull_request:13296", + "pull_request:13294|pull_request:13562", + "pull_request:2279|pull_request:3746", + "pull_request:4944|pull_request:6868", + "pull_request:6757|pull_request:8419", + "pull_request:7068|pull_request:7712", + "pull_request:7712|pull_request:8836", + "pull_request:8452|pull_request:8652", + "pull_request:8564|pull_request:9473", + "pull_request:10808|pull_request:12119", + "pull_request:11655|pull_request:12534", + "pull_request:3968|pull_request:4346", + "pull_request:5713|pull_request:5887", + "pull_request:5298|pull_request:5439", + "pull_request:10259|pull_request:10808", + "pull_request:11869|pull_request:12387", + "pull_request:12324|pull_request:13208", + "pull_request:10326|pull_request:10557", + "pull_request:10326|pull_request:10894", + "pull_request:10354|pull_request:9206", + "pull_request:10557|pull_request:9206", + "pull_request:12418|pull_request:13365", + "pull_request:6929|pull_request:7062", + "pull_request:6929|pull_request:8684", + "pull_request:11726|pull_request:11913" + ], + "prepared_review_unit_hash": "0298a94aa8925f18a35a7faf9a0012366843c688f5053eec85e5b1068c38bb58", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 104893, + "estimated_input_tokens": 26224, + "estimated_eval_tokens": 52704 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous set of unrelated PRs; only two plausible duplicate threads stand out: bnb deterministic test stabilization (11663/11738) and SDXL watermarker optionality (3968/4346). The rest are different fixes/features despite superficial subsystem overlap.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits the whole set cleanly; the cluster is mostly false positives with only small duplicate subthreads.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:3968 is the broadest representative of the only clear duplicate thread, covering the SDXL watermarker optionality change that 4346 narrows further.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11663", + "right": "pull_request:11738", + "accept": true, + "reason": "Same bnb test determinism fix on the same files; 11738 is a refactor/consolidation of the change introduced in 11663." + }, + { + "left": "pull_request:1129", + "right": "pull_request:1206", + "accept": false, + "reason": "Unrelated: DreamBooth Flax sample batch sizing vs. LDM checkpoint layer-name conversion." + }, + { + "left": "pull_request:13134", + "right": "pull_request:13296", + "accept": false, + "reason": "Unrelated: Cosmos transformer typing vs. CI release-test transformer version pinning." + }, + { + "left": "pull_request:13294", + "right": "pull_request:13562", + "accept": false, + "reason": "Unrelated: WanVACE flaky test cleanup vs. CI tokenizers install policy." + }, + { + "left": "pull_request:2279", + "right": "pull_request:3746", + "accept": false, + "reason": "Unrelated: lint-tool migration vs. from_ckpt path joining fix." + }, + { + "left": "pull_request:4944", + "right": "pull_request:6868", + "accept": false, + "reason": "Related subsystem, but different changes: initial IP-Adapter support vs. adding ip_adapter_image_embeds to pipeline calls." + }, + { + "left": "pull_request:6757", + "right": "pull_request:8419", + "accept": false, + "reason": "Unrelated: SDXL k-diffusion sigma device placement vs. dynamic module download backend change." + }, + { + "left": "pull_request:7068", + "right": "pull_request:7712", + "accept": false, + "reason": "Unrelated: docstring/typo cleanup vs. import path fixes." + }, + { + "left": "pull_request:7712", + "right": "pull_request:8836", + "accept": false, + "reason": "Unrelated: import fixes vs. sliced attention loop bug fix." + }, + { + "left": "pull_request:8452", + "right": "pull_request:8652", + "accept": false, + "reason": "Unrelated: single-file loading error fix vs. adding Lumina-T2X." + }, + { + "left": "pull_request:8564", + "right": "pull_request:9473", + "accept": false, + "reason": "Both mention training/T5, but they fix different scripts and different bugs." + }, + { + "left": "pull_request:10808", + "right": "pull_request:12119", + "accept": false, + "reason": "Unrelated: peft config key handling vs. Qwen-Image-Lightning LoRA loading support." + }, + { + "left": "pull_request:11655", + "right": "pull_request:12534", + "accept": false, + "reason": "Same broad Flux area, but different concrete problems: 8-bit LoRA loading vs. NPU positional embedding inference." + }, + { + "left": "pull_request:3968", + "right": "pull_request:4346", + "accept": true, + "reason": "Same SDXL watermarker optionality change on overlapping files; 3968 is the broader umbrella and 4346 is a focused follow-up on the same code path." + }, + { + "left": "pull_request:5713", + "right": "pull_request:5887", + "accept": false, + "reason": "5887 extends IP-Adapter to one pipeline, but it is not the same change as the core IP-Adapter feature in 5713." + }, + { + "left": "pull_request:5298", + "right": "pull_request:5439", + "accept": false, + "reason": "Unrelated: LoRA conv torch.compile fix vs. single-file tokenizer/config loading." + }, + { + "left": "pull_request:10259", + "right": "pull_request:10808", + "accept": false, + "reason": "Unrelated: Flux LoRA loading support vs. PEFT config-update logic." + }, + { + "left": "pull_request:11869", + "right": "pull_request:12387", + "accept": false, + "reason": "Both involve kernels, but one adds GGUF CUDA kernels and the other adds flash-attention backend support; different code paths." + }, + { + "left": "pull_request:12324", + "right": "pull_request:13208", + "accept": false, + "reason": "Unrelated model additions: Wan2.2 VACE-Fun vs. Helios video pipelines." + }, + { + "left": "pull_request:10326", + "right": "pull_request:10557", + "accept": false, + "reason": "Unrelated CI workflow fix vs. single-file slow-test fix." + }, + { + "left": "pull_request:10326", + "right": "pull_request:10894", + "accept": false, + "reason": "Unrelated CI workflow fix vs. docs toctree sorting." + }, + { + "left": "pull_request:10354", + "right": "pull_request:9206", + "accept": false, + "reason": "Unrelated: Flux IP-Adapter tester dtype/device fix vs. xformers test skips." + }, + { + "left": "pull_request:10557", + "right": "pull_request:9206", + "accept": false, + "reason": "Unrelated test areas and failure modes." + }, + { + "left": "pull_request:12418", + "right": "pull_request:13365", + "accept": false, + "reason": "Unrelated: xfail CI tests vs. Claude review workflow support for forks." + }, + { + "left": "pull_request:6929", + "right": "pull_request:7062", + "accept": false, + "reason": "Unrelated docs changes: community pipelines loading vs. imagic docstring correction." + }, + { + "left": "pull_request:6929", + "right": "pull_request:8684", + "accept": false, + "reason": "Unrelated docs/example fixes in different pipelines." + }, + { + "left": "pull_request:11726", + "right": "pull_request:11913", + "accept": false, + "reason": "Both Wan-related, but one is LoRA conversion support and the other is the modular Wan pipeline; not the same change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: most pairs are clearly unrelated, and the two accepted threads have plausible same-code-path overlap with the later PRs looking like narrow follow-ups. The verdicts are generally conservative, and I don\u2019t see a clear overstatement of evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 171, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10114", + "pull_request:10419", + "pull_request:10723", + "pull_request:10758", + "pull_request:10922", + "pull_request:10943", + "pull_request:11095", + "pull_request:11252", + "pull_request:1129", + "pull_request:1171", + "pull_request:11770", + "pull_request:11913", + "pull_request:11995", + "pull_request:12062", + "pull_request:12395", + "pull_request:1250", + "pull_request:12934", + "pull_request:1309", + "pull_request:13229", + "pull_request:1338", + "pull_request:1588", + "pull_request:1968", + "pull_request:2037", + "pull_request:2339", + "pull_request:2347", + "pull_request:2384", + "pull_request:2489", + "pull_request:2529", + "pull_request:2565", + "pull_request:2590", + "pull_request:2593", + "pull_request:2861", + "pull_request:3121", + "pull_request:326", + "pull_request:3299", + "pull_request:3506", + "pull_request:3563", + "pull_request:3589", + "pull_request:3681", + "pull_request:3711", + "pull_request:4112", + "pull_request:4395", + "pull_request:4465", + "pull_request:530", + "pull_request:5499", + "pull_request:788", + "pull_request:9968" + ], + "soft_pairs": [ + "pull_request:10723|pull_request:10758", + "pull_request:11913|pull_request:11995", + "pull_request:10114|pull_request:11252", + "pull_request:10419|pull_request:11770", + "pull_request:10922|pull_request:10943", + "pull_request:11095|pull_request:9968", + "pull_request:1129|pull_request:1171", + "pull_request:12062|pull_request:12395", + "pull_request:1250|pull_request:326", + "pull_request:12934|pull_request:13229", + "pull_request:1309|pull_request:788", + "pull_request:1338|pull_request:530", + "pull_request:1588|pull_request:2861", + "pull_request:1968|pull_request:2565", + "pull_request:2037|pull_request:326", + "pull_request:2339|pull_request:4395", + "pull_request:2347|pull_request:4112", + "pull_request:2384|pull_request:3711", + "pull_request:2489|pull_request:3121", + "pull_request:2529|pull_request:4465", + "pull_request:2590|pull_request:3681", + "pull_request:2593|pull_request:3506", + "pull_request:3299|pull_request:3589", + "pull_request:3563|pull_request:5499" + ], + "prepared_review_unit_hash": "844999e387b261f711415d318517fa6ff5c9a305c627a8bd1d7f18322bb346f2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 104852, + "estimated_input_tokens": 26213, + "estimated_eval_tokens": 52682 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft-edge pairs are largely false positives: they share broad keywords or subsystems (Wan, ControlNet, docs, CI, DreamBooth) but do not appear to fix the same concrete bug or implement the same change. I would not merge any of these pairs as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out because the cluster is heterogeneous; if forced, PR 11995 is the broadest self-contained feature, but it is not a true duplicate anchor.", + "best_issue_reason": null, + "best_pr_reason": "PR 11995 is the closest thing to a broad representative, but this set is too mixed for any PR to serve as a strong canonical duplicate target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10723", + "right": "pull_request:10758", + "accept": false, + "reason": "Different changes: ControlNet Union scaling vs from_pretrained type-checking docs/code." + }, + { + "left": "pull_request:11913", + "right": "pull_request:11995", + "accept": false, + "reason": "Both are modular-pipeline work, but they target different models and features (Wan vs Flux)." + }, + { + "left": "pull_request:10114", + "right": "pull_request:11252", + "accept": false, + "reason": "Unrelated edits: Stable Audio formatting vs a timeout constant fix in a utility script." + }, + { + "left": "pull_request:10419", + "right": "pull_request:11770", + "accept": false, + "reason": "Both touch docs, but one fixes scheduler docstrings and the other cleans up LoRA docs/base helpers." + }, + { + "left": "pull_request:10922", + "right": "pull_request:10943", + "accept": false, + "reason": "Same Wan ecosystem, but one adds Wan pipeline/model support and the other adds LoRA support; distinct changes." + }, + { + "left": "pull_request:11095", + "right": "pull_request:9968", + "accept": false, + "reason": "A small Flux ControlNet doc fix is not the same as a broad ControlNet module refactor." + }, + { + "left": "pull_request:1129", + "right": "pull_request:1171", + "accept": false, + "reason": "DreamBooth batch-size bug fix vs a typography/doc cleanup; unrelated underlying changes." + }, + { + "left": "pull_request:12062", + "right": "pull_request:12395", + "accept": false, + "reason": "Different CI concerns: GPU selection in workflows vs prerelease dependency installation." + }, + { + "left": "pull_request:1250", + "right": "pull_request:326", + "accept": false, + "reason": "Community pipeline addition is unrelated to a PNDM type-hint/doc change." + }, + { + "left": "pull_request:12934", + "right": "pull_request:13229", + "accept": false, + "reason": "Both are video-model releases, but they introduce different model support and code paths." + }, + { + "left": "pull_request:1309", + "right": "pull_request:788", + "accept": false, + "reason": "Both mention PIL, but one improves PIL handling broadly while the other fixes a specific warning issue." + }, + { + "left": "pull_request:1338", + "right": "pull_request:530", + "accept": false, + "reason": "ONNX inpainting cast order and img2img test flakiness are unrelated." + }, + { + "left": "pull_request:1588", + "right": "pull_request:2861", + "accept": false, + "reason": "FP16 test adjustments and StableUnCLIP image_embeds handling are different bugs." + }, + { + "left": "pull_request:1968", + "right": "pull_request:2565", + "accept": false, + "reason": "CPU offload docs/code and a UNet1D doc correction are unrelated." + }, + { + "left": "pull_request:2037", + "right": "pull_request:326", + "accept": false, + "reason": "A Stable UnCLIP community pipeline and PNDM type hints are not the same change." + }, + { + "left": "pull_request:2339", + "right": "pull_request:4395", + "accept": false, + "reason": "Both are DreamBooth-related, but one is checkpointing/docs and the other is class-prompt pre-encoding logic." + }, + { + "left": "pull_request:2347", + "right": "pull_request:4112", + "accept": false, + "reason": "Scheduler noising-schedule work is unrelated to adding rank to DreamBooth LoRA." + }, + { + "left": "pull_request:2384", + "right": "pull_request:3711", + "accept": false, + "reason": "A Pix2Pix0 utility revert and a doc-build workflow revert are unrelated." + }, + { + "left": "pull_request:2489", + "right": "pull_request:3121", + "accept": false, + "reason": "Core ML docs and ControlNet global pooling fix different concerns entirely." + }, + { + "left": "pull_request:2529", + "right": "pull_request:4465", + "accept": false, + "reason": "Stable Diffusion checkpoint conversion and ControlNet-to-TensorRT conversion are different conversion pipelines." + }, + { + "left": "pull_request:2590", + "right": "pull_request:3681", + "accept": false, + "reason": "CI deps-table checking and checkpoint-limit handling are unrelated." + }, + { + "left": "pull_request:2593", + "right": "pull_request:3506", + "accept": false, + "reason": "ControlNet checkpoint conversion and DreamBooth checkpoint resume logic are unrelated." + }, + { + "left": "pull_request:3299", + "right": "pull_request:3589", + "accept": false, + "reason": "DreamBooth LoRA doc/example updates and README reference tricks are separate doc changes." + }, + { + "left": "pull_request:3563", + "right": "pull_request:5499", + "accept": false, + "reason": "A VQ pipeline bug fix and general docs link fixes do not share the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 172, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10324", + "pull_request:10630", + "pull_request:11124", + "pull_request:11469", + "pull_request:11995", + "pull_request:12041", + "pull_request:12277", + "pull_request:12416", + "pull_request:12523", + "pull_request:1277", + "pull_request:1579", + "pull_request:1966", + "pull_request:2727", + "pull_request:3850", + "pull_request:3928", + "pull_request:4040", + "pull_request:4078", + "pull_request:4690", + "pull_request:5053", + "pull_request:5122", + "pull_request:5312", + "pull_request:5440", + "pull_request:5695", + "pull_request:5698", + "pull_request:6007", + "pull_request:6081", + "pull_request:6099", + "pull_request:6145", + "pull_request:6217", + "pull_request:652", + "pull_request:6658", + "pull_request:6893", + "pull_request:6918", + "pull_request:6965", + "pull_request:7158", + "pull_request:7335", + "pull_request:7455", + "pull_request:754", + "pull_request:7916", + "pull_request:7919", + "pull_request:7930", + "pull_request:8282", + "pull_request:8531", + "pull_request:8761", + "pull_request:9206", + "pull_request:9473", + "pull_request:9751" + ], + "soft_pairs": [ + "pull_request:3850|pull_request:5695", + "pull_request:3850|pull_request:6099", + "pull_request:3928|pull_request:4040", + "pull_request:4078|pull_request:5122", + "pull_request:4690|pull_request:6081", + "pull_request:5312|pull_request:6658", + "pull_request:5698|pull_request:6145", + "pull_request:6007|pull_request:8282", + "pull_request:6217|pull_request:8531", + "pull_request:652|pull_request:754", + "pull_request:6893|pull_request:8761", + "pull_request:6918|pull_request:7919", + "pull_request:6965|pull_request:7158", + "pull_request:7335|pull_request:7930", + "pull_request:7455|pull_request:9473", + "pull_request:7916|pull_request:9751", + "pull_request:1966|pull_request:2727", + "pull_request:12041|pull_request:12523", + "pull_request:11995|pull_request:12416", + "pull_request:5053|pull_request:5440", + "pull_request:10324|pull_request:9206", + "pull_request:10630|pull_request:11124", + "pull_request:11469|pull_request:12277", + "pull_request:1277|pull_request:1579" + ], + "prepared_review_unit_hash": "dca7a9469f7b4427eb618e5e1af86ee4217f477149b7058c930c6f0ee6e41955", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 101413, + "estimated_input_tokens": 25354, + "estimated_eval_tokens": 50964 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated test tweaks, docs edits, scheduler/model fixes, and feature PRs across many subsystems. I don\u2019t see any true duplicate pair here, so there is no sensible canonical artifact for the cluster.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:3850", + "right": "pull_request:5695", + "accept": false, + "reason": "Different pipelines and fixes: AudioDiffusion test setup vs PixArt-Alpha mask-feature logic." + }, + { + "left": "pull_request:3850", + "right": "pull_request:6099", + "accept": false, + "reason": "AudioDiffusion test fix is unrelated to TextToVideoZero constructor/super-class argument handling." + }, + { + "left": "pull_request:3928", + "right": "pull_request:4040", + "accept": false, + "reason": "ControlNet guidance-timestep bug and `_use_default_values` type change are different code paths and bug classes." + }, + { + "left": "pull_request:4078", + "right": "pull_request:5122", + "accept": false, + "reason": "Single-file accelerate requirement is unrelated to workflow Python-version bumps." + }, + { + "left": "pull_request:4690", + "right": "pull_request:6081", + "accept": false, + "reason": "DPMSolver scheduler refactor is not the same issue as the Stable Diffusion XL torch.compile sigma workaround." + }, + { + "left": "pull_request:5312", + "right": "pull_request:6658", + "accept": false, + "reason": "Docs/typehint cleanup in models is unrelated to adding memory-footprint utilities." + }, + { + "left": "pull_request:5698", + "right": "pull_request:6145", + "accept": false, + "reason": "LCM img2img bug fix is unrelated to comment cleanup in distillation scripts." + }, + { + "left": "pull_request:6007", + "right": "pull_request:8282", + "accept": false, + "reason": "Both are docs, but they cover different topics and changes, not the same underlying change." + }, + { + "left": "pull_request:6217", + "right": "pull_request:8531", + "accept": false, + "reason": "Torch-compile VAE fix and sharding/device_map loading fix affect different execution paths." + }, + { + "left": "pull_request:652", + "right": "pull_request:754", + "accept": false, + "reason": "Gradient-checkpointing test tolerance change is unrelated to community-pipeline dynamic module handling." + }, + { + "left": "pull_request:6893", + "right": "pull_request:8761", + "accept": false, + "reason": "3D UNet attention refactor and Hunyuan embedding refactor touch different models and problems." + }, + { + "left": "pull_request:6918", + "right": "pull_request:7919", + "accept": false, + "reason": "Hub-token security fix in training scripts is unrelated to offload behavior under `device_map`." + }, + { + "left": "pull_request:6965", + "right": "pull_request:7158", + "accept": false, + "reason": "A PEFT workflow test move and an LCM benchmark fix are unrelated test changes." + }, + { + "left": "pull_request:7335", + "right": "pull_request:7930", + "accept": false, + "reason": "ControlNet SDXL config/test cleanup and official callback tests do not fix the same bug." + }, + { + "left": "pull_request:7455", + "right": "pull_request:9473", + "accept": false, + "reason": "Removing distutils and Flux Dreambooth LR logic are unrelated changes." + }, + { + "left": "pull_request:7916", + "right": "pull_request:9751", + "accept": false, + "reason": "Both add NPU support, but for different models/pipelines and different code paths." + }, + { + "left": "pull_request:1966", + "right": "pull_request:2727", + "accept": false, + "reason": "Flax inpainting implementation and Flax ControlNet are separate feature additions." + }, + { + "left": "pull_request:12041", + "right": "pull_request:12523", + "accept": false, + "reason": "Both touch AutoencoderKLWan, but one fixes patchify logic while the other changes skip-key handling; not the same concrete bug." + }, + { + "left": "pull_request:11995", + "right": "pull_request:12416", + "accept": false, + "reason": "Different modular-pipeline additions: Flux text-to-image vs QwenImage Edit Plus." + }, + { + "left": "pull_request:5053", + "right": "pull_request:5440", + "accept": false, + "reason": "Validation-image arg parsing and single-file loading are unrelated changes." + }, + { + "left": "pull_request:10324", + "right": "pull_request:9206", + "accept": false, + "reason": "Kandinsky CPU offload API fix is unrelated to broken xformers test suppression." + }, + { + "left": "pull_request:10630", + "right": "pull_request:11124", + "accept": false, + "reason": "Flax test slice updates and CogView4 tokenizer test fix are unrelated test-only changes." + }, + { + "left": "pull_request:11469", + "right": "pull_request:12277", + "accept": false, + "reason": "Both disable tests, but for different pipelines and reasons; they do not share one concrete bug." + }, + { + "left": "pull_request:1277", + "right": "pull_request:1579", + "accept": false, + "reason": "INT8 quantization docs and Intel optimization example additions are different documentation/features." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 173, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10188", + "pull_request:10318", + "pull_request:10360", + "pull_request:10378", + "pull_request:10719", + "pull_request:10921", + "pull_request:11087", + "pull_request:11196", + "pull_request:11319", + "pull_request:11421", + "pull_request:11479", + "pull_request:11504", + "pull_request:12099", + "pull_request:12363", + "pull_request:12397", + "pull_request:12491", + "pull_request:12493", + "pull_request:12687", + "pull_request:13029", + "pull_request:13119", + "pull_request:13204", + "pull_request:1331", + "pull_request:1968", + "pull_request:2467", + "pull_request:3023", + "pull_request:3297", + "pull_request:3446", + "pull_request:349", + "pull_request:3607", + "pull_request:3673", + "pull_request:3708", + "pull_request:3804", + "pull_request:3847", + "pull_request:3912", + "pull_request:3957", + "pull_request:4040", + "pull_request:4596", + "pull_request:5057", + "pull_request:5653", + "pull_request:5754", + "pull_request:5948", + "pull_request:6144", + "pull_request:6150", + "pull_request:833", + "pull_request:8691", + "pull_request:8964", + "pull_request:9878" + ], + "soft_pairs": [ + "pull_request:5057|pull_request:5653", + "pull_request:10188|pull_request:8691", + "pull_request:10318|pull_request:10719", + "pull_request:10360|pull_request:8964", + "pull_request:10378|pull_request:11319", + "pull_request:10921|pull_request:11087", + "pull_request:11196|pull_request:12363", + "pull_request:11421|pull_request:9878", + "pull_request:11479|pull_request:11504", + "pull_request:12099|pull_request:12687", + "pull_request:12397|pull_request:12491", + "pull_request:12493|pull_request:13119", + "pull_request:13029|pull_request:13204", + "pull_request:1331|pull_request:349", + "pull_request:1968|pull_request:3673", + "pull_request:2467|pull_request:4040", + "pull_request:3023|pull_request:3957", + "pull_request:3023|pull_request:833", + "pull_request:3297|pull_request:4596", + "pull_request:3446|pull_request:5754", + "pull_request:3607|pull_request:3804", + "pull_request:3708|pull_request:5948", + "pull_request:3847|pull_request:6150", + "pull_request:3912|pull_request:6144" + ], + "prepared_review_unit_hash": "297f21d7ebf9c63c4c14a0f01d93deefdb658c26653a223fc7f4971e7f6bbfbd", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 97525, + "estimated_input_tokens": 24382, + "estimated_eval_tokens": 49020 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft duplicate pair looks like the same underlying bug/change. Most pairs only share a broad subsystem or theme (LoRA, Wan, docs, CI, attention), but the concrete edits and intent differ, so this cluster should not be merged into a single canonical PR.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical representative here: the items are unrelated fixes/features across different code paths and docs, not one duplicate change.", + "best_issue_reason": null, + "best_pr_reason": "No PR stands out as a global best duplicate target because the cluster is not one coherent topic.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5057", + "right": "pull_request:5653", + "accept": false, + "reason": "Different changes: SDXL clip-skip support vs PEFT-aware model unwrapping in pipeline loading." + }, + { + "left": "pull_request:10188", + "right": "pull_request:8691", + "accept": false, + "reason": "Both touch LoRA loading, but one fixes mixed LoRA expansion handling in Flux and the other changes vanilla fine-tuned LoRA loading logic." + }, + { + "left": "pull_request:10318", + "right": "pull_request:10719", + "accept": false, + "reason": "Integration tests for Flux LoRA expansion are unrelated to adding provider_options to from_pretrained." + }, + { + "left": "pull_request:10360", + "right": "pull_request:8964", + "accept": false, + "reason": "Release workflow TestPyPI URL fix is unrelated to AudioLDM2 cache-position generation fix." + }, + { + "left": "pull_request:10378", + "right": "pull_request:11319", + "accept": false, + "reason": "Dreambooth LoRA training support is unrelated to adding a community-project docs entry." + }, + { + "left": "pull_request:10921", + "right": "pull_request:11087", + "accept": false, + "reason": "Wan 2.1 support is a large feature PR; the other is a small Wan I2V quality fix in one function." + }, + { + "left": "pull_request:11196", + "right": "pull_request:12363", + "accept": false, + "reason": "Safe-serialization behavior in ModelMixin is unrelated to QwenImageEdit Plus docs/tests." + }, + { + "left": "pull_request:11421", + "right": "pull_request:9878", + "accept": false, + "reason": "Interpolation default in dreambooth training is unrelated to gradient-checkpointing behavior across eval mode." + }, + { + "left": "pull_request:11479", + "right": "pull_request:11504", + "accept": false, + "reason": "Intel Gaudi doc rename is unrelated to dependency-table and CI trigger updates." + }, + { + "left": "pull_request:12099", + "right": "pull_request:12687", + "accept": false, + "reason": "Qwen device placement typo fix is unrelated to an autoencoder_dc typo rename." + }, + { + "left": "pull_request:12397", + "right": "pull_request:12491", + "accept": false, + "reason": "CI disabling main-branch installs is unrelated to adding a documentation link-check job." + }, + { + "left": "pull_request:12493", + "right": "pull_request:13119", + "accept": false, + "reason": "LTX-Video 0.9.8 docs/code updates are unrelated to Flux2 LoRA key conversion handling." + }, + { + "left": "pull_request:13029", + "right": "pull_request:13204", + "accept": false, + "reason": "Modular quickstart docs are unrelated to modular pipeline trust_remote_code handling." + }, + { + "left": "pull_request:1331", + "right": "pull_request:349", + "accept": false, + "reason": "Inpainting scheduler-order fix is unrelated to DDPM type-hint cleanup." + }, + { + "left": "pull_request:1968", + "right": "pull_request:3673", + "accept": false, + "reason": "CPU offload doc/correction work is unrelated to Dreambooth multi-GPU config handling." + }, + { + "left": "pull_request:2467", + "right": "pull_request:4040", + "accept": false, + "reason": "Latent output type feature PR is unrelated to keeping _use_default_values as a list." + }, + { + "left": "pull_request:3023", + "right": "pull_request:3957", + "accept": false, + "reason": "Both are attention-related, but one adds KV attention processor behavior and the other is an attn-name correction WIP; different code paths and goals." + }, + { + "left": "pull_request:3023", + "right": "pull_request:833", + "accept": false, + "reason": "Attention/KV processor work is unrelated to a release-version bump." + }, + { + "left": "pull_request:3297", + "right": "pull_request:4596", + "accept": false, + "reason": "Retrieval-augmented diffusion models project code is unrelated to an SDXL checkpoint conversion script." + }, + { + "left": "pull_request:3446", + "right": "pull_request:5754", + "accept": false, + "reason": "Pipeline loading warning-message cleanup is unrelated to styling fixes in example training scripts." + }, + { + "left": "pull_request:3607", + "right": "pull_request:3804", + "accept": false, + "reason": "Both mention attention, but one fixes temb attention and the other changes legacy attention block weight conversion for device maps." + }, + { + "left": "pull_request:3708", + "right": "pull_request:5948", + "accept": false, + "reason": "Custom release-version sorting is unrelated to from_pretrained variant warning logic." + }, + { + "left": "pull_request:3847", + "right": "pull_request:6150", + "accept": false, + "reason": "Soft-dependency tests/CI setup are unrelated to IP-Adapter converter implementation." + }, + { + "left": "pull_request:3912", + "right": "pull_request:6144", + "accept": false, + "reason": "Dreambooth LoRA text-encoder finetuning feature is unrelated to unwrapped UNet LoRA state saving." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 174, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10086", + "pull_request:10096", + "pull_request:10119", + "pull_request:10132", + "pull_request:10188", + "pull_request:10342", + "pull_request:10544", + "pull_request:10922", + "pull_request:11243", + "pull_request:12109", + "pull_request:12215", + "pull_request:12416", + "pull_request:12519", + "pull_request:13352", + "pull_request:3001", + "pull_request:3128", + "pull_request:4352", + "pull_request:4417", + "pull_request:4855", + "pull_request:4986", + "pull_request:5261", + "pull_request:5361", + "pull_request:5419", + "pull_request:5771", + "pull_request:5938", + "pull_request:6262", + "pull_request:6263", + "pull_request:6264", + "pull_request:6303", + "pull_request:6432", + "pull_request:6904", + "pull_request:6975", + "pull_request:7044", + "pull_request:7216", + "pull_request:7359", + "pull_request:7455", + "pull_request:7662", + "pull_request:7738", + "pull_request:8467", + "pull_request:8496", + "pull_request:8618", + "pull_request:8627", + "pull_request:9060", + "pull_request:9175", + "pull_request:9569", + "pull_request:9918", + "pull_request:9938", + "pull_request:9985" + ], + "soft_pairs": [ + "pull_request:4855|pull_request:6904", + "pull_request:5771|pull_request:6262", + "pull_request:5771|pull_request:6263", + "pull_request:5771|pull_request:6264", + "pull_request:6432|pull_request:7216", + "pull_request:7044|pull_request:7662", + "pull_request:7455|pull_request:7662", + "pull_request:7738|pull_request:8496", + "pull_request:8467|pull_request:9569", + "pull_request:8618|pull_request:9918", + "pull_request:9175|pull_request:9985", + "pull_request:5938|pull_request:6975", + "pull_request:12519|pull_request:13352", + "pull_request:4352|pull_request:4417", + "pull_request:5361|pull_request:5419", + "pull_request:6303|pull_request:7359", + "pull_request:3001|pull_request:4986", + "pull_request:3128|pull_request:5261", + "pull_request:12109|pull_request:12416", + "pull_request:12215|pull_request:12416", + "pull_request:10086|pull_request:10922", + "pull_request:10096|pull_request:8627", + "pull_request:10119|pull_request:10544", + "pull_request:10132|pull_request:9060", + "pull_request:10188|pull_request:11243", + "pull_request:10342|pull_request:9938" + ], + "prepared_review_unit_hash": "29fdd1a6f55ef8dc7d4a6370a5e3db034d1b8993cf14f78a2aca82e6fbe333c0", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 104279, + "estimated_input_tokens": 26070, + "estimated_eval_tokens": 52396 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft-edge candidates are mostly false positives: they share broad subsystems or adjacent themes, but they do not fix the same concrete bug or implement the same change. No issue is present, and there is no strong canonical PR anchor; the Flux/ControlNet pair is the closest on surface area but still too broad and different.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents a duplicate cluster here; the items span unrelated fixes/features, so there is no strong canonical PR anchor.", + "best_issue_reason": null, + "best_pr_reason": "PR 9985 is the broadest and most central feature PR in the set, but it is not a good duplicate representative because it bundles multiple Flux features rather than one specific fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4855", + "right": "pull_request:6904", + "accept": false, + "reason": "Unrelated fixes: CPU offload/state-hook handling vs PIL image loading support. Same repo area, different code paths." + }, + { + "left": "pull_request:5771", + "right": "pull_request:6262", + "accept": false, + "reason": "Refactor of loaders vs moving Panorama out of stable_diffusion; different subsystems and different changes." + }, + { + "left": "pull_request:5771", + "right": "pull_request:6263", + "accept": false, + "reason": "Refactor of loaders vs moving LDM3D out of stable_diffusion; not the same bug or change." + }, + { + "left": "pull_request:5771", + "right": "pull_request:6264", + "accept": false, + "reason": "Refactor of loaders vs moving SAG out of stable_diffusion; unrelated code paths." + }, + { + "left": "pull_request:6432", + "right": "pull_request:7216", + "accept": false, + "reason": "README learning-rate update vs from_single_file signature/documentation change. No shared concrete fix." + }, + { + "left": "pull_request:7044", + "right": "pull_request:7662", + "accept": false, + "reason": "Unet refactor restoration vs workflow cleanup in flax tests; unrelated changes." + }, + { + "left": "pull_request:7455", + "right": "pull_request:7662", + "accept": false, + "reason": "Distutils removal vs removing redundant packages in CI. Same general maintenance theme, but not the same change." + }, + { + "left": "pull_request:7738", + "right": "pull_request:8496", + "accept": false, + "reason": "LoRA device-placement fix vs SD3 warning-log guard. Different bugs and code paths." + }, + { + "left": "pull_request:8467", + "right": "pull_request:9569", + "accept": false, + "reason": "SEGA pipeline device fix vs CogVideoX autoencoder decode fix; unrelated models and failures." + }, + { + "left": "pull_request:8618", + "right": "pull_request:9918", + "accept": false, + "reason": "SD3 tiny autoencoder shift_factor change vs adding a server example; totally different deliverables." + }, + { + "left": "pull_request:9175", + "right": "pull_request:9985", + "accept": false, + "reason": "Both touch Flux, but one adds Union ControlNet support while the other adds Fill/Canny/Depth/Redux variants. Broadly related, but not the same concrete feature or mergeable single PR." + }, + { + "left": "pull_request:5938", + "right": "pull_request:6975", + "accept": false, + "reason": "ControlNet LoRA support vs single-file ControlNet inpaint loading fix. Different functionality and code paths." + }, + { + "left": "pull_request:12519", + "right": "pull_request:13352", + "accept": false, + "reason": "Toctree sorting fix vs adding a checkout step to Claude review workflow; unrelated CI/docs changes." + }, + { + "left": "pull_request:4352", + "right": "pull_request:4417", + "accept": false, + "reason": "Kandinsky combined pipeline fixes vs Zero123 community pipeline support; unrelated features." + }, + { + "left": "pull_request:5361", + "right": "pull_request:5419", + "accept": false, + "reason": "Temporary PEFT integration testing PR vs adding PEFT CUDA slow tests on push. Same testing domain, but not the same change." + }, + { + "left": "pull_request:6303", + "right": "pull_request:7359", + "accept": false, + "reason": "ADD training script vs ResAdapter training script; both research examples, but different projects and code." + }, + { + "left": "pull_request:3001", + "right": "pull_request:4986", + "accept": false, + "reason": "Karras support in one scheduler vs a broad sigma-based refactor across multiple schedulers. Overlapping area, but not the same underlying change." + }, + { + "left": "pull_request:3128", + "right": "pull_request:5261", + "accept": false, + "reason": "Img2img timestep selection fix vs callback step indexing fix. Both scheduler-adjacent, but distinct bugs." + }, + { + "left": "pull_request:12109", + "right": "pull_request:12416", + "accept": false, + "reason": "UniPC custom sigmas/timesteps support vs QwenImage Edit Plus modular pipeline support; unrelated features." + }, + { + "left": "pull_request:12215", + "right": "pull_request:12416", + "accept": false, + "reason": "Qwen-Image ControlNet support vs QwenImage Edit Plus modular support. Same product family, but different concrete capabilities." + }, + { + "left": "pull_request:10086", + "right": "pull_request:10922", + "accept": false, + "reason": "SD3 skip-guidance bug fix vs Wan2.1 continuation PR; unrelated." + }, + { + "left": "pull_request:10096", + "right": "pull_request:8627", + "accept": false, + "reason": "Stale-bot label exemption vs SD3 ControlNet test trimming. No shared change." + }, + { + "left": "pull_request:10119", + "right": "pull_request:10544", + "accept": false, + "reason": "Docs for load_lora_adapter vs ParaAttention docs. Both documentation-only, but unrelated topics." + }, + { + "left": "pull_request:10132", + "right": "pull_request:9060", + "accept": false, + "reason": "Tokenizers CI update vs removing a debug print in transformer_flux. Different concerns and code paths." + }, + { + "left": "pull_request:10188", + "right": "pull_request:11243", + "accept": false, + "reason": "Flux LoRA error handling/test vs Wan LoRA conversion support. Both LoRA-related, but they fix different model families and behaviors." + }, + { + "left": "pull_request:10342", + "right": "pull_request:9938", + "accept": false, + "reason": "get_parameter_dtype behavior update vs SDXL reference pipeline docs/example update. Unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 175, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10351", + "pull_request:10547", + "pull_request:10557", + "pull_request:10810", + "pull_request:10894", + "pull_request:11059", + "pull_request:11252", + "pull_request:11743", + "pull_request:11813", + "pull_request:11894", + "pull_request:12018", + "pull_request:12231", + "pull_request:12315", + "pull_request:12424", + "pull_request:12450", + "pull_request:13119", + "pull_request:13120", + "pull_request:13204", + "pull_request:13338", + "pull_request:13536", + "pull_request:1414", + "pull_request:1415", + "pull_request:1468", + "pull_request:1494", + "pull_request:1945", + "pull_request:1951", + "pull_request:1965", + "pull_request:2101", + "pull_request:225", + "pull_request:2314", + "pull_request:2470", + "pull_request:2529", + "pull_request:2553", + "pull_request:2637", + "pull_request:2730", + "pull_request:2734", + "pull_request:3245", + "pull_request:3398", + "pull_request:3614", + "pull_request:3766", + "pull_request:4188", + "pull_request:889", + "pull_request:8964", + "pull_request:908", + "pull_request:957", + "pull_request:9676", + "pull_request:9684", + "pull_request:9687" + ], + "soft_pairs": [ + "pull_request:10351|pull_request:10557", + "pull_request:10547|pull_request:8964", + "pull_request:10557|pull_request:9676", + "pull_request:10810|pull_request:9684", + "pull_request:10810|pull_request:9687", + "pull_request:10894|pull_request:9676", + "pull_request:11059|pull_request:12315", + "pull_request:11252|pull_request:12424", + "pull_request:11743|pull_request:11813", + "pull_request:11894|pull_request:12450", + "pull_request:12018|pull_request:12231", + "pull_request:12315|pull_request:13119", + "pull_request:13120|pull_request:13204", + "pull_request:13338|pull_request:13536", + "pull_request:1414|pull_request:957", + "pull_request:1415|pull_request:2529", + "pull_request:1468|pull_request:1494", + "pull_request:1945|pull_request:3766", + "pull_request:1951|pull_request:225", + "pull_request:1965|pull_request:908", + "pull_request:2101|pull_request:2734", + "pull_request:2314|pull_request:3614", + "pull_request:2470|pull_request:4188", + "pull_request:2553|pull_request:889", + "pull_request:2637|pull_request:3245", + "pull_request:2637|pull_request:3398", + "pull_request:2730|pull_request:3245", + "pull_request:2730|pull_request:3398" + ], + "prepared_review_unit_hash": "32ce370ca94c5130e0be4829bf0abdc24c22302abe7b3e5d77042fdba5aa6ae7", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 97523, + "estimated_input_tokens": 24381, + "estimated_eval_tokens": 49018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is a grab bag of unrelated PRs across tests, docs, CI, training scripts, and loader/model fixes. None of the soft pairs look like the same underlying change, so all should be rejected as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "pull_request:10810 is the strongest standalone code fix in the set: it updates LoRA conversion logic in src/diffusers/loaders/lora_conversion_utils.py for Flux support, making it the most substantive and representative PR here.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10351", + "right": "pull_request:10557", + "accept": false, + "reason": "Different test targets: LoRA NaN-fusing test cleanup vs CogVideoX/CogView3 offload test tuning." + }, + { + "left": "pull_request:10547", + "right": "pull_request:8964", + "accept": false, + "reason": "Unrelated subsystems: transformer offload test settings vs AudioLDM2 generation cache-position fix." + }, + { + "left": "pull_request:10557", + "right": "pull_request:9676", + "accept": false, + "reason": "One is a LoRA test helper tweak; the other is a CI import-path cleanup for a benchmark script." + }, + { + "left": "pull_request:10810", + "right": "pull_request:9684", + "accept": false, + "reason": "Different fixes: Flux LoRA conversion support vs CogVideoX VAE batching behavior." + }, + { + "left": "pull_request:10810", + "right": "pull_request:9687", + "accept": false, + "reason": "Both are model-related but not the same change: Flux LoRA conversion vs Xlabs ControlNet support." + }, + { + "left": "pull_request:10894", + "right": "pull_request:9676", + "accept": false, + "reason": "Docs toctree sorting is unrelated to the benchmark import fix." + }, + { + "left": "pull_request:11059", + "right": "pull_request:12315", + "accept": false, + "reason": "Non-diffusers WAN LoRA conversion support is unrelated to pruning scheduler setup in LoRA tests." + }, + { + "left": "pull_request:11252", + "right": "pull_request:12424", + "accept": false, + "reason": "Timeout constant fix in a utility script is unrelated to Dockerfile definition cleanup." + }, + { + "left": "pull_request:11743", + "right": "pull_request:11813", + "accept": false, + "reason": "Different training artifacts: a shortened Flux research script vs Kontext training implementation and docs." + }, + { + "left": "pull_request:11894", + "right": "pull_request:12450", + "accept": false, + "reason": "Both touch CI/dependencies, but one pins k-diffusion and the other broadly revises CI installation flow." + }, + { + "left": "pull_request:12018", + "right": "pull_request:12231", + "accept": false, + "reason": "A docs link change is not the same as a typo fix across community example docs." + }, + { + "left": "pull_request:12315", + "right": "pull_request:13119", + "accept": false, + "reason": "LoRA test simplification and Flux2 LoRA key conversion address different code paths and failure modes." + }, + { + "left": "pull_request:13120", + "right": "pull_request:13204", + "accept": false, + "reason": "Shared modular-pipeline test file aside, one adds AutoModel load-id tests while the other changes trust_remote_code handling in pipeline code." + }, + { + "left": "pull_request:13338", + "right": "pull_request:13536", + "accept": false, + "reason": "Same workflow file, but distinct permission changes and workflow behavior; not one mergeable bug fix." + }, + { + "left": "pull_request:1414", + "right": "pull_request:957", + "accept": false, + "reason": "Very old clip-guided pipeline fix vs a training-scheduler bug fix in text-to-image; unrelated." + }, + { + "left": "pull_request:1415", + "right": "pull_request:2529", + "accept": false, + "reason": "Stable Diffusion docs/code cleanup across many files is unrelated to the specific SD checkpoint conversion bug." + }, + { + "left": "pull_request:1468", + "right": "pull_request:1494", + "accept": false, + "reason": "Saving trained betas in schedulers is unrelated to adding safetensor serialization support." + }, + { + "left": "pull_request:1945", + "right": "pull_request:3766", + "accept": false, + "reason": "UNet block tests do not match a Kandinsky UNet conversion-script update." + }, + { + "left": "pull_request:1951", + "right": "pull_request:225", + "accept": false, + "reason": "ColossalAI dependency update is unrelated to README release-note edits." + }, + { + "left": "pull_request:1965", + "right": "pull_request:908", + "accept": false, + "reason": "Checkpoint merger cleanup is unrelated to swapping the inpaint pipeline in the community mega script." + }, + { + "left": "pull_request:2101", + "right": "pull_request:2734", + "accept": false, + "reason": "Pix2Pix docs/example update is unrelated to Habana documentation number updates." + }, + { + "left": "pull_request:2314", + "right": "pull_request:3614", + "accept": false, + "reason": "Both are test-related, but one adjusts numerical expectations and the other touches a renamed community script and determinism in a different test." + }, + { + "left": "pull_request:2470", + "right": "pull_request:4188", + "accept": false, + "reason": "MPS test skips and SDXL ControlNet dataset fingerprinting are different fixes in different areas." + }, + { + "left": "pull_request:2553", + "right": "pull_request:889", + "accept": false, + "reason": "CFG flag addition in Flax pipelines is unrelated to adding a components accessor for Stable Diffusion." + }, + { + "left": "pull_request:2637", + "right": "pull_request:3245", + "accept": false, + "reason": "Quicktour typo cleanup is unrelated to Chinese docs navigation/tree updates." + }, + { + "left": "pull_request:2637", + "right": "pull_request:3398", + "accept": false, + "reason": "Both are doc typos, but they fix different files and different wording; not the same underlying change." + }, + { + "left": "pull_request:2730", + "right": "pull_request:3245", + "accept": false, + "reason": "English training-doc newline/grammar cleanup is unrelated to the zh translated docs overhaul." + }, + { + "left": "pull_request:2730", + "right": "pull_request:3398", + "accept": false, + "reason": "These are separate documentation typos in different pages with no shared bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 176, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2915", + "pull_request:2919", + "pull_request:3003", + "pull_request:3037", + "pull_request:3047", + "pull_request:3241", + "pull_request:3370", + "pull_request:3536", + "pull_request:3568", + "pull_request:3675", + "pull_request:3691", + "pull_request:3850", + "pull_request:4058", + "pull_request:4211", + "pull_request:4319", + "pull_request:4564", + "pull_request:4666", + "pull_request:4749", + "pull_request:4787", + "pull_request:5030", + "pull_request:5053", + "pull_request:508", + "pull_request:51", + "pull_request:5122", + "pull_request:5157", + "pull_request:5275", + "pull_request:5393", + "pull_request:5396", + "pull_request:5506", + "pull_request:5579", + "pull_request:5626", + "pull_request:564", + "pull_request:5672", + "pull_request:5699", + "pull_request:5707", + "pull_request:6049", + "pull_request:6065", + "pull_request:6075", + "pull_request:6163", + "pull_request:6215", + "pull_request:6305", + "pull_request:6450", + "pull_request:6588", + "pull_request:6997", + "pull_request:7464", + "pull_request:748", + "pull_request:7679" + ], + "soft_pairs": [ + "pull_request:2915|pull_request:3691", + "pull_request:2915|pull_request:5053", + "pull_request:2919|pull_request:3037", + "pull_request:3003|pull_request:4058", + "pull_request:3047|pull_request:3241", + "pull_request:3370|pull_request:3568", + "pull_request:3536|pull_request:3675", + "pull_request:3850|pull_request:5396", + "pull_request:4211|pull_request:5699", + "pull_request:4319|pull_request:4564", + "pull_request:4666|pull_request:6450", + "pull_request:4749|pull_request:5393", + "pull_request:4787|pull_request:5157", + "pull_request:5030|pull_request:6075", + "pull_request:508|pull_request:564", + "pull_request:51|pull_request:748", + "pull_request:5122|pull_request:5579", + "pull_request:5275|pull_request:6305", + "pull_request:5506|pull_request:7679", + "pull_request:5626|pull_request:6163", + "pull_request:5672|pull_request:6215", + "pull_request:5707|pull_request:6997", + "pull_request:6049|pull_request:7464", + "pull_request:6065|pull_request:6588" + ], + "prepared_review_unit_hash": "8bc81a82cfc011c3d9b1aa3989de392b627ba97d59cfc8380cb2b778de9337d2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 93137, + "estimated_input_tokens": 23285, + "estimated_eval_tokens": 46826 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of mostly unrelated PRs that share only superficial signals like \u201cfix\u201d, tests, docs, or common subsystems such as SDXL/LoRA. There is no single underlying bug or change tying them together, so the soft duplicate edges should be rejected.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents a duplicate cluster here; the items span many unrelated fixes. If one must be chosen as the most substantive core-code change, PR 3003 is the best candidate because it fixes a concrete scheduler bug across multiple implementations and includes tests/docs, but it is not a duplicate anchor for the rest of the set.", + "best_issue_reason": null, + "best_pr_reason": "PR 3003 is the strongest standalone representative of a real code-path fix: it corrects dynamic thresholding behavior in several schedulers and adds coverage. It is broader and more canonical than the mostly test/doc/metadata-only PRs, though it still does not unify the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2915", + "right": "pull_request:3691", + "accept": false, + "reason": "One is a test threshold relaxation; the other changes LoRA text-encoder processor selection. Different code paths and different problems." + }, + { + "left": "pull_request:2915", + "right": "pull_request:5053", + "accept": false, + "reason": "A slow-test tweak and a validation_image argument fix are unrelated changes." + }, + { + "left": "pull_request:2919", + "right": "pull_request:3037", + "accept": false, + "reason": "Both touch tests, but one speeds up a UNet3D test and the other skips an unstable MPS test in Stable UnCLIP. Not the same bug." + }, + { + "left": "pull_request:3003", + "right": "pull_request:4058", + "accept": false, + "reason": "Scheduler dynamic-thresholding fixes are unrelated to SDXL documentation updates." + }, + { + "left": "pull_request:3047", + "right": "pull_request:3241", + "accept": false, + "reason": "Fixing report names in CI workflows is unrelated to loosening an UnCLIP test assertion." + }, + { + "left": "pull_request:3370", + "right": "pull_request:3568", + "accept": false, + "reason": "Both are docs PRs, but they fix different pages and different link/text issues." + }, + { + "left": "pull_request:3536", + "right": "pull_request:3675", + "accept": false, + "reason": "DeepFloyd IF PIL normalization and a UniDiffuser test skip are unrelated." + }, + { + "left": "pull_request:3850", + "right": "pull_request:5396", + "accept": false, + "reason": "Audio diffusion test adjustments and a timezone-aware stale-bot fix do not share an underlying bug." + }, + { + "left": "pull_request:4211", + "right": "pull_request:5699", + "accept": false, + "reason": "Docstring cleanup for SDXL inpainting is unrelated to style-only refactors in UnCLIP/Versatile Diffusion." + }, + { + "left": "pull_request:4319", + "right": "pull_request:4564", + "accept": false, + "reason": "Both mention SDXL, but one updates training-script licensing and the other fixes a ControlNet repo name in docs. Different changes." + }, + { + "left": "pull_request:4666", + "right": "pull_request:6450", + "accept": false, + "reason": "Enabling LoRA for SDXL ControlNets is a feature change; the other is LoRA test/deprecation cleanup. Not the same fix." + }, + { + "left": "pull_request:4749", + "right": "pull_request:5393", + "accept": false, + "reason": "SDXL DreamBooth LoRA checkpointing and HFApi updates are separate example/script changes." + }, + { + "left": "pull_request:4787", + "right": "pull_request:5157", + "accept": false, + "reason": "Both affect loaders.py and LoRA handling, but one tightens weight-file selection while the other fixes Kohya/ControlNet name conversion. Different concrete bugs." + }, + { + "left": "pull_request:5030", + "right": "pull_request:6075", + "accept": false, + "reason": "Model CPU offload key handling and backend cache clearing are unrelated runtime/test issues." + }, + { + "left": "pull_request:508", + "right": "pull_request:564", + "accept": false, + "reason": "Both involve schedulers, but one adds Flax scheduler implementations and the other updates Flax scheduler docs. Not the same change." + }, + { + "left": "pull_request:51", + "right": "pull_request:748", + "accept": false, + "reason": "Packaging dependency correction and push_to_hub example fixes are unrelated." + }, + { + "left": "pull_request:5122", + "right": "pull_request:5579", + "accept": false, + "reason": "Workflow Python-version changes are unrelated to bumping the transformers minimum version for PEFT integration." + }, + { + "left": "pull_request:5275", + "right": "pull_request:6305", + "accept": false, + "reason": "Adding an SDXL Flax example and adding zero-terminal-SNR support to DDPM are different tasks." + }, + { + "left": "pull_request:5506", + "right": "pull_request:7679", + "accept": false, + "reason": "PEFT/xformers attention handling and CI installation method changes are unrelated." + }, + { + "left": "pull_request:5626", + "right": "pull_request:6163", + "accept": false, + "reason": "CPU offload test fixes and LoRA test additions are not the same bug or change." + }, + { + "left": "pull_request:5672", + "right": "pull_request:6215", + "accept": false, + "reason": "Both touch PixArt tests, but one implements non-square image support and the other changes VAE test initialization/expected outputs. Not a duplicate pair." + }, + { + "left": "pull_request:5707", + "right": "pull_request:6997", + "accept": false, + "reason": "LCM test additions and UNetSpatioTemporalCondition LoRA support are unrelated." + }, + { + "left": "pull_request:6049", + "right": "pull_request:7464", + "accept": false, + "reason": "DemoFusion documentation/code adjustments and an LDM3D import fix are distinct changes." + }, + { + "left": "pull_request:6065", + "right": "pull_request:6588", + "accept": false, + "reason": "SDXL Turbo docs and SDXL k-diffusion test updates are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 177, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11067", + "pull_request:11522", + "pull_request:11809", + "pull_request:12119", + "pull_request:1229", + "pull_request:12416", + "pull_request:12445", + "pull_request:12530", + "pull_request:13169", + "pull_request:2034", + "pull_request:2339", + "pull_request:2959", + "pull_request:3313", + "pull_request:3859", + "pull_request:4278", + "pull_request:5057", + "pull_request:5261", + "pull_request:5579", + "pull_request:5807", + "pull_request:5823", + "pull_request:6229", + "pull_request:6389", + "pull_request:6764", + "pull_request:6772", + "pull_request:7189", + "pull_request:7328", + "pull_request:7359", + "pull_request:7442", + "pull_request:7653", + "pull_request:7771", + "pull_request:7778", + "pull_request:7825", + "pull_request:8386", + "pull_request:8627", + "pull_request:8660", + "pull_request:8825", + "pull_request:8856", + "pull_request:9065", + "pull_request:9242", + "pull_request:9307", + "pull_request:9311", + "pull_request:9442", + "pull_request:9446", + "pull_request:9592", + "pull_request:9593", + "pull_request:9705", + "pull_request:9718" + ], + "soft_pairs": [ + "pull_request:6229|pull_request:7771", + "pull_request:6229|pull_request:8386", + "pull_request:6389|pull_request:7653", + "pull_request:6764|pull_request:7778", + "pull_request:6772|pull_request:8825", + "pull_request:7328|pull_request:9065", + "pull_request:7359|pull_request:7442", + "pull_request:7825|pull_request:9442", + "pull_request:8627|pull_request:9307", + "pull_request:8660|pull_request:9592", + "pull_request:8660|pull_request:9593", + "pull_request:9242|pull_request:9718", + "pull_request:9446|pull_request:9705", + "pull_request:3313|pull_request:4278", + "pull_request:12445|pull_request:13169", + "pull_request:7189|pull_request:8856", + "pull_request:12119|pull_request:12416", + "pull_request:5261|pull_request:5807", + "pull_request:8856|pull_request:9311", + "pull_request:2034|pull_request:2959", + "pull_request:3859|pull_request:5057", + "pull_request:5579|pull_request:5823", + "pull_request:11067|pull_request:11522", + "pull_request:11809|pull_request:12530", + "pull_request:1229|pull_request:2339" + ], + "prepared_review_unit_hash": "0489b7eca5d77e6e79750c316b68ba476101daf1748845e9cda1b1cebaad1bef", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 102612, + "estimated_input_tokens": 25653, + "estimated_eval_tokens": 51562 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most pairs are only superficially related by broad subsystem, shared filenames, or both being tests/docs/CI. I don\u2019t see a single underlying bug or change that ties the set together, so no canonical duplicate anchor stands out.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No single PR is a strong canonical representative; the items span unrelated fixes across CI, docs, tests, and pipeline features rather than one mergeable change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6229", + "right": "pull_request:7771", + "accept": false, + "reason": "Unrelated: CI dependency install vs a flaky UNetControlNetXS test." + }, + { + "left": "pull_request:6229", + "right": "pull_request:8386", + "accept": false, + "reason": "Unrelated: CI dependency install vs broad test-shape refactors." + }, + { + "left": "pull_request:6389", + "right": "pull_request:7653", + "accept": false, + "reason": "Different problems: SVD API typing vs a configuration-utils decorator fix." + }, + { + "left": "pull_request:6764", + "right": "pull_request:7778", + "accept": false, + "reason": "Docs note about serialization is unrelated to a debugging workflow workflow file." + }, + { + "left": "pull_request:6772", + "right": "pull_request:8825", + "accept": false, + "reason": "Both mention ControlNet/IP-Adapter, but they fix different code paths and features, not the same bug." + }, + { + "left": "pull_request:7328", + "right": "pull_request:9065", + "accept": false, + "reason": "Unrelated dependency/workflow change vs adding sentencepiece as an optional dependency." + }, + { + "left": "pull_request:7359", + "right": "pull_request:7442", + "accept": false, + "reason": "Different community projects: ResAdapter training vs Differential Diffusion." + }, + { + "left": "pull_request:7825", + "right": "pull_request:9442", + "accept": false, + "reason": "Different test/CI issues; no shared failure mode." + }, + { + "left": "pull_request:8627", + "right": "pull_request:9307", + "accept": false, + "reason": "SD3 control test sizing vs rotary embedding refactor are unrelated." + }, + { + "left": "pull_request:8660", + "right": "pull_request:9592", + "accept": false, + "reason": "Pipeline input validation vs docstring cleanup; not the same change." + }, + { + "left": "pull_request:8660", + "right": "pull_request:9593", + "accept": false, + "reason": "Pipeline validation bug vs unrelated docs refactor." + }, + { + "left": "pull_request:9242", + "right": "pull_request:9718", + "accept": false, + "reason": "XPU test backend support is unrelated to a prompt-scheduling community-script update." + }, + { + "left": "pull_request:9446", + "right": "pull_request:9705", + "accept": false, + "reason": "TPU training script updates vs a README tweak for a Flux LoRA script." + }, + { + "left": "pull_request:3313", + "right": "pull_request:4278", + "accept": false, + "reason": "Same file, but different bugs: torch.compile break fixes vs prompt=None handling in upscale." + }, + { + "left": "pull_request:12445", + "right": "pull_request:13169", + "accept": false, + "reason": "Different Flux-related changes: modular alignment vs LoRA format conversion." + }, + { + "left": "pull_request:7189", + "right": "pull_request:8856", + "accept": false, + "reason": "Both touch img2img helpers, but one is IP-Adapter embed repetition and the other is batch/image duplication handling." + }, + { + "left": "pull_request:12119", + "right": "pull_request:12416", + "accept": false, + "reason": "QwenImage ecosystem overlap only; one adds Lightning LoRA support, the other adds QwenImage Edit Plus modular support." + }, + { + "left": "pull_request:5261", + "right": "pull_request:5807", + "accept": false, + "reason": "Callback-step fix vs prompt-embedding duplication for lpw SDXL; different issues." + }, + { + "left": "pull_request:8856", + "right": "pull_request:9311", + "accept": false, + "reason": "Image batch duplication bug vs SDXL ControlNet CFG cutoff callback; unrelated." + }, + { + "left": "pull_request:2034", + "right": "pull_request:2959", + "accept": false, + "reason": "Different model test fixes with no shared underlying defect." + }, + { + "left": "pull_request:3859", + "right": "pull_request:5057", + "accept": false, + "reason": "Adding SDXL pipelines is not the same as adding clip-skip support; too broad to merge as duplicates." + }, + { + "left": "pull_request:5579", + "right": "pull_request:5823", + "accept": false, + "reason": "Transformers minimum-version bump vs LoRA adapter fusion API; unrelated changes." + }, + { + "left": "pull_request:11067", + "right": "pull_request:11522", + "accept": false, + "reason": "Benchmark version pinning vs Audioldm2 transformers compatibility fix are separate CI/runtime issues." + }, + { + "left": "pull_request:11809", + "right": "pull_request:12530", + "accept": false, + "reason": "Different test adjustments for different failures; no common bug." + }, + { + "left": "pull_request:1229", + "right": "pull_request:2339", + "accept": false, + "reason": "Both DreamBooth docs, but one is install guidance and the other is checkpointing behavior/tests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 178, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10245", + "pull_request:10482", + "pull_request:11066", + "pull_request:11556", + "pull_request:11823", + "pull_request:11896", + "pull_request:12805", + "pull_request:12960", + "pull_request:13211", + "pull_request:13330", + "pull_request:13352", + "pull_request:13355", + "pull_request:153", + "pull_request:1901", + "pull_request:2037", + "pull_request:2295", + "pull_request:2327", + "pull_request:243", + "pull_request:274", + "pull_request:3189", + "pull_request:3321", + "pull_request:3632", + "pull_request:3707", + "pull_request:3795", + "pull_request:3852", + "pull_request:3859", + "pull_request:4397", + "pull_request:4595", + "pull_request:4695", + "pull_request:4715", + "pull_request:4749", + "pull_request:4761", + "pull_request:4869", + "pull_request:4887", + "pull_request:5003", + "pull_request:5057", + "pull_request:6099", + "pull_request:6208", + "pull_request:6225", + "pull_request:6405", + "pull_request:6772", + "pull_request:6965", + "pull_request:6972", + "pull_request:7158", + "pull_request:7712", + "pull_request:7720", + "pull_request:7932", + "pull_request:8282" + ], + "soft_pairs": [ + "pull_request:12960|pull_request:13211", + "pull_request:13330|pull_request:13355", + "pull_request:153|pull_request:2037", + "pull_request:2295|pull_request:2327", + "pull_request:4749|pull_request:6099", + "pull_request:4761|pull_request:6972", + "pull_request:4887|pull_request:6208", + "pull_request:5003|pull_request:7158", + "pull_request:6225|pull_request:7932", + "pull_request:6772|pull_request:8282", + "pull_request:7712|pull_request:7720", + "pull_request:11823|pull_request:11896", + "pull_request:4715|pull_request:6965", + "pull_request:3859|pull_request:4695", + "pull_request:10245|pull_request:11556", + "pull_request:10482|pull_request:11066", + "pull_request:12805|pull_request:13352", + "pull_request:1901|pull_request:3852", + "pull_request:243|pull_request:274", + "pull_request:3189|pull_request:4397", + "pull_request:3321|pull_request:3707", + "pull_request:3632|pull_request:5057", + "pull_request:3795|pull_request:4869", + "pull_request:4595|pull_request:6405" + ], + "prepared_review_unit_hash": "755d6b604cf26cdee5c5be12695a035e44a4b315f4224bd9e04d77ab21d32a86", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 96622, + "estimated_input_tokens": 24156, + "estimated_eval_tokens": 48568 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: most pairs are superficial similarity matches across unrelated docs, CI, example, and model changes. None of the proposed soft edges look like true duplicates or the same concrete change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits this cluster; the items span unrelated subsystems and change types, so the similarity links are spurious rather than duplicate PRs.", + "best_issue_reason": null, + "best_pr_reason": "PR 13211 is the strongest standalone bug fix in the set because it targets a concrete runtime offloading issue and includes a focused test, but it still does not serve as a duplicate anchor for the rest of the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12960", + "right": "pull_request:13211", + "accept": false, + "reason": "Unrelated fixes: one determines FSDP handling in DreamBooth examples, the other patches group offloading prefetch behavior. Different code paths and failure modes." + }, + { + "left": "pull_request:13330", + "right": "pull_request:13355", + "accept": false, + "reason": "Both touch torchao, but one is a test/workflow tweak and the other is a version guard bump. Not the same bug or mergeable change." + }, + { + "left": "pull_request:153", + "right": "pull_request:2037", + "accept": false, + "reason": "Issue templates vs a community pipeline addition; no shared underlying change." + }, + { + "left": "pull_request:2295", + "right": "pull_request:2327", + "accept": false, + "reason": "Both concern UNet internals, but one adds configurability and the other adds input validation. Different changes, not duplicates." + }, + { + "left": "pull_request:4749", + "right": "pull_request:6099", + "accept": false, + "reason": "DreamBooth SDXL checkpointing and TextToVideoZero superclass argument fixes are unrelated code paths." + }, + { + "left": "pull_request:4761", + "right": "pull_request:6972", + "accept": false, + "reason": "Paint-by-example encoder initialization and IPAdapter unload test behavior are different issues." + }, + { + "left": "pull_request:4887", + "right": "pull_request:6208", + "accept": false, + "reason": "Kandinsky prior_emb2emb API cleanup vs Stable Diffusion half/bias dtype runtime fix; unrelated." + }, + { + "left": "pull_request:5003", + "right": "pull_request:7158", + "accept": false, + "reason": "LoRA fuse/unfuse logic and an LCM benchmark cleanup are not the same bug or change." + }, + { + "left": "pull_request:6225", + "right": "pull_request:7932", + "accept": false, + "reason": "Both are LoRA-related, but one fixes alpha parsing and the other adjusts test expectations/side effects. Not mergeable as one PR." + }, + { + "left": "pull_request:6772", + "right": "pull_request:8282", + "accept": false, + "reason": "ControlNet-XS implementation work and I2VGenXL docs are completely different artifacts." + }, + { + "left": "pull_request:7712", + "right": "pull_request:7720", + "accept": false, + "reason": "Import fixes and dead-code removal are separate maintenance changes." + }, + { + "left": "pull_request:11823", + "right": "pull_request:11896", + "accept": false, + "reason": "Both involve LoRA infrastructure, but one adds Fal Kontext conversion support while the other fixes hook disabling. Different concrete problems." + }, + { + "left": "pull_request:4715", + "right": "pull_request:6965", + "accept": false, + "reason": "File renames for example test discovery and moving an SDXL LoRA test into PEFT workflow are unrelated." + }, + { + "left": "pull_request:3859", + "right": "pull_request:4695", + "accept": false, + "reason": "Both are pipeline additions, but they implement different pipelines and code paths; too broad to treat as duplicates." + }, + { + "left": "pull_request:10245", + "right": "pull_request:11556", + "accept": false, + "reason": "PEFT inference docs and regional compilation docs are unrelated documentation updates." + }, + { + "left": "pull_request:10482", + "right": "pull_request:11066", + "accept": false, + "reason": "Same HunyuanVideo file, but one fixes NaNs in attention masking while the other adds a new I2V model variant. Different underlying changes." + }, + { + "left": "pull_request:12805", + "right": "pull_request:13352", + "accept": false, + "reason": "PyPI publish workflow changes and Claude review workflow checkout are unrelated CI edits." + }, + { + "left": "pull_request:1901", + "right": "pull_request:3852", + "accept": false, + "reason": "Both are docs-only, but they address different documentation pages and issues." + }, + { + "left": "pull_request:243", + "right": "pull_request:274", + "accept": false, + "reason": "RePaint pipeline WIP vs Stable Diffusion README addition; no shared bug/change." + }, + { + "left": "pull_request:3189", + "right": "pull_request:4397", + "accept": false, + "reason": "AudioLDM waveform dtype fix and SDXL docstring correction are unrelated." + }, + { + "left": "pull_request:3321", + "right": "pull_request:3707", + "accept": false, + "reason": "Generation docs update and doc-build secrets/workflow changes are separate concerns." + }, + { + "left": "pull_request:3632", + "right": "pull_request:5057", + "accept": false, + "reason": "Panorama view batching and SDXL clip-skip support are different features." + }, + { + "left": "pull_request:3795", + "right": "pull_request:4869", + "accept": false, + "reason": "Overview-page docs updates and combined-pipeline tests are not the same change." + }, + { + "left": "pull_request:4595", + "right": "pull_request:6405", + "accept": false, + "reason": "Different doc-builder/testing work; no concrete shared bug or feature." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 179, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10069", + "pull_request:10119", + "pull_request:10148", + "pull_request:10187", + "pull_request:10188", + "pull_request:10230", + "pull_request:10358", + "pull_request:10370", + "pull_request:10544", + "pull_request:10630", + "pull_request:10970", + "pull_request:11081", + "pull_request:11130", + "pull_request:11196", + "pull_request:11231", + "pull_request:11469", + "pull_request:11869", + "pull_request:11995", + "pull_request:12226", + "pull_request:12330", + "pull_request:12534", + "pull_request:1355", + "pull_request:2727", + "pull_request:4899", + "pull_request:5842", + "pull_request:5880", + "pull_request:5887", + "pull_request:5922", + "pull_request:6217", + "pull_request:628", + "pull_request:6578", + "pull_request:6646", + "pull_request:670", + "pull_request:7034", + "pull_request:7247", + "pull_request:7919", + "pull_request:7924", + "pull_request:8229", + "pull_request:8480", + "pull_request:8496", + "pull_request:8691", + "pull_request:8696", + "pull_request:8702", + "pull_request:8765", + "pull_request:8856", + "pull_request:9105", + "pull_request:9736", + "pull_request:9862" + ], + "soft_pairs": [ + "pull_request:5842|pull_request:7924", + "pull_request:5880|pull_request:6217", + "pull_request:5922|pull_request:7924", + "pull_request:628|pull_request:670", + "pull_request:6578|pull_request:8480", + "pull_request:7034|pull_request:8765", + "pull_request:7247|pull_request:8229", + "pull_request:7919|pull_request:9736", + "pull_request:8496|pull_request:8702", + "pull_request:11081|pull_request:11130", + "pull_request:11231|pull_request:11995", + "pull_request:11869|pull_request:12534", + "pull_request:6646|pull_request:8856", + "pull_request:10630|pull_request:11469", + "pull_request:12226|pull_request:12330", + "pull_request:1355|pull_request:2727", + "pull_request:4899|pull_request:5887", + "pull_request:10187|pull_request:9862", + "pull_request:10069|pull_request:11196", + "pull_request:10119|pull_request:9105", + "pull_request:10148|pull_request:8696", + "pull_request:10188|pull_request:11196", + "pull_request:10230|pull_request:8691", + "pull_request:10358|pull_request:10544", + "pull_request:10370|pull_request:10970" + ], + "prepared_review_unit_hash": "33ea2c9a8ff00389d839eaf0cad48ef67f20535bb78b54038a8d027f288f3752", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 103578, + "estimated_input_tokens": 25895, + "estimated_eval_tokens": 52046 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous set of mostly unrelated PRs (docs updates, new model/pipeline additions, LoRA/quantization changes, and isolated bug fixes). The soft edges are mostly surface-level similarity and do not indicate duplicate underlying changes.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:5842", + "right": "pull_request:7924", + "accept": false, + "reason": "Both touch image-generation pipeline behavior, but one fixes PixArt 512px resolution mapping and the other fixes IP-Adapter handling in SDXL inpaint; different models, code paths, and bugs." + }, + { + "left": "pull_request:5880", + "right": "pull_request:6217", + "accept": false, + "reason": "Both involve Stable Video Diffusion/VAE-related code, but 5880 is a broad VAE compatibility/config change while 6217 is a torch.compile-specific VAE forward introspection fix; not the same bug." + }, + { + "left": "pull_request:5922", + "right": "pull_request:7924", + "accept": false, + "reason": "Both mention adapter/inpaint flows, but 5922 fixes MultiAdapter image preprocessing in a community example and 7924 fixes added_cond_kwargs for IP-Adapter in the SDXL ControlNet pipeline; distinct changes." + }, + { + "left": "pull_request:628", + "right": "pull_request:670", + "accept": false, + "reason": "These are unrelated documentation index tweaks from early repo history; they do not fix the same underlying issue or change." + }, + { + "left": "pull_request:6578", + "right": "pull_request:8480", + "accept": false, + "reason": "One is a docs heading/indent fix for ControlNet; the other adds HunyuanDiT ControlNet support. Same broad area, but not the same change." + }, + { + "left": "pull_request:7034", + "right": "pull_request:8765", + "accept": false, + "reason": "7034 fixes `load_model_dict_into_meta` usage in single-file ControlNet loading, while 8765 corrects a docs API argument typo; unrelated." + }, + { + "left": "pull_request:7247", + "right": "pull_request:8229", + "accept": false, + "reason": "7247 fixes config/type checking in single-file loading tests, while 8229 changes Stable Video Diffusion resizing behavior. Different defects and code paths." + }, + { + "left": "pull_request:7919", + "right": "pull_request:9736", + "accept": false, + "reason": "One fixes offload behavior with device maps; the other adds the Allegro text-to-video model/pipeline. No shared underlying bug or change." + }, + { + "left": "pull_request:8496", + "right": "pull_request:8702", + "accept": false, + "reason": "Both relate to SD3, but 8496 only narrows a warning log condition in the transformer, while 8702 adds a conversion script. Not mergeable as one PR." + }, + { + "left": "pull_request:11081", + "right": "pull_request:11130", + "accept": false, + "reason": "11081 implements `record_stream` for group offloading; 11130 introduces pipeline-level quantization config and tests. Different features." + }, + { + "left": "pull_request:11231", + "right": "pull_request:11995", + "accept": false, + "reason": "Both add large model families, but HiDream Image and modular Flux are separate model integrations with unrelated implementation work." + }, + { + "left": "pull_request:11869", + "right": "pull_request:12534", + "accept": false, + "reason": "11869 adds CUDA kernel support for GGUF inference, while 12534 adjusts Flux pos-embedding behavior on Ascend NPU. Different hardware targets and fixes." + }, + { + "left": "pull_request:6646", + "right": "pull_request:8856", + "accept": false, + "reason": "Both concern image/video resolution or img2img handling, but 6646 changes SVD resolution constraints and 8856 fixes batch duplication error handling across img2img pipelines; not the same bug." + }, + { + "left": "pull_request:10630", + "right": "pull_request:11469", + "accept": false, + "reason": "Both are test-only changes, but they address different failing tests on different pipelines and PyTorch versions." + }, + { + "left": "pull_request:12226", + "right": "pull_request:12330", + "accept": false, + "reason": "One adds a fast Qwen-Image ControlNet test; the other cleans up unrelated group offloading tests. No duplicate change." + }, + { + "left": "pull_request:1355", + "right": "pull_request:2727", + "accept": false, + "reason": "Both are Flax pipeline additions, but one adds Stable Diffusion img2img and the other adds Flax ControlNet; separate features, not duplicates." + }, + { + "left": "pull_request:4899", + "right": "pull_request:5887", + "accept": false, + "reason": "4899 is a work-in-progress ControlNet LoRA support change, while 5887 adds IP-Adapter support to ControlNet inpaint. Different loader paths and goals." + }, + { + "left": "pull_request:10187", + "right": "pull_request:9862", + "accept": false, + "reason": "Both are LoRA-related, but 10187 improves no-op warnings during loading, whereas 9862 adds `save_lora_adapter()`. Different functionality." + }, + { + "left": "pull_request:10069", + "right": "pull_request:11196", + "accept": false, + "reason": "10069 adds device-type validation in model loading; 11196 reverts `save_model` usage in `save_pretrained`. Distinct model serialization/loading issues." + }, + { + "left": "pull_request:10119", + "right": "pull_request:9105", + "accept": false, + "reason": "10119 is docs for `load_lora_adapter`; 9105 is a docs deprecation cleanup. No shared underlying change." + }, + { + "left": "pull_request:10148", + "right": "pull_request:8696", + "accept": false, + "reason": "An OmniGen feature addition and a Marigold test-flakiness annotation are unrelated." + }, + { + "left": "pull_request:10188", + "right": "pull_request:11196", + "accept": false, + "reason": "10188 fixes Flux LoRA expansion/no-op loading behavior, while 11196 changes safe serialization behavior in model saving. Different code paths." + }, + { + "left": "pull_request:10230", + "right": "pull_request:8691", + "accept": false, + "reason": "10230 updates LoRA loader docs; 8691 fixes vanilla fine-tuned LoRA loading. One is documentation only, the other is runtime behavior." + }, + { + "left": "pull_request:10358", + "right": "pull_request:10544", + "accept": false, + "reason": "Both are documentation additions, but one adds a community project entry and the other adds ParaAttention optimization docs; not the same artifact." + }, + { + "left": "pull_request:10370", + "right": "pull_request:10970", + "accept": false, + "reason": "10370 is a formatting/style change in torchao quantizer code; 10970 swaps in a reusable style-bot GitHub Action workflow. Different changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 180, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10506", + "pull_request:10579", + "pull_request:11001", + "pull_request:11113", + "pull_request:11186", + "pull_request:11743", + "pull_request:11843", + "pull_request:12062", + "pull_request:12363", + "pull_request:12397", + "pull_request:12493", + "pull_request:12642", + "pull_request:13100", + "pull_request:1331", + "pull_request:1494", + "pull_request:153", + "pull_request:1945", + "pull_request:2009", + "pull_request:2122", + "pull_request:2279", + "pull_request:274", + "pull_request:2793", + "pull_request:2827", + "pull_request:2861", + "pull_request:2928", + "pull_request:2959", + "pull_request:3023", + "pull_request:3127", + "pull_request:3208", + "pull_request:3275", + "pull_request:3297", + "pull_request:335", + "pull_request:3934", + "pull_request:4276", + "pull_request:4595", + "pull_request:4596", + "pull_request:4655", + "pull_request:4887", + "pull_request:4963", + "pull_request:5053", + "pull_request:5111", + "pull_request:5615", + "pull_request:5697", + "pull_request:5853", + "pull_request:6114", + "pull_request:8981", + "pull_request:9307" + ], + "soft_pairs": [ + "pull_request:10506|pull_request:9307", + "pull_request:10579|pull_request:8981", + "pull_request:11001|pull_request:11186", + "pull_request:11113|pull_request:11743", + "pull_request:11843|pull_request:12493", + "pull_request:12062|pull_request:12397", + "pull_request:12363|pull_request:13100", + "pull_request:12642|pull_request:13100", + "pull_request:1331|pull_request:335", + "pull_request:1494|pull_request:2122", + "pull_request:153|pull_request:1945", + "pull_request:2009|pull_request:274", + "pull_request:2009|pull_request:3275", + "pull_request:2279|pull_request:2827", + "pull_request:2793|pull_request:2928", + "pull_request:2793|pull_request:4887", + "pull_request:2861|pull_request:3208", + "pull_request:2959|pull_request:5053", + "pull_request:3023|pull_request:3127", + "pull_request:3275|pull_request:5111", + "pull_request:3297|pull_request:4595", + "pull_request:3934|pull_request:4963", + "pull_request:3934|pull_request:5615", + "pull_request:4276|pull_request:5853", + "pull_request:4596|pull_request:6114", + "pull_request:4655|pull_request:5697" + ], + "prepared_review_unit_hash": "52a9285fd84d27b333321d2e5dd8ef98f16f58df1a8edf80c3f52fdf8a41680d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 100689, + "estimated_input_tokens": 25173, + "estimated_eval_tokens": 50602 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: the soft edges mostly connect unrelated PRs that only share broad themes like tests, CI, docs, or pipeline work. I rejected all pairings.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a valid canonical duplicate anchor here; the items span unrelated changes across tests, docs, CI, training scripts, and core pipeline/model code.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR for deduplication: none of these PRs subsumes the others or represents the same underlying change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10506", + "right": "pull_request:9307", + "accept": false, + "reason": "Unrelated changes: a torchao offload test vs a rotary-embedding refactor." + }, + { + "left": "pull_request:10579", + "right": "pull_request:8981", + "accept": false, + "reason": "Different topics: an XL attentive eraser community pipeline vs LoRA loader mixin exports/docs." + }, + { + "left": "pull_request:11001", + "right": "pull_request:11186", + "accept": false, + "reason": "Both are device-portability related, but one enables bnb tests on xpu while the other only removes a hard-coded cuda call in one test." + }, + { + "left": "pull_request:11113", + "right": "pull_request:11743", + "accept": false, + "reason": "Different training scripts and goals; one fixes AutoencoderKL training, the other adds a quantized Flux DreamBooth script." + }, + { + "left": "pull_request:11843", + "right": "pull_request:12493", + "accept": false, + "reason": "A Wan VACE LoRA test is unrelated to LTX-Video 0.9.8 pipeline/docs changes." + }, + { + "left": "pull_request:12062", + "right": "pull_request:12397", + "accept": false, + "reason": "Both touch CI workflows, but one changes GPU allocation and the other stops installing transformers/accelerate from main; not the same fix." + }, + { + "left": "pull_request:12363", + "right": "pull_request:13100", + "accept": false, + "reason": "QwenImage Edit Plus docs/tests are unrelated to modular pipeline state documentation." + }, + { + "left": "pull_request:12642", + "right": "pull_request:13100", + "accept": false, + "reason": "Docs/toctree maintenance is not the same change as modular pipeline state docs and code updates." + }, + { + "left": "pull_request:1331", + "right": "pull_request:335", + "accept": false, + "reason": "Different code paths: inpainting scheduler input ordering vs PNDM type hints." + }, + { + "left": "pull_request:1494", + "right": "pull_request:2122", + "accept": false, + "reason": "Safetensors saving support and safetensors docs are related only by topic, not the same underlying change." + }, + { + "left": "pull_request:153", + "right": "pull_request:1945", + "accept": false, + "reason": "Issue templates are unrelated to 2D UNet block tests." + }, + { + "left": "pull_request:2009", + "right": "pull_request:274", + "accept": false, + "reason": "Textual inversion loader integration is unrelated to a Stable Diffusion README addition." + }, + { + "left": "pull_request:2009", + "right": "pull_request:3275", + "accept": false, + "reason": "Textual inversion loader support and upsample_size support in UNet blocks are different features." + }, + { + "left": "pull_request:2279", + "right": "pull_request:2827", + "accept": false, + "reason": "Both concern lint/style tooling, but these are different repo-wide hygiene changes, not one concrete bug or change." + }, + { + "left": "pull_request:2793", + "right": "pull_request:2928", + "accept": false, + "reason": "A docs removal about cuDNN benchmark is unrelated to allowing custom local datasets in ControlNet training." + }, + { + "left": "pull_request:2793", + "right": "pull_request:4887", + "accept": false, + "reason": "No shared code-path problem: docs cleanup vs removing latents from Kandinsky prior_emb2emb." + }, + { + "left": "pull_request:2861", + "right": "pull_request:3208", + "accept": false, + "reason": "A Stable UnCLIP test for image_embeds=None is unrelated to Token Merging docs." + }, + { + "left": "pull_request:2959", + "right": "pull_request:5053", + "accept": false, + "reason": "A UNet3D norm-group test fix is unrelated to the validation_image CLI argument change." + }, + { + "left": "pull_request:3023", + "right": "pull_request:3127", + "accept": false, + "reason": "Different subsystems: attention processor/KV handling vs safety checker image masking." + }, + { + "left": "pull_request:3275", + "right": "pull_request:5111", + "accept": false, + "reason": "UNet block upsample_size plumbing and Karras sigma scheduler support are unrelated." + }, + { + "left": "pull_request:3297", + "right": "pull_request:4595", + "accept": false, + "reason": "A retrieval-augmented diffusion model project is not the same as an SDXL-to-original conversion script." + }, + { + "left": "pull_request:3934", + "right": "pull_request:4963", + "accept": false, + "reason": "These are unrelated reverts: automatic chunking vs accelerate-main install changes." + }, + { + "left": "pull_request:3934", + "right": "pull_request:5615", + "accept": false, + "reason": "Different revert scopes: text-to-video chunking behavior vs SDXL adapter LoRA loading." + }, + { + "left": "pull_request:4276", + "right": "pull_request:5853", + "accept": false, + "reason": "A pipeline import smoke test and doctests workflow are different test infrastructure changes." + }, + { + "left": "pull_request:4596", + "right": "pull_request:6114", + "accept": false, + "reason": "A conversion script for SDXL-to-original checkpoints is unrelated to lpw-xl img2img/inpaint support." + }, + { + "left": "pull_request:4655", + "right": "pull_request:5697", + "accept": false, + "reason": "Adding data_dir to load_dataset is unrelated to the PEFT LoRA xformers bugfix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 181, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10105", + "pull_request:1269", + "pull_request:13294", + "pull_request:13365", + "pull_request:1470", + "pull_request:1966", + "pull_request:4365", + "pull_request:4749", + "pull_request:4793", + "pull_request:4862", + "pull_request:5038", + "pull_request:5085", + "pull_request:5204", + "pull_request:530", + "pull_request:5393", + "pull_request:5697", + "pull_request:5771", + "pull_request:5833", + "pull_request:5834", + "pull_request:594", + "pull_request:6237", + "pull_request:6261", + "pull_request:6426", + "pull_request:6633", + "pull_request:6752", + "pull_request:6864", + "pull_request:6929", + "pull_request:7044", + "pull_request:7116", + "pull_request:7153", + "pull_request:7216", + "pull_request:7247", + "pull_request:7295", + "pull_request:7339", + "pull_request:748", + "pull_request:7913", + "pull_request:8267", + "pull_request:8564", + "pull_request:8591", + "pull_request:8618", + "pull_request:8627", + "pull_request:8631", + "pull_request:8739", + "pull_request:9103", + "pull_request:9400", + "pull_request:9447", + "pull_request:9536", + "pull_request:9735" + ], + "soft_pairs": [ + "pull_request:530|pull_request:748", + "pull_request:5393|pull_request:7216", + "pull_request:5771|pull_request:6261", + "pull_request:5833|pull_request:7339", + "pull_request:5834|pull_request:7339", + "pull_request:6237|pull_request:6864", + "pull_request:6426|pull_request:8267", + "pull_request:6633|pull_request:7247", + "pull_request:7044|pull_request:7295", + "pull_request:7116|pull_request:7153", + "pull_request:7913|pull_request:9447", + "pull_request:8564|pull_request:8631", + "pull_request:8591|pull_request:8618", + "pull_request:8591|pull_request:8627", + "pull_request:8618|pull_request:9735", + "pull_request:8627|pull_request:9735", + "pull_request:9400|pull_request:9536", + "pull_request:1966|pull_request:594", + "pull_request:5038|pull_request:5204", + "pull_request:6752|pull_request:8739", + "pull_request:10105|pull_request:9103", + "pull_request:1269|pull_request:1470", + "pull_request:13294|pull_request:13365", + "pull_request:4365|pull_request:4749", + "pull_request:4793|pull_request:4862", + "pull_request:5085|pull_request:5697", + "pull_request:6237|pull_request:6929" + ], + "prepared_review_unit_hash": "a4735a384a5cd3ecf21fdf35f8b6bb4eccba665f8cd1fa48a98ee15976f77182", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 102667, + "estimated_input_tokens": 25667, + "estimated_eval_tokens": 51590 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is a grab bag of unrelated diffusers PRs spanning docs, CI, LoRA, single-file loading, SD3, and examples. None of the soft-edge pairs look like the same underlying bug or change, so I would not merge any of them as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a true duplicate of the others. If a representative PR is needed, #5771 is the broadest core-code change in the set (large loaders/LoRA refactor across many files), so it is the best stand-in for the cluster.", + "best_issue_reason": null, + "best_pr_reason": "#5771 is the strongest representative because it is a substantial, cross-cutting code change in the core loader/LoRA path and is more central than the many narrow docs/tests/CI fixes in this group.", + "soft_edge_verdicts": [ + { + "left": "pull_request:530", + "right": "pull_request:748", + "accept": false, + "reason": "Different fixes in different areas: img2img test adjustment vs push_to_hub call cleanup for training scripts." + }, + { + "left": "pull_request:5393", + "right": "pull_request:7216", + "accept": false, + "reason": "Both touch docs/examples around hub/file handling, but one updates API usage while the other changes the from_single_file signature and loader behavior." + }, + { + "left": "pull_request:5771", + "right": "pull_request:6261", + "accept": false, + "reason": "Unrelated changes: a broad loaders/LoRA refactor vs moving Attend-and-Excite into its own module." + }, + { + "left": "pull_request:5833", + "right": "pull_request:7339", + "accept": false, + "reason": "Single-file ControlNet import fix and release-workflow dependency changes are unrelated." + }, + { + "left": "pull_request:5834", + "right": "pull_request:7339", + "accept": false, + "reason": "Single-file import path correction is unrelated to release workflow dependency setup." + }, + { + "left": "pull_request:6237", + "right": "pull_request:6864", + "accept": false, + "reason": "Docs-only changes about seeds vs LoRA/torch.compile guidance; different concerns and code paths." + }, + { + "left": "pull_request:6426", + "right": "pull_request:8267", + "accept": false, + "reason": "Both mention deprecation, but one is a broad LoRA deprecation cleanup and the other is specific resume_download handling in single-file loaders." + }, + { + "left": "pull_request:6633", + "right": "pull_request:7247", + "accept": false, + "reason": "One fixes checkpoint config file reading; the other changes single-file config-shape tests. Not the same bug." + }, + { + "left": "pull_request:7044", + "right": "pull_request:7295", + "accept": false, + "reason": "Different loader/refactor changes: restoring a UNet refactor vs cleaning copy-paste residue in the UNet single-file loader." + }, + { + "left": "pull_request:7116", + "right": "pull_request:7153", + "accept": false, + "reason": "CI workflow install tooling vs documentation path filtering; not the same underlying change." + }, + { + "left": "pull_request:7913", + "right": "pull_request:9447", + "accept": false, + "reason": "ControlNet SDXL timestep/sigma support is unrelated to pipeline save_pretrained sharding defaults." + }, + { + "left": "pull_request:8564", + "right": "pull_request:8631", + "accept": false, + "reason": "SD3 training docs/code for T5 sequence length vs a single-file loader fix for T5 checkpoints; same model family, different problems." + }, + { + "left": "pull_request:8591", + "right": "pull_request:8618", + "accept": false, + "reason": "SD3 training utility refactor vs adding shift_factor to the tiny autoencoder; different code paths and purposes." + }, + { + "left": "pull_request:8591", + "right": "pull_request:8627", + "accept": false, + "reason": "Training-utility refactor is unrelated to adding SD3 ControlNet docs/models/tests." + }, + { + "left": "pull_request:8618", + "right": "pull_request:9735", + "accept": false, + "reason": "Tiny autoencoder shift_factor addition and SD3 ControlNet example update are separate features." + }, + { + "left": "pull_request:8627", + "right": "pull_request:9735", + "accept": false, + "reason": "SD3 ControlNet model/docs addition is broader and different from updating a training example." + }, + { + "left": "pull_request:9400", + "right": "pull_request:9536", + "accept": false, + "reason": "Device-agnostic test updates and ChatGLMTokenizer padding signature fixes are unrelated." + }, + { + "left": "pull_request:1966", + "right": "pull_request:594", + "accept": false, + "reason": "Flax inpainting implementation vs Flax UNet/DDIM embedding/scheduler fix are different feature areas." + }, + { + "left": "pull_request:5038", + "right": "pull_request:5204", + "accept": false, + "reason": "LoRA-related, but one makes optional args explicit in Transformer2D while the other fixes text encoder scaling across many pipelines." + }, + { + "left": "pull_request:6752", + "right": "pull_request:8739", + "accept": false, + "reason": "Distinct community pipeline additions with different pipelines and goals." + }, + { + "left": "pull_request:10105", + "right": "pull_request:9103", + "accept": false, + "reason": "Broken doc link fix vs distributed LR scheduler logic correction; unrelated." + }, + { + "left": "pull_request:1269", + "right": "pull_request:1470", + "accept": false, + "reason": "Stable-diffusion refactor finalization vs DreamBooth AltDiffusion compatibility are separate changes." + }, + { + "left": "pull_request:13294", + "right": "pull_request:13365", + "accept": false, + "reason": "Flaky WanVACE test cleanup vs fork-support CI workflow changes; no shared bug/change." + }, + { + "left": "pull_request:4365", + "right": "pull_request:4749", + "accept": false, + "reason": "Both mention SDXL LoRA, but one adds a new training example while the other fixes checkpoint saving/loading for an existing script." + }, + { + "left": "pull_request:4793", + "right": "pull_request:4862", + "accept": false, + "reason": "Auto-pipeline kwargs forwarding and ControlNet SDXL MultiControlNet height/width handling are unrelated bugs." + }, + { + "left": "pull_request:5085", + "right": "pull_request:5697", + "accept": false, + "reason": "Different LoRA issues: offloading incompatible checkpoints vs xformers-enabled PEFT scaling." + }, + { + "left": "pull_request:6237", + "right": "pull_request:6929", + "accept": false, + "reason": "Seed batching docs and community pipeline docs are different documentation topics with no shared code change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 182, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10042", + "pull_request:10114", + "pull_request:10115", + "pull_request:10132", + "pull_request:10148", + "pull_request:10245", + "pull_request:10299", + "pull_request:10312", + "pull_request:10318", + "pull_request:10322", + "pull_request:10326", + "pull_request:10420", + "pull_request:10430", + "pull_request:10585", + "pull_request:10648", + "pull_request:10784", + "pull_request:10810", + "pull_request:10837", + "pull_request:10894", + "pull_request:11080", + "pull_request:11319", + "pull_request:11366", + "pull_request:11428", + "pull_request:11639", + "pull_request:11647", + "pull_request:11778", + "pull_request:11997", + "pull_request:12024", + "pull_request:12909", + "pull_request:2954", + "pull_request:3125", + "pull_request:4780", + "pull_request:5038", + "pull_request:5202", + "pull_request:5823", + "pull_request:7096", + "pull_request:7906", + "pull_request:8618", + "pull_request:8684", + "pull_request:8685", + "pull_request:8985", + "pull_request:9015", + "pull_request:9107", + "pull_request:9765", + "pull_request:9769", + "pull_request:9807", + "pull_request:9832", + "pull_request:9934" + ], + "soft_pairs": [ + "pull_request:2954|pull_request:3125", + "pull_request:5038|pull_request:5823", + "pull_request:7096|pull_request:7906", + "pull_request:4780|pull_request:5202", + "pull_request:10837|pull_request:11428", + "pull_request:11639|pull_request:11997", + "pull_request:12024|pull_request:12909", + "pull_request:10042|pull_request:9832", + "pull_request:10114|pull_request:10894", + "pull_request:10115|pull_request:9107", + "pull_request:10132|pull_request:11366", + "pull_request:10148|pull_request:11319", + "pull_request:10148|pull_request:8618", + "pull_request:10245|pull_request:8685", + "pull_request:10299|pull_request:9934", + "pull_request:10312|pull_request:11080", + "pull_request:10318|pull_request:11319", + "pull_request:10322|pull_request:8985", + "pull_request:10326|pull_request:8684", + "pull_request:10420|pull_request:9015", + "pull_request:10430|pull_request:9769", + "pull_request:10585|pull_request:10784", + "pull_request:10648|pull_request:11647", + "pull_request:10784|pull_request:11778", + "pull_request:10784|pull_request:9765", + "pull_request:10810|pull_request:9807" + ], + "prepared_review_unit_hash": "ccc0b56992256f6a0d5018cd8b449985cfaaa5089c9c8ccefc008eaa2378418d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 97099, + "estimated_input_tokens": 24275, + "estimated_eval_tokens": 48806 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft edges look like false positives: these PRs mostly cover unrelated docs, CI, tests, or distinct feature/fix work within broad subsystems. No single canonical duplicate pair emerges.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the cluster is heterogeneous, and the apparent similarities are mostly broad subsystem overlap rather than the same underlying change.", + "best_issue_reason": null, + "best_pr_reason": "PR 10148 is the strongest standalone representative by scope and review activity, but it is still not a duplicate anchor for the other items.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2954", + "right": "pull_request:3125", + "accept": false, + "reason": "Different changes: TextToVideoZero pipeline addition vs multi-control-image handling in ControlNet inpaint docs/example code." + }, + { + "left": "pull_request:5038", + "right": "pull_request:5823", + "accept": false, + "reason": "Both relate to LoRA, but one is an API signature cleanup and the other adds adapter_names support; different code paths and goals." + }, + { + "left": "pull_request:7096", + "right": "pull_request:7906", + "accept": false, + "reason": "Both are training/examples work, but they target different projects and scripts; not the same bug or change." + }, + { + "left": "pull_request:4780", + "right": "pull_request:5202", + "accept": false, + "reason": "PEFT/LoRA-related, but one is broad PEFT integration and the other adds Kohya checkpoint key mapping; not mergeable as one fix." + }, + { + "left": "pull_request:10837", + "right": "pull_request:11428", + "accept": false, + "reason": "Both touch HunyuanVideo docs/pipeline surface area, but one adds SkyReels T2V/I2V and the other adds Framepack; distinct features." + }, + { + "left": "pull_request:11639", + "right": "pull_request:11997", + "accept": false, + "reason": "Wan-related, but one standardizes v2v VAE sampling while the other adds modular I2V/FLF2V blocks; separate changes." + }, + { + "left": "pull_request:12024", + "right": "pull_request:12909", + "accept": false, + "reason": "Both modify Wan transformer code, but they fix different issues: timestep reshaping vs context-parallel handling." + }, + { + "left": "pull_request:10042", + "right": "pull_request:9832", + "accept": false, + "reason": "Both CI/workflow-related, but one removes a workflow and the other adds quantization nightly tests; unrelated." + }, + { + "left": "pull_request:10114", + "right": "pull_request:10894", + "accept": false, + "reason": "One is a formatting fix in Stable Audio; the other is docs toctree sorting. No shared bug or change." + }, + { + "left": "pull_request:10115", + "right": "pull_request:9107", + "accept": false, + "reason": "Flux rotary embedding bugfix vs SD3 LoRA loading hook fix; different models and code paths." + }, + { + "left": "pull_request:10132", + "right": "pull_request:11366", + "accept": false, + "reason": "Tokenizers update in PEFT backend workflow vs Hidream transformer output shape/API change; unrelated." + }, + { + "left": "pull_request:10148", + "right": "pull_request:11319", + "accept": false, + "reason": "OmniGen feature addition vs a community-projects docs entry; not the same underlying change." + }, + { + "left": "pull_request:10148", + "right": "pull_request:8618", + "accept": false, + "reason": "OmniGen is a large new pipeline/model feature; the other changes SD3 tiny autoencoder config. Different work." + }, + { + "left": "pull_request:10245", + "right": "pull_request:8685", + "accept": false, + "reason": "PEFT docs text change vs SD3 float8 note; both documentation-only but about different topics." + }, + { + "left": "pull_request:10299", + "right": "pull_request:9934", + "accept": false, + "reason": "GGUF docs generator placement fix vs Mochi docs expansion; unrelated documentation updates." + }, + { + "left": "pull_request:10312", + "right": "pull_request:11080", + "accept": false, + "reason": "Attention head_to_batch_dim bugfix vs Kandinsky image resize/divisibility fix; different subsystems and bugs." + }, + { + "left": "pull_request:10318", + "right": "pull_request:11319", + "accept": false, + "reason": "Flux LoRA integration tests vs community projects docs entry; no shared implementation or bug." + }, + { + "left": "pull_request:10322", + "right": "pull_request:8985", + "accept": false, + "reason": "Sana 2K model conversion support vs reducing Lumina test size; unrelated." + }, + { + "left": "pull_request:10326", + "right": "pull_request:8684", + "accept": false, + "reason": "MPS workflow quoting fix vs Pillow import docs fix; no common code-path problem." + }, + { + "left": "pull_request:10420", + "right": "pull_request:9015", + "accept": false, + "reason": "LTX checkpoint ID docs fix vs PIA example cleanup; unrelated doc updates." + }, + { + "left": "pull_request:10430", + "right": "pull_request:9769", + "accept": false, + "reason": "TeaCache core implementation vs Mochi T2V model/pipeline addition; not the same change." + }, + { + "left": "pull_request:10585", + "right": "pull_request:10784", + "accept": false, + "reason": "Mochi VAE annotation cleanup vs Lumina2 input validation fix; different models and bugs." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11647", + "accept": false, + "reason": "DDPM docs adding sigmoid schedule vs adding Finegrained FP8 quantization; unrelated." + }, + { + "left": "pull_request:10784", + "right": "pull_request:11778", + "accept": false, + "reason": "Lumina2 input validation vs Wan VACE mask padding; separate bugs in different pipelines." + }, + { + "left": "pull_request:10784", + "right": "pull_request:9765", + "accept": false, + "reason": "Lumina2 validation bugfix vs SD3 docstring typo fix; different scope and no shared code-path." + }, + { + "left": "pull_request:10810", + "right": "pull_request:9807", + "accept": false, + "reason": "Flux LoRA conversion improvements vs AutoencoderKL LoRA adapter support; both LoRA-related but distinct fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 183, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10819", + "pull_request:10834", + "pull_request:10837", + "pull_request:10885", + "pull_request:11066", + "pull_request:11067", + "pull_request:11243", + "pull_request:11290", + "pull_request:11366", + "pull_request:11537", + "pull_request:11564", + "pull_request:11583", + "pull_request:11647", + "pull_request:11722", + "pull_request:1194", + "pull_request:12018", + "pull_request:12062", + "pull_request:12134", + "pull_request:12142", + "pull_request:12143", + "pull_request:12231", + "pull_request:12233", + "pull_request:1229", + "pull_request:12330", + "pull_request:12360", + "pull_request:12391", + "pull_request:12420", + "pull_request:12434", + "pull_request:12450", + "pull_request:12551", + "pull_request:12631", + "pull_request:12644", + "pull_request:12675", + "pull_request:12686", + "pull_request:12705", + "pull_request:12797", + "pull_request:12856", + "pull_request:13022", + "pull_request:13076", + "pull_request:13164", + "pull_request:13218", + "pull_request:13454", + "pull_request:13562", + "pull_request:2983", + "pull_request:3172", + "pull_request:45", + "pull_request:9687", + "pull_request:9735" + ], + "soft_pairs": [ + "pull_request:10819|pull_request:11564", + "pull_request:10834|pull_request:11537", + "pull_request:10837|pull_request:11066", + "pull_request:10885|pull_request:11647", + "pull_request:11067|pull_request:12062", + "pull_request:11243|pull_request:9687", + "pull_request:11290|pull_request:12420", + "pull_request:11366|pull_request:9735", + "pull_request:11583|pull_request:12134", + "pull_request:11722|pull_request:12233", + "pull_request:1194|pull_request:3172", + "pull_request:1194|pull_request:45", + "pull_request:12018|pull_request:12391", + "pull_request:12018|pull_request:12797", + "pull_request:12142|pull_request:13022", + "pull_request:12143|pull_request:13022", + "pull_request:12231|pull_request:12705", + "pull_request:1229|pull_request:2983", + "pull_request:12330|pull_request:13218", + "pull_request:12360|pull_request:12434", + "pull_request:12450|pull_request:13164", + "pull_request:12551|pull_request:13454", + "pull_request:12551|pull_request:13562", + "pull_request:12631|pull_request:13076", + "pull_request:12644|pull_request:12675", + "pull_request:12686|pull_request:12856" + ], + "prepared_review_unit_hash": "5c3af2a27dd4e7c14b52d9a7b5cbdd816132453e0582b76a0eebc19c30f5983b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 97146, + "estimated_input_tokens": 24287, + "estimated_eval_tokens": 48830 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a set of largely unrelated PRs that only look superficially similar by subsystem or docs/CI overlap. I do not see any true duplicate/mergeable PR pairs here, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10819", + "right": "pull_request:11564", + "accept": false, + "reason": "Both touch docs/example-related content, but one adds LoRA docs entries while the other adds a dreambooth dependency; no shared bug or change." + }, + { + "left": "pull_request:10834", + "right": "pull_request:11537", + "accept": false, + "reason": "GGUF isinstance fix vs adding a torch.compile test for UNet are unrelated code/test changes." + }, + { + "left": "pull_request:10837", + "right": "pull_request:11066", + "accept": false, + "reason": "Both are HunyuanVideo-related, but one adds SkyReels Hunyuan T2V/I2V support and the other introduces a new HunyuanVideo-I2V model; different functionality." + }, + { + "left": "pull_request:10885", + "right": "pull_request:11647", + "accept": false, + "reason": "Both are docs/feature additions, but CogVideoX schedulers and Finegrained FP8 quantization are separate subsystems and changes." + }, + { + "left": "pull_request:11067", + "right": "pull_request:12062", + "accept": false, + "reason": "Both modify CI/benchmarking, but one pins transformers for benchmarks while the other enables all GPUs in CI containers; not the same issue." + }, + { + "left": "pull_request:11243", + "right": "pull_request:9687", + "accept": false, + "reason": "Musubi Wan LoRA conversion support and Xlabs ControlNet support are different model-path features with different code paths." + }, + { + "left": "pull_request:11290", + "right": "pull_request:12420", + "accept": false, + "reason": "Dockerfile torch pinning and conditional import of torch.distributed are unrelated fixes." + }, + { + "left": "pull_request:11366", + "right": "pull_request:9735", + "accept": false, + "reason": "Hidream transformer output-shape change and SD3 ControlNet example updates are unrelated." + }, + { + "left": "pull_request:11583", + "right": "pull_request:12134", + "accept": false, + "reason": "Both are docs removals/refreshes, but one removes the fast diffusion tutorial and the other rewrites the Stable Diffusion intro; different content." + }, + { + "left": "pull_request:11722", + "right": "pull_request:12233", + "accept": false, + "reason": "Both are documentation edits, but a misleading code comment in a script is not the same as renaming a quantization docs title." + }, + { + "left": "pull_request:1194", + "right": "pull_request:3172", + "accept": false, + "reason": "Different doc-builder workflow updates years apart; one updates PR docs actions, the other documents deterministic algorithms." + }, + { + "left": "pull_request:1194", + "right": "pull_request:45", + "accept": false, + "reason": "Both touch doc workflows, but 45 is the initial docs workflow setup while 1194 changes action references/secrets; not the same change." + }, + { + "left": "pull_request:12018", + "right": "pull_request:12391", + "accept": false, + "reason": "A link-text fix in quantization overview is not the same as a broad quantization docs reorganization/rewrite." + }, + { + "left": "pull_request:12018", + "right": "pull_request:12797", + "accept": false, + "reason": "Quantization link text fix vs Kandinsky 5 docs fixes are unrelated." + }, + { + "left": "pull_request:12142", + "right": "pull_request:13022", + "accept": false, + "reason": "WAN VACE bf16/fp16 mask handling and CUDA/xformers Dockerfile fixes are unrelated." + }, + { + "left": "pull_request:12143", + "right": "pull_request:13022", + "accept": false, + "reason": "The WAN VACE test/mask dtype fix is not the same as Dockerfile CUDA/xformers adjustments." + }, + { + "left": "pull_request:12231", + "right": "pull_request:12705", + "accept": false, + "reason": "A broad typo cleanup across many example scripts is not the same change as a smaller typo cleanup in assorted docs/examples." + }, + { + "left": "pull_request:1229", + "right": "pull_request:2983", + "accept": false, + "reason": "Dreambooth README dependency guidance and CONTRIBUTING test-dependency instructions are different documentation changes." + }, + { + "left": "pull_request:12330", + "right": "pull_request:13218", + "accept": false, + "reason": "Both are test-related, but one cleans group offloading tests while the other skips an invalid Helios save/load case; different failures." + }, + { + "left": "pull_request:12360", + "right": "pull_request:12434", + "accept": false, + "reason": "Chroma doc wording fixes and text-to-image training PEFT version changes are unrelated." + }, + { + "left": "pull_request:12450", + "right": "pull_request:13164", + "accept": false, + "reason": "Both alter CI installs, but 12450 revisits overall CI installation strategy whereas 13164 removes test deps from CI jobs; not the same concrete change." + }, + { + "left": "pull_request:12551", + "right": "pull_request:13454", + "accept": false, + "reason": "Skipping Sana layerwise-casting tests in CI and xfailing CLIP-related pipeline tests address different failures." + }, + { + "left": "pull_request:12551", + "right": "pull_request:13562", + "accept": false, + "reason": "Sana-specific CI skips and tokenizers installation stability in CI are unrelated." + }, + { + "left": "pull_request:12631", + "right": "pull_request:13076", + "accept": false, + "reason": "Distributed inference docs and a quantization syntax fix are unrelated documentation changes." + }, + { + "left": "pull_request:12644", + "right": "pull_request:12675", + "accept": false, + "reason": "AutoModel docs and Sana video typo fixes are unrelated docs edits." + }, + { + "left": "pull_request:12686", + "right": "pull_request:12856", + "accept": false, + "reason": "Kandinsky5 pipeline implementation and z-image-omni-base implementation are completely different model introductions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 184, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:12686", + "pull_request:1292", + "pull_request:13567", + "pull_request:1397", + "pull_request:1423", + "pull_request:1437", + "pull_request:1508", + "pull_request:1697", + "pull_request:1765", + "pull_request:1788", + "pull_request:1850", + "pull_request:1903", + "pull_request:1951", + "pull_request:2009", + "pull_request:2103", + "pull_request:2129", + "pull_request:2183", + "pull_request:2257", + "pull_request:2330", + "pull_request:2349", + "pull_request:2365", + "pull_request:2384", + "pull_request:2405", + "pull_request:2490", + "pull_request:2598", + "pull_request:2811", + "pull_request:2817", + "pull_request:2959", + "pull_request:3006", + "pull_request:3172", + "pull_request:3201", + "pull_request:3226", + "pull_request:326", + "pull_request:3265", + "pull_request:3331", + "pull_request:3561", + "pull_request:3709", + "pull_request:3716", + "pull_request:396", + "pull_request:416", + "pull_request:533", + "pull_request:564", + "pull_request:570", + "pull_request:703", + "pull_request:748", + "pull_request:806", + "pull_request:840", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:12686|pull_request:13567", + "pull_request:1292|pull_request:2405", + "pull_request:1397|pull_request:1423", + "pull_request:1437|pull_request:3331", + "pull_request:1508|pull_request:3561", + "pull_request:1508|pull_request:748", + "pull_request:1697|pull_request:1951", + "pull_request:1765|pull_request:564", + "pull_request:1788|pull_request:570", + "pull_request:1850|pull_request:2959", + "pull_request:1903|pull_request:3006", + "pull_request:2009|pull_request:2183", + "pull_request:2103|pull_request:806", + "pull_request:2129|pull_request:3716", + "pull_request:2257|pull_request:840", + "pull_request:2330|pull_request:3226", + "pull_request:2349|pull_request:396", + "pull_request:2349|pull_request:416", + "pull_request:2365|pull_request:326", + "pull_request:2384|pull_request:3201", + "pull_request:2384|pull_request:3265", + "pull_request:2384|pull_request:533", + "pull_request:2490|pull_request:3709", + "pull_request:2598|pull_request:3172", + "pull_request:2811|pull_request:703", + "pull_request:2817|pull_request:975" + ], + "prepared_review_unit_hash": "1022873c5465a39f544ced1f83a934a4116852901b1472880fc8bb92cee34ba2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 101889, + "estimated_input_tokens": 25473, + "estimated_eval_tokens": 51202 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft edges are false positives: the paired PRs are mostly unrelated docs edits, feature additions, or separate reverts in different code paths. No duplicate cluster emerges from this set.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue artifacts are present. If a PR must be chosen as the closest representative of the set, #2009 is the strongest core-library change with broad downstream impact and high discussion activity, but it is not a duplicate of the others.", + "best_issue_reason": null, + "best_pr_reason": "#2009 is the best representative PR because it is a substantial, central library change affecting loaders and many pipelines; the rest are mostly isolated docs/typo/test/revert changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12686", + "right": "pull_request:13567", + "accept": false, + "reason": "Kandinsky5 pipeline docs/code vs agents documentation update; unrelated scopes." + }, + { + "left": "pull_request:1292", + "right": "pull_request:2405", + "accept": false, + "reason": "Both are reverts, but they revert different changes in different areas and do not fix the same bug." + }, + { + "left": "pull_request:1397", + "right": "pull_request:1423", + "accept": false, + "reason": "SD2 attention slicing vs Flax SD2 support; different concrete changes." + }, + { + "left": "pull_request:1437", + "right": "pull_request:3331", + "accept": false, + "reason": "Separate typo fixes in different pipeline/docs files; not the same issue." + }, + { + "left": "pull_request:1508", + "right": "pull_request:3561", + "accept": false, + "reason": "Dreambooth docs clarification vs Dreambooth IF stage II upscaling; unrelated changes." + }, + { + "left": "pull_request:1508", + "right": "pull_request:748", + "accept": false, + "reason": "Docs clarification vs push_to_hub fix; different bugs." + }, + { + "left": "pull_request:1697", + "right": "pull_request:1951", + "accept": false, + "reason": "Community pipeline update vs ColossalAI update; no shared concrete code-path problem." + }, + { + "left": "pull_request:1765", + "right": "pull_request:564", + "accept": false, + "reason": "Dreambooth Flax fixes and Flax scheduler fixes are different subsystems and bugs." + }, + { + "left": "pull_request:1788", + "right": "pull_request:570", + "accept": false, + "reason": "DDIM sampler bug fix vs BaseOutput dict initialization; unrelated code paths." + }, + { + "left": "pull_request:1850", + "right": "pull_request:2959", + "accept": false, + "reason": "Test additions for different models; not the same underlying issue." + }, + { + "left": "pull_request:1903", + "right": "pull_request:3006", + "accept": false, + "reason": "Comment cleanup in depth2img vs DDPM variance logic fix; unrelated." + }, + { + "left": "pull_request:2009", + "right": "pull_request:2183", + "accept": false, + "reason": "Textual inversion embedding loading vs logging hotfix; different functionality." + }, + { + "left": "pull_request:2103", + "right": "pull_request:806", + "accept": false, + "reason": "LoRA model tag docs vs Dreambooth EMA training work; no duplicate bug/change." + }, + { + "left": "pull_request:2129", + "right": "pull_request:3716", + "accept": false, + "reason": "LoRA pipeline support vs ParaDiGMS pipeline addition; distinct features." + }, + { + "left": "pull_request:2257", + "right": "pull_request:840", + "accept": false, + "reason": "Both are community examples, but they implement different pipelines and changes." + }, + { + "left": "pull_request:2330", + "right": "pull_request:3226", + "accept": false, + "reason": "Ethical guidelines docs vs ControlNet v1.1 docs; unrelated documentation updates." + }, + { + "left": "pull_request:2349", + "right": "pull_request:396", + "accept": false, + "reason": "KarrasDiffusionSchedulers overview docs vs DDIM docs; different documentation targets." + }, + { + "left": "pull_request:2349", + "right": "pull_request:416", + "accept": false, + "reason": "KarrasDiffusionSchedulers docs note vs general models docs; unrelated." + }, + { + "left": "pull_request:2365", + "right": "pull_request:326", + "accept": false, + "reason": "Type-note change for Karras schedulers vs PNDM pipeline typing; not the same change." + }, + { + "left": "pull_request:2384", + "right": "pull_request:3201", + "accept": false, + "reason": "Revert of a Pix2Pix Zero utility vs revert of lpw pipeline update; different underlying PRs." + }, + { + "left": "pull_request:2384", + "right": "pull_request:3265", + "accept": false, + "reason": "Different revert targets and code paths; not duplicates." + }, + { + "left": "pull_request:2384", + "right": "pull_request:533", + "accept": false, + "reason": "Pix2Pix Zero revert vs DDIM scheduler typehint revert; unrelated." + }, + { + "left": "pull_request:2490", + "right": "pull_request:3709", + "accept": false, + "reason": "MultiDiffusion docs addition vs image processor docs; different documentation areas." + }, + { + "left": "pull_request:2598", + "right": "pull_request:3172", + "accept": false, + "reason": "Paint By Example docs vs deterministic algorithms docs; no shared bug or change." + }, + { + "left": "pull_request:2811", + "right": "pull_request:703", + "accept": false, + "reason": "ONNX test expectation update vs CI workflow change; unrelated." + }, + { + "left": "pull_request:2817", + "right": "pull_request:975", + "accept": false, + "reason": "UNet1D documentation vs OpenVINO title typo; different docs changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 185, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2894", + "pull_request:3042", + "pull_request:3183", + "pull_request:3226", + "pull_request:3294", + "pull_request:3295", + "pull_request:3331", + "pull_request:3410", + "pull_request:3462", + "pull_request:3541", + "pull_request:3589", + "pull_request:3608", + "pull_request:3673", + "pull_request:3681", + "pull_request:3786", + "pull_request:3806", + "pull_request:382", + "pull_request:3820", + "pull_request:386", + "pull_request:388", + "pull_request:390", + "pull_request:391", + "pull_request:392", + "pull_request:396", + "pull_request:404", + "pull_request:4058", + "pull_request:4078", + "pull_request:4132", + "pull_request:4155", + "pull_request:416", + "pull_request:417", + "pull_request:4196", + "pull_request:4211", + "pull_request:4227", + "pull_request:4259", + "pull_request:4284", + "pull_request:4347", + "pull_request:4498", + "pull_request:4652", + "pull_request:5024", + "pull_request:5201", + "pull_request:5384", + "pull_request:5505", + "pull_request:5698", + "pull_request:5747", + "pull_request:5768", + "pull_request:6163", + "pull_request:6485" + ], + "soft_pairs": [ + "pull_request:2894|pull_request:3226", + "pull_request:3042|pull_request:4078", + "pull_request:3183|pull_request:3541", + "pull_request:3294|pull_request:5201", + "pull_request:3295|pull_request:5505", + "pull_request:3331|pull_request:4211", + "pull_request:3410|pull_request:4078", + "pull_request:3462|pull_request:3673", + "pull_request:3589|pull_request:5024", + "pull_request:3608|pull_request:5505", + "pull_request:3681|pull_request:4132", + "pull_request:3786|pull_request:4347", + "pull_request:3806|pull_request:5747", + "pull_request:382|pull_request:392", + "pull_request:382|pull_request:404", + "pull_request:382|pull_request:417", + "pull_request:3820|pull_request:5768", + "pull_request:386|pull_request:391", + "pull_request:388|pull_request:396", + "pull_request:388|pull_request:416", + "pull_request:390|pull_request:391", + "pull_request:4058|pull_request:4155", + "pull_request:4058|pull_request:6485", + "pull_request:4196|pull_request:4227", + "pull_request:4259|pull_request:5698", + "pull_request:4284|pull_request:5768", + "pull_request:4498|pull_request:5384", + "pull_request:4652|pull_request:6163" + ], + "prepared_review_unit_hash": "66a990464013a2f10242ab1f63b8b62b3d50712f28b1e93f7664883328c924d6", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 95607, + "estimated_input_tokens": 23902, + "estimated_eval_tokens": 48060 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a false positive: the PRs span unrelated docs edits, example-script fixes, dependency bumps, loader changes, and scheduler/controlnet logic changes. There is no single underlying bug or change family to deduplicate.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR fits the cluster; the items are too heterogeneous across unrelated files and code paths, so any representative would be arbitrary.", + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:2894", + "right": "pull_request:3226", + "accept": false, + "reason": "Both are docs/example-related, but one adds a flax requirements file while the other edits ControlNet docs; different changes and no shared bug." + }, + { + "left": "pull_request:3042", + "right": "pull_request:4078", + "accept": false, + "reason": "Both involve accelerate, but 3042 raises a minimum version across many examples while 4078 only adds an install guard for from_single_file; different issues." + }, + { + "left": "pull_request:3183", + "right": "pull_request:3541", + "accept": false, + "reason": "Different DreamBooth problems: one fixes LoRA checkpointing, the other fixes attention mask collation in the base training script." + }, + { + "left": "pull_request:3294", + "right": "pull_request:5201", + "accept": false, + "reason": "Both touch LoRA internals, but 3294 adds lora prefix loading support and 5201 fixes xformers/processor offload behavior; not the same concrete change." + }, + { + "left": "pull_request:3295", + "right": "pull_request:5505", + "accept": false, + "reason": "These are unrelated typo/docs fixes in different files and subsystems." + }, + { + "left": "pull_request:3331", + "right": "pull_request:4211", + "accept": false, + "reason": "Both are doc typos, but they target different SDXL/inpainting docs and different example text; no same underlying bug." + }, + { + "left": "pull_request:3410", + "right": "pull_request:4078", + "accept": false, + "reason": "One fixes deepspeed plugin retrieval in a training script; the other makes accelerate mandatory for single-file loading. Different code paths." + }, + { + "left": "pull_request:3462", + "right": "pull_request:3673", + "accept": false, + "reason": "Same DreamBooth script area, but one changes robustness around orig UNet handling and the other fixes multi-GPU behavior via unwrap_model; distinct bugs." + }, + { + "left": "pull_request:3589", + "right": "pull_request:5024", + "accept": false, + "reason": "Unrelated topics: reference-trick docs/example updates versus scheduler step-index bookkeeping." + }, + { + "left": "pull_request:3608", + "right": "pull_request:5505", + "accept": false, + "reason": "Both are typo-related, but they affect different examples/docs and different text; not a duplicate change." + }, + { + "left": "pull_request:3681", + "right": "pull_request:4132", + "accept": false, + "reason": "Both concern accelerate usage, but 3681 replaces project configuration handling in training scripts while 4132 makes accelerate optional in checkpoint conversion; different problems." + }, + { + "left": "pull_request:3786", + "right": "pull_request:4347", + "accept": false, + "reason": "PR template changes are unrelated to scheduler step-index logic." + }, + { + "left": "pull_request:3806", + "right": "pull_request:5747", + "accept": false, + "reason": "One is a README note about loading from checkpoints; the other adds a data_dir parameter to dataset loading. Different concerns." + }, + { + "left": "pull_request:382", + "right": "pull_request:392", + "accept": false, + "reason": "Both are early docs expansions, but they cover different pages and unrelated content areas." + }, + { + "left": "pull_request:382", + "right": "pull_request:404", + "accept": false, + "reason": "Stable Diffusion pipeline docs vs fp16 optimization docs are separate documentation changes, not the same issue." + }, + { + "left": "pull_request:382", + "right": "pull_request:417", + "accept": false, + "reason": "These are different documentation pages for different user guides; no shared underlying change." + }, + { + "left": "pull_request:3820", + "right": "pull_request:5768", + "accept": false, + "reason": "LDM3D typo/docs cleanup is unrelated to consistency decoder documentation fixes." + }, + { + "left": "pull_request:386", + "right": "pull_request:391", + "accept": false, + "reason": "Two different pipeline docs (Stable Diffusion vs PNDM); same general docs theme but not the same change." + }, + { + "left": "pull_request:388", + "right": "pull_request:396", + "accept": false, + "reason": "Different pipeline documentation pages and code references; no shared bug or feature." + }, + { + "left": "pull_request:388", + "right": "pull_request:416", + "accept": false, + "reason": "Unconditional latent diffusion docs versus model API docs are unrelated." + }, + { + "left": "pull_request:390", + "right": "pull_request:391", + "accept": false, + "reason": "Different optimization/docs pages (MPS hardware guidance vs PNDM pipeline docs)." + }, + { + "left": "pull_request:4058", + "right": "pull_request:4155", + "accept": false, + "reason": "SDXL docs cleanup and SDXL ControlNet pipeline follow-up fixes are not the same change; one is documentation-only, the other changes code and tests." + }, + { + "left": "pull_request:4058", + "right": "pull_request:6485", + "accept": false, + "reason": "SDXL docs correction and PR template reviewer update are unrelated." + }, + { + "left": "pull_request:4196", + "right": "pull_request:4227", + "accept": false, + "reason": "SDXL docs improvements versus a missing import in textual inversion inference are different files and bugs." + }, + { + "left": "pull_request:4259", + "right": "pull_request:5698", + "accept": false, + "reason": "Circular-import fix for from_single_file loaders is unrelated to latent-consistency img2img behavior." + }, + { + "left": "pull_request:4284", + "right": "pull_request:5768", + "accept": false, + "reason": "Both are LDM3D-related docs fixes, but they touch different files and separate typos; not the same concrete issue." + }, + { + "left": "pull_request:4498", + "right": "pull_request:5384", + "accept": false, + "reason": "Different feature work: SDXL reference control support versus basic hypernetwork support for StableDiffusionPipeline." + }, + { + "left": "pull_request:4652", + "right": "pull_request:6163", + "accept": false, + "reason": "Both are LoRA tests, but they cover different test suites and behaviors; not a single fix that could be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 186, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4669", + "pull_request:5188", + "pull_request:5239", + "pull_request:5242", + "pull_request:5456", + "pull_request:5552", + "pull_request:5698", + "pull_request:5707", + "pull_request:5771", + "pull_request:5822", + "pull_request:5895", + "pull_request:5982", + "pull_request:5988", + "pull_request:6002", + "pull_request:6053", + "pull_request:6085", + "pull_request:6119", + "pull_request:6192", + "pull_request:6205", + "pull_request:6215", + "pull_request:6217", + "pull_request:6308", + "pull_request:6342", + "pull_request:6437", + "pull_request:6643", + "pull_request:6656", + "pull_request:6775", + "pull_request:6997", + "pull_request:7077", + "pull_request:7103", + "pull_request:7124", + "pull_request:7153", + "pull_request:7166", + "pull_request:7206", + "pull_request:7319", + "pull_request:7328", + "pull_request:7410", + "pull_request:7634", + "pull_request:7662", + "pull_request:7827", + "pull_request:7941", + "pull_request:8229", + "pull_request:8230", + "pull_request:8303", + "pull_request:8425", + "pull_request:8510", + "pull_request:8521", + "pull_request:9247" + ], + "soft_pairs": [ + "pull_request:4669|pull_request:6437", + "pull_request:5188|pull_request:7328", + "pull_request:5239|pull_request:6775", + "pull_request:5242|pull_request:6775", + "pull_request:5456|pull_request:6192", + "pull_request:5456|pull_request:6656", + "pull_request:5456|pull_request:7634", + "pull_request:5552|pull_request:6085", + "pull_request:5698|pull_request:5707", + "pull_request:5707|pull_request:5822", + "pull_request:5707|pull_request:5895", + "pull_request:5771|pull_request:7410", + "pull_request:5895|pull_request:6053", + "pull_request:5982|pull_request:6053", + "pull_request:5982|pull_request:6643", + "pull_request:5988|pull_request:7319", + "pull_request:6002|pull_request:8230", + "pull_request:6119|pull_request:8425", + "pull_request:6205|pull_request:7827", + "pull_request:6215|pull_request:7941", + "pull_request:6215|pull_request:8303", + "pull_request:6217|pull_request:8510", + "pull_request:6308|pull_request:8229", + "pull_request:6342|pull_request:8521", + "pull_request:6997|pull_request:7206", + "pull_request:7077|pull_request:7124", + "pull_request:7103|pull_request:9247", + "pull_request:7153|pull_request:7662", + "pull_request:7166|pull_request:9247" + ], + "prepared_review_unit_hash": "6fa1018a61a6f3798a85e5d204a369fdaa700fea11f9559aef4134e39f32ef9a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 96959, + "estimated_input_tokens": 24240, + "estimated_eval_tokens": 48736 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated docs, CI, training-script, scheduler, pipeline, and loader changes. None of the soft pairs look like the same underlying bug or mergeable change, so there is no duplicate group to collapse.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items span many unrelated concerns, and no single PR meaningfully represents the rest as a duplicate fix/change.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong global representative; the largest/broadest PRs are still about different subsystems and goals, not the same concrete code-path problem.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4669", + "right": "pull_request:6437", + "accept": false, + "reason": "Different problems: LoRA rank handling in text-encoder loading vs excluding a PEFT-specific test from CI." + }, + { + "left": "pull_request:5188", + "right": "pull_request:7328", + "accept": false, + "reason": "Unrelated scopes: adding compute_snr to training utils vs adding torch to a PyPI workflow step." + }, + { + "left": "pull_request:5239", + "right": "pull_request:6775", + "accept": false, + "reason": "Different artifacts: AutoencoderKL encode docstring vs DDPMScheduler parameter docs." + }, + { + "left": "pull_request:5242", + "right": "pull_request:6775", + "accept": false, + "reason": "Different artifacts: AutoencoderKL decode docstring vs DDPMScheduler parameter docs." + }, + { + "left": "pull_request:5456", + "right": "pull_request:6192", + "accept": false, + "reason": "RealFill example/package addition vs IP-Adapter unload behavior in loaders are unrelated changes." + }, + { + "left": "pull_request:5456", + "right": "pull_request:6656", + "accept": false, + "reason": "Different features: RealFill example vs textual inversion unload API." + }, + { + "left": "pull_request:5456", + "right": "pull_request:7634", + "accept": false, + "reason": "Different domains: RealFill example vs high-res UNet fix implementation." + }, + { + "left": "pull_request:5552", + "right": "pull_request:6085", + "accept": false, + "reason": "FreeU disable-method bugfix vs scheduler add_noise indexing bugfix; different code paths." + }, + { + "left": "pull_request:5698", + "right": "pull_request:5707", + "accept": false, + "reason": "LCM img2img pipeline fix vs separate LCM test additions; not the same change or bug." + }, + { + "left": "pull_request:5707", + "right": "pull_request:5822", + "accept": false, + "reason": "Test coverage for LCM vs ControlNet/Adapter support for LCM; different concrete implementations." + }, + { + "left": "pull_request:5707", + "right": "pull_request:5895", + "accept": false, + "reason": "LCM tests vs SVD feature PR; no shared underlying fix." + }, + { + "left": "pull_request:5771", + "right": "pull_request:7410", + "accept": false, + "reason": "LoRA loaders refactor vs FreeInit behavior fix; unrelated code paths and goals." + }, + { + "left": "pull_request:5895", + "right": "pull_request:6053", + "accept": false, + "reason": "Stable Video Diffusion feature work vs step_loss logging in a training example; not the same bug/change." + }, + { + "left": "pull_request:5982", + "right": "pull_request:6053", + "accept": false, + "reason": "SDXL Turbo docs vs training loss logging; unrelated documentation and example code." + }, + { + "left": "pull_request:5982", + "right": "pull_request:6643", + "accept": false, + "reason": "Both are docs, but for different models/pages (SDXL Turbo vs UViT2D), so not duplicates." + }, + { + "left": "pull_request:5988", + "right": "pull_request:7319", + "accept": false, + "reason": "Kandinsky 3 docs vs EDM scheduler docs; unrelated documentation additions." + }, + { + "left": "pull_request:6002", + "right": "pull_request:8230", + "accept": false, + "reason": "SVD guidance-scale fix vs CPU offloading usage/docs cleanup; different problems." + }, + { + "left": "pull_request:6119", + "right": "pull_request:8425", + "accept": false, + "reason": "LoRA training fp16 fix vs workflow rename; no shared code path." + }, + { + "left": "pull_request:6205", + "right": "pull_request:7827", + "accept": false, + "reason": "T2I-Adapter blog URL docs fix vs ControlNet docstring CPU-offload example fix; unrelated documentation edits." + }, + { + "left": "pull_request:6215", + "right": "pull_request:7941", + "accept": false, + "reason": "PixArt test seeding/expected values vs marking a single-file test class slow; different test concerns." + }, + { + "left": "pull_request:6215", + "right": "pull_request:8303", + "accept": false, + "reason": "PixArt test determinism fix vs philosophy typo edit; unrelated." + }, + { + "left": "pull_request:6217", + "right": "pull_request:8510", + "accept": false, + "reason": "SVD torch.compile VAE fix vs DreamBooth LR scheduler fix; different model/pipeline issues." + }, + { + "left": "pull_request:6308", + "right": "pull_request:8229", + "accept": false, + "reason": "LoRA PEFT dummy-component test fixes vs SVD VideoProcessor resize fix; unrelated test/setup vs runtime behavior." + }, + { + "left": "pull_request:6342", + "right": "pull_request:8521", + "accept": false, + "reason": "Installation docs note vs requirements file formatting; different files and concerns." + }, + { + "left": "pull_request:6997", + "right": "pull_request:7206", + "accept": false, + "reason": "Draft LoRA support for UNetSpatioTemporalConditionModel vs training-script tagging metadata; unrelated." + }, + { + "left": "pull_request:7077", + "right": "pull_request:7124", + "accept": false, + "reason": "IP-Adapter head_to_batch_dim fix vs DreamBooth LoRA kwarg rename; different code paths." + }, + { + "left": "pull_request:7103", + "right": "pull_request:9247", + "accept": false, + "reason": "ControlNet check_image bugfix vs pipeline_loading_utils variant detection bugfix; different subsystems and failures." + }, + { + "left": "pull_request:7153", + "right": "pull_request:7662", + "accept": false, + "reason": "Documentation workflow path filter fix vs removing redundant packages from flax PR tests; unrelated CI changes." + }, + { + "left": "pull_request:7166", + "right": "pull_request:9247", + "accept": false, + "reason": "SDXL LoRA loading fix vs variant-folder identification fix; different loader logic and symptoms." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 187, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10324", + "pull_request:10325", + "pull_request:10837", + "pull_request:11475", + "pull_request:11534", + "pull_request:11639", + "pull_request:11975", + "pull_request:12519", + "pull_request:12551", + "pull_request:12611", + "pull_request:13055", + "pull_request:13365", + "pull_request:3770", + "pull_request:3797", + "pull_request:4262", + "pull_request:4370", + "pull_request:4498", + "pull_request:5331", + "pull_request:570", + "pull_request:5807", + "pull_request:5813", + "pull_request:6237", + "pull_request:6316", + "pull_request:6893", + "pull_request:7181", + "pull_request:7410", + "pull_request:7638", + "pull_request:7652", + "pull_request:7653", + "pull_request:7893", + "pull_request:7913", + "pull_request:7919", + "pull_request:7938", + "pull_request:826", + "pull_request:8272", + "pull_request:8322", + "pull_request:8386", + "pull_request:8467", + "pull_request:8522", + "pull_request:8591", + "pull_request:8627", + "pull_request:8702", + "pull_request:8722", + "pull_request:9100", + "pull_request:9408", + "pull_request:9620", + "pull_request:9721", + "pull_request:9735" + ], + "soft_pairs": [ + "pull_request:7181|pull_request:7653", + "pull_request:7638|pull_request:7893", + "pull_request:7652|pull_request:8627", + "pull_request:7913|pull_request:8467", + "pull_request:7919|pull_request:8722", + "pull_request:7938|pull_request:8386", + "pull_request:8272|pull_request:8322", + "pull_request:8322|pull_request:9100", + "pull_request:8522|pull_request:9721", + "pull_request:8591|pull_request:8702", + "pull_request:8702|pull_request:9735", + "pull_request:9408|pull_request:9620", + "pull_request:11639|pull_request:12611", + "pull_request:5807|pull_request:6316", + "pull_request:12551|pull_request:13055", + "pull_request:3770|pull_request:4498", + "pull_request:10837|pull_request:11534", + "pull_request:10324|pull_request:10325", + "pull_request:11475|pull_request:11975", + "pull_request:12519|pull_request:13365", + "pull_request:3797|pull_request:4370", + "pull_request:4262|pull_request:5331", + "pull_request:570|pull_request:826", + "pull_request:5813|pull_request:6237", + "pull_request:6893|pull_request:7410" + ], + "prepared_review_unit_hash": "2c3ea0fb16be38b251a7b07c28327f2d7a563678fdabc1c2221bb7388d43d016", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 104543, + "estimated_input_tokens": 26136, + "estimated_eval_tokens": 52528 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a true duplicate cluster: the items span unrelated fixes across docs, CI, Docker, WAN, Hunyuan, SD3, ControlNet, and test-only changes. The soft edges are mostly superficial theme matches and should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:7181", + "right": "pull_request:7653", + "accept": false, + "reason": "Both touch config-related code, but 7181 is docs/example cleanup while 7653 fixes a classmethod bug in configuration_utils; different problems." + }, + { + "left": "pull_request:7638", + "right": "pull_request:7893", + "accept": false, + "reason": "Both are docs/infra maintenance, but one fixes a broken CONTRIBUTING link and the other upgrades Python in Dockerfiles; unrelated changes." + }, + { + "left": "pull_request:7652", + "right": "pull_request:8627", + "accept": false, + "reason": "Docker buildx workflow setup is unrelated to the SD3 ControlNet feature work." + }, + { + "left": "pull_request:7913", + "right": "pull_request:8467", + "accept": false, + "reason": "ControlNet SDXL timestep support and SEGA device handling are different code paths and bugs." + }, + { + "left": "pull_request:7919", + "right": "pull_request:8722", + "accept": false, + "reason": "Offload/device_map behavior fixes are unrelated to the embeddings refactor." + }, + { + "left": "pull_request:7938", + "right": "pull_request:8386", + "accept": false, + "reason": "Docker workflow status reporting and test shape-property refactors are unrelated." + }, + { + "left": "pull_request:8272", + "right": "pull_request:8322", + "accept": false, + "reason": "Both are typo/errata cleanup, but they fix different text across different files and are not the same patch." + }, + { + "left": "pull_request:8322", + "right": "pull_request:9100", + "accept": false, + "reason": "Both are typo fixes, but they are separate wording corrections in unrelated docs/examples." + }, + { + "left": "pull_request:8522", + "right": "pull_request:9721", + "accept": false, + "reason": "Transformers version pin and torch version pin are separate dependency changes." + }, + { + "left": "pull_request:8591", + "right": "pull_request:8702", + "accept": false, + "reason": "SD3 training weighting refactor and the SD3 conversion script are different deliverables with different purposes." + }, + { + "left": "pull_request:8702", + "right": "pull_request:9735", + "accept": false, + "reason": "A conversion script and an example update for SD3 ControlNet do not fix the same code-path problem." + }, + { + "left": "pull_request:9408", + "right": "pull_request:9620", + "accept": false, + "reason": "Flux ControlNet img2img/inpaint additions and interrupt support are related to ControlNet but solve different issues." + }, + { + "left": "pull_request:11639", + "right": "pull_request:12611", + "accept": false, + "reason": "Wan video-to-video sampling-mode cleanup and the modular WAN refactor are not the same concrete bug/change." + }, + { + "left": "pull_request:5807", + "right": "pull_request:6316", + "accept": false, + "reason": "LPW SDXL prompt embedding batching and moving ControlNetXS to community are unrelated." + }, + { + "left": "pull_request:12551", + "right": "pull_request:13055", + "accept": false, + "reason": "Skipping Sana layerwise-casting tests in CI is unrelated to WAN 2.2 transformer fallback handling." + }, + { + "left": "pull_request:3770", + "right": "pull_request:4498", + "accept": false, + "reason": "Control guidance start/stop and SDXL reference-attention/adain are different feature areas and code paths." + }, + { + "left": "pull_request:10837", + "right": "pull_request:11534", + "accept": false, + "reason": "Both are HunyuanVideo-related, but one adds SkyReels T2V/I2V support and the other adds Framepack F1; not the same change." + }, + { + "left": "pull_request:10324", + "right": "pull_request:10325", + "accept": false, + "reason": "Pipeline offload API fix and EMA test adjustment are unrelated." + }, + { + "left": "pull_request:11475", + "right": "pull_request:11975", + "accept": false, + "reason": "WAN recompilation fix and a Flux training style correction are unrelated." + }, + { + "left": "pull_request:12519", + "right": "pull_request:13365", + "accept": false, + "reason": "Toctree sorting/CI formatting and Claude review workflow fork support are unrelated CI changes." + }, + { + "left": "pull_request:3797", + "right": "pull_request:4370", + "accept": false, + "reason": "Bad attention naming in model init and a dummy import path fix are not the same underlying bug." + }, + { + "left": "pull_request:4262", + "right": "pull_request:5331", + "accept": false, + "reason": "Both touch DreamBooth LoRA, but they address different issues: multiple training fixes vs replacing deprecated attention LoRA procs." + }, + { + "left": "pull_request:570", + "right": "pull_request:826", + "accept": false, + "reason": "BaseOutput dict initialization and DreamBooth fp16 loss casting are unrelated." + }, + { + "left": "pull_request:5813", + "right": "pull_request:6237", + "accept": false, + "reason": "Loader API docs split and batched seed documentation are separate documentation updates." + }, + { + "left": "pull_request:6893", + "right": "pull_request:7410", + "accept": false, + "reason": "3D UNet attention refactor and FreeInit behavior fix target different subsystems and bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 188, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10132", + "pull_request:10200", + "pull_request:1038", + "pull_request:10494", + "pull_request:10597", + "pull_request:10633", + "pull_request:1147", + "pull_request:11504", + "pull_request:11613", + "pull_request:11728", + "pull_request:12363", + "pull_request:12425", + "pull_request:1269", + "pull_request:12973", + "pull_request:13029", + "pull_request:13150", + "pull_request:13332", + "pull_request:13365", + "pull_request:1508", + "pull_request:1582", + "pull_request:1583", + "pull_request:1910", + "pull_request:1926", + "pull_request:2345", + "pull_request:2347", + "pull_request:2398", + "pull_request:241", + "pull_request:2470", + "pull_request:2565", + "pull_request:2655", + "pull_request:2919", + "pull_request:3275", + "pull_request:3297", + "pull_request:3657", + "pull_request:3928", + "pull_request:402", + "pull_request:4188", + "pull_request:4214", + "pull_request:4223", + "pull_request:453", + "pull_request:4819", + "pull_request:4944", + "pull_request:5033", + "pull_request:51", + "pull_request:538", + "pull_request:5713", + "pull_request:630", + "pull_request:9269" + ], + "soft_pairs": [ + "pull_request:4819|pull_request:5033", + "pull_request:4944|pull_request:5713", + "pull_request:10132|pull_request:10494", + "pull_request:10200|pull_request:9269", + "pull_request:1038|pull_request:51", + "pull_request:10597|pull_request:10633", + "pull_request:1147|pull_request:630", + "pull_request:11504|pull_request:11728", + "pull_request:11613|pull_request:12425", + "pull_request:12363|pull_request:13029", + "pull_request:1269|pull_request:3297", + "pull_request:12973|pull_request:13365", + "pull_request:13150|pull_request:13332", + "pull_request:1508|pull_request:453", + "pull_request:1582|pull_request:538", + "pull_request:1583|pull_request:538", + "pull_request:1910|pull_request:402", + "pull_request:1926|pull_request:241", + "pull_request:2345|pull_request:3275", + "pull_request:2347|pull_request:2398", + "pull_request:2470|pull_request:4223", + "pull_request:2565|pull_request:2655", + "pull_request:2565|pull_request:402", + "pull_request:2919|pull_request:4188", + "pull_request:3657|pull_request:3928", + "pull_request:3928|pull_request:4214" + ], + "prepared_review_unit_hash": "9c30df62caa737b77a1493ae3778d3d7d6cc479e1186d297fc9a18351e1dc402", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 102064, + "estimated_input_tokens": 25516, + "estimated_eval_tokens": 51288 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All soft-edge candidates are false positives: they may touch related areas, but each pair targets a different bug, feature, or documentation/test update and should not be deduplicated.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:4819", + "right": "pull_request:5033", + "accept": false, + "reason": "Both are inpainting-related, but one adds support for 4-channel image inputs while the other fixes batch replication/mask handling. Different code-path issues." + }, + { + "left": "pull_request:4944", + "right": "pull_request:5713", + "accept": false, + "reason": "Same IP-Adapter theme, but 4944 is core loader/model support and 5713 is broader feature/docs integration across pipelines. Not the same concrete change." + }, + { + "left": "pull_request:10132", + "right": "pull_request:10494", + "accept": false, + "reason": "Both edit the CI workflow, but they fix different tokenizers install commands for different test contexts." + }, + { + "left": "pull_request:10200", + "right": "pull_request:9269", + "accept": false, + "reason": "Unrelated refactors in different subsystems: ControlNet Union vs CogVideoX 3D RoPE." + }, + { + "left": "pull_request:1038", + "right": "pull_request:51", + "accept": false, + "reason": "Flax dtype bugfix vs setup.py dependency typo; completely different changes." + }, + { + "left": "pull_request:10597", + "right": "pull_request:10633", + "accept": false, + "reason": "Both are docs, but for different pipelines and different features." + }, + { + "left": "pull_request:1147", + "right": "pull_request:630", + "accept": false, + "reason": "One fixes Flax example loading from subfolders; the other cleans up Flax from_pretrained mismatch handling. Different problems." + }, + { + "left": "pull_request:11504", + "right": "pull_request:11728", + "accept": false, + "reason": "Dependency/workflow update vs Dockerfile quant-lib additions. Same broad tooling area, but not the same bug or change." + }, + { + "left": "pull_request:11613", + "right": "pull_request:12425", + "accept": false, + "reason": "A revert of torch.compile CI/tests is unrelated to guarding distributed imports." + }, + { + "left": "pull_request:12363", + "right": "pull_request:13029", + "accept": false, + "reason": "QwenImageEdit Plus docs/tests vs Modular Diffusers quickstart rewrite; unrelated content." + }, + { + "left": "pull_request:1269", + "right": "pull_request:3297", + "accept": false, + "reason": "Stable Diffusion refactor follow-up vs retrieval-augmented diffusion model feature; different artifacts entirely." + }, + { + "left": "pull_request:12973", + "right": "pull_request:13365", + "accept": false, + "reason": "GLM-Image model support vs Claude review CI fork support. No overlap in underlying change." + }, + { + "left": "pull_request:13150", + "right": "pull_request:13332", + "accept": false, + "reason": "Test helper recompile-limit behavior vs dependency-testing workflow tightening; separate concerns." + }, + { + "left": "pull_request:1508", + "right": "pull_request:453", + "accept": false, + "reason": "DreamBooth docs clarification vs installation typo fix. Different documentation issues." + }, + { + "left": "pull_request:1582", + "right": "pull_request:538", + "accept": false, + "reason": "RePaint scheduler bugfix vs hub version bump. Unrelated." + }, + { + "left": "pull_request:1583", + "right": "pull_request:538", + "accept": false, + "reason": "Legacy inpaint pipeline fix vs hub version bump. Unrelated changes." + }, + { + "left": "pull_request:1910", + "right": "pull_request:402", + "accept": false, + "reason": "Korean docs initialization and toctree changes vs intro-section docs cleanup. Same docs area, but not the same change." + }, + { + "left": "pull_request:1926", + "right": "pull_request:241", + "accept": false, + "reason": "One is a large inpainting support/support-support PR, the other is an example script addition. Not mergeable as one duplicate change." + }, + { + "left": "pull_request:2345", + "right": "pull_request:3275", + "accept": false, + "reason": "xFormers warning in training scripts vs adding upsample_size to attention blocks. Different code paths." + }, + { + "left": "pull_request:2347", + "right": "pull_request:2398", + "accept": false, + "reason": "Scheduler noising-schedule changes vs Pix2Pix Zero docs updates; unrelated." + }, + { + "left": "pull_request:2470", + "right": "pull_request:4223", + "accept": false, + "reason": "MPS test skips vs ControlNet SDXL training fixes. Different problems." + }, + { + "left": "pull_request:2565", + "right": "pull_request:2655", + "accept": false, + "reason": "UNet1D docs correction vs LoRA save/load logic change. Different scope and behavior." + }, + { + "left": "pull_request:2919", + "right": "pull_request:4188", + "accept": false, + "reason": "Test speed-up changes vs SDXL ControlNet embedding-cache follow-up fixes. Not the same issue." + }, + { + "left": "pull_request:3657", + "right": "pull_request:3928", + "accept": false, + "reason": "ControlNet checkpoint parsing tweaks vs out-of-list runtime fix in ControlNet pipelines/schedulers. Different underlying bugs." + }, + { + "left": "pull_request:3928", + "right": "pull_request:4214", + "accept": false, + "reason": "ControlNet runtime fix vs bf16 error resolution in Custom Diffusion. Unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 189, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1006", + "pull_request:12086", + "pull_request:12117", + "pull_request:12418", + "pull_request:12519", + "pull_request:2476", + "pull_request:3652", + "pull_request:3681", + "pull_request:4070", + "pull_request:4099", + "pull_request:4245", + "pull_request:4595", + "pull_request:4887", + "pull_request:4987", + "pull_request:51", + "pull_request:5301", + "pull_request:5385", + "pull_request:5615", + "pull_request:5788", + "pull_request:6078", + "pull_request:6114", + "pull_request:6119", + "pull_request:6208", + "pull_request:6217", + "pull_request:6225", + "pull_request:6229", + "pull_request:6265", + "pull_request:6759", + "pull_request:6886", + "pull_request:6952", + "pull_request:7115", + "pull_request:7158", + "pull_request:7202", + "pull_request:7353", + "pull_request:7778", + "pull_request:7790", + "pull_request:816", + "pull_request:8303", + "pull_request:8378", + "pull_request:8389", + "pull_request:8452", + "pull_request:8510", + "pull_request:8564", + "pull_request:8702", + "pull_request:8973", + "pull_request:9323", + "pull_request:9569", + "pull_request:9896" + ], + "soft_pairs": [ + "pull_request:4070|pull_request:4245", + "pull_request:4099|pull_request:6078", + "pull_request:4595|pull_request:6114", + "pull_request:4887|pull_request:7115", + "pull_request:4987|pull_request:6265", + "pull_request:4987|pull_request:6886", + "pull_request:51|pull_request:816", + "pull_request:5385|pull_request:7202", + "pull_request:5615|pull_request:7790", + "pull_request:5788|pull_request:7158", + "pull_request:6119|pull_request:8378", + "pull_request:6208|pull_request:8510", + "pull_request:6217|pull_request:8303", + "pull_request:6229|pull_request:7778", + "pull_request:6759|pull_request:6952", + "pull_request:7158|pull_request:8389", + "pull_request:7353|pull_request:9323", + "pull_request:8452|pull_request:9569", + "pull_request:8510|pull_request:9569", + "pull_request:8564|pull_request:8702", + "pull_request:8973|pull_request:9896", + "pull_request:12086|pull_request:12117", + "pull_request:1006|pull_request:2476", + "pull_request:12418|pull_request:12519", + "pull_request:3652|pull_request:3681", + "pull_request:5301|pull_request:6225" + ], + "prepared_review_unit_hash": "7e9e4c8be89a19c42195b3c5efff1aa687743de3304105ee9028106f0b98cd7a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 94167, + "estimated_input_tokens": 23542, + "estimated_eval_tokens": 47340 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is a heterogeneous mix of unrelated PRs across docs, pipelines, training scripts, benchmarks, and utilities. The soft pairs mostly share broad subsystems or filenames, but they do not fix the same concrete bug/change, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:4070", + "right": "pull_request:4245", + "accept": false, + "reason": "Both are docs-only PRs, but one fixes SDXL doc links while the other adds safetensors examples across many docs; not the same change." + }, + { + "left": "pull_request:4099", + "right": "pull_request:6078", + "accept": false, + "reason": "Both touch model/pipeline internals, but one fixes SDXL prompt embedding handling and the other changes Euler scheduler sigma imports/ordering; different bugs." + }, + { + "left": "pull_request:4595", + "right": "pull_request:6114", + "accept": false, + "reason": "One is an unfinished XL converter scaffold; the other adds img2img/inpaint support to a community LPW script. No shared concrete fix." + }, + { + "left": "pull_request:4887", + "right": "pull_request:7115", + "accept": false, + "reason": "Kandinsky prior emb2emb latent removal and Dockerfile cleanup are unrelated code paths and goals." + }, + { + "left": "pull_request:4987", + "right": "pull_request:6265", + "accept": false, + "reason": "Both are import-structure refactors, but one removes a duplicate GLIGEN export and the other moves GLIGEN into its own package. Different scope." + }, + { + "left": "pull_request:4987", + "right": "pull_request:6886", + "accept": false, + "reason": "Duplicate GLIGEN import cleanup versus deprecation cleanup for `to(dtype=...)`; unrelated changes." + }, + { + "left": "pull_request:51", + "right": "pull_request:816", + "accept": false, + "reason": "A packaging pin fix and a training-loss dtype cast are not the same underlying issue." + }, + { + "left": "pull_request:5385", + "right": "pull_request:7202", + "accept": false, + "reason": "Both are CI/example maintenance, but one updates fix-copies logic for method signatures while the other removes a dependency from example requirements." + }, + { + "left": "pull_request:5615", + "right": "pull_request:7790", + "accept": false, + "reason": "Both concern SDXL-adjacent community code, but one reverts `load_lora_weights` behavior and the other adapts a masked img2img pipeline for SDXL." + }, + { + "left": "pull_request:5788", + "right": "pull_request:7158", + "accept": false, + "reason": "LCM scheduler error handling and benchmark workflow changes are unrelated; shared benchmark naming is incidental." + }, + { + "left": "pull_request:6119", + "right": "pull_request:8378", + "accept": false, + "reason": "One fixes LoRA alpha parsing in training scripts; the other fixes AsymmetricAutoencoderKL forward decode arguments. Different code paths and bugs." + }, + { + "left": "pull_request:6208", + "right": "pull_request:8510", + "accept": false, + "reason": "One fixes Stable Diffusion VAE dtype mismatch at decode time; the other changes LR scheduler math in DreamBooth LoRA training." + }, + { + "left": "pull_request:6217", + "right": "pull_request:8303", + "accept": false, + "reason": "Torch-compile VAE signature handling and a documentation typo fix are unrelated." + }, + { + "left": "pull_request:6229", + "right": "pull_request:7778", + "accept": false, + "reason": "CI dependency installation and a debugging workflow are both workflow-related, but they address different purposes and code paths." + }, + { + "left": "pull_request:6759", + "right": "pull_request:6952", + "accept": false, + "reason": "Both touch example/community code, but one fixes image embedding repetition in an InstantID example and the other rewrites textual inversion model card handling." + }, + { + "left": "pull_request:7158", + "right": "pull_request:8389", + "accept": false, + "reason": "Both involve benchmarking, but one fixes an LCM benchmark test and the other rewrites the benchmarking workflow/container/reporting. Not the same bug." + }, + { + "left": "pull_request:7353", + "right": "pull_request:9323", + "accept": false, + "reason": "One adds FromSingleFileMixin to SDXL ControlNet img2img; the other changes single-file default checkpoint repos. Different functionality." + }, + { + "left": "pull_request:8452", + "right": "pull_request:9569", + "accept": false, + "reason": "Single-file loading strictness and CogVideoX autoencoder decode batching are unrelated." + }, + { + "left": "pull_request:8510", + "right": "pull_request:9569", + "accept": false, + "reason": "DreamBooth LR scheduler fixes and CogVideoX decode fixes are different subsystems with no shared concrete bug." + }, + { + "left": "pull_request:8564", + "right": "pull_request:8702", + "accept": false, + "reason": "Both are SD3-related, but one adds T5 token-limit documentation/training support while the other adds an SD3 conversion script. Different changes." + }, + { + "left": "pull_request:8973", + "right": "pull_request:9896", + "accept": false, + "reason": "A test-size reduction and a Flux revert are unrelated; neither fixes the same code-path problem." + }, + { + "left": "pull_request:12086", + "right": "pull_request:12117", + "accept": false, + "reason": "Both add Flux/QwenImage modular pipeline support, but they target different features and filesets: Flux I2I modularization versus QwenImage img2img/inpaint pipelines." + }, + { + "left": "pull_request:1006", + "right": "pull_request:2476", + "accept": false, + "reason": "Both mention interpolation/image-size issues, but one fixes large-batch `F.interpolate()` behavior in resnet, while the other adds an `attn_res` parameter for Attend-and-Excite. Not the same bug." + }, + { + "left": "pull_request:12418", + "right": "pull_request:12519", + "accept": false, + "reason": "One xfails a failing test in CI; the other fixes a docs toctree sorting issue. Shared CI context only." + }, + { + "left": "pull_request:3652", + "right": "pull_request:3681", + "accept": false, + "reason": "Both involve checkpoint naming in training/example code, but one refactors checkpoint directory layout in tests/examples, while the other manually enforces checkpoint limits instead of relying on Accelerate." + }, + { + "left": "pull_request:5301", + "right": "pull_request:6225", + "accept": false, + "reason": "One fixes SDXL pipelines without text encoders; the other fixes LoRA alpha parsing across training scripts. Related area, but not the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 190, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10322", + "pull_request:10731", + "pull_request:1089", + "pull_request:10975", + "pull_request:11142", + "pull_request:11522", + "pull_request:11595", + "pull_request:11999", + "pull_request:12188", + "pull_request:12424", + "pull_request:12581", + "pull_request:12805", + "pull_request:12960", + "pull_request:1299", + "pull_request:12998", + "pull_request:130", + "pull_request:13056", + "pull_request:1386", + "pull_request:1968", + "pull_request:1977", + "pull_request:2279", + "pull_request:2314", + "pull_request:3723", + "pull_request:5038", + "pull_request:5111", + "pull_request:5697", + "pull_request:5842", + "pull_request:5974", + "pull_request:6241", + "pull_request:6311", + "pull_request:6332", + "pull_request:6365", + "pull_request:6673", + "pull_request:6787", + "pull_request:7206", + "pull_request:7507", + "pull_request:788", + "pull_request:7924", + "pull_request:7934", + "pull_request:8467", + "pull_request:8618", + "pull_request:8627", + "pull_request:8722", + "pull_request:9160", + "pull_request:9442", + "pull_request:9736", + "pull_request:978" + ], + "soft_pairs": [ + "pull_request:5842|pull_request:5974", + "pull_request:6241|pull_request:7924", + "pull_request:6787|pull_request:7507", + "pull_request:7934|pull_request:9442", + "pull_request:10322|pull_request:11142", + "pull_request:10731|pull_request:10975", + "pull_request:1089|pull_request:1386", + "pull_request:11142|pull_request:12188", + "pull_request:11522|pull_request:12424", + "pull_request:11595|pull_request:11999", + "pull_request:12424|pull_request:12960", + "pull_request:12581|pull_request:13056", + "pull_request:12805|pull_request:12998", + "pull_request:1299|pull_request:130", + "pull_request:1968|pull_request:2314", + "pull_request:1977|pull_request:2279", + "pull_request:2279|pull_request:3723", + "pull_request:5038|pull_request:5697", + "pull_request:5111|pull_request:6365", + "pull_request:6311|pull_request:8627", + "pull_request:6332|pull_request:8467", + "pull_request:6673|pull_request:6787", + "pull_request:7206|pull_request:9160", + "pull_request:788|pull_request:978", + "pull_request:8618|pull_request:9736", + "pull_request:8722|pull_request:9736" + ], + "prepared_review_unit_hash": "a584c62610311dfce2bf3579bd0eee266069c4986d715d3099ab7fce85e37c97", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 101768, + "estimated_input_tokens": 25442, + "estimated_eval_tokens": 51140 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly false-positive similarity matches: many pairs share a broad subsystem or even a filename, but they target different bugs, features, or cleanup changes. I don\u2019t see a single coherent duplicate PR/issue to canonicalize around.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:5842", + "right": "pull_request:5974", + "accept": false, + "reason": "Both touch PixArtAlpha, but one fixes 512px resolution binning while the other adapts the pipeline for pixart-lcm; different changes, not one bug." + }, + { + "left": "pull_request:6241", + "right": "pull_request:7924", + "accept": false, + "reason": "Unrelated failures: generic dtype mismatch runtime error vs IP-Adapter `added_cond_kwargs` handling in ControlNet inpaint." + }, + { + "left": "pull_request:6787", + "right": "pull_request:7507", + "accept": false, + "reason": "Different pipelines and problems: Marigold input normalization vs adding an InstantID SDXL img2img pipeline." + }, + { + "left": "pull_request:7934", + "right": "pull_request:9442", + "accept": false, + "reason": "Single-file config loading with dicts is unrelated to CI changes for deprecated attention block tests." + }, + { + "left": "pull_request:10322", + "right": "pull_request:11142", + "accept": false, + "reason": "Both modify the Sana converter, but they add support for different model families/variants and different scheduler setup changes." + }, + { + "left": "pull_request:10731", + "right": "pull_request:10975", + "accept": false, + "reason": "Type-hint/output cleanup across video pipelines is not the same as fixing Wan I2V pipeline bugs." + }, + { + "left": "pull_request:1089", + "right": "pull_request:1386", + "accept": false, + "reason": "Docs for swapping schedulers are unrelated to implementing v-prediction in schedulers." + }, + { + "left": "pull_request:11142", + "right": "pull_request:12188", + "accept": false, + "reason": "Sana Sprint converter updates do not match the Qwen-Image-Edit dimension bug fix." + }, + { + "left": "pull_request:11522", + "right": "pull_request:12424", + "accept": false, + "reason": "AudioLDM2 transformers-version compatibility is unrelated to Dockerfile/workflow definition cleanup." + }, + { + "left": "pull_request:11595", + "right": "pull_request:11999", + "accept": false, + "reason": "Both are LoRA-related, but one fixes Flux weight loading logic and the other adapts PEFT injection APIs across loaders and tests." + }, + { + "left": "pull_request:12424", + "right": "pull_request:12960", + "accept": false, + "reason": "Docker/workflow cleanup is unrelated to the `is_fsdp` detection fix in Flux2 training scripts." + }, + { + "left": "pull_request:12581", + "right": "pull_request:13056", + "accept": false, + "reason": "Both mention LoRA support, but they address different models and workflows: Qwen LoRA conversion vs Z-Image DreamBooth training." + }, + { + "left": "pull_request:12805", + "right": "pull_request:12998", + "accept": false, + "reason": "PyPI publish workflow changes are unrelated to docstring/style cleanup in several pipelines." + }, + { + "left": "pull_request:1299", + "right": "pull_request:130", + "accept": false, + "reason": "AltDiffusion feature addition and an examples README note about accelerate config are unrelated." + }, + { + "left": "pull_request:1968", + "right": "pull_request:2314", + "accept": false, + "reason": "CPU offload behavior fixes are unrelated to adjusting fast-test expected values." + }, + { + "left": "pull_request:1977", + "right": "pull_request:2279", + "accept": false, + "reason": "Training script seed/launch behavior and tooling migration from flake8 to ruff are different changes." + }, + { + "left": "pull_request:2279", + "right": "pull_request:3723", + "accept": false, + "reason": "Repository lint tooling changes are unrelated to the reference pipeline example update." + }, + { + "left": "pull_request:5038", + "right": "pull_request:5697", + "accept": false, + "reason": "Both involve LoRA scale handling, but they patch different code paths: transformer projection args vs attention processor/xformers backend behavior." + }, + { + "left": "pull_request:5111", + "right": "pull_request:6365", + "accept": false, + "reason": "Karras-sigma support and MPS dtype handling in KDPM2AncestralDiscreteScheduler are separate issues." + }, + { + "left": "pull_request:6311", + "right": "pull_request:8627", + "accept": false, + "reason": "These are different documentation additions for different feature areas; shared toctree edits are incidental." + }, + { + "left": "pull_request:6332", + "right": "pull_request:8467", + "accept": false, + "reason": "Rerender-A-Video community pipeline and SEGA pipeline fixes are unrelated community examples." + }, + { + "left": "pull_request:6673", + "right": "pull_request:6787", + "accept": false, + "reason": "InstantID pipeline addition and Marigold input-range correction are unrelated." + }, + { + "left": "pull_request:7206", + "right": "pull_request:9160", + "accept": false, + "reason": "Training script tags/metadata and Flux DreamBooth latent caching are not the same change." + }, + { + "left": "pull_request:788", + "right": "pull_request:978", + "accept": false, + "reason": "PIL resampling warning fixes are unrelated to small typo/documentation cleanup." + }, + { + "left": "pull_request:8618", + "right": "pull_request:9736", + "accept": false, + "reason": "SD3 tiny autoencoder shift_factor and Allegro T2V are different features with no shared bug." + }, + { + "left": "pull_request:8722", + "right": "pull_request:9736", + "accept": false, + "reason": "Large embeddings refactor and Allegro model addition only overlap on touched file names, not on the underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 191, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10354", + "pull_request:10497", + "pull_request:10557", + "pull_request:10894", + "pull_request:11067", + "pull_request:11243", + "pull_request:11975", + "pull_request:12021", + "pull_request:12095", + "pull_request:1229", + "pull_request:12416", + "pull_request:12424", + "pull_request:12519", + "pull_request:12960", + "pull_request:13126", + "pull_request:13321", + "pull_request:13338", + "pull_request:13364", + "pull_request:1458", + "pull_request:1582", + "pull_request:1583", + "pull_request:1913", + "pull_request:1942", + "pull_request:2001", + "pull_request:2655", + "pull_request:3034", + "pull_request:3128", + "pull_request:3294", + "pull_request:3446", + "pull_request:3590", + "pull_request:3673", + "pull_request:3681", + "pull_request:4062", + "pull_request:4262", + "pull_request:4690", + "pull_request:4706", + "pull_request:4744", + "pull_request:4793", + "pull_request:5393", + "pull_request:5438", + "pull_request:5807", + "pull_request:676", + "pull_request:6994", + "pull_request:823", + "pull_request:9175", + "pull_request:9696", + "pull_request:9807", + "pull_request:9816" + ], + "soft_pairs": [ + "pull_request:9175|pull_request:9816", + "pull_request:12416|pull_request:13126", + "pull_request:12519|pull_request:13321", + "pull_request:4062|pull_request:5438", + "pull_request:12021|pull_request:12095", + "pull_request:10354|pull_request:9696", + "pull_request:10497|pull_request:11067", + "pull_request:10557|pull_request:9696", + "pull_request:10894|pull_request:11975", + "pull_request:11243|pull_request:9807", + "pull_request:1229|pull_request:1458", + "pull_request:12424|pull_request:13338", + "pull_request:12960|pull_request:13364", + "pull_request:1582|pull_request:676", + "pull_request:1583|pull_request:676", + "pull_request:1913|pull_request:823", + "pull_request:1942|pull_request:2001", + "pull_request:2655|pull_request:3294", + "pull_request:3034|pull_request:3446", + "pull_request:3128|pull_request:3673", + "pull_request:3590|pull_request:5807", + "pull_request:3681|pull_request:4690", + "pull_request:4262|pull_request:4706", + "pull_request:4744|pull_request:5393", + "pull_request:4793|pull_request:6994" + ], + "prepared_review_unit_hash": "58dbe707b5866d8fe3958db77bd0474d134995b28b287975a224293a3481027d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 96357, + "estimated_input_tokens": 24090, + "estimated_eval_tokens": 48436 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of unrelated PRs across docs, CI, loaders, models, and pipelines. The soft-edge pairs mostly share only a broad subsystem or keyword (e.g. Flux, LoRA, LTX) but not the same concrete bug or change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 9816 is the strongest standalone representative because it is a large, self-contained feature addition with clear scope and substantial review activity; the rest are mostly small, unrelated fixes or infra/docs changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9175", + "right": "pull_request:9816", + "accept": false, + "reason": "Both are Flux-related, but one adds Union ControlNet support and the other adds an RF-inversion Flux pipeline; different code paths and user-facing features." + }, + { + "left": "pull_request:12416", + "right": "pull_request:13126", + "accept": false, + "reason": "Both involve model-adjacent modular/LoRA work, but they target different projects and problems: QwenImage Edit Plus vs Z-Image distilled LoRAs." + }, + { + "left": "pull_request:12519", + "right": "pull_request:13321", + "accept": false, + "reason": "One is a docs toctree sorting cleanup, the other fixes an unguarded torchvision import in Cosmos; unrelated changes." + }, + { + "left": "pull_request:4062", + "right": "pull_request:5438", + "accept": false, + "reason": "Shap-E mesh output support and Latent Consistency support are different features with no shared bug or mergeable unified patch." + }, + { + "left": "pull_request:12021", + "right": "pull_request:12095", + "accept": false, + "reason": "Both touch LTX, but one adds attention backend support in core code while the other introduces a new LTX Video model release and pipeline/docs updates." + }, + { + "left": "pull_request:10354", + "right": "pull_request:9696", + "accept": false, + "reason": "A Flux IP-Adapter tester fix and a workflow input-name update are unrelated." + }, + { + "left": "pull_request:10497", + "right": "pull_request:11067", + "accept": false, + "reason": "Both are CI-related, but one adjusts tokenizers installation and the other pins transformers for benchmarking; different workflows and problems." + }, + { + "left": "pull_request:10557", + "right": "pull_request:9696", + "accept": false, + "reason": "One fixes StableDiffusion single-file slow tests; the other changes a PR test workflow. No shared underlying bug." + }, + { + "left": "pull_request:10894", + "right": "pull_request:11975", + "accept": false, + "reason": "Docs toctree sorting and a style-only training script edit are not the same change." + }, + { + "left": "pull_request:11243", + "right": "pull_request:9807", + "accept": false, + "reason": "Both concern LoRA, but they fix different conversion/loading paths for different model families; not the same concrete bug." + }, + { + "left": "pull_request:1229", + "right": "pull_request:1458", + "accept": false, + "reason": "A README install note and an ONNX runtime candidate list update are unrelated." + }, + { + "left": "pull_request:12424", + "right": "pull_request:13338", + "accept": false, + "reason": "Dockerfile cleanup and adding id-token permissions to a workflow are separate infra fixes." + }, + { + "left": "pull_request:12960", + "right": "pull_request:13364", + "accept": false, + "reason": "DreamBooth FSDP detection and MotionConv2d dtype casting are different code paths and bugs." + }, + { + "left": "pull_request:1582", + "right": "pull_request:676", + "accept": false, + "reason": "RePaint scheduler logic and a variable rename in resnet are unrelated." + }, + { + "left": "pull_request:1583", + "right": "pull_request:676", + "accept": false, + "reason": "A Stable Diffusion inpaint timestep fix and a resnet variable rename are not the same issue." + }, + { + "left": "pull_request:1913", + "right": "pull_request:823", + "accept": false, + "reason": "Input-type validation in img2img pipelines and negative-prompt type checking are distinct validation fixes." + }, + { + "left": "pull_request:1942", + "right": "pull_request:2001", + "accept": false, + "reason": "A conversion-script flag rename and a ColossalAI seed-launch fix are unrelated." + }, + { + "left": "pull_request:2655", + "right": "pull_request:3294", + "accept": false, + "reason": "Both are LoRA loader changes, but one corrects loading/saving behavior while the other adds support for a new 'lora_' prefix; they are related only at a broad subsystem level." + }, + { + "left": "pull_request:3034", + "right": "pull_request:3446", + "accept": false, + "reason": "Wrong accelerate parameter name vs a warning-message wording fix; different concrete bugs." + }, + { + "left": "pull_request:3128", + "right": "pull_request:3673", + "accept": false, + "reason": "Img2img timestep selection and multi-GPU DreamBooth behavior are unrelated." + }, + { + "left": "pull_request:3590", + "right": "pull_request:5807", + "accept": false, + "reason": "StableDiffusionCanvasPipeline support and lpw SDXL prompt-embedding duplication are separate community-pipeline features." + }, + { + "left": "pull_request:3681", + "right": "pull_request:4690", + "accept": false, + "reason": "Checkpoint limit handling in training scripts and a DPMSolver scheduler refactor are unrelated." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4706", + "accept": false, + "reason": "An SDXL DreamBooth LoRA training patch set and a missing-default fix for fc alphas are different LoRA issues." + }, + { + "left": "pull_request:4744", + "right": "pull_request:5393", + "accept": false, + "reason": "A docstring correction in AudioLDM and a training/docs API update are not the same change." + }, + { + "left": "pull_request:4793", + "right": "pull_request:6994", + "accept": false, + "reason": "AutoPipeline kwargs passthrough and passing torch_dtype in from_single_file are different loader fixes with no shared failing code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 192, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10122", + "pull_request:10126", + "pull_request:10140", + "pull_request:10188", + "pull_request:10326", + "pull_request:10358", + "pull_request:10431", + "pull_request:10595", + "pull_request:10723", + "pull_request:11916", + "pull_request:11967", + "pull_request:12109", + "pull_request:12226", + "pull_request:12301", + "pull_request:12490", + "pull_request:12534", + "pull_request:12549", + "pull_request:12736", + "pull_request:12919", + "pull_request:13321", + "pull_request:13490", + "pull_request:2193", + "pull_request:2452", + "pull_request:2902", + "pull_request:3587", + "pull_request:3770", + "pull_request:4499", + "pull_request:4500", + "pull_request:4501", + "pull_request:4502", + "pull_request:5033", + "pull_request:5301", + "pull_request:5559", + "pull_request:5697", + "pull_request:5898", + "pull_request:6217", + "pull_request:7062", + "pull_request:7410", + "pull_request:7461", + "pull_request:7714", + "pull_request:7720", + "pull_request:8691", + "pull_request:8897", + "pull_request:9247", + "pull_request:9257", + "pull_request:9311", + "pull_request:9592", + "pull_request:9723" + ], + "soft_pairs": [ + "pull_request:5697|pull_request:7461", + "pull_request:5898|pull_request:7714", + "pull_request:7062|pull_request:7461", + "pull_request:7410|pull_request:7720", + "pull_request:9247|pull_request:9723", + "pull_request:12534|pull_request:13490", + "pull_request:11916|pull_request:11967", + "pull_request:12549|pull_request:12919", + "pull_request:10723|pull_request:9311", + "pull_request:12226|pull_request:12736", + "pull_request:12490|pull_request:13321", + "pull_request:2193|pull_request:2452", + "pull_request:3587|pull_request:5559", + "pull_request:3770|pull_request:4499", + "pull_request:3770|pull_request:4500", + "pull_request:3770|pull_request:4501", + "pull_request:3770|pull_request:4502", + "pull_request:5301|pull_request:6217", + "pull_request:10140|pull_request:8897", + "pull_request:12109|pull_request:12301", + "pull_request:2902|pull_request:5033", + "pull_request:10122|pull_request:10188", + "pull_request:10126|pull_request:9257", + "pull_request:10326|pull_request:8691", + "pull_request:10358|pull_request:9592", + "pull_request:10431|pull_request:10595" + ], + "prepared_review_unit_hash": "cce8983c8de946dac4aa5c05c9a52dde0e1b01458b4049019bd799221d2af53a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 101778, + "estimated_input_tokens": 25445, + "estimated_eval_tokens": 51146 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a set of mostly unrelated pull requests; the soft-similarity signals are broad topical overlaps, not true duplicates or mergeable variants of the same fix.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:5697", + "right": "pull_request:7461", + "accept": false, + "reason": "Completely different areas: LoRA/xformers attention bug vs a testing-utils helper definition fix." + }, + { + "left": "pull_request:5898", + "right": "pull_request:7714", + "accept": false, + "reason": "Both are docs-only, but they document different features and would not be merged as the same change." + }, + { + "left": "pull_request:7062", + "right": "pull_request:7461", + "accept": false, + "reason": "Imagic docstring cleanup is unrelated to testing-utils boolean parsing." + }, + { + "left": "pull_request:7410", + "right": "pull_request:7720", + "accept": false, + "reason": "FreeInit behavior fix vs dead-code/f-string cleanup; different code paths and bugs." + }, + { + "left": "pull_request:9247", + "right": "pull_request:9723", + "accept": false, + "reason": "Different fixes in different modules: model variant detection vs Matryoshka schedule_shifted_power usage." + }, + { + "left": "pull_request:12534", + "right": "pull_request:13490", + "accept": false, + "reason": "Both touch NPU-related code, but one fixes Flux rotary embeddings on Ascend and the other adjusts NPU attention mask shapes; separate problems." + }, + { + "left": "pull_request:11916", + "right": "pull_request:11967", + "accept": false, + "reason": "Shared attention-dispatch area, but one is a large refactor and the other changes runtime backend warning/error behavior." + }, + { + "left": "pull_request:12549", + "right": "pull_request:12919", + "accept": false, + "reason": "Different backends/platforms: AITER backend support vs NPU attention Ulysses enablement." + }, + { + "left": "pull_request:10723", + "right": "pull_request:9311", + "accept": false, + "reason": "Both are ControlNet-related, but one adds per-control scaling and the other adds CFG cutoff; distinct features." + }, + { + "left": "pull_request:12226", + "right": "pull_request:12736", + "accept": false, + "reason": "A fast pipeline test and a regional compilation flag for another model are unrelated." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13321", + "accept": false, + "reason": "Test fixture correction for AudioLDM2 is unrelated to a guarded torchvision import in Cosmos." + }, + { + "left": "pull_request:2193", + "right": "pull_request:2452", + "accept": false, + "reason": "Both are docs additions, but they cover entirely different documentation sections and topics." + }, + { + "left": "pull_request:3587", + "right": "pull_request:5559", + "accept": false, + "reason": "Community README update vs adding a latent-consistency img2img pipeline; not the same artifact." + }, + { + "left": "pull_request:3770", + "right": "pull_request:4499", + "accept": false, + "reason": "ControlNet guidance start/stop and SDXL reference-attn are different pipeline features." + }, + { + "left": "pull_request:3770", + "right": "pull_request:4500", + "accept": false, + "reason": "ControlNet guidance start/stop and SDXL reference-attn are different pipeline features." + }, + { + "left": "pull_request:3770", + "right": "pull_request:4501", + "accept": false, + "reason": "ControlNet guidance start/stop and SDXL reference-attn are different pipeline features." + }, + { + "left": "pull_request:3770", + "right": "pull_request:4502", + "accept": false, + "reason": "ControlNet guidance start/stop and SDXL reference-attn are different pipeline features." + }, + { + "left": "pull_request:5301", + "right": "pull_request:6217", + "accept": false, + "reason": "SDXL text-encoder handling and Stable Video Diffusion torch-compile support are unrelated." + }, + { + "left": "pull_request:10122", + "right": "pull_request:10188", + "accept": false, + "reason": "Flux attention-mask plumbing vs Flux LoRA-loading error handling are separate concerns." + }, + { + "left": "pull_request:10126", + "right": "pull_request:9257", + "accept": false, + "reason": "LoRA save_attn_procs deprecation is unrelated to a DreamBooth Flux training-script fix." + }, + { + "left": "pull_request:10326", + "right": "pull_request:8691", + "accept": false, + "reason": "MPS workflow quoting fix vs LoRA loading behavior; no common bug." + }, + { + "left": "pull_request:10358", + "right": "pull_request:9592", + "accept": false, + "reason": "Community projects docs update and Flax docstring refactor are unrelated docs changes." + }, + { + "left": "pull_request:10431", + "right": "pull_request:10595", + "accept": false, + "reason": "Different model-specific fixes: Sana positional embedding initialization vs bf16 casting in quadratic attention." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 193, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10539", + "pull_request:10704", + "pull_request:11025", + "pull_request:11059", + "pull_request:11066", + "pull_request:11067", + "pull_request:11236", + "pull_request:11325", + "pull_request:11614", + "pull_request:11743", + "pull_request:11767", + "pull_request:11770", + "pull_request:11786", + "pull_request:11814", + "pull_request:11892", + "pull_request:11894", + "pull_request:12005", + "pull_request:12137", + "pull_request:1215", + "pull_request:12231", + "pull_request:1230", + "pull_request:12363", + "pull_request:12425", + "pull_request:12527", + "pull_request:12530", + "pull_request:12586", + "pull_request:12805", + "pull_request:12998", + "pull_request:1311", + "pull_request:13134", + "pull_request:13472", + "pull_request:1360", + "pull_request:1425", + "pull_request:1625", + "pull_request:171", + "pull_request:1780", + "pull_request:1849", + "pull_request:1968", + "pull_request:1978", + "pull_request:227", + "pull_request:2386", + "pull_request:2640", + "pull_request:2715", + "pull_request:796", + "pull_request:8932", + "pull_request:9103", + "pull_request:9688", + "pull_request:9769" + ], + "soft_pairs": [ + "pull_request:10539|pull_request:8932", + "pull_request:10704|pull_request:9103", + "pull_request:11025|pull_request:11066", + "pull_request:11059|pull_request:9769", + "pull_request:11067|pull_request:11614", + "pull_request:11236|pull_request:9688", + "pull_request:11325|pull_request:11770", + "pull_request:11743|pull_request:11814", + "pull_request:11767|pull_request:12005", + "pull_request:11786|pull_request:11894", + "pull_request:11892|pull_request:12805", + "pull_request:12137|pull_request:12998", + "pull_request:1215|pull_request:1968", + "pull_request:12231|pull_request:12527", + "pull_request:1230|pull_request:796", + "pull_request:12363|pull_request:12530", + "pull_request:12425|pull_request:13134", + "pull_request:12586|pull_request:13472", + "pull_request:1311|pull_request:227", + "pull_request:1360|pull_request:1978", + "pull_request:1425|pull_request:2386", + "pull_request:1625|pull_request:2640", + "pull_request:171|pull_request:1849", + "pull_request:1780|pull_request:2715" + ], + "prepared_review_unit_hash": "350fe8614020bb454791f03aeb7d1c89332d4429fe311faddf50b60f7176aaf6", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 97581, + "estimated_input_tokens": 24396, + "estimated_eval_tokens": 49048 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft-similarity pairs are mostly superficial matches on keywords or broad subsystems (Flux, docs, CI, tests), but they do not describe the same underlying bug or change. No duplicate pair is convincing here.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10539", + "right": "pull_request:8932", + "accept": false, + "reason": "Flux CFG logic fix vs. Temporal Transformer test coverage; unrelated changes." + }, + { + "left": "pull_request:10704", + "right": "pull_request:9103", + "accept": false, + "reason": "Documentation edit for img2img vs. distributed LR scheduler fix; different code paths and goals." + }, + { + "left": "pull_request:11025", + "right": "pull_request:11066", + "accept": false, + "reason": "Wan LoRA conversion support vs. new HunyuanVideo I2V model support; separate features." + }, + { + "left": "pull_request:11059", + "right": "pull_request:9769", + "accept": false, + "reason": "Wan LoRA conversion tweak vs. Mochi model/pipeline addition; not the same change." + }, + { + "left": "pull_request:11067", + "right": "pull_request:11614", + "accept": false, + "reason": "Benchmark transformers pinning vs. docker CI security/workflow fix; unrelated CI changes." + }, + { + "left": "pull_request:11236", + "right": "pull_request:9688", + "accept": false, + "reason": "Two different documentation updates for different pipelines; no shared underlying bug or feature." + }, + { + "left": "pull_request:11325", + "right": "pull_request:11770", + "accept": false, + "reason": "HiDream docstring decorator fix vs. LoRA docs cleanup; both docs-oriented but not the same issue." + }, + { + "left": "pull_request:11743", + "right": "pull_request:11814", + "accept": false, + "reason": "Research training script addition vs. Flux gradient-checkpointing bug fix; unrelated." + }, + { + "left": "pull_request:11767", + "right": "pull_request:12005", + "accept": false, + "reason": "Shared modeling tests file, but one fixes group-offloading disk addressing and the other sharded checkpoint local-only behavior." + }, + { + "left": "pull_request:11786", + "right": "pull_request:11894", + "accept": false, + "reason": "Big-GPU test marker correction vs. k-diffusion dependency pin; different subsystems." + }, + { + "left": "pull_request:11892", + "right": "pull_request:12805", + "accept": false, + "reason": "Removing k-diffusion from core deps vs. PyPI publish workflow edits; not the same change." + }, + { + "left": "pull_request:12137", + "right": "pull_request:12998", + "accept": false, + "reason": "Parallel-loading constant refactor vs. broad style/docstring cleanup; unrelated." + }, + { + "left": "pull_request:1215", + "right": "pull_request:1968", + "accept": false, + "reason": "DDIM eta/cpu-generator bug fix vs. CPU offload doc corrections; different artifact types and issues." + }, + { + "left": "pull_request:12231", + "right": "pull_request:12527", + "accept": false, + "reason": "Typo fix in example docs vs. Kandinsky5 CFG bug fix; no overlap beyond generic cleanup." + }, + { + "left": "pull_request:1230", + "right": "pull_request:796", + "accept": false, + "reason": "MPS fast-test fix vs. adding Apple M1 tests; related platform, but not the same concrete bug/change." + }, + { + "left": "pull_request:12363", + "right": "pull_request:12530", + "accept": false, + "reason": "QwenImageEdit Plus docs/test addition vs. Stable Cascade test xfail; unrelated." + }, + { + "left": "pull_request:12425", + "right": "pull_request:13134", + "accept": false, + "reason": "torch.distributed import guard vs. Cosmos transformer typing fix; different code areas." + }, + { + "left": "pull_request:12586", + "right": "pull_request:13472", + "accept": false, + "reason": "Documentation ordering change vs. agents docs float64 guidance; unrelated docs changes." + }, + { + "left": "pull_request:1311", + "right": "pull_request:227", + "accept": false, + "reason": "AltDiffusion test additions vs. scheduler add_noise implementation; unrelated." + }, + { + "left": "pull_request:1360", + "right": "pull_request:1978", + "accept": false, + "reason": "k-diffusion community pipeline docs/code vs. push_to_hub implementation work; different features." + }, + { + "left": "pull_request:1425", + "right": "pull_request:2386", + "accept": false, + "reason": "Simple docs typo fixes vs. UniPC docs rewrite; not the same underlying issue." + }, + { + "left": "pull_request:1625", + "right": "pull_request:2640", + "accept": false, + "reason": "k-diffusion pipeline sizing fix vs. telemetry removal in pipeline loading; unrelated." + }, + { + "left": "pull_request:171", + "right": "pull_request:1849", + "accept": false, + "reason": "LDM eta-condition fix vs. cross_attention_kwargs support in a different attention module; not duplicates." + }, + { + "left": "pull_request:1780", + "right": "pull_request:2715", + "accept": false, + "reason": "Upscale preprocessing comment fix vs. broad typo cleanup; superficial text similarity only." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 194, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1850", + "pull_request:2122", + "pull_request:2126", + "pull_request:2365", + "pull_request:2529", + "pull_request:2530", + "pull_request:2553", + "pull_request:2598", + "pull_request:2604", + "pull_request:2693", + "pull_request:2757", + "pull_request:2861", + "pull_request:2968", + "pull_request:3127", + "pull_request:327", + "pull_request:3473", + "pull_request:3572", + "pull_request:3607", + "pull_request:3642", + "pull_request:3644", + "pull_request:3797", + "pull_request:386", + "pull_request:3942", + "pull_request:3957", + "pull_request:3989", + "pull_request:3995", + "pull_request:400", + "pull_request:401", + "pull_request:402", + "pull_request:4058", + "pull_request:4106", + "pull_request:4203", + "pull_request:4285", + "pull_request:4442", + "pull_request:5036", + "pull_request:5098", + "pull_request:5276", + "pull_request:5369", + "pull_request:5396", + "pull_request:5526", + "pull_request:5699", + "pull_request:5724", + "pull_request:5738", + "pull_request:5844", + "pull_request:628", + "pull_request:796", + "pull_request:819" + ], + "soft_pairs": [ + "pull_request:1850|pull_request:2861", + "pull_request:2122|pull_request:402", + "pull_request:2126|pull_request:2693", + "pull_request:2365|pull_request:327", + "pull_request:2529|pull_request:2757", + "pull_request:2530|pull_request:796", + "pull_request:2553|pull_request:3642", + "pull_request:2598|pull_request:628", + "pull_request:2604|pull_request:628", + "pull_request:2968|pull_request:819", + "pull_request:3127|pull_request:5396", + "pull_request:3473|pull_request:4442", + "pull_request:3572|pull_request:5036", + "pull_request:3607|pull_request:5526", + "pull_request:3644|pull_request:5844", + "pull_request:3797|pull_request:3957", + "pull_request:386|pull_request:401", + "pull_request:3942|pull_request:5098", + "pull_request:3942|pull_request:5276", + "pull_request:3942|pull_request:5724", + "pull_request:3989|pull_request:5699", + "pull_request:3995|pull_request:4058", + "pull_request:3995|pull_request:4285", + "pull_request:400|pull_request:401", + "pull_request:4106|pull_request:5738", + "pull_request:4203|pull_request:5369" + ], + "prepared_review_unit_hash": "b6d2e0fb0a29c9420f49ecfb1b0f9d99754908a0d4af32894eb6836314e300c4", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 98789, + "estimated_input_tokens": 24698, + "estimated_eval_tokens": 49652 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Almost all soft-similarity pairs are false positives across unrelated docs/tests/maintenance changes. The only plausible mergeable duplicate is the attention-naming pair `3797`/`3957`, which both appear to address the same core model attention-head naming correction.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 3797 is the strongest representative PR: it is merged and makes a broad attention-head naming correction across core model files. It should not be described as explicitly issue-linked, because the packet shows `explicit_issue_links: 0`.", + "best_issue_reason": null, + "best_pr_reason": "PR 3797 is the best standalone PR in this set because it is a concrete core-model bug fix affecting many attention-related files, whereas the other PRs are mostly docs/tests or narrower maintenance changes. Its standalone strength does not rely on any explicit issue link.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1850", + "right": "pull_request:2861", + "accept": false, + "reason": "Different test additions for different features: ResnetBlock2D coverage vs StableUnCLIP `image_embeds=None` handling." + }, + { + "left": "pull_request:2122", + "right": "pull_request:402", + "accept": false, + "reason": "Both are docs-related, but they cover unrelated topics: safetensors usage vs a broad docs intro/toctree overhaul." + }, + { + "left": "pull_request:2126", + "right": "pull_request:2693", + "accept": false, + "reason": "Unrelated doc fixes: duplicate `torch_dtype` docstring cleanup vs an inpaint image-link correction." + }, + { + "left": "pull_request:2365", + "right": "pull_request:327", + "accept": false, + "reason": "A scheduler enum note is not the same change as adding type hints to the PNDM pipeline." + }, + { + "left": "pull_request:2529", + "right": "pull_request:2757", + "accept": false, + "reason": "Different code paths: OpenCLIP checkpoint conversion fallback vs depth-map batching in depth2img." + }, + { + "left": "pull_request:2530", + "right": "pull_request:796", + "accept": false, + "reason": "Both add tests, but they target different features and failure modes; not the same bug fix." + }, + { + "left": "pull_request:2553", + "right": "pull_request:3642", + "accept": false, + "reason": "CFG flag handling in Flax Stable Diffusion is unrelated to adding TensorRT inpainting pipeline support." + }, + { + "left": "pull_request:2598", + "right": "pull_request:628", + "accept": false, + "reason": "Separate documentation edits for different pages/features; no shared underlying change." + }, + { + "left": "pull_request:2604", + "right": "pull_request:628", + "accept": false, + "reason": "A colab-link fix in the index is unrelated to the DDPM colab placement change." + }, + { + "left": "pull_request:2968", + "right": "pull_request:819", + "accept": false, + "reason": "Different bugs in different code paths: img2img post-processing vs DDIM timestep dtype on Windows." + }, + { + "left": "pull_request:3127", + "right": "pull_request:5396", + "accept": false, + "reason": "Safety checker tensor handling is unrelated to the stale bot timezone datetime fix." + }, + { + "left": "pull_request:3473", + "right": "pull_request:4442", + "accept": false, + "reason": "Both add docs, but for different model features and tutorial pages; not one underlying change." + }, + { + "left": "pull_request:3572", + "right": "pull_request:5036", + "accept": false, + "reason": "ControlNet inpaint docs and import/dummy-object plumbing are unrelated maintenance changes." + }, + { + "left": "pull_request:3607", + "right": "pull_request:5526", + "accept": false, + "reason": "Core attention processor changes are unrelated to the second-order scheduler logic fix." + }, + { + "left": "pull_request:3644", + "right": "pull_request:5844", + "accept": false, + "reason": "Both are docs updates, but they document different examples and features." + }, + { + "left": "pull_request:3797", + "right": "pull_request:3957", + "accept": true, + "reason": "Both concern the same attention-head naming/correction work in the same UNet model code paths, and `3957` looks like a plausible follow-up piece of the same fix." + }, + { + "left": "pull_request:386", + "right": "pull_request:401", + "accept": false, + "reason": "Different pipeline docs and code: Stable Diffusion pipeline overview vs Karras VE docs." + }, + { + "left": "pull_request:3942", + "right": "pull_request:5098", + "accept": false, + "reason": "A test harness fix for np outputs is unrelated to Kandinsky precision test coverage." + }, + { + "left": "pull_request:3942", + "right": "pull_request:5276", + "accept": false, + "reason": "Test-output handling and scheduler 1D support are unrelated changes." + }, + { + "left": "pull_request:3942", + "right": "pull_request:5724", + "accept": false, + "reason": "A common test helper change does not match a LoRA backend test/decorator change." + }, + { + "left": "pull_request:3989", + "right": "pull_request:5699", + "accept": false, + "reason": "Slow-test fixes in SDXL do not share the same underlying change as a styling-only stable unclip/text-unet cleanup." + }, + { + "left": "pull_request:3995", + "right": "pull_request:4058", + "accept": false, + "reason": "SDXL prompt-embedding plumbing is unrelated to a docs image/table update." + }, + { + "left": "pull_request:3995", + "right": "pull_request:4285", + "accept": false, + "reason": "Core SDXL pipeline changes are unrelated to ONNX/OpenVINO documentation work." + }, + { + "left": "pull_request:400", + "right": "pull_request:401", + "accept": false, + "reason": "They are different docs pages for different pipelines; no shared concrete change." + }, + { + "left": "pull_request:4106", + "right": "pull_request:5738", + "accept": false, + "reason": "A new Tortoise TTS pipeline and LoRA adapter deletion are entirely separate features." + }, + { + "left": "pull_request:4203", + "right": "pull_request:5369", + "accept": false, + "reason": "DreamBooth README typo cleanup and general docs minor fixes are separate doc edits, not one fix." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The evaluation is mostly conservative, but it overstates the evidence for PR 3797. In the packet, PR 3797 is merged and clearly about the attention naming fix, but it does **not** have an explicit issue link (`explicit_issue_links: 0`), so calling it \u201cexplicitly issue-linked\u201d and citing \u201cclearest issue context\u201d is unsupported. The borderline 3797\u21943957 pair is plausibly the same attention-name correction, but the standalone PR reasoning needs to stay grounded in the packet." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 195, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4211", + "pull_request:4347", + "pull_request:4397", + "pull_request:4499", + "pull_request:4500", + "pull_request:4811", + "pull_request:4856", + "pull_request:5097", + "pull_request:5164", + "pull_request:5275", + "pull_request:5369", + "pull_request:5384", + "pull_request:5391", + "pull_request:5420", + "pull_request:5526", + "pull_request:5602", + "pull_request:5653", + "pull_request:5698", + "pull_request:5741", + "pull_request:5831", + "pull_request:5982", + "pull_request:6179", + "pull_request:6198", + "pull_request:6221", + "pull_request:6305", + "pull_request:6308", + "pull_request:6360", + "pull_request:6365", + "pull_request:6369", + "pull_request:6498", + "pull_request:6564", + "pull_request:6997", + "pull_request:6998", + "pull_request:7247", + "pull_request:7260", + "pull_request:7328", + "pull_request:7353", + "pull_request:7371", + "pull_request:7473", + "pull_request:7857", + "pull_request:8298", + "pull_request:8404", + "pull_request:8452", + "pull_request:9242", + "pull_request:9247", + "pull_request:9426", + "pull_request:9718" + ], + "soft_pairs": [ + "pull_request:4211|pull_request:6198", + "pull_request:4347|pull_request:5275", + "pull_request:4397|pull_request:5391", + "pull_request:4499|pull_request:5384", + "pull_request:4500|pull_request:5384", + "pull_request:4811|pull_request:6360", + "pull_request:4856|pull_request:6365", + "pull_request:5097|pull_request:6221", + "pull_request:5164|pull_request:7328", + "pull_request:5369|pull_request:5831", + "pull_request:5420|pull_request:5741", + "pull_request:5526|pull_request:6179", + "pull_request:5602|pull_request:5982", + "pull_request:5653|pull_request:7371", + "pull_request:5698|pull_request:7473", + "pull_request:6305|pull_request:6997", + "pull_request:6308|pull_request:6369", + "pull_request:6498|pull_request:7857", + "pull_request:6564|pull_request:8404", + "pull_request:6998|pull_request:7260", + "pull_request:7247|pull_request:9247", + "pull_request:7353|pull_request:9426", + "pull_request:7857|pull_request:9242", + "pull_request:7857|pull_request:9718", + "pull_request:8298|pull_request:8452" + ], + "prepared_review_unit_hash": "fdb82ad4f1c62775f8b14ec3f684621b99fe52398142a7344e614e2a1ed0bd71", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 98415, + "estimated_input_tokens": 24604, + "estimated_eval_tokens": 49464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a bundle of unrelated PRs spanning docs, schedulers, pipelines, tests, and CI. The only clear duplicate pair is the SDXL reference-attention PR submission. Everything else should stay separate.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a good canonical for the whole cluster because the items are unrelated; if forced to pick one representative, the near-identical SDXL reference-attention PR (#4499) is the clearest duplicate anchor.", + "best_issue_reason": null, + "best_pr_reason": "PR #4499 is the strongest canonical candidate only because it has an exact twin (#4500) with the same title, files, and diff intent. Outside that pair, the cluster is too heterogeneous for a meaningful global canonical PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4211", + "right": "pull_request:6198", + "accept": false, + "reason": "Different fixes: docstring wording for SDXL inpainting vs benchmarking/reporting code. No shared bug or change path." + }, + { + "left": "pull_request:4347", + "right": "pull_request:5275", + "accept": false, + "reason": "A scheduler step-index change and adding an SDXL research example are unrelated." + }, + { + "left": "pull_request:4397", + "right": "pull_request:5391", + "accept": false, + "reason": "Both touch docs, but one is an SDXL docstring tweak and the other is broad typehint/docs cleanup across models; not the same change." + }, + { + "left": "pull_request:4499", + "right": "pull_request:5384", + "accept": false, + "reason": "SDXL reference-attn/adain support is a feature PR; hypernetwork support is a different loader/feature area." + }, + { + "left": "pull_request:4500", + "right": "pull_request:5384", + "accept": false, + "reason": "Same as above: unrelated features in different code paths." + }, + { + "left": "pull_request:4811", + "right": "pull_request:6360", + "accept": false, + "reason": "ControlNet inpaint pipeline addition vs adding padding_mask_crop to inpaint pipelines are distinct changes." + }, + { + "left": "pull_request:4856", + "right": "pull_request:6365", + "accept": false, + "reason": "Documentation link fix is unrelated to an MPS scheduler bugfix." + }, + { + "left": "pull_request:5097", + "right": "pull_request:6221", + "accept": false, + "reason": "Float16 test adjustments and slow-test cleanup do not fix the same underlying issue." + }, + { + "left": "pull_request:5164", + "right": "pull_request:7328", + "accept": false, + "reason": "FreeU core mechanism vs CI workflow dependency update are unrelated." + }, + { + "left": "pull_request:5369", + "right": "pull_request:5831", + "accept": false, + "reason": "Minor docs examples vs a docs title fix are different documentation edits, not one bug." + }, + { + "left": "pull_request:5420", + "right": "pull_request:5741", + "accept": false, + "reason": "A follow-up training-doc correction and PEFT unload speedup are not the same code-path problem." + }, + { + "left": "pull_request:5526", + "right": "pull_request:6179", + "accept": false, + "reason": "Both affect diffusion pipelines, but one fixes 2nd-order scheduler handling and the other adds fused attention projections; not mergeable as one bug fix." + }, + { + "left": "pull_request:5602", + "right": "pull_request:5982", + "accept": false, + "reason": "Different docs additions for different topics/models." + }, + { + "left": "pull_request:5653", + "right": "pull_request:7371", + "accept": false, + "reason": "Both mention PEFT/LoRA, but one changes maybe_raise_or_warn and the other adds DoRA LoRA support; different underlying changes." + }, + { + "left": "pull_request:5698", + "right": "pull_request:7473", + "accept": false, + "reason": "LCM img2img fix and missing raise statements in check_inputs are unrelated." + }, + { + "left": "pull_request:6305", + "right": "pull_request:6997", + "accept": false, + "reason": "DDPM zero-SNR scheduler support and LoRA support for a spatio-temporal UNet are unrelated." + }, + { + "left": "pull_request:6308", + "right": "pull_request:6369", + "accept": false, + "reason": "Both are LoRA-related tests/backend work, but one fixes dummy components and the other removes deprecated LoRA APIs; different changes." + }, + { + "left": "pull_request:6498", + "right": "pull_request:7857", + "accept": false, + "reason": "Inpaint docs for mask blur/padding mask crop and PixArtSigma docs are unrelated." + }, + { + "left": "pull_request:6564", + "right": "pull_request:8404", + "accept": false, + "reason": "IP-Adapter path fix and Latte video model addition are unrelated." + }, + { + "left": "pull_request:6998", + "right": "pull_request:7260", + "accept": false, + "reason": "A docs callout fix is unrelated to SAG IP-Adapter support." + }, + { + "left": "pull_request:7247", + "right": "pull_request:9247", + "accept": false, + "reason": "Single-file loading config tests and model variant identification fixes are different issues." + }, + { + "left": "pull_request:7353", + "right": "pull_request:9426", + "accept": false, + "reason": "ControlNet SDXL img2img pipeline update and artifact-uploader CI changes are unrelated." + }, + { + "left": "pull_request:7857", + "right": "pull_request:9242", + "accept": false, + "reason": "PixArtSigma documentation and xpu testing support do not address the same underlying bug/change." + }, + { + "left": "pull_request:7857", + "right": "pull_request:9718", + "accept": false, + "reason": "Different docs/community-script changes with no shared concrete bug." + }, + { + "left": "pull_request:8298", + "right": "pull_request:8452", + "accept": false, + "reason": "CLI env simplification and single-file loading error fix are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 196, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10148", + "pull_request:10351", + "pull_request:10360", + "pull_request:10567", + "pull_request:10622", + "pull_request:10630", + "pull_request:10985", + "pull_request:11130", + "pull_request:11290", + "pull_request:11582", + "pull_request:11738", + "pull_request:11979", + "pull_request:11997", + "pull_request:12275", + "pull_request:12383", + "pull_request:13063", + "pull_request:13216", + "pull_request:1327", + "pull_request:1513", + "pull_request:3252", + "pull_request:399", + "pull_request:4744", + "pull_request:5057", + "pull_request:5073", + "pull_request:5470", + "pull_request:6447", + "pull_request:6789", + "pull_request:6816", + "pull_request:6929", + "pull_request:7058", + "pull_request:7213", + "pull_request:7990", + "pull_request:8496", + "pull_request:8616", + "pull_request:8685", + "pull_request:8777", + "pull_request:8964", + "pull_request:9060", + "pull_request:9159", + "pull_request:9242", + "pull_request:9256", + "pull_request:9649", + "pull_request:9696", + "pull_request:9709", + "pull_request:9718", + "pull_request:9862", + "pull_request:9968", + "pull_request:9988" + ], + "soft_pairs": [ + "pull_request:8496|pull_request:8964", + "pull_request:9060|pull_request:9256", + "pull_request:9242|pull_request:9649", + "pull_request:9649|pull_request:9718", + "pull_request:9696|pull_request:9968", + "pull_request:5057|pull_request:5073", + "pull_request:10567|pull_request:11130", + "pull_request:11738|pull_request:11979", + "pull_request:5470|pull_request:6789", + "pull_request:6447|pull_request:8616", + "pull_request:6816|pull_request:7058", + "pull_request:12275|pull_request:12383", + "pull_request:10351|pull_request:10630", + "pull_request:11290|pull_request:9709", + "pull_request:1327|pull_request:399", + "pull_request:1513|pull_request:3252", + "pull_request:4744|pull_request:6929", + "pull_request:6929|pull_request:8685", + "pull_request:7213|pull_request:7990", + "pull_request:11582|pull_request:11997", + "pull_request:9159|pull_request:9988", + "pull_request:13063|pull_request:13216", + "pull_request:10985|pull_request:9862", + "pull_request:10148|pull_request:8777", + "pull_request:10360|pull_request:10622" + ], + "prepared_review_unit_hash": "0d9d290ebf2ae127b19fda6ec8468c0d4772611fa8a0cf47cffabdcb5ade77c2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 102955, + "estimated_input_tokens": 25739, + "estimated_eval_tokens": 51734 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated Diffusers PRs spanning different pipelines, docs, CI, tests, and utilities. I do not see any true duplicate PR pairs; the soft matches are mostly superficial topic overlap or shared subsystems, not the same concrete bug/change.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "No PR is a clear duplicate anchor because the cluster does not center on one underlying change; the items are heterogeneous and the soft edges are not strong enough to merge them.", + "best_issue_reason": null, + "best_pr_reason": "PR 11130 is the strongest representative only because it is a substantial, well-scoped merged change with broad impact and review activity. It is not a duplicate canonical in the strict sense, just the best single representative among unrelated PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8496", + "right": "pull_request:8964", + "accept": false, + "reason": "Different fixes in different code paths: SD3 warning-log behavior vs AudioLDM2 cache-position handling." + }, + { + "left": "pull_request:9060", + "right": "pull_request:9256", + "accept": false, + "reason": "Unrelated changes: removing a debug print in Flux vs fixing an image type check in Kandinsky2.2." + }, + { + "left": "pull_request:9242", + "right": "pull_request:9649", + "accept": false, + "reason": "Different concerns: xpu test backend support vs adding pred_original_sample to scheduler tuple returns." + }, + { + "left": "pull_request:9649", + "right": "pull_request:9718", + "accept": false, + "reason": "One is scheduler return behavior; the other is a docs update for a community-script callback. Not the same change." + }, + { + "left": "pull_request:9696", + "right": "pull_request:9968", + "accept": false, + "reason": "CI workflow input validation vs ControlNet module refactor; no shared underlying bug." + }, + { + "left": "pull_request:5057", + "right": "pull_request:5073", + "accept": false, + "reason": "Both touch SDXL, but one adds clip skip support and the other fixes multi-batch prompt embeds. Different concrete problems." + }, + { + "left": "pull_request:10567", + "right": "pull_request:11130", + "accept": false, + "reason": "HunyuanVideo true-CFG feature work vs pipeline-level quantization config. Different subsystems and goals." + }, + { + "left": "pull_request:11738", + "right": "pull_request:11979", + "accept": false, + "reason": "Test determinism for bnb quantization vs torch-version gating for compile tests. Not the same bug or change." + }, + { + "left": "pull_request:5470", + "right": "pull_request:6789", + "accept": false, + "reason": "ControlNet inpaint doc/fix work and PixArt 256px inference conversion are unrelated." + }, + { + "left": "pull_request:6447", + "right": "pull_request:8616", + "accept": false, + "reason": "K-Diffusion pipeline addition vs LoRA fusion support in SD3. Different features and code paths." + }, + { + "left": "pull_request:6816", + "right": "pull_request:7058", + "accept": false, + "reason": "SDXL LoRA training/DDP fix vs copying LoRA state dicts before mutation. Related area, but not the same concrete bug." + }, + { + "left": "pull_request:12275", + "right": "pull_request:12383", + "accept": false, + "reason": "TorchAO config support vs modular single-file support. Different feature work." + }, + { + "left": "pull_request:10351", + "right": "pull_request:10630", + "accept": false, + "reason": "LoRA test adjustment vs flax test slice update. Both are test-only changes, but not duplicates." + }, + { + "left": "pull_request:11290", + "right": "pull_request:9709", + "accept": false, + "reason": "One unpins torch in an ONNX Dockerfile, the other caps torch for CI. Opposite policy in different contexts." + }, + { + "left": "pull_request:1327", + "right": "pull_request:399", + "accept": false, + "reason": "Custom local pipeline loading vs ONNX Stable Diffusion exporter/pipeline. No shared underlying change." + }, + { + "left": "pull_request:1513", + "right": "pull_request:3252", + "accept": false, + "reason": "Accelerate config docs vs stateful offload notes. Different documentation topics." + }, + { + "left": "pull_request:4744", + "right": "pull_request:6929", + "accept": false, + "reason": "AudioLDM docstring fix vs community pipeline docs. Both docs, but not the same issue." + }, + { + "left": "pull_request:6929", + "right": "pull_request:8685", + "accept": false, + "reason": "Community pipeline loading docs vs SD3 float8 note. Different documentation updates." + }, + { + "left": "pull_request:7213", + "right": "pull_request:7990", + "accept": false, + "reason": "Merge-LoRAs docs vs scheduler-features docs. Same broad docs area, but unrelated content." + }, + { + "left": "pull_request:11582", + "right": "pull_request:11997", + "accept": false, + "reason": "Both are Wan-related, but one adds Wan VACE support and the other refactors Wan modular I2V/FLF2V support. Not mergeable as one duplicate PR." + }, + { + "left": "pull_request:9159", + "right": "pull_request:9988", + "accept": false, + "reason": "Flux sharding support vs Flux Redux conversion/pipeline work. Different code paths and goals." + }, + { + "left": "pull_request:13063", + "right": "pull_request:13216", + "accept": false, + "reason": "Different modular pipeline families: Wan refactor vs Helios addition." + }, + { + "left": "pull_request:10985", + "right": "pull_request:9862", + "accept": false, + "reason": "ComfyUI LoRA compatibility improvements vs saving LoRA adapters. Same area, but distinct functionality and bug scope." + }, + { + "left": "pull_request:10148", + "right": "pull_request:8777", + "accept": false, + "reason": "OmniGen feature addition vs dummy attention processor placeholders. Unrelated changes." + }, + { + "left": "pull_request:10360", + "right": "pull_request:10622", + "accept": false, + "reason": "Release TestPyPI install fix vs general uv installation docs. Different workflows and fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 197, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1038", + "pull_request:10838", + "pull_request:11254", + "pull_request:11894", + "pull_request:12005", + "pull_request:12425", + "pull_request:12438", + "pull_request:12586", + "pull_request:12714", + "pull_request:12863", + "pull_request:12911", + "pull_request:12914", + "pull_request:12960", + "pull_request:12973", + "pull_request:130", + "pull_request:1309", + "pull_request:13100", + "pull_request:13120", + "pull_request:13364", + "pull_request:1525", + "pull_request:1554", + "pull_request:1878", + "pull_request:1920", + "pull_request:2001", + "pull_request:2274", + "pull_request:2325", + "pull_request:2481", + "pull_request:2584", + "pull_request:2793", + "pull_request:2861", + "pull_request:3127", + "pull_request:3275", + "pull_request:3410", + "pull_request:3506", + "pull_request:3607", + "pull_request:3681", + "pull_request:3691", + "pull_request:4034", + "pull_request:4262", + "pull_request:4276", + "pull_request:4315", + "pull_request:4498", + "pull_request:4655", + "pull_request:5695", + "pull_request:5747", + "pull_request:630", + "pull_request:9838", + "pull_request:9875" + ], + "soft_pairs": [ + "pull_request:1038|pull_request:630", + "pull_request:10838|pull_request:9875", + "pull_request:11254|pull_request:9838", + "pull_request:11894|pull_request:12911", + "pull_request:12005|pull_request:12438", + "pull_request:12425|pull_request:12960", + "pull_request:12586|pull_request:13100", + "pull_request:12714|pull_request:13364", + "pull_request:12863|pull_request:12914", + "pull_request:12973|pull_request:13120", + "pull_request:130|pull_request:1309", + "pull_request:1525|pull_request:3410", + "pull_request:1554|pull_request:2793", + "pull_request:1878|pull_request:3506", + "pull_request:1920|pull_request:3275", + "pull_request:2001|pull_request:2274", + "pull_request:2001|pull_request:2481", + "pull_request:2001|pull_request:3410", + "pull_request:2325|pull_request:4498", + "pull_request:2584|pull_request:3127", + "pull_request:2584|pull_request:4276", + "pull_request:2861|pull_request:3681", + "pull_request:3607|pull_request:5695", + "pull_request:3691|pull_request:4655", + "pull_request:4034|pull_request:5747", + "pull_request:4262|pull_request:4315" + ], + "prepared_review_unit_hash": "3b3063609fee916416360b91f1980be35d8deda5e5e93b095a8ac460b5dd7304", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 96696, + "estimated_input_tokens": 24174, + "estimated_eval_tokens": 48604 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The candidate pairs are largely unrelated changes across different subsystems, versions, and goals; I don\u2019t see any pair that clearly fixes the same concrete bug or implements the same change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:1038", + "right": "pull_request:630", + "accept": false, + "reason": "One fixes a Flax training dtype mismatch; the other cleans up Flax checkpoint key handling. Different bugs, different code paths." + }, + { + "left": "pull_request:10838", + "right": "pull_request:9875", + "accept": false, + "reason": "FlashVideo pipeline addition and Meissonic support are separate model/pipeline features, not the same change." + }, + { + "left": "pull_request:11254", + "right": "pull_request:9838", + "accept": false, + "reason": "ConsisID import dependency fix is unrelated to Flux LoRA training behavior." + }, + { + "left": "pull_request:11894", + "right": "pull_request:12911", + "accept": false, + "reason": "Pinning k-diffusion for CI is unrelated to adding discrete diffusion pipelines and docs." + }, + { + "left": "pull_request:12005", + "right": "pull_request:12438", + "accept": false, + "reason": "Both touch loading utilities, but one handles local_files_only for sharded checkpoints while the other drops offload_state_dict for newer transformers models." + }, + { + "left": "pull_request:12425", + "right": "pull_request:12960", + "accept": false, + "reason": "Torch distributed import guarding is unrelated to how is_fsdp is computed in Flux2 training scripts." + }, + { + "left": "pull_request:12586", + "right": "pull_request:13100", + "accept": false, + "reason": "Doc ordering change vs modular pipeline documentation updates; not the same underlying issue." + }, + { + "left": "pull_request:12714", + "right": "pull_request:13364", + "accept": false, + "reason": "LoRA NaN test adjustments are unrelated to Wan Animate blur-kernel dtype casting." + }, + { + "left": "pull_request:12863", + "right": "pull_request:12914", + "accept": false, + "reason": "Cosmos conversion support is a separate feature from correcting a revision typo in pipeline docs." + }, + { + "left": "pull_request:12973", + "right": "pull_request:13120", + "accept": false, + "reason": "GLM-Image model support and modular loading tests are unrelated changes." + }, + { + "left": "pull_request:130", + "right": "pull_request:1309", + "accept": false, + "reason": "Example docs onboarding step and PIL handling improvements have no shared bug or change target." + }, + { + "left": "pull_request:1525", + "right": "pull_request:3410", + "accept": false, + "reason": "Upscaling batch-size fix and DeepSpeed state access fix are different training bugs." + }, + { + "left": "pull_request:1554", + "right": "pull_request:2793", + "accept": false, + "reason": "One corrects API docs, the other removes a cuDNN benchmark suggestion from optimization docs." + }, + { + "left": "pull_request:1878", + "right": "pull_request:3506", + "accept": false, + "reason": "Checkpoint format handling and global-step restoration are separate training/checkpointing issues." + }, + { + "left": "pull_request:1920", + "right": "pull_request:3275", + "accept": false, + "reason": "Documentation translation file addition and UNet upsample_size support are unrelated." + }, + { + "left": "pull_request:2001", + "right": "pull_request:2274", + "accept": false, + "reason": "ColossalAI typo fixes and a transforms naming clash are different example changes." + }, + { + "left": "pull_request:2001", + "right": "pull_request:2481", + "accept": false, + "reason": "ColossalAI typo fixes are unrelated to the unconditional training test fix." + }, + { + "left": "pull_request:2001", + "right": "pull_request:3410", + "accept": false, + "reason": "No shared code path or bug: ColossalAI typo fix vs accelerator state retrieval." + }, + { + "left": "pull_request:2325", + "right": "pull_request:4498", + "accept": false, + "reason": "ONNX VAE conversion support and SDXL reference pipeline support are separate features." + }, + { + "left": "pull_request:2584", + "right": "pull_request:3127", + "accept": false, + "reason": "Community ControlNet pipeline addition is unrelated to the img2img safety-checker fix." + }, + { + "left": "pull_request:2584", + "right": "pull_request:4276", + "accept": false, + "reason": "Pipeline addition and pipeline import tests are not the same concrete change." + }, + { + "left": "pull_request:2861", + "right": "pull_request:3681", + "accept": false, + "reason": "StableUnCLIP None-case test and checkpoint-limit handling are unrelated." + }, + { + "left": "pull_request:3607", + "right": "pull_request:5695", + "accept": false, + "reason": "Temb attention fix and PixArt masking condition fix address different model behaviors." + }, + { + "left": "pull_request:3691", + "right": "pull_request:4655", + "accept": false, + "reason": "LoRA processor selection for PyTorch 2.0 is unrelated to adding data_dir to dataset loading." + }, + { + "left": "pull_request:4034", + "right": "pull_request:5747", + "accept": false, + "reason": "LoRA unload utility and dataset data_dir plumbing are unrelated." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4315", + "accept": false, + "reason": "SDXL DreamBooth LoRA training fixes do not match the torch.compile graph-break fix in LoRA layers; different concrete problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 198, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11261", + "pull_request:11268", + "pull_request:11582", + "pull_request:11894", + "pull_request:11979", + "pull_request:12215", + "pull_request:12220", + "pull_request:12324", + "pull_request:12830", + "pull_request:1369", + "pull_request:3846", + "pull_request:4343", + "pull_request:4395", + "pull_request:4629", + "pull_request:4655", + "pull_request:4714", + "pull_request:4947", + "pull_request:5024", + "pull_request:5508", + "pull_request:5668", + "pull_request:5788", + "pull_request:5833", + "pull_request:5834", + "pull_request:6213", + "pull_request:6254", + "pull_request:6346", + "pull_request:6511", + "pull_request:6759", + "pull_request:6764", + "pull_request:7062", + "pull_request:7096", + "pull_request:7103", + "pull_request:7134", + "pull_request:7200", + "pull_request:7295", + "pull_request:7627", + "pull_request:7650", + "pull_request:7797", + "pull_request:7803", + "pull_request:819", + "pull_request:8378", + "pull_request:8422", + "pull_request:8702", + "pull_request:8981", + "pull_request:9569", + "pull_request:9662", + "pull_request:9875", + "pull_request:9918" + ], + "soft_pairs": [ + "pull_request:4343|pull_request:5508", + "pull_request:4395|pull_request:4629", + "pull_request:4655|pull_request:6511", + "pull_request:4947|pull_request:5024", + "pull_request:5788|pull_request:6254", + "pull_request:6213|pull_request:6346", + "pull_request:6759|pull_request:7096", + "pull_request:6764|pull_request:8981", + "pull_request:7062|pull_request:8378", + "pull_request:7103|pull_request:8378", + "pull_request:7134|pull_request:7295", + "pull_request:7200|pull_request:7650", + "pull_request:7627|pull_request:9569", + "pull_request:7797|pull_request:9662", + "pull_request:7803|pull_request:9662", + "pull_request:8422|pull_request:9918", + "pull_request:8702|pull_request:8981", + "pull_request:12215|pull_request:12220", + "pull_request:11582|pull_request:12324", + "pull_request:11261|pull_request:9875", + "pull_request:11268|pull_request:11894", + "pull_request:11979|pull_request:12830", + "pull_request:1369|pull_request:819", + "pull_request:3846|pull_request:5833", + "pull_request:3846|pull_request:5834", + "pull_request:4714|pull_request:5668" + ], + "prepared_review_unit_hash": "416940ba83e124cee87f341a41c862cce9ff53b596ae61c3b2f2c712153b5070", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 98032, + "estimated_input_tokens": 24508, + "estimated_eval_tokens": 49272 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most pairs are clearly unrelated PRs that just share a subsystem or file pattern. I found no true duplicate issue/PR pair; the strongest representative PR is the broad Qwen modular work (#12220), but it is not a duplicate of the other Qwen/Wan items.", + "confidence": 0.31, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single underlying change dominates the cluster; if one PR must stand in as the representative artifact, #12220 is the most substantial and central merged PR in the set.", + "best_issue_reason": null, + "best_pr_reason": "#12220 has the broadest scope and highest review signal, so it is the best standalone representative of this mixed PR cluster, though it should not be treated as a duplicate target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4343", + "right": "pull_request:5508", + "accept": false, + "reason": "Both touch DreamBooth LoRA SDXL, but one adds a rank argument while the other adds several training features; different changes, not the same bug/fix." + }, + { + "left": "pull_request:4395", + "right": "pull_request:4629", + "accept": false, + "reason": "4395 fixes pre-encoding class prompts in DreamBooth; 4629 is an unrelated SDXL long weighted prompt/push-to-hub cleanup effort." + }, + { + "left": "pull_request:4655", + "right": "pull_request:6511", + "accept": false, + "reason": "One passes data_dir to load_dataset; the other makes checkpointing compatible with torch.compile. Different code paths and goals." + }, + { + "left": "pull_request:4947", + "right": "pull_request:5024", + "accept": false, + "reason": "Training-script tests for T2I-Adapter versus scheduler step-index logic; no shared underlying bug." + }, + { + "left": "pull_request:5788", + "right": "pull_request:6254", + "accept": false, + "reason": "LCM error-message guard and randn_like/tracing behavior are unrelated changes." + }, + { + "left": "pull_request:6213", + "right": "pull_request:6346", + "accept": false, + "reason": "Text-to-video pipeline cleanup and LoRA state-dict conversion follow-up are different changes." + }, + { + "left": "pull_request:6759", + "right": "pull_request:7096", + "accept": false, + "reason": "InstantID batch embedding fix is unrelated to ControlNet training-script enhancements." + }, + { + "left": "pull_request:6764", + "right": "pull_request:8981", + "accept": false, + "reason": "Documentation about saving after fusing adapters is not the same change as adding LoraLoaderMixin to inits." + }, + { + "left": "pull_request:7062", + "right": "pull_request:8378", + "accept": false, + "reason": "A community pipeline docstring fix and an asymmetric autoencoder forward fix do not share the same underlying bug." + }, + { + "left": "pull_request:7103", + "right": "pull_request:8378", + "accept": false, + "reason": "ControlNet input validation fix and autoencoder forward fix are separate code-path issues." + }, + { + "left": "pull_request:7134", + "right": "pull_request:7295", + "accept": false, + "reason": "UNet type-hint cleanup is unrelated to single-file loader residue cleanup." + }, + { + "left": "pull_request:7200", + "right": "pull_request:7650", + "accept": false, + "reason": "Transformer norm-type validation and SDXL InstructPix2Pix `is_cosxl_edit` are distinct changes." + }, + { + "left": "pull_request:7627", + "right": "pull_request:9569", + "accept": false, + "reason": "Playground img2img latent scaling for SDXL is unrelated to CogVideoX autoencoder decode batching." + }, + { + "left": "pull_request:7797", + "right": "pull_request:9662", + "accept": false, + "reason": "DDPM fast-test size reduction and a transformers-version test bump are separate CI/test adjustments." + }, + { + "left": "pull_request:7803", + "right": "pull_request:9662", + "accept": false, + "reason": "DDIM fast-test size reduction is unrelated to the transformers version gating change." + }, + { + "left": "pull_request:8422", + "right": "pull_request:9918", + "accept": false, + "reason": "A subfolder-loading test suite and a server example are different additions, not duplicate fixes." + }, + { + "left": "pull_request:8702", + "right": "pull_request:8981", + "accept": false, + "reason": "SD3 conversion script creation is unrelated to the LoRA loader init/docs refactor." + }, + { + "left": "pull_request:12215", + "right": "pull_request:12220", + "accept": false, + "reason": "Both are Qwen-Image related, but one adds ControlNet support while the other builds a modular Qwen pipeline; different concrete features." + }, + { + "left": "pull_request:11582", + "right": "pull_request:12324", + "accept": false, + "reason": "Wan VACE initial support and Wan2.2 VACE-Fun conversion updates are related follow-ups, but not the same underlying PR/change." + }, + { + "left": "pull_request:11261", + "right": "pull_request:9875", + "accept": false, + "reason": "VidTok autoencoders and Meissonic support are unrelated feature PRs." + }, + { + "left": "pull_request:11268", + "right": "pull_request:11894", + "accept": false, + "reason": "Relaxing UnCLIP test tolerance and pinning k-diffusion for CI are unrelated test/packaging fixes." + }, + { + "left": "pull_request:11979", + "right": "pull_request:12830", + "accept": false, + "reason": "Torch version gating in compilation tests is unrelated to attention-backend test enhancements." + }, + { + "left": "pull_request:1369", + "right": "pull_request:819", + "accept": false, + "reason": "Flax sinusoidal embedding behavior and DDIM Windows int64 timesteps are unrelated bugs." + }, + { + "left": "pull_request:3846", + "right": "pull_request:5833", + "accept": false, + "reason": "Random module seeding and a docs-only PixArt change do not address the same code-path." + }, + { + "left": "pull_request:3846", + "right": "pull_request:5834", + "accept": false, + "reason": "Random module seeding and a ControlNet single-file import fix are unrelated." + }, + { + "left": "pull_request:4714", + "right": "pull_request:5668", + "accept": false, + "reason": "Both concern missing-Transformers behavior, but one fixes a dummy-object error path while the other broadly enables diffusers/DDPM without Transformers; not the same concrete fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 199, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10008", + "pull_request:10020", + "pull_request:10119", + "pull_request:1012", + "pull_request:10230", + "pull_request:10494", + "pull_request:10497", + "pull_request:10633", + "pull_request:10730", + "pull_request:10894", + "pull_request:10898", + "pull_request:11004", + "pull_request:11095", + "pull_request:11243", + "pull_request:11261", + "pull_request:1129", + "pull_request:11556", + "pull_request:11979", + "pull_request:11994", + "pull_request:11999", + "pull_request:12117", + "pull_request:12119", + "pull_request:12288", + "pull_request:12301", + "pull_request:12416", + "pull_request:12439", + "pull_request:12445", + "pull_request:12611", + "pull_request:13216", + "pull_request:1496", + "pull_request:1920", + "pull_request:1942", + "pull_request:1966", + "pull_request:2040", + "pull_request:2347", + "pull_request:2467", + "pull_request:2516", + "pull_request:3850", + "pull_request:4098", + "pull_request:6003", + "pull_request:6841", + "pull_request:7158", + "pull_request:750", + "pull_request:8229", + "pull_request:8981", + "pull_request:9175", + "pull_request:9257", + "pull_request:9709" + ], + "soft_pairs": [ + "pull_request:6841|pull_request:8981", + "pull_request:7158|pull_request:8229", + "pull_request:4098|pull_request:6003", + "pull_request:1966|pull_request:2040", + "pull_request:12301|pull_request:12445", + "pull_request:12117|pull_request:12416", + "pull_request:10020|pull_request:9175", + "pull_request:12119|pull_request:12439", + "pull_request:12611|pull_request:13216", + "pull_request:10008|pull_request:11095", + "pull_request:10119|pull_request:11556", + "pull_request:1012|pull_request:1129", + "pull_request:10230|pull_request:11004", + "pull_request:10494|pull_request:10497", + "pull_request:10633|pull_request:10898", + "pull_request:10730|pull_request:9257", + "pull_request:10894|pull_request:9709", + "pull_request:11243|pull_request:11999", + "pull_request:11261|pull_request:12288", + "pull_request:11979|pull_request:11994", + "pull_request:1496|pull_request:2516", + "pull_request:1920|pull_request:750", + "pull_request:1942|pull_request:3850", + "pull_request:2347|pull_request:2467" + ], + "prepared_review_unit_hash": "ba86b03b4d550b24a09574bf8dfb868e6e97caf6eabed86b3ef636261cbe1715", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 24, + "serialized_chars": 103827, + "estimated_input_tokens": 25957, + "estimated_eval_tokens": 52170 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs, CI, scheduler, LoRA, and pipeline PRs across many different code paths. There is no single canonical artifact that cleanly represents the cluster.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are heterogeneous and do not share one underlying bug or change.", + "best_issue_reason": null, + "best_pr_reason": "No best PR: none of the PRs is representative of the whole cluster, and selecting one would be arbitrary.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6841", + "right": "pull_request:8981", + "accept": false, + "reason": "Different changes: one restores BasicTransformerBlock params; the other is LoRA docs cleanup. Same broad area, not same bug." + }, + { + "left": "pull_request:7158", + "right": "pull_request:8229", + "accept": false, + "reason": "Unrelated fixes: benchmark cleanup vs Stable Video Diffusion resize behavior." + }, + { + "left": "pull_request:4098", + "right": "pull_request:6003", + "accept": false, + "reason": "Both are inpainting-related, but for different pipelines and different code paths; not the same concrete change." + }, + { + "left": "pull_request:1966", + "right": "pull_request:2040", + "accept": false, + "reason": "Different pipeline additions: Flax inpainting vs InstructPix2Pix. Not the same underlying change." + }, + { + "left": "pull_request:12301", + "right": "pull_request:12445", + "accept": false, + "reason": "Qwen ControlNet inpaint support and Flux modular alignment are separate features." + }, + { + "left": "pull_request:12117", + "right": "pull_request:12416", + "accept": false, + "reason": "QwenImage pipelines vs QwenImage Edit Plus modular support are distinct changes." + }, + { + "left": "pull_request:10020", + "right": "pull_request:9175", + "accept": false, + "reason": "Different features: SD3.5 ControlNet conversion tooling vs Flux ControlNet support." + }, + { + "left": "pull_request:12119", + "right": "pull_request:12439", + "accept": false, + "reason": "LoRA loading support and attention backend support are unrelated code paths." + }, + { + "left": "pull_request:12611", + "right": "pull_request:13216", + "accept": false, + "reason": "Separate modular pipeline families (Wan vs Helios), not duplicates." + }, + { + "left": "pull_request:10008", + "right": "pull_request:11095", + "accept": false, + "reason": "Docs table fix for CogVideoX vs Flux ControlNet example fix are unrelated." + }, + { + "left": "pull_request:10119", + "right": "pull_request:11556", + "accept": false, + "reason": "Docs for load_lora_adapter vs regional compilation docs; no shared bug/change." + }, + { + "left": "pull_request:1012", + "right": "pull_request:1129", + "accept": false, + "reason": "Both touch DreamBooth, but one is PyTorch text-encoder train mode and the other is Flax sample batch sizing." + }, + { + "left": "pull_request:10230", + "right": "pull_request:11004", + "accept": false, + "reason": "Docs update for LoRA mixins and PEFT key-prefix handling are different issues." + }, + { + "left": "pull_request:10494", + "right": "pull_request:10497", + "accept": false, + "reason": "Both address tokenizers install in CI, but they propose different fixes to the same line; related, but not the same concrete change." + }, + { + "left": "pull_request:10633", + "right": "pull_request:10898", + "accept": false, + "reason": "Flux IP-Adapter docs and remote_decode infrastructure are unrelated." + }, + { + "left": "pull_request:10730", + "right": "pull_request:9257", + "accept": false, + "reason": "Hook-unwrapping fix and DreamBooth Flux script bug fix are separate problems." + }, + { + "left": "pull_request:10894", + "right": "pull_request:9709", + "accept": false, + "reason": "Toctree sorting/docs ordering is unrelated to the torch version pin." + }, + { + "left": "pull_request:11243", + "right": "pull_request:11999", + "accept": false, + "reason": "Musubi Wan LoRA conversion support and PEFT LoRA injection changes are different LoRA issues." + }, + { + "left": "pull_request:11261", + "right": "pull_request:12288", + "accept": false, + "reason": "VidTok autoencoder addition and InfiniteTalk pipeline addition are unrelated model/pipeline features." + }, + { + "left": "pull_request:1496", + "right": "pull_request:2516", + "accept": false, + "reason": "Habana Gaudi docs and evaluation docs are unrelated documentation additions." + }, + { + "left": "pull_request:1920", + "right": "pull_request:750", + "accept": false, + "reason": "Translating docs and LMS scheduler back-compat are completely unrelated." + }, + { + "left": "pull_request:1942", + "right": "pull_request:3850", + "accept": false, + "reason": "Script flag rename and audio_diffusion test fixes do not share an underlying bug." + }, + { + "left": "pull_request:2347", + "right": "pull_request:2467", + "accept": false, + "reason": "Scheduler schedule updates and latent output-type plumbing are unrelated changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 200, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:237", + "pull_request:2487", + "pull_request:3260", + "pull_request:3370", + "pull_request:3552", + "pull_request:3561", + "pull_request:3711", + "pull_request:3806", + "pull_request:3877", + "pull_request:4242", + "pull_request:4276", + "pull_request:4285", + "pull_request:4365", + "pull_request:4422", + "pull_request:4508", + "pull_request:4635", + "pull_request:4887", + "pull_request:492", + "pull_request:5097", + "pull_request:5180", + "pull_request:5196", + "pull_request:5204", + "pull_request:5391", + "pull_request:5420", + "pull_request:5421", + "pull_request:5456", + "pull_request:5602", + "pull_request:5614", + "pull_request:5626", + "pull_request:5724", + "pull_request:5754", + "pull_request:5880", + "pull_request:5974", + "pull_request:5988", + "pull_request:6081", + "pull_request:6105", + "pull_request:6140", + "pull_request:6163", + "pull_request:6174", + "pull_request:6213", + "pull_request:6346", + "pull_request:6908", + "pull_request:6909", + "pull_request:7134", + "pull_request:7359", + "pull_request:7780", + "pull_request:934" + ], + "soft_pairs": [ + "pull_request:237|pull_request:934", + "pull_request:2487|pull_request:492", + "pull_request:3260|pull_request:4887", + "pull_request:3370|pull_request:5196", + "pull_request:3552|pull_request:5180", + "pull_request:3561|pull_request:3877", + "pull_request:3711|pull_request:5614", + "pull_request:3806|pull_request:5456", + "pull_request:4242|pull_request:5880", + "pull_request:4276|pull_request:4508", + "pull_request:4276|pull_request:5456", + "pull_request:4285|pull_request:4422", + "pull_request:4365|pull_request:6346", + "pull_request:4635|pull_request:5602", + "pull_request:5097|pull_request:6163", + "pull_request:5204|pull_request:5754", + "pull_request:5391|pull_request:7134", + "pull_request:5420|pull_request:6213", + "pull_request:5421|pull_request:5626", + "pull_request:5724|pull_request:6174", + "pull_request:5974|pull_request:6908", + "pull_request:5974|pull_request:6909", + "pull_request:5988|pull_request:6140", + "pull_request:6081|pull_request:6213", + "pull_request:6105|pull_request:7359", + "pull_request:6140|pull_request:7780" + ], + "prepared_review_unit_hash": "a989fb38b8c981937581eb7aaab3acffa991d08e4c83331d7ae2e39a5aeb1ab8", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 101304, + "estimated_input_tokens": 25326, + "estimated_eval_tokens": 50908 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set: it contains many unrelated pull requests spanning docs, tests, example scripts, and feature work, with only superficial thematic similarity in a few pairs.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a good canonical representative; the items are heterogeneous and do not share one concrete bug or change.", + "best_issue_reason": null, + "best_pr_reason": "No PR here is a strong global representative because the cluster is broad and the soft-linked pairs still address different code paths or documentation updates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:237", + "right": "pull_request:934", + "accept": false, + "reason": "One adds CI/fixes a test; the other fixes MPS test expectations. Different maintenance changes, not the same change or bug." + }, + { + "left": "pull_request:2487", + "right": "pull_request:492", + "accept": false, + "reason": "Docs for JAX/Flax stable diffusion vs initial Flax PNDM scheduler support; different features and code paths." + }, + { + "left": "pull_request:3260", + "right": "pull_request:4887", + "accept": false, + "reason": "Removing a required CLI arg for ControlNet training is unrelated to removing latents from Kandinsky prior_emb2emb." + }, + { + "left": "pull_request:3370", + "right": "pull_request:5196", + "accept": false, + "reason": "Both touch docs about optimization, but one updates PyTorch 2.0 docs and the other edits xformers/memory docs; not the same change." + }, + { + "left": "pull_request:3552", + "right": "pull_request:5180", + "accept": false, + "reason": "General docs maintenance versus a small Wuerstchen citation fix; unrelated documentation edits." + }, + { + "left": "pull_request:3561", + "right": "pull_request:3877", + "accept": false, + "reason": "DreamBooth IF stage II upscaling changes are unrelated to the SDXL DreamBooth LoRA WIP implementation." + }, + { + "left": "pull_request:3711", + "right": "pull_request:5614", + "accept": false, + "reason": "Reverting a docs-build secrets workflow change is unrelated to reverting SDXL training width/height ordering." + }, + { + "left": "pull_request:3806", + "right": "pull_request:5456", + "accept": false, + "reason": "A README note about checkpoint loading is unrelated to adding the RealFill training/inference project." + }, + { + "left": "pull_request:4242", + "right": "pull_request:5880", + "accept": false, + "reason": "Allowing VAE loading from single-file checkpoints is a loader change; the VAE PR fixes config support across latent-diffusion models, not the same bug." + }, + { + "left": "pull_request:4276", + "right": "pull_request:4508", + "accept": false, + "reason": "A pipeline import test and a fix for use_onnx in from_pretrained are different concerns." + }, + { + "left": "pull_request:4276", + "right": "pull_request:5456", + "accept": false, + "reason": "Pipeline import coverage is unrelated to the RealFill training project." + }, + { + "left": "pull_request:4285", + "right": "pull_request:4422", + "accept": false, + "reason": "Both edit ONNX/OpenVINO docs, but they are separate documentation updates for different content and examples." + }, + { + "left": "pull_request:4365", + "right": "pull_request:6346", + "accept": false, + "reason": "Both touch SDXL LoRA-related files, but one adds the example/docs and the other fixes state-dict conversion; not the same concrete change." + }, + { + "left": "pull_request:4635", + "right": "pull_request:5602", + "accept": false, + "reason": "MultiControlNet docs vs table-format cleanup in stable diffusion overview; unrelated documentation work." + }, + { + "left": "pull_request:5097", + "right": "pull_request:6163", + "accept": false, + "reason": "Both are test fixes for float16/LoRA-related paths, but they target different test cases and failure modes." + }, + { + "left": "pull_request:5204", + "right": "pull_request:5754", + "accept": false, + "reason": "LoRA text-encoder scaling fix versus minor styling cleanup in training scripts; not the same issue." + }, + { + "left": "pull_request:5391", + "right": "pull_request:7134", + "accept": false, + "reason": "General typehint/docs cleanup in model classes versus UNet type-hint adjustments; separate documentation/style changes." + }, + { + "left": "pull_request:5420", + "right": "pull_request:6213", + "accept": false, + "reason": "A docs example fix for create_repo/repo_id is unrelated to a text-to-video pipeline cleanup." + }, + { + "left": "pull_request:5421", + "right": "pull_request:5626", + "accept": false, + "reason": "Import/deprecation fix in versatile diffusion is unrelated to the CPU offload test correction." + }, + { + "left": "pull_request:5724", + "right": "pull_request:6174", + "accept": false, + "reason": "LoRA test decorator changes are unrelated to adding a missing backslash in a docs command." + }, + { + "left": "pull_request:5974", + "right": "pull_request:6908", + "accept": false, + "reason": "PixArtAlpha pipeline adaptation is unrelated to SDXL ControlNet model-card standardization." + }, + { + "left": "pull_request:5974", + "right": "pull_request:6909", + "accept": false, + "reason": "PixArtAlpha pipeline adaptation is unrelated to ControlNet Flax model-card standardization." + }, + { + "left": "pull_request:5988", + "right": "pull_request:6140", + "accept": false, + "reason": "Kandinsky 3 docs/pipeline work is unrelated to the IP-Adapter API doc addition." + }, + { + "left": "pull_request:6081", + "right": "pull_request:6213", + "accept": false, + "reason": "Adjusting SDXL scheduler sigmas for torch.compile is unrelated to text-to-video pipeline cleanup." + }, + { + "left": "pull_request:6105", + "right": "pull_request:7359", + "accept": false, + "reason": "SDE Drag community pipeline addition is unrelated to the ResAdapter training script." + }, + { + "left": "pull_request:6140", + "right": "pull_request:7780", + "accept": false, + "reason": "IP-Adapter API doc introduction versus an IP-Adapter masking wording fix; same area but not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 201, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10578", + "pull_request:11004", + "pull_request:11595", + "pull_request:11802", + "pull_request:11894", + "pull_request:12226", + "pull_request:12618", + "pull_request:13294", + "pull_request:13352", + "pull_request:1945", + "pull_request:2002", + "pull_request:2347", + "pull_request:2617", + "pull_request:2994", + "pull_request:3900", + "pull_request:3924", + "pull_request:4215", + "pull_request:4508", + "pull_request:4744", + "pull_request:4859", + "pull_request:5035", + "pull_request:5057", + "pull_request:5164", + "pull_request:5579", + "pull_request:6003", + "pull_request:6209", + "pull_request:6369", + "pull_request:6757", + "pull_request:6927", + "pull_request:6998", + "pull_request:7230", + "pull_request:7662", + "pull_request:7780", + "pull_request:7913", + "pull_request:8322", + "pull_request:8531", + "pull_request:8571", + "pull_request:8755", + "pull_request:8825", + "pull_request:8937", + "pull_request:9040", + "pull_request:9041", + "pull_request:9213", + "pull_request:9234", + "pull_request:9707", + "pull_request:9807", + "pull_request:9968" + ], + "soft_pairs": [ + "pull_request:6209|pull_request:6927", + "pull_request:6757|pull_request:7913", + "pull_request:6998|pull_request:7780", + "pull_request:8322|pull_request:8571", + "pull_request:8531|pull_request:9234", + "pull_request:8825|pull_request:9807", + "pull_request:8937|pull_request:9707", + "pull_request:9040|pull_request:9041", + "pull_request:5057|pull_request:7230", + "pull_request:10578|pull_request:9213", + "pull_request:11004|pull_request:11802", + "pull_request:11595|pull_request:12226", + "pull_request:11894|pull_request:12618", + "pull_request:13294|pull_request:13352", + "pull_request:1945|pull_request:2617", + "pull_request:2002|pull_request:2347", + "pull_request:2994|pull_request:3924", + "pull_request:4215|pull_request:4744", + "pull_request:4508|pull_request:4859", + "pull_request:5035|pull_request:5579", + "pull_request:5164|pull_request:5579", + "pull_request:6369|pull_request:7662", + "pull_request:8755|pull_request:9968", + "pull_request:3900|pull_request:6003" + ], + "prepared_review_unit_hash": "bf1832a5dbb444a623fc4fb318b883753536856f88723c0c6e94dad831704ccc", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 100160, + "estimated_input_tokens": 25040, + "estimated_eval_tokens": 50336 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The proposed soft-edge pairs are not true duplicates: they mostly share a subsystem or release theme, but each PR targets a different concrete change. I rejected all soft edges and would treat PR #9213 as the best representative only because it is the broadest, most central PR in the set.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a true canonical duplicate for this set; if a representative is needed, #9213 is the strongest anchor because it is the broad quantization feature PR with the most discussion and inbound references.", + "best_issue_reason": null, + "best_pr_reason": "#9213 is the best global representative because it is the largest, most impactful, and most referenced PR in the cluster, whereas the others are smaller point fixes or unrelated docs/CI changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6209", + "right": "pull_request:6927", + "accept": false, + "reason": "Different changes: README example fix vs community prompt2prompt import refactor. Same docs area, not the same underlying bug or change." + }, + { + "left": "pull_request:6757", + "right": "pull_request:7913", + "accept": false, + "reason": "Both touch timestep/sigma handling, but one is a device-placement fix and the other adds custom timesteps/sigmas for ControlNet. Not mergeable as one duplicate change." + }, + { + "left": "pull_request:6998", + "right": "pull_request:7780", + "accept": false, + "reason": "Both edit the IP-Adapter docs, but they fix different doc sections and issues: callout markup vs masking/face-id wording." + }, + { + "left": "pull_request:8322", + "right": "pull_request:8571", + "accept": false, + "reason": "Both are typo/style errata, but they are broad, unrelated documentation cleanup PRs rather than the same concrete change." + }, + { + "left": "pull_request:8531", + "right": "pull_request:9234", + "accept": false, + "reason": "Completely different fixes: sharding/device_map behavior versus safetensors compatibility regression." + }, + { + "left": "pull_request:8825", + "right": "pull_request:9807", + "accept": false, + "reason": "Unrelated code paths: k-diffusion pipeline feature work versus AutoencoderKL LoRA adapter support." + }, + { + "left": "pull_request:8937", + "right": "pull_request:9707", + "accept": false, + "reason": "One fixes long-prompt handling in AuraFlow code; the other is a docstring refactor in a different model file." + }, + { + "left": "pull_request:9040", + "right": "pull_request:9041", + "accept": false, + "reason": "Same project area but different additions: Differential Diffusion vs an Img2Img variant for HunyuanDiT." + }, + { + "left": "pull_request:5057", + "right": "pull_request:7230", + "accept": false, + "reason": "Different features: SDXL clip-skip support versus single-file checkpoint loading support." + }, + { + "left": "pull_request:10578", + "right": "pull_request:9213", + "accept": false, + "reason": "#10578 is a specific 4-bit Flux LoRA loading fix; #9213 is a broad bitsandbytes quantization framework addition. Related topic, not the same bug/change." + }, + { + "left": "pull_request:11004", + "right": "pull_request:11802", + "accept": false, + "reason": "Both involve LoRA cleanup, but one removes a PEFT prefix import path and the other removes deprecated lora_base methods. Not the same underlying change." + }, + { + "left": "pull_request:11595", + "right": "pull_request:12226", + "accept": false, + "reason": "Different domains entirely: Flux LoRA loading fix versus a Qwen-Image ControlNet test addition." + }, + { + "left": "pull_request:11894", + "right": "pull_request:12618", + "accept": false, + "reason": "Both are CI-related, but one pins k-diffusion and the other fixes a uv install typo. Different failure modes." + }, + { + "left": "pull_request:13294", + "right": "pull_request:13352", + "accept": false, + "reason": "Unrelated CI/workflow fixes: WanVACE test flakiness versus adding a checkout step to Claude review workflow." + }, + { + "left": "pull_request:1945", + "right": "pull_request:2617", + "accept": false, + "reason": "Tests for UNet blocks versus introducing the image processor are unrelated changes." + }, + { + "left": "pull_request:2002", + "right": "pull_request:2347", + "accept": false, + "reason": "One adds a LoRA training example; the other changes scheduler noise scheduling. No shared concrete bug." + }, + { + "left": "pull_request:2994", + "right": "pull_request:3924", + "accept": false, + "reason": "ControlNet training-script casting fix versus a doc example parameter update. Same example family, different change type and problem." + }, + { + "left": "pull_request:4215", + "right": "pull_request:4744", + "accept": false, + "reason": "Different documentation updates in different pipelines; not the same underlying issue." + }, + { + "left": "pull_request:4508", + "right": "pull_request:4859", + "accept": false, + "reason": "`use_onnx` loading behavior and an SDXL inpaint fast test tweak are unrelated." + }, + { + "left": "pull_request:5035", + "right": "pull_request:5579", + "accept": false, + "reason": "One avoids forcing transformers installation; the other bumps the minimum transformers version for PEFT. Different dependency policy changes." + }, + { + "left": "pull_request:5164", + "right": "pull_request:5579", + "accept": false, + "reason": "FreeU architecture changes are unrelated to a PEFT/transformers minimum-version bump." + }, + { + "left": "pull_request:6369", + "right": "pull_request:7662", + "accept": false, + "reason": "LoRA API deprecation cleanup versus removing redundant test dependencies from Flax workflows. Different code paths." + }, + { + "left": "pull_request:8755", + "right": "pull_request:9968", + "accept": false, + "reason": "Different model areas and fixes: SD3 DreamBooth LoRA training versus ControlNet module refactor." + }, + { + "left": "pull_request:3900", + "right": "pull_request:6003", + "accept": false, + "reason": "Video-to-video pipeline addition versus Stable Video Diffusion inpaint support. Similar high-level area, but not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 202, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10322", + "pull_request:10358", + "pull_request:10370", + "pull_request:10537", + "pull_request:10854", + "pull_request:10983", + "pull_request:11025", + "pull_request:11059", + "pull_request:1128", + "pull_request:11524", + "pull_request:11535", + "pull_request:11556", + "pull_request:11582", + "pull_request:11715", + "pull_request:11894", + "pull_request:12151", + "pull_request:12541", + "pull_request:12611", + "pull_request:1283", + "pull_request:1355", + "pull_request:1496", + "pull_request:164", + "pull_request:1945", + "pull_request:1977", + "pull_request:2048", + "pull_request:2279", + "pull_request:2314", + "pull_request:2325", + "pull_request:2398", + "pull_request:2489", + "pull_request:3054", + "pull_request:3334", + "pull_request:3546", + "pull_request:4090", + "pull_request:4132", + "pull_request:4597", + "pull_request:4712", + "pull_request:5053", + "pull_request:5057", + "pull_request:5127", + "pull_request:538", + "pull_request:5388", + "pull_request:5736", + "pull_request:5754", + "pull_request:5853", + "pull_request:5991", + "pull_request:6816" + ], + "soft_pairs": [ + "pull_request:1128|pull_request:1355", + "pull_request:4597|pull_request:5127", + "pull_request:5053|pull_request:5991", + "pull_request:5388|pull_request:6816", + "pull_request:11582|pull_request:12611", + "pull_request:10983|pull_request:11025", + "pull_request:10322|pull_request:10537", + "pull_request:10358|pull_request:11556", + "pull_request:10370|pull_request:11059", + "pull_request:10854|pull_request:11894", + "pull_request:11524|pull_request:11535", + "pull_request:11556|pull_request:11715", + "pull_request:12151|pull_request:12541", + "pull_request:1283|pull_request:2325", + "pull_request:1496|pull_request:164", + "pull_request:1945|pull_request:2314", + "pull_request:1977|pull_request:538", + "pull_request:2048|pull_request:4132", + "pull_request:2279|pull_request:3054", + "pull_request:2398|pull_request:4090", + "pull_request:2489|pull_request:3334", + "pull_request:3546|pull_request:5057", + "pull_request:4712|pull_request:5853", + "pull_request:5736|pull_request:5754" + ], + "prepared_review_unit_hash": "18a36dc7a4aaa4a62b93adb56f3987184041f4a83ab74dccda739db35b436b06", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 100518, + "estimated_input_tokens": 25130, + "estimated_eval_tokens": 50516 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very noisy: the soft edges mostly link unrelated PRs that only share broad keywords or a subsystem name. None of the candidate pairs looks like the same concrete fix/change, so all soft edges should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the cluster is heterogeneous, and the candidate pairings do not describe the same underlying change.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a safe global representative; the items span unrelated docs, tests, LoRA, Wan, SDXL, and scheduler changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1128", + "right": "pull_request:1355", + "accept": false, + "reason": "Inpaint tensor-input support and a Flax img2img pipeline are unrelated features with different pipelines and codepaths." + }, + { + "left": "pull_request:4597", + "right": "pull_request:5127", + "accept": false, + "reason": "MultiControlNet SDXL support and multi-adapter SDXL adapter handling are different integration changes, not the same bug fix." + }, + { + "left": "pull_request:5053", + "right": "pull_request:5991", + "accept": false, + "reason": "`validation_image` CLI handling and LoRA prefix serialization are unrelated fixes in different areas." + }, + { + "left": "pull_request:5388", + "right": "pull_request:6816", + "accept": false, + "reason": "Both touch DreamBooth LoRA, but one adds PEFT support and the other fixes DDP embedding grad handling; different code paths and goals." + }, + { + "left": "pull_request:11582", + "right": "pull_request:12611", + "accept": false, + "reason": "Both are Wan-related, but Wan VACE and modular Wan are separate feature sets and not the same change." + }, + { + "left": "pull_request:10983", + "right": "pull_request:11025", + "accept": false, + "reason": "Hunyuan I2V model support and Wan LoRA support are unrelated model integrations." + }, + { + "left": "pull_request:10322", + "right": "pull_request:10537", + "accept": false, + "reason": "Sana 2K model conversion and Sana 4K pipeline changes are different model-size additions, not one underlying fix." + }, + { + "left": "pull_request:10358", + "right": "pull_request:11556", + "accept": false, + "reason": "A community-projects docs entry and regional compilation docs are unrelated documentation updates." + }, + { + "left": "pull_request:10370", + "right": "pull_request:11059", + "accept": false, + "reason": "TorchAO quantizer style/import cleanup and Wan LoRA conversion logic are unrelated changes." + }, + { + "left": "pull_request:10854", + "right": "pull_request:11894", + "accept": false, + "reason": "Hunyuan DiT test-module renaming and k-diffusion pinning for CI are unrelated CI/test maintenance tasks." + }, + { + "left": "pull_request:11524", + "right": "pull_request:11535", + "accept": false, + "reason": "Cosmos torchvision import gating and Framepack transformer initialization order are different model implementation fixes." + }, + { + "left": "pull_request:11556", + "right": "pull_request:11715", + "accept": false, + "reason": "Regional compilation docs and bnb test annotations do not fix the same issue." + }, + { + "left": "pull_request:12151", + "right": "pull_request:12541", + "accept": false, + "reason": "Flax deprecation and late-step Wan LoRA conversion are unrelated changes." + }, + { + "left": "pull_request:1283", + "right": "pull_request:2325", + "accept": false, + "reason": "Versatile Diffusion model support and a VAE ONNX converter are separate features." + }, + { + "left": "pull_request:1496", + "right": "pull_request:164", + "accept": false, + "reason": "Habana Gaudi docs and DDPM/PNDM scaled_linear schedule support are unrelated." + }, + { + "left": "pull_request:1945", + "right": "pull_request:2314", + "accept": false, + "reason": "UNet block tests and fast-test tolerance updates are different test additions." + }, + { + "left": "pull_request:1977", + "right": "pull_request:538", + "accept": false, + "reason": "DreamBooth ColossalAI script adjustments and hub-version pinning are unrelated updates." + }, + { + "left": "pull_request:2048", + "right": "pull_request:4132", + "accept": false, + "reason": "Accelerate checkpoint-hook loading and making accelerate optional in a converter solve different problems." + }, + { + "left": "pull_request:2279", + "right": "pull_request:3054", + "accept": false, + "reason": "Ruff/black tooling changes and a contribution-doc wording tweak are unrelated maintenance edits." + }, + { + "left": "pull_request:2398", + "right": "pull_request:4090", + "accept": false, + "reason": "Pix2Pix Zero docs and BentoML integration docs are unrelated documentation additions." + }, + { + "left": "pull_request:2489", + "right": "pull_request:3334", + "accept": false, + "reason": "Core ML docs and a docstring formatting fix are not the same underlying change." + }, + { + "left": "pull_request:3546", + "right": "pull_request:5057", + "accept": false, + "reason": "Panorama scheduler compatibility and SDXL clip-skip support are distinct pipeline features." + }, + { + "left": "pull_request:4712", + "right": "pull_request:5853", + "accept": false, + "reason": "Tests fetcher infrastructure and doctest workflows are separate test-system changes." + }, + { + "left": "pull_request:5736", + "right": "pull_request:5754", + "accept": false, + "reason": "RealFill torchvision compatibility and style cleanup in training scripts are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 203, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10119", + "pull_request:10188", + "pull_request:10324", + "pull_request:10376", + "pull_request:10378", + "pull_request:1038", + "pull_request:10473", + "pull_request:10579", + "pull_request:10622", + "pull_request:10747", + "pull_request:10808", + "pull_request:10966", + "pull_request:11039", + "pull_request:11124", + "pull_request:11261", + "pull_request:11273", + "pull_request:11684", + "pull_request:11823", + "pull_request:11869", + "pull_request:12040", + "pull_request:12439", + "pull_request:2421", + "pull_request:2561", + "pull_request:2593", + "pull_request:3386", + "pull_request:4690", + "pull_request:5807", + "pull_request:6006", + "pull_request:6213", + "pull_request:6355", + "pull_request:6787", + "pull_request:6929", + "pull_request:7086", + "pull_request:7117", + "pull_request:7156", + "pull_request:7246", + "pull_request:7541", + "pull_request:819", + "pull_request:8229", + "pull_request:8797", + "pull_request:9257", + "pull_request:9311", + "pull_request:9426", + "pull_request:9507", + "pull_request:9735", + "pull_request:9968", + "pull_request:9985" + ], + "soft_pairs": [ + "pull_request:5807|pull_request:6213", + "pull_request:6355|pull_request:6787", + "pull_request:7117|pull_request:7541", + "pull_request:7156|pull_request:8229", + "pull_request:7246|pull_request:8229", + "pull_request:9735|pull_request:9968", + "pull_request:11684|pull_request:11823", + "pull_request:2561|pull_request:3386", + "pull_request:9507|pull_request:9985", + "pull_request:10747|pull_request:9311", + "pull_request:10324|pull_request:8797", + "pull_request:11869|pull_request:12439", + "pull_request:2421|pull_request:819", + "pull_request:4690|pull_request:6006", + "pull_request:6929|pull_request:7086", + "pull_request:10119|pull_request:10622", + "pull_request:10188|pull_request:9257", + "pull_request:10376|pull_request:10808", + "pull_request:10378|pull_request:10473", + "pull_request:1038|pull_request:2593", + "pull_request:10579|pull_request:11261", + "pull_request:10808|pull_request:12040", + "pull_request:10966|pull_request:11124", + "pull_request:10966|pull_request:9426", + "pull_request:11039|pull_request:11273" + ], + "prepared_review_unit_hash": "de0c771330a4c7f40573d1f16eb2d55495e9230cfe6814247db2c4d753697aba", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 99311, + "estimated_input_tokens": 24828, + "estimated_eval_tokens": 49912 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: the soft pairs mostly connect unrelated docs, tests, scheduler, LoRA, and feature PRs rather than true duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9985 is the broadest and most central change in the set by scope and discussion volume, though the cluster itself is not a true duplicate group.", + "best_issue_reason": null, + "best_pr_reason": "PR 9985 is the strongest representative by breadth and activity, but it should not be treated as a duplicate canonical for the other items.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5807", + "right": "pull_request:6213", + "accept": false, + "reason": "Both are community pipeline work, but they fix different areas: LPW SDXL embeddings vs. text-to-video pipeline cleanup." + }, + { + "left": "pull_request:6355", + "right": "pull_request:6787", + "accept": false, + "reason": "Unrelated fixes: one adjusts widget metadata, the other corrects Marigold input normalization." + }, + { + "left": "pull_request:7117", + "right": "pull_request:7541", + "accept": false, + "reason": "Both touch nightly workflows, but one adds Slack/reporting support while the other makes the workflow dispatchable for release tests." + }, + { + "left": "pull_request:7156", + "right": "pull_request:8229", + "accept": false, + "reason": "Completely different code paths: stale-bot label handling vs. Stable Video Diffusion resize behavior." + }, + { + "left": "pull_request:7246", + "right": "pull_request:8229", + "accept": false, + "reason": "These are unrelated tests/pipeline fixes; no shared underlying bug or change." + }, + { + "left": "pull_request:9735", + "right": "pull_request:9968", + "accept": false, + "reason": "One updates an SD3 ControlNet example, the other is a broad ControlNet module refactor; not the same concrete change." + }, + { + "left": "pull_request:11684", + "right": "pull_request:11823", + "accept": false, + "reason": "Different LoRA features and model families: GGUF single-file loading vs. Fal-Kontext LoRA conversion." + }, + { + "left": "pull_request:2561", + "right": "pull_request:3386", + "accept": false, + "reason": "Both mention ControlNet inpainting, but one adds community inpaint scripts and the other refactors core ControlNet pipelines; not mergeable as one fix." + }, + { + "left": "pull_request:9507", + "right": "pull_request:9985", + "accept": false, + "reason": "Both are Flux-related, but 9507 fixes a specific controlnet batch-addition bug while 9985 adds multiple Flux feature pipelines." + }, + { + "left": "pull_request:10747", + "right": "pull_request:9311", + "accept": false, + "reason": "Different ControlNet changes: union model support vs. SDXL CFG cutoff callback." + }, + { + "left": "pull_request:10324", + "right": "pull_request:8797", + "accept": false, + "reason": "Unrelated areas: Kandinsky CPU offload API fix vs. test sharding behavior." + }, + { + "left": "pull_request:11869", + "right": "pull_request:12439", + "accept": false, + "reason": "Both mention kernels, but GGUF CUDA inference and attention backend support are separate features with different code paths." + }, + { + "left": "pull_request:2421", + "right": "pull_request:819", + "accept": false, + "reason": "Unrelated fixes in different subsystems: deprecate-message cleanup vs. DDIM timestep dtype on Windows." + }, + { + "left": "pull_request:4690", + "right": "pull_request:6006", + "accept": false, + "reason": "Related scheduler area, but not the same change: a draft DPMSolver refactor vs. a broad addition of `self.sigmas` across schedulers." + }, + { + "left": "pull_request:6929", + "right": "pull_request:7086", + "accept": false, + "reason": "Different community-pipeline changes: custom-pipeline loading docs vs. AnimateDiff img2vid/controlnet updates." + }, + { + "left": "pull_request:10119", + "right": "pull_request:10622", + "accept": false, + "reason": "Both are docs-only, but they cover different topics: LoRA adapter loading vs. uv installation." + }, + { + "left": "pull_request:10188", + "right": "pull_request:9257", + "accept": false, + "reason": "Different problems: Flux LoRA loading/expansion test behavior vs. DreamBooth Flux training script bugs." + }, + { + "left": "pull_request:10376", + "right": "pull_request:10808", + "accept": false, + "reason": "Separate LoRA work: HunyuanVideo format conversion vs. PEFT config key filtering." + }, + { + "left": "pull_request:10378", + "right": "pull_request:10473", + "accept": false, + "reason": "One adds distributed training support for SD3 LoRA; the other is just README wording updates." + }, + { + "left": "pull_request:1038", + "right": "pull_request:2593", + "accept": false, + "reason": "Different tasks entirely: Flax dtype fix vs. ControlNet checkpoint conversion tooling." + }, + { + "left": "pull_request:10579", + "right": "pull_request:11261", + "accept": false, + "reason": "Unrelated additions: an attentive eraser community pipeline vs. new VidTok autoencoder models." + }, + { + "left": "pull_request:10808", + "right": "pull_request:12040", + "accept": false, + "reason": "Different LoRA concerns: PEFT config adjustment vs. Wan Lightx2v LoRA/model support." + }, + { + "left": "pull_request:10966", + "right": "pull_request:11124", + "accept": false, + "reason": "The test PR is only a harness update; the bug-fix PR changes CogView4 input validation, so they are not the same change." + }, + { + "left": "pull_request:10966", + "right": "pull_request:9426", + "accept": false, + "reason": "Unrelated: CogView4 validation logic vs. CI artifact uploader version bump." + }, + { + "left": "pull_request:11039", + "right": "pull_request:11273", + "accept": false, + "reason": "Completely unrelated: single-file torch_dtype handling vs. Slack notification bot timeout handling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 204, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11290", + "pull_request:11366", + "pull_request:1147", + "pull_request:11504", + "pull_request:11728", + "pull_request:11894", + "pull_request:1192", + "pull_request:12151", + "pull_request:12491", + "pull_request:12493", + "pull_request:12541", + "pull_request:12973", + "pull_request:130", + "pull_request:13042", + "pull_request:13150", + "pull_request:13255", + "pull_request:1338", + "pull_request:1494", + "pull_request:1525", + "pull_request:1706", + "pull_request:1945", + "pull_request:1956", + "pull_request:2048", + "pull_request:2341", + "pull_request:2584", + "pull_request:2595", + "pull_request:26", + "pull_request:3334", + "pull_request:3471", + "pull_request:3476", + "pull_request:3586", + "pull_request:3589", + "pull_request:3614", + "pull_request:3675", + "pull_request:3716", + "pull_request:4034", + "pull_request:4044", + "pull_request:4177", + "pull_request:4188", + "pull_request:5299", + "pull_request:538", + "pull_request:5702", + "pull_request:5741", + "pull_request:5747", + "pull_request:6450", + "pull_request:6511", + "pull_request:788" + ], + "soft_pairs": [ + "pull_request:11290|pull_request:11728", + "pull_request:11366|pull_request:11504", + "pull_request:1147|pull_request:130", + "pull_request:11894|pull_request:12491", + "pull_request:1192|pull_request:2048", + "pull_request:12151|pull_request:12973", + "pull_request:12493|pull_request:12541", + "pull_request:12493|pull_request:13255", + "pull_request:13042|pull_request:13150", + "pull_request:1338|pull_request:788", + "pull_request:1494|pull_request:2341", + "pull_request:1525|pull_request:26", + "pull_request:1706|pull_request:538", + "pull_request:1945|pull_request:3614", + "pull_request:1956|pull_request:3334", + "pull_request:2584|pull_request:3586", + "pull_request:2595|pull_request:3675", + "pull_request:3471|pull_request:5299", + "pull_request:3476|pull_request:3716", + "pull_request:3589|pull_request:5747", + "pull_request:4034|pull_request:6450", + "pull_request:4044|pull_request:5702", + "pull_request:4177|pull_request:6511", + "pull_request:4188|pull_request:5741" + ], + "prepared_review_unit_hash": "418c6ad1ce79cf07ed308a1592ce9f36a7f6b9b3af16b1d387b88f7cbb268217", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 91919, + "estimated_input_tokens": 22980, + "estimated_eval_tokens": 46216 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are only superficially similar; they span unrelated PRs across Dockerfiles, docs, training scripts, schedulers, loaders, and tests. None look like the same concrete bug or change, so all soft edges should be rejected.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:11290", + "right": "pull_request:11728", + "accept": false, + "reason": "Both touch Dockerfiles, but one unpins torch in an onnxruntime CPU image while the other adds quantization libraries to the doc-builder image; different purpose and code path." + }, + { + "left": "pull_request:11366", + "right": "pull_request:11504", + "accept": false, + "reason": "Hidream transformer output-shape fix vs dependency-table/CI update; no shared underlying bug or change." + }, + { + "left": "pull_request:1147", + "right": "pull_request:130", + "accept": false, + "reason": "Flax text encoder loading fix vs adding accelerate config docs; unrelated feature areas and no duplicate behavior." + }, + { + "left": "pull_request:11894", + "right": "pull_request:12491", + "accept": false, + "reason": "k-diffusion pinning for CI is unrelated to the documentation link-check workflow change." + }, + { + "left": "pull_request:1192", + "right": "pull_request:2048", + "accept": false, + "reason": "General loading edge cases in pipeline utils vs accelerate checkpoint hooks and Dreambooth training flow; different problem statements." + }, + { + "left": "pull_request:12151", + "right": "pull_request:12973", + "accept": false, + "reason": "Deprecating Flax support is the opposite of adding GLM-Image support; clearly different changes." + }, + { + "left": "pull_request:12493", + "right": "pull_request:12541", + "accept": false, + "reason": "LTX-Video pipeline/docs addition vs Wan LoRA conversion fix; different models and code paths." + }, + { + "left": "pull_request:12493", + "right": "pull_request:13255", + "accept": false, + "reason": "LTX-Video support and Z-Image LoRA-loading fix are unrelated model/pipeline changes." + }, + { + "left": "pull_request:13042", + "right": "pull_request:13150", + "accept": false, + "reason": "QwenImage prompt-isolation fix vs test helper change for recompile limits; not the same bug or feature." + }, + { + "left": "pull_request:1338", + "right": "pull_request:788", + "accept": false, + "reason": "ONNX inpainting cast-order fix vs PIL sampler warning fix in different pipelines; not duplicative." + }, + { + "left": "pull_request:1494", + "right": "pull_request:2341", + "accept": false, + "reason": "Safetensors save support vs EMA checkpoint-saving test adjustments; different functionality and no shared code defect." + }, + { + "left": "pull_request:1525", + "right": "pull_request:26", + "accept": false, + "reason": "Upscaling batch-size fix vs DDPM alpha_cumprod fix; both numerical but entirely different algorithms and code paths." + }, + { + "left": "pull_request:1706", + "right": "pull_request:538", + "accept": false, + "reason": "Main docs cleanup vs hub dependency version bump; documentation and packaging changes are unrelated." + }, + { + "left": "pull_request:1945", + "right": "pull_request:3614", + "accept": false, + "reason": "Adding 2D UNet block tests vs tweaking an existing test file/comment; not the same underlying bug/change." + }, + { + "left": "pull_request:1956", + "right": "pull_request:3334", + "accept": false, + "reason": "Imagic typo fix in example code vs pipeline docstring cleanup; different files and no common bug." + }, + { + "left": "pull_request:2584", + "right": "pull_request:3586", + "accept": false, + "reason": "Community ControlNet img2img pipeline vs StableDiffusionTilingPipeline community example; separate additions." + }, + { + "left": "pull_request:2595", + "right": "pull_request:3675", + "accept": false, + "reason": "Scheduler switching/config normalization PR vs a test fix for V100; unrelated." + }, + { + "left": "pull_request:3471", + "right": "pull_request:5299", + "accept": false, + "reason": "Dreambooth docs note about torch.compile vs typehint/docs improvements in models; not the same change." + }, + { + "left": "pull_request:3476", + "right": "pull_request:3716", + "accept": false, + "reason": "Adding use_karras_sigmas to a scheduler vs introducing ParaDiGMS pipeline and parallel schedulers; different feature areas." + }, + { + "left": "pull_request:3589", + "right": "pull_request:5747", + "accept": false, + "reason": "README/reference-tricks docs update vs dataset data_dir parameter in training scripts; unrelated docs vs training input handling." + }, + { + "left": "pull_request:4034", + "right": "pull_request:6450", + "accept": false, + "reason": "LoRA unload utility implementation vs LoRA deprecation test refactor; same broad subsystem but not the same concrete bug/change." + }, + { + "left": "pull_request:4044", + "right": "pull_request:5702", + "accept": false, + "reason": "AudioLDM example docstring typo fix vs AnimateDiff prompt bug fix; different pipelines and issues." + }, + { + "left": "pull_request:4177", + "right": "pull_request:6511", + "accept": false, + "reason": "Making safetensors required dependency vs torch.compile-compatible checkpointing for SDXL LoRA training; unrelated." + }, + { + "left": "pull_request:4188", + "right": "pull_request:5741", + "accept": false, + "reason": "SDXL ControlNet training cache/fingerprint fix vs speeding up PEFT LoRA unload; different code paths and goals." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 205, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10126", + "pull_request:10127", + "pull_request:12933", + "pull_request:12934", + "pull_request:4214", + "pull_request:4595", + "pull_request:4944", + "pull_request:5030", + "pull_request:5157", + "pull_request:5388", + "pull_request:5424", + "pull_request:5653", + "pull_request:5665", + "pull_request:5713", + "pull_request:5754", + "pull_request:5788", + "pull_request:5823", + "pull_request:6081", + "pull_request:6195", + "pull_request:6231", + "pull_request:6332", + "pull_request:6432", + "pull_request:6448", + "pull_request:6490", + "pull_request:6511", + "pull_request:6772", + "pull_request:6884", + "pull_request:689", + "pull_request:7016", + "pull_request:7115", + "pull_request:7200", + "pull_request:7359", + "pull_request:7458", + "pull_request:7638", + "pull_request:7652", + "pull_request:819", + "pull_request:8303", + "pull_request:8404", + "pull_request:8512", + "pull_request:8566", + "pull_request:8722", + "pull_request:8932", + "pull_request:8973", + "pull_request:9400", + "pull_request:9827", + "pull_request:9896", + "pull_request:9987" + ], + "soft_pairs": [ + "pull_request:4214|pull_request:5754", + "pull_request:4595|pull_request:6195", + "pull_request:5030|pull_request:5424", + "pull_request:5157|pull_request:6448", + "pull_request:5157|pull_request:6772", + "pull_request:5665|pull_request:6772", + "pull_request:5788|pull_request:6231", + "pull_request:6081|pull_request:6332", + "pull_request:6332|pull_request:8404", + "pull_request:6432|pull_request:7638", + "pull_request:6432|pull_request:8303", + "pull_request:6511|pull_request:7652", + "pull_request:689|pull_request:819", + "pull_request:7115|pull_request:8512", + "pull_request:7200|pull_request:8722", + "pull_request:7458|pull_request:7638", + "pull_request:8566|pull_request:9987", + "pull_request:9400|pull_request:9896", + "pull_request:4944|pull_request:5653", + "pull_request:6884|pull_request:7016", + "pull_request:12933|pull_request:12934", + "pull_request:5388|pull_request:5823", + "pull_request:6490|pull_request:7359", + "pull_request:5713|pull_request:7016", + "pull_request:10126|pull_request:9827", + "pull_request:10127|pull_request:8932", + "pull_request:10127|pull_request:8973" + ], + "prepared_review_unit_hash": "0c920336df852c3d431874248e774115307633986f3114f5037bbfa56fbaa7b8", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 103034, + "estimated_input_tokens": 25759, + "estimated_eval_tokens": 51774 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy soft-similarity matches across unrelated diffusers PRs. None of the candidate pairs look like true duplicates or the same concrete code change, so all soft edges should be rejected.", + "confidence": 0.69, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5713 is the broadest foundational change in the cluster: the main IP-Adapter feature rollout with wide pipeline/loaders impact and multiple later adapter-related follow-ups nearby. It is the best representative PR among an otherwise noisy set.", + "best_issue_reason": null, + "best_pr_reason": "PR #5713 is the strongest standalone representative of the cluster because it is substantive, merged, broadly referenced, and spans the widest relevant adapter-related surface area.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4214", + "right": "pull_request:5754", + "accept": false, + "reason": "Custom Diffusion bf16 fix vs docs/style cleanup; completely different changes and code paths." + }, + { + "left": "pull_request:4595", + "right": "pull_request:6195", + "accept": false, + "reason": "A WIP SDXL conversion effort vs AnimateDiff SDXL; unrelated features." + }, + { + "left": "pull_request:5030", + "right": "pull_request:5424", + "accept": false, + "reason": "Model offload key handling vs slow-test pipeline fetcher logic; no shared bug/change." + }, + { + "left": "pull_request:5157", + "right": "pull_request:6448", + "accept": false, + "reason": "ControlNet LoRA conversion fix vs LoRA test assertion cleanup; different objectives and diffs." + }, + { + "left": "pull_request:5157", + "right": "pull_request:6772", + "accept": false, + "reason": "ControlNet LoRA conversion fix vs ControlNet-XS implementation; same broad area but not the same concrete change." + }, + { + "left": "pull_request:5665", + "right": "pull_request:6772", + "accept": false, + "reason": "ControlNet img2img height/width preprocessing vs ControlNet-XS feature implementation; unrelated." + }, + { + "left": "pull_request:5788", + "right": "pull_request:6231", + "accept": false, + "reason": "LCM timestep error handling vs SDXL fp16 training fix; different subsystems and problems." + }, + { + "left": "pull_request:6081", + "right": "pull_request:6332", + "accept": false, + "reason": "SDXL sigma handling vs community Rerender-A-Video pipeline; unrelated." + }, + { + "left": "pull_request:6332", + "right": "pull_request:8404", + "accept": false, + "reason": "Rerender-A-Video community pipeline vs Latte video model/pipeline; different features." + }, + { + "left": "pull_request:6432", + "right": "pull_request:7638", + "accept": false, + "reason": "README learning-rate tweak vs CONTRIBUTING link fix; unrelated docs changes." + }, + { + "left": "pull_request:6432", + "right": "pull_request:8303", + "accept": false, + "reason": "DreamBooth README update vs philosophy typo fix; unrelated." + }, + { + "left": "pull_request:6511", + "right": "pull_request:7652", + "accept": false, + "reason": "torch.compile checkpointing support vs Docker buildx workflow change; no shared code path." + }, + { + "left": "pull_request:689", + "right": "pull_request:819", + "accept": false, + "reason": "Attention slow-test optimization vs DDIM int64 timestep fix; different bugs and code paths." + }, + { + "left": "pull_request:7115", + "right": "pull_request:8512", + "accept": false, + "reason": "Dockerfile package-install tweak vs deprecated mask helper removal; unrelated." + }, + { + "left": "pull_request:7200", + "right": "pull_request:8722", + "accept": false, + "reason": "Transformer2D norm/patch-input error handling vs embeddings refactor; not the same bug or change." + }, + { + "left": "pull_request:7458", + "right": "pull_request:7638", + "accept": false, + "reason": "Skip dynamo tests on Python 3.12 vs docs link correction; unrelated." + }, + { + "left": "pull_request:8566", + "right": "pull_request:9987", + "accept": false, + "reason": "SD3 ControlNet/Multi-ControlNet vs SD3.5 IP-Adapter integration; both SD3-related but distinct features and code paths." + }, + { + "left": "pull_request:9400", + "right": "pull_request:9896", + "accept": false, + "reason": "Device-agnostic single-file pipeline tests vs Flux device/typecast revert; unrelated." + }, + { + "left": "pull_request:4944", + "right": "pull_request:5653", + "accept": false, + "reason": "IP-Adapter support feature vs PEFT-aware maybe_raise_or_warn helper; same area only loosely, not the same change." + }, + { + "left": "pull_request:6884", + "right": "pull_request:7016", + "accept": false, + "reason": "Scale-length validation in set_ip_adapter_scale vs image-embedding refactor/load skipping; separate IP-Adapter follow-ups, not one mergeable change." + }, + { + "left": "pull_request:12933", + "right": "pull_request:12934", + "accept": false, + "reason": "LTX-2 LoRA support vs LTX2 distilled checkpoint/two-stage generation support; related project, but distinct feature work." + }, + { + "left": "pull_request:5388", + "right": "pull_request:5823", + "accept": false, + "reason": "PEFT example-script adaptation vs adding adapter_names to fuse_lora; different LoRA/PEFT changes." + }, + { + "left": "pull_request:6490", + "right": "pull_request:7359", + "accept": false, + "reason": "RAVE video editing community pipeline vs ResAdapter training script; unrelated community additions." + }, + { + "left": "pull_request:5713", + "right": "pull_request:7016", + "accept": false, + "reason": "Initial IP-Adapter feature rollout vs later image-embed refactor/optimization; same subsystem, but not the same concrete code-path problem." + }, + { + "left": "pull_request:10126", + "right": "pull_request:9827", + "accept": false, + "reason": "save_attn_procs deprecation vs LoRA device-map loading fix; both LoRA-related but separate issues." + }, + { + "left": "pull_request:10127", + "right": "pull_request:8932", + "accept": false, + "reason": "xfail for incompatible SD configs vs temporal transformer test coverage; unrelated test changes." + }, + { + "left": "pull_request:10127", + "right": "pull_request:8973", + "accept": false, + "reason": "xfail for incompatible SD configs vs lumina model-size test reduction; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 206, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10230", + "pull_request:10342", + "pull_request:10506", + "pull_request:10539", + "pull_request:10719", + "pull_request:10966", + "pull_request:11124", + "pull_request:11292", + "pull_request:1136", + "pull_request:11613", + "pull_request:11823", + "pull_request:11877", + "pull_request:11892", + "pull_request:12275", + "pull_request:1229", + "pull_request:12315", + "pull_request:12530", + "pull_request:12782", + "pull_request:1309", + "pull_request:13120", + "pull_request:13330", + "pull_request:13332", + "pull_request:13364", + "pull_request:1508", + "pull_request:1564", + "pull_request:1706", + "pull_request:1910", + "pull_request:1945", + "pull_request:2002", + "pull_request:2062", + "pull_request:2122", + "pull_request:2274", + "pull_request:2325", + "pull_request:2347", + "pull_request:2467", + "pull_request:2481", + "pull_request:2793", + "pull_request:2795", + "pull_request:326", + "pull_request:3410", + "pull_request:3850", + "pull_request:4276", + "pull_request:4499", + "pull_request:4500", + "pull_request:689", + "pull_request:9244", + "pull_request:9687" + ], + "soft_pairs": [ + "pull_request:10230|pull_request:10719", + "pull_request:10342|pull_request:10966", + "pull_request:10506|pull_request:11124", + "pull_request:10539|pull_request:9244", + "pull_request:10719|pull_request:9687", + "pull_request:11292|pull_request:12315", + "pull_request:1136|pull_request:2002", + "pull_request:11613|pull_request:12530", + "pull_request:11823|pull_request:12275", + "pull_request:11877|pull_request:11892", + "pull_request:1229|pull_request:1309", + "pull_request:12782|pull_request:13364", + "pull_request:1309|pull_request:1945", + "pull_request:13120|pull_request:13332", + "pull_request:13330|pull_request:13332", + "pull_request:1508|pull_request:2122", + "pull_request:1508|pull_request:2793", + "pull_request:1564|pull_request:1910", + "pull_request:1706|pull_request:2795", + "pull_request:2062|pull_request:3410", + "pull_request:2274|pull_request:3850", + "pull_request:2274|pull_request:689", + "pull_request:2325|pull_request:4499", + "pull_request:2325|pull_request:4500", + "pull_request:2347|pull_request:4276", + "pull_request:2467|pull_request:326", + "pull_request:2481|pull_request:3850", + "pull_request:2481|pull_request:689" + ], + "prepared_review_unit_hash": "572b2b0a712e7229fb01a4381ccf23762513c3ce8bbb7309f19051028c66f131", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 100738, + "estimated_input_tokens": 25185, + "estimated_eval_tokens": 50626 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These PRs are a heterogeneous mix of unrelated docs, tests, workflow, and code changes; the soft-similarity edges look like false positives rather than duplicate fixes.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:10230", + "right": "pull_request:10719", + "accept": false, + "reason": "Unrelated docs-only LoRA mixin documentation vs ONNX provider_options plumbing; different change goals and code paths." + }, + { + "left": "pull_request:10342", + "right": "pull_request:10966", + "accept": false, + "reason": "Generic utility dtype helper update vs CogView4 input validation fix; no shared bug or mergeable common change." + }, + { + "left": "pull_request:10506", + "right": "pull_request:11124", + "accept": false, + "reason": "TorchAO sequential offload test vs CogView4 tokenizer test update; both are tests only and address different features." + }, + { + "left": "pull_request:10539", + "right": "pull_request:9244", + "accept": false, + "reason": "Both touch Flux, but one fixes true CFG gating in the pipeline while the other adds single-file loading support across Flux pipelines; separate changes." + }, + { + "left": "pull_request:10719", + "right": "pull_request:9687", + "accept": false, + "reason": "ONNX provider_options loading path vs Flux ControlNet model support; unrelated subsystems and functionality." + }, + { + "left": "pull_request:11292", + "right": "pull_request:12315", + "accept": false, + "reason": "LoRA SDXL conversion logic vs a sweep of scheduler selection changes in tests; not the same concrete bug or change." + }, + { + "left": "pull_request:1136", + "right": "pull_request:2002", + "accept": false, + "reason": "Old README link fix vs LoRA training documentation/code support; only broad docs overlap." + }, + { + "left": "pull_request:11613", + "right": "pull_request:12530", + "accept": false, + "reason": "CI/docker/test revert for torch.compile vs xfail of one stable-cascade test; completely different maintenance actions." + }, + { + "left": "pull_request:11823", + "right": "pull_request:12275", + "accept": false, + "reason": "Fal Kontext LoRA conversion/support implementation vs TorchAO quantization feature support; unrelated loader/quantization areas." + }, + { + "left": "pull_request:11877", + "right": "pull_request:11892", + "accept": false, + "reason": "Deprecating a pipeline class vs removing a dependency from setup; both chores, but not the same fix." + }, + { + "left": "pull_request:1229", + "right": "pull_request:1309", + "accept": false, + "reason": "DreamBooth install-doc wording vs broad PIL handling overhaul; same repository area, different underlying changes." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13364", + "accept": false, + "reason": "HunyuanVideo offloading test tweak vs MotionConv2d blur-kernel dtype fix; test adjustment is not the same code-path bug." + }, + { + "left": "pull_request:1309", + "right": "pull_request:1945", + "accept": false, + "reason": "PIL handling overhaul across pipelines vs adding 2D UNet block tests; unrelated scope." + }, + { + "left": "pull_request:13120", + "right": "pull_request:13332", + "accept": false, + "reason": "Modular pipeline AutoModel tagging tests vs dependency-testing workflow hardening; both test-related but different purposes and files." + }, + { + "left": "pull_request:13330", + "right": "pull_request:13332", + "accept": false, + "reason": "TorchAO tests/workflow changes vs dependency testing workflow hardening; no shared concrete bug fix." + }, + { + "left": "pull_request:1508", + "right": "pull_request:2122", + "accept": false, + "reason": "DreamBooth prior-preservation clarification vs safetensors docs; both documentation, but unrelated content." + }, + { + "left": "pull_request:1508", + "right": "pull_request:2793", + "accept": false, + "reason": "DreamBooth docs clarification vs cuDNN benchmark docs removal; different docs topics and goals." + }, + { + "left": "pull_request:1564", + "right": "pull_request:1910", + "accept": false, + "reason": "Default accelerate config guidance vs Korean docs init workflow; unrelated documentation changes." + }, + { + "left": "pull_request:1706", + "right": "pull_request:2795", + "accept": false, + "reason": "Main README/docs update vs text-to-image LoRA training code fix; not the same bug or feature." + }, + { + "left": "pull_request:2062", + "right": "pull_request:3410", + "accept": false, + "reason": "Paint-by-example CPU offload fix vs DeepSpeed plugin retrieval fix in a training script; different code paths and problems." + }, + { + "left": "pull_request:2274", + "right": "pull_request:3850", + "accept": false, + "reason": "Function naming clash fix in unconditional training vs audio diffusion test setup fix; unrelated examples and failures." + }, + { + "left": "pull_request:2274", + "right": "pull_request:689", + "accept": false, + "reason": "Unconditional training script naming fix vs attention performance change; completely different area and intent." + }, + { + "left": "pull_request:2325", + "right": "pull_request:4499", + "accept": false, + "reason": "ONNX VAE converter addition vs SDXL reference-attention community example; no shared underlying change." + }, + { + "left": "pull_request:2325", + "right": "pull_request:4500", + "accept": false, + "reason": "ONNX VAE converter addition vs duplicate SDXL reference example; unrelated artifacts." + }, + { + "left": "pull_request:2347", + "right": "pull_request:4276", + "accept": false, + "reason": "Scheduler noising schedule updates vs pipeline import test; different functionality and files." + }, + { + "left": "pull_request:2467", + "right": "pull_request:326", + "accept": false, + "reason": "Both are broad pipeline/model refactors, but one adds latent output handling and the other is type-hint cleanup; not the same concrete bug/change." + }, + { + "left": "pull_request:2481", + "right": "pull_request:3850", + "accept": false, + "reason": "Training script logger fix vs audio diffusion test adjustment; different failure modes and locations." + }, + { + "left": "pull_request:2481", + "right": "pull_request:689", + "accept": false, + "reason": "Training script logger version guard vs attention implementation change; no common code-path problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 207, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2641", + "pull_request:2793", + "pull_request:2994", + "pull_request:3034", + "pull_request:3080", + "pull_request:326", + "pull_request:3321", + "pull_request:3339", + "pull_request:3397", + "pull_request:3657", + "pull_request:3675", + "pull_request:3691", + "pull_request:3806", + "pull_request:3852", + "pull_request:4099", + "pull_request:4272", + "pull_request:4301", + "pull_request:453", + "pull_request:4572", + "pull_request:4690", + "pull_request:4862", + "pull_request:4987", + "pull_request:5056", + "pull_request:5526", + "pull_request:5672", + "pull_request:5837", + "pull_request:5898", + "pull_request:5974", + "pull_request:6231", + "pull_request:6330", + "pull_request:6432", + "pull_request:6834", + "pull_request:7096", + "pull_request:7309", + "pull_request:7738", + "pull_request:7790", + "pull_request:7938", + "pull_request:816", + "pull_request:826", + "pull_request:8660", + "pull_request:8875", + "pull_request:8964", + "pull_request:9060", + "pull_request:9175", + "pull_request:9447", + "pull_request:9735", + "pull_request:9918", + "pull_request:9938" + ], + "soft_pairs": [ + "pull_request:2641|pull_request:3397", + "pull_request:2793|pull_request:3691", + "pull_request:2994|pull_request:3657", + "pull_request:3034|pull_request:4572", + "pull_request:3080|pull_request:3675", + "pull_request:326|pull_request:826", + "pull_request:3321|pull_request:3339", + "pull_request:3675|pull_request:5526", + "pull_request:3806|pull_request:5056", + "pull_request:3852|pull_request:5898", + "pull_request:4099|pull_request:6231", + "pull_request:4272|pull_request:4987", + "pull_request:4301|pull_request:6432", + "pull_request:453|pull_request:816", + "pull_request:4690|pull_request:4862", + "pull_request:5672|pull_request:5974", + "pull_request:5837|pull_request:7096", + "pull_request:6330|pull_request:6834", + "pull_request:7309|pull_request:9060", + "pull_request:7738|pull_request:8964", + "pull_request:7790|pull_request:9447", + "pull_request:7938|pull_request:9735", + "pull_request:8660|pull_request:9938", + "pull_request:9735|pull_request:9918", + "pull_request:8875|pull_request:9175" + ], + "prepared_review_unit_hash": "36fcc3a8268a85c169a3529def5e33314ae358618cf09eff323813aff1df37f2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 91985, + "estimated_input_tokens": 22997, + "estimated_eval_tokens": 46250 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a heterogeneous grab-bag of unrelated docs, examples, loader, scheduler, and pipeline changes. None of the soft pairs look like true duplicates; they mostly share only superficial wording patterns.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the items span unrelated docs, examples, schedulers, loaders, and model/pipeline changes.", + "best_issue_reason": null, + "best_pr_reason": "PR 9175 is the most substantial and highest-discussion item, so it is the best cluster representative even though it is not a duplicate hub.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2641", + "right": "pull_request:3397", + "accept": false, + "reason": "Different doc tasks: JAX notebook build setup vs a LoRA training doc note." + }, + { + "left": "pull_request:2793", + "right": "pull_request:3691", + "accept": false, + "reason": "Unrelated docs removal vs a code change in LoRA text-encoder loading." + }, + { + "left": "pull_request:2994", + "right": "pull_request:3657", + "accept": false, + "reason": "Different bugs in different code paths: ControlNet type casting vs checkpoint parser tweaks." + }, + { + "left": "pull_request:3034", + "right": "pull_request:4572", + "accept": false, + "reason": "Same theme of parameter-name fixes, but different scripts and APIs; not one concrete fix." + }, + { + "left": "pull_request:3080", + "right": "pull_request:3675", + "accept": false, + "reason": "Pipeline save logic vs a test-only UniDiffuser fix; unrelated changes." + }, + { + "left": "pull_request:326", + "right": "pull_request:826", + "accept": false, + "reason": "Old type-hint refactor across many files vs DreamBooth fp16 loss casting; not the same bug." + }, + { + "left": "pull_request:3321", + "right": "pull_request:3339", + "accept": false, + "reason": "Docs update vs dependency pinning; no shared underlying change." + }, + { + "left": "pull_request:3675", + "right": "pull_request:5526", + "accept": false, + "reason": "A V100 test fix is unrelated to scheduler order handling." + }, + { + "left": "pull_request:3806", + "right": "pull_request:5056", + "accept": false, + "reason": "Checkpoint-loading note vs LoRA fusion documentation; different documentation topics." + }, + { + "left": "pull_request:3852", + "right": "pull_request:5898", + "accept": false, + "reason": "Contributor credit note vs a prose cleanup in unrelated docs." + }, + { + "left": "pull_request:4099", + "right": "pull_request:6231", + "accept": false, + "reason": "SDXL prompt-embedding bug fix vs LoRA fp16 training stability; different code paths." + }, + { + "left": "pull_request:4272", + "right": "pull_request:4987", + "accept": false, + "reason": "Missing optional imports cleanup vs removing a duplicate GLIGEN export; unrelated maintenance." + }, + { + "left": "pull_request:4301", + "right": "pull_request:6432", + "accept": false, + "reason": "Kandinsky docs update vs an SDXL dreambooth learning-rate edit; no overlap." + }, + { + "left": "pull_request:453", + "right": "pull_request:816", + "accept": false, + "reason": "Installation typo in docs vs deepspeed loss casting in training code." + }, + { + "left": "pull_request:4690", + "right": "pull_request:4862", + "accept": false, + "reason": "Different scheduler redesign/refactor vs an SDXL ControlNet img2img height/width fix." + }, + { + "left": "pull_request:5672", + "right": "pull_request:5974", + "accept": false, + "reason": "PixArt non-square image support vs adapting PixArtAlpha for a different model variant." + }, + { + "left": "pull_request:5837", + "right": "pull_request:7096", + "accept": false, + "reason": "Textual inversion multi-token support vs ControlNet example enhancements; unrelated functionality." + }, + { + "left": "pull_request:6330", + "right": "pull_request:6834", + "accept": false, + "reason": "SVD docs rendering fix vs callback API typo fix; different docs-only changes." + }, + { + "left": "pull_request:7309", + "right": "pull_request:9060", + "accept": false, + "reason": "Scheduler order validation vs removing a stray debug print in Flux transformer." + }, + { + "left": "pull_request:7738", + "right": "pull_request:8964", + "accept": false, + "reason": "LoRA device placement fix vs AudioLDM2 cache-position fix; different subsystems." + }, + { + "left": "pull_request:7790", + "right": "pull_request:9447", + "accept": false, + "reason": "Community SDXL pipeline adaptation vs pipeline save shard-size behavior; not the same issue." + }, + { + "left": "pull_request:7938", + "right": "pull_request:9735", + "accept": false, + "reason": "Docker Slack reporting workflow vs SD3 ControlNet example updates; unrelated CI vs example code." + }, + { + "left": "pull_request:8660", + "right": "pull_request:9938", + "accept": false, + "reason": "SD3 ControlNet input validation vs SDXL reference pipeline update; different pipelines and bugs." + }, + { + "left": "pull_request:9735", + "right": "pull_request:9918", + "accept": false, + "reason": "SD3 ControlNet example update vs adding a server example; no common code path." + }, + { + "left": "pull_request:8875", + "right": "pull_request:9175", + "accept": false, + "reason": "Different feature work: PAG inpaint support vs Flux Union ControlNet support; not mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 208, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10072", + "pull_request:10292", + "pull_request:10552", + "pull_request:10719", + "pull_request:10747", + "pull_request:10911", + "pull_request:10951", + "pull_request:11004", + "pull_request:11614", + "pull_request:11818", + "pull_request:11823", + "pull_request:11975", + "pull_request:12424", + "pull_request:1282", + "pull_request:1309", + "pull_request:13217", + "pull_request:13229", + "pull_request:1564", + "pull_request:1579", + "pull_request:1787", + "pull_request:1920", + "pull_request:2345", + "pull_request:2367", + "pull_request:2389", + "pull_request:2452", + "pull_request:2565", + "pull_request:26", + "pull_request:274", + "pull_request:3252", + "pull_request:3632", + "pull_request:3831", + "pull_request:4078", + "pull_request:4099", + "pull_request:4262", + "pull_request:4404", + "pull_request:4408", + "pull_request:4461", + "pull_request:453", + "pull_request:4847", + "pull_request:4910", + "pull_request:5127", + "pull_request:530", + "pull_request:5695", + "pull_request:689", + "pull_request:6946", + "pull_request:9212", + "pull_request:9244", + "pull_request:9507" + ], + "soft_pairs": [ + "pull_request:10747|pull_request:10951", + "pull_request:13217|pull_request:13229", + "pull_request:5127|pull_request:6946", + "pull_request:10552|pull_request:11823", + "pull_request:3632|pull_request:4408", + "pull_request:10072|pull_request:9507", + "pull_request:10292|pull_request:9212", + "pull_request:10719|pull_request:9244", + "pull_request:10911|pull_request:11975", + "pull_request:11004|pull_request:11818", + "pull_request:11614|pull_request:12424", + "pull_request:1282|pull_request:2389", + "pull_request:1309|pull_request:274", + "pull_request:1564|pull_request:3252", + "pull_request:1579|pull_request:274", + "pull_request:1787|pull_request:689", + "pull_request:1920|pull_request:2452", + "pull_request:2345|pull_request:4461", + "pull_request:2367|pull_request:3252", + "pull_request:2565|pull_request:453", + "pull_request:26|pull_request:530", + "pull_request:3831|pull_request:4847", + "pull_request:4078|pull_request:4910", + "pull_request:4099|pull_request:5695", + "pull_request:4262|pull_request:4404" + ], + "prepared_review_unit_hash": "c1bbf9cd84a6a07b525305cac40679291cefcde71631d2b359ba294dac688ab0", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 97771, + "estimated_input_tokens": 24443, + "estimated_eval_tokens": 49142 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose similarity cluster of mostly unrelated PRs across docs, pipelines, loaders, CI, and model support. The shared-file or shared-subsystem overlaps are coincidental rather than the same concrete bug/change, so all soft edges should be rejected.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the items are heterogeneous and do not share one underlying fix or feature.", + "best_issue_reason": null, + "best_pr_reason": "PR 13217 is the most substantial and coherent standalone change in the set, with the strongest review/discussion footprint, so it is the best representative if one must pick a single PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10747", + "right": "pull_request:10951", + "accept": false, + "reason": "Both are SDXL ControlNet-related, but one adds a union model and the other adds a tile upscaler pipeline; different features and codepaths." + }, + { + "left": "pull_request:13217", + "right": "pull_request:13229", + "accept": false, + "reason": "LTX-2.3 model support vs Sana-video VAE integration; different model families and conversion paths." + }, + { + "left": "pull_request:5127", + "right": "pull_request:6946", + "accept": false, + "reason": "MultiAdapter support and IP-Adapter low_cpu_mem_usage loading are unrelated loader/pipeline changes." + }, + { + "left": "pull_request:10552", + "right": "pull_request:11823", + "accept": false, + "reason": "From_single_file support for StableDiffusionAdapterPipeline is unrelated to Flux Kontext LoRA conversion support." + }, + { + "left": "pull_request:3632", + "right": "pull_request:4408", + "accept": false, + "reason": "Panorama view batching and SDXL from_single_file inpainting support are separate changes." + }, + { + "left": "pull_request:10072", + "right": "pull_request:9507", + "accept": false, + "reason": "SD3.5 controlnet offloading and Flux controlnet batch handling are different bugs in different codepaths." + }, + { + "left": "pull_request:10292", + "right": "pull_request:9212", + "accept": false, + "reason": "Same file, but distinct registrations for different pipeline families; not the same change." + }, + { + "left": "pull_request:10719", + "right": "pull_request:9244", + "accept": false, + "reason": "ONNX provider_options plumbing and Flux single-file support are unrelated loader features." + }, + { + "left": "pull_request:10911", + "right": "pull_request:11975", + "accept": false, + "reason": "A Lumina2 LoRA test fix and a training-script style fix are not the same underlying bug." + }, + { + "left": "pull_request:11004", + "right": "pull_request:11818", + "accept": false, + "reason": "PEFT key-prefix cleanup and Flux denoising sync removal affect different subsystems and problems." + }, + { + "left": "pull_request:11614", + "right": "pull_request:12424", + "accept": false, + "reason": "Both touch docker/CI, but one fixes workflow file handling and the other changes Dockerfile definitions; not one concrete change." + }, + { + "left": "pull_request:1282", + "right": "pull_request:2389", + "accept": false, + "reason": "Different documentation typos in different pages; too granular and unrelated to merge." + }, + { + "left": "pull_request:1309", + "right": "pull_request:274", + "accept": false, + "reason": "PIL handling improvements across many components and adding an SD README are unrelated." + }, + { + "left": "pull_request:1564", + "right": "pull_request:3252", + "accept": false, + "reason": "Default accelerate config docs and stateful offloading notes are different documentation topics." + }, + { + "left": "pull_request:1579", + "right": "pull_request:274", + "accept": false, + "reason": "Intel-optimized research examples and the Stable Diffusion README are unrelated additions." + }, + { + "left": "pull_request:1787", + "right": "pull_request:689", + "accept": false, + "reason": "UnCLIP num_images_per_prompt behavior and attention speed changes are different bugs/optimizations." + }, + { + "left": "pull_request:1920", + "right": "pull_request:2452", + "accept": false, + "reason": "A translation guide and a core API docs scaffold are separate doc additions." + }, + { + "left": "pull_request:2345", + "right": "pull_request:4461", + "accept": false, + "reason": "xFormers training warning and pipeline_class_name support in conversion scripts are unrelated fixes." + }, + { + "left": "pull_request:2367", + "right": "pull_request:3252", + "accept": false, + "reason": "Checkpoint limit plumbing in training scripts and offloading documentation are not the same change." + }, + { + "left": "pull_request:2565", + "right": "pull_request:453", + "accept": false, + "reason": "UNet1D doc correction and installation typo fix are unrelated documentation edits." + }, + { + "left": "pull_request:26", + "right": "pull_request:530", + "accept": false, + "reason": "A scheduler math bugfix and a flaky-test assertion change are different issues." + }, + { + "left": "pull_request:3831", + "right": "pull_request:4847", + "accept": false, + "reason": "New DDPM inversion pipeline and prompt-free diffusion community example are distinct artifacts." + }, + { + "left": "pull_request:4078", + "right": "pull_request:4910", + "accept": false, + "reason": "Accelerate enforcement for from_single_file and an inpaint docs typo are unrelated." + }, + { + "left": "pull_request:4099", + "right": "pull_request:5695", + "accept": false, + "reason": "SDXL prompt-embedding bugfix and PixArt masking-condition fix are different model-specific bugs." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4404", + "accept": false, + "reason": "SDXL DreamBooth LoRA fixes and removing textual inversion mentions from SDXL docs are not the same code-path problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 209, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10633", + "pull_request:11017", + "pull_request:11421", + "pull_request:11645", + "pull_request:1256", + "pull_request:1258", + "pull_request:12618", + "pull_request:1282", + "pull_request:13164", + "pull_request:2040", + "pull_request:241", + "pull_request:2994", + "pull_request:3137", + "pull_request:3189", + "pull_request:4617", + "pull_request:4632", + "pull_request:4783", + "pull_request:4833", + "pull_request:4910", + "pull_request:4927", + "pull_request:5057", + "pull_request:5085", + "pull_request:5393", + "pull_request:5407", + "pull_request:5506", + "pull_request:5615", + "pull_request:5813", + "pull_request:6099", + "pull_request:6208", + "pull_request:6787", + "pull_request:6841", + "pull_request:6893", + "pull_request:7134", + "pull_request:7213", + "pull_request:7632", + "pull_request:8616", + "pull_request:8685", + "pull_request:8702", + "pull_request:8722", + "pull_request:9049", + "pull_request:9065", + "pull_request:9175", + "pull_request:9311", + "pull_request:9423", + "pull_request:9569", + "pull_request:9918", + "pull_request:9988" + ], + "soft_pairs": [ + "pull_request:4632|pull_request:5393", + "pull_request:4783|pull_request:5813", + "pull_request:4833|pull_request:6099", + "pull_request:4910|pull_request:5407", + "pull_request:4927|pull_request:5615", + "pull_request:5085|pull_request:5506", + "pull_request:6208|pull_request:6787", + "pull_request:6841|pull_request:8702", + "pull_request:6841|pull_request:9065", + "pull_request:6893|pull_request:8722", + "pull_request:7134|pull_request:7213", + "pull_request:7632|pull_request:9569", + "pull_request:8685|pull_request:9918", + "pull_request:9049|pull_request:9988", + "pull_request:2040|pull_request:241", + "pull_request:4617|pull_request:5057", + "pull_request:8616|pull_request:9423", + "pull_request:9175|pull_request:9311", + "pull_request:10633|pull_request:11017", + "pull_request:11421|pull_request:11645", + "pull_request:1256|pull_request:1258", + "pull_request:12618|pull_request:13164", + "pull_request:1282|pull_request:2994", + "pull_request:3137|pull_request:3189" + ], + "prepared_review_unit_hash": "dbc07c4dcb7d7fa3b8d918109d04c0a0972211aabec166f6991c1762fe0c6eb9", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 97276, + "estimated_input_tokens": 24319, + "estimated_eval_tokens": 48894 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous: docs, CI, examples, and unrelated model/pipeline fixes. None of the soft edges except one look like true duplicates; they mostly share broad keywords only. The only likely duplicate pair is the two tiny pipeline-test PRs about adding negative_prompt handling.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents the cluster because the items span unrelated changes. If a representative is needed, pull_request:1258 is the clearest true-duplicate candidate in the only exact same-change pair.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:1258 is the best overall PR candidate because it is the cleaner, explicit version of the only near-identical change (negative_prompt handling in the pipeline test).", + "soft_edge_verdicts": [ + { + "left": "pull_request:4632", + "right": "pull_request:5393", + "accept": false, + "reason": "Different changes: SDXL LoRA checkpointing/casting bugfix vs unrelated HFApi docs/example updates." + }, + { + "left": "pull_request:4783", + "right": "pull_request:5813", + "accept": false, + "reason": "One is SDXL ControlNet CFG-cutoff docs/code, the other is a loaders docs reorganization; same subsystem, not same change." + }, + { + "left": "pull_request:4833", + "right": "pull_request:6099", + "accept": false, + "reason": "Unfuse LoRA bugfix vs TextToVideoZero constructor args; unrelated code paths." + }, + { + "left": "pull_request:4910", + "right": "pull_request:5407", + "accept": false, + "reason": "Inpainting docs typo fix vs mixed-precision training example fix; not the same underlying bug." + }, + { + "left": "pull_request:4927", + "right": "pull_request:5615", + "accept": false, + "reason": "Both are reverts, but they revert different features in different pipelines and address different problems." + }, + { + "left": "pull_request:5085", + "right": "pull_request:5506", + "accept": false, + "reason": "LoRA offloading incompatibility fix vs xformers/PEFT attention scale argument fix; distinct issues." + }, + { + "left": "pull_request:6208", + "right": "pull_request:6787", + "accept": false, + "reason": "Stable Diffusion latent dtype mismatch vs Marigold input range normalization; unrelated." + }, + { + "left": "pull_request:6841", + "right": "pull_request:8702", + "accept": false, + "reason": "BasicTransformerBlock parameter restore vs SD3 conversion script; no shared code-path bug." + }, + { + "left": "pull_request:6841", + "right": "pull_request:9065", + "accept": false, + "reason": "Ada-layer-norm parameter fix vs sentencepiece soft-dependency wiring; unrelated changes." + }, + { + "left": "pull_request:6893", + "right": "pull_request:8722", + "accept": false, + "reason": "3D UNet attention refactor vs embeddings module refactor; broad architectural overlap only." + }, + { + "left": "pull_request:7134", + "right": "pull_request:7213", + "accept": false, + "reason": "UNet type-hint cleanup vs LoRA docs update; not the same artifact or bug." + }, + { + "left": "pull_request:7632", + "right": "pull_request:9569", + "accept": false, + "reason": "DreamBooth VAE loading fix vs CogVideoX decode batch-size fix; different model paths." + }, + { + "left": "pull_request:8685", + "right": "pull_request:9918", + "accept": false, + "reason": "A documentation note about float8 vs a new server example; unrelated." + }, + { + "left": "pull_request:9049", + "right": "pull_request:9988", + "accept": false, + "reason": "Flux prompt-encoding fix when called separately vs broader Flux Redux conversion/architecture work; not mergeable as one bugfix." + }, + { + "left": "pull_request:2040", + "right": "pull_request:241", + "accept": false, + "reason": "InstructPix2Pix pipeline addition vs inpainting example script; different features." + }, + { + "left": "pull_request:4617", + "right": "pull_request:5057", + "accept": false, + "reason": "Encode-prompt refactor vs adding clip-skip support; same area but different change and not a single fix." + }, + { + "left": "pull_request:8616", + "right": "pull_request:9423", + "accept": false, + "reason": "SD3 LoRA fusion support vs Kolors differential diffusion example; unrelated." + }, + { + "left": "pull_request:9175", + "right": "pull_request:9311", + "accept": false, + "reason": "Flux ControlNet support vs SDXL ControlNet CFG cutoff callback; different pipelines and behaviors." + }, + { + "left": "pull_request:10633", + "right": "pull_request:11017", + "accept": false, + "reason": "Flux docs IP-Adapter example vs VAE encode hybrid-inference feature; unrelated docs/features." + }, + { + "left": "pull_request:11421", + "right": "pull_request:11645", + "accept": false, + "reason": "DreamBooth interpolation default change vs torch-version sentinel when torch is disabled; unrelated." + }, + { + "left": "pull_request:1256", + "right": "pull_request:1258", + "accept": true, + "reason": "Same tiny test change in the same file/test: both add negative_prompt handling to the pipeline test; 1258 is the clearer keyworded version." + }, + { + "left": "pull_request:12618", + "right": "pull_request:13164", + "accept": false, + "reason": "One is a uv install typo fix; the other is broader CI dependency pruning. Same workflow area, different change." + }, + { + "left": "pull_request:1282", + "right": "pull_request:2994", + "accept": false, + "reason": "Docs link fix vs ControlNet training type-casting fix; unrelated." + }, + { + "left": "pull_request:3137", + "right": "pull_request:3189", + "accept": false, + "reason": "UNet/class-label dtype cast vs AudioLDM waveform dtype fix; different code paths and symptoms." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative on the soft edges. The only accepted pair, 1256/1258, is plausibly the same bugfix in the same test file, while the other pairwise rejections match the shown diffs and titles. No obvious overstatement beyond minor wording around \u2018same-change,\u2019 which is acceptable here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 210, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10056", + "pull_request:10630", + "pull_request:10983", + "pull_request:11475", + "pull_request:11485", + "pull_request:11588", + "pull_request:11786", + "pull_request:11979", + "pull_request:11994", + "pull_request:11995", + "pull_request:12041", + "pull_request:12249", + "pull_request:12330", + "pull_request:1327", + "pull_request:1355", + "pull_request:1458", + "pull_request:1470", + "pull_request:241", + "pull_request:2557", + "pull_request:4352", + "pull_request:4461", + "pull_request:4712", + "pull_request:4714", + "pull_request:4793", + "pull_request:5021", + "pull_request:5085", + "pull_request:5115", + "pull_request:5261", + "pull_request:5299", + "pull_request:5823", + "pull_request:5880", + "pull_request:6236", + "pull_request:6424", + "pull_request:6673", + "pull_request:7062", + "pull_request:7158", + "pull_request:7216", + "pull_request:7632", + "pull_request:7924", + "pull_request:821", + "pull_request:823", + "pull_request:8688", + "pull_request:8722", + "pull_request:8952", + "pull_request:9099", + "pull_request:9473", + "pull_request:9638", + "pull_request:9838" + ], + "soft_pairs": [ + "pull_request:4352|pull_request:4793", + "pull_request:5021|pull_request:7216", + "pull_request:5085|pull_request:5823", + "pull_request:5299|pull_request:6236", + "pull_request:5880|pull_request:7632", + "pull_request:6673|pull_request:7062", + "pull_request:8722|pull_request:8952", + "pull_request:9099|pull_request:9638", + "pull_request:9473|pull_request:9838", + "pull_request:10056|pull_request:11485", + "pull_request:11475|pull_request:11994", + "pull_request:11979|pull_request:12330", + "pull_request:1327|pull_request:821", + "pull_request:1458|pull_request:823", + "pull_request:1470|pull_request:2557", + "pull_request:11485|pull_request:12041", + "pull_request:1355|pull_request:241", + "pull_request:4461|pull_request:5115", + "pull_request:7158|pull_request:8688", + "pull_request:10630|pull_request:11786", + "pull_request:4712|pull_request:6673", + "pull_request:4714|pull_request:5261", + "pull_request:10983|pull_request:11588", + "pull_request:11995|pull_request:12249", + "pull_request:6424|pull_request:7924" + ], + "prepared_review_unit_hash": "4340fa2873ff93cc33cab18ec7ee44e1c1e7f7787b8d8e79b71565dcdafa7229", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 102064, + "estimated_input_tokens": 25516, + "estimated_eval_tokens": 51288 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a heterogeneous mix of unrelated PRs across pipelines, LoRA/loading, schedulers, docs, CI, and tests. The soft pairs are mostly broad-subsystem similarities or shared filenames, not the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:4352", + "right": "pull_request:4793", + "accept": false, + "reason": "Different problems: one adds combined Kandinsky pipeline helpers, the other fixes kwargs propagation in auto_pipeline. Same general area only." + }, + { + "left": "pull_request:5021", + "right": "pull_request:7216", + "accept": false, + "reason": "Both touch conversion/mapping-related code, but one updates scheduler docs mapping and the other changes from_single_file signature/documentation. No shared bug." + }, + { + "left": "pull_request:5085", + "right": "pull_request:5823", + "accept": false, + "reason": "Both involve LoRA loaders, but one fixes incompatible LoRA offloading behavior and the other adds adapter_names support for fuse_lora. Distinct changes." + }, + { + "left": "pull_request:5299", + "right": "pull_request:6236", + "accept": false, + "reason": "Both are small fixes, but one is typehints/docs in model utilities and the other corrects ControlNet inpainting example values. Not the same issue." + }, + { + "left": "pull_request:5880", + "right": "pull_request:7632", + "accept": false, + "reason": "Both concern VAE-related loading, but one is a broad latent-diffusion compatibility fix and the other is a specific DreamBooth VAE path issue. Different code paths." + }, + { + "left": "pull_request:6673", + "right": "pull_request:7062", + "accept": false, + "reason": "Both are community example/doc updates, but they address different pipeline docstrings and different functionality. No duplicate bug/change." + }, + { + "left": "pull_request:8722", + "right": "pull_request:8952", + "accept": false, + "reason": "One is a large embeddings refactor; the other adds QKV fusion support. Same model area, but not the same change or bug." + }, + { + "left": "pull_request:9099", + "right": "pull_request:9638", + "accept": false, + "reason": "Both relate to ControlNet/SD3, but one adds SD3 controlnet inpainting while the other adds Xlabs controlnet support. Different features and code paths." + }, + { + "left": "pull_request:9473", + "right": "pull_request:9838", + "accept": false, + "reason": "Both touch Flux DreamBooth LoRA training, but one fixes LR handling for T5 in the base script and the other adjusts repetition/memory behavior in advanced training. Not the same bug." + }, + { + "left": "pull_request:10056", + "right": "pull_request:11485", + "accept": false, + "reason": "One adds Flux Redux prompt/multi-image support; the other adds Flex.2 pipeline files. Different pipelines and functionality." + }, + { + "left": "pull_request:11475", + "right": "pull_request:11994", + "accept": false, + "reason": "One fixes WAN recompilation behavior; the other fixes HF Hub login/test failures in docs and CI. No overlap in underlying issue." + }, + { + "left": "pull_request:11979", + "right": "pull_request:12330", + "accept": false, + "reason": "One enforces torch version for compile tests; the other cleans group offloading tests. Both are test maintenance only and unrelated." + }, + { + "left": "pull_request:1327", + "right": "pull_request:821", + "accept": false, + "reason": "Both are early pipeline additions, but one improves loading local custom pipelines and the other adds a mega Stable Diffusion example. Different artifacts and goals." + }, + { + "left": "pull_request:1458", + "right": "pull_request:823", + "accept": false, + "reason": "One adds an onnxruntime candidate package; the other fixes negative-prompt type mismatch across SD pipelines. Different subsystems." + }, + { + "left": "pull_request:1470", + "right": "pull_request:2557", + "accept": false, + "reason": "Both touch DreamBooth, but one adds AltDiffusion compatibility and the other adds intermediate logging. Different changes, not duplicate fixes." + }, + { + "left": "pull_request:11485", + "right": "pull_request:12041", + "accept": false, + "reason": "Different features: Flex.2 pipeline addition versus WAN VAE patch fix. Same broad release period only." + }, + { + "left": "pull_request:1355", + "right": "pull_request:241", + "accept": false, + "reason": "Both are pipeline additions, but one adds Flax img2img and the other adds an inpainting example script. Not the same underlying capability." + }, + { + "left": "pull_request:4461", + "right": "pull_request:5115", + "accept": false, + "reason": "Both modify conversion scripts, but one adds a pipeline_class_name argument and the other adds config_files parsing. Related tooling, different fixes." + }, + { + "left": "pull_request:7158", + "right": "pull_request:8688", + "accept": false, + "reason": "One fixes benchmark LCM test cleanup; the other fixes DoRA LoRA conversion. No shared bug or code path." + }, + { + "left": "pull_request:10630", + "right": "pull_request:11786", + "accept": false, + "reason": "Both are test/CI adjustments, but one updates Flax scheduler expected slices and the other changes the big accelerator marker. Different failures." + }, + { + "left": "pull_request:4712", + "right": "pull_request:6673", + "accept": false, + "reason": "One adds a tests fetcher utility; the other adds InstantID pipeline docs/example. No common change." + }, + { + "left": "pull_request:4714", + "right": "pull_request:5261", + "accept": false, + "reason": "One changes DummyObject error handling for missing transformers; the other fixes callback step indexing in community schedulers. Unrelated." + }, + { + "left": "pull_request:10983", + "right": "pull_request:11588", + "accept": false, + "reason": "Both touch Hunyuan/Wan-style model support, but one adds Hunyuan I2V and the other adds WAN LoRA support for I2V. Different models and fix scopes." + }, + { + "left": "pull_request:11995", + "right": "pull_request:12249", + "accept": false, + "reason": "One adds modular Flux text-to-image support; the other adds Step1X-Edit pipeline and converter. Different pipeline families." + }, + { + "left": "pull_request:6424", + "right": "pull_request:7924", + "accept": false, + "reason": "Both are model/pipeline fixes, but one is a PixArt weight conversion bug and the other is an IP-Adapter conditional kwargs fix in SDXL ControlNet inpaint. Not the same issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 211, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1", + "pull_request:10005", + "pull_request:10058", + "pull_request:10069", + "pull_request:10086", + "pull_request:101", + "pull_request:10867", + "pull_request:10970", + "pull_request:11130", + "pull_request:11231", + "pull_request:11564", + "pull_request:11660", + "pull_request:11797", + "pull_request:12024", + "pull_request:12074", + "pull_request:12109", + "pull_request:12119", + "pull_request:12387", + "pull_request:12584", + "pull_request:12723", + "pull_request:12897", + "pull_request:12919", + "pull_request:13063", + "pull_request:13208", + "pull_request:2129", + "pull_request:2385", + "pull_request:2721", + "pull_request:2954", + "pull_request:3351", + "pull_request:3386", + "pull_request:3476", + "pull_request:4098", + "pull_request:428", + "pull_request:5298", + "pull_request:5440", + "pull_request:7", + "pull_request:7359", + "pull_request:7541", + "pull_request:8412", + "pull_request:8425", + "pull_request:8640", + "pull_request:8739", + "pull_request:9099", + "pull_request:9507", + "pull_request:9542", + "pull_request:9573", + "pull_request:9988" + ], + "soft_pairs": [ + "pull_request:2385|pull_request:3351", + "pull_request:5298|pull_request:5440", + "pull_request:7359|pull_request:8739", + "pull_request:7541|pull_request:8640", + "pull_request:2129|pull_request:2954", + "pull_request:9099|pull_request:9573", + "pull_request:13063|pull_request:13208", + "pull_request:11130|pull_request:11797", + "pull_request:10867|pull_request:9542", + "pull_request:2721|pull_request:3476", + "pull_request:3386|pull_request:4098", + "pull_request:12109|pull_request:12119", + "pull_request:12584|pull_request:12723", + "pull_request:11231|pull_request:12119", + "pull_request:11660|pull_request:12387", + "pull_request:12024|pull_request:12074", + "pull_request:9507|pull_request:9988", + "pull_request:12897|pull_request:12919", + "pull_request:1|pull_request:7", + "pull_request:10005|pull_request:11564", + "pull_request:10058|pull_request:8412", + "pull_request:10069|pull_request:10970", + "pull_request:10086|pull_request:8425", + "pull_request:101|pull_request:428" + ], + "prepared_review_unit_hash": "dfff5b20b2c5b127c59986faed864f6d3c191804eca746b1044fe2e84f4c7be4", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 105234, + "estimated_input_tokens": 26309, + "estimated_eval_tokens": 52874 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These PRs are only superficially similar by subsystem or shared filenames. They cover distinct features, refactors, docs updates, and separate bug fixes, so none of the suggested soft edges should be merged as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:2385", + "right": "pull_request:3351", + "accept": false, + "reason": "One adds a Pix2Pix Zero utility method; the other adds Karras sigmas to LMSDiscreteScheduler. Different components and unrelated changes." + }, + { + "left": "pull_request:5298", + "right": "pull_request:5440", + "accept": false, + "reason": "These fix different LoRA/compile and single-file loading issues in different code paths; same broad area, but not the same bug." + }, + { + "left": "pull_request:7359", + "right": "pull_request:8739", + "accept": false, + "reason": "ResAdapter training script versus ConsistencyTTA community example. Both are community content, but the models and goals are unrelated." + }, + { + "left": "pull_request:7541", + "right": "pull_request:8640", + "accept": false, + "reason": "Workflow dispatch/release test changes versus a Flax scheduler test tolerance tweak. No shared underlying defect." + }, + { + "left": "pull_request:2129", + "right": "pull_request:2954", + "accept": false, + "reason": "One enables LoRA from pipeline; the other adds TextToVideoZeroPipeline. Both are pipeline additions, but not the same change or bug." + }, + { + "left": "pull_request:9099", + "right": "pull_request:9573", + "accept": false, + "reason": "Both touch SD3 controlnet/inpainting, but one adds SD3 controlnet inpainting support and the other fixes height/width divisibility. Different concrete issues." + }, + { + "left": "pull_request:13063", + "right": "pull_request:13208", + "accept": false, + "reason": "Wan modular pipeline refactor versus Helios video generation pipelines. Shared infra files do not indicate the same change." + }, + { + "left": "pull_request:11130", + "right": "pull_request:11797", + "accept": false, + "reason": "Pipeline-level quantization config is a general feature; Higgs quantization is a separate quantizer implementation and loading adjustment." + }, + { + "left": "pull_request:10867", + "right": "pull_request:9542", + "accept": false, + "reason": "Multi IP-Adapter for Flux and Kohya Flux LoRA text-encoder support are distinct Flux loader changes, not one bug fix." + }, + { + "left": "pull_request:2721", + "right": "pull_request:3476", + "accept": false, + "reason": "Model editing pipeline addition versus Karras sigmas for DPMSolverSinglestepScheduler. Different functionality entirely." + }, + { + "left": "pull_request:3386", + "right": "pull_request:4098", + "accept": false, + "reason": "ControlNet refactor/additions versus SDXL inpainting. They both expand pipelines, but they do not fix the same concrete problem." + }, + { + "left": "pull_request:12109", + "right": "pull_request:12119", + "accept": false, + "reason": "Custom sigma/timestep support in UniPCMultistepScheduler versus Qwen-Image-Lightning LoRA loading. Separate scheduler and loader work." + }, + { + "left": "pull_request:12584", + "right": "pull_request:12723", + "accept": false, + "reason": "Both are SANA-Video related, but one adds pretrained 5s inference and the other adds LongSANA minute-length generation with causal transformer changes. Not mergeable as one PR." + }, + { + "left": "pull_request:11231", + "right": "pull_request:12119", + "accept": false, + "reason": "HiDream Image pipeline/model support versus Qwen-Image Lightning LoRA loading. Different models and code paths." + }, + { + "left": "pull_request:11660", + "right": "pull_request:12387", + "accept": false, + "reason": "HiDream LoRA conversion support and FlashAttention-through-kernels support are unrelated features in different subsystems." + }, + { + "left": "pull_request:12024", + "right": "pull_request:12074", + "accept": false, + "reason": "Wan follow-up fixes versus Wan 2.2 second-transformer LoRA loading. Same model family, but different concrete behavior." + }, + { + "left": "pull_request:9507", + "right": "pull_request:9988", + "accept": false, + "reason": "Flux ControlNet control-mode batching fix versus Flux Redux conversion/runtime changes. Shared Flux branding only; not the same bug." + }, + { + "left": "pull_request:12897", + "right": "pull_request:12919", + "accept": false, + "reason": "NPU pos-embed optimization for Flux versus NPU attention backend support. Different NPU code paths and different objectives." + }, + { + "left": "pull_request:1", + "right": "pull_request:7", + "accept": false, + "reason": "These are foundational early PRs adding GLIDE modeling and VQ-VAE support. Both are initial model bring-up, but not the same artifact." + }, + { + "left": "pull_request:10005", + "right": "pull_request:11564", + "accept": false, + "reason": "Docs-only Flux pipeline additions versus a requirements update for DreamBooth. No shared underlying bug or feature." + }, + { + "left": "pull_request:10058", + "right": "pull_request:8412", + "accept": false, + "reason": "Both are docs fixes about single-file/loading or links, but they address different documentation pages and issues." + }, + { + "left": "pull_request:10069", + "right": "pull_request:10970", + "accept": false, + "reason": "BnB quantizer device handling versus PR style bot workflow swap. Completely unrelated." + }, + { + "left": "pull_request:10086", + "right": "pull_request:8425", + "accept": false, + "reason": "Stable Diffusion 3 Skip Guidance Layers bugfix versus a workflow name change. No overlap beyond being repository maintenance." + }, + { + "left": "pull_request:101", + "right": "pull_request:428", + "accept": false, + "reason": "Early core model/pipeline bootstrap PR versus general docs cleanup for using diffusers. Not the same change set." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 212, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10109", + "pull_request:10114", + "pull_request:10115", + "pull_request:10119", + "pull_request:10126", + "pull_request:10140", + "pull_request:10148", + "pull_request:1017", + "pull_request:10188", + "pull_request:10197", + "pull_request:102", + "pull_request:1020", + "pull_request:10206", + "pull_request:10218", + "pull_request:10220", + "pull_request:10245", + "pull_request:10246", + "pull_request:10251", + "pull_request:1027", + "pull_request:10293", + "pull_request:10358", + "pull_request:10506", + "pull_request:10730", + "pull_request:11096", + "pull_request:11129", + "pull_request:11475", + "pull_request:11531", + "pull_request:11635", + "pull_request:11647", + "pull_request:1885", + "pull_request:2007", + "pull_request:2322", + "pull_request:3172", + "pull_request:3230", + "pull_request:628", + "pull_request:70", + "pull_request:8497", + "pull_request:8600", + "pull_request:8627", + "pull_request:8774", + "pull_request:9021", + "pull_request:9036", + "pull_request:9116", + "pull_request:9266", + "pull_request:9286", + "pull_request:9709", + "pull_request:9769", + "pull_request:9893" + ], + "soft_pairs": [ + "pull_request:10109|pull_request:9286", + "pull_request:10114|pull_request:10730", + "pull_request:10114|pull_request:11475", + "pull_request:10115|pull_request:8497", + "pull_request:10119|pull_request:10245", + "pull_request:10126|pull_request:8774", + "pull_request:10140|pull_request:9769", + "pull_request:10148|pull_request:9893", + "pull_request:1017|pull_request:1885", + "pull_request:10188|pull_request:10506", + "pull_request:10197|pull_request:11096", + "pull_request:10197|pull_request:11129", + "pull_request:10197|pull_request:11531", + "pull_request:10197|pull_request:8600", + "pull_request:10197|pull_request:9266", + "pull_request:102|pull_request:2007", + "pull_request:102|pull_request:70", + "pull_request:1020|pull_request:2322", + "pull_request:1020|pull_request:3172", + "pull_request:1020|pull_request:628", + "pull_request:10206|pull_request:8627", + "pull_request:10218|pull_request:9036", + "pull_request:10218|pull_request:9116", + "pull_request:10220|pull_request:11635", + "pull_request:10245|pull_request:10358", + "pull_request:10246|pull_request:11647", + "pull_request:10251|pull_request:9021", + "pull_request:1027|pull_request:3230", + "pull_request:1027|pull_request:628", + "pull_request:10293|pull_request:9709" + ], + "prepared_review_unit_hash": "6dfc1f355b5426a481ac9cb65fa7fa4b730897116d54e3017bd83e8a327b651d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 98746, + "estimated_input_tokens": 24687, + "estimated_eval_tokens": 49630 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a set of mostly unrelated PRs that only share broad topical similarity (docs, CI, LoRA, new-model additions, small bugfixes). I don\u2019t see any pair that looks like the same underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a credible duplicate anchor here; the cluster is heterogeneous and the similarities are too coarse to treat any PR as canonical.", + "best_issue_reason": null, + "best_pr_reason": "If forced to pick a representative, PR #10140 is the broadest substantial code addition, but it is still not a duplicate target for the rest of the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10109", + "right": "pull_request:9286", + "accept": false, + "reason": "Different subsystems and goals: Stable Audio XLA support vs __init__ export cleanup." + }, + { + "left": "pull_request:10114", + "right": "pull_request:10730", + "accept": false, + "reason": "Formatting-only Stable Audio edit vs hook-unwrapping bugfix in pipeline utils." + }, + { + "left": "pull_request:10114", + "right": "pull_request:11475", + "accept": false, + "reason": "Stable Audio formatting cleanup vs WAN recompilation fix; unrelated code paths." + }, + { + "left": "pull_request:10115", + "right": "pull_request:8497", + "accept": false, + "reason": "Flux positional embedding bugfix vs spelling/doc cleanup in a scheduler; not the same change." + }, + { + "left": "pull_request:10119", + "right": "pull_request:10245", + "accept": false, + "reason": "Both are docs about LoRA APIs, but they cover different methods and different behavior." + }, + { + "left": "pull_request:10126", + "right": "pull_request:8774", + "accept": false, + "reason": "Deprecating save_attn_procs is separate from the LoraBaseMixin refactor." + }, + { + "left": "pull_request:10140", + "right": "pull_request:9769", + "accept": false, + "reason": "Different model launches (ConsisID vs Mochi); both are new feature additions, not duplicates." + }, + { + "left": "pull_request:10148", + "right": "pull_request:9893", + "accept": false, + "reason": "OmniGen implementation vs SDXL ControlNet reference community pipeline; distinct features." + }, + { + "left": "pull_request:1017", + "right": "pull_request:1885", + "accept": false, + "reason": "Unrelated DreamBooth script type-hint tweak vs a typo fix in DreamBooth inpaint." + }, + { + "left": "pull_request:10188", + "right": "pull_request:10506", + "accept": false, + "reason": "LoRA expansion logging/test fix vs TorchAO sequential offload test; different concerns." + }, + { + "left": "pull_request:10197", + "right": "pull_request:11096", + "accept": false, + "reason": "CI TPU workflow update vs Flux TPU README docs; different artifact types and changes." + }, + { + "left": "pull_request:10197", + "right": "pull_request:11129", + "accept": false, + "reason": "CI workflow changes vs modeling import cleanup; unrelated." + }, + { + "left": "pull_request:10197", + "right": "pull_request:11531", + "accept": false, + "reason": "CI workflow changes vs inpaint documentation tweak; unrelated." + }, + { + "left": "pull_request:10197", + "right": "pull_request:8600", + "accept": false, + "reason": "CI workflow changes vs DreamBooth script formatting; no shared bug/change." + }, + { + "left": "pull_request:10197", + "right": "pull_request:9266", + "accept": false, + "reason": "CI workflow changes vs community-projects docs entry; unrelated." + }, + { + "left": "pull_request:102", + "right": "pull_request:2007", + "accept": false, + "reason": "Training README/example updates vs broad Black formatting sweep; not the same underlying change." + }, + { + "left": "pull_request:102", + "right": "pull_request:70", + "accept": false, + "reason": "Training example docs vs an old UNet/attention refactor; unrelated." + }, + { + "left": "pull_request:1020", + "right": "pull_request:2322", + "accept": false, + "reason": "Training/fine-tuning docs update vs a separate inpaint example update." + }, + { + "left": "pull_request:1020", + "right": "pull_request:3172", + "accept": false, + "reason": "Both are docs, but one is training docs and the other is reproducibility guidance; not the same change." + }, + { + "left": "pull_request:1020", + "right": "pull_request:628", + "accept": false, + "reason": "Different documentation targets and content; no shared bug or patch." + }, + { + "left": "pull_request:10206", + "right": "pull_request:8627", + "accept": false, + "reason": "Flux Control LoRA unload support vs SD3 ControlNet docs/model work; unrelated." + }, + { + "left": "pull_request:10218", + "right": "pull_request:9036", + "accept": false, + "reason": "ControlNetUnion callback tensor fix vs nightly dependency pinning; unrelated." + }, + { + "left": "pull_request:10218", + "right": "pull_request:9116", + "accept": false, + "reason": "ControlNetUnion callback inputs fix vs a dead-link doc fix; unrelated." + }, + { + "left": "pull_request:10220", + "right": "pull_request:11635", + "accept": false, + "reason": "Mochi test rename vs Flux Kontext feature addition; unrelated." + }, + { + "left": "pull_request:10245", + "right": "pull_request:10358", + "accept": false, + "reason": "PEFT docs cleanup vs community projects listing update; different docs areas and goals." + }, + { + "left": "pull_request:10246", + "right": "pull_request:11647", + "accept": false, + "reason": "Missing AttnProcessors docs/code exposure vs adding Finegrained FP8 quantization; unrelated." + }, + { + "left": "pull_request:10251", + "right": "pull_request:9021", + "accept": false, + "reason": "Regional prompting typo fix vs SD3 ControlNet transform-layer bugfix; unrelated." + }, + { + "left": "pull_request:1027", + "right": "pull_request:3230", + "accept": false, + "reason": "DDPM training example option addition vs large IF pipeline implementation; not the same change." + }, + { + "left": "pull_request:1027", + "right": "pull_request:628", + "accept": false, + "reason": "Both touch training-example adjacent content, but one adds an argument and the other updates README launch commands." + }, + { + "left": "pull_request:10293", + "right": "pull_request:9709", + "accept": false, + "reason": "TorchAO version floor bump vs torch upper-bound pinning for CI; different dependency issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 213, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10312", + "pull_request:10318", + "pull_request:10324", + "pull_request:10332", + "pull_request:10351", + "pull_request:10364", + "pull_request:1041", + "pull_request:10420", + "pull_request:1043", + "pull_request:10481", + "pull_request:10497", + "pull_request:10502", + "pull_request:10506", + "pull_request:10537", + "pull_request:10544", + "pull_request:10548", + "pull_request:10554", + "pull_request:10576", + "pull_request:10579", + "pull_request:10651", + "pull_request:10852", + "pull_request:10889", + "pull_request:10898", + "pull_request:11004", + "pull_request:11124", + "pull_request:11196", + "pull_request:11249", + "pull_request:11262", + "pull_request:11319", + "pull_request:11564", + "pull_request:21", + "pull_request:22", + "pull_request:268", + "pull_request:2759", + "pull_request:3108", + "pull_request:545", + "pull_request:828", + "pull_request:886", + "pull_request:8932", + "pull_request:8964", + "pull_request:8973", + "pull_request:9077", + "pull_request:9244", + "pull_request:9709", + "pull_request:9739", + "pull_request:9821", + "pull_request:9827" + ], + "soft_pairs": [ + "pull_request:10312|pull_request:10651", + "pull_request:10318|pull_request:11004", + "pull_request:10324|pull_request:11124", + "pull_request:10332|pull_request:11564", + "pull_request:10351|pull_request:11196", + "pull_request:10351|pull_request:8964", + "pull_request:10364|pull_request:11249", + "pull_request:1041|pull_request:268", + "pull_request:1041|pull_request:2759", + "pull_request:1041|pull_request:3108", + "pull_request:1041|pull_request:886", + "pull_request:10420|pull_request:9077", + "pull_request:1043|pull_request:21", + "pull_request:1043|pull_request:22", + "pull_request:1043|pull_request:545", + "pull_request:1043|pull_request:828", + "pull_request:10481|pull_request:10537", + "pull_request:10497|pull_request:9709", + "pull_request:10502|pull_request:10579", + "pull_request:10506|pull_request:11196", + "pull_request:10506|pull_request:8932", + "pull_request:10506|pull_request:8973", + "pull_request:10544|pull_request:11319", + "pull_request:10548|pull_request:10554", + "pull_request:10548|pull_request:9077", + "pull_request:10548|pull_request:9739", + "pull_request:10554|pull_request:10852", + "pull_request:10554|pull_request:10889", + "pull_request:10554|pull_request:11262", + "pull_request:10576|pull_request:9821", + "pull_request:10576|pull_request:9827", + "pull_request:10579|pull_request:10898", + "pull_request:10579|pull_request:9244" + ], + "prepared_review_unit_hash": "4d9607cc9131acaaf5b1e5df2e3d90e8671b22c3635d6cf590ec22de29a0ccbc", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 33, + "serialized_chars": 90710, + "estimated_input_tokens": 22678, + "estimated_eval_tokens": 45612 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated PRs (tests, docs, small fixes, and feature work). The only vaguely related area is LoRA/device-map handling, but the paired items are still fix-vs-test or different subproblems, so I would not merge any soft edges.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9827 is the strongest canonical anchor if one must pick a representative: it is a substantive code fix around LoRA/device-map behavior, with broad downstream impact and an explicit problem statement.", + "best_issue_reason": null, + "best_pr_reason": "PR 9827 is the best single PR candidate because it addresses a concrete runtime bug in shared loading logic, unlike the mostly test/doc-only items around it.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10312", + "right": "pull_request:10651", + "accept": false, + "reason": "Different bugs in different pipelines: Attention reshape docs vs Lumina input validation." + }, + { + "left": "pull_request:10318", + "right": "pull_request:11004", + "accept": false, + "reason": "One is a Flux LoRA integration test, the other changes PEFT prefix handling; related topic, not the same fix." + }, + { + "left": "pull_request:10324", + "right": "pull_request:11124", + "accept": false, + "reason": "Kandinsky offload test helper vs a CogView4 tokenizer test; unrelated changes." + }, + { + "left": "pull_request:10332", + "right": "pull_request:11564", + "accept": false, + "reason": "Shape error message fix vs adding a dependency; no shared underlying change." + }, + { + "left": "pull_request:10351", + "right": "pull_request:11196", + "accept": false, + "reason": "LoRA test cleanup vs save_pretrained serialization behavior; unrelated." + }, + { + "left": "pull_request:10351", + "right": "pull_request:8964", + "accept": false, + "reason": "Different code paths: LoRA fuse test vs AudioLDM2 cache-position fix." + }, + { + "left": "pull_request:10364", + "right": "pull_request:11249", + "accept": false, + "reason": "Sana XLA support and Flux Xlab ControlNet support are different features in different pipelines." + }, + { + "left": "pull_request:1041", + "right": "pull_request:268", + "accept": false, + "reason": "Both touch tests, but they fix different historical failures and different pipeline/model issues." + }, + { + "left": "pull_request:1041", + "right": "pull_request:2759", + "accept": false, + "reason": "General test cleanup vs a large test-run plumbing change; not the same concrete bug." + }, + { + "left": "pull_request:1041", + "right": "pull_request:3108", + "accept": false, + "reason": "No evidence they address the same issue; one is a broad test fix, the other is an opaque checkpoint fix." + }, + { + "left": "pull_request:1041", + "right": "pull_request:886", + "accept": false, + "reason": "Different test fixes for different components; no common underlying bug." + }, + { + "left": "pull_request:10420", + "right": "pull_request:9077", + "accept": false, + "reason": "LTX checkpoint ID doc fix vs typo cleanup across unrelated files." + }, + { + "left": "pull_request:1043", + "right": "pull_request:21", + "accept": false, + "reason": "Better print helper vs a model test addition; not duplicate changes." + }, + { + "left": "pull_request:1043", + "right": "pull_request:22", + "accept": false, + "reason": "Better print helper vs GLIDE test coverage; unrelated." + }, + { + "left": "pull_request:1043", + "right": "pull_request:545", + "accept": false, + "reason": "Test utility print formatting vs configuration logging behavior; different concerns." + }, + { + "left": "pull_request:1043", + "right": "pull_request:828", + "accept": false, + "reason": "Test utility change vs Flax test completion; not the same bug or change." + }, + { + "left": "pull_request:10481", + "right": "pull_request:10537", + "accept": false, + "reason": "Both involve Sana, but 10481 only flips a default while 10537 adds a broader Sana-4K update and extra pipeline changes." + }, + { + "left": "pull_request:10497", + "right": "pull_request:9709", + "accept": false, + "reason": "Tokenizers install path fix vs torch version pin; unrelated CI adjustments." + }, + { + "left": "pull_request:10502", + "right": "pull_request:10579", + "accept": false, + "reason": "Serialized type name support in model outputs vs a community-project docs entry; unrelated." + }, + { + "left": "pull_request:10506", + "right": "pull_request:11196", + "accept": false, + "reason": "Sequential CPU offload regression test vs a save_pretrained serialization revert; not the same problem." + }, + { + "left": "pull_request:10506", + "right": "pull_request:8932", + "accept": false, + "reason": "TorchAO offload test vs temporal transformer test coverage; different subsystems." + }, + { + "left": "pull_request:10506", + "right": "pull_request:8973", + "accept": false, + "reason": "TorchAO offload test vs Lumina test size reduction; unrelated." + }, + { + "left": "pull_request:10544", + "right": "pull_request:11319", + "accept": false, + "reason": "ParaAttention docs vs community-project listing; no shared underlying change." + }, + { + "left": "pull_request:10548", + "right": "pull_request:10554", + "accept": false, + "reason": "HunyuanVideo batch-mask fix vs SD3 LoRA training script fix; different code paths and bugs." + }, + { + "left": "pull_request:10548", + "right": "pull_request:9077", + "accept": false, + "reason": "Transformer batch masking bug vs typo/doc cleanup; unrelated." + }, + { + "left": "pull_request:10548", + "right": "pull_request:9739", + "accept": false, + "reason": "HunyuanVideo runtime fix vs documentation typo/update; unrelated." + }, + { + "left": "pull_request:10554", + "right": "pull_request:10852", + "accept": false, + "reason": "SD3 LoRA state-dict key fix vs issue-template wording; unrelated." + }, + { + "left": "pull_request:10554", + "right": "pull_request:10889", + "accept": false, + "reason": "SD3 LoRA loading bug vs fp16 unscaling in a different training script; not duplicate." + }, + { + "left": "pull_request:10554", + "right": "pull_request:11262", + "accept": false, + "reason": "SD3 LoRA loader fix vs Wan ftfy import fix; different models and failures." + }, + { + "left": "pull_request:10576", + "right": "pull_request:9821", + "accept": false, + "reason": "Regression test for 8-bit LoRA loading vs a broader device_map behavior fix; test-only versus implementation and not the same change." + }, + { + "left": "pull_request:10576", + "right": "pull_request:9827", + "accept": false, + "reason": "This is a test for LoRA loading in quantized models, while 9827 is the actual device_map fix; they are related but not the same PR-worthy change." + }, + { + "left": "pull_request:10579", + "right": "pull_request:10898", + "accept": false, + "reason": "Community-project docs entry vs remote_decode feature/docs/implementation; unrelated." + }, + { + "left": "pull_request:10579", + "right": "pull_request:9244", + "accept": false, + "reason": "Attentive eraser pipeline addition vs Flux single-file support; different features in different areas." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 214, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10622", + "pull_request:10648", + "pull_request:10676", + "pull_request:10698", + "pull_request:1073", + "pull_request:10730", + "pull_request:1077", + "pull_request:10808", + "pull_request:10852", + "pull_request:1086", + "pull_request:10885", + "pull_request:10889", + "pull_request:1089", + "pull_request:109", + "pull_request:11039", + "pull_request:1104", + "pull_request:11044", + "pull_request:11087", + "pull_request:11194", + "pull_request:11681", + "pull_request:11838", + "pull_request:11841", + "pull_request:12005", + "pull_request:12142", + "pull_request:12143", + "pull_request:1352", + "pull_request:1489", + "pull_request:164", + "pull_request:21", + "pull_request:2130", + "pull_request:2188", + "pull_request:2337", + "pull_request:2428", + "pull_request:289", + "pull_request:3230", + "pull_request:34", + "pull_request:774", + "pull_request:839", + "pull_request:842", + "pull_request:87", + "pull_request:9060", + "pull_request:909", + "pull_request:9442", + "pull_request:9447", + "pull_request:9449", + "pull_request:9601", + "pull_request:9739" + ], + "soft_pairs": [ + "pull_request:10622|pull_request:11841", + "pull_request:10648|pull_request:11194", + "pull_request:10648|pull_request:11838", + "pull_request:10676|pull_request:9060", + "pull_request:10698|pull_request:9601", + "pull_request:1073|pull_request:34", + "pull_request:10730|pull_request:11681", + "pull_request:1077|pull_request:774", + "pull_request:1077|pull_request:909", + "pull_request:10808|pull_request:9449", + "pull_request:10852|pull_request:9739", + "pull_request:1086|pull_request:1489", + "pull_request:1086|pull_request:2130", + "pull_request:1086|pull_request:2188", + "pull_request:1086|pull_request:2337", + "pull_request:1086|pull_request:2428", + "pull_request:1086|pull_request:289", + "pull_request:1086|pull_request:839", + "pull_request:1086|pull_request:842", + "pull_request:10885|pull_request:11194", + "pull_request:10885|pull_request:11838", + "pull_request:10889|pull_request:9739", + "pull_request:1089|pull_request:164", + "pull_request:109|pull_request:1352", + "pull_request:109|pull_request:21", + "pull_request:109|pull_request:87", + "pull_request:11039|pull_request:9442", + "pull_request:11039|pull_request:9447", + "pull_request:1104|pull_request:3230", + "pull_request:11044|pull_request:12005", + "pull_request:11087|pull_request:12142", + "pull_request:11087|pull_request:12143" + ], + "prepared_review_unit_hash": "b9b87e4608558a0f953bc335b5760a56cc53c926b41a0265cfb414070d142a5d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 98575, + "estimated_input_tokens": 24644, + "estimated_eval_tokens": 49544 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very noisy and mostly unrelated: it mixes docs-only edits, test cleanups, scheduler/docs updates, LoRA/device-map fixes, and separate pipeline feature changes. I do not see any true duplicate pairs among the soft edges.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9449 is the strongest representative of the cluster because it fixes a concrete runtime bug in LoRA loading with device-mapped models and spans the most substantive code paths among the candidates.", + "best_issue_reason": null, + "best_pr_reason": "PR 9449 is the best global PR representative: it addresses a real functional issue with clear user impact, whereas most other items are docs, tests, or unrelated feature/refactor changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10622", + "right": "pull_request:11841", + "accept": false, + "reason": "Both are docs PRs, but one updates installation instructions while the other adds batch generation docs; unrelated changes." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11194", + "accept": false, + "reason": "Scheduler docs vs a torch_dtype mapping docs update; different topics and no shared bug/change." + }, + { + "left": "pull_request:10648", + "right": "pull_request:11838", + "accept": false, + "reason": "One is scheduler documentation, the other marks pipelines deprecated; not the same change." + }, + { + "left": "pull_request:10676", + "right": "pull_request:9060", + "accept": false, + "reason": "Different pipelines and different purposes: SPMD inference example vs a transformer debug-print cleanup." + }, + { + "left": "pull_request:10698", + "right": "pull_request:9601", + "accept": false, + "reason": "Instruct Pix2Pix transform bugfix and SD3 VAE dtype handling are unrelated code-paths." + }, + { + "left": "pull_request:1073", + "right": "pull_request:34", + "accept": false, + "reason": "Dreambooth README typo fix vs a scheduler comment typo fix; purely separate documentation edits." + }, + { + "left": "pull_request:10730", + "right": "pull_request:11681", + "accept": false, + "reason": "Both touch pipeline_utils, but one fixes hook unwrapping and the other clarifies device_map docs/tests; not the same bug." + }, + { + "left": "pull_request:1077", + "right": "pull_request:774", + "accept": false, + "reason": "Scheduler test refactor vs pipeline test assertions; different test areas and changes." + }, + { + "left": "pull_request:1077", + "right": "pull_request:909", + "accept": false, + "reason": "These are unrelated test changes in different eras and different behaviors." + }, + { + "left": "pull_request:10808", + "right": "pull_request:9449", + "accept": false, + "reason": "PEFT config key restriction and LoRA loading with device-mapped models are distinct issues." + }, + { + "left": "pull_request:10852", + "right": "pull_request:9739", + "accept": false, + "reason": "Issue template label change vs Stable Diffusion 3.5 single-file docs; unrelated." + }, + { + "left": "pull_request:1086", + "right": "pull_request:1489", + "accept": false, + "reason": "Attention module refactor vs a test comment removal; no shared underlying bug or change." + }, + { + "left": "pull_request:1086", + "right": "pull_request:2130", + "accept": false, + "reason": "Attention refactor and allow_patterns cleanup are unrelated implementation changes." + }, + { + "left": "pull_request:1086", + "right": "pull_request:2188", + "accept": false, + "reason": "Model attention refactor vs inpaint docs license removal; unrelated." + }, + { + "left": "pull_request:1086", + "right": "pull_request:2337", + "accept": false, + "reason": "Attention implementation change vs test removals; not the same fix." + }, + { + "left": "pull_request:1086", + "right": "pull_request:2428", + "accept": false, + "reason": "Attention refactor and panorama docs cleanup are unrelated." + }, + { + "left": "pull_request:1086", + "right": "pull_request:289", + "accept": false, + "reason": "Attention module refactor and seed-related pipeline refactor are different code changes." + }, + { + "left": "pull_request:1086", + "right": "pull_request:839", + "accept": false, + "reason": "Model refactor vs DreamBooth README edit; no duplicate relation." + }, + { + "left": "pull_request:1086", + "right": "pull_request:842", + "accept": false, + "reason": "Attention refactor and pipeline test cleanup are unrelated." + }, + { + "left": "pull_request:10885", + "right": "pull_request:11194", + "accept": false, + "reason": "CogVideoX scheduler docs and torch_dtype mapping docs are separate documentation additions." + }, + { + "left": "pull_request:10885", + "right": "pull_request:11838", + "accept": false, + "reason": "Scheduler docs addition vs deprecated-pipelines docs; different documentation goals." + }, + { + "left": "pull_request:10889", + "right": "pull_request:9739", + "accept": false, + "reason": "Dreambooth fp16 validation fix and SD3.5 single-file docs are not the same change." + }, + { + "left": "pull_request:1089", + "right": "pull_request:164", + "accept": false, + "reason": "Stable Diffusion docs about scheduler usage vs an old scheduler schedule implementation change; different levels and concerns." + }, + { + "left": "pull_request:109", + "right": "pull_request:1352", + "accept": false, + "reason": "Major model renaming and AltDiffusion sample-model docs are unrelated." + }, + { + "left": "pull_request:109", + "right": "pull_request:21", + "accept": false, + "reason": "Big model renaming vs adding model tests; not a duplicate bug/change." + }, + { + "left": "pull_request:109", + "right": "pull_request:87", + "accept": false, + "reason": "Both are large refactors, but they address different model architecture migrations and cannot be merged as one concrete fix." + }, + { + "left": "pull_request:11039", + "right": "pull_request:9442", + "accept": false, + "reason": "Single-file dtype handling and CI behavior for deprecated attention tests are unrelated." + }, + { + "left": "pull_request:11039", + "right": "pull_request:9447", + "accept": false, + "reason": "Single-file loader dtype behavior vs pipeline save_pretrained sharding defaults; different code paths." + }, + { + "left": "pull_request:1104", + "right": "pull_request:3230", + "accept": false, + "reason": "CI test additions for MPS and the IF model addition are entirely different changes." + }, + { + "left": "pull_request:11044", + "right": "pull_request:12005", + "accept": false, + "reason": "LoRA no-op warning text and sharded-checkpoint local_files_only handling are unrelated." + }, + { + "left": "pull_request:11087", + "right": "pull_request:12142", + "accept": false, + "reason": "Wan I2V quality fix vs Wan VACE bf16/fp16 mask casting; different pipeline variants and fixes." + }, + { + "left": "pull_request:11087", + "right": "pull_request:12143", + "accept": false, + "reason": "Wan I2V quality work and Wan VACE dtype handling are separate changes, not one bug fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 215, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11095", + "pull_request:11096", + "pull_request:11113", + "pull_request:11115", + "pull_request:11129", + "pull_request:11237", + "pull_request:11244", + "pull_request:11250", + "pull_request:11257", + "pull_request:11260", + "pull_request:11261", + "pull_request:11262", + "pull_request:11269", + "pull_request:113", + "pull_request:1130", + "pull_request:1133", + "pull_request:11350", + "pull_request:11353", + "pull_request:11366", + "pull_request:1147", + "pull_request:11522", + "pull_request:11529", + "pull_request:11531", + "pull_request:11537", + "pull_request:11546", + "pull_request:11556", + "pull_request:11564", + "pull_request:11635", + "pull_request:11736", + "pull_request:11747", + "pull_request:11841", + "pull_request:11897", + "pull_request:11983", + "pull_request:12244", + "pull_request:12391", + "pull_request:12586", + "pull_request:1278", + "pull_request:1751", + "pull_request:2009", + "pull_request:42", + "pull_request:580", + "pull_request:703", + "pull_request:840", + "pull_request:885", + "pull_request:9506", + "pull_request:9739", + "pull_request:9760", + "pull_request:9938" + ], + "soft_pairs": [ + "pull_request:11095|pull_request:11350", + "pull_request:11096|pull_request:11129", + "pull_request:11096|pull_request:11531", + "pull_request:11113|pull_request:9506", + "pull_request:11115|pull_request:11250", + "pull_request:11129|pull_request:11531", + "pull_request:11129|pull_request:11983", + "pull_request:11129|pull_request:9760", + "pull_request:11237|pull_request:12391", + "pull_request:11244|pull_request:11522", + "pull_request:11250|pull_request:12244", + "pull_request:11257|pull_request:11635", + "pull_request:11260|pull_request:11897", + "pull_request:11261|pull_request:11269", + "pull_request:11262|pull_request:9739", + "pull_request:113|pull_request:1751", + "pull_request:1130|pull_request:840", + "pull_request:1133|pull_request:1278", + "pull_request:1133|pull_request:42", + "pull_request:1133|pull_request:580", + "pull_request:1133|pull_request:703", + "pull_request:1133|pull_request:885", + "pull_request:11353|pull_request:11546", + "pull_request:11366|pull_request:9938", + "pull_request:1147|pull_request:2009", + "pull_request:11529|pull_request:11531", + "pull_request:11537|pull_request:11736", + "pull_request:11556|pull_request:11841", + "pull_request:11556|pull_request:12586", + "pull_request:11564|pull_request:11747" + ], + "prepared_review_unit_hash": "c3c2569f39c679b11a7c561d1d0777c822f97fc74191e6afd0dd3e0ffeb18a8d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 95325, + "estimated_input_tokens": 23832, + "estimated_eval_tokens": 47920 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of unrelated pull requests, mostly small docs, CI, and model/pipeline fixes. I do not see a single underlying bug or change theme that would justify deduping any pair.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:11095", + "right": "pull_request:11350", + "accept": false, + "reason": "Same file, but different problems: one fixes a pipeline example variable in docs, the other changes ControlNet union forward logic." + }, + { + "left": "pull_request:11096", + "right": "pull_request:11129", + "accept": false, + "reason": "README/docs wording for Flux TPU vs internal modeling import cleanup; unrelated changes." + }, + { + "left": "pull_request:11096", + "right": "pull_request:11531", + "accept": false, + "reason": "Both are docs, but one is Flux TPU README content and the other is an inpaint usage example; different topics." + }, + { + "left": "pull_request:11113", + "right": "pull_request:9506", + "accept": false, + "reason": "AutoencoderKL training bugfix vs unrelated docs/path correction; no shared code-path problem." + }, + { + "left": "pull_request:11115", + "right": "pull_request:11250", + "accept": false, + "reason": "Code export/import work for AutoModel vs adding AutoModel docs; the latter is documentation only." + }, + { + "left": "pull_request:11129", + "right": "pull_request:11531", + "accept": false, + "reason": "Model import refactor vs inpaint documentation tweak; not the same underlying change." + }, + { + "left": "pull_request:11129", + "right": "pull_request:11983", + "accept": false, + "reason": "Modeling import cleanup and installation docs edit are unrelated." + }, + { + "left": "pull_request:11129", + "right": "pull_request:9760", + "accept": false, + "reason": "Model import changes vs README wording cleanup; no shared bug or fix." + }, + { + "left": "pull_request:11237", + "right": "pull_request:12391", + "accept": false, + "reason": "Minor dtype-map docs update vs broad quantization docs restructuring; unrelated documentation changes." + }, + { + "left": "pull_request:11244", + "right": "pull_request:11522", + "accept": false, + "reason": "Both touch AudioLDM2, but they fix different concrete issues: model typing/offload handling versus transformer-version cache-position compatibility." + }, + { + "left": "pull_request:11250", + "right": "pull_request:12244", + "accept": false, + "reason": "AutoModel docs addition vs removing Flax from many docs; different documentation scopes and goals." + }, + { + "left": "pull_request:11257", + "right": "pull_request:11635", + "accept": false, + "reason": "Workflow token test change vs Flux Kontext feature implementation; completely different concerns." + }, + { + "left": "pull_request:11260", + "right": "pull_request:11897", + "accept": false, + "reason": "Temporary modeling_utils dummy edit vs large modular-diffusers CI/docs merge test; unrelated." + }, + { + "left": "pull_request:11261", + "right": "pull_request:11269", + "accept": false, + "reason": "VidTok autoencoder support vs ONNX Runtime package detection updates; different subsystems and bugs." + }, + { + "left": "pull_request:11262", + "right": "pull_request:9739", + "accept": false, + "reason": "Wan ftfy import guard vs Stable Diffusion 3 single-file doc/example fix; unrelated." + }, + { + "left": "pull_request:113", + "right": "pull_request:1751", + "accept": false, + "reason": "SDE-VE scheduler init bug vs UnCLIP generator plumbing; different schedulers and code paths." + }, + { + "left": "pull_request:1130", + "right": "pull_request:840", + "accept": false, + "reason": "lpw_stable_diffusion offload/weighting fixes vs a dummy one-step U-Net example; unrelated." + }, + { + "left": "pull_request:1133", + "right": "pull_request:1278", + "accept": false, + "reason": "Windows test workflow addition vs CI cache test workflow changes; different CI goals." + }, + { + "left": "pull_request:1133", + "right": "pull_request:42", + "accept": false, + "reason": "Windows test workflow vs early modeling-utils test cleanup; no common bug." + }, + { + "left": "pull_request:1133", + "right": "pull_request:580", + "accept": false, + "reason": "Windows CI workflow vs Flax scheduler implementation/test work; unrelated." + }, + { + "left": "pull_request:1133", + "right": "pull_request:703", + "accept": false, + "reason": "Windows test workflow vs CUDA cache CI changes; different test infra problems." + }, + { + "left": "pull_request:1133", + "right": "pull_request:885", + "accept": false, + "reason": "Windows CI workflow vs new DPM encoder scheduler; unrelated." + }, + { + "left": "pull_request:11353", + "right": "pull_request:11546", + "accept": false, + "reason": "Wan FLF2V feature support vs Hunyuan LoRA-loading support; different model families and changes." + }, + { + "left": "pull_request:11366", + "right": "pull_request:9938", + "accept": false, + "reason": "HiDream transformer output-shape fix vs SDXL reference pipeline docs/example update; unrelated." + }, + { + "left": "pull_request:1147", + "right": "pull_request:2009", + "accept": false, + "reason": "Flax example loading-from-subfolder fix vs broad textual inversion loader integration; different feature scopes." + }, + { + "left": "pull_request:11529", + "right": "pull_request:11531", + "accept": false, + "reason": "No meaningful diff is shown for 11529, and 11531 is a specific inpaint docs edit; insufficient evidence of sameness." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11736", + "accept": false, + "reason": "Torch.compile test coverage for UNet2DConditionModel vs quantizer compileability API; different layers and problems." + }, + { + "left": "pull_request:11556", + "right": "pull_request:11841", + "accept": false, + "reason": "Regional compilation docs vs batch inference docs; both documentation, but not the same underlying change." + }, + { + "left": "pull_request:11556", + "right": "pull_request:12586", + "accept": false, + "reason": "Regional compilation docs vs doc-toctree sorting; unrelated docs maintenance." + }, + { + "left": "pull_request:11564", + "right": "pull_request:11747", + "accept": false, + "reason": "Dreambooth FLUX dependency addition vs HiDream license header addition; unrelated PRs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 216, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11566", + "pull_request:11635", + "pull_request:11636", + "pull_request:11655", + "pull_request:11660", + "pull_request:11681", + "pull_request:11684", + "pull_request:11698", + "pull_request:11703", + "pull_request:11716", + "pull_request:11728", + "pull_request:11744", + "pull_request:118", + "pull_request:11812", + "pull_request:1182", + "pull_request:1184", + "pull_request:11847", + "pull_request:11869", + "pull_request:11875", + "pull_request:11891", + "pull_request:11894", + "pull_request:11897", + "pull_request:119", + "pull_request:11920", + "pull_request:11955", + "pull_request:12062", + "pull_request:12109", + "pull_request:12115", + "pull_request:12236", + "pull_request:12244", + "pull_request:12315", + "pull_request:12337", + "pull_request:12360", + "pull_request:12398", + "pull_request:12424", + "pull_request:12438", + "pull_request:12586", + "pull_request:12637", + "pull_request:12705", + "pull_request:12848", + "pull_request:1713", + "pull_request:1868", + "pull_request:1903", + "pull_request:1939", + "pull_request:217", + "pull_request:38", + "pull_request:59", + "pull_request:75" + ], + "soft_pairs": [ + "pull_request:11566|pull_request:11812", + "pull_request:11635|pull_request:11920", + "pull_request:11635|pull_request:11955", + "pull_request:11636|pull_request:11875", + "pull_request:11655|pull_request:11684", + "pull_request:11660|pull_request:12109", + "pull_request:11681|pull_request:12315", + "pull_request:11698|pull_request:12398", + "pull_request:11703|pull_request:12244", + "pull_request:11716|pull_request:12424", + "pull_request:11716|pull_request:12438", + "pull_request:11728|pull_request:12586", + "pull_request:11744|pull_request:12848", + "pull_request:118|pull_request:1903", + "pull_request:118|pull_request:59", + "pull_request:11812|pull_request:11875", + "pull_request:11812|pull_request:12337", + "pull_request:1182|pull_request:1713", + "pull_request:1184|pull_request:38", + "pull_request:1184|pull_request:75", + "pull_request:11847|pull_request:12637", + "pull_request:11847|pull_request:12705", + "pull_request:11869|pull_request:12236", + "pull_request:11875|pull_request:12398", + "pull_request:11891|pull_request:12360", + "pull_request:11894|pull_request:12062", + "pull_request:11897|pull_request:12115", + "pull_request:119|pull_request:1868", + "pull_request:119|pull_request:1939", + "pull_request:119|pull_request:217" + ], + "prepared_review_unit_hash": "c29606e53dabb03a84ee073c8338247d50a03008b5a7eb45ac90fb99dbb96368", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 98982, + "estimated_input_tokens": 24746, + "estimated_eval_tokens": 49748 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a noisy mixed cluster of unrelated Diffusers PRs. The only coherent theme is model/pipeline work around Chroma/Flux/Kontext, but the proposed soft pairs are generally adjacent feature, docs, or test changes rather than true duplicates.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11698 is the broadest and most central substantive feature change in the cluster: it adds the Chroma pipeline/model stack, has the strongest integration footprint, and explicitly targets issue 11010.", + "best_issue_reason": null, + "best_pr_reason": "PR 11698 is the best representative artifact because it is the largest real product change here and anchors the Chroma/Flux family work better than the smaller docs/test/follow-up PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11566", + "right": "pull_request:11812", + "accept": false, + "reason": "Both are FLUX-family changes, but one adds Chroma variant loading and the other adds the Flux Kontext pipeline; different artifacts and code paths." + }, + { + "left": "pull_request:11635", + "right": "pull_request:11920", + "accept": false, + "reason": "Both are test-only PRs, but they cover different pipelines (Flux Kontext vs Wan) and unrelated behaviors." + }, + { + "left": "pull_request:11635", + "right": "pull_request:11955", + "accept": false, + "reason": "Different test additions for different pipelines (Flux Kontext vs Cosmos); same broad pattern, not the same change." + }, + { + "left": "pull_request:11636", + "right": "pull_request:11875", + "accept": false, + "reason": "One is general docs for custom community pipelines; the other wires Kontext into auto-pipeline. Related topic, but not a duplicate change." + }, + { + "left": "pull_request:11655", + "right": "pull_request:11684", + "accept": false, + "reason": "Both touch loader/quantization areas, but one adds bnb 8-bit LoRA support and the other adds GGUF single-file loading; distinct fixes." + }, + { + "left": "pull_request:11660", + "right": "pull_request:12109", + "accept": false, + "reason": "HiDream LoRA conversion work and UniPC scheduler custom sigma support are unrelated code changes." + }, + { + "left": "pull_request:11681", + "right": "pull_request:12315", + "accept": false, + "reason": "Model/pipeline device_map docs versus LoRA test cleanup; no shared bug or implementation path." + }, + { + "left": "pull_request:11698", + "right": "pull_request:12398", + "accept": false, + "reason": "Chroma pipeline integration and Mirage pipeline implementation are different feature areas despite overlapping __init__ export churn." + }, + { + "left": "pull_request:11703", + "right": "pull_request:12244", + "accept": false, + "reason": "Docs PR about quantization/compile guidance vs docs PR removing Flax; unrelated documentation changes." + }, + { + "left": "pull_request:11716", + "right": "pull_request:12424", + "accept": false, + "reason": "Flux LoRA metadata fix and Dockerfile cleanup are unrelated." + }, + { + "left": "pull_request:11716", + "right": "pull_request:12438", + "accept": false, + "reason": "LoRA metadata handling and offload_state_dict loading logic address different bugs." + }, + { + "left": "pull_request:11728", + "right": "pull_request:12586", + "accept": false, + "reason": "Doc-builder Docker image dependencies and doc tree sorting are unrelated." + }, + { + "left": "pull_request:11744", + "right": "pull_request:12848", + "accept": false, + "reason": "DreamBooth LoRA metadata handling and adding ChromaInpaintPipeline are separate features." + }, + { + "left": "pull_request:118", + "right": "pull_request:1903", + "accept": false, + "reason": "Old pipeline rename/fix and SD2 depth2img comment reordering are not the same underlying change." + }, + { + "left": "pull_request:118", + "right": "pull_request:59", + "accept": false, + "reason": "Different old-core refactors in unrelated pipelines/models." + }, + { + "left": "pull_request:11812", + "right": "pull_request:11875", + "accept": false, + "reason": "Adding the Flux Kontext pipeline and registering it in auto-pipeline are related follow-ups, but not the same concrete change." + }, + { + "left": "pull_request:11812", + "right": "pull_request:12337", + "accept": false, + "reason": "Flux Kontext pipeline creation and xformers attention fix for Flux are different code-path fixes." + }, + { + "left": "pull_request:1182", + "right": "pull_request:1713", + "accept": false, + "reason": "Both are documentation-oriented, but one is LPW docs and the other is conversion script cleanup; not duplicates." + }, + { + "left": "pull_request:1184", + "right": "pull_request:38", + "accept": false, + "reason": "Both involve attention code historically, but one is flash-attention integration and the other is early attention module unification; different changes separated by scope and time." + }, + { + "left": "pull_request:1184", + "right": "pull_request:75", + "accept": false, + "reason": "Flash-attention work and Glide attention fix are distinct bug fixes." + }, + { + "left": "pull_request:11847", + "right": "pull_request:12637", + "accept": false, + "reason": "Wan single-file example docs and a copy-comment cleanup in Sana video are unrelated." + }, + { + "left": "pull_request:11847", + "right": "pull_request:12705", + "accept": false, + "reason": "Wan single-file docs and typo fixes across many files are not the same change." + }, + { + "left": "pull_request:11869", + "right": "pull_request:12236", + "accept": false, + "reason": "GGUF CUDA kernel support and hub flash-attn kernel support are both performance/backend work, but they target different mechanisms and code paths." + }, + { + "left": "pull_request:11875", + "right": "pull_request:12398", + "accept": false, + "reason": "Kontext auto-pipeline registration and Mirage pipeline implementation are unrelated." + }, + { + "left": "pull_request:11891", + "right": "pull_request:12360", + "accept": false, + "reason": "Both are Chroma docs, but one fixes a usage typo and the other rewrites guidance text for guidance_scale; not a duplicate docs fix." + }, + { + "left": "pull_request:11894", + "right": "pull_request:12062", + "accept": false, + "reason": "k-diffusion pinning for CI and enabling all GPUs in CI are unrelated infrastructure changes." + }, + { + "left": "pull_request:11897", + "right": "pull_request:12115", + "accept": false, + "reason": "A large test merge and an AnimateDiff sparse control test addition are different scopes and different failures." + }, + { + "left": "pull_request:119", + "right": "pull_request:1868", + "accept": false, + "reason": "Very old unrelated var-name fix vs EMA decay fix; not the same bug." + }, + { + "left": "pull_request:119", + "right": "pull_request:1939", + "accept": false, + "reason": "Very old unrelated fixes in hub utils vs README image path." + }, + { + "left": "pull_request:119", + "right": "pull_request:217", + "accept": false, + "reason": "Hub util variable-name fix and NSFW checker logic fix are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 217, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:119", + "pull_request:11953", + "pull_request:11968", + "pull_request:1198", + "pull_request:11986", + "pull_request:120", + "pull_request:12018", + "pull_request:12030", + "pull_request:1208", + "pull_request:12099", + "pull_request:121", + "pull_request:1210", + "pull_request:1211", + "pull_request:1215", + "pull_request:1218", + "pull_request:1219", + "pull_request:12226", + "pull_request:12467", + "pull_request:12560", + "pull_request:12628", + "pull_request:12644", + "pull_request:12686", + "pull_request:12846", + "pull_request:12848", + "pull_request:12932", + "pull_request:12959", + "pull_request:12984", + "pull_request:1326", + "pull_request:1415", + "pull_request:1610", + "pull_request:1759", + "pull_request:19", + "pull_request:1971", + "pull_request:2007", + "pull_request:2345", + "pull_request:2701", + "pull_request:3023", + "pull_request:312", + "pull_request:3343", + "pull_request:408", + "pull_request:41", + "pull_request:423", + "pull_request:433", + "pull_request:498", + "pull_request:693", + "pull_request:70", + "pull_request:75", + "pull_request:921" + ], + "soft_pairs": [ + "pull_request:119|pull_request:312", + "pull_request:119|pull_request:41", + "pull_request:119|pull_request:498", + "pull_request:119|pull_request:693", + "pull_request:119|pull_request:75", + "pull_request:11953|pull_request:12686", + "pull_request:11968|pull_request:12644", + "pull_request:1198|pull_request:1415", + "pull_request:11986|pull_request:12560", + "pull_request:11986|pull_request:12628", + "pull_request:11986|pull_request:12932", + "pull_request:11986|pull_request:12959", + "pull_request:120|pull_request:19", + "pull_request:120|pull_request:2007", + "pull_request:120|pull_request:70", + "pull_request:12018|pull_request:12467", + "pull_request:12030|pull_request:12628", + "pull_request:12030|pull_request:12846", + "pull_request:12030|pull_request:12848", + "pull_request:12030|pull_request:12984", + "pull_request:1208|pull_request:1759", + "pull_request:1208|pull_request:2701", + "pull_request:12099|pull_request:12226", + "pull_request:121|pull_request:1971", + "pull_request:121|pull_request:433", + "pull_request:1210|pull_request:3343", + "pull_request:1211|pull_request:1326", + "pull_request:1215|pull_request:2345", + "pull_request:1215|pull_request:3023", + "pull_request:1218|pull_request:1610", + "pull_request:1218|pull_request:423", + "pull_request:1219|pull_request:408", + "pull_request:1219|pull_request:921" + ], + "prepared_review_unit_hash": "20bd9f73142c4a89d5349012c60cf583546c716632fc30a70e72b8006cca41be", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 104498, + "estimated_input_tokens": 26125, + "estimated_eval_tokens": 52506 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of unrelated PRs (docs typos, small bug fixes, tests, modular pipeline work). None of the soft pairs appear to address the same concrete bug or change, so all should be rejected and there is no coherent canonical issue/PR.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:119", + "right": "pull_request:312", + "accept": false, + "reason": "Different changes: a variable-name bug fix in hub_utils.py vs a README link/docs edit." + }, + { + "left": "pull_request:119", + "right": "pull_request:41", + "accept": false, + "reason": "Unrelated fixes: model-card arg typo vs resnet comment cleanup." + }, + { + "left": "pull_request:119", + "right": "pull_request:498", + "accept": false, + "reason": "Different code paths and problems; one is a model-card typo, the other is an attention-slicing API fix." + }, + { + "left": "pull_request:119", + "right": "pull_request:693", + "accept": false, + "reason": "README BibTeX citation formatting is unrelated to the hub_utils variable-name fix." + }, + { + "left": "pull_request:119", + "right": "pull_request:75", + "accept": false, + "reason": "Different model code; attention implementation changes are not the same as the hub_utils typo fix." + }, + { + "left": "pull_request:11953", + "right": "pull_request:12686", + "accept": false, + "reason": "Test-slice update for HiDream is unrelated to Kandinsky5 pipeline implementation." + }, + { + "left": "pull_request:11968", + "right": "pull_request:12644", + "accept": false, + "reason": "Hook refactor/shared constants and AutoModel docs are different artifacts with no shared bug/change." + }, + { + "left": "pull_request:1198", + "right": "pull_request:1415", + "accept": false, + "reason": "Both touch diffusion pipelines, but one fixes CycleDiffusion invocation while the other is a broad stable-diffusion cleanup; not the same concrete issue." + }, + { + "left": "pull_request:11986", + "right": "pull_request:12560", + "accept": false, + "reason": "Both are modular-pipeline work, but they target different features (LTX vs general modular updates) and are not mergeable as one fix." + }, + { + "left": "pull_request:11986", + "right": "pull_request:12628", + "accept": false, + "reason": "LTX modular pipeline implementation is unrelated to the modular block-name consistency check." + }, + { + "left": "pull_request:11986", + "right": "pull_request:12932", + "accept": false, + "reason": "Different Qwen/LTX modular-pipeline changes; one adds LTX pipeline code, the other only improves docstrings/types." + }, + { + "left": "pull_request:11986", + "right": "pull_request:12959", + "accept": false, + "reason": "LTX modular pipeline work is unrelated to WAN I2V parameter additions." + }, + { + "left": "pull_request:120", + "right": "pull_request:19", + "accept": false, + "reason": "Both touch README, but one adds examples and the other renders math/formatting; not the same underlying change." + }, + { + "left": "pull_request:120", + "right": "pull_request:2007", + "accept": false, + "reason": "README examples and black formatting updates are unrelated." + }, + { + "left": "pull_request:120", + "right": "pull_request:70", + "accept": false, + "reason": "Main README examples vs attention-model refactor are completely different changes." + }, + { + "left": "pull_request:12018", + "right": "pull_request:12467", + "accept": false, + "reason": "A docs link fix is unrelated to a QwenImage batch-handling code fix." + }, + { + "left": "pull_request:12030", + "right": "pull_request:12628", + "accept": false, + "reason": "Flux IP-Adapter support and modular pipeline validation are different features/problems." + }, + { + "left": "pull_request:12030", + "right": "pull_request:12846", + "accept": false, + "reason": "Flux IP-Adapter support is unrelated to adding AutoPipelineForText2Video." + }, + { + "left": "pull_request:12030", + "right": "pull_request:12848", + "accept": false, + "reason": "Flux IP-Adapter support is unrelated to adding ChromaInpaintPipeline." + }, + { + "left": "pull_request:12030", + "right": "pull_request:12984", + "accept": false, + "reason": "Flux IP-Adapter support and Flux2 Klein docs/code are different changes." + }, + { + "left": "pull_request:1208", + "right": "pull_request:1759", + "accept": false, + "reason": "One adds OneFlow/DPMsolver support; the other only guards attn2=None in xformers handling." + }, + { + "left": "pull_request:1208", + "right": "pull_request:2701", + "accept": false, + "reason": "OneFlow/DPMsolver support is unrelated to ONNX Runtime ROCm support." + }, + { + "left": "pull_request:12099", + "right": "pull_request:12226", + "accept": false, + "reason": "A device-variable typo fix in QwenImage is unrelated to a new QwenImage ControlNet test." + }, + { + "left": "pull_request:121", + "right": "pull_request:1971", + "accept": false, + "reason": "README examples and a scheduler docstring cleanup are not the same underlying change." + }, + { + "left": "pull_request:121", + "right": "pull_request:433", + "accept": false, + "reason": "README examples and broad black-style formatting changes are unrelated." + }, + { + "left": "pull_request:1210", + "right": "pull_request:3343", + "accept": false, + "reason": "Slow-test fixes for cycle diffusion/schedulers are unrelated to a DPMSolver config-doc fix." + }, + { + "left": "pull_request:1211", + "right": "pull_request:1326", + "accept": false, + "reason": "Flax usage-comment cleanup is unrelated to gpu_id handling in sequential CPU offload." + }, + { + "left": "pull_request:1215", + "right": "pull_request:2345", + "accept": false, + "reason": "DDIM eta/cpu-generator logic fix is unrelated to xformers warning messaging in training scripts." + }, + { + "left": "pull_request:1215", + "right": "pull_request:3023", + "accept": false, + "reason": "DDIM variance-noise logic and attention head/batch-dim changes are different code paths and problems." + }, + { + "left": "pull_request:1218", + "right": "pull_request:1610", + "accept": false, + "reason": "Conversion-script scheduler support is unrelated to attention logging/test changes." + }, + { + "left": "pull_request:1218", + "right": "pull_request:423", + "accept": false, + "reason": "A conversion script update is not the same as general output/docs syntax cleanup." + }, + { + "left": "pull_request:1219", + "right": "pull_request:408", + "accept": false, + "reason": "DDIM docs edits and general docs typo fixes are too broad and not the same specific change." + }, + { + "left": "pull_request:1219", + "right": "pull_request:921", + "accept": false, + "reason": "DDIM docs only vs a large v1-5 docs update; not a duplicate change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 218, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1221", + "pull_request:12236", + "pull_request:12275", + "pull_request:12305", + "pull_request:12307", + "pull_request:12337", + "pull_request:12360", + "pull_request:12398", + "pull_request:1240", + "pull_request:12434", + "pull_request:12467", + "pull_request:12487", + "pull_request:12491", + "pull_request:12527", + "pull_request:12541", + "pull_request:12552", + "pull_request:12560", + "pull_request:12578", + "pull_request:12628", + "pull_request:12637", + "pull_request:12675", + "pull_request:12686", + "pull_request:12687", + "pull_request:12705", + "pull_request:12782", + "pull_request:12846", + "pull_request:12848", + "pull_request:12868", + "pull_request:12883", + "pull_request:12919", + "pull_request:12932", + "pull_request:12984", + "pull_request:13038", + "pull_request:13055", + "pull_request:13100", + "pull_request:13146", + "pull_request:13150", + "pull_request:13158", + "pull_request:13164", + "pull_request:13174", + "pull_request:13259", + "pull_request:13275", + "pull_request:13331", + "pull_request:13344", + "pull_request:13443", + "pull_request:2631", + "pull_request:393", + "pull_request:427" + ], + "soft_pairs": [ + "pull_request:1221|pull_request:393", + "pull_request:12236|pull_request:12919", + "pull_request:12275|pull_request:12305", + "pull_request:12307|pull_request:12637", + "pull_request:12307|pull_request:12705", + "pull_request:12307|pull_request:13146", + "pull_request:12337|pull_request:12398", + "pull_request:12337|pull_request:12637", + "pull_request:12337|pull_request:13146", + "pull_request:12360|pull_request:12527", + "pull_request:12360|pull_request:12637", + "pull_request:12360|pull_request:13146", + "pull_request:12398|pull_request:13331", + "pull_request:1240|pull_request:2631", + "pull_request:1240|pull_request:427", + "pull_request:12434|pull_request:12467", + "pull_request:12434|pull_request:12487", + "pull_request:12434|pull_request:12552", + "pull_request:12491|pull_request:13164", + "pull_request:12527|pull_request:12883", + "pull_request:12541|pull_request:13055", + "pull_request:12560|pull_request:12628", + "pull_request:12560|pull_request:12932", + "pull_request:12578|pull_request:13100", + "pull_request:12628|pull_request:12846", + "pull_request:12628|pull_request:12848", + "pull_request:12628|pull_request:12932", + "pull_request:12628|pull_request:13259", + "pull_request:12637|pull_request:13038", + "pull_request:12637|pull_request:13443", + "pull_request:12675|pull_request:13158", + "pull_request:12686|pull_request:13344", + "pull_request:12687|pull_request:13275", + "pull_request:12705|pull_request:13038", + "pull_request:12705|pull_request:13443", + "pull_request:12782|pull_request:13150", + "pull_request:12846|pull_request:12848", + "pull_request:12846|pull_request:12984", + "pull_request:12846|pull_request:13259", + "pull_request:12848|pull_request:12984", + "pull_request:12848|pull_request:13259", + "pull_request:12868|pull_request:13174" + ], + "prepared_review_unit_hash": "ebd167387e5c1afdc49acfdc829c7b397cdd945ce99d2fcde236c3dc03178196", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 42, + "serialized_chars": 107761, + "estimated_input_tokens": 26941, + "estimated_eval_tokens": 54138 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs, CI, modular-pipeline cleanup, model feature additions, and separate bug fixes. I would not merge any of the soft pairs; they only share broad subsystem labels or filenames.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are only loosely related by topic and do not fix the same concrete change.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a strong global representative because the cluster spans unrelated docs, CI, feature work, and model fixes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1221", + "right": "pull_request:393", + "accept": false, + "reason": "Weight reuse docs vs textual inversion docs: unrelated documentation changes." + }, + { + "left": "pull_request:12236", + "right": "pull_request:12919", + "accept": false, + "reason": "Both touch attention code, but they fix different backends and different code paths." + }, + { + "left": "pull_request:12275", + "right": "pull_request:12305", + "accept": false, + "reason": "TorchAO config support and single-file allocator warmup are unrelated features." + }, + { + "left": "pull_request:12307", + "right": "pull_request:12637", + "accept": false, + "reason": "Modular pipeline tweaks vs Sana video copy cleanup; different problems and files." + }, + { + "left": "pull_request:12307", + "right": "pull_request:12705", + "accept": false, + "reason": "One is modular-pipeline logic, the other is typo cleanup across docs/examples." + }, + { + "left": "pull_request:12307", + "right": "pull_request:13146", + "accept": false, + "reason": "Modular pipeline fixes vs a tokenizer-type cleanup in PAG hunyuan code." + }, + { + "left": "pull_request:12337", + "right": "pull_request:12398", + "accept": false, + "reason": "xFormers attention enablement and Mirage pipeline work are unrelated." + }, + { + "left": "pull_request:12337", + "right": "pull_request:12637", + "accept": false, + "reason": "Different pipelines and different fixes; only the broad theme is code cleanup." + }, + { + "left": "pull_request:12337", + "right": "pull_request:13146", + "accept": false, + "reason": "Flux attention fix and Hunyuan tokenizer cleanup are distinct changes." + }, + { + "left": "pull_request:12360", + "right": "pull_request:12527", + "accept": false, + "reason": "Chroma docs edits and Kandinsky5 no-CFG fix are different artifacts." + }, + { + "left": "pull_request:12360", + "right": "pull_request:12637", + "accept": false, + "reason": "Docs wording changes do not match the Sana copy-fix PR." + }, + { + "left": "pull_request:12360", + "right": "pull_request:13146", + "accept": false, + "reason": "Chroma docs and PAG type cleanup are unrelated." + }, + { + "left": "pull_request:12398", + "right": "pull_request:13331", + "accept": false, + "reason": "Different model families and different pipeline additions; not the same feature." + }, + { + "left": "pull_request:1240", + "right": "pull_request:2631", + "accept": false, + "reason": "OneFlow image-to-image port and training main-process checks are unrelated." + }, + { + "left": "pull_request:1240", + "right": "pull_request:427", + "accept": false, + "reason": "Both involve tests/pipelines, but the fixes target different behaviors." + }, + { + "left": "pull_request:12434", + "right": "pull_request:12467", + "accept": false, + "reason": "PEFT version pinning and QwenImage batch handling are unrelated changes." + }, + { + "left": "pull_request:12434", + "right": "pull_request:12487", + "accept": false, + "reason": "Training requirement pin vs broken-link docs cleanup; no shared bug." + }, + { + "left": "pull_request:12434", + "right": "pull_request:12552", + "accept": false, + "reason": "PEFT requirement bump and Kandinsky5 docs typo are unrelated." + }, + { + "left": "pull_request:12491", + "right": "pull_request:13164", + "accept": false, + "reason": "Both touch CI, but one adds link checking while the other trims test deps." + }, + { + "left": "pull_request:12527", + "right": "pull_request:12883", + "accept": false, + "reason": "Kandinsky5 no-CFG fix and torchao typo correction address different issues." + }, + { + "left": "pull_request:12541", + "right": "pull_request:13055", + "accept": false, + "reason": "LoRA conversion compatibility and Wan transformer fallback fix different code paths." + }, + { + "left": "pull_request:12560", + "right": "pull_request:12628", + "accept": false, + "reason": "General modular updates vs an added consistency check; related area, different change." + }, + { + "left": "pull_request:12560", + "right": "pull_request:12932", + "accept": false, + "reason": "Code-path modular updates and docstring cleanup are not the same fix." + }, + { + "left": "pull_request:12578", + "right": "pull_request:13100", + "accept": false, + "reason": "Auto CPU offload error handling and modular docs update are unrelated." + }, + { + "left": "pull_request:12628", + "right": "pull_request:12846", + "accept": false, + "reason": "Modular pipeline validation and AutoPipelineForText2Video are different features." + }, + { + "left": "pull_request:12628", + "right": "pull_request:12848", + "accept": false, + "reason": "A modular check and Chroma inpaint pipeline addition do not share the same bug." + }, + { + "left": "pull_request:12628", + "right": "pull_request:12932", + "accept": false, + "reason": "Structural modular validation vs docstring improvements only overlap broadly." + }, + { + "left": "pull_request:12628", + "right": "pull_request:13259", + "accept": false, + "reason": "Pipeline validation code and repository AGENTS docs are unrelated." + }, + { + "left": "pull_request:12637", + "right": "pull_request:13038", + "accept": false, + "reason": "Sana copy cleanup and Wan patchification fix different pipeline behaviors." + }, + { + "left": "pull_request:12637", + "right": "pull_request:13443", + "accept": false, + "reason": "Different model families and unrelated attention/backend tweaks." + }, + { + "left": "pull_request:12675", + "right": "pull_request:13158", + "accept": false, + "reason": "Kandinsky5 docs typo and QwenImage layered docs addition are different doc changes." + }, + { + "left": "pull_request:12686", + "right": "pull_request:13344", + "accept": false, + "reason": "Kandinsky5 pipeline expansion and GLM Image transformer tests are unrelated." + }, + { + "left": "pull_request:12687", + "right": "13275", + "accept": false, + "reason": "Autoencoder typo fix and Hunyuan test kwargs cleanup are unrelated." + }, + { + "left": "pull_request:12705", + "right": "pull_request:13150", + "accept": false, + "reason": "Typo cleanup across docs/examples and a test helper API change are different issues." + }, + { + "left": "pull_request:12846", + "right": "pull_request:12848", + "accept": false, + "reason": "Both add pipelines, but for different models and different code paths." + }, + { + "left": "pull_request:12846", + "right": "pull_request:12984", + "accept": false, + "reason": "AutoPipeline support and Flux2 Klein docs are unrelated." + }, + { + "left": "pull_request:12846", + "right": "pull_request:13259", + "accept": false, + "reason": "Auto pipeline support and AGENTS.md are different changes entirely." + }, + { + "left": "pull_request:12848", + "right": "pull_request:12984", + "accept": false, + "reason": "Chroma inpaint pipeline and Flux2 Klein docs do not match." + }, + { + "left": "pull_request:12848", + "right": "pull_request:13259", + "accept": false, + "reason": "New pipeline implementation vs contributor guidance docs; unrelated." + }, + { + "left": "pull_request:12868", + "right": "pull_request:13174", + "accept": false, + "reason": "QwenImage attention splitting and attention-backend tests are not the same concrete fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 219, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1291", + "pull_request:1292", + "pull_request:1299", + "pull_request:12998", + "pull_request:13038", + "pull_request:13039", + "pull_request:13081", + "pull_request:13100", + "pull_request:1311", + "pull_request:13146", + "pull_request:13150", + "pull_request:13204", + "pull_request:1322", + "pull_request:1325", + "pull_request:13259", + "pull_request:13269", + "pull_request:1327", + "pull_request:1330", + "pull_request:13340", + "pull_request:1338", + "pull_request:1341", + "pull_request:13436", + "pull_request:13443", + "pull_request:1352", + "pull_request:1360", + "pull_request:1382", + "pull_request:1406", + "pull_request:15", + "pull_request:1624", + "pull_request:17", + "pull_request:1907", + "pull_request:1939", + "pull_request:21", + "pull_request:2516", + "pull_request:2529", + "pull_request:2530", + "pull_request:2540", + "pull_request:2667", + "pull_request:27", + "pull_request:2958", + "pull_request:3397", + "pull_request:432", + "pull_request:483", + "pull_request:51", + "pull_request:62", + "pull_request:7", + "pull_request:842" + ], + "soft_pairs": [ + "pull_request:1291|pull_request:2529", + "pull_request:1292|pull_request:1907", + "pull_request:1299|pull_request:483", + "pull_request:12998|pull_request:13081", + "pull_request:13038|pull_request:13146", + "pull_request:13039|pull_request:13150", + "pull_request:13100|pull_request:13204", + "pull_request:1311|pull_request:2530", + "pull_request:13146|pull_request:13443", + "pull_request:1322|pull_request:1330", + "pull_request:1322|pull_request:1624", + "pull_request:1322|pull_request:1939", + "pull_request:1322|pull_request:2540", + "pull_request:1322|pull_request:432", + "pull_request:1325|pull_request:1382", + "pull_request:1325|pull_request:15", + "pull_request:1325|pull_request:17", + "pull_request:1325|pull_request:27", + "pull_request:1325|pull_request:62", + "pull_request:13259|pull_request:13269", + "pull_request:1327|pull_request:3397", + "pull_request:1330|pull_request:1624", + "pull_request:1330|pull_request:2667", + "pull_request:1330|pull_request:2958", + "pull_request:1330|pull_request:432", + "pull_request:13340|pull_request:13436", + "pull_request:1338|pull_request:51", + "pull_request:1341|pull_request:7", + "pull_request:1352|pull_request:1406", + "pull_request:1352|pull_request:21", + "pull_request:1352|pull_request:842", + "pull_request:1360|pull_request:2516" + ], + "prepared_review_unit_hash": "4074d0a2107aba40f59084f8d17b2ec547f6d9b42e592e65fd8049db2aede7a7", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 99949, + "estimated_input_tokens": 24988, + "estimated_eval_tokens": 50232 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most PRs are unrelated docs, tests, or different feature/fix areas. None of the soft-similarity pairs looks like the same underlying bug or change, so all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a valid canonical for this cluster; the items cover unrelated changes across docs, pipelines, models, tests, and tooling.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong global representative because the soft-similarity links are mostly superficial and do not share the same concrete code-path problem.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1291", + "right": "pull_request:2529", + "accept": false, + "reason": "Different fixes: README notebook link vs SD checkpoint conversion logic; no shared bug." + }, + { + "left": "pull_request:1292", + "right": "pull_request:1907", + "accept": false, + "reason": "One is OneFlow integration code, the other is a docker/build repo change; unrelated changes." + }, + { + "left": "pull_request:1299", + "right": "pull_request:483", + "accept": false, + "reason": "AltDiffusion feature addition vs typo/doc cleanup; not the same underlying change." + }, + { + "left": "pull_request:12998", + "right": "pull_request:13081", + "accept": false, + "reason": "Different style/runtime issues in different pipelines; not one code-path problem." + }, + { + "left": "pull_request:13038", + "right": "pull_request:13146", + "accept": false, + "reason": "Wan patchification fix vs tokenizer type fix; separate failures in different components." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13150", + "accept": false, + "reason": "CPU upcasting bug in a model forward pass vs a test helper tweak; unrelated." + }, + { + "left": "pull_request:13100", + "right": "pull_request:13204", + "accept": false, + "reason": "Modular pipeline docs/state handling vs trust_remote_code behavior; different concerns." + }, + { + "left": "pull_request:1311", + "right": "pull_request:2530", + "accept": false, + "reason": "AltDiffusion tests vs EMAModel tests; same area only at a very broad level." + }, + { + "left": "pull_request:13146", + "right": "pull_request:13443", + "accept": false, + "reason": "Tokenizer type fix vs attention backend version check; no common bug path." + }, + { + "left": "pull_request:1322", + "right": "pull_request:1330", + "accept": false, + "reason": "Both mention scheduler/path usage, but one fixes examples' pretrained path and the other inpaint scaling; not the same bug." + }, + { + "left": "pull_request:1322", + "right": "pull_request:1624", + "accept": false, + "reason": "Examples scheduler path fix vs DPM timestep computation; different code paths." + }, + { + "left": "pull_request:1322", + "right": "pull_request:1939", + "accept": false, + "reason": "Scheduler path fix vs README logo path; only superficial path wording overlap." + }, + { + "left": "pull_request:1322", + "right": "pull_request:2540", + "accept": false, + "reason": "Scheduler path fix vs docs section correction; unrelated." + }, + { + "left": "pull_request:1322", + "right": "pull_request:432", + "accept": false, + "reason": "Different documentation/example link fixes; not the same concrete issue." + }, + { + "left": "pull_request:1325", + "right": "pull_request:1382", + "accept": false, + "reason": "OneFlow image2image fix vs dependency/import gating; different bugs and scopes." + }, + { + "left": "pull_request:1325", + "right": "pull_request:15", + "accept": false, + "reason": "No overlap beyond being repo housekeeping; unrelated changes." + }, + { + "left": "pull_request:1325", + "right": "pull_request:17", + "accept": false, + "reason": "OneFlow code vs typo fix; not mergeable as one PR." + }, + { + "left": "pull_request:1325", + "right": "pull_request:27", + "accept": false, + "reason": "OneFlow integration vs README checklist formatting; unrelated." + }, + { + "left": "pull_request:1325", + "right": "pull_request:62", + "accept": false, + "reason": "OneFlow image2image fix vs LDM unconditional pipeline fix; different pipelines and bugs." + }, + { + "left": "pull_request:13259", + "right": "pull_request:13269", + "accept": false, + "reason": "AGENTS.md introduction vs later skill/doc expansion; related docs, but not the same change." + }, + { + "left": "pull_request:1327", + "right": "pull_request:3397", + "accept": false, + "reason": "Local custom pipeline loading behavior vs LoRA local directory docs; different functionality." + }, + { + "left": "pull_request:1330", + "right": "pull_request:1624", + "accept": false, + "reason": "Inpaint scaling fix vs DPM scheduler timesteps; unrelated." + }, + { + "left": "pull_request:1330", + "right": "pull_request:2667", + "accept": false, + "reason": "Scheduler/path issue in training examples vs implementing inpaint strength behavior; not the same bug." + }, + { + "left": "pull_request:1330", + "right": "pull_request:2958", + "accept": false, + "reason": "Example scheduler path fix vs controlnet inpaint addition; different features." + }, + { + "left": "pull_request:1330", + "right": "pull_request:432", + "accept": false, + "reason": "Inpaint scaling change vs docs link corrections; unrelated." + }, + { + "left": "pull_request:13340", + "right": "pull_request:13436", + "accept": false, + "reason": "Model test refactor vs a trivial README test edit; not the same underlying issue." + }, + { + "left": "pull_request:1338", + "right": "pull_request:51", + "accept": false, + "reason": "ONNX inpaint cast-order fix vs setup dependency pin; completely unrelated." + }, + { + "left": "pull_request:1341", + "right": "pull_request:7", + "accept": false, + "reason": "Stable Diffusion repaint pipeline vs VQ-VAE model addition; different features entirely." + }, + { + "left": "pull_request:1352", + "right": "pull_request:1406", + "accept": false, + "reason": "AltDiffusion sample checkpoint/docs update vs deprecating sample size; different docs/product changes." + }, + { + "left": "pull_request:1352", + "right": "pull_request:21", + "accept": false, + "reason": "AltDiffusion docs vs early model tests; no shared bug or change." + }, + { + "left": "pull_request:1352", + "right": "pull_request:842", + "accept": false, + "reason": "AltDiffusion sample model docs vs removing ['sample'] from pipeline outputs; unrelated." + }, + { + "left": "pull_request:1360", + "right": "pull_request:2516", + "accept": false, + "reason": "Community pipeline code/README update vs evaluation docs page; no common code-path problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 220, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1373", + "pull_request:1374", + "pull_request:1382", + "pull_request:1386", + "pull_request:1402", + "pull_request:1406", + "pull_request:1412", + "pull_request:1424", + "pull_request:1425", + "pull_request:1470", + "pull_request:1487", + "pull_request:1489", + "pull_request:15", + "pull_request:1510", + "pull_request:1520", + "pull_request:1525", + "pull_request:153", + "pull_request:154", + "pull_request:1547", + "pull_request:159", + "pull_request:1605", + "pull_request:1610", + "pull_request:1756", + "pull_request:1788", + "pull_request:1903", + "pull_request:2130", + "pull_request:2188", + "pull_request:227", + "pull_request:2496", + "pull_request:286", + "pull_request:289", + "pull_request:299", + "pull_request:3080", + "pull_request:3230", + "pull_request:34", + "pull_request:3608", + "pull_request:396", + "pull_request:408", + "pull_request:423", + "pull_request:43", + "pull_request:570", + "pull_request:662", + "pull_request:718", + "pull_request:768", + "pull_request:774", + "pull_request:800", + "pull_request:839", + "pull_request:842" + ], + "soft_pairs": [ + "pull_request:1373|pull_request:299", + "pull_request:1373|pull_request:800", + "pull_request:1374|pull_request:43", + "pull_request:1382|pull_request:286", + "pull_request:1382|pull_request:662", + "pull_request:1382|pull_request:718", + "pull_request:1382|pull_request:774", + "pull_request:1386|pull_request:227", + "pull_request:1402|pull_request:1425", + "pull_request:1406|pull_request:842", + "pull_request:1412|pull_request:3230", + "pull_request:1424|pull_request:1903", + "pull_request:1424|pull_request:396", + "pull_request:1424|pull_request:408", + "pull_request:1425|pull_request:159", + "pull_request:1425|pull_request:718", + "pull_request:1470|pull_request:3080", + "pull_request:1487|pull_request:1756", + "pull_request:1489|pull_request:2130", + "pull_request:1489|pull_request:2188", + "pull_request:1489|pull_request:2496", + "pull_request:1489|pull_request:289", + "pull_request:1489|pull_request:34", + "pull_request:1489|pull_request:839", + "pull_request:1489|pull_request:842", + "pull_request:15|pull_request:286", + "pull_request:15|pull_request:662", + "pull_request:15|pull_request:718", + "pull_request:1510|pull_request:1610", + "pull_request:1510|pull_request:423", + "pull_request:1520|pull_request:3608", + "pull_request:1525|pull_request:570", + "pull_request:153|pull_request:1788", + "pull_request:154|pull_request:1605", + "pull_request:1547|pull_request:768" + ], + "prepared_review_unit_hash": "14c15f67111aa2dd1fb4533de8a8b2b4d228909348e927f2bb1bfb69e4381a3e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 35, + "serialized_chars": 107187, + "estimated_input_tokens": 26797, + "estimated_eval_tokens": 53850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: the PRs cover unrelated docs, examples, schedulers, pipeline fixes, and feature additions. The soft links are mostly superficial overlaps in shared subsystem files, not true duplicate bugs/changes.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a canonical representative; the items are unrelated changes across docs, examples, schedulers, and different pipelines.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good global representative for duplicate triage because the cluster is not about one underlying change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1373", + "right": "pull_request:299", + "accept": false, + "reason": "Different changes: Versatile Diffusion extension vs a PNDM scheduler hotfix." + }, + { + "left": "pull_request:1373", + "right": "pull_request:800", + "accept": false, + "reason": "Unrelated features: Versatile Diffusion docs/code vs VQ Diffusion work-in-progress." + }, + { + "left": "pull_request:1374", + "right": "pull_request:43", + "accept": false, + "reason": "Unrelated examples/ops changes vs model card/template and scheduler refactor work." + }, + { + "left": "pull_request:1382", + "right": "pull_request:286", + "accept": false, + "reason": "Import/dependency fix vs a docs link correction; not the same code path." + }, + { + "left": "pull_request:1382", + "right": "pull_request:662", + "accept": false, + "reason": "Bad-dependency import cleanup vs DreamBooth/textual inversion content; different concerns." + }, + { + "left": "pull_request:1382", + "right": "pull_request:718", + "accept": false, + "reason": "Import fix vs typo fixes; no shared underlying bug." + }, + { + "left": "pull_request:1382", + "right": "pull_request:774", + "accept": false, + "reason": "Import fix vs pipeline test adjustments; not a duplicate change." + }, + { + "left": "pull_request:1386", + "right": "pull_request:227", + "accept": false, + "reason": "v-prediction scheduler support vs adding add_noise methods; different scheduler behavior." + }, + { + "left": "pull_request:1402", + "right": "pull_request:1425", + "accept": false, + "reason": "Pipeline loading fix vs docs typo cleanup; only broad repo overlap." + }, + { + "left": "pull_request:1406", + "right": "pull_request:842", + "accept": false, + "reason": "Config/deprecation behavior change vs output-access test cleanup; unrelated." + }, + { + "left": "pull_request:1412", + "right": "pull_request:3230", + "accept": false, + "reason": "SD2 inpainting tests vs IF pipeline feature addition; different pipelines and purposes." + }, + { + "left": "pull_request:1424", + "right": "pull_request:1903", + "accept": false, + "reason": "Stable Diffusion 2 docs vs depth2img comment cleanup; not the same change." + }, + { + "left": "pull_request:1424", + "right": "pull_request:396", + "accept": false, + "reason": "SD2 docs vs DDIM docs/pipeline work; unrelated documentation sets." + }, + { + "left": "pull_request:1424", + "right": "pull_request:408", + "accept": false, + "reason": "SD2 docs vs general docs typo cleanup; no duplicate bug/change." + }, + { + "left": "pull_request:1425", + "right": "pull_request:159", + "accept": false, + "reason": "Docs typo fixes in different docs areas; not the same underlying issue." + }, + { + "left": "pull_request:1425", + "right": "pull_request:718", + "accept": false, + "reason": "Docs typo cleanup vs broader typo/example fixes; too different to merge as one." + }, + { + "left": "pull_request:1470", + "right": "pull_request:3080", + "accept": false, + "reason": "DreamBooth alt-diffusion compatibility vs pipeline save_pretrained behavior; different code paths." + }, + { + "left": "pull_request:1487", + "right": "pull_request:1756", + "accept": false, + "reason": "xformers docs update vs attention-mask code fix; one is docs only, the other runtime behavior." + }, + { + "left": "pull_request:1489", + "right": "pull_request:2130", + "accept": false, + "reason": "Removing a reminder comment vs pipeline allow_patterns cleanup; unrelated maintenance." + }, + { + "left": "pull_request:1489", + "right": "pull_request:2188", + "accept": false, + "reason": "Removing a comment vs deleting a docs license note; different files and goals." + }, + { + "left": "pull_request:1489", + "right": "pull_request:2496", + "accept": false, + "reason": "Comment cleanup vs a code comment rename in UnCLIP; not the same issue." + }, + { + "left": "pull_request:1489", + "right": "pull_request:289", + "accept": false, + "reason": "Comment cleanup vs refactoring seed handling; unrelated changes." + }, + { + "left": "pull_request:1489", + "right": "pull_request:34", + "accept": false, + "reason": "A single test comment removal vs a typo fix in DDPM scheduler comments; not duplicates." + }, + { + "left": "pull_request:1489", + "right": "pull_request:839", + "accept": false, + "reason": "Comment cleanup vs dreambooth README auth-token change; unrelated." + }, + { + "left": "pull_request:1489", + "right": "pull_request:842", + "accept": false, + "reason": "Comment cleanup vs pipeline output API cleanup; different scope and code." + }, + { + "left": "pull_request:15", + "right": "pull_request:286", + "accept": false, + "reason": "Example output-path rename vs stable diffusion docs link fix; no shared bug/change." + }, + { + "left": "pull_request:15", + "right": "pull_request:662", + "accept": false, + "reason": "Example path rename vs textual inversion changes; unrelated example content." + }, + { + "left": "pull_request:15", + "right": "pull_request:718", + "accept": false, + "reason": "Example path rename vs typo fixes; only superficial repo overlap." + }, + { + "left": "pull_request:1510", + "right": "pull_request:1610", + "accept": false, + "reason": "Scheduler/docs conversion-script work vs pipeline test improvements; not the same issue." + }, + { + "left": "pull_request:1510", + "right": "pull_request:423", + "accept": false, + "reason": "Conversion script/scheduler additions vs output syntax updates; different feature sets." + }, + { + "left": "pull_request:1520", + "right": "pull_request:3608", + "accept": false, + "reason": "Imagic example fixes vs text-to-image training script typo fix; different examples and behavior." + }, + { + "left": "pull_request:1525", + "right": "pull_request:570", + "accept": false, + "reason": "Upscaling batch-size bug vs BaseOutput dict initialization bug; unrelated code paths." + }, + { + "left": "pull_request:153", + "right": "pull_request:1788", + "accept": false, + "reason": "Issue templates vs DDIM prediction_type bug fix; no overlap beyond being repo maintenance." + }, + { + "left": "pull_request:154", + "right": "pull_request:1605", + "accept": false, + "reason": "Stable diffusion conversion script vs Versatile Diffusion upcast_attention; different models/features." + }, + { + "left": "pull_request:1547", + "right": "pull_request:768", + "accept": false, + "reason": "DreamBooth collate_fn bugfix vs broad dreambooth enhancement; not the same concrete fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 221, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1548", + "pull_request:1558", + "pull_request:1564", + "pull_request:1582", + "pull_request:1583", + "pull_request:159", + "pull_request:1591", + "pull_request:1601", + "pull_request:1610", + "pull_request:1623", + "pull_request:1624", + "pull_request:1627", + "pull_request:17", + "pull_request:1706", + "pull_request:171", + "pull_request:1713", + "pull_request:1756", + "pull_request:1761", + "pull_request:1763", + "pull_request:1765", + "pull_request:1827", + "pull_request:1849", + "pull_request:1971", + "pull_request:2122", + "pull_request:217", + "pull_request:2279", + "pull_request:2282", + "pull_request:2467", + "pull_request:2542", + "pull_request:258", + "pull_request:2691", + "pull_request:3230", + "pull_request:3241", + "pull_request:3343", + "pull_request:35", + "pull_request:3653", + "pull_request:3712", + "pull_request:3723", + "pull_request:408", + "pull_request:423", + "pull_request:432", + "pull_request:433", + "pull_request:535", + "pull_request:568", + "pull_request:61", + "pull_request:63", + "pull_request:662", + "pull_request:828" + ], + "soft_pairs": [ + "pull_request:1548|pull_request:1971", + "pull_request:1548|pull_request:433", + "pull_request:1558|pull_request:3653", + "pull_request:1564|pull_request:2122", + "pull_request:1582|pull_request:1623", + "pull_request:1583|pull_request:1623", + "pull_request:159|pull_request:17", + "pull_request:159|pull_request:408", + "pull_request:159|pull_request:568", + "pull_request:1591|pull_request:3241", + "pull_request:1601|pull_request:535", + "pull_request:1610|pull_request:423", + "pull_request:1623|pull_request:2279", + "pull_request:1624|pull_request:3343", + "pull_request:1624|pull_request:35", + "pull_request:1624|pull_request:432", + "pull_request:1627|pull_request:3230", + "pull_request:17|pull_request:662", + "pull_request:1706|pull_request:3723", + "pull_request:171|pull_request:217", + "pull_request:1713|pull_request:61", + "pull_request:1713|pull_request:63", + "pull_request:1756|pull_request:1761", + "pull_request:1763|pull_request:2691", + "pull_request:1763|pull_request:3653", + "pull_request:1763|pull_request:3712", + "pull_request:1765|pull_request:2282", + "pull_request:1765|pull_request:2542", + "pull_request:1765|pull_request:828", + "pull_request:1827|pull_request:2467", + "pull_request:1849|pull_request:258" + ], + "prepared_review_unit_hash": "2d22dc04aba9ac67b221ec3f4a4d7a40a4b73543af5bdaf140b3cc6fb1f3181f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 103700, + "estimated_input_tokens": 25925, + "estimated_eval_tokens": 52106 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated PRs grouped by superficial similarity (typos, docs, scheduler/model fixes). No true duplicate pair stands out; all soft edges should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear canonical PR: the set mixes unrelated docs, lint, scheduler, model, and example changes rather than one duplicated change.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR: there is no single PR that meaningfully represents the rest of the cluster because they target different code paths and maintenance tasks.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1548", + "right": "pull_request:1971", + "accept": false, + "reason": "Both are text-only edits, but one updates an issue template and the other a scheduler docstring; not the same change." + }, + { + "left": "pull_request:1548", + "right": "pull_request:433", + "accept": false, + "reason": "Issue-template wording and black/version updates are unrelated maintenance changes." + }, + { + "left": "pull_request:1558", + "right": "pull_request:3653", + "accept": false, + "reason": "Both are small cleanup PRs, but one fixes typos in pipelines and the other is a lint refactor in an example; different scope." + }, + { + "left": "pull_request:1564", + "right": "pull_request:2122", + "accept": false, + "reason": "Both touch docs, but one adds an accelerate config note and the other adds safetensors documentation; unrelated topics." + }, + { + "left": "pull_request:1582", + "right": "pull_request:1623", + "accept": false, + "reason": "A RePaint scheduler bug fix and dependency version bumps are unrelated." + }, + { + "left": "pull_request:1583", + "right": "pull_request:1623", + "accept": false, + "reason": "An inpaint-legacy masking fix and requirements updates do not address the same bug." + }, + { + "left": "pull_request:159", + "right": "pull_request:17", + "accept": false, + "reason": "Both are typo cleanup, but they update different documents in different areas and do not share one underlying defect." + }, + { + "left": "pull_request:159", + "right": "pull_request:408", + "accept": false, + "reason": "Two separate documentation typo PRs; no shared code-path or single fix." + }, + { + "left": "pull_request:159", + "right": "pull_request:568", + "accept": false, + "reason": "Scheduler README typos and broader typo cleanup are different maintenance changes." + }, + { + "left": "pull_request:1591", + "right": "pull_request:3241", + "accept": false, + "reason": "One fixes slice-attention upcasting; the other loosens a fast test threshold. Test adjustment is not the same bug fix." + }, + { + "left": "pull_request:1601", + "right": "pull_request:535", + "accept": false, + "reason": "Classifier-free-guidance control flow and a resnet upsample implementation fix are unrelated code paths." + }, + { + "left": "pull_request:1610", + "right": "pull_request:423", + "accept": false, + "reason": "An attention/xformers-related change and docs/output syntax cleanup are unrelated." + }, + { + "left": "pull_request:1623", + "right": "pull_request:2279", + "accept": false, + "reason": "Both are maintenance/tooling changes, but one bumps requirements while the other swaps linters and black versions; not the same change." + }, + { + "left": "pull_request:1624", + "right": "pull_request:3343", + "accept": false, + "reason": "Both are DPM scheduler-related, but one corrects timestep generation and the other fixes a config reference; different concrete bugs." + }, + { + "left": "pull_request:1624", + "right": "pull_request:35", + "accept": false, + "reason": "Scheduler timestep logic and an analytic DPM pipeline addition are unrelated." + }, + { + "left": "pull_request:1624", + "right": "pull_request:432", + "accept": false, + "reason": "A scheduler algorithm fix and a docs link correction are unrelated." + }, + { + "left": "pull_request:1627", + "right": "pull_request:3230", + "accept": false, + "reason": "Re-enabling xformers attention on UNet2DCondition and adding IF are unrelated feature/fix changes." + }, + { + "left": "pull_request:17", + "right": "pull_request:662", + "accept": false, + "reason": "Both mention README/docs, but one is a typo fix and the other is a 6GB textual inversion tweak; different change sets." + }, + { + "left": "pull_request:1706", + "right": "pull_request:3723", + "accept": false, + "reason": "Main docs refresh and reference-example handling are separate documentation/example updates." + }, + { + "left": "pull_request:171", + "right": "pull_request:217", + "accept": false, + "reason": "Eta-condition logic in latent diffusion and a safety checker bug are unrelated." + }, + { + "left": "pull_request:1713", + "right": "pull_request:61", + "accept": false, + "reason": "Different conversion scripts for different pipelines; not the same underlying bug or change." + }, + { + "left": "pull_request:1713", + "right": "pull_request:63", + "accept": false, + "reason": "Both are conversion helpers, but they target different source pipelines and conversion logic." + }, + { + "left": "pull_request:1756", + "right": "pull_request:2691", + "accept": false, + "reason": "Attention-mask support and module renaming touch the same file area but solve different problems." + }, + { + "left": "pull_request:1763", + "right": "pull_request:3653", + "accept": false, + "reason": "Versatile-diffusion attention-mask handling and a linting cleanup are unrelated." + }, + { + "left": "pull_request:1763", + "right": "pull_request:3712", + "accept": false, + "reason": "A model behavior fix and workflow secret renaming are unrelated." + }, + { + "left": "pull_request:1765", + "right": "pull_request:2282", + "accept": false, + "reason": "DreamBooth Flax fixes and miscellaneous converter/docs fixes do not share the same code-path problem." + }, + { + "left": "pull_request:1765", + "right": "pull_request:2542", + "accept": false, + "reason": "Flax DreamBooth training fixes and ControlNet docs fixes are unrelated." + }, + { + "left": "pull_request:1765", + "right": "pull_request:828", + "accept": false, + "reason": "Both involve Flax, but one changes training logic and the other updates tests; not the same concrete bug." + }, + { + "left": "pull_request:1827", + "right": "pull_request:2467", + "accept": false, + "reason": "Adding add_noise to SDE-VE and adding latent output types are unrelated pipeline/scheduler changes." + }, + { + "left": "pull_request:1849", + "right": "pull_request:258", + "accept": false, + "reason": "Versatile Diffusion cross_attention_kwargs plumbing and an inpainting script cleanup are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 222, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1861", + "pull_request:1868", + "pull_request:1907", + "pull_request:1910", + "pull_request:1920", + "pull_request:1937", + "pull_request:1939", + "pull_request:1942", + "pull_request:1950", + "pull_request:1951", + "pull_request:1971", + "pull_request:2004", + "pull_request:2006", + "pull_request:2007", + "pull_request:2048", + "pull_request:2057", + "pull_request:2086", + "pull_request:21", + "pull_request:217", + "pull_request:2190", + "pull_request:2208", + "pull_request:2489", + "pull_request:258", + "pull_request:2641", + "pull_request:286", + "pull_request:3069", + "pull_request:3074", + "pull_request:312", + "pull_request:3127", + "pull_request:3262", + "pull_request:3343", + "pull_request:3402", + "pull_request:3433", + "pull_request:3471", + "pull_request:3653", + "pull_request:3681", + "pull_request:3712", + "pull_request:3804", + "pull_request:4177", + "pull_request:433", + "pull_request:662", + "pull_request:718", + "pull_request:75", + "pull_request:774", + "pull_request:798", + "pull_request:822", + "pull_request:828" + ], + "soft_pairs": [ + "pull_request:1861|pull_request:433", + "pull_request:1868|pull_request:217", + "pull_request:1868|pull_request:312", + "pull_request:1868|pull_request:3343", + "pull_request:1868|pull_request:3402", + "pull_request:1868|pull_request:3433", + "pull_request:1868|pull_request:75", + "pull_request:1907|pull_request:2190", + "pull_request:1910|pull_request:3471", + "pull_request:1920|pull_request:2641", + "pull_request:1937|pull_request:3804", + "pull_request:1939|pull_request:217", + "pull_request:1939|pull_request:312", + "pull_request:1939|pull_request:3402", + "pull_request:1939|pull_request:3433", + "pull_request:1942|pull_request:3127", + "pull_request:1950|pull_request:286", + "pull_request:1950|pull_request:3653", + "pull_request:1950|pull_request:3712", + "pull_request:1950|pull_request:662", + "pull_request:1950|pull_request:718", + "pull_request:1950|pull_request:774", + "pull_request:1951|pull_request:822", + "pull_request:1971|pull_request:3069", + "pull_request:1971|pull_request:3074", + "pull_request:1971|pull_request:798", + "pull_request:2004|pull_request:258", + "pull_request:2006|pull_request:3402", + "pull_request:2006|pull_request:3433", + "pull_request:2007|pull_request:2208", + "pull_request:2007|pull_request:3262", + "pull_request:2048|pull_request:3681", + "pull_request:2057|pull_request:4177", + "pull_request:2086|pull_request:2489", + "pull_request:21|pull_request:828" + ], + "prepared_review_unit_hash": "74f17c9999791889be88050c279595498fe2fbc5231dd9a0c50848d746f2615d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 35, + "serialized_chars": 98594, + "estimated_input_tokens": 24649, + "estimated_eval_tokens": 49554 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a noisy mix of unrelated pull requests across docs, formatting, example scripts, conversion utilities, schedulers, and training helpers. The soft edges are false positives driven by shared files or broad subsystem overlap, not the same underlying bug/change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits: the items span unrelated fixes and features, and the apparent overlaps are mostly superficial (shared files, docs, or style cleanup).", + "best_issue_reason": null, + "best_pr_reason": "If a single anchor is needed, PR 2048 is the most substantial cross-cutting maintenance change in the set, but it still is not a duplicate target for the other PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1861", + "right": "pull_request:433", + "accept": false, + "reason": "Both touch training/example code, but one changes loss computation while the other is a black-version/lint update; different changes." + }, + { + "left": "pull_request:1868", + "right": "pull_request:217", + "accept": false, + "reason": "EMA decay logic and the safety checker bug are unrelated code paths." + }, + { + "left": "pull_request:1868", + "right": "pull_request:312", + "accept": false, + "reason": "Different fixes: EMA scheduling vs safety-checker image handling." + }, + { + "left": "pull_request:1868", + "right": "pull_request:3343", + "accept": false, + "reason": "Training EMA decay and a scheduler config typo are not the same underlying issue." + }, + { + "left": "pull_request:1868", + "right": "pull_request:3402", + "accept": false, + "reason": "EMA decay logic and a Dockerfile repair are unrelated." + }, + { + "left": "pull_request:1868", + "right": "pull_request:3433", + "accept": false, + "reason": "One is EMA behavior in training code; the other is doc formatting cleanup." + }, + { + "left": "pull_request:1868", + "right": "pull_request:75", + "accept": false, + "reason": "Different model/code-path changes; no concrete bug overlap." + }, + { + "left": "pull_request:1907", + "right": "pull_request:2190", + "accept": false, + "reason": "A Dockerfile commit and a large notebook/example merge are unrelated." + }, + { + "left": "pull_request:1910", + "right": "pull_request:3471", + "accept": false, + "reason": "Korean docs initialization and a DreamBooth compile note are different documentation changes." + }, + { + "left": "pull_request:1920", + "right": "pull_request:2641", + "accept": false, + "reason": "Docs translation guide vs notebook-build workflow/config change are different tasks." + }, + { + "left": "pull_request:1937", + "right": "pull_request:3804", + "accept": false, + "reason": "EMA-weight extraction in the conversion script is unrelated to legacy attention weight conversion." + }, + { + "left": "pull_request:1939", + "right": "pull_request:217", + "accept": false, + "reason": "Logo path fix and safety-checker logic are unrelated." + }, + { + "left": "pull_request:1939", + "right": "pull_request:312", + "accept": false, + "reason": "A README asset path fix is not the same as the safety-checker/img2img bug fix." + }, + { + "left": "pull_request:1939", + "right": "pull_request:3402", + "accept": false, + "reason": "README logo path and Docker build changes are unrelated." + }, + { + "left": "pull_request:1939", + "right": "pull_request:3433", + "accept": false, + "reason": "A path correction in README is unrelated to style-rendering tweaks." + }, + { + "left": "pull_request:1942", + "right": "pull_request:3127", + "accept": false, + "reason": "Converter CLI flag cleanup and img2img safety-checker behavior are distinct changes." + }, + { + "left": "pull_request:1950", + "right": "pull_request:286", + "accept": false, + "reason": "OneFlow integration and a README link fix are unrelated." + }, + { + "left": "pull_request:1950", + "right": "pull_request:3653", + "accept": false, + "reason": "OneFlow feature code and a linting-only change are not the same fix." + }, + { + "left": "pull_request:1950", + "right": "pull_request:3712", + "accept": false, + "reason": "OneFlow integration and CI secret wiring are unrelated." + }, + { + "left": "pull_request:1950", + "right": "pull_request:662", + "accept": false, + "reason": "They both touch example/pipeline plumbing, but one adds OneFlow support while the other is a 6GB textual-inversion workaround; different features." + }, + { + "left": "pull_request:1950", + "right": "pull_request:718", + "accept": false, + "reason": "OneFlow integration and typo cleanup are unrelated." + }, + { + "left": "pull_request:1950", + "right": "pull_request:774", + "accept": false, + "reason": "A OneFlow port and a test-adjustment PR are different changes." + }, + { + "left": "pull_request:1951", + "right": "pull_request:822", + "accept": false, + "reason": "Latest ColossalAI integration and Flax scheduler API updates are unrelated." + }, + { + "left": "pull_request:1971", + "right": "pull_request:3069", + "accept": false, + "reason": "PNDM docstring wording and Flax TPU test updates are unrelated." + }, + { + "left": "pull_request:1971", + "right": "pull_request:3074", + "accept": false, + "reason": "A scheduler docstring edit and a Transformer2DModel docstring correction are separate documentation fixes." + }, + { + "left": "pull_request:1971", + "right": "pull_request:798", + "accept": false, + "reason": "Different docstring cleanup in different modules; not the same issue." + }, + { + "left": "pull_request:2004", + "right": "pull_request:258", + "accept": false, + "reason": "DreamBooth README typos and an inpainting script fix are unrelated." + }, + { + "left": "pull_request:2006", + "right": "pull_request:3402", + "accept": false, + "reason": "Checkpoint-merger logic and Dockerfile changes are unrelated." + }, + { + "left": "pull_request:2006", + "right": "pull_request:3433", + "accept": false, + "reason": "Checkpoint-merger fixes are not the same as docs style rendering changes." + }, + { + "left": "pull_request:2007", + "right": "pull_request:2208", + "accept": false, + "reason": "Black-version update and xFormers docs updates are unrelated maintenance changes." + }, + { + "left": "pull_request:2007", + "right": "pull_request:3262", + "accept": false, + "reason": "A formatting bump and IF model-name renaming are unrelated." + }, + { + "left": "pull_request:2048", + "right": "pull_request:3681", + "accept": false, + "reason": "Both concern training/checkpointing, but 2048 changes accelerate save/load hooks while 3681 enforces manual checkpoint limits across scripts; different concrete fixes." + }, + { + "left": "pull_request:2057", + "right": "pull_request:4177", + "accept": false, + "reason": "Safetensors download gating and making safetensors a required dependency are distinct concerns." + }, + { + "left": "pull_request:2086", + "right": "pull_request:2489", + "accept": false, + "reason": "LoRA docs and Core ML docs are separate documentation additions." + }, + { + "left": "pull_request:21", + "right": "pull_request:828", + "accept": false, + "reason": "These are different test suites and different model families; no shared bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 223, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2118", + "pull_request:2122", + "pull_request:2126", + "pull_request:2130", + "pull_request:2134", + "pull_request:2137", + "pull_request:2138", + "pull_request:217", + "pull_request:2188", + "pull_request:2197", + "pull_request:2198", + "pull_request:22", + "pull_request:2208", + "pull_request:2227", + "pull_request:2267", + "pull_request:2282", + "pull_request:2293", + "pull_request:2322", + "pull_request:2327", + "pull_request:2329", + "pull_request:2335", + "pull_request:2337", + "pull_request:2542", + "pull_request:286", + "pull_request:2861", + "pull_request:289", + "pull_request:3012", + "pull_request:3262", + "pull_request:3321", + "pull_request:3331", + "pull_request:3370", + "pull_request:3663", + "pull_request:3798", + "pull_request:3972", + "pull_request:3976", + "pull_request:41", + "pull_request:4211", + "pull_request:4214", + "pull_request:4239", + "pull_request:4320", + "pull_request:4380", + "pull_request:693", + "pull_request:737", + "pull_request:75", + "pull_request:828", + "pull_request:839", + "pull_request:842" + ], + "soft_pairs": [ + "pull_request:2118|pull_request:4239", + "pull_request:2122|pull_request:3798", + "pull_request:2126|pull_request:3321", + "pull_request:2130|pull_request:289", + "pull_request:2130|pull_request:3972", + "pull_request:2130|pull_request:839", + "pull_request:2130|pull_request:842", + "pull_request:2134|pull_request:4239", + "pull_request:2137|pull_request:4239", + "pull_request:2138|pull_request:3012", + "pull_request:2138|pull_request:3331", + "pull_request:2138|pull_request:3976", + "pull_request:217|pull_request:41", + "pull_request:217|pull_request:693", + "pull_request:217|pull_request:75", + "pull_request:2188|pull_request:289", + "pull_request:2188|pull_request:3972", + "pull_request:2188|pull_request:839", + "pull_request:2188|pull_request:842", + "pull_request:2197|pull_request:2227", + "pull_request:2198|pull_request:4214", + "pull_request:22|pull_request:828", + "pull_request:2208|pull_request:3262", + "pull_request:2208|pull_request:3370", + "pull_request:2227|pull_request:4320", + "pull_request:2267|pull_request:737", + "pull_request:2282|pull_request:2542", + "pull_request:2282|pull_request:4380", + "pull_request:2293|pull_request:4211", + "pull_request:2322|pull_request:3663", + "pull_request:2327|pull_request:2861", + "pull_request:2329|pull_request:289", + "pull_request:2329|pull_request:828", + "pull_request:2335|pull_request:286", + "pull_request:2337|pull_request:289", + "pull_request:2337|pull_request:3972", + "pull_request:2337|pull_request:839", + "pull_request:2337|pull_request:842" + ], + "prepared_review_unit_hash": "9be03f22a11b43905feda38e467580634a57ccf33261c0c93ae85b05f40ec7d7", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 38, + "serialized_chars": 95329, + "estimated_input_tokens": 23833, + "estimated_eval_tokens": 47922 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated tiny PRs: typo fixes, docs edits, test cleanup, and a few small code changes. The surface similarity is high, but the underlying changes do not match closely enough to treat them as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:2118", + "right": "pull_request:4239", + "accept": false, + "reason": "Different naming fixes in different files; one corrects an import-utils typo, the other changes SDXL training script imports/logic." + }, + { + "left": "pull_request:2122", + "right": "pull_request:3798", + "accept": false, + "reason": "Both are docs-related, but one adds safetensors docs while the other fixes comments in training scripts; no same underlying change." + }, + { + "left": "pull_request:2126", + "right": "pull_request:3321", + "accept": false, + "reason": "Unrelated docstring cleanup in pipeline_utils vs a large docs expansion for controlling generation." + }, + { + "left": "pull_request:2130", + "right": "pull_request:289", + "accept": false, + "reason": "Different fixes in pipeline_utils allow_patterns vs scheduler/generator refactor across multiple pipelines." + }, + { + "left": "pull_request:2130", + "right": "pull_request:3972", + "accept": false, + "reason": "Different code paths: allow_patterns cleanup versus GitHub Actions workflow refactor." + }, + { + "left": "pull_request:2130", + "right": "pull_request:839", + "accept": false, + "reason": "Different files and issues; pipeline loading patterns vs example README token usage." + }, + { + "left": "pull_request:2130", + "right": "pull_request:842", + "accept": false, + "reason": "Different scope; pipeline_utils loader cleanup does not match removing ['sample'] indexing in tests." + }, + { + "left": "pull_request:2134", + "right": "pull_request:4239", + "accept": false, + "reason": "Both are typo-oriented, but one is a README spelling fix and the other is a code-level SDXL training script change." + }, + { + "left": "pull_request:2137", + "right": "pull_request:4239", + "accept": false, + "reason": "Loaders.py typo/docstring cleanup is unrelated to the SDXL text_encoder code changes." + }, + { + "left": "pull_request:2138", + "right": "pull_request:3012", + "accept": false, + "reason": "Docs typo fix in loaders.mdx versus a comment typo in a training script; unrelated content." + }, + { + "left": "pull_request:2138", + "right": "pull_request:3331", + "accept": false, + "reason": "One is loaders API docs, the other is an inpainting docs variable-name typo; different documents and fixes." + }, + { + "left": "pull_request:2138", + "right": "pull_request:3976", + "accept": false, + "reason": "Different typo fix areas: loaders docs vs safetensors error-message typos in code." + }, + { + "left": "pull_request:217", + "right": "pull_request:41", + "accept": false, + "reason": "Both touch comments/cleanup, but one fixes a safety checker bug and the other only comment formatting in resnet files." + }, + { + "left": "pull_request:217", + "right": "pull_request:693", + "accept": false, + "reason": "Safety-check logic bug fix is unrelated to a BibTeX citation formatting change." + }, + { + "left": "pull_request:217", + "right": "pull_request:75", + "accept": false, + "reason": "Code bug in safety checker vs attention initialization/refactor; different concrete problems." + }, + { + "left": "pull_request:2188", + "right": "pull_request:289", + "accept": false, + "reason": "Docs removal of license text is unrelated to the generator/scheduler refactor in sampling pipelines." + }, + { + "left": "pull_request:2188", + "right": "pull_request:3972", + "accept": false, + "reason": "Docs-only license removal vs workflow refactor; no shared bug/change." + }, + { + "left": "pull_request:2188", + "right": "pull_request:839", + "accept": false, + "reason": "Different docs: inpaint license warning removal vs token usage cleanup in DreamBooth README." + }, + { + "left": "pull_request:2188", + "right": "pull_request:842", + "accept": false, + "reason": "Inpaint docs license section removal is unrelated to test output-access cleanup." + }, + { + "left": "pull_request:2197", + "right": "pull_request:2227", + "accept": false, + "reason": "Both touch prompt argument docs, but one is a typo fix in a single inpaint pipeline; the other updates negative_prompt docs across many pipelines." + }, + { + "left": "pull_request:2198", + "right": "pull_request:4214", + "accept": false, + "reason": "Missing-key load error handling is unrelated to the custom diffusion bf16 error fix." + }, + { + "left": "pull_request:22", + "right": "pull_request:828", + "accept": false, + "reason": "Very different changes: model loading warnings/error messages vs Flax test completion and tolerance updates." + }, + { + "left": "pull_request:2208", + "right": "pull_request:3262", + "accept": false, + "reason": "xFormers docs update for installation is unrelated to IF model naming changes." + }, + { + "left": "pull_request:2208", + "right": "pull_request:3370", + "accept": false, + "reason": "Both are optimization docs, but one is a targeted xFormers install update and the other is a broad PyTorch 2.0 doc rewrite." + }, + { + "left": "pull_request:2227", + "right": "pull_request:4320", + "accept": false, + "reason": "Negative_prompt typo docs across pipelines is unrelated to a small ONNX Runtime docs wording fix." + }, + { + "left": "pull_request:2267", + "right": "pull_request:737", + "accept": false, + "reason": "Both edit training examples, but one removes datasets logging/imports and the other removes use_auth_token instructions." + }, + { + "left": "pull_request:2282", + "right": "pull_request:2542", + "accept": false, + "reason": "Mixed script/doc typo fixes in conversion tooling vs ControlNet docs link fixes; not the same change." + }, + { + "left": "pull_request:2282", + "right": "pull_request:4380", + "accept": false, + "reason": "Conversion-script typo fixes are unrelated to SDXL training docs additions." + }, + { + "left": "pull_request:2293", + "right": "pull_request:4211", + "accept": false, + "reason": "Docstring fixes in convert_from_ckpt.py are unrelated to SDXL inpainting docstring corrections." + }, + { + "left": "pull_request:2322", + "right": "pull_request:3663", + "accept": false, + "reason": "These are not the same underlying update; one is a new example app and the other is a docs addition for textual inversion utilities." + }, + { + "left": "pull_request:2327", + "right": "pull_request:2861", + "accept": false, + "reason": "UNet input validation changes are unrelated to adding a Stable UnCLIP test for image_embeds=None." + }, + { + "left": "pull_request:2329", + "right": "pull_request:289", + "accept": false, + "reason": "Workflow refactor and seed-handling refactor are different kinds of changes with no shared bug." + }, + { + "left": "pull_request:2329", + "right": "pull_request:828", + "accept": false, + "reason": "CI workflow refactor is unrelated to Flax test changes." + }, + { + "left": "pull_request:2335", + "right": "pull_request:286", + "accept": false, + "reason": "Issue template/forum link updates are unrelated to a broken README link fix." + }, + { + "left": "pull_request:2337", + "right": "pull_request:289", + "accept": false, + "reason": "Test pruning in common pipeline tests is unrelated to generator/seeding refactors in pipelines." + }, + { + "left": "pull_request:2337", + "right": "pull_request:3972", + "accept": false, + "reason": "Removing unnecessary tests is unrelated to a documentation workflow restructure." + }, + { + "left": "pull_request:2337", + "right": "pull_request:839", + "accept": false, + "reason": "Different scope: test cleanup versus README auth-token cleanup." + }, + { + "left": "pull_request:2337", + "right": "pull_request:842", + "accept": false, + "reason": "Test removal in one area is not the same as changing pipeline return handling in another test file." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 224, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2349", + "pull_request:2362", + "pull_request:2367", + "pull_request:2386", + "pull_request:2398", + "pull_request:2400", + "pull_request:2415", + "pull_request:2428", + "pull_request:2436", + "pull_request:2467", + "pull_request:2481", + "pull_request:2487", + "pull_request:2489", + "pull_request:2533", + "pull_request:2539", + "pull_request:2542", + "pull_request:2557", + "pull_request:2569", + "pull_request:2641", + "pull_request:2715", + "pull_request:274", + "pull_request:2873", + "pull_request:289", + "pull_request:3022", + "pull_request:327", + "pull_request:3271", + "pull_request:3389", + "pull_request:340", + "pull_request:3561", + "pull_request:3687", + "pull_request:3757", + "pull_request:391", + "pull_request:3972", + "pull_request:403", + "pull_request:4050", + "pull_request:4320", + "pull_request:4352", + "pull_request:4380", + "pull_request:4442", + "pull_request:4457", + "pull_request:4508", + "pull_request:4746", + "pull_request:4807", + "pull_request:710", + "pull_request:800", + "pull_request:839", + "pull_request:842" + ], + "soft_pairs": [ + "pull_request:2349|pull_request:2873", + "pull_request:2349|pull_request:391", + "pull_request:2349|pull_request:710", + "pull_request:2362|pull_request:3389", + "pull_request:2367|pull_request:274", + "pull_request:2386|pull_request:2715", + "pull_request:2398|pull_request:340", + "pull_request:2400|pull_request:4457", + "pull_request:2415|pull_request:4050", + "pull_request:2415|pull_request:4508", + "pull_request:2428|pull_request:289", + "pull_request:2428|pull_request:3972", + "pull_request:2428|pull_request:839", + "pull_request:2428|pull_request:842", + "pull_request:2436|pull_request:3022", + "pull_request:2436|pull_request:3271", + "pull_request:2467|pull_request:327", + "pull_request:2481|pull_request:4352", + "pull_request:2487|pull_request:4442", + "pull_request:2489|pull_request:2641", + "pull_request:2533|pull_request:3687", + "pull_request:2533|pull_request:4746", + "pull_request:2533|pull_request:800", + "pull_request:2539|pull_request:4807", + "pull_request:2542|pull_request:3757", + "pull_request:2542|pull_request:403", + "pull_request:2542|pull_request:4380", + "pull_request:2557|pull_request:3561", + "pull_request:2569|pull_request:4320" + ], + "prepared_review_unit_hash": "b7d0770b35d0ff45e981578bf530826b69e997f02e957e7ac3c41fa1ec8eb002", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 97436, + "estimated_input_tokens": 24359, + "estimated_eval_tokens": 48974 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a set of mostly unrelated docs, test, and small code PRs that were linked only by weak textual similarity; I do not see any true duplicate pair among the soft edges.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue exists in this cluster, so there is no canonical issue. As a representative PR, #4508 is the clearest self-contained code fix with a concrete runtime behavior change and a matching test update.", + "best_issue_reason": null, + "best_pr_reason": "#4508 is the strongest standalone PR in the set: it fixes a specific `from_pretrained` argument-forwarding bug (`use_onnx`) and includes a targeted test update, unlike the many docs-only or unrelated maintenance PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2349", + "right": "pull_request:2873", + "accept": false, + "reason": "Both are docs PRs, but one adds scheduler docs while the other edits contribution wording; different topics and no shared bug/change." + }, + { + "left": "pull_request:2349", + "right": "pull_request:391", + "accept": false, + "reason": "One documents Karras schedulers, the other adds PNDM docs; separate documentation work, not the same change." + }, + { + "left": "pull_request:2349", + "right": "pull_request:710", + "accept": false, + "reason": "Karras scheduler overview docs vs a docstring fix for UNet models; unrelated content." + }, + { + "left": "pull_request:2362", + "right": "pull_request:3389", + "accept": false, + "reason": "MPS test skip helper vs attention refactor integration testing; different purposes and code paths." + }, + { + "left": "pull_request:2367", + "right": "pull_request:274", + "accept": false, + "reason": "Checkpoint-limit training script updates vs adding a Stable Diffusion README; unrelated." + }, + { + "left": "pull_request:2386", + "right": "pull_request:2715", + "accept": false, + "reason": "UniPC docs cleanup vs broad typo fixes across models/pipelines; not the same concrete issue." + }, + { + "left": "pull_request:2398", + "right": "pull_request:340", + "accept": false, + "reason": "Pix2Pix Zero docs expansion vs re-adding LMS scheduler support to img2img; different artifacts and behavior." + }, + { + "left": "pull_request:2400", + "right": "pull_request:4457", + "accept": false, + "reason": "Both touch community docs, but they introduce different pipelines and examples; not the same underlying change." + }, + { + "left": "pull_request:2415", + "right": "pull_request:4050", + "accept": false, + "reason": "Pipeline utility bug handling vs README demo table updates; unrelated." + }, + { + "left": "pull_request:2415", + "right": "pull_request:4508", + "accept": false, + "reason": "Same file, but different bugs: one guards missing transformers in version checks, the other forwards `use_onnx`; they are distinct fixes." + }, + { + "left": "pull_request:2428", + "right": "pull_request:289", + "accept": false, + "reason": "Docs-only cleanup for panorama/pix2pix pages vs a refactor removing `set_seed`; no shared code-path problem." + }, + { + "left": "pull_request:2428", + "right": "pull_request:3972", + "accept": false, + "reason": "Docs page cleanup vs workflow restructuring; unrelated." + }, + { + "left": "pull_request:2428", + "right": "pull_request:839", + "accept": false, + "reason": "Docs cleanup vs removing `use_auth_token` from Dreambooth README; unrelated." + }, + { + "left": "pull_request:2428", + "right": "pull_request:842", + "accept": false, + "reason": "Docs cleanup vs test assertion change in pipelines; unrelated." + }, + { + "left": "pull_request:2436", + "right": "pull_request:3022", + "accept": false, + "reason": "Adding a demo link in docs vs adding `AttnProcessor2_0` to a type union; different scope and target." + }, + { + "left": "pull_request:2436", + "right": "pull_request:3271", + "accept": false, + "reason": "Attend-and-Excite docs update vs adding training-script links across docs; unrelated documentation changes." + }, + { + "left": "pull_request:2467", + "right": "pull_request:327", + "accept": false, + "reason": "Both involve pipeline output/API shape, but one adds latent output support broadly while the other is type hints for PNDM; not the same concrete fix." + }, + { + "left": "pull_request:2481", + "right": "pull_request:4352", + "accept": false, + "reason": "Unconditional training-script test fix vs Kandinsky combined pipeline feature additions; unrelated." + }, + { + "left": "pull_request:2487", + "right": "pull_request:4442", + "accept": false, + "reason": "JAX/Flax Stable Diffusion docs vs distilled SD docs; both docs, but different guides and topics." + }, + { + "left": "pull_request:2489", + "right": "pull_request:2641", + "accept": false, + "reason": "CoreML docs addition vs doc-builder workflow/notebook config changes; unrelated." + }, + { + "left": "pull_request:2533", + "right": "pull_request:3687", + "accept": false, + "reason": "RC1 testing workflow/setup pin vs a doc-builder test PR; different changes and intent." + }, + { + "left": "pull_request:2533", + "right": "pull_request:4746", + "accept": false, + "reason": "Dependency/workflow test PR vs SDXL inpainting WIP code changes; unrelated." + }, + { + "left": "pull_request:2533", + "right": "pull_request:800", + "accept": false, + "reason": "RC1 setup/workflow test PR vs VQ-Diffusion implementation work; unrelated." + }, + { + "left": "pull_request:2539", + "right": "pull_request:4807", + "accept": false, + "reason": "A small doc default-value fix in code vs adding Chinese Stable Diffusion docs; unrelated." + }, + { + "left": "pull_request:2542", + "right": "pull_request:3757", + "accept": false, + "reason": "ControlNet docs/overview updates vs a ControlNet model parameter change; same subsystem but not the same bug/change." + }, + { + "left": "pull_request:2542", + "right": "pull_request:403", + "accept": false, + "reason": "ControlNet docs fixes vs philosophy wording cleanup; unrelated." + }, + { + "left": "pull_request:2542", + "right": "pull_request:4380", + "accept": false, + "reason": "ControlNet docs changes vs broader training-doc edits; not the same underlying change." + }, + { + "left": "pull_request:2557", + "right": "pull_request:3561", + "accept": false, + "reason": "Both touch Dreambooth, but one adds intermediate logging and the other changes Dreambooth IF stage II upscaling behavior; distinct PRs." + }, + { + "left": "pull_request:2569", + "right": "pull_request:4320", + "accept": false, + "reason": "OpenVINO docs addition vs ONNX docs typo cleanup; different optimization docs and different edits." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 225, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:26", + "pull_request:2608", + "pull_request:2667", + "pull_request:2677", + "pull_request:2693", + "pull_request:27", + "pull_request:2759", + "pull_request:2764", + "pull_request:2767", + "pull_request:2794", + "pull_request:2795", + "pull_request:2808", + "pull_request:2823", + "pull_request:2846", + "pull_request:286", + "pull_request:289", + "pull_request:2897", + "pull_request:2912", + "pull_request:2958", + "pull_request:2959", + "pull_request:2968", + "pull_request:3054", + "pull_request:3172", + "pull_request:3449", + "pull_request:3757", + "pull_request:3846", + "pull_request:4556", + "pull_request:4614", + "pull_request:4741", + "pull_request:4807", + "pull_request:4811", + "pull_request:5063", + "pull_request:570", + "pull_request:662", + "pull_request:689", + "pull_request:705", + "pull_request:718", + "pull_request:768", + "pull_request:774", + "pull_request:78", + "pull_request:802", + "pull_request:823", + "pull_request:824", + "pull_request:839", + "pull_request:842", + "pull_request:87", + "pull_request:948" + ], + "soft_pairs": [ + "pull_request:26|pull_request:689", + "pull_request:2608|pull_request:4741", + "pull_request:2667|pull_request:2958", + "pull_request:2677|pull_request:4556", + "pull_request:2693|pull_request:570", + "pull_request:27|pull_request:286", + "pull_request:27|pull_request:662", + "pull_request:27|pull_request:718", + "pull_request:27|pull_request:774", + "pull_request:2759|pull_request:3449", + "pull_request:2764|pull_request:3172", + "pull_request:2767|pull_request:948", + "pull_request:2794|pull_request:768", + "pull_request:2795|pull_request:3054", + "pull_request:2808|pull_request:705", + "pull_request:2808|pull_request:824", + "pull_request:2823|pull_request:4807", + "pull_request:2846|pull_request:5063", + "pull_request:286|pull_request:78", + "pull_request:286|pull_request:802", + "pull_request:289|pull_request:839", + "pull_request:289|pull_request:842", + "pull_request:289|pull_request:87", + "pull_request:2897|pull_request:4614", + "pull_request:2912|pull_request:3757", + "pull_request:2958|pull_request:3757", + "pull_request:2958|pull_request:4811", + "pull_request:2959|pull_request:3846", + "pull_request:2968|pull_request:823" + ], + "prepared_review_unit_hash": "5fa264394462f3b72f36e882c2a70cd1408f7b4feacc6595ad50e931b005049a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 94862, + "estimated_input_tokens": 23716, + "estimated_eval_tokens": 47688 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated PRs with a few superficial title/file overlaps; I don\u2019t see any true duplicate pairs here. The strongest single change is the post-processing fix in PR 2968, but it is not a canonical duplicate target for the rest of the set.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a clear duplicate center for this mixed set; the items span docs typos, example scripts, tests, and unrelated API/features. If a representative PR must be chosen, 2968 is the cleanest concrete bug fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 2968 is the strongest standalone fix: it corrects a concrete post-processing bug in two image-to-image pipelines and is a plausible mergeable unit.", + "soft_edge_verdicts": [ + { + "left": "pull_request:26", + "right": "pull_request:689", + "accept": false, + "reason": "Both touch diffusion internals, but one fixes a scheduler state bug and the other is a performance refactor in attention; different underlying changes." + }, + { + "left": "pull_request:2608", + "right": "pull_request:4741", + "accept": false, + "reason": "Both are small docs/import cleanup style PRs, but they affect unrelated areas and do not fix the same bug or change." + }, + { + "left": "pull_request:2667", + "right": "pull_request:2958", + "accept": false, + "reason": "Both mention inpainting/controlnet-related pipelines, but they implement different features in different code paths and are not the same change." + }, + { + "left": "pull_request:2677", + "right": "pull_request:4556", + "accept": false, + "reason": "Both edit docs navigation, but one adds training docs and the other adds API docs for a different ControlNet variant." + }, + { + "left": "pull_request:2693", + "right": "pull_request:570", + "accept": false, + "reason": "An image link typo fix and a BaseOutput dict-init fix are unrelated." + }, + { + "left": "pull_request:27", + "right": "pull_request:286", + "accept": false, + "reason": "Both are minor README edits, but one fixes a checklist format and the other fixes a pipeline doc link." + }, + { + "left": "pull_request:27", + "right": "pull_request:662", + "accept": false, + "reason": "Checklist formatting in the top-level README is unrelated to the textual inversion example overhaul." + }, + { + "left": "pull_request:27", + "right": "pull_request:718", + "accept": false, + "reason": "Both are typo/cleanup PRs, but they change different files and different content; not the same bug." + }, + { + "left": "pull_request:27", + "right": "pull_request:774", + "accept": false, + "reason": "A checklist formatting tweak and a test adjustment are unrelated changes." + }, + { + "left": "pull_request:2759", + "right": "pull_request:3449", + "accept": false, + "reason": "Both are test/CI related, but one changes model training/test coverage and the other reworks workflow jobs; not the same fix." + }, + { + "left": "pull_request:2764", + "right": "pull_request:3172", + "accept": false, + "reason": "Both touch docs about modern PyTorch features, but one is a benchmark/docs update and the other is a reproducibility guide update tied to deterministic algorithms." + }, + { + "left": "pull_request:2767", + "right": "pull_request:948", + "accept": false, + "reason": "A LoRA script argument rename is unrelated to a DDPM doc comment typo." + }, + { + "left": "pull_request:2794", + "right": "pull_request:768", + "accept": false, + "reason": "Both modify dreambooth training, but one adds validation support while the other is a separate default-args enhancement." + }, + { + "left": "pull_request:2795", + "right": "pull_request:3054", + "accept": false, + "reason": "One is a LoRA training-script bugfix; the other is a docs wording cleanup. Different artifacts and change types." + }, + { + "left": "pull_request:2808", + "right": "pull_request:705", + "accept": false, + "reason": "Both involve tests/CI, but one relaxes model tests while the other changes CUDA workflow infrastructure; not the same issue." + }, + { + "left": "pull_request:2808", + "right": "pull_request:824", + "accept": false, + "reason": "Test relaxations in different pipelines are not the same concrete bug." + }, + { + "left": "pull_request:2823", + "right": "pull_request:4807", + "accept": false, + "reason": "Both are Stable UnCLIP docs, but one fixes usage examples and the other adds Chinese docs; different changes." + }, + { + "left": "pull_request:2846", + "right": "pull_request:5063", + "accept": false, + "reason": "Both touch Stable UnCLIP docs/tests, but one fixes test call order while the other updates guidance_rescale documentation." + }, + { + "left": "pull_request:286", + "right": "pull_request:78", + "accept": false, + "reason": "A broken README link fix is unrelated to a mid-block model refactor." + }, + { + "left": "pull_request:286", + "right": "pull_request:802", + "accept": false, + "reason": "Both are documentation edits, but one fixes a link and the other fixes code-example indentation in a different guide." + }, + { + "left": "pull_request:289", + "right": "pull_request:839", + "accept": false, + "reason": "One removes set_seed from pipeline code; the other removes use_auth_token from dreambooth docs. No shared bug." + }, + { + "left": "pull_request:289", + "right": "pull_request:842", + "accept": false, + "reason": "A refactor removing set_seed is unrelated to a test API cleanup." + }, + { + "left": "pull_request:289", + "right": "pull_request:87", + "accept": false, + "reason": "Both are broad refactors, but they target entirely different subsystems and eras; not the same underlying change." + }, + { + "left": "pull_request:2897", + "right": "pull_request:4614", + "accept": false, + "reason": "Both concern docs/loaders around Stable UnCLIP or single-file loading, but one adds an example and the other adds a config-file parameter; different behavior." + }, + { + "left": "pull_request:2912", + "right": "pull_request:3757", + "accept": false, + "reason": "A punctuation doc fix and a ControlNet constructor API change are unrelated." + }, + { + "left": "pull_request:2958", + "right": "pull_request:3757", + "accept": false, + "reason": "ControlNet inpaint pipeline addition and configurable conditioning channels are separate features, not one bug fix." + }, + { + "left": "pull_request:2958", + "right": "pull_request:4811", + "accept": false, + "reason": "Both add inpaint/controlnet pipelines, but one is for ControlNet in general and the other is SDXL-specific; they are different concrete implementations." + }, + { + "left": "pull_request:2959", + "right": "pull_request:3846", + "accept": false, + "reason": "A UNet3D test correction and a random-module seeding fix are unrelated." + }, + { + "left": "pull_request:2968", + "right": "pull_request:823", + "accept": false, + "reason": "Both adjust pipeline behavior, but one fixes post-processing placement and the other fixes negative-prompt type handling/tests; different code-path problems." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 226, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2983", + "pull_request:299", + "pull_request:3022", + "pull_request:3042", + "pull_request:3068", + "pull_request:3069", + "pull_request:3108", + "pull_request:312", + "pull_request:3126", + "pull_request:3166", + "pull_request:3187", + "pull_request:3230", + "pull_request:325", + "pull_request:3252", + "pull_request:327", + "pull_request:3294", + "pull_request:3318", + "pull_request:3326", + "pull_request:3331", + "pull_request:3334", + "pull_request:3337", + "pull_request:3341", + "pull_request:3433", + "pull_request:344", + "pull_request:3471", + "pull_request:3474", + "pull_request:3806", + "pull_request:3894", + "pull_request:3924", + "pull_request:4034", + "pull_request:41", + "pull_request:4172", + "pull_request:4215", + "pull_request:432", + "pull_request:4508", + "pull_request:4690", + "pull_request:4741", + "pull_request:4746", + "pull_request:5239", + "pull_request:5242", + "pull_request:5506", + "pull_request:75", + "pull_request:800", + "pull_request:826", + "pull_request:909", + "pull_request:972", + "pull_request:975" + ], + "soft_pairs": [ + "pull_request:2983|pull_request:4508", + "pull_request:299|pull_request:800", + "pull_request:3022|pull_request:3474", + "pull_request:3042|pull_request:3924", + "pull_request:3068|pull_request:909", + "pull_request:3069|pull_request:909", + "pull_request:3108|pull_request:4741", + "pull_request:3108|pull_request:972", + "pull_request:3108|pull_request:975", + "pull_request:312|pull_request:41", + "pull_request:312|pull_request:432", + "pull_request:312|pull_request:75", + "pull_request:3126|pull_request:4034", + "pull_request:3166|pull_request:3326", + "pull_request:3187|pull_request:3230", + "pull_request:3230|pull_request:3894", + "pull_request:3230|pull_request:4172", + "pull_request:3230|pull_request:5239", + "pull_request:3230|pull_request:5242", + "pull_request:325|pull_request:344", + "pull_request:3252|pull_request:3471", + "pull_request:3252|pull_request:3806", + "pull_request:327|pull_request:826", + "pull_request:3294|pull_request:5506", + "pull_request:3318|pull_request:4746", + "pull_request:3326|pull_request:3331", + "pull_request:3334|pull_request:4215", + "pull_request:3337|pull_request:4690", + "pull_request:3341|pull_request:3433" + ], + "prepared_review_unit_hash": "343c1f1476866205fddce8f967c0c9d2ad03f6f06b2cdbafd5d5f82b143a1849", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 94087, + "estimated_input_tokens": 23522, + "estimated_eval_tokens": 47300 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of unrelated diffusers PRs: docs edits, type-hint cleanups, scheduler tweaks, example updates, and several distinct feature additions. The soft links are mostly superficial filename/subsystem overlap rather than the same underlying fix or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good canonical representative because the cluster spans many unrelated changes with no common bug or shared code-path fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2983", + "right": "pull_request:4508", + "accept": false, + "reason": "Docs for installing test deps vs ONNX parameter handling in from_pretrained; unrelated changes." + }, + { + "left": "pull_request:299", + "right": "pull_request:800", + "accept": false, + "reason": "PNDM timestep fix vs VQ Diffusion model import/codebase addition; different problems and code paths." + }, + { + "left": "pull_request:3022", + "right": "pull_request:3474", + "accept": false, + "reason": "Both touch attention_processor.py, but one fixes a type-export union and the other adds docs pages and exports; not the same change." + }, + { + "left": "pull_request:3042", + "right": "pull_request:3924", + "accept": false, + "reason": "Accelerate minimum version bump for example requirements vs a docs parameter rename; unrelated." + }, + { + "left": "pull_request:3068", + "right": "pull_request:909", + "accept": false, + "reason": "Textual inversion docs update vs inpaint test model-id update; different artifacts and purpose." + }, + { + "left": "pull_request:3069", + "right": "pull_request:909", + "accept": false, + "reason": "Flax TPU test expectation changes vs pipeline inpaint test updates; unrelated test maintenance." + }, + { + "left": "pull_request:3108", + "right": "pull_request:4741", + "accept": false, + "reason": "Empty/unknown checkpoint-fix PR vs AudioLDM2 dummy-import fix; no shared bug." + }, + { + "left": "pull_request:3108", + "right": "pull_request:972", + "accept": false, + "reason": "Empty/unknown checkpoint-fix PR vs torch_dtype typo fix in README; unrelated." + }, + { + "left": "pull_request:3108", + "right": "pull_request:975", + "accept": false, + "reason": "Empty/unknown checkpoint-fix PR vs documentation title typo; unrelated." + }, + { + "left": "pull_request:312", + "right": "pull_request:41", + "accept": false, + "reason": "README link typo fix vs comment cleanup in resnet/unet files; different maintenance areas." + }, + { + "left": "pull_request:312", + "right": "pull_request:432", + "accept": false, + "reason": "Both are docs/link fixes, but one updates README community link and the other rewrites notebook links across many docs; not the same concrete change." + }, + { + "left": "pull_request:312", + "right": "pull_request:75", + "accept": false, + "reason": "README link fix vs attention implementation behavior change for Glide; unrelated." + }, + { + "left": "pull_request:3126", + "right": "pull_request:4034", + "accept": false, + "reason": "Multi-GPU example support docs/scripts vs LoRA unload utility and tests; distinct features." + }, + { + "left": "pull_request:3166", + "right": "pull_request:3326", + "accept": false, + "reason": "Textual inversion docs fix vs generic model-adaptation docs; unrelated documentation topics." + }, + { + "left": "pull_request:3187", + "right": "pull_request:3230", + "accept": false, + "reason": "Upcaler CPU offload change vs large IF pipeline addition; different pipeline and bug scope." + }, + { + "left": "pull_request:3230", + "right": "pull_request:3894", + "accept": false, + "reason": "IF pipeline addition vs Flax UNet dropout feature; unrelated subsystems." + }, + { + "left": "pull_request:3230", + "right": "pull_request:4172", + "accept": false, + "reason": "IF pipeline addition vs Flax casting utility refactor; unrelated code paths." + }, + { + "left": "pull_request:3230", + "right": "pull_request:5239", + "accept": false, + "reason": "IF pipeline addition vs AutoencoderKL encode docstring; unrelated." + }, + { + "left": "pull_request:3230", + "right": "pull_request:5242", + "accept": false, + "reason": "IF pipeline addition vs AutoencoderKL decode docstring; unrelated." + }, + { + "left": "pull_request:325", + "right": "pull_request:344", + "accept": false, + "reason": "Score SDE VE type-hint cleanup vs VAE type-hint cleanup; same style category, but different files and no shared bug." + }, + { + "left": "pull_request:3252", + "right": "pull_request:3471", + "accept": false, + "reason": "Docs for stateful offload methods vs DreamBooth torch.compile note; unrelated docs." + }, + { + "left": "pull_request:3252", + "right": "pull_request:3806", + "accept": false, + "reason": "Offload-state docs vs checkpoint-loading note for text-to-image training; different topics." + }, + { + "left": "pull_request:327", + "right": "pull_request:826", + "accept": false, + "reason": "PNDM pipeline type hints vs DreamBooth fp16/prior loss fix; unrelated code paths." + }, + { + "left": "pull_request:3294", + "right": "pull_request:5506", + "accept": false, + "reason": "LoRA prefix implementation work vs xformers/PEFT scale handling fix; not the same concrete fix." + }, + { + "left": "pull_request:3318", + "right": "pull_request:4746", + "accept": false, + "reason": "WIP IF textual inversion vs SDXL inpainting WIP; different model families and goals." + }, + { + "left": "pull_request:3326", + "right": "pull_request:3331", + "accept": false, + "reason": "New training guide for adapting a model vs an inpainting doc typo; unrelated docs." + }, + { + "left": "pull_request:3334", + "right": "pull_request:4215", + "accept": false, + "reason": "Docstring punctuation/links fix vs controlled-generation docs cleanup; different documentation changes." + }, + { + "left": "pull_request:3337", + "right": "pull_request:4690", + "accept": false, + "reason": "Pipeline postprocessing refactor vs DPMSolverMultistepScheduler refactor; different components and code paths." + }, + { + "left": "pull_request:3341", + "right": "pull_request:3433", + "accept": false, + "reason": "Dependency/style fix PR vs doc rendering spacing fixes; unrelated maintenance." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 227, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3345", + "pull_request:3359", + "pull_request:3388", + "pull_request:345", + "pull_request:3471", + "pull_request:3472", + "pull_request:352", + "pull_request:3524", + "pull_request:3541", + "pull_request:3547", + "pull_request:3563", + "pull_request:3572", + "pull_request:3577", + "pull_request:3587", + "pull_request:3607", + "pull_request:3614", + "pull_request:3625", + "pull_request:3657", + "pull_request:3676", + "pull_request:3709", + "pull_request:3727", + "pull_request:3766", + "pull_request:38", + "pull_request:3806", + "pull_request:382", + "pull_request:3866", + "pull_request:390", + "pull_request:3924", + "pull_request:3979", + "pull_request:4081", + "pull_request:4196", + "pull_request:4343", + "pull_request:4352", + "pull_request:4547", + "pull_request:4607", + "pull_request:4807", + "pull_request:5024", + "pull_request:5312", + "pull_request:5396", + "pull_request:5470", + "pull_request:5478", + "pull_request:5526", + "pull_request:5622", + "pull_request:5698", + "pull_request:5729", + "pull_request:7", + "pull_request:75", + "pull_request:819" + ], + "soft_pairs": [ + "pull_request:3345|pull_request:3524", + "pull_request:3345|pull_request:5622", + "pull_request:3359|pull_request:3866", + "pull_request:3388|pull_request:5478", + "pull_request:345|pull_request:819", + "pull_request:3471|pull_request:5312", + "pull_request:3472|pull_request:4343", + "pull_request:352|pull_request:7", + "pull_request:3524|pull_request:4081", + "pull_request:3541|pull_request:3979", + "pull_request:3547|pull_request:4607", + "pull_request:3563|pull_request:5698", + "pull_request:3572|pull_request:4196", + "pull_request:3577|pull_request:5729", + "pull_request:3587|pull_request:4352", + "pull_request:3607|pull_request:3924", + "pull_request:3607|pull_request:5396", + "pull_request:3614|pull_request:5526", + "pull_request:3625|pull_request:4547", + "pull_request:3625|pull_request:4807", + "pull_request:3657|pull_request:5470", + "pull_request:3676|pull_request:4547", + "pull_request:3676|pull_request:4807", + "pull_request:3709|pull_request:4807", + "pull_request:3727|pull_request:4547", + "pull_request:3766|pull_request:4352", + "pull_request:38|pull_request:75", + "pull_request:3806|pull_request:5024", + "pull_request:382|pull_request:390" + ], + "prepared_review_unit_hash": "d8daef4c57f67c0df18ef02843857cccdd0bb7d67c9229d8d3a9a672225e8b34", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 97637, + "estimated_input_tokens": 24410, + "estimated_eval_tokens": 49076 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All candidate pairs look like unrelated PRs rather than duplicates. They span separate docs edits, example-script changes, and distinct bug fixes/features, often only sharing a broad subsystem or filename.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the cluster is heterogeneous and does not represent one duplicate change set.", + "best_issue_reason": null, + "best_pr_reason": "No single PR is a good representative duplicate target because the items are unrelated and should stay separate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3345", + "right": "pull_request:3524", + "accept": false, + "reason": "Both touch README docs, but one is a specific pipeline-class fix in community examples and the other is a broad README rewrite; different changes." + }, + { + "left": "pull_request:3345", + "right": "pull_request:5622", + "accept": false, + "reason": "Completely unrelated: a small README pipeline-class fix vs a large examples/training-ops overhaul." + }, + { + "left": "pull_request:3359", + "right": "pull_request:3866", + "accept": false, + "reason": "Same example file, but different change: disabling safety_checker vs adding LoRA rank args; not the same bug or fix." + }, + { + "left": "pull_request:3388", + "right": "pull_request:5478", + "accept": false, + "reason": "Different documentation areas and purposes: install docs update vs full Japanese docs addition." + }, + { + "left": "pull_request:345", + "right": "pull_request:819", + "accept": false, + "reason": "Different pipelines and issues: type hints cleanup vs Windows int64 timestep bug fix." + }, + { + "left": "pull_request:3471", + "right": "pull_request:5312", + "accept": false, + "reason": "Dreambooth docs note vs model-library typehints/docs refactor; unrelated changes." + }, + { + "left": "pull_request:3472", + "right": "pull_request:4343", + "accept": false, + "reason": "One adds IF training docs, the other adds a rank CLI arg for SDXL LoRA training; different changes." + }, + { + "left": "pull_request:352", + "right": "pull_request:7", + "accept": false, + "reason": "Commands/env CLI addition vs initial VQ-VAE model implementation; unrelated." + }, + { + "left": "pull_request:3524", + "right": "pull_request:4081", + "accept": false, + "reason": "Both edit README.md, but they add different content sections/features and are not the same change." + }, + { + "left": "pull_request:3541", + "right": "pull_request:3979", + "accept": false, + "reason": "Different bug fixes in different code paths: Dreambooth attention-mask collation vs SDXL attention-processor check." + }, + { + "left": "pull_request:3547", + "right": "pull_request:4607", + "accept": false, + "reason": "Bibtex references file vs docs overview for PushToHubMixin; unrelated documentation work." + }, + { + "left": "pull_request:3563", + "right": "pull_request:5698", + "accept": false, + "reason": "Different model/pipeline fixes in unrelated code paths; no shared underlying bug." + }, + { + "left": "pull_request:3572", + "right": "pull_request:4196", + "accept": false, + "reason": "Both are docs-related, but one corrects ControlNet inpaint docs and the other improves SDXL API docs; different artifacts." + }, + { + "left": "pull_request:3577", + "right": "pull_request:5729", + "accept": false, + "reason": "README link fixes vs updating minimum version checks across training scripts; not the same issue." + }, + { + "left": "pull_request:3587", + "right": "pull_request:4352", + "accept": false, + "reason": "Different Kandinsky-related changes: a new community pipeline vs fixes to combined pipeline helper methods." + }, + { + "left": "pull_request:3607", + "right": "pull_request:3924", + "accept": false, + "reason": "Different subsystems: temb attention fix vs a docs code-snippet parameter rename." + }, + { + "left": "pull_request:3607", + "right": "pull_request:5396", + "accept": false, + "reason": "Attention/LoRA runtime fix vs stale-bot timezone handling; unrelated." + }, + { + "left": "pull_request:3614", + "right": "pull_request:5526", + "accept": false, + "reason": "Test cleanup vs scheduler step-count logic fix; not the same bug or change." + }, + { + "left": "pull_request:3625", + "right": "pull_request:4547", + "accept": false, + "reason": "Doc-builder workflow test PR vs a Flax scheduler implementation PR; unrelated." + }, + { + "left": "pull_request:3625", + "right": "pull_request:4807", + "accept": false, + "reason": "Doc-builder workflow test PR vs adding Chinese docs; unrelated." + }, + { + "left": "pull_request:3657", + "right": "pull_request:5470", + "accept": false, + "reason": "ControlNet checkpoint parsing tweaks vs ControlNet inpaint docs; different code paths and goals." + }, + { + "left": "pull_request:3676", + "right": "pull_request:4547", + "accept": false, + "reason": "Testing doc-builder change vs Flax scheduler implementation; unrelated." + }, + { + "left": "pull_request:3676", + "right": "pull_request:4807", + "accept": false, + "reason": "Testing doc-builder change vs adding Japanese docs; unrelated." + }, + { + "left": "pull_request:3709", + "right": "pull_request:4807", + "accept": false, + "reason": "New image-processor doc page vs a full Japanese docs translation; different changes." + }, + { + "left": "pull_request:3727", + "right": "pull_request:4547", + "accept": false, + "reason": "IF Dreambooth fine-tuning work vs scheduler implementation; unrelated." + }, + { + "left": "pull_request:3766", + "right": "pull_request:4352", + "accept": false, + "reason": "Both Kandinsky-related, but one updates conversion script and the other adds combined-pipeline methods/tests; not the same concrete fix." + }, + { + "left": "pull_request:38", + "right": "pull_request:75", + "accept": false, + "reason": "PR 38 is a broad attention-module unification; PR 75 is a later specific Glide attention fix. Related area, but not the same underlying change." + }, + { + "left": "pull_request:3806", + "right": "pull_request:5024", + "accept": false, + "reason": "Checkpoint-loading note vs scheduler step-index/noise-sampler fix; unrelated." + }, + { + "left": "pull_request:382", + "right": "pull_request:390", + "accept": false, + "reason": "DDPM pipeline docs vs Apple Silicon/MPS optimization docs; different documentation topics." + }, + { + "left": "pull_request:3866", + "right": "pull_request:3924", + "accept": false, + "reason": "LoRA rank argument enhancement vs a docs parameter-name update; unrelated changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 228, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:3850", + "pull_request:388", + "pull_request:391", + "pull_request:3928", + "pull_request:3931", + "pull_request:396", + "pull_request:397", + "pull_request:3971", + "pull_request:3972", + "pull_request:3979", + "pull_request:3985", + "pull_request:3989", + "pull_request:4", + "pull_request:401", + "pull_request:4010", + "pull_request:4058", + "pull_request:408", + "pull_request:4082", + "pull_request:41", + "pull_request:4106", + "pull_request:4112", + "pull_request:415", + "pull_request:416", + "pull_request:418", + "pull_request:4196", + "pull_request:4237", + "pull_request:4238", + "pull_request:4276", + "pull_request:4396", + "pull_request:4748", + "pull_request:4807", + "pull_request:4850", + "pull_request:498", + "pull_request:4984", + "pull_request:508", + "pull_request:5097", + "pull_request:5456", + "pull_request:5602", + "pull_request:5853", + "pull_request:6075", + "pull_request:6150", + "pull_request:6198", + "pull_request:6322", + "pull_request:6337", + "pull_request:6410", + "pull_request:693", + "pull_request:75", + "pull_request:921" + ], + "soft_pairs": [ + "pull_request:3850|pull_request:5853", + "pull_request:388|pull_request:391", + "pull_request:388|pull_request:401", + "pull_request:3928|pull_request:4396", + "pull_request:3931|pull_request:4748", + "pull_request:396|pull_request:397", + "pull_request:396|pull_request:415", + "pull_request:396|pull_request:418", + "pull_request:396|pull_request:508", + "pull_request:397|pull_request:408", + "pull_request:397|pull_request:416", + "pull_request:397|pull_request:921", + "pull_request:3971|pull_request:4010", + "pull_request:3971|pull_request:4196", + "pull_request:3971|pull_request:4984", + "pull_request:3972|pull_request:4850", + "pull_request:3979|pull_request:6410", + "pull_request:3985|pull_request:4807", + "pull_request:3989|pull_request:5097", + "pull_request:3989|pull_request:6075", + "pull_request:3989|pull_request:6198", + "pull_request:4|pull_request:75", + "pull_request:4058|pull_request:4237", + "pull_request:4058|pull_request:4238", + "pull_request:4058|pull_request:5602", + "pull_request:408|pull_request:415", + "pull_request:408|pull_request:418", + "pull_request:4082|pull_request:6322", + "pull_request:4082|pull_request:6337", + "pull_request:41|pull_request:498", + "pull_request:41|pull_request:693", + "pull_request:41|pull_request:75", + "pull_request:4106|pull_request:4276", + "pull_request:4106|pull_request:6150", + "pull_request:4112|pull_request:5456" + ], + "prepared_review_unit_hash": "bd72a1ce5e3ec82db4ab37a234220b206cf0458bc61ba4dc87112d8e5e693b82", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 35, + "serialized_chars": 107300, + "estimated_input_tokens": 26825, + "estimated_eval_tokens": 53906 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: docs refreshes, unrelated bug fixes, and test-only tweaks across different subsystems. I found no soft-edge pair that clearly describes the same underlying change, so I rejected all candidates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4237 is the strongest representative if one must be chosen: it fixes a concrete, shared SDXL code-path bug, has an explicit issue target, and includes regression tests.", + "best_issue_reason": null, + "best_pr_reason": "PR 4237 is the best standalone triage candidate: it is a real bug fix with tests and a clear problem statement, unlike the many docs/test-only PRs in the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3850", + "right": "pull_request:5853", + "accept": false, + "reason": "Audio Diffusion test fix vs doctest workflow/CI additions; unrelated areas and different underlying problems." + }, + { + "left": "pull_request:388", + "right": "pull_request:391", + "accept": false, + "reason": "Both are docs PRs, but they document different pipelines (latent diffusion vs PNDM), not the same change." + }, + { + "left": "pull_request:388", + "right": "pull_request:401", + "accept": false, + "reason": "Different pipeline docs and code updates: latent diffusion unconditional vs Karras-VE, no shared bug or change." + }, + { + "left": "pull_request:3928", + "right": "pull_request:4396", + "accept": false, + "reason": "Both touch SD/ControlNet code, but one fixes timestep gating and the other fixes prompt embedding dtype; distinct bugs." + }, + { + "left": "pull_request:3931", + "right": "pull_request:4748", + "accept": false, + "reason": "Checkpoint loading logic vs packaging/release instructions; no overlap in underlying issue." + }, + { + "left": "pull_request:396", + "right": "pull_request:397", + "accept": false, + "reason": "DDIM docs vs Quicktour docs; both documentation-only and not the same artifact or problem." + }, + { + "left": "pull_request:396", + "right": "pull_request:415", + "accept": false, + "reason": "Different docs areas: DDIM pipeline docs versus training docs." + }, + { + "left": "pull_request:396", + "right": "pull_request:418", + "accept": false, + "reason": "DDIM pipeline docs vs DiffusionPipeline docs; separate documentation changes." + }, + { + "left": "pull_request:396", + "right": "pull_request:508", + "accept": false, + "reason": "Docs PR versus scheduler implementation/Flax scheduler updates; not the same underlying change." + }, + { + "left": "pull_request:397", + "right": "pull_request:408", + "accept": false, + "reason": "Quicktour rewrite vs typo cleanup across docs; related only at a very broad docs level." + }, + { + "left": "pull_request:397", + "right": "pull_request:416", + "accept": false, + "reason": "Quicktour docs vs models API docs; different content and purpose." + }, + { + "left": "pull_request:397", + "right": "pull_request:921", + "accept": false, + "reason": "Both are docs, but one is quicktour/training content and the other is v1.5 release docs updates; not duplicate work." + }, + { + "left": "pull_request:3971", + "right": "pull_request:4010", + "accept": false, + "reason": "SDXL docs cleanup versus sequential offload code fix; different artifacts and different issues." + }, + { + "left": "pull_request:3971", + "right": "pull_request:4196", + "accept": false, + "reason": "Both are SDXL docs, but they cover different sections and edits; not the same concrete change." + }, + { + "left": "pull_request:3971", + "right": "pull_request:4984", + "accept": false, + "reason": "Docs workflow/content fix vs PR template edit; unrelated." + }, + { + "left": "pull_request:3972", + "right": "pull_request:4850", + "accept": false, + "reason": "Documentation workflow refactor vs deprecation-warning code change; unrelated." + }, + { + "left": "pull_request:3979", + "right": "pull_request:6410", + "accept": false, + "reason": "SDXL pipeline bug fix vs RL module registration refactor; no shared code path." + }, + { + "left": "pull_request:3985", + "right": "pull_request:4807", + "accept": false, + "reason": "Both are docs, but one is SDXL docs and the other is Chinese stable diffusion docs; different documentation targets." + }, + { + "left": "pull_request:3989", + "right": "pull_request:5097", + "accept": false, + "reason": "Sequential offload fix vs float16 test adjustments; test changes do not show the same bug being fixed." + }, + { + "left": "pull_request:3989", + "right": "pull_request:6075", + "accept": false, + "reason": "SDXL CPU offload fix vs backend cache-clearing test cleanup; unrelated problems." + }, + { + "left": "pull_request:3989", + "right": "pull_request:6198", + "accept": false, + "reason": "Pipeline offload bug fix vs benchmark reporting helper; different code paths and intent." + }, + { + "left": "pull_request:4", + "right": "pull_request:75", + "accept": false, + "reason": "Glide weight conversion utility vs attention-layer constructor fix; same broad model family, but not the same change." + }, + { + "left": "pull_request:4058", + "right": "pull_request:4237", + "accept": false, + "reason": "SDXL docs correction vs SDXL prompt-encoding bug fix; different type of change and no shared underlying defect." + }, + { + "left": "pull_request:4058", + "right": "pull_request:4238", + "accept": false, + "reason": "Docs-only PR vs SDXL ControlNet inference fix; not the same artifact or issue." + }, + { + "left": "pull_request:4058", + "right": "pull_request:5602", + "accept": false, + "reason": "Both touch docs, but one is SDXL prose/image formatting and the other is stable-diffusion overview layout; not duplicate." + }, + { + "left": "pull_request:408", + "right": "pull_request:415", + "accept": false, + "reason": "General doc typo cleanup versus training docs rewrite; too broad to be the same change." + }, + { + "left": "pull_request:408", + "right": "pull_request:418", + "accept": false, + "reason": "Typo fixes across docs versus DiffusionPipeline API docs rewrite; not the same underlying work." + }, + { + "left": "pull_request:4082", + "right": "pull_request:6322", + "accept": false, + "reason": "Requirement pin fix vs training-script scheduler change; unrelated." + }, + { + "left": "pull_request:4082", + "right": "pull_request:6337", + "accept": false, + "reason": "SDXL requirement adjustment vs retrieve_latents usage fix; different issues." + }, + { + "left": "pull_request:41", + "right": "pull_request:498", + "accept": false, + "reason": "Comment cleanup in model code versus attention-slicing API fix; unrelated." + }, + { + "left": "pull_request:41", + "right": "pull_request:693", + "accept": false, + "reason": "Comment formatting cleanup versus BibTeX citation correction; unrelated." + }, + { + "left": "pull_request:41", + "right": "pull_request:75", + "accept": false, + "reason": "Both involve Glide/attention code, but one changes comments and the other changes runtime constructor logic; not the same fix." + }, + { + "left": "pull_request:4106", + "right": "pull_request:4276", + "accept": false, + "reason": "New Tortoise TTS feature work vs import test coverage; different goals and code paths." + }, + { + "left": "pull_request:4106", + "right": "pull_request:6150", + "accept": false, + "reason": "Tortoise TTS implementation vs IP-Adapter image projection conversion; unrelated subsystems." + }, + { + "left": "pull_request:4112", + "right": "pull_request:5456", + "accept": false, + "reason": "DreamBooth LoRA rank enhancement vs RealFill feature addition; separate projects and changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 229, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4113", + "pull_request:415", + "pull_request:416", + "pull_request:418", + "pull_request:4187", + "pull_request:4188", + "pull_request:4196", + "pull_request:42", + "pull_request:421", + "pull_request:4211", + "pull_request:422", + "pull_request:423", + "pull_request:4239", + "pull_request:4275", + "pull_request:4276", + "pull_request:43", + "pull_request:4320", + "pull_request:4339", + "pull_request:4347", + "pull_request:4370", + "pull_request:4395", + "pull_request:4417", + "pull_request:4437", + "pull_request:4495", + "pull_request:4563", + "pull_request:4581", + "pull_request:4596", + "pull_request:46", + "pull_request:4635", + "pull_request:4690", + "pull_request:5456", + "pull_request:5552", + "pull_request:5707", + "pull_request:5768", + "pull_request:5937", + "pull_request:6002", + "pull_request:6053", + "pull_request:6078", + "pull_request:6326", + "pull_request:6337", + "pull_request:6360", + "pull_request:6437", + "pull_request:6591", + "pull_request:6800", + "pull_request:7", + "pull_request:91", + "pull_request:921", + "pull_request:96" + ], + "soft_pairs": [ + "pull_request:4113|pull_request:6002", + "pull_request:415|pull_request:921", + "pull_request:416|pull_request:418", + "pull_request:418|pull_request:921", + "pull_request:4187|pull_request:4690", + "pull_request:4188|pull_request:4370", + "pull_request:4196|pull_request:4635", + "pull_request:42|pull_request:96", + "pull_request:421|pull_request:422", + "pull_request:421|pull_request:91", + "pull_request:4211|pull_request:6002", + "pull_request:422|pull_request:423", + "pull_request:4239|pull_request:6326", + "pull_request:4275|pull_request:6337", + "pull_request:4275|pull_request:6591", + "pull_request:4276|pull_request:4417", + "pull_request:4276|pull_request:6437", + "pull_request:43|pull_request:7", + "pull_request:4320|pull_request:6337", + "pull_request:4320|pull_request:6591", + "pull_request:4339|pull_request:5937", + "pull_request:4347|pull_request:6053", + "pull_request:4395|pull_request:6078", + "pull_request:4437|pull_request:5768", + "pull_request:4495|pull_request:6360", + "pull_request:4495|pull_request:6800", + "pull_request:4563|pull_request:5707", + "pull_request:4563|pull_request:6053", + "pull_request:4581|pull_request:5552", + "pull_request:4596|pull_request:5456", + "pull_request:46|pull_request:96" + ], + "prepared_review_unit_hash": "5b7be3e4d407a036485d5742252cb2ee61b22fde953bd371cc15985df8dca220", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 104577, + "estimated_input_tokens": 26145, + "estimated_eval_tokens": 52546 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated docs, examples, tests, and small bugfix PRs. The only near-match is the DPMSolverMultistepScheduler pair, but even that looks like two separate scheduler changes rather than one duplicate fix.", + "confidence": 0.36, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4187 is the most concrete bug-fix style change in the only plausible overlap area (DPMSolverMultistepScheduler), while the rest are mostly unrelated docs/tests or separate feature work.", + "best_issue_reason": null, + "best_pr_reason": "PR 4187 is the best representative because it targets a specific code-path issue and is closer to a canonical bugfix than the surrounding documentation-heavy PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4113", + "right": "pull_request:6002", + "accept": false, + "reason": "Different pipelines and different problems: SDXL inpaint dtype handling vs Stable Video Diffusion guidance-scale logic." + }, + { + "left": "pull_request:415", + "right": "pull_request:921", + "accept": false, + "reason": "Both are docs, but they cover different sections and scope; not the same underlying change." + }, + { + "left": "pull_request:416", + "right": "pull_request:418", + "accept": false, + "reason": "Separate documentation updates for different API areas; no shared bug or change." + }, + { + "left": "pull_request:418", + "right": "pull_request:921", + "accept": false, + "reason": "Broad docs work in both cases, but the targets are different and not a single mergeable fix." + }, + { + "left": "pull_request:4187", + "right": "pull_request:4690", + "accept": false, + "reason": "Same scheduler file, but one is about duplicate timestep/index handling and the other is a sigma-based refactor; related, not the same concrete fix." + }, + { + "left": "pull_request:4188", + "right": "pull_request:4370", + "accept": false, + "reason": "ControlNet training cache fingerprinting vs SDXL dummy import cleanup are unrelated changes." + }, + { + "left": "pull_request:4196", + "right": "pull_request:4635", + "accept": false, + "reason": "SDXL docs cleanup vs MultiControlNet documentation; different documentation topics." + }, + { + "left": "pull_request:42", + "right": "pull_request:96", + "accept": false, + "reason": "Both touch early latent-diffusion modeling code, but they are different pieces of model/test work rather than one duplicate bug." + }, + { + "left": "pull_request:421", + "right": "pull_request:422", + "accept": false, + "reason": "Scheduler docs and BaseOutput docs are separate documentation changes." + }, + { + "left": "pull_request:421", + "right": "pull_request:91", + "accept": false, + "reason": "One is a docs fix, the other is a broad scheduler/model API implementation change." + }, + { + "left": "pull_request:4211", + "right": "pull_request:6002", + "accept": false, + "reason": "Different SDXL docs vs a Stable Video Diffusion runtime fix; no common underlying issue." + }, + { + "left": "pull_request:422", + "right": "pull_request:423", + "accept": false, + "reason": "Outputs docs versus a wider syntax/docs cleanup across many pipeline pages; not the same concrete change." + }, + { + "left": "pull_request:4239", + "right": "pull_request:6326", + "accept": false, + "reason": "DreamBooth text-encoder naming fix and TensorRT guidance-scale assignment fix are unrelated." + }, + { + "left": "pull_request:4275", + "right": "pull_request:6337", + "accept": false, + "reason": "Autopipeline docs fix vs a TensorRT latent-retrieval code fix; different concerns." + }, + { + "left": "pull_request:4275", + "right": "pull_request:6591", + "accept": false, + "reason": "Autopipeline docs and the text typo sweep across many examples are separate doc/code cleanup efforts." + }, + { + "left": "pull_request:4276", + "right": "pull_request:4417", + "accept": false, + "reason": "Pipeline import test coverage vs Zero123 feature support are unrelated." + }, + { + "left": "pull_request:4276", + "right": "pull_request:6437", + "accept": false, + "reason": "Pipeline import test vs CI exclusion change; not the same bug/change." + }, + { + "left": "pull_request:43", + "right": "pull_request:7", + "accept": false, + "reason": "Old model-card/template work and VQ-VAE addition are different features." + }, + { + "left": "pull_request:4320", + "right": "pull_request:6337", + "accept": false, + "reason": "ONNX docs typo fix vs latent retrieval code fix are unrelated." + }, + { + "left": "pull_request:4320", + "right": "pull_request:6591", + "accept": false, + "reason": "Different doc cleanup efforts with no shared code-path." + }, + { + "left": "pull_request:4339", + "right": "pull_request:5937", + "accept": false, + "reason": "A guidance-scale default typo in one example is unrelated to Kandinsky 3 docs." + }, + { + "left": "pull_request:4347", + "right": "pull_request:6053", + "accept": false, + "reason": "Scheduler step-index work versus training loss logging are unrelated changes." + }, + { + "left": "pull_request:4395", + "right": "pull_request:6078", + "accept": false, + "reason": "DreamBooth class-prompt encoding fix and Euler sigma fix are separate code paths." + }, + { + "left": "pull_request:4437", + "right": "pull_request:5768", + "accept": false, + "reason": "Consistency-model sampling bugs vs consistency decoder documentation; not the same change." + }, + { + "left": "pull_request:4495", + "right": "pull_request:6360", + "accept": false, + "reason": "A community SDXL reference example and inpaint padding-mask support are different features." + }, + { + "left": "pull_request:4495", + "right": "pull_request:6800", + "accept": false, + "reason": "Example pipeline work versus import-utility availability changes; unrelated." + }, + { + "left": "pull_request:4563", + "right": "pull_request:5707", + "accept": false, + "reason": "Prompt2Prompt feature addition and LCM test additions do not fix the same issue." + }, + { + "left": "pull_request:4563", + "right": "pull_request:6053", + "accept": false, + "reason": "Prompt2Prompt pipeline work versus training loss logging are unrelated." + }, + { + "left": "pull_request:4581", + "right": "pull_request:5552", + "accept": false, + "reason": "UniPC sigma handling fix and FreeU disable-method fix are unrelated code paths." + }, + { + "left": "pull_request:4596", + "right": "pull_request:5456", + "accept": false, + "reason": "SDXL checkpoint conversion script and RealFill training docs/code are different features." + }, + { + "left": "pull_request:46", + "right": "pull_request:96", + "accept": false, + "reason": "ResNet refactor and latent-diffusion finalization are separate architectural changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 230, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:4632", + "pull_request:4635", + "pull_request:4655", + "pull_request:4666", + "pull_request:4700", + "pull_request:4706", + "pull_request:4721", + "pull_request:4741", + "pull_request:4746", + "pull_request:4748", + "pull_request:4761", + "pull_request:4793", + "pull_request:4807", + "pull_request:4822", + "pull_request:4847", + "pull_request:4862", + "pull_request:4881", + "pull_request:4886", + "pull_request:4934", + "pull_request:4955", + "pull_request:4973", + "pull_request:498", + "pull_request:4984", + "pull_request:5", + "pull_request:5011", + "pull_request:5098", + "pull_request:5307", + "pull_request:5499", + "pull_request:5506", + "pull_request:5682", + "pull_request:5698", + "pull_request:5699", + "pull_request:5754", + "pull_request:5853", + "pull_request:5922", + "pull_request:5937", + "pull_request:5982", + "pull_request:6017", + "pull_request:6049", + "pull_request:6389", + "pull_request:6466", + "pull_request:6490", + "pull_request:6591", + "pull_request:6800", + "pull_request:7", + "pull_request:7153", + "pull_request:75" + ], + "soft_pairs": [ + "pull_request:4632|pull_request:5922", + "pull_request:4635|pull_request:5982", + "pull_request:4655|pull_request:5506", + "pull_request:4666|pull_request:5853", + "pull_request:4700|pull_request:5499", + "pull_request:4706|pull_request:4862", + "pull_request:4721|pull_request:6591", + "pull_request:4741|pull_request:6049", + "pull_request:4746|pull_request:5682", + "pull_request:4748|pull_request:5307", + "pull_request:4761|pull_request:5098", + "pull_request:4793|pull_request:5506", + "pull_request:4793|pull_request:5754", + "pull_request:4807|pull_request:5937", + "pull_request:4807|pull_request:6017", + "pull_request:4807|pull_request:6466", + "pull_request:4822|pull_request:5699", + "pull_request:4822|pull_request:6389", + "pull_request:4822|pull_request:6591", + "pull_request:4847|pull_request:6490", + "pull_request:4881|pull_request:4934", + "pull_request:4886|pull_request:7153", + "pull_request:4955|pull_request:5698", + "pull_request:4973|pull_request:6800", + "pull_request:498|pull_request:75", + "pull_request:4984|pull_request:5698", + "pull_request:5|pull_request:7", + "pull_request:5011|pull_request:5098" + ], + "prepared_review_unit_hash": "c7e6407593ac1aa585f33053100983e559c04b27cb36da4562e3ac4d57f33b3a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 97249, + "estimated_input_tokens": 24313, + "estimated_eval_tokens": 48882 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab-bag of unrelated PRs across docs, examples, workflows, and distinct pipeline/model bugs. The soft similarities are mostly lexical (e.g. SDXL, ControlNet, LoRA) rather than the same concrete fix, so none of the candidate edges should be merged.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR represents a common underlying change; the set is heterogeneous and spans unrelated fixes, docs, and tests.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a good global anchor because there is no shared bug/change theme across the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4632", + "right": "pull_request:5922", + "accept": false, + "reason": "Different problems: SDXL LoRA checkpointing/casting vs MultiAdapter inpainting handling." + }, + { + "left": "pull_request:4635", + "right": "pull_request:5982", + "accept": false, + "reason": "Unrelated docs additions: MultiControlNet guide vs SDXL Turbo docs." + }, + { + "left": "pull_request:4655", + "right": "pull_request:5506", + "accept": false, + "reason": "Different code paths: load_dataset data_dir plumbing vs PEFT/xformers attention scaling." + }, + { + "left": "pull_request:4666", + "right": "pull_request:5853", + "accept": false, + "reason": "SDXL ControlNet LoRA support vs adding doctest CI workflows; not the same change." + }, + { + "left": "pull_request:4700", + "right": "pull_request:5499", + "accept": false, + "reason": "Different documentation topics: Shap-E guide vs general link fixes across many docs." + }, + { + "left": "pull_request:4706", + "right": "pull_request:4862", + "accept": false, + "reason": "Different bugs: missing LoRA alpha defaults vs MultiControlNet img2img size handling." + }, + { + "left": "pull_request:4721", + "right": "pull_request:6591", + "accept": false, + "reason": "A Korean docs typo fix is unrelated to a broad typo cleanup across examples and pipelines." + }, + { + "left": "pull_request:4741", + "right": "pull_request:6049", + "accept": false, + "reason": "AudioLDM2 dummy import fix vs DemoFusion pipeline/README changes; unrelated." + }, + { + "left": "pull_request:4746", + "right": "pull_request:5682", + "accept": false, + "reason": "WIP SDXL inpainting work vs AnimateDiff checks; different features entirely." + }, + { + "left": "pull_request:4748", + "right": "pull_request:5307", + "accept": false, + "reason": "Packaging/setup instructions vs an unrelated new shell.py script." + }, + { + "left": "pull_request:4761", + "right": "pull_request:5098", + "accept": false, + "reason": "Paint-by-example proj_size fix vs Kandinsky precision test updates; different code paths." + }, + { + "left": "pull_request:4793", + "right": "pull_request:5506", + "accept": false, + "reason": "Auto-pipeline config kwargs vs attention processor LoRA/xformers behavior." + }, + { + "left": "pull_request:4793", + "right": "pull_request:5754", + "accept": false, + "reason": "Auto-pipeline loading fix vs formatting/data_dir changes in training scripts." + }, + { + "left": "pull_request:4807", + "right": "pull_request:5937", + "accept": false, + "reason": "Chinese docs additions vs Kandinsky3 docs; unrelated documentation work." + }, + { + "left": "pull_request:4807", + "right": "pull_request:6017", + "accept": false, + "reason": "Chinese docs vs a doc-builder test WIP; not the same underlying change." + }, + { + "left": "pull_request:4807", + "right": "pull_request:6466", + "accept": false, + "reason": "Chinese docs vs SVD doc cleanup; different pages and purposes." + }, + { + "left": "pull_request:4822", + "right": "pull_request:5699", + "accept": false, + "reason": "LoRA warning text typo vs stable_unclip styling refactors; unrelated." + }, + { + "left": "pull_request:4822", + "right": "pull_request:6389", + "accept": false, + "reason": "LoRA warning text typo vs SVD parameter type fix; different issue." + }, + { + "left": "pull_request:4822", + "right": "pull_request:6591", + "accept": false, + "reason": "Single warning-string typo vs broad typo cleanup across many files." + }, + { + "left": "pull_request:4847", + "right": "pull_request:6490", + "accept": false, + "reason": "Prompt Free Diffusion community example vs RAVE video editing community example; different additions." + }, + { + "left": "pull_request:4881", + "right": "pull_request:4934", + "accept": false, + "reason": "SDXL pix2pix VAE precision option vs T2I-Adapter training script; not the same feature." + }, + { + "left": "pull_request:4886", + "right": "pull_request:7153", + "accept": false, + "reason": "Upscale safety checker bug vs documentation workflow path filtering." + }, + { + "left": "pull_request:4955", + "right": "pull_request:5698", + "accept": false, + "reason": "Argument type fix in pix2pix script vs LCM img2img fix; unrelated." + }, + { + "left": "pull_request:4973", + "right": "pull_request:6800", + "accept": false, + "reason": "README badge update vs adding is_torchvision_available utility; different concerns." + }, + { + "left": "pull_request:498", + "right": "pull_request:75", + "accept": false, + "reason": "Attention-slicing typo fix vs a much older Glide attention implementation change." + }, + { + "left": "pull_request:4984", + "right": "pull_request:5698", + "accept": false, + "reason": "PR template editor names vs LCM img2img bug fix; no shared code path." + }, + { + "left": "pull_request:5", + "right": "pull_request:7", + "accept": false, + "reason": "Foundational latent UNet addition vs VQVAE addition; separate model components, not duplicates." + }, + { + "left": "pull_request:5011", + "right": "pull_request:5098", + "accept": false, + "reason": "Flax/PT class-name handling vs Kandinsky precision tests; unrelated changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 231, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5021", + "pull_request:5030", + "pull_request:5038", + "pull_request:5056", + "pull_request:5058", + "pull_request:5063", + "pull_request:51", + "pull_request:5111", + "pull_request:5122", + "pull_request:5185", + "pull_request:5190", + "pull_request:5201", + "pull_request:5250", + "pull_request:5267", + "pull_request:5286", + "pull_request:5289", + "pull_request:5304", + "pull_request:5307", + "pull_request:5312", + "pull_request:5331", + "pull_request:5391", + "pull_request:5424", + "pull_request:5454", + "pull_request:5506", + "pull_request:570", + "pull_request:5729", + "pull_request:5780", + "pull_request:5823", + "pull_request:5833", + "pull_request:5834", + "pull_request:5910", + "pull_request:6085", + "pull_request:6236", + "pull_request:6364", + "pull_request:6658", + "pull_request:6754", + "pull_request:6765", + "pull_request:6766", + "pull_request:6800", + "pull_request:6861", + "pull_request:6875", + "pull_request:7062", + "pull_request:7161", + "pull_request:7166", + "pull_request:7272", + "pull_request:7307", + "pull_request:7328", + "pull_request:7339" + ], + "soft_pairs": [ + "pull_request:5021|pull_request:6364", + "pull_request:5030|pull_request:7272", + "pull_request:5038|pull_request:5506", + "pull_request:5056|pull_request:5111", + "pull_request:5058|pull_request:7161", + "pull_request:5063|pull_request:5190", + "pull_request:51|pull_request:570", + "pull_request:5111|pull_request:7328", + "pull_request:5122|pull_request:5833", + "pull_request:5122|pull_request:5834", + "pull_request:5122|pull_request:7339", + "pull_request:5185|pull_request:5823", + "pull_request:5201|pull_request:5331", + "pull_request:5201|pull_request:6085", + "pull_request:5250|pull_request:5910", + "pull_request:5250|pull_request:6800", + "pull_request:5267|pull_request:6861", + "pull_request:5286|pull_request:6800", + "pull_request:5289|pull_request:6875", + "pull_request:5304|pull_request:5729", + "pull_request:5307|pull_request:5780", + "pull_request:5307|pull_request:6754", + "pull_request:5307|pull_request:6765", + "pull_request:5307|pull_request:6766", + "pull_request:5307|pull_request:7307", + "pull_request:5312|pull_request:6236", + "pull_request:5331|pull_request:7166", + "pull_request:5391|pull_request:6658", + "pull_request:5424|pull_request:7062", + "pull_request:5454|pull_request:6861" + ], + "prepared_review_unit_hash": "c613bef807317c672b6b4907f8489843eda73907c1cf0f09b57600897207d53b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 97239, + "estimated_input_tokens": 24310, + "estimated_eval_tokens": 48876 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are false positives spanning unrelated docs, workflow, test, and LoRA/IP-Adapter changes. The only clear duplicate-like pair is 6765/6766, which both patch HF cache path construction for the same underlying path-handling issue.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "Among the accepted duplicate-like pair, 6765 is the more central change: it fixes the main cache-path logic in hub_utils, while 6766 is the companion constants adjustment.", + "best_issue_reason": null, + "best_pr_reason": "6765 is the best representative PR for this cluster because it contains the core cache-path fix and matches the only strong duplicate pair.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5021", + "right": "pull_request:6364", + "accept": false, + "reason": "Both are docs, but they cover different topics: scheduler mapping vs ControlNet inpaint overlay guidance." + }, + { + "left": "pull_request:5030", + "right": "pull_request:7272", + "accept": false, + "reason": "Different bugs in different loader paths; not the same concrete fix." + }, + { + "left": "pull_request:5038", + "right": "pull_request:5506", + "accept": false, + "reason": "Both touch PEFT/LoRA code, but one makes arguments explicit and the other fixes xformers-enabled LoRA calls." + }, + { + "left": "pull_request:5056", + "right": "pull_request:5111", + "accept": false, + "reason": "Unrelated changes: LoRA docs vs Karras sigma scheduler support." + }, + { + "left": "pull_request:5058", + "right": "pull_request:7161", + "accept": false, + "reason": "Both are PEFT docs, but they describe different workflows and fixes." + }, + { + "left": "pull_request:5063", + "right": "pull_request:5190", + "accept": false, + "reason": "Both edit docstrings, but they target different functions and files with unrelated content." + }, + { + "left": "pull_request:51", + "right": "pull_request:570", + "accept": false, + "reason": "Completely unrelated: packaging metadata vs BaseOutput initialization behavior." + }, + { + "left": "pull_request:5111", + "right": "pull_request:7328", + "accept": false, + "reason": "Scheduler feature work vs release workflow dependency changes; different problems." + }, + { + "left": "pull_request:5122", + "right": "pull_request:5833", + "accept": false, + "reason": "Workflow/Python version changes are unrelated to the PixArt docs change shown here." + }, + { + "left": "pull_request:5122", + "right": "pull_request:5834", + "accept": false, + "reason": "Workflow setup updates vs a single-file import fix; no shared code-path bug." + }, + { + "left": "pull_request:5122", + "right": "pull_request:7339", + "accept": false, + "reason": "Different release/workflow dependency changes with distinct goals." + }, + { + "left": "pull_request:5185", + "right": "pull_request:5823", + "accept": false, + "reason": "Both are PEFT-related, but one is a CI fix and the other adds adapter_names to fuse_lora." + }, + { + "left": "pull_request:5201", + "right": "pull_request:5331", + "accept": false, + "reason": "Shared LoRA vocabulary, but one fixes xformers processor removal and the other updates example training code." + }, + { + "left": "pull_request:5201", + "right": "pull_request:6085", + "accept": false, + "reason": "Different code paths: LoRA/xformers processor handling vs scheduler add_noise logic." + }, + { + "left": "pull_request:5250", + "right": "pull_request:5910", + "accept": false, + "reason": "Docs for DDPO vs IPAdapterFull loader support; unrelated changes." + }, + { + "left": "pull_request:5250", + "right": "pull_request:6800", + "accept": false, + "reason": "DDPO docs and adding torchvision availability are unrelated." + }, + { + "left": "pull_request:5267", + "right": "pull_request:6861", + "accept": false, + "reason": "README formatting cleanup vs a large SDXL pipeline update; no duplicate bug/change." + }, + { + "left": "pull_request:5286", + "right": "pull_request:6800", + "accept": false, + "reason": "Japanese documentation addition vs import-utility availability helper; unrelated." + }, + { + "left": "pull_request:5289", + "right": "pull_request:6875", + "accept": false, + "reason": "Both adjust test expectations, but they concern different pipelines and different failures." + }, + { + "left": "pull_request:5304", + "right": "pull_request:5729", + "accept": false, + "reason": "DeepFloyd IF docstrings vs example min-version updates; no same underlying change." + }, + { + "left": "pull_request:5307", + "right": "pull_request:5780", + "accept": false, + "reason": "Completely unrelated new file vs checkpoint_merger doc cleanup." + }, + { + "left": "pull_request:5307", + "right": "pull_request:6754", + "accept": false, + "reason": "Unrelated: shell sort demo file vs DPO training loss fix." + }, + { + "left": "pull_request:5307", + "right": "pull_request:6765", + "accept": false, + "reason": "No relation: new shell.py demo vs cache path handling." + }, + { + "left": "pull_request:5307", + "right": "pull_request:6766", + "accept": false, + "reason": "No relation: new shell.py demo vs constants cache path handling." + }, + { + "left": "pull_request:5307", + "right": "pull_request:7307", + "accept": false, + "reason": "Unrelated demo script vs TensorRT/example environment update." + }, + { + "left": "pull_request:5312", + "right": "pull_request:6236", + "accept": false, + "reason": "Both are docs-ish, but one improves model docs/typehints and the other fixes a ControlNet example." + }, + { + "left": "pull_request:5331", + "right": "pull_request:7166", + "accept": false, + "reason": "Both mention LoRA/dreambooth, but they fix different issues in different scripts." + }, + { + "left": "pull_request:5391", + "right": "pull_request:6658", + "accept": false, + "reason": "General model docs cleanup vs introducing a memory-footprint utility; unrelated." + }, + { + "left": "pull_request:5424", + "right": "pull_request:7062", + "accept": false, + "reason": "Slow-test pipeline fetcher vs a community pipeline docstring fix." + }, + { + "left": "pull_request:5454", + "right": "pull_request:6861", + "accept": false, + "reason": "Scheduler docs link tweak vs SDXL pipeline feature update; unrelated." + }, + { + "left": "pull_request:6765", + "right": "pull_request:6766", + "accept": true, + "reason": "Both address the same cache-path construction issue around HF cache locations and are natural companions under issue 6585." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 232, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5478", + "pull_request:5499", + "pull_request:5533", + "pull_request:5568", + "pull_request:56", + "pull_request:5614", + "pull_request:5622", + "pull_request:5626", + "pull_request:5682", + "pull_request:5707", + "pull_request:5723", + "pull_request:5727", + "pull_request:5750", + "pull_request:5780", + "pull_request:5788", + "pull_request:5818", + "pull_request:5831", + "pull_request:5842", + "pull_request:5844", + "pull_request:5855", + "pull_request:5858", + "pull_request:5895", + "pull_request:5910", + "pull_request:5911", + "pull_request:5920", + "pull_request:5961", + "pull_request:5974", + "pull_request:5982", + "pull_request:6051", + "pull_request:6065", + "pull_request:6192", + "pull_request:6241", + "pull_request:6276", + "pull_request:6326", + "pull_request:6339", + "pull_request:6498", + "pull_request:6643", + "pull_request:6656", + "pull_request:6800", + "pull_request:6861", + "pull_request:6941", + "pull_request:7062", + "pull_request:7077", + "pull_request:7307", + "pull_request:7324", + "pull_request:774", + "pull_request:7767", + "pull_request:909" + ], + "soft_pairs": [ + "pull_request:5478|pull_request:5831", + "pull_request:5499|pull_request:6065", + "pull_request:5533|pull_request:5707", + "pull_request:5533|pull_request:5818", + "pull_request:5568|pull_request:5961", + "pull_request:56|pull_request:774", + "pull_request:56|pull_request:909", + "pull_request:5614|pull_request:5858", + "pull_request:5622|pull_request:6326", + "pull_request:5622|pull_request:7077", + "pull_request:5626|pull_request:7767", + "pull_request:5682|pull_request:6339", + "pull_request:5707|pull_request:7324", + "pull_request:5723|pull_request:6051", + "pull_request:5723|pull_request:6498", + "pull_request:5723|pull_request:6800", + "pull_request:5727|pull_request:5911", + "pull_request:5727|pull_request:6192", + "pull_request:5727|pull_request:6656", + "pull_request:5750|pull_request:5895", + "pull_request:5780|pull_request:6861", + "pull_request:5780|pull_request:7307", + "pull_request:5788|pull_request:5974", + "pull_request:5842|pull_request:6241", + "pull_request:5842|pull_request:7062", + "pull_request:5844|pull_request:6643", + "pull_request:5855|pull_request:6861", + "pull_request:5895|pull_request:5982", + "pull_request:5910|pull_request:6941", + "pull_request:5911|pull_request:6192", + "pull_request:5911|pull_request:6656", + "pull_request:5920|pull_request:6276" + ], + "prepared_review_unit_hash": "16178bd8f469dbfb1fe5d22d95a63315a65030ed0a0748525251bc56515d8538", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 100495, + "estimated_input_tokens": 25124, + "estimated_eval_tokens": 50504 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: docs cleanups, test adjustments, new model/pipeline features, and unrelated loader/scheduler fixes. None of the soft edges look like true duplicates or mergeable into one PR, so I reject them all.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents a duplicate cluster here; the items span unrelated docs, tests, and feature work rather than one underlying change.", + "best_issue_reason": null, + "best_pr_reason": "PR 5911 is the most substantive, self-contained implementation change with a clear issue target, but even it does not unify the rest of the set into one duplicate family.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5478", + "right": "pull_request:5831", + "accept": false, + "reason": "Japanese docs versus a small title-only docs fix; different changes and no shared bug." + }, + { + "left": "pull_request:5499", + "right": "pull_request:6065", + "accept": false, + "reason": "Broad docs link cleanup versus SDXL Turbo docs addition; same area, different change." + }, + { + "left": "pull_request:5533", + "right": "pull_request:5707", + "accept": false, + "reason": "Both are test-related, but one speeds up SDXL expert tests and the other adds LCM tests; unrelated code paths." + }, + { + "left": "pull_request:5533", + "right": "pull_request:5818", + "accept": false, + "reason": "Speeding up expert tests and restructuring PR test workflow are different testing infrastructure changes." + }, + { + "left": "pull_request:5568", + "right": "pull_request:5961", + "accept": false, + "reason": "Custom pipeline loading fix versus a PixArt docstring typo; not the same bug or change." + }, + { + "left": "pull_request:56", + "right": "pull_request:774", + "accept": false, + "reason": "Both touch tests, but they fix different assertions in different eras and files; not the same issue." + }, + { + "left": "pull_request:56", + "right": "pull_request:909", + "accept": false, + "reason": "General test tolerance tweaks versus inpaint test model/path updates; unrelated fixes." + }, + { + "left": "pull_request:5614", + "right": "pull_request:5858", + "accept": false, + "reason": "Both are revert-style docs/code edits, but they revert different changes in different places." + }, + { + "left": "pull_request:5622", + "right": "pull_request:6326", + "accept": false, + "reason": "Example script optimization/config work versus a guidance_scale setter fix in a community TensorRT pipeline." + }, + { + "left": "pull_request:5622", + "right": "pull_request:7077", + "accept": false, + "reason": "Different test/optimization work versus a specific attention processor fix; no shared underlying bug." + }, + { + "left": "pull_request:5626", + "right": "pull_request:7767", + "accept": false, + "reason": "CPU offload test fix versus PixArt Sigma nits; unrelated test and pipeline changes." + }, + { + "left": "pull_request:5682", + "right": "pull_request:6339", + "accept": false, + "reason": "AnimateDiff checks implementation versus AnimateDiff docs example correction; one is feature/checks, the other documentation." + }, + { + "left": "pull_request:5707", + "right": "pull_request:7324", + "accept": false, + "reason": "LCM test additions versus Stable Cascade test updates; different models and failures." + }, + { + "left": "pull_request:5723", + "right": "pull_request:6051", + "accept": false, + "reason": "LCM documentation and Flax LCM implementation are related broadly, but they are not the same concrete change." + }, + { + "left": "pull_request:5723", + "right": "pull_request:6498", + "accept": false, + "reason": "LCM docs versus inpaint mask blur docs; only both are docs, not the same change." + }, + { + "left": "pull_request:5723", + "right": "pull_request:6800", + "accept": false, + "reason": "LCM docs versus utility availability export; completely different concerns." + }, + { + "left": "pull_request:5727", + "right": "pull_request:5911", + "accept": false, + "reason": "LCM scripts versus IP-Adapter Full support; different features and code paths." + }, + { + "left": "pull_request:5727", + "right": "pull_request:6192", + "accept": false, + "reason": "LCM training scripts versus unloading IP-Adapter support; unrelated model/tooling changes." + }, + { + "left": "pull_request:5727", + "right": "pull_request:6656", + "accept": false, + "reason": "LCM scripts versus unload_textual_inversion; different loader features." + }, + { + "left": "pull_request:5750", + "right": "pull_request:5895", + "accept": false, + "reason": "Reference-only docs introduction versus a full Stable Video Diffusion feature PR; not the same underlying change." + }, + { + "left": "pull_request:5780", + "right": "pull_request:6861", + "accept": false, + "reason": "Checkpoint merger doc cleanup versus a ControlNet/SDXL pipeline update; unrelated." + }, + { + "left": "pull_request:5780", + "right": "pull_request:7307", + "accept": false, + "reason": "Checkpoint merger comment cleanup versus Tensorrt example update; different example codebases." + }, + { + "left": "pull_request:5788", + "right": "pull_request:5974", + "accept": false, + "reason": "LCM scheduler error handling versus PixArt-LCM pipeline adaptation; same model family only in name, but different concrete fixes." + }, + { + "left": "pull_request:5842", + "right": "pull_request:6241", + "accept": false, + "reason": "PixArt resolution bug fix versus a runtime dtype mismatch report with no diff shown; no evidence they are the same issue." + }, + { + "left": "pull_request:5842", + "right": "pull_request:7062", + "accept": false, + "reason": "PixArt resolution logic versus a community Imagic docstring fix; unrelated." + }, + { + "left": "pull_request:5844", + "right": "pull_request:6643", + "accept": false, + "reason": "Docs reference to Japanese SDXL versus UViT2D docs; different documentation additions." + }, + { + "left": "pull_request:5855", + "right": "pull_request:6861", + "accept": false, + "reason": "Community README typo fix versus SDXL ControlNet update; unrelated edits." + }, + { + "left": "pull_request:5895", + "right": "pull_request:5982", + "accept": false, + "reason": "Stable Video Diffusion feature PR versus SDXL Turbo docs; different products and deliverables." + }, + { + "left": "pull_request:5910", + "right": "pull_request:6941", + "accept": false, + "reason": "IPAdapterFull support and SDXL ControlNet IP-Adapter support share a subsystem, but they fix different pipelines and are not one mergeable change." + }, + { + "left": "pull_request:5911", + "right": "pull_request:6192", + "accept": false, + "reason": "Adding IPAdapterFull support versus unloading IP-Adapter support; opposite lifecycle operations, not the same bug." + }, + { + "left": "pull_request:5911", + "right": "pull_request:6656", + "accept": false, + "reason": "IP-Adapter support versus textual inversion unload support; different loader mechanisms." + }, + { + "left": "pull_request:5920", + "right": "pull_request:6276", + "accept": false, + "reason": "IPEX LoRA support versus IPAdapter FaceID community pipeline; different features and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 233, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:5983", + "pull_request:6", + "pull_request:6002", + "pull_request:62", + "pull_request:6236", + "pull_request:6262", + "pull_request:6263", + "pull_request:6264", + "pull_request:6296", + "pull_request:6308", + "pull_request:6322", + "pull_request:6326", + "pull_request:6332", + "pull_request:6337", + "pull_request:6339", + "pull_request:6344", + "pull_request:6373", + "pull_request:6389", + "pull_request:6410", + "pull_request:6433", + "pull_request:6453", + "pull_request:6462", + "pull_request:6487", + "pull_request:6498", + "pull_request:6564", + "pull_request:662", + "pull_request:6759", + "pull_request:6800", + "pull_request:6910", + "pull_request:7", + "pull_request:7077", + "pull_request:7087", + "pull_request:7510", + "pull_request:7653", + "pull_request:7720", + "pull_request:78", + "pull_request:8268", + "pull_request:8272", + "pull_request:8419", + "pull_request:8425", + "pull_request:8426", + "pull_request:8481", + "pull_request:8483", + "pull_request:8491", + "pull_request:8654", + "pull_request:8656", + "pull_request:8698", + "pull_request:96" + ], + "soft_pairs": [ + "pull_request:5983|pull_request:7510", + "pull_request:6|pull_request:7", + "pull_request:6|pull_request:96", + "pull_request:6002|pull_request:7653", + "pull_request:6002|pull_request:8272", + "pull_request:62|pull_request:662", + "pull_request:62|pull_request:96", + "pull_request:6236|pull_request:6910", + "pull_request:6262|pull_request:8419", + "pull_request:6263|pull_request:8419", + "pull_request:6264|pull_request:8419", + "pull_request:6296|pull_request:6344", + "pull_request:6296|pull_request:6462", + "pull_request:6308|pull_request:8425", + "pull_request:6322|pull_request:7510", + "pull_request:6322|pull_request:8481", + "pull_request:6326|pull_request:6410", + "pull_request:6332|pull_request:6759", + "pull_request:6337|pull_request:7510", + "pull_request:6337|pull_request:8481", + "pull_request:6339|pull_request:7510", + "pull_request:6339|pull_request:8481", + "pull_request:6373|pull_request:8698", + "pull_request:6389|pull_request:7087", + "pull_request:6410|pull_request:7077", + "pull_request:6410|pull_request:8426", + "pull_request:6410|pull_request:8491", + "pull_request:6410|pull_request:8656", + "pull_request:6433|pull_request:6487", + "pull_request:6453|pull_request:6800", + "pull_request:6462|pull_request:7653", + "pull_request:6487|pull_request:8483", + "pull_request:6498|pull_request:8268", + "pull_request:6498|pull_request:8654", + "pull_request:6564|pull_request:7720", + "pull_request:662|pull_request:78" + ], + "prepared_review_unit_hash": "98e23c0f4cda5170230f61e898d3743245ed11250dcde2528ff98815c269a47e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 36, + "serialized_chars": 102149, + "estimated_input_tokens": 25538, + "estimated_eval_tokens": 51332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of mostly unrelated PRs across docs, workflow cleanup, model imports, and feature work. There are a few same-subsystem follow-ups, but none look like true duplicates of the same concrete change.", + "confidence": 0.77, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 8483 is the broadest and most central artifact here: it adds Stable Diffusion 3 and drives several downstream follow-up fixes/docs around the same model family.", + "best_issue_reason": null, + "best_pr_reason": "PR 8483 is the strongest representative PR in the set because it is a major feature addition with substantial code and docs impact, unlike the smaller one-off fixes and docs tweaks.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5983", + "right": "pull_request:7510", + "accept": false, + "reason": "Different bugs: SVD docs cleanup vs a VAE tiling OOM test fix." + }, + { + "left": "pull_request:6", + "right": "pull_request:7", + "accept": false, + "reason": "Both are import/init updates, but they expose different models and solve different changes." + }, + { + "left": "pull_request:6", + "right": "pull_request:96", + "accept": false, + "reason": "Different model-setup changes in the early LDM stack; not the same code change." + }, + { + "left": "pull_request:6002", + "right": "pull_request:7653", + "accept": false, + "reason": "SVD guidance-scale logic vs a configuration decorator fix; unrelated code paths." + }, + { + "left": "pull_request:6002", + "right": "pull_request:8272", + "accept": false, + "reason": "Different fixes: guidance-scale handling versus grammar cleanup in error strings." + }, + { + "left": "pull_request:62", + "right": "pull_request:662", + "accept": false, + "reason": "Early latent-diffusion pipeline fix vs a separate 6GB textual-inversion/docs change." + }, + { + "left": "pull_request:62", + "right": "pull_request:96", + "accept": false, + "reason": "Different latent-diffusion refactors; one fixes unconditional decode behavior, the other finalizes model wiring." + }, + { + "left": "pull_request:6236", + "right": "pull_request:6910", + "accept": false, + "reason": "Both mention ControlNet, but one is an inpainting doc example and the other is model-card generation in training code." + }, + { + "left": "pull_request:6262", + "right": "pull_request:8419", + "accept": false, + "reason": "Panorama refactor versus dynamic module download logic; unrelated changes." + }, + { + "left": "pull_request:6263", + "right": "pull_request:8419", + "accept": false, + "reason": "LDM3D refactor versus dynamic module download logic; unrelated changes." + }, + { + "left": "pull_request:6264", + "right": "pull_request:8419", + "accept": false, + "reason": "SAG refactor versus dynamic module download logic; unrelated changes." + }, + { + "left": "pull_request:6296", + "right": "pull_request:6344", + "accept": false, + "reason": "Same feature area (aMUSEd), but 6344 is only a later link-update follow-up, not the same change." + }, + { + "left": "pull_request:6296", + "right": "pull_request:6462", + "accept": false, + "reason": "Both touch aMUSEd docs, but one adds the pipeline and the other only updates the paper link." + }, + { + "left": "pull_request:6308", + "right": "pull_request:8425", + "accept": false, + "reason": "LoRA dummy-component test fix versus a workflow filename/name tweak; unrelated." + }, + { + "left": "pull_request:6322", + "right": "pull_request:7510", + "accept": false, + "reason": "Training-script scheduler change versus VAE tiling OOM test fix; different issues." + }, + { + "left": "pull_request:6322", + "right": "pull_request:8481", + "accept": false, + "reason": "Training-script fix versus image-processor error-message fix; unrelated." + }, + { + "left": "pull_request:6326", + "right": "pull_request:6410", + "accept": false, + "reason": "TensorRT guidance_scale bug fix versus module-registration refactor in experimental RL code." + }, + { + "left": "pull_request:6332", + "right": "pull_request:6759", + "accept": false, + "reason": "Community video-rerender pipeline versus an InstantID batching fix; different features." + }, + { + "left": "pull_request:6337", + "right": "pull_request:7510", + "accept": false, + "reason": "retrieve_latents bug fix versus a VAE tiling OOM test fix; unrelated." + }, + { + "left": "pull_request:6337", + "right": "pull_request:8481", + "accept": false, + "reason": "Different code paths: latent retrieval in TensorRT img2img versus image input validation messaging." + }, + { + "left": "pull_request:6339", + "right": "pull_request:7510", + "accept": false, + "reason": "AnimateDiff docs update versus VAE tiling OOM test fix; unrelated." + }, + { + "left": "pull_request:6339", + "right": "pull_request:8481", + "accept": false, + "reason": "AnimateDiff docs versus image-processor validation messaging; unrelated." + }, + { + "left": "pull_request:6373", + "right": "pull_request:8698", + "accept": false, + "reason": "Workflow deletion versus transformer output deprecation cleanup; unrelated." + }, + { + "left": "pull_request:6389", + "right": "pull_request:7087", + "accept": false, + "reason": "Both are SVD-related, but one changes a parameter type and the other adds docs/example content." + }, + { + "left": "pull_request:6410", + "right": "pull_request:7077", + "accept": false, + "reason": "Experimental RL module registration versus IP-Adapter attention reshaping bug; different problems." + }, + { + "left": "pull_request:6410", + "right": "pull_request:8426", + "accept": false, + "reason": "Experimental RL cleanup versus workflow dependency changes; unrelated." + }, + { + "left": "pull_request:6410", + "right": "pull_request:8491", + "accept": false, + "reason": "Experimental RL module registration versus DualTransformer2D import-path fix; unrelated." + }, + { + "left": "pull_request:6410", + "right": "pull_request:8656", + "accept": false, + "reason": "Experimental RL module registration versus shard-checkpoint loading fixes; unrelated." + }, + { + "left": "pull_request:6433", + "right": "pull_request:6487", + "accept": false, + "reason": "Community gluegen pipeline versus Stable Cascade/Wuerstchen release work; unrelated." + }, + { + "left": "pull_request:6453", + "right": "pull_request:6800", + "accept": false, + "reason": "aMUSEd paper-link docs update versus adding a torchvision availability helper; unrelated." + }, + { + "left": "pull_request:6462", + "right": "pull_request:7653", + "accept": false, + "reason": "aMUSEd docs link correction versus config initializer decorator fix; unrelated." + }, + { + "left": "pull_request:6487", + "right": "pull_request:8483", + "accept": false, + "reason": "Both are major model-release PRs, but they implement different model families and code paths." + }, + { + "left": "pull_request:6498", + "right": "pull_request:8268", + "accept": false, + "reason": "Inpainting docs about mask blur/padding versus from_single_file documentation cleanup; unrelated docs changes." + }, + { + "left": "pull_request:6498", + "right": "pull_request:8654", + "accept": false, + "reason": "Different docs topics: inpainting mask handling versus SD3 token-limit guidance." + }, + { + "left": "pull_request:6564", + "right": "pull_request:7720", + "accept": false, + "reason": "IP-Adapter path handling versus dead-code/f-string cleanup; unrelated." + }, + { + "left": "pull_request:662", + "right": "pull_request:78", + "accept": false, + "reason": "Textual inversion 6GB fix versus a mid-block architecture fix; different areas and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 234, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:662", + "pull_request:6645", + "pull_request:6754", + "pull_request:6759", + "pull_request:6765", + "pull_request:678", + "pull_request:6800", + "pull_request:6837", + "pull_request:688", + "pull_request:689", + "pull_request:6904", + "pull_request:6929", + "pull_request:693", + "pull_request:6989", + "pull_request:6998", + "pull_request:7", + "pull_request:7024", + "pull_request:7034", + "pull_request:705", + "pull_request:7077", + "pull_request:7113", + "pull_request:7134", + "pull_request:7156", + "pull_request:7259", + "pull_request:7307", + "pull_request:74", + "pull_request:75", + "pull_request:7618", + "pull_request:7709", + "pull_request:79", + "pull_request:802", + "pull_request:822", + "pull_request:824", + "pull_request:8378", + "pull_request:8452", + "pull_request:8483", + "pull_request:8607", + "pull_request:8684", + "pull_request:8688", + "pull_request:8705", + "pull_request:8836", + "pull_request:8920", + "pull_request:8990", + "pull_request:9065", + "pull_request:921", + "pull_request:978", + "pull_request:979" + ], + "soft_pairs": [ + "pull_request:662|pull_request:802", + "pull_request:6645|pull_request:8378", + "pull_request:6754|pull_request:7307", + "pull_request:6759|pull_request:6929", + "pull_request:6765|pull_request:7307", + "pull_request:678|pull_request:921", + "pull_request:6800|pull_request:6989", + "pull_request:6800|pull_request:7259", + "pull_request:6800|pull_request:7709", + "pull_request:6800|pull_request:8483", + "pull_request:6800|pull_request:8705", + "pull_request:6837|pull_request:8607", + "pull_request:688|pull_request:822", + "pull_request:689|pull_request:978", + "pull_request:6904|pull_request:8688", + "pull_request:693|pull_request:75", + "pull_request:6989|pull_request:8990", + "pull_request:6998|pull_request:8920", + "pull_request:7|pull_request:74", + "pull_request:7|pull_request:79", + "pull_request:7|pull_request:979", + "pull_request:7024|pull_request:8607", + "pull_request:7034|pull_request:8452", + "pull_request:705|pull_request:824", + "pull_request:7077|pull_request:8836", + "pull_request:7113|pull_request:9065", + "pull_request:7134|pull_request:8684", + "pull_request:7156|pull_request:7618", + "pull_request:7156|pull_request:8688" + ], + "prepared_review_unit_hash": "02cb5c579305005ee78e4fa98942a7624ce1448b99d13cab918d6baef49410a5", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 96408, + "estimated_input_tokens": 24102, + "estimated_eval_tokens": 48460 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a false positive: the PRs cover unrelated docs, CI, dependency, and code-fix changes. None of the soft-edge pairs look like the same underlying bug or change, so no merge candidate is safe.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:662", + "right": "pull_request:802", + "accept": false, + "reason": "Different docs/code changes in different areas; not the same bug or fix." + }, + { + "left": "pull_request:6645", + "right": "pull_request:8378", + "accept": false, + "reason": "Community README edit vs asymmetric autoencoder code fix; unrelated." + }, + { + "left": "pull_request:6754", + "right": "pull_request:7307", + "accept": false, + "reason": "DPO loss correction vs TensorRT example update; different problems." + }, + { + "left": "pull_request:6759", + "right": "pull_request:6929", + "accept": false, + "reason": "InstantID prompt image embedding repetition vs docs for community pipelines; unrelated." + }, + { + "left": "pull_request:6765", + "right": "pull_request:7307", + "accept": false, + "reason": "Cache-path utility fix vs TensorRT example/diffusers.yml update; not same change." + }, + { + "left": "pull_request:678", + "right": "pull_request:921", + "accept": false, + "reason": "DreamBooth notebook/model updates vs v1-5 docs refresh; far broader and different." + }, + { + "left": "pull_request:6800", + "right": "pull_request:6989", + "accept": false, + "reason": "Torchvision availability helper vs benchmark workflow dependency; unrelated." + }, + { + "left": "pull_request:6800", + "right": "pull_request:7259", + "accept": false, + "reason": "Torchvision helper addition vs TCD docs; no shared code-path bug." + }, + { + "left": "pull_request:6800", + "right": "pull_request:7709", + "accept": false, + "reason": "Import helper addition vs CI/Tailscale workflow; different subsystems." + }, + { + "left": "pull_request:6800", + "right": "pull_request:8483", + "accept": false, + "reason": "Torchvision helper addition vs SD3 feature work; not the same bug or mergeable fix." + }, + { + "left": "pull_request:6800", + "right": "pull_request:8705", + "accept": false, + "reason": "Import helper addition vs new benchmarking Docker container; unrelated." + }, + { + "left": "pull_request:6837", + "right": "pull_request:8607", + "accept": false, + "reason": "I2VGen docs import order vs SD3 TAESD docs; only both are docs." + }, + { + "left": "pull_request:688", + "right": "pull_request:822", + "accept": false, + "reason": "Flax/PyTorch import handling vs Flax scheduler API update; different underlying issues." + }, + { + "left": "pull_request:689", + "right": "pull_request:978", + "accept": false, + "reason": "Attention performance tweak vs typo/docs cleanup; unrelated." + }, + { + "left": "pull_request:6904", + "right": "pull_request:8688", + "accept": false, + "reason": "PIL-image support in load_image vs LoRA DoRA conversion logic; different code paths." + }, + { + "left": "pull_request:693", + "right": "pull_request:75", + "accept": false, + "reason": "BibTeX citation fix vs Glide/attention model plumbing; unrelated." + }, + { + "left": "pull_request:6989", + "right": "pull_request:8990", + "accept": false, + "reason": "Benchmark workflow dependency install vs new PAG pipeline implementation; different scope." + }, + { + "left": "pull_request:6998", + "right": "pull_request:8920", + "accept": false, + "reason": "Callout markdown fix vs PAG guide text edits; both docs but not the same change." + }, + { + "left": "pull_request:7", + "right": "pull_request:74", + "accept": false, + "reason": "Initial VQ-VAE addition vs Grad-TTS mid-block change; distinct model features." + }, + { + "left": "pull_request:7", + "right": "pull_request:79", + "accept": false, + "reason": "VQ-VAE addition vs unconditional image generation; only shares early repo history." + }, + { + "left": "pull_request:7", + "right": "pull_request:979", + "accept": false, + "reason": "VQ-VAE model addition vs missing import in README; unrelated." + }, + { + "left": "pull_request:7024", + "right": "pull_request:8607", + "accept": false, + "reason": "AnimateLCM docs vs SD3 TAESD docs; different topics." + }, + { + "left": "pull_request:7034", + "right": "pull_request:8452", + "accept": false, + "reason": "Both touch single_file_utils, but one fixes ControlNet dtype handling and the other CLIP load strictness; not same bug." + }, + { + "left": "pull_request:705", + "right": "pull_request:824", + "accept": false, + "reason": "CI CUDA workflow changes vs Flax scheduler/test updates; unrelated." + }, + { + "left": "pull_request:7077", + "right": "pull_request:8836", + "accept": false, + "reason": "IP-Adapter head_to_batch_dim fix vs SlicedAttnProcessor loop fix; different attention bugs." + }, + { + "left": "pull_request:7113", + "right": "pull_request:9065", + "accept": false, + "reason": "Docker Slack notification workflow vs sentencepiece optional dependency support; unrelated." + }, + { + "left": "pull_request:7134", + "right": "pull_request:8684", + "accept": false, + "reason": "UNet type hints vs Pillow import docs fix; no shared underlying bug." + }, + { + "left": "pull_request:7156", + "right": "pull_request:7618", + "accept": false, + "reason": "Stalebot label handling vs CPU offload hook cleanup; unrelated automation vs pipeline behavior." + }, + { + "left": "pull_request:7156", + "right": "pull_request:8688", + "accept": false, + "reason": "Stalebot logic vs LoRA DoRA conversion; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 235, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:718", + "pull_request:7212", + "pull_request:7213", + "pull_request:7259", + "pull_request:7339", + "pull_request:7458", + "pull_request:7464", + "pull_request:7472", + "pull_request:748", + "pull_request:7510", + "pull_request:7520", + "pull_request:7564", + "pull_request:7594", + "pull_request:7627", + "pull_request:7709", + "pull_request:774", + "pull_request:7778", + "pull_request:7790", + "pull_request:7791", + "pull_request:78", + "pull_request:7857", + "pull_request:7913", + "pull_request:7919", + "pull_request:7938", + "pull_request:7977", + "pull_request:800", + "pull_request:819", + "pull_request:8224", + "pull_request:8225", + "pull_request:824", + "pull_request:828", + "pull_request:8412", + "pull_request:8426", + "pull_request:8467", + "pull_request:8491", + "pull_request:8836", + "pull_request:8990", + "pull_request:9015", + "pull_request:9036", + "pull_request:904", + "pull_request:9116", + "pull_request:9251", + "pull_request:9463", + "pull_request:9649", + "pull_request:9684", + "pull_request:9721", + "pull_request:9769" + ], + "soft_pairs": [ + "pull_request:718|pull_request:78", + "pull_request:7212|pull_request:8836", + "pull_request:7213|pull_request:8412", + "pull_request:7259|pull_request:8990", + "pull_request:7339|pull_request:7778", + "pull_request:7458|pull_request:9251", + "pull_request:7464|pull_request:8426", + "pull_request:7464|pull_request:8491", + "pull_request:7472|pull_request:9036", + "pull_request:748|pull_request:819", + "pull_request:7510|pull_request:9036", + "pull_request:7510|pull_request:9116", + "pull_request:7520|pull_request:9463", + "pull_request:7564|pull_request:9036", + "pull_request:7564|pull_request:9116", + "pull_request:7594|pull_request:8224", + "pull_request:7627|pull_request:7913", + "pull_request:7709|pull_request:8990", + "pull_request:774|pull_request:78", + "pull_request:7790|pull_request:8467", + "pull_request:7791|pull_request:8836", + "pull_request:7857|pull_request:9649", + "pull_request:7919|pull_request:9684", + "pull_request:7919|pull_request:9769", + "pull_request:7938|pull_request:9721", + "pull_request:7977|pull_request:9036", + "pull_request:7977|pull_request:9116", + "pull_request:800|pull_request:904", + "pull_request:8225|pull_request:9015", + "pull_request:824|pull_request:828" + ], + "prepared_review_unit_hash": "78cb2a596b55b568694a77965810cc92e784316e0579f5df9f1e7ac009c6ce76", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 96503, + "estimated_input_tokens": 24126, + "estimated_eval_tokens": 48508 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes typo/docs fixes, workflow changes, test-only adjustments, and unrelated core model/pipeline feature work. I don\u2019t see a single underlying bug/change tying the items together, so there is no true duplicate canonical artifact here.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "If forced to pick one representative PR, 9769 is the most substantial and self-contained change set, with an explicit issue target and broad, coherent feature work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:718", + "right": "pull_request:78", + "accept": false, + "reason": "Typos/docs cleanup versus a TemporalBasicTransformerBlock code fix; unrelated changes." + }, + { + "left": "pull_request:7212", + "right": "pull_request:8836", + "accept": false, + "reason": "One removes duplicated block code; the other fixes a sliced-attention loop boundary bug. Different concrete problems." + }, + { + "left": "pull_request:7213", + "right": "pull_request:8412", + "accept": false, + "reason": "Merge-LoRAs docs versus single-file loading docs; both docs, but different features and intent." + }, + { + "left": "pull_request:7259", + "right": "pull_request:8990", + "accept": false, + "reason": "TCD intro docs versus PAG img2img feature work; different pipelines and deliverables." + }, + { + "left": "pull_request:7339", + "right": "pull_request:7778", + "accept": false, + "reason": "Release dependency workflow fix versus a debugging SSH workflow; unrelated CI changes." + }, + { + "left": "pull_request:7458", + "right": "pull_request:9251", + "accept": false, + "reason": "Skipping dynamo tests on Python 3.12 is unrelated to triggering fast tests on patch branches." + }, + { + "left": "pull_request:7464", + "right": "pull_request:8426", + "accept": false, + "reason": "LDM3D import fix versus mirror-community workflow dependency fix; no shared bug." + }, + { + "left": "pull_request:7464", + "right": "pull_request:8491", + "accept": false, + "reason": "LDM3D fix and dual-transformer import move touch different modules and issues." + }, + { + "left": "pull_request:7472", + "right": "pull_request:9036", + "accept": false, + "reason": "Broken link in examples README versus nightly deps workflow fix; different scope." + }, + { + "left": "pull_request:748", + "right": "pull_request:819", + "accept": false, + "reason": "push_to_hub call-site fix in examples versus DDIM timestep int64 fix; unrelated." + }, + { + "left": "pull_request:7510", + "right": "pull_request:9036", + "accept": false, + "reason": "OOM-related test tweak versus nightly dependency workflow update; no common bug." + }, + { + "left": "pull_request:7510", + "right": "pull_request:9116", + "accept": false, + "reason": "Test VAE tiling memory fix versus dead-link docs fix; unrelated." + }, + { + "left": "pull_request:7520", + "right": "pull_request:9463", + "accept": false, + "reason": "HD-Painter community pipeline addition versus PAG SD img2img support; different features." + }, + { + "left": "pull_request:7564", + "right": "pull_request:9036", + "accept": false, + "reason": "Test adjustments for Python 3.12/dynamo versus a workflow dependency change." + }, + { + "left": "pull_request:7564", + "right": "pull_request:9116", + "accept": false, + "reason": "Test/runtime compatibility fixes versus a documentation dead-link fix." + }, + { + "left": "pull_request:7594", + "right": "pull_request:8224", + "accept": false, + "reason": "Callback docs bugfix versus DeepFloyd training docs; different documentation topics." + }, + { + "left": "pull_request:7627", + "right": "pull_request:7913", + "accept": false, + "reason": "Playground img2img latent handling fix versus custom sigmas/timesteps support; separate pipeline behaviors." + }, + { + "left": "pull_request:7709", + "right": "pull_request:8990", + "accept": false, + "reason": "Workflow tailscale setup versus a large Mochi model/pipeline feature addition." + }, + { + "left": "pull_request:774", + "right": "pull_request:78", + "accept": false, + "reason": "Tiny test assertion tweaks versus a mid-block model refactor; unrelated." + }, + { + "left": "pull_request:7790", + "right": "pull_request:8467", + "accept": false, + "reason": "Masked SDXL community pipeline adaptation versus SEGA pipeline device placement fix; different code paths." + }, + { + "left": "pull_request:7791", + "right": "pull_request:8836", + "accept": false, + "reason": "AnimateDiff/IP-Adapter compatibility fix versus sliced-attention loop bug; separate bugs." + }, + { + "left": "pull_request:7857", + "right": "pull_request:9649", + "accept": false, + "reason": "PixArtSigma docs/exports versus scheduler return-value API change; unrelated." + }, + { + "left": "pull_request:7919", + "right": "pull_request:9684", + "accept": false, + "reason": "Device-map offload behavior fix versus CogVideoX frame batching improvements; not the same issue." + }, + { + "left": "pull_request:7919", + "right": "pull_request:9769", + "accept": false, + "reason": "Offload behavior fix versus Mochi model/pipeline feature work; completely different changes." + }, + { + "left": "pull_request:7938", + "right": "pull_request:9721", + "accept": false, + "reason": "Slack reporting workflow refactor versus pinning torch versions in Dockerfiles; unrelated infra updates." + }, + { + "left": "pull_request:7977", + "right": "pull_request:9036", + "accept": false, + "reason": "Attention typo doc fix versus nightly deps workflow fix; no overlap." + }, + { + "left": "pull_request:7977", + "right": "pull_request:9116", + "accept": false, + "reason": "Attention typo docs fix versus dead-link docs fix; both docs, but different problems." + }, + { + "left": "pull_request:800", + "right": "pull_request:904", + "accept": false, + "reason": "WIP VQ Diffusion implementation versus stable diffusion inpainting; entirely different model families." + }, + { + "left": "pull_request:8225", + "right": "pull_request:9015", + "accept": false, + "reason": "Ignoring checkpoint uploads during push versus fixing a PIA example snippet; unrelated." + }, + { + "left": "pull_request:824", + "right": "pull_request:828", + "accept": false, + "reason": "Both are Flax-related test work, but 824 fixes pipeline scheduler formatting while 828 expands/adjusts tests and imports; not the same concrete bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 236, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:6532", + "pull_request:6534", + "pull_request:7030", + "pull_request:7031", + "pull_request:7613", + "pull_request:7615", + "pull_request:8298", + "pull_request:8310", + "pull_request:8386", + "pull_request:839", + "pull_request:842", + "pull_request:8426", + "pull_request:8478", + "pull_request:8481", + "pull_request:8483", + "pull_request:8491", + "pull_request:8521", + "pull_request:8522", + "pull_request:8583", + "pull_request:8607", + "pull_request:8618", + "pull_request:8652", + "pull_request:8656", + "pull_request:8685", + "pull_request:8705", + "pull_request:8737", + "pull_request:8779", + "pull_request:8851", + "pull_request:8978", + "pull_request:8990", + "pull_request:9036", + "pull_request:9070", + "pull_request:9077", + "pull_request:9083", + "pull_request:9130", + "pull_request:9196", + "pull_request:9249", + "pull_request:9426", + "pull_request:9536", + "pull_request:9569", + "pull_request:9592", + "pull_request:9658", + "pull_request:9662", + "pull_request:9739", + "pull_request:9934", + "pull_request:9938", + "pull_request:9994" + ], + "soft_pairs": [ + "pull_request:8298|pull_request:8652", + "pull_request:8310|pull_request:8478", + "pull_request:8386|pull_request:9536", + "pull_request:839|pull_request:842", + "pull_request:8426|pull_request:9739", + "pull_request:8481|pull_request:9036", + "pull_request:8483|pull_request:8990", + "pull_request:8491|pull_request:9077", + "pull_request:8491|pull_request:9739", + "pull_request:8521|pull_request:8978", + "pull_request:8522|pull_request:9662", + "pull_request:8583|pull_request:8607", + "pull_request:8607|pull_request:8618", + "pull_request:8607|pull_request:9249", + "pull_request:8656|pull_request:9077", + "pull_request:8656|pull_request:9739", + "pull_request:8685|pull_request:9592", + "pull_request:8705|pull_request:8990", + "pull_request:8737|pull_request:9070", + "pull_request:8779|pull_request:9569", + "pull_request:8851|pull_request:9934", + "pull_request:8990|pull_request:9130", + "pull_request:8990|pull_request:9196", + "pull_request:9083|pull_request:9658", + "pull_request:9083|pull_request:9994", + "pull_request:9426|pull_request:9938", + "pull_request:7030|pull_request:7031", + "pull_request:6532|pull_request:6534", + "pull_request:7613|pull_request:7615" + ], + "prepared_review_unit_hash": "92489cec02a03a3beec180a37ce2851ae7bfda5fbcfb4a04706d1acf6d210f60", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 96590, + "estimated_input_tokens": 24148, + "estimated_eval_tokens": 48552 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft-edge suggestions are mostly false positives across unrelated feature, docs, CI, and test changes. Only three pairs are true duplicates of the same concrete change: the DreamBooth torch.compile compatibility fix, the SDXL ControlNet Inpaint IP-Adapter support, and the advanced SDXL LoRA model-card standardization.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR cleanly spans this mixed cluster; if a fallback is needed, PR 8483 is the strongest general representative because it is a large merged feature PR with substantial review and downstream references.", + "best_issue_reason": null, + "best_pr_reason": "PR 8483 is the best overall anchor among these items for the same reason: it is the most substantial merged artifact in the set, whereas most other PRs are narrow and unrelated to one another.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8298", + "right": "pull_request:8652", + "accept": false, + "reason": "Unrelated changes: CLI env platform info cleanup vs a new Lumina-T2X feature PR." + }, + { + "left": "pull_request:8310", + "right": "pull_request:8478", + "accept": false, + "reason": "Both are Hunyuan-related, but one is a model refactor and the other adds controlnet/pipeline support; not the same concrete change." + }, + { + "left": "pull_request:8386", + "right": "pull_request:9536", + "accept": false, + "reason": "Test helper shape cleanup vs ChatGLM tokenizer padding fix; no shared bug or code path." + }, + { + "left": "pull_request:839", + "right": "pull_request:842", + "accept": false, + "reason": "README auth-token cleanup and pipeline test sample-output cleanup are unrelated." + }, + { + "left": "pull_request:8426", + "right": "pull_request:9739", + "accept": false, + "reason": "Mirror-community CI fix vs SD3 typo/single-file docs change; different concerns." + }, + { + "left": "pull_request:8481", + "right": "pull_request:9036", + "accept": false, + "reason": "Image processor error-message tweak vs nightly dependency workflow fix; unrelated." + }, + { + "left": "pull_request:8483", + "right": "pull_request:8990", + "accept": false, + "reason": "Major SD3 feature addition vs SDXL ControlNet PAG img2img pipeline addition; same broad area, different feature and code paths." + }, + { + "left": "pull_request:8491", + "right": "pull_request:9077", + "accept": false, + "reason": "Dual-transformer import fix vs typo cleanup and controlnet_xs changes; unrelated." + }, + { + "left": "pull_request:8491", + "right": "pull_request:9739", + "accept": false, + "reason": "Import-path fix vs SD3 typo/single-file docs change; unrelated." + }, + { + "left": "pull_request:8521", + "right": "pull_request:8978", + "accept": false, + "reason": "Requirements pin cleanup vs broad notebook/example update; not the same underlying issue." + }, + { + "left": "pull_request:8522", + "right": "pull_request:9662", + "accept": false, + "reason": "Both touch transformers versions, but one changes package pins and the other adjusts a test requirement; not the same concrete change." + }, + { + "left": "pull_request:8583", + "right": "pull_request:8607", + "accept": false, + "reason": "XFormers SD3 support vs TAESD3 docs; same ecosystem, different changes." + }, + { + "left": "pull_request:8607", + "right": "pull_request:8618", + "accept": false, + "reason": "Docs for TAESD3 usage vs code support for shift_factor; closely related, but not the same artifact or bug fix." + }, + { + "left": "pull_request:8607", + "right": "pull_request:9249", + "accept": false, + "reason": "TAESD3 docs vs SD3 ControlNet training example; unrelated feature docs." + }, + { + "left": "pull_request:8656", + "right": "pull_request:9077", + "accept": false, + "reason": "Shard-checkpoint loading fix vs typo cleanup/controlnet_xs edits; unrelated." + }, + { + "left": "pull_request:8656", + "right": "pull_request:9739", + "accept": false, + "reason": "Sharded checkpoint fix vs SD3 typo/single-file docs update; unrelated." + }, + { + "left": "pull_request:8685", + "right": "pull_request:9592", + "accept": false, + "reason": "SD3 float8 note vs Flax docstring refactor; both docs-oriented but not the same change." + }, + { + "left": "pull_request:8705", + "right": "pull_request:8990", + "accept": false, + "reason": "Docker benchmarking container addition vs SDXL ControlNet PAG pipeline feature; unrelated." + }, + { + "left": "pull_request:8737", + "right": "pull_request:9070", + "accept": false, + "reason": "Large CI/benchmarking overhaul vs FLUX img2img pipeline draft; different scopes and code paths." + }, + { + "left": "pull_request:8779", + "right": "pull_request:9569", + "accept": false, + "reason": "SD3 img2img test tweak vs CogVideoX autoencoder decode fix; unrelated." + }, + { + "left": "pull_request:8851", + "right": "pull_request:9934", + "accept": false, + "reason": "AuraFlow docs vs Mochi docs; same doc-only category, but different models/features." + }, + { + "left": "pull_request:8990", + "right": "pull_request:9130", + "accept": false, + "reason": "SDXL ControlNet PAG img2img pipeline vs textual inversion README update; unrelated." + }, + { + "left": "pull_request:8990", + "right": "pull_request:9196", + "accept": false, + "reason": "Pipeline feature addition vs PixArt attention-processor chore; not the same underlying change." + }, + { + "left": "pull_request:9083", + "right": "pull_request:9658", + "accept": false, + "reason": "Flux single-file transformer support vs a Flux ControlNet community example; related model family, but different code paths and goals." + }, + { + "left": "pull_request:9083", + "right": "pull_request:9994", + "accept": false, + "reason": "Flux single-file loader support vs Omnigen model/pipeline addition; unrelated." + }, + { + "left": "pull_request:9426", + "right": "pull_request:9938", + "accept": false, + "reason": "CI artifact uploader version bump vs SDXL reference pipeline update; unrelated." + }, + { + "left": "pull_request:7030", + "right": "pull_request:7031", + "accept": true, + "reason": "Same SDXL ControlNet Inpaint pipeline change adding IPAdapterMixin/imports, and both target the same issue." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6534", + "accept": true, + "reason": "Same DreamBooth torch.compile compatibility fix across the SD and LoRA scripts, with the same explicit issue target." + }, + { + "left": "pull_request:7613", + "right": "pull_request:7615", + "accept": true, + "reason": "Same advanced diffusion training model-card standardization work, split only by SD15 vs SDXL script variants." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 237, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10364", + "pull_request:10365", + "pull_request:10783", + "pull_request:10830", + "pull_request:10888", + "pull_request:10889", + "pull_request:11395", + "pull_request:11426", + "pull_request:11427", + "pull_request:11463", + "pull_request:11491", + "pull_request:11492", + "pull_request:11744", + "pull_request:11783", + "pull_request:4659", + "pull_request:4661", + "pull_request:5065", + "pull_request:5066", + "pull_request:5299", + "pull_request:5303", + "pull_request:5304", + "pull_request:5312", + "pull_request:5910", + "pull_request:5911", + "pull_request:6195", + "pull_request:6328", + "pull_request:6509", + "pull_request:6525", + "pull_request:6532", + "pull_request:6534", + "pull_request:6555", + "pull_request:6556", + "pull_request:6721", + "pull_request:6765", + "pull_request:6766", + "pull_request:6767", + "pull_request:6768", + "pull_request:6769", + "pull_request:7052", + "pull_request:7307", + "pull_request:7865", + "pull_request:7898", + "pull_request:9579", + "pull_request:9582", + "pull_request:9583", + "pull_request:9606", + "pull_request:9724", + "pull_request:9815" + ], + "soft_pairs": [ + "pull_request:5910|pull_request:5911", + "pull_request:6195|pull_request:6721", + "pull_request:5299|pull_request:5312", + "pull_request:11426|pull_request:11427", + "pull_request:5065|pull_request:5066", + "pull_request:7865|pull_request:7898", + "pull_request:6555|pull_request:6556", + "pull_request:6534|pull_request:6555", + "pull_request:6525|pull_request:6555", + "pull_request:6532|pull_request:6555", + "pull_request:6525|pull_request:6532", + "pull_request:6525|pull_request:6534", + "pull_request:11395|pull_request:11491", + "pull_request:10364|pull_request:10365", + "pull_request:9583|pull_request:9724", + "pull_request:6534|pull_request:6556", + "pull_request:6532|pull_request:6556", + "pull_request:7052|pull_request:7307", + "pull_request:10783|pull_request:10889", + "pull_request:11491|pull_request:11492", + "pull_request:11744|pull_request:11783", + "pull_request:6765|pull_request:6766", + "pull_request:6765|pull_request:6767", + "pull_request:6765|pull_request:6768", + "pull_request:6765|pull_request:6769", + "pull_request:6766|pull_request:6767", + "pull_request:6766|pull_request:6768", + "pull_request:6767|pull_request:6768", + "pull_request:11463|pull_request:11491", + "pull_request:9724|pull_request:9815", + "pull_request:6328|pull_request:6509", + "pull_request:5303|pull_request:5304", + "pull_request:9579|pull_request:9606", + "pull_request:11395|pull_request:11492", + "pull_request:10830|pull_request:10888", + "pull_request:9582|pull_request:9583", + "pull_request:9582|pull_request:9815", + "pull_request:6766|pull_request:6769", + "pull_request:6767|pull_request:6769", + "pull_request:4659|pull_request:4661" + ], + "prepared_review_unit_hash": "aa28d71954a969def8cf352fcdac7324709d4d4cb9885b20ffbfbc611e3e08eb", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 40, + "serialized_chars": 108355, + "estimated_input_tokens": 27089, + "estimated_eval_tokens": 54434 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes several unrelated PR families (XLA support, torch.compile compatibility, interpolation defaults, metadata handling, docstring cleanup, and test tweaks). Only a subset are true duplicates or direct follow-ups; many same-issue or same-subsystem PRs are too different in concrete code path to merge safely.", + "confidence": 0.54, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR cleanly covers the whole cluster because the items split into multiple unrelated change families. If forced to pick one representative, PR 5911 is the strongest merged exemplar for the IPAdapterFull feature because it is the completed, user-facing implementation with docs/tests.", + "best_issue_reason": null, + "best_pr_reason": "PR 5911 is the most complete representative artifact in the set: merged, clearly scoped, and backed by docs/tests. It best fits a canonical representative for one of the true duplicate subclusters.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5910", + "right": "pull_request:5911", + "accept": true, + "reason": "Same IPAdapterFull feature update in the same loader/model paths; 5911 looks like the completed merged version." + }, + { + "left": "pull_request:6195", + "right": "pull_request:6721", + "accept": true, + "reason": "Same AnimateDiff SDXL feature work; 6721 is the finished PR for the same pipeline change." + }, + { + "left": "pull_request:5299", + "right": "pull_request:5312", + "accept": true, + "reason": "Both are broad model-doc/typehint cleanups in the same `src/diffusers/models` area and could be combined." + }, + { + "left": "pull_request:11426", + "right": "pull_request:11427", + "accept": true, + "reason": "Same interpolation-mode feature for closely related text-to-image training scripts; one PR could cover both files." + }, + { + "left": "pull_request:5065", + "right": "pull_request:5066", + "accept": true, + "reason": "Same attention-processor typo fix; same underlying correction with examples and core code." + }, + { + "left": "pull_request:7865", + "right": "pull_request:7898", + "accept": false, + "reason": "Different test targets (Dance Diffusion vs DiT) and no shared concrete bug beyond reducing test size." + }, + { + "left": "pull_request:6555", + "right": "pull_request:6556", + "accept": true, + "reason": "Same torch.compile compatibility fix applied to sibling text-to-image LoRA scripts." + }, + { + "left": "pull_request:6534", + "right": "pull_request:6555", + "accept": false, + "reason": "Different training scripts and code paths; same issue tag isn't enough." + }, + { + "left": "pull_request:6525", + "right": "pull_request:6555", + "accept": false, + "reason": "ControlNet vs DreamBooth are separate scripts and fixes." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6555", + "accept": false, + "reason": "Different script families; too broad to be a single duplicate change." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6534", + "accept": false, + "reason": "DreamBooth SD and LoRA are distinct implementations, not the same PR." + }, + { + "left": "pull_request:6525", + "right": "pull_request:6534", + "accept": false, + "reason": "Different training scripts and code paths." + }, + { + "left": "pull_request:11395", + "right": "pull_request:11491", + "accept": true, + "reason": "Same interpolation default change across DreamBooth-style scripts; same feature and same kind of edit." + }, + { + "left": "pull_request:10364", + "right": "pull_request:10365", + "accept": false, + "reason": "Separate XLA support patches for different pipelines, not one concrete code path." + }, + { + "left": "pull_request:9583", + "right": "pull_request:9724", + "accept": false, + "reason": "Unrelated docstring refactors in different modules." + }, + { + "left": "pull_request:6534", + "right": "pull_request:6556", + "accept": false, + "reason": "Different script families; same compile-compat theme only." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6556", + "accept": false, + "reason": "Different script families and files." + }, + { + "left": "pull_request:7052", + "right": "pull_request:7307", + "accept": true, + "reason": "Same stable_diffusion_tensorrt_txt2img update; 7307 looks like the follow-up/expanded version of 7052." + }, + { + "left": "pull_request:10783", + "right": "pull_request:10889", + "accept": true, + "reason": "Same fp16 unscaling fix in the same dreambooth_lora_sdxl script; 10889 is the merged final PR." + }, + { + "left": "pull_request:11491", + "right": "pull_request:11492", + "accept": false, + "reason": "Different training-script families; same interpolation theme but not the same concrete change." + }, + { + "left": "pull_request:11744", + "right": "pull_request:11783", + "accept": false, + "reason": "Sana vs SD3 are different scripts/models; metadata support is analogous but not a duplicate." + }, + { + "left": "pull_request:6765", + "right": "pull_request:6766", + "accept": false, + "reason": "Repo-wide path cleanup in different modules, not one shared code path." + }, + { + "left": "pull_request:6765", + "right": "pull_request:6767", + "accept": false, + "reason": "Different utility modules; same path-handling style change only." + }, + { + "left": "pull_request:6765", + "right": "pull_request:6768", + "accept": false, + "reason": "Different utility modules and code paths." + }, + { + "left": "pull_request:6765", + "right": "pull_request:6769", + "accept": false, + "reason": "Different utility modules." + }, + { + "left": "pull_request:6766", + "right": "pull_request:6767", + "accept": false, + "reason": "Different utility modules." + }, + { + "left": "pull_request:6766", + "right": "pull_request:6768", + "accept": false, + "reason": "Different utility modules." + }, + { + "left": "pull_request:6767", + "right": "pull_request:6768", + "accept": false, + "reason": "Different utility modules." + }, + { + "left": "pull_request:11463", + "right": "pull_request:11491", + "accept": false, + "reason": "ControlNet SDXL and DreamBooth Lora Lumina2 are different training scripts." + }, + { + "left": "pull_request:9724", + "right": "pull_request:9815", + "accept": true, + "reason": "Same docstring refactor/fix series in `training_utils.py`; 9815 is the more complete follow-up." + }, + { + "left": "pull_request:6328", + "right": "pull_request:6509", + "accept": false, + "reason": "AnimateDiff video-to-video vs img-to-video are related but separate features." + }, + { + "left": "pull_request:5303", + "right": "pull_request:5304", + "accept": true, + "reason": "Same DeepFloyd IF docstring/typehint cleanup in the same pipeline file set." + }, + { + "left": "pull_request:9579", + "right": "pull_request:9606", + "accept": false, + "reason": "Different modules and unrelated doc fixes." + }, + { + "left": "pull_request:9582", + "right": "pull_request:9583", + "accept": false, + "reason": "Different modules; same docstring cleanup style only." + }, + { + "left": "pull_request:9582", + "right": "pull_request:9815", + "accept": false, + "reason": "Different modules and issue scope." + }, + { + "left": "pull_request:4659", + "right": "pull_request:4661", + "accept": true, + "reason": "Same SDXL long weighted prompt pipeline change; 4661 is clearly the replacement/follow-up to 4659." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 238, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11431", + "pull_request:11455", + "pull_request:11465", + "pull_request:11471", + "pull_request:11472", + "pull_request:11491", + "pull_request:11511", + "pull_request:11512", + "pull_request:2046", + "pull_request:2383", + "pull_request:2384", + "pull_request:3633", + "pull_request:3952", + "pull_request:4595", + "pull_request:4596", + "pull_request:5382", + "pull_request:5506", + "pull_request:5614", + "pull_request:5697", + "pull_request:5725", + "pull_request:5744", + "pull_request:5748", + "pull_request:5906", + "pull_request:5914", + "pull_request:6208", + "pull_request:6241", + "pull_request:6259", + "pull_request:6525", + "pull_request:6526", + "pull_request:6556", + "pull_request:6558", + "pull_request:6576", + "pull_request:6577", + "pull_request:6654", + "pull_request:6655", + "pull_request:7072", + "pull_request:7233", + "pull_request:7235", + "pull_request:8916", + "pull_request:8919", + "pull_request:8932", + "pull_request:8987", + "pull_request:9591", + "pull_request:9706", + "pull_request:9707", + "pull_request:9714", + "pull_request:9715" + ], + "soft_pairs": [ + "pull_request:7233|pull_request:7235", + "pull_request:4595|pull_request:4596", + "pull_request:9707|pull_request:9714", + "pull_request:5725|pull_request:5748", + "pull_request:5906|pull_request:5914", + "pull_request:6208|pull_request:6259", + "pull_request:6241|pull_request:6259", + "pull_request:5725|pull_request:5744", + "pull_request:6208|pull_request:6241", + "pull_request:5382|pull_request:5614", + "pull_request:11455|pull_request:11465", + "pull_request:9591|pull_request:9706", + "pull_request:6558|pull_request:6576", + "pull_request:6654|pull_request:6655", + "pull_request:8916|pull_request:8919", + "pull_request:11431|pull_request:11512", + "pull_request:6526|pull_request:6576", + "pull_request:9714|pull_request:9715", + "pull_request:6526|pull_request:6577", + "pull_request:6576|pull_request:6577", + "pull_request:9707|pull_request:9715", + "pull_request:11431|pull_request:11511", + "pull_request:5506|pull_request:5697", + "pull_request:11471|pull_request:11491", + "pull_request:11511|pull_request:11512", + "pull_request:2383|pull_request:2384", + "pull_request:3633|pull_request:3952", + "pull_request:8919|pull_request:8932", + "pull_request:6526|pull_request:6556", + "pull_request:6556|pull_request:6576", + "pull_request:2046|pull_request:3633", + "pull_request:11471|pull_request:11472", + "pull_request:2046|pull_request:3952", + "pull_request:6525|pull_request:6526", + "pull_request:7072|pull_request:7233", + "pull_request:11472|pull_request:11491", + "pull_request:7072|pull_request:7235", + "pull_request:8932|pull_request:8987" + ], + "prepared_review_unit_hash": "acfce77d04156eab0f7b24f8eeac1e293cceb25ceaa4980e011d821038d3ad5a", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 38, + "serialized_chars": 99899, + "estimated_input_tokens": 24975, + "estimated_eval_tokens": 50206 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not one coherent duplicate cluster; it splits into several small near-duplicate groups around the same feature/fix family (DoRA, torch.compile compatibility, interpolation-mode flags, docs sweeps), plus many unrelated PRs. I accepted only pairs that look like the same concrete change or a direct revert; I rejected pairs that only share a broad theme like docs, tests, or training-scripts in the same area.", + "confidence": 0.71, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:7233", + "right": "pull_request:7235", + "accept": true, + "reason": "Same DreamBooth LoRA SDXL DoRA feature; one is the code change, the other adds the matching docs/script follow-up in the same file family." + }, + { + "left": "pull_request:4595", + "right": "pull_request:4596", + "accept": true, + "reason": "Both are about adding the SDXL conversion script; 4595 looks like the WIP precursor and 4596 is the realized implementation." + }, + { + "left": "pull_request:9707", + "right": "pull_request:9714", + "accept": true, + "reason": "Same docstring refactor in the same AuraFlow transformer file with essentially the same patch." + }, + { + "left": "pull_request:5725", + "right": "pull_request:5748", + "accept": false, + "reason": "Both are docs cleanup sweeps, but they are split into different page/file sets; too broad to treat as the same concrete change." + }, + { + "left": "pull_request:5906", + "right": "pull_request:5914", + "accept": true, + "reason": "Same SDXL LoRA README/template bug in the same script, with 5914 extending/finalizing the same fix." + }, + { + "left": "pull_request:6208", + "right": "pull_request:6259", + "accept": false, + "reason": "Same error text, but different code paths and different files; these are not the same bug fix." + }, + { + "left": "pull_request:6241", + "right": "pull_request:6259", + "accept": false, + "reason": "No concrete diff/file evidence for 6241, and nothing shows it is the same fix as 6259." + }, + { + "left": "pull_request:5725", + "right": "pull_request:5744", + "accept": false, + "reason": "This is still a docs sweep split into chunks; similar topic, but not one identical change set." + }, + { + "left": "pull_request:6208", + "right": "pull_request:6241", + "accept": false, + "reason": "No shared concrete implementation or file-level evidence tying them to the same fix." + }, + { + "left": "pull_request:5382", + "right": "pull_request:5614", + "accept": true, + "reason": "Direct revert/fix of the same SDXL training original-size ordering bug in the same pair of scripts." + }, + { + "left": "pull_request:11455", + "right": "pull_request:11465", + "accept": true, + "reason": "Same interpolation-mode feature in closely related SDXL training scripts; effectively the same change pattern." + }, + { + "left": "pull_request:9591", + "right": "pull_request:9706", + "accept": false, + "reason": "Both are docs-oriented, but they touch different files and different model docs; too general to be a duplicate change." + }, + { + "left": "pull_request:6558", + "right": "pull_request:6576", + "accept": false, + "reason": "Both add torch.compile compatibility, but to different training scripts and code paths; same theme, not the same change." + }, + { + "left": "pull_request:6654", + "right": "pull_request:6655", + "accept": false, + "reason": "Related DreamBooth modularization work, but for different script variants and not one single concrete fix." + }, + { + "left": "pull_request:8916", + "right": "pull_request:8919", + "accept": false, + "reason": "Both add model test coverage, but they are for different models and different files; too broad to merge as duplicates." + }, + { + "left": "pull_request:11431", + "right": "pull_request:11512", + "accept": false, + "reason": "Both are torch.compile test additions, but for different transformer models and different test files." + }, + { + "left": "pull_request:6526", + "right": "pull_request:6576", + "accept": false, + "reason": "Same broad compile-compatibility idea, but different pipelines and different code paths." + }, + { + "left": "pull_request:9714", + "right": "pull_request:9715", + "accept": true, + "reason": "Same AuraFlow docstring refactor in the same file family; these look like the same change iteration." + }, + { + "left": "pull_request:6526", + "right": "pull_request:6577", + "accept": false, + "reason": "Both are compile-compat fixes, but they target different training scripts and are not the same code-path problem." + }, + { + "left": "pull_request:6576", + "right": "pull_request:6577", + "accept": false, + "reason": "Different training scripts with a similar unwrap-model pattern; related, but not the same concrete fix." + }, + { + "left": "pull_request:9707", + "right": "pull_request:9715", + "accept": true, + "reason": "Same AuraFlow transformer docstring refactor in the same file, with highly overlapping edits." + }, + { + "left": "pull_request:11431", + "right": "pull_request:11511", + "accept": false, + "reason": "Same testing theme, but different transformer models and different files." + }, + { + "left": "pull_request:5506", + "right": "pull_request:5697", + "accept": true, + "reason": "Same attention_processor PEFT/xformers bug fix; 5697 is the cleaner follow-up of the same code-path issue." + }, + { + "left": "pull_request:11471", + "right": "pull_request:11491", + "accept": false, + "reason": "Same interpolation-mode idea, but applied to different DreamBooth variants and different training scripts." + }, + { + "left": "pull_request:11511", + "right": "pull_request:11512", + "accept": false, + "reason": "Both are compile-test additions, but they cover different models and are separate test PRs." + }, + { + "left": "pull_request:2383", + "right": "pull_request:2384", + "accept": true, + "reason": "Exact same Pix2Pix Zero utility-function addition and its revert; clearly the same change." + }, + { + "left": "pull_request:3633", + "right": "pull_request:3952", + "accept": false, + "reason": "Both pin huggingface-hub RCs in setup.py, but they are different version experiments and not the same change." + }, + { + "left": "pull_request:8919", + "right": "pull_request:8932", + "accept": false, + "reason": "Model test coverage additions for different models; same category, not the same artifact." + }, + { + "left": "pull_request:6526", + "right": "pull_request:6556", + "accept": false, + "reason": "Both address compile compatibility, but on different scripts and different training flows." + }, + { + "left": "pull_request:6556", + "right": "pull_request:6576", + "accept": false, + "reason": "Similar compile-compat pattern, but different scripts and not one shared fix." + }, + { + "left": "pull_request:2046", + "right": "pull_request:3633", + "accept": false, + "reason": "Different dependency pin experiments in setup.py, separated by time and target version." + }, + { + "left": "pull_request:11471", + "right": "pull_request:11472", + "accept": true, + "reason": "Same new interpolation-mode option added to closely related advanced DreamBooth training variants." + }, + { + "left": "pull_request:2046", + "right": "pull_request:3952", + "accept": false, + "reason": "Both are temporary dependency pins, but they target different RC versions and different experiments." + }, + { + "left": "pull_request:6525", + "right": "pull_request:6526", + "accept": false, + "reason": "Same general compile-compat theme, but separate SD and SDXL training scripts; not the same concrete change." + }, + { + "left": "pull_request:7072", + "right": "pull_request:7233", + "accept": false, + "reason": "Both mention DoRA for DreamBooth LoRA, but they are different script variants and not one code patch." + }, + { + "left": "pull_request:11472", + "right": "pull_request:11491", + "accept": false, + "reason": "Similar interpolation default change, but across different model-specific training scripts; too broad to merge as a duplicate." + }, + { + "left": "pull_request:9706", + "right": "pull_request:9715", + "accept": true, + "reason": "Same docstring refactor in the same AuraFlow transformer file, with strongly overlapping edits." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 239, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10299", + "pull_request:10316", + "pull_request:11449", + "pull_request:11492", + "pull_request:11782", + "pull_request:11783", + "pull_request:2825", + "pull_request:2826", + "pull_request:3201", + "pull_request:3237", + "pull_request:3265", + "pull_request:3285", + "pull_request:4629", + "pull_request:4659", + "pull_request:4661", + "pull_request:4899", + "pull_request:5938", + "pull_request:6526", + "pull_request:6532", + "pull_request:6534", + "pull_request:6556", + "pull_request:6558", + "pull_request:6559", + "pull_request:6569", + "pull_request:6577", + "pull_request:6654", + "pull_request:6655", + "pull_request:7603", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7833", + "pull_request:7846", + "pull_request:7849", + "pull_request:8283", + "pull_request:8335", + "pull_request:8773", + "pull_request:8810", + "pull_request:8864", + "pull_request:8916", + "pull_request:8932", + "pull_request:8973", + "pull_request:8976", + "pull_request:8985", + "pull_request:9706", + "pull_request:9707", + "pull_request:9714" + ], + "soft_pairs": [ + "pull_request:11782|pull_request:11783", + "pull_request:8810|pull_request:8864", + "pull_request:7797|pull_request:7803", + "pull_request:7797|pull_request:7804", + "pull_request:7797|pull_request:7833", + "pull_request:7797|pull_request:7846", + "pull_request:7803|pull_request:7804", + "pull_request:7803|pull_request:7833", + "pull_request:7803|pull_request:7846", + "pull_request:7804|pull_request:7833", + "pull_request:7804|pull_request:7846", + "pull_request:7833|pull_request:7846", + "pull_request:4629|pull_request:4659", + "pull_request:4629|pull_request:4661", + "pull_request:7797|pull_request:7849", + "pull_request:7803|pull_request:7849", + "pull_request:7804|pull_request:7849", + "pull_request:7833|pull_request:7849", + "pull_request:7846|pull_request:7849", + "pull_request:9706|pull_request:9707", + "pull_request:3201|pull_request:3265", + "pull_request:8773|pull_request:8976", + "pull_request:8283|pull_request:8335", + "pull_request:10299|pull_request:10316", + "pull_request:11449|pull_request:11492", + "pull_request:3237|pull_request:3285", + "pull_request:6559|pull_request:6654", + "pull_request:6559|pull_request:7603", + "pull_request:6559|pull_request:8283", + "pull_request:6654|pull_request:7603", + "pull_request:6654|pull_request:8283", + "pull_request:7603|pull_request:8283", + "pull_request:6526|pull_request:6558", + "pull_request:4899|pull_request:5938", + "pull_request:9706|pull_request:9714", + "pull_request:2825|pull_request:2826", + "pull_request:8916|pull_request:8932", + "pull_request:6526|pull_request:6532", + "pull_request:8973|pull_request:8985", + "pull_request:6556|pull_request:6577", + "pull_request:6569|pull_request:6655", + "pull_request:6569|pull_request:7603", + "pull_request:6526|pull_request:6534", + "pull_request:6532|pull_request:6558" + ], + "prepared_review_unit_hash": "717e10bb0c087ffc4e4b17d1484cb487922992d9ec1665025d9e0b5d374453b2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 44, + "serialized_chars": 108616, + "estimated_input_tokens": 27154, + "estimated_eval_tokens": 54564 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set mostly splits into a few clear duplicate chains: exact/near-exact reruns of the same PR, replacement PRs for the same feature, and a number of similar-but-distinct example/test updates that should stay separate. The strongest duplicate clusters are the SDXL long weighted prompt pipeline, ControlNet LoRA support, disk offload support, and a few same-file doc/test follow-ups.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4661 is the best canonical PR: it is the merged final replacement for the SDXL long weighted prompt pipeline work and most directly represents the finished change in that cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 4661 is the strongest overall representative among these PRs because it is merged, final, and corresponds to a coherent feature rather than a partial draft or a narrow follow-up.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11782", + "right": "pull_request:11783", + "accept": true, + "reason": "Same DreamBooth LoRA SD3 feature and same files; this looks like the same PR/change in duplicate form." + }, + { + "left": "pull_request:8810", + "right": "pull_request:8864", + "accept": true, + "reason": "Same PAG support for SD ControlNet Img2Img, same implementation area and same concrete change." + }, + { + "left": "pull_request:7797", + "right": "pull_request:7803", + "accept": false, + "reason": "Both shrink fast tests, but they target different pipelines (DDPM vs DDIM) and are separate changes." + }, + { + "left": "pull_request:7797", + "right": "pull_request:7804", + "accept": false, + "reason": "Different test suites and different model types; only the general goal is shared." + }, + { + "left": "pull_request:7797", + "right": "pull_request:7833", + "accept": false, + "reason": "Same optimization theme, but this is a different pipeline test with a distinct code path." + }, + { + "left": "pull_request:7797", + "right": "pull_request:7846", + "accept": false, + "reason": "Different audio pipeline test; not the same underlying bug or change." + }, + { + "left": "pull_request:7803", + "right": "pull_request:7804", + "accept": false, + "reason": "Different test files/pipelines, so this is not one mergeable change." + }, + { + "left": "pull_request:7803", + "right": "pull_request:7833", + "accept": false, + "reason": "Separate pipeline test reductions, not the same concrete update." + }, + { + "left": "pull_request:7803", + "right": "pull_request:7846", + "accept": false, + "reason": "Different fast test and different model setup." + }, + { + "left": "pull_request:7804", + "right": "pull_request:7833", + "accept": false, + "reason": "Only broadly related test-size reductions; they are distinct PRs." + }, + { + "left": "pull_request:7804", + "right": "pull_request:7846", + "accept": false, + "reason": "Different pipelines and test files; not a duplicate change." + }, + { + "left": "pull_request:7833", + "right": "pull_request:7846", + "accept": false, + "reason": "Separate audio pipeline test reductions, not one concrete fix." + }, + { + "left": "pull_request:4629", + "right": "pull_request:4659", + "accept": true, + "reason": "Same SDXL long weighted prompt pipeline work; 4659 is a replacement of 4629." + }, + { + "left": "pull_request:4629", + "right": "pull_request:4661", + "accept": true, + "reason": "Same feature and same affected area; 4661 is the final merged version of the same work." + }, + { + "left": "pull_request:7797", + "right": "pull_request:7849", + "accept": false, + "reason": "Different pipelines and test files; only the broad goal of shrinking tests matches." + }, + { + "left": "pull_request:7803", + "right": "pull_request:7849", + "accept": false, + "reason": "Different fast test and different model architecture." + }, + { + "left": "pull_request:7804", + "right": "pull_request:7849", + "accept": false, + "reason": "Unrelated test suites, so not the same underlying change." + }, + { + "left": "pull_request:7833", + "right": "pull_request:7849", + "accept": false, + "reason": "Different pipeline tests with different model reductions." + }, + { + "left": "pull_request:7846", + "right": "pull_request:7849", + "accept": false, + "reason": "Different test suite and different pipeline." + }, + { + "left": "pull_request:9706", + "right": "pull_request:9707", + "accept": true, + "reason": "Same file and same docstring refactor in AuraFlow transformer docs." + }, + { + "left": "pull_request:3201", + "right": "pull_request:3265", + "accept": true, + "reason": "Direct revert / re-revert of the same community pipeline change in the same file." + }, + { + "left": "pull_request:8773", + "right": "pull_request:8976", + "accept": true, + "reason": "Both are the same LoRA base mixin revert work and touch the same core files and functionality." + }, + { + "left": "pull_request:8283", + "right": "pull_request:8335", + "accept": false, + "reason": "Both modularize validation/inference, but they target different training scripts and code paths." + }, + { + "left": "pull_request:10299", + "right": "pull_request:10316", + "accept": false, + "reason": "Same documentation pattern, but different pipeline docs and separate edits." + }, + { + "left": "pull_request:11449", + "right": "pull_request:11492", + "accept": false, + "reason": "Both set LANCZOS defaults, but for different training scripts; too broad to treat as one duplicate." + }, + { + "left": "pull_request:3237", + "right": "pull_request:3285", + "accept": true, + "reason": "Same disk-offload feature in the same core offload codepath; 3285 looks like the focused implementation behind the broader PR." + }, + { + "left": "pull_request:6559", + "right": "pull_request:6654", + "accept": false, + "reason": "Different training scripts and separate modularization changes." + }, + { + "left": "pull_request:6559", + "right": "pull_request:7603", + "accept": false, + "reason": "Different example scripts; same general refactor pattern is not enough." + }, + { + "left": "pull_request:6559", + "right": "pull_request:8283", + "accept": false, + "reason": "Different pipelines and different training examples." + }, + { + "left": "pull_request:6654", + "right": "pull_request:7603", + "accept": false, + "reason": "Different examples and code paths despite similar validation refactors." + }, + { + "left": "pull_request:6654", + "right": "pull_request:8283", + "accept": false, + "reason": "Separate training scripts, not one mergeable fix." + }, + { + "left": "pull_request:7603", + "right": "pull_request:8283", + "accept": false, + "reason": "Different scripts and separate inference workflows." + }, + { + "left": "pull_request:6526", + "right": "pull_request:6558", + "accept": false, + "reason": "Different training scripts; compile-compatibility is the theme, but not the same code change." + }, + { + "left": "pull_request:4899", + "right": "pull_request:5938", + "accept": true, + "reason": "Same ControlNet LoRA support work touching the same LoRA/controlnet internals." + }, + { + "left": "pull_request:9706", + "right": "pull_request:9714", + "accept": true, + "reason": "Same docstring refactor in the same file; 9714 is another version of the same change." + }, + { + "left": "pull_request:2825", + "right": "pull_request:2826", + "accept": true, + "reason": "Same DreamBooth offset-noise feature in the same script; 2826 is the completed implementation." + }, + { + "left": "pull_request:8916", + "right": "pull_request:8932", + "accept": false, + "reason": "Different test modules for different model families; only both are coverage additions." + }, + { + "left": "pull_request:6526", + "right": "pull_request:6532", + "accept": false, + "reason": "Both are torch.compile compatibility fixes, but for different training scripts and model paths." + }, + { + "left": "pull_request:8973", + "right": "pull_request:8985", + "accept": true, + "reason": "Same Lumina test-size reduction in the same test file; these are effectively the same fast-test change." + }, + { + "left": "pull_request:6556", + "right": "pull_request:6577", + "accept": false, + "reason": "Different example scripts and different model families; not one concrete change." + }, + { + "left": "pull_request:6569", + "right": "pull_request:6655", + "accept": false, + "reason": "Different training scripts and separate inference/validation refactors." + }, + { + "left": "pull_request:6569", + "right": "pull_request:7603", + "accept": false, + "reason": "Different example pipelines; same refactor pattern is too broad." + }, + { + "left": "pull_request:6526", + "right": "pull_request:6534", + "accept": false, + "reason": "Different scripts (ControlNet SDXL vs DreamBooth LoRA SD), so not the same underlying fix." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6558", + "accept": false, + "reason": "Separate training scripts and separate compile-compatibility changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 240, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10268", + "pull_request:10386", + "pull_request:10387", + "pull_request:10781", + "pull_request:11421", + "pull_request:11451", + "pull_request:11472", + "pull_request:11537", + "pull_request:11552", + "pull_request:12269", + "pull_request:12454", + "pull_request:1346", + "pull_request:1347", + "pull_request:1788", + "pull_request:2094", + "pull_request:3458", + "pull_request:3460", + "pull_request:4499", + "pull_request:4500", + "pull_request:4501", + "pull_request:4502", + "pull_request:5299", + "pull_request:5312", + "pull_request:5391", + "pull_request:5833", + "pull_request:5834", + "pull_request:6207", + "pull_request:6266", + "pull_request:6534", + "pull_request:6558", + "pull_request:6559", + "pull_request:6654", + "pull_request:6655", + "pull_request:6940", + "pull_request:6942", + "pull_request:6944", + "pull_request:6947", + "pull_request:7603", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7833", + "pull_request:8283", + "pull_request:8335", + "pull_request:8973", + "pull_request:9083", + "pull_request:9706", + "pull_request:9715" + ], + "soft_pairs": [ + "pull_request:1346|pull_request:1347", + "pull_request:6207|pull_request:6266", + "pull_request:6534|pull_request:6558", + "pull_request:1788|pull_request:2094", + "pull_request:10386|pull_request:10387", + "pull_request:3458|pull_request:3460", + "pull_request:4499|pull_request:4500", + "pull_request:4499|pull_request:4501", + "pull_request:4499|pull_request:4502", + "pull_request:4500|pull_request:4501", + "pull_request:4500|pull_request:4502", + "pull_request:4501|pull_request:4502", + "pull_request:6942|pull_request:6944", + "pull_request:5833|pull_request:5834", + "pull_request:6940|pull_request:6944", + "pull_request:6942|pull_request:6947", + "pull_request:6944|pull_request:6947", + "pull_request:7797|pull_request:8973", + "pull_request:7803|pull_request:8973", + "pull_request:7804|pull_request:8973", + "pull_request:7833|pull_request:8973", + "pull_request:11421|pull_request:11472", + "pull_request:11451|pull_request:11552", + "pull_request:10268|pull_request:9083", + "pull_request:5299|pull_request:5391", + "pull_request:5312|pull_request:5391", + "pull_request:12269|pull_request:12454", + "pull_request:11451|pull_request:11537", + "pull_request:6559|pull_request:8335", + "pull_request:6654|pull_request:8335", + "pull_request:6655|pull_request:8335", + "pull_request:9706|pull_request:9715", + "pull_request:6559|pull_request:6655", + "pull_request:6655|pull_request:7603", + "pull_request:6655|pull_request:8283", + "pull_request:10268|pull_request:10781" + ], + "prepared_review_unit_hash": "244b6941e6f3e5e5734fad2abed774de1323ba02d7c2efa21534421b6291be97", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 36, + "serialized_chars": 103523, + "estimated_input_tokens": 25881, + "estimated_eval_tokens": 52018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is really several smaller duplicate subgroups rather than one single bug. I accepted only the pairs that are the same concrete fix/feature on the same code path; most other pairs are just similar maintenance work in different scripts or subsystems.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4502 is the best canonical representative for the large SDXL Reference subcluster: it is the merged, most complete version of the same exact feature added by the other identical PRs.", + "best_issue_reason": null, + "best_pr_reason": "PR 4502 is the strongest overall representative here because it is merged, has the most discussion/inbound references, and sits at the center of the largest exact-duplicate group in this cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1346", + "right": "pull_request:1347", + "accept": false, + "reason": "Both touch the same README, but they fix different example snippets and different issues; not the same underlying change." + }, + { + "left": "pull_request:6207", + "right": "pull_request:6266", + "accept": false, + "reason": "Both are LoRA-script cleanup, but they remove different deprecated helpers from different training scripts; too broad to merge as one duplicate." + }, + { + "left": "pull_request:6534", + "right": "pull_request:6558", + "accept": false, + "reason": "Same torch.compile theme, but they target different training scripts and code paths, so they are not the same fix." + }, + { + "left": "pull_request:1788", + "right": "pull_request:2094", + "accept": true, + "reason": "Both fix the DDIM scheduler handling for prediction_type='sample' in the same code path." + }, + { + "left": "pull_request:10386", + "right": "pull_request:10387", + "accept": true, + "reason": "Same title, same file, same new example; these are effectively duplicate submissions of the same loading-support change." + }, + { + "left": "pull_request:3458", + "right": "pull_request:3460", + "accept": true, + "reason": "Both implement the same inpaint LoRA scale/cross-attention support in the same pipeline, with 3460 adding the supporting tests/import cleanup." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4500", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4501", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set." + }, + { + "left": "pull_request:4499", + "right": "pull_request:4502", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set; 4502 is the merged final version." + }, + { + "left": "pull_request:4500", + "right": "pull_request:4501", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set." + }, + { + "left": "pull_request:4500", + "right": "pull_request:4502", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set; 4502 is the merged final version." + }, + { + "left": "pull_request:4501", + "right": "pull_request:4502", + "accept": true, + "reason": "Same SDXL reference_attn/reference_adain feature, same files, same change set; 4502 is the merged final version." + }, + { + "left": "pull_request:6942", + "right": "pull_request:6944", + "accept": false, + "reason": "Both are model-card standardization work, but they target different training scripts and different model families." + }, + { + "left": "pull_request:5833", + "right": "pull_request:5834", + "accept": false, + "reason": "These are both single-file/import related, but they fix different paths and different problems; not the same concrete bug." + }, + { + "left": "pull_request:6940", + "right": "pull_request:6944", + "accept": false, + "reason": "Both standardize model cards, but one is for plain T2I LoRA and the other is for SDXL LoRA; separate changes." + }, + { + "left": "pull_request:6942", + "right": "pull_request:6947", + "accept": false, + "reason": "Same general model-card cleanup theme, but they apply to different scripts and adapter types, not one shared change." + }, + { + "left": "pull_request:6944", + "right": "pull_request:6947", + "accept": false, + "reason": "Different training scripts and model-card targets; only the broad maintenance area overlaps." + }, + { + "left": "pull_request:7797", + "right": "pull_request:8973", + "accept": false, + "reason": "Both reduce fast-test model sizes, but they are for different pipeline tests and unrelated code paths." + }, + { + "left": "pull_request:7803", + "right": "pull_request:8973", + "accept": false, + "reason": "Different test files and models; same broad test-speed theme only." + }, + { + "left": "pull_request:7804", + "right": "pull_request:8973", + "accept": false, + "reason": "Different test files and models; same broad test-speed theme only." + }, + { + "left": "pull_request:7833", + "right": "pull_request:8973", + "accept": false, + "reason": "Different test files and models; same broad test-speed theme only." + }, + { + "left": "pull_request:11421", + "right": "pull_request:11472", + "accept": false, + "reason": "Same interpolation default idea, but applied to different training scripts and pipelines; not one concrete change." + }, + { + "left": "pull_request:11451", + "right": "pull_request:11552", + "accept": false, + "reason": "Both add torch.compile tests, but for different model classes and test files." + }, + { + "left": "pull_request:10268", + "right": "pull_request:9083", + "accept": false, + "reason": "Both add single-file support, but for different models (Mochi vs Flux), so these are distinct features." + }, + { + "left": "pull_request:5299", + "right": "pull_request:5391", + "accept": false, + "reason": "Both improve docs/type hints, but they touch different files and a broad documentation sweep rather than one shared bug." + }, + { + "left": "pull_request:5312", + "right": "pull_request:5391", + "accept": false, + "reason": "Same broad docs/type-hint cleanup, but not the same concrete change set or code path." + }, + { + "left": "pull_request:12269", + "right": "pull_request:12454", + "accept": true, + "reason": "Both add the same Flux Kontext modular pipeline support across the same modular-pipeline files." + }, + { + "left": "pull_request:11451", + "right": "pull_request:11537", + "accept": false, + "reason": "One adds a Wan compile test and the other targets UNet2DConditionModel; same testing category only." + }, + { + "left": "pull_request:6559", + "right": "pull_request:8335", + "accept": false, + "reason": "Both modularize training inference, but for different scripts and generations; not the same underlying fix." + }, + { + "left": "pull_request:6654", + "right": "pull_request:8335", + "accept": false, + "reason": "Different training script and pipeline family; related refactor pattern only." + }, + { + "left": "pull_request:6655", + "right": "pull_request:8335", + "accept": false, + "reason": "Different training script and pipeline family; related refactor pattern only." + }, + { + "left": "pull_request:9706", + "right": "pull_request:9715", + "accept": true, + "reason": "Same AuraFlow docstring refactor in the same file; this is the same underlying documentation change." + }, + { + "left": "pull_request:6559", + "right": "pull_request:6655", + "accept": false, + "reason": "Both are training-script inference refactors, but they affect different model families and code paths." + }, + { + "left": "pull_request:6655", + "right": "pull_request:7603", + "accept": false, + "reason": "Same broad modularization pattern, but different scripts and different inference flows." + }, + { + "left": "pull_request:6655", + "right": "pull_request:8283", + "accept": false, + "reason": "Same broad modularization pattern, but different scripts and different inference flows." + }, + { + "left": "pull_request:10268", + "right": "pull_request:10781", + "accept": false, + "reason": "Both are single-file support work, but for different transformer models and different code changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 241, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10288", + "pull_request:10504", + "pull_request:10505", + "pull_request:11407", + "pull_request:11421", + "pull_request:11471", + "pull_request:11476", + "pull_request:11492", + "pull_request:11537", + "pull_request:11552", + "pull_request:11744", + "pull_request:11782", + "pull_request:121", + "pull_request:12387", + "pull_request:12439", + "pull_request:2046", + "pull_request:252", + "pull_request:2602", + "pull_request:3197", + "pull_request:3201", + "pull_request:3265", + "pull_request:3952", + "pull_request:6119", + "pull_request:6231", + "pull_request:6526", + "pull_request:6532", + "pull_request:6550", + "pull_request:6555", + "pull_request:6556", + "pull_request:6559", + "pull_request:6569", + "pull_request:6577", + "pull_request:6654", + "pull_request:6944", + "pull_request:7615", + "pull_request:7797", + "pull_request:7803", + "pull_request:7804", + "pull_request:7833", + "pull_request:8283", + "pull_request:8335", + "pull_request:8985", + "pull_request:9083", + "pull_request:9257", + "pull_request:9473", + "pull_request:9592", + "pull_request:9707" + ], + "soft_pairs": [ + "pull_request:11744|pull_request:11782", + "pull_request:6944|pull_request:7615", + "pull_request:2602|pull_request:3952", + "pull_request:11537|pull_request:11552", + "pull_request:10288|pull_request:9083", + "pull_request:6526|pull_request:6555", + "pull_request:6532|pull_request:6577", + "pull_request:9257|pull_request:9473", + "pull_request:12387|pull_request:12439", + "pull_request:3197|pull_request:3201", + "pull_request:3197|pull_request:3265", + "pull_request:6569|pull_request:8335", + "pull_request:6119|pull_request:6231", + "pull_request:11471|pull_request:11492", + "pull_request:7797|pull_request:8985", + "pull_request:7803|pull_request:8985", + "pull_request:7804|pull_request:8985", + "pull_request:7833|pull_request:8985", + "pull_request:11421|pull_request:11471", + "pull_request:6550|pull_request:6556", + "pull_request:6559|pull_request:6569", + "pull_request:6569|pull_request:6654", + "pull_request:6569|pull_request:8283", + "pull_request:9592|pull_request:9707", + "pull_request:2046|pull_request:2602", + "pull_request:10504|pull_request:10505", + "pull_request:11407|pull_request:11476", + "pull_request:121|pull_request:252" + ], + "prepared_review_unit_hash": "fb6af76d9774c6a6c613965551bc7d0d9ffdc0cb7175bc8eb26242dfa5fe7f96", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 101050, + "estimated_input_tokens": 25263, + "estimated_eval_tokens": 50782 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of similarly-shaped PRs (test cleanups, training-script refactors, docs tweaks) rather than true duplicates. I only accepted the exact or near-exact same-change pairs: the repeated lpw_stable_diffusion revert/update cycle, the identical print-removal test PRs, and the shared fp16 LoRA upcasting fix.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10505 is the cleanest canonical representative: it is a merged copy of the exact same test print-removal change as 10504, with identical touched files and diff intent.", + "best_issue_reason": null, + "best_pr_reason": "PR 10505 is the strongest overall representative because it is an exact duplicate of 10504, already merged, and the change is unambiguous and self-contained.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11744", + "right": "pull_request:11782", + "accept": false, + "reason": "Both add LoRA metadata handling to DreamBooth-related scripts, but Sana and SD3 are separate model-specific code paths; this is similar work, not the same change." + }, + { + "left": "pull_request:6944", + "right": "pull_request:7615", + "accept": false, + "reason": "Both standardize model-card handling, but they target different training scripts and different implementations; too broad to treat as duplicates." + }, + { + "left": "pull_request:2602", + "right": "pull_request:3952", + "accept": false, + "reason": "These are unrelated temporary dependency-pin test PRs for different huggingface-hub RC versions, not the same bug or change." + }, + { + "left": "pull_request:11537", + "right": "pull_request:11552", + "accept": false, + "reason": "Both add torch.compile tests, but for different models and test files; same chore pattern, not the same underlying artifact." + }, + { + "left": "pull_request:10288", + "right": "pull_request:9083", + "accept": false, + "reason": "Both add Flux single-file support, but for different model variants and code paths (Canny/Depth/Fill vs Transformer)." + }, + { + "left": "pull_request:6526", + "right": "pull_request:6555", + "accept": false, + "reason": "Both make training scripts torch.compile-compatible, but they patch different scripts and different unwrap/save paths; not the same PR." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6577", + "accept": false, + "reason": "Different training scripts and different compile-compatibility fixes; similar subsystem, but not the same change." + }, + { + "left": "pull_request:9257", + "right": "pull_request:9473", + "accept": false, + "reason": "Both touch Flux DreamBooth code, but one changes prompt/tokenization flow and the other fixes Prodigy learning-rate handling; distinct bugs." + }, + { + "left": "pull_request:12387", + "right": "pull_request:12439", + "accept": false, + "reason": "Both concern attention backends, but they add different backend support and different core changes; too broad to merge as duplicates." + }, + { + "left": "pull_request:3197", + "right": "pull_request:3201", + "accept": false, + "reason": "3201 is a revert of 3197, so the changes are opposite, not duplicates." + }, + { + "left": "pull_request:3197", + "right": "pull_request:3265", + "accept": true, + "reason": "3265 restores the exact same lpw_stable_diffusion update as 3197; same file and same change set." + }, + { + "left": "pull_request:6569", + "right": "pull_request:8335", + "accept": false, + "reason": "Both add validation/inference modularization, but in different example scripts with different logic; not the same code-path fix." + }, + { + "left": "pull_request:6119", + "right": "pull_request:6231", + "accept": true, + "reason": "Both apply the same fp16 LoRA fix pattern: upcast trainable parameters to fp32 to avoid gradient-scaling issues in LoRA training scripts." + }, + { + "left": "pull_request:11471", + "right": "pull_request:11492", + "accept": false, + "reason": "Same interpolation-default idea, but implemented in different DreamBooth training scripts; related refactor, not a duplicate change." + }, + { + "left": "pull_request:7797", + "right": "pull_request:8985", + "accept": false, + "reason": "These are fast-test size reductions for different pipelines; same optimization theme, but not the same test or bug." + }, + { + "left": "pull_request:7803", + "right": "pull_request:8985", + "accept": false, + "reason": "Different pipeline tests and different model-size reductions; not the same concrete change." + }, + { + "left": "pull_request:7804", + "right": "pull_request:8985", + "accept": false, + "reason": "Different amused vs lumina test simplifications; broad test-speedup theme only." + }, + { + "left": "pull_request:7833", + "right": "pull_request:8985", + "accept": false, + "reason": "Different audio vs lumina test reductions; not the same underlying artifact." + }, + { + "left": "pull_request:11421", + "right": "pull_request:11471", + "accept": false, + "reason": "Both set LANCZOS defaults, but in different training scripts and contexts; similar refactor, not a duplicate." + }, + { + "left": "pull_request:6550", + "right": "pull_request:6556", + "accept": false, + "reason": "Both address torch.compile compatibility, but for different SD/SDXL training scripts and different code paths." + }, + { + "left": "pull_request:6559", + "right": "pull_request:6569", + "accept": false, + "reason": "Both are training/inference workflow refactors, but they touch different scripts and solve different behavior." + }, + { + "left": "pull_request:6569", + "right": "pull_request:6654", + "accept": false, + "reason": "Different example scripts and different inference modularization changes; same family, not same PR." + }, + { + "left": "pull_request:9592", + "right": "pull_request:9707", + "accept": false, + "reason": "Both are docstring cleanup PRs, but for unrelated modules and separate changes." + }, + { + "left": "pull_request:2046", + "right": "pull_request:2602", + "accept": false, + "reason": "These are different temporary hub version test bumps at different versions, not duplicates." + }, + { + "left": "pull_request:10504", + "right": "pull_request:10505", + "accept": true, + "reason": "These are effectively identical print-removal test PRs across the same files, with no meaningful change differences." + }, + { + "left": "pull_request:11407", + "right": "pull_request:11476", + "accept": false, + "reason": "Both are typo-only sweeps, but across different sets of files and unrelated strings/comments." + }, + { + "left": "pull_request:121", + "right": "pull_request:252", + "accept": false, + "reason": "Both update the README, but the content and scope are different example additions rather than the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 242, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1157", + "pull_request:1186", + "pull_request:11920", + "pull_request:11955", + "pull_request:12142", + "pull_request:12143", + "pull_request:12307", + "pull_request:13115", + "pull_request:13254", + "pull_request:13315", + "pull_request:13335", + "pull_request:2812", + "pull_request:2813", + "pull_request:326", + "pull_request:327", + "pull_request:3554", + "pull_request:3555", + "pull_request:4022", + "pull_request:4684", + "pull_request:5828", + "pull_request:6261", + "pull_request:6262", + "pull_request:6263", + "pull_request:6264", + "pull_request:6265", + "pull_request:6267", + "pull_request:6908", + "pull_request:6909", + "pull_request:6910", + "pull_request:6939", + "pull_request:6940", + "pull_request:6942", + "pull_request:6944", + "pull_request:6947", + "pull_request:7692", + "pull_request:7746", + "pull_request:7876", + "pull_request:7903", + "pull_request:8064", + "pull_request:8426", + "pull_request:8427", + "pull_request:8816", + "pull_request:8820", + "pull_request:8946", + "pull_request:9591", + "pull_request:9592", + "pull_request:9593", + "pull_request:9707" + ], + "soft_pairs": [ + "pull_request:12307|pull_request:13115", + "pull_request:2812|pull_request:2813", + "pull_request:3554|pull_request:3555", + "pull_request:8816|pull_request:8820", + "pull_request:12142|pull_request:12143", + "pull_request:326|pull_request:327", + "pull_request:6908|pull_request:6910", + "pull_request:6909|pull_request:6910", + "pull_request:6939|pull_request:6940", + "pull_request:6262|pull_request:6263", + "pull_request:6262|pull_request:6264", + "pull_request:6263|pull_request:6264", + "pull_request:6908|pull_request:6909", + "pull_request:6939|pull_request:6942", + "pull_request:6939|pull_request:6944", + "pull_request:4022|pull_request:4684", + "pull_request:5828|pull_request:7746", + "pull_request:6261|pull_request:6262", + "pull_request:6261|pull_request:6263", + "pull_request:6261|pull_request:6264", + "pull_request:6262|pull_request:6267", + "pull_request:6263|pull_request:6267", + "pull_request:6264|pull_request:6267", + "pull_request:7903|pull_request:8946", + "pull_request:8064|pull_request:8946", + "pull_request:8426|pull_request:8427", + "pull_request:6940|pull_request:6942", + "pull_request:7692|pull_request:7876", + "pull_request:9592|pull_request:9593", + "pull_request:6261|pull_request:6267", + "pull_request:6265|pull_request:6267", + "pull_request:6939|pull_request:6947", + "pull_request:6940|pull_request:6947", + "pull_request:9591|pull_request:9707", + "pull_request:1157|pull_request:1186", + "pull_request:11920|pull_request:11955", + "pull_request:13254|pull_request:13315", + "pull_request:13254|pull_request:13335" + ], + "prepared_review_unit_hash": "155add5443c79604821105006f29fa128234062fd220ba1907f7aa310ba0b4aa", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 38, + "serialized_chars": 110612, + "estimated_input_tokens": 27653, + "estimated_eval_tokens": 55562 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster actually contains several unrelated subtopics (modular pipeline refactors, model-card standardization, translation/docs updates, test slices, etc.). I only accepted soft edges where the diffs look like the same concrete change or a clear rework of the same file path; most lookalike PRs are distinct follow-up changes and should stay separate.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR spans the whole set; the items split into multiple unrelated duplicate subclusters. If forced to anchor on one, the PAG controlnet PR pair is the clearest true duplicate cluster.", + "best_issue_reason": null, + "best_pr_reason": "8820 is the strongest single PR anchor among the true duplicate pairs: it is the more complete, merged version of the PAG controlnet change, including code, docs, and tests.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12307", + "right": "pull_request:13115", + "accept": false, + "reason": "Related modular-pipeline cleanup, but not the same concrete change; 13115 is a follow-up export tweak, not a duplicate of the larger 12307 fix." + }, + { + "left": "pull_request:2812", + "right": "pull_request:2813", + "accept": true, + "reason": "Same doc update to the same files; the diffs are effectively identical." + }, + { + "left": "pull_request:3554", + "right": "pull_request:3555", + "accept": true, + "reason": "Same generated example/script bundle with the same files and content; clearly duplicated." + }, + { + "left": "pull_request:8816", + "right": "pull_request:8820", + "accept": true, + "reason": "Same PAG controlnet change on the same code path; 8820 is the fuller follow-up with docs/tests." + }, + { + "left": "pull_request:12142", + "right": "pull_request:12143", + "accept": true, + "reason": "Same bf16/fp16 fix in `pipeline_wan_vace.py` on the same line; 12143 just packages it with extra test changes." + }, + { + "left": "pull_request:326", + "right": "pull_request:327", + "accept": true, + "reason": "Same PNDM typing change in the same pipeline file; 327 is a narrower rework of the same edit." + }, + { + "left": "pull_request:6908", + "right": "pull_request:6910", + "accept": false, + "reason": "Both are model-card standardizations, but they touch different training scripts and are not the same underlying change." + }, + { + "left": "pull_request:6909", + "right": "pull_request:6910", + "accept": false, + "reason": "Different ControlNet training scripts; similar cleanup pattern, but not a duplicate change." + }, + { + "left": "pull_request:6939", + "right": "pull_request:6940", + "accept": false, + "reason": "Same style of model-card refactor, but applied to different scripts and different outputs." + }, + { + "left": "pull_request:6262", + "right": "pull_request:6263", + "accept": false, + "reason": "Both move different Stable Diffusion variants out into separate packages; same refactor pattern, but distinct pipelines." + }, + { + "left": "pull_request:6262", + "right": "pull_request:6264", + "accept": false, + "reason": "Different pipeline split-outs (Panorama vs SAG), so not the same change." + }, + { + "left": "pull_request:6263", + "right": "pull_request:6264", + "accept": false, + "reason": "Different pipeline split-outs with separate targets; related refactor family, not duplicates." + }, + { + "left": "pull_request:6908", + "right": "pull_request:6909", + "accept": false, + "reason": "Same standardization theme, but different training scripts and file targets." + }, + { + "left": "pull_request:6939", + "right": "pull_request:6942", + "accept": false, + "reason": "Both standardize model cards, but for different scripts (T2I vs SDXL) and different metadata." + }, + { + "left": "pull_request:6939", + "right": "pull_request:6944", + "accept": false, + "reason": "Different LoRA vs base training scripts; similar cleanup, not the same PR." + }, + { + "left": "pull_request:4022", + "right": "pull_request:4684", + "accept": false, + "reason": "Both are Korean doc updates, but they are separate translation passes with different content and pages." + }, + { + "left": "pull_request:5828", + "right": "pull_request:7746", + "accept": false, + "reason": "Different tests and different bugs; one is a no-grad annotation, the other fixes VAE tiling dtype." + }, + { + "left": "pull_request:6261", + "right": "pull_request:6262", + "accept": false, + "reason": "Same refactor family, but different pipelines and different code paths." + }, + { + "left": "pull_request:6261", + "right": "pull_request:6263", + "accept": false, + "reason": "Different moved pipeline modules; not duplicate changes." + }, + { + "left": "pull_request:6261", + "right": "pull_request:6264", + "accept": false, + "reason": "Different moved pipeline modules; same broad refactor, not the same PR." + }, + { + "left": "pull_request:6262", + "right": "pull_request:6267", + "accept": false, + "reason": "Both relocate pipelines, but they target different variants and files." + }, + { + "left": "pull_request:6263", + "right": "pull_request:6267", + "accept": false, + "reason": "Different pipeline refactors; similar packaging work, not duplicates." + }, + { + "left": "pull_request:6264", + "right": "pull_request:6267", + "accept": false, + "reason": "Different moved pipeline targets; not the same underlying change." + }, + { + "left": "pull_request:7903", + "right": "pull_request:8946", + "accept": false, + "reason": "Unrelated test changes in different areas; no shared concrete bug/change." + }, + { + "left": "pull_request:8064", + "right": "pull_request:8946", + "accept": false, + "reason": "Different model/test areas and different intent; not duplicates." + }, + { + "left": "pull_request:8426", + "right": "pull_request:8427", + "accept": false, + "reason": "Both touch the same workflow file, but one changes install steps while the other adds debug output; separate fixes." + }, + { + "left": "pull_request:6940", + "right": "pull_request:6942", + "accept": false, + "reason": "Similar model-card standardization, but different scripts and different training contexts." + }, + { + "left": "pull_request:7692", + "right": "pull_request:7876", + "accept": true, + "reason": "Same community PAG pipeline behavior change; 7876 is a reworked version of the same fix." + }, + { + "left": "pull_request:9592", + "right": "pull_request:9593", + "accept": false, + "reason": "Docstring refactors in different files; same style, different code." + }, + { + "left": "pull_request:6261", + "right": "pull_request:6267", + "accept": false, + "reason": "Both are modular pipeline refactors, but they move different pipelines and add different exports." + }, + { + "left": "pull_request:6265", + "right": "pull_request:6267", + "accept": false, + "reason": "Different stable-diffusion subpipeline moves; related but not duplicate." + }, + { + "left": "pull_request:6939", + "right": "pull_request:6947", + "accept": false, + "reason": "Both standardize model cards, but for different pipelines and different code paths." + }, + { + "left": "pull_request:6940", + "right": "pull_request:6947", + "accept": false, + "reason": "Different training scripts and different model-card metadata; not the same change." + }, + { + "left": "pull_request:9591", + "right": "pull_request:9707", + "accept": false, + "reason": "Different docstring cleanups in different models; no duplicate bug/change." + }, + { + "left": "pull_request:1157", + "right": "pull_request:1186", + "accept": false, + "reason": "Similar Flax training-script fixes, but they target different examples and separate code paths." + }, + { + "left": "pull_request:11920", + "right": "pull_request:11955", + "accept": false, + "reason": "Different model test-slice additions for different pipelines; same pattern, not duplicates." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13315", + "accept": false, + "reason": "Different transformer test refactors for different models; not the same underlying change." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13335", + "accept": false, + "reason": "Different transformer test refactors for different models; related CI cleanup, not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 243, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10109", + "pull_request:10183", + "pull_request:10268", + "pull_request:10364", + "pull_request:10365", + "pull_request:10402", + "pull_request:10434", + "pull_request:10991", + "pull_request:11920", + "pull_request:11954", + "pull_request:11955", + "pull_request:13069", + "pull_request:13071", + "pull_request:13253", + "pull_request:13254", + "pull_request:13315", + "pull_request:13335", + "pull_request:13341", + "pull_request:13342", + "pull_request:13346", + "pull_request:1513", + "pull_request:1564", + "pull_request:2812", + "pull_request:2813", + "pull_request:2815", + "pull_request:3568", + "pull_request:3577", + "pull_request:5104", + "pull_request:5105", + "pull_request:5239", + "pull_request:5242", + "pull_request:5833", + "pull_request:5834", + "pull_request:6907", + "pull_request:6908", + "pull_request:6909", + "pull_request:6910", + "pull_request:6942", + "pull_request:6944", + "pull_request:6947", + "pull_request:6963", + "pull_request:7615", + "pull_request:7903", + "pull_request:7934", + "pull_request:8064", + "pull_request:9083", + "pull_request:9591", + "pull_request:9715" + ], + "soft_pairs": [ + "pull_request:13254|pull_request:13341", + "pull_request:13254|pull_request:13342", + "pull_request:13315|pull_request:13335", + "pull_request:13315|pull_request:13341", + "pull_request:13315|pull_request:13342", + "pull_request:13335|pull_request:13341", + "pull_request:13335|pull_request:13342", + "pull_request:13341|pull_request:13342", + "pull_request:5104|pull_request:5105", + "pull_request:5239|pull_request:5242", + "pull_request:7903|pull_request:8064", + "pull_request:13069|pull_request:13071", + "pull_request:6907|pull_request:6908", + "pull_request:6907|pull_request:6909", + "pull_request:6907|pull_request:6910", + "pull_request:10109|pull_request:10365", + "pull_request:10183|pull_request:10268", + "pull_request:11920|pull_request:11954", + "pull_request:11954|pull_request:11955", + "pull_request:13253|pull_request:13254", + "pull_request:13253|pull_request:13315", + "pull_request:13253|pull_request:13335", + "pull_request:13253|pull_request:13341", + "pull_request:13253|pull_request:13342", + "pull_request:13254|pull_request:13346", + "pull_request:13315|pull_request:13346", + "pull_request:13335|pull_request:13346", + "pull_request:13341|pull_request:13346", + "pull_request:2812|pull_request:2815", + "pull_request:2813|pull_request:2815", + "pull_request:3568|pull_request:3577", + "pull_request:6908|pull_request:6942", + "pull_request:10402|pull_request:10434", + "pull_request:6908|pull_request:6944", + "pull_request:6908|pull_request:6947", + "pull_request:6908|pull_request:6963", + "pull_request:6908|pull_request:7615", + "pull_request:9591|pull_request:9715", + "pull_request:6942|pull_request:6963", + "pull_request:6944|pull_request:6963", + "pull_request:6947|pull_request:6963", + "pull_request:1513|pull_request:1564", + "pull_request:5833|pull_request:7934", + "pull_request:5834|pull_request:7934", + "pull_request:10109|pull_request:10364", + "pull_request:10183|pull_request:10991", + "pull_request:10183|pull_request:9083" + ], + "prepared_review_unit_hash": "dcabbfaa1e574170cc1957a5995e1b933a2d4c6974ca4fd3a1d715113e90da69", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 47, + "serialized_chars": 107733, + "estimated_input_tokens": 26934, + "estimated_eval_tokens": 54124 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous maintenance cluster: many PRs are similar in style but apply to different models, pipelines, or example scripts. The only clear duplicate-looking subgroup is the docs link cleanup, especially the Stable UnCLIP and Hunyuan docs fixes.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "10434 is the strongest cluster representative: it is issue-linked, broad, and covers the same docs-link cleanup theme across many pages, making it the best canonical PR in this mixed set.", + "best_issue_reason": null, + "best_pr_reason": "10434 has the clearest scope and broadest coverage among the merged PRs here, so it is the best overall representative even though the cluster contains several unrelated subfamilies.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13254", + "right": "pull_request:13341", + "accept": false, + "reason": "Same refactor style, but different model test files (LTX vs Bria); not the same change." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13342", + "accept": false, + "reason": "Different model families and files; both are test refactors, not duplicates." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13335", + "accept": false, + "reason": "Different models (Sana vs Cosmos); similar refactor pattern only." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13341", + "accept": false, + "reason": "Different test targets and code paths; not the same underlying change." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13342", + "accept": false, + "reason": "Different models/pipelines; only the CI refactor pattern matches." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13341", + "accept": false, + "reason": "Different model test files; same refactor family, not a duplicate change." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13342", + "accept": false, + "reason": "Different model targets and test code; not the same PR." + }, + { + "left": "pull_request:13341", + "right": "pull_request:13342", + "accept": false, + "reason": "Different model families and files; too broad to treat as the same change." + }, + { + "left": "pull_request:5104", + "right": "pull_request:5105", + "accept": false, + "reason": "Same idea, but separate training scripts (SDXL LoRA vs base DreamBooth); not one concrete code-path change." + }, + { + "left": "pull_request:5239", + "right": "pull_request:5242", + "accept": false, + "reason": "Same class, but different methods (encode vs decode) and separate docstring fixes." + }, + { + "left": "pull_request:7903", + "right": "pull_request:8064", + "accept": false, + "reason": "Unrelated work: one is a video processor test PR, the other is a Marigold pipeline addition/test PR." + }, + { + "left": "pull_request:13069", + "right": "pull_request:13071", + "accept": false, + "reason": "Different transformer models and different test refactors; same CI pattern only." + }, + { + "left": "pull_request:6907", + "right": "pull_request:6908", + "accept": false, + "reason": "Both are ControlNet model-card cleanups, but for different example scripts and scopes." + }, + { + "left": "pull_request:6907", + "right": "pull_request:6909", + "accept": false, + "reason": "Different training entrypoints; same documentation standardization theme only." + }, + { + "left": "pull_request:6907", + "right": "pull_request:6910", + "accept": false, + "reason": "Same broad docs work, but different script and change set; not duplicates." + }, + { + "left": "pull_request:10109", + "right": "pull_request:10365", + "accept": false, + "reason": "Both add torch_xla support, but to different pipelines and code paths." + }, + { + "left": "pull_request:10183", + "right": "pull_request:10268", + "accept": false, + "reason": "Different single-file loaders for different models; same loader subsystem only." + }, + { + "left": "pull_request:11920", + "right": "pull_request:11954", + "accept": false, + "reason": "Both add test slices, but for different pipelines and expected outputs." + }, + { + "left": "pull_request:11954", + "right": "pull_request:11955", + "accept": false, + "reason": "Different pipeline test updates; same testing style, not the same bug/change." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13254", + "accept": false, + "reason": "Different model test refactors (Z-Image vs LTX); not the same concrete change." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13315", + "accept": false, + "reason": "Different models and files; same refactor pattern only." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13335", + "accept": false, + "reason": "Different model test targets; not a duplicate." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13341", + "accept": false, + "reason": "Different model families and tests; no shared concrete change." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13342", + "accept": false, + "reason": "Different model test refactors, not the same PR." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13346", + "accept": false, + "reason": "Different model test files and model families; only the refactor pattern overlaps." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13346", + "accept": false, + "reason": "Different models and test code paths; not duplicates." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13346", + "accept": false, + "reason": "Different model test refactors; same theme but not the same change." + }, + { + "left": "pull_request:13341", + "right": "pull_request:13346", + "accept": false, + "reason": "Different model families and files; not the same underlying fix." + }, + { + "left": "pull_request:2812", + "right": "pull_request:2815", + "accept": true, + "reason": "Same Stable UnCLIP docs cleanup theme, same docs file family, and 2815 is the broader final docs update." + }, + { + "left": "pull_request:2813", + "right": "pull_request:2815", + "accept": true, + "reason": "Same Stable UnCLIP docs fix set; 2815 looks like the consolidated/broader version." + }, + { + "left": "pull_request:3568", + "right": "pull_request:3577", + "accept": false, + "reason": "Both fix broken links, but in different documentation pages and contexts." + }, + { + "left": "pull_request:6908", + "right": "pull_request:6942", + "accept": false, + "reason": "Different example scripts and model-card templates; same standardization theme only." + }, + { + "left": "pull_request:6908", + "right": "pull_request:6944", + "accept": false, + "reason": "Different training scripts and docs content; not one concrete change." + }, + { + "left": "pull_request:6908", + "right": "pull_request:6947", + "accept": false, + "reason": "Same broad model-card cleanup category, but different pipelines and code paths." + }, + { + "left": "pull_request:6908", + "right": "pull_request:6963", + "accept": false, + "reason": "Different example scripts; similar docs standardization, not duplicates." + }, + { + "left": "pull_request:6908", + "right": "pull_request:7615", + "accept": false, + "reason": "Both standardize model cards, but for different training recipes and files." + }, + { + "left": "pull_request:10402", + "right": "pull_request:10434", + "accept": true, + "reason": "Same broken-links/docs-fix family; 10434 includes the same hunyuan pages and broadens the cleanup across more docs." + }, + { + "left": "pull_request:6942", + "right": "pull_request:6963", + "accept": false, + "reason": "Different training examples and model-card text; not the same change." + }, + { + "left": "pull_request:6944", + "right": "pull_request:6963", + "accept": false, + "reason": "Different scripts and card generation logic; only the docs standardization theme matches." + }, + { + "left": "pull_request:6947", + "right": "pull_request:6963", + "accept": false, + "reason": "Different training entrypoints and model-card content; not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 244, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10259", + "pull_request:10268", + "pull_request:1073", + "pull_request:10947", + "pull_request:10991", + "pull_request:11655", + "pull_request:11920", + "pull_request:11953", + "pull_request:11954", + "pull_request:11955", + "pull_request:1291", + "pull_request:13253", + "pull_request:13254", + "pull_request:13315", + "pull_request:13335", + "pull_request:13341", + "pull_request:13344", + "pull_request:13345", + "pull_request:13346", + "pull_request:13347", + "pull_request:13348", + "pull_request:1420", + "pull_request:178", + "pull_request:1998", + "pull_request:2486", + "pull_request:2487", + "pull_request:2488", + "pull_request:260", + "pull_request:2839", + "pull_request:309", + "pull_request:555", + "pull_request:5871", + "pull_request:5979", + "pull_request:6133", + "pull_request:6534", + "pull_request:6589", + "pull_request:6940", + "pull_request:6942", + "pull_request:6944", + "pull_request:6963", + "pull_request:7425", + "pull_request:7613", + "pull_request:7615", + "pull_request:7903", + "pull_request:7910", + "pull_request:9591", + "pull_request:9714" + ], + "soft_pairs": [ + "pull_request:10268|pull_request:10947", + "pull_request:11920|pull_request:11953", + "pull_request:11953|pull_request:11955", + "pull_request:1291|pull_request:309", + "pull_request:13253|pull_request:13346", + "pull_request:13254|pull_request:13345", + "pull_request:13315|pull_request:13345", + "pull_request:13335|pull_request:13345", + "pull_request:13341|pull_request:13345", + "pull_request:2486|pull_request:2488", + "pull_request:2487|pull_request:555", + "pull_request:5871|pull_request:7903", + "pull_request:5979|pull_request:6133", + "pull_request:6534|pull_request:6589", + "pull_request:7425|pull_request:7910", + "pull_request:9591|pull_request:9714", + "pull_request:10259|pull_request:11655", + "pull_request:6942|pull_request:7615", + "pull_request:6940|pull_request:7613", + "pull_request:6940|pull_request:7615", + "pull_request:6944|pull_request:7613", + "pull_request:6963|pull_request:7615", + "pull_request:10268|pull_request:10991", + "pull_request:1073|pull_request:260", + "pull_request:11953|pull_request:11954", + "pull_request:13253|pull_request:13344", + "pull_request:13253|pull_request:13345", + "pull_request:13254|pull_request:13347", + "pull_request:13254|pull_request:13348", + "pull_request:13315|pull_request:13347", + "pull_request:13315|pull_request:13348", + "pull_request:13335|pull_request:13347", + "pull_request:13335|pull_request:13348", + "pull_request:13341|pull_request:13347", + "pull_request:13341|pull_request:13348", + "pull_request:13344|pull_request:13346", + "pull_request:1420|pull_request:178", + "pull_request:1998|pull_request:2839" + ], + "prepared_review_unit_hash": "266da5cbaf8d179a337081a621b7ddd812a56d06020bb7e0b8583861896abe6c", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 38, + "serialized_chars": 102933, + "estimated_input_tokens": 25734, + "estimated_eval_tokens": 51724 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes unrelated docs typos, pipeline/model feature additions, and model-test refactors across many different components. No soft edge looks like the same underlying bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits well: the items span many unrelated PRs and codepaths, so there is no coherent duplicate anchor for the cluster.", + "best_issue_reason": null, + "best_pr_reason": "No best PR: the cluster is heterogeneous, and the PRs address different models, scripts, docs, and loader behaviors rather than one shared change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10268", + "right": "pull_request:10947", + "accept": false, + "reason": "Different single-file targets: Mochi Transformer vs SANA Transformer." + }, + { + "left": "pull_request:11920", + "right": "pull_request:11953", + "accept": false, + "reason": "Different model test slices: Wan vs HiDream-Image." + }, + { + "left": "pull_request:11953", + "right": "pull_request:11955", + "accept": false, + "reason": "Different model families: HiDream-Image vs Cosmos." + }, + { + "left": "pull_request:1291", + "right": "pull_request:309", + "accept": false, + "reason": "Both are doc link fixes, but they touch different notebooks/URLs and are unrelated changes." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13346", + "accept": false, + "reason": "Different transformer models and test suites: Z Image vs CogVideoX/CogView." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test refactors: LTX vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test refactors: Sana vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test refactors: Cosmos vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:13341", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test refactors: Bria vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:2486", + "right": "pull_request:2488", + "accept": false, + "reason": "Both change test mixin behavior, but they address different helpers and different concrete behaviors." + }, + { + "left": "pull_request:2487", + "right": "pull_request:555", + "accept": false, + "reason": "Unrelated: a Flax/JAX guide versus a VAE implementation." + }, + { + "left": "pull_request:5871", + "right": "pull_request:7903", + "accept": false, + "reason": "Unrelated: deprecated scheduler docs/rebase work versus video processor tests." + }, + { + "left": "pull_request:5979", + "right": "pull_request:6133", + "accept": false, + "reason": "Different new features: DDGAN pipeline/scheduler vs UFOGen pipeline/scheduler." + }, + { + "left": "pull_request:6534", + "right": "pull_request:6589", + "accept": false, + "reason": "Similar torch.compile theme, but different training scripts and code paths." + }, + { + "left": "pull_request:7425", + "right": "pull_request:7910", + "accept": false, + "reason": "Different documentation areas: PEFT tutorial snippet vs image processor API docs." + }, + { + "left": "pull_request:9591", + "right": "pull_request:9714", + "accept": false, + "reason": "Unrelated docstring edits in different model classes." + }, + { + "left": "pull_request:10259", + "right": "pull_request:11655", + "accept": false, + "reason": "Same loader file, but different problems: Flux LoRA loading vs bnb 8-bit dequantization support." + }, + { + "left": "pull_request:6942", + "right": "pull_request:7615", + "accept": false, + "reason": "Both standardize model cards, but for different training scripts and model variants." + }, + { + "left": "pull_request:6940", + "right": "pull_request:7613", + "accept": false, + "reason": "Different model-card updates for different training scripts and outputs." + }, + { + "left": "pull_request:6940", + "right": "pull_request:7615", + "accept": false, + "reason": "Different model-card updates for different training scripts and outputs." + }, + { + "left": "pull_request:6944", + "right": "pull_request:7613", + "accept": false, + "reason": "Different training scripts: T2I LoRA SDXL vs advanced SD15 LoRA." + }, + { + "left": "pull_request:6963", + "right": "pull_request:7615", + "accept": false, + "reason": "Different training scripts: textual inversion SDXL vs advanced SDXL LoRA." + }, + { + "left": "pull_request:10268", + "right": "pull_request:10991", + "accept": false, + "reason": "Different single-file targets: Mochi Transformer vs Wan models." + }, + { + "left": "pull_request:1073", + "right": "pull_request:260", + "accept": false, + "reason": "Different README typo fixes in unrelated example folders." + }, + { + "left": "pull_request:11953", + "right": "pull_request:11954", + "accept": false, + "reason": "Different fast test slices for different pipelines: HiDream-Image vs Hunyuan Video." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13344", + "accept": false, + "reason": "Different model test suites: Z Image vs GLM Image." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13345", + "accept": false, + "reason": "Different model test suites: Z Image vs Chroma/LongCat/HiDream." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13347", + "accept": false, + "reason": "Different transformer test groups and different models." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13348", + "accept": false, + "reason": "Different transformer test groups and different models." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13347", + "accept": false, + "reason": "Different transformer test groups and different models." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13348", + "accept": false, + "reason": "Different transformer test groups and different models." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13347", + "accept": false, + "reason": "Different transformer test groups and different models." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13348", + "accept": false, + "reason": "Different transformer test groups and different models." + }, + { + "left": "pull_request:13341", + "right": "pull_request:13347", + "accept": false, + "reason": "Different transformer test groups and different models." + }, + { + "left": "pull_request:13341", + "right": "pull_request:13348", + "accept": false, + "reason": "Different transformer test groups and different models." + }, + { + "left": "pull_request:13344", + "right": "pull_request:13346", + "accept": false, + "reason": "Different model tests: GLM Image vs CogVideoX/CogView." + }, + { + "left": "pull_request:1420", + "right": "pull_request:178", + "accept": false, + "reason": "Both are Stable Diffusion tests, but they target different variants and different test goals." + }, + { + "left": "pull_request:1998", + "right": "pull_request:2839", + "accept": false, + "reason": "Unrelated README updates in different top-level/example docs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 245, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10183", + "pull_request:10279", + "pull_request:10288", + "pull_request:10323", + "pull_request:10568", + "pull_request:10781", + "pull_request:11059", + "pull_request:11243", + "pull_request:11516", + "pull_request:11571", + "pull_request:12215", + "pull_request:12301", + "pull_request:13082", + "pull_request:13119", + "pull_request:13253", + "pull_request:13255", + "pull_request:13346", + "pull_request:13347", + "pull_request:13348", + "pull_request:3168", + "pull_request:3676", + "pull_request:3923", + "pull_request:3941", + "pull_request:4513", + "pull_request:5093", + "pull_request:5220", + "pull_request:5626", + "pull_request:5671", + "pull_request:5871", + "pull_request:6532", + "pull_request:6534", + "pull_request:6555", + "pull_request:6589", + "pull_request:7542", + "pull_request:7544", + "pull_request:7984", + "pull_request:7988", + "pull_request:8571", + "pull_request:8572", + "pull_request:8755", + "pull_request:8926", + "pull_request:9152", + "pull_request:9257", + "pull_request:9318", + "pull_request:9473", + "pull_request:9553", + "pull_request:9993" + ], + "soft_pairs": [ + "pull_request:3168|pull_request:3676", + "pull_request:4513|pull_request:5671", + "pull_request:5626|pull_request:7544", + "pull_request:5671|pull_request:5871", + "pull_request:6532|pull_request:7984", + "pull_request:6532|pull_request:7988", + "pull_request:11516|pull_request:11571", + "pull_request:11059|pull_request:11243", + "pull_request:8755|pull_request:9993", + "pull_request:6532|pull_request:6589", + "pull_request:8571|pull_request:8572", + "pull_request:13119|pull_request:13255", + "pull_request:10183|pull_request:10288", + "pull_request:10183|pull_request:10781", + "pull_request:13082|pull_request:13346", + "pull_request:13253|pull_request:13347", + "pull_request:13253|pull_request:13348", + "pull_request:3923|pull_request:3941", + "pull_request:6534|pull_request:7984", + "pull_request:6534|pull_request:7988", + "pull_request:9152|pull_request:9553", + "pull_request:12215|pull_request:12301", + "pull_request:10279|pull_request:10323", + "pull_request:5626|pull_request:7542", + "pull_request:5093|pull_request:5220", + "pull_request:8755|pull_request:9473", + "pull_request:10568|pull_request:9318", + "pull_request:6555|pull_request:6589", + "pull_request:8926|pull_request:9257" + ], + "prepared_review_unit_hash": "fffb1cf07a5458f235ee7eaa913fb301f7b52231e85791b776d6154e93a03eb1", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 99334, + "estimated_input_tokens": 24834, + "estimated_eval_tokens": 49924 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mix of unrelated docs, tests, training-script, and model-support PRs. I only found a few true overlap pairs: doc-builder header tweaks, CPU offload test assertions, and quantization link fixes.", + "confidence": 0.72, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR for the whole set: the items split into several unrelated subclusters. Among the overlapping docs-link fixes, PR 10279 is the cleanest representative.", + "best_issue_reason": null, + "best_pr_reason": "PR 10279 is the strongest representative of the small overlap subcluster because it fixes the same quantization overview links as 10323 and also includes the related GGUF doc correction.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3168", + "right": "pull_request:3676", + "accept": true, + "reason": "Same doc-builder WIP on the same docs index header change; same underlying docs-only tweak." + }, + { + "left": "pull_request:4513", + "right": "pull_request:5671", + "accept": false, + "reason": "Unrelated changes: safetensors/version pinning in setup/deps vs dependency-test workflow debugging." + }, + { + "left": "pull_request:5626", + "right": "pull_request:7544", + "accept": true, + "reason": "Both fix the CPU offload test assertions in test_pipelines_common.py; same underlying offload-test bug." + }, + { + "left": "pull_request:5671", + "right": "pull_request:5871", + "accept": false, + "reason": "Different targets: dependency-test workflow tweaks vs deprecated-scheduler doc/test rebasing." + }, + { + "left": "pull_request:6532", + "right": "pull_request:7984", + "accept": false, + "reason": "Different code paths: Dreambooth compile-compatibility vs VAE compile-compatibility." + }, + { + "left": "pull_request:6532", + "right": "pull_request:7988", + "accept": false, + "reason": "Different code paths and model surfaces: Dreambooth training script vs IPAdapter compile compatibility." + }, + { + "left": "pull_request:11516", + "right": "pull_request:11571", + "accept": false, + "reason": "Related project, but not the same change: initial LTX 0.9.7 support vs later distilled-model addition and docs." + }, + { + "left": "pull_request:11059", + "right": "pull_request:11243", + "accept": false, + "reason": "Same converter file, but different WAN LoRA formats (t2v vs musubi) and different concrete key layouts." + }, + { + "left": "pull_request:8755", + "right": "pull_request:9993", + "accept": false, + "reason": "Separate SD3 Dreambooth LoRA fixes: text-encoder training bug vs checkpoint-saving bug." + }, + { + "left": "pull_request:6532", + "right": "pull_request:6589", + "accept": false, + "reason": "Both are compile-related, but they fix different training scripts and different unwrap/save behavior." + }, + { + "left": "pull_request:8571", + "right": "pull_request:8572", + "accept": false, + "reason": "Both are broad typo/style errata, but they touch different docs and code areas; not the same underlying change." + }, + { + "left": "pull_request:13119", + "right": "pull_request:13255", + "accept": false, + "reason": "Different model-specific LoRA loading fixes: Flux2 key handling vs Z-Image loading." + }, + { + "left": "pull_request:10183", + "right": "pull_request:10288", + "accept": false, + "reason": "Different single-file support additions for different models (AutoencoderDC vs Flux variants)." + }, + { + "left": "pull_request:10183", + "right": "pull_request:10781", + "accept": false, + "reason": "Different single-file support additions for different models (AutoencoderDC vs Lumina2 transformer)." + }, + { + "left": "pull_request:13082", + "right": "pull_request:13346", + "accept": false, + "reason": "Unrelated test-refactor PRs for different transformer families; same broad theme, not same bug/change." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13347", + "accept": false, + "reason": "Different test refactors for different transformer families; not the same underlying change." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13348", + "accept": false, + "reason": "Different test refactors for different transformer families; not the same underlying change." + }, + { + "left": "pull_request:3923", + "right": "pull_request:3941", + "accept": false, + "reason": "Different documentation fixes on different pages; both are torch-import/doc-link cleanups, but not the same change." + }, + { + "left": "pull_request:6534", + "right": "pull_request:7984", + "accept": false, + "reason": "Dreambooth LoRA compile compatibility vs VAE compile compatibility; separate fixes." + }, + { + "left": "pull_request:6534", + "right": "pull_request:7988", + "accept": false, + "reason": "Dreambooth LoRA compile compatibility vs IPAdapter compile compatibility; separate fixes." + }, + { + "left": "pull_request:9152", + "right": "pull_request:9553", + "accept": false, + "reason": "Same docs file and topic, but different distributed-inference enhancements rather than one shared fix." + }, + { + "left": "pull_request:12215", + "right": "pull_request:12301", + "accept": false, + "reason": "Qwen-Image ControlNet vs ControlNet-inpaint are different features and code paths." + }, + { + "left": "pull_request:10279", + "right": "pull_request:10323", + "accept": true, + "reason": "Both make the same quantization overview link corrections; 10279 just includes an extra related GGUF docs change." + }, + { + "left": "pull_request:5626", + "right": "pull_request:7542", + "accept": false, + "reason": "Same offload test file, but different assertions: CPU-offload device check vs a sequential-offload typo fix." + }, + { + "left": "pull_request:5093", + "right": "pull_request:5220", + "accept": false, + "reason": "Both move slow tests to nightly, but they cover different test batches and are not the same change." + }, + { + "left": "pull_request:8755", + "right": "pull_request:9473", + "accept": false, + "reason": "Different SD3 Dreambooth LoRA fixes in the same script family; not the same bug." + }, + { + "left": "pull_request:10568", + "right": "pull_request:9318", + "accept": false, + "reason": "Both update HF token usage in CI, but one is fast GPU tests and the other is nightly tests; separate workflows." + }, + { + "left": "pull_request:6555", + "right": "pull_request:6589", + "accept": false, + "reason": "Different scripts and different fixes; same compile-compatibility theme only at a very high level." + }, + { + "left": "pull_request:8926", + "right": "pull_request:9257", + "accept": false, + "reason": "Different Dreambooth/Flux fixes touching different scripts and failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 246, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10268", + "pull_request:10288", + "pull_request:10437", + "pull_request:10508", + "pull_request:10781", + "pull_request:1087", + "pull_request:10947", + "pull_request:11058", + "pull_request:11386", + "pull_request:11391", + "pull_request:1210", + "pull_request:1218", + "pull_request:12236", + "pull_request:12387", + "pull_request:12933", + "pull_request:13217", + "pull_request:1510", + "pull_request:2193", + "pull_request:2257", + "pull_request:2271", + "pull_request:2400", + "pull_request:2740", + "pull_request:2741", + "pull_request:2751", + "pull_request:2846", + "pull_request:2924", + "pull_request:2939", + "pull_request:3472", + "pull_request:3525", + "pull_request:4498", + "pull_request:4499", + "pull_request:4500", + "pull_request:4501", + "pull_request:4502", + "pull_request:4946", + "pull_request:5267", + "pull_request:5855", + "pull_request:6526", + "pull_request:7934", + "pull_request:7984", + "pull_request:7988", + "pull_request:8452", + "pull_request:8490", + "pull_request:8860", + "pull_request:888", + "pull_request:9255", + "pull_request:9355", + "pull_request:9401" + ], + "soft_pairs": [ + "pull_request:12933|pull_request:13217", + "pull_request:10437|pull_request:11058", + "pull_request:10268|pull_request:10288", + "pull_request:10508|pull_request:9401", + "pull_request:10781|pull_request:10947", + "pull_request:2257|pull_request:2400", + "pull_request:3472|pull_request:4946", + "pull_request:6526|pull_request:7984", + "pull_request:6526|pull_request:7988", + "pull_request:9255|pull_request:9355", + "pull_request:7934|pull_request:8452", + "pull_request:2193|pull_request:888", + "pull_request:12236|pull_request:12387", + "pull_request:1087|pull_request:1210", + "pull_request:1087|pull_request:2271", + "pull_request:1087|pull_request:2846", + "pull_request:11386|pull_request:11391", + "pull_request:1210|pull_request:2271", + "pull_request:1210|pull_request:2846", + "pull_request:1218|pull_request:1510", + "pull_request:2271|pull_request:2846", + "pull_request:2740|pull_request:2741", + "pull_request:2740|pull_request:2751", + "pull_request:2741|pull_request:2751", + "pull_request:2924|pull_request:2939", + "pull_request:3525|pull_request:5267", + "pull_request:3525|pull_request:5855", + "pull_request:5267|pull_request:5855", + "pull_request:8490|pull_request:8860", + "pull_request:4498|pull_request:4499", + "pull_request:4498|pull_request:4500", + "pull_request:4498|pull_request:4501", + "pull_request:4498|pull_request:4502" + ], + "prepared_review_unit_hash": "67c067db2c909601ae2bb2fd223c017f0db4506f29c5208ce80caaeb13f62ced", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 101307, + "estimated_input_tokens": 25327, + "estimated_eval_tokens": 50910 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mix of unrelated PRs that happen to share broad subsystems or doc/test themes. The only clear duplicate set is the SDXL reference-pipeline PRs, plus one near-duplicate text-inversion docs edit; most other soft edges are false positives because they touch different concrete features or code paths.", + "confidence": 0.71, + "canonical_issue_reason": null, + "canonical_pr_reason": "4502 is the clearest merged final version of the repeated SDXL reference-pipeline PRs and the strongest representative of the only obvious duplicate group in this set.", + "best_issue_reason": null, + "best_pr_reason": "4502 has the most complete implementation and discussion among the true duplicate candidates, so it is the best PR anchor here.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12933", + "right": "pull_request:13217", + "accept": false, + "reason": "Different features: LTX-2 LoRA support vs LTX-2.3 model support. Same area, but not the same underlying change." + }, + { + "left": "pull_request:10437", + "right": "pull_request:11058", + "accept": false, + "reason": "Both are test-infra refactors, but one adds xpu/device-agnostic support while the other adds multi-accelerator decorators and offload tweaks. Not the same concrete change." + }, + { + "left": "pull_request:10268", + "right": "pull_request:10288", + "accept": false, + "reason": "Both extend single-file support, but for different models (Mochi vs Flux Canny/Depth/Fill)." + }, + { + "left": "pull_request:10508", + "right": "pull_request:9401", + "accept": false, + "reason": "Different test suites and different device-agnostic changes; shared testing utilities are not enough to make this a duplicate." + }, + { + "left": "pull_request:10781", + "right": "pull_request:10947", + "accept": false, + "reason": "Different model support additions: Lumina Image 2.0 vs SANA Transformer." + }, + { + "left": "pull_request:2257", + "right": "pull_request:2400", + "accept": false, + "reason": "Related community pipelines, but one is text interpolation and the other is image interpolation. Different pipeline changes." + }, + { + "left": "pull_request:3472", + "right": "pull_request:4946", + "accept": false, + "reason": "Different docs updates for different training topics; not the same change." + }, + { + "left": "pull_request:6526", + "right": "pull_request:7984", + "accept": false, + "reason": "Different torch.compile fixes in different code paths: ControlNet SDXL training vs VAE decode." + }, + { + "left": "pull_request:6526", + "right": "pull_request:7988", + "accept": false, + "reason": "Different compile-compatibility work in different models; same theme, not the same fix." + }, + { + "left": "pull_request:9255", + "right": "pull_request:9355", + "accept": false, + "reason": "Same workflow file, but one changes branch triggering and naming while the other adds workflow_dispatch. Separate changes." + }, + { + "left": "pull_request:7934", + "right": "pull_request:8452", + "accept": false, + "reason": "Both are single-file loading fixes, but one handles dict config loading and the other relaxes state_dict strictness. Different bugs." + }, + { + "left": "pull_request:2193", + "right": "pull_request:888", + "accept": false, + "reason": "Different pipelines entirely: SAG Stable Diffusion vs Cycle Diffusion." + }, + { + "left": "pull_request:12236", + "right": "pull_request:12387", + "accept": false, + "reason": "Both involve kernels/flash attention, but they target different backend capabilities and code paths." + }, + { + "left": "pull_request:1087", + "right": "pull_request:1210", + "accept": false, + "reason": "Both are slow-test fixes, but they touch different tests and expected outputs." + }, + { + "left": "pull_request:1087", + "right": "pull_request:2271", + "accept": false, + "reason": "Different slow-test adjustments in different files and pipelines." + }, + { + "left": "pull_request:1087", + "right": "pull_request:2846", + "accept": false, + "reason": "Different test fixes with only superficial overlap in the 'Fix slow tests' theme." + }, + { + "left": "pull_request:11386", + "right": "pull_request:11391", + "accept": false, + "reason": "Same filename/title, but 11391 has no diff/files shown here, so there isn't enough evidence it is the same concrete change." + }, + { + "left": "pull_request:1210", + "right": "pull_request:2271", + "accept": false, + "reason": "Both are test-output updates, but they modify different pipelines/schedulers and are not the same change." + }, + { + "left": "pull_request:1210", + "right": "pull_request:2846", + "accept": false, + "reason": "Different slow-test fixes in different test modules." + }, + { + "left": "pull_request:1218", + "right": "pull_request:1510", + "accept": false, + "reason": "Both touch conversion tooling, but one adds scheduler options while the other improves scheduler/documentation coverage. Different PRs." + }, + { + "left": "pull_request:2271", + "right": "pull_request:2846", + "accept": false, + "reason": "Different slow-test fixes for different test files." + }, + { + "left": "pull_request:2740", + "right": "pull_request:2741", + "accept": false, + "reason": "2740 includes an additional DreamBooth docs edit, so it is not the same change as the text-inversion-only PR." + }, + { + "left": "pull_request:2740", + "right": "pull_request:2751", + "accept": false, + "reason": "2740 spans DreamBooth plus text inversion; 2751 is only the text-inversion docs edit." + }, + { + "left": "pull_request:2741", + "right": "pull_request:2751", + "accept": true, + "reason": "Same text_inversion.md wording fix on the same sentence; 2751 looks like the later merged revision of the same docs change." + }, + { + "left": "pull_request:2924", + "right": "pull_request:2939", + "accept": false, + "reason": "Both are tiny docs tweaks for VQ Diffusion, but they modify different docs pages and are not the same edit." + }, + { + "left": "pull_request:3525", + "right": "pull_request:5267", + "accept": false, + "reason": "Both are README edits, but they address different content and different sections." + }, + { + "left": "pull_request:3525", + "right": "pull_request:5855", + "accept": false, + "reason": "Unrelated README edits; same file alone is not enough." + }, + { + "left": "pull_request:5267", + "right": "pull_request:5855", + "accept": false, + "reason": "Different documentation fixes in the same README, not the same change." + }, + { + "left": "pull_request:8490", + "right": "pull_request:8860", + "accept": false, + "reason": "Same README filename, but the previews indicate different SD3 README edits and not a single duplicate change." + }, + { + "left": "pull_request:4498", + "right": "pull_request:4499", + "accept": true, + "reason": "Same SDXL reference-pipeline feature on the same files; effectively the same change." + }, + { + "left": "pull_request:4498", + "right": "pull_request:4500", + "accept": true, + "reason": "Same SDXL reference-pipeline implementation and docs, just a parallel PR variant." + }, + { + "left": "pull_request:4498", + "right": "pull_request:4501", + "accept": true, + "reason": "Same underlying SDXL reference-pipeline addition with the same touched files." + }, + { + "left": "pull_request:4498", + "right": "pull_request:4502", + "accept": true, + "reason": "Same SDXL reference-pipeline feature; 4502 is the merged final variant of the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 247, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10464", + "pull_request:10465", + "pull_request:10678", + "pull_request:10784", + "pull_request:10849", + "pull_request:10852", + "pull_request:1087", + "pull_request:1131", + "pull_request:11407", + "pull_request:11416", + "pull_request:11476", + "pull_request:1210", + "pull_request:13071", + "pull_request:13082", + "pull_request:1425", + "pull_request:17", + "pull_request:2271", + "pull_request:2526", + "pull_request:3169", + "pull_request:344", + "pull_request:365", + "pull_request:3676", + "pull_request:3687", + "pull_request:386", + "pull_request:3863", + "pull_request:388", + "pull_request:408", + "pull_request:5201", + "pull_request:5439", + "pull_request:5440", + "pull_request:5506", + "pull_request:6261", + "pull_request:6262", + "pull_request:6263", + "pull_request:6264", + "pull_request:6265", + "pull_request:6962", + "pull_request:703", + "pull_request:705", + "pull_request:7131", + "pull_request:7451", + "pull_request:831", + "pull_request:8720", + "pull_request:8946", + "pull_request:8992", + "pull_request:9040", + "pull_request:9131" + ], + "soft_pairs": [ + "pull_request:8992|pull_request:9040", + "pull_request:3676|pull_request:3687", + "pull_request:703|pull_request:705", + "pull_request:5439|pull_request:5440", + "pull_request:11407|pull_request:11416", + "pull_request:11416|pull_request:11476", + "pull_request:3169|pull_request:3863", + "pull_request:6962|pull_request:8720", + "pull_request:344|pull_request:365", + "pull_request:6262|pull_request:6265", + "pull_request:6263|pull_request:6265", + "pull_request:6264|pull_request:6265", + "pull_request:10678|pull_request:10784", + "pull_request:8946|pull_request:9131", + "pull_request:6261|pull_request:6265", + "pull_request:386|pull_request:388", + "pull_request:5201|pull_request:5506", + "pull_request:10464|pull_request:10465", + "pull_request:7131|pull_request:7451", + "pull_request:10849|pull_request:10852", + "pull_request:1087|pull_request:2526", + "pull_request:1131|pull_request:831", + "pull_request:1210|pull_request:2526", + "pull_request:13071|pull_request:13082", + "pull_request:1425|pull_request:408", + "pull_request:17|pull_request:408", + "pull_request:2271|pull_request:2526" + ], + "prepared_review_unit_hash": "945003d9d172f514de6ded6d9d806a3c733d576dab2427ae3bc563d6a60e8758", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 103573, + "estimated_input_tokens": 25894, + "estimated_eval_tokens": 52044 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "A few pairs are clear duplicates or follow-up fixes in the same file/change, but most are merely similar PRs in the same subsystem and should stay separate.", + "confidence": 0.71, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10784 is the cleanest canonical representative: it is merged, issue-linked, and contains a tight self-contained fix to the same validation bug.", + "best_issue_reason": null, + "best_pr_reason": "PR 10784 is the best overall candidate because it is a focused merged fix with an explicit issue target and minimal scope.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8992", + "right": "pull_request:9040", + "accept": false, + "reason": "Same feature area, but one is the implementation PR and the other is a follow-up/community README update; not the same concrete change." + }, + { + "left": "pull_request:3676", + "right": "pull_request:3687", + "accept": true, + "reason": "Same file and identical doc-builder test edit; these look like duplicate/overlapping PRs for the same docs title change." + }, + { + "left": "pull_request:703", + "right": "pull_request:705", + "accept": false, + "reason": "Both touch CI CUDA testing, but they change different workflows and different failure modes." + }, + { + "left": "pull_request:5439", + "right": "pull_request:5440", + "accept": false, + "reason": "Related single-file loading fixes, but one is config/tokenizer loading while the other is local_files_only propagation." + }, + { + "left": "pull_request:11407", + "right": "pull_request:11416", + "accept": false, + "reason": "Both are typo cleanup PRs, but they cover different text sets and docs sections rather than one shared fix." + }, + { + "left": "pull_request:11416", + "right": "pull_request:11476", + "accept": false, + "reason": "Different typo batches across docs versus code/tests; same cleanup style, not the same change." + }, + { + "left": "pull_request:3169", + "right": "pull_request:3863", + "accept": false, + "reason": "Same Habana doc file, but the versions/config updates are different follow-on documentation revisions." + }, + { + "left": "pull_request:6962", + "right": "pull_request:8720", + "accept": false, + "reason": "Both target flaky tests, but they fix different pipelines with different assertions and root causes." + }, + { + "left": "pull_request:344", + "right": "pull_request:365", + "accept": true, + "reason": "Same VAE model file and same type-hinting effort; 365 looks like a narrow follow-up to the broader PR." + }, + { + "left": "pull_request:6262", + "right": "pull_request:6265", + "accept": false, + "reason": "Same refactor theme, but these move different pipelines out of stable_diffusion, so they are not the same change." + }, + { + "left": "pull_request:6263", + "right": "pull_request:6265", + "accept": false, + "reason": "Same refactor theme, but different pipeline extraction work in different modules." + }, + { + "left": "pull_request:6264", + "right": "pull_request:6265", + "accept": false, + "reason": "Same refactor theme, but this is a separate pipeline move, not the same concrete change." + }, + { + "left": "pull_request:10678", + "right": "pull_request:10784", + "accept": true, + "reason": "Both patch the same height/width divisibility validation bug pattern with nearly identical code in different pipelines." + }, + { + "left": "pull_request:11407", + "right": "pull_request:11416", + "accept": false, + "reason": "Docs typo batches overlap in style only; they are not the same text fix." + }, + { + "left": "pull_request:11416", + "right": "pull_request:11476", + "accept": false, + "reason": "One is docs/comments typos, the other is example/code string typos; too different to merge as duplicates." + }, + { + "left": "pull_request:10849", + "right": "pull_request:10852", + "accept": true, + "reason": "Same remote VAE issue template file, with 10852 clearly a follow-up fix to the template introduced in 10849." + }, + { + "left": "pull_request:1087", + "right": "pull_request:2526", + "accept": false, + "reason": "Both are slow-test fixes, but they touch different tests and different nondeterminism sources." + }, + { + "left": "pull_request:1131", + "right": "pull_request:831", + "accept": false, + "reason": "Both bump versions, but they are distinct release bumps to different versions." + }, + { + "left": "pull_request:1210", + "right": "pull_request:2526", + "accept": false, + "reason": "Both adjust test expectations, but they are separate flake fixes in unrelated pipelines." + }, + { + "left": "pull_request:13071", + "right": "pull_request:13082", + "accept": false, + "reason": "Both are model test refactors, but for different model families and code paths." + }, + { + "left": "pull_request:1425", + "right": "pull_request:408", + "accept": false, + "reason": "Both fix typos in docs, but they are separate content edits across different pages." + }, + { + "left": "pull_request:17", + "right": "pull_request:408", + "accept": false, + "reason": "General typo cleanup, but on different docs with no shared concrete bug." + }, + { + "left": "pull_request:2271", + "right": "pull_request:2526", + "accept": false, + "reason": "Both are test tolerance fixes, but they address different pipelines and numerical expectations." + }, + { + "left": "pull_request:5201", + "right": "pull_request:5506", + "accept": false, + "reason": "Same file, but one fixes LoRA offload/set_processor behavior while the other fixes xformers+PEFT scaling; different bugs." + }, + { + "left": "pull_request:10464", + "right": "pull_request:10465", + "accept": true, + "reason": "Same FLUX NPU adaptation theme and overlapping transformer/controlnet changes; 10465 reads like a narrowed duplicate/follow-up of 10464." + }, + { + "left": "pull_request:7131", + "right": "pull_request:7451", + "accept": false, + "reason": "Both are docs typo fixes, but they update different documentation areas and are not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 248, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1131", + "pull_request:12117", + "pull_request:12225", + "pull_request:12521", + "pull_request:12562", + "pull_request:13069", + "pull_request:13082", + "pull_request:13259", + "pull_request:13567", + "pull_request:1415", + "pull_request:1490", + "pull_request:2290", + "pull_request:2388", + "pull_request:2446", + "pull_request:2757", + "pull_request:2864", + "pull_request:2925", + "pull_request:325", + "pull_request:327", + "pull_request:333", + "pull_request:335", + "pull_request:349", + "pull_request:359", + "pull_request:360", + "pull_request:3625", + "pull_request:3676", + "pull_request:4330", + "pull_request:4472", + "pull_request:4612", + "pull_request:4652", + "pull_request:6017", + "pull_request:6018", + "pull_request:61", + "pull_request:63", + "pull_request:6647", + "pull_request:6909", + "pull_request:6910", + "pull_request:6939", + "pull_request:7903", + "pull_request:8064", + "pull_request:831", + "pull_request:8470", + "pull_request:8478", + "pull_request:8479", + "pull_request:912", + "pull_request:9131", + "pull_request:9760" + ], + "soft_pairs": [ + "pull_request:2864|pull_request:2925", + "pull_request:3625|pull_request:3676", + "pull_request:3676|pull_request:6017", + "pull_request:3676|pull_request:6018", + "pull_request:4330|pull_request:4472", + "pull_request:4652|pull_request:6647", + "pull_request:6017|pull_request:6018", + "pull_request:61|pull_request:63", + "pull_request:8470|pull_request:9760", + "pull_request:8478|pull_request:8479", + "pull_request:12117|pull_request:12225", + "pull_request:327|pull_request:335", + "pull_request:7903|pull_request:9131", + "pull_request:8064|pull_request:9131", + "pull_request:6910|pull_request:6939", + "pull_request:325|pull_request:333", + "pull_request:359|pull_request:360", + "pull_request:335|pull_request:349", + "pull_request:6909|pull_request:6939", + "pull_request:12521|pull_request:12562", + "pull_request:1131|pull_request:1490", + "pull_request:1131|pull_request:912", + "pull_request:13069|pull_request:13082", + "pull_request:13259|pull_request:13567", + "pull_request:1415|pull_request:2757", + "pull_request:1490|pull_request:831", + "pull_request:2290|pull_request:2446", + "pull_request:2388|pull_request:4612" + ], + "prepared_review_unit_hash": "10279bfafd289cc48f1c2e1dcc71988410ceac53c96736ae248e8e7ed8c63a1f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 96890, + "estimated_input_tokens": 24223, + "estimated_eval_tokens": 48702 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated PRs spanning docs tweaks, type-hint refactors, release bumps, and separate feature additions; only the HunyuanDiT controlnet/pipeline pair looks like an exact duplicate.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 8478 is the cleanest canonical choice because PR 8479 appears to be the same HunyuanDiT controlnet/pipeline patch duplicated; 8478 is the earlier, representative copy.", + "best_issue_reason": null, + "best_pr_reason": "PR 8478 best represents the only true duplicate-style pair in the set, and its content matches 8479 closely enough to serve as the canonical PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2864", + "right": "pull_request:2925", + "accept": false, + "reason": "Both are tiny docs punctuation tweaks, but they hit different overview pages and are not the same change." + }, + { + "left": "pull_request:3625", + "right": "pull_request:3676", + "accept": false, + "reason": "Both are doc-builder-related, but one changes workflow config while the other changes the docs index title; different fixes." + }, + { + "left": "pull_request:3676", + "right": "pull_request:6017", + "accept": false, + "reason": "Same broad docs test theme, but the edits are different title strings in separate PRs, not one underlying change." + }, + { + "left": "pull_request:3676", + "right": "pull_request:6018", + "accept": false, + "reason": "These are separate docs-index title tweaks with different test strings, not a single duplicated patch." + }, + { + "left": "pull_request:4330", + "right": "pull_request:4472", + "accept": false, + "reason": "Different README files and different content; they are not the same bug or change." + }, + { + "left": "pull_request:4652", + "right": "pull_request:6647", + "accept": false, + "reason": "Unrelated test fixes in different areas; sharing the word \"fix\" is not enough to make them duplicates." + }, + { + "left": "pull_request:6017", + "right": "pull_request:6018", + "accept": false, + "reason": "Both are doc-builder test edits to the same file family, but they are distinct title changes and not the same patch." + }, + { + "left": "pull_request:61", + "right": "pull_request:63", + "accept": false, + "reason": "Different conversion scripts for different pipelines and checkpoints; not the same underlying work." + }, + { + "left": "pull_request:8470", + "right": "pull_request:9760", + "accept": false, + "reason": "Both touch README text, but they are separate documentation updates for different content." + }, + { + "left": "pull_request:8478", + "right": "pull_request:8479", + "accept": true, + "reason": "These look like the same HunyuanDiT controlnet/pipeline addition duplicated almost verbatim." + }, + { + "left": "pull_request:12117", + "right": "pull_request:12225", + "accept": false, + "reason": "Related QwenImage work, but one adds img2img/inpaint while the other adds edit-inpaint; different concrete code paths." + }, + { + "left": "pull_request:327", + "right": "pull_request:335", + "accept": false, + "reason": "One is a pipeline type-hint refactor and the other is a scheduler type-hint refactor; not the same change." + }, + { + "left": "pull_request:7903", + "right": "pull_request:9131", + "accept": false, + "reason": "A video/image processor test refactor versus a dependency pin are unrelated changes." + }, + { + "left": "pull_request:8064", + "right": "pull_request:9131", + "accept": false, + "reason": "Marigold test changes and a setup dependency change are unrelated." + }, + { + "left": "pull_request:6910", + "right": "pull_request:6939", + "accept": false, + "reason": "Both standardize model-card code, but for different training scripts and different artifacts." + }, + { + "left": "pull_request:325", + "right": "pull_request:333", + "accept": false, + "reason": "Different pipelines, different files, and different type-hint refactors." + }, + { + "left": "pull_request:359", + "right": "pull_request:360", + "accept": false, + "reason": "Both are scheduler annotations, but they target different schedulers and are separate changes." + }, + { + "left": "pull_request:335", + "right": "pull_request:349", + "accept": false, + "reason": "Type-hint updates for different schedulers; not the same concrete bug or patch." + }, + { + "left": "pull_request:12521", + "right": "pull_request:12562", + "accept": false, + "reason": "Different Wan crash fixes: one is tiling decode, the other is context-parallel handling." + }, + { + "left": "pull_request:1131", + "right": "pull_request:1490", + "accept": false, + "reason": "Both are version/release-related, but they are different bumps with different accompanying code changes." + }, + { + "left": "pull_request:2290", + "right": "pull_request:2446", + "accept": false, + "reason": "Separate docs snippet fixes for different examples; not the same issue." + }, + { + "left": "pull_request:2388", + "right": "pull_request:4612", + "accept": false, + "reason": "One adds the controlling-generation doc page, the other fixes links within it; related but not duplicate." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 249, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10206", + "pull_request:10810", + "pull_request:1525", + "pull_request:2539", + "pull_request:2839", + "pull_request:3099", + "pull_request:3214", + "pull_request:326", + "pull_request:327", + "pull_request:333", + "pull_request:335", + "pull_request:343", + "pull_request:3525", + "pull_request:359", + "pull_request:360", + "pull_request:3709", + "pull_request:3710", + "pull_request:3819", + "pull_request:4078", + "pull_request:4132", + "pull_request:4451", + "pull_request:5220", + "pull_request:5267", + "pull_request:5280", + "pull_request:5292", + "pull_request:5420", + "pull_request:5855", + "pull_request:6327", + "pull_request:6342", + "pull_request:6346", + "pull_request:6356", + "pull_request:6438", + "pull_request:6453", + "pull_request:6462", + "pull_request:6617", + "pull_request:6620", + "pull_request:7118", + "pull_request:7246", + "pull_request:7451", + "pull_request:7544", + "pull_request:7767", + "pull_request:7825", + "pull_request:7827", + "pull_request:7966", + "pull_request:831", + "pull_request:8775", + "pull_request:912", + "pull_request:9760" + ], + "soft_pairs": [ + "pull_request:2539|pull_request:3099", + "pull_request:2839|pull_request:3525", + "pull_request:3525|pull_request:3819", + "pull_request:3709|pull_request:3710", + "pull_request:3819|pull_request:5267", + "pull_request:3819|pull_request:5855", + "pull_request:4451|pull_request:5220", + "pull_request:5280|pull_request:5292", + "pull_request:6342|pull_request:6438", + "pull_request:6453|pull_request:6462", + "pull_request:6617|pull_request:6620", + "pull_request:7118|pull_request:7451", + "pull_request:7544|pull_request:7827", + "pull_request:7767|pull_request:7966", + "pull_request:831|pull_request:912", + "pull_request:8775|pull_request:9760", + "pull_request:6327|pull_request:6356", + "pull_request:326|pull_request:335", + "pull_request:4078|pull_request:4132", + "pull_request:343|pull_request:359", + "pull_request:343|pull_request:360", + "pull_request:10206|pull_request:10810", + "pull_request:1525|pull_request:3214", + "pull_request:5420|pull_request:6346", + "pull_request:7246|pull_request:7825", + "pull_request:327|pull_request:333" + ], + "prepared_review_unit_hash": "1863e2ec5701a907aaa9ce1654ae50d4fc5af4fc64e710efcb9b112bbad2709d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 96056, + "estimated_input_tokens": 24014, + "estimated_eval_tokens": 48284 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous mix of unrelated docs, tests, version bumps, and small code fixes. Only a few pairs are true duplicates around the exact same documentation change; most other soft edges should be rejected under strict PR triage.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits the whole cluster: the items span unrelated edits across docs, tests, schedulers, LoRA utilities, and examples. The cluster is too mixed for one representative PR to safely stand in for the rest.", + "best_issue_reason": null, + "best_pr_reason": "No strong global best PR; if forced, the cleanest representative is one of the accepted doc-fix PRs, but the overall cluster is not cohesive enough to justify a single canonical PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2539", + "right": "pull_request:3099", + "accept": false, + "reason": "Same broad theme of default-value docs, but they touch different pipeline files and different parameters; not the same underlying fix." + }, + { + "left": "pull_request:2839", + "right": "pull_request:3525", + "accept": false, + "reason": "Both update README content, but they are different files and different changes (examples table vs general README copy)." + }, + { + "left": "pull_request:3525", + "right": "pull_request:3819", + "accept": false, + "reason": "Unrelated README edits in different locations; one is a general README rewrite, the other adds an FAQ section in examples." + }, + { + "left": "pull_request:3709", + "right": "pull_request:3710", + "accept": true, + "reason": "Same new documentation page and same toctree wiring; this is the same underlying doc addition with minor follow-up wording changes." + }, + { + "left": "pull_request:3819", + "right": "pull_request:5267", + "accept": false, + "reason": "Different documentation areas and different intent; FAQ expansion in examples is not the same change as README formatting guidance." + }, + { + "left": "pull_request:3819", + "right": "pull_request:5855", + "accept": false, + "reason": "Different README and different content; one is a FAQ addition, the other is a typo fix in a community example." + }, + { + "left": "pull_request:4451", + "right": "pull_request:5220", + "accept": false, + "reason": "Both retag tests, but they affect different test suites and different subsets; not the same concrete change." + }, + { + "left": "pull_request:5280", + "right": "pull_request:5292", + "accept": false, + "reason": "Both speed up ControlNet-related tests, but they target different test files and different setups, so they are not the same fix." + }, + { + "left": "pull_request:6342", + "right": "pull_request:6438", + "accept": true, + "reason": "Same installation doc and same Windows/PyTorch version-compatibility note; this is effectively the same doc fix with refined wording." + }, + { + "left": "pull_request:6453", + "right": "pull_request:6462", + "accept": true, + "reason": "Same file and same aMUSEd paper-link correction; the second PR updates the exact same documentation bug." + }, + { + "left": "pull_request:6617", + "right": "pull_request:6620", + "accept": false, + "reason": "Insufficient evidence on the left artifact and the visible right diff is a docs-arguments fix; not enough to conclude the same change." + }, + { + "left": "pull_request:7118", + "right": "pull_request:7451", + "accept": false, + "reason": "Both are typo/doc cleanup PRs, but they span different files and broad documentation sets, so they are not one concrete change." + }, + { + "left": "pull_request:7544", + "right": "pull_request:7827", + "accept": false, + "reason": "Different bug classes: CPU offload test assertions versus a docstring example fix." + }, + { + "left": "pull_request:7767", + "right": "pull_request:7966", + "accept": false, + "reason": "One changes PixArt pipeline code, the other updates PixArt Sigma tests; related area but not the same underlying bug or patch." + }, + { + "left": "pull_request:831", + "right": "pull_request:912", + "accept": false, + "reason": "Both are versioning/type-hint-era maintenance, but they correspond to different release bumps and different scheduler edits; not the same PR-worthy change." + }, + { + "left": "pull_request:8775", + "right": "pull_request:9760", + "accept": false, + "reason": "Same top-level README only in the broadest sense; the content changes are unrelated." + }, + { + "left": "pull_request:6327", + "right": "pull_request:6356", + "accept": false, + "reason": "Overlapping advanced SDXL LoRA script area, but they fix different issues and include different code paths; too broad to merge as duplicates." + }, + { + "left": "pull_request:326", + "right": "pull_request:335", + "accept": false, + "reason": "Both add typing to PNDM-related code, but one is pipeline typing and the other is scheduler typing; not the same change." + }, + { + "left": "pull_request:4078", + "right": "pull_request:4132", + "accept": false, + "reason": "Opposite behaviors in the same file: one forces accelerate to be present, the other makes accelerate optional." + }, + { + "left": "pull_request:343", + "right": "pull_request:359", + "accept": false, + "reason": "Both are scheduler typing cleanups, but they affect different schedulers and code paths." + }, + { + "left": "pull_request:343", + "right": "pull_request:360", + "accept": false, + "reason": "Same as above; different schedulers, so not the same underlying fix." + }, + { + "left": "pull_request:10206", + "right": "pull_request:10810", + "accept": false, + "reason": "Both are LoRA/Flux-related, but one is unload logic and the other is conversion support; related subsystem, different concrete changes." + }, + { + "left": "pull_request:1525", + "right": "pull_request:3214", + "accept": false, + "reason": "Different bugs in different components: batch sizing in the upscaler versus attention head reshaping." + }, + { + "left": "pull_request:5420", + "right": "pull_request:6346", + "accept": false, + "reason": "Both touch training examples, but they address different example scripts and different behaviors; not a single mergeable fix." + }, + { + "left": "pull_request:7246", + "right": "pull_request:7825", + "accept": false, + "reason": "Different test scopes and different failure modes; one is VAE tiling, the other is device-map/offload coverage." + }, + { + "left": "pull_request:327", + "right": "pull_request:333", + "accept": false, + "reason": "Both are typing updates for different pipelines, but they are separate code paths and not one duplicate change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 250, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10319", + "pull_request:10323", + "pull_request:10434", + "pull_request:11257", + "pull_request:11701", + "pull_request:11879", + "pull_request:11895", + "pull_request:12490", + "pull_request:12675", + "pull_request:12883", + "pull_request:1291", + "pull_request:13254", + "pull_request:13255", + "pull_request:13293", + "pull_request:13313", + "pull_request:13315", + "pull_request:13335", + "pull_request:13338", + "pull_request:13340", + "pull_request:13341", + "pull_request:13352", + "pull_request:1415", + "pull_request:1423", + "pull_request:1490", + "pull_request:1508", + "pull_request:1564", + "pull_request:178", + "pull_request:1998", + "pull_request:2406", + "pull_request:2569", + "pull_request:309", + "pull_request:325", + "pull_request:327", + "pull_request:3470", + "pull_request:3472", + "pull_request:3525", + "pull_request:4188", + "pull_request:4285", + "pull_request:4612", + "pull_request:555", + "pull_request:6119", + "pull_request:6346", + "pull_request:7166", + "pull_request:889", + "pull_request:9040", + "pull_request:912", + "pull_request:9423", + "pull_request:9765" + ], + "soft_pairs": [ + "pull_request:13255|pull_request:13313", + "pull_request:4188|pull_request:6346", + "pull_request:10323|pull_request:10434", + "pull_request:11879|pull_request:11895", + "pull_request:12490|pull_request:13293", + "pull_request:13338|pull_request:13352", + "pull_request:1508|pull_request:1564", + "pull_request:6119|pull_request:7166", + "pull_request:9040|pull_request:9423", + "pull_request:10319|pull_request:9765", + "pull_request:11257|pull_request:11701", + "pull_request:12675|pull_request:12883", + "pull_request:1291|pull_request:1415", + "pull_request:13254|pull_request:13340", + "pull_request:13315|pull_request:13340", + "pull_request:13335|pull_request:13340", + "pull_request:13340|pull_request:13341", + "pull_request:1415|pull_request:309", + "pull_request:1423|pull_request:555", + "pull_request:1490|pull_request:912", + "pull_request:178|pull_request:555", + "pull_request:178|pull_request:889", + "pull_request:1998|pull_request:3525", + "pull_request:2406|pull_request:4612", + "pull_request:2569|pull_request:4285", + "pull_request:325|pull_request:327", + "pull_request:3470|pull_request:3472" + ], + "prepared_review_unit_hash": "05f016a4814fbb95dac2d7bf1cbffa3e1331a7ae3328279c034292343d9f2843", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 96652, + "estimated_input_tokens": 24163, + "estimated_eval_tokens": 48582 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is highly heterogeneous: docs nits, workflow tweaks, test refactors, and unrelated bug fixes across different code paths. I don\u2019t see any soft pair that is truly the same underlying change, so I reject all duplicate links. If a single canonical PR is needed, #13255 is the cleanest concrete bug fix in a core utility.", + "confidence": 0.61, + "canonical_issue_reason": null, + "canonical_pr_reason": "#13255 is a small, focused code fix in a central loader utility, with a clear bug description and the strongest signal of a real standalone change among this mixed set.", + "best_issue_reason": null, + "best_pr_reason": "#13255 is the best representative PR here because it is a precise, mergeable bug fix with minimal scope and clearer code-path impact than the docs/test/workflow-only PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13255", + "right": "pull_request:13313", + "accept": false, + "reason": "Same file, but different LoRA-loading fixes and different bug shapes; not the same concrete change." + }, + { + "left": "pull_request:4188", + "right": "pull_request:6346", + "accept": false, + "reason": "Both touch training examples, but they fix different issues in different scripts and are not one mergeable bug fix." + }, + { + "left": "pull_request:10323", + "right": "pull_request:10434", + "accept": false, + "reason": "Both are docs-link cleanups, but they target different pages and different broken-link sets." + }, + { + "left": "pull_request:11879", + "right": "pull_request:11895", + "accept": false, + "reason": "Both remove deprecated tests, but they are separate cleanup batches for different pipelines." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13293", + "accept": false, + "reason": "These affect different files and different bugs; one is a test fixture change, the other is pipeline cache-position logic." + }, + { + "left": "pull_request:13338", + "right": "pull_request:13352", + "accept": false, + "reason": "Same workflow file, but one changes permissions and the other adds a checkout step; distinct fixes." + }, + { + "left": "pull_request:1508", + "right": "pull_request:1564", + "accept": false, + "reason": "Both are DreamBooth docs edits, but they cover unrelated documentation topics." + }, + { + "left": "pull_request:6119", + "right": "pull_request:7166", + "accept": false, + "reason": "Both are SDXL LoRA-related, but they fix different code paths: gradient scaling vs loading state dict into the correct text encoder." + }, + { + "left": "pull_request:9040", + "right": "pull_request:9423", + "accept": false, + "reason": "Different community pipelines and different features; not the same underlying change." + }, + { + "left": "pull_request:10319", + "right": "pull_request:9765", + "accept": false, + "reason": "Different pipeline docs and different docstring mistakes; no shared bug/change." + }, + { + "left": "pull_request:11257", + "right": "pull_request:11701", + "accept": false, + "reason": "Both edit the style-bot workflow token, but they point to different secrets and are not the same exact fix." + }, + { + "left": "pull_request:12675", + "right": "pull_request:12883", + "accept": false, + "reason": "Both are typo fixes in docs, but for unrelated quantization pages and different strings." + }, + { + "left": "pull_request:1291", + "right": "pull_request:1415", + "accept": false, + "reason": "One fixes a single README link; the other is a broad stable-diffusion cleanup PR." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13340", + "accept": false, + "reason": "Different transformer models and different test refactors; not the same bug or change." + }, + { + "left": "pull_request:13315", + "right": "pull_request:13340", + "accept": false, + "reason": "Different model test refactors with no shared concrete code-path problem." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13340", + "accept": false, + "reason": "Different transformer test suites; these are separate refactors." + }, + { + "left": "pull_request:13340", + "right": "pull_request:13341", + "accept": false, + "reason": "Different models and test code; not mergeable as one duplicate fix." + }, + { + "left": "pull_request:1415", + "right": "pull_request:309", + "accept": false, + "reason": "Both mention Stable Diffusion docs, but one is a broad cleanup and the other is a specific notebook link fix." + }, + { + "left": "pull_request:1423", + "right": "pull_request:555", + "accept": false, + "reason": "One adds SD2 Flax support, the other adds a Flax VAE; separate features, not duplicates." + }, + { + "left": "pull_request:1490", + "right": "pull_request:912", + "accept": false, + "reason": "Both are version/deprecation releases, but for different versions and unrelated release work." + }, + { + "left": "pull_request:178", + "right": "pull_request:555", + "accept": false, + "reason": "Different artifacts and different scope: a pipeline test vs a Flax model implementation." + }, + { + "left": "pull_request:178", + "right": "pull_request:889", + "accept": false, + "reason": "One is a pipeline test, the other adds components support; unrelated changes." + }, + { + "left": "pull_request:1998", + "right": "pull_request:3525", + "accept": false, + "reason": "Both touch README.md, but they are unrelated content changes." + }, + { + "left": "pull_request:2406", + "right": "pull_request:4612", + "accept": false, + "reason": "Both are docs in the controlling-generation area, but they fix different text and link issues." + }, + { + "left": "pull_request:2569", + "right": "pull_request:4285", + "accept": false, + "reason": "Different optimization docs and different runtime backends; not the same change." + }, + { + "left": "pull_request:325", + "right": "pull_request:327", + "accept": false, + "reason": "Different pipelines and different type-hint updates; not the same bug fix." + }, + { + "left": "pull_request:3470", + "right": "pull_request:3472", + "accept": false, + "reason": "Both add IF DreamBooth material, but one is the training docs and the other is the README; separate documentation updates." + }, + { + "left": "pull_request:13254", + "right": "pull_request:13340", + "accept": false, + "reason": "Different model test refactors with no shared concrete code-path problem." + }, + { + "left": "pull_request:13335", + "right": "pull_request:13341", + "accept": false, + "reason": "Different transformer test suites; these are separate refactors." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 251, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10021", + "pull_request:10330", + "pull_request:10378", + "pull_request:11428", + "pull_request:11534", + "pull_request:11913", + "pull_request:11995", + "pull_request:11997", + "pull_request:12086", + "pull_request:12660", + "pull_request:12909", + "pull_request:2464", + "pull_request:2795", + "pull_request:325", + "pull_request:3525", + "pull_request:359", + "pull_request:4262", + "pull_request:4351", + "pull_request:4451", + "pull_request:4472", + "pull_request:4574", + "pull_request:4597", + "pull_request:4694", + "pull_request:4749", + "pull_request:4833", + "pull_request:5003", + "pull_request:5093", + "pull_request:5201", + "pull_request:5267", + "pull_request:555", + "pull_request:5697", + "pull_request:5855", + "pull_request:6314", + "pull_request:6462", + "pull_request:6647", + "pull_request:6837", + "pull_request:6907", + "pull_request:6939", + "pull_request:7335", + "pull_request:7425", + "pull_request:8755", + "pull_request:8798", + "pull_request:889", + "pull_request:8913", + "pull_request:9249", + "pull_request:9860", + "pull_request:9893", + "pull_request:9938" + ], + "soft_pairs": [ + "pull_request:3525|pull_request:4351", + "pull_request:4351|pull_request:5267", + "pull_request:4351|pull_request:5855", + "pull_request:4472|pull_request:4574", + "pull_request:555|pull_request:889", + "pull_request:6314|pull_request:6462", + "pull_request:6647|pull_request:7335", + "pull_request:6837|pull_request:7425", + "pull_request:9249|pull_request:9860", + "pull_request:11913|pull_request:11997", + "pull_request:4262|pull_request:4749", + "pull_request:8798|pull_request:8913", + "pull_request:11995|pull_request:12086", + "pull_request:4451|pull_request:5093", + "pull_request:4833|pull_request:5003", + "pull_request:4597|pull_request:4694", + "pull_request:12660|pull_request:12909", + "pull_request:6907|pull_request:6939", + "pull_request:11428|pull_request:11534", + "pull_request:10378|pull_request:8755", + "pull_request:325|pull_request:359", + "pull_request:5201|pull_request:5697", + "pull_request:9893|pull_request:9938", + "pull_request:10021|pull_request:10330", + "pull_request:2464|pull_request:2795" + ], + "prepared_review_unit_hash": "47f6ab87fce74ed74f7e0bdd38c0204b8ce774197182beec70c426c7803868b3", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 102375, + "estimated_input_tokens": 25594, + "estimated_eval_tokens": 51444 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous feature/docs PRs, not a single duplicate set. Only two soft-edge pairs look like true same-change follow-ups: LoRA fuse/unfuse and sharded-checkpoint subfolder loading.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the items span unrelated features and docs, so there isn\u2019t one PR that cleanly represents the whole cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 8798 is the best anchor among the few real near-duplicates: it\u2019s a focused fix in the shared checkpoint-loading utility, has a direct follow-up pair, and includes test coverage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3525", + "right": "pull_request:4351", + "accept": false, + "reason": "Both touch README.md, but they are unrelated docs edits with different goals and no shared bug/change." + }, + { + "left": "pull_request:4351", + "right": "pull_request:5267", + "accept": false, + "reason": "Different documentation files and different edits; not the same underlying change." + }, + { + "left": "pull_request:4351", + "right": "pull_request:5855", + "accept": false, + "reason": "Generic README cleanup versus a community example typo fix; unrelated." + }, + { + "left": "pull_request:4472", + "right": "pull_request:4574", + "accept": false, + "reason": "Different example READMEs with different content changes; not the same change." + }, + { + "left": "pull_request:555", + "right": "pull_request:889", + "accept": false, + "reason": "Flax VAE addition versus DiffusionPipeline components support are distinct features." + }, + { + "left": "pull_request:6314", + "right": "pull_request:6462", + "accept": false, + "reason": "Both are paper-link doc fixes, but for different docs and different papers; not duplicates." + }, + { + "left": "pull_request:6647", + "right": "pull_request:7335", + "accept": false, + "reason": "One updates SDXL k-diffusion test outputs; the other adjusts ControlNet SDXL config/docs. Different problems." + }, + { + "left": "pull_request:6837", + "right": "pull_request:7425", + "accept": false, + "reason": "Different docs fixes in unrelated pipelines/tutorials." + }, + { + "left": "pull_request:9249", + "right": "pull_request:9860", + "accept": false, + "reason": "Initial SD3 ControlNet example versus later SD3.5 support extension; related area, but not the same concrete change." + }, + { + "left": "pull_request:11913", + "right": "pull_request:11997", + "accept": false, + "reason": "Both are Wan modular-pipeline work, but one introduces the base pipeline and the other adds I2V/FLF2V support; separate feature changes." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4749", + "accept": false, + "reason": "Both touch the SDXL DreamBooth LoRA script, but they address different issues and code paths." + }, + { + "left": "pull_request:8798", + "right": "pull_request:8913", + "accept": true, + "reason": "Same checkpoint-loading bug family: both fix sharded checkpoint loading from subfolders in the shared hub/model-loading path." + }, + { + "left": "pull_request:4451", + "right": "pull_request:5093", + "accept": false, + "reason": "Both move tests to nightly, but they affect different test suites and are not the same underlying issue." + }, + { + "left": "pull_request:4833", + "right": "pull_request:5003", + "accept": true, + "reason": "Same LoRA fuse/unfuse bug in the same code path and files; these are plausibly one merged fix." + }, + { + "left": "pull_request:4597", + "right": "pull_request:4694", + "accept": false, + "reason": "Same broad ControlNet area, but one is MultiControlNet SDXL support and the other is SDXL inpainting support; distinct features." + }, + { + "left": "pull_request:12660", + "right": "pull_request:12909", + "accept": false, + "reason": "Both tweak parallel_config handling, but they fix different model processors and different code paths." + }, + { + "left": "pull_request:6907", + "right": "pull_request:6939", + "accept": false, + "reason": "Both standardize model cards, but they are separate scripts and separate doc/model-card updates." + }, + { + "left": "pull_request:11428", + "right": "pull_request:11534", + "accept": false, + "reason": "Same family of Hunyuan/Framepack work, but the PRs target different variants and documentation additions rather than the same exact change." + }, + { + "left": "pull_request:10378", + "right": "pull_request:8755", + "accept": false, + "reason": "Both modify the SD3 DreamBooth LoRA script, but they solve different training/saving issues." + }, + { + "left": "pull_request:325", + "right": "pull_request:359", + "accept": false, + "reason": "Both are type-hint cleanups, but on unrelated files and APIs." + }, + { + "left": "pull_request:5201", + "right": "pull_request:5697", + "accept": false, + "reason": "Both involve LoRA/xformers behavior, but they fix different mechanisms in different code paths." + }, + { + "left": "pull_request:9893", + "right": "pull_request:9938", + "accept": false, + "reason": "Both update SDXL reference community docs, but they describe different pipelines and changes." + }, + { + "left": "pull_request:10021", + "right": "pull_request:10330", + "accept": false, + "reason": "Same LTX video model family, but one introduces the feature and the other updates it to 0.9.1; related follow-up, not a duplicate." + }, + { + "left": "pull_request:2464", + "right": "pull_request:2795", + "accept": false, + "reason": "Different training scripts and different example fixes; not the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 252, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10005", + "pull_request:10109", + "pull_request:10183", + "pull_request:10222", + "pull_request:10278", + "pull_request:10394", + "pull_request:1041", + "pull_request:10434", + "pull_request:10819", + "pull_request:1087", + "pull_request:10947", + "pull_request:1103", + "pull_request:1104", + "pull_request:1137", + "pull_request:11431", + "pull_request:11473", + "pull_request:11505", + "pull_request:11522", + "pull_request:11537", + "pull_request:11634", + "pull_request:11857", + "pull_request:121", + "pull_request:1210", + "pull_request:12272", + "pull_request:1230", + "pull_request:12442", + "pull_request:12487", + "pull_request:12490", + "pull_request:12526", + "pull_request:12775", + "pull_request:12856", + "pull_request:1291", + "pull_request:13253", + "pull_request:13340", + "pull_request:1437", + "pull_request:1467", + "pull_request:1998", + "pull_request:2271", + "pull_request:2362", + "pull_request:2396", + "pull_request:2757", + "pull_request:2766", + "pull_request:3127", + "pull_request:3166", + "pull_request:353", + "pull_request:5674", + "pull_request:6122", + "pull_request:9015" + ], + "soft_pairs": [ + "pull_request:11522|pull_request:12490", + "pull_request:12442|pull_request:12526", + "pull_request:11431|pull_request:11537", + "pull_request:5674|pull_request:6122", + "pull_request:1137|pull_request:3127", + "pull_request:10005|pull_request:10819", + "pull_request:10109|pull_request:10222", + "pull_request:10183|pull_request:10947", + "pull_request:10278|pull_request:11473", + "pull_request:10278|pull_request:9015", + "pull_request:10394|pull_request:11634", + "pull_request:1041|pull_request:2396", + "pull_request:10434|pull_request:11505", + "pull_request:1087|pull_request:1467", + "pull_request:1103|pull_request:3166", + "pull_request:1104|pull_request:2362", + "pull_request:11505|pull_request:12487", + "pull_request:11857|pull_request:12272", + "pull_request:121|pull_request:1998", + "pull_request:1210|pull_request:1467", + "pull_request:1230|pull_request:2766", + "pull_request:12775|pull_request:12856", + "pull_request:1291|pull_request:2757", + "pull_request:13253|pull_request:13340", + "pull_request:1437|pull_request:353", + "pull_request:1467|pull_request:2271" + ], + "prepared_review_unit_hash": "c852f9c8cd6002b9969422e5b06a763f3f5d941fcbea33359bdc0c045884d3c4", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 102079, + "estimated_input_tokens": 25520, + "estimated_eval_tokens": 51296 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated docs/test/infra fixes with one true continuation pair for Wan Animate. Accept only the Wan Animate continuation; the rest are too broad or cover different concrete changes.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12526 is the clearest representative: it explicitly continues #12442, targets the same Wan Animate implementation/docs/tests, and is the more complete follow-up.", + "best_issue_reason": null, + "best_pr_reason": "PR 12526 is the best overall anchor for this cluster because it is the largest substantive change here and the only pair with explicit continuation/overlap strong enough to justify duplicate-style grouping.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11522", + "right": "pull_request:12490", + "accept": false, + "reason": "Both are small docs edits, but one is an AudioLDM2 transformers-version compatibility fix and the other is a ClapConfig test cleanup; different problems and code paths." + }, + { + "left": "pull_request:12442", + "right": "pull_request:12526", + "accept": true, + "reason": "Same Wan Animate feature work: 12526 is explicitly a continuation of 12442 and overlaps heavily in files, tests, and implementation scope." + }, + { + "left": "pull_request:11431", + "right": "pull_request:11537", + "accept": false, + "reason": "Both add torch.compile-related tests, but they target different models and different test refactors; not the same concrete fix." + }, + { + "left": "pull_request:5674", + "right": "pull_request:6122", + "accept": false, + "reason": "Both are broad documentation typo passes, but they touch different docs sets and do not represent one shared change." + }, + { + "left": "pull_request:1137", + "right": "pull_request:3127", + "accept": false, + "reason": "Both involve safety checker behavior, but one fixes inpaint legacy dtype handling and the other fixes img2img safety-checker processing; distinct bugs." + }, + { + "left": "pull_request:10005", + "right": "pull_request:10819", + "accept": false, + "reason": "Both are docs updates, but one adds missing flux pipelines and the other adds LoRA loader entries; separate content." + }, + { + "left": "pull_request:10109", + "right": "pull_request:10222", + "accept": false, + "reason": "Both add torch_xla support, but in different pipelines and code paths; not mergeable as one PR." + }, + { + "left": "pull_request:10183", + "right": "pull_request:10947", + "accept": false, + "reason": "Both are single-file loading work, but they implement support for different models with different converters and tests." + }, + { + "left": "pull_request:10278", + "right": "pull_request:11473", + "accept": false, + "reason": "Same docs file, but different torchao doc fixes: one updates the example, the other fixes fp8 granular quantization naming." + }, + { + "left": "pull_request:10278", + "right": "pull_request:9015", + "accept": false, + "reason": "Both are docs/example updates, but for unrelated components and unrelated issues." + }, + { + "left": "pull_request:10394", + "right": "pull_request:11634", + "accept": false, + "reason": "Both are docs link fixes, but they affect different guides and different link targets; no shared underlying change." + }, + { + "left": "pull_request:1041", + "right": "pull_request:2396", + "accept": false, + "reason": "Both touch tests, but one is general failing-test cleanup and the other is UniPC-specific warning/trained_beta test fixes." + }, + { + "left": "pull_request:10434", + "right": "pull_request:11505", + "accept": false, + "reason": "Both fix docs links, but they are separate link sets across different doc areas and not one code/doc change." + }, + { + "left": "pull_request:1087", + "right": "pull_request:1467", + "accept": false, + "reason": "Both are test-stability PRs, but they patch different test suites and different failure modes; too broad to merge." + }, + { + "left": "pull_request:1103", + "right": "pull_request:3166", + "accept": false, + "reason": "Both are textual-inversion-related docs/code updates, but one fixes a README URL and the other fixes loader usage docs; different scope." + }, + { + "left": "pull_request:1104", + "right": "pull_request:2362", + "accept": false, + "reason": "One adds slow MPS tests, the other adds an MPS skip decorator; related area, but not the same fix." + }, + { + "left": "pull_request:11505", + "right": "pull_request:12487", + "accept": false, + "reason": "Both are docs link-fix PRs, but they cover different documentation families and different broken links." + }, + { + "left": "pull_request:11857", + "right": "pull_request:12272", + "accept": false, + "reason": "Both touch Flux docs/pipelines, but one fixes pipeline references while the other fixes modular pipeline height/width handling; different changes." + }, + { + "left": "pull_request:121", + "right": "pull_request:1998", + "accept": false, + "reason": "Both are README/example updates, but one adds usage examples to the main README and the other links a DreamBooth blog post; unrelated." + }, + { + "left": "pull_request:1210", + "right": "pull_request:1467", + "accept": false, + "reason": "Both are test fixes, but one is cycle-diffusion/scheduler reproducibility work and the other is stable-diffusion slow-test tolerance changes." + }, + { + "left": "pull_request:1230", + "right": "pull_request:2766", + "accept": false, + "reason": "Both address MPS-related test behavior, but one fixes generator handling in DDIM/DDPM tests and the other changes UNet block expectations for torch 2.0." + }, + { + "left": "pull_request:12775", + "right": "pull_request:12856", + "accept": false, + "reason": "The docs PR and the implementation PR are about Z-Image, but they are not the same artifact or concrete change; one is documentation, the other is model implementation." + }, + { + "left": "pull_request:1291", + "right": "pull_request:2757", + "accept": false, + "reason": "Both are Stable Diffusion-related, but one fixes a README link and the other fixes depth batching behavior; unrelated problems." + }, + { + "left": "pull_request:13253", + "right": "pull_request:13340", + "accept": false, + "reason": "Both refactor transformer tests, but for different models (Z-Image vs SD3) and different test mixins; not the same change." + }, + { + "left": "pull_request:1437", + "right": "pull_request:353", + "accept": false, + "reason": "Both are typo fixes, but in different files and on different messages; no shared bug/change." + }, + { + "left": "pull_request:1467", + "right": "pull_request:2271", + "accept": false, + "reason": "Both are slow-test maintenance, but they change different expected outputs in different test suites and were not one merged fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 253, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1467", + "pull_request:168", + "pull_request:178", + "pull_request:1937", + "pull_request:1998", + "pull_request:2271", + "pull_request:2362", + "pull_request:2383", + "pull_request:2446", + "pull_request:252", + "pull_request:2595", + "pull_request:2792", + "pull_request:2846", + "pull_request:2935", + "pull_request:3119", + "pull_request:3217", + "pull_request:353", + "pull_request:386", + "pull_request:395", + "pull_request:3987", + "pull_request:3989", + "pull_request:4145", + "pull_request:4175", + "pull_request:4472", + "pull_request:4495", + "pull_request:4540", + "pull_request:4652", + "pull_request:4797", + "pull_request:5051", + "pull_request:5065", + "pull_request:5066", + "pull_request:5332", + "pull_request:5411", + "pull_request:5844", + "pull_request:6122", + "pull_request:6775", + "pull_request:679", + "pull_request:6837", + "pull_request:688", + "pull_request:715", + "pull_request:7425", + "pull_request:7472", + "pull_request:8470", + "pull_request:8775", + "pull_request:8923", + "pull_request:8946", + "pull_request:948" + ], + "soft_pairs": [ + "pull_request:1467|pull_request:2846", + "pull_request:168|pull_request:178", + "pull_request:178|pull_request:386", + "pull_request:1937|pull_request:2595", + "pull_request:1998|pull_request:252", + "pull_request:2271|pull_request:3989", + "pull_request:2362|pull_request:2792", + "pull_request:2383|pull_request:3217", + "pull_request:2446|pull_request:3987", + "pull_request:2846|pull_request:3989", + "pull_request:2935|pull_request:3119", + "pull_request:353|pull_request:948", + "pull_request:395|pull_request:679", + "pull_request:4145|pull_request:4175", + "pull_request:4472|pull_request:4540", + "pull_request:4495|pull_request:5844", + "pull_request:4652|pull_request:4797", + "pull_request:5051|pull_request:6122", + "pull_request:5065|pull_request:5332", + "pull_request:5066|pull_request:5332", + "pull_request:5411|pull_request:7472", + "pull_request:6775|pull_request:6837", + "pull_request:6775|pull_request:7425", + "pull_request:688|pull_request:715", + "pull_request:8470|pull_request:8775", + "pull_request:8923|pull_request:8946" + ], + "prepared_review_unit_hash": "5d3cff56a9eef1f5a3e26fea380874c399eb5abe209d6cefe26c933f84a54220", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 92869, + "estimated_input_tokens": 23218, + "estimated_eval_tokens": 46692 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated maintenance/doc/test PRs. The only credible duplicate pair is the two resnet variable-renaming cleanups; everything else changes different files or fixes different problems.", + "confidence": 0.78, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 395 is the clearest representative of the only real duplicate pair: a self-contained refactor in src/diffusers/models/resnet.py that renames internal variables for readability.", + "best_issue_reason": null, + "best_pr_reason": "PR 395 is the best cluster representative because it is the broadest, cleanest code-only cleanup among the plausible duplicate pair and has the most complete local context.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1467", + "right": "pull_request:2846", + "accept": false, + "reason": "Both are flaky-test fixes, but they target different pipelines/files and different failure modes; not the same concrete bug." + }, + { + "left": "pull_request:168", + "right": "pull_request:178", + "accept": false, + "reason": "One adds the Stable Diffusion pipeline implementation, the other adds tests for it; related work, but not the same change or bug." + }, + { + "left": "pull_request:178", + "right": "pull_request:386", + "accept": false, + "reason": "Tests for Stable Diffusion vs docs for Stable Diffusion; different artifacts and different changes." + }, + { + "left": "pull_request:1937", + "right": "pull_request:2595", + "accept": false, + "reason": "EMA extraction in the conversion script vs scheduler config mapping are unrelated code paths." + }, + { + "left": "pull_request:1998", + "right": "pull_request:252", + "accept": false, + "reason": "Both are README/docs edits, but they update different pages for different purposes and are not the same underlying change." + }, + { + "left": "pull_request:2271", + "right": "pull_request:3989", + "accept": false, + "reason": "Both fix slow tests, but in different pipelines and with different adjustments; not the same bug." + }, + { + "left": "pull_request:2362", + "right": "pull_request:2792", + "accept": false, + "reason": "One adds the skip_mps helper and applies it to some tests; the other just applies it to text-to-video tests. Related, but not a duplicate change." + }, + { + "left": "pull_request:2383", + "right": "pull_request:3217", + "accept": false, + "reason": "Different utility additions in different files and for different pipelines; not the same fix." + }, + { + "left": "pull_request:2446", + "right": "pull_request:3987", + "accept": false, + "reason": "Two unrelated docs snippet fixes for different pipelines." + }, + { + "left": "pull_request:2846", + "right": "pull_request:3989", + "accept": false, + "reason": "Different slow-test fixes in different pipelines; same broad theme only." + }, + { + "left": "pull_request:2935", + "right": "pull_request:3119", + "accept": false, + "reason": "Both touch RePaint docs, but they fix different doc snippets and are not the same concrete issue." + }, + { + "left": "pull_request:353", + "right": "pull_request:948", + "accept": false, + "reason": "Two unrelated typo fixes in different pipeline/model files." + }, + { + "left": "pull_request:395", + "right": "pull_request:679", + "accept": true, + "reason": "Both are small resnet.py refactors renaming single-letter variables in the same code path; they could plausibly be merged into one cleanup PR." + }, + { + "left": "pull_request:4145", + "right": "pull_request:4175", + "accept": false, + "reason": "A draft UnCLIP pipeline toy change vs a ControlNet webdataset training example; unrelated." + }, + { + "left": "pull_request:4472", + "right": "pull_request:4540", + "accept": false, + "reason": "Both edit the same README_sdxl.md, but one removes a login note and the other adds a Colab link; separate documentation updates." + }, + { + "left": "pull_request:4495", + "right": "pull_request:5844", + "accept": false, + "reason": "Different docs/examples additions for different SDXL references; not the same change." + }, + { + "left": "pull_request:4652", + "right": "pull_request:4797", + "accept": false, + "reason": "One is a LoRA test tweak, the other changes the LoRA loader mapping logic; not the same bug or fix." + }, + { + "left": "pull_request:5051", + "right": "pull_request:6122", + "accept": false, + "reason": "Different documentation typo sweeps across different pages/files." + }, + { + "left": "pull_request:5065", + "right": "pull_request:6837", + "accept": false, + "reason": "LoRA/attention-processor cleanup vs i2vgen docs dependency tweak; unrelated." + }, + { + "left": "pull_request:5066", + "right": "pull_request:5332", + "accept": false, + "reason": "One is a broader code cleanup in attention-related examples/files; the other is a DreamBooth LoRA comment typo. Different changes." + }, + { + "left": "pull_request:5411", + "right": "pull_request:7472", + "accept": false, + "reason": "Both are broken-link fixes in docs, but they target different pages and links." + }, + { + "left": "pull_request:6775", + "right": "pull_request:6837", + "accept": false, + "reason": "Missing parameter docs in DDPM vs missing deps in I2VGen docs; unrelated." + }, + { + "left": "pull_request:6775", + "right": "pull_request:7425", + "accept": false, + "reason": "Different docs fixes in different parts of the docs tree." + }, + { + "left": "pull_request:688", + "right": "pull_request:715", + "accept": false, + "reason": "Both address importability without PyTorch, but in different modules and different failure points; too broad to treat as the same bug." + }, + { + "left": "pull_request:8470", + "right": "pull_request:8775", + "accept": false, + "reason": "Different README updates for different research projects; not a duplicate change." + }, + { + "left": "pull_request:8923", + "right": "pull_request:8946", + "accept": false, + "reason": "CI/debug workflow changes vs a test expectation change; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 254, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10323", + "pull_request:1087", + "pull_request:11438", + "pull_request:11473", + "pull_request:11512", + "pull_request:11537", + "pull_request:11737", + "pull_request:1210", + "pull_request:1218", + "pull_request:12265", + "pull_request:12288", + "pull_request:12526", + "pull_request:12705", + "pull_request:1277", + "pull_request:12915", + "pull_request:13217", + "pull_request:1324", + "pull_request:1510", + "pull_request:1582", + "pull_request:1583", + "pull_request:1713", + "pull_request:2608", + "pull_request:326", + "pull_request:3286", + "pull_request:3313", + "pull_request:333", + "pull_request:4099", + "pull_request:4862", + "pull_request:5506", + "pull_request:5982", + "pull_request:6065", + "pull_request:6216", + "pull_request:6291", + "pull_request:6299", + "pull_request:6300", + "pull_request:6301", + "pull_request:6308", + "pull_request:6907", + "pull_request:7118", + "pull_request:7131", + "pull_request:718", + "pull_request:7301", + "pull_request:7613", + "pull_request:774", + "pull_request:7747", + "pull_request:9077", + "pull_request:9739" + ], + "soft_pairs": [ + "pull_request:326|pull_request:333", + "pull_request:7118|pull_request:7131", + "pull_request:3286|pull_request:3313", + "pull_request:10323|pull_request:11473", + "pull_request:11438|pull_request:11737", + "pull_request:5982|pull_request:6065", + "pull_request:7301|pull_request:7747", + "pull_request:4099|pull_request:4862", + "pull_request:5506|pull_request:6308", + "pull_request:12915|pull_request:13217", + "pull_request:11512|pull_request:11537", + "pull_request:6216|pull_request:6291", + "pull_request:6907|pull_request:7613", + "pull_request:12288|pull_request:12526", + "pull_request:2608|pull_request:718", + "pull_request:6299|pull_request:6300", + "pull_request:6299|pull_request:6301", + "pull_request:6300|pull_request:6301", + "pull_request:9077|pull_request:9739", + "pull_request:1277|pull_request:1324", + "pull_request:1582|pull_request:1583", + "pull_request:1087|pull_request:774", + "pull_request:1210|pull_request:774", + "pull_request:1218|pull_request:1713", + "pull_request:12265|pull_request:12705", + "pull_request:1510|pull_request:1713" + ], + "prepared_review_unit_hash": "78724315c2b28bc4e6a022f134f920f94fab64bce028a5ef3e7ea010dfc1d95d", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 106361, + "estimated_input_tokens": 26591, + "estimated_eval_tokens": 53438 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly unrelated PRs across different docs, tests, and feature work. The only clear duplicate group is the identical text-to-image training PR trio (6299/6300/6301); SDXL Turbo docs (5982/6065) look like a plausible continuation of the same docs rollout, but most other soft edges are false positives.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 6299 is the best canonical PR candidate because it is part of the only exact-duplicate trio in the set (6299/6300/6301) with identical filenames and diff shape.", + "best_issue_reason": null, + "best_pr_reason": "PR 6299 is the strongest overall PR candidate here: it has an exact duplicate match and is the clearest representative of a duplicate cluster, unlike the other mostly unrelated PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:326", + "right": "pull_request:333", + "accept": false, + "reason": "Different pipelines and different type-hint cleanups; same theme, but not the same underlying change." + }, + { + "left": "pull_request:7118", + "right": "pull_request:7131", + "accept": false, + "reason": "Both are docs typo fixes, but they touch different documentation areas and correct different wording issues." + }, + { + "left": "pull_request:3286", + "right": "pull_request:3313", + "accept": false, + "reason": "Both target torch.compile compatibility, but they fix different code paths and different breakages." + }, + { + "left": "pull_request:10323", + "right": "pull_request:11473", + "accept": false, + "reason": "Different quantization docs changes: link formatting vs a torchao fp8 terminology correction." + }, + { + "left": "pull_request:11438", + "right": "pull_request:11737", + "accept": false, + "reason": "One adds aspect-ratio bucketing, the other changes DeepSpeed support behavior; not the same bug or feature." + }, + { + "left": "pull_request:5982", + "right": "pull_request:6065", + "accept": true, + "reason": "Both are part of the SDXL Turbo docs rollout and could plausibly be merged into one documentation PR." + }, + { + "left": "pull_request:7301", + "right": "pull_request:7747", + "accept": false, + "reason": "Unrelated test fixes: a VAE scaling constant vs a VAE tiling assertion." + }, + { + "left": "pull_request:4099", + "right": "pull_request:4862", + "accept": false, + "reason": "Different SDXL bugs in different code paths: prompt embedding handling vs MultiControlNet img2img sizing." + }, + { + "left": "pull_request:5506", + "right": "pull_request:6308", + "accept": false, + "reason": "Different LoRA-related fixes; one is xformers/PEFT interaction, the other is dummy component test setup." + }, + { + "left": "pull_request:12915", + "right": "pull_request:13217", + "accept": false, + "reason": "Same family of LTX2 work, but these are different model-version additions and not the same concrete change." + }, + { + "left": "pull_request:11512", + "right": "pull_request:11537", + "accept": false, + "reason": "Different torch.compile test additions for different model classes." + }, + { + "left": "pull_request:6216", + "right": "pull_request:6291", + "accept": false, + "reason": "Different test fixes in the same file; one handles PEFT key remapping, the other changes a LoRA unfusion assertion." + }, + { + "left": "pull_request:6907", + "right": "pull_request:7613", + "accept": false, + "reason": "Different model-card standardization work for different training scripts." + }, + { + "left": "pull_request:12288", + "right": "pull_request:12526", + "accept": false, + "reason": "Both add new video-generation pipelines, but they are different models/features and not one mergeable fix." + }, + { + "left": "pull_request:2608", + "right": "pull_request:718", + "accept": false, + "reason": "Both are typo fixes, but they address different text and different locations; too broad to treat as the same change." + }, + { + "left": "pull_request:6299", + "right": "pull_request:6300", + "accept": true, + "reason": "Identical diff preview and filenames; these are effectively the same PR change." + }, + { + "left": "pull_request:6299", + "right": "pull_request:6301", + "accept": true, + "reason": "Identical diff preview and filenames; same underlying change." + }, + { + "left": "pull_request:6300", + "right": "pull_request:6301", + "accept": true, + "reason": "Identical diff preview and filenames; same underlying change." + }, + { + "left": "pull_request:9077", + "right": "pull_request:9739", + "accept": false, + "reason": "Both are typo/docs PRs, but they touch unrelated docs and different content changes." + }, + { + "left": "pull_request:1277", + "right": "pull_request:1324", + "accept": false, + "reason": "One adds INT8 quantization documentation, the other adds an INT8 Stable Diffusion feature path; related topic but not the same change." + }, + { + "left": "pull_request:1582", + "right": "pull_request:1583", + "accept": false, + "reason": "Both touch repaint/inpaint logic, but they fix different scheduler/pipeline steps and are not the same bug." + }, + { + "left": "pull_request:1087", + "right": "pull_request:774", + "accept": false, + "reason": "Different slow-test fixes affecting different tests and code paths." + }, + { + "left": "pull_request:1210", + "right": "pull_request:774", + "accept": false, + "reason": "Same as above: unrelated test/scheduler fixes." + }, + { + "left": "pull_request:1218", + "right": "pull_request:1713", + "accept": false, + "reason": "Both improve the conversion script, but they change different conversion behaviors and are not one concrete fix." + }, + { + "left": "pull_request:12265", + "right": "pull_request:12705", + "accept": false, + "reason": "Both are typo PRs, but they are broad unrelated cleanup passes over different files." + }, + { + "left": "pull_request:1510", + "right": "pull_request:1713", + "accept": false, + "reason": "Different conversion-script changes with different scheduler/model support; not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 255, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10517", + "pull_request:10537", + "pull_request:10581", + "pull_request:10589", + "pull_request:10759", + "pull_request:10774", + "pull_request:12004", + "pull_request:12006", + "pull_request:12915", + "pull_request:13121", + "pull_request:2007", + "pull_request:2637", + "pull_request:2642", + "pull_request:2647", + "pull_request:2684", + "pull_request:2685", + "pull_request:2686", + "pull_request:2687", + "pull_request:2688", + "pull_request:2690", + "pull_request:2730", + "pull_request:2740", + "pull_request:2741", + "pull_request:2742", + "pull_request:2746", + "pull_request:2748", + "pull_request:2749", + "pull_request:2751", + "pull_request:2752", + "pull_request:2862", + "pull_request:2863", + "pull_request:2864", + "pull_request:2865", + "pull_request:2870", + "pull_request:2923", + "pull_request:2954", + "pull_request:3690", + "pull_request:3695", + "pull_request:433", + "pull_request:4695", + "pull_request:5058", + "pull_request:5151", + "pull_request:7068", + "pull_request:7246", + "pull_request:7247", + "pull_request:7411", + "pull_request:9423", + "pull_request:9597" + ], + "soft_pairs": [ + "pull_request:2007|pull_request:433", + "pull_request:2637|pull_request:2730", + "pull_request:2642|pull_request:2647", + "pull_request:2642|pull_request:2684", + "pull_request:2642|pull_request:2685", + "pull_request:2642|pull_request:2686", + "pull_request:2642|pull_request:2687", + "pull_request:2642|pull_request:2688", + "pull_request:2642|pull_request:2690", + "pull_request:2642|pull_request:2740", + "pull_request:2642|pull_request:2741", + "pull_request:2642|pull_request:2742", + "pull_request:2642|pull_request:2746", + "pull_request:2642|pull_request:2749", + "pull_request:2642|pull_request:2751", + "pull_request:2642|pull_request:2752", + "pull_request:2642|pull_request:2862", + "pull_request:2642|pull_request:2863", + "pull_request:2642|pull_request:2864", + "pull_request:2642|pull_request:2865", + "pull_request:2642|pull_request:2870", + "pull_request:2642|pull_request:2923", + "pull_request:2647|pull_request:2684", + "pull_request:2647|pull_request:2685", + "pull_request:2647|pull_request:2686", + "pull_request:2647|pull_request:2687", + "pull_request:2647|pull_request:2688", + "pull_request:2647|pull_request:2690", + "pull_request:2647|pull_request:2742", + "pull_request:2647|pull_request:2749", + "pull_request:2647|pull_request:2752", + "pull_request:2647|pull_request:2862", + "pull_request:2647|pull_request:2863", + "pull_request:2647|pull_request:2864", + "pull_request:2647|pull_request:2865", + "pull_request:2647|pull_request:2870", + "pull_request:2647|pull_request:2923", + "pull_request:2684|pull_request:2685", + "pull_request:2684|pull_request:2686", + "pull_request:2684|pull_request:2687", + "pull_request:2684|pull_request:2688", + "pull_request:2684|pull_request:2690", + "pull_request:2684|pull_request:2742", + "pull_request:2684|pull_request:2749", + "pull_request:2684|pull_request:2752", + "pull_request:2684|pull_request:2862", + "pull_request:2684|pull_request:2863", + "pull_request:2684|pull_request:2865", + "pull_request:2684|pull_request:2870", + "pull_request:2684|pull_request:2923", + "pull_request:7068|pull_request:7411", + "pull_request:10759|pull_request:10774", + "pull_request:12915|pull_request:13121", + "pull_request:10517|pull_request:10537", + "pull_request:12004|pull_request:12006", + "pull_request:2954|pull_request:4695", + "pull_request:7246|pull_request:7247", + "pull_request:5058|pull_request:5151", + "pull_request:2642|pull_request:2748", + "pull_request:2647|pull_request:2748", + "pull_request:2684|pull_request:2748", + "pull_request:9423|pull_request:9597", + "pull_request:10581|pull_request:10589", + "pull_request:3690|pull_request:3695" + ], + "prepared_review_unit_hash": "ac3cc3bd4dc987a1fab6b5688e0c7ee9516906704ed50f152f9cf60c853e894e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 64, + "serialized_chars": 114078, + "estimated_input_tokens": 28520, + "estimated_eval_tokens": 57296 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a noisy mixed cluster: many items are unrelated docs/typo or feature PRs that only share broad subsystem/file overlap. Only a few pairs look like true duplicates or near-duplicates: the black update, the Sana 4K pipeline fix, and the Kandinsky docstring example fix.", + "confidence": 0.74, + "canonical_issue_reason": null, + "canonical_pr_reason": "10537 is the clearest representative: it is a merged, concrete Sana 4K pipeline fix with an earlier matching PR (10517) that appears to be the same change.", + "best_issue_reason": null, + "best_pr_reason": "10537 is the best overall PR candidate because it fixes an actual code path in the Sana 4K pipeline, unlike the many docs-only or broad-sweep PRs in this cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2007", + "right": "pull_request:433", + "accept": true, + "reason": "Same black-version bump / formatting sweep; 433 and 2007 are effectively the same maintenance change." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2730", + "accept": false, + "reason": "Both are doc edits, but they touch different pages and unrelated wording fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2647", + "accept": false, + "reason": "Different docs typo fixes in unrelated files; same style, not the same change." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2684", + "accept": false, + "reason": "Unrelated docs edits in different pages; no shared concrete bug or change." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2685", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2686", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2687", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2688", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2690", + "accept": false, + "reason": "Different documentation pages with separate typo fixes." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2740", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2741", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2742", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2746", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2748", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2749", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2751", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2752", + "accept": false, + "reason": "Separate docs fixes; no shared concrete change beyond general proofreading." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2862", + "accept": false, + "reason": "Different docs edits in unrelated conceptual pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2863", + "accept": false, + "reason": "Different docs edits in unrelated API pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2864", + "accept": false, + "reason": "Different docs edits in unrelated API pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2865", + "accept": false, + "reason": "Different docs edits in unrelated API pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2870", + "accept": false, + "reason": "Different docs edits in unrelated API pages." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2923", + "accept": false, + "reason": "Different docs edits in unrelated API pages." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2684", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2685", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2686", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2687", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2688", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2690", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2742", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2749", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2752", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2862", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2863", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2864", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2865", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2870", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2923", + "accept": false, + "reason": "Different docs pages and separate typo fixes." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2685", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2686", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2687", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2688", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2690", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2742", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2749", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2752", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2862", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2863", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2865", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2870", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2923", + "accept": false, + "reason": "Separate docs typo fixes, not the same underlying change." + }, + { + "left": "pull_request:7068", + "right": "pull_request:7411", + "accept": false, + "reason": "Both are typo sweeps, but they touch different file sets and different documentation/code areas." + }, + { + "left": "pull_request:10759", + "right": "pull_request:10774", + "accept": false, + "reason": "One adds the community pipeline, the other fixes flattened image elements; related files but distinct changes." + }, + { + "left": "pull_request:12915", + "right": "pull_request:13121", + "accept": false, + "reason": "New LTX2 video pipelines versus a later inference bug fix in the same family; not the same change." + }, + { + "left": "pull_request:10517", + "right": "pull_request:10537", + "accept": true, + "reason": "Same Sana 4K PAG pipeline bug fix; 10517 and 10537 target the same underlying issue." + }, + { + "left": "pull_request:12004", + "right": "pull_request:12006", + "accept": false, + "reason": "Broad Wan2.2 rollout versus a specific 5B I2V addition; different scope and code paths." + }, + { + "left": "pull_request:2954", + "right": "pull_request:4695", + "accept": false, + "reason": "TextToVideoZero and TextToVideoZeroSDXLPipeline are related but separate features, not one duplicate fix." + }, + { + "left": "pull_request:7246", + "right": "pull_request:7247", + "accept": false, + "reason": "Different test fixes: one is device handling for VAE tiling, the other config checking for single-file conversion." + }, + { + "left": "pull_request:5058", + "right": "pull_request:5151", + "accept": false, + "reason": "Both are PEFT-related, but one integrates the text encoder and the other integrates the UNet; distinct code paths." + }, + { + "left": "pull_request:9423", + "right": "pull_request:9597", + "accept": false, + "reason": "Different community pipelines for different backends; not the same underlying bug or change." + }, + { + "left": "pull_request:10581", + "right": "pull_request:10589", + "accept": false, + "reason": "Separate Stable Diffusion 3 pipeline variants with different IP-Adapter plumbing." + }, + { + "left": "pull_request:3690", + "right": "pull_request:3695", + "accept": true, + "reason": "Both correct Kandinsky docstring examples and usage snippets; they look like the same documentation fix family and could be merged." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 256, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1041", + "pull_request:1087", + "pull_request:11096", + "pull_request:12004", + "pull_request:12006", + "pull_request:12024", + "pull_request:1210", + "pull_request:1219", + "pull_request:13039", + "pull_request:13042", + "pull_request:13275", + "pull_request:13293", + "pull_request:13442", + "pull_request:17", + "pull_request:1724", + "pull_request:21", + "pull_request:2137", + "pull_request:2138", + "pull_request:2175", + "pull_request:2197", + "pull_request:22", + "pull_request:2208", + "pull_request:2415", + "pull_request:2464", + "pull_request:260", + "pull_request:2608", + "pull_request:2624", + "pull_request:3066", + "pull_request:327", + "pull_request:3470", + "pull_request:353", + "pull_request:365", + "pull_request:396", + "pull_request:407", + "pull_request:429", + "pull_request:4320", + "pull_request:500", + "pull_request:5552", + "pull_request:6952", + "pull_request:6978", + "pull_request:7068", + "pull_request:718", + "pull_request:7627", + "pull_request:774", + "pull_request:798", + "pull_request:8467", + "pull_request:8572", + "pull_request:9760" + ], + "soft_pairs": [ + "pull_request:327|pull_request:365", + "pull_request:12006|pull_request:12024", + "pull_request:7068|pull_request:8572", + "pull_request:13039|pull_request:13042", + "pull_request:6952|pull_request:6978", + "pull_request:2415|pull_request:2464", + "pull_request:7627|pull_request:8467", + "pull_request:5552|pull_request:7627", + "pull_request:12004|pull_request:12024", + "pull_request:1041|pull_request:774", + "pull_request:1087|pull_request:3066", + "pull_request:11096|pull_request:9760", + "pull_request:1210|pull_request:3066", + "pull_request:1219|pull_request:396", + "pull_request:13275|pull_request:13293", + "pull_request:13275|pull_request:13442", + "pull_request:17|pull_request:718", + "pull_request:1724|pull_request:2208", + "pull_request:21|pull_request:22", + "pull_request:2137|pull_request:2608", + "pull_request:2137|pull_request:718", + "pull_request:2138|pull_request:260", + "pull_request:2138|pull_request:353", + "pull_request:2138|pull_request:407", + "pull_request:2138|pull_request:798", + "pull_request:2175|pull_request:2624", + "pull_request:2175|pull_request:3470", + "pull_request:2175|pull_request:429", + "pull_request:2175|pull_request:500", + "pull_request:2197|pull_request:4320" + ], + "prepared_review_unit_hash": "ff3743af661516ddb86abe46c8526694d3d9f29841a01b65059ff5b990503f4b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 95218, + "estimated_input_tokens": 23805, + "estimated_eval_tokens": 47866 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of small maintenance PRs: typo/doc cleanup, test-only fixes, and a few unrelated model/pipeline patches. Only the Wan2.2 follow-up PRs look like they belong to the same concrete change set; the rest are too different to merge as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 8572 is the broadest maintenance cleanup in the set, touching many docs/examples/code/test files and best matches the cluster\u2019s overall typo/maintenance character.", + "best_issue_reason": null, + "best_pr_reason": "PR 12004 is the most substantive standalone change here: a large Wan2.2 support PR spanning converter, core model/pipelines, and tests, so it is the strongest representative code change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:327", + "right": "pull_request:365", + "accept": false, + "reason": "Both are type-hint PRs, but they fix different pipelines/models and are unrelated changes." + }, + { + "left": "pull_request:12006", + "right": "pull_request:12024", + "accept": true, + "reason": "Same Wan2.2 image-to-video feature series; 12024 is a follow-up fix on the same pipeline/code path and could plausibly have been one PR." + }, + { + "left": "pull_request:7068", + "right": "pull_request:8572", + "accept": false, + "reason": "Both are typo/errata cleanups, but they target different files and different scopes of maintenance work." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13042", + "accept": false, + "reason": "Different bugs in different areas: Wan Animate CPU upcasting vs QwenImage prompt isolation tests." + }, + { + "left": "pull_request:6952", + "right": "pull_request:6978", + "accept": false, + "reason": "Both touch training/model-card code, but they are separate scripts and not the same underlying bug/change." + }, + { + "left": "pull_request:2415", + "right": "pull_request:2464", + "accept": false, + "reason": "One changes pipeline loading/version checks; the other removes xFormers enablement from a LoRA example. Not the same change." + }, + { + "left": "pull_request:7627", + "right": "pull_request:8467", + "accept": false, + "reason": "Both are img2img pipeline fixes, but they address different concrete issues in different pipelines." + }, + { + "left": "pull_request:5552", + "right": "pull_request:7627", + "accept": false, + "reason": "FreeU disable logic vs Playground img2img latent handling are unrelated code paths." + }, + { + "left": "pull_request:12004", + "right": "pull_request:12024", + "accept": true, + "reason": "Both are Wan2.2 support changes; 12024 is a follow-up patch to the same transformer/pipeline/test work introduced by 12004." + }, + { + "left": "pull_request:1041", + "right": "pull_request:774", + "accept": false, + "reason": "Both tweak tests, but they cover different failures and different pipelines." + }, + { + "left": "pull_request:1087", + "right": "pull_request:3066", + "accept": false, + "reason": "Different test fixes in different areas; no shared concrete bug." + }, + { + "left": "pull_request:11096", + "right": "pull_request:9760", + "accept": false, + "reason": "Both are README edits, but one is Flux TPU instructions and the other is a generic README wording update." + }, + { + "left": "pull_request:1210", + "right": "pull_request:3066", + "accept": false, + "reason": "Both are test-adjustment PRs, but they fix different failures and unrelated code paths." + }, + { + "left": "pull_request:1219", + "right": "pull_request:396", + "accept": false, + "reason": "Same DDIM docs file, but these are not clearly the same underlying change; one is a later doc tweak while the other adds the docs page/code reference." + }, + { + "left": "pull_request:13275", + "right": "pull_request:13293", + "accept": false, + "reason": "Hunyuan test kwargs vs AudioLDM2 cache-position handling are unrelated." + }, + { + "left": "pull_request:13275", + "right": "pull_request:13442", + "accept": false, + "reason": "Both are test-related, but they fix different models/scripts and different issues." + }, + { + "left": "pull_request:17", + "right": "pull_request:718", + "accept": false, + "reason": "General typo fixes in separate documents; not the same artifact or bug." + }, + { + "left": "pull_request:1724", + "right": "pull_request:2208", + "accept": false, + "reason": "Both concern xFormers docs, but one adds/reworks the section while the other updates installation details; too broad to treat as duplicates." + }, + { + "left": "pull_request:21", + "right": "pull_request:22", + "accept": false, + "reason": "Different model/test additions: UNet tests vs GLIDESuperResUNetModel tests." + }, + { + "left": "pull_request:2137", + "right": "pull_request:2608", + "accept": false, + "reason": "Unrelated typo fixes in loaders.py vs README.md." + }, + { + "left": "pull_request:2137", + "right": "pull_request:718", + "accept": false, + "reason": "Both are typo cleanups, but in different files and contexts." + }, + { + "left": "pull_request:2138", + "right": "pull_request:260", + "accept": false, + "reason": "A docs typo change and a README typo change are not the same underlying change." + }, + { + "left": "pull_request:2138", + "right": "pull_request:353", + "accept": false, + "reason": "Docs wording vs unet_blocks error-message typo are unrelated." + }, + { + "left": "pull_request:2138", + "right": "pull_request:407", + "accept": false, + "reason": "Docs wording vs attention-slicing typo are unrelated changes." + }, + { + "left": "pull_request:2138", + "right": "pull_request:798", + "accept": false, + "reason": "Docs wording vs unet docstring typo are different fixes." + }, + { + "left": "pull_request:2175", + "right": "pull_request:2624", + "accept": false, + "reason": "Both are docs, but one adds a pipeline API entry while the other documents cache_dir; not the same change." + }, + { + "left": "pull_request:2175", + "right": "pull_request:3470", + "accept": false, + "reason": "Different docs/features: pipeline API docs vs IF DreamBooth README additions." + }, + { + "left": "pull_request:2175", + "right": "pull_request:429", + "accept": false, + "reason": "Docs API entry vs configuration docs are unrelated." + }, + { + "left": "pull_request:2175", + "right": "pull_request:500", + "accept": false, + "reason": "Both are docs-related, but they document different configuration concepts and file sets." + }, + { + "left": "pull_request:2197", + "right": "pull_request:4320", + "accept": false, + "reason": "StableDiffusionInpaint typo fix and ONNX docs typo fix are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 257, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2271", + "pull_request:2290", + "pull_request:2540", + "pull_request:2587", + "pull_request:2608", + "pull_request:2624", + "pull_request:299", + "pull_request:3066", + "pull_request:3525", + "pull_request:3942", + "pull_request:408", + "pull_request:421", + "pull_request:4275", + "pull_request:429", + "pull_request:432", + "pull_request:4320", + "pull_request:4472", + "pull_request:4572", + "pull_request:4822", + "pull_request:500", + "pull_request:5003", + "pull_request:5201", + "pull_request:5454", + "pull_request:568", + "pull_request:5780", + "pull_request:6192", + "pull_request:6339", + "pull_request:6438", + "pull_request:6656", + "pull_request:6754", + "pull_request:6759", + "pull_request:6765", + "pull_request:6766", + "pull_request:6767", + "pull_request:6768", + "pull_request:6769", + "pull_request:7052", + "pull_request:7068", + "pull_request:7131", + "pull_request:718", + "pull_request:7181", + "pull_request:7236", + "pull_request:7472", + "pull_request:8600", + "pull_request:904", + "pull_request:9077", + "pull_request:9116" + ], + "soft_pairs": [ + "pull_request:2271|pull_request:3066", + "pull_request:2271|pull_request:3942", + "pull_request:2290|pull_request:421", + "pull_request:2290|pull_request:4275", + "pull_request:2540|pull_request:432", + "pull_request:2587|pull_request:429", + "pull_request:2587|pull_request:500", + "pull_request:2608|pull_request:568", + "pull_request:2624|pull_request:500", + "pull_request:299|pull_request:904", + "pull_request:3525|pull_request:5454", + "pull_request:408|pull_request:718", + "pull_request:4275|pull_request:6339", + "pull_request:429|pull_request:500", + "pull_request:4320|pull_request:4822", + "pull_request:4472|pull_request:6438", + "pull_request:4572|pull_request:6759", + "pull_request:5003|pull_request:5201", + "pull_request:568|pull_request:718", + "pull_request:5780|pull_request:6754", + "pull_request:5780|pull_request:6765", + "pull_request:5780|pull_request:6766", + "pull_request:5780|pull_request:6767", + "pull_request:5780|pull_request:6768", + "pull_request:5780|pull_request:6769", + "pull_request:5780|pull_request:7052", + "pull_request:5780|pull_request:7236", + "pull_request:6192|pull_request:6656", + "pull_request:6754|pull_request:6765", + "pull_request:6754|pull_request:6766", + "pull_request:6754|pull_request:6767", + "pull_request:6754|pull_request:6768", + "pull_request:6754|pull_request:6769", + "pull_request:6754|pull_request:7052", + "pull_request:6754|pull_request:7236", + "pull_request:6754|pull_request:8600", + "pull_request:6765|pull_request:7052", + "pull_request:6765|pull_request:7236", + "pull_request:6765|pull_request:8600", + "pull_request:6766|pull_request:7052", + "pull_request:6766|pull_request:7236", + "pull_request:6766|pull_request:8600", + "pull_request:6767|pull_request:8600", + "pull_request:7068|pull_request:7131", + "pull_request:7068|pull_request:7181", + "pull_request:7181|pull_request:9077", + "pull_request:7472|pull_request:9116" + ], + "prepared_review_unit_hash": "27bbbe38d8e1901c4283c705c0ecdd15fd76fb046fe36bc27b035505f0abd649", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 47, + "serialized_chars": 97914, + "estimated_input_tokens": 24479, + "estimated_eval_tokens": 49214 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose similarity cluster, not a true duplicate set: it mixes typo/docs edits, test expectation refreshes, and unrelated code-path fixes. I would reject all soft edges and avoid forcing a single canonical duplicate target.", + "confidence": 0.21, + "canonical_issue_reason": null, + "canonical_pr_reason": "No exact canonical duplicate stands out. If a representative is needed, PR 7181 is the broadest generic cleanup, but it is still not a true duplicate target for the rest of the cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 7181 is the best overall representative of the cluster\u2019s dominant low-risk docs/typo-cleanup theme, but the cluster is too heterogeneous for it to serve as a genuine duplicate target.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2271", + "right": "pull_request:3066", + "accept": false, + "reason": "Both touch tests, but they fix different assertions in different files and different pipelines." + }, + { + "left": "pull_request:2271", + "right": "pull_request:3942", + "accept": false, + "reason": "Separate test failures in unrelated code paths; not the same bug." + }, + { + "left": "pull_request:2290", + "right": "pull_request:421", + "accept": false, + "reason": "Both are docs fixes, but they target different pages and different typo sets." + }, + { + "left": "pull_request:2290", + "right": "pull_request:4275", + "accept": false, + "reason": "Different documentation areas and unrelated wording changes." + }, + { + "left": "pull_request:2540", + "right": "pull_request:432", + "accept": false, + "reason": "Both are docs-only, but they fix different pages and different link/text issues." + }, + { + "left": "pull_request:2587", + "right": "pull_request:429", + "accept": false, + "reason": "A translated docs add-on and a configuration docs rewrite are not the same change." + }, + { + "left": "pull_request:2587", + "right": "pull_request:500", + "accept": false, + "reason": "Different documentation changes with no shared concrete bug." + }, + { + "left": "pull_request:2608", + "right": "pull_request:568", + "accept": false, + "reason": "Both are typo cleanups, but they affect unrelated files and unrelated wording." + }, + { + "left": "pull_request:2624", + "right": "pull_request:500", + "accept": false, + "reason": "Different docs topics: cache_dir API docs versus subfolder docs." + }, + { + "left": "pull_request:299", + "right": "pull_request:904", + "accept": false, + "reason": "A scheduler implementation hotfix and an inpainting pipeline rewrite are unrelated code paths." + }, + { + "left": "pull_request:3525", + "right": "pull_request:5454", + "accept": false, + "reason": "A README cleanup and a scheduler docs edit are not the same underlying change." + }, + { + "left": "pull_request:408", + "right": "pull_request:718", + "accept": false, + "reason": "Both are broad docs typo fixes, but they patch different files and different mistakes." + }, + { + "left": "pull_request:4275", + "right": "pull_request:6339", + "accept": false, + "reason": "Both touch docs, but for different pipelines and different example text." + }, + { + "left": "pull_request:429", + "right": "pull_request:500", + "accept": false, + "reason": "Both are configuration docs, but they address different API surfaces and different additions." + }, + { + "left": "pull_request:4320", + "right": "pull_request:4822", + "accept": false, + "reason": "A docs typo fix and a loader warning typo fix are different changes in different files." + }, + { + "left": "pull_request:4472", + "right": "pull_request:6438", + "accept": false, + "reason": "One removes SDXL login instructions; the other adds installation platform guidance." + }, + { + "left": "pull_request:4572", + "right": "pull_request:6759", + "accept": false, + "reason": "Different training/example bugs in different scripts and code paths." + }, + { + "left": "pull_request:5003", + "right": "pull_request:5201", + "accept": false, + "reason": "Both involve LoRA, but one fixes fuse/unfuse internals while the other changes processor removal behavior." + }, + { + "left": "pull_request:568", + "right": "pull_request:718", + "accept": false, + "reason": "Generic typo fixes, but they touch different subsystems and unrelated wording." + }, + { + "left": "pull_request:5780", + "right": "pull_request:6754", + "accept": false, + "reason": "A community example typo and a diffusion DPO loss fix are unrelated." + }, + { + "left": "pull_request:5780", + "right": "pull_request:6765", + "accept": false, + "reason": "Example doc wording versus cache-path utility logic; not the same bug." + }, + { + "left": "pull_request:5780", + "right": "pull_request:6766", + "accept": false, + "reason": "Example doc wording versus constants path handling; unrelated changes." + }, + { + "left": "pull_request:5780", + "right": "pull_request:6767", + "accept": false, + "reason": "Example docs and Flax model path handling are different code paths." + }, + { + "left": "pull_request:5780", + "right": "pull_request:6768", + "accept": false, + "reason": "Example typo fix and config path handling are unrelated." + }, + { + "left": "pull_request:5780", + "right": "pull_request:6769", + "accept": false, + "reason": "Example typo fix and dynamic module path handling are unrelated." + }, + { + "left": "pull_request:5780", + "right": "pull_request:7052", + "accept": false, + "reason": "A small example typo and a large TensorRT example rewrite are not the same change." + }, + { + "left": "pull_request:5780", + "right": "pull_request:7236", + "accept": false, + "reason": "Different community/example edits with no shared concrete bug." + }, + { + "left": "pull_request:6192", + "right": "pull_request:6656", + "accept": false, + "reason": "Both add unload-related APIs, but for different loaders and different model components." + }, + { + "left": "pull_request:6754", + "right": "pull_request:6765", + "accept": false, + "reason": "Different fixes: DPO loss math versus cache-path construction." + }, + { + "left": "pull_request:6754", + "right": "pull_request:6766", + "accept": false, + "reason": "Different code paths and unrelated bugs." + }, + { + "left": "pull_request:6754", + "right": "pull_request:6767", + "accept": false, + "reason": "A training-loss fix and a Flax path fix are unrelated." + }, + { + "left": "pull_request:6754", + "right": "pull_request:6768", + "accept": false, + "reason": "Different functionality: DPO loss calculation versus config path joining." + }, + { + "left": "pull_request:6754", + "right": "pull_request:6769", + "accept": false, + "reason": "Different functionality and different subsystems." + }, + { + "left": "pull_request:6754", + "right": "pull_request:7052", + "accept": false, + "reason": "A loss-function fix and a TensorRT example update are unrelated." + }, + { + "left": "pull_request:6754", + "right": "pull_request:7236", + "accept": false, + "reason": "Different DreamBooth/DPO training scripts and different bugs." + }, + { + "left": "pull_request:6754", + "right": "pull_request:8600", + "accept": false, + "reason": "Different training scripts and unrelated fixes." + }, + { + "left": "pull_request:6765", + "right": "pull_request:7052", + "accept": false, + "reason": "Cache-path utility logic and a TensorRT example are unrelated." + }, + { + "left": "pull_request:6765", + "right": "pull_request:7236", + "accept": false, + "reason": "Different code paths and different bug classes." + }, + { + "left": "pull_request:6765", + "right": "pull_request:8600", + "accept": false, + "reason": "Cache-path handling and DreamBooth text-encoder cleanup are unrelated." + }, + { + "left": "pull_request:6766", + "right": "pull_request:7052", + "accept": false, + "reason": "Constants path construction and a TensorRT example rewrite are not the same change." + }, + { + "left": "pull_request:6766", + "right": "pull_request:7236", + "accept": false, + "reason": "Different files, different behavior, no shared concrete bug." + }, + { + "left": "pull_request:6766", + "right": "pull_request:8600", + "accept": false, + "reason": "Cross-platform constants fix versus DreamBooth script cleanup are unrelated." + }, + { + "left": "pull_request:6767", + "right": "pull_request:8600", + "accept": false, + "reason": "Flax model path handling and DreamBooth cleanup are unrelated code paths." + }, + { + "left": "pull_request:7068", + "right": "pull_request:7131", + "accept": false, + "reason": "Both are typo/doc fixes, but they update different files and different content." + }, + { + "left": "pull_request:7068", + "right": "pull_request:7181", + "accept": false, + "reason": "Overlapping style, but not the same underlying change or files." + }, + { + "left": "pull_request:7181", + "right": "pull_request:9077", + "accept": false, + "reason": "Both are typo sweeps, but they touch different areas and different exact edits." + }, + { + "left": "pull_request:7472", + "right": "pull_request:9116", + "accept": false, + "reason": "A broken link in examples README and a dead link in stable audio docs are separate fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 258, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10136", + "pull_request:10351", + "pull_request:10454", + "pull_request:10810", + "pull_request:11124", + "pull_request:11254", + "pull_request:1137", + "pull_request:11892", + "pull_request:11894", + "pull_request:11995", + "pull_request:12269", + "pull_request:12425", + "pull_request:12611", + "pull_request:12782", + "pull_request:13039", + "pull_request:13042", + "pull_request:13063", + "pull_request:13275", + "pull_request:13424", + "pull_request:2693", + "pull_request:325", + "pull_request:326", + "pull_request:343", + "pull_request:349", + "pull_request:365", + "pull_request:4572", + "pull_request:4780", + "pull_request:4783", + "pull_request:483", + "pull_request:5058", + "pull_request:5188", + "pull_request:5407", + "pull_request:5993", + "pull_request:6009", + "pull_request:6764", + "pull_request:7247", + "pull_request:7627", + "pull_request:7825", + "pull_request:7919", + "pull_request:7977", + "pull_request:8308", + "pull_request:8310", + "pull_request:8498", + "pull_request:9257", + "pull_request:9736", + "pull_request:9769", + "pull_request:978" + ], + "soft_pairs": [ + "pull_request:7977|pull_request:8498", + "pull_request:8308|pull_request:8310", + "pull_request:11995|pull_request:12269", + "pull_request:11892|pull_request:11894", + "pull_request:12611|pull_request:13063", + "pull_request:7247|pull_request:7825", + "pull_request:10810|pull_request:9257", + "pull_request:4572|pull_request:5407", + "pull_request:4783|pull_request:6764", + "pull_request:9736|pull_request:9769", + "pull_request:12782|pull_request:13039", + "pull_request:483|pull_request:978", + "pull_request:326|pull_request:365", + "pull_request:13275|pull_request:13424", + "pull_request:10351|pull_request:11124", + "pull_request:325|pull_request:343", + "pull_request:10136|pull_request:10454", + "pull_request:5993|pull_request:6009", + "pull_request:5188|pull_request:5407", + "pull_request:7627|pull_request:7919", + "pull_request:4780|pull_request:5058", + "pull_request:326|pull_request:349", + "pull_request:11254|pull_request:12425", + "pull_request:1137|pull_request:2693", + "pull_request:13042|pull_request:13275" + ], + "prepared_review_unit_hash": "5f298f24a0fec59103b992769911a2f954707d83af3de6bcac196883d103f486", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 106152, + "estimated_input_tokens": 26538, + "estimated_eval_tokens": 53332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated PRs. The only plausible duplicate-like pair is the two Hunyuan model refactor PRs; the rest are separate fixes, docs tweaks, or feature additions that should stay distinct.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 8310 is the clearest representative of the only plausible duplicate pair: it is the more specific Hunyuan DiT refactor and overlaps the same file set as 8308.", + "best_issue_reason": null, + "best_pr_reason": "PR 8310 is the best single PR to anchor the only real duplicate candidate in this cluster; the other items are mostly unrelated changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7977", + "right": "pull_request:8498", + "accept": false, + "reason": "Both are typo/doc fixes, but they fix different typos in different files and are unrelated changes." + }, + { + "left": "pull_request:8308", + "right": "pull_request:8310", + "accept": true, + "reason": "Same Hunyuan/Hunyuan DiT model refactor over the same file set; these look like the same underlying change and could plausibly be one PR." + }, + { + "left": "pull_request:11995", + "right": "pull_request:12269", + "accept": false, + "reason": "Both involve modular Flux work, but one adds Flux modular T2I and the other adds Flux Kontext T2I/I2I support; different changes." + }, + { + "left": "pull_request:11892", + "right": "pull_request:11894", + "accept": false, + "reason": "One removes k-diffusion, the other pins it; these are opposite dependency changes, not duplicates." + }, + { + "left": "pull_request:12611", + "right": "pull_request:13063", + "accept": false, + "reason": "Both concern Wan modular pipelines, but the second is a later task-based refactor with new Wan22/Wan i2v blocks, not the same change." + }, + { + "left": "pull_request:7247", + "right": "pull_request:7825", + "accept": false, + "reason": "Config-checking tests and device/offload tests are different test fixes for different behavior." + }, + { + "left": "pull_request:10810", + "right": "pull_request:9257", + "accept": false, + "reason": "Flux LoRA conversion support and a DreamBooth Flux prompt/tokenization fix are separate code paths." + }, + { + "left": "pull_request:4572", + "right": "pull_request:5407", + "accept": false, + "reason": "Different training/example fixes: network_alpha naming versus mixed-precision unconditional training." + }, + { + "left": "pull_request:4783", + "right": "pull_request:6764", + "accept": false, + "reason": "One adds SDXL embedding docs, the other adds a serialization note for fused adapters; unrelated documentation changes." + }, + { + "left": "pull_request:9736", + "right": "pull_request:9769", + "accept": false, + "reason": "Different model releases entirely: Allegro T2V versus Mochi T2V." + }, + { + "left": "pull_request:326", + "right": "pull_request:365", + "accept": false, + "reason": "Type-hint cleanup in different modules; no shared bug or change." + }, + { + "left": "pull_request:13275", + "right": "pull_request:13424", + "accept": false, + "reason": "Separate test fixes in different areas: Hunyuan video tests versus AutoencoderDC tests." + }, + { + "left": "pull_request:10351", + "right": "pull_request:11124", + "accept": false, + "reason": "A LoRA test fix and a CogView4 tokenizer test fix are unrelated." + }, + { + "left": "pull_request:325", + "right": "pull_request:343", + "accept": false, + "reason": "Different scheduler/pipeline type-hint updates with no shared underlying issue." + }, + { + "left": "pull_request:10136", + "right": "pull_request:10454", + "accept": false, + "reason": "The first adds Hunyuan Video support; the second fixes an attention-mask bug later. Same subsystem, but not the same change." + }, + { + "left": "pull_request:5993", + "right": "pull_request:6009", + "accept": false, + "reason": "These fix different constructor/super-call issues in different community examples." + }, + { + "left": "pull_request:5188", + "right": "pull_request:5407", + "accept": false, + "reason": "Compute-SNR utility work and unconditional training mixed-precision behavior are distinct changes." + }, + { + "left": "pull_request:7627", + "right": "pull_request:7919", + "accept": false, + "reason": "Playground img2img latent handling and device_map/offload behavior are separate fixes." + }, + { + "left": "pull_request:4780", + "right": "pull_request:5058", + "accept": false, + "reason": "PEFT integration draft work and later LoRA backend support are related, but not the same concrete change." + }, + { + "left": "pull_request:11254", + "right": "pull_request:12425", + "accept": false, + "reason": "ConsisID import fixes and torch.distributed import guarding are unrelated dependency/import changes." + }, + { + "left": "pull_request:1137", + "right": "pull_request:2693", + "accept": false, + "reason": "A pipeline dtype fix and a docs image-link typo are unrelated." + }, + { + "left": "pull_request:13042", + "right": "pull_request:13275", + "accept": false, + "reason": "Prompt-isolation tests for QwenImage and Llava-kwargs fixes for Hunyuan tests are different test changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 259, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10259", + "pull_request:11534", + "pull_request:11546", + "pull_request:12782", + "pull_request:13039", + "pull_request:13042", + "pull_request:1312", + "pull_request:13275", + "pull_request:13293", + "pull_request:13352", + "pull_request:13365", + "pull_request:13442", + "pull_request:13469", + "pull_request:2602", + "pull_request:2637", + "pull_request:2642", + "pull_request:2647", + "pull_request:2684", + "pull_request:2685", + "pull_request:2686", + "pull_request:2687", + "pull_request:2688", + "pull_request:2690", + "pull_request:2740", + "pull_request:2741", + "pull_request:2742", + "pull_request:2746", + "pull_request:2749", + "pull_request:2751", + "pull_request:2752", + "pull_request:2862", + "pull_request:2863", + "pull_request:2864", + "pull_request:2865", + "pull_request:2870", + "pull_request:2915", + "pull_request:3825", + "pull_request:3850", + "pull_request:4343", + "pull_request:4862", + "pull_request:4947", + "pull_request:5188", + "pull_request:6021", + "pull_request:7050", + "pull_request:7103", + "pull_request:7118", + "pull_request:7411", + "pull_request:9985" + ], + "soft_pairs": [ + "pull_request:2637|pull_request:3825", + "pull_request:2915|pull_request:3850", + "pull_request:4947|pull_request:5188", + "pull_request:13042|pull_request:13293", + "pull_request:13042|pull_request:13442", + "pull_request:10259|pull_request:9985", + "pull_request:1312|pull_request:2602", + "pull_request:4343|pull_request:6021", + "pull_request:7050|pull_request:7411", + "pull_request:7118|pull_request:7411", + "pull_request:13293|pull_request:13469", + "pull_request:13352|pull_request:13365", + "pull_request:4862|pull_request:7103", + "pull_request:11534|pull_request:11546", + "pull_request:12782|pull_request:13042", + "pull_request:12782|pull_request:13275", + "pull_request:13039|pull_request:13275", + "pull_request:2637|pull_request:2642", + "pull_request:2637|pull_request:2647", + "pull_request:2637|pull_request:2684", + "pull_request:2637|pull_request:2685", + "pull_request:2637|pull_request:2686", + "pull_request:2637|pull_request:2687", + "pull_request:2637|pull_request:2688", + "pull_request:2637|pull_request:2690", + "pull_request:2637|pull_request:2740", + "pull_request:2637|pull_request:2741", + "pull_request:2637|pull_request:2742", + "pull_request:2637|pull_request:2746", + "pull_request:2637|pull_request:2749", + "pull_request:2637|pull_request:2751", + "pull_request:2637|pull_request:2752", + "pull_request:2637|pull_request:2862", + "pull_request:2637|pull_request:2863", + "pull_request:2637|pull_request:2864", + "pull_request:2637|pull_request:2865", + "pull_request:2637|pull_request:2870" + ], + "prepared_review_unit_hash": "0145d76f7c5ea9a693cdfb068a0491ea0730cbd89ea00ffadd564667f53cfc94", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 37, + "serialized_chars": 98009, + "estimated_input_tokens": 24503, + "estimated_eval_tokens": 49262 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a pile of superficially similar but unrelated PRs: docs typo cleanups, test adjustments, CI workflow edits, and distinct model/loader feature changes. The soft edges are driven by overlapping keywords or shared files, not the same underlying bug or change, so none should be merged as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical representative: the set spans unrelated docs, tests, CI, and feature work, and the apparent overlaps are too shallow to indicate one shared change.", + "best_issue_reason": null, + "best_pr_reason": "No PR is a strong global duplicate anchor. If forced, PR 9985 is the broadest and most central feature change, but it is still not equivalent to the others in the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2637", + "right": "pull_request:3825", + "accept": false, + "reason": "Both are doc updates, but they edit different pages with unrelated wording changes; this is not the same underlying change." + }, + { + "left": "pull_request:2915", + "right": "pull_request:3850", + "accept": false, + "reason": "Different test suites for different pipelines; they address separate test fixes, not one shared bug." + }, + { + "left": "pull_request:4947", + "right": "pull_request:5188", + "accept": false, + "reason": "Both touch training examples, but they fix different scripts and different behaviors." + }, + { + "left": "pull_request:13042", + "right": "pull_request:13293", + "accept": false, + "reason": "One fixes QwenImage prompt-isolation tests; the other changes AudioLDM2 cache-position handling. Different code paths." + }, + { + "left": "pull_request:13042", + "right": "pull_request:13442", + "accept": false, + "reason": "Different training/test fixes in different subsystems; no shared concrete bug or merged change." + }, + { + "left": "pull_request:10259", + "right": "pull_request:9985", + "accept": false, + "reason": "Both are Flux-related, but one is LoRA-loading support for control/fill and the other is a much broader Flux feature expansion. Not the same change." + }, + { + "left": "pull_request:1312", + "right": "pull_request:2602", + "accept": false, + "reason": "Both pin huggingface-hub prereleases, but they target different versions and different release-testing contexts." + }, + { + "left": "pull_request:4343", + "right": "pull_request:6021", + "accept": false, + "reason": "Both add a rank argument to LoRA training scripts, but they modify different scripts with different surrounding logic." + }, + { + "left": "pull_request:7050", + "right": "pull_request:7411", + "accept": false, + "reason": "Both are typo/doc cleanups, but they span different doc files and unrelated edits." + }, + { + "left": "pull_request:7118", + "right": "pull_request:7411", + "accept": false, + "reason": "Different documentation pages and different typo fixes; not one underlying change." + }, + { + "left": "pull_request:13293", + "right": "pull_request:13469", + "accept": false, + "reason": "Unrelated fixes: one is AudioLDM2 pipeline behavior, the other is attention-processor test deprecation handling." + }, + { + "left": "pull_request:13352", + "right": "pull_request:13365", + "accept": false, + "reason": "Same workflow file, but the changes are different: one adds checkout, the other rewrites fork support and sanitization. Too distinct to treat as one fix." + }, + { + "left": "pull_request:4862", + "right": "pull_request:7103", + "accept": false, + "reason": "Both involve ControlNet, but they fix different bugs in different functions and should not be merged as duplicates." + }, + { + "left": "pull_request:11534", + "right": "pull_request:11546", + "accept": false, + "reason": "Framepack feature/docs vs Hunyuan LoRA-loading support are separate changes, not one code-path problem." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13042", + "accept": false, + "reason": "Different test areas for different models; no shared underlying bug." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13275", + "accept": false, + "reason": "Both are Hunyuan-related tests, but they address different test failures and different setups." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13275", + "accept": false, + "reason": "Different models and different issues: WAN CPU upcasting vs Hunyuan test kwargs." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2642", + "accept": false, + "reason": "Separate documentation pages with unrelated prose edits." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2647", + "accept": false, + "reason": "Different docs pages and different wording fixes; not the same change." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2684", + "accept": false, + "reason": "Unrelated documentation edits on different pages." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2685", + "accept": false, + "reason": "Different doc page and content; only the general docs-maintenance theme overlaps." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2686", + "accept": false, + "reason": "Different docs page and unrelated wording correction." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2687", + "accept": false, + "reason": "Separate documentation cleanup on another page." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2688", + "accept": false, + "reason": "Different docs page and edits; not a duplicate." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2690", + "accept": false, + "reason": "Different documentation content and target page." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2740", + "accept": false, + "reason": "Different training docs with distinct edits; no shared underlying fix." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2741", + "accept": false, + "reason": "Different docs file and separate wording change." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2742", + "accept": false, + "reason": "Dreambooth docs vs quicktour docs; unrelated changes." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2746", + "accept": false, + "reason": "Different docs page and typo fix only." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2749", + "accept": false, + "reason": "Different optimization docs with unrelated wording edits." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2751", + "accept": false, + "reason": "Different training documentation page and different change." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2752", + "accept": false, + "reason": "Philosophy docs vs quicktour docs; separate edits." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2862", + "accept": false, + "reason": "Different conceptual docs page and different wording cleanup." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2863", + "accept": false, + "reason": "Different docs page and standalone typo fix." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2864", + "accept": false, + "reason": "Different docs page; only topical similarity exists." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2865", + "accept": false, + "reason": "Different pipeline docs pages and separate typo fixes." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2870", + "accept": false, + "reason": "Different docs page and unrelated formatting/title cleanup." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 260, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:12069", + "pull_request:12416", + "pull_request:12454", + "pull_request:12490", + "pull_request:12782", + "pull_request:13039", + "pull_request:13042", + "pull_request:13293", + "pull_request:13424", + "pull_request:13442", + "pull_request:154", + "pull_request:2637", + "pull_request:2642", + "pull_request:2647", + "pull_request:2684", + "pull_request:2730", + "pull_request:2923", + "pull_request:2928", + "pull_request:3214", + "pull_request:356", + "pull_request:3607", + "pull_request:3798", + "pull_request:4546", + "pull_request:4621", + "pull_request:4749", + "pull_request:4822", + "pull_request:4947", + "pull_request:4984", + "pull_request:5164", + "pull_request:5188", + "pull_request:5552", + "pull_request:5615", + "pull_request:5768", + "pull_request:6276", + "pull_request:6346", + "pull_request:6433", + "pull_request:6485", + "pull_request:7062", + "pull_request:7247", + "pull_request:7353", + "pull_request:7746", + "pull_request:7825", + "pull_request:7934", + "pull_request:8467", + "pull_request:8797", + "pull_request:9060", + "pull_request:9206", + "pull_request:9696" + ], + "soft_pairs": [ + "pull_request:2637|pull_request:2923", + "pull_request:2642|pull_request:2730", + "pull_request:2647|pull_request:2730", + "pull_request:2684|pull_request:2730", + "pull_request:2928|pull_request:3798", + "pull_request:3214|pull_request:3607", + "pull_request:4546|pull_request:4822", + "pull_request:4984|pull_request:6485", + "pull_request:5768|pull_request:7746", + "pull_request:6276|pull_request:6433", + "pull_request:7247|pull_request:7934", + "pull_request:7247|pull_request:8797", + "pull_request:9206|pull_request:9696", + "pull_request:12782|pull_request:13293", + "pull_request:12782|pull_request:13442", + "pull_request:13039|pull_request:13293", + "pull_request:13039|pull_request:13442", + "pull_request:4621|pull_request:4947", + "pull_request:4749|pull_request:5615", + "pull_request:5164|pull_request:5552", + "pull_request:7062|pull_request:8467", + "pull_request:7353|pull_request:9060", + "pull_request:12069|pull_request:12490", + "pull_request:154|pull_request:356", + "pull_request:7247|pull_request:9206", + "pull_request:12416|pull_request:12454", + "pull_request:13042|pull_request:13424", + "pull_request:7825|pull_request:8797", + "pull_request:5188|pull_request:6346" + ], + "prepared_review_unit_hash": "4a5fc1219889358300bbeb15d5d09c6853a46930942188be71ed731e578aea0e", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 98002, + "estimated_input_tokens": 24501, + "estimated_eval_tokens": 49258 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set; it mixes unrelated docs tweaks, test fixes, small bug fixes, and feature work across different subsystems. Most soft-edge pairs should stay separate.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "If a single anchor is needed, PR 5164 is the strongest representative: it is the broadest substantive code change, spans many files, and had the most discussion/review activity. But the set is heterogeneous, so no real canonical PR exists.", + "best_issue_reason": null, + "best_pr_reason": "PR 5164 is the best overall representative because it is the largest, most impactful core-library change in the set and has the clearest product significance.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2637", + "right": "pull_request:2923", + "accept": false, + "reason": "Both are docs edits, but they change different pages and different wording; not the same underlying change." + }, + { + "left": "pull_request:2642", + "right": "pull_request:2730", + "accept": false, + "reason": "Different documentation pages with unrelated edits; no shared bug or change request." + }, + { + "left": "pull_request:2647", + "right": "pull_request:2730", + "accept": false, + "reason": "Unrelated docs wording fix versus training docs newline/wording cleanup." + }, + { + "left": "pull_request:2684", + "right": "pull_request:2730", + "accept": false, + "reason": "Different docs files and different edits; not a duplicate." + }, + { + "left": "pull_request:2928", + "right": "pull_request:3798", + "accept": false, + "reason": "ControlNet dataset-loading change is unrelated to the comment-only training-script cleanup." + }, + { + "left": "pull_request:3214", + "right": "pull_request:3607", + "accept": false, + "reason": "Attention reshape bug fix is a concrete code-path issue; the other PR is a separate temb-attention change with different scope." + }, + { + "left": "pull_request:4546", + "right": "pull_request:4822", + "accept": false, + "reason": "Both are typo/warning text fixes, but in different scripts and different messages; not the same change." + }, + { + "left": "pull_request:4984", + "right": "pull_request:6485", + "accept": false, + "reason": "Same template file, but the reviewer-list update is a later separate maintenance edit, not the same underlying fix." + }, + { + "left": "pull_request:5768", + "right": "pull_request:7746", + "accept": false, + "reason": "One adjusts docs/config handling for consistency decoder; the other changes a VAE test dtype. Different problems." + }, + { + "left": "pull_request:6276", + "right": "pull_request:6433", + "accept": false, + "reason": "Community README entry versus a separate community pipeline addition; same area, but not the same artifact or bug." + }, + { + "left": "pull_request:7247", + "right": "pull_request:7934", + "accept": false, + "reason": "Both involve single-file/loading utilities, but they fix different code paths and different failure modes." + }, + { + "left": "pull_request:7247", + "right": "pull_request:8797", + "accept": false, + "reason": "Single-file config normalization and sharding tests are unrelated." + }, + { + "left": "pull_request:9206", + "right": "pull_request:9696", + "accept": false, + "reason": "Broken xformers tests versus workflow input validation; different concerns and not mergeable as one fix." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13293", + "accept": false, + "reason": "Hunyuan offloading test tweak is unrelated to AudioLDM2 cache-position logic." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13442", + "accept": false, + "reason": "Different test suites and different model families; no shared underlying bug." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13293", + "accept": false, + "reason": "WAN layerwise upcasting on CPU is unrelated to AudioLDM2 cache-position handling." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13442", + "accept": false, + "reason": "Different model-specific test fixes; not the same issue." + }, + { + "left": "pull_request:4621", + "right": "pull_request:4947", + "accept": false, + "reason": "Multi-adapter downscale validation and T2I adapter training support are separate changes." + }, + { + "left": "pull_request:4749", + "right": "pull_request:5615", + "accept": false, + "reason": "SDXL LoRA checkpointing and SDXL Adapter load-lora revert address different code paths." + }, + { + "left": "pull_request:5164", + "right": "pull_request:5552", + "accept": false, + "reason": "FreeU mechanism introduction and fixing disable_freeu are related, but the PRs are different changes with different concrete fixes." + }, + { + "left": "pull_request:7062", + "right": "pull_request:8467", + "accept": false, + "reason": "Community imagic docstring cleanup versus SEGA pipeline fix are unrelated." + }, + { + "left": "pull_request:7353", + "right": "pull_request:9060", + "accept": false, + "reason": "Adding FromSingleFileMixin to a ControlNet XL pipeline is unrelated to removing a debug print in Flux." + }, + { + "left": "pull_request:12069", + "right": "pull_request:12490", + "accept": false, + "reason": "Both are AudioLDM2-related, but one fixes pipeline compatibility logic while the other only adjusts test component construction." + }, + { + "left": "pull_request:154", + "right": "pull_request:356", + "accept": false, + "reason": "Stable Diffusion conversion script creation and training README/implementation work are not the same change." + }, + { + "left": "pull_request:7247", + "right": "pull_request:9206", + "accept": false, + "reason": "Single-file config conversion and xformers test failures are unrelated." + }, + { + "left": "pull_request:12416", + "right": "pull_request:12454", + "accept": false, + "reason": "Both add modular pipeline support, but for different models/features; not the same underlying bug or change." + }, + { + "left": "pull_request:7825", + "right": "pull_request:8797", + "accept": false, + "reason": "Model offload guards and sharded-checkpoint test fixes are separate test issues." + }, + { + "left": "pull_request:5188", + "right": "pull_request:6346", + "accept": false, + "reason": "compute_snr utility addition and LoRA state-dict conversion are unrelated training-example fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 261, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:102", + "pull_request:1020", + "pull_request:1077", + "pull_request:11588", + "pull_request:1162", + "pull_request:11726", + "pull_request:1182", + "pull_request:119", + "pull_request:1194", + "pull_request:120", + "pull_request:121", + "pull_request:1218", + "pull_request:12265", + "pull_request:12307", + "pull_request:12490", + "pull_request:12552", + "pull_request:12573", + "pull_request:12584", + "pull_request:12628", + "pull_request:12634", + "pull_request:12705", + "pull_request:12763", + "pull_request:12782", + "pull_request:12932", + "pull_request:13039", + "pull_request:13042", + "pull_request:13115", + "pull_request:13424", + "pull_request:1360", + "pull_request:1487", + "pull_request:15", + "pull_request:1610", + "pull_request:168", + "pull_request:1724", + "pull_request:1861", + "pull_request:2006", + "pull_request:2208", + "pull_request:2329", + "pull_request:252", + "pull_request:2624", + "pull_request:312", + "pull_request:3470", + "pull_request:415", + "pull_request:429", + "pull_request:61", + "pull_request:63", + "pull_request:8631", + "pull_request:8798" + ], + "soft_pairs": [ + "pull_request:8631|pull_request:8798", + "pull_request:12584|pull_request:12634", + "pull_request:12490|pull_request:13042", + "pull_request:11588|pull_request:11726", + "pull_request:102|pull_request:121", + "pull_request:102|pull_request:1861", + "pull_request:102|pull_request:252", + "pull_request:1020|pull_request:415", + "pull_request:1077|pull_request:2329", + "pull_request:1162|pull_request:2006", + "pull_request:1162|pull_request:312", + "pull_request:1182|pull_request:1610", + "pull_request:119|pull_request:15", + "pull_request:1194|pull_request:2208", + "pull_request:120|pull_request:121", + "pull_request:120|pull_request:252", + "pull_request:1218|pull_request:61", + "pull_request:1218|pull_request:63", + "pull_request:12265|pull_request:12307", + "pull_request:12265|pull_request:13115", + "pull_request:12552|pull_request:12705", + "pull_request:12573|pull_request:12932", + "pull_request:12628|pull_request:12763", + "pull_request:12782|pull_request:13424", + "pull_request:13039|pull_request:13424", + "pull_request:1360|pull_request:168", + "pull_request:1487|pull_request:1724", + "pull_request:1487|pull_request:2208", + "pull_request:1487|pull_request:2624", + "pull_request:1487|pull_request:3470", + "pull_request:1487|pull_request:429" + ], + "prepared_review_unit_hash": "3dfa5666dfab9ad7f3b3df8607435b3c02977a16441cd7d0e16a01a73d707759", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 102733, + "estimated_input_tokens": 25684, + "estimated_eval_tokens": 51624 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous, with a few superficially similar docs/tests/feature PRs but no true duplicate pairs. The only close-looking pair is the Sana-Video feature work, but the PRs target different concrete capabilities (initial inference vs image-to-video support).", + "confidence": 0.19, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue exists and the PRs do not form a real duplicate set; if one representative must be chosen, PR 12584 is the most substantial self-contained feature landing in the cluster, with broad SANA-Video docs/code/tests and high review activity.", + "best_issue_reason": null, + "best_pr_reason": "PR 12584 is the strongest single representative because it is a complete, well-reviewed feature addition with broad scope; the rest of the items are mostly unrelated docs, tests, or small fixes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8631", + "right": "pull_request:8798", + "accept": false, + "reason": "Both are loading-related bugfixes, but they touch different code paths: single-file T5 checkpoint creation vs sharded checkpoint loading from subfolders." + }, + { + "left": "pull_request:12584", + "right": "pull_request:12634", + "accept": false, + "reason": "Same SANA-Video family, but different concrete changes: 12584 adds 5s 480p inference, while 12634 adds image-to-video pipeline support and new I2V code paths." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13042", + "accept": false, + "reason": "Unrelated test fixes for different models/pipelines: AudioLDM2 ClapConfig setup vs QwenImage prompt isolation." + }, + { + "left": "pull_request:11588", + "right": "pull_request:11726", + "accept": false, + "reason": "Both concern Wan LoRA handling, but one fixes I2V LoRA support and the other expands conversion support for VACE loras; not the same change." + }, + { + "left": "pull_request:102", + "right": "pull_request:121", + "accept": false, + "reason": "Both edit examples docs, but one updates training examples while the other updates the main README examples with code blocks; separate docs changes." + }, + { + "left": "pull_request:102", + "right": "pull_request:1861", + "accept": false, + "reason": "Training examples README updates vs loss-computation changes in training scripts are different artifacts and fixes." + }, + { + "left": "pull_request:102", + "right": "pull_request:252", + "accept": false, + "reason": "Both touch README/example content, but they are different documentation updates with different scope and content." + }, + { + "left": "pull_request:1020", + "right": "pull_request:415", + "accept": false, + "reason": "Both are training-docs PRs, but 1020 is a broad docs refresh across training/fine-tuning pages while 415 is an earlier restructuring of training docs." + }, + { + "left": "pull_request:1077", + "right": "pull_request:2329", + "accept": false, + "reason": "Different test infrastructure changes: scheduler integration test refactor vs push-tests workflow refactor." + }, + { + "left": "pull_request:1162", + "right": "pull_request:2006", + "accept": false, + "reason": "Different community maintenance fixes: link updates vs checkpoint-merger behavior and loading fixes." + }, + { + "left": "pull_request:1162", + "right": "pull_request:312", + "accept": false, + "reason": "Both fix links, but they target different pages and link targets; not the same underlying issue." + }, + { + "left": "pull_request:1182", + "right": "pull_request:1610", + "accept": false, + "reason": "Docs for the LPW pipeline vs attention/xFormers pipeline test changes are unrelated." + }, + { + "left": "pull_request:119", + "right": "pull_request:15", + "accept": false, + "reason": "A variable-name bugfix in hub_utils is unrelated to the examples README output-path rename." + }, + { + "left": "pull_request:1194", + "right": "pull_request:2208", + "accept": false, + "reason": "Different documentation/workflow edits: PR docs action wiring vs xFormers docs updates." + }, + { + "left": "pull_request:120", + "right": "pull_request:121", + "accept": false, + "reason": "Both update README examples, but 120 adds the examples section and 121 adds code samples; they are complementary, not duplicates." + }, + { + "left": "pull_request:120", + "right": "pull_request:252", + "accept": false, + "reason": "Both are README example updates, but they add different example content in different edits." + }, + { + "left": "pull_request:1218", + "right": "pull_request:61", + "accept": false, + "reason": "Both are conversion-script work, but one extends the Stable Diffusion conversion script and the other adds an LDM unconditional conversion script." + }, + { + "left": "pull_request:1218", + "right": "pull_request:63", + "accept": false, + "reason": "Different conversion scripts for different pipelines (Stable Diffusion vs BDDM), so not the same fix." + }, + { + "left": "pull_request:12265", + "right": "pull_request:12307", + "accept": false, + "reason": "One is typo cleanup across assorted files; the other is a large modular-pipeline feature/fix set." + }, + { + "left": "pull_request:12265", + "right": "pull_request:13115", + "accept": false, + "reason": "Typos/cleanup PR vs modular-pipeline import/export additions for Flux2 are not the same change." + }, + { + "left": "pull_request:12552", + "right": "pull_request:12705", + "accept": false, + "reason": "Both are typo fixes, but they modify different docs/code locations and address different wording issues." + }, + { + "left": "pull_request:12573", + "right": "pull_request:12932", + "accept": false, + "reason": "Different small maintenance changes in modular pipelines: warning text vs docstring/output metadata updates." + }, + { + "left": "pull_request:12628", + "right": "pull_request:12763", + "accept": false, + "reason": "A validation check in modular pipeline internals is not the same as adding the Flux2 modular pipeline feature." + }, + { + "left": "pull_request:12782", + "right": "pull_request:13424", + "accept": false, + "reason": "Unrelated test fixes for HunyuanVideo offloading vs AutoencoderDC dtype inference." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13424", + "accept": false, + "reason": "Different test fixes for Wan layerwise upcasting vs AutoencoderDC tests; no shared code path." + }, + { + "left": "pull_request:1360", + "right": "pull_request:168", + "accept": false, + "reason": "Community k-diffusion pipeline addition is a separate feature from the original stable diffusion pipeline landing." + }, + { + "left": "pull_request:1487", + "right": "pull_request:1724", + "accept": false, + "reason": "Both mention xFormers docs, but 1487 adds xFormers API docs while 1724 recommends xFormers broadly; not duplicate work." + }, + { + "left": "pull_request:1487", + "right": "pull_request:2208", + "accept": false, + "reason": "Different xFormers documentation edits at different times and scopes." + }, + { + "left": "pull_request:1487", + "right": "pull_request:2624", + "accept": false, + "reason": "xFormers docs are unrelated to adding cache_dir documentation to pipeline_utils." + }, + { + "left": "pull_request:1487", + "right": "pull_request:3470", + "accept": false, + "reason": "xFormers documentation and IF DreamBooth documentation are unrelated changes." + }, + { + "left": "pull_request:1487", + "right": "pull_request:429", + "accept": false, + "reason": "xFormers docs vs configuration docs touch different parts of the documentation and codebase." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 262, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:1487", + "pull_request:1510", + "pull_request:168", + "pull_request:1763", + "pull_request:1950", + "pull_request:1968", + "pull_request:2062", + "pull_request:21", + "pull_request:2197", + "pull_request:2290", + "pull_request:2333", + "pull_request:2367", + "pull_request:2388", + "pull_request:2406", + "pull_request:2433", + "pull_request:2533", + "pull_request:2540", + "pull_request:2569", + "pull_request:2587", + "pull_request:260", + "pull_request:2604", + "pull_request:2608", + "pull_request:2715", + "pull_request:2788", + "pull_request:2812", + "pull_request:2814", + "pull_request:3166", + "pull_request:3343", + "pull_request:353", + "pull_request:356", + "pull_request:3676", + "pull_request:3744", + "pull_request:3798", + "pull_request:3821", + "pull_request:407", + "pull_request:408", + "pull_request:421", + "pull_request:4275", + "pull_request:429", + "pull_request:43", + "pull_request:4479", + "pull_request:4721", + "pull_request:500", + "pull_request:61", + "pull_request:63", + "pull_request:718", + "pull_request:75", + "pull_request:798" + ], + "soft_pairs": [ + "pull_request:1487|pull_request:500", + "pull_request:1510|pull_request:61", + "pull_request:1510|pull_request:63", + "pull_request:168|pull_request:356", + "pull_request:1763|pull_request:75", + "pull_request:1950|pull_request:3343", + "pull_request:1968|pull_request:2062", + "pull_request:21|pull_request:43", + "pull_request:2197|pull_request:260", + "pull_request:2197|pull_request:353", + "pull_request:2197|pull_request:407", + "pull_request:2197|pull_request:798", + "pull_request:2290|pull_request:3166", + "pull_request:2290|pull_request:408", + "pull_request:2333|pull_request:421", + "pull_request:2333|pull_request:4275", + "pull_request:2367|pull_request:3798", + "pull_request:2388|pull_request:2406", + "pull_request:2433|pull_request:429", + "pull_request:2433|pull_request:500", + "pull_request:2533|pull_request:3676", + "pull_request:2540|pull_request:2788", + "pull_request:2540|pull_request:3744", + "pull_request:2569|pull_request:3821", + "pull_request:2587|pull_request:3821", + "pull_request:2604|pull_request:4275", + "pull_request:2604|pull_request:4721", + "pull_request:2608|pull_request:2715", + "pull_request:2608|pull_request:4479", + "pull_request:2715|pull_request:718", + "pull_request:2812|pull_request:2814" + ], + "prepared_review_unit_hash": "1f350d5380b8f4e4ec5e9a615a1d9934b2b5ee0244d0c7425d41c42c222237e5", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 100221, + "estimated_input_tokens": 25056, + "estimated_eval_tokens": 50368 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a false-positive cluster: the items span unrelated docs nits, code fixes, example changes, and translation updates. The soft edges are driven by shared filenames or broad subsystem names, not the same underlying bug/change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR covers the cluster; the items are unrelated docs and code changes with only superficial topic overlap.", + "best_issue_reason": null, + "best_pr_reason": "No clear best PR. The largest PRs are still one-off fixes or docs updates, not a representative duplicate target for the whole cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1487", + "right": "pull_request:500", + "accept": false, + "reason": "Unrelated docs changes: xformers API docs vs configuration subfolder docs." + }, + { + "left": "pull_request:1510", + "right": "pull_request:61", + "accept": false, + "reason": "Both are conversion-related, but one improves a broad converter while the other adds an unconditional conversion script." + }, + { + "left": "pull_request:1510", + "right": "pull_request:63", + "accept": false, + "reason": "Different conversion targets and code paths: scheduler/pipeline conversion work vs BDDM conversion script." + }, + { + "left": "pull_request:168", + "right": "pull_request:356", + "accept": false, + "reason": "Stable diffusion pipeline addition is unrelated to text-to-image fine-tuning example code." + }, + { + "left": "pull_request:1763", + "right": "pull_request:75", + "accept": false, + "reason": "Both touch attention code, but one fixes mask handling and the other changes Glide attention initialization." + }, + { + "left": "pull_request:1950", + "right": "pull_request:3343", + "accept": false, + "reason": "Different bugs: one is UNet graph config / OneFlow support, the other is a DPM-Solver config reference fix." + }, + { + "left": "pull_request:1968", + "right": "pull_request:2062", + "accept": false, + "reason": "Both are CPU offload fixes, but for different pipelines and different offload code paths." + }, + { + "left": "pull_request:21", + "right": "pull_request:43", + "accept": false, + "reason": "Different models/pipelines and different work: model tests vs model card/template and pipeline migration." + }, + { + "left": "pull_request:2197", + "right": "pull_request:260", + "accept": false, + "reason": "Inpaint pipeline typo fix vs README typo fix; unrelated artifacts and scopes." + }, + { + "left": "pull_request:2197", + "right": "pull_request:353", + "accept": false, + "reason": "Different typo fixes in unrelated pipeline code; same broad repo area is not enough." + }, + { + "left": "pull_request:2197", + "right": "pull_request:407", + "accept": false, + "reason": "Different stable-diffusion pipeline typos; one is an inpaint signature fix, the other attention slicing API cleanup." + }, + { + "left": "pull_request:2197", + "right": "pull_request:798", + "accept": false, + "reason": "Different docstring/signature typo fixes in separate pipeline/model files." + }, + { + "left": "pull_request:2290", + "right": "pull_request:3166", + "accept": false, + "reason": "Pix2Pix docs typo vs textual inversion loader doc/example fix." + }, + { + "left": "pull_request:2290", + "right": "pull_request:408", + "accept": false, + "reason": "Pix2Pix docs formatting fix is unrelated to general docs typo cleanup." + }, + { + "left": "pull_request:2333", + "right": "pull_request:421", + "accept": false, + "reason": "Different documentation pages and different fixes: ethical guidelines formatting vs scheduler docs typo cleanup." + }, + { + "left": "pull_request:2333", + "right": "pull_request:4275", + "accept": false, + "reason": "Ethical guidelines docs fix is unrelated to AutoPipeline docs." + }, + { + "left": "pull_request:2367", + "right": "pull_request:3798", + "accept": false, + "reason": "Training-script feature addition vs comment cleanup; same files but not the same underlying change." + }, + { + "left": "pull_request:2388", + "right": "pull_request:2406", + "accept": false, + "reason": "Same docs page, but 2388 adds the whole page and 2406 adds follow-up nits; not the same underlying change." + }, + { + "left": "pull_request:2433", + "right": "pull_request:429", + "accept": false, + "reason": "SDEdit note in img2img docs vs configuration docs; unrelated documentation topics." + }, + { + "left": "pull_request:2433", + "right": "pull_request:500", + "accept": false, + "reason": "Img2Img docs note vs configuration subfolder docs; no shared bug or change." + }, + { + "left": "pull_request:2533", + "right": "pull_request:3676", + "accept": false, + "reason": "Slow-test workflow/rc1 setup is unrelated to a doc-builder test PR." + }, + { + "left": "pull_request:2540", + "right": "pull_request:2788", + "accept": false, + "reason": "ControlNet section docs vs training section deduplication; different pages and objectives." + }, + { + "left": "pull_request:2540", + "right": "pull_request:3744", + "accept": false, + "reason": "ControlNet docs correction vs workflow secret rename; unrelated changes." + }, + { + "left": "pull_request:2569", + "right": "pull_request:3821", + "accept": false, + "reason": "OpenVINO docs vs translated Korean docs; different content and purpose." + }, + { + "left": "pull_request:2587", + "right": "pull_request:3821", + "accept": false, + "reason": "Chinese translated docs set vs Korean translated docs page; separate translation additions." + }, + { + "left": "pull_request:2604", + "right": "pull_request:4275", + "accept": false, + "reason": "Colab link fix in index vs AutoPipeline docs correction; unrelated docs changes." + }, + { + "left": "pull_request:2604", + "right": "pull_request:4721", + "accept": false, + "reason": "Colab link fix vs Korean DreamBooth translation fix; different documents and edits." + }, + { + "left": "pull_request:2608", + "right": "pull_request:2715", + "accept": false, + "reason": "README punctuation fixes vs broad code/docs typo cleanup across models and pipelines." + }, + { + "left": "pull_request:2608", + "right": "pull_request:4479", + "accept": false, + "reason": "README punctuation fixes are unrelated to miscellaneous pipeline typo fixes." + }, + { + "left": "pull_request:2715", + "right": "pull_request:718", + "accept": false, + "reason": "Broad typo cleanup across model files is not the same change as the earlier mixed docs/example typo PR." + }, + { + "left": "pull_request:2812", + "right": "pull_request:2814", + "accept": false, + "reason": "Stable unCLIP docs update vs Stable unCLIP implementation/offload changes; related feature area but different artifacts and scope." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 263, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:2812", + "pull_request:2813", + "pull_request:2814", + "pull_request:2823", + "pull_request:2861", + "pull_request:2922", + "pull_request:3", + "pull_request:3068", + "pull_request:3069", + "pull_request:3166", + "pull_request:3183", + "pull_request:3241", + "pull_request:3318", + "pull_request:3470", + "pull_request:3554", + "pull_request:3625", + "pull_request:3687", + "pull_request:37", + "pull_request:3850", + "pull_request:3979", + "pull_request:3985", + "pull_request:407", + "pull_request:4082", + "pull_request:42", + "pull_request:4276", + "pull_request:4380", + "pull_request:4652", + "pull_request:4721", + "pull_request:5250", + "pull_request:5280", + "pull_request:531", + "pull_request:5407", + "pull_request:5723", + "pull_request:5727", + "pull_request:580", + "pull_request:5911", + "pull_request:5983", + "pull_request:6017", + "pull_request:6018", + "pull_request:6215", + "pull_request:6466", + "pull_request:6475", + "pull_request:6759", + "pull_request:75", + "pull_request:774", + "pull_request:824", + "pull_request:828", + "pull_request:88" + ], + "soft_pairs": [ + "pull_request:2812|pull_request:2922", + "pull_request:2813|pull_request:2814", + "pull_request:2813|pull_request:2922", + "pull_request:2814|pull_request:2823", + "pull_request:2861|pull_request:4276", + "pull_request:3|pull_request:88", + "pull_request:3068|pull_request:3318", + "pull_request:3069|pull_request:828", + "pull_request:3166|pull_request:4721", + "pull_request:3183|pull_request:3979", + "pull_request:3241|pull_request:5280", + "pull_request:3318|pull_request:3554", + "pull_request:3470|pull_request:5723", + "pull_request:3625|pull_request:3687", + "pull_request:3687|pull_request:6017", + "pull_request:3687|pull_request:6018", + "pull_request:37|pull_request:75", + "pull_request:3850|pull_request:6215", + "pull_request:3979|pull_request:4082", + "pull_request:3985|pull_request:4380", + "pull_request:407|pull_request:75", + "pull_request:4082|pull_request:4652", + "pull_request:42|pull_request:824", + "pull_request:4380|pull_request:6466", + "pull_request:5250|pull_request:5911", + "pull_request:531|pull_request:774", + "pull_request:5407|pull_request:6759", + "pull_request:5723|pull_request:5727", + "pull_request:580|pull_request:824", + "pull_request:5983|pull_request:6466", + "pull_request:5983|pull_request:6475" + ], + "prepared_review_unit_hash": "e4aa36be9bfda169cfd363e31446f77c02b94c6f282de70453b191518271d24b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 94631, + "estimated_input_tokens": 23658, + "estimated_eval_tokens": 47572 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated docs, tests, and small feature/fix PRs. The only clear duplicate-like item is the near-identical Stable unCLIP docs patch pair (2812/2813, not one of the provided soft edges); most other soft pairs are same subsystem at best, but not the same concrete bug/change.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 2812 is the cleanest duplicate representative in the only obvious exact-match pair (2812/2813): same Stable unCLIP docs patch, same hunk and filenames, created minutes apart.", + "best_issue_reason": null, + "best_pr_reason": "PR 2814 is the most substantive representative here: it actually completes the Stable UnCLIP pipeline code path across multiple source files, whereas most other items are docs/test tweaks.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2812", + "right": "pull_request:2922", + "accept": false, + "reason": "Both are Stable unCLIP-related docs, but they change different pages and different content; not the same change." + }, + { + "left": "pull_request:2813", + "right": "pull_request:2814", + "accept": false, + "reason": "Docs-only Stable unCLIP update vs core pipeline implementation; related area, not the same bug or change." + }, + { + "left": "pull_request:2813", + "right": "pull_request:2922", + "accept": false, + "reason": "Different docs edits on different files; same topic, but not duplicate work." + }, + { + "left": "pull_request:2814", + "right": "pull_request:2823", + "accept": false, + "reason": "Code completion PR vs follow-up docs improvement; not the same underlying change." + }, + { + "left": "pull_request:2861", + "right": "pull_request:4276", + "accept": false, + "reason": "StableUnCLIPImg2Img test fix vs generic pipeline import test; unrelated." + }, + { + "left": "pull_request:3", + "right": "pull_request:88", + "accept": false, + "reason": "Foundational hub-loading work vs DDPM implementation work; different code paths and goals." + }, + { + "left": "pull_request:3068", + "right": "pull_request:3318", + "accept": false, + "reason": "Textual inversion docs cleanup vs IF textual inversion feature work; not the same artifact." + }, + { + "left": "pull_request:3069", + "right": "pull_request:828", + "accept": false, + "reason": "Both are Flax test updates, but they target different test suites and fixes." + }, + { + "left": "pull_request:3166", + "right": "pull_request:4721", + "accept": false, + "reason": "Text inversion loader doc example vs unrelated Korean DreamBooth doc edits." + }, + { + "left": "pull_request:3183", + "right": "pull_request:3979", + "accept": false, + "reason": "DreamBooth LoRA checkpoint bug vs SDXL attention bug; different subsystems." + }, + { + "left": "pull_request:3241", + "right": "pull_request:5280", + "accept": false, + "reason": "UnCLIP test tolerance change vs ControlNet fast-test resizing; unrelated tests." + }, + { + "left": "pull_request:3318", + "right": "pull_request:3554", + "accept": false, + "reason": "IF textual inversion WIP vs separate textual inversion/example branch; not the same change." + }, + { + "left": "pull_request:3470", + "right": "pull_request:5723", + "accept": false, + "reason": "IF DreamBooth docs vs LCM docs; different model families and docs." + }, + { + "left": "pull_request:3625", + "right": "pull_request:3687", + "accept": false, + "reason": "Both are doc-builder test PRs, but they modify different files with different test intents." + }, + { + "left": "pull_request:3687", + "right": "pull_request:6017", + "accept": false, + "reason": "Same broad doc-builder testing theme, but different edits and different PRs." + }, + { + "left": "pull_request:3687", + "right": "pull_request:6018", + "accept": false, + "reason": "Same broad doc-builder testing theme, but different edits and different PRs." + }, + { + "left": "pull_request:37", + "right": "pull_request:75", + "accept": false, + "reason": "Attention refactor/merge vs Glide-specific attention fix; same subsystem, not same change." + }, + { + "left": "pull_request:3850", + "right": "pull_request:6215", + "accept": false, + "reason": "Audio diffusion test fix vs PixArt test initialization fix; unrelated test suites." + }, + { + "left": "pull_request:3979", + "right": "pull_request:4082", + "accept": false, + "reason": "SDXL bug fix vs requirement pin fix; different problem classes." + }, + { + "left": "pull_request:3985", + "right": "pull_request:4380", + "accept": false, + "reason": "SDXL doc polish vs broader docs cleanup in other training pages; not duplicate content." + }, + { + "left": "pull_request:407", + "right": "pull_request:75", + "accept": false, + "reason": "Attention slicing typo fix vs Glide attention implementation fix; different changes." + }, + { + "left": "pull_request:4082", + "right": "pull_request:4652", + "accept": false, + "reason": "Package requirement tweak vs LoRA SDXL test adjustment; unrelated." + }, + { + "left": "pull_request:42", + "right": "pull_request:824", + "accept": false, + "reason": "Old LDM unconditional test addition vs Flax pipeline test work; different code paths." + }, + { + "left": "pull_request:4380", + "right": "pull_request:6466", + "accept": false, + "reason": "Different docs pages and edits; both minor docs fixes, but not the same change." + }, + { + "left": "pull_request:5250", + "right": "pull_request:5911", + "accept": false, + "reason": "DDPO docs entry vs IP-Adapter Full support implementation; unrelated." + }, + { + "left": "pull_request:531", + "right": "pull_request:774", + "accept": false, + "reason": "Two separate test-fix PRs in different eras and files; not the same bug." + }, + { + "left": "pull_request:5407", + "right": "pull_request:6759", + "accept": false, + "reason": "Mixed-precision training fix vs InstantID prompt-embedding shape fix; different pipelines and bugs." + }, + { + "left": "pull_request:5723", + "right": "pull_request:5727", + "accept": false, + "reason": "LCM docs vs LCM scripts; same feature area, but docs and implementation are separate changes." + }, + { + "left": "pull_request:580", + "right": "pull_request:824", + "accept": false, + "reason": "Early Flax scheduler test work vs later Flax pipeline test work; different changes." + }, + { + "left": "pull_request:5983", + "right": "pull_request:6466", + "accept": false, + "reason": "Both touch SVD docs, but they make different edits to different sections; not a duplicate change." + }, + { + "left": "pull_request:5983", + "right": "pull_request:6475", + "accept": false, + "reason": "SVD doc cleanup vs Stable Video Diffusion VAE type fix; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 264, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:11428", + "pull_request:11522", + "pull_request:11546", + "pull_request:12069", + "pull_request:12490", + "pull_request:130", + "pull_request:13275", + "pull_request:13442", + "pull_request:13469", + "pull_request:1564", + "pull_request:2001", + "pull_request:2062", + "pull_request:327", + "pull_request:3397", + "pull_request:343", + "pull_request:3471", + "pull_request:349", + "pull_request:4262", + "pull_request:4666", + "pull_request:4749", + "pull_request:5201", + "pull_request:5639", + "pull_request:5983", + "pull_request:6645", + "pull_request:6647", + "pull_request:6759", + "pull_request:70", + "pull_request:7015", + "pull_request:7024", + "pull_request:7050", + "pull_request:7068", + "pull_request:7118", + "pull_request:7206", + "pull_request:7451", + "pull_request:7564", + "pull_request:78", + "pull_request:7825", + "pull_request:7977", + "pull_request:800", + "pull_request:8498", + "pull_request:8656", + "pull_request:8845", + "pull_request:9040", + "pull_request:9086", + "pull_request:9206", + "pull_request:9597", + "pull_request:971" + ], + "soft_pairs": [ + "pull_request:5983|pull_request:7015", + "pull_request:6647|pull_request:7564", + "pull_request:70|pull_request:78", + "pull_request:7050|pull_request:7068", + "pull_request:7050|pull_request:7451", + "pull_request:7068|pull_request:7118", + "pull_request:7564|pull_request:8656", + "pull_request:7825|pull_request:9206", + "pull_request:7977|pull_request:8845", + "pull_request:800|pull_request:971", + "pull_request:8498|pull_request:8845", + "pull_request:11522|pull_request:12069", + "pull_request:12490|pull_request:13442", + "pull_request:327|pull_request:343", + "pull_request:3397|pull_request:3471", + "pull_request:4262|pull_request:4666", + "pull_request:11428|pull_request:11546", + "pull_request:2001|pull_request:2062", + "pull_request:4749|pull_request:5201", + "pull_request:6645|pull_request:6759", + "pull_request:327|pull_request:349", + "pull_request:13275|pull_request:13469", + "pull_request:5639|pull_request:7024", + "pull_request:7206|pull_request:9086", + "pull_request:9040|pull_request:9597", + "pull_request:130|pull_request:1564" + ], + "prepared_review_unit_hash": "a91719b292ba1a940b22d79d10ff9f62e9c78c0d80feea855a7ecd34706f6fe4", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 97871, + "estimated_input_tokens": 24468, + "estimated_eval_tokens": 49192 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated docs/test/feature PRs that only look similar at the subsystem or wording level. The only strong duplicate-like pair is the two AudioLDM2 fixes, which both patch the same `generate_language_model` compatibility path for newer Transformers.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:12069 is the best canonical PR because it is the more complete AudioLDM2 compatibility fix: it keeps the version guard, moves `model_kwargs` handling into the shared path, and includes the test update, making it a better representative than the earlier partial fix in 11522.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:12069 is the strongest representative of the only true duplicate pair in this cluster; it fixes the concrete failing AudioLDM2 code path rather than a broader or unrelated docs/test change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5983", + "right": "pull_request:7015", + "accept": false, + "reason": "Both are docs edits, but they target different pages and unrelated fixes (SVD docs vs single-file SDXL docs)." + }, + { + "left": "pull_request:6647", + "right": "pull_request:7564", + "accept": false, + "reason": "Both touch tests, but they cover different pipelines and different failure modes; not the same bug." + }, + { + "left": "pull_request:70", + "right": "pull_request:78", + "accept": false, + "reason": "Both are broad architectural mid-block changes, but not the same concrete change or failure; too coarse to merge as duplicates." + }, + { + "left": "pull_request:7050", + "right": "pull_request:7068", + "accept": false, + "reason": "Both are typo/docs PRs, but they edit different files and unrelated content." + }, + { + "left": "pull_request:7050", + "right": "pull_request:7451", + "accept": false, + "reason": "Same general docs-typo theme, but different documentation areas and unrelated edits." + }, + { + "left": "pull_request:7068", + "right": "pull_request:7118", + "accept": false, + "reason": "Both are docs typo cleanups, but they fix different docs pages and examples." + }, + { + "left": "pull_request:7564", + "right": "pull_request:8656", + "accept": false, + "reason": "Both are test maintenance PRs, but they address different model behaviors and checkpoint handling." + }, + { + "left": "pull_request:7825", + "right": "pull_request:9206", + "accept": false, + "reason": "Different test suites and different issues: device-map/offload tests vs xformers attention tests." + }, + { + "left": "pull_request:7977", + "right": "pull_request:8845", + "accept": false, + "reason": "Both are typo fixes, but they are in different docs/test contexts and not the same change." + }, + { + "left": "pull_request:800", + "right": "pull_request:971", + "accept": false, + "reason": "Both relate to diffusion model additions, but they are different models/features (VQ Diffusion vs bit diffusion)." + }, + { + "left": "pull_request:8498", + "right": "pull_request:8845", + "accept": false, + "reason": "Unrelated fixes: a single doc typo versus typo corrections in PAG tests." + }, + { + "left": "pull_request:11522", + "right": "pull_request:12069", + "accept": true, + "reason": "Same AudioLDM2 `generate_language_model` compatibility bug for Transformers main/newer versions; 12069 is a direct follow-up/complete fix to 11522." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13442", + "accept": false, + "reason": "Different topics: a CLAP config test fix vs training-script test fixes." + }, + { + "left": "pull_request:327", + "right": "pull_request:343", + "accept": false, + "reason": "Both are type-hint cleanup PRs, but they touch different schedulers/pipelines and are not the same underlying change." + }, + { + "left": "pull_request:3397", + "right": "pull_request:3471", + "accept": false, + "reason": "Both are docs, but one is a local-path note and the other is a torch.compile note for DreamBooth." + }, + { + "left": "pull_request:4262", + "right": "pull_request:4666", + "accept": false, + "reason": "Both involve SDXL/LoRA, but one is DreamBooth training/model fixes and the other is ControlNet SDXL LoRA support; different code paths." + }, + { + "left": "pull_request:11428", + "right": "pull_request:11546", + "accept": false, + "reason": "FramePack pipeline addition vs a LoRA-loader mapping for that new model; related feature area, but not the same change." + }, + { + "left": "pull_request:2001", + "right": "pull_request:2062", + "accept": false, + "reason": "Completely different scopes: a ColossalAI example typo fix vs Paint by Example CPU offload behavior." + }, + { + "left": "pull_request:4749", + "right": "pull_request:5201", + "accept": false, + "reason": "Both are LoRA-related, but one is SDXL DreamBooth checkpointing and the other is xformers/LoRA processor behavior." + }, + { + "left": "pull_request:6645", + "right": "pull_request:6759", + "accept": false, + "reason": "Both are community example edits, but they implement different community pipelines with no shared bug/change." + }, + { + "left": "pull_request:327", + "right": "pull_request:349", + "accept": false, + "reason": "Both are type-hint PRs, but they affect different scheduler classes and are unrelated." + }, + { + "left": "pull_request:13275", + "right": "pull_request:13469", + "accept": false, + "reason": "Different test fixes in different areas: Hunyuan/LLaVA kwargs vs deprecated attention-processor testing." + }, + { + "left": "pull_request:5639", + "right": "pull_request:7024", + "accept": false, + "reason": "Both are Animatediff docs, but they add different feature sections (Motion LoRAs vs AnimateLCM)." + }, + { + "left": "pull_request:7206", + "right": "pull_request:9086", + "accept": false, + "reason": "Both touch training scripts, but one is tag metadata and the other is actual Flux DreamBooth/LoRA training code." + }, + { + "left": "pull_request:9040", + "right": "pull_request:9597", + "accept": false, + "reason": "Same general 'differential diffusion' concept, but for different community pipelines/models; not the same patch." + }, + { + "left": "pull_request:130", + "right": "pull_request:1564", + "accept": false, + "reason": "Both mention accelerate config, but one adds a setup step and the other adds a default config option for DreamBooth docs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 265, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10858", + "pull_request:12006", + "pull_request:12018", + "pull_request:12041", + "pull_request:2172", + "pull_request:2406", + "pull_request:2490", + "pull_request:2526", + "pull_request:2595", + "pull_request:2637", + "pull_request:2693", + "pull_request:2748", + "pull_request:3351", + "pull_request:3476", + "pull_request:3586", + "pull_request:3589", + "pull_request:3590", + "pull_request:3942", + "pull_request:3945", + "pull_request:4262", + "pull_request:4343", + "pull_request:4498", + "pull_request:4572", + "pull_request:4626", + "pull_request:4749", + "pull_request:4833", + "pull_request:4862", + "pull_request:5003", + "pull_request:5157", + "pull_request:5201", + "pull_request:5615", + "pull_request:5665", + "pull_request:5768", + "pull_request:5828", + "pull_request:6342", + "pull_request:6759", + "pull_request:7050", + "pull_request:7131", + "pull_request:7246", + "pull_request:7825", + "pull_request:7827", + "pull_request:788", + "pull_request:8230", + "pull_request:8797", + "pull_request:8913", + "pull_request:9506", + "pull_request:9696" + ], + "soft_pairs": [ + "pull_request:2693|pull_request:788", + "pull_request:4833|pull_request:5201", + "pull_request:8913|pull_request:9506", + "pull_request:10858|pull_request:12018", + "pull_request:2172|pull_request:2595", + "pull_request:2406|pull_request:2490", + "pull_request:2526|pull_request:3942", + "pull_request:2637|pull_request:2748", + "pull_request:3945|pull_request:4343", + "pull_request:4626|pull_request:6342", + "pull_request:4862|pull_request:5665", + "pull_request:5003|pull_request:5157", + "pull_request:5768|pull_request:5828", + "pull_request:7246|pull_request:8797", + "pull_request:7825|pull_request:9696", + "pull_request:7827|pull_request:8230", + "pull_request:12006|pull_request:12041", + "pull_request:3351|pull_request:3476", + "pull_request:3589|pull_request:4498", + "pull_request:3586|pull_request:3590", + "pull_request:4262|pull_request:5615", + "pull_request:4572|pull_request:4749", + "pull_request:4749|pull_request:6759", + "pull_request:7050|pull_request:7131" + ], + "prepared_review_unit_hash": "1765ba9f485742a1de799413f2913d2088f3e3129277a29a2a4c874d5e335af5", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 94485, + "estimated_input_tokens": 23622, + "estimated_eval_tokens": 47500 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous cluster, not a duplicate set: it mixes unrelated docs edits, example updates, scheduler/LoRA fixes, ControlNet bugs, tests, and workflow changes. There is no single underlying bug or change tying these PRs together.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits the cluster. The items span many unrelated code paths; the broadest docs cleanup (PR 7050) is only a loose representative, not a true duplicate anchor.", + "best_issue_reason": null, + "best_pr_reason": "PR 7050 is the closest representative only because it is a broad cross-file docs typo cleanup, but it still does not subsume the rest of the unrelated fixes in this cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2693", + "right": "pull_request:788", + "accept": false, + "reason": "One is an inpaint doc image-link fix; the other is a PIL sampler warning/code compatibility change. Different artifacts and no shared bug." + }, + { + "left": "pull_request:4833", + "right": "pull_request:5201", + "accept": false, + "reason": "Both involve LoRA, but PR 4833 fixes unfuse/save behavior while PR 5201 fixes xformers/attention processor LoRA handling across models. Different concrete code paths." + }, + { + "left": "pull_request:8913", + "right": "pull_request:9506", + "accept": false, + "reason": "PR 8913 fixes sharded checkpoint loading from a subfolder; PR 9506 is a docs change adding an import and path fix. Unrelated." + }, + { + "left": "pull_request:10858", + "right": "pull_request:12018", + "accept": false, + "reason": "Both are docs edits, but one changes a pipeline overview table entry while the other fixes a quantization docs link label. Not the same issue." + }, + { + "left": "pull_request:2172", + "right": "pull_request:2595", + "accept": false, + "reason": "Multi-GPU validation in training scripts and scheduler switching for DEIS/DPM/UniPC are unrelated problems in different subsystems." + }, + { + "left": "pull_request:2406", + "right": "pull_request:2490", + "accept": false, + "reason": "These are separate documentation updates in the same file, but one is a generic nits edit and the other adds a MultiDiffusion section. Not duplicates." + }, + { + "left": "pull_request:2526", + "right": "pull_request:3942", + "accept": false, + "reason": "PR 2526 adjusts slow-test tolerances; PR 3942 fixes xformers-related NumPy test behavior. Different test failures and causes." + }, + { + "left": "pull_request:2637", + "right": "pull_request:2748", + "accept": false, + "reason": "These are unrelated typo fixes in different docs pages; they do not reflect the same underlying change." + }, + { + "left": "pull_request:3945", + "right": "pull_request:4343", + "accept": false, + "reason": "Both add a LoRA rank argument, but to different DreamBooth training scripts (base vs SDXL). Similar feature family, not the same concrete change." + }, + { + "left": "pull_request:4626", + "right": "pull_request:6342", + "accept": false, + "reason": "One fixes text2image docs links; the other adds an installation note about Python 3.12 on Windows. Unrelated." + }, + { + "left": "pull_request:4862", + "right": "pull_request:5665", + "accept": false, + "reason": "Different ControlNet img2img bugs: one is an SDXL MultiControlNetModel height/width issue, the other passes dimensions into image preprocessing. Not the same bug." + }, + { + "left": "pull_request:5003", + "right": "pull_request:5157", + "accept": false, + "reason": "PR 5003 fixes LoRA fuse/unfuse logic; PR 5157 fixes LoRA conversion naming for ControlNet/Kohya formats. Different code paths." + }, + { + "left": "pull_request:5768", + "right": "pull_request:5828", + "accept": false, + "reason": "Docs cleanup versus a memory test annotation for the consistency decoder. No shared bug or change." + }, + { + "left": "pull_request:7246", + "right": "pull_request:8797", + "accept": false, + "reason": "VAE tiling device handling and sharded checkpoint tests are unrelated test fixes." + }, + { + "left": "pull_request:7825", + "right": "pull_request:9696", + "accept": false, + "reason": "Model offload test guards and a GitHub Actions workflow update are unrelated." + }, + { + "left": "pull_request:7827", + "right": "pull_request:8230", + "accept": false, + "reason": "Both mention CPU offloading, but one is a docstring example fix and the other is a broad usage/docs update. Not the same concrete issue." + }, + { + "left": "pull_request:12006", + "right": "pull_request:12041", + "accept": false, + "reason": "One adds Wan 5b i2v support; the other fixes VAE patchify behavior. Different components and bugs." + }, + { + "left": "pull_request:3351", + "right": "pull_request:3476", + "accept": false, + "reason": "Both add Karras sigmas, but to different schedulers with separate implementations. Similar enhancement theme, not one duplicate change." + }, + { + "left": "pull_request:3589", + "right": "pull_request:4498", + "accept": false, + "reason": "PR 3589 updates README examples/tricks; PR 4498 adds SDXL reference pipeline support. Docs-only vs implementation work, not duplicates." + }, + { + "left": "pull_request:3586", + "right": "pull_request:3590", + "accept": false, + "reason": "These are separate community pipelines/example additions (tiling vs canvas), not the same underlying feature." + }, + { + "left": "pull_request:4262", + "right": "pull_request:5615", + "accept": false, + "reason": "Different SDXL-related changes in different pipelines: DreamBooth LoRA fixes versus reverting load_lora behavior in the T2I adapter." + }, + { + "left": "pull_request:4572", + "right": "pull_request:4749", + "accept": false, + "reason": "One renames network_alpha handling in the text-to-image LoRA script; the other fixes SDXL DreamBooth LoRA checkpoint loading. Different problems." + }, + { + "left": "pull_request:4749", + "right": "pull_request:6759", + "accept": false, + "reason": "Checkpoint loading for SDXL DreamBooth LoRA and prompt embedding repetition in InstantID are unrelated." + }, + { + "left": "pull_request:7050", + "right": "pull_request:7131", + "accept": false, + "reason": "Both are typo/docs cleanup PRs, but they touch different documentation areas and do not address the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 266, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10192", + "pull_request:11353", + "pull_request:11479", + "pull_request:11655", + "pull_request:11726", + "pull_request:11728", + "pull_request:1229", + "pull_request:1269", + "pull_request:1508", + "pull_request:1527", + "pull_request:154", + "pull_request:1554", + "pull_request:1564", + "pull_request:168", + "pull_request:1788", + "pull_request:1951", + "pull_request:2009", + "pull_request:2367", + "pull_request:2565", + "pull_request:2693", + "pull_request:2892", + "pull_request:2928", + "pull_request:2976", + "pull_request:3370", + "pull_request:3446", + "pull_request:3471", + "pull_request:3568", + "pull_request:3641", + "pull_request:4272", + "pull_request:4621", + "pull_request:4856", + "pull_request:5090", + "pull_request:5299", + "pull_request:5391", + "pull_request:5411", + "pull_request:5759", + "pull_request:6119", + "pull_request:6645", + "pull_request:7200", + "pull_request:7319", + "pull_request:7451", + "pull_request:7857", + "pull_request:7919", + "pull_request:822", + "pull_request:827", + "pull_request:9207", + "pull_request:98" + ], + "soft_pairs": [ + "pull_request:7451|pull_request:9207", + "pull_request:11353|pull_request:11726", + "pull_request:11479|pull_request:11728", + "pull_request:1229|pull_request:1564", + "pull_request:1229|pull_request:2367", + "pull_request:1527|pull_request:2693", + "pull_request:154|pull_request:168", + "pull_request:154|pull_request:98", + "pull_request:1554|pull_request:2565", + "pull_request:1788|pull_request:3641", + "pull_request:1951|pull_request:3370", + "pull_request:2367|pull_request:2928", + "pull_request:2892|pull_request:3446", + "pull_request:2892|pull_request:3641", + "pull_request:2892|pull_request:4272", + "pull_request:2976|pull_request:4856", + "pull_request:3568|pull_request:5411", + "pull_request:4621|pull_request:6119", + "pull_request:5090|pull_request:6645", + "pull_request:5391|pull_request:5759", + "pull_request:7319|pull_request:7857", + "pull_request:822|pull_request:827", + "pull_request:10192|pull_request:11655", + "pull_request:1269|pull_request:2009", + "pull_request:1508|pull_request:3471", + "pull_request:5299|pull_request:5759", + "pull_request:7200|pull_request:7919" + ], + "prepared_review_unit_hash": "45a9ba5aed339a066c7cdd04aa5be3f3d8699ce6b40bc8595835d15bd7884339", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 101394, + "estimated_input_tokens": 25349, + "estimated_eval_tokens": 50954 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a false-positive grouping of unrelated PRs: docs tweaks, feature additions, and bug fixes across different subsystems. None of the soft edges look like the same underlying change, so I would not deduplicate any pair.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the set is heterogeneous and does not share one concrete bug/change. If forced to pick a representative, PR 5759 is the broadest docs/type-hint cleanup, but it is not a true duplicate target.", + "best_issue_reason": null, + "best_pr_reason": "PR 5759 is the closest to an umbrella representative because it spans the broadest docs/type-hint maintenance pass, but the cluster still lacks a true common underlying fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7451", + "right": "pull_request:9207", + "accept": false, + "reason": "Both are docs-only, but one fixes typos across training docs while the other updates CPU offloading usage guidance; different change targets and intent." + }, + { + "left": "pull_request:11353", + "right": "pull_request:11726", + "accept": false, + "reason": "Different Wan work: FLF2V support versus Wan LoRA conversion fixes. They touch the same project area but not the same bug or feature." + }, + { + "left": "pull_request:11479", + "right": "pull_request:11728", + "accept": false, + "reason": "One is a doc title rename for Intel Gaudi; the other adds quantization libraries to the doc-builder Docker image. Unrelated changes." + }, + { + "left": "pull_request:1229", + "right": "pull_request:1564", + "accept": false, + "reason": "Both touch DreamBooth docs, but one is about bitsandbytes installation and the other about default accelerate config. Different user problems." + }, + { + "left": "pull_request:1229", + "right": "pull_request:2367", + "accept": false, + "reason": "Install guidance for bitsandbytes is unrelated to adding checkpoint-retention controls in training scripts." + }, + { + "left": "pull_request:1527", + "right": "pull_request:2693", + "accept": false, + "reason": "Stable Diffusion inpaint tensor-input support is a code fix; the other is only a broken image link in docs." + }, + { + "left": "pull_request:154", + "right": "pull_request:168", + "accept": false, + "reason": "A conversion script for original SD checkpoints is not the same as the initial Stable Diffusion pipeline integration." + }, + { + "left": "pull_request:154", + "right": "pull_request:98", + "accept": false, + "reason": "These are different conversion scripts for different model families (LDM/Stable Diffusion vs NCSN++)." + }, + { + "left": "pull_request:1554", + "right": "pull_request:2565", + "accept": false, + "reason": "Both are docs corrections, but they target different model APIs and different dimensionality docs; no shared underlying bug." + }, + { + "left": "pull_request:1788", + "right": "pull_request:3641", + "accept": false, + "reason": "DDIM sample-prediction handling and inpainting initial-latent handling are separate scheduler/pipeline bugs." + }, + { + "left": "pull_request:1951", + "right": "pull_request:3370", + "accept": false, + "reason": "ColossalAI upgrade work is unrelated to PyTorch 2.0 documentation updates." + }, + { + "left": "pull_request:2367", + "right": "pull_request:2928", + "accept": false, + "reason": "Checkpoint-limit plumbing in training scripts and ControlNet local dataset loading are independent features." + }, + { + "left": "pull_request:2892", + "right": "pull_request:3446", + "accept": false, + "reason": "Allowing embed inputs in the upscaler pipeline is unrelated to a warning-message wording fix." + }, + { + "left": "pull_request:2892", + "right": "pull_request:3641", + "accept": false, + "reason": "These are different inpainting/upscaler behaviors with distinct code paths and failure modes." + }, + { + "left": "pull_request:2892", + "right": "pull_request:4272", + "accept": false, + "reason": "Upgrader input validation is unrelated to fixing missing optional imports in pipeline folders." + }, + { + "left": "pull_request:2976", + "right": "pull_request:4856", + "accept": false, + "reason": "Both are docs link fixes, but they address different pages and different broken links, not the same change." + }, + { + "left": "pull_request:3568", + "right": "pull_request:5411", + "accept": false, + "reason": "Both involve documentation links, but one fixes general tutorial links and the other is specifically about textual inversion training links." + }, + { + "left": "pull_request:4621", + "right": "pull_request:6119", + "accept": false, + "reason": "Multi-adapter downscale-factor validation and fp16 LoRA gradient scaling are unrelated training/runtime issues." + }, + { + "left": "pull_request:5090", + "right": "pull_request:6645", + "accept": false, + "reason": "A T2I Adapter bugfix is not the same as a community README cleanup." + }, + { + "left": "pull_request:5391", + "right": "pull_request:5759", + "accept": false, + "reason": "Both are broad docs/type-hint cleanups, but they are different edit sets with different scopes and not one concrete fix." + }, + { + "left": "pull_request:7319", + "right": "pull_request:7857", + "accept": false, + "reason": "Adding EDM scheduler docs and adding PixArt-Sigma docs are separate documentation additions." + }, + { + "left": "pull_request:822", + "right": "pull_request:827", + "accept": false, + "reason": "Flax scheduler API changes and PyTorch/Flax checkpoint-loading alignment are different compatibility fixes." + }, + { + "left": "pull_request:10192", + "right": "pull_request:11655", + "accept": false, + "reason": "Flux Control Inpaint pipeline support and bnb 8-bit LoRA loading support are different code paths and different features." + }, + { + "left": "pull_request:1269", + "right": "pull_request:2009", + "accept": false, + "reason": "Stable diffusion refactor work and textual inversion loading support are unrelated." + }, + { + "left": "pull_request:1508", + "right": "pull_request:3471", + "accept": false, + "reason": "Both are DreamBooth docs changes, but they cover different topics: prior-preservation vs torch.compile guidance." + }, + { + "left": "pull_request:5299", + "right": "pull_request:5759", + "accept": false, + "reason": "These overlap only at a high level (docs/type hints). The concrete files and edits do not indicate the same underlying bug/change." + }, + { + "left": "pull_request:7200", + "right": "pull_request:7919", + "accept": false, + "reason": "Transformers2D norm-type validation and offload behavior with device_map are separate core changes with different failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 267, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10539", + "pull_request:10810", + "pull_request:10941", + "pull_request:10983", + "pull_request:11066", + "pull_request:1136", + "pull_request:11704", + "pull_request:11739", + "pull_request:11744", + "pull_request:11856", + "pull_request:11997", + "pull_request:12006", + "pull_request:12490", + "pull_request:12523", + "pull_request:12611", + "pull_request:12782", + "pull_request:13039", + "pull_request:13055", + "pull_request:13121", + "pull_request:13217", + "pull_request:13275", + "pull_request:1508", + "pull_request:1525", + "pull_request:1982", + "pull_request:2001", + "pull_request:2339", + "pull_request:2693", + "pull_request:326", + "pull_request:3546", + "pull_request:359", + "pull_request:360", + "pull_request:3607", + "pull_request:3632", + "pull_request:5407", + "pull_request:5898", + "pull_request:6330", + "pull_request:6346", + "pull_request:6994", + "pull_request:7068", + "pull_request:7246", + "pull_request:774", + "pull_request:8571", + "pull_request:8631", + "pull_request:8797", + "pull_request:8992", + "pull_request:9206", + "pull_request:9423", + "pull_request:9676" + ], + "soft_pairs": [ + "pull_request:10941|pull_request:9676", + "pull_request:326|pull_request:359", + "pull_request:326|pull_request:360", + "pull_request:5898|pull_request:6330", + "pull_request:7246|pull_request:9206", + "pull_request:8797|pull_request:9206", + "pull_request:13121|pull_request:13217", + "pull_request:8992|pull_request:9423", + "pull_request:12490|pull_request:13275", + "pull_request:3546|pull_request:3632", + "pull_request:11739|pull_request:11744", + "pull_request:12490|pull_request:12782", + "pull_request:12490|pull_request:13039", + "pull_request:13039|pull_request:13055", + "pull_request:1508|pull_request:2339", + "pull_request:11997|pull_request:12611", + "pull_request:10539|pull_request:10810", + "pull_request:1136|pull_request:2001", + "pull_request:1136|pull_request:2693", + "pull_request:11704|pull_request:11856", + "pull_request:5407|pull_request:6346", + "pull_request:7068|pull_request:8571", + "pull_request:1982|pull_request:774", + "pull_request:10983|pull_request:11066", + "pull_request:12006|pull_request:12523", + "pull_request:1525|pull_request:3607", + "pull_request:6994|pull_request:8631" + ], + "prepared_review_unit_hash": "624ece7fc0b1f4e388f48fe4efc3c6828816a17ee1a12cdaf3663ddd3348956f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 100657, + "estimated_input_tokens": 25165, + "estimated_eval_tokens": 50586 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous: the proposed soft edges pair up superficially similar PRs (same subsystem, docs/tests, or related model families) but not the same underlying bug or change. I rejected all candidate duplicate links.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 13217 is the most substantial and activity-heavy item, so it is the best representative anchor if one must be chosen, but this set is not truly a duplicate cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10941", + "right": "pull_request:9676", + "accept": false, + "reason": "Flux true-cfg logic fix vs an import cleanup; unrelated changes." + }, + { + "left": "pull_request:326", + "right": "pull_request:359", + "accept": false, + "reason": "Type-hint cleanup in a pipeline vs scheduler typing; same style, different code paths." + }, + { + "left": "pull_request:326", + "right": "pull_request:360", + "accept": false, + "reason": "Different scheduler typing refactors, not the same change." + }, + { + "left": "pull_request:5898", + "right": "pull_request:6330", + "accept": false, + "reason": "Both are docs fixes, but on different pages and different issues." + }, + { + "left": "pull_request:7246", + "right": "pull_request:9206", + "accept": false, + "reason": "Different test fixes in different pipelines; only broad test hygiene overlaps." + }, + { + "left": "pull_request:8797", + "right": "pull_request:9206", + "accept": false, + "reason": "Unrelated model/modeling test adjustments, not one bug." + }, + { + "left": "pull_request:13121", + "right": "pull_request:13217", + "accept": false, + "reason": "LTX-2 CFG fix vs broader LTX-2.3 model support; related area, but not the same concrete change." + }, + { + "left": "pull_request:8992", + "right": "pull_request:9423", + "accept": false, + "reason": "Different community differential-diffusion pipelines for different models." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13275", + "accept": false, + "reason": "Audioldm2 test config fix vs Hunyuan test kwargs fix; unrelated." + }, + { + "left": "pull_request:3546", + "right": "pull_request:3632", + "accept": false, + "reason": "Panorama scheduler compatibility vs view-batch feature/docs; same subsystem, different functionality." + }, + { + "left": "pull_request:11739", + "right": "pull_request:11744", + "accept": false, + "reason": "Both touch LoRA metadata/alpha, but on different training scripts and example flows." + }, + { + "left": "pull_request:12490", + "right": "pull_request:12782", + "accept": false, + "reason": "Different test-only fixes for different models." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13039", + "accept": false, + "reason": "Audioldm2 test config vs WAN CPU upcasting test behavior; unrelated." + }, + { + "left": "pull_request:13039", + "right": "pull_request:13055", + "accept": false, + "reason": "One fixes layerwise upcasting on CPU; the other fixes WAN pipeline fallback when a transformer is missing." + }, + { + "left": "pull_request:1508", + "right": "pull_request:2339", + "accept": false, + "reason": "Dreambooth docs clarification vs checkpointing docs/tests; related topic, not same issue." + }, + { + "left": "pull_request:11997", + "right": "pull_request:12611", + "accept": false, + "reason": "Both are Wan-related, but one is a modular pipeline feature PR and the other is a broad guider/modular update; not the same concrete change." + }, + { + "left": "pull_request:10539", + "right": "pull_request:10810", + "accept": false, + "reason": "True-cfg fix in Flux vs LoRA support conversion changes; unrelated." + }, + { + "left": "pull_request:1136", + "right": "pull_request:2001", + "accept": false, + "reason": "404 link fix in examples README vs ColossalAI typo/seed fix; unrelated." + }, + { + "left": "pull_request:1136", + "right": "pull_request:2693", + "accept": false, + "reason": "Different documentation pages and different broken links." + }, + { + "left": "pull_request:11704", + "right": "pull_request:11856", + "accept": false, + "reason": "Both touch Wan LoRA conversion, but they fix different conversion paths and failure modes." + }, + { + "left": "pull_request:5407", + "right": "pull_request:6346", + "accept": false, + "reason": "Mixed-precision training example fix vs LoRA state-dict conversion follow-up; not the same bug." + }, + { + "left": "pull_request:7068", + "right": "pull_request:8571", + "accept": false, + "reason": "Generic typo/style cleanup PRs; too broad and unrelated." + }, + { + "left": "pull_request:1982", + "right": "pull_request:774", + "accept": false, + "reason": "Example test path tweak vs pipeline callback/test tolerance changes; unrelated." + }, + { + "left": "pull_request:10983", + "right": "pull_request:11066", + "accept": false, + "reason": "Both are HunyuanVideo I2V changes, but they target different model variants/releases and different code paths." + }, + { + "left": "pull_request:12006", + "right": "pull_request:12523", + "accept": false, + "reason": "Wan model addition vs AutoencoderKLWan skip-key fix; different problems." + }, + { + "left": "pull_request:1525", + "right": "pull_request:3607", + "accept": false, + "reason": "Upscaling batch-size fix vs temb attention fix; unrelated." + }, + { + "left": "pull_request:6994", + "right": "pull_request:8631", + "accept": false, + "reason": "Two separate from_single_file loader fixes touching different conversion utilities." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 268, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10105", + "pull_request:10200", + "pull_request:10579", + "pull_request:10704", + "pull_request:1128", + "pull_request:11366", + "pull_request:11384", + "pull_request:11571", + "pull_request:11572", + "pull_request:11635", + "pull_request:11726", + "pull_request:11897", + "pull_request:12315", + "pull_request:12324", + "pull_request:12496", + "pull_request:12522", + "pull_request:12523", + "pull_request:12982", + "pull_request:13002", + "pull_request:13150", + "pull_request:13208", + "pull_request:13223", + "pull_request:1338", + "pull_request:13454", + "pull_request:1926", + "pull_request:2322", + "pull_request:2691", + "pull_request:2693", + "pull_request:2968", + "pull_request:3298", + "pull_request:391", + "pull_request:416", + "pull_request:4242", + "pull_request:4294", + "pull_request:433", + "pull_request:4412", + "pull_request:5098", + "pull_request:6489", + "pull_request:6490", + "pull_request:689", + "pull_request:6962", + "pull_request:7247", + "pull_request:7941", + "pull_request:8322", + "pull_request:8761", + "pull_request:9008", + "pull_request:9065" + ], + "soft_pairs": [ + "pull_request:13208|pull_request:13223", + "pull_request:11635|pull_request:11897", + "pull_request:2322|pull_request:433", + "pull_request:2691|pull_request:4294", + "pull_request:391|pull_request:416", + "pull_request:12982|pull_request:13002", + "pull_request:11571|pull_request:11572", + "pull_request:12496|pull_request:12523", + "pull_request:10105|pull_request:10704", + "pull_request:1338|pull_request:689", + "pull_request:1128|pull_request:1926", + "pull_request:11366|pull_request:11384", + "pull_request:10579|pull_request:9065", + "pull_request:2693|pull_request:4412", + "pull_request:2968|pull_request:4412", + "pull_request:5098|pull_request:6962", + "pull_request:12522|pull_request:13454", + "pull_request:11726|pull_request:12324", + "pull_request:12315|pull_request:13150", + "pull_request:3298|pull_request:4242", + "pull_request:6489|pull_request:6490", + "pull_request:8322|pull_request:9008", + "pull_request:7247|pull_request:7941", + "pull_request:10200|pull_request:8761" + ], + "prepared_review_unit_hash": "8865b013118fcaab7d0b9f3ad3f0fef2e0121c5173dfe86abb2c14aaaa7b1719", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 24, + "serialized_chars": 105037, + "estimated_input_tokens": 26260, + "estimated_eval_tokens": 52776 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous cluster of unrelated pull requests across different features, docs, tests, and refactors. The soft edges are mostly superficial filename/subsystem matches rather than the same underlying change.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a safe canonical representative: the cluster spans unrelated PRs from different subsystems and release trains, with no shared concrete bug/change theme.", + "best_issue_reason": null, + "best_pr_reason": "No PR is globally best-suited for duplicate collapsing here; the items do not form one coherent fix or feature.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13208", + "right": "pull_request:13223", + "accept": false, + "reason": "Same Helios area, but one adds the pipelines and docs while the other changes context-parallel behavior in the transformer; different changes, not one duplicate fix." + }, + { + "left": "pull_request:11635", + "right": "pull_request:11897", + "accept": false, + "reason": "Both touch docs/pipelines, but 11635 adds Flux Kontext while 11897 is a broad modular-diffusers/test merge; unrelated work." + }, + { + "left": "pull_request:2322", + "right": "pull_request:433", + "accept": false, + "reason": "Different eras and completely different concerns: inpaint example/update vs black/versioning cleanup." + }, + { + "left": "pull_request:2691", + "right": "pull_request:4294", + "accept": false, + "reason": "One renames attention code; the other updates SDXL checkpoint/docs naming. No shared code-path bug." + }, + { + "left": "pull_request:391", + "right": "pull_request:416", + "accept": false, + "reason": "PNDM docs versus core model docs restructuring; unrelated documentation changes." + }, + { + "left": "pull_request:12982", + "right": "pull_request:13002", + "accept": false, + "reason": "Both are about FLUX.2 klein, but one is training/docs/pipeline support and the other is modular-pipeline implementation; not the same concrete change." + }, + { + "left": "pull_request:11571", + "right": "pull_request:11572", + "accept": false, + "reason": "Same LTX-Video theme, but one adds distilled model/docs and the other adds LoRA conversion support; different code paths." + }, + { + "left": "pull_request:10105", + "right": "pull_request:10704", + "accept": false, + "reason": "Broken README link fix versus img2img docs cleanup; unrelated doc edits." + }, + { + "left": "pull_request:1338", + "right": "pull_request:689", + "accept": false, + "reason": "ONNX inpaint cast-order fix is unrelated to an attention performance tweak." + }, + { + "left": "pull_request:1128", + "right": "pull_request:1926", + "accept": false, + "reason": "Both mention inpainting, but one adds tensor input support and the other is OneFlow dev support; different implementations." + }, + { + "left": "pull_request:11366", + "right": "pull_request:11384", + "accept": false, + "reason": "Same Hidream file, but one changes model outputs and the other only shifts deprecation versions; not the same bug/change." + }, + { + "left": "pull_request:10579", + "right": "pull_request:9065", + "accept": false, + "reason": "Community pipeline addition versus sentencepiece optional dependency plumbing; unrelated." + }, + { + "left": "pull_request:2693", + "right": "pull_request:4412", + "accept": false, + "reason": "Doc link fix versus test fixture cleanup; no shared underlying issue." + }, + { + "left": "pull_request:2968", + "right": "pull_request:4412", + "accept": false, + "reason": "Post-processing bug fix versus a test setup change; unrelated." + }, + { + "left": "pull_request:5098", + "right": "pull_request:6962", + "accept": false, + "reason": "Different pipelines and different goals: Kandinsky precision tests versus PixArt slow-test stabilization." + }, + { + "left": "pull_request:12522", + "right": "pull_request:13454", + "accept": false, + "reason": "Toctree styling cleanup versus xfail markers for CLIP/transformers compatibility; unrelated." + }, + { + "left": "pull_request:11726", + "right": "pull_request:12324", + "accept": false, + "reason": "Both involve Wan, but one extends LoRA conversion support and the other adds a new VACE-Fun model/config path; distinct changes." + }, + { + "left": "pull_request:12315", + "right": "pull_request:13150", + "accept": false, + "reason": "One normalizes scheduler usage in LoRA tests; the other changes a test helper API. Not the same bug or feature." + }, + { + "left": "pull_request:3298", + "right": "pull_request:4242", + "accept": false, + "reason": "Image aspect-ratio support in IF super-resolution is unrelated to allowing VAE loading in from_single_file." + }, + { + "left": "pull_request:6489", + "right": "pull_request:6490", + "accept": false, + "reason": "Two different community pipelines; no shared code-path or duplicate fix." + }, + { + "left": "pull_request:8322", + "right": "pull_request:9008", + "accept": false, + "reason": "Generic errata/typo cleanup versus broader typo and formatting fixes; separate maintenance work." + }, + { + "left": "pull_request:7247", + "right": "pull_request:7941", + "accept": false, + "reason": "Config-checking test fix versus adding a slow-test decorator; unrelated test maintenance." + }, + { + "left": "pull_request:10200", + "right": "pull_request:8761", + "accept": false, + "reason": "ControlNet Union refactor and Hunyuan embedding refactor are separate model changes with no common bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 269, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10330", + "pull_request:10431", + "pull_request:10537", + "pull_request:10815", + "pull_request:11252", + "pull_request:11556", + "pull_request:11916", + "pull_request:11986", + "pull_request:12006", + "pull_request:12092", + "pull_request:12387", + "pull_request:12496", + "pull_request:12530", + "pull_request:12549", + "pull_request:12647", + "pull_request:12763", + "pull_request:12961", + "pull_request:13039", + "pull_request:13331", + "pull_request:13454", + "pull_request:1926", + "pull_request:1966", + "pull_request:1982", + "pull_request:2367", + "pull_request:2464", + "pull_request:2470", + "pull_request:2593", + "pull_request:2892", + "pull_request:2959", + "pull_request:327", + "pull_request:344", + "pull_request:3586", + "pull_request:3657", + "pull_request:382", + "pull_request:3912", + "pull_request:396", + "pull_request:4098", + "pull_request:416", + "pull_request:4408", + "pull_request:4412", + "pull_request:4780", + "pull_request:5151", + "pull_request:5384", + "pull_request:6633", + "pull_request:6961", + "pull_request:802", + "pull_request:8452", + "pull_request:9696" + ], + "soft_pairs": [ + "pull_request:327|pull_request:344", + "pull_request:3586|pull_request:5384", + "pull_request:2470|pull_request:2959", + "pull_request:382|pull_request:396", + "pull_request:382|pull_request:416", + "pull_request:1926|pull_request:1966", + "pull_request:6961|pull_request:8452", + "pull_request:4098|pull_request:4408", + "pull_request:11916|pull_request:12549", + "pull_request:12530|pull_request:13454", + "pull_request:2892|pull_request:4412", + "pull_request:4780|pull_request:5151", + "pull_request:2464|pull_request:3912", + "pull_request:12006|pull_request:12496", + "pull_request:10330|pull_request:10815", + "pull_request:10431|pull_request:10537", + "pull_request:11556|pull_request:12092", + "pull_request:11986|pull_request:12763", + "pull_request:12647|pull_request:13039", + "pull_request:12961|pull_request:13331", + "pull_request:1982|pull_request:802", + "pull_request:2367|pull_request:3657", + "pull_request:2593|pull_request:3657", + "pull_request:6633|pull_request:6961", + "pull_request:11252|pull_request:9696", + "pull_request:11916|pull_request:12387" + ], + "prepared_review_unit_hash": "3d5b69faf66793c7a1dfe4d5f9580a25f8266af242ef97202d33163aec38588f", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 102040, + "estimated_input_tokens": 25510, + "estimated_eval_tokens": 51276 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous: many unrelated docs, test, and feature PRs. The only strong duplicate-like pair is the PEFT integration work (4780 and 5151).", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 5151 is the best representative because it is the clearest mature continuation of the only truly overlapping feature thread in the set (PEFT integration with 4780).", + "soft_edge_verdicts": [ + { + "left": "pull_request:327", + "right": "pull_request:344", + "accept": false, + "reason": "Different type-hint cleanup PRs for unrelated codepaths (PNDM vs VAE)." + }, + { + "left": "pull_request:3586", + "right": "pull_request:5384", + "accept": false, + "reason": "Both are broad feature PRs, but they target different subsystems and problems (tiling pipeline vs hypernetwork support)." + }, + { + "left": "pull_request:2470", + "right": "pull_request:2959", + "accept": false, + "reason": "Both are test fixes, but for different models and failure modes." + }, + { + "left": "pull_request:382", + "right": "pull_request:396", + "accept": false, + "reason": "Docs for different pipelines (DDPM vs DDIM), not the same change." + }, + { + "left": "pull_request:382", + "right": "pull_request:416", + "accept": false, + "reason": "DDPM/DDIM docs versus general model docs; unrelated documentation updates." + }, + { + "left": "pull_request:1926", + "right": "pull_request:1966", + "accept": false, + "reason": "Both involve inpainting, but they are separate framework/backend implementations and not one concrete fix." + }, + { + "left": "pull_request:6961", + "right": "pull_request:8452", + "accept": false, + "reason": "Both touch single-file loading, but one adds tests and the other changes loader behavior for a different bug." + }, + { + "left": "pull_request:4098", + "right": "pull_request:4408", + "accept": false, + "reason": "One adds SDXL inpainting; the other adds from_single_file support for that feature. Related area, but distinct PRs." + }, + { + "left": "pull_request:11916", + "right": "pull_request:12549", + "accept": false, + "reason": "Attention-dispatch refactor versus adding a new AITER backend; same subsystem, different concrete changes." + }, + { + "left": "pull_request:12530", + "right": "pull_request:13454", + "accept": false, + "reason": "Separate xfail annotations for unrelated failing tests." + }, + { + "left": "pull_request:2892", + "right": "pull_request:4412", + "accept": false, + "reason": "Unrelated changes: upscaler input validation versus a float16 test fix." + }, + { + "left": "pull_request:4780", + "right": "pull_request:5151", + "accept": true, + "reason": "Both are PEFT integration work for UNet; 5151 looks like the completed follow-through of the same feature thread." + }, + { + "left": "pull_request:2464", + "right": "pull_request:3912", + "accept": false, + "reason": "Same training script file, but different changes: xformers cleanup versus adding text-encoder finetuning support." + }, + { + "left": "pull_request:12006", + "right": "pull_request:12496", + "accept": false, + "reason": "Different Wan i2v changes: feature expansion for 5B support versus a shape-mismatch bug fix." + }, + { + "left": "pull_request:10330", + "right": "pull_request:10815", + "accept": false, + "reason": "LTX Video release/docs work versus Enhance-A-Video hook implementation; not the same bug or change." + }, + { + "left": "pull_request:10431", + "right": "pull_request:10537", + "accept": false, + "reason": "Both are Sana-related, but one fixes positional embeddings and the other adds 4K pipeline adjustments." + }, + { + "left": "pull_request:11556", + "right": "pull_request:12092", + "accept": false, + "reason": "Different docs topics: regional compilation versus GGUF checkpoint conversion." + }, + { + "left": "pull_request:11986", + "right": "pull_request:12763", + "accept": false, + "reason": "Different modular pipelines for different model families (LTX vs Flux2)." + }, + { + "left": "pull_request:12647", + "right": "pull_request:13039", + "accept": false, + "reason": "Both concern layerwise-casting tests, but they are model-specific and fix different failures." + }, + { + "left": "pull_request:12961", + "right": "pull_request:13331", + "accept": false, + "reason": "Different pipeline additions altogether (PhotoDoodle vs CogVideoX inpaint)." + }, + { + "left": "pull_request:1982", + "right": "pull_request:802", + "accept": false, + "reason": "Example test path fix versus an indentation fix in a custom pipeline example." + }, + { + "left": "pull_request:2367", + "right": "pull_request:3657", + "accept": false, + "reason": "Both touch training/conversion scripts, but they address unrelated features and scripts." + }, + { + "left": "pull_request:2593", + "right": "pull_request:3657", + "accept": false, + "reason": "Same ControlNet converter script, but different supported checkpoint behaviors; not the same concrete change." + }, + { + "left": "pull_request:6633", + "right": "pull_request:6961", + "accept": false, + "reason": "Different single-file loader fixes: YAML config reading versus test scaffolding changes." + }, + { + "left": "pull_request:11252", + "right": "pull_request:9696", + "accept": false, + "reason": "Unrelated utility/workflow updates: timeout constant cleanup versus PR test workflow input changes." + }, + { + "left": "pull_request:11916", + "right": "pull_request:12387", + "accept": false, + "reason": "Attention dispatcher refactor versus kernels-backed flash-attention backend support; distinct features." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 270, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10330", + "pull_request:10537", + "pull_request:10538", + "pull_request:10547", + "pull_request:11174", + "pull_request:11995", + "pull_request:12454", + "pull_request:12593", + "pull_request:12644", + "pull_request:12679", + "pull_request:12934", + "pull_request:12983", + "pull_request:13158", + "pull_request:13208", + "pull_request:13216", + "pull_request:13223", + "pull_request:1788", + "pull_request:1913", + "pull_request:2327", + "pull_request:2367", + "pull_request:2593", + "pull_request:2706", + "pull_request:2919", + "pull_request:2959", + "pull_request:3292", + "pull_request:3607", + "pull_request:3806", + "pull_request:382", + "pull_request:386", + "pull_request:390", + "pull_request:391", + "pull_request:392", + "pull_request:404", + "pull_request:417", + "pull_request:4187", + "pull_request:5506", + "pull_request:5665", + "pull_request:6426", + "pull_request:6759", + "pull_request:6994", + "pull_request:7116", + "pull_request:7200", + "pull_request:7246", + "pull_request:7825", + "pull_request:7941", + "pull_request:8937", + "pull_request:9251", + "pull_request:9676" + ], + "soft_pairs": [ + "pull_request:6426|pull_request:7116", + "pull_request:1788|pull_request:3607", + "pull_request:2367|pull_request:4187", + "pull_request:6759|pull_request:8937", + "pull_request:12593|pull_request:12679", + "pull_request:7825|pull_request:9676", + "pull_request:1913|pull_request:3292", + "pull_request:2327|pull_request:2706", + "pull_request:2593|pull_request:3806", + "pull_request:5665|pull_request:6994", + "pull_request:2919|pull_request:2959", + "pull_request:7246|pull_request:7941", + "pull_request:11995|pull_request:12454", + "pull_request:12934|pull_request:12983", + "pull_request:13208|pull_request:13216", + "pull_request:10330|pull_request:11174", + "pull_request:12644|pull_request:13158", + "pull_request:382|pull_request:391", + "pull_request:386|pull_request:392", + "pull_request:386|pull_request:404", + "pull_request:386|pull_request:417", + "pull_request:390|pull_request:392", + "pull_request:390|pull_request:404", + "pull_request:390|pull_request:417", + "pull_request:5506|pull_request:7200", + "pull_request:13216|pull_request:13223", + "pull_request:10537|pull_request:10538", + "pull_request:10547|pull_request:9251" + ], + "prepared_review_unit_hash": "402ba18f5bc06a7cbb0b725be80c6ea23f56a5e559c3acac6bb813e862a0ad5b", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 103629, + "estimated_input_tokens": 25908, + "estimated_eval_tokens": 52072 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a set of mostly unrelated PRs that only look similar at a very broad level (docs, tests, model/pipeline additions). None of the soft-pair candidates appear to fix the same concrete bug or implement the same change, so they should not be merged as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR stands out as a duplicate target; the items are not a coherent duplicate cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 13216 is the strongest single representative if one must be chosen: it is a broad, self-contained Helios modular pipeline feature with supporting code and tests.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6426", + "right": "pull_request:7116", + "accept": false, + "reason": "LoRA deprecation cleanup vs workflow install changes; unrelated changes and not the same code-path problem." + }, + { + "left": "pull_request:1788", + "right": "pull_request:3607", + "accept": false, + "reason": "DDIM scheduler bug fix vs attention/PEFT/xformers fix; different components and failure modes." + }, + { + "left": "pull_request:2367", + "right": "pull_request:4187", + "accept": false, + "reason": "Training checkpoint retention option vs DPM-Solver timestep handling; unrelated functionality." + }, + { + "left": "pull_request:6759", + "right": "pull_request:8937", + "accept": false, + "reason": "InstantID example fix vs long-prompt handling in AuraFlow; different pipelines and issues." + }, + { + "left": "pull_request:12593", + "right": "pull_request:12679", + "accept": false, + "reason": "ChronoEdit feature addition vs ChronoEdit LoRA support; related project, but not the same concrete change." + }, + { + "left": "pull_request:7825", + "right": "pull_request:9676", + "accept": false, + "reason": "Model offload test guard vs benchmark import cleanup; no shared bug or implementation target." + }, + { + "left": "pull_request:1913", + "right": "pull_request:3292", + "accept": false, + "reason": "Input type validation for img2img pipelines vs training input perturbation; separate concerns." + }, + { + "left": "pull_request:2327", + "right": "pull_request:2706", + "accept": false, + "reason": "UNet config length validation vs EMA save/restore in training; different bugs and files." + }, + { + "left": "pull_request:2593", + "right": "pull_request:3806", + "accept": false, + "reason": "ControlNet checkpoint conversion support vs README inference note; docs/support update, not same fix." + }, + { + "left": "pull_request:5665", + "right": "pull_request:6994", + "accept": false, + "reason": "ControlNet img2img preprocessing fix vs single-file dtype propagation; distinct code paths." + }, + { + "left": "pull_request:2919", + "right": "pull_request:2959", + "accept": false, + "reason": "Two different test adjustments for UNet3D; both are test-only and not the same underlying change." + }, + { + "left": "pull_request:7246", + "right": "pull_request:7941", + "accept": false, + "reason": "VAE tiling device fix vs slow-test decorator; unrelated test maintenance." + }, + { + "left": "pull_request:11995", + "right": "pull_request:12454", + "accept": false, + "reason": "Flux modular pipeline addition vs Kontext modular support; same family, but distinct features and pipelines." + }, + { + "left": "pull_request:12934", + "right": "pull_request:12983", + "accept": false, + "reason": "LTX2 distilled checkpoint feature support vs single-file loader support; related release but not the same concrete change." + }, + { + "left": "pull_request:13208", + "right": "pull_request:13216", + "accept": false, + "reason": "Helios docs/pipeline registration vs Helios modular pipeline implementation; different product surfaces." + }, + { + "left": "pull_request:13216", + "right": "pull_request:13223", + "accept": false, + "reason": "Helios modular pipeline addition vs context-parallelism fix in Helios transformer; separate implementation and bug fix." + }, + { + "left": "pull_request:10330", + "right": "pull_request:11174", + "accept": false, + "reason": "LTX Video 0.9.1 update vs LTX0.9.5 conditioning refactor; different versions and different changes." + }, + { + "left": "pull_request:12644", + "right": "pull_request:13158", + "accept": false, + "reason": "AutoModel docs vs QwenImageLayered docs; unrelated documentation changes." + }, + { + "left": "pull_request:382", + "right": "pull_request:391", + "accept": false, + "reason": "DDPM docs vs PNDM docs; separate pipeline documentation pages." + }, + { + "left": "pull_request:386", + "right": "pull_request:392", + "accept": false, + "reason": "Stable Diffusion docs vs conceptual contribution docs; unrelated documentation scope." + }, + { + "left": "pull_request:386", + "right": "pull_request:404", + "accept": false, + "reason": "Stable Diffusion docs vs fp16/memory-speed docs; both docs, but different topics and targets." + }, + { + "left": "pull_request:390", + "right": "pull_request:392", + "accept": false, + "reason": "Apple Silicon MPS guide vs contribution docs; unrelated documentation areas." + }, + { + "left": "pull_request:390", + "right": "pull_request:404", + "accept": false, + "reason": "MPS optimization guide vs fp16/memory-speed guide; different optimization topics." + }, + { + "left": "pull_request:390", + "right": "pull_request:417", + "accept": false, + "reason": "MPS optimization guide vs inference pipeline docs; not the same change." + }, + { + "left": "pull_request:5506", + "right": "pull_request:7200", + "accept": false, + "reason": "PEFT/xformers LoRA fix vs transformer norm-type validation; different attention/model issues." + }, + { + "left": "pull_request:10537", + "right": "pull_request:10538", + "accept": false, + "reason": "Sana PAG pipeline behavior tweak vs auto-pipeline registration; separate feature plumbing." + }, + { + "left": "pull_request:10547", + "right": "pull_request:9251", + "accept": false, + "reason": "Offload test data for CogVideoX/CogView3 vs CI trigger branch change; unrelated test vs workflow update." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 271, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10021", + "pull_request:10063", + "pull_request:10105", + "pull_request:10136", + "pull_request:10245", + "pull_request:10326", + "pull_request:10815", + "pull_request:10983", + "pull_request:11613", + "pull_request:11614", + "pull_request:12387", + "pull_request:12490", + "pull_request:12549", + "pull_request:13338", + "pull_request:13365", + "pull_request:13454", + "pull_request:1878", + "pull_request:2593", + "pull_request:3160", + "pull_request:326", + "pull_request:3351", + "pull_request:344", + "pull_request:3471", + "pull_request:3806", + "pull_request:4242", + "pull_request:4262", + "pull_request:4395", + "pull_request:4572", + "pull_request:4963", + "pull_request:5053", + "pull_request:5393", + "pull_request:5615", + "pull_request:5665", + "pull_request:5788", + "pull_request:5837", + "pull_request:6426", + "pull_request:6743", + "pull_request:6764", + "pull_request:6886", + "pull_request:7113", + "pull_request:7200", + "pull_request:7627", + "pull_request:7938", + "pull_request:8512", + "pull_request:8652", + "pull_request:8684", + "pull_request:8797" + ], + "soft_pairs": [ + "pull_request:1878|pull_request:2593", + "pull_request:4242|pull_request:5837", + "pull_request:4262|pull_request:5837", + "pull_request:4963|pull_request:5615", + "pull_request:6764|pull_request:8652", + "pull_request:6886|pull_request:8512", + "pull_request:12387|pull_request:12549", + "pull_request:10021|pull_request:10815", + "pull_request:13338|pull_request:13365", + "pull_request:326|pull_request:344", + "pull_request:3471|pull_request:3806", + "pull_request:3806|pull_request:4395", + "pull_request:4572|pull_request:5393", + "pull_request:5053|pull_request:6743", + "pull_request:5665|pull_request:7627", + "pull_request:5788|pull_request:7200", + "pull_request:6426|pull_request:8512", + "pull_request:10063|pull_request:10105", + "pull_request:10245|pull_request:8684", + "pull_request:3160|pull_request:3351", + "pull_request:10326|pull_request:8797", + "pull_request:7113|pull_request:7938", + "pull_request:11613|pull_request:11614", + "pull_request:12490|pull_request:13454", + "pull_request:10136|pull_request:10983" + ], + "prepared_review_unit_hash": "c6810c295ae11c0bf0dcda8c47966b19d67584d5a6683651493994d9bbc92d96", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 25, + "serialized_chars": 98849, + "estimated_input_tokens": 24713, + "estimated_eval_tokens": 49682 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly false-positive similarity links across unrelated PRs; no duplicate pair is strong enough to accept under strict PR duplicate rules.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents this cluster: the linked PRs span unrelated docs, CI, scheduler, model, and example changes rather than one shared bug or feature.", + "best_issue_reason": null, + "best_pr_reason": "No global best PR is suitable because the cluster is heterogeneous. If forced, #10136 is the broadest substantive feature PR, but it still does not match the rest of the items.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1878", + "right": "pull_request:2593", + "accept": false, + "reason": "Different changes: checkpoint loading guard in a conversion script vs a separate ControlNet conversion script. Same general area only." + }, + { + "left": "pull_request:4242", + "right": "pull_request:5837", + "accept": false, + "reason": "Unrelated loader docs/code: allowing VAE loading vs allowing multiple textual inversion tokens." + }, + { + "left": "pull_request:4262", + "right": "pull_request:5837", + "accept": false, + "reason": "Different example fixes in different training paths; not the same underlying bug." + }, + { + "left": "pull_request:4963", + "right": "pull_request:5615", + "accept": false, + "reason": "One PR installs accelerate main; the other reverts LoRA loading in an SDXL adapter pipeline. No shared bug." + }, + { + "left": "pull_request:6764", + "right": "pull_request:8652", + "accept": false, + "reason": "A serialization note for PEFT adapters is unrelated to adding the Lumina-T2X model/pipeline." + }, + { + "left": "pull_request:6886", + "right": "pull_request:8512", + "accept": false, + "reason": "Deprecation cleanup for `to(dtype=...)` is unrelated to removing a deprecated mask helper." + }, + { + "left": "pull_request:12387", + "right": "pull_request:12549", + "accept": false, + "reason": "Both touch attention backends, but they add different backends (`kernels` flash vs AITER). Not the same concrete change." + }, + { + "left": "pull_request:10021", + "right": "pull_request:10815", + "accept": false, + "reason": "LTX Video core support and Enhance-A-Video hooks are different feature areas, not one bug fix." + }, + { + "left": "pull_request:13338", + "right": "pull_request:13365", + "accept": false, + "reason": "Same workflow file, but one adds `id-token` permission while the other changes fork-review behavior. Different changes." + }, + { + "left": "pull_request:326", + "right": "pull_request:344", + "accept": false, + "reason": "Both are type-hint cleanup PRs, but they target different modules and problems." + }, + { + "left": "pull_request:3471", + "right": "pull_request:3806", + "accept": false, + "reason": "Different docs examples: one adds a torch.compile note, the other explains loading from a checkpoint." + }, + { + "left": "pull_request:3806", + "right": "pull_request:4395", + "accept": false, + "reason": "Checkpoint-loading guidance is unrelated to the DreamBooth class-prompt pre-encoding fix." + }, + { + "left": "pull_request:4572", + "right": "pull_request:5393", + "accept": false, + "reason": "`network_alpha` naming fix in an example is unrelated to the HFApi docs update." + }, + { + "left": "pull_request:5053", + "right": "pull_request:6743", + "accept": false, + "reason": "Different training script fixes: validation image argument vs hflip argument naming." + }, + { + "left": "pull_request:5665", + "right": "pull_request:7627", + "accept": false, + "reason": "Both are img2img-related, but one fixes control image preprocessing dimensions and the other handles latents stats for Playground." + }, + { + "left": "pull_request:5788", + "right": "pull_request:7200", + "accept": false, + "reason": "LCM scheduler error handling and Transformer2D layernorm/patch-input validation are unrelated changes." + }, + { + "left": "pull_request:6426", + "right": "pull_request:8512", + "accept": false, + "reason": "LoRA deprecation handling is unrelated to removing the deprecated mask helper function." + }, + { + "left": "pull_request:10063", + "right": "pull_request:10105", + "accept": false, + "reason": "Both are docs nits, but they fix different pages and different broken text/links." + }, + { + "left": "pull_request:10245", + "right": "pull_request:8684", + "accept": false, + "reason": "PEFT adapter docs cleanup is unrelated to the Pillow import fix." + }, + { + "left": "pull_request:3160", + "right": "pull_request:3351", + "accept": false, + "reason": "Same feature family (`use_karras_sigmas`) but different scheduler implementations; not one concrete bug/change." + }, + { + "left": "pull_request:10326", + "right": "pull_request:8797", + "accept": false, + "reason": "Workflow quoting fix and sharded-checkpoint test fixes are unrelated." + }, + { + "left": "pull_request:7113", + "right": "pull_request:7938", + "accept": false, + "reason": "Both touch Docker CI notifications, but one adds Slack reporting and the other swaps the notification implementation. Not the same change." + }, + { + "left": "pull_request:11613", + "right": "pull_request:11614", + "accept": false, + "reason": "A revert of torch.compile CI changes is unrelated to a Docker CI security fix." + }, + { + "left": "pull_request:12490", + "right": "pull_request:13454", + "accept": false, + "reason": "A CLAP config test fix is unrelated to xfail markers for CLIP/transformers compatibility issues." + }, + { + "left": "pull_request:10136", + "right": "pull_request:10983", + "accept": false, + "reason": "Related Hunyuan family features, but one adds core Hunyuan Video while the other adds the I2V pipeline/model extension; separate feature work." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 266, + "review_unit_index": 272, + "review_unit_count": 272, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1-3019", + "nodes": [ + "pull_request:10148", + "pull_request:10431", + "pull_request:11001", + "pull_request:11058", + "pull_request:11142", + "pull_request:11231", + "pull_request:11269", + "pull_request:11660", + "pull_request:12450", + "pull_request:12491", + "pull_request:1878", + "pull_request:2892", + "pull_request:2915", + "pull_request:3298", + "pull_request:3586", + "pull_request:3589", + "pull_request:3806", + "pull_request:4412", + "pull_request:4749", + "pull_request:5672", + "pull_request:6450", + "pull_request:6752", + "pull_request:6886", + "pull_request:7116", + "pull_request:7359", + "pull_request:8386", + "pull_request:8719", + "pull_request:9049", + "pull_request:9159" + ], + "soft_pairs": [ + "pull_request:11001|pull_request:11058", + "pull_request:2915|pull_request:4412", + "pull_request:11231|pull_request:11660", + "pull_request:9049|pull_request:9159", + "pull_request:6752|pull_request:7359", + "pull_request:10148|pull_request:11269", + "pull_request:10431|pull_request:11142", + "pull_request:12450|pull_request:12491", + "pull_request:1878|pull_request:3806", + "pull_request:2892|pull_request:3298", + "pull_request:3586|pull_request:3589", + "pull_request:3586|pull_request:5672", + "pull_request:4749|pull_request:6450", + "pull_request:6886|pull_request:8719", + "pull_request:7116|pull_request:8386" + ], + "prepared_review_unit_hash": "4fa0c962b7c274ffa1ae75b0c0ac23b28ba3a05c808a954ddd68e2a28116ebb2", + "component_budget": { + "node_count": 3019, + "item_count": 3019, + "soft_pair_count": 7653, + "serialized_chars": 8728126, + "estimated_input_tokens": 2182032, + "estimated_eval_tokens": 4364320 + }, + "budget": { + "node_count": 29, + "item_count": 29, + "soft_pair_count": 15, + "serialized_chars": 62349, + "estimated_input_tokens": 15588, + "estimated_eval_tokens": 31432 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a true duplicate set. The PRs span unrelated features, CI/doc changes, and isolated test fixes across different subsystems; the soft edges are false positives from generic similarity (tests, workflows, model family names, or docs).", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a valid canonical here: the items are not the same underlying bug or change, so picking one would misrepresent the cluster.", + "best_issue_reason": null, + "best_pr_reason": "No best PR candidate exists for deduplication because there is no coherent duplicate theme across the set; the cluster is heterogeneous.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11001", + "right": "pull_request:11058", + "accept": false, + "reason": "Both touch test/device handling, but one enables bnb tests on xpu while the other makes CUDA-only tests device-agnostic; different fixes and code paths." + }, + { + "left": "pull_request:2915", + "right": "pull_request:4412", + "accept": false, + "reason": "Both are test-only adjustments, but they fix different failures in different tests; not the same bug or change." + }, + { + "left": "pull_request:11231", + "right": "pull_request:11660", + "accept": false, + "reason": "Same model family, but one adds the HiDream Image pipeline and the other adds HiDream LoRA conversion support; separate features." + }, + { + "left": "pull_request:9049", + "right": "pull_request:9159", + "accept": false, + "reason": "Flux encode_prompt dtype handling and Flux transformer sharding are unrelated code-path changes." + }, + { + "left": "pull_request:6752", + "right": "pull_request:7359", + "accept": false, + "reason": "Different research-project additions: PromptDiffusion pipeline vs ResAdapter training script." + }, + { + "left": "pull_request:10148", + "right": "pull_request:11269", + "accept": false, + "reason": "OmniGen integration and onnxruntime package-detection changes are unrelated." + }, + { + "left": "pull_request:10431", + "right": "pull_request:11142", + "accept": false, + "reason": "Both are Sana-related, but one fixes positional embedding behavior in the transformer and the other adjusts a converter script for Sana Sprint checkpoints; not the same issue." + }, + { + "left": "pull_request:12450", + "right": "pull_request:12491", + "accept": false, + "reason": "Both are CI workflow edits, but they address different pipeline/install concerns; not mergeable as one bug fix." + }, + { + "left": "pull_request:1878", + "right": "pull_request:3806", + "accept": false, + "reason": "Stable diffusion checkpoint parsing fix vs documentation note about loading from checkpoints; different artifacts and purposes." + }, + { + "left": "pull_request:2892", + "right": "pull_request:3298", + "accept": false, + "reason": "One fixes upscaler input validation for embeds, the other adds arbitrary aspect ratio support to a different pipeline." + }, + { + "left": "pull_request:3586", + "right": "pull_request:3589", + "accept": false, + "reason": "A new community pipeline and README reference-trick updates are not the same underlying change." + }, + { + "left": "pull_request:3586", + "right": "pull_request:5672", + "accept": false, + "reason": "Completely different features: a community Stable Diffusion tiling pipeline vs PixArt non-square image support." + }, + { + "left": "pull_request:4749", + "right": "pull_request:6450", + "accept": false, + "reason": "DreamBooth SDXL LoRA checkpointing and LoRA deprecation test cleanup are unrelated changes." + }, + { + "left": "pull_request:6886", + "right": "pull_request:8719", + "accept": false, + "reason": "Both are maintenance-oriented, but one removes deprecated torch_dtype usage across converters/tests while the other updates VQ model deprecation docs; different changes." + }, + { + "left": "pull_request:7116", + "right": "pull_request:8386", + "accept": false, + "reason": "CI installation workflow changes and test input-shape refactors are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 2, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9562-4", + "nodes": [ + "pull_request:10009", + "pull_request:10163", + "pull_request:10347", + "pull_request:9562" + ], + "soft_pairs": [ + "pull_request:10009|pull_request:9562", + "pull_request:10163|pull_request:9562", + "pull_request:10347|pull_request:9562", + "pull_request:10009|pull_request:10163", + "pull_request:10009|pull_request:10347", + "pull_request:10163|pull_request:10347" + ], + "prepared_review_unit_hash": "3e73d48468252583a1dfe71f158bbfd33123f7aa4e76bea94aa942134590b530", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 15324, + "estimated_input_tokens": 3831, + "estimated_eval_tokens": 7918 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 15324, + "estimated_input_tokens": 3831, + "estimated_eval_tokens": 7918 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are thematically related by optimization/hook infrastructure, but they are not duplicates: TorchAO quantization, Pyramid Attention Broadcast, FasterCache, and Layerwise Upcasting are distinct changes with different code paths and goals. No soft edges should be accepted.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the items are separate feature PRs in the same optimization area, not the same underlying change.", + "best_issue_reason": null, + "best_pr_reason": "No single best PR for deduplication; 9562 is only the earliest/broadest infra PR, but it does not represent the other distinct features.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10009", + "right": "pull_request:9562", + "accept": false, + "reason": "Different features: TorchAO quantization vs Pyramid Attention Broadcast. Shared docs/init wiring is not enough to indicate the same change." + }, + { + "left": "pull_request:10163", + "right": "pull_request:9562", + "accept": false, + "reason": "Both are cache-related, but FasterCache is a different caching algorithm and issue than Pyramid Attention Broadcast." + }, + { + "left": "pull_request:10347", + "right": "pull_request:9562", + "accept": false, + "reason": "Layerwise FP8 casting is unrelated to Pyramid Attention Broadcast; they modify different mechanisms and code paths." + }, + { + "left": "pull_request:10009", + "right": "pull_request:10163", + "accept": false, + "reason": "TorchAO quantization and FasterCache solve different problems and would not plausibly merge into one PR." + }, + { + "left": "pull_request:10009", + "right": "pull_request:10347", + "accept": false, + "reason": "Quantization support and layerwise casting are distinct optimization features, not the same bug/change." + }, + { + "left": "pull_request:10163", + "right": "pull_request:10347", + "accept": false, + "reason": "FasterCache and layerwise upcasting target different execution optimizations; overlap is only in shared hook/util plumbing." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 3, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-999-2", + "nodes": [ + "pull_request:1001", + "pull_request:999" + ], + "soft_pairs": [ + "pull_request:1001|pull_request:999" + ], + "prepared_review_unit_hash": "21fc13bced0843287492bb4ff77d426165bf9c0628a30b48fcdf2edc0edfcbab", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4259, + "estimated_input_tokens": 1065, + "estimated_eval_tokens": 2386 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4259, + "estimated_input_tokens": 1065, + "estimated_eval_tokens": 2386 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests add Flax Stable Diffusion fine-tuning examples, but for different training tasks: text-to-image vs DreamBooth. They are related thematically yet not the same underlying change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #999 is the broader, more central Stable Diffusion Flax finetuning addition and is a better representative of the cluster than the more specific DreamBooth follow-up.", + "best_issue_reason": null, + "best_pr_reason": "PR #999 is the best single PR to represent the cluster because it introduces the core Flax finetuning example for Stable Diffusion; PR #1001 is a separate DreamBooth example rather than a duplicate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1001", + "right": "pull_request:999", + "accept": false, + "reason": "Different concrete use cases and code paths: DreamBooth finetuning vs text-to-image finetuning. They touch separate example scripts/READMEs and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 4, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9905-11", + "nodes": [ + "pull_request:10032", + "pull_request:10094", + "pull_request:10217", + "pull_request:10499", + "pull_request:10713", + "pull_request:10846", + "pull_request:11128", + "pull_request:9905", + "pull_request:9935", + "pull_request:9952", + "pull_request:9960" + ], + "soft_pairs": [ + "pull_request:9905|pull_request:9952", + "pull_request:10032|pull_request:9905", + "pull_request:10094|pull_request:9905", + "pull_request:10499|pull_request:9905", + "pull_request:10713|pull_request:9905", + "pull_request:9935|pull_request:9960", + "pull_request:10217|pull_request:9935", + "pull_request:10032|pull_request:9952", + "pull_request:10094|pull_request:9952", + "pull_request:10499|pull_request:9952", + "pull_request:10713|pull_request:9952", + "pull_request:10217|pull_request:9960", + "pull_request:9905|pull_request:9960", + "pull_request:9952|pull_request:9960", + "pull_request:10032|pull_request:9960", + "pull_request:10032|pull_request:10094", + "pull_request:10032|pull_request:10499", + "pull_request:10032|pull_request:10713", + "pull_request:10094|pull_request:10499", + "pull_request:10094|pull_request:10713", + "pull_request:10499|pull_request:10713", + "pull_request:10846|pull_request:9905", + "pull_request:10846|pull_request:9952", + "pull_request:10032|pull_request:10846", + "pull_request:10094|pull_request:10846", + "pull_request:10499|pull_request:10846", + "pull_request:11128|pull_request:9905", + "pull_request:11128|pull_request:9952", + "pull_request:10032|pull_request:11128", + "pull_request:10094|pull_request:11128", + "pull_request:10499|pull_request:11128" + ], + "prepared_review_unit_hash": "6dbb48b42d869cfdfcb7f2e55e37f85b67a83855053f799022f03fe15da18f06", + "component_budget": { + "node_count": 11, + "item_count": 11, + "soft_pair_count": 31, + "serialized_chars": 32189, + "estimated_input_tokens": 8048, + "estimated_eval_tokens": 16352 + }, + "budget": { + "node_count": 11, + "item_count": 11, + "soft_pair_count": 31, + "serialized_chars": 32189, + "estimated_input_tokens": 8048, + "estimated_eval_tokens": 16352 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is really two unrelated themes: a sequence of incremental docs-only PRs adding notebook links to community-script READMEs, and separate PRs moving individual research projects. None of the soft pairs are true duplicates of the same underlying change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10094 is the strongest representative of the community-scripts notebook series: it updates the main community README and the companion community-scripts README, with higher review/discussion activity than the surrounding incremental follow-ups.", + "best_issue_reason": null, + "best_pr_reason": "PR #10094 is the best single representative for the dominant repeated-docs theme, but the overall set is heterogeneous and should not be collapsed into one duplicate change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9905", + "right": "pull_request:9952", + "accept": false, + "reason": "Same broad notebook/README area, but different files and different content; these are related follow-ups, not the same change." + }, + { + "left": "pull_request:10032", + "right": "pull_request:9905", + "accept": false, + "reason": "Both are community notebook README updates, but they touch different README generations and add different entries." + }, + { + "left": "pull_request:10094", + "right": "pull_request:9905", + "accept": false, + "reason": "Overlapping documentation area, but not the same concrete edit; 10094 updates the main community README plus the companion file." + }, + { + "left": "pull_request:10499", + "right": "pull_request:9905", + "accept": false, + "reason": "Sequential community notebook updates, but the diffs are not identical and add different notebook links." + }, + { + "left": "pull_request:10713", + "right": "pull_request:9905", + "accept": false, + "reason": "Same docs series, but this PR adds a different notebook row and is not a duplicate of 9905." + }, + { + "left": "pull_request:9935", + "right": "pull_request:9960", + "accept": false, + "reason": "Both move scripts into research_projects, but they are for different projects (Wuerstchen vs IP Adapter), so not the same change." + }, + { + "left": "pull_request:10217", + "right": "pull_request:9935", + "accept": false, + "reason": "Different migration targets and different codebases; only the general move-to-research-project theme matches." + }, + { + "left": "pull_request:10032", + "right": "pull_request:9952", + "accept": false, + "reason": "Same README family, but these are distinct notebook additions in separate PRs." + }, + { + "left": "pull_request:10094", + "right": "pull_request:9952", + "accept": false, + "reason": "Related notebook-link updates, but 10094 edits a broader set of docs and is not the same patch as 9952." + }, + { + "left": "pull_request:10499", + "right": "pull_request:9952", + "accept": false, + "reason": "Both are notebook README changes, but they introduce different rows and cannot be merged as one duplicate PR." + }, + { + "left": "pull_request:10713", + "right": "pull_request:9952", + "accept": false, + "reason": "Same subsystem, different notebook entry and later incremental update." + }, + { + "left": "pull_request:10217", + "right": "pull_request:9960", + "accept": false, + "reason": "Separate project relocation PRs; they do not fix or move the same project." + }, + { + "left": "pull_request:9905", + "right": "pull_request:9960", + "accept": false, + "reason": "Different artifact family: community README notebook links versus moving IP Adapter scripts." + }, + { + "left": "pull_request:9952", + "right": "pull_request:9960", + "accept": false, + "reason": "Different concrete changes in different paths; only loosely similar by being example docs/workflow related." + }, + { + "left": "pull_request:10032", + "right": "pull_request:9960", + "accept": false, + "reason": "Community README notebook updates are unrelated to the IP Adapter move PR." + }, + { + "left": "pull_request:10032", + "right": "pull_request:10094", + "accept": false, + "reason": "Both touch the same docs area, but they are separate incremental notebook-link additions, not the same patch." + }, + { + "left": "pull_request:10032", + "right": "pull_request:10499", + "accept": false, + "reason": "Different notebook-link updates in the same README series; not identical enough to merge as duplicates." + }, + { + "left": "pull_request:10032", + "right": "pull_request:10713", + "accept": false, + "reason": "Same series, but each PR adds different documentation entries." + }, + { + "left": "pull_request:10094", + "right": "pull_request:10499", + "accept": false, + "reason": "Sequential docs-only updates, not the same underlying change." + }, + { + "left": "pull_request:10094", + "right": "pull_request:10713", + "accept": false, + "reason": "Related community notebook README updates, but distinct content and diffs." + }, + { + "left": "pull_request:10499", + "right": "pull_request:10713", + "accept": false, + "reason": "Same broad README/table area, but different added notebook links and separate PRs." + }, + { + "left": "pull_request:10846", + "right": "pull_request:9905", + "accept": false, + "reason": "Broadly similar community notebook docs work, but different files and different rows; not a duplicate." + }, + { + "left": "pull_request:10846", + "right": "pull_request:9952", + "accept": false, + "reason": "Same documentation theme, but 10846 is a later expansion across multiple READMEs, not the same change as 9952." + }, + { + "left": "pull_request:10032", + "right": "pull_request:10846", + "accept": false, + "reason": "Both update community/example READMEs, but they are separate incremental additions with different content." + }, + { + "left": "pull_request:10094", + "right": "pull_request:10846", + "accept": false, + "reason": "Related docs updates, but 10846 also touches examples/README.md and adds different notebook links." + }, + { + "left": "pull_request:10499", + "right": "pull_request:10846", + "accept": false, + "reason": "Not the same patch; 10846 is a broader docs update than 10499." + }, + { + "left": "pull_request:11128", + "right": "pull_request:9905", + "accept": false, + "reason": "Only loosely related by community notebook docs; they modify different rows and files." + }, + { + "left": "pull_request:11128", + "right": "pull_request:9952", + "accept": false, + "reason": "Same README series, but this is a later distinct notebook addition." + }, + { + "left": "pull_request:10032", + "right": "pull_request:11128", + "accept": false, + "reason": "Incremental updates to the same documentation family, but not the same underlying change." + }, + { + "left": "pull_request:10094", + "right": "pull_request:11128", + "accept": false, + "reason": "Later notebook-link update versus an earlier broader docs update; not duplicates." + }, + { + "left": "pull_request:10499", + "right": "pull_request:11128", + "accept": false, + "reason": "Same subsystem, different notebook entries and separate PRs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 5, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10044-3", + "nodes": [ + "pull_request:10044", + "pull_request:10051", + "pull_request:10053" + ], + "soft_pairs": [ + "pull_request:10044|pull_request:10051", + "pull_request:10051|pull_request:10053", + "pull_request:10044|pull_request:10053" + ], + "prepared_review_unit_hash": "e3c9351d48ea9ea03f2aeabf9217f4dbb90a8b6fbc0c77ee179454d2be7acf84", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7492, + "estimated_input_tokens": 1873, + "estimated_eval_tokens": 4002 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7492, + "estimated_input_tokens": 1873, + "estimated_eval_tokens": 4002 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs are variations of the same Flux/SD3 attention-mask work in `attention_processor.py`. PR 10044 is the broadest implementation; 10051 and 10053 are narrower follow-ups around wiring `attn_mask` into the Flux attention call and related conversion/pipeline plumbing.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10044 is the best canonical PR: it has the broadest scope, explicit Flux & SD3 coverage, the most discussion/review activity, and looks like the main implementation rather than a narrow follow-up.", + "best_issue_reason": null, + "best_pr_reason": "PR 10044 is the strongest representative of the cluster because it addresses the feature at the source with the most complete scope and evidence of review.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10044", + "right": "pull_request:10051", + "accept": true, + "reason": "Both target the same underlying attention-mask support in `attention_processor.py` for Flux, and both are tied to the same issue. 10051 looks like a narrower implementation slice of the broader 10044 fix." + }, + { + "left": "pull_request:10051", + "right": "pull_request:10053", + "accept": true, + "reason": "These are effectively the same Flux attention-mask change: both modify the same attention call to pass `attn_mask`, with only minor surrounding plumbing differences." + }, + { + "left": "pull_request:10044", + "right": "pull_request:10053", + "accept": false, + "reason": "They overlap on `attention_processor.py`, but 10044 is a broader Flux/SD3 mask-preparation change while 10053 is a narrower Flux-only wiring change; not enough evidence that they are the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 6, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-916-36", + "nodes": [ + "pull_request:1007", + "pull_request:1110", + "pull_request:1225", + "pull_request:1834", + "pull_request:1980", + "pull_request:2031", + "pull_request:2173", + "pull_request:2478", + "pull_request:2818", + "pull_request:2859", + "pull_request:2978", + "pull_request:3360", + "pull_request:3425", + "pull_request:3550", + "pull_request:3560", + "pull_request:3810", + "pull_request:3896", + "pull_request:4016", + "pull_request:4038", + "pull_request:4097", + "pull_request:4505", + "pull_request:5778", + "pull_request:6422", + "pull_request:6514", + "pull_request:6554", + "pull_request:6751", + "pull_request:6840", + "pull_request:7126", + "pull_request:7242", + "pull_request:8446", + "pull_request:8461", + "pull_request:8476", + "pull_request:8743", + "pull_request:916", + "pull_request:9308", + "pull_request:9316" + ], + "soft_pairs": [ + "pull_request:8461|pull_request:9316", + "pull_request:8446|pull_request:8461", + "pull_request:8461|pull_request:9308", + "pull_request:8476|pull_request:9308", + "pull_request:8446|pull_request:9308", + "pull_request:6514|pull_request:7126", + "pull_request:4016|pull_request:5778", + "pull_request:4097|pull_request:6514", + "pull_request:3896|pull_request:4097", + "pull_request:6554|pull_request:7126", + "pull_request:4016|pull_request:4505", + "pull_request:6554|pull_request:7242", + "pull_request:4097|pull_request:4505", + "pull_request:5778|pull_request:6514", + "pull_request:2031|pull_request:4097", + "pull_request:4038|pull_request:4097", + "pull_request:4097|pull_request:5778", + "pull_request:3425|pull_request:3560", + "pull_request:5778|pull_request:7126", + "pull_request:4505|pull_request:5778", + "pull_request:6422|pull_request:7126", + "pull_request:2478|pull_request:2978", + "pull_request:2818|pull_request:2859", + "pull_request:5778|pull_request:7242", + "pull_request:6751|pull_request:7126", + "pull_request:6422|pull_request:6554", + "pull_request:4016|pull_request:4038", + "pull_request:3550|pull_request:3896", + "pull_request:1110|pull_request:1980", + "pull_request:1225|pull_request:1980", + "pull_request:3550|pull_request:4016", + "pull_request:6840|pull_request:7126", + "pull_request:1007|pull_request:1980", + "pull_request:6422|pull_request:6514", + "pull_request:3360|pull_request:3550", + "pull_request:5778|pull_request:6422", + "pull_request:4016|pull_request:6422", + "pull_request:3810|pull_request:4016", + "pull_request:3896|pull_request:5778", + "pull_request:4038|pull_request:5778", + "pull_request:5778|pull_request:6751", + "pull_request:1980|pull_request:2031", + "pull_request:6514|pull_request:8446", + "pull_request:2031|pull_request:4016", + "pull_request:2818|pull_request:4038", + "pull_request:3810|pull_request:4505", + "pull_request:3360|pull_request:3425", + "pull_request:1980|pull_request:916", + "pull_request:6514|pull_request:8476", + "pull_request:2031|pull_request:3896", + "pull_request:2859|pull_request:4505", + "pull_request:3896|pull_request:4038", + "pull_request:2859|pull_request:2978", + "pull_request:1834|pull_request:2978", + "pull_request:3810|pull_request:3896", + "pull_request:2978|pull_request:4016", + "pull_request:2478|pull_request:4505", + "pull_request:2031|pull_request:2173", + "pull_request:1834|pull_request:3810", + "pull_request:7242|pull_request:8743", + "pull_request:8461|pull_request:8476", + "pull_request:8446|pull_request:8476", + "pull_request:6514|pull_request:6554", + "pull_request:8476|pull_request:9316", + "pull_request:8446|pull_request:9316", + "pull_request:9308|pull_request:9316", + "pull_request:3896|pull_request:4016", + "pull_request:7126|pull_request:7242", + "pull_request:6514|pull_request:6840", + "pull_request:4016|pull_request:4097", + "pull_request:6514|pull_request:7242", + "pull_request:4505|pull_request:6751" + ], + "prepared_review_unit_hash": "ca89445835212f449f9c3c0d90fe9a6bdb9da42e76196307d69353d5bbb5e9c9", + "component_budget": { + "node_count": 36, + "item_count": 36, + "soft_pair_count": 83, + "serialized_chars": 105750, + "estimated_input_tokens": 26438, + "estimated_eval_tokens": 53132 + }, + "budget": { + "node_count": 36, + "item_count": 36, + "soft_pair_count": 72, + "serialized_chars": 95388, + "estimated_input_tokens": 23847, + "estimated_eval_tokens": 47950 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mixed cluster with a few real duplicate families: distributed LR-scheduler fixes across example scripts, and fp16/state-dict resume fixes in LoRA training. Most other pairs are adjacent feature work or docs and should stay separate. The 9308 links to ControlNet scheduler-fix PRs are too weak to merge and should be rejected.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 8461 is the cleanest anchor for the strongest duplicate family: a focused one-file distributed LR-scheduler bug fix with a direct sibling in 8476 and the same fix pattern repeated elsewhere.", + "best_issue_reason": null, + "best_pr_reason": "PR 8461 is the best global representative because it is a crisp, concrete bug fix rather than a feature/docs change, and it sits at the center of the clearest duplicate subcluster in this packet.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8461", + "right": "pull_request:9316", + "accept": true, + "reason": "Same distributed LR-scheduler math bug and same fix pattern in sibling example training scripts." + }, + { + "left": "pull_request:8446", + "right": "pull_request:8461", + "accept": true, + "reason": "Same scheduler computation bug in a closely related example training script; the patch shape is the same." + }, + { + "left": "pull_request:8461", + "right": "pull_request:9308", + "accept": false, + "reason": "They share a broad scheduler-fix theme, but the packet marks this edge non-deterministic and the scripts are different enough that this is not a single mergeable change." + }, + { + "left": "pull_request:8476", + "right": "pull_request:9308", + "accept": false, + "reason": "Same broad training-scheduler theme, but different example script and not the same concrete code path; too weak for duplicate triage." + }, + { + "left": "pull_request:8446", + "right": "pull_request:9308", + "accept": false, + "reason": "Shared bug class only; the packet does not ground this as the same mergeable patch, so reject conservatively." + }, + { + "left": "pull_request:6514", + "right": "pull_request:7126", + "accept": false, + "reason": "Different fixes in different training flows: fp16 resume/state-dict restoration versus EDM-style training support." + }, + { + "left": "pull_request:4016", + "right": "pull_request:5778", + "accept": false, + "reason": "Different SDXL training features in different scripts; related domain, but not the same underlying change." + }, + { + "left": "pull_request:4097", + "right": "pull_request:6514", + "accept": false, + "reason": "One is text-encoder fine-tuning support, the other is an fp16 resume fix; not the same bug or change." + }, + { + "left": "pull_request:3896", + "right": "pull_request:4097", + "accept": false, + "reason": "WIP SDXL DreamBooth LoRA support versus a later text-encoder-finetuning follow-up; related, but not a duplicate." + }, + { + "left": "pull_request:6554", + "right": "pull_request:7126", + "accept": false, + "reason": "Both are DreamBooth/LoRA training-related, but they fix different problems in different scripts." + }, + { + "left": "pull_request:4016", + "right": "pull_request:4505", + "accept": false, + "reason": "Different SDXL training scripts and different feature goals; not the same change." + }, + { + "left": "pull_request:6422", + "right": "pull_request:7126", + "accept": false, + "reason": "Separate research/training recipes with different functionality; only superficially related." + }, + { + "left": "pull_request:2478", + "right": "pull_request:2978", + "accept": false, + "reason": "Original example addition versus a later multi-GPU bug fix; not duplicates." + }, + { + "left": "pull_request:2818", + "right": "pull_request:2859", + "accept": false, + "reason": "Both are ControlNet JAX-related, but one adds the training script and the other adds streaming support; different changes." + }, + { + "left": "pull_request:5778", + "right": "pull_request:7242", + "accept": false, + "reason": "Different SDXL DreamBooth LoRA follow-ups; the prior-preservation fix is not the same change as the EDM-style training work." + }, + { + "left": "pull_request:6751", + "right": "pull_request:7126", + "accept": false, + "reason": "Different SDXL training examples and different fix scopes." + }, + { + "left": "pull_request:6422", + "right": "pull_request:6554", + "accept": false, + "reason": "Different training recipes and different bug fixes; no duplicate code path." + }, + { + "left": "pull_request:4016", + "right": "pull_request:4038", + "accept": false, + "reason": "Different model families and training features; not the same underlying bug or change." + }, + { + "left": "pull_request:3550", + "right": "pull_request:3896", + "accept": false, + "reason": "Debug/discussion PR versus broader SDXL LoRA feature work; not the same artifact." + }, + { + "left": "pull_request:5778", + "right": "pull_request:7126", + "accept": false, + "reason": "Different training approaches and different code paths." + }, + { + "left": "pull_request:4505", + "right": "pull_request:5778", + "accept": false, + "reason": "Different text-to-image SDXL workstreams; only loosely related." + }, + { + "left": "pull_request:6514", + "right": "pull_request:6840", + "accept": true, + "reason": "Same fp16 resume/state-dict restoration bug pattern in LoRA training scripts." + }, + { + "left": "pull_request:2478", + "right": "pull_request:4505", + "accept": false, + "reason": "Different example families and different functionality." + }, + { + "left": "pull_request:2818", + "right": "pull_request:4038", + "accept": false, + "reason": "One is JAX ControlNet training, the other is PyTorch SDXL ControlNet support; not the same change." + }, + { + "left": "pull_request:3810", + "right": "pull_request:4016", + "accept": false, + "reason": "Model-card/docs cleanup versus SDXL DreamBooth LoRA feature work; not duplicates." + }, + { + "left": "pull_request:3896", + "right": "pull_request:5778", + "accept": false, + "reason": "Unrelated SDXL training efforts with different goals." + }, + { + "left": "pull_request:4038", + "right": "pull_request:4097", + "accept": false, + "reason": "Different pipeline/feature work, not the same underlying bug or change." + }, + { + "left": "pull_request:4505", + "right": "pull_request:6751", + "accept": false, + "reason": "Both are text-to-image SDXL-related, but they are different additions/fixes in different files." + }, + { + "left": "pull_request:5778", + "right": "pull_request:6514", + "accept": false, + "reason": "Different training recipes and fixes; no shared concrete bug." + }, + { + "left": "pull_request:4016", + "right": "pull_request:4097", + "accept": true, + "reason": "Same SDXL DreamBooth LoRA script line and the later PR is a direct follow-up on the same training implementation." + }, + { + "left": "pull_request:6514", + "right": "pull_request:6554", + "accept": true, + "reason": "Same fp16 resume bug pattern applied to another LoRA training script." + }, + { + "left": "pull_request:8461", + "right": "pull_request:8476", + "accept": true, + "reason": "Same distributed LR-scheduler fix across sibling example scripts." + }, + { + "left": "pull_request:8446", + "right": "pull_request:8476", + "accept": true, + "reason": "Same scheduler bug and same fix pattern in a related example training script." + }, + { + "left": "pull_request:8476", + "right": "pull_request:9316", + "accept": true, + "reason": "Same LR-scheduler bug fix pattern in another example training script." + }, + { + "left": "pull_request:8446", + "right": "pull_request:9316", + "accept": true, + "reason": "Same scheduler-math bug and same style of fix, just in a different example script." + }, + { + "left": "pull_request:9308", + "right": "pull_request:9316", + "accept": true, + "reason": "Same distributed scheduler computation bug across example training scripts." + }, + { + "left": "pull_request:3896", + "right": "pull_request:4016", + "accept": true, + "reason": "Same SDXL DreamBooth LoRA feature family; 4016 is the polished continuation of the earlier WIP implementation." + }, + { + "left": "pull_request:7126", + "right": "pull_request:7242", + "accept": false, + "reason": "EDM-style training support versus a prior-preservation bug fix: related file, different concrete change." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "Reject: the analyst overstates mergeability in the distributed LR-scheduler cluster. In the packet, the pairs pull_request:8461\u2013pull_request:9308, pull_request:8476\u2013pull_request:9308, and pull_request:8446\u2013pull_request:9308 all have deterministic_accept=false, but the analyst marked them accept. The summary is broadly aligned, but these soft-edge verdicts are not conservative enough and are not grounded by the packet." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 6, + "component_count": 266, + "review_unit_index": 2, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-916-36", + "nodes": [ + "pull_request:1007", + "pull_request:1110", + "pull_request:1225", + "pull_request:5778", + "pull_request:6514", + "pull_request:6554", + "pull_request:6751", + "pull_request:6840", + "pull_request:7242", + "pull_request:916" + ], + "soft_pairs": [ + "pull_request:6554|pull_request:6840", + "pull_request:1110|pull_request:1225", + "pull_request:1007|pull_request:1110", + "pull_request:1007|pull_request:1225", + "pull_request:6751|pull_request:6840", + "pull_request:6751|pull_request:7242", + "pull_request:5778|pull_request:6840", + "pull_request:1007|pull_request:916", + "pull_request:1110|pull_request:916", + "pull_request:1225|pull_request:916", + "pull_request:6514|pull_request:6751" + ], + "prepared_review_unit_hash": "02027bb584d3f21bd19ea9f5271bfc24795f8200e39ec81a07a1a1bf3d833034", + "component_budget": { + "node_count": 36, + "item_count": 36, + "soft_pair_count": 83, + "serialized_chars": 105750, + "estimated_input_tokens": 26438, + "estimated_eval_tokens": 53132 + }, + "budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 11, + "serialized_chars": 23329, + "estimated_input_tokens": 5833, + "estimated_eval_tokens": 11922 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is actually two loose themes: an ORT integration subcluster around unconditional image generation, and several separate fp16 resume-fix PRs for different LoRA/DreamBooth-style training scripts. I treat only the ORT PRs as true duplicates/same change; the resume-fix PRs are related but not the same concrete code path.", + "confidence": 0.73, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 1110 is the merged, most complete ORT integration: it adds the dedicated ORT example, docs, and the supporting pipeline change, so it best represents the final change in this subcluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 1110 is the strongest representative overall because it is merged and contains the finished ORT example plus documentation, making it the clearest canonical PR for the ORT work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6554", + "right": "pull_request:6840", + "accept": false, + "reason": "Both address fp16 resume issues, but they touch different training scripts and code paths (DreamBooth SD vs. LCM consistency distillation), so they are related fixes rather than the same duplicate change." + }, + { + "left": "pull_request:1110", + "right": "pull_request:1225", + "accept": true, + "reason": "Both are ORT integrations for the unconditional image generation training example and aim at the same underlying change: making that training flow work with ONNX Runtime." + }, + { + "left": "pull_request:1007", + "right": "pull_request:1110", + "accept": true, + "reason": "Both implement the same ORT-enabled unconditional training support, with overlapping edits to the example and shared pipeline support, so they look like alternate versions of the same change." + }, + { + "left": "pull_request:1007", + "right": "pull_request:1225", + "accept": true, + "reason": "These are both ORT integration variants for the same unconditional training script and supporting model code, differing mainly in implementation shape rather than underlying goal." + }, + { + "left": "pull_request:6751", + "right": "pull_request:6840", + "accept": false, + "reason": "They both touch LoRA training resume behavior, but they are for different example scripts and different training scenarios, so they are not the same concrete fix." + }, + { + "left": "pull_request:6751", + "right": "pull_request:7242", + "accept": false, + "reason": "These address unrelated problems: mixed-precision fine-tuning resume vs. prior-preservation collate handling in DreamBooth SDXL." + }, + { + "left": "pull_request:5778", + "right": "pull_request:6840", + "accept": false, + "reason": "Different feature areas and different files; one is an LCM LoRA datasets-version training addition, the other is a resume-from-fp16 bug fix." + }, + { + "left": "pull_request:1007", + "right": "pull_request:916", + "accept": true, + "reason": "Both are about integrating ORT into the unconditional image generation training stack and share the same supporting core changes, so they appear to be the same underlying feature work." + }, + { + "left": "pull_request:1110", + "right": "pull_request:916", + "accept": true, + "reason": "Both are ORT integration work for unconditional training, with overlapping example/core changes; they look like different iterations of the same change." + }, + { + "left": "pull_request:1225", + "right": "pull_request:916", + "accept": true, + "reason": "Both target ORT support for the unconditional training example and core model/pipeline compatibility, so they are close enough to treat as duplicate change work." + }, + { + "left": "pull_request:6514", + "right": "pull_request:6751", + "accept": false, + "reason": "They are both resume-related LoRA fixes, but they affect different example scripts and are not the same code-path issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 7, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-52-3", + "nodes": [ + "pull_request:1008", + "pull_request:52", + "pull_request:85" + ], + "soft_pairs": [ + "pull_request:52|pull_request:85", + "pull_request:1008|pull_request:52" + ], + "prepared_review_unit_hash": "d1fea38def82d49544e2200ee780c04bfa34624593bdcea3112979be47713bc2", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6605, + "estimated_input_tokens": 1652, + "estimated_eval_tokens": 3560 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6605, + "estimated_input_tokens": 1652, + "estimated_eval_tokens": 3560 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These three PRs are only loosely similar by the word \u201cclean\u201d; they target different areas (docs, UNetNCSNpp/resnet internals, and unconditional UNet refactoring) and are not duplicates of the same change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 52 is the most concrete code-path change in the cluster; PR 85 is a separate UNet refactor and PR 1008 is docs-only.", + "best_issue_reason": null, + "best_pr_reason": "PR 52 is the best representative PR here because it makes the clearest implementation change, while the others are unrelated cleanup work in different subsystems.", + "soft_edge_verdicts": [ + { + "left": "pull_request:52", + "right": "pull_request:85", + "accept": false, + "reason": "Different code paths and goals: PR 52 cleans up UNetNCSNpp/resnet internals, while PR 85 refactors unconditional UNet modules and tests. They are not the same underlying fix and would not plausibly merge as one PR." + }, + { + "left": "pull_request:1008", + "right": "pull_request:52", + "accept": false, + "reason": "Docs cleanup vs model implementation cleanup. PR 1008 only edits documentation pages, while PR 52 changes resnet and score-estimation model code; no shared bug or concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 8, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10081-2", + "nodes": [ + "pull_request:10081", + "pull_request:10116" + ], + "soft_pairs": [ + "pull_request:10081|pull_request:10116" + ], + "prepared_review_unit_hash": "956ec3a44cd96cd78ff98cb4faaf8fe7ae0b4ff0d2effa2292cce032a7832873", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5368, + "estimated_input_tokens": 1342, + "estimated_eval_tokens": 2940 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5368, + "estimated_input_tokens": 1342, + "estimated_eval_tokens": 2940 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related API-surface updates adding `sigmas` support, but they target different pipeline families and different code paths, so they are not duplicates of the same concrete change.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10116 is the broader, more representative change: it updates multiple FlowMatch-based pipelines and reflects the shared API shift more comprehensively than the Flux-only PR.", + "best_issue_reason": null, + "best_pr_reason": "PR #10116 is the best representative artifact for this cluster because it covers the wider set of affected pipelines and the common `sigmas` API update pattern.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10081", + "right": "pull_request:10116", + "accept": false, + "reason": "Both add `sigmas`, but one is Flux-specific and the other targets FlowMatch-based pipelines in different files and code paths. They are conceptually similar, not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 9, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9776-2", + "nodes": [ + "pull_request:10131", + "pull_request:9776" + ], + "soft_pairs": [ + "pull_request:10131|pull_request:9776" + ], + "prepared_review_unit_hash": "1e418c24bcd8073afdd21e35c0eda6465a2ae5502d42ff9358281db8b77a2282", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6117, + "estimated_input_tokens": 1530, + "estimated_eval_tokens": 3316 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6117, + "estimated_input_tokens": 1530, + "estimated_eval_tokens": 3316 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both artifacts are the same ControlNetUnion feature addition against issue #9772. PR #10131 is the merged implementation and PR #9776 is an earlier, unmerged attempt that touches the same files and code paths.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10131 is the best canonical PR because it is the merged, later completion of the same ControlNetUnion feature and supersedes the earlier unmerged attempt.", + "best_issue_reason": null, + "best_pr_reason": "PR #10131 is the strongest representative of the cluster: it is merged, has the same explicit issue target, and covers the same ControlNetUnion implementation plus docs/API wiring as PR #9776.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10131", + "right": "pull_request:9776", + "accept": true, + "reason": "Same title, same explicit issue target, and substantial overlap in the exact ControlNetUnion files and pipelines; this looks like the same underlying feature change rather than merely related work." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 10, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10155-3", + "nodes": [ + "pull_request:10155", + "pull_request:10156", + "pull_request:10161" + ], + "soft_pairs": [ + "pull_request:10155|pull_request:10161", + "pull_request:10156|pull_request:10161" + ], + "prepared_review_unit_hash": "cb199d9273ffdaa612480432927b1ec4bddcd7c84dc69a98953180a3665d2e1c", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7436, + "estimated_input_tokens": 1859, + "estimated_eval_tokens": 3974 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7436, + "estimated_input_tokens": 1859, + "estimated_eval_tokens": 3974 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate but closely related PRs that all add device/torch output support to positional-embedding helpers and then update downstream call sites. They share a pattern, but they do not fix the same concrete code path.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10156 is the broadest and most central change: it updates shared 2D/3D sincos embedding helpers in `src/diffusers/models/embeddings.py` and propagates the new torch/device behavior into multiple consumers. It is the best representative of the cluster\u2019s underlying refactor pattern.", + "best_issue_reason": null, + "best_pr_reason": "PR 10156 is the best overall representative because it touches the shared embedding utilities rather than a single pipeline, and it addresses the core implementation pattern that the other PRs follow in narrower scopes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10155", + "right": "pull_request:10161", + "accept": false, + "reason": "Both are about positional embeddings, but one changes `get_2d_rotary_pos_embed` and the other changes `get_3d_rotary_pos_embed`/Allegro. Different helper functions and different downstream pipelines, so they are related refactors rather than the same bug fix." + }, + { + "left": "pull_request:10156", + "right": "pull_request:10161", + "accept": false, + "reason": "PR 10156 updates 2D/3D sincos embedding helpers, while PR 10161 updates 3D rotary embedding helpers. They touch different code paths and would not naturally be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 11, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10173-2", + "nodes": [ + "pull_request:10173", + "pull_request:10509" + ], + "soft_pairs": [ + "pull_request:10173|pull_request:10509" + ], + "prepared_review_unit_hash": "f0b8af3575739dfd3a8e761ec00656307c2f7d5f198b743c87ddd29e9cf318fa", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4386, + "estimated_input_tokens": 1097, + "estimated_eval_tokens": 2450 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4386, + "estimated_input_tokens": 1097, + "estimated_eval_tokens": 2450 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs only overlap at a broad file/subsystem level in the modular SDXL pipeline. PR #10173 adds Modular APG guidance changes, while PR #10509 adds SDXL ControlNet Union denoising logic. They are different features and not the same underlying fix/change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the PRs implement different features and should not be treated as duplicates.", + "best_issue_reason": null, + "best_pr_reason": "#10509 is the better representative only in the sense that it is a self-contained, clearly scoped modular SDXL feature; #10173 is unrelated APG work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10173", + "right": "pull_request:10509", + "accept": false, + "reason": "Reject: they touch the same broad modular SDXL code, but one is for APG guidance and the other is for ControlNet Union denoising. They do not appear to fix the same concrete code-path problem and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 12, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10189-2", + "nodes": [ + "pull_request:10189", + "pull_request:10714" + ], + "soft_pairs": [ + "pull_request:10189|pull_request:10714" + ], + "prepared_review_unit_hash": "cfd9d4f226a65ff97593a7f97f1eb43bd0ee94316a99ff70fc5715970991f50b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4024, + "estimated_input_tokens": 1006, + "estimated_eval_tokens": 2268 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4024, + "estimated_input_tokens": 1006, + "estimated_eval_tokens": 2268 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: one PR fixes runtime model-type validation in `from_pretrained`, while the other only adds a `Self` return type annotation for type inference. They touch the same file but different problems and would not be merged as one change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10189 is the better canonical PR because it addresses the concrete runtime bug in `from_pretrained` by checking the correct model type; it is the more substantive and behavior-changing fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #10189 is the best representative of the cluster since it fixes the actual code-path problem rather than only improving typing.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10189", + "right": "pull_request:10714", + "accept": false, + "reason": "Different underlying changes: #10189 validates the passed model type at runtime, while #10714 only annotates `from_pretrained` with `Self` for static type checking." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 13, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9093-8", + "nodes": [ + "pull_request:10228", + "pull_request:10254", + "pull_request:9093", + "pull_request:9150", + "pull_request:9302", + "pull_request:9312", + "pull_request:9389", + "pull_request:9482" + ], + "soft_pairs": [ + "pull_request:9093|pull_request:9150", + "pull_request:9093|pull_request:9302", + "pull_request:9093|pull_request:9389", + "pull_request:9093|pull_request:9482", + "pull_request:9150|pull_request:9302", + "pull_request:9150|pull_request:9389", + "pull_request:9150|pull_request:9482", + "pull_request:9150|pull_request:9312", + "pull_request:9302|pull_request:9482", + "pull_request:9302|pull_request:9389", + "pull_request:10254|pull_request:9302", + "pull_request:10228|pull_request:9482", + "pull_request:10254|pull_request:9482", + "pull_request:10228|pull_request:10254", + "pull_request:10254|pull_request:9150" + ], + "prepared_review_unit_hash": "ecf3ccbcaef459b4f8b032517f1fdb1ea24c2eeb72e26a75e4dd21a1a3660102", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 15, + "serialized_chars": 24940, + "estimated_input_tokens": 6235, + "estimated_eval_tokens": 12726 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 15, + "serialized_chars": 24940, + "estimated_input_tokens": 6235, + "estimated_eval_tokens": 12726 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is broad CogVideoX/video-LoRA related work, but the PRs are not true duplicates of one another; they target different models, pipelines, or feature additions.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "No exact duplicate stands out. If one PR must represent the cluster, 9302 is the broadest CogVideoX LoRA training PR and is the most central to the shared example/training theme.", + "best_issue_reason": null, + "best_pr_reason": "9302 is the best representative for the cluster theme because it is the main CogVideoX LoRA training PR and has the clearest overlap with the only adjacent training-example follow-up (9482), while the others are separate model or pipeline features.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9093", + "right": "pull_request:9150", + "accept": false, + "reason": "Different changes: 9093 adds dynamic CFG for CogVideoX, while 9150 is CogVideoX follow-ups plus tiled decoding support." + }, + { + "left": "pull_request:9093", + "right": "pull_request:9302", + "accept": false, + "reason": "9093 is a pipeline/model behavior change; 9302 is a LoRA training example for CogVideoX." + }, + { + "left": "pull_request:9093", + "right": "pull_request:9389", + "accept": false, + "reason": "Dynamic CFG and FreeNoise are separate CogVideoX experiments/features, not the same concrete fix." + }, + { + "left": "pull_request:9093", + "right": "pull_request:9482", + "accept": false, + "reason": "9093 changes inference behavior; 9482 adds CogVideoX-I2V LoRA training scripts and docs." + }, + { + "left": "pull_request:9150", + "right": "pull_request:9302", + "accept": false, + "reason": "9150 is a pipeline refactor/tiled decoding change, not the LoRA training work in 9302." + }, + { + "left": "pull_request:9150", + "right": "pull_request:9389", + "accept": false, + "reason": "CogVideoX tiled-decoding follow-ups and FreeNoise are distinct pipeline experiments." + }, + { + "left": "pull_request:9150", + "right": "pull_request:9482", + "accept": false, + "reason": "9150 targets CogVideoX pipeline behavior; 9482 targets I2V LoRA training examples." + }, + { + "left": "pull_request:9150", + "right": "pull_request:9312", + "accept": false, + "reason": "9150 is a CogVideoX pipeline feature PR; 9312 is just a small rotary-embedding refactor/comment/dtype cleanup." + }, + { + "left": "pull_request:9302", + "right": "pull_request:9482", + "accept": false, + "reason": "Both are training-related, but 9302 is text-to-video LoRA while 9482 is image-to-video LoRA; separate scripts and pipeline support." + }, + { + "left": "pull_request:9302", + "right": "pull_request:9389", + "accept": false, + "reason": "9302 is a LoRA training example; 9389 is a FreeNoise experiment on the CogVideoX inference stack." + }, + { + "left": "pull_request:10254", + "right": "pull_request:9302", + "accept": false, + "reason": "Different models and tasks: HunyuanVideo LoRA support versus CogVideoX LoRA training." + }, + { + "left": "pull_request:10228", + "right": "pull_request:9482", + "accept": false, + "reason": "LTX Video LoRA support and CogVideoX I2V LoRA training are unrelated model families and changes." + }, + { + "left": "pull_request:10254", + "right": "pull_request:9482", + "accept": false, + "reason": "HunyuanVideo LoRA support is not the same change as CogVideoX I2V LoRA training." + }, + { + "left": "pull_request:10228", + "right": "pull_request:10254", + "accept": false, + "reason": "These support different video models entirely: LTX Video versus HunyuanVideo." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 14, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10229-3", + "nodes": [ + "pull_request:10229", + "pull_request:10488", + "pull_request:11414" + ], + "soft_pairs": [ + "pull_request:10229|pull_request:10488", + "pull_request:10488|pull_request:11414" + ], + "prepared_review_unit_hash": "fd6069cc69d641b65ccc923a5effa3fbf20681395e2b6c1e2a86d4746f544b6a", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6576, + "estimated_input_tokens": 1644, + "estimated_eval_tokens": 3544 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6576, + "estimated_input_tokens": 1644, + "estimated_eval_tokens": 3544 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "No duplicate cluster here: one docs-only LTX PR, one LTX VAE feature PR, and one Wan VAE feature PR. The only suggested soft match is too broad because it targets a different model and code path.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:10488 is the clearest canonical artifact because it makes the concrete LTX Video VAE code change (framewise encoding/decoding) and has the most direct relation to the cluster\u2019s shared LTX VAE topic.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:10488 is the best representative PR: it is a substantive implementation change in the LTX Video VAE, while pull_request:10229 is just documentation and pull_request:11414 is a similar feature for a different model (Wan), not the same underlying change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10229", + "right": "pull_request:10488", + "accept": false, + "reason": "Docs-only LTX video changes are not the same underlying bug/change as implementing framewise VAE encoding/decoding in code." + }, + { + "left": "pull_request:10488", + "right": "pull_request:11414", + "accept": false, + "reason": "These are similar feature themes, but they modify different models and code paths (LTX Video VAE vs Wan VAE), so they are not duplicates and would not plausibly merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 15, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10234-7", + "nodes": [ + "pull_request:10234", + "pull_request:11111", + "pull_request:11281", + "pull_request:11381", + "pull_request:11494", + "pull_request:11858", + "pull_request:12056" + ], + "soft_pairs": [ + "pull_request:10234|pull_request:11494", + "pull_request:10234|pull_request:11281", + "pull_request:10234|pull_request:11111", + "pull_request:10234|pull_request:11381", + "pull_request:11111|pull_request:11281", + "pull_request:11111|pull_request:11381", + "pull_request:11281|pull_request:11381", + "pull_request:11381|pull_request:11494", + "pull_request:11381|pull_request:12056", + "pull_request:11494|pull_request:12056", + "pull_request:11858|pull_request:12056" + ], + "prepared_review_unit_hash": "5e5d3ffd129d4a01ddd34eaedd42273c19c8b9e8bf8f816a3e244b178ee62c57", + "component_budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 11, + "serialized_chars": 19502, + "estimated_input_tokens": 4876, + "estimated_eval_tokens": 10008 + }, + "budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 11, + "serialized_chars": 19502, + "estimated_input_tokens": 4876, + "estimated_eval_tokens": 10008 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Loose topical cluster around DreamBooth/LoRA training and docs, but each PR targets a different model family or a different follow-up feature. None of these look like true duplicates of the same underlying change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "#11281 is the best representative PR because it is the broadest, most complete LoRA integration change in the set (core loader/pipeline changes plus docs/tests). That said, the cluster is not composed of duplicates, so it should not be treated as a canonical duplicate target.", + "best_issue_reason": null, + "best_pr_reason": "#11281 is the strongest overall representative for this cluster: it implements a substantial end-to-end LoRA feature for HiDream with code, docs, and tests. The others are narrower model-specific additions or follow-up README/script tweaks, not the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10234", + "right": "pull_request:11494", + "accept": false, + "reason": "Different model/features: SANA LoRA support vs HiDream LoRA quantization support. Not the same concrete change." + }, + { + "left": "pull_request:10234", + "right": "pull_request:11281", + "accept": false, + "reason": "Both are LoRA work, but for different model families (SANA vs HiDream) and different code paths." + }, + { + "left": "pull_request:10234", + "right": "pull_request:11111", + "accept": false, + "reason": "SANA LoRA support is unrelated to Flux LoRA script fixes; same subsystem, but not the same underlying bug/change." + }, + { + "left": "pull_request:10234", + "right": "pull_request:11381", + "accept": false, + "reason": "SANA LoRA support vs HiDream README/script optimizations. Different model and different change scope." + }, + { + "left": "pull_request:11111", + "right": "pull_request:11281", + "accept": false, + "reason": "Flux LoRA script fixes and HiDream LoRA support address different model pipelines and different problems." + }, + { + "left": "pull_request:11111", + "right": "pull_request:11381", + "accept": false, + "reason": "Flux script fixes are not the same as HiDream training README updates/optimizations." + }, + { + "left": "pull_request:11281", + "right": "pull_request:11381", + "accept": false, + "reason": "Same model family, but #11281 adds core HiDream LoRA support while #11381 is a narrower README/example optimization follow-up. Not one identical change." + }, + { + "left": "pull_request:11381", + "right": "pull_request:11494", + "accept": false, + "reason": "Both touch HiDream, but one is a README/example optimization update and the other adds quantization support in the training flow; separate changes." + }, + { + "left": "pull_request:11381", + "right": "pull_request:12056", + "accept": false, + "reason": "HiDream README/script updates vs Qwen Image LoRA support; different model families and unrelated code paths." + }, + { + "left": "pull_request:11494", + "right": "pull_request:12056", + "accept": false, + "reason": "HiDream quantization support is not the same as Qwen Image LoRA support." + }, + { + "left": "pull_request:11858", + "right": "pull_request:12056", + "accept": false, + "reason": "Kontext I2I training support vs Qwen Image LoRA support; different model features and not mergeable as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 16, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1024-2", + "nodes": [ + "pull_request:1024", + "pull_request:1085" + ], + "soft_pairs": [ + "pull_request:1024|pull_request:1085" + ], + "prepared_review_unit_hash": "552d74a71454a5b287614d739731f0038b34edad2728d471b0b6fbd2c081a506", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4649, + "estimated_input_tokens": 1163, + "estimated_eval_tokens": 2582 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4649, + "estimated_input_tokens": 1163, + "estimated_eval_tokens": 2582 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about sequential CPU offloading in Stable Diffusion, but they are not the same change: #1024 primarily documents the feature and touches the base pipeline/docs, while #1085 adds the offload method to img2img/inpaint plus tests. Related feature work, not duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1085 is the stronger canonical representative because it contains the concrete code change and tests for enabling sequential CPU offload in additional Stable Diffusion pipelines.", + "best_issue_reason": null, + "best_pr_reason": "PR #1085 is the best overall PR in this cluster since it implements the feature in code paths and adds validation, whereas #1024 is mostly documentation/supporting pipeline work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1024", + "right": "pull_request:1085", + "accept": false, + "reason": "Related to the same offload feature area, but #1024 is documentation/base-pipeline support and #1085 extends the method to img2img/inpaint with tests; they are not the same concrete change and would not plausibly merge as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 17, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9891-2", + "nodes": [ + "pull_request:10252", + "pull_request:9891" + ], + "soft_pairs": [ + "pull_request:10252|pull_request:9891" + ], + "prepared_review_unit_hash": "795339f249bd146d1de2c0ac2bc2a67e871c29b2e59ebec44debaa83f5b19a58", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2898, + "estimated_input_tokens": 725, + "estimated_eval_tokens": 1706 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2898, + "estimated_input_tokens": 725, + "estimated_eval_tokens": 1706 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs both mention checkpoint-saving behavior, but they target different problems: PR #10252 is a concrete bugfix in the SD3 DreamBooth LoRA save hook, while PR #9891 is a broader checkpointing-step fix tied to issue #2606. I do not consider them duplicates.", + "confidence": 0.83, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10252 is the stronger canonical PR because it has a specific code-path and diff: it fixes the LoRA checkpoint save hook in `examples/dreambooth/train_dreambooth_lora_sd3.py`, including model-type matching and optional upcasting before saving.", + "best_issue_reason": null, + "best_pr_reason": "PR #10252 is the best representative of this cluster because it is narrowly scoped, has an explicit diff, and addresses a specific saving bug rather than a broad checkpointing concern.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10252", + "right": "pull_request:9891", + "accept": false, + "reason": "They both relate to saving/checkpointing, but the overlap is too broad. #10252 fixes a specific LoRA save-hook bug in one DreamBooth script, while #9891 is a general checkpointing-steps fix with no evidence it touches the same code path or bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 18, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1028-3", + "nodes": [ + "pull_request:1028", + "pull_request:1071", + "pull_request:1142" + ], + "soft_pairs": [ + "pull_request:1028|pull_request:1071", + "pull_request:1028|pull_request:1142", + "pull_request:1071|pull_request:1142" + ], + "prepared_review_unit_hash": "4f919d2c2f19e396e9607838a26a9637716251a7c86d8cac55940c1e61d40adc", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 9303, + "estimated_input_tokens": 2326, + "estimated_eval_tokens": 4908 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 9303, + "estimated_input_tokens": 2326, + "estimated_eval_tokens": 4908 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These three PRs are closely related versions of the same multilingual Stable Diffusion community pipeline work. PR #1142 looks like the final, merged, most focused form of the change; #1071 is a broader unmerged variant; #1028 is an earlier proposal of the same feature.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1142 is the best canonical PR because it is the merged, most focused implementation of the multilingual Stable Diffusion community pipeline, while #1071 is a broader unmerged draft and #1028 is an earlier proposal.", + "best_issue_reason": null, + "best_pr_reason": "PR #1142 is the best representative of the cluster: it captures the core feature with minimal extra noise and is already merged, making it the most stable reference point.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1028", + "right": "pull_request:1071", + "accept": true, + "reason": "Both PRs center on the multilingual Stable Diffusion community pipeline; #1071 appears to be a later, expanded version of the same feature, so they represent the same underlying change." + }, + { + "left": "pull_request:1028", + "right": "pull_request:1142", + "accept": true, + "reason": "Both PRs target the same multilingual Stable Diffusion community pipeline addition, with #1142 looking like the finalized, narrower version of the same change." + }, + { + "left": "pull_request:1071", + "right": "pull_request:1142", + "accept": true, + "reason": "They have the same title and share the key pipeline file, indicating the same core multilingual Stable Diffusion change; #1142 is a slimmer follow-up/finalized variant of #1071." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 19, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10284-2", + "nodes": [ + "pull_request:10284", + "pull_request:10300" + ], + "soft_pairs": [ + "pull_request:10284|pull_request:10300" + ], + "prepared_review_unit_hash": "7e0f66c26476a14d90d046bb3165287a4234520026d991ca54e3b509bab9e9f4", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3533, + "estimated_input_tokens": 884, + "estimated_eval_tokens": 2024 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3533, + "estimated_input_tokens": 884, + "estimated_eval_tokens": 2024 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two Mochi PRs are unrelated code changes: one makes empty-prompt zeroing configurable in the pipeline, while the other enables splitting the Mochi transformer across multiple GPUs. They should not be clustered as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10284 is the better canonical representative because it is a concrete end-user behavior change in the Mochi pipeline and is more directly tied to a single bug/feature area than the GPU-sharding transformer change.", + "best_issue_reason": null, + "best_pr_reason": "PR #10284 is the best single PR to keep as the representative change in this cluster; PR #10300 addresses a separate model-parallelism concern and does not overlap the same code-path problem.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10284", + "right": "pull_request:10300", + "accept": false, + "reason": "Different fixes in different files and code paths: configurable empty-prompt embedding zeroing in the pipeline versus adding split support for the transformer. They share only the broad Mochi subsystem, not the same underlying bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 20, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10308-3", + "nodes": [ + "pull_request:10308", + "pull_request:11515", + "pull_request:11520" + ], + "soft_pairs": [ + "pull_request:10308|pull_request:11515", + "pull_request:10308|pull_request:11520", + "pull_request:11515|pull_request:11520" + ], + "prepared_review_unit_hash": "023f58fff308298dea14b790beeb598a4e10993a3b67bd044f53ea979f9d3d2b", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7322, + "estimated_input_tokens": 1831, + "estimated_eval_tokens": 3918 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7322, + "estimated_input_tokens": 1831, + "estimated_eval_tokens": 3918 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are only loosely similar at the theme level (model-related changes and tests), but they address different code paths: `_no_split_modules` support, group offloading training coverage, and a framepack gradient-checkpointing fix. They should not be deduplicated.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10308 is the clearest concrete code fix in the cluster: it adds `_no_split_modules` support to specific models and includes targeted tests. The other PRs are different changes focused on test coverage or a separate model attribute fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #10308 is the best representative because it contains the most substantive functional change and is already merged; the others are separate, narrower changes rather than the same underlying fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10308", + "right": "pull_request:11515", + "accept": false, + "reason": "Different problems: #10308 adds `_no_split_modules` support for model splitting/device_map, while #11515 adds a training-time group-offloading test. Same general subsystem, but not the same bug or change." + }, + { + "left": "pull_request:10308", + "right": "pull_request:11520", + "accept": false, + "reason": "#10308 is about `_no_split_modules`; #11520 changes a framepack model attribute from `use_gradient_checkpointing` to `gradient_checkpointing` and adds tests. Distinct fixes, not mergeable as one PR." + }, + { + "left": "pull_request:11515", + "right": "pull_request:11520", + "accept": false, + "reason": "Both touch tests, but they target different models and issues: group offloading training coverage versus a framepack gradient-checkpointing/model-init fix. Not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 21, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10363-2", + "nodes": [ + "pull_request:10363", + "pull_request:10561" + ], + "soft_pairs": [ + "pull_request:10363|pull_request:10561" + ], + "prepared_review_unit_hash": "b72ee631837d16c6edd7feae58a8641267cc1ae6bf36e0ea40858517aff3e916", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4532, + "estimated_input_tokens": 1133, + "estimated_eval_tokens": 2522 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4532, + "estimated_input_tokens": 1133, + "estimated_eval_tokens": 2522 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs add IP-Adapter support to closely related Stable Diffusion 3 ControlNet pipelines, but they target different concrete code paths: the base ControlNet pipeline vs. the inpainting variant. They are related feature siblings, not true duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical duplicate PR: these are separate implementations for different SD3 ControlNet pipeline variants, not the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR 10363 is the best representative because it is the earlier base-pipeline IP-Adapter addition, has an explicit issue link, and establishes the core implementation pattern later mirrored by the inpainting PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10363", + "right": "pull_request:10561", + "accept": false, + "reason": "Same feature family, but different pipeline classes and files; they do not fix the same concrete code-path problem and would not naturally collapse into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 22, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6255-18", + "nodes": [ + "pull_request:10377", + "pull_request:10766", + "pull_request:10984", + "pull_request:11234", + "pull_request:11364", + "pull_request:11415", + "pull_request:11429", + "pull_request:11664", + "pull_request:11686", + "pull_request:11748", + "pull_request:11924", + "pull_request:12279", + "pull_request:6255", + "pull_request:6539", + "pull_request:7635", + "pull_request:7873", + "pull_request:8971", + "pull_request:9956" + ], + "soft_pairs": [ + "pull_request:6255|pull_request:7635", + "pull_request:6255|pull_request:7873", + "pull_request:6255|pull_request:6539", + "pull_request:6539|pull_request:7873", + "pull_request:6539|pull_request:7635", + "pull_request:7635|pull_request:7873", + "pull_request:7635|pull_request:8971", + "pull_request:7873|pull_request:8971", + "pull_request:10377|pull_request:8971", + "pull_request:8971|pull_request:9956", + "pull_request:11234|pull_request:9956", + "pull_request:11415|pull_request:9956", + "pull_request:11429|pull_request:9956", + "pull_request:10377|pull_request:9956", + "pull_request:10984|pull_request:9956", + "pull_request:10377|pull_request:10984", + "pull_request:10377|pull_request:11234", + "pull_request:10377|pull_request:10766", + "pull_request:10377|pull_request:11664", + "pull_request:10766|pull_request:11364", + "pull_request:10766|pull_request:11664", + "pull_request:10766|pull_request:10984", + "pull_request:10766|pull_request:11429", + "pull_request:10984|pull_request:11234", + "pull_request:10984|pull_request:11364", + "pull_request:10984|pull_request:11664", + "pull_request:11234|pull_request:11429", + "pull_request:11234|pull_request:11686", + "pull_request:11364|pull_request:11664", + "pull_request:11364|pull_request:12279", + "pull_request:11364|pull_request:11429", + "pull_request:11415|pull_request:11924", + "pull_request:11415|pull_request:11429", + "pull_request:11415|pull_request:11748", + "pull_request:11234|pull_request:11415", + "pull_request:11429|pull_request:11664", + "pull_request:11429|pull_request:11686", + "pull_request:11429|pull_request:11748", + "pull_request:11664|pull_request:12279", + "pull_request:11686|pull_request:11748", + "pull_request:11686|pull_request:11924", + "pull_request:10377|pull_request:11686", + "pull_request:11234|pull_request:11748", + "pull_request:11748|pull_request:11924", + "pull_request:11234|pull_request:11924", + "pull_request:11429|pull_request:11924", + "pull_request:11429|pull_request:12279", + "pull_request:11748|pull_request:12279", + "pull_request:11415|pull_request:12279" + ], + "prepared_review_unit_hash": "599b30c037f90866f02306d131c43061aace2367c19dd324782e76e5374d4dac", + "component_budget": { + "node_count": 18, + "item_count": 18, + "soft_pair_count": 49, + "serialized_chars": 45396, + "estimated_input_tokens": 11349, + "estimated_eval_tokens": 22954 + }, + "budget": { + "node_count": 18, + "item_count": 18, + "soft_pair_count": 49, + "serialized_chars": 45396, + "estimated_input_tokens": 11349, + "estimated_eval_tokens": 22954 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a mix of unrelated dependency-bump PRs across two different example projects (realfill and server). The soft similarities are mostly superficial because the PRs change different packages and/or different version steps, so I would not merge any of them as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11429 is the best representative overall: it is merged, materially updates both the server input and lockfile, and sits near the center of the server-side dependency refresh chain. That said, the cluster is heterogeneous and not a true duplicate set.", + "best_issue_reason": null, + "best_pr_reason": "11429 is the strongest single PR to anchor the cluster because it is a concrete, merged dependency update with broader file impact than the others; the rest are distinct package/version bumps rather than the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6255", + "right": "pull_request:7635", + "accept": false, + "reason": "Same file and project, but different transformer version steps (4.34\u21924.36 vs 4.36\u21924.38); sequential updates, not duplicates." + }, + { + "left": "pull_request:6255", + "right": "pull_request:7873", + "accept": false, + "reason": "Different package changes in the same file (transformers vs Jinja2)." + }, + { + "left": "pull_request:6255", + "right": "pull_request:6539", + "accept": false, + "reason": "Different package changes in the same file (transformers vs Jinja2)." + }, + { + "left": "pull_request:6539", + "right": "pull_request:7873", + "accept": false, + "reason": "Same file/project, but different Jinja2 version steps (3.1.2\u21923.1.3 vs 3.1.3\u21923.1.4); not the same change." + }, + { + "left": "pull_request:6539", + "right": "pull_request:7635", + "accept": false, + "reason": "Different package changes in the same file (Jinja2 vs transformers)." + }, + { + "left": "pull_request:7635", + "right": "pull_request:7873", + "accept": false, + "reason": "Different package changes in the same file (transformers vs Jinja2)." + }, + { + "left": "pull_request:7635", + "right": "pull_request:8971", + "accept": false, + "reason": "Same project/file, but different Torch version steps (2.0.1\u21922.2.0 vs 2.2.0\u21922.6.0); not duplicates." + }, + { + "left": "pull_request:7873", + "right": "pull_request:8971", + "accept": false, + "reason": "Different package changes in the same file (Jinja2 vs torch)." + }, + { + "left": "pull_request:10377", + "right": "pull_request:8971", + "accept": false, + "reason": "Same project/file, but different dependency changes (Jinja2 vs torch) and different fix goals." + }, + { + "left": "pull_request:8971", + "right": "pull_request:9956", + "accept": false, + "reason": "Different projects and different packages (realfill torch vs server aiohttp)." + }, + { + "left": "pull_request:11234", + "right": "pull_request:9956", + "accept": false, + "reason": "Same server requirements file, but different package bumps (Jinja2 vs aiohttp)." + }, + { + "left": "pull_request:11415", + "right": "pull_request:9956", + "accept": false, + "reason": "Same server requirements file, but different package bumps (h11 vs aiohttp)." + }, + { + "left": "pull_request:11429", + "right": "pull_request:9956", + "accept": false, + "reason": "Different package bumps in server requirements (torch vs aiohttp) and different scope." + }, + { + "left": "pull_request:10377", + "right": "pull_request:9956", + "accept": false, + "reason": "Different projects and different packages (realfill Jinja2 vs server aiohttp)." + }, + { + "left": "pull_request:10984", + "right": "pull_request:9956", + "accept": false, + "reason": "Different projects and different packages (realfill Jinja2 vs server aiohttp)." + }, + { + "left": "pull_request:10377", + "right": "pull_request:10984", + "accept": false, + "reason": "Same project/file, but different Jinja2 version steps (3.1.4\u21923.1.5 vs 3.1.5\u21923.1.6); sequential, not duplicate." + }, + { + "left": "pull_request:10377", + "right": "pull_request:11234", + "accept": false, + "reason": "Different projects and different packages (realfill Jinja2 vs server Jinja2), not the same change." + }, + { + "left": "pull_request:10377", + "right": "pull_request:10766", + "accept": false, + "reason": "Different projects and different packages (realfill Jinja2 vs server transformers)." + }, + { + "left": "pull_request:10377", + "right": "pull_request:11664", + "accept": false, + "reason": "Different projects and different packages (realfill Jinja2 vs realfill torch)." + }, + { + "left": "pull_request:10766", + "right": "pull_request:11364", + "accept": false, + "reason": "Same file/project, but different packages (transformers vs torch)." + }, + { + "left": "pull_request:10766", + "right": "pull_request:11664", + "accept": false, + "reason": "Same file/project, but different packages (transformers vs torch)." + }, + { + "left": "pull_request:10766", + "right": "pull_request:10984", + "accept": false, + "reason": "Different package changes in the same file (transformers vs Jinja2)." + }, + { + "left": "pull_request:10766", + "right": "pull_request:11429", + "accept": false, + "reason": "Different projects and different packages (realfill transformers vs server torch)." + }, + { + "left": "pull_request:10984", + "right": "pull_request:11234", + "accept": false, + "reason": "Different projects and different packages (realfill Jinja2 vs server Jinja2), not the same change." + }, + { + "left": "pull_request:10984", + "right": "pull_request:11364", + "accept": false, + "reason": "Same project/file, but different packages (Jinja2 vs torch)." + }, + { + "left": "pull_request:10984", + "right": "pull_request:11664", + "accept": false, + "reason": "Same project/file, but different packages (Jinja2 vs torch)." + }, + { + "left": "pull_request:11234", + "right": "pull_request:11429", + "accept": false, + "reason": "Same server file, but different dependency bumps (Jinja2 vs torch)." + }, + { + "left": "pull_request:11234", + "right": "pull_request:11686", + "accept": false, + "reason": "Different server dependency bumps (Jinja2 vs requests)." + }, + { + "left": "pull_request:11364", + "right": "pull_request:11664", + "accept": false, + "reason": "Same project/file, but different Torch version steps (2.2.0\u21922.6.0 vs 2.2.0\u21922.7.1); not duplicates." + }, + { + "left": "pull_request:11364", + "right": "pull_request:12279", + "accept": false, + "reason": "Same project/file, but different Torch version steps (2.6.0 vs 2.8.0 target); distinct updates." + }, + { + "left": "pull_request:11364", + "right": "pull_request:11429", + "accept": false, + "reason": "Different projects and different packages (realfill torch vs server torch), with different lockfile contexts." + }, + { + "left": "pull_request:11415", + "right": "pull_request:11924", + "accept": false, + "reason": "Same server file, but different packages (h11 vs aiohttp)." + }, + { + "left": "pull_request:11415", + "right": "pull_request:11429", + "accept": false, + "reason": "Same server file, but different package changes (h11 vs torch)." + }, + { + "left": "pull_request:11415", + "right": "pull_request:11748", + "accept": false, + "reason": "Different server dependency bumps (h11 vs urllib3)." + }, + { + "left": "pull_request:11234", + "right": "pull_request:11415", + "accept": false, + "reason": "Same server file, but different package bumps (Jinja2 vs h11)." + }, + { + "left": "pull_request:11429", + "right": "pull_request:11664", + "accept": false, + "reason": "Different projects and different Torch update targets (server torch vs realfill torch)." + }, + { + "left": "pull_request:11429", + "right": "pull_request:11686", + "accept": false, + "reason": "Different server dependency bumps (torch vs requests)." + }, + { + "left": "pull_request:11429", + "right": "pull_request:11748", + "accept": false, + "reason": "Different server dependency bumps (torch vs urllib3)." + }, + { + "left": "pull_request:11429", + "right": "pull_request:11924", + "accept": false, + "reason": "Different server dependency bumps (torch vs aiohttp)." + }, + { + "left": "pull_request:11429", + "right": "pull_request:12279", + "accept": false, + "reason": "Different Torch updates in different projects/files; not the same concrete fix." + }, + { + "left": "pull_request:11664", + "right": "pull_request:12279", + "accept": false, + "reason": "Same project/file, but different Torch version steps (2.2.0\u21922.7.1 vs 2.2.0\u21922.8.0); sequential updates, not duplicates." + }, + { + "left": "pull_request:11686", + "right": "pull_request:11748", + "accept": false, + "reason": "Different server dependency bumps (requests vs urllib3)." + }, + { + "left": "pull_request:11686", + "right": "pull_request:11924", + "accept": false, + "reason": "Different server dependency bumps (requests vs aiohttp)." + }, + { + "left": "pull_request:11748", + "right": "pull_request:11924", + "accept": false, + "reason": "Different server dependency bumps (urllib3 vs aiohttp)." + }, + { + "left": "pull_request:11234", + "right": "pull_request:11924", + "accept": false, + "reason": "Same server file, but different package bumps (Jinja2 vs aiohttp)." + }, + { + "left": "pull_request:10377", + "right": "pull_request:11686", + "accept": false, + "reason": "Different projects and different packages (realfill Jinja2 vs server requests)." + }, + { + "left": "pull_request:11748", + "right": "pull_request:12279", + "accept": false, + "reason": "Different projects and different packages (server urllib3 vs realfill torch)." + }, + { + "left": "pull_request:11415", + "right": "pull_request:12279", + "accept": false, + "reason": "Different projects and different packages (server h11 vs realfill torch)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 23, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10400-2", + "nodes": [ + "pull_request:10400", + "pull_request:11189" + ], + "soft_pairs": [ + "pull_request:10400|pull_request:11189" + ], + "prepared_review_unit_hash": "f93c4fde766781e0d772bfb763dfd980af10092321c77c501cde568ee3d4b0a4", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3868, + "estimated_input_tokens": 967, + "estimated_eval_tokens": 2190 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3868, + "estimated_input_tokens": 967, + "estimated_eval_tokens": 2190 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs touch related pipeline-component validation logic, but they fix different code paths and are not the same underlying change. PR #10400 addresses AutoPipeline.from_pipe using the source pipeline\u2019s optional-components set instead of the target pipeline class, while PR #11189 fixes a generic optional-kwargs set conversion in pipeline_utils.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10400 is the more canonical representative because it targets a concrete, narrowly scoped AutoPipeline bug, includes an explicit issue link, and directly fixes the specific failing code path described in the title and diff.", + "best_issue_reason": null, + "best_pr_reason": "PR #10400 is the best representative overall because it is the more specific, clearly described bug fix and is tied to an explicit issue; PR #11189 is broader and addresses a different validation-path detail.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10400", + "right": "pull_request:11189", + "accept": false, + "reason": "Related subsystem, but different concrete bugs and different functions/files: #10400 fixes AutoPipeline.from_pipe optional-component comparison, while #11189 fixes set handling in pipeline_utils mandatory/optional component checking. They would not plausibly be merged into one PR as the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 24, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10401-2", + "nodes": [ + "pull_request:10401", + "pull_request:10574" + ], + "soft_pairs": [ + "pull_request:10401|pull_request:10574" + ], + "prepared_review_unit_hash": "7e1270f4133b83a0126d0b166ef7f5fc455cd7a97786dbcfb4b442973e4b8c4a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4568, + "estimated_input_tokens": 1142, + "estimated_eval_tokens": 2540 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4568, + "estimated_input_tokens": 1142, + "estimated_eval_tokens": 2540 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two bitsandbytes-related PRs in the same area, but they address different changes: one refactors int8 dequantization, the other enables multi-backend bitsandbytes support and removes the CUDA-only gate.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10574 is the broader, more central change: it updates environment validation and backend selection to support multi-backend bitsandbytes, with related test and utility changes.", + "best_issue_reason": null, + "best_pr_reason": "PR #10574 is the best representative of the cluster because it covers the main behavioral shift in bitsandbytes quantization support, whereas #10401 is a narrower helper refactor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10401", + "right": "pull_request:10574", + "accept": false, + "reason": "They only overlap on bitsandbytes utilities; #10401 changes dequantization logic, while #10574 changes backend validation and GPU requirement handling. Different concrete code paths, so not a duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 25, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9986-2", + "nodes": [ + "pull_request:10417", + "pull_request:9986" + ], + "soft_pairs": [ + "pull_request:10417|pull_request:9986" + ], + "prepared_review_unit_hash": "85840d0978af8616d58750911860eff03c7d1d30c4f25005ad16515e5f425c67", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4428, + "estimated_input_tokens": 1107, + "estimated_eval_tokens": 2470 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4428, + "estimated_input_tokens": 1107, + "estimated_eval_tokens": 2470 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related model-search example PRs, but they are not the same concrete change: #9986 adds the model-search example and supporting setup, while #10417 later refactors the example to use auto-load helpers and update README snippets. I would keep them separate and treat #9986 as the canonical PR for the feature.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #9986 is the broader, original feature addition for the model-search example and related setup; it best represents the underlying change cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR #9986 is the strongest representative because it introduces the feature end-to-end, whereas #10417 is a narrower follow-up/example refinement.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10417", + "right": "pull_request:9986", + "accept": false, + "reason": "Same area and files, but different change intent: #9986 adds the model-search example stack, while #10417 only updates the example to use auto-load helpers. They do not look like the same fix/change and would not naturally merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 26, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10440-2", + "nodes": [ + "pull_request:10440", + "pull_request:10441" + ], + "soft_pairs": [ + "pull_request:10440|pull_request:10441" + ], + "prepared_review_unit_hash": "35e2afc7a49217c977dd3a1a41db81f99d316357bb89ebf1db7828affb70cf20", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6918, + "estimated_input_tokens": 1730, + "estimated_eval_tokens": 3716 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6918, + "estimated_input_tokens": 1730, + "estimated_eval_tokens": 3716 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two distinct merged PRs that both harden community pipelines against missing components, but they fix different problems: one adds guards for missing UNet, the other for missing VAE. Similar pattern, not the same change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #10440 is the better canonical representative because it is the broader, earlier fix for the 'without unet' case and touches the larger set of affected pipelines.", + "best_issue_reason": null, + "best_pr_reason": "PR #10440 is the strongest overall representative of the cluster theme: it addresses a concrete initialization failure when pipelines are used without a UNet, across many community pipelines.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10440", + "right": "pull_request:10441", + "accept": false, + "reason": "Reject: these are different bug fixes. #10440 guards UNet-dependent version/sample-size checks when UNet is absent; #10441 guards VAE-dependent scale-factor setup when VAE is absent. They share a broad pattern but not the same underlying code path or mergeable change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 27, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9818-2", + "nodes": [ + "pull_request:10456", + "pull_request:9818" + ], + "soft_pairs": [ + "pull_request:10456|pull_request:9818" + ], + "prepared_review_unit_hash": "6abfcfef02c86d6d07537a9ea48f7a2aa8172a47a74142befd5e42a58f66caa9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4271, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4271, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs add support for forwarding custom attention kwargs, but they do so in different model-specific code paths (SD3 vs CogVideoX). They are related in theme, not the same concrete change, so this is not a duplicate pair.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #9818 is the better canonical PR because it is the earlier, more general implementation of kwargs plumbing for custom attention processors and has an explicit issue target; PR #10456 is a separate CogVideoX-specific follow-up.", + "best_issue_reason": null, + "best_pr_reason": "PR #9818 is the strongest representative of this change family: it patches the shared SD3 attention path and is tied to an explicit issue, whereas PR #10456 is a separate model-specific adaptation in CogVideoX.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10456", + "right": "pull_request:9818", + "accept": false, + "reason": "Same broad feature area, but different model code paths and different concrete edits. This is not the same underlying bug/change and would not naturally be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 28, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9428-2", + "nodes": [ + "pull_request:10476", + "pull_request:9428" + ], + "soft_pairs": [ + "pull_request:10476|pull_request:9428" + ], + "prepared_review_unit_hash": "a8bc1aa56bd5626754480e7bb46b8a340f711b56723621013919a5a333b22da3", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6574, + "estimated_input_tokens": 1644, + "estimated_eval_tokens": 3544 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6574, + "estimated_input_tokens": 1644, + "estimated_eval_tokens": 3544 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are both PRs about updating Stable Diffusion model references, but they target different model IDs and different scopes: #10476 updates code/examples from runwayml/stable-diffusion-v1-5 to stable-diffusion-v1-5 paths, while #9428 is a docs-only replacement to Lykon/dreamshaper-8. They are related in theme but not the same change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "#10476 is the stronger canonical PR for this cluster because it makes the broader code-path change across loaders/pipelines/examples, whereas #9428 is primarily documentation updates with a different model target.", + "best_issue_reason": null, + "best_pr_reason": "#10476 best represents the underlying maintenance pattern in this cluster: fixing hardcoded Stable Diffusion v1.5 repo paths in executable code, not just docs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10476", + "right": "pull_request:9428", + "accept": false, + "reason": "Different concrete changes and targets: #10476 updates code to new Stable Diffusion v1.5 paths, while #9428 replaces docs references with Lykon/dreamshaper-8. Same broad area, but not the same bug or mergeable change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 29, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1052-2", + "nodes": [ + "pull_request:1052", + "pull_request:1113" + ], + "soft_pairs": [ + "pull_request:1052|pull_request:1113" + ], + "prepared_review_unit_hash": "b7adf1ccd5581747f5826d110ac9de0264c0ca5fa5daffb62ec8098bcc723419", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4553, + "estimated_input_tokens": 1139, + "estimated_eval_tokens": 2534 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4553, + "estimated_input_tokens": 1139, + "estimated_eval_tokens": 2534 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are merged PRs about improving diffusion test precision, but they touch different test areas and implement different fixes. They are related in theme, not duplicates.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 1052 is the cleaner representative: it has an explicit issue target and a focused change adding a helper for inpainting integration-test precision updates.", + "best_issue_reason": null, + "best_pr_reason": "PR 1052 is the best cluster representative because it is the most concrete, well-scoped change with direct tracking context; PR 1113 is broader and less specific.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1052", + "right": "pull_request:1113", + "accept": false, + "reason": "Reject: both are test-precision improvements, but they modify different tests and expected artifacts, with no evidence they fix the same underlying code-path or could be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 30, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10527-3", + "nodes": [ + "pull_request:10527", + "pull_request:11771", + "pull_request:11793" + ], + "soft_pairs": [ + "pull_request:10527|pull_request:11771", + "pull_request:10527|pull_request:11793", + "pull_request:11771|pull_request:11793" + ], + "prepared_review_unit_hash": "d613ed63b1e4ddbb5c85e78cdfca347dd2d8f6b810b25c44712997856d86c3fc", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7606, + "estimated_input_tokens": 1902, + "estimated_eval_tokens": 4060 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7606, + "estimated_input_tokens": 1902, + "estimated_eval_tokens": 4060 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are all related test-skip cleanup PRs, but they target different test cases and code paths. They share a theme, not the same underlying change, so none should be merged as duplicates.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #11771 is the best representative of the cluster because it updates shared test infrastructure in `tests/models/test_modeling_common.py` and clearly fixes a concrete skip path with `pytest.skip()` rather than a broad test-only cleanup.", + "best_issue_reason": null, + "best_pr_reason": "PR #11771 is the strongest canonical PR candidate: it addresses a specific runtime skip behavior in a central shared test helper, whereas #10527 is a broad cross-file test annotation cleanup and #11793 is a separate UNet-only hotswapping test gate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10527", + "right": "pull_request:11771", + "accept": false, + "reason": "Both are test-skip related, but #10527 is a broad multi-file `unittest.skip()` cleanup while #11771 changes `test_modeling_common.py` for gradient-checkpointing skip behavior. Different tests and different concrete changes." + }, + { + "left": "pull_request:10527", + "right": "pull_request:11793", + "accept": false, + "reason": "#10527 adds `unittest.skip()` to several unsupported tests across many files; #11793 changes two UNet-only hotswapping tests in `test_modeling_common.py`. Same theme, not the same bug/change." + }, + { + "left": "pull_request:11771", + "right": "pull_request:11793", + "accept": false, + "reason": "Both touch `test_modeling_common.py`, but they fix different paths: gradient-checkpointing support vs UNet-only hotswapping skips. They are not the same concrete change and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 31, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10615-3", + "nodes": [ + "pull_request:10615", + "pull_request:11741", + "pull_request:12001" + ], + "soft_pairs": [ + "pull_request:10615|pull_request:11741", + "pull_request:11741|pull_request:12001" + ], + "prepared_review_unit_hash": "e967bd7266311c698a232d1d5f8f79038ae649b6f404f11da838b1c321c6eece", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10045, + "estimated_input_tokens": 2512, + "estimated_eval_tokens": 5280 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10045, + "estimated_input_tokens": 2512, + "estimated_eval_tokens": 5280 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related licensing-chore PRs, but they are not duplicates: two only bump copyright years, while 12001 also adds the missing license line.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a true canonical duplicate. If one must be picked, 12001 is the broadest license-header cleanup, but it is a different change from the year-only updates.", + "best_issue_reason": null, + "best_pr_reason": "12001 is the most representative of the cluster\u2019s licensing work because it completes the license statement across several scripts, though it is still distinct from the year-only edits.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10615", + "right": "pull_request:11741", + "accept": false, + "reason": "Both are copyright-year bumps (2024\u21922025) in different file sets, but they are separate repo-wide chore passes rather than the same concrete change." + }, + { + "left": "pull_request:11741", + "right": "pull_request:12001", + "accept": false, + "reason": "11741 only updates the year in test headers, while 12001 adds the missing license sentence and touches different scripts; related licensing cleanup, but not a duplicate." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 32, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9080-3", + "nodes": [ + "pull_request:10626", + "pull_request:9080", + "pull_request:9082" + ], + "soft_pairs": [ + "pull_request:9080|pull_request:9082", + "pull_request:10626|pull_request:9082" + ], + "prepared_review_unit_hash": "05c7025acf1132f3e8f1f9aeaf326aa946ea05fbe58547a4c5cb3cc989a21ce7", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10580, + "estimated_input_tokens": 2645, + "estimated_eval_tokens": 5546 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10580, + "estimated_input_tokens": 2645, + "estimated_eval_tokens": 5546 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The CogVideoX PRs (9080 and 9082) are the same underlying model-integration change and should be clustered; the EasyAnimate PR is a separate model addition and should not be merged with CogVideoX.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9082 is the clearest representative: it has the explicit CogVideoX title, is merged, and contains the full end-to-end model/pipeline/docs integration for that feature.", + "best_issue_reason": null, + "best_pr_reason": "PR 9082 is the best PR for the cluster because it is the more clearly scoped and documented CogVideoX implementation, while 9080 appears to be an earlier/less descriptive variant of the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9080", + "right": "pull_request:9082", + "accept": true, + "reason": "Same concrete feature: both add CogVideoX text-to-video support with the same core files and pipeline/model conversion pieces, so they look like duplicate/parallel versions of one PR." + }, + { + "left": "pull_request:10626", + "right": "pull_request:9082", + "accept": false, + "reason": "Different features: 10626 adds EasyAnimateV5.1 support, while 9082 adds CogVideoX support. They touch overlapping framework files, but not the same underlying model or code-path change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 33, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8899-4", + "nodes": [ + "pull_request:10721", + "pull_request:8899", + "pull_request:9037", + "pull_request:9969" + ], + "soft_pairs": [ + "pull_request:8899|pull_request:9037", + "pull_request:8899|pull_request:9969", + "pull_request:10721|pull_request:9969" + ], + "prepared_review_unit_hash": "a25ff1884df92a5bf0358ba81bf4ceaad2cc18b1e1a92cd78b4f3c4815e9a129", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 8551, + "estimated_input_tokens": 2138, + "estimated_eval_tokens": 4532 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 8551, + "estimated_input_tokens": 2138, + "estimated_eval_tokens": 4532 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The only near-duplicate pair is the two TensorRT community example PRs; the other PRs are unrelated changes and should stay separate.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:9037 is the best representative of the shared TensorRT community-pipeline update because it covers the same example/API migration theme as 8899, but across more of the community pipelines.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:9037 is the strongest canonical representative for the cluster\u2019s only cohesive change family; it best captures the TensorRT community example refresh, while the other PRs are unrelated dependency/library updates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8899", + "right": "pull_request:9037", + "accept": true, + "reason": "Both PRs update TensorRT community pipeline examples and README usage to the newer DiffusionPipeline-style API; they are the same change family and could plausibly be merged into one PR." + }, + { + "left": "pull_request:8899", + "right": "pull_request:9969", + "accept": false, + "reason": "8899 is an example/pipeline code update, while 9969 is only a dependency version bump for huggingface-hub; different underlying change." + }, + { + "left": "pull_request:10721", + "right": "pull_request:9969", + "accept": false, + "reason": "10721 is a broad internal/latest-diffusers update touching loaders and model code, not the same concrete issue as a standalone dependency pin bump in 9969." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 34, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10742-3", + "nodes": [ + "pull_request:10742", + "pull_request:10811", + "pull_request:11192" + ], + "soft_pairs": [ + "pull_request:10742|pull_request:11192", + "pull_request:10742|pull_request:10811", + "pull_request:10811|pull_request:11192" + ], + "prepared_review_unit_hash": "338e45e79434b74161e8400928c801158b8cecc54beba920abefee34c0c9c7a0", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6728, + "estimated_input_tokens": 1682, + "estimated_eval_tokens": 3620 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6728, + "estimated_input_tokens": 1682, + "estimated_eval_tokens": 3620 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are all closely related type-annotation cleanup for `Self` in diffusers mixin APIs, but they change different methods and code paths, so they are not duplicates of the same fix/change.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11192 is the broadest representative of this cluster: it applies `Self` typing to two core mixin/config/scheduler APIs, making it the best single exemplar of the overall annotation cleanup theme.", + "best_issue_reason": null, + "best_pr_reason": "PR 11192 is the most representative single PR because it covers multiple related core utilities and reflects the broader API-typing sweep better than the more narrow single-method PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10742", + "right": "pull_request:11192", + "accept": false, + "reason": "Both add `Self` annotations, but they touch different APIs (`ModelMixin.from_pretrained` vs `ConfigMixin.from_config` / scheduler utilities) and are not the same concrete change." + }, + { + "left": "pull_request:10742", + "right": "pull_request:10811", + "accept": false, + "reason": "Different targets and behaviors: one annotates `ModelMixin.from_pretrained`, the other updates single-file loading mixins and pipeline utilities. Same theme, not the same PR." + }, + { + "left": "pull_request:10811", + "right": "pull_request:11192", + "accept": false, + "reason": "Both are type-hint refactors, but they modify different loader/mixin paths versus config/scheduler APIs; they would not plausibly merge into one bug-fix PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 35, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1078-4", + "nodes": [ + "pull_request:1078", + "pull_request:1118", + "pull_request:2166", + "pull_request:2258" + ], + "soft_pairs": [ + "pull_request:1078|pull_request:1118", + "pull_request:1078|pull_request:2166", + "pull_request:2166|pull_request:2258" + ], + "prepared_review_unit_hash": "6bd5a5f5800cf3d73ee41ebaabba8117a66286bd4282df35eca945029820caf9", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 7707, + "estimated_input_tokens": 1927, + "estimated_eval_tokens": 4110 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 7707, + "estimated_input_tokens": 1927, + "estimated_eval_tokens": 4110 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are separate diffusers fixes. The only close pair is the two `safety_checker=None` PRs, but one handles CPU offload and the other handles saving/loading pipelines, so they should not be deduplicated. The other pairs are clearly unrelated.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 1118 is the broadest and most self-contained `None`-component fix in the set, covering save/load behavior for pipelines with `safety_checker=None` and generalizing beyond a single runtime path.", + "best_issue_reason": null, + "best_pr_reason": "PR 1118 is the best representative of the cluster because it addresses a concrete, reusable pipeline serialization edge case and has the clearest generalized scope among the PRs shown.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1078", + "right": "pull_request:1118", + "accept": false, + "reason": "Both mention `safety_checker=None`, but they fix different code paths: 1078 prevents CPU-offload crashes, while 1118 handles saving/loading pipelines with missing components. Not the same underlying change." + }, + { + "left": "pull_request:1078", + "right": "pull_request:2166", + "accept": false, + "reason": "Unrelated fixes: one is about optional safety checker offload, the other removes an unnecessary model copy during sample generation in training." + }, + { + "left": "pull_request:2166", + "right": "pull_request:2258", + "accept": false, + "reason": "Different subsystems and bugs: training-time model unwrapping/copying versus an MPS cross-attention dtype issue. No shared underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 36, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10836-4", + "nodes": [ + "pull_request:10836", + "pull_request:11836", + "pull_request:12150", + "pull_request:12733" + ], + "soft_pairs": [ + "pull_request:10836|pull_request:11836", + "pull_request:10836|pull_request:12150", + "pull_request:11836|pull_request:12733" + ], + "prepared_review_unit_hash": "423149ec084f37f36c9c30a99ad329b855711af17cb886c0d244b82b77a8f002", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 7440, + "estimated_input_tokens": 1860, + "estimated_eval_tokens": 3976 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 7440, + "estimated_input_tokens": 1860, + "estimated_eval_tokens": 3976 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are superficially similar only because they contain small cleanup-style changes, but they target different code paths and different underlying changes. The cluster does not form a true duplicate set.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10836 is the best representative of the cluster because it is a merged, generic debug-print cleanup and covers more than one file, making it the broadest example of the shared theme.", + "best_issue_reason": null, + "best_pr_reason": "PR 10836 is the strongest global representative: it is merged, clearly a cleanup PR, and its changes are directly about removing stray prints without introducing unrelated behavior changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10836", + "right": "pull_request:11836", + "accept": false, + "reason": "Both remove debug output, but in different modules and unrelated code paths: model loading vs SCM timesteps. They are not the same underlying fix and would not be one coherent PR." + }, + { + "left": "pull_request:10836", + "right": "pull_request:12150", + "accept": false, + "reason": "This is not a duplicate cleanup PR; 12150 changes CogView4 normalization behavior and architecture details, while 10836 only removes prints. Different bug/change entirely." + }, + { + "left": "pull_request:11836", + "right": "pull_request:12733", + "accept": false, + "reason": "Different subsystems and different debug artifacts: scheduler print removal vs Flux2 saving a tensor to disk. Same broad theme, but not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 37, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10880-4", + "nodes": [ + "pull_request:10880", + "pull_request:10907", + "pull_request:10974", + "pull_request:11073" + ], + "soft_pairs": [ + "pull_request:10880|pull_request:10907", + "pull_request:10907|pull_request:11073", + "pull_request:10907|pull_request:10974", + "pull_request:10974|pull_request:11073" + ], + "prepared_review_unit_hash": "b371a2aa98d327f878bd6718c177fe5e80694385a446ce498b3248a1e27da426", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 9719, + "estimated_input_tokens": 2430, + "estimated_eval_tokens": 5116 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 9719, + "estimated_input_tokens": 2430, + "estimated_eval_tokens": 5116 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are thematically similar, but they fix different callback-tensor omissions in different pipeline families and code paths. I would not cluster them as duplicates; the closest representative is the SD ControlNet callback-input fix in PR 10907.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 10907 is the broadest and most central SD ControlNet callback-input fix here: it updates the base SD ControlNet pipeline and related img2img/inpaint variants to expose the missing `image` tensor to callbacks.", + "best_issue_reason": null, + "best_pr_reason": "PR 10907 is the best overall representative because it addresses a core ControlNet callback-input omission across multiple SD pipelines, while the others are narrower or for different model families.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10880", + "right": "pull_request:10907", + "accept": false, + "reason": "Different underlying fixes: 10880 targets SDXL ControlNet inpaint/img2img and adds `control_image`; 10907 targets SD ControlNet pipelines and adds `image`." + }, + { + "left": "pull_request:10907", + "right": "pull_request:11073", + "accept": false, + "reason": "11073 is a narrower SD inpaint-specific expansion adding `mask` and `masked_image_latents`; 10907 is about exposing `image` in SD ControlNet callbacks." + }, + { + "left": "pull_request:10907", + "right": "pull_request:10974", + "accept": false, + "reason": "Different pipeline families and tensors: 10907 is SD ControlNet with `image`, while 10974 is Flux ControlNet with `control_image`." + }, + { + "left": "pull_request:10974", + "right": "pull_request:11073", + "accept": false, + "reason": "These fix different code paths in different families: Flux ControlNet callback inputs vs. SD ControlNet inpaint callback inputs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 38, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10881-2", + "nodes": [ + "pull_request:10881", + "pull_request:10882" + ], + "soft_pairs": [ + "pull_request:10881|pull_request:10882" + ], + "prepared_review_unit_hash": "b3d994c5cb7533df9e93e5e826c875b7b337577e60d6df6bc039b4176409eb2d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4290, + "estimated_input_tokens": 1073, + "estimated_eval_tokens": 2402 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4290, + "estimated_input_tokens": 1073, + "estimated_eval_tokens": 2402 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are both refactors in diffusers, but they target different model families and unrelated code paths: Flux vs SD3. The soft similarity is superficial, so they should not be clustered as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR #10882 is the broader and more self-contained refactor, but it is not a duplicate of #10881; it is the better representative only by scope, not by sameness.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10881", + "right": "pull_request:10882", + "accept": false, + "reason": "Different subsystems and code paths: #10881 removes extra Flux transformer code, while #10882 updates SD3 docs and SD3-specific model/attention code. They are not the same concrete change and would not plausibly merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 39, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10939-2", + "nodes": [ + "pull_request:10939", + "pull_request:10946" + ], + "soft_pairs": [ + "pull_request:10939|pull_request:10946" + ], + "prepared_review_unit_hash": "4be7965c63c9be9a588f249406c08473caffc99f3889725ab035ef6cb14e1646", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4185, + "estimated_input_tokens": 1047, + "estimated_eval_tokens": 2350 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4185, + "estimated_input_tokens": 1047, + "estimated_eval_tokens": 2350 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs touch the same remote decode test file, but they are not the same change: #10939 just swaps a VAE decode endpoint, while #10946 introduces shared endpoint constants and adds a slow test case. They look related in sequence, not duplicate.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "#10946 is the better representative because it contains the broader, more explicit test change set: shared endpoint constants plus a new slow test; #10939 is only a single endpoint update.", + "best_issue_reason": null, + "best_pr_reason": "#10946 best captures the cluster\u2019s theme of remote VAE decode test maintenance and has the most complete, self-contained change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10939", + "right": "pull_request:10946", + "accept": false, + "reason": "Same file and same endpoint area, but different underlying changes: #10939 updates one endpoint, while #10946 refactors endpoint constants and adds a slow test. They could not plausibly be merged as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 40, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10944-2", + "nodes": [ + "pull_request:10944", + "pull_request:11622" + ], + "soft_pairs": [ + "pull_request:10944|pull_request:11622" + ], + "prepared_review_unit_hash": "4d484c463433a826f30799b99b4f376fba335a84dfaa317fef84a762fbe22a32", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3690, + "estimated_input_tokens": 923, + "estimated_eval_tokens": 2102 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3690, + "estimated_input_tokens": 923, + "estimated_eval_tokens": 2102 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are not duplicates: one fixes a CogView4 text-encoder device reference, while the other adds IP-Adapter loading support to Flux Fill. They touch different pipelines and different code paths.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR for deduplication; these are unrelated pipeline-specific changes with different fixes and intents.", + "best_issue_reason": null, + "best_pr_reason": "PR 11622 is the stronger representative of its own change because it has a concrete issue target and a substantive feature addition, but it is not a duplicate of PR 10944.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10944", + "right": "pull_request:11622", + "accept": false, + "reason": "Different pipelines and different bugs: CogView4 device placement vs. Flux Fill IP-Adapter loading. They do not look like the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 41, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10973-3", + "nodes": [ + "pull_request:10973", + "pull_request:11239", + "pull_request:11240" + ], + "soft_pairs": [ + "pull_request:10973|pull_request:11239", + "pull_request:10973|pull_request:11240", + "pull_request:11239|pull_request:11240" + ], + "prepared_review_unit_hash": "b6ca1bb692d9284286b462e04b460e813ca8d7b275e5d3a3ee3c4ea0f1ab2920", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7045, + "estimated_input_tokens": 1762, + "estimated_eval_tokens": 3780 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7045, + "estimated_input_tokens": 1762, + "estimated_eval_tokens": 3780 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs are tightly related variants of the same scheduler bug fix in duplicated DreamBooth training scripts. Two explicitly target issue #8384; the third applies the same corrected math to the LoRA variant. No separate issue is present.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #11239 is the best canonical representative: it has an explicit issue link to #8384, a clear title, and the core scheduler fix in the main DreamBooth script. The other two are near-identical sibling variants for related example files.", + "best_issue_reason": null, + "best_pr_reason": "PR #11239 is the most suitable global representative because it is the most directly traceable to the tracked bug and uses the primary DreamBooth code path. PR #11240 is essentially the SDXL counterpart, and #10973 is the LoRA counterpart, so neither is a better single anchor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:10973", + "right": "pull_request:11239", + "accept": true, + "reason": "Same underlying scheduler miscalculation in duplicated DreamBooth training code; both patch the same training-step/warmup math with essentially the same fix pattern and could plausibly be combined." + }, + { + "left": "pull_request:10973", + "right": "pull_request:11240", + "accept": true, + "reason": "Same bug and same fix pattern applied to closely related DreamBooth example scripts; this is a concrete code-path change, not just a shared topic." + }, + { + "left": "pull_request:11239", + "right": "pull_request:11240", + "accept": true, + "reason": "Both are near-identical fixes for the same LR scheduler counting bug in parallel DreamBooth variants and share the same explicit issue target #8384." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 42, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11079-2", + "nodes": [ + "pull_request:11079", + "pull_request:11587" + ], + "soft_pairs": [ + "pull_request:11079|pull_request:11587" + ], + "prepared_review_unit_hash": "2aba173a58c652712bbb424ee7d37ad07e6599639a224f1fa191c8cbdfa105e7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3985, + "estimated_input_tokens": 997, + "estimated_eval_tokens": 2250 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3985, + "estimated_input_tokens": 997, + "estimated_eval_tokens": 2250 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests fix different bugs in different parts of Diffusers: one changes PixArt one-step sampling behavior, the other fixes safetensors/variant loading compatibility. They are not duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11587 is the better cluster representative because it addresses a broader, reusable loading bug and includes tests; PR 11079 is a narrow model-specific inference fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 11587 is the strongest representative overall due to its general applicability and clearer bug scope; PR 11079 is more specialized to PixArt inference.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11079", + "right": "pull_request:11587", + "accept": false, + "reason": "Different underlying problems and code paths: PixArt scheduler output selection for num_inference_steps=1 vs. safetensors/variant file compatibility in pipeline loading. They do not look mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 43, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-977-2", + "nodes": [ + "pull_request:1108", + "pull_request:977" + ], + "soft_pairs": [ + "pull_request:1108|pull_request:977" + ], + "prepared_review_unit_hash": "73b7401ad5f7d1239ae2e954813f66d2055f599bf8e3d76f09449ff95639cb17", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4152, + "estimated_input_tokens": 1038, + "estimated_eval_tokens": 2332 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4152, + "estimated_input_tokens": 1038, + "estimated_eval_tokens": 2332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two closed PRs make the same small parameter-name fix in similar UNet downsampler code, but they touch different files and appear to be parallel/duplicated fixes rather than one mergeable change. The merged PR is the better canonical representative.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1108 is the merged, completed fix and is the best single representative of this parameter-naming correction in the downsampler code.", + "best_issue_reason": null, + "best_pr_reason": "PR #1108 is preferable as the canonical PR because it was merged and cleanly captures the change in one concrete code path/file set; PR #977 is the earlier, unmerged counterpart.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1108", + "right": "pull_request:977", + "accept": false, + "reason": "They make the same small fix but in different files (`unet_2d_blocks.py` vs `unet_blocks.py`), so they are analogous duplicated edits rather than the same concrete change in one mergeable code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 44, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11094-2", + "nodes": [ + "pull_request:11094", + "pull_request:11097" + ], + "soft_pairs": [ + "pull_request:11094|pull_request:11097" + ], + "prepared_review_unit_hash": "0a17c9d31c57a3401cbf3dbf940fec34b634b5711758a0d587eca976ce2a21de", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4206, + "estimated_input_tokens": 1052, + "estimated_eval_tokens": 2360 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4206, + "estimated_input_tokens": 1052, + "estimated_eval_tokens": 2360 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs both modify `group_offloading.py`, but they address different problems: one changes onload/offload tensor movement, while the other changes execution-order tracking for streamed group offloading. They are related in subsystem, not the same concrete fix, so this is not a duplicate pair.", + "confidence": 0.78, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11097 is the better canonical representative because it targets a specific bug ('using streams') and explicitly links to issue 11041, making its scope clearer than the broader 'improvements' PR 11094.", + "best_issue_reason": null, + "best_pr_reason": "PR 11097 is the best single PR here because it is a focused bug fix with explicit issue linkage; PR 11094 is a separate change to offload/onload behavior and is not the same underlying problem.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11094", + "right": "pull_request:11097", + "accept": false, + "reason": "Shared file and subsystem only. 11094 updates parameter/buffer device transfers during onload/offload, while 11097 adjusts execution-order callbacks for stream-based group offloading. They do not appear to fix the same code-path bug and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 45, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-111-2", + "nodes": [ + "pull_request:111", + "pull_request:112" + ], + "soft_pairs": [ + "pull_request:111|pull_request:112" + ], + "prepared_review_unit_hash": "ae8e72c07ebad60c2d63438e796c5b291bc54c69c81b8d610f36c5740b09f32d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5426, + "estimated_input_tokens": 1357, + "estimated_eval_tokens": 2970 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5426, + "estimated_input_tokens": 1357, + "estimated_eval_tokens": 2970 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are near-duplicates around the same PIL output refactor: both add `numpy_to_pil` in `pipeline_utils.py` and update the same pipeline classes/tests to return PIL images. PR 112 is the broader/expanded version, but it is the same underlying change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 111 is the cleaner canonical representative because it is the merged, concrete implementation of the PIL-output refactor across the shared pipeline code path.", + "best_issue_reason": null, + "best_pr_reason": "PR 111 is the strongest overall PR to keep: it is merged and addresses the same core pipeline-output conversion change as PR 112.", + "soft_edge_verdicts": [ + { + "left": "pull_request:111", + "right": "pull_request:112", + "accept": true, + "reason": "Both PRs implement the same code-path change: adding `numpy_to_pil` and converting diffusion pipeline outputs to PIL. The file overlap and diff preview indicate the same underlying refactor, with PR 112 just covering one additional pipeline class." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analyst\u2019s summary is grounded in the packet: both PRs share the same `pipeline_utils.py` `numpy_to_pil` addition and the same PIL-output refactor across overlapping pipeline/test files, while PR 112 only extends the change to `ddim`. The canonical-pr choice is also conservative because PR 111 is the merged implementation. The soft-edge acceptance is reasonable, and the changes do look mergeable as one bugfix PR." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 46, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11190-2", + "nodes": [ + "pull_request:11190", + "pull_request:11228" + ], + "soft_pairs": [ + "pull_request:11190|pull_request:11228" + ], + "prepared_review_unit_hash": "f3a71af22b5030daadaea3338e101ae9002395bc24a351408f30ba82836f6052", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3718, + "estimated_input_tokens": 930, + "estimated_eval_tokens": 2116 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3718, + "estimated_input_tokens": 930, + "estimated_eval_tokens": 2116 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two pull requests are present and they are not duplicates: one adjusts a quantization test's autocast context, the other changes SDXL img2img VAE dtype restoration logic. They touch different code paths and address different bugs.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #11190 is the better canonical PR because it is merged, narrowly scoped, and the diff cleanly addresses a specific autocast test issue.", + "best_issue_reason": null, + "best_pr_reason": "PR #11190 is the best overall PR candidate because it is merged and has the clearest, most self-contained fix. PR #11228 is unmerged and fixes a separate VAE dtype problem.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11190", + "right": "pull_request:11228", + "accept": false, + "reason": "Reject: these are different changes. #11190 fixes a test autocast context in bitsandbytes quantization tests, while #11228 changes VAE dtype restoration in the Stable Diffusion XL img2img pipeline. They do not appear to be the same underlying bug or mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 47, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1121-2", + "nodes": [ + "pull_request:1121", + "pull_request:1313" + ], + "soft_pairs": [ + "pull_request:1121|pull_request:1313" + ], + "prepared_review_unit_hash": "f73d91fdfa9888ccbe9a301edfdda474c7936c8e629ce7df501e9067b14453bb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4189, + "estimated_input_tokens": 1048, + "estimated_eval_tokens": 2352 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4189, + "estimated_input_tokens": 1048, + "estimated_eval_tokens": 2352 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about v-prediction support, but they touch different code paths and are not duplicates: one updates DDPM internals, the other updates DDIM plus adds an example.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1313 is the more representative v-prediction change: it explicitly targets DDIM, updates the scheduler, and includes an example demonstrating the feature.", + "best_issue_reason": null, + "best_pr_reason": "PR #1313 is the best single PR for this cluster because it is the clearest end-to-end implementation of v-prediction behavior, not just a low-level helper change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1121", + "right": "pull_request:1313", + "accept": false, + "reason": "Both discuss v-prediction, but #1121 changes DDPM scheduler logic while #1313 changes DDIM and adds an example. They are related feature work, not the same concrete bug or mergeable duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 48, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11219-15", + "nodes": [ + "pull_request:11219", + "pull_request:11245", + "pull_request:11263", + "pull_request:11264", + "pull_request:11265", + "pull_request:11266", + "pull_request:11310", + "pull_request:11404", + "pull_request:11444", + "pull_request:11458", + "pull_request:11459", + "pull_request:12350", + "pull_request:12842", + "pull_request:13270", + "pull_request:13535" + ], + "soft_pairs": [ + "pull_request:11219|pull_request:12350", + "pull_request:11219|pull_request:11458", + "pull_request:11245|pull_request:11263", + "pull_request:11245|pull_request:11264", + "pull_request:11245|pull_request:11266", + "pull_request:11245|pull_request:11310", + "pull_request:11264|pull_request:11266", + "pull_request:11264|pull_request:11310", + "pull_request:11264|pull_request:11458", + "pull_request:11264|pull_request:11404", + "pull_request:11265|pull_request:11459", + "pull_request:11265|pull_request:12350", + "pull_request:11266|pull_request:11310", + "pull_request:11266|pull_request:11458", + "pull_request:11266|pull_request:11444", + "pull_request:11444|pull_request:11458", + "pull_request:11458|pull_request:12350", + "pull_request:11264|pull_request:12350", + "pull_request:11266|pull_request:12350", + "pull_request:12350|pull_request:12842", + "pull_request:12842|pull_request:13270", + "pull_request:13270|pull_request:13535" + ], + "prepared_review_unit_hash": "f0fb87b45db6cf21431e38131db8569f4857eb0c1efc04897ce7ac8319630b63", + "component_budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 22, + "serialized_chars": 35734, + "estimated_input_tokens": 8934, + "estimated_eval_tokens": 18124 + }, + "budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 22, + "serialized_chars": 35734, + "estimated_input_tokens": 8934, + "estimated_eval_tokens": 18124 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is not a true duplicate set: it mixes many unrelated XPU/A100 test-porting PRs across different pipelines plus a few real code fixes. They share a broad accelerator-testing theme, but not the same underlying bug or change.", + "confidence": 0.22, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR cleanly represents the whole cluster because the items are mostly unrelated. If one must be chosen, PR #11458 is the broadest accelerator-portability test fix, but it still only covers one slice of the cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR #11458 is the best broad representative of the shared theme because it updates test setup/helpers for accelerator-agnostic execution and spans multiple test cases, but it is still not a duplicate of the other PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11219", + "right": "pull_request:12350", + "accept": false, + "reason": "Both are XPU-related test fixes, but they target different components and failures: mixed int8 autocast vs Marigold depth tests." + }, + { + "left": "pull_request:11219", + "right": "pull_request:11458", + "accept": false, + "reason": "Different test failures and code paths; one is a bnb autocast test tweak, the other is Safe Diffusion test setup for XPU/A100." + }, + { + "left": "pull_request:11245", + "right": "pull_request:11263", + "accept": false, + "reason": "Both adjust test expectations for accelerator runs, but Flux Redux inference and LoRA finetuning are unrelated pipelines with different failures." + }, + { + "left": "pull_request:11245", + "right": "pull_request:11264", + "accept": false, + "reason": "Different pipeline tests and outputs; no shared bug or merged change beyond broad XPU test flakiness." + }, + { + "left": "pull_request:11245", + "right": "pull_request:11266", + "accept": false, + "reason": "Flux Redux inference and IP-Adapter masking failures are separate pipeline behaviors and should not be merged as one fix." + }, + { + "left": "pull_request:11245", + "right": "pull_request:11310", + "accept": false, + "reason": "These are distinct pipeline test failures in different models; only the accelerator target is shared." + }, + { + "left": "pull_request:11264", + "right": "pull_request:11266", + "accept": false, + "reason": "Stable Diffusion inpaint vs IP-Adapter multiple masks are different failing tests and different code paths." + }, + { + "left": "pull_request:11264", + "right": "pull_request:11310", + "accept": false, + "reason": "Both touch image-generation tests, but the failing assertions and pipeline implementations are unrelated." + }, + { + "left": "pull_request:11264", + "right": "pull_request:11404", + "accept": false, + "reason": "One is a test expectation update; the other includes GGUF loader/quantizer code changes plus tests, so they are not the same fix." + }, + { + "left": "pull_request:11264", + "right": "pull_request:11458", + "accept": false, + "reason": "Different Stable Diffusion variants and different failures; both are XPU test adjustments, not one underlying bug." + }, + { + "left": "pull_request:11264", + "right": "pull_request:12350", + "accept": false, + "reason": "Separate pipelines and assertions; no evidence of the same concrete bug." + }, + { + "left": "pull_request:11265", + "right": "pull_request:11459", + "accept": false, + "reason": "i2vgen-xl tuple output tolerance and semantic/panorama accelerator setup are unrelated test fixes." + }, + { + "left": "pull_request:11265", + "right": "pull_request:12350", + "accept": false, + "reason": "Different models, different test names, different behavior changes." + }, + { + "left": "pull_request:11266", + "right": "pull_request:11310", + "accept": false, + "reason": "IP-Adapter mask outputs and SDXL K-Diffusion sigmas are unrelated test failures." + }, + { + "left": "pull_request:11266", + "right": "pull_request:11458", + "accept": false, + "reason": "Both are XPU test fixes, but they target different pipelines and output slices." + }, + { + "left": "pull_request:11266", + "right": "pull_request:11444", + "accept": false, + "reason": "IP-Adapter masking vs Unidiffuser device/offload tests are separate issues." + }, + { + "left": "pull_request:11266", + "right": "pull_request:12350", + "accept": false, + "reason": "Different model families and different failing expectations." + }, + { + "left": "pull_request:11444", + "right": "pull_request:11458", + "accept": false, + "reason": "Unidiffuser test portability and Safe Diffusion test portability are different test suites with no shared concrete bug." + }, + { + "left": "pull_request:11458", + "right": "pull_request:12350", + "accept": false, + "reason": "Safe Diffusion XPU/A100 test changes do not overlap with Marigold depth test changes." + }, + { + "left": "pull_request:12350", + "right": "pull_request:12842", + "accept": false, + "reason": "Marigold depth and PixArt test expectations are unrelated pipeline tests." + }, + { + "left": "pull_request:12842", + "right": "pull_request:13270", + "accept": false, + "reason": "PixArt test expectation updates and parallelism infrastructure changes are not the same underlying issue." + }, + { + "left": "pull_request:13270", + "right": "pull_request:13535", + "accept": false, + "reason": "Parallelism backend selection and PEFT hotswap compilation behavior are distinct code paths and fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 49, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11405-3", + "nodes": [ + "pull_request:11405", + "pull_request:11406", + "pull_request:11759" + ], + "soft_pairs": [ + "pull_request:11405|pull_request:11406", + "pull_request:11405|pull_request:11759" + ], + "prepared_review_unit_hash": "28508053a86849bc8a8981a0e4801ad2366bebffdc4db2a6765842ed88ff29cd", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6755, + "estimated_input_tokens": 1689, + "estimated_eval_tokens": 3634 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6755, + "estimated_input_tokens": 1689, + "estimated_eval_tokens": 3634 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related XPU CI/test-enablement PRs, but they address different test failures and code paths. PR 11759 is the broadest representative; the other two are narrower, distinct fixes.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11759 is the best canonical representative because it spans the widest set of XPU CI test adjustments across multiple files and subsumes the general theme of fixing test stability on XPU.", + "best_issue_reason": null, + "best_pr_reason": "PR 11759 is the strongest standalone choice for the cluster since it is the most comprehensive and directly aimed at getting multiple CI test cases passing on XPU.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11405", + "right": "pull_request:11406", + "accept": false, + "reason": "Both are XPU-related test enablement changes, but they fix different things: one updates test decorators for group offload/quanto, the other adds backend synchronization and updates layerwise-casting memory tests. They do not look like the same concrete fix." + }, + { + "left": "pull_request:11405", + "right": "pull_request:11759", + "accept": false, + "reason": "They overlap only at a broad XPU test-maintenance level and share one test file, but the concrete changes differ: 11405 targets group offload and quanto tests, while 11759 tweaks several unrelated pipeline test expectations and test setup. Not duplicate-level." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 50, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11445-15", + "nodes": [ + "pull_request:11445", + "pull_request:11461", + "pull_request:11503", + "pull_request:11507", + "pull_request:11523", + "pull_request:11526", + "pull_request:11527", + "pull_request:11620", + "pull_request:11654", + "pull_request:11988", + "pull_request:12158", + "pull_request:12214", + "pull_request:12345", + "pull_request:12408", + "pull_request:12460" + ], + "soft_pairs": [ + "pull_request:11445|pull_request:11503", + "pull_request:11445|pull_request:11523", + "pull_request:11445|pull_request:11527", + "pull_request:11445|pull_request:11654", + "pull_request:11445|pull_request:12345", + "pull_request:11461|pull_request:12158", + "pull_request:11445|pull_request:11461", + "pull_request:11461|pull_request:11503", + "pull_request:11461|pull_request:11507", + "pull_request:11461|pull_request:11526", + "pull_request:11503|pull_request:11523", + "pull_request:11503|pull_request:11527", + "pull_request:11503|pull_request:11654", + "pull_request:11503|pull_request:12345", + "pull_request:11445|pull_request:11507", + "pull_request:11503|pull_request:11507", + "pull_request:11507|pull_request:11523", + "pull_request:11507|pull_request:11526", + "pull_request:11507|pull_request:11527", + "pull_request:11523|pull_request:11654", + "pull_request:11523|pull_request:11526", + "pull_request:11523|pull_request:12214", + "pull_request:11526|pull_request:12214", + "pull_request:11526|pull_request:11527", + "pull_request:11445|pull_request:11526", + "pull_request:11503|pull_request:11526", + "pull_request:11527|pull_request:11654", + "pull_request:11527|pull_request:11988", + "pull_request:11620|pull_request:12214", + "pull_request:11620|pull_request:12158", + "pull_request:11445|pull_request:11620", + "pull_request:11503|pull_request:11620", + "pull_request:11507|pull_request:11620", + "pull_request:11654|pull_request:12345", + "pull_request:11654|pull_request:11988", + "pull_request:11988|pull_request:12214", + "pull_request:11445|pull_request:11988", + "pull_request:11503|pull_request:11988", + "pull_request:11445|pull_request:12158", + "pull_request:11503|pull_request:12158", + "pull_request:11507|pull_request:12158", + "pull_request:11527|pull_request:12214", + "pull_request:11654|pull_request:12214", + "pull_request:11527|pull_request:12345", + "pull_request:11461|pull_request:12345", + "pull_request:11445|pull_request:12408", + "pull_request:11461|pull_request:12408", + "pull_request:11503|pull_request:12408", + "pull_request:11507|pull_request:12408", + "pull_request:11526|pull_request:12408", + "pull_request:11445|pull_request:12460", + "pull_request:11461|pull_request:12460", + "pull_request:11503|pull_request:12460", + "pull_request:11507|pull_request:12460", + "pull_request:11527|pull_request:12460" + ], + "prepared_review_unit_hash": "f9ac3942fe321462be4d7111c3944a8c7cb3ca7035b85a382d2cfba0daa172ef", + "component_budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 55, + "serialized_chars": 51479, + "estimated_input_tokens": 12870, + "estimated_eval_tokens": 25996 + }, + "budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 55, + "serialized_chars": 51479, + "estimated_input_tokens": 12870, + "estimated_eval_tokens": 25996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related XPU/accelerator enablement and regression-fix PRs, but they target different tests, utilities, or code paths. I do not see true duplicate PRs here; they are best treated as separate maintenance changes within the same broad migration theme.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11654 is the broadest and most central change: it updates shared quantization/testing logic and touches many XPU-sensitive tests, making it the best representative of the cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 11654 best captures the cluster\u2019s main theme because it includes a shared code-path fix plus many downstream test updates, unlike the narrower test-only XPU enablement PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11445", + "right": "pull_request:11503", + "accept": false, + "reason": "Both are XPU-oriented test enablement, but they touch different integration suites and different files; not the same underlying fix." + }, + { + "left": "pull_request:11445", + "right": "pull_request:11523", + "accept": false, + "reason": "Marigold intrinsics XPU support vs DiT XPU support are separate test targets and code paths." + }, + { + "left": "pull_request:11445", + "right": "pull_request:11527", + "accept": false, + "reason": "Both are test migration work, but one is Marigold-specific while the other spans pipeline utilities and multiple other tests." + }, + { + "left": "pull_request:11445", + "right": "pull_request:11654", + "accept": false, + "reason": "Marigold test-only XPU enablement is not the same change as shared quantization/config and multi-test XPU support." + }, + { + "left": "pull_request:11445", + "right": "pull_request:12345", + "accept": false, + "reason": "Different subsystems: Marigold integration tests vs component-manager/offloading and LoRA-related XPU work." + }, + { + "left": "pull_request:11461", + "right": "pull_request:12158", + "accept": false, + "reason": "Group offloading stream/device handling and GGUF XPU behavior are unrelated code paths." + }, + { + "left": "pull_request:11445", + "right": "pull_request:11461", + "accept": false, + "reason": "Both mention XPU, but one is Marigold tests and the other is group offloading infrastructure plus several tests." + }, + { + "left": "pull_request:11461", + "right": "pull_request:11503", + "accept": false, + "reason": "Group offloading changes are not the same as broad test-suite XPU enablement across multiple pipelines." + }, + { + "left": "pull_request:11461", + "right": "pull_request:11507", + "accept": false, + "reason": "Group offloading code changes vs environment-printing updates are different underlying fixes." + }, + { + "left": "pull_request:11461", + "right": "pull_request:11526", + "accept": false, + "reason": "Offloading infrastructure and several pipeline XPU test updates are related only at the accelerator level, not the same bug." + }, + { + "left": "pull_request:11503", + "right": "pull_request:11523", + "accept": false, + "reason": "These enable different pipeline suites on XPU; they are separate test migrations." + }, + { + "left": "pull_request:11503", + "right": "pull_request:11527", + "accept": false, + "reason": "Both are broad XPU test changes, but 11527 also changes shared loading utilities and targets different tests." + }, + { + "left": "pull_request:11503", + "right": "pull_request:11654", + "accept": false, + "reason": "Broad XPU test enablement is similar in theme, but 11654 also changes shared quantization logic and a different set of tests." + }, + { + "left": "pull_request:11503", + "right": "pull_request:12345", + "accept": false, + "reason": "Different scopes: pipeline/test enablement versus component-manager offloading and LoRA-related work." + }, + { + "left": "pull_request:11507", + "right": "pull_request:11523", + "accept": false, + "reason": "Environment-reporting support for XPU is unrelated to DiT pipeline test enablement." + }, + { + "left": "pull_request:11507", + "right": "pull_request:11526", + "accept": false, + "reason": "Print-env changes and pipeline integration test migrations do not share a concrete fix path." + }, + { + "left": "pull_request:11507", + "right": "pull_request:11527", + "accept": false, + "reason": "Different artifacts and code paths: env reporting vs test utility/loading and pipeline updates." + }, + { + "left": "pull_request:11507", + "right": "pull_request:11620", + "accept": false, + "reason": "print_env XPU support is not the same issue as group_offloading test and helper changes." + }, + { + "left": "pull_request:11523", + "right": "pull_request:11526", + "accept": false, + "reason": "DiT integration tests and several other pipeline tests are separate suites, not duplicates." + }, + { + "left": "pull_request:11523", + "right": "pull_request:11654", + "accept": false, + "reason": "DiT XPU test enablement is distinct from torchao/quantization config changes and broader test updates." + }, + { + "left": "pull_request:11523", + "right": "pull_request:12214", + "accept": false, + "reason": "Different pipeline targets: DiT vs Bria." + }, + { + "left": "pull_request:11526", + "right": "pull_request:12214", + "accept": false, + "reason": "Both are pipeline XPU enablement PRs, but they cover different model families and test files." + }, + { + "left": "pull_request:11526", + "right": "pull_request:11527", + "accept": false, + "reason": "Related XPU migration work, but 11527 includes shared helper changes and different test groups." + }, + { + "left": "pull_request:11445", + "right": "pull_request:11526", + "accept": false, + "reason": "Marigold-specific test enablement is separate from the broader set of pipeline tests in 11526." + }, + { + "left": "pull_request:11503", + "right": "pull_request:11526", + "accept": false, + "reason": "Two different bundles of XPU test cases; same theme, not the same concrete bug." + }, + { + "left": "pull_request:11527", + "right": "pull_request:11654", + "accept": false, + "reason": "Both touch test utilities, but 11527 focuses on load_pt/test migrations while 11654 also changes quantization capability logic." + }, + { + "left": "pull_request:11527", + "right": "pull_request:11988", + "accept": false, + "reason": "Pipeline test utility changes vs quantization compile tests are different scopes and failures." + }, + { + "left": "pull_request:11620", + "right": "pull_request:12214", + "accept": false, + "reason": "Group offloading and Bria integration tests are unrelated code paths." + }, + { + "left": "pull_request:11620", + "right": "pull_request:12158", + "accept": false, + "reason": "Group offloading tests and GGUF inference tests are not the same underlying issue." + }, + { + "left": "pull_request:11445", + "right": "pull_request:11620", + "accept": false, + "reason": "Marigold XPU test enablement and group offloading XPU support are separate areas." + }, + { + "left": "pull_request:11503", + "right": "pull_request:11620", + "accept": false, + "reason": "Broad XPU test enablement vs group offloading-specific XPU changes." + }, + { + "left": "pull_request:11507", + "right": "pull_request:11620", + "accept": false, + "reason": "Environment printing changes are unrelated to group offloading fixes." + }, + { + "left": "pull_request:11654", + "right": "pull_request:12345", + "accept": false, + "reason": "Quantization config / torchao support is a different code path from components-manager offloading and LoRA handling." + }, + { + "left": "pull_request:11654", + "right": "pull_request:11988", + "accept": false, + "reason": "Torchao quantization config/test updates are distinct from quant compile tests." + }, + { + "left": "pull_request:11988", + "right": "pull_request:12214", + "accept": false, + "reason": "Quant compile tests and Bria integration tests are different model/test paths." + }, + { + "left": "pull_request:11445", + "right": "pull_request:11988", + "accept": false, + "reason": "Marigold XPU test enablement is unrelated to quant compile XPU support." + }, + { + "left": "pull_request:11503", + "right": "pull_request:11988", + "accept": false, + "reason": "Different suites and failures: multi-pipeline test enablement vs quant compile tests." + }, + { + "left": "pull_request:11445", + "right": "pull_request:12158", + "accept": false, + "reason": "Marigold integration tests and GGUF XPU test behavior are unrelated." + }, + { + "left": "pull_request:11503", + "right": "pull_request:12158", + "accept": false, + "reason": "General XPU test enablement is not the same as GGUF-specific inference/tolerance updates." + }, + { + "left": "pull_request:11507", + "right": "pull_request:12158", + "accept": false, + "reason": "print_env XPU support and GGUF model tests do not share a concrete fix." + }, + { + "left": "pull_request:11527", + "right": "pull_request:12214", + "accept": false, + "reason": "Shared test utility migration vs Bria-specific XPU enablement are different changes." + }, + { + "left": "pull_request:11654", + "right": "pull_request:12214", + "accept": false, + "reason": "Torchao/quantization and Bria pipeline support are unrelated code paths." + }, + { + "left": "pull_request:11527", + "right": "pull_request:12345", + "accept": false, + "reason": "Load/helper updates for test migration are not the same as component-manager offloading changes." + }, + { + "left": "pull_request:11461", + "right": "pull_request:12345", + "accept": false, + "reason": "Group offloading stream handling and components-manager offloading are different implementation paths." + }, + { + "left": "pull_request:11445", + "right": "pull_request:12408", + "accept": false, + "reason": "Marigold XPU test enablement is unrelated to the later PyTorch-regression fixes in 12408." + }, + { + "left": "pull_request:11461", + "right": "pull_request:12408", + "accept": false, + "reason": "Group offloading infrastructure and latest-PyTorch test fixes are distinct issues." + }, + { + "left": "pull_request:11503", + "right": "pull_request:12408", + "accept": false, + "reason": "Broad XPU test enablement is not the same as a targeted regression fix for a newer PyTorch version." + }, + { + "left": "pull_request:11507", + "right": "pull_request:12408", + "accept": false, + "reason": "Environment-reporting support and test-failure fixes from latest PyTorch are unrelated." + }, + { + "left": "pull_request:11526", + "right": "pull_request:12408", + "accept": false, + "reason": "Different pipeline test bundles and different failure mode: generic XPU enablement vs PyTorch-update regressions." + }, + { + "left": "pull_request:11445", + "right": "pull_request:12460", + "accept": false, + "reason": "Marigold XPU test support and a tolerance adjustment for Kandinsky controlnet are separate issues." + }, + { + "left": "pull_request:11461", + "right": "pull_request:12460", + "accept": false, + "reason": "Group offloading work is unrelated to a single-test tolerance loosening." + }, + { + "left": "pull_request:11503", + "right": "pull_request:12460", + "accept": false, + "reason": "Broad XPU test enablement is not the same as a precision-tolerance change for one test." + }, + { + "left": "pull_request:11507", + "right": "pull_request:12460", + "accept": false, + "reason": "print_env XPU support has nothing to do with Kandinsky controlnet thresholds." + }, + { + "left": "pull_request:11527", + "right": "pull_request:12460", + "accept": false, + "reason": "Different targets: pipeline utility/test migration vs a single test's XPU tolerance tweak." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 51, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11567-2", + "nodes": [ + "pull_request:11567", + "pull_request:12583" + ], + "soft_pairs": [ + "pull_request:11567|pull_request:12583" + ], + "prepared_review_unit_hash": "ec21dc9e8245d3779bc6cc6d893feb4eb415dd3128c68f169dcdf9ec397c605d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5153, + "estimated_input_tokens": 1289, + "estimated_eval_tokens": 2834 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5153, + "estimated_input_tokens": 1289, + "estimated_eval_tokens": 2834 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are broad link-migration cleanups from arXiv URLs to Hugging Face Papers, but they appear to cover different file sets and scopes rather than one identical change.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12583 is the better representative because its title is explicit and its scope spans examples plus source files, making it the broader 'ArXiv -> HF Papers' cleanup.", + "best_issue_reason": null, + "best_pr_reason": "PR #12583 is the strongest single representative of this duplicate-like family of paper-link migrations; #11567 is similar but more narrowly centered on docs pages.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11567", + "right": "pull_request:12583", + "accept": false, + "reason": "Same general URL migration theme, but the PRs touch different areas and do not look like one concrete change that could be merged as a single PR without being overbroad." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 52, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11593-2", + "nodes": [ + "pull_request:11593", + "pull_request:12057" + ], + "soft_pairs": [ + "pull_request:11593|pull_request:12057" + ], + "prepared_review_unit_hash": "7b473d22a0a3a7a24ad80ce22a14163114bd115dad18ab9bc33186e5e1336661", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5712, + "estimated_input_tokens": 1428, + "estimated_eval_tokens": 3112 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5712, + "estimated_input_tokens": 1428, + "estimated_eval_tokens": 3112 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are unrelated: one refactors First Block Cache internals and docs, while the other adds/adjusts QwenImage tests and minor refactors. They do not fix the same concrete code-path problem.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:12057 is the stronger canonical PR because it is merged, feature-specific, and self-contained around QwenImage; it also has tests and docs for the same change set.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:12057 is the best representative of this cluster since it is the completed merged change, whereas pull_request:11593 is an unrelated unmerged cache refactor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11593", + "right": "pull_request:12057", + "accept": false, + "reason": "Different subsystems and purposes: First Block Cache refactor vs QwenImage tests/minor refactor. No shared concrete bug or mergeable unified fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 53, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11665-3", + "nodes": [ + "pull_request:11665", + "pull_request:11666", + "pull_request:11667" + ], + "soft_pairs": [ + "pull_request:11665|pull_request:11667", + "pull_request:11666|pull_request:11667" + ], + "prepared_review_unit_hash": "e4b65cb896f2a3f19a7c8121e9511b211d56ebeb6a5bb666ed62c4cb24e2d5b0", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6473, + "estimated_input_tokens": 1619, + "estimated_eval_tokens": 3494 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6473, + "estimated_input_tokens": 1619, + "estimated_eval_tokens": 3494 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three pull requests are standalone performance tweaks in different subsystems (Wan autoencoder cache clearing, BLIP image postprocessing, Kandinsky3 group norm). They are not duplicates and should not be clustered together.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #11665 is the strongest canonical representative only because it is merged and has the clearest, self-contained optimization in the set; the other PRs are unrelated changes in different code paths.", + "best_issue_reason": null, + "best_pr_reason": "PR #11665 is the best representative of the group: it is merged, concrete, and the most established example of the performance-optimization pattern in this set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11665", + "right": "pull_request:11667", + "accept": false, + "reason": "Both are speedup PRs, but they modify different models and different execution paths: AutoencoderKLWan.clear_cache versus Kandinsky3ConditionalGroupNorm.forward. They are not the same bug/change and would not plausibly be merged into one PR." + }, + { + "left": "pull_request:11666", + "right": "pull_request:11667", + "accept": false, + "reason": "These target unrelated code paths in different components: BLIP image postprocessing versus Kandinsky3 group norm broadcasting. Similar performance intent is too broad to treat as a duplicate." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 54, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-76-2", + "nodes": [ + "pull_request:117", + "pull_request:76" + ], + "soft_pairs": [ + "pull_request:117|pull_request:76" + ], + "prepared_review_unit_hash": "97659fe59dde236127f9e72ca83f58fb1d4aaeb917099f20ddedb650eb299ff3", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4259, + "estimated_input_tokens": 1065, + "estimated_eval_tokens": 2386 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4259, + "estimated_input_tokens": 1065, + "estimated_eval_tokens": 2386 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are not duplicates: PR #117 only updates README quickstart documentation, while PR #76 is a code refactor across UNet model files. They do not appear to fix the same underlying bug or implement the same change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #76 is the better canonical PR for this cluster because it represents the substantive code change; PR #117 is just documentation and is not representative of the same change set.", + "best_issue_reason": null, + "best_pr_reason": "PR #76 is the best overall representative among the PRs because it changes core model code, whereas PR #117 is a README-only update and should not be clustered with it.", + "soft_edge_verdicts": [ + { + "left": "pull_request:117", + "right": "pull_request:76", + "accept": false, + "reason": "README quickstart documentation update vs UNet model code refactor; different artifacts and no shared concrete bug/change path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 55, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11723-2", + "nodes": [ + "pull_request:11723", + "pull_request:11780" + ], + "soft_pairs": [ + "pull_request:11723|pull_request:11780" + ], + "prepared_review_unit_hash": "e1bb1e6b9dfafaea8e9d0d123237ace4b4d11351d7defdd283f3c1a7adb33541", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4370, + "estimated_input_tokens": 1093, + "estimated_eval_tokens": 2442 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4370, + "estimated_input_tokens": 1093, + "estimated_eval_tokens": 2442 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged pull requests that both add LoRA alpha/rank guidance to example READMEs, but they target different example flows and different files; this is topical similarity, not the same underlying fix/change.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #11723 is the broader representative: it updates the advanced diffusion training docs and associated training/test files, while #11780 is a narrower README-only Sana documentation tweak.", + "best_issue_reason": null, + "best_pr_reason": "PR #11723 is the best single representative because it covers the more complete LoRA metadata update across example docs and training code, not just one README.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11723", + "right": "pull_request:11780", + "accept": false, + "reason": "Both PRs discuss LoRA alpha/rank, but they change different README files for different example pipelines and do not fix the same concrete code-path problem. This is related documentation work, not a duplicate change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 56, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11822-2", + "nodes": [ + "pull_request:11822", + "pull_request:11834" + ], + "soft_pairs": [ + "pull_request:11822|pull_request:11834" + ], + "prepared_review_unit_hash": "847ac0822113c8667c098cd296031212db0c0a5c4b620259547779bfc7eeca0c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4224, + "estimated_input_tokens": 1056, + "estimated_eval_tokens": 2368 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4224, + "estimated_input_tokens": 1056, + "estimated_eval_tokens": 2368 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two unrelated merged LoRA-related PRs: one fixes unloading state for PEFT/Lora removal, the other optimizes LoRA scale expansion in the UNet loader. They do not look like the same underlying bug or change, so this is not a duplicate cluster.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:11822", + "right": "pull_request:11834", + "accept": false, + "reason": "Different code paths and goals: 11822 changes LoRA unload/PEFT cleanup state in peft.py, while 11834 is a speed optimization in unet_loader_utils for expanding LoRA scales. They are not the same fix and would not be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 57, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11837-2", + "nodes": [ + "pull_request:11837", + "pull_request:11952" + ], + "soft_pairs": [ + "pull_request:11837|pull_request:11952" + ], + "prepared_review_unit_hash": "91073d682d5d94511af18f5aaf2804c799220325cf7d75961eea0d4f80093a77", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4198, + "estimated_input_tokens": 1050, + "estimated_eval_tokens": 2356 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4198, + "estimated_input_tokens": 1050, + "estimated_eval_tokens": 2356 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged docs PRs both touch `docs/source/en/optimization/fp16.md`, but they are separate documentation updates: one rewrites/expands the regional compilation section, the other adds a tip linking a blog post. They are not the same underlying change and should not be deduplicated.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 11837 is the better canonical representative because it makes the larger, more substantive documentation update to the optimization guide, while 11952 is a smaller follow-up link insertion.", + "best_issue_reason": null, + "best_pr_reason": "PR 11837 is the best single representative for the cluster since it covers the broader content change; 11952 is a narrow, unrelated docs addition in the same file.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11837", + "right": "pull_request:11952", + "accept": false, + "reason": "Both are documentation-only PRs in the same file, but they address different sections and different goals: regional compilation wording vs. adding a diffusion performance blog link. They do not look like one merged change or the same underlying bug/fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 58, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11860-4", + "nodes": [ + "pull_request:11860", + "pull_request:11862", + "pull_request:11868", + "pull_request:11893" + ], + "soft_pairs": [ + "pull_request:11860|pull_request:11862", + "pull_request:11860|pull_request:11868", + "pull_request:11860|pull_request:11893", + "pull_request:11862|pull_request:11868", + "pull_request:11862|pull_request:11893", + "pull_request:11868|pull_request:11893" + ], + "prepared_review_unit_hash": "82a70eac423d605b2d65c9dae91c93973442b3845ef7c9ccc6bd2c405f8030a7", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 12357, + "estimated_input_tokens": 3090, + "estimated_eval_tokens": 6436 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 12357, + "estimated_input_tokens": 3090, + "estimated_eval_tokens": 6436 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a set of related Diffusers pull requests around modular guider/cache work plus a later branch-merge PR. They share files because they evolve the same subsystem, but they are not the same concrete change set.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #11893 is the broad umbrella integration PR that merges the modular diffusers branch back into main, so it best represents the cluster\u2019s endpoint and subsumes the other modular changes at a high level.", + "best_issue_reason": null, + "best_pr_reason": "PR #11893 is the most useful representative because it is the largest integration point and touches the widest set of affected files; the other PRs are narrower feature/infrastructure changes within that broader merge stream.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11860", + "right": "pull_request:11862", + "accept": false, + "reason": "Related subsystem, but different changes: PAG guider registration vs. adding ConfigMixin support across guiders. They are not the same underlying bug or fix and would not naturally be one PR." + }, + { + "left": "pull_request:11860", + "right": "pull_request:11868", + "accept": false, + "reason": "11860 adds a new guider, while 11868 changes config serialization behavior. They overlap in guider files but address different code-paths and should remain separate." + }, + { + "left": "pull_request:11860", + "right": "pull_request:11893", + "accept": false, + "reason": "11893 is an umbrella merge-back PR, not the same concrete feature as 11860. Shared filenames alone are insufficient to treat them as duplicates." + }, + { + "left": "pull_request:11862", + "right": "pull_request:11868", + "accept": false, + "reason": "Both are modular-guider infrastructure work, but one adds config registration and the other fixes serialization of custom config objects. Different underlying changes." + }, + { + "left": "pull_request:11862", + "right": "pull_request:11893", + "accept": false, + "reason": "11893 is a broad integration merge, not the same change as the ConfigMixin PR. They may overlap in touched files, but not in a single concrete bug or change." + }, + { + "left": "pull_request:11868", + "right": "pull_request:11893", + "accept": false, + "reason": "11868 implements config serialization support, whereas 11893 is a merge of modular diffusers with main. Not the same fix/change even though they share some files." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 59, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-11887-2", + "nodes": [ + "pull_request:11887", + "pull_request:11926" + ], + "soft_pairs": [ + "pull_request:11887|pull_request:11926" + ], + "prepared_review_unit_hash": "8ce57cfb3def4ec5fed75d49f2355cc115fe86b6b519c2037aba9490d006d251", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3452, + "estimated_input_tokens": 863, + "estimated_eval_tokens": 1982 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3452, + "estimated_input_tokens": 863, + "estimated_eval_tokens": 1982 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are both CI-speedup PRs, but they target different concrete changes: #11887 only raises GPU job parallelism, while #11926 makes broader PR CI edits across CPU/GPU workflows and tests. They are related in theme, not duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "#11887 is the best canonical PR: it is merged, focused on one specific CI bottleneck, and its scope is clearer than the broader WIP PR.", + "best_issue_reason": null, + "best_pr_reason": "#11887 is the strongest representative of the cluster because it is a completed, narrow change to a single workflow and best captures the underlying CI speedup theme without conflating unrelated edits.", + "soft_edge_verdicts": [ + { + "left": "pull_request:11887", + "right": "pull_request:11926", + "accept": false, + "reason": "Both touch CI workflows, but #11887 only increases GPU job concurrency while #11926 changes multiple workflows and test execution settings; they do not fix the same underlying bug/change and would not plausibly merge as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 60, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-928-2", + "nodes": [ + "pull_request:1189", + "pull_request:928" + ], + "soft_pairs": [ + "pull_request:1189|pull_request:928" + ], + "prepared_review_unit_hash": "58dac52bdf63081234b00c047d701fe83d4fc8ba2863196df9d0bd2945ca0544", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4109, + "estimated_input_tokens": 1028, + "estimated_eval_tokens": 2312 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4109, + "estimated_input_tokens": 1028, + "estimated_eval_tokens": 2312 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are not duplicates. #1189 fixes a concrete runtime/device bug in DDIMScheduler noise creation, while #928 only adjusts tests and pipeline device placement in a test helper.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1189 is the canonical PR: it directly changes the DDIM scheduler code path to fix noise-device handling and MPS reproducibility, which is the substantive bug fix in this cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR #1189 is the best representative because it addresses the underlying implementation bug; #928 is test-only and not the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1189", + "right": "pull_request:928", + "accept": false, + "reason": "Reject: #1189 fixes DDIM noise device logic in src/diffusers/schedulers/scheduling_ddim.py, while #928 only updates tests and moves test pipelines to the device. They do not fix the same concrete code-path problem and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 61, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12023-2", + "nodes": [ + "pull_request:12023", + "pull_request:12314" + ], + "soft_pairs": [ + "pull_request:12023|pull_request:12314" + ], + "prepared_review_unit_hash": "0ccf95438c59143f1364e7c140c3f0a5e6237f269edb5ea9987cd18749bf9dc9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3449, + "estimated_input_tokens": 863, + "estimated_eval_tokens": 1982 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3449, + "estimated_input_tokens": 863, + "estimated_eval_tokens": 1982 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs touch the Wan pipeline\u2019s ftfy handling, but they are not the same change: one forces an unconditional import, the other guards the text-cleaning call on ftfy availability. They target the same symptom area, yet they are different fixes and should not be deduplicated as one PR.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12314 is the better canonical PR because it is merged and fixes the ftfy error in the actual runtime path by guarding use of ftfy when unavailable.", + "best_issue_reason": null, + "best_pr_reason": "PR 12314 is the best PR representative since it is the merged, safer fix for the Wan pipeline ftfy error.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12023", + "right": "pull_request:12314", + "accept": false, + "reason": "Same file and related ftfy symptom, but 12023 changes the import strategy while 12314 guards the call site; they are distinct fixes and not merge-equivalent." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 62, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12032-3", + "nodes": [ + "pull_request:12032", + "pull_request:12179", + "pull_request:12194" + ], + "soft_pairs": [ + "pull_request:12032|pull_request:12194", + "pull_request:12179|pull_request:12194" + ], + "prepared_review_unit_hash": "b08e4dc0e84ba31853b7e745498f50c263648c5b51bb66fe90c0abc46cbe10f6", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8943, + "estimated_input_tokens": 2236, + "estimated_eval_tokens": 4728 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8943, + "estimated_input_tokens": 2236, + "estimated_eval_tokens": 4728 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related Chinese documentation translation PRs in the same area, but they are successive follow-up translation passes rather than the same underlying change. The broadest/latest PR is 12194, while the soft edges should be rejected as non-duplicates.", + "confidence": 0.74, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:12194 is the broadest and most recent zh translation pass, with the largest doc surface and substantial _toctree expansion, so it is the best representative PR in this cluster.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:12194 best represents the cluster because it covers the widest set of translated zh docs and subsystem navigation updates, making it the most comprehensive translation PR here.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12032", + "right": "pull_request:12194", + "accept": false, + "reason": "Both are zh translation PRs, but 12194 is a later continuation that adds different docs and toctree entries; they are related follow-ups, not the same concrete change." + }, + { + "left": "pull_request:12179", + "right": "pull_request:12194", + "accept": false, + "reason": "They overlap on zh documentation translation and _toctree.yml, but 12179 focuses on modular diffusers translations while 12194 expands broader optimization/inference docs; not a single duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 63, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12049-2", + "nodes": [ + "pull_request:12049", + "pull_request:12076" + ], + "soft_pairs": [ + "pull_request:12049|pull_request:12076" + ], + "prepared_review_unit_hash": "980e0d4fe897082aa582969364b271248365bf490c06eed3ab80e63d449b97c3", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4197, + "estimated_input_tokens": 1050, + "estimated_eval_tokens": 2356 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4197, + "estimated_input_tokens": 1050, + "estimated_eval_tokens": 2356 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests are effectively the same gguf quantization support change in the same file, with nearly identical added code. The later merged PR is the better canonical representative.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12076 is the merged version of the same change, with the more complete patch and the strongest evidence of being the final canonical implementation.", + "best_issue_reason": null, + "best_pr_reason": "PR #12076 is the best overall PR because it was merged and contains the finalized form of the same code-path update.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12049", + "right": "pull_request:12076", + "accept": true, + "reason": "Same title, same target file, and nearly identical diff adding IQ4_NL dequantization support; this looks like the same underlying change rather than separate PRs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 64, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12058-2", + "nodes": [ + "pull_request:12058", + "pull_request:12615" + ], + "soft_pairs": [ + "pull_request:12058|pull_request:12615" + ], + "prepared_review_unit_hash": "b064b7fd95b9bfe37179b5d4fbee83c102d128f2db743030a7273513f0ca4bb8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3807, + "estimated_input_tokens": 952, + "estimated_eval_tokens": 2160 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3807, + "estimated_input_tokens": 952, + "estimated_eval_tokens": 2160 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are similarly titled but address different model-specific attention implementations in different files: Wan vs. LTX. They do not look mergeable into one change or the same underlying fix.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No shared canonical PR: PR #12058 modifies Wan attention internals in transformer_wan.py, while PR #12615 modifies LTX attention behavior in transformer_ltx.py. They target different code paths and model families.", + "best_issue_reason": null, + "best_pr_reason": "PR #12058 is the better canonical PR only by virtue of being a merged, concrete implementation, but it is not a duplicate of #12615; the latter is a separate model-specific change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12058", + "right": "pull_request:12615", + "accept": false, + "reason": "Same broad theme (cross-attention) but different models, files, and code paths. The Wan PR sets an attention flag in transformer_wan.py; the LTX PR initializes cross-attention state in transformer_ltx.py. They are not the same bug/change and should not be merged as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 65, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1209-2", + "nodes": [ + "pull_request:1209", + "pull_request:1245" + ], + "soft_pairs": [ + "pull_request:1209|pull_request:1245" + ], + "prepared_review_unit_hash": "f79b48efa7a534c8d71033b15f07ea0104f55c580e8508da14fd4673d3819a2d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4552, + "estimated_input_tokens": 1138, + "estimated_eval_tokens": 2532 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4552, + "estimated_input_tokens": 1138, + "estimated_eval_tokens": 2532 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two related but distinct PRs for the unconditional image generation example: one updates docs/dependencies to include wandb, the other adds a logger CLI option in the training script. They share a subsystem but do not look like the same mergeable change.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 1245 is the more canonical code change because it directly implements the wandb logger selection in `train_unconditional.py`; the other PR is a supporting docs/dependency update.", + "best_issue_reason": null, + "best_pr_reason": "PR 1245 is the best representative of the cluster because it contains the main functional behavior change, while PR 1209 is ancillary setup/documentation work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1209", + "right": "pull_request:1245", + "accept": false, + "reason": "Related to wandb support, but not the same concrete change: 1209 updates example docs/.gitignore and installs wandb, while 1245 changes the training script to add a logger argument and logging behavior. They could be merged separately and are not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 66, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12103-2", + "nodes": [ + "pull_request:12103", + "pull_request:12298" + ], + "soft_pairs": [ + "pull_request:12103|pull_request:12298" + ], + "prepared_review_unit_hash": "580ca9df7e53bef6b942a2b30149ad1a2b7c427623f204e8f4cd5724fcc4c8c8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4326, + "estimated_input_tokens": 1082, + "estimated_eval_tokens": 2420 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4326, + "estimated_input_tokens": 1082, + "estimated_eval_tokens": 2420 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are both LoRA test-related, but they address different test behaviors: one fixes device placement for non-denoiser components during group-offloading tests, while the other removes redundant non-LoRA baseline output checks and caches outputs across several LoRA test files. They are not the same underlying change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12298 is the broader, more representative test cleanup/change: it touches multiple LoRA test files and the shared test utility, all around caching non-LoRA outputs.", + "best_issue_reason": null, + "best_pr_reason": "PR #12298 is the best representative of the cluster because it is the larger and more central test refactor, whereas #12103 is a narrower device-placement fix in a single test path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12103", + "right": "pull_request:12298", + "accept": false, + "reason": "They target different problems: #12103 fixes device placement for non-denoiser components in group-offloading LoRA tests, while #12298 removes redundant non-LoRA baseline inference and caches outputs across LoRA tests. Shared test utility usage is not enough to make them the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 67, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12191-2", + "nodes": [ + "pull_request:12191", + "pull_request:12753" + ], + "soft_pairs": [ + "pull_request:12191|pull_request:12753" + ], + "prepared_review_unit_hash": "ff8aedf7832bd1168a9610c5f30be50ee8de5ca2b40e73e129db8475f52ca0a7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4345, + "estimated_input_tokens": 1087, + "estimated_eval_tokens": 2430 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4345, + "estimated_input_tokens": 1087, + "estimated_eval_tokens": 2430 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs both touch AutoEncoderKLWan tiling, but they fix different concrete bugs: one changes patchify/encode-decode path selection, the other fixes spatial compression ratio handling in tiled encode. They should not be treated as duplicates.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12753 is the better canonical PR because it is merged and directly fixes a specific tiled-encode compression-ratio bug in AutoEncoderKLWan.", + "best_issue_reason": null, + "best_pr_reason": "PR #12753 is the strongest representative of the cluster: it is merged, narrowly scoped, and addresses a concrete code-path error. PR #12191 is open and targets a different tiling/patchify behavior.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12191", + "right": "pull_request:12753", + "accept": false, + "reason": "Same broad model class and file, but different bugs and code paths: #12191 changes when to patchify during encode/decode, while #12753 fixes tiled-encode latent compression ratio math. They are not the same underlying change and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 68, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12201-2", + "nodes": [ + "pull_request:12201", + "pull_request:12212" + ], + "soft_pairs": [ + "pull_request:12201|pull_request:12212" + ], + "prepared_review_unit_hash": "04d20d9b00a053cd70591ba63fa4c1d4e3ef44f0320b1bfb3098a4a25278438e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4232, + "estimated_input_tokens": 1058, + "estimated_eval_tokens": 2372 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4232, + "estimated_input_tokens": 1058, + "estimated_eval_tokens": 2372 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are merged documentation PRs, but they address different topics: one rewrites the community pipelines page, the other adds callback docs/API entries. They do not look like the same underlying change, so the soft duplicate edge should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:12201", + "right": "pull_request:12212", + "accept": false, + "reason": "Different doc updates in different areas of the docs tree; no shared concrete bug/change and not mergeable as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 69, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12206-4", + "nodes": [ + "pull_request:12206", + "pull_request:13075", + "pull_request:13093", + "pull_request:13514" + ], + "soft_pairs": [ + "pull_request:12206|pull_request:13093", + "pull_request:13093|pull_request:13514", + "pull_request:13075|pull_request:13093", + "pull_request:13075|pull_request:13514" + ], + "prepared_review_unit_hash": "9e74dc5604f91cced30c697cef29ca8f7d9600731d52b3bed69ff4f14f7dfe06", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 9259, + "estimated_input_tokens": 2315, + "estimated_eval_tokens": 4886 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 9259, + "estimated_input_tokens": 2315, + "estimated_eval_tokens": 4886 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "`13093` and `13514` are the same gather-device compatibility fix in `_modeling_parallel.py` and should cluster together. `13075` is related by target/file but its evidence is only soft similarity plus the shared issue target, so it should stay separate rather than being auto-merged with the other two. `12206` is a different compatibility fix in `attention_dispatch.py` and is unrelated.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "`13514` is the best representative of the gather fix: it addresses the same `gather_size_by_comm` path, explicitly states the PyTorch < 2.6 compatibility constraint, and gives the most direct fallback behavior.", + "best_issue_reason": null, + "best_pr_reason": "`13514` is the strongest cluster representative because it is the clearest, most self-contained version of the shared distributed-gather compatibility patch in `_modeling_parallel.py`.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12206", + "right": "pull_request:13093", + "accept": false, + "reason": "Different subsystem and different bug: `12206` guards `torch.library.custom_op`/`register_fake` in `attention_dispatch.py`, while `13093` changes distributed gather device selection in `_modeling_parallel.py`." + }, + { + "left": "pull_request:13093", + "right": "pull_request:13514", + "accept": true, + "reason": "Same concrete code path and same compatibility bug in `gather_size_by_comm`; both patch older-PyTorch device selection in the same file and are plausible variants of one mergeable fix." + }, + { + "left": "pull_request:13075", + "right": "pull_request:13093", + "accept": false, + "reason": "They share the file and issue target, but the pairwise evidence is only soft similarity plus the common target; that is not enough to treat them as the same mergeable bugfix." + }, + { + "left": "pull_request:13075", + "right": "pull_request:13514", + "accept": false, + "reason": "Related context-parallel gather changes, but the evidence is still only shared target/file plus soft similarity, so this is not strong enough for a deterministic duplicate/merge decision." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The overall clustering is mostly grounded, but the soft-edge verdicts are too strong for 13075. In the packet, 13075 only shares the issue target/file with 13093 and 13514; its pairwise signal is not a deterministic accept, so saying those pairs are definitely mergeable variants of the same bugfix overstates the evidence. 13093\u219413514 is well supported; 12206 is correctly separated." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 70, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12267-2", + "nodes": [ + "pull_request:12267", + "pull_request:12289" + ], + "soft_pairs": [ + "pull_request:12267|pull_request:12289" + ], + "prepared_review_unit_hash": "5640e475f4d483e3886fd6aef331490b164466c11e924e018c57cc525f560993", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6149, + "estimated_input_tokens": 1538, + "estimated_eval_tokens": 3332 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6149, + "estimated_input_tokens": 1538, + "estimated_eval_tokens": 3332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests are the same type-hint cleanup across the same transformer code paths; PR 12289 appears to be the merged, more complete superset of PR 12267.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12289 is the merged, more complete version of the same change set, with nearly identical edits plus an additional transformer file.", + "best_issue_reason": null, + "best_pr_reason": "PR 12289 is the best representative because it was actually merged and covers the same underlying fix with broader completion.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12267", + "right": "pull_request:12289", + "accept": true, + "reason": "Same underlying code change: the titles, touched files, and diff preview overlap heavily, and 12289 looks like a later merged superset of 12267 rather than a different PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 71, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12320-3", + "nodes": [ + "pull_request:12320", + "pull_request:12394", + "pull_request:12486" + ], + "soft_pairs": [ + "pull_request:12320|pull_request:12394", + "pull_request:12320|pull_request:12486", + "pull_request:12394|pull_request:12486" + ], + "prepared_review_unit_hash": "185c1f2872e606ebe5b4e54498af0584fbc4f892f0964276b75c02d94e5c9d0b", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6967, + "estimated_input_tokens": 1742, + "estimated_eval_tokens": 3740 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6967, + "estimated_input_tokens": 1742, + "estimated_eval_tokens": 3740 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs touch the same docs page, but they are sequential, distinct documentation updates rather than duplicate changes; PR #12320 is the base introduction of the page.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12320 creates the new attention backends doc page and adds it to the toctree, making it the foundational artifact in this set.", + "best_issue_reason": null, + "best_pr_reason": "PR #12320 is the most representative because it introduces the page itself, while the others are later incremental edits to that page.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12320", + "right": "pull_request:12394", + "accept": false, + "reason": "Same documentation file, but not the same change: #12320 adds the page, while #12394 only makes small wording/tone edits to it." + }, + { + "left": "pull_request:12320", + "right": "pull_request:12486", + "accept": false, + "reason": "Different doc updates on the same page: #12486 adds a new checks section, so it is not a duplicate of the page-creation PR." + }, + { + "left": "pull_request:12394", + "right": "pull_request:12486", + "accept": false, + "reason": "Both are follow-up edits to the same docs page, but they address different content (minor wording tweaks vs. adding attention checks)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 72, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12322-2", + "nodes": [ + "pull_request:12322", + "pull_request:12820" + ], + "soft_pairs": [ + "pull_request:12322|pull_request:12820" + ], + "prepared_review_unit_hash": "1cff384175f2cabfac3d970de118c58b2286aa64e3835ca43fc823ee37f9776a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5871, + "estimated_input_tokens": 1468, + "estimated_eval_tokens": 3192 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5871, + "estimated_input_tokens": 1468, + "estimated_eval_tokens": 3192 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are effectively two versions of the same pull request: same title, same broad refactor, and the same set of community pipeline files. The newer open PR looks like a continuation/replacement of the older closed draft PR.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12820 is the newer open version of the same refactor, so it is the best canonical PR candidate; #12322 appears to be an earlier closed draft of the same work.", + "best_issue_reason": null, + "best_pr_reason": "#12820 is the best representative because it is current, open, and matches the same scope/files as #12322 without adding evidence of a different change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12322", + "right": "pull_request:12820", + "accept": true, + "reason": "Same title, same file set, and same underlying refactor to introduce pipeline-specific mixins; #12820 looks like a continuation of #12322 rather than a different PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 73, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12336-2", + "nodes": [ + "pull_request:12336", + "pull_request:12342" + ], + "soft_pairs": [ + "pull_request:12336|pull_request:12342" + ], + "prepared_review_unit_hash": "803b026798dffeb06c00a77f65ed83919a8b1d525890040e8f382fac5e141ef9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4269, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4269, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two pull requests are effectively duplicates: they propose the same tiling fix in `autoencoder_kl_wan.py` with the same diff hunk and near-identical titles.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12342 is the better canonical PR because it is the open, current instance of the same fix; #12336 is a closed duplicate of the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR #12342 best represents the cluster since it is the active PR and matches the same concrete code-path fix as #12336.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12336", + "right": "pull_request:12342", + "accept": true, + "reason": "Same file, same hunk, same tiling bug fix, and the diffs appear identical; these are the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 74, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12351-2", + "nodes": [ + "pull_request:12351", + "pull_request:12366" + ], + "soft_pairs": [ + "pull_request:12351|pull_request:12366" + ], + "prepared_review_unit_hash": "52a862763200c976dd61f2ea9eb9008a4ee8915af0af651da733069391c5e3f4", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4435, + "estimated_input_tokens": 1109, + "estimated_eval_tokens": 2474 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4435, + "estimated_input_tokens": 1109, + "estimated_eval_tokens": 2474 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are near-duplicate documentation PRs adding CacheDiT to Diffusers docs. PR #12366 is the merged, more complete version and is the best canonical representative.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12366 is the merged implementation and the fuller, later docs addition for the same CacheDiT community optimization page and toctree entry.", + "best_issue_reason": null, + "best_pr_reason": "PR #12366 best represents the cluster because it actually landed, covers the same two files, and appears to supersede the earlier duplicate docs PR #12351.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12351", + "right": "pull_request:12366", + "accept": true, + "reason": "Both PRs add the same CacheDiT documentation page and toctree entry under the same path; they are effectively the same documentation change and could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 75, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12352-3", + "nodes": [ + "pull_request:12352", + "pull_request:12379", + "pull_request:12621" + ], + "soft_pairs": [ + "pull_request:12352|pull_request:12621", + "pull_request:12379|pull_request:12621" + ], + "prepared_review_unit_hash": "64fc505d31fa89a216d8bce2d3e0ba6c101d24b5e97d280093839a0f403fa901", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7841, + "estimated_input_tokens": 1961, + "estimated_eval_tokens": 4178 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7841, + "estimated_input_tokens": 1961, + "estimated_eval_tokens": 4178 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The three PRs are related by a broad theme of removing `unittest` from test infrastructure, but they modify different code paths: shared testing utilities, single-file test mixins, and model tester helpers. They are not duplicates of the same change.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12621 is the best canonical PR for the cluster because it changes the shared `tests/testing_utils.py` helpers used by many tests, replacing `unittest.skipUnless` with `pytest.mark.skipif` in the core slow/nightly decorators.", + "best_issue_reason": null, + "best_pr_reason": "PR #12621 is the strongest representative of the group: it is merged, touches the common test utility layer, and looks like the most central infrastructure change among the three. The other PRs are adjacent refactors in narrower test files rather than the same underlying change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12352", + "right": "pull_request:12621", + "accept": false, + "reason": "Both are test-framework modernization work, but #12352 is a single-file model tester refactor while #12621 changes shared decorator utilities; they do not fix the same concrete bug or change the same code path." + }, + { + "left": "pull_request:12379", + "right": "pull_request:12621", + "accept": false, + "reason": "#12379 only removes `unittest` usage in `test_modeling_common.py`, whereas #12621 updates shared `testing_utils.py`; these are related cleanup tasks, not one duplicate fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 76, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1236-2", + "nodes": [ + "pull_request:1236", + "pull_request:1457" + ], + "soft_pairs": [ + "pull_request:1236|pull_request:1457" + ], + "prepared_review_unit_hash": "aaa9ff6a2dbc505385e0eb47e735dbde058ed504a0696a720ab3b30a2eb95110", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4530, + "estimated_input_tokens": 1133, + "estimated_eval_tokens": 2522 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4530, + "estimated_input_tokens": 1133, + "estimated_eval_tokens": 2522 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs target DDIM sampling, but they are distinct changes: #1236 adds pipeline support for predict_x0 and uneven jumps, while #1457 introduces a more extensive paired-timestep DDIM scheduler rewrite with helper/tests. They are related but not duplicates.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1457 is the stronger representative of the cluster because it is the more complete DDIM scheduler/pipeline change and includes supporting helper code and tests.", + "best_issue_reason": null, + "best_pr_reason": "PR #1457 is the best overall candidate: it is broader, more self-contained, and better validated than the narrower pipeline-only tweak in #1236.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1236", + "right": "pull_request:1457", + "accept": false, + "reason": "Same subsystem, but not the same concrete fix. #1236 focuses on pipeline support for predict_x0 and uneven jumps; #1457 implements a different paired-timestep DDIM scheduler extension with additional helper/test changes. They are related but would not plausibly collapse into one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 77, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12382-5", + "nodes": [ + "pull_request:12382", + "pull_request:12639", + "pull_request:13234", + "pull_request:13242", + "pull_request:13491" + ], + "soft_pairs": [ + "pull_request:12382|pull_request:13234", + "pull_request:12639|pull_request:13491", + "pull_request:13234|pull_request:13242", + "pull_request:13234|pull_request:13491" + ], + "prepared_review_unit_hash": "c49fdcb900520ab99e689428046357671d0ff63eb4b553d4c19f628b871973ea", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 10394, + "estimated_input_tokens": 2599, + "estimated_eval_tokens": 5454 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 10394, + "estimated_input_tokens": 2599, + "estimated_eval_tokens": 5454 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Mostly loose topical overlap around group offloading and test maintenance, but these PRs fix different problems: a Qwen-specific test skip removal, a PEFT delete_adapters regression, pytest migration work, and a separate group-offloading-with-disk test adjustment. I don\u2019t see true duplicate PR pairs here.", + "confidence": 0.66, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 13491 is the closest representative of the shared theme because it directly fixes a concrete group-offloading-with-disk test path, but the cluster is not actually one duplicate change set.", + "best_issue_reason": null, + "best_pr_reason": "PR 13491 is the strongest standalone candidate since it targets a specific failing test path in group offloading with disk and makes a concrete behavioral adjustment, unlike the broader test-migration PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12382", + "right": "pull_request:13234", + "accept": false, + "reason": "Different scope: 12382 only removes a QwenImageTransformer2DModel test skip, while 13234 is a broad pytest migration for group offloading tests." + }, + { + "left": "pull_request:12639", + "right": "pull_request:13491", + "accept": false, + "reason": "Different underlying bug: 12639 fixes delete_adapters breaking leaf-level offloading, while 13491 adjusts test setup for group offloading with disk." + }, + { + "left": "pull_request:13234", + "right": "pull_request:13242", + "accept": false, + "reason": "Both are pytest migration/refactor PRs, but they touch different test files and different subsystems; not the same concrete change." + }, + { + "left": "pull_request:13234", + "right": "pull_request:13491", + "accept": false, + "reason": "13234 is a test-framework migration, whereas 13491 is a specific group-offloading test fix; similar area, but not the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 78, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12384-2", + "nodes": [ + "pull_request:12384", + "pull_request:12389" + ], + "soft_pairs": [ + "pull_request:12384|pull_request:12389" + ], + "prepared_review_unit_hash": "f67325c08532017ecb078b125805975b2983d8c8653924c02cddcec0afd4b774", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5943, + "estimated_input_tokens": 1486, + "estimated_eval_tokens": 3228 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5943, + "estimated_input_tokens": 1486, + "estimated_eval_tokens": 3228 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about adapting to Hugging Face Hub v1, but they solve different problems: one is temporary CI/test environment pinning, the other is the actual library compatibility change. They should not be clustered as duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12389 is the substantive, merged code change that implements v0.x/v1.x compatibility in the library and tests. PR #12384 is a draft, temporary workflow-only adaptation for testing and should not be canonical.", + "best_issue_reason": null, + "best_pr_reason": "PR #12389 is the best representative of the underlying change because it directly updates dependencies and exception handling to support both huggingface_hub major versions, and it was merged.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12384", + "right": "pull_request:12389", + "accept": false, + "reason": "Related to the same ecosystem shift, but not the same change: #12384 only tweaks CI/workflows for temporary test compatibility, while #12389 changes runtime/library code to support huggingface_hub v0.x and v1.x. They would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 79, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12478-3", + "nodes": [ + "pull_request:12478", + "pull_request:12582", + "pull_request:12664" + ], + "soft_pairs": [ + "pull_request:12478|pull_request:12664", + "pull_request:12582|pull_request:12664" + ], + "prepared_review_unit_hash": "f633aca9d3ad75d12f77fc1bc938101a4389ae7786505a78062cb9287e8a52f2", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8446, + "estimated_input_tokens": 2112, + "estimated_eval_tokens": 4480 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8446, + "estimated_input_tokens": 2112, + "estimated_eval_tokens": 4480 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are related to the Kandinsky 5 family, but they are not duplicates: 12478 introduces the original Kandinsky 5 integration, 12582 only fixes docs/navigation, and 12664 adds new image/video variants plus tests. The cluster is better treated as a theme group than a duplicate set.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12664 is the most complete representative of the Kandinsky 5 feature set, covering both image and video variants with the broadest code/test surface.", + "best_issue_reason": null, + "best_pr_reason": "PR 12664 is the best single representative because it reflects the most mature, user-facing Kandinsky 5 implementation; 12478 is the initial integration and 12582 is docs-only.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12478", + "right": "pull_request:12664", + "accept": false, + "reason": "Related subsystem, but different concrete changes: 12478 is the initial Kandinsky 5 add-on, while 12664 adds separate image/video lite/pro pipelines and tests. They could not plausibly be merged as the same PR." + }, + { + "left": "pull_request:12582", + "right": "pull_request:12664", + "accept": false, + "reason": "12582 is a docs-only rename/navigation cleanup, whereas 12664 introduces substantial new pipeline code and tests. Not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 80, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12545-2", + "nodes": [ + "pull_request:12545", + "pull_request:12930" + ], + "soft_pairs": [ + "pull_request:12545|pull_request:12930" + ], + "prepared_review_unit_hash": "6718d9c10aa00d499ca99c60bda81796ba8bed4cb31578c93d1fcd528bbfbdb5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5664, + "estimated_input_tokens": 1416, + "estimated_eval_tokens": 3088 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5664, + "estimated_input_tokens": 1416, + "estimated_eval_tokens": 3088 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are related but not duplicates: #12545 adds the core Bria Fibo model/pipeline, while #12930 adds a separate Bria Fibo Edit pipeline built on the same family. They share some boilerplate and integration points, but they implement different concrete features and code paths.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12545 is the broader, more foundational Bria Fibo addition and has substantially more review activity; it is the better representative PR for this cluster, but it is not the same change as #12930.", + "best_issue_reason": null, + "best_pr_reason": "PR #12545 is the strongest representative because it introduces the base Bria Fibo pipeline/model and the associated docs/tests, making it the more central artifact in the set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12545", + "right": "pull_request:12930", + "accept": false, + "reason": "Different concrete changes: #12545 adds the base Bria Fibo model/pipeline, while #12930 adds a distinct Bria Fibo Edit pipeline. Shared package wiring and docs do not make them the same underlying fix or mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 81, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12555-2", + "nodes": [ + "pull_request:12555", + "pull_request:13518" + ], + "soft_pairs": [ + "pull_request:12555|pull_request:13518" + ], + "prepared_review_unit_hash": "52579e2b9b3033b3a9e7a1d411d830156a2de8a5359e1d3e04d0a544a23a8f1d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3725, + "estimated_input_tokens": 932, + "estimated_eval_tokens": 2120 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3725, + "estimated_input_tokens": 932, + "estimated_eval_tokens": 2120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two PRs reference the same issue, but they fix different parts of onnxruntime support: one changes provider handling in `onnx_utils.py`, the other adds `onnxruntime-windowsml` to version detection in `import_utils.py`. They are related, but not the same concrete bug/change.", + "confidence": 0.83, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12555 is the broader, earlier fix and directly addresses the runtime provider-path change for `onnxruntime-windowsml`, making it the better representative of the cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 12555 is the strongest candidate because it changes the actual loading logic and is the more complete implementation of the feature/support request.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12555", + "right": "pull_request:13518", + "accept": false, + "reason": "Both mention issue 13062, but they modify different code paths: `onnx_utils.load_model` provider selection vs `import_utils` package detection. Related, but not the same fix and not something I\u2019d merge as a duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 82, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12563-2", + "nodes": [ + "pull_request:12563", + "pull_request:12829" + ], + "soft_pairs": [ + "pull_request:12563|pull_request:12829" + ], + "prepared_review_unit_hash": "5f8ec820d1192d62b0b53bbad209983d92021658a1b65007a5a8da6e91c53e0e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4217, + "estimated_input_tokens": 1055, + "estimated_eval_tokens": 2366 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4217, + "estimated_input_tokens": 1055, + "estimated_eval_tokens": 2366 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related attention-dispatch PRs, but they fix different concrete paths: one enables ulysses in the generic native attention path, the other adds CP support in native flash attention. Same file/subsystem, not the same change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No duplicate canonical PR: PR 12829 is a separate flash-attention CP support change, while PR 12563 targets the generic native attention path for ulysses enablement.", + "best_issue_reason": null, + "best_pr_reason": "PR 12829 is the better standalone representative because it has a narrower, clearer scope around native flash attention support and is not conflated with the separate native-attention ulysses change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12563", + "right": "pull_request:12829", + "accept": false, + "reason": "Different code paths and goals: native attention vs native flash attention. Related subsystem, but not the same underlying bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 83, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12617-3", + "nodes": [ + "pull_request:12617", + "pull_request:12671", + "pull_request:13481" + ], + "soft_pairs": [ + "pull_request:12617|pull_request:12671", + "pull_request:12617|pull_request:13481" + ], + "prepared_review_unit_hash": "486a95f599e55ac6fe51ad463714ea3aec4797b1e3e116a2f4071fc5397b16ef", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6758, + "estimated_input_tokens": 1690, + "estimated_eval_tokens": 3636 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6758, + "estimated_input_tokens": 1690, + "estimated_eval_tokens": 3636 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two PRs are closely related CI fixes for the same prerelease dependency issue in workflow setup; the BnB quantization PR is unrelated and should not be clustered with them.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:12671 is the better cluster representative because it targets the same CI dependency regression more broadly across nightly/release workflows, whereas pull_request:12617 is a narrower push-tests-only patch.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:12671 is the strongest PR candidate for the shared CI regression: it applies the same transformers/huggingface_hub prerelease install workaround in the more comprehensive workflow context.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12617", + "right": "pull_request:12671", + "accept": true, + "reason": "Both PRs apply the same underlying CI fix: reinstalling prerelease transformers and removing conflicting huggingface_hub/accelerate state in workflow setup to unblock tests. They could plausibly be merged into one PR touching multiple workflow files." + }, + { + "left": "pull_request:12617", + "right": "pull_request:13481", + "accept": false, + "reason": "These address different problems: 12617 is a CI workflow dependency workaround, while 13481 changes Wan attention code and related tests to fix BnB quantization behavior. They are not the same underlying bug or change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster call is grounded in the packet: 12617 and 12671 are both CI workflow patches that reinstall prerelease transformers and remove conflicting huggingface_hub/accelerate state, so treating them as mergeable into one broader CI-fix PR is reasonable. The rejection of 12617 with 13481 is also conservative and supported by the very different model/quantization scope. Minor caution: the wording about \"nightly/release workflows\" leans on the PR title more than the diff snippet, but it does not overstate the evidence enough to invalidate the result." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 84, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12622-22", + "nodes": [ + "pull_request:12622", + "pull_request:12623", + "pull_request:12651", + "pull_request:12654", + "pull_request:12676", + "pull_request:12678", + "pull_request:12710", + "pull_request:12726", + "pull_request:12766", + "pull_request:12767", + "pull_request:12796", + "pull_request:12798", + "pull_request:12871", + "pull_request:12928", + "pull_request:12931", + "pull_request:12936", + "pull_request:12992", + "pull_request:13010", + "pull_request:13020", + "pull_request:13023", + "pull_request:13024", + "pull_request:13027" + ], + "soft_pairs": [ + "pull_request:12623|pull_request:12654", + "pull_request:12623|pull_request:12676", + "pull_request:12623|pull_request:12678", + "pull_request:12623|pull_request:12710", + "pull_request:12623|pull_request:12726", + "pull_request:12623|pull_request:12766", + "pull_request:12623|pull_request:12767", + "pull_request:12623|pull_request:12796", + "pull_request:12623|pull_request:12798", + "pull_request:12623|pull_request:12871", + "pull_request:12623|pull_request:12928", + "pull_request:12623|pull_request:12931", + "pull_request:12623|pull_request:12936", + "pull_request:12623|pull_request:12992", + "pull_request:12623|pull_request:13010", + "pull_request:12623|pull_request:13020", + "pull_request:12623|pull_request:13023", + "pull_request:12623|pull_request:13024", + "pull_request:12623|pull_request:13027", + "pull_request:12654|pull_request:12676", + "pull_request:12654|pull_request:12678", + "pull_request:12654|pull_request:12710", + "pull_request:12654|pull_request:12726", + "pull_request:12654|pull_request:12766", + "pull_request:12654|pull_request:12767", + "pull_request:12654|pull_request:12796", + "pull_request:12654|pull_request:12798", + "pull_request:12654|pull_request:12871", + "pull_request:12654|pull_request:12928", + "pull_request:12654|pull_request:12931", + "pull_request:12654|pull_request:12936", + "pull_request:12654|pull_request:12992", + "pull_request:12654|pull_request:13010", + "pull_request:12654|pull_request:13020", + "pull_request:12654|pull_request:13023", + "pull_request:12654|pull_request:13024", + "pull_request:12654|pull_request:13027", + "pull_request:12676|pull_request:12678", + "pull_request:12676|pull_request:12710", + "pull_request:12676|pull_request:12726", + "pull_request:12676|pull_request:12766", + "pull_request:12676|pull_request:12767", + "pull_request:12676|pull_request:12796", + "pull_request:12676|pull_request:12798", + "pull_request:12676|pull_request:12871", + "pull_request:12676|pull_request:12928", + "pull_request:12676|pull_request:12931", + "pull_request:12676|pull_request:12936", + "pull_request:12676|pull_request:12992", + "pull_request:12676|pull_request:13010", + "pull_request:12676|pull_request:13020", + "pull_request:12676|pull_request:13023", + "pull_request:12676|pull_request:13024", + "pull_request:12676|pull_request:13027", + "pull_request:12622|pull_request:12651", + "pull_request:12651|pull_request:12654", + "pull_request:12651|pull_request:12676", + "pull_request:12651|pull_request:12678", + "pull_request:12651|pull_request:12710", + "pull_request:12651|pull_request:12726", + "pull_request:12651|pull_request:12766", + "pull_request:12651|pull_request:12767", + "pull_request:12651|pull_request:12796", + "pull_request:12651|pull_request:12798", + "pull_request:12651|pull_request:12871", + "pull_request:12651|pull_request:12928", + "pull_request:12651|pull_request:12931", + "pull_request:12651|pull_request:12936", + "pull_request:12651|pull_request:12992", + "pull_request:12651|pull_request:13010", + "pull_request:12651|pull_request:13020", + "pull_request:12651|pull_request:13023" + ], + "prepared_review_unit_hash": "47c4f514b6f62f0081d1c6cccac36d8e93cecf2365c1ee561bb5018f98d99672", + "component_budget": { + "node_count": 22, + "item_count": 22, + "soft_pair_count": 95, + "serialized_chars": 90548, + "estimated_input_tokens": 22637, + "estimated_eval_tokens": 45530 + }, + "budget": { + "node_count": 22, + "item_count": 22, + "soft_pair_count": 72, + "serialized_chars": 74155, + "estimated_input_tokens": 18539, + "estimated_eval_tokens": 37334 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a set of related but separate scheduler maintenance PRs: mostly docstring and type-hint cleanups across different files. They are similar in style and often touch shared copied helpers, but they do not represent one duplicate bug/change. I would not merge any of the soft-edge pairs as duplicates.", + "confidence": 0.83, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12651 is the best canonical representative: it is one of the broadest cleanup PRs in the set and touches the shared scheduler/helper surface that many later file-specific refactors copy from.", + "best_issue_reason": null, + "best_pr_reason": "PR 12651 is the strongest global representative because it spans multiple scheduler modules and shared helper code, making it the most central example of the refactor pattern in this cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12623", + "right": "pull_request:12654", + "accept": false, + "reason": "Different scheduler modules (`amused.py` vs `euler_discrete.py`); same maintenance style, but not the same concrete change." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12676", + "accept": false, + "reason": "Separate file-specific doc/type-hint cleanup in unrelated scheduler code; not one underlying bug/change." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12678", + "accept": false, + "reason": "Different code paths (`amused.py` vs `dpmsolver_sde.py`/`lms_discrete.py`); too broad to treat as duplicates." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12710", + "accept": false, + "reason": "Same refactor theme, but across different scheduler families and copied helpers; not mergeable as one PR." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12726", + "accept": false, + "reason": "Unrelated scheduler file cleanup; no shared concrete bug or code-path fix." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12766", + "accept": false, + "reason": "Different scheduler implementation and different edits; only the docstring/type-hint pattern is shared." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12767", + "accept": false, + "reason": "Separate file-level cleanup in a different scheduler module; not the same change." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12796", + "accept": false, + "reason": "Different target file and refactor scope; no evidence of the same underlying fix." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12798", + "accept": false, + "reason": "Only stylistically similar; the PRs touch different schedulers and code paths." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12871", + "accept": false, + "reason": "Different scheduler families and helper updates; not a single code-path bug or fix." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12928", + "accept": false, + "reason": "`amused.py` vs `consistency_decoder.py`; different modules, so not duplicates." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12931", + "accept": false, + "reason": "Separate scheduler docs/type-hint cleanup in a different file; not the same change." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12936", + "accept": false, + "reason": "Different scheduler file and scope; only the maintenance pattern is shared." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12992", + "accept": false, + "reason": "Despite overlapping scheduler themes, these are broad but distinct file-family cleanups, not one merged PR." + }, + { + "left": "pull_request:12623", + "right": "pull_request:13010", + "accept": false, + "reason": "Different file and different implementation surface; not the same concrete bug/change." + }, + { + "left": "pull_request:12623", + "right": "pull_request:13020", + "accept": false, + "reason": "`amused.py` vs `ddim_inverse.py`; unrelated scheduler cleanup targets." + }, + { + "left": "pull_request:12623", + "right": "pull_request:13023", + "accept": false, + "reason": "Different scheduler modules and no shared concrete fix beyond style cleanup." + }, + { + "left": "pull_request:12623", + "right": "pull_request:13024", + "accept": false, + "reason": "Different file-specific cleanup; not the same underlying change." + }, + { + "left": "pull_request:12623", + "right": "pull_request:13027", + "accept": false, + "reason": "Different scheduler family and code path; too broad to deduplicate." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12676", + "accept": false, + "reason": "Both are doc/type-hint cleanups, but they target different schedulers (`euler_discrete.py` vs `pndm.py`)." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12678", + "accept": false, + "reason": "Different scheduler helper/code paths; not the same bug or change." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12710", + "accept": false, + "reason": "Overlapping style, but distinct multi-file scheduler cleanup PRs; not one concrete fix." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12726", + "accept": false, + "reason": "Different scheduler modules and no shared code-path issue." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12766", + "accept": false, + "reason": "Separate maintenance PR for another scheduler; not duplicate-worthy." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12767", + "accept": false, + "reason": "Different file-specific cleanup target; same theme only." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12796", + "accept": false, + "reason": "Different scheduler family and distinct edits; not the same change." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12798", + "accept": false, + "reason": "Different module and change set; only broadly similar refactoring." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12871", + "accept": false, + "reason": "Touching different scheduler implementations; no single shared bug/change." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12928", + "accept": false, + "reason": "Different files and different scheduler logic; not duplicates." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12931", + "accept": false, + "reason": "`euler_discrete.py` cleanup vs `consistency_models.py` cleanup; unrelated targets." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12936", + "accept": false, + "reason": "Different scheduler module and scope; same style of change, not same change." + }, + { + "left": "pull_request:12654", + "right": "pull_request:12992", + "accept": false, + "reason": "Both are broad scheduler cleanups, but they are separate PRs over different file sets and not one concrete fix." + }, + { + "left": "pull_request:12654", + "right": "pull_request:13010", + "accept": false, + "reason": "Different module and refactor surface; no evidence of same underlying bug/change." + }, + { + "left": "pull_request:12654", + "right": "pull_request:13020", + "accept": false, + "reason": "Separate scheduler files with distinct edits; not duplicate fixes." + }, + { + "left": "pull_request:12654", + "right": "pull_request:13023", + "accept": false, + "reason": "Different scheduler target (`euler_discrete.py` vs `ddim_parallel.py`); not the same change." + }, + { + "left": "pull_request:12654", + "right": "pull_request:13024", + "accept": false, + "reason": "Different scheduler family and code path; only similar doc/type-hint cleanup." + }, + { + "left": "pull_request:12654", + "right": "pull_request:13027", + "accept": false, + "reason": "Separate PRs touching different schedulers and helpers; not one underlying fix." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12678", + "accept": false, + "reason": "`pndm.py` and `lms_discrete.py`/`dpmsolver_sde.py` are different scheduler implementations; not duplicates." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12710", + "accept": false, + "reason": "Different scheduler families and file sets; similar documentation cleanup only." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12726", + "accept": false, + "reason": "Separate file-specific refactors; not the same concrete change." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12766", + "accept": false, + "reason": "Different scheduler module and edits; no same underlying bug/change." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12767", + "accept": false, + "reason": "Different file and scope; just the same cleanup pattern." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12796", + "accept": false, + "reason": "Separate scheduler cleanup PRs; not mergeable as one bugfix." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12798", + "accept": false, + "reason": "Different code paths in different scheduler implementations." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12871", + "accept": false, + "reason": "Different target schedulers and helper changes; not the same change." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12928", + "accept": false, + "reason": "Different modules; no shared concrete fix." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12931", + "accept": false, + "reason": "`pndm.py` cleanup vs `consistency_models.py` cleanup; unrelated." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12936", + "accept": false, + "reason": "Different scheduler file and change scope; same style only." + }, + { + "left": "pull_request:12676", + "right": "pull_request:12992", + "accept": false, + "reason": "Broad scheduler cleanups, but separate file families and distinct PRs." + }, + { + "left": "pull_request:12676", + "right": "pull_request:13010", + "accept": false, + "reason": "Different file-specific cleanup target; not the same underlying bug/change." + }, + { + "left": "pull_request:12676", + "right": "pull_request:13020", + "accept": false, + "reason": "Different scheduler implementations and edits; not duplicates." + }, + { + "left": "pull_request:12676", + "right": "pull_request:13023", + "accept": false, + "reason": "Different schedulers (`pndm.py` vs `ddim_parallel.py`); not one concrete fix." + }, + { + "left": "pull_request:12676", + "right": "pull_request:13024", + "accept": false, + "reason": "Different module and change set; only style similarity." + }, + { + "left": "pull_request:12676", + "right": "pull_request:13027", + "accept": false, + "reason": "Different scheduler family and code path; not the same PR/change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12651", + "accept": false, + "reason": "Both are scheduler cleanup PRs, but they touch different file groups and are not the same concrete refactor." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12654", + "accept": false, + "reason": "Overlapping scheduler helper surface, but the file sets and edits are distinct enough that these are separate cleanup PRs." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12676", + "accept": false, + "reason": "Different scheduler families and change sets; same documentation/type-hint theme only." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12678", + "accept": false, + "reason": "Different helper/code-path edits in different scheduler modules; not one bug/change." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12710", + "accept": false, + "reason": "Both are broad scheduler cleanups, but not the same underlying change and not safely mergeable into one PR." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12726", + "accept": false, + "reason": "Different scheduler file families; no identical change target." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12766", + "accept": false, + "reason": "Different scheduler modules and distinct cleanup scope." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12767", + "accept": false, + "reason": "Separate file-specific cleanup PRs; not duplicates." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12796", + "accept": false, + "reason": "Different scheduler implementations and distinct doc/type-hint changes." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12798", + "accept": false, + "reason": "Different modules and edit sets; not the same change." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12871", + "accept": false, + "reason": "Different scheduler families and code paths; only similar refactoring style." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12928", + "accept": false, + "reason": "`consistency_decoder.py` vs the broad scheduler cleanup in 12651; different change targets." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12931", + "accept": false, + "reason": "Different file and module; not the same underlying bug/change." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12936", + "accept": false, + "reason": "Different scheduler cleanup target; similar but not duplicate." + }, + { + "left": "pull_request:12651", + "right": "pull_request:12992", + "accept": false, + "reason": "Both are broad scheduler doc/type-hint sweeps, but they touch different file groups and should remain separate." + }, + { + "left": "pull_request:12651", + "right": "pull_request:13010", + "accept": false, + "reason": "Different scheduler file and change scope; not one concrete fix." + }, + { + "left": "pull_request:12651", + "right": "pull_request:13020", + "accept": false, + "reason": "Different scheduler module and edits; not duplicates." + }, + { + "left": "pull_request:12651", + "right": "pull_request:13023", + "accept": false, + "reason": "Different file-specific cleanup target; not the same underlying change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12651", + "accept": false, + "reason": "Different scheduler cleanup targets; the overlap is stylistic, not a shared code-path bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 84, + "component_count": 266, + "review_unit_index": 2, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12622-22", + "nodes": [ + "pull_request:12622", + "pull_request:12623", + "pull_request:12651", + "pull_request:12654", + "pull_request:12676", + "pull_request:12678", + "pull_request:12710", + "pull_request:12726", + "pull_request:12766", + "pull_request:12767", + "pull_request:12796", + "pull_request:12798", + "pull_request:12871", + "pull_request:12928", + "pull_request:12931", + "pull_request:12936", + "pull_request:12992", + "pull_request:13010", + "pull_request:13020", + "pull_request:13023", + "pull_request:13024", + "pull_request:13027" + ], + "soft_pairs": [ + "pull_request:12651|pull_request:13024", + "pull_request:12651|pull_request:13027", + "pull_request:12622|pull_request:12654", + "pull_request:12622|pull_request:12676", + "pull_request:12622|pull_request:12678", + "pull_request:12622|pull_request:12710", + "pull_request:12622|pull_request:12726", + "pull_request:12622|pull_request:12766", + "pull_request:12622|pull_request:12767", + "pull_request:12622|pull_request:12796", + "pull_request:12622|pull_request:12798", + "pull_request:12622|pull_request:12871", + "pull_request:12622|pull_request:12928", + "pull_request:12622|pull_request:12931", + "pull_request:12622|pull_request:12936", + "pull_request:12622|pull_request:12992", + "pull_request:12622|pull_request:13010", + "pull_request:12622|pull_request:13020", + "pull_request:12622|pull_request:13023", + "pull_request:12622|pull_request:13024", + "pull_request:12622|pull_request:13027", + "pull_request:12623|pull_request:12651", + "pull_request:12622|pull_request:12623" + ], + "prepared_review_unit_hash": "ff2c233d2e1f36047c1a00ae82b2a56187708bf9612723e3506beaa3c0524aa0", + "component_budget": { + "node_count": 22, + "item_count": 22, + "soft_pair_count": 95, + "serialized_chars": 90548, + "estimated_input_tokens": 22637, + "estimated_eval_tokens": 45530 + }, + "budget": { + "node_count": 22, + "item_count": 22, + "soft_pair_count": 23, + "serialized_chars": 55930, + "estimated_input_tokens": 13983, + "estimated_eval_tokens": 28222 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are mostly related scheduler docstring/type-hint cleanup PRs, but they target different files and code paths; they are not duplicates of the same underlying change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12622 is the best representative because it is an early, broad scheduler cleanup touching several related files and reflects the common pattern in the cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 12622 is the strongest cluster representative: broad scope, early in the sequence, and closest to the shared docstring/type-hint cleanup theme across schedulers.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12651", + "right": "pull_request:13024", + "accept": false, + "reason": "Both are scheduler cleanup PRs, but they edit different modules and different concrete helpers; same style of refactor, not the same change." + }, + { + "left": "pull_request:12651", + "right": "pull_request:13027", + "accept": false, + "reason": "Different target files and different modifications; they share a general cleanup theme but not a single underlying bug or change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12654", + "accept": false, + "reason": "Both are broad docstring/type-hint edits across schedulers, but they are separate file-specific cleanup passes, not one mergeable fix." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12676", + "accept": false, + "reason": "This is the same general documentation cleanup pattern, but it targets different scheduler code and is not the same concrete change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12678", + "accept": false, + "reason": "Different scheduler modules and different helpers are being updated; not a duplicate code-path fix." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12710", + "accept": false, + "reason": "Overlapping subsystem, but distinct scheduler APIs and file sets; these are separate cleanup PRs." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12726", + "accept": false, + "reason": "Both are doc/type-hint cleanups, but they concern different scheduler implementations and changes." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12766", + "accept": false, + "reason": "The shared scheduler area is too broad; the concrete edits are in different files and functions." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12767", + "accept": false, + "reason": "Separate cleanup to a different scheduler module; same theme, not the same underlying change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12796", + "accept": false, + "reason": "Different scheduler code paths and file scope; not a duplicate PR." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12798", + "accept": false, + "reason": "These are separate scheduler docstring/type-hint updates, not one same fix." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12871", + "accept": false, + "reason": "The PRs touch different scheduler implementations and update different methods; not the same concrete change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12928", + "accept": false, + "reason": "Different module and different target edits; same cleanup style only." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12931", + "accept": false, + "reason": "This is a different scheduler file with its own typing/docstring updates, not a duplicate." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12936", + "accept": false, + "reason": "Same broad cleanup wave, but separate code paths and no shared concrete change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12992", + "accept": false, + "reason": "Both are scheduler cleanup PRs, but they are distinct file groups and not one mergeable fix." + }, + { + "left": "pull_request:12622", + "right": "pull_request:13010", + "accept": false, + "reason": "Different target module and unrelated docstring/type-hint edits; not a duplicate." + }, + { + "left": "pull_request:12622", + "right": "pull_request:13020", + "accept": false, + "reason": "Both touch DDIM-family code, but they update different files and helpers; not the same underlying bug/change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:13023", + "accept": false, + "reason": "Different scheduler implementation and different concrete edits; same refactor theme only." + }, + { + "left": "pull_request:12622", + "right": "pull_request:13024", + "accept": false, + "reason": "These are separate file-specific cleanup PRs in the same subsystem, not the same change." + }, + { + "left": "pull_request:12622", + "right": "pull_request:13027", + "accept": false, + "reason": "Different scheduler files and different API/type-hint updates; not a duplicate." + }, + { + "left": "pull_request:12623", + "right": "pull_request:12651", + "accept": false, + "reason": "Both are docstring/type-hint cleanups, but they affect different scheduler files and functions." + }, + { + "left": "pull_request:12622", + "right": "pull_request:12623", + "accept": false, + "reason": "Same general cleanup campaign, but different scheduler modules and no shared concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 85, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12703-4", + "nodes": [ + "pull_request:12703", + "pull_request:12715", + "pull_request:12828", + "pull_request:13317" + ], + "soft_pairs": [ + "pull_request:12703|pull_request:12715", + "pull_request:12703|pull_request:12828", + "pull_request:12703|pull_request:13317", + "pull_request:12715|pull_request:12828" + ], + "prepared_review_unit_hash": "6292cf709b4e49b8a9fa5a6af77b1e15c7da3ed29b22d5753976c4d97665e1b5", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 12182, + "estimated_input_tokens": 3046, + "estimated_eval_tokens": 6348 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 12182, + "estimated_input_tokens": 3046, + "estimated_eval_tokens": 6348 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are four distinct pull requests: one adds Z-Image support, one is a Z-Image follow-up bugfix/test PR, and the other two add unrelated model families (LongCat-Image and NucleusMoE-Image). None should be merged as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12703 is the broadest and most central artifact here: it introduces the Z-Image series support and adds the core files and exports. It is the best representative of the only feature area that has a follow-up PR.", + "best_issue_reason": null, + "best_pr_reason": "PR 12703 is the best single representative because it is the foundational Z-Image implementation, while 12715 is a later bugfix/test follow-up and 12828/13317 are separate model additions.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12703", + "right": "pull_request:12715", + "accept": false, + "reason": "Related to the same Z-Image codebase, but not the same change: 12703 adds support, while 12715 fixes dtype/mask handling and tests. This is a follow-up bugfix PR, not a duplicate." + }, + { + "left": "pull_request:12703", + "right": "pull_request:12828", + "accept": false, + "reason": "Different underlying feature and model family: Z-Image versus LongCat-Image. Shared scaffolding files do not make them the same change." + }, + { + "left": "pull_request:12703", + "right": "pull_request:13317", + "accept": false, + "reason": "Different underlying feature and model family: Z-Image versus NucleusMoE-Image. They are not the same bug or change." + }, + { + "left": "pull_request:12715", + "right": "pull_request:12828", + "accept": false, + "reason": "Different model families and different code paths; 12715 is a Z-Image fix/test PR, while 12828 adds LongCat-Image support." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 86, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12724-10", + "nodes": [ + "pull_request:12724", + "pull_request:13031", + "pull_request:13044", + "pull_request:13083", + "pull_request:13085", + "pull_request:13122", + "pull_request:13127", + "pull_request:13130", + "pull_request:13160", + "pull_request:13198" + ], + "soft_pairs": [ + "pull_request:12724|pull_request:13031", + "pull_request:12724|pull_request:13044", + "pull_request:12724|pull_request:13083", + "pull_request:12724|pull_request:13085", + "pull_request:12724|pull_request:13122", + "pull_request:13044|pull_request:13083", + "pull_request:13044|pull_request:13085", + "pull_request:13044|pull_request:13122", + "pull_request:13044|pull_request:13127", + "pull_request:13044|pull_request:13130", + "pull_request:13083|pull_request:13085", + "pull_request:13083|pull_request:13122", + "pull_request:13083|pull_request:13127", + "pull_request:13083|pull_request:13130", + "pull_request:13085|pull_request:13122", + "pull_request:13085|pull_request:13127", + "pull_request:13085|pull_request:13130", + "pull_request:13122|pull_request:13127", + "pull_request:13122|pull_request:13130", + "pull_request:13127|pull_request:13130", + "pull_request:13044|pull_request:13160", + "pull_request:13083|pull_request:13160", + "pull_request:13085|pull_request:13160", + "pull_request:13122|pull_request:13160", + "pull_request:13127|pull_request:13160", + "pull_request:13044|pull_request:13198", + "pull_request:13083|pull_request:13198", + "pull_request:13085|pull_request:13198", + "pull_request:13122|pull_request:13198", + "pull_request:13127|pull_request:13198" + ], + "prepared_review_unit_hash": "e4445c6b6a408665a05bdb1bbca8ae0198fdfbe31bd2a1ecb43877bae5d1db89", + "component_budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 30, + "serialized_chars": 33362, + "estimated_input_tokens": 8341, + "estimated_eval_tokens": 16938 + }, + "budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 30, + "serialized_chars": 33362, + "estimated_input_tokens": 8341, + "estimated_eval_tokens": 16938 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a set of unrelated scheduler PRs that mostly share broad docs/typing cleanup patterns, not the same underlying change. No duplicate issue is present; the most central PR is a broad scheduler-docstring cleanup, but it is still distinct from the others.", + "confidence": 0.64, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:13083 is the broadest and most central scheduler-docstring cleanup in the set, touching several of the recurring DPMSolver-family files.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:13083 is the best representative of the cluster\u2019s theme because it spans multiple scheduler docstring edits, but it is not a duplicate of the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12724", + "right": "pull_request:13031", + "accept": false, + "reason": "Different targets: paper-link docs update across schedulers vs Flax deprecation warnings." + }, + { + "left": "pull_request:12724", + "right": "pull_request:13044", + "accept": false, + "reason": "Different scheduler files and different doc edits; only broad docs similarity." + }, + { + "left": "pull_request:12724", + "right": "pull_request:13083", + "accept": false, + "reason": "Both are scheduler docs PRs, but they change different files and different text." + }, + { + "left": "pull_request:12724", + "right": "pull_request:13085", + "accept": false, + "reason": "Docs link update vs code/typing changes in DPMSolver-SDE; not the same change." + }, + { + "left": "pull_request:12724", + "right": "pull_request:13122", + "accept": false, + "reason": "Different files and different doc content; no shared underlying bug/change." + }, + { + "left": "pull_request:13044", + "right": "pull_request:13083", + "accept": false, + "reason": "Both are docs-only scheduler PRs, but they edit different files and unrelated docstrings." + }, + { + "left": "pull_request:13044", + "right": "pull_request:13085", + "accept": false, + "reason": "One is docs cleanup; the other changes DPMSolver-SDE typing/runtime code." + }, + { + "left": "pull_request:13044", + "right": "pull_request:13122", + "accept": false, + "reason": "Different scheduler modules and different docstring updates." + }, + { + "left": "pull_request:13044", + "right": "pull_request:13127", + "accept": false, + "reason": "Different files and different docstring/typing changes; not mergeable as one PR." + }, + { + "left": "pull_request:13044", + "right": "pull_request:13130", + "accept": false, + "reason": "Different scheduler classes and distinct docstring edits." + }, + { + "left": "pull_request:13044", + "right": "pull_request:13160", + "accept": false, + "reason": "Different files; one updates a paper link, the other revises FlowMatchLCM docs/types." + }, + { + "left": "pull_request:13044", + "right": "pull_request:13198", + "accept": false, + "reason": "Different scheduler modules and unrelated docstring changes." + }, + { + "left": "pull_request:13083", + "right": "pull_request:13085", + "accept": false, + "reason": "Docstring cleanup in multiple schedulers vs DPMSolver-SDE typing/runtime edits." + }, + { + "left": "pull_request:13083", + "right": "pull_request:13122", + "accept": false, + "reason": "Overlap is only broad scheduler docs; these are separate files and separate edits." + }, + { + "left": "pull_request:13083", + "right": "pull_request:13127", + "accept": false, + "reason": "Different scheduler families and different docstring/type changes." + }, + { + "left": "pull_request:13083", + "right": "pull_request:13130", + "accept": false, + "reason": "Different files and different docstring adjustments; not the same underlying change." + }, + { + "left": "pull_request:13083", + "right": "pull_request:13160", + "accept": false, + "reason": "Different scheduler implementations and unrelated documentation changes." + }, + { + "left": "pull_request:13083", + "right": "pull_request:13198", + "accept": false, + "reason": "Distinct scheduler files with different docstring updates." + }, + { + "left": "pull_request:13085", + "right": "pull_request:13122", + "accept": false, + "reason": "Different code paths: DPMSolver-SDE typing changes vs EDM DPMSolver docs cleanup." + }, + { + "left": "pull_request:13085", + "right": "pull_request:13127", + "accept": false, + "reason": "Different scheduler implementations and different edit types." + }, + { + "left": "pull_request:13085", + "right": "pull_request:13130", + "accept": false, + "reason": "DPMSolver-SDE code/typing change is unrelated to FlowMatch-Heun docs edits." + }, + { + "left": "pull_request:13085", + "right": "pull_request:13160", + "accept": false, + "reason": "Different scheduler families and no shared concrete bug/change." + }, + { + "left": "pull_request:13085", + "right": "pull_request:13198", + "accept": false, + "reason": "Different modules and different change types; not duplicates." + }, + { + "left": "pull_request:13122", + "right": "pull_request:13127", + "accept": false, + "reason": "Shared docs theme only; these touch different schedulers and different text." + }, + { + "left": "pull_request:13122", + "right": "pull_request:13130", + "accept": false, + "reason": "Different scheduler classes and unrelated docstring changes." + }, + { + "left": "pull_request:13122", + "right": "pull_request:13160", + "accept": false, + "reason": "EDM DPMSolver docs vs FlowMatchLCM docs/types; different underlying changes." + }, + { + "left": "pull_request:13122", + "right": "pull_request:13198", + "accept": false, + "reason": "Different scheduler files and separate docstring edits." + }, + { + "left": "pull_request:13127", + "right": "pull_request:13130", + "accept": false, + "reason": "Different scheduler implementations and different APIs/types." + }, + { + "left": "pull_request:13127", + "right": "pull_request:13160", + "accept": false, + "reason": "Different files and different docstring/type updates; no common fix." + }, + { + "left": "pull_request:13127", + "right": "pull_request:13198", + "accept": false, + "reason": "Different schedulers and unrelated docstring changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 87, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12725-2", + "nodes": [ + "pull_request:12725", + "pull_request:12808" + ], + "soft_pairs": [ + "pull_request:12725|pull_request:12808" + ], + "prepared_review_unit_hash": "4eda8d772220ba783cdf39a77a87999fd74c136ebc0d293ecae918ccafb034c6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4853, + "estimated_input_tokens": 1214, + "estimated_eval_tokens": 2684 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4853, + "estimated_input_tokens": 1214, + "estimated_eval_tokens": 2684 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are both z-image-related PRs, but they address different changes: one is a RoPE implementation refactor inside the transformer, the other is the modular z-image pipeline addition and package exports. They should not be clustered as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12808 is the stronger canonical PR because it is the merged, broader z-image feature integration and includes the actual module/package wiring; #12725 is an earlier, narrower transformer refactor that did not merge.", + "best_issue_reason": null, + "best_pr_reason": "PR #12808 is the best representative of the cluster because it is merged, more complete, and central to the z-image modular pipeline work. PR #12725 is a separate internal refactor and is not the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12725", + "right": "pull_request:12808", + "accept": false, + "reason": "Both mention z-image, but the diffs target different code paths and purposes: #12725 changes RoPE application in the transformer, while #12808 adds modular pipeline plumbing and exports. They are not the same underlying fix/change and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 88, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12743-2", + "nodes": [ + "pull_request:12743", + "pull_request:12770" + ], + "soft_pairs": [ + "pull_request:12743|pull_request:12770" + ], + "prepared_review_unit_hash": "d327fcf1439e5ce8ada4e0924da24c4fe3ec4c6da9996c01fbc532d289eb6ea1", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4345, + "estimated_input_tokens": 1087, + "estimated_eval_tokens": 2430 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4345, + "estimated_input_tokens": 1087, + "estimated_eval_tokens": 2430 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs touch the same file but fix different empty-tensor edge cases in different code paths: PR 12743 guards image padding position-id creation, while PR 12770 adjusts caption padding mask handling for TPU/XLA repeat behavior. They are related but not duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 12770 is the stronger canonical PR because it is merged, has a clear concrete bug fix, and the diff directly addresses the empty-dimension/XLA repeat failure path.", + "best_issue_reason": null, + "best_pr_reason": "PR 12770 is the best representative because it is merged and fixes the more explicit compatibility bug; PR 12743 appears to be a separate unmerged padding fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12743", + "right": "pull_request:12770", + "accept": false, + "reason": "Different underlying bugs: 12743 changes image padding position IDs when image_padding_len > 0, while 12770 changes caption pad-mask handling for empty dimensions / TPU repeat compatibility. Same file and theme, but not the same concrete fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 89, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12762-2", + "nodes": [ + "pull_request:12762", + "pull_request:13103" + ], + "soft_pairs": [ + "pull_request:12762|pull_request:13103" + ], + "prepared_review_unit_hash": "4e485bc26adeeb6e9a48b02ccfc6f7e23f2e4488c4695a8b6995e0171bc5a8a1", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3202, + "estimated_input_tokens": 801, + "estimated_eval_tokens": 1858 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3202, + "estimated_input_tokens": 801, + "estimated_eval_tokens": 1858 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address the same bug in `from_pipe()` where a new pipeline was being forced to `float32` instead of preserving the source pipeline\u2019s dtype. They touch the same file, target the same issue, and the second PR is a clearer direct implementation of the same fix.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 13103 is the cleaner canonical PR: it directly sets the default to `pipeline.dtype` and clearly states the intended behavior, while PR 12762 uses `None` as an intermediate mechanism to achieve the same outcome.", + "best_issue_reason": null, + "best_pr_reason": "PR 13103 is the best representative because it most explicitly and directly fixes the dtype-preservation behavior in `from_pipe()` with minimal ambiguity.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12762", + "right": "pull_request:13103", + "accept": true, + "reason": "Same concrete bug in `from_pipe()` defaulting to float32; both PRs aim to preserve the source pipeline dtype and could be merged into one fix." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both PRs target the same issue, touch the same file, and each changes the `from_pipe()` default away from `torch.float32` to preserve the source pipeline dtype. The soft-edge verdict is conservative enough, since the two diffs look like alternative implementations of the same bugfix and appear mergeable into one PR." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 90, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12784-5", + "nodes": [ + "pull_request:12784", + "pull_request:13520", + "pull_request:13522", + "pull_request:13524", + "pull_request:13528" + ], + "soft_pairs": [ + "pull_request:12784|pull_request:13520", + "pull_request:13520|pull_request:13522", + "pull_request:13520|pull_request:13524", + "pull_request:13524|pull_request:13528" + ], + "prepared_review_unit_hash": "4336a87c9508e5e13bd5daf147961446f64fc8c6254df0649b5aa147ac4f6326", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 9725, + "estimated_input_tokens": 2432, + "estimated_eval_tokens": 5120 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 4, + "serialized_chars": 9725, + "estimated_input_tokens": 2432, + "estimated_eval_tokens": 5120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are mostly unrelated example-script fixes. Two PRs in the same Flux2 Dreambooth script point at the same AttributeError theme but propose conflicting arg names, so they are not duplicates; the diffusion_dpo, onnxruntime, and textual_inversion changes are separate bugs in different code paths.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the items address different scripts and different bugs, and the Flux2 pair is conflicting rather than a mergeable duplicate.", + "best_issue_reason": null, + "best_pr_reason": "No strong global best PR for deduplication; 12784 and 13520 are the closest in wording but their fixes conflict, while the others are unrelated.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12784", + "right": "pull_request:13520", + "accept": false, + "reason": "Same file and similar title, but they change opposite sides of the argument-name mismatch (`validation_image_path` vs `validation_image`). They are not the same mergeable change." + }, + { + "left": "pull_request:13520", + "right": "pull_request:13522", + "accept": false, + "reason": "Different example scripts and different bugs: one fixes a validation-image attribute error, the other fixes `beta_dpo` usage in the IPO loss calculation." + }, + { + "left": "pull_request:13520", + "right": "pull_request:13524", + "accept": false, + "reason": "Unrelated code paths in different scripts; one is Flux2 Dreambooth validation, the other is onnxruntime logging/argument handling." + }, + { + "left": "pull_request:13524", + "right": "pull_request:13528", + "accept": false, + "reason": "Different scripts and different fixes: one changes logger/reporting checks, the other adds a missing `--report_to` argument to textual inversion bf16." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 91, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12788-3", + "nodes": [ + "pull_request:12788", + "pull_request:13239", + "pull_request:13240" + ], + "soft_pairs": [ + "pull_request:13239|pull_request:13240", + "pull_request:12788|pull_request:13239", + "pull_request:12788|pull_request:13240" + ], + "prepared_review_unit_hash": "e7e26e42da3248c3e4506b502e042b7a299e747e2e0687348b74e20b2b669b41", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7373, + "estimated_input_tokens": 1844, + "estimated_eval_tokens": 3944 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7373, + "estimated_input_tokens": 1844, + "estimated_eval_tokens": 3944 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "#13239 and #13240 are near-identical fixes for the same DPMSolverMultistep timestep-duplication bug. #12788 is related to the same scheduler and issue target, but the evidence only supports loose similarity; it is not strong enough to treat it as the same concrete fix or fold it into the same PR cluster.", + "confidence": 0.66, + "canonical_issue_reason": null, + "canonical_pr_reason": "#13240 is the cleanest representative of the near-identical pair: it cleanly targets the timestep rounding/duplication logic in the shared scheduler code with the most focused patch.", + "best_issue_reason": null, + "best_pr_reason": "#13240 is the best overall representative because it is the most focused and clearly scoped version of the shared fix, while #13239 is effectively a close variant.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13239", + "right": "pull_request:13240", + "accept": true, + "reason": "Same file, same issue target, and effectively the same timestep-rounding fix for the same scheduler code path; these look like near-identical variants of one change." + }, + { + "left": "pull_request:12788", + "right": "pull_request:13239", + "accept": false, + "reason": "They share the issue target and file, but the evidence is only soft similarity and the patch in #12788 is broader/different enough that it is not clearly the same concrete fix." + }, + { + "left": "pull_request:12788", + "right": "pull_request:13240", + "accept": false, + "reason": "Shared scheduler context and issue target are not enough here; #12788 does not look confidently mergeable as the same underlying bugfix as #13240." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary overstates the evidence for PR #12788: the packet only shows shared issue/file and generic soft similarity with #13239/#13240, while the pairwise signals are explicitly non-deterministic for those comparisons. It is safe to group #13239 and #13240 as near-identical fixes, but not to claim all three PRs are the same underlying bugfix or that #12788 could clearly be folded into the same PR. The soft-edge verdicts for #12788 should be more conservative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 92, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12865-2", + "nodes": [ + "pull_request:12865", + "pull_request:12866" + ], + "soft_pairs": [ + "pull_request:12865|pull_request:12866" + ], + "prepared_review_unit_hash": "b496d937306ebfff8984152463b574284139ddca08f2332bf56971366458d75a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5259, + "estimated_input_tokens": 1315, + "estimated_eval_tokens": 2886 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5259, + "estimated_input_tokens": 1315, + "estimated_eval_tokens": 2886 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related workflow-maintenance PRs, but they are not duplicates: one is a broad Node 24 compatibility upgrade across many GitHub Actions, while the other is a smaller follow-up that bumps a few specific actions to newer major versions.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12865 is the broader, more central workflow-actions compatibility upgrade and covers the main Node 24 migration across many workflows.", + "best_issue_reason": null, + "best_pr_reason": "PR #12865 is the best representative of the cluster because it addresses the larger underlying maintenance change; #12866 is narrower and partially overlapping.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12865", + "right": "pull_request:12866", + "accept": false, + "reason": "They overlap in workflow files, but the concrete changes differ: #12865 upgrades many actions for Node 24 compatibility, while #12866 separately bumps Docker and typos actions to newer versions. They are related maintenance PRs, not the same fix/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 93, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12869-2", + "nodes": [ + "pull_request:12869", + "pull_request:12870" + ], + "soft_pairs": [ + "pull_request:12869|pull_request:12870" + ], + "prepared_review_unit_hash": "2908506d9b3ba1841d1fe9c1f6e0a79572bdf45683c4c950dccbeb15eec6209c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5650, + "estimated_input_tokens": 1413, + "estimated_eval_tokens": 3082 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5650, + "estimated_input_tokens": 1413, + "estimated_eval_tokens": 3082 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about the same QwenImage split-attention work. PR #12870 is the cleaner, focused implementation; PR #12869 is a draft and includes broader docs/benchmarking plus related pipeline edits.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #12870 is the best canonical PR because it is the more focused code change: it updates attention backend dispatch and the QwenImage transformer/pipeline support without the extra draft/docs churn seen in #12869.", + "best_issue_reason": null, + "best_pr_reason": "PR #12870 is the best representative of the cluster since it targets the core backend implementation directly and is less noisy than #12869, which mixes in documentation and performance notes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12869", + "right": "pull_request:12870", + "accept": true, + "reason": "Strong title and filename overlap around QwenImage split attention backends, including shared core files. #12869 looks like the broader draft/documentation-heavy version of the same feature work that #12870 continues in a more focused form." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 94, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12876-2", + "nodes": [ + "pull_request:12876", + "pull_request:13257" + ], + "soft_pairs": [ + "pull_request:12876|pull_request:13257" + ], + "prepared_review_unit_hash": "732acf7eda77575adc423dfb983dcbc51e4960182d74964f4d70aa983519f7ad", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3303, + "estimated_input_tokens": 826, + "estimated_eval_tokens": 1908 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3303, + "estimated_input_tokens": 826, + "estimated_eval_tokens": 1908 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are both small additions to `AUTO_TEXT2IMAGE_PIPELINES_MAPPING`, but they add different pipeline registrations (`OvisImagePipeline` vs `PRXPipeline`). They touch the same file and pattern, yet they fix distinct missing entries and are not the same underlying change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: each PR registers a different pipeline class in `auto_pipeline.py`, so they are separate changes rather than duplicates.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a duplicate of the other; if forced to pick a representative, PR 12876 is the earlier, simpler mapping addition, but it does not subsume PR 13257.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12876", + "right": "pull_request:13257", + "accept": false, + "reason": "Same subsystem and same file, but different concrete mappings/classes. They do not fix the same code-path problem and would not be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 95, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-12978-2", + "nodes": [ + "pull_request:12978", + "pull_request:13051" + ], + "soft_pairs": [ + "pull_request:12978|pull_request:13051" + ], + "prepared_review_unit_hash": "3dd7d0b3eb1d18264d32db8dbfab12c45cf048b9bf6799270bdaaac5ad7d696a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4563, + "estimated_input_tokens": 1141, + "estimated_eval_tokens": 2538 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4563, + "estimated_input_tokens": 1141, + "estimated_eval_tokens": 2538 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two Mellon-related PRs overlap on one utility file, but they are not the same change: one is a narrow schema tweak, the other is a broader docs/integration update.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 13051 is the better canonical PR because it is the more complete Mellon's work item, spanning code plus docs and related pipeline changes, while 12978 is a smaller utility-only tweak.", + "best_issue_reason": null, + "best_pr_reason": "PR 13051 is the strongest representative of the cluster\u2019s theme: it covers the broader Mellon's feature/doc integration and is more suitable as the primary reference than the narrowly scoped 12978.", + "soft_edge_verdicts": [ + { + "left": "pull_request:12978", + "right": "pull_request:13051", + "accept": false, + "reason": "Shared filename alone is not enough: 12978 only adds a new optional field and excludes it from serialization, while 13051 is a broader Mellon's docs/integration PR touching multiple files. They do not look like the same concrete bug fix or a single mergeable change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 96, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13048-2", + "nodes": [ + "pull_request:13048", + "pull_request:13405" + ], + "soft_pairs": [ + "pull_request:13048|pull_request:13405" + ], + "prepared_review_unit_hash": "b96bc58301700ea3d2505150ef1b581650b13e626919aae44916d77c30fd9ed5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4421, + "estimated_input_tokens": 1106, + "estimated_eval_tokens": 2468 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4421, + "estimated_input_tokens": 1106, + "estimated_eval_tokens": 2468 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are both torchao-related cleanups, but they target different changes: #13048 removes autoquant docs/references, while #13405 removes AffineQuantizedTensor-specific repr handling and updates tests. The shared file is not enough to treat them as the same underlying change.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue is present; as a canonical PR, #13405 is the more code-path-focused change because it alters torchao quantizer behavior and tests, whereas #13048 is primarily documentation/cleanup.", + "best_issue_reason": null, + "best_pr_reason": "#13405 is the best representative PR for the cluster because it makes the substantive quantizer implementation change, while #13048 is a separate docs-focused cleanup.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13048", + "right": "pull_request:13405", + "accept": false, + "reason": "Different underlying changes: autoquant doc/reference removal versus AffineQuantizedTensor repr cleanup in the quantizer. They do not fix the same concrete bug and would not naturally merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 97, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13094-2", + "nodes": [ + "pull_request:13094", + "pull_request:13183" + ], + "soft_pairs": [ + "pull_request:13094|pull_request:13183" + ], + "prepared_review_unit_hash": "f750981785ddaed591804927101ea6d2d226a08954383712796a8a3adeadcdaf", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3680, + "estimated_input_tokens": 920, + "estimated_eval_tokens": 2096 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3680, + "estimated_input_tokens": 920, + "estimated_eval_tokens": 2096 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs make the same fix in `ZImagePipeline`: change `do_classifier_free_guidance` from `> 1` to `> 0` in `src/diffusers/pipelines/z_image/pipeline_z_image.py`. They appear to be duplicate implementations of the same concrete bug fix, with #13183 being the merged version.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13183 is the best canonical PR because it is the merged, closed implementation of the exact same code change and has slightly richer completion signals (review/inbound reference).", + "best_issue_reason": null, + "best_pr_reason": "PR #13183 is the best representative PR for this fix because it is already merged and directly resolves the threshold bug in the affected code path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13094", + "right": "pull_request:13183", + "accept": true, + "reason": "Same file, same function, same one-line threshold change, and both target the same underlying Z-Image guidance bug; they are effectively duplicate PRs for the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 98, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13102-3", + "nodes": [ + "pull_request:13102", + "pull_request:13128", + "pull_request:13136" + ], + "soft_pairs": [ + "pull_request:13102|pull_request:13136", + "pull_request:13128|pull_request:13136", + "pull_request:13102|pull_request:13128" + ], + "prepared_review_unit_hash": "7919ac1ae48cfb484de2f67f51ea0f1c935933980b77774d0043ddefad67c22c", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7360, + "estimated_input_tokens": 1840, + "estimated_eval_tokens": 3936 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7360, + "estimated_input_tokens": 1840, + "estimated_eval_tokens": 3936 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The cluster centers on ZImageControlNet.from_transformer, but only 13102 and 13128 are true duplicate fixes for the weight-sharing bug; 13136 takes a different freezing-based approach.", + "confidence": 0.83, + "canonical_issue_reason": null, + "canonical_pr_reason": "13128 is the strongest representative: it replaces shared module assignment with a state_dict-based reconstruction and adds tests, making it the most complete fix for the weight-sharing bug.", + "best_issue_reason": null, + "best_pr_reason": "13128 is the best PR to represent the cluster because it addresses the same bug as the other duplicate fix, but with a more robust implementation and test coverage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13102", + "right": "pull_request:13128", + "accept": true, + "reason": "Same underlying bug in from_transformer: both aim to stop controlnet/transformer weight sharing. 13128 is just a more complete implementation of that fix." + }, + { + "left": "pull_request:13102", + "right": "pull_request:13136", + "accept": false, + "reason": "13102 removes shared references, while 13136 intentionally keeps sharing and freezes modules instead. Different concrete change and not the same fix." + }, + { + "left": "pull_request:13128", + "right": "pull_request:13136", + "accept": false, + "reason": "13128 reconstructs modules to avoid sharing; 13136 keeps shared modules and adds freezing logic. These are different remedies for the same area, not the same change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: all three PRs target the same file and issue, and the distinction between 13102/13128 (two ways to eliminate weight sharing) versus 13136 (intentional sharing plus freezing) is conservative. The soft-edge decisions are supported by the diffs and titles, and the representative PR choice for 13128 is justified by the added state_dict-based reconstruction and tests." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 99, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13108-2", + "nodes": [ + "pull_request:13108", + "pull_request:13125" + ], + "soft_pairs": [ + "pull_request:13108|pull_request:13125" + ], + "prepared_review_unit_hash": "b462529f9a7f18530752d5788489633245cffc4a4fa7243a7e2e5da8493f05b2", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4296, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4296, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two related PRs fix an incorrect prev_timestep calculation in scheduler step logic, but they target different scheduler implementations and are not the same concrete change.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13108 is the cleaner canonical PR: it has an explicit issue target, a narrowly scoped one-file fix, and directly addresses the bug in DDIMScheduler.step().", + "best_issue_reason": null, + "best_pr_reason": "PR #13108 is the best representative of the cluster because it is the most specific fix and has the strongest linkage to a tracked issue; #13125 is a separate scheduler-port of the same pattern.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13108", + "right": "pull_request:13125", + "accept": false, + "reason": "They share the same bug pattern, but they change different scheduler code paths (DDIM vs CogVideoX/DPM CogVideoX) and would not plausibly be merged into one PR without broadening scope." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 100, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13109-2", + "nodes": [ + "pull_request:13109", + "pull_request:13517" + ], + "soft_pairs": [ + "pull_request:13109|pull_request:13517" + ], + "prepared_review_unit_hash": "c587fd507c56a5ac432a7b5567a2a86c56d1e06e61c6b71943bdac03275a816d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4301, + "estimated_input_tokens": 1076, + "estimated_eval_tokens": 2408 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4301, + "estimated_input_tokens": 1076, + "estimated_eval_tokens": 2408 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs touch Sana guidance handling, but they address different failure points and are not safe duplicates: one defaults missing guidance inside the embedding module, while the other fixes forward-path dispatch and explicit validation when `guidance_embeds=True`.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13517 is the better canonical PR: it targets the model\u2019s actual forward dispatch bug for `guidance_embeds=True` and handles the non-guidance pipeline mismatch explicitly, rather than only papering over missing guidance inside one helper.", + "best_issue_reason": null, + "best_pr_reason": "PR #13517 is the best fit because it addresses the concrete code-path problem in `SanaTransformer2DModel.forward` and is the more complete, safer fix for the crash scenario.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13109", + "right": "pull_request:13517", + "accept": false, + "reason": "Related theme, but not the same change: #13109 makes `SanaCombinedTimestepGuidanceEmbeddings` tolerate missing guidance by defaulting to ones, while #13517 changes the transformer forward routing and raises when `guidance_embeds=True` without guidance. They fix different layers and could conflict in behavior." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 101, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13110-2", + "nodes": [ + "pull_request:13110", + "pull_request:13516" + ], + "soft_pairs": [ + "pull_request:13110|pull_request:13516" + ], + "prepared_review_unit_hash": "e91fa0e881a0fd250df3d84cf74084f0c5dacf378046e87d60e460859553a71f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4179, + "estimated_input_tokens": 1045, + "estimated_eval_tokens": 2346 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4179, + "estimated_input_tokens": 1045, + "estimated_eval_tokens": 2346 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests implement the same fix for AuraFlow\u2019s positional-embedding bounds bug: they add a guard that rejects latent sizes larger than the pretrained grid to prevent out-of-range indexing / CUDA asserts. They should be treated as duplicates, with one canonical PR.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 13516 is the stronger canonical choice because it contains the same bounds check plus a fuller explanation of the crash mode and failure context for oversized latents.", + "best_issue_reason": null, + "best_pr_reason": "PR 13516 best represents the cluster because it targets the same concrete code path, explains the CUDA/device-side-assert failure, and adds the protective validation needed to fix it.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13110", + "right": "pull_request:13516", + "accept": true, + "reason": "Same file, same issue target, and same underlying fix: reject latents larger than the positional embedding grid before centered indexing can go out of range. These are duplicate PRs for the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 102, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13118-3", + "nodes": [ + "pull_request:13118", + "pull_request:13188", + "pull_request:13189" + ], + "soft_pairs": [ + "pull_request:13118|pull_request:13188", + "pull_request:13118|pull_request:13189", + "pull_request:13188|pull_request:13189" + ], + "prepared_review_unit_hash": "013514ac4172069ffb140e210b516b97a30bdb8e979ffcb260dc7578dface92c", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7235, + "estimated_input_tokens": 1809, + "estimated_eval_tokens": 3874 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7235, + "estimated_input_tokens": 1809, + "estimated_eval_tokens": 3874 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are all about Kohya LoRA loading in the same conversion code, but they split into two concrete fixes: a lora_te1_ prefix handling fix and a Flux.2/Musubi tuner conversion fix. Only the Flux.2-focused pair looks like the same underlying change; the TE-prefix PR is a related but separate bugfix.", + "confidence": 0.74, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13118 is the broadest umbrella for the shared Flux/Kohya loader work and explicitly covers both the TE-prefix and Flux.2 dev cases, so it best represents the cluster theme.", + "best_issue_reason": null, + "best_pr_reason": "PR #13118 is the best single representative because it spans both concrete loader fixes in the same code path, while the other two PRs are narrower slices of that work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13118", + "right": "pull_request:13188", + "accept": false, + "reason": "Related area, but not the same underlying change: #13118 bundles Flux.2 conversion changes as well, while #13188 is specifically the lora_te1_ prefix fix." + }, + { + "left": "pull_request:13118", + "right": "pull_request:13189", + "accept": true, + "reason": "Both target the Flux.2/Kohya loader conversion path and the diff suggests the same code-path fix around Flux.2 support; they look like overlapping implementations of the same concrete change." + }, + { + "left": "pull_request:13188", + "right": "pull_request:13189", + "accept": false, + "reason": "They address different bugs: one is TE-prefix recognition, the other is Flux.2/Musubi tuner loading. Same subsystem, but not the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 103, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13182-2", + "nodes": [ + "pull_request:13182", + "pull_request:13457" + ], + "soft_pairs": [ + "pull_request:13182|pull_request:13457" + ], + "prepared_review_unit_hash": "675c6c3b2dde9c57eeb917a722a6fee9587b7741f0808947607caff6e750461a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4458, + "estimated_input_tokens": 1115, + "estimated_eval_tokens": 2486 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4458, + "estimated_input_tokens": 1115, + "estimated_eval_tokens": 2486 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests are unrelated: one fixes attention backend behavior in core model code, the other patches a specific PyTorch/XLA Flux inference example for TPU execution.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:13182", + "right": "pull_request:13457", + "accept": false, + "reason": "Different bugs and code paths: PR 13182 changes core attention dispatch logic to handle LSE shape across flash/cudnn backends, while PR 13457 only adjusts a TPU inference example and XLA step handling for Flux. They do not look like the same underlying change and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 104, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13213-3", + "nodes": [ + "pull_request:13213", + "pull_request:13214", + "pull_request:13222" + ], + "soft_pairs": [ + "pull_request:13213|pull_request:13214", + "pull_request:13213|pull_request:13222", + "pull_request:13214|pull_request:13222" + ], + "prepared_review_unit_hash": "7ab3f441fe6c9886af7b2d403b7eb53cf4f368b9668324b9402a417dc84e3c83", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6369, + "estimated_input_tokens": 1593, + "estimated_eval_tokens": 3442 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6369, + "estimated_input_tokens": 1593, + "estimated_eval_tokens": 3442 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate Helios pull requests: a documentation paper-link fix, a code-path float conversion fix, and a broader docs optimization update. They are related by topic but do not look like the same underlying change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 13222 is the broadest Helios docs change and shares the main pipeline documentation file with 13213, so it is the best single representative of the cluster. It still does not subsume the code fix in 13214.", + "best_issue_reason": null, + "best_pr_reason": "PR 13222 is the best representative artifact here because it is the largest and most central documentation update, but it should not be treated as a duplicate of the other two PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13213", + "right": "pull_request:13214", + "accept": false, + "reason": "Different changes: docs paper-link cleanup vs. a float-casting fix in optimized_scale. They touch different files and different code paths." + }, + { + "left": "pull_request:13213", + "right": "pull_request:13222", + "accept": false, + "reason": "Both are Helios docs changes, but one fixes a paper link while the other changes memory-usage guidance and example code. Not the same underlying change." + }, + { + "left": "pull_request:13214", + "right": "pull_request:13222", + "accept": false, + "reason": "One is a code fix in the pipeline implementation; the other is documentation optimization guidance. Shared subsystem alone is not enough to merge them as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 105, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13230-2", + "nodes": [ + "pull_request:13230", + "pull_request:13233" + ], + "soft_pairs": [ + "pull_request:13230|pull_request:13233" + ], + "prepared_review_unit_hash": "73ca127fa26cad86f61cd7cad333160d94c497a7903f01c564c02fe7fb47e9be", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2480, + "estimated_input_tokens": 620, + "estimated_eval_tokens": 1496 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2480, + "estimated_input_tokens": 620, + "estimated_eval_tokens": 1496 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs add `permissions: contents: read` to different CI workflows to satisfy workflow-permissions scanning, but they are separate workflow files and not the same concrete change.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13233 is the cleaner representative: it targets the main PR test workflow directly and uses a broader, more general title for the permissions fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #13233 best captures the underlying pattern of adding workflow permissions to PR tests, whereas #13230 is a narrower GPU-workflow-specific variant.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13230", + "right": "pull_request:13233", + "accept": false, + "reason": "Same class of workflow-permissions fix, but applied to different workflow files with different scopes; they are not the same underlying change and would not naturally merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 106, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13246-2", + "nodes": [ + "pull_request:13246", + "pull_request:13512" + ], + "soft_pairs": [ + "pull_request:13246|pull_request:13512" + ], + "prepared_review_unit_hash": "2076e6f2a419f6e2e56946e9cb015062e3f298d9e8975317d464c3f27395a67d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4469, + "estimated_input_tokens": 1118, + "estimated_eval_tokens": 2492 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4469, + "estimated_input_tokens": 1118, + "estimated_eval_tokens": 2492 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs fix the same FlowMatchEulerDiscreteScheduler bug: sigma_min/sigma_max were being stored after shifting, which caused set_timesteps to apply the shift twice. PR #13246 is the stronger canonical candidate because it includes tests and is the more complete patch; PR #13512 is a near-identical follow-up/duplicate on the same code path.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13246 is the best canonical PR: it fixes the bug in the scheduler and includes the accompanying test file, making it the more complete and representative change.", + "best_issue_reason": null, + "best_pr_reason": "PR #13246 is the best overall PR because it contains the core code fix plus tests, whereas #13512 is the same fix with only a slightly different explanation/commenting and no broader completeness.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13246", + "right": "pull_request:13512", + "accept": true, + "reason": "Accept: both PRs modify the same scheduler to move sigma_min/sigma_max assignment before shifting, addressing the same double-shift bug in the same code path." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: both PRs touch the same scheduler file, target the same issue, and implement the same unshifted sigma_min/sigma_max fix. The soft-edge acceptance is conservative enough, and choosing #13246 as canonical because it also includes tests is supported by the evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 107, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13250-2", + "nodes": [ + "pull_request:13250", + "pull_request:13260" + ], + "soft_pairs": [ + "pull_request:13250|pull_request:13260" + ], + "prepared_review_unit_hash": "a25c6036e163c2cc5a13732c6c36aefebb89c59bd5126f0ccebb0a9f2d549fe5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4040, + "estimated_input_tokens": 1010, + "estimated_eval_tokens": 2276 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4040, + "estimated_input_tokens": 1010, + "estimated_eval_tokens": 2276 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two separate PRs in the same LoRA conversion helper: one adds LoKr/Z-image key support, the other makes alpha-key lookup tolerant of missing keys. They address different bugs and should not be deduplicated.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "13250 is the stronger canonical PR for this cluster: it has the more concrete feature/bug fix, explicit issue linkage, and a clearer end-to-end code-path change. 13260 is a separate defensive fix in the same helper.", + "best_issue_reason": null, + "best_pr_reason": "13250 is the best representative of the cluster because it is the more substantial and specific change, with direct LoKr format handling rather than a small guard around alpha lookup.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13250", + "right": "pull_request:13260", + "accept": false, + "reason": "Different underlying problems in the same file: 13250 adds LoKr/Z-image format conversion support, while 13260 only changes missing-alpha handling. They would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 108, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13272-2", + "nodes": [ + "pull_request:13272", + "pull_request:13513" + ], + "soft_pairs": [ + "pull_request:13272|pull_request:13513" + ], + "prepared_review_unit_hash": "48d81b9f9d08f65442c1ebd5f13edcc71be7dd97dd5bb1d59a3a9c53b3264bfd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4362, + "estimated_input_tokens": 1091, + "estimated_eval_tokens": 2438 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4362, + "estimated_input_tokens": 1091, + "estimated_eval_tokens": 2438 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests address the same Lumina2 bug: `image_seq_len` was incorrectly derived from latent channels instead of spatial patch tokens. The fixes are effectively the same code change against the same file and issue target, so they should be clustered together.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13513 is the cleaner canonical PR: it makes the same fix as #13272, includes a more explicit explanation of the bug, and is the later refinement of the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR #13513 is the best representative because it is the same functional fix with clearer rationale and no extra unrelated scope.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13272", + "right": "pull_request:13513", + "accept": true, + "reason": "Same underlying Lumina2 bug, same explicit issue target, same file, and the diff changes the same computation of `image_seq_len` from channels to spatial patch tokens. The second PR is essentially a rewrite/commented version of the first, so they could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 109, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13273-2", + "nodes": [ + "pull_request:13273", + "pull_request:13510" + ], + "soft_pairs": [ + "pull_request:13273|pull_request:13510" + ], + "prepared_review_unit_hash": "3e1663c393bb740e2679c854980fb6473e94a49f3851b0d126cc05452681b742", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5213, + "estimated_input_tokens": 1304, + "estimated_eval_tokens": 2864 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5213, + "estimated_input_tokens": 1304, + "estimated_eval_tokens": 2864 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two DreamBooth PRs share the same tracking issue, but they fix different bugs in the validation path: one prevents gradient tracking during validation, the other restores fp32 training params after validation mutates model dtype. Not duplicates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13510 is the more concrete, failure-specific fix: it directly addresses the fp16 crash after validation by re-upcasting trainable params after `log_validation` mutates model dtype.", + "best_issue_reason": null, + "best_pr_reason": "PR #13510 is the best representative of the cluster because it targets the explicit crash mode described in its title and diff, rather than a broader validation hygiene change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13273", + "right": "pull_request:13510", + "accept": false, + "reason": "Related through the same issue target, but not the same underlying bug. #13273 wraps validation in `torch.no_grad()`, while #13510 fixes dtype corruption after validation causes fp16 training to crash. Different code paths and failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 110, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13289-2", + "nodes": [ + "pull_request:13289", + "pull_request:13485" + ], + "soft_pairs": [ + "pull_request:13289|pull_request:13485" + ], + "prepared_review_unit_hash": "89b635b7200fead473031d06fab7f7c60554b3fa918e03068936fdee4a2978e7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6149, + "estimated_input_tokens": 1538, + "estimated_eval_tokens": 3332 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6149, + "estimated_input_tokens": 1538, + "estimated_eval_tokens": 3332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two highly overlapping Neuron PRs touch mostly the same files and some near-identical setup code, but the titles suggest different goals: broad Trainium/Inferentia device support vs torch.compile compatibility. They look related, not the same concrete change.", + "confidence": 0.77, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 13289 is the broader umbrella change for official Neuron device support and is the better representative of this cluster; 13485 reads like a narrower follow-up around compile compatibility.", + "best_issue_reason": null, + "best_pr_reason": "PR 13289 is the most canonical-looking artifact because it frames the overall device support effort and appears to subsume the shared infrastructure changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13289", + "right": "pull_request:13485", + "accept": false, + "reason": "High file overlap, but the PRs target different outcomes and are not clearly the same concrete fix; they could be related follow-ups rather than duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 111, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13307-3", + "nodes": [ + "pull_request:13307", + "pull_request:13396", + "pull_request:13415" + ], + "soft_pairs": [ + "pull_request:13307|pull_request:13396", + "pull_request:13307|pull_request:13415", + "pull_request:13396|pull_request:13415" + ], + "prepared_review_unit_hash": "ff63e04a78dfe8512cdbf072af7114e332ee53261bec3230448b8276fdd60e6a", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7216, + "estimated_input_tokens": 1804, + "estimated_eval_tokens": 3864 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7216, + "estimated_input_tokens": 1804, + "estimated_eval_tokens": 3864 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are all about the same DreamBooth prior-preservation repeat-count bug pattern, but only the two Flux2 PRs are close enough to be duplicate candidates. The Qwen-image PR fixes a separate script/code path and should stay separate.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13415 is the best canonical PR: it is merged, directly addresses the Flux2 repeat bug, and is the cleanest representative of the fix in this cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR #13415 is the strongest overall representative because it is merged and narrowly fixes the same concrete Flux2 prior-preservation repeat logic without extra scope.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13307", + "right": "pull_request:13396", + "accept": false, + "reason": "Same bug pattern, but different training script/code path (Flux2 vs Qwen-image). Too broad to treat as the same PR-level duplicate." + }, + { + "left": "pull_request:13307", + "right": "pull_request:13415", + "accept": true, + "reason": "Both fix the Flux2 DreamBooth prior-preservation repeat-count bug in the same code path, and the changes are plausibly mergeable into one PR." + }, + { + "left": "pull_request:13396", + "right": "pull_request:13415", + "accept": false, + "reason": "They address similar behavior, but in different example scripts with different implementations; not the same concrete change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The summary correctly distinguishes the shared prior-preservation repeat bug pattern from PR-level duplicate scope, and the soft-edge verdicts are cautious: Flux2 PRs are plausibly mergeable as the same fix, while the Qwen-image PR is a separate script/code path. No evidence is overstated." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 112, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1334-2", + "nodes": [ + "pull_request:1334", + "pull_request:1426" + ], + "soft_pairs": [ + "pull_request:1334|pull_request:1426" + ], + "prepared_review_unit_hash": "02f12c7b984493089235690154053c2deac2dba99928dd7379050e94ad88561c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6506, + "estimated_input_tokens": 1627, + "estimated_eval_tokens": 3510 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6506, + "estimated_input_tokens": 1627, + "estimated_eval_tokens": 3510 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests are effectively the same feature branch: PR #1426 is a rebased/updated version of PR #1334, with the same title, nearly identical file set, and the same audio pipeline additions. They should be clustered together as one underlying change.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1426 is the best canonical PR because it is the later merged version of the same change set, while #1334 is the earlier unmerged predecessor.", + "best_issue_reason": null, + "best_pr_reason": "PR #1426 is the strongest representative because it is the finalized merged implementation of the audio pipeline feature and supersedes #1334.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1334", + "right": "pull_request:1426", + "accept": true, + "reason": "Same underlying change: same feature title, almost identical modified files, and #1426 appears to be the updated/merged continuation of #1334." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 113, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13354-2", + "nodes": [ + "pull_request:13354", + "pull_request:13374" + ], + "soft_pairs": [ + "pull_request:13354|pull_request:13374" + ], + "prepared_review_unit_hash": "accbd0f4b78ffafe0ed104fd593cbcd8d6ed364198e84542ed855866ffebfc84", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4168, + "estimated_input_tokens": 1042, + "estimated_eval_tokens": 2340 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4168, + "estimated_input_tokens": 1042, + "estimated_eval_tokens": 2340 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related workflow-maintenance PRs on the same file/title, but they are not the same underlying change. PR #13354 is a smaller Claude action/workflow tweak, while PR #13374 makes a different, broader update (checkout behavior, config sanitization, and fetching the PR diff) and was merged.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13374 is the better canonical PR because it is the merged, more complete implementation of a later workflow update, whereas #13354 is an earlier, smaller variant with different edits.", + "best_issue_reason": null, + "best_pr_reason": "PR #13374 is the best PR in this cluster: it is merged and represents the fuller workflow change set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13354", + "right": "pull_request:13374", + "accept": false, + "reason": "Same file and similar title, but the diffs are materially different: #13354 changes permissions, pins the action, and edits the prompt; #13374 changes checkout flow, restores base config, and adds a PR diff step. They do not look mergeable as one single fix/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 114, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13378-2", + "nodes": [ + "pull_request:13378", + "pull_request:13389" + ], + "soft_pairs": [ + "pull_request:13378|pull_request:13389" + ], + "prepared_review_unit_hash": "5fa8cfe709386c7d890e6b1a39ce9134999dca34e49284bef3046f0f28001e44", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6034, + "estimated_input_tokens": 1509, + "estimated_eval_tokens": 3274 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6034, + "estimated_input_tokens": 1509, + "estimated_eval_tokens": 3274 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two distinct feature PRs: one adds an LTX Video modular pipeline and the other adds a HunyuanVideo 1.5 modular pipeline. They share framework scaffolding but not the same underlying model integration or code path.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the changes target different model-specific pipelines and different module directories, so they are not duplicates.", + "best_issue_reason": null, + "best_pr_reason": "If a representative PR is needed, #13389 is slightly more central because it has more inbound references and is the broader follow-on integration, but it is still a separate feature from #13378.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13378", + "right": "pull_request:13389", + "accept": false, + "reason": "Different underlying additions (LTX Video vs HunyuanVideo 1.5) with distinct pipeline files and model-specific blocks; shared scaffolding is not enough to merge them as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 115, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13379-2", + "nodes": [ + "pull_request:13379", + "pull_request:13449" + ], + "soft_pairs": [ + "pull_request:13379|pull_request:13449" + ], + "prepared_review_unit_hash": "ed58cac4034ce4fd2339a2fe7414959ff302ae9bf232973fa7627bd2900401aa", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6685, + "estimated_input_tokens": 1672, + "estimated_eval_tokens": 3600 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6685, + "estimated_input_tokens": 1672, + "estimated_eval_tokens": 3600 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs target QwenImage negative-prompt handling, but they fix different concrete failure modes: #13379 fixes the embedded-text path when the mask is absent, while #13449 changes CFG gating so precomputed negative embeddings still count as a negative prompt. Related, but not the same change.", + "confidence": 0.83, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13379 is the best canonical PR because it is merged, has explicit issue linkage, and appears to be the broader accepted fix for the negative-prompt/missing-mask path across QwenImage variants.", + "best_issue_reason": null, + "best_pr_reason": "PR #13379 is the strongest representative of the cluster: it is merged, touches the relevant code paths, and addresses the concrete negative_prompt_embeds-with-none-mask bug directly.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13379", + "right": "pull_request:13449", + "accept": false, + "reason": "They are adjacent QwenImage CFG/negative-prompt fixes, but not the same underlying bug. #13379 focuses on handling a missing mask in embedding extraction; #13449 changes the has_neg_prompt condition so CFG is not silently disabled when negative embeddings are supplied without a mask." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 116, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13390-2", + "nodes": [ + "pull_request:13390", + "pull_request:13511" + ], + "soft_pairs": [ + "pull_request:13390|pull_request:13511" + ], + "prepared_review_unit_hash": "a621f5608f9ba657d09cf815feca4d96a9b71b1c66443edb9cb6b8578a3f0f41", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5587, + "estimated_input_tokens": 1397, + "estimated_eval_tokens": 3050 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5587, + "estimated_input_tokens": 1397, + "estimated_eval_tokens": 3050 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two unrelated feature PRs: one adds the LongCat-AudioDiT pipeline, the other adds AutoPipelineForText2Audio support. They overlap on package wiring/docs, but not on the same code-path change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13390 is the better canonical PR because it is merged and contains a complete, concrete feature implementation for a new LongCat-AudioDiT pipeline, with model, pipeline, docs, and tests. PR #13511 is a separate auto-pipeline feature and not the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR #13390 is the best representative of the cluster: it is the larger, merged, self-contained change. PR #13511 is only loosely related via shared framework wiring and should not be grouped as the same PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13390", + "right": "pull_request:13511", + "accept": false, + "reason": "Do not merge: #13390 adds a specific LongCat-AudioDiT pipeline implementation, while #13511 adds AutoPipelineForText2Audio support. They touch some shared init/dummy/docs files, but the underlying changes and code paths are different and would not plausibly be one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 117, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13402-6", + "nodes": [ + "pull_request:13402", + "pull_request:13403", + "pull_request:13407", + "pull_request:13408", + "pull_request:13413", + "pull_request:13414" + ], + "soft_pairs": [ + "pull_request:13402|pull_request:13408", + "pull_request:13402|pull_request:13414", + "pull_request:13402|pull_request:13403", + "pull_request:13402|pull_request:13413", + "pull_request:13403|pull_request:13407", + "pull_request:13408|pull_request:13414", + "pull_request:13408|pull_request:13413", + "pull_request:13413|pull_request:13414" + ], + "prepared_review_unit_hash": "f7507b413fa09976f04a284c1405b2844955f342d44a786b88cbb50bd4d61431", + "component_budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 8, + "serialized_chars": 18863, + "estimated_input_tokens": 4716, + "estimated_eval_tokens": 9688 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 8, + "serialized_chars": 18863, + "estimated_input_tokens": 4716, + "estimated_eval_tokens": 9688 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a set of loosely related cleanup PRs (docstring grammar, duplicate words, and missing f-string prefixes), but most pairs are too broad to be true duplicates.", + "confidence": 0.78, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 13408 is the best cluster representative: it is a broad, closed docstring/error-message cleanup and most closely matches the shared theme of the cluster without being overly narrow.", + "best_issue_reason": null, + "best_pr_reason": "PR 13408 is the strongest overall representative because it has the broadest cleanup scope and best reflects the cluster\u2019s recurring grammar/docstring fix pattern.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13402", + "right": "pull_request:13408", + "accept": false, + "reason": "Same general cleanup theme, but the edits are in different files with different wording fixes; not the same concrete bug or change." + }, + { + "left": "pull_request:13402", + "right": "pull_request:13414", + "accept": false, + "reason": "Both are grammar/docstring cleanup PRs, but they touch different modules and fix different text issues; too broad to merge as one change." + }, + { + "left": "pull_request:13402", + "right": "pull_request:13403", + "accept": false, + "reason": "13403 fixes missing f-string prefixes in BnB quantizer logs, which is a different issue than 13402\u2019s duplicate-word/docstring cleanup." + }, + { + "left": "pull_request:13402", + "right": "pull_request:13413", + "accept": false, + "reason": "Both edit model docs, but they address different models and different wording mistakes; not the same underlying fix." + }, + { + "left": "pull_request:13403", + "right": "pull_request:13407", + "accept": false, + "reason": "Both involve f-string corrections, but across different pipelines and code paths; this looks like separate cleanup work, not one duplicate change." + }, + { + "left": "pull_request:13408", + "right": "pull_request:13414", + "accept": false, + "reason": "They share a broad grammar-cleanup theme, but the actual edits are different across different files; not the same concrete PR." + }, + { + "left": "pull_request:13408", + "right": "pull_request:13413", + "accept": false, + "reason": "13413 is a focused possessive-grammar cleanup in autoencoder docs, while 13408 is a wider mix of docstring/error-message edits; not a duplicate." + }, + { + "left": "pull_request:13413", + "right": "pull_request:13414", + "accept": false, + "reason": "Both are wording fixes, but they target different subsystems and different exact text changes; too little evidence they are the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 118, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13412-2", + "nodes": [ + "pull_request:13412", + "pull_request:13466" + ], + "soft_pairs": [ + "pull_request:13412|pull_request:13466" + ], + "prepared_review_unit_hash": "37eb644352ed066b7899ed7c770118fc782f2ef3129fd9ae836b68783d70b2ea", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4600, + "estimated_input_tokens": 1150, + "estimated_eval_tokens": 2556 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4600, + "estimated_input_tokens": 1150, + "estimated_eval_tokens": 2556 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two related PRs target the same scheduler, but they are not the same change: #13412 adds the monotonicity validation plus docs/tests, while #13466 is a narrow follow-up tightening the explicit sigma-schedule check. Do not deduplicate them as one PR.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "#13412 is the broader, original scheduler update with docs and tests; it best represents the main change set compared with the later follow-up fix in #13466.", + "best_issue_reason": null, + "best_pr_reason": "#13412 is the best PR candidate because it contains the full feature/fix scope and validation coverage, whereas #13466 only patches the explicit sigma monotonicity guard.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13412", + "right": "pull_request:13466", + "accept": false, + "reason": "Related to the same scheduler, but not the same underlying change. #13412 introduces the monotonicity validation and supporting docs/tests; #13466 only refines that check for explicit sigma schedules. They would not plausibly be merged into one PR as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 119, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13417-2", + "nodes": [ + "pull_request:13417", + "pull_request:13441" + ], + "soft_pairs": [ + "pull_request:13417|pull_request:13441" + ], + "prepared_review_unit_hash": "44b323fcc5eac212b89652f0f80dfdcfa617bbed6e009ce72975eec3b6da0597", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4474, + "estimated_input_tokens": 1119, + "estimated_eval_tokens": 2494 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4474, + "estimated_input_tokens": 1119, + "estimated_eval_tokens": 2494 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two unrelated DreamBooth PRs: one fixes a VAE offload device mismatch during encoding across several scripts, the other fixes prompt embedding repetition/order handling in the Qwen Image DreamBooth prior-preservation path. They do not look mergeable into one change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13417 is the better canonical PR because it fixes a concrete runtime device-mismatch bug and covers multiple DreamBooth scripts with the same code-path issue.", + "best_issue_reason": null, + "best_pr_reason": "PR #13417 is the strongest representative: broader impact, clearer bug-fix scope, and a single underlying defect in VAE encode/offload handling.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13417", + "right": "pull_request:13441", + "accept": false, + "reason": "Different underlying bugs and code paths: VAE encode device placement vs. prompt embedding repeat_interleave for prior-preservation ordering. They are not the same change and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 120, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13420-2", + "nodes": [ + "pull_request:13420", + "pull_request:13422" + ], + "soft_pairs": [ + "pull_request:13420|pull_request:13422" + ], + "prepared_review_unit_hash": "97d5ffb46e72430e4267dcad3d2b9a8040e471a0142b48d2da56aca64d3e34a8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4231, + "estimated_input_tokens": 1058, + "estimated_eval_tokens": 2372 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4231, + "estimated_input_tokens": 1058, + "estimated_eval_tokens": 2372 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two closed PRs that both add a `negative_prompt` parameter, but they target different pipelines and different implementation paths, so they are not duplicates of the same underlying change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13420 is the stronger canonical representative because it has an explicit issue link, a clear pipeline-specific API change, and more direct evidence of the intended fix scope.", + "best_issue_reason": null, + "best_pr_reason": "PR #13420 is the best representative for the cluster because it is self-contained, clearly scoped to `Flux2KleinPipeline`, and has an explicit linked issue; PR #13422 is similar in theme but addresses a different pipeline and code path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13420", + "right": "pull_request:13422", + "accept": false, + "reason": "Both add `negative_prompt`, but #13420 changes the Flux2 Klein pipeline call signature while #13422 changes GLM Image prompt encoding/CFG handling. They do not look like the same concrete bug or a single mergeable PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 121, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13455-2", + "nodes": [ + "pull_request:13455", + "pull_request:13509" + ], + "soft_pairs": [ + "pull_request:13455|pull_request:13509" + ], + "prepared_review_unit_hash": "75add2e745c8be349b1207202ecc5720a6e095a39dd23ea9a80359c2532d432c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3185, + "estimated_input_tokens": 797, + "estimated_eval_tokens": 1850 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3185, + "estimated_input_tokens": 797, + "estimated_eval_tokens": 1850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests are effectively the same change: they update the Flux 2 Klein DreamBooth script\u2019s `text_encoder_out_layers` default from `[10, 20, 30]` to `[9, 18, 27]` and both target issue #13445. They are duplicates and can be merged into one PR.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13455 is the better canonical PR because it is earlier, explicitly states the fix target, and has the same exact code change as the other PR.", + "best_issue_reason": null, + "best_pr_reason": "PR #13455 is the strongest representative of the cluster: same file, same diff, same issue target, and a clearer fix-oriented title.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13455", + "right": "pull_request:13509", + "accept": true, + "reason": "Identical code change in the same file, same default values, and both explicitly target the same issue (#13445); these are the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 122, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13456-2", + "nodes": [ + "pull_request:13456", + "pull_request:13464" + ], + "soft_pairs": [ + "pull_request:13456|pull_request:13464" + ], + "prepared_review_unit_hash": "5950003273e3813ad760a1479a04e247b24233b96ba2b0a24f6b473bdd56f061", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5026, + "estimated_input_tokens": 1257, + "estimated_eval_tokens": 2770 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5026, + "estimated_input_tokens": 1257, + "estimated_eval_tokens": 2770 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates. PR #13464 is a targeted bug fix in `transformer_ernie_image.py` for dtype handling in the ERNIE-Image forward/rope path, while PR #13456 is a much larger feature/docs addition that happens to touch the same file but is mainly about introducing ERNIE-Image models/pipelines. Do not merge them as the same change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13464 is the clearer canonical PR: it is merged, small, and directly fixes the concrete dtype regression in the ERNIE-Image transformer code path.", + "best_issue_reason": null, + "best_pr_reason": "PR #13464 is the best representative for the underlying bug fix because it precisely targets the dtype issue in the relevant implementation and is already merged.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13456", + "right": "pull_request:13464", + "accept": false, + "reason": "They are not the same underlying change: #13456 is a broad ERNIE-Image feature/docs PR, while #13464 is a narrow dtype bug fix in the transformer forward/rope path. Shared filename alone is not enough to treat them as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 123, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13467-2", + "nodes": [ + "pull_request:13467", + "pull_request:13497" + ], + "soft_pairs": [ + "pull_request:13467|pull_request:13497" + ], + "prepared_review_unit_hash": "0ec3a40426b14cbd948e4a077a5cb106677f2ff1c34ec07d8d79c3f1d8764943", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4105, + "estimated_input_tokens": 1027, + "estimated_eval_tokens": 2310 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4105, + "estimated_input_tokens": 1027, + "estimated_eval_tokens": 2310 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs target the same cache/iteration logic in `pyramid_attention_broadcast.py` and are likely alternate fixes for the same underlying hook bug. PR 13497 is the cleaner representative because it explicitly removes the redundant `iteration == 0` guard and clears stale cache outside the timestep range.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 13497 is the best canonical PR: it is open, narrowly focused, and directly addresses the same cache invalidation / redundant-guard logic in the hook.", + "best_issue_reason": null, + "best_pr_reason": "PR 13497 is the strongest cluster representative because it presents the same fix more cleanly and explicitly than the older closed PR 13467.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13467", + "right": "pull_request:13497", + "accept": true, + "reason": "Accept: both PRs modify the same file and appear to fix the same concrete code path in the PAB hook\u2014redundant iteration-0 recomputation plus stale-cache handling outside the timestep range." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 124, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13468-2", + "nodes": [ + "pull_request:13468", + "pull_request:13471" + ], + "soft_pairs": [ + "pull_request:13468|pull_request:13471" + ], + "prepared_review_unit_hash": "4d5004b596d674f64b889e19bca8ddc584d69884cce50ad6b95ee29d0a945811", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4149, + "estimated_input_tokens": 1038, + "estimated_eval_tokens": 2332 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4149, + "estimated_input_tokens": 1038, + "estimated_eval_tokens": 2332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two related profiling PRs address method-decoration isolation in `examples/profiling/profiling_utils.py`, but they are not the same change: one proposes a broad class-level patching strategy, while the other applies a narrower LTX2-only fix and was merged.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13471 is the better canonical PR because it was merged and contains the final, narrower fix for the concrete isolation bug in the profiling helper.", + "best_issue_reason": null, + "best_pr_reason": "PR #13471 is the best representative artifact: it resolves the bug in the shipped code path, is merged, and its diff reflects the implemented solution rather than an exploratory/generalized approach.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13468", + "right": "pull_request:13471", + "accept": false, + "reason": "They touch the same file and issue target, but the implementations differ materially: #13468 patches sub-component methods at the class level for all pipelines, while #13471 applies an LTX2-specific workaround and preserves original behavior elsewhere. That is related, not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 125, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13476-3", + "nodes": [ + "pull_request:13476", + "pull_request:13493", + "pull_request:13555" + ], + "soft_pairs": [ + "pull_request:13476|pull_request:13493", + "pull_request:13476|pull_request:13555", + "pull_request:13493|pull_request:13555" + ], + "prepared_review_unit_hash": "a83784478f05ccb157a10690ab0d93aeb01f327cd94f3792ee00a3ce16875935", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 5029, + "estimated_input_tokens": 1258, + "estimated_eval_tokens": 2772 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 5029, + "estimated_input_tokens": 1258, + "estimated_eval_tokens": 2772 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two PRs are direct duplicates of the same documentation workflow maintenance change; the upload-workflow bump is separate. The merged main-doc workflow PR is the best representative.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13555 is the strongest canonical PR: it is merged, touches the main documentation build workflow, and is the cleanest representative of the repeated doc-builder SHA bump change.", + "best_issue_reason": null, + "best_pr_reason": "PR #13555 is the best overall PR because it is merged and implements the same workflow SHA bump in the main doc build path, making it the most complete and authoritative instance.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13476", + "right": "pull_request:13493", + "accept": false, + "reason": "Different workflows and different files: one updates the PR documentation upload workflow, the other the main documentation build workflow. Same kind of dependency bump, but not the same underlying change." + }, + { + "left": "pull_request:13476", + "right": "pull_request:13555", + "accept": false, + "reason": "These target different doc-builder workflows and different YAML files, so they are not the same concrete bug/change despite both being SHA bumps." + }, + { + "left": "pull_request:13493", + "right": "pull_request:13555", + "accept": true, + "reason": "Both PRs update the same `.github/workflows/build_documentation.yml` doc-builder reference for the main doc build workflow, with nearly identical intent and only the SHA differing. They are the same underlying maintenance change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 126, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13529-2", + "nodes": [ + "pull_request:13529", + "pull_request:13531" + ], + "soft_pairs": [ + "pull_request:13529|pull_request:13531" + ], + "prepared_review_unit_hash": "e8ba520ca6b4a4d4c28149133fc149c6d5fbc32a0c8301a81c0e24fd49b35a2d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3804, + "estimated_input_tokens": 951, + "estimated_eval_tokens": 2158 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3804, + "estimated_input_tokens": 951, + "estimated_eval_tokens": 2158 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two unrelated PRs in different example training scripts: one fixes a Wuerstchen model-card dtype/AttributeError issue, the other fixes sana_sprint model-card save by removing an unsupported argument. They share a broad save-path theme but not the same underlying bug or patch.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a true duplicate of the other. If one must be surfaced as the most representative item, 13531 is slightly more specific about the failure mode, but it does not subsume 13529.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13529", + "right": "pull_request:13531", + "accept": false, + "reason": "Reject: different scripts, different concrete fixes, and different failure modes. 13529 hardcodes float16 in Wuerstchen model-card generation; 13531 removes instance_prompt from sana_sprint save_model_card. They are not the same underlying bug and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 127, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13530-2", + "nodes": [ + "pull_request:13530", + "pull_request:13541" + ], + "soft_pairs": [ + "pull_request:13530|pull_request:13541" + ], + "prepared_review_unit_hash": "824d7eb0631ea0b1adf2bf737a36c7b02bc6fb2c5f746dac0a0d36372a1e9c52", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4434, + "estimated_input_tokens": 1109, + "estimated_eval_tokens": 2474 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4434, + "estimated_input_tokens": 1109, + "estimated_eval_tokens": 2474 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two open PRs share one test file but address different problems: TorchAO quantization/XPU support vs. shrinking WanAnimate test tensors to avoid OOM. They should not be deduplicated.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #13530 is the broader, more substantive TorchAO quantization change and best represents the cluster\u2019s main code-path work; #13541 is a narrow test-only size reduction.", + "best_issue_reason": null, + "best_pr_reason": "#13530 is the best PR to keep as the representative because it changes the quantizer implementation and associated tests, whereas #13541 only adjusts test inputs to reduce memory use.", + "soft_edge_verdicts": [ + { + "left": "pull_request:13530", + "right": "pull_request:13541", + "accept": false, + "reason": "Shared test-file overlap is incidental; one PR adds TorchAO dequantization/XPU-related behavior, the other only reduces WanAnimate test tensor sizes to prevent OOM. They are not the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 128, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-157-2", + "nodes": [ + "pull_request:1361", + "pull_request:157" + ], + "soft_pairs": [ + "pull_request:1361|pull_request:157" + ], + "prepared_review_unit_hash": "38354450f6336f323860fd95a93f0c235514a75e7919de652ad013f833722846", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5769, + "estimated_input_tokens": 1443, + "estimated_eval_tokens": 3142 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5769, + "estimated_input_tokens": 1443, + "estimated_eval_tokens": 3142 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two pull requests are not duplicates: one standardizes `image` argument usage across diffusion pipelines and docs, while the other updates tensor creation to use the `device` argument when available. They touch different code paths and solve unrelated problems.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1361 is the stronger canonical PR for this cluster because it is merged, has broad code/docs coverage, and explicitly targets issue #1257; PR #157 is an unrelated device-argument cleanup.", + "best_issue_reason": null, + "best_pr_reason": "PR #1361 is the best PR to represent the cluster overall due to its clear scope, merge status, and explicit issue linkage. PR #157 is not a fit for the same cluster topic.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1361", + "right": "pull_request:157", + "accept": false, + "reason": "Different underlying changes: #1361 renames an image parameter across pipelines/docs, while #157 changes random tensor/device handling in training and model code. They do not fix the same bug or implement the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 129, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1383-3", + "nodes": [ + "pull_request:1383", + "pull_request:1637", + "pull_request:3274" + ], + "soft_pairs": [ + "pull_request:1383|pull_request:1637", + "pull_request:1383|pull_request:3274", + "pull_request:1637|pull_request:3274" + ], + "prepared_review_unit_hash": "8d24bdf19bf1b15805d529332c77ddf1637b831ba988044a9a62c2f55ccb0aba", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7816, + "estimated_input_tokens": 1954, + "estimated_eval_tokens": 4164 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7816, + "estimated_input_tokens": 1954, + "estimated_eval_tokens": 4164 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate patch-release PRs for different versions and unrelated code/doc changes. They share the same release-maintenance pattern, but not the same underlying fix or change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1637 is the strongest representative because it is the only merged PR and includes a concrete code change plus tests, whereas the others are unmerged release bumps with different surrounding edits.", + "best_issue_reason": null, + "best_pr_reason": "PR #1637 is the best global PR candidate: merged, substantive, and more representative than the other two version-bump release PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1383", + "right": "pull_request:1637", + "accept": false, + "reason": "Both are patch-release PRs, but they target different release versions and different surrounding files; they do not appear to fix or deliver the same concrete change." + }, + { + "left": "pull_request:1383", + "right": "pull_request:3274", + "accept": false, + "reason": "Same broad release-bump pattern, but 0.8.1 and 0.16.1 are different releases with different doc/code updates, so they are not the same underlying change." + }, + { + "left": "pull_request:1637", + "right": "pull_request:3274", + "accept": false, + "reason": "Both are patch-release PRs, but one is a modeling_utils/test change for 0.10.1 and the other is IF/docs-related for 0.16.1; these are distinct changes, not mergeable duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 130, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1446-3", + "nodes": [ + "pull_request:1446", + "pull_request:1653", + "pull_request:2298" + ], + "soft_pairs": [ + "pull_request:1446|pull_request:1653", + "pull_request:1653|pull_request:2298" + ], + "prepared_review_unit_hash": "80848fa741d84796bcf65694831cb3e17c284f60a006fcc01428e48a08ce625e", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8361, + "estimated_input_tokens": 2091, + "estimated_eval_tokens": 4438 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8361, + "estimated_input_tokens": 2091, + "estimated_eval_tokens": 4438 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three related-but-distinct PRs in diffusion pipelines: one removes `**kwargs` and speeds tests, one fixes img2img timestep offset handling, and one removes an unused latent upscale noise argument. They share the same general area but not the same underlying change.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1653 is the most concrete bugfix: it changes the img2img timestep calculation across several pipeline variants and has a clear, focused purpose. The other PRs are broader cleanup/speed or unrelated unused-parameter removal.", + "best_issue_reason": null, + "best_pr_reason": "PR #1653 is the best representative of the cluster because it targets a specific code-path defect in image-to-image timestep logic, whereas #1446 is mostly API/test cleanup and #2298 is a separate latent-upscale signature cleanup.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1446", + "right": "pull_request:1653", + "accept": false, + "reason": "Both touch pipeline call paths, but #1446 removes `**kwargs` and speeds a test, while #1653 fixes img2img timestep offset logic. Different underlying problems and not safely mergeable as one PR." + }, + { + "left": "pull_request:1653", + "right": "pull_request:2298", + "accept": false, + "reason": "They are different fixes in different pipelines: #1653 adjusts timestep/offset computation, while #2298 removes an unused `noise_level` parameter in latent upscale. No shared bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 131, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1557-2", + "nodes": [ + "pull_request:1557", + "pull_request:1799" + ], + "soft_pairs": [ + "pull_request:1557|pull_request:1799" + ], + "prepared_review_unit_hash": "93278210b9916602fafbcac32767abfa49085e84f3f3eb29a9dbf1b27e6fce72", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3708, + "estimated_input_tokens": 927, + "estimated_eval_tokens": 2110 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3708, + "estimated_input_tokens": 927, + "estimated_eval_tokens": 2110 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged PRs in the same example file, but they address different ORT training bugs: one removes an incompatible DDPMScheduler argument, the other reorders ORTModule wrapping and adjusts return_dict usage.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1799 is the better representative of the cluster because it fixes the later, broader runtime issue in the example training flow and includes the concrete ORTModule call-path correction.", + "best_issue_reason": null, + "best_pr_reason": "PR #1799 is the strongest standalone representative: it directly fixes the model execution path in the example and is more substantial than the narrower scheduler-argument compatibility change in #1557.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1557", + "right": "pull_request:1799", + "accept": false, + "reason": "Same example file, but not the same underlying bug. #1557 removes a deprecated/incompatible DDPMScheduler parameter; #1799 changes when ORTModule is wrapped and how model outputs are returned. They affect different code paths and would not naturally be one merged PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 132, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1567-2", + "nodes": [ + "pull_request:1567", + "pull_request:1618" + ], + "soft_pairs": [ + "pull_request:1567|pull_request:1618" + ], + "prepared_review_unit_hash": "aa4e23a542efb8e550bc62dd407cd5070a8573d299bb9c453dd52869407e8077", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4314, + "estimated_input_tokens": 1079, + "estimated_eval_tokens": 2414 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4314, + "estimated_input_tokens": 1079, + "estimated_eval_tokens": 2414 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are near-identical fixes for the same DreamBooth fp32-wrapper checkpointing crash (#1566). They touch the same file, implement the same inspect/signature-based guard for keep_fp32_wrapper, and have the same title; one is the merged version and the other is a duplicate/unmerged attempt.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1618 is the better canonical PR because it is merged and contains the same concrete fix as #1567, making it the authoritative completed change.", + "best_issue_reason": null, + "best_pr_reason": "PR #1618 is the strongest PR overall since it is merged, targets the exact crash in train_dreambooth.py, and is effectively the finalized version of the duplicate patch.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1567", + "right": "pull_request:1618", + "accept": true, + "reason": "Same title, same target file, same underlying fp32-wrapper crash fix, and nearly identical patch; these are duplicate PRs for the same change and could plausibly be one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 133, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1691-2", + "nodes": [ + "pull_request:1691", + "pull_request:1888" + ], + "soft_pairs": [ + "pull_request:1691|pull_request:1888" + ], + "prepared_review_unit_hash": "77bd2feb9939488664ffe2a1b18e59deda376c6974a9556a27fa983636b4f71c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4391, + "estimated_input_tokens": 1098, + "estimated_eval_tokens": 2452 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4391, + "estimated_input_tokens": 1098, + "estimated_eval_tokens": 2452 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are both in training scripts, but they fix different things: one removes duplicated training-step bookkeeping in `train_text_to_image.py`, while the other updates `report_to` help text across several scripts. They are not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 1888 is the better representative artifact: it is merged, spans multiple training scripts, and contains the more complete user-facing fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 1888 is the strongest overall choice because it is merged and addresses a concrete, user-visible documentation/help-text issue across the training scripts.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1691", + "right": "pull_request:1888", + "accept": false, + "reason": "Different underlying changes: 1691 adjusts max-train-steps/epoch logic in `examples/text_to_image/train_text_to_image.py`, while 1888 fixes `report_to` CLI help text in several scripts. Shared file overlap is incidental, not the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 134, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1694-3", + "nodes": [ + "pull_request:1694", + "pull_request:1775", + "pull_request:2033" + ], + "soft_pairs": [ + "pull_request:1694|pull_request:1775", + "pull_request:1694|pull_request:2033", + "pull_request:1775|pull_request:2033" + ], + "prepared_review_unit_hash": "6247552883bff0b647b438eb16cb45ebcfb68d513b3032c51cb02279c94d006b", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7060, + "estimated_input_tokens": 1765, + "estimated_eval_tokens": 3786 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7060, + "estimated_input_tokens": 1765, + "estimated_eval_tokens": 3786 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs touch the same example file, but they are different follow-up edits rather than the same fix/change. #1694 is the broadest merged update, while #1775 and #2033 are narrower subsequent modifications.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue is present; as a cluster anchor, #1694 is the best representative because it is the earliest merged substantive update to the file and covers the widest change set.", + "best_issue_reason": null, + "best_pr_reason": "#1694 is the most suitable representative PR for this cluster: it is merged and materially larger than the others, which are smaller incremental edits to the same script.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1694", + "right": "pull_request:1775", + "accept": false, + "reason": "Same file, but different changes: #1694 adds substantial helper/update logic, while #1775 mainly adds logging/CLI options. They do not look like the same underlying bug or mergeable fix." + }, + { + "left": "pull_request:1694", + "right": "pull_request:2033", + "accept": false, + "reason": "#2033 is a small cleanup/adjustment to the script, not the same concrete change as #1694's larger manual update. Shared filename alone is not enough." + }, + { + "left": "pull_request:1775", + "right": "pull_request:2033", + "accept": false, + "reason": "Both are follow-up edits to the same example script, but they address different concerns (logging args vs. minor cleanup/scheduler tweak), so they are not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 135, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1710-3", + "nodes": [ + "pull_request:1710", + "pull_request:1770", + "pull_request:2600" + ], + "soft_pairs": [ + "pull_request:1710|pull_request:2600", + "pull_request:1770|pull_request:2600" + ], + "prepared_review_unit_hash": "73cf4143c9736a078d9afec38463a9aafcf3a09aa5927b3056db9fbf6bf7729b", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8894, + "estimated_input_tokens": 2224, + "estimated_eval_tokens": 4704 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8894, + "estimated_input_tokens": 2224, + "estimated_eval_tokens": 4704 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate documentation-focused PRs: one adds the docs README/build scaffolding, one updates Stable Diffusion pipeline docs, and one adds Flax pipeline API docs. They overlap in the docs subsystem but do not look like the same underlying change.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1770 is the broadest and most central documentation update in this set, covering Stable Diffusion pipeline docs and related code/doc examples, so it is the best representative if one PR must be chosen.", + "best_issue_reason": null, + "best_pr_reason": "PR #1770 is the strongest global representative because it has the widest doc surface area and sits closest to the main Stable Diffusion API documentation work, while #1710 is infrastructure/docs setup and #2600 is a narrower Flax-specific doc addition.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1710", + "right": "pull_request:2600", + "accept": false, + "reason": "#1710 adds the docs README and documentation build/navigation files; #2600 adds Flax Stable Diffusion API docs and docstring examples. They target different documentation problems and are not one mergeable fix." + }, + { + "left": "pull_request:1770", + "right": "pull_request:2600", + "accept": false, + "reason": "#1770 updates Stable Diffusion pipeline docs and related examples, while #2600 adds Flax pipeline API docs. They overlap in topic but fix different doc surfaces and should not be treated as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 136, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1711-2", + "nodes": [ + "pull_request:1711", + "pull_request:1714" + ], + "soft_pairs": [ + "pull_request:1711|pull_request:1714" + ], + "prepared_review_unit_hash": "32c01e9c5eaa4d16ce8a80cc387c7832f00f7caff74f419e17c0861b28313e69", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4252, + "estimated_input_tokens": 1063, + "estimated_eval_tokens": 2382 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4252, + "estimated_input_tokens": 1063, + "estimated_eval_tokens": 2382 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests are near-identical DreamBooth changes adding interim image generation logging to the same code path in `train_dreambooth.py`; they look like duplicate attempts at the same patch, with 1714 the more complete/non-draft version.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #1714 is the better canonical PR because it is non-draft and the diff matches #1711 almost exactly, indicating it is the cleaner final version of the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR #1714 is the best overall representative of the cluster since it is the non-draft submission and implements the same DreamBooth interim generation logging change as #1711.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1711", + "right": "pull_request:1714", + "accept": true, + "reason": "Accepted: both PRs modify the same file with the same added argument, seeded generator setup, and interim generation logging logic; they are effectively duplicate implementations of the same concrete change and could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 137, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1803-2", + "nodes": [ + "pull_request:1803", + "pull_request:1935" + ], + "soft_pairs": [ + "pull_request:1803|pull_request:1935" + ], + "prepared_review_unit_hash": "042b32af098bb95988f6cf20edf6ede031575ad92afe29c782274acd7ae92d06", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6697, + "estimated_input_tokens": 1675, + "estimated_eval_tokens": 3606 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6697, + "estimated_input_tokens": 1675, + "estimated_eval_tokens": 3606 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs add OneFlow nn.graph support and share a large amount of infrastructure, but they target different concrete pipeline changes: 1803 is for AltDiffusion, while 1935 is for Stable Diffusion inpainting. They are similar in theme, not the same underlying change.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 1935 is the better representative of the cluster because it is a later, more specific OneFlow graph support change centered on Stable Diffusion inpainting.", + "best_issue_reason": null, + "best_pr_reason": "PR 1935 is the strongest representative here: it targets a concrete pipeline feature rather than the more separate AltDiffusion path, and it appears to be a more focused implementation of the same broad OneFlow effort.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1803", + "right": "pull_request:1935", + "accept": false, + "reason": "Reject: they share OneFlow helper/infrastructure files, but the actual feature work is for different pipelines and code paths (AltDiffusion vs Stable Diffusion inpainting)." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 138, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1870-2", + "nodes": [ + "pull_request:1870", + "pull_request:2434" + ], + "soft_pairs": [ + "pull_request:1870|pull_request:2434" + ], + "prepared_review_unit_hash": "49cb7bc176f5b7870ed0f98569156be9338b22cb4f8edd841449086199a614ae", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5245, + "estimated_input_tokens": 1312, + "estimated_eval_tokens": 2880 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5245, + "estimated_input_tokens": 1312, + "estimated_eval_tokens": 2880 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two unrelated documentation PRs: #1870 is a broad API/docs cleanup for diffusion pipeline pages, while #2434 adds pix2pix-zero and controlling-generation details. They do not look like the same underlying change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: #1870 revises pipeline API reference docs, while #2434 expands pix2pix-zero usage guidance and adds a demo link; different doc targets and intent.", + "best_issue_reason": null, + "best_pr_reason": "No single best PR for duplicate triage here; the PRs are not duplicates and cover separate documentation updates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1870", + "right": "pull_request:2434", + "accept": false, + "reason": "Different documentation changes in different files and topics; they are not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 139, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1893-2", + "nodes": [ + "pull_request:1893", + "pull_request:1948" + ], + "soft_pairs": [ + "pull_request:1893|pull_request:1948" + ], + "prepared_review_unit_hash": "2b04c8bf5080cb18f5a215df09fb5f214ffed0f7cd00bb5b810b8f65a9f069d0", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4364, + "estimated_input_tokens": 1091, + "estimated_eval_tokens": 2438 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4364, + "estimated_input_tokens": 1091, + "estimated_eval_tokens": 2438 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests are about DreamBooth support for multiple subjects/prompts, but they are different implementations in different locations: one patches the existing DreamBooth example, the other adds a separate research-project script and README. They are related but not the same concrete change, so the soft edge should not be merged.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:1948 is the stronger canonical PR because it is merged, self-contained, and adds a dedicated multi-subject DreamBooth implementation with its own script and documentation.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:1948 is the best representative of the cluster because it fully implements the multi-subject DreamBooth feature, whereas pull_request:1893 is a narrower, unmerged doc/code tweak to the existing example.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1893", + "right": "pull_request:1948", + "accept": false, + "reason": "Related DreamBooth multi-concept work, but not the same underlying change: 1893 modifies the existing example/README, while 1948 adds a separate research-project script and docs. They could not plausibly be merged as one PR without substantial restructuring." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 140, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-1906-2", + "nodes": [ + "pull_request:1906", + "pull_request:1924" + ], + "soft_pairs": [ + "pull_request:1906|pull_request:1924" + ], + "prepared_review_unit_hash": "a7a02e58b281895b1ab4ba4e9d6430399d6ff38975666d998bdd30d645c7adfa", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6882, + "estimated_input_tokens": 1721, + "estimated_eval_tokens": 3698 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6882, + "estimated_input_tokens": 1721, + "estimated_eval_tokens": 3698 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related reproducibility PRs in the same series, but they are not duplicates: one is a code-path RNG consistency change, the other is a broader docs/tests/reproducibility follow-up.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 1924 is the more complete reproducibility change: it adds the docs page, testing utilities, and broader test coverage, whereas PR 1906 is a narrower implementation step replacing torch.randn with randn_tensor in several code paths.", + "best_issue_reason": null, + "best_pr_reason": "PR 1924 is the best representative of the cluster because it packages the reproducibility work in a user-facing, fully documented and tested change. PR 1906 is an earlier, narrower prerequisite-style code update.", + "soft_edge_verdicts": [ + { + "left": "pull_request:1906", + "right": "pull_request:1924", + "accept": false, + "reason": "Same theme and some overlapping files, but different underlying changes: 1906 focuses on making sampling use randn_tensor for reproducibility, while 1924 adds reproducibility documentation, helpers, and tests. They are complementary follow-ups, not the same concrete fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 141, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2067-3", + "nodes": [ + "pull_request:2067", + "pull_request:2096", + "pull_request:2147" + ], + "soft_pairs": [ + "pull_request:2067|pull_request:2096", + "pull_request:2067|pull_request:2147", + "pull_request:2096|pull_request:2147" + ], + "prepared_review_unit_hash": "82935e60182d57dabd2f2815ebf7f72eade0be9ac1efd565f421e03d0be4bafa", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7403, + "estimated_input_tokens": 1851, + "estimated_eval_tokens": 3958 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7403, + "estimated_input_tokens": 1851, + "estimated_eval_tokens": 3958 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three pull requests target the same legacy Stable Diffusion inpaint bug/fix set: resizing the mask tensor to latent size and preventing extra noise from being added to the final latents. PR 2147 appears to be the finalized, merged version with tests; 2067 and 2096 are earlier overlapping attempts.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 2147 is the canonical PR: it is merged, covers the same concrete legacy inpaint code path, and looks like the most complete/finalized fix with accompanying tests.", + "best_issue_reason": null, + "best_pr_reason": "PR 2147 is the best representative overall because it is merged and most fully addresses the underlying bug, while the other two are earlier overlapping iterations.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2067", + "right": "pull_request:2096", + "accept": true, + "reason": "Same legacy inpaint bug/fix set in the same file: both address scaling the mask tensor to latent size and avoiding noise on final latents, so they look like overlapping iterations of one fix." + }, + { + "left": "pull_request:2067", + "right": "pull_request:2147", + "accept": true, + "reason": "Same concrete bug in the same code path; PR 2147 is a later, more complete version of the same legacy inpaint mask-resize/noise fix." + }, + { + "left": "pull_request:2096", + "right": "pull_request:2147", + "accept": true, + "reason": "Both target the same underlying legacy inpaint correction, and PR 2147 clearly refines the same fix into the merged final form." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: all three PRs target the same legacy inpaint code path and the same mask-resize / final-latents noise fix, and PR 2147 is the only merged one with a test file. The pairwise overlap verdicts are conservative enough based on the shared file and closely related diffs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 142, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2072-2", + "nodes": [ + "pull_request:2072", + "pull_request:2079" + ], + "soft_pairs": [ + "pull_request:2072|pull_request:2079" + ], + "prepared_review_unit_hash": "90d922c1e117e2ca2494c32daca511b27238f048073969834d821ffb53a534ff", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4744, + "estimated_input_tokens": 1186, + "estimated_eval_tokens": 2628 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4744, + "estimated_input_tokens": 1186, + "estimated_eval_tokens": 2628 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address resume/checkpoint handling in training scripts, but they fix different concrete bugs and are not duplicates of the same change.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #2079 is the broader, more representative fix: it updates resume/epoch behavior across multiple training scripts and includes the checkpoint-missing handling path.", + "best_issue_reason": null, + "best_pr_reason": "PR #2079 best captures the cluster\u2019s main theme because it spans the general resume-state bug across training scripts, while #2072 is a narrower textual_inversion-specific variant.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2072", + "right": "pull_request:2079", + "accept": false, + "reason": "Related area, but not the same underlying change: #2072 fixes textual_inversion resume state with gradient checkpointing, while #2079 fixes resume epoch logic and missing-checkpoint handling across several scripts. They would not plausibly merge into one PR without broadening scope." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 143, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2125-2", + "nodes": [ + "pull_request:2125", + "pull_request:2127" + ], + "soft_pairs": [ + "pull_request:2125|pull_request:2127" + ], + "prepared_review_unit_hash": "ac08870da3005bd2120892b1306b71613095c6ffeed5ff25762205268c235967", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4270, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4270, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs both touch `examples/community/checkpoint_merger.py` and both improve reporting, but they are not the same concrete change: one focuses on logger usage for reporting, while the other also changes safetensors import handling and status/error messaging. They are related but should not be merged as duplicates.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #2127 is the more central checkpoint-merger reporting change and includes the concrete import/reporting adjustments in the example itself.", + "best_issue_reason": null, + "best_pr_reason": "PR #2127 is the best representative because it directly addresses the example\u2019s status/error reporting path; PR #2125 is a narrower logger refactor that also touches another utility file.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2125", + "right": "pull_request:2127", + "accept": false, + "reason": "Related reporting updates in the same example, but they change different aspects of the code path: #2125 is a logger/reporting refactor, while #2127 also alters safetensors import behavior and error/status handling. Not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 144, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-44-3", + "nodes": [ + "pull_request:218", + "pull_request:44", + "pull_request:67" + ], + "soft_pairs": [ + "pull_request:218|pull_request:44", + "pull_request:218|pull_request:67" + ], + "prepared_review_unit_hash": "0ace2f0c0a405b5bcaaeae25bfdcffc9bea3e162716f3e915c55889b982b73ff", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6543, + "estimated_input_tokens": 1636, + "estimated_eval_tokens": 3528 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6543, + "estimated_input_tokens": 1636, + "estimated_eval_tokens": 3528 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs touch `resnet.py` and are cleanup-oriented, but they address different code removals/refactors: one deletes `set_weight`, one changes `Normalize` and related comments, and one removes `Normalize`/`nonlinearity`/`TimestepBlock` helpers entirely. They are related in theme, not duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 67 is the best representative cleanup PR: it removes multiple obsolete helper definitions from `resnet.py` and is the broadest, clearest dead-code cleanup in the set.", + "best_issue_reason": null, + "best_pr_reason": "PR 67 is the strongest single PR to anchor the cluster because it targets the core helper-class dead-code cleanup in `resnet.py`; the other PRs are narrower or different refactors and are not merge-equivalent.", + "soft_edge_verdicts": [ + { + "left": "pull_request:218", + "right": "pull_request:44", + "accept": false, + "reason": "Both touch `resnet.py`, but 218 removes a stale `set_weight` helper while 44 changes `Normalize`\u2019s signature and related formatting. Different code paths and not plausibly one merged fix." + }, + { + "left": "pull_request:218", + "right": "pull_request:67", + "accept": false, + "reason": "218 deletes only `set_weight`; 67 deletes `Normalize`, `nonlinearity`, and `TimestepBlock`. Same file and cleanup theme, but not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 145, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2206-4", + "nodes": [ + "pull_request:2206", + "pull_request:2260", + "pull_request:2842", + "pull_request:2962" + ], + "soft_pairs": [ + "pull_request:2206|pull_request:2842", + "pull_request:2206|pull_request:2962", + "pull_request:2206|pull_request:2260", + "pull_request:2842|pull_request:2962" + ], + "prepared_review_unit_hash": "088e4ea398a2d8d0edb0a0e1044056e88ae3f209dcd53128c7a47adb91065e86", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 9353, + "estimated_input_tokens": 2339, + "estimated_eval_tokens": 4934 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 9353, + "estimated_input_tokens": 2339, + "estimated_eval_tokens": 4934 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "There are two duplicate groups here: the img2img prompt/prompt_embeds input-check fix (2206, 2260) and the separate legacy inpaint / k-diffusion prompt_embeds-only support changes (2842, 2962). These are related thematically but not the same concrete code change across pipelines.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "2206 is the best canonical representative for the img2img fix: it is merged, touches the exact code path, and its title/body describe the prompt input-check behavior being corrected. The other PR in that pair appears to be the same change in the same file.", + "best_issue_reason": null, + "best_pr_reason": "2206 is the strongest single representative overall because it is merged, clearly scoped, and directly addresses the prompt validation / batch-size inference bug in the img2img pipeline.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2206", + "right": "pull_request:2842", + "accept": false, + "reason": "Different pipelines and code paths: img2img versus legacy inpaint. Both involve prompt_embeds support, but they are separate fixes and should not be merged as one duplicate PR." + }, + { + "left": "pull_request:2206", + "right": "pull_request:2962", + "accept": false, + "reason": "Different pipelines and code paths: img2img versus k-diffusion. Same broad theme, but not the same concrete bug or patch." + }, + { + "left": "pull_request:2206", + "right": "pull_request:2260", + "accept": true, + "reason": "Same file, same prompt input-check / prompt_embeds inference fix, and the diff preview shows the same underlying edit. These look like duplicate implementations of the same change." + }, + { + "left": "pull_request:2842", + "right": "pull_request:2962", + "accept": true, + "reason": "Both change different pipelines in the same way to allow prompt_embeds without requiring a prompt, which is the same concrete functional fix pattern and could plausibly be consolidated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and the soft-edge decisions are appropriately conservative. The 2206/2260 pair is well-supported as the same img2img fix in the same file, and the 2842/2962 pair is plausibly mergeable as the same prompt_embeds-only support change across two closely related pipelines. The cross-pipeline non-duplicates are correctly rejected." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 146, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2217-2", + "nodes": [ + "pull_request:2217", + "pull_request:2259" + ], + "soft_pairs": [ + "pull_request:2217|pull_request:2259" + ], + "prepared_review_unit_hash": "94ea7a00927ce14ac5b47b5fe74f8de5bc9215f35030af08d65b22ac14efd68b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3210, + "estimated_input_tokens": 803, + "estimated_eval_tokens": 1862 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3210, + "estimated_input_tokens": 803, + "estimated_eval_tokens": 1862 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests target the same feature change: exposing center crop and random flip flags for unconditional image generation. PR #2259 is the merged implementation and the stronger canonical PR; PR #2217 appears to be the same change or an earlier duplicate attempt.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #2259 is the merged PR with concrete file changes and the same title as #2217, so it best represents the actual completed change.", + "best_issue_reason": null, + "best_pr_reason": "PR #2259 is the best representative because it is merged, has the implemented diff, and directly matches the requested behavior change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2217", + "right": "pull_request:2259", + "accept": true, + "reason": "Same title and same underlying feature request; #2259 shows the implemented code/docs change, so these are effectively duplicate PRs for the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 147, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2281-2", + "nodes": [ + "pull_request:2281", + "pull_request:2318" + ], + "soft_pairs": [ + "pull_request:2281|pull_request:2318" + ], + "prepared_review_unit_hash": "80feda53ac44d461ab4e83cc82baf01b49f9e6c60d4d018d8623b1bc00a74050", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7188, + "estimated_input_tokens": 1797, + "estimated_eval_tokens": 3850 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7188, + "estimated_input_tokens": 1797, + "estimated_eval_tokens": 3850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about loading Stable Diffusion from checkpoints, but they are not the same concrete change: #2281 is an earlier Stable Diffusion-specific refactor, while #2318 generalizes the idea into a FromCkptMixin and extends it across multiple pipelines/docs. Treat them as related but not duplicates.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "#2318 is the better canonical PR because it is merged, broader in scope, and appears to be the finalized implementation of the checkpoint-loading mixin across pipelines.", + "best_issue_reason": null, + "best_pr_reason": "#2318 is the strongest representative of the feature set: it is merged, has the wider code and docs coverage, and reflects the more complete API shape.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2281", + "right": "pull_request:2318", + "accept": false, + "reason": "Related feature work, but not the same underlying change. #2281 is a Stable Diffusion-specific from_checkpoint refactor; #2318 introduces a FromCkptMixin and expands support to multiple pipelines, so they would not plausibly be one single PR without substantial redesign." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 148, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-302-2", + "nodes": [ + "pull_request:2313", + "pull_request:302" + ], + "soft_pairs": [ + "pull_request:2313|pull_request:302" + ], + "prepared_review_unit_hash": "b6bd3f9155d38f66de489b313c4ba274067d517321f0b419c7df69d16b7beb74", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4108, + "estimated_input_tokens": 1027, + "estimated_eval_tokens": 2310 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4108, + "estimated_input_tokens": 1027, + "estimated_eval_tokens": 2310 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are related to fast testing but are not duplicates: PR #302 adds fast pipeline tests in the test suite, while PR #2313 changes CI workflow to run fast CPU tests on main. They touch different layers and solve different problems.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #302 is the more substantive, self-contained change because it implements the fast pipeline tests themselves; PR #2313 is only workflow wiring around those tests.", + "best_issue_reason": null, + "best_pr_reason": "PR #302 is the better representative of the cluster\u2019s core testing change. PR #2313 depends on that work and is not the same code-path change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2313", + "right": "pull_request:302", + "accept": false, + "reason": "Different changes: #2313 updates the GitHub Actions workflow to schedule CPU tests on main, while #302 adds fast pipeline tests in tests/test_pipelines.py. Shared testing theme is not enough to treat them as the same PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 149, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2417-2", + "nodes": [ + "pull_request:2417", + "pull_request:2418" + ], + "soft_pairs": [ + "pull_request:2417|pull_request:2418" + ], + "prepared_review_unit_hash": "bcd0ad5527f1b435b661c837fad2f3dbaeceff10ab3c106f2b4d519f757cccfd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4032, + "estimated_input_tokens": 1008, + "estimated_eval_tokens": 2272 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4032, + "estimated_input_tokens": 1008, + "estimated_eval_tokens": 2272 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests target the same bug in `get_indices` inside the Attend-and-Excite pipeline; PR 2418 is the merged, more complete version with the same fix plus an example update.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 2418 is merged and contains the actual `get_indices` fix, plus the accompanying documentation/example update.", + "best_issue_reason": null, + "best_pr_reason": "PR 2418 is the best representative because it directly fixes the bug, is merged, and subsumes the narrower PR 2417 change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2417", + "right": "pull_request:2418", + "accept": true, + "reason": "Same underlying `get_indices` bug in the same file; PR 2418 includes the same code-path fix and could plausibly absorb PR 2417 into one change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both PRs change the same `get_indices` line in the Attend-and-Excite pipeline, and PR 2418 adds only a small documentation/example update on top of that fix. The soft-edge acceptance is conservative and the two changes look mergeable into one PR for the same bugfix." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 150, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2450-2", + "nodes": [ + "pull_request:2450", + "pull_request:2451" + ], + "soft_pairs": [ + "pull_request:2450|pull_request:2451" + ], + "prepared_review_unit_hash": "bfb87f9ec2b71a13393a64e0f97bf89480deeba62f954471fd5e298043933284", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4233, + "estimated_input_tokens": 1059, + "estimated_eval_tokens": 2374 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4233, + "estimated_input_tokens": 1059, + "estimated_eval_tokens": 2374 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests are near-duplicate implementations of the same DreamBooth `--save_every` feature in the same file, with essentially the same diff and intent.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #2451 is a slightly better canonical representative because it is the later, more engaged duplicate with more discussion/review activity while implementing the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR #2451 is the best overall representative of this cluster: it targets the same concrete code path, has the same feature diff, and appears to be the more developed version of the duplicate pair.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2450", + "right": "pull_request:2451", + "accept": true, + "reason": "Both PRs add the same `--save_every` option and same save-hook logic in `examples/dreambooth/train_dreambooth.py`; they are effectively duplicate implementations of one change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 151, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2527-2", + "nodes": [ + "pull_request:2527", + "pull_request:2537" + ], + "soft_pairs": [ + "pull_request:2527|pull_request:2537" + ], + "prepared_review_unit_hash": "0595f7847f262b1ed43264c4e473f94692b0469fa4b947a22a6d81dcd3c9b960", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4285, + "estimated_input_tokens": 1072, + "estimated_eval_tokens": 2400 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4285, + "estimated_input_tokens": 1072, + "estimated_eval_tokens": 2400 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related documentation PRs, but they do not fix the same underlying thing: one migrates and renames LoRA training docs, the other restructures the text-to-image training page. The overlap in _toctree.yml is incidental.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the two pull requests touch different docs pages and implement different content changes, so they are not duplicates.", + "best_issue_reason": null, + "best_pr_reason": "PR 2527 is the cleaner standalone docs migration/update for the LoRA training page, but it is only a representative PR for the cluster, not a duplicate of 2537.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2527", + "right": "pull_request:2537", + "accept": false, + "reason": "Different pages and purposes: 2527 moves/rewrites LoRA training docs, while 2537 renames and updates the text-to-image training doc. Shared navigation edits do not make them the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 152, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2547-3", + "nodes": [ + "pull_request:2547", + "pull_request:2576", + "pull_request:2592" + ], + "soft_pairs": [ + "pull_request:2547|pull_request:2576", + "pull_request:2576|pull_request:2592" + ], + "prepared_review_unit_hash": "dd36dae3d2aecb12b1c3446721cd8d2e9d049e540ff2b9a152c8a63e9ebceb58", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6467, + "estimated_input_tokens": 1617, + "estimated_eval_tokens": 3490 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6467, + "estimated_input_tokens": 1617, + "estimated_eval_tokens": 3490 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate documentation PRs for different training topics (DreamBooth, Textual Inversion, and Unconditional Image Generation). They share a similar docs-migration style, but they do not appear to be the same underlying change or bug fix.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR fits the cluster: each PR updates a different training doc page and toctree entry, so they are distinct changes rather than duplicates.", + "best_issue_reason": null, + "best_pr_reason": "None of the PRs is a strong global representative for a duplicate cluster because the items are unrelated topic-specific docs updates; if forced, 2547 is only the earliest merged docs move, not a true canonical duplicate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2547", + "right": "pull_request:2576", + "accept": false, + "reason": "Different doc pages and different training methods; similar docs-migration pattern only, not the same change." + }, + { + "left": "pull_request:2576", + "right": "pull_request:2592", + "accept": false, + "reason": "Both are documentation updates, but they target different training topics and files, so they are not the same underlying PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 153, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2621-6", + "nodes": [ + "pull_request:2621", + "pull_request:4084", + "pull_request:4131", + "pull_request:4138", + "pull_request:4861", + "pull_request:5638" + ], + "soft_pairs": [ + "pull_request:2621|pull_request:4861", + "pull_request:4084|pull_request:4861", + "pull_request:4131|pull_request:4138", + "pull_request:4131|pull_request:5638", + "pull_request:4861|pull_request:5638" + ], + "prepared_review_unit_hash": "97487a0bf8ec069d6623f3859b9eb0b6331b01c059c7e576ddbf60298c235beb", + "component_budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 5, + "serialized_chars": 15909, + "estimated_input_tokens": 3978, + "estimated_eval_tokens": 8212 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 5, + "serialized_chars": 15909, + "estimated_input_tokens": 3978, + "estimated_eval_tokens": 8212 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Cluster is centered on AutoPipeline/controlnet work. One pair is clearly the same feature branch (initial AutoPipeline PR and the later completed AutoPipeline PR); the other controlnet-related PRs are related but address different changes or bugfixes.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4138 is the best canonical PR: it is merged, broader, and represents the completed AutoPipeline feature set with docs, exports, mappings, and pipeline support.", + "best_issue_reason": null, + "best_pr_reason": "PR 4138 is the strongest representative for the cluster because it is the largest merged change and subsumes the core AutoPipeline work that the later related PRs build on.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2621", + "right": "pull_request:4861", + "accept": false, + "reason": "Different underlying changes: 2621 adds multi-ControlNet support in the ControlNet pipeline, while 4861 changes AutoPipeline.from_pipe behavior for controlnet counterparts." + }, + { + "left": "pull_request:4084", + "right": "pull_request:4861", + "accept": false, + "reason": "Different features: 4084 adds single-file loading for ControlNet/VAE, not AutoPipeline conversion logic." + }, + { + "left": "pull_request:4131", + "right": "pull_request:4138", + "accept": true, + "reason": "Same underlying AutoPipeline feature set; 4138 appears to complete and expand the draft AutoPipeline work started in 4131, with shared exports and pipeline wiring." + }, + { + "left": "pull_request:4131", + "right": "pull_request:5638", + "accept": false, + "reason": "Related area but different change: 4131 introduces AutoPipeline, while 5638 is a narrow bugfix in from_pipe controlnet class-name handling." + }, + { + "left": "pull_request:4861", + "right": "pull_request:5638", + "accept": false, + "reason": "Both touch AutoPipeline.from_pipe, but 4861 adds controlnet counterpart support and 5638 fixes a later class-name replacement bug; they are not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 154, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2723-2", + "nodes": [ + "pull_request:2723", + "pull_request:2779" + ], + "soft_pairs": [ + "pull_request:2723|pull_request:2779" + ], + "prepared_review_unit_hash": "6fbc1437e282ef52468317aa2a29be438a85e5f8c25eedc00a3cc5475fed2bde", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4392, + "estimated_input_tokens": 1098, + "estimated_eval_tokens": 2452 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4392, + "estimated_input_tokens": 1098, + "estimated_eval_tokens": 2452 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related to the same community CLIP-guided pipelines, but they address different changes: PR 2723 adds CLIP guidance for the img2img pipeline, while PR 2779 adds DPMSolverMultistepScheduler support to the CLIP-guided pipelines. They should not be merged as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 2779 is the cleaner representative if you need one cluster PR: it makes a concrete compatibility change across the CLIP-guided pipeline implementations, with focused code updates rather than a new feature introduction.", + "best_issue_reason": null, + "best_pr_reason": "PR 2779 is the stronger standalone change for the cluster topic because it targets a specific scheduler-compatibility code path in the existing CLIP-guided pipelines and is more self-contained.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2723", + "right": "pull_request:2779", + "accept": false, + "reason": "Not the same underlying change: 2723 adds CLIP guidance for stable diffusion img2img, while 2779 adds DPMSolverMultistepScheduler support to the CLIP-guided pipelines. Shared filenames alone do not make them duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 155, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-926-2", + "nodes": [ + "pull_request:2771", + "pull_request:926" + ], + "soft_pairs": [ + "pull_request:2771|pull_request:926" + ], + "prepared_review_unit_hash": "6588bf855d02e037647c7e6b46b97409806eb0526f62ef79a90ced2324caecfb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4740, + "estimated_input_tokens": 1185, + "estimated_eval_tokens": 2626 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4740, + "estimated_input_tokens": 1185, + "estimated_eval_tokens": 2626 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are MPS-related pull requests, but they address different changes: PR #926 is a broader PyTorch 1.13 MPS enablement/update, while PR #2771 removes the warmup pass workaround later on. They share subsystem/context, not the same concrete fix.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #2771 is the more focused artifact: it targets a single MPS behavior change (\u201cremove warmup passes\u201d) and is narrower/specific compared with PR #926\u2019s broader MPS documentation and code updates for PyTorch 1.13.", + "best_issue_reason": null, + "best_pr_reason": "PR #2771 is the best representative of the cluster because its title and diff align cleanly with one concrete MPS issue, making it the most self-contained candidate.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2771", + "right": "pull_request:926", + "accept": false, + "reason": "Related MPS work, but not the same change. #926 updates MPS support and docs for PyTorch 1.13 with code changes; #2771 removes the warmup-pass workaround and updates docs/tests. They would not plausibly be merged as one PR without conflating distinct fixes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 156, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2918-2", + "nodes": [ + "pull_request:2918", + "pull_request:3594" + ], + "soft_pairs": [ + "pull_request:2918|pull_request:3594" + ], + "prepared_review_unit_hash": "cd067c883e8a41c1d6b6ef199475e573ed6d42301646d7e31c2d86b57be44e9b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4699, + "estimated_input_tokens": 1175, + "estimated_eval_tokens": 2606 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4699, + "estimated_input_tokens": 1175, + "estimated_eval_tokens": 2606 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests are LoRA-related, but they address different changes: PR #2918 adds limited LoRA support for the text encoder, while PR #3594 adds a LoRA attention processor for PyTorch 2.0 and related training/doc updates. They are adjacent feature work, not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #2918 is the broader, earlier LoRA integration change and is the better representative for this cluster; PR #3594 is a separate follow-up feature for a different code path.", + "best_issue_reason": null, + "best_pr_reason": "PR #2918 is the best overall representative because it is the more general foundational LoRA PR, whereas PR #3594 is narrower and targets LoRA attention processor 2.0 support.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2918", + "right": "pull_request:3594", + "accept": false, + "reason": "Reject: these PRs touch the same LoRA area but fix different concrete problems. #2918 adds limited text-encoder LoRA loading support; #3594 adds LoRAAttnProcessor2_0 and related training/docs changes. They would not plausibly merge as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 157, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-2963-3", + "nodes": [ + "pull_request:2963", + "pull_request:3492", + "pull_request:3992" + ], + "soft_pairs": [ + "pull_request:2963|pull_request:3492", + "pull_request:3492|pull_request:3992" + ], + "prepared_review_unit_hash": "dbfbed88c00b3c87c79596c36423e0602f6290afeaaf87b695795d486810a5d5", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8356, + "estimated_input_tokens": 2089, + "estimated_eval_tokens": 4434 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 8356, + "estimated_input_tokens": 2089, + "estimated_eval_tokens": 4434 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Three PRs around Diffusers feature additions, but they target different models/features: UniDiffuser, Consistency Models core pipeline, and a Consistency Models distillation example. The overlap is mostly broad subsystem/file-level, not the same underlying change, so no soft duplicate edges are accepted.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No true duplicate PR stands out; PR #3492 is the strongest representative of the cluster because it is the main merged Consistency Models implementation and centralizes the shared scheduler work.", + "best_issue_reason": null, + "best_pr_reason": "PR #3492 is the best cluster representative: it is merged, substantial, and the core implementation PR for Consistency Models, whereas #2963 is a different model family and #3992 is a separate example/training script.", + "soft_edge_verdicts": [ + { + "left": "pull_request:2963", + "right": "pull_request:3492", + "accept": false, + "reason": "They add different features for different models (UniDiffuser vs. Consistency Models). Shared docs/init file edits are broad integration changes, not the same underlying code change." + }, + { + "left": "pull_request:3492", + "right": "pull_request:3992", + "accept": false, + "reason": "#3492 adds the core Consistency Models pipeline/scheduler, while #3992 adds a distillation example training script. They touch one common scheduler file, but they are not the same concrete fix/change and would not be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 158, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3024-2", + "nodes": [ + "pull_request:3024", + "pull_request:3048" + ], + "soft_pairs": [ + "pull_request:3024|pull_request:3048" + ], + "prepared_review_unit_hash": "4819770414e8e2c6480d10ad30acc1a354c5055110df21d77a9791810dfede54", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4519, + "estimated_input_tokens": 1130, + "estimated_eval_tokens": 2516 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4519, + "estimated_input_tokens": 1130, + "estimated_eval_tokens": 2516 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are related at a high level (time embedding behavior), but they change different code paths and are not duplicates: PR 3024 modifies ResNet time-activation handling and output scaling, while PR 3048 adds a UNet time embedding activation option.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR for deduping: the changes target different modules and different mechanisms, so they should not be collapsed into one artifact.", + "best_issue_reason": null, + "best_pr_reason": "PR 3048 is the more focused representative of the shared theme because it directly introduces a time embedding activation option; PR 3024 also bundles unrelated ResNet skip-activation and output-scale changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3024", + "right": "pull_request:3048", + "accept": false, + "reason": "Different concrete changes in different code paths: ResNet skip-time-activation/output-scale vs UNet time_embedding_act_fn. Related theme, but not the same fix/change and not mergeable as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 159, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3083-2", + "nodes": [ + "pull_request:3083", + "pull_request:4180" + ], + "soft_pairs": [ + "pull_request:3083|pull_request:4180" + ], + "prepared_review_unit_hash": "5032ccf868ac90f794e737703fdb1b9c0a5f920cf0fb6adb37e327c2701d9e40", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3173, + "estimated_input_tokens": 794, + "estimated_eval_tokens": 1844 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3173, + "estimated_input_tokens": 794, + "estimated_eval_tokens": 1844 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are both in ControlNet example/docs code, but they address different changes: one updates a Flax pipeline doc example to use a flax revision instead of from_pt, and the other removes safety_checker=None from an SDXL training example. They are not the same underlying fix.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:3083", + "right": "pull_request:4180", + "accept": false, + "reason": "Different edits and purposes: PR 3083 is a Flax doc example tweak, while PR 4180 changes a training example's validation pipeline. They do not fix the same code-path problem and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 160, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3084-2", + "nodes": [ + "pull_request:3084", + "pull_request:3459" + ], + "soft_pairs": [ + "pull_request:3084|pull_request:3459" + ], + "prepared_review_unit_hash": "62b4e52d8cdbeb9e8e9bcd96d58c00a228dfd9b47f17a615efff67bb5c90f92a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4274, + "estimated_input_tokens": 1069, + "estimated_eval_tokens": 2394 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4274, + "estimated_input_tokens": 1069, + "estimated_eval_tokens": 2394 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests implement the same Min-SNR weighting addition for `examples/text_to_image/train_text_to_image_lora.py`; PR 3459 is the merged follow-up and appears to supersede PR 3084.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 3459 is the best canonical PR because it is merged and makes the same code change as PR 3084: adding `--snr_gamma` and SNR computation to the LoRA text-to-image training script.", + "best_issue_reason": null, + "best_pr_reason": "PR 3459 is the strongest representative of the cluster because it is the merged implementation of the Min-SNR feature, while PR 3084 is an earlier unmerged version of the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3084", + "right": "pull_request:3459", + "accept": true, + "reason": "Same underlying feature/change in the same file with nearly identical diff: both add `snr_gamma` and Min-SNR loss weighting to the LoRA training script. They could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 161, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3308-3", + "nodes": [ + "pull_request:3308", + "pull_request:3783", + "pull_request:3845" + ], + "soft_pairs": [ + "pull_request:3308|pull_request:3783", + "pull_request:3308|pull_request:3845", + "pull_request:3783|pull_request:3845" + ], + "prepared_review_unit_hash": "97e671eecc243ee84105f30fd2fff9b4cac43703fe71574987d3f7cda48caff2", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 8061, + "estimated_input_tokens": 2016, + "estimated_eval_tokens": 4288 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 8061, + "estimated_input_tokens": 2016, + "estimated_eval_tokens": 4288 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related Kandinsky PRs, but they are not duplicates: one is the initial Kandinsky 2.1 integration, one adds DDPM scheduler support for the text-to-image path, and one refactors the img2img path for DDPM variance handling.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #3308 is the broad foundational Kandinsky 2.1 implementation and the clearest umbrella change; the other two are later follow-ups on specific pipeline behavior.", + "best_issue_reason": null, + "best_pr_reason": "PR #3308 is the best representative of the cluster because it introduces the core feature set, while #3783 and #3845 are narrower incremental changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3308", + "right": "pull_request:3783", + "accept": false, + "reason": "#3308 adds the full Kandinsky 2.1 feature set; #3783 is a later, narrow DDPM scheduler/docs update for the text-to-image pipeline. Related, but not the same concrete change." + }, + { + "left": "pull_request:3308", + "right": "pull_request:3845", + "accept": false, + "reason": "#3308 is the initial Kandinsky integration; #3845 only refactors img2img DDPM variance handling. Different code path and different fix scope." + }, + { + "left": "pull_request:3783", + "right": "pull_request:3845", + "accept": false, + "reason": "Both mention DDPM Kandinsky, but one changes text-to-image scheduler behavior/docs and the other changes img2img variance handling. Too different to treat as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 162, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3330-3", + "nodes": [ + "pull_request:3330", + "pull_request:3373", + "pull_request:3426" + ], + "soft_pairs": [ + "pull_request:3330|pull_request:3373", + "pull_request:3373|pull_request:3426" + ], + "prepared_review_unit_hash": "394d0a74e2404d0b160933a2da484282c42b0c315e4a402b230a2c956dd919f6", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7224, + "estimated_input_tokens": 1806, + "estimated_eval_tokens": 3868 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7224, + "estimated_input_tokens": 1806, + "estimated_eval_tokens": 3868 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three are separate Kandinsky feature PRs: decoder/model support, image encoder + multilingual CLIP, and img2img pipeline. They share the same subsystem and one shared file, but they do not fix the same concrete bug/change and would not plausibly be merged as one PR.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single PR is a clear duplicate-canonical choice; each PR targets a distinct Kandinsky capability with different codepaths and deliverables.", + "best_issue_reason": null, + "best_pr_reason": "PR 3426 is the most self-contained user-facing feature addition (img2img pipeline plus exports/tests), so it is the best representative artifact among these, but it is not a duplicate of the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3330", + "right": "pull_request:3373", + "accept": false, + "reason": "Different changes: 3330 adds the Kandinsky decoder/VQ model path, while 3373 adds the image encoder and MultilingualCLIP model. They touch the same pipeline file but address separate components and are not the same concrete fix." + }, + { + "left": "pull_request:3373", + "right": "pull_request:3426", + "accept": false, + "reason": "3373 introduces the image encoder/Multi-clip model, whereas 3426 adds the Kandinsky img2img pipeline and exports/tests. Shared subsystem only; different feature scopes and codepaths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 163, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3414-2", + "nodes": [ + "pull_request:3414", + "pull_request:3415" + ], + "soft_pairs": [ + "pull_request:3414|pull_request:3415" + ], + "prepared_review_unit_hash": "a9bd58c3ada49e0769151c7dbb3f0c8f4b677f850a16855ae3cf2c6fcdce823d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3783, + "estimated_input_tokens": 946, + "estimated_eval_tokens": 2148 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3783, + "estimated_input_tokens": 946, + "estimated_eval_tokens": 2148 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two CI workflow PRs make the same change: stop installing transformers/accelerate from GitHub source in test workflows. They are close enough to cluster as one change set.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #3415 is the better representative because it covers the same source-install cleanup and also updates the conda-based workflow to use released pip packages, making it slightly more complete.", + "best_issue_reason": null, + "best_pr_reason": "PR #3415 is the strongest standalone representative of this change because it addresses the same underlying CI dependency-install problem across its workflow path and includes the more complete package-install replacement.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3414", + "right": "pull_request:3415", + "accept": true, + "reason": "Both PRs target the same underlying CI issue: removing GitHub source installs of transformers/accelerate from workflow dependency setup. They are functionally compatible parts of one broader workflow cleanup and could plausibly be merged into a single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 164, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3435-2", + "nodes": [ + "pull_request:3435", + "pull_request:3508" + ], + "soft_pairs": [ + "pull_request:3435|pull_request:3508" + ], + "prepared_review_unit_hash": "0bb0e1c420eff651049e316faef7f101c1d7dd5b59a1ea70e094e1aa655b4743", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4405, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4405, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related community example PRs around the Stable Diffusion reference pipeline, but they are not the same change: PR 3435 adds the reference-only control example, while PR 3508 adds a separate ControlNet reference example and updates the README accordingly.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR for duplication: 3435 and 3508 touch overlapping files but introduce different example variants and different functionality.", + "best_issue_reason": null, + "best_pr_reason": "PR 3508 is the broader follow-up and more representative of the later state of the community examples, but it is still distinct from 3435.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3435", + "right": "pull_request:3508", + "accept": false, + "reason": "Related but distinct PRs: 3435 introduces the reference-only control example, whereas 3508 adds a ControlNet reference example and README updates. They do not fix the same concrete code-path problem and would not be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 165, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3437-2", + "nodes": [ + "pull_request:3437", + "pull_request:3756" + ], + "soft_pairs": [ + "pull_request:3437|pull_request:3756" + ], + "prepared_review_unit_hash": "44a286a39298fca201ab91ef06587190f876409166c3d2f951304232175f0ba7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4749, + "estimated_input_tokens": 1188, + "estimated_eval_tokens": 2632 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4749, + "estimated_input_tokens": 1188, + "estimated_eval_tokens": 2632 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs work on Kohya-ss-style LoRA loading, but they are not the same concrete change: #3437 adds limited support and docs/tests, while #3756 is a broader draft refactor for unrestricted loading. This looks like related follow-up work rather than duplicate PRs.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #3437 is the best canonical representative because it is the merged, completed implementation of the feature and establishes the initial code path for Kohya-ss LoRA support. PR #3756 is a later, broader draft and not the same exact change.", + "best_issue_reason": null, + "best_pr_reason": "PR #3437 is the best overall PR to anchor the cluster: it is merged, more complete, and its scope is the clearest match to the shared feature area. PR #3756 is unfinished and appears to extend rather than duplicate it.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3437", + "right": "pull_request:3756", + "accept": false, + "reason": "Related feature area, but not the same underlying change. #3437 implements limited Kohya-ss LoRA support with docs/tests; #3756 is a broader draft loader refactor aimed at unrestricted support. They would not cleanly collapse into one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 166, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3479-2", + "nodes": [ + "pull_request:3479", + "pull_request:3535" + ], + "soft_pairs": [ + "pull_request:3479|pull_request:3535" + ], + "prepared_review_unit_hash": "a985963d9e0b1896b41c7d1799c55ae151b44bf647f867c35a663ccce447f32d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4153, + "estimated_input_tokens": 1039, + "estimated_eval_tokens": 2334 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4153, + "estimated_input_tokens": 1039, + "estimated_eval_tokens": 2334 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related only at a very high level: both fix broadcasted kernel assignment, but they touch different code paths and modules (1D up/downsample blocks vs ResNet 2D conv/transposed-conv). They are not duplicates and should not be merged together.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #3535 is the better canonical PR for this cluster because its title and diff clearly describe the general fix pattern\u2014making explicit broadcasts for assignment\u2014and it applies to the more broadly used ResNet code path.", + "best_issue_reason": null, + "best_pr_reason": "PR #3535 is the best PR representative overall: the change is well-scoped, clearly stated, and fixes the concrete broadcast assignment issue in a widely used model block.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3479", + "right": "pull_request:3535", + "accept": false, + "reason": "Both PRs use the same broadcast-assignment pattern, but they modify different implementations and different tensor ranks/shapes in separate files. They do not appear to fix the same underlying bug or be plausibly merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 167, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-361-2", + "nodes": [ + "pull_request:361", + "pull_request:727" + ], + "soft_pairs": [ + "pull_request:361|pull_request:727" + ], + "prepared_review_unit_hash": "1050cf547764823651539a32769efcb98e26a6f8f9fa436fcab747e3f77d7547", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3094, + "estimated_input_tokens": 774, + "estimated_eval_tokens": 1804 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3094, + "estimated_input_tokens": 774, + "estimated_eval_tokens": 1804 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "PR 361 is the substantive merged implementation for adding accelerate-based low-memory loading; PR 727 is only a later closed PR with the same general theme but no diff shown and a different loading path.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 361 is the only merged, code-changing PR here and it directly updates model loading in `modeling_utils.py` plus tests. PR 727 is closed without changes in the provided data, so it is not a better canonical PR.", + "best_issue_reason": null, + "best_pr_reason": "PR 361 is the strongest representative of this change because it contains the actual implementation and was merged; PR 727 is too underspecified to serve as the main reference.", + "soft_edge_verdicts": [ + { + "left": "pull_request:361", + "right": "pull_request:727", + "accept": false, + "reason": "Both mention low-memory loading with accelerate, but 361 fixes `from_pretrained` model loading while 727 targets pipeline loading. That is not clearly the same concrete code-path change, and 727 has no diff evidence here." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 168, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3668-2", + "nodes": [ + "pull_request:3668", + "pull_request:3875" + ], + "soft_pairs": [ + "pull_request:3668|pull_request:3875" + ], + "prepared_review_unit_hash": "e3143c3e407d66feaf95c025ac43ec64e0907306f1b0bb991ad5026d0baa2892", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4909, + "estimated_input_tokens": 1228, + "estimated_eval_tokens": 2712 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4909, + "estimated_input_tokens": 1228, + "estimated_eval_tokens": 2712 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related LDM3D pull requests, but they are not duplicates: PR 3668 introduces the new LDM3D pipeline/docs, while PR 3875 is a later compatibility fix for new checkpoints in image processing. The shared files reflect the same subsystem, not the same change.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 3668 is the foundational feature PR for adding LDM3D support, including the pipeline, docs, exports, and tests. PR 3875 is a narrower follow-up fix rather than the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR 3668 is the best representative PR for the cluster because it captures the primary LDM3D feature introduction; PR 3875 is a secondary maintenance update for checkpoint compatibility.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3668", + "right": "pull_request:3875", + "accept": false, + "reason": "They touch the same LDM3D area, but 3668 adds the pipeline and docs while 3875 adjusts depth/image handling for newer checkpoints. These are related but distinct changes and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 169, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3692-2", + "nodes": [ + "pull_request:3692", + "pull_request:4444" + ], + "soft_pairs": [ + "pull_request:3692|pull_request:4444" + ], + "prepared_review_unit_hash": "4635b77db6bf4ecbf2c85b8fafe5d4c4cf1432f6fdd833fbefc1f0074d8aed44", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5938, + "estimated_input_tokens": 1485, + "estimated_eval_tokens": 3226 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5938, + "estimated_input_tokens": 1485, + "estimated_eval_tokens": 3226 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related-image-processing refactors in diffusers, but they address different concrete changes: one adds deprecation warnings to x4 upscaler preprocess helpers, the other refactors prepare_mask_and_masked_image across many pipelines to use VaeImageProcessor. They should not be deduplicated.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #4444 is the better canonical representative because it is the broader, more general VaeImageProcessor refactor spanning many pipelines and tests, whereas #3692 is a narrower upscaler-specific deprecation cleanup.", + "best_issue_reason": null, + "best_pr_reason": "PR #4444 is the best standalone PR for the cluster since it covers the main image-processor migration work across the codebase; #3692 is a smaller, separate upscaler warning refactor and is not the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3692", + "right": "pull_request:4444", + "accept": false, + "reason": "Both touch image-processing code, but the underlying changes differ: #3692 only deprecates preprocess in the x4 upscaler pipelines, while #4444 refactors prepare_mask_and_masked_image and image input handling across many pipelines. They are not the same concrete fix and would not be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 170, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3751-3", + "nodes": [ + "pull_request:3751", + "pull_request:4939", + "pull_request:4941" + ], + "soft_pairs": [ + "pull_request:3751|pull_request:4939", + "pull_request:4939|pull_request:4941" + ], + "prepared_review_unit_hash": "4a2c9e8596e926aa8d93e75258ae20984e70c8e2fdb07ed6364410126ccf920e", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6201, + "estimated_input_tokens": 1551, + "estimated_eval_tokens": 3358 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6201, + "estimated_input_tokens": 1551, + "estimated_eval_tokens": 3358 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are related only by broad Stable Diffusion pipeline work, not by the same bug or change. PR 3751 adds LoRA loading to the KDiffusion pipeline, PR 4939 adds LoRA loading to the SDXL Adapter pipeline, and PR 4941 adds negative micro-conditioning arguments to that same SDXL Adapter pipeline.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No true duplicate PR exists here. If one PR must anchor the cluster, PR 4939 is the closest representative because it is a concrete SDXL Adapter pipeline feature change and shares the same file/context with PR 4941, but it is still a different change from both other PRs.", + "best_issue_reason": null, + "best_pr_reason": "PR 4939 is the best global representative of the cluster because it is a substantial, clearly-scoped pipeline enhancement and sits in the shared SDXL Adapter codepath; however, it should not be treated as a duplicate of PR 4941 or PR 3751.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3751", + "right": "pull_request:4939", + "accept": false, + "reason": "Different pipelines and different features: KDiffusion LoRA loading vs SDXL Adapter LoRA loading. Same general LoRA theme, but not the same code-path problem or mergeable as one PR." + }, + { + "left": "pull_request:4939", + "right": "pull_request:4941", + "accept": false, + "reason": "Same file, but the changes fix different things: LoRA loading/unloading for the adapter pipeline versus adding negative micro-conditioning parameters. These are separate enhancements, not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 171, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3904-2", + "nodes": [ + "pull_request:3904", + "pull_request:3905" + ], + "soft_pairs": [ + "pull_request:3904|pull_request:3905" + ], + "prepared_review_unit_hash": "626872c355e2458e051813e52a7b2e13d0c3b88618c5c265dd9d50951450f8a1", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6257, + "estimated_input_tokens": 1565, + "estimated_eval_tokens": 3386 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6257, + "estimated_input_tokens": 1565, + "estimated_eval_tokens": 3386 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two docs-oriented PRs in the same pipelines area, but they address different cleanup scopes: one is a narrower API docs wording update, the other is a broader pipeline docs/title normalization pass. I would not treat them as duplicates.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #3905 is the stronger canonical representative: it is merged, has much broader review/activity, and covers the larger, finished documentation cleanup across pipeline pages.", + "best_issue_reason": null, + "best_pr_reason": "PR #3905 is the best overall representative because it is completed and clearly the broader, more central docs cleanup change; PR #3904 is draft and narrower.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3904", + "right": "pull_request:3905", + "accept": false, + "reason": "Both are docs cleanup PRs with overlapping pipeline files, but they do not appear to fix the same concrete issue; #3904 is a narrower wording/API docs tweak while #3905 is a broader title and docs normalization pass." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 172, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-3930-2", + "nodes": [ + "pull_request:3930", + "pull_request:3949" + ], + "soft_pairs": [ + "pull_request:3930|pull_request:3949" + ], + "prepared_review_unit_hash": "9483fb5f6c294a8cd68da4dee181c9a53fc050b1265d7bc9be46567e6d22a530", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4926, + "estimated_input_tokens": 1232, + "estimated_eval_tokens": 2720 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4926, + "estimated_input_tokens": 1232, + "estimated_eval_tokens": 2720 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related text-to-video PRs, but they fix different problems: one improves memory use via attention chunking, the other adds `torch.compile()` compatibility by changing return handling. Not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 3930 is the stronger canonical representative because it is the more substantive text-to-video memory optimization and its diff is centered on the core attention/unet path; PR 3949 is a separate compile-compatibility change.", + "best_issue_reason": null, + "best_pr_reason": "PR 3930 best represents the cluster\u2019s text-to-video runtime improvement theme, but the cluster is not a true duplicate set, so this is only a representative choice.", + "soft_edge_verdicts": [ + { + "left": "pull_request:3930", + "right": "pull_request:3949", + "accept": false, + "reason": "Different underlying changes: 3930 adds feed-forward chunking/memory optimizations, while 3949 makes 3D blocks/pipelines `torch.compile()`-compatible by returning tuples instead of `.sample`. Shared files alone are not enough to treat them as the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 173, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4032-2", + "nodes": [ + "pull_request:4032", + "pull_request:4454" + ], + "soft_pairs": [ + "pull_request:4032|pull_request:4454" + ], + "prepared_review_unit_hash": "4ecaef953ab7d738e128a970f6050f8c4da7b0eac5ce3083462d3f50af9d36ab", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4907, + "estimated_input_tokens": 1227, + "estimated_eval_tokens": 2710 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4907, + "estimated_input_tokens": 1227, + "estimated_eval_tokens": 2710 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address singleton MultiControlNet handling in ControlNet pipelines, but they fix different concrete failures: PR 4032 fixes keep/list shape selection, while PR 4454 fixes conditioning-scale typing during multiplication. They are related follow-ups, not duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: both are merged fixes for singleton MultiControlNet behavior, but they patch different code paths and do not subsume each other.", + "best_issue_reason": null, + "best_pr_reason": "PR 4454 is the narrower, more explicit singleton-MultiControlNet TypeError fix, but it does not cover PR 4032's separate keep-shape bug.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4032", + "right": "pull_request:4454", + "accept": false, + "reason": "Different failure sites and fixes: 4032 changes how controlnet_keep is shaped for a singleton ControlNetModel, while 4454 coerces controlnet_conditioning_scale before scalar multiplication. Same subsystem, but not the same underlying bug or mergeable change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 174, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4076-2", + "nodes": [ + "pull_request:4076", + "pull_request:4079" + ], + "soft_pairs": [ + "pull_request:4076|pull_request:4079" + ], + "prepared_review_unit_hash": "d203a3d179fa5b014fa022f54b9b3202a065bf2d84a367aa269b02f4c143544c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5075, + "estimated_input_tokens": 1269, + "estimated_eval_tokens": 2794 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5075, + "estimated_input_tokens": 1269, + "estimated_eval_tokens": 2794 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about adding Stable Diffusion XL InstructPix2Pix support. PR #4079 is the more complete and merged version, containing the same core code changes as #4076 plus docs, initialization, dummy objects, and tests.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #4079 is the canonical PR because it is the merged, more complete implementation of the SDXL InstructPix2Pix feature and includes the core files from #4076 plus the supporting integration work.", + "best_issue_reason": null, + "best_pr_reason": "PR #4079 is the best PR overall because it is merged and appears to subsume #4076's change set while adding the necessary surrounding updates and tests.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4076", + "right": "pull_request:4079", + "accept": true, + "reason": "Same underlying SDXL InstructPix2Pix feature/change; #4079 contains the same core files as #4076 and extends them with the remaining supporting changes, so these could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 175, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4221-2", + "nodes": [ + "pull_request:4221", + "pull_request:6046" + ], + "soft_pairs": [ + "pull_request:4221|pull_request:6046" + ], + "prepared_review_unit_hash": "ee3a428c9ac39deae79e274c966647f124e12b201cacc7998f5cd7d2b0d7244e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3876, + "estimated_input_tokens": 969, + "estimated_eval_tokens": 2194 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3876, + "estimated_input_tokens": 969, + "estimated_eval_tokens": 2194 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate PRs in the same loader area, but they fix different bugs: one passes `network_alpha` for old-format UNet LoRA loading, the other makes IP-Adapter/LoRA loading work when the UNet attribute is not named `unet`.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #6046 is the better representative of the cluster because it addresses the broader custom-UNet-name compatibility issue across loader code paths and has the stronger multi-file fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #6046 is the best overall artifact here since it is the broader, more general loader compatibility fix and more likely to be the canonical representative of the related-but-distinct loader work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4221", + "right": "pull_request:6046", + "accept": false, + "reason": "Different underlying bugs and code paths: #4221 fixes missing `network_alpha` when loading old-format LoRA weights into UNet, while #6046 fixes loader access when the UNet attribute is not literally named `unet` (plus related scale handling). They are not plausibly one mergeable PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 176, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4369-2", + "nodes": [ + "pull_request:4369", + "pull_request:4411" + ], + "soft_pairs": [ + "pull_request:4369|pull_request:4411" + ], + "prepared_review_unit_hash": "9317a9541b16f981ce142302cdea9a12d20f48f2a177b89e2d16b6361cf84ce8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4076, + "estimated_input_tokens": 1019, + "estimated_eval_tokens": 2294 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4076, + "estimated_input_tokens": 1019, + "estimated_eval_tokens": 2294 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate cleanup PRs in different code paths: one removes duplicate lines in SDXL prompt encoding, the other removes duplicated ControlNet img2img code. They are not the same underlying change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:4369", + "right": "pull_request:4411", + "accept": false, + "reason": "Different bugs and files: PR 4369 fixes duplicate encode_prompt lines across SDXL pipelines, while PR 4411 removes a duplicated ControlNet img2img assignment. They are cleanup-only but not the same code-path problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 177, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4415-2", + "nodes": [ + "pull_request:4415", + "pull_request:4455" + ], + "soft_pairs": [ + "pull_request:4415|pull_request:4455" + ], + "prepared_review_unit_hash": "b76d4c9eccf7ea36347155553c358fcd113a153ecf4995e5597469cbba0572cc", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4672, + "estimated_input_tokens": 1168, + "estimated_eval_tokens": 2592 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4672, + "estimated_input_tokens": 1168, + "estimated_eval_tokens": 2592 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are test-only flaky-test cleanups in Stable Diffusion and both add `set_default_attn_processor()`, but they target different test cases and code paths. They look related by theme, not like one duplicate change that should be merged together.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4415 is the broader, earlier cleanup across more Stable Diffusion tests; 4455 is a narrower follow-up focused on attention-slicing tests, so there is no true duplicate canonical PR.", + "best_issue_reason": null, + "best_pr_reason": "PR 4415 is the best representative of the cluster because it spans more of the shared flaky-test cleanup work and captures the broader pattern behind the changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4415", + "right": "pull_request:4455", + "accept": false, + "reason": "Same general Stable Diffusion test-flakiness theme, but different concrete test paths and fixes. 4415 focuses on compile-related image tests across several pipelines; 4455 focuses on attention-slicing behavior. Too broad and separate to treat as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 178, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4475-2", + "nodes": [ + "pull_request:4475", + "pull_request:4493" + ], + "soft_pairs": [ + "pull_request:4475|pull_request:4493" + ], + "prepared_review_unit_hash": "4e28f819efa5b3c60609118a3ab33b3d233077490c31058a9a464cd3ee2115d5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4231, + "estimated_input_tokens": 1058, + "estimated_eval_tokens": 2372 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4231, + "estimated_input_tokens": 1058, + "estimated_eval_tokens": 2372 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests make the same DreamBooth TensorBoard logging change (use `global_step` instead of `epoch` in `train_dreambooth.py`). PR #4493 is the canonical PR because it is the merged copy of the identical change.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #4493 is the merged instance of the same patch and title as #4475, so it is the best canonical PR for this change.", + "best_issue_reason": null, + "best_pr_reason": "PR #4493 is the strongest representative because it was merged and its diff matches #4475 exactly.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4475", + "right": "pull_request:4493", + "accept": true, + "reason": "Identical title, identical file change, and identical diff replacing `epoch` with `global_step`; these are the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 179, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4485-2", + "nodes": [ + "pull_request:4485", + "pull_request:4486" + ], + "soft_pairs": [ + "pull_request:4485|pull_request:4486" + ], + "prepared_review_unit_hash": "b7dfbf9daee6150e8433aee4e27a43097b97a2cf9db0ab2b6d34e114513c983e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4187, + "estimated_input_tokens": 1047, + "estimated_eval_tokens": 2350 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4187, + "estimated_input_tokens": 1047, + "estimated_eval_tokens": 2350 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two closed PRs appear to be near-duplicate fixes for the same DreamBooth LoRA SDXL batch-size/data-loader issue in the same file, with essentially the same code changes.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #4486 is the better canonical PR because its title clearly states the concrete fix ('training with arbitrary batch sizes') and its diff matches the underlying bug fix more explicitly than the vague 'Lora fix' in #4485.", + "best_issue_reason": null, + "best_pr_reason": "PR #4486 is the best representative of the cluster because it is the most descriptively scoped and directly explains the user-visible problem being fixed, while still matching the shared code-path changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4485", + "right": "pull_request:4486", + "accept": true, + "reason": "Both PRs modify the same file with essentially the same DataLoader changes (shuffle/drop_last) to address the same training/batch-size behavior, so they look like duplicate fixes for the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 180, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4544-2", + "nodes": [ + "pull_request:4544", + "pull_request:4590" + ], + "soft_pairs": [ + "pull_request:4544|pull_request:4590" + ], + "prepared_review_unit_hash": "77cd1c9d60847118b213f984686aa2782da2c9c9eb579a81a229940a10664ded", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2962, + "estimated_input_tokens": 741, + "estimated_eval_tokens": 1738 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2962, + "estimated_input_tokens": 741, + "estimated_eval_tokens": 1738 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two pull requests are effectively duplicates: same title, same diff, same file, and same fix for `pipeline_class_name`. The merged PR is the better canonical representative.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4590 is the better canonical PR because it is the merged version of the exact same code change and therefore represents the resolved history of this fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 4590 is the best PR overall because it was merged and has the same implementation as PR 4544, making it the definitive instance of this change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4544", + "right": "pull_request:4590", + "accept": true, + "reason": "Accept: same title, same target file, same patch, and same underlying bug fix; one is simply the merged duplicate of the other." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 181, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4558-2", + "nodes": [ + "pull_request:4558", + "pull_request:4603" + ], + "soft_pairs": [ + "pull_request:4558|pull_request:4603" + ], + "prepared_review_unit_hash": "73c6676d38ccd9849e07c716a7a5518edbeb44102916e2b65c03d07a11a2bb99", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4460, + "estimated_input_tokens": 1115, + "estimated_eval_tokens": 2486 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4460, + "estimated_input_tokens": 1115, + "estimated_eval_tokens": 2486 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs patch the same guidance_rescale default-value bug in `pipeline_stable_diffusion.py`; PR 4558 is the broader, issue-linked fix and is the better canonical choice.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4558 is the stronger canonical PR: it has an explicit issue link, broader coverage across related pipelines/examples/tests, and more review/discussion activity while fixing the same default-value bug.", + "best_issue_reason": null, + "best_pr_reason": "PR 4558 best represents the underlying fix because it addresses the bug across multiple affected pipelines and includes validation, making it more complete than the narrow single-file follow-up in PR 4603.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4558", + "right": "pull_request:4603", + "accept": true, + "reason": "Accept: both PRs fix the same concrete bug\u2014incorrect `guidance_rescale` default in the Stable Diffusion code path\u2014and the narrower PR 4603 is a subset of the broader fix in PR 4558." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both PRs change the `guidance_rescale` default from `0.0` to `0.7` in `pipeline_stable_diffusion.py`, and PR 4558 also extends that same fix to additional pipelines/examples/tests with an explicit issue link. The soft-edge accept is conservative because PR 4603 looks like a narrow subset of the broader same-bug fix in PR 4558, so the pair is plausibly mergeable into one PR." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 182, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-471-8", + "nodes": [ + "pull_request:471", + "pull_request:611", + "pull_request:618", + "pull_request:619", + "pull_request:629", + "pull_request:631", + "pull_request:680", + "pull_request:683" + ], + "soft_pairs": [ + "pull_request:471|pull_request:683", + "pull_request:611|pull_request:683", + "pull_request:611|pull_request:619", + "pull_request:611|pull_request:629", + "pull_request:618|pull_request:629", + "pull_request:619|pull_request:631", + "pull_request:619|pull_request:629", + "pull_request:629|pull_request:631", + "pull_request:631|pull_request:680" + ], + "prepared_review_unit_hash": "3ae356ea11a13d675c33921aa6949010e95712117ef9ed939bb0c230464b110d", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 9, + "serialized_chars": 17922, + "estimated_input_tokens": 4481, + "estimated_eval_tokens": 9218 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 9, + "serialized_chars": 17922, + "estimated_input_tokens": 4481, + "estimated_eval_tokens": 9218 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This set is a loose theme cluster of documentation/readability/refactor PRs, but they are not duplicates of the same underlying change. The only close pairs share generic cleanup language, not the same code path or doc section.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "No true canonical PR: these are separate cleanup changes affecting different files and sections, so none safely subsumes the others as a duplicate anchor.", + "best_issue_reason": null, + "best_pr_reason": "PR 629 is the broadest representative cleanup in the set, spanning multiple source/docs files, but it still does not duplicate the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:471", + "right": "pull_request:683", + "accept": false, + "reason": "Both are docs table-formatting fixes, but they touch different docs pages and different pipeline topics; same style of edit, not the same change." + }, + { + "left": "pull_request:611", + "right": "pull_request:683", + "accept": false, + "reason": "One fixes a copyright typo in scheduler files; the other rewrites a docs table. Related only by being small cleanup PRs." + }, + { + "left": "pull_request:611", + "right": "pull_request:619", + "accept": false, + "reason": "Both are minor cleanup PRs, but they modify different files and different text; no shared bug or concrete fix." + }, + { + "left": "pull_request:611", + "right": "pull_request:629", + "accept": false, + "reason": "These are unrelated cleanup edits in different code areas; the overlap is only generic wording/readability improvement." + }, + { + "left": "pull_request:618", + "right": "pull_request:629", + "accept": false, + "reason": "One updates a docs reference link, the other makes broad readability edits across Flax/utils files; not the same issue." + }, + { + "left": "pull_request:619", + "right": "pull_request:631", + "accept": false, + "reason": "Both are readability fixups, but they touch completely different files and content; too broad to be a duplicate." + }, + { + "left": "pull_request:619", + "right": "pull_request:629", + "accept": false, + "reason": "Shared theme is documentation/readability cleanup, but the file sets and edits are distinct and not mergeable as one fix." + }, + { + "left": "pull_request:629", + "right": "pull_request:631", + "accept": false, + "reason": "Both are cleanup PRs, but 629 targets multiple diffusers source/docs files while 631 only fixes custom_init_isort comments/identifiers." + }, + { + "left": "pull_request:631", + "right": "pull_request:680", + "accept": false, + "reason": "One is a refactor/readability cleanup; the other changes a specific ldm-bert config URL. They address different concrete changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 183, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4776-2", + "nodes": [ + "pull_request:4776", + "pull_request:6988" + ], + "soft_pairs": [ + "pull_request:4776|pull_request:6988" + ], + "prepared_review_unit_hash": "dd88bdd230591495f195333a5bd8ce5fc42f3312372e9385b1e01c7e38c4e88e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4206, + "estimated_input_tokens": 1052, + "estimated_eval_tokens": 2360 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4206, + "estimated_input_tokens": 1052, + "estimated_eval_tokens": 2360 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are not duplicates: one adds denoising_end support to the ControlNet SDXL pipeline, while the other only updates the Stable Diffusion inpaint pipeline example/docstring. They touch different code paths and solve different problems.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear canonical PR for a duplicate set here; the PRs are unrelated. If forced, PR #4776 is the more substantive code-change candidate, but it does not match #6988.", + "best_issue_reason": null, + "best_pr_reason": "PR #4776 is the stronger standalone change because it adds actual pipeline behavior (denoising_end handling) rather than a documentation/example update, but it is not a valid duplicate partner for #6988.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4776", + "right": "pull_request:6988", + "accept": false, + "reason": "Different pipelines and different kinds of changes: ControlNet SDXL denoising logic vs Stable Diffusion inpaint example/docstring. They do not look like the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 184, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4810-2", + "nodes": [ + "pull_request:4810", + "pull_request:4891" + ], + "soft_pairs": [ + "pull_request:4810|pull_request:4891" + ], + "prepared_review_unit_hash": "3739a69fed67d7de884b5bade2c8a4490b97b677adbcdb981178c72ec43d0155", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5439, + "estimated_input_tokens": 1360, + "estimated_eval_tokens": 2976 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5439, + "estimated_input_tokens": 1360, + "estimated_eval_tokens": 2976 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests target the same bug: handling image latents when the VAE returns latents directly instead of a latent distribution. PR 4891 is the cleaner canonical choice because it includes the concrete code fix and tests, while PR 4810 appears broader and more documentation-heavy.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4891 is the best canonical PR because it directly patches the affected pipeline code paths and adds tests for the latent-retrieval behavior.", + "best_issue_reason": null, + "best_pr_reason": "PR 4891 is the strongest representative of this change set: it fixes the concrete sampling bug in the relevant pipelines and verifies it with tests.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4810", + "right": "pull_request:4891", + "accept": true, + "reason": "Same underlying bug and same fix theme across the same image-to-image/inpaint code paths; the PRs are sufficiently identical to be treated as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 185, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4903-2", + "nodes": [ + "pull_request:4903", + "pull_request:4994" + ], + "soft_pairs": [ + "pull_request:4903|pull_request:4994" + ], + "prepared_review_unit_hash": "dc6c52dae1ddab1eee12a3c3710907a99361ec9573d0bc220f9080c3cfef1417", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4680, + "estimated_input_tokens": 1170, + "estimated_eval_tokens": 2596 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4680, + "estimated_input_tokens": 1170, + "estimated_eval_tokens": 2596 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are both merged Diffusers PRs that touch `src/diffusers/loaders.py`, but they address different changes: PR #4903 relaxes textual inversion loading for SDXL pipelines, while PR #4994 speeds up LoRA loading and refactors loader internals. They do not look like the same underlying fix and should not be clustered together as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No strong canonical PR choice for duplicate-triage purposes; #4994 is broader and more central to loader performance, but it is not the same change as #4903.", + "best_issue_reason": null, + "best_pr_reason": "PR #4994 is the better representative only in a loose topical sense because it has broader loader impact and more references, but it is not a duplicate of #4903 and should not be merged with it.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4903", + "right": "pull_request:4994", + "accept": false, + "reason": "Different features/fixes: #4903 changes textual inversion loading and SDXL pipeline handling, while #4994 targets LoRA loading speed and related loader internals. Shared file `loaders.py` is too broad to indicate the same underlying bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 186, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4917-2", + "nodes": [ + "pull_request:4917", + "pull_request:5528" + ], + "soft_pairs": [ + "pull_request:4917|pull_request:5528" + ], + "prepared_review_unit_hash": "6af7e5a7a17f2fd5032c1c8d28ae454bec453e6acd8c55300f623be3ce100d82", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2674, + "estimated_input_tokens": 669, + "estimated_eval_tokens": 1594 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2674, + "estimated_input_tokens": 669, + "estimated_eval_tokens": 1594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs fix a similar `torch_dtype` misuse in `from_single_file`, but they target different model paths (`VAE` vs `ControlNetModel`) and are not the same concrete bug.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #4917 is the broader and clearer representative: it fixes the `dtype` argument bug in `from_single_file` with a generic VAE path and the title directly states the issue.", + "best_issue_reason": null, + "best_pr_reason": "PR #4917 is the best representative of the cluster because it addresses the underlying `torch_dtype`/`dtype` bug pattern in the shared loader code path more generally than the ControlNet-specific follow-up.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4917", + "right": "pull_request:5528", + "accept": false, + "reason": "They are related bug fixes in the same file, but they fix different model branches and not the same underlying code-path bug, so they should not be merged as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 187, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4938-3", + "nodes": [ + "pull_request:4938", + "pull_request:5020", + "pull_request:5210" + ], + "soft_pairs": [ + "pull_request:4938|pull_request:5020", + "pull_request:4938|pull_request:5210", + "pull_request:5020|pull_request:5210" + ], + "prepared_review_unit_hash": "12052854ea4e525d4d050b42385fc17a70cf6237d30d79fc6a34df50035aacf3", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6810, + "estimated_input_tokens": 1703, + "estimated_eval_tokens": 3662 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6810, + "estimated_input_tokens": 1703, + "estimated_eval_tokens": 3662 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related docs-only pull requests improving different guides (text-to-image, image-to-image, inpaint), but they do not fix the same underlying bug or implement the same change. Treat them as separate PRs, not duplicates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5210 is the best cluster representative because it is the broadest docs update in the set, touching both img2img and inpaint guides and reflecting the same documentation-improvement theme.", + "best_issue_reason": null, + "best_pr_reason": "PR #5210 is the strongest overall representative of the cluster because it covers multiple docs files and is more central than the single-guide updates, but it is still not a duplicate of the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4938", + "right": "pull_request:5020", + "accept": false, + "reason": "Both are documentation improvements, but one is for text-to-image and the other for image-to-image; they update different guides and do not represent the same change." + }, + { + "left": "pull_request:4938", + "right": "pull_request:5210", + "accept": false, + "reason": "These are separate guide rewrites for different topics (text-to-image vs inpaint/img2img docs), so they are not the same concrete change." + }, + { + "left": "pull_request:5020", + "right": "pull_request:5210", + "accept": false, + "reason": "They overlap on img2img docs wording, but #5210 is primarily an inpaint-doc update plus a shared tip tweak, not the same underlying fix or a mergeable single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 188, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-4995-3", + "nodes": [ + "pull_request:4995", + "pull_request:4997", + "pull_request:4998" + ], + "soft_pairs": [ + "pull_request:4995|pull_request:4997", + "pull_request:4997|pull_request:4998" + ], + "prepared_review_unit_hash": "215bf76ac5ea62f8f2684ceb9791abcc74cb38d1e8b0b88334f0a30ff7ea9fde", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7812, + "estimated_input_tokens": 1953, + "estimated_eval_tokens": 4162 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7812, + "estimated_input_tokens": 1953, + "estimated_eval_tokens": 4162 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly about Stable Diffusion CPU offload behavior, but only one PR is the actual code fix. PR 4998 is the best canonical PR; PR 4997 is a test-only adjustment around offloading, and PR 4995 is unrelated image-variation test cleanup.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 4998 directly fixes the offload implementation by excluding safety_checker from CPU offload across affected pipelines, which is the concrete behavior change the cluster centers on.", + "best_issue_reason": null, + "best_pr_reason": "PR 4998 is the strongest representative because it changes the production code path, not just test setup, and it applies the fix across multiple pipelines.", + "soft_edge_verdicts": [ + { + "left": "pull_request:4995", + "right": "pull_request:4997", + "accept": false, + "reason": "Different test fixes for different concerns: 4995 adjusts image-variation test inputs, while 4997 tweaks a Stable Diffusion offload test. They do not appear to address the same underlying bug or change." + }, + { + "left": "pull_request:4997", + "right": "pull_request:4998", + "accept": false, + "reason": "Related offload theme, but not the same change: 4997 is a test-only workaround, while 4998 changes the pipeline implementation to exclude safety_checker from CPU offload. They could coexist, but they are not the same concrete fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 189, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5106-2", + "nodes": [ + "pull_request:5106", + "pull_request:5107" + ], + "soft_pairs": [ + "pull_request:5106|pull_request:5107" + ], + "prepared_review_unit_hash": "687f0b855a38b067c23d407b21bcada0ed6922e2b5d939f74fcccf625923e95e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4417, + "estimated_input_tokens": 1105, + "estimated_eval_tokens": 2466 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4417, + "estimated_input_tokens": 1105, + "estimated_eval_tokens": 2466 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are related by Min-SNR / v-prediction training logic, but they target different training scripts and code paths. They are not the same fix and should not be merged as duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5106 is the broader, more representative SNR-gamma/v-prediction fix across multiple text-to-image/controlnet training scripts.", + "best_issue_reason": null, + "best_pr_reason": "PR #5106 is the best cluster representative because it addresses the core v-prediction SNR-weighting bug in several example trainers, whereas #5107 is a separate Dreambooth-specific support change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5106", + "right": "pull_request:5107", + "accept": false, + "reason": "Both concern Min-SNR gamma, but #5106 patches v_prediction SNR weighting in text-to-image/controlnet trainers while #5107 adds Dreambooth-specific support with new compute_snr logic; different concrete code paths and not a single mergeable PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 190, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5148-2", + "nodes": [ + "pull_request:5148", + "pull_request:5235" + ], + "soft_pairs": [ + "pull_request:5148|pull_request:5235" + ], + "prepared_review_unit_hash": "f89181d504779e2588ce21593611211510c4a1243cf08eeeb3421873c8eb22a9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4829, + "estimated_input_tokens": 1208, + "estimated_eval_tokens": 2672 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4829, + "estimated_input_tokens": 1208, + "estimated_eval_tokens": 2672 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are related to the new compile-test/Docker image effort, but they address different fixes: one introduces CI/build changes and test exclusions, the other repairs the compile CUDA Dockerfile and a specific test. They are adjacent follow-ups, not the same mergeable change.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5235 is the more focused representative of the cluster because it directly fixes the compile test environment via the Dockerfile and a targeted test adjustment, whereas #5148 is broader CI/test plumbing work.", + "best_issue_reason": null, + "best_pr_reason": "PR #5235 best represents the underlying theme of compile-test breakage and environment repair; it is narrower and more concrete than #5148, which mainly adds CI wiring and skips compile tests from push tests.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5148", + "right": "pull_request:5235", + "accept": false, + "reason": "Both concern compile-testing infrastructure, but #5148 adds the compile CUDA image/CI workflow changes and skips compile tests, while #5235 fixes the Dockerfile Python setup and one test. They are related follow-up changes, not the same concrete bug or a single PR merge candidate." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 191, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5174-2", + "nodes": [ + "pull_request:5174", + "pull_request:5183" + ], + "soft_pairs": [ + "pull_request:5174|pull_request:5183" + ], + "prepared_review_unit_hash": "7b8cdabbf076d4efe448fe5eb6c84aae91c00eefdc7d12203b98737ddc7d8902", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4483, + "estimated_input_tokens": 1121, + "estimated_eval_tokens": 2498 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4483, + "estimated_input_tokens": 1121, + "estimated_eval_tokens": 2498 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two pull requests about model CPU offload, but they address different bugs: one is BLIPDiffusion-specific offload plumbing, the other fixes the shared offload GPU-id handling in pipeline_utils with tests. They are not duplicate changes.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5183 is the better canonical representative because it fixes the core offload implementation in pipeline_utils and includes regression tests, making it the broader and more central bug fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #5183 is the strongest overall representative of the cluster since it corrects a shared CPU offload code path rather than a single pipeline-specific integration.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5174", + "right": "pull_request:5183", + "accept": false, + "reason": "Reject: #5174 is a BLIPDiffusion-specific offload sequence/device fix, while #5183 changes generic offload GPU-id selection in pipeline_utils and related tests. Different underlying bugs and not plausibly one merged PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 192, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5214-2", + "nodes": [ + "pull_request:5214", + "pull_request:5305" + ], + "soft_pairs": [ + "pull_request:5214|pull_request:5305" + ], + "prepared_review_unit_hash": "b9ef5495d33440806b52eb7476ead00d04307c5c6462eff22e1eefadddf2d8b0", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6149, + "estimated_input_tokens": 1538, + "estimated_eval_tokens": 3332 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6149, + "estimated_input_tokens": 1538, + "estimated_eval_tokens": 3332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests are effectively the same fix for the GLIGEN in-painting 0-d tensor iteration TypeError, target the same issue, and touch the same code paths/files. PR #5305 is the merged/final version, while #5214 is an earlier unmerged duplicate.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5305 is the better canonical PR because it is merged, has the same issue target and nearly identical scope as #5214, and appears to be the finalized version of the same fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #5305 is the strongest representative because it is merged and implements the same concrete bug fix more definitively than the unmerged earlier duplicate #5214.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5214", + "right": "pull_request:5305", + "accept": true, + "reason": "Same title/issue target, same core GLIGEN in-painting bug, and overlapping file changes indicate the same underlying change; #5305 is the merged continuation of #5214." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 193, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5243-2", + "nodes": [ + "pull_request:5243", + "pull_request:5255" + ], + "soft_pairs": [ + "pull_request:5243|pull_request:5255" + ], + "prepared_review_unit_hash": "7cd3fc526c44d400242acfeb7212fa825e3566bf3331dfbc645d6fa183e00e18", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4675, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4675, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two nearly identical PRs fixing the same Wuerstchen doc/defaults mismatch in the same files; the later merged PR is the best canonical representative.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5255 is the merged version of the same change, touches the same files with the same diff pattern, and has substantially more review activity than #5243.", + "best_issue_reason": null, + "best_pr_reason": "PR #5255 is the best representative because it is merged and appears to be the finalized duplicate of the same fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5243", + "right": "pull_request:5255", + "accept": true, + "reason": "Same underlying change: identical affected files and matching diff preview adjusting the Wuerstchen pipeline docs/defaults. The merged PR #5255 is a direct duplicate/finalized version of #5243." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 194, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-527-2", + "nodes": [ + "pull_request:527", + "pull_request:603" + ], + "soft_pairs": [ + "pull_request:527|pull_request:603" + ], + "prepared_review_unit_hash": "4ab1d9182d04dd575181c1dbaa7e165eff6c71a03de191ea1e44c9fe7297d1ae", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4421, + "estimated_input_tokens": 1106, + "estimated_eval_tokens": 2468 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4421, + "estimated_input_tokens": 1106, + "estimated_eval_tokens": 2468 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged PRs in the same Flax loading area, but they address different changes: PR #527 adds `from_pt` support and related conversion utilities, while PR #603 fixes a specific `from_pretrained` file-existence check bug for PyTorch weights. They are related but not duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #527 is the broader, foundational change that introduces `from_pt` support and the associated Flax/PyTorch conversion path, making it the best representative of the cluster theme.", + "best_issue_reason": null, + "best_pr_reason": "PR #527 is the most representative artifact because it covers the main feature addition; PR #603 is a narrower follow-up bugfix in the same code path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:527", + "right": "pull_request:603", + "accept": false, + "reason": "Not the same underlying change: #527 adds `from_pt` loading support and conversion utilities, while #603 only corrects a conditional/file check in `from_pretrained`. They touch the same file but fix different problems and would not plausibly merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 195, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5278-2", + "nodes": [ + "pull_request:5278", + "pull_request:5279" + ], + "soft_pairs": [ + "pull_request:5278|pull_request:5279" + ], + "prepared_review_unit_hash": "1731c6493ea034b78e05e67506a1b8e26676a5e7de624362873e292b9c1ad0fe", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4400, + "estimated_input_tokens": 1100, + "estimated_eval_tokens": 2456 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4400, + "estimated_input_tokens": 1100, + "estimated_eval_tokens": 2456 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs target SDXL ControlNet training/validation fixes, but they are not the same change set: #5278 is mostly a README/default-model and narrow script tweak, while #5279 changes validation behavior, adds a negative prompt, and also touches t2i_adapter and dependency files.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "#5279 is the better canonical PR candidate because it is broader, has real review activity, and contains the more substantive code-path fix in `train_controlnet_sdxl.py` rather than mostly documentation/default-value edits.", + "best_issue_reason": null, + "best_pr_reason": "#5279 is the strongest PR in the cluster: it addresses the core validation/checkpointing path, has supporting changes in related files, and appears more vetted than #5278.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5278", + "right": "pull_request:5279", + "accept": false, + "reason": "They overlap on `train_controlnet_sdxl.py`, but the concrete edits differ: #5278 changes the model/README and removes a validation-path xformers guard, while #5279 restores xformers handling, adds a negative prompt, and includes unrelated dependency/t2i_adapter updates. Not the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 196, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5288-3", + "nodes": [ + "pull_request:5288", + "pull_request:5366", + "pull_request:5370" + ], + "soft_pairs": [ + "pull_request:5366|pull_request:5370", + "pull_request:5288|pull_request:5366", + "pull_request:5288|pull_request:5370" + ], + "prepared_review_unit_hash": "f6e546aeadad56660f552fe9626c8c66e3dc9a4d40d97fc653602d6130f04ed3", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6749, + "estimated_input_tokens": 1688, + "estimated_eval_tokens": 3632 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6749, + "estimated_input_tokens": 1688, + "estimated_eval_tokens": 3632 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three closely related PRs around shrinking Stable Diffusion fast-test dummy components for issue #5262. PR 5370 is the merged, most complete implementation; 5288 is only a TODO/commentary change; 5366 is an earlier competing implementation with different dummy-model settings.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 5370 is the only merged change and appears to be the final, concrete fix for the fast-test size reduction in the StableDiffusion test fixture.", + "best_issue_reason": null, + "best_pr_reason": "PR 5370 is the best representative because it is merged, touches the exact target code path, and implements the substantive model-size reductions rather than just comments or an alternative draft.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5366", + "right": "pull_request:5370", + "accept": true, + "reason": "Both PRs change the same StableDiffusion fast-test dummy component setup in the same file to reduce model size for issue #5262; 5370 looks like the final version of the same underlying fix." + }, + { + "left": "pull_request:5288", + "right": "pull_request:5366", + "accept": false, + "reason": "5288 only adds a TODO/comment about making the fast test smaller, while 5366 makes actual component-config changes; they are related but not the same change." + }, + { + "left": "pull_request:5288", + "right": "pull_request:5370", + "accept": false, + "reason": "5288 is just a placeholder/commentary PR, whereas 5370 contains the real code changes to the test fixture; not the same underlying change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary stays within the packet: 5288 is only a comment/TODO, 5370 is the merged implementation, and 5366 is an earlier alternative change in the same test file tied to the same issue. The 5366\u21945370 linkage is conservative enough given the shared target file and issue." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 197, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5293-2", + "nodes": [ + "pull_request:5293", + "pull_request:5389" + ], + "soft_pairs": [ + "pull_request:5293|pull_request:5389" + ], + "prepared_review_unit_hash": "b40008559b1bd16405fffa1c7ac58684cde46ef257a6a0acc0543537138eb7a5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4039, + "estimated_input_tokens": 1010, + "estimated_eval_tokens": 2276 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4039, + "estimated_input_tokens": 1010, + "estimated_eval_tokens": 2276 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are independent documentation-only PRs in the same file, but they target different classes and different docstrings, so they are not duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5389 is a broader, clearer representation of the docstring-only change pattern in this cluster, but it is still distinct from #5293.", + "best_issue_reason": null, + "best_pr_reason": "PR #5389 is the better standalone representative because it covers a larger, self-contained documentation update with a clearer class target; however, it is not a duplicate of #5293.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5293", + "right": "pull_request:5389", + "accept": false, + "reason": "Different targets and different changes: #5293 documents AutoencoderTinyBlock, while #5389 documents UNetMidBlock2D. They share a file but not the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 198, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5297-3", + "nodes": [ + "pull_request:5297", + "pull_request:6780", + "pull_request:8603" + ], + "soft_pairs": [ + "pull_request:5297|pull_request:6780", + "pull_request:6780|pull_request:8603" + ], + "prepared_review_unit_hash": "7c815a68c4a131e94809a21fe4609e11192a38584696ae7343ffd13963bdd1bd", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7016, + "estimated_input_tokens": 1754, + "estimated_eval_tokens": 3764 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7016, + "estimated_input_tokens": 1754, + "estimated_eval_tokens": 3764 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs are related only at a high level as dependency pinning in Docker/CI images, but they target different packages and different underlying compatibility problems: torch version pinning in one image, torch upper-bound pinning across test runners, and a temporary numpy pin across CI images.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #6780 is the broadest and most central dependency-fix PR in this set: it updates multiple runner Dockerfiles plus setup/dependency metadata to constrain torch versions consistently.", + "best_issue_reason": null, + "best_pr_reason": "PR #6780 is the best representative because it spans the shared torch compatibility issue across several images and repository metadata, making it the most cluster-central change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5297", + "right": "pull_request:6780", + "accept": false, + "reason": "Both pin torch, but #5297 is a narrow pin to specific versions in one CUDA image, while #6780 applies a repo-wide torch <2.2.0 constraint across several test-runner images and dependency files. Different scope and likely different compatibility trigger." + }, + { + "left": "pull_request:6780", + "right": "pull_request:8603", + "accept": false, + "reason": "These address different packages and failure modes: torch pinning versus numpy pinning. They overlap only in being CI/container dependency fixes, not in the same underlying bug or mergeable code-path change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 199, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5322-2", + "nodes": [ + "pull_request:5322", + "pull_request:5590" + ], + "soft_pairs": [ + "pull_request:5322|pull_request:5590" + ], + "prepared_review_unit_hash": "99854b0c76ff215b8730f87c9ed602cae1e3b194c1eb03e4fd10265954be13ae", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4186, + "estimated_input_tokens": 1047, + "estimated_eval_tokens": 2350 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4186, + "estimated_input_tokens": 1047, + "estimated_eval_tokens": 2350 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate documentation PRs in the same inpaint guide: one adds a mask-creation helper section, the other adds a comparison showing non-inpaint checkpoints. They touch the same file but address different guide improvements, so they are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5590 is the better canonical PR for the cluster because it is the larger, more self-contained docs expansion and reads as the stronger representative of the inpaint-guide updates.", + "best_issue_reason": null, + "best_pr_reason": "PR #5590 is the best PR candidate overall because it makes the more substantial and standalone change to the inpaint documentation, while #5322 is a smaller auxiliary addition.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5322", + "right": "pull_request:5590", + "accept": false, + "reason": "They modify the same documentation page but add different content for different purposes: one explains creating a mask image, the other compares regular vs inpaint checkpoints. Same file/subsystem is not enough to treat them as the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 200, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5340-2", + "nodes": [ + "pull_request:5340", + "pull_request:5355" + ], + "soft_pairs": [ + "pull_request:5340|pull_request:5355" + ], + "prepared_review_unit_hash": "291803bd61417968e28287b44f073efe481d2d7d1d2579f1edf75b57556fb7fc", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4324, + "estimated_input_tokens": 1081, + "estimated_eval_tokens": 2418 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4324, + "estimated_input_tokens": 1081, + "estimated_eval_tokens": 2418 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two pull requests in the same general multi-GPU training area, but they fix different concrete bugs in different example scripts. They should not be clustered as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5355 is the better canonical PR: it has the clearer bug description, broader scope across two SDXL LoRA example scripts, and directly addresses the concrete multi-GPU DDP error via `find_unused_parameters=True`.", + "best_issue_reason": null, + "best_pr_reason": "PR #5355 is the strongest representative of the cluster because it fixes a specific runtime failure in a well-defined code path and has the clearer, more actionable title and diff.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5340", + "right": "pull_request:5355", + "accept": false, + "reason": "Both mention multi-GPU, but they fix different bugs: #5340 changes checkpointing behavior in `train_unconditional.py` around `accelerator.is_main_process`, while #5355 changes DDP kwargs in SDXL LoRA training to address `find_unused_parameters`. Different code paths and not mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 201, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5347-2", + "nodes": [ + "pull_request:5347", + "pull_request:5349" + ], + "soft_pairs": [ + "pull_request:5347|pull_request:5349" + ], + "prepared_review_unit_hash": "a4510e958961524200475710839fac2adf7a001eb653d8ce565fd54846f3e203", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5211, + "estimated_input_tokens": 1303, + "estimated_eval_tokens": 2862 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5211, + "estimated_input_tokens": 1303, + "estimated_eval_tokens": 2862 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests appear to be near-identical implementations of the same xformers test-runner move; 5349 is the merged version and is the best canonical PR.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5349 is merged and covers the same xformers runner/workflow change, making it the best canonical PR for this cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR #5349 is the strongest representative because it is merged, reviewed, and implements the same concrete workflow/docker/test-runner change as the other PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5347", + "right": "pull_request:5349", + "accept": true, + "reason": "Both PRs change the same workflow/docker/test files to add a dedicated xformers CUDA test runner, with highly overlapping file coverage and matching diff structure; they look like the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 202, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5426-2", + "nodes": [ + "pull_request:5426", + "pull_request:5597" + ], + "soft_pairs": [ + "pull_request:5426|pull_request:5597" + ], + "prepared_review_unit_hash": "7e573c30149e5007583d443526feac9be22ff0c73f9215f445bfa93b687e9328", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3943, + "estimated_input_tokens": 986, + "estimated_eval_tokens": 2228 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3943, + "estimated_input_tokens": 986, + "estimated_eval_tokens": 2228 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are both Docker image maintenance changes, but they target different images and different concrete dependency sets: one updates the slow CUDA test images, the other updates the xformers CUDA image and unpins torch/torchvision. They do not look like the same underlying change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5426 is the better representative of the cluster because it makes the broader base-image update across two CUDA test Dockerfiles, so it is the more general maintenance change.", + "best_issue_reason": null, + "best_pr_reason": "PR #5426 is the strongest standalone representative: it has the larger, clearer image-base upgrade and affects multiple Dockerfiles, making it more useful as the cluster\u2019s canonical PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5426", + "right": "pull_request:5597", + "accept": false, + "reason": "Different Docker images and different dependency updates; they share only broad CUDA/docker maintenance context, not the same concrete bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 203, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5464-4", + "nodes": [ + "pull_request:5464", + "pull_request:5582", + "pull_request:5810", + "pull_request:7340" + ], + "soft_pairs": [ + "pull_request:5464|pull_request:5582", + "pull_request:5464|pull_request:5810", + "pull_request:5464|pull_request:7340", + "pull_request:5582|pull_request:5810", + "pull_request:5582|pull_request:7340", + "pull_request:5810|pull_request:7340" + ], + "prepared_review_unit_hash": "7e62c4f245e5ed48aac05adc1044f4b9edc6309079ed8faecbe86a630789f150", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 10144, + "estimated_input_tokens": 2536, + "estimated_eval_tokens": 5328 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 10144, + "estimated_input_tokens": 2536, + "estimated_eval_tokens": 5328 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are all unrelated maintenance changes: one scheduler test fix, one Alt Diffusion whitespace cleanup, and two separate documentation/template wording tweaks. The soft-similarity signal is misleading here; they should not be deduplicated together.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue-linked PRs here; among the PRs, #7340 is the most substantive and concrete because it fixes scheduler test behavior for LMSDiscreteScheduler, while the others are formatting or wording-only edits.", + "best_issue_reason": null, + "best_pr_reason": "#7340 is the best representative PR because it targets a specific code-path behavior in scheduler tests. The others are unrelated doc/template cleanup or trivial whitespace changes, so they are not good canonical choices for the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5464", + "right": "pull_request:5582", + "accept": false, + "reason": "Unrelated changes: Alt Diffusion pipeline whitespace vs. issue/PR template and repository text edits." + }, + { + "left": "pull_request:5464", + "right": "pull_request:5810", + "accept": false, + "reason": "Unrelated changes: Alt Diffusion pipeline whitespace vs. README wording/capitalization updates." + }, + { + "left": "pull_request:5464", + "right": "pull_request:7340", + "accept": false, + "reason": "Different domains and intent: Alt Diffusion formatting cleanup vs. scheduler test logic for LMSDiscreteScheduler." + }, + { + "left": "pull_request:5582", + "right": "pull_request:5810", + "accept": false, + "reason": "Both are docs/metadata edits, but they change different files for different purposes and are not the same underlying change." + }, + { + "left": "pull_request:5582", + "right": "pull_request:7340", + "accept": false, + "reason": "Template/documentation cleanup is unrelated to a scheduler test behavior fix." + }, + { + "left": "pull_request:5810", + "right": "pull_request:7340", + "accept": false, + "reason": "README wording edits are unrelated to a scheduler test code-path change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 204, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5486-3", + "nodes": [ + "pull_request:5486", + "pull_request:5717", + "pull_request:7065" + ], + "soft_pairs": [ + "pull_request:5486|pull_request:5717", + "pull_request:5486|pull_request:7065" + ], + "prepared_review_unit_hash": "2f6e7f796e01523e0c844d44cdeef97e472266bd1b468231db1b6c11ad79b062", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6053, + "estimated_input_tokens": 1514, + "estimated_eval_tokens": 3284 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6053, + "estimated_input_tokens": 1514, + "estimated_eval_tokens": 3284 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three pull requests are about different fixes: a docs typo, a ControlNet example signature fix, and a TensorRT guidance_scale bug fix. They are not duplicates and should not be clustered as the same change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 7065 is the strongest representative because it fixes a concrete runtime bug across two example pipelines, whereas the others are either documentation-only or a separate ControlNet example compatibility fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 7065 is the best overall representative of the cluster because it is the most substantive code change and the most clearly user-facing bug fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5486", + "right": "pull_request:5717", + "accept": false, + "reason": "Different kinds of changes: 5486 only corrects ControlNet documentation text, while 5717 changes the forward signature in a community example to accept extra args/kwargs. They do not fix the same bug or code path." + }, + { + "left": "pull_request:5486", + "right": "pull_request:7065", + "accept": false, + "reason": "5486 is a docs-only typo fix; 7065 fixes TensorRT pipeline state handling for guidance_scale in two separate example scripts. No shared underlying bug or mergeable single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 205, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5514-2", + "nodes": [ + "pull_request:5514", + "pull_request:5659" + ], + "soft_pairs": [ + "pull_request:5514|pull_request:5659" + ], + "prepared_review_unit_hash": "6d22d8cddcbe4c31316eba3baa3e4b9a5d938b5389ea386eef30d6be2398410c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3925, + "estimated_input_tokens": 982, + "estimated_eval_tokens": 2220 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3925, + "estimated_input_tokens": 982, + "estimated_eval_tokens": 2220 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests are not duplicates: one changes the community LCM example\u2019s scheduler handling, while the other fixes custom pipeline loading to respect repo revisions. They touch different code paths and solve different problems.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5659 is the better canonical PR for the cluster because it is merged, addresses a concrete framework-level bug in custom pipeline loading, and has a clearer, more general impact than the example-only LCM backward-compatibility tweak.", + "best_issue_reason": null, + "best_pr_reason": "PR #5659 is the strongest PR candidate overall: it is merged, fixes a specific bug in `pipeline_utils.py`, and is more broadly applicable than PR #5514, which only affects one community example.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5514", + "right": "pull_request:5659", + "accept": false, + "reason": "Different underlying changes: #5514 modifies the LCM community example\u2019s scheduler initialization, while #5659 fixes repository revision handling for custom pipelines in `pipeline_utils.py`. They do not target the same bug or code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 206, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5521-7", + "nodes": [ + "pull_request:5521", + "pull_request:5583", + "pull_request:5584", + "pull_request:5585", + "pull_request:5586", + "pull_request:5587", + "pull_request:5588" + ], + "soft_pairs": [ + "pull_request:5521|pull_request:5583", + "pull_request:5521|pull_request:5584", + "pull_request:5521|pull_request:5585", + "pull_request:5521|pull_request:5586", + "pull_request:5521|pull_request:5587", + "pull_request:5583|pull_request:5584", + "pull_request:5583|pull_request:5585", + "pull_request:5583|pull_request:5586", + "pull_request:5583|pull_request:5587", + "pull_request:5584|pull_request:5585", + "pull_request:5584|pull_request:5586", + "pull_request:5584|pull_request:5587", + "pull_request:5584|pull_request:5588", + "pull_request:5585|pull_request:5586", + "pull_request:5585|pull_request:5587", + "pull_request:5585|pull_request:5588", + "pull_request:5586|pull_request:5587", + "pull_request:5586|pull_request:5588", + "pull_request:5587|pull_request:5588", + "pull_request:5521|pull_request:5588" + ], + "prepared_review_unit_hash": "81bd6d1212638523053999727e10c5a52aaaa10987034dbda1ade47e7dfe5447", + "component_budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 20, + "serialized_chars": 25163, + "estimated_input_tokens": 6291, + "estimated_eval_tokens": 12838 + }, + "budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 20, + "serialized_chars": 25163, + "estimated_input_tokens": 6291, + "estimated_eval_tokens": 12838 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are closely related but distinct documentation-cleanup PRs, not duplicates of the same change. They all fix typos or wording in different doc sections, so the high similarity is from shared editorial intent rather than one underlying bug or patch.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 5588 is the broadest merged docs cleanup in the set, touching main-page and repository-level files, so it is the best single representative of this cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 5588 is the strongest cluster anchor: merged, broad in scope, and representative of the general docs typo/update theme without being tied to one narrow subsection.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5521", + "right": "pull_request:5583", + "accept": false, + "reason": "Both are docs typo sweeps, but they cover different file sets and sections; not the same concrete change." + }, + { + "left": "pull_request:5521", + "right": "pull_request:5584", + "accept": false, + "reason": "One targets templates/main docs, the other targets loading and hub docs; separate cleanup batches." + }, + { + "left": "pull_request:5521", + "right": "pull_request:5585", + "accept": false, + "reason": "Different doc areas (templates/main files vs conceptual guides); similarity is only topical." + }, + { + "left": "pull_request:5521", + "right": "pull_request:5586", + "accept": false, + "reason": "Tutorial-page cleanup versus template/main-file cleanup; distinct PRs with different patch scopes." + }, + { + "left": "pull_request:5521", + "right": "pull_request:5587", + "accept": false, + "reason": "Getting-started docs and docs README changes are unrelated to the template/main-file typo sweep." + }, + { + "left": "pull_request:5521", + "right": "pull_request:5588", + "accept": false, + "reason": "Both are broad docs edits, but they touch different repository areas and are not one unified code/docs change." + }, + { + "left": "pull_request:5583", + "right": "pull_request:5584", + "accept": false, + "reason": "Separate documentation sections: pipeline/examples pages versus loading/hub pages." + }, + { + "left": "pull_request:5583", + "right": "pull_request:5585", + "accept": false, + "reason": "API/example doc typos and conceptual-guide typos are different cleanup sets." + }, + { + "left": "pull_request:5583", + "right": "pull_request:5586", + "accept": false, + "reason": "Different targets: pipeline/example docs versus tutorials docs." + }, + { + "left": "pull_request:5583", + "right": "pull_request:5587", + "accept": false, + "reason": "Pipeline/example docs and getting-started docs are distinct patches." + }, + { + "left": "pull_request:5584", + "right": "pull_request:5585", + "accept": false, + "reason": "Loading/hub documentation changes are not the same change as conceptual-guide edits." + }, + { + "left": "pull_request:5584", + "right": "pull_request:5586", + "accept": false, + "reason": "Loading/hub page fixes and tutorial fixes address different docs and wording." + }, + { + "left": "pull_request:5584", + "right": "pull_request:5587", + "accept": false, + "reason": "These are separate docs batches for loading/hub versus getting-started pages." + }, + { + "left": "pull_request:5584", + "right": "pull_request:5588", + "accept": false, + "reason": "One is focused on loading/hub docs, the other on main-page and .github repository files." + }, + { + "left": "pull_request:5585", + "right": "pull_request:5586", + "accept": false, + "reason": "Conceptual-guide edits and tutorial edits are different documentation changes." + }, + { + "left": "pull_request:5585", + "right": "pull_request:5587", + "accept": false, + "reason": "Conceptual docs versus getting-started docs; not the same underlying patch." + }, + { + "left": "pull_request:5585", + "right": "pull_request:5588", + "accept": false, + "reason": "Conceptual guide cleanup is unrelated to the main-page/.github cleanup batch." + }, + { + "left": "pull_request:5586", + "right": "pull_request:5587", + "accept": false, + "reason": "Tutorial-page typo fixes and getting-started-page fixes are separate documentation updates." + }, + { + "left": "pull_request:5586", + "right": "pull_request:5588", + "accept": false, + "reason": "Tutorial docs changes and main-page/.github changes are different cleanup scopes." + }, + { + "left": "pull_request:5587", + "right": "pull_request:5588", + "accept": false, + "reason": "Getting-started docs and repository-level .github/main-page edits are distinct PRs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 207, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5589-2", + "nodes": [ + "pull_request:5589", + "pull_request:5680" + ], + "soft_pairs": [ + "pull_request:5589|pull_request:5680" + ], + "prepared_review_unit_hash": "044cd7309a5e0437983935dfb5baea66f33708380b1dc4ff62be32650f7e70bf", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6226, + "estimated_input_tokens": 1557, + "estimated_eval_tokens": 3370 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6226, + "estimated_input_tokens": 1557, + "estimated_eval_tokens": 3370 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address final CPU/model hook offloading in diffusion pipelines, but they are separate scoped updates rather than the same mergeable change. PR 5589 is a narrower pipeline set; PR 5680 is a later, broader free-model-hook cleanup across additional pipelines.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 5680 is the better canonical representative because it generalizes the offload-hook cleanup to a wider set of pipelines and uses the newer maybe_free_model_hooks pattern consistently.", + "best_issue_reason": null, + "best_pr_reason": "PR 5680 is the strongest standalone PR: broader coverage, clearer hook cleanup, and it consolidates the offloading behavior more completely than 5589.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5589", + "right": "pull_request:5680", + "accept": false, + "reason": "Related theme, but not the same concrete change. 5589 updates final offloading for a subset of pipelines, while 5680 removes specific offload calls and switches to maybe_free_model_hooks across a different, broader pipeline set." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 208, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5696-2", + "nodes": [ + "pull_request:5696", + "pull_request:5704" + ], + "soft_pairs": [ + "pull_request:5696|pull_request:5704" + ], + "prepared_review_unit_hash": "32628422260c7d3b363e5564297ff4ab50d83120c67d2f7153e58c54b4e00e0b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4519, + "estimated_input_tokens": 1130, + "estimated_eval_tokens": 2516 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4519, + "estimated_input_tokens": 1130, + "estimated_eval_tokens": 2516 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are related to Wuerstchen + PEFT support, but they are not the same concrete patch: #5696 changes generic attention/offload logic plus one WuerstchenPrior file, while #5704 updates Wuerstchen layer classes and adds tests. #5704 is the stronger canonical PR for the cluster.", + "confidence": 0.74, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5704 is merged, directly targets the reported Wuerstchen/PEFT breakage, and includes the broader model-layer fix plus tests; it best represents the completed change set.", + "best_issue_reason": null, + "best_pr_reason": "PR #5704 is the best representative because it is merged and covers the full Wuerstchen PEFT compatibility fix in the affected code paths, with test coverage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5696", + "right": "pull_request:5704", + "accept": false, + "reason": "They share the same issue target and subsystem, but the diffs are not the same change: #5696 also edits generic attention_processor LoRA-offload logic, while #5704 changes Wuerstchen model layers and tests. Related, but not duplicate enough to merge as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 209, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5739-2", + "nodes": [ + "pull_request:5739", + "pull_request:5752" + ], + "soft_pairs": [ + "pull_request:5739|pull_request:5752" + ], + "prepared_review_unit_hash": "8efd8411bc5cf4639fa1537008a2ce211fb45459cd04dd3dc299715b3ce1daba", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4515, + "estimated_input_tokens": 1129, + "estimated_eval_tokens": 2514 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4515, + "estimated_input_tokens": 1129, + "estimated_eval_tokens": 2514 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs touch the PixArt-Alpha pipeline and tests, but they address different problems: PR #5739 adds resolution binning, while PR #5752 fixes multi-image generation and prompt mask handling. They are related by subsystem, not the same underlying change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5752 is the more specific bug fix and directly addresses a concrete runtime failure in multi-image generation.", + "best_issue_reason": null, + "best_pr_reason": "PR #5752 is the best representative of the cluster because it targets a user-visible correctness bug rather than a feature addition.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5739", + "right": "pull_request:5752", + "accept": false, + "reason": "Same pipeline/files, but different code-paths and goals: resolution binning vs. multi-image prompt/mask handling. They could not plausibly be merged as one duplicate fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 210, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-583-2", + "nodes": [ + "pull_request:583", + "pull_request:600" + ], + "soft_pairs": [ + "pull_request:583|pull_request:600" + ], + "prepared_review_unit_hash": "0b0bb28c01125b4861af62e6842b6240752f7d31cea82ee80cf6548414f89c05", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4422, + "estimated_input_tokens": 1106, + "estimated_eval_tokens": 2468 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4422, + "estimated_input_tokens": 1106, + "estimated_eval_tokens": 2468 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged PRs touch the same Flax Stable Diffusion pipeline file but address different problems: one adds PNDM scheduler support, the other adds dtype passthrough during Flax model init. They are not duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #583 is the more central cluster member: it makes the Flax Stable Diffusion pipeline work with PNDM by wiring in scheduler state and loop changes, which is the larger concrete code-path change.", + "best_issue_reason": null, + "best_pr_reason": "PR #583 is the best representative of the cluster because it implements a substantive pipeline capability rather than a small config convenience fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:583", + "right": "pull_request:600", + "accept": false, + "reason": "Different changes: #583 adds PNDM scheduler support in the Flax pipeline, while #600 only propagates dtype from config into initialization. Shared filename alone is not enough to merge them as the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 211, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-587-2", + "nodes": [ + "pull_request:587", + "pull_request:664" + ], + "soft_pairs": [ + "pull_request:587|pull_request:664" + ], + "prepared_review_unit_hash": "906b8a6bd22ca7d9a16e29fb1382a9291f4fa173c21bac554341dc1335b17e03", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4472, + "estimated_input_tokens": 1118, + "estimated_eval_tokens": 2492 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4472, + "estimated_input_tokens": 1118, + "estimated_eval_tokens": 2492 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are related only by the LMS/stable-diffusion area, but they address different problems: #587 adds K-LMS support to img2img/inpainting, while #664 fixes a separate LMS scheduler PyTorch regression. They should not be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No true canonical PR: #587 is a feature/support change for inpainting, while #664 is a separate scheduler regression fix.", + "best_issue_reason": null, + "best_pr_reason": "#664 is the cleaner standalone bug-fix PR if one item must represent the cluster, but it is not a duplicate of #587.", + "soft_edge_verdicts": [ + { + "left": "pull_request:587", + "right": "pull_request:664", + "accept": false, + "reason": "Different underlying changes and code paths: #587 adds K-LMS inpainting/img2img support, whereas #664 fixes LMS scheduler sigma handling and a dependency regression. They do not look like the same concrete bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 212, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5878-3", + "nodes": [ + "pull_request:5878", + "pull_request:5879", + "pull_request:6031" + ], + "soft_pairs": [ + "pull_request:5878|pull_request:5879", + "pull_request:5879|pull_request:6031" + ], + "prepared_review_unit_hash": "c813d1d8e0d1f1dfa1be87550ab16273d40ba9c1e1c1d6217d6ac7d97941b8c0", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10140, + "estimated_input_tokens": 2535, + "estimated_eval_tokens": 5326 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10140, + "estimated_input_tokens": 2535, + "estimated_eval_tokens": 5326 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "PR #5878 and #5879 are near-duplicate fixes for the same VAE scale-factor robustness issue across the same set of pipelines. PR #6031 is unrelated LoRA padding behavior and should not be clustered with them.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5878 is the cleaner canonical representative: it has the clearer title and the same VAE fallback fix applied consistently across the affected pipelines.", + "best_issue_reason": null, + "best_pr_reason": "PR #5878 is the best overall PR in this cluster because it directly addresses the VAE config compatibility bug with a consistent cross-pipeline change; #5879 is essentially the same fix with a noisier title and near-identical implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5878", + "right": "pull_request:5879", + "accept": true, + "reason": "Same underlying bug: make pipeline VAE scale-factor handling work with configs that expose decoder_block_out_channels instead of block_out_channels. The diffs overlap on the same files and the code change is functionally equivalent." + }, + { + "left": "pull_request:5879", + "right": "pull_request:6031", + "accept": false, + "reason": "Different problem and code path: #5879 fixes VAE config compatibility in diffusion pipelines, while #6031 changes LoRACompatibleConv padding_mode behavior in lora.py." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 213, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5935-2", + "nodes": [ + "pull_request:5935", + "pull_request:6040" + ], + "soft_pairs": [ + "pull_request:5935|pull_request:6040" + ], + "prepared_review_unit_hash": "aba7272e74d2c29a5f5c3e2ff584420c3d26a6c2eb88ae5ff7349bf8f81f3b0d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4424, + "estimated_input_tokens": 1106, + "estimated_eval_tokens": 2468 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4424, + "estimated_input_tokens": 1106, + "estimated_eval_tokens": 2468 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged PRs touch the same training script, but they address different changes: PR #5935 adds LoRA/SDXL training-script fixes including a text-encoder LoRA state-dict helper, while PR #6040 only changes `token_abstraction` parsing/help text. They are related, but not duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #5935 is the broader, more substantive change set in the cluster; it fixes actual training-script behavior, whereas PR #6040 is a narrow CLI-argument tweak.", + "best_issue_reason": null, + "best_pr_reason": "PR #5935 is the best representative for this cluster because it covers the main bug-fix work and has the wider scope of impact in the shared script.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5935", + "right": "pull_request:6040", + "accept": false, + "reason": "Different underlying changes. #5935 adds LoRA/text-encoder helper logic and other training-script fixes; #6040 only changes the `--token_abstraction` argument type/help text. Same file, but not the same bug or mergeable duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 214, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-5970-2", + "nodes": [ + "pull_request:5970", + "pull_request:5971" + ], + "soft_pairs": [ + "pull_request:5970|pull_request:5971" + ], + "prepared_review_unit_hash": "cbf3f798c1febcece245a3234e04234bd26e2100dd497dd3c6f248c6410f94c5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3660, + "estimated_input_tokens": 915, + "estimated_eval_tokens": 2086 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3660, + "estimated_input_tokens": 915, + "estimated_eval_tokens": 2086 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests implement the same fix in the same file and location: moving the DreamBooth SDXL components back to CPU and clearing CUDA memory before final inference. PR 5970 is slightly more complete because it also cleans up LoRA layers when training the text encoder.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 5970 is the better canonical PR because it appears first and includes the broader cleanup, including conditional deletion of text-encoder LoRA layers in addition to the shared GPU-memory release steps.", + "best_issue_reason": null, + "best_pr_reason": "PR 5970 is the strongest representative of the cluster: same code path, same intent, and the most complete version of the fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:5970", + "right": "pull_request:5971", + "accept": true, + "reason": "Same underlying change: both free UNet/text encoder memory before final inference in the same script. The differences are minor cleanup/comment wording, and the 5970 version is a superset with extra LoRA-layer deletion." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the diffs: both PRs add the same GPU-memory cleanup at the same spot in `train_dreambooth_lora_sdxl.py`, and PR 5970 does include the extra LoRA-layer deletion gated on `args.train_text_encoder`. The soft-edge acceptance is conservative enough, and the two changes look mergeable as one bugfix PR." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 215, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6043-3", + "nodes": [ + "pull_request:6043", + "pull_request:7343", + "pull_request:7344" + ], + "soft_pairs": [ + "pull_request:6043|pull_request:7344", + "pull_request:7343|pull_request:7344" + ], + "prepared_review_unit_hash": "a9f94348d72402c8b1e6e4ca05a606128ce985a4719bca68684762a30b1d0c2a", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6414, + "estimated_input_tokens": 1604, + "estimated_eval_tokens": 3464 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6414, + "estimated_input_tokens": 1604, + "estimated_eval_tokens": 3464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "No duplicate cluster here: the three pull requests address different changes. PR #7343 is a broad code/doc/test refactor around config attribute access, #7344 is a small test-only config-loading adjustment, and #6043 fixes a deprecation warning in single_file loading. None should be merged as the same underlying change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR #7343 is the best representative because it is the most substantial code change and touches the underlying config-attribute access pattern across multiple real code paths, unlike the smaller test-only or warning-fix PRs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6043", + "right": "pull_request:7344", + "accept": false, + "reason": "Different code paths and purposes: #6043 changes a dtype call in single_file loading, while #7344 only rewrites a config-loading test. They do not fix the same bug or merge cleanly into one PR." + }, + { + "left": "pull_request:7343", + "right": "pull_request:7344", + "accept": false, + "reason": "Related only in a broad config area, but #7343 updates production and docs code for accessing config attributes, whereas #7344 changes one test's loading pattern. They are not the same concrete change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 216, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6138-2", + "nodes": [ + "pull_request:6138", + "pull_request:6959" + ], + "soft_pairs": [ + "pull_request:6138|pull_request:6959" + ], + "prepared_review_unit_hash": "b7c0afbb8dd01bcf3d0ea925ae23f881ce586895d2b4465fa9be544999b225ae", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4286, + "estimated_input_tokens": 1072, + "estimated_eval_tokens": 2400 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4286, + "estimated_input_tokens": 1072, + "estimated_eval_tokens": 2400 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are unrelated pull requests: one fixes SDXL `from_single_file` behavior for inpaint/model components, the other adds `config_file` support for ControlNet `from_single_file`. They touch the same broad loader area but not the same bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #6959 is the better canonical PR for this cluster because it is the merged, self-contained change, while #6138 addresses a different SDXL-specific loader bug.", + "best_issue_reason": null, + "best_pr_reason": "PR #6959 is the best representative artifact here: it is merged, narrowly scoped, and clearly describes a single loader enhancement. PR #6138 is a separate SDXL fix and should not be conflated with it.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6138", + "right": "pull_request:6959", + "accept": false, + "reason": "Different changes and different code paths: #6138 fixes SDXL single-file loading for inpaint-related components, while #6959 adds `config_file` passthrough for ControlNet `from_single_file`. They are not the same underlying bug or mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 217, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6286-2", + "nodes": [ + "pull_request:6286", + "pull_request:6334" + ], + "soft_pairs": [ + "pull_request:6286|pull_request:6334" + ], + "prepared_review_unit_hash": "f2515184f881517c6986ea322f5fc059bd79e06c0434d390cb84d20b9fe5ab5c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4418, + "estimated_input_tokens": 1105, + "estimated_eval_tokens": 2466 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4418, + "estimated_input_tokens": 1105, + "estimated_eval_tokens": 2466 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs touch `pipeline_controlnet.py`, but they address different problems: one fixes a typo in a nested-image `ValueError`, while the other changes multi-ControlNet prompt/image handling and tests. They should not be deduplicated.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #6334 is the broader, behavior-changing fix for multi-image/multi-ControlNet input handling and includes tests; PR #6286 is only a message typo correction.", + "best_issue_reason": null, + "best_pr_reason": "PR #6334 is the best representative of the underlying bug/change because it fixes the actual ControlNet input-handling logic rather than just editing error text.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6286", + "right": "pull_request:6334", + "accept": false, + "reason": "Not the same underlying change: #6286 only adjusts an error string, while #6334 changes input validation behavior for multi-ControlNet prompts/images and adds tests." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 218, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6338-3", + "nodes": [ + "pull_request:6338", + "pull_request:6629", + "pull_request:6637" + ], + "soft_pairs": [ + "pull_request:6338|pull_request:6629", + "pull_request:6338|pull_request:6637", + "pull_request:6629|pull_request:6637" + ], + "prepared_review_unit_hash": "74adefec81054e60296200895d4e0d81486d6931e5453616387699328ae8db2b", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 8933, + "estimated_input_tokens": 2234, + "estimated_eval_tokens": 4724 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 8933, + "estimated_input_tokens": 2234, + "estimated_eval_tokens": 4724 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate Japanese docs translation PRs that all touch the JA toctree, but they target different tutorial pages/content rather than the same concrete change. I would not cluster them as duplicates.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 6338 is the best anchor for the cluster because it is merged and makes the broadest initial JA docs/tutorials update, whereas the other two are distinct follow-on translations.", + "best_issue_reason": null, + "best_pr_reason": "PR 6338 is the most suitable representative overall: merged, broadly scoped, and closest to the common JA docs navigation update theme.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6338", + "right": "pull_request:6629", + "accept": false, + "reason": "Both edit the JA toctree, but 6338 updates the tutorials overview/index while 6629 adds an AutoPipeline tutorial page; these are separate documentation changes, not the same underlying PR." + }, + { + "left": "pull_request:6338", + "right": "pull_request:6637", + "accept": false, + "reason": "6338 is a tutorials/index translation update, while 6637 translates loading/Hub docs; they share the same docs area but not the same concrete change." + }, + { + "left": "pull_request:6629", + "right": "pull_request:6637", + "accept": false, + "reason": "These PRs add different JA tutorial documentation topics (AutoPipeline vs loading docs); shared toctree edits are too broad to treat as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 219, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6390-4", + "nodes": [ + "pull_request:6390", + "pull_request:7596", + "pull_request:7639", + "pull_request:9365" + ], + "soft_pairs": [ + "pull_request:6390|pull_request:7639", + "pull_request:7596|pull_request:7639", + "pull_request:7639|pull_request:9365" + ], + "prepared_review_unit_hash": "3dbf81a5c12a3a7a8025d0753affa71a03a395b73c9ca3ede503014c0bad362d", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 7558, + "estimated_input_tokens": 1890, + "estimated_eval_tokens": 4036 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 7558, + "estimated_input_tokens": 1890, + "estimated_eval_tokens": 4036 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs all live under the optimization/docs area, but they are for different underlying features: DeepCache, TGATE, xDiT, and a separate TGATE enablement PR. They share section-level similarity only, not the same concrete change or bug.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 7639 is the best central representative of the cluster because it introduces the TGATE optimization page and sits between the other optimization-doc additions, but it is still a distinct change from DeepCache, xDiT, and the TGATE enablement PR.", + "best_issue_reason": null, + "best_pr_reason": "PR 7639 is the strongest standalone representative of the optimization-docs subgroup: it is a complete, merged docs addition with clear scope and the most explicit topical overlap with the others, even though it is not a duplicate of them.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6390", + "right": "pull_request:7639", + "accept": false, + "reason": "DeepCache docs vs TGATE docs: same parent section, but different optimization methods and different pages. Shared _toctree placement is not enough to indicate the same change." + }, + { + "left": "pull_request:7596", + "right": "pull_request:7639", + "accept": false, + "reason": "Both mention TGATE, but 7596 is a feature/enablement PR for PixArt-Alpha while 7639 is a documentation PR. Different change type and no evidence they fix the same code path." + }, + { + "left": "pull_request:7639", + "right": "pull_request:9365", + "accept": false, + "reason": "TGATE and xDiT are separate optimization technologies with different docs pages and scope; they only overlap in the optimization docs tree." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 220, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6477-2", + "nodes": [ + "pull_request:6477", + "pull_request:7817" + ], + "soft_pairs": [ + "pull_request:6477|pull_request:7817" + ], + "prepared_review_unit_hash": "af0c403a7a507f95689145252abe48b06d2d1fa5ec5e9d8c400d09d9cdfd91c8", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6352, + "estimated_input_tokens": 1588, + "estimated_eval_tokens": 3432 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6352, + "estimated_input_tokens": 1588, + "estimated_eval_tokens": 3432 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items are scheduler-related PRs, but they address different changes: PR #6477 fixes a specific DPM Solver `use_karras_sigmas`/final-sigma behavior, while PR #7817 adds broader custom `timesteps` and `sigmas` support across many pipelines and docs. They are not duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #6477 is the better canonical PR because it targets a narrow, concrete scheduler bug fix with clear code-path changes in DPM solver implementations, whereas #7817 is a broader feature PR spanning many pipelines and documentation.", + "best_issue_reason": null, + "best_pr_reason": "PR #6477 best represents the specific underlying scheduler fix in this cluster; #7817 is more of a general feature expansion than the same bug/change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6477", + "right": "pull_request:7817", + "accept": false, + "reason": "They overlap on scheduler internals, but #6477 fixes a specific DPM Solver `use_karras_sigmas` final-sigma issue while #7817 adds general custom timestep/sigma support across many pipelines. Different concrete changes, so not the same PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 221, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-648-2", + "nodes": [ + "pull_request:648", + "pull_request:767" + ], + "soft_pairs": [ + "pull_request:648|pull_request:767" + ], + "prepared_review_unit_hash": "499910d43a2a7c4c8a6a83d1275e1f9b807b7fa8733d14ee20800a27f87278c2", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4564, + "estimated_input_tokens": 1141, + "estimated_eval_tokens": 2538 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4564, + "estimated_input_tokens": 1141, + "estimated_eval_tokens": 2538 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests touch the same scheduler `add_noise` path, but they are not the same fix: PR #648 addresses device placement for `timesteps`/`alphas_cumprod`, while PR #767 expands the handling to dtype as well and broader scheduler coverage. They are related, but not duplicates.", + "confidence": 0.9, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #767 is the better canonical PR for this cluster because it is the later, broader `add_noise` fix and explicitly covers the dtype/device handling that subsumes the narrower device-only adjustment in PR #648.", + "best_issue_reason": null, + "best_pr_reason": "PR #767 is the strongest representative of the cluster because it clearly states the `add_noise` dtype/device bug, touches multiple schedulers, and appears to be the more complete implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:648", + "right": "pull_request:767", + "accept": false, + "reason": "Related scheduler fixes, but not the same underlying change: #648 fixes device movement for `add_noise`, while #767 adds dtype handling and broader scheduler updates. They should not be merged as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 222, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6551-6", + "nodes": [ + "pull_request:6551", + "pull_request:8275", + "pull_request:8789", + "pull_request:8948", + "pull_request:8972", + "pull_request:9337" + ], + "soft_pairs": [ + "pull_request:6551|pull_request:8789", + "pull_request:8275|pull_request:8789", + "pull_request:8789|pull_request:8972", + "pull_request:8948|pull_request:8972", + "pull_request:8972|pull_request:9337" + ], + "prepared_review_unit_hash": "d474a4028d9a030fb2c9dad785b443cd511913b60504a909d75436dba92febc1", + "component_budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 5, + "serialized_chars": 15684, + "estimated_input_tokens": 3921, + "estimated_eval_tokens": 8098 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 5, + "serialized_chars": 15684, + "estimated_input_tokens": 3921, + "estimated_eval_tokens": 8098 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are all related to AnimateDiff, but they are distinct feature/fix efforts: long-context generation, PAG integration, FreeNoise, ControlNet-to-core migration, and vid2vid+ControlNet. The overlaps are mostly in shared pipeline/docs files, not the same underlying change.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the cluster contains several separate AnimateDiff changes rather than duplicates of one fix or feature.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:9337 is the most self-contained and mature concrete pipeline addition (AnimateDiff video-to-video + ControlNet), with substantial discussion and a clear feature boundary, so it is the best representative if one must choose one PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6551", + "right": "pull_request:8789", + "accept": false, + "reason": "Different changes: 6551 fixes multicontrolnet support in the community example, while 8789 adds PAG support for AnimateDiff. Same ecosystem, not the same bug or code path." + }, + { + "left": "pull_request:8275", + "right": "pull_request:8789", + "accept": false, + "reason": "Long-context video generation vs. PAG variant are separate AnimateDiff features with different implementation goals." + }, + { + "left": "pull_request:8789", + "right": "pull_request:8972", + "accept": false, + "reason": "PAG support and moving ControlNet to core are unrelated changes; they share package wiring but not the same underlying feature or fix." + }, + { + "left": "pull_request:8948", + "right": "pull_request:8972", + "accept": false, + "reason": "FreeNoise adds temporal/context sampling behavior, while 8972 migrates AnimateDiff ControlNet into core. Shared files do not imply duplicate PRs." + }, + { + "left": "pull_request:8972", + "right": "pull_request:9337", + "accept": false, + "reason": "Both concern AnimateDiff ControlNet, but one is a core migration and the other is a new video-to-video ControlNet pipeline. They could not plausibly be merged as one duplicate change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 223, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6587-9", + "nodes": [ + "pull_request:6587", + "pull_request:6590", + "pull_request:6608", + "pull_request:6672", + "pull_request:6719", + "pull_request:6737", + "pull_request:6740", + "pull_request:6781", + "pull_request:6803" + ], + "soft_pairs": [ + "pull_request:6587|pull_request:6590", + "pull_request:6587|pull_request:6608", + "pull_request:6587|pull_request:6803", + "pull_request:6590|pull_request:6608", + "pull_request:6590|pull_request:6781", + "pull_request:6590|pull_request:6803", + "pull_request:6590|pull_request:6740", + "pull_request:6608|pull_request:6781", + "pull_request:6608|pull_request:6803", + "pull_request:6608|pull_request:6737", + "pull_request:6672|pull_request:6737", + "pull_request:6672|pull_request:6740", + "pull_request:6672|pull_request:6781", + "pull_request:6672|pull_request:6803", + "pull_request:6737|pull_request:6740", + "pull_request:6737|pull_request:6781", + "pull_request:6737|pull_request:6803", + "pull_request:6740|pull_request:6781", + "pull_request:6740|pull_request:6803", + "pull_request:6781|pull_request:6803", + "pull_request:6672|pull_request:6719" + ], + "prepared_review_unit_hash": "6b5d94e14e7088aa40ed5240b5881af74ca01ce6925a77881d946488aa8b8f00", + "component_budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 21, + "serialized_chars": 21891, + "estimated_input_tokens": 5473, + "estimated_eval_tokens": 11202 + }, + "budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 21, + "serialized_chars": 21891, + "estimated_input_tokens": 5473, + "estimated_eval_tokens": 11202 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a set of related Windows path-compatibility PRs across different diffusers modules, not one single duplicate change. Two pairs are close enough to treat as the same underlying fix; the rest are related but distinct code-paths.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #6737 is the strongest canonical representative: it is merged, explicitly targets issue 6585, and makes a focused, concrete Windows path fix in one code path.", + "best_issue_reason": null, + "best_pr_reason": "PR #6737 is the best overall representative of the cluster because it is merged, issue-linked, and has a clear minimal path-normalization change; the other PRs are adjacent but separate Windows-support fixes in different files.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6587", + "right": "pull_request:6590", + "accept": false, + "reason": "Same general Windows-path theme, but different files and different helper code paths; not the same concrete change." + }, + { + "left": "pull_request:6587", + "right": "pull_request:6608", + "accept": false, + "reason": "Different modules and different load/save logic; too broad to count as the same bug fix." + }, + { + "left": "pull_request:6587", + "right": "pull_request:6803", + "accept": true, + "reason": "Both modify src/diffusers/utils/testing_utils.py to fix Windows-incompatible path joining; these are the same underlying helper-file bug and could plausibly be combined." + }, + { + "left": "pull_request:6590", + "right": "pull_request:6608", + "accept": false, + "reason": "Both are Windows path cleanups, but they fix different loaders in different files and are not one concrete code-path problem." + }, + { + "left": "pull_request:6590", + "right": "pull_request:6781", + "accept": false, + "reason": "Different save/load paths in different modules; same theme, but not the same underlying change." + }, + { + "left": "pull_request:6590", + "right": "pull_request:6803", + "accept": false, + "reason": "Different file and different helper functions; not the same bug fix." + }, + { + "left": "pull_request:6590", + "right": "pull_request:6740", + "accept": false, + "reason": "Different files and no shared concrete code path; only a broad Windows-support similarity." + }, + { + "left": "pull_request:6608", + "right": "pull_request:6781", + "accept": false, + "reason": "Both touch path handling, but they affect different model-saving/loading code paths in different modules." + }, + { + "left": "pull_request:6608", + "right": "pull_request:6803", + "accept": false, + "reason": "Different modules and different helper logic; related theme only." + }, + { + "left": "pull_request:6608", + "right": "pull_request:6737", + "accept": false, + "reason": "Both are in onnx_utils, but the diffs are not the same concrete fix and change different logic/conditions." + }, + { + "left": "pull_request:6672", + "right": "pull_request:6737", + "accept": false, + "reason": "They share issue target 6585 and Windows-support intent, but they fix different files and different code paths." + }, + { + "left": "pull_request:6672", + "right": "pull_request:6740", + "accept": false, + "reason": "Different files and no shared concrete bug; only the same broad path-normalization theme." + }, + { + "left": "pull_request:6672", + "right": "pull_request:6781", + "accept": false, + "reason": "Different modules and different save logic; not mergeable as one duplicate fix." + }, + { + "left": "pull_request:6672", + "right": "pull_request:6803", + "accept": false, + "reason": "Different files and unrelated helper functions; too broad to accept." + }, + { + "left": "pull_request:6737", + "right": "pull_request:6740", + "accept": false, + "reason": "One is onnx_utils and the other is pipeline_flax_utils; same issue target is not enough to make them the same fix." + }, + { + "left": "pull_request:6737", + "right": "pull_request:6781", + "accept": false, + "reason": "Both are Windows path fixes, but they apply to different model serialization/loading code paths." + }, + { + "left": "pull_request:6737", + "right": "pull_request:6803", + "accept": false, + "reason": "Different files and different helper routines; not the same underlying bug." + }, + { + "left": "pull_request:6740", + "right": "pull_request:6781", + "accept": false, + "reason": "Different modules and different functionality; only a shared Windows-path motif." + }, + { + "left": "pull_request:6740", + "right": "pull_request:6803", + "accept": false, + "reason": "Different files and code paths; not the same concrete change." + }, + { + "left": "pull_request:6781", + "right": "pull_request:6803", + "accept": false, + "reason": "Different modules and different save/load helpers; same theme only." + }, + { + "left": "pull_request:6672", + "right": "pull_request:6719", + "accept": true, + "reason": "Both patch src/diffusers/loaders/unet.py with the same Windows path-normalization fix for model saving; 6719 is a refinement of the same underlying change and could be merged into one PR." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: it frames the cluster as related Windows path-compatibility fixes rather than one duplicate change. The two accepted soft pairs are reasonable same-file/same-helper or same-function refinements, and the rejected pairs are cautious given the different files/code paths." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 224, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6627-2", + "nodes": [ + "pull_request:6627", + "pull_request:6849" + ], + "soft_pairs": [ + "pull_request:6627|pull_request:6849" + ], + "prepared_review_unit_hash": "913011688bec7951a4fc2fb947c08238e3fe47b093c558a80009db4d408bc8a6", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3233, + "estimated_input_tokens": 809, + "estimated_eval_tokens": 1874 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3233, + "estimated_input_tokens": 809, + "estimated_eval_tokens": 1874 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs patch the same `load_hf_numpy` path-construction bug in `src/diffusers/utils/testing_utils.py`; PR 6627 is the broader, more complete fix, while PR 6849 is a narrower follow-up. They are close enough to treat as the same change cluster.", + "confidence": 0.81, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 6627 is the better canonical PR because it fixes the full path-handling bug, including the broken HTTP/HTTPS check and the Posix-path construction, whereas PR 6849 only swaps out the path construction line.", + "best_issue_reason": null, + "best_pr_reason": "PR 6627 is the strongest representative: it addresses the underlying failure mode more completely and is the most self-contained fix for the testing utility path bug.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6627", + "right": "pull_request:6849", + "accept": true, + "reason": "Same function, same file, same underlying bug in test data URL/path construction; the second PR is a narrower variant of the first and could plausibly be merged into one fix." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The summary stays within the packet: both PRs touch the same function/file and overlapping path-handling logic, and PR 6627 is indeed the broader fix while PR 6849 is narrower. The soft-edge acceptance is conservative enough because the two diffs could plausibly be merged into one PR for the same bugfix." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 225, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6665-7", + "nodes": [ + "pull_request:6665", + "pull_request:8240", + "pull_request:8265", + "pull_request:8396", + "pull_request:8399", + "pull_request:8694", + "pull_request:8747" + ], + "soft_pairs": [ + "pull_request:6665|pull_request:8396", + "pull_request:8240|pull_request:8747", + "pull_request:8240|pull_request:8265", + "pull_request:8265|pull_request:8694", + "pull_request:8265|pull_request:8747", + "pull_request:8396|pull_request:8399", + "pull_request:8396|pull_request:8747" + ], + "prepared_review_unit_hash": "d2aedc61c9ea41e6d7c1671c3524c2c7598db5aef27ee9bbaea2a1036c87c236", + "component_budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 7, + "serialized_chars": 19723, + "estimated_input_tokens": 4931, + "estimated_eval_tokens": 10118 + }, + "budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 7, + "serialized_chars": 19723, + "estimated_input_tokens": 4931, + "estimated_eval_tokens": 10118 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a family of related HunyuanDiT pull requests, but they cover distinct follow-up changes: initial model/pipeline introduction, attention refactors, QKV fusion, memory optimization, ControlNet support, and v1.2 support. They should not be deduplicated into one change set.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8240 is the foundational HunyuanDiT introduction, adding the core transformer, pipeline, exports, and tests that the later PRs build on.", + "best_issue_reason": null, + "best_pr_reason": "PR #8240 is the best representative PR for the cluster because it establishes the base HunyuanDiT feature set; the other PRs are narrower follow-up enhancements or variants rather than the same concrete change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6665", + "right": "pull_request:8396", + "accept": false, + "reason": "Different changes: #6665 introduces the I2VGen-XL image-to-video pipeline, while #8396 only adds QKV fusion support for HunyuanDiT attention. They affect unrelated code paths and would not be merged as one PR." + }, + { + "left": "pull_request:8240", + "right": "pull_request:8747", + "accept": false, + "reason": "#8240 is the initial HunyuanDiT integration; #8747 adds HunyuanDiT-v1.2 support via embedding/model adjustments. Related project area, but not the same underlying change." + }, + { + "left": "pull_request:8240", + "right": "pull_request:8265", + "accept": false, + "reason": "#8265 is a refactor of the Hunyuan DiT attention implementation, not the same concrete feature as the initial HunyuanDiT add in #8240." + }, + { + "left": "pull_request:8265", + "right": "pull_request:8694", + "accept": false, + "reason": "#8265 is an internal attention refactor, while #8694 adds Hunyuan-DiT ControlNet inference. Different feature sets and different code paths." + }, + { + "left": "pull_request:8265", + "right": "pull_request:8747", + "accept": false, + "reason": "#8265 changes attention internals; #8747 adds v1.2 support through embedding/model updates. These are distinct follow-up changes, not duplicates." + }, + { + "left": "pull_request:8396", + "right": "pull_request:8399", + "accept": false, + "reason": "#8396 enables fused QKV projections in attention, whereas #8399 is a VRAM/memory-optimization change for the pipeline. Same subsystem, but different concrete fixes." + }, + { + "left": "pull_request:8396", + "right": "pull_request:8747", + "accept": false, + "reason": "#8396 is an attention performance tweak; #8747 is a model-version support update. They do not address the same bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 226, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6683-2", + "nodes": [ + "pull_request:6683", + "pull_request:8643" + ], + "soft_pairs": [ + "pull_request:6683|pull_request:8643" + ], + "prepared_review_unit_hash": "d649888309cc620f0a52db170ab2636cc93fa552a8951dc9273b39d5b3d83c10", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4331, + "estimated_input_tokens": 1083, + "estimated_eval_tokens": 2422 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4331, + "estimated_input_tokens": 1083, + "estimated_eval_tokens": 2422 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate merged community-pipeline PRs: one adds SDXL IPEX CPU acceleration, the other adds AnimateDiff IPEX CPU acceleration. They share only the community README and the broad IPEX/CPU theme, but they target different pipelines and code paths.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:6683", + "right": "pull_request:8643", + "accept": false, + "reason": "Different concrete changes: SDXL IPEX vs AnimateDiff IPEX. Shared README edits and broad subsystem are not enough to treat these as the same underlying fix or merge them into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 227, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6690-2", + "nodes": [ + "pull_request:6690", + "pull_request:6692" + ], + "soft_pairs": [ + "pull_request:6690|pull_request:6692" + ], + "prepared_review_unit_hash": "97dfb8c2a842acaed78f3dd6c5c610a35efc0efb39b4c0861089b7573712732c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3673, + "estimated_input_tokens": 919, + "estimated_eval_tokens": 2094 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3673, + "estimated_input_tokens": 919, + "estimated_eval_tokens": 2094 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs target the same UNet1D timestep-embedding dimension bug, but 6692 is the more complete fix and subsumes 6690's core change.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 6692 is the better canonical PR because it includes the shared embedding-size fix plus additional related corrections in the same code path, making it the more complete representative change.", + "best_issue_reason": null, + "best_pr_reason": "PR 6692 is the strongest overall representative: it directly fixes the timestep embedding dimension and also corrects downstream shape handling, so it best captures the underlying bug and final intended patch.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6690", + "right": "pull_request:6692", + "accept": true, + "reason": "Same underlying UNet1D timestep-embedding bug; 6692 contains the same embedding_size change as 6690 and extends it with related fixes in the same constructor path, so they are close enough to treat as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 228, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6738-2", + "nodes": [ + "pull_request:6738", + "pull_request:8768" + ], + "soft_pairs": [ + "pull_request:6738|pull_request:8768" + ], + "prepared_review_unit_hash": "09a630ce3565a5ca4efedd978db35cd27a7a41a29f420f7f8a9bd81abd7c3b74", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6135, + "estimated_input_tokens": 1534, + "estimated_eval_tokens": 3324 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6135, + "estimated_input_tokens": 1534, + "estimated_eval_tokens": 3324 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are not duplicates: one is a test directory refactor for VAE/UNet tests, the other is a broad ControlNet module reorganization in core library code and docs.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8768 is the more central code change in this cluster: it introduces the new `models.controlnets` module and updates core imports/exports. PR #6738 is a narrower test-only structure cleanup.", + "best_issue_reason": null, + "best_pr_reason": "PR #8768 is the best representative artifact because it captures the substantive library-level refactor, while #6738 is just a related but separate test layout change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6738", + "right": "pull_request:8768", + "accept": false, + "reason": "Different underlying changes: #6738 moves and rewires test files, while #8768 splits out the ControlNet model code into a new module and updates docs/imports. They do not fix the same code-path problem and would not be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 229, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6844-2", + "nodes": [ + "pull_request:6844", + "pull_request:7005" + ], + "soft_pairs": [ + "pull_request:6844|pull_request:7005" + ], + "prepared_review_unit_hash": "40412e77f5ff1e4bc25e767efd63ec045fcdcb101a4fb44a4151ed93cd4b0298", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5159, + "estimated_input_tokens": 1290, + "estimated_eval_tokens": 2836 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5159, + "estimated_input_tokens": 1290, + "estimated_eval_tokens": 2836 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two closed PRs for MotionCtrl on Stable Video Diffusion overlap in topic and some core files, but 7005 looks like a follow-up/community packaging of the feature rather than the same standalone change as 6844. I would not deduplicate them as the same PR.", + "confidence": 0.71, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 6844 is the earlier, more central implementation of MotionCtrl SVD, with the conversion script and core model/pipeline changes needed for the feature.", + "best_issue_reason": null, + "best_pr_reason": "PR 6844 is the best representative because it is the original feature implementation and the most directly tied to the underlying code-path changes; PR 7005 reads more like a later community/example-focused follow-up.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6844", + "right": "pull_request:7005", + "accept": false, + "reason": "Same MotionCtrl SVD feature area, but 7005 adds examples/research-project packaging and references 6844 as prior work; it is not clearly the same concrete code change or a single mergeable duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 230, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6846-2", + "nodes": [ + "pull_request:6846", + "pull_request:7149" + ], + "soft_pairs": [ + "pull_request:6846|pull_request:7149" + ], + "prepared_review_unit_hash": "fea7b99eaaf7cdfdb8ad6cfae57e47b303798795749e2d8d1a01a19eec3e61e1", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4206, + "estimated_input_tokens": 1052, + "estimated_eval_tokens": 2360 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4206, + "estimated_input_tokens": 1052, + "estimated_eval_tokens": 2360 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are unrelated: one patches a training example script for distributed setup, while the other adds `callback_on_step_end` behavior and a noise-rescaling helper in the LDM3D pipeline. They do not target the same bug or change and should not be clustered together.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #7149 is the stronger canonical PR because it is a merged, well-scoped code change with an explicit issue target and clear pipeline functionality addition.", + "best_issue_reason": null, + "best_pr_reason": "PR #7149 is the best representative of the cluster only because it is the only substantive merged change; PR #6846 appears to be an unrelated training-script edit with no clear issue linkage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6846", + "right": "pull_request:7149", + "accept": false, + "reason": "Different code paths and purposes: one modifies an example training script for accelerator/process-group setup, the other changes the StableDiffusionLDM3D pipeline to add callback support and guidance rescaling. They are not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 231, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-686-2", + "nodes": [ + "pull_request:686", + "pull_request:721" + ], + "soft_pairs": [ + "pull_request:686|pull_request:721" + ], + "prepared_review_unit_hash": "207f82451439868eff24d44a7bad231f0378002b84681763a4927c3593c2cadb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3270, + "estimated_input_tokens": 818, + "estimated_eval_tokens": 1892 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3270, + "estimated_input_tokens": 818, + "estimated_eval_tokens": 1892 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged PRs with nearly identical wording, but they fix the same gradient-clipping pattern in different example training scripts, for different issue targets. They are related, but not the same underlying change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #686 is the better canonical PR: it has higher discussion/inbound activity and appears slightly more central, while still representing the same fix pattern as the other PR.", + "best_issue_reason": null, + "best_pr_reason": "PR #686 is a strong representative for the cluster because it is the earlier, more referenced fix and clearly shows the core pattern being corrected in code.", + "soft_edge_verdicts": [ + { + "left": "pull_request:686", + "right": "pull_request:721", + "accept": false, + "reason": "Both PRs add the same guard around clip_grad_norm_, but they apply it to different example scripts, different model code paths, and different issue targets. This is a shared fix pattern, not the same concrete bug/change that should be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 232, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6873-3", + "nodes": [ + "pull_request:6873", + "pull_request:6980", + "pull_request:6983" + ], + "soft_pairs": [ + "pull_request:6873|pull_request:6980", + "pull_request:6873|pull_request:6983", + "pull_request:6980|pull_request:6983" + ], + "prepared_review_unit_hash": "385d264c4b1e3b2a8b0e95c66728b9107c7dc6fff3079028dacc9f4719b88252", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7325, + "estimated_input_tokens": 1832, + "estimated_eval_tokens": 3920 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7325, + "estimated_input_tokens": 1832, + "estimated_eval_tokens": 3920 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two PRs are the same underlying 2D attention-name fix and differ mainly by scope; the 3D UNet PR is a separate change and should not be clustered with them.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 6983 is the most complete representative of the 2D attention-parameter naming fix: it covers the same ControlNet/UNet2DConditionModel path as PR 6980 and extends it to another affected model file, making it the best canonical PR for this cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR 6983 is the best overall representative because it is the broader, later version of the same concrete bug fix and touches the full set of affected 2D code paths.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6873", + "right": "pull_request:6980", + "accept": false, + "reason": "Different code path and model family: 6873 is about UNet3DConditionModel attention naming, while 6980 fixes the 2D condition/controlnet naming issue." + }, + { + "left": "pull_request:6873", + "right": "pull_request:6983", + "accept": false, + "reason": "6873 targets the 3D UNet naming change, whereas 6983 targets the 2D UNet/ControlNet naming bug; these are related in theme but not the same fix." + }, + { + "left": "pull_request:6980", + "right": "pull_request:6983", + "accept": true, + "reason": "Both PRs fix the same concrete 2D attention-head parameter naming bug in the same core code paths; 6983 is effectively a superset/extension of 6980 and could plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 233, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6902-2", + "nodes": [ + "pull_request:6902", + "pull_request:7981" + ], + "soft_pairs": [ + "pull_request:6902|pull_request:7981" + ], + "prepared_review_unit_hash": "ef85dd2c04368efb2fbffa54f43191ad7c776c5051f2761bf3650d4a34d5f656", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3176, + "estimated_input_tokens": 794, + "estimated_eval_tokens": 1844 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3176, + "estimated_input_tokens": 794, + "estimated_eval_tokens": 1844 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are both small metadata/housekeeping PRs, but they fix different things: one updates the license year across docs, the other changes the package author email in setup.py. They should not be treated as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 7981 is the narrower, more concrete change and is a better representative of this cluster than the broad documentation/license-year sweep in 6902.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6902", + "right": "pull_request:7981", + "accept": false, + "reason": "Different underlying changes: 6902 updates license/copyright text in documentation files, while 7981 edits setup.py author_email. They do not fix the same code-path or merge into one PR cleanly." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 234, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-6933-3", + "nodes": [ + "pull_request:6933", + "pull_request:8183", + "pull_request:9288" + ], + "soft_pairs": [ + "pull_request:6933|pull_request:8183", + "pull_request:8183|pull_request:9288" + ], + "prepared_review_unit_hash": "b9b4024c0dad7049fb11e67a9f114391bce71da34fd40ebee5dffc14a5991e8f", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6611, + "estimated_input_tokens": 1653, + "estimated_eval_tokens": 3562 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6611, + "estimated_input_tokens": 1653, + "estimated_eval_tokens": 3562 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These three PRs are not duplicates: they address different bugs in different code paths (SD conversion, logger typo handling, and FreeNoise/AnimateDiff V3 compatibility).", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No single PR cleanly represents the cluster because the changes are unrelated; if forced, PR #8183 is the most self-contained and explicitly issue-linked, but it is not a good cluster canonical.", + "soft_edge_verdicts": [ + { + "left": "pull_request:6933", + "right": "pull_request:8183", + "accept": false, + "reason": "Shared file alone is not enough; one PR fixes SD conversion reshaping/text encoder conversion, the other fixes a logger.warning typo in checkpoint loading." + }, + { + "left": "pull_request:8183", + "right": "pull_request:9288", + "accept": false, + "reason": "Different bugs and different behavior: logger warning attribute fix versus FreeNoise mid_block handling for AnimateDiff V3." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 235, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7025-2", + "nodes": [ + "pull_request:7025", + "pull_request:7067" + ], + "soft_pairs": [ + "pull_request:7025|pull_request:7067" + ], + "prepared_review_unit_hash": "107bd2cb3c906786e9ba58023940ab3063c39ce7dd14871a804ae52f6899ce62", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4388, + "estimated_input_tokens": 1097, + "estimated_eval_tokens": 2450 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4388, + "estimated_input_tokens": 1097, + "estimated_eval_tokens": 2450 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two unrelated ControlNet training-script PRs: one fixes validation image logging/output labels, the other changes dataset/train_data_dir argument handling. They overlap in filenames but not in the underlying bug/change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #7067 is the better representative because it is merged and implements a concrete argument-handling fix across training scripts, whereas #7025 is an unmerged logging-only update.", + "best_issue_reason": null, + "best_pr_reason": "PR #7067 is the strongest cluster representative: it is merged, touches the main training path in a substantive way, and is more likely to reflect the final resolved change set.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7025", + "right": "pull_request:7067", + "accept": false, + "reason": "Reject: #7025 changes validation image logging labels/indexing; #7067 changes dataset/train_data_dir validation and dataset loading. Same subsystem, but different code paths and not the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 236, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7071-4", + "nodes": [ + "pull_request:7071", + "pull_request:7211", + "pull_request:8376", + "pull_request:8387" + ], + "soft_pairs": [ + "pull_request:7071|pull_request:8376", + "pull_request:7071|pull_request:8387", + "pull_request:7071|pull_request:7211", + "pull_request:8376|pull_request:8387", + "pull_request:7211|pull_request:8387", + "pull_request:7211|pull_request:8376" + ], + "prepared_review_unit_hash": "fe63bf9d1112b4d0dffea637cacf1e4d1ea4db943e3ba807854485c61f8a61b9", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 10323, + "estimated_input_tokens": 2581, + "estimated_eval_tokens": 5418 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 10323, + "estimated_input_tokens": 2581, + "estimated_eval_tokens": 5418 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two PRs are a near-exact duplicate pair for the same file (`regional_prompting_stable_diffusion.py`), while the other PRs are separate refactors for different example pipelines and should not be merged just because they share issue 6984.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "`pull_request:8376` is the best canonical PR for the duplicate pair: it is the non-draft, actively reviewed version of the same `regional_prompting_stable_diffusion.py` refactor and is essentially identical to 8387.", + "best_issue_reason": null, + "best_pr_reason": "`pull_request:8376` is the strongest representative overall because it is the exact duplicate target of 8387 and has the more complete, non-draft PR state. The other PRs are separate code changes in different example files.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7071", + "right": "pull_request:8376", + "accept": false, + "reason": "Both target issue 6984, but they refactor different example pipelines (`stable_diffusion_reference.py` vs `regional_prompting_stable_diffusion.py`) and are not the same concrete change." + }, + { + "left": "pull_request:7071", + "right": "pull_request:8387", + "accept": false, + "reason": "Same shared issue target only; different file and different example pipeline, so this is not a single mergeable change." + }, + { + "left": "pull_request:7071", + "right": "pull_request:7211", + "accept": false, + "reason": "These are separate refactors for different community examples, not one underlying bug fix or change." + }, + { + "left": "pull_request:8376", + "right": "pull_request:8387", + "accept": true, + "reason": "Near-identical diffs in the same file, same change intent, same issue target; 8387 is effectively a draft/repost of 8376." + }, + { + "left": "pull_request:7211", + "right": "pull_request:8387", + "accept": false, + "reason": "They only share issue 6984; the PRs touch different example pipelines and should not be collapsed together." + }, + { + "left": "pull_request:7211", + "right": "pull_request:8376", + "accept": false, + "reason": "Different files and different pipeline refactors; shared issue target is not enough to treat them as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 237, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7092-3", + "nodes": [ + "pull_request:7092", + "pull_request:8687", + "pull_request:8867" + ], + "soft_pairs": [ + "pull_request:7092|pull_request:8687", + "pull_request:8687|pull_request:8867" + ], + "prepared_review_unit_hash": "77f5a43fb8aca9f45458479a803dae787bd085d7dcba5f3c77b45e717dbe8de3", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6324, + "estimated_input_tokens": 1581, + "estimated_eval_tokens": 3418 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6324, + "estimated_input_tokens": 1581, + "estimated_eval_tokens": 3418 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This is a small cluster of PRs that all tweak GitHub issue/PR templates. The first two are close enough to treat as the same template-maintenance change, but the last one is a separate follow-on edit.", + "confidence": 0.78, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 8687 is the best representative: it makes substantive updates to the same template files and sits between the earlier and later template edits, so it most cleanly captures the shared template-maintenance line.", + "best_issue_reason": null, + "best_pr_reason": "PR 8687 is the strongest single PR to keep: it is broader than 7092 and more central than 8867, while still clearly in the same template-edit sequence.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7092", + "right": "pull_request:8687", + "accept": true, + "reason": "Both PRs edit the same issue/PR template files with the same general goal of updating template contact assignments and prompts, so they look like the same underlying template-maintenance change and could plausibly be merged." + }, + { + "left": "pull_request:8687", + "right": "pull_request:8867", + "accept": false, + "reason": "They share filenames, but the actual edits are different follow-on template tweaks (SD3/improving quality vs. TTV/PEFT contact updates). Shared template files alone are not enough to treat them as the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 238, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7125-2", + "nodes": [ + "pull_request:7125", + "pull_request:8266" + ], + "soft_pairs": [ + "pull_request:7125|pull_request:8266" + ], + "prepared_review_unit_hash": "eea037eeb946d2d67aeaaada108eb0c1d7b4216cd792943df7bcd847654f840d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3841, + "estimated_input_tokens": 961, + "estimated_eval_tokens": 2178 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3841, + "estimated_input_tokens": 961, + "estimated_eval_tokens": 2178 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both items modify the same documentation workflow files, but they address different problems: one limits when docs workflows run, the other changes the container used by those workflows. They are not duplicates and should not be merged.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 8266 is the better representative of the cluster because it makes the more central workflow infrastructure change, updating the docs jobs to use a custom container in both documentation workflows.", + "best_issue_reason": null, + "best_pr_reason": "PR 8266 is the strongest standalone PR here because it directly updates the execution environment for the documentation workflows and is a coherent, self-contained change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7125", + "right": "pull_request:8266", + "accept": false, + "reason": "Different underlying changes: 7125 adds path filters to avoid unnecessary doc workflow runs, while 8266 configures a custom container for docs builds. They touch the same files but fix unrelated workflow behaviors." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 239, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7144-2", + "nodes": [ + "pull_request:7144", + "pull_request:9010" + ], + "soft_pairs": [ + "pull_request:7144|pull_request:9010" + ], + "prepared_review_unit_hash": "723b145efe8df5019642ad30bb558a8a9904cbae6e911712aa213a00d418c942", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4415, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4415, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two pull requests are unrelated: PR #7144 adds NPU support in core diffusers utilities, while PR #9010 updates the SDXL textual inversion example and README to support the second text encoder. They do not fix the same bug or implement the same change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the cluster contains two unrelated changes in different areas of the codebase (NPU support vs. SDXL textual inversion example updates).", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a good global canonical representative because there is no duplicate relationship here; if one must be chosen, #9010 is the more self-contained user-facing example change, but it does not represent the cluster as a whole.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7144", + "right": "pull_request:9010", + "accept": false, + "reason": "Different changes and code paths: #7144 adds NPU support wiring in core utilities, while #9010 modifies textual inversion SDXL example docs/code to support the second text encoder. They are not mergeable as one PR and do not address the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 240, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7162-2", + "nodes": [ + "pull_request:7162", + "pull_request:7878" + ], + "soft_pairs": [ + "pull_request:7162|pull_request:7878" + ], + "prepared_review_unit_hash": "ecdecedba955ea86e4b7607a83d91794aef966173bcaeb24ca26695259e1c6ea", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3117, + "estimated_input_tokens": 780, + "estimated_eval_tokens": 1816 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3117, + "estimated_input_tokens": 780, + "estimated_eval_tokens": 1816 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged PRs touch the same workflow file but make opposite CI concurrency changes: one removes `max-parallel`, the other adds `max-parallel: 8`. They are related historically, but not the same underlying change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:7162", + "right": "pull_request:7878", + "accept": false, + "reason": "Reject: these PRs change the same workflow setting in opposite directions and do not look like the same concrete fix; they would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 241, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7352-2", + "nodes": [ + "pull_request:7352", + "pull_request:7875" + ], + "soft_pairs": [ + "pull_request:7352|pull_request:7875" + ], + "prepared_review_unit_hash": "35a0d0444ee8654c2f70e1bfa68e5f519a0b80a65fbc0f86327019bf0d1b0ae0", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4573, + "estimated_input_tokens": 1144, + "estimated_eval_tokens": 2544 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4573, + "estimated_input_tokens": 1144, + "estimated_eval_tokens": 2544 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related only at a broad LoRA-scaling level, but they address different problems: PR #7352 adds blockwise adapter strength control, while PR #7875 fixes a specific missing-weight fallback for some LoRA modules. Not duplicates.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #7352 is the broader, feature-level change and the more central representative of the cluster; PR #7875 is a narrower bugfix in the same utility code path.", + "best_issue_reason": null, + "best_pr_reason": "PR #7352 best represents the cluster because it introduces the main blockwise-LoRA capability, while #7875 is a separate targeted fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7352", + "right": "pull_request:7875", + "accept": false, + "reason": "They share LoRA scaling code, but #7352 implements blockwise adapter strength control and #7875 fixes a separate fallback for missing module weights; not the same underlying change or bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 242, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7553-3", + "nodes": [ + "pull_request:7553", + "pull_request:8949", + "pull_request:8980" + ], + "soft_pairs": [ + "pull_request:7553|pull_request:8980", + "pull_request:8949|pull_request:8980" + ], + "prepared_review_unit_hash": "49980c84c6c4a07142ed9ac9a6450d277237fd2b677128728bd297f4007fb828", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6451, + "estimated_input_tokens": 1613, + "estimated_eval_tokens": 3482 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 6451, + "estimated_input_tokens": 1613, + "estimated_eval_tokens": 3482 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two PRs are about the same AuraFlow cleanup removing a stray `i` argument; the broader linear/conv refactor PR is unrelated and should not cluster with them.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8980 is the clearest representative of the AuraFlow `i` cleanup: it removes the argument from the transformer forward and the attention call, matching the underlying code-path fix.", + "best_issue_reason": null, + "best_pr_reason": "#8980 is the most complete and specific AuraFlow fix in the cluster, so it best represents the shared change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7553", + "right": "pull_request:8980", + "accept": false, + "reason": "#7553 is a broad refactor removing class assignments for linear/conv across many model files; it does not target the AuraFlow stray-`i` bug or the same code path." + }, + { + "left": "pull_request:8949", + "right": "pull_request:8980", + "accept": true, + "reason": "Both PRs address the same AuraFlow stray-`i` cleanup and could plausibly be merged together as one fix to remove the unused argument from the shared call chain." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 243, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7739-2", + "nodes": [ + "pull_request:7739", + "pull_request:7761" + ], + "soft_pairs": [ + "pull_request:7739|pull_request:7761" + ], + "prepared_review_unit_hash": "2a5dd8ad5786506b22e641af0714cd3cdd0192a0000ec537bde855af3725e05d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5753, + "estimated_input_tokens": 1439, + "estimated_eval_tokens": 3134 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5753, + "estimated_input_tokens": 1439, + "estimated_eval_tokens": 3134 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about callback support in Diffusers, but they are not the same concrete change: #7739 introduces the callback framework/proposal, while #7761 adds official callback implementations and docs across many pipelines. Treat #7761 as the stronger canonical PR; do not dedupe the pair.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #7761 is the merged, broader implementation of the callbacks feature, with official callback classes/docs and multi-pipeline integration.", + "best_issue_reason": null, + "best_pr_reason": "PR #7761 is the best representative because it is merged, more complete, and directly ships the feature rather than only proposing the callback infrastructure.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7739", + "right": "pull_request:7761", + "accept": false, + "reason": "Related feature area, but not the same underlying change: #7739 adds generic callback infrastructure, while #7761 implements official callbacks and documentation across multiple pipelines. They could coexist as sequential work, not a duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 244, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7838-2", + "nodes": [ + "pull_request:7838", + "pull_request:9033" + ], + "soft_pairs": [ + "pull_request:7838|pull_request:9033" + ], + "prepared_review_unit_hash": "864cd52498b77ab2b11423152017541c798150ff2bbc28a0a58ef8bea715b6cd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2773, + "estimated_input_tokens": 694, + "estimated_eval_tokens": 1644 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2773, + "estimated_input_tokens": 694, + "estimated_eval_tokens": 1644 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged PRs that both tweak test-fetcher workflow dependencies, but they target different workflow files and make different install changes; they are related, not duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #9033 is the better representative because it is the more general pipeline test fetcher dependency update and is less tied to the specific venv/uv workflow rewrite in #7838.", + "best_issue_reason": null, + "best_pr_reason": "PR #9033 is the cleaner cluster representative: same subsystem and broader title, but still a distinct change from #7838 rather than the same fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7838", + "right": "pull_request:9033", + "accept": false, + "reason": "Different workflows (.github/workflows/push_tests.yml vs .github/workflows/nightly_tests.yml) and different dependency/install adjustments; they touch the same area but do not look like the same concrete bug fix or mergeable single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 245, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7839-2", + "nodes": [ + "pull_request:7839", + "pull_request:7840" + ], + "soft_pairs": [ + "pull_request:7839|pull_request:7840" + ], + "prepared_review_unit_hash": "19e10d5b33113d80a43e2849ed7bfeca7875f8500407d4c6fa22a3681f245eeb", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4125, + "estimated_input_tokens": 1032, + "estimated_eval_tokens": 2320 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4125, + "estimated_input_tokens": 1032, + "estimated_eval_tokens": 2320 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two merged PRs both update GitHub Actions runner labels, but they target different workflows and hardware classes: one switches PR test fetching to new CPU runners, the other switches nightly/push GPU workflows to new GPU runners. They are related by theme, not the same change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #7840 is the better canonical example for this cluster because it is the broader runner migration across multiple GPU workflows, making it more representative than the smaller CPU-only workflow update in #7839.", + "best_issue_reason": null, + "best_pr_reason": "PR #7840 is the strongest representative PR in the set: it touches multiple workflow files and updates the GPU runner configuration in a cohesive way, whereas #7839 is a narrower CPU runner-only workflow change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7839", + "right": "pull_request:7840", + "accept": false, + "reason": "These are not the same underlying change. #7839 updates CPU runners for pr_test_fetcher, while #7840 updates GPU runners for nightly/push workflows. They affect different code paths and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 246, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7915-2", + "nodes": [ + "pull_request:7915", + "pull_request:7917" + ], + "soft_pairs": [ + "pull_request:7915|pull_request:7917" + ], + "prepared_review_unit_hash": "8f778775ef2426de0f8f11f90f65cb862e6cab1c875a551b29ef75f2a7830842", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2867, + "estimated_input_tokens": 717, + "estimated_eval_tokens": 1690 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2867, + "estimated_input_tokens": 717, + "estimated_eval_tokens": 1690 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs both touch SDXL training, but they address different changes: one is a broad Ascend NPU support effort, the other is a specific deepspeed bug fix in the SDXL example. I would not cluster them as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 7917 is the more concrete, narrowly scoped change: it fixes specific deepspeed-related bugs in the SDXL training script with a small targeted patch. PR 7915 is a broader hardware-support PR with a different goal and is not the same underlying change.", + "best_issue_reason": null, + "best_pr_reason": "PR 7917 is the best representative for this cluster because it has a clear code-path fix and an actual diff; PR 7915 is broader and less clearly aligned to the same bug/change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7915", + "right": "pull_request:7917", + "accept": false, + "reason": "Different underlying changes: 7915 is about Ascend NPU support for SDXL, while 7917 is a targeted deepspeed bug fix in the SDXL training script. Same area, but not the same concrete fix and not something that would naturally merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 247, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-7996-2", + "nodes": [ + "pull_request:7996", + "pull_request:9444" + ], + "soft_pairs": [ + "pull_request:7996|pull_request:9444" + ], + "prepared_review_unit_hash": "37190c7a028d0a8d55998910190dc1a8d0e6067cbabcd1c591f78402c4c80d5b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4391, + "estimated_input_tokens": 1098, + "estimated_eval_tokens": 2452 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4391, + "estimated_input_tokens": 1098, + "estimated_eval_tokens": 2452 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are not duplicates: one removes Stable Cascade UNet tests, the other updates CogVideoX single-file docs and tests. They touch different artifacts and different change intents.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: PR 7996 and PR 9444 address unrelated changes in different parts of the repo.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a strong cluster representative for duplication purposes because they do not share the same underlying bug or change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:7996", + "right": "pull_request:9444", + "accept": false, + "reason": "Different code paths and purposes: 7996 deletes Stable Cascade UNet tests, while 9444 edits CogVideoX docs/tests. They are not the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 248, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8383-2", + "nodes": [ + "pull_request:8383", + "pull_request:8783" + ], + "soft_pairs": [ + "pull_request:8383|pull_request:8783" + ], + "prepared_review_unit_hash": "b72da46cc29a5534d691a9b952ae2a953ef11fc60b6e559498c2bf58908e6d22", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4451, + "estimated_input_tokens": 1113, + "estimated_eval_tokens": 2482 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4451, + "estimated_input_tokens": 1113, + "estimated_eval_tokens": 2482 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs are related to HunyuanDiT, but they are not the same change: one is documentation for the model/pipeline, the other adds checkpoint conversion scripts and a ControlNet model implementation. They should not be deduplicated.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8783 is the stronger representative because it introduces the substantive code-path change (conversion scripts and ControlNet/model updates), while #8383 is docs-only.", + "best_issue_reason": null, + "best_pr_reason": "PR #8783 is the best cluster representative since it affects implementation and model conversion behavior rather than only documentation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8383", + "right": "pull_request:8783", + "accept": false, + "reason": "Different artifact types and different scopes: #8383 adds docs pages/nav for HunyuanDiT, while #8783 adds conversion scripts and ControlNet/model code. Related topic, but not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 249, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8517-2", + "nodes": [ + "pull_request:8517", + "pull_request:8602" + ], + "soft_pairs": [ + "pull_request:8517|pull_request:8602" + ], + "prepared_review_unit_hash": "0dad51ffb708b3ef6aed0d592d42f7962c9b4382675380e007dbe483194f33b7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3871, + "estimated_input_tokens": 968, + "estimated_eval_tokens": 2192 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3871, + "estimated_input_tokens": 968, + "estimated_eval_tokens": 2192 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: PR #8517 adds SD3 single-file support and related code/docs, while PR #8602 is a small documentation wording fix about loading SD3 with T5.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8517 is the substantive change in the cluster, adding the SD3 single-file support code path and associated docs; #8602 is only a narrow docs correction.", + "best_issue_reason": null, + "best_pr_reason": "PR #8517 is the best representative of the cluster because it contains the actual feature implementation rather than a follow-up wording tweak.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8517", + "right": "pull_request:8602", + "accept": false, + "reason": "Different underlying changes: #8517 expands single-file support in code and docs, while #8602 only renames a section title from 'without T5' to 'with T5'. They do not fix the same concrete code-path problem and would not plausibly merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 250, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8558-4", + "nodes": [ + "pull_request:8558", + "pull_request:8610", + "pull_request:9406", + "pull_request:9550" + ], + "soft_pairs": [ + "pull_request:8558|pull_request:8610", + "pull_request:9406|pull_request:9550", + "pull_request:8558|pull_request:9406" + ], + "prepared_review_unit_hash": "f02a0937c4c55f53c989bf5029d9cce61537293918e47ae09d99380d354478ac", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 8886, + "estimated_input_tokens": 2222, + "estimated_eval_tokens": 4700 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 8886, + "estimated_input_tokens": 2222, + "estimated_eval_tokens": 4700 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is heterogeneous: one Stable Diffusion 3 shape fix, one attention-state bug fix, and two Flux ControlNet-related PRs that still address different concrete behaviors. None look like true duplicates of each other.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No single PR cleanly represents the cluster; if forced, PR 9406 is the most concrete bug fix, but the rest are unrelated or only loosely adjacent.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8558", + "right": "pull_request:8610", + "accept": false, + "reason": "Different subsystems and bugs: SD3 prompt-embed shape handling vs attention processor state being wrong across runs." + }, + { + "left": "pull_request:9406", + "right": "pull_request:9550", + "accept": false, + "reason": "Both touch Flux ControlNet, but one fixes `control_mode` batch expansion while the other adds image-dimension validation; different code paths and not mergeable as one duplicate change." + }, + { + "left": "pull_request:8558", + "right": "pull_request:9406", + "accept": false, + "reason": "Unrelated pipelines and failure modes: SD3 `text_encoder_3=None` shape mismatch vs Flux ControlNet `control_mode` batch-size handling." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 251, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8568-2", + "nodes": [ + "pull_request:8568", + "pull_request:9498" + ], + "soft_pairs": [ + "pull_request:8568|pull_request:9498" + ], + "prepared_review_unit_hash": "ccd0cc7441f1bf0179fc7f6197342b0686ccb9570d2cb4b80feccbd545787117", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3307, + "estimated_input_tokens": 827, + "estimated_eval_tokens": 1910 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3307, + "estimated_input_tokens": 827, + "estimated_eval_tokens": 1910 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs fix the same kind of SD3 checkpointing return-value unpacking mistake, but they apply it in different code paths: SD3 transformer blocks vs SD3 ControlNet training. They are related patterns, not the same mergeable change.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #9498 is the better canonical PR because it is merged and directly addresses the concrete ControlNet training path with the checkpointing fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #9498 is the best representative of the cluster since it is merged and clearly documents the SD3 gradient-checkpointing bug in an active training path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8568", + "right": "pull_request:9498", + "accept": false, + "reason": "Reject: both fix a similar checkpointing unpack bug, but in different modules/code paths (transformer_sd3 vs controlnet_sd3). They are related variants, not the same underlying change that should be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 252, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8690-3", + "nodes": [ + "pull_request:8690", + "pull_request:8914", + "pull_request:8938" + ], + "soft_pairs": [ + "pull_request:8690|pull_request:8914", + "pull_request:8690|pull_request:8938", + "pull_request:8914|pull_request:8938" + ], + "prepared_review_unit_hash": "ff4186dc1ca7bfc1360afa38d09f69d6cda9f5b4320e8c80b28405959df9b415", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 12206, + "estimated_input_tokens": 3052, + "estimated_eval_tokens": 6360 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 12206, + "estimated_input_tokens": 3052, + "estimated_eval_tokens": 6360 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three are separate Korean documentation PRs that add or revise anchors and wording on different conceptual pages. They share a similar \u201creflect few contributions\u201d pattern and the same upstream tracking issue reference, but they are not the same underlying change and should not be merged as duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8938 is the best representative because it is the most substantial of the three, has by far the most review activity, and updates the contribution guide rather than a smaller single-page edit.", + "best_issue_reason": null, + "best_pr_reason": "PR #8938 is the strongest standalone PR in the cluster: it has the broadest scope among these translations, the highest review activity, and is the clearest candidate to anchor the group if one must be chosen.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8690", + "right": "pull_request:8914", + "accept": false, + "reason": "Both are Korean doc updates, but they edit different files (`philosophy.md` vs `ethical_guidelines.md`) with different content and anchors; this is similar maintenance, not the same fix." + }, + { + "left": "pull_request:8690", + "right": "pull_request:8938", + "accept": false, + "reason": "These touch different conceptual pages (`philosophy.md` vs `contribution.md`) and make page-specific wording/anchor changes, so they are not the same underlying change." + }, + { + "left": "pull_request:8914", + "right": "pull_request:8938", + "accept": false, + "reason": "Although both are translation/documentation cleanup, they modify different docs and different sections; the shared tracking issue is not enough to treat them as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 253, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8701-2", + "nodes": [ + "pull_request:8701", + "pull_request:8735" + ], + "soft_pairs": [ + "pull_request:8701|pull_request:8735" + ], + "prepared_review_unit_hash": "d8e9e7ae86d9947f58fda5e1451f8947699ccb70334124fbfdb2866ecf6f1a3f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4347, + "estimated_input_tokens": 1087, + "estimated_eval_tokens": 2430 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4347, + "estimated_input_tokens": 1087, + "estimated_eval_tokens": 2430 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are both documentation updates, but they cover different features and user workflows: PR 8701 adds PAG `from_pipe` usage docs, while PR 8735 adds SDXL refiner tips for HunyuanDiT/PixArt and related examples.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR in this cluster; the two PRs do not describe the same change or bug fix.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a strong global representative for the cluster because they are unrelated docs changes. If forced, PR 8735 is slightly broader in scope, but it is not a duplicate of PR 8701.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8701", + "right": "pull_request:8735", + "accept": false, + "reason": "Different documentation topics and code paths: PAG `enable_pag` via `from_pipe` versus SDXL refiner guidance for HunyuanDiT/PixArt. They are not the same underlying change and would not plausibly merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 254, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8762-2", + "nodes": [ + "pull_request:8762", + "pull_request:9943" + ], + "soft_pairs": [ + "pull_request:8762|pull_request:9943" + ], + "prepared_review_unit_hash": "b965b6efafd06d78586e0730e9eb6250a39242e72413f4f503bb0629c8e6b1b5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4676, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4676, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are both LoRA-related, but they target different models and codepaths: #8762 adds an SD3 DreamBooth LoRA Colab/research example, while #9943 implements LoRA support for Mochi-1 in the core library. They are not duplicates and should not be merged together.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:8762", + "right": "pull_request:9943", + "accept": false, + "reason": "Different underlying changes: SD3 Colab training example vs. Mochi-1 LoRA framework support. They touch different models, files, and codepaths, so they are not the same fix/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 255, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8814-2", + "nodes": [ + "pull_request:8814", + "pull_request:8815" + ], + "soft_pairs": [ + "pull_request:8814|pull_request:8815" + ], + "prepared_review_unit_hash": "492428d5e63572d44b17b7547bc615de1603e8fae527fd906bcc7ecdde4495a2", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5381, + "estimated_input_tokens": 1346, + "estimated_eval_tokens": 2948 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5381, + "estimated_input_tokens": 1346, + "estimated_eval_tokens": 2948 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs appear to be the same Lumina feature branch/theme, with identical titles and overlapping Lumina model/test files. PR #8814 is the broader, more complete implementation; PR #8815 looks like a narrower follow-up touching the same model path. No issue artifacts are present.", + "confidence": 0.77, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8814 is the better canonical PR because it includes the full Lumina integration surface (docs, pipeline, scheduler, exports, and model files), making it the most complete representation of the change.", + "best_issue_reason": null, + "best_pr_reason": "PR #8814 is the best representative of the cluster: it is the more comprehensive, end-to-end change and is less likely to be an incomplete slice of the feature than #8815.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8814", + "right": "pull_request:8815", + "accept": true, + "reason": "Same title, same Lumina code area, and overlapping model/test files indicate the same underlying feature work; the narrower PR #8815 looks like a subset/follow-up to the broader Lumina implementation in #8814." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 256, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8884-3", + "nodes": [ + "pull_request:8884", + "pull_request:8888", + "pull_request:8893" + ], + "soft_pairs": [ + "pull_request:8884|pull_request:8893", + "pull_request:8888|pull_request:8893" + ], + "prepared_review_unit_hash": "3c4b009bf40d5b28711c49393f0315cf8d77d63dc46e6fceb2cbdb7e5d7c42ea", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 5194, + "estimated_input_tokens": 1299, + "estimated_eval_tokens": 2854 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 5194, + "estimated_input_tokens": 1299, + "estimated_eval_tokens": 2854 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three items are sequential pull requests touching the same workflow, but they are not duplicates of the same concrete change. PR 8884 introduces the SSH runner workflow, PR 8888 removes an unnecessary GPU option for the CPU runner, and PR 8893 removes the NVIDIA-SMI step. They look like follow-up refinements rather than the same PR.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 8884 is the best canonical PR because it is the original, most complete implementation of the SSH runner workflow and the later PRs are smaller follow-up adjustments to it.", + "best_issue_reason": null, + "best_pr_reason": "PR 8884 is the strongest representative of the cluster: it introduces the feature itself, whereas 8888 and 8893 are incremental fixes/cleanup to that workflow.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8884", + "right": "pull_request:8893", + "accept": false, + "reason": "Related workflow changes, but not the same concrete fix. 8884 adds the SSH runner workflow; 8893 only removes the NVIDIA-SMI step as a follow-up adjustment." + }, + { + "left": "pull_request:8888", + "right": "pull_request:8893", + "accept": false, + "reason": "Both modify the same workflow file, but they address different details: 8888 drops the GPU container option, while 8893 removes a diagnostic step. These are incremental edits, not duplicate PRs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 257, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-8961-2", + "nodes": [ + "pull_request:8961", + "pull_request:8966" + ], + "soft_pairs": [ + "pull_request:8961|pull_request:8966" + ], + "prepared_review_unit_hash": "8b6e51cd6d833d8488ac2f5a7bdd36454b24f657aa26d05e991b003272f4065e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3430, + "estimated_input_tokens": 858, + "estimated_eval_tokens": 1972 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3430, + "estimated_input_tokens": 858, + "estimated_eval_tokens": 1972 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs patch the same tutorial snippet for gradient accumulation in `basic_training.md` and target the same issue, with 8966 refining the earlier approach into the final idiomatic `accelerator.sync_gradients` version.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #8966 is the better canonical PR because it is the later, more correct revision of the same documentation fix and matches the final merged snippet more closely.", + "best_issue_reason": null, + "best_pr_reason": "PR #8966 is the best representative of the cluster: same issue, same file, same underlying documentation bug, but with the cleaner final implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:8961", + "right": "pull_request:8966", + "accept": true, + "reason": "Same tutorial code path and same issue target; 8966 is a direct refinement of 8961 rather than a different change, so they are duplicates at the PR level." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both PRs touch the same file, link to the same issue target (8960), and the diffs show 8966 refining the gradient-accumulation snippet introduced by 8961. The soft-edge verdict is conservative enough here because the later PR is a direct revision of the same doc fix rather than a separate bugfix." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 258, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9050-2", + "nodes": [ + "pull_request:9050", + "pull_request:9845" + ], + "soft_pairs": [ + "pull_request:9050|pull_request:9845" + ], + "prepared_review_unit_hash": "703f08d740bfcc539351cc49768042be50dd8babb8481649e6c6b2b2b0684107", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4618, + "estimated_input_tokens": 1155, + "estimated_eval_tokens": 2566 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4618, + "estimated_input_tokens": 1155, + "estimated_eval_tokens": 2566 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related test-enablement PRs in the Flux area, but they are not the same change: 9050 is a Flux-specific test/run fix, while 9845 is a broader big-GPU LoRA/integration-test marker cleanup that happens to touch one shared Flux test file.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9845 is the broader and more representative test-infrastructure change, spanning multiple suites and fixing the big-GPU gating/teardown path rather than just Flux-specific behavior.", + "best_issue_reason": null, + "best_pr_reason": "PR 9845 is the best cluster representative because it addresses the shared test execution setup across several affected tests; 9050 is narrower and not the same concrete change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9050", + "right": "pull_request:9845", + "accept": false, + "reason": "Do not merge: 9050 enables Flux tests to run by changing Flux model/pipeline test code, while 9845 updates big-GPU LoRA integration test markers and cleanup across multiple suites. They overlap on one file but fix different test-run issues and would not plausibly be one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 259, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-92-2", + "nodes": [ + "pull_request:92", + "pull_request:94" + ], + "soft_pairs": [ + "pull_request:92|pull_request:94" + ], + "prepared_review_unit_hash": "7db1261e7625195395a96c7176cc5fe109d431af4b53467c9b3d8c24748f8a8e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4320, + "estimated_input_tokens": 1080, + "estimated_eval_tokens": 2416 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4320, + "estimated_input_tokens": 1080, + "estimated_eval_tokens": 2416 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two related conversion-script PRs, but not the same underlying change: one adds an LDM converter, the other adds DDPM conversion plus shared utilities/tests. The overlap is too shallow to treat them as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 94 is the better canonical PR for the cluster because it is the broader, more complete change set: it adds the DDPM converter, also carries the LDM converter, and includes a test update.", + "best_issue_reason": null, + "best_pr_reason": "PR 94 is the strongest representative because it covers the larger conversion feature and adds test coverage, while PR 92 is a narrower single-script addition.", + "soft_edge_verdicts": [ + { + "left": "pull_request:92", + "right": "pull_request:94", + "accept": false, + "reason": "They are related conversion-script PRs, but PR 94 is not just a variant of PR 92: it introduces a separate DDPM converter and test changes, so they are not the same concrete code-path fix/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 260, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9333-2", + "nodes": [ + "pull_request:9333", + "pull_request:9671" + ], + "soft_pairs": [ + "pull_request:9333|pull_request:9671" + ], + "prepared_review_unit_hash": "68f35a306cf1c021067b03cec33dda439d0d18775932f4035f3b5d352ffa3a44", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5262, + "estimated_input_tokens": 1316, + "estimated_eval_tokens": 2888 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5262, + "estimated_input_tokens": 1316, + "estimated_eval_tokens": 2888 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two distinct CogVideoX feature PRs that overlap only in shared package/docs wiring. PR #9333 adds Video-to-Video support; PR #9671 adds CogVideoX-Fun pose/control generation. They are not the same underlying change and should not be clustered as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #9333 is the more canonical representative because it introduces the core CogVideoX video-to-video pipeline support and shared plumbing, which is broader and more foundational than the pose-control add-on in #9671.", + "best_issue_reason": null, + "best_pr_reason": "PR #9333 is the best single representative for the cluster since it is the earlier, more general CogVideoX pipeline expansion; #9671 is a separate feature on a different concrete code path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9333", + "right": "pull_request:9671", + "accept": false, + "reason": "Reject: they implement different features and code paths. #9333 adds CogVideoX video-to-video support, while #9671 adds CogVideoX-Fun pose controllable generation. Shared docs/init changes and the same subsystem are not enough to treat them as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 261, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9499-2", + "nodes": [ + "pull_request:9499", + "pull_request:9509" + ], + "soft_pairs": [ + "pull_request:9499|pull_request:9509" + ], + "prepared_review_unit_hash": "3cdd30150a517917c036a45f9ce25124d597eea86195630e027cff50b4d6828d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4272, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4272, + "estimated_input_tokens": 1068, + "estimated_eval_tokens": 2392 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate feature PRs in the same scheduler file: one adds exponential sigmas, the other adds beta sigmas. They are related but not duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:9499", + "right": "pull_request:9509", + "accept": false, + "reason": "Different concrete changes: exponential noise schedule vs beta noise schedule. Same subsystem/file, but not the same underlying bug or mergeable duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 262, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9518-2", + "nodes": [ + "pull_request:9518", + "pull_request:9538" + ], + "soft_pairs": [ + "pull_request:9518|pull_request:9538" + ], + "prepared_review_unit_hash": "42a868e6cf7b04ea9190e5a94f231e3f146aca9192f315fddd883c1ef7dcc5ac", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6414, + "estimated_input_tokens": 1604, + "estimated_eval_tokens": 3464 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6414, + "estimated_input_tokens": 1604, + "estimated_eval_tokens": 3464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs are highly similar in structure and touch many of the same scheduler files, but they implement different features: one adds exponential sigmas and the other adds beta sigmas. They are related, not duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "No true duplicate here. If a representative PR is needed, #9538 is the cleaner canonical because it is a single-feature follow-up that mirrors the same scheduler/docs pattern, but it remains distinct from #9518.", + "best_issue_reason": null, + "best_pr_reason": "#9538 is the better representative of the cluster because it is a focused enhancement PR with the same code shape as #9518, but it still addresses a different sigma type and should not absorb the other.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9518", + "right": "pull_request:9538", + "accept": false, + "reason": "Reject: despite heavy overlap in touched files and docs, these PRs add different scheduler options (exponential vs beta sigmas). They are related changes in the same subsystem, not the same concrete fix/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 263, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9564-3", + "nodes": [ + "pull_request:9564", + "pull_request:9854", + "pull_request:9855" + ], + "soft_pairs": [ + "pull_request:9564|pull_request:9854", + "pull_request:9564|pull_request:9855", + "pull_request:9854|pull_request:9855" + ], + "prepared_review_unit_hash": "907c8f32db51ca1eccb73e1430620a4104cfa027928aed54ec0e78f686c513f7", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7038, + "estimated_input_tokens": 1760, + "estimated_eval_tokens": 3776 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7038, + "estimated_input_tokens": 1760, + "estimated_eval_tokens": 3776 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The three PRs appear to cover the same PEFT compatibility guard in the same loader files. 9854 and 9855 are clearly identical; 9564 looks like the same change as well, but the evidence supports a likely match rather than a deterministic exact duplicate.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9564 is the best canonical PR because it is the earliest submission and has the most descriptive title, while the other two are fork-sync style reposts of the same loader change.", + "best_issue_reason": null, + "best_pr_reason": "PR 9564 is the strongest representative of the cluster since it is earliest and most clearly labeled, with the same change content as the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9564", + "right": "pull_request:9854", + "accept": true, + "reason": "Very likely the same change: matching diff preview, same touched files, and the same PEFT version guard in the same code path." + }, + { + "left": "pull_request:9564", + "right": "pull_request:9855", + "accept": true, + "reason": "Very likely the same change: matching diff preview and filenames indicate the same loader fix, though the evidence is soft rather than deterministic." + }, + { + "left": "pull_request:9854", + "right": "pull_request:9855", + "accept": true, + "reason": "These appear to be identical PR contents with matching diffs and filenames, so they can reasonably be treated as the same change." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The canonical choice is plausible, but the soft-edge verdicts are too strong for the evidence shown. For 9564 vs 9854 and 9564 vs 9855, the packet only provides soft_similarity plus matching diff previews/filenames, not a deterministic exact-match signal, so calling them \"exact same patch\" / \"direct duplicates\" overstates confidence. The 9854 vs 9855 pair is fine to accept, but the other two should be phrased more conservatively as likely the same change rather than confirmed duplicates." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 264, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9631-2", + "nodes": [ + "pull_request:9631", + "pull_request:9642" + ], + "soft_pairs": [ + "pull_request:9631|pull_request:9642" + ], + "prepared_review_unit_hash": "fe0a48c8f340933d0153d92cb3cd00df5dda918f3a8c38b24435e0836674b92a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2888, + "estimated_input_tokens": 722, + "estimated_eval_tokens": 1700 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 2888, + "estimated_input_tokens": 722, + "estimated_eval_tokens": 1700 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests have the same title and very similar NPU-performance intent; the merged PR 9642 includes the concrete code changes in the same file, while 9631 appears to be an earlier/duplicate version.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9642 is the better canonical PR because it is the merged change and includes the actual implementation in examples/text_to_image/train_text_to_image_sdxl.py.", + "best_issue_reason": null, + "best_pr_reason": "PR 9642 is the best representative because it is merged, specific, and contains the concrete fix; PR 9631 looks like a duplicate/unmerged predecessor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9631", + "right": "pull_request:9642", + "accept": true, + "reason": "Same title and same NPU/performance change context; 9642 is the implemented/merged version of the same underlying change in the same code path." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 265, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9651-2", + "nodes": [ + "pull_request:9651", + "pull_request:9652" + ], + "soft_pairs": [ + "pull_request:9651|pull_request:9652" + ], + "prepared_review_unit_hash": "215c46aec8c3bf3e6b08863dd1459da42e09db4a5c63cc39d54b19eb3e241660", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3180, + "estimated_input_tokens": 795, + "estimated_eval_tokens": 1846 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3180, + "estimated_input_tokens": 795, + "estimated_eval_tokens": 1846 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two small merged PRs apply the same pattern\u2014wrapping list/tuple ControlNet inputs in the appropriate MultiControlNetModel\u2014but they target different pipeline implementations and concrete code paths.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: these are parallel fixes in different pipelines, not the same change to one code path.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a better duplicate representative of the other; each is the best match for its own pipeline-specific fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9651", + "right": "pull_request:9652", + "accept": false, + "reason": "Same broad pattern, but different pipeline files, model classes, and execution paths; they do not look like one underlying bug that could be merged into a single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 266, + "component_count": 266, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-9842-2", + "nodes": [ + "pull_request:9842", + "pull_request:9879" + ], + "soft_pairs": [ + "pull_request:9842|pull_request:9879" + ], + "prepared_review_unit_hash": "d3ebf81e7705f49fe385546b931c82d35f72709a6d372e809eeb658fa5e9bfef", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4393, + "estimated_input_tokens": 1099, + "estimated_eval_tokens": 2454 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4393, + "estimated_input_tokens": 1099, + "estimated_eval_tokens": 2454 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs target the same DreamBooth SDXL LoRA config bug in the same file and same code path; PR 9879 appears to be the merged, more complete version of the same fix.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 9879 is merged and addresses the same TypeError path with an added compatibility guard for DoRA/peft, making it the stronger canonical PR.", + "best_issue_reason": null, + "best_pr_reason": "PR 9879 is the best representative because it is merged, fixes the same underlying LoRA config construction bug, and includes the extra version check needed for DoRA support.", + "soft_edge_verdicts": [ + { + "left": "pull_request:9842", + "right": "pull_request:9879", + "accept": true, + "reason": "Same file, same explicit issue target, same LoRA config TypeError in DreamBooth SDXL; 9879 looks like a superseding refinement of 9842 rather than a different change." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both PRs touch the same file, share the same issue target, and modify the same LoRA config construction path. It is also supported that 9879 is merged and adds an additional peft version guard on top of the base fix. The soft-edge pairing is conservative enough here, since the two changes look like closely related revisions that could be merged into one PR for the same bugfix." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + } + ] +} diff --git a/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/manifest.json b/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..4834baaff04b44e8ab8a23073478445609bd1894 --- /dev/null +++ b/snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/manifest.json @@ -0,0 +1,14 @@ +{ + "analysis_id": "hybrid-model-20260427t113512z", + "artifacts": { + "hybrid": "snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "model": null, + "published_at": "2026-04-27T12:02:06Z", + "repo": "huggingface/diffusers", + "schema_version": 1, + "snapshot_id": "20260427T113512Z", + "variant": "hybrid" +} diff --git a/snapshots/20260427T113512Z/manifest.json b/snapshots/20260427T113512Z/manifest.json index 4d910fb463acb9a315015108923d0f94115ff8e1..a76933008f07891bd7d41df635485c824c1e353c 100644 --- a/snapshots/20260427T113512Z/manifest.json +++ b/snapshots/20260427T113512Z/manifest.json @@ -30,6 +30,24 @@ "timeline_events": 94 }, "extracted_at": "2026-04-27T11:35:12Z", + "published_analysis": { + "canonical_analysis_id": "hybrid-model-20260427t113512z", + "runs": { + "hybrid-model-20260427t113512z": { + "analysis_id": "hybrid-model-20260427t113512z", + "artifacts": { + "hybrid": "snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "manifest_path": "snapshots/20260427T113512Z/analysis-runs/hybrid-model-20260427t113512z/manifest.json", + "model": null, + "published_at": "2026-04-27T12:02:06Z", + "variant": "hybrid" + } + }, + "schema_version": 1 + }, "repo": "huggingface/diffusers", "snapshot_id": "20260427T113512Z", "watermark": {